Datasets:

marin-community
/

ar5iv-warning-markdown

id stringlengths 27 33	source stringclasses 1 value	format stringclasses 1 value	text stringlengths 13 1.81M
warning/0001/math-ph0001010.html	ar5iv	text	# References Osterwalder–Schrader axioms—Wightman Axioms—The mathematical axiom systems for quantum field theory (QFT) grew out of Hilbert’s sixth problem , that of stating the problems of quantum theory in precise mathematical terms. There have been several competing mathematical systems of axioms, and here we shall deal with those of A.S. Wightman and of K. Osterwalder and R. Schrader , stated in historical order. They are centered around group symmetry, relative to unitary representations of Lie groups in Hilbert space. We also mention how the Osterwalder–Schrader axioms have influenced the theory of unitary representations of groups, making connection with . Wightman’s axioms involve: (1) a unitary representation $`U`$ of $`G:=\mathrm{SL}(2,)^4`$ as a cover of the Poincaré group of relativity, and a vacuum state vector $`\psi _0`$ fixed by the representation, (2) quantum fields $`\phi _1\left(f\right),\mathrm{},\phi _n\left(f\right)`$, say, as operator-valued distributions, $`f`$ running over a specified space of test functions, and the operators $`\phi _i\left(f\right)`$ defined on a dense and invariant domain $`D`$ in $`𝐇`$ (the Hilbert space of quantum states), and $`\psi _0D`$, (3) a transformation law which states that $`U\left(g\right)\phi _j\left(f\right)U\left(g^1\right)`$ is a finite-dimensional representation $`R`$ of the group $`G`$ acting on the fields $`\phi _i\left(f\right)`$, i.e., $`_iR_{ji}\left(g^1\right)\phi _i\left(g\left[f\right]\right)`$, $`g`$ acting on space-time and $`g\left[f\right]\left(x\right)=f\left(g^1x\right)`$, $`x^4`$. (4) The fields $`\phi _j\left(f\right)`$ are assumed to satisfy locality and one of the two canonical commutation relations of $`[A,B]_\pm =AB\pm BA`$, for fermions, resp., bosons; and (5) it is assumed that there is scattering with asymptotic completeness, in the sense $`𝐇=𝐇^{\text{in}}=𝐇^{\text{out}}`$. The Wightman axioms were the basis for many of the spectacular developments in QFT in the seventies, see, e.g., , and the Osterwalder–Schrader axioms came in response to the dictates of path space measures. The constructive approach involved some variant of the Feynman measure. But the latter has mathematical divergences that can be resolved with an analytic continuation so that the mathematically well-defined Wiener measure becomes instead the basis for the analysis. Two analytical continuations were suggested in this connection: in the mass-parameter, and in the time-parameter, i.e., $`t\sqrt{1}t`$. With the latter, the Newtonian quadratic form on space-time turns into the form of relativity, $`x_1^2+x_2^2+x_3^2t^2`$. We get a stochastic process $`𝐗_t`$: symmetric, i.e., $`𝐗_t𝐗_t`$; stationary, i.e., $`𝐗_{t+s}𝐗_s`$; and Osterwalder–Schrader positive, i.e., $`_\mathrm{\Omega }f_1𝐗_{t_1}f_2𝐗_{t_2}\mathrm{}f_n𝐗_{t_n}𝑑P0`$, $`f_1,\mathrm{},f_n`$ test functions, $`\mathrm{}<t_1t_2\mathrm{}t_n<\mathrm{}`$, and $`P`$ denoting a path space measure. Specifically: If $`t/2<t_1t_2\mathrm{}t_n<t/2`$, then (1) $$\begin{array}{c}\mathrm{\Omega }A_1e^{\left(t_2t_1\right)\widehat{H}}A_2e^{\left(t_3t_2\right)\widehat{H}}A_3\mathrm{}A_n\mathrm{\Omega }\hfill \\ \hfill =\underset{t\mathrm{}}{lim}\underset{k=1}{\overset{n}{}}A_k\left(q\left(t_k\right)\right)d\mu _t\left(q()\right).\end{array}$$ By Minlos’ theorem, there is a measure $`\mu `$ on $`𝒟^{}`$ such that (2) $$\underset{t\mathrm{}}{lim}e^{iq\left(f\right)}d\mu _t\left(q\right)=e^{iq\left(f\right)}d\mu \left(q\right)=:S\left(f\right)$$ for all $`f𝒟`$. Since $`\mu `$ is a positive measure, we have $$\underset{k}{}\underset{l}{}\overline{c}_kc_lS\left(f_k\overline{f}_l\right)0$$ for all $`c_1,\mathrm{},c_n`$, and all $`f_1,\mathrm{},f_n𝒟`$. When combining (1) and (2), we note that this limit-measure $`\mu `$ then accounts for the time-ordered $`n`$-point functions which occur on the left-hand side in formula (1). This observation is further used in the analysis of the stochastic process $`𝐗_t`$, $`𝐗_t\left(q\right)=q\left(t\right)`$. But, more importantly, it can be checked from the construction that we also have the following reflection positivity: Let $`\left(\theta f\right)\left(s\right):=f\left(s\right)`$, $`f𝒟`$, $`s`$, and set $$𝒟_+=\{f𝒟f\text{ real valued, }f\left(s\right)=0\text{ for }s<0\}.$$ Then $$\underset{k}{}\underset{l}{}\overline{c}_kc_lS\left(\theta \left(f_k\right)f_l\right)0$$ for all $`c_1,\mathrm{},c_n`$, and all $`f_1,\mathrm{},f_n𝒟_+`$, which is one version of Osterwalder–Schrader positivity. Since the Killing form of Lie theory may serve as a finite-dimensional metric, the Osterwalder–Schrader idea turned out also to have implications for the theory of unitary representations of Lie groups. In , the authors associate to Riemannian symmetric spaces $`G/K`$ of tube domain type, a duality between complementary series representations of $`G`$ on one side, and highest weight representations of a $`c`$-dual $`G^c`$ on the other side. The duality $`GG^c`$ involves analytic continuation, in a sense which generalizes $`t\sqrt{1}t`$, and the reflection positivity of the Osterwalder–Schrader axiom system. What results is a new Hilbert space where the new representation of $`G^c`$ is “physical” in the sense that there is positive energy and causality, the latter concept being defined from certain cones in the Lie algebra of $`G`$. A unitary representation $`\pi `$ acting on a Hilbert space $`𝐇(\pi )`$ is said to be reflection symmetric if there is a unitary operator $`J:𝐇(\pi )𝐇(\pi )`$ such that * $`J^2=\text{id}`$. * $`J\pi (g)=\pi (\tau (g))J,gG`$, where $`\tau \mathrm{Aut}\left(G\right)`$, $`\tau ^2=id`$, and $`H:=\{gG\tau \left(g\right)=g\}`$. A closed convex cone $`C𝔮`$ is hyperbolic if $`C^o\mathrm{}`$ and if $`\mathrm{ad}X`$ is semisimple with real eigenvalues for every $`XC^o`$. Assume the following for $`(G,\pi ,\tau ,J)`$: * $`\pi `$ is reflection symmetric with reflection $`J`$. * There is an $`H`$-invariant hyperbolic cone $`C𝔮`$ such that $`S(C)=H\mathrm{exp}C`$ is a closed semigroup and $`S(C)^o=H\mathrm{exp}C^o`$ is diffeomorphic to $`H\times C^o`$. * There is a subspace $`0𝐊_0𝐇(\pi )`$ invariant under $`S(C)`$ satisfying the positivity condition $$v\text{ }v_J:=v\text{ }J(v)0,v𝐊_0.$$ Assume that $`(\pi ,C,𝐇,J)`$ satisfies (PR1)–(PR3). Then the following hold: * $`S(C)`$ acts via $`s\stackrel{~}{\pi }(s)`$ by contractions on $`𝐊`$ ($`=`$ the Hilbert space obtained by completion of $`𝐊_0`$ in the norm from (PR3)). * Let $`G^c`$ be the simply connected Lie group with Lie algebra $`𝔤^c`$. Then there exists a unitary representation $`\stackrel{~}{\pi }^c`$ of $`G^c`$ such that $`d\stackrel{~}{\pi }^c(X)=d\stackrel{~}{\pi }(X)`$ for $`X𝔥`$ and $`id\stackrel{~}{\pi }^c(Y)=d\stackrel{~}{\pi }(iY)`$ for $`YC`$, where $`𝔥:=\{X𝔤\tau \left(X\right)=X\}`$. * The representation $`\stackrel{~}{\pi }^c`$ is irreducible if and only if $`\stackrel{~}{\pi }`$ is irreducible. Palle E.T. Jorgensen: jorgen@math.uiowa.edu Gestur Ólafsson: olafsson@math.lsu.edu
warning/0001/nucl-th0001031.html	ar5iv	text	# 1 Introduction ## 1 Introduction The success of the shell model in predicting the nuclear magic numbers is related to the presence of a strong spin-orbit term in the nuclear average potential. Few years after the formulation of the nuclear shell–model , evidences of spin-orbit terms in the nuclear interaction were identified by analyzing polarized proton scattering data off complex nuclei . The strong spin–orbit term of the nuclear average potential should be generated by an analogous term present in the nucleon–nucleon interaction. However, the connection between realistic nucleon–nucleon interactions, i.e. those built to reproduce the two nucleon scattering data and the deuteron properties, and the effective interactions, those used in nuclear structure effective theories, is still unclear. In the present article we present a study on the relationship between the spin–orbit terms used in realistic nucleon–nucleon interaction and the analogous terms used in Hartree–Fock (HF) calculations. Our work has been done within the non relativistic framework where the spin–orbit terms can be easily isolated. In effect, in our HF calculations, we have used the explicit spin-orbit terms of the Argonne–Urbana nucleon-nucleon realistic potentials describing, within the non relativistic framework, nucleon–nucleon elastic scattering data up to energies of about 300 MeV. The spin–orbit interactions commonly used in HF calculations are of zero–range type and, in general, they are parametrized following the expressions proposed by Skyrme . Also the Gogny interaction , which has a finite range for all the other channels, uses a Skyrme–like expression for the spin–orbit term. We explicitly develop the expressions for a finite range spin–orbit interaction to be used in HF calculations. As expected, in addition to the direct term these expressions produce a contribution also in the Fock–Dirac exchange term of the HF equations. This term is not present when zero–range interactions are used. Also in the direct term there are some new contributions with respect to the expression obtained with the Skyrme interaction. In the next paragraphs we present the detailed expressions of the HF equations when finite range spin–orbit terms are considered, then we discuss the importance of the finite–range and the effect of using spin–orbit terms taken from realistic nucleon–nucleon interactions. Finally we draw our conclusions. ## 2 The formalism In ref. we made an explicit presentation of the HF formalism with finite–range interactions. In the present article we extend the formalism in order to treat also the spin–orbit terms. For this reason we recall here only those parts of the formalism involving the spin-orbit terms. The effective interaction used in our calculations has the form: $$V(𝐫_1,𝐫_2)=\underset{p=1}{\overset{8}{}}V_p(𝐫_1,𝐫_2)O^p(1,2)$$ (1) where to the first 6 components used in ref. we have added the spin–orbit terms defined as $`O^7(1,2)=𝐋𝐒`$ and $`O^8(1,2)=𝐋𝐒𝝉_1𝝉_2`$ with $`𝐋=(𝐫_1𝐫_2)\times (𝐩_1𝐩_2)`$ and $`𝐒=\frac{1}{2}(𝝈_1+𝝈_2)`$. To take advantage of the spherical symmetry of the problem, we describe the single particle wave functions by separating the angular and the radial parts $`\varphi _k(𝐫)\stackrel{~}{k}(\mathrm{\Omega })u_k(r)/r`$ where the subindex $`k`$ indicates all the quantum numbers necessary to identify the state and $`\mathrm{\Omega }`$ the two angular coordinates $`\theta `$ and $`\varphi `$. The specific expression of the single particle wave functions allows us to reduce the HF equations into a set of differential equations of the type: $$\frac{\mathrm{}^2}{2m_k}\left(\frac{\mathrm{d}^2}{\mathrm{d}r^2}\frac{l_k(l_k+1)}{r^2}\right)u_k(r)+U_k(r)u_k(r)W_k(r)=ϵ_ku_k(r),$$ (2) where we have defined: $$U_k(r)=\underset{i}{}dr^{}u_i^{}(r^{})d\mathrm{\Omega }d\mathrm{\Omega }^{}\stackrel{~}{k}^{}(\mathrm{\Omega })\stackrel{~}{i}^{}(\mathrm{\Omega }^{})V(\|𝐫𝐫^{}\|)\stackrel{~}{k}(\mathrm{\Omega })\stackrel{~}{i}(\mathrm{\Omega }^{})u_i(r^{}),$$ (3) and $$W_k(r)=\underset{i}{}dr^{}u_i^{}(r^{})d\mathrm{\Omega }d\mathrm{\Omega }^{}\stackrel{~}{k}^{}(\mathrm{\Omega })\stackrel{~}{i}^{}(\mathrm{\Omega }^{})V(\|𝐫𝐫^{}\|)\stackrel{~}{i}(\mathrm{\Omega })\stackrel{~}{k}(\mathrm{\Omega }^{})u_i(r)u_k(r^{}).$$ (4) While the interaction depends from the relative distance between two nucleons, the HF equations (2) depend upon the distance of the particles from the origin of the reference system. The implementation of finite range interactions in the HF equations requires the separation of the coordinate variables in the interaction. For the central and tensor channels this separation is done by considering the interaction in coordinate space as Fourier transform of the interaction expressed in momentum space (see ref. for details). For the spin–orbit channels ($`p=7,8`$) we use a different strategy consisting in expanding in multipoles the interaction: $$V_p(r_{12})=4\pi \underset{LM}{}\frac{1}{\widehat{L}^2}𝒱_L^p(r_1,r_2)Y_{LM}^{}(\widehat{r}_1)Y_{LM}(\widehat{r}_2).$$ (5) with $`\widehat{L}\sqrt{2L+1}`$. ¿From the previous equation, making use of the orthogonality of the spherical harmonics, we obtain a close expression for the coefficients of the expansion: $$𝒱_L^p(r_1,r_2)=\frac{\widehat{L}^2}{2}_1^1\mathrm{d}\mathrm{cos}\theta _{12}V_p(r_{12})P_L(\mathrm{cos}\theta _{12}),$$ (6) In the previous equation we have indicated with $`P_L`$ the Legendre polynomials and with $`\theta _{12}`$ the angle between $`𝐫_1`$ and $`𝐫_2`$. The details of the calculations of the spin–orbit matrix elements are given in the Appendix. We obtain for the direct term in the HF equations the following result: $$\left[U_k(r)\right]_{p=7,8}=\mathrm{\hspace{0.17em}2}\pi I_k^pdr^{}r^2\left\{\left[j_k\left(j_k+1\right)l_k\left(l_k+1\right)\frac{3}{4}\right]𝒰_C^p(r,r^{})+𝒰_{LS}^p(r,r^{})\right\},$$ (7) where $$I_k^p=\{\begin{array}{cc}1,\hfill & p\text{=7}\hfill \\ 2t_k,\hfill & p\text{=8},\hfill \end{array}$$ (8) and the potentials $`𝒰_C^p`$ and $`𝒰_{LS}^p`$ are given by: $$𝒰_C^p(r,r^{})=\left[𝒱_0^p(r,r^{})\frac{1}{3}\frac{r^{}}{r}𝒱_1^p(r,r^{})\right]\mathrm{\Omega }_C^p(r^{}),p=7,8$$ (9) and $$𝒰_{LS}^p(r,r^{})=\left[𝒱_0^p(r,r^{})\frac{1}{3}\frac{r}{r^{}}𝒱_1^p(r,r^{})\right]\mathrm{\Omega }_{LS}^p(r^{}),p=7,8.$$ (10) The function $`\mathrm{\Omega }_C^p(r)`$ used in the previous equations has been defined as: $$\mathrm{\Omega }_C^p(r)=\{\begin{array}{cc}\rho (r),\hfill & p\text{=7}\hfill \\ & \\ \rho ^\pi (r)\rho ^\nu (r),\hfill & p\text{=8},\hfill \end{array}$$ (11) where $`\rho ^\pi (r)`$ and $`\rho ^\nu (r)`$ are the proton and neutron densities such as $`\rho ^\pi (r)+\rho ^\nu (r)=\rho (r)`$. The other function used in eq. (10) has been defined as: $$\mathrm{\Omega }_{LS}^p(r)=\{\begin{array}{cc}\rho _{LS}(r),\hfill & p\text{=7}\hfill \\ & \\ \rho _{LS}^\pi (r)\rho _{LS}^\nu (r),\hfill & p\text{=8},\hfill \end{array}$$ (12) where $`\rho _{LS}`$ is the nucleon spin-density, $$\rho _{LS}(r)=\frac{1}{4\pi }\underset{i}{}\left[j_i(j_i+1)l_i(l_i+1)\frac{3}{4}\right]\widehat{j}_i^2\left(\frac{u_i(r)}{r}\right)^2,$$ (13) and $`\rho _{LS}^\pi (r)`$ and $`\rho _{LS}^\nu (r)`$ are the analogous functions for protons and neutrons respectively. For the exchange terms of eq. (2) we obtain: $$\left[W_k(r)\right]_{p=7,8}=\underset{iL}{}I_{ki}^p\underset{\alpha =1,5}{}\epsilon _{kiL}^{(\alpha )}dr^{}𝒲_{kiL}^{p(\alpha )}(r,r^{}),$$ (14) where $$I_{ki}^p=\{\begin{array}{cc}\delta _{t_k,t_i},\hfill & p=7\hfill \\ & \\ 2\delta _{t_k,t_i}+\delta _{t_k,t_i},\hfill & p=8.\hfill \end{array}$$ (15) The new five functions $`\epsilon `$ have been defined as: $$\epsilon _{kiL}^{(\alpha )}=\sqrt{3}(1)^{j_i+l_i+\frac{1}{2}}\widehat{l}_k\widehat{l}_i\widehat{j}_i^2\underset{K}{}(1)^K\widehat{K}^2\left(\begin{array}{ccc}1& K& L\\ 1& 1& 0\end{array}\right)\zeta _{ki}^{(\alpha )}(L,K),\alpha =1,\mathrm{},5,$$ (16) with $`\zeta _{ki}^{(\alpha )}(L,K)`$ given by: $`\zeta _{ki}^{(1)}(L,K)`$ $`=`$ $`\xi (l_k+l_i+L)𝒯_{ki}(L,K)`$ $`\left[\sqrt{l_i(l_i+1)}\left(\begin{array}{ccc}l_i& l_k& K\\ 1& 0& 1\end{array}\right)\sqrt{l_k(l_k+1)}\left(\begin{array}{ccc}l_i& l_k& K\\ 0& 1& 1\end{array}\right)\right]`$ $`\zeta _{ki}^{(\alpha )}(L,K)`$ $`=`$ $`𝒢_{ki}(L,K)\{\begin{array}{cc}(2)\sqrt{l_i(l_i+1)}\left(\begin{array}{ccc}l_i& l_k& K\\ 1& 0& 1\end{array}\right)\left(\begin{array}{ccc}1& K& L\\ 1& 1& 0\end{array}\right),\hfill & \alpha =2\hfill \\ & \\ 2\sqrt{l_k(l_k+1)}\left(\begin{array}{ccc}l_i& l_k& K\\ 0& 1& 1\end{array}\right)\left(\begin{array}{ccc}1& K& L\\ 1& 1& 0\end{array}\right),\hfill & \alpha =3\hfill \\ & \\ (\sqrt{2})\left(\begin{array}{ccc}l_i& l_k& K\\ 0& 0& 0\end{array}\right)\left(\begin{array}{ccc}1& K& L\\ 0& 0& 0\end{array}\right),\hfill & \alpha =4,5.\hfill \end{array}`$ In the previous expressions we have used the following definitions: $`𝒯_{ki}(L,K)`$ $`=`$ $`\left\{\begin{array}{ccc}l_k& \frac{1}{2}& j_k\\ l_i& \frac{1}{2}& j_i\\ K& 1& L\end{array}\right\}\left(\begin{array}{ccc}j_k& j_i& L\\ \frac{1}{2}& \frac{1}{2}& 0\end{array}\right)`$ $`\widehat{l}_k\widehat{l}_i\left\{\begin{array}{ccc}l_k& \frac{1}{2}& j_k\\ l_i& \frac{1}{2}& j_i\\ L& 1& K\end{array}\right\}\left\{\begin{array}{ccc}l_k& l_i& K\\ j_i& j_k& \frac{1}{2}\end{array}\right\}\left(\begin{array}{ccc}l_k& l_i& L\\ 0& 0& 0\end{array}\right),`$ $`𝒢_{ki}(L,K)`$ $`=`$ $`\xi (l_k+l_i+L+1){\displaystyle \underset{L^{}}{}}\xi (L+L^{}+1)\widehat{L^{}}^2\left(\begin{array}{ccc}1& K& L^{}\\ 1& 1& 0\end{array}\right)𝒯_{ki}(L^{},K).`$ In the eqs. (14) we have used five new potentials: $`𝒲_{kiL}^{p(\alpha )}(r,r^{})`$ $`=`$ $`u_i^{}(r^{})𝒱_L^p(r,r^{})u_k(r^{})u_i(r)\{\begin{array}{cc}1,\hfill & \alpha =1,\hfill \\ & \\ \frac{r^{}}{r},\hfill & \alpha =2,\hfill \\ & \\ \frac{r}{r^{}},\hfill & \alpha =3,\hfill \end{array}`$ $`𝒲_{kiL}^{p(\alpha )}(r,r^{})`$ $`=`$ $`\{\begin{array}{cc}r^{}u_i^{}(r^{})u_k(r^{})𝒱_L^p(r,r^{})\frac{\text{d}}{\text{d}r}u_i(r),\hfill & \alpha =4,\hfill \\ & \\ ru_k(r^{})\frac{\text{d}}{\text{d}r^{}}[u_i^{}(r^{})𝒱_L^p(r,r^{})]u_i(r),\hfill & \alpha =5.\hfill \end{array}`$ Like in ref. the numerical solution of eq. (2) has been obtained iteratively using the plane wave expansion method of refs. . The center of mass motion has been considered in its simplest approximation, consisting in inserting the nucleon reduced mass in the hamiltonian. The single particle wave functions used to start the iterative procedure have been generated by a Saxon–Woods potential without spin–orbit and Coulomb terms. Therefore the starting wave functions for spin–orbit partners are the same. ## 3 Results In the same spirit of the work of ref. we are more interested in investigating the validity of the commonly used approximations rather than proposing a new effective interaction to be used in HF calculations. This study has been conducted by adding different kinds of spin–orbit terms to a basic interaction composed by the four central terms of the force. These terms are described as a sum of two gaussians: $$V_p(r)=\underset{i=1}{\overset{2}{}}A_{pi}exp(b_ir^2),$$ (27) with $`p=1,2,3,4`$. The parameters of this part of the interaction, which we call $`B1a`$, are compared in tab. 1 with the parameterization $`B1`$ of Brink and Boeker . The small differences are due to the fact that we have considered the Coulomb interaction and therefore we had to readjust the parameters of the force in order to reproduce the binding energy of <sup>4</sup>He. We added to the $`B1a`$ interaction a finite range spin–orbit term of gaussian form: $$V_7(r_{12})=A_7exp(b_7r_{12}^2).$$ (28) The finite range effects have been investigated by comparing the results obtained with the above interaction with those produced by adding to the $`B1a`$ force a zero range spin-orbit term of the form: $$V_7(r_{12})=A_7\delta ^3(𝐫_\mathrm{𝟏}𝐫_\mathrm{𝟐}).$$ (29) The straightforward insertion of this expression in our formalism gives a contribution exactly equal to zero. The reason of this result can be traced back to the fact that we have developed our expressions using $`𝐋=(𝐫_1𝐫_2)\times (𝐩_1𝐩_2)`$. To get results different from zero for a zero-range spin–orbit interaction we set to zero the quantity $`𝒱_1^p(r,r^{})`$ in eqs. (9) and (10), and after inserting eq. (29) we obtained: $$𝒰_{C,LS}^7(r,r^{})=\frac{A_7}{2r}\mathrm{\Omega }_{C,LS}^7(r^{})\delta (rr^{}).$$ (30) The calculations done with this approach are labelled as $`z`$. In addition to these effective interactions we have also used spin–orbit terms taken from microscopic forces: the Urbana V14 , Argonne V14 , and Argonne V18 potentials. Our study has been restricted to the investigation of the doubly magic nuclei <sup>12</sup>C, <sup>16</sup>O, <sup>40</sup>Ca, <sup>48</sup>Ca and <sup>208</sup>Pb. The finite range interaction (28) has been used to study the role played by the various terms of the spin–orbit potential. In a first set of calculations, labelled as $`c`$, only the $`𝒰_C`$ term of eq. (7) has been used. This is the only spin–orbit term present in shell–model calculations. In another set of calculations, denoted as $`d`$, we considered the full direct term, and, finally, the results identified with $`so`$ have been obtained with all the spin–orbit terms. These calculations have been done by changing every time the parameters of the force (28) to reproduce the 6.3 MeV splitting between the protons 1p levels in <sup>16</sup>O. The parameter $`b_7`$ was fixed to the arbitrary value of 1.2 fm<sup>-2</sup> and the fit of the splitting was obtained by changing $`A_7`$. The values of $`A_7`$ obtained in this way are, -108.75, -107.50, -97.86 MeV for the $`c`$, $`d`$, and $`so`$ calculations respectively. The three interactions do not differ very much as it is shown in the panel I of the figure. This result indicates that the largest contribution from the spin–orbit force is coming from the $`𝒰_C`$ factor of the direct term. Since the other terms are small we have explored the possibility of avoiding their explicit calculation by simulating their effects with a readjustment of the force parameters. This is the reason why each type of calculation has been done with a different parametrization of the force, each of them reproducing the same empirical quantity. In tab. 2 we compare the binding energies obtained with our calculations with the experimental ones . In spite of the fact that we handle with a non–linear problem, the spin–orbit terms acts on the binding energies as expected. The main contribution to the binding energy is obtained by the sum of the single particle energies. In nuclei where all the spin–orbit partners are occupied the spin–orbit term lowers the energy of the $`l+1/2`$ level and increases that of the $`l1/2`$ level, in such a way that the contribution to the nuclear binding energy is almost zero. In table 2 this is observed by looking at the values of the energies of <sup>16</sup>O and <sup>40</sup>Ca which are practically the same, independently from the spin-orbit force used. Clearly those nuclei where not all the spin–orbit partners are occupied are sensitive to the spin–orbit force, since the single particle energy of the last occupied level is lowered. The effect is seen in <sup>12</sup>C, <sup>48</sup>Ca and <sup>208</sup>Pb where the binding energy increases, in absolute value, the stronger the spin-orbit force is. The quantity most sensitive to the spin–orbit interaction is the energy splitting between spin–orbit partners levels. The energy splittings calculated for the various nuclei under investigation with the interactions proposed are compared in tabs. 3 and 4 with the Skyrme III results and with the empirical values . The experimental spectrum is more compressed than the theoretical one. This fact is well known , and it is related to the intrinsic limitations of the HF theory in the description of an interacting many–body system. As expected, the splittings increase with increasing value of $`l`$. The splittings obtained with the zero range interaction $`z`$ become larger than those obtained with finite range interaction as the mass number of the nucleus increases. The results obtained with zero–range Skyrme interaction do not present this effect. In the Skyrme interaction there are velocity dependent terms generating spin–orbit like contributions which add to those produced by the genuine spin–orbit term. These velocity dependent terms simulate the effects of the finite range. We observe that the value of the splittings obtained with the Skyrme III interaction are comparable with those obtained with our finite range interactions. ¿From the comparison of the results of the $`c`$ and $`d`$ columns of tabs. 3 and 4 we infer information on the role of the terms $`𝒰_{LS}^p`$ in eq. (7). The inclusion of $`𝒰_{LS}^p`$ increases the splitting for all the nuclei considered but the magnitude of this increase is rather different for the various nuclei. We should not consider in our analysis the nucleus <sup>16</sup>O since it has been used to fit the interaction. We notice that the addition of $`𝒰_{LS}^p`$ produces quite small differences in the splitting of <sup>40</sup>Ca and <sup>208</sup>Pb nuclei while they are remarkable in <sup>12</sup>C and <sup>48</sup>Ca. These results can be understood by considering that $`𝒰_{LS}^p`$ is related to the nuclear spin density, eq. (13). If we assume that the radial wave functions $`u(r)`$ are the same for spin–orbit partners levels, the contribution of these two levels to the spin density is exactly zero. In real calculations these wave functions are slightly different, but the contribution to the spin density remains small. This explains the small increase of the splitting in <sup>40</sup>Ca and the relatively large modifications produced in <sup>12</sup>C and <sup>48</sup>Ca. One should remark that only the unoccupied levels contribute to the spin density. For this reason the effect of $`𝒰_{LS}^p`$ is relatively large with respect to that of $`𝒰_C^p`$ in <sup>12</sup>C and <sup>48</sup>Ca where the number of single particle levels is relatively small. In a heavy nucleus like <sup>208</sup>Pb there are many levels contributing in $`𝒰_C^p`$ and the effects of $`𝒰_{LS}^p`$ produced by a single level is relatively small. The contribution of the exchange term can be seen by comparing the results of the $`d`$ and $`so`$ columns. The variations with respect to the calculations done with only the direct terms can be as big as 10-15%, but not all of them have the same sign. It seems that for all the $`p`$ states the splitting is reduced when the exchange term is considered, but it is increased in the $`f`$, $`g`$ and $`h`$ states. The situation for the $`d`$ states is even more complicated, since the splitting is reduced for the $`1d`$ states in <sup>208</sup>Pb but it has increased for all the other $`d`$ states. The contribution of the exchange term cannot be taken into account in calculations with the direct term only by modifying the force parameters. The values of the splittings produced by the Urbana (U), and Argonne V<sub>18</sub> interactions are comparable with those of our interactions, while the Argonne V<sub>14</sub> (A14) generates smaller values. The radial dependence of the spin–orbit terms of these interactions are shown in the panel II of the figure. It is remarkable that the results of U and A18 are similar in spite of the large difference in the depth. The depth value of A14 is intermediate between those of the previous two forces, but its splittings are smaller. Th U and A18 forces have similar range, while that of A14 is smaller. These facts indicate that our calculations are more sensitive to the range of the interaction than to its minimum value. In effect we recall that, in our calculations, a zero–range interaction does not produce any splitting. The calculations done with the microscopic interactions include both spin–orbit and spin–orbit isospin terms. In order to study the importance of the isospin part of the spin–orbit interaction we have repeated each calculation leaving out this terms. The differences of the results obtained with the full interaction and those without the isospin part are very small. In order to avoid a long list of numbers we give in tab. 5, for each nucleus under investigation, the minimum, the maximum and the average difference, in absolute value, between the calculated splittings. It appears clear the relatively small importance of this term of the interaction. This fact can be understood considering that the major contribution to the spin–orbit interaction is coming from the direct $`𝒰_C^p`$ term. For the spin-orbit isospin term of the interaction, the case $`p=8`$, the $`𝒰_C^p`$ term contains a function which is given by the difference between the proton and neutron density distributions, eq. (11). In all the nuclei we have considered this difference is small and particularly small in those nuclei having the same number of protons and neutron (<sup>16</sup>O and <sup>40</sup>Ca). In effect the maximum differences are larger in <sup>48</sup>Ca and <sup>208</sup>Pb than in <sup>16</sup>O and <sup>40</sup>Ca. ## 4 Summary and Conclusions In this article we have presented a formalism to treat finite range spin–orbit interactions in HF calculations. The finite range of the interaction generates additional terms with respect to the usual shell model expression. One of these is the contribution to the exchange Fock–Dirac term in the HF equation (2). Also in the direct (Hartree) term of this equation there is a new part: the $`𝒰_{LS}`$ piece of eq. (7). The major goal of our work was the investigation of the effects produced by these new components. This has been done by adding different type of spin–orbit terms to a fixed interaction active only in the four central channels. We have used a spin–orbit interaction of a gaussian form whose parameters have been fixed to reproduce the energy splitting of the proton $`1p`$ levels in <sup>16</sup>O. We have shown that the largest part of the spin–orbit effects in HF calculations is produced by the traditional shell model term, $`𝒰_C`$ in eq. (7). The contribution of the other term, $`𝒰_{LS}`$, is very small and it can be simulated by a redefinition of the parameters of the force. The role of the exchange term is more complicated: its inclusion in the calculations modifies by a maximum of 15% the values of the spin–orbit splittings. The complication arises because these modifications do not have the same sign for all the nuclei studied. In calculations done with only the direct terms, it is not possible to simulate the exchange effects by simply readjusting the parameters of the interaction. Forcing our formalism to handle zero–range spin–orbit interactions we have studied, by comparison, the importance of the finite range. We found that calculations done with zero–range interaction produce energy splittings which, in heavy nuclei, are much larger than the empirical ones. Traditional HF calculations use spin–orbit zero–range terms of Skyrme type . These expressions produce contributions to the hamiltonian which are related to the derivative of the density distribution, while our expressions depend directly from the density distribution. The dependence from the derivative of the density distribution simulate effects produced by the finite range of the force. We have done calculations with spin-orbit terms taken from microscopic interactions and we have obtained splittings close to those produced by our effective spin–orbit interactions. This would indicate that the medium does not affect the spin-orbit term of the realistic interaction, in agreement with the findings of G-matrix calculations . We would like to point out, however, that the observables we have investigated are more sensitive to the global properties of the spin-orbit potential than to its details. Modifications of the local properties of the interaction would not produce effects on our results. We have also investigated the effects of the spin–orbit isospin term of the interaction, and we found them to be very small. These terms are related to the differences between protons and neutrons density and spin–orbit density distributions. In our calculations these quantities are rather small even for a nucleus with a large neutron excess like <sup>208</sup>Pb. There are however indications for observables which seems to be sensitive to this part of the potential . The comparison of the results of our calculations with the empirical values of the splittings on the various nuclei investigated is not satisfactory. The empirical splittings are smaller than those we have obtained, except for <sup>12</sup>C. This is a well known problem of the HF theory, and it could be solved only by using theories going beyond the mean field description of the nucleus. ## 5 Appendix Since we have developed the HF equations in spherical coordinates it is necessary to express the operator $`𝐋𝐒`$ in terms of these coordinates. For this purpose we define an operator $`O(ijk)`$ as: $$O(ijk)(1)^{i+j}𝐫_i\times 𝐩_j𝐬_k=\sqrt{\frac{2\pi }{3}}(1)^{i+j}r_i\underset{\mu }{}(1)^{1\mu }Y_{1\mu }(\widehat{r}_i)\left[_j\sigma (k)\right]_\mu ^1,$$ (31) where we have set $`\mathrm{}=1`$ and the indexes $`i,j,k`$ can assume only two values, $`1`$ and $`2`$ for example. The previous formula has been obtained by expressing $`r`$ in terms of spherical harmonics and by making explicit use of the tensor product properties. Using the above operator we can write the spin–orbit channels of the force as: $`V_p(r_{12})O_p(1,2)`$ $``$ $`V_p(r_{12})^p{\displaystyle \underset{i,j,k=1,2}{}}O(ijk)`$ (32) $`=`$ $`4\pi \sqrt{{\displaystyle \frac{2\pi }{3}}}^p{\displaystyle \underset{L}{}}{\displaystyle \frac{1}{\widehat{L}^2}}𝒱_L^p(r_1,r_2){\displaystyle \underset{jk}{}}(1)^j{\displaystyle \underset{\mu }{}}(1)^{1\mu }\left[_j\sigma (k)\right]_\mu ^1`$ $`{\displaystyle \underset{iM}{}}(1)^{M+i}r_iY_{LM}(\widehat{r}_1)Y_{LM}(\widehat{r}_2)Y_{1\mu }(\widehat{r}_i),p=7,8.`$ The operator $`^p`$ has been defined as: $$^p=\{\begin{array}{cc}1,\hfill & p=7\hfill \\ & \\ 𝝉(1)𝝉(2),\hfill & p=8.\hfill \end{array}$$ (33) In eq. (32) we make the sum on $`M`$ e $`\mu `$, and we obtain a more synthetic expression: $`V_p(r_{12})O_p(1,2)`$ $`=`$ $`2\sqrt{2\pi }^p{\displaystyle \underset{LL^{}}{}}{\displaystyle \frac{\widehat{L^{}}}{\widehat{L}}}\left(\begin{array}{ccc}L& L^{}& 1\\ 0& 0& 0\end{array}\right)𝒱_L^p(r_1,r_2)`$ (37) $`{\displaystyle \underset{ijk}{}}O_{00}^{LL^{}}(ijk),p=7,8,`$ where we have defined new set of operators as: $$O_{00}^{LL^{}}(ijk)=(1)^{i+j}r_i\left[[Y_L^{}(\widehat{r}_i)Y_L(\widehat{r}_{i/})]^1\left[_j\sigma (k)\right]^1\right]_0^0$$ (38) and where we have defined: $$𝐫_{i/}=\{\begin{array}{cc}𝐫_1,\hfill & \text{for }i\text{=2}\hfill \\ 𝐫_2,\hfill & \text{for }i\text{=1}.\hfill \end{array}$$ (39) It is convenient to express these operators such as the coordinates of each particle are separated: $$O_{00}^{LL^{}}(ijk)=\sqrt{3}\underset{KM}{}(1)^{L+M}\left\{\begin{array}{ccc}L& L^{}& 1\\ 1& 1& K\end{array}\right\}\stackrel{~}{O}_{LL^{}K}^M(ijk),$$ (40) where we have used the Racah 6-$`j`$ symbol and we have defined the operators: $$\stackrel{~}{O}_{LL^{}K}^M(ijk)=e(ijk)\overline{A}_{JM}^{\left(ijk\right)}(1)\overline{B}_{JM}^{\left(ijk\right)}(2).$$ (41) The expressions of the three terms of the above equation, for each value of $`(ijk)`$ are given in tab. 6 as functions of the following operators: $`𝒞_{\lambda \mu }^L(i)`$ $`=`$ $`\left[Y_L(\widehat{r}_i)_i\right]_\mu ^\lambda `$ $`𝒮_{\lambda \mu }^{LK}(i)`$ $`=`$ $`\left[\left[Y_L(\widehat{r}_i)_i\right]^K\sigma (i)\right]_\mu ^\lambda \left[𝒞_K^L(i)\sigma (i)\right]_\mu ^\lambda `$ (42) $`_{\lambda \mu }^L(i)`$ $`=`$ $`\left[Y_L(\widehat{r}_i)\sigma (i)\right]_\mu ^\lambda .`$ At the end we express the spin–orbit terms of the interaction as: $`V_p(r_{12})O_p(1,2)`$ $`=`$ $`4\pi ^p{\displaystyle \underset{LL^{}K}{}}f(L,L^{},K)𝒱_L^p(r_1,r_2)`$ $`{\displaystyle \underset{ijk}{}}{\displaystyle \underset{M}{}}(1)^M\stackrel{~}{O}_{LL^{}K}^M(ijk),p=7,8,`$ where $`f`$ is given by: $$f(L,L^{},K)=()^{L+K}\xi (L+L^{}+1)\frac{\widehat{L^{}}}{\widehat{L}}\left(\begin{array}{ccc}1& K& L^{}\\ 1& 1& 0\end{array}\right)\left(\begin{array}{ccc}1& K& L\\ 1& 1& 0\end{array}\right).$$ (43) In the calculation of the HF equations for the spin–orbit channels we have used the results corresponding to the matrix elements $`\stackrel{~}{O}_{LL^{}K}^M(ijk)`$ which in the tab. 6 are shown to be function of $`𝒞_{LM}^K`$, $`_{LM}^K`$, $`𝒮_{LM}^{L^{}K}`$ defined in (5) and of the spherical harmonics $`Y_{LM}(\widehat{r})`$. Using the function $`\xi (l)`$ =1 if $`l`$ is even and =0 if $`l`$ is odd, we express the reduced matrix elements for the spherical harmonics as: $$l\frac{1}{2}jY_Ll^{}\frac{1}{2}j^{}=\frac{1}{\sqrt{4\pi }}(1)^{j^{}+L+\frac{3}{2}}\widehat{j}\widehat{j^{}}\widehat{L}\xi (l+l^{}+L)\left(\begin{array}{ccc}j& j^{}& L\\ \frac{1}{2}& \frac{1}{2}& 0\end{array}\right).$$ (44) For the other three operators we obtain the following operators: $`l{\displaystyle \frac{1}{2}}j𝒮_L^{L^{}K}l^{}{\displaystyle \frac{1}{2}}j^{}`$ $`=`$ $`\sqrt{{\displaystyle \frac{3}{2\pi }}}(1)^l^{}\widehat{j}\widehat{j^{}}\widehat{l}\widehat{l^{}}\widehat{L}\widehat{L^{}}\widehat{K}\left\{\begin{array}{ccc}l& \frac{1}{2}& j\\ l^{}& \frac{1}{2}& j^{}\\ K& 1& L\end{array}\right\}`$ (58) $`\{\sqrt{2}\xi (l+l^{}+L^{}+1)\left(\begin{array}{ccc}l^{}& K& l\\ 1& 1& 0\end{array}\right)\left(\begin{array}{ccc}K& 1& L^{}\\ 1& 1& 0\end{array}\right){\displaystyle \frac{\sqrt{l^{}(l^{}+1)}}{r}}`$ $`+\left(\begin{array}{ccc}l^{}& K& l\\ 0& 0& 0\end{array}\right)\left(\begin{array}{ccc}K& 1& L^{}\\ 0& 0& 0\end{array}\right){\displaystyle \frac{\mathrm{d}}{\mathrm{d}r}}\},`$ $`l{\displaystyle \frac{1}{2}}j𝒞_L^Kl^{}{\displaystyle \frac{1}{2}}j^{}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{4\pi }}}(1)^{l+l^{}+\frac{1}{2}+j^{}+L}\widehat{j}\widehat{j^{}}\widehat{l}\widehat{l^{}}\widehat{L}\widehat{K}\left\{\begin{array}{ccc}l& j& \frac{1}{2}\\ j^{}& l^{}& L\end{array}\right\}`$ (71) $`\{\sqrt{2}\xi (l+l^{}+K+1)\left(\begin{array}{ccc}l^{}& L& l\\ 1& 1& 0\end{array}\right)\left(\begin{array}{ccc}L& 1& K\\ 1& 1& 0\end{array}\right){\displaystyle \frac{\sqrt{l^{}(l^{}+1)}}{r}}`$ $`+\left(\begin{array}{ccc}l^{}& L& l\\ 0& 0& 0\end{array}\right)\left(\begin{array}{ccc}L& 1& K\\ 0& 0& 0\end{array}\right){\displaystyle \frac{\mathrm{d}}{\mathrm{d}r}}\},`$ $`l{\displaystyle \frac{1}{2}}j_L^Kl^{}{\displaystyle \frac{1}{2}}j^{}`$ $`=`$ $`\sqrt{{\displaystyle \frac{3}{2\pi }}}(1)^l\widehat{j}\widehat{j^{}}\widehat{l}\widehat{l^{}}\widehat{L}\widehat{K}\left\{\begin{array}{ccc}l& \frac{1}{2}& j\\ l^{}& \frac{1}{2}& j^{}\\ K& 1& L\end{array}\right\}\left(\begin{array}{ccc}l& K& l^{}\\ 0& 0& 0\end{array}\right).`$ (77)
warning/0001/quant-ph0001018.html	ar5iv	text	# Quantum computation with mesoscopic superposition states ## I Introduction Quantum Mechanics is now fundamental to the modern world we live and interact with, not being just the abstract realm of theoretical physics. Many new areas of emerging technology depend on the principles contained within it . One of the most striking features of quantum systems are superposition states. They have given rise to a large amount of discussion in the literature and now play a central role for the recent developments made in the area of quantum information. This is due to their possibility to encode information in a way impossible to be attained by any classical system. Quantum computation has become a significant subject within quantum information theory, due to the powerful property of superposition states to execute large parallel processing. Quantum information research has also improved significantly the understanding about the quantum systems involved on the factual realisation of a quantum computer and has raised many interesting problems such as in the encoding of information , entanglement of states and quantum cryptography . A number of core technologies are currently under investigation for constructing a quantum computer which is necessary to fully implement quantum algorithms. These include ion-traps, cavity QED, solid state and liquid state NMR to name but a few. The proposals to engineer a quantum computer or as a first step a single logic gate in the realm of quantum optics are generally based on discrete atomic states and cavity field number states of zero and one photons. A central proposal which has gained much attention in recent years is the Cirac and Zoller trapped ions scheme to encode a n-conditional gate. We also cite the proposals of Sleator and Weinfurter and of Domokos et al. based on cavity-QED (quantum electrodynamics) technology and dealing with two-bit universal gates. Experimentally, there are few initiatives for logical operations in ion traps and in NMR , which allow for a scalable implementation. These proposals require a technological domain, which to date has not been attained . In cavity-QED technology, for optical frequencies, a conditional interaction between two-modes, the idler and pump, have been proposed to encode a phase gate (P-gate) due to the high non-linearity that can be presented by single atoms. At microwave frequencies, logical elements have been demonstrated experimentally as a means of encoding a quantum memory with a single photon . In this article we are going to focus on cavity QED and the technology associated with it. Cavity QED has had a very rich past and has been instrumental in a huge amount of fundamental quantum and atom optical research . Such a system has been used for photon number quantum non demolition measurements , generation of single Fock state and generation and measurement of the time of decoherence of Schrödinger cats states . With such a rich history recent attention in cavity QED has been focused on quantum information. With the non demolition measurement of a single photon number in the cavity , the technology became available to encode qubits and realise a quantum gate . The quantum information proposals based on cavity QED technology makes use of only zero and one field number states. More recently there is the significant evidence of generation of trapped states of more than one photon which could be used in an encoding scheme. With a CNot gate based on an encoding scheme using zero and one Fock states, spontaneous errors have a disastrous effect. Quantum information is irreversibly lost. It is possible to protect the system against such errors. In fact to protect the qubit against general one qubit errors it is necessary to encode the original state by distributing its quantum information over at least five qubits. Basically the 5-qubit quantum circuit takes the initial state with four extra qubits in the state $`\|0`$ to an encoded state. This state is then protected versus all single qubit errors. Decoding this state and then applying a simple unitary transformation yields the original state. Implementing a five qubit error correcting code is quite expensive in terms of quantum resources. Other encoding schemes may allow simpler error correction circuits. There is no fundamental reason to restrict oneself to physical systems with two dimensional Hilbert spaces for the encoding. It may be more natural in some contexts to encode logical states as a superposition over a large number of basis states. Significant advances can be achieved. For instance in the protection against errors incoming due to the coupling of the qubit system to a dissipative environment. Recent work by Cochrane et. al. have proposed how macroscopically distinct quantum superposition states (Schrödinger cat states) may be used as logical qubit encoding. Spontaneous emission causes a bit-flip error in these superposition state qubit encoding, which is easily corrected by a standard 3-qubit error correction circuit (compared to five qubits for Fock states). This is particularly relevant, as the bit-flip error is much easier to fix than spontaneous emission errors in Fock state systems. Another good reason for using superposition of coherent states to encode qubits, is that they are naturally generated in any cavity system, while number states of more than one photon require a large amount of control . In this paper we propose how even and odd mesoscopic coherent superposition of states can be used to implement and encode a CNot quantum gate in a realistic superconducting cavity-QED system, where those states were already generated . We define the even cat state as the $`0`$ qubit and the odd cat state as the $`1`$ qubit. This encoding can be represented as $`\|0_L`$ $``$ $`{\displaystyle \frac{1}{N_+}}\left(\|\alpha +\|\alpha \right),`$ (1) $`\|1_L`$ $``$ $`{\displaystyle \frac{1}{N_{}}}\left(\|\alpha \|\alpha \right).`$ (2) where $`N_\pm =\sqrt{2(1\pm e^{2\left\|\alpha \right\|^2})}`$. This normalisation is important and will be retained throughout the paper. Given the generation of the two logic qubits how does one implement a quantum gate in cavity QED. Essentially any two-bit quantum gate is universal . One of these universal quantum gates is the control not gate (CNot) and consist of a conditional gate - here if the control bit is 0 the target bit will be maintained, but if the control bit is 1 the target bit will suffer a flip transform to 0. The CNot gate can be engineered by two Hadamard transforms plus a phase (P) transform . The Hadamard transform is a single qubit operation that leads to a rotation in the state while the P-transform is a conditional two-bits transform necessary to identify the state of the control bit. The question posed here is how to identify these Hadamard and P transforms in a realizable physical cavity QED system when the encoding for the qubits is in terms of odd and even cat states. To begin this paper we show how the apparatus similar to the one used to generate Schrödinger cat field states can be generalised to perform a CNot gate conditional transform involving two levels of a Rydberg atom and the field mesoscopic superposition state. Here the two levels of a Rydberg atom are considered to encode the controlled (or target) bit and the field cat state will be the control bit. Since the generation of Schrödinger even and odd cat field states in cavity QED experiments is dependent of a conditional measurement , giving a random outcome, we propose in Section III a strategy based on resonant atomic feedback which allow us to definitely prepare the state of the control bit. The essence of this proposal involves using a feedback scheme based on the injection of appropriately prepared atoms. Basically the state of the cavity is monitored indirectly via the detection of atoms that have interacted dispersively with it. If the cavity field state is not in the required state, a photon is injected into the cavity. Finally in the last section of this paper we present a reasonable detailed discussion of dissipation and their effect on the CNot gate. We explicitly discuss the advantages of encoding with superposition states over zero one photon number states used in previous proposals . Attention is focused on the decoherence phenomenon, as this is one of the main difficulties for quantum computation. ## II Superposition State Encoding In the last few years a great amount of experimental progress in cavity QED has enabled work at the level of single atoms and single photons, where only two electronic energy states of Rydberg atoms participate in the exchange of a photon with the cavity . This has enabled cavity QED technology to be responsible for a large number of interesting experiments showing, the generation of mesoscopic coherent superposition field states, called Schrödinger cat states , the decoherence phenomenon and non-local entanglement of quantum systems . These systems have gained much attention due to the quantum non demolition (QND) property of measurement on the field photon number by atomic interferometry . Our experimental proposal is based on the cavity-QED scheme to generate the field superposition states and is depicted schematically in Fig.(1). It consists of a Rydberg atom beam crossing three cavities, R$`{}_{}{}^{\varphi }{}_{1}{}^{}`$, C and R$`{}_{}{}^{\theta }{}_{2}{}^{}`$. Here R$`{}_{}{}^{\varphi }{}_{1}{}^{}`$ and R$`{}_{}{}^{\theta }{}_{2}{}^{}`$ are Ramsey zones and C is a superconducting Fabry-Perot cavity of high quality factor. To achieve our desired encoding the atoms are initially prepared at B in circular states of quantum principal number of the order of 50. Such atoms are well suited for this scheme since their lifetime is over $`3\times 10^2`$s . The R$`{}_{}{}^{\varphi }{}_{1}{}^{}`$ and R$`{}_{}{}^{\theta }{}_{2}{}^{}`$ cavities, where classical fields resonant with an atomic $`\|g`$ $``$ $`\|e`$ transition (51.099 GHz) are injected during the time of interaction with the atoms, constitute the usual setup for Ramsey interferometry . There, for a selected atomic velocity, the state of the atom will suffer a rotation in the vector space spanned by $`\{\|e,\|g\}`$. The experiment is started when one selects the initial state of an atom prepared in the $`\|g`$ or $`\|e`$ by the laser field L. This atom has a resonant interaction with the field in R$`{}_{}{}^{\varphi }{}_{1}{}^{}`$ given by $$H_I=\mathrm{}\mathrm{\Omega }\left(a_r\sigma ^++a_r^{}\sigma ^{}\right)$$ (3) where $`\sigma ^+\|eg\|`$ and $`\sigma ^{}\|ge\|`$ are the atomic pseudo-spin Pauli operators, $`a_r^{}`$ ($`a_r`$) are the creation (annihilation) operator for the mode of the field in R$`{}_{}{}^{\varphi }{}_{1}{}^{}`$ and $`\mathrm{\Omega }`$ is the one photon Rabi frequency. With a proper choice of the field phase $`\varphi `$ in R$`{}_{}{}^{\varphi }{}_{1}{}^{}`$ the atomic states $`\|g`$ and $`\|e`$ are rotated to $`\|g`$ $``$ $`{\displaystyle \frac{1}{\sqrt{2}}}\left(\|g+e^{i\varphi }\|e\right)`$ (4) $`\|e`$ $``$ $`{\displaystyle \frac{1}{\sqrt{2}}}\left(\|ee^{i\varphi }\|g\right)`$ (5) The cavity C is tuned near the resonance of the transitions between the atomic states $`\|e`$ and $`\|i`$, a reference state corresponding to the higher level from $`\|e`$. The frequency of the transition $`\|e`$ $``$$`\|i`$ is 48.18 GHz and is distinct of any transition involving the level $`\|g`$. The mode geometry inside the cavity is configured in such a way that the intensity of the field rises and decreases smoothly through with the atomic trajectory inside C. For sufficiently slow atoms and for sufficiently large cavity mode detuning from the $`\|e`$ $``$$`\|i`$ frequency transition, the atom-field evolution is adiabatic and no photonic absorption or emission occurs . On the other hand, dispersive effects emerge \- an atom in the state $`\|e`$ crossing C induces a phase shift in the cavity field which can be adjusted by a proper selection of the atomic velocity ($`100`$ m/s) . For a $`\pi `$ phase shift the coherent field $`\|\alpha `$ in C transforms to $`\|\alpha `$. On the other hand, the phase shift caused by an atom in the $`\|g`$ state is null. The atom field interaction can be written effectively as $$H_{off}=\mathrm{}\mathrm{\Omega }_2a^{}a\sigma ^+\sigma ^{}$$ (6) where $`\mathrm{\Omega }_2`$ is the effective Rabi frequency for the interaction of the atom with the field and $`a^{}`$ ($`a`$) is the creation (annihilation) operator for the field in C. After the atomic interaction with the field in C, the atom crosses the second Ramsey zone $`R_2^\theta `$ which introduces a new rotation in the atomic vector space, analogously to Eq. (4) and (5), but for the phase $`\theta `$. The atomic state is detected in D by an ionization zone detector, instantaneously giving the atomic state and the field state in C. This is due to the entanglement of their states. The important point we emphasise here is that the resonant interaction of the Ramsey zones can be used as Hadamard transform since they induce rotations in the vector space of the target bit (atomic state) and the off-resonant interaction between atom and field in C can be used for the P-transform. We begin the description of the implementation of the CNot gate by specifying that the coherent field state will be responsible for the encoding of the control bit and the atomic states $`\|g`$ and $`\|e`$ will be the target bits $`\|0_T`$ and $`\|1_T`$, respectively. The procedures to implement the CNot gate is described as follows. The laser field L prepares the target bit in $`\|g`$ or $`\|e`$; a one bit Hadamard transform is applied to the target qubit by the first Ramsey zone R$`{}_{}{}^{\varphi }{}_{1}{}^{}`$; then the two-bit P-gate is realized by the off-resonant atom-field