Modular Forms 1, 2023/2024 UDE

Jie LIN

October 13, 2023

This is the lecture notes for the course Modular Forms 1 at Universität Duisburg-
Essen. The main references are [DS05], [Miy89], [Bum98] and [Kob84].

Contents

0.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
0.1.1 Ramanujan’s τ function: . . . . . . . . . . . . . . . . . . . . . . . 2
0.1.2 Eisenstein series: . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
0.1.3 The j-function: . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
0.1.4 the θ-function: . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
0.2 Fourier transform and Poisson’s formula . . . . . . . . . . . . . . . . . . 7
0.3 Mellin transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
0.4 the Riemann zeta function . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1 Modular forms for SL2 pZq 13

1.1 Basic definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2 Eisenstein series revisited . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3 E2 and ∆ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.4 Modular curve for SL2 pZq . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.5 Dimension formula: SL2 pZq-case . . . . . . . . . . . . . . . . . . . . . . 29

2 Modular forms for congruence subgroups 34

2.1 Theta series revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.2 Basic definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.3 Some examples and applications . . . . . . . . . . . . . . . . . . . . . . 39
2.3.1 Theta series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.3.2 An interlude: old forms . . . . . . . . . . . . . . . . . . . . . . . 40

1
CONTENTS 2

2.3.3 Dedekind η function . . . . . . . . . . . . . . . . . . . . . . . . . 41

2.4 Eisenstein series for ΓpN q . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.5 Modular function and dimension formula . . . . . . . . . . . . . . . . . 48

3 Hecke operators 56
3.1 Double coset operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.2 Diamond and Hecke operators . . . . . . . . . . . . . . . . . . . . . . . . 58
3.3 Hecke operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.4 Invariant measure on H . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.5 Adjoint of Hecke operators . . . . . . . . . . . . . . . . . . . . . . . . . 68

A Representation of finite abelian group 76

B Invariant measure 77

Introduction:

0.1 Examples

0.1.1 Ramanujan’s τ function:

Definition 0.1. We define τ pnq, n ě 1, to be the n-th coefficient of the formal power
8
p1 ´ q n q24 . In other words, we have,
ś
series ∆ :“ q
n“1

8
ÿ 8
ź
n
∆“ τ pnqq “ q p1 ´ q n q24 . (0.1)
n“1 n“1

Here are some values of τ :

n 1 2 3 4 5 6 8 9 10 12
τ pnq 1 ´24 252 -1472 4830 ´6048 84480 ´113643 ´115920 ´370994

Proposition 0.2. These values have several properties:

(a) τ pnmq “ τ pnqτ pmq if pn, mq “ 1;

(b) τ ppk`1 q “ τ ppqτ ppk q ´ p11 τ ppk´1 q for p prime and k ě 1;

(c) (Ramanujan’s conjecture) |τ ppq| ď 2p11{2 .

CONTENTS 3

We will prove in our course the first two points and a weaker version of the third one
by replacing 11 “ 12 ´ 1 by 12. We remark that the Ramanujan conjecture is a special
case of one part of the Weil conjecture proved by Deligne which is the analogue of the
Riemann hypothesis on function fields.

0.1.2 Eisenstein series:

Definition 0.3. We define the Bernoulli numbers Bn P Q, n ě 0, by:

8
t ÿ Bn n
t
“ t . (0.2)
e ´ 1 n“0 n!

Here are some values of Bn :

k ě 3
n 0 1 2 4 6 8 10 12
odd
Bn 1 1 1 1 1 1 5 691 0
´ ´ ´ ´
2 6 30 42 30 66 2730
Exercise 0.4. Prove that if k ě 3 is odd, then Bk “ 0.

Definition 0.5. Let k ě 2 be an integer. We define the k-th Eisenstein series by:
8
4k ÿ
E2k :“ 1 ´ σ2k´1 pnqq n (0.3)
B2k n“1

dl for n P N and l P Z.
ř
where σl pnq :“
d|n

Example 0.6. We have:

8
ÿ
E4 “ 1 ` 240 σ3 pnqq n “ 1 ` 240pq ` 9q 2 ` 28q 3 ` 73q 4 ` ¨ ¨ ¨ q
n“1
ÿ8
E6 “ 1 ´ 504 σ5 pnqq n “ 1 ´ 504pq ` 33q 2 ` 244q 3 ` 1057q 4 ` ¨ ¨ ¨ q
n“1
ÿ8
E8 “ 1 ` 480 σ7 pnqq n “ 1 ` 480pq ` 129q 2 ` 2188q 3 ` 16513q 4 ` ¨ ¨ ¨ q
n“1
ÿ8
E10 “ 1 ´ 264 σ9 pnqq n “ 1 ´ 264pq ` 513q 2 ` 19684q 3 ` 262657q 4 ` ¨ ¨ ¨ q
n“1
8
65520 ÿ 65520
E12 “ 1` σ11 pnqq n “ 1 ` pq ` 2049q 2 ` 177148q 3 ` 4196353q 4 ` ¨ ¨ ¨ q
691 n“1 691

We will prove the following relations in our course:

CONTENTS 4

Proposition 0.7. (a) E42 “ E8 ;

(b) E4 E6 “ E10 ;

(c) E43 ´ E62 “ 1728∆ and E62 ´ E12 “ ´762048∆{691.

Corollary 0.8. (exercise) Deduce from the above properties that for any n ě 1 we have:

(a) τ pnq ” σ11 pnq mod 691;

n´1
ř
(b) σ7 pnq “ σ3 pnq ` 120 σ3 pmqσ3 pn ´ mq.
m“1

Remark 0.9. You can verify the previous relations at https://sagecell.sagemath.org. Here
are some codes example:

S.<q>=QQ[[]]

PREC = 500

Delta=q+O(q^PREC)

for n in range(1,PREC):

Delta*=(1-q^n)**24

print(Delta)

tau = list(Delta)

print(tau)

print(tau[13] * tau[25] - tau[13*25])

print(tau[9]-tau[3]*tau[3]+(3**11)*tau[1])

def E(k):

f = O(q^PREC)

for n in range(1, PREC):

f += sigma(n, k - 1) * q^n
CONTENTS 5

f *= -2*k/bernoulli(k)

f += 1

return f

for k in range(2,7):

print(E(2*k))

print(E(4)*E(4)-E(8))

print(E(4)*E(6)-E(10))

print(E(6)^2-E(12)+762048/691*Delta)

print((E(4)^3 - E(6)^2)/1728-Delta)

0.1.3 The j-function:

Definition 0.10. We define:

E43
j: “ 1728
E43 ´ E62
1
“ ` 744 ` 196884q ` 21493760q 2 ` 864299970q 3 ` 20245856256q 4 ` ¨ ¨ ¨
q
8
1 ÿ
:“ ` 744 ` an q n .
q n“1

Exercise 0.11. Show that the coefficients an , n ě 1, are all integers.

One famous application of properties of the j-function is that:

?
eπ 163 is almost an integer.

More precisely, we have:

?
eπ 163
« 6403203 ` 744.
CONTENTS 6

We now give a brief explanation of above. One can show that when |q| ă 1, the series
∆ and E2k , k ě 2, all converge absolutely (exercise). Hence they define holomorphic
functions on the complex unit disc D :“ tq P C | |q| ă 1u.
We define the Poincaré half plane by H :“ tz P C | Impzq ą 0u. The map
z ÞÑ qz :“ e2πiz is a holomorphic function from H to D. Therefore we can consider
∆ and E2k , k ě 2, as holomorphic functions and then consider j as a meromorphic
function on H.

The CM (complex multiplication) theory shows that if K is a quadratic imaginary

field with OK “ Z ` Zz, then jpzq is an algebraic integer of degree at most hK , the class
number of K. In particular, if hK “ 1, then jpzq is a rational integer (c.f. [Sil85]).

?
For example, the number field?K “ Qp 163iq has class number 1 and hence jpzq is
1 ` 163i ?
a rational integer where z :“ . Note that qz “ e2πiz “ ´e´π 163 . We have
2
? 8
ÿ ?
π 163
jpzq “ ´e ` 744 ` p´1qn an e´n 163π
P Z.
n“1
?
? ? 1` 163i
Since e´ 163π is very small, we can expect that eπ 163 « ´jp q ` 744 which is
2
an integer.
?
1` 163i
One can further more prove that jp q is a cube. In fact, we have:
2

? ? ? ? ? ? ? ?
z 1`i 1 ` 2i 1 ` 3i 1 ` 7i 1` 11i 1` 19i 1` 43i 1` 67i 1` 163i
2 2 2 2 2 2 2
jpzq 123 203 03 ´153 ´323 ´963 ´9603 ´52803 ´6403203

We also refer to [Apo90] for more applications of the j-function, for example, the
proof of Picard’s theorem which says that every non-constant entire function attains
every complex value with at most one exception.

0.1.4 the θ-function:

Definition 0.12. We define:

2
ÿ
θptq :“ e´tπn , t ą 0. (0.4)
nPZ

This series is absolutely convergent for all t ą 0 and satisfies the following functional
equation:
Proposition 0.13. The theta series satisfies the following functional equation for all t ą 0:
?
θp1{tq “ tθptq. (0.5)
CONTENTS 7

We shall prove this in section 0.2. We remark that it is also a key step in the proof
of the functional equation for the Riemann zeta function by Riemann himself.

Another application of the θ-function is the following:

Proposition 0.14. Let n be a positive integer. Let r4 pnq denote the number of ways to
write n into sum of four squares. Then
ÿ
r4 pnq “ 8 d (0.6)
d|n,4∤d

For example, let n “ 5 “ 12 ` 22 , then r4 p5q “ 4 ˚ 3 ˚ 4 which indeed equals 8p1 ` 5q.

0.2 Fourier transform and Poisson’s formula

Definition 0.15. (a) We define the space of Schwartz functions SpRq to be the space
of smooth functions f : R Ñ C such that for any integer k ą 0 and any C ą 0,
the function xC f pkq pxq converges to 0 when |x| tends to 8.

(b) Let f P SpRq be a Schwartz function. We define its Fourier transform by:
ż8
fˆpxq :“ f pyqe2πixy dy, , x P R. (0.7)
´8

The Fourier transform gives a one-to-one map from SpRq to itself. Moreover, we
ˆ
have the Fourier inversion formula fˆpxq “ f p´xq (c.f. [Kat04]).
2
Lemma 0.16. Let t ą 0. Define ft pxq :“ e´πtx for x P R. Then

1 2 1
fˆt pxq “ ? e´πx {t “ ? f1{t pxq.
t t

Proof. By definition we have:

ż8
2
fˆt pxq “ e´πty e2πixy dy
´8
ż8
2 2 {t
“ e´πtpy´ix{tq e´πx dy
´8
ż 8´ix{t
´πx2 {t 2
“ e e´πty dy
´8´ix{t
ż8
2 {t 2
“ e´πx e´πty dy
´8
ż8
1 2 2
“ ? e´πx {t e´πy dy
t ´8
1 2
“ ? e´πx {t
t
CONTENTS 8

´πy 2 dy
ş8
where the last equation is due to the fact that ´8 e ą 0 and
ż8 ż8 ż8
2 2 2
p e´πy dyq2 “ e´πy e´πx dxdy
´8 ´8 ´8
ż 8 ż 2π ż8 ż8 ż8
´πr2 ´πr2 ´πr2 2
“ e rdrdθ “ e 2πrdr “ e dpπr q “ e´x dx “ 1.
0 0 0 0 0

Proposition 0.17 (Poisson’s formula). Let f P SpRq. Then we have:

ÿ ÿ
f pnq “ fˆpnq (0.8)
nPZ nPZ

where both sums converge absolutely.

ř
Proof. Since f is Schwartz, the series f px ` nq converges uniformly on any compact
nPZ
subset of R, and hence defines a smooth function, denoted by F pxq. It is moreover
1-periodic. It has Fourier expansion
ÿ
F pxq “ am e2πimx
mPZ

wehre
ż1 ż1 ÿ ż8
am “ F pxqe ´2πimx
dx “ f px ` nqe ´2πimx
“ f pxqe´2πimx dx “ fˆp´mq.
0 0 nPZ ´8

In particular, we have
ÿ ÿ ÿ ÿ
f pnq “ F p0q “ am “ fˆp´mq “ fˆpnq.
nPZ mPZ mPZ nPZ

We can now prove Proposition 0.13, the functional equation for θ.

ř
Proof. Note that θptq “ ft pnq. By the Poisson summation formula and Lemma 0.16,
nPZ
we have:
ÿ ÿ ÿ 1 1
θptq “ ft pnq “ fˆt pnq “ ? f1{t pnq “ ? θp1{tq
nPZ nPZ nPZ
t t
?
and hence θp1{tq “ tθptq as expected.
CONTENTS 9

0.3 Mellin transform

Let f : R` Ñ C be a continuous function.

Definition 0.18. We assume that there exist A, B P R such that tA f ptq is bounded when
t Ñ 0 and tB f ptq is bounded when t Ñ 8. Then the integral
ż8
dt
M f psq :“ f ptqts
0 t

converges absolutely and uniformly on any compact subset in the range A ă Repsq ă B
and hence defines a holomorphic function in this range, called the Mellin transform
of f .
ş8dt
Example 0.19. (a) Let f “ e´t . Then M f psq “ 0 e´t ts
“: Γpsq is well-defined on
t
Repsq ą 0. One can show easily that Γps ` 1q “ sΓpsq by integration by parts. We
can hence extend Γ to a meromorphic function on the whole complex plane.

The Gamma function has simple poles at non-positive integers ´n, n ě 0, with
p´1qn
residues . Its values at positive integers are Γpnq “ pn ´ 1q! with n ě 1.
n!
π 1 ?
We remark the fact that ΓpsqΓp1 ´ sq “ . Consequently, Γp q “ π, and
sinpπsq 2
Γ has no zero everywhere. In particular, Γ´1 is an entire function.

t 8
ř tn
(b) Let f “ “ B n . Then for Repsq ą 1, we have:
et ´ 1 n“0 n!
ż8
t s´1 dt
ż8 ´t dt
s e
M f ps ´ 1q “ t
t “ t
0 e ´1 t 0 1 ´ e´t t
ż8 8 8 8
dt dt
ÿ ż ÿ
“ ts e´t e´nt “ ts e´nt
0 n“0
t 0 n“1
t
8
ÿ 8 8 8
dt dt
ż ÿ ż
“ ts e´nt “ n´s ts e´t “ ζpsqΓpsq
n“1 0
t n“1 0 t

8 1
ř
where ζpsq :“ s
is the Riemann zeta function.
n“1 n

Definition 0.20. We say f decays rapidly at 8 (resp. 0) if for all M P Z the function
tM f ptq is bounded when t Ñ 8 (resp. t Ñ 0).

If f decays rapidly at both 0 and 8, then M f defines a holomorphic function on the

whole complex plane. However, this is not always the case. The following situation is
more common:
CONTENTS 10

8
an tn
ř
Proposition 0.21. If f decays rapidly at 8 and has asymptotic expansion f ptq „
n“0
Nř
´1
near 0, i.e., for any N ą 0, |f ptq ´ an tn | “ OptN q when t Ñ 0.
n“0
Then M f defines a holomorphic function for Repsq ą 0 and can be continued to a
meromorphic function on C. It can only has simple poles at non-positive integers ´n,
n ě 0, with residues an .

Proof.
ż1 ż8
s dt
M f psq “ f ptqt ` f ptqts
0 t 1
ż1 Nÿ
´1 Nÿ
´1 ż8
n s dt an dt
“ pf ptq ´ an t qt ` ` f ptqts .
0 n“0
t n“0
n`s 1 t

ş1 Nř
´1 dt ş8 dt
The term 0 pf ptq´ an tn qts
converges absolutely for Repsq ą ´N, the term 1 f ptqts
n“0 t t
converges absolutely for all s P C. We can then conclude by tending N to infinity.

0.4 the Riemann zeta function

t 8
ř tn
Recall that Γpsqζpsq “ M f ps ´ 1q where f “ “ B n . Proposition 0.21 along
et ´ 1 n“0 n!
with the analytic properties of Γ (see example 0.19) leads to:

Theorem 0.22. (a) The Riemann zeta function can be extended to a meromorphic func-
tion on the whole complex plane.

(b) It is holomorphic everywhere except for a simple pole at s “ 1 with residue 1.

Bn`1
(c) ζp´nq “ p´1qn for n ě 0.
n`1

The Mellin transform of θ does not converge in any range because the constant term.
ř ´tπn 2
Hence we consider f :“ pθptq ´ 1q{2 “ e which decays rapidly at 8. By the
ně1
functional equation of θ (Proposition 0.13 ) , we get:
?
1 ? t´1
f p q “ tf ptq ` .
t 2

For A P R, we have p1{tqA f p1{tq “ t´A`1{2 f ptq ` t´A`1{2 p1 ´ t´1{2 q{2 is bounded
when t Ñ 8 if and only if A ą 1{2. Hence M f psq converges on Repsq ą 1{2 and equals
ż8 ÿ
2 dt ÿ 1
e´tπn ts “ s n2s
Γpsq “ π ´s ζp2sqΓpsq
0 ně1
t ně1
π
CONTENTS 11

The functional equation of f should lead to some property of the Riemann zeta
function, which is the functional equation of ζ. We now give more details.

Note that in the functional equation of f we transform t to 1{t, it is natural to sep-

arate the integral in the Mellin transform into two parts, one from 0 to 1, the other one
from 1 to 8. Note that the second integral converges absolutely for all s. We can trans-
form the first integral to an integral from 1 to 8 by the functional equation of f . We
then get the analytic continuation and the functional equation for M f at the same time.

More precisely, we have for Repsq ą 1{2:

ż8
dt
M f psq “ f ptqts
0 t
ż1 ż8
s dt dt
“ f ptqt ` f ptqts
0 t 1 t
ż8 ż8
dt dt
“ f p1{tqt´s ` f ptqts
1 t t
ż8 ? ? 1 ż8
t ´ 1 ´s dt dt
“ p tf ptq ` qt ` f ptqts
1 2 t 1 t
1
ż8 ´s ´s
1 t2 t
“ pt1{2´s ` ts qf ptq ` p 1 ´ q|8
1
1 2 2 ´ s ´s
ż8
1 1 1
“ pt1{2´s ` ts qf ptq ` p 1 ´ sq.
1 2 s´ 2
ş8
Since 1 pt1{2´s ` ts qf ptq converges absolutely for all s, M f psq can be continued to a
meromorphic function to the whole complex plane with simple poles 12 and 0 and residues
1 1
2 and ´ 2 respectively. Moreover, the last equation is obviously symmetric with respect
to 2 , i.e., M f psq “ M f p 21 ´ sq. We get another proof of the first two parts of Theorem
1

0.22 and moreover:

Theorem 0.23. The Riemann zeta function satisfies the functional equation:
s 1´s
π ´s{2 Γp qζpsq “ π ´p1´sq{2 Γp qζp1 ´ sq. (0.9)
2 2
p´1qn`1 B2n p2πq2n
Corollary 0.24. Let n ě 1 be an integer. Then ζp2nq “ .
2p2nq!
π2 π4
For example, ζp2q “ and ζp4q “ . In general, we have ζp2nq P π 2n Qˆ .
6 90

Proof. Let s “ 2n in the functional equation of ζ. We get:

1 ´ 2n
π ´n Γpnqζp2nq “ π ´p1´2nq{2 Γp qζp1 ´ 2nq.
2
CONTENTS 12

?
Recall that Γps ` 1q “ sΓpsq, Γp0q “ 0 and Γp 21 q “ π, we get Γpnq “ pn ´ 1q! and

1 ´ 2n 1 1 1 1 p´1qn 2n ? p´1qn 22n n!?

Γp q“ 1´2n 3´2n ¨ ¨ ¨ ´1 Γp q “ π “ π.
2 2 2 2
2 p2n ´ 1q!! p2nq!

B2n
We can conclude by ζp1 ´ 2nq “ ´ from Theorem 0.22.
2n
Remark 0.25. If we take s “ 2n ` 1 an odd positive integer, then the left hand side
of functional equation equals π ´n´1{2 Γpn ` 1{2qζp2n ` 1q. But on the right hand side,
B2n`1
Γpsq has a pole at s “ n and ζp´2nq “ is zero. If we want to deduce ζp2n ` 1q
2n ` 1
from the functional equation, we must know the derivation of ζ at p´2nq which is rather
difficult.
1 MODULAR FORMS FOR SL2 pZq 13

1 Modular forms for SL2 pZq

1.1 Basic definition

Recall the Poincaré half plane H :“ tz P C | Impzq ą 0u.

ˆ ˙
a b
Definition-Lemma 1.1. For z P H and γ “ P SL2 pZq. We define γz P C by:
c d

az ` b
γz :“ .
cz ` d
Then we get an action of SL2 pZq on H (exercise). More precisely, we have:
ˆ ˙
a b
• γz P H for any γ P SL2 pZq and z P H, in fact, for γ “ P SL2 pZq,
c d
Impγzq “ |cz ` d|´2 Impzq;

• I2 z “ z for any z P H;

• γpγ 1 zq “ pγγ 1 qz for any γ, γ 1 P SL2 pZq and z P H.

ˆ ˙ ˆ ˙
1 1 0 ´1
Exercise 1.2. The group SL2 pZq is generated by T “ and S “ . More-
0 1 1 0
over, we verify that T z “ z ` 1 and Sz “ ´1{z for any z P H.
ˆ ˙
a b
Definition 1.3. Let z P H and γ “ P SL2 pZq. We define a factor of automor-
c d
phy
jpγ, zq :“ cz ` d.

Let k be an integer. We write V for the space of functions from H to C. The weight
k operator, written as |k , defines a right action of SL2 pZq on V as follows:
for any γ P SL2 pZq and f P V , we define f |k γ : H Ñ C by

pf |k γqpzq :“ jpγ, zq´k f pγzq, z P H.

Exercise 1.4. 1. Prove that for each γ1 , γ2 P SL2 pZq and z P H, we have jpγ1 γ2 , zq “
jpγ1 , γ2 zqjpγ2 , zq.

2. Deduce the weight k operator indeed defines a right action of SL2 pZq on V . More
precisely, show that f |k I2 “ f , f |k pγ1 γ2 q “ pf |k γ1 q|k γ2 for any f P V , γ1 , γ2 P
SL2 pZq.

Definition 1.5. Let f be a function from H to C. We say f is a modular form (resp.

cusp form) of weight k and level SL2 pZq if it satisfies the following conditions:

(a) f is holomorphic on H;
1 MODULAR FORMS FOR SL2 pZq 14

ˆ ˙
a b
(b) (transformation property) for any γ “ P SL2 pZq, f |k γ “ f , or equiv-
c d
alently, f pγzq “ pcz ` dqk f pzq for any z P H;

(c) f is holomorphic (resp. zero) at 8.

We denote the space of modular forms and cusp forms of weight k and level SL2 pZq by
Mk pSL2 pZqq and Sk pSL2 pZqq respectively.

