jackkuo commited on
Commit
67552bf
Β·
verified Β·
1 Parent(s): bd5af05

Add files using upload-large-folder tool

Browse files
This view is limited to 50 files because it contains too many changes. Β  See raw diff
Files changed (50) hide show
  1. -NAzT4oBgHgl3EQfg_yi/content/2301.01478v1.pdf +3 -0
  2. -NAzT4oBgHgl3EQfg_yi/vector_store/index.pkl +3 -0
  3. -NFIT4oBgHgl3EQf8ysQ/vector_store/index.faiss +3 -0
  4. .gitattributes +41 -0
  5. 0NE0T4oBgHgl3EQf_QLA/content/tmp_files/2301.02825v1.pdf.txt +0 -0
  6. 0NE0T4oBgHgl3EQf_QLA/content/tmp_files/load_file.txt +0 -0
  7. 1NE1T4oBgHgl3EQfRgP7/content/tmp_files/2301.03055v1.pdf.txt +0 -0
  8. 1NE1T4oBgHgl3EQfRgP7/content/tmp_files/load_file.txt +0 -0
  9. 59E1T4oBgHgl3EQf6wVZ/content/tmp_files/2301.03526v1.pdf.txt +152 -0
  10. 59E1T4oBgHgl3EQf6wVZ/content/tmp_files/load_file.txt +325 -0
  11. AdAyT4oBgHgl3EQfq_lh/content/tmp_files/2301.00552v1.pdf.txt +2011 -0
  12. AdAyT4oBgHgl3EQfq_lh/content/tmp_files/load_file.txt +0 -0
  13. BdFJT4oBgHgl3EQfsS2B/content/tmp_files/2301.11612v1.pdf.txt +1807 -0
  14. BdFJT4oBgHgl3EQfsS2B/content/tmp_files/load_file.txt +0 -0
  15. DdAzT4oBgHgl3EQfGvsw/content/tmp_files/2301.01032v1.pdf.txt +2566 -0
  16. DdAzT4oBgHgl3EQfGvsw/content/tmp_files/load_file.txt +0 -0
  17. E9E5T4oBgHgl3EQfVA-E/content/2301.05548v1.pdf +3 -0
  18. E9E5T4oBgHgl3EQfVA-E/vector_store/index.faiss +3 -0
  19. EdE1T4oBgHgl3EQfqQXv/content/tmp_files/2301.03342v1.pdf.txt +1902 -0
  20. EdE1T4oBgHgl3EQfqQXv/content/tmp_files/load_file.txt +0 -0
  21. F9AyT4oBgHgl3EQfSveK/vector_store/index.faiss +3 -0
  22. FtE0T4oBgHgl3EQfRAB3/vector_store/index.faiss +3 -0
  23. H9FKT4oBgHgl3EQfdS4O/content/2301.11819v1.pdf +3 -0
  24. H9FKT4oBgHgl3EQfdS4O/vector_store/index.pkl +3 -0
  25. HtE1T4oBgHgl3EQfXwQA/vector_store/index.faiss +3 -0
  26. HtE1T4oBgHgl3EQfXwQA/vector_store/index.pkl +3 -0
  27. IdAyT4oBgHgl3EQf5vqw/vector_store/index.faiss +3 -0
  28. JNE3T4oBgHgl3EQfXQpr/content/tmp_files/2301.04477v1.pdf.txt +1477 -0
  29. JNE3T4oBgHgl3EQfXQpr/content/tmp_files/load_file.txt +0 -0
  30. JNFJT4oBgHgl3EQfwC0U/content/tmp_files/2301.11628v1.pdf.txt +1414 -0
  31. JNFJT4oBgHgl3EQfwC0U/content/tmp_files/load_file.txt +0 -0
  32. JdE4T4oBgHgl3EQf7Q53/content/tmp_files/2301.05338v1.pdf.txt +510 -0
  33. JdE4T4oBgHgl3EQf7Q53/content/tmp_files/load_file.txt +0 -0
  34. JtE0T4oBgHgl3EQfSADl/vector_store/index.faiss +3 -0
  35. JtE0T4oBgHgl3EQfiQGP/content/tmp_files/2301.02443v1.pdf.txt +1720 -0
  36. JtE0T4oBgHgl3EQfiQGP/content/tmp_files/load_file.txt +0 -0
  37. K9E4T4oBgHgl3EQfJgxB/content/tmp_files/2301.04921v1.pdf.txt +0 -0
  38. K9E4T4oBgHgl3EQfJgxB/content/tmp_files/load_file.txt +0 -0
  39. KdE0T4oBgHgl3EQfigFl/content/tmp_files/2301.02446v1.pdf.txt +0 -0
  40. KdE0T4oBgHgl3EQfigFl/content/tmp_files/load_file.txt +0 -0
  41. LNE2T4oBgHgl3EQfpwjv/content/tmp_files/2301.04033v1.pdf.txt +1202 -0
  42. LNE2T4oBgHgl3EQfpwjv/content/tmp_files/load_file.txt +0 -0
  43. LdE0T4oBgHgl3EQfzwKd/content/tmp_files/2301.02677v1.pdf.txt +1135 -0
  44. LdE0T4oBgHgl3EQfzwKd/content/tmp_files/load_file.txt +0 -0
  45. MNAyT4oBgHgl3EQfgfgN/content/tmp_files/2301.00358v1.pdf.txt +1317 -0
  46. MNAyT4oBgHgl3EQfgfgN/content/tmp_files/load_file.txt +0 -0
  47. MNE0T4oBgHgl3EQf0AKB/vector_store/index.faiss +3 -0
  48. MNE4T4oBgHgl3EQfKAyT/content/tmp_files/2301.04926v1.pdf.txt +1133 -0
  49. MNE4T4oBgHgl3EQfKAyT/content/tmp_files/load_file.txt +0 -0
  50. MdFQT4oBgHgl3EQfVTaW/content/2301.13300v1.pdf +3 -0
-NAzT4oBgHgl3EQfg_yi/content/2301.01478v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e52e6dd63d731abc2b187ae50d4b7a51cd4a01e2836444213b49e20db5f1287c
3
+ size 2684530
-NAzT4oBgHgl3EQfg_yi/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0bad09e2520e871e567789a40925ba94936340876be8fc0255a8ed51f29b1e1e
3
+ size 290378
-NFIT4oBgHgl3EQf8ysQ/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:083abc7caa2aad1c0e1555222c39fa1c1d0fce20ded893b6b5a46bb92539d356
3
+ size 2883629
.gitattributes CHANGED
@@ -6565,3 +6565,44 @@ Z9FQT4oBgHgl3EQffTZu/content/2301.13339v1.pdf filter=lfs diff=lfs merge=lfs -tex
6565
  g9E0T4oBgHgl3EQfXgDe/content/2301.02295v1.pdf filter=lfs diff=lfs merge=lfs -text
6566
  y9FKT4oBgHgl3EQfMS25/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6567
  z9AyT4oBgHgl3EQfPPZM/content/2301.00020v1.pdf filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6565
  g9E0T4oBgHgl3EQfXgDe/content/2301.02295v1.pdf filter=lfs diff=lfs merge=lfs -text
6566
  y9FKT4oBgHgl3EQfMS25/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6567
  z9AyT4oBgHgl3EQfPPZM/content/2301.00020v1.pdf filter=lfs diff=lfs merge=lfs -text
6568
+ jNA0T4oBgHgl3EQfIv8h/content/2301.02078v1.pdf filter=lfs diff=lfs merge=lfs -text
6569
+ c9FKT4oBgHgl3EQfqS4B/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6570
+ IdAyT4oBgHgl3EQf5vqw/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6571
+ NNE2T4oBgHgl3EQfqwgI/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6572
+ xtE0T4oBgHgl3EQftQEc/content/2301.02588v1.pdf filter=lfs diff=lfs merge=lfs -text
6573
+ g9E0T4oBgHgl3EQfXgDe/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6574
+ z9AyT4oBgHgl3EQfPPZM/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6575
+ sdE2T4oBgHgl3EQffQc2/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6576
+ FtE0T4oBgHgl3EQfRAB3/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6577
+ mdE2T4oBgHgl3EQfeQfB/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6578
+ -NFIT4oBgHgl3EQf8ysQ/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6579
+ MNE0T4oBgHgl3EQf0AKB/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6580
+ JtE0T4oBgHgl3EQfSADl/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6581
+ HtE1T4oBgHgl3EQfXwQA/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6582
+ vdFLT4oBgHgl3EQfjy_2/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6583
+ wtE3T4oBgHgl3EQfOgmo/content/2301.04394v1.pdf filter=lfs diff=lfs merge=lfs -text
6584
+ idFAT4oBgHgl3EQfZx1V/content/2301.08547v1.pdf filter=lfs diff=lfs merge=lfs -text
6585
+ wtE3T4oBgHgl3EQfOgmo/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6586
+ gtE4T4oBgHgl3EQfRwx9/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6587
+ -NAzT4oBgHgl3EQfg_yi/content/2301.01478v1.pdf filter=lfs diff=lfs merge=lfs -text
6588
+ E9E5T4oBgHgl3EQfVA-E/content/2301.05548v1.pdf filter=lfs diff=lfs merge=lfs -text
6589
+ ZdFAT4oBgHgl3EQf4B4e/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6590
+ E9E5T4oBgHgl3EQfVA-E/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6591
+ MdFQT4oBgHgl3EQfVTaW/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6592
+ MdFQT4oBgHgl3EQfVTaW/content/2301.13300v1.pdf filter=lfs diff=lfs merge=lfs -text
6593
+ o9FAT4oBgHgl3EQfeB1V/content/2301.08573v1.pdf filter=lfs diff=lfs merge=lfs -text
6594
+ ZdFAT4oBgHgl3EQf4B4e/content/2301.08723v1.pdf filter=lfs diff=lfs merge=lfs -text
6595
+ ktE0T4oBgHgl3EQfYwAL/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6596
+ sdE2T4oBgHgl3EQfLgYA/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6597
+ ldE1T4oBgHgl3EQfNwO7/content/2301.03007v1.pdf filter=lfs diff=lfs merge=lfs -text
6598
+ F9AyT4oBgHgl3EQfSveK/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6599
+ TtFJT4oBgHgl3EQfMiyz/content/2301.11474v1.pdf filter=lfs diff=lfs merge=lfs -text
6600
+ qNFAT4oBgHgl3EQfex2-/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6601
+ ktE0T4oBgHgl3EQfYwAL/content/2301.02310v1.pdf filter=lfs diff=lfs merge=lfs -text
6602
+ dtA0T4oBgHgl3EQfG__M/content/2301.02056v1.pdf filter=lfs diff=lfs merge=lfs -text
6603
+ sdE2T4oBgHgl3EQffQc2/content/2301.03924v1.pdf filter=lfs diff=lfs merge=lfs -text
6604
+ idFAT4oBgHgl3EQfZx1V/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
6605
+ qNFAT4oBgHgl3EQfex2-/content/2301.08578v1.pdf filter=lfs diff=lfs merge=lfs -text
6606
+ H9FKT4oBgHgl3EQfdS4O/content/2301.11819v1.pdf filter=lfs diff=lfs merge=lfs -text
6607
+ X9AyT4oBgHgl3EQfvflc/content/2301.00631v1.pdf filter=lfs diff=lfs merge=lfs -text
6608
+ _dAzT4oBgHgl3EQfvf3k/content/2301.01709v1.pdf filter=lfs diff=lfs merge=lfs -text
0NE0T4oBgHgl3EQf_QLA/content/tmp_files/2301.02825v1.pdf.txt ADDED
The diff for this file is too large to render. See raw diff
 
0NE0T4oBgHgl3EQf_QLA/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
1NE1T4oBgHgl3EQfRgP7/content/tmp_files/2301.03055v1.pdf.txt ADDED
The diff for this file is too large to render. See raw diff
 
1NE1T4oBgHgl3EQfRgP7/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
59E1T4oBgHgl3EQf6wVZ/content/tmp_files/2301.03526v1.pdf.txt ADDED
@@ -0,0 +1,152 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ QUANTUM MAGNETISM
2
+ Thermal Hall conductivity of 𝛼-RuCl3
3
+ Hae-Young Kee
4
+ Department of Physics, University of Toronto, Toronto, Ontario, Canada
5
+ Thermal Hall conductivity originating from topological magnons is observed in the Kitaev candidate 𝜢-RuCl3 in
6
+ broad intervals of temperature and in-plane magnetic field, raising questions on the role of the Majorana mode
7
+ in heat conduction.
8
+ The black-coloured ruthenium trichloride (Ξ±-RuCl3) has a layered honeycomb structure composed of Ru3+ with a
9
+ magnetic moment of an effective spin-1/2. Although RuCl3 compounds were discovered back in the early twentieth
10
+ century, physicists only began to perceive their connection to the Kitaev spin liquid (KSL) β€” a special kind of quantum
11
+ spin liquid (QSL) β€” in 20141. The elementary excitations of a KSL, Majorana fermions and vortices, offer a platform
12
+ for quantum memory protected from decoherence, as they cannot be annihilated locally but only through fusion with
13
+ their antiparticle2. The smoking-gun signature of the KSL is 1/2-integer quantized thermal Hall conductivity under a
14
+ magnetic field, originating from unpaired Majorana moving around the edge of the sample2. Remarkably, observation
15
+ of 1/2-integer quantized thermal Hall conductivity in narrow ranges of temperature and magnetic field has been
16
+ reported3. However, such experiments were repeated by a few groups using an in-plane magnetic field4-6, and
17
+ conflicting conclusions were drawn. Despite similar-looking data, one group concluded robust 1/2-integer
18
+ quantization4, while another reported no trace of 1/2-integer quantization5, which has generated considerable debate in
19
+ the community of quantum magnetism.
20
+
21
+ The thermal Hall experiment measures the temperature change (βˆ†T) transverse to the thermal current (JQ)
22
+ applied in the sample under a magnetic field (B) (Fig. 1a) which generally measures magnetic excitations in magnetic
23
+ materials. When spins are ordered or partially aligned by an external magnetic field, that is, a polarized state, the low-
24
+ energy excitation is a collective motion of the spins, referred to as magnon. A topological magnon is characterized by a
25
+ finite Chern number associated with the Berry phase in momentum space (Fig. 1b) and may exist in the high magnetic
26
+ field region of the phase diagram. This means that a magnon propagating transverse to the thermal current leads to a
27
+ finite thermal Hall conductivity with a temperature dependence following bosonic statistics. To differentiate the source
28
+ of heat carriers, a detailed measurement of the thermal Hall conductivity is required. Writing in Nature Materials, Peter
29
+ Czajka and colleagues7 report a comprehensive measurement of the thermal Hall conductivity over broad intervals in
30
+ temperature and in-plane magnetic field (Ba) and demonstrate that the finite but not quantized thermal Hall signal arises
31
+ from topological magnons, in contrast to the earlier report of the Majorana mode being the heat carrier. A theoretical
32
+ study also found topological magnons in the polarized state using a widely accepted set of spin exchange parameters for
33
+ Ξ±-RuCl38, consistent with the conclusion drawn by Czajka and colleagues7.
34
+
35
+ Czajka and colleagues further suggest that there may be a QSL in the intermediate field region bounded by
36
+ critical magnetic field strengths Bc1 and Bc2 at very low temperatures (Fig. 1b), where deviation from the expected
37
+ magnon occurs, and oscillation of the longitudinal thermal conductivity is observed5. As the transition between the
38
+ polarized state and the QSL at Bc2 is only well defined at T = 0 K in two dimensions, this implies that there may be a
39
+ crossover between a QSL and the polarized state as the temperature increases, where the topological magnons become
40
+ responsible for the finite thermal Hall signal.
41
+
42
+ From microscopic theory, the intermediate field-induced KSL is unexpected, as the so-called vison gap
43
+ protecting the KSL is about 0.07K, where K is the Kitaev interaction2, implying that the KSL is fragile upon
44
+ introducing other perturbations. Indeed, Ξ±-RuCl3 shows a magnetic ordering with a zigzag pattern in lieu of the KSL at
45
+ low temperatures9-11, despite the dominant Kitaev interaction1,12-15. The survival of the KSL is even less likely when the
46
+ interaction is ferromagnetic. For example, a field of about 0.02K destroys the KSL and turns it into the polarized state16,
47
+ whereas for an antiferromagnetic interaction, the KSL is extended up to a field strength of about 0.3K17-23. However, it
48
+ is possible that non-Kitaev interactions work together with the Kitaev interaction and promote a QSL under a magnetic
49
+ field. Such possibilities have been investigated by several theoretical groups using various numerical techniques24-28.
50
+ Given the huge phase space of exchange parameters, the focus was near the ferromagnetic Kitaev interaction regime
51
+ relevant for Ξ±-RuCl3. A direct transition from the zigzag order to the polarized state was found when the magnetic field
52
+ is applied in the plane25,26, contradicting the experimental observations. Strikingly, when the field is oriented out of the
53
+ plane, a magnetically disordered intermediate phase was found26-28. The strong anisotropic field response is due to the
54
+ non-Kitaev interaction called Ξ“26,28,29. Whether the intermediate state boosted by the positive Ξ“ interaction is a QSL or
55
+ not remains to be resolved, as a finite Ξ“ allows for mobile visons, and the free Majorana picture of the pure Kitaev
56
+ model does not work. However, the effects of the Ξ“ interaction and a magnetic field somehow cancel, and the field-
57
+ induced intermediate phase may map to the effective KSL with a perturbing magnetic field. While this scenario seems
58
+ unlikely, it has not been ruled out.
59
+
60
+
61
+ There are experimental challenges owing to a strong sample dependence30,31. The layers of Ξ±-RuCl3 are stacked
62
+ via a weak van der Waals interaction and different types of stacking are naturally expected10,11,30-34. Depending on the
63
+ stacking pattern of Ξ±-RuCl3, the in-plane spin exchange parameters vary, because of the changes in the Ru–Ru ion bond
64
+ length and the angle between Ru–Cl–Ru bonds32. As this sensitivity traces back to the spin–orbit entangled
65
+ wavefunction35, it is difficult to avoid. If the Kitaev interaction is antiferromagnetic in certain samples, a more robust
66
+ spin liquid and a proposed U(1) spin liquid may occur under the magnetic field21-23. If so, the moment direction in the
67
+ ordered states may differ from the samples with a dominant ferromagnetic Kitaev interaction36,37. Looking ahead,
68
+ thorough experimental studies on a given sample that give a full set of information, including the layer stackings, the
69
+ moment direction of the magnetic order, the anisotropy in the susceptibilities, the dynamic excitations and the thermal
70
+ Hall measurements in different field directions, will advance our search for a material realization of a KSL.
71
+
72
+
73
+ References:
74
+ 1. Plumb, K.W. et al. Phys. Rev. B 90, 04112(R) (2014).
75
+ 2. Kitaev, A. Ann. Phys. 321, 2 (2006).
76
+ 3. Kasahara, Y. et al. Nature 559, 227 (2018).
77
+ 4. Bruin, J. A. N. et al. Nat. Phys. 18, 401 (2022).
78
+ 5. Czajka, P. et al. Nat. Phys. 17, 915 (2021).
79
+ 6. Lefrancois, E. et al. Phys. Rev. X 12, 021025 (2022).
80
+ 7. Czajka, P. et al. Nat. Materials 22, 36 (2023): https://doi.org/10.1038/s41563-022-01397-w
81
+ 8. Zhang, E., Chern, L. E. & Kim, Y. B. Phys. Rev. B 103, 174402 (2021).
82
+ 9. Sears, J. A. et al. Phys. Rev. B 91, 144420 (2015).
83
+ 10. Johnson, R. D. et al. Phys. Rev. B 92, 235119 (2015).
84
+ 11. Cao, H. B. et al. Phys. Rev. B 93, 134423 (2016).
85
+ 12. Kim, H. S. et al. Phys. Rev. B 91, 24110 (2015).
86
+ 13. Sandilands, L. J. et al. Phys. Rev. Lett. 114, 147201 (2015).
87
+ 14. Banerjee, A. et al. Nature Materials 15, 733 (2016).
88
+ 15. Sandilands, L. J. et al. Phys. Rev. B 93, 075144 (2016).
89
+ 16. Jiang, H.-C., Gu, Z.-C., Qi, X.-L. & Trebst, S. Phys. Rev. B 83, 245104 (2011).
90
+ 17. Zhu, Z., Kimchi, I., Sheng, D. N. & Fu, L. Phys. Rev. B 97, 24110 (2018).
91
+ 18. Nasu, J., Kato, Y., Kamiya, Y. & Motome, Y. Phys. Rev. B 98, 060416 (2018).
92
+ 19. Liang, S. et al. Phys. Rev. B 98, 054433 (2018).
93
+ 20. Gohlke, M., Moessner, R. & Pollmann, F. Phys. Rev. B 98, 014418 (2018).
94
+ 21. Jiang, H.-C., Wang, C.-Y., Huang, B. & Lu, Y.-M. arXiv:1809.08247 (2018).
95
+ 22. Hickey, C. & Trebst, S. Nature Commun. 10, 530 (2019).
96
+ 23. Patel, N. D. & Trivedi, N. Proc. Natl. Acad. Sci. 116, 12199 (2019).
97
+ 24. Yadav, R. et al. Sci. Rep. 6, 37925 (2016).
98
+ 25. Winter, S. et al. Phys. Rev. Lett. 120, 077203 (2018).
99
+ 26. Gordon, J. S., Catuneanu, A., Sorensen, E. & Kee, H.-Y. Nature Commun. 10, 2470 (2019).
100
+ 27. Kaib, D. A. S., Winter, S. & Valenti., R. Phys. Rev. B 100, 14445 (2019).
101
+ 28. Lee., H.-Y. et al. Nature Commun. 11, 1639 (2020).
102
+ 29. Rau, J. G., Lee, E. K.-H. & Kee, H.-Y. Phys. Rev. Lett. 112, 077204 (2014).
103
+ 30. Yamashita, M. et al. Phys. Rev. B 102, 220404 (2020).
104
+ 31. Kasahara, Y. et al. Phys. Rev. B 106, L060410 (2022).
105
+ 32. Kim, H.-S. & Kee, H.-Y., Phys. Rev. B 93, 155143 (2016).
106
+ 33. Winter, S., Li, Y., Jeschke, H. O. & Valenti, R. Phys. Rev. B 93, 214431 (2016).
107
+ 34. Balz. C., et al. Phys. Rev. B 103, 174417 (2021).
108
+ 35. Jackeli, G. & Khaliullin, G. Phys. Rev. Lett. 102, 017205 (2009).
109
+ 36. Chaloupka, J. & Khaliullin, G. Phys. Rev. B 94, 064435 (2016).
110
+ 37. Sears, J. A. et al. Nature Physics 16, 837 (2020).
111
+
112
+
113
+
114
+
115
+
116
+
117
+
118
+
119
+
120
+
121
+
122
+
123
+
124
+
125
+
126
+ Figure 1 Thermal Measurements and Phase Diagram of Ξ±-RuCl3. a, Thermal Hall experimental set-up. The
127
+ magnetic field (B) and thermal current (JQ) are applied along the a axis (in-plane direction perpendicular to one of the
128
+ honeycomb bonds), and the change of temperature (Ξ”T) is measured across the b axis. The yellow arrow represents a
129
+ heat carrier leading to a finite thermal Hall conductivity. For the other in-plane direction, that is, the b axis (parallel to
130
+ one of the honeycomb bonds) set-up, the thermal Hall conductivity vanishes due to the ac-mirror plane (or C2 rotation
131
+ about the b axis) symmetry. b, Phase diagram with respect to the a-axis field strength (Ba) and temperature (T). At low
132
+ temperatures without the magnetic field, a magnetic order with a zigzag pattern is found, indicated by red and blue
133
+ arrows. As the field strength increases, the zigzag pattern along the layers is modified (zz2)34, indicating a weak
134
+ interlayer spin interaction altered by the in-plane field. A puzzling QSL between Bc1 and Bc2 before the polarized state
135
+ (PS) was suggested at very low temperatures. As temperature increases, a crossover to the polarized state may occur.
136
+ The rainbow-coloured hexagon in the polarized state denotes the Berry curvature in the first Brillouin zone, leading to a
137
+ finite thermal Hall conductivity.
138
+
139
+ (a)
140
+ β–³T
141
+ B
142
+ JQ
143
+ (b)
144
+ T
145
+ topologicalmagnon
146
+ zig-zag
147
+ PS
148
+ ZZ2
149
+ QSL?
150
+ Bc1
151
+ Bc2
152
+ Ba
59E1T4oBgHgl3EQf6wVZ/content/tmp_files/load_file.txt ADDED
@@ -0,0 +1,325 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf,len=324
2
+ page_content='QUANTUM MAGNETISM Thermal Hall conductivity of 𝛼-RuCl3 Hae-Young Kee Department of Physics, University of Toronto, Toronto, Ontario, Canada Thermal Hall conductivity originating from topological magnons is observed in the Kitaev candidate 𝜢-RuCl3 in broad intervals of temperature and in-plane magnetic field, raising questions on the role of the Majorana mode in heat conduction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
3
+ page_content=' The black-coloured ruthenium trichloride (Ξ±-RuCl3) has a layered honeycomb structure composed of Ru3+ with a magnetic moment of an effective spin-1/2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
4
+ page_content=' Although RuCl3 compounds were discovered back in the early twentieth century, physicists only began to perceive their connection to the Kitaev spin liquid (KSL) β€” a special kind of quantum spin liquid (QSL) β€” in 20141.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
5
+ page_content=' The elementary excitations of a KSL, Majorana fermions and vortices, offer a platform for quantum memory protected from decoherence, as they cannot be annihilated locally but only through fusion with their antiparticle2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
6
+ page_content=' The smoking-gun signature of the KSL is 1/2-integer quantized thermal Hall conductivity under a magnetic field, originating from unpaired Majorana moving around the edge of the sample2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
7
+ page_content=' Remarkably, observation of 1/2-integer quantized thermal Hall conductivity in narrow ranges of temperature and magnetic field has been reported3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
8
+ page_content=' However, such experiments were repeated by a few groups using an in-plane magnetic field4-6, and conflicting conclusions were drawn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
9
+ page_content=' Despite similar-looking data, one group concluded robust 1/2-integer quantization4, while another reported no trace of 1/2-integer quantization5, which has generated considerable debate in the community of quantum magnetism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
10
+ page_content=' The thermal Hall experiment measures the temperature change (βˆ†T) transverse to the thermal current (JQ) applied in the sample under a magnetic field (B) (Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
11
+ page_content=' 1a) which generally measures magnetic excitations in magnetic materials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
12
+ page_content=' When spins are ordered or partially aligned by an external magnetic field, that is, a polarized state, the low- energy excitation is a collective motion of the spins, referred to as magnon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
13
+ page_content=' A topological magnon is characterized by a finite Chern number associated with the Berry phase in momentum space (Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
14
+ page_content=' 1b) and may exist in the high magnetic field region of the phase diagram.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
15
+ page_content=' This means that a magnon propagating transverse to the thermal current leads to a finite thermal Hall conductivity with a temperature dependence following bosonic statistics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
16
+ page_content=' To differentiate the source of heat carriers, a detailed measurement of the thermal Hall conductivity is required.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
17
+ page_content=' Writing in Nature Materials, Peter Czajka and colleagues7 report a comprehensive measurement of the thermal Hall conductivity over broad intervals in temperature and in-plane magnetic field (Ba) and demonstrate that the finite but not quantized thermal Hall signal arises from topological magnons, in contrast to the earlier report of the Majorana mode being the heat carrier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
18
+ page_content=' A theoretical study also found topological magnons in the polarized state using a widely accepted set of spin exchange parameters for Ξ±-RuCl38, consistent with the conclusion drawn by Czajka and colleagues7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
19
+ page_content=' Czajka and colleagues further suggest that there may be a QSL in the intermediate field region bounded by critical magnetic field strengths Bc1 and Bc2 at very low temperatures (Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
20
+ page_content=' 1b), where deviation from the expected magnon occurs, and oscillation of the longitudinal thermal conductivity is observed5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
21
+ page_content=' As the transition between the polarized state and the QSL at Bc2 is only well defined at T = 0 K in two dimensions, this implies that there may be a crossover between a QSL and the polarized state as the temperature increases, where the topological magnons become responsible for the finite thermal Hall signal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
22
+ page_content=' From microscopic theory, the intermediate field-induced KSL is unexpected, as the so-called vison gap protecting the KSL is about 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
23
+ page_content='07K, where K is the Kitaev interaction2, implying that the KSL is fragile upon introducing other perturbations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
24
+ page_content=' Indeed, Ξ±-RuCl3 shows a magnetic ordering with a zigzag pattern in lieu of the KSL at low temperatures9-11, despite the dominant Kitaev interaction1,12-15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
25
+ page_content=' The survival of the KSL is even less likely when the interaction is ferromagnetic.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
26
+ page_content=' For example, a field of about 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
27
+ page_content='02K destroys the KSL and turns it into the polarized state16, whereas for an antiferromagnetic interaction, the KSL is extended up to a field strength of about 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
28
+ page_content='3K17-23.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
29
+ page_content=' However, it is possible that non-Kitaev interactions work together with the Kitaev interaction and promote a QSL under a magnetic field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
30
+ page_content=' Such possibilities have been investigated by several theoretical groups using various numerical techniques24-28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
31
+ page_content=' Given the huge phase space of exchange parameters, the focus was near the ferromagnetic Kitaev interaction regime relevant for Ξ±-RuCl3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
32
+ page_content=' A direct transition from the zigzag order to the polarized state was found when the magnetic field is applied in the plane25,26, contradicting the experimental observations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
33
+ page_content=' Strikingly, when the field is oriented out of the plane, a magnetically disordered intermediate phase was found26-28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
34
+ page_content=' The strong anisotropic field response is due to the non-Kitaev interaction called Ξ“26,28,29.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
35
+ page_content=' Whether the intermediate state boosted by the positive Ξ“ interaction is a QSL or not remains to be resolved, as a finite Ξ“ allows for mobile visons, and the free Majorana picture of the pure Kitaev model does not work.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
36
+ page_content=' However, the effects of the Ξ“ interaction and a magnetic field somehow cancel, and the field- induced intermediate phase may map to the effective KSL with a perturbing magnetic field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
37
+ page_content=' While this scenario seems unlikely, it has not been ruled out.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
38
+ page_content=' There are experimental challenges owing to a strong sample dependence30,31.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
39
+ page_content=' The layers of Ξ±-RuCl3 are stacked via a weak van der Waals interaction and different types of stacking are naturally expected10,11,30-34.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
40
+ page_content=' Depending on the stacking pattern of Ξ±-RuCl3, the in-plane spin exchange parameters vary, because of the changes in the Ru–Ru ion bond length and the angle between Ru–Cl–Ru bonds32.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
41
+ page_content=' As this sensitivity traces back to the spin–orbit entangled wavefunction35, it is difficult to avoid.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
42
+ page_content=' If the Kitaev interaction is antiferromagnetic in certain samples, a more robust spin liquid and a proposed U(1) spin liquid may occur under the magnetic field21-23.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
43
+ page_content=' If so, the moment direction in the ordered states may differ from the samples with a dominant ferromagnetic Kitaev interaction36,37.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
44
+ page_content=' Looking ahead, thorough experimental studies on a given sample that give a full set of information, including the layer stackings, the moment direction of the magnetic order, the anisotropy in the susceptibilities, the dynamic excitations and the thermal Hall measurements in different field directions, will advance our search for a material realization of a KSL.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
45
+ page_content=' References: 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
46
+ page_content=' Plumb, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
47
+ page_content='W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
48
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
49
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
50
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
51
+ page_content=' B 90, 04112(R) (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
52
+ page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
53
+ page_content=' Kitaev, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
54
+ page_content=' Ann.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
55
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
56
+ page_content=' 321, 2 (2006).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
57
+ page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
58
+ page_content=' Kasahara, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
59
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
60
+ page_content=' Nature 559, 227 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
61
+ page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
62
+ page_content=' Bruin, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
63
+ page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
64
+ page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
65
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
66
+ page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
67
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
68
+ page_content=' 18, 401 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
69
+ page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
70
+ page_content=' Czajka, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
71
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
72
+ page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
73
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
74
+ page_content=' 17, 915 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
75
+ page_content=' 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
76
+ page_content=' Lefrancois, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
77
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
78
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
79
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
80
+ page_content=' X 12, 021025 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
81
+ page_content=' 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
82
+ page_content=' Czajka, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
83
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
84
+ page_content=' Nat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
85
+ page_content=' Materials 22, 36 (2023): https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
86
+ page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
87
+ page_content='1038/s41563-022-01397-w 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
88
+ page_content=' Zhang, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
89
+ page_content=', Chern, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
90
+ page_content=' E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
91
+ page_content=' & Kim, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
92
+ page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
93
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
94
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
95
+ page_content=' B 103, 174402 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
96
+ page_content=' 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
97
+ page_content=' Sears, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
98
+ page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
99
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
100
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
101
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
102
+ page_content=' B 91, 144420 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
103
+ page_content=' 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
104
+ page_content=' Johnson, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
105
+ page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
106
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
107
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
108
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
109
+ page_content=' B 92, 235119 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
110
+ page_content=' 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
111
+ page_content=' Cao, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
112
+ page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
113
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
114
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
115
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
116
+ page_content=' B 93, 134423 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
117
+ page_content=' 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
118
+ page_content=' Kim, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
119
+ page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
120
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
121
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
122
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
123
+ page_content=' B 91, 24110 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
124
+ page_content=' 13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
125
+ page_content=' Sandilands, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
126
+ page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
127
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
128
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
129
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
130
+ page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
131
+ page_content=' 114, 147201 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
132
+ page_content=' 14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
133
+ page_content=' Banerjee, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
134
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
135
+ page_content=' Nature Materials 15, 733 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
136
+ page_content=' 15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
137
+ page_content=' Sandilands, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
138
+ page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
139
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
140
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
141
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
142
+ page_content=' B 93, 075144 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
143
+ page_content=' 16.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
144
+ page_content=' Jiang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
145
+ page_content='-C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
146
+ page_content=', Gu, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
147
+ page_content='-C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
148
+ page_content=', Qi, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
149
+ page_content='-L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
150
+ page_content=' & Trebst, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
151
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
152
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
153
+ page_content=' B 83, 245104 (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
154
+ page_content=' 17.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
155
+ page_content=' Zhu, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
156
+ page_content=', Kimchi, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
157
+ page_content=', Sheng, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
158
+ page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
159
+ page_content=' & Fu, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
160
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
161
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
162
+ page_content=' B 97, 24110 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
163
+ page_content=' 18.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
164
+ page_content=' Nasu, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
165
+ page_content=', Kato, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
166
+ page_content=', Kamiya, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
167
+ page_content=' & Motome, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
168
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
169
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
170
+ page_content=' B 98, 060416 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
171
+ page_content=' 19.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
172
+ page_content=' Liang, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
173
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
174
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
175
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
176
+ page_content=' B 98, 054433 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
177
+ page_content=' 20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
178
+ page_content=' Gohlke, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
179
+ page_content=', Moessner, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
180
+ page_content=' & Pollmann, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
181
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
182
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
183
+ page_content=' B 98, 014418 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
184
+ page_content=' 21.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
185
+ page_content=' Jiang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
186
+ page_content='-C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
187
+ page_content=', Wang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
188
+ page_content='-Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
189
+ page_content=', Huang, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
190
+ page_content=' & Lu, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
191
+ page_content='-M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
192
+ page_content=' arXiv:1809.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
193
+ page_content='08247 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
194
+ page_content=' 22.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
195
+ page_content=' Hickey, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
196
+ page_content=' & Trebst, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
197
+ page_content=' Nature Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
198
+ page_content=' 10, 530 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
199
+ page_content=' 23.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
200
+ page_content=' Patel, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
201
+ page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
202
+ page_content=' & Trivedi, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
203
+ page_content=' Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
204
+ page_content=' Natl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
205
+ page_content=' Acad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
206
+ page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
207
+ page_content=' 116, 12199 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
208
+ page_content=' 24.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
209
+ page_content=' Yadav, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
210
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
211
+ page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
212
+ page_content=' Rep.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
213
+ page_content=' 6, 37925 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
214
+ page_content=' 25.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
215
+ page_content=' Winter, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
216
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
217
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
218
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
219
+ page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
220
+ page_content=' 120, 077203 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
221
+ page_content=' 26.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
222
+ page_content=' Gordon, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
223
+ page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
224
+ page_content=', Catuneanu, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
225
+ page_content=', Sorensen, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
226
+ page_content=' & Kee, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
227
+ page_content='-Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
228
+ page_content=' Nature Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
229
+ page_content=' 10, 2470 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
230
+ page_content=' 27.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
231
+ page_content=' Kaib, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
232
+ page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
233
+ page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
234
+ page_content=', Winter, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
235
+ page_content=' & Valenti.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
236
+ page_content=', R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
237
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
238
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
239
+ page_content=' B 100, 14445 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
240
+ page_content=' 28.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
241
+ page_content=' Lee.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
242
+ page_content=', H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
243
+ page_content='-Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
244
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
245
+ page_content=' Nature Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
246
+ page_content=' 11, 1639 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
247
+ page_content=' 29.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
248
+ page_content=' Rau, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
249
+ page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
250
+ page_content=', Lee, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
251
+ page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
252
+ page_content='-H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
253
+ page_content=' & Kee, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
254
+ page_content='-Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
255
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
256
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
257
+ page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
258
+ page_content=' 112, 077204 (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
259
+ page_content=' 30.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
260
+ page_content=' Yamashita, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
261
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
262
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
263
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
264
+ page_content=' B 102, 220404 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
265
+ page_content=' 31.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
266
+ page_content=' Kasahara, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
267
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
268
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
269
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
270
+ page_content=' B 106, L060410 (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
271
+ page_content=' 32.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
272
+ page_content=' Kim, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
273
+ page_content='-S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
274
+ page_content=' & Kee, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
275
+ page_content='-Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
276
+ page_content=', Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
277
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
278
+ page_content=' B 93, 155143 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
279
+ page_content=' 33.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
280
+ page_content=' Winter, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
281
+ page_content=', Li, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
282
+ page_content=', Jeschke, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
283
+ page_content=' O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
284
+ page_content=' & Valenti, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
285
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
286
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
287
+ page_content=' B 93, 214431 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
288
+ page_content=' 34.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
289
+ page_content=' Balz.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
290
+ page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
291
+ page_content=', et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
292
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
293
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
294
+ page_content=' B 103, 174417 (2021).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
295
+ page_content=' 35.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
296
+ page_content=' Jackeli, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
297
+ page_content=' & Khaliullin, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
298
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
299
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
300
+ page_content=' Lett.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
301
+ page_content=' 102, 017205 (2009).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
302
+ page_content=' 36.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
303
+ page_content=' Chaloupka, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
304
+ page_content=' & Khaliullin, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
305
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
306
+ page_content=' Rev.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
307
+ page_content=' B 94, 064435 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
308
+ page_content=' 37.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
309
+ page_content=' Sears, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
310
+ page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
311
+ page_content=' et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
312
+ page_content=' Nature Physics 16, 837 (2020).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
313
+ page_content=' Figure 1 Thermal Measurements and Phase Diagram of Ξ±-RuCl3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
314
+ page_content=' a, Thermal Hall experimental set-up.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
315
+ page_content=' The magnetic field (B) and thermal current (JQ) are applied along the a axis (in-plane direction perpendicular to one of the honeycomb bonds), and the change of temperature (Ξ”T) is measured across the b axis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
316
+ page_content=' The yellow arrow represents a heat carrier leading to a finite thermal Hall conductivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
317
+ page_content=' For the other in-plane direction, that is, the b axis (parallel to one of the honeycomb bonds) set-up, the thermal Hall conductivity vanishes due to the ac-mirror plane (or C2 rotation about the b axis) symmetry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
318
+ page_content=' b, Phase diagram with respect to the a-axis field strength (Ba) and temperature (T).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
319
+ page_content=' At low temperatures without the magnetic field, a magnetic order with a zigzag pattern is found, indicated by red and blue arrows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
320
+ page_content=' As the field strength increases, the zigzag pattern along the layers is modified (zz2)34, indicating a weak interlayer spin interaction altered by the in-plane field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
321
+ page_content=' A puzzling QSL between Bc1 and Bc2 before the polarized state (PS) was suggested at very low temperatures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
322
+ page_content=' As temperature increases, a crossover to the polarized state may occur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
323
+ page_content=' The rainbow-coloured hexagon in the polarized state denotes the Berry curvature in the first Brillouin zone, leading to a finite thermal Hall conductivity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
324
+ page_content=' (a) β–³T B JQ (b) T topologicalmagnon zig zag PS ZZ2 QSL?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
325
+ page_content=' Bc1 Bc2 Ba' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/59E1T4oBgHgl3EQf6wVZ/content/2301.03526v1.pdf'}
AdAyT4oBgHgl3EQfq_lh/content/tmp_files/2301.00552v1.pdf.txt ADDED
@@ -0,0 +1,2011 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 1
2
+ NEURAL SOURCE/SINK PHASE CONNECTIVITY IN
3
+ DEVELOPMENTAL DYSLEXIA BY MEANS OF INTERCHANNEL CAUSALITY
4
+ I. RODRÍGUEZ-RODRÍGUEZ, A. ORTIZ, N.J. GALLEGO-MOLINA, M.A. FORMOSO
5
+ Departamento de IngenierΓ­a de Comunicaciones, Universidad de MΓ‘laga, 29004 MΓ‘laga, Spain
6
+ {ignacio.rodriguez, aortiz, njgm, marco.a.formoso}@ic.uma.es
7
+ W. L. WOO
8
+ Department of Computer and Information Sciences, Northumbria University, Newcastle upon Tyne NE1 8ST, UK
9
+ wailok.woo@northumbria.ac.uk
10
+ While the brain connectivity network can inform the understanding and diagnosis of developmental dyslexia, its
11
+ cause-effect relationships have not yet enough been examined. Employing electroencephalography signals and band-
12
+ limited white noise stimulus at 4.8 Hz (prosodic-syllabic frequency), we measure the phase Granger causalities among
13
+ channels to identify differences between dyslexic learners and controls, thereby proposing a method to calculate
14
+ directional connectivity. As causal relationships run in both directions, we explore three scenarios, namely channels’
15
+ activity as sources, as sinks, and in total. Our proposed method can be used for both classification and exploratory
16
+ analysis. In all scenarios, we find confirmation of the established right-lateralized Theta sampling network anomaly,
17
+ in line with the temporal sampling framework’s assumption of oscillatory differences in the Theta and Gamma bands.
18
+ Further, we show that this anomaly primarily occurs in the causal relationships of channels acting as sinks, where it
19
+ is significantly more pronounced than when only total activity is observed. In the sink scenario, our classifier obtains
20
+ 0.84 and 0.88 accuracy and 0.87 and 0.93 AUC for the Theta and Gamma bands, respectively.
21
+ Keywords: Developmental Dyslexia, EEG, Granger causality, functional connectivity, anomaly detection;
22
+ 1. Introduction
23
+ Developmental dyslexia (DD) is a learning difficulty that
24
+ typically causes various reading difficulties, including
25
+ letter migration and frequent spelling errors. In any given
26
+ population, between 5% and 12% of learners are likely to
27
+ have DD, depending on the test battery used 1. DD is
28
+ traditionally diagnosed using behavioral tests of reading
29
+ and writing skills, but these are vulnerable to exogenous
30
+ factors, such as attitude or disposition, leading to
31
+ diagnoses that may be fundamentally unsound 2. It is,
32
+ therefore, imperative to develop more objective metrics
33
+ that can offer a more accurate diagnosis among young
34
+ learners. A stimulus system that remains uninfluenced by
35
+ the learner’s behavior, actions and context (e.g., native
36
+ language or learning level) would be extremely valuable.
37
+ If the stimulus further involves the simulation of prosody,
38
+ i.e. the white noise at the usual frequency of the language
39
+ envelope, it may also inform our understanding of the
40
+ brain areas active in auditory processing, indicating the
41
+ differences between learners with and without dyslexia.
42
+ While various neuroscience methods for gathering
43
+ functional brain data exist, including functional magnetic
44
+ resonance imaging (fMRI) 3, magnetoencephalography
45
+ (MEG) and functional near-infrared spectroscopy
46
+ (fNIRS), electroencephalography (EEG) continues to be
47
+ the most widely used and least costly method to assess
48
+ cortical brain activity with enhanced temporal resolution.
49
+ An EEG measures several frequency bands, namely the
50
+ Delta, Theta, Alpha, Beta, and Gamma bands, which do
51
+ not experience stimulation equally, and it is thus
52
+ generally held that the stimulation of one band can
53
+ transfer to the others. Using EEG to separately
54
+ investigate the patterns emerging in these bands may
55
+ offer valuable insights for the research on DD.
56
+ EEG is well-established in DD studies exploring the
57
+ functional network connectivity and organization of the
58
+ brain. Functional connectivity means the level of
59
+ coordination between the activities in different areas of
60
+ the brain while the learner is engaging in a task. Prior
61
+ research has produced various techniques that employ
62
+ EEG to assess functional connectivity to, for example,
63
+ determine what patterns are characteristic of neurological
64
+ conditions, including Parkinson’s disease 4. Studies in
65
+ cognitive neuroscience have also used brain connectivity
66
+ to identify brain areas crucial to language and learning 5.
67
+
68
+ RodrΓ­guez-RodrΓ­guez et al.
69
+
70
+ 2
71
+ Connectivity analysis has allowed neuroscience to
72
+ provide even deeper insights
73
+ 6 by analyzing the
74
+ parameters linking two signals gathered through two
75
+ distinct channels, such as their correlation, causality and
76
+ covariance 7. Employing connectivity analysis to brain
77
+ signals measured in different regions allows us to explore
78
+ the neural network, in line with the notion that the brain
79
+ is hyper-connected 8.
80
+ Notably, brain connectivity is not solely limited to the
81
+ interactions between areas, with regions potentially
82
+ influencing each other through, e.g., phase-phase or
83
+ phase-amplitude modulations among bands 9. We here
84
+ consider areas that primarily exert an influence as
85
+ sources, while those more likely to be subject to this
86
+ influence are sinks. This novel consideration of
87
+ connectivity in terms of sources and sinks can not only
88
+ serve classification but also facilitate exploratory
89
+ analysis.
90
+ We propose extracting the frequency components of each
91
+ band from the EEG signals acquired under prosodic
92
+ auditory stimuli, subsequently using them to generate a
93
+ connectivity model based on inter-channel Granger
94
+ causality 10. By modelling connectivity as sources and
95
+ sinks, we seek to clarify the existence of abnormalities
96
+ between learners with and without DD, aiming to offer
97
+ an
98
+ enhanced understanding
99
+ of
100
+ the
101
+ mechanisms
102
+ underlying DD, ultimately allowing early diagnosis.
103
+ The remainder of the paper is structured as follows.
104
+ Section 2 describes the most relevant extant work in this
105
+ field, and Section 3 outlines the data and methodology,
106
+ including the preprocessing, Granger causality matrices
107
+ and connectivity matrices construction, and classification
108
+ algorithms. Section 4 presents our main results, leading
109
+ to the discussion in Section 5. Finally, Section 6 presents
110
+ the main conclusions and contributions of this work.
111
+ 2. Related works
112
+ Previous studies have indicated that the phonological
113
+ deficit that causes DD may be due to an impairment in
114
+ the neural encoding of low-frequency speech envelopes,
115
+ relating to speech prosody 11. There is evidence of
116
+ significant difficulties among that learners with DD in
117
+ tasks relying on prosodic awareness, e.g. identifying
118
+ syllable stress, compared to controls at an earlier reading
119
+ level 12. This indicates the presence of atypical oscillatory
120
+ functioning in low-frequency brain rhythms in DD 13.
121
+ There has been substantial research on the important role
122
+ played by the ability to perceive prosodic frequency.
123
+ After directly measuring the neural encoding of
124
+ children’s speech using EEG, Power et al.
125
+ 11
126
+ reconstructed the participants’ speech stimulus envelopes
127
+ using the emergent patterns. The EEG recordings were
128
+ done while the participants were performing a word
129
+ report task using noise-vocoded speech, i.e. still with a
130
+ low-frequency envelope yet with a degraded temporal
131
+ fine structure (TFS) of speech. Due to this degradation,
132
+ the participants necessarily derived the spoken words and
133
+ sentences from the information given by the envelope. If
134
+ the learners could accurately perceive the words and
135
+ sentences, it was possible to evaluate the functioning of
136
+ their neural encoding of the low-frequency envelopes in
137
+ speech, which is likely impaired in learners with DD
138
+ according to temporal sampling theory.
139
+ Brain activity, and thus the connectivity network, occurs
140
+ across various frequency bands, as demonstrated via the
141
+ temporal sampling framework (TSF). Temporal coding
142
+ is thought to be partially attributed to synchronous
143
+ auditory cortex activity, wherein the network neurons
144
+ synchronize
145
+ endogenous
146
+ oscillations
147
+ at
148
+ different
149
+ preferred rates while matching the temporal information
150
+ of the acoustic speech signal 14 15 16. The auditory and
151
+ visual parts of speech unfold across different timescales,
152
+ and thus, when the neurons in auditory and visual cortices
153
+ oscillate, they are believed to phase-align their activity to
154
+ match the input’s modulation rates 17.
155
+ TSF proposes that atypical oscillatory sampling at
156
+ various temporal rates may be the cause of the
157
+ phonological impairment in DD. Furthermore, a potential
158
+ biological mechanism for DD has recently been
159
+ suggested, highlighting the presence of atypical
160
+ dominant neural entrainment 18 for the slow rhythmic
161
+ prosodic (0.5–1 Hz), syllabic (4–8 Hz) and phoneme (12–
162
+ 40Hz) rhythm categories 19. Following this line of
163
+ thought, we might consider learners with DD to have
164
+ atypical oscillatory sampling for at least one temporal
165
+ rate, leading to difficulties in phonologically capturing
166
+ linguistic units such as syllables or phonemes.
167
+ However, this phenomenon is not likely to be
168
+ experienced equally across all frequency bands (i.e.
169
+ Delta, Theta, Alpha, Beta, and Gamma). Thus, it seems
170
+ pertinent to examine these bands’ connectivity patterns
171
+ separately using EEG. Prior research has indeed used
172
+ EEG or MEG to investigate the fundamental mechanisms
173
+ underlying DD, implementing speech-based stimuli
174
+ under the premise that DD is essentially derived from a
175
+ lesser awareness of individual speech units 20. Using
176
+ visual and auditory stimulus, Power et al. 21, for example,
177
+ identified differences between learners with DD and a
178
+ control group in the preferred entrainment phase of the
179
+ Delta and Theta bands. Based on changes in the
180
+ frequency, phase, and power spectrum, it thus becomes
181
+ feasible to derive measures of spectral connectivity. In
182
+ line with this, there are techniques showing the statistical
183
+
184
+
185
+ Neural source/sink phase connectivity in Developmental Dyslexia
186
+
187
+ 3
188
+ relationship between electrodes on the same frequency
189
+ band 22.
190
+ The prior research has also explored the inference from
191
+ connectivity patterns during reading tasks. For example,
192
+ Žarić et al. 23 used visual word and false font processing
193
+ tasks to investigate disruptions in the connectivity
194
+ between the visual and language processing networks.
195
+ They hereby calculated the connectivity patterns based
196
+ on how statistically significant the differences in the
197
+ power spectral density (PSD) were for each EEG band.
198
+ Language-based reading or writing-related tasks have
199
+ also been used in previous studies identifying
200
+ discriminant patterns in EEG signals. For instance, using
201
+ graph theory, GonzΓ‘lez et al. 24 compared the EEG
202
+ measurements of participants performing audiovisual
203
+ tasks or at rest to determine differences in the
204
+ connectivity patterns. Meanwhile, Stam et al. 25 used a
205
+ phase lag index to compute multiple weighted
206
+ connectivity matrices for multiple frequency bands.
207
+ Assessing the connectivity of two channels requires a
208
+ separate analysis of their respective phases. A signal’s
209
+ phase, Ο†(t), changes over time when being captured with
210
+ an electrode, and thus it must be measured for each
211
+ channel i, referring to the instantaneous phase, captured
212
+ using a Hilbert transform and computed via band-pass
213
+ filtered signals. Consequently, the phase value can be
214
+ pinpointed at each time point, allowing inter-channel
215
+ correlation and causality to be determined. Using this
216
+ method to track changes in the phase synchronization of
217
+ epileptic patients, Mormann et al. 26 showed that epileptic
218
+ episodes are often preceded by characteristic changes in
219
+ synchronization.
220
+ Following this, we can estimate the inter-channel
221
+ connectivity based on the cause-effect relationships. The
222
+ Granger causality test can hereby show whether one of
223
+ the factors is a time series, allowing the characteristics of
224
+ additional time series to be predicted. First employed in
225
+ the 1980s in the economics field, Granger causality is a
226
+ statistical hypothesis test that has been used to produce
227
+ good results in a wide range of other fields 10.
228
+ Neuroscience
229
+ research
230
+ has
231
+ applied
232
+ it
233
+ to
234
+ EEG
235
+ measurements, producing findings on brain activity in
236
+ emotion recognition 27, Vagus nerve stimulation 28, and
237
+ pain perception 29.
238
+ Connectivity based on causality implies cause-effect
239
+ relationships between various areas of the brain, but these
240
+ are not necessarily bidirectional. Thus, some brain areas
241
+ will be very active because they are influencing others,
242
+ and other areas may be very active because they are being
243
+ influenced by remote areas. Likewise, it could be the case
244
+ that high activity may be due to both situations. While
245
+ this concept of sources/sinks is not new, it has been
246
+ subject to a variety of different approaches. For example,
247
+ Rimehaug et al. 30 integrated it into their model of the
248
+ visual cortex’s local field potential, while Sotero et al. 31
249
+ used it to explain the laminar distribution of phase-
250
+ amplitude coupling of spontaneous current sources and
251
+ sinks in rat brains. However, neither of those studies
252
+ based their modeling of sources and sinks on causality
253
+ relationships, instead using the electrical activity in the
254
+ cerebral cortex.
255
+ The concepts of Granger causality and source/sink
256
+ relationships have been used to address the clinical issue
257
+ of surgical resection planning by capturing high-
258
+ frequency ictal and preictal oscillations on an intracranial
259
+ EEG 32, although no connectivity maps were constructed;
260
+ furthermore, the study did not use machine learning to
261
+ examine whether this approach could be applied in the
262
+ differential diagnosis of impairments.
263
+ Building on the work outlined above, we apply machine
264
+ learning classification algorithms to assess the potential
265
+ of diagnosing DD via a learner’s sources, sinks and total
266
+ activity under stimulus, identified using Granger
267
+ causality matrices. Due to the impenetrable nature of
268
+ EEG signal classification and the complexity of the
269
+ problem being addressed, machine learning is highly
270
+ suitable 33. Briefly, we seek to demonstrate that different
271
+ connectivity patterns are induced in certain brain
272
+ networks by low-level auditory processing. To this end,
273
+ we delineate this connectivity by establishing the source
274
+ and sink relationships through the application of Granger
275
+ causality to the phase synchronization among EEG
276
+ channels.
277
+ 3. Materials and methods
278
+ 3.1. Data acquisition
279
+ The dataset comprised EEG data from the University of
280
+ MΓ‘laga’s Leeduca Study Group 34, gathered from 48 age-
281
+ matched child participants (32 skilled readers and 16
282
+ dyslexic readers) (t(1) = -1.4, p > 0.05, age range: 88-100
283
+ months). All participants were righthanded native
284
+ Spanish speakers with normal or corrected-to-normal
285
+ vision; none had a hearing impairment. All participants
286
+ in the dyslexic group had been formally diagnosed with
287
+ dyslexia at school. All participants in the skilled reader
288
+ group were free from reading and writing difficulties and
289
+ had not been formally diagnosed with dyslexia. The
290
+ participants’
291
+ legal
292
+ guardians
293
+ expressed
294
+ their
295
+ understanding of the study, gave their written consent,
296
+ and were present throughout the experiment.
297
+ All participants experienced an auditory stimulus in 15-
298
+ minute sessions. The stimulus, which was modulated at
299
+
300
+ RodrΓ­guez-RodrΓ­guez et al.
301
+
302
+ 4
303
+ 4.8 Hz (prosodic-syllabic frequency) in 2.5-minute
304
+ segments, was band-limited white noise. This type of
305
+ stimulus was chosen to identify what synchronicity
306
+ patterns the low-level auditory processing would induce
307
+ and on the basis of the expert knowledge of linguistic
308
+ psychologists
309
+ concerning
310
+ the
311
+ main
312
+ frequency
313
+ components representing words in the human voice. The
314
+ participants’ EEG signals were recorded with a
315
+ BrainVision actiCHamp Plus with 32 active electrodes
316
+ (actiCAP, Brain Products GmbH, Germany) at a 500 Hz
317
+ sampling rate. The 10–20 standardized system was used
318
+ to place the 32 electrodes.
319
+ 3.2. Preprocessing
320
+ The preprocessing involved removing all eye-blinking
321
+ and movement/impedance variation artifacts from the
322
+ EEG signals. The former were eliminated via
323
+ independent component analysis (ICA) 35 based on the
324
+ eye movements observed in the EOG channel, while for
325
+ the latter the relevant EEG segments were excluded. The
326
+ channels were then referenced to the Cz channel.
327
+ Then, a band-pass filter was applied to the EEG channels
328
+ to collect information for the five EEG frequency bands
329
+ (Delta, 1.5–4 Hz; Theta, 4–8 Hz, Alpha, 8–13 Hz; Beta,
330
+ 13–30 Hz; and Gamma, 30–80 Hz). We used finite
331
+ impulse response (FIR) filters because these ensure a
332
+ constant phase lag that can later be corrected. To be
333
+ specific, each signal was sent forward and backward
334
+ through the two-way zero-phase lag band-pass FIR least-
335
+ squares filter, producing a zero-lag phase in the overall
336
+ filtering process that addressed the issue of phase lag 36.
337
+ As low-pass filtering with an 80 Hz threshold was
338
+ employed, we added a 50 Hz notch filter during
339
+ preprocessing to eliminate this frequency component.
340
+ 3.3. Hilbert Transform
341
+ A Hilbert transform (HT) transforms real signals into
342
+ analytic signals, i.e. complex-valued time series without
343
+ negative frequency components, allowing the time-
344
+ varying amplitude, phase and frequency, i.e., the
345
+ instantaneous amplitude, phase and frequency, to be
346
+ calculated from the analytic signal.
347
+
348
+ We define HT for a signal x(t) as:
349
+
350
+ β„‹[π‘₯(𝑑)] = 1
351
+ πœ‹ ∫
352
+ π‘₯(𝑑)
353
+ 𝑑 βˆ’ 𝜏 π‘‘πœ
354
+ +∞
355
+ βˆ’βˆž
356
+
357
+ (1)
358
+
359
+ and we obtain the analytic signal zi(t) for signal x(t) as:
360
+
361
+ 𝑧𝑖(𝑑) = π‘₯𝑖(𝑑) + 𝑗ℋ{π‘₯𝑖(𝑑)} = π‘Ž(𝑑)π‘’π‘—πœ™(𝑑)
362
+ (2)
363
+
364
+ From zi(t), computing the instantaneous amplitude is
365
+ straightforward:
366
+
367
+ π‘Ž(𝑑) = βˆšπ‘Ÿπ‘’(𝑧𝑖(𝑑))2 + π‘–π‘š(𝑧𝑖(𝑑))2
368
+ (3)
369
+
370
+ with the instantaneous, unwrapped phase as:
371
+
372
+ πœ™(𝑑) = π‘‘π‘Žπ‘›βˆ’1 π‘–π‘š(𝑧𝑖(𝑑))
373
+ π‘Ÿπ‘’(𝑧𝑖(𝑑))
374
+ (4)
375
+
376
+ The above technique gives the phase value for each time
377
+ point, allowing the inter-channel synchronization to be
378
+ estimated based on the phase variation.
379
+ 3.4. Granger Causality test
380
+ Developed for the field of econometrics by Clive
381
+ Granger,
382
+ Granger
383
+ causality
384
+ 37
385
+ describes
386
+ causal
387
+ interactions occurring between continuous-valued time
388
+ series. As a statistical hypothesis test, it essentially states
389
+ that β€œthe past and present may cause the future, but the
390
+ future cannot cause the past”; hence, knowing a cause
391
+ will be more helpful in predicting future effects than an
392
+ auto-regression will. Specifically, variable x will
393
+ Granger-cause y if the auto-regression for y that uses past
394
+ values of x and y is significantly more accurate than one
395
+ using only past values of y. We may exemplify this by
396
+ taking two stationary time-series sequences, xt and yt,
397
+ whereby xtβˆ’k and ytβˆ’k are, respectively, the past k values
398
+ of xt and yt. We then use two regressions to perform
399
+ Granger causality:
400
+
401
+ 𝑦𝑑̂ 1 = βˆ‘ π‘Žπ‘˜
402
+ 𝑙
403
+ π‘˜=1
404
+ π‘¦π‘‘βˆ’π‘˜ + πœ€π‘‘
405
+ (5)
406
+
407
+ 𝑦𝑑̂ 2 = βˆ‘ π‘Žπ‘˜
408
+ 𝑙
409
+ π‘˜=1
410
+ π‘¦π‘‘βˆ’π‘˜ + βˆ‘ π‘π‘˜
411
+ 𝑀
412
+ π‘˜=1
413
+ π‘₯π‘‘βˆ’π‘˜ + πœ‚π‘‘
414
+ (6)
415
+
416
+ where 𝑦𝑑̂ 1 and 𝑦𝑑̂ 2 are, respectively, the fitting values of
417
+ the first and second regressions; l and w are the maximum
418
+ numbers of the lagged observations of xt and yt; ak; bk ∈
419
+ R are the regression coefficient vectors estimated using
420
+ least squares; and Ξ΅t and Ξ·t are white noise (prediction
421
+ errors). Note that even though w can be infinite, due to
422
+ the finite nature of our data, we consider w finite and give
423
+ it a length well below the time series length, estimated
424
+ using model selection, such as the Akaike information
425
+ criterion (AIC) 38. Next, an F-test is applied to give a p-
426
+ value indicating whether the regression model produced
427
+
428
+
429
+ Neural source/sink phase connectivity in Developmental Dyslexia
430
+
431
+ 5
432
+ by Eq. (5) is statistically better than that of Eq. (6). If it
433
+ is, then x Granger-causes y.
434
+ We perform Granger causality testing for each
435
+ participant and evaluate the channels’ interactions,
436
+ producing an n x n square matrix of p-values (n = number
437
+ of channels).
438
+ Using Granger causality to analyze the neural network’s
439
+ directed functional connectivity intuitively demonstrates
440
+ the directionality with which information is transmitted
441
+ between neurons or brain regions. Previous studies have
442
+ already applied this technique to EEG analysis with great
443
+ success 39 40.
444
+ 3.5. Connectivity vectors
445
+ The field of neuroscience tends to consider the brain as a
446
+ network using functional information
447
+ 41
448
+ 42
449
+ 43,
450
+ culminating in the so-called connectome. This refers to
451
+ the complete mapping of all connections between brain
452
+ regions as an adjacency matrix, and often includes the
453
+ covariance, as well as other metrics, between fMRI
454
+ signals measured for different regions. Several studies
455
+ have also examined the temporal covariance between
456
+ EEG electrodes.
457
+ Once we had assembled the Granger causality matrices
458
+ for each participant subject, we established a threshold
459
+ value that evidenced a causal relationship between the
460
+ channels. Then, we formulated the three scenarios used
461
+ to produce each participant’s feature set:
462
+ β€’ Sources: Array of n x 1 elements; each element
463
+ relates each channel with the number of channels that
464
+ it influences.
465
+ β€’ Sinks: Array of n x 1 elements; each element relates
466
+ each channel with the number of channels that it is
467
+ influenced by.
468
+ β€’ Total activity: Array of n x 1 elements; the sum of
469
+ the two previous scenarios, acting as a reference for
470
+ each channel’s global activity.
471
+ By organizing the information thus, we receive the same
472
+ number of features as there are channels for each
473
+ participant, each with a number that indicates its activity
474
+ as a source, as a sink, or the total. A summary of this
475
+ process is presented in Figure 1.
476
+
477
+
478
+ Fig. 1. Assembling the source and sink connectivity arrays for
479
+ a participant, given the relevant Granger matrix. [P(k)] is an
480
+ Iverson bracket function.
481
+ 3.6. Ensemble feature selection
482
+ If the model includes many features, it will be more
483
+ complex, potentially leading to data overfitting.
484
+ Moreover, some of the features may be noise and could
485
+ adversely affect the model. Thus, we removed such
486
+ features to ensure the better generalization of the model.
487
+ We hereby selected the variables based on majority
488
+ voting through the application of several techniques. If a
489
+ variable was chosen by an algorithm, it received one
490
+ vote. The votes were then summed for each variable, and
491
+ those with the most votes were selected. (Fig. 2). This
492
+ method has been found to be suitable for datasets that are
493
+ high-dimensional yet have few instances 44. The voting
494
+ strategy used a variety of feature selection methods 45, as
495
+ outlined in the following:
496
+ Information value (IV) using weight of evidence
497
+ (WOE): This indicates the predictive power of an
498
+ independent variable concerning the dependent variable
499
+ 46. It allows a continuous independent variable to be
500
+ transformed into a set of groups or bins based on the
501
+ similarity of the dependent variable distribution (i.e.
502
+ numbers of events and non-events). Using WOE allows
503
+ outliers and missing values to be addressed and
504
+ eliminates the need for dummy variables 47:
505
+
506
+ π‘Šπ‘‚πΈ = ln (
507
+ 𝐸𝑣𝑒𝑛𝑑%
508
+ π‘π‘œπ‘› 𝐸𝑣𝑒𝑛𝑑%)
509
+ (7)
510
+
511
+ 𝐼𝑉 = Ξ£[(𝐸𝑣𝑒𝑛𝑑% βˆ’ π‘π‘œπ‘› 𝐸𝑣𝑒𝑛𝑑%) βˆ— π‘Šπ‘‚πΈ]
512
+ (8)
513
+
514
+ An IV statistic above 0.3 is held to indicate a strong
515
+ relationship between the predictor and the event/non-
516
+ event odds ratio 48.
517
+
518
+ FP1
519
+ FP2
520
+ F7
521
+ PO10
522
+ Sinks =
523
+ FP1
524
+ 1
525
+ pvalue<0.01
526
+ GrM[FP1,k],
527
+ [P(R)
528
+ FP2
529
+ 1
530
+ , GrM[F7,k].
531
+ F7
532
+ pvalue<0.01
533
+ 1
534
+ pvalue<0.01
535
+ P(a)
536
+ ...
537
+ PO10
538
+ pvalue<0.01
539
+ 1
540
+ GrM[P010, k]
541
+ P(k)
542
+ , GrM[k, FP2]
543
+ Sources =
544
+ GrM[k, FP1],
545
+ [P(k)
546
+ l0otherwiseRodrΓ­guez-RodrΓ­guez et al.
547
+
548
+ 6
549
+ Variable importance using random forest/extra trees
550
+ classifier: Calculated using a tree-based estimator, this
551
+ can be used to eliminate irrelevant features. Variable
552
+ importance is conventionally computed using the mean
553
+ decrease in impurity (i.e., gini importance
554
+ 49)
555
+ mechanism, wherein the improvement in the split
556
+ criterion for each split of each tree is the importance
557
+ measure assigned to the splitting variable. For each
558
+ variable, this is separately accumulated over all the trees
559
+ in the forest. This measure is similar to the R2 in the
560
+ training set regression.
561
+ Recursive Feature Elimination: This can be used to
562
+ select features by recursively considering feature sets
563
+ with diminishing size based on an external estimator (a
564
+ linear regression model) that assigns weights to the
565
+ features 50. The estimator is trained on the first feature
566
+ set, noting each feature’s importance based on a given
567
+ attribute. The least important features are subsequently
568
+ removed from the current set. The process is performed
569
+ recursively on the pruned set until the desired number of
570
+ features is achieved.
571
+ Chi-square best variables: This uses a chi-square (Ο‡2)
572
+ test to assess the correlations among a dataset’s features
573
+ and identify multicollinearity. The aim is revealing any
574
+ relationships between the dependent variable and any of
575
+ the independent variables 51. In the chi-square test, Hβ‚€
576
+ (null hypothesis) assumes that two features are
577
+ independent, while H₁ (alternative hypothesis) predicts
578
+ that they are related. We set a Ξ±=0.05 and a p-value of
579
+ 0.05 or greater is considered critical, anything less means
580
+ the deviations are significant hence the hypothesis must
581
+ be rejected.
582
+ L1-based feature selection: Some features can be
583
+ eliminated using a linear model with an L1 penalty. This
584
+ method involves regularization, wherein a penalty is
585
+ added to various parameters of a machine learning model
586
+ to reduce the model’s freedom and prevent overfitting.
587
+ When regularizing linear models, the penalty is applied
588
+ in addition to the coefficients multiplying the predictors
589
+ 52. Unlike other forms of regularization, L1 can reduce
590
+ some coefficients to zero, meaning the feature is
591
+ removed.
592
+ Once the best variables had been chosen by voting, we
593
+ performed a multicollinearity check on them.
594
+
595
+
596
+ Fig. 2. The feature selection procedure for the β€˜Sources’
597
+ scenario using a vote-based approach.
598
+ 3.7. Classification process
599
+ In an ensemble method, multiple models are first
600
+ generated and then integrated to produce higher-quality
601
+ results. The respective predictions are hereby combined
602
+ using weighted majority voting to make the final
603
+ prediction. At each boosting iteration, the data are
604
+ modified by applying w1, w2 , …, wn to each training
605
+ sample. As the weights are initially wi=1/N, a weak
606
+ learner is trained in the first step using the raw data. At
607
+ each successive iteration, the sample weights are
608
+ modified individually, and the algorithm is then applied
609
+ to the reweighted data. Training examples that are
610
+ incorrectly predicted relative to the previous step’s
611
+ boosted model are given increased weights; correctly
612
+ predicted examples are given decreased weights. As a
613
+ result, the examples that were difficult to predict become
614
+ increasingly influential as the number of iterations
615
+ increases, and the weak learners that follow are forced to
616
+ focus on the examples previously missed.
617
+ Ensemble methods deliver more accurate results than
618
+ single models, and are particularly suitable for improving
619
+ binary prediction on small data sets. We use the Gradient
620
+ Boosting classifier, as well as an Ada Boost for results
621
+ verification. This latter classifier 53 is a meta-estimator
622
+ that initially fits to the data, with further copies then being
623
+ fit to the same data, while incorrectly classified
624
+ instances’ weights are modified to force subsequent
625
+ classifiers to focus on them. The Gradient Boosting
626
+ classifier 54 creates an additive model based on a forward
627
+ stage-wise construction, allowing the optimization of the
628
+ arbitrary differentiable loss function. At each stage, n
629
+ regression trees are fit to the multinomial or deviance
630
+ binomial loss function’s negative gradient, with a single
631
+ regression tree being used for the special case of binary
632
+ classification. To identify the best parameter set, we
633
+ cross-validate with 20 folds and a parameter grid, as
634
+ shown in Table 1.
635
+
636
+ Method:IVusingWOE-β†’
637
+ Feature subset: f1, f2, f3,.., fn
638
+ All features for Scenario Sources:
639
+ Voting
640
+ FP1. FP2,F7... PO10
641
+ Method:Var.Imp.using RF→
642
+ Feature subset:f'1, f'2,f3, ...,f'n
643
+ M
644
+ Method:Var.Imp.using Trees β†’
645
+ Feature subset: f"1, f"2, f"3,., f"n
646
+ Method:RFE→Feature subset
647
+ Reducedranked feature
648
+ subsetbasedonvotes
649
+ β†’fr1,f2,f3,.,frn
650
+ Method:Chi Squared→Feature subset
651
+ Neural source/sink phase connectivity in Developmental Dyslexia
652
+
653
+ 7
654
+ Table 1. Parameter grid of machine learning classifiers.
655
+ Algorithm
656
+ Parameter
657
+ Range
658
+ Gradient
659
+ n_estimators
660
+ 1 to 12
661
+ Boosting
662
+ Loss
663
+ deviance, exponential
664
+
665
+ Learning rate
666
+ 0.05 to 1.5
667
+
668
+ Criterion
669
+ friedm_mse, sq_error, mse, mae
670
+
671
+ Min_samples_split
672
+ 0.01 to 3
673
+
674
+ Min_samples_leaf
675
+ 0.01 to 3
676
+
677
+
678
+ Max_depth
679
+ 1 to 4
680
+ Ada Boost
681
+ n_estimators
682
+ 1 to 25
683
+
684
+ Learning rate
685
+ 1 to 3.5
686
+
687
+ Boosting algorithm
688
+ SAMME, SAMME.R
689
+ 4. Results
690
+ Plotting each learner’s array of sources and sinks permits
691
+ the visual extraction of the respective patterns of the
692
+ dyslexic and control groups. To this end, we examined
693
+ the channel distributions for both groups by calculating
694
+ the means and dispersions and producing a box-and-
695
+ whisker plot. We also constructed a topoplot as this can
696
+ illustrate the results with greater clarity. For example,
697
+ Fig. 3 shows the Theta band connectivity of the control
698
+ and dyslexic groups specifically for total activity. Please
699
+ note that Fig. 3 and Fig. 4 do not directly represent the
700
+ electrical activity of the cerebral cortex, but rather show
701
+ the levels of the cause-effect relationships between the
702
+ channels, i.e. in one direction or in the other direction or
703
+ in total. It immediately becomes clear that despite the
704
+ similarity of the patterns, the dyslexic group has a
705
+ significantly higher activity level in the Theta band.
706
+
707
+
708
+
709
+
710
+
711
+ Fig. 3. Boxplot of the total activity in the Alpha band.
712
+ Fig 4. The equivalent graphical representation of Fig. 3 in a
713
+ topoplot.
714
+
715
+
716
+
717
+ Fig. 5. Source/sink activity in the Theta, Beta and Gamma bands in the control and dyslexic groups. Numbers represent how many
718
+ channels are affected by each channel as a source, or how many channels are affecting each channel as a sink.
719
+
720
+ ActivityofsourcesinThetaband
721
+ Control
722
+ 22.5
723
+ Dyslexic
724
+ 20.0
725
+ nels
726
+ 7.5
727
+ 5.0
728
+ 2.5 -
729
+ Fp1 Fp2 F7 F3 FZ F4 F8 FC5 FC1 FC2FC6 T7 C3 C4 T8 TP9CP5 CP1CP2CP6IP1O P7 P3 PZ P4 P8 PO9 O1 OZ O2PO10
730
+ ChannelActivityofsources inThetaband
731
+ Control group
732
+ Dyslexic group
733
+ 19
734
+ 19
735
+ Fp1
736
+ Fp2
737
+ F1
738
+ 2
739
+ F8
740
+ IFG
741
+ 13
742
+ FC6
743
+ FQ1
744
+ FG2
745
+ FG6
746
+ .
747
+ 13
748
+ TZ
749
+ T8
750
+ G3
751
+ 18
752
+ TP9CR5
753
+ CP1
754
+ CPEFP.i
755
+ CP5
756
+ CBI
757
+ Cp2
758
+ CP6
759
+ P3
760
+ RZ
761
+ P4
762
+ P8
763
+ P7
764
+ PZ
765
+ F4
766
+ P8
767
+ kod
768
+ 01
769
+ 02
770
+ Pig
771
+ 60
772
+ 01
773
+ 02
774
+ Poip
775
+ QzActivity of sources in Theta band
776
+ 19
777
+ Control group
778
+ Dyslexlc group
779
+ 19
780
+ 13
781
+ 13Activity of sources in Beta band
782
+ 19
783
+ Control group
784
+ Dyslexlc group
785
+ 19
786
+ F8
787
+ 13
788
+ 13Activityof sources in Gamma band
789
+ 19
790
+ Control group
791
+ Dyslexlc group
792
+ 19
793
+ 13
794
+ 13Activity of sinks in Theta band
795
+ 19
796
+ Control group
797
+ Dyslexic group
798
+ 19
799
+ Fp2
800
+ 13
801
+ 13Activity of sinks in Beta band
802
+ 19
803
+ Control group
804
+ Dyslexic group
805
+ 19
806
+ 13
807
+ 6
808
+ 13Activity of sinks in Gamma band
809
+ 19
810
+ Controlgroup
811
+ Dyslexicgroup
812
+ 19
813
+ Fp.
814
+ F4
815
+ 13
816
+ T651
817
+ FE6
818
+ 13
819
+ 4
820
+ 9P5
821
+ EPFTotal activity per channel in Theta band
822
+ Control group
823
+ Dyslexic group
824
+ 38
825
+ 38
826
+ 27
827
+ 27
828
+ 15
829
+ 15Total activity per channel in Beta band
830
+ Control group
831
+ Dyslexic group
832
+ 38
833
+ 38
834
+ 27
835
+ 27
836
+ CP
837
+ CPI
838
+ 15
839
+ TE
840
+ 15Total activity per channel in Gamma band
841
+ Control group
842
+ Dyslexic group
843
+ 38
844
+ 38
845
+ H
846
+ 27
847
+ Fe6
848
+ 27
849
+ 4
850
+ 15
851
+ 15RodrΓ­guez-RodrΓ­guez et al.
852
+
853
+ 8
854
+ Fig. 5 compares the channel activity in the Theta, Beta
855
+ and Gamma bands, and can be viewed separately as
856
+ sources, sinks, or total activity for both the control and
857
+ dyslexic groups. Please note that the range of
858
+ visualization is the same in all sinks/sources topoplots,
859
+ while different in the total activity ones, for better
860
+ representation. Once more, it is immediately clear that
861
+ while the patterns are broadly similar, the activity level is
862
+ higher in the dyslexic group, primarily observed in the
863
+ sink activity (less in the source activity). Thus, although
864
+ the sources, broadly speaking, behave similarly between
865
+ the groups, the dyslexic group has significantly more
866
+ concentrated sinks and more activity. Consequently, the
867
+ overall activity level is also affected.
868
+
869
+
870
+ Fig. 6. Feature importance in Theta, Beta and Gamma bands considering sources, sinks and total activity.
871
+ With as many arrays as subjects, and with each array
872
+ having as many components as channels, we performed
873
+ feature selection to identify channels that can help
874
+ differentiate between the control and dyslexic groups.
875
+ The feature selection procedure outlined above was thus
876
+ applied for the cases of sources, sinks and total activity,
877
+ according to the band. Fig. 6 presents the results for the
878
+ Theta, Beta and Gamma bands, whereby the importance
879
+ values are normalized to permit fair and simple
880
+ comparison. Channels showing a higher significance are
881
+ those with more dissimilarity between the control and
882
+
883
+ Feature importance in Theta band
884
+ 1.0
885
+ Activity of sources
886
+ Activity of sinks
887
+ Total activity per channel
888
+ 0.8
889
+ 0.2 -
890
+ 0.0-
891
+ Fp1
892
+ Fp2
893
+ F7
894
+ F3
895
+ Fz
896
+ F4
897
+ F8
898
+ FC5
899
+ FC1
900
+ FC2
901
+ FC6
902
+ T7
903
+ C3
904
+ C4
905
+ T8
906
+ CP5
907
+ CP1
908
+ CP2
909
+ CP6 TP10
910
+ P3
911
+ Pz
912
+ P4
913
+ P8
914
+ PO9
915
+ Q1
916
+ Qz
917
+ ZO
918
+ PO10
919
+ ChannelFeature importance in Beta band
920
+ 1.0
921
+ Activity of sources
922
+ I Activity of sinks
923
+ Total activity per channel
924
+ 0.8-
925
+ Fea
926
+ 0.2
927
+ 0.0-
928
+ Fp1
929
+ Fp2
930
+ F7
931
+ F3
932
+ Fz
933
+ F4
934
+ F8
935
+ FC5
936
+ FC1
937
+ FC2
938
+ FC6
939
+ T7
940
+ C3
941
+ C4
942
+ T8
943
+ CP5
944
+ CP1
945
+ CP2
946
+ CP6TP10
947
+ P7
948
+ P3
949
+ Zd
950
+ P4
951
+ P8
952
+ PO9
953
+ Q1
954
+ Qz
955
+ 02
956
+ PO10
957
+ ChannelFeatureimportanceinGammaband
958
+ 1.0
959
+ Activity of sources
960
+ Activity of sinks
961
+ Total activity per channel
962
+ 0.8
963
+ 0.2
964
+ 0.0-
965
+ Fp1
966
+ Fp2
967
+ F7
968
+ F3
969
+ Fz
970
+ F4
971
+ F8
972
+ FC5
973
+ FC1
974
+ FC2
975
+ FC6
976
+ C3
977
+ C4
978
+ T8
979
+ CP5
980
+ CP1
981
+ CP2
982
+ CP6TP10
983
+ 3
984
+ P4
985
+ P8
986
+ PO9
987
+ Q1
988
+ Qz
989
+ 02PO10
990
+ Channel
991
+ Neural source/sink phase connectivity in Developmental Dyslexia
992
+
993
+ 9
994
+ dyslexic groups, directing us to where we can find
995
+ different patterns of functioning.
996
+ After performing the feature selection for each band, for
997
+ each case (sources, sinks and total activity), we optimize
998
+ the Gradient Boosting classifier to obtain the best
999
+ performance. The results are summarized in Table 2, with
1000
+ performances achieving at least 80% marked bold.
1001
+ According to the results, the greatest differences between
1002
+ the control and dyslexic groups (i.e., the best classifier
1003
+ results) emerge in the Theta and Gamma bands when
1004
+ accounting for the activity sink role of the different
1005
+ channels, achieving accuracies of 84% and 88%,
1006
+ respectively. We also wish to highlight the results for the
1007
+ Beta band for the activity sources regarding the Area
1008
+ Under the Curve (AUC), in addition to accuracy.
1009
+ Table 2. Results of the Gradient Boosting machine
1010
+ learning classifier.
1011
+ Band
1012
+ Features set
1013
+ Accuracy
1014
+ AUC
1015
+ Delta
1016
+ Sources
1017
+ 0.77 Β± 0.14
1018
+ 0.65 Β± 0.31
1019
+
1020
+ Sinks
1021
+ 0.79 Β± 0.20
1022
+ 0.70 Β± 0.29
1023
+
1024
+ Total activity
1025
+ 0.74 Β± 0.19
1026
+ 0.76 Β± 0.25
1027
+ Theta
1028
+ Sources
1029
+ 0.77 Β± 0.17
1030
+ 0.77 Β± 0.30
1031
+
1032
+ Sinks
1033
+ 0.84 Β± 0.15
1034
+ 0.87 Β± 0.18
1035
+
1036
+ Total activity
1037
+ 0.74 Β± 0.17
1038
+ 0.72 Β± 0.28
1039
+ Alpha
1040
+ Sources
1041
+ 0.79 Β± 0.19
1042
+ 0.74 Β± 0.25
1043
+
1044
+ Sinks
1045
+ 0.76 Β± 0.21
1046
+ 0.71 Β± 0.29
1047
+
1048
+ Total activity
1049
+ 0.79 Β± 0.17
1050
+ 0.77 Β± 0.21
1051
+ Beta
1052
+ Sources
1053
+ 0.80 Β± 0.17
1054
+ 0.86 Β± 0.18
1055
+
1056
+ Sinks
1057
+ 0.79 Β± 0.24
1058
+ 0.81 Β± 0.27
1059
+
1060
+ Total activity
1061
+ 0.76 Β± 0.23
1062
+ 0.75 Β± 0.32
1063
+ Gamma
1064
+ Sources
1065
+ 0.81 Β± 0.18
1066
+ 0.83 Β± 0.22
1067
+
1068
+ Sinks
1069
+ 0.88 Β± 0.14
1070
+ 0.93 Β± 0.16
1071
+
1072
+ Total activity
1073
+ 0.82 Β± 0.12
1074
+ 0.87 Β± 0.18
1075
+
1076
+ The Receiver Operating Curve (ROC) space is a valuable
1077
+ data interpretation tool that can be used to assess the
1078
+ performance of a binary classifier, wherein it indicates
1079
+ the cutoff point at which sensitivity is traded for
1080
+ specificity. Hence, it can be used to evaluate the
1081
+ classifier’s performance in distinguishing positive and
1082
+ negative samples. Related to this, AUC is the probability
1083
+ that the classifier will assign a random positive instance
1084
+ a more extreme value than a random negative instance.
1085
+ Fig. 7 presents the ROC curves for the Theta, Beta and
1086
+ Gamma bands, to identify those with the best
1087
+ performance. Notably, the Gamma band with the
1088
+ channels’ sinks activity as the features presents a 93%
1089
+ under the curve.
1090
+ The obtained results were verified by repeating the
1091
+ classification process using the Ada Boost algorithm.
1092
+ Table 3 presents the results for the Gamma band while
1093
+ Fig. 8 shows the ROC curve. While the performance is
1094
+ slightly diminished, it remains consistent across all bands
1095
+ and cases (sources, sinks and total activity) with the
1096
+ results from the Gradient Boosting.
1097
+
1098
+
1099
+
1100
+ Fig. 7. ROC curves for the Theta, Beta and Gamma bands with
1101
+ the Gradient Boosting classifier.
1102
+
1103
+
1104
+ Thetaband
1105
+ 1.0
1106
+ : Rate (Positive label:
1107
+ 0.8
1108
+ 0.6
1109
+ Positive
1110
+ 0.4
1111
+ True
1112
+ 0.2
1113
+ Chance
1114
+ SourcesMeanROC(AUC =0.77Β±0.31)
1115
+ Sinks Mean ROC (AUC = 0.87Β±0.18)
1116
+ 0.0
1117
+ Total activityMeanROC (AUC = 0.72 Β± 0.28)
1118
+ 0.0
1119
+ 0.2
1120
+ 0.4
1121
+ 0.6
1122
+ 0.8
1123
+ 1.0
1124
+ False Positive Rate (Positive label: 1)Betaband
1125
+ 1.0
1126
+ : Rate (Positive label:
1127
+ 0.8
1128
+ 0.6
1129
+ Positive
1130
+ 0.4
1131
+ True
1132
+ 0.2
1133
+ Chance
1134
+ SourcesMeanROC(AUC=0.86Β±0.18)
1135
+ Sinks Mean ROC (AUC = 0.81 Β± 0.27)
1136
+ 0.0
1137
+ Total activityMeanROC (AUC = 0.75 Β± 0.33)
1138
+ 0.0
1139
+ 0.2
1140
+ 0.4
1141
+ 0.6
1142
+ 0.8
1143
+ 1.0
1144
+ False Positive Rate (Positive label:1)Gammaband
1145
+ 1.0
1146
+ Rate (Positive label: 1)
1147
+ 0.8
1148
+ 0.6
1149
+ Positive
1150
+ 0.4
1151
+ True
1152
+ 0.2
1153
+ Chance
1154
+ SourcesMeanROC(AUC=0.83Β±0.23)
1155
+ SinksMeanROC(AUC=0.93Β± 0.17)
1156
+ 0.0
1157
+ Total activityMean ROC (AUC = 0.87 Β± 0.18)
1158
+ 0.0
1159
+ 0.2
1160
+ 0.4
1161
+ 0.6
1162
+ 0.8
1163
+ 1.0
1164
+ False Positive Rate (Positive label: 1)RodrΓ­guez-RodrΓ­guez et al.
1165
+
1166
+ 10
1167
+ Table 3. Results for the Ada Boost classifier for
1168
+ the Gamma band.
1169
+ Band
1170
+ Feature set
1171
+ Accuracy
1172
+ AUC
1173
+ Gamma
1174
+ Sources
1175
+ 0.83 Β± 0.17
1176
+ 0.82 Β± 0.27
1177
+
1178
+ Sinks
1179
+ 0.88 Β± 0.11
1180
+ 0.86 Β± 0.21
1181
+
1182
+ Total activity
1183
+ 0.77 Β± 0.19
1184
+ 0.76 Β± 0.31
1185
+
1186
+
1187
+ Fig. 8. ROC curves for the Gamma band with the Ada Boost
1188
+ classifier.
1189
+ As is often the case in biomedical studies, statistical tests
1190
+ are required to check that the number of samples has not
1191
+ introduced bias in the classification stage (e.g., through
1192
+ overfitting). Moreover, there is a need to check the
1193
+ probability of these results having been obtained by
1194
+ chance. For large datasets, such tests need not be as
1195
+ stringent, but real-world studies demand special attention
1196
+ due to the small sample sizes and unbalanced classes.
1197
+ Specifically, in experimental studies the prevalence of
1198
+ the disorder among the population being treated must be
1199
+ taken into account. For DD, this is around 5-12%, as
1200
+ mentioned above.
1201
+ To this end, a null distribution is generated by estimating
1202
+ the classifier’s accuracy for 1000 permutations of the
1203
+ labels. This indicates the distribution for the null
1204
+ hypothesis that the features are not dependent on the
1205
+ labels, and enables the estimation of the probability that
1206
+ the classification results will be reproduced with shuffled
1207
+ labels. The result is an empirical p-value determined by:
1208
+
1209
+ 𝑝 βˆ’ π‘£π‘Žπ‘™π‘’π‘’ = #π‘π‘’π‘Ÿπ‘š π‘€π‘–π‘‘β„Ž π‘Žπ‘π‘. β„Žπ‘–π‘”β„Žπ‘’π‘Ÿ π‘‘β„Žπ‘Žπ‘› π‘π‘Žπ‘ π‘’π‘™π‘–π‘›π‘’
1210
+ #π‘π‘’π‘šπ‘π‘’π‘Ÿ π‘œπ‘“ π‘π‘’π‘Ÿπ‘šπ‘’π‘‘π‘Žπ‘‘π‘–π‘œπ‘›π‘ 
1211
+
1212
+ (9)
1213
+
1214
+ Fig. 9 gives the permutation test results for the Theta,
1215
+ Beta, and Gamma bands for sources, sinks and total
1216
+ activity.
1217
+ The
1218
+ null
1219
+ distribution
1220
+ from
1221
+ the
1222
+ label
1223
+ permutations, as outlined above, is in blue, while the
1224
+ vertical red line represents the accuracy obtained for the
1225
+ non-permuted case. At each permutation iteration, a 20-
1226
+ fold stratified cross-validation is performed, and based on
1227
+ the average of the results obtained at these 20 folds, the
1228
+ corresponding permutation iteration is determined.
1229
+ Hence, Fig. 9 presents the classification’s probability
1230
+ density. According to the permutation tests, the results
1231
+ have low p-values and are significant.
1232
+ 5. Discussion
1233
+ The participants were subjected to white noise at 4.8 Hz,
1234
+ i.e. between the syllabic and prosodic frequencies, as the
1235
+ sole stimulus. DD has been shown to link to impairments
1236
+ in syllabic and prosodic perception 55, suggesting general
1237
+ difficulties in identifying the different modulation
1238
+ frequencies. This influences the slower temporal rates of
1239
+ speech processing in particular, as well as the tracking of
1240
+ the amplitude envelope of speech, diminishing learners’
1241
+ syllabic segmentation efficiency.
1242
+ Multi-time resolution models of speech processing 16
1243
+ have evidenced that phonetic segment identification
1244
+ associates with faster temporal modulations (Gamma
1245
+ rate, 30–80 Hz), syllable identification is linked to slower
1246
+ modulations (Theta rate, 4–10 Hz), and syllable stress
1247
+ and prosodic patterning information correlates with very
1248
+ slow modulations (Delta rate, 1.5–4 Hz). Nonetheless,
1249
+ anomalies can emerge in various frequency ranges due to
1250
+ inter-band entrainment.
1251
+ As it offers adequate time resolution, examining the
1252
+ patterns occurring in EEG channels at different bands can
1253
+ unveil the speech encoding linked to problems with
1254
+ speech prosody and sensorimotor synchronization.
1255
+ Exemplifying this, previous research 18 used speech-
1256
+ based stimuli and time-frequency descriptors to reveal
1257
+ the link between speech features and neural dynamics.
1258
+ We find that the classifier performs better in the Theta
1259
+ and Gamma bands. The results for the Theta band are
1260
+ expected as the TSF suggests that the phonological
1261
+ deficit of DD – regardless of language – may be partially
1262
+ attributed to functionally atypical or impaired phonology
1263
+ entrainment mechanisms in the auditory cortex,
1264
+ especially as oscillations at slower temporal rates, i.e.
1265
+ Theta and Delta, relate to syllabic and prosodic
1266
+ processing 56.
1267
+
1268
+
1269
+
1270
+ Gammaband
1271
+ 1.0
1272
+ : Rate (Positive label:
1273
+ 0.8
1274
+ 0.6
1275
+ Positive
1276
+ 0.4
1277
+ True
1278
+ 0.2
1279
+ Chance
1280
+ SourcesMeanROC(AUC=0.78Β±0.27)
1281
+ Sinks MeanROC (AUC=0.86Β±0.21)
1282
+ 0.0
1283
+ Total activity Mean ROC(AUC=0.76Β± 0.31)
1284
+ 0.0
1285
+ 0.2
1286
+ 0.4
1287
+ 0.6
1288
+ 0.8
1289
+ 1.0
1290
+ False Positive Rate (Positive label:1)
1291
+ Neural source/sink phase connectivity in Developmental Dyslexia
1292
+
1293
+ 11
1294
+
1295
+
1296
+
1297
+ Fig. 9. Permutation tests for Gradient Boosting classifier in Theta, Beta and Gamma bands.
1298
+ As per the TSF, group differences are expected in
1299
+ neuronal oscillatory entrainment at slower rates (approx.
1300
+ 4 Hz, in line with the stimulus used) 57. Higher causality
1301
+ relationships emerged in the frontal area in all scenarios
1302
+ for the Theta band. In addition, the number of channels
1303
+ that g-causes causality is higher in the dyslexic domain,
1304
+ which was the case for the sources, sinks and total
1305
+ activity. This higher activity in terms of overall causality
1306
+ relations was evident across all bands. However, in the
1307
+ participants with DD there was significantly less
1308
+ entrainment in the auditory networks of the right
1309
+ hemisphere in the Theta band. As Fig. 6 (feature
1310
+ selection) shows, the C4 channel in the upper part, i.e. the
1311
+ Theta band, is predominantly influential for the causality
1312
+ regarding the sources, as well as the sinks and total
1313
+ activity. It has already been established that the right-
1314
+ lateralized Theta sampling network tends to involve
1315
+ slower temporal rates and codes the speech signal’s lower
1316
+ modulation frequencies 57, facilitating syllable-scale
1317
+ temporal integration. In other words, spoken sentences
1318
+ are tracked and distinguished by the Theta band phase
1319
+ pattern, allowing the incoming speech signal to be broken
1320
+ into syllable-sized packets and speech dynamics to be
1321
+ tracked through resetting and sliding, such as with
1322
+ varying rates of speech 58. Fig. 5 (topoplots) clearly
1323
+ demonstrates that the C4 channel is the most interesting
1324
+ as it has the most Granger causality (causing and being
1325
+ caused) for all scenarios for the dyslexic group. For the
1326
+ sources, the frontal area contains other noteworthy
1327
+ channels (FP2, F7, F3 and Fz) that show differences
1328
+ between the control and dyslexic groups in terms of
1329
+ activity. The most influential channels in the sinks are F3
1330
+ and F4 (frontal area) and P3.
1331
+ Hence, it seems pertinent to suggest that the main
1332
+ differences in the causality relationships of the Theta
1333
+ band lie in the so-called dorsal and ventral pathways. In
1334
+ particular, the right area seems critical, as evidenced in
1335
+
1336
+ Thetaband-Sources
1337
+ 8
1338
+ 7
1339
+ Score on original data: o.77
1340
+ 6
1341
+ (p-value:0.002)
1342
+ 5
1343
+ I
1344
+ -
1345
+ 4 -
1346
+ -
1347
+ FE
1348
+ 2
1349
+ 1
1350
+ 0-
1351
+ 0.40
1352
+ 0.45
1353
+ 0.50
1354
+ 0.55
1355
+ 0.60
1356
+ 0.65
1357
+ 0.70
1358
+ 0.75
1359
+ 0.80
1360
+ Accuracy scoreTheta band - Sinks
1361
+ -
1362
+ 8
1363
+ score
1364
+ on original data: 0.84
1365
+ 6 -
1366
+ (p-value:0.001)
1367
+ *.
1368
+ -
1369
+ -
1370
+ -
1371
+ 0
1372
+ 0.45
1373
+ 0.50
1374
+ 0.55
1375
+ 0.60
1376
+ 0.65
1377
+ 0.70
1378
+ 0.75
1379
+ 0.80
1380
+ 0.85
1381
+ Accuracy scoreTheta band -Total activity
1382
+ -
1383
+ 8 -
1384
+ -
1385
+ Score on original data: o.74
1386
+ (p-value:0.001)
1387
+ 6 -
1388
+ -
1389
+ robability
1390
+ 4
1391
+ -
1392
+ 2 -
1393
+ 0
1394
+ 0.50
1395
+ 0.55
1396
+ 0.60
1397
+ 0.65
1398
+ 0.70
1399
+ 0.75
1400
+ Accuracy scoreBeta band-Sources
1401
+ 10 -
1402
+ 8
1403
+ Score onoriginal data: o.80
1404
+ 6
1405
+ (p-value:0.001)
1406
+ 4
1407
+ 2 -
1408
+ 0
1409
+ 0.45
1410
+ 0.50
1411
+ 0.55
1412
+ 0.60
1413
+ 0.65
1414
+ 0.70
1415
+ 0.75
1416
+ 0.80
1417
+ Accuracy scoreBeta band -Sinks
1418
+ 8 -
1419
+ !!
1420
+ 7 -
1421
+ Score on original data: o.79
1422
+ 6 -
1423
+ (p-value:0.001)
1424
+ .
1425
+ 4.
1426
+ E
1427
+ 2
1428
+ 1
1429
+ 0
1430
+ 0.45
1431
+ 0.50
1432
+ 0.55
1433
+ 0.60
1434
+ 0.65
1435
+ 0.70
1436
+ 0.75
1437
+ 0.80
1438
+ Accuracy scoreBeta band -Total activity
1439
+ 7
1440
+ Score on original data: o.76
1441
+ 6
1442
+ D-
1443
+ value:0.002
1444
+ 5
1445
+ 3 -
1446
+ 2 -
1447
+ 1
1448
+ 0
1449
+ 0.45
1450
+ 0.50
1451
+ 0.55
1452
+ 0.60
1453
+ 0.65
1454
+ 0.70
1455
+ 0.75
1456
+ Accuracy scoreGammaband-Sources
1457
+ 10
1458
+ 8
1459
+ Score on original data: 0.81
1460
+ (p-value: 0.001)
1461
+ 4
1462
+ 2 .
1463
+ 0
1464
+ 0.50
1465
+ 0.55
1466
+ 0.60
1467
+ 0.65
1468
+ 0.70
1469
+ 0.75
1470
+ 0.80
1471
+ Accuracy scoreGammaband-Sinks
1472
+ 8
1473
+ Score 0n original data: 0.88
1474
+ 6
1475
+ (p-value:0.001)
1476
+ 2
1477
+ 0
1478
+ 0.45
1479
+ 0.50
1480
+ 0.55
1481
+ 0.60
1482
+ 0.65
1483
+ 0.70
1484
+ 00.75
1485
+ 0.80
1486
+ 0.85
1487
+ Accuracy scoreGammaband-Totalactivity
1488
+ -8
1489
+ 7
1490
+ Scoreor
1491
+ n original data: 0.82
1492
+ 6 Β·
1493
+ (p-value:0.001)
1494
+ 3
1495
+ 2
1496
+ 1 -
1497
+ 0
1498
+ 0.50
1499
+ 0.55
1500
+ 0.60
1501
+ 0.65
1502
+ 0.70
1503
+ 0.75
1504
+ 0.80
1505
+ Accuracy scoreRodrΓ­guez-RodrΓ­guez et al.
1506
+
1507
+ 12
1508
+ the prior research and especially demonstrated here with
1509
+ the sinks scenario.
1510
+ Another interesting result worth discussing is that for the
1511
+ Beta band. Here, more activity was observed for all three
1512
+ scenarios in the dyslexic group; this agrees with the
1513
+ results for the Theta band as well as those from previous
1514
+ studies 59. For the sources, differences in the causal
1515
+ relationships were mainly identified in the C3 and C4
1516
+ channels, pointing to areas responsible for motor
1517
+ processing 11. It is becoming increasingly clear that
1518
+ speech perception is at least partially located in the motor
1519
+ areas, especially under less-than-optimal listening
1520
+ conditions. This cruciality of the C4 channel was
1521
+ similarly seen in the Theta band and is in line with prior
1522
+ research evidencing the important role played by the
1523
+ lower frequency bands in general and Beta band coupling
1524
+ in particular 60. Hence, inefficient phase locking in the
1525
+ auditory cortex may affect visual and motor processing
1526
+ development, which may in turn cause some of the
1527
+ visual, motor and attentional difficulties seen in DD 61.
1528
+ It should be noted, however, that the C3-C4 interaction is
1529
+ mostly relevant for the sources and is not important for
1530
+ either the C3 for the sinks or, as a result, for the total
1531
+ activity. Meanwhile, the causal activity in the Beta band
1532
+ is different in the occipital area in the sinks scenario, and
1533
+ it is remarkably different in the frontal area, especially in
1534
+ FP1 for all three scenarios and in the F3 channel for the
1535
+ sinks scenario.
1536
+ In the Gamma band the activity is higher than in the Theta
1537
+ band for maximum values, although the occipital area
1538
+ shows more concentrated activity among the causality
1539
+ relations, as Fig. 5 shows. Nevertheless, the effect is
1540
+ different between the control and dyslexic groups,
1541
+ whereby the participants with DD show higher activity
1542
+ for the sinks, which increases their total activity.
1543
+ For the sources, the channels with the most explicit
1544
+ differences are FC1 and, more generally, TP9 in the left
1545
+ temporal area. In the case of sinks, this is also an
1546
+ important channel, although O1 and, as highlighted
1547
+ above, C3 also play a role.
1548
+ Meanwhile,
1549
+ in
1550
+ the
1551
+ Gamma
1552
+ band,
1553
+ despite
1554
+ the
1555
+ discrepancies between the dorsal and ventral pathways,
1556
+ the latter offers the main difference for the classification
1557
+ of TP9 for both sources and sinks. FC1 is linked to
1558
+ sources and C3 to sinks, suggesting a significant cause-
1559
+ effect relationship, albeit with potentially less activity in
1560
+ the dyslexic group, facilitating classification.
1561
+ We can confirm that the classifier performs better in the
1562
+ Theta and Gamma bands, which can evidence atypical
1563
+ oscillatory differences based on both speech and non-
1564
+ speech stimuli 56. According to Leong’s models 62, the
1565
+ slower rates (Delta and Theta) temporally constrain
1566
+ entrainment at the faster rates, such as Gamma.
1567
+ Lehongre et al. 65 contended that the oscillatory nesting
1568
+ seen between the Theta/Delta phase and the Gamma
1569
+ power 63 64 offers a way to integrate information at the
1570
+ phonemic (Gamma) rate into the syllabic rate.
1571
+ Meanwhile, the integration of the various acoustic
1572
+ features that contribute to the same phoneme being
1573
+ perceived may be hindered by impairments in the phase
1574
+ locking by Theta generators. Otherwise, flaws in certain
1575
+ Theta mechanisms could influence the development of
1576
+ the phonological system, which thus tends to code
1577
+ information bilaterally with the Gamma oscillations
1578
+ independently and then link them perceptually with the
1579
+ Theta oscillator output. In this case, the impaired phase
1580
+ locking of the right hemisphere Theta oscillatory
1581
+ networks causes difficulties with lower frequency
1582
+ modulations 17 66.
1583
+ In addition, the spontaneous oscillatory neural activity
1584
+ identified in the auditory cortex in both the Theta and
1585
+ Gamma bands is known to associate with spontaneous
1586
+ activity in the visual and premotor areas 66.
1587
+ A bilateral Gamma sampling network codes the signal’s
1588
+ higher frequency modulations, thereby facilitating
1589
+ temporal integration at the phonetic (i.e., phoneme) scale.
1590
+ If we apply this model to DD, it is indicated that impaired
1591
+ processing at the syllable level (i.e., less efficient Theta
1592
+ phase locking) occurs alongside unimpaired Gamma
1593
+ sampling, meaning more weight is assigned to phonetic
1594
+ feature information during phonological development.
1595
+ Hence, as is the case in typical infant development,
1596
+ children with DD may have sensitivity to all phonetic
1597
+ contrasts of human languages 67.
1598
+ Leong and Goswami 62 found that learners with DD show
1599
+ a preference for different phase alignment between
1600
+ amplitude modulations (AMs) when these respectively
1601
+ convey syllable and phoneme information (Theta and
1602
+ Gamma-AMs). A different phase locking angle suggests
1603
+ a discrepancy in the integration of speech information
1604
+ that arrives at a temporal rate different to that of the final
1605
+ perception of the speech 14. Our results concerning the
1606
+ interaction between the Theta and Gamma bands support
1607
+ this.
1608
+ Finally, our results also seem to confirm that the dyslexic
1609
+ brain is less efficient at encoding the amplitude
1610
+ modulation hierarchy’s highest levels, i.e. those bearing
1611
+ information on the prosodic-syllabic structure, leading to
1612
+ cascade effects that impact the encoding of the
1613
+ phonological structure’s levels nested within the Delta
1614
+ band, such as the syllable-level (Theta band) and
1615
+ phoneme-level (Gamma band) AM information.
1616
+ Importantly, our results have been validated using a
1617
+ demanding permutation test, with the aim of ensuring
1618
+ that the results are not coincidental, despite the medium
1619
+ sample size.
1620
+
1621
+
1622
+ Neural source/sink phase connectivity in Developmental Dyslexia
1623
+
1624
+ 13
1625
+ 6. Conclusion and future works
1626
+ Our results support the main assumption of the TSF that
1627
+ DD involves a specific deficit in the low-frequency phase
1628
+ locking mechanisms in the auditory cortex, thereby
1629
+ potentially affecting phonological development 56.
1630
+ In confirmation of this, we find an anomaly that emerges
1631
+ primarily in the causal relationships of channels that
1632
+ function as sinks, which is significantly more pronounced
1633
+ than when only the total activity is considered. Hence, it
1634
+ is reasonable to consider a division into Granger-causing
1635
+ or Granger-caused relationships. This, in turn, suggests
1636
+ that the main differences contributing to DD emerge
1637
+ when certain brain areas must function as receptors in the
1638
+ interactions between channels.
1639
+ Furthermore, our results are in line with previous
1640
+ research, which has already detected an anomaly in the
1641
+ right-lateralized Theta band. We have clearly identified
1642
+ this here across all three scenarios (sources, sinks, total
1643
+ activity).
1644
+ We also find confirmation for the higher brain activity in
1645
+ learners with DD, although differences are more
1646
+ significant for the sinks in the Theta and Gamma bands,
1647
+ in turn leading to more total activity. The highest
1648
+ classifier performance (accuracy and AUC) is hereby
1649
+ found in the sink scenario. For the Beta band, the
1650
+ difference in activity is more consistent across all three
1651
+ scenarios. The classifier also performs well for the Beta
1652
+ band in all three scenarios, with few differences
1653
+ observed, thereby confirming the important role played
1654
+ by this band in the sensorimotor coding of speech.
1655
+ The results reflect the causal activity generated in the
1656
+ brain subjected to prosodic-syllabic stimulus at 4.8 Hz.
1657
+ Consequently, future work could consider the Granger
1658
+ causality relationships in the phases across channels and
1659
+ bands using higher frequency stimuli to stimulate
1660
+ syllabic-phonetic and phonetic activity.
1661
+ Acknowledgements
1662
+ This work was supported by projects PGC2018-098813-
1663
+ B-C32 (Spanish β€œMinisterio de Ciencia, InnovaciΓ³n y
1664
+ Universidades”), UMA20-FEDERJA-086 (ConsejerΓ­a de
1665
+ econnomΓ­a y conocimiento,Junta de AndalucΓ­a) and by
1666
+ European Regional Development Funds (ERDF).
1667
+
1668
+ References
1669
+ 1. Peterson RL, Pennington BF. Developmental dyslexia.
1670
+ The lancet. 2012;379:1997–2007.
1671
+ 2. Thompson PA, Hulme C, Nash HM, Gooch D, Hayiou-
1672
+ Thomas E, Snowling MJ. Developmental dyslexia:
1673
+ predicting individual risk. Journal of Child Psychology
1674
+ and Psychiatry. 2015;56:976–987.
1675
+ 3. Clark KA, Helland T, Specht K, et al. Neuroanatomical
1676
+ precursors of dyslexia identified from pre-reading through
1677
+ to age 11. Brain. 2014;137:3136–3141.
1678
+ 4. Chaturvedi M, Bogaarts JG, Kozak VV, et al. Phase lag
1679
+ index and spectral power as QEEG features for
1680
+ identification of patients with mild cognitive impairment
1681
+ in
1682
+ Parkinson's
1683
+ disease.
1684
+ Clinical
1685
+ Neurophysiology.
1686
+ 2019;130:1937–1944.
1687
+ 5. Romeo RR, Segaran J, Leonard JA, et al. Language
1688
+ exposure relates to structural neural connectivity in
1689
+ childhood. Journal of Neuroscience. 2018;38:7870–7877.
1690
+ 6. Schmidt C, Piper D, Pester B, Mierau A, Witte H.
1691
+ Tracking the reorganization of module structure in time-
1692
+ varying weighted brain functional connectivity networks.
1693
+ International
1694
+ journal
1695
+ of
1696
+ neural
1697
+ systems.
1698
+ 2018;28:1750051.
1699
+ 7. Yaqub MA, Hong KS, Zafar A, Kim CS. Control of
1700
+ transcranial direct current stimulation duration by
1701
+ assessing
1702
+ functional
1703
+ connectivity
1704
+ of
1705
+ near-infrared
1706
+ spectroscopy signals. International Journal of Neural
1707
+ Systems. 2022;32:2150050.
1708
+ 8. Martinez-Murcia FJ, Ortiz A, Gorriz JM, et al. EEG
1709
+ connectivity analysis using denoising autoencoders for the
1710
+ detection of dyslexia. International Journal of Neural
1711
+ Systems. 2020;30:2050037.
1712
+ 9. Osterhage H, Mormann F, Wagner T, Lehnertz K.
1713
+ Measuring the directionality of coupling: phase versus
1714
+ state space dynamics and application to EEG time series.
1715
+ International journal of neural systems. 2007;17:139–
1716
+ 148.
1717
+ 10. Ding M, Chen Y, Bressler SL. Granger causality: basic
1718
+ theory and application to neuroscience. Handbook of time
1719
+ series analysis: recent theoretical developments and
1720
+ applications. 2006:437–460.
1721
+ 11. Power AJ, Colling LJ, Mead N, Barnes L, Goswami U.
1722
+ Neural encoding of the speech envelope by children with
1723
+ developmental
1724
+ dyslexia.
1725
+ Brain
1726
+ and
1727
+ Language.
1728
+ 2016;160:1–10.
1729
+ 12. Goswami U, Mead N, Fosker T, Huss M, Barnes L, Leong
1730
+ V. Impaired perception of syllable stress in children with
1731
+ dyslexia: A longitudinal study. Journal of memory and
1732
+ language. 2013;69:1–17.
1733
+ 13. SoltΓ©sz F, SzΕ±cs D, Leong V, White S, Goswami U.
1734
+ Differential entrainment of neuroelectric delta oscillations
1735
+ in developmental dyslexia. PLoS One. 2013;8:e76608.
1736
+ 14. Poeppel D. The analysis of speech in different temporal
1737
+ integration
1738
+ windows:
1739
+ cerebral
1740
+ lateralization
1741
+ as
1742
+ β€˜asymmetric sampling in time’. Speech communication.
1743
+ 2003;41:245–255.
1744
+ 15. Lakatos P, Karmos G, Mehta AD, Ulbert I, Schroeder CE.
1745
+ Entrainment of neuronal oscillations as a mechanism of
1746
+ attentional selection. science. 2008;320:110–113.
1747
+ 16. Giraud AL, Poeppel D. Speech perception from a
1748
+ neurophysiological perspective. The human auditory
1749
+ cortex: Springer; 2012.
1750
+
1751
+ RodrΓ­guez-RodrΓ­guez et al.
1752
+
1753
+ 14
1754
+ 17. Luo H, Liu Z, Poeppel D. Auditory cortex tracks both
1755
+ auditory and visual stimulus dynamics using low-
1756
+ frequency neuronal phase modulation. PLoS biology.
1757
+ 2010;8:e1000445.
1758
+ 18. Di Liberto GM, Peter V, Kalashnikova M, Goswami U,
1759
+ Burnham D, Lalor EC. Atypical cortical entrainment to
1760
+ speech in the right hemisphere underpins phonemic
1761
+ deficits in dyslexia. Neuroimage. 2018;175:70–79.
1762
+ 19. Flanagan
1763
+ S,
1764
+ Goswami
1765
+ U.
1766
+ The
1767
+ role
1768
+ of
1769
+ phase
1770
+ synchronisation between low frequency amplitude
1771
+ modulations in child phonology and morphology speech
1772
+ tasks. The Journal of the Acoustical Society of America.
1773
+ 2018;143:1366–1375.
1774
+ 20. Molinaro N, Lizarazu M, Lallier M, Bourguignon M,
1775
+ Carreiras M. Out-of-synchrony speech entrainment in
1776
+ developmental
1777
+ dyslexia.
1778
+ Human
1779
+ brain
1780
+ mapping.
1781
+ 2016;37:2767–2783.
1782
+ 21. Power AJ, Mead N, Barnes L, Goswami U. Neural
1783
+ entrainment to rhythmic speech in children with
1784
+ developmental
1785
+ dyslexia.
1786
+ Frontiers
1787
+ in
1788
+ human
1789
+ neuroscience. 2013;7:777.
1790
+ 22. Unde SA, Shriram R. Coherence analysis of EEG signal
1791
+ using power spectral density. Paper presented at: 2014
1792
+ Fourth International Conference on Communication
1793
+ Systems and Network Technologies, 2014.
1794
+ 23. Žarić G, Correia JM, GonzÑlez GF, et al. Altered patterns
1795
+ of directed connectivity within the reading network of
1796
+ dyslexic children and their relation to reading dysfluency.
1797
+ Developmental cognitive neuroscience. 2017;23:1–13.
1798
+ 24. GonzÑlez GF, Van der Molen MJW, Žarić G, et al. Graph
1799
+ analysis of EEG resting state functional networks in
1800
+ dyslexic
1801
+ readers.
1802
+ Clinical
1803
+ Neurophysiology.
1804
+ 2016;127:3165–3175.
1805
+ 25. Stam CJ, Nolte G, Daffertshofer A. Phase lag index:
1806
+ assessment of functional connectivity from multi channel
1807
+ EEG and MEG with diminished bias from common
1808
+ sources. Human brain mapping. 2007;28:1178–1193.
1809
+ 26. Mormann F, Lehnertz K, David P, Elger CE. Mean phase
1810
+ coherence as a measure for phase synchronization and its
1811
+ application to the EEG of epilepsy patients. Physica D:
1812
+ Nonlinear Phenomena. 2000;144:358–369.
1813
+ 27. Gao Y, Wang X, Potter T, Zhang J, Zhang Y. Single-trial
1814
+ EEG
1815
+ emotion
1816
+ recognition
1817
+ using
1818
+ Granger
1819
+ Causality/Transfer
1820
+ Entropy
1821
+ analysis.
1822
+ Journal
1823
+ of
1824
+ Neuroscience Methods. 2020;346:108904.
1825
+ 28. Uchida T, Fujiwara K, Inoue T, Maruta Y, Kano M,
1826
+ Suzuki M. Analysis of VNS effect on EEG connectivity
1827
+ with granger causality and graph theory. Paper presented
1828
+ at: 2018 Asia-Pacific Signal and Information Processing
1829
+ Association Annual Summit and Conference (APSIPA
1830
+ ASC), 2018.