interaction in C and the second Hadamard transform is realized by R$`{}_{}{}^{\theta }{}_{2}{}^{}`$. Finally the atom is detected simultaneously specifying the atomic and field states. The effective unitary operator related to the evolution of the atom-field in cavity C entangled state, due to the sequential interaction of the atom with the field in $`R_1^\varphi `$, C and $`R_2^\theta `$ is given by $$U(\varphi ,\theta )=U_2^\theta \mathrm{exp}\left[i\mu a^{}a\sigma ^+\sigma ^{}\right]U_1^\varphi $$ (7) where $`U_1^\varphi `$ and $`U_2^\theta `$ are the unitary operators related to the evolution of the joint state in $`R_1^\varphi `$ and $`R_2^\theta `$, respectively. In eqn. (7), $`\mu =\mathrm{\Omega }_2t`$, where $`t`$ is the time interval for the off-resonant interaction. Proceeding through the immediate states generated by the atomic passing through each of the cavities it is easy to show, for $`\varphi =\pi `$ and $`\theta =0`$, the following table \| Input \| $`R_1^\varphi `$ \| \| C \| \| $`R_2^\theta `$ \| Output \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| $`\|g\|0_L`$ \| $``$ \| $`\frac{1}{\sqrt{2}}\left(\|g\|e\right)\|0_L`$ \| $``$ \| $`\frac{1}{\sqrt{2}}\left(\|g\|e\right)\|0_L`$ \| $``$ \| $`\|g\|0_L`$ \| \| $`\|e\|0_L`$ \| $``$ \| $`\frac{1}{\sqrt{2}}\left(\|e+\|g\right)\|0_L`$ \| $``$ \| $`\frac{1}{\sqrt{2}}\left(\|e+\|g\right)\|0_L`$ \| $``$ \| $`\|e\|0_L`$ \| \| $`\|g\|1_L`$ \| $``$ \| $`\frac{1}{\sqrt{2}}\left(\|g\|e\right)\|1_L`$ \| $``$ \| $`\frac{1}{\sqrt{2}}\left(\|e+\|g\right)\|1_L`$ \| $``$ \| $`\|e\|1_L`$ \| \| $`\|e\|1_L`$ \| $``$ \| $`\frac{1}{\sqrt{2}}\left(\|e+\|g\right)\|1_L`$ \| $``$ \| $`\frac{1}{\sqrt{2}}\left(\|g\|e\right)\|1_L`$ \| $``$ \| $`\|g\|1_L`$ \| which verifies the standard CNot truth-table. Above we have discussed a setup where the atoms encode the target qubit and the cavity field mode encodes the control qubit. Nevertheless, it is also possible to proceed with atoms responsible by both the control and target qubit. In this second case, the state of the control atom must be transferred to the cavity C and with a proper selection of the cavity state (to what we address to the next section) the procedure for implementing the CNot gate follows as above. After the second atom, which encodes the target qubit interaction in the process described above, a third atom is sent across the system to read the cavity state in a process similar to the scheme already proposed by Sleator and Weinfurter. To envisage a quantum network, i.e., the interconnection of quantum gates, the carriers of qubits between gates can be achieved by atoms transferring the state of one cavity to another , or even by the coupling of these cavities by superconducting wave-guides which can be responsible by an exchange of states between two gates. It is important for this proposal to include a brief discussion of the realistic parameters. We first note that an atom crosses the cavity in a time of order of 10<sup>-4</sup> s, which is well below the relaxation time of the field inside C (typically of the order of 10<sup>-3</sup>-10<sup>-2</sup>s for Niobium superconducting cavities ) and below the atomic spontaneous emission time of (3$`\times 10^2`$s) . Therefore, the limits considered in that proposal must be far away from the problematic limits found in those experiments. Our entire proposal for encoding a CNot gate discussed here is reliant on being able to generate the zero ($`\|0_L`$) and one ($`\|1_L`$) logical states. For this reason we address in Section (III) a strategy for guaranteeing the exact choice of the initial cavity field state. Without such a strategy, the logical states can only theoretically be generated with a $`50\%`$ probability. More explicitly there is a $`50\%`$ probability that the $`\|0_L`$ state actually contains only even photon number states and a $`50\%`$ probability that it contains only odd photon number states. ## III Initial Conditions for the Control Bit Our generation of the CNot gate outlined in the previous section relies on our ability to be able to generate the coherent logical state encoding with a high degree of certainty. The initial state of the control bit (the field state of the cavity) has to be prepared with a probability greater than 50% as usually occurs in the preparation of superposition field states by Rydberg atoms. The state of the field in the cavity is $`\|0_L`$ or $`\|1_L`$ conditioned by the measurement of the atomic $`\|g`$ or $`\|e`$ state in the process of generation of superposition states. Such a scheme is analogous to the depicted in fig.(1), however here we have $`\theta =\pi `$ in the second Ramsey zone and for the initial cavity state a coherent one, considering that the atom was prepared in the $`\|e`$ state. Let us suppose we are interested in preparing the state $`\|0_L`$ for the control bit. If the atomic state $`\|e`$ was detected, then our scheme would have failed. For it to succeed we have to apply a process conditioned to the measurement of the atomic $`\|e`$ state to guarantee the flip of the cavity field state from $`\|1_L`$ to $`\|0_L`$. Analogously we have to apply a process conditioned to the measurement of the atomic $`\|g`$ state to guarantee the flip of the cavity field state from $`\|0_L`$ to $`\|1_L`$ if we are interested in prepare the control bit in the $`\|1_L`$ state. First noting the fact that an atom interacting resonantly with the field in C, with a controlled velocity, can exchange a single photon and regarding that a single photon emission by the cavity field causes $`a\|0_L`$ $`=`$ $`\alpha {\displaystyle \frac{N_{}}{N_+}}\|1_L\alpha \|1_L(\alpha \mathrm{large})`$ (8) $`a\|1_L`$ $``$ $`\alpha \|0_L.`$ (9) We can now formulate an atomic feedback scheme that operates whenever the atomic detector clicks, if we are interested in the control qubit $`\|0_L`$ or $`\|1_L`$. In fact this process is very similar to the stroboscopic feedback proposed by Vitali et al. for the suppression of decoherence of superposition field states. Of course we do not need a stroboscopic action, but just one event conditioned to the atomic state measurement. The scheme proposed is depicted schematically in fig.(2), where B<sub>2</sub> is a source of atoms which are tuned in resonance with the field in C by the Stark shift conditioned to the atomic state measurement made in the ionization zones D<sub>e</sub> or D<sub>g</sub>. The resonant atom-field interaction is given by the Hamiltonian $$H_I=\mathrm{}\mathrm{\Gamma }\left(a\sigma _f^++a^{}\sigma _f^{}\right)$$ (10) where $`\mathrm{\Gamma }`$ is the coupling constant between the field and atomic variables. Here $`\sigma _f^+`$ and $`\sigma _f^{}`$ are rising and lowering operator for the feedback atom. If the feedback atom is prepared in the state $`\|e`$ then the field state is given by $`\rho _f^{\left(\genfrac{}{}{0pt}{}{g}{e}\right)}`$ $`=`$ $`\mathrm{cos}(\mathrm{\Gamma }\tau \sqrt{a^{}a+1})\rho _C^{\left(\genfrac{}{}{0pt}{}{g}{e}\right)}\mathrm{cos}(\mathrm{\Gamma }\tau \sqrt{a^{}a+1})+a^{}{\displaystyle \frac{\mathrm{sin}(\mathrm{\Gamma }\tau \sqrt{a^{}a+1})}{\sqrt{a^{}a+1}}}\rho _C^{\left(\genfrac{}{}{0pt}{}{g}{e}\right)}{\displaystyle \frac{\mathrm{sin}(\mathrm{\Gamma }\tau \sqrt{a^{}a+1})}{\sqrt{a^{}a+1}}}a.`$ (11) where $`\rho _C^{\left(\genfrac{}{}{0pt}{}{g}{e}\right)}`$ is the density operator associated with the field state in C before the feedback action. Here $`\rho _f^g`$ ($`\rho _f^e`$) explicitly is the ground (excited) state density operator. $`\tau `$ is the time of interaction of the feedback atom with the field. As a measure of the field state in the cavity a second atom is sent through the setup and again measured in D<sub>g</sub> or D<sub>e</sub> . The conditional probability $`P^{\left(\genfrac{}{}{0pt}{}{g}{e}\right)}(T)`$ that the second atom will be detected in the $`\|g`$ or $`\|e`$ state, at the time $`T`$ after detection of the first atom follows $`P^{\left(\genfrac{}{}{0pt}{}{g}{e}\right)}(T)`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left\{1\pm {\displaystyle \frac{1}{1+\mathrm{cos}\phi \text{e}^{2\left\|\alpha \right\|^2}}}\left[\text{e}^{2\left\|\alpha \right\|^2\text{e}^{\gamma T}}+\mathrm{cos}\phi \text{e}^{2\left\|\alpha \right\|^2\left(1\text{e}^{\gamma T}\right)}\right]\right\},`$ (12) conditioned to $`\phi =0`$ \[$`\pi `$\] if the first atom is detected in the $`\|g_1`$ \[$`\|e_1`$\] state and to the signal + \[-\] for the second atom be detected in the $`\|g_2`$ \[$`\|e_2`$\] state. For the computation of Eq. (12) at time T we have included the relaxation of the field state due to dissipation. Considering a reservoir at zero temperature, this state is now given by $`\rho _C^{\left(\genfrac{}{}{0pt}{}{g}{e}\right)}(T)`$ $`=`$ $`{\displaystyle \frac{1}{N_\pm ^2}}\{\|\alpha \text{e}^{\gamma T/2}><\alpha \text{e}^{\gamma T/2}\|+\|\alpha \text{e}^{\gamma T/2}><\alpha \text{e}^{\gamma T/2}\|`$ (13) $`\pm `$ $`\text{e}^{2\|\alpha \|^2(1\text{e}^{\gamma T})}[\|\alpha \text{e}^{\gamma T/2}><\alpha \text{e}^{\gamma T/2}\|+\|\alpha \text{e}^{\gamma T/2}><\alpha \text{e}^{\gamma T/2}\|]\}.`$ (14) where $`\gamma `$ is the relaxation constant of the field. By analyzing Eq.(12) we observe that if the second atom is detected instantaneously after the first one ($`\gamma T1`$) then $$P^{\left(\genfrac{}{}{0pt}{}{g}{e}\right)}(T)=\frac{1}{2}\left[1\pm \mathrm{cos}\phi \right],$$ (15) again with $`\phi =0`$ $`(\pi )`$. This gives the conditional probability of detection of the first atom in $`\|e_1`$ and the second atom in $`\|e_2`$ as $`P(e,e)P^{\left(\genfrac{}{}{0pt}{}{g}{e}\right)}\|^{\pi ,}=1`$ and the probability of detection of the first in $`\|g_1`$and the second in $`\|e_2`$ as $`P(g,e)P^{\left(\genfrac{}{}{0pt}{}{g}{e}\right)}\|^{0,}=0`$. Analogously $`P(g,g)P^{\left(\genfrac{}{}{0pt}{}{g}{e}\right)}\|^{0,+}=1`$ and $`P(e,g)P^{\left(\genfrac{}{}{0pt}{}{g}{e}\right)}\|^{\pi ,+}=0`$. This is a signature of the measurement for the state in which the cavity field was prepared. It presents our undesired results $`P(g,e)`$ and $`P(e,g)`$ equal to zero, that is no probability of them occurring. However if the feedback loop is taken in to account in the calculation of the probabilities $`P(g,e)`$ and $`P(e,g)`$ then, instead of using $`\rho _C^{\left(\genfrac{}{}{0pt}{}{g}{e}\right)}`$ in Eq. (12) we must use $`\rho _f^{\left(\genfrac{}{}{0pt}{}{g}{e}\right)}(T+\tau )`$ from Eq. (11), where $`T^{}=T+\tau `$. Substituting Eq. (14) for the field relaxation into Eq. (11) it follows $`P_f^{\left(\genfrac{}{}{0pt}{}{g}{e}\right)}(T+\tau )`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left\{1\pm {\displaystyle \frac{2}{N^2}}\text{e}^{\left\|\alpha \right\|^2}{\displaystyle \underset{m}{}}{\displaystyle \frac{\left(\left\|\alpha \right\|^2\text{e}^{\gamma T}\right)^m}{m!}}\left[1+(1)^m\mathrm{cos}\phi \text{e}^{2\left\|\alpha \right\|^2\left(1\text{e}^{\gamma T}\right)}\right]\mathrm{cos}\left(2\mathrm{\Gamma }\tau \sqrt{m+1}\right)\right\},`$ (16) which accounts for the conditional probability of detection of the second atom in the state $`\|g_2`$ \[$`\|e_2`$\] if the first atom was detected in the state $`\|e_1`$ \[$`\|g_1`$\]. In Fig. (3) we show the respective four conditional probability of atomic detection, $`P(e,e)`$ and $`P(g,g)`$ without feedback and $`P_f(e,g)`$ and $`P_f(g,e)`$ considering the feedback loop for some values of $`\mathrm{\Gamma }\tau `$. This shows the feasibility of the feedback to control the initial state of the cavity. The figure is plotted until $`\gamma T=1`$ since there is no reason to consider times longer than this once the decoherence of the state has already taken place. In fact the scale of time to be taken into account in figures 3(c) and 3(d) is $`T^{}=T+\tau `$, the time interval after the detection of the first atom plus the time interval of the feedback atom. In these figures the continuous solid line represents the absence of feedback. As can be seen there is an optimum value for the feedback process at $`\mathrm{\Gamma }\tau =\pi /6`$ which gives a 93% of chance for the cavity field qubit to be prepared in the right state. It also must be noted that an optimal value is possible only when the feedback atom is sent instantaneously after the click of the respective detector. The performance of the setup decreases considerably when a time delay exists, as can be observed in Figs. 3(c) and 3(d) for $`\gamma T^{}0.1`$. The limit of those curves around 0.5 means that the field state already decohered, and so there is 50% of chance again for generation of the $`\|0_L`$ or $`\|1_L`$ states. For $`\gamma T^{}>1.0`$ (not shown in the figures) the effect of dissipation implies amplitude damping. The field asymptotically tends to be in a vacuum state, and when this occurs is easily shown through Eq. (16) that the second atom will always be detected in the $`\|g`$ state. With the feedback it tends to always be detected in $`\|e`$ state. ## IV Efficiency and Sources of Error This section discusses in detail the advantages and disadvantages of encoding qubits in superposition states instead of number states of only one and zero photon and the effect of dissipation on these. As is already well known for cavity QED experiments the dominant source of error that will affect the implementation of quantum logic elements is cavity damping. Since the cavities are not isolated, when the states $`\|0_L`$ or $`\|1_L`$ are constructed, the presence of dissipative effects will alter the free evolution of the cavity field state introducing amplitude damping as well as coherence loss. The zero temperature master equation describing the bosonic damping is simply $$\frac{d\rho }{dt}=\frac{\gamma }{2}\left(2a\rho a^{}a^{}a\rho \rho a^{}a\right),$$ (17) and its solution for any initial state can be written as $$\rho (t)=\underset{k=0}{\overset{\mathrm{}}{}}\mathrm{{\rm Y}}_k(t)\rho (0)\mathrm{{\rm Y}}_k^{}(t),$$ (18) where $$\mathrm{{\rm Y}}_k(t)=\underset{n=k}{\overset{\mathrm{}}{}}\sqrt{\left(\genfrac{}{}{0pt}{}{n}{k}\right)}\left(e^{\gamma t}\right)^{(nk)/2}\left(1e^{\gamma t}\right)^{k/2}\|nkn\|.$$ (19) We are interested in the effect of dissipation on the information encoded in the qubits. For that we will consider first a superposition of Schrödinger cats qubits and thereafter a superposition of one and zero photon number states qubit encoding. The action of a single decay event $`\mathrm{{\rm Y}}_1`$ on the state $$\|\psi _1=E_1\|0_L+E_2\|1_L,$$ (20) leads to $$\mathrm{{\rm Y}}_1\|\psi _1=\alpha \left(1e^{\gamma t}\right)^{1/2}e^{\|\alpha \|^2\left(1e^{\gamma t}\right)/2}\left(E_1\frac{N_{}}{N_+}\|\stackrel{~}{1}_L+E_2\frac{N_+}{N_{}}\|\stackrel{~}{0}_L\right),$$ (21) that is, a simple bit-flip occurs. Here $`\|\stackrel{~}{0}_L\frac{1}{N_+}\left(\|e^{\gamma t/2}\alpha +\|e^{\gamma t/2}\alpha \right)`$ and $`\|\stackrel{~}{1}_L\frac{1}{N_{}}\left(\|e^{\gamma t/2}\alpha \|e^{\gamma t/2}\alpha \right)`$, account for the amplitude damping. A simple unitary process will transform Eq. (21) back to Eq. (20), meaning the reversibility of the process. Under a double decay event $`\mathrm{{\rm Y}}_2`$, $$\mathrm{{\rm Y}}_2\|\psi _1=\alpha ^2\left(1e^{\gamma t}\right)e^{\|\alpha \|^2\left(1e^{\gamma t}\right)/2}\left(E_1\|\stackrel{~}{0}_L+E_2\|\stackrel{~}{1}_L\right),$$ (22) which is exactly our initial state but with amplitude damping. This special superposition is invariant under even number of decay events. This fact brings one important information about these states. However a single decay event, $`\mathrm{{\rm Y}}_1`$ on the Fock superposition state $$\|\psi _2=F_1\|0+F_2\|1,$$ (23) leads to $$\mathrm{{\rm Y}}_1\|\psi _2=F_2\left(1e^{\gamma t}\right)^{1/2}\|0.$$ (24) No unitary operation can recover (23) indicating the irreversibility of the process. This means that in one photon state information processing schemes, one single photon decay is fatal, since there is no way in which the resulting error can be corrected once it occurs. However for qubits consisting of superpositions of odd and even number states, one decay event cause a bit-flip, which could be, in principle be corrected. So here, we classified two different kind of error arising from dissipation, one impossible to be corrected (called irreversible error) and the other a bit-flip which can be corrected (reversible error) by unitary processes. There is a number of error correction schemes that protect the quantum information against single errors. As we have mentioned previously a spontaneous emission error for the Schrödinger cat encoding results in a bit-flip. It is well known that such errors can easily be prevented by a 3-qubit error correction circuit (schematically depicted in Fig (4a)). This circuit is reasonably simple and the superposition state it produces is relatively simple. In fact for an arbitrary qubit $`\|\psi =E_1\|0_L+E_2\|1_L`$ the correction circuit generates the encoded superposition state $`\|\psi `$ $`=`$ $`E_1\|000+E_2\|111.`$ (25) To protect against arbitrary error requires normally a 5-qubit error correction circuit (schematically depicted in Fig (4b)). For an arbitrary qubit $`\|\psi =F_1\|0+F_2\|1`$, the correction circuit generates the superposition state $`\|\psi `$ $`=`$ $`F_1\left[\|00000+\|00110+\|01001\|01111+\|10011+\|10101+\|11010+\|11100\right]`$ (26) $`+`$ $`F_2\left[\|00011\|00101\|01010\|01100\|10000+\|10110+\|11001+\|11111\right].`$ (27) This is quite a complicated superposition state to create (as can be seen from the quantum circuit). The 3-qubit correction circuit is much simpler and hence we see the advantage of the Schrödinger cat encoding. Also, while here we are only discussing a single gate, a reasonable quantum computer has to be constituted of many gates. Then, if the above 5-qubit protection circuit has to be implemented, this will become much more expensive in terms of qubits in comparison to the 3-qubit circuit for bit-flip protection. The bit-flip protection scheme saves 2 qubits at each needed qubit in comparison to the 5-qubit protection circuit described above. It does however only protect against a specific type of error. An unavoidable error incoming from dissipation over superposition states is decoherence. Let us consider the general effect of dissipation on the quantum coherent superposition state. At zero temperature the state of the cavity field is described by the density operator $`\rho _C^\pm (t)`$ $`=`$ $`{\displaystyle \frac{1}{N_\pm ^2}}\{\|\alpha \text{e}^{\gamma t/2}><\alpha \text{e}^{\gamma t/2}\|+\|\alpha \text{e}^{\gamma t/2}><\alpha \text{e}^{\gamma t/2}\|\pm \text{e}^{2\|\alpha \|^2(1\text{e}^{\gamma t})}`$ (29) $`\times [\|\alpha \text{e}^{\gamma t/2}><\alpha \text{e}^{\gamma t/2}\|+\|\alpha \text{e}^{\gamma t/2}><\alpha \text{e}^{\gamma t/2}\|]\}.`$ We see that two characteristic times are involved in this evolution. The first one, the decoherence time is the time in which the pure state given by Eq.(7) is turned into a statistic mixture $$\rho _C(t)\frac{1}{2}\left\{\|\alpha ><\alpha \|+\|\alpha ><\alpha \|\right\},$$ (30) the second is the damping time or relaxation time of the field $`t_c`$ =$`\gamma ^1`$, the time that the dissipative effect reduces the energy of the field leading it in to a vacuum state. The decoherence of the field state is characterized by the $`\mathrm{exp}\left[2\left\|\alpha \right\|^2\left(1\text{e}^{\gamma t}\right)\right]`$ factor, that for short times, $`\gamma t1`$, turns to be $`\mathrm{exp}\left[2\left\|\alpha \right\|^2\gamma t\right]`$ and the coherence decays with the time $`t_d=\left(2\gamma \left\|\alpha \right\|^2\right)^1`$. Unfortunately the coherence time depends on inversely on $`\left\|\alpha \right\|^2`$ and hence the larger the $`\left\|\alpha \right\|^2`$ the smaller the coherence time. Decoherence constitutes the main obstacle to quantum computation , since the encoding is completely based in the purity of the field state. The relaxation time of microwave fields in superconducting cavities is of the order of $`10^2`$ s , what means $`t_d10^2\left\|\alpha \right\|^2`$ s. So all the interactions involved in this proposal must consider this time and more specifically the number of photons as critical quantities. Moreover, the initial information encoded in the superpositions given by Eqs. (20) and (23) also suffer the effect of decoherence, which for $`\|\alpha \|1`$ occur at the same time for both encoding schemes. Again, decoherence prevention schemes play a crucial role for any quantum information encoding. One favorable point for the superposition state encoding is that proposals for sustaining the coherence of these field states have already been considered which could be well adapted for our case. For example the stroboscopic feedback proposal of Vitali et al. . This proposal is particularly appropriate here since it guarantee that at each single decay event a feedback atom is sent through the setup restituting the coherence and the state parity. In fact in the authors claim that the coherence is restored but for a slightly different state. For the proposal presented here what is important is not the original superposition of states, but the original parity of the state, if it was originally a superposition of even or odd photon number states. It is important to emphasize the experimentally critical values for the physical elements involved. The time of flight of the atom across the setup (10<sup>-4</sup> s), the relaxation time of the field (10<sup>-3</sup>-10<sup>-2</sup>s for Niobium superconducting cavities) and the atomic spontaneous emission time (3$`\times 10^2`$s) . ## V Conclusion In conclusion we have presented a feasible scheme to encode the CNot quantum gate, based on a field superposition of states. These states have been already generated in superconducting microwave cavities which constitute a system almost dominated by the current technology . The proposal here to encode the CNot gate based on a superposition of states is less susceptible to irreversible errors due to dissipative effect imposed by the environment than number states . The generation of these kind of states is dependent of a conditional measurement giving a random assignment of initial control bits, which would be useless if no further process is considered. Hence we propose a conditional feedback scheme, which guarantees that the initial control bit is prepared in the required state. Once the amplitude damping of a coherent state (at zero temperature) still constitutes a coherent state the method proposed works until the inevitable effect of decoherence takes place. For that a reset of the qubits must be done after a time of the order of the time of the decoherence or a coherence control scheme must be applied. The reset process is done repeating the process here described. The state of a logic unit can be transferred to another logic unit (if the time of decoherence is respected), constituting a sort of quantum memory circuit . That can be attained by the proper choice of atomic interactions between atoms and the field in the microwave cavity or even by the direct photonic process of coupling two cavities by a superconducting wave-guide, which permits an exchange of information (exchange of states) between the coupled units . This problem is of fundamental importance for the engineering of quantum networks. The major sources of error here are the loss of coherence of the field state and the control bit-flip due to the dissipative effect. Analysis of these kind of errors on quantum networks constituted by the basic element here described is left for further investigation. ###### Acknowledgements. MCO thanks the Fundação de Amparo à Pesquisa do Estado de São Paulo (Brazil) for financial support while WJM acknowledges the support of the Australian Research Council.
