Compositio Math.

140 (2004) 552–566

DOI: 10.1112/S0010437X03000721

The arithmetic of the values of modular functions and

the divisors of modular forms

Jan H. Bruinier, Winfried Kohnen and Ken Ono

Abstract
We investigate the arithmetic and combinatorial significance of the values of the polyno-
mials jn (x) defined by the q-expansion
∞

E4 (z)2 E6 (z) 1
jn (x)q n := · .
∆(z) j(z) − x
n=0
They allow us to provide an explicit description of the action of the Ramanujan Theta-
operator on modular forms. There are a substantial number of consequences for this result.
We obtain recursive formulas for coefficients of modular forms, formulas for the infinite
product exponents of modular forms, and new p-adic class number formulas.

1. Introduction and statement of results

Let j(z) = q −1 + 744 + 196 884q + · · · denote the usual elliptic modular function
√ on SL2 (Z) (q :=
e2πiz throughout). We shall refer to a complex number τ of the form τ = (−b + b2 − 4ac)/2a with
a, b, c ∈ Z, gcd(a, b, c) = 1 and b2 − 4ac < 0 as a Heegner point, and we denote its discriminant by
the integer dτ := b2 − 4ac. The values of j at such points are known as singular moduli, and they
play a substantial role in classical and modern number theory. For example, the theory of complex
√ with discriminant dτ , then j(τ ) is an algebraic
multiplication implies that if τ is a Heegner point
integer which generates a ring class field of Q( dτ ).
Singular moduli also play an important role in Borcherds’ [Bor95a, Bor95b] recent work on the
infinite product expansions of certain modular forms. A meromorphic modular form f on SL2 (Z),
by definition, has a Heegner divisor if its zeros and poles are supported at the cusp at infinity
and Heegner points. In particular, Borcherds obtains an elegant description of the infinite product
expansion of those meromorphic modular forms on SL2 (Z) with a Heegner divisor.
Here we consider the values of a specific sequence of elliptic modular functions jn , where
j1 = j − 744. In an important recent paper [Zag02], Zagier expressed the traces of the values
of jn at Heegner points in terms of Fourier coefficients of half integral weight modular forms.
Here we consider the more general case of the sums of the values of jn over divisors of meromorphic
modular forms. We show that the ‘traces’ of these values (see Theorem 1) dictate the properties
of modular forms on SL2 (Z). This result is obtained using a jn -weighted version of the proof of the
classical valence formula for modular forms on SL2 (Z).
Theorem 1 provides a very useful link relating the values of j to the arithmetic of the Fourier
coefficients of modular forms. Naturally, one then expects a wide variety of consequences. Here we

Received 25 February 2002, accepted in final form 14 May 2002.

2000 Mathematics Subject Classification 11F03, 11F11, 11F33.
Keywords: Borcherds products, singular moduli, modular forms and functions.
The first and third authors thank the Number Theory Foundation for its generous support, and the third author
is grateful for the support of an Alfred P. Sloan Fellowship, a David and Lucile Packard Fellowship, an H. I. Romnes
Fellowship, a John Guggenheim Fellowship and a grant from the National Science Foundation.
This journal is
c Foundation Compositio Mathematica 2004.

Downloaded from https://www.cambridge.org/core. IP address: 201.124.245.18, on 17 Jan 2021 at 22:41:35, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1112/S0010437X03000721
 Modular functions and divisors of modular forms

begin by considering such consequences in connection with the algebraicity of j-values, congruence
properties and bounds for class numbers of imaginary quadratic fields, infinite product expansions
of modular forms, and recurrence relations for Fourier coefficients. For example, we show that there
are universal recursion formulas for the Fourier coefficients of every modular form on SL2 (Z) (see
Theorem 3). We also obtain formulas for the exponents in the infinite product expansion of every
modular form on SL2 (Z) (see Theorem 5), and we obtain new p-adic formulas for class numbers as
traces of j-values (see Theorem 9).
Our investigation begins with a careful analysis of Ramanujan’s Theta-operator, the differential
operator defined by

∞ 
∞
n
Θ a(n)q := na(n)q n . (1.1)
n=h n=h
We refer to Θ as Ramanujan’s operator since he first observed [Ram16] that
Θ(E4 ) = (E4 E2 − E6 )/3 and Θ(E6 ) = (E6 E2 − E8 )/2, (1.2)
where Ek , for every even integer k
2, is the standard Eisenstein series
∞
2k

Ek (z) := 1 − σk−1 (n)q n . (1.3)
Bk
n=1

Here Bk denotes the usual kth Bernoulli number and σk−1 (n) := d|n dk−1 . If k > 2, then Ek is
a weight k modular form on SL2 (Z). As usual, let ∆ := (E43 − E62 )/1728, the unique normalized
weight 12 cusp form on SL2 (Z).
Although the Eisenstein series
∞