ˆWe need
˙ some explanation on part (c) above. By part (b), we have f pz ` 1q “
1 1
fp zq “ f pzq. Hence f is 1-periodic and has Fourier expansion:
0 1
ÿ
f pzq “ an q n (1.1)
nPZ

where q “ e2πiz .
We say f is holomorphic at 8 if an “ 0 for n ă 0. In this case, we write f p8q :“ a0 .
In particular, we say f is zero at 8 if an “ 0 for all n ď 0.

an e´2πny e2πinx we see that

ř
Remark 1.6. By the formula f px ` iyq “
nPZ
ż1
an e´2πny “ f px ` iyqe´2πinx dx.
0

In particular, if f is bounded uniformly in x when y Ñ 8, then an “ 0 for any n ă 0.

8
an q n is a modular form, then an is of
ř
Conversely we shall show that if f “
n“0
polynomial growth in n. In particular, f px ` iyq converges uniformly in x to a0 when
y Ñ 8. In particular, it is bounded uniformly in x when y Ñ 8. The definition
f p8q “ a0 then makes sense.

Remark 1.7. Let γ and γ 1 P SL2 pZq, k an integer and f be a complex valued function on
H. If f |k γ “ f |k γ 1 “ f , then by the second part of Exercise 1.4, we have

f |k pγγ 1 q “ pf |k γq|k γ 1 “ f |k γ 1 “ f.

Hence one only need to verify condition (b) in Definition 1.5 for any family of generators
of SL2 pZq. By Exercise 1.2, S and T generates SL2 pZq. Hence we can change condition
(b) in Definition 1.5 to the following:

pb1 q f pz ` 1q “ f pzq and f p´1{zq “ z k f pzq.

Remark 1.8. (a) Both Mk pSL2 pZqq and Sk pSL2 pZqq are vector spaces over C.

(b) If k is odd, then Mk pSL2 pZqq “ Sk pSL2 pZqq “ t0u.

(c) Let k, l be two integers. If f P Mk pSL2 pZqq and g P Ml pSL2 pZqq, then f ˚ g P
Mk`l pSL2 pZqq and f {g P Mk´l pSL2 pZqq provided that g has no zero on H Y t8u.
1 MODULAR FORMS FOR SL2 pZq 15

ř
(d) Let MpSL2 pZqq :“ Mk pSL2 pZqq be the C-vector space generated by modular
kPZ
forms of any weight and level SL2 pZq. By (c), it is an algebra over C endowed
with the natural multiplication of functions and with unity the constant function
1. One can show that modular forms of different weights are linearly independent.
Hence à
MpSL2 pZqq “ Mk pSL2 pZqq.
kPZ

1.2 Eisenstein series revisited

Definition 1.9. Let k ě 2 be an integer. We define:

ÿ
1
G2k pzq :“ pmz ` nq´2k
pm,nqPZ2
ř 1
where denotes the sum where pm, nq runs through all the couples pm, nq in
pm,nqPZ2
Z2 ´ tp0, 0qu.

We shall show in the exercise session that the series converges absolutely and uni-
formly on any compact subset of H and hence defines a holomorphic function on H.
ˆ ˙
a b
Lemma 1.10. For any γ “ P SL2 pZq, we have G2k pγzq “ pcz ` dq2k G2k pzq
c d

Proof. We have
˜ ¸´2k
az ` b
ÿ1
G2k pγzq “ m `n
cz ` d
pm,nqPZ2
ÿ1
“ pcz ` dq2k ppma ` ncqz ` pmb ` ndqq´2k .
pm,nqPZ2

Since γ P SL2 pZq, the map

Z2 ´ tp0, 0qu Ñ Z2 ´ tp0, 0qu, pm, nq Ñ pma ` nc, mb ` ndq

is bijective. We conclude that the last equation equals pcz ` dq2k G2k as expected.

To show G2k is a modular form, it remains to show that it is holomorphic at 8. By

remark 1.6, it is enough to show that G2k is bounded uniformly in x when y Ñ 8.
In fact, we have
ÿ ÿ ÿ ÿ ÿ
G2k pzq “ n´2k ` pmz ` nq´2k “ 2ζp´2kq ` pmz ` nq´2k
n‰0 m‰0 nPZ m‰0 nPZ

where the second term converges uniformly in x to zero then y Ñ 8. We refer to

Exercise 4 of the first exercise sheet for more details.
One can moreover show that:
1 MODULAR FORMS FOR SL2 pZq 16

Proposition 1.11. G2k “ 2ζp2kqE2k . Consequently, both G2k and E2k are modular forms
of weight 2k and level SL2 pZq.

We only prove the following lemma here and leave the rest of the proof in the exercise
session.

Lemma 1.12. Let z P H and k ě 2 be an integer. Then

8
ÿ p2πiqk ÿ
pn ´ zq´k “ nk´1 e2πinz . (1.2)
nPZ
pk ´ 1q! n“1

Proof. Let f puq “ pu ´ zq´k for u P R. Then for v P R,

ż
ˆ
f pvq “ pu ´ zq´k e2πiuv du.
R

Let M ą |z| be a large enough real number.

If v ą 0, we consider the path CM “ γM ` lM , where γM is the upper half circle
with center 0 and radius M , lM is the line segment from ´M to M , as indicated on the
following graph.
y
γM

‚z

lM M x

The function u Ñ pu´zq´k e2πiuv has a poles at z “ u. By Cauchy’s residue formula,

we have
ż ´ ¯
pu ´ zq´k e2πiuv du “ 2πiResu“z pu ´ zq´k e2πiuv
CM
˜ ¸
8 n
ÿ p2πipu ´ zqvq
“ 2πie2πivz Resu“z pu ´ zq´k
n“0
n!
p2πiqk k´1 2πivz
“ v e .
pk ´ 1q!

Moreover,
ż żπ
iθ
| pu ´ zq´k e2πiuv du| “ | pM eiθ ´ zq´k e2πiM e v iM eiθ dθ|
γM 0
żπ
iθ
ď |pM eiθ ´ zq´k e2πiM e v iM eiθ |dθ
0
1 MODULAR FORMS FOR SL2 pZq 17

żπ
“ |M eiθ ´ z|´k e´2π sin θM v M dθ
ż0π
M Ñ`8
ď M |M eiθ ´ z|´k dθ ÝÝÝÝÝÑ 0
0

where the last inequality is due to the fact that v ą 0.

Hence
p2πiqk k´1 2πivz
ż ż
fˆpvq “ lim pu´zq ´k 2πiuv
e du “ lim pu´zq ´k 2πiuv
e du “ v e .
M Ñ8 γM M Ñ8 γM `lM pk ´ 1q!

If v ď 0, we consider the following path and get fˆpvq “ 0.

y
‚z

lM Mx

1
γM

We can finally conclude by applying Poisson’s formula (c.f. Proposition 0.17).

Remark 1.13. Careful reader may notice that Proposition 0.17 considers the case where
the function is Schwartz. But here the function u Ñ pu ´ zq´k is not. We remark that
Poisson’s formula still holds for continuous function f on R if there exists c ą 1 such
that both f and f pxqxc are bounded on R.

We will give another proof of this lemma in the exercise session.

Corollary 1.14. For each k ě 4 even, we have a C-vector space decomposition:

à
Mk pSL2 pZqq “ Sk pSL2 pZqq CEk .

8
an q n P Mk pSL2 pZqq. Then f ´ a0 Ek P Mk pSL2 pZqq by part (c)
ř
Proof. Let f “
n“0
of remark 1.8. Moreover, pf ´ a0 Ek qp8q “ f p8q ´ a0 Ek p8q “ a0 ´ a0 “ 0. Hence
f ´ a0 Ek P Sk pSL2 pZqq as expected.

In fact we can replace Ek by any element inside Sk pSL2 pZqq´Mk pSL2 pZqq. However,
the above decomposition is canonical in the sense that Ek is orthogonal to Sk pSL2 pZqq
with respect to the Petersson inner product that we are going to define later (see Remark
3.32).
1 MODULAR FORMS FOR SL2 pZq 18

À
Recall that MpSL2 pZqq “ kPZ Mk pSL2 pZqq is an algebra over C. We shall show
later that this algebra is generated by E4 and E6 . We show here that E4 and E6 are
algebraically independent. Consequently, we will get
à
Mk pSL2 pZqq “ CrE4 , E6 s – CrX, Y s.
kě0
?
´1` 3i
Proposition 1.15. Let ρ “ , a cubic root of 1. We have:
2
(a) E6 piq “ 0 and E4 pρq “ 0;
(b) E4 piq ‰ 0;
(c) E4 and E6 are algebraically independent.
ˆ ˙
0 ´1
Proof. (a) Note that for any f P Mk pSL2 pZqq, we have f piq “ f p iq “ ik f piq.
1 0
Hence f piq “ 0 if 4 ∤ k. In particular, E6 piq “ 0.
ˆ ˙
´1 ´1
Similarly, f pρq “ f p ρq “ ρk f pρq. Hence f pρq “ 0 if 3 ∤ k. In particular,
1 0
E4 pρq “ 0.
8
ř
(b) In fact, E4 piq “ 1 ` 240 σ3 pnqe´2πn is positive real and hence non-zero.
n“1

(c) We now assume by contradiction that E4 and E6 are not algebraically independent.
Note that for a ě 0 and b ě 0, the function E4a E6b P M4a`6b pSL2 pZqq. By looking
at the weights, there exist an integer k ě 0 and a family of complex numbers
tca,b uaě0,bě0,4a`6b“k , not all zero, such that
ÿ
ca,b E4a E6b pzq “ 0.
aě0,bě0,4a`6b“k

We may assume that k is minimal. Take z “ i, since E4 piq ‰ 0, we get ca,b “ 0 if

b “ 0. In particular,
ÿ
ca,b E4a E6b´1 pzq “ 0
aě0,pb´1qě0,4a`6pb´1q“k´6

where the family tca,b uaě0,pb´1qě0,4a`6pb´1q“k´6 are not all zero which contradicts
to the fact that k is minimal.

1.3 E2 and ∆
ř 1 pmz
You may wonder what happens if k “ 1. In this case, the series ` nq´2k
pm,nqPZ2
does not converge absolutely. We arrange the sum as:
ÿ ÿ
G2 pzq :“ pmz ` nq´2 (1.3)
mPZ nPZm
1 MODULAR FORMS FOR SL2 pZq 19

where Zm “ Z if m ‰ 0, and Zm “ Z ´ t0u otherwise. It is easy to see that:

ÿ ÿ
G2 pzq :“ 2ζp2q ` pmz ` nq´2 .
m‰0 nPZ

Lemma 1.16. The above series converges conditionally for all z, and satisfies:
8 8
ÿ π2 ÿ
G2 pzq “ 2ζp2q ´ 8π 2 σ1 pnqq n “ p1 ´ 24 σ1 pnqq n q
n“1
3 n“1

where q “ e2πiz . Moreover, the right hand side converges absolutely and uniformly on
any compact subset of H. Hence G2 is holomorphic on H.

Proof. By Lemma 1.12, we have:

ÿ 8
ÿ
pn ´ zq´2 “ p2πiq2 ke2πikz .
nPZ k“1

Hence
ÿ ÿ ÿ ÿ ÿ
pmz ` nq´2 “ n´2 ` 2 pn ´ mzq´2
mPZ nPZm 1 nPZ0 1 mą0 nPZ
8 ÿ
ÿ 8
“ 2ζp2q ´ 8π 2 ke2πimkz .
m“1 k“1

8
8 ř
ke2πimkz converges absolutely and uniformly on any
ř
It remains to show that
m“1 k“1
compact subset of H. We leave this and the remaining as an exercise.

Apparently we have G2 pz ` 1q “ G2 pzq. We have shown that G2 is holomorphic on

H as well as at infinity. Is G2 a modular form of weight 2 then? By Remark 1.7, G2 is
a modular form if and only if G2 p´1{zq “ z 2 Gpzq.

Lemma 1.17. For all z P H, we have

z ´2 G2 p´1{zq “ G2 pzq ´ 2πiz ´1 . (1.4)

Proof. We have:
ÿ ÿ ÿ ÿ
z ´2 G2 p´1{zq “ z ´2 p´mz ´1 ` nq´2 “ p´m ` nzq´2
mPZ nPZm mPZ nPZm
ÿ ÿ ÿ ÿ
´2
“ pm ` nzq “ pn ` mzq´2
mPZ nPZm nPZ mPZn
ÿ ÿ ÿ ÿ
´2
“ pmz ` nq “ 2ζp2q ` pmz ` nq´2 .
nPZ mPZn nPZ m‰0
1 MODULAR FORMS FOR SL2 pZq 20

ř ř
Here pmz `nq´2 does not converge absolutely, hence we can not change the order
nPZ m‰0 ř ř
of summand. But pmz ` nq´2 pmz ` n ` 1q´1 converges absolutely. Therefore
nPZ m‰0
ÿ ÿ ÿ ÿ
pmz ` nq´2 pmz ` n ` 1q´1 “ pmz ` nq´2 pmz ` n ` 1q´1 . (1.5)
nPZ m‰0 m‰0 nPZ

The right hand side of equaton (1.5) equals:

ÿ ÿ
ppmz ` nq´2 ´ pmz ` nq´1 pmz ` n ` 1q´1 q
m‰0 nPZ
ÿ ÿ ÿ
“ p pmz ` nq´2 ´ pmz ` nq´1 pmz ` n ` 1q´1 q
m‰0 nPZ nPZ
ÿ ÿ ÿ
“ p pmz ` nq´2 ´ ppmz ` nq´1 ´ pmz ` n ` 1q´1 q
m‰0 nPZ nPZ
ÿ ÿ
´2
“ pmz ` nq .
m‰0 nPZ

The left hand side of equation (1.5) equals

ÿ ÿ
ppmz ` nq´2 ´ pmz ` nq´1 pmz ` n ` 1q´1 q
nPZ m‰0
ÿ ÿ ÿ
“ p ppmz ` nq´2 ´ pmz ` nq´1 pmz ` n ` 1q´1 q
nPZ m‰0 m‰0
ÿ ÿ ÿ ÿ
´2
“ pmz ` nq ´ pmz ` nq´1 pmz ` n ` 1q´1
nPZ m‰0 nPZ m‰0
ÿ ÿ ÿ ÿ
´2
“ pmz ` nq ´2 pmz ` nq´1 pmz ` n ` 1q´1
nPZ m‰0 nPZ mě1

ř ř
It remains to show that pmz ` nq´1 pmz ` n ` 1q´1 “ ´πiz ´1 . Indeed, we
nPZ mě1
have:
ÿ ÿ Nÿ
´1 ÿ
´1 ´1
pmz ` nq pmz ` n ` 1q “ lim pmz ` nq´1 pmz ` n ` 1q´1
N Ñ8
nPZ mě1 n“´N mě1
ÿ Nÿ
´1
“ lim pmz ` nq´1 pmz ` n ` 1q´1
N Ñ8
mě1 n“´N
ÿ
“ lim ppmz ´ N q´1 ´ pmz ` N q´1 q.
N Ñ8
mě1
ÿ
´1
“ z lim ppm ´ N z ´1 q´1 ´ pm ` N z ´1 q´1 q
N Ñ8
mě1
ÿ
´1
“ ´z lim ppN z ´1 ´ mq´1 ` pN z ´1 ` mq´1 q.
N Ñ8
mě1

1 8
ř
Recall that π cotpπzq “ ` ppz ´ mq´1 ` pz ` mq´1 q.
z m“1
1 MODULAR FORMS FOR SL2 pZq 21

Hence equation (1.6) equals

´1 i ´1 i
´1 ´1 ´1 ´1 eπN z ` e´πN z
´z lim pπ cotpπN z q ´ zN q “ ´πiz lim “ ´πiz ´1
N Ñ8 N Ñ8 eπN z ´1 i ´ e´πN z ´1 i
as expected.

π
Corollary 1.18. Define G
r 2 pzq :“ G2 pzq ´ for z “ x ` iy P H. Then G r 2 satisfies the
y
condition (b’) in Remark 1.7, and hence the transformation property in Definition 1.5.

Proof. Indeed, we have

π π
G
r 2 pz ` 1q “ G2 pz ` 1q ´ “ G2 pzq ´ “ G
r 2 pzq,
y y

π πz z̄ πz
G
r 2 p´1{zq “ G2 p´1{zq´ “ z 2 G2 pzq´2πiz´ “ z 2 G2 pzq´ p2yi`z̄q “ z 2 G
r 2 pzq.
Imp´1{zq y y

Remark 1.19. The function G r 2 is an almost holomorphic modular from. By def-

inition, an almost holomorphic modular form is a function on the upper half plane
which are polynomials in 1{y whose coefficients are holomorphic functions on z. The
holomorphic part of an almost holomorphic modular from, for example G2 , is called a
quasi-modular form.

G2 8
σ1 pnqq n . It generalizes the definition of E2k ,
ř
We define E2 :“ “ 1 ´ 24
2ζp2q n“1
k ě 2 (see equation 0.3). Then E2 verifies

6iz
E2 p´1{zq “ z 2 E2 pzq ´ . (1.6)
π

We can use this equation to show that ∆ is a cusp form of weight 12 and level SL2 pZq.
8
2πiz ś
We define ηpzq :“ e 24 p1 ´ q n q, the Dedekind eta function.
n“1
We may define log, a holomorphic function on Repsq ą 0, such that log1 “ 0. Or
more precisely, let z “ reiθ with r ą 0, θ P p´π{2, π{2q, we define logpzq :“ logprq ` iθ.
?
We can then define ? the square root on the above right half plane by z :“ elogpzq{2 . In
particular, we have 1 “ 1.

Lemma 1.20. We have: ?

ηp´1{zq “ ´izηpzq.
1 MODULAR FORMS FOR SL2 pZq 22

8
logp1 ´ q n q converges absolutely when |q| ă 1. We get:
ř
Proof. We use the fact that
n“1

8
d πi ÿ nq n´1 q
logpηpzqq “ ´ 2πi
dz 12 n“1
1 ´ qn
8 ÿ 8
πi ÿ
“ ´ 2πi nq n q mn
12 n“1 m“0
8 ÿ 8
πi ÿ
“ ´ 2πi nq mn
12 n“1 m“1
πi
“ E2 pzq.
12
d πi πi 6i πi 1 d ?
Hence logpηp´1{zqq “ z ´2 E2 p´1{zq “ pE2 pzq´ q “ E2 pzq` “ logp ´izηpzqq.
dz 12 12 ? πz 12 2z dz
In particular, ηp´1{zq is a constant multiple of ´izηpzq. Let z “ i we obtain that this
constant is nothing but 1.
8
Corollary 1.21. ∆ “ η 24 “ q p1 ´ q n q24 is a cusp form of weight 12 and level SL2 pZq.
ś
n“1

8
p1 ´ q n q24 converges absolutely and uniformly on
ś
We remark that the series q
n“1
any compact subset of the unit disc |q| ă 1. In particular, ∆ has no zero on H. Since
∆ has an zero at 8 of order 1, one can show that for any f P Sk pSL2 pZqq, we have
f {∆ P Mk´12 pSL2 pZqq. We shall use this to determine the structure of MpSL2 pZqq.

1.4 Modular curve for SL2 pZq

The quotient SL2 pZqzH inherits a quotient topology from H, which is the finest topology
on SL2 pZqzH such that the natural projection

π : H Ñ SL2 pZqzH

is continuous. Or equivalently, a subset of SL2 pZqzH is open if and only if its preimage
in H is open.

In particuler, let U be an open subet of H. Since the preimahe of πpU q is YγPSL2 pZq γU
which is open in H, we know πpU q is open in SL2 pZqzH. Here we notice that the action
of SL2 pZq on H is continuous and hence each γU is an open subset of H.

We let Γ :“ SL2 pZq in this section. We call Y pΓq :“ SL2 pZqzH the modular curve
(attached to Γ).
Lemma 1.22. The group Γ acts discontinuously on H, i.e., for any K1 , K2 Ă H compact,
we have
#tγ P Γ | K2 X γpK1 q ‰ ∅u ă 8.
1 MODULAR FORMS FOR SL2 pZq 23

Proof. We first prove that there are only finitely many number of possible bottom rows
of γ P SL2 pZq such that K2 X γpK1 q ‰ ∅.
Indeed, let ε ą 0 and C
ˆ ą 0˙such that C ą Impzq ą ε for any z P K1 or K2 . For any
a b
z “ x ` iy P K2 and γ “ P SL2 pZq, if γz P K1 , then
c d

ε ă Impγzq “ |cz ` d|´2 Impzq ă |cz ` d|´2 C.

Hence |cz ` d|2 “ pcx ` dq2 ` pcyq2 ă C{ε. Therefore there are only finitely many
number of possible c as y ą ε, and for each possible c, there are only finitely many
number of possible d.

It remains to prove that for each fixed bottom row, there exists only finitely many
number of possible γ P SL2 pZq with this fixed bottom row such that K2 X γpK1 q ‰ ∅,
we fix one such γ if it exists (of course there is nothing to say if such γ does not exist). If
γ 1 is another element
ˆ in˙ SL2 pZq which has theˆ fixed˙bottom row then there exists n P Z
1 n 1 n
such that γ 1 “ γ. Then γ 1 pK1 q “ γpK1 q “ γpK1 q ` n. Since K2 is
0 1 0 1
compact, there are only finitely many number of possible n such that pγpK1 q ` nq X K2
is nonempty which finishes the proof.

Definition 1.23. For any z P H, define Γz :“ tγ P Γ, γz “ zu, the stabilizer of z in Γ.

One can show that for Γz is a subgroup of Γ and Let z P H and γ P Γ. Prove that
Γγz “ γΓz γ ´1 for any γ P Γ.

Corollary 1.24. (a) For any z P H, Γz is discrete in H.

(b) For any z P H, Γz is finite.

(c) For any z, w P H, there exist an open neighbourhood U of z and an open neigh-
bourhood V of w, such that for any γ P Γ, if γpU q X V ‰ ∅, then γpzq “ w.
Consequently, the quotient ΓzH is Hausdorff.

Proof. (a) Let K1 “ K2 “ K be any open neighbourhood of z with compact closure.

The set tγ P Γ | γz P Ku is a subset of tγ P Γ | K X γpKq ‰ ∅u, hence is finite.
Therefore Γz X K is finite. For each w P Γz X K with w ‰ z, there
Ş exists Uw an
open neighbourhood of z such that w R Uw . Let Uz “ K X Uw . Then
wPΓzXK,w‰z
Uz is an open neighbourhood of z such that Uz X Γz “ tzu. The discreteness is
clear.

(b) It is enough to take K1 “ K2 “ tzu in the previous lemma.

