This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

On the 2D Yang-Mills/Hurwitz Correspondence

Jonathan Novak

1. Introduction

1.1. Graph theory

Let G\mathrm{G} be a connected finite simple graph with vertices π,ρ,σ,\pi,\rho,\sigma,\dots. Let G\mathbb{C}\mathrm{G} be the free Hilbert space over the vertices of G\mathrm{G} modeled as the space of kets |π,|ρ,|σ,|\pi\rangle,|\rho\rangle,|\sigma\rangle,\dots. The adjacency operator KEndGK\in\operatorname{End}\mathbb{C}\mathrm{G} encodes the edges of G\mathrm{G},

(1) σ|K|ρ={1, if ρ,σ adjacent0, otherwise.\langle\sigma|K|\rho\rangle=\begin{cases}1,\text{ if }\rho,\sigma\text{ adjacent}\\ 0,\text{ otherwise}.\end{cases}

More generally, the sequence of operators K1,K2,K3,K^{1},K^{2},K^{3},\dots is meaningful in that σ|Kr|ρ\langle\sigma|K^{r}|\rho\rangle counts rr-step walks ρσ\rho\to\sigma in G\mathrm{G}. The exponential adjacency operator Ψt=etK\Psi_{t}=e^{-tK}, which for regular graphs is essentially the heat kernel [13], encodes this information in its matrix elements,

(2) σ|Ψt|ρ=r=0(t)rr!σ|Kr|ρ.\langle\sigma|\Psi_{t}|\rho\rangle=\sum_{r=0}^{\infty}\frac{(-t)^{r}}{r!}\langle\sigma|K^{r}|\rho\rangle.

The algebraic approach to counting walks in G\mathrm{G} is to find a basis of G\mathbb{C}\mathrm{G} in which KK and hence Ψt\Psi_{t} act diagonally [53].

Instead of asking how many walks there are between two given vertices in G\mathrm{G}, we may ask how far apart they are. This information is encoded by the distance operator DEndGD\in\operatorname{End}\mathbb{C}\mathrm{G}, whose matrix

(3) σ|D|ρ=d(ρ,σ)\langle\sigma|D|\rho\rangle=\mathrm{d}(\rho,\sigma)

in the vertex basis tabulates geodesic distance d\mathrm{d} in G\mathrm{G}. Spectral properties of DD were first explored by Graham and Lovasz [32] and have been much studied since; see [2].

Another way to encode distance in graphs is to use linear operators L1,L2,L3,L_{1},L_{2},L_{3},\dots corresponding to metric level sets,

(4) σ|Lr|ρ={1, if d(ρ,σ)=r0, otherwise.\langle\sigma|L_{r}|\rho\rangle=\begin{cases}1,\text{ if }\mathrm{d}(\rho,\sigma)=r\\ 0,\text{ otherwise}.\end{cases}

Note that L1=KL_{1}=K and LrL_{r} is zero for all rr larger than the diameter of G\mathrm{G}. The exponential distance operator,

(5) Ωq=r=0qrLr,\Omega_{q}=\sum_{r=0}^{\infty}q^{r}L_{r},

is the entrywise exponential of DD,

(6) σ|Ωq|ρ=qd(ρ,σ).\langle\sigma|\Omega_{q}|\rho\rangle=q^{\mathrm{d}(\rho,\sigma)}.

This matrix is natural from a statistical physics perspective: it tabulates the Boltzmann weight eβde^{-\beta\mathrm{d}} corresponding to goedesic distance at inverse temperature β=logq\beta=-\log q.

The general study of exponential distance operators in graph theory is just beginning; see [10]. To the best of this author’s knowledge, the first appearance of Ωq\Omega_{q} was in Zagier’s study of deformed commutation relations in quantum physics [58], where the exponential distance operator on the Cayley graph of the symmetric group as generated by the set of Coxeter transpositions (ii+1)(i\ i+1) plays a pivotal role. The analysis in [58] culminates in an explicit formula for the determinant of Ωq\Omega_{q} showing that its zeros lie on the unit circle, implying the existence of a Hilbert space representation of the qq-commutation relations for all 1<q<1-1<q<-1, interpolating between bosons and fermions.

1.2. Hurwitz theory

The setting of this paper is the Cayley graph of the symmetric group as generated by the full conjugacy class of transpositions, which has the feature that its exponential adjacency and distance operators Ψt\Psi_{t} and Ωq\Omega_{q} are simultaneously and explicitly diagonalizable. This is not the case for the Coxeter-Cayley graph of the symmetric group. On the other hand, in the all-transpositions case the zeros of detΩq\det\Omega_{q} are unstable: they converge on the origin as the degree of the symmetric group increases.

The all-transpositions Cayley graph is intimately linked to Hurwitz theory, a classical branch of enumerative geometry concerned with counting maps between compact Riemann surfaces [11], and we therefore refer to it as the Hurwitz-Cayley graph. Cover counting is associated with the exponential adjacency operator Ψt\Psi_{t} on this graph. For example, loop enumeration on the Hurwitz-Cayley graph is algebraically equivalent to computing a diagonal matrix element of Ψt\Psi_{t} and geometrically equivalent to counting simple branched covers of the Riemann sphere, a classical and much-studied problem [21].

It is a special feature of the Hurwitz-Cayley graph that matrix elements of the exponential distance operator Ωq\Omega_{q} and its inverse also count walks [48]. The walks enumerated by σ|Ωq±1|ρ\langle\sigma|\Omega_{q}^{\pm 1}|\rho\rangle are “monotone” with respect to an edge labeling of the Hurwitz-Cayley graph related to the representation theory [46] and order theory [45] of the symmetric group. Thus Ωq\Omega_{q} corresponds to a monotone version of Hurwitz theory [24, 25, 26, 27, 28] which turns out to have many applications in random matrix theory and related areas.

1.3. Yang-Mills theory

The idea that two-dimensional Yang-Mills theory with UN\mathrm{U}_{N} gauge group becomes Hurwitz theory in the large NN limit [33] has been intensively studied by physicists [14, 15, 35, 36, 54], who noticed that eigenvalues of the exponential adjacency operator Ψt\Psi_{t} of the Hurwitz-Cayley graph appear in the partition function of YM2YM_{2}. The Yang-Mills/Hurwitz correspondence works perfectly in the Calabi-Yau case, where spacetime is a torus, even manifesting mirror symmetry [19]. For non-torus spacetimes, Hurwitz theory is not an exact match for the large NN limit of YM2YM_{2} due to the presence of certain “Ω\Omega-factors” which appear in the partition function with multiplicity equal to the Euler characteristic of spacetime but have no obvious Hurwitz-theoretic meaning. The large NN limit of YM2YM_{2} is necessarily a completion of classical Hurwitz theory which accounts for these Ω\Omega-factors.

The purpose of this paper is to point out that the Ω\Omega-factors in YM2YM_{2} are exactly the eigenvalues of the exponential adjacency operator Ωq\Omega_{q} of the Hurwitz-Cayley graph, and that consequently the large NN limit of YM2YM_{2} is the union of classical and monotone Hurwitz theory: mixed Hurwitz theory. The mixed Hurwitz theory of a genus zero target was introduced in [27] as a way to interpolate between results in Hurwitz theory [50] and random matrix theory [60]. It was generalized to higher genus targets in [40].

1.4. Organization

In Section 2 we introduce the Hurwitz-Cayley graph and explain what is perhaps its most important feature: a combinatorial duality called the Jucys-Murphy correspondence. In Section 3 we derive the Gross-Taylor formula for the chiral partition function of YM2YM_{2} on compact orientable surfaces [14, 35], and explain its relationship to the graph operators Ψt\Psi_{t} and Ωq\Omega_{q}, which constitutes the Yang-Mills/Hurwitz correspondence.

The remaining sections of the paper illustrate the Yang-Mills/Hurwitz correspondence for specific choices of spacetime. This is done first for the finitely many cases which can be described using only classical Hurwitz theory: the cylinder and surfaces obtained from it by identifying boundaries or plugging holes (torus, disc, sphere). We then consider the three-holed sphere and one-holed torus as representative examples of the infinitely many remaining compact orientable spacetimes, where monotone Hurwitz theory is inescapable. In fact, the area zero limit of YM2YM_{2}, a much-studied degeneration [14, 15, 54], is pure monotone Hurwitz theory. This implies a relationship between monotone Hurwitz numbers and Euler characteristics of Hurwitz moduli spaces [22] which seems to be very interesting, but which we do not explore in detail here.

2. Hurwitz-Cayley Graph

2.1. Elementary features

Let Sd=Aut{1,,d}\mathrm{S}^{d}=\mathrm{Aut}\{1,\dots,d\} be the symmetric group of degree dd\in\mathbb{N}, where we define the group law as π1π2=π2π1\pi_{1}\pi_{2}=\pi_{2}\circ\pi_{1} so that permutations are multiplied left to right. Let K={(ij):1i<jd}K=\{(i\ j)\colon 1\leq i<j\leq d\} be the conjugacy class of transpositions in Sd\mathrm{S}^{d}. The Hurwitz-Cayley graph has vertex set Sd\mathrm{S}^{d}, with ρ,σ\rho,\sigma adjacent if and only if ρτ=σ\rho\tau=\sigma for τK\tau\in K. We henceforth identify Sd\mathrm{S}^{d} with the Hurwitz-Cayley graph.

Geodesic distance in Cayley graphs is implemented by word norm. For the Hurwitz-Cayley grpah, d(ρ,σ)=|ρ1σ|\mathrm{d}(\rho,\sigma)=|\rho^{-1}\sigma| where |π||\pi| is the minimal length of a factorization of π\pi into transpositions. The Join-Cut Lemma [29] gives |π|=d(π)|\pi|=d-\ell(\pi), where (π)\ell(\pi) is the number of factors in the unique decomposition of π\pi as a product of disjoint cycles. Accordingly, the Hurwitz-Cayley graph decomposes as

(7) Sd=r=0d1Lr,\mathrm{S}^{d}=\bigsqcup_{r=0}^{d-1}L_{r},

where LrL_{r} is the set of permutations consisting of drd-r disjoint cycles. Every edge of the Hurwitz-Cayley graph spans consecutive levels Lr,Lr+1L_{r},L_{r+1}.

The Cayley graph of any group with respect to any symmetric generating set decomposes into independent sets made up of identity-centered spheres, giving a graded graph structure. The Hurwitz-Cayley graph has the non-generic feature that its levels are unions of conjugacy classes: we have

(8) Lr=α𝖸d(α)=drKα,L_{r}=\bigsqcup_{\begin{subarray}{c}\alpha\in\mathsf{Y}^{d}\\ \ell(\alpha)=d-r\end{subarray}}K_{\alpha},

where 𝖸d\mathsf{Y}^{d} is the set of Young diagrams with exactly dd cells, (α)\ell(\alpha) denotes the number of rows of α𝖸d\alpha\in\mathsf{Y}^{d}, and KαK_{\alpha} is the conjugacy class of permutations in Sd\mathrm{S}^{d} of cycle type α\alpha.

2.2. Jucys-Murphy correspondence: combinatorial form

There are in general many geodesics between permutations ρ,σSd\rho,\sigma\in\mathrm{S}^{d}. The number of geodesics ρσ\rho\to\sigma equals the number of minimal transposition factorizations of ρ1σ\rho^{-1}\sigma, and this number was computed independently by Hurwitz and Cayley; see [44]. It is

(9) (drα11,,αr1)i=1rCayαi1,{d-r\choose\alpha_{1}-1,\dots,\alpha_{r}-1}\prod_{i=1}^{r}\mathrm{Cay}_{\alpha_{i}-1},

where α1,,αr\alpha_{1},\dots,\alpha_{r} are the row lengths of the Young diagram α𝖸d\alpha\in\mathsf{Y}^{d} encoding the cycle type of ρ1σ\rho^{-1}\sigma and Cayn=(n+1)n1\mathrm{Cay}_{n}=(n+1)^{n-1} is the Cayley number. The reasoning behind this formula is elementary. For each i=1,,ri=1,\dots,r, factor cylce ii of ρ1σ\rho^{-1}\sigma into αi1\alpha_{i}-1 transpositions, the minimal number required; this can be done in Cayαi1\operatorname{Cay}_{\alpha_{i}-1} ways. The result is a string of drd-r transpositions divided into rr blocks, the factored cycles of ρ1σ\rho^{-1}\sigma; by minimality the factors in distinct blocks are disjoint. The multinomial coefficient counts shuffles of these factors which maintain the relative order within each block, preserving the total product.

A fundamental feature of the Hurwitz-Cayley graph is that it can be Euclideanized by choosing a distinguished geodesic between every pair of points in a systematic way. This construction involves an edge labeling of the symmetric group introduced, implicitly and independently, by Jucys [41] and Murphy [47]. Let us mark each edge corresponding to the transposition (ij)(i\ j) with jj, the larger of the two numbers interchanged. Thus, emanating from every vertex of Sd\mathrm{S}^{d} we have one 22-edge, two 33-edges, three 44-edges, etc. See Figure 1 for the case d=4d=4.

Refer to caption
Figure 1. Jucys-Murphy labeling of S4\mathrm{S}^{4}.

A walk on Sd\mathrm{S}^{d} is said to be strictly monotone if the labels of the edges it traverses form a strictly increasing sequence. Clearly the length of any such walk is at most d1d-1, the number of edge labels and the diameter of Sd\mathrm{S}^{d}.

Theorem 2.1.

There exists a unique strictly monotone walk between every pair of permutations, and this walk is a geodesic.

As with the Hurwitz-Cayley geodesic count, the reasoning behind the Jucys-Murphy geodesic count is elementary. A strictly monotone walk ρσ\rho\to\sigma is a factorization of ρ1σ\rho^{-1}\sigma into transpositions of the form

(10) ρ1σ=(i1j1)(irjr),j1<<jr.\rho^{-1}\sigma=(i_{1}\ j_{1})\dots(i_{r}\ j_{r}),\quad j_{1}<\dots<j_{r}.

In order to establish Theorem 2.1, it is sufficient to establish that a cycle admits a unique strictly monotone factorization, and show that the length of this factorization is minimal — monotonicity kills the shuffle factor present in the unrestricted geodesic count. For an inductive proof of a statement equivalent to Theorem 2.1, see [18].

Theorem 2.1 says that for any permutation ρ\rho, the points of the sphere of radius rr centered at ρ\rho correspond bijectively to endpoints of strictly monotone rr-step walks emanating from ρ\rho. We refer to this bijection the Jucys-Murphy correspondence; see Figure 2.

Refer to caption
Figure 2. Jucys-Murphy correspondence in S4\mathrm{S}^{4}.

Instead of defining a strictly monotone walk by the condition that the labels of its edges are strictly increasing, we could have stipulated that they strictly decrease along the walk. Since path reversal defines a bijection between strictly decreasing walks ρσ\rho\to\sigma and strictly increasing walks σρ\sigma\to\rho, Theorem 2.1 implies that there is one and only one strictly decreasing walk between any two vertices of the Hurwitz-Cayley graph, and that it is a geodesic. Theorem 2.1 is thus valid in both senses of the word “monotone.”

Theorem 2.1 allows us to view an arbitrary product π1π2πk\pi_{1}\pi_{2}\dots\pi_{k} of permutations graphically, as a concatenation of strictly monotone walks

(11) ιπ1π1π2π1π2πk.\iota\to\pi_{1}\to\pi_{1}\pi_{2}\to\dots\to\pi_{1}\pi_{2}\dots\pi_{k}.

2.3. Jucys-Murphy correspondence: matricial form

Since Sd\mathrm{S}^{d} is a group, Sd\mathbb{C}\mathrm{S}^{d} is an algebra, the product of kets being |π1|π2=|π1π2|\pi_{1}\rangle|\pi_{2}\rangle=|\pi_{1}\pi_{2}\rangle. The group algebra Sd\mathbb{C}\mathrm{S}^{d} acts on itself by right multiplication: every |ASd|A\rangle\in\mathbb{C}\mathrm{S}^{d} gives a linear operator AEndSdA\in\operatorname{End}\mathbb{C}\mathrm{S}^{d} defined by

(12) A|ρ=πSdπ|A|ρπ.A|\rho\rangle=\sum_{\pi\in\mathrm{S}^{d}}\langle\pi|A\rangle|\rho\pi\rangle.

The matrix elements of AEndSdA\in\operatorname{End}\mathbb{C}\mathrm{S}^{d} in the permutation basis are

(13) σ|A|ρ=πSdπ|Aσ|ρπ=ρ1σ|A.\langle\sigma|A|\rho\rangle=\sum_{\pi\in\mathrm{S}^{d}}\langle\pi|A\rangle\langle\sigma|\rho\pi\rangle=\langle\rho^{-1}\sigma|A\rangle.