E2 (z) = 1 − 24 σ1 (n)q n (1.4)
n=1

is not a modular form, it plays an important role. If f (z) = ∞ n
n=h a(n)q is a weight k meromorphic
modular form on SL2 (Z), then
Θ(f ) = (f˜ + kf E2 )/12, (1.5)
where f˜ is a meromorphic modular form of weight k + 2 on SL2 (Z). (Note that the formulas in (1.2)
imply (1.5).) Because of this fact, the Θ-operator is fundamental in the theory of p-adic modular
forms and modular forms modulo p. For instance, if f is a p-adic modular form of weight k, then
since E2 is a p-adic modular form of weight 2, Θ(f ) is a p-adic modular form of weight k + 2 [Ser73,
Theorem 5].
Although Θ is simple to define, its arithmetic nature is much deeper and is dictated by the
˜
f appearing in (1.5). We derive an explicit formula for Θ(f ) in terms of a natural sequence of
modular functions jm (z). Let j0 (z) := 1, and for every positive integer m let jm (z) be the
unique modular function which is holomorphic on H, the upper half of the complex plane, whose
Fourier expansion is of the form
∞


jm (z) = q −m + cm (n)q n . (1.6)
n=1

Note that if m is a positive integer, then jm (z) = j1 (z) | T0 (m), where T0 (m) is the usual normalized
mth weight zero Hecke operator. The first few jm are
j0 (z) = 1,
j1 (z) = j(z) − 744 = q −1 + 196 884q + · · · ,
553
Downloaded from https://www.cambridge.org/core. IP address: 201.124.245.18, on 17 Jan 2021 at 22:41:35, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1112/S0010437X03000721
 J. H. Bruinier, W. Kohnen and K. Ono

j2 (z) = j(z)2 − 1488j(z) + 159 768 = q −2 + 42 987 520q + · · · ,

j3 (z) = j(z)3 − 2232j(z)2 + 1069 956j(z) − 36 866 976 = q −3 + 2 592 899 910q + · · · .
Each jm is a monic degree m polynomial in j with integer coefficients.
Let F denote the usual fundamental domain of the √ action of SL2 (Z) on√H. By assumption, F
does not include the cusp at ∞. Throughout, let i = −1 and let ω := (1 + −3)/2. If τ ∈ F, then
define eτ by


1/2 if τ = i,
eτ := 1/3 if τ = ω, (1.7)


1 otherwise.
For every point τ ∈ H, Asai, Kaneko, and Ninomiya [AKN97, Theorem 3] proved that
∞

E42 (z)E6 (z) 1
Hτ (z) := jn (τ )q n = · . (1.8)
∆(z) j(z) − j(τ )
n=0

For τ = i and ω, we have the following beautiful formulas:

∞
E6
Hω = = jn (ω)q n , (1.9)
E4
n=0

∞
E8
Hi = = jn (i)q n . (1.10)
E6 n=0

In particular, for every τ it turns out that Hτ is a weight 2 meromorphic modular form. The utility
of (1.8) was already known; for example, it can be used to prove that

∞ 
pn
j(τ ) − j(z) = p−1 exp − jn (z) · ,
n
n=1

where p = e2πiτ . This identity is equivalent to the famous denominator formula for the monster Lie
algebra

j(τ ) − j(z) = p−1 (1 − pm q n )c(mn) ,
m>0 and n∈Z

where the exponents c(n) are defined as the coefficients of j1 = ∞ n
n=−1 c(n)q .
Here we obtain a new proof of (1.8) and consider many of its number theoretic consequences.

Theorem 1. If f = ∞ n
n=h af (n)q is a non-zero weight k meromorphic modular form on SL2 (Z)
for which af (h) = 1, then
kE2 f
Θ(f ) = − f fΘ ,
12
where fΘ is defined by


fΘ := eτ ordτ (f )Hτ (z).
τ ∈F

Theorem 1 easily reveals some algebraic information about the jn evaluated at the finite points
of the divisor of any meromorphic modular form. A celebrated result of Schneider asserts that, if τ is
an algebraic number of degree >2, then j(τ ) is transcendental. Under certain conditions, we observe
that the values of j at the points in the divisor of an algebraic modular form are algebraic. Although
there are more direct ways of establishing this result, it follows rather nicely from Theorem 1.

554
Downloaded from https://www.cambridge.org/core. IP address: 201.124.245.18, on 17 Jan 2021 at 22:41:35, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1112/S0010437X03000721
 Modular functions and divisors of modular forms
∞ n
Corollary 2. Let f = n=h af (n)q be a meromorphic modular form on SL2 (Z) for which
af (h) = 1. If τ0 ∈ F is a point for which ordτ0 (f ) = 0 and the coefficients of f are in a number field
K, then j(τ0 ) is algebraic.

Using Borcherds’ work on infinite product expansions of modular forms, this corollary generalizes
the classical fact that j(τ ) is algebraic whenever τ is a Heegner point.
We consider the arithmetic of the Fourier coefficients of meromorphic modular forms. If k
4
is aneven integer and p is prime, then let Tk (p) be the usual Hecke operator. In particular, if
f= ∞ n=0 af (n)q ∈ Mk (1), the space of holomorphic modular forms of weight k on SL2 (Z), then
n

∞


f | Tk (p) := (af (np) + pk−1 af (n/p))q n . (1.11)
n=0
If f ∈ Sk (1), the space of weight k cusp forms on SL2 (Z), then f | Tk (p) ∈ Sk (1). If Tk (p, x) denotes
the characteristic polynomial of Tk (p) on Sk , then it is well known that Tk (p, x) ∈ Z[x]. There is
wide speculation that Tk (p, x) is irreducible for every prime p, and has the additional property that
the Galois group of its splitting field is the symmetric group Sdk , where dk denotes the dimension of
Sk (1). Here we express these polynomials in terms of the values of jn at the zeros of the eigenforms
in Sk (1) (see (1.12)). We begin with the following universal recursion relation for certain modular
forms.
Theorem 3. For every n
2 define Fn (x1 , . . . , xn−1 ) ∈ Q[x1 , . . . , xn−1 ] by
2x1 σ1 (n − 1)

Fn (x1 , . . . , xn−1 ) := − + (−1)m1 +···+mn−2
n−1
m1 ,...,mn−2
0,
m1 +2m2 +···+(n−2)mn−2 =n−1
(m1 + · · · + mn−2 − 1)! m1 mn−2
· · x2 · · · xn−1 .
m1 ! · · · mn−2 !