1831
+ 29. Tripanpitak K, He S, Sânmezışık I, Morant T, Huang SY,
1832
+ Yu W. Granger Causality-Based Pain Classification Using
1833
+ EEG Evoked by Electrical Stimulation Targeting
1834
+ Nociceptive
1835
+ AΞ΄
1836
+ and
1837
+ C
1838
+ Fibers.
1839
+ IEEE
1840
+ Access.
1841
+ 2021;9:10089–10106.
1842
+ 30. Rimehaug AE, Stasik AJ, Hagen E, et al. Uncovering
1843
+ circuit mechanisms of current sinks and sources with
1844
+ biophysical simulations of primary visual cortex. bioRxiv.
1845
+ 2022.
1846
+ 31. Sotero RC, Bortel A, Naaman S, et al. Laminar
1847
+ distribution of phase-amplitude coupling of spontaneous
1848
+ current sources and sinks. Frontiers in neuroscience.
1849
+ 2015;9:454.
1850
+ 32. Epstein CM, Adhikari BM, Gross R, Willie J, Dhamala M.
1851
+ Application of high-frequency Granger causality to
1852
+ analysis of epileptic seizures and surgical decision
1853
+ making. Epilepsia. 2014;55:2038–2047.
1854
+ 33. Hinton G, LeCun Y, Bengio Y. Deep learning. Nature.
1855
+ 2015;521:436–444.
1856
+ 34. Ortiz A, Martinez-Murcia FJ, Luque JL, GimΓ©nez A,
1857
+ Morales-Ortega R, Ortega J. Dyslexia diagnosis by EEG
1858
+ temporal and spectral descriptors: an anomaly detection
1859
+ approach. International Journal of Neural Systems.
1860
+ 2020;30:2050029.
1861
+ 35. Li R, Principe JC. Blinking artifact removal in cognitive
1862
+ EEG data using ICA. Paper presented at: 2006
1863
+ International Conference of the IEEE Engineering in
1864
+ Medicine and Biology Society, 2006.
1865
+ 36. Robertson DGE, Dowling JJ. Design and responses of
1866
+ Butterworth and critically damped digital filters. Journal
1867
+ of Electromyography and Kinesiology. 2003;13:569–573.
1868
+ 37. Granger
1869
+ CWJ.
1870
+ Investigating
1871
+ causal
1872
+ relations
1873
+ by
1874
+ econometric
1875
+ models
1876
+ and
1877
+ cross-spectral
1878
+ methods.
1879
+ Econometrica: journal of the Econometric Society.
1880
+ 1969:424–438.
1881
+ 38. Akaike H. A new look at the statistical model
1882
+ identification. IEEE transactions on automatic control.
1883
+ 1974;19:716–723.
1884
+ 39. Hu F, Wang H, Wang Q, Feng N, Chen J, Zhang T.
1885
+ Acrophobia quantified by EEG based on CNN
1886
+ incorporating Granger causality. International Journal of
1887
+ Neural Systems. 2021;31:2050069.
1888
+ 40. Wang X, Chen Y, Bressler SL, Ding M. Granger causality
1889
+ between multiple interdependent neurobiological time
1890
+ series: blockwise versus pairwise methods. International
1891
+ journal of neural systems. 2007;17:71–78.
1892
+ 41. Schoffelen JM, Gross J. Source connectivity analysis with
1893
+ MEG and EEG. Human brain mapping. 2009;30:1857–
1894
+ 1865.
1895
+ 42. Sakkalis V. Review of advanced techniques for the
1896
+ estimation
1897
+ of
1898
+ brain
1899
+ connectivity
1900
+ measured
1901
+ with
1902
+ EEG/MEG. Computers in biology and medicine.
1903
+ 2011;41:1110–1117.
1904
+ 43. Verly M, Verhoeven J, Zink I, et al. Altered functional
1905
+ connectivity of the language network in ASD: role of
1906
+ classical language areas and cerebellum. NeuroImage:
1907
+ Clinical. 2014;4:374–382.
1908
+ 44. Chamakura L, Saha G. An instance voting approach to
1909
+ feature selection. Information Sciences. 2019;504:449–
1910
+ 469.
1911
+
1912
+
1913
+ Neural source/sink phase connectivity in Developmental Dyslexia
1914
+
1915
+ 15
1916
+ 45. Seijo-Pardo B, Porto-Dı́az I, Bolón-Canedo V, Alonso-
1917
+ Betanzos A. Ensemble feature selection: homogeneous
1918
+ and
1919
+ heterogeneous
1920
+ approaches.
1921
+ Knowledge-Based
1922
+ Systems. 2017;118:124–139.
1923
+ 46. Weed DL. Weight of evidence: a review of concept and
1924
+ methods. Risk Analysis: An International Journal.
1925
+ 2005;25:1545–1557.
1926
+ 47. KolΓ‘cek J, RezΓ‘c M. Assessment of scoring models using
1927
+ information value. Paper presented at: 19th International
1928
+ Conference on Computational Statistics, Paris France,
1929
+ 2010.
1930
+ 48. Intelligent I, Siddiqi N. Credit Risk Scorecards. 2006.
1931
+ 49. Nembrini S, KΓΆnig IR, Wright MN. The revival of the
1932
+ Gini importance? Bioinformatics. 2018;34:3711–3718.
1933
+ 50. Darst BF, Malecki KC, Engelman CD. Using recursive
1934
+ feature elimination in random forest to account for
1935
+ correlated variables in high dimensional data. BMC
1936
+ genetics. 2018;19:1–6.
1937
+ 51. Jović A, Brkić K, Bogunović N. A review of feature
1938
+ selection methods with applications. Paper presented at:
1939
+ 2015 38th international convention on information and
1940
+ communication
1941
+ technology,
1942
+ electronics
1943
+ and
1944
+ microelectronics (MIPRO), 2015.
1945
+ 52. Ng AY. Feature selection, L 1 vs. L 2 regularization, and
1946
+ rotational invariance. Paper presented at: Proceedings of
1947
+ the twenty-first international conference on Machine
1948
+ learning, 2004.
1949
+ 53. Freund
1950
+ Y,
1951
+ Schapire
1952
+ RE.
1953
+ A
1954
+ decision-theoretic
1955
+ generalization of on-line learning and an application to
1956
+ boosting. Journal of computer and system sciences.
1957
+ 1997;55:119–139.
1958
+ 54. Friedman
1959
+ JH.
1960
+ Stochastic
1961
+ gradient
1962
+ boosting.
1963
+ Computational statistics & data analysis. 2002;38:367–
1964
+ 378.
1965
+ 55. Goswami U, Gerson D, Astruc L. Amplitude envelope
1966
+ perception, phonology and prosodic sensitivity in children
1967
+ with developmental dyslexia. Reading and Writing.
1968
+ 2010;23:995–1019.
1969
+ 56. Goswami U. A temporal sampling framework for
1970
+ developmental dyslexia. Trends in cognitive sciences.
1971
+ 2011;15:3–10.
1972
+ 57. Poeppel D, Idsardi WJ, Van Wassenhove V. Speech
1973
+ perception at the interface of neurobiology and linguistics.
1974
+ Philosophical Transactions of the Royal Society B:
1975
+ Biological Sciences. 2008;363:1071–1086.
1976
+ 58. Luo H, Poeppel D. Phase patterns of neuronal responses
1977
+ reliably discriminate speech in human auditory cortex.
1978
+ Neuron. 2007;54:1001–1010.
1979
+ 59. Poeppel D. The neuroanatomic and neurophysiological
1980
+ infrastructure for speech and language. Current opiniΓ³n in
1981
+ neurobiology. 2014;28:142–149.
1982
+ 60. Arnal LH, Doelling KB, Poeppel D. Delta–beta coupled
1983
+ oscillations underlie temporal prediction accuracy.
1984
+ Cerebral Cortex. 2015;25:3077–3085.
1985
+ 61. Facoetti A, Trussardi AN, Ruffino M, et al. Multisensory
1986
+ spatial attention deficits are predictive of phonological
1987
+ decoding skills in developmental dyslexia. Journal of
1988
+ cognitive neuroscience. 2010;22:1011–1025.
1989
+ 62. Leong V, Goswami U. Assessment of rhythmic
1990
+ entrainment at multiple timescales in dyslexia: evidence
1991
+ for disruption to syllable timing. Hearing research.
1992
+ 2014;308:141–161.
1993
+ 63. Lehongre K, Ramus F, Villiermet N, Schwartz D, Giraud
1994
+ AL. Altered low-gamma sampling in auditory cortex
1995
+ accounts for the three main facets of dyslexia. Neuron.
1996
+ 2011;72:1080–1090.
1997
+ 64. Schroeder CE, Lakatos P. Low-frequency neuronal
1998
+ oscillations as instruments of sensory selection. Trends in
1999
+ neurosciences. 2009;32:9–18.
2000
+ 65. Canolty RT, Edwards E, Dalal SS, et al. High gamma
2001
+ power is phase-locked to theta oscillations in human
2002
+ neocortex. science. 2006;313:1626–1628.
2003
+ 66. Giraud AL, Kleinschmidt A, Poeppel D, Lund TE,
2004
+ Frackowiak RSJ, Laufs H. Endogenous cortical rhythms
2005
+ determine cerebral specialization for speech perception
2006
+ and production. Neuron. 2007;56:1127–1134.
2007
+ 67. Kuhl PK. Early language acquisition: cracking the speech
2008
+ code. Nature reviews neuroscience. 2004;5:831–843.
2009
+
2010
+
2011
+
AdAyT4oBgHgl3EQfq_lh/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
BdFJT4oBgHgl3EQfsS2B/content/tmp_files/2301.11612v1.pdf.txt ADDED
@@ -0,0 +1,1807 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ A neural network potential with self-trained atomic fingerprints:
2
+ a test with the mW water potential
3
+ Francesco Guidarelli Mattioli, Francesco Sciortino, and John Russoβˆ—
4
+ Sapienza University of Rome, Piazzale Aldo Moro 2, 00185 Rome, Italy
5
+ (Dated: January 30, 2023)
6
+ We present a neural network (NN) potential based on a new set of atomic fingerprints built upon
7
+ two- and three-body contributions that probe distances and local orientational order respectively.
8
+ Compared to existing NN potentials, the atomic fingerprints depend on a small set of tuneable
9
+ parameters which are trained together with the neural network weights. To tackle the simultaneous
10
+ training of the atomic fingerprint parameters and neural network weights we adopt an annealing
11
+ protocol that progressively cycles the learning rate, significantly improving the accuracy of the NN
12
+ potential. We test the performance of the network potential against the mW model of water, which
13
+ is a classical three-body potential that well captures the anomalies of the liquid phase. Trained on
14
+ just three state points, the NN potential is able to reproduce the mW model in a very wide range of
15
+ densities and temperatures, from negative pressures to several GPa, capturing the transition from
16
+ an open random tetrahedral network to a dense interpenetrated network. The NN potential also
17
+ reproduces very well properties for which it was not explicitly trained, such as dynamical properties
18
+ and the structure of the stable crystalline phases of mW.
19
+ I.
20
+ INTRODUCTION
21
+ Machine learning (ML) potentials represent one of the
22
+ emerging trends in condensed matter physics and are
23
+ revolutionising the landscape of computational research.
24
+ Nowadays, different methods to derive ML potentials
25
+ have been proposed, providing a powerful methodology
26
+ to model liquids and solid phases in a large variety of
27
+ molecular systems [1–16, 16, 17]. Among these methods,
28
+ probably the most successful representation of a ML po-
29
+ tential so far is given by Neural Network (NN) potentials,
30
+ where the potential energy surface is the output of a feed-
31
+ forward neural network [18–35].
32
+ In short, the idea underlying NN potentials construc-
33
+ tion is to train a neural network to represent the po-
34
+ tential energy surface of a target system.
35
+ The model
36
+ is initially trained on a set of configurations generated
37
+ ad-hoc, for which total energies and forces are known,
38
+ by minimizing a suitable defined loss-function based on
39
+ the error in the energy and force predictions.
40
+ If the
41
+ training set is sufficiently broad and representative, the
42
+ model can then be used to evaluate the total energy and
43
+ forces of any related atomic configuration with an accu-
44
+ racy comparable to the original potential. Typically the
45
+ original potential will include additional degrees of free-
46
+ dom, such as the electron density for DFT calculations,
47
+ or solvent atoms in protein simulations, which make the
48
+ full computation very expensive.
49
+ By training the net-
50
+ work only on a subset of the original degrees of freedom
51
+ one obtaines a coarse-grained representation that can be
52
+ simulated at a much reduced computational cost. NN
53
+ potentials thus combine the best of two worlds, retain-
54
+ ing the accuracy of the underlying potential model, at
55
+ the much lower cost of coarse-grained classical molecu-
56
+ βˆ—Corresponding author: john.russo@uniroma1.it
57
+ lar dynamics simulations. The accuracy of the NN po-
58
+ tential depends crucially on how local atomic positions
59
+ are encoded in the input of the neural network, which
60
+ needs to retain the symmetries of the underlying Hamil-
61
+ tonian, i.e. rotational, translational, and index permu-
62
+ tation invariance. Several methods have been proposed
63
+ in the literature [12, 36], such as the approaches based
64
+ on the Behler-Parrinello (BP) symmetry functions [18],
65
+ the Smooth Overlap of Atomic Positions (SOAP) [37],
66
+ N-body iterative contraction of equivariants (NICE) [38]
67
+ and polynomial symmetry functions [39], or frameworks
68
+ like the DeepMD [23], SchNet [22] and RuNNer [18]. In
69
+ all cases, atomic positions are transformed into atomic
70
+ fingerprints (AFs). The choice of the AFs is particularly
71
+ relevant, as it greatly affects the accuracy and generality
72
+ of the resulting NN potential.
73
+ We develop here a fully learnable NN potential in
74
+ which the AFs, while retaining the simplicity of typi-
75
+ cal local fingerprints, do not need to be fixed beforehand
76
+ but instead are learned during the training procedure.
77
+ The coupled training of the atomic fingerprint param-
78
+ eters and of the network weights makes the NN train-
79
+ ing process more efficient since the NN representation is
80
+ spontaneously built on a variable atomic fingerprint rep-
81
+ resentation. To tackle the combined minimization of the
82
+ AF parameters and of the network weights we adopt an
83
+ efficient annealing procedure, that periodically cycles the
84
+ learning rate, i.e. the step size of the minimization algo-
85
+ rithm, resulting in a fast and accurate training process.
86
+ We validate the NN potential on the mW model of
87
+ water [40], which is a one-site classical potential that
88
+ has found widespread adoption to study water’s anoma-
89
+ lies [41, 42] and crystallization phenomena [43, 44]. Since
90
+ the first pioneering MD simulations [45, 46], water is of-
91
+ ten chosen as a prototypical case study, as the large num-
92
+ ber of distinct local structures that are compatible with
93
+ its tetrahedral coordination make it the molecule with
94
+ the most complex thermodynamic behavior [47], for ex-
95
+ arXiv:2301.11612v1 [cond-mat.soft] 27 Jan 2023
96
+
97
+ 2
98
+ ample displaying a liquid-liquid critical point at super-
99
+ cooled conditions [48–52]. NN potentials for water have
100
+ been developed starting from density functional calcu-
101
+ lations, with different levels of accuracy [53–60].
102
+ NN
103
+ potentials have also been proposed to parametrise accu-
104
+ rate classical models for water with the aim of speeding
105
+ up the calculations when multi-body interactions are in-
106
+ cluded [61], as in the MBpol model [62–64] or for testing
107
+ the relevance of the long range interactions, as for the
108
+ SPC/E model [65]. We choose the mW potential as our
109
+ benchmark system because its explicit three-body poten-
110
+ tial term offers a challenge to the NN representation that
111
+ is not found in molecular models built from pair-wise
112
+ interactions. We stress that we train the NN-potential
113
+ against data which can be generated easily and for which
114
+ structural and dynamic properties are well known (or
115
+ can be evaluated with small numerical errors) in a wide
116
+ range of temperatures and densities. In this way, we can
117
+ perform a quantitative accurate comparison between the
118
+ original mW model and the hereby proposed NN model.
119
+ Our results show that training the NN potential at
120
+ even just one density-temperature state point provides
121
+ an accurate description of the mW model in a surround-
122
+ ing phase space region that is approximately a hundred
123
+ kelvins wide. A training based on three different state
124
+ points extends the convergence window extensively, ac-
125
+ curately reproducing state points at extreme conditions,
126
+ i.e.
127
+ large negative and (crushingly) positive pressures.
128
+ We will show that the NN reproduces thermodynamic,
129
+ structural and dynamical properties of the mW liquid
130
+ state, as well as structural properties of all the stable
131
+ crystalline phases of mW water.
132
+ The paper is organized as follows. In Section II we de-
133
+ scribe the new atomic fingerprints and the details about
134
+ the Neural Network potential implementation, including
135
+ the warm restart procedure used to train the weights
136
+ and the fingerprints at the same time.
137
+ In Section III
138
+ we present the results, which include the accuracy of the
139
+ models built from training sets that include one or three
140
+ state points, and a comparison of the thermodynamic,
141
+ structural and dynamic properties with those of the orig-
142
+ inal mW model. We conclude in Section IV.
143
+ II.
144
+ THE NEURAL NETWORK MODEL
145
+ The most important step in the design of a feed-
146
+ forward neural network potential is the choice on how to
147
+ define the first and the last layers of the network, respec-
148
+ tively named the input and output layers. We start with
149
+ the output layer, as it determines the NN potential ar-
150
+ chitecture to be constructed. Here we follow the Behler
151
+ Parrinello NN potential architecture [18], in which the
152
+ total energy of the system is decomposed as the sum of
153
+ local fields (Ei), each one representing the contribution of
154
+ a local environment centered around atom i. Being this
155
+ a many-body contribution, it is important to note that
156
+ Ei is not the energy of the single atom i, but of all its
157
+ environment (see also the Appendix A). With this choice,
158
+ the total energy of the system is simply the sum over all
159
+ atoms, E = οΏ½ Ei, and the force βƒ—fi acting on atom i is
160
+ the negative gradient of the total energy with respect to
161
+ the coordinates Ξ½ of atom i, e.g. fiΞ½ = βˆ‚E/βˆ‚xiΞ½. We
162
+ have to point out that a NN potential is differentiable
163
+ and hence it is possible to evaluate the gradient of the
164
+ energy analytically. This allows to compute forces of the
165
+ NN potential in the same way of other force fields, e.g.
166
+ by the negative gradient of the total potential energy.
167
+ The input layer is built from two-body (distances) and
168
+ three-body (angles) descriptors of the local environment,
169
+ βƒ—D(i) and T (i) respectively, ensuring translational and ro-
170
+ tational invariance. The first layer of the neural network
171
+ is the Atomic Fingerprint Constructor (AFC), as shown
172
+ in Fig. 1, which applies an exponential weighting on the
173
+ atomic descriptors, restoring the invariance under per-
174
+ mutations of atomic indexes. The outputs of this first
175
+ layer are the atomic fingerprints (AFs) and in turn these
176
+ are given to the first hidden layer. We will show how
177
+ this organization of the AFC layer allows for the inter-
178
+ nal parameters of the exponential weighting to be trained
179
+ together with the weights in the hidden layers of the net-
180
+ work. In the following we describe in detail the construc-
181
+ tion of the inputs and the calculation flow in the first
182
+ layers.
183
+ A.
184
+ The atomic fingerprints
185
+ The choice of input layer presents considerably more
186
+ freedom, and it is here that we deviate from previous NN
187
+ potentials. The data in this layer should retain all the in-
188
+ formation needed to properly evaluate forces and energies
189
+ of the particles in the system, possibly exploiting the in-
190
+ ternal symmetries of the Hamiltonian (which in isotropic
191
+ fluids are the rotational, translational and permutational
192
+ invariance) to reduce the number of degenerate inputs.
193
+ Given that the output was chosen as Ei, the energy of the
194
+ atomic environment surrounding atom i, the input uses
195
+ an atom-centered representation of the local environment
196
+ of atom i.
197
+ In the input layer, we define an atom-centered repre-
198
+ sentation of the local environment of atom i, consider-
199
+ ing both the distances rij with the nearest neighbours j
200
+ within a spatial cut-off Rc, and the angles ΞΈjik between
201
+ atom i and the pair of neighbours jk that are within a
202
+ cut-off Rcβ€². More precisely, for each atom j within Rc
203
+ from i we calculate the following descriptors
204
+ D(i)
205
+ j (rij; Rc) =
206
+ οΏ½
207
+ 1
208
+ 2
209
+ οΏ½
210
+ 1 + cos
211
+ οΏ½
212
+ Ο€ rij
213
+ Rc
214
+ οΏ½οΏ½
215
+ rij ≀ Rc
216
+ 0
217
+ rij > Rc
218
+ (1)
219
+ and, for each triplet j βˆ’ i βˆ’ k within Rcβ€² from i,
220
+ T (i)
221
+ jk (rij, rik, ΞΈjik) =
222
+ (2)
223
+ 1
224
+ 2 [1 + cos (ΞΈjik)] D(i)
225
+ j (rij; Rc
226
+ β€²) D(i)
227
+ k (rik; Rc
228
+ β€²)
229
+
230
+ 3
231
+ C
232
+ Ei
233
+ Ξ±
234
+ Ξ³
235
+ Ξ΄
236
+ Ξ²
237
+ Compression
238
+ βƒ—
239
+ D(i)
240
+ βƒ—
241
+ D(i)
242
+ T(i)
243
+ ΞΈjik
244
+ rij
245
+ ≑ (
246
+ βƒ—
247
+ D(i), T(i))
248
+ A
249
+ B
250
+ Atomic Fingerprint
251
+ Constructor
252
+ Hidden
253
+ layers
254
+ FIG. 1: Schematic representation of the Neural Network Potential flow. (A) Starting from the relative distances and the triplets
255
+ angles between neighbouring atoms, the input layer evaluates the atomic descriptors βƒ—D(i) = {D(i)
256
+ j } (Eq. 1) and T (i) = {T (i)
257
+ jk }
258
+ (Eq. 2). (B) The first layer is the Atomic Fingerprint Constructor (AFC) and it combines the atomic descriptors into atomic
259
+ fingerprints, weighting them with an exponential function. The red nodes perform the calculation of Eq. 5, where from the two-
260
+ body descriptors a weighting vector βƒ—D(i)
261
+ w (Ξ±) = {eΞ±D(i)
262
+ j } is calculated (square with Ξ±) and then the scalar product βƒ—D(i) Β· βƒ—D(i)
263
+ w (Ξ±)
264
+ is computed (square with point) and finally a logarithm is applied (circle). The blue nodes perform the calculation of Eq. 7,
265
+ where two weighting vectors are calculated from the two-body descriptors namely βƒ—D(i)
266
+ w (Ξ³) and βƒ—D(i)
267
+ w (Ξ΄) and one weighting
268
+ matrix from the three-body descriptors T (i)
269
+ w (Ξ²) = {eΞ²T (i)
270
+ jk /2}. Finally in the compression unit (Eq 6) values are combined as
271
+ 0.5[ βƒ—D(i) β—¦ βƒ—D(i)
272
+ w (Ξ³)]T [T (i) β—¦T (i)
273
+ w (Ξ²)][ βƒ—D(i) β—¦ βƒ—D(i)
274
+ w (Ξ΄)] where we use the circle symbol for the element-wise multiplication. The output
275
+ value of the compression unit is given to the logarithm function (circle). The complete network (D) is made of ten AFC units
276
+ and two hidden layers with 25 nodes per layer and here is depicted 2.5 times smaller.
277
+ Here i indicates the label of i-th particle while in-
278
+ dex j and k run over all other particles in the system.
279
+ In Eq. 1, D(i)
280
+ j (rij; Rc) is a function that goes continu-
281
+ ously to zero at the cut-off (including its derivatives).
282
+ The choice of this functional form guarantees that D(i)
283
+ j
284
+ is able to express contributions even from neighbours
285
+ close to the cut-off.
286
+ Other choices, based on polyno-
287
+ mials or other non-linear functions, have been tested
288
+ in the past [31].
289
+ For example, we tested a parabolic
290
+ cutoff function which produced considerably worse re-
291
+ sults than the cutoff function in Eq. 1.
292
+ The function
293
+ T (i)
294
+ jk (rij, rik, ΞΈjik) is also continuous at the triplet cutoff
295
+ Rβ€²
296
+ c.
297
+ The angular function
298
+ 1
299
+ 2 [1 + cos (ΞΈjik)] guarantees
300
+ that 0 ≀ T (i)
301
+ jk (rij, rik, ΞΈjik) ≀ 1. We note that the use of
302
+ relative distances and angles in Eq. 1-2 guarantees trans-
303
+ lational and rotational invariance.
304
+ The pairs and triplets descriptors are then fed to the
305
+ AFC layer to compute the atomic fingerprints, AFs.
306
+ These are computed by projecting the D(i)
307
+ j
308
+ and T (i)
309
+ jk de-
310
+ scriptors on a exponential set of functions defined by
311
+
312
+ 4
313
+ D
314
+ (i)(Ξ±) = ln
315
+ οΏ½
316
+ οΏ½οΏ½
317
+ jΜΈ=i
318
+ D(i)
319
+ j eΞ±D(i)
320
+ j
321
+ + Ο΅
322
+ οΏ½
323
+ οΏ½ βˆ’ ZΞ±
324
+ (3)
325
+ T
326
+ (i)(Ξ², Ξ³, Ξ΄) = ln
327
+ οΏ½
328
+ οΏ½ οΏ½
329
+ jΜΈ=kΜΈ=i
330
+ T (i)
331
+ jk eΞ²T (i)
332
+ jk eΞ³D(i)
333
+ j eΞ΄D(i)
334
+ k
335
+ 2
336
+ + Ο΅
337
+ οΏ½
338
+ οΏ½(4)
339
+ βˆ’ZΞ²Ξ³Ξ΄
340
+ These AFs are built summing over all pairs and all
341
+ triplets involving particle i, making them invariant un-
342
+ der permutations, and multiplying each descriptor by an
343
+ exponential filter whose parameters are called α for dis-
344
+ tance AFs, and Ξ², Ξ³, Ξ΄ for the triplet AFs. These param-
345
+ eters play the role of feature selectors, i.e. by choosing
346
+ an appropriate list of Ξ±, Ξ², Ξ³, Ξ΄ the AFs can extract the
347
+ necessary information from the atomic descriptors. The
348
+ best choice of Ξ±, Ξ², Ξ³, Ξ΄ will emerge automatically dur-
349
+ ing the training stage. In Eqs. 3-4, the number Ο΅ is set to
350
+ 10βˆ’3 and fixes the value of energy in the rare event that
351
+ no neighbors are found inside the cutoff. Parameters ZΞ±
352
+ and ZΞ²Ξ³Ξ΄ are optimized during the training process, shift-
353
+ ing the AFs towards positive or negative values, and act
354
+ as normalization factors that improve the representation
355
+ of the NN.
356
+ The definitions in equations 3-4 can be reformulated in
357
+ terms of product between vectors and matrices in the fol-
358
+ lowing way. The descriptors in equations 1-2 for particle i
359
+ can be represented as a vector βƒ—D(i) = {D(i)
360
+ j } and a matrix
361
+ T (i) = {T (i)
362
+ jk } respectively. Given a choice of Ξ±, Ξ², Ξ³ and
363
+ Ξ΄, three weighting vector βƒ—D(i)
364
+ w (Ξ±) = {eΞ±D(i)
365
+ j }, βƒ—D(i)
366
+ w (Ξ³) =
367
+ {eΞ³D(i)
368
+ j } and βƒ—D(i)
369
+ w (Ξ΄) = {eΞ΄D(i)
370
+ j } and one weighting ma-
371
+ trix T (i)
372
+ w (Ξ²) = {eΞ²T (i)
373
+ jk /2} are calculated from βƒ—D(i) and
374
+ T (i).
375
+ The 2-body atomic fingerprint (Eq. 3) is finally
376
+ computed as
377
+ D
378
+ (i)(Ξ±) = ln
379
+ οΏ½
380
+ βƒ—D(i) Β· βƒ—D(i)
381
+ w (Ξ±) + Ο΅
382
+ οΏ½
383
+ βˆ’ ZΞ±
384
+ (5)
385
+ The 3-body atomic fingerprint (Eq. 4) is computed first
386
+ by what we call compression step in Fig. 1 as
387
+ T c
388
+ (i) = [ βƒ—D(i) β—¦ βƒ—D(i)
389
+ w (Ξ³)]T [T (i) β—¦ T (i)
390
+ w (Ξ²)][ βƒ—D(i) β—¦ βƒ—D(i)
391
+ w (Ξ΄)]
392
+ 2
393
+ (6)
394
+ and finally by
395
+ T
396
+ (i)(Ξ², Ξ³, Ξ΄) = ln
397
+ οΏ½
398
+ T c
399
+ (i)(Ξ², Ξ³, Ξ΄) + Ο΅
400
+ οΏ½
401
+ βˆ’ ZΞ²Ξ³Ξ΄
402
+ (7)
403
+ where we use the circle symbol for the element-wise mul-
404
+ tiplication. The NN potential flow is depicted in Figure
405
+ 1 following the vectorial representation.
406
+ In summary, our AFs select the local descriptors use-
407
+ ful for the reconstruction of the potential by weight-
408
+ ing them with an exponential factor tuned with expo-
409
+ nents Ξ±, Ξ², Ξ³, Ξ΄. A similar weighting procedure has been
410
+ showed to be extremely powerful in the selection of com-
411
+ plex patterns and is widely applied in the so-called atten-
412
+ tion layer first introduced by Google Brain [66]. However
413
+ the AFC layer imposes additionally physically motivated
414
+ constraints on the neural network representation.
415
+ We note that the expression for the system energy is a
416
+ sum over the fields Ei, but the local fields Ei are not addi-
417
+ tive energies, involving all the pair distances and triplets
418
+ angles within the cut-off sphere centered on particle i.
419
+ This non-additive feature favours the NN ability to cap-
420
+ ture higher order correlations (multi-body contribution
421
+ to the energy), and has been shown to outperform ad-
422
+ ditive models in complex datasets [67].
423
+ The NN non-
424
+ additivity requires the derivative of the whole energy E
425
+ (as opposed to Ei) to estimate the force on a particle i. In
426
+ this way, contributions to the force on particle i come not
427
+ only from the descriptors of i but also from the descrip-
428
+ tors of all particles who have i as a neighbour, de facto
429
+ enlarging the effective region in space where interaction
430
+ between particles are included. This allows the network
431
+ to include contributions from length-scales larger than
432
+ the cutoffs that define the atomic descriptors. The Ap-
433
+ pendix A provides further information on this point.
434
+ B.
435
+ Hidden layers
436
+ We employ a standard feed-forward fully-connected
437
+ neural network composed of two hidden layers with 25
438
+ nodes per layer and using the hyperbolic tangent (tanh)
439
+ as the activation function. The nodes of the first hidden
440
+ layer are fully connected to the ones in the second layer,
441
+ and these connections have associated weights W which
442
+ are optimized during the training stage.
443
+ The input of the first hidden layer is given by the AFC
444
+ layer where we used five nodes for the two-body AFs
445
+ (Eq. 3) and five nodes for the three-body AFs (Eq. 4)
446
+ for a total of 10 AFs for each atom.
447
+ We explore the
448
+ performance of some combinations for the number of two-
449
+ body and three-body AF in Appendix D and we find that
450
+ the choice of five and five is the more efficient.
451
+ The output is the local field Ei, for each atomic envi-
452
+ ronment i, whose sum E = οΏ½N
453
+ i=1 Ei represents the NN
454
+ estimate of the potential energy E of the whole system.
455
+ C.
456
+ Loss function and training strategy
457
+ To train the NN-potential we minimize a loss function
458
+ computed over nf frames, i.e. the number of independent
459
+ configurations extracted from an equilibrium simulation
460
+ of the liquid phase of the target potential (in our case
461
+ the mW potential). The loss function is the sum of two
462
+ contributions.
463
+ The first contribution, H[{βˆ†Ο΅k, βˆ†f k
464
+ iΞ½}], expresses the
465
+ difference in each frame k between the NN estimates and
466
+ the target values for both the total potential energy (nor-
467
+ malized by total number of atoms) Ο΅k and the atomic
468
+
469
+ 5
470
+ forces f k
471
+ iΞ½ acting in direction Ξ½ on atom i. The nf energy
472
+ Ο΅k values and 3Nnf force f k
473
+ iΞ½ values are combined in the
474
+ following expression
475
+ H[{βˆ†Ο΅k, βˆ†f k
476
+ iΞ½}] = pe
477
+ nf
478
+ nf
479
+ οΏ½
480
+ k=1
481
+ hHuber(βˆ†Ο΅k) +
482
+ pf
483
+ 3Nnf
484
+ nf
485
+ οΏ½
486
+ k=1
487
+ N
488
+ οΏ½
489
+ i=1
490
+ 3
491
+ οΏ½
492
+ Ξ½=1
493
+ hHuber(βˆ†f k
494
+ iΞ½)
495
+ (8)
496
+ where pe = 0.1 and pf = 1 control the relative contri-
497
+ bution of the energy and the forces to the loss function,
498
+ and hHuber(x) is the so-called Huber function
499
+ hHuber(x) =
500
+ οΏ½
501
+ 0.5x2 if |x| ≀ 1
502
+ 0.5 + (|x| βˆ’ 1) if |x| > 1
503
+ (9)
504
+ pe and pf are hyper-parameters of the model, and we se-
505
+ lected them with some preliminary tests that found those
506
+ values to be near the optimal ones.
507
+ The Huber func-
508
+ tion [68] is an optimal choice whenever the exploration
509
+ of the loss function goes through large errors caused by
510
+ outliers, i.e. data points that differ significantly from pre-
511
+ vious inputs. Indeed when a large deviation between the
512
+ model and data occur, a mean square error minimization
513
+ may gives rise to an anomalous trajectory in parameters
514
+ space, largely affecting the stability of the training pro-
515
+ cedure. This may happen especially in the first part of
516
+ the training procedure when the parameter optimization,
517
+ relaxing both on the energy and forces error surfaces may
518
+ experience some instabilities.
519
+ The second contribution to the loss function is a reg-
520
+ ularization function, R[{Ξ±l, Ξ²m, Ξ³m, Ξ΄m}], that serves to
521
+ limit the range of positive values of Ξ±l and of the triplets
522
+ βm, γm, δm (where the indexes l and m run over the five
523
+ different values of Ξ± and five different triplets of values
524
+ for Ξ², Ξ³ and Ξ΄) in the window βˆ’βˆž to 5. To this aim we
525
+ select the commonly used relu function
526
+ rrelu(x) =
527
+ οΏ½
528
+ x βˆ’ 5
529
+ if x > 5
530
+ 0
531
+ if x ≀ 5
532
+ (10)
533
+ (11)
534
+ and write
535
+ R[{Ξ±l, Ξ²m, Ξ³m, Ξ΄m}] =
536
+ 5
537
+ οΏ½
538
+ l=1
539
+ rrelu(Ξ±l) +
540
+ 5
541
+ οΏ½
542
+ m=1
543
+ [rrelu(Ξ²m) + rrelu(Ξ³m) + rrelu(Ξ΄m)]
544
+ (12)
545
+ Thus, the R function is activated whenever one param-
546
+ eters of the AFC layer becomes, during the minimization,
547
+ larger than 5.
548
+ To summarize, the global loss function L used in the
549
+ training of the NN is
550
+ L[Ο΅, f] = H[{βˆ†Ο΅k, βˆ†f k
551
+ iΞ½}] + pbR[{Ξ±l, Ξ²m, Ξ³m, Ξ΄m}] (13)
552
+ where pb = 1 weights the relative contribution of R com-
553
+ pared to H.
554
+ Compared to a standard NN-potential, we train not
555
+ only the network weights W but also the AFs param-
556
+ eters Ξ£ ≑ {Ξ±l, Ξ²m, Ξ³m, Ξ΄m} at the same time. The si-
557
+ multaneous optimization of the weights W and AFs Ξ£
558
+ prevents possible bottleneck in the optimisation of W at
559
+ fixed representation of Σ. Other NN potential approaches
560
+ implement a separate initial procedure to optimise the Ξ£
561
+ parameters followed by the optimisation of W at fixed
562
+ Ξ£ [69]. The two-step procedure not only requires a spe-
563
+ cific methodological choice for optimising Σ, but also may
564
+ not result in the optimal values, compared to a search in
565
+ the full parameter space (i.e. both Ξ£ and W). Since the
566
+ complexity of the loss function has increased, we have
567
+ investigated in some detail some efficient strategies that
568
+ lead to a fast and accurate training. Firstly, we initial-
569
+ ize the parameters W via the Xavier algorithm, in which
570
+ the weights are extracted from a random uniform distri-
571
+ bution [70].
572
+ To initialize the Ξ£ parameters we used a
573
+ uniform distribution in interval [βˆ’5, 5]. We then mini-
574
+ mize the loss function using the warm restart procedure
575
+ proposed in reference [71]. In this procedure, the learn-
576
+ ing rate Ξ· is reinitialized at every cycle l and inside each
577
+ cycle it decays as a function of the number of training
578
+ steps t following
579
+ Ξ·(l)(t) = Al
580
+ οΏ½(1 βˆ’ ΞΎf)
581
+ 2
582
+ οΏ½
583
+ 1 + cos
584
+ οΏ½Ο€t
585
+ Tl
586
+ οΏ½οΏ½
587
+ + ΞΎf
588
+ οΏ½
589
+ (14)
590
+ 0 ≀ t ≀ Tl
591
+ where ΞΎf = 10βˆ’7, Al = Ξ·0ΞΎl
592
+ 0 is the initial learning rate of
593
+ the l-th cycle with Ξ·0 = 0.01 and ΞΎ0 = 0.9, Tl = bΟ„ l is
594
+ the period of the l-th cycle with Ο„ = 1.4 and b = 40. The
595
+ absolute number of training steps n during cycle l can be
596
+ calculated summing over the length of all previous cycles
597
+ as n = Ο„ + οΏ½lβˆ’1
598
+ m=0 Tm.
599
+ We also select to evaluate the loss function for groups
600
+ of four frames (mini-batch) and we randomly select 200
601
+ frames nf = 200 for a system of 1000 atoms and hence
602
+ we split this dataset in 160 frames (%80) for the training
603
+ set and the 40 frames (%20) for the test set.
604
+ In Fig. 2(A) we represent the typical decay of the learn-
605
+ ing rate of the warm restart procedure, which will be
606
+ compared to the standard exponential decay protocol in
607
+ the Results section.
608
+ D.
609
+ The Target Model
610
+ To test the quality of the proposed novel NN we train
611
+ the NN with data produced with the mW [40] model
612
+
613
+ 6
614
+ 100
615
+ 101
616
+ 102
617
+ 103
618
+ 104
619
+ 105
620
+ n
621
+ 0.000
622
+ 0.005
623
+ 0.010
624
+ A
625
+ 0
626
+ 1000
627
+ 2000
628
+ 3000
629
+ ne
630
+ 10
631
+ 1
632
+ 100
633
+ B
634
+ Validation Loss
635
+ Training Loss
636
+ 0
637
+ 1000
638
+ 2000
639
+ 3000
640
+ ne
641
+ 10
642
+ 2
643
+ 10
644
+ 1
645
+ 100
646
+ 101
647
+ (kcal mol
648
+ 1)
649
+ C
650
+ 0
651
+ 1000
652
+ 2000
653
+ 3000
654
+ ne
655
+ 101
656
+ f (kcal mol
657
+ 1 nm
658
+ 1)
659
+ D
660
+ FIG. 2:
661
+ Model convergence properties:
662
+ (A) Learning rate
663
+ schedule (Eq. 14) as a function of the absolute training step n
664
+ (one step is defined as an update of the network parameters).
665
+ (B) The training and validation loss (see L[Ο΅, f] in Eq. 13) evo-
666
+ lution during the training procedure, reported as a function
667
+ of the number of epoch ne (an epoch is defined as a complete
668
+ evaluation of the training dataset). Root mean square (RMS)
669
+ error of the total potential energy per particle (C) and of the
670
+ force cartesian components (D) during the training evaluated
671
+ in the test dataset. Data in panels B-C-D refers to the NN3
672
+ model and the green point shows the best model location.
673
+ of water.
674
+ This potential, a re-parametrization of the
675
+ Stillinger-Weber model for silicon [72], uses a combina-
676
+ tion of pairwise functions complemented with an additive
677
+ three-body potential term
678
+ 0
679
+ 20
680
+ 40
681
+ 60
682
+ Seed
683
+ 0.01
684
+ 0.02
685
+ 0.03
686
+ 0.04
687
+ 0.05
688
+ 0.06
689
+ 0.07
690
+ (kcal mol
691
+ 1)
692
+ A
693
+ Exponential
694
+ Warm restart
695
+ 0
696
+ 20
697
+ 40
698
+ 60
699
+ Seed
700
+ 1.4
701
+ 1.7
702
+ 2.0
703
+ 2.3
704
+ 2.6
705
+ 2.9
706
+ 3.2
707
+ 3.5
708
+ 3.8
709
+ f (kcal mol
710
+ 1 nm
711
+ 1)
712
+ B
713
+ FIG. 3: Comparison of the root mean square error calculated
714
+ on the validation set for 60 replicas differing in the initial
715
+ seed of the training procedure using both an exponential de-
716
+ cay of the learning rate (points) and the warm restart method
717
+ (squares), for the energy (panel A) and for the forces (panel
718
+ B). For the forces, a significant improvement both in the av-
719
+ erage error and in its variance is found for the warm restart
720
+ schedule.
721
+ E =
722
+ οΏ½
723
+ i
724
+ οΏ½
725
+ j>i
726
+ U2(rij)+Ξ»
727
+ οΏ½
728
+ i
729
+ οΏ½
730
+ jΜΈ=i
731
+ οΏ½
732
+ j>k
733
+ U3 (rij, rik, ΞΈjik) (15)
734
+ where the two body contribution between two particles
735
+ i and j at relative distance rij is a generalized Lennard-
736
+ Jones potential
737
+ U2 (rij) = AΟ΅
738
+ οΏ½
739
+ B
740
+ οΏ½ Οƒ
741
+ rij
742
+ οΏ½p
743
+ βˆ’
744
+ οΏ½ Οƒ
745
+ rij
746
+ οΏ½qοΏ½
747
+ exp
748
+ οΏ½
749
+ Οƒ
750
+ rij βˆ’ aΟƒ
751
+ οΏ½
752
+ (16)
753
+ where the p = 12 and q = 6 powers are substituted
754
+ by q = 0 and p = 4, multiplied by an exponential cut-
755
+ off that brings the potential to zero at aσ, with a =
756
+ 1.8 and Οƒ = 2.3925 ˚A. AΟ΅ (with A = 7.049556277 and
757
+ Ο΅ = 6.189 kcal molβˆ’1) controls the strength of the two
758
+ body part. B controls the two-body repulsion (with B =
759
+ 0.6022245584).
760
+ The three body contribution is computed from all pos-
761
+ sible ordered triplets formed by the central particle with
762
+ the interacting neighbors (with the same cut-off aσ as the
763
+ two-body term) and favours the tetrahedral coordination
764
+ of the atoms via the following functional form
765
+ U3 (rij, rik, ΞΈjik) = Ο΅ [cos (ΞΈjik) βˆ’ cos (ΞΈ0)]2 Γ—
766
+ exp
767
+ οΏ½
768
+ Ξ³Οƒ
769
+ rij βˆ’ aΟƒ
770
+ οΏ½
771
+ exp
772
+ οΏ½
773
+ Ξ³Οƒ
774
+ rik βˆ’ aΟƒ
775
+ οΏ½
776
+ (17)
777
+ where ΞΈjik is the angle formed in the triplet jik and
778
+ Ξ³ = 1.2 controls the smoothness of the cut-off function
779
+ on approaching the cut-off. Finally, ΞΈ0 = 109.47β—¦ and
780
+ Ξ» = 23.15 controls the strength of the angular part of
781
+ the potential.
782
+
783
+ 7
784
+ The mW model, with its three-body terms centered
785
+ around a specific angle and non-monotonic radial interac-
786
+ tions, is based on a functional form which is quite differ-
787
+ ent from the radial and angular descriptors selected in the
788
+ NN model. The NN is thus agnostic with respect to the
789
+ functional form that describes the physical system (the
790
+ mW in this case). But having a reference model with ex-
791
+ plicit three body contributions offers a more challenging
792
+ target for the NN potential compared to potential models
793
+ built entirely from pairwise interactions. The mW model
794
+ is thus an excellent candidate to test the performance of
795
+ the proposed NN potential.
796
+ III.
797
+ RESULTS
798
+ A.
799
+ Training
800
+ We study two different NN models, indicated with the
801
+ labels NN1 and NN3, differing in the number of state
802
+ points included in the training set. These two models are
803
+ built with a cut-off of Rc = 4.545 ˚A
804
+ for the two-body
805
+ atomic descriptors and a cut-off of Rβ€²
806
+ c = 4.306 ˚A
807
+ for
808
+ the three-body atomic descriptors. Rβ€²
809
+ c is the same as the
810
+ mW cutoff while Rc was made slightly larger to miti-
811
+ gate the suppression of information at the boundaries by
812
+ the cutoff functions.
813
+ The NN1 model uses only train-
814
+ ing information based on mW equilibrium configurations
815
+ from one state point at ρ1 = 1.07 g cmβˆ’3, T1 = 270.9 K
816
+ where the stable phase is the liquid.
817
+ The NN3 model
818
+ uses training information based on mW liquid configura-
819
+ tions in three different state points, two state points at
820
+ ρ1 = 0.92 g cmβˆ’3, T1 = 221.1 K and ρ2 = 0.92 g cmβˆ’3,
821
+ T2 = 270.9 K where the stable solid phase is the clathrate
822
+ Si34/Si136 [73] and one state point at ρ3 = 1.15 g cmβˆ’3,
823
+ T2 = 270.9 K.
824
+ This choice of points in the phase diagram is aimed to
825
+ improve agreement with the low temperature-low density
826
+ as well as high density regions of the phase diagram. Im-
827
+ portantly, all configurations come from either stable or
828
+ metastable liquid state configurations. Indeed, the point
829
+ at ρ2 = 0.92 g cmβˆ’3, T2 = 270.9 K is quite close to
830
+ the limit of stability (respect to cavitation) of the liquid
831
+ state.
832
+ To generate the training set, we simulate a system of
833
+ N = 1000 mW particles with a standard molecular dy-
834
+ namics code in the NVT ensemble, where we use a time
835
+ step of 4 fs and run 107 steps for each state point. From
836
+ these trajectory, we randomly select 200 configurations
837
+ (frames) to create a dataset of positions, total energies
838
+ and forces. We then split the dataset in the training and
839
+ in the test data sets, the first one containing 80% of the
840
+ data. We then run the training for 4000 epochs with a
841
+ minibatch of 4 frames. At the end of every epoch, we
842
+ check if the validation loss is improved and we save the
843
+ model parameters. In Fig. 2 we plot the loss function for
844
+ the training and test datasets (B), the root mean square
845
+ error of the total energy per particle (C), and of the force
846
+ (D) for the NN3 model. The results show that the learn-
847
+ ing rate schedule of Eq. 14 is very effective in reducing
848
+ both the loss and error functions.
849
+ Interestingly, the neural network seems to avoid over-
850
+ fitting (i.e. the validation loss is decreasing at the same
851
+ rate as the loss on the training data), and the best model
852
+ (deepest local minimum explored), in a given window of
853
+ training steps, is always found at the end of that win-
854
+ dow, which also indicates that the accuracy could be fur-
855
+ ther improved by running more training steps. Indeed we
856
+ found that by increasing the number of training steps by
857
+ one order of magnitude the error in the forces decreases
858
+ by a further 30%. Similar accuracy of the training stage
859
+ is obtained also for the NN1 model (not shown).
860
+ The training procedure always terminates with an
861
+ error
862
+ on
863
+ the
864
+ test
865
+ set
866
+ equal
867
+ or
868
+ less
869
+ than
870
+ βˆ†Ο΅
871
+ ≃
872
+ 0.01 kcal molβˆ’1 (0.43 meV) for the energy, and of βˆ†f ≃
873
+ 1.55 kcal molβˆ’1 nmβˆ’1 (6.72 meV ˚Aβˆ’1) for the forces.
874
+ These values are comparable to the state-of-the-art NN
875
+ potentials [23, 54, 55, 61], and within the typical accuracy
876
+ of DFT calculations [74].
877
+ We can compare the precision of our model with that
878
+ of alternative NN potentials trained on a range of water
879
+ models. An alternative mW neural network potential has
880
+ been trained on a dataset made of 1991 configurations
881
+ of 128 particles system at different pressure and tem-
882
+ perature (including both liquid and ice structures) with
883
+ Behler-Parinello symmetry functions [24].
884
+ The train-
885
+ ing of this model (which uses more atomic fingerprints
886
+ and a larger cutoff radius) converged to an error in en-
887
+ ergy of βˆ†Ο΅ ≃ 0.0062 kcal molβˆ’1 (0.27 meV), and βˆ†f ≃
888
+ 3.46 kcal molβˆ’1 nmβˆ’1 (15.70 meV ˚Aβˆ’1) for the forces. In
889
+ a recent work searching for liquid-liquid transition signa-
890
+ tures in an ab-initio water NN model [55], a dataset of
891
+ configurations spanning a temperature range of 0βˆ’600 K
892
+ and a pressure range of 0 βˆ’ 50 GPa was selected. For
893
+ a system of 192 particles, the training converged to an
894
+ error in energy of βˆ†Ο΅ ≃ 0.010 kcal molβˆ’1 (0.46 meV),
895
+ and βˆ†f
896
+ ≃
897
+ 9.96 kcal molβˆ’1 nmβˆ’1 (43.2 meV ˚Aβˆ’1)
898
+ for the forces.
899
+ In the NN model of MB-POL [61],
900
+ a dataset spanning a temperature range from 198 K
901
+ to 368 K at ambient pressure was selected.
902
+ In this
903
+ case, for a system of 256 water molecule, an accu-
904
+ racy of βˆ†Ο΅ ≃ 0.01 kcal molβˆ’1 (0.43 meV) and βˆ†f ≃
905
+ 10 kcal molβˆ’1 nmβˆ’1 (43.36 meV ˚Aβˆ’1) was reached. Fi-
906
+ nally, the NN for water at T = 300 K used in Ref. [54],
907
+ reached precisions of βˆ†Ο΅ ≃ 0.046 kcal molβˆ’1 (2 meV) and
908
+ βˆ†f ≃ 25.36 kcal molβˆ’1 nmβˆ’1 (110 meV ˚Aβˆ’1).
909
+ While a direct comparison between NN potentials
910
+ trained on different reference potentials is not a valid test
911
+ to rank the respective accuracies, the comparisons above
912
+ show that our NN potential reaches a similar precision
913
+ in energies, and possibly an improved error in the force
914
+ estimation.
915
+ The accuracy of the NN potential could be further
916
+ improved by extending the size of the dataset and the
917
+ choice of the state points. In fact, while the datasets in
918
+ Ref. [54, 55, 61] have been built with optimized proce-
919
+
920
+ 8
921
+ dures, the dataset used in this study was prepared by
922
+ sampling just one (NN1) or three (NN3) state-points.
923
+ Also the size of the datasets used in the present work is
924
+ smaller or comparable to the ones of Ref. [54, 55, 61].
925
+ In Fig. 3 we compare the error in the energies (A)
926
+ and the forces (B) between sixty independent training
927
+ runs using the standard exponential decay of the learn-
928
+ ing rate (points) and the warm restart protocol (squares).
929
+ The figure shows that while the errors in the energy com-
930
+ putations are comparable between the two methods, the
931
+ warm restart protocol allows the forces to be computed
932
+ with higher accuracy. Moreover we found that the warm
933
+ restart procedure is less dependent on the initial seed and
934
+ that it reaches deeper basins than the standard exponen-
935
+ tial cooling rate.
936
+ B.
937
+ Comparing NN1 with NN3
938
+ The NN potential model was implemented in a custom
939
+ MD code that makes use of the tensorflow C API [75].
940
+ We adopted the same time step (4 fs), the same number
941
+ of particles (N = 1000) and the same number of steps
942
+ (107) as for the simulations in the mW model.
943
+ As described in the Training Section, we compare the
944
+ accuracy of two different training strategies: NN1 which
945
+ was trained on a single state point, and NN3 which is
946
+ instead trained on three different state point. In Fig. 4
947
+ we plot the energy error (βˆ†Ο΅) between the NN potential
948
+ and the mW model with both NN1 (panel A) and NN3
949
+ (panel B). Starting from NN1, we see that the model
950
+ already provides an excellent accuracy for a large range
951
+ of temperatures and for densities close to the training
952
+ density. The biggest shortcoming of the NN1 model is
953
+ at densities lower than the trained density, where the
954
+ NN potential model cavitates and does not retain the
955
+ long-lived metastable liquid state displayed by the mW
956
+ model. We speculate that this behaviour is due to the
957
+ absence of low density configurations in the training set,
958
+ which prevents the NN potential model from correctly
959
+ reproducing the attractive tails of the mW potential.
960
+ To overcome this limitation we have included two ad-
961
+ ditional state points at low density in the NN3 model. In
962
+ this case, Fig. 4B shows that NN3 provides a quite ac-
963
+ curate reproduction of the energy in the entire explored
964
+ density and temperature window (despite being trained
965
+ only with data at ρ = 0.92 g cmβˆ’3 and ρ = 1.15 g cmβˆ’3).
966
+ We can also compare the accuracy obtained during
967
+ production runs against the accuracy reached during
968
+ training, which was βˆ†Ο΅ ≃ 0.01 kcal molβˆ’1. Fig. 4B shows
969
+ the error is of the order of 0.032 kcal molβˆ’1 (1.3 meV), for
970
+ density above the training set density. But in the density
971
+ region between 0.92 and 1.15, the error is even smaller,
972
+ around 0.017 kcal molβˆ’1 (0.7 meV) at the lowest density
973
+ boundary.
974
+ We can thus conclude that the NN3 model, which adds
975
+ to the NN1 model information at lower density and tem-
976
+ perature, in the region where tetrahedality in the wa-
977
+ 0.9
978
+ 1.0
979
+ 1.1
980
+ 1.2
981
+ (g cm
982
+ 3)
983
+ 385
984
+ 365
985
+ 345
986
+ 325
987
+ 305
988
+ 285
989
+ 265
990
+ 245
991
+ 225
992
+ T (K)
993
+ A
994
+ 0.9
995
+ 1.0
996
+ 1.1
997
+ 1.2
998
+ (g cm
999
+ 3)
1000
+ B
1001
+ 0.001
1002
+ 0.010
1003
+ 0.020
1004
+ 0.030
1005
+ 0.040
1006
+ (kcal mol
1007
+ 1)
1008
+ FIG. 4: Comparison between the mW total energy and the
1009
+ NN1 model (A) and NN3 model (B) for different temperatures
1010
+ and densities. While the NN3 model is able to reproduce the
1011
+ mW total energy with a good agreement in a wide region of
1012
+ densities and temperatures, the NN1 provide a good repre-
1013
+ sentation only in a limited region of density and temperature
1014
+ values. Blue squares represent the state points used for build-
1015
+ ing the NN models.
1016
+ ter structure is enhanced, is indeed capable to represent,
1017
+ with only three state points, a quite large region of the
1018
+ phase space, encompassing dense and stretched liquid
1019
+ states. This suggests that a training based on few state
1020
+ points at the boundary of the density/temperature re-
1021
+ gion which needs to be studied is sufficient to produce a
1022
+ high quality NN model. In the following we focus entirely
1023
+ on the NN3 model.
1024
+ C.
1025
+ Comparison of thermodynamic, structural and
1026
+ dynamical quantities
1027
+ In Fig. 5 we present a comparison of thermodynamic
1028
+ data between the mW model (squares) and its NN poten-
1029
+ tial representation (points) across a wide range of state
1030
+ points. Fig. 5A plots the energy as function of density
1031
+ for temperatures ranging from melting to deeply super-
1032
+ cooled conditions. Perhaps the most interesting result is
1033
+ that the NN potential is able to capture the energy min-
1034
+ imum, also called the optimal network forming density,
1035
+ which is a distinctive anomalous property of water and
1036
+ other empty liquids [76].
1037
+ Fig. 5(B) shows the pressure as a function of the tem-
1038
+ perature for different densities, comparing the mW with
1039
+ the NN3 model. Also the pressure shows a good agree-
1040
+ ment between the two models in the region of densities
1041
+ between ρ = 0.92 g cmβˆ’3 and ρ = 1.15 g cmβˆ’3, which, as
1042
+ for the energy, tends to deteriorate at ρ = 1.22 g cmβˆ’3.
1043
+ In the large density region explored, the structure of
1044
+ the liquid changes considerably. On increasing density, a
1045
+ transition from tetrahedral coordinated local structure,
1046
+ prevalent at low T and low ρ, towards denser local envi-
1047
+
1048
+ 9
1049
+ 0.85
1050
+ 0.90
1051
+ 0.95
1052
+ 1.00
1053
+ 1.05
1054
+ 1.10
1055
+ 1.15
1056
+ 1.20
1057
+ 1.25
1058
+ (g cm
1059
+ 3)
1060
+ 10.0
1061
+ 9.5
1062
+ 9.0
1063
+ 8.5
1064
+ (kcal mol
1065
+ 1)
1066
+ 221.1K
1067
+ 233.6K
1068
+ 246.0K
1069
+ 258.5K
1070
+ 271.0K
1071
+ 299.0K
1072
+ 311.4K
1073
+ 373.7K
1074
+ A
1075
+ mW
1076
+ NN3
1077
+ 200
1078
+ 225
1079
+ 250
1080
+ 275
1081
+ 300
1082
+ 325
1083
+ 350
1084
+ 375
1085
+ T (K)
1086
+ 0
1087
+ 1
1088
+ 2
1089
+ 3
1090
+ 4
1091
+ P (GPa)
1092
+ 0.92 g cm
1093
+ 3
1094
+ 0.99 g cm
1095
+ 3
1096
+ 1.07 g cm
1097
+ 3
1098
+ 1.15 g cm
1099
+ 3
1100
+ 1.19 g cm
1101
+ 3
1102
+ 1.22 g cm
1103
+ 3
1104
+ B
1105
+ FIG. 5: Comparison between the mW total energy and the
1106
+ NN3 total energy as a function of density along different
1107
+ isotherm (A) and comparison between the mW pressure and
1108
+ the NN3 pressure as a function of temperature along differ-
1109
+ ent isochores (B). The relative error of the NN vs the mW
1110
+ potential grows with density, but remains within 3% even for
1111
+ densities larger than the densities used in the training set.
1112
+ ronments with interstitial molecules included in the first
1113
+ coordination shell takes place. This structural change is
1114
+ well displayed in the radial distribution function, shown
1115
+ for different densities at fixed temperature in Fig. 6.
1116
+ Fig. 6 also shows the progressive onset of a peak around
1117
+ 3.5 ˚A
1118
+ developing on increasing pressure, which signals
1119
+ the growth of interstitial molecules, coexisting with open
1120
+ tetrahedral local structures [77, 78]. At the highest den-
1121
+ sity, the tetrahedral peak completely merges with the
1122
+ interstitial peak. The NN3 model reproduces quite ac-
1123
+ curately all features of the radial distribution functions,
1124
+ maxima and minima positions and their relative ampli-
1125
+ tudes, at all densities, from the tetrahedral-dominated to
1126
+ the interstitial-dominated limits. In general, NN3 model
1127
+ reproduces quite well the mW potential in energies, pres-
1128
+ sure and structures and it appreciably deviates from mW
1129
+ pressures and energies quantities only at densities (above
1130
+ 1.15 g/cm3) which are outside of the training region.
1131
+ To assess the ability of NN potential to correctly de-
1132
+ scribe also the crystal phases of the mW potential, we
1133
+ compare in Fig. 7 the g(r) of mW with the g(r) of the
1134
+ NN3 model for four different stable solid phases [73]:
1135
+ hexagonal and cubic ice (ρ = 1.00 g cmβˆ’3 and T =
1136
+ 246 K), the dense crystal SC16 (ρ = 1.20 g cmβˆ’3 and T =
1137
+ 234 K) and the clathrate phase Si136 (ρ = 0.80 g cmβˆ’3
1138
+ and T = 221 K). The results, shown in Fig. 7, show that,
1139
+ despite no crystal configurations have been included in
1140
+ the training set, a quite accurate representation of the
1141
+ crystal structure at finite temperature is provided by the
1142
+ NN3 model for all distinct sampled lattices.
1143
+ 0
1144
+ 2
1145
+ 4
1146
+ 6
1147
+ 8
1148
+ 10
1149
+ 12
1150
+ 14
1151
+ 16
1152
+ R (Γ…)
1153
+ 0
1154
+ 1
1155
+ 2
1156
+ 3
1157
+ 4
1158
+ 5
1159
+ 6
1160
+ 7
1161
+ 8
1162
+ RDF
1163
+ 0.92 g cm
1164
+ 3
1165
+ 0.99 g cm
1166
+ 3
1167
+ 1.15 g cm
1168
+ 3
1169
+ 1.22 g cm
1170
+ 3
1171
+ mW
1172
+ NN3
1173
+ FIG. 6:
1174
+ Comparison between the mW radial distribution
1175
+ functions g(r) and the NN3 g(r) at T = 270.9 K for four
1176
+ different densities.
1177
+ The tetrahedral structure (signalled by
1178
+ the peak at 4.54 ˚A ) progressively weakens in favour of an in-
1179
+ terstitial peak progressively growing at 3.5 βˆ’ 3.8 ˚A . Different
1180
+ g(r) have been progressively shifted by two to improve clarity.
1181
+ 0
1182
+ 5
1183
+ 10
1184
+ 15
1185
+ R (Γ…)
1186
+ 0.0
1187
+ 2.5
1188
+ 5.0
1189
+ 7.5
1190
+ 10.0
1191
+ 12.5
1192
+ 15.0
1193
+ 17.5
1194
+ RDF
1195
+ Hexagonal diamond
1196
+ Cubic diamond
1197
+ Si136 Clathrate
1198
+ SC16
1199
+ mW
1200
+ NN3
1201
+ FIG. 7:
1202
+ Comparison between the mW radial distribution
1203
+ functions g(r) and the NN3 g(r) for four different lattices:
1204
+ (A) hexagonal diamond (the oxygen positions of the ice Ih);
1205
+ (B) cubic diamond (the oxygen positions of the ice Ic; (C)
1206
+ the SC16 crystal (the dense crystal form stable at large pres-
1207
+ sures in the mW model) and (D) the Si136 clathrate structure,
1208
+ which is stable at negative pressures in the mW model. Dif-
1209
+ ferent g(r) have been progressively shifted by four to improve
1210
+ clarity.
1211
+ Finally, we compare in Fig. 8 the diffusion coefficient
1212
+ (evaluated from the long time limit of the mean square
1213
+ displacement) for the mW and the NN3 model, in a wide
1214
+ range of temperatures and densities, where water displays
1215
+ a diffusion anomaly.
1216
+ Fig. 8 shows again that, also for
1217
+
1218
+ 10
1219
+ 0.85
1220
+ 0.90
1221
+ 0.95
1222
+ 1.00
1223
+ 1.05
1224
+ 1.10
1225
+ 1.15
1226
+ 1.20
1227
+ (g cm
1228
+ 3)
1229
+ 0
1230
+ 100
1231
+ 200
1232
+ 300
1233
+ 400
1234
+ 500
1235
+ 600
1236
+ 700
1237
+ 800
1238
+ D (Γ…2 ns
1239
+ 1)
1240
+ 221.1K
1241
+ 233.6K
1242
+ 246.0K
1243
+ 258.5K
1244
+ 271.0K
1245
+ 299.0K
1246
+ 311.4K
1247
+ 373.7K
1248
+ mW
1249
+ NN3
1250
+ FIG. 8: Comparison between the mW diffusion coefficient D
1251
+ and the NN3 corresponding quantity for different tempera-
1252
+ tures and densities, in the interval 221 βˆ’ 271 K. In this dy-
1253
+ namic quantity, the relative error is, for all temperatures,
1254
+ around 8%. Note also that in this T window the diffusion
1255
+ coefficient shows a clear maximum, reproducing one of the
1256
+ well-know diffusion anomaly of water. Diffusion coefficients
1257
+ have been calculated in the NVT ensemble using the same
1258
+ Andersen thermostat algorithm [79] for mW and NN3 poten-
1259
+ tial.
1260
+ dynamical quantities, the NN potential offers an excellent
1261
+ representation of the mW potential, despite the fact that
1262
+ no dynamical quantity was included in the training set. A
1263
+ comparison between fluctuations of energy and pressure
1264
+ of mW and NN3 potential is reported in Appendix B.
1265
+ IV.
1266
+ CONCLUSIONS
1267
+ In this work we have presented a novel neural net-
1268
+ work (NN) potential based on a new set of atomic fin-
1269
+ gerprints (AFs) built from two- and three-body local de-
1270
+ scriptors that are combined in a permutation-invariant
1271
+ way through an exponential filter (see Eq. 3-4). One of
1272
+ the distinctive advantages of our scheme is that the AF’s
1273
+ parameters are optimized during the training procedure,
1274
+ making the present algorithm a self-training network that
1275
+ automatically selects the best AFs for the potential of in-
1276
+ terest.
1277
+ We have shown that the added complexity in the con-
1278
+ current training of the AFs and of the NN weights can
1279
+ be overcome with an annealing procedure based on the
1280
+ warm restart method [71], where the learning rate goes
1281
+ through damped oscillatory ramps.
1282
+ This strategy not
1283
+ only gives better accuracy compared to the commonly
1284
+ implemented exponential learning rate decay, but also
1285
+ allows the training procedure to converge rapidly inde-
1286
+ pendently from the initialisation strategies of the model’s
1287
+ parameters.
1288
+ Moreover we show in Appendix C that the potential
1289
+ hyper-surface of the NN model has the same smoothness
1290
+ as the target model, as confirmed by (i) the possibility to
1291
+ use the same timestep in the NN and in the target model
1292
+ when integrating the equation of motion and (ii) by the
1293
+ possibility of simulate the NN model even in the NVE
1294
+ ensemble with proper energy conservation.
1295
+ We test the novel NN on the mW model [40], a
1296
+ one-component model system commonly used to de-
1297
+ scribe water in classical simulations. This model, a re-
1298
+ parametrization of the Stillinger-Weber model for sili-
1299
+ con [72], while treating the water molecule as a simple
1300
+ point, is able to reproduce the characteristic tetrahedral
1301
+ local structure of water (and its distortion on increasing
1302
+ density) via the use of three-body interactions. Indeed
1303
+ water changes from a liquid of tetrahedrally coordinated
1304
+ molecules to a denser liquid, in which a relevant fraction
1305
+ of interstitial molecules are present in the first nearest-
1306
+ neighbour shell. The complexity of the mW model, both
1307
+ due to its functional form as well as to the variety of dif-
1308
+ ferent local structures which characterise water, makes it
1309
+ an ideal benchmark system to test our NN potential.
1310
+ We find that a training based on configurations ex-
1311
+ tracted by three different state points is able to pro-
1312
+ vide a quite accurate representation of the mW poten-
1313
+ tial hyper-surface, when the densities and temperatures
1314
+ of the training state points delimit the region of in which
1315
+ the NN potential is expected to work. We also find that
1316
+ the error in the NN estimate of the total energy is low,
1317
+ always smaller than 0.03 kcal molβˆ’1, with a mean error
1318
+ of 0.013 kcal molβˆ’1. The NN model reproduces very well
1319
+ not only the thermodynamic properties but also struc-
1320
+ tural properties, as quantified by the radial distribution
1321
+ function, and the dynamic properties, as expressed by
1322
+ the diffusion coefficient, in the extended density interval
1323
+ from ρ = 0.92 g cmβˆ’3 to ρ = 1.22 g cmβˆ’3.
1324
+ Interestingly, we find that the NN model, trained only
1325
+ on disordered configurations, is also able to properly
1326
+ describe the radial distribution of the ordered lattices
1327
+ which characterise the mW phase diagram, encompass-
1328
+ ing the cubic and hexagonal ices, the SC16 and the Si136
1329
+ clathrate structure [73]. In this respect, the ability of the
1330
+ NN model to properly represent crystal states suggests
1331
+ that, in the case of the mW, and as such probably in the
1332
+ case of water, the geometrical information relevant to the
1333
+ ordered structures is contained in the sampling of phase
1334
+ space typical of the disordered liquid phase. These find-
1335
+ ings have been recently discussed in reference [80] where
1336
+ it has been demonstrated that liquid water contains all
1337
+ the building blocks of diverse ice phases.
1338
+ We conclude by noticing that the present approach can
1339
+ be generalized to multicomponent systems, following the
1340
+ same strategy implemented by previous approaches [18,
1341
+ 23]. Work in this direction is underway.
1342
+ Acknowledgments
1343
+ FGM and JR acknowledge support from the European
1344
+ Research Council Grant DLV-759187 and CINECA grant
1345
+ ISCRAB NNPROT.
1346
+
1347
+ 11
1348
+ [1] K. Hansen, F. Biegler, R. Ramakrishnan, W. Pronobis,
1349
+ O. A. Von Lilienfeld, K.-R. M¨uller, and A. Tkatchenko,
1350
+ The journal of physical chemistry letters 6, 2326 (2015).
1351
+ [2] S. Chmiela, A. Tkatchenko, H. E. Sauceda, I. Poltavsky,
1352
+ K. T. Sch¨utt, and K.-R. M¨uller, Science advances 3,
1353
+ e1603015 (2017).
1354
+ [3] M. Haghighatlari and J. Hachmann, Current Opinion in
1355
+ Chemical Engineering 23, 51 (2019).
1356
+ [4] A. Glielmo, C. Zeni, Β΄A. Fekete, and A. D. Vita, in Ma-
1357
+ chine Learning Meets Quantum Physics (Springer, 2020),
1358
+ pp. 67–98.
1359
+ [5] S. Manzhos and T. Carrington Jr, Chemical Reviews
1360
+ 121, 10187 (2020).
1361
+ [6] P. Gkeka, G. Stoltz, A. Barati Farimani, Z. Belkacemi,
1362
+ M. Ceriotti, J. D. Chodera, A. R. Dinner, A. L. Fergu-
1363
+ son, J.-B. Maillet, H. Minoux, et al., Journal of chemical
1364
+ theory and computation 16, 4757 (2020).
1365
+ [7] Z. L. Glick, D. P. Metcalf, A. Koutsoukas, S. A. Spronk,
1366
+ D. L. Cheney, and C. D. Sherrill, The Journal of Chem-
1367
+ ical Physics 153, 044112 (2020).
1368
+ [8] M. Benoit, J. Amodeo, S. Combettes, I. Khaled, A. Roux,
1369
+ and J. Lam, Machine Learning: Science and Technology
1370
+ 2, 025003 (2020).
1371
+ [9] M. Dijkstra and E. Luijten, Nature materials 20, 762
1372
+ (2021).
1373
+ [10] O. T. Unke, S. Chmiela, H. E. Sauceda, M. Gastegger,
1374
+ I. Poltavsky, K. T. Sch¨utt, A. Tkatchenko, and K.-R.
1375
+ M¨uller, Chemical Reviews 121, 10142 (2021).
1376
+ [11] G. Campos-Villalobos, E. Boattini, L. Filion, and M. Di-
1377
+ jkstra, The Journal of Chemical Physics 155, 174902
1378
+ (2021).
1379
+ [12] F. Musil, A. Grisafi, A. P. Bart´ok, C. Ortner, G. Cs´anyi,
1380
+ and M. Ceriotti, Chemical Reviews 121, 9759 (2021).
1381
+ [13] G. Campos-Villalobos, G. Giunta, S. MarΒ΄Δ±n-Aguilar, and
1382
+ M. Dijkstra, The Journal of Chemical Physics 157,
1383
+ 024902 (2022).
1384
+ [14] J. Goniakowski, S. Menon, G. Laurens, and J. Lam, The
1385
+ Journal of Physical Chemistry C 126, 17456 (2022).
1386
+ [15] A. Glielmo, C. Zeni, B. Cheng, G. CsΒ΄anyi, and A. Laio,
1387
+ PNAS Nexus 1, pgac039 (2022).
1388
+ [16] S. Batzner, A. Musaelian, L. Sun, M. Geiger, J. P.
1389
+ Mailoa, M. Kornbluth, N. Molinari, T. E. Smidt, and
1390
+ B. Kozinsky, Nature communications 13, 1 (2022).
1391
+ [17] G. Tallec, G. Laurens, O. Fresse-Colson, and J. Lam,
1392
+ Quantum Chemistry in the Age of Machine Learning pp.
1393
+ 253–277 (2023).
1394
+ [18] J. Behler and M. Parrinello, Physical review letters 98,
1395
+ 146401 (2007).
1396
+ [19] J. Behler, Journal of Physics:
1397
+ Condensed Matter 26,
1398
+ 183001 (2014).
1399
+ [20] J. S. Smith, O. Isayev, and A. E. Roitberg, Chemical
1400
+ science 8, 3192 (2017).
1401
+ [21] J. S. Smith, O. Isayev, and A. E. Roitberg, Scientific data
1402
+ 4, 1 (2017).
1403
+ [22] K. T. Sch¨utt, F. Arbabzadah, S. Chmiela, K. R. M¨uller,
1404
+ and A. Tkatchenko, Nature communications 8, 1 (2017).
1405
+ [23] L. Zhang, J. Han, H. Wang, W. Saidi, R. Car, et al.,
1406
+ Advances in Neural Information Processing Systems 31
1407
+ (2018).
1408
+ [24] A. Singraber, J. Behler, and C. Dellago, Journal of chem-
1409
+ ical theory and computation 15, 1827 (2019).
1410
+ [25] A. Singraber, T. Morawietz, J. Behler, and C. Dellago,
1411
+ Journal of chemical theory and computation 15, 3075
1412
+ (2019).
1413
+ [26] B. E. Husic, N. E. Charron, D. Lemm, J. Wang, A. PΒ΄erez,
1414
+ M. Majewski, A. Kr¨amer, Y. Chen, S. Olsson, G. de Fab-
1415
+ ritiis, et al., The Journal of chemical physics 153, 194101
1416
+ (2020).
1417
+ [27] B. Cheng, G. Mazzola, C. J. Pickard, and M. Ceriotti,
1418
+ Nature 585, 217 (2020).
1419
+ [28] L. Zhang, H. Wang, R. Car, and E. Weinan, Physical
1420
+ review letters 126, 236001 (2021).
1421
+ [29] D. Lu, H. Wang, M. Chen, L. Lin, R. Car, E. Weinan,
1422
+ W. Jia, and L. Zhang, Computer Physics Communica-
1423
+ tions 259, 107624 (2021).
1424
+ [30] D. Tisi, L. Zhang, R. Bertossa, H. Wang, R. Car, and
1425
+ S. Baroni, Physical Review B 104, 224202 (2021).
1426
+ [31] J. Behler, Chemical Reviews 121, 10037 (2021).
1427
+ [32] T. Zubatiuk and O. Isayev, Accounts of Chemical Re-
1428
+ search 54, 1575 (2021).
1429
+ [33] L. D. Jacobson, J. M. Stevenson, F. Ramezanghorbani,
1430
+ D. Ghoreishi, K. Leswing, E. D. Harder, and R. Abel,
1431
+ Journal of Chemical Theory and Computation 18, 2354
1432
+ (2022).
1433
+ [34] T. E. Gartner III, P. M. Piaggi, R. Car, A. Z. Pana-
1434
+ giotopoulos, and P. G. Debenedetti, arXiv preprint
1435
+ arXiv:2208.13633 (2022).
1436
+ [35] C. Malosso, L. Zhang, R. Car, S. Baroni, and D. Tisi,
1437
+ arXiv preprint arXiv:2203.01262 (2022).
1438
+ [36] T. W. Ko, J. A. Finkler, S. Goedecker, and J. Behler,
1439
+ Nature communications 12, 1 (2021).
1440
+ [37] A. P. BartΒ΄ok, R. Kondor, and G. CsΒ΄anyi, Physical Review
1441
+ B 87, 184115 (2013).
1442
+ [38] J. Nigam, S. Pozdnyakov, and M. Ceriotti, The Journal
1443
+ of Chemical Physics 153, 121101 (2020).
1444
+ [39] M. P. Bircher, A. Singraber, and C. Dellago, Machine
1445
+ Learning: Science and Technology 2, 035026 (2021).
1446
+ [40] V. Molinero and E. B. Moore, The Journal of Physical
1447
+ Chemistry B 113, 4008 (2009).
1448
+ [41] J. Russo, K. Akahane, and H. Tanaka, Proceedings of the
1449
+ National Academy of Sciences 115, E3333 (2018).
1450
+ [42] V. Holten, D. T. Limmer, V. Molinero, and M. A.
1451
+ Anisimov, The Journal of chemical physics 138, 174501
1452
+ (2013).
1453
+ [43] E. B. Moore and V. Molinero, Nature 479, 506 (2011).
1454
+ [44] M. B. Davies, M. Fitzner, and A. Michaelides, Pro-
1455
+ ceedings of the National Academy of Sciences 119,
1456
+ e2205347119 (2022).
1457
+ [45] J. Barker and R. Watts, Chemical Physics Letters 3, 144
1458
+ (1969).
1459
+ [46] A. Rahman and F. H. Stillinger, The Journal of Chemical
1460
+ Physics 55, 3336 (1971).
1461
+ [47] H. Tanaka, Journal of Non-Crystalline Solids:
1462
+ X 13,
1463
+ 100076 (2022).
1464
+ [48] P. H. Poole, F. Sciortino, U. Essmann, and H. E. Stanley,
1465
+ Nature 360, 324 (1992).
1466
+ [49] G. A. Cisneros, K. T. Wikfeldt, L. Ojam¨ae, J. Lu, Y. Xu,
1467
+ H. Torabifard, A. P. BartΒ΄ok, G. CsΒ΄anyi, V. Molinero, and
1468
+ F. Paesani, Chemical reviews 116, 7501 (2016).
1469
+ [50] P. G. Debenedetti, F. Sciortino, and G. H. Zerze, Science
1470
+ 369, 289 (2020).
1471
+ [51] K. H. Kim,
1472
+ K. Amann-Winkel,
1473
+ N. Giovambattista,
1474
+
1475
+ 12
1476
+ A. Sp¨ah, F. Perakis, H. Pathak, M. L. Parada, C. Yang,
1477
+ D. Mariedahl, T. Eklund, et al., Science 370, 978 (2020).
1478
+ [52] J. Weis, F. Sciortino, A. Z. Panagiotopoulos, and P. G.
1479
+ Debenedetti, The Journal of Chemical Physics 157,
1480
+ 024502 (2022).
1481
+ [53] T. T. Nguyen, E. SzΒ΄ekely, G. Imbalzano, J. Behler,
1482
+ G. Cs´anyi, M. Ceriotti, A. W. G¨otz, and F. Paesani,
1483
+ The Journal of chemical physics 148, 241725 (2018).
1484
+ [54] B. Cheng, E. A. Engel, J. Behler, C. Dellago, and M. Ce-
1485
+ riotti, Proceedings of the National Academy of Sciences
1486
+ 116, 1110 (2019).
1487
+ [55] T. E. Gartner III, L. Zhang, P. M. Piaggi, R. Car, A. Z.
1488
+ Panagiotopoulos, and P. G. Debenedetti, Proceedings of
1489
+ the National Academy of Sciences 117, 26040 (2020).
1490
+ [56] O. Wohlfahrt, C. Dellago, and M. Sega, The Journal of
1491
+ Chemical Physics 153, 144710 (2020).
1492
+ [57] A. Torres, L. S. Pedroza, M. Fernandez-Serra, and A. R.
1493
+ Rocha, The Journal of Physical Chemistry B 125, 10772
1494
+ (2021).
1495
+ [58] A. Reinhardt and B. Cheng, Nature communications 12,
1496
+ 1 (2021).
1497
+ [59] E. Lambros, S. Dasgupta, E. Palos, S. Swee, J. Hu, and
1498
+ F. Paesani, Journal of Chemical Theory and Computa-
1499
+ tion 17, 5635 (2021).
1500
+ [60] P.
1501
+ M.
1502
+ Piaggi,
1503
+ J.
1504
+ Weis,
1505
+ A.
1506
+ Z.
1507
+ Panagiotopoulos,
1508
+ P.
1509
+ G.
1510
+ Debenedetti,
1511
+ and
1512
+ R.
1513
+ Car,
1514
+ arXiv
1515
+ preprint
1516
+ arXiv:2203.01376 (2022).
1517
+ [61] Y. Zhai, A. Caruso, S. L. Bore, and F. Paesani (2022).
1518
+ [62] V. Babin, C. Leforestier, and F. Paesani, Journal of
1519
+ chemical theory and computation 9, 5395 (2013).
1520
+ [63] V. Babin, G. R. Medders, and F. Paesani, Journal of
1521
+ chemical theory and computation 10, 1599 (2014).
1522
+ [64] G. R. Medders, V. Babin, and F. Paesani, Journal of
1523
+ chemical theory and computation 10, 2906 (2014).
1524
+ [65] S. Yue, M. C. Muniz, M. F. Calegari Andrade, L. Zhang,
1525
+ R. Car, and A. Z. Panagiotopoulos, The Journal of
1526
+ Chemical Physics 154, 034111 (2021).
1527
+ [66] A. Vaswani,
1528
+ N. Shazeer,
1529
+ N. Parmar,
1530
+ J. Uszkoreit,
1531
+ L. Jones, A. N. Gomez, οΏ½L. Kaiser, and I. Polosukhin,
1532
+ Advances in neural information processing systems 30
1533
+ (2017).
1534
+ [67] S. N. Pozdnyakov, M. J. Willatt, A. P. BartΒ΄ok, C. Ortner,
1535
+ G. CsΒ΄anyi, and M. Ceriotti, Physical Review Letters 125,
1536
+ 166001 (2020).
1537
+ [68] P. J. Huber, The Annals of Mathematical Statistics
1538
+ 35, 73
1539
+ (1964), URL https://doi.org/10.1214/aoms/
1540
+ 1177703732.
1541
+ [69] G. Imbalzano, A. Anelli, D. GiofrΒ΄e, S. Klees, J. Behler,
1542
+ and M. Ceriotti, The Journal of chemical physics 148,
1543
+ 241730 (2018).
1544
+ [70] X. Glorot and Y. Bengio, in Proceedings of the thir-
1545
+ teenth international conference on artificial intelligence
1546
+ and statistics (JMLR Workshop and Conference Pro-
1547
+ ceedings, 2010), pp. 249–256.
1548
+ [71] I.
1549
+ Loshchilov
1550
+ and
1551
+ F.
1552
+ Hutter,
1553
+ arXiv
1554
+ preprint
1555
+ arXiv:1608.03983 (2016).
1556
+ [72] F. H. Stillinger and T. A. Weber, Physical review B 31,
1557
+ 5262 (1985).
1558
+ [73] F. Romano, J. Russo, and H. Tanaka, Physical Review
1559
+ B 90, 014204 (2014).
1560
+ [74] M. J. Gillan, D. Alfe, and A. Michaelides, The Journal
1561
+ of chemical physics 144, 130901 (2016).
1562
+ [75] TensorFlow,
1563
+ Tensorflow
1564
+ c
1565
+ 2.7,
1566
+ URL
1567
+ https://www.
1568
+ tensorflow.org/install/lang_c.
1569
+ [76] J. Russo, F. Leoni, F. Martelli, and F. Sciortino, Reports
1570
+ on Progress in Physics (2021).
1571
+ [77] R. Foffi and F. Sciortino, Physical Review Letters 127,
1572
+ 175502 (2021).
1573
+ [78] R. Foffi, J. Russo, and F. Sciortino, The Journal of Chem-
1574
+ ical Physics 154, 184506 (2021).
1575
+ [79] H. C. Andersen, The Journal of chemical physics 72, 2384
1576
+ (1980).
1577
+ [80] B. Monserrat, J. G. Brandenburg, E. A. Engel, and
1578
+ B. Cheng, Nature communications 11, 1 (2020).
1579
+ V.
1580
+ APPENDIX A
1581
+ In this appendix we discuss the effective spacial range
1582
+ covered by a NN potential whose fingerprints are defined
1583
+ based on pair information confined within a sphere of
1584
+ cutoff radius Rc.
1585
+ As noted in reference [31], multi-body potentials and
1586
+ especially non-additive multibody potentials induce lo-
1587
+ cal interactions beyond the cut-off radius, enlarging the
1588
+ sphere of interaction.
1589
+ Indeed, the force on particle i
1590
+ comes from the derivative of the local field of i and of
1591
+ all its neighbours with respect to the coordinates of par-
1592
+ ticle i.
1593
+ 2
1594
+ 1
1595
+ 3
1596
+ 2
1597
+ 1
1598
+ 3
1599
+ 4
1600
+ 5
1601
+ A
1602
+ B
1603
+ FIG. 9: (A) Two-body interactions and (B) three-body inter-
1604
+ actions in a non linear local field model Ei. The non linearity
1605
+ of the local field enlarges the interaction cut-off where a neigh-
1606
+ bour particle (blue) makes a bridge between non-neighboring
1607
+ particle (red and blue).
1608
+ Fig. 9 graphically explains the effective role of Rc in
1609
+ the NN potential.
1610
+ In panel A, we describe particle 1
1611
+ with only one neighbour (particle 2) within Rc. We also
1612
+ represent the sphere centered on particle 3, which also
1613
+ includes particle 2 as one of its neighbour. In this case,
1614
+ the energy of the system will be represented as a sum
1615
+ over the local fields E1, E2 and E3. Due to the intrinsic
1616
+ non-linearity of the NN, the field Ei mixes together the
1617
+ AFs, and consequently the distances and angles entering
1618
+ in the AFs are non-linearly mixed in Ei. The force on
1619
+ atom 1 is then written as
1620
+ f1Ξ½ = βˆ’βˆ‚E1(r12)
1621
+ βˆ‚x1Ξ½
1622
+ βˆ’ βˆ‚E2(r21, r23)
1623
+ βˆ‚x1Ξ½
1624
+ = βˆ’βˆ‚E1(r12)
1625
+ βˆ‚x1Ξ½
1626
+ βˆ’βˆ‚E2(r21, r23)
1627
+ βˆ‚r21
1628
+ βˆ‚r21
1629
+ βˆ‚x1Ξ½
1630
+ βˆ’ βˆ‚E2(r21, r23)
1631
+ βˆ‚r23
1632
+ βˆ‚r23
1633
+ βˆ‚x1Ξ½
1634
+ (A1)
1635
+ While the last term vanishes, the next to the last retains
1636
+
1637
+ 13
1638
+ A
1639
+ B
1640
+ FIG. 10: (A) Standard deviation of total energy (normal-
1641
+ ized with the number of particles) and (B) standard deviation
1642
+ of virial pressure for both NN3 model (red) and mW model
1643
+ (black).
1644
+ an intrinsic dependence on the coordinates both of par-
1645
+ ticle 2 as well as of particle 3, if the local field E2 is non
1646
+ linear. Thus, even if particle 3 is further than Rc, it en-
1647
+ ters in the determination of the force acting on particle
1648
+ 1. A similar effect is also present in the angular part of
1649
+ the AFs, as shown graphically in panel B. Indeed, for the
1650
+ angular component of the AF the force on particle 1 is
1651
+ f1Ξ½ = βˆ’βˆ‚E1(ΞΈ512)
1652
+ βˆ‚x1Ξ½
1653
+ βˆ’ βˆ‚E2(ΞΈ123, ΞΈ124, ΞΈ324)
1654
+ βˆ‚x1Ξ½
1655
+ .
1656
+ (A2)
1657
+ Also in this case two contributions can be separated: (i)
1658
+ the interaction of particle 1 with triplets 123 and 124 is
1659
+ an effect of the three-body AF and it is present also in
1660
+ additive-models such as the mW model, (ii) the inter-
1661
+ action of particle 1 with triplet 324 is an effect of the
1662
+ non-additive nature of the NN local field Ei.
1663
+ VI.
1664
+ APPENDIX B
1665
+ In this Appendix we provide further thermodynam-
1666
+ ics comparisons between mW and NN3 potential focus-
1667
+ ing on the pressure and energy fluctuations. We depict
1668
+ in Fig. 10 the standard deviations of the total energy
1669
+ (normalized by N) in panel (A) and the standard devia-
1670
+ tion of virial pressure in panel (B). Energy fluctuations
1671
+ of NN3 follow qualitatively and quantitatively the trend
1672
+ of mW potential.
1673
+ Pressure fluctuations of NN3 are in
1674
+ good agreement with the mW model but, as for the pres-
1675
+ sure (Fig. 5.B), the accuracy decreases approaching state
1676
+ points outside the density range used for the training.
1677
+ 0.0
1678
+ 0.5
1679
+ 1.0
1680
+ 1.5
1681
+ 2.0
1682
+ 2.5
1683
+ 3.0
1684
+ 3.5
1685
+ 4.0
1686
+ Time (ns)
1687
+ 10.2
1688
+ 10.0
1689
+ 9.8
1690
+ 9.6
1691
+ 9.4
1692
+ (kcal mol
1693
+ 1)
1694
+ NN3 Total Energy
1695
+ NN3 Potential Energy
1696
+ mW Total Energy
1697
+ mW Potential Energy
1698
+ FIG. 11: NVE molecular dynamics at T = 299 K and ρ =
1699
+ 1.07 g cmβˆ’3 for both NN3 and mW model. The time step is
1700
+ dt = 4 fs for both models.
1701
+ VII.
1702
+ APPENDIX C
1703
+ In this Appendix we show a comparison between the
1704
+ mW and NN3 potentials in terms of the energy conser-
1705
+ vation in the NVE ensemble. In Fig. 11 we depict both
1706
+ total energy and potential energy for mW and NN3 po-
1707
+ tential. The potential energy and total energy of the two
1708
+ models are in good agreement.
1709
+ VIII.
1710
+ APPENDIX D
1711
+ In this Appendix we investigate the efficiency of the
1712
+ training over different choices for the number and types
1713
+ of atomic fingerprints introduced in the Neural Network
1714
+ Model section. We start by using only one three-body
1715
+ (n3b = 1) and one two-body (n2b = 1) AF and subse-
1716
+ quently increasing the number of the AF. For every com-
1717
+ bination of n2b and n3b, we run a 4000 epochs training
1718
+ and at the end of each training we extract the best model.
1719
+ We summarized these results in table I where we com-
1720
+ pare the error on forces over the all investigated model.
1721
+ From table I it emerges that the choice of n3b = 5 and
1722
+ n2b = 5 is the more convenient both for accuracy and
1723
+ computational efficiency.
1724
+ Doubling the number of the
1725
+ three-body AF marginally improves the error on forces
1726
+ while increases the computational cost due to the increase
1727
+ in the size of the input layer of the first hidden layer and
1728
+ due to the additional time to compute the three-body
1729
+ AF. Moreover in the RESULTS section we show that the
1730
+ choice n3b = 5 and n2b = 5 is sufficient to represent the
1731
+ target potential. Finally the accuracy of the training af-
1732
+ ter doubling the configurations in the dataset reaches an
1733
+ error on forces of βˆ†f = 5.85 meV ˚Aβˆ’1 that is 0.87 times
1734
+ the error value found with a half of the dataset.
1735
+
1736
+ 14
1737
+ TABLE I:
1738
+ Table of errors on forces at the end of the 4000
1739
+ epoch-long training procedure for different combination of the
1740
+ number and type of the AF.
1741
+ n3b
1742
+ n2b
1743
+ βˆ†f (meV ˚Aβˆ’1)
1744
+ n3b
1745
+ n2b
1746
+ βˆ†f (meV ˚Aβˆ’1)
1747
+ 1
1748
+ 1
1749
+ 72.79
1750
+ 5
1751
+ 1
1752
+ 16.53
1753
+ 1
1754
+ 2
1755
+ 67.92
1756
+ 5
1757
+ 2
1758
+ 7.53
1759
+ 1
1760
+ 5
1761
+ 56.25
1762
+ 5
1763
+ 5
1764
+ 6.72
1765
+ 1
1766
+ 10
1767
+ 56.00
1768
+ 5
1769
+ 10
1770
+ 6.87
1771
+ 1
1772
+ 15
1773
+ 56.02
1774
+ 5
1775
+ 15
1776
+ 6.95
1777
+ 2
1778
+ 1
1779
+ 53.76
1780
+ 10
1781
+ 1
1782
+ 7.98
1783
+ 2
1784
+ 2
1785
+ 43.95
1786
+ 10
1787
+ 2
1788
+ 7.17
1789
+ 2
1790
+ 5
1791
+ 32.43
1792
+ 10
1793
+ 5
1794
+ 5.79
1795
+ 2
1796
+ 10
1797
+ 32.39
1798
+ 10
1799
+ 10
1800
+ 6.55
1801
+ 2
1802
+ 15
1803
+ 24.70
1804
+ 10
1805
+ 15
1806
+ 6.19
1807
+
BdFJT4oBgHgl3EQfsS2B/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
DdAzT4oBgHgl3EQfGvsw/content/tmp_files/2301.01032v1.pdf.txt ADDED
@@ -0,0 +1,2566 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ON THE LIFTING PROBLEM OF REPRESENTATIONS OF A
2
+ METACYCLIC GROUP.
3
+ ARISTIDES KONTOGEORGIS AND ALEXIOS TEREZAKIS
4
+ Abstract. We give a necessary and sufficient condition for a modular rep-
5
+ resentation of a group G = Cph β‹Š Cm in a field of characteristic zero to be
6
+ lifted to a representation over local principal ideal domain of characteristic
7
+ zero containing the ph roots of unity.
8
+ 1. Introduction
9
+ The lifting problem for a representation
10
+ ρ : G β†’ GLn(k),
11
+ where k is a field of characteristic p > 0, is about finding a local ring R of char-
12
+ acteristic 0, with maximal ideal mR such that R/mR = k, so that the following
13
+ diagram is commutative:
14
+ GLn(R)
15
+ οΏ½
16
+ G
17
+ οΏ½
18
+ οΏ½
19
+ GLn(k)
20
+ Equivalently one asks if there is a free R-module V , which is also an R[G]-module
21
+ such that V βŠ—RR/mR is the k[G]-module corresponding to our initial representation.
22
+ We know that projective k[G]-modules lift in characteristic zero, [16, chap. 15], but
23
+ for a general k[G]-module such a lifting is not always possible, for example, see [10,
24
+ prop. 15]. This article aims to study the lifting problem for the group G = Cqβ‹ŠCm,
25
+ where Cq is a cyclic group of order ph and Cm is a cyclic group of order m, (p, m) = 1
26
+ and give necessary and sufficient condition in order to lift. We assume that the local
27
+ ring R contains the q-roots of unity and k is algebraically closed, and we might need
28
+ to consider a ramified extension of R, in order to ensure that certain q-roots of unit
29
+ are distant in the mR-topology, see remark 35. An example of such a ring R is the
30
+ ring of Witt vectors W(k)[ΞΆq] with the q-roots of unity adjoined to it.
31
+ We notice that a decomposable R[G]-module V gives rise to a decomposable
32
+ R-module modulo mR and also an indecomposable R[G]-module can break in the
33
+ reduction modulo mR into a direct sum of indecomposable k[G]-summands. We
34
+ also give a classification of k[Cq β‹Š Cm]-modules in terms of Jordan decomposition
35
+ and give the relation with the more usual uniserial description in terms of their
36
+ socle [1].
37
+ Date: January 4, 2023.
38
+ Key words and phrases. Lifting of representations, modular representation theory, integral
39
+ representation theory, Generalized Oort conjecture, metacyclic groups.
40
+ 1
41
+ arXiv:2301.01032v1 [math.AG] 3 Jan 2023
42
+
43
+ 2
44
+ A. KONTOGEORGIS AND A. TEREZAKIS
45
+ Our interest to this problem comes from the problem of lifting local actions. The
46
+ local lifting problem considers the following question: Does there exist an extension
47
+ Ξ›/W(k), and a representation
48
+ ˜ρ : G οΏ½β†’ Aut(Ξ›[[T]]),
49
+ such that if t is the reduction of T, then the action of G on Ξ›[[T]] reduces to the
50
+ action of G on k[[t]]?
51
+ If the answer to the above question is positive, then we say that the G-action
52
+ lifts to characteristic zero. A group G for which every local G-action on k[[t]] lifts to
53
+ characteristic zero is called a local Oort group for k. Notice that cyclic groups are
54
+ always local Oort groups. This result was known as the β€œOort conjecture”, which
55
+ was recently proved by F. Pop [15] using the work of A. Obus and S. Wewers [14].
56
+ There are a lot of obstructions that prevent a local action to lift in characteristic
57
+ zero. Probably the most important of these obstructions in the KGB-obstruction
58
+ [4]. It is believed that this is the only obstruction for the local lifting problem, see
59
+ [11], [12]. In [10, Thm. 3] the authors have given a criterion for the local lifting
60
+ which involves the lifting of a linear representation of the same group.
61
+ The case
62
+ G = Cq β‹ŠCm and especially the case of dihedral groups Dq = Cq β‹ŠC2, is a problem
63
+ of current interest in the theory of local liftings, see [12], [6], [18]. For more details
64
+ on the local lifting problem we refer to [3], [4], [5], [11].
65
+ Keep also in mind that the Cq β‹Š Cm groups were important to the study of
66
+ group actions in holomorphic differentials of curves defined over fields of positive
67
+ characteristic p, where the group involved has cyclic p-Sylow subgroup, see [2].
68
+ Let us now describe the method of proof. For understanding the splitting of
69
+ indecomposable R[G]-modules modulo mR, we develop a version of Jordan normal
70
+ form in lemma 16 for endomorphisms T : V β†’ V of order ph, where V is a free
71
+ module of rank d. We give a way to select this basis, by selecting an initial suitable
72
+ element E ∈ V , see lemma 15.
73
+ The normal form (as given in eq.
74
+ (9)) of the
75
+ element T of order q, determines the decomposition of the reduction. We show
76
+ that for every indecomposable summand Vi of V , we can select E as an eigenvalue
77
+ of the generator Οƒ of Cm and then by forcing the relation Ξ“T = T Ξ±Ξ“ to hold, we
78
+ see how the action of Οƒ can be extended recursivelly to an action of Οƒ on Vi, this is
79
+ done in lemma 24. Proving that this gives indeed a well defined action is a technical
80
+ computation and is done in lemmata 26, 27, 28, 32, 33.
81
+ The important thing here is that the definition of the action of Οƒ on E is the
82
+ β€œinitial condition” of a dynamical system that determines the action of Cm on the
83
+ indecomposable summand Vi. The R[Cqβ‹ŠCm] indecomposable module Vi can break
84
+ into a direct sum VΞ±(ϡν, ΞΊΞ½)-modules 1 ≀ Ξ½ ≀ s (for a precise definition of them
85
+ see definition 9, notice that ΞΊi denotes the dimension). The action of Οƒ on each
86
+ VΞ±(ϡν, ΞΊΞ½) can be uniquely determined by the action of Οƒ on an initial basis element
87
+ as shown in section 3, again by a β€œdynamical system” approach, where we need s
88
+ initial conditions, one for each Vα(ϡν, κν). The lifting condition essentially means
89
+ that the indecomposable summands Vα(ϡ, κ) of the special fibre, should be able
90
+ to be rearranged in a suitable way, so that they can be obtained as reductions of
91
+ indecomposable R[Cq β‹Š Cq]-modules. The precise expression of our lifting criterion
92
+ is given in the following proposition:
93
+ Proposition 1. Consider a k[G]-module M which is decomposed as a direct sum
94
+ M = VΞ±(Ο΅1, ΞΊ1) βŠ• Β· Β· Β· βŠ• VΞ±(Ο΅s, ΞΊs).
95
+
96
+ ON THE LIFTING PROBLEM OF REPRESENTATIONS OF A METACYCLIC GROUP.
97
+ 3
98
+ The module lifts to an R[G]-module if and only if the set {1, . . . , s} can be written
99
+ as a disjoint union of sets IΞ½, 1 ≀ Ξ½ ≀ t so that
100
+ a οΏ½
101
+ ¡∈IΞ½ ΞΊΒ΅ ≀ q, for all 1 ≀ Ξ½ ≀ t.
102
+ b οΏ½
103
+ ¡∈IΞ½ ΞΊΒ΅ ≑ a modm for all 1 ≀ Ξ½ ≀ t, where a ∈ {0, 1}.
104
+ c For each Ξ½, 1 ≀ Ξ½ ≀ t there is an enumeration Οƒ : {1, . . . , #IΞ½} β†’ IΞ½ βŠ‚
105
+ {1, .., s}, such that
106
+ ϡσ(2) = ϡσ(1)Ξ±ΞΊΟƒ(1), ϡσ(3) = ϡσ(3)Ξ±ΞΊΟƒ(3), . . . ϡσ(s) = ϡσ(sβˆ’1)Ξ±ΞΊΟƒ(sβˆ’1)
107
+ In the above proposition, each set IΞ½ corresponds to a collection of modules
108
+ VΞ±(ϡ¡, ΞΊΒ΅), Β΅ ∈ IΞ½ which come as the reduction of an indecomposable R[Cq β‹Š Cm]-
109
+ module VΞ½ of V .
110
+ Aknowledgements A. Terezakis is a recipient of financial support in the context
111
+ of a doctoral thesis (grant number MIS-5113934). The implementation of the doc-
112
+ toral thesis was co-financed by Greece and the European Union (European Social
113
+ Fund-ESF) through the Operational Programmeβ€”Human Resources Development,
114
+ Education and Lifelong Learningβ€”in the context of the Actβ€”Enhancing Human
115
+ Resources Research Potential by undertaking a Doctoral Researchβ€”Sub-action 2:
116
+ IKY Scholarship Programme for Ph.D. candidates in the Greek Universities.
117
+ 2. Notation
118
+ Let Ο„ be a generator of the cyclic group Cq and Οƒ be a generator of the cyclic
119
+ group Cm. The group G is given in terms of generators and relations as follows:
120
+ G = βŸ¨Οƒ, Ο„|Ο„ q = 1, Οƒm = 1, ΟƒΟ„Οƒβˆ’1 = Ο„ Ξ± for some Ξ± ∈ N, 1 ≀ Ξ± ≀ ph βˆ’ 1, (Ξ±, p) = 1⟩.
121
+ The integer α satisfies the following congruence:
122
+ (1)
123
+ Ξ±m ≑ 1 modq
124
+ as one sees by computing Ο„ = ΟƒmΟ„Οƒβˆ’m = Ο„ Ξ±m. Also the integer Ξ± can be seen as
125
+ an element in the finite field Fp, and it is a (p βˆ’ 1)-th root of unity, not necessarily
126
+ primitive. In particular the following holds:
127
+ Lemma 2. Let ΢m ∈ k be a fixed primitive m-th root of unity. There is a natural
128
+ number a0, 0 ≀ a0 < m βˆ’ 1 such that Ξ± = ΞΆa0
129
+ m .
130
+ Proof. The integer α if we see it as an element in k is an element in the finite field
131
+ Fp βŠ‚ k, therefore Ξ±pβˆ’1 = 1 as an element in Fp. Let ordp(Ξ±) be the order of Ξ± in Fβˆ—
132
+ p.
133
+ By eq. (1) we have that ordp(Ξ±) | pβˆ’1 and ordp(Ξ±) | m, that is ordp(Ξ±) | (pβˆ’1, m).
134
+ The primitive m-th root of unity ΢m generates a finite field Fp(΢m) = Fpν for
135
+ some integer Ξ½, which has cyclic multiplicative group FpΞ½\{0} containing both the
136
+ cyclic groups ⟨΢m⟩ and ⟨α⟩. Since for every divisor δ of the order of a cyclic group
137
+ C there is a unique subgroup Cβ€² < C of order Ξ΄ we have that Ξ± ∈ ⟨΢m⟩, and the
138
+ result follows.
139
+ β–‘
140
+ Definition 3. For each pi | q we define ordpiα to be the smallest natural number
141
+ o such that Ξ±o ≑ 1 modpi.
142
+
143
+ EZNA
144
+ OperationalProgramme
145
+ HumanResourcesDevelopment
146
+ 2014-2020
147
+ 士
148
+ EducationandLifelongLearning
149
+ avantuen-epyaoia-aaanaeun
150
+ Eupwaikn'Evwon
151
+ Co-financed byGreece and the European Union
152
+ European Social Fund4
153
+ A. KONTOGEORGIS AND A. TEREZAKIS
154
+ It is clear that for ν ∈ N
155
+ Ξ±Ξ½ ≑ 1 modpi β‡’ Ξ±Ξ½ ≑ 1 modpj for all j ≀ i.
156
+ Therefore
157
+ ordpjΞ± | ordpiΞ± for j ≀ i.
158
+ On the other hand Ξ± ∈ N and Ξ±pβˆ’1 ≑ 1 modp so ordpΞ± | p βˆ’ 1.
159
+ Also since
160
+ ΟƒtΟ„Οƒβˆ’t = Ο„ Ξ±t we have that Ξ±m ≑ 1 modph, therefore ordpΞ± | ordpiΞ± | ordphΞ± | m,
161
+ for 1 ≀ i ≀ h.
162
+ Lemma 4. The center CentG(Ο„) = βŸ¨Ο„, Οƒordphα⟩. Moreover
163
+ |CentG(Ο„)|
164
+ ph
165
+ =
166
+ m
167
+ ordph(Ξ±) =: mβ€²
168
+ Proof. The result follows by observing (Ο„ Ξ½Οƒt)Ο„(Ο„ Ξ½Οƒt)βˆ’1 = Ο„ Ξ±t, for all 1 ≀ Ξ½ ≀ q,
169
+ 1 ≀ t ≀ m.
170
+ β–‘
171
+ Remark 5. If ordpΞ± = m then ordpiΞ± = m for all 1 ≀ i ≀ h.
172
+ Lemma 6. If the group G = Cq β‹Š Cm is a subgroup of Aut(k[[t]]), then all orders
173
+ ordpiΞ± = m/mβ€², for all 1 ≀ i ≀ h.
174
+ Proof. We will use the notation of the book of J.P.Serre on local fields [17]. By
175
+ [13, Th.1.1b] we have that the first gap i0 in the lower ramification filtration of the
176
+ cyclic group Cq satisfies (m, i0) = mβ€².
177
+ The ramification relation [17, prop. 9 p. 69]
178
+ Ξ±ΞΈi0(Ο„) = ΞΈi0(Ο„ Ξ±) = ΞΈi0(ΟƒΟ„Οƒβˆ’1) = ΞΈ0(Οƒ)i0ΞΈi0(Ο„),
179
+ implies that ΞΈ0(Οƒ)i0 = Ξ± ∈ N. From (m, i0) = mβ€² and the fact that ordΞΈ0(Οƒ) = m
180
+ we obtain
181
+ m
182
+ mβ€² = ordΞΈ0(Οƒ)i0 = ordp(Ξ±).
183
+ Thus
184
+ m
185
+ mβ€² = ordpΞ±|ordpiΞ±|ordphΞ± = m
186
+ mβ€² .
187
+ Hence all orders ordpiΞ± = m/mβ€².
188
+ β–‘
189
+ Remark 7. If the KGB-obstruction vanishes and Ξ± ΜΈ= 1, then by [11][prop. 5.9]
190
+ i0 ≑ βˆ’1 modm and ordpiΞ± = m for all 1 ≀ i ≀ h.
191
+ 3. Indecomposable Cq β‹Š Cm modules, modular representation theory
192
+ In this section we will describe the indecomposable Cq β‹Š Cm-modules. We will
193
+ give two methods in studying them. The first one is needed since it is in accordance
194
+ to the method we will give in order to describe indecomposable R[Cqβ‹ŠCm]-modules.
195
+ The second one, using the structure of the socle, is the standard method of describ-
196
+ ing k[Cq β‹Š Cm]-modules in modular representation theory.
197
+
198
+ ON THE LIFTING PROBLEM OF REPRESENTATIONS OF A METACYCLIC GROUP.
199
+ 5
200
+ 3.1. Linear algebra method. The indecomposable modules of the Cq are deter-
201
+ mined by the Jordan normal forms of the generator Ο„ of the cyclic group Cq. So
202
+ for each 1 ≀ ΞΊ ≀ ph there is exactly one Cq indecomposable module denoted by
203
+ JΞΊ. Therefore we have the following decomposition of an indecomposable Cq β‹ŠCm-
204
+ module M considered as a Cq-module.
205
+ (2)
206
+ M = JΞΊ1 βŠ• Β· Β· Β· βŠ• JΞΊr.
207
+ Lemma 8. In the indecomposable module JΞΊ for every element E such that
208
+ (Ο„ βˆ’ IdΞΊi)ΞΊiβˆ’1E ΜΈ= 0
209
+ the elements B = {E, (Ο„ βˆ’ IdΞΊ)E, . . . , (Ο„ βˆ’ IdΞΊ)ΞΊβˆ’1E} form a basis of JΞΊ such that
210
+ the matrix of Ο„ with respect to this basis is given by
211
+ (3)
212
+ Ο„ = IdΞΊ +
213
+ οΏ½
214
+ οΏ½
215
+ οΏ½
216
+ οΏ½
217
+ οΏ½
218
+ οΏ½
219
+ οΏ½
220
+ οΏ½
221
+ οΏ½
222
+ 0
223
+ Β· Β· Β·
224
+ Β· Β· Β·
225
+ Β· Β· Β·
226
+ 0
227
+ 1
228
+ ...
229
+ ...
230
+ 0
231
+ ...
232
+ ...
233
+ ...
234
+ ...
235
+ ...
236
+ 1
237
+ 0
238
+ ...
239
+ 0
240
+ Β· Β· Β·
241
+ 0
242
+ 1
243
+ 0
244
+ οΏ½
245
+ οΏ½
246
+ οΏ½
247
+ οΏ½
248
+ οΏ½
249
+ οΏ½
250
+ οΏ½
251
+ οΏ½
252
+ οΏ½
253
+ .
254
+ Proof. Since the set B has k-elements it is enough to prove that it consists of linear
255
+ independent elements. Indeed, consider a linear relation
256
+ Ξ»0E + Ξ»1(Ο„ βˆ’ IdΞΊ)E + Β· Β· Β· + Ξ»ΞΊβˆ’1(Ο„ βˆ’ IdΞΊ)ΞΊβˆ’1E = 0.
257
+ By applying (Ο„ βˆ’IdΞΊ)ΞΊβˆ’1 we obtain Ξ»0(Ο„ βˆ’IdΞΊ)ΞΊβˆ’1 = 0, which gives us Ξ»0 = 0. We
258
+ then apply (Ο„ βˆ’ IdΞΊ)ΞΊβˆ’2 to the linear relation and by the same argument we obtain
259
+ Ξ»1 = 0 and we continue this way proving that Ξ»0 = Β· Β· Β· = Ξ»ΞΊβˆ’1 = 0. The matrix
260
+ form of Ο„ in this basis is immediate.
261
+ β–‘
262
+ We will now prove that Οƒ acts on each JΞΊ of eq.
263
+ (2) proving that r = 1.
264
+ Since the field k is algebraically closed and (m, p) = 1 we know that there is
265
+ a basis of M consisting of eigenvectors of Οƒ.
266
+ There is an eigenvector E of Οƒ,
267
+ which is not in the kernel of (Ο„ βˆ’ IdΞΊ)ΞΊ1βˆ’1. Then the elements of the set B =
268
+ {E, (Ο„ βˆ’ IdΞΊ)E, . . . , (Ο„ βˆ’ IdΞΊ)ΞΊ1βˆ’1E} are linearly independent and form a direct Cq
269
+ summand of M isomorphic to JΞΊ1.
270
+ We will now show that this module is an k[Cq β‹Š Cm]-module. For this, we have
271
+ to show that the generator Οƒ of Cm acts on the basis B. Observe that for every
272
+ 0 ≀ i ≀ ΞΊ1 βˆ’ 1 < ph
273
+ Οƒ(Ο„ βˆ’ 1)iβˆ’1 = (Ο„ Ξ± βˆ’ 1)iβˆ’1Οƒ.
274
+ Set e = E1 and ΞΊ = ΞΊ1. This means that the action of Οƒ on e determines the action
275
+ of Οƒ on all other basis elements eΞ½ := (Ο„ βˆ’ 1)Ξ½βˆ’1e, 1 ≀ Ξ½ ≀ ΞΊ1.
276
+ Let us compute:
277
+ Οƒei+1 = Οƒ(Ο„ βˆ’ 1)ie = (Ο„ Ξ± βˆ’ 1)iΞΆΞ»
278
+ me
279
+
280
+ 6
281
+ A. KONTOGEORGIS AND A. TEREZAKIS
282
+ On the basis {e1, . . . , eΞΊ1} the matrix Ο„ is given by eq. (3) hence using the binomial
283
+ formula we compute
284
+ (4)
285
+ Ο„ Ξ± =
286
+ οΏ½
287
+ οΏ½
288
+ οΏ½
289
+ οΏ½
290
+ οΏ½
291
+ οΏ½
292
+ οΏ½
293
+ οΏ½
294
+ οΏ½
295
+ οΏ½
296
+ οΏ½
297
+ οΏ½
298
+ 1
299
+ 0
300
+ Β· Β· Β·
301
+ Β· Β· Β·
302
+ Β· Β· Β·
303
+ 0
304
+ οΏ½Ξ±
305
+ 1
306
+ οΏ½
307
+ 1
308
+ ...
309
+ ...
310
+ οΏ½Ξ±
311
+ 2
312
+ οΏ½
313
+ οΏ½Ξ±
314
+ 1
315
+ οΏ½
316
+ ...
317
+ ...
318
+ ...
319
+ οΏ½Ξ±
320
+ 3
321
+ οΏ½
322
+ οΏ½Ξ±
323
+ 2
324
+ οΏ½
325
+ ...
326
+ 1
327
+ ...
328
+ ...
329
+ ...
330
+ ...
331
+ ...
332
+ οΏ½Ξ±
333
+ 1
334
+ οΏ½
335
+ 1
336
+ 0
337
+ οΏ½Ξ±
338
+ k
339
+ οΏ½
340
+ οΏ½ Ξ±
341
+ kβˆ’1
342
+ οΏ½
343
+ Β· Β· Β·
344
+ οΏ½Ξ±
345
+ 2
346
+ οΏ½
347
+ οΏ½Ξ±
348
+ 1
349
+ οΏ½
350
+ 1
351
+ οΏ½
352
+ οΏ½
353
+ οΏ½
354
+ οΏ½
355
+ οΏ½
356
+ οΏ½
357
+ οΏ½
358
+ οΏ½
359
+ οΏ½
360
+ οΏ½
361
+ οΏ½
362
+ οΏ½
363
+ .
364
+ Thus Ο„ Ξ± βˆ’ 1 is a nilpotent matrix A = (aij) of the form:
365
+ aij =
366
+ οΏ½οΏ½Ξ±
367
+ Β΅
368
+ οΏ½
369
+ if j = i βˆ’ Β΅ for some Β΅, 1 ≀ Β΅ ≀ ΞΊ
370
+ 0
371
+ if j β‰₯ i
372
+ The β„“-th power Aβ„“ = (a(β„“)
373
+ ij ) of A is then computed by (keep in mind that aij = 0
374
+ for i ≀ j)
375
+ a(β„“)
376
+ ij =
377
+ οΏ½
378
+ i<Ξ½1<Β·Β·Β·<Ξ½β„“βˆ’1<j
379
+ ai,Ξ½1aΞ½1,Ξ½2aΞ½2,Ξ½3 Β· Β· Β· aΞ½β„“βˆ’1,j
380
+ This means that i βˆ’ j > β„“ in order to have aij ΜΈ= 0. Moreover for i = j + β„“ (which
381
+ is the the first non zero diagonal below the main diagonal) we have
382
+ ai,i+β„“ = ai,i+1ai+1,i+2 Β· Β· Β· ai+β„“βˆ’1,i+β„“ =
383
+ οΏ½Ξ±
384
+ 1
385
+ οΏ½β„“
386
+ = Ξ±β„“.