In particular, every diagonal matrix element is equal to a common value A=ι|A\langle A\rangle=\langle\iota|A\rangle, the coefficient of |ι|\iota\rangle in |A|A\rangle. The linear functional |AA|A\rangle\mapsto\langle A\rangle is the normalized character of the regular representation,

(14) A=1d!TrA.\langle A\rangle=\frac{1}{d!}\operatorname{Tr}A.

The map |AA|A\rangle\mapsto A is a faithful linear representation of Sd\mathbb{C}\mathrm{S}^{d}, the right regular representation. By abuse of notation, if |ASd|A\rangle\in\mathbb{C}\mathrm{S}^{d} has {0,1}\{0,1\}-coefficients in the permutation basis we write AA for both the subset of Sd\mathrm{S}^{d} determined by the nonzero coefficients of |A|A\rangle, and for the operator on Sd\mathbb{C}\mathrm{S}^{d} which is the image of |A|A\rangle in the regular representation. Thus KK denotes both the conjugacy class of transpositions in Sd\mathrm{S}^{d} and the adjacency operator on the Hurwitz-Cayley graph, and LrL_{r} denotes both the sphere of radius rr centered at ι\iota in Sd\mathrm{S}^{d} and the operator which maps |ρ|\rho\rangle to the sum of all permutations |σ|\sigma\rangle on the sphere of radius rr with center ρ\rho. In matrix form, the Jucys-Murphy correspondence is as follows.

Theorem 2.2.

For any ρ,σSd\rho,\sigma\in\mathrm{S}^{d}, we have

σ|Lr|ρ=W<r(ρ,σ)\langle\sigma|L_{r}|\rho\rangle=W^{r}_{<}(\rho,\sigma)

where W<r(ρ,σ)W^{r}_{<}(\rho,\sigma) is the number of strictly monotone rr-step walks ρσ\rho\to\sigma in the Hurwitz-Cayley graph.

Let |ΩqSd|\Omega_{q}\rangle\in\mathbb{C}\mathrm{S}^{d} be the exponential distance element,

(15) |Ωq=πSdq|π||π=r=0d1qr|Lr,|\Omega_{q}\rangle=\sum_{\pi\in\mathrm{S}^{d}}q^{|\pi|}|\pi\rangle=\sum_{r=0}^{d-1}q^{r}|L_{r}\rangle,

so that ΩqEndSd\Omega_{q}\in\operatorname{End}\mathbb{C}\mathrm{S}^{d} is the exponential distance operator of the Hurwitz-Cayley graph. Theorem 2.2 immediately implies the following alternative interpretation of the matrix elements of Ωq\Omega_{q} in the vertex basis.

Corollary 2.3.

For any ρ,σSd\rho,\sigma\in\mathrm{S}^{d}, we have

σ|Ωq|ρ=q=0qrW<r(ρ,σ).\langle\sigma|\Omega_{q}|\rho\rangle=\sum_{q=0}^{\infty}q^{r}W^{r}_{<}(\rho,\sigma).

2.4. Jucys-Murphy correspondence: polynomial form

Theorem 2.2 says that

(16) |Lr=2j1<<jrd(i1=1j1|i1j1)(ir=1jr|irjr).|L_{r}\rangle=\sum_{2\leq j_{1}<\dots<j_{r}\leq d}\left(\sum_{i_{1}=1}^{j_{1}}|i_{1}\ j_{1}\rangle\right)\dots\left(\sum_{i_{r}=1}^{j_{r}}|i_{r}\ j_{r}\rangle\right).

This is the statement that

(17) |Lr=er(|J1,,|Jr),|L_{r}\rangle=e_{r}(|J_{1}\rangle,\dots,|J_{r}\rangle),

the elementary symmetric polynomial ere_{r} of degree rr evaluated on the Jucys-Murphy elements

(18) |J1=|0|J2=|1 2|J3=|1 3+|2 3|Jd=|1d++|d1d\begin{split}|J_{1}\rangle&=|0\rangle\\ |J_{2}\rangle&=|1\ 2\rangle\\ |J_{3}\rangle&=|1\ 3\rangle+|2\ 3\rangle\\ \ &\vdots\\ |J_{d}\rangle&=|1\ d\rangle+\dots+|d-1\ d\rangle\end{split}

in the group algebra Sd\mathbb{C}\mathrm{S}^{d}. These elements, which commute with one another, are of fundamental importance in the representation theory of the symmetric groups. A very readable account of their properties has been given by Diaconis and Greene [18]. Their images J1,J2,,JrEndSdJ_{1},J_{2},\dots,J_{r}\in\operatorname{End}\mathbb{C}\mathrm{S}^{d} in the regular representation are called the Jucys-Murphy operators. As a polynomial identity in Sd\mathbb{C}\mathrm{S}^{d} or EndSd\operatorname{End}\mathbb{C}\mathrm{S}^{d}, the Jucys-Murphy correspondence takes the following form.

Theorem 2.4.

For any 0rd10\leq r\leq d-1, we have

|Lr=er(|J1,,|Jd)|L_{r}\rangle=e_{r}(|J_{1}\rangle,\dots,|J_{d}\rangle)

in Sd\mathbb{C}\mathrm{S}^{d}, or equivalently

Lr=er(J1,,Jd)L_{r}=e_{r}(J_{1},\dots,J_{d})

in EndSd\operatorname{End}\mathbb{C}\mathrm{S}^{d}.

For r=1r=1, Theorem 2.4 is the obvious fact that the adjacency operator of the Hurwitz-Cayley graph is the sum of the Jucys-Murphy operators,

(19) K=J1++Jd,K=J_{1}+\dots+J_{d},

which implies that the exponential adjacency operator factors as

(20) Ψt=etK=etJ1etJd.\Psi_{t}=e^{-tK}=e^{-tJ_{1}}\dots e^{-tJ_{d}}.

An analogous but subtler factorization of the exponential distance operator of the Hurwitz-Cayley graph is obtained by combining Theorem 2.4 with the generating series

(21) r=0qrer(x1,x2,x3,)=i=1(1+qxi)\sum_{r=0}^{\infty}q^{r}e_{r}(x_{1},x_{2},x_{3},\dots)=\prod_{i=1}^{\infty}(1+qx_{i})

for the elementary symmetric functions in an alphabet x1,x2,x2,x_{1},x_{2},x_{2},\dots of commuting indeterminates [30].

Corollary 2.5.

We have

|Ωq=(|ι+q|J1)(|ι+q|Jd)|\Omega_{q}\rangle=(|\iota\rangle+q|J_{1}\rangle)\dots(|\iota\rangle+q|J_{d}\rangle)

in Sd\mathbb{C}\mathrm{S}^{d}, or equivalently

Ωq=(I+qJ1)(I+qJd)\Omega_{q}=(I+qJ_{1})\dots(I+qJ_{d})

in EndSd\operatorname{End}\mathbb{C}\mathrm{S}^{d}.

We conclude that computing the eigenvalues of the exponential adjacency and distance operators Ψt\Psi_{t} and Ωq\Omega_{q} on the Hurwitz-Cayley graph reduces to computing the eigenvalues of the Jucys-Murphy operators J1,,JdEndSdJ_{1},\dots,J_{d}\in\operatorname{End}\mathbb{C}\mathrm{S}^{d}. Before addressing this spectral problem, we consider a further implication of the Jucys-Murphy correspondence.

2.5. Weakly monotone walks

Combining the polynomial form of the Jucys-Murphy correspondence with Newton’s theorem on symmetric polynomials, every symmetric polynomial function of the Jucys-Murphy elements |J1,,|JdSd|J_{1}\rangle,\dots,|J_{d}\rangle\in\mathbb{C}\mathrm{S}^{d} is a polynomial function of the levels |Lr=er(|J1,,|Jd)|L_{r}\rangle=e_{r}(|J_{1}\rangle,\dots,|J_{d}\rangle) of the Hurwitz-Cayley graph. As explained above, each level is a sum of conjugacy classes,

(22) |Lr=α𝖸d(α)=dr|Kα,0rd1.|L_{r}\rangle=\sum_{\begin{subarray}{c}\alpha\in\mathsf{Y}^{d}\\ \ell(\alpha)=d-r\end{subarray}}|K_{\alpha}\rangle,\quad 0\leq r\leq d-1.

Consequently, every symmetric polynomial function f(|J1,,|Jd)f(|J_{1}\rangle,\dots,|J_{d}\rangle) is a linear combination of conjugacy classes: we have

(23) f(|J1,,|Jd)=α𝖸dcα(f)|Kαf(|J_{1}\rangle,\dots,|J_{d}\rangle)=\sum_{\alpha\in\mathsf{Y}^{d}}c_{\alpha}(f)|K_{\alpha}\rangle

for some coefficients cα(f)c_{\alpha}(f), which are integers if ff is an integral polynomial in the ere_{r}’s.

A particularly interesting case is that of complete symmetric polynomials in Jucys-Murphy elements,

(24) |Mr=hr(|J1,,|Jd)=2j1jrd|Jj1|Jjr.|M_{r}\rangle=h_{r}(|J_{1}\rangle,\dots,|J_{d}\rangle)=\sum_{2\leq j_{1}\leq\dots\leq j_{r}\leq d}|J_{j_{1}}\rangle\dots|J_{j_{r}}\rangle.

We say that a walk on the Hurwitz-Cayley graph is weakly monotone if the labels of the edges it traverses form a weakly increasing sequence, and write Wr(ρ,σ)W^{r}_{\leq}(\rho,\sigma) for the number of weakly monotone rr-step walks ρσ\rho\to\sigma between given permutations ρ,σSd\rho,\sigma\in\mathrm{S}^{d}. By definition of the Jucys-Murphy elements and the complete symmetric polynomials, we have that

(25) π|Mr=Wr(ι,π)\langle\pi|M_{r}\rangle=W^{r}_{\leq}(\iota,\pi)

is the number of weakly monotone rr-step walks ιπ\iota\to\pi. Equivalently, for MrEndSdM_{r}\in\operatorname{End}\mathbb{C}\mathrm{S}^{d} the image of |Mr|M_{r}\rangle in the regular representation, we have the matrix formula

(26) σ|Mr|ρ=Wr(ρ,σ),\langle\sigma|M_{r}|\rho\rangle=W^{r}_{\leq}(\rho,\sigma),

the number of rr-step weakly monotone walks ρσ\rho\to\sigma in the Hurwitz-Cayley graph.

The fact that |Mr|M_{r}\rangle is a central element in Sd\mathbb{C}\mathrm{S}^{d} implies that Wr(ι,π)W^{r}_{\leq}(\iota,\pi) depends only on the conjugacy class of πSd\pi\in\mathrm{S}^{d}, which is equivalent to the statement that Wr(ρ,σ)W^{r}_{\leq}(\rho,\sigma) depends only on the cycle type of ρ1σ\rho^{-1}\sigma. In particular, we have

(27) Wr(σ,ρ)=Wr(ι,σ1ρ)=Wr(ι,ρ1σ)=Wr(ρ,σ),W^{r}_{\leq}(\sigma,\rho)=W^{r}_{\leq}(\iota,\sigma^{-1}\rho)=W^{r}_{\leq}(\iota,\rho^{-1}\sigma)=W^{r}_{\leq}(\rho,\sigma),

which by path reversal implies that

(28) Wr(ρ,σ)=Wr(ρ,σ),W^{r}_{\leq}(\rho,\sigma)=W^{r}_{\geq}(\rho,\sigma),

where Wr(ρ,σ)W^{r}_{\geq}(\rho,\sigma) is the number of rr-step walks ρσ\rho\to\sigma on the Hurwitz-Cayley graph whose step labels form a weakly decreasing sequence. Thus at the enumerative level there is no difference between decreasing and increasing walks in the Hurwitz-Cayley graph, whether strictly or weakly.

We have seen above that the total number of geodesics ρσ\rho\to\sigma in the Hurwitz-Cayley graph is given by a product of Cayley numbers along the cycles of ρ1σ\rho^{-1}\sigma times a shuffle factor. We have also seen that the number of strictly monotone geodesics ρσ\rho\to\sigma is exactly one. The number of weakly monotone geodesics ρσ\rho\to\sigma, i.e. the number of minimal weakly monotone factorizations of ρ1σ\rho^{-1}\sigma into transpositions, must lie between these two extremes.

Theorem 2.6 ([48]).

The number of weakly monotone geodesics ρσ\rho\to\sigma in the Hurwitz-Cayley graph Sd\mathrm{S}^{d} is

i=1(α)Catαi1,\prod_{i=1}^{\ell(\alpha)}\operatorname{Cat}_{\alpha_{i}-1},

where α𝖸d\alpha\in\mathsf{Y}^{d} is the cycle type of ρ1σ\rho^{-1}\sigma and Catn=1n+1(2nn)\operatorname{Cat}_{n}=\frac{1}{n+1}{2n\choose n} is the Catalan number.

The proof of Theorem 2.6 is again elementary, reducing to counting minimal monotone factorizations of a cycle; that this is a Catalan number was first discovered by Gewurz and Merola [23], and the lack of shuffle factor is again due to monotonicity. A more general result counting weakly monotone factorizations with a given “signature,” corresponding to evaluation of monomial symmetric polynomials in Jucys-Murphy elements, was proved in [46]. This result can in turn be used to develop an enumerative theory of weighted walks in the Hurwitz-Cayley graph [39], which turns out to be very rich [1].

2.6. Fourier transform

As with the group algebra of any finite group [52], we have an orthogonal decomposition

(29) Sd=λ𝖸d(dim𝐕λ)𝐕λ\mathbb{C}\mathrm{S}^{d}=\bigoplus_{\lambda\in\mathsf{Y}^{d}}(\dim\mathbf{V}^{\lambda})\mathbf{V}^{\lambda}

where 𝐕λ\mathbf{V}^{\lambda} is an enumeration of the irreducible representations of Sd\mathbb{C}\mathrm{S}^{d}. By Schur’s Lemma, if |ASd|A\rangle\in\mathbb{C}\mathrm{S}^{d} is a linear combination of conjugacy classes then AEndSdA\in\operatorname{End}\mathbb{C}\mathrm{S}^{d} acts in 𝐕λ\mathbf{V}^{\lambda} as a scalar operator whose eigenvalue we denote A^(λ)\hat{A}(\lambda). This associates to every central element |ASd|A\rangle\in\mathbb{C}\mathrm{S}^{d} a function A^:𝖸d\hat{A}\colon\mathsf{Y}^{d}\to\mathbb{C}, the Fourier transform of |A|A\rangle. The map |AA^|A\rangle\to\hat{A} is an isometric algebra isomorphism from the center of Sd\mathbb{C}\mathrm{S}^{d} to the function algebra L2(𝖸d)L^{2}(\mathsf{Y}^{d}) — we have

(30) Kα|Kβ=λ𝖸d(dim𝐕λ)2d!K^α(λ)K^β(λ)=δαβ|Kα|.\langle K_{\alpha}|K_{\beta}\rangle=\sum_{\lambda\in\mathsf{Y}^{d}}\frac{(\dim\mathbf{V}^{\lambda})^{2}}{d!}\hat{K}_{\alpha}(\lambda)\hat{K}_{\beta}(\lambda)=\delta_{\alpha\beta}|K_{\alpha}|.

This gives the Plancherel expectation formula

(31) A=1d!TrA=λ𝖸d(dim𝐕λ)2d!A^(λ).\langle A\rangle=\frac{1}{d!}\operatorname{Tr}A=\sum_{\lambda\in\mathsf{Y}^{d}}\frac{(\dim\mathbf{V}^{\lambda})^{2}}{d!}\hat{A}(\lambda).

for central elements |ASd|A\rangle\in\mathbb{C}\mathrm{S}^{d}.

A remarkable formula for computing the Fourier transform of a symmetric polynomial function in Jucys-Murphy elements was discovered by Jucys and Murphy; see [18]. Recall that the content c()c(\Box) of a cell \Box in a Young diagram λ\lambda is defined to be its column index minus its row index. Thus, filling the cells of λ\lambda with their contents produces a jagged λ\lambda-shaped corner of the infinite Toeplitz matrix [ji]i,j=1[j-i]_{i,j=1}^{\infty}.

Theorem 2.7.