If f = q + ∞ n
n=2 af (n)q is a weight k meromorphic modular form on SL2 (Z), then for every integer
n
2 we have
1

af (n) = Fn (k, af (2), . . . , af (n − 1)) − eτ ordτ (f ) · jn−1 (τ ).
n−1
τ ∈F

It is simple to modify Theorem 3 for any modular form with leading coefficient 1.
The first few polynomials Fn are
F2 (x1 ) := −2x1 ,
x22
F3 (x1 , x2 ) := −3x1 +
,
2
8x1 x32
F4 (x1 , x2 , x3 ) := − − + x2 x3 ,
3 3
7x1 x4 x2
F5 (x1 , x2 , x3 , x4 ) := − − x22 x3 + x2 x4 + 2 + 3 .
2 4 2
By arguing inductively with Theorem 3, it turns out that every Fourier coefficient af (n) is a
Q-rational expression in the weight k and the values of j at the points in the divisor of f .

Remark. Theorem 3 includes a simple recursion for the coefficients of ∆ = ∞ n
n=1 τ (n)q . Since ∆
has no zeros in F, for every n
2 we find that
τ (n) = Fn (12, τ (2), . . . , τ (n − 1)).
As a special case of Theorem 3, we obtain the following strange formula.
555
Downloaded from https://www.cambridge.org/core. IP address: 201.124.245.18, on 17 Jan 2021 at 22:41:35, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1112/S0010437X03000721
 J. H. Bruinier, W. Kohnen and K. Ono
∞ n
Corollary 4. If f = q + n=2 af (n)q is a meromorphic modular form of weight k on SL2 (Z),
then

af (2) = 60k − 744 − eτ ordτ (f ) · j(τ ).
τ ∈F

As an immediate consequence of Theorem 3, we obtain an expression for Tk (p, x). If dk is the

dimension of Sk (1), then for 1  s  dk let
∞


fs = q + afs (n)q n
n=2
be the normalized Hecke eigenforms in Sk (1). For every prime p, we have
dk  
1

Tk (p, x) = x − Fp (k, afs (2), . . . , afs (p − 1)) + eτ ordτ (fs ) · jp−1 (τ ) . (1.12)
p−1
s=1 τ ∈F

These results are closely related to Borcherds’ recent work on the infinite product expansions
of modular forms. Borcherds [Bor95a, Bor95b] provided a striking description for the exponents in
the infinite product expansion for those modular forms with a Heegner divisor. For example, if the
integers c(n) are defined by
∞

∞

−240
E4 (z) = 1 + 240 σ3 (n)q = (1 − q)
n
(1 − q )
2 26 760
··· = (1 − q n )c(n) ,
n=1 n=1
then Borcherds’ theorem implies that there is a weight 1/2 meromorphic modular form


G(z) = b(n)q n = q −3 + 4 − 240q + 26 760q 4 + · · · − 4 096 240q 9 + · · ·
n
−3
on Γ0 (4) with the property that c(n) = b(n2 ) for every positive integer n. We obtain an arithmetic
formula for the exponents of the infinite product expansion of every meromorphic modular form
on SL2 (Z).

Theorem 5. Suppose that f = ∞ n
n=h af (n)q is a weight k meromorphic modular form on SL2 (Z)
for which af (h) = 1, and let c(n) denote the complex numbers for which
∞
f =q h
(1 − q n )c(n) .
n=1
If n is a positive integer, then



c(d)d = 2kσ1 (n) + eτ ordτ (f ) · jn (τ ).
d|n τ ∈F

In an important paper [GZ85], Gross and Zagier described the divisibility properties of differences
of singular moduli. More recently [Zag02], Zagier described the arithmetic of the traces of singular
moduli in terms of the Fourier coefficients of modular forms of half integral weight. Since the modular
functions jn play an important role, we consider their divisibility and congruence properties. We
consider the arithmetic of the values of jn as we vary n. First we obtain the following theorem for
the special values at τ = ω and τ = i.
Theorem 6. If τ = ω, then let M be a positive integer which is not divisible by a prime p ≡ 1
(mod 3). If τ = i, then suppose that M is a positive integer which is not divisible by a prime
p ≡ 1 (mod 4). Then there is a positive real number α(M ) for which
 
X
#{1  n  X : jn (τ ) ≡ 0 (mod M )} = O .
(log X)α(M )
In particular, for almost all n we have jn (τ ) ≡ 0 (mod M ).

556
Downloaded from https://www.cambridge.org/core. IP address: 201.124.245.18, on 17 Jan 2021 at 22:41:35, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1112/S0010437X03000721
 Modular functions and divisors of modular forms

In addition to results of this type, there are examples of explicit congruences. For example,
congruences with modulus 2k relating such values to Borcherds exponents follow immediately from
Theorem 5. We highlight two further types of congruence properties.
Theorem 7. If k
4 is even, then for every positive integer n we have

 
eτ ordτ (Ek ) · jn (τ ) ≡ −2kσ1 (n) mod 4 p .
τ ∈F p−1|k
5p prime

Theorem 8. Let f = ∞ n
n=h af (n)q be a weight k meromorphic modular form on SL2 (Z) whose
coefficients are in OK , the ring of algebraic integers in a number field K. Suppose that af (h) = 1
and that f has a Heegner divisor whose Heegner points in F are τ1 , τ2 , . . . , τt . Furthermore, suppose
that p ∈ {2, 3, 5, 7} has the property that for all 1  s  t we have



3 (mod 8)

if p = 2,
1 (mod 3) if p = 3,
|dτs | ≡

2, 3 (mod 5) if p = 5,


1, 2, 4 (mod 7) if p = 7.

