On the Number of Branched Covers of an Elliptic Curve

Walter Andrés Páez Gaviria

1 Introduction
Let E be an elliptic curve over C, and g ≥ 1, m ≥ 2 integers.
A pair (C, f ) consisting of a smooth complex curve f and a holomorphic map f : C → E is called an
simple m-branched cover if C is connected, every ramification point has index m, and different ramification
points have different images.
(m)
Let Xg,d be the set of isomorphism classes of m-branched covers of E of degree d and genus g, and let

(m)
X 1
Ng,d = .
(m)
|Aut(C, f )|
(C,f )∈Xg,d

Define
(m)
X
Fg(m) (q) = Ng,d q d ,
d≥1

for g ≥ 2. The case g = 1 must be treated separately, since we have to consider the contribution of
constant maps. We define
1 X (m)
F1 (q) = − log q + N1,d q d .
24
d≥1

In this document will present an sketch of the proof the case m = 2 of the next theorem; for the general
case we refer to [3].
(m)
Theorem 1 (Dijkgraaf m = 2 [1], Ochiai m ≥ 2 [3]) For g ≥ 2, Fg (q) is a quasimodular form for the full
modular subgroup SL(2Z. In the case m = 2, Fg (q) has weight 6g − 6.

From now on we will try to omit m = 2.

2 Quasimodular forms
Let’s recall what a quasimodular form is. Let Y (τ ) = 4πIm(τ ). Kaneko and Zagier define an almost
homolomorphic modular form of weight k to be a function on the upper half plane of the form
n
X
fm (τ )Y −m ,
m=0

where each fm is holomorphic and grows polynomially in 1/Y as Y → ∞, and such that F satisfies the usual
weight k indentity for the full modular group SL(2, Z). A quasimodular form of degree k is defined to be the
coefficient f0 (τ ) in a holomorphic modular form of weight k.
It can be proved that the graded algebra of quasimodular forms is C[E2 , E4 , E6 ], with deg(Ei ) = i. Two
(m)
different approaches to this can be found in [2], [4] and [5]. So theorem 1 asserts that Fg (q), for g, m ≥ 2
can be written as polinomials in E2 , E4 and E6 . Some explicit computations can be found in [6].

1
3 Partition functions
One of the steps in the proof of theorem 1 is to allow possibly disconnected covers of E, so let X̂g,d and N̂g,d
be defined as in the introduction, but allowing possibly disconnected smooth curves C. N̂g,d and Ng,d are
finite by proposition 2 in the next section. Now we define partition functions for the connected and possibly
disconnected case, respectively, by
XX Ng,d X Fg (q)
Z(q, λ) = q d λ2g−2 = λ2g−2 ,
(2g − 2)! (2g − 2)!
g≥1 d≥1 g≥1

and
XX N̂g,d
Ẑ(q, λ) = q d λ2g−2 .
(2g − 2)!
g≥1 d≥1

The strategy is to relate this two partitions functions by means of the next proposition.

Proposition 1 Ẑ(q, λ) = exp(Z(q, λ)) − 1.

The proof of it is stricly combinatorial and can be considered as a consequence of the exponential formula
in combinatorics. See, for example, corollary 3.4.1. in [7].

4 Reducing N̂g,d to the trace on an operator

Let us choose b different points P1 , . . . , Pb in E, where b = 2g − 2, that is, by Hurwitz formula, the number
of ramification points of a 2-branched cover of E, and let π1b be the fundamental group of the b-punctured
elliptic curve E − P1 , . . . , Pb . Then

π1b = hγ1 , . . . , γb , α, β | γ1 · · · γb = αβα−1 β −1 i

.
Let ctrans be the conjugation class of the transpositions in S d . Define

Φg,d = {φ ∈ Hom(π1b , Sd ) | φ(γi ) ∈ ctrans for every i}.

Then Sd acts on Φg,d by conjugation.

