Modular Forms

Sarah Zerbes

Spring semester 2021/22

 The notes are based on the lecture notes of David Loeffler and Marc Masdeu.

2
Contents
0 Prologue 4

1 The modular group 6

1.1 The upper half-plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 The modular group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Modular forms and modular functions . . . . . . . . . . . . . . . . . . . 10
1.4 Eisenstein series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.5 The valence formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.6 Applications to modular forms . . . . . . . . . . . . . . . . . . . . . . . . 19
1.7 The q-expansion of ∆ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2 Modular forms of higher level 27

2.1 Congruence subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2 Fundamental domains and cusps . . . . . . . . . . . . . . . . . . . . . . . 28
2.3 Weakly modular forms for congruence subgroups . . . . . . . . . . . . . . 33
2.4 q-expansion at ∞ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.5 q-expansion at a cusp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.6 The valence formula in arbitrary levels . . . . . . . . . . . . . . . . . . . 36
2.7 Eisenstein series revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3 Hecke operators 44
3.1 Double cost operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.2 The Hecke algebra of Γ1 (N ) . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.2.1 Diamond operators . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.2.2 Hecke operators on q-expansions . . . . . . . . . . . . . . . . . . . 52
3.2.3 The Hecke algebra . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3 The Petersson product . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.4 Hecke operators and the Petersson product . . . . . . . . . . . . . . . . . 63
3.5 Old and new modular forms . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.6 L-functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.6.1 Basic definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.6.2 L-functions of cusp forms . . . . . . . . . . . . . . . . . . . . . . 75
3.6.3 Relation to elliptic curves . . . . . . . . . . . . . . . . . . . . . . 77

3
0 Prologue
Example 0.0.1. Let z ∈ C, =(z) > 0. Let q = e2πiz and define Ramanujan’s tau
function Y
∆(z) = q · (1 − q n )24 .
n∈N

This is one of the simplest examples of a modular form. Note that we can ”multiply
out” the product above which leads us to
X
∆(z) = τ (n)q n
n∈N

for some integers τ (n).

Facts 0.0.2.
(1) Known to Weierstrass, 1850:
 
−12 1
∆(z) = z ·∆ −
z

(2) Ramanujan proved in 1916 that the integers τ (n) satisfy the equation
X
τ (n) = d11 mod 691.
d|n

(3) Ramanujan also conjectured τ (nm) = τ (n)τ (m) for n, m coprime. This was proved
by Mordell in 1917.
(4) In 1972 Swinnerton-Dyer proved τ (n) satisfies congruences like (2) modulo 2, 3, 5,
7, 23 and 691 but no other primes.
(5) Ramanujan conjectured in 1916 for p prime holds |τ (p)| < 2 p11/2 . This was proved
in 1974 by Deligne.
(6) The quantity
τ (p)
∈ [−1, 1]
2p11/2
√
is distributed in the interval [−1, 1] with density function proportional to 1 − x2 .
This was conjectured by Sato and Tate (1960s) and proved by Barnet-Lamb, Ger-
aghty, Harris and Taylor in 2009 using Bau Chau Ngo’s Fundamental Lemma which
got Ngo the 2010 Fields Medal.

4
Example 0.0.3. We now consider another modular form
∞
Y
f (z) = q (1 − q n )2 (1 − q 11n )2
n=1
= q − 2q 2 − q 3 + 2q 4 + q 5 + 2q 6 + . . .
X∞
0
= a(n)q n with a(n) ∈ N
n=1

We will later prove the following results:

Theorem.

1. We have a(mn) = a(m)(n) for all m, n ≥ 1 with (m, n) = 1.

√
2. We have |a(p)| ≤ 2 p for all primes p.

It turns out that this modular form is closely related to the elliptic curve

E : Y 2 + Y = X 3 − X 2 − 10X − 20.

For p prime, denote by N (p) the number of points on the elliptic curve in Fp . It is easy
to see heuristically tat N (p) ' p.

Theorem. (Hasse) We have

√
|p − N (p)| ≤ 2 p.

The theory of modular forms allows one to prove that the elliptic curve E and the
modular form f ‘correspond’ to each other in the following sense:

Theorem. For all primes p, we have

a(p) = p − N (p).

In particular, using the properties of the modular form f , we can easily calculate the
quantity N (p) for all p, so f ‘knows’ about the behaviour of the elliptic curve over Fp .
We say that the elliptic curve E is modular. It is generally not too difficult to attach
an elliptic curve to a modular form (this is called ”Eichler–Shimura”); however, it is
very difficult indeed to reverse this process, and this is the basis of Andrew Wiles’ work
on Fermat’s Last Theorem. The proof of this result was later completed by Breuil–
Conrad–Diamond–Taylor. I will talk a bit more about this when we discuss L-functions
of modular forms.

5
1 The modular group
1.1 The upper half-plane
Definition 1.1.1. Let H = {z ∈ C : =(z) > 0} the upper half-plane.
Proposition 1.1.2. The special linear group SL2 (R) = {A ∈ GL2 (R) : det(A) = 1}
acts on H via  
a b az + b
.z = .
c d cz + d
Proof. For z ∈ H is =(z) > 0 and either c or d is nonzero, so cz + d 6= 0. Moreover
 
az + b 1
= = = ((az + b)(cz + d)) .
cz + d |cz + d|2
Say z = x + iy for x, y ∈ R.
 
az + b 1 2


= = = (ax + b)(cx + d) + acy +i (ad − bc) y
cz + d |cz + d|2 | {z } | {z }
∈R =1
1
= =(z) > 0
|cz + d|2
Therefore az+b


a b ∈ SL (R).
∈ H for any z ∈ H, 2
cz+d
1 0

c d
Also it is easy to check that 0 1 z = z and A(Bz) = (AB)z for any z ∈ H and for
any A, B ∈ SL2 (R). Thus SL2 (R) acts on H.
Note 1.1.3. The matrix −1 0


0 −1 ∈ SL2 (R) acts trivially on H, so the action of SL2 (R)
on H factors through the quotient PSL2 (R) = SL2 (R)/(±1), the projective special
linear group.
Definition 1.1.4. The automorphy factor is the function
j : SL2 (R) × H → C,
 
a b
(g, z) 7→ cz + d for g =
c d
Proposition 1.1.5. For any k ∈ Z, we can define a right action of SL2 (R) on the set
of holomorphic functions H → C given by
(f |k g) (z) := j(g, z)−k f (gz)


where f : H → C holomorphic, g = ac db ∈ SL2 (R). We will call this the weight k
action.

6
Proof. Firstly we need to show that f |k g is a well-defined holomorphic function H → C.
But this is obvious since cz + d 6= 0 and gz ∈ H for all z ∈ H. Clearly also the
equation f |k 1 = f holds. Therefore it remains to show (f |k g)|k h = f |k (gh) for arbitrary
g, h ∈ SL2 (R). The left hand side of the equation can be rewritten as
(f |k g)|k h = j(h, z)−k ((f |k g)(hz))
= j(h, z)−k j(g, hz)−k f (g(hz))
and the right hand side results in
f |k (gh) = j(gh, z)−k f ((gh)z).
We already know (gh)z = g(hz). So it remains to show j(gh, z) = j(h, z)j(g, hz). This
is the so called cocycle relation and can be checked easily.

1.2 The modular group

Definition 1.2.1. The modular group is the group
   
a b
SL2 (Z) = A = ; a, b, c, d ∈ Z, det(A) = 1 .
c d
The projective modular group is PSL2 (Z) = SL2 (Z)/(±1).
0 −1

1 1


Theorem 1.2.2. (a) The group SL2 (Z) is generated by S = 1 0 and T = 0 1 .
(b) Every orbit of SL2 (Z) acting on H contains a point of the set D defined by
 
1 1
D = z ∈ H : − ≤ <(z) ≤ and |z| ≥ 1 .
2 2

(c) If z ∈ D and gz ∈ D for some g ∈ SL2 (Z), then either g = ±1 and gz = z or z

lies on the boundary of D.
(d) The stabilizer of z ∈ H in PSL2 (Z) is trivial unless z is in the orbit of i or in the
orbit of ρ = e2πi/3 .
Proof. We will prove all of these statements in 4 steps using a very elegant argument of
Serre. Let G = SL2 (Z) and G0 = hS, T i ≤ G.
Step 1. Every G0 orbit in H contains a point of D.
Proof of Step 1. Let z ∈ H. Since |cz+d| ≥ |c =(z)| and |cz+d| ≥ |c
<(z)+d| there exist
only finitely many (c, d) ∈ Z2 such that |cz + d| < 1. Recall =( ac db z) = |cz + d|−2 =(z).
This implies there are only finitely many g ∈ G0 such that =(gz) > =(z). So the G0
orbit of z contains a point of maximal imaginary part. Let this point be z.
We can assume <(z) ∈ [− 21 , 12 ] since T z = z + 1. Moreover =(Sz) = |z|−2 =(z). But
z is a point of maximal imaginary part in the orbit of G0 , so we get |z|−2 =(z) ≤ =(z)
implying |z| ≥ 1. Thus z ∈ D. Clearly this proves part (b) of the theorem.

7
Step 2. If z ∈ D and gz ∈ D, where g ∈ G, then one of the following holds:

1. g = ±Id

2. g = ±S and |z| = 1

3. g = ±T and <(z) = − 12 , or g = ±T −1 and <(z) = 21

4. g = ±ST = ± 01 −1 or g = ±T −1 S = ± −1 −1
or g = ±ST −1 S = ± −1 0




1 1 0 −1 −1
and z = ρ

5. g = ±T S = ± 11 −1 or g = ±ST −1 = ± 01 −1 or g = ±ST S = ± −1 0




0 −1 1 −1 and
z =ρ+1

Proof of Step 2. Let z ∈ D and g = ac db ∈ G such that z 0 = gz ∈ D. Being free

to replace g by g −1 and z by z 0 we can assume that =(z 0 ) ≥ =(z). Again recalling

=(gz) = |cz + d|−2 =(z) we gain |cz + d| ≤ 1. Furthermore we have
√
3
|cz + d| ≥ |c| =(z) ≥ |c| =(ρ) = |c|.
2
√
Thus |c| ≤ 2/ 3 < 2. As c ∈ Z we get c = 0 or c = ±1.

• Let c = 0. Since 1 ≥ |cz + d| = |d| we have d = 0 or d = ±1.
But c = d = 0 is

impossible. So d = ±1 and hence a = ±1. Therefore g = ±1 b
0 ±1 is the translation
by b. But since  
1 1
<(z), <(gz) ∈ − , ,
2 2
this implies that b = 0 or b = ±1. So either g = ±Id (case 1) or g = ±T and
<(z) = − 12 or g = ±T −1 and <(z) = 12 .

• Let c = 1. Assuming |d| ≥ 2 leads to the following contradiction:

1 3
1 ≥ |cz + d| = |z + d| ≥ |d| − <(z) ≥ |d| − ≥
2 2
Thus we have d = 0 or d = ±1.
Let d = 0. Then 1 ≥ |cz + d| = |z|. On the other hand |z| ≥ 1 as z ∈ D and
therefore |z| = 1 (cases 2, 4 or 5 – exercise sheet 1).
Let d = 1. Then 1 ≥ |z + 1|. This is only possible for z ∈ D if z = ρ (exercise).
Since a − b = 1, we deduce that wither (a, b) = (1, 0) or (a, b) = (0, −1) (case 4).
Analogue d = −1 implies z = ρ + 1 (case 5).

• The case c = −1 is analogous to the case c = 1.

Since there are no further cases this shows Step 2 (it remains to check the matrices in
case 4 and 5 – see exercise sheet 1) and therefore part (c) of the theorem.

8
Step 3. Let z ∈ D such that the stabilizer Gz of z is not ±Id. Then z = i, z = ρ or
z = ρ + 1.

Proof of Step 3. This follows directly from Step 2 by checking gz = z for all possible
g’s. Step 3 proves part (d) of the theorem.

Step 4. It remains to show that SL2 (Z) is generated by S and T .

Proof of Step 4. Let g ∈ G and let z be an arbitrary point of the interior of D. Then
gz ∈ H and by Step 1 exists g 0 ∈ G0 such that g 0 (gz) ∈ D. Moreover Step 2 implies that
either g 0 g ∈ {± Id} or z is on the boundary of D which is by assumption not the case.
Thus either g 0 g = Id or g 0 g = − Id. Since S 2 = − Id ∈ G0 , we deduce that g ∈ G0 , so
SL2 (Z) is generated by S and T . This proves part (a) of the theorem.

Therefore the theorem is proved.

Remark 1.2.3. We have seen in the proof of Theorem 1.2.2 that SL2 (Z) is generated
by the elements S and T . These satisfy the relations

S 4 = Id (ST )3 = S 2 ,

and one can show that these generate all the relations, i.e. that

hS, T | S 4 , S −2 (ST )3 i

is a presentation of the group SL2 (Z).

Remark 1.2.4. The set D is called the fundamental domain. The figure below
represents D itself and the transforms of D by some group elements of SL2 (Z). Part (c)
of the theorem shows that two sets gD and g 0 D where g, g 0 ∈ SL2 (Z) are either equal
(if g 0 = ±g) or only intersect along their edges. Furthermore part (a) implies that H is
covered by the sets {gD : g ∈ SL2 (Z)}: they form a tesselation of H.

9
1.3 Modular forms and modular functions
Definition 1.3.1. A weakly modular function of weight k and level 1 is a mero-
morphic function H → C such that f |k α = f for all α ∈ SL2 (Z), or equivalent
 
az + b
f = (cz + d)k f (z)
cz + d


for all z ∈ H and for all ac db ∈ SL2 (Z).

Note 1.3.2. Since SL2 (Z) is generated by the matrices S and T , it is sufficient to check
invariance under these two matrices, i.e. that

f (z + 1) = f (z) and f (−1/z) = z k f (z)

for all z ∈ H.

Lemma 1.3.3. There are no nonzero weakly modular functions of odd weight.

Proof. Let k be odd and let f be a weakly modular function of weight k. As shown
in (2) we have f (z) = f (z
+ 1) for all z ∈ H. Moreover we get f (z) = −f (z + 1) for
all z ∈ H, since f |k −1 −1
0 −1 = −f (· + 1). So f (z) = −f (z) and thus f (z) = 0 for all
z ∈ H.

Define the function

q : H → C,
z 7→ exp(2πiz).

Note 1.3.4. Now let f be weakly periodic of weight k. Then f is periodic with period
1, so it can be written in the form

f (z) = f˜(exp(2πiz)),

where f˜ is a meromorphic function on the punctured unit disk

D∗ = {q ∈ C : 0 < |q| < 1}.

Note 1.3.5. The function f˜ is defined by

 
log q
f˜(q) = f .
2πi

Observe that the logarithm is multi-valued, but choosing a different value of the logar-
ithm is the same as adding an integer to log
2πi
q
. The periodicity of f hence implies that
f˜(q) does not depend on the chosen value of the logarithm.

10
Note 1.3.6. Any weakly modular function can be written as
∞
X
f (z) = an q n
n=−∞

for some an ∈ C where q = e2πiz ; we call this the q-expansion of f . This is just the
Laurent series of f˜ around q = 0, which converges for 0 < |q| < ε for ε sufficiently small
(⇔ =(z)  0)

Definition 1.3.7.

• We say that f is meromorphic at ∞ if an = 0 for n < −N and some N ∈ N.

• We say that f is holomorphic at ∞ if an = 0 for n < 0. In this case, we define the

value of f at ∞ to be f (∞) = f˜(0) = a0 .

Definition 1.3.8. Let f be a weakly modular function of weight k and level 1.

1. If f is meromorphic on H ∪ {∞} we say f is a modular function (of weight k

and level 1).

2. If f is holomorphic on H ∪ {∞} we say f is a modular form (of weight k and

level 1).

3. If f is holomorphic on H ∪ {∞} and f (∞) = 0 we say f is a cuspidal modular

form or cusp form.

Note 1.3.9. If f and g are modular forms (resp. modular functions) of level 1 and
weights k and `, then the product f g is a modular form (resp. modular function) of
weight k + `.

1.4 Eisenstein series

Definition 1.4.1. Let k ≥ 4 even. Define the Eisenstein series of weight k to be
the function Gk : H → C given by
X 1
Gk (z) = . (1.1)
(mz + n)k
(m,n)∈Z2 \{0}

Recall the following result from complex analysis:

Proposition 1.4.2. Let U be an open subset of C, and let (fn )n ≥ 0 be a sequence of

holomorphic functions on U that converges uniformly on compact subsets of U . Then
the limit function U → C is holomorphic.

11
Lemma 1.4.3. The series defining Gk (z) converges absolutely and uniformly on subsets
of H of the form
Rr,s = {x + iy : |x| ≤ r, y ≥ s}.
It hence converges to a holomorphic function on H.

Proof. Let z = x + iy ∈ Rr,s . We have

|mz + n|2 = (mx + n)2 + m2 y 2 ≥ (mx + n)2 + m2 s2 .

For fixed m and n, we distinguish the cases |n| ≤ 2r|m| and |n| ≥ 2r|m|. In the first
case, we have

s2 2 s2
 2 2
2 2 2 2 s s
|mz + n| ≥ m s ≥ m + 2
n ≥ min , 2 · (m2 + n2 ).
2 2(2r) 2 8r

In the second case, the triangle inequality implies

 2  
2 2 2 2 |n| 2 2 1 2
|mz + n| ≥ (|mx| − |n|) + m s ≥ + m s ≥ min ,s · (m2 + n2 ).
2 4

Combining both cases and putting

s2 s2 1 2
 
c = min , , ,s ,
2 8r2 4

we get the inequality

|mz + n| ≥ c1/2 (m2 + n2 )1/2 for all m, n ∈ Z, z ∈ Rr,s .

Hence for all z ∈ Rr,s , we have

1 X 1
Gk (z) ≤ .
ck/2 (m2 + n2 )k/2
(m,n)6=(0,0)

We rearrange the sum by grouping together, for each fixed j = 1, 2, 3, . . . , all pairs (m, n)
with max{|m|, |n|} = j. We note that for each j there are 8j such pairs (m, n), each of
which satisfies
j 2 ≤ m2 + n2 .
Hence
∞ ∞
1 X 8j 8 X 1
|Gk (z)| ≤ k/2 = k/2 ,
c j=1
jk c j=1
j k−1

which is finite and independent of z ∈ Rr,s , so Gk (z) converges absolutely and uniformly
on Rr,s . Since every compact subset of H is contained in some Rr,s , this finishes the
proof by Proposition 1.4.2.