If |ASd|A\rangle\in\mathbb{C}\mathrm{S}^{d} is such that |A=f(|J1,,|Jd)|A\rangle=f(|J_{1}\rangle,\dots,|J_{d}\rangle) for a symmetric polynomial ff, then the Fourier transform of |A|A\rangle is

A^(λ)=f(c():λ),\hat{A}(\lambda)=f(c(\Box)\colon\Box\in\lambda),

the evaluation of ff on the multiset of conents of λ\lambda. Equivalently, for each λ𝖸d\lambda\in\mathsf{Y}^{d} the eigenvalue of A=f(J1,,Jd)EndSdA=f(J_{1},\dots,J_{d})\in\operatorname{End}\mathbb{C}\mathrm{S}^{d} acting in 𝐕λ\mathbf{V}^{\lambda} is f(c():λ)f(c(\Box)\colon\Box\in\lambda).

According to Theorem 2.7, the transposition class |K=|J1++|Jd|K\rangle=|J_{1}\rangle+\dots+|J_{d}\rangle has Fourier transform

(32) K^(λ)=λc().\hat{K}(\lambda)=\sum_{\Box\in\lambda}c(\Box).

The spectrum of the adjacency operator of the Hurwitz-Cayley graph consists of these numbers as λ\lambda ranges over 𝖸d\mathsf{Y}^{d}, with the multiplicity of K^(λ)\hat{K}(\lambda) being dim𝐕λ\dim\mathbf{V}^{\lambda}. The eigenvalues of the exponential adjacency operator are thus

(33) Ψ^t(λ)=λetc(),λ𝖸d.\hat{\Psi}_{t}(\lambda)=\prod_{\Box\in\lambda}e^{-tc(\Box)},\quad\lambda\in\mathsf{Y}^{d}.

To find the eigenvalues of the exponential distance operator Ωq\Omega_{q} of the Hurwitz-Cayley graph, we need to compute the Fourier of every level of the graph. Theorem 2.7 gives the Fourier transform of |Lr=er(|J1,,|Jd)|L_{r}\rangle=e_{r}(|J_{1}\rangle,\dots,|J_{d}\rangle) as

(34) L^r(λ)=er(c():λ).\hat{L}_{r}(\lambda)=e_{r}(c(\Box)\colon\Box\in\lambda).

The eigenvalues of the exponential adjacency operator

(35) Ωq=r=0d1qrLr\Omega_{q}=\sum_{r=0}^{d-1}q^{r}L_{r}

are thus

(36) Ω^q(λ)=λ(1+qc()),λ𝖸d,\hat{\Omega}_{q}(\lambda)=\prod_{\Box\in\lambda}(1+qc(\Box)),\quad\lambda\in\mathsf{Y}^{d},

and

(37) detΩq=λ𝖸dΩ^q(λ)dim𝐕λ=λ𝖸dλ(1+qc())dim𝐕λ.\det\Omega_{q}=\prod_{\lambda\in\mathsf{Y}^{d}}\hat{\Omega}_{q}(\lambda)^{\dim\mathbf{V}^{\lambda}}=\prod_{\lambda\in\mathsf{Y}^{d}}\prod_{\Box\in\lambda}(1+qc(\Box))^{\dim\mathbf{V}^{\lambda}}.

One can give an analogous formula for any coefficient of the characteristic polynomial of Ωq\Omega_{q}, but we will not do so here. The determinant formula is enough to see the following.

Theorem 2.8.

The exponential distance operator Ωq\Omega_{q} of the Hurwitz-Cayley graph Sd\mathrm{S}^{d} is singular if and only if d>1d>1 and qq or q-q is one of the unit fractions

1d1,1d2,,11.\frac{1}{d-1},\frac{1}{d-2},\dots,\frac{1}{1}.

2.7. The operator Ωq1\Omega_{q}^{-1}

For qq away from the above unit fractions, the exponential distance operator Ωq\Omega_{q} of the Hurwitz-Cayley graph is invertible, and its inverse acts in 𝐕λ\mathbf{V}^{\lambda} as multiplication by

(38) Ω^q1(λ)=λ11+qc().\hat{\Omega}_{q}^{-1}(\lambda)=\prod_{\Box\in\lambda}\frac{1}{1+qc(\Box)}.

For |q|<1d1|q|<\frac{1}{d-1}, this is the absolutely convergent series

(39) Ω^q1(λ)=r=0(q)rhr(c():λ)=r=0(q)rM^r(λ),\hat{\Omega}_{q}^{-1}(\lambda)=\sum_{r=0}^{\infty}(-q)^{r}h_{r}(c(\Box)\colon\Box\in\lambda)=\sum_{r=0}^{\infty}(-q)^{r}\hat{M}_{r}(\lambda),

where |Mr=hr(|J1,,|Jd)|M_{r}\rangle=h_{r}(|J_{1}\rangle,\dots,|J_{d}\rangle). Thus, for |q|<1d1|q|<\frac{1}{d-1}, we have

(40) Ωq1=r=0(q)rMr\Omega_{q}^{-1}=\sum_{r=0}^{\infty}(-q)^{r}M_{r}

in EndSd\operatorname{End}\mathbb{C}\mathrm{S}^{d}, where Mr=hr(J1,,Jd)M_{r}=h_{r}(J_{1},\dots,J_{d}) is the complete symmetric polynomial of degree rr in the Jucys-Murphy operators. The matrix elements of the inverse exponential distance operator are therefore

(41) σ|Ωq1|ρ=r=0(q)rσ|Mr|ρ,\langle\sigma|\Omega_{q}^{-1}|\rho\rangle=\sum_{r=0}^{\infty}(-q)^{r}\langle\sigma|M_{r}|\rho\rangle,

and we get a combinatorial interpretation for the matrix elements of Ωq1\Omega_{q}^{-1} as generating functions for weakly monotone walks in the Hurwitz-Cayley graph.

Theorem 2.9.

For |q|<1d1|q|<\frac{1}{d-1}, the matrix elements of the inverse exponential distance operator on Ωq\Omega_{q} are absolutely convergent generating functions

σ|Ωq1|ρ=r=0(q)rWr(ρ,σ)\langle\sigma|\Omega_{q}^{-1}|\rho\rangle=\sum_{r=0}^{\infty}(-q)^{r}W^{r}_{\leq}(\rho,\sigma)

for weakly monotone walks in the Hurwitz-Cayley graph Sd\mathrm{S}^{d},

2.8. Summary

The exponential adjacency and distance matrices Ψt\Psi_{t} and Ωq\Omega_{q} of the Hurwitz-Cayley graph are generating functions for walks in Sd\mathrm{S}^{d}. The matrix elements of the exponential adjacency matrix are

(42) σ|Ψt|ρ=r=0(t)rr!Wr(ρ,σ),\langle\sigma|\Psi_{t}|\rho\rangle=\sum_{r=0}^{\infty}\frac{(-t)^{r}}{r!}W^{r}(\rho,\sigma),

where Wr(ρ,σ)W^{r}(\rho,\sigma) is the total number of walks ρσ\rho\to\sigma. This is a general fact about graphs. The Hurwitz-Cayley graph has the special feature that

(43) σ|Ωq|ρ=r=0qrW<r(ρ,σ)andσ|Ωq1|ρ=r=0(q)rWr(ρ,σ)\langle\sigma|\Omega_{q}|\rho\rangle=\sum_{r=0}^{\infty}q^{r}W^{r}_{<}(\rho,\sigma)\quad\text{and}\quad\langle\sigma|\Omega_{q}^{-1}|\rho\rangle=\sum_{r=0}^{\infty}(-q)^{r}W^{r}_{\leq}(\rho,\sigma)

are generating functions enumerating strictly and weakly monotone walks ρσ\rho\to\sigma in Sd\mathrm{S}^{d}. The first series consists of a single nonzero term, namely qd(ρ,σ)q^{\mathrm{d}(\rho,\sigma)}. The second series has infinitely many nonzero terms and requires |q|<1d1|q|<\frac{1}{d-1} to ensure absolute convergence.

3. Yang-Mills Theory

3.1. Partition functions

The partition function of Yang-Mills theory on a compact two-dimensional orientable spacetime of area tt with mm holes and nn handles admits a dual representation in terms of the gauge group, UN\mathrm{U}_{N}. The elegant formula reads [14]

(44) 𝒵N=λ1λNSλ(U1,,Um)(dim𝐖Nλ)m+2n2et2NC^(λ).\mathcal{Z}_{N}=\sum_{\lambda_{1}\geq\dots\geq\lambda_{N}}\frac{S_{\lambda}(U_{1},\dots,U_{m})}{(\dim\mathbf{W}_{N}^{\lambda})^{m+2n-2}}e^{-\frac{t}{2N}\hat{C}(\lambda)}.

The sum is over integer vectors λ=(λ1,,λN)\lambda=(\lambda_{1},\dots,\lambda_{N}) with weakly decreasing coordinates, each labeling an irreducible representation 𝐖Nλ\mathbf{W}_{N}^{\lambda} of UN\mathrm{U}_{N}. The function Sλ:UNmS_{\lambda}\colon\mathrm{U}_{N}^{m}\to\mathbb{C} is the character of (𝐖Nλ)m(\mathbf{W}_{N}^{\lambda})^{\otimes m}. The unitary matrices U1,,UmU_{1},\dots,U_{m} represent boundary holonomies: setting any one of them to the identity cancels a factor of dim𝐖Nλ\dim\mathbf{W}_{N}^{\lambda} and plugs a hole. The exponential factor is a discrete Gaussian weight: C^(λ)\hat{C}(\lambda) is the 𝐖Nλ\mathbf{W}_{N}^{\lambda}-eigenvalue of the quadratic Casimir CC, a central element in the universal enveloping algebra of 𝔲N\mathfrak{u}_{N} corresponding to the Laplacian on UN\mathrm{U}_{N}. If spacetime is a cylinder, then 𝒵N\mathcal{Z}_{N} is the UN\mathrm{U}_{N} heat kernel [42, 59].

It was perceived by Gross that (44) provides a promising path to a string description of YM2YM_{2} at large NN, which ought to be Hurwitz theory [33]. This insight was fully developed by Gross and Taylor [35, 36], who argued that as NN\to\infty the partition function 𝒵N\mathcal{Z}_{N} factorizes into two copies of a chiral partition function, ZNZ_{N}, obtained by restricting the sum (44) to nonnegative λ\lambda. The chiral partition function may be viewed as a sum over the set 𝖸N\mathsf{Y}_{N} of Young diagrams with at most NN rows, and thus decomposes as

(45) ZN=1+d=1ZNd,Z_{N}=1+\sum_{d=1}^{\infty}Z_{N}^{d},

where the microchiral partition function

(46) ZNd=λ𝖸NdSλ(U1,,Um)(dim𝐖Nλ)m+2n2et2NC^(λ)Z_{N}^{d}=\sum_{\lambda\in\mathsf{Y}_{N}^{d}}\frac{S_{\lambda}(U_{1},\dots,U_{m})}{(\dim\mathbf{W}_{N}^{\lambda})^{m+2n-2}}e^{-\frac{t}{2N}\hat{C}(\lambda)}

is a sum over the finite set 𝖸Nd\mathsf{Y}_{N}^{d} of Young diagrams with at most NN rows and exactly dd cells. The microchiral partition function (46) is the core of YM2YM_{2}. What we seek is a 1/N1/N expansion of ZNdZ_{N}^{d} whose coefficients count maps of worldsheets into spacetime. Once this string signature has been found, we sum on dd to recover the chiral partition function ZNZ_{N}, and then square to recover the full partition function 𝒵N\mathcal{Z}_{N}.

3.2. Gross-Taylor formula

Gross and Taylor’s fundamental insight [35, 36] is that ZNdZ_{N}^{d} can be presented entirely in terms of the representation theory of the symmetric group Sd\mathrm{S}^{d}. This is accomplished in three steps.

3.2.1. Laplacian swap

The first observation of Gross and Taylor [35] is that

(47) C^(λ)=dN+2K^(λ),λ𝖸Nd,\hat{C}(\lambda)=dN+2\hat{K}(\lambda),\quad\lambda\in\mathsf{Y}_{N}^{d},

where K^(λ)\hat{K}(\lambda) is the Fourier transform of the transposition class |KSd|K\rangle\in\mathbb{C}\mathrm{S}^{d}. Thus (47) effectively trades the Laplacian on UN\mathrm{U}_{N} for the Laplacian on Sd\mathrm{S}^{d}, yielding

(48) ZNd=zdλ𝖸NdSλ(U1,,Um)(dim𝐖Nλ)m+2n2Ψ^tN(λ),Z_{N}^{d}=z^{d}\sum_{\lambda\in\mathsf{Y}_{N}^{d}}\frac{S_{\lambda}(U_{1},\dots,U_{m})}{(\dim\mathbf{W}_{N}^{\lambda})^{m+2n-2}}\hat{\Psi}_{\frac{t}{N}}(\lambda),

where z=et/2z=e^{-t/2} and Ψ^t(λ)=etK^(λ)\hat{\Psi}_{t}(\lambda)=e^{-t\hat{K}(\lambda)}.

3.2.2. Character swap

According to Frobenius [58], the character sλs_{\lambda} of 𝐖Nλ\mathbf{W}_{N}^{\lambda} is given by

(49) sλ(U)=dim𝐕λd!α𝖸dpα(U)K^α(λ),s_{\lambda}(U)=\frac{\dim\mathbf{V}^{\lambda}}{d!}\sum_{\alpha\in\mathsf{Y}^{d}}p_{\alpha}(U)\hat{K}_{\alpha}(\lambda),

where K^α(λ)\hat{K}_{\alpha}(\lambda) is the Fourier transform of the conjugacy class |KαSd|K_{\alpha}\rangle\in\mathbb{C}\mathrm{S}^{d} and

(50) pα(U)=i=1(α)TrUαip_{\alpha}(U)=\prod_{i=1}^{\ell(\alpha)}\operatorname{Tr}U^{\alpha_{i}}

is the corresponding trace invariant. We can thus replace Sλ=sλsλS_{\lambda}=s_{\lambda}\otimes\dots\otimes s_{\lambda} with

(51) Sλ=(dim𝐕λd!)mα1,,αm𝖸dPα1αmK^α1αm(λ),S_{\lambda}=\left(\frac{\dim\mathbf{V}^{\lambda}}{d!}\right)^{m}\sum_{\alpha^{1},\dots,\alpha^{m}\in\mathsf{Y}^{d}}P_{\alpha^{1}\dots\alpha^{m}}\hat{K}_{\alpha^{1}\dots\alpha^{m}}(\lambda),

where

(52) Pα1αm(U1,,Um)=pα1(U1)pαm(Um)P_{\alpha^{1}\dots\alpha^{m}}(U_{1},\dots,U_{m})=p_{\alpha^{1}}(U_{1})\dots p_{\alpha^{m}}(U_{m})

is a product of trace invariants on UN\mathrm{U}_{N} and

(53) K^α1αm(λ)=K^α1(λ)K^αm(λ)\hat{K}_{\alpha^{1}\dots\alpha^{m}}(\lambda)=\hat{K}_{\alpha^{1}}(\lambda)\dots\hat{K}_{\alpha^{m}}(\lambda)

is a product of central characters of Sd\mathrm{S}^{d}. This gives the microchiral partition function as

(54) ZNd=zdα1,,αm𝖸dPα1αmλ𝖸Nd(dim𝐕λd!)mK^α1αm(λ)(dim𝐖Nλ)m+2n2Ψ^tN(λ),Z_{N}^{d}=z^{d}\sum_{\alpha^{1},\dots,\alpha^{m}\in\mathsf{Y}^{d}}P_{\alpha^{1}\dots\alpha^{m}}\sum_{\lambda\in\mathsf{Y}_{N}^{d}}\left(\frac{\dim\mathbf{V}^{\lambda}}{d!}\right)^{m}\frac{\hat{K}_{\alpha^{1}\dots\alpha^{m}}(\lambda)}{(\dim\mathbf{W}_{N}^{\lambda})^{m+2n-2}}\hat{\Psi}_{\frac{t}{N}}(\lambda),

where we now view ZNdZ_{N}^{d} as a function on UNm\mathrm{U}_{N}^{m}.