By the next proposition, computing N̂g,d is equivalent to computing |Φg,d |.

Proposition 2 a. Monodromy induces a bijection X̂g,d ∼

= Φg,d /Sd ;

b. N̂g,d = |Φg,d |/d!.

We want now want to compute |Φg,d |. Since we have a description of π1b by generators and relations, we
get a bijective correspondence between Φg,d and the set

{(τ1 , . . . , τb , τ, σ) ∈ (Sd )b+2 | τ1 · · · τb = τ στ −1 σ −1 , τi ∈ ctrans for every i}.

So we need to count the elements of this set. Fix σ ∈ Sd . Let

Pg,d (σ) = {(τ1 , . . . , τb ) ∈ (Sd )b | τ1 · · · τb σ and σ are conjugate, τi ∈ ctrans for every i}.

Since given a τ such that τ1 · · · τb = τ στ −1 σ −1 we can get all the other τ with such a property by
multiplying it with an element of the conjugation class of σ, c(σ), we have
X d!
|Φg,d | = |Pg,d (σ)|.
|c(σ)|
σ∈Sd

Since Pg,d (τ στ −1 ) = τ Pg,d (σ)τ −1 , |Pg,d | : Sd → N is a class function, therefore

X
|Φg,d | = d!|Pg,d (σ)|,
σ∈Cd

2
 P
and N̂g,d = c∈Cd |Pg,d (c)|.
For now on, let’s fix a complete set of representatives {σ1 , . . . , σm } of the conjugation classes Cd =
{c1 , . . . , cm }, c(σi ) = ci . Observe that m is the number of partitions of d (that we call part(d)) since the
conjugacy class of an element of Sd is determined by the partition induced by its cycle decomposition Consider
the matrix
(Md )ij = |{τ ∈ ctrans | τ σi ∈ cj }|.
An straightforward induction argument, with the base case being the last definition gives us

(Mdk )ij = |{(τ1 , . . . , τk ) ∈ (Sd )k | τ1 , · · · τk σi ∈ cj , τl ∈ ctrans for every l}|.

From this observation we get (Mdb )i,i = |Pg,d (σi )| and therefore we have N̂g,d = T r(Mdb ). We also conclude
that (Mdk )i,i = 0 for odd k. Therefore we have the following expression for Ẑ:
X X T r(M b ) X
Ẑ(q, λ) = d
q d λ2g−2 = (exp(Md · λ))q d .
(2g − 2)!
g≥1 d≥1 d≥1

Pm k
Let αd,i be the eigenvalues of Md . Then N̂g,d = i=1 αd,i . So the last expression gives us the following
proposition.
Ppart(d)
exp(αd,i λ)q d .
P
Proposition 3 .Ẑ(q, λ) = d≥1 i≥1

5 Diagonalizing Md
Recall that the class algebra Hd , which is the center of the groupP algebra C[Sd ] (i.e., functions Sd → C
constant on conjugation classes) has a basis (zc )c∈Cd , where zc = σ∈c σ. Since the inverse of a transposition
is a transposition, we have the following easy proposition.

Proposition 4 The matrix of the operator on Hd defined by multiplication by zctrans is the transpose of Md .

Now, the eigenvalues of this operator follow from Schur orthogonality relations:
For any irreducible character χ of Sd , define
dim(χ) X dim(χ) X
wχ = χ(c−1 )zc = χ(σ −1 )σ.
d! d!
c∈Cd σ∈Cd

Schur orthogonality relations tell us that wχ · wχ0 = δχ,χ0 wχ (this is Kronecker’s delta), and
X |c−1 |χ(c−1 )

zc = = wχ .
χ
dim(χ)

From this identities it follows immediately that (wχ )χ∈Rd is basis of Hd which diagonalizes multiplication
by zctrans . The eigenvector wχ has eigenvalue
d


2 χ(c)
.
dim(χ)