387
+ Therefore, the matrix of Aβ„“ is of the following form:
388
+ (5)
389
+ οΏ½
390
+ οΏ½
391
+ οΏ½
392
+ οΏ½
393
+ οΏ½
394
+ οΏ½
395
+ οΏ½
396
+ οΏ½
397
+ οΏ½
398
+ οΏ½
399
+ οΏ½
400
+ οΏ½
401
+ οΏ½
402
+ οΏ½
403
+ k βˆ’ β„“
404
+ οΏ½
405
+ οΏ½οΏ½
406
+ οΏ½
407
+ 0
408
+ Β· Β· Β·
409
+ Β· Β· Β·
410
+ 0
411
+ β„“
412
+ οΏ½
413
+ οΏ½οΏ½
414
+ οΏ½
415
+ 0
416
+ Β· Β· Β·
417
+ 0
418
+ ...
419
+ ...
420
+ ...
421
+ ...
422
+ 0
423
+ Β· Β· Β·
424
+ Β· Β· Β·
425
+ 0
426
+ 0
427
+ Β· Β· Β·
428
+ 0
429
+ Ξ±β„“
430
+ ...
431
+ 0
432
+ ...
433
+ ...
434
+ βˆ—
435
+ Ξ±β„“
436
+ ...
437
+ ...
438
+ ...
439
+ ...
440
+ ...
441
+ ...
442
+ ...
443
+ 0
444
+ ...
445
+ ...
446
+ βˆ—
447
+ Β· Β· Β·
448
+ βˆ—
449
+ Ξ±β„“
450
+ 0
451
+ Β· Β· Β·
452
+ 0
453
+ οΏ½
454
+ οΏ½
455
+ οΏ½
456
+ οΏ½
457
+ οΏ½
458
+ οΏ½
459
+ οΏ½
460
+ οΏ½
461
+ οΏ½
462
+ οΏ½
463
+ οΏ½
464
+ οΏ½
465
+ οΏ½
466
+ οΏ½
467
+ Definition 9. We will denote by Vα(λ, κ) the indecomposable κ-dimensional G-
468
+ module given by the basis elements {(Ο„ βˆ’ 1)Ξ½e, Ξ½ = 0, . . . , ΞΊ βˆ’ 1}, where Οƒe = ΞΆΞ»
469
+ me.
470
+ This definition is close to the notation used in [9].
471
+ Lemma 10. The action of Οƒ on the basis element ei of VΞ±(Ξ», ΞΊ) is given by:
472
+ (6)
473
+ Οƒei = Ξ±iβˆ’1ΞΆΞ»
474
+ mei +
475
+ ΞΊ
476
+ οΏ½
477
+ Ξ½=i+1
478
+ aΞ½eΞ½,
479
+ for some coefficients ai ∈ k. In particular the matrix of Οƒ with respect to the basis
480
+ e1, . . . , eΞΊ is lower triangular.
481
+
482
+ ON THE LIFTING PROBLEM OF REPRESENTATIONS OF A METACYCLIC GROUP.
483
+ 7
484
+ Proof. Recall that ei = (Ο„ βˆ’ 1)iβˆ’1e1. Therefore
485
+ Οƒei = Οƒ(Ο„ βˆ’ 1)iβˆ’1e1 = (Ο„ Ξ± βˆ’ 1)iβˆ’1Οƒe1 = ΞΆΞ»
486
+ m(Ο„ Ξ± βˆ’ 1)iβˆ’1e1.
487
+ The result follows by eq. (5)
488
+ β–‘
489
+ We have constructed a set of indecomposable modules VΞ±(Ξ», ΞΊ).
490
+ Apparently
491
+ VΞ±(Ξ», ΞΊ) can not be isomorphic to VΞ±(Ξ»β€², ΞΊβ€²) if ΞΊ ΜΈ= ΞΊβ€², since they have different
492
+ dimensions.
493
+ Assume now that ΞΊ = ΞΊβ€². Can the modules VΞ±(Ξ», ΞΊ) and VΞ±(Ξ»β€², ΞΊ) be isomorphic
494
+ for Ξ» ΜΈ= Ξ»β€²?
495
+ The eigenvalues of the prime to p generator Οƒ on VΞ±(Ξ», ΞΊ)are
496
+ ΞΆΞ»
497
+ m, Ξ±ΞΆΞ»
498
+ m, . . . , Ξ±ΞΊβˆ’1ΞΆΞ»
499
+ m.
500
+ Similarly the eigenvalues for Οƒ when acting on VΞ±(Ξ»β€², ΞΊ) are
501
+ ΞΆΞ»β€²
502
+ m , Ξ±ΞΆΞ»β€²
503
+ m , . . . , Ξ±ΞΊβˆ’1ΞΆΞ»β€²
504
+ m .
505
+ If the two sets of eigenvalues are different then the modules can not be isomorphic.
506
+ But even if Ξ» ΜΈ= Ξ»β€² modn the two sets of eigenvalues can still be equal. Even in this
507
+ case the modules can not be isomorphic.
508
+ Lemma 11. The modules VΞ±(Ξ»1, ΞΊ) and VΞ±(Ξ»2, ΞΊ) are isomorphic if and only if
509
+ Ξ»1 ≑ Ξ»2 modm.
510
+ Proof. Indeed, the module VΞ±(Ξ»1, ΞΊ) has an eigenvector for the action of Οƒ which
511
+ generates the VΞ±(Ξ»1, ΞΊ) by powers of (Ο„ βˆ’ 1), i.e. the vectors
512
+ (7)
513
+ e, (Ο„ βˆ’ 1)e, (Ο„ βˆ’ 1)2e, . . . , (Ο„ βˆ’ 1)ΞΊβˆ’1e
514
+ form a basis of VΞ±(Ξ»1, ΞΊ).
515
+ The elements E which can generate VΞ±(Ξ»1, ΞΊ) by powers of (Ο„ βˆ’ 1) are linear
516
+ combinations
517
+ E =
518
+ ΞΊβˆ’1
519
+ οΏ½
520
+ Ξ½=0
521
+ Ξ»i(Ο„ βˆ’ 1)Ξ½e,
522
+ for λi ∈ k and λ0 ̸= 0.
523
+ On the other hand using eq. (6) we see that Οƒ with respect to the basis given in
524
+ eq. (7) admits the matrix form:
525
+ οΏ½
526
+ οΏ½
527
+ οΏ½
528
+ οΏ½
529
+ οΏ½
530
+ οΏ½
531
+ οΏ½
532
+ οΏ½
533
+ ΞΆΞ»
534
+ m
535
+ 0
536
+ Β· Β· Β·
537
+ Β· Β· Β·
538
+ 0
539
+ 0
540
+ Ξ±ΞΆΞ»
541
+ m
542
+ 0
543
+ Β· Β· Β·
544
+ 0
545
+ ...
546
+ ...
547
+ ...
548
+ ...
549
+ ...
550
+ ...
551
+ ...
552
+ ...
553
+ ...
554
+ 0
555
+ Β· Β· Β·
556
+ Β· Β· Β·
557
+ 0
558
+ Ξ±ΞΊβˆ’1ΞΆΞ»
559
+ m
560
+ οΏ½
561
+ οΏ½
562
+ οΏ½
563
+ οΏ½
564
+ οΏ½
565
+ οΏ½
566
+ οΏ½
567
+ οΏ½
568
+ .
569
+ It is now easy to see from the above matrix that every eigenvector of the eigenvalue
570
+ Ξ±Ξ½Ξ»1, Ξ½ > 1 is expressed as a linear combination of the basis given in eq. (7), where
571
+ the coefficient of e is zero.
572
+ Therefore, the eigenvector of the eigenvalue Ξ±Ξ½ΞΆm can not generate the module
573
+ VΞ±(Ξ», ΞΊ) by powers of (Οƒ βˆ’ 1)Ξ½.
574
+ β–‘
575
+
576
+ 8
577
+ A. KONTOGEORGIS AND A. TEREZAKIS
578
+ 3.2. The uniserial description. We will now give an alternative description of
579
+ the indecomposable Cq β‹Š Cm-modules, which is used in [2].
580
+ It is known that Aut(Cq) ∼= Fβˆ—
581
+ p Γ— Q, for some abelian p-group Q. The repre-
582
+ sentation ψ : Cm β†’ Aut(Cq) given by the action of Cm on Cq is known to factor
583
+ through a character Ο‡ : Cm β†’ Fβˆ—
584
+ p. The order of Ο‡ divides pβˆ’1 and Ο‡pβˆ’1 = Ο‡βˆ’(pβˆ’1)
585
+ is the trivial one dimensional character.
586
+ For all i ∈ Z, Ο‡i defines a simple k[Cm]-module of k dimension one, which we
587
+ will denote by SΟ‡i. For 0 ≀ β„“ ≀ m βˆ’ 1 denote by Sβ„“ the simple module where
588
+ on which Οƒ acts as ΞΆβ„“
589
+ m. Both SΟ‡i, Sβ„“ can be seen as k[Cq β‹Š Cm]-modules using
590
+ inflation. Finally for 0 ≀ β„“ ≀ m βˆ’ 1 we define Ο‡i(β„“) ∈ {0, 1, . . . , m βˆ’ 1} such that
591
+ SΟ‡i(β„“) ∼= Sβ„“ βŠ—k SΟ‡i.
592
+ There are q Β· m isomorphism classes of indecomposable k[Cq β‹Š Cm]-modules and
593
+ are all uniserial. An indecomposable k[Cq β‹Š Cm]-module U is unique determined
594
+ by its socle, which is the kernel of the action of Ο„ βˆ’ 1 on U, and its k-dimension.
595
+ For 0 ≀ β„“ ≀ m βˆ’ 1 and 1 ≀ Β΅ ≀ q, let Uβ„“,Β΅ be the indecomposable k[Cq β‹Š Cm]
596
+ module with socle Sa and k-dimension Β΅. Then Uβ„“,Β΅ is uniserial and its Β΅ ascending
597
+ composition factors are the first ¡ composition factors of the sequence
598
+ Sβ„“, SΟ‡βˆ’1(β„“), SΟ‡βˆ’2(β„“), . . . , SΟ‡βˆ’(pβˆ’2)(β„“), Sβ„“, SΟ‡βˆ’1(β„“), SΟ‡βˆ’2(β„“), . . . , SΟ‡βˆ’(pβˆ’2)(β„“).
599
+ Notice that in our notation VΞ±(Ξ», ΞΊ) = UΞ»+ΞΊ,ΞΊ.
600
+ Remark 12. The condition ordpi = m for all 1 ≀ i ≀ h, is equivalent to requiring
601
+ that ψi : Cm β†’ Aut(Cpi) is faithful for all i.
602
+ 4. Lifting of representations
603
+ Proposition 13. Let G = Cq β‹Š Cm. Assume that for all 1 ≀ i ≀ h, ordpia = m.
604
+ If the G-module V lifts to an R[G]-module ˜V , where K = Quot(R) is a field of
605
+ characterstic zero, then
606
+ m |
607
+ οΏ½
608
+ dim( ˜V βŠ—R K) βˆ’ dim( ˜V βŠ—R K)CqοΏ½
609
+ .
610
+ Moreover, if ˜V (΢αiκ
611
+ q
612
+ ) is the eigenspace of the eigenvalue ΞΆΞ±iΞΊ
613
+ q
614
+ of T acting on ˜V ,
615
+ then
616
+ dim ˜V (΢κ
617
+ q ) = dim ˜V (΢ακ
618
+ q ) = dim ˜V (΢α2κ
619
+ q
620
+ ) = Β· Β· Β· = dim ˜V (ΞΆΞ±mβˆ’1ΞΊ
621
+ q
622
+ ).
623
+ Proof. Consider a lifting ˜V of V . The generator Ο„ of the cyclic part Cq has eigen-
624
+ values Ξ»1, . . . , Ξ»s which are pn-roots of unity. Let ΞΆq be a primitive q-root of unity.
625
+ Consider any eigenvalue Ξ» ΜΈ= 1. It is of the form Ξ» = ΞΆΞΊ
626
+ q for some κ ∈ N, q ∀ κ. If E
627
+ is an eigenvector of T corresponding to Ξ», that is Ο„E = ΞΆΞΊ
628
+ q E then
629
+ Ο„Οƒβˆ’1E = Οƒβˆ’1Ο„ Ξ±E = ΞΆΞΊΞ±mβˆ’1
630
+ q
631
+ Οƒβˆ’1E
632
+ and we have a series of eigenvectors E, Οƒβˆ’1E, Οƒβˆ’2E, Β· Β· Β· with corresponding eigen-
633
+ values ΞΆΞΊ
634
+ q , ΞΆΞΊΞ±
635
+ q , ΞΆΞΊa2
636
+ q
637
+ Β· Β· Β· , ΞΆΞΊΞ±o
638
+ q
639
+ , where o = ordq/(q,k). Indeed, the integer o satisfies
640
+ the
641
+ ΞΊΞ±o ≑ ΞΊ modq β‡’ Ξ±m ≑ 1 mod
642
+ q
643
+ (q, k).
644
+ Therefore the eigenvalues Ξ» ΜΈ= 1 form orbits of size m, while the eigenspace of the
645
+ eigenvalue 1 is just the invariant space V G and the result follows.
646
+ β–‘
647
+
648
+ ON THE LIFTING PROBLEM OF REPRESENTATIONS OF A METACYCLIC GROUP.
649
+ 9
650
+ 5. Indecomposable Cq β‹Š Cm modules, integral representation theory
651
+ From now on V be a free R-module, where R is an integral local principal ideal
652
+ domain with maximal ideal mR, R has characteristic zero and that R contains all
653
+ q-th roots of unity and has characteristic zero. Let K = Quot(R).
654
+ The indecomposable modules for a cyclic group both in the ordinary and in the
655
+ modular case are described by writing down the Jordan normal form of a generator
656
+ of the cyclic group. Since in integral representation theory there are infinitely many
657
+ non-isomorphic indecomposable Cq-modules for q = ph, h β‰₯ 3, one is not expecting
658
+ to have a theory of Jordan normal forms even if one works over complete local
659
+ principal ideal domains [7], [8].
660
+ Lemma 14. Let T be an element of order q = ph in End(V ), then the minimal
661
+ polynomial of T has simple eigenvalues and T is diagonalizable when seen as an
662
+ element in End(V βŠ— K).
663
+ Proof. Since T q = IdV , the minimal polynomial of T divides xq βˆ’ 1, which has
664
+ simple roots over a field of characteristic zero. This ensures that T ∈ End(V βŠ— K)
665
+ is diagonalizable.
666
+ β–‘
667
+ Lemma 15. Let f(x) = (x βˆ’ Ξ»1)(x βˆ’ Ξ»2) Β· Β· Β· (x βˆ’ Ξ»d) be the minimal polynomial of
668
+ T on V . There is an element E ∈ V , such that
669
+ E, (T βˆ’ Ξ»1IdV )E, (T βˆ’ Ξ»2IdV )(T βˆ’ Ξ»1IdV )E, . . . , (T βˆ’ Ξ»dβˆ’1IdV ) Β· Β· Β· (T βˆ’ Ξ»1IdV )E
670
+ are linear independent elements in V βŠ— K.
671
+ Proof. Consider the endomorphisms for i = 1, . . . , d
672
+ Ξ i =
673
+ d
674
+ οΏ½
675
+ Ξ½=1
676
+ Ξ½ΜΈ=i
677
+ (T βˆ’ λνIdV ).
678
+ In the above product notice that T βˆ’ Ξ»iIdV , T βˆ’ Ξ»jIdV are commuting endomor-
679
+ phisms. Since the minimal polynomial of T has degree d all R-modules KerΞ i are
680
+ strictly less than V . Moreover there is an element E such that E ̸∈ Ker(Πi) for all
681
+ 1 ≀ i ≀ d. Consider a relation
682
+ (8)
683
+ d
684
+ οΏ½
685
+ Β΅=0
686
+ Ξ³Β΅
687
+ Β΅
688
+ οΏ½
689
+ Ξ½=0
690
+ (T βˆ’ λ¡IdV )E,
691
+ where οΏ½0
692
+ Ξ½=0(T βˆ’ λνIdV )E = E. We fist apply the operator οΏ½d
693
+ Ξ½=2(T βˆ’ λνIdV ) to
694
+ eq. (8) and we obtain
695
+ 0 = Ξ³0Ξ 1E,
696
+ and by the selection of E we have that a0 = 0. We now apply οΏ½d
697
+ Ξ½=3(T βˆ’ λνIdV )
698
+ to eq. (8). We obtain that
699
+ 0 = Ξ³1
700
+ d
701
+ οΏ½
702
+ Ξ½=3
703
+ (T βˆ’ λνIdV )(T βˆ’ Ξ»1IdV ) = Ξ³1Ξ 2E,
704
+ and by the selection of E we have that Ξ³1 = 0. We now apply οΏ½d
705
+ Ξ½=4(T βˆ’ λνIdV )
706
+ to eq. (8) and we obtain
707
+ 0 = Ξ³2
708
+ d
709
+ οΏ½
710
+ Ξ½=4
711
+ (T βˆ’ λνIdV )(T βˆ’ Ξ»2IdV )(T βˆ’ Ξ»1IdV )E = Ξ³2Ξ 3E
712
+
713
+ 10
714
+ A. KONTOGEORGIS AND A. TEREZAKIS
715
+ and by the selection of E we obtain γ3 = 0. Continuing this way we finally arrive
716
+ at Ξ³0 = Ξ³1 = Β· Β· Β· = Ξ³dβˆ’1 = 0.
717
+ β–‘
718
+ Lemma 16. Let V be a free R-module of rank R acted on by an automorphism
719
+ T : V β†’ V of order ph. Assume that the minimal polynomial of T is of degree d
720
+ and has roots Ξ»1, . . . , Ξ»d. Then T can be written as a matrix with respect to the
721
+ basis as follows:
722
+ (9)
723
+ οΏ½
724
+ οΏ½
725
+ οΏ½
726
+ οΏ½
727
+ οΏ½
728
+ οΏ½
729
+ οΏ½
730
+ οΏ½
731
+ οΏ½
732
+ Ξ»1
733
+ 0
734
+ Β· Β· Β·
735
+ Β· Β· Β·
736
+ 0
737
+ a1
738
+ Ξ»2
739
+ ...
740
+ ...
741
+ 0
742
+ a2
743
+ Ξ»3
744
+ ...
745
+ ...
746
+ ...
747
+ ...
748
+ ...
749
+ ...
750
+ 0
751
+ 0
752
+ Β· Β· Β·
753
+ 0
754
+ adβˆ’1
755
+ Ξ»d
756
+ οΏ½
757
+ οΏ½
758
+ οΏ½
759
+ οΏ½
760
+ οΏ½
761
+ οΏ½
762
+ οΏ½
763
+ οΏ½
764
+ οΏ½
765
+ Proof. By lemma 15 the elements
766
+ E, (T βˆ’ Ξ»1IdV )E, (T βˆ’ Ξ»2IdV )(T βˆ’ Ξ»1IdV )E, . . . , (T βˆ’ Ξ»dβˆ’1IdV ) Β· Β· Β· (T βˆ’ Ξ»1IdV )E
767
+ form a free submodule of V of rank d. The theory of submodules of principal ideal
768
+ domains, there is a basis E1, E2, . . . , Ed of the free module V such that
769
+ E1 = E,
770
+ (10)
771
+ a1E2 = (T βˆ’ Ξ»1IdV )E1,
772
+ a2E3 = (T βˆ’ Ξ»2IdV )E2,
773
+ . . .
774
+ asβˆ’1Ed = (T βˆ’ Ξ»dβˆ’1IdV )Edβˆ’1.
775
+ Let us consider the module V1 = ⟨E1, . . . , Ed⟩ βŠ‚ V . By construction, the map T
776
+ restricts to an automorphism V1 β†’ V1 with respect to the basis E1, . . . , Ed has the
777
+ desired form. We then consider the free module V/V1 and we repeat the procedure
778
+ for the minimal polynomial of T, which again acts on V/V1. The desired result
779
+ follows.
780
+ β–‘
781
+ Remark 17. The element T as defined in eq. (9) has order equal to the higher
782
+ order of the eigenvalues Ξ»1, . . . , Ξ»d involved. Indeed, since we have assumed that
783
+ the eigenvalues are different the matrix is diagonalizable in Quot(R) and has order
784
+ equal to the maximal order of the eigenvalues involved. In particular it has order q
785
+ if there is at least one Ξ»i that is a primitive q-root of unity. The statement about
786
+ the order of T is not necessarily true if some of the eigenvalues are the same. For
787
+ instance the matrix
788
+ οΏ½
789
+ 1
790
+ 0
791
+ 1
792
+ 1
793
+ οΏ½
794
+ has infinite order over a field of characteristic zero.
795
+ Remark 18. The number of indecomposable R[T]-summands of V is given by
796
+ #{i : ai = 0} + 1.
797
+ A lift of a sum of indecomposable kCq-modules JΞΊ1 βŠ• Β· Β· Β· βŠ• JΞΊn can form an
798
+ indecomposable RCq-module. For example the indecomposable module where the
799
+
800
+ ON THE LIFTING PROBLEM OF REPRESENTATIONS OF A METACYCLIC GROUP.
801
+ 11
802
+ generator T of Cq has the form
803
+ T =
804
+ οΏ½
805
+ οΏ½
806
+ οΏ½
807
+ οΏ½
808
+ οΏ½
809
+ οΏ½
810
+ οΏ½
811
+ οΏ½
812
+ οΏ½
813
+ Ξ»1
814
+ 0
815
+ Β· Β· Β·
816
+ Β· Β· Β·
817
+ 0
818
+ a1
819
+ Ξ»2
820
+ ...
821
+ ...
822
+ 0
823
+ a2
824
+ Ξ»3
825
+ ...
826
+ ...
827
+ ...
828
+ ...
829
+ ...
830
+ ...
831
+ 0
832
+ 0
833
+ Β· Β· Β·
834
+ 0
835
+ asβˆ’1
836
+ Ξ»d
837
+ οΏ½
838
+ οΏ½
839
+ οΏ½
840
+ οΏ½
841
+ οΏ½
842
+ οΏ½
843
+ οΏ½
844
+ οΏ½
845
+ οΏ½
846
+ where a1 = Β· Β· Β· = aΞΊ1βˆ’1 = 1, aΞΊ1 ∈ mR, aΞΊ1+1, . . . , aΞΊ2+ΞΊ1βˆ’1 = 1, aΞΊ2+ΞΊ1 ∈
847
+ mR , etc reduces to a decomposable direct sum of Jordan normal forms of sizes
848
+ JΞΊ1, JΞΊ2βˆ’ΞΊ1, Β· Β· Β· .
849
+ Remark 19. It is an interesting question to classify these matrices up to conju-
850
+ gation with a matrix in GLd(R). It seems that the valuation of elements ai should
851
+ also play a role.
852
+ Definition 20. Let hi(x1, . . . , xj) be the complete symmetric polynomial of degree
853
+ i in the variables x1, . . . , xj. For instance
854
+ h3(x1, x2, x3) = x3
855
+ 1 + x2
856
+ 1x2 + x2
857
+ 1x3 + x1x2
858
+ 2 + x1x2x3 + x1x2
859
+ 3 + x3
860
+ 2 + x2
861
+ 2x3 + x2x2
862
+ 3 + x3
863
+ 3.
864
+ Set
865
+ L(ΞΊ, j, Ξ½) = hΞΊ(Ξ»j, Ξ»j+1, . . . , Ξ»j+Ξ½)
866
+ A(i, j) =
867
+ οΏ½
868
+ aiai+1 Β· Β· Β· ai+j
869
+ if j β‰₯ 0
870
+ 0
871
+ if j < 0
872
+ Lemma 21. The matrix T Ξ± = (t(Ξ±)
873
+ ij ) is given by the following formula:
874
+ t(Ξ±)
875
+ ij
876
+ =
877
+ οΏ½
878
+ οΏ½
879
+ οΏ½
880
+ οΏ½
881
+ οΏ½
882
+ λα
883
+ i
884
+ if i = j
885
+ A(j, i βˆ’ j βˆ’ 1) Β· L(Ξ± βˆ’ (i βˆ’ j), j, i βˆ’ j)
886
+ if j < i
887
+ 0
888
+ if j > i
889
+ Proof. For j β‰₯ i the proof is trivial. When j < i and Ξ± = 1 it is immediate, since
890
+ L(x, Β·, Β·) ≑ 0, for every x ≀ 0. Assume this holds for Ξ± = n. If Ξ± = n + 1,
891
+ t(n+1)
892
+ ij
893
+ = t(n)
894
+ ij tij =
895
+ r
896
+ οΏ½
897
+ k=1
898
+ t(Ξ±)
899
+ ik tkj = Ξ»jt(Ξ±)
900
+ ij
901
+ + ajt(Ξ±)
902
+ ij+1 = Ξ»jA(j, i βˆ’ j βˆ’ 1)L(Ξ± βˆ’ (i βˆ’ j), j, i βˆ’ j)+
903
+ + ajA(j + 1, i βˆ’ j βˆ’ 2)L(Ξ± βˆ’ (i βˆ’ j βˆ’ 1), j + 1, i βˆ’ j βˆ’ 1) =
904
+ = A(j, i βˆ’ j βˆ’ 1)
905
+ οΏ½
906
+ Ξ»jhΞ±βˆ’(iβˆ’j)(Ξ»j, . . . , Ξ»j) + hΞ±βˆ’(iβˆ’j)+1(Ξ»j+1, . . . , Ξ»i)
907
+ οΏ½
908
+ =
909
+ = A(j, i βˆ’ j βˆ’ 1)hΞ±βˆ’(iβˆ’j)+1(Ξ»j, . . . , Ξ»i) =
910
+ = A(j, i βˆ’ j βˆ’ 1)L(Ξ± βˆ’ (i βˆ’ j) + 1, i, i βˆ’ j).
911
+ β–‘
912
+ Remark 22. The space of homogeneous polynomials of degree k in n-variables
913
+ has dimension
914
+ οΏ½nβˆ’1+c
915
+ nβˆ’1
916
+ οΏ½
917
+ . Since all q-roots of unity are reduced to 1 modulo mR the
918
+ quantity L(Ξ± βˆ’ (i βˆ’ j), j, i βˆ’ j) is reduced to n = (i βˆ’ j) + 1, c = Ξ± βˆ’ (i βˆ’ j)
919
+ οΏ½n βˆ’ 1 + c
920
+ n βˆ’ 1
921
+ οΏ½
922
+ =
923
+ οΏ½ Ξ±
924
+ i βˆ’ j
925
+ οΏ½
926
+ .
927
+ This equation is compatible with the computation of Ο„ Ξ± given in eq. (4).
928
+
929
+ 12
930
+ A. KONTOGEORGIS AND A. TEREZAKIS
931
+ Lemma 23. There is an eigenvector E of the generator Οƒ of the cyclic group Cm
932
+ which is not an element in
933
+ sοΏ½
934
+ i=1
935
+ Ker(Ξ i βŠ— K).
936
+ Proof. The eigenvectors E1, . . . , Ed of Οƒ form a basis of the space V βŠ— K.
937
+ By
938
+ multiplying by certain elements in R, if necessary, we can assume that all Ei are in
939
+ V and their reductions Ei βŠ— R/mR, 1 ≀ i ≀ d give rise to a basis of eigenvectors of
940
+ a generator of the cyclic group Cm acting on V βŠ— R/mR. If every eigenvector Ei is
941
+ an element of some Ker(Ξ Ξ½) for 1 ≀ i ≀ d, then their reductions will be elements
942
+ in Ker(T βˆ’ 1)dβˆ’1, a contradiction since the later kernel has dimension < d.
943
+ β–‘
944
+ Lemma 24. Let V be a free Cq β‹Š Cm-module, which is indecomposable as a Cq-
945
+ module. Consider the basis given in lemma 16. Then the value of Οƒ(E1) determines
946
+ Οƒ(Ei) for 2 ≀ i ≀ d.
947
+ Proof. Let Οƒ be a generator of the cyclic group Cm. We will use the notation of
948
+ lemma 15. We use lemma 23 in order to select a suitable eigenvector of E1 of Οƒ
949
+ and then form the basis E1, E2, . . . , Ed as given in eq. (10). We can compute the
950
+ action of Οƒ on all basis elements Ei by
951
+ (11)
952
+ Οƒ(aiβˆ’1Ei) = Οƒ(T βˆ’ Ξ»iβˆ’1IdV )Eiβˆ’1 = (T a βˆ’ Ξ»iβˆ’1IdV )Οƒ(Eiβˆ’1).
953
+ This means that one can define recursively the action of Οƒ on all elements Ei.
954
+ Indeed, assume that
955
+ Οƒ(Eiβˆ’1) =
956
+ d
957
+ οΏ½
958
+ Ξ½=1
959
+ Ξ³Ξ½,iβˆ’1EΞ½.
960
+ We now have
961
+ (T a βˆ’ Ξ»iβˆ’1IdV )EΞ½ =
962
+ d
963
+ οΏ½
964
+ Β΅=1
965
+ t(Ξ±)
966
+ Β΅,Ξ½EΒ΅ βˆ’ Ξ»iβˆ’1EΞ½
967
+ = (λα
968
+ Ξ½ βˆ’ Ξ»iβˆ’1)EΞ½ +
969
+ d
970
+ οΏ½
971
+ Β΅=Ξ½+1
972
+ t(Ξ±)
973
+ Β΅,Ξ½EΒ΅
974
+ We combine all the above to
975
+ aiβˆ’1Οƒ(Ei) =
976
+ d
977
+ οΏ½
978
+ Ξ½=1
979
+ Ξ³Ξ½,iβˆ’1(λα
980
+ Ξ½ βˆ’ Ξ»iβˆ’1)EΞ½ +
981
+ d
982
+ οΏ½
983
+ Ξ½=1
984
+ Ξ³Ξ½,iβˆ’1
985
+ d
986
+ οΏ½
987
+ Β΅=Ξ½+1
988
+ t(Ξ±)
989
+ Β΅,Ξ½EΒ΅
990
+ =
991
+ d
992
+ οΏ½
993
+ Ξ½=1
994
+ ˜γν,iEν,
995
+ (12)
996
+ for a selection of elements γν,i ∈ R, which can be explicitly computed by collecting
997
+ the coefficients of the basis elements E1, . . . , Ed.
998
+ Observe that the quantity on the right hand side of eq. (12) must be divisible
999
+ by aiβˆ’1. Indeed, let v be the valuation of the local principal ideal domain R. Set
1000
+ e0 = min
1001
+ 1≀ν≀d{v(˜γν,i)}.
1002
+ If e0 < v(aiβˆ’1) then we divide eq. (12) by Ο€e0 where Ο€ is the local uniformizer of
1003
+ R, that is mR = Ο€R. We then consider the divided equation modulo mR to obtain
1004
+
1005
+ ON THE LIFTING PROBLEM OF REPRESENTATIONS OF A METACYCLIC GROUP.
1006
+ 13
1007
+ a linear dependence relation among the elements Ei βŠ— k, which is a contradiction.
1008
+ Therefore e0 β‰₯ v(aiβˆ’1) and we obtain an equation
1009
+ Οƒ(Ei) =
1010
+ d
1011
+ οΏ½
1012
+ Ξ½=1
1013
+ ˜γν,i
1014
+ aiβˆ’1
1015
+ EΞ½ =
1016
+ d
1017
+ οΏ½
1018
+ Ξ½=1
1019
+ Ξ³Ξ½,iEΞ½.
1020
+ β–‘
1021
+ For example Οƒ(E1) = ΞΆΟ΅
1022
+ mE1. We compute that
1023
+ a1Οƒ(E2) = (T Ξ± βˆ’ Ξ»1Id)Οƒ(E1)
1024
+ and
1025
+ Οƒ(E2) = (λα
1026
+ 1 βˆ’ Ξ»1)
1027
+ a1
1028
+ ΞΆΟ΅
1029
+ Β΅E1 + ΞΆΟ΅
1030
+ m
1031
+ d
1032
+ οΏ½
1033
+ Β΅=2
1034
+ t(Ξ±)
1035
+ Β΅,1
1036
+ a1
1037
+ EΒ΅
1038
+ = (λα
1039
+ 1 βˆ’ Ξ»1)
1040
+ a1
1041
+ ΞΆΟ΅
1042
+ Β΅E1 + ΞΆΟ΅
1043
+ m
1044
+ d
1045
+ οΏ½
1046
+ Β΅=2
1047
+ A(1, Β΅ βˆ’ 2)L(Ξ± βˆ’ (Β΅ βˆ’ 1), 1, Β΅ βˆ’ 1)
1048
+ a1
1049
+ EΒ΅
1050
+ = (λα
1051
+ 1 βˆ’ Ξ»1)
1052
+ a1
1053
+ ΞΆΟ΅
1054
+ Β΅E1 + ΞΆΟ΅
1055
+ m
1056
+ d
1057
+ οΏ½
1058
+ Β΅=2
1059
+ a1a2 Β· Β· Β· aΒ΅βˆ’1hΞ±βˆ’(Β΅βˆ’1)(Ξ»1, Ξ»2, . . . , λ¡)
1060
+ a1
1061
+ EΒ΅.
1062
+ Proposition 25. Assume that no element a1, . . . , adβˆ’1 given in eq. (9) is zero.
1063
+ Given Ξ± ∈ N, Ξ± β‰₯ 1 and an element E1, which is not an element in οΏ½d
1064
+ i=1 Ker(Ξ i βŠ—
1065
+ K), if there is a matrix Ξ“ = (Ξ³ij), such that Ξ“TΞ“βˆ’1 = T Ξ± and Ξ“E1 = ΞΆΟ΅
1066
+ mE1, then
1067
+ this matrix Ξ“ is unique.
1068
+ Proof. We will use the idea leading to equation (11) replacing Οƒ with Ξ“. We will
1069
+ compute recursively and uniquely the entries Ξ³Β΅,i, arriving at the explicit formula
1070
+ of eq. (18).
1071
+ Observe that trivially Ξ³Ξ½,1 = 0 for all Ξ½ < 1 since we only allow 1 ≀ Ξ½ ≀ d. We
1072
+ compute
1073
+ ˜γ¡,i = Ξ³Β΅,iβˆ’1(λα
1074
+ Β΅ βˆ’ Ξ»iβˆ’1) +
1075
+ Β΅βˆ’1
1076
+ οΏ½
1077
+ Ξ½=1
1078
+ Ξ³Ξ½,iβˆ’1t(Ξ±)
1079
+ Β΅,Ξ½
1080
+ (13)
1081
+ = Ξ³Β΅,iβˆ’1(λα
1082
+ Β΅ βˆ’ Ξ»iβˆ’1) +
1083
+ Β΅βˆ’1
1084
+ οΏ½
1085
+ Ξ½=1
1086
+ Ξ³Ξ½,iβˆ’1A(Ξ½, Β΅ βˆ’ Ξ½ βˆ’ 1)L
1087
+ οΏ½
1088
+ Ξ± βˆ’ (Β΅ βˆ’ Ξ½), Ξ½, Β΅ βˆ’ Ξ½)
1089
+ = Ξ³Β΅,iβˆ’1(λα
1090
+ Β΅ βˆ’ Ξ»iβˆ’1) +
1091
+ Β΅βˆ’1
1092
+ οΏ½
1093
+ Ξ½=1
1094
+ Ξ³Ξ½,iβˆ’1aΞ½aΞ½+1 Β· Β· Β· aΒ΅βˆ’1hΞ±βˆ’Β΅+Ξ½(λν, λν+1, . . . , λ¡)
1095
+ Define
1096
+ [λα
1097
+ m βˆ’ Ξ»x]j
1098
+ i =
1099
+ jοΏ½
1100
+ x=i
1101
+ (λα
1102
+ Β΅ βˆ’ Ξ»x)
1103
+ [a]j
1104
+ i =
1105
+ jοΏ½
1106
+ x=i
1107
+ ax
1108
+ for i ≀ j. If i > j then both of the above quantities are defined to be equal to 1.
1109
+
1110
+ 14
1111
+ A. KONTOGEORGIS AND A. TEREZAKIS
1112
+ Observe that for Β΅ = 1 eq. (13) becomes
1113
+ (14)
1114
+ Ξ³1,i =
1115
+ 1
1116
+ aiβˆ’1
1117
+ Ξ³1,iβˆ’1(λα
1118
+ 1 βˆ’ Ξ»iβˆ’1)
1119
+ and we arrive at (assuming that Ξ“(E1) = ΞΆΟ΅
1120
+ mE1)
1121
+ (15)
1122
+ Ξ³1,i =
1123
+ ΞΆΟ΅
1124
+ m
1125
+ a1a2 Β· Β· Β· aiβˆ’1
1126
+ iβˆ’1
1127
+ οΏ½
1128
+ x=1
1129
+ (λα
1130
+ 1 βˆ’ Ξ»x) =
1131
+ ΞΆΟ΅
1132
+ m
1133
+ a1a2 Β· Β· Β· aiβˆ’1
1134
+ [λα
1135
+ 1 βˆ’ Ξ»x]iβˆ’1
1136
+ 1
1137
+ .
1138
+ For Β΅ β‰₯ 2 we have Ξ³Β΅,1 = 0, since by assumption TE1 = ΞΆΟ΅
1139
+ mE1. Therefore eq. (13)
1140
+ gives us
1141
+ Ξ³Β΅,i =
1142
+ iβˆ’2
1143
+ οΏ½
1144
+ ΞΊ1=0
1145
+ [λα
1146
+ Β΅ βˆ’ Ξ»x]iβˆ’1
1147
+ iβˆ’ΞΊ1
1148
+ [a]iβˆ’1
1149
+ iβˆ’1βˆ’ΞΊ1
1150
+ Β΅βˆ’1
1151
+ οΏ½
1152
+ Β΅2=1
1153
+ Ξ³Β΅2,iβˆ’1βˆ’ΞΊ1[a]Β΅βˆ’1
1154
+ Β΅2 hΞ±βˆ’Β΅+Β΅2(λ¡2, . . . , λ¡)
1155
+ =
1156
+ Β΅βˆ’1
1157
+ οΏ½
1158
+ Β΅2=1
1159
+ [a]Β΅βˆ’1
1160
+ Β΅2 hΞ±βˆ’Β΅+Β΅2(λ¡2, . . . , λ¡)
1161
+ iβˆ’2
1162
+ οΏ½
1163
+ ΞΊ1=0
1164
+ [λα
1165
+ Β΅ βˆ’ Ξ»x]iβˆ’1
1166
+ iβˆ’ΞΊ1
1167
+ [a]iβˆ’1
1168
+ iβˆ’1βˆ’ΞΊ1
1169
+ Ξ³Β΅2,iβˆ’1βˆ’ΞΊ1.
1170
+ (16)
1171
+ We will now prove eq. (16) by induction on i. For i = 2, Β΅ β‰₯ 2 we have
1172
+ Ξ³Β΅,2 = 1
1173
+ a1
1174
+ γ¡,1(λα
1175
+ Β΅ βˆ’ Ξ»1) + 1
1176
+ a1
1177
+ Β΅βˆ’1
1178
+ οΏ½
1179
+ Β΅2=1
1180
+ Ξ³Β΅2,1[a]Β΅βˆ’1
1181
+ Β΅2 hΞ±βˆ’Β΅+Β΅2(λ¡2, . . . , λ¡)
1182
+ = 1
1183
+ a1
1184
+ [a]Β΅βˆ’1
1185
+ 1
1186
+ hΞ±βˆ’Β΅+1(Ξ»1, . . . , λ¡)Ξ³1,1.
1187
+
1188
+ ON THE LIFTING PROBLEM OF REPRESENTATIONS OF A METACYCLIC GROUP.
1189
+ 15
1190
+ Assume now that eq. (16) holds for computing Ξ³Β΅,iβˆ’1. We will treat the Ξ³Β΅,i case.
1191
+ We have
1192
+ γ¡,i = (λα
1193
+ Β΅ βˆ’ Ξ»iβˆ’1)
1194
+ aiβˆ’1
1195
+ Ξ³Β΅,iβˆ’1 +
1196
+ 1
1197
+ aiβˆ’1
1198
+ Β΅βˆ’1
1199
+ οΏ½
1200
+ Β΅2=1
1201
+ Ξ³Β΅2,iβˆ’1[a]Β΅βˆ’1
1202
+ Β΅2 hΞ±βˆ’Β΅+Β΅2(λ¡2, . . . , λ¡)
1203
+ = (λα
1204
+ Β΅ βˆ’ Ξ»iβˆ’1)
1205
+ aiβˆ’1
1206
+ Β΅βˆ’1
1207
+ οΏ½
1208
+ Β΅2=1
1209
+ [a]Β΅βˆ’1
1210
+ Β΅2 hΞ±βˆ’Β΅+Β΅2(λ¡2, . . . , λ¡)
1211
+ iβˆ’3
1212
+ οΏ½
1213
+ ΞΊ1=0
1214
+ [λα
1215
+ Β΅ βˆ’ Ξ»x]iβˆ’2
1216
+ iβˆ’1βˆ’ΞΊ1
1217
+ [a]iβˆ’2
1218
+ iβˆ’2βˆ’ΞΊ1
1219
+ Ξ³Β΅2,iβˆ’2βˆ’ΞΊ1
1220
+ +
1221
+ 1
1222
+ aiβˆ’1
1223
+ Β΅βˆ’1
1224
+ οΏ½
1225
+ Β΅2=1
1226
+ Ξ³Β΅2,iβˆ’1[a]Β΅βˆ’1
1227
+ Β΅2 hΞ±βˆ’Β΅+Β΅2(λ¡2, . . . , λ¡)
1228
+ =
1229
+ Β΅βˆ’1
1230
+ οΏ½
1231
+ Β΅2=1
1232
+ [a]Β΅βˆ’1
1233
+ Β΅2 hΞ±βˆ’Β΅+Β΅2(λ¡2, . . . , λ¡)
1234
+ iβˆ’3
1235
+ οΏ½
1236
+ ΞΊ1=0
1237
+ [λα
1238
+ Β΅ βˆ’ Ξ»x]iβˆ’1
1239
+ iβˆ’1βˆ’ΞΊ1
1240
+ [a]iβˆ’1
1241
+ iβˆ’2βˆ’ΞΊ1
1242
+ Ξ³Β΅2,iβˆ’2βˆ’ΞΊ1
1243
+ +
1244
+ 1
1245
+ aiβˆ’1
1246
+ Β΅βˆ’1
1247
+ οΏ½
1248
+ Β΅2=1
1249
+ Ξ³Β΅2,iβˆ’1[a]Β΅βˆ’1
1250
+ Β΅2 hΞ±βˆ’Β΅+Β΅2(λ¡2, . . . , λ¡)
1251
+ =
1252
+ Β΅βˆ’1
1253
+ οΏ½
1254
+ Β΅2=1
1255
+ [a]Β΅βˆ’1
1256
+ Β΅2 hΞ±βˆ’Β΅+Β΅2(λ¡2, . . . , λ¡)
1257
+ iβˆ’2
1258
+ οΏ½
1259
+ ΞΊ1=1
1260
+ [λα
1261
+ Β΅ βˆ’ Ξ»x]iβˆ’1
1262
+ iβˆ’ΞΊ1
1263
+ [a]iβˆ’1
1264
+ iβˆ’1βˆ’ΞΊ1
1265
+ Ξ³Β΅2,iβˆ’1βˆ’ΞΊ1
1266
+ +
1267
+ Β΅βˆ’1
1268
+ οΏ½
1269
+ Β΅2=1
1270
+ [a]Β΅βˆ’1
1271
+ Β΅2 hΞ±βˆ’Β΅+Β΅2(λ¡2, . . . , λ¡)
1272
+ 1
1273
+ aiβˆ’1
1274
+ Ξ³Β΅2,iβˆ’1
1275
+ =
1276
+ Β΅βˆ’1
1277
+ οΏ½
1278
+ Β΅2=1
1279
+ [a]Β΅βˆ’1
1280
+ Β΅2 hΞ±βˆ’Β΅+Β΅2(λ¡2, . . . , λ¡)
1281
+ iβˆ’2
1282
+ οΏ½
1283
+ ΞΊ1=0
1284
+ [λα
1285
+ Β΅ βˆ’ Ξ»x]iβˆ’1
1286
+ iβˆ’ΞΊ1
1287
+ [a]iβˆ’1
1288
+ iβˆ’1βˆ’ΞΊ1
1289
+ Ξ³Β΅2,iβˆ’1βˆ’ΞΊ1
1290
+ and equation (16) is now proved.
1291
+ We proceed recursively applying eq. (16) to each of the summands Ξ³Β΅2,iβˆ’1βˆ’ΞΊ1
1292
+ if Β΅2 > 1 and i βˆ’ 1 βˆ’ ΞΊ1 > 1. If Β΅2 = 1, then Ξ³Β΅2,iβˆ’1βˆ’ΞΊ1 is computed by eq. (14)
1293
+ and if Β΅2 > 1 and i βˆ’ 1 βˆ’ ΞΊ1 ≀ 1 then Ξ³Β΅2,iβˆ’1βˆ’ΞΊ1 = 0. We can classify all iterations
1294
+ needed by the set Σ¡ of sequences (Β΅s, Β΅sβˆ’1, . . . , Β΅3, Β΅2) such that
1295
+ (17)
1296
+ 1 = Β΅s < Β΅sβˆ’1 < Β· Β· Β· < Β΅3 < Β΅2 < Β΅ = Β΅1.
1297
+ For example for Β΅ = 5 the set of such sequences is given by
1298
+ Σ¡ = {(1), (1, 2), (1, 3), (1, 2, 3), (1, 4), (1, 2, 4), (1, 3, 4), (1, 2, 3, 4)}
1299
+ corresponding to the tree of iterations given in figure 1. The length of the sequence
1300
+ (Β΅s, Β΅sβˆ’1, . . . , Β΅2) is given in eq. (17) is s βˆ’ 1. In each iteration the i changes to
1301
+ i βˆ’ 1 βˆ’ k thus we have the following sequence of indices
1302
+ i1 = i β†’ i2 = iβˆ’1βˆ’ΞΊ1 β†’ i3 = iβˆ’2βˆ’(ΞΊ1+ΞΊ2) β†’ Β· Β· Β· β†’ is = iβˆ’(sβˆ’1)βˆ’(ΞΊ1+Β· Β· Β·+ΞΊsβˆ’1)
1303
+ For the sequence i1, i2, . . . , we might have it = 1 for t < s βˆ’ 1. But in this case,
1304
+ we will arrive at the element Ξ³Β΅t+1,it = Ξ³Β΅t,1 = 0 since Β΅t > 1. This means that we
1305
+ will have to consider only selections ΞΊ1, . . . , ΞΊsβˆ’1 such that isβˆ’1 β‰₯ 1. Therefore we
1306
+
1307
+ 16
1308
+ A. KONTOGEORGIS AND A. TEREZAKIS
1309
+ Β΅ = 5
1310
+ Β΅2 = 1
1311
+ Β΅2 = 2
1312
+ Β΅3 = 1
1313
+ Β΅2 = 3
1314
+ Β΅3 = 1
1315
+ Β΅3 = 2
1316
+ Β΅4 = 1
1317
+ Β΅2 = 4
1318
+ Β΅3 = 1
1319
+ Β΅3 = 2
1320
+ Β΅4 = 1
1321
+ Β΅3 = 3
1322
+ Β΅4 = 1
1323
+ Β΅4 = 2
1324
+ Β΅5 = 1
1325
+ Figure 1.
1326
+ Iteration tree for Β΅ = 5
1327
+ arrive at the following expression for Β΅ β‰₯ 2
1328
+ Ξ³Β΅,i =
1329
+ οΏ½
1330
+ (¡s,...,¡2)∈Σ¡
1331
+ [a]Β΅βˆ’1
1332
+ Β΅2 [a]Β΅2βˆ’1
1333
+ Β΅3
1334
+ Β· Β· Β· [a]Β΅sβˆ’1βˆ’1
1335
+ Β΅s
1336
+ s
1337
+ οΏ½
1338
+ Ξ½=2
1339
+ hΞ±βˆ’Β΅Ξ½βˆ’1+¡ν(λ¡ν, . . . , Ξ»Β΅Ξ½βˆ’1)
1340
+ Β·
1341
+ οΏ½
1342
+ i=i1>i2>Β·Β·Β·>isβ‰₯1
1343
+ sβˆ’1
1344
+ οΏ½
1345
+ Ξ½=1
1346
+ [λα
1347
+ ¡ν βˆ’ Ξ»x]iΞ½βˆ’1
1348
+ iΞ½+1+1
1349
+ [a]iΞ½βˆ’1
1350
+ iΞ½+1
1351
+ Β· Ξ³1,is.
1352
+ =
1353
+ οΏ½
1354
+ (¡s,...,¡2)∈Σ¡
1355
+ s
1356
+ οΏ½
1357
+ Ξ½=2
1358
+ hΞ±βˆ’Β΅Ξ½βˆ’1+¡ν(λ¡ν, . . . , Ξ»Β΅Ξ½βˆ’1)
1359
+ Β·
1360
+ οΏ½
1361
+ i=i1>i2>Β·Β·Β·>isβ‰₯1
1362
+ [a]Β΅βˆ’1
1363
+ 1
1364
+ [a]iβˆ’1
1365
+ is
1366
+ sβˆ’1
1367
+ οΏ½
1368
+ Ξ½=1
1369
+ [λα
1370
+ ¡ν βˆ’ Ξ»x]iΞ½βˆ’1
1371
+ iΞ½+1+1
1372
+ ΞΆΟ΅
1373
+ m[λα
1374
+ 1 βˆ’ Ξ»x]isβˆ’1
1375
+ 1
1376
+ [a]isβˆ’1
1377
+ 1
1378
+ =
1379
+ οΏ½
1380
+ (¡s,...,¡2)∈Σ¡
1381
+ s
1382
+ οΏ½
1383
+ Ξ½=2
1384
+ hΞ±βˆ’Β΅Ξ½βˆ’1+¡ν(λ¡ν, . . . , Ξ»Β΅Ξ½βˆ’1)[a]Β΅βˆ’1
1385
+ 1
1386
+ [a]iβˆ’1
1387
+ 1
1388
+ ΞΆΟ΅
1389
+ m
1390
+ οΏ½
1391
+ i=i1>i2>Β·Β·Β·>isβ‰₯1
1392
+ s
1393
+ οΏ½
1394
+ Ξ½=1
1395
+ [λα
1396
+ ¡ν βˆ’ Ξ»x]iΞ½βˆ’1
1397
+ iΞ½+1+1
1398
+ (18)
1399
+ where is+1 + 1 = 1 that is is+1 = 0.
1400
+ β–‘
1401
+ We will now prove that the matrix Ξ“ of lemma 25 exists by cheking that Ξ“T =
1402
+ T Ξ±Ξ“. Set (aΒ΅,i) = Ξ“T, (bΒ΅,i) = T Ξ±Ξ“. For i < d we have
1403
+ aΒ΅,i =
1404
+ d
1405
+ οΏ½
1406
+ Ξ½=1
1407
+ Ξ³Β΅,Ξ½tΞ½,i = Ξ³Β΅,itii + Ξ³Β΅,i+1ti+1,i
1408
+ = γ¡,iλi + γ¡,i(λα
1409
+ Β΅ βˆ’ Ξ»i) +
1410
+ Β΅βˆ’1
1411
+ οΏ½
1412
+ Ξ½=1
1413
+ Ξ³Ξ½,it(Ξ±)
1414
+ Β΅,Ξ½
1415
+ = γ¡,iλα
1416
+ Β΅ +
1417
+ Β΅βˆ’1
1418
+ οΏ½
1419
+ Ξ½=1
1420
+ Ξ³Ξ½,it(Ξ±)
1421
+ Β΅,Ξ½ =
1422
+ Β΅
1423
+ οΏ½
1424
+ Ξ½=1
1425
+ t(Ξ±)
1426
+ Β΅,Ξ½Ξ³Ξ½,i = bΒ΅,i.
1427
+ For i = d we have:
1428
+ aΒ΅,d =
1429
+ d
1430
+ οΏ½
1431
+ Ξ½=1
1432
+ Ξ³Β΅,Ξ½tΞ½,d = Ξ³Β΅,dtd,d = Ξ³Β΅,dΞ»d
1433
+
1434
+ ON THE LIFTING PROBLEM OF REPRESENTATIONS OF A METACYCLIC GROUP.
1435
+ 17
1436
+ while
1437
+ bΒ΅,d =
1438
+ d
1439
+ οΏ½
1440
+ Ξ½=1
1441
+ t(Ξ±)
1442
+ Β΅,Ξ½Ξ³Ξ½,d =
1443
+ Β΅βˆ’1
1444
+ οΏ½
1445
+ Ξ½=1
1446
+ t(Ξ±)
1447
+ ¡,νγν,d + λα
1448
+ ¡γ¡,d
1449
+ This gives us the relation
1450
+ (19)
1451
+ (Ξ»d βˆ’ Ξ»a
1452
+ Β΅)Ξ³Β΅,d =
1453
+ Β΅βˆ’1
1454
+ οΏ½
1455
+ Ξ½=1
1456
+ t(Ξ±)
1457
+ Β΅,Ξ½Ξ³Ξ½,d
1458
+ For Β΅ = 1 using eq. (15) we have
1459
+ γ1,dλd = γ1,dλα
1460
+ 1 β‡’ [λα
1461
+ 1 βˆ’ Ξ»x]d
1462
+ 1 = 0.
1463
+ This relation is satisfied if λα
1464
+ 1 is one of {Ξ»1, . . . , Ξ»d}. Without loss of generality we
1465
+ assume that
1466
+ (20)
1467
+ Ξ»(a)
1468
+ i
1469
+ =
1470
+ οΏ½
1471
+ Ξ»i+1
1472
+ if m ∀ i
1473
+ Ξ»iβˆ’m+1
1474
+ if m | i
1475
+ We have the following conditions:
1476
+ Β΅ = 2
1477
+ (Ξ»d βˆ’ λα
1478
+ 2 )Ξ³2,d = t(Ξ±)
1479
+ 2,1 Ξ³1,d
1480
+ Β΅ = 3
1481
+ (Ξ»d βˆ’ λα
1482
+ 3 )Ξ³3,d = t(Ξ±)
1483
+ 3,1 Ξ³1,d + t(Ξ±)
1484
+ 3,2 Ξ³2,d
1485
+ Β΅ = 4
1486
+ (Ξ»d βˆ’ λα
1487
+ 4 )Ξ³4,d = t(Ξ±)
1488
+ 4,1 Ξ³1,d + t(Ξ±)
1489
+ 4,2 Ξ³2,d + t(Ξ±)
1490
+ 4,3 Ξ³3,d
1491
+ ...
1492
+ ...
1493
+ Β΅ = d βˆ’ 1
1494
+ (Ξ»d βˆ’ λα
1495
+ dβˆ’1)Ξ³dβˆ’1,d = t(Ξ±)
1496
+ dβˆ’1,1Ξ³1,d + t(Ξ±)
1497
+ dβˆ’1,2Ξ³2,d + Β· Β· Β· + t(Ξ±)
1498
+ dβˆ’1,dβˆ’2Ξ³dβˆ’1,d
1499
+ All these equations are true provided that Ξ³1,d, . . . , Ξ³dβˆ’2,d = 0. Finally, for Β΅ = d,
1500
+ we have
1501
+ (21)
1502
+ (Ξ»d βˆ’ λα
1503
+ d )Ξ³d,d =
1504
+ dβˆ’1
1505
+ οΏ½
1506
+ Ξ½=1
1507
+ t(Ξ±)
1508
+ d,Ξ½Ξ³Ξ½,d
1509
+ which is true provided that (Ξ»d βˆ’ λα
1510
+ d )Ξ³d,d = t(a)
1511
+ d,dβˆ’1Ξ³dβˆ’1,d.
1512
+ Lemma 26. For n β‰₯ 2 the vertical sum Sn of the products of every line of the
1513
+ following array
1514
+ y
1515
+ 1
1516
+ 1
1517
+ (x1 βˆ’ x2)
1518
+ (x1 βˆ’ x3)
1519
+ Β· Β· Β·
1520
+ Β· Β· Β·
1521
+ (x1 βˆ’ xn)
1522
+ 2
1523
+ (z βˆ’ x1)
1524
+ 1
1525
+ (x1 βˆ’ x3)
1526
+ Β· Β· Β·
1527
+ Β· Β· Β·
1528
+ (x1 βˆ’ xn)
1529
+ 3
1530
+ (z βˆ’ x1)
1531
+ (z βˆ’ x2)
1532
+ 1
1533
+ ...
1534
+ ...
1535
+ ...
1536
+ ...
1537
+ ...
1538
+ ...
1539
+ ...
1540
+ ...
1541
+ ...
1542
+ ...
1543
+ ...
1544
+ ...
1545
+ ...
1546
+ n βˆ’ 1
1547
+ (z βˆ’ x1)
1548
+ (z βˆ’ x2)
1549
+ Β· Β· Β·
1550
+ (z βˆ’ xnβˆ’2)
1551
+ 1
1552
+ (x1 βˆ’ xn)
1553
+ n
1554
+ (z βˆ’ x1)
1555
+ (z βˆ’ x2)
1556
+ Β· Β· Β·
1557
+ (z βˆ’ xnβˆ’2)
1558
+ (z βˆ’ xnβˆ’1)
1559
+ 1
1560
+ is given by
1561
+ Sn =
1562
+ n
1563
+ οΏ½
1564
+ y=1
1565
+ n
1566
+ οΏ½
1567
+ Ξ½=y+1
1568
+ (x1 βˆ’ xΞ½)
1569
+ yβˆ’1
1570
+ οΏ½
1571
+ Β΅=1
1572
+ (z βˆ’ xΒ΅) = (z βˆ’ x2) Β· Β· Β· (z βˆ’ xn).
1573
+
1574
+ 18
1575
+ A. KONTOGEORGIS AND A. TEREZAKIS
1576
+ In particular when z = xn the sum is zero.
1577
+ Proof. We will prove the lemma by induction. For n = 2 we have S2 = (x1 βˆ’ x2) +
1578
+ (zβˆ’x1) = zβˆ’x2. Assume that the equality holds for n. The sum Sn+1 corresponds
1579
+ to the array:
1580
+ y
1581
+ 1
1582
+ 1
1583
+ (x1 βˆ’ x2)
1584
+ (x1 βˆ’ x3)
1585
+ Β· Β· Β·
1586
+ (x1 βˆ’ xn)
1587
+ (x1 βˆ’ xn+1)
1588
+ 2
1589
+ (z βˆ’ x1)
1590
+ 1
1591
+ (x1 βˆ’ x3)
1592
+ Β· Β· Β·
1593
+ (x1 βˆ’ xn)
1594
+ (x1 βˆ’ xn+1)
1595
+ 3
1596
+ (z βˆ’ x1)
1597
+ (z βˆ’ x2)
1598
+ 1
1599
+ ...
1600
+ ...
1601
+ ...
1602
+ ...
1603
+ ...
1604
+ ...
1605
+ ...
1606
+ ...
1607
+ ...
1608
+ n βˆ’ 1
1609
+ (z βˆ’ x1)
1610
+ Β· Β· Β·
1611
+ (z βˆ’ xnβˆ’2)
1612
+ 1
1613
+ (x1 βˆ’ xn)
1614
+ (x1 βˆ’ xn+1)
1615
+ n
1616
+ (z βˆ’ x1)
1617
+ (z βˆ’ x2)
1618
+ Β· Β· Β·
1619
+ (z βˆ’ xnβˆ’1)
1620
+ 1
1621
+ (x1 βˆ’ xn+1)
1622
+ n + 1
1623
+ (z βˆ’ x1)
1624
+ (z βˆ’ x2)
1625
+ Β· Β· Β·
1626
+ (z βˆ’ xnβˆ’1)
1627
+ (z βˆ’ xn)
1628
+ 1
1629
+ We have by definition Sn+1 = Sn(x1 βˆ’ xn+1) + (z βˆ’ x1)(z βˆ’ x2) Β· Β· Β· (z βˆ’ xn), which
1630
+ by induction gives
1631
+ Sn+1 = (z βˆ’ x2) Β· Β· Β· (z βˆ’ xn)(x1 βˆ’ xn+1) + (z βˆ’ x1)(z βˆ’ x2) Β· Β· Β· (z βˆ’ xn)
1632
+ = (z βˆ’ x2) Β· Β· Β· (z βˆ’ xn)(x1 βˆ’ xn+1 + z βˆ’ x1)
1633
+ and gives the desired result.
1634
+ β–‘
1635
+ Lemma 27. Consider A < l < L < B. The quantity
1636
+ οΏ½
1637
+ l≀y≀L
1638
+ [Ξ»a βˆ’ Ξ»x]yβˆ’1
1639
+ A
1640
+ Β· [Ξ»b βˆ’ Ξ»x]B
1641
+ y+1
1642
+ equals to
1643
+ [Ξ»a βˆ’ Ξ»x]lβˆ’1
1644
+ A
1645
+ Β· [Ξ»b βˆ’ Ξ»x]B
1646
+ L+1 Β· [Ξ»a βˆ’ Ξ»x]L
1647
+ l βˆ’ [Ξ»b βˆ’ Ξ»x]L
1648
+ l
1649
+ (Ξ»a βˆ’ Ξ»b)
1650
+ Proof. We write
1651
+ οΏ½
1652
+ l≀y≀L
1653
+ [Ξ»a βˆ’ Ξ»x]yβˆ’1
1654
+ A
1655
+ Β· [Ξ»b βˆ’ Ξ»x]B
1656
+ y+1
1657
+ = [Ξ»a βˆ’ Ξ»x]lβˆ’1
1658
+ A
1659
+ Β· [Ξ»b βˆ’ Ξ»x]B
1660
+ L+1 Β·
1661
+ οΏ½
1662
+ l≀y≀L
1663
+ [Ξ»a βˆ’ Ξ»x]yβˆ’1
1664
+ l
1665
+ Β· [Ξ»b βˆ’ Ξ»x]L
1666
+ y+1
1667
+ The last sum can be read as the vertical sum S of the products of every line in the
1668
+ following array:
1669
+ y
1670
+ l
1671
+ 1
1672
+ (Ξ»b βˆ’ Ξ»l+1)(Ξ»b βˆ’ Ξ»l+2)
1673
+ Β· Β· Β·
1674
+ (Ξ»b βˆ’ Ξ»Lβˆ’1)(Ξ»b βˆ’ Ξ»L)
1675
+ l + 1 (Ξ»a βˆ’ Ξ»l)
1676
+ 1
1677
+ (Ξ»b βˆ’ Ξ»l+2)
1678
+ Β· Β· Β·
1679
+ (Ξ»b βˆ’ Ξ»Lβˆ’1)(Ξ»b βˆ’ Ξ»L)
1680
+ l + 2 (Ξ»a βˆ’ Ξ»l)(Ξ»a βˆ’ Ξ»l+1)
1681
+ 1
1682
+ ...
1683
+ ...
1684
+ ...
1685
+ ...
1686
+ ...
1687
+ ...
1688
+ ...
1689
+ ...
1690
+ ...
1691
+ L βˆ’ 2(Ξ»a βˆ’ Ξ»l)(Ξ»a βˆ’ Ξ»l+1)
1692
+ Β· Β· Β·
1693
+ 1
1694
+ (Ξ»b βˆ’ Ξ»Lβˆ’1)(Ξ»b βˆ’ Ξ»L)
1695
+ L βˆ’ 1(Ξ»a βˆ’ Ξ»l)(Ξ»a βˆ’ Ξ»l+1)
1696
+ Β· Β· Β·
1697
+ (Ξ»a βˆ’ Ξ»Lβˆ’2)
1698
+ 1
1699
+ (Ξ»b βˆ’ Ξ»L)
1700
+ L
1701
+ (Ξ»a βˆ’ Ξ»l)(Ξ»a βˆ’ Ξ»l+1)
1702
+ Β· Β· Β·
1703
+ (Ξ»a βˆ’ Ξ»Lβˆ’2)(Ξ»a βˆ’ Ξ»Lβˆ’1)
1704
+ 1
1705
+ If l = b, then lemma 26 implies that S = [Ξ»a βˆ’ Ξ»x]L
1706
+ b+1. Furthermore, if L = a then
1707
+ S = 0.
1708
+
1709
+ ON THE LIFTING PROBLEM OF REPRESENTATIONS OF A METACYCLIC GROUP.
1710
+ 19
1711
+ The quantity S cannot be directly computed using lemma 26, if l ΜΈ= b.
1712
+ We
1713
+ proceed by forming the array:
1714
+ y
1715
+ b
1716
+ 1
1717
+ (Ξ»b βˆ’ Ξ»b+1)
1718
+ Β· Β· Β·
1719
+ (Ξ»b βˆ’ Ξ»l)
1720
+ Β· Β· Β·
1721
+ Β· Β· Β·
1722
+ Β· Β· Β·
1723
+ Β· Β· Β·
1724
+ (Ξ»b βˆ’ Ξ»L)
1725
+ ...
1726
+ ...
1727
+ ...
1728
+ l βˆ’ 1 (Ξ»a βˆ’ Ξ»b)
1729
+ Β· Β· Β·
1730
+ 1
1731
+ (Ξ»b βˆ’ Ξ»l)
1732
+ Β· Β· Β·
1733
+ Β· Β· Β·
1734
+ Β· Β· Β·
1735
+ Β· Β· Β·
1736
+ (Ξ»b βˆ’ Ξ»L)
1737
+ l
1738
+ (Ξ»a βˆ’ Ξ»b)
1739
+ Β· Β· Β·
1740
+ (Ξ»a βˆ’ Ξ»lβˆ’1)
1741
+ 1
1742
+ (Ξ»b βˆ’ Ξ»l+1)(Ξ»b βˆ’ Ξ»l+2)
1743
+ Β· Β· Β·
1744
+ (Ξ»b βˆ’ Ξ»Lβˆ’1)(Ξ»b βˆ’ Ξ»L)
1745
+ l + 1 (Ξ»a βˆ’ Ξ»b)
1746
+ Β· Β· Β·
1747
+ (Ξ»a βˆ’ Ξ»lβˆ’1)(Ξ»a βˆ’ Ξ»l)
1748
+ 1
1749
+ (Ξ»b βˆ’ Ξ»l+2)
1750
+ Β· Β· Β·
1751
+ (Ξ»b βˆ’ Ξ»Lβˆ’1)(Ξ»b βˆ’ Ξ»L)
1752
+ l + 2 (Ξ»a βˆ’ Ξ»b)
1753
+ Β· Β· Β·
1754
+ (Ξ»a βˆ’ Ξ»lβˆ’1)(Ξ»a βˆ’ Ξ»l)(Ξ»a βˆ’ Ξ»l+1)
1755
+ 1
1756
+ ...
1757
+ ...
1758
+ ...
1759
+ ...
1760
+ ...
1761
+ ...
1762
+ ...
1763
+ ...
1764
+ ...
1765
+ L βˆ’ 2(Ξ»a βˆ’ Ξ»b)
1766
+ Β· Β· Β·
1767
+ (Ξ»a βˆ’ Ξ»lβˆ’1)(Ξ»a βˆ’ Ξ»l)(Ξ»a βˆ’ Ξ»l+1)
1768
+ Β· Β· Β·
1769
+ 1
1770
+ (Ξ»b βˆ’ Ξ»Lβˆ’1)(Ξ»b βˆ’ Ξ»L)
1771
+ L βˆ’ 1(Ξ»a βˆ’ Ξ»b)
1772
+ Β· Β· Β·
1773
+ (Ξ»a βˆ’ Ξ»lβˆ’1)(Ξ»a βˆ’ Ξ»l)(Ξ»a βˆ’ Ξ»l+1)
1774
+ Β· Β· Β·
1775
+ (Ξ»a βˆ’ Ξ»Lβˆ’2)
1776
+ 1
1777
+ (Ξ»b βˆ’ Ξ»L)
1778
+ L
1779
+ (Ξ»a βˆ’ Ξ»b)
1780
+ Β· Β· Β·
1781
+ (Ξ»a βˆ’ Ξ»lβˆ’1)(Ξ»a βˆ’ Ξ»l)(Ξ»a βˆ’ Ξ»l+1)
1782
+ Β· Β· Β·
1783
+ (Ξ»a βˆ’ Ξ»Lβˆ’2)(Ξ»a βˆ’ Ξ»Lβˆ’1)
1784
+ 1
1785
+ The value of this array is computed using lemma 26 to be equal to [Ξ»a βˆ’Ξ»x]L
1786
+ b+1. We
1787
+ observe that the sum of the products of the top left array can be computed using
1788
+ lemma 26, while the sum of the products of the lower right array is S.
1789
+ [Ξ»a βˆ’ Ξ»x]lβˆ’1
1790
+ b
1791
+ Β· S + [Ξ»a βˆ’ Ξ»x]lβˆ’1
1792
+ b+1 Β· [Ξ»b βˆ’ Ξ»x]L
1793
+ l = [Ξ»a βˆ’ Ξ»x]L
1794
+ b+1
1795
+ we arrive at
1796
+ [Ξ»a βˆ’ Ξ»x]lβˆ’1
1797
+ b
1798
+ S = [Ξ»a βˆ’ Ξ»x]lβˆ’1
1799
+ b+1
1800
+ οΏ½
1801
+ [Ξ»a βˆ’ Ξ»x]L
1802
+ l βˆ’ [Ξ»b βˆ’ Ξ»x]L
1803
+ l
1804
+ οΏ½
1805
+ or equivalently
1806
+ (Ξ»a βˆ’ Ξ»b) Β· S = [Ξ»a βˆ’ Ξ»x]L
1807
+ l βˆ’ [Ξ»b βˆ’ Ξ»x]L
1808
+ l
1809
+ β–‘
1810
+ Lemma 28. For all 1 ≀ Β΅ ≀ d βˆ’ 2 we have Ξ³Β΅,d = 0.
1811
+ Proof. Let ¡1 = ¡ > ¡2 > · · · > ¡s = 1 ∈ Σ¡ be a selection of iterations and
1812
+ d = i1 > i2 > Β· Β· Β· Β· Β· Β· is β‰₯ 1 > is+1 = 0 be the sequence of i’s. Using eq. (20) we
1813
+ see that the quantity [λα
1814
+ ¡ν βˆ’ Ξ»x]iΞ½βˆ’1
1815
+ iΞ½+1+1 ΜΈ= 0 if and only if one of the following two
1816
+ inequalities hold:
1817
+ either
1818
+ iΞ½+1 >¡ν βˆ’ mf(¡ν)
1819
+ (22)
1820
+ or
1821
+ iΞ½ <¡ν + 2 βˆ’ mf(¡ν),
1822
+ (23)
1823
+ where
1824
+ f(x) =
1825
+ οΏ½
1826
+ 1
1827
+ if m | x
1828
+ 0
1829
+ if m ∀ x
1830
+ We will denote the above two inequalities by (22)Ξ½,(23)Ξ½ when applied for the
1831
+ integer Ξ½. Assume, that for all 1 ≀ Ξ½ ≀ s one of the two inequalities (22)Ξ½,(23)Ξ½
1832
+ hold, that is [λα
1833
+ ¡ν βˆ’ Ξ»x]iΞ½βˆ’1
1834
+ iΞ½+1+1 ΜΈ= 0. Inequality (22)s can not hold for Ξ½ = s since it
1835
+ gives us 0 = is+1 > 1 = ¡s, we have m ∀ 1 = ¡s.
1836
+ We will keep the sequence ¯¡ : ¡1 > ¡2 > · · · > ¡s fixed and we will sum over all
1837
+ possible selections of sequences of i1 > Β· Β· Β· is > is+1 = 0, that is we will show that
1838
+ the sum
1839
+ (24)
1840
+ Γ¯¡,i :=
1841
+ οΏ½
1842
+ i=i1>i2>Β·Β·Β·>isβ‰₯1
1843
+ s
1844
+ οΏ½
1845
+ Ξ½=1
1846
+ [λα
1847
+ ¡ν βˆ’ Ξ»x]iΞ½βˆ’1
1848
+ iΞ½+1+1
1849
+ is zero, which will show that Ξ³Β΅,d = 0 using eq. (18).
1850
+
1851
+ 20
1852
+ A. KONTOGEORGIS AND A. TEREZAKIS
1853
+ Observe now that if (23)Ξ½ holds and m ∀ Ξ½, Ξ½ βˆ’1, then (23)Ξ½βˆ’1 also holds. Indeed
1854
+ the combination of (23)Ξ½ and (22)Ξ½βˆ’1 gives the impossible inequality
1855
+ ¡ν + 2
1856
+ (23)Ξ½
1857
+ >
1858
+ iΞ½
1859
+ (22)Ξ½βˆ’1
1860
+ >
1861
+ Β΅Ξ½βˆ’1.
1862
+ Assume now that m | Ξ½ and (23)Ξ½ holds, then (23)Ξ½βˆ’1 also holds.
1863
+ Indeed the
1864
+ combination of (23)Ξ½ and (22)Ξ½βˆ’1 gives us
1865
+ ¡ν + 2 βˆ’ m
1866
+ (23)Ξ½
1867
+ >
1868
+ iΞ½
1869
+ (22)Ξ½βˆ’1
1870
+ >
1871
+ Β΅Ξ½βˆ’1 βˆ’ mf(Β΅Ξ½βˆ’1).
1872
+ If m ∀ Β΅Ξ½βˆ’1, then the above inequality is impossible since it implies that
1873
+ ¡ν + 2 βˆ’ m > Β΅Ξ½βˆ’1 > ¡ν.
1874
+ If m | Β΅Ξ½βˆ’1, then the inequality is also impossible since it implies that ¡ν + 2 >
1875
+ Β΅Ξ½βˆ’1 so if we write Β΅Ξ½βˆ’1 = kβ€²m and ¡ν = km, k, kβ€² ∈ N, kβ€² > k, we arrive at
1876
+ 2 > (kβ€² βˆ’ k)m β‰₯ m. This proves the following
1877
+ Lemma 29. The inequality (22)Ξ½βˆ’1 might be correct only in cases where m | Β΅Ξ½βˆ’1,
1878
+ m ∀ ¡ν.
1879
+ Assume that for all Ξ½ inequality (23) holds. Then for Ξ½ = 1 it gives us (recall
1880
+ that Β΅ ≀ d βˆ’ 2)
1881
+ (25)
1882
+ Β΅ + 2 ≀ d = i1 < Β΅1 + 2 βˆ’ mf(Β΅1) = Β΅ + 2 βˆ’ mf(Β΅),
1883
+ which is impossible. Therefore either there are Ξ½ such that none of the two inequal-
1884
+ ities (22)Ξ½, (23)Ξ½ hold (in this case the contribution to the sum is zero) or there are
1885
+ cases where (22) holds.
1886
+ The sumands appearing in eq. (24) can be zero, for example the sequence Β΅1 =
1887
+ m > Β΅2 = 1 with i2 = 2 < i1 = d, s = 2 give the contribution
1888
+ [λα
1889
+ Β΅2 βˆ’ Ξ»x]i2βˆ’1
1890
+ 1
1891
+ [λα
1892
+ Β΅1 βˆ’ Ξ»x]dβˆ’1
1893
+ i2
1894
+ = [λα
1895
+ 1 βˆ’ Ξ»x]1
1896
+ 1[λα
1897
+ m βˆ’ Ξ»x]dβˆ’1
1898
+ i2+1 = (Ξ»2 βˆ’ Ξ»1)[Ξ»1 βˆ’ Ξ»x]dβˆ’1
1899
+ 3
1900
+ while for i2 = 1 < i1 = d it gives the contribution
1901
+ [λα
1902
+ Β΅2 βˆ’ Ξ»x]i2βˆ’1
1903
+ 1
1904
+ [λα
1905
+ Β΅1 βˆ’ Ξ»x]dβˆ’1
1906
+ i2+1 = [λα
1907
+ 1 βˆ’ Ξ»x]0
1908
+ 1[λα
1909
+ m βˆ’ Ξ»x]dβˆ’1
1910
+ 2
1911
+ = [Ξ»1 βˆ’ Ξ»x]dβˆ’1
1912
+ 2
1913
+ It is clear that these non-zero contributions cancel out when added.
1914
+ Lemma 30. Assume that m | ¡ν0βˆ’1 and m ∀ ¡ν0, where (23)Ξ½0 and (22)Ξ½0βˆ’1 hold.
1915
+ Then, we can eliminate ¡ν0βˆ’1 and iΞ½0 from both selections of the sequence of ¡’s
1916
+ and i’s, i.e. we can form the sequence of length s βˆ’ 1
1917
+ Β―Β΅sβˆ’1 = Β΅s < Β―Β΅sβˆ’2 = Β΅sβˆ’1 < Β· Β· Β· < ¯¡ν0βˆ’1 = ¡ν0 < ¯¡ν0οΏ½οΏ½οΏ½2 = ¡ν0βˆ’2 < Β· Β· Β· < Β―Β΅1 = Β΅1.
1918
+ and the corresponding sequence of equal length
1919
+ Β―isβˆ’1 = is < Β―isβˆ’2 = isβˆ’1 < Β· Β· Β· < Β―iΞ½0βˆ’1 = iΞ½0βˆ’1 < Β―iΞ½0 = iΞ½0+1 < Β· Β· Β· < Β―i1 = i1 = d,
1920
+ so that
1921
+ Γ¯¡,i =
1922
+ οΏ½
1923
+ i1>Β·Β·Β·>is
1924
+ s
1925
+ οΏ½
1926
+ Ξ½=1
1927
+ [λα
1928
+ ¡ν βˆ’ Ξ»x]iΞ½βˆ’1
1929
+ iΞ½+1+1 = (⋆)
1930
+ οΏ½
1931
+ Β―i1>Β·Β·Β·>Β―isβˆ’1
1932
+ s
1933
+ οΏ½
1934
+ Ξ½=1
1935
+ Ξ½ΜΈ=Ξ½0βˆ’1
1936
+ [λα
1937
+ ¡ν βˆ’ Ξ»x]iΞ½βˆ’1
1938
+ iΞ½+1+1,
1939
+ where (⋆) is a non zero element.
1940
+
1941
+ ON THE LIFTING PROBLEM OF REPRESENTATIONS OF A METACYCLIC GROUP.
1942
+ 21
1943
+ Proof. (of lemma 30) We are in the case m | ¡ν0βˆ’1 and m ∀ ¡ν0, where (23)Ξ½0 and
1944
+ (22)Ξ½0βˆ’1 hold,
1945
+ (26)
1946
+ ¡ν0βˆ’1 βˆ’ m
1947
+ (22)Ξ½0βˆ’1
1948
+ <
1949
+ iΞ½0
1950
+ (23)Ξ½0
1951
+ <
1952
+ ¡ν0 + 2,
1953
+ or equivalently
1954
+ Β΅0 := ¡ν0βˆ’1 βˆ’ m + 1 ≀ iΞ½0 ≀ ¡ν0 + 1
1955
+ For iΞ½0+1 the inequality (22)Ξ½0 iΞ½0+1 > ¡ν0 βˆ’ mf(¡ν0) can not hold, since it implies
1956
+ iΞ½0+1 < iΞ½0
1957
+ (23)Ξ½0
1958
+ <
1959
+ ¡ν0 + 2 < iν0+1 + 2.
1960
+ Observe that also
1961
+ iΞ½0+1 + 1 ≀ iΞ½0 ≀ iΞ½0βˆ’1 βˆ’ 1.
1962
+ Set l = max{Β΅0, iΞ½0+1 + 1} and L = min{¡ν0 + 1, iΞ½0βˆ’1 βˆ’ 1}. Then y = iΞ½0 satisfies
1963
+ l ≀ y ≀ L.
1964
+ By lemma 27 the quantity
1965
+ οΏ½
1966
+ l≀y≀L
1967
+ [λ¡ν0+1 βˆ’ Ξ»x]yβˆ’1
1968
+ iΞ½0+1+1 Β· [λ¡0 βˆ’ Ξ»x]
1969
+ iΞ½0βˆ’1βˆ’1
1970
+ y+1
1971
+ equals to
1972
+ [λ¡ν0+1 βˆ’ Ξ»x]lβˆ’1
1973
+ iΞ½0+1+1 Β· [λ¡0 βˆ’ Ξ»x]
1974
+ iΞ½0βˆ’1βˆ’1
1975
+ L+1
1976
+ Β· [λ¡ν0+1 βˆ’ Ξ»x]L
1977
+ l βˆ’ [λ¡0 βˆ’ Ξ»x]L
1978
+ l
1979
+ (λ¡ν0+1 βˆ’ λ¡0)
1980
+ (27)
1981
+ [λ¡ν0+1 βˆ’ Ξ»x]L
1982
+ iΞ½0+1+1 Β· [λ¡0 βˆ’ Ξ»x]
1983
+ iΞ½0βˆ’1βˆ’1
1984
+ L+1
1985
+ βˆ’ [λ¡ν0+1 βˆ’ Ξ»x]lβˆ’1
1986
+ iΞ½0+1+1 Β· [λ¡0 βˆ’ Ξ»x]
1987
+ iΞ½0βˆ’1βˆ’1
1988
+ l
1989
+ (λ¡ν0+1 βˆ’ λ¡0)
1990
+ Case A1 l = Β΅0 β‰₯ iΞ½0+1 + 1. Then [λ¡0 βˆ’ Ξ»x]L
1991
+ l = 0.
1992
+ Case A2 l = iΞ½0+1 + 1 > Β΅0. We set z := iΞ½0+1, which is bounded by eq. (23)Ξ½0+1
1993
+ that is
1994
+ Β΅0
1995
+ Case A2
1996
+ ≀
1997
+ z
1998
+ (23)Ξ½0+1
1999
+ ≀
2000
+ ¡ν0+1 + 1.
2001
+ Notice that in this case m ∀ ¡ν0+1. Indeed, we have assumed that inequality (23)ν0+1
2002
+ holds wich gives us
2003
+ ¡ν0βˆ’1 βˆ’ m = Β΅0 βˆ’ 1
2004
+ (Case A2)
2005
+ <
2006
+ iΞ½0+1
2007
+ (23)Ξ½0+1
2008
+ <
2009
+ ¡ν0+1 + 2 βˆ’ m,
2010
+ which implies that ¡ν0βˆ’1 < ¡ν0+1 + 2, a contradiction. Thus for l = z + 1 we
2011
+ compute
2012
+ οΏ½
2013
+ Β΅0≀z≀¡ν0+1+1
2014
+ [λα
2015
+ ¡ν0+1 βˆ’ Ξ»x]
2016
+ iΞ½0+1βˆ’1
2017
+ iΞ½0+2+1 Β· [λ¡0 βˆ’ Ξ»x]L
2018
+ l =
2019
+ =
2020
+ οΏ½
2021
+ Β΅0≀z≀¡ν0+1+1
2022
+ [λ¡ν0+1+1 βˆ’ Ξ»x]zβˆ’1
2023
+ iΞ½0+2+1 Β· [λ¡0 βˆ’ Ξ»x]L
2024
+ z+1 =
2025
+ = (⋆) Β· [λ¡ν0+1+1 βˆ’ Ξ»x]
2026
+ ¡ν0+1+1
2027
+ Β΅0
2028
+ βˆ’ [λ¡0 βˆ’ Ξ»x]
2029
+ ¡ν0+1+1
2030
+ Β΅0
2031
+ λ¡ν0+1+1 βˆ’ λ¡0+1
2032
+ = 0.
2033
+ Case B1 L = ¡ν0 + 1 ≀ iΞ½0βˆ’1 βˆ’ 1. In this case [λ¡ν0+1 βˆ’ Ξ»x]L
2034
+ l = 0.
2035
+
2036
+ 22
2037
+ A. KONTOGEORGIS AND A. TEREZAKIS
2038
+ Case B2 L = iΞ½0βˆ’1 βˆ’ 1 < ¡ν0 + 1. In this case eq. (27) is reduced to
2039
+ [λ¡ν0+1 βˆ’ Ξ»x]
2040
+ iΞ½0βˆ’1βˆ’1
2041
+ iΞ½0+1+1
2042
+ (λ¡ν0+1 βˆ’ λ¡0)
2043
+ This means that we have erased the ¡ν0βˆ’1 from the product and we have
2044
+ οΏ½
2045
+ i1>Β·Β·Β·>is
2046
+ s
2047
+ οΏ½
2048
+ Ξ½=1
2049
+ [λα
2050
+ ¡ν βˆ’ Ξ»x]iΞ½βˆ’1
2051
+ iΞ½+1+1 = (⋆)
2052
+ οΏ½
2053
+ i1>Β·Β·Β·>is
2054
+ s
2055
+ οΏ½
2056
+ Ξ½=1
2057
+ Ξ½ΜΈ=Ξ½0βˆ’1
2058
+ [λα
2059
+ ¡ν βˆ’ Ξ»x]iΞ½βˆ’1
2060
+ iΞ½+1+1,
2061
+ where (⋆) is a non zero element. This procedure gives us that the original quantity
2062
+ [λα
2063
+ ¡ν0 βˆ’ Ξ»x]
2064
+ iΞ½0βˆ’1
2065
+ iν0+1+1 · [λα
2066
+ ¡ν0βˆ’1 βˆ’ Ξ»x]
2067
+ iΞ½0βˆ’1βˆ’1
2068
+ iΞ½0+1
2069
+ after summing over iΞ½0 becomes the quantity
2070
+ [λα
2071
+ ¡ν0 βˆ’ Ξ»x]
2072
+ iΞ½0βˆ’1βˆ’1
2073
+ iν0+1+1 = [λα
2074
+ ¯¡ν0βˆ’1 βˆ’ Ξ»x]
2075
+ Β―iΞ½0βˆ’1βˆ’1
2076
+ Β―iΞ½0+1
2077
+ ,
2078
+ that is we have eliminated the ¡ν0βˆ’1 and iΞ½0 from both selections of the sequence
2079
+ of ¡’s and i’s, i.e. we have the sequence of length s βˆ’ 1
2080
+ Β―Β΅sβˆ’1 = Β΅s < Β―Β΅sβˆ’2 = Β΅sβˆ’1 < Β· Β· Β· < ¯¡ν0βˆ’1 = ¡ν0 < ¯¡ν0βˆ’2 = ¡ν0βˆ’2 < Β· Β· Β· < Β―Β΅1 = Β΅1.
2081
+ and the corresponding sequence of equal length
2082
+ Β―isβˆ’1 = is < Β―isβˆ’2 = isβˆ’1 < Β· Β· Β· < Β―iΞ½0βˆ’1 = iΞ½0βˆ’1 < Β―iΞ½0 = iΞ½0+1 < Β· Β· Β· < Β―i1 = i1 = d.
2083
+ β–‘
2084
+ Remark 31. One should be careful here since Β―iΞ½0βˆ’1 = iΞ½0βˆ’1 > iΞ½0 > Β―iΞ½0 = iΞ½0+1,
2085
+ so Β―iΞ½0βˆ’1 > Β―iΞ½0 + 1. This means that the new sequence of Β―isβˆ’1 > Β· Β· Β· > Β―i1 satisfies a
2086
+ stronger inequality in the Ξ½0 position, unless Ξ½0 βˆ’ 1 = d in the computation of Ξ³d,d.
2087
+ Consider the set s, s βˆ’ 1, . . . , Ξ½0 such that m ∀ ¡ν for s β‰₯ Ξ½ β‰₯ Ξ½0 and assume
2088
+ that m | ¡ν0βˆ’1 and (23)Ξ½0 and (22)Ξ½0βˆ’1 hold. We apply lemma 30 and we obtain
2089
+ a new sequence of ¡’s with ¡ν0βˆ’1 removed, provided that Ξ½0 βˆ’ 1 > 1. We continue
2090
+ this way and in the sequence of ¡’s we eliminate all possible inequalities like (26)
2091
+ obtaining a series of Β΅ which involves only inequalities of type (23). But this is not
2092
+ possible if Β΅ ≀ d βˆ’ 2, according to equation (25). This proves that all Ξ³Β΅,d = 0 for
2093
+ 1 ≀ Β΅ ≀ d βˆ’ 2, this completes the proof of lemma 28.
2094
+ β–‘
2095
+ Lemma 32. If Β΅2 ΜΈ= d βˆ’ 1, then the contribution of the corresponding summand
2096
+ Γ¯¡,i to Ξ³d,d is zero.
2097
+ Proof. We are in the case Β΅ = d = i. We begin the procedure of eliminating all
2098
+ sequences of inequalities of the form (23)Ξ½0, (22)Ξ½0βˆ’1, where m | Ξ½0βˆ’1, m ∀ Ξ½0, using
2099
+ lemma 30. For Ξ½ = 1 inequality (23)1 can not hold since it implies the impossible
2100
+ inequality d = i1 < d + 2 βˆ’ m. Therefore, (22)1 holds, that is i2 > d βˆ’ m. On the
2101
+ other hand we can assume that (23)2 holds by the elimination process, so we have
2102
+ d βˆ’ m
2103
+ (22)1
2104
+ < i2
2105
+ (23)2
2106
+ < Β΅2 + 2.
2107
+ Following the analysis of the proof of lemma 28 we see that the contribution to Ξ³d,d
2108
+ is non zero if case B2 holds, that is (Ξ½0 = 2 in this case) d βˆ’ 1 = iΞ½0βˆ’1 βˆ’ 1 < Β΅2 + 1,
2109
+ obtaining that Β΅2 = d βˆ’ 1.
2110
+ β–‘
2111
+
2112
+ ON THE LIFTING PROBLEM OF REPRESENTATIONS OF A METACYCLIC GROUP.
2113
+ 23
2114
+ Lemma 33. Equation (21) holds, that is
2115
+ (Ξ»d βˆ’ λα
2116
+ d )Ξ³d,d =
2117
+ dβˆ’1
2118
+ οΏ½
2119
+ Ξ½=1
2120
+ t(Ξ±)
2121
+ d,Ξ½Ξ³Ξ½,d = t(Ξ±)
2122
+ d,dβˆ’1Ξ³dβˆ’1,d.
2123
+ Proof. We will use the procedure of the proof of lemma 30. We recall that for
2124
+ each fixed sequence of ¡s > · · · > ¡1 we summed over all possible sequences i1 >
2125
+ Β· Β· Β· > is+1 = 0.
2126
+ In the final step the inequality (26) appears, for ν0 = 2, and
2127
+ ¡ν0 = Β΅2 = d βˆ’ 1 and Ξ½0 βˆ’ 1 = 1 and ¡ν0βˆ’1 = Β΅ = d, that is:
2128
+ 0 = ¡ν0βˆ’1 βˆ’ m
2129
+ (22)2
2130
+ < iΞ½0
2131
+ (23)1
2132
+ < ¡ν0 + 2 = d + 1.
2133
+ As in the proof of lemma 30 we sum over y = iΞ½ and the result is either zero in case
2134
+ B1 or in the B2 case, where ¡ν0 = Β΅2 = d βˆ’ 1 and Β΅0 = ¡ν0βˆ’1 βˆ’ m + 1 = d βˆ’ m + 1,
2135
+ the contribution is computed to be equal to
2136
+ [λα
2137
+ ¡ν0+1 βˆ’ Ξ»x]
2138
+ iΞ½0βˆ’1βˆ’1
2139
+ iΞ½0+1+1
2140
+ (λ¡ν0+1 βˆ’ Ξ»m0)
2141
+ = [λα
2142
+ d βˆ’ Ξ»x]dβˆ’1
2143
+ i3+1
2144
+ Ξ»d βˆ’ λα
2145
+ d
2146
+ .
2147
+ The last ¡ν0βˆ’1 = Β΅1 = d is eliminated in the above expression. This means that
2148
+ for a fixed sequence ¡1 > . . . > ¡s the contribution of the inner sum in eq. (24) is
2149
+ given by
2150
+ 1
2151
+ Ξ»d βˆ’ λα
2152
+ d
2153
+ Β·
2154
+ οΏ½
2155
+ dβˆ’1=i2>i3>Β·Β·Β·>isβ‰₯1
2156
+ s
2157
+ οΏ½
2158
+ Ξ½=2
2159
+ [λα
2160
+ ¡ν βˆ’ Ξ»x]iΞ½βˆ’1
2161
+ iΞ½+1+1.
2162
+ Observe that Β΅1 = d does not appear in this expression and this expression corre-
2163
+ sponds to the sequence Β―Β΅1 = Β΅2 = d βˆ’ 1 > Β―Β΅2 = Β΅3 > Β· Β· Β· > Β―Β΅sβˆ’1 = Β―Β΅s = 1. Notice,
2164
+ also that the problem described in remark 31 does not appear here, sence we erased
2165
+ i1 which is not between some i’s but the first one. Therefore, we can relate it to
2166
+ a similar expression that contributes to Ξ³dβˆ’1,d. Conversely every contribution of
2167
+ Ξ³dβˆ’1,d gives rise to a contribution in Ξ³d,d, by multiplying by Ξ»d βˆ’ λα
2168
+ d . The desired
2169
+ result follows by the expression of Ξ³Β΅,d given in eq. (18).
2170
+ β–‘
2171
+ We have shown so far how to construct matrices Ξ“, T so that
2172
+ (28)
2173
+ T q = 1, Ξ“TΞ“βˆ’1 = T Ξ±.
2174
+ We will now prove that Ξ“ has order m. By equation (28) Ξ“k should satisfy equation
2175
+ Ξ“kTΞ“βˆ’k = T Ξ±k.
2176
+ Using proposition 25 asserting the uniqueness of such Ξ“k with Ξ± replaced by Ξ±k we
2177
+ have that the matrix multiplication of the entries of Ξ“ giving rise to (Ξ³(k)
2178
+ Β΅,i ) = Ξ“k
2179
+ coincide to the values by the the recursive method of proposition (28) applied
2180
+ for Ξ“β€² = Ξ“k, Ξ±β€² = Ξ±k and Ξ“β€²E1 = ΞΆΟ΅k
2181
+ m E1.
2182
+ In particular for k = m, we have
2183
+ Ξ±m ≑ 1 modpΞ½ for all 1 ≀ Ξ½ ≀ h, that is the matrix Ξ“k should be recursively
2184
+ constructed using proposition (28) for the relation Ξ“mTΞ“m = T, Ξ“mE1 = E1,
2185
+ leading to the conclusion Ξ“m = Id.
2186
+ Notice that the first eigenvalue of Ξ“ is a
2187
+ primitive root of unity, therefore Ξ“ has order exactly m.
2188
+ By lemma 10 the action of Οƒ in the special fibre is given by a lower triangular
2189
+ matrix. Therefore, we must have
2190
+ (29)
2191
+ γν,i ∈ mr for ν < i.
2192
+
2193
+ 24
2194
+ A. KONTOGEORGIS AND A. TEREZAKIS
2195
+ Proposition 34. If
2196
+ (30)
2197
+ v(Ξ»i βˆ’ Ξ»j) > v(aΞ½) for all 1 ≀ i, j ≀ d and 1 ≀ Ξ½ ≀ d βˆ’ 1
2198
+ then the matrix (Ξ³Β΅,i) has entries in the ring R and is lower triangular modulo mR.
2199
+ Proof. Assume that the condition of eq. (30) holds. In equation (18) we compute
2200
+ the fraction
2201
+ (31)
2202
+ [a]Β΅βˆ’1
2203
+ 1
2204
+ [a]iβˆ’1
2205
+ 1
2206
+ =
2207
+ οΏ½
2208
+ οΏ½
2209
+ οΏ½
2210
+ οΏ½
2211
+ οΏ½
2212
+ 1
2213
+ [a]iβˆ’1
2214
+ Β΅
2215
+ if i > Β΅
2216
+ 1
2217
+ if i = Β΅
2218
+ [a]Β΅βˆ’1
2219
+ i
2220
+ if i < Β΅
2221
+ The number of (λα
2222
+ Β΅ βˆ’ Ξ»x) factors in the numerator is equal to (recall that is+1 = 0)
2223
+ s
2224
+ οΏ½
2225
+ Ξ½=1
2226
+ (iΞ½ βˆ’ 1 βˆ’ iΞ½+1 βˆ’ 1 + 1) = i βˆ’ s,
2227
+ and i > Β΅ β‰₯ s, so i βˆ’ s > 0. Therefore, for the upper part of the matrix i > Β΅ we
2228
+ have i βˆ’ s factors of the form (λα
2229
+ i βˆ’ Ξ»j) in the numerator and i βˆ’ Β΅ factors ax in
2230
+ the denominator. Their difference is equal to (i βˆ’ s) βˆ’ (i βˆ’ Β΅) = Β΅ βˆ’ s β‰₯ 0. By
2231
+ assumption the matrix reduces to an upper triangular matrix modulo mR.
2232
+ β–‘
2233
+ Remark 35. The condition given in equation (30) can be satisfied in the following
2234
+ way: It is clear that Ξ»i βˆ’ Ξ»j ∈ mR. Even in the case vmR(Ξ»i βˆ’ Ξ»j) = 1 we can
2235
+ consider a ramified extension Rβ€² of the ring R with ramification index e, in order to
2236
+ make the valuation vmRβ€² (Ξ»i βˆ’ Ξ»j) = e and then there is space to select vmRβ€² (ai) <
2237
+ vmRβ€² (Ξ»i βˆ’ Ξ»j).
2238
+ Proposition 36. We have that
2239
+ (32)
2240
+ Ξ³i,i ≑ ΞΆΟ΅
2241
+ mΞ±iβˆ’1 modmR
2242
+ Let A = {a1, . . . , adβˆ’1} ∈ R be the set of elements below the diagonal in eq. (9). If
2243
+ ai ∈ mR, then
2244
+ γ¡,i ∈ mR for ¡ ̸= i,
2245
+ that is Ei is an eigenvector for the reduced action of Ξ“ modulo mR. If aΞΊ1, . . . , aΞΊr
2246
+ the elements of the set A which are in mR, then the reduced matrix of Ξ“ has the
2247
+ form:
2248
+ οΏ½
2249
+ οΏ½
2250
+ οΏ½
2251
+ οΏ½
2252
+ οΏ½
2253
+ οΏ½
2254
+ Ξ“1
2255
+ 0
2256
+ Β· Β· Β·
2257
+ 0
2258
+ 0
2259
+ Ξ“2
2260
+ ...
2261
+ ...
2262
+ ...
2263
+ ...
2264
+ ...
2265
+ 0
2266
+ 0
2267
+ Β· Β· Β·
2268
+ 0
2269
+ Ξ“r
2270
+ οΏ½
2271
+ οΏ½
2272
+ οΏ½
2273
+ οΏ½
2274
+ οΏ½
2275
+ οΏ½
2276
+ where Ξ“1, Ξ“2, . . . , Ξ“r+1 for 1 ≀ Ξ½ ≀ r + 1 are (ΞΊΞ½ βˆ’ ΞΊΞ½βˆ’1) Γ— (ΞΊΞ½ βˆ’ ΞΊΞ½βˆ’1) lower
2277
+ triangular matrices (we set ΞΊ0 = 0, ΞΊr+1 = d).
2278
+
2279
+ ON THE LIFTING PROBLEM OF REPRESENTATIONS OF A METACYCLIC GROUP.
2280
+ 25
2281
+ Proof. Consider the matrix Ξ“:
2282
+ οΏ½
2283
+ οΏ½
2284
+ οΏ½
2285
+ οΏ½
2286
+ οΏ½
2287
+ οΏ½
2288
+ οΏ½
2289
+ οΏ½
2290
+ οΏ½
2291
+ οΏ½
2292
+ οΏ½
2293
+ οΏ½
2294
+ οΏ½
2295
+ οΏ½
2296
+ οΏ½
2297
+ οΏ½
2298
+ οΏ½
2299
+ οΏ½
2300
+ οΏ½
2301
+ οΏ½
2302
+ Ξ³11
2303
+ ...
2304
+ ...
2305
+ 0
2306
+ Ξ³ΞΊ1,1
2307
+ Β· Β· Β·
2308
+ Ξ³ΞΊ1,ΞΊ1
2309
+ Ξ³11
2310
+ Ξ³ΞΊ1+1,ΞΊ1+1
2311
+ ...
2312
+ ...
2313
+ Ξ³Β΅,i
2314
+ Ξ³ΞΊ2,ΞΊ1+1
2315
+ Β· Β· Β·
2316
+ Ξ³ΞΊ2,ΞΊ2
2317
+ Ξ³ΞΊ1+1,ΞΊ1+1
2318
+ Ξ³Β΅,i
2319
+ ...
2320
+ Ξ³ΞΊr+1,ΞΊr+1
2321
+ Β· Β· Β·
2322
+ ...
2323
+ ...
2324
+ Ξ³ΞΊ1,ΞΊ1
2325
+ Ξ³ΞΊ2,ΞΊ2
2326
+ Ξ³d,ΞΊr+1
2327
+ Β· Β· Β· Ξ³d,d
2328
+ οΏ½
2329
+ οΏ½
2330
+ οΏ½
2331
+ οΏ½
2332
+ οΏ½
2333
+ οΏ½
2334
+ οΏ½
2335
+ οΏ½
2336
+ οΏ½
2337
+ οΏ½
2338
+ οΏ½
2339
+ οΏ½
2340
+ οΏ½
2341
+ οΏ½
2342
+ οΏ½
2343
+ οΏ½
2344
+ οΏ½
2345
+ οΏ½
2346
+ οΏ½
2347
+ οΏ½
2348
+ 1 ≀ i ≀ ΞΊ1 < m ≀ d
2349
+ ΞΊ1 < i ≀ ΞΊ2 < Β΅ ≀ d
2350
+ We have that ¡ = i and the only element in Σ¡ which does not have any factor of
2351
+ the form (λα
2352
+ y βˆ’ Ξ»x) is the sequence
2353
+ 1 = Β΅s = Β΅sβˆ’1 βˆ’ 1 < Β΅sβˆ’1 < Β· Β· Β· < Β΅2 = Β΅1 βˆ’ 1 < Β΅1 = Β΅
2354
+ For this sequence eq. (18) becomes
2355
+ Ξ³i,i =
2356
+ s
2357
+ οΏ½
2358
+ Ξ½=2
2359
+ hΞ±βˆ’1(λ¡ν, Ξ»Β΅Ξ½βˆ’1)ΞΆΟ΅
2360
+ m modmR,
2361
+ which gives the desired result since hΞ±βˆ’1(λ¡ν, Ξ»Β΅Ξ½βˆ’1) ≑
2362
+ οΏ½Ξ±
2363
+ 1
2364
+ οΏ½
2365
+ = Ξ± modmR.
2366
+ For proving that all entries Ξ³Β΅,i ∈ mR for ΞΊΞ½ < i ≀ ΞΊΞ½+1 < Β΅ ≀ d, that is for
2367
+ all entries bellow the central blocks, we observe that from equation (18) combined
2368
+ with eq. (31) that Ξ³Β΅,i is divisible by [a]Β΅βˆ’1
2369
+ i
2370
+ = aiai+1 Β· Β· Β· aΞΊΞ½+1 Β· Β· Β· aΒ΅βˆ’1 ∈ mR.
2371
+ β–‘
2372
+ Recall that by lemma 2 there is an 1 ≀ a0 ≀ m such that Ξ± = ΞΆa0
2373
+ m .
2374
+ Proposition 37. The indecomposable module V modulo mR breaks into a direct
2375
+ sum of r + 1 indecomposable k[Cq β‹Š Cm] modules VΞ½, 1 ≀ Ξ½ ≀ r + 1. Each VΞ½ is
2376
+ isomorphic to VΞ±(Ο΅ + a0ΞΊΞ½βˆ’1, ΞΊΞ½ βˆ’ ΞΊΞ½βˆ’1).
2377
+ Proof. By eq. (32) the first eigenvalue of the reduced matrix block Γν is
2378
+ ΞΆΟ΅
2379
+ mΞ±ΞΊΞ½βˆ’1 = ΞΆΟ΅+(ΞΊΞ½βˆ’1)a0
2380
+ m
2381
+ .
2382
+ Since that first eigenvalue together with the size of the block determine the last
2383
+ eigenvalue, that is the action of Cm on the socle the reduced block is uniquely
2384
+ determined up to isomorphism.
2385
+ β–‘
2386
+ This way we arrive at a new obstruction.
2387
+ Assume that the indecomposable
2388
+ representation given by the matrix T as in lemma 16 reduces modulo mR to a sum
2389
+ of Jordan blocks. Then the Οƒ action on the leading elements of each Jordan block
2390
+ in the special fibre should be described by the corresponding action of Οƒ on the
2391
+ leading eigenvector E of V . The corresponding actions on the special fibre should
2392
+ be compatible.
2393
+ This observation is formally given in proposition 1, which we now prove: Each set
2394
+ IΞ½, 1 ≀ Ξ½ ≀ t corresponds to an indecomposable R[G]-module, which decomposes
2395
+ to the indecomposables Vα(ϡ¡, κ¡), ν ∈ Iν of the special fiber. Indecomposable
2396
+
2397
+ 26
2398
+ A. KONTOGEORGIS AND A. TEREZAKIS
2399
+ summands have different roots of unity in R, therefore οΏ½
2400
+ ¡∈IΞ½ kΞ½ ≀ q, this is con-
2401
+ dition (1.a). The second condition (1.b) comes from proposition 13. If 1 is one of
2402
+ the possible eigenvalues of the lift T, then οΏ½
2403
+ ¡∈IΞ½ ΞΊΒ΅ ≑ 1 modm. If all eigenvalues
2404
+ of the lift T are different than one, then οΏ½
2405
+ ¡∈IΞ½ ΞΊΒ΅ ≑ 0 modm. If #IΞ½ = q, then
2406
+ there is one zero eigenvalue and the sum equals 1 modm.
2407
+ It is clear by eq. (32) that condition (1.c) is a necessary condition. On the other
2408
+ hand if (1.c) is satisfied we can write (after a permutation if necessary) the set
2409
+ {1, . . . , S}, S = οΏ½t
2410
+ Ξ½=1 #IΞ½ as
2411
+ J1 = {1, 2, . . . , ΞΊ(1)
2412
+ 1 , ΞΊ(1)
2413
+ 1
2414
+ + 1, . . . , ΞΊ(1)
2415
+ 1
2416
+ + ΞΊ(1)
2417
+ 2 , . . . ,
2418
+ r1
2419
+ οΏ½
2420
+ j=1
2421
+ ΞΊ(1)
2422
+ j
2423
+ = b1}, I1 = {ΞΊ(1)
2424
+ 1 , . . . , ΞΊ(1)
2425
+ r1 }
2426
+ J2 = {b1 + 1, b1 + 2, . . . , b2 = b1 +
2427
+ r2
2428
+ οΏ½
2429
+ j=1
2430
+ ΞΊ(2)
2431
+ j }, I2 = {ΞΊ(2)
2432
+ 1 , . . . , ΞΊ(2)
2433
+ r2 }
2434
+ Β· Β· Β· Β· Β· Β·
2435
+ Js = {bsβˆ’1 + 1, btβˆ’1 + 2, . . . , bt = S}, Is = {ΞΊ(s)
2436
+ 1 , . . . , ΞΊ(s)
2437
+ rs }
2438
+ The matrix given in eq. (9), where
2439
+ ai =
2440
+ οΏ½
2441
+ οΏ½
2442
+ οΏ½
2443
+ οΏ½
2444
+ οΏ½
2445
+ 0
2446
+ if i ∈ {b1, . . . , bsβˆ’1}
2447
+ Ο€
2448
+ if i ∈ {κ(ν)
2449
+ 1 , ΞΊ(Ξ½)
2450
+ 1
2451
+ + ΞΊ(Ξ½)
2452
+ 2 , ΞΊ(Ξ½)
2453
+ 1
2454
+ + ΞΊ(Ξ½)
2455
+ 2
2456
+ + ΞΊ(Ξ½)
2457
+ 3 , . . . , ΞΊ(Ξ½)
2458
+ 1
2459
+ + ΞΊ(Ξ½)
2460
+ 2
2461
+ + Β· Β· Β· + ΞΊ(Ξ½)
2462
+ rΞ½βˆ’1}
2463
+ 1
2464
+ otherwise
2465
+ lifts the Ο„ generator, and by (12) there is a well defined extended action of the Οƒ
2466
+ as well.
2467
+ Example: Consider the group q = 52, m = 4, Ξ± = 7,
2468
+ G = C52 β‹Š C4 = βŸ¨Οƒ, Ο„|Οƒ4 = Ο„ 25 = 1, ΟƒΟ„Οƒβˆ’1 = Ο„ 7⟩.
2469
+ Observe that ord57 = ord527 = 4.
2470
+ β€’ The module V (Ο΅, 25) is projective and is known to lift in characteristic zero.
2471
+ This fits well with proposition 1, since 4 | 25 βˆ’ 1 = 4 Β· 6.
2472
+ β€’ The modules V (Ο΅, ΞΊ) do not lift in characteristic zero if 4 ∀ ΞΊ or 4 ∀ ΞΊ βˆ’ 1.
2473
+ Therefore only V (Ο΅, 1), V (Ο΅, 4), V (Ο΅, 5), V (Ο΅, 8), V (Ο΅, 9), V (Ο΅, 12), V (Ο΅, 13),
2474
+ V (Ο΅, 16), V (Ο΅, 17), V (Ο΅, 20), V (Ο΅, 21), V (Ο΅, 24), V (Ο΅, 25) lift.
2475
+ β€’ The module V (1, 2) βŠ• V (3, 2) lift to characteristic zero, where the matrix
2476
+ of T with respect to a basis E1, E2, E3, E4 is given by
2477
+ T =
2478
+ οΏ½
2479
+ οΏ½
2480
+ οΏ½
2481
+ οΏ½
2482
+ ΞΆq
2483
+ 0
2484
+ 0
2485
+ 0
2486
+ 1
2487
+ ΞΆ2
2488
+ q
2489
+ 0
2490
+ 0
2491
+ 0
2492
+ Ο€
2493
+ ΞΆ3
2494
+ q
2495
+ 0
2496
+ 0
2497
+ 0
2498
+ 1
2499
+ ΞΆ4
2500
+ q
2501
+ οΏ½
2502
+ οΏ½
2503
+ οΏ½
2504
+ οΏ½
2505
+ and Οƒ(E1) = ΞΆqE1.
2506
+ β€’ The module V (1, 2) βŠ• V (1, 2) does not lift in characteristic zero. There is
2507
+ no way to permute the direct summands so that the eigenvalues of Οƒ are
2508
+ given by ΞΆΟ΅
2509
+ m, Ξ±ΞΆΟ΅
2510
+ m, Ξ±2ΞΆΟ΅
2511
+ m, Ξ±3ΞΆΟ΅
2512
+ m. Notice that Ξ± = 2 = ΞΆm.
2513
+ β€’ The module V (Ο΅1, 21)βŠ•V (221Β·Ο΅1, 23) does not lift in characteristic zero. The
2514
+ sum 21+24 is divisible by 4, Ο΅2 = 221Ο΅1 is compatible, but 21+23 = 44 > 25
2515
+ so the representation of T in the supposed indecomposable module formed
2516
+
2517
+ ON THE LIFTING PROBLEM OF REPRESENTATIONS OF A METACYCLIC GROUP.
2518
+ 27
2519
+ by their sum can not have different eigenvalues which should be 25-th roots
2520
+ of unity.
2521
+ References
2522
+ [1] J. L. Alperin. Local representation theory, volume 11 of Cambridge Studies in Advanced
2523
+ Mathematics. Cambridge University Press, Cambridge, 1986. Modular representations as an
2524
+ introduction to the local representation theory of finite groups.
2525
+ [2] Frauke M. Bleher, Ted Chinburg, and Aristides Kontogeorgis. Galois structure of the holo-
2526
+ morphic differentials of curves. J. Number Theory, 216:1–68, 2020.
2527
+ [3] T. Chinburg, R. Guralnick, and D. Harbater. Oort groups and lifting problems. Compos.
2528
+ Math., 144(4):849–866, 2008.
2529
+ [4] Ted Chinburg, Robert Guralnick, and David Harbater. The local lifting problem for actions
2530
+ of finite groups on curves. Ann. Sci. Β΄Ec. Norm. SupΒ΄er. (4), 44(4):537–605, 2011.
2531
+ [5] Ted Chinburg, Robert Guralnick, and David Harbater. Global Oort groups. J. Algebra,
2532
+ 473:374–396, 2017.
2533
+ [6] Huy Dang, Soumyadip Das, Kostas Karagiannis, Andrew Obus, and Vaidehee Thatte. Local
2534
+ oort groups and the isolated differential data criterion, 2019.
2535
+ [7] A. Heller and I. Reiner. Representations of cyclic groups in rings of integers. I. Ann. of Math.
2536
+ (2), 76:73–92, 1962.
2537
+ [8] A. Heller and I. Reiner. Representations of cyclic groups in rings of integers. II. Ann. of Math.
2538
+ (2), 77:318–328, 1963.
2539
+ [9] Sotiris Karanikolopoulos and Aristides Kontogeorgis. Representation of cyclic groups in pos-
2540
+ itive characteristic and Weierstrass semigroups. J. Number Theory, 133(1):158–175, 2013.
2541
+ [10] Aristides Kontogeorgis and Alexios Terezakis. The canonical ideal and the deformation theory
2542
+ of curves with automorphisms, 2021.
2543
+ [11] Andrew Obus. The (local) lifting problem for curves. In Galois-Teichm¨uller theory and arith-
2544
+ metic geometry, volume 63 of Adv. Stud. Pure Math., pages 359–412. Math. Soc. Japan,
2545
+ Tokyo, 2012.
2546
+ [12] Andrew Obus. A generalization of the Oort conjecture. Comment. Math. Helv., 92(3):551–
2547
+ 620, 2017.
2548
+ [13] Andrew Obus and Rachel Pries. Wild tame-by-cyclic extensions. J. Pure Appl. Algebra,
2549
+ 214(5):565–573, 2010.
2550
+ [14] Andrew Obus and Stefan Wewers. Cyclic extensions and the local lifting problem. Ann. of
2551
+ Math. (2), 180(1):233–284, 2014.
2552
+ [15] Florian Pop. The Oort conjecture on lifting covers of curves. Ann. of Math. (2), 180(1):285–
2553
+ 322, 2014.
2554
+ [16] Jean-Pierre Serre. Linear representations of finite groups. Springer-Verlag, New York, 1977.
2555
+ Translated from the second French edition by Leonard L. Scott, Graduate Texts in Mathe-
2556
+ matics, Vol. 42.
2557
+ [17] Jean-Pierre Serre. Local fields. Springer-Verlag, New York, 1979. Translated from the French
2558
+ by Marvin Jay Greenberg.
2559
+ [18] Bradley Weaver. The local lifting problem for D4. Israel J. Math., 228(2):587–626, 2018.
2560
+ Department of Mathematics, National and Kapodistrian University of Athens Pane-
2561
+ pistimioupolis, 15784 Athens, Greece
2562
+ Email address: kontogar@math.uoa.gr
2563
+ Department of Mathematics, National and Kapodistrian University of Athens, Panepis-
2564
+ timioupolis, 15784 Athens, Greece
2565
+ Email address: aleksistere@math.uoa.gr
2566
+
DdAzT4oBgHgl3EQfGvsw/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
E9E5T4oBgHgl3EQfVA-E/content/2301.05548v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:50aa7a1f7bed2011dd59bf963a023f46f215354158229c3f3d872d8fdd156553
3
+ size 834963
E9E5T4oBgHgl3EQfVA-E/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:38c9a5f8297dbed285fe4ba74319ef0b05ff9d66dc9c80eaf4853a532a32f220
3
+ size 3801133
EdE1T4oBgHgl3EQfqQXv/content/tmp_files/2301.03342v1.pdf.txt ADDED
@@ -0,0 +1,1902 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ JOURNAL OF LATEX CLASS FILES, VOL. XX, NO. X, FEB. 2019
2
+ 1
3
+ Real-time Feedback Based Online Aggregate EV
4
+ Power Flexibility Characterization
5
+ Dongxiang Yan, Shihan Huang, and Yue Chen, Member, IEEE
6
+ Abstractβ€”As an essential measure to combat global warming,
7
+ electric vehicles (EVs) have witnessed rapid growth. Meanwhile,
8
+ thanks to the flexibility of EV charging, vehicle-to-grid (V2G)
9
+ interaction has captured great attention. However, the direct con-
10
+ trol of individual EVs is challenging due to their small capacity,
11
+ large number, and private information. Hence, it is the aggregator
12
+ that interacts with the grid on behalf of EVs by characterizing
13
+ their aggregate flexibility. In this paper, we focus on the aggregate
14
+ EV power flexibility characterization problem. First, an offline
15
+ model is built to obtain the lower and upper bounds of the
16
+ aggregate power flexibility region. It ensures that any trajectory
17
+ within the region is feasible. Then, considering that parameters
18
+ such as real-time electricity prices and EV arrival/departure
19
+ times are not known in advance, an online algorithm is developed
20
+ based on Lyapunov optimization techniques. We prove that the
21
+ charging time delays for EVs always meet the requirement even
22
+ if they are not considered explicitly. Furthermore, real-time
23
+ feedback is designed and integrated into the proposed online
24
+ algorithm to better unlock EV power flexibility. Comprehensive
25
+ performance comparisons are carried out to demonstrate the
26
+ advantages of the proposed method.