12
Remark 1.4.4. This proof clearly fails for k = 2. One can show that for k = 2, the
series (1.1) is conditionally but not absolutely convergent. We will come back to this
issue later in the course.
Proposition 1.4.5. For every even integer k ≥ 4, the function Gk is a modular form
of weight k and level 1. The q-expansion of Gk is given by
∞
2 · (2πi)k X
Gk (z) = 2 ζ(k) + · σk−1 (n)q n
(k − 1)! n=1
P∞ 1
dk−1 .
P
where ζ(k) = n=1 nk (the Riemann zeta function) and σk−1 (n) = d|n

Proof. One easily checks that Gk (z + 1) = Gk (z). Moreover, we have

 
1 X 1
Gk − = 1
z (m(− z ) + n)k
(m,n)∈Z2 \{0}
X 1
= zk
2
(−m + nz)k
(m,n)∈Z \{0}
k
= z Gk (z).

Hence Gk |k S = Gk and Gk |k T = Gk , so Gk |k α = Gk for all α ∈ SL2 (Z) by Theorem

1.2.2 (a). Thus Gk is a weakly modular function of weight k and level 1.
P at1 ∞. Therefore we will determine the
It remains to show that Gk is holomorphic
q-expansion of Gk . Consider the formula n∈Z z+n = π · cot(πz). Thus we obtain
∞
e2πiz + 1
   
X 1 2 X
= π · cot(πz) = iπ = iπ 1 + = iπ − 2πi qn,
n∈Z
z+n e2πiz − 1 q−1 n=0

∂
where q = e2πiz . Differentiating (k − 1) times with respect to z, and using that ∂z
=
∂
2πiq ∂q , leads to
∞
!
X −(k − 1)! ∂ k−1 X
= k−1 iπ − 2πi qn
n∈Z
(z + n)k ∂z n=0
∞
X
= −2πi (2πin)k−1 q n
n=1
∞
X
k
= −(2πi) nk−1 q n
n=1

(We are using here that k is even; for k odd we get an additional − sign.)
Hence we get
∞
X 1 (2πi)k X k−1 2πinz
tk (z) := = n e .
n∈Z
(z + n)k (k − 1)! n=1

13
Now we can split up the original sum of the function Gk into two parts, one where
m = 0 and one where m 6= 0. Afterwards we will simplify both parts using symmetry
(remember again that k is even) of the sums and the above formula:
X 1 X X 1
Gk (z) = k
+
n n∈Z
(mz + n)k
n∈Z\{0} m∈Z\{0}
∞ ∞ X
X 1 X 1
=2 k
+2
n=1
n m=1 n∈Z
(mz + n)k
∞
X
= 2ζ(k) + 2 tk (mz)
m=1
∞ ∞
(2πi)k X k−1 2πinmz
X
= 2ζ(k) + 2 n e
m=1
(k − 1)! n=1
∞ ∞
2 · (2πi)k X X k−1 nm
= 2ζ(k) + n q
(k − 1)! m=1 n=1

From there we obtain the proposed q-expansion by resorting the last sum:
∞
2 · (2πi)k X X k−1 l
Gk (z) = 2ζ(k) + d q
(k − 1)! l=1
d|l
| {z }
σk−1 (l)

And since Gk has a q-expansion without any negative powers of q, Gk is holomorphic at

∞. Thus Gk is indeed a modular form.
Definition 1.4.6. The Bernoulli numbers are the rational numbers Bk , for k ≥ 0,
defined by the equation
∞
t X Bk
= tk ∈ Q[[t]].
exp(t) − 1 k=0 k!

Remark 1.4.7. The Bernoulli numbers are of great importance in mathematics. Barry
Mazur once said: “When a Bernoulli number sneezes, the tremors can be felt in all of
mathematics.”
Lemma 1.4.8. We have
Bk 6= 0 ⇔ k = 1 or k is even.
Proof. Exercise sheet 2.
Example 1.4.9. The first few non-zero Bernoulli numbers
1 1 1 1
B0 = 0, B1 = − , B2 = , B4 = − , B6 = ,
2 6 3 42
1 5 691
B8 = − , B10 = , B12 = − .
30 66 2730

14
Lemma 1.4.10. If k ≥ 2 is an even integer, then

(2πi)k Bk
ζ(k) = − .
2 · k!
Proof. Exercise sheet 2.

Definition 1.4.11. Let k ≥ 4 be even. The normalised Eisenstein series of weight k

is given by
∞
1 2k X
Ek (z) = Gk (z) = 1 − σk−1 (n)q n .
2ζ(k) Bk n=1

1.5 The valence formula

Definition 1.5.1. Let f 6= 0 be a meromorphic function H → C and let P ∈ H. The
unique integer n such that (z −P )−n f (z) is holomorphic and non-vanishing at P is called
the order of f at P and denoted by vP (f ). We say f has a zero of order n at P if
n is positive, and f has a pole of order n at P if n is negative.

Definition 1.5.2. Consider the Laurent expansion of f around P

X
f (z) = cn (z − P )n .
n≥n0

Then the residue of f at P is ResP (f ) = c−1 ∈ C.

Lemma 1.5.3. If f is meromorphic around a point P , then

ResP (f /f 0 ) = vP (f ).

Proof. Exercise.
We recall without proof the following results from complex analysis:

Theorem 1.5.4. (Cauchy’s integral formula) Let g be a holomorphic function on an

open subset U ⊆ C and let C be a contour in U . Then for each P ∈ U , we have
Z
g(z)
dz = 2πi · g(P ).
C z −P

Corollary 1.5.5. Let C(P, r, α) be an arc of a circle of radius r and angle α around a
point P . If g is holomorphic at P , then
Z
g(z)
lim dz = αi · g(P ).
r→0 C(P,r,α) z − P

(Here, we integrate counterclockwise.)

15
 The following result relates the contour integral of the logarithmic derivative of f to
the orders of f at the interior points:
Theorem 1.5.6. (Argument principle) Let f be a meromorphic function on an open
subset U ⊆ C, and let C be a contour in U not passing through any zeros or poles of f .
Then Z 0
f (z) X
dz = 2πi vP (f ).
C f (z) P ∈int(C)

Note 1.5.7. By Lemma 1.5.3, we have

Z 0
f (z) X
dz = 2πi ResP (f 0 /f ). (1.2)
C f (z) P ∈int(C)

Corollary 1.5.8. Let C(P, r, α) be an arc of a circle of radius r and angle α around a
point P . If f is meromorphic at P , then
f 0 (z)
Z
lim dz = αi · vP (f ).
r→0 C(P,r,α) f (z)

Now assume that f is a weakly modular funktion (of weight k and level 1).
Remark 1.5.9. Since f |k α = f for all α ∈ SL2 (Z), we have vαP (f ) = vp (f ). Hence
vP (f ) is well-defined for P being a SL2 (Z) orbit in H.
Moreover, if f is meromorphic at ∞, we can define the order of f at ∞ by
v∞ (f ) := v0 (f˜).
The following theorem is fundamental for studying the spaces of modular forms:
Theorem 1.5.10. (The valence formula) Let f 6= 0 be a modularfunction of weight k
and level 1. Then f has finitely many SL2 (Z)-orbits of zeros and poles in H, and
1 1 X k
v∞ (f ) + vi (f ) + vρ (f ) + vP (f ) = , (1.3)
2 3 P ∈W
12

where ρ = e2πi/3 and W is the set of all SL2 (Z)-orbits in H except the orbits of i and ρ.
Proof. Recall the fundamental domain from 1.2.2 and let C be the contour as shown in
the figure below. Here =(A) = =(E) = R (we will later let R → +∞) and the three
small circles have radius r. We assume that R is sufficiently large and r sufficiently small
that the interior of C contains all the zeros and poles of f except those at i, ρ, ρ + 1 and
∞.
Simplifying assumption: We assume for simplicity f has no zeros or poles on the
boundary of the fundamental domain, except possibly at i and ρ. (In the case where it
does contain zeros or poles of f , the contour has to be modified using additional small
arcs going around these zeros or poles in the counterclockwise direction.)
R 0 (z)
We will now calculate C ff (z) dz in two different ways and compose the results after-
wards.

16
(1) Computing the integral using Theorem 1.5.6, we get

f 0 (z)
Z X X
dz = 2πi vP (f ) = 2πi vP (f ),
C f (z) P ∈W
P ∈interior(C)

where W is the set described in the stated theorem. The last equality is satisfied
by the simplifying assumption, so the interior of the fundamental domain contains
exactly one representative of every pole or zero SL2 (Z)-orbit of H.

(2) Secondly, we estimate the integral by splitting up the contour in 8 parts. Let C1
be the part from E to A, C2 be the part from A to B, and so on, such that in the
end C8 is the part from D0 to E.

(i) Note that since f is a modular function, we have f (z) = f (z + 1). Hence also
f 0 (z) = f 0 (z + 1), and we have

f 0 (z) f 0 (z + 1) f 0 (z)
Z Z Z
dz = dz = − dz,
C2 f (z) C2 f (z + 1) C8 f (z)

so
f 0 (z) f 0 (z)
Z Z
dz + dz = 0.
C2 f (z) C8 f (z)

(ii) Now we consider C1 and change the variable by q(z) = e2πiz . This maps C1 to
a clockwise oriented circle around the origin with radius e−2πR . Furthermore
we have f (z) = f˜(q(z)), thus f 0 (z) = f˜0 (q(z)) q 0 (z) and since f is a modular

17
 function, f˜ is meromorphic at 0. Therefore
Z ˜0
f 0 (z) f (q(z))q 0 (z)
Z
dz = dz
C1 f (z) C1 f˜(q(z))
f˜0 (q)
Z
= dq
˜
q(C1 ) f (q)
!
f˜0
= −2πi Res0
f˜
= −2πi v0 (f˜)
= −2πi v∞ (f ).

(iii) C5 is half of a circle around i. We deduce from Corollary 1.5.8 that

f 0 (z)
Z
1
lim dz = − 2πi vi (f ).
r→0 C f (z) 2
5

Similarly we get
f 0 (z)
Z
1
lim dz = − 2πi vρ (f )
r→0 f (z) 6
ZC3 0
f (z) 1 1
lim dz = − 2πi vρ+1 (f ) = − 2πi vρ (f ).
r→0 C7 f (z) 6 6

(iv) So it remains to study C4 and C6 . Therefore consider u(z) = − z1 . This maps

C6 to −C4 and we have f (z) = z −k f (u(z)), hence
f 0 (z) = −kz −k−1 f (u(z)) + z −k f 0 (u(z))u0 (z).
So
f 0 (z) −k f 0 (u(z))u0 (z)
Z Z Z
dz = dz + dz
C4 f (z) C4 z C4 f (u(z))
f 0 (u)
Z
2πik
= + du
12 u(C4 ) f (u)
f 0 (u)
Z
2πik
= − du
12 C6 f (u)

and thus
f 0 (z) f 0 (z)
Z Z
k
dz + dz = 2πi .
C4 f (z) C6 f (z) 12
Composing (i) to (iv) yields
Z 0  
f (z) k 1 1
dz = 2πi − vρ (f ) − vi (f ) − v∞ (f ) .
C f (z) 12 3 2

Combining this with the result in (1) gives us exactly the proposed formula.

18
1.6 Applications to modular forms
The valence formula provides some interesting consequences to spaces of modular forms
which we will investigate below.

Definition 1.6.1. Let Mk be the set of all modular forms of weight k and level 1 and
let Sk be the set of all cusp forms of weight k and level 1.

Remark 1.6.2. It can be easily checked that these are both vector spaces over C.

Lemma 1.6.3.

(a) Mk = {0} for k < 0 and k = 2.

(b) Sk = {0} for k < 12.

(c) M0 is the set of all constant functions H → C and thus isomorphic to C.

Proof. (a) Let f ∈ Mk , f 6= 0. Then vz (f ) ≥ 0 for all z ∈ H ∪ {∞}. So by the valence

formula we get k ≥ 0. Moreover a sum of non-negative integer multiples of 21 and
1
3
can’t equal 16 . Thus k 6= 2.

(b) Let f ∈ Sk , f 6= 0. Then v∞ (f ) ≥ 1, hence k ≥ 12 by valence formula.

(c) Let f ∈ M0 . Then the constant function g := f (∞) is also in M0 , so f − g ∈ S0

and therefore f = g since S0 = {0}.

Definition 1.6.4. Define

E43 − E62
∆= .
1728
Remark 1.6.5. In the prologue of this lecture we defined ∆ = q · n∈N (1 − q n )24 . We
Q
will prove later that this is indeed the same ∆ as the one in Definition 1.6.4.

Note 1.6.6. Since E4 and E6 are modular forms of weight 4 and 6, respectively, ∆ is
a modular form of weight 12. Since the q-expansion has zero constant coefficient, it is
indeed a cusp form.

Lemma 1.6.7. The modular form ∆ has a simple zero at ∞ and no other zeros.

Proof. Using the known q-expansions of E4 and E6 , one can compute the q-expansion
of ∆ as

∆ = q − 24q 2 + 252q 3 − 1472q 4 + 4830q 5 − 6048q 6 − 16744q 7 + . . . ,

so ∆ has a simple zero at ∞. Now since ∆ is a modular form, all the quantities v? (∆)
occurring in Theorem 1.5.10 are non-negative, so the only way to get equality is if there
are no zeros apart from the one at ∞.

19
Proposition 1.6.8. S12 is one-dimensional over C and spanned by ∆.
Proof. Let f ∈ S12 and define a function g by
f (i)
g(z) = f (z) − ∆(z).
∆(i)
This function is well-defined since ∆ does not vanish on H, so ∆(i) 6= 0. Clearly g ∈ S12
and g(i) = 0. Using the valence formula yields
1 1 X
v∞ (g) + vi (g) + vρ (g) + vp (g) = 1.
2 3 p∈W

But this is a contradiction since v∞ (g) ≥ 1 and vi (g) ≥ 1. Therefore g has to be zero
and
f (i)
f= ∆ ∈ C · ∆.
∆(i)

Corollary 1.6.9.
1. For all k ∈ Z, the map
Mk → Sk+12 , f 7→ f · ∆
is an isomorphism.
2. For k ≥ 4 we have Mk = Sk ⊕ (C · Ek ).
Proof. The first statement is trivial for k < 0 since then Mk = Sk+12 = {0} by Lemma
1.6.3 (a), (b). So let k ≥ 0. As ∆ is non-vanishing the given map is clearly an injection.
Now let g ∈ Sk+12 . Then ∆g is weakly modular of weight (k + 12) − 12 = k and
holomorphic on H since ∆ is non-vanishing. Furthermore v∞ (g) ≥ 1 by assumption, so
g
v∞ = v∞ (g) − v∞ (∆) = v∞ (g) − 1 ≥ 0.
∆
So ∆g ∈ Mk . Therefore the given map is also onto, thus bijectiv.
For the second part of the corollary we just have to note that Sk is the kernel of the
linear map Mk → C, f 7→ f (∞). Thus we have dim(Mk /Sk ) ≤ 1. On the other hand
we know that Ek ∈ Mk \ SK since Ek (∞) 6= 0. So Mk = Sk ⊕ (C Ek ).
Theorem 1.6.10.
(a) The space Mk is finite dimensional over C for all k ∈ Z.
(b) Let k ≥ 0 even. Then
( k
1 + 12 , k=6 2 mod 12,
dim(Mk ) =  k 
12
, k = 2 mod 12.

Otherwise Mk = {0}.

20
 (c) A basis for Mk is given by {E4a E6b : a, b ∈ N0 , 4a + 6b = k}.
Proof. (a) This is a consequence of part (b).
(b) We will prove this by induction on k. First of all note that the statement is clear
for odd k since there aren’t any nonzero weakly modular functions of odd weight.
Moreover we already know that dim(M0 ) = 1, dim(M2 ) = 0 and dim(Mk ) = 0
for k < 0 by Lemma 1.6.3 (a) and (c). In addition we have dim(Mk ) = 1 for
k = 4, . . . , 10 since dim(Mk ) = dim(Sk ) + 1 by Corollary 1.6.9 and Sk = {0} for
these k’s by Lemma 1.6.3 (b). Hence the statement is true for k = 0, . . . , 10.
Let now k ≥ 12. Then
dim(Mk ) = dim(Mk−12 ) + 1
since dim(Sk ) = dim(Mk−12 ) by Corollary 1.6.9. So the statement is true for all k
by induction in steps of 12.
(c) We will use again induction to prove the statement. Note that there is nothing to
show for odd k, k < 0 and k = 2 since in these cases Mk = {0}. The case k = 0
is also trivial because M0 is the set of all constant functions, hence generated by
1 = E40 E60 .
Let now k ≥ 4 be even. Obviously there is always a pair (a, b) such that a, b ∈ Z≥0
and 4a + 6b = k. Pick such a pair. Let f ∈ Mk . Then f can be written in the form
f = λ E4a E6b + g
for some λ ∈ C and g ∈ Sk since the modular form E4a E6b is in Mk and does not
vanish at infinity. So there is an h ∈ Mk−12 such that g = h · ∆ by corollary 1.6.9
and by induction we may assume h to be a linear combination of E4r E6s where
r, s ∈ Z≥0 and 4r + 6s = k − 12. Hence
 3
E4 − E62

h·∆=h·
1726
is a linear combination of E4r+3 E6s and E4r E6s+2 and since
4(r + 3) + 6s = 4r + 6(s + 2) = k
the function h is a linear combination of E4p E6q with 4p + 6q = k. So the linear
span of these functions contains g and hence also f . Therefore
Mk = span{E4a E6b : a, b ∈ N0 , 4a + 6b = k}.
To show that the given set is indeed a basis of Mk it suffices to check that

 (a, b) ∈ Z2≥0 : 4a + 6b = k  = dim(Mk ).


This can again be easily seen by induction in steps of 12 (exercise).