warning/0001/nucl-th0001025.html	ar5iv	text	# 1 Introduction ## 1 Introduction Some time ago a bright phenomenon of pion condensation predicted in attracted the common attention and was widely investigated -. The reason for this prediction was the fact that one of the solutions of the pion dispersion equation in medium (let it be called $`\omega _c`$) turns to zero $`\omega _c^2(k)=0`$ (at some momentum $`k0`$) while the density of medium increases . This meant an appearance of excitations with zero energy in medium. In this case the ground state should be reconstructed in correspondence with phase transition. One variant of reconstruction was to include the pion condensate (taken in one or another form) into the ground state. This resulted in a prediction for pion condensation. A number of efforts was devoted to the search for this phenomenon. The result of discussions given in the book of A.B.Migdal et. al. was that the pion condensation manifested itself weakly and probably was absent (at least not observed) at normal density. Nevertheless, this phenomenon could influence the equation of state at high densities achieved in heavy ion collisions or neutron stars. Below, we consider in detail the solutions of pion dispersion equation $$\omega ^2k^2m_\pi ^2\mathrm{\Pi }(\omega ,k,p_F)=0$$ (1) in the complex plane of the pion frequency $`\omega `$. In (1) $`\mathrm{\Pi }`$ is the pion polarization operator (pion self-energy) in the matter. The consideration in the complex $`\omega `$-plane allows us to obtain additional information about well-known solutions. The main content is as follows. We consider excitations with quantum numbers $`0^{}`$ in symmetrical nuclear matter. At the equilibrium density there are three branches of solutions of (1). When the density is larger than critical one ($`\rho >\rho _c`$) the fourth branch appears on the physical sheet. The equation (1) has logarithmic cuts on the physical sheet of the complex $`\omega `$-plane, determined by the structure of the polarization operator $`\mathrm{\Pi }`$. Different branches of solutions are considered on physical and unphysical sheets of the complex $`\omega `$-plane. With the change of $`k`$ they move from physical to unphysical sheet (and backward) through the cuts. Below, to explain the obtained results, we deal with the case when isobar width in medium is equal to that in vacuum, i.e. 115 MeV. In this case all solutions of (1) and certain cuts move from the real (or imaginary) axis to the complex plane. This helps us to trace the $`k`$-dependence of the solutions. For certain important cases the influence of $`\mathrm{\Gamma }_\mathrm{\Delta }`$ on the behaviour of $`\omega _i(k)`$ is studied. The well-known branches of solutions (1) are: 1) spin-isospin sound branch $`\omega _s(k)`$; 2) pion branch $`\omega _\pi (k)`$; 3) isobar branch $`\omega _\mathrm{\Delta }(k)`$. At $`0kk_f`$ they are located on the physical sheet (the momentum $`k_f`$ is different for each branch), then with further increase of $`k`$ they move to unphysical sheet across the cuts. For these branches Re($`\omega _i^2)>0`$ everywhere on the physical sheet. Our investigations show that there is the whole set of solutions at unphysical sheets. With increasing density certain solutions come to the physical sheet. In such a way the fourth branch, $`\omega _c(k)`$, appears on the physical sheet at $`p_F283`$ MeV<sup>1</sup><sup>1</sup>1For our values of the parameters, which are presented below.. Below $`\omega _c(k)`$ is referred as the condensate branch. At $`k=0`$ $`\omega _c(k)`$ is located at the same point as $`\omega _\pi (k)`$; with increasing of $`k`$ the branch goes onto the unphysical sheet. At some $`k=k_1`$ it appears on the physical sheet and at $`k=k_2`$ leaves it for the same unphysical sheet. The values of $`k_1`$ and $`k_2`$ depend on the density. At critical density, $`\rho _c`$, corresponding to $`p_F=283`$ MeV ($`\rho _c1.2\rho _0`$) there is the equality $`k_1=k_2=1.8m_\pi `$. It is $`\omega _c`$ which obeys the inequality Re$`\omega _c^20`$. All branches depend on the isobar width: at $`\mathrm{\Gamma }_\mathrm{\Delta }=0`$ the branch $`\omega _c(k)`$ is pure imaginary and $`\omega _c^20`$. Recall that $`\rho _0`$ is the equilibrium density of the matter, $`\rho _0=2p_{F0}^3/3\pi ^2`$, $`p_{F0}=268`$ MeV. The paper is organized as follows. In section 2 the particle-hole polarization operator $`\mathrm{\Pi }`$ is considered in the complex $`\omega `$-plane. Then the branches of solutions of the pion dispersion equation (1) are presented on the physical and unphysical sheets. It is shown how the branches of solutions go across the logarithmic cuts in the complex plane. The condensate branch $`\omega _c`$ is shown in details at different densities $`\rho `$ and isobar width $`\mathrm{\Gamma }_\mathrm{\Delta }`$. ## 2 Polarization operator and its singularities Here we write down the formulae for the polarization operator used. Our expressions, as is shown below, differ in some points from the well-known expressions . Only $`S`$\- and $`P`$-waves of the $`\pi NN`$ ($`\pi N\mathrm{\Delta }`$) interactions are taken into account. In this case $`\mathrm{\Pi }`$ is the sum of scalar, $`\mathrm{\Pi }_S`$, and vector, $`\mathrm{\Pi }_P`$, terms: $$\mathrm{\Pi }(\omega ,k)=\mathrm{\Pi }_S(\omega ,k)+\mathrm{\Pi }_P(\omega ,k).$$ (2) The scalar polarization operator $`\mathrm{\Pi }_S`$ is constructed using linear PCAC equation obtained by Gell-Mann–Oakes–Renner (GMOR) for the pion mass squared in the matter: $$m_\pi ^2=\frac{NM\|\overline{q}q\|NM(m_u+m_d)}{2f_\pi ^2}.$$ (3) The value $`\kappa =NM\|\overline{q}q\|NM`$ is the scalar quark condensate calculated in nuclear matter ; $`m_u,m_d`$ are masses of the current $`u`$\- and $`d`$-quarks; $`f_\pi ^{}`$ is the pion decay constant in medium. Here we put $`f_\pi ^{}=f_\pi =92`$ MeV (more details can be found in ). The scalar quark condensate $`\kappa `$ can be expanded in a power series with respect to $`\rho `$ $$\kappa =\kappa _0+\rho N\|\overline{q}q\|N+termswithhigherdegreesof\rho ,$$ (4) where $`\kappa _0`$ is the value of scalar quark condensate in vacuum, $`\kappa _0=0.03`$ GeV<sup>3</sup>; $`N\|\overline{q}q\|N`$ is the matrix element for scalar quark condensate in nucleon, $`N\|\overline{q}q\|N8`$. In this place it is enough to keep the first two terms in (4), this correspond to the gas approximation for $`\kappa `$ . Then we get from (3) $$m_\pi ^2=m_\pi ^2\rho \frac{N\|\overline{q}q\|N(m_u+m_d)}{2f_\pi ^2}.$$ (5) On the other hand, in the dispersion equation (1) $`m_\pi ^2`$ is defined as $$m_\pi ^2=m_\pi ^2+\mathrm{\Pi }(\omega ,k=0).$$ (6) Whilst $`\mathrm{\Pi }_P(k=0)=0`$, we get from (5) and (6) the following form for $`\mathrm{\Pi }_S`$: $$\mathrm{\Pi }_S=\rho \frac{N\|\overline{q}q\|N(m_u+m_d)}{2f_\pi ^2}.$$ (7) The P-wave polarization operator $`\mathrm{\Pi }_P`$ can be written following the papers as a sum of nucleon and isobar polarization operators $$\mathrm{\Pi }_P=\mathrm{\Pi }_N+\mathrm{\Pi }_\mathrm{\Delta }.$$ (8) Here $`\mathrm{\Pi }_N(\mathrm{\Pi }_\mathrm{\Delta }`$) is equal to the sum of the nucleon-hole and isobar-hole loops without pion in the intermediate states : $$\mathrm{\Pi }_N=\mathrm{\Pi }_N^0\frac{1+(\gamma _\mathrm{\Delta }\gamma _{\mathrm{\Delta }\mathrm{\Delta }})\mathrm{\Pi }_\mathrm{\Delta }^0/k^2}{E},\mathrm{\Pi }_\mathrm{\Delta }=\mathrm{\Pi }_\mathrm{\Delta }^0\frac{1+(\gamma _\mathrm{\Delta }\gamma _{NN})\mathrm{\Pi }_N^0/k^2}{E},$$ (9) $$E=1\gamma _{NN}\frac{\mathrm{\Pi }_N^0}{k^2}\gamma _{\mathrm{\Delta }\mathrm{\Delta }}\frac{\mathrm{\Pi }_\mathrm{\Delta }^0}{k^2}+\left(\gamma _{NN}\gamma _{\mathrm{\Delta }\mathrm{\Delta }}\gamma _\mathrm{\Delta }^2\right)\frac{\mathrm{\Pi }_N^0\mathrm{\Pi }_\mathrm{\Delta }^0}{k^4}.$$ The expressions for the nucleon-hole loop , $`\mathrm{\Pi }_N^0`$, is as follows: $$\mathrm{\Pi }_N^0(\omega ,k)=\mathrm{Sp}\frac{d^3p}{(2\pi )^3}\mathrm{\Gamma }_{\pi NN}^2\left[\frac{\theta (pp_F)\theta (p_F\|\stackrel{}{p}+\stackrel{}{k}\|)}{E_{\stackrel{}{p}+\stackrel{}{k}}E_p\omega }+\frac{\theta (p_Fp)\theta (\|\stackrel{}{p}+\stackrel{}{k}\|p_F)}{E_pE_{\stackrel{}{p}+\stackrel{}{k}}+\omega }\right].$$ (10) Here $`E_p=p^2/2m^{}`$. The expression for $`\mathrm{\Pi }_\mathrm{\Delta }^0`$ is analogous. The $`\pi NB`$ vertex $`\mathrm{\Gamma }_{\pi NB}`$ with $`B`$ labelling nucleon or $`\mathrm{\Delta }`$-isobar is $$\mathrm{\Gamma }_{\pi NB}=\mathrm{\Gamma }_{\pi NB}^0d_B(k),$$ (11) $$\mathrm{\Gamma }_{\pi NN}^0=i\frac{g_A}{\sqrt{2}f_\pi }\chi ^{}(\stackrel{}{\sigma }\stackrel{}{k})\chi ,\mathrm{\Gamma }_{\pi N\mathrm{\Delta }}^0=f_{\mathrm{\Delta }/N}i\frac{g_A}{\sqrt{2}f_\pi }\chi ^\alpha (\stackrel{}{S}_\alpha ^+\stackrel{}{k})\chi ,$$ (12) where $`\chi `$ is nucleon 2-spinors, $`\stackrel{}{\sigma }`$ is nucleon spin, $`\stackrel{}{S}^+`$ turns the spin $`3/2`$ into $`1/2`$. In order to take into account the non-zero baryon size, the vertex $`\mathrm{\Gamma }_{\pi NB}^0`$ is multiplied by the form factor $`d_B(k)`$ taken in the form $`d_B=(1m_\pi ^2/\mathrm{\Lambda }_B^2)/(1+k^2/\mathrm{\Lambda }_B^2)`$. The constants $`\gamma _{NN},\gamma _\mathrm{\Delta },\gamma _{\mathrm{\Delta }\mathrm{\Delta }}`$ are $$\gamma _{NN}=C_0g_{NN}^{}\left(\frac{\sqrt{2}f_\pi }{g_A}\right)^2,\gamma _\mathrm{\Delta }=\frac{C_0g_{N\mathrm{\Delta }}^{}}{f_{\mathrm{\Delta }/N}}\left(\frac{\sqrt{2}f_\pi }{g_A}\right)^2,\gamma _{\mathrm{\Delta }\mathrm{\Delta }}=\frac{C_0g_{\mathrm{\Delta }\mathrm{\Delta }}^{}}{f_{\mathrm{\Delta }/N}^2}\left(\frac{\sqrt{2}f_\pi }{g_A}\right)^2,$$ where $`C_0`$ is the normalization factor ($`C_0=\pi ^2/(p_Fm^{})`$) and $`g^{}s`$ are the constants of the effective quasi-particle–quasi-hole interaction in nuclear matter . The calculation results are given for the following set of parameters : $$f_{\mathrm{\Delta }/N}2,\mathrm{\Lambda }_N=0.667GeV,\mathrm{\Lambda }_\mathrm{\Delta }=1GeV,g_A=1,f_\pi =92MeV,$$ (13) $$g_{NN}^{}=1.0,g_{N\mathrm{\Delta }}^{}=0.2,g_{\mathrm{\Delta }\mathrm{\Delta }}^{}=0.8.$$ When the parameters $`f_{\mathrm{\Delta }/N},g^{},\mathrm{\Lambda }_B`$ vary within the experimentally acceptable limits the dispersion equation solutions change quantitatively but not qualitatively. Later on, we can perform integration in (10) in two ways. 1. To integrate separately the first and the second terms. In this way there appears the expression for $`\mathrm{\Pi }_N^0`$ as follows: $`\mathrm{\Pi }_N^0(\omega ,k)`$ $`=`$ $`4\left({\displaystyle \frac{g_A}{\sqrt{2}f_\pi }}\right)^2k^2\left[\mathrm{\Phi }_N(\omega ,k)+\mathrm{\Phi }_N(\omega ,k)\right]d_N^2(k),`$ (14) $`\mathrm{\Phi }_N(\omega ,k)`$ $`=`$ $`{\displaystyle \frac{m^{}}{k}}{\displaystyle \frac{1}{4\pi ^2}}({\displaystyle \frac{\omega m^{}+kp_F}{2}}\omega m^{}\mathrm{ln}\left({\displaystyle \frac{\omega m^{}}{\omega m^{}kp_F+k^2/2}}\right)+`$ (15) $`+`$ $`{\displaystyle \frac{(kp_F)^2(\omega m^{}k^2/2)^2}{2k^2}}\mathrm{ln}\left({\displaystyle \frac{\omega m^{}kp_Fk^2/2}{\omega m^{}kp_F+k^2/2}}\right))`$ at $`0k2p_F`$. At $`k2p_F`$ $`\mathrm{\Phi }_N(\omega ,k)`$ is Migdal’s function: $$\mathrm{\Phi }_N(\omega ,k)=\frac{1}{4\pi ^2}\frac{m^3}{k^3}\left[\frac{a^2b^2}{2}\mathrm{ln}\left(\frac{a+b}{ab}\right)ab\right]$$ (16) where $`a=\omega (k^2/2m^{})`$, $`b=kp_F/m^{}`$. Consider now which cuts has the polarization operator $`\mathrm{\Pi }_N^0(\omega ,k)`$ (14)–(16) in the $`\omega `$-plane. We can see that at $`k2p_F`$ there are two cuts (define them as $`I`$ and $`II`$). They are related to the first and second logarithms in (15). The cuts are situated within the intervals: $$I:0\omega \frac{kp_F}{m^{}}\frac{k^2}{2m^{}},II:\frac{kp_F}{m^{}}\frac{k^2}{2m^{}}\omega \frac{kp_F}{m^{}}+\frac{k^2}{2m^{}}.$$ (17) Since $`\mathrm{\Pi }_N^0`$ is symmetrical under the replacement $`\omega \omega `$, the cuts of $`\mathrm{\Phi }_N(\omega ,k)`$ are placed symmetrically on the negative semiaxis. Thus $`\mathrm{\Pi }_N^0`$ has four cuts in the complex $`\omega `$-plane, they are shown in Fig.1. 2. The other way of integration in (10) gives a well-known expression of $`\mathrm{\Pi }_N^0`$ through Migdal’s functions. To follow it, let us do a substitution in (10) $`\theta (pp_F)1\theta (p_Fp)=1n(p)`$. Then, after the integration, $`\mathrm{\Phi }_N(\omega ,k)`$ takes a well-known form (16) for all values of $`k`$ . Undoubtedly, the expression (14) is the same for both integration ways. Now $`\mathrm{\Pi }_N^0`$ has not four cuts in $`\omega `$-plane but two (overlapping). It can be seen from the expressions (10), (15), that the cut $`I`$ is the sum of two overlapping cuts. The search for dispersion equation solutions becomes more convenient and obvious when we work with one cut but not with two overlapping ones. It is difficult to follow the solution in the case 2, therefore we use equations (14)–(16) for $`\mathrm{\Pi }_N^0`$. Now let us turn to the polarization operator $`\mathrm{\Pi }_\mathrm{\Delta }^0`$, which is the isobar–nucleon-hole loop. Since the isobar Fermi surface is absent at the nuclear densities and isobar momentum is unrestricted, there is no problem discussed above and $`\mathrm{\Pi }_\mathrm{\Delta }^0(\omega ,k)`$ reads: $$\mathrm{\Pi }_\mathrm{\Delta }^0=\frac{16}{9}\left(\frac{g_A}{\sqrt{2}f_\pi }\right)^2f_{\mathrm{\Delta }/N}^2k^2\left[\mathrm{\Phi }_\mathrm{\Delta }(\omega ,k)+\mathrm{\Phi }_\mathrm{\Delta }(\omega ,k)\right]d_\mathrm{\Delta }^2(k).$$ (18) The function $`\mathrm{\Phi }_\mathrm{\Delta }(\omega ,k)`$ is expressed through Migdal’s functions (16) with $`a=\omega (k^2/2m^{})\mathrm{\Delta }m`$, $`b=kp_F/m^{}`$. The mass difference, $`\mathrm{\Delta }m=m_\mathrm{\Delta }m`$, is the following: Re$`(\mathrm{\Delta }m)=292`$ MeV and Im$`(\mathrm{\Delta }m)=\mathrm{\Gamma }_\mathrm{\Delta }/2`$. The cuts of $`\mathrm{\Phi }_\mathrm{\Delta }^0(\omega ,k)`$ are shown in Fig.1. At $`\omega 0`$ the cut is in the interval $$\frac{k^2}{2m^{}}+\mathrm{\Delta }m\frac{kp_F}{m^{}}\omega \frac{k^2}{2m^{}}+\mathrm{\Delta }m+\frac{kp_F}{m^{}}.$$ (19) The cut is shifted into the complex plane in the value $`i\mathrm{\Gamma }_\mathrm{\Delta }/2`$. In this paper Landau equation is used for the nucleon effective mass $$m^{}=\frac{m}{1+(2mp_F/\pi ^2)f_1}.$$ (20) Unknown parameter $`f_1`$ is fixed by the condition $`m^{}(p_F=p_{F0})=0.8m`$. ## 3 Solutions of the dispersion equation In this section the solutions of the dispersion equation (1) are presented. The solution branch $`\omega _c`$ emerges on the physical sheet of the complex $`\omega `$-plane at $`p_F=283`$ MeV for the parameter values (13). The figures for the zero sound branch $`\omega _s(k)`$, pion branch $`\omega _\pi (k)`$ and isobar branch $`\omega _\mathrm{\Delta }(k)`$ are presented for $`\mathrm{\Gamma }_\mathrm{\Delta }=115`$ MeV at $`p_F=`$268 and 290 MeV (i.e. at the equilibrium density and at density slightly larger than critical one). Some special cases, with the other values of $`p_F`$ and $`\mathrm{\Gamma }_\mathrm{\Delta }`$, are considered as well. Branch $`\omega _s(k)`$. (Fig.2a) The branch $`\omega _s(k)`$ is shown for $`p_F=268`$ and $`290`$ MeV (curves 1 and 2 correspondingly). The branch begins at $`\omega _s(k=0)=0`$, then while $`k`$ increases, moves practically along the real axis. At $`k_f=0.430m_\pi `$ for $`p_F=290`$ MeV (at $`k_f=0.436m_\pi `$ for $`p_F=268`$ MeV) goes under the cut $`II`$ (17), this corresponds to the decay of $`\omega _s`$ into real nucleon and the hole. Branch $`\omega _\mathrm{\Delta }(k)`$. (Fig.2b) The isobar branch $`\omega _\mathrm{\Delta }(k)`$ begins at $`\omega =\mathrm{\Delta }m`$ at $`k=0`$ and ends on the isobar cut (19) at $`k_f=5.1m_\pi `$ for $`p_F=268`$ MeV ($`k_f=4.8m_\pi `$ for $`p_F=290`$ MeV). Branch $`\omega _\pi (k)`$. (Fig.2c) The pion branch starts at $`k=0`$ in $`\omega _\pi =m_\pi ^{}`$ (see (5),(6)). The beginning of $`\omega _\pi (k=0)`$ is shifted to the smaller then $`m_\pi `$ values when GMOR is used to determine $`m_\pi ^{}`$. The pion branch ends on physical sheet under the isobar cut, this corresponds to the decay of pion into isobar and nucleon hole. It takes place at $`k_f=3.5m_\pi `$ for $`p_F=268`$ MeV $`(k_f=3.8m_\pi `$, $`p_F=290`$ MeV). Branch $`\omega _c(k)`$. (Fig.2d, 3a,b) While the density increases one more branch of solutions, $`\omega _c(k)`$, appears on the physical sheet. It emerges at $`p_F283`$ MeV when parameter values (13) are used. In Fig.2d the pion branch $`\omega _\pi (k)`$ and condensate branch $`\omega _c(k)`$ are presented at $`p_F=290`$ MeV. The dashed piece of $`\omega _c(k)`$ belongs to the upper unphysical sheet of the logarithmic cut $`I`$ (17) (Fig.1). The branch $`\omega _c(k)`$ starts at $`k=0`$ at the same point as $`\omega _\pi (k)`$ and moves onto the unphysical sheet; at $`k=k_1=1.3m_\pi `$ the branch goes down to the physical sheet and at $`k=k_2=2.3m_\pi `$ moves back to the same unphysical sheet. In the momentum interval $`(k_1,k_2)`$ one can follow over all the branches shown in Fig.2 and check that $`\omega _c`$ does not belong to any branches studied before ($`\omega _s,\omega _\pi ,\omega _\mathrm{\Delta }`$). The branch $`\omega _c`$ depends on the isobar width $`\mathrm{\Gamma }_\mathrm{\Delta }`$. Decreasing the isobar width we see that the isobar cuts move to the real axis and $`\omega _s`$, $`\omega _\pi `$ and $`\omega _\mathrm{\Delta }`$ have smaller imaginary parts. When $`\mathrm{\Gamma }_\mathrm{\Delta }=0`$ the branches $`\omega _s`$, $`\omega _\pi `$ and $`\omega _\mathrm{\Delta }`$ are real. On the contrary, $`\omega _c(k)`$ moves to the imaginary axis with decreasing $`\mathrm{\Gamma }_\mathrm{\Delta }`$ (Fig.3a). At $`\mathrm{\Gamma }_\mathrm{\Delta }=0`$ we have pure imaginary solutions on the physical sheet: $`\omega _c^20`$. In Fig.3b the branch $`\omega _c(k)`$ is shown at the different densities: $`p_F=280,290,300,360`$ MeV. When $`p_F=280`$ MeV the whole branch (curve 1) is located on unphysical sheet. At $`p_F=283`$ MeV the branch touches the real axis (not shown). For $`p_F>283`$ Mev the part of $`\omega _c(k)`$ in the interval $`(k_1,k_2)`$ is placed on the physical sheet. At the critical density $`\rho =rho_c`$ and $`\mathrm{\Gamma }_\mathrm{\Delta }=0`$ $`\omega _c(k)`$ touches the real axis at the point $`\omega _c(k)=0.`$ It was shown in paper that the appearance of $`\omega _c`$ on the physical sheet results not only in pion condensation but in restoration of chiral symmetry in the nuclear matter at critical density as well. ## 4 Conclusion In the paper the solutions of pion dispersion equation are considered in details in the complex $`\omega `$-plane. It is shown that, besides the well-known solutions with quantum numbers $`0^{}`$ (zero spin-isospin sound, pion and isobar waves), there exists the fourth branch $`\omega _c(k)`$. It is the branch which obeys the condition $`\omega _c^20`$, therefore it is responsible for instability of the ground state. Such instability can indicate the beginning of ’pion condensation’. We demonstrate that at the density less than critical one, $`\rho <\rho _c`$ the branch $`\omega _c(k)`$ is situated on unphysical sheet and at $`\rho \rho _c`$ it comes on physical one. ### Acknowledgments I am grateful to M.G. Ryskin for the important and fruitful discussions during the work. I thanks E.G. Drukarev and E.E. Saperstein for useful discussions. This work was supported by RFFI grant 96-15-96764. ## 5 Figure captions Fig.1. The cuts on the physical sheet of the complex $`\omega `$-plane of polarization operators $`\mathrm{\Pi }_N^0`$, $`\mathrm{\Pi }_\mathrm{\Delta }^0`$, corresponding to equations (14), (17), (18), (19). The cuts are presented at $`p_F=290`$ MeV, $`k=m_\pi `$. Fig.2. Branches of solutions of (1) in the complex $`\omega `$-plane. Curves 1 and 2 stand for $`p_F=268,290`$ MeV, correspondingly. The dashed pieces of curves are situated on unphysical sheets. a) Zero spin-isospin sound branch $`\omega _s(k)`$. The curves are presented up to 1.6$`m_\pi `$. b) The isobar branch $`\omega _\mathrm{\Delta }`$. The horizontal dashed line is a logarithmic cut (19) for $`p_F=290`$ MeV at momentum $`k`$ when $`\omega _\mathrm{\Delta }(k)`$ is on the cut. c) The pion branch $`\omega _\pi `$. The horizontal dashed line is a logarithmic cut (19) for $`p_F=290`$ MeV at momentum $`k`$ when $`\omega _\pi `$ is on the cut. d) The total picture at $`p_F=290`$ MeV for pion branch $`\omega _\pi (k)`$ and condensate branch $`\omega _c`$(k). Fig.3. Condensate branch $`\omega _c`$ in the complex $`\omega `$-plane. Here the dashed pieces of branches are on the physical sheet, but solid lines belong to unphysical sheet. a) The branch $`\omega _c`$ is presented at $`p_F=290`$ MeV for different values of isobar width: $`\mathrm{\Gamma }_\mathrm{\Delta }=0,10,50,115`$ MeV (curves 1,2,3,4, correspondingly); $`\omega _c(k=0)=0.744m_\pi `$. b) The branch $`\omega _c`$ is presented at $`\mathrm{\Gamma }_\mathrm{\Delta }=115`$ MeV for different values of Fermi momenta $`p_F=280,290,300,360`$ MeV (curves 1,2,3,4, correspondingly). For $`p_F`$=300 and 360 MeV only the part of the branch, which is placed on the physical sheet, is shown (the curves 3 and 4). The whole branch for $`p_F=280`$ MeV (curve 1) is on unphysical sheet.
warning/0001/math0001162.html	ar5iv	text	# Combed 3-Manifolds with Concave Boundary, Framed Links, and Pseudo-Legendrian Links ## Introduction This paper describes combinatorial realizations, based on the machinery of branched standard spines (see Section 1) of the following three topological categories (in which manifolds and diffeomorphisms are oriented by default): 1. Combed $`3`$-manifolds with concave boundary, that is pairs $`(M,v)`$, where $`M`$ is a compact 3-manifold (possibly with boundary), and $`v`$ is a nowhere-zero vector field on $`M`$ with simple tangency circles of concave type on $`M`$, up to diffeomorphism of $`M`$ and homotopy of $`v`$ through fields of the same sort; 2. Framed links in $`3`$-manifolds, that is pairs $`(M,L)`$, where $`M`$ is as above and $`L`$ is a framed link in $`M`$, up to diffeomorphism of $`M`$ and framed isotopy of $`L`$; 3. Pseudo-Legendrian links in combed $`3`$-manifolds, that is triples $`(M,v,L)`$, where $`(M,v)`$ is as above and $`L`$ is transversal to $`v`$, up to diffeomorphism of $`M`$ and ‘pseudo-Legendrian isotopy’ of $`(v,L)`$, i.e. simultaneous homotopy of $`v`$ and isotopy of $`L`$ through pairs $`(v,L)`$ of the same type. We will denote these categories respectively by $`\mathrm{Comb}`$, $`\mathrm{Fram}`$ and $`\mathrm{PLeg}`$. (Regarding names, recall that a non-zero vector field up to homotopy is often called a combing, and that if $`\xi `$ is an oriented contact structure and $`L`$ is Legendrian in $`\xi `$, then $`(M,\xi ^{},L)`$ defines an element of $`\mathrm{PLeg}`$.) Our realizations are given according to the by now popular scheme in 3-dimensional topology, namely: 1. A class of combinatorial objects, each of which can be specified by a finite set of data, and a surjective reconstruction map which assigns to a combinatorial object a topological one; 2. A finite set of local combinatorial moves on objects, finite combinations of which give the equivalence relation induced by the reconstruction map. In the definitions of the topological categories given above we have been forced to include the action of diffeomorphisms, because we use spines, which determine manifolds only up to diffeomorphism. However if a certain manifold $`M`$ is given we can restrict to spines embedded in $`M`$ (rather than abstract ones), and get formally identical combinatorial realizations of the refined categories where only diffeomorphisms of $`M`$ isotopic to the identity are considered. We will mention how to do this in Section 2 for combings, but a similar refinement could easily be stated for framed links and for pseudo-Legendrian links. Rather than providing precise statements of our realizations, in this introduction we give some general background and motivations, starting with $`\mathrm{Comb}`$. A combinatorial realization of the subcategory $`\mathrm{Comb}^{\mathrm{cl}}`$ of $`\mathrm{Comb}`$ given by pairs $`(M,v)`$ with closed $`M`$ was given in . In Section 2 we extend the arguments of to the bounded case, and we actually refine the results proved there, by showing that some of the moves previously considered may actually be neglected. The realization of $`\mathrm{Comb}^{\mathrm{cl}}`$ in was the basis for the treatment of other refinements of the category of 3-manifolds, involving spin structures and framings. These realizations proved fruitful in connection with spin-refined Turaev-Viro invariants (see Section 8.3 in ) and G. Kuperberg’s invariants for combed and framed manifolds, of which a very constructive description is given in . Our main motivation here comes from , where we have developed a theory of Euler structures with simple boundary and their Reidemeister-Turaev torsion (see ). The surjectivity of the reconstruction map of the realization of $`\mathrm{Comb}`$ was used in to construct an explicit canonical $`H_1`$-equivariant bijection from the space of smooth Euler structures to the space of combinatorial Euler structures, and to exhibit a canonical Euler chain for the structure carried by a branched spine. The subcategory of $`\mathrm{Fram}`$ consisting of framed links in closed manifolds was combinatorially realized by Turaev in terms of link diagrams on a given standard spine, and moves on these diagrams (including the classical framed Reidemeister moves). In Section 3 we modify the situation considered by Turaev by taking a branched standard spine of the manifold, and restricting to link diagrams which are $`\mathrm{C}^1`$ with respect to the branching. On one hand, this allows to simplify the encoding of the framing, because the field carried by the spine is automatically transverse to the link, while Turaev needs to add half-twists. On the other hand, some technical complications emerge, because only $`\mathrm{C}^1`$ moves can be used. Nevertheless, a result formally analogous to Turaev’s turns out to be true, yielding the presentation of $`\mathrm{Fram}`$ discussed in Section 3. In Section 4 we exploit the fact that if a branched spine defines a global field on a manifold, according to the scheme given for $`\mathrm{Comb}`$, then the link defined by a $`\mathrm{C}^1`$ diagram on the spine is automatically pseudo-Legendrian with respect to the field. This leads us to the presentation of $`\mathrm{PLeg}`$. Comparing the presentations of $`\mathrm{Fram}`$ and $`\mathrm{PLeg}`$ one notices a rather remarkable feature: the former is obtained from the latter just by adding the ‘curl’ (first Reidemeister) move. This fact has two interesting interpretations: * it is a perfect combinatorial analogue of the imitation of a framed isotopy by a Legendrian isotopy in a contact manifold; * it allows a partial extension of the notion of winding number of a link diagram. The imitation mentioned in (a) plays a central role in the comparison, due to Fuchs-Tabachnikov and Tchernov of framed and Legendrian finite-order invariants, and we believe that our combinatorial realizations could be of some help in the understanding of these invariants. In particular, we conjecture that the right environment in which finite-order invariants should be considered in precisely our category $`\mathrm{PLeg}`$ (we will provide an exact statement and some evidence in Section 5). Concerning (b), recall first that if a knot $`K:S^1\text{}^3`$ is transverse to the constant vertical field $`/z`$, then its equivalence class up to isotopy transverse to $`/z`$ is determined by the framed isotopy class and by the ‘winding’ number (the degree of $`\pi K^{}`$, where $`\pi `$ is the obvious projection on the horizontal unit circle). Using our presentations of $`\mathrm{Fram}`$ and $`\mathrm{Comb}`$ we can show that a partial analogue of this fact is true in any combed manifold with concave boundary, provided one allows a homotopy of the field simultaneous with the isotopy of the knot. In the general setting, however, the winding number only exists as a relative object, and we can prove that it leads to a well-defined invariant only under the assumption that the knot is ‘good.’ The notion of ‘goodness’ for knots emerged in our study of torsion as a relative invariant of pairs of pseudo-Legendrian knots which are framed-isotopic . Many knots are good: for instance, all knots are good if the ambient manifold is a homology sphere, and most knots with hyperbolic complement are good. In Section 5 we will give some applications of the notion of relative winding number, in connection with torsion and finite-order invariants. In particular we will show the following: ###### Proposition 0.1 Let $`M`$ be a homology sphere, let $`v`$ be a field on $`M`$ and consider two pseudo-Legendrian knots in $`(M,v)`$ which are isotopic as framed knots. Then the following conditions are pairwise equivalent: 1. the knots are pseudo-Legendrian isotopic; 2. the relative winding number vanishes; 3. the knots have the same Maslov index; 4. the knots cannot be distinguished by the relative torsion invariants of ; 5. the knots are homotopic as pseudo-Legendrian immersions. Moreover we will prove that torsion invariants cannot distinguish the pairs of framed-isotopic Legendrian knots given in , which Tchernov shows to be distinguished by finite-order invariants. We conclude by giving another perspective of the realization of $`\mathrm{PLeg}`$. Recall that $`\mathrm{PLeg}`$ comes as a refinement of Turaev’s presentation of $`\mathrm{Fram}`$, which was the starting point of his beautiful theory of 4-dimensional shadows. We believe that the extra structure given by the branching of the spine, which underlies the presentation of $`\mathrm{PLeg}`$, should have 4-dimensional counterparts. Our intuition is that “4-dimensional branched shadows”, which are not quite defined yet, should correspond to $`\mathrm{Spin}^\mathrm{c}`$ structures on 4-manifolds and allow to treat their invariants. This intuition is supported by the fact that in dimension three branched spines indeed are a good framework to treat torsion of Euler structures (and hence, in particular, Seiberg-Witten invariants of closed 3-manifolds with $`\mathrm{Spin}^\mathrm{c}`$ structures, see ). Acknowledgment. Section 5 owes a lot to very useful discussions we had with Vladimir Tchernov. ## 1 Branched spines and combings This section contains many definitions used below and reviews the theory developed in . #### Manifolds and fields All the manifolds we will consider are 3-dimensional, oriented, and compact, with or without boundary. Using the Hauptvermutung, we will somewhat intermingle the differentiable and piecewise linear viewpoints. Maps will always respect orientations. All vector fields mentioned in this paper will be non-singular, and they will be termed just fields for the sake of brevity. A field $`v`$ on a manifold $`M`$ is called traversing if its orbits eventually intersect $`M`$ transversely in both directions (in other words, orbits are compact intervals or points). A point where $`v`$ is tangent to $`M`$ is called simple if it appears in a cross-section as in Fig. 1. The field is called concave if it is tangent to $`M`$ only in a concave fashion, as shown on the left in the figure. Given a concave field $`v`$ on $`M`$, the boundary of $`M`$ naturally splits into the region on which $`v`$ points outside $`M`$ (which we denote by $`B`$ and call the black region), and the region on which $`v`$ points inside (denoted by $`W`$ and called white). Note that $`B=W`$ is a union of circles. The pair $`(B,W)`$, which is actually determined by any two of its elements, is called a boundary pattern on $`M`$. (This definition is a simplified version of that given in , because here we do not allow convex tangency.) Starting from the Poincaré-Hopf theorem one can show that a given boundary pattern $`𝒫=(B,W)`$ on $`M`$, i.e. a splitting of $`M`$ into two surfaces with common boundary, arises from a concave field if and only if $`\chi (W)=\chi (M)`$. See . #### Standard spines A simple polyhedron $`P`$ is a finite, connected, purely 2-dimensional polyhedron with singularity of stable nature (triple lines and points where six non-singular components meet). Such a $`P`$ is called standard if all the components of the natural stratification given by singularity are open cells. Depending on dimension, we will call the components vertices, edges and regions. A standard spine of a $`3`$-manifold $`M`$ with $`M\mathrm{}`$ is a standard polyhedron $`P`$ embedded in $`\mathrm{Int}(M)`$ so that $`M`$ collapses onto $`P`$. Standard spines of oriented $`3`$-manifolds are characterized among standard polyhedra by the property of carrying an orientation, defined (see Definition 2.1.1 in ) as a “screw-orientation” along the edges (as in the left-hand-side of Fig. 2), with the obvious compatibility at vertices (as in the centre of Fig. 2). It is the starting point of the theory of standard spines that every oriented $`3`$-manifold $`M`$ with $`M\mathrm{}`$ has an oriented standard spine, and can be reconstructed (uniquely up to equivalence) from any of its oriented standard spines. See for the non-oriented version of this result and or Proposition 2.1.2 in for the (slight) oriented refinement. We will denote by $`M(P)`$ the manifold defined by $`P`$. Note that $`M(P)\mathrm{}`$. To recover closed manifolds one considers spines $`P`$ such that $`M(P)S^2`$, and defines $`\widehat{M}(P)`$ as $`M(P)_fD^3`$ with $`f:M(P)S^2`$ a diffeomorphism. Note that this definition makes sense also when $`M(P)`$ has more than one component, but at least one is a sphere. #### Moves for standard spines The fundamental move for standard spines, which (in both directions) preserves the topological type of the associated manifold, is the Matveev-Piergallini MP move, see and Fig. 3. Counting the vertices involved one is naturally led to call the positive MP a “2-to-3” move. The MP-move and its inverse are actually not sufficient to relate spines of the same manifold, because they obviously cannot apply to spines with one vertex. However, as soon as one decides to dismiss these “MP-rigid” spines (not the corresponding manifolds, which have plenty of other spines), the MP-move does become sufficient . To deal with spines with one vertex the “0-to-2” move of Fig. 4 (and its inverse) must be added. #### Branched spines A branching on a standard polyhedron $`P`$ is an orientation for each region of $`P`$, such that no edge is induced the same orientation three times. See the right-hand side of Fig. 2 and Definition 3.1.1 in for the geometric meaning of this notion. An oriented standard spine $`P`$ endowed with a branching is shortly named branched spine. We will never use specific notations for the extra structures: they will be considered to be part of $`P`$. The following result, proved as Theorem 4.1.9 in , is the starting point of our constructions. ###### Proposition 1.1 To every branched spine $`P`$ there corresponds a manifold $`M(P)`$ with non-empty boundary and a concave traversing field $`v(P)`$ on $`M(P)`$. The pair $`(M(P),v(P))`$ is well-defined up to equivalence, and an embedding $`i:P\mathrm{Int}(M(P))`$ is defined with the property that $`v(P)`$ is positively transversal to $`i(P)`$. The topological construction which underlies this proposition is actually quite simple, and it is illustrated in Fig. 5. Concerning the last assertion of the proposition, note that the branching allows to define an oriented tangent plane at each point of $`P`$. #### Non-traversing fields and closed manifolds As noted above, standard spines do not directly represent closed manifolds, but one can use spines of manifolds bounded by $`S^2`$ and cap off this sphere to get a closed manifold, or, viewing things the other way around, one can remove an open ball from a given closed manifold to get a bounded one. When one is interested in a manifold equipped with a field, one can try to use branched spines, but of course one sees that they are inadequate to give a direct description both when the manifold is closed and when the field is non-traversing. This limitation is circumvented again by removing a ball, with a proviso on the field on that ball. Let $`P`$ be a branched standard spine, and assume that in $`M(P)`$ there is only one component which is diffeomorphic to $`S^2`$ and is split by the tangency line of $`v(P)`$ to $`M(P)`$ into two discs. Such a component will be denoted by $`S_{\mathrm{triv}}^2`$. Now, notice that $`S_{\mathrm{triv}}^2`$ is also the boundary of the closed $`3`$-ball with constant vertical field, denoted by $`B_{\mathrm{triv}}^3`$. This shows that we can cap off $`S_{\mathrm{triv}}^2`$ by attaching a copy of $`B_{\mathrm{triv}}^3`$, getting a compact manifold $`\widehat{M}(P)`$ and a concave field $`\widehat{v}(P)`$ on $`\widehat{M}(P)`$. If we denote by $`\widehat{𝒫}(P)`$ the boundary pattern of $`\widehat{v}(P)`$ on $`\widehat{M}(P)`$, we easily see that the pair $`(\widehat{M}(P),\widehat{v}(P))`$ is only well-defined up to diffeomorphism of $`\widehat{M}(P)`$ and homotopy of $`\widehat{v}(P)`$ through fields compatible with $`\widehat{𝒫}(P)`$. #### Standard sliding moves Let $`PP^{}`$ be a positive MP-move (so, $`P^{}`$ has one vertex and one region more than $`P`$). If $`P`$ has a branching, all the regions of $`P^{}`$, except for the new one, already have an orientation, and it is a fact that the new region can always be given an orientation (sometimes not a unique one) so to get a branching on $`P^{}`$. Each of the moves on branched spines arising like this will be called a branched MP-move, and it will be called a sliding MP-move if moreover it does not modify the boundary pattern of the associated concave field. One can actually see that each sliding-MP-move can be realized within a certain pair $`(M,v)`$ as a continuous deformation through branched spines of $`M`$ transverse to $`v`$, with only one singularity at which the spine is non-standard but transversality is preserved. This deformation is shown in Fig. 6, and it justifies the term ‘sliding’ quite clearly. Since in Fig. 6 we are showing portions of spines embedded in $`\text{}^3`$, to give a completely intrinsic description of the moves we should specify in each portion whether the screw-orientation of the spine is equal or opposite to that induced by $`\text{}^3`$, and whether the upward vertical field is positively or negatively transversal to the spine. As a result, the complete list of sliding-MP-moves contains 16 different ones, but the essential physical modifications are only those shown in Fig. 6. From this figure one also sees quite clearly that if $`\widehat{v}(P)`$ and $`\widehat{v}(P^{})`$ can be defined then they coincide (up to homotopy through fields compatible with $`\widehat{𝒫}(P)=\widehat{𝒫}(P^{})`$). Another move which obviously has the same property, and will be needed below, is the branched version of the 0-to-2 move, shown Fig. 7, and called the snake move in the sequel. As above, if one takes orientations into account, there is another essentially different snake move, obtained by mirroring Fig. 7. Since also the snake move involves a sliding, we will call standard sliding move any sliding-MP or snake move. ## 2 A calculus for combed manifolds <br>with concave boundary In this section we will extend and refine the main results of Chapter 5 of . The extension consists in passing from the closed to the bounded case, and the refinement comes from the shortening of the list of moves to be considered. More precisely, we will show that compact manifolds with concave combings are combinatorially described by (suitable) branched spines up to certain moves, namely the standard sliding (snake and sliding-MP) moves shown above. Moreover, we will show that spines which are rigid with respect to sliding-MP-moves can be dismissed with no harm, and that the sliding-MP-moves suffice to generate the equivalence on the remaining spines. This implies that our result is a perfect combed analogue of the Matveev-Piergallini theorem (but our proof is self-contained). #### Definitions and statements We will denote by $`\mathrm{Comb}`$ the set of all pairs $`(M,v)`$, where $`M`$ is a compact oriented manifold and $`v`$ is a concave field on $`M`$, viewed up to diffeomorphism of $`M`$ and homotopy of $`v`$ through concave fields. A class $`[M,v]\mathrm{Comb}`$ is called a combing on the diffeomorphism class of the manifold $`M`$. Note that the boundary pattern on $`M`$ evolves isotopically during a homotopy of $`v`$, so a pair $`(M,𝒫)`$, viewed up to diffeomorphism of $`M`$, can be associated to each $`[M,v]\mathrm{Comb}`$. In particular, $`\mathrm{Comb}`$ naturally splits as the disjoint union of subsets $`\mathrm{Comb}([M,𝒫])`$, consisting of combings on $`M`$ compatible with $`𝒫`$. For a technical reason we actually rule out from $`\mathrm{Comb}`$ the set of those classes $`[M,v]`$ such that the corresponding boundary pattern contains components of the type $`S_{\mathrm{triv}}^2`$. This is actually not a serious restriction, because each $`S_{\mathrm{triv}}^2`$ component can be capped off by a $`B_{\mathrm{triv}}^3`$, and the result is well-defined up to homotopy. Note that we do accept pairs $`(M,v)`$ with closed $`M`$, and pairs in which $`v`$ has no tangency at all to $`M`$. Let us denote now by $``$ the set of all branched spines $`P`$ (up to PL isomorphism) such that $`𝒫(P)`$ contains only one $`S_{\mathrm{triv}}^2`$. Such a $`P`$ being given, $`\widehat{M}(P)`$ and $`\widehat{v}(P)`$ can be considered, and the pair $`(\widehat{M}(P),\widehat{v}(P))`$ gives rise to a well-defined element of $`\mathrm{Comb}`$, which we denote by $`C(P)`$. The following will be shown below: ###### Theorem 2.1 The map $`C:\mathrm{Comb}`$ is surjective, and the equivalence relation defined by $`C`$ on $``$ is generated by sliding-MP-moves and snake moves. ###### Remark 2.2 The following interpretation of the surjectivity of $`C`$ is perhaps useful. Note first that the dynamics of a field, even a concave one, can be very complicated, whereas the dynamics of a traversing field (in particular, $`B_{\mathrm{triv}}^3`$) is simple. Surjectivity of $`C`$ means that for any (complicated) concave field there exists a sphere $`S^2`$ which splits the field into two (simple) pieces: a standard $`B_{\mathrm{triv}}^3`$ and a concave traversing field. Actually, a 1-parameter version of this statement also holds (see Remark 2.6): we will need it to show that the $`C`$-equivalence is the same as the sliding equivalence. As announced, we state now the sliding analogue of the fact that the MP moves suffice. Let us denote by $``$ the subset of $``$ consisting of the branched spines which are “rigid” from the point of view of the sliding-MP-moves, i.e. the spines to which no such move applies. An explicit description of $``$ is given in the proof of the next result. In the statement we only emphasize the most important consequences of this description. ###### Proposition 2.3 1. For every surface $`\mathrm{\Sigma }`$ and pattern $`𝒫`$ on $`\mathrm{\Sigma }`$ there are at most two spines $`P`$ such that $`(M(P))(\mathrm{\Sigma },𝒫)`$. 2. If two elements of $``$ are related through sliding-MPmoves and snake moves, they are also related through sliding-MP-moves only. 3. Every $`P`$ is related by a snake move to an element of $``$. This proposition shows that in the statement of Theorem 2.1 one may remove $``$ from $``$ and forget the snake move. #### Embedding-refined calculus We spell out in this paragraph the embedding-refined version of our calculus, which allows to neglect the action of automorphisms. Let a certain manifold $`M`$ be given, and consider the set $`\mathrm{Comb}(M)`$ of concave vector fields on $`M`$, up to homotopy. Let $`(M)`$ consist of the elements of $``$ which are smoothly embedded in $`M`$ as spines of $`M`$ minus a ball $`B_{\mathrm{triv}}^3`$. Each element $`P`$ of $`(M)`$ is viewed up to isotopy in $`M`$, and gives rise to a well-defined element $`C_M(P)`$ of $`\mathrm{Comb}(M)`$. Moreover sliding-MP-moves and snake moves are well-defined in $`(M)`$, because they can be realized as embedded moves. The embedded analogue of Theorem 2.1 states that $`C_M:(M)\mathrm{Comb}(M)`$ is surjective, and the relation it defines is generated by the embedded moves. The proof of this result is a refinement of the proof of the general statement, along the lines explained in (4.1.12, 4.1.13, 4.3.5, and 5.2.1.) #### Normal sections of a concave field The proof of Theorem 2.1 is an extension of the argument given in Chapter 5 of , and it is based on the following technical notion, which extends ideas originally due to Ishii . Let $`v`$ be a concave field on $`M`$. Let $`B_1,\mathrm{},B_k`$ be the black components of the splitting of $`M`$, i.e. the regions on which $`v`$ points outwards. A normal section for $`(M,v)`$ is a compact surface $`\mathrm{\Sigma }`$ with boundary, embedded in the interior of $`M`$, with the following properties: 1. $`v`$ is transverse to $`\mathrm{\Sigma }`$; 2. $`\mathrm{\Sigma }`$ has exactly $`k+1`$ components $`\mathrm{\Sigma }_0,\mathrm{},\mathrm{\Sigma }_k`$, with $`\mathrm{\Sigma }_0D^2`$; 3. For $`i>0`$, the projection of $`B_i`$ on $`\mathrm{\Sigma }`$ along the orbits of $`v`$ is well-defined and yields a diffeomorphism between $`B_i`$ and a surface $`B_i^{}`$ contained in the interior of $`\mathrm{\Sigma }_i`$, with $`\mathrm{\Sigma }_iB_i^{}`$ being a collar on $`\mathrm{\Sigma }_i`$ (so $`B_i^{}=\mathrm{\Sigma }_i`$ if $`\mathrm{\Sigma }_i=\mathrm{}`$); 4. Each positive half-orbit of $`v`$ meets either the interior of some $`B_i`$ (where it stops), or the interior of some $`\mathrm{\Sigma }_i`$; 5. $`\mathrm{\Sigma }`$ meets itself generically along $`v`$ (i.e. each orbit of $`v`$ meets $`\mathrm{\Sigma }`$ at most two consecutive times on $`\mathrm{\Sigma }`$, and, if so, transversely); 6. Let $`P_\mathrm{\Sigma }`$ be the union of $`\mathrm{\Sigma }`$ with all the orbit segments starting on $`\mathrm{\Sigma }`$ and ending on $`\mathrm{\Sigma }`$. Then $`\mathrm{\Sigma }`$, which is a simple polyhedron by the previous point, is actually standard. The next two lemmas show that normal sections of $`(M,v)`$ correspond bijectively to spines $`P`$ such that $`C(P)=[M,v]`$. The proof of surjectivity of $`C`$ and the discussion of its non-injectivity will be based on these lemmas. ###### Lemma 2.4 If $`(M,v)`$, $`\mathrm{\Sigma }`$ and $`P_\mathrm{\Sigma }`$ are as above, then $`P_\mathrm{\Sigma }`$ can be given a structure of branched spine such that $`C([P_\mathrm{\Sigma }])=[M,v]`$. Proof of2.4. We orient $`\mathrm{\Sigma }`$ so that $`v\|^+\mathrm{\Sigma }`$ (by default $`M`$ is oriented). Every region of $`P_\mathrm{\Sigma }`$ contains some open portion of $`\mathrm{\Sigma }`$, so it can be oriented accordingly; with the obvious screw-orientation, this turns $`P_\mathrm{\Sigma }`$ into a branched spine of its regular neighbourhood in $`M`$. We show that $`C([P_\mathrm{\Sigma }])=[M,v]`$ by embedding the abstract manifold $`M(P_\mathrm{\Sigma })`$ in $`M`$, in such a way that the field carried by $`P_\mathrm{\Sigma }`$ on $`M(P_\mathrm{\Sigma })M`$ is just the restriction of $`v`$. By construction, $`MM(P_\mathrm{\Sigma })`$ will consist of a copy of $`B_{\mathrm{triv}}^3`$, together with a collar on $`M`$ which can be parameterized as $`(M)\times [0,1]`$ in such a way that $`v`$ is constant in the $`[0,1]`$-direction. This easily implies that $`C([P_\mathrm{\Sigma }])=[M,v]`$ indeed. We illustrate the embedding of $`M(P_\mathrm{\Sigma })`$ in $`M`$ pictorially in one dimension less. Figure 8 shows how $`\mathrm{\Sigma }_0`$ gives rise to a $`B_{\mathrm{triv}}^3`$. In the figure we describe $`v`$ by dotted lines, $`\mathrm{\Sigma }`$ by thick lines, portions of $`P_\mathrm{\Sigma }\mathrm{\Sigma }`$ by thin lines, and $`(M(P_\mathrm{\Sigma }))`$ by a thick dashed line. Note also that the portions of $`P_\mathrm{\Sigma }\mathrm{\Sigma }`$ have been slightly modified so to become positively transversal to $`v`$, which allows us to represent the branching as usual, i.e. as a $`\mathrm{C}^1`$ structure on $`P_\mathrm{\Sigma }`$. Figure 9 shows the collar based on a component of $`M`$. We use the same conventions as in the previous figure, and in addition we represent the black and white components of $`M`$ by thick and thin lines respectively. This description concludes the proof. 2.4 ###### Lemma 2.5 Let $`[P]`$ and $`C([P])=[M,v]\mathrm{Comb}`$, with $`P`$ embedded in $`(M,v)`$ according to the geometric description of $`C`$. Let $`\mathrm{\Sigma }`$ be obtained from $`P`$ as suggested (in one dimension less) in Fig. 10. Then $`\mathrm{\Sigma }`$ is a normal section of $`(M,v)`$, and $`P_\mathrm{\Sigma }`$ is isomorphic to $`\mathrm{\Sigma }`$. Proof of2.5. The construction suggested by Fig. 10 is obviously the inverse of the construction in the proof of Lemma 2.4. 2.5 #### The concave combing calculus Using normal sections we can now show the main result of this section. Proof of2.1. We start with the proof of surjectivity. So, let us consider a combed manifold $`(M,v)`$, subject to the usual restrictions. By Lemma 2.4 it is natural to try and construct a normal section for $`(M,v)`$. Let $`B_1,\mathrm{},B_k`$ be the black regions in $`M`$. Slightly translate each $`B_i`$ along $`v`$, getting $`B_i^{}`$. Add to each $`B_i^{}`$ a small collar normal to $`v`$, getting $`\mathrm{\Sigma }_i`$ (if $`B_i=\mathrm{}`$, we set $`\mathrm{\Sigma }_i=B_i^{}`$). Select finitely many discs $`\{D_n\}`$ disjoint from each other and from all the $`\mathrm{\Sigma }_i`$’s, such that all positive orbits of $`v`$, except for the small segments between $`B_i^{}`$ and $`B_i`$, meet $`(_{i1}\mathrm{\Sigma }_i)(D_n)`$ in some interior point. Connect the $`D_n`$’s together by strips transversal to $`v`$ and disjoint from $`_{i1}\mathrm{\Sigma }_i`$, getting a disc $`\mathrm{\Sigma }_0`$. Up to a generic small perturbation, the surface $`\mathrm{\Sigma }=_{i0}\mathrm{\Sigma }_i`$ satisfies all axioms of a normal section for $`(M,v)`$, except axiom 6. Now, even if it is not standard, $`P_\mathrm{\Sigma }`$ can be defined, and the proof of Lemma 2.4 shows that it is a simple branched spine of $`(MB^3,v)`$. In particular, $`P_\mathrm{\Sigma }`$ is connected and its singular locus is non-empty. We recall now that in Chapter 4 of we have considered a set of local moves on simple branched spines, called ‘simple sliding moves’, which preserve the transversal field (and hence the splitting of the boundary), but do not require or preserve the cellularity condition. Knowing that $`P_\mathrm{\Sigma }`$ is connected and $`S(P_\mathrm{\Sigma })\mathrm{}`$, it is not too hard to see that there exists a sequence of (abstract) simple sliding moves which turns $`P_\mathrm{\Sigma }`$ into a standard spine (see , Section 4.4). If we physically realize these moves within $`M`$, preserving transversality to $`v`$, the result is a standard branched spine $`P`$ such that $`C([P])=[M,v]`$. We are left to show that if $`C([P_0])=C([P_1])`$ then $`P_0`$ and $`P_1`$ are related by sliding-MP-moves and snake moves (‘sliding-equivalent’ for short). By the definition of $`\mathrm{Comb}`$ and $`C`$, using also the above lemmas, there exists a manifold $`M`$ and a homotopy $`(v_t)`$ of concave fields on $`M`$, such that $`P_0`$ and $`P_1`$ are defined by normal sections $`\mathrm{\Sigma }^{(0)}`$ and $`\mathrm{\Sigma }^{(1)}`$ of $`(M,v_0)`$ and $`(M,v_1)`$ respectively. We prove that $`P_0`$ and $`P_1`$ are sliding-equivalent first in the special case where $`v_0=v_1=v`$. The general case will be an easy consequence. For $`j=0,1`$, let $`\mathrm{\Sigma }^{(j)}=_{i0}\mathrm{\Sigma }_i^{(j)}`$. Proceeding as in the above proof of surjectivity, for each black region $`B_i`$ of $`M`$, we consider a collared negative translate $`\overline{\mathrm{\Sigma }}_i`$ of $`B_i`$. We choose $`\overline{\mathrm{\Sigma }}_i`$ so close to $`B_i`$ that $`\overline{\mathrm{\Sigma }}_i\mathrm{\Sigma }^{(j)}=\mathrm{}`$, and the negative integration of $`v`$ yields a diffeomorphism from $`\overline{\mathrm{\Sigma }}_i`$ to a subset of $`\mathrm{\Sigma }_i^{(j)}`$. Step I. For $`j=0,1`$, there exists a disc $`D_j`$ such that $`D_j(_{i1}\overline{\mathrm{\Sigma }}_i)`$ is a normal section of $`(M,v)`$, and the associated branched spine is sliding-equivalent to $`P_j`$. To prove this, we temporarily drop the index $`j`$. We first isotope each $`\mathrm{\Sigma }_i`$, without changing the associated spine, until it contains $`\overline{\mathrm{\Sigma }}_i`$, as suggested in Fig. 11. Note that if $`B_i=\mathrm{}`$ we automatically have $`\mathrm{\Sigma }_i=\overline{\mathrm{\Sigma }}_i`$. Otherwise, we concentrate on one of the annuli $`A`$ of which $`\mathrm{\Sigma }_i\overline{\mathrm{\Sigma }}_i`$ consists. Note that we cannot just shrink $`A`$ leaving the rest of the section unchanged, because we could spoil axiom 4 of the definition of normal section. To actually shrink $`A`$ we first need to “insulate” it, toward the positive direction of $`v`$, by adding to the disc $`\mathrm{\Sigma }_0`$ a strip normal to $`v`$. Figure 12 suggests how to do this. As we modify $`\mathrm{\Sigma }_0`$ as suggested, it is clear that we keep having a “quasi-normal” section, i.e. all axioms except 6 hold. Moreover the corresponding simple branched spines are obtained from each other by the simple sliding moves already mentioned above. To conclude we apply, as above, the fact that a simple branched spine can be transformed via simple sliding moves to a standard one, and the technical result established in , Proposition 4.5.6, according to which standard spines which are equivalent under simple sliding moves are also sliding-equivalent. This proves Step I. The conclusion will now follow quite closely the argument in , Theorem 5.2.1. Step II. There exist discs $`D_0^{}`$ and $`D_1^{}`$ such that $`D_j^{}(_{i1}\overline{\mathrm{\Sigma }}_i)`$ is a normal section of $`(M,v)`$ for $`j=0,1`$, and $`D_0D_0^{}=D_0^{}D_1^{}=D_1^{}D_1=\mathrm{}`$. Choosing a metric on $`M`$, one can construct $`D_0^{}`$ and $`D_1^{}`$ by first taking many very small discs almost orthogonal to $`v`$, and then connecting these discs by strips transversal to $`v`$. Step III. Conclusion in the case $`v_0=v_1`$. If we connect $`D_0`$ and $`D_0^{}`$ by a strip orthogonal to $`v`$, we get a bigger disc $`\stackrel{~}{D}_0`$ such that $`\stackrel{~}{D}_0(_{i1}\overline{\mathrm{\Sigma }}_i)`$ is still a normal section of $`(M,v)`$. We can actually imagine a dynamical process, in which $`D_0`$ is first enlarged to $`\stackrel{~}{D}_0`$, and then is reduced to $`D_0^{}`$, as in Fig. 13. If the transformation is chosen generic enough, at all times axioms 123 and 4 will hold, and axiom 5 will hold at all but finitely many times. This means that the corresponding branched spines are related by simple sliding moves. Similarly, we can replace $`D_0^{}`$ first by $`D_1^{}`$ and then by $`D_1`$. Using the facts quoted above, the conclusion follows. We are left to deal with the general case, where $`(v_t)`$ is a non-constant homotopy. It is then sufficient to take a partition $`0=t_0<t_1<\mathrm{}<t_n=1`$ of $`[0,1]`$, fine enough that $`(M,v_{t_{k1}})`$ and $`(M,v_{t_k})`$ admit a common normal section which gives rise to isomorphic branched spines. 