27
+ Index Termsβ€”Aggregate flexibility, charging station, electric
28
+ vehicle, Lyapunov optimization, online algorithm.
29
+ I. INTRODUCTION
30
+ T
31
+ HANKS to the low carbon emissions, electric vehicles
32
+ (EVs) have been considered a promising solution to
33
+ climate change and proliferate in recent years [1]. However,
34
+ the uncontrolled charging of a large number of EVs can cause
35
+ voltage deviation, line overload, and huge transmission loss
36
+ [2], threatening the reliability of the power system. Unlike
37
+ inelastic loads, the charging power and charging period of
38
+ EVs are more flexible [3]. Therefore, unlocking the power
39
+ flexibility hidden in EVs is a promising way to lessen the
40
+ adverse impact of EVs on the power grid.
41
+ There are extensive literature aiming to design coordinated
42
+ charging strategies to optimally schedule EV charging. For
43
+ example, to promote local renewable generation consumption,
44
+ a dynamic charging strategy was proposed to allow the EV
45
+ charging power to dynamically track the PV generation [4]
46
+ and wind generation [5]. To save the electricity cost, a
47
+ deterministic optimal charging strategy was proposed for a
48
+ home energy management system based on the time-of-use
49
+ tariffs [6]. A model predictive control (MPC) algorithm was
50
+ proposed to minimize the operational cost of EV charging
51
+ stations [7] relying on short-term forecasts. To address the
52
+ uncertainties related to EV charging, reference [8] proposed a
53
+ D. Yan, S. Huang, and Y. Chen are with the Department of Me-
54
+ chanical and Automation Engineering, the Chinese University of Hong
55
+ Kong,
56
+ Hong
57
+ Kong
58
+ SAR,
59
+ China
60
+ (e-mail:
61
+ dongxiangyan@cuhk.edu.hk,
62
+ shhuang@link.cuhk.edu.hk, yuechen@mae.cuhk.edu.hk).
63
+ stochastic charging strategy based on the probabilistic model
64
+ related to EV daily travels. A combined robust and stochastic
65
+ MPC method was developed in [9] to handle the uncertain EV
66
+ charging behaviors and renewable generations. A multi-stage
67
+ energy management strategy including day-ahead and real-
68
+ time stages was developed for a charging station integrated
69
+ with PV generation and energy storage [10]. In addition,
70
+ a pricing mechanism was suggested in [11] to guide EVs
71
+ for economical charging. A double-layer optimization model
72
+ was built to reduce the voltage violations caused by EV
73
+ charging [12]. Despite the efforts mentioned above that intend
74
+ to determine the EV charging power, it is challenging to
75
+ directly control a large number of individual EVs due to the
76
+ high computational complexity.
77
+ To get around this problem, some other literature en-
78
+ deavored to characterize the EV charging power flexibility.
79
+ Reference [13] proposed to model the aggregate EV charging
80
+ flexibility region by the lower and upper bounds of power and
81
+ cumulative energy. This aggregate EV model was adopted by
82
+ [14] to evaluate the achievable vehicle-to-grid capacity of an
83
+ EV fleet and by [15] to quantify the value of EV flexibility in
84
+ terms of maintaining distribution system reliability. Reference
85
+ [16] further considered the spatio-temporal distribution of the
86
+ probability that an EV is available for charging during the
87
+ aggregation and clustering processes. An EV dispatchable
88
+ region was proposed to allow charging stations to participate in
89
+ market bidding [17]. In addition, the aggregate flexibility issue
90
+ was also studied in the fields of thermostatically controllable
91
+ loads (TCLs) [18], distributed energy resources [19], and
92
+ virtual power plant [20]. For example, a geometric approach
93
+ was utilized to model the aggregate flexibility of TCLs [21].
94
+ An inner box approximation method was proposed to charac-
95
+ terize the power flexibility region of various distributed energy
96
+ resources [22].
97
+ The above works provide sound techniques for evaluating
98
+ EV flexibility in an offline manner. It means that the aggregator
99
+ is assumed to have complete information of future uncertainty
100
+ realizations, e.g., EV arrival/departure time, and electricity
101
+ prices. In practice, those data are usually unavailable or
102
+ inaccurate, making the obtained region fail to reflect the actual
103
+ EV aggregate flexibility in real-time. Thus, an online algorithm
104
+ is desired. A straightforward approach is the greedy algorithm
105
+ that decomposes the offline problem into subproblems in each
106
+ time slot by neglecting the time-coupling constraints [23].
107
+ Obviously, the result could be far from optimum. Hence, we
108
+ resort to another approach, Lyapunov optimization, that can
109
+ run in an online manner but with an outcome near to the
110
+ offline optimum [24]. Lyapunov optimization has been used in
111
+ arXiv:2301.03342v1 [math.OC] 9 Jan 2023
112
+
113
+ JOURNAL OF LATEX CLASS FILES, VOL. XX, NO. X, FEB. 2019
114
+ 2
115
+ microgrid control [25], energy storage sharing [26], and data
116
+ center energy management [27]. For EV charging, a charging
117
+ strategy based on Lyapunov optimization was proposed to
118
+ minimize the total electricity cost [28]. However, it cannot
119
+ guarantee that EVs will depart with desired amount of energy.
120
+ To meet the EV charging requirement, virtual delay queues
121
+ were introduced to minimize the charging cost under uncertain
122
+ renewable generations and electricity prices [29], [30].
123
+ Though the optimal online EV charging strategy has been
124
+ widely studied as above, the online aggregate EV power
125
+ flexibility characterization problem has not been well explored
126
+ yet. The latter problem is more complicated than the former
127
+ one, requiring a new model and algorithm design. This paper
128
+ proposes a real-time feedback based online aggregate EV
129
+ power flexibility characterization method. Our main contribu-
130
+ tions are two-fold:
131
+ 1) Model. We first propose an offline optimization model
132
+ to characterize the aggregate EV power flexibility region.
133
+ It decomposes the time-coupled flexibility region into each
134
+ time slot and gives their lower and upper bounds. We prove
135
+ that any trajectory within the region is achievable. Then,
136
+ by categorizing the EVs according to their allowable time
137
+ delays, we develop a counterpart of the offline model that
138
+ enables the further utilization of the Lyapunov optimization
139
+ framework. The proposed model has not been reported in
140
+ previous literature.
141
+ 2) Algorithm. A real-time feedback based online algorithm
142
+ is developed to derive the aggregate EV power flexibility
143
+ region sequentially. First, to fit into the Lyapunov optimization
144
+ framework, charging task queues and delay-aware virtual
145
+ queues are introduced to reformulate the model. Then, a
146
+ drift-plus-penalty term is constructed and by minimizing its
147
+ upper bound, an online algorithm is developed. We prove
148
+ that the charging time delays for EVs will not exceed their
149
+ maximum allowable values even if they are not explicitly
150
+ considered. The bound of optimality gap between the offline
151
+ and online outcomes is provided theoretically. Furthermore,
152
+ real-time dispatch strategy based feedback is designed and
153
+ integrated into the online algorithm. The proposed real-time
154
+ feedback based online algorithm is prediction free and can
155
+ adapt to uncertainties such as random electricity prices and EV
156
+ charging behaviors. Furthermore, it can make use of the most
157
+ recent information, allowing it to even outperform the offline
158
+ model with full knowledge of future uncertainty realizations
159
+ but without the updated dispatch information.
160
+ The rest of this paper is organized as follows. Section
161
+ II formulates the offline model for deriving aggregate EV
162
+ charging power flexibility region. Section III and IV introduce
163
+ the Lyapunov optimization method and real-time feedback
164
+ design, respectively, to generate flexibility region in an online
165
+ manner. Simulation results are presented in Section V. Finally,
166
+ Section VI concludes this paper.
167
+ II. PROBLEM FORMULATION
168
+ In this section, we first introduce the concept of aggregate
169
+ EV power flexibility and then formulate an offline optimization
170
+ problem to approximate it.
171
+ A. Aggregate EV Charging Power Flexibility
172
+ As shown in Fig. 1, when an EV v ∈ V arrives at the
173
+ charging station, it submits its charging task to the aggregator.
174
+ The task is described by (ta
175
+ v, td
176
+ v, ea
177
+ v, ed
178
+ v), where ta
179
+ v is its arrival
180
+ time, td
181
+ v is its departure time, ea
182
+ v is the initial battery energy
183
+ level at ta
184
+ v, and ed
185
+ v is the desired energy level when it leaves.
186
+ For the EV v, the maximum allowable charging time delay is
187
+ Rv = td
188
+ vβˆ’ta
189
+ v. The EV charging task needs to be finished within
190
+ this declared time duration. With the submitted information,
191
+ the aggregator can flexibly schedule the EV charging to meet
192
+ the charging requirement. Two possible trajectories to meet the
193
+ EV charging need are depicted in Fig. 1. Let {pc
194
+ v,t, οΏ½οΏ½t} be the
195
+ charging power of EV v over time. The range that the charging
196
+ power can vary within is called the power flexibility of EV
197
+ v. If we sum the power flexibility of all EVs in a charging
198
+ station up, we can get the aggregate EV power flexibility of
199
+ the charging station.
200
+ (𝑑1
201
+ π‘Ž, 𝑑1
202
+ 𝑑,
203
+ 𝑒1
204
+ π‘Ž, 𝑒1
205
+ 𝑑)
206
+ (𝑑2
207
+ π‘Ž, 𝑑2
208
+ 𝑑,
209
+ 𝑒2
210
+ π‘Ž, 𝑒2
211
+ 𝑑)
212
+ (π‘‘π‘£π‘Ž, 𝑑𝑣𝑑,
213
+ π‘’π‘£π‘Ž, 𝑒𝑣𝑑)
214
+ Aggregator
215
+ Distribution System Operator
216
+ Aggregate
217
+ dispatch power
218
+ Aggregate power
219
+ flexibility region
220
+ Ƽ𝑝𝑑,𝑑, Ƹ𝑝𝑑,𝑑
221
+ 𝑝𝑑,𝑑
222
+ 𝑑𝑖𝑠𝑝
223
+ 𝑝1,𝑑
224
+ 𝑑𝑖𝑠𝑝
225
+ 𝑝2,𝑑
226
+ 𝑑𝑖𝑠𝑝
227
+ 𝑝𝑣,𝑑
228
+ 𝑑𝑖𝑠𝑝
229
+ EV 1
230
+ EV 2
231
+ EV v
232
+ 1. Generate EV aggregate power flexibility region
233
+ 2. Disaggregation
234
+ Fig. 1. System diagram and illustration of EV power flexibility.
235
+ However, it is difficult to characterize the EV power flexibil-
236
+ ity for each time slot due to the temporal-coupled EV charging
237
+ constraints. The EV power flexibility in the current time slot is
238
+ affected by those in the past time slots and further affects those
239
+ in the future time slots. This is different from the traditional
240
+ controllable generators whose flexibility can be described by
241
+ the minimum and maximum power outputs in each time slot.
242
+ In the following, we aim to derive an aggregate EV power
243
+ flexibility region that: 1) is time-decoupled so that it can be
244
+ used in real-time power system operation; and 2) any trajectory
245
+ within it can meet the EV charging requirement.
246
+ B. Offline Problem Formulation
247
+ Suppose there are T time slots, indexed by t ∈ T
248
+ =
249
+ {1, ..., T}. The desired time-decoupled aggregate EV power
250
+ flexibility region can be represented by a series of intervals
251
+ [Λ‡pd,t, Λ†pd,t], βˆ€t ∈ T . The intervals can be specified by a lower
252
+ power trajectory {Λ‡pd,t, βˆ€t} and an upper power trajectory
253
+ {Λ†pd,t, βˆ€t}. To obtain the lower and upper power trajectories,
254
+ we formulate the following offline optimization problem:
255
+ P1 :
256
+ max
257
+ Λ†pd,t,Λ‡pd,t,βˆ€t lim
258
+ T β†’βˆž
259
+ 1
260
+ T
261
+ T
262
+ οΏ½
263
+ t=1
264
+ E
265
+ οΏ½
266
+ Ft
267
+ οΏ½
268
+ ,
269
+ (1a)
270
+
271
+ Power
272
+ Aggregate power
273
+ flexibility region
274
+ flexibilitJOURNAL OF LATEX CLASS FILES, VOL. XX, NO. X, FEB. 2019
275
+ 3
276
+ where
277
+ Ft = Ο€t(Λ†pd,t βˆ’ Λ‡pd,t), βˆ€t,
278
+ (1b)
279
+ subject to
280
+ Λ†pd,t =
281
+ οΏ½
282
+ v∈V
283
+ Λ†pc
284
+ v,t, βˆ€t,
285
+ (1c)
286
+ 0 ≀ Λ†pc
287
+ v,t ≀ pmax
288
+ v
289
+ , βˆ€v, βˆ€t,
290
+ (1d)
291
+ Λ†ev,t+1 = Λ†ev,t + Λ†pc
292
+ v,tβˆ†t, βˆ€v, βˆ€t ΜΈ= T,
293
+ (1e)
294
+ Λ†ev,tav = eini
295
+ v , Λ†ev,tdv β‰₯ ed
296
+ v, βˆ€v,
297
+ (1f)
298
+ emin
299
+ v
300
+ ≀ Λ†ev,t ≀ emax
301
+ v
302
+ , βˆ€v, βˆ€t,
303
+ (1g)
304
+ Λ‡pd,t =
305
+ οΏ½
306
+ n∈V
307
+ Λ‡pc
308
+ v,t, βˆ€t,
309
+ (1h)
310
+ 0 ≀ Λ‡pc
311
+ v,t ≀ pmax
312
+ v
313
+ , βˆ€v, βˆ€t,
314
+ (1i)
315
+ Λ‡ev,t+1 = Λ‡ev,t + Λ‡pc
316
+ v,tβˆ†t, βˆ€v, βˆ€t ΜΈ= T,
317
+ (1j)
318
+ Λ‡ev,tav = eini
319
+ v , Λ‡ev,tdv β‰₯ ed
320
+ v, βˆ€v,
321
+ (1k)
322
+ emin
323
+ v
324
+ ≀ Λ‡ev,t ≀ emax
325
+ v
326
+ , βˆ€v, βˆ€t,
327
+ (1l)
328
+ Λ‡pd,t ≀ Λ†pd,t, βˆ€t,
329
+ (1m)
330
+ Λ†pc
331
+ v,t/Λ‡pc
332
+ v,t =
333
+ οΏ½ Λ†pc
334
+ v,t/Λ‡pc
335
+ v,t,
336
+ if t ∈ [ta
337
+ v, td
338
+ v]
339
+ 0,
340
+ if t < ta
341
+ v βˆͺ t > td
342
+ v
343
+ .
344
+ (1n)
345
+ In the objective function (1a)-(1b), Ο€t, βˆ€t are the real-time
346
+ electricity prices, showing the unit value of power flexibility
347
+ in different time slots. Hence, the objective function aims to
348
+ maximize the value of total aggregate EV power flexibility.
349
+ Constraint (1c) defines the upper bound of aggregate EV
350
+ power flexibility region. The charging power of an EV v is
351
+ limited by (1d), where pmax
352
+ v
353
+ is the maximum charging power.
354
+ Constraint (1f) defines the EV’s initial energy level and the
355
+ charging requirement. (1e) and (1g) describe the EV’s energy
356
+ dynamics and battery capacity. Similarly, (1h)-(1l) are the
357
+ constraints related to the lower bound of the aggregate EV
358
+ power flexibility. (1m) is the joint constraint to ensure that
359
+ {Λ†pd,t, βˆ€t} and {Λ‡pd,t, βˆ€t} provide the upper and lower bounds,
360
+ respectively. (1n) limits that charging only happens during the
361
+ EV’s declared parking time.
362
+ Proposition 1: Any aggregate EV charging power trajectory
363
+ within [Λ‡pd,1, Λ†pd,1] Γ— Β· Β· Β· Γ— [Λ‡pd,T , Λ†pd,T ] is achievable.
364
+ The proof of Proposition 1 can be found in Appendix A.
365
+ Despite this nice property, the offline optimization problem
366
+ above cannot be solved directly since it requires complete
367
+ knowledge of the future EV charging tasks and future elec-
368
+ tricity prices, which are usually not available in practice.
369
+ Therefore, an online algorithm is necessary. To this end, in
370
+ the next section, we will first propose a closely related but
371
+ more flexible form of the problem studied. Then, we adopt the
372
+ Lyapunov optimization framework to reformulate the offline
373
+ problem into an online one. We construct charging task queues
374
+ and delay-aware virtual queues to ensure the satisfaction of
375
+ charging requirements. Furthermore, considering the impact
376
+ of real-time dispatch decisions on the future aggregate EV
377
+ power flexibility, a real-time feedback based online flexibility
378
+ characterization method is developed in Section IV to avoid
379
+ the potential underestimate of EV power flexibility.
380
+ III. ONLINE ALGORITHM
381
+ In this section, we adopt the Lyapunov optimization frame-
382
+ work to solve the offline problem P1 in an online manner.
383
+ The proposed algorithm can output an aggregate EV power
384
+ flexibility value with an economic value close to P1.
385
+ A. Problem Modification
386
+ As mentioned above, the charging station serves dozens of
387
+ EVs every day, and each EV arrives along with a charging task,
388
+ i.e., (ta
389
+ v, td
390
+ v, ea
391
+ v, ed
392
+ v). Those EV charging tasks can be first stored
393
+ in a queue and be served later according to a first-in-first-
394
+ out basis. Since different EVs may have different allowable
395
+ charging time delays, we use multiple queues to classify and
396
+ collect the EV charging tasks. Suppose there are G types
397
+ of charging time delays Rgs, each of which is indexed by
398
+ g ∈ {1, 2, ..., G}. Correspondingly, we construct G queues to
399
+ collect the respective charging tasks, and each queue is denoted
400
+ by Qg. For queue Qg, Qg,t refers to its charging task backlog
401
+ in time slot t. The queue backlog growth is described by
402
+ Qg,t+1 = max[Qg,t βˆ’ xg,t, 0] + ag,t,
403
+ (2)
404
+ where xg,t is the charging power for EVs in group g at time
405
+ t, and ag,t is the arrival rate of EV charging tasks of group g
406
+ at time t. In particular, ag,t sums up the energy demand of all
407
+ EVs that arrive at the beginning of time t, i.e.,
408
+ ag,t =
409
+ οΏ½
410
+ v∈Vg
411
+ ag,v,t,
412
+ (3)
413
+ where ag,v,t is the charging demand of EV v of group g in
414
+ time slot t. Vg is the set of EVs in group g.
415
+ Recalling that our target in P1 is to derive an upper bound
416
+ and a lower bound for the aggregate EV power flexibility
417
+ region, we correspondingly define the upper bound queue Λ†Qg,t
418
+ and the lower bound queue Λ‡Qg,t. Similar to (2), we have
419
+ Λ†Qg,t+1 = max[ Λ†Qg,t βˆ’ Λ†xg,t, 0] + Λ†ag,t,
420
+ (4)
421
+ Λ‡Qg,t+1 = max[ Λ‡Qg,t βˆ’ Λ‡xg,t, 0] + Λ‡ag,t,
422
+ (5)
423
+ where Λ†xg,t and Λ‡xg,t are the charging power for upper and
424
+ lower bound queues, respectively, i.e., Λ†xg,t = οΏ½
425
+ v∈Vg Λ†pc
426
+ v,t and
427
+ Λ‡xg,t = οΏ½
428
+ v∈Vg Λ‡pc
429
+ v,t.
430
+ The upper and lower bounds of arriving charging demand,
431
+ i.e., Λ†ag,t and Λ‡ag,t, are determined by
432
+ Λ†ag,t =
433
+ οΏ½
434
+ v∈Vg
435
+ Λ†ag,v,t, Λ‡ag,t =
436
+ οΏ½
437
+ v∈Vg
438
+ Λ‡ag,v,t,
439
+ (6)
440
+ Particularly, the lower bound of arriving charging demand
441
+ Λ‡ag,v,t can be determined in the following charging as soon as
442
+ possible way,
443
+ Λ‡ag,v,t =
444
+ οΏ½
445
+ οΏ½
446
+ οΏ½
447
+ pmax
448
+ v
449
+ ,
450
+ ta
451
+ v ≀ t < βŒŠΛ‡tmin
452
+ v
453
+ βŒ‹ + ta
454
+ v
455
+ Λ‡echa
456
+ v
457
+ /Ξ·c βˆ’ βŒŠΛ‡tmin
458
+ v
459
+ βŒ‹pmax
460
+ v
461
+ ,
462
+ t = βŒŠΛ‡tmin
463
+ v
464
+ βŒ‹ + ta
465
+ v
466
+ 0,
467
+ otherwise
468
+ ,
469
+ (7)
470
+ where Λ‡echa
471
+ v
472
+ = ed
473
+ v βˆ’ea
474
+ v, Λ‡tmin
475
+ v
476
+ is the minimum required charging
477
+ time determined by Λ‡tmin
478
+ v
479
+ =
480
+ Λ‡echa
481
+ v
482
+ pmax
483
+ v
484
+ Ξ·c , and ⌊.βŒ‹ means rounding
485
+ down to the nearest integer.
486
+
487
+ JOURNAL OF LATEX CLASS FILES, VOL. XX, NO. X, FEB. 2019
488
+ 4
489
+ Different from Λ‡ag,v,t, the upper bound of arrival charging
490
+ demand Λ†ag,v,t is determined using the maximum charging
491
+ demand emax
492
+ v
493
+ instead of ed
494
+ v. Denote Λ†echa
495
+ v
496
+ = emax
497
+ v
498
+ βˆ’ ea
499
+ v,
500
+ Λ†tmin
501
+ v
502
+ =
503
+ Λ†echa
504
+ v
505
+ pmax
506
+ v
507
+ Ξ·c , and then Λ†ag,v,t can be determined by
508
+ Λ†ag,v,t =
509
+ οΏ½
510
+ οΏ½
511
+ οΏ½
512
+ pmax
513
+ v
514
+ ,
515
+ ta
516
+ v ≀ t < βŒŠΛ†tmin
517
+ v
518
+ βŒ‹ + ta
519
+ v
520
+ Λ†echa
521
+ v
522
+ /Ξ·c βˆ’ βŒŠΛ†tmin
523
+ v
524
+ βŒ‹pmax
525
+ v
526
+ ,
527
+ t = βŒŠΛ†tmin
528
+ v
529
+ βŒ‹ + ta
530
+ v
531
+ 0,
532
+ otherwise.
533
+ (8)
534
+ We then formulate the aggregate EV power flexibility char-
535
+ acterization problem as follows:
536
+ P2 :
537
+ min
538
+ Λ†xg,t,Λ‡xg,t lim
539
+ T β†’βˆž
540
+ 1
541
+ T
542
+ T
543
+ οΏ½
544
+ t=1
545
+ E
546
+ οΏ½
547
+ βˆ’ Ft
548
+ οΏ½
549
+ ,
550
+ (9a)
551
+ subject to
552
+ lim
553
+ T β†’βˆž
554
+ 1
555
+ T
556
+ T
557
+ οΏ½
558
+ t=1
559
+ E[Λ†ag,t βˆ’ Λ†xg,t] ≀ 0, βˆ€g
560
+ (9b)
561
+ lim
562
+ T β†’βˆž
563
+ 1
564
+ T
565
+ T
566
+ οΏ½
567
+ t=1
568
+ E[Λ‡ag,t βˆ’ Λ‡xg,t] ≀ 0, βˆ€g
569
+ (9c)
570
+ 0 ≀ Λ†xg,t ≀ min{xg,max, Λ†Qg,t}, βˆ€g
571
+ (9d)
572
+ 0 ≀ Λ‡xg,t ≀ min{xg,max, Λ‡Qg,t}, βˆ€g
573
+ (9e)
574
+ Λ†xg,t β‰₯ Λ‡xg,t, βˆ€g
575
+ (9f)
576
+ where xg,max = οΏ½
577
+ v∈Vg pmax
578
+ v
579
+ . Constraint (9b) ensures that if
580
+ using the upper bound trajectory {Λ†xg,t, βˆ€t}, the total charg-
581
+ ing requirement can be satisfied in the long run. Constraint
582
+ (9c) poses a similar requirement for the lower bound trajec-
583
+ tory {Λ‡xg,t, βˆ€t}. Based on (9b), (9c), and the definitions of
584
+ Λ†Qg,t+1, Λ‡Qg,t+1 in (4)-(5), we can prove that the queues Λ†Qg,t
585
+ and Λ‡Qg,t are mean rate stable. To be specific,
586
+ Λ†Qg,t+1 βˆ’ Λ†ag,t β‰₯ Λ†Qg,t βˆ’ Λ†xg,t, βˆ€t
587
+ (10)
588
+ Summing (10) up over all t and divide both sides by T yields
589
+ 0 ≀ E[ Λ†Qg,T ]
590
+ T
591
+ ≀
592
+ οΏ½T
593
+ t=1 E[Λ†ag,t βˆ’ Λ†xg,t]
594
+ T
595
+ (11)
596
+ Hence, lim
597
+ T β†’βˆž
598
+ E[ Λ†
599
+ Qg,T ]
600
+ T
601
+ = 0. Similarly, lim
602
+ T β†’βˆž
603
+ E[ Λ‡
604
+ Qg,T ]
605
+ T
606
+ = 0.
607
+ Constraints (9d) and (9e) give the upper and lower bounds
608
+ of the aggregate EV charging power for group g, respectively.
609
+ The upper bound is no less than the lower bound, as shown in
610
+ (9f). The P2 provides a counterpart problem for P1. Similar
611
+ to the proof of Proposition 1, we can prove that any trajectory
612
+ between [Λ‡xg,t, Λ†xg,t] is achievable. However, the allowable
613
+ charging delay is not considered in P2, which may result in
614
+ unfulfilled EV charging tasks upon departure.
615
+ B. Construct Virtual Queues
616
+ To overcome the aforementioned charging delay issue, we
617
+ introduce delay-aware virtual queues,
618
+ Λ†Zg,t+1 = max{ Λ†Zg,t + Ξ·g
619
+ Rg
620
+ I Λ†
621
+ Qg,t>0 βˆ’ Λ†xg,t, 0}, βˆ€g, βˆ€t
622
+ (12)
623
+ Λ‡Zg,t+1 = max{ Λ‡Zg,t + Ξ·g
624
+ Rg
625
+ I Λ‡
626
+ Qg,t>0 βˆ’ Λ‡xg,t, 0}, βˆ€g, βˆ€t
627
+ (13)
628
+ where I Λ†
629
+ Qg,t>0 and I Λ‡
630
+ Qg,t>0 are indicator functions of Λ†Qg,t and
631
+ Λ‡Qg,t, respectively. They are equal to 1 if there exists unserved
632
+ charging tasks in the queues, i.e., Λ†Qg,t > 0 and Λ‡Qg,t > 0.
633
+ Using Ξ·g
634
+ Rg to times it, this whole term constitutes a penalty to
635
+ the virtual queue backlog. ηg is a user-defined parameter that
636
+ can adjust the growth rate of the virtual queues. For instance,
637
+ increasing the value of Ξ·g leads to a fast queue growth and a
638
+ larger backlog value, calling for more attention to accelerate
639
+ the charging process. We prove that, when Qg,t and Zg,t have
640
+ finite upper bounds, with a proper ηg, the charging time delay
641
+ for EVs in group g is bounded.
642
+ Proposition 2: Suppose Λ†Qg,t, Λ‡Qg,t, Λ†Zg,t, and Λ‡Zg,t have finite
643
+ upper bounds, e.g., Λ†Qg,t ≀ Λ†Qg,max, Λ‡Qg,t ≀ Λ‡Qg,max Λ†Zg,t ≀
644
+ Λ†Zg,max, and Λ‡Zg,t ≀ Λ‡Zg,max. The charging time delay of all
645
+ EVs in group g is upper bounded by Λ†Ξ΄g,max and Λ‡Ξ΄g,max time
646
+ slots, where
647
+ Λ†Ξ΄g,max := ( Λ†Qg,max + Λ†Zg,max)Rg
648
+ Ξ·g
649
+ ,
650
+ (14)
651
+ Λ‡Ξ΄g,max := ( Λ‡Qg,max + Λ‡Zg,max)Rg
652
+ Ξ·g
653
+ .
654
+ (15)
655
+ The proof of Proposition 2 can be found in Appendix B.
656
+ It ensures that the charging tasks can always be fulfilled
657
+ within the available charging periods by properly setting the
658
+ parameters Ξ·g, βˆ€g.
659
+ C. Lyapunov Optimization
660
+ Based on the charging task queues and delay-aware virtual
661
+ queues, the Lyapunov optimization framework is applied as
662
+ follows.
663
+ 1) Lyapunov
664
+ Function:
665
+ First,
666
+ we
667
+ define
668
+ Θt
669
+ =
670
+ ( Λ†
671
+ Qt, Λ†
672
+ Zt, Λ‡
673
+ Qt, Λ‡
674
+ Zt) as the concatenated vector of queues,
675
+ where
676
+ Λ†
677
+ Qt = ( Λ†Q1,t, ..., Λ†QG,t),
678
+ (16a)
679
+ Λ†
680
+ Zt = ( Λ†Z1,t, ..., Λ†ZG,t),
681
+ (16b)
682
+ Λ‡
683
+ Qt = ( Λ‡Q1,t, ..., Λ‡QG,t),
684
+ (16c)
685
+ Λ‡
686
+ Zt = ( Λ‡Z1,t, ..., Λ‡ZG,t).
687
+ (16d)
688
+ The Lyapunov function is then defined as
689
+ L(Θt) = 1
690
+ 2
691
+ οΏ½
692
+ g∈G
693
+ Λ†Q2
694
+ g,t + 1
695
+ 2
696
+ οΏ½
697
+ g∈G
698
+ Λ†Z2
699
+ g,t + 1
700
+ 2
701
+ οΏ½
702
+ g∈G
703
+ Λ‡Q2
704
+ g,t + 1
705
+ 2
706
+ οΏ½
707
+ g∈G
708
+ Λ‡Z2
709
+ g,t,
710
+ (17)
711
+ where L(Θt) can be considered as a measure of the queue
712
+ size. A smaller L(Θt) is preferred to push (virtual) queues
713
+ Λ†Qg,t, Λ†Zg,t, Λ‡Qg,t, and Λ‡Zg,t to be less congested.
714
+ 2) Lyapunov Drift: The conditional one-time slot Lyapunov
715
+ drift is defined as follows:
716
+ βˆ†(Θt) = E[L(Θt+1) βˆ’ L(Θt)|Θt],
717
+ (18)
718
+ where the expectation is taken with respect to the random Θt.
719
+ The Lyapunov drift is a measure of the expectation of the
720
+ queue size growth given the current state Θt. Intuitively, by
721
+ minimizing the Lyapunov drift, virtual queues are expected
722
+ to be stabilized. However, only minimizing the Lyapunov
723
+ drift may lead to a low aggregate EV power flexibility value.
724
+
725
+ JOURNAL OF LATEX CLASS FILES, VOL. XX, NO. X, FEB. 2019
726
+ 5
727
+ Therefore, we include the expected aggregate flexibility value
728
+ (1b) for the time slot t to (18). The drift-plus-penalty term is
729
+ obtained, i.e.,
730
+ βˆ†(Θt) + V E[βˆ’Ft|Θt],
731
+ (19)
732
+ where V is a weight parameter that controls the trade-off
733
+ between (virtual) queues stability and aggregate EV power
734
+ flexibility maximization.
735
+ 3) Minimizing the Upper Bound: (19) is still time-coupled
736
+ due to the definition of βˆ†(Θt). To adapt to online imple-
737
+ mentation, instead of directly minimizing the drift-plus-penalty
738
+ term, we minimize the upper bound to obtain the upper and
739
+ lower bounds of aggregate EV power flexibility region. We
740
+ first calculate the one-time slot Lyapunov drift:
741
+ L(Θt+1) βˆ’ L(Θt)
742
+ = 1
743
+ 2
744
+ οΏ½
745
+ g∈G
746
+ οΏ½ οΏ½
747
+ Λ†Q2
748
+ g,t+1 βˆ’ Λ†Q2
749
+ g,t
750
+ οΏ½
751
+ +
752
+ οΏ½
753
+ Λ†Z2
754
+ g,t+1 βˆ’ Λ†Z2
755
+ g,t
756
+ οΏ½
757
+ +
758
+ οΏ½ Λ‡Q2
759
+ g,t+1 βˆ’ Λ‡Q2
760
+ g,t
761
+ οΏ½
762
+ +
763
+ οΏ½ Λ‡Z2
764
+ g,t+1 βˆ’ Λ‡Z2
765
+ g,t
766
+ οΏ½ οΏ½
767
+ .
768
+ (20)
769
+ Use queue Λ†Qg,t as an example, based on the queue update
770
+ equations (4), we have
771
+ Λ†Q2
772
+ g,t+1 = {max[ Λ†Qg,t βˆ’ Λ†xg,t, 0] + Λ†ag,t}2
773
+ ≀ Λ†Q2
774
+ g,t + Λ†a2
775
+ g,max + Λ†x2
776
+ g,max + 2 Λ†Qg,t(Λ†ag,t βˆ’ Λ†xg,t).
777
+ (21)
778
+ Thus,
779
+ 1
780
+ 2
781
+ οΏ½
782
+ Λ†Q2
783
+ g,t+1 βˆ’ Λ†Q2
784
+ g,t
785
+ οΏ½
786
+ ≀ 1
787
+ 2
788
+ οΏ½
789
+ Λ†x2
790
+ g,max + Λ†a2
791
+ g,max
792
+ οΏ½
793
+ + Λ†Qg,t (Λ†ag,t βˆ’ Λ†xg,t) .
794
+ (22)
795
+ Similarly, for queue Λ‡Qg,t, Λ†Zg,t, and Λ‡Zg,t, we have
796
+ 1
797
+ 2
798
+ οΏ½ Λ‡Q2
799
+ g,t+1 βˆ’ Λ‡Q2
800
+ g,t
801
+ οΏ½
802
+ ≀ 1
803
+ 2
804
+ οΏ½
805
+ Λ‡x2
806
+ g,max + Λ‡a2
807
+ g,max
808
+ οΏ½
809
+ + Λ‡Qg,t (Λ‡ag,t βˆ’ Λ‡xg,t) .
810
+ (23)
811
+ 1
812
+ 2[ Λ†Z2
813
+ g,t+1 βˆ’ Λ†Z2
814
+ g,t] ≀ 1
815
+ 2 max[( Ξ·g
816
+ Rg
817
+ )2, Λ†x2
818
+ g,max]
819
+ + Λ†Zg,t[ Ξ·g
820
+ Rg
821
+ βˆ’ Λ†xg,t].
822
+ (24)
823
+ 1
824
+ 2[ Λ‡Z2
825
+ g,t+1 βˆ’ Λ‡Z2
826
+ g,t] ≀ 1
827
+ 2 max[( Ξ·g
828
+ Rg
829
+ )2, Λ‡x2
830
+ g,max]
831
+ + Λ‡Zg,t[ Ξ·g
832
+ Rg
833
+ βˆ’ Λ‡xg,t].
834
+ (25)
835
+ We then substitute inequalities (22),(23), (24) and (25) into
836
+ drift-plus-penalty term and yield
837
+ βˆ†(Θt) + V E[βˆ’Ft|Θt]
838
+ ≀ A + V E[βˆ’Ft|Θt] +
839
+ οΏ½
840
+ g∈G
841
+ Λ†Qg,tE [Λ†ag,t βˆ’ Λ†xg,t|Θt]
842
+ +
843
+ οΏ½
844
+ g∈G
845
+ Λ‡Qg,tE [Λ‡ag,t βˆ’ Λ‡xg,t|Θt] +
846
+ οΏ½
847
+ g∈G
848
+ Λ†Zg,tE [βˆ’Λ†xg,t|Θt]
849
+ +
850
+ οΏ½
851
+ g∈G
852
+ Λ‡Zg,tE [βˆ’Λ‡xg,t|Θt] ,
853
+ (26)
854
+ where A is a constant, i.e.,
855
+ A = 1
856
+ 2
857
+ οΏ½
858
+ g∈G
859
+ (Λ†x2
860
+ g,max + Λ†a2
861
+ g,max) + 1
862
+ 2
863
+ οΏ½
864
+ g∈G
865
+ max[( Ξ·g
866
+ Rg
867
+ )2, Λ†x2
868
+ g,max]
869
+ + 1
870
+ 2
871
+ οΏ½
872
+ g∈G
873
+ (Λ‡x2
874
+ g,max + Λ‡a2
875
+ g,max) + 1
876
+ 2
877
+ οΏ½
878
+ g∈G
879
+ max[( Ξ·g
880
+ Rg
881
+ )2, Λ‡x2
882
+ g,max]
883
+ +
884
+ οΏ½
885
+ g∈G
886
+ [ Λ†Zg,max
887
+ Ξ·g
888
+ Rg
889
+ ] +
890
+ οΏ½
891
+ g∈G
892
+ [ Λ‡Zg,max
893
+ Ξ·g
894
+ Rg
895
+ ].
896
+ By reorganizing the expression in (26) and ignoring the con-
897
+ stant terms, we can obtain the following online optimization
898
+ problem,
899
+ P3 :
900
+ min
901
+ Λ†xg,t,Λ‡xg,t,βˆ€g,βˆ€t
902
+ οΏ½
903
+ gοΏ½οΏ½G
904
+ (βˆ’V Ο€t βˆ’ Λ†Qg,t βˆ’ Λ†Zg,t)Λ†xg,t
905
+ +
906
+ οΏ½
907
+ g∈G
908
+ (V Ο€t βˆ’ Λ‡Qg,t βˆ’ Λ‡Zg,t)Λ‡xg,t,
909
+ (27)
910
+ s.t. (9d) βˆ’ (9f),
911
+ where Λ†Qg,t, Λ‡Qg,t, Λ‡Zg,t, and Λ‡Zg,t are first updated based on
912
+ (4),(5),(12), and (13) before solving P3 in each time slot.
913
+ In each time slot t, given the current system queue state
914
+ Θt, the proposed method determines the current upper and
915
+ lower aggregate EV power flexibility bounds Λ†xg,t and Λ‡xg,t by
916
+ solving problem P3. Hence, the original offline optimization
917
+ problem P1 has been decoupled into simple online (real-time)
918
+ problems, which can be executed in each time slot without
919
+ requiring prior knowledge of future uncertain states. Since
920
+ the modified problem P3 is slightly different from the offline
921
+ one P1, an important issue we care about is: what’s the gap
922
+ between the optimal solutions of the online problem P3 the
923
+ and offline problem P1?
924
+ Proposition 3: Denote the obtained long-term time-average
925
+ aggregate EV power flexibility value of P1 and P3 by F βˆ—
926
+ and F pro, respectively. We have
927
+ 0 ≀ βˆ’F pro + F βˆ— ≀ 1
928
+ V A,
929
+ (28)
930
+ where A is a constant defined in (26).
931
+ The proof of Proposition 3 can be found in Appendix C.
932
+ The optimality gap can be controlled by the parameter V . A
933
+ bigger V value leads to a smaller optimality gap but increased
934
+ queue sizes. In contrast, a smaller V value makes the queues
935
+ more stable but results in a larger optimality gap.
936
+ IV. DISAGGREGATION
937
+ AND REAL-TIME FEEDBACK DESIGN
938
+ In each time slot t, given the aggregate EV power flexibility
939
+ region [οΏ½
940
+ g Λ‡xβˆ—
941
+ g,t, οΏ½
942
+ g Λ†xβˆ—
943
+ g,t], the distribution system operator
944
+ (DSO) can determine the optimal aggregate dispatch strategy
945
+ for EVs. This aggregate dispatch strategy should be further
946
+ disaggregated to obtain the control strategy for each EV,
947
+ which is studied in this section. Moreover, considering that
948
+ the current dispatch strategy will influence the aggregate EV
949
+ power flexibility in future time slots, real-time feedback is
950
+ designed and integrated with the proposed online flexibility
951
+ characterization method in Section III.
952
+
953
+ JOURNAL OF LATEX CLASS FILES, VOL. XX, NO. X, FEB. 2019
954
+ 6
955
+ A. Disaggregation
956
+ Suppose the dispatch strategy in time slot t is pdisp
957
+ agg,t ∈
958
+ [οΏ½
959
+ g Λ‡xβˆ—
960
+ g,t, οΏ½
961
+ g Λ†xβˆ—
962
+ g,t]. This can be determined by the DSO by
963
+ solving an economic dispatch problem based on the up-to-
964
+ date information (e.g., the electricity price Ο€t, the grid-side
965
+ renewable generation). Since this paper focuses on the online
966
+ characterization of aggregate EV power flexibility, the eco-
967
+ nomic dispatch problem by the DSO is omitted for simplicity.
968
+ Interested readers can refer to [31].
969
+ Let the dispatch ratio Ξ±t be
970
+ Ξ±t =
971
+ pdisp
972
+ agg,t βˆ’ οΏ½
973
+ g Λ‡xβˆ—
974
+ g,t
975
+ οΏ½
976
+ g Λ†xβˆ—
977
+ g,t βˆ’ οΏ½
978
+ g Λ‡xβˆ—
979
+ g,t
980
+ .
981
+ (29)
982
+ Then, the dispatched power pdisp
983
+ g,t
984
+ for each group g can be
985
+ determined according to the ratio, i.e.,
986
+ pdisp
987
+ g,t
988
+ = (1 βˆ’ Ξ±t)Λ‡xβˆ—
989
+ g,t + Ξ±tΛ†xβˆ—
990
+ g,t,
991
+ (30)
992
+ which satisfies
993
+ pdisp
994
+ agg,t =
995
+ οΏ½
996
+ g
997
+ pdisp
998
+ g,t .
999
+ The next step is to allocate pdisp
1000
+ g,t
1001
+ to the EVs in the group
1002
+ g. All EVs in the group g are sorted according to their arrival
1003
+ time. Then, we follow the first-in-first-service principle to
1004
+ allocate the energy; namely, the EV that comes earlier will
1005
+ be charged with the maximum charging power. We have
1006
+ pdisp
1007
+ v,t
1008
+ = min
1009
+ οΏ½
1010
+ pdisp
1011
+ g,t , pmax
1012
+ v
1013
+ , emax
1014
+ v
1015
+ βˆ’ ev,t
1016
+ βˆ†t
1017
+ οΏ½
1018
+ , βˆ€v ∈ Vg,
1019
+ (31)
1020
+ where Vg refers to the set of EVs in group g. The third term
1021
+ on the right side is used to ensure that the EV will not exceed
1022
+ its allowable maximum energy level.
1023
+ After this charging assignment for an earlier EV is com-
1024
+ pleted, the following update procedures will execute
1025
+ pdisp
1026
+ g,t
1027
+ ← (pdisp
1028
+ g,t
1029
+ βˆ’ pdisp
1030
+ v,t ),
1031
+ (32)
1032
+ which means deducting pdisp
1033
+ v,t
1034
+ from the total remaining dis-
1035
+ patched power pdisp
1036
+ g,t .
1037
+ Then, the pdisp
1038
+ g,t
1039
+ is allocated to the next earlier arrival EVs
1040
+ until the aggregate EV charging power is completely assigned.
1041
+ At this time, the disaggregation is finished.
1042
+ The dispatched power disaggregation algorithm is presented
1043
+ in Algorithm 1.
1044
+ B. State Update to Improve Power Flexibility Region
1045
+ Following the disaggregation procedures in Algorithm 1, we
1046
+ can get the actual EV dispatched charging power pdisp
1047
+ v,t , βˆ€v.
1048
+ By now, we can move on to the next time slot t + 1 and
1049
+ evaluate the EV power flexibility by solving problem P3,
1050
+ determine the dispatch strategy, disaggregate the dispatched
1051
+ power, and so on. But considering that the current actual
1052
+ dispatched EV charging power can affect the future aggregate
1053
+ EV power flexibility, which is ignored in the aforementioned
1054
+ processes. Therefore, we propose a real-time feedback method
1055
+ to integrate the current actual EV dispatched charging power
1056
+ into the future aggregate EV power flexibility characterization.
1057
+ Algorithm 1 EV Dispatched Power Disaggregation
1058
+ 1: Initialization: aggregate EV dispatched power pdisp
1059
+ agg,t.
1060
+ 2: Calculate the dispatched aggregate charging power pdisp
1061
+ g,t
1062
+ for each group g using (30).
1063
+ 3: for Each group g ∈ G do
1064
+ 4:
1065
+ for Each EV v in group g do
1066
+ 5:
1067
+ if the EV is not available for charging then
1068
+ 6:
1069
+ Let EV v’s charging power pdisp
1070
+ v,t
1071
+ = 0, βˆ€v.
1072
+ 7:
1073
+ else
1074
+ 8:
1075
+ Calculate pdisp
1076
+ v,t
1077
+ according to (31).
1078
+ 9:
1079
+ Update the remaining aggregate power via (32).
1080
+ 10:
1081
+ if If the updated pdisp
1082
+ g,t
1083
+ = 0 then
1084
+ 11:
1085
+ Break and return to Step 3.
1086
+ 12:
1087
+ end if
1088
+ 13:
1089
+ end if
1090
+ 14:
1091
+ end for
1092
+ 15: end for
1093
+ Disaggregation
1094
+ State update
1095
+ Dispatch
1096
+ Aggregate power flexibility
1097
+ region
1098
+ State
1099
+ feedback
1100
+ t=t+1
1101
+ Fig. 2. Overall procedure of the proposed method.
1102
+ The overall procedure is shown in Fig. 2 with the right-hand
1103
+ side blue box showing the real-time feedback.
1104
+ To be specific, we change the constraints (9d)-(9e) for time
1105
+ slot t into
1106
+ Λ†xg,t = pdisp
1107
+ g,t , Λ‡xg,t = pdisp
1108
+ g,t .
1109
+ (33)
1110
+ Since pdisp
1111
+ g,t
1112
+ ∈ [Λ‡xβˆ—
1113
+ g,t, Λ†xβˆ—
1114
+ g,t], after replacing (9d)-(9e) with (33),
1115
+ the problem P3 is still solvable and the optimal solution is
1116
+ Λ†xupdateβˆ—
1117
+ g,t
1118
+ = pdisp
1119
+ g,t , Λ‡xupdateβˆ—
1120
+ g,t
1121
+ = pdisp
1122
+ g,t , βˆ€g. With these updated
1123
+ lower and upper bounds, we update the queues Λ†Qg,t+1, Λ‡Qg,t+1,
1124
+ Λ†Zg,t+1, Λ‡Zg,t+1 according to (4), (5), (12) and (13), respectively.
1125
+ Then, we move on the solve problem P3 for time slot t + 1
1126
+ using the updated Λ†Qg,t+1, Λ‡Qg,t+1, Λ†Zg,t+1, Λ‡Zg,t+1.
1127
+ So far, we have developed a real-time feedback based online
1128
+ aggregate EV power flexibility characterization method as well
1129
+ as the EV dispatched charging power disaggregation approach.
1130
+ A completed description of the proposed method is shown in
1131
+ Algorithm 2.
1132
+ V. SIMULATION RESULTS AND DISCUSSIONS
1133
+ In this section, we evaluate the performance of the proposed
1134
+ online algorithm and compare it with other approaches.
1135
+ A. System Setup
1136
+ The time resolution is set as 10 minutes. The entire sim-
1137
+ ulation duration considered is 24 hours, i.e., 144 time slots
1138
+
1139
+ JOURNAL OF LATEX CLASS FILES, VOL. XX, NO. X, FEB. 2019
1140
+ 7
1141
+ Algorithm 2 Real-time Feedback Based Online Aggregate EV
1142
+ Power Flexibility Characterization and Disaggregation
1143
+ I. Aggregation
1144
+ 1: Aggregator classifies the arriving EVs and pushes them
1145
+ into different queues Λ†Qg, Λ‡Qg, Λ†Zg and Λ‡Zg according to
1146
+ their declared charging delay time Rg.
1147
+ 2: Solve problem P3 and obtain the aggregate EV power
1148
+ flexibility region [Λ‡xβˆ—
1149
+ g,t, Λ†xβˆ—
1150
+ g,t].
1151
+ 3: Update queues Λ†Qg,t+1, Λ‡Qg,t+1, Λ†Zg,t+1, and Λ‡Zg,t+1 ac-
1152
+ cording to (4), (5), (12) and (13), respectively.
1153
+ II. Dispatch and Disaggregation
1154
+ 4: Receive the dispatch decision from the DSO, and decom-
1155
+ pose it to each group according to (29) and (30).
1156
+ 5: Perform EV dispatched power disaggregation according to
1157
+ Algorithm 1.
1158
+ III. Real-Time Feedback and Update
1159
+ 6: Update the lower and upper bounds Λ†xupdateβˆ—
1160
+ g,t
1161
+ , Λ‡xupdateβˆ—
1162
+ g,t
1163
+ , βˆ€g
1164
+ of EV aggregate power flexibility region.
1165
+ 7: Update queues Λ†Qg,t+1, Λ‡Qg,t+1, Λ†Zg,t+1, and Λ‡Zg,t+1 ac-
1166
+ cording to (4), (5), (12) and (13), respectively.
1167
+ 8: Move to the next time slot t = t+1, and repeat the above
1168
+ steps I-III.
1169
+ each with a time interval of 10 minutes. To reflect the actual
1170
+ fluctuations in the electricity prices, we use the real-time
1171
+ electricity price data obtained from the PJM market [32]. The
1172
+ dynamic electricity price data profile is shown in Fig. 3. For
1173
+ the setting of EVs, we consider 30 EVs that are divided into
1174
+ three groups with different allowable charging time delays, i.e.,
1175
+ G = 3. Each group has 10 EVs. The EVs in the same group
1176
+ have identical allowable time delay, i.e., Rg. Particularlly,
1177
+ R1 = 8 hours, R2 = 6 hours, and R3 = 7 hours. In addition,
1178
+ for EV battery parameters, we refer to the Nissan Leaf EV
1179
+ model with a battery pack of 40 kWh and a maximum charging
1180
+ power of 6.6 kW [33]. Considering that the EV charging
1181
+ behavior is uncertain, the EVs’ arriving times are randomly
1182
+ generated. The initial battery energy level of each EV is
1183
+ selected from a uniform distribution in [0.3, 0.5] Γ— 40 kWh
1184
+ randomly [29]. We set the required state-of-charge (SOC) 1
1185
+ upon departure as 0.5 and the maximum SOC upon departure
1186
+ as 0.9. The weight parameter value of V is chosen as 1000,
1187
+ and the value of Ξ·g is set as 648, 540, and 756 for the three
1188
+ groups, respectively.
1189
+ Fig. 3. Real-time electricity price profile.
1190
+ 1The SOC of an EV is the ratio between the battery energy level and the
1191
+ battery capacity.
1192
+ B. Effectiveness of the proposed method
1193
+ We first show how the obtained aggregate EV power
1194
+ flexibility region looks like. Since the power grid dispatch
1195
+ determined by the DSO is beyond the scope of this paper,
1196
+ here the dispatch ratio Ξ±t in (29) is assumed to be randomly
1197
+ generated within the range of [0, 1] in each time slot, as shown
1198
+ in Fig. 4. Based on the generated dispatch ratio, we apply
1199
+ the proposed online flexibility characterization method and
1200
+ real-time feedback in turns (as in Algorithm 2) to obtain the
1201
+ aggregate EV power flexibility region (grey area) over time for
1202
+ each group and the charging station as a whole. The results are
1203
+ shown in Fig. 5. As seen, the power flexibility region varies
1204
+ over time. This is because EVs dynamically arrive and leave.
1205
+ Fig. 4. Randomly generated dispatch ratio Ξ±t.
1206
+ Fig. 5. The obtained aggregate EV power flexibility region.
1207
+ To validate the effectiveness of the proposed algorithm, dis-
1208
+ aggregation of the dispatched EV charging power is performed
1209
+ and we check if the SOC curves of EVs satisfy the charging
1210
+ requirements. Here, if the final EV SOC value can reach
1211
+ or exceed the EV owner’s requirement (SOC β‰₯ 0.5) upon
1212
+ leaving, then it means that the proposed method is effective.
1213
+ The left of Fig. 6 shows the actual EV charging SOC curves
1214
+ under the randomly generated dispatch ratio Ξ±t in Fig. 4. Each
1215
+ curve represents an EV. As we can see from the figure, all
1216
+ EVs’ final SOC is between 0.58 and 0.7, greater than the
1217
+ required value 0.5 and less than the maximum value 0.9. The
1218
+ right-hand side of Fig. 6 shows the number of delayed time
1219
+ slots (NDTS) to reach the requirement SOC=0.5. We can find
1220
+ that the maximum NDST is 10 for group 1, 7 for group 2,
1221
+ and 12 for group 3. All of them are within their respective
1222
+ declared allowable charging delay, i.e. Rg. This validates the
1223
+ proposed algorithm in providing maximum power flexibility
1224
+ while meeting the charging requirement.
1225
+ Furthermore, Fig. 7 shows the queue backlog evolution of
1226
+ the three groups over time. Taking group 1 for example, the
1227
+ lower and upper bound queues Λ‡Q1 and Λ†Q1 first increase be-
1228
+ cause EVs continue to arrive with their charging tasks pushing
1229
+
1230
+ JOURNAL OF LATEX CLASS FILES, VOL. XX, NO. X, FEB. 2019
1231
+ 8
1232
+ Rg=8h=48
1233
+ time slots
1234
+ Rg=6h=36
1235
+ time slots
1236
+ Rg=7h=42
1237
+ time slots
1238
+ Fig. 6.
1239
+ EV charging SOC of each group and the number of delayed time
1240
+ slots needed to reach SOC=0.5.
1241
+ into the queues. Then as time moves on, through the EV
1242
+ charging dispatch pdisp
1243
+ v,t , βˆ€v, βˆ€t determined by disaggregation,
1244
+ the EV SOC gradually increases and reaches the minimum
1245
+ charging requirement 0.5. Hence, the lower bound queue Λ‡Q1
1246
+ becomes zero because the Λ‡ag,t, βˆ€g, βˆ€t is set using the charging
1247
+ as soon as possible method to meet the minimum charging
1248
+ requirement as in (7). The upper bound queue Λ†Q1 is still larger
1249
+ than zero since the EV SOC has not reach the maximum value
1250
+ 0.9 (see Fig. 6), so there is charging flexibility. At the same
1251
+ time, since Λ†Q1 is nonnegative, the delay aware upper bound
1252
+ queue Λ†Z1 keeps growing, aiming to increase the charging
1253
+ power. The queue evolution in groups 2 and 3 can be analyzed
1254
+ similarly.
1255
+ Fig. 7. Queue backlog of each group.
1256
+ C. Performance Evaluation
1257
+ To show the advantage of the proposed online algorithm,
1258
+ two widely used benchmarks in the literature are performed.
1259
+ β€’ Benchmark 1 (B1): This is a greedy algorithm that EVs
1260
+ start charging at the maximum charging power upon
1261
+ arrival. Let us denote the arrival time as t0. When the EV
1262
+ SOC reaches the minimum charging requirement 0.5, the
1263
+ lower bound of charging power Λ‡pd,t turns to be zero (time:
1264
+ t1), and the upper bound of charging power Λ†pd,t remains
1265
+ the maximum charging power until the EV SOC reaches
1266
+ the maximum value 0.9 (time: t2). The aggregate power
1267
+ flexibility region for [t0, t1] is empty and for t ∈ [t1, t2] is
1268
+ the region between 0 and the maximum charging power.
1269
+ β€’ Benchmark 2 (B2): This is the offline method. It directly
1270
+ solves P1 to obtain the aggregate EV power flexibility
1271
+ regions over the whole time horizon by assuming known
1272
+ future information. Though not realistic, it provides a
1273
+ theoretical benchmark to verify the performance of other
1274
+ methods. But it is worth noting that since it does not
1275
+ take into account the real-time actual dispatch strategy
1276
+ when calculating the aggregate flexibility, its performance
1277
+ may be worse than the proposed real-time feedback based
1278
+ method even though it is an offline method.
1279
+ Fig. 8 shows the accumulated flexibility values (οΏ½t
1280
+ Ο„=1 FΟ„)
1281
+ under the three different methods, and TABLE I summarizes
1282
+ the total flexibility value (οΏ½T
1283
+ t=1 Ft) under different methods.
1284
+ The B1, i.e., greedy algorithm, has the worst performance and
1285
+ the lowest total flexibility value due to the myopic strategy. For
1286
+ B2, since it has complete future knowledge of EV behaviors
1287
+ and real-time electricity prices, it outperforms B1. However,
1288
+ this method is usually impossible in practice since the accurate
1289
+ future information is hardly available. Though predictions on
1290
+ future uncertainty realizations may be obtained, the potential
1291
+ prediction errors limit B2’s performance. In contrast, the
1292
+ proposed online algorithm achieves the best performance with
1293
+ the highest total power flexibility value. This is owing to the
1294
+ fact that it runs in a online manner with real-time feedback
1295
+ that allows it to utilize the most recent dispatch information
1296
+ to update its state. In addition, compared to the offline method
1297
+ B2, it does not require prior knowledge of future information
1298
+ or forecasts, which is more practical.
1299
+ Fig. 8. Accumulated flexibility value under different methods.
1300
+ TABLE I
1301
+ TOTAL FLEXIBILITY VALUE COMPARISON BETWEEN B1, B2, AND THE
1302
+ PROPOSED ALGORITHM (UNIT: USD).
1303
+ Methods
1304
+ B1
1305
+ B2
1306
+ Proposed
1307
+ Value
1308
+ 517.69
1309
+ 586
1310
+ 647.21
1311
+ Improvement
1312
+ -
1313
+ 13.2%
1314
+ 25%
1315
+ The above result is obtained under the random dispatch ratio
1316
+ Ξ±t (see Fig. 4). In fact, the dispatch ratio can affect the actual
1317
+ charging power of each EV and further affect their aggregate
1318
+ power flexibility. Therefore, it is interesting to investigate the
1319
+
1320
+ JOURNAL OF LATEX CLASS FILES, VOL. XX, NO. X, FEB. 2019
1321
+ 9
1322
+ impact of Ξ±t on the aggregate EV power flexibility (or the total
1323
+ power flexibility value (9)). Here, we use a uniform Ξ± over
1324
+ time, i.e., Ξ±t = Ξ±, βˆ€t. We change Ξ± from 0 to 1 and record the
1325
+ total power flexibility value in Fig. 9. As seen, the total power
1326
+ flexibility value depends on the dispatch ratio Ξ±. Generally,
1327
+ a larger Ξ± leads to a larger power flexibility value. However,
1328
+ this increase is nonlinear. When the dispatch ratio Ξ± exceeds
1329
+ 0.5, the total power flexibility value no longer increases. In the
1330
+ extreme case when Ξ± = 0, the total power flexibility value is
1331
+ 520 USD, which is still larger than the greedy algorithm B1.
1332
+ In addition, it can be concluded that if the average dispatch
1333
+ ratio Ξ± is greater than 0.2, then the proposed online algorithm
1334
+ more likely outperforms the offline method. This demonstrates
1335
+ the advantage of the proposed algorithm. We also present the
1336
+ aggregate EV power flexibility region under different Ξ± in Fig.
1337
+ 10. As Ξ± decreases, the aggregate EV power flexibility region
1338
+ gradually narrows. This is because under a low dispatch ratio,
1339
+ the EVs are charged at a low charging rate and more likely to
1340
+ fail to meet the charging requirement; hence, the lower bound
1341
+ of aggregate EV power flexibility region is raised to ensure
1342
+ the EVs can meet the charging requirement in the remaining
1343
+ time.
1344
+ Fig. 9. The impact of Ξ± on total power flexibility value.
1345
+ 0
1346
+ 50
1347
+ Power
1348
+ [kW]
1349
+ ub
1350
+ lb
1351
+ =0
1352
+ 0
1353
+ 50
1354
+ Power
1355
+ [kW]
1356
+ ub
1357
+ lb
1358
+ =0.1
1359
+ 0
1360
+ 50
1361
+ Power
1362
+ [kW]
1363
+ ub
1364
+ lb
1365
+ =0.2
1366
+ 0
1367
+ 50
1368
+ Power
1369
+ [kW]
1370
+ ub
1371
+ lb
1372
+ =0.3
1373
+ 0
1374
+ 50
1375
+ Power
1376
+ [kW]
1377
+ ub
1378
+ lb
1379
+ =0.4
1380
+ 0
1381
+ 20
1382
+ 40
1383
+ 60
1384
+ 80
1385
+ 100
1386
+ 120
1387
+ 140
1388
+ Time [10 min]
1389
+ 0
1390
+ 50
1391
+ Power
1392
+ [kW]
1393
+ ub
1394
+ lb
1395
+ =0.5
1396
+ 0
1397
+ 50
1398
+ Power
1399
+ [kW]
1400
+ ub
1401
+ lb
1402
+ =0
1403
+ 0
1404
+ 50
1405
+ Power
1406
+ [kW]
1407
+ ub
1408
+ lb
1409
+ =0.1
1410
+ 0
1411
+ 50
1412
+ Power
1413
+ [kW]
1414
+ ub
1415
+ lb
1416
+ =0.2
1417
+ 0
1418
+ 50
1419
+ Power
1420
+ [kW]
1421
+ ub
1422
+ lb
1423
+ =0.3
1424
+ 0
1425
+ 50
1426
+ Power
1427
+ [kW]
1428
+ ub
1429
+ lb
1430
+ =0.4
1431
+ 0
1432
+ 20
1433
+ 40
1434
+ 60
1435
+ 80
1436
+ 100
1437
+ 120
1438
+ 140
1439
+ Time [10 min]
1440
+ 0
1441
+ 50
1442
+ Power
1443
+ [kW]
1444
+ ub
1445
+ lb
1446
+ =0.5
1447
+ Fig. 10. The aggregate EV power flexibility under different Ξ±.
1448
+ D. Impact of Parameters
1449
+ According to (19), the parameter V controls the trade-
1450
+ off between stabilizing the queues and maximizing the total
1451
+ power flexibility value in the objective function (27). Here, we
1452
+ change the value of V to investigate its impact on the total
1453
+ power flexibility value. As shown in Fig. 11, the total power
1454
+ flexibility value becomes larger with an increasing V .
1455
+ Fig. 11. The impact of V on the total power flexibility value.
1456
+ Fig. 12. The impact of V on the average time delay.
1457
+ Fig. 12 depicts the impact of V on the number of time slots
1458
+ needed for EVs to meet their required battery energy level ed
1459
+ v.
1460
+ We calculate the maximum/minimum/average number of time
1461
+ slots for the EVs in each group. As seen, with the growth
1462
+ of V , the number of delayed time slots slightly increases.
1463
+ This is because a larger V means putting more emphasis on
1464
+ maximizing the total power flexibility, which may result in a
1465
+ reduced lower bound Λ‡xg,tof the aggregate EV power flexibility
1466
+ region. Consequently, the charging time needed to reach the
1467
+ required energy level becomes longer. Comparing the three
1468
+ groups, we can find that the time delays of groups 1 and 3 are
1469
+ generally longer than that of group 2, which is owing to the
1470
+ shorter allowable time delay Rg.
1471
+ Fig. 13 shows the impact of Ξ·g on the total flexibility value
1472
+ and the number of time slots needed for EVs to meet their
1473
+ required battery energy level ed
1474
+ v. We can find that a larger ηg
1475
+ results in a lower total power flexibility value and less number
1476
+ of delayed time slots. This is because a larger Ξ·g forces the
1477
+ virtual delay-aware queue Λ‡Zg to grow rapidly, allowing EVs
1478
+ to get charged quickly. Meanwhile, the power flexibility is
1479
+ sacrificed.
1480
+ VI. CONCLUSION
1481
+ With the proliferation of EVs, it is necessary to better
1482
+ utilize their charging power flexibility, making them valuable
1483
+ resources rather than burdens on the power grid. This paper
1484
+ proposes a real-time feedback based online aggregate EV
1485
+ power flexibility characterization method. It can output the
1486
+ aggregate flexibility region for each time slot in an online
1487
+ manner, with a total flexibility value over time similar to the
1488
+ offline counterpart. We prove that by choosing an aggregate
1489
+
1490
+ JOURNAL OF LATEX CLASS FILES, VOL. XX, NO. X, FEB. 2019
1491
+ 10
1492
+ Fig. 13.
1493
+ The impact of Ξ·g on the total flexibility value and average time
1494
+ delay.
1495
+ dispatch strategy within the obtained flexibility region for
1496
+ each time slot, the corresponding disaggregated EV control
1497
+ strategies allow all EVs to satisfy their charging requirements.
1498
+ Simulations demonstrate the effectiveness and benefits of the
1499
+ proposed method. It is worth noting that the proposed method
1500
+ can even outperform the offline method in some cases since
1501
+ it can utilize up-to-date dispatch information via real-time
1502
+ feedback. Future research may take into account the conflicting
1503
+ interests between the operator, aggregator, and EVs when
1504
+ deriving the flexibility region.
1505
+ REFERENCES
1506
+ [1] S. Wang, S. Bi, Y.-J. A. Zhang, and J. Huang, β€œElectrical vehicle
1507
+ charging station profit maximization: Admission, pricing, and online
1508
+ scheduling,” IEEE Trans. Sustain. Energy, vol. 9, no. 4, pp. 1722–1731,
1509
+ 2018.
1510
+ [2] A. A. Munshi and Y. A.-R. I. Mohamed, β€œExtracting and defining
1511
+ flexibility of residential electrical vehicle charging loads,” IEEE Trans.
1512
+ Ind. Informat., vol. 14, no. 2, pp. 448–461, 2018.
1513
+ [3] L. Gan, U. Topcu, and S. H. Low, β€œOptimal decentralized protocol for
1514
+ electric vehicle charging,” IEEE Trans. Power Syst., vol. 28, no. 2, pp.
1515
+ 940–951, 2012.
1516
+ [4] G. C. Mouli, P. Bauer, and M. Zeman, β€œSystem design for a solar
1517
+ powered electric vehicle charging station for workplaces,” Appl. Energy,
1518
+ vol. 168, pp. 434–443, 2016.
1519
+ [5] Y. Yang, Q.-S. Jia, G. Deconinck, X. Guan, Z. Qiu, and Z. Hu,
1520
+ β€œDistributed coordination of EV charging with renewable energy in a
1521
+ microgrid of buildings,” IEEE Trans. Smart Grid, vol. 9, no. 6, pp.
1522
+ 6253–6264, 2018.
1523
+ [6] X. Wu, X. Hu, Y. Teng, S. Qian, and R. Cheng, β€œOptimal integration
1524
+ of a hybrid solar-battery power source into smart home nanogrid with
1525
+ plug-in electric vehicle,” J. Power Sources, vol. 363, pp. 277–283, 2017.
1526
+ [7] Y. Zheng, Y. Song, D. J. Hill, and K. Meng, β€œOnline distributed
1527
+ MPC-based optimal scheduling for EV charging stations in distribution
1528
+ systems,” IEEE Trans. Ind. Informat., vol. 15, no. 2, pp. 638–649, Feb.
1529
+ 2019.
1530
+ [8] X. Wu, X. Hu, X. Yin, and S. J. Moura, β€œStochastic optimal energy
1531
+ management of smart home with pev energy storage,” IEEE Trans. Smart
1532
+ Grid, vol. 9, no. 3, pp. 2065–2075, 2018.
1533
+ [9] F. Jiao, Y. Zou, X. Zhang, and B. Zhang, β€œOnline optimal dispatch
1534
+ based on combined robust and stochastic model predictive control for a
1535
+ microgrid including EV charging station,” Energy, vol. 247, p. 123220,
1536
+ 2022.
1537
+ [10] Q. Yan, B. Zhang, and M. Kezunovic, β€œOptimized operational cost
1538
+ reduction for an EV charging station integrated with battery energy
1539
+ storage and PV generation,” IEEE Trans. Smart Grid, vol. 10, no. 2,
1540
+ pp. 2096–2106, 2018.
1541
+ [11] T. Zhao, Y. Li, X. Pan, P. Wang, and J. Zhang, β€œReal-time optimal
1542
+ energy and reserve management of electric vehicle fast charging station:
1543
+ Hierarchical game approach,” IEEE Trans. Smart Grid, vol. 9, no. 5, pp.
1544
+ 5357–5370, Sep. 2018.
1545
+ [12] X. Dong, Y. Mu, X. Xu, H. Jia, J. Wu, X. Yu, and Y. Qi, β€œA charging
1546
+ pricing strategy of electric vehicle fast charging stations for the voltage
1547
+ control of electricity distribution networks,” Appl. Energy, vol. 225, pp.
1548
+ 857–868, 2018.
1549
+ [13] Z. Xu, D. S. Callaway, Z. Hu, and Y. Song, β€œHierarchical coordination
1550
+ of heterogeneous flexible loads,” IEEE Trans. Power Syst., vol. 31, no. 6,
1551
+ pp. 4206–4216, 2016.
1552
+ [14] H. Zhang, Z. Hu, Z. Xu, and Y. Song, β€œEvaluation of achievable vehicle-
1553
+ to-grid capacity using aggregate PEV model,” IEEE Trans. Power Syst.,
1554
+ vol. 32, no. 1, pp. 784–794, 2016.
1555
+ [15] L. Wang, J. Kwon, N. Schulz, and Z. Zhou, β€œEvaluation of aggregated
1556
+ EV flexibility with tso-dso coordination,” IEEE Trans. Sustain. Energy,
1557
+ vol. 13, no. 4, pp. 2304–2315, 2022.
1558
+ [16] X. Shi, Y. Xu, Q. Guo, and H. Sun, β€œOptimal dispatch based on aggre-
1559
+ gated operation region of EV considering spatio-temporal distribution,”
1560
+ IEEE Trans. Sustain. Energy, vol. 13, no. 2, pp. 715–731, 2021.
1561
+ [17] M. Zhou, Z. Wu, J. Wang, and G. Li, β€œForming dispatchable region
1562
+ of electric vehicle aggregation in microgrid bidding,” IEEE Trans. Ind.
1563
+ Informat., vol. 17, no. 7, pp. 4755–4765, 2021.
1564
+ [18] G. Chen, H. Zhang, H. Hui, N. Dai, and Y. Song, β€œScheduling ther-
1565
+ mostatically controlled loads to provide regulation capacity based on a
1566
+ learning-based optimal power flow model,” IEEE Trans. Sustain. Energy,
1567
+ vol. 12, no. 4, pp. 2459–2470, 2021.
1568
+ [19] Z. Yi, Y. Xu, W. Gu, L. Yang, and H. Sun, β€œAggregate operation model
1569
+ for numerous small-capacity distributed energy resources considering
1570
+ uncertainty,” IEEE Trans. Smart Grid, vol. 12, no. 5, pp. 4208–4224,
1571
+ 2021.
1572
+ [20] S. Wang and W. Wu, β€œAggregate flexibility of virtual power plants with
1573
+ temporal coupling constraints,” IEEE Trans. Smart Grid, vol. 12, no. 6,
1574
+ pp. 5043–5051, 2021.
1575
+ [21] L. Zhao, W. Zhang, H. Hao, and K. Kalsi, β€œA geometric approach
1576
+ to aggregate flexibility modeling of thermostatically controlled loads,”
1577
+ IEEE Trans. Power Syst., vol. 32, no. 6, pp. 4721–4731, 2017.
1578
+ [22] X. Chen, E. Dall’Anese, C. Zhao, and N. Li, β€œAggregate power flexibility
1579
+ in unbalanced distribution systems,” IEEE Trans. Smart Grid, vol. 11,
1580
+ no. 1, pp. 258–269, 2020.
1581
+ [23] Z. Guo, P. Pinson, Q. Wu, S. Chen, Q. Yang, and Z. Yang, β€œAn
1582
+ asynchronous online negotiation mechanism for real-time peer-to-peer
1583
+ electricity markets,” IEEE Trans. Power Syst., pp. 1–13, 2021 early
1584
+ access.
1585
+ [24] S. Fan, G. He, X. Zhou, and M. Cui, β€œOnline optimization for net-
1586
+ worked distributed energy resources with time-coupling constraints,”
1587
+ IEEE Trans. Smart Grid, vol. 12, no. 1, pp. 251–267, 2020.
1588
+ [25] W. Shi, N. Li, C. Chu, and R. Gadh, β€œReal-time energy management in
1589
+ microgrids,” IEEE Trans. Smart Grid, vol. 8, no. 1, pp. 228–238, Jan.
1590
+ 2017.
1591
+ [26] W. Zhong, K. Xie, Y. Liu, C. Yang, S. Xie, and Y. Zhang, β€œOnline
1592
+ control and near-optimal algorithm for distributed energy storage sharing
1593
+ in smart grid,” IEEE Trans. Smart Grid, vol. 11, no. 3, pp. 2552–2562,
1594
+ 2020.
1595
+ [27] L. Yu, T. Jiang, and Y. Zou, β€œDistributed real-time energy management
1596
+ in data center microgrids,” IEEE Trans. Smart Grid, vol. 9, no. 4, pp.
1597
+ 3748–3762, 2018.
1598
+ [28] Z. Shen, C. Wu, L. Wang, and G. Zhang, β€œReal-time energy management
1599
+ for microgrid with EV station and CHP generation,” IEEE Trans.
1600
+ Network Science Engineering, vol. 8, no. 2, pp. 1492–1501, 2021.
1601
+ [29] C. Jin, X. Sheng, and P. Ghosh, β€œOptimized electric vehicle charging
1602
+ with intermittent renewable energy sources,” IEEE J. Sel. Top. Signal
1603
+ Proc.., vol. 8, no. 6, pp. 1063–1072, 2014.
1604
+ [30] Y. Zhou, D. K. Yau, P. You, and P. Cheng, β€œOptimal-cost scheduling of
1605
+ electrical vehicle charging under uncertainty,” IEEE Trans. Smart Grid,
1606
+ vol. 9, no. 5, pp. 4547–4554, 2018.
1607
+ [31] Y. Chen and W. Wei, β€œRobust generation dispatch with strategic re-
1608
+ newable power curtailment and decision-dependent uncertainty,” IEEE
1609
+ Trans. Power Syst., 2022.
1610
+ [32] PJM-Data Miner 2, β€œReal-time five minute LMPs,” 2022, accessed Mar.,
1611
+ 2022. [Online]. Available: https://dataminer2.pjm.com/.
1612
+ [33] L. Affolabi, M. Shahidehpour, W. Gan, M. Yan, B. Chen, S. Pandey,
1613
+ A. Vukojevic, E. A. Paaso, A. Alabdulwahab, and A. Abusorrah,
1614
+ β€œOptimal transactive energy trading of electric vehicle charging stations
1615
+ with on-site PV generation in constrained power distribution networks,”
1616
+ IEEE Trans. Smart Grid, vol. 13, no. 2, pp. 1427–1440, 2022.
1617
+ APPENDIX A
1618
+ PROOF OF PROPOSITION 1
1619
+ Let {pd,t, βˆ€t} be the aggregate power trajectory. For each
1620
+ time slot t ∈ T , since pd,t ∈ [Λ‡pβˆ—
1621
+ d,t, Λ†pβˆ—
1622
+ d,t], we can define an
1623
+
1624
+ JOURNAL OF LATEX CLASS FILES, VOL. XX, NO. X, FEB. 2019
1625
+ 11
1626
+ auxiliary coefficient:
1627
+ Ξ²t :=
1628
+ Λ†pβˆ—
1629
+ d,t βˆ’ pd,t
1630
+ Λ†pβˆ—
1631
+ d,t βˆ’ Λ‡pβˆ—
1632
+ d,t
1633
+ ∈ [0, 1]
1634
+ (A.1)
1635
+ so that pd,t = Ξ²tΛ‡pβˆ—
1636
+ d,t + (1 βˆ’ Ξ²t)Λ†pβˆ—
1637
+ d,t. Then, we can construct a
1638
+ feasible EV dispatch strategy by letting
1639
+ pc
1640
+ v,t = Ξ²tΛ‡pcβˆ—
1641
+ v,t + (1 βˆ’ Ξ²t)Λ†pcβˆ—
1642
+ v,t,
1643
+ (A.2a)
1644
+ ev,t = Ξ²tΛ‡ecβˆ—
1645
+ v,t + (1 βˆ’ Ξ²t)Λ†ecβˆ—
1646
+ v,t.
1647
+ (A.2b)
1648
+ for all time slots t ∈ T .
1649
+ We prove that it is a feasible EV dispatch strategy as
1650
+ follows,
1651
+ pd,t = Ξ²tΛ‡pβˆ—
1652
+ d,t + (1 βˆ’ Ξ²t)Λ†pβˆ—
1653
+ d,t
1654
+ = Ξ²t
1655
+ οΏ½
1656
+ v∈V
1657
+ Λ‡pcβˆ—
1658
+ v,t + (1 βˆ’ Ξ²t)
1659
+ οΏ½
1660
+ v∈V
1661
+ Λ†pcβˆ—
1662
+ v,t
1663
+ =
1664
+ οΏ½
1665
+ v∈V
1666
+ οΏ½
1667
+ Ξ²tΛ‡pcβˆ—
1668
+ v,t + (1 βˆ’ Ξ²t)Λ†pcβˆ—
1669
+ v,t
1670
+ οΏ½
1671
+ =
1672
+ οΏ½
1673
+ v∈V
1674
+ pc
1675
+ v,t
1676
+ (A.3)
1677
+ Hence, constraint (1c) holds for pd,t and pc
1678
+ v,t, βˆ€v. Similarly,
1679
+ we can prove that constraints (1d)-(1g) are met. Therefore,
1680
+ we have constructed a feasible EV dispatch strategy, which
1681
+ completes the proof.
1682
+ β– 
1683
+ APPENDIX B
1684
+ PROOF OF PROPOSITION 2
1685
+ Here, we use the contradiction. If a charging request Λ†ag,t
1686
+ arrives in time slot t cannot be fulfilled on or before time
1687
+ slot t + Λ†Ξ΄g,max. Then, queue Λ†Qg,Ο„ > 0always holds for Ο„ ∈
1688
+ [t + 1, ..., t + Λ†Ξ΄g,max]. Thus, we have I Λ†
1689
+ Qg,t>0 = 1. According
1690
+ to delay virtual queue dynamics (12), for all Ο„ ∈ [t+1, ..., t+
1691
+ Λ†Ξ΄g,max], we have
1692
+ Λ†Zg,Ο„+1 β‰₯ Λ†Zg,Ο„ + Ξ·g
1693
+ Rg
1694
+ βˆ’ Λ†xg,Ο„, βˆ€g, βˆ€t.
1695
+ (B.1)
1696
+ By summing the above inequalities over Ο„ ∈ [t + 1, ..., t +
1697
+ Λ†Ξ΄g,max], we have
1698
+ Λ†Zg,t+Λ†Ξ΄g,max+1 βˆ’ Λ†Zg,t+1 β‰₯ Ξ·g
1699
+ Rg
1700
+ Λ†Ξ΄g,max +
1701
+ t+Λ†Ξ΄g,max
1702
+ οΏ½
1703
+ Ο„=t+1
1704
+ (βˆ’Λ†xg,Ο„).
1705
+ (B.2)
1706
+ Since Λ†Zg,t+Λ†Ξ΄g,max+1 ≀ Λ†Zg,max and Λ†Zg,t+1 β‰₯ 0, we have
1707
+ Λ†Zg,max β‰₯ Ξ·g
1708
+ Rg
1709
+ Λ†Ξ΄g,max +
1710
+ t+Λ†Ξ΄g,max
1711
+ οΏ½
1712
+ Ο„=t+1
1713
+ (βˆ’Λ†xg,Ο„).
1714
+ (B.3)
1715
+ Since the charging tasks are processed in a first-in-first-out
1716
+ manner, and the charging request is not fulfilled by t+Λ†Ξ΄g,max,
1717
+ we have
1718
+ t+Λ†Ξ΄g,max
1719
+ οΏ½
1720
+ Ο„=t+1
1721
+ (Λ†xg,Ο„) < Λ†Qg,max
1722
+ (B.4)
1723
+ Combining the above two inequalities, we obtain
1724
+ Λ†Zg,max > Ξ·g
1725
+ Rg
1726
+ Λ†Ξ΄g,max βˆ’ Λ†Qg,max,
1727
+ (B.5)
1728
+ which implies
1729
+ Λ†Ξ΄g,max < ( Λ†Qg,max + Λ†Zg,max)Rg
1730
+ Ξ·g
1731
+ .
1732
+ (B.6)
1733
+ However, this result contradicts the definition of Λ†Ξ΄g,max in
1734
+ (14). Therefore, the worst case delay should be less than or
1735
+ equal to Λ†Ξ΄g,max as defined in (14).
1736
+ The proof of (15) follows a similar procedure, and we omit
1737
+ it here for brevity.
1738
+ β– 
1739
+ APPENDIX C
1740
+ PROOF OF PROPOSITION 3
1741
+ Denote the solution of P3 by the proposed algorithm by
1742
+ Λ†xpro
1743
+ g,t and Λ‡xpro
1744
+ g,t , and the optimal solution of P1 by Λ†xβˆ—
1745
+ g,t and
1746
+ Λ‡xβˆ—
1747
+ g,t. According to (26), we have
1748
+ βˆ†(Θt) + V E[βˆ’F pro
1749
+ t
1750
+ |Θt]
1751
+ ≀ A + V E[βˆ’F pro
1752
+ t
1753
+ |Θt] +
1754
+ οΏ½
1755
+ g∈G
1756
+ Λ†Qg,tE
1757
+ οΏ½
1758
+ Λ†ag,t βˆ’ Λ†xpro
1759
+ g,t |Θt
1760
+ οΏ½
1761
+ +
1762
+ οΏ½
1763
+ g∈G
1764
+ Λ‡Qg,tE
1765
+ οΏ½
1766
+ Λ‡ag,t βˆ’ Λ‡xpro
1767
+ g,t |Θt
1768
+ οΏ½
1769
+ +
1770
+ οΏ½
1771
+ g∈G
1772
+ Λ†Zg,tE
1773
+ οΏ½
1774
+ βˆ’Λ†xpro
1775
+ g,t |Θt
1776
+ οΏ½
1777
+ +
1778
+ οΏ½
1779
+ g∈G
1780
+ Λ‡Zg,tE
1781
+ οΏ½
1782
+ βˆ’Λ‡xpro
1783
+ g,t |Θt
1784
+ οΏ½
1785
+ ,
1786
+ ≀ A + V E[βˆ’F βˆ—
1787
+ t |Θt] +
1788
+ οΏ½
1789
+ g∈G
1790
+ Λ†Qg,tE
1791
+ οΏ½
1792
+ Λ†ag,t βˆ’ Λ†xβˆ—
1793
+ g,t|Θt
1794
+ οΏ½
1795
+ +
1796
+ οΏ½
1797
+ g∈G
1798
+ Λ‡Qg,tE
1799
+ οΏ½
1800
+ Λ‡ag,t βˆ’ Λ‡xβˆ—
1801
+ g,t|Θt
1802
+ οΏ½
1803
+ +
1804
+ οΏ½
1805
+ g∈G
1806
+ Λ†Zg,tE
1807
+ οΏ½
1808
+ βˆ’Λ†xβˆ—
1809
+ g,t|Θt
1810
+ οΏ½
1811
+ +
1812
+ οΏ½
1813
+ g∈G
1814
+ Λ‡Zg,tE
1815
+ οΏ½
1816
+ βˆ’Λ‡xβˆ—
1817
+ g,t|Θt
1818
+ οΏ½
1819
+ ,
1820
+ ≀ A + V E[βˆ’F βˆ—
1821
+ t |Θt]
1822
+ (C.1)
1823
+ The result is based on the fact that
1824
+ lim
1825
+ T β†’βˆž
1826
+ 1
1827
+ T
1828
+ T
1829
+ οΏ½
1830
+ t=1
1831
+ E [Λ†ag,t βˆ’ Λ†xg,t|Θt] ≀ 0
1832
+ (C.2)
1833
+ lim
1834
+ T β†’βˆž
1835
+ 1
1836
+ T
1837
+ T
1838
+ οΏ½
1839
+ t=1
1840
+ E [Λ‡ag,t βˆ’ Λ‡xg,t|Θt] ≀ 0
1841
+ (C.3)
1842
+ lim
1843
+ T β†’βˆž
1844
+ 1
1845
+ T
1846
+ T
1847
+ οΏ½
1848
+ t=1
1849
+ E [βˆ’Λ†xg,t|Θt] ≀ 0
1850
+ (C.4)
1851
+ lim
1852
+ T β†’βˆž
1853
+ 1
1854
+ T
1855
+ T
1856
+ οΏ½
1857
+ t=1
1858
+ E [βˆ’Λ‡xg,t|Θt] ≀ 0
1859
+ (C.5)
1860
+ which is due to constraints (9b)-(9e).
1861
+ By summing the above inequality (C.1) over time slots t ∈
1862
+ {1, 2, . . . , T}, we have
1863
+ T
1864
+ οΏ½
1865
+ t=1
1866
+ V E[βˆ’F pro
1867
+ t
1868
+ ]
1869
+ ≀ AT + V
1870
+ T
1871
+ οΏ½
1872
+ t=1
1873
+ E[βˆ’F βˆ—
1874
+ t ] βˆ’ E[L(ΘT +1)] + E[L(Θ1)].
1875
+
1876
+ JOURNAL OF LATEX CLASS FILES, VOL. XX, NO. X, FEB. 2019
1877
+ 12
1878
+ Based on the fact that L(ΘT +1) and L(Θ1) are finite, we
1879
+ divide both sides of the above inequalities by V T and let
1880
+ T β†’ ∞, then we have
1881
+ lim
1882
+ T β†’βˆž
1883
+ 1
1884
+ T
1885
+ T
1886
+ οΏ½
1887
+ t=1
1888
+ E(βˆ’F pro
1889
+ t
1890
+ ) ≀ A
1891
+ V + lim
1892
+ T β†’βˆž
1893
+ 1
1894
+ T
1895
+ T
1896
+ οΏ½
1897
+ t=1
1898
+ E(βˆ’F βˆ—
1899
+ t ),
1900
+ which completes the proof.
1901
+ β– 
1902
+
EdE1T4oBgHgl3EQfqQXv/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
F9AyT4oBgHgl3EQfSveK/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:455ab431799dfef75e2510aaad0f970aca1e62798fea6003d4984970d4f48172
3
+ size 8192045
FtE0T4oBgHgl3EQfRAB3/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9cb41541f0cfe50b2c3afa1365baf692c7be7552d74823b84bb981367c50f1a9
3
+ size 2555949
H9FKT4oBgHgl3EQfdS4O/content/2301.11819v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9539256f7f7a32ce2d8ff16c6efa43cd05f1ff848d0ff0ed6bb1a58f9ed4ec1b
3
+ size 824757
H9FKT4oBgHgl3EQfdS4O/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:578d3e5d67efeabc2bea4ea48e7a1ec9e5d4f892b7b6ba60cc733d4bb9b3b68c
3
+ size 183584
HtE1T4oBgHgl3EQfXwQA/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:007dc85cf297d43f4190ada9788c94c12741592609aa5809053992a6d52c6d56
3
+ size 3997741
HtE1T4oBgHgl3EQfXwQA/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5620bfc446f34e043cba9f6d0199e13c1e460ec22b1d0c875225086ebb0cf0ca
3
+ size 169492
IdAyT4oBgHgl3EQf5vqw/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:150985bfb49eb80c77fabd8f436dcd9c726900d6e10afc940a95fbb7b9800738
3
+ size 3735597
JNE3T4oBgHgl3EQfXQpr/content/tmp_files/2301.04477v1.pdf.txt ADDED
@@ -0,0 +1,1477 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ arXiv:2301.04477v1 [gr-qc] 11 Jan 2023
2
+ Reconstruction Methods and the Amplification of the Inflationary Spectrum
3
+ Leonardo Chataignier,βˆ— Alexander Yu. Kamenshchik,† Alessandro Tronconi,‑ and Giovanni VenturiΒ§
4
+ Dipartimento di Fisica e Astronomia, UniversitΓ  di Bologna, via Irnerio 46, 40126 Bologna, Italy
5
+ I.N.F.N., Sezione di Bologna, I.S. FLAG, viale B. Pichat 6/2, 40127 Bologna, Italy
6
+ We analyze the consequences of different evolutions of the Hubble parameter on the spectrum of
7
+ scalar inflationary perturbations. The analysis is restricted to inflationary phases described by a
8
+ transient evolution, when uncommon features arise in the inflationary spectra which may lead to
9
+ an amplitude enhancement. We then discuss how the spectrum is, respectively, amplified or blue-
10
+ tilted in the presence or absence of a growing solution of the Mukhanov-Sasaki equation. The cases
11
+ of general relativity with a minimally coupled inflaton and that of induced gravity are considered
12
+ explicitly. Finally, some remarks on constant roll inflation are discussed.
13
+ I.
14
+ INTRODUCTION
15
+ The possibility that a large amount of the Dark Matter (DM) content in our Universe is made of (primordial)
16
+ black holes (PBHs) has been seriously considered in the last few years [1]. This idea seems compelling because it
17
+ could improve our understanding of cosmological evolution and, in particular, of inflation [2]. Moreover, the PBH
18
+ hypothesis is also intriguing due to the increasing amount of direct and indirect observations of black holes (BHs)
19
+ out of the astrophysical range, as well as the current lack of evidence for particle models of DM that go beyond the
20
+ Standard Model of particle physics.
21
+ According to the present observational bounds [3], it is possible that even the whole DM content of the Universe
22
+ today is comprised of PBHs originated from the collapse of matter overdensities in a certain wavelength interval
23
+ of inflationary perturbations. In this scenario, the abundance of PBHs is related to the amplitude of the inflaton
24
+ fluctuations, the enhancement of which must be by several orders of magnitude with respect to (w.r.t.) the amplitude
25
+ probed by Cosmic Microwave Background (CMB) radiation. Nonetheless, the microscopic physics that originate such
26
+ a mechanism of amplification is still debated. For example, the amplification needed can be generated by a phase of
27
+ ultra slow roll (USR) inflation in the presence of an inflection point of the inflaton potential [4]. This USR phase
28
+ is the consequence of a transient period of inflatonary evolution, when slow-roll conditions are violated, and the
29
+ inflaton then relaxes towards a de Sitter attractor. In contrast to the case of the fluctuations imprinted in the CMB,
30
+ the perturbations [5] do not freeze at horizon exit in this case, as a growing solution of the Mukhanov-Sasaki (MS)
31
+ equation is present, and it is responsible for the amplification of the modes. Other possibilities have been considered
32
+ in the literature, such as an inflationary model able to generate a blue-tilted spectrum without the presence of the
33
+ growing solution [6].
34
+ In this article, these two mechanisms of amplification are considered. Instead of analyzing the possible consequences
35
+ of different inflationary models obtained by varying the form of the inflaton-gravity action, we shall here consider
36
+ different evolutions of the Hubble parameter and correspondingly obtain the inflaton action. Within this approach,
37
+ even if the inflaton potential cannot be exactly reconstructed, the features of the resulting spectra can still be
38
+ calculated, and one may verify whether their amplitude is amplified. For simplicity, our starting point is the case
39
+ of a minimally coupled inflaton, then some non-minimally coupled models are also investigated. Moreover, different
40
+ techniques for the reconstruction are adopted.
41
+ The article is organised as follows. In Section II, we review the general formalism of the dynamics of the inflationary
42
+ perturbations adopting a slightly unconventional formalism, and we derive the conditions for the existence of a growing
43
+ solution in the MS equation or simply a blue-tilted spectrum in the absence of this solution. Furthermore, a useful
44
+ relation between the odd and even slow-roll (SR) parameters in a certain hierarchy is obtained. This relation is valid
45
+ for transient periods described by a certain time evolution, and it will be employed across the entire article. In Section
46
+ III, different models are analysed, and the procedure for reconstruction is illustrated. In Section IV, the application
47
+ of the formalism to constant roll inflation is studied. Finally, the conclusions are drawn in Section V.
48
+ βˆ— leonardo.chataignier@unibo.it
49
+ † kamenshchik@bo.infn.it
50
+ ‑ tronconi@bo.infn.it
51
+ Β§ giovanni.venturi@bo.infn.it
52
+
53
+ 2
54
+ II.
55
+ INFLATIONARY PERTURBATIONS
56
+ Let us first review the formalism of the inflationary perturbations. On adopting a slightly unconventional approach,
57
+ we find the conditions which must hold in order to have an amplification of the inflationary spectrum either as the
58
+ wavenumber k grows or as time evolves. In a realistic inflationary scenario, wherein amplification starts at some
59
+ given time, both mechanisms essentially lead to an enhancement of the shortest wavelength part of the spectrum
60
+ (k > kCMB).
61
+ The conditions are then expressed in a model-independent form, which is valid provided the SR
62
+ parameters are β€œconstant”, and we use it in what follows to discuss different scenarios.
63
+ In general, after some manipulations, the Mukhanov-Sasaki (MS) equation takes the following form
64
+ vβ€²β€²
65
+ k +
66
+ οΏ½
67
+ k2 βˆ’ zβ€²β€²
68
+ z
69
+ οΏ½
70
+ vk = 0 ,
71
+ (1)
72
+ where the prime denotes the derivative w.r.t. conformal time Ξ· and z is a time-dependent function that depends on
73
+ the specific model of inflation. For example, in the case of general relativity (GR) with a minimally coupled inflaton,
74
+ one has z = a√ǫ1, which leads to (see e.g. [7])
75
+ zβ€²β€²
76
+ z = a2H2
77
+ οΏ½
78
+ 2 βˆ’ Η«1 + Η«2
79
+ οΏ½3
80
+ 2 + Η«2
81
+ 4 βˆ’ Η«1
82
+ 2 + Η«3
83
+ 2
84
+ οΏ½οΏ½
85
+ ≑ a2H2fMS(Η«i) ,
86
+ (2)
87
+ with Η«1 = βˆ’ Λ™H/H2, Η«i+1 = Η«βˆ’1
88
+ i dΗ«i/dN for i > 0 and N = ln a. The infinite set of Η«i’s form the so-called hierarchy of
89
+ β€œHubble flow functions” of SR parameters. It is important to note that, depending on the model of inflation, other
90
+ hierarchies are commonly used, and they are associated with the evolution of different (homogeneous) degrees of
91
+ freedom.
92
+ In general, one has
93
+ zβ€²β€²
94
+ z ≑ a2H2fMS ,
95
+ (3)
96
+ where fMS is a dimensionless quantity that takes a different form depending on the inflationary model. It can then
97
+ be expressed as a function of the SR parameters Η«i’s.
98
+ It is now convenient to define the new independent variable ΞΎ = k/(aH), where dΞΎ/dΞ· = βˆ’aH(1 βˆ’ Η«1)ΞΎ < 0 during
99
+ inflation. Due to
100
+ d
101
+ dΞ· = βˆ’aH(1 βˆ’ Η«1)ΞΎ d
102
+ dΞΎ
103
+ (4)
104
+ and
105
+ d2
106
+ dΞ·2 = a2H2(1 βˆ’ Η«1)2
107
+ οΏ½
108
+ ΞΎ2 d2
109
+ dΞΎ2 +
110
+ Η«1Η«2
111
+ (1 βˆ’ Η«1)2 ΞΎ d
112
+ dΞΎ
113
+ οΏ½
114
+ ,
115
+ (5)
116
+ we are led to
117
+ οΏ½
118
+ ΞΎ2 d2vk
119
+ dΞΎ2 +
120
+ Η«1Η«2
121
+ (1 βˆ’ Η«1)2 ΞΎ dvk
122
+ dΞΎ
123
+ οΏ½
124
+ + ΞΎ2 βˆ’ fMS(Η«i)
125
+ (1 βˆ’ Η«1)2
126
+ vk = 0 .
127
+ (6)
128
+ On rewriting the MS equation in terms of ΞΎ one eliminates its explicit dependence on aH.
129
+ In the regime where the SR parameters are constant, and in the long wavelength limit (ΞΎ β†’ 0) Eq. (6) can be
130
+ algebraically solved and the features of the primordial spectra can be derived in a straightforward manner. Indeed,
131
+ in this limit, the two independent solutions of Eq. (6) have the form vk = ξα, where α satisfies the algebraic equation
132
+ Ξ±2 +
133
+ οΏ½
134
+ Η«1Η«2
135
+ (1 βˆ’ Η«1)2 βˆ’ 1
136
+ οΏ½
137
+ Ξ± βˆ’ fMS(Η«i)
138
+ (1 βˆ’ Η«1)2 = 0 ,
139
+ (7)
140
+ with
141
+ Ξ±1,2 =
142
+ βˆ’
143
+ οΏ½
144
+ Η«1Η«2
145
+ (1βˆ’Η«1)2 βˆ’ 1
146
+ οΏ½
147
+ Β±
148
+ οΏ½οΏ½
149
+ Η«1Η«2
150
+ (1βˆ’Η«1)2 βˆ’ 1
151
+ οΏ½2
152
+ + 4 fMS(Η«i)
153
+ (1βˆ’Η«1)2
154
+ 2
155
+ .
156
+ (8)
157
+
158
+ 3
159
+ For instance, when fMS is defined by Eq. (2), and in the pure de Sitter case (ǫi = 0), we obtain
160
+ Ξ±1,2 = 1 Β± 3
161
+ 2
162
+ .
163
+ (9)
164
+ For this case, the positive solution, Ξ±1 = 2, decreases in time, while the negative solution, Ξ±2 = βˆ’1, increases, and it
165
+ remains nontrivial in the limit ΞΎ β†’ 0, which leads to
166
+ vk,dS ∼ kβˆ’1/2
167
+ οΏ½ k
168
+ aH
169
+ οΏ½βˆ’1
170
+ , Rk,dS ∼ kβˆ’3/2 aH
171
+ z
172
+ = kβˆ’3/2H ,
173
+ (10)
174
+ where Rk ≑ vk/z is the curvature perturbation (z = a in the de Sitter case), and the prefactor kβˆ’1/2 is essentially
175
+ fixed by the initial (Bunch-Davies) conditions. The quantity Rk is independent of time, and the spectral index can
176
+ be straightforwardly computed to be
177
+ ns βˆ’ 1 = d ln βˆ†2
178
+ s
179
+ d ln k
180
+ ,
181
+ (11)
182
+ with βˆ†2
183
+ s ≑ |Rk,dS|2k3/(2Ο€2), which leads to the well known de Sitter result (ns βˆ’ 1)dS = 0.
184
+ In the SR case (|Η«i| β‰ͺ 1), the SR parameters can be approximated by constants and the expressions (8) are still
185
+ valid but must be expanded to first order for consistency. One then obtains
186
+ α1,2 = 1 ± √9 + 12ǫ1 + 6ǫ2
187
+ 2
188
+ ≃ 1 Β± (3 + 2Η«1 + Η«2)
189
+ 2
190
+ ,
191
+ (12)
192
+ which implies (ns βˆ’ 1)SR = βˆ’2Η«1 βˆ’ Η«2.
193
+ We note that there is a caveat one must take into account for USR. In this case, one finds the same solutions for
194
+ the α’s as the de Sitter case, but the definition of the curvature perturbations is different, since zUSR ∝ a√ǫ1 β†’ 0.
195
+ Then, the amplitude of primordial curvature perturbations depends on time and is amplified. In the USR case, the
196
+ spectral index cannot be calculated analytically with the same procedure as illustrated for de Sitter and SR.
197
+ One can better illustrate the differences among the three cases just mentioned by solving the equation for Rk,
198
+ Rβ€²β€²
199
+ k + 2zβ€²
200
+ z Rk + k2Rk = 0 .
201
+ (13)
202
+ In terms of ΞΎ, Eq. (13) can be conveniently rewritten as
203
+ ΞΎ2 d2Rk
204
+ dΞΎ2
205
+ +
206
+ οΏ½
207
+ Η«1Η«2 βˆ’ 2 (1 βˆ’ Η«1) d ln z
208
+ dN
209
+ (1 βˆ’ Η«1)2
210
+ οΏ½
211
+ ΞΎ dRk
212
+ dΞΎ
213
+ +
214
+ ΞΎ2
215
+ (1 βˆ’ Η«1)2 Rk = 0 .
216
+ (14)
217
+ In GR with a minimally coupled inflaton, we have d ln z/dN = 1 + Η«2/2. Then, for constant SR parameters and in
218
+ the long wavelength limit, the last term is negligible, and the equation admits a constant solution and a solution
219
+ proportional to ΞΎΞ², where
220
+ Ξ² = 3 βˆ’ 4Η«1 + Η«2 + Η«1 (Η«1 βˆ’ 2Η«2)
221
+ (1 βˆ’ Η«1)2
222
+ .
223
+ (15)
224
+ If ΞΎΞ² decreases in time, the constant solution dominates in the ΞΎ β†’ 0 limit. This is what happens for de Sitter and
225
+ SR. In contrast, if ΞΎΞ² increases in time, it dominates in the ΞΎ β†’ 0 limit. This is what occurs for USR leading to
226
+ results that are very different from de Sitter and SR, namely, an amplitude of the spectrum that increases in time.
227
+ The non-constant solution is
228
+ Rk ∝
229
+ οΏ½ k
230
+ aH
231
+ οΏ½Ξ²
232
+ ∼ eβˆ’Ξ²(1βˆ’Η«1) N ,
233
+ (16)
234
+ and it increases or decreases depending on the sign of
235
+ Ξ¦ ≑ Ξ² (1 βˆ’ Η«1) = 3 βˆ’ 4Η«1 + Η«2 + Η«1 (Η«1 βˆ’ 2Η«2)
236
+ (1 βˆ’ Η«1)
237
+ ,
238
+ (17)
239
+
240
+ 4
241
+ increasing if Ξ¦ < 0 and decreasing if Ξ¦ > 0. Only in the latter case the spectral index of primordial spectrum can be
242
+ analytically calculated by using the definition (11). For a general inflationary model, one finds
243
+ βˆ†2
244
+ s ∝ k2+2α2 = k
245
+ 2βˆ’
246
+ οΏ½
247
+ Η«1Η«2
248
+ (1βˆ’Η«1)2 βˆ’1
249
+ οΏ½
250
+ βˆ’
251
+ οΏ½οΏ½
252
+ Η«1Η«2
253
+ (1βˆ’Η«1)2 βˆ’1
254
+ οΏ½2+4 fMS(Η«i)
255
+ (1βˆ’Η«1)2 ,
256
+ (18)
257
+ and
258
+ ns βˆ’ 1 = 2 βˆ’
259
+ οΏ½
260
+ Η«1Η«2
261
+ (1 βˆ’ Η«1)2 βˆ’ 1
262
+ οΏ½
263
+ βˆ’
264
+ οΏ½οΏ½
265
+ Η«1Η«2
266
+ (1 βˆ’ Η«1)2 βˆ’ 1
267
+ οΏ½2
268
+ + 4 fMS(Η«i)
269
+ (1 βˆ’ Η«1)2 .
270
+ (19)
271
+ A.
272
+ Evolutions with β€œConstant” SR Parameters
273
+ Let us now illustrate an important point.
274
+ The results obtained above are exact when the SR parameters are
275
+ constant. However, given the recursive definition of the SR parameters (Η«i+1 = Η«βˆ’1
276
+ i dΗ«i/dN), a constant set of Η«i’s
277
+ corresponds either to H = const and Η«i = 0 (de Sitter case), or Η«1 = const and Η«i = 0 for i > 1 (power law inflation).
278
+ It may thus seem redundant to present the general formalism for such a restricted range of applications. Nevertheless,
279
+ we note that the above results can be applied to a wider set of problems. First, as we already mentioned, the general
280
+ results for Ξ¦ and ns βˆ’ 1 can be applied to the SR case, in which the expressions must be expanded to the first order
281
+ for consistency, since the SR parameters are approximately constant when they are small. Furthermore, the large a
282
+ limit of some transient phase (such as the USR phase) leads to non-trivial sequences of β€œconstant” SR parameters. In
283
+ these cases, one obtains a hierarchy of, for example, Η«i’s with constant, non-zero SR parameters for either even or odd
284
+ values of i, while the remaining SR parameters are zero. For instance, let Η«i
285
+ Nβ†’βˆž
286
+ =
287
+ li +Li(N) with limNβ†’βˆž Li(N) = 0.
288
+ Then, due to their recursive definition, one obtains
289
+ Η«iΗ«i+1 ≑ dΗ«i
290
+ dN
291
+ aβ†’βˆž
292
+ =
293
+ Li,N(N) ,
294
+ (20)
295
+ which leads to limNβ†’βˆž Η«i+1 = 0, provided limNβ†’βˆž Li,N(N) = 0, and, in particular,
296
+ Η«i+1
297
+ Nβ†’βˆž
298
+ =
299
+ Li,N(N)
300
+ li + Li(N) .
301
+ (21)
302
+ Moreover,
303
+ Η«i+2 ≑ dΗ«i+1/dN
304
+ Η«i+1
305
+ Nβ†’βˆž
306
+ =
307
+ Li,NN(N)
308
+ Li,N(N) + Η«i+1 .
309
+ (22)
310
+ Let us now suppose Li(N) ∝ eβˆ’Ξ³N ∼ aβˆ’Ξ³, with Ξ³ > 0. In this case:
311
+ Η«i+2
312
+ Nβ†’βˆž
313
+ =
314
+ βˆ’Ξ³ + Η«i+1 ,
315
+ (23)
316
+ and the subsequent terms of the hierarchy take values equal to zero and βˆ’Ξ³:
317
+ lim
318
+ Nβ†’βˆž Η«i = li,
319
+ lim
320
+ Nβ†’βˆž Η«i+1+2n = 0,
321
+ lim
322
+ Nβ†’βˆž Η«i+2n = βˆ’Ξ³ .
323
+ (24)
324
+ Therefore, due to their definition, an infinite sequence of SR parameters may take alternate β€œconstant” values in the
325
+ large a limit. This property is crucial in the analysis that follows, and it depends on the form of Li(N). Indeed,
326
+ exponential forms lead to the result (24) but, in contrast, if Li ∝ N βˆ’Ξ³, then the sequence obtained is limNβ†’βˆž Η«j = 0
327
+ for j > i.
328
+ It is also worthwhile to mention that similar results can be generalized to other hierarchies of SR parameters because
329
+ they only depend on the recursive definition of the SR parameters [analogously to Eq. (20)] and on the form of Li.
330
+ For example, the same results can be extended to the hierarchy of β€œscalar field flow functions” that is defined by
331
+ Ξ΄0 = Ο†/Ο†0 and Ξ΄iΞ΄i+1 = dΞ΄i/dN. In general, the Η«i’s and the Ξ΄i’s are related through the homogeneous Friedmann
332
+ and Klein-Gordon equations, and, in some scenarios, it is useful to use one or both hierarchies.
333
+
334
+ 5
335
+ III.
336
+ MODEL RECONSTRUCTION
337
+ We are interested in reconstructing scalar field potentials that describe transient inflationary solutions, which are
338
+ associated with varying SR parameters with a β€œconstant” behaviour in the future (and necessarily Η«1 < 1). Therefore,
339
+ the results illustrated in the previous section can be adopted to study such models and to verify whether they can
340
+ generate an amplification of the primordial spectrum. Finding the entire evolution of the scalar field is not necessary
341
+ for this purpose, and we will only calculate the potential and the asymptotic behaviour of the homogeneous quantities
342
+ in terms of the corresponding SR parameters. The potentials that lead to an amplification can then be used to build
343
+ an inflationary model that fits the CMB observations and which produces a large amount of DM in the form of PBHs
344
+ at the end of inflation.
345
+ A.
346
+ GR with a Minimally Coupled Inflaton
347
+ In order to proceed with the reconstruction, let us first briefly review the homogeneous Einstein equation,
348
+ H2 =
349
+ 1
350
+ 3MP
351
+ 2
352
+ οΏ½1
353
+ 2
354
+ Λ™Ο†2 + V (Ο†)
355
+ οΏ½
356
+ ,
357
+ (25)
358
+ Λ™H = βˆ’
359
+ Λ™Ο†2
360
+ 2MP
361
+ 2 ,
362
+ (26)
363
+ which leads to
364
+ MP
365
+ 2H2 (3 βˆ’ Η«1) = V .
366
+ (27)
367
+ This last equation can be used to reconstruct the potential. The Eqs. (26) and (27) can be conveniently used for the
368
+ reconstructions starting from some ansatz for H = H(a). In this case, Eq. (26) becomes
369
+ Η«1 =
370
+ 1
371
+ 2MP
372
+ 2
373
+ � dφ
374
+ d ln a
375
+ οΏ½2
376
+ ,
377
+ (28)
378
+ which can be integrated to obtain, when possible, a = a(Ο†).
379
+ Let us first consider the following evolution of the Hubble constant:
380
+ H = H0
381
+ οΏ½
382
+ Ξ± + A
383
+ an
384
+ οΏ½m
385
+ ,
386
+ (29)
387
+ where A, Ξ±, n > 0. Similarly to USR, the evolution described by Eq. (29) has a de Sitter attractor in the future,
388
+ and, indeed, H(a) is that of USR when n = 6 and m = 1/2. (It is interesting to note that this evolution represents
389
+ a general solution in the model with a minimally coupled scalar field and a constant potential, or, in other words, in
390
+ a universe driven by a mixture of two fluids: a cosmological constant and stiff matter. It is curious that n = 6 and
391
+ m = 1/4 yield the general solution for the universe driven by the Chaplygin gas [8].) We also note that the transient
392
+ is described by A/an ∼ eβˆ’nN and that a result similar to Eq. (24) is then expected. This is easily verified if we
393
+ explicitly calculate the hierarchy of SR parameters:
394
+ Η«1 = m Β· n
395
+ A
396
+ Ξ±an + A = m Η«3 = m Η«5 = . . .
397
+ aβ†’+∞
398
+ βˆ’β†’ 0 ,
399
+ (30)
400
+ and
401
+ Η«2 = βˆ’n
402
+ Ξ±an
403
+ Ξ±an + A = Η«4 = Η«6 = . . .
404
+ aβ†’+∞
405
+ βˆ’β†’ βˆ’n ,
406
+ (31)
407
+ where a > [(m Β· n βˆ’ 1)A/Ξ±]1/n is necessary for inflation to occur. We can integrate and invert Eq. (28) to obtain
408
+ exp
409
+ οΏ½Ο† βˆ’ Ο†0
410
+ MP
411
+ οΏ½ n
412
+ 2m
413
+ οΏ½
414
+ = x + 1
415
+ x βˆ’ 1
416
+ x0 βˆ’ 1
417
+ x0 + 1 ,
418
+ (32)
419
+
420
+ 6
421
+ with x ≑ Aβˆ’1/2√
422
+ Ξ±an + A and x, x0 > 1. Notice that Ο† = Ο†0 when x = x0. Conversely, Ο† = Ο†βˆž, with
423
+ Ο†βˆž ≑ Ο†0 + MP
424
+ οΏ½
425
+ 2m
426
+ n ln B0 ,
427
+ (33)
428
+ for x β†’ ∞. Eq. (32) can be solved for x, which yields
429
+ x = e
430
+ Ο†βˆ’Ο†0
431
+ MP
432
+ √ n
433
+ 2m + B0
434
+ e
435
+ Ο†βˆ’Ο†0
436
+ MP
437
+ √ n
438
+ 2m βˆ’ B0
439
+ ,
440
+ (34)
441
+ where B0 = (x0 βˆ’ 1)/(x0 + 1), and the reconstructed potential is finally
442
+ V = H2
443
+ 0
444
+ οΏ½ Ξ±x2
445
+ x2 βˆ’ 1
446
+ οΏ½2m οΏ½
447
+ 3 βˆ’ n Β· m
448
+ x2
449
+ οΏ½
450
+ .
451
+ (35)
452
+ For n = 6 and m = 1/2, one recovers a constant potential and the USR evolution, as expected. For other choices of the
453
+ parameters n and m, the expression for the potential in terms of Ο† is a complicated function with exponentials that
454
+ need not be written here explicitly. However, this cumbersome expression is exact. Since the asymptotic behaviour
455
+ of the potential at Ο† ∼ Ο†βˆž determines the limiting values of the SR parameters, we simply give the form of V around
456
+ Ο†βˆž, which is
457
+ V ≃ 3H2
458
+ 0Ξ±2m
459
+ οΏ½
460
+ 1 + n
461
+ 4
462
+ οΏ½
463
+ 1 βˆ’ n
464
+ 2
465
+ οΏ½ οΏ½Ο† βˆ’ Ο†βˆž
466
+ MP
467
+ οΏ½2οΏ½
468
+ .
469
+ (36)
470
+ Finally, let us calculate the consequences of the background evolution given by Eq; (29) on the inflationary spectrum.
471
+ The value of Ξ¦ is
472
+ Ξ¦ = 3 βˆ’ 4Η«1 + Η«2 + Η«1 (Η«1 βˆ’ 2Η«2)
473
+ (1 βˆ’ Η«1)
474
+ aβ†’+∞
475
+ βˆ’β†’ 3 βˆ’ n ,
476
+ (37)
477
+ and for n > 3 the curvature perturbations Rk are amplified, after their horizon exit, as time passes. In contrast, if
478
+ 0 < n < 3, from the constant solution for Rk, one finds
479
+ ns βˆ’ 1 = n > 0 ,
480
+ (38)
481
+ which implies the amplitude is that of a blue-tilted spectrum, which grows as the wavenumber k increases.
482
+ We
483
+ conclude that for GR with a minimally coupled inflaton, the inflationary evolution described by Eq. (29), with a
484
+ transient phase and a de Sitter attractor in the future, leads to an inflationary enhancement. The corresponding
485
+ inflaton dynamics is driven by the potential (35) and similar behaviours can be obtained from potentials of the form
486
+ (36) with the field close to Ο†βˆž.
487
+ B.
488
+ Power Law solutions
489
+ In this section, we generalise the results obtained from Eq. (29), and study the transient phase with a Power Law
490
+ inflation attractor. For this case, in contrast to de Sitter, it is only possible to reconstruct the inflaton potential
491
+ exactly for particular choices of the parameters. Close to the attractor, an approximate reconstruction can always
492
+ be obtained, and that is enough for the purposes of model building. The amplification of the primordial spectrum
493
+ can still be studied in full generality, as it depends on the asymptotic values of the SR parameters, which can be
494
+ calculated exactly. In this case, and in the large a limit, one obtains
495
+ Η«1 β†’ const + L(a) ,
496
+ (39)
497
+ with L(a) β†’ 0. In analogy to the previous case, we consider
498
+ Η«1 =
499
+ οΏ½
500
+ Ξ² + B
501
+ an
502
+ οΏ½m
503
+ β†’ Ξ²m ,
504
+ (40)
505
+ with β, B, n > 0 and βm < 1 (so as to have acceleration close to the attractor). Notice that, when β = 0, one finds
506
+ a transient phase with a de Sitter attractor, but Η«1 in Eq. (40) is different from that in the set (30). This case is
507
+
508
+ 7
509
+ expected to generate a hierarchy of the form (24) in the large a limit. Indeed, the ansatz (40) leads to the following
510
+ hierarchy of SR parameters:
511
+ Η«2 = βˆ’mΗ«4 = βˆ’mΗ«6 = Β· Β· Β· = βˆ’ n m B
512
+ B + Ξ² an β†’ 0 ,
513
+ (41)
514
+ and
515
+ Η«3 = Η«5 = Β· Β· Β· = βˆ’
516
+ n Ξ²an
517
+ B + Ξ² an β†’ βˆ’n ,
518
+ (42)
519
+ where, in contrast to the de Sitter case examined in the previous section, now the even SR parameters tend to zero.