(c) Let K (resp. L) be any open neighbourhood of z (resp. w) with compact closure.
By the previous lemma, the set S :“ tγ P Γ | L X γK ‰ ∅, γz ‰ wu is finite.
For each γ P S, since γz ‰ w, then there exists an open neighbourhood Uγ of z
and an open neighbourhood Vγ of w such that γUγ X Vγ “ ∅.
1 MODULAR FORMS FOR SL2 pZq 24

Ş Ş
Let U “ K X Uγ and V “ L X Vγ . If γ P Γ such that γpU q X V ‰ ∅, then
γPS γPS
γ R S and hence γz “ w as expected.
Consequently, if Γz ‰ Γw, we take U and V as above, then there is no γ P Γ
such that γz “ w and hence for all γ P Γ we have γpU q X V “ ∅. In particular,
πpU q X πpV q “ ∅ and πpU q and πpV q are disjoint open neibourhoods of Γz and
Γw respectively in ΓzH. In other words, the quotient ΓzH is Hausdorff.

Definition 1.25. We define the period of an element z in H by #t˘IuΓz {t˘Iu, denoted

by hz . If hz ‰ 1, we say that z is an elliptic point for Γ.
Note that different elements in one orbit have the same period by the previous exercise.
We can hence define the period for an element Γz inside the modular curve SL2 pZqzH
by hΓz :“ hw for any w P Γz. We say Γz is elliptic if its period is not 1.
Proposition 1.26. We define the region D :“ tz P H | |Repzq| ď 1{2, |z| ě 1u. Then the
projection π induces a surjective map D Ñ SL2 pZqzH.
Moreover, for z1 ‰ z2 P D, we have Γz1 “ Γz2 if and only if either

(i) |Repz1 q| “ |Repz2 q| “ 1{2, |z1 ´ z2 | “ 1; or

(ii) |z1 | “ |z2 | “ 1 and z2 “ ´1{z1 .

Such region D is called a fundamental domain for Γ.

To show the map is surjective, one first shows that the ˆ

imaginary
˙ part of Γz reaches
a b
its maximum. We can do that by showing that for γ “ , if c or d tends to 8
c d
then Impγzq tends to zero. We can then take γ0 P γ such that Impγ0 zq ě Impγzq for
1 MODULAR FORMS FOR SL2 pZq 25

ˆ ˙
1 n
any γ P Γ. We may assume that |Repγ0 zq| ď 1{2 by replacing γ0 with γ for
0 1 0
suitable n P Z. It remains to show that |γ0 z| ě 1. We leave the full proof as an exercise.

Proposition 1.27. For any z P H, there exists an open neighbourhood Uz of z, which is

stable under the action of Γz , such that for any γ P Γ, if γUz X Uz ‰ ∅, then γ P Γz .
Moreover, the open set Uz contains no elliptic point other than z.
Consequently, the natural projection π on Uz factors through Γz zUz and induces a
homeomorphism from Γz zUz onto an open subset of ΓzH. In particular, if z P H is not
an elliptic point for Γ, then the projection gives a homeomophsim from Uz to πpUz q.

Proof. Fix any z P H. Let K be any open neighborhood of z with compact closure. Let
S :“ tγ P Γ | K X γK ‰ Hu. By Lemma 1.22, S is finite.
For any δ P S such that δz ‰ z, we can choose Vδ , an open neighborhood of z, and
Wδ , an open neighborhood of δz, such that Vδ X Wδ “ H.
Then Uδ :“ Vδ X δ ´1 Wδ is an open neighborhood of z.
Ş
Let U :“ K X δPS,δz‰z Uδ be an open neighborhood of z. For any γ P S, if
γU X U ‰ H, then γK X K ‰ H and γUδ X Uδ ‰ H for any δ P S such that δz ‰ z.
The first part implies that γ P S and the second part implies that γ ‰ δ for δz ‰ z.
Necessarily we have γz “ z.
Ş
To construct an open neighborhood invariant under Γz , we take Uz :“ γPΓz γU .
Since Γz is finite, it is then an open neighborhood of z which satisfies the conditions in
the statement.
Now let w P Uz be an elliptic point. There exists γ P Γ such that γw “ w. In
particular, w
ˆ P Uz˙X γUz and the latter is non-empty. Hence γ P Γz , i.e., γz “ z as well.
a b
Write γ “ . Then both z and w are imaginary roots with positive imaginary
c d
part of the real polynomial cz 2 ` pd ´ aqz ´ b “ 0. They must be equal.
Finally, it is easy to see that the restriction of π to Uz factors through Γz zUz . For
w, w1 P Uz , if Γz w and Γz w1 has the same image under π, then Γw “ Γw1 . In particular,
there exists γ P Γ such that γw “ w1 . Hence Uz X γUz contains w1 and hence non-empty.
In particular γ P Γz and Γz w “ Γz w1 . We deduce that π induces a bijective map from
Γz zUz to its image. The bi-continuity is clear.

Lemma 1.28. If z P D is an elliptic point for Γ, then z “ i, ρ or ρ ` 1 where ρ :“ e2πi{3

is a cubic root of 1. Equivalently, SL2 pZqzH has only two elliptic points, one is Γi who
has period 2, the other is Γρ who has period 3. In both cases, Γz is cyclic.
ˆ ˙
a b
Proof. Let z “ x ` yi P D be an elliptic point. Take γ “ P Γz ´ t˘Iu. By
c d
multiplying ´I, we may assume that c ě 0.
The relation γz “ z implies that cz 2 ` pd ´ aqz ´ b “ 0. In particular, since z is not
real, we have c ‰ 0 and pd ´ aq2 ´ 4bc “ pd ` aq2 ´ 4 ă 0. Hence c ą 0 and |a ` d| “ 0
1 MODULAR FORMS FOR SL2 pZq 26

or 1.
a
4 ´ pd ` aq2 1 ?
Moreover, we have y “ ď . Since z P D, we have y ě 3{2 and
2c c
hence c must be 1.
?
If |a ` d| “ 1, then y “ 3{2 and z must be ρ or ρ ` 1.
a´d
If |a ` d| “ 0, then y “ 1. Moreover, x “ “ a. Since z P D, we get a “ 0.
2
Hence z “ i as expected.

It remains to calculate their periods. For z “ i, note that γz “ z if and only if a “ d

and b “ ´c. Then detpγq “ ad ´ bc “ a2 ` b2 “ 1. Hence either a “ 0 or b “ 0. If a “ 0,
then γ “ ˘S. If a “ 1, then γ “ ˘I. Hence z has period 2. Moreover, it is clear that
Γz is generated by S in this case.
For z “ ρ, note that γz “ z ðñ aρ ` b “ cρ2 ` dρ ðñ pa ` c ´ dqρ ` pb ` cq “ 0 ðñ
d “ a ` c “ d, b “ ´c. Hence detpγq “ ad ´ bc “ apa ` cq ` c2 “ pa ` c{2q2 ˆ ` 3c2 {4 ˙
“ 1.
0 ´1
Therefore c “ ˘1 or 0. If c “ ˘1, then a “ 0 or ´c. In this case, γ “ ˘ or
1 1
ˆ ˙
´1 ´1
˘ . If c “ 0, then a “ ˘1. In this case, γ “ ˘I. We conclude that the period
1 0
ˆ ˙
0 ´1
of ρ is 3. Moreover, it is clear that Γz is generated by in this case.
1 1
Remark 1.29. Let Y be a topological space and n be a positive integer. An n-dimensional
complex structure on Y is a family pUz , ϕz qzPY , where Uz is an open neighborhood of
z in Y and ϕz is a homeomorphism from Uz onto an open subset of the unit disc of Cn
sending z to 0 such that the transition map between different ϕz ’s are holomorphic. We
call ϕz a local coordinate chart of Y and the family an atlas of charts.
We say a Hausdorff, second countable topological space is a complex manifold of
dimension n if it is endowed with an n-dimensional complex structure.
We remark that by a theorem of Radó (c.f. [?]), if a connected Hausdorff topological
space has a one dimensional complex structure then it is second countable. We call such
a space a Riemann surface, i.e., a Riemann surface is a connected Hausdorff space
endowed with a one dimensional complex structure.
For example, H is a Riemann surface. Let D be the unit disc of C. We can define
w´z
the local chart at z P H as δz : H Ñ D, w Ñ .
w´z
The quotient Y pΓq “ ΓzH inherits a complex structure from H. More precisely, for
any z P H, take Uz as in Proposition 1.27. We can define a local coordinate chart at
Γz P Y pΓq as
ϕz : πpUz q Ñ D, Γw ÞÑ pδz pwqqhz .

We now show that ϕz is well-defined and is injective. Let w1 , w2 P Uz . Then Γw1 “

Γw2 if and only if Γz w1 “ Γz w2 as in the last paragraph of the proof of Proposition 1.27.
This is equivalent to that δz Γz δz´1 pδz pw1 qq “ δz Γz δz´1 pδz pw2 qq. Note that δz Γz δz´1 is a
1 MODULAR FORMS FOR SL2 pZq 27

cyclic group of degree hz consisting of fractional linear map which fixes both 0 and 8.
We know δz Γz δz´1 is the set of linear functions z ÞÑ ζz where ζ runs over all roots of
unity of degree hz . In particular, we see that δz Γz δz´1 pδz pw1 qq “ δz Γz δz´1 pδz pw2 qq if and
only if pδz pw1 qqhz “ pδz pw2 qqhz which is equivalent to ϕpw1 q “ ϕpw2 q.
We conclude that ϕz induces a bijection from Uz to its image. The bi-continuity is
clear. We still need to show that this family is compatible. We leave the details to the
reader. We recall that Uz does not contain any elliptic point other than z.
Finally, it is easy to see that Y pΓq “ πpDq is connected since D is. We conclude
that Y pΓq is a Riemann surface.

Remark 1.30. We actually have a more natural choice for the complex structure of the
quotient ΓzH. We first take pUz , δz qzPH as a local chart of H. We define
ź
ψz : Γz zUz Ñ D, Γz w ÞÑ δz pγwq.
γPt˘IuΓz {t˘Iu

One can show that it is a homeomorphism onto the image. Recall that π induces a
homeomorphism from Γz zUz to πpUz q. We can show that pπpUz q, π ´1 ˝ψz qzPΓzH defines a
local coordinate chart on Y pΓq. For example, under this construction, the local coordinate
chart at i is w ÞÑ δi pwqδi pSwq “ ´δi pwq2 and the local coordinate chart at ρ is w ÞÑ
δρ pwq3 .
Such a construction can be extended in the same way to the quotient of a Riemann
surface by a discrete group where the action is discontinuous and effective (c.f. chapter
3 of [Mir95]). Here we say an action of a group G on a space X is effective if the only
element in G which acts trivially on the whole space is the identity. Note that Γ does
not act effectively on H but Γ{t˘Iu does.

Apparently Y pΓq is not compact. To compactify it, we consider the extended upper
half plane
H˚ :“ H Y P 1 pQq “ H Y Q Y t8u.
ˆ ˙
a b
For γ “ P SL2 pZq and rx : ys P P 1 pQq, we define
c d

γrx : ys “ rax ` by, cx ` dys.

The elements in ΓzP 1 pQq are called cusps.

Remark 1.31. (topological structure of H˚ ) For M ą 0, we define NM :“ tz P C |

Impzq ą Mu Ă H and NM ˚ :“ N ˚ ˚
M X t8u Ă H . We let tNM uM ą0 be a family of open
˚
basis at 8 in H .

For any α P Q, there exists δα P SL2 pZq, such that δα α “ 8 (exercise). We let
tα´1 pNM
˚ qu
M ą0be a family of open basis at α P H˚ .
1 MODULAR FORMS FOR SL2 pZq 28

One can show that (exercise) α´1 pNM q is an open disc tangent to the real axis at α.

We remark that H˚ is a subset of the complex projective line P1 pCq and hence inherits
a subset topology from P1 pCq. Notice that the topology above is not the same as the subset
topology. In fact, it is strictly finer than the subset topology.

We define XpΓq :“ ΓzH˚ , a topological space with the quotient topology, also called
a modular curve. Note that XpΓq “ Y pΓq X tΓ8u since Γ “ SL2 pZq acts transitively on
P 1 pQq. We still use π to denote the natural projection H˚ Ñ ΓzH˚ .
Lemma 1.32. For any z P H and γ P SL2 pZq, we have Impγzq ď maxtImpzq, 1{Impzqu.

We leave the proof as an exercise.

Corollary 1.33. The quotient XpΓq is Hausdorff.

Proof. We only need to check that πp8q can be separated from other points. More
precisely, we need to show that for any z P H, there exists M ą 0 such that ΓzXNM “ ∅.
Indeed, let M ą maxtImpzq, 1{Impzquq and the above relation holds by the previous
lemma.
Exercise 1.34. Prove that XpΓq is compact (hint: use the fundamental domain).

The extended upper half plane H˚ does not have a complex structure itself, but the
quotient XpΓq has.
ˆ ˙
1 n
Exercise 1.35. The group Γ8 , stabilizer of 8 in Γ, equals ˘t | n P Zu.
0 1
Lemma 1.36. The open subset N2˚ is stable under Γ8 . The natural projection factors
through Γ8 zN2˚ and gives a homeomorphism from Γ8 zN2˚ onto an open subset in XpΓq.

Proof. The first part is obvious. We now prove the second part.
Since N2˚ is open, πpN2˚ q is open. It remains to show that π induces a homeomor-
phism from Γ8 zN2˚ “ Γ8 zN2 Y tΓ8 8u to πpN2 q Y tΓ8u.
The
ˆ surjectivity
˙ is clear. We now show the injectivity. Let z, w P N2 . If there exists
a b
γ“ P SL2 pZq, such that γz “ w, by Lemma 1.32, we know Impwq “ Impγzq ď
c d
Impzq and similarly, Impzq “ Impγ ´1 wq ď Impwq. We get Impzq “ Impwq. Recall that
Impwq “ Impzq{|cz ` d|2 . Hence |cz ` d| “ 1. Note that Impzq ą 2, necessarily we have
c “ 0 and |d| “ 1. Hence γ P Γ8 as expected. The bicontinuety is also clear.

Remark 1.37. (complex structure at cusps) It is easy to see that q : H˚ Ñ D, z ÞÑ

e2πiz induces a homeomorphism from Γ8 zN2˚ onto the disc tz P C | |z| ă e´4π u. Hence
q gives a local coordinate at Γ8 P XpΓq.
The modular curve XpΓq is then a compact, connected 1-dimensional complex mani-
fold. In other words, XpΓq is a compact Riemann surface. We shall show that XpSL2 pZqq
is isomorphic to P 1 pCq and Y pSL2 pZqq is isomorphic to A1 pCq as complex manifolds.
1 MODULAR FORMS FOR SL2 pZq 29

1.5 Dimension formula: SL2 pZq-case

Definition 1.38. Let X be a 1-dimensional complex manifold. We may define the notion
of holomorphic or meromorphic function on X via the local charts.
„
More precisely, we write tUx , ϕx : Ux Ý
Ñ Vx uxPX for the atlas of local charts. Here
Vx is an open subset of the unit disc D. We may assume that ϕx pxq “ 0.
A function f : X Ñ C is called holomorphic if for each x, the function f ˝ ϕ´1
x is a
holomorphic function on Vx . A meromorphic function on X is a function f defined
on X except a set of isolated points, say Σ, such that f ˝ ϕ´1
x is a meromorphic function
on Vx with poles in Vx X ϕx pΣq.
an tn
ř
Let f be meromorphic. For each x, f ˝ ϕ´1x has Laurent series of the form
n"´8
where t is the local coordinate at x. If f is not constantly zero on any connected com-
ponent, then for any x, there exists n P Z such that an ‰ 0. In this case, we define
vx pf q :“ mintn | an ‰ 0u. In other words, vx pf q is the order of zero of f at x, or minus
the order of pole of f at x.

For example, let z1 , ¨ ¨ ¨ , zk , w1 , ¨ ¨ ¨ , wl be different points in C. Let n1 , ¨ ¨ ¨ , nk ,

m1 , ¨ ¨ ¨ , ml be positive integers. Consider the rational function on P 1 pCq
k
pz ´ zi qni
ś
i“1
f pzq :“ l
.
ś
pz ´ wj qmj
j“1

Then vzi pf q “ ni and vwj pf q “ ´mi for each i and j. Note that t “ 1{z is the local
coordinate at 8 and
k k
p1{t ´ zi qni p1 ´ tzi qni
ś l k
ś
ř ř
mj ´ ni
i“1 i“1
f ptq “ l
“ tj“1 i“1
l
.
ś ś
p1{t ´ wj qmj p1 ´ twj qmj
j“1 j“1

l
ř k
ř ř
We get v8 pf q “ mj ´ ni . In particular, we see that vx pf q “ 0.
j“1 i“1 xPP 1 pCq

Recall a Riemann surface is a connected complex manifold of dimension 1.

Proposition 1.39. Let X be a compact Riemann manifold and f ‰ 0 be a meromorphic
function on X. Then the set tx P X | vx pf q ‰ 0u is finite. Moreover, we have
ÿ
vx pf q “ 0.
xPX

Proof. For any meromorphic function on complex manifold, the set of zeros and poles
is discrete and closed, hence is finite if the manifold is compact.
If f 1 “ 0, then f is a constant and the equation holds automatically. We assume in
the following that f 1 ‰ 0 and hence the set of zeros and poles of f 1 is also finite.
1 MODULAR FORMS FOR SL2 pZq 30

Now let C be a closed rectifiable curve in X which does not intersect any poles or
zeros of f 1 or f . Then XzC has two connected component, say U and V . We may assume
that C is positively oriented as boundary of U , and negatively oriented as boundary of
V.
By Cauchy’s residue theorem, we have:
1 f1 f1 f1
ż ÿ ÿ
“ Resz“x “ ´ Resz“x .
2πi C f xPU
f xPV
f

8
an tn with am ‰ 0
ř
Fix x P X and let m :“ vx pf q. We can then write f ˝ ϕ´1
x “
n“m
where ϕx is the local chart at x, and t is the local coordinate at x. Then
8
nan tn´1
ř
f1 n“m m
˝ ϕ´1
x ptq “ 8 P ` Crrtss.
f ř t
an t n
n“m

f1
Hence Resz“x “ vx pf q. We conclude that
f
ÿ ÿ ÿ
vx pf q “ vx pf q ` vx pf q “ 0.
xPX xPU xPV

We refer to Proposition II.4.12 of [Mir95] for another proof. See also Corollary
IV.3.18 of the loc.cit.
Corollary 1.40. Let X be a compact Riemann surface. Then the only holomorphic func-
tions on X are the constant functions.

Proof. Let c be any value of f . Then f ´ c is a holomorphic function on X which has

zero but no pole. By Proposition 1.39, f ´ c must be constantly 0.

Proposition 1.41. (a) M0 pSL2 pZqq “ C and S0 pSL2 pZqq “ t0u.

(b) If k ă 0, or if k ą 0 odd, then Mk pSL2 pZqq “ t0u
(c) For k ě 0 even, if k ı 2 mod 12, then
« ff « ff
k k
dim Mk pSL2 pZqq “ ` 1, and dim Sk pSL2 pZqq “ .
12 12

Consequently, we have E43 ´ E62 “ 1728∆.

(d) For k ě 0 even and k ” 2 mod 12, we have M2 pSL2 pZqq “ t0u and
« ff « ff
k k
dim Mk pSL2 pZqq “ , dim Sk pSL2 pZqq “ ´ 1 if k ą 2.
12 12
1 MODULAR FORMS FOR SL2 pZq 31

Proof. (a) Note that a modular form of weight 0 is nothing but a holomorphic function
on the compact Riemann surface XpΓq, and hence must be constant by the previous
corollary.

(b) If k ă 0, let f P Mk pSL2 pZqq. Then f 12 P M12k pSL2 pZqq and hence ∆´k f 12 P
S0 pSL2 pZqq “ t0u. Therefore f “ 0 as expected.
The case k odd is already shown in the exercise. One only needs to notice that
´I P SL2 pZq.

(c) Recall that ∆ is a cusp form of weight 12 which has no zero on H and has rank 1 at
8. We can verify easily that for all k, the following linear map is an isomorphism:

Mk´12 pSL2 pZqq Ñ Sk pSL2 pZqq, f ÞÑ ∆f. (1.7)

À
Moreover, recall that for any k ě 4 even, Mk pSL2 pZqq “ Sk pSL2 pZqq CEk . We
get for all k ě 4 even,

dim Mk pSL2 pZqq “ dim Sk pSL2 pZqq ` 1 “ dim Mk´12 pSL2 pZqq ` 1.

Hence by recurrence we only need to prove (c) for 0 ď k ď 10, k ‰ 2. The case
k “ 0 are known by (a). For 4 ď k ď 10, k even, we have dim Mk pSL2 pZqq “
dim Mk´12 pSL2 pZqq ` 1 “ 0 ` 1 “ 1 by (b).

(d) For the last equation, notice that both E43 ´ E62 and ∆ are in S12 pSL2 pZqq which
is one dimensional. By comparing the first term of the Fourier expansions we get
E43 ´ E62 “ 1728∆.
Recall that E4 pρq “ 0. Hence E6 pρq ‰ 0 otherwise ∆pρq “ 1728pE43 pρq´E62 pρqq “ 0
which is impossible.
We can now prove that M2 pSL2 pZqq “ t0u. Indeed, let f P M2 pSL2 pZqq, then
f E4 P M6 pSL2 pZqq “ CE6 . There exists λ P C such that f E4 “ λE6 . In
particular, f pρqE4 pρq “ λE6 pρq. Hence λ “ 0 an f “ 0 as expected.
We can then apply equation (1.7) and get the dimension formula for k ą 2, k ” 2
mod 12 recurrently.

We can now deduce Proposition 0.7 easily, for example E4 ˚ E4 “ E8 and E4 ˚ E6 “

E10 , as well as:
3 1 1
Exercise 1.42. Prove that E6 “ E4 E2 ´ E4 and E8 “ E6 E2 ´ E61 .
2πi πi

6iz
(Recall equation (1.5): E2 p´1{zq “ z 2 E2 pzq ´ .)
π
À
Exercise 1.43. The natural map CrE4 , E6 s Ñ Mk pSL2 pZqq is surjective. Hence the
kPZ
algebra à
Mk pSL2 pZqq “ CrE4 , E6 s – CrX, Y s.
kPZ
1 MODULAR FORMS FOR SL2 pZq 32

(Hint: use equation (1.7) and the fact that any even integer ě 4 can be written as
4a ` 6b with a, b ě 0.)

Definition 1.44. Let X be a compact Riemann surface, we define CpXq to be the field of
meromorphic functions on X.

Lemma 1.45. The meromorphic functions on P 1 pCq are the rational fractions, i.e. CpP 1 pCqq “
Cpzq.

Proof. Let f ‰ 0 be a meromorphic function on P 1 pCq. Let z1 , ¨ ¨ ¨ , zN be the zeros or

N
pz ´ zi qvzi pf q . Then f {g has no pole or zero on H. By
ś
poles of f in C. Define g :“
i“1
Proposition 1.39, it also has no pole or zero at t8u, hence must be a constant.

