Journal of Number Theory 95, 253258 (2002)

doi:10.1006/jnth.2001.2765

Representations of Integers as Sums of Squares

Ken Ono
Department of Mathematics, University of Wisconsin, Madison, Wisconsin 53706
E-mail : ono@math:wisc:edu
Communicated by A. Granville
Received February 28, 2001
DEDICATED TO THE MEMORY OF ROBERT RANKIN

1. INTRODUCTION AND STATEMENT OF RESULTS

If s is a positive integer, then let rs; n denote the number of
representations
integer n as a sum of s integer squares.
P of a non-negative
n n2
2piz
If Yz : 1
1
q
q
:
e
throughout), then
n1
1
X

1n rs; nqn Yzs :

1:1

n0

For small s, there are well-known formulas such as Jacobis four-squares

theorem:
1
X
n0

1n r4; nqn 1 8

1 X
X
n1

1n dqn :

djn;
4[d:

The general problem of determining exact formulas for rs; n is classical

in number theory. One may consult the popular book by Grosswald [1] for a
thorough account (as of the early 1980s) of the subject complete with
references. The series Yzs is a modular form, and so there are abstract
formulas for rs; n as the Fourier coecients of modular forms. Specically,
it is well known that Yzs Es*z cs z, where Es* z is an Eisenstein
series with explicit coecients and cs z is a cusp form. Using this fact, one
may deduce asymptotic information for rs; n. Rankin proved [5] that cs z
is non-trivial for every s > 8. Therefore, the problem of computing nontrivial formulas for rs; n remains since the coecients of cusp forms,
although small, rarely have simple descriptions.
253
0022-314X/02 $35.00
# 2002 Elsevier Science (USA)
All rights reserved.

254

KEN ONO

In a startling turnabout, Milne [3] announced formulas for r4s2 ; n and

r4s2 4s; n for every s. His formulas were obtained by combining a variety
of methods and observations from the theory of elliptic functions, continued
fractions, Lie algebras, Schur functions, and hypergeometric functions. The
proofs of his formulas appear in [4].
Also in [4], he proves (via similar methods) conjectures of Kac and
Wakimoto on the number of representations of positive integers as sums of
triangular numbers. These conjectures were born out of observations arising
in the theory of Lie algebras. In a recent paper [6], Zagier also proves these
conjectures. His method involves an elegant and surprisingly simple
argument. Zagier notices that the generating functions in the Kac and
Wakimoto Conjectures are modular forms on G0 2 whose zeros are
supported on the cusp at innity. Two forms sharing this property with the
same weight must be multiples of each other. Zagier then observes that the
specializations of suitable polynomials with certain Eisenstein series yield
such forms. Therefore, these specializations equal the relevant generating
functions up to easily computable constants.
For r4s2 ; n and r4s2 4s; n, it turns out that a similar analysis applies.
The powers of Yz are modular forms on G0 2 whose zeros are supported
at the cusp inequivalent to innity. Arguing as above with E  k; z (see (1.5)
and (1.6)) and the polynomials in Zagiers work, one easily obtains new
formulas for r4s2 ; n and r4s2 4s; n (see Corollary 2). These formulas are
sums of products of divisor functions, and are simpler than those of Milne.
His formulas involve Schur functions and determinants of Lambert series.
Instead of this approach, we use the fact that the map sending z to 1=2z
swaps (see Proposition 2.1) Yz and the generating function for triangular
numbers. Since the fundamental domain of G0 2 has two cusps which are
interchanged by this map, we obtain our formulas from Zagiers work on
the KacWakimoto conjectures. This is completely elementary.
For every s, let A
s l denote the coecients of the polynomials
s
Y

Xi

i1
s
Y
i1

Xi2  Xj2 2

Xi2  Xj2 2

1:2

a1
as
A
s lX1 Xs :

1:3

la1 ;...;as

14i5j4s

As usual, let sn n :
dened by

a1
as
A
s lX1 Xs ;

la1 ;...;as

14i5j4s

Xi3

djn

d n , and let fBk g denote the Bernoulli numbers

1
X
k0

Bk tk =k! : t=et  1:

1:4

255

SUMS OF SQUARES

If k52 is an even integer, then dene weight k modular forms E  k; z by

!
1
X

2k1 Bk
4n

sk1 nq
E k; z : 2
2k
n1
!
1
X
B
k
1:5

 2k1
1n sk1 nqn ;
2k
n1


E k; z : 2

1
Bk X

sk1 nq2n
2k n1

1s 4s
Q
s! 2s1
j1 j!

Yz

1:6

A
s lE a1 1; z E as 1; z;

1:7

la1 ;...;as

4s2 4s

!
1
Bk X
n
sk1 nq :

2k n1

If s is a positive integer, then

Theorem 1.
Yz4s

22s 3s
Q2s
s! j1 j!