21
Example 1.6.11. For the first few values of k, the dimensions of Mk and Sk are given
by
k dim Mk dim Sk
0 1 0
2 0 0
4 1 0
6 1 0
8 1 0
10 1 0
12 2 1
14 1 0
16 2 1
Example 1.6.12. Both, E42 and E8 are in M8 . But dim(M8 ) = 1 by Theorem 1.6.10 (b).
Hence E42 and E8 are linearly dependent and as both are 1 at infinity, we can conclude
that E42 and E8 are equal. So
∞
!2 ∞
X X
1 + 240 σ3 (n)q n = E42 = E8 = 1 + 480 σ7 (n)q n
n=1 n=1

, so
n−1
X
σ7 (n) = σ3 (n) + 120 σ3 (m)σ3 (n − m).
m=1

This is very hard to prove (or even conjecture!) without using the theory of modular
forms.
Example 1.6.13. From the theorem, we deduce that

M30 = CE30 ⊕ C∆E18 ⊕ C∆2 E6 .

I claim that another basis for the same space is given by

M30 = CE65 ⊕ C∆E63 ⊕ C∆2 E62 .

Note that these forms are linearly independent (exercise), so since dim(M30 ) = 3, they
form a basis.
The following theorem is a very useful consequence of the fact that the spaces of
modular forms are finite-dimensional:
Theorem 1.6.14. Let f be a modular form of weight k and level 1 with q-expansion
P∞ n
n=0 an q . Suppose that

aj = 0 for all j = 0, . . . , bk/12c.

Then f = 0.

22
Proof. Suppose that f 6= 0. Then the hypothesis implies that
v∞ (f ) ≥ bk/12c + 1 > k/12.
Hence the left-hand side of (1.3) is strictly greater than k/12, which gives a contradiction.

Corollary 1.6.15.
P∞ Letn f, g be
P∞ modular forms of the same weight k and level 1, with
q-expansions n=0 an q and n=0 bn q n , respectively. Suppose that
aj = bj for all j = 0, ..., bk/12c.
Then f = g.
Corollary 1.6.15 is a very powerful tool: it allows us to conclude that two modular
forms are identical if we only know a priori that their q-expansions agree to a certain
finite precision.

1.7 The q-expansion of ∆

The aim of this section is to prove the product formula for the q-expansion of ∆. We
start with the following definition:
Definition 1.7.1. We define
 
X X 1
G2 (z) =  
m∈Z
(mz + n)2
n∈Z,(m,n)6=0

3
and E2 (z) = π2
· G2 (z).
Lemma 1.7.2.
1. The series in Definition 1.7.1 is convergent, but not absolutely convergent, and
defines a holomorphic function on H1 .
2. We have ∞
X
G2 (z) = 2ζ(2) − 8π σ1 (n)q n .
n=1

Proof. 1. Exercise.
2. Argue as in the proof of proposition 1.4.5.
Proposition 1.7.3. The functions G2 and E2 satisfies the transformation property
 
−2 1
z G2 − = G2 (z) − 2πiz, (1.4)
z
 
−2 1 6i
z E2 − = E2 (z) − . (1.5)
z πz
1
It is not a modular form, however: it can’t be, since M2 = {0}.

23
 The proof of this result is based on the following lemma, which gives an example of
two double series that contain the same terms but sum to different values due to the
order of summation being different.
Lemma 1.7.4. For all z ∈ H, we have
XX 1 1

− = 0, (1.6)
m6=0 n∈Z
mz + n mz + n + 1
XX 1 1

2πi
− =− . (1.7)
n∈Z m6=0
mz + n mz + n + 1 z

Proof. We start with the sum

X  1 1

1 1
− = − .
−N ≤n<N
mz + n mz + n + 1 mz − N mz + N

Using this, we compute the inner sum of (1.6) as

X 1 1
 X  1 1

− = lim − (1.8)
n∈Z
mz + n mz + n + 1 N →∞
−N ≤n<N
mz + n mz + n + 1
1 1
= lim − . (1.9)
N →∞ mz − N mz + N
= 0, (1.10)
which implies (1.6).
The proof of the second formula is more complicated, and I will not give the proof
here. For a reference, see Serre’s ”A course in Arithmetic”.
We can now prove Proposition 1.7.3:
Proof. Recall that
XX 1
G2 (z) = 2ζ(2) + .
m6=0 n∈Z
(mz + n)2
Subtracting (1.6) and simplifying, we obtain the alternative expression
XX 1
G2 (z) = 2ζ(2) + 2
. (1.11)
m6=0 n∈Z
(mz + n) (mz + n + 1)

Also, we have
XX 1
z −2 G2 (−1/z) = 2ζ(2)z −2 + (1.12)
m6=0 n∈Z
(nz − m)2
XX 1
= 2ζ(2) + (1.13)
m∈Z n6=0
(nz − m)2
XX 1
= 2ζ(2) + 2
; (1.14)
n∈Z m6=0
(mz + n)

24
note that in the second equality we just relabelled the parameters, but did not change
the order of summation.
Subtracting (1.7) and simplifying, we obtain
2πi XX 1
z −2 G2 (−1/z) + = 2ζ(2) + 2
, (1.15)
z n∈Z m6=0
(mz + n) (mz + n + 1)

and by imitating the proof of Lemma 1.4.3 one can show that the sum on the right-hand
side is absolutely convergent. We can hence change the order of summation, and we see
that (1.15) is equal to (1.11).
Corollary 1.7.5. The q-expansion of ∆ is given by
Y
∆=q (1 − q n )24 .
n≥1

Proof. Let D(z) = q n≥1 (1 − q n )24 .

Q

Let D(z) = q · ∞ n 24
where q = e2πiz as usual. We can check that this
Q
n=1 (1 − q )
product converges sufficiently fast for D to be defined and holomorphic on H. Evidently
D(z + 1) = D(z) and D(z) → 0 as =(z) → ∞. So to check that it is a modular form of
weight 12 (clearly cuspidal), it suffices to show that D(− z1 ) = z 12 D(z). The result then
follow immediately, since we already know that S12 is 1-dimensional.
Recall that ∂d
∂z
∂
= 2πiq ∂q . Then
∞
!
∂ ∂ X
(log(D(z))) = log(q) + 24 log(1 − q n )
∂z ∂z n=1
∞
X −2πinq n
= 2πi + 24
n=1
1 − qn
∞ ∞
!
X X
= 2πi 1 − 24 nq n qr
n=1 r=0
∞ ∞
!
XX
nr
= 2πi 1 − 24 nq
n=1 r=0
∞
!
X
= 2πi 1 − 24 σ1 (n)q n
n=1
= 2πi E2 (z).
Hence finally
    
∂ D(−1/z) 1 1 12
log 12
= 2 2πi E2 − − − 2πi E2 (z)
∂z z D(z) z z z
    
2πi 1 2 6z
= 2 E2 − − z E2 (z) +
z z iπ
= 0.

25
So there is a constant λ sucht that D(− z1 ) = λz 12 D(z) for all z ∈ H. For z = i solves
this to D(i) = D(− 1i ) = λD(i), and since D(i) 6= 0 we have λ = 1, and therefore
D(− z1 ) = z 12 D(z).
We can now expand the product formula for ∆(z) as
X
∆(z) = τ (n)q n for some τ (n) ∈ Z.
n≥1

Conjecture 1.7.6. (Ramanujan, 1916)

1. For m, n coprime, we have τ (mn) = τ (m)τ (n).

2. For p prime and n > 0, we have

τ (pn+1 ) = τ (p)τ (pn ) − p1 1τ (pn−1 ).

11
3. We have |τ (p)| ≤ 2p 2 for all primes p.

We will see a proof of properties 1) and 2) later in the course, in the section on Hecke
operators. Property 3) was proved by Deligne in 1974 as a consequence of his proof of
the Weil conjectures, for which he was awarded the Fields medal in 1978.

26
2 Modular forms of higher level
The idea is to look at functions transforming nicely under subgroups of SL2 (Z).

2.1 Congruence subgroups

Definition 2.1.1. For N ∈ N define the subgroup
      
a b a b 1 0
Γ(N ) = ∈ SL2 (Z) : = mod N .
c d c d 0 1

We will call this group the principal congruence subgroup of level N .

Note 2.1.2. Γ(N ) is the kernel of the group homomorphism induced by the reduction
map Z → Z/N Z:
πN : SL2 (Z) → SL2 (Z/N Z).
It is hence a normal subgroup of finite index. (Ex: show that πN is sujective. This
statement goes by the name of ”strong approximation for SL2 ”. It can be shown to be
false for GL2 (Z).)

Definition 2.1.3. A subgroup Γ of SL2 (Z) is called a congruence subgroup if there

exists N ≥ 1 such that Γ(N ) ⊆ Γ. The least such N is called the level of Γ.

Lemma 2.1.4. Any congruence subgroup has finite index in SL2 (Z).

Proof. It sufficies to show that [SL2 (Z) : Γ(N )] < ∞ for all N ∈ N. But this is clear as
SL2 (Z)/Γ(N ) ,→ SL2 (Z/N Z) and SL2 (Z/N Z) is finite.

Remark 2.1.5. The converse to Lemma 2.1.4 is false. There exist finite index Γ ⊆
SL2 (Z) which don’t contain Γ(N ) for any N . (For example there is one of index 7.) But
every finite index subgroup of SLn (Z) is congruence for n ≥ 3. So SL2 is quite unusual.
(Bass-Serre-Milnor theorem)

Definition 2.1.6. Other standard congruence subgroups of level N are given by

      
a b a b 1 ∗
• Γ1 (N ) = ∈ SL2 (Z) : = mod N ,
c d c d 0 1
      
a b a b ∗ ∗
• Γ0 (N ) = ∈ SL2 (Z) : = mod N .
c d c d 0 ∗

27
Note 2.1.7. We have a chain of inclusions
Γ(N ) ⊆ Γ1 (N ) ⊆ Γ0 (N ) ⊆ SL2 (Z).
These inclusions are in general strict; however, all of them are equalities for N = 1, and
Γ0 (2) = Γ1 (2).
Lemma 2.1.8. For N ≥ 1, we have
Y 1

[Γ1 (N ) : Γ(N )] = N, [Γ0 (M ) : Γ1 (N )] = N 1− ,
p
p|N
Y 1

[SL2 (Z) : Γ0 (M )] = N 1+ .
p
p|N

Definition 2.1.9. Let Γ be a congruence subgroup. We say that Γ is even (resp. odd)
if − Id ∈ Γ (resp. Id 6∈ Γ). We define the projective index of Γ to be
dΓ = [PSL2 (Z) : Γ̄],
where Γ̄ is the image of Γ in PSL2 (Z).

2.2 Fundamental domains and cusps

Proposition 2.2.1. Let Γ be a congruence subgroup of SL2 (Z), and let R be a set of
coset representatives for the quotient Γ\ SL2 (Z). Then the set
[
DΓ = γD
γ∈R

has the property that for any z ∈ H there exists γ ∈ Γ such that γz ∈ DΓ . Furthermore,
γ is unique up to multiplication by an element of Γ ∩ {± Id}, except possibly if γz lies
on the boundary of D. We call DΓ a fundamental domain for Γ.
Proof. If z ∈ H, thern there exists g ∈ SL2 (Z) and z0 ∈ D such that g.z = z0 . The coset
decomposition implies that we can express g uniquely as γ −1 γ 0 with γ ∈ Γ and γ 0 ∈ R.
We now have
γ.z = γg.z0 = γ 0 .z0 ∈ DΓ .
The uniqueness is left as an exercise.
Example 2.2.2. Let Γ = Γ0 (2). A system of representatives for the quotient Γ\ SL2 (Z)
is      
1 0 0 −1 0 −1
, , = {Id, S, ST }.
0 1 1 0 1 1
Using this, one can draw the fundamental domain for Γ.
Note that there are now two points in its closure which do not belong to H: the cusp
∞, as well as 0.

28
Definition 2.2.3. The set P1 (Q), the projective line over Q, consists of Q ∪ {∞}.
We give this an action of SL2 (Z) via
 
a b ax + b
.x =
c d cx + d
a
where the right-hand-side is interpreted as c
if x = ∞, and as ∞ if cx + d = 0.

Proposition 2.2.4. SL2 (Z) acts transitively on P1 (Q).

Proof. Clearly it sufficies to show that for any x ∈ P1 (Q) we can map ∞ to x. For
x = ∞ we have ∞.1 = ∞. So let
x = ac with a, c ∈ Z coprime.

Then there are r, s ∈ Z
a −s a −s
such that ar + cs = 1, thus c r ∈ SL2 (Z) and c r .∞ = x.

Note 2.2.5. An easy computation shows that the stabiliser of ∞ in SL2 (Z) is the
subgroup    
1 b
SL2 (Z)∞ = ± : b∈Z .
0 1
It follows from Proposition 2.2.4 that we hence have a bijection

SL2 (Z)/ SL2 (Z)∞ → P1 (Q),

γ SL2 (Z)∞ 7→ γ∞.

Definition 2.2.6. For Γ ≤ SL2 (Z) of finite index we define the set of cusps of Γ,
denoted by Cusps(Γ), as the set of Γ-orbits in P1Q .

Example 2.2.7. Let p be prime. Then Cusps(Γ0 (p)) = {[∞], [0]}.

u

v ∈ Q with u, vu ∈
Proof. Let
coprime.
Z
u
Then there are r, s ∈ Z such that ur + vs = 1,
so v r ∈ SL2 (Z) and v −s
u −s
r .∞ = v
. We will distinguish two cases:

29
 ∈ Γ0 (p), so uv ∈ [∞]. Conversly, if γ ∈ Γ0 (p) then p


(1) If p divides v then uv −sr
divides the denominator of γ.∞ by definition. So the orbit of ∞ is given by all
fractions uv with p dividing the denominator v.

(2) If v is not divisible by p we can note that

u(r + λv) + v(s − λu) = 1

0
and since p is not
s0 u

a divisor of v 0we find λ ∈ Z such
s0 u

that ru = r + λv ∈ pZ.
Therefore −r0 v ∈ Γ0 (p) where s = s − λu and −r0 v .0 = v by definition. So
u a b


v
∈ [0]. Conversly, if c d ∈ Γ0 (p) then p does not divide d since ad − bc = 1.
Thus p cannot divide the denominator of γ.0. Therefore the orbit of 0 is given by
all fractions uv with p not dividing the denominator v.

So this is everything and there are exactly two distinct orbits as claimed.

Note 2.2.8. By Note 2.2.5, we see that

Cusps(γ) = Γ\ SL2 (Z)/ SL2 (Z)∞ .

In particular, we have a sujective map

SL2 (Z)/ SL2 (Z)∞  Cusps(Γ).

Definition 2.2.9. If P = [t] ∈ Cusps(Γ), denote by Γt the stabilizer for t in Γ.

Lemma 2.2.10. Choose γt ∈ SL2 (Z) such that γt (∞) = t. Then

Γt = Γ ∩ γt SL2 (Z)∞ γt−1 .

Proof. Let h ∈ Γ. Then

h ∈ Γt ⇔ h.t = t
⇔ hγt (∞) = γt (∞)
⇔ γt−1 hγt (∞) = ∞
⇔ γt−1 hγt ∈ SL2 (Z)∞
⇔ h ∈ γt SL2 (Z)∞ γP t−1.

Note 2.2.11. It follows from the proof that we have an injection

Γt \(γt−1 SL2 (Z)∞ γt ) ,→ Γ\ SL2 (Z),

so Γt has finite index in γt−1 SL2 (Z)∞ γt .

30
Lemma 2.2.12. The subgroup

HP = γt−1 Γγt ∩ SL2 (Z)∞ ⊆ SL2 (Z)

does not depend on the choice of representative for P , and it has finite index in SL2 (Z)∞ .

Proof. We first show that if we have elements γt and γ̃t in SL2 (Z) such that γt .∞ = t
and γ̃t .∞ = t, then

γt−1 Γγt ∩ SL2 (Z)∞ = γ̃t−1 Γγ̃t ∩ SL2 (Z)∞ .

Note that γt−1 γ̃t fixes ∞, so it is an element in SL2 (Z)∞ , say γt−1 γ̃t = g ∈ SL2 (Z)∞ .
Then

γ̃t−1 Γγ̃t ∩ SL2 (Z)∞ = g −1 γt−1 Γγt g ∩ SL2 (Z)∞

= g −1 γt−1 Γγt ∩ g SL2 (Z)∞ g −1 g

= γt−1 Γγt ∩ SL2 (Z)∞ .

Here, we get the last equality since γt−1 Γγt ∩ g SL2 (Z)∞ g −1 ⊆ SL2 (Z)∞ and hence is
commutative, so in particular its elements commute with g.
Suppose now that we choose another element t in the Γ-orbit of t, and let γt0 ∈ SL2 (Z)
such that γt0 .∞ = t0 . Then we can write γt0 = gγt for some g ∈ Γ which satisfies g.t = t0 .
Then
γt−1 −1 −1
0 Γγt0 = γt g Γgγt = γt−1 Γγt−1 ,
and hence
γt−1 −1
0 Γγt ∩ SL2 (Z)∞ = γt Γγt0 ∩ SL2 (Z)∞ .

Lemma 2.2.13. Let H be a subgroup of finite index in SL2 (Z)∞ , and let h be the index
of ±H in SL2 (Z)∞ . Then H is one of the following:

1 h

h( 0 1 )i


−1 h
H= 0 −1 = {(−1)t ( 10 th
1 ) : t ∈ Z}

 1 h
{± Id} × h( 0 1 )i

Proof. Exercise.

Definition 2.2.14. For H = HP , the integer hΓ (P ) = h in Lemma 2.2.13 is called the

width of the cusp P for Γ. The cusp P is

• irregular if HP is of the form −1

h


0 −1 (then Γ is necessarily odd),

• regular if HP is of the form h( 10 h1 )i (so Γ is odd), of if HP is of the form {± Id} ×

h( 10 h1 )i (so Γ is even).

31
Remark 2.2.15. If Γ is normal in SL2 (Z), the subgroup HP does not depend on the
cusp P , and hence all the cusps have the same width and regularity.

Example 2.2.16. Let us determine the width of the two cusps in Cusps(Γ0 (p)).

• c = [∞]: we need to determine the smallest h ≥ 1 such that ( 10 h1 ) or −1 h

0 −1 are
in Γ0 (p). Hence hΓ0 (p) (∞) = 1, since ( 10 11 ) ∈ Γ0 (p).

• c = [0]: note that g.∞ = 0 for g = 01 −1

0 . Moreover
   
a b −1 d −c
g g = ,
c d −b a

∈ g −1 Γ0 (p)g if and only if b = 0 mod p. In particular,

a b


so c d
 
−1

1 pZ
(Γ0 (p))[0] = g Γ0 (p)g ∩ P∞ =± .
0 1

So the width of the cusp 0 is p.