If ν is a positive integer, then there is a positive real number α(p, ν) for which


t   
X
# 1nX : eτc ordτc (f ) · jn (τc ) ≡ 0 (mod p ) = O
ν
.
c=1
(log X)α(p,ν)

In particular, for almost all n we have tc=1 eτc ordτc ·jn (τc ) ≡ 0 (mod pν ).

The p-adic properties of the values of the jn are closely related to the arithmetic of class numbers
of imaginary quadratic fields. Let H(−D) be the Hurwitz class number for the discriminant −D.
Theorem 9. Suppose that −D < −4 is a fundamental discriminant of an imaginary quadratic field,
and let τ be any Heegner point of discriminant −D. If K = Q(j(τ )), then the following are true:
1) If D ≡ 3 (mod 8), then as 2-adic numbers we have
1
H(−D) = lim TrK/Q (j2n (τ )).
24 n→+∞

2) If D ≡ 1 (mod 3), then as 3-adic numbers we have

1
H(−D) = lim TrK/Q (j3n (τ )).
12 n→+∞

3) If D ≡ 2, 3 (mod 5), then as 5-adic numbers we have

1
H(−D) = lim TrK/Q (j5n (τ )).
6 n→+∞

4) If D ≡ 1, 2, 4 (mod 7), then as 7-adic numbers we have

1
H(−D) = lim TrK/Q (j7n (τ )).
4 n→+∞

Remark. Analogs of Theorem 9 hold for −D = −3 (respectively −D = −4). Subject to the same
congruence conditions on D, these results simply require replacing jpn (τ ) by jpn (ω)/3 (respectively
jpn (i)/2). Moreover, simple analogs hold for every −D, not just those which are fundamental.
More generally, there are analogs of Theorems 8 and 9 for primes p
11, but these results are more
complicated to state.

If D ≡ 3 (mod 8), it turns out that the 2-adic behavior of these traces, for all jn , are also
controlled by the class number H(−D). Using the fact that the Hecke algebra for holomorphic
557
Downloaded from https://www.cambridge.org/core. IP address: 201.124.245.18, on 17 Jan 2021 at 22:41:35, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1112/S0010437X03000721
 J. H. Bruinier, W. Kohnen and K. Ono

modular forms is locally nilpotent at 2, we obtain the following 2-divisibility results. Let ω(n)
denote the number of distinct prime factors of n.
Theorem 10. Suppose that −3 = −D ≡ 5 (mod 8) is a fundamental discriminant of an imaginary
quadratic field, and suppose that τ is a Heegner point of discriminant −D. If K = Q(j(τ )) and
s
4, then
TrK/Q (jn (τ )) ≡ 0 (mod 2s )
for every positive square-free integer n for which
ω(n) > 2s−4 H(−D).
Theorem 10 yields theoretical lower bounds for H(−D). To state these results, for D ≡ 0, 3
(mod 4), let
∞

F (D; z) = q −H(−D) (1 − q n )cD (n) (1.13)
n=1
be the unique weight zero modular function on SL2 (Z), with leading coefficient 1, whose divisor con-
sists of a pole of order H(−D) at z = ∞ and a simple zero at each Heegner point with discriminant
−D. These functions have integer coefficients. Consider the formal power series

∞
∞

Θ(F (D; z))
:= −H(−D) − A(D; n)q n = −H(−D) − cD (d) dq n . (1.14)
F (D; z)
n=0 n=1 d|n

Corollary 11. Suppose that −3 = −D ≡ 5 (mod 8) is a fundamental discriminant of an imagi-

nary quadratic field. If s
4 and there is an odd square-free integer n for which ord2 (A(D; n)) < s,
then
ω(n) 1
H(−D) > s−4 − .
2 s 3 · 2s−3
It will be extremely interesting to see whether a detailed study of the Hecke algebra modulo
powers of 2, perhaps combined with further 2-adic arguments, can be used to transform Corollary
11 into a lower bound like the celebrated bound due to Goldfeld, Gross and Zagier.
In § 2 we prove Theorems 1, 3, and 5, and Corollaries 2 and 4. In § 3 we prove Theorems 6, 7,
8, and 9. There we consider the p-adic behavior of the Θ-operator under certain conditions. In § 4
we prove Theorem 10 and Corollary 11 using an analysis of the behavior of the Hecke algebra on
modular forms modulo 2.

2. Proof of Theorems 1, 3, and 5 and Corollaries 2 and 4

For convenience, we begin by proving Theorem 5 on the infinite product expansion of generic
modular forms. Before we prove Theorem 5, we call attention to earlier work of Eholzer and Skoruppa
[ES96] which also considers product expansions of modular forms.

Proposition 2.1. Let f = ∞ n
n=h af (n)q be a meromorphic function in a neighborhood of q = 0,
and suppose that af (h) = 1. Then there are uniquely determined complex numbers c(n) such that
∞

f =q h
(1 − q n )c(n) ,
n=1
where the product converges in a small neighborhood of q = 0. Moreover, the following identity is
true:

∞

Θ(f )
=h− c(d) dq n .
f n=1 d|n

558
Downloaded from https://www.cambridge.org/core. IP address: 201.124.245.18, on 17 Jan 2021 at 22:41:35, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1112/S0010437X03000721
 Modular functions and divisors of modular forms

Proof. As usual, we understand that complex powers are defined by the principal branch of the
complex logarithm. If F (q) := f (z), then the function qF  (q)/F (q) is holomorphic at q = 0. Write
its Taylor expansion as


qF  (q)/F (q) = h − α(n)q n (|q| < ), (2.1)
n
1
and for n
1 let
1

c(n) := α(d)µ(n/d),
n
d|n
where µ denotes the Möbius function. This implies that


α(n) = c(d)d. (2.2)
d|n

Obviously, the numbers c(n) are uniquely determined by f .