A
s lE a1 1; z E as 1; z:

1:8

la1 ;...;as

If t is an odd integer, then define divisor functions s

t n

Corollary 2.
by

(
s
t n

2
(

s
t n :

Bt1
2t  22t1 2t2
2t1

if n 0;
n

st n=4  2 1 st n otherwise;

Bt1
1  2t1 2t2

t1

if n 0;

st n=2  st n otherwise:

If s is a positive integer, then for every non-negative integer n, we have

r4s2 ; n

1sn 4s
Q
s! 2s1
j1 j!

la1 ;...;as

r4s2 4s; n

A
s l

1n 22s 3s
Q
s! 2s
j1 j!

X
la1 ;...;as

s
a1 m1 sas ms ;

m1 ms n;
mi 50:

A
s l

X
m1 ms n;
mi 50:


s
a1 m1 sas ms :

256

KEN ONO

2.

PROOFS

If k52 is an even integer, then let Gk z denote the weight k Eisenstein

series
Gk z 

1
Bk X

sk1 nqn :
2k n1

2:1

If k54, then Gk is a weight k modular form on SL2 Z. As usual, let Zz

Zz : q1=24

1
Y

1  qn

2:2

n1

be Dedekinds eta-function. It is well known that

Yz Z2 z=Z2z:

2:3

Similarly, it is also well known that

!
1
X
Z2 2z
1=8
n2 n=2
q
Tz :
:
q
Zz
n0

2:4

Up to the factor q1=8 ; Tz is the generating function for the triangular

numbers.
Proposition 2.1.

If s is a positive integer and Imz > 0, then

T1=2z4s

1s z2s
Yz4s :
22s

Proof. In view of (2.3) and (2.4), the proposition follows from the fact
that [2, p. 121]
p
Z1=z z=i Zz: ]
Proposition 2.2.

If k54 is an even integer and Imz > 0, then

(1) Gk 1=4z 4zk Gk 4z,

(2) Gk 4z1 12 2zk Gk z 12.
Proof.

Since

0
1

1
0

and

1
2 1

2 SL2 Z, the modularity of Gk z implies

Gk 1=z zk Gz;

2:5

SUMS OF SQUARES

z1
2z  1

Gk

2z  1k Gk z:

257

2:6

Claim (1) follows from (2.5), and claim (2) follows by replacing z by z 12
in (2.6). ]
If Imz > 0, then
24z
.
(1) G2 1=4z 4z2 G2 4z
pi
24z
(2) G2 4z1 12 2z2 G2 z 12
.
pi

Proposition 2.3.


Proof. Let S : 01 10 and T : 10 11 be the standard generators of
SL2 Z. Claim (1) follows from the fact that [2, p. 113]
G2 Sz G2 1=z z2 G2 z 6z=pi:


1 1
Since G2 Tz G2 z, (2.7) and
ST 2 S1 T 1 , implies
2


G2

z1
2z  1

2:7

1

2z  12 G2 z 122z  1=pi:

Claim (2) follows by replacing z by z 12.

Proof of Theorem 1. First we prove (1.7). If k52 is even, then dene

g k; z by



1
z1
Gk z=2  Gk
g k; z :
:
2
2

2:8

Zagier [6] proved that

1
Q2s1

Tz4s

4ss1 s!

j1

X
j!

A
s lg a1 1; z g as 1; z: 2:9

la1 ;...;as

By replacing z by 1=2z, Proposition 2.1 implies

2

Yz4s

1s 4s
Q
z2s2 s! 2s1
j1 j!

A
s lg a1 1; 1=2z

la1 ;...;as

g as 1; 1=2z:

2:10

258

KEN ONO

By (2.8), Propositions 2.2 and 2.3, we nd that




1
1 1

g k; 1=2z Gk 1=4z  Gk 
2
4z 2



1
k
2k1
k1
z 2
Gk 4z  2 Gk z
zk E k; z:
2
In view of (2.10), this implies (1.7).
To prove (1.8), we begin
g k; z Gk z  Gk 2z, then
Tz4ss1

s!

2s
Q2s

j1 j!

with

Zagiers

formula



A
s lg a1 1; z g as 1; z:

[6].

If

2:11

la1 ;...;as

By (2.5), it is easy to see that

g k; 1=2z 2k zk Gk 2z  zk Gk z zk E  k; z:
By Proposition 2.1, (2.11) and (2.12) implies (1.8).

2:12

ACKNOWLEDGMENTS
The author thanks the Number Theory Foundation, the Alfred P. Sloan Foundation, the
David and Lucile Packard Foundation, and the National Science Foundation for their generous
support.

REFERENCES
1. E. Grosswald, Representations of Integers as Sums of Squares, Springer-Verlag, Berlin,
1984.
2. N. Koblitz, Introduction to Elliptic Curves and Modular Forms, Springer-Verlag, Berlin,
1984.
3. S. Milne, New innite families of exact sums of squares formulas, Jacobi elliptic functions
and Ramanujans tau function, Proc. Nat. Acad. Sci. USA 93 (1996), 1500415008.
4. S. Milne, Innite families of exact sums of squares formulas, Jacobi elliptic functions,
continued fractions, and Schur functions, Ramanujan J. 6 (2002), 7149.
5. R. Rankin, Sums of squares and cusp forms, Amer. J. Math. 87 (1965), 857860.
6. D. Zagier, A proof of the KacWakimoto ane denominator formula for the strange series,
Math. Res. Lett. 7 (2000), 597604.