Recall that
E43 1
j“ “ ` 744 ` 196884q ` 21493760q 2 ` 864299970q 3 ` 20245856256q 4 ` ¨ ¨ ¨
∆ q

is a holomorphic function on H. Since both E43 and ∆ are modular forms of weight
12, we know j is a modular function of weight 0, and hence can be considered as a
meromorphic function on the compact modular curve XpSL2 pZqq “ SL2 pZqzH˚ .

Proposition 1.46. The j-function XpSL2 pZqq Ñ P 1 pCq defines an isomorphism of com-
pact Riemann surfaces.
Consequently, the field CpXpSL2 pZqqq “ Cpjq.

Proof. For any c P C, we have v8 pj ´ cq “ ´1. Hence by Proposition 1.39, there exists
a unique z P XpSL2 pZqq such that vz pj ´ cq “ 1. In other words, z is the unique zero of
j ´ c in XpSL2 pZqq. Hence j is a holomorphic bijection as expected.

Remark 1.47. Any compact Riemann surface X has an algebraic structure, i.e., it is iso-
morphic to the set of common zeros of a family of homogeneous polynomials in Pn pCq for
some positive integer n (c.f. [Gun66]) (we can in fact take n “ 3). By GAGA (Géome-
trie Algébrique et Géométrie Analytique) we know the algebraic structure is unique up
to birational maps and moreover, every meromorphic function on X is a rational map.
Recall that the map X ÞÑ CpXq defines an equivalence of categories between the category
of birational classes of curves over C and extensions of C of transcendence degree 1
([Sil85] II.2). Hence in order to determine the algebraic structure of X, we only need to
calculate CpXq.
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 33

2 Modular forms for congruence subgroups

2.1 Theta series revisited

Recall that
2
ÿ
θptq :“ e´tπn , t ą 0.
nPZ
converges absolutely for all t ą 0 and satisfies
?
θp1{tq “ tθptq.

We may consider t as a complex variable. It is easy to see that the above series converges
absolutely and uniformly on any compact subset of Repzq ą 0.
?
Recall that z can be extended analytically to the right half plane Repzq ą 0. The
functional equation ?
θp1{zq “ zθpzq
?
still holds when Repzq ą 0. Indeed, the holomorphic function θp1{zq ´ zθpzq has zeros
at Rą0 and hence must be zero on the domain of definition.

For z P H, Rep´2izq ą 0. Hence we may define

2
ÿ ÿ 2
Θpzq :“ θp´2izq “ e2πizn “ 1 ` 2 qn , q “ e2πiz .
nPZ ně1

Obviously Θpz ` 1q “ Θpzq for all z P H.

We can check (exercise) that Θ satisfies the following functional equation:
?
Θp´1{4zq “ ´2ziΘpzq,

and then deduce that ?

Θpz{p4z ` 1qq “ 4z ` 1Θpzq.
(Hint: recall that Θpz ` 1q “ Θpzq.)
ˆ ˙
1 0
In other words, Θp zq “ p4z ` 1q1{2 Θpzq.
4 1
We define gpzq :“ Θpzq4 , a holomorphic function on H. Then we have:
ˆ ˙
1 0
gpz ` 1q “ gpzq, gp zq “ p4z ` 1q2 gpzq.
4 1
ˆ
˙ ˆ ˙
1 1 1 0
Let Γ be the subgroup of SL2 pZq generated by , and ´I. Then g
0 1 4 1
satisfies the weight 2 transformation property in Definition 1.5 for all γ P Γ. Obviously,
g is holomorphic on H and os 1-periodic whose Fourier expansion has no negative terms.
Moreover, the Fourier coefficients of g are of polynomial growth. In this case, we say g
is a modular form of weight 2 and level Γ.
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 34

2.2 Basic definition

Let N be a positive integer. The natural projection Z Ñ Z{N Z induces a group mor-
phism:
SL2 pZq Ñ SL2 pZ{N Zq. (2.1)
ˆ ˙
1 0
We define its kernel as ΓpN q :“ tγ P SL2 pZq | γ ” mod N u.
0 1

Exercise 2.1. (a) Show that the map (2.1) is surjective. Deduce that ΓpN q is a normal
subgroup of SL2 pZq with index #SL2 pZ{N Zq.
(Hint: for any c, d P Z with pc, d, N q “ 1. Write c “ pep . We can split c “ c1 c2
ś
ś ep ś ep p
where c1 “ p and c2 “ p . In particular, pc2 , dq “ 1, pc1 , c2 q “ 1.
p|pc,dq p∤pc,dq
Check that pc, d ` c2 N q “ pc1 c2 , d ` c2 N q “ 1. )

(b) Prove that #SL2 pZ{N Zq “ N 3

ś
p1 ´ p´2 q where p runs over the primes which
p|N
divides N . (Hint: calculate the case N “ pe by induction on e first.)

Definition 2.2. A subgroup of SL2 pZq is called a congruence subgroup if it contains

ΓpN q for some N . In this case, we say it is of level N . In particular, it is finite order
in SL2 pZq.

Exercise 2.3. (a) Let γ P SL2 pZq. Show that if Γ is a congruence subgroup of level N ,
then γΓγ ´1 is also a congruence subgroup of level N .

(b) The intersection of two congruence subgroups is still a congruence subgroup.

There are two important classes of congruence subgroups:

ˆ ˙
a b
Γ0 pN q :“ tγ “ P SL2 pZq | c ” 0 mod N u;
c d
ˆ ˙
a b
and Γ1 pN q :“ tγ “ P SL2 pZq | a ” 1, c ” 0, d ” 1 mod N u.
c d

We have ΓpN q Ă Γ1 pN q Ă Γ0 pN q Ă SL2 pZq.

Exercise 2.4. (a) Prove that both Γ0 pN q and Γ1 pN q are subgroups of SL2 pZq.

(b) Show that the map ˆ ˙

ˆ a b
Γ0 pN q Ñ pZ{N q , ÞÑ d
c d
is a surjective group morphism with kernelś
Γ1 pN q. Deduce that Γ1 pN q is a normal
subgroup of Γ0 pN q with index ϕpN q :“ N p1 ´ p´1 q.
p|N
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 35

(c) Show that the map ˆ ˙

a b
Γ1 pN q Ñ Z{N, ÞÑ b
c d
is a surjective group morphism with kernel ΓpN q. Deduce that ΓpN q is a normal
subgroup of Γ1 pN q with index N .

(d) Prove that S ´1 Γ0 pN qS “ Γ0 pN qT , the transpose of Γ0 pN q. Deduce that Γ0 pN q

is not
ś normal in SL2 pZq. Finally show that the index of Γ0 pN q in SL2 pZq is
N p1 ` p´1 q (Hint: use (b), (c) and Exercise 2.1.)
p|N

In brief, we have
ΓpN q ◁ Γ1 pN q ◁ Γ0 pN q ă SL2 pZq.

ˆ ˙ ˆ ˙
1 1 1 0
Exercise 2.5. Prove that ´I, and generate Γ0 p4q.
0 1 4 1

Definition 2.6. Let Γ be a congruence subgroup of level N . Let f be a function from H

to C. We say f is a modular form (resp. cusp form) of weight k and level Γ if it
satisfies the following conditions:

(a) f is holomorphic on H;
ˆ ˙
a b
(b) (transformation property) for any γ “ P Γ, f |k γ “ f , or equivalently,
c d
f pγzq “ pcz ` dqk f pzq for any z P H;

(c) f is holomorphic (resp. zero) at any α P P 1 pQq.

We denote the space of modular forms and cusp forms of weight k and level Γ by Mk pΓq
and Sk pΓq respectively.

We need to explanation partˆ(c) above.

˙ We first look at the point 8. Since Γ is
1 N
congruence of level N , we know P Γ.
0 1
ˆ ˙
1 N
By part (b), we have f pz ` N q “ f p zq “ f pzq. Moreover, f is holomorphic
0 1
on H. Hence f has Fourier expansion:
ÿ
n
f pzq “ a n qN (2.2)
nPZ

where qN “ e2πiz{N . We say f is holomorphic at 8 if an “ 0 for n ă 0. In this case, we

write f p8q :“ a0 . In particular, we say f is zero at 8 if an “ 0 for all n ď 0.
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 36

For α P Q, take δ P SL2 pZq such that α “ δ8. The function f |k δ satisfies the
transformation property for δ ´1 Γδ (exercise) which is still a congruence subgroup of
level N . We say f is holomorphic (resp. zero) at α if f |k δ is holomorphic (resp. zero)
at 8. One can show that (exercise) f |k δ does not depend on the choice of δ, and hence
the above definition of holomorphy is well-defined.
We can thus rewrite (c) as follows:

(c1 ) For any δ P SL2 pZq, the function f |k δ has Fourier expansion of the form f pzq “
8 8
n (resp. f pzq “ n ).
ř ř
a n qN an qN
n“0 n“1

Remark 2.7. Let α P P 1 pQq and γ P Γ. We take δ P SL2 pZq as above, namely α “ δ8.
For any γ P Γ, since f |k δ “ f |k γδ, we know that if f is holomorphic (resp. zero) at
α, then it is also holomorphic (resp. zero) at γα. Hence we only need to check pcq
in the above definition for any set of representatives for ΓzP 1 pQq. For example, when
Γ “ SL2 pZq, Γ acts transitively on P 1 pQq and we only need to check pcq for 8.

The C-vector spaces Mk pΓq and Sk pΓq have similar properties as in exercise 1.8. We
leave the reader to complete them.
When Γ is small, ΓzP 1 pQq may contain many elements. It is still not easy to write
down representatives of each orbits in ΓzP 1 pQq and then verify (c) in Definition 2.6 for
them. The following proposition allow us to simplify this condition.

Proposition 2.8. Let Γ be a congruence subgroup of level N . Let k ě 0 be an integer.

Let f be a function on H which satisfies conditions (a) and (b) of Definition 2.6. If f
is holomorphic at 8 and the Fourier coefficients are of polynomial growth, i.e. f pzq “
8
n such that there exist C, r ą 0 with |a | ď Cnr for all n, then f also verifies
ř
a n qN n
n“0
condition (c) of Definition 2.6. In other words, f is a modular form.

8
nr e´2πny{N .
ř
Proof. Write z “ x ` iy. Then |f pzq| ď |a0 | ` C
n“1

Write gptq :“ tr e´2πty{N , t ą 0. Then logpgptqq “ rlogptq ´ 2πty 1

N and logpgptqq “
r 2πy rN
t ´ N . We know that logpgq, and hence g, increases when t ă 2πy and decreases when
rN
t ą 2πy .We get
8 ż8
ÿ
r ´2πny{N 1 1
n e ď C p gptqdt ` r q.
n“1
y
0

Note that
ż8 ż8 ż8
r ´2πty{N C2 dt Γpr ` 1qC 2
gptqdt “ t e dt “ r`1 tr`1 e´t “ .
y t y r`1
0 0 0

By replacing C and r with larger numbers, we get |f pzq| ď C ` Cy ´r when y ą 1.

2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 37

ˆ ˙
a b
Then for any δ “ P SL2 pZq, we want to show that f |k δ is holomorphic at
c d
8. Equivalently, we need to show that

lim |pf |k δqpzqqN | “ lim |pf |k δqpzqe´2πy{N | “ 0 (2.3)

yÞÑ8 yÞÑ8

uniformly for 0 ď x ď N .
Recall that Impδzq “ y|cz ` d|´2 . Hence

|f |k δpzq| “ |pcz ` dq´k |f pδz| ď |cz ` d|´k pC ` Cy ´r |cz ` d|2r q.

Note that |cz ` d|2 “ pcx ` dq2 ` c2 y 2 . Hence f |k δpzq is of polynomial growth in y.
Equation 2.3 then follows.

We shall show later that the inverse is also true. For example, for Ek , the Fourier
coefficients are bounded by Cσk pnq, and hence of polynomial growth. In other words,
we may replace condition pcq of Definition 2.6 by:

(c2 ) The function f is holomorphic at 8 and the Fourier coefficients are of polynomial
growth.

Proposition 2.9. Let Γ1 Ă Γ be two congruence subgroups. Let γ P SL2 pZq. Let tβj u1ďjďn
n
Ů
be a set of representatives of Γ1 zΓ, i.e., Γ “ Γ1 βj .
j“1

(a) The map f ÞÑ f |k γ gives isomorphisms Mk pΓq – Mk pγ ´1 Γγq, and Sk pΓq –

Sk pγ ´1 Γγq.

(b) We have a natural inclusion Mk pΓq Ă Mk pΓ1 q and Sk pΓq Ă Sk pΓ1 q.

n
ř
(c) The function f ÞÑ f |k βj defines a surjective map from Mk pΓ1 q to Mk pΓq, and
j“1
a surjective map from Sk pΓ1 q to Sk pΓq, called the trace map.

To prove the transformation property in the last part, one can first show that
n
ř
f |k βj does not depend on the choice on the set of representatives and notice that
j“1
tβi γu1ďiďn is still a set of representatives for any γ P Γ. Moreover, one can use condition
(c’) instead of condition (c) in Definition 2.6. We leave the details to the reader.

Definition 2.10. For Γ a congruence subgroup, define Y pΓq :“ Γ and XpΓq “ ΓzH˚ .
Then Y pΓq is a complex manifold and XpΓq is a compact Riemann surface where the
compact structure is similar to the case SL2 pZqzH˚ . The proof is exactly the same as
the full level case.

Corollary 2.11. Let Γ be a congruence subgroup. Let k be an integer.

2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 38

(a) M0 pΓq “ C and S0 pΓq “ t0u.

(b) Mk pΓq “ t0u for k ă 0.

Proof. (a) A modular form of weight 0 is nothing but a holomorphic function on XpΓq,
hence must be constant by Corollary 1.40.

(b) Let f P Mk pΓq. Then f 12 P M12k pΓq. Note that ∆´k P S´12k pSL2 pZqq Ă S´12k pΓq.
Hence f 12 ∆´k P S0 pΓq “ t0u. We get f “ 0 as expected.

2.3 Some examples and applications

2.3.1 Theta series

Our first example is Θ4 . One can Deduce deduce directly from Proposition 2.8 and
Exercise 2.5 that it is a modular form of weight 2 and level Γ0 p4q.

π2 8
σ1 pnqq n q is a quasi-modular form of weight 2. We
ř
Recall that G2 pzq “ p1 ´ 24
3 n“1
have G2 pz ` 1q “ Gpzq and z ´2 G2 p´1{zq “ G2 pzq ´ 2πiz ´1 .
ˆ ˙
a b
One can verify easily (exercise) that for all γ “ P SL2 pZq, we have:
c d

2πic
pG2 |2 γqpzq “ G2 pzq ´ .
cz ` d
Definition 2.12. For N ě 1, we define G2,N pzq :“ G2 pzq ´ N G2 pN zq a holomorphic
function on z P H.

Exercise 2.13. The function G2,N is a modular form of weight 2 and level Γ0 pN q.

π2
In particular, we have G2,2 “ ´ p1 ` 24q ` ¨ ¨ ¨ q P M2 pΓ0 p2qq Ă M2 pΓ0 p4qq and
3
G0,4 “ ´π 2 p1 ` 8q ` ¨ ¨ ¨ q Ă M2 pΓ0 p4qq. We shall show later that dimpM2 pΓ0 p4qqq “ 2.
By observing the Fourier coefficients, we see that G2,2 and G2,4 are not fractional, hence
generate M2 pΓ0 p4q. Recall that Θ4 is also in this vector space. By comparing the
Fourier coefficients, we get:

Proposition 2.14. The C-vector space dimpMk pΓ0 p4qqq is generated by G2,2 and G2,4 .
1
Moreover, Θ4 “ ´ 2 G2,4 .
π
Corollary 2.15. Forřeach n ą 0, r4 pnq, the number of ways to write n into sum of four
squares, equals 8 d.
d|n,4∤d
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 39

2
8
q n . Hence Θ4 “ 1 `
ř ř
Proof. Recall that Θ :“ θp´2izq “ r4 pnq.
nPZ n“1
On the other hand,
8 8
1 1 ÿ
n 4 ÿ
´ G 2,4 “ ´ p1 ´ 24 σ 1 pnqq q ` p1 ´ 24 σ1 pnqq 4n q
π2 3 n“1
3 n“1
8
ÿ 8
ÿ
“ 1 ` 8p σ1 pnqq n ´ 4 σ1 pnqq 4n q.
n“1 n“1

We let σ1 paq “ 0 if a R Z. Then

ÿ ÿ
r4 pnq “ 8pσ1 pnq ´ 4σ1 pn{4qq “ 8p d ´ 4 dq
d|n d|n{4
ÿ ÿ ÿ
“ 8p d´ 4dq “ 8 d
d|n 4d|n d|n,4∤d

as expected.

2.3.2 An interlude: old forms

Recall that G2 is a quasi-modular form of weight 2 and level SL2 pZq “ Γ0 p1q, and the
function G2 pN zq appears in the definition of G2,N which is a modular form of level
Γ0 pN q. In general, one can construct a modular form from small level to big level in
this way.

`
˙ we note that GL2 pQq acts on H the same way as SL2 pZq does. For
Moreˆprecisely,
a b
any γ “ P GL2 pQq` , we may define the weight k operator by
c d

az ` b
f |k γ : H Ñ C, z ÞÑ pdet γqk{2 jpγ, zq´k f pγzq “ pdet γqk{2 pcz ` dq´k f p q
cz ` d
for any function f : H Ñ C. Here jpγ, zq “ cz ` d.
ˆ ˙ ˆ ˙
a 0 d 0 k
For example, for any a P Qˆ , f |k “ f , and f |k pzq “ d´ 2 f pdzq.
0 a 0 1
We also notice that this definition is compatible with Definition 1.3. Similarly, one
can show that the above definition defines a right action of GL2 pQq` on the space of
functions on H.

Lemma 2.16. Let M and d be two positive integers. Let N “ M d. If f : H Ñ C satisfies

the transformation property of weight k and level Γ0 pM q, then the map z ÞÑ f pdzq
satisfies the transformation property of weight k and level Γ0 pN q. One can replace Γ0
by Γ or Γ1 above.
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 40

ˆ ˙
d 0
Proof. The map z ÞÑ f pdzq equals d´k{2 f | k . We only need to show that
0 1
ˆ ˙
d 0
f |k is invariant under the weight k operator by Γ0 pN q.
0 1
ˆ ˙
A B
Indeed, for any P Γ0 pN q, we have
C D
ˆ ˙ˆ ˙ˆ ˙´1 ˆ ˙
d 0 A B d 0 A Bd
“ P Γ0 pM q.
0 1 C D 0 1 C{d D
ˆ ˙ ˆ ˙ ˆ ˙ ˆ ˙ ˆ ˙
d 0 A B A Bd d 0 d 0
Hence pf |k q|k “ pf |k q|k “ f |k as ex-
0 1 C D C{d D 0 1 0 1
pected.

Lemma 2.17. Let f be a holomorphic function on H which is moreover periodic with

integer period. We say f is holomorphic (resp. zero) at 8 if all its negative (resp.
non-positive) Fourier coefficients are zero.
Assume that for any γ P SL2 pZq, f |k γ is periodic with some integer period, then so
is f |k δ for any δ P GL2 pQq` .
In this case, if f |k γ is holomorphic (resp. zero) for any γ P SL, then so is f |k δ
for any δ P GL2 pQq`
˙ ˆ
a b
One can first show that any δ P GL2 pQq`
can be written as γ for some
0 d
γ P SL and a, b, d P Z. We leave the proof as an exercise.
The above two lemmas imply immediately that
Corollary 2.18. Let d and M be positive integers. The map f ÞÑ pz ÞÑ f pdzqq gives a
map from Mk pΓ0 pM qq to Mk pΓ0 pM dqq. It sends moreover cusp forms to cusp forms.
One can replace Γ0 by Γ or Γ1 .
Remark 2.19. Let f be a modular form of weight k and level Γ0 pM q. Then both f and
the function z ÞÑ f pdzq are in Sk pΓ0 pM dqq. We call them old forms.

2.3.3 Dedekind η function

2πiz 8
p1 ´ q n q, the Dedekind eta function, satisfies:
ś
Recall that ηpzq :“ e 24
n“1
?
ηp´1{zq “ ´izηpzq.
2πi
Note that ηpz ` 1q “ e 24 ηpzq. Hence η ”almost” satisfies the transformation property
of weight 1{2 and level Γ1 “ Γ1 p1q.

Let N be a positive integer. Then the function ηpzqηpN zq should ”almost” satisfies
the transformation property of weight 1 and level Γ0 pN q. We note that ηpz ` 1qηpN pz `
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 41

2πipN `1q
1qq “ e 24 ηpzqηpN zq.

If k is a positive integer such that kpN ` 1q “ 24, we have pηpz ` 1qηpN pz ` 1qqqk “
pηpzqηpN zqqk . Hence we could expect that ϕk :“ pηpzqηpN pzqqqk is a modular form of
weight k and level Γ0 pN q. We shall show that the values of ϕk at any point in H˚ is
minimal among all the elements in Sk pΓ0 pN qq. In particular, by Corollary 2.11 (c), we
get Sk pΓ0 pN qq (if k is even) or Mk pΓ1 pN qq (if k is odd) equals Cϕk .

For example, if N “ 1 and k “ 12, we get the ∆ function which is a cusp form of
weight 12 and level SL2 pZq. Recall ∆ has rank 0 at any point in H, and has rank 1 at
8. This implies that S12 pSL2 pZqq “ C∆.

Proposition 2.20. Let k and N be positive integers such that kpN ` 1q “ 24. Let Γ “
Γ0 pN q or Γ1 pN q. We define ϕk “ pηpzqηpN pzqqqk . If Sk pΓq ‰ 0, then Sk pΓq “ Cϕk .

Proof. For simplicity, we assume that Γ “ Γ0 pN q. The other case is similar. The idea is
to show that ϕk has minimal rank at every point in H˚ . Here we can use the fact that
the natural projection Γ1 pN qzH˚ Ñ Γ0 pN qzH˚ is unramified at all cusps, i.e., for any
α P P 1 pQq, a local coordinate at α in Γ0 pN qzH˚ can be lifted to a local coordinate at α
in Γ1 pN qzH˚ by the natural projection.
When pk, N q “ p12, 1q, ϕ12 “ ∆. The result is known. For the other cases, N is
always a prime number. We shall show in the exercise session that in this case (where
N is a prime) that Γ0 pN q has two cusps, one is Γ0ˆpN q8 with
˙ parameter q, the other
0 ´1
one is Γ0 pN q0 with parameter S ´1 ˝ qN . Here S “ such that S0 “ 8.
1 0
ˆ ˙
1 1
Assume that Sk pΓq ‰ 0 and let f be any non-zero element inside. Note that P
0 1
ˆ ˙ ˆ ˙
1 N 1 0
Γ, f is 1-periodic. Note also S S ´1 “ P Γ, we know that f |k S is
0 1 ´N 1
N -periodic.
If η satisfies the transformation property of weight 1{2 and level SL2 pZq, then ϕk is
a cusp form of weight k and level Γ0 pN q. It has minimal rank at all points among forms
in Sk pΓ0 pN qq. Indeed, it is non-zero on H and moreover

8 8
2πikpN `1q ź ź
z
ϕk “ e 24 rp1 ´ q n qp1 ´ q N n qs “ q rp1 ´ q n qp1 ´ q N n qs;
n“1 n“1
? ´k ? ´k
pϕk |k Sqzq “ z pηp´1{zqηp´N {zqq “ ´N rηpzqηpz{N qsk “
´k k
´N ϕk pz{N q
? 8
´k ź
n Nn
“ ´N qN rp1 ´ qN qp1 ´ qN qs.
n“1

In particular, f {ϕk is holomorphic on H˚ and is then a holomorphic function on the

2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 42

compact Riemann surface XpΓq. In particular, it is a constant. Hence Sk pΓq “ Cϕk .