For fixed q0 with |q0 | <  we have α(n) = O(|q0 |−n ) for all n, and this easily implies that the
double series


c(n)nq mn
m,n
1
is absolutely convergent in |q| < |q0 |, hence in |q| < .
In the following, suppose that |q| < . From the above we see that
d F  (q) h
log(F (q)q −h ) = −
dq F (q) q

 
d
q mn
=− c(n)
dq m
n
1 m
1
 
d

= c(n) log(1 − q n ) ,
dq
n
1

the interchange of differentiation and summation being justified because of local uniform convergence
as can easily be seen in a similar way as above.
We thus obtain


log(F (q)q −h ) = c(n) log(1 − q n ).
n
1

The values c(n) log(1 − and log(1 − qn) q n )c(n)

differ by integer multiples of 2πi. Since c(n)
log(1 − q ) → 0 (n → ∞) the same is true for log(1 − q n )c(n) . Hence we see that there is an
n

integer N such that

log(F (q)q −h ) = log(1 − q n )c(n) + 2πiN.
n
1
Taking the exponential on both sides proves our claim.
Proof of Theorem 5. Let
F := {z ∈ H : |z|
1, |Re(z)|  12 }
be the standard fundamental domain for the action of SL2 (Z) on H. We cut off F by a horizontal
line L := {iC − t : − 12  t  12 } where C > 0 is chosen so large that all poles and zeros of f , apart
from those at the cusp at infinity, are contained in {z ∈ H : Im(z) < C}.
For simplicity, suppose that f has no zero or pole on the boundary ∂F except possibly i or ω
(if not, one has to modify the arguments in the same way as in the classical proof of the ‘k/12-
identity’).
559
Downloaded from https://www.cambridge.org/core. IP address: 201.124.245.18, on 17 Jan 2021 at 22:41:35, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1112/S0010437X03000721
 J. H. Bruinier, W. Kohnen and K. Ono

We let γ be the closed path with positive orientation consisting of L and γ1 where γ1 is the part
of ∂F below L modified in the usual way: in a small neighborhood U of ω (respectively i; respectively
−ρ) we replace U ∩ ∂F by F ∩ Cω (respectively F ∩ Ci ; respectively F ∩ C−ω ) where Cω (respectively
Ci ; respectively C−ω ) are small circles with radius r around ω (respectively i; respectively −ω).
We integrate
1 f  (z)
jn (z)
2πi f (z)
along γ. By the residue theorem, taking into account that jn (z) is holomorphic on H, this integral
is equal to


ordτ (f )jn (τ ).
τ ∈F−{ω,i}

On the other hand, the integral can be evaluated separately along the different pieces of γ, in a
well-known way. If we let r tend to zero, we then find that

1 1
ordτ (f )jn (τ ) = − ordω (f )jn (ω) − ordi (f )jn (i)
3 2
τ ∈F−{ω,i}
 
1 F  (q) k jn (z)
+ Jn (q) dq − dz. (2.3)
2πi ρ F (q) 2πi σ z
Here F (q) = f (z) as before and Jn (q) := jn (z). Furthermore, ρ is a small circle around q = 0 with
negative orientation and not containing any pole or zero of F (q) except possibly 0, and σ is the
part of the unit circle in the upper half-plane that connects ω and i, with positive orientation.
By Proposition 2.1, for |q| <  we see that


∞
qF  (q) Θ(f )
= =h− c(d) dq n ,
F (q) f
n=1 d|n

where h is the order of F at q = 0. Hence recalling that Jn (q) = q −n + O(q) we find that


1 F  (q)
Jn (q) dq = c(d)d. (2.4)
2πi ρ F (q)
d|n

We cannot directly evaluate the last integral on the right-hand side of (2.3). Instead we proceed
as follows. Formula (2.3) in particular is valid for the function f = ∆ of weight 12. In this case we
have


c(d)d = 24σ1 (m) (m
1),
d|m

by definition. Since ∆ has no zeros on H, we obtain from (2.3) that


1 jn (z)
dz = 2σ1 (n). (2.5)
2πi σ z
Inserting (2.4) and (2.5) into (2.3), we deduce the theorem.

Proof of Theorem 1. We begin by proving that if

Θ(f ) kE2
= − fΘ , (2.6)
f 12
then fΘ has the claimed form. If n is a positive integer, then Proposition 2.1 and Theorem 5 imply
560
Downloaded from https://www.cambridge.org/core. IP address: 201.124.245.18, on 17 Jan 2021 at 22:41:35, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1112/S0010437X03000721
 Modular functions and divisors of modular forms

that the coefficient of q n in Θ(f )/f is −2kσ1 (n) − τ ∈F eτ ordτ (f ) · jn (τ ). Since the E2 is given by
∞


E2 = 1 − 24 σ1 (n)q n ,
n=1

(2.6) verifies the truth of Theorem 1 for every coefficient with the exception of the constant term.
 term in Θ(f )/f is h = ord∞ (f ). However, the constant term of kE2 /12 − fΘ is
The constant
(k/12) − τ ∈F eτ ordτ (f ), which equals h by the classical ‘k/12’ valence formula.

