Topological Recursion, explicit ABCD tensors and implementation

B. Eynard · 2025

Read Topological Recursion, explicit ABCD tensors and implementation by B. Eynard (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-00128.txt.

Introduction

Topological recursion is a method that builds complex quantities step by step from a starting curve. B. Eynard presents explicit expressions for its ABCD tensors in a 2025 study, covering the most classical classes of spectral curves. The study also discusses how to implement the recursion algorithmically.

Topological recursion was invented to compute large-size expansions of random matrices. It has since proved useful across mathematical physics, from enumerative geometry and string theory to integrable systems and combinatorics. Explicit tensor formulas turn an abstract machine into something researchers can actually compute with, widening its use in those fields. Spectral curves are the input data from which the recursion builds its answers. For a new reader, the value is in the worked forms. They turn a grand scheme into sums that a small team can run, test, and check by hand.

Catalog record

CREATORB. Eynard
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE82 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-00128

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Topological Recursion, explicit ABCD tensors and implementation B. Eynard Abstract Abstract We provide explicit expressions of ABCD tensors for the most classical classes of spectral curves. And we discuss algorithmic implementation of Topological Recursion. 1 Introduction Topological Recursion was invented in order to compute recursively the large size asymptotic expansion of random matrices [ 6 ] , and then it was realized that Topological Recursion is ubiquitous in mathematical physics, beyond random matrices, it has applications to enumerative geometry, string theory, integrable systems, conformal field theory, combinatorics of maps, statistical physics…etc. In [ 5 ] Topological Recursion was formulated as a process that to a spectral curve (a Riemann surface with extra structure) associates a sequence of differential forms, called the invariants of the spectral curve. These invariants often coincide with classical geometric invariants. For example: spectral curve \to T.R. invariants Airy y=\sqrt{x} Witten-Kontsevich intersection numbers Sine y=\sin{(2\pi z)} Weil-Petersson volumes Mirror of toric CY3-fold P(e^{x},e^{y})=0 Gromov-Witten invariants semi-circle (GUE) y=\sqrt{4-x^{2}} higher Catalan numbers A-polynomial of knot P(e^{x},e^{y})=0 Jones polynomial classical stress–energy y^{2}=T(x) CFT conformal blocs tensor etc… There exists many lectures, introductory lecture notes and review articles on topological recursion and its applications. In [ 8 ] , Topological Recursion was reformulated as ” Quantum Airy Structure ”, in terms of 4 tensors called A,B,C,D , which was further developed in [ 1 ] . The ABCD tensor formulation amounts to decomposing the invariants on a basis of forms, and consider the recursion for the coefficients. This is how Topological recursion is actually implemented algorithmically. In this article, we give explicit expressions of the ABCD tensors for many classical spectral curves, in a way used for actual implementation in a python package. We also discuss the algorithmic issues and implementation. 