520
+ By proceeding with reconstruction and integrating Eqs. (40) and (28), one finds, respectively,
521
+ H = H0 exp
522
+ οΏ½
523
+ βˆ’
524
+ οΏ½
525
+ Ξ² + B
526
+ an
527
+ οΏ½1+m
528
+ (1 + m)nΞ²
529
+ 2F1
530
+ οΏ½
531
+ 1, 1 + m, 2 + m, 1 + B
532
+ Ξ²an
533
+ οΏ½οΏ½
534
+ ,
535
+ (43)
536
+ and
537
+ Ο† βˆ’ Ο†0 = f(a) βˆ’ f(a0) ,
538
+ (44)
539
+ where
540
+ f(a) =
541
+ 2
542
+ √
543
+ 2MP
544
+ οΏ½
545
+ Ξ² + B
546
+ an
547
+ οΏ½ 2+m
548
+ 2
549
+ 2F1
550
+ οΏ½
551
+ 1, 1 + m
552
+ 2 , 2 + m
553
+ 2 , 1 +
554
+ B
555
+ Ξ²an
556
+ οΏ½
557
+ (2 + m)nΞ²
558
+ .
559
+ (45)
560
+ In this case, the exact reconstruction of the potential is rather complicated unless one adopts simplifying assumptions.
561
+ For example, let m = βˆ’1 and 0 < Ξ² < 1. Then,
562
+ H =
563
+ H0
564
+ [n (B + Ξ²an)]
565
+ 1
566
+ nΞ² ,
567
+ (46)
568
+ and
569
+ Ο† βˆ’ Ο†0 = MP
570
+ ln
571
+ οΏ½
572
+ 1+
573
+ οΏ½
574
+ Η«1(a)
575
+ Ξ²
576
+ 1βˆ’
577
+ οΏ½
578
+ Η«1(a)
579
+ Ξ²
580
+ 1βˆ’
581
+ οΏ½
582
+ Η«1(a0)
583
+ Ξ²
584
+ 1+
585
+ οΏ½
586
+ Η«1(a0)
587
+ Ξ²
588
+ οΏ½
589
+ n√β
590
+ .
591
+ (47)
592
+ In the a β†’ ∞ limit, one obtains
593
+ Ο†βˆž = Ο†0 + MP
594
+ ln
595
+ οΏ½
596
+ Ξ²+1
597
+ Ξ²βˆ’1A0
598
+ οΏ½
599
+ n√β
600
+ ,
601
+ (48)
602
+ with A0 ≑
603
+ οΏ½
604
+ 1 βˆ’
605
+ οΏ½
606
+ Η«1(a0)/Ξ²
607
+ οΏ½
608
+ /
609
+ οΏ½
610
+ 1 +
611
+ οΏ½
612
+ Η«1(a0)/Ξ²
613
+ οΏ½
614
+ . The relation (47) can be inverted to obtain an = an(Ο†):
615
+ an = an
616
+ 0
617
+ οΏ½οΏ½
618
+ 1 βˆ’
619
+ οΏ½
620
+ Η«1(a0)
621
+ Ξ²
622
+ οΏ½
623
+ +
624
+ οΏ½
625
+ 1 +
626
+ οΏ½
627
+ Η«1(a0)
628
+ Ξ²
629
+ οΏ½
630
+ en√β(Ο†βˆ’Ο†0)/MP
631
+ οΏ½2
632
+ 4en√β(Ο†βˆ’Ο†0)/MP
633
+ ,
634
+ (49)
635
+ and finally the potential can be reconstructed, provided we substitute Eq. (49) into Eq. (27). In terms of an, it then
636
+ takes the following form:
637
+ V =
638
+ H2
639
+ 0
640
+ [n (B + Ξ²an)]
641
+ 2
642
+ nΞ²
643
+ οΏ½
644
+ 3 βˆ’
645
+ οΏ½
646
+ Ξ² + B
647
+ an
648
+ οΏ½mοΏ½
649
+ .
650
+ (50)
651
+ The expression in terms of Ο† is cumbersome and it will not be needed. It is also worthwhile to note that such a
652
+ potential depends on the homogeneous inflaton through the exponential function exp
653
+ οΏ½
654
+ nβˆšΞ²Ο†/MP
655
+ οΏ½
656
+ . This functional
657
+ dependence is expected as it is the generalization of the standard Power-Law inflation potential, which only contains
658
+ one exponential function of the inflaton. Moreover, various approximate reconstruction methods can be used to obtain
659
+
660
+ 8
661
+ the shape of the potential close to the attractor but we omit this discussion here. Whereas the exact reconstruction can
662
+ be obtained for certain values of the parameters, the behaviour of the resulting inflationary spectra can be calculated
663
+ exactly from Eqs. (41) and (42). For generic values of m, one may compute
664
+ Ξ¦ = Ξ²2m βˆ’ 4Ξ²m + 3
665
+ (1 βˆ’ Ξ²m)
666
+ = 3 βˆ’ Ξ²m > 0 .
667
+ (51)
668
+ This shows the absence of the growing solution for Eq. (13). The spectral index is then simply given by
669
+ ns βˆ’ 1 = βˆ’ 2Ξ²m
670
+ 1 βˆ’ Ξ²m < 0 .
671
+ (52)
672
+ This is the same result as the one obtained for the Power Law attractor solution. In contrast to the de Sitter case,
673
+ the resulting primordial spectrum, if evaluated on the trajectory which approaches the attractor (and close to it),
674
+ coincides with the spectrum calculated on the attractor itself, and no amplification or peculiar features emerge. It is
675
+ also noteworthy that this result is not restricted to the evolution given by Eq. (40), as it only depends on the limits
676
+ (41) and (42), which are not partcular to Eq. (40). For instance, starting from the ansatz
677
+ H(a) = H0
678
+ οΏ½a0
679
+ a
680
+ οΏ½Ξ²m οΏ½
681
+ 1 + A0
682
+ an
683
+ οΏ½m
684
+ ,
685
+ (53)
686
+ where n, A0 > 0, the resulting spectra are the same.
687
+ The observed absence of amplification in the cases of Power-Law inflation considered here is relevant because Power-
688
+ Law is the exactly solvable inflationary model that is most akin to SR. One may then conjecture that similar results
689
+ (and, in particular, the lack of amplification) hold for SR inflation when the inflaton approaches the attractor solution,
690
+ and the enhancement is a peculiarity of de Sitter.
691
+ C.
692
+ Non-Minimally Coupled Inflaton
693
+ Let us now consider the different scenario of a non-minimally coupled inflaton. In order to perform the reconstruc-
694
+ tion, we first review the basic homogeneous equations for this model:
695
+ H2 =
696
+ 1
697
+ 3F(Ο†)
698
+ οΏ½ Λ™Ο†2
699
+ 2 + V βˆ’ 3HF,Ο† Λ™Ο†
700
+ οΏ½
701
+ ,
702
+ (54)
703
+ and
704
+ Λ™H = βˆ’
705
+ 1
706
+ 2F(Ο†)
707
+ οΏ½
708
+ (1 + F,φφ) Λ™Ο†2 + F,Ο†
709
+ οΏ½
710
+ ¨φ βˆ’ H Λ™Ο†
711
+ οΏ½οΏ½
712
+ ,
713
+ (55)
714
+ where F(Ο†) represents a general non-minimal coupling and F = MP reproduces the minimally coupled case. In contrast
715
+ to the previous cases, the homogeneous equations and the reconstruction procedure now become more involved. Then,
716
+ for simplicity, we shall henceforth limit our study to the induced gravity (IG) case, where F(Ο†) = ΞΎΟ†2 [9, 10]. This
717
+ simplifying choice is also justified by the fact that both Higgs inflation and Starobinsky inflation (in the Einstein
718
+ Frame) occur in a regime that is very close to pure IG.
719
+ Reconstructing the inflaton potential for a given H(a) is not as straightforward as for GR with a minimally coupled
720
+ inflaton, and we found exact potentials only for certain values of the parameters and for the de Sitter attractor case
721
+ [cf. Eq. (29)]. Nevertheless, we can still predict the shape of the inflationary spectra or, at least, the possibility of an
722
+ amplification in the large a limit.
723
+ D.
724
+ De Sitter limit
725
+ Let us consider H(a) given by Eq. (29). In IG, the following exact relations hold between some SR parameters:
726
+ Η«1 =
727
+ Ξ΄1
728
+ 1 + Ξ΄1
729
+ οΏ½ Ξ΄1
730
+ 2ΞΎ + 2Ξ΄1 + Ξ΄2 βˆ’ 1
731
+ οΏ½
732
+ ,
733
+ (56)
734
+
735
+ 9
736
+ Η«1 =
737
+ 1
738
+ 2ΞΎ(1 + 6ΞΎ)
739
+ οΏ½
740
+ (1 + 2ΞΎ)Ξ΄2
741
+ 1 βˆ’ 8ΞΎΞ΄1 βˆ’ 6ΞΎ2
742
+ οΏ½
743
+ 1 + 2Ξ΄1 βˆ’ Ξ΄2
744
+ 1
745
+ 6ΞΎ
746
+ οΏ½ οΏ½d ln V
747
+ d ln Ο† βˆ’ 4
748
+ οΏ½οΏ½
749
+ .
750
+ (57)
751
+ Before discussing the reconstruction of the inflaton potential, we must first calculate the asymptotic values of the SR
752
+ parameters, which are pivotal in the analysis of the amplification of the spectrum. From Eq. (54), the potential can
753
+ then be obtained as
754
+ V = 3ΞΎΟ†2H2
755
+ οΏ½
756
+ 1 + 2Ξ΄1 βˆ’ 1
757
+ 6ΞΎ Ξ΄2
758
+ 1
759
+ οΏ½
760
+ ,
761
+ (58)
762
+ provided H = H(Ο†) and Ξ΄1 = Ξ΄1(Ο†) are known [indeed, Eq. (58) is the IG counterpart of Eq. (27) in GR].
763
+ Let us first calculate the SR parameters in the large a limit. Since limaβ†’βˆž Η«1 = 0, one either has limaβ†’βˆž Ξ΄1 = 0
764
+ and limaβ†’βˆž Ξ΄2 ΜΈ= 0, or limaβ†’βˆž Ξ΄2 = 0 and limaβ†’βˆž Ξ΄1 ΜΈ= 0 but satisfying the relation
765
+ δ1,∞ =
766
+ 2ΞΎ
767
+ 1 + 4ΞΎ .
768
+ (59)
769
+ These results follow from the functional dependence of H(a) on a inherited by Η«1 and Ξ΄i’s and by the general result
770
+ given in Eq. (24), which is applied here to the SR hierarchy Ξ΄i. Notice that, in contrast to GR, two different de Sitter
771
+ trajectories are present in IG, and they are associated with two different evolutions of the inflaton field. Using Eq.
772
+ (57) in the same limit for a, one obtains that the potential, on the attractor, must satisfy
773
+ d ln V∞
774
+ d ln Ο† βˆ’ 4 = 0 β‡’ V∞ ∝ Ο†4
775
+ (60)
776
+ in the former case and
777
+ d ln V∞
778
+ d ln Ο† βˆ’ 4 = 0 β‡’ V∞ ∝ Ο†2
779
+ (61)
780
+ in the latter case.
781
+ We can now proceed to evaluate the full hierarchy of Ξ΄i’s. Starting from Eq. (56) and differentiating, we find
782
+ Η«2 = Ξ΄2
783
+ οΏ½
784
+ (1 + 4ΞΎ)Ξ΄2
785
+ 1 + 2ΞΎ (Ξ΄2 + Ξ΄3 βˆ’ 1) + 2Ξ΄1 (1 + 4ΞΎ + ΞΎΞ΄3)
786
+ οΏ½
787
+ (1 + Ξ΄1) [(1 + 4ΞΎ)Ξ΄1 + 2ΞΎ(Ξ΄2 βˆ’ 1)]
788
+ ,
789
+ (62)
790
+ and, by further differentiating, the Η«i’s with arbitrary large i can be obtained. In the large-a limit, we have already
791
+ calculated Η«2i = βˆ’n and Η«2i+1 = 0 [cf. Eqs. (30) and (31)], and one then obtains two possible hierarchies for the Ξ΄i’s:
792
+ Ξ΄2i+1,∞ = 0, Ξ΄2i,∞ = Η«2,∞ = βˆ’n ,
793
+ (63)
794
+ and
795
+ δ1,∞ =
796
+ 2ΞΎ
797
+ 1 + 4ΞΎ , Ξ΄2i+1,∞ = βˆ’n, Ξ΄2i,∞ = 0 .
798
+ (64)
799
+ This latter statement cannot be simply verified by substitution because the limits involved do not commute. For
800
+ example, on substituting first δ2 = 0 in Eq. (62), one obtains ǫ2 = 0, which is not correct. One must solve (at least
801
+ perturbatively in the large-a limit) Eq. (56) and then evaluate the limits with the help of the solution found. The
802
+ above results are correctly reproduced only if we proceed in this manner.
803
+ The exact reconstruction of the inflaton potential is not possible in general. Nonetheless, in specific cases, the
804
+ potential may be derived exactly as follows. Consider the following ansatz for Ξ΄1:
805
+ Ξ΄1 = n0 + n1aβˆ’n
806
+ d0 + d1aβˆ’n ,
807
+ (65)
808
+ which is suggested by the expression for Η«1 and Eq. (56). If n0 = 0 and n1 ΜΈ= 0, then Eq. (65) can be integrated, and
809
+ the resulting Ο†(a) is inverted as follows
810
+ Ο†(a) = Ο†0
811
+ οΏ½
812
+ d0 + d1aβˆ’nοΏ½βˆ’ n1
813
+ n d1 β‡’ aβˆ’n =
814
+ οΏ½
815
+ Ο†(a)
816
+ Ο†0
817
+ οΏ½βˆ’ n d1
818
+ n1 βˆ’ d0
819
+ d1
820
+ .
821
+ (66)
822
+
823
+ 10
824
+ The coefficients n1, d0, d1 and ΞΎ can finally be fixed by the requirement that Eq. (65) be a solution of Eq. (56). Two
825
+ nontrivial solutions can be found:
826
+ d0 = βˆ’(1 + n) n1Ξ±
827
+ A m n
828
+ , d1 = βˆ’(1 + n)n1
829
+ m n
830
+ , ΞΎ =
831
+ m
832
+ 2(1 βˆ’ 3m + n + m n) ,
833
+ (67)
834
+ or
835
+ d0 = βˆ’(1 + n) n1Ξ±
836
+ A m n
837
+ , d1 = βˆ’(1 + n + m n)n1
838
+ m n
839
+ , ΞΎ =
840
+ m
841
+ 2(1 βˆ’ 2m + n + m n) .
842
+ (68)
843
+ Notice that more exact solutions for Ξ΄1 can be found if we start from the ansatz (65) and n0 ΜΈ= 0. However, by further
844
+ integrating these solutions to obtain Ο†(a), one is led to non-invertible functions, and the reconstruction cannot be
845
+ completed. For both Eqs. (67) and (68) one has
846
+ δ1,∞ = 0 ,
847
+ (69)
848
+ and one can explicitly verify that the hierarchies belong to the set (63). Notice that n1 in Eqs. (67) and (68) can be
849
+ arbitrarily chosen, as should be due to the form of the ansatz (65). Let us, for simplicity, complete the reconstruction
850
+ choosing n and m to reproduce USR in the IG context (n = 6, m = 1/2). In this case, Eqs. (67) and (68) take the
851
+ following form:
852
+ n0 = 0, d0 = βˆ’ 7Ξ±
853
+ 3An1, d1 = βˆ’7
854
+ 3n1, ΞΎ = 1/10 β‡’ Ξ΄1 = βˆ’
855
+ 3A
856
+ 7 (A + Ξ±a6) ,
857
+ (70)
858
+ n0 = 0, d0 = βˆ’ 7Ξ±
859
+ 3An1, d1 = βˆ’10
860
+ 3 n1, ΞΎ = 1/36 β‡’ Ξ΄1 = βˆ’
861
+ 3A
862
+ 10A + 7Ξ±a6 .
863
+ (71)
864
+ From Eq. (29), a(Ο†) in (66) and Eq. (70), one finds
865
+ Ξ΄1 = 3
866
+ 7
867
+ οΏ½
868
+ Ξ±
869
+ οΏ½Ο†0
870
+ Ο†
871
+ οΏ½14
872
+ βˆ’ 1
873
+ οΏ½
874
+ and
875
+ H2 = H2
876
+ 0
877
+ οΏ½ Ο†
878
+ Ο†0
879
+ οΏ½14
880
+ ,
881
+ (72)
882
+ with Ο†/Ο†0
883
+ aβ†’βˆž
884
+ βˆ’β†’ Ξ±1/14 and Ο† > Ο†0, while for Eq. (71) one finds
885
+ Ξ΄1 = 3
886
+ 10
887
+ οΏ½
888
+ Ξ±
889
+ οΏ½Ο†0
890
+ Ο†
891
+ οΏ½20
892
+ βˆ’ 1
893
+ οΏ½
894
+ and
895
+ H2 = H2
896
+ 0
897
+ οΏ½οΏ½ 7
898
+ 10
899
+ Ο†
900
+ Ο†0
901
+ οΏ½20
902
+ + 3Ξ±
903
+ 10
904
+ οΏ½
905
+ ,
906
+ (73)
907
+ with Ο†/Ο†0
908
+ aβ†’βˆž
909
+ βˆ’β†’ Ξ±1/20 and Ο† > Ο†0. Finally, by using Eq. (58), one obtains
910
+ V = βˆ’
911
+ 3H2
912
+ 0
913
+ 490Ο†12Ο†14
914
+ 0
915
+ οΏ½
916
+ 8Ο†28 βˆ’ 72Ξ±Ο†14
917
+ 0 Ο†14 + 15Ξ±2Ο†28
918
+ 0
919
+ οΏ½
920
+ (74)
921
+ for the first exact solution, and
922
+ V = βˆ’
923
+ H2
924
+ 0
925
+ 6000Ο†38Ο†20
926
+ 0
927
+ οΏ½
928
+ 7Ο†20 + 3Ξ±Ο†20
929
+ 0
930
+ οΏ½ οΏ½
931
+ 7Ο†40 βˆ’ 84Ξ±Ο†20
932
+ 0 Ο†20 + 27Ξ±2Ο†40
933
+ 0
934
+ οΏ½
935
+ (75)
936
+ for the second. In the a β†’ ∞ limit, the potentials (74) and (75) satisfy the condition d ln V/d ln Ο† = 4. The potential
937
+ can have negative values but, in the vicinity of Ο† ≃ Ο†βˆž, the potential is positive and V∞ > 0.
938
+ We discuss at last the behaviour of the primordial scalar curvature spectrum. The general formulae illustrated in
939
+ the Sec. II can be easily generalised to the IG case wherein
940
+ zIG = aφδ1
941
+ οΏ½
942
+ 1 + 6ΞΎ
943
+ 1 + Ξ΄1
944
+ ,
945
+ (76)
946
+ and Ξ¦ is given by
947
+ Ξ¦ =
948
+ οΏ½
949
+ 1 βˆ’ Η«1 βˆ’
950
+ Η«1Η«2
951
+ (1 βˆ’ Η«1) +
952
+ οΏ½
953
+ 2 + 2Ξ΄1 + 2Ξ΄2 βˆ’ Ξ΄1Ξ΄2
954
+ 1 + Ξ΄1
955
+ οΏ½οΏ½
956
+ .
957
+ (77)
958
+
959
+ 11
960
+ If we evaluate Ξ¦ w.r.t. to the hierarchies (63) and (64), one observes that only constants and terms linear in the SR
961
+ parameters remain. Moreover ǫ1,∞ = 0 and Φ then simplifies to
962
+ Ξ¦ = 3 + 2Ξ΄1 + 2Ξ΄2 ,
963
+ (78)
964
+ which can be negative only for the hierarchy (63) but is strictly positive for the hierarchy (64), provided we restrict
965
+ ourselves to positive values of the non-minimal coupling ΞΎ. In the former case, Ξ¦ = 3 βˆ’ 2n, which implies that the
966
+ growing solution exists for n > 3/2.
967
+ If no growing solution exists [as is the case for (64) or (63) with n < 3/2], an amplification of the spectrum is only
968
+ possible if the spectrum is blue-tilted. Let us then evaluate ns βˆ’ 1. In the IG case, fMS(Η«i) in the MS equation is
969
+ given by
970
+ fMS = Ξ΄2
971
+ 1 + Ξ΄2
972
+ 2 + (3 βˆ’ Η«1) (1 + Ξ΄1 + Ξ΄2) + Ξ΄2Ξ΄3 +
973
+ Ξ΄1Ξ΄2
974
+ οΏ½
975
+ Η«1 + Ξ΄1 βˆ’ 3Ξ΄2 βˆ’ Ξ΄3 + 2Ξ΄1Ξ΄2
976
+ 1+Ξ΄1 βˆ’ 2
977
+ οΏ½
978
+ 1 + Ξ΄1
979
+ βˆ’ 1 ,
980
+ (79)
981
+ and, as usual, it can be simplified to obtain the following expression for the scalar spectral index:
982
+ ns βˆ’ 1 = 3 βˆ’
983
+ οΏ½
984
+ 1 + 4 (Ξ΄2
985
+ 1 + Ξ΄2
986
+ 2 + 3 (1 + Ξ΄1 + Ξ΄2) βˆ’ 1) .
987
+ (80)
988
+ Then, for the hierarchy (63) and n < 3/2, we obtain
989
+ ns βˆ’ 1 = 3 βˆ’ |3 βˆ’ 2n| = 2n ,
990
+ (81)
991
+ which is indeed blue-tilted, while for the hierarchy (64), we find
992
+ ns βˆ’ 1 = βˆ’
993
+ 4ΞΎ
994
+ 1 + 4ΞΎ ,
995
+ (82)
996
+ which is red-tilted.
997
+ We therefore conclude that solutions having H of the form given in Eq. (29), in the IG context, may lead to a
998
+ spectrum enhancement for evolutions asymptotically described by the hierarchy (63) and either for n > 3/2 (due to
999
+ the presence of the growing solution) or 0 < n < 3/2 (in absence of the growing solution but with the blue-tilted
1000
+ spectrum).
1001
+ IV.
1002
+ APPLICATIONS
1003
+ We have so far studied the consequences of cosmological evolutions with a transient phase, which is crucial to
1004
+ potentially obtain the amplification required by the formation of PBHs. Indeed, the presence of the transient generates,
1005
+ in the large-a limit, a sequence of values for the SR parameters that is otherwise not obtained. We then reconstructed,
1006
+ when possible, the potentials that led to the desired evolution. In this section, our approach will be slightly different,
1007
+ as we shall study the presence of the transient solutions in the particular dynamical regime of constant roll (CR)
1008
+ inflation [11], which is the natural generalisation of USR.
1009
+ CR solutions satisfy the equation
1010
+ ¨φ + BH Λ™Ο† = 0 ,
1011
+ (83)
1012
+ where B > 0, and one recovers the USR solution for B = 3, while the case of |B| β‰ͺ 1 reproduces standard SR. We
1013
+ observe that the CR condition (83) can be rewritten, in terms of the SR parameters, as
1014
+ Ξ΄2 + Ξ΄1 βˆ’ Η«1 + B = 0 .
1015
+ (84)
1016
+ Eq. (84) is model independent, since it only depends on the definitions of Η«i’s and Ξ΄i’s, and can be easily integrated
1017
+ to obtain
1018
+ dφ
1019
+ d ln aH
1020
+ οΏ½ a
1021
+ a0
1022
+ οΏ½B
1023
+ = C3 β‡’ Λ™Ο† = C3
1024
+ οΏ½a0
1025
+ a
1026
+ οΏ½B
1027
+ ,
1028
+ (85)
1029
+ where C3 is an integration constant.
1030
+ In the minimally coupled case, CR can generate an amplification of the primordial scalar spectrum. In what follows,
1031
+ after a revision of this result (which was analysed in [7]) we shall consider CR in IG and study its consequences.
1032
+
1033
+ 12
1034
+ A.
1035
+ Constant Roll in GR with a Minimally Coupled Inflaton
1036
+ In GR with a minimally coupled inflaton, on imposing CR conditions and adopting the Hamilton-Jacobi (HJ)
1037
+ formalism, it is possible to reconstruct the evolution of the Hubble parameter and the corresponding potential [7]. In
1038
+ particular, one finds that H(Ο†) is the following superposition of two exponential functions:
1039
+ H(Ο†) = C1 exp
1040
+ οΏ½οΏ½
1041
+ B
1042
+ 2
1043
+ Ο†
1044
+ MP
1045
+ οΏ½
1046
+ + C2 exp
1047
+ οΏ½
1048
+ βˆ’
1049
+ οΏ½
1050
+ B
1051
+ 2
1052
+ Ο†
1053
+ MP
1054
+ οΏ½
1055
+ .
1056
+ (86)
1057
+ In [7], the solution (86) with one exponential (C1 = 0 or C2 = 0), as well as the cosh and sinh cases, are analysed
1058
+ with the aim of finding the exact solutions compatible with CMB observations [12] (and thus not amplified).
1059
+ Here, in a slightly different approach, we consider the general case, and we study the power enhancement of the
1060
+ spectrum. Eq. (26) can be rewritten in terms of the SR parameters
1061
+ Η«1 =
1062
+ Ο†2
1063
+ 2MP
1064
+ 2 Ξ΄2
1065
+ 1 ,
1066
+ (87)
1067
+ from which, using the chain rule, we obtain
1068
+ Η«1 = βˆ’Ξ΄1
1069
+ d ln H
1070
+ d ln Ο† .
1071
+ (88)
1072
+ Eq. (87) then becomes
1073
+ Η«1 = 2MP
1074
+ 2
1075
+ Ο†2
1076
+ οΏ½d ln H
1077
+ d ln Ο†
1078
+ οΏ½
1079
+ .
1080
+ (89)
1081
+ The potential can subsequently be reconstructed by substituting Eqs. (86) and (89) into Eq. (27):
1082
+ V (Ο†) = MP
1083
+ 2H(Ο†)2
1084
+ οΏ½
1085
+ 3 βˆ’ 2MP
1086
+ 2
1087
+ Ο†2
1088
+ οΏ½d ln H
1089
+ d ln Ο†
1090
+ οΏ½2οΏ½
1091
+ .
1092
+ (90)
1093
+ In order to obtain the corresponding evolution, one must integrate and invert the equation
1094
+ Ξ΄1 = βˆ’2MP
1095
+ 2
1096
+ Ο†2
1097
+ d ln H
1098
+ d ln Ο† ,
1099
+ (91)
1100
+ which can be easily derived from Eq. (87) by using (88). One finds
1101
+ οΏ½ a
1102
+ a0
1103
+ οΏ½B
1104
+ =
1105
+ x
1106
+ B (C2 βˆ’ C1x2) ,
1107
+ (92)
1108
+ where x = exp
1109
+ οΏ½οΏ½
1110
+ B
1111
+ 2
1112
+ Ο†
1113
+ MP
1114
+ οΏ½
1115
+ . It is straightforward to invert Eq. (92) so as to obtain x = x(a). Correspondingly, one has
1116
+ H(a) = Β±
1117
+ 4C1C2 +
1118
+ οΏ½ a0
1119
+ a
1120
+ οΏ½2B βˆ“
1121
+ οΏ½ a0
1122
+ a
1123
+ οΏ½B οΏ½
1124
+ 4C1C2 +
1125
+ οΏ½ a0
1126
+ a
1127
+ οΏ½2B
1128
+ βˆ“
1129
+ οΏ½ a0
1130
+ a
1131
+ οΏ½B +
1132
+ οΏ½
1133
+ 4C1C2 +
1134
+ οΏ½ a0
1135
+ a
1136
+ οΏ½2B
1137
+ aβ†’βˆž
1138
+ βˆ’β†’ Β±8C1C2 +
1139
+ οΏ½ a0
1140
+ a
1141
+ οΏ½2B
1142
+ 4√C1C2
1143
+ .
1144
+ (93)
1145
+ Notice that the same result can be obtained if one uses the CR definition (84) instead of Eq. (26).
1146
+ The last, approximate, equality in Eq. (93) is the large-a limit of H(a), and this shows that the CR evolution is
1147
+ asymptotically equivalent to the evolution given in Eq. (29) with m = 1 and n = 2B. The results obtained in the
1148
+ Sec. II for large a are therefore inherited by CR. Thus, one obtains ǫ2i+1,∞ = 0 and ǫ2i,∞ = 2B. Correspondingly,
1149
+ Ξ¦ = 3 βˆ’ 2B ,
1150
+ (94)
1151
+ which shows that the curvature perturbations are amplified for B > 3/2 due to the presence of a growing solution. In
1152
+ contrast, if 0 < B < 3/2, one finds a blue-tilted spectrum
1153
+ ns βˆ’ 1 = 3 βˆ’
1154
+ οΏ½
1155
+ (3 βˆ’ 2B)2 = 2B > 0 ,
1156
+ (95)
1157
+ i.e., a spectrum enhancement in the absence of growing solutions. Therefore, CR inflation admits transient solutions
1158
+ that always lead to an amplification. Finally, it is worthwhile to mention that the solutions with C1 = 0 or C2 = 0
1159
+ simply correspond to the attractor solutions for power-law inflation, and thus they are not associated with any
1160
+ amplification effect.
1161
+
1162
+ 13
1163
+ B.
1164
+ Constant Roll with a Non-Minimally Coupled Inflaton
1165
+ Let us now consider CR in the IG context. For this case, the HJ formalism leads to [13]
1166
+ H(Ο†) = C1Ο†(B+p)/2 +
1167
+ C2
1168
+ Ο†(pβˆ’B)/2 ,
1169
+ (96)
1170
+ where p =
1171
+ οΏ½
1172
+ (B + 2)2 + 2B(2 + ΞΎβˆ’1), and (B + p)/2 and (p βˆ’ B)/2 are both positive with (p βˆ’ B)/2 < (B + p)/2.
1173
+ For simplicity, we shall take C1,2 > 0 and we restrict the analysis to the Ο† > 0 interval. Studying the spectrum
1174
+ enhancement for CR in the IG case is more complicated than for GR. This is essentially a consequence of the
1175
+ complicated form of Eq. (56) in comparison to Eq. (26) in the GR case. However, the simple relation (84) holds, and
1176
+ it can be used to simplify the equations. First, with Eq. (84), one may eliminate Ξ΄2 from Eq. (56) and obtain
1177
+ Η«1 = 1 + 2ΞΎ
1178
+ 2ΞΎ
1179
+ Ξ΄2
1180
+ 1 βˆ’ (B + 1)Ξ΄1 .
1181
+ (97)
1182
+ Subsequently, by using Eq. (88), one finds
1183
+ Ξ΄1 =
1184
+ 2ΞΎ
1185
+ 1 + 2ΞΎ
1186
+ οΏ½
1187
+ B + 1 βˆ’ d ln H
1188
+ d ln Ο†
1189
+ οΏ½
1190
+ ,
1191
+ (98)
1192
+ and the potential can be reconstructed by substituting Eqs. (96) and (98) into Eq. (58).
1193
+ The evolution could be obtained by integrating Eq. (98) and inverting the result. However, analytically inverting
1194
+ the resulting equation for arbitrary values of the parameters is impossible. As we are only interested in the asymptotic
1195
+ form of H(a), one can employ a perturbative approach. Integration of Eq. (98) yields
1196
+ οΏ½a0
1197
+ a
1198
+ οΏ½B
1199
+ = Ο†
1200
+ 2+B
1201
+ 2
1202
+ οΏ½
1203
+ (B + p + 2) C1Ο†
1204
+ p
1205
+ 2 + (B βˆ’ p + 2) C2
1206
+ Ο†
1207
+ p
1208
+ 2
1209
+ οΏ½
1210
+ ,
1211
+ (99)
1212
+ where B + p + 2 > 0 and B βˆ’ p + 2 < 0. Therefore, in the large-a limit, the inversion of Eq. (99) leads to
1213
+ Ο†(a) = Ο†βˆž +
1214
+ οΏ½
1215
+ i>1
1216
+ Ο†i
1217
+ οΏ½a0
1218
+ a
1219
+ οΏ½i B
1220
+ ∼ Ο†βˆž + Ο†1
1221
+ οΏ½a0
1222
+ a
1223
+ οΏ½B
1224
+ ,
1225
+ (100)
1226
+ where Ο†βˆž is positive. By substituting Eq. (100) into Eq. (96) and expanding for large a (properly accounting for the
1227
+ next-to-leading-order contributions), one finally obtains the asymptotic form of H(a), which reads
1228
+ H ∼ H∞ + H1
1229
+ οΏ½a0
1230
+ a
1231
+ οΏ½B
1232
+ .
1233
+ (101)
1234
+ Comparison to Eq. (29) shows that m = 1 and n = B, and the corresponding hierarchy of Ξ΄i’s is given by Eq.(63)
1235
+ since
1236
+ Ξ΄1,∞ = limaβ†’βˆž Λ™Ο†
1237
+ HβˆžΟ†βˆž
1238
+ = 0 ,
1239
+ (102)
1240
+ where Λ™Ο† is given by (85). One then obtains
1241
+ Ξ¦ = 3 βˆ’ 2B, ns βˆ’ 1 = 2B ,
1242
+ (103)
1243
+ and, when 0 < B < 3/2,
1244
+ ns βˆ’ 1 = 2B ,
1245
+ (104)
1246
+ which are the same results as GR with a minimally coupled inflaton. Indeed, in the a β†’ ∞ limit, the homogeneous
1247
+ inflaton is frozen at a certain value and one essentially recovers the evolution of the minimally coupled case, where
1248
+ β€œNewton’s constant” is now reproduced by the (constant) asymptotic value of the inflaton. Furthermore the a depen-
1249
+ dence of the solution is a consequence of the fact that the CR condition (84) is independent of the specific inflationary
1250
+ model, provided H∞ and Ο†βˆž are found to be (finite) constants.
1251
+
1252
+ 14
1253
+ C.
1254
+ Jordan and Einstein frame mapping
1255
+ In the previous section, we found the same asymptotic behaviour for the spectra in the minimally coupled case and
1256
+ in the IG case. This result was obtained in spite of the fact that CR condition is not frame invariant; i.e., the CR
1257
+ condition in the Einstein Frame (EF) is not mapped, in general, into a CR condition in the Jordan Frame (JF). In
1258
+ this section, we briefly review this statement and discuss its consequences.
1259
+ It is well known that, by a suitable conformal transformation and a redefinition of the scalar field (inflaton), one
1260
+ can map a minimally coupled theory (defined in the so-called EF) into a non-minimally coupled one (in the JF). In
1261
+ partcular, for IG, the mapping is given by the following transformation rules (see e.g. [14]):
1262
+ a(t) = MP
1263
+ βˆšΞΎΟƒ ˜a(t), N(t) = MP
1264
+ βˆšΞΎΟƒ
1265
+ ˜
1266
+ N(t) ,
1267
+ (105)
1268
+ and
1269
+ Ο† = MP
1270
+ οΏ½
1271
+ 1 + 6ΞΎ
1272
+ ΞΎ
1273
+ ln Οƒ
1274
+ Οƒ0
1275
+ , ˜V (Ο†(Οƒ)) = MP
1276
+ 2
1277
+ ΞΎ2Οƒ4 V (Οƒ) ,
1278
+ (106)
1279
+ where the tilde refers to the Einstein frame, Ο† is the scalar field in the EF, Οƒ is that in the JF. The mapping induces
1280
+ the following transformations of the Hubble parameter:
1281
+ ˜H = d˜a/dt
1282
+ ˜N˜a
1283
+ = (1 + Ξ΄1) MP
1284
+ βˆšΞΎΟƒ H ,
1285
+ (107)
1286
+ where
1287
+ H(t) = da(t)/dt
1288
+ a(t)N(t), Η«i+1 =
1289
+ dΗ«i/dt
1290
+ Η«iN(t)H(t), Ξ΄i+1 =
1291
+ dΞ΄i/dt
1292
+ Ξ΄iN(t)H(t)
1293
+ (108)
1294
+ are the Hubble and SR parameters in the JF. From the relation (105), one also finds that
1295
+ d
1296
+ d ln ˜a = (1 + Ξ΄1)βˆ’1
1297
+ d
1298
+ d ln a .
1299
+ (109)
1300
+ It is now straightforward to obtain the relations between SR parameters in the two frames:
1301
+ ˜ǫ1 ≑ βˆ’d ln ˜H
1302
+ d ln ˜a = βˆ’ (1 + Ξ΄1)βˆ’1
1303
+ d
1304
+ d ln a ln
1305
+ οΏ½
1306
+ (1 + Ξ΄1) MP
1307
+ βˆšΞΎΟƒ H
1308
+ οΏ½
1309
+ =
1310
+ Ξ΄1 + Η«1 βˆ’ Ξ΄1Ξ΄2
1311
+ 1+Ξ΄1
1312
+ 1 + Ξ΄1
1313
+ .
1314
+ (110)
1315
+ Given the relation (56), one then finds
1316
+ ˜ǫ1 = (1 + 6ξ)δ2
1317
+ 1
1318
+ 2ΞΎ(1 + Ξ΄1)2 .
1319
+ (111)
1320
+ From Eqs.
1321
+ (56) and (111), given that CR for a minimally coupled inflaton has ˜ǫ1,∞ = 0, one concludes that,
1322
+ correspondingly, in the JF one has Ξ΄1,∞ = 0 and Η«1,∞ = 0. If we differentiate Eq. (111), we obtain the following
1323
+ relations among other SR parameters in the two frames
1324
+ ˜ǫ2 =
1325
+ 2Ξ΄2
1326
+ (1 + Ξ΄1)2 ,
1327
+ (112)
1328
+ ˜ǫ3 = Ξ΄3 βˆ’ 2Ξ΄1Ξ΄2 + Ξ΄1Ξ΄3
1329
+ (1 + Ξ΄1)2
1330
+ ,
1331
+ (113)
1332
+ ˜ǫ4 =
1333
+ Ξ΄1Ξ΄2
1334
+ οΏ½
1335
+ 2Ξ΄2 βˆ’ 2Ξ΄1Ξ΄2 + 3Ξ΄3 + 3Ξ΄1Ξ΄3 βˆ’ (1 + Ξ΄1)2 Ξ΄3Ξ΄4
1336
+ οΏ½
1337
+ (1 + Ξ΄1)2 (2Ξ΄1Ξ΄2 βˆ’ Ξ΄1Ξ΄3 βˆ’ Ξ΄3)
1338
+ ,
1339
+ (114)
1340
+ and further ˜ǫi’s can be found by iterating the procedure but are useless for what follows.
1341
+
1342
+ 15
1343
+ Similarly, one can directly calculate the relations of the ˜δi’s with the dynamical variables in the JF:
1344
+ ˜δ1 ≑
1345
+ Λ™Ο†
1346
+ ˜N ˜HΟ†
1347
+ =
1348
+ οΏ½
1349
+ 1 + 6ΞΎ
1350
+ ΞΎ
1351
+ MP
1352
+ Ο†
1353
+ Ξ΄1
1354
+ 1 + Ξ΄1
1355
+ ,
1356
+ (115)
1357
+ and
1358
+ ˜δ2 ≑ d˜δ1/dt
1359
+ ˜N ˜H˜δ1
1360
+ = βˆ’ΛœΞ΄1 +
1361
+ Ξ΄2
1362
+ (1 + Ξ΄1)2 .
1363
+ (116)
1364
+ From the last relation and Eq. (56), one has that the CR condition in the EF,
1365
+ ˜δ2 + ˜δ1 βˆ’ ˜ǫ1 + B = 0 ,
1366
+ (117)
1367
+ is mapped into the following condition in the JF
1368
+ Ξ΄2 + (B βˆ’ 1) Ξ΄1 βˆ’ Η«1 + B = 0 .
1369
+ (118)
1370
+ Notice that only for B = 2 the CR condition is frame invariant. Nonetheless, both equations reduce to δ2,∞ = ˜δ2,∞ =
1371
+ βˆ’B at late times, and the evolution is indistinguishable, at least as far as the homogeneous degrees of freedom and
1372
+ the inflationary spectra are concerned.
1373
+ We conclude that, whereas the scalar spectral index ns βˆ’ 1 is frame invariant, Ξ¦ is generally not frame invariant.
1374
+ This can be checked directly by substitution. However, assuming CR holds in the EF, one verifies that Ξ¦ and ns βˆ’ 1
1375
+ are both frame invariant in the asymptotic regime.
1376
+ TABLE I. Results summary
1377
+ Inflation Asymptotic Growing
1378
+ Blue-Tilted
1379
+ Model
1380
+ Svolution
1381
+ Solution Spectral Index
1382
+ GR
1383
+ dS
1384
+ n > 3
1385
+ 0 < n < 3
1386
+ GR
1387
+ PL
1388
+ βˆ’
1389
+ βˆ’
1390
+ IG
1391
+ dS, δ1,∞ = 0
1392
+ n > 3/2
1393
+ 0 < n < 3/2
1394
+ IG
1395
+ dS, δ1,∞ ̸= 0
1396
+ -
1397
+ -
1398
+ CR+GR
1399
+ dS
1400
+ B > 3/2
1401
+ 0 < B < 3/2
1402
+ CR+IG
1403
+ dS, δ1,∞ = 0
1404
+ B > 3/2
1405
+ 0 < B < 3/2
1406
+ V.
1407
+ CONCLUSIONS
1408
+ In this article, we have analyzed the effects of different transient phases, which may occur during inflation due to a
1409
+ particularity of the inflaton potential, on the primordial inflationary spectrum of scalar perturbations. These transients
1410
+ have been studied in the last few years as sources of amplification of the amplitude of the curvature spectrum. It
1411
+ is important to notice that if the amplitude of scalar perturbations grows large enough, it may induce gravitational
1412
+ collapse and consequently seed the formation of primordial black holes after inflation ends. In the literature, several
1413
+ mechanisms for such an amplification during inflation have been proposed. In particular, the presence of an ultra
1414
+ slow-roll or, more generally, a constant-roll phase has been studied. Whereas in the former case the amplification is
1415
+ due to the existence of a growing solution to the equation of motion of the curvature perturbations, in the latter case
1416
+ the amplification can also be generated by a blue-tilted spectrum in absence of the growing solution.
1417
+ The purpose of this paper was precisely to examine general features of the aforementioned models starting from
1418
+ a rather generic ansatz for the Hubble parameter as a function of the scale factor.
1419
+ This general description of
1420
+ the transient phase is model independent, and many results obtained can be readily applied to several modified
1421
+ gravity models. The matter-gravity dynamics is described in terms of the hierarchies of SR parameters, both at the
1422
+ homogeneous level and at the level of perturbations. These hierarchies, when the transient phase that describes the
1423
+ approach to some inflationary attractor is considered, have been shown to take a peculiar form wherein either odd or
1424
+ even terms of the hierarchy are null and the remaining ones are different for zero. This general feature is a peculiarity
1425
+ of the asymptotic form of the SR parameters close to the attractor, and it is then used as a simplifying assumption
1426
+
1427
+ 16
1428
+ throughout the entire article. The resulting hierarchies, in the large-a limit and for the cases considered, were used
1429
+ to calculate the behavior of the primordial curvature spectrum as the parametrisation of H(a) was varied. Then,
1430
+ when possible, the corresponding inflaton potential was fully reconstructed. An overview of the spectra enhancement
1431
+ results was presented in Table (I).
1432
+ For simplicity, only the induced gravity case has been considered here as a generalisation of general relativity with
1433
+ a minimally coupled inflaton. Induced gravity is particularly relevant since both Higgs and Starobinsky inflationary
1434
+ models (which are in good agreement with observations) take place in the β€˜induced gravity phase’. We note that while
1435
+ transient evolutions that have the de Sitter universe as a limit (such as USR) can lead to an amplification, the results
1436
+ differ when power-law inflation is considered as the limit of a transitory dynamics and, for the cases we were able to
1437
+ solve explicitly, no modification of the scalar spectrum was obtained. Finally, the constant-roll case was discussed in
1438
+ more detail as an application of the preceding results, and the issue of the transition from the Einstein frame to the
1439
+ Jordan frame was also scrutinized.
1440
+ [1] G. F. Chapline, Nature 253 (1975) no.5489, 251-252 doi:10.1038/253251a0; P. Meszaros, Astron. Astrophys. 38 (1975),
1441
+ 5-13
1442
+ [2] A.A. Starobinsky. Springer. in H.J. De Vega and N. Sanchez (eds.) Current trends in field theory quantum gravity and
1443
+ strings, Lecture Notes in Physics 246 Verlag, Heidelberg, 1986), pp. 107-126. A.D. Linde. Academic. Particle Physics and
1444
+ Inflationary Cosmology (Harwood New York, 1990).
1445
+ [3] B. Carr and F. Kuhnel, Ann. Rev. Nucl. Part. Sci. 70 (2020), 355-394 doi:10.1146/annurev-nucl-050520-125911
1446
+ [4] J. Garcia-Bellido and E. Ruiz Morales, Phys. Dark Univ. 18 (2017) 47 doi:10.1016/j.dark.2017.09.007 [arXiv:1702.03901
1447
+ [astro-ph.CO]]. G. Ballesteros and M. Taoso, Phys. Rev. D 97 (2018) no.2, 023501 doi:10.1103/PhysRevD.97.023501
1448
+ [arXiv:1709.05565 [hep-ph]]. M. Cicoli, V. A. Diaz and F. G. Pedro, JCAP 1806 (2018) no.06, 034 doi:10.1088/1475-
1449
+ 7516/2018/06/034 [arXiv:1803.02837 [hep-th]]. H. Motohashi and W. Hu, Phys. Rev. D 96 (2017) no.6, 063503
1450
+ doi:10.1103/PhysRevD.96.063503 [arXiv:1706.06784 [astro-ph.CO]]. C. Germani and T. Prokopec, Phys. Dark Univ. 18
1451
+ (2017) 6 doi:10.1016/j.dark.2017.09.001 [arXiv:1706.04226 [astro-ph.CO]].
1452
+ [5] V.F. Mukhanov, Sov. Phys. JETP 68, 1297 (1988); J. M. Maldacena, JHEP 0305 (2003) 013; V. F. Mukhanov,
1453
+ H. A. Feldman and R. H. Brandenberger, Phys. Rept. 215 (1992) 203. V.F. Mukhanov, Phys. Lett. B 218, 17 (1989);
1454
+ J. M. Bardeen, Phys. Rev. D 22, 1882 (1980). doi:10.1103/PhysRevD.22.1882; M. Sasaki, Prog. Theor. Phys. 70 (1983)
1455
+ 394. doi:10.1143/PTP.70.394
1456
+ [6] A. Y. Kamenshchik, A. Tronconi and G. Venturi, JCAP 01 (2022) no.01, 051 doi:10.1088/1475-7516/2022/01/051
1457
+ [arXiv:2110.08112 [gr-qc]] A. Y. Kamenshchik, A. Tronconi, T. Vardanyan and G. Venturi, Phys. Lett. B 791 (2019),
1458
+ 201-205 doi:10.1016/j.physletb.2019.02.036 [arXiv:1812.02547 [gr-qc]] H. Motohashi, S. Mukohyama and M. Oliosi, JCAP
1459
+ 03 (2020), 002 doi:10.1088/1475-7516/2020/03/002 [arXiv:1910.13235 [gr-qc]]
1460
+ [7] H. Motohashi,
1461
+ A. A. Starobinsky and J. Yokoyama,
1462
+ JCAP 09 (2015),
1463
+ 018 doi:10.1088/1475-7516/2015/09/018
1464
+ [arXiv:1411.5021 [astro-ph.CO]]
1465
+ [8] A.Yu. Kamenshchik, U. Moschella and V. Pasquier, Phys. Lett. B 511 (2001), 265 [arXiv: gr-qc/0103004]
1466
+ [9] A.D. Sakharov, Sov. Phys. Dokl. 12 (1968), 1040; S.L. Adler, Rev. Mod. Phys. 54 (1982) 729.
1467
+ [10] A. Cerioni, F. Finelli, A. Tronconi and G. Venturi, Phys. Rev. D 81 (2010), 123505 doi:10.1103/PhysRevD.81.123505
1468
+ [arXiv:1005.0935 [gr-qc]]
1469
+ [11] H. Motohashi and A. A. Starobinsky, EPL 117 (2017) no.3, 39001 doi:10.1209/0295-5075/117/39001 [arXiv:1702.05847
1470
+ [astro-ph.CO]]
1471
+ [12] P. A. R. Ade et al. [Planck Collaboration], arXiv:1502.02114 [astro-ph.CO]. P. A. R. Ade et al. [Planck Collaboration],
1472
+ arXiv:1502.01589 [astro-ph.CO]. Y. Akrami et al. [Planck Collaboration], arXiv:1807.06211 [astro-ph.CO]. N. Aghanim et
1473
+ al. [Planck Collaboration], arXiv:1807.06209 [astro-ph.CO].
1474
+ [13] H. Motohashi and A. A. Starobinsky, JCAP 11 (2019), 025 doi:10.1088/1475-7516/2019/11/025 [arXiv:1909.10883 [gr-qc]]
1475
+ [14] A. Y. Kamenshchik, E. O. Pozdeeva, A. Tronconi, G. Venturi and S. Y. Vernov, Class. Quant. Grav. 31 (2014), 105003
1476
+ doi:10.1088/0264-9381/31/10/105003 [arXiv:1312.3540 [hep-th]]
1477
+
JNE3T4oBgHgl3EQfXQpr/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
JNFJT4oBgHgl3EQfwC0U/content/tmp_files/2301.11628v1.pdf.txt ADDED
@@ -0,0 +1,1414 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Astronomy & Astrophysics manuscript no. bt_spidi2
2
+ Β©ESO 2023
3
+ January 30, 2023
4
+ Spectroscopic and interferometric signatures of magnetospheric
5
+ accretion in young stars
6
+ B. Tessore1, A. Soulain1, G. Pantolmos1, J. Bouvier1, C. Pinte1, 2, and K. Perraut1
7
+ 1 UniversitΓ© Grenoble Alpes, CNRS, IPAG, 38000 Grenoble, France
8
+ 2 School of Physics and Astronomy, Monash University, VIC 3800, Australia
9
+ xx/xx/xx; yy/yy/yy
10
+ ABSTRACT
11
+ Aims. We aim to assess the complementarity between spectroscopic and interferometric observations in the characterisation of the
12
+ inner star-disc interaction region of young stars.
13
+ Methods. We use the code MCFOST to solve the non-LTE problem of line formation in non-axisymmetric accreting magnetospheres.
14
+ We compute the BrΞ³ line profile originating from accretion columns for models with different magnetic obliquities. We also derive
15
+ monochromatic synthetic images of the Brγ line emitting region across the line profile. This spectral line is a prime diagnostics of
16
+ magnetospheric accretion in young stars and is accessible with the long baseline near-infrared interferometer GRAVITY installed at
17
+ the ESO Very Large Telescope Interferometer.
18
+ Results. We derive Brγ line profiles as a function of rotational phase and compute interferometric observables, visibilities and phases,
19
+ from synthetic images. The line profile shape is modulated along the rotational cycle, exhibiting inverse P Cygni profiles at the time
20
+ the accretion shock faces the observer. The size of the line’s emission region decreases as the magnetic obliquity increases, which is
21
+ reflected in a lower line flux. We apply interferometric models to the synthetic visibilities in order to derive the size of the line-emitting
22
+ region. We find the derived interferometric size to be more compact than the actual size of the magnetosphere, ranging from 50 to
23
+ 90% of the truncation radius. Additionally, we show that the rotation of the non-axisymmetric magnetosphere is recovered from the
24
+ rotational modulation of the BrΞ³-to-continuum photo-centre shifts, as measured by the differential phase of interferometric visibilities.
25
+ Conclusions. Based on the radiative transfer modelling of non-axisymmetric accreting magnetospheres, we show that simultaneous
26
+ spectroscopic and interferometric measurements provide a unique diagnostics to determine the origin of the BrΞ³ line emitted by young
27
+ stellar objects and are ideal tools to probe the structure and dynamics of the star-disc interaction region.
28
+ Key words. Radiative transfer – Line: profiles – Stars: variables: T Tauri, Herbig Ae/Be – Accretion, accretion disks
29
+ 1. Introduction
30
+ The early evolution of low mass stars (Mβˆ— < 2 MβŠ™) during the
31
+ classical T Tauri (CTT) phase depends on the interaction be-
32
+ tween the star and its accretion disc, on a distance of a few stel-
33
+ lar radii. At the truncation radius, matter from the disc surface is
34
+ channelled onto the stellar surface following the magnetic field
35
+ lines and forming an accretion funnel or column (Ghosh et al.
36
+ 1977; Zanni & Ferreira 2009; Romanova & Owocki 2016; Pan-
37
+ tolmos et al. 2020). The star-disc interaction is responsible for
38
+ accretion and ejection phenomena that have a strong impact on
39
+ spectral lines formed in the close vicinity of the star’s surface.
40
+ Ghosh et al. (1977) developed an analytical model of mag-
41
+ netospheric accretion around a rotating neutron star with a dipo-
42
+ lar magnetic field. Hartmann et al. (1994) applied this magne-
43
+ tospheric accretion model to the formation of emission lines in
44
+ the spectrum of T Tauri stars. This fundamental paper sets the
45
+ general theoretical framework for the density and temperature
46
+ distributions in aligned axisymmetric magnetospheres. The cou-
47
+ pling between this representation of magnetospheric accretion in
48
+ T Tauri systems with radiative transfer calculations has provided
49
+ a crucial tool to interpret spectroscopic, photometric, and inter-
50
+ ferometric observations. The sensitivity of hydrogen lines to the
51
+ parameters of the magnetospheric models was studied in detail
52
+ by Muzerolle et al. (2001), improving the earlier calculations by
53
+ Hartmann et al. (1994).
54
+ Near-infrared observations of the Brackett Ξ³ (BrΞ³) line with
55
+ the Very Large Telescope Interferometer (VLTI) GRAVITY in-
56
+ strument (Gravity Collaboration et al. 2017) also probe the inner
57
+ part of the star-disc interaction region (Gravity Collaboration
58
+ et al. 2020; Bouvier et al. 2020a). However, it is still difficult
59
+ to associate the characteristic sizes derived from interferometry
60
+ with the actual size of the magnetospheric accretion region, a
61
+ key parameter in our understanding of the star-disc interaction.
62
+ In this paper, we aim at studying the formation of the BrΞ³
63
+ line and compute its spectroscopic and interferometric signa-
64
+ tures for non-axisymmetric models of the inner star-disc inter-
65
+ action region, akin to state-of-art MHD simulations (Romanova
66
+ & Owocki 2016). In particular, we want to clarify the meaning
67
+ of the sizes inferred through near-infrared interferometric obser-
68
+ vations and how they compare with the overall size of the mag-
69
+ netospheric accretion region.
70
+ In sections Β§2 and Β§3, we describe the model used to com-
71
+ pute the line formation in accreting magnetospheres. We discuss
72
+ spectroscopic and interferometric signatures in sections Β§4 and
73
+ Β§5, respectively.
74
+ 2. Radiative transfer framework
75
+ We use the code MCFOST 1(Pinte et al. 2006, 2009; Tessore
76
+ et al. 2021) to compute emergent line fluxes from multidimen-
77
+ 1 https://github.com/cpinte/mcfost
78
+ Article number, page 1 of 13
79
+ arXiv:2301.11628v1 [astro-ph.SR] 27 Jan 2023
80
+
81
+ A&A proofs: manuscript no. bt_spidi2
82
+ sional models of magnetospheres for a 20-level hydrogen atom.
83
+ The atomic model, with 19 bound levels and the ground state of
84
+ HII, consists of 171 bound-bound transitions (atomic lines) and
85
+ 19 bound-free transitions (continua). We focus here on the BrΞ³
86
+ line at 2.1661 Β΅m although, the Balmer lines HΞ± and HΞ² and
87
+ the Paschen β line (Paβ) are modelled as well. These specific
88
+ hydrogen lines are commonly used to characterise accretion and
89
+ ejection phenomena in young systems (Folha & Emerson 2001;
90
+ Alencar et al. 2012; Bouvier et al. 2020a; Pouilly et al. 2020;
91
+ Sousa et al. 2021). The method to solve for the non-LTE pop-
92
+ ulations of hydrogen and the microphysics are the same as in
93
+ Tessore et al. (2021). The updated version of the code we use
94
+ now simultaneously solves the charge equation and the statis-
95
+ tical equilibrium equations, which has been proven to increase
96
+ the convergence in chromospheric conditions (Leenaarts et al.
97
+ 2007). We tested our code for different magnetospheric models
98
+ taken as benchmarks in Muzerolle et al. (2001) and Kurosawa
99
+ et al. (2006). The results of this comparison are presented and
100
+ discussed in Appendix A.
101
+ 3. Magnetospheric accretion model
102
+ Matter from the circumstellar disc is channelled onto the stel-
103
+ lar surface along the dipolar magnetic field lines. The stellar
104
+ magnetic field truncates the disc at a distance Rt from the star,
105
+ the truncation radius. In practice, the interaction between the
106
+ stellar magnetic field and the disc takes place over a small re-
107
+ gion between Rt and Rt + Ξ΄r. Both Rt and Ξ΄r are used to de-
108
+ fine the size of the disc region magnetically connected to the
109
+ star. As the gas approaches the stellar surface, it decelerates in
110
+ a shock and is heated at coronal temperatures. Theoretical mod-
111
+ els of accretion shocks by Calvet & Gullbring (1998) show that
112
+ the optically thin emission of the pre/post-shock dominates be-
113
+ low the Balmer jump and that the optically thick emission of
114
+ the heated photosphere contributes to the total continuum emis-
115
+ sion at larger wavelengths. In the following, we only consider
116
+ the contribution of the heated photosphere to the shock radia-
117
+ tion. The shock2 temperature is computed from the energy of the
118
+ gas infalling onto the stellar surface following the prescription
119
+ of Romanova et al. (2004) unless specified. This approach as-
120
+ sumes energy conservation and that the shock radiates as a black
121
+ body, meaning that its temperature is determined by the specific
122
+ kinetic energy and enthalpy of the gas deposited at the stellar
123
+ surface. The shock temperature hence derived is of the order of
124
+ 4500 K - 6000 K.
125
+ 3.1. The stellar surface
126
+ The stellar surface is considered as the inner boundary of the
127
+ model and emits as a blackbody whose temperature is deter-
128
+ mined by the stellar parameters. Throughout the paper, the stellar
129
+ parameters are Tβˆ— = 4, 000 K, Mβˆ— = 0.5 MβŠ™, and Rβˆ— = 2 RβŠ™. We
130
+ set the distance to the star at 140 pc, which is typical of the near-
131
+ est star forming regions such as Upper Scorpius (β‰ˆ 146 pc Galli
132
+ et al. 2018a) or Taurus (β‰ˆ 130 pc Galli et al. 2018b).
133
+ 3.2. Geometry of the accretion funnels
134
+ We consider 3D non-axisymmetric models of the magneto-
135
+ spheric accretion region. These models are parametrised by the
136
+ same set of parameters as the axisymmetric magnetospheric
137
+ 2 We assume that the shock region is unresolved and is part of the
138
+ stellar surface.
139
+ Fig. 1: Density distribution of a non-axisymmetric model with an
140
+ obliquity of 10β—¦. The rotation axis of the star Ω is shown with a
141
+ white arrow and the dipole axis, Β΅, with a red arrow. The density
142
+ is computed from Eqs. (1) and (2). The colour map scales with
143
+ the density.
144
+ model of Hartmann et al. (1994) (see also Muzerolle et al. 1998,
145
+ 2001; Kurosawa et al. 2006; Lima et al. 2010; Kurosawa et al.
146
+ 2011; Dmitriev et al. 2019).The density and the velocity fields of
147
+ the accretion columns are fully described with a set of indepen-
148
+ dent parameters: the mass accretion rate Λ™M, the rotation period
149
+ Prot, Rt, and Ξ΄r.
150
+ For our study, the value of Λ™M, Rt and Ξ΄r, and of the temper-
151
+ ature of the magnetosphere are fixed. The impact of these pa-
152
+ rameters on the line formation has been discussed thoroughly in
153
+ Muzerolle et al. (1998, 2001, see also App. A). The line’s re-
154
+ sponse to the mass accretion rate and to the temperature is an es-
155
+ sential proxy for understanding the physics of the star-disc inter-
156
+ action region. We use a mass accretion rate Λ™M = 10βˆ’8 MβŠ™ yrβˆ’1,
157
+ a truncation radius Rt = 4 Rβˆ—, and Ξ΄r = 1 Rβˆ—. The value of the
158
+ rotation period is deduced from the maximum truncation ra-
159
+ dius (Rt + Ξ΄r), imposing that stable accretion occurs at 90% of
160
+ the corotation radius, consistent with the work of Blinova et al.
161
+ (2016). The rotation period is therefore fixed at Prot = 6 days,
162
+ corresponding to slowly rotating T Tauri stars (see Herbst et al.
163
+ 2007; Bouvier et al. 2014, for a review). The rotational velocity
164
+ for that period is thus of the order of 80 km sβˆ’1 at the outer edge
165
+ of the magnetosphere.
166
+ When the magnetic field axis (¡) is misaligned with respect
167
+ to the rotational axis (Ω) of the star, the geometry of the accre-
168
+ tion flow changes dramatically. The equations for the magnetic
169
+ field components of a non-axisymmetric dipole, i.e. with a non-
170
+ zero obliquity, are provided in Mahdavi & Kenyon (1998). The
171
+ parameter Ξ²ma describes the angle between the dipole moment
172
+ and the star’s rotational axis, the magnetic obliquity.
173
+ We approximate the density, ρ, along the non-axisymmetric
174
+ magnetic field lines with,
175
+ ρ = α B
176
+ � = αB ρaxi
177
+ Baxi
178
+ ,
179
+ (1)
180
+ where α is a constant along a given field line and B the ana-
181
+ lytic misaligned dipolar field. �, ρaxi, and Baxi denote the veloc-
182
+ ity field, density, and dipolar magnetic field, respectively, and
183
+ they are taken from the axisymmetric model of Hartmann et al.
184
+ (1994). In other words, the 3D density structure is computed
185
+ from Eq. (1) under the assumption that the infalling gas has a
186
+ velocity field on the poloidal plane. The value of α is computed
187
+ Article number, page 2 of 13
188
+
189
+ 1.2e-08
190
+ 5e-9
191
+ [s-'b]
192
+ 2e-9
193
+ 1e-9
194
+ 5e-10
195
+ 2.3e-10
196
+ RtB. Tessore & A. Soulain et al.: Spectroscopic and interferometric signatures of magnetospheric accretion
197
+ from the numerical integration 3 of the mass flux over the shock
198
+ area,
199
+ Λ™M =
200
+ οΏ½
201
+ ρv · dS,
202
+ (2)
203
+ where dS is the surface element and v the velocity field. In our
204
+ model, the value of Λ™M is an input parameter and is held constant.
205
+ Therefore, Ξ± is obtained to ensure consistency between Eqs. (1)
206
+ and (2). We compute five models with an obliquity βma ranging
207
+ from five to forty degrees in step of ten degrees, representative
208
+ of what has been measured for T Tauri stars with spectroscopy
209
+ (McGinnis et al. 2020) and spectropolarimetry (Donati et al.
210
+ 2008, 2010, 2013; Johnstone et al. 2014; Pouilly et al. 2020). For
211
+ these non-axisymmetric models, the shortest field lines – defin-
212
+ ing the main accretion columns4 – carry most of the gas density.
213
+ We remove the longest field lines – the secondary columns –
214
+ in our modelling as in Esau et al. (2014). This yields models
215
+ with one crescent-shaped accretion spot per stellar hemisphere
216
+ reminiscent of numerical simulations of misaligned dipoles (Ro-
217
+ manova et al. 2003).
218
+ Figure 1 shows the density of a non-axisymmetric magneto-
219
+ sphere with an obliquity of 10β—¦.
220
+ 3.3. Temperature of the funnels
221
+ The temperature of the magnetospheric accretion region is not
222
+ well constrained. The determination of the temperature by Mar-
223
+ tin (1996) from first principles was not able to reproduce the
224
+ observations. A self-consistent calculation of the temperature of
225
+ the magnetosphere is beyond the scope of this paper. Instead,
226
+ we adopt here the temperature profile of Hartmann et al. (1994),
227
+ which has been extensively used in the past to model line fluxes
228
+ from accreting T Tauri stars. The temperature is computed using
229
+ a volumetric heating rate (∝ rβˆ’3) and balancing the energy input
230
+ with the radiative cooling rates of Hartmann et al. (1982). The
231
+ exact balance between the heating and cooling mechanisms is
232
+ unknown. Instead, the temperature profile is normalised to a free
233
+ parameter, Tmax, that sets the value of the maximum temperature
234
+ in the funnel flow. In the following, we have set the temperature
235
+ maximum to Tmax = 8, 000 K.
236
+ 4. Spectroscopic signatures
237
+ Thanks to the Doppler shift of the funnel flow, it is possible
238
+ to reconstruct the origin of the emission line by looking at the
239
+ brightness maps in various velocity channels. Figure 2 shows
240
+ the contribution of the different parts of the magnetosphere to
241
+ the total integrated BrΞ³ line flux at a given velocity for an in-
242
+ clination of 30β—¦, matching the model illustrated in Fig. 1. At
243
+ those density and temperature, the continuum emission comes
244
+ from the stellar surface (Isurf/Imag > 100). Locally, the contin-
245
+ uum emission from the shock is three times larger than the emis-
246
+ sion from the star. Overall, given the small covering area of the
247
+ accretion shock (around 1%), the total continuum emission at the
248
+ frequency of the BrΞ³ line is dominated by the star’s radiation,
249
+ Fshock/Fβˆ— = 3%. The low-velocity components (< 50 km sβˆ’1) of
250
+ the line form in the regions where the projected velocity along
251
+ the line-of-sight is close to zero and near the disc. The geometry
252
+ 3 For this 3D magnetospheric accretion model, an explicit formula for
253
+ the shock area does not exist (see also Mahdavi & Kenyon 1998)
254
+ 4 Geometrically, the shortest field lines obey the following criterion
255
+ cos Ο†β€² Γ— z > 0 where Ο†β€² is the azimuth in the frame aligned with the
256
+ dipole axis and z the coordinate parallel to the rotation axis.
257
+ of the non-axisymmetric model, defined in §3, is responsible for
258
+ a rotational modulation of the integrated line flux. Many classical
259
+ T Tauri stars shows modulated photometric variability (e.g Cody
260
+ et al. 2014) and, more directly related to the magnetospheric re-
261
+ gion, many also show rotational modulation of the longitudinal
262
+ component of the stellar magnetic field (e.g Donati et al. 2020).
263
+ Indeed, the periodic variability of optical and emission line pro-
264
+ files has been reported in various systems (for instance Sousa
265
+ et al. 2016; Alencar et al. 2018; Bouvier et al. 2020a), which in-
266
+ dicates that the emission region is stable on a timescale of several
267
+ rotation periods.
268
+ Figure 3 shows the variability of the BrΞ³ line at different
269
+ phases of rotation at an inclination of 60β—¦ for different obliq-
270
+ uities. The origin of the rotational phase is defined such that at
271
+ phase 0.5, the accretion shock is facing the observer. The red-
272
+ shifted absorption seen for the BrΞ³ line at phases 0.250, 0.47 and
273
+ 0.69, results from a lower source function of the gas above the
274
+ shock (see App. A). From observations, red-shifted absorption
275
+ in the PaΞ² and BrΞ³ lines are seen in less than 34% and 20% of
276
+ the line profiles, respectively (Folha & Emerson 2001). The in-
277
+ verse P Cygni profile disappears when the shock, or a significant
278
+ fraction of it, is hidden on the opposite side of the star. The line,
279
+ with either a double-peaked profile or a moderate red-shifted ab-
280
+ sorption, is reminiscent of Reipurth et al. (1996) cases II and IV.
281
+ While the profiles with redshifted absorption agree with observa-
282
+ tions, those that display an M-shape are usually not observed in
283
+ young stellar objects. This suggests that magnetospheric accre-
284
+ tion is not the only contribution to the profile, which can also be
285
+ impacted by various types of outflows (e.g., stellar, interface, and
286
+ disk winds Lima et al. 2010; Kurosawa et al. 2011). The optically
287
+ thick accretion disc is not included in our models. The effect of
288
+ the disc emission and absorption on the spectroscopic and inter-
289
+ ferometric observables will be discussed in a subsequent paper.
290
+ We also observe a decrease of the line flux as the obliquity
291
+ increases. Figure 4 shows the radius encompassing 90% of the
292
+ total line flux, R90, at an inclination of 60β—¦ for non-axisymmetric
293
+ models with different obliquities for the HΞ±, HΞ², PaΞ² and BrΞ³
294
+ lines.
295
+ 5
296
+ 10
297
+ 15
298
+ 20
299
+ 25
300
+ 30
301
+ 35
302
+ 40
303
+ ma [ ]
304
+ 2.8
305
+ 3.0
306
+ 3.2
307
+ 3.4
308
+ 3.6
309
+ 3.8
310
+ R90 [R ]
311
+ H
312
+ H
313
+ Pa
314
+ Br
315
+ Fig. 4: Radius encompassing 90% of the total flux (R90) for each
316
+ line as a function of the obliquity, Ξ²ma. Hydrogen lines are la-
317
+ belled with different colours.
318
+ As Ξ²ma increases, the volume of the magnetospheric accre-
319
+ tion region decreases because the arc length of the accreting field
320
+ lines shortens. Therefore, the total flux, for all lines, decreases
321
+ accordingly, independently of the viewing angle of the system.
322
+ However, we also note a dependence of R90 with the line. The
323
+ Article number, page 3 of 13
324
+
325
+ A&A proofs: manuscript no. bt_spidi2
326
+ II
327
+ III
328
+ IV
329
+ I
330
+ 200
331
+ 0
332
+ 200
333
+ v [km. s
334
+ 1]
335
+ 1.0
336
+ 1.2
337
+ 1.4
338
+ F/Fc
339
+ I
340
+ II
341
+ III
342
+ IV
343
+ V
344
+ V
345
+ 0%
346
+ 5%
347
+ 30%
348
+ 50%
349
+ 80%
350
+ Fig. 2: Origin of the emission seen across the Brackett Ξ³ line. The contribution of individual images to the total line flux is indicated
351
+ on the central image showing the line profile. The brightness maps are in units of the maximum emission. The emission of the stellar
352
+ surface is saturated. Orange to red colours indicates the regions of maximum emission. The system is seen at an inclination of 30β—¦
353
+ and an rotational phase of ∼0.25, similar to Fig. 1.
354
+ value of R90 represents the size of the emitting region in a given
355
+ line, which is a function of density and temperature, and of the
356
+ viewing angle.
357
+ 5. Interferometric signatures
358
+ In this section, we compute the size of the BrΞ³ line-emitting re-
359
+ gion inferred from interferometric observations, and we compare
360
+ it to model flux radii (see Β§4).
361
+ 5.1. Interferometric observables
362
+ The interferometric observables are derived from the radiative
363
+ transfer (RT) model using the ASPRO25 software developed by
364
+ the Jean-Marie Mariotti Center (JMMC). These observables rep-
365
+ resent what would be observed with GRAVITY in the near-
366
+ infrared. We consider the configuration obtained with the Very
367
+ Large Telescope (i.e. 4x8m telescopes), encompassing a range
368
+ of baselines from 35 to 135 m. With a typical night of 8 hours,
369
+ we compute one observing point per hour for the six baselines
370
+ 5 Available at https://www.jmmc.fr
371
+ of the VLTI to increase the Fourier sampling, namely u-v cov-
372
+ erage, which is crucial for the fitting part of our approach. As
373
+ described in Bourgès & Duvert (2016), we derive the observ-
374
+ ables from the RT images (see Fig. 2) by computing the com-
375
+ plex visibility in each spectral channel around the BrΞ³ line and
376
+ interpolating them to match GRAVITY’s spectral resolution (R
377
+ = 4000). Specifically, we simulate a total of 37 spectral chan-
378
+ nels (from 2.161 to 2.171 Β΅m with a step of 2.8 10βˆ’4 Β΅m) for the
379
+ six projected baselines repeated eight times. Within this range,
380
+ 31 spectral channels are used to measure the K-band continuum
381
+ and six channels sample the BrΞ³ line emitting region.
382
+ Figure B.1 illustrates the resulting u-v plane projected on-
383
+ sky for a typical object observed at the VLTI with a declination
384
+ of -34β—¦ (e.g. TW Hydrae).
385
+ Figure 5 shows the interferometric observables along the ro-
386
+ tational cycle for a model with an inclination of 60β—¦ and an obliq-
387
+ uity of 10β—¦. The two main observables are: the modulus of the
388
+ complex visibility – the visibility amplitude – and the differen-
389
+ tial phase – its argument – dispersed in wavelength. The phase
390
+ is normalised to zero in the continuum. The visibility amplitude
391
+ can then be used to estimate the object’s size, while the phase
392
+ measures the photo-centre shifts between the line-emitting re-
393
+ Article number, page 4 of 13
394
+
395
+ B. Tessore & A. Soulain et al.: Spectroscopic and interferometric signatures of magnetospheric accretion
396
+ 250
397
+ 125
398
+ 0
399
+ 125
400
+ 250
401
+ 0.8
402
+ 0.9
403
+ 1.0
404
+ 1.1
405
+ 1.2
406
+ 1.3
407
+ 1.4
408
+ phase = 0.03
409
+ 250
410
+ 125
411
+ 0
412
+ 125
413
+ 250
414
+ phase = 0.25
415
+ 250
416
+ 125
417
+ 0
418
+ 125
419
+ 250
420
+ phase = 0.47
421
+ 250
422
+ 125
423
+ 0
424
+ 125
425
+ 250
426
+ phase = 0.69
427
+ 250
428
+ 125
429
+ 0
430
+ 125
431
+ 250
432
+ phase = 0.92
433
+ ma = 5
434
+ ma = 10
435
+ ma = 20
436
+ ma = 30
437
+ ma = 40
438
+ v [km s
439
+ 1]
440
+ F/Fc
441
+ Fig. 3: Brackett γ line variability along the rotational cycle. Each column corresponds to a specific rotational phase. At phase 0,
442
+ the shock area is unseen on the stellar surface, while phase of 0.5, the shock is fully seen on the visible hemisphere. The colours
443
+ correspond to different values of the obliquity. All fluxes are computed with an inclination of 60β—¦.
444
+ gion and the continuum. The phase can only be used as a rela-
445
+ tive measurement (e.g. between the line and the continuum), the
446
+ absolute phase being lost due to a combination of atmospheric
447
+ and instrumental effects. We repeat the simulated observations
448
+ and compute nine datasets over a rotational cycle sampled every
449
+ 40 degrees ( 0.11 in phase). In this study, we are interested in the
450
+ line’s emitting region only. Therefore, we use pure line quanti-
451
+ ties, instead of total visibilities and phases, to remove the contri-
452
+ bution from the stellar surface (see appendix B for the derivation
453
+ of the pure line interferometric quantities).
454
+ 5.2. Physical characteristic and sizes
455
+ Once the interferometric observables are computed, we apply
456
+ standard modelling methods to interpret the data (Berger 2003).
457
+ Firstly, we average the visibility amplitude of the six spectral
458
+ channels available within the BrΞ³ line6. We use the average vis-
459
+ ibilities to recover the global size of the BrΞ³ emitting region,
460
+ where the different velocities probe specific parts of the mov-
461
+ ing material within the magnetosphere. Then, we fit the aver-
462
+ aged visibility amplitude using elongated Gaussian or uniform
463
+ disc models. Such models are typically used in interferometry to
464
+ estimate the system’s characteristic size and on-sky orientation.
465
+ The source’s brightness distribution is defined by its half-flux ra-
466
+ dius in the case of a Gaussian disc or its radius for the uniform
467
+ disc model, and an elongation factor and a position angle. In the
468
+ following, we adopt the definition of "radius" for both models,
469
+ which corresponds to the half-flux semi-major axis for the Gaus-
470
+ sian model and the semi-major axis for the uniform disc model.
471
+ The recovered sizes and orientations are represented in the top
472
+ panel of Fig. 5. While neither model can fully account for the
473
+ size of the magnetosphere, the uniform disc probes a larger area
474
+ of the magnetosphere, while the Gaussian disc seems limited to
475
+ the most luminous parts. We note that the fit of the visibility is
476
+ 6 Five and four spectral channels only were used at phase 0.25 and
477
+ 0.47, respectively, due to a limited line-to-continuum ratio (see Ap-
478
+ pendix B for details).
479
+ equally good for both models and, thus, does not allow us to dis-
480
+ criminate between the models from the synthetic visibilities only
481
+ (middle-top, Fig. 5).
482
+ In order to quantify the physical meaning of the interfero-
483
+ metric measurements, we compare the interferometric sizes with
484
+ reference flux radii of the RT models. We set these radii to repre-
485
+ sent 50, 80, 90, and 99% of the total flux emitted by the magne-
486
+ tospheric accretion region. Figure 6 compares the sizes derived
487
+ with interferometry to the characteristic radii of the RT models.
488
+ We find that the size derived from the uniform disc model is
489
+ modulated around an average value of 3.5 Rβˆ— corresponding to
490
+ 90% of the BrΞ³ emitting region. The size obtained by interferom-
491
+ etry appears to be modulated by the position of the funnel flows
492
+ close to the star, with a minimum located around phase 0.8. The
493
+ Gaussian model exhibits the same modulation but with a lower
494
+ amplitude (2.1 Β± 0.4 Rβˆ—) and appears sensitive to the magneto-
495
+ sphere’s innermost region, close to the 50% flux radius. The size
496
+ derived from the uniform disc model emerges as being the most
497
+ appropriate to recover the reference model size, accounting for
498
+ at least 80% of the total flux emitted by the magnetosphere.
499
+ Article number, page 5 of 13
500
+
501
+ A&A proofs: manuscript no. bt_spidi2
502
+ 0.0
503
+ 0.2
504
+ 0.4
505
+ 0.6
506
+ 0.8
507
+ Phase
508
+ 2.0
509
+ 2.5
510
+ 3.0
511
+ 3.5
512
+ 4.0
513
+ 4.5
514
+ 5.0
515
+ Radius [R
516
+ ]
517
+ Rt +
518
+ r
519
+ 50%
520
+ 80%
521
+ 90%
522
+ 99%
523
+ Uniform disc
524
+ Gaussian disc
525
+ Fig. 6: Interferometric radii as a function of the rotational phase.
526
+ Uniform and Gaussian disc models are shown with green and
527
+ yellow markers, respectively. Blue lines correspond to the radii
528
+ encompassing 50, 80, 90 and 99% of the total RT model’s flux.
529
+ The blue shaded areas represent the standard deviation of these
530
+ radii across the rotational phase.
531
+ The red shaded area indicates the inner (Rt) and outer radius (Rt+
532
+ Ξ΄r) of the RT model.
533
+ The derived orientations obtained from interferometry seem
534
+ to be particularly representative of the position of the accretion
535
+ funnel flow and the on-sky orientation of the BrΞ³ emitting re-
536
+ gion (Fig. 5). The measured position angle agrees with the mag-
537
+ netosphere’s orientation, particularly when the shock faces the
538
+ observer (phase = 0.5). Nevertheless, it appears somewhat haz-
539
+ ardous to decipher the shape and orientation of the emitting re-
540
+ gion across the rotational cycle from this observable only, as
541
+ different magnetospheric configurations can be described by a
542
+ very similar interferometric model (e.g. phases 0.03 and 0.25).
543
+ A stronger constraint on the orientation of the funnel flows arises
544
+ from differential phase measurements.
545
+ 5.3. Differential phases and photo-centre shifts
546
+ From the differential phases, we can derive the photo-centre shift
547
+ between the continuum and the BrΞ³ line emitting region. In the
548
+ regime of marginally resolved sources, there is a direct relation-
549
+ ship between the projected photo-centre displacement vector (P)
550
+ and the phase along each baseline (Lachaume 2003):
551
+ Ο†i = βˆ’2Ο€ Bi
552
+ Ξ» P,
553
+ (3)
554
+ where Ο†i is the differential phase measured for the ith baseline,
555
+ Bi is the length of the corresponding baseline, and Ξ» is the effec-
556
+ tive wavelength of the spectral channel. A four telescope beam-
557
+ combiner like GRAVITY gives access to six projected baselines
558
+ that enable us to accurately retrieve the value and orientation
559
+ of the photo-centre shifts in each spectral channel (Le Bouquin
560
+ et al. 2009; Waisberg et al. 2017). Such a measurement results in
561
+ a position-velocity plot of the displacement of the photo-centre
562
+ across the BrΞ³ line relative to the continuum. This is illustrated
563
+ in the bottom panels of Figure 5.
564
+ The photo-centre shifts trace the accretion funnel flow’s di-
565
+ rection and follow the stellar rotation. For instance, when the
566
+ northern accretion shock (N-shock) is located behind the star
567
+ (phase = 0), the accreting material falls onto the stellar surface in
568
+ the direction of the observer. Accordingly, the photo-centre mea-
569
+ sured in the blue-shifted part of the line profile (≃ -75 km sβˆ’1)
570
+ lies on the blue-shifted part of the velocity map, corresponding
571
+ to the approaching funnel flow. Equivalently, the photo-centre
572
+ measured in positive velocity channels of the line profile (≃
573
+ +75 km sβˆ’1) is shifted towards the receding funnel flow. In con-
574
+ trast, when the shock faces the observer (phase = 0.5), the veloc-
575
+ ity map goes from blue to red in the east-west direction, and the
576
+ photo-centre shifts recover this trend as demonstrated at phase
577
+ 0.47.
578
+ We can thus identify three privileged directions and shapes
579
+ of the photo-centre shifts: – linear north-south at phase ≃ 0 (N-
580
+ shock behind), – S-shape at phase 0.25 and 0.69 and – linear
581
+ east-west at phase ≃ 0.5 (N-shock in front). The differential
582
+ phase is, therefore, a key ingredient to recover the geometry and
583
+ orientation of the line-emitting region, tracing the moving mate-
584
+ rial along a rotational cycle.
585
+ 5.4. Signal-to-noise considerations
586
+ As a proof-of-concept, the results presented above assume infi-
587
+ nite signal-to-noise ratio. The goal is to predict the spectroscopic
588
+ and interferometric signatures of the magnetospheric accretion
589
+ process. Thus, the models predict typical visibility amplitudes
590
+ ranging from 1 down to 0.97 (see Fig. 5). Such a modest inter-
591
+ ferometric signal requires a measurement accuracy of about 1%
592
+ to be securely detected. Similarly, the models predict a deviation
593
+ of the differential phases by 1 to 2 degrees (Fig. 5), which re-
594
+ quires an accuracy of order of a fraction of a degree to yield a
595
+ robust detection. Recent interferometric studies performed with
596
+ VLTI/GRAVITY in the K-band demonstrate that these levels of
597
+ accuracy can be routinely obtained indeed with reasonable ex-
598
+ posure times on young stellar objects (e.g. Bouvier et al. 2020b;
599
+ Gravity Collaboration et al. 2020, 2022), or active galactic nuclei
600
+ (Gravity Collaboration et al. 2018).
601
+ 6. Summary and conclusion
602
+ We presented non-LTE radiative transfer modelling of the Brack-
603
+ ett Ξ³ line emission for non-axisymmetric models of accreting
604
+ magnetospheres. We used the equations of a misaligned dipo-
605
+ lar magnetic field to derive the geometry of the magnetospheric
606
+ accretion region for different obliquities of the magnetic dipole.
607
+ We used MCFOST to compute radiative signatures of the BrΞ³
608
+ line along a full stellar rotational cycle. Further, we derived near-
609
+ infrared interferometric observables for the line, comparable to
610
+ what the GRAVITY instrument has already measured for T Tauri
611
+ stars.
612
+ The main conclusions of this study are the following:
613
+ 1) The total flux in the line, and the line-to-continuum ratio,
614
+ depends on the obliquity of the dipole. As the obliquity in-
615
+ creases, the size of the emitting region decreases, leading
616
+ to a lower integrated flux. Also, projection effects make the
617
+ emission region of lines forming close to the stellar surface
618
+ appearing narrower.
619
+ 2) The BrΞ³ line total flux varies with the rotational phase due to
620
+ the non-axisymmetry of the models induced by the magnetic
621
+ obliquity. The line profiles exhibit a red-shifted absorption,
622
+ that is an inverse P Cygni profile, when a significant fraction
623
+ of the accretion shock is aligned with the observer’s line of
624
+ sight. When the shock is hidden on the opposite side of the
625
+ star, the line profiles exhibit a double-peaked shape, reminis-
626
+ cent of the lines formed in rotating envelope. The latter is
627
+ due to the relatively large rotational velocity of the magneto-
628
+ spheric model (∼80 km sβˆ’1).
629
+ Article number, page 6 of 13
630
+
631
+ B. Tessore & A. Soulain et al.: Spectroscopic and interferometric signatures of magnetospheric accretion
632
+ 3) Near-infrared interferometric observations in the BrΞ³ line di-
633
+ rectly probe the size of the magnetospheric accretion region.
634
+ The Gaussian disc model is sensitive to the brightest parts of
635
+ the magnetosphere, up to 50% of the truncation radius, while
636
+ a uniform disc model grasps 90% of the magnetosphere. It is
637
+ of prime importance to consider this aspect when estimating
638
+ magnetospheric radius from interferometric measurements.
639
+ In both cases, the measured radius varies with the rotational
640
+ phase (due to the non-axisymmetry of the dipole). A robust
641
+ interferometric estimate of the magnetospheric radius there-
642
+ fore requires monitoring the system over a full rotational cy-
643
+ cle.
644
+ 4) The combined knowledge of the differential phase and the
645
+ associated photo-centre shifts gives hints on the object ori-
646
+ entation and geometry. More specifically, the relative direc-
647
+ tion of the photo-centre shifts indicates the changing orien-
648
+ tation of the accreting material along the rotational cycle in
649
+ the non-axisymmetric case.
650
+ Near-infrared interferometry of the Brackett Ξ³ line is used
651
+ to characterise the inner star-disc interaction region, and offers a
652
+ good estima te of the size of the line’s forming region, at sub-au
653
+ precision. Comparing this size with reference model radii, such
654
+ as the truncation radius, allows us to distinguish between mul-
655
+ tiple origins of the BrΞ³ line, within or beyond these radii (e.g.
656
+ magnetosphere, stellar and disc winds, jets). Further, simultane-
657
+ ous spectroscopic and interferometric observations along a rota-
658
+ tional cycle, have the potential to unveil the geometry and ori-
659
+ entation of the line’s emitting region. The variability of the line
660
+ associated with the photo-centre shifts, provides a unique and
661
+ unambiguous proxy of the physical processes occurring in the
662
+ magnetosphere of young accreting systems, within a few hun-
663
+ dredths of an astronomical unit around the central star.
664
+ Article number, page 7 of 13
665
+
666
+ A&A proofs: manuscript no. bt_spidi2
667
+ Fig. 5: Synthetic interferometric measurements and modelling across the rotational phase of the system. Top: integrated images over the BrΞ³ line. Green (uniform disc) and
668
+ yellow (Gaussian disc) ellipses are the characteristic sizes measured with a GRAVITY-like instrument. Middle-top: Pure line visibility amplitude observables associated with
669
+ the corresponding models. The visibility variation (for a given baseline) as the u-v plane rotates is the specific signature of an elongated object. Middle-bottom: Pure phase
670
+ visibility across the line profile for the six baselines of the VLTI. Colours encode the observing time. Bottom: Velocity map of the radiative transfer model. The coloured dots
671
+ represent the measurement of the photo-centre derived from the phase visibility in each available spectral channel (see appendix B). In each figure, the magnetosphere and the
672
+ stellar surface have been normalised independently, for display purpose.
673
+ Article number, page 8 of 13
674
+
675
+ phase = 0.03
676
+ phase = 0.47
677
+ phase = 0.92
678
+ rud = 3.60 R*
679
+ phase = 0.25
680
+ rud = 3.88 R*
681
+ phase = 0.69
682
+ rud = 3.06 R*
683
+ rud = 3.30 R*
684
+ rgd = 2.12 R
685
+ rgd = 2.43 R,
686
+ rgd = 2.29 R*
687
+ rgd = 1.80 R*
688
+ 1.95 R
689
+ 0.8
690
+ 0.8
691
+ 0.8
692
+ 0.8
693
+ 0.8
694
+ 0.6
695
+ 0.6
696
+ 0.6
697
+ 0.6
698
+ 0.6
699
+ 0.
700
+ 0.01 AU
701
+ 0.01 AU
702
+ 0.01 AU
703
+ 0.01 AU
704
+ 0.01 AU
705
+ 1.00-
706
+ 1.00-
707
+ 1.00
708
+ 1.00
709
+ 1.00
710
+ 0.99
711
+ m0.991
712
+ β‰₯0.99 -
713
+ visibil
714
+ visibil
715
+ visibil
716
+ visibil
717
+ visibil
718
+ .-GD fit
719
+ ..GD fit
720
+ .. GD fit
721
+ .. GD fit
722
+ .-GD fit
723
+ UD fit
724
+ UD fit
725
+ UD fit
726
+ UD fit
727
+ UD fit
728
+ UT1-UT2
729
+ In-TIn
730
+ UT1-UT2
731
+ UT1-UT2
732
+ UT1-UT2
733
+ UT1-UT3
734
+ UT1-UT4
735
+ UT1-UT4
736
+ UT1-UT4
737
+ UT1-UT4
738
+ UT1-UT4
739
+ UT2-UT3
740
+ UT2-UT3
741
+ UT2-UT3
742
+ UT2-UT3
743
+ UT2-UT3
744
+ -L60
745
+ F260
746
+ L60
747
+ L60
748
+ -L60
749
+ .
750
+ UT2-UT4
751
+ UT2-UT4
752
+ UT2-UT4
753
+ UT2-UT4
754
+ UT3-UT4
755
+ 40
756
+ 120
757
+ 2 0
758
+ 50
759
+ 50
760
+ 50
761
+ Diff. Ξ¦ [deg]
762
+ Diff. [deg]
763
+ Diff. [deg]
764
+ Diff. [deg]
765
+ Diff. [deg
766
+ UT2
767
+ Ξ¦ [deg]
768
+ p[deg]
769
+ [deg]
770
+ [deg]
771
+ [deg]
772
+ Diff.
773
+ [deg]
774
+ [deg]
775
+ [deg]
776
+ [deg]
777
+ [deg]
778
+ Diff.
779
+ Diff.
780
+ Diff.
781
+ Diff.
782
+ Diff.
783
+ 100
784
+ 100
785
+ 100
786
+ 100
787
+ 100
788
+ 100
789
+ 100
790
+ 100
791
+ 100
792
+ 100
793
+ 100
794
+ .00
795
+ 100
796
+ 100
797
+ elocity [km/s]
798
+ Velocity [km/s]
799
+ elocity [km/s]
800
+ F00m
801
+ F00m
802
+ F00m
803
+ -00m
804
+ -00m
805
+ 75
806
+ 75
807
+ 75
808
+ 75
809
+ 200-
810
+ 200-
811
+ Photocenter shift[ΞΌas]
812
+ -002
813
+ 200
814
+ Photocenter shift [ΞΌas]
815
+ 50
816
+ [ΞΌas]
817
+ [ΞΌas]
818
+ [ΞΌas]
819
+ 100-
820
+ 100-
821
+ 100-
822
+ 100-
823
+ hotocenter shift ["
824
+ 100-
825
+ hotocenter shift [
826
+ hotocenter shift [
827
+ 25
828
+ 25
829
+ 25
830
+ 25
831
+ -0
832
+ -100
833
+ -100-
834
+ -100-
835
+ -100-
836
+ -100-
837
+ -50
838
+ -50
839
+ -50
840
+ -50
841
+ -50
842
+ -200-
843
+ -200-
844
+ -200-
845
+ 200-
846
+ -75
847
+ -75
848
+ -75
849
+ -75
850
+ -75
851
+ -300-
852
+ -300-
853
+ -300-
854
+ -300 -200 -
855
+ -300 -200 -
856
+ -300 -200 -
857
+ -300 -200 -
858
+ -300 -200 -
859
+ -100
860
+ 300
861
+ -100
862
+ 300
863
+ -100
864
+ 300
865
+ -100
866
+ 300
867
+ -100
868
+ 300
869
+ Photocenter shift [ΞΌasB. Tessore & A. Soulain et al.: Spectroscopic and interferometric signatures of magnetospheric accretion
870
+ Appendix A: benchmark
871
+ We present here the comparison between line profiles obtained
872
+ with MCFOST and previous studies. The magnetospheric model
873
+ corresponds to the axisymmetric and compact configuration of
874
+ Muzerolle et al. (2001) with a fixed shock temperature at 8 000
875
+ K, a rotation period of 5 days and the following canonical T Tauri
876
+ parameters: Mβˆ— = 0.8 MβŠ™, Rβˆ— = 2 RβŠ™ and Tβˆ— = 4 000 K. The in-
877
+ clination of the system is 60 degrees. The continuum emission
878
+ of the stellar surface (shock and photosphere) is constant for all
879
+ models. Figures A.1, A.2, A.3, and A.4 show the HΞ±, HΞ², PaΞ²
880
+ and BrΞ³ lines profiles for different values of Tmax and Λ™M. An in-
881
+ verse P Cygni profile, with a red-shifted absorption, is seen for
882
+ all lines although it is dependent on the value of the mass ac-
883
+ cretion rate and of the maximum temperature. For a given mass
884
+ accretion rate, an increase of the maximum temperature results
885
+ in a higher line emission peak and a shallower red-shifted ab-
886
+ sorption. As the temperature increases, the line source function
887
+ increases, which is the cause of a higher emission above the
888
+ continuum emission. The appearance of the red-shifted absorp-
889
+ tion component is caused by absorption from the gas above the
890
+ stellar surface. It is controlled by the ratio between the source
891
+ function of the line in the accretion funnel and that of the un-
892
+ derlying continuum from the stellar surface, especially at low
893
+ mass accretion rates and temperatures. Eventually, for the high-
894
+ est mass accretion rate and temperature, the lines become so op-
895
+ tically thick that the red-shifted absorption is washed out by the
896
+ large wings of the line. The red-shifted absorption is more pro-
897
+ nounced for lines forming closer to the accretion shock like the
898
+ HΞ² line. At a temperature larger than 8,000 K and a mass accre-
899
+ tion rate above 10βˆ’8 MβŠ™ yrβˆ’1, the continuum emission from the
900
+ magnetosphere becomes important and the line-to-continuum ra-
901
+ tio decreases. This effect is seen for instance in the HΞ± line (see
902
+ Fig. A.1). When the mass accretion rate increases for a given
903
+ temperature, the density of the magnetosphere increases. As a
904
+ consequence, the line source function increases. At high temper-
905
+ ature and high density, the background continuum emission of
906
+ the magnetosphere dominates for certain wavelengths, and ab-
907
+ sorption occurs. The latter effect is seen in the PaΞ² (Fig. A.3) and
908
+ BrΞ³ (Fig. A.4) lines where the strong continuum contribution at
909
+ the disc surface leads to absorption at low velocities, where the
910
+ lines source function is small. These results are consistent with
911
+ the previous studies and demonstrate the robustness of our code
912
+ for modelling the close environment of T Tauri stars (Tessore
913
+ et al. 2021).
914
+ Appendix B: Derivation of the interferometric
915
+ pure-line phase and visibility
916
+ We focus on the magnetospheric emission probed by the BrΞ³ line
917
+ and, therefore, aim to remove any additional contributions (stel-
918
+ lar photosphere, the accretion shocks, dusty disc, etc...). Follow-
919
+ ing Kraus et al. (2008); Bouvier et al. (2020b), we compute the
920
+ continuum-subtracted observables, the so-called pure line visi-
921
+ bility and phase, by using the emission line profiles computed in
922
+ Sect. 3. This is of prime importance in the case of BrΞ³ line as
923
+ the magnetospheric emission is quite faint in the infrared (β‰ˆ 1.3
924
+ excess flux compared to the continuum, Fig. 3). The deriva-
925
+ tion of the pure line quantities is only possible if the source is
926
+ marginally resolved (i.e, size < Ξ»/2B).
927
+ In this case, the pure line visibility Vline and phase Ξ¦line are
928
+ given by:
929
+ VLine = FL/CVTot βˆ’ VCont
930
+ FL/C βˆ’ 1
931
+ ,
932
+ (B.1)
933
+ Ξ¦Line = arcsin
934
+ οΏ½
935
+ FL/C
936
+ FL/C βˆ’ 1
937
+ VTot
938
+ VLine
939
+ sin Ξ¦Tot
940
+ οΏ½
941
+ .
942
+ (B.2)
943
+ Where FL/C denotes the line-to-continuum flux ratio as taken
944
+ from the normalised spectrum (Fig. 3), VCont is the visibility
945
+ computed in the continuum (star+shock only), and VTot, Ξ¦Tot
946
+ are the total complex quantities measured by GRAVITY. In
947
+ Eq. (B.1), we note that in the case when FL/C is close to one,
948
+ the derived VLine cannot exist (converges to infinity). Such non-
949
+ ideal profiles appear if the red absorption becomes too important.
950
+ Therefore, we assume to discard the affected spectral channels
951
+ for phases 0.25 and 0.47, where FL/C is too close to one: – one
952
+ point (v = 53 km sβˆ’1) at phases 0.25 and – two points (v = 15,
953
+ 53 km sβˆ’1) at phase 0.47.
954
+ Acknowledgements. The authors thank Claudio Zanni, Lucas Labadie, Cather-
955
+ ine Dougados, and Alexander Wojtczak for fruitful discussions. This project
956
+ has received funding from the European Research Council (ERC) under the
957
+ European Union’s Horizon 2020 research and innovation programme (grant
958
+ agreement No 742095; SPIDI: Star-Planets-Inner Disk-Interactions, http://
959
+ www.spidi-eu.org). B. Tessore thanks the french minister of Europe and
960
+ foreign affairs and the minister of superior education, research and innova-
961
+ tion (MEAE and MESRI) for research funding through FASIC partnership.
962
+ C. Pinte acknowledges funding from the Australian Research Council via
963
+ FT170100040 and DP180104235. The numerical simulations presented in this
964
+ paper were performed with the Dahu supercomputer of the GRICAD infrastruc-
965
+ ture (https://gricad.univ-grenoble-alpes.fr), which is supported by Grenoble re-
966
+ search communities.
967
+ Article number, page 9 of 13
968
+
969
+ A&A proofs: manuscript no. bt_spidi2
970
+ 0.8
971
+ 1.0
972
+ 1.2
973
+ M = 10
974
+ 9.0 M /yr
975
+ 1.0
976
+ 1.5
977
+ M = 10
978
+ 8.0 M /yr
979
+ 1
980
+ 2
981
+ 6000 K
982
+ M = 10
983
+ 7.0 M /yr
984
+ 0.75
985
+ 1.00
986
+ 1.25
987
+ 1
988
+ 2
989
+ 2
990
+ 4
991
+ 7000 K
992
+ 1.0
993
+ 1.5
994
+ 2.0
995
+ 2
996
+ 4
997
+ 6
998
+ 5
999
+ 10
1000
+ 8000 K
1001
+ 1000
1002
+ 500
1003
+ 0
1004
+ 500
1005
+ 1000
1006
+ 2
1007
+ 4
1008
+ 1000
1009
+ 500
1010
+ 0
1011
+ 500
1012
+ 1000
1013
+ 5
1014
+ 10
1015
+ 15
1016
+ 1000
1017
+ 500
1018
+ 0
1019
+ 500
1020
+ 1000
1021
+ 1
1022
+ 2
1023
+ 10000 K
1024
+ H
1025
+ v [km/s]
1026
+ F/Fc
1027
+ v [km/s]
1028
+ F/Fc
1029
+ v [km/s]
1030
+ F/Fc
1031
+ v [km/s]
1032
+ F/Fc
1033
+ v [km/s]
1034
+ F/Fc
1035
+ v [km/s]
1036
+ F/Fc
1037
+ v [km/s]
1038
+ F/Fc
1039
+ v [km/s]
1040
+ F/Fc
1041
+ v [km/s]
1042
+ F/Fc
1043
+ v [km/s]
1044
+ F/Fc
1045
+ v [km/s]
1046
+ F/Fc
1047
+ v [km/s]
1048
+ F/Fc
1049
+ Fig. A.1: Dependence of HΞ± line flux with mass accretion rates Λ™M and maximum temperatures Tmax. The inclination of the system
1050
+ is 60β—¦.
1051
+ 0.9
1052
+ 1.0
1053
+ M = 10
1054
+ 9.0 M /yr
1055
+ 0.8
1056
+ 1.0
1057
+ M = 10
1058
+ 8.0 M /yr
1059
+ 0.75
1060
+ 1.00
1061
+ 1.25
1062
+ 6000 K
1063
+ M = 10
1064
+ 7.0 M /yr
1065
+ 0.8
1066
+ 1.0
1067
+ 1.0
1068
+ 1.5
1069
+ 1
1070
+ 2
1071
+ 3
1072
+ 7000 K
1073
+ 0.75
1074
+ 1.00
1075
+ 1.25
1076
+ 1
1077
+ 2
1078
+ 3
1079
+ 2
1080
+ 4
1081
+ 6
1082
+ 8000 K
1083
+ 1000
1084
+ 500
1085
+ 0
1086
+ 500
1087
+ 1000
1088
+ 1.0
1089
+ 1.5
1090
+ 2.0
1091
+ 1000
1092
+ 500
1093
+ 0
1094
+ 500
1095
+ 1000
1096
+ 2.5
1097
+ 5.0
1098
+ 7.5
1099
+ 1000
1100
+ 500
1101
+ 0
1102
+ 500
1103
+ 1000
1104
+ 1
1105
+ 2
1106
+ 3
1107
+ 10000 K
1108
+ H
1109
+ v [km/s]
1110
+ F/Fc
1111
+ v [km/s]
1112
+ F/Fc
1113
+ v [km/s]
1114
+ F/Fc
1115
+ v [km/s]
1116
+ F/Fc
1117
+ v [km/s]
1118
+ F/Fc
1119
+ v [km/s]
1120
+ F/Fc
1121
+ v [km/s]
1122
+ F/Fc
1123
+ v [km/s]
1124
+ F/Fc
1125
+ v [km/s]
1126
+ F/Fc
1127
+ v [km/s]
1128
+ F/Fc
1129
+ v [km/s]
1130
+ F/Fc
1131
+ v [km/s]
1132
+ F/Fc
1133
+ Fig. A.2: Same as Fig. A.1 for HΞ²
1134
+ Article number, page 10 of 13
1135
+
1136
+ B. Tessore & A. Soulain et al.: Spectroscopic and interferometric signatures of magnetospheric accretion
1137
+ 1.00
1138
+ 1.02
1139
+ M = 10
1140
+ 9.0 M /yr
1141
+ 0.98
1142
+ 1.00
1143
+ 1.02
1144
+ M = 10
1145
+ 8.0 M /yr
1146
+ 0.9
1147
+ 1.0
1148
+ 1.1
1149
+ 6000 K
1150
+ M = 10
1151
+ 7.0 M /yr
1152
+ 0.98
1153
+ 1.00
1154
+ 1.02
1155
+ 1.0
1156
+ 1.2
1157
+ 1.0
1158
+ 1.5
1159
+ 7000 K
1160
+ 1.0
1161
+ 1.1
1162
+ 1.0
1163
+ 1.5
1164
+ 1
1165
+ 2
1166
+ 8000 K
1167
+ 1000
1168
+ 500
1169
+ 0
1170
+ 500
1171
+ 1000
1172
+ 1.0
1173
+ 1.2
1174
+ 1.4
1175
+ 1000
1176
+ 500
1177
+ 0
1178
+ 500
1179
+ 1000
1180
+ 1
1181
+ 2
1182
+ 1000
1183
+ 500
1184
+ 0
1185
+ 500
1186
+ 1000
1187
+ 1.0
1188
+ 1.1
1189
+ 1.2
1190
+ 10000 K
1191
+ Pa
1192
+ v [km/s]
1193
+ F/Fc
1194
+ v [km/s]
1195
+ F/Fc
1196
+ v [km/s]
1197
+ F/Fc
1198
+ v [km/s]
1199
+ F/Fc
1200
+ v [km/s]
1201
+ F/Fc
1202
+ v [km/s]
1203
+ F/Fc
1204
+ v [km/s]
1205
+ F/Fc
1206
+ v [km/s]
1207
+ F/Fc
1208
+ v [km/s]
1209
+ F/Fc
1210
+ v [km/s]
1211
+ F/Fc
1212
+ v [km/s]
1213
+ F/Fc
1214
+ v [km/s]
1215
+ F/Fc
1216
+ Fig. A.3: Same as Fig. A.1 for PaΞ².
1217
+ 0.99
1218
+ 1.00
1219
+ 1.01
1220
+ M = 10
1221
+ 9.0 M /yr
1222
+ 0.99
1223
+ 1.00
1224
+ 1.01
1225
+ M = 10
1226
+ 8.0 M /yr
1227
+ 1.00
1228
+ 1.05
1229
+ 6000 K
1230
+ M = 10
1231
+ 7.0 M /yr
1232
+ 0.99
1233
+ 1.00
1234
+ 1.01
1235
+ 0.95
1236
+ 1.00
1237
+ 1.05
1238
+ 1.0
1239
+ 1.2
1240
+ 1.4
1241
+ 7000 K
1242
+ 1.000
1243
+ 1.025
1244
+ 1.0
1245
+ 1.2
1246
+ 1.0
1247
+ 1.5
1248
+ 8000 K
1249
+ 1000
1250
+ 500
1251
+ 0
1252
+ 500
1253
+ 1000
1254
+ 1.0
1255
+ 1.1
1256
+ 1.2
1257
+ 1000
1258
+ 500
1259
+ 0
1260
+ 500
1261
+ 1000
1262
+ 1.0
1263
+ 1.5
1264
+ 1000
1265
+ 500
1266
+ 0
1267
+ 500
1268
+ 1000
1269
+ 1.00
1270
+ 1.05
1271
+ 1.10
1272
+ 10000 K
1273
+ Br
1274
+ v [km/s]
1275
+ F/Fc
1276
+ v [km/s]
1277
+ F/Fc
1278
+ v [km/s]
1279
+ F/Fc
1280
+ v [km/s]
1281
+ F/Fc
1282
+ v [km/s]
1283
+ F/Fc
1284
+ v [km/s]
1285
+ F/Fc
1286
+ v [km/s]
1287
+ F/Fc
1288
+ v [km/s]
1289
+ F/Fc
1290
+ v [km/s]
1291
+ F/Fc
1292
+ v [km/s]
1293
+ F/Fc
1294
+ v [km/s]
1295
+ F/Fc
1296
+ v [km/s]
1297
+ F/Fc
1298
+ Fig. A.4: Same as Fig. A.1 for BrΞ³.
1299
+ Article number, page 11 of 13
1300
+
1301
+ A&A proofs: manuscript no. bt_spidi2
1302
+ 50
1303
+ 25
1304
+ 0
1305
+ 25
1306
+ 50
1307
+ U [M ]
1308
+ 60
1309
+ 40
1310
+ 20
1311
+ 0
1312
+ 20
1313
+ 40
1314
+ 60
1315
+ V [M ]
1316
+ UT1-UT2
1317
+ UT1-UT3
1318
+ UT1-UT4
1319
+ UT2-UT3
1320
+ UT2-UT4
1321
+ UT3-UT4
1322
+ 150
1323
+ 100
1324
+ 50
1325
+ 0
1326
+ 50
1327
+ 100
1328
+ 150
1329
+ V [m] - East
1330
+ 150
1331
+ 100
1332
+ 50
1333
+ 0
1334
+ 50
1335
+ 100
1336
+ 150
1337
+ U [m] (2.17 Β΅m) - North
1338
+ Fig. B.1: Fourier sampling (u-v coverage) of the simulated data.
1339
+ The different colours correspond to the six different baselines
1340
+ of the VLTI. The eight points per baseline represent a typical
1341
+ observational sequence with one data point per hour.
1342
+ Article number, page 12 of 13
1343
+
1344
+ B. Tessore & A. Soulain et al.: Spectroscopic and interferometric signatures of magnetospheric accretion
1345
+ References
1346
+ Alencar, S. H. P., Bouvier, J., Donati, J. F., et al. 2018, A&A, 620, A195
1347
+ Alencar, S. H. P., Bouvier, J., Walter, F. M., et al. 2012, A&A, 541, A116
1348
+ Berger, J. P. 2003, in EAS Publications Series, Vol. 6, EAS Publications Series,
1349
+ ed. G. Perrin & F. Malbet, 23
1350
+ Blinova, A. A., Romanova, M. M., & Lovelace, R. V. E. 2016, MNRAS, 459,
1351
+ 2354
1352
+ Bourgès, L. & Duvert, G. 2016, in Society of Photo-Optical Instrumentation En-
1353
+ gineers (SPIE) Conference Series, Vol. 9907, Optical and Infrared Interfer-
1354
+ ometry and Imaging V, ed. F. Malbet, M. J. Creech-Eakman, & P. G. Tuthill,
1355
+ 990711
1356
+ Bouvier, J., Alecian, E., Alencar, S. H. P., et al. 2020a, A&A, 643, A99
1357
+ Bouvier, J., Matt, S. P., Mohanty, S., et al. 2014, in Protostars and Planets VI,
1358
+ ed. H. Beuther, R. S. Klessen, C. P. Dullemond, & T. Henning, 433
1359
+ Bouvier, J., Perraut, K., Le Bouquin, J. B., et al. 2020b, A&A, 636, A108
1360
+ Calvet, N. & Gullbring, E. 1998, ApJ, 509, 802
1361
+ Cody, A. M., Stauffer, J., Baglin, A., et al. 2014, AJ, 147, 82
1362
+ Dmitriev, D. V., Grinin, V. P., & Katysheva, N. A. 2019, Astronomy Letters, 45,
1363
+ 371
1364
+ Donati, J. F., Bouvier, J., Alencar, S. H., et al. 2020, MNRAS, 491, 5660
1365
+ Donati, J. F., Gregory, S. G., Alencar, S. H. P., et al. 2013, MNRAS, 436, 881
1366
+ Donati, J. F., Jardine, M. M., Gregory, S. G., et al. 2008, MNRAS, 386, 1234
1367
+ Donati, J. F., Skelly, M. B., Bouvier, J., et al. 2010, MNRAS, 409, 1347
1368
+ Esau, C. F., Harries, T. J., & Bouvier, J. 2014, MNRAS, 443, 1022
1369
+ Folha, D. F. M. & Emerson, J. P. 2001, A&A, 365, 90
1370
+ Galli, P. A. B., Joncour, I., & Moraux, E. 2018a, MNRAS, 477, L50
1371
+ Galli, P. A. B., Loinard, L., Ortiz-LΓ©on, G. N., et al. 2018b, ApJ, 859, 33
1372
+ Ghosh, P., Lamb, F. K., & Pethick, C. J. 1977, ApJ, 217, 578
1373
+ Gravity Collaboration, Abuter, R., Accardo, M., et al. 2017, A&A, 602, A94
1374
+ Gravity Collaboration, Garcia Lopez, R., Natta, A., et al. 2020, Nature, 584, 547
1375
+ Gravity Collaboration, Sturm, E., Dexter, J., et al. 2018, Nature, 563, 657
1376
+ Gravity Collaboration, Wojtczak, J. A., Labadie, L., et al. 2022, arXiv e-prints,
1377
+ arXiv:2210.13095
1378
+ Hartmann, L., Avrett, E., & Edwards, S. 1982, ApJ, 261, 279
1379
+ Hartmann, L., Hewett, R., & Calvet, N. 1994, ApJ, 426, 669
1380
+ Herbst, W., Eislâffel, J., Mundt, R., & Scholz, A. 2007, in Protostars and Planets
1381
+ V, ed. B. Reipurth, D. Jewitt, & K. Keil, 297
1382
+ Johnstone, C. P., Jardine, M., Gregory, S. G., Donati, J. F., & Hussain, G. 2014,
1383
+ MNRAS, 437, 3202
1384
+ Kraus, S., Hofmann, K. H., Benisty, M., et al. 2008, A&A, 489, 1157
1385
+ Kurosawa, R., Harries, T. J., & Symington, N. H. 2006, MNRAS, 370, 580
1386
+ Kurosawa, R., Romanova, M. M., & Harries, T. J. 2011, MNRAS, 416, 2623
1387
+ Lachaume, R. 2003, A&A, 400, 795
1388
+ Le Bouquin, J. B., Absil, O., Benisty, M., et al. 2009, A&A, 498, L41
1389
+ Leenaarts, J., Carlsson, M., Hansteen, V., & Rutten, R. J. 2007, A&A, 473, 625
1390
+ Lima, G. H. R. A., Alencar, S. H. P., Calvet, N., Hartmann, L., & Muzerolle, J.
1391
+ 2010, A&A, 522, A104
1392
+ Mahdavi, A. & Kenyon, S. J. 1998, ApJ, 497, 342
1393
+ Martin, S. C. 1996, ApJ, 470, 537
1394
+ McGinnis, P., Bouvier, J., & Gallet, F. 2020, MNRAS, 497, 2142
1395
+ Muzerolle, J., Calvet, N., & Hartmann, L. 1998, ApJ, 492, 743
1396
+ Muzerolle, J., Calvet, N., & Hartmann, L. 2001, ApJ, 550, 944
1397
+ Pantolmos, G., Zanni, C., & Bouvier, J. 2020, A&A, 643, A129
1398
+ Pinte, C., Harries, T. J., Min, M., et al. 2009, A&A, 498, 967
1399
+ Pinte, C., MΓ©nard, F., DuchΓͺne, G., & Bastien, P. 2006, A&A, 459, 797
1400
+ Pouilly, K., Bouvier, J., Alecian, E., et al. 2020, A&A, 642, A99
1401
+ Reipurth, B., Pedrosa, A., & Lago, M. T. V. T. 1996, A&AS, 120, 229
1402
+ Romanova, M. M. & Owocki, S. P. 2016, Accretion, Outflows, and Winds of
1403
+ Magnetized Stars, Vol. 54, 347
1404
+ Romanova, M. M., Ustyugova, G. V., Koldoba, A. V., & Lovelace, R. V. E. 2004,
1405
+ ApJ, 610, 920
1406
+ Romanova, M. M., Ustyugova, G. V., Koldoba, A. V., Wick, J. V., & Lovelace,
1407
+ R. V. E. 2003, ApJ, 595, 1009
1408
+ Sousa, A. P., Alencar, S. H. P., Bouvier, J., et al. 2016, A&A, 586, A47
1409
+ Sousa, A. P., Bouvier, J., Alencar, S. H. P., et al. 2021, A&A, 649, A68
1410
+ Tessore, B., Pinte, C., Bouvier, J., & MΓ©nard, F. 2021, A&A, 647, A27
1411
+ Waisberg, I., Dexter, J., Pfuhl, O., et al. 2017, ApJ, 844, 72
1412
+ Zanni, C. & Ferreira, J. 2009, A&A, 508, 1117
1413
+ Article number, page 13 of 13
1414
+
JNFJT4oBgHgl3EQfwC0U/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
JdE4T4oBgHgl3EQf7Q53/content/tmp_files/2301.05338v1.pdf.txt ADDED
@@ -0,0 +1,510 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ arXiv:2301.05338v1 [cs.DS] 13 Jan 2023
2
+ Computing matching statistics on Wheeler DFAs
3
+ Alessio Conte1, Nicola Cotumaccio2,3, Travis Gagie3, Giovanni Manzini1,
4
+ Nicola Prezza4 and Marinella Sciortino5
5
+ 1 University of Pisa, Italy, alessio.conte@unipi.it, giovanni.manzini@unipi.it
6
+ 2 GSSI, L’Aquila, Italy, nicola.cotumaccio@gssi.it
7
+ 3 Dalhousie University, Halifax, Canada, nicola.cotumaccio@dal.ca, travis.gagie@dal.ca
8
+ 4 Ca’ Foscari Unversity, Venice, Italy, nicola.prezza@unive.it
9
+ 5 University of Palermo, Italy, marinella.sciortino@unipa.it
10
+ Abstract
11
+ Matching statistics were introduced to solve the approximate string matching problem, which is
12
+ a recurrent subroutine in bioinformatics applications.