We now want to count the number of cusps for a given congruence subgroup. We need
the following group-theoretic result:

Proposition 2.2.17. Let G be a group acting transitively on a set X, and let H be a

subgroup of finite index in G.

(i) For any x ∈ X, StabH (x) has finite index in StabG (x), and we have an injection

StabH (x)\ StabG (x) ,→ H\G

with image H\H StabG (x).

(ii) Let x0 ∈ X. Then there is a surjective map

H\G  H\X,
Hg 7→ Hg.x0

and for each x ∈ X, the cardinality of the fibre of this map over Hx equals the
index [StabG (x) : StabH (x)].

(iii) If R is a set of orbit representatives for the quotient H\X, we have

X
[StabG (x) : StabH (x)] = [G : H].
x∈R

Proof. (i) is standard.

32
 For (ii), the transitivity of the G-action on X implies that for all x ∈ X, we can choose
an element gx ∈ G such that gx .x0 = x, so the map H\G → H\X is surjective. Denote
by THx the fibre of this map over Hx, i.e.

THx = {Hg ∈ H\G | Hg.x0 = H.x} .

Writing g as g 0 gx , we obtain a bijection

THx ∼
= {Hg 0 ∈ H\G | Hg 0 gx .x0 = H.x}
= {Hg 0 ∈ H\G | Hg 0 .x = Hx}
= H\(H StabG (x))
∼
= StabH (x)\ StabG (x),

where the last equality follows from (i).

(iii) Summing over R and using (ii), we obtain
X X
[G : H] = |H\G| = |THx | = [StabG (x) : StabH (x)],
x∈R r∈R

which finishes the proof.

Corollary 2.2.18. Let Γ be a congruence subgroup. Then

X
hΓ (P ) = dΓ .
P ∈Cusps(Γ)

Proof. Apply Proposition 2.2.17 to G = PSL2 (Z), H = Γ̄ and X = P1 (Q).

2.3 Weakly modular forms for congruence subgroups

Definition 2.3.1. Let Γ ≤ SL2 (Z) be a congruence subgroup, and let k ∈ Z. A
function f : H → C is a weakly modular function of weight k and level Γ if f is
meromorphic on H and f |k γ = f for all γ ∈ Γ.

Remark 2.3.2. Let k be odd and Γ be even. Let f be a weakly modular function
of
weight k and level Γ. By Lemmas 2.2.12 and 2.2.13 there is h ∈ N such that ± 10 h1 ∈ Γ,
so
f = f |k −1 −h



f = f |k 10 h1 = f (· + h) and 0 −1 = −f (· + h).

Hence f (z) = −f (z) for all z ∈ H and therefore f = 0.

Example 2.3.3. Let f be weakly modular of level SL2 (Z) and weight k. Then f (N z)
is weakly modular of level Γ0 (N ) and weight k.

33
Proof. We have
     
az + b aN z + bN aN z + bN
f N =f =f c .
cz + d cz + d N
Nz + d
a Nb
a b



If c d ∈ Γ0 (N ) then c/N d ∈ SL2 (Z) and hence
 
aN z + bN  c  k
f c = (N z) + d f (N z) = (cz + d)k f (N z)
N
Nz + d N

as required. So z 7→ f (N z) is weakly modular of level Γ0 (N ).

2.4 q-expansion at ∞
Proposition 2.4.1. Let f : H → C be weakly modular of weight k and level Γ and let
h = hΓ (∞).

• If k is even or if k is odd, Γ is odd and ∞ is a regular cusp, then there is a

meromorphic function f˜ on the punctured disc D∗ such that f (z) = f˜(qh (z)) for
all z ∈ B where qh (z) = e2πiz/h .

• If k is odd, Γ is odd and ∞ is irregular, then there is a meromorphic function F̃

on D∗ such that f (z) = eπiz/h F̃ (qh (z)) for all z ∈ H where qh (z) = e2πiz/h .


Proof. By Lemma 2.2.13, at least one of ± 10 h1 lies in Γ, so

(z) = (±1)k f (z + h)



f (z) = f |k ± 1 h
0 1

for all z ∈ H.
If
k is even then (±1)k = 1, so f = f (· + h), and if Γ is odd and ∞ is regular, then
1 h ∈ Γ, so we also have f = f (· + h). In both cases we can argue as in section 1.3.
0 1

If k is odd and Γ is odd but ∞ is irregular, then − 10 h1 ∈ Γ and therefore

f (z) = −f (z + h) ∀z ∈ H.

Define a function F on H by F (z) = f (z)e−πiz/h . Then

F (z + h) = e−πi f (z + h)e−πiz/h = f (z)e−πiz/h = F (z).

So we can argue for F as before and get f (z) = eπiz/h F̃ (qh (z)).

Remark 2.4.2. We can hence write f (z) as a q-expansion at ∞:

(P
a∞,n qhn if k is even or if k is odd and Γ is odd and regular at ∞
f (z) = Pn∈Z n
n∈ 1 +Z a∞,n qh if k is odd and Γ is odd and irregular at ∞
2

34
Definition 2.4.3. Let f : H → C be weakly modular of weight k and level Γ. We say
that f is meromorphic at ∞ if f˜ is meromorphic at 0. Similarly we define f to be
holomorphic at ∞ if f˜ is holomorphic at 0. If f is meromorphic at ∞, we define
1
v∞,Γ (f ) = min{n ∈ Z : a∞,n 6= 0.}
2
We then say f is vanishing at ∞ if v∞,Γ (f ) > 0. If f is holomorphic at ∞ we define
(
f˜(0) if k is even or if k is odd, Γ is odd and ∞ is regular
f (∞) =
0, if k is odd and Γ is odd and irregular at ∞.

1
Remark 2.4.4. To motivate the definition v∞,Γ (f ) = v0 (F̃ ) + 2
in the irregular case
note that the additional 12 term ensures

v∞,Γ (f g) = v∞,Γ (f ) + v∞,Γ (g)

since this would fail for f , g with f (z) = eπiz/h f˜(qh ) and g(z) = eπiz/h g̃(qh ) without this
extra term. Moreover, note that in the irregular case f being holomorphic at ∞ implies
f vanishes at ∞.

2.5 q-expansion at a cusp

To define the q-expansion at a general cusp, we need the following result:

Lemma 2.5.1. Let f : H → C be weakly modular of weight k and level Γ and let g ∈
SL2 (Z)∞ but not necessarily in H∞ . Then f |k g is meromorphic at ∞ if and only if f is.
Moreover v∞,g−1 Γg (f |k g) = v∞,Γ (f ) and (f |k g)(∞) = f (∞) if defined and if k is even.

Proof. We check that f |k g is indeed weakly modular of weight k and level g −1 Γg since

(f |k g) |k g −1 γg = (f |k γ) |k g = f |k g.

Moreover we have
h i h i
hg−1 Γg (∞) = SL2 (Z)∞ : g −1 H∞ g = SL2 (Z)∞ : H ∞

1 t

and g ∈ SL2 (Z)∞ .
since SL2 (Z)∞ is abelian
Now let g = ± 0 1 . Then
(
(±1)k f˜ e2πit/h q ,


if k is even or if k is odd, Γ is odd and ∞ is regular,
(f |k g)(z) = k it/h 2πit/h

.
(±1) e F̃ e q , if k is odd and Γ is odd and irregular at ∞.

So f |k g is meromorphic or holomorphic at ∞ if and only if so is f , and the orders of

vanishing are equal.

35
Definition 2.5.2. Let f be weakly modular of weight k and level Γ. Let P ∈ Cusps(Γ)
be represented by an element t ∈ P1 (Q) and choose some γt ∈ SL2 (Z) such that γt .∞ = t.
Define vP,Γ (f ) = v∞,γt−1 Γγt (f |k γt ).

The following proposition shows that vP,Γ (f ) is well-defined.

Proposition 2.5.3. vP,Γ (P ) is well-defined.
Proof. Suppose that γt0 ∈ SL2 (Z) also satisfies γt0 .∞ = t. then γt−1 γt0 ∈ SL2 (Z)∞ , so by
Lemma 2.5.1 applied to F |k γt we deduce that (f |kγt )|k γt−1 γt0 = f |k γt0 is meromorphic at
∞ if and only if so f |kγt , with the same order of vanishing.
Now let s be another representative of P , and let γs ∈ SL2 (Z) such that γs .∞ = s.
Then there exists g ∈ Γ such that g.s = t, so g.γs .∞ = t, so fk γt is meromorphic at ∞
if and only if so is f |k (gγs ) = fk γs , with the same order of vanishing.
Note 2.5.4. Note that we can define f (P ) = (f |k g)(∞) if f is holomorphic at P and if
k is even, but if k is odd, then f (P ) is only defined up to change of sign.
Definition 2.5.5. We say that f is holomorphic at P if vP,Γ (f ) ≥ 0 and that f is
vanishing at P if vP,Γ (f ) > 0.
Definition 2.5.6. We say f is a modular function if f is meromorphic at every cusp,
f is a modular form if f is holomorphic on H and at every cusp, and f is a cusp
form if f is holomorphic on H and vanishes at every cusp.
Definition 2.5.7. Define Mk (Γ) to be the space of modular forms of level Γ and Sk (Γ)
to be the space of cusp forms of level Γ.
Clearly they are both complex vector spaces.

2.6 The valence formula in arbitrary levels

Definition 2.6.1. For z ∈ H and Γ ≤ SL2 (Z) of finite index we let

nΓ (z) = |stabΓ (z)| .

If nΓ (z) > 1, we say z is an elliptic point of Γ.

Note 2.6.2. Clearly nΓ (z) is 1, 2 or 3, and it is 1 unless z ∈ SL2 (Z)-orbit of i or ρ.
There exist only finitely many Γ-orbits of elliptic points for any Γ, often even none at
all, for example for Γ1 (N ) if N ≥ 4.
Theorem 2.6.3 (The valence formula). If f is a modular function of weight k and level
Γ and f 6= 0 then
X vz (f ) X k dΓ
+ vP,Γ (f ) = .
nΓ (z) 12
z∈Γ\H P ∈Cusps(Γ)

Here, dΓ is the projective index as defined in Definition 2.1.9.

36
 The proof of this will take us a while.
Definition 2.6.4. Let VΓ (f ) = z∈Γ\H nvzΓ(f )
P P
(z)
+ P ∈Cusps(Γ) vP,Γ (f ).

Lemma 2.6.5. Let f be a modular function of level Γ, f 6= 0, and let Γ0 ≤ Γ be another

finite index subgroup of SL2 (Z). Then
d Γ0
VΓ0 (f ) = · VΓ (f ).
dΓ
Proof. Let z ∈ H. We apply Proposition 2.2.17 with X being the Γ-orbit of z, G = Γ
and H = Γ0 . This yields
X nΓ (w) X | stab (w)|
Γ
=
nΓ0 (w) | stabΓ0 (w)|
w∈Γ0 \H w∈H\X
[w]=[z] mod Γ
X   h 0
i d 0
Γ
= stabΓ (w) : stabΓ0 (w) = Γ : Γ = ,
dΓ
w∈H\X

and since nΓ (w) = nΓ (z) for all w ∈ Rz , we have

X 1 1 d Γ0
= .
w∈Rz
nΓ0 (w) nΓ (z) dΓ

Hence we have
!
X vw (f ) X X 1 dΓ0 X vz (f )
= vz (f ) = .
nΓ0 (w) nΓ0 (w) dΓ nΓ (z)
w∈H\X z∈Γ\H w∈Γ0 \H z∈Γ\H
[w]=[z] mod Γ

Similarily we can argue at the cusps: If P ∈ Cusps(Γ) and Q ∈ Cusps(Γ0 ) which maps
to P under the natural map Cusps(Γ0 ) → Cusps(Γ), then we have by definition
hΓ0 (Q)
vQ,Γ0 (f ) = vP,Γ (f ).
hΓ (P )
Therefore we get again by Proposition 2.2.17
X X hΓ0 (Q) d Γ0
vQ,Γ0 (f ) = vP,Γ (f ) = vP,Γ (f ) .
hΓ (P ) dΓ
Q∈Cusps(Γ0 ) Q∈Cusps(Γ0 )
Q=P in Cusps(Γ) Q=P in Cusps(Γ)

and thus
X X X dΓ0 X
vQ,Γ0 (f ) = vQ,Γ0 (f ) = vP,Γ (f ).
dΓ
Q∈Cusps(Γ0 ) P ∈Cusps(Γ) Q∈Cusps(Γ0 ) P ∈Cusps(Γ)
Q=P in Cusps(Γ)

This finishes the proof.

37
Lemma 2.6.6. For any g ∈ SL2 (Z) we have

Vg−1 Γg (f |k g) = VΓ (f ).

Proof. We clearly have vz (f |k g) = vgz (f ) for any z ∈ H and ng−1 Γg (z) = nΓ (gz) since
stabΓ (gz) = g(stabg−1 Γg (z))g −1 . Hence
X vz (f |k g) X vgz (f )
= .
ng−1 Γg (z) nΓ (gz)
z∈(g −1 Γg)\H gz∈Γ\H

This deals with the non-cusp terms in the valence formular. But similarily we can check
that vP (f |k g) = vgP (f ) for all P ∈ Cusps(Γ), so the cusp terms in Vg−1 Γg (f |k g) and
VΓ (f ) are also equal.
Now we can finally proof the valence formula.
Proof of theorem 2.6.3. Let Γ0 be any finite index subgroup of SL2 (Z) which is normal
and contained in Γ. (Note that such a group exists since Γ is a congruence subgroup.)
Then
dΓ
VΓ (f ) = · VΓ0 (f )
dΓ0
by Lemma 2.6.5. Let d = dΓ0 and choose g1 , . . . , gd ∈ SL2 (Z) such that g1 , . . . , gd are
coset representatives for PSL2 (Z)/Γ0 . Define
d
Y
F (z) = (f |k gi )(z).
i=1

Then F is weakly modular of weight dk for the full modular group SL2 (Z), and mero-
morphic at ∞. Hence by Theorem 1.5.10, we have
dk k
VSL2 (Z) (F ) = ⇒ VΓ0 (F ) = d2
12 12
since VΓ0 (F ) = d VSL2 (Z) (F ) by Lemma 2.6.5 But we can easily check that
d
X d
X
VΓ0 (F ) = VΓ0 (f |k gi ) = Vgi−1 Γ0 gi (f |k gi ) = dVΓ0 (f )
i=1 i=1

where we obtain the last two equalities since Γ0 is normal and applying Lemma 2.6.6.
Hence
dk kdΓ
VΓ0 (f ) = ⇒ VΓ (f ) = ,
12 12
which finishes the proof.
Corollary 2.6.7. Mk (Γ) is empty for any k < 0 and for any Γ.
Proof. Clear since the left hand side of the valence formula must be non-negative.

38
Corollary 2.6.8 (”The unreasonable effectiveness of modular forms in number theory”).
Let k ∈ Z and suppose f and g are modular forms of weight k and level Γ, and their
q-expansions agree up to degree k12dΓ , so up to and including qhm where m = b k12dΓ c and
h = h∞ (Γ). Then f = g.
Proof. We have v∞,Γ (f − g) ≥ 1 + b k12dΓ c > k dΓ
12
, which yields a contradiction to Theorem
2.6.3 unless f − g = 0.
Corollary 2.6.9. For any k ≥ 0 and any finite index subgroup Γ ≤ SL2 (Z) we have
 
k dΓ
dim (Mk (Γ)) ≤ 1 + .
12
In particular Mk (Γ) is finite dimensional.
Proof. Let m = b k12dΓ c and h = h∞ (Γ). Consider the linear map Mk (Γ) → Cm+1 mapping
f to the coefficients up to qhm in its q-expansion. By Corollary 2.6.8 this map is injective,
hence dim(Mk (Γ)) ≤ m + 1.
Remark 2.6.10.
(i) It can be shown that if −1 ∈ Γ and k is any non-negative even integer or if Γ is
odd and k is any non-negative integer then
k
dim(Mk (Γ)) ≥ ( − 1)dΓ .
12

(ii) In Diamond & Shurman there are precise formulae for the dimsion of Mk (Γ).

2.7 Eisenstein series revisited

Recall that SL2 (Z)∞ = ± 1 Z
0 1 , and let SL2 (Z)+
∞ =
1 Z
0 1 ⊆ SL2 (Z)∞ .
0 0
Proposition 2.7.1. (a) Let g, g 0 ∈ SL2 (Z), g = ac d and g 0 = ac0 db 0 . Then c = c0
b

and d = d0 if and only if there is an g∞ ∈ SL2 (Z)+ 0

∞ such that g = g∞ g.

(b) For (c, d) ∈ Z2 there exists γ ∈ SL2 (Z) with bottom row (c, d) if and only if
gcd(c, d) = 1.
Proof. For (a) note that
 0 0    0
a d − b0 c −a0 b + b0 a 1 ab0 − a0 b
  
0 −1 a b d −b
gg = = = ∈ SL2 (Z)+
∞.
c d −c a 0 −cb + da 0 1

Part (b) is clear since gcd(c, d) divides det(γ).

Corollary 2.7.2. The mapping ac db 7→ (c, d) gives a bijection

SL2 (Z)+ 2
∞ \ SL2 (Z) → {(c, d) ∈ Z : gcd(c, d) = 1}.

39
 We will now motivate the definition of a generalised Eisenstein series using this bijec-
tion.

Note 2.7.3. Observe that 1|k ac db = (cz + d)−k , so 1 is SL2 (Z)+

∞ -invariant. Hence the
unnormalised level 1 Eisenstenstein series Gk (z) can be written as
∞
!
X 1 X X 1
=
2
(cz + d)k r=1 2
(cz + d)k
(c,d)∈Z \{0} (c,d)∈Z
gcd(c,d)=r
∞
!
X 1 X 1
=
r=1
rk (cz + d)k
(c,d)∈Z2
gcd(c,d)=1
∞
! !
X 1 X
= j(γ, z)−k
r=1
rk
[γ]∈SL2 (Z)+
∞ \ SL2 (Z)
X
= ζ(k) j(γ, z)−k .
[γ]∈SL2 (Z)+
∞ \ SL2 (Z)

Definition 2.7.4. Let Γ be a congruence subgroup of SL2 (Z), and let Γ+ +

∞ = Γ∩SL2 (Z)∞ .
For k ≥ 3, define X
Gk,Γ,∞ = j(γ, z)−k .
γ∈Γ+
∞ \Γ

Proposition 2.7.5. The function Gk,Γ,∞ is a weakly modular function of weight k and
level Γ.