Unfortunately η does not satisfies the transformation property of weight 1{2 and
2πi
level SL2 pZq as ηpz ` 1q “ e 24 ηpzq ‰ ηpzq. To solve this problem, we consider
ϕN
k
`1
“ pηpzqηpN zqq24 “ ∆pzq∆pN zq.

Note that both ∆pzq and ∆pN zq are in S12 pΓq. We have ϕkN `1 P S24 pΓq. Similarly
as before, one can show that it has minimal rank at all cusps.
Now for any 0 ď g P Sk pΓq, g N `1 P S24 pΓq. The ratio pg{ϕk qN `1 is a holomorphic
function on the compact Riemann surface XpΓq, and thus must be a non-zero constant.
Consequently, g{ϕk must be a non-zero constant as expected. In particular, ϕk is a
non-zero multiple of g and is hence in Sk pΓq, and Sk pΓq “ Cϕk .

Remark 2.21. We shall calculate the dimension of Sk pΓq later and we will get:

(a) If k is even, i.e., k “ 2, 4, 6, 8 or 12, then Sk pΓ0 pN qq “ Sk pΓ1 pN qq “ Cϕk

(b) If k is odd, i.e., k “ 1 or 3, then Sk pΓ0 pN qq “ 0 and Sk pΓ1 pN qq “ Cϕk .

2.4 Eisenstein series for ΓpN q

ř
Recall for k ě 2, G2k pzq “ pmz ` nq´2k and E2k pzq “ G2k pzq{2ζp2kq are the
pm,nqPZ2
Eisenstein series for SL2 pZq.
2
ˆ One ˙ can verify easily that the surjective map SL2 pZq Ñ tpc, dq P Z | pc, dq “ 1uu,
a b
ÞÑ pc, dq gives a bijection between SL2 pZq8 zSL2 pZq and tpc, dq P Z2 | pc, dq “
c d
1u{˘. We can hence rearrange the sum in G2k and get
ÿ ÿ
G2k pzq “ pdmz ` dnq´2k
dPZ,d‰0 pm,nqPZ2 ,pm,nq“1
ÿ
“ ζp2kq pmz ` nq´2k
pm,nqPZ2 ,pm,nq“1
ÿ
“ 2ζp2kq jpγ, zq´2k .
γPSL2 pZq8 zSL2 pZq

Consequently, we get:
ÿ
E2k pzq “ jpγ, zq´2k . (2.4)
γPSL2 pZq8 zSL2 pZq

Definition 2.22. Let N ě 1 be an integer and k ě 3 be an integer. When ´I P ΓpN q we

assume that k is even. Then for any δ P SL2 pZq, we can define:
ÿ
Ekδ pzq :“ jpγ, zq´k . (2.5)
γPΓpN q8 zΓpN qδ
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 43

Note that j is not invariant under the action of ´I, but j k is when k is even. Hence
it is natural to assume that k is even if ´I P ΓpN q. Moreover, we know that Mk pΓq “ 0
if k is odd and ´I P Γ. Hence there is no harm to exclude this case. We also remark
that ´I P ΓpN q if and only if N “ 1 or 2.

We want to show the Ekδ is a modular form of weight k and level ΓpN q.
Firstly, since ΓpN q8 zΓpN qδ Ă ΓpN q8 zSL2 pZq, we have
ÿ ÿ
|jpγ, zq´k | ď rSL2 pZq8 : ΓpN q8 s |jpγ, zq´k |
γPΓpN q8 zΓpN qδ γPSL2 pZq8 zSL2 pZq
ÿ
“ rSL2 pZq8 : ΓpN q8 s{2 |mz ` n|´k
pm,nqPZ2 ,pm,nq“1

which converges absolutely and uniformly on any compact subset of H. Hence Ekδ
defines a holomorphic function on H.
Secondly, Ekδ satisfies the weight k transformation property for ΓpN q. Indeed, for
any δ 1 P SL2 pZq, we have:
ÿ ÿ
Ekδ |k δ 1 “ rjpδ 1 , zqjpγ, δ 1 zqs´k “ rjpγδ 1 , zqs´k
γPΓpN q8 zΓpN qδ γPΓpN q8 zΓpN qδ
ÿ 1
“ ´k
jpγ, zq “ Ekδδ . (2.6)
γPΓpN q8 zΓpN qδδ 1

When δ 1 P ΓpN q, since ΓpN q is normal in SL2 pZq, we have ΓpN qδδ 1 “ δΓpN qδ 1 “
1
δΓpN q “ ΓpN qδ. Hence Ekδ |k δ 1 “ Ekδδ “ Ekδ .

Finally, we want to analyse the values of Ekδ at cusps. We first determine the cusps
of ΓpN qzH˚ .
Lemma 2.23. Let A, C, A1 , C 1 be integers such that pA, Cq “ 1 and pA1 , C 1 q “ 1. We
have:

(a) A{C “ A1 {C 1 if and only if pA, Cq “ ˘pA1 , C 1 q;

(b) ΓpN qA{C “ ΓpN qA1 {C 1 if and only if pA, Cq ” ˘pA1 , C 1 q mod N .

Proof. (a) If A{C “ A1 {C 1 and then AC 1 “ CA1 . Since pA, Cq “ 1, we have A | A1 .

Similarly, A1 | A and hence A “ ˘A1 as expected. The converse is trivial.
ˆ ˙
A B
(b) We may extend the column pA, Cq to a matrix δ “ and extend the
C D
ˆ 1 ˙
1 1 1 A B1
column pA , C q to a matrix δ “ , both in SL2 pZq.
C 1 D1
By 1 1
ˆ (a),
˙ ΓpN qA{C “ ΓpN qA {C if and only if there exists there exists γ “
a b
P ΓpN q, such that γδ has the same first column as δ 1 or ´δ 1 .
c d
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 44

This implies pA, Cq ” ˘pA1 , C 1 q mod N .

Conversely, if pA, Cq ” ˘pA1 , C 1 q mod N , we first look at the case where pA, Cq “
p1, 0q and pA1 , C 1 q ” p1, 0q mod N . Since A1 ” ˆ 1 1 mod N , ˙we may take b, d P Z
A bN
such that A1 d ´ C 1 b “ p1 ´ A1 q{N . Then γ “ is in ΓpN q, and γδ
C 1 1 ` dN
has the same column as δ 1 . If pA, Cq “ p1, 0q and pA1 , C 1 q ” ´p1, 0q mod N , we
repeat the previous construction for p´A1 , ´C 1 q then we get a δ P ΓpN q such that
γδ has the same column as ´δ 1 .
ˆ 1 ˙
1´1 D A ´ B1C ˚
In general, we consider δ δ “ . Since D1 A ´ B 1 C ” D1 A1 ´
C 1 A ´ A1 C ˚
B 1 C 1 “ ˘1 mod N and C 1 A ´ A1 C ” CA ´ AC “ 0 mod N , there exists γ 1 P
ΓpN q such that γ 1 has the same first column as δ 1´1 δ or ´δ 1´1 δ. Hence δ 1 γ 1 “
δ 1 γδ 1´1 δ 1 has the same first column as δ or ´δ. Let γ “ pδ 1 γδ 1´1 q´1 which is still
in ΓpN q. We obtain that γδ has the same first column as δ 1 or ´δ 1 as expected.

ˆ ˙
A B
Lemma 2.24. Let δ “ P SL2 pZq. Let εN “ 1 if N ě 3, and εN “ 1{2 if N “ 1
C D
or 2. Then ÿ
Ekδ pzq “ εN pcz ` dq´k . (2.7)
pc,dq“1,pc,dq”pC,Dq mod N

ˆ ˙
a b
Proof. We need to show that the map ÞÑ pc, dq gives a bijection between
c d
Γ8 pN qzΓpN qδ and tpc, dq P Z2 | pc, dq “ 1, pc, dq ” pC, Dq mod N u (resp. tpc, dq P
Z2 | pc, dq “ 1, pc, dq ” pC, Dq mod N uz˘) if N ě 3 (resp. if N “ 1 or 2). Note that
´I P Γ if and only if N “ 1 or 2.
ˆ ˙
1 nN
We only prove the case where N ě 3 here. In this case, Γ8 “ t | n P Zu.
0 1
ˆ ˙
1 nN
Firstly, the map is well defined since multiplying any element of the form
0 1
on the left does not affect the second row.

We then show the map is bijective. For the surjectivity, let pc, dq P Z2 such that
1
pc, dq “ 1, pc, dq ” pC, Dq mod N . Then pc, dq can be extended to a matrix δ P
SL2 pZq with second row pc, dq. Hence δ 1 δ ´1 P Γ1 pN q. Note that the natural map
ΓpN q8 zΓpN q Ñ Γ1 pN q8 zΓ1 pN q is bijective. We get the surjectivity.

The injectivity is easy to verify.

Definition 2.25. We define ΣN :“ tpC, Dq P pZ{N Zq2 | pC, Dq of order N u{˘ and SN
be a set of representatives of ΣN in tpC, Dq P pZ{N Zq2 | pC, Dq of order N u, i.e., for
any tpC, Dq P pZ{N Zq2 | pC, Dq of order N u, either pC, Dq P SN or p´C, ´Dq P SN .
Apparently the natural map from SN to ΣN is bijective.
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 45

ˆ ˙
A B
Remark 2.26. We see immediately that Ekδ ,
with δ “ P SL2 pZq, only depends
C D
on
ˆ pC, Dq˙ mod N . Conversely, for any pC, Dq P ΣN , we may lift it first to an ˆ element
˙
A B A B
P SL2 pZ{N Zq, and then lift it to en element δ in SL, still denoted by .
C D C D
Recall that the natural map SL2 pZq Ñ SL2 pZ{N Zq is surjective. We can then define
pC,Dq
ÿ
Ek :“ Ekδ “ εN pcz ` dq´k . (2.8)
pc,dq“1,pc,dq”pC,Dq mod N

pC,Dq p´C,´Dq
We see immediately that Ek “ p´1qk Ek .
We shall show that there are no other linear relations among the Eisenstein series.
More precisely, we shall show that there is a bijection between the cusps of XpΓpN qq
pC,Dq
and SN , and for each pC, Dq P SN , Ek is non-zero at the corresponding cusp and
pC,Dq
zero at other cusps. Consequently, the set tEk upC,DqPSN is linearly independant and
generates the space Mk pΓpN qqzSk pΓpN qq
Lemma 2.27. For each α P P1 pQq, we can write α in terms of C{D where C, D P Z with
gcdpC, Dq “ 1. The map α “ C{D ÞÑ pC, ´Dq mod N induces a bijection:

ΓpN qzP1 pQq Ñ ΣN .

Proof. We first show that the map is well-defined. Let C, D P Z with gcdpC, Dq “ 1. Let
l be the order of pC, Dq mod N . Then N | lC and N | lD hence N | l ˚ gcdpC, Dq “ l
and l “ N as expected.
Now if C 1 , D1 P Z with gcdpC 1 , D1 q “ 1 such that ΓpN qC{D “ ΓpN qC 1 {D1 , then by
Lemma 2.23, pC, Dq ” pC 1 , D1 q mod N . The map is then well-defined.
The injectivity is also corollary to Lemma 2.23. The surjectivity follows from the
surjectivity from SL2 pZq to SL2 pZ{NZq.

We identify ΣN with SN , and get a bijection:

Φ : ΓpN qzP1 pQq Ñ SN
ˆ ˙
A B
Proposition 2.28. Let δ “ P SL2 pZq. Then
C D
"
δ δ p˘1qk if pC, Dq ” ˘p0, 1q mod N ;
Ek p8q “ lim Ek pzq “
yÑ8 0 otherwise.

Consequently, for any α P P 1 pQq, we have

"
δ p˘1qk if α P ΓpN qp´D{Cq;
Ek pαq “
0 otherwise.
pC,Dq
Consequently, for any pC, Dq P SN , Ek is non-zero at the cusp Φ´1 ppC, Dqq, and is
zero at other cusps. In particular, they are linearly independent.
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 46

Proof. Recall that

ÿ
Ekδ pzq “ εN pcz ` dq´k .
pc,dq“1,pc,dq”pC,Dq mod N

(equation 2.7)
If pC, Dq ı ˘p0, 1q mod N , then for any pc, dq P Z2 with pc, dq “ 1 and pc, dq ”
pC, Dq mod N , we have c ‰ 0. Hence
ÿ
|cz ` d|´k
pc,dq“1,pc,dq”pC,Dq mod N
ÿ ÿ
´k
ď |cz ` d| “2 |cz ` d|´k (2.9)
c‰0,dPZ cě1,dPZ

We want to control the last term. We write z “ x ` iy. Since Ekδ is N -periodic, we may
assume that 0 ď x ď N . One can show easily that there exists C ą 0, such that for any
y ą 1,c ě 1 and d P Z, we have | yz ` cy
d d
| ą Csupt1, | cy |u ě C. Hence |cz `d| ą Ccy ě Cy
and |cz ` d| ě suptcy, |d|u ě suptc, |d|u.
Then
ÿ ÿ ÿ
|cz`d|´k “ |cz`d|´1{2 |cz`d|´k`1{2 ă pCyq´1{2 psuptc, |d|uq´k`1{2 Ñ 0
cě1,dPZ cě1,dPZ cě1,dPZ

when y Ñ 8.
When pC, Dq ” ˘p0, 1q mod N , we have a constant term where pc, dq “ p0, ˘1q
with value p´1qk . The sum of other terms tends to 0 as before.

For general α P P 1 pQq, take δα P SL2 pZq such that δα 8 “ α. Then

Ekδ pαq “ Ekδ |k δα p8q “ Ekδδα p8q. (2.10)

It remains to show that the second row of δδα is congruent to ˘p0, 1q modulo N if
and only if α P ΓpN qp´D{Cq. In fact, the second row of δδα is congruent to ˘p0, 1q
modulo N if and only if the second row of δ is congruent to ˘p0, 1qδα´1 modulo N .
ˆ ˙
a b
Write write δα “ , then α “ a{c, ˘p0, 1qδα´1 “ p´c, aq. Note that pC, Dq ”
c d
˘p´c, aq mod N if and only if pC, ´Dq ” ˘pc, aq mod N , if and only if ΓpN qα “
ΓpN qa{c “ ΓpN qp´D{Cq which completes the proof.
pC,Dq
Remark 2.29. The Fourier coefficients for each Ek are of polynomial growth.
Proposition 2.30. Let Eisk pΓpN qq be the C-vector space generated by the Eisenstein
pδq
series Ek , δ P SL2 pZq. We have a natural decomposition
à pC,Dq
Eisk pΓpN qq “ CEk .
pC,DqPSN

Moreover, à
Mk pΓpN qq “ Sk pΓpN q Eisk pΓpN qq.
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 47

Proof. The first part is clear, For the second part, take any f P Mk pΓpN qq, consider
ÿ f p´D{Cq pC,Dq
f´ pC,Dq
Ek .
pC,DqPSN Ek p´D{Cq

It is a modular form which is zero at all cusps, hence a cusp form as expected. We get
Mk pΓpN qq “ Sk pΓpN q ` Eisk pΓpN qq.
ř pC,Dq
It remains to show that the sum is direct. Indeed, if λpC,Dq Ek is a cusp
pC,DqPS
form for some λpC,Dq P C, then its value at the cusp ´D{C must be zero. Namely
pC,Dq
λC,D Ek p´D{Cq “ 0 and hence λpC,Dq “ 0 for all pC, Dq P SN as expected.

Definition 2.31. We define Ek pΓpN qq :“ Mk pΓpN qqzSk pΓpN qq, called the Eisenstein
space.
We write ε8 pΓpN qq for the number of cusps for ΓpN q.

Corollary 2.32. We have for any k ě 3, dim Ek pΓpN qq “ ε8 pΓpN qq.

2.5 Modular function and dimension formula

Let Γ be a congruence subgroup. Recall that XpΓq “ ΓzH˚ is a compact Riemann

surface. The aim of this section is to calculate the dimension of Mk pΓq and Sk pΓq for
general k.

Definition 2.33. For any z P H, we define hz :“ #t˘Γz uzt˘I2 u, called the period of z
for Γ. We say z is elliptic if hz ‰ 1.

Example 2.34. For any ellptic point, hz “ 2 or 3.

We then explain the complex structure at cusps. For the cusp 8, as the SL2 pZq-case,
Γ8 zN2˚ Ñ ΓzH˚ is a homeomorphism ˆ onto˙ an open subset. Note that t˘IuΓ8 is a finite
1 n
index subgroup of SL2 pZq8 “ ˘t | n P Zu. We obtain a positive integer h8 ,
0 1
ˆ ˙
1 h8 n
called the width of 8 in XpΓq, such that t˘IuΓ8 “ ˘t | n P Zu. The the
0 1
map z ÞÑ qh8 :“ e2πiz{h8 is a local coordinate at 8.

In general, let α P Q, take δα P SL2 pZq such that δα α “ 8. Then δα Γα δα´1 is a

finite index subgroup
ˆ of SL˙ 2 pZq8 . Hence there exists hα , a positive integer, such that
1 h α n
˘δα Γα δα´1 “ ˘t | n P Zu. We call hα the width of α in XpΓq. We can
0 1
show that it does not depend on the choice of δα (exercise). The automorphism δα
gives a homeomorphism Γα zδα´1 pN2˚ q Ñ δα Γα δα´1 zN2˚ . The map ϕα : z ÞÑ qhα is a local
coordinate on the right hand side, and hence ϕα ˝ δα is a local coordinate on the left
hand side.
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 48

Let f be a holomorphic
ˆ function
˙ on H which satisfies the transformation
ˆ ˙ property
1 hα 1 2hα
for Γ. We remark that might not in Γ, but its square is always in
0 1 0 1
Γ. So f |k δα is always 2hα -periodic.

Definition 2.35. Let f be a complex-valued function defined on H except a set of isolated

points. We say f is a modular function of weight k and level Γ if it satisfies the
following conditions:

(a) f is meromorphic on H;

(b) (transformation property) for any γ P Γ, f |k γ “ f ;

(c) f is meromorphic at cusps, i.e., for any δ P SL2 pZq, f |k δ has Fourier expansion
of the form ÿ
n
f |k δpzq “ an q2h (2.11)
n"´8

where h is the width of δ8. One can verify easily that mintn | an ‰ 0u only
depends on δ8. We hence define

vδ8 pf q :“ mintn | an ‰ 0u{2,

and called the valuation of f at Γδ8 P XpΓq.

Finally, we denote the space of modular function of weight k and level Γ by Ak pΓq.

The following lemma is easy and we leave the proof to the reader:

Lemma 2.36. Let z P H˚ and γ P Γ. Let f be a modular function of level Γ. It is easy

to see that vz pf q “ vγz pf q.

Since both the period and the valuation are Γ-invariant, although f is not a function
on XpΓq, we may define its valuation on XpΓq as follows:

Definition 2.37. Let f be a modular function of level Γ. For z P H, we define vΓz pf q :“

vw pf q
for any w P Γz. For α P P1 pQq, we define vΓα pf q :“ vα pf q.
hw

Note that when f P A0 pΓq, f is Γ-invariant and hence factors through ΓzH. It is
easy to see that for any x P XpΓq, vx pf q is the valuation of f at x as a function on XpΓq.
We then deduce from Proposition 1.39 that:

Corollary 2.38. If 0 ‰ f P A0 pΓq “ CpXq, then

ÿ
vx pf q “ 0.
xPΓzH˚

Remark 2.39. For each k, Ak pΓq is a C-vector space. Moreover, if f P Ak pΓq and
0 ‰ g P Al pΓq, then f g P Ak`l pΓq and f {g P Ak´l pΓq.
A modular form is just a modular function such that vz pf q ě 0 for all z P H˚ .
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 49

Exercise 2.40. (a) Show that for any k even, Ak pSL2 pZqq ‰ 0 and hence Ak pΓq ‰ t0u
for any Γ. (Hint: consider Ek`12m {∆m for m big enough).

(b) Take any f ‰ 0 P Ak pΓq, prove that the map A0 pΓq Ñ Ak pΓq, g ÞÑ gf is an
isomorphism of C-vector spaces.

(c) Conclude that for any k even, Ak pSL2 pZqq – A0 pΓq “ CpXpΓqq as C-vector spaces.
In particular, Ak pSL2 pZqq – Cpjq as C-vector spaces.

As a consequence, we see that M0 pΓq “ C and S0 pΓq “ t0u.

To calculate the dimension for general k, we first realise modular forms geometrically.

Definition 2.41. Let X be a compact Riemann surface.

ř
A rational divisor (resp. integral divisor) D on X is a formal finite sum nx x
xPX
where nx P Q (resp. nx P Z) and nx “ 0 for all but finitely many x P X.
ř ř
Let D “ nx x be a rational divisor. We define rDs :“ rnx sx an integral
xPX xPX
divisor.
ř ř
We define the degree of a divisor D “ nx x by degpDq :“ nx .
xPX xPX
We say D is positive if nx ě 0 for all x, and we write D ě 0.
ř ř 1
For D “ nx x and D1 “ nx x two rational divisors, we define D ` D1 :“
ř xPX ř xPX
pnx ` n1x qx and D ´ D1 :“ pnx ´ n1x qx.
xPX xPX
Then the set of rational (resp. integral) divisors, denoted by DivQ pXq (resp. DivQ pXq),
is an additive abelian group. The degree map is a group morphism from DivQ pXq to Q.

Example 2.42. (a) Let f be a meromorphic function on X. Then

ÿ
divpf q :“ vx pf qx
xPX

is an integral divisor on X with degree 0, called a principal divisor.

(b) Let f be a modular function of level Γ and weight k. Although f is not defined on
XpΓq, we may define vx pf q for all x P XpΓq by Definition 2.37. We may define a
rational divisor on X as follows:
ÿ
divpf q :“ vx pf qx.
xPXpΓq

Theorem 2.43. (Valence’s formula) Let 0 ‰ f P Ak pΓq. Define d :“ rSL : ˘Γs. Then
kd
degpdivpf qq “ .
12
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 50

We sketch the idea of proof here. For k “ 0, it is nothing but Corollary 2.38. We
then prove it for ∆ P A12 pΓq. Note that ∆ has no zero on H. Let π be the natural
projection from XpΓq to XpSL2 pZqq. Hence
ÿ ÿ
degpdivp∆qq “ vα p∆q “ vα pπq “ degpπq “ d.
αPΓzP1 pQq xPπ ´1 pαq

For general f , consider f 12 {∆k P A0 pΓq.