2 Topological Recursion and ABCD tensors Consider a Riemann surface \overset{\circ}{{\Sigma}} called the base curve. Most often \overset{\circ}{{\Sigma}}={\mathbb{C}}P^{1}={\mathbb{C}}\cup\{\infty\} the Riemann sphere, and this is what we shall consider in all examples below. 2.1 Spectral curve A spectral curve is the following data \left\{\begin{array}[]{l}{\Sigma}=\text{Riemann surface}\cr x:{\Sigma}\to\overset{\circ}{{\Sigma}}\quad=\text{ramified covering map}\cr{\rm y}=ydx=\text{ meromorphic 1-form on }{\Sigma}\cr B(z_{1},z_{2})=\text{fundamental 2nd kind differential },B\in H^{0}({\Sigma}\times{\Sigma},K_{\Sigma}\overset{\text{Sym}}{\boxtimes}K_{\Sigma})_{\mathop{\,\rm Res\,}=1}\cr R=\{\text{ramification points}\}\cr a\in R\ :\ \ \sigma_{a},\ \text{local anayltic involution such that }x(\sigma_{a}(z))=x(z)\text{ and }\sigma_{a}(a)=a.\end{array}\right. (2-1) For regular spectral curves, ramification points are the zeros of dx , and we assume that they are simple zeros. Therefore near z=a , x is 2:1 , and there exists a unique local involution \sigma_{a}\neq\text{Id} , such that \sigma_{a}(a)=a and x(\sigma_{a}(z))=x(z) in a neighborhood of a . We shall consider non-regular spectral curves in section 7 below. Then we define the recursion kernel : K_{a}(z_{1},z)=\frac{dS(z_{1},z)}{(y(z)-y(\sigma_{a}(z)))dx(z)}\quad\text{where}\quad dS(z_{1},z)=\frac{1}{2}\int_{\sigma_{a}(z)}^{z}B(z_{1},.) (2-2) which is meromorphic for z in a neighborhood of the ramification point a , with a simple pole at z=a . 2.2 TR invariants Topological Recursion is a recursive procedure that to a spectral curve, associates its invariants \omega_{g,n} defined as follows: \displaystyle\omega_{0,1} \displaystyle= \displaystyle ydx (2-3) \displaystyle\omega_{0,2} \displaystyle= \displaystyle B (2-4) \displaystyle\omega_{0,3}(z_{1},z_{2},z_{3}) \displaystyle= \displaystyle\sum_{a\in R}\mathop{\,\rm Res\,}_{z\to a}K_{a}(z_{1},z)\left[B(z,z_{2})B(\sigma_{a}(z),z_{3})+B(z,z_{3})B(\sigma_{a}(z),z_{2})\right] (2-5) \displaystyle= \displaystyle-2\sum_{a\in R}\mathop{\,\rm Res\,}_{z\to a}K_{a}(z_{1},z)B(z,z_{2})B(z,z_{3}) (2-6) \displaystyle= \displaystyle-\sum_{a\in R}\mathop{\,\rm Res\,}_{z\to a}\frac{B(z,z_{1})B(z,z_{2})B(z,z_{3})}{dx(z)dy(z)} (2-7) \displaystyle\omega_{1,1}(z_{1}) \displaystyle= \displaystyle\sum_{a\in R}\mathop{\,\rm Res\,}_{z\to a}K_{a}(z_{1},z)B(z,\sigma_{a}(z)) (2-8) and for n\geq 1 and 2g-2+n>1 \displaystyle\omega_{g,n}(z_{1},\dots,z_{n}) \displaystyle= \displaystyle 2\sum_{j=2}^{n}\sum_{a\in R}\mathop{\,\rm Res\,}_{z\to a}K_{a}(z_{1},z) (2-12) \displaystyle\qquad