3.2.3. Dimension swap

Finally, we eliminate dim𝐖Nλ\dim\mathbf{W}_{N}^{\lambda} using the proportionality

(55) dim𝐕λdim𝐖Nλ=d!(N)λ,λ𝖸Nd,\frac{\dim\mathbf{V}^{\lambda}}{\dim\mathbf{W}_{N}^{\lambda}}=\frac{d!}{(N)_{\lambda}},\quad\lambda\in\mathsf{Y}_{N}^{d},

where

(56) (x)λ=λ(x+c())(x)_{\lambda}=\prod_{\Box\in\lambda}(x+c(\Box))

is the generalized Pochammer symbol, which after renormalization is an eigenvalue

(57) Ω^q(λ)=λ(1+qc())\hat{\Omega}_{q}(\lambda)=\prod_{\Box\in\lambda}(1+qc(\Box))

of the exponential distance operator Ωq\Omega_{q} of the Hurwitz-Cayley graph Sd\mathrm{S}^{d}. We thus have

(58) dim𝐕λdim𝐖Nλ=d!NdΩ^1N1(λ),\frac{\dim\mathbf{V}^{\lambda}}{\dim\mathbf{W}_{N}^{\lambda}}=\frac{d!}{N^{d}}\hat{\Omega}_{\frac{1}{N}}^{-1}(\lambda),

where Ω^1N1(λ)\hat{\Omega}^{-1}_{\frac{1}{N}}(\lambda) is well-defined for any Young diagram with at most NN rows. The microchiral partition function thus assumes its final form.

Theorem 3.1 (Gross-Taylor formula).

The microchiral partition function of YM2YM_{2} on a compact orientable surface of area tt with mm holes and nn handles is

ZNd=(zN22nm)dα1,,αm𝖸dPα1αmλ𝖸Nd(dim𝐕λd!)22nK^α1αm(λ)Ψ^tN(λ)Ω^1N22nm(λ).Z_{N}^{d}=\left(zN^{2-2n-m}\right)^{d}\sum_{\alpha^{1},\dots,\alpha^{m}\in\mathsf{Y}^{d}}P_{\alpha^{1}\dots\alpha^{m}}\sum_{\lambda\in\mathsf{Y}_{N}^{d}}\left(\frac{\dim\mathbf{V}^{\lambda}}{d!}\right)^{2-2n}\hat{K}_{\alpha^{1}\dots\alpha^{m}}(\lambda)\hat{\Psi}_{\frac{t}{N}}(\lambda)\hat{\Omega}_{\frac{1}{N}}^{2-2n-m}(\lambda).

3.3. Great expectations

The Gross-Taylor formula reveals that the existence of a 1/N1/N expansion of ZNdZ_{N}^{d}, and the Hurwitz-theoretic interpretation of its coefficients, is a somewhat subtle issue. The internal sum in Theorem 3.1 can be written

(59) 1d!λ𝖸Nd(dim𝐕λ)2d!H^n(λ)K^α1αm(λ)Ψ^tN(λ)Ω^1N22nm(λ),\frac{1}{d!}\sum_{\lambda\in\mathsf{Y}_{N}^{d}}\frac{(\dim\mathbf{V}^{\lambda})^{2}}{d!}\hat{H}^{n}(\lambda)\hat{K}_{\alpha^{1}\dots\alpha^{m}}(\lambda)\hat{\Psi}_{\frac{t}{N}}(\lambda)\hat{\Omega}_{\frac{1}{N}}^{2-2n-m}(\lambda),

where |HSd|H\rangle\in\mathbb{C}\mathrm{S}^{d} is the central element whose Fourier transform is H^(λ)=(d!dim𝐕λ)2\hat{H}(\lambda)=\left(\frac{d!}{\dim\mathbf{V}^{\lambda}}\right)^{2}. This element is the commutator sum [58]

(60) |H=ρ,σSdρ1σ1ρσ.|H\rangle=\sum_{\rho,\sigma\in\mathrm{S}^{d}}\rho^{-1}\sigma^{-1}\rho\sigma.

In the stable range, where 1dN1\leq d\leq N, we have 𝖸Nd=𝖸d\mathsf{Y}_{N}^{d}=\mathsf{Y}^{d} and the sum (59) runs over all irreducible representations of Sd\mathbb{C}\mathrm{S}^{d} and is a Plancherel expectation.

Definition 3.2.

For any integers 1dN1\leq d\leq N and m,n0m,n\in\mathbb{N}_{0}, the corresponding Gross-Taylor expectation is the function on (𝖸d)m(\mathsf{Y}^{d})^{m} defined by

Nd(α1,,αm;n)=HnKα1αmΨtNΩ1N22nm.\mathcal{E}_{N}^{d}(\alpha_{1},\dots,\alpha_{m};n)=\langle H^{n}K_{\alpha^{1}\dots\alpha^{m}}\Psi_{\frac{t}{N}}\Omega_{\frac{1}{N}}^{2-2n-m}\rangle.

The Gross-Taylor expectation Nd(α1,,αm;n)\mathcal{E}_{N}^{d}(\alpha_{1},\dots,\alpha_{m};n) admits an absolutely convergent series expansion in powers of 1/N1/N determined by the 1/N1/N expansions of the exponential adjacency operator

(61) ΨtN=r=0(t)rr!NrKr\Psi_{\frac{t}{N}}=\sum_{r=0}^{\infty}\frac{(-t)^{r}}{r!N^{r}}K^{r}

and exponential distance operator and its inverse,

(62) Ω1N=r=0d11NrLrandΩ1N1=r=0(1)rNrMr,\Omega_{\frac{1}{N}}=\sum_{r=0}^{d-1}\frac{1}{N^{r}}L_{r}\quad\text{and}\quad\Omega_{\frac{1}{N}}^{-1}=\sum_{r=0}^{\infty}\frac{(-1)^{r}}{N^{r}}M_{r},

of the Hurwitz-Cayley graph Sd\mathrm{S}^{d}. The coefficients

(63) Lr=er(J1,,Jd)andMr=hr(J1,,Jd)L_{r}=e_{r}(J_{1},\dots,J_{d})\quad\text{and}\quad M_{r}=h_{r}(J_{1},\dots,J_{d})

are elementary and complete symmetric polynomial functions of the Jucys-Murphy operators, as discussed in Section 2. The coefficients of the 1/N1/N expansion of any Gross-Taylor expectation are expectations of words in the commuting operators H,Kα,Lr,MsH,K_{\alpha},L_{r},M_{s}.

In the unstable range, where d>Nd>N, two problems arise. The first is that 𝖸Nd\mathsf{Y}_{N}^{d} is a proper subset of 𝖸d\mathsf{Y}^{d} and the internal sum in the Gross-Taylor formula is not a Plancherel expectation. The second is that, if the number of holes in spacetime exceeds its Euler characteristic, the exponential distance operator Ω1N\Omega_{\frac{1}{N}} appears to a negative power and the resulting 1/N1/N expansion is divergent.

The standard approach to these issues in YM2YM_{2} is to ignore them and say that we take NN\to\infty. This is perfectly fine at the microchiral level, where dd\in\mathbb{N} is fixed and we only need NdN\geq d. However, if we wish to obtain a 1/N1/N expansion for the microchiral partition function ZN=1+dZNdZ_{N}=1+\sum_{d}Z_{N}^{d} by summing the 1/N1/N expansion of ZNdZ_{N}^{d}, an interchange of limits is required. What is actually done in the literature is to deceptively replace the numerical quantity with an indeterminate deceptively named 1/N1/N. This leads to a formal power series, the chiral Gross-Taylor series [14, 15], and understanding when this operation produces a quantitatively correct NN\to\infty asymptotic expansion of ZNZ_{N} is an apparent gap in the YM2YM_{2} literature.

3.4. Lattice Yang-Mills

There is an interesting analogy with Yang-Mills on the lattice, where the role of the chiral partition function is roughly played by the partition function of the Bars-Green/Brézin-Gross-Witten/Wadia unitary matrix model [3, 5, 37, 55],

(64) WN=UNezNTr(AU+BU)dUW_{N}=\int_{\mathrm{U}_{N}}e^{\sqrt{z}N\operatorname{Tr}(AU+BU^{*})}\mathrm{d}U

which plays a basic role in lattice gauge theory of any dimension. By character expansion [51],

(65) WN=1+d=1WNd,W_{N}=1+\sum_{d=1}^{\infty}W_{N}^{d},

where

(66) WNd=(zN)dd!α𝖸dpα(AB)λ𝖸Nd(dim𝐕λ)2d!Kα(λ)Ω1N1(λ)W_{N}^{d}=\frac{(zN)^{d}}{d!}\sum_{\alpha\in\mathsf{Y}^{d}}p_{\alpha}(AB)\sum_{\lambda\in\mathsf{Y}_{N}^{d}}\frac{(\dim\mathbf{V}^{\lambda})^{2}}{d!}K_{\alpha}(\lambda)\Omega_{\frac{1}{N}}^{-1}(\lambda)

is the lattice counterpart of the microchiral partition function ZNdZ_{N}^{d} in YM2YM_{2}. The internal sum over λ\lambda in (66) has come to be known as the Weingarten function; see [12].

In the one-plaquette model, Ω\Omega is the sole source of 1/N1/N — everything is determined by the spectral theory of the exponential distance operator on the Hurwitz-Cayley graph. The divergence of the 1/N1/N expansion, which is a consequence of the instability of the zeros of detΩq\det\Omega_{q}, was understood early on to be a fact of life on the lattice [17], and diagrammatic expansions are very complicated [6, 16, 49, 56] if one works directly with WNdW_{N}^{d} as a sum of link integrals,

(67) WNd=Aj1i1AjdidBi1j1BidjdUNUi1j1UidjdUi1j1Uidjd¯dU.W_{N}^{d}=\sum A_{j_{1}i_{1}}\dots A_{j_{d}i_{d}}B_{i^{\prime}_{1}j^{\prime}_{1}}\dots B_{i^{\prime}_{d}j^{\prime}_{d}}\int_{\mathrm{U}_{N}}U_{i_{1}j_{1}}\dots U_{i_{d}j_{d}}\overline{U_{i^{\prime}_{1}j^{\prime}_{1}}\dots U_{i^{\prime}_{d}j^{\prime}_{d}}}\ \mathrm{d}U.

The use of monotone walks in the Hurwitz-Cayley graph as “dual” Feynman diagrams for unitary matrix integrals was developed, in the language of permutation factorizations, in [46, 48]. It can be quite useful in physical applications, see e.g. [4].

3.5. Summary

We have derived the Gross-Taylor formula, Theorem 3.1, which presents the microchiral partition function ZNdZ_{N}^{d} of YM2YM_{2} on any compact orientable two-dimensional space time in terms of the representation theory of the symmetric group Sd\mathrm{S}^{d}. A key feature of the formula is the appearance of the eigenvalues of the exponential adjacency and distance operators Ψt\Psi_{t} and Ωq\Omega_{q} of the Hurwitz-Cayley graph, with tt/Nt\rightsquigarrow t/N and q1/Nq\rightsquigarrow 1/N. Iin the stable range, 1dN1\leq d\leq N, this combines with the power series expansions of these operators in tt and qq to yield an absolutely convergent expansion ZNdZ_{N}^{d} in powers of 1/N1/N, whose coefficients are joint Plancherel moments of four basic commuting operators H,Kα,Lr,MsH,K_{\alpha},L_{r},M_{s}. These basic expectations admit combinatorial interpretations as counting trajectories in the Hurwitz-Cayley graph.

In the remainder of the paper, we calculate the 1/N1/N expansion of the microchiral partition function of YM2YM_{2} for representative choices of compact orientable spacetimes, and compute the corresponding chiral Gross-Taylor series ZZ in each case. We consider the large NN asymptotic expansion of ZNZ_{N} predicted by the Gross-Taylor series, but do not address its quantitative correctness, though this is a very interesting topic [9, 59].

4. Cylinder and Torus

The Gross-Taylor formula is free of Ω\Omega-factors whenever the number of holes in spacetime is equal to its Euler characteristic. The equation m=22nm=2-2n does not have many solutions in nonnegative integers.

4.1. Cylinder

If m=2m=2 and n=0n=0, spacetime is a cylinder of area tt and the microchiral partition function is

(68) ZNd=zdd!α,β𝖸dPαβλ𝖸Nd(dim𝐕λ)2d!K^αβ(λ)Ψ^tN(λ),Z_{N}^{d}=\frac{z^{d}}{d!}\sum_{\alpha,\beta\in\mathsf{Y}^{d}}P_{\alpha\beta}\sum_{\lambda\in\mathsf{Y}_{N}^{d}}\frac{(\dim\mathbf{V}^{\lambda})^{2}}{d!}\hat{K}_{\alpha\beta}(\lambda)\hat{\Psi}_{\frac{t}{N}}(\lambda),

where Pαβ(U1,U2)=pα(U1)pα(U2)P_{\alpha\beta}(U_{1},U_{2})=p_{\alpha}(U_{1})p_{\alpha}(U_{2}) is a product of trace invariants and K^αβ(λ)=K^α(λ)K^β(λ)\hat{K}_{\alpha\beta}(\lambda)=\hat{K}_{\alpha}(\lambda)\hat{K}_{\beta}(\lambda) is a product of central characters. For 1dN1\leq d\leq N, the corresponding Gross-Taylor expectation is (two holes, zero handles)

(69) Nd(α,β;0)=KαβΨtN.\mathcal{E}_{N}^{d}(\alpha,\beta;0)=\langle K_{\alpha\beta}\Psi_{\frac{t}{N}}\rangle.

4.1.1. Microchiral 1/N1/N expansion

The cylinder expectation is

(70) Nd(α,β;0)=r=0(t)rr!NrKαβKr,\mathcal{E}_{N}^{d}(\alpha,\beta;0)=\sum_{r=0}^{\infty}\frac{(-t)^{r}}{r!N^{r}}\langle K_{\alpha\beta}K^{r}\rangle,

where a finite sum over λ\lambda has been interchanged with an absolutely convergent series in 1/N1/N. The Plancherel expectation

(71) KαβKr=KαKrKβ=Kα|Kr|Kβ\langle K_{\alpha\beta}K^{r}\rangle=\langle K_{\alpha}K^{r}K_{\beta}\rangle=\langle K_{\alpha}|K^{r}|K_{\beta}\rangle

counts walks on the Hurwitz-Cayley graph which begin at a point of the conjugacy class KαK_{\alpha} and end at a point of the conjugacy class KβK_{\beta}. We thus find that the cylinder expectation (69) admits an absolutely convergent series expansion in powers of 1/N1/N whose coefficients enumerate walks between specified conjugacy classes of Sd\mathrm{S}^{d}.

The path count KαβKr\langle K_{\alpha\beta}K^{r}\rangle can be understood topologically by inverting the monodromy correspondence [11]. More precisely, the normalized path count 1d!KαβKr\frac{1}{d!}\langle K_{\alpha\beta}K^{r}\rangle counts orbits of the action of Sd\mathrm{S}^{d} on rr-step walks KαKβK_{\alpha}\to K_{\beta} in the Hurwitz-Cayley graph by simultaneous conjugation of steps and endpoints, each orbit being weighted by the reciprocal of the cardinality of the corresponding stabilizer. Length rr walks KαKβK_{\alpha}\to K_{\beta} modulo conjugation are in bijection with equivalence classes of pairs (X,f)(X,f) consisting of a compact but not necessarily connected Riemann surface XX together with a degree dd holomorphic function f:XYf\colon X\to Y to the Riemann sphere with a fixed branch locus {yα,yβ,y1,,yr}\{y_{\alpha},y_{\beta},y_{1},\dots,y_{r}\} on YY such that the ramification profile of ff over the branch points yαy_{\alpha} and yβy_{\beta} is given by the Young diagrams α\alpha and β\beta, respectively, and the remaining branch points y1,,yry_{1},\dots,y_{r} are simple. The normalized expectations 1d!KαβKr\frac{1}{d!}\langle K_{\alpha\beta}K^{r}\rangle were called double Hurwitz numbers in [50], owing to the fact that Hurwitz had considered these numbers in the case where one of Kα,KβK_{\alpha},K_{\beta} is the identity class; see [44]. We will use the term “double Hurwitz number” for the raw count KαβKr\langle K_{\alpha\beta}K^{r}\rangle of rr-step walks KαKβK_{\alpha}\to K_{\beta} on the Hurwitz-Cayley graph.

To conclude, inverting the usual flow of information

(72) SurfacesMonodromyPermutationsFourierCharacters,\begin{CD}\text{Surfaces}@>{\text{Monodromy}}>{}>\text{Permutations}@>{\text{Fourier}}>{}>\text{Characters},\end{CD}

we find that the microchiral partition function of YM2YM_{2} on the cylinder is a generating function for double Hurwitz numbers.

Theorem 4.1.

For any integers 1dN1\leq d\leq N, the microchiral partition function ZNdZ_{N}^{d} of YM2YM_{2} on a cylinder of area tt admits the absolutely convergent series expansion

ZNd=zdd!α,β𝖸dPαβr=0(t)rr!NrKαβKr,Z_{N}^{d}=\frac{z^{d}}{d!}\sum_{\alpha,\beta\in\mathsf{Y}^{d}}P_{\alpha\beta}\sum_{r=0}^{\infty}\frac{(-t)^{r}}{r!N^{r}}\langle K_{\alpha\beta}K^{r}\rangle,

where KαβKr=Kα|Kr|Kβ\langle K_{\alpha\beta}K^{r}\rangle=\langle K_{\alpha}|K^{r}|K_{\beta}\rangle is a double Hurwitz number.

