
NON-RECURSIVE COUNTS OF GRAPHS ON SURFACES 3
counts are each divisible by j! (2ν)j. Hence, in this case the quotient counts the number of coarser
map equivalence classes without labeling.
As previously mentioned, the method we employ to calculate map counts stems from statistical
physics and more precisely from application of the Wick calculus to perturbation expansions of the
random matrix partition function (for hermitian ensembles) and its correlation functions (referred
to as genus expansions). This gives rise to a hierarchy of equations that recover generating functions
for the map counts as coefficients in the genus expansions [EM03]. These functions have the general
form ∞
X
j=0
N(g, j)
j!tj.
In what follows we denote the generating functions for counts of maps of genus g(g-maps) in case
(a) by eg, and in case (b) by zg. The two are intimately related. In particular, the egmay be
determined from the zgby solving a forced Cauchy-Euler equation [EMP08].
There are two approaches in the literature for deriving recursion equations for the map gener-
ating functions from the genus expansions. One of these is based on the matrix resolvent (Green’s
function) for random matrices [ACKM93,Ey16]. This is often referred to as the topological recur-
sion. The other is based on the recurrence matrix resolvent of orthogonal polynomials associated
to the probability measure of the random matrix ensemble [EM03,EMP08,BD13,EW22]. In the
latter approach one is able to bring to bear rigorous analytical methods from the classical theory
of orthogonal polynomials as well as Riemann-Hilbert analysis in order to get detailed estimates
on the validity of the genus expansion and associated recurrence equations. This enables one to
establish that the generating functions of the form displayed above, initially taken to be formal,
are in fact convergent [EM03] and to determine their maximal domains of holomorphy. This is
essential for the derivation of explicit closed rational expressions for the zginitiated in [Er11] and
continued in [ELT22b] that we use. We refer the reader to the last citation for additional details
and will take as our starting point the rational expressions for zgand egthat were developed in
[EMP08,Er11,Er14].
While map enumeration has stemmed from recursion schemes of the type just mentioned, what
we do in this paper is decidedly different. We aim to derive expressions for the map counts N(g, j)
that are of closed form in terms of known classical combinatorial functions. This is what we refer to
as a non-recursive count. A few examples are available in the literature: N2ν,z(0, j) and N2ν,e(0, j)
were derived in [EMP08], N4,e(2, j) was introduced in [BIZ80], and more recently, N4,e(3, j) was
obtained in [BGM21]. To the best of our knowledge, the only other instances of known closed-form
expressions are N3,e(0, j) and N3,e(1, j), for triangulations on a surface of genus 0 or 1 [BD13].
This article establishes general expressions for N2ν,z(g, j) and N2ν,e(g, j) in terms of a linear com-
bination of a specified set of coefficients appearing in the rational forms for zgand eg, weighted
by hypergeometric functions depending on the genus gand the number of vertices j. In addition,
we introduce a dynamic perspective which, in the case of 4-valent maps, leads to a remarkable
analytical simplification that enables us to efficiently express the desired non-recursive counts in
terms of polynomials, powers, and factorials of the number of vertices j, for fixed but arbitrary
genus, as shown in Tables 1and 2below. These tables suggest a general structure for the counts
N4,z(g, j) and N4,e(g, j) of the form
N4,z(g, j) = (−1)g12j 2g−2
Y
k=0
(j−k)!(2j)!
j!P(g)
bg/2c(j)−4jj!P(g)
b(g−1)/2c(j), g ≥1, j ≥0
and
N4,e(g, j) = (−1)g12j−1 2g−3
Y
k=1
(j−k)!(2j)!
j!Q(g)
bg/2c(j)−4j−1j!Q(g)
b(g−1)/2c(j), g ≥2, j ≥1