6 Frobenius formula and Haneko-Zagier result

It is known that irreducible representations of Sd are in one-to-one correspondence with partitions of d,
(for example, via the Sprecht representation). The particular construction of this representations will not be
relevant to us, but a fomula of Frobenius will. Let χ be an irreducible representation of Sd and λ = (λ1 , . . . , λl )
its corresponding partition. Define ui = λi − i + 12 and vi defined as for ui , but by considering the transpose
partition of λ.. The Frobenius formula we will use affirms that
d


2 χ(c) 1X 2
= (ui − vi2 ),
dim(χ) 2 i

3
so it computes the eigenvalues found in the last section.
Consider the product
Y u2
Y −v 2
(1 + ζq u e 2 λ ) (1 + ζ −1 q u e 2 λ )
u∈N+ 21 v∈N+ 21

We can see it as a Laurent formal series in ζ. Propositions 3 and 4 gives us

u2 −v 2
Proposition 5 Ẑ(q, λ) = coef. of ζ 0 in + ζq u e λ
+ ζ −1 q u e λ
Q Q
u∈N+ 12 (1
2 ) v∈N+ 21 (1
2 ) − 1.

In [2], Kaneko and Zagier considered the series

Y Y u2
Y −v 2
Θ(q, λ, z) = (1 − q n ) (1 + ζq u e 2 λ ) (1 + ζ −1 q u e 2 λ ).
n≥1 u∈N+ 21 v∈N+ 12

Let Θ0 (q, λ) be the coefficient of ζ 0 in Θ(q, λ, z). Then, since Θ0 (q, λ) = Θ0 (q, −λ), we have an expansion
X
Θ0 (q, λ) = An (q)2n .
n≥0

Kaneko and Zagier prove (theorem 1, [2]) that An (q) is a quasimodular form of weight 6n. This implies that
the coefficient of λ2g−2 in log Θ0 (q, λ) is a modular form of weight 6g − 6, since, by using the taylor series of
the logarithm around −1, we have
X X (−1)m+1 X
log Θ0 (q, λ) = log(1 + An (q)2n ) = ( An (q)2n )m .
m
n≥1 m≥1 n≥1

On the other hand, by propositions 1 and 5, we get

Y X X
log Θ0 (q, λ) = log( (1 − q n )(Ẑ(q) + 1) = log(1 − q n ) + log(Ẑ(q, λ) + 1)) = log(1 − q n ) + Z(q, λ).
n≥1 n≥1 n≥1

Since the coefficient of λ2g−2 in this last sum is the one coming from Z(q, λ), i.e.,
Fg (q)
,
(2g − 2)!
the proof of theorem 1 for m = 2 is concluded.

References
[1] E. Dijkgraaf. Mirror symmetry and elliptic curves. In The Moduli Space of Curves (Texel Island,
1994), vol 129 of Prog. Math. Birkhäuser Boston, Boston, MA, 1995.
[2] M. Kaneko, Don Zagier A generalized Jacobi theta function and quasimodular forms. In The Moduli
Space of Curves (Texel Island, 1994), vol 129 of Prog. Math. Birkhäuser Boston, Boston, MA, 1995.
[3] H. Ochiai. Counting functions for branched covers and quasimodular forms. RIMS Kokyuroko 1218.
!53-167 (2001).
[4] H. Movasati. On differential modular forms and some analytic relations between Eisenstein series.
Ramanujan J., 17(1):53-76, 2008.
[5] H. Movasati. Quasi modular forms attached to ellipic curves, I. Annales Mathematique Blaise Pascal,
19:307-377, 2012.
[6] M. Roth, N. Yui. Mirror Symmetry for Elliptic Curves: The A-Model (Fermionic) Counting. In
Motives, Quantum Field Theory, and Pseudodifferential Operators, vol 12 of Clay Mathematics Pro-
ceedings. AMS. 2010.
[7] H. S. Wilf. generatingfunctionology.