Yang-Mills on the cylinder is particularly significant given its relation to the heat kernel on the unitary group [20, 42, 34, 59]. Double Hurwitz numbers are major objects of study in Hurwitz theory due to their deep connections with enumerative geometry and integrable systems [8, 38, 43, 50]. However, the relationship between YM2YM_{2} on the cylinder and double Hurwitz numbers given by Theorem 4.1 does not seem to be present in the existing literature on either subject.

Note that while Theorem 4.1 pertains to YM2YM_{2} with cylindrical spacetime, it is degree dd maps to the sphere which are counted by the 1/N1/N expansion of ZNdZ_{N}^{d}. Nevertheless, 1/N1/N respects the boundaries of spacetime by remembering them as two distinguished points on the sphere, which we have taken to be 0 and \infty, and counting maps to the sphere which may have any ramification profile over these points.

4.1.2. Chiral Gross-Taylor series

Theorem 4.1 suggests that the large NN limit of YM2YM_{2} on the cylinder is double Hurwitz theory, the enumerative theory of maps from compact Riemann surfaces to the sphere with two branch points of arbitrary ramification, but it does not directly imply such a statement in any quantitatively meaningful way. What we can do “for free” with Theorem 4.1 is formally set N=N=\infty.

Definition 4.2.

The chiral Gross-Taylor series of the cylinder is the formal power series

Z=1+d=1zdd!α,β𝖸dPαβr=0(t)rr!KαβKr,Z=1+\sum_{d=1}^{\infty}\frac{z^{d}}{d!}\sum_{\alpha,\beta\in\mathsf{Y}^{d}}P_{\alpha\beta}\sum_{r=0}^{\infty}\frac{(-t\hbar)^{r}}{r!}\langle K_{\alpha\beta}K^{r}\rangle,

where z,t,,p1,p2,p3,z,t,\hbar,p_{1},p_{2},p_{3},\dots are commuting indeterminates and

Pαβ=pαpβ=(i=1(α)pαi)(i=1(β)pβi).P_{\alpha\beta}=p_{\alpha}\otimes p_{\beta}=\left(\prod_{i=1}^{\ell(\alpha)}p_{\alpha_{i}}\right)\otimes\left(\prod_{i=1}^{\ell(\beta)}p_{\beta_{i}}\right).

It is important to understand that the chiral Gross-Taylor series is just a convenient way to organize the information in Theorem 4.1 and does not have any additional meaning. However, ZZ can be brought to a form which suggests what the large NN asymptotics of the chiral partition function ZNZ_{N} may actually be. The indeterminate zz is an exponential marker for the degree dd of the symmetric group, while t-t\hbar is an exponential marker for the length rr of a walk KαKβK_{\alpha}\to K_{\beta} on Sd\mathrm{S}^{d} counted by KαβKr\langle K_{\alpha\beta}K^{r}\rangle. From an enumerative perspective we could dispense with \hbar, but in order to anticipate asymptotics it is useful to keep it as an infinitesimal form of 1/N1/N. The tensor PαβP_{\alpha\beta} is a multiplicative marker for the boundary conditions of a walk KαKβK_{\alpha}\to K_{\beta} on Sd\mathrm{S}^{d}. The Exponential Formula [30] gives us the formal logarithm of ZZ as a generating series enumerating “connected” walks on the Cayley graph.

Theorem 4.3.

The chiral Gross-Taylor series of the cylinder is given by Z=eFZ=e^{F}, where

F=d=1zdd!α,β𝖸dPαβr=0(t)rr!KαβKrcF=\sum_{d=1}^{\infty}\frac{z^{d}}{d!}\sum_{\alpha,\beta\in\mathsf{Y}^{d}}P_{\alpha\beta}\sum_{r=0}^{\infty}\frac{(-t\hbar)^{r}}{r!}\langle K_{\alpha\beta}K^{r}\rangle_{c}

and the cumulant KαβKrc\langle K_{\alpha\beta}K^{r}\rangle_{c} is a connected double Hurwitz number.

The combinatorial meaning of the connected double Hurwitz number KαβKr\langle K_{\alpha\beta}K^{r}\rangle is that it counts rr-step walks KαKβK_{\alpha}\to K_{\beta} on the Hurwitz-Cayley graph whose steps and endpoints together generate a transitive subgroup of Sd\mathrm{S}^{d}. Topologically, 1d!KαβKrc\frac{1}{d!}\langle K_{\alpha\beta}K^{r}\rangle_{c} enumerates degree dd covers as above, but with XX irreducible. By the Riemann-Hurwitz formula, KαβKrc\langle K_{\alpha\beta}K^{r}\rangle_{c} vanishes unless r=2g2+(α)+(β)r=2g-2+\ell(\alpha)+\ell(\beta) with g0g\geq 0 the genus of XX. Setting

(73) Hg(α,β)=KαβK2g2+(α)+(β)c,H_{g}(\alpha,\beta)=\langle K_{\alpha\beta}K^{2g-2+\ell(\alpha)+\ell(\beta)}\rangle_{c},

the logarithm of the chiral Gross-Taylor series becomes

(74) F=d=1zdd!α,β𝖸dPαβg=0(t)2g2+(α)+(β)(2g2+(α)+(β))!Hg(α,β).F=\sum_{d=1}^{\infty}\frac{z^{d}}{d!}\sum_{\alpha,\beta\in\mathsf{Y}^{d}}P_{\alpha\beta}\sum_{g=0}^{\infty}\frac{(-t\hbar)^{2g-2+\ell(\alpha)+\ell(\beta)}}{(2g-2+\ell(\alpha)+\ell(\beta))!}H_{g}(\alpha,\beta).

This formal power series is the main object of study in [38, 50]. Since it is a formal series, we are free to sum first over genus and then over degree to get a “genus expansion.”

Theorem 4.4.

The chiral Gross-Taylor series ZZ of YM2YM_{2} on a cylinder admits the topological expansion

Z=expg=02g2Fg,Z=\exp\sum_{g=0}^{\infty}\hbar^{2g-2}F_{g},

where FgF_{g} is given by

Fg=t2g2d=1zdd!α,β𝖸dPαβ(t)(α)+(β)(2g2+(α)+(β))!Hg(α,β),F_{g}=t^{2g-2}\sum_{d=1}^{\infty}\frac{z^{d}}{d!}\sum_{\alpha,\beta\in\mathsf{Y}^{d}}P_{\alpha\beta}\frac{(-t\hbar)^{\ell(\alpha)+\ell(\beta)}}{(2g-2+\ell(\alpha)+\ell(\beta))!}H_{g}(\alpha,\beta),

a generating series for connected double Hurwitz numbers of genus gg.

4.1.3. Large NN

Theorem 4.4 is a formal statement, but it offers a prediction on the NN\to\infty behavior of the chiral partition function

(75) ZN(U1,U2)=1+λ𝖸Nsλ(U1)sλ(U2)et2NC^(λ)Z_{N}(U_{1},U_{2})=1+\sum_{\lambda\in\mathsf{Y}_{N}}s_{\lambda}(U_{1})s_{\lambda}(U_{2})e^{-\frac{t}{2N}\hat{C}(\lambda)}

of YM2YM_{2} on a cylinder of area tt. Remembering that =1/N\hbar=1/N with N=N=\infty, it is natural to conjecture on the basis of Theorem 4.4 that

(76) logZN(U1,U2)g=0N22gFg(U1,U2)\log Z_{N}(U_{1},U_{2})\sim\sum_{g=0}^{\infty}N^{2-2g}F_{g}(U_{1},U_{2})

as NN\to\infty, where

(77) Fg(U1,U2)=t2g2d=1etd2d!α,β𝖸Ndpα(U1)N(α)pβ(U2)N(β)(t)(α)+(β)(2g2+(α)+(β))!Hg(α,β).F_{g}(U_{1},U_{2})=t^{2g-2}\sum_{d=1}^{\infty}\frac{e^{-\frac{td}{2}}}{d!}\sum_{\alpha,\beta\in\mathsf{Y}_{N}^{d}}\frac{p_{\alpha}(U_{1})}{N^{\ell(\alpha)}}\frac{p_{\beta}(U_{2})}{N^{\ell(\beta)}}\frac{(-t)^{\ell(\alpha)+\ell(\beta)}}{(2g-2+\ell(\alpha)+\ell(\beta))!}H_{g}(\alpha,\beta).

This putative NN\to\infty asymptotic expansion of logZN\log Z_{N} is inspired by Theorem 4.4, not implied by it. Indeed, one cannot even claim that the right hand side of (76) is an asymptotic series without first demonstrating that coefficients FgF_{g} defined by (77) converge for tt in some non-trivial and gg-independent set 𝒯\mathcal{T} of positive number (this convergence set is related to the behavior of Brownian motion on UN\mathrm{U}_{N} as NN\to\infty). We will discuss this further below for YM2YM_{2} with spherical spacetime, where the relationship between the chiral Gross-Taylor series and a true large NN approximation is even clearer.

4.2. Torus

If m=0m=0 and n=1n=1, spacetime is a torus of area tt and the microchiral partition function is

(78) ZNd=zdd!λ𝖸Nd(dim𝐕λ)2d!H^(λ)Ψ^tN(λ).Z_{N}^{d}=\frac{z^{d}}{d!}\sum_{\lambda\in\mathsf{Y}_{N}^{d}}\frac{(\dim\mathbf{V}^{\lambda})^{2}}{d!}\hat{H}(\lambda)\hat{\Psi}_{\frac{t}{N}}(\lambda).

For 1dN1\leq d\leq N, the corresponding Gross-Taylor expectation is (no holes, one handle)

(79) Nd(1)=HΨtN.\mathcal{E}_{N}^{d}(1)=\langle H\Psi_{\frac{t}{N}}\rangle.

Since a torus has no boundaries there are no Young diagrams in the expectation (79).

4.2.1. Microchiral 1/N1/N expansion

The torus expectation is

(80) Nd(1)=r=0(t)rr!NrHKr,\mathcal{E}_{N}^{d}(1)=\sum_{r=0}^{\infty}\frac{(-t)^{r}}{r!N^{r}}\langle HK^{r}\rangle,

where a finite sum over λ\lambda has been interchanged with an absolutely convergent series in 1/N1/N. The Plancherel expectation HKr\langle HK^{r}\rangle counts the number of solutions of the equation

(81) ρστ1τr=ρσ\rho\sigma\tau_{1}\dots\tau_{r}=\rho\sigma

in the symmetric group Sd\mathrm{S}^{d} such that ρ,σ\rho,\sigma are arbitrary and the τi\tau_{i}’s are transpositions.

Solutions of the equation (81) have a well-known topological interpretation which we will discuss momentarily, but first we explain how they can be interpreted graph-theoretically which does not seem to be well-known. Invoking the Jucys-Murphy correspondence, every two-factor product ρσ\rho\sigma corresponds uniquely to a concatenation of two strictly monotone walks ρ\rho and σ\sigma on the Hurwitz-Cayley graph such that the initial point of ρ\rho is ι\iota and the initial point of σ\sigma is the terminal point of ρ\rho. Therefore, the Plancherel expectation HKr\langle HK^{r}\rangle is the number of walks on on the Hurwitz-Cayley graph which begin at the identity ι\iota and consist of a concatenation of two strictly monotone walks ρ\rho and σ\sigma, each of any length from 0 to d1d-1, followed by loop with steps τ1,,τr\tau_{1},\dots,\tau_{r} based at ρσ\rho\sigma.

The topological interpretation of HKr\langle HK^{r}\rangle is classical: after normalizing by 1d!\frac{1}{d!} this expectation counts isomorphism classes of degree dd maps f:XYf\colon X\to Y from a compact but not necessarily connected Riemann surface XX to the torus YY which have rr simple branch points y1,,yry_{1},\dots,y_{r} on YY and no other branching [58]. For this reason 1d!HKr\frac{1}{d!}\langle HK^{r}\rangle is called a simple Hurwitz number with torus target, but we will use this term for the raw broken-stick loop count HKr\langle HK^{r}\rangle.

Theorem 4.5.

For any integers 1dN1\leq d\leq N, the microchiral partition function ZNdZ_{N}^{d} of YM2YM_{2} on a torus of area tt admits the absolutely convergent series expansion

ZNd=zdd!r=0(t)rr!NrHKr,Z_{N}^{d}=\frac{z^{d}}{d!}\sum_{r=0}^{\infty}\frac{(-t)^{r}}{r!N^{r}}\langle HK^{r}\rangle,

where HKr\langle HK^{r}\rangle is a simple Hurwitz number with torus target.

4.2.2. Gross-Taylor series

Definition 4.6.

The chiral Gross-Taylor series of the torus is the trivariate formal power series

Z=d=0zdd!r=0(t)rr!HKr.Z=\sum_{d=0}^{\infty}\frac{z^{d}}{d!}\sum_{r=0}^{\infty}\frac{(-t\hbar)^{r}}{r!}\langle HK^{r}\rangle.

Once again, zz is an exponential marker for degree and t-t\hbar is an exponential marker for the cardinality of the branch locus of a map with simple branch points at prescribed locations, or equivlanelty the length of a loop in the Hurwitz-Cayley graph Sd\mathrm{S}^{d} attached to a broken stick. The Exponential Formula gives the formal logarithm of ZZ as the solution to a connected enumeration problem.

Theorem 4.7.

We have Z=eFZ=e^{F}, where

F=d=1zdd!r=0(t)rr!HKrc,F=\sum_{d=1}^{\infty}\frac{z^{d}}{d!}\sum_{r=0}^{\infty}\frac{(-t\hbar)^{r}}{r!}\langle HK^{r}\rangle_{c},

and the cumulant HKrc\langle HK^{r}\rangle_{c} is a connected simple Hurwitz number over the torus.

By the Riemann-Hurwitz formula, the cumulant HKrc\langle HK^{r}\rangle_{c} vanishes unless r=2g2r=2g-2 with g1g\geq 1 the genus of XX, so we set

(82) Hgd=HK2g2c.H_{g}^{d}=\langle HK^{2g-2}\rangle_{c}.

Then, we have the following genus expansion of the chiral Gross-Taylor series.

Theorem 4.8.

The chiral Gross-Taylor series ZZ of YM2YM_{2} on a torus admits the topological expansion

Z=expg=12g2Fg,Z=\exp\sum_{g=1}^{\infty}\hbar^{2g-2}F_{g},

where FgF_{g} with

Fg=t2g2(2g2)!d=1zdd!Hgd.F_{g}=\frac{t^{2g-2}}{(2g-2)!}\sum_{d=1}^{\infty}\frac{z^{d}}{d!}H_{g}^{d}.

The univariate formal power series F~g=d=1zdd!Hgd\tilde{F}_{g}=\sum_{d=1}^{\infty}\frac{z^{d}}{d!}H_{g}^{d} is the main object of study in [19], where it is shown that for g2g\geq 2 the series F~g\tilde{F}_{g} is a quasimodular form of weight 6g66g-6, i.e. a polynomial in the Eisenstein series E2,E4,E6E_{2},E_{4},E_{6} of degree 6g66g-6 with respect to the grading degEk=k\deg E_{k}=k. We refer to [19] for a full discussion of this remarkable fact, which gives a way to understand the convergence properties of the genus gg free energies FgF_{g} in terms of properties of quasimodular forms.

5. Sphere and Disc

The Gross-Taylor formula contains a positive power of Ω\Omega whenever the number of holes in spacetime is less than its Euler characteristic. The inequality m<22nm<2-2n does not have many solutions in nonnegative integers.

5.1. Sphere

If m=0m=0 and n=0n=0, spacetime is a sphere of area tt and the YM2YM_{2} microchiral partition function is

(83) ZNd=(zN2)dd!λ𝖸Nd(dim𝐕λ)2d!Ψ^tN(λ)Ω^1N2(λ).Z_{N}^{d}=\frac{(zN^{2})^{d}}{d!}\sum_{\lambda\in\mathsf{Y}_{N}^{d}}\frac{(\dim\mathbf{V}^{\lambda})^{2}}{d!}\hat{\Psi}_{\frac{t}{N}}(\lambda)\hat{\Omega}_{\frac{1}{N}}^{2}(\lambda).

The corresponding Gross-Taylor expectation is (no holes, no handles)

(84) Nd=ΨtNΩ1N2.\mathcal{E}_{N}^{d}=\langle\Psi_{\frac{t}{N}}\Omega_{\frac{1}{N}}^{2}\rangle.