We can also prove it using integrals of f 1 {f along the fundamental domain. For Γ “
SL2 pZq, see Proposition III.2.8 of [Kob84]. We can then deduce the general case
ś from the
SL2 pZq-case by considering the norm map Ak pΓq Ñ Adk pSL2 pZqq, f ÞÑ f |k δ.
δP˘ΓzSL2 pZq

We now calculate the dimension of Mk pΓq and Sk pΓq. Assume for simplicity that k
is even in the following of the section. We refer to [DS05] Section 3.6 for the case where
k is odd.

Let n be a positive integer. Recall that D is the unit disc of the complex plane. Let
V be an open subset of D. We define an n-meromorphic differential form on V by:
Ωbn pV q :“ tω “ f pzqpdzqn | f meromorphic on V u.

Let X be a compact Riemann surface with pϕx : Ux Ñ Vx Ă DqxPX an atlas of

charts. We define an n-meromorphic differential form by
Ωbn pXq :“ tω “ pωx P Ωbn pVx qqxPX | the family pωx qxPX is compatible with the transition mapsu.

We refer to [DS05] for more details.

Recall that Mk pΓq “ 0 if k ă 0 and M0 pΓq “ C. We may assume that k ě 2.

We note that for any γ P SL2 pZq, dpγzq “ pcz ` dq´2 dz. Hence if f is a modular
function of weight k, then
k k k
f pγzqpdpγzqq 2 “ pcz ` dqk f pzqpcz ` dq´k pdzq 2 “ f pzqpdzq 2 .
k
In other words, f pdzq 2 is Γ-invariant and then defines a k{2-meromorphic differential
k
form on XpΓq. Conversely, if a meromorphic differential form f pdzq 2 on XpΓq is Γ-
invariant, the lifting of f on H must be a modular function of weight k. We then
obtain:
k
Proposition 2.44. For k even, the map f ÞÑ f pdzq 2 gives an isomorphism between Ak pΓq
and Ωbk{2 pXpΓqq.
k
For any 0 ‰ ω P Ωbk{2 pXpΓqq and any x P X, we write ωx “ fx pzx qpdzx q 2 near
x where zx is the local coordinate at x. We may define vx pωq by vx pfx q. Since X is
compact, one can show easily that there are only finitely many pointes x P X such that
vx pfx q ‰ 0. We can hence define an integral divisor on X by:
ÿ
divpωq :“ vx pωqx.
xPXpΓq
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 51

Definition 2.45. We write E2 for the set of elliptic point of period 2, E3 for the set of
elliptic point of period 3, and E8 for the set of cusps in XpΓq. We write ε˚ for the
cardinality of E˚ where ˚ “ 2, 3 or 8.

? 1 ? 1 h n´1
Combine d z “ ? dz, d 3 z “ ? dz and the fact that qhn dz “ q dqh , we
2 z 3 2
3 z 2πi h
obtain:

Lemma 2.46. For any f P Ak pΓq, we have

k k
divpf pdzq 2 q “ divpf q ` divpdzq
2
1 ř 2 ř ř
where dz :“ ´ x´ x´ x.
2 xPE2 3 xPE3 xPE8

k ř k ř k ř k
Consequently, rdivpf qs “ divpf pdzq 2 q ` r sx ` r sx ` ` r sx.
xPE2 4 xPE3 3 xPE8 2

Now fix weight k and level Γ. We are going to calculate the dimension of Mk pΓq.
By Exercise 2.40, there exists 0 ‰ f P Ak pΓq and the map f ÞÑ f f0 is an isomorphism
from A0 pΓq “ CpXpΓqq to Ak pΓq.
Therefore, we get:

Mk pΓq – tf0 P CpXpΓqq | f0 “ 0 or divpf0 f q ě 0u

“ tf0 P CpXpΓqq | f0 “ 0 or divpf0 q ` divpf q ě 0u
“ tf0 P CpXpΓqq | f0 “ 0 or divpf0 q ` rdivpf qs ě 0u (2.12)
ř
where rdivpf qs is the integral part of divpf q, namely, rdivpf qs “ rvx pf qsx.
xPXpΓq

Similarly, we get
ÿ
Sk pΓq – tf0 P CpXpΓqq | f0 “ 0 or divpf0 q ` rdivpf qs ´ x ě 0u. (2.13)
xPE8

Let X be a compact Riemann surface and D be an integral divisor on X. We define

the linear space

LpDq :“ tf P CpXq | f “ 0 or divpf q ` D ě 0u.

We can verify easily that LpDq is a C-vector space (exercise) and we denote by lpDq its
dimension.

If D and D1 are two integral divisors on X such that D1 ´ D is a principal divisor,

say divpf q, then the map g ÞÑ gf gives an isomorphism between LpD1 q and LpDq. In
particular, we have lpD1 q “ lpDq.
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 52

Let 0 ‰ ω1 P Ω1 pXq. The divisor K :“ divpω1 q is called a canonical divisor of X. We

remark that two canonical divisors can only differ by a principal divisor. In particular,
lpKq is well-defined and does not depend on the choice of ω1

ř
We fix any 0 ‰ f P Ak pΓq and define Dk “ rdivpf qs, Dk1 “ rdivf s ´ x. We can
xPE8
rewrite equation (2.12) and equation (2.13) as Mk pΓq – LpDk q and Sk pΓq – LpDk1 q.
Consequently, we have dimpMk pΓqq “ lpDk q and dimpSk pΓqq “ lpDk1 q.

Theorem 2.47. (Riemann-Roch) Let X and D as above. Let K be a canonical divisor

and g be the genus of X. Then

lpDq “ degpDq ´ g ` 1 ` lpK ´ Dq. (2.14)

Corollary 2.48. We keep the notation above and have:

(a) lpKq “ g.

(b) degpKq “ 2g ´ 2.

(c) If degpDq ă 0, then lpDq “ 0.

(d) If degpDq ą 2g ´ 2, then lpDq “ degpDq ´ g ` 1.

Proof. (a) We first notice that f P Lp0q if and only if f is holomorphic, and hence f
is constant. We get lp0q “ 1.
Then let D “ 0 in equation (2.14), we get 1 “ ´g ` 1 ` lpKq and hence lpKq “ g.

(b) Let D “ K in equation (2.14), we get g “ degpKq ´ g ` 1 ` 1 and hence degpKq “

2g ´ 2.

(c) We assume by contradiction that there exists 0 ‰ f P LpDq. Then D ` divpf q ě 0.

In particular, degpDq ` degpdivpf qq “ degpDq ě 0. Contradiction!

(d) When degpDq ą 2g ´ 2, we have degpK ´ Dq “ degpKq ´ degpDq ă 0. Hence

lpDq “ degpDq ´ g ` 1 ` lpK ´ Dq “ degpDq ´ g ` 1.

k{2
Notice that any ω P Ωbk{2 pXq is of the form f ω1 for some f P CpXq and 0 ‰ ω1 P
Ω1 pXq. We deduce that:
Lemma 2.49. Let 0 ‰ ω P Ωbk{2 pXq. Then degpωq “ kpg ´ 1q.
k
We go back to the modular curve case. Let g :“ gpXpΓqq. Recall that f pdzq 2 P
k
Ωbk{2 pXpΓqq. We get: degpdivpf qq ` degpdivpdzqq “ kpg ´ 1q. Combine with Va-
2
kd
lence’s formula that degpdivpf qq “ , we obtain that:
12
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 53

Corollary 2.50.
d ε2 ε3 ε8
gpXpΓqq “ 1 ` ´ ´ ´ .
12 4 3 2
ř
Recall that Dk “ rdivpf qs and Dk1 “ rdivf s ´ x. We can also deduce from
xPE8
Lemma 2.46 that:

Lemma 2.51.
k k k
degpDk q “ kpg ´ 1q ` r sϵ2 ` r sϵ3 ` r sϵ8 ;
4 3 2
k k k
and degpDk1 q “ kpg ´ 1q ` r sϵ2 ` r sϵ3 ` r ´ 1sϵ8 .
4 3 2
Lemma 2.52. When k ě 2 we have degpDk q ą 2pg ´ 1q; when k ě 4, we have degpDk1 q ą
2pg ´ 1q.

d
Proof. It is clear if g ě 1. It remains the case where g “ 0. Recall that g “ 1 ` ´
12
ε2 ε3 ε8 ε2 ε3 ε8 d
´ ´ . Hence ` ` “1` .
4 3 2 4 3 2 12
d k k kd k k
We have degpDk q “ kpg ´ 1q ` kp1 ` q ´ t uε2 ´ t uε3 “ ´ t uε2 ´ t u
12 4 3 12 4 3
kd
If k “ 2, it is equal to and hence bigger than ´2.
2
kd 3 2
If k ě 4, we have degpDk q ě ´ ε2 ´ ε3 .
12 4 3
ε2 ε3 1 d
Since ε8 ě 1, we have ` ď ` .
4 3 2 12
kd 3 2 kd ε2 ε3 pk ´ 3qd 3
Therefore ´ ε2 ´ ε3 ě ´ 3p ` q ě ´ ą ´2 as expected.
12 4 3 2 4 3 12 2
The case for Dk1 is similar and we leave the details to the reader.

We can now calculate dimpMk pΓqq and dimpSk pΓqq. The above cases can be de-
duced from Corollary 2.48. It remains the case where k “ 2. In this case, we have
S2 pΓq – Ω1hol pXpΓqq “ LpKq. Hence dimpS2 pΓqq “ lpKq “ gpXpΓqq.

We can now summarize the results as follows:

Theorem 2.53. Let k be an even integer, Γ be a congruence subgroup, ε2 , ε3 and ε8 be

the number of elliptic points of order 2, the number of elliptic points of order 3 and the
number of cusps respectively. Let g be the genus of XpΓq. We then have:
2 MODULAR FORMS FOR CONGRUENCE SUBGROUPS 54

dimpMk pΓqq dimpSk pΓqq

kă0 0 0
k“0 1 0
k“2 g ´ 1 ` ε8 « ff « ff g « ff « ff
kě4 k k k k k k
pk ´ 1qpg ´ 1q ` ε2 ` ε3 ` ε8 pk ´1qpg ´1q` ε2 ` ε3 `p ´1qε8
4 3 2 4 3 2
3 HECKE OPERATORS 55

3 Hecke operators

3.1 Double coset operator

Recall that GL2 pQq` acts on H, and we can also define the j-factor and the weight k
operator for elements in GL2 pQq` . See section 2.3.2.

Definition-Lemma 3.1. Let Γ1 and Γ2 be two congruence subgroups and let `

Ů α P GL2 pQq .
Take tβj P Γ1 αΓ2 u a set of representatives of Γ1 zΓ1 αΓ2 , i.e., Γ1 αΓ2 “ Γ1 βj . We will
j
show in Lemma 3.5 that Γ1 zΓ1 αΓ2 is finite.
ř
Then the map |k Γ1 αΓ2 : f ÞÑ f |k βj is well-defined and gives a linear map from
j
Mk pΓ1 q to Mk pΓ2 q. Moreover, it sends Sk pΓ1 q to Sk pΓ2 q.

Proof. Since each |k βj is linear, the map |k Γ1 αΓ2 is linear.

ř
Let f P Mk pΓ1 q. We want to show f |k βj is in Mk pΓ2 q.
j
ř
It is clear that f |k βj is holomorphic on H.
j

Moreover, since f |k γ1 “ f if γ1 P Γ1 , we see that the sum does not depend on the
choice of representatives βj .
ř ř
Then for any γ2 P Γ2 , p f |k βj q|k γ2 “ f |k pβj γ2 q.
j j
Ů
Note that Γ1 αΓ2 “ Γ1 αΓ2 γ2 “ Γ1 βj γ2 . In other words, tβj γ2 u is another set of
jř ř
representatives of Γ1 zΓ1 αΓ2 . Hence p f |k βj q|k γ2 “ f |k βj as expected.
j j
ř ř
Finally, for any δ P SL2 pZq, we have p f |k βj q|k δ “ f |k pβj δq is holomorphic at 8,
j j
and is moreover 0 if f is a cusp form.

Example 3.2. (a) If α “ I, Γ1 Ą Γ2 , then Γ1 αΓ2 “ Γ1 . The map |Γ1 αΓ2 is the natural
inclusion Mk pΓ1 q ãÑ Mk pΓ2 q.

(b) If α “ I, Γ1 Ă Γ2 , then Γ1 αΓ2 “ Γ2 . The map |Γ1 αΓ2 is the trace map from
Mk pΓ1 q to Mk pΓ2 q.

(c) If Γ2 “ α´1 Γ1 α, then Γ1 αΓ2 “ Γ1 α. The map |Γ1 αΓ2 is just |k α which sends
Mk pΓ1 q to Mk pα´1 Γ1 αq.

Exercise 3.3. Any double coset operator is a composition of the operators as in the
previous example. (Hint: consider Γ3 “ α´1 Γ1 α X Γ2 and Γ4 “ αΓ2 α´1 .)

Lemma 3.4. Let Γ be a congruence subgroup. Let α P GL2 pQq` . Then α´1 Γα X SL2 pZq
is still a congruence subgroup.

Proof. Take N P N such that Γ Ą ΓpN q. Let M be a positive integer such that both
M α and M α´1 have integral coefficients.
3 HECKE OPERATORS 56

We claim that α´1 Γα contains ΓpM 2 N q.

Indeed, for any γ P ΓpM 2 N q, write γ “ 1 ` M 2 N δ with δ P M2 pZq. Then

αγα´1 “ I ` N pM αqδpM α´1 q P M2 pZq

and congruent to I modulo N .

Hence αγα´1 P ΓpN q and then γ P α´1 ΓpN qα X SL2 pZq Ă α´1 Γα X SL2 pZq. In
particular, the latter is a congruence subgroup.

Lemma 3.5. Let Γ1 , Γ2 and α be as above. Then the left multiplication

ϕα : Γ2 Ñ Γ1 αΓ2 , γ2 ÞÑ αγ2

gives a natural bijection from pα´1 Γ1 α X Γ2 qzΓ2 to Γ1 zΓ1 αΓ2 .

Consequently, Γ1 zΓ1 αΓ2 is a finite set.

Proof. Apparently ϕα induces a surjective map Γ2 Ñ Γ1 zΓ1 αΓ2 .

For any γ2 , γ21 P Γ2 , ϕα γ2 “ ϕα γ21 if and only if there exists γ1 P Γ1 such that
γ1 αγ2 “ αγ21 .
It is equivalent to that γ21 “ α´1 γ1 αγ2 ðñ γ21 P pα´1 Γ1 α X Γ2 qγ2 as expected.
Finally, since α´1 Γ1 α X SL2 pZq is a congruence subgroup by the previous lemma, we
know that α´1 Γ1 α X Γ2 is a congruence subgroup, and hence is finite index inside Γ2 .
Consequently, Γ1 zΓ1 αΓ2 is finite.

Remark 3.6. (Geometric explanation of the double coset operator) Recall that Mk pΓq –
Ωbk{2 pXq when k is even. The double coset operator can be interpreted in terms of cor-
respondences between modular curves.

Roughly speaking, a correspondence is a multi-valued function. More precisely, let X1

and X2 be two compact Riemann surface, a correspondence from X2 to X1 , written as
X2 ⇝ X1 , is a triple pX3 , ϕ, ψq where ϕ : X3 Ñ X2 and ψ : X3 Ñ X1 are holomorphic
maps.
The map ϕ˚ ˝ ψ ˚ then induces a liner map from Ωl pX1 q to Ωl pX2 q for any l ě 0.
ř
More precisely, we define a map F : X2 Ñ DivpX1 q by z ÞÑ vw pϕqψpwq. Let
wPϕ´1 pzq
ω P Ωl pX1 q. Then the map ϕ˚ ˝ ψ ˚ sends ω to the differential form z ÞÑ
ř
nx pF zqωpxq
xPX1
in Ωl pX2 q.

Now let Γ1 , Γ2 and α as before. Define Γ3 “ α´1 Γ1 α X Γ2 .

The natural inclusion Γ3 ãÑ Γ2 gives a projection ϕ : XpΓ3 q Ñ XpΓ2 q. The inclusion
Γ3 ãÑ Γ1 , γ ÞÑ αγα´1 induces a map ψ : XpΓ3 q Ñ XpΓ1 q sending Γ3 z to Γ1 αz.
3 HECKE OPERATORS 57

Ů
Write Γ2 “ Γ3 γj . Then by Lemma 3.5, the family tβj :“ αγj u is a set of repre-
j
sentatives of Γ1 zΓ1 αΓ2 .

řLet z P H. Then
ř ϕ´1 pΓ2 zq “ tΓ3 γj zu. Hence the map X2 Ñ DivpX1 q sending Γ2 z
to Γ1 αγj z “ Γ1 βj z.
j j
k
For k even, let f P Mk pΓ1 q. Then f pdzq 2 is a k{2-differential form on XpΓ1 q.
k ř k ř
The map ϕ˚ ˝ ψ ˚ sending f pdzq 2 to z ÞÑ f pβj zqpdpβj zqq 2 “ pf |k βj qpzq. The last
j j
differential form corresponds to the modular form f |k Γ1 αΓ2 .

3.2 Diamond and Hecke operators

ˆ ˙
a b
Recall that Γ1 pN q is the kernel of the map Γ0 pN q ÞÑ pZ{N Zqˆ , ÞÑ d mod N .
c d
In particular, Γ1 pN q is a normal subgroup of Γ0 pN q.

We shall study the finite dimensional space Mk pΓ1 pN qq by defining linear operators
on it. We may assume in the following that k is even. Note that since ´I2 P Γ1 pN q we
have Mk pΓ1 pN qq “ t0u for k even.

Definition 3.7. ˆ
ˆ ˙ (Diamond operator) Let d P pZ{N Zq . Take α P Γ0 pN q such that
˚ ˚
α ” mod N . We have α´1 Γ1 pN qα “ Γ1 pN q. We are in the situation of
0 d
Example 3.2 (c). We get a linear operator, called a Diamond operator:

xdy : Mk pΓ1 pN qq Ñ Mk pΓ1 pN qq, f ÞÑ f |k α.

It sends Sk pΓ1 pN qq to Sk pΓ1 pN qq. One can show easily that xdy does not depend on the
choice of α.
Moreover, for d, d1 P pZ{N Zqˆ , we have xdyxd1 y “ xdd1 y. We hence get an action of
the finite abelian group pZ{N Zqˆ on the finite-dimensional C-vector spaces Mk pΓ1 pN qq
and Sk pΓ1 pN qq.

Assume that f P Mk pΓ1 pN qq is an eigenvector for all Diamond operators. It is easy

to see that taking eigenvalues defines a group morphism from pZ{N Zqˆ to S 1 :“ tz P
C | |z| “ 1u. We call such a morphism a character of pZ{N Zqˆ .
For G a finite abelian group, the set of its characters also forms an abelian group,
denoted by G p and called its Pontryagin dual. It is easy to see that there is a (non-
canonical) isomorphism between G and G p (c.f. Appendix A).
We remark any finite-dimensional irreducible representation of a finite abelian group
is one dimensional, and hence is given by a character of the finite group (c.f. Appendix
A). We can then decompose Mk pΓ1 pN qq and Sk pΓ1 pN qq into characters of pZ{N Zqˆ .

Definition 3.8. Let χ : pZ{N Zqˆ Ñ S 1 be a character where S 1 “ tz P C | |z| “ 1u. We

3 HECKE OPERATORS 58

define

Mk pN, χq :“ tf P Mk pΓ1 pN qq, xdyf “ χpdqf, for any d P pZ{N Zqˆ u

and Sk pN, χq :“ tf P Sk pΓ1 pN qq, xdyf “ χpdqf, for any d P pZ{N Zqˆ u

In particular, Mk pN, 1q “ Mk pΓ0 pN qq and Sk pN, 1q “ Sk pΓ0 pN qq.

Lemma 3.9. We have decompositions
à
Mk pΓ1 pN qq “ Mk pN, χq
Ź

pZ{N Zqˆ
à
and Sk pΓ1 pN qq “ Sk pN, χq.
Ź

χPpZ{N Zqˆ

For the reader who are not familiar with the representation theory of finite abelian
groups, we refer to Appendix A.
Definition 3.10. For any positive integer n, we define
"
xn mod N y if pn, N q “ 1;
xny :“
0 otherwise.

They are operators on Mk pΓ1 pN qq and Sk pΓ1 pN qq. It is easy to see that xnmy “
xnyxmy “ xmyxny for any positive integers n, m.

3.3 Hecke operators

Definition 3.11. (Hecke operator) Let p beˆ a rational

˙ prime. Let N be a positive
1 0
integer. Consider Γ1 “ Γ2 “ Γ1 pN q and α “ . We define
0 p
k
Tp :“ p 2 ´1 |k Γ1 αΓ2 ,

a linear operator on Mk pΓ1 pN qq as well as on Sk pΓ1 pN qq.