B(z,z_{j})\omega_{g,n}(\sigma_{a}(z),z_{2},\dots,z_{j-1},z_{j+1},\dots,z_{n}) \displaystyle+\sum_{a\in R}\mathop{\,\rm Res\,}_{z\to a}K_{a}(z_{1},z)\Big[\omega_{g-1,n+1}(z,\sigma_{a}(z),z_{2},\dots,z_{n}) \displaystyle+\sum_{g_{1}+g_{2}=g}\sum^{\text{stable}}_{I_{1}\sqcup I_{2}=\{z_{2},\dots,z_{n}\}}\omega_{g_{1},1+|I_{1}|}(z,I_{1})\omega_{g_{2},1+|I_{2}|}(\sigma_{a}(z),I_{2})\Big] For n=0 , the TR invariants are defined for g\geq 2 as \displaystyle\mathcal{F}_{g}=\omega_{g,0} \displaystyle= \displaystyle\frac{1}{2-2g}\sum_{a\in R}\mathop{\,\rm Res\,}_{z\to a}\omega_{g,1}(z)F_{0,1}(z) (2-13) where dF_{0,1}=\omega_{0,1} (any antiderivative of \omega_{-0,1}={\rm y}=ydx gives the same result). Here we won’t consider \mathcal{F}_{0} and \mathcal{F}_{1} . Homogeneity property : If we rescale ydx\to\lambda ydx (i.e. we rescale either x or y or both), then \omega_{g,n}\mapsto\lambda^{2-2g-n}\omega_{g,n} (2-14) Moreover \omega_{g,n} is invariant under translations x\mapsto x+c . 2.3 Basis of 1-forms Consider a vector space V , subspace of the space of meromorphic 1-forms on {\Sigma} , such that for all 2g-2+n>0 we have \omega_{g,n}\in V\otimes\dots\otimes V. V can be chosen as a subspace of meromorphic 1-forms that have poles at the ramification points, generated by the coefficients of the Taylor expansion of B(z,z^{\prime}) at z^{\prime}\in R , i.e. by derivatives of B(z^{\prime},z) with respect to z^{\prime} , at z^{\prime}\in R : V\subset\mathop{\operatorname{Span}}_{a\in R,d\in{\mathbb{Z}}_{+}}\Big\langle(d/dx(z^{\prime}))^{d}(B(z^{\prime},z)/d\zeta_{a}(z^{\prime}))_{z^{\prime}=a}\Big\rangle (2-15) (using the local coordinate \zeta_{a}(z^{\prime})=\sqrt{x(z^{\prime})-x(a)} near a\in R ). V can be chosen smaller for example when the spectral curve has symmetries. Consider a basis \{d\xi_{\alpha}\}_{\alpha\in J} of V . \omega_{g,n} can be decomposed on the basis: \displaystyle\omega_{g,n}(z_{1},\dots,z_{n}) \displaystyle= \displaystyle\sum_{\alpha_{1};\dots;\alpha_{n}}F_{g,n}[\alpha_{1};\dots;\alpha_{n}]\prod_{i=1}^{n}d\xi_{\alpha_{i}}(z_{i}) (2-16) where we assume that the basis is chosen so that the number of terms is finite. Despite our notation d\xi_{\alpha} needs not be an exact form. The goal is to be able to compute the coefficients F_{g,n}[\alpha_{1};\dots;\alpha_{n}] . We also consider the antiderivative (only defined locally): \displaystyle\mathcal{F}_{g,n}(z_{1},\dots,z_{n}) \displaystyle= \displaystyle\sum_{\alpha_{1};\dots;\alpha_{n}}F_{g,n}[\alpha_{1};\dots;\alpha_{n}]\prod_{i=1}^{n}\xi_{\alpha_{i}}(z_{i}). (2-17) Examples of basis: • any basis of 1-forms d\xi_{a,d}(z) , such that d\xi_{a,d}(z) has a pole of degree 2d+2 at z=a , and d\xi_{a,d}(z)+d\xi_{a,d}(\sigma_{a}(z)) has no pole at z=a . • d\xi_{a,d}(z)=\mathop{\,\rm Res\,}_{z^{\prime}\to a}(x(z^{\prime})-x(a))^{-(d+1/2)}B(z,z^{\prime}) • d\xi_{a,0}(z)=\mathop{\,\rm Res\,}_{z^{\prime}\to a}(x(z^{\prime})-x(a))^{-1/2}B(z,z^{\prime}) and d\xi_{a,d+1}(z)=d(d\xi_{a,d}(z)/dx(z)) . • many other choices are possible… 