5.1.1. Microchiral partition function

Both sources of 1/N1/N are now in play. Write the Gross-Taylor expectation (84) as Ω1NΨtNΩ1N\langle\Omega_{\frac{1}{N}}\Psi_{\frac{t}{N}}\Omega_{\frac{1}{N}}\rangle and recall from Section 2 that the exponential distance element in Sd\mathbb{C}\mathrm{S}^{d} is

(85) |Ωq=a=0d1qa|La|\Omega_{q}\rangle=\sum_{a=0}^{d-1}q^{a}|L_{a}\rangle

with |La|L_{a}\rangle the aath level of the Hurwitz-Cayley graph Sd\mathrm{S}^{d}. Then

(86) Ω1NΨtNΩ1N=a,b=0d11Na+br=0(t)rr!NrLaKrLb,\langle\Omega_{\frac{1}{N}}\Psi_{\frac{t}{N}}\Omega_{\frac{1}{N}}\rangle=\sum_{a,b=0}^{d-1}\frac{1}{N^{a+b}}\sum_{r=0}^{\infty}\frac{(-t)^{r}}{r!N^{r}}\langle L_{a}K^{r}L_{b}\rangle,

where the Plancherel expectation

(87) LaKrLb=La|Kr|Lb\langle L_{a}K^{r}L_{b}\rangle=\langle L_{a}|K^{r}|L_{b}\rangle

counts rr-step walks LaLbL_{a}\to L_{b} in the Hurwitz-Cayley graph. This is a sum of double Hurwitz numbers with spherical target,

(88) LaKrLb=α,β𝖸dd(α)=a,d(β)=bKαKrKβ,\langle L_{a}K^{r}L_{b}\rangle=\sum_{\begin{subarray}{c}\alpha,\beta\in\mathsf{Y}^{d}\\ d-\ell(\alpha)=a,\ d-\ell(\beta)=b\end{subarray}}\langle K_{\alpha}K^{r}K_{\beta}\rangle,

and instead of depending on a pair of partitions α,β𝖸d\alpha,\beta\in\mathsf{Y}^{d} marking conjugacy classes in Sd\mathrm{S}^{d} it depends on a pair of integers a,b{0,1,,d1}a,b\in\{0,1,\dots,d-1\} marking levels in Sd\mathrm{S}^{d}. We therefore call LaKrLb\langle L_{a}K^{r}L_{b}\rangle a coarse double Hurwitz number over the sphere.

Theorem 5.1.

For any integers 1dN1\leq d\leq N, the microchiral partition function ZNdZ_{N}^{d} of YM2YM_{2} on a sphere of area tt admits the absolutely convergent series expansion

ZNd=(zN2)dd!a,b=0d11Na+br=0(t)rr!NrLaKrLb,Z_{N}^{d}=\frac{(zN^{2})^{d}}{d!}\sum_{a,b=0}^{d-1}\frac{1}{N^{a+b}}\sum_{r=0}^{\infty}\frac{(-t)^{r}}{r!N^{r}}\langle L_{a}K^{r}L_{b}\rangle,

where LaKrLb\langle L_{a}K^{r}L_{b}\rangle is a coarse double Hurwitz number.

Considerable effort has gone into trying to guess the form of the 1/N1/N expansion of YM2YM_{2} on the sphere [14, 15, 54]. While Theorem 5.1 gives a mathematically simple answer, the physical meaning of this answer is non-obvious. On one hand, the case of a spherical spacetime is like the case of a torus spacetime in that the 1/N1/N expansion actually enumerates worldsheet maps to spacetime. On the other hand, the spherical spacetime seems to conceive of itself as a cylinder with two nonexistent boundaries which manifest as two distinguished points over which the ramification type of the map is arbitrary; the 1/N1/N expansion sees only the number of cycles in the ramification profile of these points.

5.1.2. Chiral Gross-Taylor series

Since the chiral Gross-Taylor series is by definition a formal series, we may enrich it with additional parameters which are useful enumerative markers not present in the quantitative Gross-Taylor formula.

Definition 5.2.

The chiral Gross-Taylor series of the sphere is defined by

Z=1+d=1(z2)dd!a,b=0d1(u)a(v)br=0(t)rr!LaKrLb,Z=1+\sum_{d=1}^{\infty}\frac{(z\hbar^{-2})^{d}}{d!}\sum_{a,b=0}^{d-1}(u\hbar)^{a}(v\hbar)^{b}\sum_{r=0}^{\infty}\frac{(-t\hbar)^{r}}{r!}\langle L_{a}K^{r}L_{b}\rangle,

where z,,t,u,vz,\hbar,t,u,v are formal variables.

As in the case of the chiral Gross-Taylor series of the cylinder, the indeterminate \hbar seems to be redundant, but ultimately it serves a purpose. The indeterminate zz is once again an exponential marker for the degree of Sd\mathrm{S}^{d}, and tt is again an exponential marker for the number of steps in a length rr walk LaLbL_{a}\to L_{b}. The indeterminates uu and vv are ordinary markers for the boundary conditions a,b{0,,d1}a,b\in\{0,\dots,d-1\}. The Jucys-Murphy correspondence gives a second interpretation of the expectation LaKrLb\langle L_{a}K^{r}L_{b}\rangle as counting loops of length a+r+ba+r+b on the Hurwitz-Cayley graph based at ι\iota which consist of aa strictly monotone increasing steps, followed by rr unrestricted steps, followed by bb strictly monotone decreasing steps. the monotone walks which make up the loop carry ordinary weight because their steps cannot be shuffled, whereas unrestricted walks carry exponential weight because their steps can be shuffled.

Theorem 5.3.

We have Z=eFZ=e^{F}, where

F=d=1(z2)dd!a,b=0d1(u)a(v)br=0(t)rr!LaKrLbcF=\sum_{d=1}^{\infty}\frac{(z\hbar^{-2})^{d}}{d!}\sum_{a,b=0}^{d-1}(u\hbar)^{a}(v\hbar)^{b}\sum_{r=0}^{\infty}\frac{(-t\hbar)^{r}}{r!}\langle L_{a}K^{r}L_{b}\rangle_{c}

and the cumulant LaKrLbc\langle L_{a}K^{r}L_{b}\rangle_{c} is the number of rr-step walks LaLbL_{a}\to L_{b} on the Hurwitz-Cayley graph whose steps and endpoints generate a transitive subgroup of Sd\mathrm{S}^{d}.

The cumulant LaKrLbc\langle L_{a}K^{r}L_{b}\rangle_{c} counts connected branched covers with dad-a and dbd-b cycles in the ramification over two branch points and the remaining rr branch points simple. From the Riemann-Hurwitz formula, we see that the cumulant LaKrLbc\langle L_{a}K^{r}L_{b}\rangle_{c} is zero unless

(89) r=2g2+(da)+(db)r=2g-2+(d-a)+(d-b)

for a nonnegative integer gg, the genus of the cover. Therefore, it is natural to parameterize the connected coarse double Hurwitz numbers by genus, writing

(90) Hg(a,b)=LaK2g2+2dabLb.H_{g}(a,b)=\langle L_{a}K^{2g-2+2d-a-b}L_{b}\rangle.

In the genus parameterization, the net power of \hbar in the coefficient of zdz^{d} in FF is

(91) 2dab2g2+2dab=2g2.\hbar^{-2d}\hbar^{a}\hbar^{b}\hbar^{2g-2+2d-a-b}=\hbar^{2g-2}.

This gives a genus expansion of the chiral Gross-Taylor free energy of YM2YM_{2} on the sphere in the physical specialization u=v=1u=v=1 corresponding to the expectation value Ω1NΨtNΩ1N\langle\Omega_{\frac{1}{N}}\Psi_{\frac{t}{N}}\Omega_{\frac{1}{N}}\rangle.

Theorem 5.4.

We have

F|u=v=1=g=02g2Fg,F\big{|}_{u=v=1}=\sum_{g=0}^{\infty}\hbar^{2g-2}F_{g},

where

Fg=t2g2d=1zdd!a,b=0d1t2dabHg(a,b)(2g2+2dab)!.F_{g}=t^{2g-2}\sum_{d=1}^{\infty}\frac{z^{d}}{d!}\sum_{a,b=0}^{d-1}t^{2d-a-b}\frac{H_{g}(a,b)}{(2g-2+2d-a-b)!}.

5.1.3. Large NN

As in the case of the cylinder, Theorem 5.4 predicts an NN\to\infty approximation of the chiral partition function ZNZ_{N} of YM2YM_{2} on a sphere of area t>0t>0. The predicted asymptotics are

(92) logZNg=0N22gFg,\log Z_{N}\sim\sum_{g=0}^{\infty}N^{2-2g}F_{g},

where

(93) Fg=t2g2d=1etd2d!a,b=0d1t2dabHg(a,b)(2g2+2dab)!F_{g}=t^{2g-2}\sum_{d=1}^{\infty}\frac{e^{-\frac{td}{2}}}{d!}\sum_{a,b=0}^{d-1}t^{2d-a-b}\frac{H_{g}(a,b)}{(2g-2+2d-a-b)!}

is a positive series. In order for the prediction 92 to even make sense, it is necessary that there exists a set 𝒯0\mathcal{T}\subseteq\mathbb{R}_{\geq 0} the series Fg=Fg(t)F_{g}=F_{g}(t) given by (93) converges for every g0g\in\mathbb{N}_{0}. The formal power series

(94) F~g=d=1zdd!a,b=0d1t2dabHg(a,b)(2g2+2dab)!\tilde{F}_{g}=\sum_{d=1}^{\infty}\frac{z^{d}}{d!}\sum_{a,b=0}^{d-1}t^{2d-a-b}\frac{H_{g}(a,b)}{(2g-2+2d-a-b)!}

with zz independent of tt does indeed have radius of convergence bounded below by a positive constant independent of gg, see [28] and [7]. Setting z=et/2z=e^{-t/2} with t>0t>0 means that the convergence set 𝒯0\mathcal{T}\subseteq\mathbb{R}_{\geq 0} consists of two disjoint intervals, as predicted in [54]. This is related to the large NN behavior of Brownian loops in UN\mathrm{U}_{N}.

5.2. Disc

In addition to the sphere, there is one and only one other spacetime for which the Gross-Taylor formula contains a positive power of Ω\Omega. If m=1m=1 and n=0n=0, spacetime is a disc of area tt and the YM2YM_{2} microchiral partition function is

(95) ZNd=(zN)dd!α𝖸dpαλ𝖸Nd(dim𝐕λ)2d!K^α(λ)Ψ^tN(λ)Ω^1N(λ).Z_{N}^{d}=\frac{(zN)^{d}}{d!}\sum_{\alpha\in\mathsf{Y}^{d}}p_{\alpha}\sum_{\lambda\in\mathsf{Y}_{N}^{d}}\frac{(\dim\mathbf{V}^{\lambda})^{2}}{d!}\hat{K}_{\alpha}(\lambda)\hat{\Psi}_{\frac{t}{N}}(\lambda)\hat{\Omega}_{\frac{1}{N}}(\lambda).

The corresponding Gross-Taylor expectation is (one hole, no handles)

(96) Nd(α;0)=KαΨtNΩ1N.\mathcal{E}_{N}^{d}(\alpha;0)=\langle K_{\alpha}\Psi_{\frac{t}{N}}\Omega_{\frac{1}{N}}\rangle.

The 1/N1/N expansion of (96) has coefficients given by the Plancherel expecations KαKrLb\langle K_{\alpha}K^{r}L_{b}\rangle, which are “half-coarse” double Hurwitz numbers counting rr-step walks KαLbK_{\alpha}\to L_{b} from a specified conjugacy class to a specified level in the Hurwitz-Cayley graph, or equivalently branched covers of the sphere whose profile over 0 is α\alpha while the ramification over \infty has dbd-b cycles. Being almost identical to what has come before, the details are left to the interested reader.

6. Three-Holed Sphere and One-Holed Torus

The Gross-Taylor formula contains a negative power of Ω\Omega whenever the number of holes in spacetime is greater than its Euler characteristic. The inequality m>22nm>2-2n has infinitely many solutions in nonnegative integers. We now examine the two extremal solutions in which the number of holes in spacetime exceeds the Euler characteristic of spacetime by exactly one.

6.1. Three-holed sphere

If m=3m=3 and n=0n=0, spacetime is a size tt pair of pants and the YM2YM_{2} microchiral partition function is

(97) ZNd=(zN1)dd!α,β,γ𝖸dPαβγλ𝖸Nd(dim𝐕λ)2d!K^αβγ(λ)Ψ^tN(λ)Ω^1N1(λ),Z_{N}^{d}=\frac{(zN^{-1})^{d}}{d!}\sum_{\alpha,\beta,\gamma\in\mathsf{Y}^{d}}P_{\alpha\beta\gamma}\sum_{\lambda\in\mathsf{Y}_{N}^{d}}\frac{(\dim\mathbf{V}^{\lambda})^{2}}{d!}\hat{K}_{\alpha\beta\gamma}(\lambda)\hat{\Psi}_{\frac{t}{N}}(\lambda)\hat{\Omega}_{\frac{1}{N}}^{-1}(\lambda),

where Pαβγ(U1,U2,U3)=pα(U1)pβ(U2)pγ(U3)P_{\alpha\beta\gamma}(U_{1},U_{2},U_{3})=p_{\alpha}(U_{1})p_{\beta}(U_{2})p_{\gamma}(U_{3}) is a product of three trace invariants and K^αβγ(λ)=K^α(λ)K^β(λ)K^γ(λ)\hat{K}_{\alpha\beta\gamma}(\lambda)=\hat{K}_{\alpha}(\lambda)\hat{K}_{\beta}(\lambda)\hat{K}_{\gamma}(\lambda) is a product of three central characters. The corresponding Gross-Taylor expectation for 1dN1\leq d\leq N is (three holes, no handles)

(98) Nd(α,β,γ;0)=KαβγΨtNΩ1N1.\mathcal{E}_{N}^{d}(\alpha,\beta,\gamma;0)=\langle K_{\alpha\beta\gamma}\Psi_{\frac{t}{N}}\Omega_{\frac{1}{N}}^{-1}\rangle.

6.1.1. Microchiral partition function

In order to obtain the 1/N1/N expansion of the Gross-Taylor expectation (98), we must use the result from Section 2 that the exponential distance operator Ωq\Omega_{q} on the Hurwitz-Cayley graph Sd\mathrm{S}^{d} is invertible for |q|<1d1|q|<\frac{1}{d-1} with inverse given by the absolutely convergent series

(99) Ωq1=r=0(q)rMr,\Omega_{q}^{-1}=\sum_{r=0}^{\infty}(-q)^{r}M_{r},

where Mr=hr(J1,,Jd)M_{r}=h_{r}(J_{1},\dots,J_{d}) is the complete symmetric polynomial of degree rr in the Jucys-Murphy operators. Thus for 1dN1\leq d\leq N we have the absolutely convergent 1/N1/N expansion

(100) Nd(α,β,γ;0)=r,s=0(t)r(1)sr!Nr+sKαβγKrMs,\mathcal{E}_{N}^{d}(\alpha,\beta,\gamma;0)=\sum_{r,s=0}^{\infty}\frac{(-t)^{r}(-1)^{s}}{r!N^{r+s}}\langle K_{\alpha\beta\gamma}K^{r}M_{s}\rangle,

where the Plancherel expectation KαβγKrMs\langle K_{\alpha\beta\gamma}K^{r}M_{s}\rangle counts (3+r+s)(3+r+s)-tuples (π,ρ,σ,τ1,,τr,μ1,,μs)(\pi,\rho,\sigma,\tau_{1},\dots,\tau_{r},\mu_{1},\dots,\mu_{s}) from the symmetric group Sd\mathrm{S}^{d} such that

(101) πρστ1τrμ1μs=ι,\pi\rho\sigma\tau_{1}\dots\tau_{r}\mu_{1}\dots\mu_{s}=\iota,

with

(102) π|Kα=ρ|Kβ=σ|Kγ=1\langle\pi|K_{\alpha}\rangle=\langle\rho|K_{\beta}\rangle=\langle\sigma|K_{\gamma}\rangle=1

and τ1,,τr,μ1,,μsK\tau_{1},\dots,\tau_{r},\mu_{1},\dots,\mu_{s}\in K transpositions, such that moreover

(103) μ1μs=(i1j1)(isjs),ik<jk,j1js.\mu_{1}\dots\mu_{s}=(i_{1}j_{1})\dots(i_{s}\ j_{s}),\quad i_{k}<j_{k},\ j_{1}\leq\dots\leq j_{s}.