2.1 ###### Remark 2.6 Along the lines of the previous proof we have established the following topological fact, whose statement does not involve spines. Let $`(v_t)`$ be a homotopy of concave fields on $`M`$, let $`B_0,B_1M`$ be balls with $`(B_j,v_j)B_{\mathrm{triv}}^3`$ and $`v_j`$ traversing on $`MB_j`$ for $`j=0,1`$. Then there exist another homotopy $`(v_t^{})`$ between $`v_0`$ and $`v_1`$ and an isotopy $`(B_t)`$ with $`(B_t,v_t)B_{\mathrm{triv}}^3`$ and $`v_t`$ traversing on $`MB_t`$ for all $`t`$. #### Sufficiency of the sliding-MP-moves To show Proposition 2.3 we will find it convenient to use the graphic representation of branched spines introduced in , Section 3.2, but we do not reproduce here the technicalities needed to introduce this representation. Proof of2.3. We start by listing rigid spines. Note first that if a negative sliding-MP-move applies to a spine then also a positive one does, so we only need to consider positive rigidity. The spines with one vertex, shown in Fig. 14, are of course rigid. Using , Proposition 3.3.5, one easily checks that $`(M(P))`$ is $`S_{\mathrm{triv}}^2`$ for the first two spines, and $`S_{\mathrm{triv}}^2S_{\mathrm{triv}}^2`$ for the other two. Now we turn to rigid spines with more than one vertex. Rigidity implies that all edges with distinct endpoints should appear as on the left in Fig. 15. It is not hard to deduce that rigid spines come in a sequence $`P_1^{\mathrm{rig}}`$, $`P_2^{\mathrm{rig}}`$, $`\mathrm{}`$ as shown in the rest of Fig. 15, where $`P_k^{\mathrm{rig}}`$ has $`2k`$ vertices, and $`(M(P_k^{\mathrm{rig}}))`$ is the union of $`S_{\mathrm{triv}}^2`$ together with $`k`$ copies of $`S_{\mathrm{white}}^2`$ and $`k`$ copies of $`S_{\mathrm{black}}^2`$. This classification proves (i). To show (ii) we must prove that: 1. Sequences which contain rigid spines can be replaced by sequences which do not. 2. If two non-rigid spines are related by one snake move then they are also related by a sequence of sliding-MP-moves. For (ii-a), we note that the result of a positive snake move is never rigid. So if a rigid spine $`P`$ appears in a sequence of moves then $`P`$ is the result of a negative snake move $`\mu _1^1:P_1P`$, and a positive snake move $`\mu _2:PP_2`$ is applied to $`P`$. Since all edges of a spine survive through a snake move, there is a version $`\stackrel{~}{\mu }_2`$ of $`\mu _2`$ which applies to $`P_1`$ and a version $`\stackrel{~}{\mu }_1`$ of $`\mu _1`$ which applies to $`P_2`$, and the result $`\stackrel{~}{P}`$ is the same. So can replace the segment $`(P_1,P,P_2)`$ by $`(P_1,\stackrel{~}{P},P_2)`$, and now all the spines involved are non-rigid. Let us turn to (ii-b). The proof results from three steps, to describe which we introduce in Figure 16 another move, called sliding-vertex move, whose unbranched version was already considered in and . Again, taking into account orientations, there are two versions of the move (for each vertex type), but we will ignore this detail. Step 1: if $`v`$ is a vertex of a branched spine $`P`$, $`e`$ is any one of the edges incident to $`v`$, $`P_v`$ is obtained from $`P`$ via the sliding-vertex move at $`v`$, and $`P_e`$ is obtained from $`P`$ via the snake move on $`e`$, then $`P_v`$ and $`P_e`$ are related by sliding-MP-moves. This is proved by an easy case-by-case analysis. It turns out that two MP-moves (a positive and a negative one) are always sufficient. Step 2: let $`v`$, $`P`$ and $`P_v`$ be as above. If $`P`$ and $`P_v`$ are related by sliding-MP-moves, the same is true for $`P`$ and any spine obtained from $`P`$ by a snake move. To see this, use step 1 to successively transform sliding-vertex moves into snake moves and conversely, until the desired snake move is reached. Step 3: if $`P`$ is non-rigid then there exists a vertex $`v`$ such that $`P`$ and $`P_v`$ are related by MP-moves. The vertex $`v`$ is chosen to be an endpoint of an edge to which the positive MP-move applies. The argument is again a long case-by-case one, which refines in a branched context the argument given by Piergallini in . The sequence always consists of three positive moves followed by a negative one. This concludes the proof of (ii), whereas (iii) is evident. 2.3 ## 3 A calculus for framed links We fix in this section a compact manifold $`M`$ and consider the set $`\mathrm{Fram}(M)`$ of isotopy classes of framed links in $`M`$. Since a link isotopy generically avoids a fixed 3-ball, $`\mathrm{Fram}(M)`$ and $`\mathrm{Fram}(\widehat{M})`$ are canonically isomorphic when $`M=S^2`$, so we can restrict to non-closed $`M`$’s and include the closed case as usual. #### Statement Let us fix a branched standard spine $`P`$ of $`M`$. The fact that such a spine always exists was proved as Theorem 3.4.9 in . We call $`\mathrm{C}^1`$ link diagram on $`P`$ an immersion of a disjoint union of circles into $`P`$, with generic intersection with $`S(P)`$ appearing as in Fig. 17, generic self-intersections (crossings), and the usual under-over marking at crossings (as shown in the same figure; here ‘under’ and ‘over’ refer to the field positively transversal to $`P`$). The set of all $`\mathrm{C}^1`$ link diagrams on $`P`$ will be denoted by $`𝒟(P)`$. An element of $`𝒟(P)`$ obviously defines a link. Moreover $`v(P)`$ is transversal to this link, so it defines a framing, and we get an (obviously well-defined) map $`F_{(P,M)}:𝒟(P)\mathrm{Fram}(M)`$. Besides isotopy on $`P`$ through immersions having the same configuration of crossings and intersections with $`S(P)`$, there are several combinatorial moves which of course do not modify the isotopy class of the framed link defined by a diagram. We show a list of moves having this property in Fig. 18. For a reason to be given below, which also explains the apparently weird notation, we call these moves $`\mathrm{C}^1`$-Turaev moves. As we did when we described the sliding-MP-moves in Fig. 6, we are showing in Fig. 18 only the essential physical modifications, without specifying the screw-orientation of the spine and the orientation of its regions. ###### Theorem 3.1 The map $`F_{(P,M)}:𝒟(P)\mathrm{Fram}(M)`$ is surjective, and the equivalence relation it defines is generated by $`\mathrm{C}^1`$-Turaev moves. Our argument, after the easy proof of surjectivity, goes along the following lines: 1. We state the analogue of Theorem 3.1 for non-branched spines, due to Turaev (we include a quick proof for the sake of completeness); 2. We modify Turaev’s result to the case of a branched spine, but allowing non-$`\mathrm{C}^1`$ diagrams; 3. We prove our theorem, showing how to canonically replace each non-$`\mathrm{C}^1`$ diagram by a $`\mathrm{C}^1`$ one along a sequence of modifications. #### Surjectivity Since $`M`$ and $`P`$ are fixed, we write $`F`$ for short. Given a framed link $`L`$ in $`M`$, we can prove that it is contained in the image of $`F`$ as follows: * First, forget the framing and take a generic projection on $`P`$, recalling that $`MPM\times (0,1]`$; * Next, eliminate non-$`\mathrm{C}^1`$ intersections with $`S(P)`$ as shown in Fig. 19 (left). * Finally, give the resulting projection the right framing by adding the necessary numbers of curls. (Here we use the fact that two framings on a given knot differ at most by a finite number of full rotations.) It may be noted that surjectivity of $`F`$ is preserved by restriction to the set of diagrams without crossings. This follows quite easily from the fact that all the regions of $`P`$ have non-empty boundary, as suggested in Fig. 19 (right). This property will not be used below. #### Turaev moves on standard spines The ideas and results of this paragraph are due to Turaev . We temporarily allow $`P`$ to be any standard spine of $`M`$, not a branched one. If each region of $`P`$ is given an arbitrary transverse orientation, the definition of a link diagram $`D`$ makes sense also on $`P`$, but $`D`$ may not define a framing on the associated link, because the strip which runs along a component of $`D`$ on $`P`$ need not be a cylinder, it may be a Möbius strip. So we attach to each component $`D_i`$ of $`D`$ a full or half-integer $`a_i/2`$, depending on the topology of the strip, and we define the framing by giving $`a_i`$ positive half-twists to the strip (recall that $`M`$ is oriented). By diagram on $`P`$ we will actually mean one such pair $`(\{D_i\},\{a_i/2\})`$. We call Turaev moves those shown in Fig. 20, together with the $`\mathrm{R}_{\mathrm{I}\mathrm{I}}`$ and $`\mathrm{R}_{\mathrm{I}\mathrm{I}\mathrm{I}}`$ already shown above. In Fig. 20, for $`\mathrm{R}_\mathrm{I}^{}`$ and $`\mathrm{T}_{\mathrm{I}\mathrm{I}\mathrm{I}}`$, the local orientation must be that of $`\text{}^3`$. ###### Theorem 3.2 Every isotopy class of framed link in $`M`$ is defined by some diagram on $`P`$, and two diagrams define the same class if and only if they are obtained from each other by a sequence of Turaev moves. Proof of3.2. Recall that a framed link can be thought of as an embedded cylinder. Moreover $`M`$ projects onto $`P`$, and the projection of a cylinder generically appears as in Fig. 21. Such a projection easily defines a diagram, the half-integers being sums of $`\pm 1/2`$’s corresponding to the bends of the projection. Moreover, the elementary catastrophes along an isotopy of a projection translate into the moves of Fig. 20, or simple combinations of them. 3.2 #### Turaev moves on branched spines Going back to the case where $`P`$ is branched, we can still apply Theorem 3.2, but now the list of moves becomes slightly longer, if we want to take the branching into account. ###### Proposition 3.3 If $`P`$ is a branched spine then any Turaev move for a diagram on $`P`$ can be expressed as a combination (including inverses) of the moves $`\mathrm{R}_\mathrm{I}^{}`$, $`\mathrm{R}_{\mathrm{I}\mathrm{I}}`$, $`\mathrm{R}_{\mathrm{I}\mathrm{I}\mathrm{I}}`$, $`\mathrm{T}_\mathrm{I}^{}`$, $`\mathrm{T}_{\mathrm{I}^{\prime \prime }}`$, $`\mathrm{T}_{\mathrm{I}\mathrm{I}^{}}`$, $`\mathrm{T}_{\mathrm{I}\mathrm{V}^{}}`$, $`\mathrm{T}_\mathrm{V}^{}`$ shown above, together with the moves $`\mathrm{T}_{\mathrm{I}^{\prime \prime \prime }}`$, $`\mathrm{T}_{\mathrm{I}\mathrm{I}^{\prime \prime }}`$, $`\mathrm{T}_{\mathrm{I}\mathrm{I}\mathrm{I}^{}}`$, $`\mathrm{T}_{\mathrm{I}\mathrm{V}^{\prime \prime }}`$, $`\mathrm{T}_{\mathrm{V}^{\prime \prime }}`$ shown in Fig. 22 Proof of3.3. The branching can be interpreted as a loss of symmetry of a spine, so each of Turaev’s moves, when viewed as a move on a branched spine, generates many different ones according to the position of the diagram with respect to the branching. The result is a list much longer than that given in the statement, but one can show that all the moves omitted from the statement are generated by the moves included. Two examples are provided in Figg. 23 and 24. 3.3 #### $`𝐂^\mathrm{𝟏}`$ moves We can now conclude the proof of Theorem 3.1 (surjectivity having been shown above). We are left to show that if $`D`$ and $`D^{}`$ are $`\mathrm{C}^1`$-diagrams on $`P`$ which define the same framed link, then they are related by a sequence of $`\mathrm{C}^1`$ Turaev moves. By Theorem 3.2 and Proposition 3.3, there exists a sequence $`D=D_0D_1\mathrm{}D_{n1}D_n=D^{}`$ where each move $`D_{ii}D_i`$ is one of those listed in Proposition 3.3. In particular the $`D_i`$’s with $`0<i<n`$ can be non-$`\mathrm{C}^1`$ and can have a non-zero half-integer attached to them. We will now show how to construct a modified sequence $`D=\stackrel{~}{D}_0\stackrel{~}{D}_1\mathrm{}\stackrel{~}{D}_{n1}\stackrel{~}{D}_n`$ with the following properties: 1. each $`\stackrel{~}{D}_i`$ is a $`\mathrm{C}^1`$ diagram with number 0 attached; 2. each $`\stackrel{~}{D}_i`$ is obtained from $`\stackrel{~}{D}_{i1}`$ by a sequence of $`\mathrm{C}^1`$ Turaev moves; 3. each $`\stackrel{~}{D}_i`$ differs from $`D_i`$ for the presence of some extra curls; in particular each component $`\stackrel{~}{D}_i^{(j)}`$ of $`\stackrel{~}{D}_i`$ has a natural companion $`D_i^{(j)}`$ in $`D_i`$, with the property that, as unframed knots, the knots associated to $`\stackrel{~}{D}_i^{(j)}`$ and $`D_i^{(j)}`$ are both contained in a solid torus $`T_i^{(j)}`$ and parallel to the core of the torus; 4. the framed knots associated to $`\stackrel{~}{D}_i^{(j)}`$ and $`D_i^{(j)}`$ are framed-isotopic within $`T^{(j)}`$. Requirement 4 is the crucial technical point of our proof. To verify that the requirement is stronger than just framed isotopy, note that in $`D^2\times S^1`$ the framings on the core $`\{0\}\times S^1`$ are parameterized by the integers, but, when $`D^2\times S^1`$ is mirrored in its boundary to get $`S^2\times S^1`$, only two inequivalent framings remain (corresponding to even and odd integers). We assume for a moment the sequence $`\stackrel{~}{D}_i`$ to exist, and we show how to conclude. The transformation from $`D`$ to $`\stackrel{~}{D}_n`$ is made with $`\mathrm{C}^1`$ Turaev moves, so we only need to compare $`\stackrel{~}{D}_n`$ and $`D^{}=D_n`$, which by assumption differ for some curls. Using $`\mathrm{C}^1`$ Turaev moves we can easily make all these curls slide until they are consecutive on the diagram. We recall now that there are four local pictures for a curl, depending on its local contributions $`\pm `$ to the framing and to the winding number . Assumption 4 now implies that the algebraic sum of local contributions to the framing vanishes. Therefore we can cancel the curls in pairs, either by moves $`\mathrm{R}_\mathrm{I}^{}`$ (when the local contributions to the winding number are the same), or by a combination of moves $`\mathrm{R}_{\mathrm{I}\mathrm{I}}`$ and $`\mathrm{R}_{\mathrm{I}\mathrm{I}\mathrm{I}}`$ (when the contributions cancel). This shows the conclusion. We are left to define the sequence $`\stackrel{~}{D}_i`$. The idea is simply not to perform the moves which change the half-integer or introduce cusps, and show that the sequence of moves can be followed anyway. While doing this we need to keep track of the portions where the new diagram $`\stackrel{~}{D}_i`$ differs from $`D_i`$, which we do by marking a neighbourhood of the portion as a shadowed box. The moves which change the colour or introduce cusps are $`\mathrm{R}_\mathrm{I}^{}`$ (in both directions), $`\mathrm{T}_{\mathrm{I}^{\prime \prime \prime }}`$, $`\mathrm{T}_{\mathrm{I}\mathrm{I}\mathrm{I}^{}}`$, and $`\mathrm{T}_{\mathrm{V}^{\prime \prime }}`$, and we show in Figg. 25 and 26 what we replace them with. For move $`\mathrm{R}_\mathrm{I}^{}`$ it has been necessary to be more specific because, in the original definition of the move, two different ones were actually defined at the same time. To show that $`\{\stackrel{~}{D}_i\}`$ can indeed be constructed we must now show that after performing the construction up to some level $`k`$ we can still still follow the rest of the sequence and go on with the replacements of Figg. 25 and 26. By construction $`\stackrel{~}{D}_k`$ differs from $`D_k`$ only within some shadowed boxes. We denote by $`\mu _k`$ the move $`D_kD_{k+1}`$ and explain how to lift it to a move on $`\stackrel{~}{D}_k`$. First, note that a shadowed box lying within a region of $`P`$ and containing a curl does not interfere with $`\mu _k`$ whatever its type (but it may be necessary to add some $`\mathrm{R}_{\mathrm{I}\mathrm{I}}`$’s and $`\mathrm{R}_{\mathrm{I}\mathrm{I}\mathrm{I}}`$’s to replace isotopy supported within the region). We fix now our attention on a shadowed box $`B`$ which lies on $`S(P)`$ and contains a smoothed cusp, and examine the various instances for $`\mu _k`$ with respect to $`B`$. If $`\mu _k`$ is of type $`\mathrm{R}_{\mathrm{I}\mathrm{I}}^{\pm 1}`$, $`\mathrm{R}_{\mathrm{I}\mathrm{I}\mathrm{I}}^{\pm 1}`$, $`\mathrm{T}_\mathrm{I}^{}^{\pm 1}`$, $`\mathrm{T}_{\mathrm{I}^{\prime \prime }}^{\pm 1}`$ $`\mathrm{T}_{\mathrm{I}\mathrm{I}^{}}^{\pm 1}`$, or $`\mathrm{T}_{\mathrm{I}\mathrm{V}^{}}^{\pm 1}`$, then it obviously does not interfere with $`B`$, so we can just perform $`\mu _k`$ on $`\stackrel{~}{D}_k`$. If $`\mu _k`$ is one of the moves $`\mathrm{R}_\mathrm{I}^{\pm 1}`$, $`\mathrm{T}_{\mathrm{I}^{\prime \prime \prime }}`$, $`\mathrm{T}_{\mathrm{I}\mathrm{I}\mathrm{I}^{}}`$ or $`\mathrm{T}_{\mathrm{V}^{\prime \prime }}`$ then again it does not interfere with $`B`$, and we can perform the appropriate replacement from Figg. 25 or 26, getting a move from $`\stackrel{~}{D}_k`$ to $`\stackrel{~}{D}_{k+1}`$. If $`\mu _k`$ is a move of type $`\mathrm{T}_{\mathrm{I}\mathrm{I}^{\prime \prime }}^{\pm 1}`$ or $`\mathrm{T}_{\mathrm{I}\mathrm{V}^{\prime \prime }}^{\pm 1}`$ then it may interfere with $`B`$. However, since it does not create or destroy cusps, $`\mu _k`$ can be translated on $`\stackrel{~}{D}_k`$ as a combination of moves which do not involve cusps, i.e. allowed from the statement. We are only left to deal with the case where $`\mu _k`$ is one of the moves $`\mathrm{T}_{\mathrm{I}^{\prime \prime \prime }}^1`$, $`\mathrm{T}_{\mathrm{I}\mathrm{I}\mathrm{I}^{}}^1`$ or $`\mathrm{T}_{\mathrm{V}^{\prime \prime }}^1`$ which destroy cusps. By construction the cusp(s) to be destroyed still appear in $`\stackrel{~}{D}_k`$ as smoothed cusps within shadowed boxes, so we can destroy them also from $`\stackrel{~}{D}_k`$ by means of allowed moves. The crucial properties 3 and 4 hold by construction, and the conclusion eventually follows. 3.1 ## 4 A calculus for pseudo-Legendrian links <br>in combed manifolds We will deal in this section with the set $`\mathrm{PLeg}`$ of equivalence classes of triples $`(M,v,L)`$ already described in the introduction. Its combinatorial counterpart will be given by the set $$=\{(P,D):P,D𝒟(P)\}$$ where $``$ is as in Section 2 and $`𝒟(P)`$ is as in Section 3. The reconstruction map $`(P,D)L(P,D)`$ is here defined by noting that $`D`$ defines a link transversal to $`v(P)`$ in $`M(P)`$, and hence also a link transversal to $`\widehat{v}(P)`$ in $`\widehat{M}(P)`$. According to what we stated after Proposition 2.3 we will actually drop from $``$ the sliding-MP-rigid spines, and ignore the snake move. If for a fixed $`P`$ we consider the effect on $`L(P,D)`$ of the $`\mathrm{C}^1`$-Turaev moves on $`D`$, we see that the class of $`L(P,D)`$ is in general modified by the first Reidemeister move $`\mathrm{R}_\mathrm{I}^{}`$ of Fig. 18 (see also Section 5), but not by the other moves, which we will therefore call pseudo-Legendrian Turaev moves. Other moves which obviously do not change $`L(P,D)`$ up to equivalence are the sliding-MP-moves on $`P`$ which do not involve $`D`$ (these moves permit to follow $`D`$ along the modification of $`P`$, so they are well-defined for pairs). It is not hard to see that before performing a sliding-MP-move on $`P`$ it is always possible to modify $`D`$ by pseudo-Legendrian Turaev moves to a diagram which is not involved in the sliding-MP-move, and the diagram after the sliding-MP-move is well-defined up to pseudo-Legendrian Turaev moves. For this reason we will freely speak of sliding-MP-moves also for pairs. The following will be established below. ###### Theorem 4.1 The map $`L:\mathrm{PLeg}`$ is surjective, and the equivalence relation defined by $`L`$ is generated by pseudo-Legendrian Turaev moves and sliding-MP-moves. #### Fixed-spine statement Recall from the definition that the map $`L`$ of the statement of Theorem 4.1 involves the passage from $`P`$ to $`(M(P),v(P))`$ and then to $`(\widehat{M}(P),\widehat{v}(P))`$. As already pointed out, this is necessary if one wants to be able to deal with non-traversing fields. However, if one happens to have a concave traversing field, one can directly encode this field by a spine, without first removing a ball, and one can investigate how isotopy of links transversal to the field reflects on link diagrams on the spine. The following is shown below: ###### Proposition 4.2 Let $`P`$ be a branched spine. Fix a representative of $`(M(P),v(P))`$ and an embedding of $`P`$ in $`M(P)`$ transversal to $`v(P)`$. Then every link transversal to $`v(P)`$ is represented by a $`\mathrm{C}^1`$ diagram on $`P`$. Moreover two $`\mathrm{C}^1`$ diagrams define the same link up to isotopy through links transversal to $`v(P)`$ if and only if they are related by pseudo-Legendrian Turaev moves. #### From fixed to variable spine We show in this paragraph how to deduce Theorem 4.1 from Proposition 4.2. First of all, to prove surjectivity, we consider a triple $`(M,v,L)`$ representing an element of $`\mathrm{PLeg}`$. Using a normal section as in Proposition 2.4, we can obtain a spine $`P`$ which encodes the equivalence class of $`(M,v)`$ in the sense of Theorem 2.1. Moreover $`P`$ comes with an embedding in $`M`$ transversal to $`v`$. Now, a neighbourhood of $`P`$ can be identified to $`M(P)`$ and its complement is isomorphic to $`B_{\mathrm{triv}}^3`$. Using the flow generated by $`v`$ in this ball we can now isotope $`L`$ through links transversal to $`v`$ to a link which lies in $`M(P)`$, and the first assertion of Proposition 4.2 implies that $`L`$ is represented by a diagram $`D`$ on $`P`$. Summing up, we see that $`(M,v,L)`$ is represented by $`(P,D)`$, and surjectivity of $`L`$ is proved. To conclude we must now show that two pairs $`(P_0,D_0)`$ and $`(P_1,D_1)`$ are equivalent via pseudo-Legendrian Turaev moves when $`L(P_0,D_0)=L(P_1,D_1)`$. Spelling out the relation of pseudo-Legendrian isotopy, which defines $`\mathrm{PLeg}`$, we assume that $`P_0`$ and $`P_1`$ embed in the same manifold $`M`$ and that there exist a field homotopy $`(v_t)_{t[0,1]}`$ and a link isotopy $`(L_t)_{t[0,1]}`$ on $`M`$ such that: 1. for $`i=0,1`$, the link $`L_i`$ is the one defined by $`D_i`$, and the field $`v_i`$ is positively transversal to $`P_i`$ and restricts to $`B_{\mathrm{triv}}^3`$ on the complement of $`P_i`$; 2. $`L_t`$ is transversal to $`v_t`$ for all $`t`$. Given $`t[0,1]`$, we note that for $`\|ts\|1`$ the link $`L_s`$ is transversal to $`v_t`$, and that a branched spine for $`v_t`$ (in the sense repeatedly used above) is a branched spine also for $`v_s`$. So we can subdivide $`[0,1]`$ into subintervals $`[t_{i1},t_i]`$ so that: 1. $`L_{t_{i1}}`$ is isotopic to $`L_{t_i}`$ through links transversal to $`v_{t_i}`$; 2. $`v_{t_i}`$ has a spine $`P_{t_i}`$ which is also a spine for $`v_{t_{i1}}`$. Now let $`D_{t_i}`$ be a diagram for $`L_{t_i}`$ on $`P_{t_i}`$. Since both $`P_{t_{i1}}`$ and $`P_{t_i}`$ are spines for $`v_{t_i}`$, we can transform $`P_{t_{i1}}`$ into $`P_{t_i}`$ via sliding-MP-moves. Using pseudo-Legendrian Turaev moves we can now follow $`D_{t_{i1}}`$ along this sequence of moves, getting a diagram $`D_{t_i}^{}`$ on $`P_{t_i}`$. Since the sequence of sliding-MP-moves can be realized in $`M`$ so that each spine of the sequence is a branched spine for $`v_{t_i}`$, we deduce that the link defined by $`D_{t_i}^{}`$ is isotopic to $`L_{t_{i1}}`$, and hence to $`L_{t_i}`$, through links transversal to $`v_{t_i}`$. Proposition 4.2 now implies that $`D_{t_i}^{}`$ and $`D_{t_i}`$ are related by pseudo-Legendrian Turaev moves on $`P_{t_i}`$. This shows that $`(P_{t_i},D_{t_i})`$ is obtained from $`(P_{t_{i1}},D_{t_{i1}})`$ via the moves of the statement, and the conclusion follows by iteration. 