2.4 TR and ABCD Tensors Topological Recursion ( 2-12 ), written for the coefficients amounts to: \displaystyle F_{0,3}[\alpha_{1};\alpha_{2};\alpha_{3}] \displaystyle= \displaystyle A[\alpha_{1};\alpha_{2};\alpha_{3}] (2-18) \displaystyle F_{1,1}[\alpha_{1}] \displaystyle= \displaystyle D[\alpha_{1}] (2-19) and for 2g-2+n>1 : \displaystyle F_{g,n}[\alpha_{1};\dots;\alpha_{n}] \displaystyle= \displaystyle 2\sum_{i=2}^{n}\sum_{\alpha\in J}B[\alpha_{1};\alpha_{i}|\alpha]\ F_{g,n-1}[\alpha;\alpha_{2};\dots;\widehat{\alpha_{i}};\dots;\alpha_{n}] (2-22) \displaystyle\sum_{\alpha,\alpha^{\prime}\in J}C[\alpha_{1}|\alpha;\alpha^{\prime}]\ \Big[F_{g-1,n+1}[\alpha;\alpha^{\prime};\alpha_{2};\dots;\alpha_{n}] \displaystyle+\sum_{g_{1}+g_{2}=g}\sum^{\text{stable}}_{I_{1}\cup I_{2}=\{\alpha_{2};\dots;\alpha_{n}\}}F_{g_{1},1+|I_{1}|}(\alpha;I_{1})F_{g_{2},1+|I_{2}|}(\alpha^{\prime};I_{2})\Big] where we have introduced the following tensors: \bullet A\in V^{*}\otimes V^{*}\otimes V^{*} : \displaystyle\sum_{\alpha_{1},\alpha_{2},\alpha_{3}}A[\alpha_{1},\alpha_{2},\alpha_{3}]\ d\xi_{\alpha_{1}}(z_{1})d\xi_{\alpha_{2}}(z_{2})d\xi_{\alpha_{3}}(z_{3}) (2-24) \displaystyle= \displaystyle\sum_{a\in R}\mathop{\,\rm Res\,}_{z\to a}\frac{B(z,z_{1})B(z,z_{2})B(z,z_{3})}{dx(z)dy(z)} (2-25) \displaystyle= \displaystyle-2\sum_{a\in R}\mathop{\,\rm Res\,}_{z\to a}K_{a}(z_{1},z)B(z,z_{1})B(z,z_{2}) (2-26) \bullet B\in V^{*}\otimes V^{*}\otimes V : \displaystyle\sum_{\alpha_{1}}\sum_{\alpha_{2}}B[\alpha_{1},\alpha_{2}|\alpha_{3}]d\xi_{\alpha_{1}}(z_{1})d\xi_{\alpha_{2}}(z_{2}) \displaystyle= \displaystyle\sum_{a\in R}\mathop{\,\rm Res\,}_{z\to a}K_{a}(z_{1},z)B(z,z_{2})d\xi_{\alpha_{3}}(\sigma_{a}(z)) (2-28) \displaystyle= \displaystyle-\sum_{a\in R}\mathop{\,\rm Res\,}_{z\to a}K_{a}(z_{1},z)B(z,z_{2})d\xi_{\alpha_{3}}(z) (2-29) \bullet C\in V^{*}\otimes V\otimes V : \displaystyle\sum_{\alpha_{1}}C[\alpha_{1}|\alpha_{2},\alpha_{3}]d\xi_{\alpha_{1}}(z_{1}) \displaystyle= \displaystyle\sum_{a\in R}\mathop{\,\rm Res\,}_{z\to a}K_{a_{1}}(z_{1},z)d\xi_{\alpha_{2}}(z)d\xi_{\alpha_{3}}(\sigma_{a}(z)) (2-31) \bullet D\in V^{*} : \displaystyle\sum_{\alpha_{1}}D[\alpha_{1}]d\xi_{\alpha_{1}}(z_{1}) \displaystyle= \displaystyle\sum_{a\in R}\mathop{\,\rm Res\,}_{z\to a}K_{a}(z_{1},z)B(z,\sigma_{a}(z)). (2-33) These tensors are the same as those of Quantum Airy Structures of Kontsevich Soibelman [ 8 ] and [ 1 ] . Now we shall compute explicitely A,B,C,D for several examples of spectral curves. 3 Airy and KdV 3.1 Spectral curve The KdV spectral curve with times \{\tilde{t}_{k}\}_{k\in{\mathbb{Z}}_{+}} is defined as \left\{\begin{array}[]{l}{\Sigma}={\mathbb{C}}\cr x=z^{2}+\tilde{t}_{1}\cr y=-\frac{1}{2}\sum_{k\geq 3}\tilde{t}_{k}z^{k-2}\cr B=\frac{dz_{1}dz_{2}}{(z_{1}-z_{2})^{2}}\cr R=\{0\},\ \ \sigma_{0}(z)=-z\end{array}\right. (3-1) Airy is the case \tilde{t}_{k}=-2\delta_{k,3}. (3-2) The KP-times are related to KdV-times as follows: \displaystyle t_{k}=-\delta_{k,1}\tilde{t}_{1}+\sum_{j\geq 