The expectation KαβγKrMs\langle K_{\alpha\beta\gamma}K^{r}M_{s}\rangle is a mixed triple Hurwitz number. To interpret it graphically as counting walks in the Hurwitz-Cayley graph, we first decompose

(104) Kαβγ=KαKβKγ=η𝖸dcβγηKαKη,K_{\alpha\beta\gamma}=K_{\alpha}K_{\beta}K_{\gamma}=\sum_{\eta\in\mathsf{Y}^{d}}c_{\beta\gamma}^{\eta}K_{\alpha}K_{\eta},

where cβγηc_{\beta\gamma}^{\eta} are the connection coefficients of the class algebra, i.e. the structure constants of the center of Sd\mathbb{C}\mathrm{S}^{d} relative the class basis, which are nonnegative integers. This gives the mixed triple Hurwitz number as the nonnegative integral linear combination

(105) KαβγKrMs=η𝖸dcβγηKαKrMsKη,\langle K_{\alpha\beta\gamma}K^{r}M_{s}\rangle=\sum_{\eta\in\mathsf{Y}^{d}}c_{\beta\gamma}^{\eta}\langle K_{\alpha}K^{r}M_{s}K_{\eta}\rangle,

where now each term KαKrMsKη\langle K_{\alpha}K^{r}M_{s}K_{\eta}\rangle is a mixed double Hurwitz number [27] which counts length r+sr+s walks KαKηK_{\alpha}\to K_{\eta} in the Hurwitz-Cayley graph whose first rr streps are unrestricted and whose last ss steps are weakly monotone in the sense of Section 2. In this very simple-minded sense, the connection coefficients of the class algebra can be viewed as giving an analogue of the Wick formula which allows to write an arbitrary Plancherel expectation Kα1αm\langle K_{\alpha_{1}\dots\alpha_{m}}\dots\rangle as a nonnegative integral linear combination of Plancherel expectations Kαβ\langle K_{\alpha\beta}\dots\rangle.

In the case s=0s=0, the mixed triple Hurwitz number KαβγKrMs\langle K_{\alpha\beta\gamma}K^{r}M_{s}\rangle becomes the purely classical triple Hurwitz number KαβγKr\langle K_{\alpha\beta\gamma}K^{r}\rangle whose topological meaning is the same as that of the classical double Hurwitz numbers, except that the covers f:XYf\colon X\to Y of the sphere being counted have three points with arbitrary ramification profile in their branch locus. In the general case, one can say that the mixed triple Hurwitz number KαβγKrMs\langle K_{\alpha\beta\gamma}K^{r}M_{s}\rangle counts some subset of the covers counted by the classical triple Hurwitz number KαβγKr+s\langle K_{\alpha\beta\gamma}K^{r+s}\rangle. A reasonable objection to this is that the subset of covers counted depends on a labeling of the points in the fibre over the unbranched basepoint used to perform the monodromy construction, and so is not geometrically defined. A reasonable response is that the cardinality of the restricted subset of covers does not depend on the choice of a labeling because of the centrality of symmetric polynomial functions of Jucys-Murphy elements.

Theorem 6.1.

For any integers 1dN1\leq d\leq N, the microchiral partition function ZNdZ_{N}^{d} of YM2YM_{2} on a sphere of area tt admits the absolutely convergent series expansion

ZNd=(zN1)dd!α,β,γ𝖸dPαβγr,s=0(t)r(1)sr!Nr+sKαβγKrMs,Z_{N}^{d}=\frac{(zN^{-1})^{d}}{d!}\sum_{\alpha,\beta,\gamma\in\mathsf{Y}^{d}}P_{\alpha\beta\gamma}\sum_{r,s=0}^{\infty}\frac{(-t)^{r}(-1)^{s}}{r!N^{r+s}}\langle K_{\alpha\beta\gamma}K^{r}M_{s}\rangle,

where KαβγKrMs\langle K_{\alpha\beta\gamma}K^{r}M_{s}\rangle is a mixed triple Hurwitz number.

6.1.2. Chiral Gross-Taylor series

Definition 6.2.

The chiral Gross-Taylor series of the three-holed sphere is

Z=d=0(z)dd!α,β,γ𝖸dPαβγr,s=0(t)rr!(u)sKαβγKrMs,Z=\sum_{d=0}^{\infty}\frac{(z\hbar)^{d}}{d!}\sum_{\alpha,\beta,\gamma\in\mathsf{Y}^{d}}P_{\alpha\beta\gamma}\sum_{r,s=0}^{\infty}\frac{(-t\hbar)^{r}}{r!}(-u\hbar)^{s}\langle K_{\alpha\beta\gamma}K^{r}M_{s}\rangle,

where z,t,,p1,p2,p3,z,t,\hbar,p_{1},p_{2},p_{3},\dots are indeterminates and

Pαβγ=(i=1(α)pαi)(i=1(β)pβi)(i=1(γ)pγ).P_{\alpha\beta\gamma}=\left(\prod_{i=1}^{\ell(\alpha)}p_{\alpha_{i}}\right)\otimes\left(\prod_{i=1}^{\ell(\beta)}p_{\beta_{i}}\right)\otimes\left(\prod_{i=1}^{\ell(\gamma)}p_{\gamma}\right).
Theorem 6.3.

We have Z=eFZ=e^{F}, where

F=d=0(z1)dd!α,β,γ𝖸dPαβγr,s=0(t)rr!(u)sKαβγKrMsF=\sum_{d=0}^{\infty}\frac{(z\hbar^{-1})^{d}}{d!}\sum_{\alpha,\beta,\gamma\in\mathsf{Y}^{d}}P_{\alpha\beta\gamma}\sum_{r,s=0}^{\infty}\frac{(-t\hbar)^{r}}{r!}(-u\hbar)^{s}\langle K_{\alpha\beta\gamma}K^{r}M_{s}\rangle

and the cumulant KαβγKrMsc\langle K_{\alpha\beta\gamma}K^{r}M_{s}\rangle_{c} is a connected mixed triple Hurwitz number.

From the Riemann-Hurwitz formula, KαβγKrMsc\langle K_{\alpha\beta\gamma}K^{r}M_{s}\rangle_{c} is zero unless

(106) r+s=2g2+d+(α)+(β)+(γ),r+s=2g-2+d+\ell(\alpha)+\ell(\beta)+\ell(\gamma),

so that the net power of \hbar at degree dd in FF is

(107) d2g2d+(α)+(β)+(γ)=2g2(α)+(β)+(γ),\hbar^{d}\hbar^{2g-2-d+\ell(\alpha)+\ell(\beta)+\ell(\gamma)}=\hbar^{2g-2}\hbar^{\ell(\alpha)+\ell(\beta)+\ell(\gamma)},

and we obtain the topological expansion of the chiral Gross-Taylor series of the three-holed sphere.

Theorem 6.4.

The chiral Gross-Taylor series ZZ of YM2YM_{2} on a three-holed sphere admits the topological expansion

Z=expg=02g2Fg,Z=\exp\sum_{g=0}^{\infty}\hbar^{2g-2}F_{g},

where FgF_{g} is given by

Fg=d=1zdd!α,β,γ𝖸dPαβγ(α)+(β)+(γ)r+s=2g2d+(α)+(β)+(γ)(t)rr!(u)sKαβγKrMsc.F_{g}=\sum_{d=1}^{\infty}\frac{z^{d}}{d!}\sum_{\alpha,\beta,\gamma\in\mathsf{Y}^{d}}P_{\alpha\beta\gamma}\hbar^{\ell(\alpha)+\ell(\beta)+\ell(\gamma)}\sum_{r+s=2g-2-d+\ell(\alpha)+\ell(\beta)+\ell(\gamma)}\frac{(-t)^{r}}{r!}(-u)^{s}\langle K_{\alpha\beta\gamma}K^{r}M_{s}\rangle_{c}.

a generating series for connected double Hurwitz numbers of genus gg.

Note that the internal sum of FgF_{g} is finite, and its extreme terms correspond to connected classical and monotone triple Hurwitz numbers: the s=0s=0 term is

(108) (t)(α)+(β)+(γ)(2g2d+(α)+(β)+(γ))!Hg(α,β,γ)\frac{(-t)^{\ell(\alpha)+\ell(\beta)+\ell(\gamma)}}{(2g-2-d+\ell(\alpha)+\ell(\beta)+\ell(\gamma))!}H_{g}(\alpha,\beta,\gamma)

with

(109) Hg(α,β,γ)=KαβγK2g2d+(α)+(β)+(γ)cH_{g}(\alpha,\beta,\gamma)=\langle K_{\alpha\beta\gamma}K^{2g-2-d+\ell(\alpha)+\ell(\beta)+\ell(\gamma)}\rangle_{c}

a connected classical triple Hurwitz number, and the r=0r=0 term is

(110) (u)(α)+(β)+(γ)Hg(α,β,γ)(-u)^{\ell(\alpha)+\ell(\beta)+\ell(\gamma)}H_{g}^{\leq}(\alpha,\beta,\gamma)

with

(111) Hg(α,β,γ)=KαβγM2g2d+(α)+(β)+(γ)cH_{g}^{\leq}(\alpha,\beta,\gamma)=\langle K_{\alpha\beta\gamma}M_{2g-2-d+\ell(\alpha)+\ell(\beta)+\ell(\gamma)}\rangle_{c}

a connected monotone triple Hurwitz number.

6.2. One-holed torus

If m=1m=1 and n=1n=1, spacetime is a one-holed torus of area tt and the YM2YM_{2} microchiral partition function is

(112) ZNd=(zN1)dd!α𝖸dpαλ𝖸Nd(dim𝐕λ)2d!H^(α)K^α(λ)Ψ^tN(λ)Ω^1N1(λ).Z_{N}^{d}=\frac{(zN^{-1})^{d}}{d!}\sum_{\alpha\in\mathsf{Y}^{d}}p_{\alpha}\sum_{\lambda\in\mathsf{Y}_{N}^{d}}\frac{(\dim\mathbf{V}^{\lambda})^{2}}{d!}\hat{H}(\alpha)\hat{K}_{\alpha}(\lambda)\hat{\Psi}_{\frac{t}{N}}(\lambda)\hat{\Omega}_{\frac{1}{N}}^{-1}(\lambda).

The corresponding Gross-Taylor expectation for 1dN1\leq d\leq N is (one hole, one handle)

(113) Nd(α;1)=HKαΨtNΩ1N1\mathcal{E}_{N}^{d}(\alpha;1)=\langle HK_{\alpha}\Psi_{\frac{t}{N}}\Omega_{\frac{1}{N}}^{-1}\rangle

6.2.1. Microchiral partition function

Based on our experience above we can easily predict what will happen in this instance of YM2YM_{2}: since spacetime has one hole, the corresponding Hurwitz theory must pertain to covers with one free ramification point and simple branching elsewhere; since spacetime has Euler characteristic zero the base curve of this Hurwitz theory must be the torus; since the number of holes in spacetime exceeds its Euler characteristic this Hurwitz theory must be mixed.

This is borne out by calculations entirely analogous to those already performed. In the stable range, 1dN1\leq d\leq N, the expectation (113) admits the absolutely convergent 1/N1/N expansion

(114) Nd(α;1)=r,s=0(t)r(1)sr!Nr+sHKαKrMs.\mathcal{E}_{N}^{d}(\alpha;1)=\sum_{r,s=0}^{\infty}\frac{(-t)^{r}(-1)^{s}}{r!N^{r+s}}\langle HK_{\alpha}K^{r}M_{s}\rangle.

Reorganizing this into HKrMsKα\langle HK^{r}M_{s}K_{\alpha}\rangle, the expectation counts tuples (ρ,σ,τ1,,τr,μ1,,μs,π)(\rho,\sigma,\tau_{1},\dots,\tau_{r},\mu_{1},\dots,\mu_{s},\pi) of permutations from S(d)\mathrm{S}(d) such that

(115) ρ1σ1ρστ1τrμ1μs=π.\rho^{-1}\sigma^{-1}\rho\sigma\tau_{1}\dots\tau_{r}\mu_{1}\dots\mu_{s}=\pi.

The target permutation π\pi belongs to the conjugacy class KαK_{\alpha}, while τ1,,τr,μ1,,μs\tau_{1},\dots,\tau_{r},\mu_{1},\dots,\mu_{s} are transpositions with the μ\mu’s weakly monotone increasing. The transpositions ρ,σ\rho,\sigma coming from the commutator sum |H|H\rangle are arbitrary. Using the Jucys-Murphy correspondence, we can interpret the product ρ1σ1ρσ\rho^{-1}\sigma^{-1}\rho\sigma as counting a concatenation of four strictly monotone walks, the first two legs of which are strictly monotone decreasing and the final two are strictly monotone increasing, being the first two trajectories after path reversal. All together, this gives a walk in the Hurwitz-Cayley graph from ι\iota to π\pi in 2(|ρ|+|σ|)+r+s2(|\rho|+|\sigma|)+r+s steps, where the first 2(|ρ|+|σ|)2(|\rho|+|\sigma|) steps form a walk of shape \searrow\searrow\nearrow\nearrow, the next rr steps are arbitrary, and the final ss steps are weakly monotone.

The topological interpretation of the normalized expectation 1d!HKαKrMs\frac{1}{d!}\langle HK_{\alpha}K^{r}M_{s}\rangle is that it counts a subset of the covers counted by 1d!HKαKr+s\frac{1}{d!}\langle HK_{\alpha}K^{r+s}\rangle, which are branched covers of a torus (with no hole) having 1+r+s1+r+s total branch points at fixed locations, one of which has ramification type α\alpha and the rest are simple.

Theorem 6.5.

For any integers 1dN1\leq d\leq N, the microchiral partition function ZNdZ_{N}^{d} of YM2YM_{2} on a one-holed torus of area tt admits the absolutely convergent series expansion

ZNd=(zN1)dd!r,s=0(t)rr!Nr(1)sNsα𝖸dpαHKαKrMsZ_{N}^{d}=\frac{(zN^{-1})^{d}}{d!}\sum_{r,s=0}^{\infty}\frac{(-t)^{r}}{r!N^{r}}\frac{(-1)^{s}}{N^{s}}\sum_{\alpha\in\mathsf{Y}^{d}}p_{\alpha}\langle HK_{\alpha}K^{r}M_{s}\rangle

where HKαKrMs\langle HK_{\alpha}K^{r}M_{s}\rangle is a mixed single Hurwitz number with torus target.

6.2.2. Chiral Gross-Taylor series

Definition 6.6.

The chiral Gross-Taylor series of the one-hold torus is

Z=d=0(z)dd!α𝖸dpαr,s=0(t)rr!(u)sHKαKrMs,Z=\sum_{d=0}^{\infty}\frac{(z\hbar)^{d}}{d!}\sum_{\alpha\in\mathsf{Y}^{d}}p_{\alpha}\sum_{r,s=0}^{\infty}\frac{(-t\hbar)^{r}}{r!}(-u\hbar)^{s}\langle HK_{\alpha}K^{r}M_{s}\rangle,

where z,t,u,,p1,p2,p3,z,t,u,\hbar,p_{1},p_{2},p_{3},\dots are commuting indeterminates.

Theorem 6.7.

We have Z=eFZ=e^{F}, where

F=d=0(z)dd!α𝖸dpαr,s=0(t)rr!(u)sHKαKrMsF=\sum_{d=0}^{\infty}\frac{(z\hbar)^{d}}{d!}\sum_{\alpha\in\mathsf{Y}^{d}}p_{\alpha}\sum_{r,s=0}^{\infty}\frac{(-t\hbar)^{r}}{r!}(-u\hbar)^{s}\langle HK_{\alpha}K^{r}M_{s}\rangle

and the cumulant HKαKrMsc\langle HK_{\alpha}K^{r}M_{s}\rangle_{c} is a connected mixed single Hurwitz number with torus target.

The genus expansion is now obtained from the Riemann-Hurwitz formula just as in all above computations, e.g. as in the case of a torus without a hole. The quasimodularity of the genus specific free energies FgF_{g}, g1g\geq 1, which are generating functions for connected mixed single Hurwitz numbers with torus target and fixed genus gg of the covering surface, is a recent result of [40].

7. Area Zero

The microchiral partition function of YM2YM_{2} on an area zero compact orientable surface of with mm holes and nn handles is

(116) WNd=(zN22nm)dd!α1,,αm𝖸dPα1αmλ𝖸Nd(dim𝐕λ)2d!HnKα1αm(λ)Ω1N22nm(λ).W_{N}^{d}=\frac{\left(zN^{2-2n-m}\right)^{d}}{d!}\sum_{\alpha^{1},\dots,\alpha^{m}\in\mathsf{Y}^{d}}P_{\alpha^{1}\dots\alpha^{m}}\sum_{\lambda\in\mathsf{Y}_{N}^{d}}\frac{(\dim\mathbf{V}^{\lambda})^{2}}{d!}H^{n}K_{\alpha^{1}\dots\alpha^{m}}(\lambda)\Omega_{\frac{1}{N}}^{2-2n-m}(\lambda).