13
+ In 2010, Ohlebusch et al. [SPIRE 2010]
14
+ proposed a time and space efficient algorithm for computing matching statistics which relies on
15
+ some components of a compressed suffix tree - notably, the longest common prefix (LCP) array. In
16
+ this paper, we show how their algorithm can be generalized from strings to Wheeler deterministic
17
+ finite automata. Most importantly, we introduce a notion of LCP array for Wheeler automata, thus
18
+ establishing a first clear step towards extending (compressed) suffix tree functionalities to labeled
19
+ graphs.
20
+ Introduction
21
+ Given a string T and a pattern Ο€, the classical formulation of the pattern matching problem
22
+ requires to decide whether the pattern Ο€ occurs in the string T and, possibly, count the
23
+ number of such occurrences and report the positions where they occur. The invention of
24
+ the FM-index [1], which is based on the Burrows-Wheeler transform [2], opened a new
25
+ line of research in the pattern matching field. The indexing and compression techniques
26
+ behind the FM-index deeply rely on the idea of suffix sorting, and over the years have
27
+ been generalized from strings to trees [3], De Brujin graphs [4,5], Wheeler graphs [6,7] and
28
+ arbitrary graphs [8, 9]. In particular, the class of Wheeler graphs is probably the one that
29
+ captures the intuition behind the FM-index in the simplest way, and indeed the notion of
30
+ Wheeler order has relevant consequences in automata theory [7,10].
31
+ However, in bioinformatics we are not only interested in exact pattern matching, but
32
+ also in a myriad of variations of the pattern matching problem [11]. In particular, matching
33
+ statistics were introduced to solve the approximate pattern matching problem [12]. A pow-
34
+ erful data structure that is able to address the variations of the pattern matching problem
35
+ at once is the suffix tree [13]. The main drawback of the suffix tree is its space consumption,
36
+ which is non-negligible both in theory and in practice. As a consequence, the suffix tree has
37
+ been replaced by the suffix array [14]. While suffix arrays do not have all the functionalities
38
+ of suffix trees, it has been shown that they can be augmented with some additional data
39
+ structures β€” notably, the longest common prefix (LCP) array β€” so that it is possible to
40
+ retrieve the full functionalities of a suffix trees [15]. All these components can be successfully
41
+ compressed, leading to the so-called compressed suffix trees [16].
42
+
43
+ The natural question is whether it is possible to provide suffix tree functionalities not
44
+ only to strings, but also to graphs, and in particular Wheeler graphs. In this paper, we
45
+ provide a first partial affirmative answer by considering the problem of computing matching
46
+ statistics. In 2010, Ohlebusch et al. [17] proposed a time and space efficient algorithm for
47
+ computing matching statistics which relies on some components of a compressed suffix tree.
48
+ In this paper, we show how their algorithm can be generalized from strings to Wheeler deter-
49
+ ministic finite automata. Most importantly, we introduce a notion of longest common prefix
50
+ (LCP) array for Wheeler automata, thus establishing an important step towards extending
51
+ (compressed) suffix tree functionalities to labeled graphs.
52
+ Notation and first definitions
53
+ Throughout the paper, we consider an alphabet Ξ£ and a fixed total order βͺ― on Ξ£. We
54
+ denote by Ξ£βˆ— the set of all finite strings on Ξ£ and by Σω the set of all (countably) infinite
55
+ strings on Ξ£. The empty word is Η«. If Ξ± ∈ Ξ£βˆ—, then Ξ±R is the reverse string of Ξ±. We extend
56
+ the total order βͺ― from Ξ£ to Ξ£βˆ— βˆͺ Σω lexicographically. If i and j are integers, with i ≀ j,
57
+ define [i, j] = {i, i + 1, . . . , j βˆ’ 1, j}. If T is a string, the i-th character of T is T[i], and
58
+ T[i..j] = T[i]..T[j].
59
+ We will consider deterministic automata A = (Q, E, s0, F), where Q is the set of states,
60
+ E βŠ† Q Γ— Q Γ— Ξ£ is the set of labeled edges, s0 ∈ Q is the initial state and F βŠ† Q is the set
61
+ of final states. The definition implies that for every u ∈ Q and for every a ∈ Σ there exists
62
+ at most one edge labeled a leaving u. Following [7,10], we assume that s0 has no incoming
63
+ edges, and every state is reachable from the initial state; moreover, all edges entering the
64
+ same state have the same label (input-consistency), so that for every u ∈ Q \ {s0} we can
65
+ let ��(u) be the label of all edges entering u. We define λ(s0) = #, where # ̸∈ Σ is a special
66
+ character for which we assume # β‰Ί a for every a ∈ Ξ£ (the character # is an analogous of
67
+ the termination character $ used for suffix trees and suffix arrays). As a consequence, an
68
+ edge (uβ€², u, a) can be simply written as (uβ€², u), because it must be a = Ξ»(u).
69
+ We assume familiarity with the notions of suffix array (SA), Burrows Wheeler transform
70
+ (BWT), FM-index and backward search [1].
71
+ The matching statistics of a pattern Ο€ = Ο€[1..m] with respect to a string T = T[1..n] are
72
+ defined as follows. Assume that T[n] = $ ̸∈ Ξ£, where $ β‰Ί a for every a ∈ Ξ£. Determining
73
+ the matching statistics of Ο€ with respect to T means determining, for 1 ≀ i ≀ m, (i) the
74
+ longest prefix Ο€β€² of Ο€[i..m] which occurs in T, and (ii) the interval corresponding to the
75
+ set of all strings starting with Ο€β€² in the list of all lexicographically sorted suffixes. We can
76
+ describe (i) and (ii) by means of three values: the length β„“i of Ο€β€², and the endpoints li and
77
+ ri of the interval considered in (ii). For example, let T = mississippi$ (see Figure 1), and
78
+ Ο€ = stpissi. For i = 1, we have Ο€β€² = s, so β„“1 = 1 and [l1, r1] = [9, 12] (suffixes starting with
79
+ s). For i = 2, we have Ο€β€² = Η«, so β„“2 = 0 and [l2, r2] = [1, n] = [1, 12] (all suffixes start with
80
+ the empty string). For i = 3, we have Ο€β€² = pi, so β„“3 = 2, and [l3, r3] = [7, 7] (suffixes starting
81
+ with pi). For i = 4, we have Ο€β€² = issi, so β„“4 = 4, and [l4, r4] = [4, 5] (suffixes starting with
82
+ issi). One can proceed analogously for i = 5, 6, 7.
83
+
84
+ i
85
+ Sorted suffixes
86
+ LCP
87
+ SA
88
+ BWT
89
+ 1
90
+ $
91
+ 12
92
+ i
93
+ 2
94
+ i$
95
+ 0
96
+ 11
97
+ p
98
+ 3
99
+ ippi$
100
+ 1
101
+ 8
102
+ s
103
+ 4
104
+ issippi$
105
+ 1
106
+ 5
107
+ s
108
+ 5
109
+ ississippi$
110
+ 4
111
+ 2
112
+ m
113
+ 6
114
+ mississippi$
115
+ 0
116
+ 1
117
+ $
118
+ 7
119
+ pi$
120
+ 0
121
+ 10
122
+ p
123
+ 8
124
+ ppi$
125
+ 1
126
+ 9
127
+ i
128
+ 9
129
+ sippi$
130
+ 0
131
+ 7
132
+ s
133
+ 10
134
+ sissippi$
135
+ 2
136
+ 4
137
+ s
138
+ 11
139
+ ssippi$
140
+ 1
141
+ 6
142
+ i
143
+ 12
144
+ ssissippi$
145
+ 3
146
+ 3
147
+ i
148
+ Figure 1: The sorted suffixes of β€œmississippi$” and the LCP, SA, and BWT arrays.
149
+ Computing matching statistics for strings
150
+ We will first describe the algorithm by Ohlebusch et al. [17], emphasizing the ideas that we
151
+ will generalize when switching to Wheeler DFAs. The algorithms computes the matching
152
+ statistics using a number of iterations linear in m by exploiting the backward search. We
153
+ start from the end of Ο€, and we use the backward search (starting from the interval [1, n]
154
+ which corresponds to the set of suffixes prefixed by the empty string) to find the interval of
155
+ all occurrences of the last character of Ο€ in T (if any). Then, starting from the new interval,
156
+ we use the backward search to find all the occurrences of the suffix of length 2 of Ο€ in T (if
157
+ any), and so on. At some point, it may happen that for some i ≀ m+1 we have that Ο€[i..m]
158
+ occurs in T, but the next application of the backward search returns the empty interval, so
159
+ that Ο€[i βˆ’ 1..m] does not occur in T (the case i = m + 1 corresponds to the initial setting
160
+ when Ο€[i..m] is the empty string). We distinguish two cases:
161
+ β€’ (Case 1) If li = 1 and ri = n, this means that all suffixes of T are prefixed by Ο€[i..m].
162
+ This may happen in particular if i = m + 1: this means that the first backward search
163
+ has been unsuccessful. We immediately conclude that character Ο€[iβˆ’1] does not occur
164
+ in T, so β„“iβˆ’1 = 0 and [liβˆ’1, riβˆ’1] = [1, n] (because all suffixes start with the empty
165
+ string). In this case, in the following iterations of the algorithm, we can simply discard
166
+ Ο€[i βˆ’ 1, m]: when for iβ€² ≀ i βˆ’ 2 we will be searching for the longest prefix of Ο€[iβ€², m]
167
+ occurring in T, it will suffice to search for the longest prefix of Ο€[iβ€², i βˆ’ 2] occurring in
168
+ T.
169
+ β€’ (Case 2) If li > 1 or ri < n, this means that the number of suffixes of T starting with
170
+ Ο€[i..m] is less than n. Now, every suffix starting with Ο€[i..m] also starts with Ο€[i..mβˆ’1].
171
+ If the number of suffixes starting with Ο€[i..m βˆ’ 1] is equal to the number of suffixes
172
+ starting with Ο€[i..m], then also Ο€[iβˆ’1..mβˆ’1] does not occur in T. More in general, for
173
+ j ≀ mβˆ’1 we can have that Ο€[iβˆ’1..j] occurs in T only if the number of suffixes starting
174
+ with Ο€[i..j] is larger than the number of suffixes starting with Ο€[i..m]. Since we are
175
+ interested in maximal matches, we want j to be as large as possible: we will show later
176
+
177
+ how to compute the largest integer j such that the number of suffixes starting with
178
+ Ο€[i..j] is larger than the number of suffixes starting with Ο€[i..m]. Notice that j always
179
+ exists, because all n suffixes start with the empty string, but less than n suffixes start
180
+ with Ο€[i..m]. After determining j we discard Ο€[j + 1..m] (so in the following iterations
181
+ of the algorithm we will simply consider Ο€[1..j]), and we recursively apply the backward
182
+ search starting from the interval associated with the occurrences of Ο€[i..j] β€” we will
183
+ also see how to compute this interval.
184
+ Let us apply the above algorithm to T = mississippi$ and Ο€ = stpissi. We start with
185
+ the interval [1, n] = [1, 12], corresponding to the empty pattern, and character Ο€[7] = i. A
186
+ backward step yields the interval [l7, r7] = [2, 5] (suffixes starting with i), so β„“7 = 1. Now, we
187
+ apply a backward step from [2, 5] and Ο€[6] = s, obtaining [l6, r6] = [9, 10] (suffixes starting
188
+ with si), so β„“6 = 2. Again, we apply a backward step from [9, 10] and Ο€[5] = s, obtaining
189
+ [l5, r5] = [11, 12] (suffixes starting with ssi), so β„“5 = 3. Again, we apply a backward step
190
+ from [11, 12] and Ο€[4] = i, obtaining [l4, r4] = [4, 5] (suffixes starting with issi), so β„“4 = 4.
191
+ We now apply a backward step from [4, 5] and Ο€[3] = p, and we obtain the empty interval.
192
+ This means that no suffix starts with pissi. Notice in Figure 1 that the number of suffixes
193
+ starting with issi is equal to the number of suffixes starting with iss or is, but the number
194
+ of suffixes starting with i is bigger. As a consequence, we consider the interval of all suffixes
195
+ starting with i β€” which is [2, 5] β€” and we apply a backward step with Ο€[3] = p. This time
196
+ the backward step is successful, and we obtain [l3, r3] = [7, 7] (suffixes starting with pi), and
197
+ β„“3 = 2. We now apply a backward step from [7, 7] and Ο€[2] = t, obtaining the empty interval.
198
+ This means that no suffix starts with tpi. Notice in Figure 1 that the number of suffixes
199
+ starting with p is bigger than the number of suffixes starting with pi. The corresponding
200
+ interval is [7, 8], but a backward step with Ο€[2] = t is still unsuccessful (so no suffix starts
201
+ with tp).
202
+ The number of suffixes starting with p is smaller than the number of suffixes
203
+ starting with the empty string (which is equal to n = 12), so we apply a backward step with
204
+ [1, 12] and Ο€[2] = t. Since the backward step is still unsuccessful, we conclude that Ο€[2] = t
205
+ does not occur in S, so [l2, r2] = [1, n] = [1, 12] and β„“2 = 0. Finally, we start again from the
206
+ whole interval [1, 12], and a backward step with Ο€[1] = s returns [l1, r1] = [9, 12] (suffixes
207
+ starting with s), so β„“1 = 1.
208
+ It is easy to see that the number of iterations is linear in m. Indeed, every time we apply
209
+ a backward step, either we move to the left across Ο€ to compute a new matching statistic,
210
+ or we increase by at least 1 the length of the suffix of Ο€ which is forever discarded. This
211
+ implies that the number of iterations is bounded by 2|Ο€| = 2m.
212
+ We are only left with showing (i) how to compute j and (ii) the interval of all suffixes
213
+ starting with Ο€[i..j] in Case 2 of the algorithm.
214
+ To this end, we introduce the longest
215
+ common prefix (LCP) array LCP = LCP[2, n] of T. We define LCP[i] to be the length of the
216
+ longest common prefix of the (i βˆ’ 1)-st lexicographically smallest suffix of T and the i-th
217
+ lexicographically smallest suffix of T. In Figure 1 we have LCP[5] = 4 because the fourth
218
+ lexicographically smallest suffix of T is issippi$, the fifth lexicographically smallest suffix of
219
+ T is ississippi$, and the longest common prefix of issippi$ and ississippi$ is issi, which has
220
+ length 4. Remember that in the example the backward search starting from [4, 5] (suffixes
221
+ starting with issi) and p was unsuccessful, so computing j means determining the longest
222
+ prefix of issi such that the the number of suffixes starting with such a prefix is bigger than 2.
223
+
224
+ 2
225
+ 5
226
+ 6
227
+ 7
228
+ 8
229
+ 9
230
+ 3
231
+ 4
232
+ 10
233
+ 11
234
+ 12
235
+ 13
236
+ 14
237
+ 15
238
+ 16
239
+ 17
240
+ 18
241
+ 19
242
+ 1
243
+ start
244
+ a
245
+ a
246
+ a
247
+ a
248
+ a
249
+ b
250
+ b
251
+ b
252
+ c
253
+ c
254
+ d
255
+ d
256
+ e
257
+ e
258
+ e
259
+ f
260
+ g
261
+ h
262
+ i
263
+ l
264
+ a
265
+ Figure 2: A Wheeler DFA. States are numbered according to their positions in the Wheeler
266
+ order.
267
+ This is easy to compute by using the LCP array: the longest such prefix is the one of length
268
+ max{LCP[4], LCP[6]} = max{1, 0} = 1, so that the desired prefix is i. As a consequence,
269
+ we are only left with showing how to compute the interval of all suffixes starting with the
270
+ prefix i β€” which is [2, 5]. Notice that in order to compute this interval, it is enough to
271
+ expand the interval [4, 6] in both directions as long as the LCP value does not go below 1.
272
+ Since LCP[4] = 1, LCP[3] = 1, and LCP[2] = 0, and we already know that LCP[6] = 0, we
273
+ conclude that the desired interval is [2, 5]. In other words, given a position t, we must be
274
+ able to compute the biggest integer k less than t such that LCP[k] < LCP[t], and the smallest
275
+ integer k bigger than t such that LCP[k] < LCP[t] (in our case, t = 4). These queries are
276
+ called PSV (β€œprevious smaller value”) and NSV (β€œnext smaller value”) queries. The LCP
277
+ array can be augmented in such a way that PSV and NSV queries can be solved efficiently:
278
+ different space-time trade-offs are possible, we refer the reader to [17] for details.
279
+ Matching statistics for Wheeler DFAs
280
+ Let us define Wheeler DFAs [7].
281
+ Definition 1. Let A = (Q, E, s0, F) be a DFA. A Wheeler order on A is a total order ≀ on
282
+ Q such that s0 ≀ u for every u ∈ Q and:
283
+ (Axiom 1) If u, v ∈ Q and u < v, then Ξ»(u) βͺ― Ξ»(v).
284
+ (Axiom 2) If (uβ€², u), (vβ€², v) ∈ E, Ξ»(u) = Ξ»(v) and u < v, then uβ€² < vβ€².
285
+ A DFA A is Wheeler if it admits a Wheeler order.
286
+ It is immediate to check that this definition is equivalent to the one in [7], where it was
287
+ shown that if a DFA A admits a Wheeler order ≀, then ≀ is uniquely determined (that is,
288
+
289
+ ≀ is the Wheeler order on A). In the following, we fix a Wheeler DFA A = (Q, E, s0, F),
290
+ where we assume Q = {u1, . . . , un}, with u1 < u2 < Β· Β· Β· < un in the Wheeler order, and u1
291
+ coincides with the initial state s0. See Figure 2 for an example.
292
+ We now show that a Wheeler order can be seen of as a permutation of the set of all states
293
+ playing the same role as the suffix array of a string. In the following, it will be expedient
294
+ to (conceptually) assume that s0 has a self-loop labeled # (this is consistent with Axiom 1,
295
+ because # β‰Ί a for every a ∈ Ξ£). This implies that every state has at least one incoming
296
+ edge, so for every state ui there exists at least one infinite string Ξ± ∈ Σω that can be read
297
+ starting from ui and following edges in a backward fashion (for example, in Figure 2 for u9
298
+ such a string is cel### . . . ). We denote by Iui the set of all such strings. Formally:
299
+ Definition 2. Let i ∈ [1, n]. For every state ui ∈ Q define:
300
+ Iui = {Ξ± ∈ Σω | there exist integers f1, f2, . . . in [1, n] such that (i) f1 = i,
301
+ (ii) (ufk+1, ufk) ∈ E for every k β‰₯ 1 and (iii) Ξ± = Ξ»(uf1)Ξ»(uf2) . . . }.
302
+ For example, in Figure 2 we have Iu3 = {abdg### . . . , abeh### . . . , acei### . . . }.
303
+ The following lemma shows that the permutation of the states defined by the Wheeler
304
+ order is the one lexicographically sorting the strings entering each state, just like the permu-
305
+ tation defined by the suffix array lexicographically sorts the suffixes of the strings (a suffix
306
+ is seen as a string β€œleaving” a text position).
307
+ Lemma 3. Let i, j ∈ [1, n], with i < j. Let Ξ± ∈ Iui and Ξ² ∈ Iuj. Then, Ξ± βͺ― Ξ².
308
+ Proof. Let f1, f2, . . . in [1, n] be such that (i) f1 = i, (ii) (ufk+1, ufk) ∈ E for every k β‰₯ 1 and
309
+ (iii) Ξ± = Ξ»(uf1)Ξ»(uf2) . . . . Analogously, let g1, g2, . . . in [1, n] be such that (i) g1 = j, (ii)
310
+ (ugk+1, ugk) ∈ E for every k β‰₯ 1 and (iii) Ξ² = Ξ»(ug1)Ξ»(ug2) . . . . Let Ξ± ΜΈ= Ξ². We must prove
311
+ that Ξ± β‰Ί Ξ². Let p β‰₯ 1 be the smallest integer such that the p-th character of Ξ± is different
312
+ than the p-th character of Ξ². In other words, we know that Ξ»(uf1) = Ξ»(ug1), Ξ»(uf2) = Ξ»(ug2),
313
+ . . . , Ξ»(ufpβˆ’1) = Ξ»(ugpβˆ’1), but Ξ»(ufp) ΜΈ= Ξ»(ugp). We must prove that Ξ»(ufp) β‰Ί Ξ»(ugp). Since
314
+ λ(uf1) = λ(ug1) f1 = i < j = g1, and (uf2, uf1), (ug2, ug1) ∈ E, from Axiom 2 we obtain
315
+ f2 < g2. Since λ(uf2) = λ(ug2), f2 < g2, and (uf3, uf2), (ug3, ug2) ∈ E, from Axiom 2 we
316
+ obtain f3 < g3. By iterating this argument, we conclude fp < gp. By Axiom 1, we obtain
317
+ Ξ»(ufp) βͺ― Ξ»(ugp). Since Ξ»(ufp) ΜΈ= Ξ»(ugp), we conclude Ξ»(ufp) β‰Ί Ξ»(ugp).
318
+ If we think of a string as a labeled path, then the suffix array sorts the strings that can
319
+ be read from each position by moving forward (that is, the suffixes of the string), while the
320
+ Wheeler order sorts the strings that can be read from each position by moving backward
321
+ towards the initial state. The underlying idea is the same: the forward vs backward difference
322
+ is only due to historical reasons [6]. To compute the matching statistics on Wheeler DFA
323
+ we reason as in the previous section replacing backward search with the forward search [6]
324
+ defined as follows: given an interval [i, j] in [1, n] and a ∈ Σ, find the (possibly empty)
325
+ interval [iβ€², jβ€²] in [1, n] such that a state vkβ€² is reachable from some state vk, with i ≀ k ≀ j,
326
+ through an edge labeled a, if and only if iβ€² ≀ kβ€² ≀ jβ€² (this easily follows by using the axioms of
327
+ Definition 1). For a constant size alphabet, given [i, j] and a then [iβ€², jβ€²] can be determined in
328
+ constant time. Given a string Ο€ ∈ Ξ£βˆ—, if we start from the whole set of states and repeatedly
329
+ apply the forward search we reach the set of all states ui for which there exists α ∈ Iui
330
+
331
+ prefixed by Ο€R; this is an interval with respect to the Wheeler order: in the following we call
332
+ this interval T(Ο€).
333
+ Because of the forward vs backward difference the problem of matching statistics will be
334
+ defined in a symmetrical way on Wheeler DFAs. Given a pattern Ο€ = Ο€[1..m], for every
335
+ 1 ≀ i ≀ m we want to determine (i) the longest suffix Ο€β€² of Ο€[1..i] which occurs in the
336
+ Wheeler DFA A (that is, that can be read somewhere on A by concatenating edges), and
337
+ (ii) the endpoints of the interval T(Ο€β€²).
338
+ Broadly speaking, we can apply the same idea of the algorithm for strings, but in a
339
+ symmetrical way. We start from the beginning of Ο€ (not from the end of Ο€), and initially we
340
+ consider the whole set of states. We repeatedly apply the forward search (not the backward
341
+ search), until the forward search returns the empty interval for some i β‰₯ 0. This means that
342
+ Ο€[1..i+1] does not occur in A. Then, if T(Ο€[1..i]) is the whole set of states, we conclude that
343
+ the character Ο€[i + 1] labels no edge in the graph. Otherwise, we must find the smallest j
344
+ such that T(Ο€[1..i]) is strictly contained in T(Ο€[j..i]) (that is, we must determine the longest
345
+ suffix Ο€[j..i] of Ο€[1..i] which reaches more states than Ο€[1..i]). Then we must determine the
346
+ endpoints of the interval T(Ο€[j..i]) so that we can go on with the forward search.
347
+ The challenge now is to find a way to solve the same subproblems that we identified in
348
+ Case 2 of the algorithm for strings. In other words, we must find a way to determine j and
349
+ find the endpoints of the interval T(Ο€[j..i]). We will show that the solution is not as simple
350
+ as the one for the algorithm on strings.
351
+ The LCP array and matching statistics for Wheeler DFAs
352
+ We start observing that Iui may be an infinite set. For example, in Figure 2, we have
353
+ Iu2 = {aaaaa . . . , abdf### . . . , aabdf### . . . , aaabdf### . . . , . . . }.
354
+ In general, an infinite set of (lexicographically sorted) strings in Σω need not admit a
355
+ minimum or a maximum. For example, the set {baaaa . . . , abaaa . . . , aabaa . . . , aaaba . . . }
356
+ does not admit a minimum (but only the infimum string aaaaa . . . ). Nonetheless, Lemma 3
357
+ implies that each Iui admits both a minimum and a maximum. For example, the minimum
358
+ is obtained as follows. Let f1 = i, and for every k β‰₯ 1, recursively let fk+1 be the smallest
359
+ integer in [1, n] such that (ufk+1, ufk) ∈ E. Then, the minimum of Iui is λ(uf1)λ(uf2) . . . ,
360
+ and analogously one can determine the maximum.
361
+ In the following, we will denote the minimum and the maximum of Iui by mini and maxi,
362
+ respectively (for example, in Figure 2 we have min2 = aaaaa . . . , and max2 = abdf### . . . ).
363
+ Lemma 3 implies that:
364
+ min1 βͺ― max1 βͺ― min2 βͺ― max2 βͺ― Β· Β· Β· βͺ― maxnβˆ’1 βͺ― minn βͺ― maxn.
365
+ This suggests to generalize the LCP array as follows. Given Ξ±, Ξ² ∈ Ξ£βˆ— βˆͺ Σω, let lcp(Ξ±, Ξ²) be
366
+ the length of the longest common prefix of Ξ± and Ξ² (if Ξ± = Ξ² ∈ Σω, define lcp(Ξ±, Ξ²) = ∞).
367
+ Definition 4. The LCP-array of a Wheeler automaton A is the array LCPA = LCPA[2, 2n]
368
+ which contains the following 2n βˆ’ 1 values in this order: lcp(min1, max1), lcp(max1, min2),
369
+ lcp(min2, max2), . . . , lcp(maxnβˆ’1, minn), lcp(minn, maxn).
370
+
371
+ From the above characterization of mini and maxi, one can prove that for every entry
372
+ either LCPA[i] = ∞ or LCPA[i] < 3n (it follows from Fine and Wilf Theorem [18,19]), and
373
+ one can design a polynomial time algorithm to compute LCPA.
374
+ Unfortunately, the array LCPA alone is not sufficient for computing matching statistics.
375
+ Assume that T(Ο€) = {ur, ur+1, . . . , usβˆ’1, us}, and that when we apply the forward search by
376
+ adding a character c, we obtain T(Ο€c) = βˆ…. We must then determine the largest suffix Ο€β€²
377
+ of T(Ο€) such that T(Ο€) is strictly contained in T(Ο€β€²). Suppose that every string in Iur is
378
+ prefixed by Ο€R, and every string in Ius is prefixed by Ο€R. In particular, both minr and maxs
379
+ are prefixed by Ο€R. In this case, we can proceed like in the algorithm for strings: the desired
380
+ suffix Ο€β€² is the one having length max{lcp(maxrβˆ’1, minr), lcp(maxs, mins+1)}, which can be
381
+ determined using LCPA. However, in general, even if some string in Iur must be prefixed
382
+ by Ο€R, the string minr need not be prefixed by Ο€R, and similarly maxs need not be prefixed
383
+ by Ο€R. The worst-case scenario occurs when r = s. Consider Figure 2, and assume that
384
+ Ο€ = heba. Then, we have r = s = 3 (note that abeh### . . . is a string in Iu3 prefixed by
385
+ Ο€R). However, both min3 = abdg### . . . , and max3 = acei### . . . , are not prefixed by
386
+ Ο€R. Notice that lcp(max2, min3) = 3 and lcp(max3, min4) = 3, but Ο€β€² is not the suffix of
387
+ length 3 of Ο€. Indeed, since min3 is only prefixed by the prefix of Ο€R of length 2, and max3
388
+ is only prefixed by the prefix of Ο€R of length 1, we conclude that it must be |Ο€β€²| = 2. In
389
+ general, the desired suffix Ο€β€² is the one having length |Ο€β€²| given by:
390
+ max
391
+ οΏ½
392
+ min{lcp(maxrβˆ’1, minr),lcp(minr, Ο€R)}, min{lcp(Ο€R, maxs),lcp(maxs, mins+1)}
393
+ οΏ½
394
+ . (1)
395
+ The above formula shows that, in order to compute Ο€β€², in addition to LCPA it suffices to
396
+ know the values lcp(minr, Ο€R) and lcp(Ο€R, maxs) (Ο€β€² is a suffix of Ο€, so it is determined by
397
+ its length). We now show how our algorithm can efficiently maintain the current pattern Ο€,
398
+ the set T(Ο€) = {ur, ur+1, . . . , usβˆ’1, us} and the values lcp(minr, Ο€R) and lcp(Ο€R, maxs) during
399
+ the computation of the matching statistics. We assume that the input automaton is encoded
400
+ with the rank/select data structures supporting the execution of a step of forward search in
401
+ O(log |Ξ£|) time, see [6] for details. In addition, we will use the following result.
402
+ Lemma 5. Let A[1, n] be a sequence of values over an ordered alphabet Ξ£. Consider the
403
+ following queries: (i) given i, j ∈ [1..n], compute the minimum value in S[i..j], and (ii)
404
+ given t ∈ [1..n] and c ∈ Σ, determine the biggest k < t (or the smallest k > t) such that
405
+ A[k] < c. Then, A can be augumented with a data structure of 2n+o(n) bits such that query
406
+ (i) can be answered in constant time and query (ii) can be answered in O(log n) time.
407
+ Proof. There exists a data structure of 2n + o(n) bits that allows to solve range minimum
408
+ queries in constant time [20], so using A we can solve queries (i) in constant time. Now, let
409
+ us show how to solve queries (ii). Let f1 be the answer of query (i) on input i = ⌈t/2βŒ‰ and
410
+ j = t βˆ’ 1. If f1 < c, then we must keep searching in the interval [⌈t/2βŒ‰, t βˆ’ 1], otherwise, we
411
+ must keep searching in the interval [1, ⌈t/2βŒ‰ βˆ’ 1]. In other words, we can answer a query (ii)
412
+ by means of a binary search on [1, t βˆ’ 1], which takes O(log t) (and so O(log n)) time.
413
+ Notice that query (ii) can be seen as a variant of PSV and NSV queries. In the following,
414
+ we assume that the array LCPA has been augmented with the data structure of Lemma 5.
415
+ At the beginning we have Ο€ = Η«, so T(Η«) = {1, 2, . . . , n} and trivially lcp(minr, Ο€R) =
416
+ lcp(Ο€R, maxs) = 0. At each iteration we perform a step of forward search computing T(Ο€c)
417
+ given T(Ο€); then we distinguish two cases according to whether T(Ο€c) is empty or not.
418
+
419
+ Case 1. T(Ο€c) = {urβ€², urβ€²+1, . . . , usβ€²βˆ’1, usβ€²} is not empty. In that case Ο€c will become the
420
+ pattern at the next iteration. Since we already have T(Ο€c) we are left with the task of com-
421
+ puting lcp(minrβ€², cΟ€R) and lcp(cΟ€R, maxsβ€²). We only show how to compute lcp(minrβ€², cΟ€R),
422
+ the latter computation being analogous. Let k be the smallest integer in [1, n] such that
423
+ (uk, urβ€²) ∈ E. Notice that we can easily compute k by means of standard rank/select opera-
424
+ tions on the compact data structure used to encode A. Since urβ€² ∈ T(Ο€c), it must be k ≀ s.
425
+ Moreover, the characterization of minrβ€² that we described above implies that minrβ€² = c mink,
426
+ hence lcp(minrβ€², cΟ€R) = lcp(c mink, cΟ€R) = 1 + lcp(mink, Ο€R). To compute lcp(mink, Ο€R) we
427
+ distinguish two subcases:
428
+ a) k > r, hence r < k ≀ s. Since ur, us ∈ T(Ο€), there exist Ξ± ∈ Iur and Ξ² ∈ Ius both
429
+ prefixed by Ο€R. But Ξ± βͺ― maxr βͺ― mink βͺ― mins βͺ― Ξ², so mink is also prefixed by Ο€R,
430
+ and we conclude lcp(mink, Ο€R) = |Ο€|.
431
+ b) k ≀ r. In this case, we have mink βͺ― maxk βͺ― mink+1 β‰Ί maxk+1 βͺ― Β· Β· Β· βͺ― minr β‰Ί Ο€R,
432
+ and therefore lcp(mink, Ο€R) is equal to
433
+ min{lcp(mink, maxk), lcp(maxk, mink+1), lcp(mink+1, maxk+1), . . . , lcp(minr, Ο€R)}.
434
+ With the above formula we can compute lcp(mink, Ο€R) using query (i) of Lemma 5 over
435
+ the range LCPA[2k, 2r βˆ’ 1] and the value lcp(minr, Ο€R).
436
+ Case 2. T(Ο€c) is empty. In this case at the next iteration the pattern will be largest suffix
437
+ Ο€β€² of Ο€ such that T(Ο€) is strictly contained in T(Ο€β€²) = {urβ€²β€², . . . , usβ€²β€²}. We compute |Ο€β€²|
438
+ using (1); if |Ο€β€²| > lcp(minr, Ο€R) we set rβ€²β€² = r, otherwise we apply query (ii) of Lemma 5 to
439
+ find the rightmost entry rβ€²β€² in LCPA[2, 2r βˆ’ 1] smaller than |Ο€β€²|. Computing sβ€²β€² is analogous.
440
+ Given T(Ο€β€²) = {urβ€²β€², urβ€²β€²+1, . . . , usβ€²β€²βˆ’1, usβ€²β€²}, where rβ€²β€² ≀ r, s ≀ sβ€²β€², and at least one inequal-
441
+ ity is strict, we want to compute lcp(minrβ€²β€², (Ο€β€²)R) and lcp((Ο€β€²)R, maxsβ€²β€²). We only consider
442
+ lcp(minrβ€²β€², (Ο€β€²)R), the latter computation being analogous. We distinguish two subcases:
443
+ a) rβ€²β€² = r. Then lcp(minrβ€²β€², (Ο€β€²)R) = lcp(minr, (Ο€β€²)R) = min{lcp(minr, Ο€R), |Ο€β€²|}.
444
+ b) rβ€²β€² < r. In particular, since urβ€²β€² is the left endpoint of T(Ο€β€²) and |T(Ο€β€²)| β‰₯ 2, one can
445
+ prove like in Case 1a) that maxrβ€²β€² is prefixed by (Ο€β€²)R. We immediately conclude that
446
+ lcp(minrβ€²β€², (Ο€β€²)R) = min{lcp(minrβ€²β€², maxrβ€²β€²), |Ο€β€²|}, which can be immediately computed
447
+ since lcp(minrβ€²β€², maxrβ€²β€²) is a value stored in LCPA.
448
+ We can summarize the above discussion as follows.
449
+ Theorem 6. Given a Wheeler DFA A, there exists a data structure occupying O(|A|) words
450
+ which can compute the pattern matching statistics of a pattern P in time O(|P| log |A|).
451
+ Funding
452
+ TG funded by National Institutes of Health (NIH) NIAID (grant no. HG011392),
453
+ the National Science Foundation NSF IIBR (grant no. 2029552) and a Natural Science and
454
+ Engineering Research Council (NSERC) Discovery Grant (grant no. RGPIN-07185-2020).
455
+ GM funded by the Italian Ministry of University and Research (PRIN 2017WR7SHH). MS
456
+ funded by the INdAM-GNCS Project (CUP E55F22000270001). NP funded by the European
457
+ Union (ERC, REGINDEX, 101039208). Views and opinions expressed are however those
458
+ of the author(s) only and do not necessarily reflect those of the European Union or the
459
+ European Research Council. Neither the European Union nor the granting authority can be
460
+ held responsible for them.
461
+
462
+ References
463
+ [1] P. Ferragina and G. Manzini, β€œOpportunistic data structures with applications,” in Proc. 41st
464
+ Annual Symposium on Foundations of Computer Science (FOCS’00), 2000, pp. 390–398.
465
+ [2] M. Burrows and D. J. Wheeler, β€œA block-sorting lossless data compression algorithm,” Tech.
466
+ Rep., 1994.
467
+ [3] P. Ferragina, F. Luccio, G. Manzini, and S. Muthukrishnan, β€œStructuring labeled trees for
468
+ optimal succinctness, and beyond,” in proc. 46th Annual IEEE Symposium on Foundations of
469
+ Computer Science (FOCS’05), 2005, pp. 184–193.
470
+ [4] A. Bowe, T. Onodera, K. Sadakane, and T. Shibuya, β€œSuccinct de Bruijn graphs,” in Algo-
471
+ rithms in Bioinformatics, Berlin, Heidelberg, 2012, pp. 225–235, Springer Berlin Heidelberg.
472
+ [5] V. MΒ¨akinen, N. VΒ¨alimΒ¨aki, and J. SirΒ΄en, β€œIndexing graphs for path queries with applications in
473
+ genome research,” IEEE/ACM Transactions on Computational Biology and Bioinformatics,
474
+ vol. 11, pp. 375–388, 2014.
475
+ [6] T. Gagie, G. Manzini, and J. SirΒ΄en, β€œWheeler graphs: A framework for BWT-based data
476
+ structures,” Theoret. Comput. Sci., vol. 698, pp. 67 – 78, 2017.
477
+ [7] J. Alanko, G. D’Agostino, A. Policriti, and N. Prezza, β€œRegular languages meet prefix sorting,”
478
+ in Proc. of the 31st Symposium on Discrete Algorithms, (SODA’20). 2020, pp. 911–930, SIAM.
479
+ [8] N. Cotumaccio and N. Prezza, β€œOn indexing and compressing finite automata,” in Proc. of
480
+ the 32nd Symposium on Discrete Algorithms, (SODA’21). 2021, pp. 2585–2599, SIAM.
481
+ [9] N. Cotumaccio, β€œGraphs can be succinctly indexed for pattern matching in O(|E|2 + |V |5/2)
482
+ time,” in 2022 Data Compression Conference (DCC), 2022, pp. 272–281.
483
+ [10] J. Alanko, G. D’Agostino, A. Policriti, and N. Prezza, β€œWheeler languages,” Information and
484
+ Computation, vol. 281, pp. 104820, 2021.
485
+ [11] D. Gusfield, Algorithms on Strings, Trees, and Sequences: Computer Science and Computa-
486
+ tional Biology, Cambridge University Press, 1997.
487
+ [12] W. I. Chang and E. L. Lawler, β€œSublinear approximate string matching and biological appli-
488
+ cations,” Algorithmica, vol. 12, pp. 327–344, 2005.
489
+ [13] P. Weiner, β€œLinear pattern matching algorithms,” in Proc. 14th IEEE Annual Symposium on
490
+ Switching and Automata Theory, 1973, pp. 1–11.
491
+ [14] U. Manber and G. Myers, β€œSuffix arrays: A new method for on-line string searches,” SIAM
492
+ J. Comput., vol. 22, no. 5, pp. 935–948, 1993.
493
+ [15] M. I. Abouelhoda, S. Kurtz, and E. Ohlebusch, β€œReplacing suffix trees with enhanced suffix
494
+ arrays,” J. of Discrete Algorithms, vol. 2, no. 1, pp. 53–86, 2004.
495
+ [16] K. Sadakane, β€œCompressed suffix trees with full functionality,” Theor. Comp. Sys., vol. 41,
496
+ no. 4, pp. 589–607, 2007.
497
+ [17] E. Ohlebusch, S. Gog, and A. KΒ¨ugell, β€œComputing matching statistics and maximal exact
498
+ matches on compressed full-text indexes,” in Proceedings of the 17th International Confer-
499
+ ence on String Processing and Information Retrieval (SPIRE’10), Berlin, Heidelberg, 2010, p.
500
+ 347–358, Springer-Verlag.
501
+ [18] N. J. Fine and H. S. Wilf, β€œUniqueness theorem for periodic functions,” Proc. Amer. Math.
502
+ Soc., , no. 16, pp. 109–114, 1965.
503
+ [19] S. Mantaci, A. Restivo, G. Rosone, and M. Sciortino, β€œAn extension of the Burrows-Wheeler
504
+ transform,” Theor. Comput. Sci., vol. 387, no. 3, pp. 298–312, 2007.
505
+ [20] Johannes Fischer,
506
+ β€œOptimal succinctness for range minimum queries,”
507
+ in LATIN 2010:
508
+ Theoretical Informatics, Alejandro LΒ΄opez-Ortiz, Ed., Berlin, Heidelberg, 2010, pp. 158–169,
509
+ Springer Berlin Heidelberg.
510
+
JdE4T4oBgHgl3EQf7Q53/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
JtE0T4oBgHgl3EQfSADl/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f59368c19d801f0e2558e91d5333ee73c6a2f9a032a1cf6896a1d13bee534513
3
+ size 3735597
JtE0T4oBgHgl3EQfiQGP/content/tmp_files/2301.02443v1.pdf.txt ADDED
@@ -0,0 +1,1720 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 1
2
+
3
+ Exploring Euroleague History using Basic Statistics
4
+
5
+
6
+
7
+ Christos Katris1,2
8
+
9
+ 1Adjunct Lecturer, Department of Mathematics, University of Patras
10
+ 2Customs Officer (Statistician), Independent Authority for Public Revenue, Greece
11
+
12
+ 1chriskatris@upatras.gr, 2c.katris1@aade.gr
13
+
14
+
15
+
16
+
17
+
18
+
19
+
20
+
21
+
22
+
23
+
24
+
25
+
26
+ 2
27
+
28
+ Abstract
29
+
30
+ In this paper are used historical statistical data to track the evolution of the game in the European-wide
31
+ top-tier level professional basketball club competition (until 2017-2018 season) and also are answered
32
+ questions by analyzing them. The term basic is referred because of the nature of the data (not available
33
+ detailed statistics) and of the level of aggregation (not disaggregation to individual level). We are
34
+ examining themes such as the dominance per geographic area, the level of the competition in the game,
35
+ the evolution of scoring pluralism and possessions in the finals, the effect of a top scorer in the
36
+ performance of a team and the existence of unexpected outcomes in final fours. For each theme under
37
+ consideration, available statistical data is specified and suitable statistical analysis is applied. The analysis
38
+ allows us to handle and answer the above themes and interesting conclusions are drawn. This paper can
39
+ be an example of statistical thinking in basketball problems by the means of using efficiently available
40
+ statistical data.
41
+
42
+ Keywords: Statistical analysis, basketball statistics, Euroleague evolution.
43
+
44
+ 1. Introduction
45
+ The field of basketball is ideal for the application of statistical methods in order to extract useful
46
+ conclusions which can help in analyzing many aspects of the game. The origin of many ideas is from
47
+ persons outside academia. The book of Oliver (2004) was a worthy attempt to develop and apply
48
+ statistical concepts in the area of basketball. Much information is included on this book and can offer to a
49
+ reader a statistical way of thinking for the game of basketball. There are also many academic papers
50
+ which use advanced statistical methods for basketball analysis in themes such as performance evaluation
51
+ of players and teams, home advantage effect etc. The field of basketball analytics is not yet entirely
52
+ unified and new ideas which are based on quantitative analysis are appearing continuously from diverse
53
+ academic fields. In many cases, there are used advanced statistics for the analysis of many situations. The
54
+ majority of studies – not only with USA origin - are related to NBA and this is not just a coincidence. The
55
+
56
+ 3
57
+
58
+ tracking system of statistics is superior to other leagues in terms of quality (calculation of more advanced
59
+ statistics) and quantity (calculation of more statistical categories), and the discrepancy was larger
60
+ especially in the past.
61
+
62
+ The paper of Kubatko et al (2007) presents the general accepted basics of the analysis of basketball.
63
+ Furthermore, most of the statistics are based on the concept of possessions. However, this is not the case
64
+ for other leagues, including Euroleague. Given the available statistical data is difficult or even impossible
65
+ (for older years) to calculate neither possessions nor advanced statistics. Only after 2001 in the modern
66
+ era of Euroleague, plenty of statistics are available.
67
+
68
+ This paper is an attempt to utilize available statistical information through statistical analysis in order to
69
+ explore the evolution of the game in Euroleague. It is demonstrated that even simple available statistics
70
+ can offer insights about the game and can be extracted useful conclusions. Graphical analysis, statistical
71
+ hypothesis testing and correlation measures are our weapons in this chase of insights related to the
72
+ evolution of Euroleague. The next section is a brief description of Euroleague and are referred the sources
73
+ of statistical data. Section 3 is the main part of the paper and contains statistical analysis and methods to
74
+ deal with questions related to the historical evolution of the tournament. Finally, in Section 4 are
75
+ presented the conclusions of the analysis.
76
+
77
+ 2. A Brief History of Euroleague and Statistical Data
78
+ In this paper is examined the evolution of the European-wide top-tier level professional basketball club
79
+ competition. Briefly the history of the competition is following. The FIBA European Champions Cup
80
+ competition has established in 1958 and FIBA was organizing its operation until 2000. Then Euroleague
81
+ Basketball was created. The next year, the two competitions were unified again under the umbrella of
82
+ Euroleague Basketball (for more details: https://en.wikipedia.org/wiki/EuroLeague). Also the competition
83
+ has changed names across time. From 1958 to 1991 was the FIBA European Champions Cup, from 1991
84
+ to 1996 the name of the competition was FIBA European League, from 1996 to 2000 the name was FIBA
85
+
86
+ 4
87
+
88
+ Euroleague. In season 2000-2001 there were 2 competitions: FIBA Suproleague which was organized by
89
+ FIBA and Euroleague which was organized by Euroleague Basketball. From the next year there was a
90
+ unique competition for the top-tier level under the name Euroleague which was organized by Euroleague
91
+ Basketball. In 2016 the name changed to EuroLeague. For the rest of the article the name Euroleague is
92
+ used for the whole competition. The concept of final four applied for 1965-1966 and 1966-1967 seasons
93
+ and was included permanently in the competition from the season of 1987-1988. In this paper, we
94
+ consider as final four teams before 1986-1987, the teams which have reached the semi-finals in order to
95
+ generate a consistent system for studying the evolution of the tournament.
96
+
97
+ There is not a unique data source which contains all information from the beginning of the tournament in
98
+ 1958. Statistical data sources which were used are: Wikipedia, http://pearlbasket.altervista.org,
99
+ http://www.linguasport.com and http://www.fibaeurope.com/ and http://www.euroleague.net/ for stats
100
+ after 2001.
101
+
102
+ 3. Statistical Analysis of Euroleague Historical Data
103
+ In this section is made an attempt to shed light to questions related to the historical evolution of the game
104
+ with the use of suitable statistical methods. The graphs are created in excel, whilst for the implementation
105
+ of the methods is used statistical software R.
106
+
107
+
108
+ 3.1 Dominance on the Game per Geographic Area
109
+ Firstly, we can derive some quick conclusions about the dominance in the game in terms of geographic
110
+ location. Table 1 displays per country the winners, the runners-up and the number of teams which had
111
+ appeared to final fours. We consider only the teams which participated to final fours since 1958. From
112
+ this limited statistical information we will explore briefly the game over time.
113
+
114
+
115
+ 5
116
+
117
+ Based on Table 1, we consider the following Geographic areas: Spain and Italy which are leading the
118
+ table in all categories are considered separately, Ex USSR and ex Yugoslavian countries form the next
119
+ area and every other country is assigned to a fourth group (other).
120
+ Fig.1 displays the titles per time period of teams from each geographic area and Fig.2 displays the
121
+ appearances in final fours of teams from each geographic area.
122
+
123
+ Table 1. Titles and appearances per country
124
+ Country
125
+ Winner
126
+ Runner-Up
127
+ Final Four Appearances
128
+ Number of Teams
129
+ Spain
130
+ 13
131
+ 16
132
+ 57
133
+ 6
134
+ Italy
135
+ 13
136
+ 13
137
+ 44
138
+ 9
139
+ Greece
140
+ 9
141
+ 7
142
+ 13
143
+ 5
144
+ Russia1
145
+ 7
146
+ 6
147
+ 17
148
+ 2
149
+ Israel
150
+ 6
151
+ 9
152
+ 20
153
+ 1
154
+ Croatia2
155
+ 5
156
+ 1
157
+ 3
158
+ 3
159
+ Latvia1
160
+ 3
161
+ 1
162
+ 4
163
+ 1
164
+ Turkey
165
+ 1
166
+ 2
167
+ 6
168
+ 2
169
+ Lithuania1
170
+ 1
171
+ 1
172
+ 3
173
+ 1
174
+ Georgia1
175
+ 1
176
+ 1
177
+ 3
178
+ 1
179
+ Bosnia2
180
+ 1
181
+ 0
182
+ 4
183
+ 1
184
+ Serbia2
185
+ 1
186
+ 0
187
+ 10
188
+ 4
189
+ France
190
+ 1
191
+ 0
192
+ 9
193
+ 4
194
+ Czech Republic3
195
+ 0
196
+ 3
197
+ 9
198
+ 2
199
+ Bulgaria
200
+ 0
201
+ 2
202
+ 2
203
+ 1
204
+ Slovenia2
205
+ 0
206
+ 0
207
+ 3
208
+ 1
209
+ Poland
210
+ 0
211
+ 0
212
+ 2
213
+ 2
214
+ Romania
215
+ 0
216
+ 0
217
+ 1
218
+ 1
219
+ Netherlands
220
+ 0
221
+ 0
222
+ 1
223
+ 1
224
+ Hungary
225
+ 0
226
+ 0
227
+ 1
228
+ 1
229
+ Belgium
230
+ 0
231
+ 0
232
+ 1
233
+ 1
234
+ 1 Trophies won before 1991 were under the umbrella of Soviet Union
235
+ 2 Trophies won before 1995 were under the umbrella of Yugoslavia
236
+ 3 Trophies won before 1991 were under the umbrella of Czechoslovakia
237
+
238
+
239
+
240
+
241
+
242
+
243
+
244
+
245
+
246
+
247
+
248
+
249
+
250
+
251
+
252
+
253
+
254
+
255
+ 6
256
+
257
+ Fig.1. Titles evolution per geographic area
258
+
259
+
260
+
261
+ Fig.2. Appearances to Final Four per geographic area
262
+
263
+
264
+
265
+
266
+
267
+
268
+ 1958-1970
269
+ 1971-1980
270
+ 1981-1990
271
+ 1991-2000
272
+ 2001-2010
273
+ 2011-2018
274
+ Spain
275
+ 4
276
+ 3
277
+ 0
278
+ 2
279
+ 2
280
+ 2
281
+ Italy
282
+ 2
283
+ 4
284
+ 5
285
+ 1
286
+ 1
287
+ 0
288
+ Ex USSR and Yugoslavian
289
+ 7
290
+ 2
291
+ 4
292
+ 3
293
+ 2
294
+ 1
295
+ Other
296
+ 0
297
+ 1
298
+ 1
299
+ 4
300
+ 6
301
+ 5
302
+ 0
303
+ 1
304
+ 2
305
+ 3
306
+ 4
307
+ 5
308
+ 6
309
+ 7
310
+ 8
311
+ Titles
312
+ Titles Evolution per Geographic Area
313
+ 1958-1970
314
+ 1971-1980
315
+ 1981-1990
316
+ 1991-2000
317
+ 2001-2010
318
+ 2011-2018
319
+ Spain
320
+ 10
321
+ 9
322
+ 7
323
+ 11
324
+ 10
325
+ 10
326
+ Italy
327
+ 6
328
+ 12
329
+ 9
330
+ 7
331
+ 9
332
+ 1
333
+ Ex USSR and Yugoslavian
334
+ 20
335
+ 9
336
+ 12
337
+ 6
338
+ 10
339
+ 9
340
+ Other
341
+ 16
342
+ 10
343
+ 12
344
+ 16
345
+ 15
346
+ 12
347
+ 0
348
+ 5
349
+ 10
350
+ 15
351
+ 20
352
+ 25
353
+ Appearances
354
+ Appearances to Final Four per geographic area
355
+
356
+ 7
357
+
358
+ To test formally if there are significant differences to the appearances and to the titles per geographic
359
+ area, we perform Friedman tests with titles (or appearances) per geographic area as treatments and time
360
+ periods as blocks (a blocking factor is a source of variability which is not of primary interest). We want to
361
+ check for significant differences to the titles and appearances per geographic area. Note that we want to
362
+ overall check the titles and appearances and not the trend, and we consider the time periods as blocks in
363
+ order to reduce their effect to the variability of titles and appearances respectively. The non-parametric
364
+ Friedman test is used in order not to have distributional assumptions, because normality assumption (data
365
+ to follow normal distribution) does not seem very likely. Details about the test can be found in every book
366
+ of non-parametric statistics such as that of (Hollander and Wolfe, 1999).
367
+
368
+ The null hypothesis (𝐻0) is that apart of the effect of time period (blocks) there is no difference in titles
369
+ (or appearances) are even between the considered regions. The level of significance is considered at 5%
370
+ (0.05). To reject the null hypothesis, the p-value should be less than 0.05.
371
+ Friedman Test
372
+
373
+ Statistic
374
+ df
375
+ p-value
376
+ Appearances
377
+ 6.5789
378
+ 3
379
+ 0.0866
380
+ Titles
381
+ 0.7627
382
+ 3
383
+ 0.8584
384
+
385
+ From the application of the test we do not have enough evidence to suppose significant differences
386
+ between the performance of geographic areas in appearances and titles. However there are trends which
387
+ have been described graphically and discussed previously.
388
+
389
+
390
+ 3.2 Dominance of the Champion
391
+ Next, is examined the dominance of the champion to its opponents and is measured in terms of scoring
392
+ points. The considered data are the points per game for and against the champions after the quarterfinals
393
+ because the potential existence of weak teams in earlier rounds may lead to instability of point
394
+ performance. In Fig.3 there are displayed the Points per Game (PPG) of the champion team and of their
395
+ opponents, while Fig.4 displays the Point difference as % of the points of the opponents of the champion
396
+
397
+ 8
398
+
399
+ team. This is considered as a metric of the dominance of the champion team against its opponents in
400
+ terms of scoring.
401
+
402
+
403
+ Fig.3. PPG for and against the champion team
404
+
405
+
406
+
407
+
408
+
409
+
410
+ Fig.4. Point difference as % of the opponents of the champion team
411
+
412
+
413
+
414
+
415
+ 30.00%
416
+ 40.00%
417
+ 50.00%
418
+ 60.00%
419
+ 70.00%
420
+ 80.00%
421
+ 90.00%
422
+ 1958-1970 1971-1980 1981-1990 1991-2000 2001-2010 2011-2018
423
+ CR4
424
+ CR5
425
+ CR6
426
+ CR7
427
+ CR8
428
+ 0
429
+ 20
430
+ 40
431
+ 60
432
+ 80
433
+ 100
434
+ 120
435
+ 1958 1962 1966 1970 1974 1978 1982 1986 1990 1994 1998 2001 2005 2009 2013 2017
436
+ Points per Game
437
+ Points per game for and against the Champions
438
+ champion
439
+ opponent
440
+
441
+ 9
442
+
443
+ The above graph displays the points scored by the champion minus the points scored by opponents on
444
+ average, as percentage of opponent points. This could show in a sense how dominant was a champion.
445
+ Only in six seasons the champions scored more than 20% of their opponent points with Real Madrid to be
446
+ the only team which scored more than 30% of their opponents’ points in 1978. Extreme cases like this
447
+ should be examined in more detail. For example this team scored only 75 points in the final.
448
+ To better track the change of the game over time, we calculate the average points of the champions and
449
+ opponents in every decade and the average points per team and we draw their evolution across time in
450
+ Table 2 and Fig.5.
451
+ Table 2. Euroleague For and Against Points for Champion after Quarterfinals on average
452
+ Time Period
453
+ Average Points per Time Period
454
+ champion
455
+ opponent
456
+ Points per Team
457
+ 1958-1970
458
+ 82.61
459
+ 72.90
460
+ 77.75
461
+ 1971- 1980
462
+ 91.20
463
+ 77.88
464
+ 84.54
465
+ 1981-1990
466
+ 88.27
467
+ 81.92
468
+ 85.09
469
+ 1991-2000
470
+ 72.92
471
+ 65.61
472
+ 69.27
473
+ 2001-2010
474
+ 82.94
475
+ 74.38
476
+ 78.66
477
+ 2011-2018
478
+ 81.86
479
+ 75.18
480
+ 78.52
481
+
482
+
483
+
484
+ Fig.5. Evolution of Average Points per team
485
+
486
+
487
+
488
+
489
+
490
+ 60
491
+ 65
492
+ 70
493
+ 75
494
+ 80
495
+ 85
496
+ 90
497
+ Points per Team
498
+ Points per Team
499
+
500
+ 10
501
+
502
+
503
+ From the above graph we can see the changes of the mentality of the game across time. From 1958 to
504
+ 1990 there was an upward trend in scoring, with a sudden drop in 90s, something which indicates a
505
+ significant change in game mentality, and a return to the levels of 1958-1970 period.
506
+ Finally, the Fig.6 displays the decade average of the difference of points scored by champions minus the
507
+ points scored by opponents as percentage of the opponent points. It’s an indication of the dominance of
508
+ the champions of every decade. On average, the champions were more dominant in 60s and 70s, but in
509
+ the 80s they scored only 8% more than their opponents, a clear sign that the competition was more intense
510
+ in this decade. We also notice that the competitiveness of the last years (2010-2018) tends to similar
511
+ levels.
512
+
513
+ Fig.6. Evolution of Point difference as % of the opponents of the champion team
514
+
515
+
516
+
517
+
518
+
519
+
520
+ 3.3 Analyze Scoring Pluralism in the Finals: Evolution of the style of the game
521
+ In this subsection we want to follow the evolution of the game as pictured in finals. We use raw data which
522
+ are the first scorers and the team points of the finals since 1958. It is commonly assumed that in the runner-up
523
+ team there is a more dominant scorer, in terms of first scorer points as % of team points. To test this
524
+ hypothesis we perform a Wilcoxon rank sum test for pairs of observations for data from all finals since
525
+ 1958. The null hypothesis (𝐻0) of the test is that the differences between the pairs follows a symmetric
526
+ 0.00%
527
+ 2.00%
528
+ 4.00%
529
+ 6.00%
530
+ 8.00%
531
+ 10.00%
532
+ 12.00%
533
+ 14.00%
534
+ 16.00%
535
+ 18.00%
536
+ 1958-1970 1971- 1980 1981-1990 1991-2000 2001-2010 2011-2018
537
+ Dominance of the Champion
538
+ Point Difference pct.
539
+
540
+ 11
541
+
542
+ distribution around zero. This is a non parametric test and through its application we avoid the
543
+ distributional assumption of normality of the data. A detailed description of the test can be found in
544
+ (Hollander and Wolfe, 1999). The test suggests that there is no reason to assume that in a specific year is
545
+ more probable the runner-up team to have a more dominant player in the scoring in final.
546
+
547
+ Wilcoxon Signed Rank Test for paired Samples
548
+ Statistic
549
+ p-value
550
+ 824
551
+ 0.1494
552
+
553
+ Additionally, we explore whether first scorer in terms of % of team points appears randomly or is more
554
+ probable to appear in sequences either from the champion or from the runner-up team. This can be
555
+ achieved through the application of a runs test in the difference of first scorer points as % of team points
556
+ between the two teams and we generate from this variable a sequence of + signs (if the variable is larger
557
+ than a threshold) and – signs (if the variable is smaller than a threshold). A run of a sequence is defined as
558
+ a series consisting of adjacent equal elements. We are testing the null hypothesis (H0) that each element
559
+ in the sequence is independently drawn from the same distribution. The threshold in our case is set to
560
+ zero. A description of the test can be found in (Gibbons and Chakraborti, 2003) and its implementation
561
+ performed via the randtests package of R (Caeiro and Mateus, 2014).
562
+ Through the application of the test, we can decide if over and under zero values are random. There is no
563
+ sign of non-randomness for this variable, so we can assume that the first scorer appears randomly from
564
+ the champion or from the runner-up team.
565
+ Runs Test
566
+ Statistic
567
+ Observations>0
568
+ Observations<0
569
+ Runs
570
+ p-value
571
+ ~0
572
+ 24
573
+ 40
574
+ 31
575
+ ~1
576
+
577
+ The above tests are for the whole time period and they don’t reveal anything about the evolution of the game. The
578
+ rest of this section examines the evolution of the game and the statistical tests are adjusted accordingly.
579
+
580
+ At first, we present graphically the 10 year moving average of the points scored by the first scorer in final as % of
581
+ team points in Fig.7. It is displayed the 10 year moving average for decreasing the effect of extreme cases
582
+ and is easier to follow the trend of the game.
583
+
584
+ 12
585
+
586
+ Fig.7. Moving average (10-year) of the points scored by the first scorer in final as % of team points
587
+
588
+
589
+
590
+
591
+
592
+
593
+ We observe that the game has been transformed over time from finals with offences which are based on
594
+ top scorers to finals with more pluralism. From the beginning until the 80s the trend was the one player
595
+ star in scoring, but since then, there was a slow but continuing turn to games based on pluralism.
596
+
597
+ At the next step, in Fig. 8 we present graphically the 10 year moving average of the difference between
598
+ the points of first scorer as % of team points for champion team minus the same metric of runner-up team.
599
+
600
+ Fig.8. Moving average (10-year) of the difference of points of first scorers as % of their team points
601
+
602
+
603
+
604
+
605
+ 0%
606
+ 5%
607
+ 10%
608
+ 15%
609
+ 20%
610
+ 25%
611
+ 30%
612
+ 35%
613
+ 40%
614
+ 45%
615
+ 1966-1967
616
+ 1981-1982
617
+ 1996-1997
618
+ 2008-2009
619
+ Points scored by 1st scorer as % of team points
620
+ 10 year Moving Avarage
621
+ -8.00%
622
+ -6.00%
623
+ -4.00%
624
+ -2.00%
625
+ 0.00%
626
+ 2.00%
627
+ 4.00%
628
+ 1966-1967
629
+ 1978-1979
630
+ 1990-1991
631
+ 2002-2003
632
+ 2013-2014
633
+ 10-year Moving Average of difference
634
+ 10-year Moving Average
635
+ of difference
636
+
637
+ 13
638
+
639
+ From 1998 there is a downward trend until 2009 and a new cycle begins after 2009 and evolves but in a
640
+ lower level than the past. After 2001 there is no single year where the champion team has a more
641
+ dominant scorer than the runner-up team in terms of 10 year moving averages.
642
+
643
+ We make an assumption that there is a structural break in this variable and is very important to specify the
644
+ time when it happened because is a clue that the game has changed at this moment. To achieve this, is
645
+ performed a Zivot-Andrew test (Zivot and Andrews, 1992) to test for the existence of a structural break
646
+ (null hypothesis 𝐻0) against the hypothesis of nonstationarity.
647
+
648
+
649
+
650
+ Andrew-Zivot Test*
651
+ Statistic
652
+ p-value
653
+ Potential Break
654
+ -4.3499
655
+ >0.1
656
+ 1997-1998 final
657
+ *We assume both level and linear trend and 5 lags
658
+
659
+
660
+ The existence of a structural break is in favour compared to non-stationarity. Moreover, it is important to
661
+ specify when the structural break occurs. The potential structural break occurs in 1997-1998 final. The
662
+ history of Euroleague can be break into 2 periods: before and after 1998, let’s say after 1998 is the
663
+ modern period of Euroleague. For this reason, we perform again the Wilcoxon sign rank test and the Runs
664
+ test for the modern period of Euroleague. In the modern period of Euroleague, we can assume that there is
665
+ a more dominant scorer in the runner-up team, but we cannot predict this fact for a specific year.
666
+
667
+ Wilcoxon Signed Rank Test for paired Samples
668
+ Statistic
669
+ p-value
670
+ 42.5
671
+ 0.02056
672
+
673
+
674
+
675
+ Runs Test
676
+ Statistic
677
+ Observations>0
678
+ Observations<0
679
+ Runs
680
+ p-value
681
+ 1.5607
682
+ 5
683
+ 12
684
+ 11
685
+ 0.1186
686
+
687
+
688
+
689
+
690
+
691
+
692
+
693
+ 14
694
+
695
+ 3.4 Pace in the Finals: The concept of possessions
696
+ In this subsection we include to our analysis the central concept of possessions (Kubatko et al, 2007).
697
+ Larger number of possessions displays a quicker pace of a game and the intension is to track the evolution
698
+ of the game.
699
+
700
+ We assume that both teams have the same number of possessions, but there is no unique formula for the
701
+ calculation of exact possessions in a game. For this reason, there are considered two formulas and we
702
+ average them in order to approximate more accurately the actual possessions. The used formulas are the
703
+ possessions lost (1) and the possessions gained (2) respectively:
704
+ 𝑃𝑂𝑆𝑆𝑑 = 𝐹𝐺𝐴𝑑 + πœ† Γ— 𝐹𝑇𝐴𝑑 βˆ’ 𝑂𝑅𝐸𝐡𝑑 + 𝑇𝑂𝑑 (1)
705
+ 𝑃𝑂𝑆𝑆𝑑 = 𝐹𝐺𝑀𝑑 + πœ† Γ— 𝐹𝑇𝑀𝑑 + π·π‘…πΈπ΅π‘œ + 𝑇𝑂𝑑 (2)
706
+
707
+ After the calculation of the positions, we perform a line graph for the 5 year moving average of the
708
+ possessions in order to track their evolution (Fig.9). There is a downward trend and stability in low game
709
+ pace in the 90s, but from the beginning of the millennium there is a growing trend in game pace and from
710
+ 2002 only four times there were fewer than 70 positions.
711
+
712
+ Fig.9. Evolution of possessions as 5-year moving average
713
+
714
+
715
+
716
+ 60
717
+ 62
718
+ 64
719
+ 66
720
+ 68
721
+ 70
722
+ 72
723
+ 74
724
+ 76
725
+ 1987-1988
726
+ 1997-1998
727
+ 2006-2007
728
+ 2016-2017
729
+ Possessions
730
+ 5 year moving average
731
+
732
+ 15
733
+
734
+ Considering a break at 1997-1998 we perform a Mann-Whitney test for the equality of possessions before
735
+ and after 1998. This test is a non parametric equivalent of a t-test for comparing the means of 2 groups
736
+ when the data do not follow Normal distribution. Details can be found on (Hollander and Wolfe, 1999).
737
+
738
+ Possessions
739
+ Before 1998
740
+ After 1998
741
+ 66.25
742
+ 71.33
743
+ Mann-Whitney Test for possessions
744
+ Statistic
745
+ p-value
746
+ 84
747
+ 0.01028
748
+
749
+
750
+ We detect a significant difference in possessions before and after 1998 finals at the 5% significance level.
751
+ This result is in accordance with our assumption that the game has changed after 1998 final and supports
752
+ the assumption that the triumph of Zalgiris in 1999 was the start of the change.
753
+
754
+
755
+
756
+ 3.5 Correlation of First Scorer with Team Performance
757
+ Another interesting question is whether the existence of a first scorer of a tournament is correlated with
758
+ the performance of the team. The most popular opinion is that first scorers belong to weak teams which
759
+ do not have offensive many good offensive players. The other opinion is that a very gifted scorer can
760
+ affect the performance of the team positively and relies to the coach to keep the balance of the team. We
761
+ have the first scorers of the tournament after 1992 and we measure the strength of the link between their
762
+ scoring performances (PPG) with the success of their team in the season using the Pearson (r) and
763
+ Spearman (ρ) correlation coefficients (Hollander and Wolfe, 1999; Best and Roberts, 1975). The
764
+ Spearman coefficient is non parametric and correlates the ranks of the variables and assesses monotonic
765
+ relationships between them (instead of linear relationships which are assessed from Pearson correlation).
766
+ Except from the coefficient, we perform a statistical test for testing the null hypothesis that the coefficient
767
+ (either r or ρ) is zero, thus there is not significant correlation between the variables. We assign values for
768
+ the performance of the teams: 1 for regular season, 2 for Top 16, 3 for quarterfinals, 4 for final four, 4.5 if
769
+
770
+ 16
771
+
772
+ the team was runner-up and 5 if the team won the trophy. Table 3 displays the first scorer, the team
773
+ position and the assigned values of the position.
774
+
775
+
776
+ π‘―πŸŽ: 𝝆 = 𝟎 𝒐𝒓 𝒓 = 𝟎
777
+
778
+ Coefficient
779
+ Statistic
780
+ p-value
781
+ Pearson Correlation
782
+ -0.4002
783
+ -2.2271
784
+ 0.03481
785
+ Spearman Correlation
786
+ -0.3925
787
+ 5088.032
788
+ 0.03886
789
+
790
+ Both Pearson and Spearman correlation coefficients indicate that there is a significant negative
791
+ relationship between the first scorer and the performance of the team. This finding rather favors the first
792
+ opinion where the first scorers are rarely parts of top teams (exception of Nando De Colo in 2015-2016
793
+ confirms the general rule).
794
+ Table 3. First scorer of the tournament, team position and assigned values of the position
795
+ Season
796
+ Player
797
+ PPG
798
+ Team
799
+ Performance
800
+ Assigned Score
801
+ 1991-1992
802
+ Nikos Galis
803
+ 32.3
804
+ Aris
805
+ Regular season
806
+ 1
807
+ 1992-1993
808
+ Zdravko Radulović
809
+ 23.9
810
+ Cibona
811
+ Regular season
812
+ 1
813
+ 1993-1994
814
+ Nikos Galis
815
+ 23.8
816
+ Panathinaikos
817
+ 3rd place
818
+ 4
819
+ 1994-1995
820
+ Saőha Danilović
821
+ 22.1
822
+ Buckler Bologna
823
+ Quarterfinals
824
+ 3
825
+ 1995-1996
826
+ Joe Arlauckas
827
+ 26.4
828
+ Real Madrid
829
+ 4th place
830
+ 4
831
+ 1996-1997
832
+ Carlton Myers
833
+ 22.9
834
+ Teamsystem Bologna
835
+ Quarterfinals
836
+ 3
837
+ 1997-1998
838
+ Peja Stojaković
839
+ 20.9
840
+ PAOK
841
+ Top 16
842
+ 2
843
+ 1998-1999
844
+ Δ°brahim Kutluay
845
+ 21.4
846
+ FenerbahΓ§e
847
+ Top 16
848
+ 2
849
+ 1999-2000
850
+ Miljan Goljović
851
+ 20.2
852
+ Pivovarna LaΕ‘ko
853
+ Regular season
854
+ 1
855
+ 2000-2001
856
+ (FIBA)
857
+ Miroslav Berić
858
+ 23.3
859
+ Partizan
860
+ Top 16
861
+ 2
862
+ 2000-2001
863
+ (Euroleague)
864
+ Alphonso Ford
865
+ 26
866
+ Peristeri
867
+ Top 16
868
+ 2
869
+ 2001-2002
870
+ Alphonso Ford
871
+ 24.8
872
+ Olympiacos
873
+ Top 16
874
+ 2
875
+ 2002-2003
876
+ Miloő Vujanić
877
+ 25.8
878
+ Partizan
879
+ Regular season
880
+ 1
881
+ 2003-2004
882
+ Lynn Greer
883
+ 25.1
884
+ ŚlΔ…sk WrocΕ‚aw
885
+ Regular season
886
+ 1
887
+ 2004-2005
888
+ Charles Smith
889
+ 20.7
890
+ Scavolini Pesaro
891
+ Quarterfinals
892
+ 3
893
+ 2005-2006
894
+ Drew Nicholas
895
+ 18.5
896
+ Benetton Treviso
897
+ Top 16
898
+ 2
899
+ 2006-2007
900
+ Igor RakočeviΔ‡
901
+ 16.2
902
+ Tau CerΓ‘mica
903
+ 4th place
904
+ 4
905
+ 2007-2008
906
+ Marc Salyers
907
+ 21.8
908
+ Roanne
909
+ Regular season
910
+ 1
911
+ 2008-2009
912
+ Igor RakočeviΔ‡
913
+ 18
914
+ Tau CerΓ‘mica
915
+ Quarterfinals
916
+ 3
917
+ 2009-2010
918
+ Linas Kleiza
919
+ 17.1
920
+ Olympiacos
921
+ 2nd place
922
+ 4.5
923
+ 2010-2011
924
+ Igor RakočeviΔ‡
925
+ 17.2
926
+ Efes Pilsen
927
+ Top 16
928
+ 2
929
+ 2011-2012
930
+ Bo McCalebb
931
+ 16.9
932
+ Montepaschi Siena
933
+ Quarterfinals
934
+ 3
935
+ 2012-2013
936
+ Bobby Brown
937
+ 18.8
938
+ Montepaschi Siena
939
+ Top 16
940
+ 2
941
+
942
+ 17
943
+
944
+ 2013-2014
945
+ Keith Langford
946
+ 17.6
947
+ EA7 Milano
948
+ Quarterfinals
949
+ 3
950
+ 2014-2015
951
+ Taylor Rochestie
952
+ 18.9
953
+ Nizhny Novgorod
954
+ Top 16
955
+ 2
956
+ 2015-2016
957
+ Nando de Colo
958
+ 18.9
959
+ CSKA Moscow
960
+ Winner
961
+ 5
962
+ 2016-2017
963
+ Keith Langford
964
+ 21.8
965
+ UNICS
966
+ Regular season
967
+ 1
968
+ 2017-2018
969
+ Alexey Shved
970
+ 21.8
971
+ Khimki
972
+ Quarterfinals
973
+ 3
974
+
975
+ 3.6 Unexpected Outcomes in the Final Fours
976
+ In this section it is examined the unexpected of the Final-Fours in terms of outcomes based on previous
977
+ attempts with the use of Binomial Distribution. Can we make the assumption that each final four is an
978
+ experiment with each team to have the same probabilities of winning the tournament (25%)?
979
+ To answer this question, we consider each final four as an experiment and teams are considered as
980
+ independent random variables. Each experiment can be described by the binomial distribution and the
981
+ whole situation with multinomial distribution (Forbes et al, 2011) which is a generalization of binomial
982
+ distribution and describes n trials. There is performed a multinomial goodness of fit test and to strengthen
983
+ the results a binomial test for each team, in order to decide if there is any significant difference from
984
+ binomial distribution.
985
+ Multinomial Testing
986
+ p-value
987
+
988
+ 0.54499Β±0.001575
989
+ Binomial Testing*
990
+
991
+
992
+ p-value
993
+ Cibona
994
+ 2 attempts - 2 trophies
995
+ 0.0625
996
+ Jugoplastica
997
+ 4 attempts - 3 trophies
998
+ 0.05078
999
+ ASK Riga
1000
+ 4 attempts - 3 trophies
1001
+ 0.05078
1002
+ Panathinaikos
1003
+ 12 attempts - 6 trophies
1004
+ 0.08608
1005
+ *Only cases with p-value<0.1
1006
+
1007
+ Table on the appendix displays the final four teams, the expected titles according to Binomial distribution,
1008
+ the observed values and their difference. Indeed there is no evidence that there are significant
1009
+ discrepancies from the binomial distribution at the 5% level of significance.
1010
+ In the modern period of Euroleague (1999-2018), again there is no evidence of significant discrepancy
1011
+ from the multinomial distribution, but the case of Panathinaikos could be seen as an exception, with
1012
+ significant larger success rate than the expected.
1013
+
1014
+
1015
+ 18
1016
+
1017
+ Multinomial Testing
1018
+ p-value
1019
+
1020
+ 0.68173Β±0.001473
1021
+ Binomial Testing*
1022
+
1023
+
1024
+ p-value
1025
+ Panathinaikos
1026
+ 8 attempts - 5 trophies
1027
+ 0.02730
1028
+ *Only cases with p-value<0.1
1029
+
1030
+
1031
+ 4 Summary and Conclusions
1032
+ To sum up, in this paper is made an attempt to address questions related to historical evolution of
1033
+ Euroleague using statistical analysis to draw conclusions. One main problem is the lack of plenty
1034
+ available statistical data from the beginning of the competition. This paper demonstrates that by applying
1035
+ suitable statistical designs we can draw interesting conclusions even with limited data. Firstly, is made a
1036
+ brief exploration of the historical evolution of the Euroleague and the tracking statistics.
1037
+ Then, some questions are answered and some conclusions are drawn which are briefly the following:
1038
+ Although overall there is no difference in success between more traditional powers such as Italy, Spain
1039
+ and ex USSR and Yugoslavian countries and other countries, there is a clear trend of other countries to
1040
+ expand their presence (in terms of titles and final four appearances) in the tournament after the 90s. In
1041
+ terms of scoring, there was an upward trend from the beginning of the competition, with a sudden drop in
1042
+ 90s, something which indicates a significant change in game mentality in terms of defence and/or game
1043
+ pace. The champions were more dominant in 60s and 70s, but in the 80s they scored only 8% more than
1044
+ their opponent, which indicates that the competition was more intense in this decade. The last years
1045
+ (2010-2018), the competitiveness of the tournament tends to similar levels.
1046
+ There is a popular belief that the first scorer in the majority of cases come from the runner-up team.
1047
+ However, there is no reason to assume that in a specific year is more probable the runner-up team to have
1048
+ a more dominant player in the scoring in the final. In the modern period of Euroleague (after 1998), we
1049
+ can assume that there is a more dominant scorer in the runner-up team, but we cannot predict this fact for
1050
+ a specific year. According to the game evolution in finals, we observe that the game has been transformed
1051
+ from finals with offences which are based on top scorers to finals with more pluralism. From the
1052
+ beginning until the 80s the trend was the one player star in scoring, but since then, there was a slow but
1053
+
1054
+ 19
1055
+
1056
+ continuing turn to games based on pluralism. Moreover, we detect a significant difference in possessions
1057
+ before and after 1998 finals, which is in accordance with our assumption that the game has changed after
1058
+ 1998 final and supports the assumption that the triumph of Zalgiris in 1999 was the start of the change.
1059
+ Furthermore, it is found a significant negative relationship between the first scorer and the performance of
1060
+ his team. This finding favors the opinion that the first scorers are rarely parts of top teams. Finally, there
1061
+ is no evidence to reject the hypothesis that in a final four there are equal chances of winning overall. In
1062
+ modern era, again the hypothesis of the final four as a random experiment is not rejected, however in the
1063
+ case of Panathinaikos we observe significantly higher success rate than the expected.
1064
+
1065
+
1066
+ References
1067
+ Best, D. J., & Roberts, D. E. 1975. β€œAlgorithm AS 89: the upper tail probabilities of Spearman's rho.” Journal of the
1068
+ Royal Statistical Society. Series C (Applied Statistics), 24(3), 377-379.
1069
+ Caeiro, F., & Mateus, A. 2014. randtests: Testing randomness in R. R package version, 1.
1070
+ Forbes, C., Evans, M., Hastings, N., & Peacock, B. 2011.Statistical distributions. John Wiley & Sons.
1071
+ Gibbons, J. D., & Chakraborti, S. 2003. Nonparametric Statistical Inference. Marcel Dekker. Inc. New York.
1072
+ Hollander, M., & Wolfe, D. A. 1999. Nonparametric statistical methods. 2nd Edition, John Wiley & Sons.
1073
+ Kubatko, J., Oliver, D., Pelton, K., & Rosenbaum, D. T. 2007. β€œA starting point for analyzing basketball statistics.”
1074
+ Journal of Quantitative Analysis in Sports, 3(3).
1075
+ Oliver, D. (2004). Basketball on paper: rules and tools for performance analysis. Potomac Books, Inc.
1076
+ Zivot, E., & Andrews, D. W. K. 2002. β€œFurther evidence on the great crash, the oil-price shock, and the unit-root
1077
+ hypothesis.” Journal of business & economic statistics, 20(1), 25-44.
1078
+
1079
+
1080
+
1081
+
1082
+
1083
+ 20
1084
+
1085
+ Appendix
1086
+ Table. Euroleague Final Four Teams
1087
+ Year
1088
+ Winner
1089
+ Runner-Up
1090
+ 3rd Place
1091
+ 4th Place
1092
+ 1958
1093
+ ASK Riga
1094
+ Academic
1095
+ Honved
1096
+ Real Madrid
1097
+ 1958- 1959
1098
+ ASK Riga
1099
+ Academic
1100
+ Lech Poznan
1101
+ OKK Beograd
1102
+ 1959-1960
1103
+ ASK Riga
1104
+ Dinamo Tbilisi
1105
+ Pologna Warzawa
1106
+ Slovan Orbis Praha
1107
+ 1960-1961
1108
+ CSKA Moscow
1109
+ ASK Riga
1110
+ Real Madrid
1111
+ Steaua Bucarest
1112
+ 1961-1962
1113
+ Dinamo Tbilisi
1114
+ Real Madrid
1115
+ ASK Olimpija
1116
+ CSKA Moscow
1117
+ 1962-1963
1118
+ CSKA Moscow
1119
+ Real Madrid
1120
+ Dinamo Tbilisi
1121
+ Spartak Brno
1122
+ 1963-1964
1123
+ Real Madrid
1124
+ Spartak Brno
1125
+ OKK Beograd
1126
+ Simmenthal Milano
1127
+ 1964-1965
1128
+ Real Madrid
1129
+ CSKA Moscow
1130
+ Ignis Varese
1131
+ OKK Beograd
1132
+ 1965-1966
1133
+ Simmenthal Milano
1134
+ Slavia Praha
1135
+ CSKA Moscow
1136
+ AEK
1137
+ 1966-1967
1138
+ Real Madrid
1139
+ Simmenthal Milano
1140
+ ASK Olimpija
1141
+ Slavia Praha
1142
+ 1967-1968
1143
+ Real Madrid
1144
+ Spartak Brno
1145
+ Simmenthal Milano
1146
+ Zadar
1147
+ 1968-1969
1148
+ CSKA Moscow
1149
+ Real Madrid
1150
+ Spartak Brno
1151
+ Standard Liege
1152
+ 1969-1970
1153
+ Ignis Varese
1154
+ CSKA Moscow
1155
+ Real Madrid
1156
+ Slavia Praha
1157
+ 1970-1971
1158
+ CSKA Moscow
1159
+ Ignis Varese
1160
+ Real Madrid
1161
+ Slavia Praha
1162
+ 1971-1972
1163
+ Ignis Varese
1164
+ Jugoplastica1
1165
+ Real Madrid
1166
+ Panathinaikos
1167
+ 1972-1973
1168
+ Ignis Varese
1169
+ CSKA Moscow
1170
+ Crvena Zvezda
1171
+ Simmenthal Milano
1172
+ 1973-1974
1173
+ Real Madrid
1174
+ Ignis Varese
1175
+ Berck
1176
+ Radniski Belgrade
1177
+ 1974-1975
1178
+ Ignis Varese
1179
+ Real Madrid
1180
+ Berck
1181
+ Zadar
1182
+ 1975-1976
1183
+ Mobilgirgi Varese
1184
+ Real Madrid
1185
+ ASVEL
1186
+ Forst CantΓΉ
1187
+ 1976-1977
1188
+ Maccabi Tel Aviv
1189
+ Mobilgirgi Varese
1190
+ CSKA Moscow
1191
+ Real Madrid
1192
+ 1977-1978
1193
+ Real Madrid
1194
+ Mobilgirgi Varese
1195
+ ASVEL
1196
+ Maccabi Tel Aviv
1197
+ 1978-1979
1198
+ Bosna
1199
+ Emerson Varese
1200
+ Maccabi Tel Aviv
1201
+ Real Madrid
1202
+ 1979-1980
1203
+ Real Madrid
1204
+ Maccabi Tel Aviv
1205
+ Bosna
1206
+ Sinudyne Bologna3
1207
+ 1980-1981
1208
+ Maccabi Tel Aviv
1209
+ Sinudyne Bologna3
1210
+ Nashua EBBC
1211
+ Bosna
1212
+ 1981-1982
1213
+ Squibb CantΓΉ
1214
+ Maccabi Tel Aviv
1215
+ Partizan
1216
+ FC Barcelona
1217
+ 1982-1983
1218
+ Ford CantΓΉ
1219
+ Billy Milano
1220
+ Real Madrid
1221
+ CSKA Moscow
1222
+ 1983-1984
1223
+ Virtus Roma
1224
+ FC Barcelona
1225
+ Jollycolombani CantΓΉ
1226
+ Bosna
1227
+ 1984-1985
1228
+ Cibona
1229
+ Real Madrid
1230
+ Maccabi Tel Aviv
1231
+ CSKA Moscow
1232
+ 1985-1986
1233
+ Cibona
1234
+ Zalgiris
1235
+ Simac Milano
1236
+ Real Madrid
1237
+ 1986-1987
1238
+ Tracer Milano
1239
+ Maccabi Tel Aviv
1240
+ Orthez
1241
+ Zadar
1242
+ 1987-1988
1243
+ Tracer Milano
1244
+ Maccabi Tel Aviv
1245
+ Partizan
1246
+ Aris
1247
+ 1988-1989
1248
+ Jugoplastica1
1249
+ Maccabi Tel Aviv
1250
+ Aris
1251
+ FC Barcelona
1252
+ 1989-1990
1253
+ Jugoplastica1
1254
+ FC Barcelona
1255
+ Limoges
1256
+ Aris
1257
+ 1990-1991
1258
+ POP 84 1
1259
+ FC Barcelona
1260
+ Maccabi Tel Aviv
1261
+ Scavolini Pezaro
1262
+ 1991-1992
1263
+ Partizan
1264
+ Joventut
1265
+ Phillips Milano
1266
+ Estudiantes
1267
+ 1992-1993
1268
+ Limoges
1269
+ Benneton Treviso
1270
+ PAOK
1271
+ Real Madrid
1272
+ 1993-1994
1273
+ Joventut
1274
+ Olympiacos
1275
+ Panathinaikos
1276
+ FC Barcelona
1277
+ 1994-1995
1278
+ Real Madrid
1279
+ Olympiacos
1280
+ Panathinaikos
1281
+ Limoges
1282
+ 1995-1996
1283
+ Panathinaikos
1284
+ FC Barcelona
1285
+ CSKA Moscow
1286
+ Real Madrid
1287
+ 1996-1997
1288
+ Olympiacos
1289
+ FC Barcelona
1290
+ Smelt Olimpija
1291
+ ASVEL
1292
+ 1997-1998
1293
+ Kinder Bologna3
1294
+ AEK
1295
+ Benneton Treviso
1296
+ Partizan
1297
+ 1998-1999
1298
+ Zalgiris
1299
+ Kinder Bologna3
1300
+ Olympiacos
1301
+ Teamsystem Bologna4
1302
+ 1999-2000
1303
+ Panathinaikos
1304
+ Maccabi Tel Aviv
1305
+ Efes Pilsen
1306
+ FC Barcelona
1307
+ 2000-2001 (FIBA)
1308
+ Kinder Bologna3
1309
+ TAU Ceramica2
1310
+ AEK
1311
+ Paf Wennington Bologna4
1312
+ 2000-2001 (Euroleague)
1313
+ Maccabi Tel Aviv
1314
+ Panathinaikos
1315
+ Efes Pilsen
1316
+ CSKA Moscow
1317
+ 2001-2002
1318
+ Panathinaikos
1319
+ Kinder Bologna3
1320
+ Benneton Treviso
1321
+ Maccabi Tel Aviv
1322
+ 2002-2003
1323
+ FC Barcelona
1324
+ Benneton Treviso
1325
+ Montepaschi Siena
1326
+ CSKA Moscow
1327
+ 2003-2004
1328
+ Maccabi Tel Aviv
1329
+ Skipper Bologna4
1330
+ CSKA Moscow
1331
+ Montepaschi Siena
1332
+ 2004-2005
1333
+ Maccabi Tel Aviv
1334
+ TAU Ceramica2
1335
+ Panathinaikos
1336
+ CSKA Moscow
1337
+ 2005-2006
1338
+ CSKA Moscow
1339
+ Maccabi Tel Aviv
1340
+ TAU Ceramica2
1341
+ FC Barcelona
1342
+ 2006-2007
1343
+ Panathinaikos
1344
+ CSKA Moscow
1345
+ Unicaja Malaga
1346
+ TAU Ceramica2
1347
+ 2007-2008
1348
+ CSKA Moscow
1349
+ Maccabi Tel Aviv
1350
+ Montepaschi Siena
1351
+ Real Madrid
1352
+ 2008-2009
1353
+ Panathinaikos
1354
+ CSKA Moscow
1355
+ FC Barcelona
1356
+ Olympiacos
1357
+ 2009-2010
1358
+ FC Barcelona
1359
+ Olympiacos
1360
+ CSKA Moscow
1361
+ Partizan
1362
+ 2010-2011
1363
+ Panathinaikos
1364
+ Maccabi Tel Aviv
1365
+ Montepaschi Siena
1366
+ Real Madrid
1367
+ 2011-2012
1368
+ Olympiacos
1369
+ CSKA Moscow
1370
+ FC Barcelona
1371
+ Panathinaikos
1372
+ 2012-2013
1373
+ Olympiacos
1374
+ Real Madrid
1375
+ CSKA Moscow
1376
+ FC Barcelona
1377
+ 2013-2014
1378
+ Maccabi Tel Aviv
1379
+ Real Madrid
1380
+ FC Barcelona
1381
+ CSKA Moscow
1382
+ 2014-2015
1383
+ Real Madrid
1384
+ Olympiacos
1385
+ CSKA Moscow
1386
+ Fenerbahce
1387
+ 2015-2016
1388
+ CSKA Moscow
1389
+ Fenerbahce
1390
+ Lokomotiv Kuban
1391
+ Laboral Kutxa2
1392
+ 2016-2017
1393
+ Fenerbahce
1394
+ Olympiacos
1395
+ CSKA Moscow
1396
+ Real Madrid
1397
+ 2017-2018
1398
+ Real Madrid
1399
+ Fenerbahce
1400
+ Zalgiris
1401
+ CSKA Moscow
1402
+ 1Croatia Split
1403
+ 2Club Deportivo Saski Baskonia, S.A.D.
1404
+ 3Virtus Pallacanestro Bologna
1405
+ 4 Fortitudo Pallacanestro Bologna 103
1406
+
1407
+ 21
1408
+
1409
+ Table . Teams and appearances to final four
1410
+ Team
1411
+ Winner
1412
+ Appearances
1413
+ Expected Titles
1414
+ Observed Titles
1415
+ Difference
1416
+ Lokomotiv Kuban
1417
+ 0
1418
+ 1
1419
+ 0.25
1420
+ 0
1421
+ -0.25
1422
+ Unicaja Malaga
1423
+ 0
1424
+ 1
1425
+ 0.25
1426
+ 0
1427
+ -0.25
1428
+ PAOK
1429
+ 0
1430
+ 1
1431
+ 0.25
1432
+ 0
1433
+ -0.25
1434
+ Estudiantes
1435
+ 0
1436
+ 1
1437
+ 0.25
1438
+ 0
1439
+ -0.25
1440
+ Scavolini Pezaro
1441
+ 0
1442
+ 1
1443
+ 0.25
1444
+ 0
1445
+ -0.25
1446
+ Orthez
1447
+ 0
1448
+ 1
1449
+ 0.25
1450
+ 0
1451
+ -0.25
1452
+ Nashua EBBC
1453
+ 0
1454
+ 1
1455
+ 0.25
1456
+ 0
1457
+ -0.25
1458
+ Radniski Belgrade
1459
+ 0
1460
+ 1
1461
+ 0.25
1462
+ 0
1463
+ -0.25
1464
+ Crvena Zvezda
1465
+ 0
1466
+ 1
1467
+ 0.25
1468
+ 0
1469
+ -0.25
1470
+ Standard Liege
1471
+ 0
1472
+ 1
1473
+ 0.25
1474
+ 0
1475
+ -0.25
1476
+ Steaua Bucarest
1477
+ 0
1478
+ 1
1479
+ 0.25
1480
+ 0
1481
+ -0.25
1482
+ Lech Poznan
1483
+ 0
1484
+ 1
1485
+ 0.25
1486
+ 0
1487
+ -0.25
1488
+ Honved
1489
+ 0
1490
+ 1
1491
+ 0.25
1492
+ 0
1493
+ -0.25
1494
+ Pologna Warzawa
1495
+ 0
1496
+ 1
1497
+ 0.25
1498
+ 0
1499
+ -0.25
1500
+ Efes Pilsen
1501
+ 0
1502
+ 2
1503
+ 0.5
1504
+ 0
1505
+ -0.5
1506
+ Berck
1507
+ 0
1508
+ 2
1509
+ 0.5
1510
+ 0
1511
+ -0.5
1512
+ Zadar
1513
+ 0
1514
+ 3
1515
+ 0.75
1516
+ 0
1517
+ -0.75
1518
+ Olimpija Ljubliana
1519
+ 0
1520
+ 3
1521
+ 0.75
1522
+ 0
1523
+ -0.75
1524
+ OKK Beograd
1525
+ 0
1526
+ 3
1527
+ 0.75
1528
+ 0
1529
+ -0.75
1530
+ ASVEL
1531
+ 0
1532
+ 3
1533
+ 0.75
1534
+ 0
1535
+ -0.75
1536
+ Aris
1537
+ 0
1538
+ 3
1539
+ 0.75
1540
+ 0
1541
+ -0.75
1542
+ Montepaschi Siena
1543
+ 0
1544
+ 4
1545
+ 1
1546
+ 0
1547
+ -1
1548
+ Fortitudo Bologna
1549
+ 0
1550
+ 3
1551
+ 0.75
1552
+ 0
1553
+ -0.75
1554
+ AEK
1555
+ 0
1556
+ 3
1557
+ 0.75
1558
+ 0
1559
+ -0.75
1560
+ Praha
1561
+ 0
1562
+ 5
1563
+ 1.25
1564
+ 0
1565
+ -1.25
1566
+ Baskonia
1567
+ 0
1568
+ 5
1569
+ 1.25
1570
+ 0
1571
+ -1.25
1572
+ Treviso
1573
+ 0
1574
+ 4
1575
+ 1
1576
+ 0
1577
+ -1
1578
+ Brno
1579
+ 0
1580
+ 4
1581
+ 1
1582
+ 0
1583
+ -1
1584
+ Academic
1585
+ 0
1586
+ 2
1587
+ 0.5
1588
+ 0
1589
+ -0.5
1590
+ Limoges
1591
+ 1
1592
+ 3
1593
+ 0.75
1594
+ 1
1595
+ 0.25
1596
+ Partizan
1597
+ 1
1598
+ 5
1599
+ 1.25
1600
+ 1
1601
+ -0.25
1602
+ Virtus Roma
1603
+ 1
1604
+ 1
1605
+ 0.25
1606
+ 1
1607
+ 0.75
1608
+ Bosna
1609
+ 1
1610
+ 4
1611
+ 1
1612
+ 1
1613
+ 0
1614
+ Zalgiris
1615
+ 1
1616
+ 3
1617
+ 0.75
1618
+ 1
1619
+ 0.25
1620
+ Joventut Badalona
1621
+ 1
1622
+ 2
1623
+ 0.5
1624
+ 1
1625
+ 0.5
1626
+ Dinamo Tbilisi
1627
+ 1
1628
+ 3
1629
+ 0.75
1630
+ 1
1631
+ 0.25
1632
+ Fenerbahce
1633
+ 1
1634
+ 4
1635
+ 1
1636
+ 1
1637
+ 0
1638
+ Cibona
1639
+ 2
1640
+ 2
1641
+ 0.5
1642
+ 2
1643
+ 1.5
1644
+ Cantu
1645
+ 2
1646
+ 4
1647
+ 1
1648
+ 2
1649
+ 1
1650
+ Virtus Bologna
1651
+ 2
1652
+ 6
1653
+ 1.5
1654
+ 2
1655
+ 0.5
1656
+ FC Barcelona
1657
+ 2
1658
+ 16
1659
+ 4
1660
+ 2
1661
+ -2
1662
+ Split
1663
+ 3
1664
+ 4
1665
+ 1
1666
+ 3
1667
+ 2
1668
+ ASK Riga (Latvia)
1669
+ 3
1670
+ 4
1671
+ 1
1672
+ 3
1673
+ 2
1674
+ Olympia Milano
1675
+ 3
1676
+ 10
1677
+ 2.5
1678
+ 3
1679
+ 0.5
1680
+ Olympiacos
1681
+ 3
1682
+ 10
1683
+ 2.5
1684
+ 3
1685
+ 0.5
1686
+ Varese
1687
+ 5
1688
+ 11
1689
+ 2.75
1690
+ 5
1691
+ 2.25
1692
+ Panathinaikos
1693
+ 6
1694
+ 12
1695
+ 3
1696
+ 6
1697
+ 3
1698
+ Maccabi Tel Aviv
1699
+ 6
1700
+ 20
1701
+ 5
1702
+ 6
1703
+ 1
1704
+ CSKA Moscow
1705
+ 7
1706
+ 29
1707
+ 7.25
1708
+ 7
1709
+ -0.25
1710
+ Real Madrid
1711
+ 10
1712
+ 32
1713
+ 8
1714
+ 10
1715
+ 2
1716
+
1717
+
1718
+
1719
+
1720
+
JtE0T4oBgHgl3EQfiQGP/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
K9E4T4oBgHgl3EQfJgxB/content/tmp_files/2301.04921v1.pdf.txt ADDED
The diff for this file is too large to render. See raw diff
 
K9E4T4oBgHgl3EQfJgxB/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
KdE0T4oBgHgl3EQfigFl/content/tmp_files/2301.02446v1.pdf.txt ADDED
The diff for this file is too large to render. See raw diff
 
KdE0T4oBgHgl3EQfigFl/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
LNE2T4oBgHgl3EQfpwjv/content/tmp_files/2301.04033v1.pdf.txt ADDED
@@ -0,0 +1,1202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 1
2
+
3
+ Numerical investigation of progressive damage and associated seismicity on a
4
+ laboratory fault
5
+ Qi Zhao1,2*, Nicola Tisato3, Aly Abdelaziz2, Johnson Ha2, and Giovanni Grasselli2
6
+ 1Department of Civil and Environmental Engineering, The Hong Kong Polytechnic University,
7
+ Hung Hom, Hong Kong SAR, China.
8
+ 2Department of Civil and Mineral Engineering, The University of Toronto, 35 St. George Street,
9
+ Toronto, Ontario M5S 1A4, Canada.
10
+ 3Department of Geological Sciences, Jackson School of Geosciences, The University of Texas at
11
+ Austin, 2305 Speedway Stop C1160, Austin, TX 78712-1692, USA.
12
+ *Corresponding author: Qi Zhao (qi.qz.zhao@polyu.edu.hk)
13
+
14
+ Abstract
15
+ Understanding rock shear failure behavior is crucial to gain insights into slip-related geohazards
16
+ such as rock avalanches, landslides, and earthquakes. However, descriptions of the progressive
17
+ damage on the shear surface are still incomplete or ambiguous. In this study, we use the hybrid
18
+ finite-discrete element method (FDEM) to simulate a shear experiment and obtain a detailed
19
+ comprehension of shear induced progressive damage and the associated seismic activity. We built
20
+ a laboratory fault model from high resolution surface scans and micro-CT imaging. Our results
21
+ show that under quasi-static shear loading, the fault surface experiences local dynamic seismic
22
+ activities. We found that the seismic activity is related to the stress concentration on interlocking
23
+ asperities. This interlocking behavior (i) causes stress concentration at the region of contact that
24
+ could reach the compressive strength, and (ii) produces tensile stress up to the tensile strength in
25
+ the region adjacent to the contact area. Thus, different failure mechanisms and damage patterns
26
+ including crushing and sub-vertical fracturing are observed on the rough surface. Asperity failure
27
+ creates rapid local slips resulting in significant stress perturbations that alter the overall stress
28
+ condition and may trigger the slip of adjacent critically stressed asperities. We found that the
29
+
30
+ 2
31
+
32
+ spatial distribution of the damaged asperities and the seismic activity is highly heterogeneous;
33
+ regions with intense asperity interactions formed gouge material, while others exhibit minimal to
34
+ no damage. These results emphasize the important role of surface roughness in controlling the
35
+ overall shear behavior and the local dynamic seismic activities on faults.
36
+ Keywords
37
+ Shear behavior; surface roughness; asperity; shear induced damage; seismicity
38
+
39
+
40
+ 3
41
+
42
+ 1 Introduction
43
+ Understanding shear behavior along rock discontinuities at various scales, such as joints
44
+ and faults, is essential to rock engineering projects and geohazard mitigation. Rock discontinuities
45
+ are planes of weakness and are responsible for many geohazards, for example, rock avalanches,
46
+ landslides, and earthquakes. Numerous laboratory shear experiments have been conducted on a
47
+ large variety of rock types under different conditions (e.g., Bandis et al., 1983; Beeler, 1996;
48
+ Marone, 1998; Di Toro et al., 2004; Grasselli, 2001; Reches and Lockner, 2010; Tisato et al., 2012;
49
+ Kim and Jeon, 2019; Zhao et al., 2020; Morad et al., 2022). Among these studies, many suggested
50
+ the importance of surface roughness and contact condition in controlling the shear behavior.
51
+ However, the progressive damaging process on faults is still not well understood because fault
52
+ surfaces cannot be observed directly during shear, with the exception of a few studies utilizing
53
+ transparent halite samples (e.g., Renard et al., 2012) and in situ and operando testes conducted
54
+ under X-ray micro-computed tomography (micro-CT) (e.g., Zhao et al, 2018; Zhao et al., 2020).
55
+ Shear processes control coseismic damage and friction on fault, but the constitutive friction
56
+ theories are not yet fully understood.
57
+ To observe and gain insights into damage processes on rough rock surfaces undergoing
58
+ shear deformation, Tatone and Grasselli (2013) used micro-CT to image the joint surfaces after
59
+ the shear test; and Crandall et al. (2017) used micro-CT to obtain geometrical information from
60
+ fractured shale core that is incrementally sheared. Recently, direct and detailed observations of the
61
+ evolution of laboratory fault were achieved by using an in situ rotary shear experimental apparatus
62
+ under X-ray micro-CT (Zhao et al., 2017). Such experimental results help draw the connections
63
+ between microscopic damage and macroscopic shear behavior, the formation and accumulation of
64
+ gouge material, and shear-induced secondary fractures (Zhao, 2017; Zhao et al., 2018; Zhao et al.,
65
+
66
+ 4
67
+
68
+ 2020). However, due to technological limitations, time-continuous observations of the shear
69
+ surface evolution and the in situ stress condition remain shortfalls.
70
+ Numerical simulation methods have been extensively used to study the shear behavior of
71
+ rock discontinuities. To simulate the interaction and breakage of asperities and the frictional
72
+ sliding behavior, numerical methods that can capture solid fracturing and interaction are typically
73
+ used. For example, the particle-based lattice solid model (Mora and Place 1993) uses a numerical
74
+ concept similar to the discrete element method (DEM) to simulate frictional behavior and
75
+ fracturing in solids. Mora and Place (1998) and Place and Mora (2000) used their model to study
76
+ the role fault gouge on the frictional behavior of faults, offering a possible explanation for the heat
77
+ flow paradox (Henyey and Wasserburg, 1971; Lachenbruch and Sass, 1992). Bonded particle-
78
+ based methods, such as the particle flow code (PFC) (Cundall and Strack, 1979), are commonly
79
+ used for simulating rock shear behavior. Park and Song (2009) used PFC3D to simulate direct
80
+ shear tests and demonstrated that the method can simulate typical rock joint shear behavior, and
81
+ they found that the peak shear strength and peak dilation angle was strongly influenced by the
82
+ friction coefficient, roughness, and bond strength, while the residual shear strength and residual
83
+ friction angle was influenced by the particle size, friction coefficient, and bond strength. Asadi et
84
+ al. (2012) used a similar approach in two dimensions (PFC2D) to simulate direct shear experiments
85
+ on synthetic joint profiles of varying roughness and boundary conditions to assess asperity
86
+ degradation and intact material damage. They showed that as the joint is sheared, highly localized
87
+ asperity interaction on the joint surface and the geometry of the asperities has a significant
88
+ influence on how joints fail. Types of failure include asperity sliding, cut-off, separation, and
89
+ crushing, typically associated with tensile failure into the intact material in conditions with steep
90
+ asperities and/or high normal stress. However, Bahaaddini et al. (2013) identified a significant
91
+
92
+ 5
93
+
94
+ shortcoming of the particle-based methods due to the unrealistic shear and dilation behavior of
95
+ joints as a result of particle interlocking due to the inherent micro-scale roughness of the joint. To
96
+ overcome this limitation, they implemented the smooth-joint model (Pierce et al., 2007), where
97
+ the blocks associated with either side of the joint are generated separately to appropriately define
98
+ and apply the smooth joint model to the interface. Lambert and Coll (2014) created a synthetic
99
+ rock joint by importing the real morphology of the joint surface into a bonded particle assembly
100
+ and studied the shear behavior using the smooth-joint model. Their results reproduced the
101
+ progressive degradation of the asperities upon shearing.
102
+ The hybrid finite-discrete element method (FDEM) is increasingly used to investigate rock
103
+ shear behavior. FDEM is a micromechanical numerical method first introduced by Munjiza et al.,
104
+ (1995) combining the finite element method and discrete element method. In doing so, the
105
+ numerical method can model the transition of a continuous material to a discontinuous material as
106
+ it deforms, yields, and breaks. Karami and Stead (2008) and Tatone (2014) used FDEM to model
107
+ direct shear tests and relate progressive asperity degradation mechanisms with the measured shear
108
+ stress and dilation during shearing. In addition, Tatone (2014) verified the numerical modelling
109
+ observations by coupling their study with X-ray micro-CT imaging on post-mortem specimens. It
110
+ was found that tensile fractures develop in asperities at and near the peak shear stress, followed by
111
+ a reduction in shear resistance as asperities continue to fail in both tension and shear, and finally,
112
+ a residual shear resistance is reached once asperities are completely broken and gouge is formed.
113
+ In this study, we used the two-dimensional (2D) FDEM to simulate an experiment on the
114
+ gradual evolution of deformation of a laboratory fault, and we improve the understanding of shear
115
+ behavior of rough faults through combined interpretation of the simulation and experimental
116
+ results. First, we provide a brief review of the FDEM, emphasizing the modeling of damage and
117
+
118
+ 6
119
+
120
+ seismic activity. Second, we develop a clustering algorithm to improve the comprehension of the
121
+ simulated fractures and seismic events. Next, we build a model based on the laboratory experiment
122
+ and analyze the simulated results focusing on three aspects that are hardly accessible by
123
+ experiments: (1) the time-continuous variation of stress conditions on the shear surface, (2) the
124
+ progressive failure of the asperities and accumulation of gouge, and (3) the seismic activity related
125
+ to shear-induced damage.
126
+ The carefully built and calibrated numerical model is able to simulate the emergent rock
127
+ mechanical and frictional behaviors. We observe that shear-induced damage and seismic activities
128
+ are heterogeneously distributed along the fault surface due to the surface roughness. Seismic
129
+ events occur at the locations of asperity failure due to the interlocking-induced stress concentration.
130
+ Such events radiate seismic waves and significantly change the overall stress conditions. Some
131
+ areas on the fault were covered by gouge material and free from damage. These results agree with
132
+ the laboratory observations and further elaborate on the importance of surface roughness in
133
+ controlling shear behavior, which is critical to rock engineering practices and earthquake studies.
134
+ 2 Material and methods
135
+ 2.1 In situ shear test under X-ray micro-CT
136
+ The numerical simulation in this study is based on the experimental work using in situ
137
+ shear tests under micro-CT reported by Zhao et al. (2018) and Zhao et al. (2020), and a brief review
138
+ is provided here for completeness. The tested specimen was a cylindrical Flowstone (microfine
139
+ calcium sulfate cement mortar) 32 mm in length and 12 mm in diameter. The specimen was divided
140
+ into top and bottom parts by a three-point bending test that created two semi-samples divided by
141
+ a discontinuity (i.e., laboratory fault) with two matching rough surfaces. An unconfined rotary
142
+ shear test was conducted on the two semi-samples by shearing the fault under the initial normal
143
+
144
+ 7
145
+
146
+ stress of 2.5 MPa. The top semi-sample was forced to slip incrementally against the fixed bottom
147
+ semi-sample. Normal force and torque were recorded during rotation and used to calculate the
148
+ friction coefficient. After each incremental slip of 6Β°, a three-dimensional (3D) micro-CT scan was
149
+ conducted, which allows for imaging of the gradual morphological evolution of the specimen (Fig.
150
+ 1a). This experimental work provided detailed information of the shear-induced secondary
151
+ fractures (Fig. 1b) and the progressive damage on the slipping surface (Fig. 1c) in the sample
152
+ volume; however, the observation of the shear surface damage evolution was only available at
153
+ discrete time points coincident with each shear step, while an actual time-continuous observation
154
+ of the shear surface evolution and the local stress condition on the rough surface was not available.
155
+
156
+ Fig. 1 Summary of the laboratory set-up and results. (a) Schematic of the in situ shear test and a
157
+ zoom-in view of the shear surface (i.e., the zone of interest). (b) 3D visualization of the
158
+ development of shear induced fractures with increasing shear displacement. (c) 2D unwrapped
159
+ micro-CT image slice showing the progressive damage on the slipping surface with increasing
160
+ shear displacement, viewed at the radius (R = 5.6 mm) corresponding to the highest asperity. Red
161
+
162
+ 8
163
+
164
+ dashed boxes in (c) indicate (from left to right) shear-induced aperture opening, fracturing, and
165
+ progressive damage and gouge formation (modified from Zhao et al. (2018) and Zhao et al. (2020)).
166
+ 2.2 The hybrid finite-discrete element method
167
+ The hybrid finite-discrete element method (FDEM) combines continuum mechanics
168
+ principles with discrete element principles to simulate interaction, deformation, and fracturing of
169
+ materials (Munjiza et al., 1995; Munjiza, 2004). FDEM has been used to investigate a wide range
170
+ of rock mechanics and geophysics problems including, but not limited to, tunneling and excavation,
171
+ micromechanics, rock joint shear behavior, hydraulic fracturing, thermal-mechanical/hydro-
172
+ thermal-mechanical coupling, and fault dynamics (e.g., Karami and Stead 2008; Mahabadi et al.,
173
+ 2012; Lisjak et al., 2014; Zhao et al., 2014; Yan et al., 2016; Huang et al., 2017; Lei et al., 2017;
174
+ Ma et al., 2017; Fukuda et al., 2019; Okubo et al., 2019; Knight et al., 2020). Simulating the entire
175
+ shear behavior and evolution of a rough surface is a challenging task that requires advanced
176
+ computational resources encompassing, for example, the 3D FDEM method. However, 3D models
177
+ explicitly capturing the surface roughness at sub-millimeter resolution and the entire shear process
178
+ is not practical due to the demanded computation power. On the other hand, 2D FDEM simulations
179
+ has the merit of reducing the computational demand, and it has been shown to provide insights
180
+ into the mechanical behavior of rock joints and faults (e.g., Karami and Stead, 2008; Tatone, 2014;
181
+ Okubo et al., 2019).
182
+ FDEM models synthesize the macroscopic behavior of materials from the interaction of
183
+ the micromechanical constituents. In a 2D FDEM model, the simulated material is first discretized
184
+ based on a finite element mesh consisting of nodes and triangular elements. Then, the finite
185
+ element mesh is enriched by inserting a four-node cohesive crack element (CCE) between each
186
+ adjacent triangular element pair. Motion for the discretized system is calculated by an explicit time
187
+
188
+ 9
189
+
190
+ integration scheme, and the nodal coordinates of the elements are updated at each simulation step
191
+ (Munjiza, 2004). FDEM models the progressive damage and failure of brittle material according
192
+ to the principles of non-linear elastic fracture mechanics (Dugdale, 1960; Barenblatt, 1962), and it
193
+ captures the fracturing behavior of solids by modeling the entire failure path, including elastic
194
+ deformation, yielding, and fracturing (Fig. 2).
195
+
196
+ Fig. 2 Schematic diagram showing the FDEM approach of simulating fracturing. (a) Propagation
197
+ of a fracture and the creation of fracture process zone (FPZ). (b) Realization of the fracturing
198
+ process in FDEM involves the yielded cohesive crack elements and broken cohesive crack
199
+ elements (BCCE).
200
+ Depending on the local stress and deformation field, the CCE undergoes elastic
201
+ deformation, yielding, and breakage, simulating the damage development of the fracture process
202
+ zone (FPZ) (Fig. 3) (Labuz et al., 1985). During elastic loading, the relationships between bonding
203
+ stresses (normal bonding stress, Οƒ and shear bonding stress, Ο„) and the corresponding crack
204
+ displacement (opening, o and slip, s) are as follows (Munjiza et al., 1999):
205
+
206
+ Cracktip10
207
+
208
+ 𝜎 = {
209
+ 2π‘œ
210
+ π‘œπ‘ 𝑓𝑑 (π‘œ < 0, compression)
211
+ [
212
+ 2π‘œ
213
+ π‘œπ‘ βˆ’ (
214
+ π‘œ
215
+ π‘œπ‘)
216
+ 2
217
+ ] 𝑓𝑑 (0 < π‘œ < π‘œπ‘, tension)
218
+
219
+
220
+
221
+ (1)
222
+ 𝜏 = [
223
+ 2𝑠
224
+ 𝑠𝑝 βˆ’ (
225
+ 𝑠
226
+ 𝑠𝑝)
227
+ 2
228
+ ] 𝑓𝑠 (|𝑠| ≀ |𝑠𝑝|, shear)
229
+
230
+
231
+
232
+ (2)
233
+ where ft and fs are the peak tensile and shear bonding strength of a CCE, respectively. The peak
234
+ shear bonding strength is calculated based on the Mohr-Coulomb failure criterion using the
235
+ cohesion (c) and internal friction angle (Ο•): 𝑓𝑠 = 𝑐 + 𝜎 tan πœ™. op and sp are the peak opening and
236
+ slip values at the peak bonding stresses calculated as π‘œπ‘ = 2β„Žπ‘“π‘‘ 𝑝𝑓
237
+ ⁄
238
+ and 𝑠𝑝 = 2β„Žπ‘“π‘  𝑝𝑓
239
+ ⁄
240
+ , where h is
241
+ the nominal element edge length, and pf is the fracture penalty value. A CCE yields once the stress
242
+ reaches the peak, then it experiences a post-peak softening behavior with the bonding stresses
243
+ gradually decreased (Munjiza et al., 1999):
244
+ 𝜎 = 𝐹(𝐷)𝑓𝑑
245
+
246
+
247
+
248
+
249
+
250
+ (3)
251
+ 𝜏 = 𝐹(𝐷)𝑓𝑠
252
+
253
+
254
+
255
+
256
+ (4)
257
+ F(D) is an empirical function that approximates the shape of the experimental stress-displacement
258
+ failure curve according to Evans and Marathe (1968):
259
+ F(𝐷) = [1 βˆ’
260
+ π‘Ž+π‘βˆ’1
261
+ π‘Ž+𝑏 exp (𝐷
262
+ π‘Ž+𝑐𝑏
263
+ (π‘Ž+𝑏)(1βˆ’π‘Žβˆ’π‘))] βˆ™ [π‘Ž(1 βˆ’ 𝐷) + 𝑏(1 βˆ’ 𝐷)𝑐]
264
+
265
+ (5)
266
+ where a, b, c are empirical curve fitting parameters equal to 0.63, 1.8, and 6.0, respectively. The
267
+ damage coefficient (D) is calculated for Mode I, II, and I-II as
268
+ 𝐷I =
269
+ π‘œβˆ’π‘œπ‘
270
+ π‘œπ‘Ÿβˆ’π‘œπ‘
271
+
272
+
273
+
274
+
275
+
276
+ (6)
277
+ 𝐷II =
278
+ π‘ βˆ’π‘ π‘
279
+ π‘ π‘Ÿβˆ’π‘ π‘
280
+
281
+
282
+
283
+
284
+
285
+ (7)
286
+
287
+ 11
288
+
289
+ 𝐷Iβˆ’II = √𝐷I
290
+ 2 + 𝐷II
291
+ 2
292
+
293
+
294
+
295
+
296
+ (8)
297
+ with the subscripts indicating the mode of failure. The CCE breaks when D = 1, which corresponds
298
+ to a residual opening (or) or a residual slip (sr), for pure Mode I or II failure, respectively. For
299
+ Mode I-II failure, DI-II = 1 corresponds to a mixed failure opening and slip (of and sf). The values
300
+ of or and sr are calculated using the predefined numerical fracture energy GfI and GfII, for opening
301
+ failure and shear failure, respectively. The failure mode of the CCE (ΞΊ) is computed as
302
+ πœ… = {
303
+ 1 (pure tensile, Mode I)
304
+ 1 + 𝐷II (mixed mode, Mode I βˆ’ II)
305
+ 2 (pure shear, Mode II)
306
+
307
+
308
+
309
+ (9)
310
+
311
+
312
+ a)
313
+ b
314
+ Mode I
315
+ Mode II
316
+ Gf
317
+ 0
318
+ Sp
319
+ Opening
320
+ Slip
321
+ O, t
322
+ Normal/tangential bonding stress
323
+ 0, s
324
+ Opening/slip
325
+ c)
326
+ f,fs
327
+ Tensile/shear strength
328
+ Internal friction angle
329
+ Pf
330
+ Fracture penalty
331
+ h
332
+ Nominal elementedgelength
333
+ Mode I-HI
334
+ G/G Energy consumed by Mode I/II fracture
335
+ fΓ—f(D)
336
+ 0
337
+ Failure path
338
+ Broken
339
+ Mode I-II
340
+ 012
341
+
342
+ Fig. 3 Deformation and failure criteria of the cohesive crack element (CCE). (a) Mode I, tensile
343
+ mode, (b) Mode II, shear mode, and (c) Mode I-II, mixed-mode. Shaded areas highlight the total
344
+ fracture energy consumed during the failure process of a CCE. The blue curve (failure path)
345
+ indicates the stress condition during the yielding and failure processes of the CCE.