0}\begin{pmatrix}-k/2\cr j\end{pmatrix}\tilde{t}_{k+2j}\tilde{t}_{1}^{j}. (3-3) In particular if \tilde{t}_{1}=0 , KdV and KP times are the same t_{k}=\tilde{t}_{k} . The Minimal model of type (p,2) is such that \tilde{t}_{k}=0 for k>p+2 . Airy is the minimal model (1,2) . The minimal model (3,2) is often called pure-gravity , or Painlevé 1 . The minimal model (5,2) is often called Lee-Yang . The case \tilde{t}_{2k+3}=t_{2k+3}=\frac{(-1)^{k}(2\pi)^{2k}}{(2k+1)!} is called Weil-Petersson . 3.2 Basis We define \displaystyle\xi_{d}(z)=-\frac{(2d-1)!!}{z^{2d+1}}\quad,\quad d\xi_{d}(z)=\frac{(2d+1)!!dz}{z^{2d+2}}. (3-4) We have \displaystyle B(z,z_{1}) \displaystyle= \displaystyle\sum_{k=1}^{\infty}kz^{k-1}dz\ \frac{dz_{1}}{z_{1}^{k+1}} (3-5) keeping only the odd part under the involution \sigma_{0}(z)=-z , we have \displaystyle B_{\text{odd}}(z,z_{1}) \displaystyle= \displaystyle\sum_{k=0}^{\infty}(2k+1)z^{2k}dz\ \frac{dz_{1}}{z_{1}^{2k+2}} (3-6) \displaystyle= \displaystyle\sum_{k=0}^{\infty}\frac{1}{(2k-1)!!}z^{2k}dz\ d\xi_{d}(z_{1}) (3-7) whose antiderivative wrt z is then \displaystyle dS_{\text{odd}}(z_{1};z) \displaystyle= \displaystyle\sum_{k=0}^{\infty}\frac{1}{(2k+1)!!}z^{2k+1}\ d\xi_{d}(z_{1}). (3-9) 3.3 Kernel The kernel is \displaystyle K_{0}(z_{1},z) \displaystyle= \displaystyle\frac{dS_{\text{odd}}(z_{1};z)}{(y(z)-y(-z))dx(z)} (3-10) \displaystyle= \displaystyle\frac{-1}{2zdz}\ \left(\sum_{k=0}^{\infty}\frac{1}{(2k+1)!!}z^{2k}\ d\xi_{d}(z_{1})\right)\left(\sum_{k\geq 0}\tilde{t}_{2k+3}z^{2k}\right)^{-1} (3-11) \displaystyle= \displaystyle\frac{-1}{2\tilde{t}_{3}}\ \left(\sum_{k=0}^{\infty}\frac{1}{(2k+1)!!}z^{2k}\ d\xi_{d}(z_{1})\right)\left(1+\sum_{k\geq 1}\frac{\tilde{t}_{2k+3}}{\tilde{t}_{3}}z^{2k}\right)^{-1}\ \frac{1}{zdz} (3-12) Let us define: \displaystyle T_{0} \displaystyle= \displaystyle 1 (3-14) \displaystyle T_{k} \displaystyle= \displaystyle\sum_{n=1}^{k}\frac{(-1)^{n}}{\tilde{t}_{3}^{n}}\sum_{j_{1}+\dots+j_{n}=k,\ j_{i}>0}\tilde{t}_{3+2j_{1}}\dots\tilde{t}_{3+2j_{n}}. (3-15) We have \displaystyle K_{0}(z_{1},z) \displaystyle= \displaystyle\frac{-1}{2\tilde{t}_{3}}\left(\sum_{d=0}^{\infty}\frac{d\xi_{d}(z_{1})}{(2d+1)!!}z^{2d}\right)\left(\sum_{k\geq 0}T_{k}z^{2k}\right)\ \frac{1}{zdz}. (3-16) 3.4 First invariants We have \displaystyle\omega_{0,3}(z_{1},z_{2},z_{3}) \displaystyle= \displaystyle\mathop{\,\rm Res\,}_{z\to 0}K_{0}(z_{1},z)(B(z,z_{2})B(-z,z_{3})+B(z,z_{3})B(-z,z_{2})) (3-17) \displaystyle= \displaystyle-2\mathop{\,\rm Res\,}_{z\to 0}K_{0}(z_{1},z)B(z,z_{2})B(z,z_{3}) (3-18) \displaystyle= \displaystyle\frac{1}{\tilde{t}_{3}}\frac{dz_{1}}{z_{1}^{2}}\mathop{\,\rm Res\,}_{z\to 0}\left(1-z^{2}/z_{1}^{2}\right)^{-1}\left(1+\sum_{k\geq 1}T_{k}z^{2k}\right)\ \frac{1}{zdz}\frac{dz^{2}dz_{2}dz_{3}}{(z-z_{2})^{2}(z-z_{3})^{2}} (3-19) \displaystyle= \displaystyle\frac{1}{\tilde{t}_{3}}\prod_{i=1}^{3}\frac{dz_{i}}{z_{i}^{2}} (3-20) \displaystyle= \displaystyle\frac{1}{\tilde{t}_{3}}\prod_{i=1}^{3}d\xi_{0}(z_{i}). (3-21)

Read the complete text