Proof of Corollary 2. We begin by fixing notation. Let τ1 , τ2 , . . . , τt ∈ F be the numbers for which
ordτ (f ) = 0. If n is a positive integer, then the coefficient of q n in kE2 /12 is the integer −2kσ1 (n).
Therefore by Theorem 1, if the Fourier coefficients of f are in a field K, then the coefficients of fΘ
and 1/f belong to K. Hence if n is a positive integer, then


t

t
jn (τs ) = Gn (j(τs )) ∈ K, (2.7)
s=1 s=1

where Gn ∈ Z[x] is a monic polynomial of degree n. Since j1 = j − 744, for every positive integer n
we have

t
j(τs )n ∈ K.
s=1
Therefore, by solving for the elementary symmetric functions in j(τ1 ), . . . , j(τt ), we find that

t
(x − j(τs )) ∈ K[x].
s=1
This proves the corollary.

Proof of Theorem 3. By Theorem 1, we have that

∞

Θ(f ) kE2
eτ ordτ (f ) jn (τ )q n = − + .
f 12
τ ∈F n=0

If n
2, then Theorem 5 gives

eτ ordτ (f )jn−1 (τ ) = c(d)d − 2kσ1 (n − 1),
τ ∈F d|n−1

where
∞

f =q (1 − q n )c(n) .
n=1

Therefore, to prove the theorem it suffices to obtain a closed formula for b(n) := d|n c(d)d in
terms of af (n). In particular, it suffices to show that, if n
1, then

(m1 + · · · + mn − 1)!
b(n) = n (−1)m1 +···+mn af (2)m1 · · · af (n + 1)mn . (2.8)
m1 ! · · · mn !
m1 ,...,mn
0,
m1 +2m2 +···+nmn =n

To prove (2.8), one observes that

0 = b(n) + b(n − 1)af (2) + b(n − 2)af (3) + · · · + b(1)af (n) + naf (n + 1),
and uses the well-known fact that
0 = sn − sn−1 σ1 + sn−2 σ2 − · · · + (−1)n−1 s1 σn−1 + (−1)n nσn .
561
Downloaded from https://www.cambridge.org/core. IP address: 201.124.245.18, on 17 Jan 2021 at 22:41:35, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1112/S0010437X03000721
 J. H. Bruinier, W. Kohnen and K. Ono

Here the σi are the elementary symmetric functions in X1 , . . . , Xn and the si are the power functions
in these variables (i.e. si := X1i + · · · + Xni ). One now obtains (2.8) by evaluating these identities at
(X1 , . . . , Xn ) = (λ(1, n), . . . , λ(n, n)) where the λ(j, n) are the roots of the polynomial
X n + af (2)X n−1 + af (3)X n−2 + · · · + af (n + 1).
One requires the fact that

(m1 + m2 + · · · + mn − 1)! m1
si = i (−1)m2 +m4 +··· σ1 · · · σnmn .
m1 !m2 ! · · · mn !
m1 ,...,mn
0,
m1 +2m2 +···+nmn =i

Proof of Corollary 4. Since j1 (z) = j(z) − 744 and τ ∈F eτ ordτ (f ) = (k/12) − 1, this result is the
n = 2 case of Theorem 3.

3. Proofs of Theorems 6, 7, 8, and 9

In this section we prove Theorems 6, 7, 8, and 9 using theorems of Serre on p-adic modular forms
and the divisibility of the Fourier coefficients of modular forms modulo M (see [Ser73, Ser76]).

Proof of Theorem 6. By (1.9) and (1.10), it suffices to prove that the coefficients of the Fourier
series

∞
E6
Hω = = jn (ω)q n = 1 − 744q + 159 768q 2 − 36 866 976q 3 + · · · , (3.1)
E4 n=0

∞
E8
Hi = = jn (i)q n = 1 + 984q + 574 488q 2 + 307 081 056q 3 + · · · (3.2)
E6
n=0

satisfy the claim.

Since z = i (respectively z = ω) is fixed by the modular transformation Sz = −1/z (respectively
Az = −(z + 1)/z), the definition of a modular form implies that, if k
4 is even, then
k ≡ 2 (mod 4) =⇒ Ek (i) = 0,
k ≡ 2, 4 (mod 6) =⇒ Ek (ω) = 0.
If p
5 is prime, then these observations together with the von Staudt–Clausen Theorem [IR90,
p. 233] and (1.3) imply that, if p ≡ 1 (mod 4), then there is an Eisenstein series Ei,p for which
Ei,p (i) = 0 and Ei,p ≡ 1 (mod 24p), (3.3)
and if p ≡ 1 (mod 3), then there is an Eisenstein series Eω,p for which
Eω,p (ω) = 0 and Eω,p ≡ 1 (mod 24p). (3.4)
s
Now observe that if H ≡ 1 (mod ), where  is prime, then H  ≡ 1 (mod s+1 ). If p1 ≡ 1
(mod 4) is prime, then for every positive integer s we have that
E8 E8 ps1
≡ ·E (mod ps+1
1 ). (3.5)
E6 E6 i,p1
Similarly, if p2 ≡ 1 (mod 3) is prime, then
E6 E6 ps2
≡ · Eω,p2 (mod ps+1
2 ). (3.6)
E4 E4
Since E4 (ω) = 0 (respectively E6 (i) = 0) and E4 (respectively E6 ) has no other zeros in F, (3.3) and
(3.5) (respectively (3.4) and (3.6)) illustrate that the relevant forms are the reduction modulo ps+1
i

562
Downloaded from https://www.cambridge.org/core. IP address: 201.124.245.18, on 17 Jan 2021 at 22:41:35, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1112/S0010437X03000721
 Modular functions and divisors of modular forms

of holomorphic integer weight modular forms on SL2 (Z). There are obvious analogous constructions
for both forms modulo powers of 2 and 3. The theorem now follows from a well-known theorem
of Serre which asserts that almost all the coefficients of a modular form with algebraic integer
coefficients are multiples of any given integer M [Ser76, Theorem 4.7].