4.1 #### Fixed-spine proof We will establish now Proposition 4.2, writing just $`M`$ and $`v`$ for $`M(P)`$ and $`v(P)`$. For the sake of simplicity we will assume that $`v`$ is tangent to $`M`$ along only one curve (denoted by $`\gamma `$), but our arguments extends almost verbatim to the general case of more than one curve. We fix in $`M`$ an annulus $`A`$ which connects $`\gamma `$ to $`S(P)`$ as shown in a cross-section in Fig. 27. Note that $`A`$ is almost but not quite embedded: it has double point at the vertices of $`P`$. Since we will only need to consider $`A`$ locally and away from vertices of $`P`$, this fact will not disturb us. We choose coordinates $`(\rho ,\theta )[0,1]\times [0,2\pi ]`$ on $`A`$, where $`\rho =0`$ corresponds to $`S(P)`$ and $`\rho =1`$ to $`\gamma `$. Near $`A`$ we can also define a coordinate $`z[\epsilon ,\epsilon ]`$ by integrating $`v`$. Now let $`L`$ be transversal to $`v`$, and assume by general position that $`L`$ intersects $`A`$ only at points with $`0<\rho <1`$, that no two such intersections have the same coordinate $`\theta `$, and that at all the intersections the tangent direction to $`L`$ has non-zero components in all three coordinates $`\rho ,\theta ,z`$. Depending on the sign of these components, we can divide the points of $`LA`$ into four types, shown in Fig. 28 ($`L`$ is dashed when it lies over $`A`$ and dotted when it lies under $`A`$). We consider now the projection $`\pi `$ of $`MA`$ onto $`P`$ along the orbits of $`v`$, as shown in Fig. 29. Of course $`LA`$ locally projects to a $`\mathrm{C}^1`$-strand on $`P`$, and by general position we can assume that $`\pi (LA)`$ locally appears as a $`\mathrm{C}^1`$-diagram. We are only left to extend the diagram at the points of $`LA`$, which we do locally in Fig. 30. The top part of this figure actually refers to a simplified situation, because other strands of $`L`$ already projected on $`P`$ may locally interfere. We show at the bottom of the same figure in one example how to deal with this fact. The resulting diagram of course represents $`L`$, and we have proved the first assertion in Proposition 4.2. To prove the second assertion we must now examine an isotopy of $`L`$ through links transversal to $`v`$, and hence examine first-order violations of genericity of $`L`$ with respect to $`A`$ and $`\pi `$. All the elementary accidents which do not involve $`A`$ of course correspond to pseudo-Legendrian Turaev moves. We are left to deal with the following accidents: 1. $`L`$ intersects $`A`$ at a point of $`S(P)A`$; 2. at a point of $`LA`$, the tangent direction to $`L`$ has vanishing $`\rho `$-coordinate; 3. similarly, with the $`\theta `$-coordinate; 4. similarly, with the $`z`$-coordinate. In Fig. 31 we show the situation just before and just after each of these accidents, and we analyze the corresponding transformations of the diagrams constructed as in Fig. 30. In all cases one easily sees that indeed the transformation is generated by pseudo-Legendrian Turaev moves: the number of moves needed is respectively one, zero (isotopy within regions), three and two. By simplicity in Fig. 31 we have ignored the possible interference of other strands of $`L`$ projected on $`P`$, but the conclusion is valid anyway (some Reidemeister moves must be added in the general case). 4.2 ## 5 Applications and speculations In this section we will discuss some consequences of the calculi described above, and mention some natural questions and problems which we put forward for further investigation. The section is split into two subsections. ### 5.1 Winding number, torsion, and finite-order invariants In this section we employ our realizations of $`\mathrm{Fram}`$ and $`\mathrm{PLeg}`$ in connection with winding number, Maslov index, torsion, and finite-order invariants of pseudo-Legendrian knots. #### Relative winding number We spell out in this paragraph the analogue of Trace’s result on knot diagrams in $`\text{}^3`$. We confine ourselves to knots for the sake of simplicity, but essentially the same holds for links. ###### Proposition 5.1 Let $`(v_0,K_0)`$ and $`(v_1,K_1)`$ be pseudo-Legendrian pairs in a manifold $`M`$, where $`K_0`$ and $`K_1`$ are oriented knots. Assume that $`v_0`$ and $`v_1`$ are homotopic relatively to $`M`$, and that $`K_0`$ and $`K_1`$ are isotopic as oriented framed knots. Then, up to pseudo-Legendrian isotopy on $`(v_1,K_1)`$, we can assume that $`v_1=v_0`$ and that $`K_1`$ differs from $`K_0`$ only within a region of $`(M,v_0)`$ isomorphic to $`(\text{}^3,/z)`$, where $`K_0`$ is a straight horizontal line and $`K_1`$ has either some positive or some negative double curls (shown in Fig. 32). Proof of5.1. Choose branched spines $`P_0`$ and $`P_1`$ of $`v_0`$ and $`v_1`$ according to Proposition 2.4, and use Proposition 4.2 to represent $`K_0`$ and $`K_1`$ on $`P_0`$ and $`P_1`$ respectively by $`\mathrm{C}^1`$-diagrams $`D_0`$ and $`D_1`$. Since $`v_0`$ and $`v_1`$ are homotopic, a sequence of sliding-MP moves connects $`P_1`$ to $`P_0`$. Following $`D_1`$ along this sequence of moves we get a pseudo-Legendrian isotopy, so we can assume that $`v_1=v_0`$ and $`P_1=P_0`$. Now $`D_0`$ and $`D_1`$ define on $`P_0`$ framed-isotopic oriented knots, so by Theorem 3.1 they are related by $`\mathrm{C}^1`$-Turaev moves. If along the sequence of moves there is no $`\mathrm{R}_\mathrm{I}^{}`$, we deduce pseudo-Legendrian equivalence of $`K_0`$ and $`K_1`$. If however there is some $`\mathrm{R}_\mathrm{I}^{}`$, we replace it by a pseudo-Legendrian move as shown in Fig. 33. One easily sees that this replacement can be done consistently along the sequence of moves. The result is a pseudo-Legendrian isotopy between $`D_1`$ and a diagram which differs from $`D_0`$ only for some double curls. These double curls can of course be slid to be consecutive. Now, it is precisely the content of that up to moves $`\mathrm{R}_{\mathrm{I}\mathrm{I}}`$ and $`\mathrm{R}_{\mathrm{I}\mathrm{I}\mathrm{I}}`$ there are only the types of double curls shown in Fig. 32, and that a positive and a negative double curl cancel out. 5.1 Under the assumptions of the previous proposition one may be tempted to define a relative winding number $`w(K_1,K_0)`$ as the (algebraic) number of double curls by which the diagram of $`K_1`$ differs from the diagram of $`K_0`$. This number is however not well-defined in general, as one easily sees in $`S^2\times S^1`$ with vector field parallel to the $`S^1`$-factor, because in this case a double curl on a diagram contained in $`\text{}^2=S^2\{\mathrm{}\}`$ can always be removed by isotoping the diagram through $`\mathrm{}`$. This seems to suggest that not $`w(K_1,K_0)`$, but maybe $`w(K_1,K_0)[\mu _{K_0}]H_1(E(K_0);\text{})`$ is well-defined, where $`E(K_0)`$ is the exterior of $`K_0`$ and $`\mu _{K_0}`$ is the meridian. We will show this fact under the additional assumption that $`K_0`$ is ‘good’ (see and below for explanations). #### Torsion invariants and good knots In we have defined the Reidemeister-Turaev torsion of an Euler structure with simple boundary, and we have applied this notion to define the torsion of pseudo-Legendrian knots. As an absolute invariant torsion contains a sort of lift of the classical Alexander invariant. We will discuss in this paragraph the information carried by torsion as a relative invariant of two pseudo-Legendrian framed-isotopic knots $`K_0,K_1`$ in the same concave combed manifold $`(M,v)`$. We recall from that this information is most easily expressed when $`K_0`$ has the property of being good. Goodness depends only on the isotopy class of $`K_0`$ as a framed knot, and it means that a certain quotient of the mapping class group of $`E(K_0)`$ acts trivially on the space of Euler structures on $`E(K_0)`$. We omit the precise definition here, but we recall that many knots indeed are good (for instance, all are good if $`M`$ is a homology sphere, and most hyperbolic knots are good). When $`K_0`$ is good, the information carried by torsion as a relative invariant depends only on $`\alpha (v\|_{E(K_0)},f(v\|_{E(K_1)})H_1(E(K_0);\text{})`$, where $`f\mathrm{Diff}_0(M)`$ maps $`K_1`$ to $`K_0`$ as framed knots, and $`\alpha `$ is the first obstruction for two vector fields to be homotopic relative to the boundary. So the next result means that for good knots the relative winding number gives a well-defined invariant, and all the information torsion can capture is contained in the relative winding number. The statement involves all the assumptions and notations of the present and previous paragraph. ###### Proposition 5.2 $`\alpha (v\|_{E(K_0)},f(v\|_{E(K_1)})=w(K_1,K_0)[\mu _{K_0}]H_1(E(K_0);\text{}).`$ Proof of5.2. We note first that $`w(K_1,K_0)`$ and $`[\mu _{K_0}]`$ depend on the choice of an orientation on $`K_0`$, but their product does not, so the statement makes sense. For the proof, note that by goodness we can just assume that $`K_0`$ and $`K_1`$ differ as in the statement of Proposition 5.1, and that $`f`$ is supported on a neighbourhood of the region where $`K_0`$ and $`K_1`$ differ. The conclusion then follows directly from Proposition 2.17 of . 5.2 Even if we have not discussed finite-order invariants yet, we note here that Proposition 5.2 implies that torsion is a weaker invariant than the finite-order ones for Legendrian knots in a given homotopy class of Legendrian immersions. To our knowledge the only known examples of framed-isotopic knots distinguished by such invariants are those due to Tchernov , and we believe that they are all good (at least, they certainly are good when the ambient manifold is $`S^2\times S^1`$). Now one sees that in all of Tchernov examples $`w(K_1,K_0)[\mu _{K_0}]=0`$, so torsion definitely cannot distinguish. On the other hand, the definition of torsion does not require fixing a homotopy class of Legendrian immersions, so torsion and finite-order invariants are in some sense complementary. We will state in the rest of this paragraph some interesting consequences of Proposition 5.2, always assuming the knots involved to be good. For simplicity, as in Proposition 5.2, we stick to knots transverse to a given field $`v`$ on a given $`M`$, but we remind that the relation of pseudo-Legendrian isotopy also involves a homotopy of $`v`$. ###### Corollary 5.3 Under the same assumptions as in Proposition 5.2, suppose furthermore that $`[\mu _{K_0}]`$ has infinite order in $`H_1(E(K_0);\text{})`$, so $`w(K_1,K_0)\text{}`$ is well-defined. Then the following facts are pairwise equivalent: 1. $`w(K_1,K_0)=0`$; 2. $`K_0`$ and $`K_1`$ have trivial relative torsion invariants; 3. $`K_0`$ and $`K_1`$ are pseudo-Legendrian isotopic. Proof of5.3. Equivalence of (1) and (3) follows from the definition of $`w(K_1,K_0)`$. Implication (1)$``$(2) follows from Proposition 5.2, and the opposite implication follows by taking the torsion associated to a representation $`\phi `$ of $`H_1(E(K_0);\text{})`$ such that $`\phi ([\mu _{K_0}])`$ has infinite order (see for details). 5.3 If $`M`$ is a homology sphere and $`K`$ is a pseudo-Legendrian knot in $`(M,v)`$ we have shown in that the rotation number $`\mathrm{rot}_v(K)`$, also called Maslov index, can be defined just as in the case where $`K`$ is Legendrian in a contact structure. Now: ###### Lemma 5.4 If $`M`$ is a homology sphere, $`K_0`$ and $`K_1`$ are pseudo-Legendrian in $`(M,v)`$ and framed isotopic, then $`w(K_1,K_0)=\frac{1}{2}(\mathrm{rot}_v(K_1)\mathrm{rot}_v(K_0))`$. (Concerning the statement, note that $`\mathrm{rot}_v(K_1)\mathrm{rot}_v(K_0)`$ must be even if $`K_0`$ and $`K_1`$ are framed-isotopic, otherwise one of $`\{K_0,K_1\}`$ would lift to a closed path in a spin structure on $`M`$, and the other one would not: a contradiction. A proof is easily obtained by isotoping $`K_1`$ as stated in Proposition 5.1.) Lemma 5.4 gives another proof of the fact that $`w(K_1,K_0)\text{}`$ can be defined when $`M`$ is a homology sphere. Moreover, it could be used to show goodness of knots in a homology sphere by a more direct argument than that given in . We conclude this paragraph by showing the result stated in the introduction and asking a question which naturally arises from it. Proof of0.1. Equivalence of (1) and (2) comes from Corollary 5.3. Equivalence of (2) and (3) comes from Lemma 5.4. Equivalence of (3) and (4) follows from Corollary5.3 and the fact that the first homology group of the complement of a knot in $`M`$ is infinite cyclic and generated by a meridian. Equivalence of (3) and (5) is an application of Gromov’s $`h`$-principle (see ). 0.1 ###### Question 5.5 Let $`(M,v)`$ be an arbitrary combed manifold, let $`K_0`$ and $`K_1`$ be pseudo-Legendrian in $`(M,v)`$ and framed-isotopic, and assume that they are homotopic through pseudo-Legendrian immersions. Does this imply that $`w(K_1,K_0)[\mu _{K_0}]=0`$? (We do not think that the opposite implication can be true in general, in particular when $`[\mu _{K_0}]`$ has finite order.) #### Absolute winding number We concentrate in this paragraph on fields $`v`$ such that $`(v^{})=0`$, where $``$ denotes the Euler class and the choice of the metric is of course immaterial. Condition $`(v^{})=0`$ is equivalent to the existence of another non-vanishing field $`x`$ always transversal to $`v`$. Since the ambient manifold is oriented, this is also equivalent to the fact that $`v`$ extends to a framing $`(v,x,y)`$, i.e. a global trivialization of the tangent bundle to $`M`$. Assume now that $`K`$ is an oriented knot transversal to $`v`$. Then, taking the projection of the tangent vector to $`K`$ on the unit sphere of the $`(x,y)`$-plane, and computing the degree, we can define a rotation number $`\mathrm{rot}_{(v,x)}(K)`$. ###### Remark 5.6 $`\mathrm{rot}_{(v,x)}(K)`$ is invariant under simultaneous homotopy $`(v_t,x_t)`$ and isotopy $`(K_t)`$ such that $`x_t`$ and $`v_t`$ are transversal to $`v_t`$ for all $`t`$. Moreover $`\mathrm{rot}_{(v,x)}(K)`$ is independent of $`x`$ when $`M`$ is a homology sphere, and it equals the Maslov index already discussed above. Assume now that $`K_0`$ and $`K_1`$ are both transversal to $`v`$. Within the proof of Proposition 5.1 we have shown that $`K_1`$ can be isotoped through knots transversal to $`v`$ to a knots which differs from $`K_0`$ by double curls only. ###### Remark 5.7 $`\mathrm{rot}_{(v,x)}(K_1)\mathrm{rot}_{(v,x)}(K_0)`$ is independent of $`x`$ and equals twice the number of double curls by which $`K_0`$ and $`K_1`$ differ, up to isotopy transversal to $`v`$. The previous remark shows that the relative winding number is well-defined (without any assumption on the knots) if one restricts to knots transversal to a given $`v`$ with $`(v^{})=0`$, and one views the knots up to isotopy transversal to $`v`$ (as opposed to pseudo-Legendrian isotopy, which involves also a homotopy of $`v`$). More on the difference between transversal isotopy and pesudo-Legendrian isotopy will be said below. ###### Remark 5.8 Combining the previous two remarks one gets yet another proof that the relative winding number is well-defined in up to pseudo-Legendrian isotopy in a homology sphere. #### Finite-order invariants We formally state and motivate in this paragraph the conjecture announced in the introduction. Let $`\xi `$ be an oriented contact structure on $`M`$ (which we assume to be closed by simplicity), and let $`v`$ be a field positively transversal to $`\xi `$. Consider the spaces $`\mathrm{Leg}(M,\xi )`$, $`\mathrm{PLeg}^{\mathrm{weak}}(M,v)`$ and $`\mathrm{Fram}(M)`$ of $`\xi `$-Legendrian, $`v`$-transverse, and framed knots in $`M`$, with the appropriate equivalence relations (namely $`\xi `$-Legendrian, pseudo-Legendrian, and framed isotopy). Enlarge these spaces by allowing immersions of $`S^1`$ rather than embeddings, and take path-connected components l, p and f, with $`\text{l}\text{p}\text{f}`$. (Concerning $`\mathrm{PLeg}`$, note that a path is a family $`(K_t,v_t)_{t[0,1]}`$ with $`v_0=v_1=v`$.) Given an Abelian group $`A`$ one can define, using the customary Vassiliev-Goussarov skein relations, the spaces $`V_\text{l}^n(A)`$, $`V_\text{p}^n(A)`$ and $`V_\text{f}^n(A)`$ of $`A`$-valued order-$`n`$ invariants under Legendrian, pseudo-Legendrian and framed isotopy respectively. Since a Legendrian isotopy is pseudo-Legendrian, and a pseudo-Legendrian isotopy is framed, using restrictions we get a commutative diagram of homomorphisms: $$\begin{array}{ccc}V_\text{f}^n(A)& \stackrel{\varphi _{\text{f},\text{p}}^n}{}& V_\text{p}^n(A)\\ & & \\ & \varphi _{\text{f},\text{l}}^n& \varphi _{\text{p},\text{l}}^n\\ & & \\ & & V_\text{l}^n(A).\end{array}$$ Tchernov’s arguments imply that all three $`\varphi `$’s are always injective, and our conjecture is that $`\varphi _{\text{p},\text{l}}^n`$ is always an isomorphism. By again, the conjecture is equivalent to showing that every finite-order Legendrian invariant is automatically invariant also under pseudo-Legendrian isotopy. The generalized Fuchs-Tabachnikov theorem (see and ) states that $`\varphi _{\text{f},\text{l}}^n`$ is an isomorphism in many cases (e.g. if $`M`$ is a homology sphere), so $`\varphi _{\text{p},\text{l}}^n`$ is also an isomorphism in these cases. Tchernov has provided the only known examples in which $`\varphi _{\text{f},\text{l}}^n`$ is not an isomorphisms, namely he has exhibited elements of $`V_\text{l}^n(A)`$ which do not lift to $`V_\text{f}^n(A)`$. Our impression is that these elements do lift to $`V_\text{p}^n(A)`$, which would imply that $`\varphi _{\text{p},\text{l}}^n`$ is indeed an isomorphism in all known cases. Truthness of our conjecture would imply that Legendrian finite-order invariants are only sensitive to the homotopy class of a contact structure, and in particular that they cannot capture tightness. ### 5.2 Pseudo-Legendrian vs. Legendrian knots After the work of Eliashberg , we know that on a closed manifold an overtwisted contact structure is determined up to isotopy by its homotopy class as a plane field. We discuss in this section the extent to which this fact extends in presence of a pseudo-Legendrian link. We start with an open question which arises from the results of Section 4 and will lead us to the connection with overtwisted contact structures. #### Fixed vs. variable spine for pseudo-Legendrian links We will adopt in this paragraph the viewpoint which allows to dismiss automorphisms of manifolds, fixing $`M`$ and considering spines and moves embedded in $`M`$, as explained after the statement of Proposition 2.3. Theorems 2.1 and 4.1 and Proposition 4.2 leave the following question open: given an embedded spine $`PM`$ representing a concave combing on $`M`$, what intrinsic topological object is represented by $`\mathrm{C}^1`$-diagrams on $`P`$ up to pseudo-Legendrian Turaev moves? Let us introduce some notation to formalize the situation. We denote by $`𝒟^{\mathrm{PLeg}}(P)`$ the set of equivalence classes of $`\mathrm{C}^1`$-diagrams on $`P`$ up to pseudo-Legendrian Turaev moves. We also fix a representative $`v`$ of the combing carried by $`P`$ (so, $`v`$ is positively transversal to $`P`$ and restricts to $`B_{\mathrm{triv}}^3`$ on the complement of $`P`$). We consider now the set of links in $`M`$ transversal to $`v`$, and we denote by $`\mathrm{PLeg}^{\mathrm{weak}}(M,v)`$ the quotient space under the relation of existence of a pseudo-Legendrian isotopy, i.e. a path $`(L_t,v_t)_{t[0,1]}`$ as usual, with $`v_0=v_1=v`$. We also denote by $`\mathrm{PLeg}^{\mathrm{strong}}(M,v)`$ the (bigger) quotient obtained by forcing $`(v_t)`$ to be constant. So $`\mathrm{PLeg}^{\mathrm{strong}}(M,v)`$ is just the set of equivalence classes of links transversal to $`v`$. Using Proposition 4.2 one sees that the operation of turning a diagram into a link defines a bijection $$\psi ^{\mathrm{strong}}:𝒟^{\mathrm{PLeg}}(P)\mathrm{PLeg}^{\mathrm{strong}}(M,v).$$ (This is not quite the content of Proposition 4.2, because here $`(M,v)`$ is $`(\widehat{M}(P),\widehat{v}(P))`$ rather that $`(M(P),v(P))`$, but a link isotopy can be modified to avoid a $`B_{\mathrm{triv}}^3`$, and the conclusion follows.) Bijectivity of $`\psi ^{\mathrm{strong}}`$ is significant if one imagines to have started with the pair $`(M,v)`$, and to have constructed $`P`$ from a normal section of $`v`$, as in Proposition 2.4. It is however less significant if one assumes only $`P`$ to be given from the beginning, because in this case $`v`$ is actually well-defined only up to homotopy, and fixing a representative looks artificial. The natural map to consider is in this case $$\psi ^{\mathrm{weak}}:𝒟^{\mathrm{PLeg}}(P)\mathrm{PLeg}^{\mathrm{weak}}(M,v),$$ obtained by composition with the projection $`\mathrm{PLeg}^{\mathrm{strong}}(M,v)\mathrm{PLeg}^{\mathrm{weak}}(M,v)`$. This map is of course surjective, and one can ask whether it is injective or not. Some remarks are in order: 1. Theorem 4.1 implies that if $`\psi ^{\mathrm{weak}}(D)=\psi ^{\mathrm{weak}}(D^{})`$ then there exists a circular sequence $`P=P_0P_1\mathrm{}P_n=P`$ of sliding-MP-moves and diagrams $`D_i,D_i^{}𝒟(P_i)`$ with $`D_0=D`$, $`D_n^{}=D^{}`$, $`D_iD_i^{}`$ a pseudo-Legendrian Turaev move, and $`D_{i+1}`$ the companion of $`D_i^{}`$ through $`P_iP_{i+1}`$. Checking the injectivity of $`\psi ^{\mathrm{weak}}`$ corresponds to the (purely combinatorial) question whether such a sequence $`(P_i,D_i)`$ can be replaced by one with constant $`P_i`$. 2. Using Theorem 2.1 and the fact that $`\mathrm{C}^1`$ diagrams can be followed through sliding-MP-moves, one sees quite easily that injectivity of $`\psi ^{\mathrm{weak}}`$ actually depends only on the combing carried by $`P`$, not on $`P`$ itself. 3. Injectivity of $`\psi ^{\mathrm{weak}}`$ is equivalent to injectivity of the projection $`\mathrm{PLeg}^{\mathrm{strong}}(M,v)\mathrm{PLeg}^{\mathrm{weak}}(M,v)`$, a purely topological question. Injectivity of this projection may appear very unlikely at first sight, since it basically corresponds to the fact that a homotopy can be replaced by an isotopy. However one can remark that injectivity of projection depends only on the homotopy class of $`v`$, rather than $`v`$ itself, so one can assume that $`v`$ is transversal to an overtwisted contact structure. For overtwisted structures, after the work of Eliashberg , it is indeed true that homotopy implies isotopy, but the presence of the link of course somewhat modifies the situation. We will expound this theme in the next paragraph. #### Overtwisted structures and overtwisted knot complements We fix in this paragraph an overtwisted contact structure $`\xi `$ on $`M`$ (which we assume to be closed by simplicity) and a field $`v`$ positively transversal to $`\xi `$. We will denote by $`\mathrm{Leg}(M,\xi )`$ the space of Legendrian links in $`(M,\xi )`$ up to Legendrian isotopy. In , having also in mind the facts mentioned in the previous paragraph, we put forward the question of whether the natural map $$\mathrm{Leg}(M,\xi )\mathrm{PLeg}^{\mathrm{weak}}(M,v)$$ is a bijection. A fact implying that this map is not injective in some cases was recently communicated to us by E. Giroux . He was able to construct triples $`(M,\xi ,K)`$ where $`\xi `$ is overtwisted, $`K`$ is $`\xi `$-Legendrian, and $`\xi \|_{MK}`$ is tight. Let us apply a Lutz twist away from $`K`$ to get a new structure $`\xi ^{}`$ such that $`\xi ^{}`$ is homotopic to $`\xi `$ as a plane field on $`M`$, and $`\xi ^{}\|_{MK}`$ is overtwisted. Using Eliashberg’s classification we consider $`\phi \mathrm{Diff}_0(M)`$ such that $`\xi ^{}=\phi ^{}(\xi )`$, and define $`K^{}=\phi (K)`$. By construction $`K`$ and $`K^{}`$ have the same image in $`\mathrm{PLeg}^{\mathrm{weak}}(M,v)`$, but of course they are inequivalent in $`\mathrm{Leg}(M,\xi )`$. To avoid the phenomenon discovered by Giroux we consider in $`\mathrm{Leg}(M,\xi )`$ the subset $`\mathrm{Leg}^{\mathrm{OT}}(M,\xi )`$ given by links whose complement is overtwisted. We start by showing: ###### Proposition 5.9 The natural map $`\mathrm{Leg}^{\mathrm{OT}}(M,\xi )\mathrm{PLeg}^{\mathrm{weak}}(M,v)`$ is surjective. Proof of5.9. Let $`L`$ be transversal to $`v`$, and fix a metric on $`M`$. Let $`\eta `$ be a positive contact structure near $`L`$ with $`\eta =v^{}`$ on $`K`$ (such an $`\eta `$ is unique up to isomorphism). Extend $`\eta `$ to any plane field homotopic to $`v^{}`$ (and hence to $`\xi `$) on $`M`$. So $`\eta `$ is a plane distribution which has a contact zone, and $`L`$ lies in this contact zone. The technique of Eliashberg now allows to homotope $`\eta `$ away from its contact zone to an overtwisted contact structure $`\xi ^{}`$. The resulting $`\xi ^{}`$ is now isotopic to $`\xi `$, again by Eliashberg’s result. If $`\phi \mathrm{Diff}_0(M)`$ and $`\xi ^{}=\phi ^{}(\xi )`$ we define $`L^{}=\phi (L)`$. By construction $`(L^{},v)`$ is pseudo-Legendrian isotopic to $`(L,v)`$, and surjectivity is proved. 5.9 We cannot presently state whether the map $`\mathrm{Leg}^{\mathrm{OT}}(M,\xi )\mathrm{PLeg}^{\mathrm{weak}}(M,v)`$ is injective or not in general. We only give a partial argument (based on the techniques of Eliashberg again), and mention where the difficulty arises. Assume that $`L_0`$ and $`L_1`$ are $`\xi `$-Legendrian with overtwisted complements and define equivalent pseudo-Legendrian links. Then there exists a continuous family $`(L_t,\xi _t)_{t[0,1]}`$, where $`\xi _0=\xi _1=\xi `$ but $`\xi _t`$ is only a plane field for $`t0,1`$. Eliashberg’s contactization methods for homotopies, together with the uniqueness of contact structures in the neighbourhood of Legendrian links, should in our opinion allow to replace such a $`(\xi _t)_{t[0,1]}`$ by another one in which each $`\xi _t`$ is a contact structure (and still contains $`L_t`$ as a Legendrian link). Applying Gray’s theorem we get an isotopy $`(\phi _t)_{t[0,1]}`$ such that $`\xi _t=\phi _t^{}(\xi _0)`$. Setting $`\stackrel{~}{L}_t=\phi _t(L_t)`$ we get a Legendrian isotopy between $`L_0`$ and $`\phi _1(L_1)`$. The question whether $`\phi _1(L_1)`$ is automatically Legendrian isotopic to $`L_1`$, at least for some classes of manifolds, now depends on the analysis of the group $`\mathrm{Aut}(M,\xi )\mathrm{Diff}_0(M)`$, which we leave unsettled for the time being. benedett@dm.unipi.it petronio@dm.unipi.it Dipartimento di Matematica Via F. Buonarroti, 2 I-56127, PISA (Italy)
warning/0001/math0001081.html	ar5iv	text	"# Trace Functionals of the Kontsevich Quantization\n\n## 1 Introduction\n\nThe gist of deformation (...TRUNCATED)
warning/0001/cond-mat0001132.html	ar5iv	text	"# Theory of magnetic order in the three-dimensional spatially anisotropic Heisenberg model\n\n## I (...TRUNCATED)
warning/0001/hep-ph0001120.html	ar5iv	text	"# 1 Introduction\n\n## 1 Introduction\n\nIn this note we study scattering of virtual and real photo(...TRUNCATED)
warning/0001/cond-mat0001156.html	ar5iv	text	"# Thermodynamics and dielectric anomalies of DMAAS and DMAGaS crystals in the phase transitions reg(...TRUNCATED)
warning/0001/hep-ph0001105.html	ar5iv	text	"# 1 Introduction\n\n## 1 Introduction\n\nRecent polarized deep-inelastic experiments have yielded v(...TRUNCATED)