Lemma 3.12.
$ ˜ ˙¸
p´1
ˆ
’ k
´1 ř 1 j
p f |k if p | N,
’
’ 2
0 p
’
&
j“0
Tp f “ ˜
p´1
ˆ ˙ ˆ ˙ ˆ ˙¸
’ k ř 1 j r s p 0
p 2 ´1 f |k ` f |k if p ∤ N
’
’
0 p N p 0 1
’
%
j“0

where r, s are any integers such that rp ´ sN “ 1 in the second case. Such integers
exist since pp, N q “ 1.
Equivalently, for any N and p, we have
˜p´1 ˆ ˙ ˆ ˙¸
k
´1
ÿ 1 j p 0
Tp f “ p 2 f |k ` pxpyf q|k .
0 p 0 1
j“0
3 HECKE OPERATORS 59

ˆ ˙
1 0
Proof. Let Γ “ Γ1 “ Γ2 “ Γ1 pN q and α “ .
0 p
Recall by Lemma 3.5, the map
pα´1 Γα X ΓqzΓ Ñ ΓzΓαΓ, pα´1 Γα X Γqγ ÞÑ Γαγ
is a bijection.
We write Γ3 :“ α´1 Γα X Γ.
It is easy to see that we only need to show that
$
p´1
ˆ ˙
’ Ů 1 j
’
’ Γ3 if p | N ;
& 0 1
Γ“ p´1
ˆ ˙ j“0 ˆ ˙ (3.1)
’ Ů 1 j Ů rp s
’
’ Γ3 Γ3 if p ∤ N.
%
j“0 0 1 N 1
ˆ ˙
˚ 0
Define Γ0 ppq :“ Γ0 ppqT “ tγ P SL2 pZq, γ ” mod p.u
˚ ˚
We claim that Γ3 “ α´1 Γα X Γ “ Γ0 ppq X Γ.
ˆ ˙
a b
Indeed, let α´1 γα P α´1 Γα X Γ where γ “ P Γ “ Γ1 pN q. Then α´1 γα “
c d
ˆ ˙
a pb
P Γ0 ppq as expected.
p´1 c d
ˆ ˙ ˆ ˙
a b 0 ´1 a b{p
Conversely, let γ “ P Γ ppq X Γ, then αγα “ has integer
c d pc d
ˆ ˙
1 ˚
coefficients and is congruent to mod N , and hence is in Γ.
0 1
ˆ ˙
1 j
We then observe immediately that for any 0 ď j ă k ď p ´ 1, we have Γ3 ‰
0 1
ˆ ˙
1 k
Γ3 .
0 1
ˆ ˙
a b
Now let γ “ P Γ. If p ∤ a, then there exists 0 ď j ď p ´ 1 such that b ” aj
c d
ˆ ˙ ˆ ˙
1 ´j a b ´ aj
mod p. Then γ “ P Γ0 ppq and hence in Γ3 . In other words,
0 1 c d ´ cj
ˆ ˙
1 j
γ P Γ3 . This is always the case if p | N . We hence obtain the first case of the
0 1
equation 3.1.
ˆ ˙ ˆ ˙ ˆ ˙ˆ ˙´1
1 j rp s rp s 1 j
If p ∤ N , we first check that Γ3 ‰ Γ3 for any j. Indeed, “
0 1 N 1 N 1 0 1
ˆ ˙
˚ s ´ rpj
R Γ3 .
˚ ˚
p´1
ˆ ˙ ˆ ˙ˆ ˙´1
Ů 1 j a b rp s
Then for any γ P Γ ´ Γ3 , we must have p|a.Then “
j“0 0 1 c d N 1
3 HECKE OPERATORS 60

ˆ ˙ˆ ˙ ˆ ˙
a b 1 ´s ˚ ´as ` brp
“ P Γ0 ppq X Γ “ Γ3 where r, s are integers as in
c d ´N rp ˚ ˚
the statement.
ˆ ˙ ˆ ˙
a b rp s
In other words, P Γ3 . We hence complete the proof.
c d N 1
ˆ ˙
1 1
Definition 3.13. Let f P Mk pΓ1 pN qq. Since P Γ1 pN q, the Fourier expansion of
0 1
f at 8 is of the form
8
ÿ
f“ an q n
n“0

where q “ e2πiz . For each n ě 0, we define an pf q :“ an .

For t P R ´ Z., we define at pf q :“ 0.

Definition 3.14. Let χ be a character of pZ{N Zqˆ . we can extend it to Z{N Z by zero
and then lift it to Z, still denoted by χ. More precisely, for any positive integer n, we
have "
χpn mod N q if pn, N q “ 1;
χpnq :“
0 otherwise.
Then χ is completely multiplicative, i.e., χpnqχpmq “ χpnmq for any n, m P Z.

Proposition 3.15. Let f P Mk pΓ1 pN qq. Let 1N be the trivial character on pZ{N Zqˆ .
Then we have:

(a)
8
ÿ 8
ÿ
n k´1
Tp pf q “ anp pf qq ` p an pxpyf qq np
n“0 n“0
ÿ8
“ panp pf q ` pk´1 an{p pxpyf qqq n .
n“0

Recall at pf q “ 0 for t P R ´ Z.

(b) If f is moreover in Mk pN, χq for some character χ, then

8
ÿ
Tp pf q “ panp pf q ` χppqpk´1 an{p pf qqq n .
n“0

Proof. Apparently pbq is a direct corollary to paq. We now prove (a). By Lemma 3.9,
we have ˜p´1 ˆ ˙ ˆ ˙¸
k
´1
ÿ 1 j p 0
Tp f “ p 2 f |k ` pxpyf q|k
0 p 0 1
j“0
.
3 HECKE OPERATORS 61

For 0 ď j ď p ´ 1, we have:
ˆ ˙
k
´1 1 j k k z`j
p 2 f |k pzq “ p 2 ´1 p 2 p´k f p q
0 p p
8
1ÿ 2πi z`j n
“ an pf qe p .
p n“0

Hence
p´1
1 p´1 8
ˆ ˙
ÿ k 1 j ÿ ÿ 2πi z`j n
p 2 ´1 f |k pzq “ an pf qe p
0 p p j“0 n“0
j“0
8 p´1
1ÿ ÿ 2πi j n 2πi z n
“ p e p qan pf qe p
p n“0 j“0
2πi z n
ÿ
“ an pf qe p
p|n
8
ÿ
“ anp pf qe2πizn .
n“0

ˆ ˙ 8
p 0
pk{2 an pxpyf qq pn .
ř
Finally, it is easy to see that pxpyf q|k “
0 1 n“0

ˆ ˙
1 0
Lemma 3.16. Let α “ . Let N be a positive integer.
0 p

(a) Γ0 pN qαΓ0 pN q “ Γ1 pN qαΓ0 pN q.

(b) Let γˆ P Γ0˙

pN qαΓ0 pN q “ Γ1 pN qαΓ0 pN q. Then γ P Γ1 pN qαΓ1 pN q if and only if
1 ˚
γ” mod N .
0 p

Proof.ˆ (a) ˙ For each 1 ď d ă N with pd, N q “ 1, take take a, b P Z such that δd :“
a b
P SL2 pZq. Then Γ1 pN qzΓ0 pN q is represented by these matrices. We only
N d
need to show that δd α are in Γ1 pN qαΓ0 pN q.
ˆ ˙´1
1 dN ` 1 ´1
If p | N , take δ “ P Γ1 pN q, then
´dN 1
ˆ ˙ˆ ˙ˆ ˙ˆ ˙ ˆ ˙
´1 1´1 1 0 dN ` 1 ´1 a b 1 0 ˚ ˚
α δ δd α “ “ P Γ0 pN q.
0 p´1 ´dN 1 N d 0 p p1 ´ adqN p´1 ˚

If p ∤ N , we may take d1 P Zˆsuch that 1

˙ d ” 1 mod N and d ” ´a mod p. Take
˚ ˚
δ 1 P Γ1 pN q such that δ 1´1 ” P SL2 pZq.
N d1
3 HECKE OPERATORS 62

Then then
ˆ ˙ˆ ˙ˆ ˙ˆ ˙ ˆ ˙
´1 1´1 1 0 ˚ ˚ a b 1 0 ˚ ˚
α δ δα “ “ P Γ0 pN q.
0 p´1 N d1 N d 0 p N pa ` d1 qp´1 ˚

(b) is easy simply by modulo N .

Remark 3.17. In fact, we can show that

ˆ ˙
1 ˚
Γ1 pN qαΓ1 pN q “ tγ P M2 pZq | γ ” mod N, detpγq “ pu.
0 p

Proposition 3.18. Diamond operators commute with Hecke operators. More precisely,
let d P pZ{N Zqˆ and p be a rational prime. We have xdyTp f “ Tp xdyf for any f P
Mk pΓ1 pN qq.
Consequently, Tp acts on Mk pN, χq for any character χ on pZ{N Zqˆ .
ˆ ˙
1 0 Ů
Proof. Let α “ . Write Γ1 pN qαΓ1 pN q “ j Γ1 pN qβj .
0 p
ˆ ˙
˚ ˚
Let γ P Γ0 pN q such that γ ” mod N . Let c “ d´1 P pZ{N Zqˆ .
0 d
ˆ ˙ˆ ˙ˆ ˙ ˆ ˙
´1 c ˚ 1 0 d ˚ 1 ˚
Then γαγ ” ” mod N . Note that γαγ ´1 P
0 d 0 p 0 c 0 p
Γ0 pN qαΓ0 pN q. Hence by Lemma 3.16, we know γαγ ´1 P Γ1 pN qαΓ1 pN q.
Consequently, we have

Γ1 pN qαΓ1 pN q “ Γ1 pN qγαγ ´1 Γ1 pN q
“ γΓ1 pN qαΓ1 pN qγ ´1
ğ
“ γ Γ1 pN qβj γ ´1
j
ğ
“ Γ1 pN qγβj γ ´1 .
j

Here we have used the fact that Γ1 pN q is normal in Γ0 pN q and hence γΓ1 pN q “
Γ1 pN qγ.
Consequently,
k ÿ ÿ
p´ 2 `1 xdyTp f “ xdy f |k γβj γ ´1 “ f |k γβj γ ´1 |k γ
j j
ÿ ÿ k
“ f |k γβj “ pf |k γq|k βj “ p´ 2 `1 Tp xdyf
j j

as expected.
3 HECKE OPERATORS 63

Corollary 3.19. Hecke operators commute with each other. More precisely, for p, q two
rational primes, we have Tp Tq f “ Tq Tp f for any f P Mk pΓ1 pN qq.

8
an e2πinz P Mk pN, χq for
ř
Proof. By Proposition 3.18, we may assume that f “
n“0
some character χ of pZ{N Zqˆ . Then by Proposition 3.15, we have an pTp f q “ anp `
χppqpk´1 an{p for any n ě 0. Similarly, an pTq f q “ anq ` χpqqq k´1 an{q .
Hence an pTp Tq f q “ anp pTq pf qq ` χppqpk´1 an{p pTq pf qq “ anpq ` χpqqq k´1 anp{q `
χppqpk´1 anq{p ` χppqχpqqpk´1 q k´1 an{pq which is symmetric with respect to p and q.
In particular, Tp Tq “ Tq Tp as expected.

Definition 3.20. For each positive integer n, we may define Tn , an operator on Mk pΓ1 pN qq
as well as on Sk pΓ1 pN qq, inductively by the following relations:

• T1 “ Id;

• Tpr “ Tp ˝ Tpr´1 ´ pk´1 xpyTpr´2 for any rational prime p and r ě 2;

• Tnm “ Tn ˝ Tm if pn, mq “ 1.

It is easy to see that the operators Tn , n ě 1, xmy, m ě 1, all commute.

8
p1 ´ Tp p´s ` xpypk´1´2s q´1 .
ř ś
Exercise 3.21. The formal series Tn n´s “
n“1 p

Proposition 3.22. Let f P Mk pΓ1 pN qq, then for each m ě 0, we have

ÿ
am pTn pf qq “ dk´1 amn{d2 pxdyf q. (3.2)
d|pm,nq

Equivalently, if f P Mk pN, χq for some character χ of pZ{N Zqˆ , then

ÿ
am pTn pf qq “ χpdqdk´1 amn{d2 pf q. (3.3)
d|pm,nq

Proof. It is easy to see that equation 3.2 is equivalent to equation 3.3. We now prove
the second equation.
We first show it for n “ pr for some rational prime p and r ě 0 by induction on r.
Namely, we shall show that
ÿ
am pTpr pf qq “ χpdqdk´1 amn{d2 pf q (3.4)
d|pm,pr q

for f P Mk pN, χq.

When r “ 0 the equation is trivial. When r “ 1, it is just Proposition 3.15 (b). We
assume that r ě 2 and the equation is known for r ´ 1 and r ´ 2.
3 HECKE OPERATORS 64

By Definition 3.20, we have Tpr “ Tp Tpr´1 ´ pk´1 xpyTpr´2 . Hence

Tpr f “ Tp Tpr´1 f ´ pk´1 χppqTpr´2 f.

Therefore, for any m ě 0,

am pTpr f q “ am pTp Tpr´1 f q ´ pk´1 χppqam pTpr´2 f q
ÿ
“ amp pTpr´1 f q ` χppqpk´1 am{p pTpr´1 f q ´ pk´1 χppq χpdqdk´1 ampr´2 {d2 pf q.
d|pm,pr´2 q

Note that by induction,

ÿ ÿ
amp pTpr´1 f q “ χpdqdk´1 ampr {d2 pf q “ ampr pf q` χpdqdk´1 ampr {d2 pf q.
d|pmp,pr´1 q d|ppm,pr´1 q,p|d

Moreover, pk´1 χppq χpdqdk´1 amn{d2 pf q “ χppdqppdqk´1 ampr {ppdq2 pf q

ř ř
d|pm,pr´2 q pd|ppm,pr´1 q

Hence amp pTpr´1 f q ´ pk´1 χppq χpdqdk´1 ampr´2 {d2 pf q “ ampr pf q.

ř
d|pm,pr´2 q

Then since
ÿ
χppqpk´1 am{p pTpr´1 f q “ χppdqppdqk´1 ampr´2 {d2
d|pm{p,pr´1 q
ÿ
“ χppdqppdqk´1 ampr {ppdq2
pd|pm,pr q
ÿ
“ χpdqpdqk´1 ampr {d2 ,
d|pm,pr q,p|d

we obtain equation (3.4) for pr .

At last, let n, l be positive integers such that pn, lq “ 1. It remains to show that if
equation (3.3) holds for n and l, then it holds for nl. We leave it as an exercise.

8
an q n P Sk pN, χq is an eigenvector for all Tn , n ě 1 with
ř
Remark 3.23. If 0 ‰ f “
n“1
eigenvalues λn . For each n ě 1, since a1 pTn f q “ an pf q, we know that an “ λn a1 . In
particular, a1 ‰ 0 since otherwise an “ 0 for all n which contradicts to the fact that
f ‰ 0.
We may hence normalise f such that a1 pf q “ 1. Then an is the eigenvalue of f for
Tn . We define the L-function of f by:
8
ÿ
Lps, f q :“ an n´s . (3.5)
n“1

We shall show that it converges absolutely for Repsq ąą 0. By exercise 3.21, we see that
Lps, f q admits an Euler product:
ź
Lps, f q “ p1 ´ ap p´s ` χppqpk´1´2s q´1 . (3.6)
p
3 HECKE OPERATORS 65

Moreover, for pn, mq “ 1, since Tn Tm “ Tnm , we obtain that an am “ anm . Similarly,

we have apr “ ap apr´1 ´ χppqpk´1 apr´2 for p prime and r ě 2.
Conversely, one can show that (exercise) for a family of numbers tan uně1 , the series
8
an n´s factors as p1 ´ ap p´s ` χppqpk´1´2s q´1 if and only if:
ř ś
n“1 p

• a1 “ 1;

• apr “ ap apr´1 ´ χppqpk´1 apr´2 for p prime and r ě 2; and

• an am “ anm if pm, nq “ 1.

For example, recall that S12 pSL2 pZqq is one-dimensional and is generated by
8
ÿ 8
ź
∆“ τ pnqq n “ q p1 ´ q n q24 .
n“1 n“1

Hence ∆ must be an eigenvector for all Hecke operators Tn . Consequently, we get part
(a) and (b) of Proposition 0.2.

Remark 3.24. (the Hecke algebras) We define

ˆ ˙
a b
∆1 pN q :“ t P M2 pZq | c ” 0, a ” 1 mod N, ad ´ bc ą 0u Ă GL2 pQq` ;
c d
ˆ ˙
a b
∆1 pN q :“ t P M2 pZq | c ” 0, pa, N q “ 1, ad ´ bc ą 0u Ă GL2 pQq` .
c d

We define Rp∆1 pN q, Γ1 pN q a free abelian group with basis tTλ | λ P Γ1 pN qz∆1 pN q{Γ1 pN qu.
We define a multiplication structure on Rp∆1 pN q, Γ1 pN qq as follows:
Ů Ů
let α, β P ∆1 pN q, write Γ1 pN qαΓ1 pN q “ Γ1 pN qαi and Γ1 pN qβΓ1 pN q “ Γ1 pN qβj .
For any γ P Γ1 pN q, define mpα, β; γq :“ #tpi, jq | Γ1 pN qαi βj “ Γ1 pN qγu. One can show
that mpα, β; γq only depends of rαs, rβs and rγs where r˚s :“ Γ1 pN qz∆1 pN q{Γ1 pN q the
class of ˚ for any ˚ P ∆1 pN q.

ř
We can then define Trαs Trβs “ mpα, β : γqTrγs which is a finite
rγsPΓ1 pN qz∆1 pN q{Γ1 pN q
sum. We can then extend the product linearly to Rp∆1 pN q, Γ1 pN qq. The latter is hence
endowed with a structure of algebra, and is called a Hecke algebra.

We can define the algebra Rp∆0 pN q, Γ0 pN qq similarly. Both Rp∆1 pN q, Γ1 pN qq and

Rp∆0 pN q, Γ0 pN qq are algebras with unity TrIds . The algebras Rp∆0 pN q, Γ0 pN qq and
Rp∆1 pN q, Γ1 pN qq act on Mk pΓ1 pN qq and Mk pΓ0 pN qq respectively by the double coset
operator. In particular, the multiplication of elements in the Hecke algebra correspon-
dence to the composition of double coset operators.
3 HECKE OPERATORS 66

Moreover, the natural injection ∆1 pN q ãÑ ∆0 pN q induces an isomorphism of alge-

bras Rp∆1 pN q, Γ1 pN qq – Rp∆0 pN q, Γ0 pN qq. One
ˆ can ˙showˆthat Rp∆ ˙ 0 pN q, Γ0 pN qq is
a b a c
commutative since the modified transpose map ÞÑ induces identity
Nc d Nb d
on the Hecke algebra Rp∆0 pN q, Γ0 pN qq. Consequently, Rp∆1 pN q, Γ1 pN qq is also com-
mutative.

The isomorphism of two algebras also imply that Rp∆0 pN q, Γ0 pN qq acts on Mk pΓ1 pN qq
as well, and hence on Mk pN, χq for each character χ of pZ{N Zqˆ by the commutativity.

The Hecke operator Tn corresponds to the element

k ÿ
T pnq :“ n 2 ´1 Trαs P Rp∆0 pN q, Γ0 pN qq – Rp∆1 pN q, Γ1 pN qq.
rαsPΓ0 pN qz∆0 pN q{Γ0 pN q,detpαq“n

For more details, we refer to section 4 of [Miy89].

3.4 Invariant measure on H

We can extend the action of SL2 pZq on H to SL2 pRq by the same formula. Namely, for
ˆ ˙
a b az ` b
γ“ P SL2 pRq and z P H, let γz :“ . We can verify similarly that γz P H.
c d cz ` d
We consider SL2 pRq as a real algebraic group, and hence is a real analytic manifold.
Lemma 3.25. The group SL2 pRq acts ˆ on H transitively
˙ and continuously. Moreover, the
cos θ sin θ
stabilizer of i in G is SO2 pRq :“ t | θ P r0, 2πqu. Consequently, the map
´ sin θ cos θ
ˆ ˙
a b ai ` b
SL2 pRq Ñ H, ÞÑ induces an homeomorphism SL2 pRq{SO2 pRq – H where
c d ci ` d
the first quotient is endowed with the quotient topology.
ˆ ˙ ˆ 1{2 ˙
1 x y 0
Proof. For any z “ x ` iy P H, consider γ “ P SL2 pRq, then
0 1 0 y ´1{2
ˆ ˙ ˆ 1{2 ˙ ˆ ˙
1 x y 0 1 x
γi “ ´1{2 i“ yi “ yi ` x “ z. Hence the action is transitive.
0 1 0 y 0 1
ˆ ˙
a b
Apparently any element in SO2 pRq fixes i. Conversely, if γ “ P Gi , then
c d
ai ` b
“ i. Then ai ` b “ ´c ` di, and hence a “ d, b “ ´c. Moreover, since ad ´ bc “ 1,
ci ` d
we getˆa2 ` b2 =1. Then
˙ there exists θ P r0, 2πq such that a “ cos θ and b “ sin θ. Hence
cos θ sin θ
γ“t P SO2 pRq as expected.
´ sin θ cos θ

Corollary 3.26. (Iwasawa

ˆ decomposition
˙ ˆ 1{2 ˙G
ˆ “ BK) Every ˙ element in SL2 pRq can
1 x y 0 cos θ sin θ
be written uniquely as with x P R, y P Rˆ
` and
0 1 0 y ´1{2 ´ sin θ cos θ
3 HECKE OPERATORS 67

θ P r0, 2πq.
˙ ˆ 1{2 ˆ ˙
y 0 1 x
Proof. For any γ P SL2 pRq, write γz “ x ` iy. Let “ γ1 . Then
0 y ´1{2 0 1
ˆ ˙
1 1´1 1´1 cos θ sin θ
γ i “ γi and hence γ γ fixes i. Consequently γ γ is of the form by
´ sin θ cos θ
the previous lemma.
The uniqueness of x and y is corollary to the fact γi “ x ` yi. The uniqueness of θ
then follows.
dxdy
Exercise 3.27. The measure dµpzq “ on H is left SL2 pRq-invariant.
y2

One can verify the previous lemma by hand. This measure comes from the (both left
dxdy
and right) Haar measure dθ on SL2 pRq. We refer to Appendix B for the existence
y2
of a such invariant measure. We remark that any left SL2 pRq-invariant measure on H is
a scalar multiple of µ since the action of SL2 pRq on H is transitive.

Exercise 3.28. Let D be the fundamental domain of SL2 pZqzH defined in Proposition
1.26. Show that µpDq ă 8.

π
In fact, one can show that µpDq “ .
3

3.5 Adjoint of Hecke operators

Definition 3.29. Let Γ be a congruence subgroup.

Let ϕ : H Ñ C be a function which is continuous, bounded and Γ-invariant.
Let DΓ˚ be a fundamental domain for ΓzH˚ .
Define ż ż
ϕpzqdµpzq :“ ϕpzqdµpzq.
XpΓq ˚
DΓ

One can show easily that it does not depend on the choice of DΓ˚ .
Ů Ť
For example, write SL2 pZq “ i t˘IuΓαi . Then we may take DΓ˚ to be i αi D˚ and
hence
ż ż ÿż
ϕpzqdµpzq “ ϕpzqdµpzq “ ϕpαi zqdµpzq.
Ť i
XpΓq αi D˚ D˚

The latter is a finite sum of finite complex numbers. In particular, we see that
3 HECKE OPERATORS 68

ş
µpDΓ˚ q “ 1dµpzq is finite. We define the latter to be
XpΓq
ż
VΓ :“ dµpzq,
XpΓq

and call it the volume of XpΓq.

We remark that VΓ “ rSL2 pZq : t˘IuΓsVSL2 pZq .

Lemma 3.30. Let f, g P Mk pΓq for some integer k such that at least one of them is
cuspidal. Define ϕ : z ÞÑ f pzqgpzqImpzqk a function on H. Then it is continuous,
bounded and Γ-invariant.

We leave the proof of this lemma to the reader.

Definition 3.31. For f, g as in the previous lemma, we define

1
ż
xf, gy ÞÑ f pzqgpzqImpzqk dµpzq.
VΓ
XpΓq

In particular, we define the Petersson inner product on Sk pΓq:

x, yΓ : Sk pΓq ˆ Sk pΓq Ñ C,
1
ż
xf, gy ÞÑ f pzqgpzqImpzqk dµpzq.
VΓ
XpΓq

We note that if f ‰ 0, then xf, f y ą 0. Hence the inner product is non-degenerate.

In particular, if T is a linear operator on Sk pΓq, then there exists a unique linear operator
T ˚ , called the adjoint of T , such that xT f, gyΓ “ xf, T ˚ gyΓ for any f, g P Sk pΓq.