Proof of Theorem 7. By (1.3) and the von Staudt–Clausen theorem, if k
4 is even, then
 
Ek ≡ 1 mod 4 p .
p−1|k
p prime

This observation and Theorem 1 imply that

 
Θ(Ek ) kE2
0≡ = − (Ek )Θ mod 4 p .
Ek 12
p−1|k
p prime

The theorem follows from (1.4).

Proof of Theorem 8. By [BO03, Corollary 3], Θ(f )/f is a p-adic modular form of weight 2. Since
the Eisenstein series E2 is also a p-adic modular form of weight 2 [Ser73], we find that

∞


Θ(f ) kE2
fΘ = − + = eτ ordτ (f ) jn (τ )q n
f 12
τ inf F n=0

is a p-adic modular form of weight 2. Therefore, fΘ (mod pν ) is the reduction modulo pν of

some holomorphic integer weight modular form on SL2 (Z). The theorem now follows from [Ser76,
Theorem 4.7].

Proof of Theorem 9. If 0 < D ≡ 0, 3 (mod 4), then there is a unique meromorphic modular form
of weight 1/2 on Γ0 (4) that is holomorphic on H whose Fourier series has the form [Bor95a,
Lemma 14.2]
∞


f (D; z) = q −D + c(D; n)q n , (3.7)
n=1
where c(D; n) = 0 for every n ≡ 2, 3 (mod 4). Borcherds’ theory [Bor95a, Bor95b] implies that
∞
2)
F (D; z) = q −H(−D) (1 − q n )c(D;n (3.8)
n=1

is a weight zero modular function on SL2 (Z) whose divisor consists of a pole of order H(−D) at
z = ∞ and a simple zero at each Heegner point with discriminant −D. For each D we consider the
following formal power series (also defined in (1.14)):
∞

∞



F(D; q) = −H(−D) − A(D; n)q n := −H(−D) − c(D; d2 ) dq n . (3.9)
n=0 n=1 d|n

If D and p satisfy the hypotheses of the theorem, then [BO03, Corollary 3] implies that F(D; q)
is a p-adic modular form of weight 2. Serre proved [Ser73, Theorem 7], for certain p-adic modular
forms, that the constant term of the Fourier expansion is essentially the p-adic limit of its Fourier
563
Downloaded from https://www.cambridge.org/core. IP address: 201.124.245.18, on 17 Jan 2021 at 22:41:35, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1112/S0010437X03000721
 J. H. Bruinier, W. Kohnen and K. Ono

coefficients at exponents which are pth powers. In these cases we obtain



24 limn→+∞ A(D; 2 ) if D ≡ 3 (mod 8),
1 n



 1 lim
n→+∞ A(D; 3 ) if D ≡ 1 (mod 3),
n
H(−D) = 12 (3.10)
 16 limn→+∞ A(D; 5n ) if D ≡ 2, 3 (mod 5),



 1 lim n if D ≡ 1, 2, 4 (mod 7).
4 n→+∞ A(D; 7 )

Since F (D; z) has weight zero, for every positive integer n Theorem 5 implies
A(D; pn ) = jpn (τ1 ) + · · · + jpn (τH(−D) ),
where τ1 , . . . , τH(−D) ∈ F are the Heegner points of discriminant −D. Since the j(τi ) are conjugates
over Q, the theorem follows from (3.10) and the fact that each jn is an integral polynomial in j.

4. Proofs of Theorem 10 and Corollary 11

We adopt the notation from the proof of Theorem 9. We begin by recalling the following theorem
which is proved in [BO03, Corollary 3].
Theorem 4.1. If 0 < D ≡ 3 (mod 8), then F(D; q) is a weight two 2-adic modular form.
Using the local nilpotency of the Hecke algebra on modular forms of SL2 (Z) modulo 2, we make
the following vital observation.

Theorem 4.2. Suppose that f = ∞ n=0 a(n)q ∈ Mk (1) has integer coefficients. If s is a positive
n

integer and t
ks/12, then for every set of odd primes p1 , p2 , . . . , pt we have
f |Tk (p1 )|Tk (p2 )| · · · |Tk (pt ) ≡ 0 (mod 2s ).
Proof. Begin by noticing that the Fourier expansion of every Eisenstein series on SL2 (Z) is congruent
to 1 modulo 2. Serre [Ser76] observed that the Hecke operators act nilpotently on Sk (1) (mod 2),
the space of cusp forms modulo 2 on SL2 (Z). If ∆ ∈ S12 (1) is the unique normalized weight 12 cusp
form
∞

∆(z) = q (1 − q n )24 = q − 24q 2 + · · · ,
n=1
then Sk (1) (mod 2) has F2 -basis
{∆i (mod 2) : 1  i  k/12
}.
Serre’s observation implies that, if j is a positive integer, then


j−1
∆j |Tk (p) ≡ α(i)∆i (mod 2),
i=1
where α(i) ∈ F2 , and so we have
f |Tk (p1 )|Tk (p2 )| · · · |Tk (pt ) ≡ 0 (mod 2) (4.1)
whenever t
k/12. One easily obtains the result by successive division by 2 and iteration of
(4.1).
As an immediate corollary, we obtain the following inequality.

Corollary 4.3. Suppose that f = ∞ n=1 a(n)q ∈ Mk (1) has integer coefficients. If s is a positive
n

integer, then
max{ω(n) : n odd and square-free with ord2 (a(n)) < s} < ks/12.