Proof. It can be shown that the sum defining Gk,Γ,∞ converges absolutely and uniformly
on compact subsets of H. Thus Gk,Γ,∞ is well-defined and holomorphic. Moreover,
Gk,Γ,∞ is also clearly Γ-invariant under the weight k action.

Proposition 2.7.6. If either k is even or if k is odd and Γ is regular at ∞, then Gk,Γ,∞

is a modular form of weight k and level Γ, which does not vanish at ∞, but at all other
cusps. Conversly, if k is odd and Γ is irregular at ∞, then Gk,Γ,∞ = 0.

Proof. First suppose that k is odd and Γ is odd and irregular at ∞, so g = −1 n

0 −1 ∈ Γ
for some n ∈ Z. Then g ∈ / Γ+
∞ and

j(γ, z)−k + j(gγ, z)−k = (cz + d)k + (−1)k (cz + d)k = 0

for all γ ∈ Γ. Hence the terms in the sum defining Gk,Γ,∞ cancel out, so Gk,Γ,∞ = 0.
Now let k be even or let k be odd and Γ regular at ∞. We compute Gk,Γ,∞ (∞). We
have (
d−k if c = 0
lim (cz + d)−k = (2.1)
=(z)→∞ 0 if c 6= 0

40
Note also that for γ = ( ac db ), we have c = 0 if and only if γ ∈ Γ∞ . Thus

Gk,Γ,∞ (∞) = lim Gk,Γ,∞ (z)

=(z)→∞
X
= lim j(γ, z)−k
=(z)→∞
γ∈Γ+
∞ \Γ∞

which takes the following values:

Γ odd Γ odd
Γ even ∞ regular ∞ irregular
k even 2 1 2
k odd 0 1 0
Γ+
∞ \Γ∞ ± Id Id (Id, ( −1 h1 ))
Now let P be a cusp different from ∞. Let t be a representative of P , and choose
γt ∈ SL2 (Z) such that γt .∞ = t.
Then by definition, we have

Gk,Γ,∞ (P ) = (Gk,Γ,∞ |k γt ) (∞).

But X X
(Gk,Γ,∞ |k γt ) (z) = j(γγt , z)−k = j(γ, z)−k .
γ∈Γ+
∞ \Γ γ∈Γ+
∞ \Γγt

Claim: any g = ( ac db ) ∈ Γγt has c 6= 0.

Proof of claim: if g = γγt had c = 0, then g ∈ SL2 (Z)∞ , so γ ∈ SL2 (Z)∞ γt−1 ∩ Γ. But
any element in SL2 (Z)∞ γt−1 maps t to ∞, which gives a contradicton since P 6= ∞, i.e.
t does not lie in the Γ-orbit of ∞. We therefore deduce from (2.1) that

Gk,Γ,∞ (P ) = (Gk,Γ,∞ |k γt ) (∞) = 0.

In particular Gk,Γ,∞ |k g is bounded as =(z) → ∞ for all g ∈ SL2 (Z), so Gk,Γ,∞ is indeed
a modular form.

Note 2.7.7. We have constructed a modular form that doesn’t vanish at ∞ for all pairs
(k, Γ) where this isn’t trivially impossible.

Corollary 2.7.8. Let Γ be a congruence subgroup, let P ∈ Cusps(Γ), and let k ≥ 3. If

k is odd, assume that P is regular and that Γ is odd. Then there is a modular form in
Mk (Γ) does not vanish at P but at all other cusps.

Proof. Let t be a representative of P , and choose γt ∈ SL2 (Z) such that γt .∞ = t. Define

Gk,Γ,P = Gk,g−1 Γg,∞ |k g −1 .

Then Gk,Γ,P is a modular form of weight k and level Γ which does not vanish at P but
at all other cusps, by Proposition 2.7.6.

41
Note 2.7.9. The Eisenstein series Gk,Γ,P is well-defined if k is even, and in this case
independent of the choice of t. But if k is odd Gk,Γ,P is only well-defined up to sign.
Definition 2.7.10. We define Ek (Γ) as the subspace of Mk (Γ) spanned by the Gk,Γ,P ’s.
Note 2.7.11. We have
(
| Cusps(Γ)|, if k is even
dim(Ek (Γ)) = .
| Cuspsreg (Γ)|, if k is odd and Γ is odd

Example 2.7.12. Let p be prime and Γ = Γ0 (p). Then Cusps(Γ) = {0, ∞}, both cusps
are regular (see Example 2.2.16), and Γ is even. So the case k odd is trivial. For k ≥ 4
an even integer there are two Eisenstein series: Gk,Γ,∞ and Gk,Γ,0 .

• Gk,Γ,∞ : by the definition of Γ0 (p) and Proposition 2.7.1 we have

X 1
Gk,Γ,∞ = .
(cz + d)k
(c,d)∈Z2
gcd(c,d)=1
p|c

• Gk,Γ,0 : note that we have S.∞ = 0 for S = 01 −1

0 and
  
−1 a b
S ΓS = : p didivdes b =: Γ0 (p).
c d
Now clearly
Γ0 (p)+
 1 p?


∞ = 0 1 ,
and Γ0 (p)+ 0
∞ \Γ (p) can be identified with the set

(c, d) ∈ Z2 − {0} : gcd(c, d) = 1, p - d



Hence
  
0 −1
Gk,Γ,0 (z) = Gk,Γ0 (p),∞ |k (z)
1 0
X 1
= z −k −1
2
(−cz + d)k
(c,d)∈Z
gcd(c,d)=1
p-d
X 1
=
(−c + dz)k
(c,d)∈Z2
gcd(c,d)=1
p-d
X 1
= .
(cz + d)k
(c,d)∈Z2
gcd(c,d)=1
p-c

42
Thus we have

Gk,Γ,∞ (z) + Gk,Γ,0 (z) = Gk,SL2 (Z),∞ (z) = 2Ek (z).

Finally consider
X 1
2Ek (pz) = .
(cpz + d)k
(c,d)∈Z2
gcd(c,d)=1

Note that if (c, d) ∈ Z2 with gcd(c, d) = 1, then gcd(pc, d) = 1 unless p divides d.

So X 1 X 1
2Ek (pz) = k
+ .
2
(cz + d) 2
(pcz + d)k
(c,d)∈Z (c,d)∈Z
gcd(c,d)=1 gcd(c,d)=1
p|c p|d

We can check that

{(pc, d) : gcd(c, d) = 1, p|d} = {(pc, pd) : gcd(c, d) = 1, p - c},

which gives us
X 1
2Ek (pz) = Gk,Γ,∞ + = Gk,Γ,∞ + p−k Gk,Γ,0 .
(pcz + pd)k
(c,d)∈Z2
gcd(c,d)=1
p-c

Hence Ek (Γ) is spanned by Ek (z) and Ek (pz). Note that we have also shown that
Ek (pz) is p−k at cusp 0.

43
3 Hecke operators
3.1 Double cost operators
It turns out that the space MK (Γ) has a very interesting structure: it is a module over
a commutative ring, classed the Hecke algebra.
Lemma 3.1.1.
1. If Γ is a congruence subgroup and α ∈ GL2 (Q)+ , then SL2 (Z) ∩ α−1 Γα is also a
congruence subgroup.
2. Any two congruence subgroups are commensurable: we have
[Γ1 : Γ1 ∩ Γ2 ] < ∞ and [Γ2 : Γ1 ∩ Γ2 ].

Proof. 1. Let N ≥ 1 such that Γ(N ) ⊆ Γ, and such that N α ∈ M2 (Z) and N α−1 ∈
M2 (Z). Then one can check that
αΓ(N 3 )α−1 ⊆ Γ(N ) ⊆ Γ,
so Γ(N 3 ) ⊆ α−1 Γα.
2. Note that there is some M ≥ 1 such that Γ(M ) ⊆ Γ1 ∩ Γ2 .
Definition 3.1.2. Let GL+ 2 (Q) denote the set of invertible 2 × 2 matrices over Q with
positive determinant. Let Γ1 , Γ2 be congruence subgroups, and let α ∈ GL+ 2 (Q). The
double coset Γ1 αΓ2 is the set
Γ1 αΓ2 = {γ1 αγ2 | γ1 ∈ Γ1 , γ2 ∈ Γ2 }.
Note 3.1.3. Multiplication gives a left (resp. right) action by Γ1 (resp. by Γ2 ) on
Γ1 αΓ2 . We can hence decompose the double coset into Γ1 -orbits:
[
Γ1 αΓ2 = Γ1 βj .
j

We will see in a moment that this decomposition is finite.

Proposition 3.1.4. Let Γ1 , Γ2 be congruence subgroups, and let α ∈ GL2 (Q)+ . Let
Γ3 = (α−1 Γ1 α) ∩ Γ2 .
Then the map γ2 7→ Γ1 αγ2 induces a bijection
Γ3 \Γ2 ∼
= Γ1 \Γ1 αΓ2 .

44
Proof. Consider the map

Γ2 → Γ2 \(Γ1 αΓ2 ), γ2 7→ Γ1 αγ2 .

The map is clearly surjective, and two elements γ2 , γ20 get mapped to the same element
if and only if
Γ1 αγ2 = Γ1 αγ20 ⇔ γ20 γ2−1 ∈ α−1 Γ1 α ∩ Γ2 .

Note 3.1.5. By Lemma 3.1.1 (2), we have [Γ2 : Γ3 ] < ∞.

S
Corollary 3.1.6. Let Γ2 = Γ3 γj be a coset decomposition of Γ3 \Γ2 . Then
[
Γ1 αΓ2 = Γ1 αγj

is an orbit decomposition (so Γ1 αγi ∩ Γ1 αγj = ∅ if i 6= j). In particular, the number of

orbits of Γ1 αΓ2 under the action of Γ1 is finite.

Note 3.1.7. Note that the action of SL2 (R) on H extends naturally to GL+
2 (R).

Definition 3.1.8.

(i) Let k ∈ Z. For a function f : H → C and ac db ∈ GL+

2 (R) we define
    
a b k−1 −k az + b
f |k (z) = (ad − bc) (cz + d) f .
c d cz + d

g ∈ GL+
(ii) Let Γ1 , Γ2 ≤ SL2 (Z) of finite index, S 2 (Q) and let β1 , . . . , βr ∈ GL2 (Q)
+

be an orbit decomposition Γ1 gΓ2 = Γ1 βi as in Corollary 3.1.6. For f weakly

modular of weight k and level Γ1 we define
r
X
f |k [Γ1 gΓ2 ] = f |k βi .
i=1

Proposition 3.1.9. f |k [Γ1 gΓ2 ] is independent of the choice of the βi ’s, and it is weakly
modular of weight k and level Γ2 .

Proof. If β10 , . . . , βs0 is another set of coset representatives then we see that s = r. So we
can reorder such that βi = γi βi0 for some γi ∈ Γ1 . Hence f |k βi = f |k βi0 , so f |k [Γ1 gΓ2 ] is
independent of the choice of the βi ’s.
In particular, if β1 , . . . , βr is one such choice then so is β1 γ, . . . , βr γ2 for any γ2 ∈ Γ2 .
Hence the sum r r r
X X X
f |k βi = f |k (βi γ) = ( f |k βi )|k γ,
i=1 i=1 i=1
Pr
so i=1 f |k βi is weakly modular of weight k and level Γ2 .

45
Note 3.1.10. Note that acting on the right of f |k [Γ1 gΓ2 ] by Γ2 is effectively permuting
summands.

Proposition 3.1.11. If f is a a modular form or a cusp form of level Γ1 then so is

f |k [Γ1 gΓ2 ] of level Γ2 .

Proof. If f is a modular function, a modular form or a cusp form of level Γ1 then so is

each term f |k βi of level βi−1 Γ1 βi ∩ SL2 (Z). Hence all the f |k βi are of the same type of
level Γ0 := SL2 (Z) ∩ ∩ri=1 βi−1 Γ1 βi ∩ Γ2 and thus so is f |k [Γ1 gΓ2 ].
But all of the properties for a function being a modular function, a modular form or
a cusp form of some level Γ are satisfied ifthese properties are already satisfied at any
smaller level Γ0 ⊆ Γ of finite index. So we can descend from Γ0 to Γ2 .

Remark 3.1.12. We thus have a map

[Γ1 gΓ2 ]
Mk (Γ1 ) −−−−→ Mk (Γ2 ).

This map preserves cusp forms and hence induces a map

Mk (Γ1 )/Sk (Γ1 ) → Mk (Γ2 )/Sk (Γ2 ).

Examples 3.1.13. (1) If g −1 Γ1 g = Γ2 then Γ1 gΓ2 = Γ1 g = gΓ2 . So the map f 7→

f |k [Γ1 gΓ2 ] is just f 7→ f |k g.

(2) More generally, if g −1 Γ1 g ⊇ Γ2 then this map is still f 7→ f |k g, but it is not an

isomorphism anymore.

(3) If Γ1 ⊃ Γ2 and g = Id, then Γ1 gΓ2 = Γ1 , and Γ1 = Γ1 . Id is an orbit decomposition.

Then fk |[Γ1 gΓ2 ] = fK Id = f . This just says that M (Γ1 ) is a subgpace of M (Γ2 ).

(4) Suppose Γ1 ⊆ Γ2 and g = 1. Then the αi ’s are just coset representatives for Γ1 \Γ2
and we are sending
X
f 7→ f |k γ.
γ∈Γ1 \Γ2

This is a surjective map Mk (Γ1 ) → Mk (Γ2 ). The restriction of this map to

Mk (Γ2 ) ⊆ Mk (Γ1 ) is just the multiplication by the index [Γ2 : Γ1 ]. (The map
is called the ”trace map” from level Γ1 to level Γ2 .)

(5) The last
example is a much more subtle one. Let Γ = Γ1 = Γ2 = SL2 (Z) and
g = 10 p0 for some prime p. Then
  
−1

0 a b
Γ ∩ g Γg = Γ (p) = : p divdes b .
c d

46


One can check that Γ0 (p)\Γ is given by the coset representatives 10 1j j=0,...,p−1 and
0 −1


1 0 . So for f ∈ Mk (Γ) we have
p−1      
X 1 0 1 j 1 0 0 −1
f |k [ΓgΓ] = f |k + f |k
0 p 0 1 0 p 1 0
j=0
p−1    
X 1 j 0 −1
= f |k + f |k
0 p p 0
j=0
p−1    
X
k−1 −k z+j k−1 −k 1
= p p f +p (pz) f − .
j=0
p pz

But f is a modular form of level SL2 (Z), so

    
−k 1 0 −1
(pz) f − = f |k (pz) = f (pz).
pz 1 0
Therefore we get
p−1  
1X z+j
f |k [ΓgΓ] = f + pk−1 f (pz).
p j=0 p

We extract the following lemma from Example (5).

Lemma 3.1.14. Let H be the subgroup of SL2 (Z/pZ) consisting of the lower-triangular
matrices. Then we have
p−1
G
H\ SL2 (Z/pZ) = H ᾱj t H β̄,
j=0

0 −1
1 j



where ᾱj = 0 1 for j = 0, . . . , p − 1 and β̄ = 1 0 .
Definition 3.1.15. (a) Let Γ1 , Γ2 ≤ SL2 (Z) of finite index. We define R(Γ1 , Γ2 ) to
be the C-vector space with basis the symbols [Γ1 gΓ2 ] for each g ∈ Γ1 \ GL+
2 (Q)/Γ2 .

(b) Let Γ1 , Γ2 , Γ3 ≤ SL2 (Z) of finite index. We define a multiplication

R(Γ1 , Γ2 ) × R(Γ2 , Γ3 ) → R(Γ1 , Γ3 ).
For [Γ1 gΓ2 ] ∈ R(Γ1 , Γ2 ) and [Γ2 hΓ3 ] ∈ R(Γ2 , Γ3 ) write
s
a t
a
Γ1 gΓ2 = Γ1 λi and Γ2 hΓ3 = Γ2 µj .
i=1 j=1

We define X
[Γ1 gΓ2 ] × [Γ2 hΓ3 ] := cγ · [Γ1 γΓ3 ]
γ∈Γ1 \ GL+
2 (Q)/Γ3

where
cγ := |{(i, j) ∈ {1, . . . , s} × {1, . . . , t} : λi µj ∈ Γ1 γ}| .

47
Remark 3.1.16. It is tedious to check that this definition is indeed well-defined, so
independent of the choice of λi and µj , and that this multiplication is associative, so
   
[Γ1 gΓ2 ] × [Γ2 hΓ3 ] × [Γ3 jΓ4 ] = [Γ1 gΓ2 ] × [Γ2 hΓ3 ] × [Γ3 jΓ4 ].

Moreover, we have to check that the introduced multiplication satisfies

   
f |k [Γ1 gΓ2 ] × [Γ2 hΓ3 ] = f |k [Γ1 gΓ2 ] |k [Γ2 hΓ3 ].

In particular, R(Γ) := R(Γ, Γ) is a ring and Mk (Γ) and Sk (Γ) are right modules over it.

3.2 The Hecke algebra of Γ1(N )

Lemma 3.2.1. Let Γ be any congruence subgroup containing Γ(N ). If p is a prime
which is comprime to N , then Γ surjects onto SL2 (Z/pZ) under reduction (mod p).
Proof. It is clearly sufficient to prove the result for Γ = Γ(N ). We know by Strong
Approximation (Question Sheet 4) that the map
SL2 (Z) → SL2 (Z/N pZ)
is sujective. Since N and p are coprime, we have
SL2 (Z/N pZ) ∼
= SL2 (Z/N Z) × SL2 (Z/pZ),
so we deduce that the map
SL2 (Z) ∼
= SL2 (Z/N Z) × SL2 (Z/pZ), x 7→ (x (mod N ), x (mod p))
is surjective. It follows that for any element Ā ∈ SL2 (Z/pZ) there exists A ∈ SL2 (Z)
which maps to (Id, Ā). Since
Γ(N ) = ker (SL2 (Z) → SL2 (Z/N Z)) ,
this finishes the proof.
Proposition 3.2.2. Let p be prime, N ≥ 1 and Γ = Γ0 (N ) or Γ = Γ1 (N ).
(i) If p divides N then
  p−1  
1 0 a 1 i
Γ Γ= Γ .
0 p 0 p
i=0

(ii) If p does not divide N then

  p−1    
1 0 a 1 i p 0
Γ Γ= Γ t Γγ
0 p 0 p 0 1
i=0

a b


where γ = 1 in the case of Γ0 (N ) and γ = N p in the case of Γ1 (N ) with a, b
being any integers such that ap − bN = 1.