This is obtained from the true Gross-Taylor formula, Theorem 3.1, by setting t=0t=0 and it a definition, not a theorem. Not that we have retained the variable zz in this definition rather than setting it to 11 as we would have to were z=et/2z=e^{-t/2}. This is done so that we can define the full chiral partition function of YM2YM_{2} on an area zero surface by

(117) WN=1+d=1WNd,W_{N}=1+\sum_{d=1}^{\infty}W_{N}^{d},

a formal power series considered in a number of papers in the YM2YM_{2} literature [14, 15]. The main result of these works is that in the formal N=N=\infty limit, the chiral partition function WNW_{N} of area zero YM2YM_{2} becomes a generating function for Euler characteristics of the Hurwitz moduli spaces of branched covers. At the same time, our mixed Hurwitz formalism shows that this degeneration of YM2YM_{2} corresponds exactly to pure monotone Hurwitz theory, leading to the conclusion that monotone Hurwitz numbers have an interpretation as Euler characteristics of Hurwitz moduli spaces. It would be extremely interesting to make this connection precise. Here we give several examples of zero area YM2YM_{2} as pure monotone Hurwitz theory.

7.1. Cylinder

The michrochiral partition function of YM2YM_{2} on a cylinder of area zero is

(118) WNd=N2dα,β𝖸dPαβλ𝖸Nd(dim𝐕λd!)2Kαβ(λ).W_{N}^{d}=N^{2d}\sum_{\alpha,\beta\in\mathsf{Y}^{d}}P_{\alpha\beta}\sum_{\lambda\in\mathsf{Y}_{N}^{d}}\left(\frac{\dim\mathbf{V}^{\lambda}}{d!}\right)^{2}K_{\alpha\beta}(\lambda).

In the stable range, 1dN1\leq d\leq N, the internal sum is the Plancherel expectation

(119) Kαβ=Kα|Kβ=δαβ|Kα|.\langle K_{\alpha\beta}\rangle=\langle K_{\alpha}|K_{\beta}\rangle=\delta_{\alpha\beta}|K_{\alpha}|.

The area zero chiral Gross-Taylor series is therefore

(120) W=1+d=1zdd!α𝖸dpαpα|Kα|,W=1+\sum_{d=1}^{\infty}\frac{z^{d}}{d!}\sum_{\alpha\in\mathsf{Y}_{d}}p_{\alpha}\otimes p_{\alpha}|K_{\alpha}|,

where the internal sum can be written

(121) α𝖸dpαpα|Kα|=πSdpt(π)pt(π),\sum_{\alpha\in\mathsf{Y}_{d}}p_{\alpha}\otimes p_{\alpha}|K_{\alpha}|=\sum_{\pi\in\mathrm{S}^{d}}p_{t(\pi)}\otimes p_{t(\pi)},

where t(π)t(\pi) is the cycle type of the permutation π\pi. This is multiplicative function on permutations, and the Exponential Formula gives

(122) W=expd=1zddpdpd.W=\exp\sum_{d=1}^{\infty}\frac{z^{d}}{d}p_{d}\otimes p_{d}.

Noting that

(123) W=1+d=1zdλ𝖸dsλsλ,W=1+\sum_{d=1}^{\infty}z^{d}\sum_{\lambda\in\mathsf{Y}^{d}}s_{\lambda}\otimes s_{\lambda},

we see that this is exactly the Cauchy identity from symmetric function theory.

7.1.1. Sphere

For a sphere of area zero, the chiral Gross-Taylor series is

(124) W=1+d=1(z2)dd!a,b=0d1(u)a(v)bLaLb.W=1+\sum_{d=1}^{\infty}\frac{(z\hbar^{-2})^{d}}{d!}\sum_{a,b=0}^{d-1}(u\hbar)^{a}(v\hbar)^{b}\langle L_{a}L_{b}\rangle.

We have that

(125) LaLb=La|Lb=δab[dda],\langle L_{a}L_{b}\rangle=\langle L_{a}|L_{b}\rangle=\delta_{ab}{d\brack d-a},

where the right hand side is the Stirling cycle number enumerating permutations in Sd\mathrm{S}^{d} with dad-a cycles. This gives

(126) W=1+d=1(z2)dd!a=0d12a[dda]W=1+\sum_{d=1}^{\infty}\frac{(z\hbar^{-2})^{d}}{d!}\sum_{a=0}^{d-1}\hbar^{2a}{d\brack d-a}

or equivalently

(127) W=1+d=1zdd!a=1d2a[da]=exp(2d=1zdd),W=1+\sum_{d=1}^{\infty}\frac{z^{d}}{d!}\sum_{a=1}^{d}\hbar^{-2a}{d\brack a}=\exp\left(\hbar^{-2}\sum_{d=1}^{\infty}\frac{z^{d}}{d}\right),

the standard generating series for Stirling cycle numbers.

7.2. Three-holed sphere

For a three-holed sphere of area zero, the chiral Gross-Taylor series is

(128) W=d=0(z)dd!α,β,γ𝖸dPαβγs=0(u)sKαβγMs,W=\sum_{d=0}^{\infty}\frac{(z\hbar)^{d}}{d!}\sum_{\alpha,\beta,\gamma\in\mathsf{Y}^{d}}P_{\alpha\beta\gamma}\sum_{s=0}^{\infty}(-u\hbar)^{s}\langle K_{\alpha\beta\gamma}M_{s}\rangle,

the total generating series for possibly disconnected monotone triple Hurwitz numbers, and the corresponding topological expansion is

(129) W=expg=02g2FgW=\exp\sum_{g=0}^{\infty}\hbar^{2g-2}F_{g}

with

(130) Fg=u2g2d=1zdd!α,β,γ𝖸dPαβγ(u)(α)+(β)+(γ)Hg(α,β,γ)F_{g}=u^{2g-2}\sum_{d=1}^{\infty}\frac{z^{d}}{d!}\sum_{\alpha,\beta,\gamma\in\mathsf{Y}^{d}}P_{\alpha\beta\gamma}(u\hbar)^{\ell(\alpha)+\ell(\beta)+\ell(\gamma)}H_{g}^{\leq}(\alpha,\beta,\gamma)

the generating series for connected monotone triple Hurwitz numbers of genus gg. Interestingly, monotone triple Hurwitz numbers also arise in the context of matrix models, specifically in relation to Jacobi ensembles [31].

References

  • [1] A. Alexandrov, G. Chapuy, B. Eynard, J. Harnad, Weighted Hurwitz numbers and topological recursion, Commun. Math. Phys. 375 (2020), 237-305.
  • [2] M. Aouchiche, P. Hansen, Distance spectra of graphs: a survey, Linear Algebra Appl. 458 (2014), 301-386.
  • [3] I. Bars, F. Green, Complete integration of U(N)U(N) lattice gauge theory in a large NN limit, Phys. Rev. D 20 (1979), 3311-3330.
  • [4] G. Berkolaiko, J. Kuipers, Combinatorial theory of the semiclassical evaluation of transport moments. I. Equivalence with the random matrix approach, J. Math. Phys. 54 (2013), 112103.
  • [5] E. Brezin, D. Gross, The external field problem in the large NN limit of QCD, Phys. Lett. 97 (1980), 120-124.
  • [6] P. Brouwer, C. Beenaker, Diagrammatic method of integration over the unitary group, with applications to quantum transport in mesoscopic systems, J. Math. Phys. 37 (1996), 4904-4934.
  • [7] V. Bonzom, G. Chapuy, S. Charbonnier, E. Garcia-Failde, Topological recursion for Orlov-Scherbin tau functions, and constellations with internal faces, https://arxiv.org/abs/2206.14768
  • [8] G. Borot, N. Do, M. Karev, D. Lewanski, E. Moskovsky, Double Hurwitz numbers: polynomiality, topological recursion and intersection theory, Math. Ann. 387 (2023), 179-243.
  • [9] A. de Monvel, M. Shcherbina, Free energy of the two-dimensional U(n)U(n) gauge field theory on the sphere, Theor. Math. Phys. 115 (1998), 670-679.
  • [10] S. Butler, E. Cooper, A. Li, K. Lorenzen, Z. Schopick, Spectral properties of the exponential distance matrix, Involve 15 (2022), 739-760.
  • [11] R. Cavalieri, E. Miles, Riemann Surfaces and Algebraic Curves: an Introduction to Hurwitz Theory.
  • [12] B. Collins, S. Matsumoto, J. Novak, The Weingarten calculus, Notices Amer. Math. Soc. 69 (2022), 734-745.
  • [13] F. Chung, The heat kernel as page rank, Proc. Nat. Acad. Sci. 104 (2007), 19735-1940.
  • [14] S. Cordes, G. Moore, S. Ramgoolam, Lectures on 2D Yang-Mills theory, equivariant cohomology, and topological field theories, Nucl. Phys. B (Proc. Suppl.) 41 (1995), 184-244.
  • [15] S. Cordes, G. Moore, S. Ramgoolam, Large NN 2D Yang-Mills theory and topological string theory, Commun. Math. Phys. 185 (1997), 543-619.
  • [16] M. Creutz, Feynman rules for lattice gauge theory, Reviews of Modern Physics, Vol. 50, No. 3, July 1978.
  • [17] B. De Wit, G. ’t Hooft, Nonconvergence of the 1/N1/N expansion for SU(N)SU(N) gauge fields on a lattice, Phys. Lett. 69 (1977), 61-64.
  • [18] P. Diaconis, C. Greene, Applications of Murphy’s elements, Technical Report 335 (1989), Department of Statistics, Stanford University.
  • [19] R. Dijkgraaf, Mirror symmetry and elliptic curves, The Moduli Space of Curves, Progress in Mathematics 129, 1995.
  • [20] B. Driver, F. Gabriel, B. Hall, T. Kemp, The Makeenko-Migdal equations for Yang-Mills theory on compact surfaces, Commun. Math. Phys. 352 (2017), 967-978.
  • [21] B. Dubrovin, D. Yang, D. Zagier, Classical Hurwitz numbers and related combinatorics, Moscow Math. J. 17 (2017), 601-633.
  • [22] N. Elkies, D. Eisenbud, J. Harris, R. Speiser, On the Hurwitz scheme and its monodromy, Compos. Math. 77 (1991), 95-117.
  • [23] D. Gewurz, F. Merola, Some factorisations counted by Catalan numbers, Eur. J. Comb. 27 (2006), 990-994.
  • [24] I. Goulden, M. Guay-Paquet, J. Novak, Monotone Hurwitz numbers in genus zero, Canad. J. Math. 65 (2013), 1020-1042.
  • [25] I. Goulden, M. Guay-Paquet, J. Novak, Polynomiality of monotone Hurwitz numbers in higher genera, Adv. Math. 238 (2013), 1-23.
  • [26] I. Goulden, M. Guay-Paquet, J. Novak, Monotone Hurwitz theory and the HCIZ integral, Ann. Math. Blaise Pascal 21 (2014), 71-89.
  • [27] I. Goulden, M. Guay-Paquet, J. Novak, Toda equations and piecewise polynomiality for mixed double Hurwitz numbers, SIGMA 12 (2016), 1-10.
  • [28] I. Goulden, M. Guay-Paquet, J. Novak, On the convergence of monotone Hurwitz generating functions, Ann. Comb. 21 (2017), 73-81.
  • [29] I. Goulden, D. Jackson, Transitive factorizations into transpositions and holomorphic maps on the sphere, Proc. Amer. Math. Soc. 125 (1997), 51-60.
  • [30] I. Goulden, D. Jackson, Combinatorial Enumeration, Dover Publications, 2004.
  • [31] M. Gisoni, T. Grava, G. Ruzza, Jacobi ensemble, Wilson polynomials and Hurwitz numbers, Lett. Math. Phys. 111 (2021), 67.
  • [32] R. Graham, L. Lovasz, Distance matrix polynomials of trees, Adv. Math. 29 (1978), 60-88.
  • [33] D. Gross, Two-dimensional QCD as a string theory, Nucl. Phys. B 400 (1993), 161-180.
  • [34] D. Gross, A. Matytsin, Some properties of large-NN two-dimensional Yang-Mills theory, Nucl. Phys. B 437 (1995), 541-584. Nucl. Phys. B 380 (1992), 168-180.
  • [35] D. Gross, W. Taylor, Two-dimensional QCD is a string theory, Nucl. Phys. B 400 (1993), 181-208.
  • [36] D. Gross, W. Taylor, Twists and Wilson loops in the string theory of two-dimensional QCD, Nucl. Phys. B 403 (1993), 395-449.
  • [37] D. Gross, E. Witten, Possible third-order phase transition in the large-NN lattice gauge theory, Phys. Rev. D 21 (1980), 446-453.
  • [38] I. Goulden, D. Jackson, R. Vakil, Towards the geometry of double Hurwitz numbers, Adv. Math. 198 (2005), 43-92.
  • [39] M. Guay-Paquet, J. Harnad, Generating functions for weighted Hurwitz numbers, J. Math. Phys. 58 (2017), 083503.
  • [40] M. Hahn, J. van Ittersum, F, Leid, Triply mixed coverings of arbitrary base curves: quasimodularity, quantum curves, and a mysterious topological recursion, Ann. Inst. H. Poincaré D 9 (2022), 239-296.
  • [41] A. Jucys, Symmetric polynomials and the center of the symmetric group ring, Rep. Math. Phys. 5 (1974), 107-112.
  • [42] V. Kazakov, T. Wynter, Large NN phase transition in the heat kernel on the U(N)U(N) group, Nucl. Phys. B 440 (1995), 407-420.
  • [43] M. Kazarian, S. Lando, An algebro-geometric proof of Witten’s conjecture, J. Amer. Math. Soc. 20 (2007), 1079-1089.
  • [44] S. Lando, Hurwitz numbers: on the edge between combinatorics and geometry, ICM Proceedings 2010.
  • [45] M. Makhina, J. Novak, Hurwitz theory and absolute order, in preparation.
  • [46] S. Matsumoto, J. Novak, Jucys-Murphy elements and unitary matrix integrals, Int. Math. Res. Not. IMRN 2 (2013), 362-397.
  • [47] G. Murphy, A new construction of Young’s seminormal representation of the symmetric group, J. Algebra 69 (1981), 287-297.
  • [48] J. Novak, Jucys-Murphy elements and the unitary Weingarten function, Banach Center Publ. 89 (2010), 231-235.
  • [49] K. O’Brien, J.-B. Zuber, Strong coupling expansion of large NN QCD and surfaces, Nucl. Phys. B. 253 (1985), 621-634.
  • [50] A. Okounkov, Toda equations for Hurwitz numbers, Math. Res. Lett. 7 (2000), 447-453.
  • [51] S. Samuel, U(N)U(N) integrals, 1/N1/N, and the De Wit-’t Hooft anomalies, J. Math. Phys. 21 (1980), 2695-2703.
  • [52] J.-P. Serre, Linear Representations of Finite Groups, Springer Graduate Texts in Mathematics 42, 1977.
  • [53] R. Stanley, Algebraic Combinatorics: Walks, Trees, Tableaux and More, Second Edition, Springer Undergraduate Texts in Mathematics, 2018.
  • [54] W. Taylor, Counting strings and phase transitions in 2D QCD, https://arxiv.org/abs/hep-th/9404175.
  • [55] S. Wadia, A study of U(N)U(N) lattice gauge theory in two dimensions, https://arxiv.org/abs/1212.2906v1
  • [56] D. Weingarten, Non-planar diagram in the limit of U(N)U(N) and SU(N)SU(N) lattice gauge theories, Phys. Lett. 90 (1980), 285-288.
  • [57] D. Zagier, Realizability of a model in infinite statistics, Commun. Math. Phys. 147 (1992), 199-210.
  • [58] D. Zagier, Applications of the Representation Theory of Finite Groups, Appendix to Graphs on Surfaces and Their Applications by S. Lando and A. Zvonkin, Encyclopedia of Mathematical Sciences 141, Springer-Verlag 2004.
  • [59] S. Zelditch, Macdonald’s identities and the large NN limit of YM2YM_{2} on the cylinder, Commun. Math. Phys. 245 (2004), 611-626.
  • [60] P. Zinn-Justin, HCIZ integral and 2D Toda lattice hierarchy, Nucl. Phys. B 634 (2002), 417-432.