346
+ When both DI and DII are satisfied at the same time, the failure is also considered as Mode
347
+ I-II, and a value of 1.5 is assigned to these events during post-processing. The broken cohesive
348
+ crack element (BCCE) is then considered as a new crack with no cohesion, and its behavior is
349
+ handled by the interaction algorithms, which are discussed in detail in the literature (Munjiza 2004;
350
+ Mahabadi et al., 2012).
351
+ 2.3 Simulation of fracture propagation and seismicity in FDEM
352
+ Modeling seismic activity in rocks can provide quantitative information of the rock failure
353
+ process, and a validated model can improve the understanding of laboratory and field seismic
354
+ observations. In FDEM, upon breakage of the CCE, the accumulated strain energy is released,
355
+ resembling seismic activity. The coordinates, failure time, kinetic energy at failure, and failure
356
+ mode of the related BCCE can be recorded (Lisjak et al., 2013). However, a limitation of this
357
+ approach is that it considers each BCCE as one single seismic event. Consequently, the properties
358
+ of the fracture and the associated seismic events are highly dependent on the mesh size and mesh
359
+ orientation (Munjiza and John, 2002). In nature, the breakage of CCEs can be regarded as acoustic
360
+ emissions associated with the breakage of several mineral grains and grain boundaries (Zhao et al.,
361
+ 2015; Abdelaziz et al., 2018). In most cases, such a mesh dependency needs to be addressed to
362
+ obtain a better physical meaning of the failure process of CCEs. Zhao et al. (2014) attempted to
363
+ mitigate the problem with a clustering algorithm considering the temporal and spatial distribution
364
+ of BCCEs. In their method, each BCCE is viewed as an advancing crack tip, and BCCEs
365
+
366
+ 13
367
+
368
+ connecting to the crack tip are clustered together as a continuous fracture. However, this
369
+ implementation did not consider the physical meaning of fracture propagation. The propagating
370
+ fracture can arrest and then continue to propagate according to the stress conditions and material
371
+ heterogeneities (Van der Pluijm and Marshak, 2004), and from an energy dissipation point of view,
372
+ choosing the yielding point of a BCCE as the fracture tip is more consistent with the cohesive
373
+ crack model (Shet and Chandra, 2002).
374
+ Stemming from Zhao et al. (2014) and Zhao (2017), we implemented a new clustering
375
+ algorithm to mimic fracture propagation process during a seismic event. Note that we consider
376
+ only seismic activities related to the formation of new fractures, and seismic events created by
377
+ slipping on existing fracture surfaces are not considered. The algorithm proceeds as follows:
378
+ (1) The first BCCE that yields at time ty and fails at time tf is considered the initial crack of a
379
+ cluster. The search algorithm is then executed to include BCCEs connecting to either side of this
380
+ BCCE (i.e., fracture tips).
381
+ (2) BCCEs that are connected to the fracture tips and yield within the time window between ty and
382
+ tf are included in the same cluster and then treated as new fracture tips. At each output frame, the
383
+ same searching criterion is applied to such new fracture tips until no new BCCEs are found. Then,
384
+ this cluster of BCCEs is considered to be one continuous fracture, whose growth has produced one
385
+ seismic event.
386
+ (3) Repeat steps 1-2, until all recorded BCCEs are processed.
387
+ (4) Calculate the source parameters of the clustered seismic events as follows (for a cluster of n
388
+ BCCEs):
389
+ (a) event time is the breakage time tf of the initial BCCE in this cluster;
390
+
391
+ 14
392
+
393
+ (b) the hypocentre location is the centre coordinates of the initial BCCE in this cluster;
394
+ (c) the kinetic energy, Ee, is calculated as the sum of the kinetic energy of all BCCEs in
395
+ this cluster, 𝐸e = βˆ‘
396
+ 𝐸k
397
+ 𝑖
398
+ 𝑛
399
+ 𝑖=1
400
+ , where 𝐸k
401
+ 𝑖 =
402
+ 1
403
+ 2 βˆ‘
404
+ π‘šπ‘—π‘£π‘—
405
+ 2
406
+ 4
407
+ 𝑗=1
408
+ is the kinetic energy of a BCCE, and
409
+ mj and vj are the nodal mass and velocity of the BCCE at the time of breakage. We adopt
410
+ the empirical relation between radiated energy and magnitude to calculate the magnitude
411
+ of the seismic events: 𝑀𝑒 =
412
+ 2
413
+ 3 (log𝐸𝑒 βˆ’ 4.8) (Gutenberg, 1956; Lisjak et al., 2013).
414
+ (d) the dominant source mechanism (ΞΆ) of each cluster is calculated as a weighted average
415
+ of the failure modes of all BCCEs in this cluster:
416
+ 𝜁 =
417
+ βˆ‘
418
+ 𝐸k
419
+ 𝑖
420
+ 𝑛
421
+ 𝑖=1
422
+ πœ…π‘–
423
+ βˆ‘
424
+ 𝐸k
425
+ 𝑖
426
+ 𝑛
427
+ 𝑖=1
428
+
429
+
430
+
431
+
432
+
433
+
434
+ (10)
435
+ Where ΞΊi is the failure mode of the ith BCCE, and its associated kinetic energy, Eki is taken
436
+ as its weight. ΞΆ = 1 and 2 represent pure tensile (Mode I) and shear events (Mode II),
437
+ respectively, while events having 1 < ΞΆ < 2 have tensile and shear failure components
438
+ (Mode I-II).
439
+ This algorithm considers multiple BCCEs created by a single fracturing event, resulting in a more
440
+ realistic representation of the source mechanism and event energy than previous studies. Note that
441
+ if a series of connected CCEs break simultaneously due to mechanisms such as crushing or
442
+ pulverization, they will also be clustered as one event under this algorithm.
443
+ 2.4 Numerical model setup
444
+
445
+ 15
446
+
447
+
448
+ Fig. 4 Preparation of 2D surface profiles for the FDEM model. (a) The top (left) and bottom (right)
449
+ parts of the sample used in the rotary shear experiment. (b)–(c) 3D surface scan of the shear
450
+ surfaces. Red dashed lines indicate the extracted profiles. (d) The initial condition by micro-CT
451
+ imaging. (e) Comparison of the profiles (red dashed curves) with the micro-CT image showing the
452
+ initial condition of the shear simulation. Note that profiles are vertically offset for clearer
453
+ illustration.
454
+ 3D shear simulations would mimic at best the deformation processes, but this is currently
455
+ impossible due to computational limitations. Instead, we built the 2D FDEM model that considers
456
+ not only the geometry of the experimental specimen but also the initial contact condition on the
457
+ rough surface. A 2D circular profile at the radius of 5.6 mm, which corresponds to the roughest
458
+ region (i.e., highest asperities) on the surface, was extracted (Fig. 4a-c). We chose such a profile
459
+ because the work by Zhao et al. (2018) suggested that this region with the largest roughness plays
460
+ an important role in controlling the shear strength and fracture development during the experiment.
461
+ To capture the geometry of the slipping surface, we digitized the top and bottom surfaces before
462
+ the experiment using a 3D surface scanner (ATOS II by GOM) at a horizontal grid interval of
463
+ 44 ΞΌm. The relative location of the two profiles were adjusted to recreate the initial contact
464
+ conditions according to the micro-CT image (Fig. 4d-e). We subsampled the profiles to a 0.1 mm
465
+
466
+ a)
467
+ b)
468
+ C)
469
+ 1 mm
470
+ 0.30.20.10.0-0.1-0.2-0.3
471
+ d)
472
+ Elevation(mm)
473
+ e)
474
+ Sheardirection
475
+ 0
476
+ 5
477
+ 10
478
+ 15
479
+ 20
480
+ 25
481
+ 30
482
+ (mm)16
483
+
484
+ nominal grid interval, which was chosen as an acceptable compromise between computation time
485
+ and accuracy in representing the surface geometry. In addition, to mimic the rotary shear behavior,
486
+ the two ends of the profiles were extended by 3 mm (i.e., the desired total shear displacement)
487
+ using the same geometry as their opposite ends to create an effective periodic boundary. These
488
+ profiles formed the initial shear surfaces of the numerical model (Fig. 4e).
489
+
490
+ Fig. 5 (a) Mesh topology and boundary conditions of the shear test simulation. The blue dotted
491
+ line indicates the location of the virtual measurement line. (b) Zoom in view of the refined mesh
492
+ at the shear surfaces, and the arrows indicate the smallest gap between top and bottom surfaces.
493
+ The bodies of the top and bottom model were 15 mm in thickness, resulting in a total
494
+ vertical height of 30 mm, similar to the sample used in the laboratory experiment (Fig. 5a). The
495
+ corners at the ends of the shear surfaces were filleted with a radius of 0.2 mm to avoid stress
496
+ concentrations that may result in unrealistic damage. To reduce computational time in applying
497
+ the normal stress during the simulation, the initial vertical distance between the top and bottom
498
+ semi-sample was adjusted to 2Γ—10βˆ’6 mm (Fig. 5b). Moreover, two rigid boxes were added to
499
+ simulate the sample holders encasing the two semi-samples. The region of interest (i.e., within
500
+ 1 mm distance from the shear surface) was discretized with a constant nominal element size of
501
+ 0.1 mm. The remaining parts of the model were meshed with linearly increasing mesh size as a
502
+
503
+ b)
504
+ 5 mm
505
+ V
506
+ 0.2 mm
507
+ x
508
+ V17
509
+
510
+ function of the distance from the shear surface, with the coarsest element size being 3 mm. As a
511
+ result, the model was meshed into 20,240 triangular elements. These elements were assigned with
512
+ the calibrated numerical properties (Table 1&2), while the shear boxes had properties of stainless
513
+ steel (Young’s modulus at 200 GPa, density at 8100 kg/m3, and Poisson’s ratio at 0.25).
514
+ Table 1 Laboratory measured macromechanical properties (i.e., calibration targets) and emergent
515
+ properties of the calibrated FDEM model (after Tatone and Grasselli, 2015; Zhao, 2017).
516
+ Properties (unit)
517
+ Laboratory
518
+ measurement
519
+ Calibrated
520
+ FDEM model
521
+ Density (kgΒ·mβˆ’3)
522
+ 1704
523
+ 1704
524
+ Young’s modulus (GPa)
525
+ 15.0
526
+ 15.0
527
+ Poisson’s ratio (-)
528
+ 0.24
529
+ 0.24
530
+ Internal friction angle (Degrees)
531
+ 23
532
+ 23
533
+ Internal cohesion (MPa)
534
+ 16.4
535
+ 16.4
536
+ Tensile strength (MPa)
537
+ 2.6
538
+ 2.7
539
+ Uniaxial compressive strength (MPa)
540
+ 50.3
541
+ 49.9
542
+
543
+ FDEM models synthesize the macroscopic behavior of materials from the interaction of
544
+ the micromechanical constituents. The overall deformation and failure behavior of the simulated
545
+ material are controlled by the combined effect of the input parameters defining the elastic
546
+ triangular elements and CCEs. As a result, the macroscopic mechanical properties (as listed in
547
+ Table 1, except for the density that needs no calibration) measured by standard laboratory tests
548
+ cannot be used directly. Rather, an iterative calibration approach is carried out to obtain input
549
+ parameters representative of the material, and the laboratory measured properties were used as the
550
+ calibration targets. In this approach, numerical compressive and tensile strength test models are
551
+ created and simulated using an initial set of input parameters. The macroscopic mechanical
552
+ properties and failure patterns are obtained from the simuation and compared against laboratory
553
+
554
+ 18
555
+
556
+ measurements. In a successful calibration, the numerical model will replicate both the macroscopic
557
+ mechanical properties measured from the experiments and the overall failure mode of the material.
558
+ If the simulation result is inadequate, the input parameters are iteratively fine-tuned until the
559
+ calibration targets are met (Tatone and Grasselli, 2015). The laboratory-measured properties and
560
+ the emergent macromechanical properties of the calibrated FDEM model are listed in Table 1, and
561
+ the calibrated FDEM model parameters are listed in Table 2.
562
+ Table 2 Calibrated FDEM model input parameters (after Zhao, 2017).
563
+ Parameter (unit)
564
+ Value
565
+ Continuum triangular elements
566
+
567
+ Density, ρ (kgΒ·mβˆ’3)
568
+ 1704
569
+ Young’s modulus, E (GPa)
570
+ 15.6
571
+ Poisson’s ratio, Ο… (-)
572
+ 0.22
573
+ Viscous damping factor, Ξ±
574
+ 1
575
+ Cohesive crack elements
576
+
577
+ Internal cohesion, c (MPa)
578
+ 17.5
579
+ Tensile strength, Οƒt (MPa)
580
+ 2.55
581
+ Friction angle, Ο• (Degree)
582
+ 24.5
583
+ Mode I fracture energy, GIc (JΒ·mβˆ’2)
584
+ 3.8
585
+ Mode II fracture energy, GIIc (JΒ·mβˆ’2)
586
+ 90
587
+ Fracture penalty, Pf (GPa)
588
+ 156
589
+ Normal contact penalty, Pn (GPa)
590
+ 156
591
+ Tangential contact penalty, Pt (GPa)
592
+ 156
593
+
594
+ 2.5 Simulation procedure and boundary conditions
595
+ The simulation was computed using the Irazu FDEM software (Geomechanica Inc., 2021)
596
+ with GPU (graphics processing unit) parallelization. The shear test simulation was conducted in
597
+ three phases (Table 3). In phase 1, the initial normal stress was applied by compressing the sample
598
+ at a constant vertical velocity of 0.2 m/s until the vertical stress reaches 2.5 MPa, which
599
+
600
+ 19
601
+
602
+ corresponds to the initial normal stress condition of the laboratory experiment. In phase 2, the top
603
+ and bottom boxes were constrained to their vertical position, and the horizontal shear velocity was
604
+ increased gradually to 0.3 m/s. This transition phase allows the oscillation induced by the
605
+ instantaneous stop of normal loading to dampen oscillations due to the shear acceleration. In
606
+ phase 3, the top and bottom boxes were fixed in their vertical positions (i.e., this is a constant
607
+ normal stiffness shear test) and moved in the horizontal direction at a constant velocity of 0.3 m/s
608
+ until the desired shear displacement of 3 mm was reached. Note that the loading velocities used in
609
+ the study are significantly higher (1000 times) than those used in laboratory experiments; however,
610
+ such a speed has been verified to provide a quasi-static loading condition while allowing a
611
+ reasonable computation time (Mahabadi, 2012). The model has 26 million simulation time steps,
612
+ and each step represents a simulation time of 4Γ—10βˆ’10 s.
613
+ Table 3 Simulation phases and boundary conditions applied to the model. The applied velocities
614
+ in the x (vx) and y (vy) directions, and the resultant shear displacement (u) are listed.
615
+ Phase Simulation steps
616
+ vx (m/s)[1] vy (m/s)[2] u (mm)
617
+ 1
618
+ 1–66,400
619
+ 0
620
+ 0.1
621
+ 0
622
+ 2
623
+ 66,401–964,000
624
+ 0–0.15[3]
625
+ 0
626
+ 0–0.02
627
+ 3
628
+ 964,000–26,000,000 0.15
629
+ 0
630
+ 0.02–3.02
631
+ [1] Positive (β†’) on the top box and negative (←) on the bottom box.
632
+ [2] Negative (↓) on the top box and positive (↑) on the bottom box.
633
+ [3] Linearly interpolated every time step to ramp up the shear velocity gradually.
634
+
635
+ Normal and shear stresses were measured along a line parallel to the fault and placed 5 mm
636
+ above the rigid box in the bottom sample (Fig. 5a). This measurement line monitored the stress
637
+ conditions every 13,000 simulation steps, equivalent to a 200 kHz monitoring rate. The recorded
638
+ stress values in all elements along the measurement line were averaged to obtain the overall normal
639
+
640
+ 20
641
+
642
+ stress (Οƒn) and shear stress (Ο„), which were used to calculate the friction coefficient ΞΌ = Ο„/Οƒn, similar
643
+ to the laboratory-measured apparent friction coefficient.
644
+ 3 Results and data analysis
645
+ 3.1 Shear behavior
646
+
647
+ Fig. 6 Calculated friction coefficient of (a) the laboratory test results plotted as a function of the
648
+ equivalent slip distance at a radius of 5.6 mm and (b) the numerical simulation. The first ~0.3 mm
649
+ are detailed in Fig. 7. Red arrows indicate significant drops of frictional resistance associated to
650
+ seismic events 1, 2, and 3, which are investigated in Section 3.2.
651
+ The simulated ΞΌ showed a similar trend with the experimental data. It reached the peak
652
+ value of 0.22 at a shear displacement of 0.32 mm, followed by a significant drop (Fig. 6b). The
653
+ simulated ΞΌ experienced many abrupt drops during the slipping process and then stabilized at
654
+ approximately 0.04 after approximately 1.7 mm of shear displacement. The simulated ΞΌ was
655
+ significantly lower than the value reported in the laboratory experiment, with many more
656
+ oscillations.
657
+
658
+ a)
659
+ Experiment
660
+ 1.5
661
+ 1.0
662
+ 0.5
663
+ 0
664
+ 0
665
+ 0.5
666
+ 1.2
667
+ 1.8
668
+ 2.4
669
+ 3.0
670
+ Equivalent slip distance (mm)
671
+ b)
672
+ Simulation
673
+ 0.2
674
+ 0.1
675
+ 0
676
+ Event 2 & 3
677
+ -0.1
678
+ -Event 1
679
+ Fig. 7b
680
+ -0.2
681
+ 0
682
+ 0.5
683
+ 1
684
+ 1.5
685
+ 2
686
+ 2.5
687
+ 3
688
+ Shear displacement (mm)21
689
+
690
+ The simulated stress conditions of the first 0.3 mm showed intriguing similarities to the
691
+ laboratory experimental data (Fig. 7a&b). In this interval, the shear behavior observed in the
692
+ experiment can be divided into four stages (Fig. 7a): (I) Ο„, Οƒn, and the resultant ΞΌ ramped up
693
+ gradually; (II) Ο„ experienced a relatively stable stage with minor change, and Οƒn decreased
694
+ continuously, causing minor change of ΞΌ; (III) Ο„ and Οƒn gradually increased to a peak shear stress,
695
+ and ΞΌ increased to the peak value; and (IV) Ο„, Οƒn, and ΞΌ dropped rapidly.
696
+
697
+ Fig. 7 Comparison of the overall normal and shear stresses and the friction coefficient of stages I-
698
+ IV between (a) the experimental data and (b) the simulated data. (c) and (d) are the zoom in views
699
+ of the local shear and normal stresses, respectively, at the asperity responsible for the stress drop
700
+ at stage IV. Orange circles numbered 1-6 indicate the horizontal shear displacements (u). (e) and
701
+ (f) are the micro-CT image of the laboratory specimen corresponding to frame 1 and 6 in (c) and
702
+ (d). The initial surface profiles from the surface scan data (red curves) are placed next to the
703
+ laboratory fault for comparison.
704
+
705
+ a)Experiment
706
+ b) Simulation
707
+ II
708
+ III
709
+ IV
710
+ II
711
+ III
712
+ IV
713
+ 70.6
714
+ T(MPa)
715
+ 2
716
+ 0.4
717
+ (MPa)
718
+ 0.2
719
+ 0
720
+ 0
721
+ 3
722
+ On (MPa)
723
+ 2.2
724
+ 6
725
+ tttttt!
726
+ 1.8
727
+ 1
728
+ 0.2
729
+ 0.1
730
+ β‰₯0.5
731
+ 、.
732
+ 0
733
+ e)
734
+ 0
735
+ 0
736
+ 0.05
737
+ 0.1
738
+ 0.15
739
+ 0.2
740
+ 0.25
741
+ 0.3
742
+ 0.05
743
+ 0.15
744
+ 0.2
745
+ 0.25
746
+ 0.3
747
+ Shear displacement (mm)
748
+ Shear displacement, u (mm)
749
+ 1
750
+ u=0mm
751
+ β‘‘u=0.092mm
752
+ β‘’ u= 0.136 mm
753
+ 4u=0.183mm
754
+ 5 u=0.320mm
755
+ 6u=0.321mm
756
+ 5mm
757
+ Localshear/normalstress(MPa)
758
+ 0
759
+ 5
760
+ 10
761
+ 2022
762
+
763
+ The numerical simulation qualitatively captured the general trend of these stages (Fig. 7b);
764
+ however, simulated Οƒn in stage I decreased gradually, and more oscillations are observed in the
765
+ curves in the simulated data. To further investigate the mechanisms behind the shear behavior
766
+ during these stages, we examined the simulated local stress conditions around the asperity whose
767
+ breakage was responsible for the large and sudden drop of frictional resistance at stage IV (Fig.
768
+ 7c&d). During stage I, the simulation shows that the shear surface is at the initial contact condition.
769
+ As the shear displacement increases, the shear stress increases gradually due to frictional resistance
770
+ of the initial contact area. Note that the numerical model did not capture the minor normal stress
771
+ increase measured in the experiment at this stage. Such an increase may be related to the interaction
772
+ of the asperities in the direction perpendicular to shear (i.e., out-of-plane motion) that does not
773
+ exist in the 2D simulation. During stage II, the top and bottom surfaces adjusted to a more
774
+ conforming contact, which resulted in the decrease of the shear and normal stress. During stage III,
775
+ new contact points were established, and asperities engage and interlock, causing the shear stress
776
+ to increase rapidly reaching the peak shear stress at the end of this stage. Asperities survived and
777
+ climbed onto each other, causing dilation that increased the normal stress. At stage IV, the highly
778
+ stressed asperity underwent high-stress concentration and failure, releasing the accumulated strain
779
+ energy that resulted in the sudden and significant drop of stresses and frictional resistance. The
780
+ simulated failure pattern, in terms of location and mechanism, resembled the laboratory
781
+ observation (Fig. 7e&f). More importantly, the numerical model can provide the evolution of
782
+ surface contacts and stress conditions throughout the shear process.
783
+ 3.2 Progressive damage, gouge formation, and seismic activity
784
+ Progressive damage on the shear surface and fault gouge formation was simulated by
785
+ BCCEs (Fig. 8). The first several BCCEs occurred when the top and bottom semi-sample were
786
+
787
+ 23
788
+
789
+ loaded with the initial normal stress. Before ⁓1 mm of shear displacement, the damage was
790
+ concentrated in the vicinity of the shear surface. After ⁓1 mm of shear displacement, a number of
791
+ sub-vertical fractures penetrated the sample body, resembling the fracturing observed in the
792
+ laboratory (Fig. 9). The distribution of the shear-induced damage was mostly concentrated close
793
+ to the fault surface and heterogeneously distributed along the fault. Broken asperities formed the
794
+ gouge layer that accumulated between the semi-samples. As a result, some portions of the bare
795
+ fracture surface were protected from wearing (Fig. 9a), and this phenomenon is also observed in
796
+ the laboratory micro-CT image (Fig. 9b).
797
+
798
+ 24
799
+
800
+
801
+ Fig. 8 Damage of the shear surface and the accumulation of gouge material with increasing shear
802
+ displacement. Damage is represented by broken cohesive crack elements.
803
+
804
+ 25
805
+
806
+
807
+ Fig. 9 (a) Zoom-in view of a portion of the simulated fault surface at 3 mm of slip. The dashed red
808
+ lines highlight intact fault walls that were not damaged. (b) Zoom-in view of the micro-CT image
809
+ of a portion of the laboratory fault at a similar location to (a) (adopted from Zhao et al., (2018)).
810
+
811
+ Fig. 10 Simulated seismic activities. (a) Magnitude, location, and failure mode of the clustered
812
+ seismic activity. (b) Event count in each bin.
813
+ A total of 7,557 CCEs were broken throughout the simulation, and they were clustered into
814
+ 1,561 seismic events. Most of the BCCEs near the shear surface failed in shear mode (Mode II),
815
+ and almost all sub-vertical fractures propagated in tensile mode (Mode I). The magnitude of these
816
+ seismic events ranged between βˆ’11.1 and βˆ’4.4, with an average magnitude of βˆ’7.3. In general,
817
+
818
+ a)
819
+ 1mm
820
+ -Intact surface
821
+ Intact surface
822
+ -Gouge
823
+ Gouge
824
+ Shear
825
+ Shear induced fracture
826
+ induced
827
+ fractures
828
+ mmMode I-IH
829
+ 20
830
+ 30
831
+ 35
832
+ 40
833
+ 2026
834
+
835
+ large magnitude events were mostly produced by shear-mode failures, while small magnitude
836
+ events mostly arose from tensile-mode failures (Fig. 10a). Along the vertical direction (y direction),
837
+ the spatial distribution of seismic activity coincides with the damage pattern: events were
838
+ concentrated within Β±4 mm of the fault. We divided the horizontal length of the fault, i.e., the x
839
+ direction, into 100 bins and examined the spatial distribution of the seismic events along such a
840
+ path (Fig. 10b). Seismic events were distributed heterogeneously along the fault: bins at x =
841
+ 29.8 mm and 34.9 mm had the largest number of events at 44; bins at x ranging 10.5 to 12.2 mm
842
+ and 20.5 to 21.0 mm had no seismic events.
843
+ Each asperity failure resulted in the sudden and significant drop of frictional resistance and
844
+ the release of accumulated strain energy. These large magnitude events caused stick-slip-like
845
+ responses and released high amplitude stress waves propagating across the model. Prior to these
846
+ dynamic seismic events, their corresponding locations experienced low shear velocity due to the
847
+ interlocking of asperities and are referred to as interlocking zones (ILZs) in the following
848
+ discussion. Stress concentrated at ILZs and eventually broke the asperities, releasing the
849
+ accumulated strain energy (see animated figures Fig. S1 in Supplementary Material for the velocity
850
+ fields). Three seismic events (Events 1-3, as indicated in Fig. 6) with distinct wave radiation
851
+ patterns are chosen as examples for further examination. Event 1 at u⁓0.3 mm was related to the
852
+ most significant friction drop. Events 2 and 3 were two consecutive events that occurred on the
853
+ slipping surface 12.5 mm apart from each other with a 0.005 ms time delay. We examined the
854
+ stress field (Fig. 11) and observed that the magnitude of seismic events was directly correlated to
855
+ the magnitude of the stress concentration at the asperities that failed. We observed that stress
856
+ concentration at the ILZs reached values as high as the compressive strength of the material,
857
+ causing compressive failure. Due to interlocking, the non-interlocking regions slightly ahead (with
858
+
859
+ 27
860
+
861
+ respect to the shear direction) of the ILZs were subjected to significant tensile stress that reached
862
+ the tensile strength of the material, thus, causing tensile fracturing. By examining the particle
863
+ velocity field (Fig. 12), we found that prior to the seismic events, the locations of the ILZ were
864
+ experiencing particle velocities lower than the loading velocity (i.e., < 0.1 m/s). As the seismic
865
+ events occurred, the source region had particle velocities that were two orders of magnitude higher
866
+ than that of the ILZs (i.e., > 10 m/s). Interestingly, considering that P- and S-wave velocities are
867
+ 2967 m/s and 1884 m/s, respectively, Event 3 occurred right after the arrival of the P-wave induced
868
+ by Event 2, but prior to the arrival of the S-wave. Therefore, Event 3 may have been triggered by
869
+ the stress perturbation from Event 2.
870
+ Fig. 11 Output frames of the numerical model showing the horizontal stress Οƒxx at (a-c) Event 1
871
+ and (d-g) Events 2 and 3. Interlocking zones (ILZs) that are related to the selected seismic events
872
+ are highlighted by the yellow arrows. Note that the simulation time interval between two frames
873
+ is 0.052 ms.
874
+
875
+
876
+ evenl
877
+ erem28
878
+
879
+
880
+ Fig. 12 Output frames of the numerical model showing the particle velocity at (a-c) Event 1 and
881
+ (d-g) Events 2 and 3. ILZs are highlighted by the yellow arrows and the P-wave wavefront of
882
+ Event 2 is labelled. Note the color map is in log scale.
883
+ 4 Discussion
884
+ The FDEM numerical model qualitatively captured the mechanical behavior observed in
885
+ the laboratory experiments, highlighting the dominant role of surface roughness on the shear
886
+ behavior of rocks at low-stress conditions. Both the laboratory experiment and the numerical
887
+ simulation show a slip weakening behavior where the friction coefficient ramps up to the peak
888
+ value and then decreases to a residual value. In the numerical simulation, the shear stress and
889
+ friction reached a steady-state and residual value around ~1.7 mm of total displacement, in
890
+ agreement with the laboratory value (Zhao et al., 2018).
891
+ During the first ~0.3 mm of shear displacement, the discrepancies between the
892
+ experimental and simulation results in stage I and the additional stress oscillations in the simulation
893
+ results may be because the 2D model was not able to capture 3D asperity interactions. However,
894
+ the overall variation trend of stresses and the damage pattern on the shear surface showed close
895
+
896
+ Event!
897
+ front
898
+ ILZ
899
+ Eyent2
900
+ Event329
901
+
902
+ similarities, suggesting that the 2D profile that we used is a proxy for the laboratory specimen.
903
+ This also supports our previous interpretation that the highest asperity was responsible for the
904
+ formation of the large secondary sub-vertical fractures and the associated sudden drop in shear
905
+ resistance (Zhao et al., 2018). These results suggest that our numerical technique, which uses a
906
+ combination of surface scanning, X-ray micro-CT imaging, and FDEM modelling, represents a
907
+ promising approach to simulate realistic fault behavior. Our simulation provides the continuous
908
+ evolution of contacts on the shear surface and the stress conditions that complement the laboratory
909
+ observations in achieving a better comprehension of how the interaction between asperities
910
+ controls the stress conditions and damage patterns in faults.
911
+ During the shear process, asperities interact in various modes including climbing onto each
912
+ other, interlocking, and breaking (Scholz, 1990). Our experimental and numerical results show
913
+ that such interactions directly influenced the stress conditions and damage patterns. When the slip
914
+ displacement is small (u < 1.5 mm), weak asperities (i.e., millimetric scale unevenness) controlled
915
+ the frictional behavior, creating gouge material. These observations agree with the laboratory
916
+ observations on the post-mortem sample and suggest the importance of surface roughness in
917
+ controlling the formation of the gouge layer. As the slip displacement increases (u > 1.5 mm), the
918
+ large-scale roughness of the shear surface (i.e., centimetric scale waviness) becomes important to
919
+ the shear behavior and damage pattern. Large scale waviness causes high stress concentration
920
+ through interlocking and climbing and may cause sub-vertical secondary fractures.
921
+ The damage and seismic event distributions are closely related to the stress heterogeneity
922
+ on the shear surface caused by the surface roughness. Depending on the geometry of the asperity,
923
+ the stress concentration at the ILZs could reach the compressive strength of the material, causing
924
+ compressive failure. This mechanism creates gouge material in the vicinity of the shear surface.
925
+
926
+ 30
927
+
928
+ On the other hand, the areas ahead of the ILZs experience tensile stress up to the tensile strength
929
+ of the material, thus, creating tensile fractures. This mechanism creates large sub-vertical
930
+ secondary fractures. Breakage of strong asperities release the accumulated strain energy in the
931
+ whole model, causing an overall shear stress drop, giving a stick-slip-like shear behavior. Such a
932
+ lock-and-fail mechanism is recently found to be the key process of stick-slip behavior of bare
933
+ surfaces (Chen et al., 2020; Morad et al., 2022). Note that the overall shear loading in our model
934
+ is considered quasi-static, but the local seismic events are dynamic activities with particle velocity
935
+ more than 100 times the quasi-static loading velocity. This suggests that on a rough shear surface,
936
+ quasi-static shear consists of numerous heterogeneously distributed local dynamic seismic
937
+ activities, and this process may complicate the slip process on rough faults and the estimation of
938
+ the energy budget (Tinti et al., 2005).
939
+ Observations on Events 2 and 3 suggest that the stress perturbation from asperities
940
+ breakage may trigger events on adjacent interlocking zones. From an earthquake perspective, there
941
+ are two possible mechanisms that may trigger seismic events in the near field: (1) static stress
942
+ redistribution (e.g., King et al., 1994; Toda et al., 1998) and (2) dynamic stress wave perturbation
943
+ (e.g., Kilb et al., 2000; Gomberg et al., 2001). In our simulation, the modeled body did not slip as
944
+ a rigid body, rather, the slipping consisted of pulses of local movements, accompanied by
945
+ numerous continuously changing of contacts and asperities breakages. When the asperity
946
+ associated to Event 1 breaks, the dynamic stress perturbation was damped out, and the static stress
947
+ concentration is transferred to nearby asperities, which eventually caused failure of other asperities.
948
+ On the other hand, Events 2 and 3 showed a more interesting correlation. Event 3 occurred
949
+ between the arrival times of P- and S-waves from Event 2. Within this time window, stress
950
+ redistribution had not reached a steady state, suggesting that the perturbation of the dynamic stress
951
+
952
+ 31
953
+
954
+ wave radiated from Event 2 may have triggered Event 3. These results imply that static stress
955
+ transfer and dynamic stress perturbation triggering may occur on the same fault and contribute to
956
+ the movement of fault slip. However, due to the limitation of the model output frequency and post-
957
+ processing method, the triggering is not conclusive, Event 2 and 3 may have been independent
958
+ seismic events occurred in a narrow time window, and more investigation is needed in future
959
+ research.
960
+ The numerical simulation has the advantage of continuously modeling the fault shear
961
+ process, fault surface damage, and associated stress conditions. However, the simulated sample
962
+ experienced more damage than the laboratory sample, which is probably related to the limitation
963
+ of 2D simulations not accounting for the motion in the third dimension. For the same reason, the
964
+ simulated stresses suffered significant fluctuations, and the friction coefficient was much lower
965
+ than the experimental measurement, which is a common limitation of 2D simulations. The
966
+ laboratory experiment by Frye and Marone (2002) and the numerical simulation by Hazzard and
967
+ Mair (2003) demonstrated that 2D numerical models exhibit friction values notably lower than 3D
968
+ models and suffer from greater stress fluctuations due to the lack of particle motion in the third
969
+ dimension. In addition, we meshed the shear surface at a relatively high resolution (0.1 mm),
970
+ resulting in a large number of asperities at various sizes. Hence, the interlocking and breakage of
971
+ these asperities caused stress oscillations (i.e., microseismic events). Even though we qualitatively
972
+ captured the shear behavior that matches the laboratory measurements, to fully capture the shear
973
+ behavior of the rotary shear experiment, a 3D model capturing the surface geometry and asperity
974
+ interaction on the entire shear surface will be required.
975
+ 5 Conclusion
976
+
977
+ 32
978
+
979
+ In this study, we used a carefully built and calibrated FDEM numerical model to simulate a
980
+ laboratory shear experiment. We introduced a new clustering algorithm to improve the
981
+ understanding of the simulated fracturing and associated seismic events. The model was able to
982
+ qualitatively capture the frictional behavior observed in the laboratory experiment, providing the
983
+ missing information in the experimental observation regarding the continuous variation of stresses
984
+ and the progressive evolution on the shear surfaces.
985
+ Our numerical model matches the experimental results particularly well at the beginning
986
+ of the shear deformation (~0.3 mm). We were able to identify similar stress variation trends and
987
+ damage patterns. The simulation results provided detailed evolution processes of the contacts on
988
+ the shear surface and the local stress conditions, which are not available in experimental
989
+ observations. Combining the numerical and experimental results, we conclude that interlocking of
990
+ asperities can cause compressive stress concentration on the front side (i.e., facing the shear
991
+ direction) of the asperity, which could induce compressive failure (e.g., crushing) near the shear
992
+ surface; on the other hand, tensile stress concentration is generated on the leeward side of the
993
+ asperity, which could cause sub-vertical tensile fractures that could propagate into the host rock.
994
+ Progressive surface damage and the associated microseismic events occur at the locations of
995
+ asperity interactions and is highly heterogeneous. Several locations experienced no damage even
996
+ after large shear displacement, these locations are either not in contact or were protected by gouge
997
+ materials.
998
+ As a result of the interlocking and breakdown of asperities, local dynamic failure events
999
+ occur, even though the overall loading is quasistatic. These events are considered microseismic
1000
+ events, and their magnitudes range between βˆ’11.1 and βˆ’4.4. Strain energy stored in the medium
1001
+ was released during these events, causing dynamic perturbation to the overall stress condition, and
1002
+
1003
+ 33
1004
+
1005
+ the particle velocity in the source reached > 10 m/s, two orders of magnitude larger than the
1006
+ surrounding regions. This high amplitude stress perturbation could even trigger the failure of
1007
+ adjacent critically stressed asperities.
1008
+ Both the numerical model and the experiment suggested the importance of shear surface
1009
+ roughness in controlling slip behavior, and we were able to explain the laboratory observations
1010
+ with the help of numerical results. Shear surface evolution is a complicated process that involves
1011
+ frictional sliding, fracturing, gouge comminution, and seismicity. The high degree of agreement
1012
+ between simulation and experiment data leads to a promising future of predicting fault behavior
1013
+ through, laboratory testing, surface characterization, and numerical simulations. These results
1014
+ improved the understanding of shear behavior and demonstrated that micromechanical based
1015
+ numerical simulation is a capable approach to study fault mechanics.
1016
+ Declaration of competing interest
1017
+ The authors declare that they have no known competing financial interests or personal
1018
+ relationships that could have appeared to influence the work reported in this paper.
1019
+ Acknowledgements
1020
+ Q. Zhao is supported by the FCE Start-up Fund for New Recruits at the Hong Kong Polytechnic
1021
+ University (Project ID P0034042) and the Early Career Scheme of the Research Grants Council of
1022
+ the Hong Kong Special Administrative Region, China (Project No. PolyU 25220021). This work
1023
+ has also been supported through the NSERC Discovery Grants 341275, CFILOF Grant 18285,
1024
+ Carbon Management Canada (CMC), and NSERC/Energi Simulation Industrial Research Chair
1025
+ Program. The authors would like to thank Geomechanica Inc. for providing the Irazu FDEM
1026
+ simulation software. Q. Zhao would like to thank Dr. Andrea Lisjak and Dr. Bin Chen for
1027
+
1028
+ 34
1029
+
1030
+ discussions and suggestions. The authors appreciate the constructive suggestions and comments
1031
+ from the editor and the reviewers.
1032
+ References
1033
+ Abdelaziz, A., Zhao, Q., & Grasselli, G. (2018). Grain based modelling of rocks using the
1034
+ combined finite-discrete element method. Computers and Geotechnics, 103, 73-81.
1035
+ Asadi, M. S., Rasouli, V., & Barla, G. (2012). A bonded particle model simulation of shear strength
1036
+ and asperity degradation for rough rock fractures. Rock Mechanics and Rock Engineering, 45,
1037
+ 649–675. doi:10.1007/s00603-012-0231-4.
1038
+ Bahaaddini, M., Sharrock, G., & Hebblewhite, B. (2013). Numerical direct shear tests to model
1039
+ the shear behaviour of rock joints. Computers and Geotechnics, 51, 101–115.
1040
+ Bandis, S., Lumsden, A., & Barton, N. (1983). Fundamentals of rock joint deformation.
1041
+ International Journal of Rock Mechanics and Mining Sciences & Geomechanics Abstracts, 20,
1042
+ 249–268.
1043
+ Barenblatt, G. I. (1962). The mathematical theory of equilibrium cracks in brittle fracture.
1044
+ Advances in applied mechanics, 7, 55–129.
1045
+ Beeler, N. M., Tullis, T. E., Blanpied, M. L., & Weeks, J. D. (1996). Frictional behavior of large
1046
+ displacement experimental faults. Journal of Geophysical Research, 101, 8697–8715.
1047
+ Chen, X., Carpenter, B.M. & Reches, Z. Asperity Failure Control of Stick–Slip Along Brittle
1048
+ Faults. Pure Appl. Geophys., 177, 3225–3242 (2020).
1049
+ Crandall, D., Moore, J., Gill, M., & Stadelman, M. (2017). CT scanning and flow measurements
1050
+ of shale fractures after multiple shearing events. International Journal of Rock Mechanics and
1051
+ Mining Sciences, 100, 177-187.
1052
+ Cundall, P. A., & Strack, O. D. (1979). A discrete numerical model for granular assemblies.
1053
+ Geotechnique, 29(1), 47-65.
1054
+ Di Toro, G., Goldsby, D. L., & Tullis, T. E. (2004). Friction falls towards zero in quartz rock as
1055
+ slip velocity approaches seismic rates. Nature, 427, 436–439.
1056
+ Dugdale, D. S. (1960). Yielding of steel sheets containing slits. Journal of the Mechanics and
1057
+ Physics of Solids, 8, 100–104.
1058
+ Evans, R., Marathe, M., 1968. Microcracking and stress-strain curves for concrete in tension.
1059
+ Materials and Structures 1, 61–64.
1060
+ Frye, K. M., & Marone, C. (2002). The effect of particle dimensionality on granular friction in
1061
+ laboratory shear zones. Geophysical Research Letters, 29, 22–1–22–4.
1062
+ Fukuda, D., Mohammadnejad, M., Liu, H., Dehkhoda, S., Chan, A., Cho, S.H., Min, G.J., Han, H.,
1063
+ Kodama, J.I. and Fujii, Y. (2019). Development of a GPGPU‐parallelized hybrid finite‐discrete
1064
+ element method for modeling rock fracture. International journal for numerical and analytical
1065
+ methods in geomechanics, 43(10), 1797-1824.
1066
+
1067
+ 35
1068
+
1069
+ Geomechanica Inc., 2021. Irazu 2D Geomechanical Simulation Software. Theory Manual.
1070
+ Gomberg, J., Reasenberg, P., Bodin, P., & Harris, R. (2001). Earthquake triggering by seismic
1071
+ waves following the landers and hector mine earthquakes. Nature, 411, 462–466.
1072
+ Grasselli, G. (2001). Shear strength of rock joints based on quantified surface description. Ph.D.
1073
+ thesis ENAC Lausanne.
1074
+ Gutenberg, B., 1956. The energy of earthquakes, Quarterly Journal of the Geological Society, 112,
1075
+ 1–14.
1076
+ Hazzard, J. F., & Mair, K. (2003). The importance of the third dimension in granular shear.
1077
+ Geophysical research letters, 30, 1708.
1078
+ Henyey, T. L., and G. J. Wasserburg, Heat flow near major strike-slip faults in central and southern
1079
+ California, J. Geophys. Res., 76, 7924-7946, 1971.
1080
+ Huang, X., Zhao, Q., Qi, S., Xia, K., Grasselli, G., & Chen, X. (2017). Numerical simulation on
1081
+ seismic response of the filled joint under high amplitude stress waves using finite-discrete element
1082
+ method (FDEM). Materials, 10(1), 13.
1083
+ Karami, A., & Stead, D. (2008). Asperity degradation and damage in the direct shear test: A hybrid
1084
+ FEM/DEM approach. Rock Mechanics and Rock Engineering, 41, 229–266. doi:10.1007/s00603-
1085
+ 007-0139-6.
1086
+ Ketcham, R. A., Slottke, D. T., & Sharp Jr, J. M. (2010). Three-dimensional measurement of
1087
+ fractures in heterogeneous materials using high-resolution X-ray computed tomography.
1088
+ Geosphere, 6(5), 499-514.
1089
+ Kilb, D., Gomberg, J., & Bodin, P. (2000). Triggering of earthquake aftershocks by dynamic
1090
+ stresses. Nature, 408(6812), 570-574.
1091
+ Kim, T., Jeon, S. Experimental Study on Shear Behavior of a Rock Discontinuity Under Various
1092
+ Thermal, Hydraulic and Mechanical Conditions. Rock Mech Rock Eng, 52, 2207–2226 (2019).
1093
+ King, G. C., Stein, R. S., & Lin, J. (1994). Static stress changes and the triggering of earthquakes.
1094
+ Bulletin of the Seismological Society of America, 84(3), 935-953.
1095
+ Knight, E.E., Rougier, E., Lei, Z., Euser, B., Chau, V., Boyce, S.H., Gao, K., Okubo, K. and
1096
+ Froment, M. (2020). HOSS: an implementation of the combined finite-discrete element method.
1097
+ Computational Particle Mechanics, 7(5), 765-787.
1098
+ Labuz, J., Shah, S., & Dowding, C. (1985). Experimental analysis of crack propagation in granite.
1099
+ International Journal of Rock Mechanics and Mining Sciences & Geomechanics Abstracts, 22,
1100
+ 85–98.
1101
+ Lachenbruch, A. H., and J. H. Sass, Heat flow from Cajon Pass, fault strength and tectonic
1102
+ implications, J. Geophys. Res., 97, 4995-5015, 1992.
1103
+ Lambert, C., & Coll, C. (2014). Discrete modeling of rock joints with a smooth-joint contact model.
1104
+ Journal of Rock Mechanics and Geotechnical Engineering, 6, 1–12.
1105
+ Lei, Q., Latham, J. P., & Tsang, C. F. (2017). The use of discrete fracture networks for modelling
1106
+ coupled geomechanical and hydrological behaviour of fractured rocks. Computers and
1107
+ Geotechnics, 85, 151-176.
1108
+
1109
+ 36
1110
+
1111
+ Lisjak, A., Liu, Q., Zhao, Q., Mahabadi, O. K., & Grasselli, G. (2013). Numerical simulation of
1112
+ acoustic emission in brittle rocks by two-dimensional finite-discrete element analysis. Geophysical
1113
+ Journal International, 195, 423–443.
1114
+ Lisjak, A., Grasselli, G., & Vietor, T. (2014). Continuum–discontinuum analysis of failure
1115
+ mechanisms around unsupported circular excavations in anisotropic clay shales. International
1116
+ Journal of Rock Mechanics and Mining Sciences, 65, 96-115.
1117
+ Ma, G., Zhou, W., Regueiro, R. A., Wang, Q., & Chang, X. (2017). Modeling the fragmentation
1118
+ of rock grains using computed tomography and combined FDEM. Powder Technology, 308, 388-
1119
+ 397.
1120
+ Mahabadi, O. K. (2012). Investigating the influence of micro-scale heterogeneity and
1121
+ microstructure on the failure and mechanical behaviour of geomaterials. Ph.D. thesis University
1122
+ of Toronto.
1123
+ Mahabadi, O. K., Lisjak, A., Munjiza, & Grasselli, G. (2012). Y-Geo: New Combined Finite-
1124
+ Discrete Element Numerical Code for Geomechanical Applications. International Journal of
1125
+ Geomechanics, 12, 676–688.
1126
+ Marone, C. (1998). Laboratory-derived friction laws and their application to seismic faulting.
1127
+ Annual Review of Earth and Planetary Sciences, 26, 643–696.
1128
+ Mora, P., & Place, D. (1993). A lattice solid model for the nonlinear dynamics of earthquakes.
1129
+ International Journal of Modern Physics C, 4(06), 1059-1074.
1130
+ Mora, P., & Place, D. (1998). Numerical simulation of earthquake faults with gouge: toward a
1131
+ comprehensive explanation for the heat flow paradox. Journal of Geophysical Research, 103,
1132
+ 21067–21089. doi:10.1029/98jb01490.
1133
+ Munjiza, A., Andrews, K.R.F., & White, J.K., (1999). Combined single and smeared crack model
1134
+ in combined finite-discrete element analysis. International Journal for Numerical Methods in
1135
+ Engineering, 44, 41–57.
1136
+ Munjiza, A., & John, N. W. M. (2002). Mesh size sensitivity of the combined FEM/DEM fracture
1137
+ and fragmentation algorithms. Engineering Fracture Mechanics, 69(2), 281-295.
1138
+ Munjiza, A. (2004). The Combined Finite-Discrete Element Method. John Wiley & Sons.
1139
+ Munjiza, A., Owen, D. R. J., & Bicanic, N. (1995). A combined finite-discrete element method in
1140
+ transient dynamics of fracturing solids. Engineering Computations, 12, 145–174.
1141
+ Morad, D., Sagy, A., Tal, Y., & Hatzor, Y. H. (2022). Fault roughness controls sliding instability.
1142
+ Earth and Planetary Science Letters, 579, 117365.
1143
+ Okubo, K., Bhat, H.S., Rougier, E., Marty, S., Schubnel, A., Lei, Z., Knight, E.E. and Klinger, Y.
1144
+ (2019). Dynamics, radiation, and overall energy budget of earthquake rupture with coseismic off‐
1145
+ fault damage. Journal of Geophysical Research, 124(11), 11771-11801.
1146
+ Park, J.-W., & Song, J.-J. (2009). Numerical simulation of a direct shear test on a rock joint using
1147
+ a bonded-particle model. International Journal of Rock Mechanics and Mining Sciences, 46,
1148
+ 1315–1328.
1149
+ Place, D., & Mora, P. (2000). Numerical simulation of localisation phenomena in a fault zone.
1150
+ Pure and Applied Geophysics. 157 (11–12), 1821–1845.
1151
+
1152
+ 37
1153
+
1154
+ Pierce, M., Cundall, P., Potyondy, D., Ivars, M.D. (2007) A synthetic rock mass model for jointed
1155
+ rock Rock Mechanics: Meeting Society's Challenges and Demands, Proceedings of the 1st
1156
+ Canada-US Rock Mechanics Symposium, Taylor & Francis, Vancouver, Canada, pp. 341-349
1157
+ Reches, Z., & Lockner, D. A. (2010). Fault weakening and earthquake instability by powder
1158
+ lubrication. Nature, 467, 452–455.
1159
+ Renard, F., Mair, K., & Gundersen, O. (2012). Surface roughness evolution on experimentally
1160
+ simulated faults. J. Struct. Geol., 45, 101-112.
1161
+ Scholz, C. H. (1990). The Mechanics of Earthquakes and Faulting. Cambridge University Press.
1162
+ Shet, C., & Chandra, N. (2002). Analysis of energy balance when using cohesive zone models to
1163
+ simulate fracture processes. Journal of Engineering Materials and Technology, 124, 440–450.
1164
+ Tatone, B.S.A. & Grasselli. G. (2013). An investigation of discontinuity roughness scale
1165
+ dependency using high-resolution surface measurements. Rock Mech Rock Eng, 46 (4), pp. 657-
1166
+ 681
1167
+ Tatone, B. S. A. (2014). Investigating the evolution of rock discontinuity asperity degradation and
1168
+ void space morphology under direct shear. Ph.D. thesis University of Toronto.
1169
+ Tatone, B. S. A., & Grasselli, G. (2015). A calibration procedure for two-dimensional laboratory-
1170
+ scale hybrid finite-discrete element simulations. International Journal of Rock Mechanics and
1171
+ Mining Sciences, 75, 56–72.
1172
+ Tinti, E., Spudich, P., and Cocco, M. (2005), Earthquake fracture energy inferred from kinematic
1173
+ rupture models on extended faults, J. Geophys. Res., 110, B12303, doi:10.1029/2005JB003644.
1174
+ Tisato, N., Di Toro, G., De Rossi, N., Quaresimin, M., & Candela, T. (2012). Experimental
1175
+ investigation of flash weakening in limestone. Journal of Structural Geology, 38, 183–199.
1176
+ Toda, S., Stein, R. S., Richards‐Dinger, K., & Bozkurt, S. B. (2005). Forecasting the evolution
1177
+ of seismicity in southern California: Animations built on earthquake stress transfer. Journal of
1178
+ Geophysical Research, 110(B5).
1179
+ Van der Pluijm, B. A., & Marshak, S. (2004). Earth structure: an introduction to structural
1180
+ geology and tectonics. New York.
1181
+ Yan, C., Zheng, H., Sun, G., & Ge, X. (2016). Combined finite-discrete element method for
1182
+ simulation of hydraulic fracturing. Rock mechanics and rock engineering, 49(4), 1389-1410.
1183
+ Zhao, Q. (2017). Investigating brittle rock failure and associated seismicity using laboratory
1184
+ experiments and numerical simulations. Ph.D. thesis University of Toronto.
1185
+ Zhao, Q., Lisjak, A., Mahabadi, O. K., Liu, Q., & Grasselli, G. (2014). Numerical simulation of
1186
+ hydraulic fracturing and associated microseismicity using finite-discrete element method. Journal
1187
+ of Rock Mechanics and Geotechnical Engineering, 6, 574–581.
1188
+ Zhao, Q., Tisato, N., Grasselli, G., Mahabadi, O. K., Lisjak, A., & Liu, Q. (2015). Influence of in
1189
+ situ stress variations on acoustic emissions: a numerical study. Geophysical Journal International,
1190
+ 203, 1246–1252.
1191
+ Zhao, Q., Tisato, N., & Grasselli, G. (2017). Rotary shear experiments under X-ray micro-
1192
+ computed tomography. Review of Scientific Instruments, 88(1), 015110.
1193
+
1194
+ 38
1195
+
1196
+ Zhao, Q.; Tisato, N.; Kovaleva, O.; and Grasselli, G. (2018). Direct Observation of Faulting by
1197
+ Means of Rotary Shear Tests Under X-Ray Micro-Computed Tomography. Journal of
1198
+ Geophysical Research, 123(9).
1199
+ Zhao, Q.; Glaser, S.; Tisato, N.; and Grasselli, G. (2020). Assessing Energy Budget of
1200
+ Laboratory Fault Slip Using Rotary Shear Experiments and Micro-Computed Tomography.
1201
+ Geophysical Research Letters, 47(1).
1202
+
LNE2T4oBgHgl3EQfpwjv/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
LdE0T4oBgHgl3EQfzwKd/content/tmp_files/2301.02677v1.pdf.txt ADDED
@@ -0,0 +1,1135 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Resonant inelastic X-ray scattering in topological semimetal FeSi
2
+ Yao Shen,1 Anirudh Chandrasekaran,2, 3 Jennifer Sears,1 Tiantian Zhang,4, 5 Xin Han,6 Youguo Shi,6
3
+ Jiemin Li,7 Jonathan Pelliciari,7 Valentina Bisogni,7 Mark P. M. Dean,1, βˆ— and Stefanos Kourtis8, 2, †
4
+ 1Condensed Matter Physics and Materials Science Department,
5
+ Brookhaven National Laboratory, Upton, New York 11973, USA
6
+ 2Department of Physics, Boston University, Boston, MA, 02215, USA
7
+ 3Department of Physics and Centre for the Science of Materials,
8
+ Loughborough University, Loughborough LE11 3TU, UK
9
+ 4Department of Physics, Tokyo Institute of Technology, Okayama, Meguro-ku, Tokyo, Japan
10
+ 5Tokodai Institute for Element Strategy, Tokyo Institute of Technology, Nagatsuta, Midori-ku, Yokohama, Kanagawa, Japan
11
+ 6Beijing National Laboratory for Condensed Matter Physics,
12
+ Institute of Physics, Chinese Academy of Sciences, Beijing 100190, China
13
+ 7National Synchrotron Light Source II, Brookhaven National Laboratory, Upton, New York 11973, USA
14
+ 8Institut quantique & DΒ΄epartement de physique, UniversitΒ΄e de Sherbrooke, J1K 2R1, QuΒ΄ebec, Canada
15
+ (Dated: January 10, 2023)
16
+ The energy spectrum of topological semimetals contains protected degeneracies in reciprocal space that corre-
17
+ spond to Weyl, Dirac, or multifold fermionic states. To exploit the unconventional properties of these states, one
18
+ has to access the electronic structure of the three-dimensional bulk. In this work, we resolve the bulk electronic
19
+ states of candidate topological semimetal FeSi using momentum-dependent resonant inelastic X-ray scattering
20
+ (RIXS) at the Fe L3 edge. We observe a broad excitation continuum devoid of sharp features, consistent with
21
+ particle-hole scattering in an underlying electronic band structure. Using density functional theory, we calculate
22
+ the electronic structure of FeSi and derive a band theory formulation of RIXS in the fast collision approximation
23
+ to model the scattering process. We find that band theory qualitatively captures the number and position of the
24
+ main spectral features, as well as the overall momentum dependence of the RIXS intensity. Our work paves the
25
+ way for targeted studies of band touchings in topological semimetals with RIXS.
26
+ I.
27
+ INTRODUCTION
28
+ In the last decade and a half, topological matter has become
29
+ a cornerstone of quantum materials science [1]. The discovery
30
+ of three-dimensional topological insulators [2], in particular,
31
+ sparked a flurry of activity in the then nascent field. Elec-
32
+ trons in these crystalline materials are effectively noninteract-
33
+ ing, giving rise to electronic bands in the bulk that are indis-
34
+ tinguishable from those of a trivial band insulator. The elec-
35
+ tronic wavefunction, however, is characterized by topologi-
36
+ cal indices that dictate the presence of symmetry-protected
37
+ Dirac states at the surface of the material, as well as nontrivial
38
+ (magneto-)transport responses.
39
+ More recently, topological semimetals have been added
40
+ to the catalogue of three-dimensional topological materi-
41
+ als [3, 4]. These systems also feature topologically protected
42
+ boundary states and nontrivial (magneto-)transport, but addi-
43
+ tionally have distinct geometric characteristics in their bulk
44
+ band structure. In the simplest case of Weyl semimetals, these
45
+ geometric characteristics are singly degenerate energy sur-
46
+ faces in reciprocal space that contain a band touching pointβ€”a
47
+ Weyl nodeβ€”around which electronic bands disperse linearly
48
+ in all three directions in reciprocal space [5–7]. Such band
49
+ touchings are Berry curvature singularities characterized by
50
+ topological indices. The value of the topological index of a
51
+ nodal point determines the geometry of the band dispersion in
52
+ βˆ— mdean@bnl.gov
53
+ † Stefanos.Kourtis@usherbrooke.ca
54
+ the vicinity of nodal points [8–10]. Conversely, the geometry
55
+ of the bulk bands becomes a proxy of topology in these ma-
56
+ terials. Measuring the electronic density of states in the bulk
57
+ can therefore reveal the topological nature of a semimetal.
58
+ Resonant inelastic X-ray scattering (RIXS) is a spectro-
59
+ scopic technique that yields momentum- and energy-resolved
60
+ spectra of charge-neutral electronic excitations. While RIXS
61
+ has been extensively used in studying magnetic excitations in
62
+ gapped materials like insulators and superconductors [11, 12],
63
+ it is increasingly applied in studies of compounds that host
64
+ itinerant carriers with small or no charge gaps [13–15]. That
65
+ RIXS can be used to map electronic bands of materials, in-
66
+ cluding semimetals, has long been established [16–23]. Im-
67
+ provements in resolution in recent years have renewed in-
68
+ terest in using RIXS to detect band structure effects at meV
69
+ energy scales in materials of technological interest, such as
70
+ unconventional superconductors [24]. There are even theo-
71
+ retical proposals to use RIXS to measure topological indices
72
+ of nodal points in topological semimetals [25, 26].
73
+ These
74
+ prospects are particularly appealing for probing the bulk of
75
+ three-dimensional materials, since alternative techniques such
76
+ as angle-resolved photoemission (ARPES) and scanning tun-
77
+ neling spectroscopy, probe predominantly the surface rather
78
+ than the bulk. Furthermore, topological nodal points may only
79
+ appear above the Fermi level or as a result of an applied mag-
80
+ netic field, settings in which the resolving power of ARPES is
81
+ limited. As these settings may be relevant to the technologi-
82
+ cal exploitation of topological materials, alternative methods
83
+ to visualize the bulk band structure and to identify topologi-
84
+ cal features are sought after. Before honing in on properties
85
+ of topological origin, however, one has to determine whether
86
+ arXiv:2301.02677v1 [cond-mat.str-el] 6 Jan 2023
87
+
88
+ 2
89
+ bulk band structure effects at large are detectable in RIXS of
90
+ topological semimetals.
91
+ In this work, we use RIXS to probe the bulk of FeSi aim-
92
+ ing to quantitatively test how bulk band structure manifests
93
+ in RIXS spectra of a putative topological semimetal [27, 28].
94
+ We observe broad continua in the RIXS spectra, consistent
95
+ with particle-hole scattering in an underlying band structure.
96
+ We model the RIXS process in the fast-collision approxima-
97
+ tion using the band structure of FeSi as determined by den-
98
+ sity functional theory (DFT) calculations. We find reasonable
99
+ agreement between experiment and theory in the number and
100
+ hierarchy of dominant spectral features. We interpret this find-
101
+ ing as evidence of a bulk band structure in FeSi, with many-
102
+ body effects playing only a secondary role in determining the
103
+ RIXS spectrum. Our results indicate that higher resolution
104
+ experiments β€” feasible with existing instruments β€” could vi-
105
+ sualize topological nodal points and thus identify and classify
106
+ topological semimetals.
107
+ II.
108
+ THEORY
109
+ A.
110
+ RIXS cross section and fast collision approximation
111
+ We briefly introduce the theoretical description of RIXS at
112
+ zero temperature. More comprehensive presentations of RIXS
113
+ can be found in Refs. 29 and 30.
114
+ In a RIXS experiment, core electrons of an ion are pro-
115
+ moted to a state above the Fermi level Ξ΅F by an intense x-
116
+ ray beam, thereby locally exciting the irradiated material into
117
+ a highly energetic and short-lived intermediate state. Subse-
118
+ quently, the core hole recombines with a valence electron. The
119
+ process imparts both energy and momentum to particle-hole
120
+ excitations in the material. In what follows, we will consider
121
+ excitation of core electrons directly into orbital(s) close to Ξ΅F ,
122
+ which give rise to the low-energy physics in the material. This
123
+ process, which is often referred to as direct RIXS, is illus-
124
+ trated in Fig. 1.
125
+ The double differential cross section is a measure of the
126
+ total RIXS intensity. Up to a constant prefactor, it is given by
127
+ I(kin, kout, Ο‰in, Ο‰out, Ο΅in, Ο΅out)
128
+ =
129
+ οΏ½
130
+ fg
131
+ |Ffg(kin, kout, Ο‰in, Ο΅in, Ο΅out)|2Ξ΄(Eg βˆ’ Ef + β„βˆ†Ο‰) ,
132
+ (1)
133
+ where β„βˆ†Ο‰ = ℏ(Ο‰in βˆ’ Ο‰out) is the energy transferred to the
134
+ material, kin and kout (Ο΅in and Ο΅out) the incoming and outgoing
135
+ photon wavevectors (polarizations) and Eg and Ef the ener-
136
+ gies corresponding to initial and final many-body states |g⟩
137
+ and |f⟩ of the valence electrons. The scattering amplitude
138
+ Ffg in the dipole approximation is
139
+ Ffg(kin, kout, Ο‰in, Ο΅in, Ο΅out)
140
+ = ⟨f| οΏ½D†(Ο΅out, kout) οΏ½G(Ο‰in) οΏ½D(Ο΅in, kin)|g⟩ ,
141
+ (2)
142
+ where οΏ½G is the intermediate-state propagator
143
+ οΏ½G(Ο‰in) = (Eg + ℏωin + iΞ“ βˆ’ οΏ½
144
+ H)βˆ’1 ,
145
+ (3)
146
+ with οΏ½
147
+ H the Hamiltonian describing the system in the interme-
148
+ diate excited state and Ξ“ the intermediate-state inverse life-
149
+ time. The dipole operators οΏ½D and οΏ½D † represent the x-ray ab-
150
+ sorption and emission, respectively. For a crystalline material,
151
+ they can be written as
152
+ οΏ½D(Ο΅, k) = Ο΅ Β· οΏ½Dk ,
153
+ (4)
154
+ οΏ½Dk =
155
+ οΏ½
156
+ Β΅,Ξ½
157
+ ⟨¡|οΏ½r βˆ’ rΒ΅|ν⟩
158
+ οΏ½
159
+ R
160
+ eikΒ·R οΏ½d †
161
+ RΒ΅ οΏ½pRΞ½ ,
162
+ (5)
163
+ where R is the lattice position. States |¡⟩ and |ν⟩ express
164
+ single-electron valence and core states respectively. The com-
165
+ bined valence (core) index Β΅ (Ξ½) encodes spin, orbital, and
166
+ sublattice degrees of freedom.
167
+ Core states |ν⟩ are conve-
168
+ niently expressed as atomic orbitals, whereas valence states
169
+ |¡⟩ can be appropriately chosen Wannier functions, both lo-
170
+ calized in space around the same position rΒ΅ of each atomic
171
+ site within the unit cell. The position operator �r, defined with
172
+ respect to each lattice position R, is the same for all ions.
173
+ For the L2/3 and M2/3 resonant edges, the operators οΏ½d †
174
+ RΒ΅ and
175
+ οΏ½pRΞ½ create a d-orbital electron and a p-orbital core hole, re-
176
+ spectively.
177
+ Physical arguments allow us to simplify the RIXS scatter-
178
+ ing amplitude. First, core holes do not hop appreciably; they
179
+ are created and annihilated at the same site. Taking this into
180
+ account, Ffg becomes [30]
181
+ Ffg(q, Ο‰in, Ο΅in, Ο΅out)
182
+ =
183
+ οΏ½
184
+ Β΅,Ξ½,Β΅β€²,Ξ½β€²
185
+ T¡ν¡′ν′(Ο΅in, Ο΅out)F¡ν¡′ν′(q, Ο‰in) ,
186
+ (6)
187
+ where q = kin βˆ’ kout. The scattering amplitude has been
188
+ factored in the atomic scattering tensor
189
+ T¡ν¡′ν′(Ο΅in, Ο΅out) = ⟨¡|Ο΅out Β· οΏ½r|Ξ½βŸ©βˆ—βŸ¨Β΅β€²|Ο΅in Β· οΏ½r|Ξ½β€²βŸ©
190
+ (7)
191
+ and the fundamental scattering amplitude
192
+ F¡ν¡′ν′(q, Ο‰in)
193
+ = ⟨f|
194
+ οΏ½
195
+ R
196
+ eβˆ’iqΒ·R οΏ½p †
197
+ RΞ½ οΏ½dRΒ΅ οΏ½G(Ο‰in) οΏ½d †
198
+ RΒ΅β€² οΏ½pRΞ½β€²|g⟩ .
199
+ (8)
200
+ The intrinsic spectral characteristics of a material are carried
201
+ by the tensor F, which is typically the main quantity of inter-
202
+ est in theoretical studies. The tensor T modulates the scatter-
203
+ ing amplitude according to the geometry of the localized core
204
+ and valence states. The entries of T can be calculated given
205
+ knowledge of the valency of the targeted ion and the symme-
206
+ try group of the crystal [31–34].
207
+ Then, within the fast-collision approximation, one assumes
208
+ that οΏ½G(Ο‰in) β‰ˆ 1/Ξ“, where Ξ“ is the inverse core-hole lifetime.
209
+ In this approximation, the RIXS process reduces to the intro-
210
+ duction of a particle-hole excitation with fixed momentum and
211
+ energy in the material β€” see Fig 1 for an example.
212
+ Before proceeding to derive the theory of RIXS in band
213
+ structures, we evaluate the geometric modulation of the RIXS
214
+ spectrum owing purely to the orbital content of the quantum
215
+
216
+ 3
217
+ Ξ΅k
218
+ k
219
+ Ξ΅F
220
+ energy
221
+ core level
222
+ Ξ΅F
223
+ Ξ΅k
224
+ k
225
+ INITIAL
226
+ energy
227
+ core level
228
+ Ξ΅F
229
+ Ξ΅k
230
+ k
231
+ FINAL
232
+ fast
233
+ collision
234
+ approximation
235
+ FIG. 1. Illustration of the direct RIXS process and reduction to effective particle-hole scattering via the fast-collision approximation.
236
+ states involved in the RIXS process. This is obtained by set-
237
+ ting the fundamental scattering amplitude F¡ν¡′ν′ to unity:
238
+ T (Ο΅in, Ο΅out) =
239
+ οΏ½οΏ½οΏ½οΏ½οΏ½οΏ½
240
+ οΏ½
241
+ Β΅,Ξ½,Β΅β€²,Ξ½β€²
242
+ T¡ν¡′ν′(Ο΅in, Ο΅out)
243
+ οΏ½οΏ½οΏ½οΏ½οΏ½οΏ½
244
+ 2
245
+ .
246
+ (9)
247
+ We shall use T as a diagnostic to disentangle contributions to
248
+ the modulation of the RIXS intensity as a function of scatter-
249
+ ing angles. The first contribution comes through the polariza-
250
+ tion vectors, which are angle-dependent β€” see Fig. 2. The
251
+ second contribution is the intrinsic momentum dependence
252
+ coming from electronic dispersion in the material. In Sec. V
253
+ we calculate the geometric modulation in Eq. (9) for FeSi and
254
+ compare it to the scattering angle dependence of RIXS inten-
255
+ sity.
256
+ B.
257
+ RIXS process in a band structure
258
+ We wish to describe the RIXS response of crystalline mate-
259
+ rials in which electrons are, to a good approximation, non-
260
+ interacting.
261
+ Valence electrons in these materials are well-
262
+ described by band theory. The states |g⟩ and |f⟩ in Eq. (2)
263
+ are then collections of Bloch modes.
264
+ For a given RIXS edge, one then sums over core states
265
+ |ν⟩ and valence Wannier states |¡⟩ and
266
+ οΏ½οΏ½Β΅β€²οΏ½
267
+ connected by the
268
+ dipole operators οΏ½D, οΏ½D†. Here we study the Fe L3 edge, hence
269
+ we consider 2p3/2 orbitals for core electrons and the 3d shell
270
+ for valence electrons.
271
+ In k-space, the band eigenbasis is given by a unitary rota-
272
+ tion of a basis of Wannier states |¡⟩ per lattice position R to a
273
+ basis of Bloch states |k¡⟩. The Wannier states have wavefunc-
274
+ tions ϕ¡(x) = ⟨x|¡⟩ that are centered about different points
275
+ in the unit cell, possibly atomic sites. Let Ο•kΒ΅(x) = ⟨x|k¡⟩
276
+ be the spatial wavefunction of |k¡⟩, which could be a spinor.
277
+ We then have
278
+ Ο•kΒ΅(x) =
279
+ 1
280
+ √
281
+ N
282
+ οΏ½
283
+ R
284
+ eikΒ·Rϕ¡(x βˆ’ R) .
285
+ (10)
286
+ The raising and lowering operators of the Bloch wavefunc-
287
+ tions are οΏ½d †
288
+ kΒ΅ and οΏ½dkΒ΅. They are defined by οΏ½d †
289
+ kΒ΅|β„¦βŸ© = |k¡⟩,
290
+ {οΏ½dkΒ΅, οΏ½dkβ€²Β΅β€²} = 0 and {οΏ½dkΒ΅, οΏ½d †
291
+ kβ€²Β΅β€²} = Ξ΄k,k′δ¡,Β΅β€², where |β„¦βŸ© is
292
+ the vacuum of valence excitations and core holes. A general
293
+ Hamiltonian describing noninteracting valence electrons is
294
+ οΏ½
295
+ Hband =
296
+ οΏ½
297
+ k∈BZ
298
+ οΏ½
299
+ Β΅,Β΅β€²
300
+ οΏ½d †
301
+ kΒ΅ H¡¡′(k) οΏ½dkΒ΅β€² ,
302
+ (11)
303
+ where H¡¡′(k) are the elements of the matrix H(k).
304
+ Let U(k) be a matrix that diagonalizes H(k), such that
305
+ U †(k)H(k)U(k) is a diagonal matrix containing the eigen-
306
+ values Ξ΅l(k), which constitute the dispersing bands. We can
307
+ then write
308
+ οΏ½dkΒ΅ =
309
+ οΏ½
310
+ l
311
+ U¡l(k) �ψkl,
312
+ (12)
313
+ where l denotes an energy band index and �ψkl annihilates the
314
+ corresponding eigenstate. The ground state at zero tempera-
315
+ ture is obtained by populating all states below the Fermi level:
316
+ |g⟩ =
317
+ οΏ½
318
+ οΏ½ οΏ½
319
+ l,k∈BZ
320
+ Θ(Ξ΅F βˆ’ Ξ΅l(k)) οΏ½Οˆβ€ 
321
+ kl
322
+ οΏ½
323
+ οΏ½ |β„¦βŸ© ,
324
+ (13)
325
+ with Θ the Heaviside step function. The Wannier lowering
326
+ operators at any lattice site R can be expressed in terms of the
327
+ band operators as
328
+ οΏ½dRΒ΅ = 1
329
+ √
330
+ N
331
+ οΏ½
332
+ k∈BZ
333
+ eβˆ’ikΒ·R οΏ½dkΒ΅
334
+ (14a)
335
+ = 1
336
+ √
337
+ N
338
+ οΏ½
339
+ l,k∈BZ
340
+ eβˆ’ikΒ·R UΒ΅l(k) �ψkl.
341
+ (14b)
342
+ We now assume, as per the fast collision approximation,
343
+ that the intermediate-state Hamiltonian is well approximated
344
+ by the band Hamiltonian, along with a core hole inverse life-
345
+ time Ξ“ in the intermediate-state propagator. Due to this as-
346
+ sumption, core electron operators cancel out and the interme-
347
+ diate state propagator becomes simply
348
+ οΏ½G(Ο‰in) =(Eg + ℏωin + iΞ“ βˆ’ οΏ½
349
+ Hband)βˆ’1 ,
350
+ (15a)
351
+ =
352
+ οΏ½
353
+ k
354
+ οΏ½
355
+ l
356
+ |k, l⟩⟨k, l|
357
+ Eg + ℏωin + iΞ“ βˆ’ Ξ΅l(k) .
358
+ (15b)
359
+
360
+ 4
361
+ where |k, l⟩ are band eigenstates, and we treat Eg and Ξ“ as
362
+ free parameters to be determined by fitting the x-ray absorp-
363
+ tion spectrum (see App. B).
364
+ Substituting the expression for οΏ½dRΒ΅ in Eq. (14b) in Eq. (8),
365
+ we obtain the fundamental RIXS scattering amplitude in a
366
+ band structure
367
+ F¡¡′(q, Ο‰in) = ⟨f|
368
+ οΏ½
369
+ k,kβ€²βˆˆBZ
370
+ οΏ½
371
+ l,lβ€²
372
+ 1
373
+ N
374
+ οΏ½
375
+ R
376
+ eβˆ’i(k+qβˆ’kβ€²)Β·R
377
+ Γ— UΒ΅l(k) U βˆ—
378
+ Β΅β€²lβ€²(kβ€²) �ψkl οΏ½G(Ο‰in) �ψ †
379
+ kβ€²lβ€²|g⟩ .
380
+ (16)
381
+ Notice that F is independent of the core orbitals at this level
382
+ of description. The sum over R evaluates to NΞ΄kβ€²,k+q, which
383
+ enforces kβ€² = k + q. When Ξ΅lβ€²(k + q) > Ξ΅F, we have that
384
+ �ψ †
385
+ k+qlβ€²|g⟩ is an eigenstate of the band Hamiltonian with en-
386
+ ergy Eg + Ξ΅lβ€²(k + q) (otherwise the single particle state is
387
+ already occupied and this term evaluates to zero). The action
388
+ of οΏ½G(Ο‰in) on �ψ †
389
+ k+qlβ€²|g⟩ is
390
+ οΏ½G(Ο‰in) �ψ †
391
+ k+qlβ€²|g⟩ =
392
+ Θ(Ξ΅lβ€²(k + q) βˆ’ Ξ΅F)
393
+ ℏωin βˆ’ Ξ΅lβ€²(k + q) + iΞ“
394
+ �ψ †
395
+ k+qlβ€²|g⟩.
396
+ (17)
397
+ Furthermore, the action of �ψkl on �ψ †
398
+ k+qlβ€²|g⟩ is non-zero only
399
+ if Ξ΅l(k) < Ξ΅F (we need this single particle level to be occupied
400
+ for the term to be non-zero). Using this we obtain
401
+ F¡¡′(q, Ο‰in) =
402
+ οΏ½
403
+ l,lβ€²
404
+ οΏ½
405
+ k∈BZ
406
+ οΏ½
407
+ ⟨f| �ψkl �ψ †
408
+ k+qlβ€²|g⟩
409
+ Γ— Θ(Ξ΅lβ€²(k + q) βˆ’ Ξ΅F)
410
+ Γ— Θ(Ξ΅F βˆ’ Ξ΅l(k))
411
+ Γ—
412
+ UΒ΅l(k) U βˆ—
413
+ Β΅β€²lβ€²(k + q)
414
+ ℏωin βˆ’ Ξ΅lβ€²(k + q) + iΞ“
415
+ οΏ½
416
+ .
417
+ (18)
418
+ The sum over final states |f⟩ can be taken over the eigenstates
419
+ of οΏ½
420
+ Hband. The pair of step functions in the fundamental scat-
421
+ tering amplitude given above in Eq. (18) ensures that there
422
+ is a unique |f⟩ that makes the inner product ⟨f| �ψkl �ψ †
423
+ k+qlβ€²|g⟩
424
+ non-zero, since the role of the operator pair is to simply cre-
425
+ ate particle-hole excitations across the Fermi level. Thus, the
426
+ final sum over |f⟩ can be replaced as
427
+ οΏ½
428
+ f
429
+ β†’
430
+ οΏ½
431
+ l,lβ€²
432
+ οΏ½
433
+ k∈BZ
434
+ Θ(Ξ΅lβ€²(k + q) βˆ’ Ξ΅F) Θ(Ξ΅F βˆ’ Ξ΅l(k)).
435
+ (19)
436
+ This corresponds to summing over final states with one
437
+ particle-hole excitation in the valence bands. The inner prod-
438
+ uct is then redundant and can be removed.
439
+ The final form of the RIXS intensity for systems well-
440
+ described by band theory is
441
+ I(q, Ο‰in, βˆ†Ο‰, Ο΅in, Ο΅out) =
442
+ οΏ½
443
+ l,lβ€²
444
+ οΏ½
445
+ k∈BZ
446
+ Θ(Ξ΅lβ€²(k + q) βˆ’ Ξ΅F) Θ(Ξ΅F βˆ’ Ξ΅l(k))
447
+ Γ—
448
+ οΏ½οΏ½οΏ½οΏ½οΏ½οΏ½
449
+ οΏ½
450
+ Β΅,Ξ½,Β΅β€²
451
+ ⟨¡|Ο΅out Β· οΏ½r|Ξ½βŸ©βˆ—βŸ¨Β΅β€²|Ο΅in Β· οΏ½r|ν⟩
452
+ UΒ΅l(k) U βˆ—
453
+ Β΅β€²lβ€²(k + q)
454
+ ℏωin βˆ’ Ξ΅lβ€²(k + q) + iΞ“
455
+ οΏ½οΏ½οΏ½οΏ½οΏ½οΏ½
456
+ 2
457
+ Γ—
458
+ Ξ·
459
+ [Ξ΅l(k) βˆ’ Ξ΅lβ€²(k + q) + β„βˆ†Ο‰]2 + Ξ·2 ,
460
+ (20)
461
+ where we have replaced the Dirac Ξ΄-function with a
462
+ Lorentzian of peak broadening η to represent finite experi-
463
+ mental resolution.
464
+ With respect to a local set of coordinate axes, the incoming
465
+ X-ray polarization Ο΅ has components (Ο΅x, Ο΅y, Ο΅z). The outgo-
466
+ ing polarization is usually not measured, and hence one sums
467
+ over either polarizations parallel to and perpendicular to the
468
+ scattering plane or over left, linear, and right polarizations.
469
+ We list the polarization matrix elements for the specific case
470
+ of the L3 edge of a 3d transition-metal in Table I.
471
+ III.
472
+ EXPERIMENTAL METHODS
473
+ A.
474
+ Sample preparation
475
+ Single crystals of FeSi were prepared using a Ga flux
476
+ method.
477
+ We mixed the starting materials in a molar ratio
478
+ of 1:1:15 in a glove box filled with argon.
479
+ This mixture
480
+ was placed in an alumina crucible and sealed in an evacuated
481
+ quartz tube. The crucible was heated to 1150β—¦C and held for
482
+ 10 h, before cooling to 950β—¦C at 2 K/h, after which the flux
483
+ was centrifuged. The crystals were washed with diluted hy-
484
+ drochloric acid in order to remove Ga flux from the surface of
485
+ the samples.
486
+
487
+ 5
488
+ B.
489
+ RIXS and experimental geometry
490
+ FIG. 2. Schematic of the RIXS setup. kin and kout respectively
491
+ denote the ingoing and outgoing scattering vectors. The components
492
+ of the ingoing and outgoing photon polarization within the scattering
493
+ plane are denoted by Ο€in and Ο€out while the Οƒ polarization direction
494
+ is the same for both. The incident angle ΞΈi is measured with respect
495
+ to the sample surface, that is the a direction in the sample coordinates
496
+ while the c direction is the normal to the sample surface.
497
+ RIXS measurements were performed at the Soft Inelas-
498
+ tic X-Ray (SIX) beamline at the National Syncrotron Light
499
+ Source-II (NSLS-II). The energy resolution was 23 meV. The
500
+ experimental setup is depicted in Fig. 2. The lab coordinates
501
+ are denoted by x, y, z while the crystallographic axes are la-
502
+ belled a, b, c. We define the incident and outgoing beam an-
503
+ gles with respect to the sample coordinate system, wherein
504
+ the c direction is the normal to the sample surface and the ac
505
+ plane is the scattering plane. Experimental data are corrected
506
+ to account for self-absorption effects.
507
+ With ΞΈi denoting the incident angle measured with respect
508
+ to the sample surface and 2ΞΈ denoting the angle between the
509
+ incident and outgoing beam, we can easily verify the follow-
510
+ ing in the sample frame:
511
+ kin = kin(βˆ’ cos ΞΈi, 0, βˆ’ sin ΞΈi) ,
512
+ (21a)
513
+ kout = kout (βˆ’ cos(2ΞΈ βˆ’ ΞΈi), 0, sin(2ΞΈ βˆ’ ΞΈi)) ,
514
+ (21b)
515
+ ϡπ,in = (βˆ’ sin ΞΈi, 0, cos ΞΈi) ,
516
+ (21c)
517
+ ϡπ,out = (sin(2ΞΈ βˆ’ ΞΈi), 0, cos(2ΞΈ βˆ’ ΞΈi)) ,
518
+ (21d)
519
+ ϡσ,in = ϡσ,out = (0, 1, 0) .
520
+ (21e)
521
+ Although in reality the ingoing and outgoing photon mo-
522
+ mentum magnitudes kin and kout is different owing to non-
523
+ zero energy transfer βˆ†Ο‰, the difference is negligible since
524
+ βˆ†Ο‰ β‰ͺ Ο‰in, and hence kin β‰ˆ kout = k. The momentum
525
+ transfer to the material q is then
526
+ q = k (cos(2ΞΈ βˆ’ ΞΈi) βˆ’ cos ΞΈi, 0, βˆ’ sin(2ΞΈ βˆ’ ΞΈi) βˆ’ sin ΞΈi) .
527
+ (22)
528
+ 0
529
+ 1
530
+ 2
531
+ 3
532
+ 4
533
+ 5
534
+ 705 706 707 708 709 710 705 706 707 708 709 710
535
+ Energy Loss (eV)
536
+ (a) Experiment
537
+ (b) Theory
538
+ Incident energy (eV)
539
+ 705 706 707 708 709 710
540
+ 0
541
+ 0.1
542
+ 0.2
543
+ 0.3
544
+ 0.4
545
+ 0.5
546
+ 0.6
547
+ 0.7
548
+ 0.8
549
+ 0.9
550
+ 1
551
+ RIXS (arb. units)
552
+ FIG. 3. Color maps of RIXS intensity with 2ΞΈ = 150β—¦ and ΞΈi = 68β—¦
553
+ for Ο€-polarized incident beam as a function of incident photon energy
554
+ ℏωin and energy loss βˆ†Ο‰ at the L3 edge of Fe in FeSi as obtained (a)
555
+ in experiment and (b) in the band theory formalism of Sec. II B.
556
+ In practice the ingoing and outgoing beam directions are set
557
+ to specific values, which defines a specific 2θ. By rotating
558
+ the sample about the y axis (or equivalently b axis), we can
559
+ change q by changing ΞΈi .
560
+ IV.
561
+ DENSITY FUNCTIONAL THEORY AND
562
+ TIGHT-BINDING MODEL
563
+ The band structure of FeSi was simulated in a similar way
564
+ to prior studies of FeSi [35]. We performed first-principles
565
+ calculations based on DFT [36] within the Perdew-Burke-
566
+ Ernzerhof exchange-correlation [37] implemented in the Vi-
567
+ enna ab initio simulation package (VASP) [38]. The plane-
568
+ wave cutoff energy was 450 eV with a 9Γ—9Γ—9 k-mesh in the
569
+ BZ for self-consistent calculation without considering spin-
570
+ orbit coupling. Maximally localized Wannier functions [39]
571
+ were used to obtain the tight-binding model of bulk FeSi with
572
+ the lattice constants a = b = c = 4.48 ˚A.
573
+ V.
574
+ RIXS SPECTRUM OF FESI
575
+ The RIXS intensity at the Fe L3 edge with Ο€-polarized in-
576
+ cident beam is shown in Fig. 3 in the incident energy-energy
577
+ loss plane. The absence of prominent sharp inelastic features
578
+ suggests a particle-hole continuum, consistent with particle-
579
+ hole excitations in a partially filled band structure.
580
+ Momentum-resolved RIXS spectra are shown in Fig. 4 for
581
+ two values of 2ΞΈ. Spectral weight from inelastic processes
582
+ lies predominantly in a window of width ∼ 5 eV. Within that
583
+ window, all spectra have similar lineshape, featuring a peak
584
+ around 2 eV that disperses to higher energies with increasing
585
+ ΞΈi and a dispersionless shoulder above 3 eV. Overall inelastic
586
+ intensity increases with increasing ΞΈi for both values of 2ΞΈ.
587
+ We use the band theory formulation of Sec. II B and
588
+ Eq. (20) to theoretically model the RIXS process in FeSi. A
589
+ fit of the absorption spectrum (see App. B) yields Ξ“ = 0.8
590
+
591
+ 6
592
+ 0
593
+ 1
594
+ 2
595
+ 3
596
+ 4
597
+ 5
598
+ 0
599
+ 1
600
+ 2
601
+ 3
602
+ 4
603
+ 5
604
+ Exp
605
+ (a) 2οΏ½ D 150Δ±
606
+ (b) 2οΏ½ D 70Δ±
607
+ RIXS (arb. units)
608
+ οΏ½i
609
+ οΏ½i
610
+ Thy
611
+ Energy loss (eV)
612
+ FIG. 4. Momentum-resolved RIXS spectra at ℏωin = 708.7 eV
613
+ with Ο€ x-ray polarization for (a) 2ΞΈ = 150β—¦ and (b) 2ΞΈ = 70β—¦
614
+ and comparison to simulations within band theory (bottom pan-
615
+ els) using Eq. (20) with Ξ“ = 0.8 eV and Ξ΅0
616
+ = 707.67 eV.
617
+ For scattering angle 2ΞΈ = 150β—¦, the incident angle values are
618
+ ΞΈi = 10β—¦, 30β—¦, 45β—¦, 68β—¦, 120β—¦, while for 2ΞΈ = 70β—¦ we have ΞΈi =
619
+ 10β—¦, 30β—¦, 60β—¦.
620
+ eV and Eg = 707.67 eV, and we choose a peak broadening
621
+ Ξ· = 100 meV. We use a 48 Γ— 48 Γ— 48 grid of k points in the
622
+ Brillouin zone for the 32-band tight binding model detailed
623
+ above.
624
+ The simulated spectra show a structure similar to that ob-
625
+ served experimentally, with inelastic weight in a ∼ 5 eV win-
626
+ dow containing a peak at βˆ†Ο‰ ∼ 2.5 eV and shoulder at
627
+ βˆ†Ο‰ ∼ 3.5 eV. As in experiment, overall inelastic intensity
628
+ increases with increasing ΞΈi for both values of 2ΞΈ, though to
629
+ a lesser extent. Compared to experiment, features are shifted
630
+ to higher energies in simulated spectra, while for 2ΞΈ = 70β—¦
631
+ and ΞΈi = 60β—¦ the main peak subsides, leading to theory and
632
+ experimental spectra that look qualitatively different. Finally,
633
+ experimental spectra also contain subdominant features close
634
+ to the elastic line (βˆ†Ο‰ < 1 eV) that, as discussed later, deviate
635
+ from the band theory predictions.
636
+ To investigate what causes the overall increase in RIXS in-
637
+ tensity with increasing ΞΈi, we calculate the atomic scattering
638
+ tensor (7). From this we obtain the modulation of the RIXS
639
+ spectrum (9) coming purely from the orbital content of core
640
+ and valence states. After summing over outgoing polariza-
641
+ tions, we obtain the behavior shown in Fig. 5. Comparing to
642
+ the ΞΈi dependence of RIXS spectra in Fig. 4, we see that ge-
643
+ ometric considerations are insufficient to explain the momen-
644
+ tum dependence of RIXS intensity in experiment: the mo-
645
+ mentum dependence of the atomic scattering tensor is differ-
646
+ ent from that observed, even showing a reverse trend in the
647
+ 0
648
+ 0.5
649
+ 1
650
+ 1.5
651
+ 2
652
+ 2.5
653
+ 0Δ±
654
+ 30Δ±
655
+ 60Δ±
656
+ 90Δ±
657
+ 120Δ± 150Δ± 0
658
+ 0.2
659
+ 0.4
660
+ 0.6
661
+ 0.8
662
+ 1
663
+ 1.2
664
+ 1.4
665
+ 0Δ±
666
+ 30Δ±
667
+ 60Δ±
668
+ 90Δ±
669
+ 120Δ± 150Δ±
670
+ Orbital RIXS
671
+ Incident angle οΏ½i
672
+ (a) 2οΏ½ D 150Δ±
673
+ (b) 2οΏ½ D 70Δ±
674
+ T .οΏ½; οΏ½/ C T .οΏ½; οΏ½/
675
+ FIG. 5. The contribution of the dipole matrix elements to the RIXS
676
+ spectrum of FeSi for Ο€ ingoing polarisation as given by Eq. (9) after
677
+ summing over Οƒ and Ο€ polarisations of the outgoing beam.
678
+ case of 2ΞΈ = 70β—¦. Reinstating the band structure fundamen-
679
+ tal scattering amplitudes in Eq. (20) yields the experimentally
680
+ observed trend of overall momentum dependence, albeit only
681
+ qualitatively, as shown in Fig. 4.
682
+ VI.
683
+ DISCUSSION & CONCLUSION
684
+ We have seen that the theoretical formulation of RIXS
685
+ based on band theory captures the overall momentum depen-
686
+ dence of the experimental Fe L3-edge RIXS spectra of FeSi
687
+ better than a calculation based on just the atomic multiplet.
688
+ Band theory also reproduces the right bandwidth for the in-
689
+ elastic part of the spectrum, as well as the right number of
690
+ features therein, at roughly the right energy.
691
+ Discrepancies between theory and experiment exist. While
692
+ the overall momentum dependence of the RIXS spectrum is
693
+ reproduced by band theory, experimental spectra depend more
694
+ sensitively on ΞΈi.
695
+ Spectral features also do not align per-
696
+ fectly between experiment and theory. This includes a fea-
697
+ ture around 0.3 eV in the experiment, which is only present
698
+ as a weak shoulder in the theory. We discuss potential rea-
699
+ sons for these discrepancies. First, the band theory of RIXS
700
+ ignores electronic correlations. The extent to which correla-
701
+ tions play a role in the 3d shell of FeSi is not clear [40, 41]. A
702
+ more detailed theoretical study of the RIXS spectrum would
703
+ require identifying the precise type, range, and magnitude of
704
+ interactions present in FeSi. Fully incorporating the effects
705
+ of interactions in theoretical studies of RIXS is nevertheless a
706
+ challenge, since we are dealing with a three-dimensional ma-
707
+ terial with 32 relevant orbitals per unit cell. Numerical sim-
708
+ ulation of the RIXS spectrum based on dynamical mean field
709
+ theory [42] may eventually be up to this task. Second, in the
710
+ fast collision approximation we ignore the effects of a finite
711
+ core-hole lifetime, which may be appreciable in 3d transition
712
+ metal compounds [43]. Future simulations could be improved
713
+ by incorporating dynamics and interactions with the core hole
714
+ in the intermediate state.
715
+ In conclusion, we have reported RIXS spectra of FeSi at
716
+
717
+ 7
718
+ the Fe L3 edge. We observe an excitation continuum without
719
+ sharp features. Through a band theory formulation of RIXS
720
+ in the fast collision approximation, we model the RIXS pro-
721
+ cess using the ab initio band structure of FeSi. We obtain
722
+ reasonable agreement for the spectrum bandwidth, as well as
723
+ the number and position of main features. Theory also repro-
724
+ duces the dispersion trend of the RIXS spectrum, albeit only
725
+ qualitatively. This work paves the path to ever finer resolution
726
+ of distinctive band structure features in topological materials
727
+ with RIXS.
728
+ ACKNOWLEDGMENTS
729
+ A.C. was supported by DOE Grant No.
730
+ DE-FG02-
731
+ 06ER46316 and EPSRC grant EP/T034351/1.
732
+ S.K. ac-
733
+ knowledges support from the Minist`ere de l’´Economie et de
734
+ l’Innovation du QuΒ΄ebec and the Canada First Research Excel-
735
+ lence Fund. Work at Brookhaven National Laboratory (x-ray
736
+ scattering and analysis) was supported by the U.S. Depart-
737
+ ment of Energy, Office of Science, Office of Basic Energy
738
+ Sciences. This research used resources at the SIX beamline
739
+ of the National Synchrotron Light Source II, a U.S. DOE Of-
740
+ fice of Science User Facility operated for the DOE Office of
741
+ Science by Brookhaven National Laboratory under Contract
742
+ No. DE-SC0012704. We acknowledge National Natural Sci-
743
+ ence Foundation of China (U2032204), the Strategic Prior-
744
+ ity Research Program of the Chinese Academy of Sciences
745
+ (XDB33010000) for funding sample synthesis. We thank Yue
746
+ Cao, Siddhant Das, Claudio Chamon, Michael El-Batanouny,
747
+ Jungho Kim, and Karl Ludwig for useful discussions.
748
+ Appendix A: Polarization matrix elements and the atomic
749
+ scattering tensor
750
+ The DFT derived tight-binding model used for the calcula-
751
+ tions presented in the paper involves thirty two basis orbitals
752
+ per unit cell of the crystal lattice. Due to the assumption of
753
+ zero spin orbit coupling for the valence bands, this gives rise
754
+ to thirty-two, two-fold spin-degenerate bands. The orbitals
755
+ used are the five 3d orbitals of each of the four Fe atoms and
756
+ the three 3p orbitals of each of the four silicon atoms within a
757
+ unit cell, giving a total of 32 orbitals per unit cell.
758
+ Since the tight binding model is expressed in terms of 3d or-
759
+ bitals whose local axes are perfectly aligned with crystal axis
760
+ for each of the four Fe atoms in the unit cell, we need to com-
761
+ pute the matrix elements of the 2p3/2 β†’ 3d transitions for
762
+ just one of the atoms. The 2p orbitals all have the same radial
763
+ part of the wavefunction, Ο†2p(r) and, likewise, the 3d orbitals
764
+ have same radial wavefunction Ο†3d(r). The radial integral of
765
+ the various matrix elements in the atomic scattering tensor is
766
+ simply the radial integral of the product Ο†2p(r) Β· Ο†3d(r) and
767
+ the radial part of the dipole transition operator. Since this is a
768
+ common term that just provides an overall multiplicative fac-
769
+ tor for the RIXS cross section, we ignore it and compute only
770
+ the azimuthal and polar integrals of the matrix elements. We
771
+ document the relevant matrix elements of the dipole operator
772
+ in Table I, which were verified by comparing to open source
773
+ RIXS code EDRIXS [44].
774
+ Appendix B: X-ray absorption spectrum and theoretical fit
775
+ To align the experimental RIXS spectra with theoretical re-
776
+ sults obtained through ab initio calculations, we need to de-
777
+ termine the absolute energy scale Eg of the initial state. We
778
+ determine Eg through a fit of the experimental X-ray absorp-
779
+ tion intensity with the calculated absorption spectrum
780
+ Iabs(q, Ο‰in, Ο΅in) =
781
+ οΏ½
782
+ Ο΅out
783
+ οΏ½
784
+ l,lβ€²
785
+ οΏ½
786
+ k∈BZ
787
+ Θ(Ξ΅lβ€²(k + q) βˆ’ Ξ΅F) Θ(Ξ΅F βˆ’ Ξ΅l(k))
788
+ οΏ½οΏ½οΏ½οΏ½οΏ½οΏ½
789
+ οΏ½
790
+ Β΅,Ξ½,Β΅β€²
791
+ ⟨¡|Ο΅out Β· οΏ½r|Ξ½βŸ©βˆ—βŸ¨Β΅β€²|Ο΅in Β· οΏ½r|ν⟩ UΒ΅l(k) U βˆ—
792
+ Β΅β€²lβ€²(k + q)
793
+ Eg + ℏωin βˆ’ Ξ΅lβ€²(k + q) + iΞ“
794
+ οΏ½οΏ½οΏ½οΏ½οΏ½οΏ½
795
+ 2
796
+ ,
797
+ (B1)
798
+ which is obtained by integrating over βˆ†Ο‰ the RIXS spectrum
799
+ in Eq. (20). In addition to Eg, we consider the core hole in-
800
+ verse lifetime Ξ“ as a tunable parameter in the fit. We sum over
801
+ outgoing polarizations since the measured spectrum is not po-
802
+ larization resolved. Fig. 6 shows the fit that minimizes the
803
+ average absolute deviation and yields the values Ξ“ = 0.8 eV
804
+ and Eg = 707.67 eV.
805
+ [1] M. Z. Hasan and C. L. Kane, Reviews of modern physics 82,
806
+ 3045 (2010).
807
+ [2] M. Z. Hasan and J. E. Moore, Annu. Rev. Condens. Matter
808
+ Phys. 2, 55 (2011).
809
+ [3] A. Burkov, Nature materials 15, 1145 (2016).
810
+ [4] N. P. Armitage, E. J. Mele,
811
+ and A. Vishwanath, Rev. Mod.
812
+ Phys. 90, 015001 (2018).
813
+ [5] J. von Neumann and E. P. Wigner, in Collect. Work. Eugene
814
+
815
+ 8
816
+ J = 3
817
+ 2, Jz = βˆ’ 3
818
+ 2 J = 3
819
+ 2, Jz = βˆ’ 1
820
+ 2
821
+ J = 3
822
+ 2, Jz = 1
823
+ 2
824
+ J = 3
825
+ 2, Jz = 3
826
+ 2
827
+ d3z2βˆ’r2↑
828
+ (0, 0, 0)
829
+ οΏ½
830
+ βˆ’ 1
831
+ 6, i
832
+ 6, 0
833
+ οΏ½
834
+ οΏ½
835
+ 0, 0, 2
836
+ 3
837
+ οΏ½
838
+ οΏ½
839
+ 1
840
+ 2
841
+ √
842
+ 3,
843
+ i
844
+ 2
845
+ √
846
+ 3, 0
847
+ οΏ½
848
+ d3z2βˆ’r2↓
849
+ οΏ½
850
+ βˆ’
851
+ 1
852
+ 2
853
+ √
854
+ 3,
855
+ i
856
+ 2
857
+ √
858
+ 3, 0
859
+ οΏ½
860
+ οΏ½
861
+ 0, 0, 2
862
+ 3
863
+ οΏ½
864
+ οΏ½ 1
865
+ 6, i
866
+ 6, 0
867
+ οΏ½
868
+ (0, 0, 0)
869
+ dxz↑
870
+ (0, 0, 0)
871
+ οΏ½
872
+ 0, 0,
873
+ 1
874
+ 2
875
+ √
876
+ 3
877
+ οΏ½
878
+ οΏ½
879
+ 1
880
+ √
881
+ 3, 0, 0
882
+ οΏ½
883
+ οΏ½
884
+ 0, 0, βˆ’ 1
885
+ 2
886
+ οΏ½
887
+ dxz↓
888
+ οΏ½
889
+ 0, 0, 1
890
+ 2
891
+ οΏ½
892
+ οΏ½
893
+ 1
894
+ √
895
+ 3, 0, 0
896
+ οΏ½
897
+ οΏ½
898
+ 0, 0, βˆ’
899
+ 1
900
+ 2
901
+ √
902
+ 3
903
+ οΏ½
904
+ (0, 0, 0)
905
+ dyz↑
906
+ (0, 0, 0)
907
+ οΏ½
908
+ 0, 0, βˆ’
909
+ i
910
+ 2
911
+ √
912
+ 3
913
+ οΏ½
914
+ οΏ½
915
+ 0,
916
+ 1
917
+ √
918
+ 3, 0
919
+ οΏ½
920
+ οΏ½
921
+ 0, 0, βˆ’ i
922
+ 2
923
+ οΏ½
924
+ dyz↓
925
+ οΏ½
926
+ 0, 0, βˆ’ i
927
+ 2
928
+ οΏ½
929
+ οΏ½
930
+ 0,
931
+ 1
932
+ √
933
+ 3, 0
934
+ οΏ½
935
+ οΏ½
936
+ 0, 0, βˆ’
937
+ i
938
+ 2
939
+ √
940
+ 3
941
+ οΏ½
942
+ (0, 0, 0)
943
+ dx2βˆ’y2↑
944
+ (0, 0, 0)
945
+ οΏ½
946
+ 1
947
+ 2
948
+ √
949
+ 3,
950
+ i
951
+ 2
952
+ √
953
+ 3, 0
954
+ οΏ½
955
+ (0, 0, 0)
956
+ οΏ½
957
+ βˆ’ 1
958
+ 2, i
959
+ 2, 0
960
+ οΏ½
961
+ dx2βˆ’y2↓
962
+ οΏ½ 1
963
+ 2, i
964
+ 2, 0
965
+ οΏ½
966
+ (0, 0, 0)
967
+ οΏ½
968
+ βˆ’
969
+ 1
970
+ 2
971
+ √
972
+ 3,
973
+ i
974
+ 2
975
+ √
976
+ 3, 0
977
+ οΏ½
978
+ (0, 0, 0)
979
+ dxy↑
980
+ (0, 0, 0)
981
+ οΏ½
982
+ βˆ’
983
+ i
984
+ 2
985
+ √
986
+ 3,
987
+ 1
988
+ 2
989
+ √
990
+ 3, 0
991
+ οΏ½
992
+ (0, 0, 0)
993
+ οΏ½
994
+ βˆ’ i
995
+ 2, βˆ’ 1
996
+ 2, 0
997
+ οΏ½
998
+ dxy↓
999
+ οΏ½
1000
+ βˆ’ i
1001
+ 2, 1
1002
+ 2, 0
1003
+ οΏ½
1004
+ (0, 0, 0)
1005
+ οΏ½
1006
+ βˆ’
1007
+ i
1008
+ 2
1009
+ √
1010
+ 3, βˆ’
1011
+ 1
1012
+ 2
1013
+ √
1014
+ 3, 0
1015
+ οΏ½
1016
+ (0, 0, 0)
1017
+ TABLE I. Dipole matrix elements relevant for the L3 edge of FeSi. Only the polar and azimuthal integrals are evaluated since the radial
1018
+ integral is the same for all the core-valence pairs, and provides only an overall prefactor to the theoretical RIXS spectrum.
1019
+ 704
1020
+ 705
1021
+ 706
1022
+ 707
1023
+ 708
1024
+ 709
1025
+ 710
1026
+ 0
1027
+ 500
1028
+ 1000
1029
+ 1500
1030
+ 2000
1031
+ Incident Energy (eV)
1032
+ XAS (arb. units)
1033
+ FIG. 6. Optimal average absolute deviation fit to the L3-edge x-
1034
+ ray absorption spectrum (XAS) of FeSi. The black dots represent
1035
+ the experimental absorption spectrum while the continuous blue line
1036
+ represents the theoretical spectrum calculated using the tight-binding
1037
+ model described in Sec IV. The fit yields Ξ“ = 0.8 eV and Eg =
1038
+ 707.67 eV.
1039
+ Paul Wigner (Springer Berlin Heidelberg, Berlin, Heidelberg,
1040
+ 1993) pp. 294–297.
1041
+ [6] C. Herring, Phys. Rev. 52, 365 (1937).
1042
+ [7] E. Blount, Solid State Phys. 13, 305 (1962).
1043
+ [8] M. Onoda and N. Nagaosa, J. Phys. Soc. Japan 71, 19 (2002).
1044
+ [9] C. Fang, M. J. Gilbert, and B. A. Bernevig, Phys. Rev. B 86,
1045
+ 115112 (2012).
1046
+ [10] B. Bradlyn, J. Cano, Z. Wang, M. Vergniory, C. Felser, R. J.
1047
+ Cava, and B. A. Bernevig, Science 353, aaf5037 (2016).
1048
+ [11] S. Fatale, S. Moser, and M. Grioni, Journal of Electron Spec-
1049
+ troscopy and Related Phenomena 200, 274 (2015), special An-
1050
+ niversary Issue: Volume 200.
1051
+ [12] M. Dean, Journal of Magnetism and Magnetic Materials 376, 3
1052
+ (2015).
1053
+ [13] N. B. Brookes, D. Betto, K. Cao, Y. Lu, K. Kummer,
1054
+ and
1055
+ F. Giustino, Phys. Rev. B 102, 064412 (2020).
1056
+ [14] J. Pelliciari, S. Lee, K. Gilmore, J. Li, Y. Gu, A. Barbour, I. Jar-
1057
+ rige, C. H. Ahn, F. J. Walker, and V. Bisogni, Nature Materials
1058
+ 20, 188 (2021).
1059
+ [15] X. Jia, A. Haldar, J. Kim, Y. Wang, G. Fabbris, K. Ludwig,
1060
+ S. Kourtis, M. Upton, Y. Liu, W. Lu, X. Luo, Y.-P. Sun, D. Casa,
1061
+ S. Sharifzadeh, P. T. Darancet, and Y. Cao, β€œInterplay of broken
1062
+ symmetry and delocalized excitations in the insulating state of
1063
+ 1t-tas2,” (2022).
1064
+ [16] Y. Ma, Phys. Rev. B 49, 5799 (1994).
1065
+ [17] P. D. Johnson and Y. Ma, Phys. Rev. B 49, 5024 (1994).
1066
+ [18] J. A. Carlisle, E. L. Shirley, E. A. Hudson, L. J. Terminello,
1067
+ T. A. Callcott, J. J. Jia, D. L. Ederer, R. C. C. Perera, and F. J.
1068
+ Himpsel, Phys. Rev. Lett. 74, 1234 (1995).
1069
+ [19] J. A. Carlisle, E. L. Shirley, L. J. Terminello, J. J. Jia, T. A.
1070
+ Callcott, D. L. Ederer, R. C. C. Perera, and F. J. Himpsel, Phys.
1071
+ Rev. B 59, 7433 (1999).
1072
+ [20] J. D. Denlinger, J. A. Clack, J. W. Allen, G.-H. Gweon, D. M.
1073
+ Poirier, C. G. Olson, J. L. Sarrao, A. D. Bianchi, and Z. Fisk,
1074
+ Phys. Rev. Lett. 89, 157601 (2002).
1075
+ [21] K. Kokko, V. Kulmala, J. A. Leiro, and W. Hergert, Phys. Rev.
1076
+ B 68, 052503 (2003).
1077
+ [22] V. N. Strocov,
1078
+ T. Schmitt,
1079
+ J.-E. Rubensson,
1080
+ P. Blaha,
1081
+ T. Paskova, and P. O. Nilsson, Phys. Status Solidi B 241, R27
1082
+ (2004).
1083
+ [23] C. Monney, M. Herzog, A. Pulkkinen, Y. Huang, J. Pelli-
1084
+ ciari, P. Olalde-Velasco, N. Katayama, M. Nohara, H. Takagi,
1085
+ T. Schmitt, and T. Mizokawa, Phys. Rev. B 102, 085148 (2020).
1086
+ [24] M. KanΒ΄asz-Nagy, Y. Shi, I. Klich, and E. A. Demler, Phys. Rev.
1087
+ B 94, 165127 (2016).
1088
+ [25] S. Kourtis, Phys. Rev. B 94, 125132 (2016).
1089
+ [26] M. SchΒ¨uler, T. Schmitt, and P. Werner, β€œProbing magnetic or-
1090
+ bitals and berry curvature with circular dichroism in resonant
1091
+ inelastic x-ray scattering,” (2022).
1092
+ [27] S. Changdar,
1093
+ S. Aswartham,
1094
+ A. Bose,
1095
+ Y. Kushnirenko,
1096
+ G. Shipunov, N. C. Plumb, M. Shi, A. Narayan, B. B¨uchner,
1097
+ and S. Thirupathaiah, Phys. Rev. B 101, 235105 (2020).
1098
+ [28] Y. Ohtsuka, N. Kanazawa, M. Hirayama, A. Matsui, T. Nomoto,
1099
+ R. Arita, T. Nakajima, T. Hanashima, V. Ukleev, H. Aoki, et al.,
1100
+ Science advances 7, eabj0498 (2021).
1101
+ [29] A. Kotani and S. Shin, Rev. Mod. Phys. 73, 203 (2001).
1102
+
1103
+ 9
1104
+ [30] L. J. P. Ament, M. van Veenendaal, T. P. Devereaux, J. P. Hill,
1105
+ and J. van den Brink, Rev. Mod. Phys. 83, 705 (2011).
1106
+ [31] M. W. Haverkort, Phys. Rev. Lett. 105, 167404 (2010).
1107
+ [32] M. W. Haverkort, N. Hollmann, I. P. Krug,
1108
+ and A. Tanaka,
1109
+ Phys. Rev. B 82, 094403 (2010).
1110
+ [33] L. J. P. Ament, G. Khaliullin, and J. van den Brink, Phys. Rev.
1111
+ B 84, 020403 (2011).
1112
+ [34] B. J. Kim and G. Khaliullin, Phys. Rev. B 96, 085108 (2017).
1113
+ [35] H. Miao, T. T. Zhang, L. Wang, D. Meyers, A. H. Said, Y. L.
1114
+ Wang, Y. G. Shi, H. M. Weng, Z. Fang, and M. P. M. Dean,
1115
+ Phys. Rev. Lett. 121, 035302 (2018).
1116
+ [36] P. Hohenberg and W. Kohn, Phys. Rev. 136, B864 (1964).
1117
+ [37] J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77,
1118
+ 3865 (1996).
1119
+ [38] G. Kresse and J. Furthm¨uller, Phys. Rev. B 54, 11169 (1996).
1120
+ [39] N. Marzari and D. Vanderbilt, Phys. Rev. B 56, 12847 (1997).
1121
+ [40] G. Wertheim, V. Jaccarino, J. Wernick, J. Seitchik, H. Williams,
1122
+ and R. Sherwood, Physics Letters 18, 89 (1965).
1123
+ [41] J. M. Tomczak, K. Haule,
1124
+ and G. Kotliar, Proceedings
1125
+ of the National Academy of Sciences 109, 3243 (2012),
1126
+ https://www.pnas.org/doi/pdf/10.1073/pnas.1118371109.
1127
+ [42] A. Hariki, M. Winder, T. Uozumi, and J. Kuneˇs, Phys. Rev. B
1128
+ 101, 115130 (2020).
1129
+ [43] V. Bisogni, S. Kourtis, C. Monney, K. Zhou, R. Kraus, C. Sekar,
1130
+ V. Strocov, B. B¨uchner, J. van den Brink, L. Braicovich,
1131
+ T. Schmitt, M. Daghofer, and J. Geck, Phys. Rev. Lett. 112,
1132
+ 147401 (2014).
1133
+ [44] Y. Wang, G. Fabbris, M. P. Dean, and G. Kotliar, Computer
1134
+ Physics Communications 243, 151 (2019).
1135
+
LdE0T4oBgHgl3EQfzwKd/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
MNAyT4oBgHgl3EQfgfgN/content/tmp_files/2301.00358v1.pdf.txt ADDED
@@ -0,0 +1,1317 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ arXiv:2301.00358v1 [gr-qc] 1 Jan 2023
2
+ Noname manuscript No.
3
+ (will be inserted by the editor)
4
+ Nonsingular Black Holes in Higher dimensions
5
+ Bikash Chandra Paul
6
+ Received: date / Accepted: date
7
+ Abstract We present a class of new nonsingular black holes in higher dimensional
8
+ theories of gravity. Assuming a specific form of the stress energy tensor exact an-
9
+ alytic solutions of the field equation are generated in general theory of relativity
10
+ (GR) and Rastall theory. The non-singular black hole solutions are obtained with
11
+ a finite pressure at the centre in D = 4 dimensions. For D > 4 the transverse pres-
12
+ sure is found finite at the centre for a set of model parameters. In the later case the
13
+ transverse pressure is more than that in the usual four dimensions. The exact ana-
14
+ lytic solution of the field equations in higher dimensions for large r coincides with
15
+ the Schwarzschild black hole solution in the usual four and in higher dimensions
16
+ which is singularity free. The different features of the generalized non-singular
17
+ black hole in GR and modified GR are explored. A new vacuum nonsingular black
18
+ hole is found in Rastall gravity. We also study the motion of massive and massless
19
+ particles around the black holes.