48
  
1 0
Proof. Let g = . For Γ = Γ0 (N ) or Γ = Γ1 (N ), let
0 p
Γ0 = Γ ∩ g −1 Γg .

We need to find representatives for the quotient Γ0 \Γ.

1. Assume p - N .
Now for Γ = Γ0 (N ), we have
  
0 a b
Γ = : p divides b, N divides c
c d
and for Γ = Γ1 (N ) that
  
0 a b p divides b, N divides c
Γ = : .
c d a = d = 1 mod N
Hence in both cases the image Γ̄0 = Γ0 (mod p) is ∗∗ 0∗ ⊂ SL2 (Z/pZ), and by

Lemma 3.1.14 we have

p−1
G
Γ̄0 \ SL2 (Z/pZ) = Γ̄0 ᾱj t Γ̄0 β̄,
j=0

By Lemma 3.2.1, we know that there exists lifts
of the coset representatives to Γ.
For ᾱj , this is easy: we take the lift αj = 10 1j .
For Γ̄0 β̄, we need to find an element β of Γ0 (N ) or Γ1 (N ) whose reduction (mod p)
lies in the coset
0 −1


∗ 0 = 0? ?? SL2 (Z/pZ).
∗ ∗ 1 0
This will be satisfied by any matrix β ∈ Γ which
• for Γ = Γ0 (N ), is of the form Npac db with a, b, c, d ∈ Z such that pad − N bc =

1;
• for Γ = Γ1 (N ), is of the form Npac db with a, b, c, d ∈ Z such that pad − N bc =

1, and such that

pa = d = 1 (mod N ).
We make the spicific choice that c = d = 1; it is then easy to see that we can
find a, b which satisfy pa − N b = 1; note that this automatically implies that
pa = 1 (mod N ).
Hence we obtain the decomposition
p−1
[
0
Γ \Γ = Γ0 αj t Γ0 β
j=0
  p−1
1 0 [
Γ0 1 0

1 0


⇒ Γ Γ= 0 p αi t Γ 0 p β.
0 p
j=0

49
 Now write      
1 0 pa b a b p 0
= ,
0 p Nc d N c pd 0 1
so we can write   
1 0

a b p 0
Γ 0 p β=Γ .
N c pd 0 1
 
a b
In the case Γ = Γ0 (N ), the matrix is an element of Γ, so we have
N c pd
 
1 0

p 0
Γ 0 p β=Γ .
0 1

For Γ = Γ1 (N ) and our choice c = d = 1, we get the claimed result.

2. Assume p|N . Then one can check that 10 1j j=0,...,p−1 is a set of coset representatives
for (Γ ∩ g −1 Γg)\Γ, so we don’t need αp since any element of Γ has diagonal entries
coprime to p.

Corollary 3.2.3. Let Γ = Γ0 (N ) or Γ = Γ1 (N ), and let f ∈ Mk (Γ).

1. If p divides N , then
    p−1  
1 0 1X z+i
Γ Γ (f ) = f
0 p p i=0 p

2. If p does not divide N , then

    p−1  
1 0 1X z+i
Γ Γ (f ) = f + pk−1 (f |k γ) (pz),
0 p p i=0 p

where γ is as in Proposition 3.2.2. In particular, in the case Γ = Γ0 (N ) the term

f |k γ reduces to f .
   
1 0
Definition 3.2.4. Write Tp for the operator Γ1 (n) Γ1 (N ) .
0 p

3.2.1 Diamond operators

Definition 3.2.5. Let N ≥ 1. A Dirichlet charachter mod N is a homomorphism
χ : (Z/N Z)× → C× .

50
Example 3.2.6. The map
(Z/4Z)× → C× , 1 7→ 1, 3 7→ −1
is a Dirichlet character mod 4. In particular, it is the only non-trivial character mod 4.
An example of a character mod 13 is the map
(Z/13Z)× → C× , 2 7→ e2πi/12 ,
which is well-defined since 2 generates (Z/13Z)× .
Note 3.2.7. If M divides N any Dirichlet character mod M induces a character mod
N.
Definition 3.2.8. We say a character χ is primitive if it is not induced from a character
(mod M ) for any M dividing N , M < N .
Example 3.2.9. The characters in Example 3.2.6 above are primitive characters. How-
ever, the character
χ : (Z/8Z)× → C× , 1, 5 7→ 1, 3, 7 7→ −1
is not primitve since it comes from the above character mod 4.
Note 3.2.10. If χ is a Dirichlet character (mod N ), it can be extended to a map
χ : Z → C by the recipe
(
χ(d (mod N )) if (d, N ) = 1
χ(d) =
0 if (d, N ) > 1
The resulting function is multiplicative: it satisfies
χ(d1 d2 ) = χ(d1 )χ(d2 ) ∀d1 , d2 ∈ Z.
Lemma 3.2.11. The map
ι : Γ0 (N ) → (Z/N Z)× , ( ac db ) 7→ d (mod N )
is well-defined, and it induces an isomorphism
Γ0 (N )/Γ1 (N ) ∼
= (Z/N Z)× .
Definition 3.2.12. Let d ∈ (Z/N Z)× , and let g ∈ Γ0 (N ) such that ι(g) = d (mod N ).
Then the diamond operator hdi is the double coset operator Γ1 (N )gΓ1 (N ) ∈ R(Γ1 (N )).
Note 3.2.13. Since Γ1 (N ) is normal in Γ0 (N ), we have
Γ1 (N )gΓ1 (N ) = Γ1 (N )g = gΓ1 (N ).
The map
(Z/N Z)× → R(Γ1 (N )), d 7→ hdi
is hence a group homomorphism, and we get an action of (Z/N Z)× by linear operators
on Mk (Γ1 (N )) and Sk (Γ1 (N )).

51
 Recall the following result from the representation theory of finite groups:

Proposition 3.2.14. Let V be any complex vector space with an action of (Z/N Z)× by
linear operators. Then
M
V = Vχ
χ : (Z/N Z)× →C×

where
V χ = v ∈ V : g.v = χ(g) · v for all g ∈ (Z/N Z)× .


Definition 3.2.15. Let χ be a Dirichlet character. We define Mk (Γ1 (N ), χ) as the

χ-eigenspace Mk (Γ1 (N ))χ for the action of (Z/N Z)× . In other words,

Mk (Γ1 (N ), χ) = {f ∈ Mk (Γ1 (N )) : hdif = χ(d)f ∀d ∈ (Z/N Z)× }.

This is called the space of modular forms of weight k, level N and character χ.
We similarly define Sk (Γ1 (N ), χ).

Example 3.2.16. If 1N is the trivial character mod N then

Mk (Γ1 (N ), 1N ) = Mk (Γ0 (N )).

To see this consider f ∈ Mk (Γ1 (N ), 1N ). Then for all g ∈ Γ0 (N ) we have

f |k g = hι(g)if = f.

Note 3.2.17. We have Mk (Γ1 (N ), χ) = {0} unless χ(−1) = (−1)k .

3.2.2 Hecke operators on q-expansions

Definition 3.2.18. Define the following two operators on formal q-expansions: let q =
e2πiz , and define
X
Up .f = anp q n ,
X
Vp .f = an q np
P∞
Lemma 3.2.19. If f = n=0 an q n , then

p−1   X p−1  
1X z+j 1 j
Up .f = f = f |k ,
p j=0 p 0 p
j=0
 
p 0
Vp .f = p1−k f |k .
0 1

52
 2πi
Proof. Note that if ζp = e p , then
p−1
(
X p if p|n
ζpnj =
j=0
0 if p - n

Now
p−1   p−1  
X 1 j k−1 −k
X z+j
f |k =p p f
0 p p
j=0 j=0
p−1 ∞
1 X X z+j
= an e2πin p
p j=0 n=0
∞ p−1
!
X 2πinz 1X
= an e p ζpnj
n=0
p j=0

= Up .f

by Lemma 3.2.19. The statement for Vp is clear.

Theorem 3.2.20. If f ∈ Mk (Γ1 (N )), then
(
Up .f if p|N
Tp .f =
Up .f + pk−1 Vp hpi.f if p - N

Proof. Immediate from Corollary 3.2.3 and Lemma 3.2.19.

Corollary 3.2.21. If f ∈ Mk (Γ1 (N ), χ), then for all p we have

Tp .f = Up .f + χ(p)pk−1 Vp .f.

Note 3.2.22. Recall that χ(p) = 0 if p | N .

Example 3.2.23. Consider the Eisenstein series
∞
2k X
Ek (z) = 1 − σk−1 (n)q n ∈ Mk (Γ1 (1)).
Bk n=1

Claim. Ek (z) is an eigenform for all Tp , and

Tp .Ek = σk−1 (p)Ek = (1 + pk−1 )Ek .

Proof of claim. By Theorem 3.2.20, we have for any f ∈ Mk (Γ1 (1))

an (Tp .f ) = an (Up .f ) + pk−1 an (Vp .f ) = anp (f ) + pk−1 an/p (f ),

where we understand that an/p (f ) = 0 if p - f . Hence

a0 (Tp Ek ) = a0 (Ek ) + pk−1 a0 (Ek ) = σk−1 (p).

53
For n ≥ 1, we get
2k
σk−1 (np) + pk−1 σk−1 (n/p) ,


an (Tp .Ek ) = −
Bk
where σk−1 (n/p) = 0 if p - n. We now want to show that

σk−1 (pn) + pk−1 σk−1 (n/p) = σk−1 (n)σk−1 (p) ∀n ≥ 1.

• For p - p, this is just the multiplicativity of σk−1 .

• if p|n, write n = pe m with p - m. Then we need to show that

σk−1 (pe+1 m) + pk−1 σk−1 (pe−1 m) = σk−1 (p)σk−1 (pe m)

⇔ σk−1 (pe+1 ) + pk−1 σk−1 (pe−1 ) = σk−1 (p)σk−1 (pe ). (3.1)

since p and m are coprime. But (3.1) can easily seen to be true.

3.2.3 The Hecke algebra

Definition 3.2.24. For λ ∈ Q× write Rλ for the Hecke operator Γ1 (N ) λ0 λ0 Γ1 (N ) .



Define T (Γ1 (N )) as the subalgebra of R(Γ1 (N )) generated by the operators Tp , Rλ and

hdi for all primes p, λ ∈ Q× and d ∈ (Z/N Z)× .

Proposition 3.2.25. The algebra T (Γ1 (N )) is commutative.

We will only sketch the proof:

Proof. The Rλ ’s commute with everything since λ0 λ0 is central in GL+


2 (Q) and the
hdi’s commute with each other. So it remains to show that the Tp ’s commute with each
other and with the hdi’s.
We will first show that for p, q distinct primes, we have
 
1 0
Tp Tq = Tq Tp = Γ1 (N ) Γ1 (N ).
0 pq

To simplify the notation, let Γ = Γ1 (N ).
Recall the multiplication

in R(Γ): write
1 0 1 0
F F
Tp = Γαi , Tq = Γβj , with αi ∈ Γ 0 p Γ, βj ∈ Γ 0 q Γ. (Of course we know what
the αi , βj are explicitly, but we do not use that here.) Then
X
Tp Tq = cγ · [ΓγΓ],
γ∈Γ\ GL+
2 (Q)/Γ

where cγ := |{(i, j) : αi βj ∈ Γγ}|.

Claim. For all α ∈ Γ 10 p0 Γ, β ∈ Γ 1 0
Γ, we have
0 q


αβ ∈ Γ 10 pq
0
.

54
 Proof of claim. Note
that αβ has determinant pq, so by the Smith normal form we
1 ∗
1 0


have αβ ∈ SL2 (Z) 0 pq SL2 (Z). One can show that since αβ = 0 pq mod N we in
fact have  
1 ∗
αβ ∈ Γ Γ.
0 pq


This proves that the product Tp Tq is a constant multiple of Γ1 (N ) 10 pq 0
Γ1 (N ) and
one can check that this constant is indeed one.

It remains to check that Tp hdi = hdiTp . Let γ = ac db as in the definition of hdi. Note
that since Γ1 (N ) is normal in Γ0 (N ) we have
   
1 0 −1
hdi Tp = (Γ1 (N )γ) Γ1 (N )γ γ Γ1 (N )
0 p
   
−1

1 0 −1
= Γ1 (N ) γΓ1 (N )γ γ γ γΓ1 (N )
0 p
   
1 0 −1
= Γ1 (N )γ γ Γ1 (N ) hdi .
0 p

But γ 10 p0 γ −1 has determinant p and is 10 p∗ mod N . By multiplying

on the right
1 1 1 0
by some power
of 0 1 ∈ Γ1 (N ) we can make this be 0 p mod N . So it is in
Γ1 (N ) 10 p0 Γ1 (N ) and thus Tp hdi = hdi Tp .

Definition 3.2.26. For a prime power n = pr , r ≥ 2, we define Tn by

(
(Tp )r , if p divides N ,
Tp =
r
Tpr−1 Tp − p Rp Tpr−2 hpi , if p does not divide N .

For general n = pr11 . . . prkk we define Tn = Tpr11 . . . Tprk .

k

Note 3.2.27. We have Tn ∈ T (Γ1 (N )) for all n ∈ N by definition. In particular all Tn ’s

commute.

Proposition 3.2.28. Let f ∈ Mk (Γ1 (N )), and let m, n be comprime. Then am (Tn f ) =
amn (f ). In particular, we have a1 (Tn f ) = an (f ).

Proof. First a prime power n = pr . By induction and using proposition 3.2.20 we get
∞
X ∞
X
Tpr (f ) = anpr (f )q n + pk−1 anpr−1 (hpi f ) q np
n=0 n=0
∞
X
2
+ p 2(k−1)
anpr−2 hpi2 f q np
n=0
∞
r
X
+ ... + p r(k−1)
an (hpir f ) q np .
n=0

55
If n = pr11 . . . prkk with gcd(n, m) = 1, then

am (Tn f ) = am (Tpr11 . . . Tprk f ) = ampr11 ((Tpr22 . . . Tprk f ) = · · · = ampr11 ...mrk (f ).

k k k

Remark 3.2.29. For general m, n (not necessarily comprime), one can show (exercise)
that X
am (Tn f ) = dk−1 a mn
d2
(hdif ) .
d|gcd(m,n)

Proposition 3.2.30. For all χ mod N the operators Tn preserve the subspaces Mk (Γ1 (N ), χ)
and Sk (Γ1 (N ), χ) of Mk (Γ1 (N )) and Sk (Γ1 (N )).
Proof. This follows from the commutativity of T (Γ1 (N )) as commuting operators pre-
serve each others eigenspaces.
Definition 3.2.31. We say f ∈ Mk (Γ1 (N )) is an Hecke eigenform (or just eigenform)
if it is a simultaneous eigenvector for all the operators in T (Γ1 (N )) (i.e. for all the Tn0 s
and hdi’s).
A normalized Hecke eigenform is an eigenform satisfying a1 (f ) = 1.
Note 3.2.32. Let f ∈ Mk (Γ1 (N )) be an eigenform, say Tn .f = λn f for all n. Then

an (f ) = a1 (Tn f ) = λn a1 (f ) ∀n ≥ 1.

It follows that if a1 (f ) = 0, then an (f ) = 0 for all n ≥ 1, so f is constant. Therefore a

non-constant eigenform must have a1 (f ) 6= 0, and it may be scaled to be a normalized
eigenform.
Theorem 3.2.33. Let f ∈ Mk (Γ1 (N )) be a normalized eigenform. Then the eigenvalues
of the Hecke operators Tn on f are the coefficients of the q-expansion of f at the cusp
∞: we have
Tn .f = an (f ) · f ∀ n ≥ 1.
Proposition 3.2.34. Let f ∈ Mk (Γ1 (N ), χ) be a modular form with q-expansion n≥0 an (f )q n
P
at ∞. Then f is a normalized eigenform if and only if
i a1 (f ) = 1;

ii amn (f ) = am (f )an (f ) for all m, n comprime;

iii apr (f ) = ap (f )apr−1 (f ) − pk−1 χ(p)apr−2 (f ) for all primes p and all r ≥ 2.
Proof. The implication ⇒ follows directly from Definition 3.2.26 and Theorem 3.2.33.
Conversely, if f ∈ Mk (Γ1 (N ), χ) satisfies properties (i)-(iii), then f is already normalized,
so we need to show that it satisfies

am (Tp .f ) = ap (f )am (f ) ∀ p prime, ∀m ≥ 1.

56
If p - m, then it follows from Proposition 3.2.28 that am (Tp .f ) = amp (f ), which by (ii)
is equal to am (f )ap (f ). If m = pr m0 with p - m0 , then by Remark 3.2.29 we have
am (Tp .f ) = apr+1 m0 (f ) + χ(p)pk−1 apr−1 m0 (f ).
Using (ii) and (ii), this can be shown to be equal to ap (f )am (f ) as required.
Question. Do such normalized eigenforms actually exist?
Example 3.2.35.
1. A non-Eisenstein eigenform is given by ∆ ∈ S12 (SL2 (Z)). This is clear since all Tn
preserve S12 and S12 is spanned by ∆. Moreover ∆ is obviously normalized. Let
τ (n) = an (∆). Then
τ (mn) = τ (m)τ (n)
for m and n coprime by Proposition 3.2.34. This shows a statement which was
made in the prologue of this lecture.
2. Similarly we can show that the cusp forms E4 ∆, E6 ∆, E42 ∆, E4 E6 ∆ and E42 E6 ∆
of level (SL2 (Z)) and weight 16, 18, 20, 22 and 26 are normalized eigenforms since
the corresponding spaces of cusp forms are one-dimensional.
3. More interesting is the case k = 24 since S24 (SL2 (Z)) is two-dimensional. It can
easily be shown that S24 (SL2 (Z)) is spanned by f1 = E43 ∆ and f2 = ∆2 . The
q-expansion of these are given by
f1 = q + 696q 2 + 162252q 3 + 128318089q 4 + . . .
f1 = q 2 − 48q 3 + 1080q 4 + . . . .
We want to know how T2 acts on this basis. By the formula in the proof of
Proposition 3.2.28, we have
T2 (f1 ) = 696q + 128318089q 2 + . . . + 223 q 2 + 696q 4 + . . .