End of preview. Expand in Data Studio

Marin Markdownified Ar5iv

Markdownified Ar5iv transforms academic papers from arXiv into clean, structured Markdown format consisting of 22.34B tokens across two splits. This dataset preserves th content while making it accessible for language model training on academic text.

	Value
Tokens	19 552 307 274
Primary source	https://sigmathling.kwarc.info/resources/ar5iv-dataset-2024/
File format	JSONL
License	C-UDA-1.0 (mirrors upstream Ar5iv licenses)

Processing and Cleaning Pipeline

Our conversion pipeline combines several sophisticated techniques to transform raw Wikipedia HTML into high-quality Markdown:

HTML Preprocessing: We start with the Ar5iv dump in Extended DOLMA format, which provides HTML representations of academic papers with metadata.
Structural Cleanup
- The abstract is transformed into a proper section heading for consistent document structure
- LaTeX equations are carefully preserved using inline ($...$) and display ($$...$$) notation
- Code blocks and listings maintain proper formatting with appropriate line breaks
Noise Reduction:
- Author information is removed
- Title page elements are streamlined to avoid redundancy
- The Ar5iv footer is removed to eliminate conversion metadata
- Figure captions are removed to focus on the main content
- Bibliography sections, footnotes, and citation links are removed
Formatting Cleanup:
- List items are cleaned to prevent duplicate numbering patterns (e.g., "1. 1.")
- Content before the first main section (typically metadata) is removed
- Equation tables are converted to inline elements for better rendering
DOM Simplification: We employ a custom-enhanced version of Resiliparse that preserves semantic HTML structure. Rather than flattening to plain text, we retain important elements like headings, paragraphs, lists while removing scripts, tracking code, and boilerplate.
Markdown Conversion: Our custom Markdownify implementation transforms the simplified DOM into clean Markdown. The final output stores each article as a JSON object containing the Markdown text and essential metadata.

Dataset Variants

The Markdownified Ar5iv dataset comes in two variants:

Ar5iv No Problem (2.74B tokens): Papers that were converted without significant issues or warnings during the HTML generation process. This subset represents the cleanest and most reliable papers.
Ar5iv Warning (19.6B tokens): Papers that generated warnings during conversion from LaTeX to HTML. While still valuable, these may contain occasional formatting artifacts.

Usage Example

from datasets import load_dataset

ds = load_dataset(
    "marin-community/ar5iv-warning-markdown",
    split="train",
    streaming=True
)

for article in ds.take(3):
    print(article["text"])

Citation

If you use this dataset in your research, please cite both the original Wikipedia contributors and our work:

@misc{markdownified_ar5iv_2024,
  title        = {Markdownified Ar5iv},
  author       = {The Marin Community},
  year         = {2024},
  url          = {https://huggingface.co/datasets/marin-community/ar5iv-warning-markdown}
}