Recall that an operator T is normal if T T ˚ “ T ˚ T . Moreover, a normal operator

acting on a finite dimensional Hermitian space is orthogonally diagonalizable.

Remark 3.32. Let Γ “ SL2 pZq and k ě 4 be an even integer. Recall that
ÿ
Ek pzq “ jpγ, zq´k .
γPΓ8 zΓ

Let f be a cusp form of weight k and level Γ. Then we have:

1
ż
xf, Ek y “ f pzqEpzqImpzqk dµpzq
VΓ
XpΓq

1
ż ÿ
“ f pzq jpγ, zq´k Impzqk dµpzq
VΓ
XpΓq γPΓ8 zΓ
3 HECKE OPERATORS 69

1
ż ÿ
“ f pzqjpγ, zq´k Impzqk dµpzq.
VΓ
XpΓq γPΓ8 zΓ

Impzq ´k
Note that f pγzqImpγzqk “ f pzqjpγ, zqk 2k
“ f pzqjpγ, zq Impzqk . We have:
|jpγ, zq|

1
ż ÿ
xf, Ek y “ f pγzqImpγzqk dµpzq
VΓ
ΓzH˚ γPΓ8 zΓ

1
ż
“ f pzqImpzqk dµpzq
VΓ
Γ8 zH˚
ż8 ż1
1 dy
“ p f px ` iyqdxqy k 2 “ 0.
VΓ y
0 0

If a congrence subgroup Γ is normal in SL2 pZq (e.g. Γ “ ΓpN q), we can define
ÿ
Ekδ pzq :“ jpγ, zq´k (3.7)
γPΓ8 zΓδ

for any k ě 3 and δ P SL2 pZq. We can show (as for the case of ΓpN q) that the above
definition only depends on the cusp Γpδ8q and is a modular form of weight k and level
Γ. Moreover, it is non-zero at one cusp and zero at other cusps. We call Eisk pΓq the
C-vector space generated by these forms. Then we have
à
Mk pΓq “ Sk pΓq Eisk pΓq.

Moreover, the above decomposition is orthogonal.

Proposition 3.33. Let Γ1 ă Γ2 be two congruence subgroups. Recall that we have the
natural inclusion Sk pΓ2 q ãÑ Sk pΓ1 q and the trace map: Sk pΓ1 q Ñ Sk pΓ2 q.

(a) If f, g P Sk pΓ2 q, then xf, gyΓ1 “ xf, gyΓ2 .

(b) If f P Sk pΓ1 q and g P Sk pΓ2 q, then VΓ1 xf, gyΓ1 “ VΓ2 xtracepf q, gyΓ2 .

Proof. For simplicity, assume that both Γ1 and Γ2 contain ´I, or none of them contains
´I.
Ů Ů Ů
We write Γ2 “ Γ1 βj . Write SL2 pZq “ t˘IuΓ2 αi . Then SL2 pZq “ t˘IuΓ1 βj αi .
j i i,j

(a) Write ϕ “ f gImpzqk . Then

1 ÿÿ
ż
xf, gyΓ1 “ ϕpβj αi zqdµpzq
VΓ1 i j
D˚
3 HECKE OPERATORS 70

1 ÿÿ
ż
“ ϕpαi zqdµpzq
VΓ1 i j
D˚
rΓ2 : Γ1 s ÿ
ż
“ ϕpαi zqdµpzq
VΓ1 i
D˚
1 ÿ
ż
“ ϕpαi zqdµpzq
VΓ2 i
D˚
“ xf, gyΓ2 .
ř
(b) Note that tracepf q “ f |k βj . Hence
j
ÿÿż
VΓ2 xtracepf q, gyΓ2 “ pf |k βj qpαi zqgpαi zqImpαi zqk dµpzq
i j D˚
ÿÿż
“ jpβj , αi zq´k f pβj αi zqjpβj , αi zq´k gpβj αi zqImpαi zqk dµpzq
i j D˚
ÿÿż
“ f pβj αi zqgpβj αi zqImpβj αi zqk dµpzq
i j D˚

“ VΓ1 xf, gyΓ1 .

Definition 3.34. Let Γ be a congruence subgroup and α P GL2 pQq` . We may define the
integral for a continuous, bounded, α´1 Γα-invariant function ϕ : H Ñ C over α´1 ΓαzH˚
as follows:
Let Dα˚´1 Γα be a fundamental domain for α´1 ΓαzH˚ .
Define ż ż
ϕpzqdµpzq :“ ϕpzqdµpzq.
α´1 ΓαzH˚ D˚´1
α Γα

Similarly to the case of XpΓq, one can show easily that it does not depend on the
choice of Dα˚´1 Γα .
Recall that the map z ÞÑ αz from H to H induces a bijection map α´1 ΓαzH˚ Ñ
ΓzH˚ . Let DΓ˚ be any fundamental domain for ΓzH˚ . Then α´1 DΓ˚ is a fundamental
domain for α´1 ΓαzH˚ . We obtain that:
ż ż
ϕpzqdµpzq “ ϕpα´1 zqdµpzq (3.8)
α´1 ΓαzH˚ ΓzH˚

Lemma 3.35. Let α P GL2 pQq` .

(a) If α´1 Γα Ă SL2 pZq, then Vα´1 Γα “ VΓ .

Consequently, rSL2 pZq : α´1 Γαs “ rSL2 pZq : Γs.
3 HECKE OPERATORS 71

(b) We have rΓ : α´1 Γα X Γs “ rΓ : αΓα´1 X Γs. Let us call this index n.

(c) Then there exists α1 , ¨ ¨ ¨ , αn P ΓαΓ such that

n
ğ n
ğ
ΓαΓ “ Γαi “ αi Γ.
i“1 i“1

Proof. (a) It is enough to take ϕ to be the constant function 1 on H in equation 3.8.

The second part follows from the first part and the fact that ´I P Γ if and only if
´I P α´1 Γα.

(b) Note that αΓα´1 X Γ is a congruence subgroup and α´1 pαΓα´1 X Γqα Ă SL2 pZq.
We can apply pbq and get rSL2 pZq : α´1 Γα X Γs “ rSL2 pZq : αΓα´1 X Γs. Hence
rΓ : α´1 Γα X Γs “ rΓ : αΓα´1 X Γs.

By Lemma 3.5, we know #ΓzΓαΓ “ rΓ : α´1 Γα X Γs and #ΓzΓα´1 Γ “ rΓ :

αΓα´1 X Γs. Hence #ΓzΓαΓ “ #ΓzΓα´1 Γ “ n. Note that the inverse map gives
a bijection between ΓzΓα´1 Γ and ΓαΓ{Γ. Hence #ΓαΓ{Γ also equals n.

Ů Ů
(c) Take β1 , ¨ ¨ ¨ , βn , γ1 , ¨ ¨ ¨ , γn P ΓαΓ such that ΓαΓ “ i Γβi and ΓαΓ “ i γi Γ.

We claim
Ů that for any i, Γβi X γi Γ ‰ ∅.Ů Otherwise there exists i such that
Γβi Ă j‰i γj Γ and hence ΓαΓ “ Γβi Γ Ă j‰i γj Γ which is not possible.

Then we can find αi P ΓβŮ

i X γi Γ for each
Ů i. Note that Γαi “ Γβi and αi Γ “ γi Γ.
We conclude that ΓαΓ “ ni“1 Γαi “ ni“1 αi Γ as expected.

Proposition 3.36. For α P GL2 pQq` . Let α1 :“ detpαqα´1 .

(a) If α´1 Γα Ă SL2 pZq, then for f P Sk pΓq and g P Sk pα´1 Γαq, we have:

xf |k α, gyα´1 Γα “ xf, g|k α1 yΓ . (3.9)

(b) For any f, g P Sk pΓq, we have:

xf |k ΓαΓ, gyΓ “ xf, g|Γα1 ΓyΓ . (3.10)

In other words, p|k ΓαΓq˚ “ |k Γα1 Γ.

Proof. (a) We first notice that α1 z “ α´1 z for any z P H. Then by the previous lemma
we have:
ż
xf |k α, gyα´1 Γα “ pf |k αqpzqgpzqImpzqk dµpzq
α´1 ΓαzH˚
3 HECKE OPERATORS 72

ż
“ detpαqk{2 jpα, zq´k f pαzqgpzqImpzqk dµpzq
α´1 ΓαzH˚
ż
“ detpαqk{2 jpα, α´1 zq´k f pzqgpα´1 zqImpα´1 zqk dµpzq
ΓzH˚
ż
“ detpαqk{2 jpα, α1 zq´k f pzqgpα1 zqImpα1 zqk dµpzq
ΓzH˚
ż
“ detpαqk{2 jpα, α1 zq´k f pzqgpα1 zq detpα1 qk |jpα1 , zq|´2k Impzqk dµpzq.
ΓzH˚

Here we used the fact that Impα1 zq “ detpα1 q|jpα1 , zq|´2 Impzq.
Note that detpα1 q “ detpαq “ jpdetpαqI2 , zq “ jpαα1 , zq “ jpα, α1 zqjpα1 , zq. We
obtain that:

detpαqk{2 jpα, α´1 zq´k detpα1 qk |jpα1 , zq|´2k “ detpα1 qk{2 jpα1 , zqk |jpα1 , zq|´2k
“ detpα1 qk{2 jpα1 , zq´k

Consequently, we have
ż
xf |k α, gyα´1 Γα “ f pzqdetpα1 qk{2 jpα1 , zq´k gpα1 zqImpzqk dµpzq
ΓzH˚
ż
“ f pzqpg|k α1 qpzqImpzqk dµpzq
ΓzH˚

“ xf, g|k α1 yΓ .

By part (c) of the previous lemma, there exits α1 , α2 , ¨ ¨ ¨ , αn P ΓαΓ such that
n
ğ n
ğ
ΓαΓ “ Γαi “ αi Γ.
i“1 i“1
Ůn ´1
In particular, we have detpαi q “ detpαq for each i. Moreover, Γα´1 “ i“1 Γαi .
Define αi1 :“ detpαi qαi´1 . Then
n
ğ n
ğ
Γα1 “ Γ detpαqα´1 “ detpαqΓα´1 “ detpαq Γαi´1 “ Γαi1 . (3.11)
i“1 i“1

pΓ X αi´1 Γαi q. For each i, f |k αi P

ř Ş
Recall that f |k ΓαΓ “ f |k αi . Let Γ1 :“
i i
Sk pΓ X αi´1 Γαi q Ă Sk pΓ1 q.
Then by Proposition 3.33 (a), we have
ÿ ÿ
xf |k ΓαΓ, gyΓ “ x f |k αi , gyΓ1 “ xf |k αi , gyΓ1 .
i i
3 HECKE OPERATORS 73

pαi Γαi´1 q X Γ. By Proposition 3.33 (a) again along with part (a) of the
Ş
Let Γ2 “
i
current proposition, we have for each i,

xf |k αi , gyΓ1 “ xf |k αi , gyΓXα´1 Γαi “ xf, g|k αi1 yαi Γα´1 XΓ “ xf, g|k αi1 yΓ2 .
i i

Finally, we deduce from equation (3.11) that:

ÿ
xf |k ΓαΓ, gyΓ “ xf, g|k αi1 yΓ2 “ xf, g|Γα1 Γ yΓ2 “ xf, g|Γα1 Γ yΓ .
i

Corollary 3.37. Let Γ “ Γ1 pN q.

(a) For d P pZ{N Zqˆ , we have xdy˚ “ xd´1 y.

For a prime p ∤ N , we have Tp˚ “ xpy´1 Tp .
ˆ ˙ ˆ ˙
0 ´1 0 ´1
(b) It is easy to verify that Γ1 pN q “ Γ1 pN q . Hence
N 0 N 0
ˆ ˙
0 ´1
wN :“ |k
N 0
˚ “ p´1qk w
defines a linear operator on Sk pΓ1 pN qq. We have wN ˚ “
N and T
´1
wN T wN for T “ Tn or xny for any n ě 1.

Proof. (a) The first part is direct.

ˆ ˙For the second part, the previous proposition
p 0
implies that Tp˚ “ |k Γ1 pN q Γ1 pN q.
0 1
Take r, s P Z such that ps ´ rN “ 1. Then
ˆ ˙´1 ˆ ˙ˆ ˙ ˆ ˙
1 r 1 0 p r p 0
“ .
N ps 0 p N s 0 1
ˆ ˙ ˆ ˙
1 r p r
Since P Γ1 pN q and P Γ0 pN q, we have
N ps N s
ˆ ˙ ˆ ˙ˆ ˙ ˆ ˙ ˆ ˙
p 0 1 0 p r 1 0 p r
Γ1 pN q Γ1 pN q “ Γ1 pN q Γ1 pN q “ Γ1 pN q Γ1 pN q .
0 1 0 p N s 0 p N s

We can deduce easily that Tp˚ “ Tp xsy “ Tp xp´1 y “ xp´1 yTp .

(b) We leave this part as an exercise.

ˆ ˙ ˆ ˙´1 ˆ ˙ˆ ˙
p 0 0 ´1 1 0 0 ´1
(Hint: show that “ .)
0 1 N 0 0 p N 0
3 HECKE OPERATORS 74

Recall that an operator T is normal if T T ˚ “ T ˚ T . Moreover, a normal operator

acting on a finite dimensional Hermitian space is orthogonally diagonalizable. The
above corollary tells us that all xny and Tn , pn, N q “ 1, are normal. Moreover, they all
commute with each other. Hence they can be diagonalized at the same time. In other
words, we get:

Theorem 3.38. The vector space Sk pΓ1 pN qq has an orthogonal basis of simultaneous
eigenforms for the Hecke operators txny, Tn | pn, N q “ 1u.
A REPRESENTATION OF FINITE ABELIAN GROUP 75

A Representation of finite abelian group

Let G be a finiteśabelian group. By the elementary divisor theorem, the group G is

isomorphic to Z{di Z with some positive integers n and d1 , ¨ ¨ ¨ , dn . We define
1ďiďn
Gp :“ HompG, S 1 q, called the dual of G. Then the group G p “ Homp ś Z{di Z, S 1 q is
ś 1ďiďn
also isomorphic to Z{di Z. We get a non-canonical isomorphism between G and its
1ďiďn
dual G.
p

Lemma A.1. Let G be a finite abelian group acting on a finite-dimensional C-vector

space À p define V χ :“ tv P V | gv “ χpgqv, for any g P Gu. Then
V. For each χ P G,
V “ V χ.
χPG
p

1 ř
Proof. For each χ, define πχ : V Ñ V, v ÞÑ χphq´1 hv.
|G| hPG
We first verify that πχ V “ V χ .
1 ř 1 ř
For any v P V and any g P G, gpπχ vq “ χphq´1 ghv “ χpg ´1 hq´1 hv “
|G| hPG |G| hPG
1 ř
χpgq χphq´1 hv “ χpgqpπχ vq. Hence πχ V Ă V χ .
|G| hPG
Conversely, if v P V χ , then πχ v “
ř ř
χphq´1 hv “ χphq´1 χphqv “ v. Hence
hPG hPG
πχ V Ą V χ as expected.
It remains to show that πχ2 “ πχ ,
ř
πχ “ 1V and πχ ˝ πχ1 “ 0 if χ ‰ χ1 P G.
p
χPG
p

1 ř ř 1 ř
Indeed, πχ2 “ 2
χphh1 q´1 hh1 “ χphq´1 h “ πχ .
|G| hPG h1 PG |G| hPG
ř 1 ř ř 1 ř ř
Moreover, πχ “ χphq´1 h “ p χphq´1 qh.
χPG
p |G| χPGp hPG |G| hPG χPGp
p such that χ0 phq ‰ 1. We have χ0 phq´1 ř
If h ‰ Id, then there exists χ0 P G χphq´1 “
χPG
p
ř ř ř
χ0 χphq´1 “ χphq´1 . Hence χphq´1 “ 0.
χPG
p χPG
p χPG
p

ř 1 ř |G|
p
We deduce that πχ “ 1“ “ 1 as expected.
χPG
p |G| χPGp |G|

1 ř ř 1 ř ř
Finally, if χ ‰ χ1 , then πχ ˝πχ1 “ 2
χphq˝χ1 ph1 qhh1 “ χph{h1 qχ1 ph1 qh “
|G| hPG h1 PG1 |G|2 hPG h1 PG1
1 ř ř 1
p χ {χph1 qqχphqh.
|G|2 hPG h1 PG1
Since χ1 {χ ‰ 1G , there exists g P G such that χ1 {χpgq ‰ 1. Moreover, note that
B INVARIANT MEASURE 76

ř ř ř ř
χ1 {χpgqp χ1 {χph1 qq “ χ1 {χpgh1 q “ χ1 {χph1 q, we get χ1 {χph1 q “ 0 and
h1 PG1 h1 PG1 h1 PG1 h1 PG1
hence πχ ˝ πχ1 “ 0 as expected.

B Invariant measure

Let G be a locally compact topological group. A Borel measure on G is called a left Haar
measure if it is finite on compact subset, strictly positive on any non-empty open subset,
and left invariant. Any locally compact topological group admits a left Haar measure
unique up to a scalar multiple. Let µL be a left Haar measure on G. In particular, for
any measurable set A, and and g P G, we have µL pgXq “ µL pXq.

For any g P G, the measure µg pXq :“ µL pXgq is another left Haar measure. Hence
there exists a positive real number δG pgq, such that µg “ δG pgq´1 µL . Namely, for any
measurable set X, we have

µL pXgq “ δG pgq´1 µL pXq. (B.1)

It is easy to verify that δG : G Ñ Rˆ

` is a group morphism, called the modulus character.

Moreover, for any g P G and X measurable, δG pgqµL pXgq “ µL pXq “ δG p1qµL pXq.
We see that δG µL is a right Haar measure on G.

Hence for any g 1 P G,

ż ż
f pgg 1 qdµL pgqq “ δG pgq´1 f pgg 1 qpδG pgqdµL pgqq
gPG gPG
ż
“ δG pgg 1´1 q´1 f pgqpδG pgqdµL pgq
gPG
ż
1
“ δG pg q f pgqdµL pgq. (B.2)
gPG

ş
Moreover, the linear functional f ÞÑ G f pg ´1 qdµL pgq also defines a right Haar mea-
sure on G. Since the right Haar measure is unique up to scalars as well, one can show
that: ż ż
f pg ´1 qdµL pgq “ f pgqδG pgqdµL pgq. (B.3)
G G

When the modulus character is trivial, we say that G is unimodular. In this case, a
left Haar measure is also a right Haar measure. For example, if G is abelian or compact
then it is unimodular. For the compact case, we only need to take X “ G in equation
(B.1).
B INVARIANT MEASURE 77

Now let H be a closed subgroup of G. Let µL be a left Haar measures on G. We fix

µL,˚ any left Haar measure on ˚ where ˚ “ G, H. We remark that µL,G |H may not be
a Haar measure on H, for example, consider the case G “ R and H “ Z. The quotient
G{H is a locally compact topological space endowed with the quotient topology. We
want to to show:

Theorem B.1. There exists a left G-invariant measure ν on G{H if and only if δG |H “
δH .

Sketch of the proof :

One one hand, we assume that such ν exists. Then the linear functional
ż ż
f P CC pGq ÞÑ f pghqdµL,H phqdνpgq
G{H
H

defines a left Haar measure on G. Consequently, we may normalize µL,G such that for
any f P CC pGq,
ż ż ż
f pgqdµL,G pgq “ f pghqdµL,H phqdνpgq. (B.4)
G G{H
H

In particular, we apply the previous equation to the function g P G ÞÑ f pgh1 q, with

any fixed h1 P H, we get
ż ż ż
1
f pgh qdµL,G pgq “ f pghh1 qdµL,H phqdν.
G G{H
H

Recall that ż ż
1 1
f pgh qdµL,G pgq “ δG ph q f pgqdµL,G pgq
G G
and ż ż
1 1
f pghh qdµL,H phq “ δH ph q f pghqdµL,H phq
H H

by equation (B.2). Hence necessarily we have δG |H “ δH .

Conversely, we assume that δG |H “ δH . We first remark that the map

ż
Ψ : CC pG, Cq Ñ CC pG, Cq, f ÞÑ pg ÞÑ f pghqdµL,H phqq
H

induces a surjective map: Φ : CC pG, Cq Ñ CC pG{H, Cq.

REFERENCES 78

ş
One can show that for any f P CC pG, Cq, the integral f pgqdµL pgq only depends
gPG ş
on Φpf q. For that, it is enough to show that if Φpf q “ 0, then f pgqdµL pgq “ 0.
gPG ş
This
ş is corollary to the fact that for any f, f 1 P CC pG, Cq, we have Ψpf qf 1 dµL,G “
f Ψpf 1 qdµL,G (Hint: use the fact δG |H “ δH and equation (B.3)).

ş
Therefore the map F Ñ f pgqdµL pgq, for F P CC pG{H, Cq, where f is any func-
gPG
tion in Φ´1 pF q, defines a measure on G{H, called ν. Apparently it is G-invariant.

We refer to [Bou04] chapter 7 for more details.

Example B.2. Let Gˆ“ SL2˙pRq. ˆ 1{2Recall that

˙ ˆ(Corollary 3.26)
˙ any element in G can be
1 x y 0 cos θ sin θ
written uniquely as with x P R, y P Rˆ` and
0 1 0 y ´1{2 ´ sin θ cos θ
dxdy
θ P r0, 2πq. One can verify that dθ 2 is a both left and right Haar measure on G. In
y
particular, G is unimodular.
ˆ ˙
cos θ sin θ
Let H :“ SO2 pRq “ t | θ P r0, 2πqu be a maximal compact subgroup
´ sin θ cos θ
of G. Then H is also unimodular since it is compact.
The condition δG |H “ δH is then satisfied automatically and hence there exists a
G-invariant measure on G{H – H.

References

[Apo90] T. M. Apostol, Modular functions and dirichlet series in number theory, GTM
41, Springer-Verlag, 1990. 6

[Bou04] N. Bourbaki, Integration ii, Springer-Verlag, 2004. 79

[Bum98] D. Bump, Automorphic forms and representations, Cambridge University

Press, 1998. 1

[DS05] F. Diamond and J. Shurman, A first course in modular forms, Springer-Verlag,

2005. 1, 51

[Gun66] Lectures on riemann surfaces, Princeton Mathematical Notes, Princeton, 1966.

33

[Kat04] Y. Katznelson, An introduction to harmonic analysis, Cambridge University

Press, 2004. 7

[Kob84] N. Koblitz, Introduction to elliptic curves and modular forms, GTM 97,
Springer, 1984. 1, 51
REFERENCES 79

[Mir95] Rick Miranda, Algebraic curves and riemann surfaces, Graduate Studies in
Mathematics, vol. 5, Amer. Math. Soc., 1995. 28, 31

[Miy89] T. Miyake, Modular forms, Springer, 1989. 1, 67

[Sil85] J. H. Silverman, The arithmetic of elliptic curves, GTM 106, Springer-Verlag,

1985. 6, 33