= 696q + 136706697q 2 + . . .
and
T2 (f2 ) = q + 1080q 2 + . . . + 223 q 4 + . . .

= q + 1080q 2 + . . .
In terms of the given basis we therefore have
T2 (f1 ) = 696f1 + 136222281f2
T2 (f2 ) = f1 + 384f2 .
Thus T2 is given by the matrix
 
696 1
.
136222281 384
Open conjecture (Maeda’s conjecture). The Galois group of the splitting field of the
characteristic polynomial of T2 on Sk (SL2 (Z)) is as large as possible, so isomorphic to
the symmetric group Sym(d) where d = dim(Sk ).

57
3.3 The Petersson product
The aim of this section is to define a Hermitian inner product on the space Sk (Γ1 (N )).

Lemma 3.3.1. Let U ⊆ H be a closed set whose boundary consists of finitely many line
segments and circle arcs. Let, f : U → C be a continuous function, and let γ ∈ SL2 (R).
Then Z Z
dx dy dx dy
f (z) 2 = f (γ.z) 2 ,
z∈U y γ −1 U y
where we write z = x + iy.


Proof. We view H as an open subset of R2 with coordinates (x, y) and γ = a b
c d as a
differentiable map H → H. Write

γ1 (x, y) = <γ(x + iy) and γ2 (x, y) = =γ(x + iy).

The Jacobian matrix of γ at a point z = x + iy is

 ∂γ1 ∂γ2 
Jγ = ∂γ ∂x ∂x .
1 ∂γ2
∂y ∂y

Since γ is holomorphic, it satisfies the Cauchy-Riemann equations

∂γ2 ∂γ1 ∂γ2 ∂γ1

= , =− ,
∂y ∂x ∂x ∂y

and we have
∂γ1 ∂γ2
Γ0 (z) = +i .
∂x ∂x
Hence  ∂γ1 ∂γ2 
= |γ 0 (z)|2 .

∂x
|Jγ | =  ∂γ ∂x 
1 ∂γ2 
∂y ∂y

On the other hand we have

 a(cz + d) − c(az + b) 2
   2
0 2 1 =(γz)
|γ (z)| =   = = ,
(cz + d)2  |cz + d|4 =(z)
=(z)
since =(gz) = |j(g,z)|2
. This yields
Z Z
dx dy dx dy
f (z) 2 = f (γ.z) |Jγ |
z∈U y −1 (=(γz))2
Zz∈γ U
dx dy
= f (γ.z) 2 .
z∈γ −1 U y

58
 dx∧dy
Remark 3.3.2. What we have shown that the differential 2-form y2
is SL2 (R)-
invariant.
Notation. Let Γ ≤ SL2 (Z) be a finite index subgroup, let R be a set of coset repres-
entatives for Γ\ PSL2 (Z) and let S
D be the fundamental domain as defined in theorem
1.2.2. We then denote the union γ∈R γD by DΓ .
Lemma 3.3.3. Let F : H → C be a continuous function with is Γ-invariant, i.e.

F (γ.z) = F (z) ∀ γ ∈ Γ, z ∈ H.

Then the value of the integral Z

dx dy
F (z)
z∈DΓ y2
does not depend on the coice of the system of coset representatives R.
Proof. Immediate from Lemma 3.3.1.
Definition 3.3.4. We define the following regions around the cusps: For Y > 0, let

UY = {x + iy ∈ H : |x| ≤ 1/2, y ≥ Y }.

Note 3.3.5. The fundamental domain DΓ is the union of some compact set K ⊂ H and
the set
γ UY {γ.z|γ ∈ R, z ∈ UY }.
Lemma 3.3.6. Let F be as in Lemma 3.3.3. Suppose that for all γ ∈ SL2 (Z) there exist
real numbers cγ > 0 and eγ < 1 such that

|F (γ.z)| ≤ cy · (=(z))eγ ∀z with =(z) sufficiently large. (3.2)

Then the integral Z

dx dy
F (z) (3.3)
z∈DΓ y2
converges.
Proof. The restriction of the integral (3.3) to K clearly converges since K is compact. It
therefore remains to show that the integral converges on each of the sets γUY for γ ∈ R.
By Lemma 3.3.1, we have
Z  Z 
 dx dy   dx dy 
 F (z) 2  =  F (γ.z) 2 

z∈γUz y z∈Uz y
Z
dx dy
≤ cγ y eγ 2
y
Zz∈U
∞
Y

= cγ y eγ −2 dy.
y=Y

This converges since eγ < 1 by assumption.

59
Note 3.3.7. Condition (3.2) is in particular satisfied if F tends to 0 exponentially at
the cusps.
Proposition 3.3.8. Let Γ ≤ SL2 (Z) be a finite index subgroup, k ≥ 1 and f, g ∈ Mk (Γ).
Define F : H → C by
F (z) = f (z)g(z)(=(z))k .
Then F is Γ-invariant. If at least one of f and g vanishes at each cusp then F tends
exponentially to 0 at each cusp and hence
Z
dx dy
F (x + iy) 2
DΓ y
converges.
=(z)
Proof. F as defined is of weight 0 Γ-invariant since f, g ∈ Mk (Γ) and =(gz) = |j(g,z)| 2.

Moreover, the decay at the cusps can easily be shown by considering the product of the
q-expansions of f and g at the corresponding cusps: the product will be a function in q
without a constant term, so it certainly tends to 0 exponentially. It hence follows from
Lemma 3.3.6 that the integral converges.
Definition 3.3.9. Let Γ ≤ SL2 (Z) be a congruence subgroup, k ≥ 1 and f, g ∈ Mk (Γ),
at least one of f and g vanishing at every cusp. Then we define the Petersson product
as Z
hf, giΓ = f (z)g(z)(=(z))k−2 dx dy.
DΓ

Note 3.3.10. The Petersson product is well-defined by Proposition 3.3.8.

Proposition 3.3.11. The Petersson inner product is a positive definite inner product
on the C-vector space Sk (Γ): it satisfies
1. ha1 f1 + a2 f2 , gi = a1 hf1 , gi + a2 hf2 , gi for all a1 , a2 ∈ C, f2 , f2 , g ∈ Sk (Γ);

2. hf, gi = hg, f i;

3. hf, f i ≥ 0 with equality if and only if f = 0.

We now want to show that the subspact of Mk (Γ) spanned by the Eisenstein series is
orthogonal to Sk (Γ).
Proposition 3.3.12. Let Γ ≤ SL2 (Z) be a congruence subgroup, k ≥ 3 and c ∈
Cusps(Γ). If k is odd, assume that Γ is regular at c . Then hGk,Γ,c , f i = 0 for all
f ∈ Sk (Γ).
We will sketch the proof of the proposition. We need some preparatory lemmas:
Lemma 3.3.13. Let f ∈ Sk (Γ). Then there is C > 0 such that
C
|f (z)| ≤ .
=(z)k/2

60
Proof. Assume first that Γ = SL2 (Z). Let f ∈ Sk (Γ). Define F (z) = |f (z)|=(z)k/2 . We
can check that g is SL2 (Z)-invariant. So it is bounded on H if and only if it is bounded
on the funcamental domain D. But D ∩ {z ∈ C : |=(z)| ≤ R} is compact and F (z)| → 0
as =(z) → ∞ since f˜ is holomorphic at 0 and vanishes there, so |f˜(q)| < C|q| for small
q and some C > 0. But q = e2πiz decreases faster than =(z)k/2 increases.
The argument easily generalizes to general congruence subgroups.

Proposition 3.3.14. Let f ∈ Sk (Γ). Then there is M > 0 such that

k
|an (f )| ≤ M n 2 .

Proof. Let f ∈ Sk (Γ) and Cy be a small circle around the origin described by e2πi(x+iy)
where y is fixed and 0 ≤ x ≤ 1. Since

f˜(qh )qh−n−1 = · · · + an qh−1 + an+1 + an+2 qh + . . . ,

the residue theorem implies that

Z ˜
1 f (q)
an (f ) = dq
2πi Cy q n+1
Z 1
= f (x + iy)qh−n dx.
0

Using Lemma 3.3.13 and using that qh = e2πiz/h with z = x + iy, we get
Z 1
C  −2πin/h(x+iy) 
|an (f )| ≤ k/2
e dx
0 y
= Cy −k/2 e2πyn/h .

This expression holds for all y > 0, so in particular for y = nh , we get

|an (f )| ≤ Ce2π nk/2 .

Setting M = Ce2π finishes the proof.

We can now prove Proposition 3.3.12.

Proof. It can easily be checked that hf, giΓ = hf |k γ, g|k γiγ −1 Γγ for all γ ∈ SL2 (Z). So
we can assume that c = ∞ without loss of generality. Let f ∈ Sk (Γ). By definition we
have
Z !
X
hf, Gk,Γ,∞ iΓ = f (z) 1|k γ(z) (=(z))k−2 dx dy.
DΓ
γ∈Γ+
∞ \Γ

61
Now we can interchange integral and sum and afterwards apply lemma 3.3.1. This yields
X Z
hf, Gk,Γ,∞ iΓ = f (z)(1|k γ(z))(=(z))k−2 dx dy
DΓ
γ∈Γ+
∞ \Γ
X Z
= (f |k γ −1 )(z)1(z)(=(z))k−2 dx dy
γDΓ
γ∈Γ+
∞ \Γ
X Z
= f (z)(=(z))k−2 dx dy
γDΓ
γ∈Γ+
∞ \Γ
Z
= f (z)(=(z))k−2 dx dy.
Γ+
∞ \H

This is well-defined since the integrand is Γ+∞ -invariant. So

Z
hf, Gk,Γ,∞ i = f (z)(=(z))k−2 dx dy
+
Γ \H
Z ∞
= f (x + iy)y k−2 dx dy
0<y<∞
0≤x≤h
Z ∞
X
= an (f )e2πin(x+iy)/h y k−2 dx dy.
0<y<∞
0≤x≤h n=1

Now since an (f ) = O(nk/2 ) by Proposition 3.3.14,

Z ∞
X  
an (f )e2πin(x+iy)/h y k−2  dx dy < ∞,
0<y<∞
0≤x≤h n=1

so we can interchange the order of summation and integration. Thus

Z X Z
k−2
f (x + iy)y dx dy = an (f ) y k−2 e2πin(x+iy)/h dx dy
0<y<∞ 0<y<∞
0≤x≤h n≥0 0≤x≤h
X Z ∞ Z h
k−2 −2πyn/h
= an (f ) y e dy e2πinx/h dx.
n≥1 0 0

Rh 1
R1
But 0
e2πinx/h dx = h 0
e2πinx dx = 0. So the product we started with was 0.
Remark 3.3.15. If c and d are distinct cusps then hGk,Γ,c , Gk,Γ,d i is well-defined, but it
is not generally 0. Moreover, for k ≥ 3 can be shown that Ek (Γ), the subspace of Mk (Γ)
spanned by the Gk,Γ,c ’s, is exactly given by the set
{f ∈ Mk (Γ) : hf, gi = 0 for all g ∈ Sk (Γ)} .
In an abuse of notation we say Ek (Γ) is ”the orthogonal complement of Sk (Γ)”, which is
not correct since h·, ·i is not well-defined on all Mk (Γ). However h·, ·i certainly defines
a positive definite innder product on Sk (Γ). In particular we can take the above set as
the definition of Ek (Γ) for k = 1, 2.

62
3.4 Hecke operators and the Petersson product
Definition 3.4.1. Let V be a finite-dimensional C-vector space equipped with a positive
definite inner product h , , i, and let T : V → V be a linear operator. The adjoint of
T is the linear operator T ∗ : V → V such that hT x, yi = hx, T ∗ yi for all x, y ∈ V .

Definition 3.4.2. For Γ a congruence subgroup, we define

Z
dx dy
covol(Γ) = .
DΓ y2

Lemma 3.4.3. Let Γ be a congruence subgroup.

1. We have Z Z
dx dy dx dy
= dΓ · ,
DΓ y2 D y2
hwereD is the fundamental domain for SL2 (Z) and dΓ is the projective index of Γ.

2. Let g ∈ GL+ −1
2 (Q) such that g Γg ⊆ SL2 (Z). Then
Z Z
dx dy dx dy
2
= ,
Dg−1 Γg y DΓ y2

and Γ and g −1 Γg have the same index in SL2 (Z).

Proof. Exercise.

Definition 3.4.4. We normalize the Petersson inner product as follows: for Γ a con-
gruence subgroup, we let
Z
1
hf, giΓ = f (z)g(z)(=(z))k−2 dx dy,
covol(Γ) DΓ

Lemma 3.4.5. If Γ0 ⊂ Γ are congruence subgroups and f, g ∈ Sk (Γ), then

hf, giΓ = hf, giΓ0 .

Proof.
The aim of this section is to compute the adjoints of Hecke operators with respect to
the Petersson inner product. More precisely, we want to prove the following theorem:

Theorem 3.4.6. Let f1 , f2 ∈ Mk (Γ) for some Γ ≤ SL2 (Z) of finite index such that at
least one of f1 , f2 is a cusp form and let g ∈ GL+
2 (Q). Then

hf1 |k [ΓgΓ], f2 iΓ = hf1 , f2 |k [Γg 0 Γ]iΓ

where g 0 = det(g) · g −1 . So Γg 0 Γ is the adjoint of ΓgΓ.

63
 To prove this theorem we first need several technical results:

Proposition 3.4.7. Let Γ ≤ SL2 (Z) be a congruence subgroup and let g ∈ GL+
2 (Q) such
−1 −1
that g Γg ⊆ SL2 (Z). Then for any f1 ∈ Mk (Γ) and f2 ∈ Mk (g Γg) we have

hf1 |k g, f2 ig−1 Γg = hf1 , f2 |k g 0 iΓ ,

where g 0 = det(g) · g −1 as above and f1 , f2 are such that both sides are defined.

Exercise 3.4.8. If one of these sides is defined, so is the other.

Proof. An explicit calculation shows that for any compact U ⊆ Dg−1 Γg ,

Z Z
k−2
(f1 |k g)(z)f2 (z)(=(z)) dx dy = f1 (z)(f2 |k g 0 )(z)(=(z))k−2 dx dy.
U gU

If we let U grow into a fundamental domain for g −1 Γg then gU grows into a fundamental
domain for Γ and we are done.

Lemma 3.4.9. For any Γ ≤ SL2 (Z) of finite index and any g ∈ GL+
2 (Q) we have

Γ : Γ ∩ g −1 Γg = Γ : Γ ∩ gΓg −1 .
   

Moreover, if r is this common value, then there are elements α1 , . . . , αr ∈ GL+

2 (Q) such
that r r
a a
ΓgΓ = Γαi = αi Γ.
i=1 i=1

Proof. We first check the index. Let Γ0 = Γ ∩ gΓg −1 . Then g −1 Γ0 g = Γ ∩ g −1 Γg. Since
these are both contained in SL2 (Z), we deduce from Lemma 3.4.3 that

[SL2 (Z) : Γ0 ] = [SL2 (Z) : g −1 Γ0 g].

Since indices are multiplicative, this proves the first part.

There hence exist γ1 , . . . , γr and γ̃1 , . . . , γ̃r such that
a a
Γ= (Γ ∩ gΓg −1 )γi = (Γ ∩ g −1 Γg)γ̃i .

We hence deduce from Corollary 3.1.6 that

a
ΓgΓ = Γgγi ,
a a
Γg −1 Γ = Γg −1 γ̃i−1 ⇒ ΓgΓ = γ̃i gΓ. (3.4)

Claim. For all 1, ≤ i, j ≤ r, we have

Γgγi ∩ γ̃j gΓ 6= ∅.

64
 Proof of claim. Suppose that Γgγi ∩ γ̃j gΓ = ∅ for some i, j. Then
a
Γgγi ⊆ γ̃k gΓ.
k6=j

Multiplying on the left by Γ implies that

a
ΓgΓ ⊆ γ̃k gΓ,
k6=j

which contradicts (3.4).

For all 1 ≤ i ≤ r, choose αi ∈ Γgγi ∩ γ̃i gΓ. Then we have
r
a r
a
ΓgΓ = Γαi = αi Γ.
i=1 i=1

We are now able to prove Theorem 3.4.6.

Proof of Theorem 3.4.6. As in Lemma 3.4.9 let α1 , . . . , αr ∈ GL+
2 (Q) be such that
r
a r
a
ΓgΓ = Γαi = αi Γ.
i=1 i=1
`r
Inverting yields Γg −1 Γ = i=1 Γαi−1 . Since Γ ⊆ SL2 (Z), we have det(αi ) = det(g) for
all i. Hence r
a
0
Γg Γ = Γαi0 ,
i=1

where g 0 = det(g)g −1 and αi0 = det(αi )αi−1 . Then

r
X
hf1 |k [ΓgΓ], f2 iΓ = hf1 |k αi , f2 iΓ∩α−1
i Γαi
i=1
Xr
= hf1 , f2 |k αi0 iΓ∩αi Γα−1
i
i=1
= hf1 , f2 |k [Γg 0 Γ]iΓ .

Corollary 3.4.10. The operators ΓgΓ for g ∈ GL+

2 (Q) preserve Ek (Γ) ⊆ Mk (Γ).

Proof. The operator Γg 0 Γ preserves the cusp forms. Using theorem 3.4.6 we can pass to
the orthogonal complement: We have for any f1 ∈ Ek (Γ) and for any f2 ∈ Sk (Γ) that
hf1 |k [ΓgΓ], f2 iΓ = hf1 , f2 |k [Γg 0 Γ]iΓ = 0
since f2 |k [Γg 0 Γ] is in Sk (Γ). Thus f1 |k [ΓgΓ] is orthogonal to every cusp form and therefore
is in Ek (Γ).

65
Definition 3.4.11. In the setting of Definition 3.4.1, the operator T is normal if it
commutes with its adjoint T ∗ : T T ∗ = T ∗ T .
Theorem 3.4.12. Let N ≥ 1, Γ = Γ1 (N ), and consider the C-vector space Sk (Γ). Let
p be a prime not dividing N . Then

hpi∗ = hpi−1 = hp−1 i Tp∗ = hpi−1 Tp .

and


Proof. Recall that hpi = ΓαΓ, where α is any matrix α = ac pb ∈ Γ0 (N ). By Theorem
3.4.6, we have
hpi∗ = Γα−1 Γ
which is clearly equal to hpi−1 = hp
−1 i.
Now Tp = ΓαΓ, where α = 10 p0 , with adjoint Tp∗ = Γα0 Γ with α0 = p0 01 . In the

proof of Proposition 3.2.2, we saw that

a b −1 1 0
p 0



pa b

0 1 = N p 0 p N 1



where a, b satisfy ap − N b = 1. Now pa N 1
b
∈ Γ, and a b
N p ∈ Γ0 (N ) normalizes Γ.
Hence

Tp∗ = Γ p0 01 Γ



−1 1 0
pa b

= Γ Na pb 0 p N 1
Γ

−1
= Γ Na pb Γ · Tp
−1
= hp iTp .