20
+ 1 Introduction
21
+ The idea that spacetime dimensions should be extended from four to higher dimen-
22
+ sions came from the seminal work of Kaluza and Klein [1,2] who first tried to unify
23
+ gravity with electromagnetism. The Kaluza-Klein approach has been revived and
24
+ considerably generalized after realizing that many interesting theories of particle
25
+ interactions need spacetime dimensions more than four for their formulation. Dur-
26
+ ing the last few decades considerable research activities in progress to understand
27
+ the quantum properties of gravity. The investigation seems to lead some people to
28
+ believe that a consistent theory of quantum gravity cannot be obtained within the
29
+ framework of point-field theories. For example, superstring theory is considered
30
+ B. C. Paul
31
+ E-mail: bcpaul@nbu.ac.in
32
+ Department of Physics, University of North Bengal, Siliguri, Dist. : Darjeeling 734 013, West
33
+ Bengal, India
34
+ and
35
+ IUCAA Centre for Astronomy Research and Development, North Bengal
36
+
37
+ 2
38
+ Bikash Chandra Paul
39
+ to be the promising candidate which may unify gravity with the other fundamen-
40
+ tal forces in nature which requires ten dimensions for consistent formulation. The
41
+ advent of string theory has opened up new and interesting possibilities in this con-
42
+ text. The discovery that a supergravity theory coupled to Yang-Mills fields with a
43
+ gauge group SO(32) or E8 Γ—E8 is anomaly-free in ten dimensions had inspired con-
44
+ siderable activities in this area. Although the expected breakthrough has not yet
45
+ come, worldwide hectic activities have served to focus on a number of issues which
46
+ need further investigations in higher dimensions. The present ideas in dimensional
47
+ reduction suggest that our cosmos may be a 3-brane evolving in a D-dimensional
48
+ spacetime. Cadeau and Woolgar [3] addressed this issue in the context of black
49
+ holes which led to homogeneous but non- FRW-braneworld cosmologies. Classical
50
+ General Relativity (GR) in more than the usual four dimensions is thus a sub-
51
+ ject of increasing attention in recent years. A successful development of counting
52
+ of the five dimensional black hole entropy [4] and the AdS/CFT correspondence
53
+ relates the properties of a D-dimensional black hole with those of a quantum field
54
+ theory in (D βˆ’ 1) dimensions [5]. There has been a growing interest to investigate
55
+ the physics of higher-dimensional black holes [6] which is markedly different, and
56
+ much richer in structure compared to four dimensions.
57
+ In connection with localized sources, higher dimensional generalization of the
58
+ spherically symmetric Schwarzschild, Reisner-Nordstr¨om black holes, Kerr black
59
+ holes can be found in the literature [7,8,9,10,11,12]. The generalization of the
60
+ rotating Kerr black hole [8,9,13,14] and black holes in compactified spacetime [7,
61
+ 12] are also found in the literature. The linearized stability of the black holes [15],
62
+ no hair theorems [16], black hole thermodynamics and Hawking radiation have
63
+ also been investigated. Mandelbrot [17] investigted the problem on the variability
64
+ of dimensions and describe how a ball of thin thread is seen as an observer changes
65
+ scale. An object which looks like a point object from a very large distance becomes
66
+ a three-dimensional ball visible at a closer distance. Therefore at various scales the
67
+ ball appears to change shape as an observer moves down. While the embedding
68
+ dimensions for the ball has not changed, the effective dimensions of the contents
69
+ however also remains same. It is possible that there are compact [18] or non-
70
+ compact [19] dimensions present at a certain point. In this case the (3 + 1) metric
71
+ is simply not true, although one obtains a valid description with general relativity.
72
+ We also probe the black hole solution in Rastall theory [20] which is prescribed
73
+ by a modification of GR to accommodate the present accelerating universe [21,22,
74
+ 23].
75
+ Regular (i.e. non-singular) black holes have been initiated by Bardeen [24] and
76
+ thereafter a number of black hole solutions in four dimensions have been obtained
77
+ [25,26,27,28,29,30,31,32,33]. In this case one can find metrics which are spher-
78
+ ically symmetric, static, asymptotically flat, with regular centres, and for which
79
+ the resulting Einstein tensor is physically reasonable, satisfying the weak energy
80
+ condition and having components which are bounded and fall off appropriately at
81
+ large distance. Dymnikova [25] obtained nonsingular Schwarzsclid black hole solu-
82
+ tion in vacuum and thereafter extended to obtain nonsingular cosmological black
83
+ hole [26] solutions to include the de Sitter solution in the usual four dimensions.
84
+ Formation and evaporation of non-singular black hole is also discussed [34] from
85
+ an initial vacuum region accommodating Bardeen-like static region supported by
86
+ finite density and pressures, subsequently its pressure vanishes rapidly at large ra-
87
+ dius which however behaves as a cosmological constant at a small radius. In 2019,
88
+
89
+ Nonsingular Black Holes in Higher dimensions
90
+ 3
91
+ Event Horizon Telescope group captured the first ever image of a supermassive
92
+ black hole at the centre of the M87 galaxy which triggers the various possibili-
93
+ ties for the state of the compact object and opened up new horizon in theoretical
94
+ physics. The motivation of the paper is to obtain a nonsingular black hole solution
95
+ in higher dimensions and investigate the different features of such black holes in
96
+ GR and beyond GR. For this we consider Higher dimensional Einstein gravity
97
+ (GR) and Rastall gravity for a comparative study.
98
+ The paper is organised as follows: In sec. 2, the Einstein field equation in a
99
+ static higher dimensional metric is obtained. In sec. 3, non-singular black holes are
100
+ obtained in Rastall theory with extension of spacetime dimensions. In sec. 4, we
101
+ present analytical set up of the non-singular black hole solution to investigate the
102
+ shadow of the black hole. The effective potential and the shadow behaviour of the
103
+ black holes are analyzed in sec. 5. Finally we summarize in sec 6.
104
+ 2 Einstein Field Equation in Higher Dimensions
105
+ We consider a higher dimensional gravitational action which is given by
106
+ I = βˆ’
107
+ 1
108
+ 16Ο€GD
109
+ οΏ½ βˆšβˆ’g dDx R + Im
110
+ (1)
111
+ where R is the Ricci scalar, GD is the D-dimensional gravitational constant and
112
+ Im represents the matter action. The Einstein field equation is given by
113
+ RAB βˆ’ 1
114
+ 2gABR = ΞΊ2TAB
115
+ (2)
116
+ where A, B = 0, 1, ...D βˆ’1 and T A
117
+ B = (βˆ’Ο, Pr, PβŠ₯, ...) the energy-momentum tensor,
118
+ ρ the energy density, Pr the radial pressure, PβŠ₯ transverse pressure, ΞΊ2 = 8Ο€GD
119
+ c2
120
+ .
121
+ We consider D dimensional spacetime metric given by
122
+ ds2 = βˆ’eΞ½dt2 + eΞ» dr2 + r2dΩ2
123
+ Dβˆ’2
124
+ (3)
125
+ where Ξ½ and Ξ» are functions of radial coordinate r and ΩDβˆ’2 is for unit sphere in
126
+ SDβˆ’2 dimensions. The components of the Einstein equations and the metric given
127
+ by eq. (26) are given by
128
+ T t
129
+ t = (D βˆ’ 2)
130
+ 2
131
+ οΏ½
132
+ eβˆ’Ξ»
133
+ οΏ½
134
+ D βˆ’ 3
135
+ r2
136
+ βˆ’ Ξ»β€²
137
+ r
138
+ οΏ½
139
+ βˆ’ D βˆ’ 3
140
+ r2
141
+ οΏ½
142
+ ,
143
+ (4)
144
+ T r
145
+ r = (D βˆ’ 2)
146
+ 2
147
+ οΏ½
148
+ eβˆ’Ξ»
149
+ οΏ½
150
+ D βˆ’ 3
151
+ r2
152
+ + Ξ½β€²
153
+ r
154
+ οΏ½
155
+ βˆ’ D βˆ’ 3
156
+ r2
157
+ οΏ½
158
+ ,
159
+ (5)
160
+ T ΞΈ1
161
+ ΞΈ1 = βˆ’(D βˆ’ 3)(D βˆ’ 4)
162
+ 2r2
163
+ + eβˆ’Ξ»
164
+ 2 Γ—
165
+ οΏ½
166
+ Ξ½β€²β€² + Ξ½β€²2
167
+ 2 βˆ’ Ξ»β€²Ξ½β€²
168
+ 2
169
+ + (D βˆ’ 3)(Ξ½β€² βˆ’ Ξ»β€²)
170
+ r
171
+ + (D βˆ’ 3)(D βˆ’ 4)
172
+ r2
173
+ οΏ½
174
+ ,
175
+ (6)
176
+ T ΞΈ2
177
+ ΞΈ2 = ...... = T ΞΈDβˆ’2
178
+ ΞΈDβˆ’2 = T ΞΈ1
179
+ ΞΈ1
180
+ (7)
181
+ for simplicity we have taken ΞΊ2 = 8Ο€GD
182
+ c2
183
+ = 1. The radial null vector lA can be
184
+ selected to have the components lt = eΞ»/2, lr = Β±eΞ½/2 and li = 0. The two
185
+
186
+ 4
187
+ Bikash Chandra Paul
188
+ radial null-null components of the Ricci tensor are equal, and given by RABlAlB =
189
+ eΞ»Rtt + eΞ½Rrr = (D βˆ’ 2) (eΞ»+Ξ½)β€²
190
+ 2eΞ»+Ξ½ , which vanishes if and only if (Ξ» + Ξ½) is a constant.
191
+ A rescaling of the time coordinate can be set to make the sum of the terms equal
192
+ to zero for black hole solution and we write
193
+ Ξ» + Ξ½ = 0
194
+ (8)
195
+ Now substituting
196
+ Ξ» = βˆ’ ln f(r)
197
+ (9)
198
+ we obtain the following components of energy momentum tensors in terms of f(r),
199
+ which are given by
200
+ T t
201
+ t = T r
202
+ r = D βˆ’ 2
203
+ 2
204
+ οΏ½
205
+ f(r)
206
+ οΏ½
207
+ D βˆ’ 3
208
+ r2
209
+ +
210
+ fβ€²
211
+ rf(r)
212
+ οΏ½
213
+ βˆ’ D βˆ’ 3
214
+ r2
215
+ οΏ½
216
+ (10)
217
+ T ΞΈ1
218
+ ΞΈ1 = f(r)
219
+ 2
220
+ οΏ½
221
+ fβ€²β€²
222
+ f + 2(D βˆ’ 3)fβ€²
223
+ rf(r)
224
+ + (D βˆ’ 3)(D βˆ’ 4)
225
+ r2
226
+ οΏ½
227
+ βˆ’ (D βˆ’ 3)(D βˆ’ 4)
228
+ 2r2
229
+ (11)
230
+ T ΞΈ2
231
+ ΞΈ2 = ...... = T ΞΈDβˆ’2
232
+ ΞΈDβˆ’2 = T ΞΈ1
233
+ ΞΈ1
234
+ (12)
235
+ where the prime denotes the derivative with respect to r. The source term satis-
236
+ fying
237
+ T t
238
+ t = T r
239
+ r ,
240
+ and
241
+ T ΞΈ2
242
+ ΞΈ2 = ...... = T ΞΈDβˆ’2
243
+ ΞΈDβˆ’2 = T ΞΈ1
244
+ ΞΈ1
245
+ (13)
246
+ and the equation of state, T A
247
+ B ;A = 0,.
248
+ Assume the density profile in higher dimensions T t
249
+ t = βˆ’Ο as
250
+ ρ = βˆ’T t
251
+ t = ρ0 e
252
+ βˆ’ rDβˆ’1
253
+ rDβˆ’1
254
+ βˆ—
255
+ (14)
256
+ where rβˆ— is a dimensional constant connected with a constant density ρ0. The
257
+ density ρ0 also permits a D dimensional de Sitter solution with its size given by
258
+ r2
259
+ 0 = (D βˆ’ 1)(D βˆ’ 2)
260
+ 2Ξ›
261
+ ,
262
+ (15)
263
+ where Ξ› = ρ0. Using the density profile given eq. (14) in eq. (10) we integrate and
264
+ obtain the metric potential which yields
265
+ f(r) = 1 βˆ’ rDβˆ’3
266
+ g
267
+ rDβˆ’3 +
268
+ 2ρ0rDβˆ’1
269
+ βˆ—
270
+ (D βˆ’ 1)(D βˆ’ 2)
271
+ 1
272
+ rDβˆ’3 e
273
+ βˆ’ rDβˆ’1
274
+ rDβˆ’1
275
+ βˆ—
276
+ (16)
277
+ where rDβˆ’3
278
+ g
279
+ =
280
+ οΏ½
281
+ 2ρ0
282
+ (Dβˆ’1)(Dβˆ’2)
283
+ οΏ½
284
+ rDβˆ’1
285
+ βˆ—
286
+ . The higher dimensional metric is now can be
287
+ written as
288
+ ds2 = βˆ’
289
+ οΏ½
290
+ 1 βˆ’ Rs(r)
291
+ rDβˆ’3
292
+ οΏ½
293
+ dt2 +
294
+ dr2
295
+ οΏ½
296
+ 1 βˆ’ Rs(r)
297
+ rDβˆ’3
298
+ οΏ½ + r2dΩ2
299
+ Dβˆ’2
300
+ (17)
301
+ where we denote
302
+ Rs(r) = rDβˆ’3
303
+ g
304
+ οΏ½
305
+ 1 βˆ’ exp
306
+ οΏ½
307
+ βˆ’rDβˆ’1
308
+ rDβˆ’1
309
+ βˆ—
310
+ οΏ½οΏ½
311
+ (18)
312
+
313
+ Nonsingular Black Holes in Higher dimensions
314
+ 5
315
+ and
316
+ rDβˆ’1
317
+ βˆ—
318
+ = r2
319
+ 0 rDβˆ’3
320
+ g
321
+ ,
322
+ (19)
323
+ where r2
324
+ 0 = (Dβˆ’1)(Dβˆ’2)
325
+ 2ρ0
326
+ . This is an exact spherically symmetric solution of the
327
+ Einstein field equations in D-dimensions. For D = 4 the solution given by eq.
328
+ (17) reduces to the solution obtained by Dymnikova [25]. The other components
329
+ of energy momentum tensor can be obtained using the Einstein’s field equations
330
+ which are given by
331
+ T ΞΈ2
332
+ ΞΈ2 = ... = T ΞΈDβˆ’2
333
+ ΞΈDβˆ’2 =
334
+ οΏ½
335
+ D βˆ’ 1
336
+ D βˆ’ 2
337
+ οΏ½ r
338
+ rβˆ—
339
+ οΏ½Dβˆ’1
340
+ βˆ’ 1
341
+ οΏ½
342
+ ρ0e
343
+ βˆ’ rDβˆ’1
344
+ rDβˆ’1
345
+ βˆ—
346
+ .
347
+ (20)
348
+ It is evident that in the usual 4 dimensions Dymnikova [25] black hole solutions
349
+ recovered with anisotropic fluid distribution when r = rβˆ—, which is true also in
350
+ higher dimensions. The nonsingular black hole (NSBH) solutions are permitted
351
+ with anisotropic fluid distributions in higher dimensions. The energy density and
352
+ radial pressure follow the vacuum configuration but the tangential pressures do
353
+ not. The tangential pressure is non-zero which remains positive definite for r > rβˆ—.
354
+ At the center the tangential pressure is negative indicating existence of exotic mat-
355
+ ter (PβŠ₯ < 0) at the center of the black hole. The nonsingular black hole solution
356
+ obtained by Dymnikova can not be described in lower dimension D = 2 + 1, how-
357
+ ever, we can extend the concept of NSBH in more than the usual four dimensions.
358
+ The generalization of the black hole solution in higher dimensions accommodates
359
+ a new class of NSBH solutions where the tangential pressure increases to a large
360
+ extent inside the non-singular black hole with a different feature but away from
361
+ the centre of the black hole it decreases exponentially.
362
+ The mass of a massive object in higher dimension is given by
363
+ m(r) = ADβˆ’2
364
+ οΏ½ r
365
+ 0
366
+ rβ€²Dβˆ’2ρ(rβ€²)drβ€²
367
+ (21)
368
+ where ADβˆ’2 =
369
+ 2Ο€
370
+ Dβˆ’1
371
+ 2
372
+ Ξ“ ( Dβˆ’1
373
+ 2
374
+ ) which at r β†’ ∞ is connected to the whole mass M con-
375
+ nected with rDβˆ’3
376
+ g
377
+ by the Schwarzschild relation. The modulus difference between
378
+ Rs(rg) and rDβˆ’3
379
+ g
380
+ is rDβˆ’3
381
+ g
382
+ e
383
+ βˆ’ rDβˆ’1
384
+ rDβˆ’1
385
+ βˆ—
386
+ . The measure of the difference between the higher
387
+ dimensional Schwarzschild mass m(r) ∼ rDβˆ’3
388
+ g
389
+ in a singular black hole and Rs of a
390
+ non-singular black hole is given by
391
+ M βˆ’ m(r)
392
+ M
393
+ = exp
394
+ οΏ½
395
+ βˆ’rDβˆ’1
396
+ rDβˆ’1
397
+ βˆ—
398
+ οΏ½
399
+ .
400
+ (22)
401
+ Here m(r) becomes M at infinite distance. It is found that the mass difference
402
+ decreases as the dimension in which black hole embedded increases. The metric
403
+ has two event horizons located at
404
+ r+ = rg
405
+ οΏ½
406
+ 1 βˆ’ O
407
+ οΏ½
408
+ exp
409
+ οΏ½
410
+ βˆ’r2g
411
+ r2
412
+ 0
413
+ οΏ½οΏ½οΏ½
414
+ ,
415
+ rβˆ’ = r0
416
+ οΏ½
417
+ 1 βˆ’ O
418
+ οΏ½
419
+ exp
420
+ οΏ½
421
+ βˆ’r0
422
+ rg
423
+ οΏ½οΏ½οΏ½
424
+ .
425
+ (23)
426
+
427
+ 6
428
+ Bikash Chandra Paul
429
+ Here r+ is the external event horizon. The metric evaluated at gtt(r+) = 0 de-
430
+ scribes an object with the similar properties properties of a black hole by a dis-
431
+ tant observer, it does not send light signals outside and could not interact with
432
+ its surroundings by the gravitational field. In four dimensions it is found that
433
+ both r+ and rβˆ’ are removable singularities of the metric. The singularities can be
434
+ eliminated by an appropriate transformation.
435
+ 3 Higher Dimensional Rastall gravity
436
+ In this section we explore NSBH solution in the Rastall theory of gravity for D β‰₯ 4
437
+ dimensions. The Rastall theory [20] is based on the modification of the Einstein
438
+ field equation for a spacetime with Ricci scalar filled by an energy momentum
439
+ source as follows:
440
+ T AB; A = Ξ»RB
441
+ (24)
442
+ where Ξ» is the Rastall parameter which is a measure for deviation from the stan-
443
+ dard GR conservation law. Consequently the Rastall field equation can be written
444
+ as
445
+ GAB + ΞΊ2Ξ»gABR = ΞΊ2TAB
446
+ (25)
447
+ where κ2 is the Rastall gravitational constant. The above field equation reduces
448
+ to that of GR in the limit Ξ» β†’ 0 and ΞΊ2 = 8Ο€G. However, for a vanishing trace
449
+ of the energy-momentum tensor, for example the electrovacuum solution can be
450
+ obtained when Ξ» = 1
451
+ 4 or R = 0. It is important to note that the former possibility
452
+ is not physically acceptable as the trace of the energy momentum tensor vanishes
453
+ T = 0 for any scalar field. Consequently the matter configuration where the energy-
454
+ momentum tensor has null trace, the relativistic solution obtained in Rastall theory
455
+ is same as that one obtains in the general theory of relativity (GR). This feature of
456
+ Rastall theory which is a modified GR led us to look for black holes solutions in a
457
+ background of matter/energy with non-vanishing trace. It may be pointed out here
458
+ that the Rastall gravity is widely used to accommodate acceptable explanation for
459
+ the current acceleration of the universe which has no solution in GR and for this
460
+ it is interesting to explore NSBH in Rastall theory.
461
+ We consider the metric for black hole solution in higher dimensions D β‰₯ 4:
462
+ ds2 = βˆ’f(r)dt2 + dr2
463
+ f(r) + r2dΩ2
464
+ Dβˆ’2.
465
+ (26)
466
+ Using the metric, we obtain the non-vanishing components of the Rastall tensor
467
+ HAB = GAB + Ξ»gABR and ΞΊ2 = 1,
468
+ Ht
469
+ t = D βˆ’ 2
470
+ 2r2
471
+ οΏ½
472
+ rfβ€² βˆ’ (D βˆ’ 3) + (D βˆ’ 3)fοΏ½ + Ξ»R,
473
+ (27)
474
+ Hr
475
+ r = D βˆ’ 2
476
+ 2r2
477
+ οΏ½
478
+ rfβ€² βˆ’ (D βˆ’ 3) + (D βˆ’ 3)fοΏ½ + Ξ»R,
479
+ (28)
480
+ HΞΈi
481
+ ΞΈi = r2fβ€²β€² + (D βˆ’ 3)(2rfβ€² + (D βˆ’ 4)(f βˆ’ 1)
482
+ 2r2
483
+ + Ξ»R
484
+ (29)
485
+ where i = 1, 2, ..., (D βˆ’ 2), and the Ricci scalar in D dimensions is given by
486
+ R = βˆ’ 1
487
+ r2
488
+ οΏ½
489
+ r2fβ€²β€² + 2(D βˆ’ 2)rfβ€² + (D βˆ’ 2)(D βˆ’ 3)(f βˆ’ 1)
490
+ οΏ½
491
+ (30)
492
+
493
+ Nonsingular Black Holes in Higher dimensions
494
+ 7
495
+ in the above we denote ()β€² to represent derivative with respect to the radial co-
496
+ ordinate r. We solve the field equation to obtain higher dimensional non-singular
497
+ black holes in Rastall theory and for this Ht
498
+ t = T t
499
+ t and Hrr = T rr yield
500
+ Pr = D βˆ’ 2
501
+ 2r2
502
+ οΏ½
503
+ rfβ€² βˆ’ (D βˆ’ 3) + (D βˆ’ 3)fοΏ½
504
+ βˆ’ Ξ»
505
+ r2
506
+ οΏ½
507
+ r2fβ€²β€² + 2(D βˆ’ 2)rfβ€² + (D βˆ’ 2)(D βˆ’ 3)(f βˆ’ 1)
508
+ οΏ½
509
+ ,
510
+ (31)
511
+ and also we consider HΞΈ1
512
+ ΞΈ1 = T ΞΈ1
513
+ ΞΈ1 , ... and HΞΈDβˆ’2
514
+ ΞΈDβˆ’2 = T ΞΈDβˆ’2
515
+ ΞΈDβˆ’2 which yield
516
+ PβŠ₯ =
517
+ 1
518
+ 2r2
519
+ οΏ½
520
+ r2fβ€²β€² + 2(D βˆ’ 3)rfβ€² + (D βˆ’ 3)(D βˆ’ 4)(f βˆ’ 1)
521
+ οΏ½
522
+ βˆ’ Ξ»
523
+ r2
524
+ οΏ½
525
+ r2fβ€²β€² + 2(D βˆ’ 2)rfβ€² + (D βˆ’ 2)(D βˆ’ 3)(f βˆ’ 1)
526
+ οΏ½
527
+ .
528
+ (32)
529
+ In this case we explore the non-singular Black hole obtained in higher dimensional
530
+ Rastall gravity, the general solution of the metric is
531
+ f(r) = 1 βˆ’ rDβˆ’3
532
+ g
533
+ rDβˆ’3 +
534
+ 2ρ0rDβˆ’1
535
+ βˆ—
536
+ (D βˆ’ 1)(D βˆ’ 2)
537
+ 1
538
+ rDβˆ’3 e
539
+ βˆ’ rDβˆ’1
540
+ rDβˆ’1
541
+ βˆ—
542
+ (33)
543
+ The energy density and radial pressure are
544
+ ρ =
545
+ 
546
+ ο£­
547
+ D βˆ’ 2 βˆ’ 2Ξ»D + 2(D βˆ’ 1)Ξ» rDβˆ’1
548
+ rDβˆ’1
549
+ βˆ—
550
+ D βˆ’ 2
551
+ ο£Ά
552
+  ρ0e
553
+ βˆ’ rDβˆ’1
554
+ rDβˆ’1
555
+ βˆ—
556
+ ,
557
+ (34)
558
+ Pr = βˆ’
559
+ 
560
+ ο£­
561
+ D βˆ’ 2 βˆ’ 2Ξ»D + 2(D βˆ’ 1)Ξ» rDβˆ’1
562
+ rDβˆ’1
563
+ βˆ—
564
+ D βˆ’ 2
565
+ ο£Ά
566
+  ρ0e
567
+ βˆ’ rDβˆ’1
568
+ rDβˆ’1
569
+ βˆ—
570
+ ,
571
+ (35)
572
+ the tangential pressure is given by
573
+ PβŠ₯ =
574
+ οΏ½
575
+ (1 βˆ’ 2Ξ»)D βˆ’ 1
576
+ D βˆ’ 2
577
+ rDβˆ’1
578
+ rDβˆ’1
579
+ βˆ—
580
+ βˆ’ D βˆ’ 2 βˆ’ 2Ξ»D
581
+ D βˆ’ 2
582
+ οΏ½
583
+ ρ0e
584
+ βˆ’ rDβˆ’1
585
+ rDβˆ’1
586
+ βˆ—
587
+ .
588
+ (36)
589
+ The energy density and the transverse pressure in Rastall gravity framework ob-
590
+ tained in eqs. (34) and (36) reduces to the eqs. (14) and (20) in GR for Ξ» β†’ 0.
591
+ The modification introduced in GR by Rastall admits nonsingular Dymnikova
592
+ [25] black hole (NSBH) with normal matter while the radial pressure corresponds
593
+ to vacuum equation of state. At the center of the NSBH the energy density is
594
+ ρ = (D βˆ’ 2 βˆ’ 2Ξ»D)ρ0, which increases as the. number of spacetime dimension in-
595
+ creases for a given range of Rastall parameter Ξ» < Dβˆ’2
596
+ 2D . It is evident that for a
597
+ given dimension, NSBH admits greater mass for lower values of Ξ» and the lower
598
+ limiting value for Ξ» < Dβˆ’2
599
+ 2D
600
+ and |Ξ»| > Dβˆ’2
601
+ 2D
602
+ (for negative Ξ»). The corresponding
603
+ tangential pressure at the center PβŠ₯ = (2DΞ» + 2 βˆ’ D)ρ0 is finite but negative. In
604
+ D = 4 dimensions, at the center of the black hole, ρ(r = 0) = 2(1 βˆ’ 4Ξ»)ρ0 and
605
+ tangential pressure PβŠ₯ = βˆ’(1βˆ’4Ξ»)ρ0 which indicates existence of exotic matter at
606
+ the centre in GR (as Ξ» = 0) as well as in Rastall theory for Ξ» > βˆ’ 1
607
+ 4. Thus NSBH
608
+ can be realized with both central radial pressure and tangential pressure negative
609
+ and equal but an anisotropy in pressure develops away from the center in Rastall
610
+
611
+ 8
612
+ Bikash Chandra Paul
613
+ gravity, normal matter exists when r >
614
+ οΏ½
615
+ Dβˆ’2βˆ’2Ξ»D
616
+ (Dβˆ’1)(1βˆ’2Ξ»
617
+ οΏ½1/(Dβˆ’1)
618
+ rβˆ—. The tangential
619
+ pressure indicates black hole surrounded by exotic matter in Rastall gravity [35]
620
+ for the range 1
621
+ 4 < Ξ» < 1
622
+ 2.. For r β†’ ∞, the energy density and pressure vanishes
623
+ asymptotically.
624
+ When Ξ» = Dβˆ’2
625
+ 2D , we get the following :
626
+ ρ = ρ0
627
+ οΏ½
628
+ D βˆ’ 1
629
+ D
630
+ rDβˆ’1
631
+ rDβˆ’1
632
+ βˆ—
633
+ οΏ½
634
+ e
635
+ βˆ’ rDβˆ’1
636
+ rDβˆ’1
637
+ βˆ—
638
+ ,
639
+ (37)
640
+ Pr = βˆ’Ο = ρ0
641
+ οΏ½
642
+ D βˆ’ 1
643
+ D
644
+ rDβˆ’1
645
+ rDβˆ’1
646
+ βˆ—
647
+ οΏ½
648
+ e
649
+ βˆ’ rDβˆ’1
650
+ rDβˆ’1
651
+ βˆ—
652
+ ,
653
+ (38)
654
+ the tangential pressure is given by
655
+ PβŠ₯ = 2ρ0
656
+ οΏ½
657
+ D βˆ’ 1
658
+ D(D βˆ’ 2)
659
+ rDβˆ’1
660
+ rDβˆ’1
661
+ βˆ—
662
+ οΏ½
663
+ e
664
+ βˆ’ rDβˆ’1
665
+ rDβˆ’1
666
+ βˆ—
667
+ .
668
+ (39)
669
+ one obtains NSBH with ρ > 0, ρ + Pr = 0 and PβŠ₯ > 0. For D = 4 dimensions,
670
+ Ξ» = 1
671
+ 4 and the NSBH can be realized in Rastall gravity with normal matter which
672
+ however is not permitted in GR. Also we note that at the centre of the NSBH the
673
+ tangential pressure vanishes, admitting a perfect vacuum NSBH in the usual four
674
+ dimensions. The result obtained in this case is also applicable in higher dimensions.
675
+ This is a new result.
676
+ 4 Analytical set up
677
+ The modified Schwarzschild metric for a non-singular black hole is given by
678
+ ds2 = βˆ’f(r) dt2 + f(r)βˆ’1 dr2 + r2dΩ2
679
+ Dβˆ’2
680
+ (40)
681
+ where f(r) = 1 βˆ’ οΏ½ rg
682
+ r
683
+ οΏ½Dβˆ’3 + οΏ½ rg
684
+ r
685
+ οΏ½Dβˆ’3 exp
686
+ οΏ½
687
+ βˆ’ οΏ½ r
688
+ rβˆ—
689
+ οΏ½Dβˆ’1οΏ½
690
+ , making use of the as-
691
+ sumption ΞΊ2 = 8Ο€ made earlier, we write rg =
692
+ οΏ½
693
+ 16Ο€M
694
+ (Dβˆ’2)ADβˆ’2
695
+ οΏ½
696
+ 1
697
+ Dβˆ’3 and the area
698
+ of D dimensional sphere ADβˆ’2 =
699
+ 2Ο€
700
+ Dβˆ’1
701
+ 2
702
+ Ξ“( Dβˆ’1
703
+ 2 ), where M represents the mass of the
704
+ non-singular Black hole. The metric function gtt = f(r), whose sign determines
705
+ gravitational trapping [34], we plot to draw a sketch to study the existence of
706
+ black hole solutions. The metric potential f(r) is plotted with r in Fig. (1) for
707
+ D = 4 and Fig. (2) for D = 10. It is evident that both the extreme black hole
708
+ and non-extreme black holes can be obtained for a given set of values of rg and
709
+ r0. We note that extreme black hole exists for rg = 1.0 and r0 = 0.57 in D = 4
710
+ and rg = 2.0 and r0 = 1.57 in D = 10. In the first case no black hole exist for
711
+ rg < 1.0 and the later case for r0 > 1.57. The photon radii are tabulated in Table-I
712
+ for D = 4 and Table-II for D = 10. It is found that for D = 4, it increases with
713
+ decrease of ρ0 ∼ 1/r2
714
+ 0 for a given mass but for a given ρ0, photon radius is found
715
+ to increase with mass. In D = 10 dimensions as ρ0 is decreases the photon radius
716
+ decreases then increases and decreases once again. In Fig (3) dimensional varia-
717
+ tion of the photon radius for M = 1 is plotted for non-singular black holes with
718
+
719
+ Nonsingular Black Holes in Higher dimensions
720
+ 9
721
+ 1
722
+ 2
723
+ 3
724
+ 4
725
+ 5
726
+ 6
727
+ r
728
+ οΏ½1.0
729
+ οΏ½0.5
730
+ 0.5
731
+ 1.0
732
+ fοΏ½rοΏ½
733
+ Fig. 1 Radial variation of f(r) for rg = 0.8 (Red), 1.0 (Black), 1.5 (Blue) for r0 = 0.57 in
734
+ D = 4.
735
+ 1
736
+ 2
737
+ 3
738
+ 4
739
+ 5
740
+ 6
741
+ r
742
+ οΏ½1.0
743
+ οΏ½0.5
744
+ 0.5
745
+ 1.0
746
+ fοΏ½rοΏ½
747
+ Fig. 2 Radial variation of the metric function f(r) in D = 10 for r0 = 1.1 (Blue), 1.57 (Black)
748
+ and 2.0 (Red) with rg = 2.
749
+ dimensions. The photon radius is maximum at D = 4 and then decreases sharply
750
+ as the dimensions is increases and remains constant.
751
+ The Lagrangian is given by
752
+ L = 1
753
+ 2gAB Λ™xA Λ™xB.
754
+ (41)
755
+ 2
756
+ 4
757
+ 6
758
+ 8
759
+ 10
760
+ οΏ½1
761
+ 0
762
+ 1
763
+ 2
764
+ 3
765
+ r
766
+ V
767
+ Fig. 3 Radial variation of the potential for J= 5 (Blue), 6 (Green), 8 (Dashed), 10 (Red) in
768
+ D = 4 for non-singular BH.
769
+
770
+ 10
771
+ Bikash Chandra Paul
772
+ where Λ™() =
773
+ d
774
+ dΟ„ and Ο„ is the affine parameter. Expanding eq. (41) we get
775
+ 2L = βˆ’f(r)Λ™t2 +
776
+ 1
777
+ f(r) Λ™r2 + (r2 Λ™ΞΈ1
778
+ 2 + sin2ΞΈ1 Λ™ΞΈ2
779
+ 2 + ......)
780
+ (42)
781
+ To obtain trajectory of light path, we set ΞΈi = Ο€
782
+ 2 where i = 1, ..., D βˆ’ 3 and ΞΈDβˆ’2
783
+ is a free parameter. The momenta are given by
784
+ Pt = βˆ‚L
785
+ βˆ‚ Λ™t = βˆ’f(r)Λ™t,
786
+ Pr = βˆ‚L
787
+ βˆ‚ Λ™r =
788
+ 1
789
+ f(r) Λ™r,
790
+ PΞΈ1 = βˆ‚L
791
+ βˆ‚ Λ™ΞΈ1
792
+ = r2 Λ™ΞΈ1, PΞΈ2 = βˆ‚L
793
+ βˆ‚ Λ™ΞΈ2
794
+ = r2sin2ΞΈ1 Λ™ΞΈ2, ....
795
+ (43)
796
+ Now as defined above, ΞΈi = Ο€
797
+ 2 , and at the equatorial plane ΞΈ1 = Ο€
798
+ 2 ,
799
+ βˆ‚L
800
+ βˆ‚ Λ™t = constant
801
+ (44)
802
+ and we determine the energy (E) and angular momentum (J) at r β†’ ∞ as
803
+ f(r)Λ™t = E, PΞΈDβˆ’2 = r2
804
+ Λ™
805
+ ΞΈDβˆ’2 = J.
806
+ (45)
807
+ The Hamilton Jacobi equation is the most general method to find the geodesic
808
+ equation of motion around black hole or a compact object, we adopt the technique
809
+ to obtain the photon orbits. In higher dimensions we get
810
+ βˆ‚S
811
+ βˆ‚Ο„ = H = βˆ’1
812
+ 2gAB βˆ‚S
813
+ βˆ‚xA
814
+ βˆ‚S
815
+ βˆ‚xB
816
+ (46)
817
+ where gAB is the inverse of the metric and S is the Jacobian. The Jacobian is
818
+ given by
819
+ S = 1
820
+ 2m2Ο„ βˆ’ E + JΞΈDβˆ’2 + Sr(r) +
821
+ Dβˆ’3
822
+ οΏ½
823
+ i=1
824
+ SΞΈi(ΞΈi)
825
+ (47)
826
+ where Sr(r) and SΞΈi(ΞΈi) are functions of r and ΞΈi respectively and m is the mass
827
+ of the test particle, it is zero for photon. The Hamilton-Jacobi eq. (46) can be
828
+ written as
829
+ r4
830
+ οΏ½
831
+ 1 βˆ’
832
+ Rs
833
+ rDβˆ’3
834
+ οΏ½2 οΏ½
835
+ βˆ‚S
836
+ βˆ‚Ο„
837
+ οΏ½
838
+ = E2r4 βˆ’ r2
839
+ οΏ½
840
+ 1 βˆ’
841
+ Rs
842
+ rDβˆ’3
843
+ οΏ½
844
+ (K + J2)
845
+ (48)
846
+ Dβˆ’3
847
+ οΏ½
848
+ i=1
849
+ 1
850
+ Ξ iβˆ’1
851
+ n=1sin2ΞΈn
852
+ οΏ½βˆ‚SΞΈi
853
+ βˆ‚ΞΈi
854
+ οΏ½2
855
+ = K βˆ’ Ξ Dβˆ’3
856
+ i=1 J2cot2ΞΈi
857
+ (49)
858
+ where K is the Carter constant [36]. Using the above eq. (43) in eq. (46) we get
859
+ the following
860
+ Λ™t =
861
+ E
862
+ f(r),
863
+ Λ™ΞΈDβˆ’2 =
864
+ J
865
+ r2Ξ Dβˆ’3
866
+ i=1 sin2ΞΈi
867
+ ;
868
+ r2 Λ™r = Β±
869
+ √
870
+ R,
871
+ r2
872
+ Dβˆ’3
873
+ οΏ½
874
+ i=1
875
+ Ξ iβˆ’1
876
+ n=1sin2ΞΈn Λ™ΞΈi = Β±
877
+ οΏ½
878
+ Θi
879
+ (50)
880
+
881
+ Nonsingular Black Holes in Higher dimensions
882
+ 11
883
+ in the above ”+” and ”-” sign corresponds to motion of photon in outgoing and
884
+ incoming radial direction and over dot represents derivative w.r.t to the affine
885
+ parameterΟ„. For the null curves the eqs.(49) can be expressed as
886
+ R(r) = E2r4 βˆ’ r2f(r)(K2 + J2),
887
+ (51)
888
+ Θi(ΞΈi) = K βˆ’ Ξ Dβˆ’3
889
+ i=1 J2cot2ΞΈi.
890
+ (52)
891
+ The characteristics of photon near the black hole can be defined by two impact
892
+ parameters, which are functions of the constants E, J and K. For general orbit we
893
+ define the impact parameters ξ = J
894
+ E and Ξ· =
895
+ K
896
+ E2 . The boundary of the shadow of
897
+ a black hole can be estimated from the effective potential. The radial null geodesic
898
+ from eqs. (48) and (50) is given by
899
+ οΏ½
900
+ dr
901
+ dΟ„
902
+ οΏ½2
903
+ + Veff = 0,
904
+ (53)
905
+ where Veff is the effective potential, for radial motion we obtain
906
+ Veff = f(r)
907
+ r2 (K + J2) βˆ’ E2
908
+ = 1
909
+ r2
910
+ οΏ½
911
+ 1 βˆ’
912
+ οΏ½rg
913
+ r
914
+ οΏ½Dβˆ’3 οΏ½
915
+ 1 βˆ’ eβˆ’( r
916
+ rβˆ— )
917
+ Dβˆ’1οΏ½οΏ½
918
+ (K + J2) βˆ’ E2.
919
+ (54)
920
+ The effective potential is identical to the classical equation describing the motion
921
+ of a massless particle in a 1-dimensional potential V (r) provided its energy is
922
+ 1
923
+ 2E2 (of course the true energy should be E), but we use this form to obtain an
924
+ expression for potential in our study. We plot radial variation of V (r) in Fig. (4) in
925
+ a four dimensional universe for singular as well as non-singular black hole. As the
926
+ angular velocity increases the photons heading towards the black hole are unstable.
927
+ In Fig. 2, it is found that there is no difference of the behaviour of the potential.
928
+ In Fig. (4) we plot radial variation of V (r) for different angular momentum, for a
929
+ non-singular BH, it is evident that as the angular momentum increases the photon
930
+ can approach near to the BH unbounded
931
+ The photon orbits are circular and unstable for a maximum value of the effec-
932
+ tive potential. The unstable circular orbit determines the boundary of the apparent
933
+ shape and can be maximized. The maximal value of the effective potential corre-
934
+ sponds to the circular orbits and the unstable photons satisfies
935
+ Veff
936
+ οΏ½οΏ½οΏ½
937
+ r=rp
938
+ = dVeff
939
+ dr
940
+ οΏ½οΏ½οΏ½
941
+ r=rp
942
+ = 0,
943
+ R(r) = dR(r)
944
+ dr
945
+ οΏ½οΏ½οΏ½
946
+ r=rc
947
+ = 0
948
+ (55)
949
+ The impact parameters are now related as Using eqs. (54) and (55), we get
950
+ f(rp)
951
+ r2p
952
+ (K + J2) βˆ’ E2 = 0
953
+ rpfβ€²(rp) βˆ’ 2f(rp)
954
+ r3p
955
+ (K + J2) = 0.
956
+ (56)
957
+ In four dimensions the potential V (r) is plotted in Fig. (4) with different angular
958
+ momentum (J) for rg = 2 and E = 1. The particles are bounded for a radius
959
+ r < rmin and unbounded for the range rmin < r < rmax. The range of values
960
+
961
+ 12
962
+ Bikash Chandra Paul
963
+ rp in
964
+ rp in
965
+ rp in
966
+ r0
967
+ M = 2
968
+ M = 5
969
+ M = 10
970
+ 0.5
971
+ 0.8757
972
+ 1.0192
973
+ 1.1532
974
+ 0.6
975
+ 1.0221
976
+ 1.1851
977
+ 1.3389
978
+ 0.8
979
+ 1.3079
980
+ 1.5053
981
+ 1.6958
982
+ 1.0
983
+ 1.5880
984
+ 1.8142
985
+ 2.0383
986
+ 1.2
987
+ 6.0000
988
+ 2.1150
989
+ 2.3702
990
+ 1.5
991
+ 6.0000
992
+ -
993
+ -
994
+ 1.8
995
+ 5.990
996
+ -
997
+ -
998
+ 2
999
+ 2.9921
1000
+ -
1001
+ -
1002
+ Table 1 The variation of the photon radius (rp) in D = 4 with r0 =
1003
+ οΏ½
1004
+ (Dβˆ’1)(Dβˆ’2)
1005
+ 4ρ0
1006
+ and the
1007
+ mass of the BH.
1008
+ rp in
1009
+ rp in
1010
+ r0
1011
+ M = 5
1012
+ M = 10
1013
+ 0.5
1014
+ 0.9369
1015
+ 1.0746
1016
+ 0.6
1017
+ 1.0035
1018
+ 1.0002
1019
+ 0.7
1020
+ 1.0641
1021
+ 0.9863
1022
+ 0.8
1023
+ 1.1203
1024
+ 1.0613
1025
+ 0.84
1026
+ 1.1417
1027
+ 24.6008
1028
+ 0.85
1029
+ 1.1470
1030
+ 0.9472
1031
+ 0.89
1032
+ 1.178
1033
+ 1.5069
1034
+ 0.9
1035
+ 1.1730
1036
+ 25.2259
1037
+ 0.91
1038
+ 1.1781
1039
+ 0.9598
1040
+ 0.95
1041
+ 1.1983
1042
+ 1.1687
1043
+ 1.0
1044
+ 1.2229
1045
+ 0.9772
1046
+ 1.2
1047
+ 1.8426
1048
+ 1.0105
1049
+ 2
1050
+ 3.5942
1051
+ 6.9106
1052
+ Table 2 The variation of the photon radius (rp) in D = 10 with r0 =
1053
+ οΏ½
1054
+ (Dβˆ’1)(Dβˆ’2)
1055
+ 4ρ0
1056
+ and the
1057
+ mass of the BH.
1058
+ 3
1059
+ 4
1060
+ 5
1061
+ 6
1062
+ 7
1063
+ 8
1064
+ 9
1065
+ 10
1066
+ D
1067
+ 2
1068
+ 4
1069
+ 6
1070
+ 8
1071
+ 10
1072
+ Rp
1073
+ [t]
1074
+ Fig. 4 Dimensional variation of the photon radius for M = 1 for a non-singular BH
1075
+ can be determined from the sketch. It is found that rmin decreases as angular
1076
+ momentum (J) increases. We note that the potentials for Schwarzschild black hole
1077
+ (singular) and that for non-singular black holes overlaps for a set of similar values
1078
+ of D, J and E.
1079
+ We draw the shadow contour of non-singular black hole in Fig. (8) for a given
1080
+ value of ρ0 (say, 0.16 unit) in all dimensions. It is shown that as the spacetime
1081
+ dimensions increases the radius of the shadow decreases.
1082
+
1083
+ Nonsingular Black Holes in Higher dimensions
1084
+ 13
1085
+ οΏ½10
1086
+ οΏ½5
1087
+ 0
1088
+ 5
1089
+ 10
1090
+ οΏ½10
1091
+ οΏ½5
1092
+ 0
1093
+ 5
1094
+ 10
1095
+ Fig. 5 Contour plot for an object having M = 2MβŠ™ with ρ0 = 0.16 unit for D = 4 (Red),
1096
+ D = 5 (Dashed) and D = 6 (Green), D = 8 (Thick)
1097
+ οΏ½40
1098
+ οΏ½20
1099
+ 0
1100
+ 20
1101
+ 40
1102
+ οΏ½40
1103
+ οΏ½20
1104
+ 0
1105
+ 20
1106
+ 40
1107
+ Fig. 6 Contour plot for an object for ρ0 = 0.04 unit in D = 4 with M = 2MβŠ™ (Black),
1108
+ M = 4MβŠ™ (Green), M = 6MβŠ™ (Red), M = 10MβŠ™ (Blue)
1109
+ 5 Effective potential and shadow behaviour
1110
+ The effective potential of the Schwarzschild-Tangherlini black holes exhibits a max-
1111
+ imum for the photon sphere radius rp corresponding to the real and the positive
1112
+ solution of the constraint obtained from eq. (56),
1113
+ rpfβ€²(rp) βˆ’ 2f(rp) = 0.
1114
+ (57)
1115
+ Defining impact parameters η and ξ that are functions of the energy E, angular
1116
+ momentum J and the Carter constant K as
1117
+ ΞΎ = J
1118
+ E ,
1119
+ Ξ· = K
1120
+ E2
1121
+ (58)
1122
+ we get from eq. (55) corresponding to Veff
1123
+ E2
1124
+ = 0 and
1125
+ R
1126
+ E2 = 0, the following
1127
+ Ξ· + ΞΎ2 =
1128
+ r2p
1129
+ f(rp), Ξ· + ΞΎ2 =
1130
+ 4r2p
1131
+ rfβ€²(rp) + 2f(rp).
1132
+ (59)
1133
+
1134
+ 14
1135
+ Bikash Chandra Paul
1136
+ Now we obtain
1137
+ Ξ· + ΞΎ2 =
1138
+ 5r2p
1139
+ rpfβ€²(rp) + 3f(rp),
1140
+ (60)
1141
+ where the right hand side corresponds to
1142
+ r2
1143
+ p
1144
+ f(rp), the observer’s frame the shadow
1145
+ can be described properly making use of the celestial coordinates Ξ± and Ξ² as
1146
+ introduced earlier [37]. Following the definition introduced by Subrahmanyan as
1147
+ follows
1148
+ Ξ± =
1149
+ lim
1150
+ rpβ†’βˆž
1151
+ οΏ½
1152
+ rpP ΞΈDβˆ’2
1153
+ P t
1154
+ οΏ½
1155
+ ,
1156
+ Ξ²i =
1157
+ lim
1158
+ rpβ†’βˆž
1159
+ οΏ½
1160
+ rpP ΞΈi
1161
+ P t
1162
+ οΏ½
1163
+ ,
1164
+ (61)
1165
+ where
1166
+ i = 1, ...(D βˆ’ 3).
1167
+ For an observer on the equatorial plane, these equations reduced to
1168
+ Ξ· + ΞΎ2 = Ξ±2 + Ξ²2 =
1169
+ r2p
1170
+ f(rp)
1171
+ (62)
1172
+ the radius of the shadow is Rbhs =
1173
+ rp
1174
+ √
1175
+ f(rp). The form of f(r) is complex and
1176
+ therefore we study numerically. The photon radius depends on the dimensions.
1177
+ The photon radius is plotted in Fig. 4, it is evident that as the mass of the black
1178
+ hole increases the radius decreases. It is maximum in D = 4 but decreases sharply
1179
+ as the dimension increases but almost constant with the increase in dimension. The
1180
+ fig (9) shows that as the mass increases the radius of the shadow also increases.
1181
+ 6 Discussion
1182
+ We obtain non-singular black hole (NSBH) solutions in the higher dimensional
1183
+ Einstein’s general theory of gravity (GR) and found that the methods in GR can
1184
+ be adopted also in Rastall gravity. Considering a specific exponential form of the
1185
+ energy density we obtain NSBH which reduces to the Dymnikova NSBH solution
1186
+ [25] obtained in the usual four dimensions (D = 4). In 2+1 dimensions no black hole
1187
+ solution exists. However, a non-rotating NSBH solutions obtained by Dymnikova
1188
+ in four dimensional GR can be accommodated in higher dimensions. We obtained
1189
+ NSBH in a vacuum described by T t
1190
+ t + T rr = 0 with pβŠ₯ < 0 (where i = 1, ..., D βˆ’ 2)
1191
+ near the center indicating requirement of exotic matter which however extends up
1192
+ to certain height thereafter pβŠ₯ > 0 for r > οΏ½ Dβˆ’2
1193
+ Dβˆ’1
1194
+ οΏ½
1195
+ 1
1196
+ Dβˆ’1 rβˆ— in GR. But in the Rastall
1197
+ gravity pβŠ₯ > 0 for r >
1198
+ οΏ½
1199
+ 2βˆ’D+2Ξ»D
1200
+ (Dβˆ’1)(2Ξ»βˆ’1)
1201
+ οΏ½
1202
+ 1
1203
+ Dβˆ’1 rβˆ— when Ξ» ΜΈ= Dβˆ’2
1204
+ 2D
1205
+ and thereafter at a
1206
+ large distance it vanishes because the pressure decreases rapidly i.e., exponentially.
1207
+ Both in GR and modified gravity it indicates existence of exotic matter near the
1208
+ center of the NSBH but in the later case the Rastall parameter plays an important
1209
+ role in determining the distance from the centre where the normal matter exists in
1210
+ the tangential direction. In the usual four dimensions at the center of the NSBH
1211
+ in the modified theory we get the following estimations ρ(r = 0) = 2(1 βˆ’ 4Ξ»)ρ0
1212
+ and tangential pressure PβŠ₯ = βˆ’2(1 βˆ’ 4Ξ») which are determined by the Rastall
1213
+ parameter Ξ». It is evident that exotic matter at the center of the black hole requires
1214
+ both in GR (as Ξ» = 0) as well as in Rastall theory with the lower limiting value
1215
+ of the Rastall parameter Ξ» < 1
1216
+ 4. The tangential pressure is negative it indicates
1217
+
1218
+ Nonsingular Black Holes in Higher dimensions
1219
+ 15
1220
+ NSBH surrounded by exotic matter in Rastall gravity, existence of BH with exotic
1221
+ matter also reported in [35]. Thus NSBH is realized with both the central radial
1222
+ pressure and tangential pressure negative and equal initially but an anisotropy
1223
+ in pressure develops away from the center in Rastall gravity with normal matter
1224
+ thereafter when r >
1225
+ οΏ½
1226
+ Dβˆ’2+2Ξ»D
1227
+ (1βˆ’2Ξ»)(Dβˆ’2)
1228
+ οΏ½
1229
+ 1
1230
+ Dβˆ’1 rβˆ—.
1231
+ However, we note a new and interesting result in Rastall theory that permits
1232
+ a NSBH with normal matter in the usual four and in higher dimensions when
1233
+ Ξ» = Dβˆ’2
1234
+ 2D , which however is not permitted in GR. In It is also noted that away from
1235
+ the center at a large distance, the tangential pressure remains positive definite at a
1236
+ maximum radial distance which is PβŠ₯ = (1βˆ’2Ξ») Dβˆ’1
1237
+ Dβˆ’2
1238
+ rDβˆ’1
1239
+ rDβˆ’1
1240
+ βˆ—
1241
+ ρ0e
1242
+ βˆ’ rDβˆ’1
1243
+ rDβˆ’1
1244
+ βˆ—
1245
+ . Thus one gets
1246
+ a physically realistic NSBH for 1
1247
+ 4 < Ξ» < 1
1248
+ 2 in D = 4 dimensions. For r β†’ ∞, both
1249
+ the energy density and pressure vanishes asymptotically. Thus the Rastall gravity
1250
+ has rich structure which unearth the structure of non-singular black hole even with
1251
+ normal matter for a restricted domain of the Rastall parameter depending on the
1252
+ embedding spacetime dimensions. Thus, we see that for Ξ» ΜΈ= 0 the Rastall theory
1253
+ plays an important role leading to distinct solutions relative to GR.
1254
+ The sketch of the potentials permissible in the theory are plotted in Figs. (1)
1255
+ and (2), which show that both extreme and non-extreme black holes exist.
1256
+ The contour plots in Fig. (5) for NSBH shows that the circular shadow radius
1257
+ decreases as the spacetime dimension is increased for a given mass. The circular
1258
+ shadow radius in Fig. (6) show that the radii increases with the mass of the
1259
+ compact objects for a given dimensions. The rotating NSBH will be taken up
1260
+ elsewhere.
1261
+ Acknowledgment The author would like to thank IUCAA , Pune and IUCAA
1262
+ Centre for Astronomy Research and Development (ICARD), NBU for extending
1263
+ research facilities and North Bengal University for a research grant. BCP acknowl-
1264
+ edge the suggestions and constructive criticism of the anonymous Referee.
1265
+ References
1266
+ 1. T. Kaluza, Sitz. Preuss. Acad. Wiss. F 1, 966 (1921)
1267
+ 2. O. Klein, Ann. Phys. 37, 895 (1926)
1268
+ 3. C. Cadeau and E. Woolgar, Class. Quant. Grav. 18 , 527, (2001)
1269
+ 4. A. Strominger and C. Vafa, Phys. Lett. B 379, 99 (1996)
1270
+ 5. O. Aharony, S. S. Gubser, J. M. Maldacena, H. Ooguri and Y. Oz, Phys. Rept. 323, 183
1271
+ (2000)
1272
+ 6. R. Emparan, H. S. Reall. Living Rev.Rel. 11, 6 (2008)
1273
+ 7. R. C. Myers and M. J. Perry, Ann. Phys. 172, 314 (1986)
1274
+ 8. Y. Shen and Z. Tan, Phys. Lett. A 142 , 341 (1989)
1275
+ 9. B. Iyer and C. V. Vishveshwara, Pramana J. Phys. 32, 749 (1989)
1276
+ 10. A. Chodos and S. Detweiler, Gen. Rel. Grav. 14, 879 (1982)
1277
+ 11. G. W. Gibbons and D. L. Wiltshire, Ann.Phys. 167, 201 (1986)
1278
+ 12. P.O. Mazur, J. Math. Phys. 28, 406 (1987)
1279
+ 13. V. P. Frolov, A. Zel’nikov and U. Bleyer, Ann. Phys. (Leipzig) 44 , 37 (1987)
1280
+ 14. D. Xu, Class. Quant. Grav. 5 , 871, (1988)
1281
+ 15. R. Gregory and R. Laflamme, Phys. Rev. D 37, 3051(988)
1282
+ 16. L. Sokolowski and B. Carr, Phys. Lett. B 176, 334 (1986)
1283
+ 17. B. Mandelbrot, The Fractal Geometry of Nature (San Francisco, CA: Freeman, 1982)
1284
+ 18. H. Yu and L. H. Ford, Phys. Lett. B 496, 107 (2000)
1285
+
1286
+ 16
1287
+ Bikash Chandra Paul
1288
+ 19. L. Randall and R. Sundrum, Phys. Rev. Lett. 83 , 4690 (1999)
1289
+ 20. P. Rastall, Phys. Rev. D 6, 3357 (1972)
1290
+ 21. H. Moradpour, Phys. Lett.
1291
+ B 757, 187 (2016); F.F. Yuan, P. Huang, Class. Quantum
1292
+ Gravity. 34, 077001 (2017)
1293
+ 22. M. Capone,V.F. Cardone, M.L. Ruggiero,
1294
+ Nuovo Cim. B 125, 1133 (2011); C.E.M.
1295
+ Batista, M.H. Daouda, J.C. Fabris, O.F. Piattella, D.C. Rodrigues, Phys.Rev. D 85, 084008
1296
+ (2012); G.F. Silva, O.F. Piattella, J.C. Fabris, L. Casarini, T.O. Barbosa, Gravit. Cosmol.
1297
+ 19, 156 (2013)
1298
+ 23. I. P. Lobo, H. Moradpour, J. P. Morais Graca and I. G. Salako, Int. Journal of Mod. Phys.
1299
+ D 27, 1850070 (2018); K. Bamba, A. Jawad, S. Rafique. H. Moradpour, Euro. Phys. J. C
1300
+ 78, 986 (2018).
1301
+ 24. J Bardeen, Proc. GR5, Tiflis, USSR (1968).
1302
+ 25. I. Dymnikova, Gen. Rel. Grav. 24 235; (1992)
1303
+ 26. I Dymnikova and B. Soltysek, AIP Conference Proceedings 453, 460 (1998)
1304
+ 27. I. Dymnikova, Gravitation & Cosmology 8 Suppl. 131 (2002)
1305
+ 28. I. Dymnikova, Int. J. Mod. Phys. D 5, 529 (1996)
1306
+ 29. I. Dymnikova, Class. Quantum Grav. 19, 725 (2002)
1307
+ 30. M. Mars, M. M. Mart’in-Prats, J. M. M. Senovilla, Class. Quantum Grav. 13, L51 (1996)
1308
+ 31. A. Borde, Phys. Rev. D 55, 7615 (1997)
1309
+ 32. E. Ay’on-Beato, A. Garci’a, Phys. Rev. Lett. 80, 5056 (1998)
1310
+ 33. M. R. Mbonye, D. Kazanas, Phys. Rev. D 72, 024016 (2005)
1311
+ 34. S. A. Hayward, Phys. Rev. Lett. 96 , 031103 (2006)
1312
+ 35. J. P. Morais Graca, I. P. Lobo, Eur. Phys. J. C 78, 101 (2018)
1313
+ 36. B. Carter, Phys. Rev. 174, 1559 (1968)
1314
+ 37. S. Vazquez, E. P. Esteban, Nuovo Cim. B 119, 489 (2004)
1315
+ 38. C. Subrahmanyan, The mathematical theory of black holes, (Oxford University Press,
1316
+ 1992)
1317
+
MNAyT4oBgHgl3EQfgfgN/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
MNE0T4oBgHgl3EQf0AKB/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:68b1648e3efbbf2041f348a461b369c1a087e4db35a35c73d263f679c5fac154
3
+ size 5832749
MNE4T4oBgHgl3EQfKAyT/content/tmp_files/2301.04926v1.pdf.txt ADDED
@@ -0,0 +1,1133 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ CLIP2Scene: Towards Label-efficient 3D Scene Understanding by CLIP
2
+ Runnan Chen1, Youquan Liu2, Lingdong Kong3, Xinge Zhu6, Yuexin Ma5,
3
+ Yikang Li4, Yuenan Hou4, Yu Qiao4, Wenping Wang7
4
+ 1The University of Hong Kong
5
+ 2Hochschule Bremerhaven
6
+ 3National University of Singapore
7
+ 4Shanghai AI Lab
8
+ 5ShanghaiTech University
9
+ 6The Chinese University of Hong Kong
10
+ 7Texas A&M University
11
+ Abstract
12
+ Contrastive
13
+ language-image
14
+ pre-training
15
+ (CLIP)
16
+ achieves promising results in 2D zero-shot and few-shot
17
+ learning.
18
+ Despite the impressive performance in 2D
19
+ tasks, applying CLIP to help the learning in 3D scene
20
+ understanding has yet to be explored. In this paper, we
21
+ make the first attempt to investigate how CLIP knowledge
22
+ benefits 3D scene understanding. To this end, we propose
23
+ CLIP2Scene, a simple yet effective framework that transfers
24
+ CLIP knowledge from 2D image-text pre-trained models to
25
+ a 3D point cloud network. We show that the pre-trained
26
+ 3D network yields impressive performance on various
27
+ downstream tasks, i.e., annotation-free and fine-tuning
28
+ with labelled data for semantic segmentation. Specifically,
29
+ built upon CLIP, we design a Semantic-driven Cross-modal
30
+ Contrastive Learning framework that pre-trains a 3D
31
+ network via semantic and spatial-temporal consistency
32
+ regularization. For semantic consistency regularization, we
33
+ first leverage CLIP’s text semantics to select the positive
34
+ and negative point samples and then employ the contrastive
35
+ loss to train the 3D network. In terms of spatial-temporal
36
+ consistency regularization, we force the consistency be-
37
+ tween the temporally coherent point cloud features and
38
+ their corresponding image features.
39
+ We conduct experi-
40
+ ments on the nuScenes and SemanticKITTI datasets. For
41
+ the first time, our pre-trained network achieves annotation-
42
+ free 3D semantic segmentation with 20.8% mIoU. When
43
+ fine-tuned with 1% or 100% labelled data, our method
44
+ significantly outperforms other self-supervised methods,
45
+ with improvements of 8% and 1% mIoU, respectively.
46
+ Furthermore, we demonstrate its generalization capability
47
+ Semantic and
48
+ Spatial-Temporal
49
+ Consistency
50
+ Regularization
51
+ Image
52
+ Encoder
53
+ Annotation-free
54
+ 1% annotation
55
+ CLIP2Scene
56
+ Text
57
+ Encoder
58
+ CLIP
59
+ How CLIP benefits
60
+ 3D scene
61
+ understanding?
62
+ 100% annotation
63
+ Semantic-driven Cross-modal Contrastive Learning
64
+ car, bus
65
+ pedestrian
66
+ car
67
+ Figure 1. We explore whether and how CLIP knowledge benefits
68
+ 3D scene understanding. To this end, we propose CLIP2Scene, a
69
+ semantic-driven cross-modal contrastive learning framework that
70
+ leverages CLIP knowledge to pre-train a 3D point cloud seg-
71
+ mentation network via semantic and spatial-temporal consistency
72
+ regularization.
73
+ CLIP2Scene yields impressive performance on
74
+ annotation-free 3D semantic segmentation and significantly out-
75
+ performs other self-supervised methods when fine-tuning on an-
76
+ notated data.
77
+ for handling cross-domain datasets.
78
+ 1. Introduction
79
+ 3D scene understanding is fundamental in autonomous
80
+ driving, robot navigation, etc [24, 26].
81
+ Current deep
82
+ arXiv:2301.04926v1 [cs.CV] 12 Jan 2023
83
+
84
+ learning-based methods have shown inspirational perfor-
85
+ mance on 3D point cloud data [37, 50, 29, 44, 15, 45].
86
+ However, some drawbacks hinder their real-world applica-
87
+ tions. The first one comes from their heavy reliance on the
88
+ large collection of the annotated point clouds, especially
89
+ when high-quality 3D annotations are expensive to acquire
90
+ [34, 40]. Besides, they typically fail to recognize novel ob-
91
+ jects that are never seen in the training data [11, 35]. As
92
+ a result, it may need extra annotation efforts to train the
93
+ model on recognizing these novel objects, which is both te-
94
+ dious and time-consuming.
95
+ Contrastive Vision-Language Pre-training (CLIP) [38]
96
+ provides a new perspective that mitigates the above issues
97
+ in 2D vision. It was trained on large-scale free-available
98
+ image-text pairs from websites and built vision-language
99
+ correlation to achieve promising open-vocabulary recogni-
100
+ tion. MaskCLIP [49] further explores semantic segmen-
101
+ tation based on CLIP. With minimal modifications to the
102
+ CLIP pre-trained network, MaskCLIP can be directly used
103
+ for the semantic segmentation of novel objects without ad-
104
+ ditional training efforts. PointCLIP [48] reveals that the
105
+ zero-shot classification ability of CLIP can be generalized
106
+ from the 2D image to the 3D point cloud. It perspectively
107
+ projects a point cloud frame into different views of 2D depth
108
+ maps that bridge the modal gap between the image and
109
+ the point cloud. The above studies indicate the potential
110
+ of CLIP on enhancing the 2D segmentation and 3D clas-
111
+ sification performance. However, whether and how CLIP
112
+ knowledge benefits 3D scene understanding is still under-
113
+ explored.
114
+ In this paper, we explore how to leverage CLIP’s 2D
115
+ image-text pre-learned knowledge for 3D scene understand-
116
+ ing. Previous cross-modal knowledge distillation methods
117
+ [40, 34] suffer from the optimization-conflict issue, i.e.,
118
+ some of the positive pairs are regarded as negative sam-
119
+ ples for contrastive learning, leading to unsatisfactory rep-
120
+ resentation learning and hammering the performance of
121
+ downstream tasks. Besides, they also ignore the tempo-
122
+ ral coherence of the multi-sweep point cloud, failing to
123
+ utilize the rich inter-sweep correspondence.
124
+ To handle
125
+ the mentioned problems, we propose a novel Semantic-
126
+ driven Cross-modal Contrastive Learning framework that
127
+ fully leverages CLIP’s semantic and visual information to
128
+ regularize a 3D network. Specifically, we propose Seman-
129
+ tic Consistency Regularization and Spatial-Temporal Con-
130
+ sistency Regularization. In semantic consistency regular-
131
+ ization, we utilize CLIP’s text semantics to select the posi-
132
+ tive and negative point samples for less-conflict contrastive
133
+ learning. For spatial-temporal consistency regularization,
134
+ we take CLIP’s image pixel feature to impose a soft con-
135
+ straint on points within local space and time. Such oper-
136
+ ation also prevents the network from degenerating due to
137
+ image-to-point calibration errors.
138
+ We conduct several downstream tasks on nuScenes to
139
+ verify how the pre-trained network benefits the 3D scene
140
+ understanding.
141
+ The first one is annotation-free semantic
142
+ segmentation. Following MaskCLIP, we place class names
143
+ into multiple hand-crafted templates as prompts and av-
144
+ erage the text embeddings generated by CLIP to conduct
145
+ the annotation-free segmentation. For the first time, our
146
+ method achieves 20.8% mIoU annotation-free 3D semantic
147
+ segmentation without any labelled data for training. Sec-
148
+ ondly, we compare with other self-supervised methods to
149
+ verify the superiority of our method in label-efficient learn-
150
+ ing. When fine-tuning the 3D network with 1% or 100% la-
151
+ belled data, our method significantly outperforms state-of-
152
+ the-art self-supervised methods, with improvements of 8%
153
+ and 1% mIoU, respectively. Besides, to verify the general-
154
+ ization capability, we pre-train the network on the nuScenes
155
+ dataset and evaluate it on the SemanticKITTI dataset. Our
156
+ method still significantly outperforms state-of-the-art meth-
157
+ ods.
158
+ The contributions of our work are summarized as fol-
159
+ lows.
160
+ β€’ The first work that distils CLIP knowledge to a 3D net-
161
+ work for 3D scene understanding.
162
+ β€’ We propose a novel Semantic-driven Cross-modal
163
+ Contrastive Learning framework that pre-trains a 3D
164
+ network via spatial-temporal and semantic consistency
165
+ regularization.
166
+ β€’ We
167
+ propose
168
+ a
169
+ novel
170
+ Semantic-guided
171
+ Spatial-
172
+ Temporal Consistency Regularization that forces the
173
+ consistency between the temporally coherent point
174
+ cloud features and their corresponding image features.
175
+ β€’ For the first time, our method achieves promising
176
+ performance on annotation-free 3D scene segmenta-
177
+ tion and significantly outperforms state-of-the-art self-
178
+ supervised methods when fine-tuning with labelled
179
+ data.
180
+ 2. Related Work
181
+ Zero-shot Learning in 3D. The objective of zero-shot
182
+ learning (ZSL) is to recognize objects that are unseen in
183
+ the training set. Many efforts have been devoted to the 2D
184
+ recognition tasks [8, 30, 47, 36, 31, 1, 43, 32, 4, 2, 19, 33,
185
+ 23], and few works concentrate on performing ZSL in the
186
+ 3D domain [18, 11, 35, 16, 17]. [18] makes the first at-
187
+ tempt to apply ZSL to 3D tasks, where they train PointNet
188
+ [37] on ”seen” samples and test on ”unseen” samples. Sub-
189
+ sequent work [16] addresses the hubness problem caused
190
+ by the low-quality point cloud features. [17] proposes the
191
+ triplet loss to boost the performance under the transductive
192
+ setting, where the ”unseen” class is observed and unlabeled
193
+
194
+ Spatial-Temporal Consistency Regularization
195
+ Image
196
+ Encoder
197
+ Text
198
+ Encoder
199
+ car, bus
200
+ Pedestrian
201
+ …
202
+ A
203
+ photo
204
+ of a { };
205
+ This is
206
+ the { }
207
+ in the
208
+ scene;
209
+ …
210
+ Semantic Consistency Regularization
211
+ Point
212
+ Encoder
213
+ CLIP
214
+ pixel-to-text mapping
215
+ pixel-point-text pairs
216
+ pixel-to-point mapping
217
+ 3D Network
218
+ point-text pairs
219
+ … … …
220
+ …
221
+ …
222
+ …
223
+ …
224
+ Multi-sweeps
225
+ calibration
226
+ …
227
+ …
228
+ grid 1
229
+ grid 2
230
+ grid 3
231
+ pulling force
232
+ Semantic-guided fusion features
233
+ text embedding
234
+ point feature
235
+ text embedding
236
+ point feature
237
+ pixel feature
238
+ point feature
239
+ prompts
240
+ 𝑃1
241
+ 𝑃2
242
+ 𝑃3
243
+ image feature
244
+ … …
245
+ …
246
+ …
247
+ grid 1
248
+ grid 2
249
+ grid 3
250
+ Figure 2. Illustration of the Semantic-driven Cross-modal Contrastive Learning. Firstly, we obtain the text embeddings ti, image pixel
251
+ feature xi, and point feature pi by text encoder, image encoder, and point encoder, respectively. Secondly, we leverage CLIP knowledge to
252
+ construct positive and negative samples for contrastive learning. Thus we obtain point-text pairs {xi, ti}M
253
+ i=1 and all pixel-point-text pairs
254
+ in a short temporal {Λ†xk
255
+ i , Λ†pk
256
+ i , tk
257
+ i }
258
+ Λ†
259
+ M,K
260
+ i=1,k=1. Here, {xi, ti}M
261
+ i=1 and {Λ†xk
262
+ i , Λ†pk
263
+ i , tk
264
+ i }
265
+ Λ†
266
+ M,K
267
+ i=1,k=1 are used for Semantic Consistency Regularization and
268
+ Spatial-Temporal Consistency Regularization, respectively. Lastly, we perform Semantic Consistency Regularization by pulling the point
269
+ features to their corresponding text embedding and Spatial-Temporal Consistency Regularization by mimicking the temporally coherent
270
+ point features to their corresponding pixel features.
271
+ in the training phase. Recently, some studies introduced
272
+ CLIP into zero-shot learning. MaskCLIP [49] investigates
273
+ the problem of utilizing CLIP to help the 2D dense pre-
274
+ diction tasks and exhibits encouraging zero-shot semantic
275
+ segmentation performance. PointCLIP [48] is the pioneer-
276
+ ing work that applies CLIP to 3D recognition. As opposed
277
+ to previous approaches that require training on the labelled
278
+ point cloud, PointCLIP is free from any 3D training and
279
+ shows impressive performance on zero-shot and few-shot
280
+ classification tasks. Our work takes a step further to inves-
281
+ tigate whether the rich semantic and visual knowledge in
282
+ CLIP can benefit the 3D semantic segmentation tasks.
283
+ Self-supervised Representation Learning. The purpose
284
+ of self-supervised learning is to learn a good representa-
285
+ tion that benefits the downstream tasks. The dominant ap-
286
+ proaches resort to contrastive learning to pre-train the net-
287
+ work [27, 25, 21, 20, 14, 13, 7, 10, 12, 9]. Recently, inspired
288
+ by the success of CLIP, leveraging the pre-trained model
289
+ of CLIP to the downstream tasks has raised the commu-
290
+ nity’s attention. DenseCLIP [39] utilizes the CLIP’s pre-
291
+ trained knowledge for dense image pixel prediction. Det-
292
+ CLIP [46] proposes a pre-training method equipped with
293
+ CLIP for open-world detection. In this paper, we make the
294
+ first attempt to pre-train a 3D network with CLIP’s knowl-
295
+ edge for 3D scene understanding.
296
+ Cross-modal Knowledge Distillation.
297
+ Recently, an in-
298
+ creasing number of researchers have focused on transferring
299
+ the knowledge in 2D images to 3D point cloud [34, 40].
300
+ PPKT [34] proposes the contrastive pixel-to-point knowl-
301
+ edge transfer to utilize the rich information in image back-
302
+ bones. SLidR [40] resorts to the InfoNCE loss to help the
303
+ 3D network distil rich knowledge from the 2D image back-
304
+ bone. Our work explores leveraging the image-text pre-
305
+ trained CLIP knowledge to help 3D scene understanding.
306
+ 3. Methodology
307
+ Considering the impressive open-vocabulary perfor-
308
+ mance achieved by CLIP in image classification and seg-
309
+ mentation, natural curiosities have been raised. Can CLIP
310
+ endow the ability to a 3D network for annotation-free
311
+ scene understanding?
312
+ And further, will it promote the
313
+ network performance when fine-tuned on labelled data?
314
+ To answer the above questions, we study the cross-modal
315
+ knowledge transfer of CLIP for 3D scene understanding,
316
+ termed CLIP2Scene. Our work is a pioneer in exploiting
317
+ CLIP knowledge for 3D scene understanding. In what fol-
318
+ lows, we revisit the CLIP applied in 2D open-vocabulary
319
+ classification and semantic segmentation, then present our
320
+ CLIP2Scene in detail.
321
+ Our approach consists of three
322
+ major components: Semantic Consistency Regularization,
323
+ Semantic-Guided Spatial-Temporal Consistency Regular-
324
+ ization, and Switchable Self-Training Strategy.
325
+
326
+ car
327
+ road
328
+ bicycle
329
+ ...
330
+ building
331
+ Text embeddings
332
+ Figure 3. Illustration of the image pixel-to-text mapping.
333
+ The
334
+ dense pixel-text correspondence {xi, ti}M
335
+ i=1 is extracted by the
336
+ off-the-shelf method MaskCLIP [49].
337
+ 3.1. Revisiting CLIP
338
+ Contrastive Vision-Language Pre-training (CLIP) miti-
339
+ gates the following drawbacks that dominate the computer
340
+ vision field: 1. Deep models need a large amount of for-
341
+ matted and labelled training data, which is expensive to ac-
342
+ quire; 2. The model’s generalization ability is weak, mak-
343
+ ing it difficult to migrate to a new scenario with unseen
344
+ objects. CLIP consists of an image encoder (ResNet [28]
345
+ or ViT [6]) and a text encoder (Transformer [42]), both
346
+ respectively project the image and text representation to a
347
+ joint embedding space. During training, CLIP constructs
348
+ positive and negative samples from 400 million image-text
349
+ pairs to train both encoders with a contrastive loss, where
350
+ the large-scale image-text pairs are free-available from the
351
+ Internet and assumed to contain every class of images and
352
+ most concepts of text. Therefore, CLIP can achieve promis-
353
+ ing open-vocabulary recognition.
354
+ For 2D zero-shot classification, CLIP first places the
355
+ class name into a pre-defined template to generate the text
356
+ embeddings and then encodes images to obtain image em-
357
+ beddings.
358
+ Next, it calculates the similarity matrices be-
359
+ tween images and text embeddings to determine the class.
360
+ MaskCLIP further extends CLIP into 2D semantic segmen-
361
+ tation. Specifically, MaskCLIP modifies the attention pool-
362
+ ing layer of the CLIP’s image encoder, thus performing
363
+ pixel-level mask prediction instead of the global image-
364
+ level prediction.
365
+ 3.2. CLIP2Scene
366
+ As shown in Fig. 2, we first leverage CLIP and 3D net-
367
+ work to respectively extract the text embeddings, image
368
+ pixel feature and point feature.
369
+ Secondly, we construct
370
+ positive and negative samples based on CLIP’s knowledge.
371
+ Lastly, we impose Semantic Consistency Regularization by
372
+ pulling the point features to their corresponding text embed-
373
+ ding. At the same time, we apply Spatial-Temporal Con-
374
+ sistency Regularization by forcing the consistency between
375
+ temporally coherent point features and their corresponding
376
+ pixel features. In what follows, we present the details and
377
+ insights.
378
+ 3.2.1
379
+ Semantic Consistency Regularization
380
+ As CLIP is pre-trained on 2D images and text, our first con-
381
+ cern is the domain gap between 2D images and the 3D point
382
+ cloud. To this end, we build dense pixel-point correspon-
383
+ dence and transfer image knowledge to the 3D point cloud
384
+ via the pixel-point pairs. Specifically, we calibrate the Li-
385
+ DAR point cloud with corresponding images captured by
386
+ six cameras. Therefore, the dense pixel-point correspon-
387
+ dence {xi, pi}M
388
+ i=1 can be obtained accordingly, where xi
389
+ and pi indicates i-th paired image feature and point feature,
390
+ which are respectively extracted by the CLIP’s image en-
391
+ coder and the 3D network. M is the number of pairs. Note
392
+ that it is an online operation and is irreverent to the image
393
+ and point data augmentation.
394
+ Previous methods [40, 34] provide a promising solution
395
+ to cross-modal knowledge transfer.
396
+ They first construct
397
+ positive pixel-point pairs {xi, pi}M
398
+ i=1 and negative pairs
399
+ {xi, pj}(i ΜΈ= j), and then pull in the positive pairs while
400
+ pushing away the negative pairs in the embedding space via
401
+ the InfoNCE loss. Despite the encourageable performance
402
+ of previous methods in transferring cross-modal knowl-
403
+ edge, they are both confronted with the same optimization-
404
+ conflict issue. For example, suppose i-th pixel xi and j-th
405
+ point pj are in the different positions of the same instance
406
+ with the same semantics. However, the InfoNCE loss will
407
+ try to push them away, which is unreasonable and ham-
408
+ mer the performance of the downstream tasks [40]. In light
409
+ of this, we propose a Semantic Consistency Regularization
410
+ that leverages the CLIP’s semantic information to allevi-
411
+ ate this issue. Specifically, we generate the dense pixel-
412
+ text pairs {xi, ti}M
413
+ i=1 by following the off-the-shelf method
414
+ MaskCLIP [49] (Fig. 3), where ti is the text embedding gen-
415
+ erated from the CLIP’s text encoder. Note that the pixel-text
416
+ mappings are free-available from CLIP without any addi-
417
+ tional training. We then transfer pixel-text pairs to point-
418
+ text pairs {pi, ti}M
419
+ i=1 and utilize the text semantics to se-
420
+ lect the positive and negative point samples for contrastive
421
+
422
+ Image 𝐼
423
+ Pixel-to-point mapping
424
+ 𝑃1
425
+ 𝑃2
426
+ 𝑃3
427
+ Multi-sweeps calibration
428
+ … … …
429
+ …
430
+ …
431
+ grid 1
432
+ grid 2
433
+ grid 3
434
+ …
435
+ …
436
+ grid 1
437
+ grid 2
438
+ grid 3
439
+ {𝑓𝑛}𝑛=1
440
+ 𝑁
441
+ ොπ‘₯𝑖
442
+ π‘˜, ΰ·œπ‘π‘–
443
+ π‘˜
444
+ 𝑛=1,π‘˜=1
445
+ 𝑁,𝐾
446
+ Text embedding
447
+ Figure 4. Illustration of the image pixel-to-point mapping (left)
448
+ and semantic-guided fusion feature generation (right). We build
449
+ the grid-wise correspondence between an image I and the tem-
450
+ porally coherent LiDAR point cloud {Pk}K
451
+ k=1 within S seconds
452
+ and generate semantic-guided fusion features for individual grids.
453
+ Both {Λ†xk
454
+ i , Λ†pk
455
+ i }
456
+ Λ†
457
+ M,K
458
+ i=1,k=1 and {fn}N
459
+ n=1 are used to perform Spatial-
460
+ Temporal Consistency Regularization.
461
+ learning. The objective function is as follows:
462
+ LS info = βˆ’
463
+ C
464
+ οΏ½
465
+ c=1
466
+ log
467
+ οΏ½
468
+ ti∈c,pi exp(D(ti, pi)/Ο„)
469
+ οΏ½
470
+ ti∈c,tj /∈c,pj exp(D(ti, pj)/Ο„),
471
+ (1)
472
+ where ti ∈ c indicates that ti is generated by c-th classes
473
+ name, and C is the number of classes. D denotes the scalar
474
+ product operation and Ο„ is a temperature term (Ο„ > 0).
475
+ Since the text is composed of class names placed into
476
+ pre-defined templates, the text embedding represents the se-
477
+ mantic information of the corresponding class. Therefore,
478
+ those points with the same semantics will be restricted near
479
+ the same text embedding, and those with different semantics
480
+ will be pushed away. To this end, our Semantic Consistency
481
+ Regularization causes less conflict in contrastive learning.
482
+ 3.2.2
483
+ Semantic-guided Spatial-temporal Consistency
484
+ Regularization
485
+ Besides semantic consistency regularization, we consider
486
+ how image pixel features help to regularize a 3D network.
487
+ The natural alternative directly pulls in the point feature
488
+ with its corresponding pixel in the embedding space. How-
489
+ ever, after trial and error, we observe that the network easily
490
+ degenerates and achieves poor performance in the down-
491
+ stream tasks when following the aforementioned strategy.
492
+ The main reason lies in the noise-assigned semantics of the
493
+ image pixel and the imperfect pixel-point mapping caused
494
+ by the calibration errors. To this end, we propose a novel
495
+ semantic-guided Spatial-Temporal Consistency Regulariza-
496
+ tion to alleviate the problem by imposing a soft constraint
497
+ on points within local space and time.
498
+ Specifically, given an image I and temporally coherent
499
+ LiDAR point cloud {Pk}K
500
+ k=1, where K is the number of
501
+ sweeps within S seconds. Note that the image is matched
502
+ to the first frame of the point cloud P1 with pixel-point pairs
503
+ {Λ†x1
504
+ i , Λ†p1
505
+ i } Λ†
506
+ M
507
+ i=1. We register the rest of the point cloud to the
508
+ first frame via the calibration matrices and map them to the
509
+ image (Fig. 4). Thus we obtain all pixel-point-text pairs
510
+ in a short temporal {Λ†xk
511
+ i , Λ†pk
512
+ i , tk
513
+ i }
514
+ Λ†
515
+ M,K
516
+ i=1,k=1. Next, we divide
517
+ the entire stitched point cloud into regular grids {gn}N
518
+ n=1,
519
+ where the temporally coherent points are located in the
520
+ same grid. We impose the spatial-temporal consistency con-
521
+ straint within individual grids by the following objective
522
+ function:
523
+ LSSR =
524
+ οΏ½
525
+ gn
526
+ οΏ½
527
+ (Λ†i,Λ†k)∈gn
528
+ (1 βˆ’ sigmoid(D(Λ†p
529
+ Λ†k
530
+ Λ†i , fn)))/N, (2)
531
+ where (Λ†i, Λ†k) ∈ gn indicates the pixel-point pair {Λ†xk
532
+ i , Λ†pk
533
+ i }
534
+ is located in the n-th grid. {fn}N
535
+ n=1 is a semantic-guided
536
+ cross-modal fusion feature formulated by:
537
+ fn =
538
+ οΏ½
539
+ (Λ†i,Λ†k)∈gn
540
+ a
541
+ Λ†k
542
+ Λ†i βˆ— Λ†x
543
+ Λ†k
544
+ Λ†i + b
545
+ Λ†k
546
+ Λ†i βˆ— Λ†p
547
+ Λ†k
548
+ Λ†i ,
549
+ (3)
550
+ where aˆk
551
+ ˆi and bˆk
552
+ Λ†i are attention weight calculated by:
553
+ a
554
+ Λ†k
555
+ Λ†i =
556
+ exp(D(ˆxˆk
557
+ Λ†i , t1
558
+ Λ†i )/Ξ»)
559
+ οΏ½
560
+ (Λ†i,Λ†k)∈gn exp(D(Λ†xΛ†k
561
+ Λ†i , t1
562
+ ˆi )/λ) + exp(D(ˆpˆk
563
+ Λ†i , t1
564
+ Λ†i )/Ξ»)
565
+ ,
566
+ b
567
+ Λ†k
568
+ Λ†i =
569
+ exp(D(ˆpˆk
570
+ Λ†i , t1
571
+ Λ†i )/Ξ»)
572
+ οΏ½
573
+ (Λ†i,Λ†k)∈gn exp(D(Λ†xΛ†k
574
+ Λ†i , t1
575
+ ˆi )/λ) + exp(D(ˆpˆk
576
+ Λ†i , t1
577
+ Λ†i )/Ξ»)
578
+ ,
579
+ (4)
580
+ where Ξ» is the temperature term.
581
+ Actually, those pixel and point features within the local
582
+ grid gn are restricted near a dynamic centre fn. Thus, such a
583
+ soft constraint alleviates the noisy prediction and calibration
584
+ error issues. At the same time, it imposes Spatio-Temporal
585
+ Regularization on the temporally coherent point features.
586
+ 3.2.3
587
+ Switchable Self-training Strategy
588
+ We combine the loss function LS info and LSSR to end-
589
+ to-end train the whole network, where the CLIP’s image
590
+ and text encoder backbone are frozen during training. We
591
+ find that method worked only when the pixel-point feature
592
+ {xi, pi}M
593
+ i=1 and {Λ†xk
594
+ i , Λ†pk
595
+ i }
596
+ Λ†
597
+ M,K
598
+ i=1,k=1, which are used in LS info
599
+ and LSSR, are generated from different learnable linear
600
+ layer. On top of that, we further put forward an effective
601
+ strategy to promote performance. Specifically, after con-
602
+ trastive learning of the 3D network for a few epochs, we
603
+ randomly switch the point labels between the paired im-
604
+ age pixel’s labels and their own predictions for self-training.
605
+ Merely training the 3D network with their own predictions
606
+ yields satisfactory performance. Essentially, such a Switch-
607
+ able Self-Training Strategy (S3) increases the number of
608
+
609
+ Table 1. Ablation study experiments on the nuScenes validation
610
+ dataset for annotation-free semantic segmentation.
611
+ Ablation target
612
+ Settings
613
+ mIoU(%)
614
+ -
615
+ baseline
616
+ 15.1
617
+ Prompts
618
+ nuScenes
619
+ 15.1
620
+ semanticKITTI
621
+ 13.9
622
+ Cityscapes
623
+ 11.3
624
+ Regularization
625
+ w/o SCR
626
+ 19.8
627
+ KL
628
+ 0
629
+ Training Strategies
630
+ w/o S3
631
+ 18.8
632
+ ST
633
+ 10.1
634
+ Sweeps
635
+ 1 sweep
636
+ 18.7
637
+ 3 sweeps
638
+ 20.8
639
+ 5 sweeps
640
+ 20.6
641
+ merged
642
+ 18.6
643
+ -
644
+ CLIP2Scene
645
+ 20.8
646
+ positive and negative samples by switching the point pseudo
647
+ labels, which benefits cross-modal knowledge distillation.
648
+ 4. Experiments
649
+ Datasets.
650
+ We conduct experiments on two large-scale
651
+ outdoor
652
+ LiDAR
653
+ segmentation
654
+ benchmarks,
655
+ i.e.,
656
+ Se-
657
+ manticKITTI [3] and nuScenes [5, 22].
658
+ The nuScenes
659
+ dataset contains 700 scenes for training, 150 scenes for
660
+ validation and 150 scenes for testing, where 16 classes
661
+ are utilized for LiDAR semantic segmentation. As to Se-
662
+ manticKITTI, it contains 19 classes for training and evalu-
663
+ ation. It has 22 sequences, where sequences 00 to 10, 08
664
+ and 11 to 21 are used for training, validation and testing,
665
+ respectively.
666
+ Implementation Details.
667
+ We use the nuScenes [5, 22]
668
+ dataset to pre-train the network.
669
+ Following SLidR, we
670
+ pre-train the network on all key frames from 600 scenes.
671
+ Besides, we fine-tune the pre-trained network on Se-
672
+ manticKITTI [3] to verify the generalization ability. We
673
+ leverage CLIP’s image encoder and text encoder to gener-
674
+ ate image features and text embedding, respectively. Fol-
675
+ lowing MaskCLIP, we modify the attention pooling layer of
676
+ the CLIP’s image encoder, thus extracting the dense pixel-
677
+ text correspondences. We take SPVCNN [41] as the 3D
678
+ network to produce the point-wise feature. The whole net-
679
+ work is trained on the PyTorch platform. The training time
680
+ is about 40 hours for 20 epochs on two NVIDIA Tesla A100
681
+ GPUs. For the switchable self-training strategy, we ran-
682
+ domly switch the point supervision signal after 10 epochs.
683
+ The optimizer is SGD with a cosine scheduler. We set the
684
+ temperature Ξ» and Ο„ to be 1 and 0.5, respectively.
685
+ The
686
+ sweep number is set to be 3 empirically. We apply sev-
687
+ eral data augmentations in contrastive learning, including
688
+ random rotation around the z-axis and random flip on the
689
+ Table 2. Comparison of different self-supervised methods for se-
690
+ mantic segmentation on the nuScenes and SemanticKITTI valida-
691
+ tion datasets.
692
+ Initialization
693
+ nuScenes
694
+ semanticKITTI
695
+ 1%
696
+ 100%
697
+ 1%
698
+ Random
699
+ 42.2
700
+ 69.1
701
+ 32.5
702
+ PPKT [34]
703
+ 48.0
704
+ 70.1
705
+ 39.1
706
+ SLidR [40]
707
+ 48.2
708
+ 70.4
709
+ 39.6
710
+ CLIP2Scene
711
+ 56.3
712
+ 71.5
713
+ 42.6
714
+ point cloud, random horizontal flip and random crop-resize
715
+ on the image.
716
+ 4.1. Annotation-free Semantic Segmentation
717
+ After pre-training the network, we show the performance
718
+ of the 3D network when it is not fine-tuned on any annota-
719
+ tions. As no previous method reports the 3D annotation-free
720
+ segmentation performance, we compare our method with
721
+ different setups (Table 1). In what follows, we describe the
722
+ experimental settings and give insights into our method and
723
+ the different settings.
724
+ Settings. We conduct experiments on the nuScenes dataset
725
+ to evaluate the annotation-free semantic segmentation per-
726
+ formance. Following MaskCLIP [49], we place the class
727
+ name into 85 hand-craft prompts and feed it into the CLIP’s
728
+ text encoder to produce multiple text features. We then av-
729
+ erage the text features and feed the averaged features to the
730
+ classifier for point-wise prediction. Besides, to explore how
731
+ to effectively transfer CLIP’s knowledge to the 3D network
732
+ for annotation-free segmentation, We conduct the following
733
+ experiments to highlight the effectiveness of different mod-
734
+ ules in our framework.
735
+ Baseline. The input of the 3D network is only one sweep,
736
+ and we pre-train the framework via semantic consistency
737
+ regularization.
738
+ Prompts (nuScenes, semanticKITTI, Cityscapes). Based
739
+ on the baseline, we respectively replace the nuScenes, se-
740
+ manticKITTI, and Cityscapes class names into the prompts
741
+ to produce the text embedding.
742
+ Regularization (w/o STR, KL). Based on the full method,
743
+ we remove the Spatial-temporal Consistency Regulariza-
744
+ tion (w/o SCR). Besides, we abuse both SR and SCR
745
+ and distill the image feature to the point cloud by Kull-
746
+ back–Leibler (KL) divergence loss.
747
+ Training Strategies (w/o S3, ST). We abuse the Switchable
748
+ Self-Training Strategy (w/o S3) in the full method. Besides,
749
+ we show the performance of only training the 3D network
750
+ by their own predictions after ten epochs (ST).
751
+ Sweeps Number (1 sweep, 3 sweeps, 5 sweeps, and
752
+ merged). We set the sweep number K to be 1, 3, and 5, re-
753
+ spectively. Besides, we also take three sweeps of the point
754
+ cloud as the input to pre-train the network.
755
+ Effect of Different Prompts.
756
+ To verify how text em-
757
+
758
+ Ground truth
759
+ Ours*
760
+ Ours
761
+ Bus
762
+ Motorcycle
763
+ Car
764
+ Truck
765
+ Figure 5. Qualitative results of annotation-free semantic segmentation on nuScenes dataset. Note that we show the results by individual
766
+ class. From the left to the right column are the bus, motorcycle, car and truck, respectively. The first row is the ground truth; The second
767
+ row (ours*) is our prediction of the highlighted target; the third row is our prediction of full classes (ours).
768
+ bedding affects the performance, we generate various text
769
+ embeddings by the class name from different datasets
770
+ (nuScenes, SemanticKITT, and Cityscapes) for pre-training
771
+ the framework.
772
+ As shown in Table 1, we find that
773
+ even learning with other datasets’ text embedding (se-
774
+ manticKITT and Cityscapes), the 3D network could still
775
+ recognize the nuScenes’s objects with decent performance
776
+ (13.9 and 11.3 mIoU, respectively). The result shows that
777
+ the 3D network is capable of open-vocabulary recognition.
778
+ Effect of Semantic and Spatial-temporal Consistency
779
+ Regularization. We remove Spatial-temporal Consistency
780
+ Regularization (w/o SCR) from our method. Experiments
781
+ show that the performance is dramatically decreased, indi-
782
+ cating the effectiveness of our design. Besides, we also dis-
783
+ till the image feature to the point cloud by KL divergence
784
+ loss, where the text embeddings calculate the logits. How-
785
+ ever, such a method fails to transfer the semantic informa-
786
+ tion from the image. The main reason is the noise-assigned
787
+ semantics of the image pixel and the imperfect pixel-point
788
+ correspondence due to the calibration error.
789
+ Effect of Switchable Self-training Strategy. To examine
790
+ the effect of the Switchable Self-Training Strategy, we ei-
791
+ ther train the network with image supervision (w/o S3) or
792
+ train the 3D network by their own predictions. Both tri-
793
+ als witness the performance drop, indicating our Switch-
794
+ able Self-Training Strategy is efficient in cross-modal self-
795
+ supervised learning. The main reason is that the number of
796
+ positive and negative samples is enlarged by switching the
797
+ supervision signal.
798
+ Effect of Sweep Numbers. Intuitively, the performance of
799
+ our method benefits from more sweeps information. There-
800
+ fore, we also show the performance when restricting sweep
801
+ size to 1, 3, and 5, respectively. However, we observe that
802
+ the performance of 5 sweeps is similar to 3 sweeps but is
803
+ more computationally expensive. Thus, we empirically set
804
+ the sweep number to be 3.
805
+ Qualitative Evaluation. We show the qualitative evalua-
806
+ tion in Fig. 5. Note that we show the results by individ-
807
+ ual class (construction vehicle, truck, and car). The results
808
+ show that our method is able to perceive the objects without
809
+ any annotation training data. However, we also observe the
810
+ false positive predictions around the ground truth objects.
811
+ We will resolve this issue in future work.
812
+ 4.2. Annotation-efficient Semantic Segmentation
813
+ Besides annotation-free semantic segmentation, the pre-
814
+ trained 3D network also boosts the performance when it
815
+ is fine-tuned on labelled data. To the best of our knowl-
816
+ edge, only one published method SLidR studies image-to-
817
+ Lidar self-supervised representation distillation. We also
818
+ compared our method with another self-supervised method
819
+ PPKT [34] for 3D network pre-training.
820
+ In the follow-
821
+ ings, we first introduce SLidR [40] and PPKT, then compare
822
+ them in detail.
823
+ PPKT. PPKT is a cross-modal self-supervised method for
824
+ the RGB-D dataset. It performs 2D-to-3D knowledge dis-
825
+ tillation via pixel-to-point contrastive loss. Since there is
826
+ no public code, we re-implement it for a fair comparison.
827
+
828
+ Input
829
+ Ground Truth
830
+ SLidR
831
+ Ours
832
+ Figure 6. Qualitative results of fine-tuning on 1% nuScenes dataset. From the first row to the last row are the input Lidar scan, ground truth,
833
+ prediction of SLidR, and our prediction, respectively. Note that we show the results by error map, where the red point indicates the wrong
834
+ prediction. Apparently, our method achieves decent performance.
835
+ Specifically, we use the same 3D network and training pro-
836
+ tocol but replace our semantic and Spatio-Temporal Reg-
837
+ ularization with InfoNCE loss. The framework is trained
838
+ on 4, 096 randomly selected image-to-point pairs for 50
839
+ epochs.
840
+ SLidR. SLidR is an image-to-Lidar self-supervised method
841
+ for autonomous driving data.
842
+ Compared with PPKT,
843
+ it introduces image super-pixel into cross-modal self-
844
+ supervised learning. For a fair comparison, we replace our
845
+ loss function with their superpixel-driven contrastive loss.
846
+ Performance. As shown in Table 2, our method signifi-
847
+ cantly outperforms the state-of-the-art methods when fine-
848
+ tuned on 1% and 100% data, with the improvement of
849
+ 8.1% and 1.1%, respectively.
850
+ Compared with the ran-
851
+ dom initialization, the improvement is 14.1% and 2.4%, re-
852
+ spectively, indicating the efficiency of our semantic-driven
853
+ cross-modal contrastive learning framework. The qualita-
854
+ tive results are shown in Fig. 6. Besides, we also verify the
855
+ cross-domain generalization ability of our method. When
856
+ pre-training the 3D network on the nuScenes dataset and
857
+ fine-tuning on 1% SemanticKITTI dataset, our method sig-
858
+ nificantly outperforms other state-of-the-art self-supervised
859
+ methods.
860
+ Discussions. PPKT and SLidR reveal that contrastive loss
861
+ is promising for transferring knowledge from image to point
862
+ cloud. Like self-supervised learning, constructing the pos-
863
+ itive and negative samples is vital to unsupervised cross-
864
+ modal knowledge distillation.
865
+ However, previous meth-
866
+ ods suffer from the optimization-conflict issue, i.e., some
867
+ of the negative paired samples are actually positive pairs.
868
+ For example, the road occupies a large proportion of the
869
+ point cloud in a scene and is supposed to have the same
870
+ semantics in the semantic segmentation task. When ran-
871
+ domly selecting training samples, most negatively defined
872
+ road-road points are actually positive. When feedforward-
873
+ ing such training samples into contrastive learning, the con-
874
+ trastive loss will push them away in the embedding space,
875
+ leading to unsatisfactory representation learning and ham-
876
+ mering the downstream tasks’ performance.
877
+ SLidR in-
878
+ troduces superpixel-driven contrastive learning to alleviate
879
+ such issues. The motivation is that the visual representation
880
+ of the image pixel and the projected points are consistent
881
+ intra-superpixel. Although avoiding selecting the negative
882
+ image-point pairs from the same superpixel, the conflict is-
883
+ sue still exists inter-superpixel. In our CLIP2Scene, we in-
884
+ troduce the free-available dense pixel-text correspondence
885
+ to alleviate the optimization conflicts. The text embedding
886
+ represents the semantic information and can be used to se-
887
+ lect more reasonable training samples for contrastive learn-
888
+ ing.
889
+ Besides training sample selection, the previous method
890
+ also ignores the temporal coherence of the multi-sweep
891
+ point cloud. Similar to multi-view consistency, multi-sweep
892
+ consistency emphasizes inter-sweep consistency along time
893
+ series. That is, for those LiDAR points mapping to the same
894
+ image pixel, their feature should be the same. Besides, con-
895
+ sidering the sparsity of the LiDAR scan and the calibration
896
+ error between the LiDAR scan and the camera image. We
897
+
898
+ δΈ“
899
+ . - relax the pixel-to-point mapping to image grid-to-point grid
900
+ mapping and calculate the dynamic centre within the indi-
901
+ vidual grid for consistency regularization. To this end, our
902
+ Spatial-temporal consistency regularization leads to a more
903
+ comprehensive point representation.
904
+ Last but not least, the previous method typically enlarges
905
+ the number of training samples by data augmentation. In
906
+ our CLIP2Scene, we find that randomly switching the su-
907
+ pervision signal benefits self-supervised learning. Essen-
908
+ tially, our Switchable Self-Training Strategy enlarges the
909
+ training samples and prevents the network from deteriorat-
910
+ ing.
911
+ 5. Conclusion
912
+ We explored how CLIP knowledge benefits 3D scene
913
+ understanding in this paper, termed CLIP2Scene. To ef-
914
+ ficiently transfer CLIP’s image feature and text feature
915
+ to a 3D network, we propose a novel Semantic-driven
916
+ Cross-modal Contrastive Learning framework including Se-
917
+ mantic Regularization and Spatial-Temporal Regulariza-
918
+ tion. For the first time, our pre-trained 3D network achieves
919
+ annotation-free 3D semantic segmentation with decent per-
920
+ formance. Besides, our method significantly outperforms
921
+ state-of-the-art self-supervised methods when fine-tuning
922
+ the 3D network with labelled data.
923
+ Potential Negative Impacts. Although our approach im-
924
+ proves the 3D semantic segmentation performance in gen-
925
+ eral, its effectiveness under adversarial attack is not con-
926
+ sidered, which could be safety-critical in practical applica-
927
+ tions, such as autonomous driving and robot navigation.
928
+ References
929
+ [1] Z. Akata, F. Perronnin, Z. Harchaoui, and C. Schmid. Label-
930
+ embedding for attribute-based classification. CVPR, pages
931
+ 819–826, 2013. 2
932
+ [2] Z. Akata, S. E. Reed, D. Walter, H. Lee, and B. Schiele. Eval-
933
+ uation of output embeddings for fine-grained image classifi-
934
+ cation. CVPR, pages 2927–2936, 2015. 2
935
+ [3] J. Behley, M. Garbade, A. Milioto, J. Quenzel, S. Behnke,
936
+ C. Stachniss, and J. Gall.
937
+ SemanticKITTI: A Dataset for
938
+ Semantic Scene Understanding of Lidar Sequences. In IEEE
939
+ International Conference on Computer Vision, pages 9297–
940
+ 9307, 2019. 6
941
+ [4] M. Bucher, S. Herbin, and F. Jurie. Generating visual repre-
942
+ sentations for zero-shot classification. ICCVW, pages 2666–
943
+ 2673, 2017. 2
944
+ [5] H. Caesar, V. Bankiti, A. H. Lang, S. Vora, V. E. Liong,
945
+ Q. Xu, A. Krishnan, Y. Pan, G. Baldan, and O. Beijbom.
946
+ NuScenes: A Multimodal Dataset for Autonomous Driving.
947
+ In IEEE Conference on Computer Vision and Pattern Recog-
948
+ nition, pages 11621–11631, 2020. 6
949
+ [6] N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov,
950
+ and S. Zagoruyko. End-to-end object detection with trans-
951
+ formers. In European conference on computer vision, pages
952
+ 213–229. Springer, 2020. 4
953
+ [7] M. Caron, I. Misra, J. Mairal, P. Goyal, P. Bojanowski, and
954
+ A. Joulin. Unsupervised learning of visual features by con-
955
+ trasting cluster assignments. Advances in Neural Informa-
956
+ tion Processing Systems, 33:9912–9924, 2020. 3
957
+ [8] S. Changpinyo, W.-L. Chao, B. Gong, and F. Sha. Synthe-
958
+ sized classifiers for zero-shot learning. CVPR, pages 5327–
959
+ 5336, 2016. 2
960
+ [9] N. Chen, L. Liu, Z. Cui, R. Chen, D. Ceylan, C. Tu, and
961
+ W. Wang. Unsupervised learning of intrinsic structural rep-
962
+ resentation points. In Proceedings of the IEEE/CVF Con-
963
+ ference on Computer Vision and Pattern Recognition, pages
964
+ 9121–9130, 2020. 3
965
+ [10] R. Chen, Z. Xinge, N. Chen, D. Wang, W. Li, Y. Ma,
966
+ R. Yang, and W. Wang. Referring self-supervised learning
967
+ on 3d point cloud. 2021. 3
968
+ [11] R. Chen, X. Zhu, N. Chen, W. Li, Y. Ma, R. Yang, and
969
+ W. Wang. Zero-shot point cloud segmentation by transfer-
970
+ ring geometric primitives. arXiv preprint arXiv:2210.09923,
971
+ 2022. 2
972
+ [12] R. Chen, X. Zhu, N. Chen, D. Wang, W. Li, Y. Ma, R. Yang,
973
+ and W. Wang. Towards 3d scene understanding by referring
974
+ synthetic models. arXiv preprint arXiv:2203.10546, 2022. 3
975
+ [13] T. Chen, S. Kornblith, M. Norouzi, and G. Hinton. A sim-
976
+ ple framework for contrastive learning of visual represen-
977
+ tations.
978
+ In International conference on machine learning,
979
+ pages 1597–1607. PMLR, 2020. 3
980
+ [14] X. Chen and K. He. Exploring simple siamese representa-
981
+ tion learning. In Proceedings of the IEEE/CVF Conference
982
+ on Computer Vision and Pattern Recognition, pages 15750–
983
+ 15758, 2021. 3
984
+ [15] R. Cheng, R. Razani, E. Taghavi, E. Li, and B. Liu. (af)2-
985
+ S3Net: Attentive Feature Fusion with Adaptive Feature Se-
986
+ lection for Sparse Semantic Segmentation Network. In IEEE
987
+ Conference on Computer Vision and Pattern Recognition,
988
+ pages 12547–12556, 2021. 2
989
+ [16] A. Cheraghian, S. Rahman, D. Campbell, and L. Petersson.
990
+ Mitigating the Hubness Problem for Zero-Shot Learning of
991
+ 3D Objects. arXiv preprint arXiv:1907.06371, 2019. 2
992
+ [17] A. Cheraghian, S. Rahman, T. F. Chowdhury, D. Campbell,
993
+ and L. Petersson. Zero-Shot Learning on 3D Point Cloud
994
+ Objects and Beyond. International Journal of Computer Vi-
995
+ sion, 130(10):2364–2384, 2022. 2
996
+ [18] A. Cheraghian, S. Rahman, and L. Petersson.
997
+ Zero-Shot
998
+ Learning of 3D Point Cloud Objects. In International Con-
999
+ ference on Machine Vision Applications (MVA), pages 1–6.
1000
+ IEEE, 2019. 2
1001
+ [19] B. Demirel, R. G. Cinbis, and N. Ikizler-Cinbis.
1002
+ At-
1003
+ tributes2classname: A discriminative model for attribute-
1004
+ based unsupervised zero-shot learning. ICCV, pages 1241–
1005
+ 1250, 2017. 2
1006
+ [20] C. Doersch, A. Gupta, and A. A. Efros. Unsupervised vi-
1007
+ sual representation learning by context prediction. In Pro-
1008
+ ceedings of the IEEE international conference on computer
1009
+ vision, pages 1422–1430, 2015. 3
1010
+
1011
+ [21] A. Dosovitskiy, J. T. Springenberg, M. Riedmiller, and
1012
+ T. Brox. Discriminative unsupervised feature learning with
1013
+ convolutional neural networks. Advances in neural informa-
1014
+ tion processing systems, 27, 2014. 3
1015
+ [22] W. K. Fong, R. Mohan, J. V. Hurtado, L. Zhou, H. Caesar,
1016
+ O. Beijbom, and A. Valada. Panoptic nuScenes: A Large-
1017
+ Scale Benchmark for LiDAR Panoptic Segmentation and
1018
+ Tracking. IEEE Robotics and Automation Letters, 7:3795–
1019
+ 3802, 2022. 6
1020
+ [23] C. Gan, M. Lin, Y. Yang, Y. Zhuang, and A. Hauptmann. Ex-
1021
+ ploring semantic inter-class relationships (sir) for zero-shot
1022
+ action recognition. In AAAI, 2015. 2
1023
+ [24] B. Gao, Y. Pan, C. Li, S. Geng, and H. Zhao. Are We Hungry
1024
+ for 3D LiDAR Data for Semantic Segmentation? A Survey
1025
+ of Datasets and Methods. IEEE Transactions on Intelligent
1026
+ Transportation Systems, 2021. 1
1027
+ [25] J.-B. Grill, F. Strub, F. AltchΒ΄e, C. Tallec, P. Richemond,
1028
+ E. Buchatskaya, C. Doersch, B. Avila Pires, Z. Guo,
1029
+ M. Gheshlaghi Azar, et al. Bootstrap your own latent-a new
1030
+ approach to self-supervised learning. Advances in neural in-
1031
+ formation processing systems, 33:21271–21284, 2020. 3
1032
+ [26] Y. Guo, H. Wang, Q. Hu, H. Liu, L. Liu, and M. Ben-
1033
+ namoun. Deep Learning for 3D Point Clouds: A Survey.
1034
+ IEEE Transactions on Pattern Analysis and Machine Intelli-
1035
+ gence, 43(12):4338–4364, 2020. 1
1036
+ [27] K. He, H. Fan, Y. Wu, S. Xie, and R. Girshick. Momentum
1037
+ contrast for unsupervised visual representation learning. In
1038
+ Proceedings of the IEEE/CVF conference on computer vi-
1039
+ sion and pattern recognition, pages 9729–9738, 2020. 3
1040
+ [28] K. He, X. Zhang, S. Ren, and J. Sun. Deep Residual Learning
1041
+ for Image Recognition. In IEEE Conference on Computer
1042
+ Vision and Pattern Recognition, pages 770–778, 2016. 4
1043
+ [29] Y. Hou, X. Zhu, Y. Ma, C. C. Loy, and Y. Li. Point-to-Voxel
1044
+ Knowledge Distillation for LiDAR Semantic Segmentation.
1045
+ In IEEE Conference on Computer Vision and Pattern Recog-
1046
+ nition, pages 8479–8488, 2022. 2
1047
+ [30] E. Kodirov, T. Xiang, and S. Gong. Semantic autoencoder
1048
+ for zero-shot learning. CVPR, pages 4447–4456, 2017. 2
1049
+ [31] C. H. Lampert, H. Nickisch, and S. Harmeling. Learning to
1050
+ detect unseen object classes by between-class attribute trans-
1051
+ fer. CVPR, pages 951–958, 2009. 2
1052
+ [32] C. H. Lampert, H. Nickisch, and S. Harmeling. Attribute-
1053
+ based classification for zero-shot visual object categoriza-
1054
+ tion. IEEE TPAMI, 36:453–465, 2014. 2
1055
+ [33] Y. Li, Z. Jia, J. Zhang, K. Huang, and T. Tan. Deep semantic
1056
+ structural constraints for zero-shot learning. In AAAI, 2018.
1057
+ 2
1058
+ [34] Y.-C. Liu, Y.-K. Huang, H.-Y. Chiang, H.-T. Su, Z.-Y. Liu,
1059
+ C.-T. Chen, C.-Y. Tseng, and W. H. Hsu. Learning From
1060
+ 2D: Contrastive Pixel-to-Point Knowledge Transfer for 3D
1061
+ Pretraining. arXiv preprint arXiv:2104.04687, 2021. 2, 3, 4,
1062
+ 6, 7
1063
+ [35] B. Michele, A. Boulch, G. Puy, M. Bucher, and R. Marlet.
1064
+ Generative zero-shot learning for semantic segmentation of
1065
+ 3d point clouds. In 3DV, pages 992–1002. IEEE, 2021. 2
1066
+ [36] A. Mishra, M. S. K. Reddy, A. Mittal, and H. A. Murthy.
1067
+ A generative model for zero shot learning using conditional
1068
+ variational autoencoders. CVPRW, pages 2269–22698, 2018.
1069
+ 2
1070
+ [37] C. R. Qi, H. Su, K. Mo, and L. J. Guibas. PointNet: Deep
1071
+ Learning on Point Sets for 3D Classification and Segmenta-
1072
+ tion. In IEEE Conference on Computer Vision and Pattern
1073
+ Recognition, pages 652–660, 2017. 2
1074
+ [38] A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh,
1075
+ S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, et al.
1076
+ Learning transferable visual models from natural language
1077
+ supervision. In International Conference on Machine Learn-
1078
+ ing, pages 8748–8763. PMLR, 2021. 2
1079
+ [39] Y. Rao, W. Zhao, G. Chen, Y. Tang, Z. Zhu, G. Huang,
1080
+ J. Zhou, and J. Lu. Denseclip: Language-guided dense pre-
1081
+ diction with context-aware prompting.
1082
+ In Proceedings of
1083
+ the IEEE/CVF Conference on Computer Vision and Pattern
1084
+ Recognition, pages 18082–18091, 2022. 3
1085
+ [40] C. Sautier, G. Puy, S. Gidaris, A. Boulch, A. Bursuc, and
1086
+ R. Marlet. Image-to-Lidar Self-Supervised Distillation for
1087
+ Autonomous Driving Data.
1088
+ In IEEE/CVF Conference on
1089
+ Computer Vision and Pattern Recognition, pages 9891–
1090
+ 9901, 2022. 2, 3, 4, 6, 7
1091
+ [41] H. Tang, Z. Liu, S. Zhao, Y. Lin, J. Lin, H. Wang, and
1092
+ S. Han. Searching Efficient 3D Architectures with Sparse
1093
+ Point-Voxel Convolution. In European Conference on Com-
1094
+ puter Vision, pages 685–702. Springer, 2020. 6
1095
+ [42] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones,
1096
+ A. N. Gomez, Ł. Kaiser, and I. Polosukhin. Attention is all
1097
+ you need. Advances in neural information processing sys-
1098
+ tems, 30, 2017. 4
1099
+ [43] Y. Xian, Z. Akata, G. Sharma, Q. N. Nguyen, M. Hein, and
1100
+ B. Schiele. Latent embeddings for zero-shot classification.
1101
+ CVPR, pages 69–77, 2016. 2
1102
+ [44] J. Xu, R. Zhang, J. Dou, Y. Zhu, J. Sun, and S. Pu. RPVnet:
1103
+ A Deep and Efficient Range-Point-Voxel Fusion Network
1104
+ for Lidar Point Cloud Segmentation. In IEEE International
1105
+ Conference on Computer Vision, pages 16024–16033, Octo-
1106
+ ber 2021. 2
1107
+ [45] X. Yan, J. Gao, C. Zheng, C. Zheng, R. Zhang, S. Cui, and
1108
+ Z. Li. 2DPASS: 2D Priors Assisted Semantic Segmentation
1109
+ on LiDAR Point Clouds. In European Conference on Com-
1110
+ puter Vision, 2022. 2
1111
+ [46] L. Yao, J. Han, Y. Wen, X. Liang, D. Xu, W. Zhang,
1112
+ Z. Li, C. Xu, and H. Xu.
1113
+ Detclip: Dictionary-enriched
1114
+ visual-concept paralleled pre-training for open-world detec-
1115
+ tion. arXiv preprint arXiv:2209.09407, 2022. 3
1116
+ [47] Β΄E. Zablocki, P. Bordes, B. Piwowarski, L. Soulier, and P. Gal-
1117
+ linari. Context-aware zero-shot learning for object recogni-
1118
+ tion. ArXiv, abs/1904.12638, 2019. 2
1119
+ [48] R. Zhang, Z. Guo, W. Zhang, K. Li, X. Miao, B. Cui, Y. Qiao,
1120
+ P. Gao, and H. Li. PointCLIP: Point Cloud Understanding
1121
+ by CLIP. In IEEE/CVF Conference on Computer Vision and
1122
+ Pattern Recognition, pages 8552–8562, 2022. 2, 3
1123
+ [49] C. Zhou, C. C. Loy, and B. Dai. Extract Free Dense Labels
1124
+ from CLIP. In European Conference on Computer Vision,
1125
+ pages 696–712. Springer, 2022. 2, 3, 4, 6
1126
+ [50] X. Zhu, H. Zhou, T. Wang, F. Hong, Y. Ma, W. Li, H. Li,
1127
+ and D. Lin.
1128
+ Cylindrical and Asymmetrical 3D Convolu-
1129
+ tion Networks for Lidar Segmentation. In IEEE Conference
1130
+
1131
+ on Computer Vision and Pattern Recognition, pages 9939–
1132
+ 9948, 2021. 2
1133
+
MNE4T4oBgHgl3EQfKAyT/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
MdFQT4oBgHgl3EQfVTaW/content/2301.13300v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:790bb02ea74c08bdd600da15621e8bc91ee550c789edccb3fb2d7a9fc3bf3dac
3
+ size 477131