564
Downloaded from https://www.cambridge.org/core. IP address: 201.124.245.18, on 17 Jan 2021 at 22:41:35, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1112/S0010437X03000721
 Modular functions and divisors of modular forms

Proof.If t
ks/12, then let p1 , p2 , . . . , pt be distinct odd primes. Let f0 := f , and for 1  i  t let
fi = ∞ n
n=0 ai (n)q be the modular forms defined inductively by

fi := fi−1 | Tk (pi ). (4.2)

By Theorem 4.2, we have
at (M ) ≡ 0 (mod 2s )
for every M . In particular, (4.2) implies that
0 ≡ at (1)
= at−1 (pt )
= at−2 (pt−1 pt )
..
.
= a(p1 p2 · · · pt ) (mod 2s ).
This completes the proof.

Theorem 4.4. If s
4 and 0 < D ≡ 3 (mod 8), then F(D; q) (mod 2s ) is the reduction modulo 2s
of a modular form with integer coefficients in Mk(D,s)(1) where
k(D, s) := 12 · 2s−4 H(−D) + 2.

Proof. By construction [BO03, Proposition 2.1], we have

Θ(F (D; z))
F(D; q) = . (4.3)
F (D; z)
We see that F(D; q) is a weight 2 meromorphic modular form on SL2 (Z) which is non-vanishing at
infinity. Moreover, it has a simple zero at each Heegner point τ with discriminant −D and no other
singularities.
It is well known that j(ω) = 0. Let τ1 , . . . , τH(−D) denote the Heegner points of discriminant
−D. For each 1  i  H(−D) define E(D, i; z) by
 
j(τi )
E(D, i; z) := E4 (z) · 1 −
3
. (4.4)
j(z)
Observe that the modular function 1 − j(τi )/j(z) has a simple pole at z = ω and a simple zero
at z = τi . Since E43 (z) has a simple zero at z = ω, the modular form E(D, i; z) is a holomorphic
modular form in M12 (1). Since E4 (z) ≡ 1 (mod 16) and j(τi ) ≡ 0 (mod 215 ) (see [GZ85]), we have
that
E(D, i; z) ≡ 1 (mod 16).
Hence, if s
4, then
s−4
E(D, i; z)2 ≡ 1 (mod 2s ).
Therefore, if s
4, then

Θ(F (D; z))

H(−D)
s−4
F(D; q) ≡ · E(D, i; z)2 (mod 2s ). (4.5)
F (D; z)
i=1

The modular form on the right-hand side of (4.5) is holomorphic and has weight
k(D, s) = 12 · 2s−4 H(−D) + 2.
This completes the proof.

565
Downloaded from https://www.cambridge.org/core. IP address: 201.124.245.18, on 17 Jan 2021 at 22:41:35, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1112/S0010437X03000721
 Modular functions and divisors of modular forms

Proofs of Theorem 10 and Corollary 11. By Corollary 4.3 and Theorem 4.4, we have that if A(D; n)
≡ 0 (mod 2s ), then
k(D, s)s
ω(n) <
12
(12 · 2s−4 H(−D) + 2)s
=
12
= 2 s · H(−D) + s/6.
s−4

Therefore, we find that

ω(n) 1
− < H(−D).
2 s 3 · 2s−3
s−4
This completes the proofs.

References
AKN97 T. Asai, M. Kaneko and H. Ninomiya, Zeros of certain modular functions and an application, Comm.
Math. Univ. Sancti Pauli 46 (1997), 93–101.
Bor95a R. E. Borcherds, Automorphic forms on Os+2,2 (R) and infinite products, Invent. Math. 120 (1995),
161–213.
Bor95b R. E. Borcherds, Automorphic forms on Os+2,2 (R)+ and generalized Kac–Moody algebras, in Proc.
Int. Congress of Mathematicians, Zürich, 1994, vol. 1, 2, ed. S. D. Chatterji (Birkhäuser, Basel,
1995), 744–752.
BO03 J. H. Bruinier and K. Ono, The arithmetic of Borcherds exponents, Math. Ann. 327 (2003), 293–303.
ES96 W. Eholzer and N.-P. Skoruppa, Product expansions of conformal characters, Phys. Lett. B. 388
(1996), 82–89.
GZ85 B. Gross and D. Zagier, On singular moduli, J. Reine Angew. Math. 355 (1985), 191–220.
IR90 K. Ireland and M. Rosen, A classical introduction to modern number theory (Springer, New York,
1990).
Ram16 S. Ramanujan, On certain arithmetical functions, Trans. Camb. Phil. Soc. 22 (1916), 159–184
(Collected Papers, No. 18).
Ser73 J.-P. Serre, Formes modulaires et fonctions zêta p-adiques, Lecture Notes in Mathematics, vol. 350
(Springer, Berlin,1973), 191–268.
Ser76 J.-P. Serre, Divisibilité de certaines fonctions arithmétiques, Enseign. Math. 22 (1976), 227–260.
Zag02 D. Zagier, Traces of singular moduli, in Motives, polylogarithms and Hodge theory, part I, Irvine,
CA, 1998, Int. Press Lecture Series, vol. 3, I, eds F. Bogomolov and L. Katzarkov (International
Press, Somerville, MA, 2002), 211–244.

Jan H. Bruinier [email protected]

Department of Mathematics, University of Wisconsin, Madison, Wisconsin 53706, USA
Current address: Mathematisches Institut, Universität Köln, Im Weyertal 86-90, D-50931 Köln,
Germany

Winfried Kohnen [email protected]

Mathematisches Institut, Universität Heidelberg, INF 288, D-69120 Heidelberg, Germany

Ken Ono [email protected]

Department of Mathematics, University of Wisconsin, Madison, Wisconsin 53706, USA

566
Downloaded from https://www.cambridge.org/core. IP address: 201.124.245.18, on 17 Jan 2021 at 22:41:35, subject to the Cambridge Core terms of use, available at
https://www.cambridge.org/core/terms. https://doi.org/10.1112/S0010437X03000721