Remark 3.4.13. More generally, we have

hhpi f, giΓ = hf, hpi−1 giΓ ,

hTp f, giΓ = hf, hpi−1 Tp giΓ

for all f, g ∈ Mk (Γ), at least one cuspidal.

Corollary 3.4.14. For n - N , the Hecke operators hni and Tn are normal.
We now recall the following result from linear algebra:
Theorem 3.4.15 (Spectral theorem). Let T be a normal operator on a finite dimen-
sional C-vector space V . Then V has an orthogonal basis of T -eigenvectors.
Corollary 3.4.16. Let N ≥ 1. Then the space Sk (Γ1 (N )) has an orthonormal basis
consisting of eigenforms for the operators hni and Tn , n coprime to N .
Proof. Clear by applying the spectral theorem to Sk (Γ1 (N )), using Corollary and the
fact that the Hecke operators commute.1
1
A family of commuting normal operators is simultaneously diagonalisable.

66
Remark 3.4.17. Considering Γ0 (N ) instead of Γ1 (N ) the same logic applies, but as
the hdi operators are trivial in this case the Tp are self-adjoint. Hence their eigenvalues
are real. Therefore Sk (Γ0 (N ) has a basis of modular forms with real eigenvalues for the
Tp ’s. In particular this applies to Γ = SL2 (Z).
The following example shows that the Tp , p not dividing N , can indeed fail to be
diagonalisable.
Example 3.4.18. Let f ∈ Sk (Γ1 (N )) and p prime not dividing N . Assume f is an
eigenvector for Tp . Now look at the space Sk (Γ1 (N p)). It contains f1 (z) := f (z) and
f2 (z) := f (pz). By comparing formulae for the Tp -action on the q-expansions (Theorem
3.2.20), which are not the same at N and N p, we find that
Tp f1 = λf1 − pk−1 χ(p)f2 , Tp f2 = f1
if f is a Tp -eigenform with eigenvalue λ and f has character χ. More generally at level
pj N , the space spanned by f (z), f (pz), . . . , f (pj z) is Tp -stable and the matrix of Tp looks
like  
λ 1 0 0 ··· 0
−pk−1 χ(p) 0 1 0 0
 
 0 0 0 1 0
..  .
 
 .. .. .. . .
 . . . . .
 
 0 0 0 0 1
0 0 0 0 ··· 0
Exercise 3.4.19. This matrix is not diagonalisable for j ≥ 3, independent of λ, χ and
k.
So there is an obstruction to diagonalise Tp for p dividing N coming from forms of
small level at p.

3.5 Old and new modular forms

Let N ≥ 1, and let p be a prime dividing N . Recall that if f (z) ∈ Sk (Γ1 (N/p)), then
f (pz) ∈ Sk (Γ1 (N )).
Definition 3.5.1. Let N ≥ 1, and let p be a prime dividing N .
1. Define
Sk (Γ1 (N ))p-old = i1,p (Sk (Γ1 (N/p))) + i2,p (Sk (Γ1 (N/p))) ,
where
i1,p : Sk (Γ1 (N/p)) ,→ Sk (Γ1 (N ))
is the natural inclusion and
i2,p : Sk (Γ1 (N/p)) ,→ Sk (Γ1 (N ))
maps f (z) to f (pz). (Note that this sum is not generally a direct sum.)

67
 2. The space of all old modular forms is defined to be y
X
Sk (Γ1 (N ))old = Sk (Γ1 (N ))p-old .
p prime, p|N

3. Define Sk (Γ1 (N ))p-new as the orthogonal complement of Sk (Γ1 (N ))p-old , and define
the space of new forms
\
Sk (Γ1 (N ))new = Sk (Γ1 (N ))p-new .
p prime, p|N

Remark 3.5.2. The space Sk (Γ1 (N ))new is precisely the orthogonal complement of
Sk (Γ1 (N ))old .

Proposition 3.5.3. The subspaces of Sk (Γ1 (N )) in Definition 3.5.1 are stable under
the operators Tn for all n ≥ 1 and hdi for all d ∈ (Z/N Z)× .

We first recall the formulae from Theorem 3.2.20: If f = n≥0 an q n ∈ Mk (Γ1 (N )),
P
then

T` .f = U` .f if `|N (3.5)
k−1
T` .f = U` .f + ` V` h`i.f if ` - N (3.6)

Here, X X
U` .f = an` q n , and V` .f = an q n` .

Note 3.5.4.

1. For ` 6= p, Vp commutes with U` and V` .

2. We have U` ◦ V` = id. (Exercise: what is V` ◦ U` ?)

Proof. It sufficies to show that the old subspaces are stable under the operators hdi, Tp
and their adjoints. By Theorem 3.4.12, the adjoints of Tp for p not dividing N and of hdi
for all d ∈ (Z/N Z)× are in the subalgebra T (Γ1 (N )), so we don’t need to worry about
them.
Firstly consider T` for ` not dividing N . Then the action of T` on Sk (Γ1 (N )) and
Sk (Γ1 (N/p)) is given by the formulae (3.6). Hence

i1,p (T` f ) = T` (i1,p f ) and i2,p (T` f ) = T` (i2,p f )

for all f ∈ Sk (Γ1 (N/p)), so T` preserves Sk (Γ1 (N ))p-old for all p dividing

N.
× a b
Now consider hdi for some d ∈ (Z/N Z) . Choose some γ = N c d ∈ Γ0 (N ). Then
γ ∈ Γ0 (N/p) also represents hdi ∈ R(Γ1 (N/p)). As functions on H we clearly have

i1,p (hdi f ) = hdi f = f |k γ = (i1,p f )|k γ = hdi (i1,p f ).

68
Furthermore
       
1−k p 0 1−k p 0 a b
hdi (i2,p f ) = hdi p f |k =p f |k |k .
0 1 0 1 Nc d

Note that      
p 0 a b a pb p 0
= .
0 1 Nc d N c/p d 0 1
Hence
    
1−k a pb p 0
hdi (i2,p f ) = p f |k |k
N c/p d 0 1
 
1−k p 0
= p (hdi f )|k
0 1
= i2,p (hdi f ).

We deduce that hdi stabilizes Sk (Γ1 (N ))p-old .

So it remains to show that Sk (Γ1 (N ))p-old is preserved under the action of Tq and Tq∗
for q dividing N . We first consider Tq :

1. If q divides N but q 6= p, then Tq is given by the same formula (3.5) at level N

and at level N/p. So we have

Tq ◦ i1,p = i1,p ◦ Tq , Tq ◦ i2,p = i2,p ◦ Tq

as in the case of q not dividing N .

2. If p2 is not dividing N (so p does not divide N/p), then for f ∈ Sk (Γ1 (N/p)) we
haveTp (i1,p f ) = Up (i1,p f ) by (3.5). On the other hand, using (3.6), we have

i1,p (Tp f ) = Up (i1,p f ) + pk−1 hpi Vp (f )

= Up (i1,p f ) + pk−1 i2,p (hpi f ),

Hence
Tp (i1,p f ) = i1,p (Tp f ) − pk−1 i2,p (hpi f ) ∈ Sk (Γ1 (N ))p-old .
On the other hand we have

Tp (i2,p f ) = Up (Vp (f )) = i1,p (f ).

Hence Tp preserves Sk (Γ1 (N ))p-old if p2 does not divide N .

3. If p2 divides N , then formula (3.5) applies for Tp both at level N and at level N/p.
So we have

Tp ◦ i1,p = i1,p ◦ Tp and Tp ◦ i2,p = Up ◦ Vp = i1,p .

69
We hence deduce that Tq preserves Sk (Γ1 (N ))p-old for all primes p, q.
Therefore we are left with checking that the adjoints Tq∗ , for q dividing N , preserve
old forms (equivalently, that the space of new forms is preserved by Tq ).
0 −1


Let wN = N 0 . This normalises Γ1 (N ), so Γ1 (N )wN Γ1 (N ) is a single left or right
coset (as for hdi’s), and defines an element of R(Γ1 (N )). We check that
   
∗ q 0 −1
Tq = Γ1 (N ) Γ1 (N ) = wN Tq wN
0 1
as    −1  
0 −1 1 0 0 −1 p 0
= .
N 0 0 q N 0 0 1
We will show that wN preserves the p-old subspace for all primes p dividing N , from
which it follows that Tq preserves the p-new subspace for all primes p, q dividing N .
As before, we compare wN with the corresponding operator at level N/p, namely wN/p .
 
k−1 −k 1
wN (i1,p f )(z) = N (N z) f −
Nz
 k−1  −k !
N N 1
= pk−1 pz f −N
p p p
pz
| {z }
=wN/p (f )(pz)
k−1
=p i2,p (wN/p f )(z).

On the other hand

 p 
wN (i2,p f )(z) = N k−1 (N z)−k f −
Nz !
 k−1  −k
−1 N N 1
=p z f −N
p p p
z
| {z }
=wN/p (f )(z)

= p−1 i1,p (wN/p f )(z).

This finishes the proof.

Exercise 3.5.5. Suppose that pj | N , but pj+1 - N . Then

Sk (Γ1 (N ))p−old = i1,p (Sk (Γ1 (N/p))) + i2,p (Sk (Γ1 (N/p))) + · · · + ij,p Sk (Γ1 (N/pj )) ,

where in,p sends f (z) to f (pn z).

Definition 3.5.6. The operator Γ1 (N )wN Γ1 (N ) ∈ R(Γ1 (N )) is called the the Atkin-
Lehner involution.
∗ ∗
Remarks 3.5.7. (i) We have wN = −wN . So wN preserves old subspaces and wN
preserves new ones.

70
 (ii) wN does not preserve character subspaces: wN maps Sk (Γ1 (N ), χ) to Sk (Γ1 (N ), χ).
Moreover, note that χ = χ−1 .
Proposition 3.5.8. Let χ be a primitive character mod N . Then

Sk (Γ1 (N ), χ) ⊆ Sk (Γ1 (N ))new .

Proof. The action of the hdi’s on Sk (Γ1 (N/p)), p dividing N , must factor through
(Z/(N/p)Z)× . The maps i1,p and i2,p commute with the hdi’s, so any hdi eigenvector in
an old subspace must have a character factoring through Z/(N/p)Z for some p. Thus it
will not be primitive.
Warning. The converse is not true: in general, it is not possible to tell from its
character whether a modular form is new.
Proposition 3.5.9 (An alternative definition of the new subspace). For p being a prime
dividing N , and assume2 that N/p 6∈ {1, 2}. Define maps
δ
1X
tr1,p : Mk (Γ1 (N )) → Mk (Γ1 (N/p)), f 7→ f |k γi ,
δ i=1
 
−1
tr2,p : Mk (Γ1 (N )) → Mk (Γ1 (N/p)), f 7→ p wN/p ◦ tr1,p ◦wN (f ),
`δ
where δ = [Γ1 (N/p) : Γ1 (N )] and γ1 , . . . , γδ such that Γ1 (N/p) = i=1 Γ1 (N )γi . Then
tr1,p ◦i1,p = id, tr2,p ◦i2,p = id and we have

Sk (Γ1 (N ))p-new = ker(tr1,p ) ∩ ker(tr2,p ).

Note that we can see δ tr1,p as the element [Γ1 (N ) · Γ1 (N/p)] in R(Γ1 (N ), Γ1 (N/p))
(c.f. Example 3.1.13 (4)).
Proof. We have for f ∈ Sk (Γ1 (N/p)) and g ∈ Sk (Γ1 (N )) that

hi1,p (f ), giΓ1 (N ) = hf, tr1,p (g)iΓ1 (N/p) ,

so the kernel of tr1,p is the orthogonal complement of the image of i1,p . Similarly
−1
hi2,p (f ), giΓ1 (N ) = hp−1 wN (i1,p (wN/p f )), giΓ1 (N )
−1 ∗
= p hwN (i1,p (wN/p f )), giΓ1 (N )
= p−1 hi1,p (wN/p f ), wN giΓ1 (N )
= p−1 hwN/p f, tr1,p (wN g)iΓ1 (N/p)
−1
= p−1 hf, wN/p (tr1,p (wN g))iΓ1 (N/p)
= p−2 hf, tr2,p (g)iΓ1 (N/p) .

So the kernel of tr2,p is the orthogonal complement of the image of i2,p .

2
This proof needs minor modifications in the case of N/p being 1 or 2 (why?).

71
Remark 3.5.10. It is an amusing fact, that this definition of ”new” and ”old” works
also for Eisenstein series, but there is an Eisenstein series of level 6 which is new and
old simultaneously.

Definition 3.5.11. A normalised eigenform in Sk (Γ1 (N ))new is called a primitive form.

Example 3.5.12. ∆ is a primitive form.

Theorem 3.5.13. [Strong Multiplicity One]

(a) For any N ≥ 1, Sk (Γ1 (N ))new has a basis of primitive forms.

(b) If f ∈ Sk (Γ1 (N ))new is an eigenvector for all T` with ` not dividing N , then f is a
scalar multiple of a primitive form.

(c) If f ∈ Sk (Γ1 (N )) and g ∈ Sk (Γ1 (M )) are primitive forms with a` (f ) = a` (g) for
all but finitely many primes `, then N = M and f = g.

We are not going to proof this theorem in the lecture. There is a nearly (but not
quite) complete proof in Diamond & Shuman and a different one in Miyake. For a full
proof see the paper of Atkin & Lehner, 1970.

Proposition 3.5.14. Let M divide N , and let f ∈ Sk (Γ1 (M )) be a primitive form.

Define Sk (Γ1 (N ))[f ] as the subspace of Sk (Γ1 (N )) spanned by all modular forms f (dz)
for some d dividing N/M . Then
M
Sk (Γ1 (N )) = Sk (Γ1 (N ))[f ].
f primitive of
level dividing N

Moreover, a form g ∈ Sk (Γ1 (N )) is an eigenvector for T` for all ` not dividing N if and
only if it lies in one of the subspaces Sk (Γ1 (N ))[f ].

Proof. We have seen that Sk (Γ1 (N ))new has a basis of primitive forms. By induction on
the number of divisors of N , the subspaces Sk (Γ1 (N ))[f ], f primitive of level dividing
N , span Sk (Γ1 (N )).
Suppose the sum is not a direct sum. Then there is a nontrivial linear relation
X
ci,j fi (di,j z) = 0
i,j

with scalars ci,j , primitive forms fi and factors di,j dividing N/level(fi ). We can suppose
without loss of generality that this relation has the least possible number of nonzero
ci,j ’s. Then all the fi ’s such that ci,j 6= 0 must have the same Tl eigenvalue for all l not
dividing N , since otherwise apllying Tl − λ for some λ would give a relation with fewer
terms. Hence all the fi ’s with ci,j 6= 0 for some j have some Tl eigenvalue for all l not

72
dividing N and thus they are equal by the Strong Multiplicity One theorem. So any
linear relation between vectors in
X
Sk (Γ1 (N ))[f ]
f primitive of
level dividing N

comes from a relation in Sk (Γ1 (N ))[f ] for a single f . Hence the sum is direct.
Note that this also shows that the vectors {f (dz) : d dividing N/level(f )} are linearly
independent. So the set

{f (dz) : f primitive of level N , d dividing N/level(f )}

is a basis. So it remains to show that any g ∈ Sk (Γ1 (N )) being an eigenvector for all T` ,
` not dividing N , is in Sk (Γ1 (N ))[f ] for some
Pmf . Suppose g is such an eigenvector for
all T` , ` not dividing N . We can write g = i=1 µi gi with gi ∈ Sk (Γ1 (N ))[fi ] for some
fi . if T` g = αg then
Xm
0 = (T` − α)(g) = µi (T` − α)(gi ).
i=1

Since vectors in subspaces Sk (Γ1 (N ))[f ] for distinct f ’s are linearly independent, all the
vectors (T` − α)(gi ) are zero. Since this holds for all ` not dividing N , Theorem 3.5.13
implies there is at most one nonzero µi , so g = µi gi ∈ Sk (Γ1 (N ))[fi ]. This finishes the
proof.

3.6 L-functions
3.6.1 Basic definitions
Definition 3.6.1. Let f ∈ Mk (Γ1 (N )) be a modular form with q-expansion f =
n
P
n≥0 an q . The L-function of f is the function in one complex variable s given by

∞
X
L(f, s) = an n−s .
n=0

Proposition 3.6.2.
1. If f ∈ Sk (Γ1 (N )), then L(f, s) converges absolutely for all s with <(s) > k/2 + 1.
2. If f ∈ Mk (Γ1 (N )) is not a cusp form, then L(f, s) converges absolutely for all s
with <(s) > k.
Proof. (1) By Proposition 3.3.14, we know that |an (f )| ≤ M nk/2 for some M > 0.
Hence, if <(s) > k2 + 1, then
 
X  X k
−s 
an n  ≤ M n 2 −<(s) < ∞.


 
n≥1 n≥1

73
 (2) For f ∈ Mk (Γ1 (N )), one can show that there exists M > 0 such that |an (f )| ≤
M nk for all n. The proof of the statement is analogous.

The L-functions of normalized eigenforms have a remarkable decomposition, the so-

called Euler product expansion. In fact, having this property characterizes normalized
eigenforms, as the following result shows:

Proposition 3.6.3. Let f ∈ Mk (Γ1 (N ), χ) be a modular form with q-expansion n≥0 an q n .

P
Then f is a normalized eigenform if and only if L(f, s) has an Euler product expansion
Y
−1
L(f, s) = 1 − ap (f )p−1 + χ(p)pk−1−2s .
p prime

Proof. By Proposition 3.2.34, we need to show that properties

i a1 (f ) = 1;

ii amn (f ) = am (f )an (f ) for all m, n comprime;

iii apr (f ) = ap (f )apr−1 (f ) − pk−1 χ(p)apr−2 (f ) for all primes p and all r ≥ 2.

are equivalent to L(f, s) having an Euler product. Suppose first that the conditions are
satisfied. Multiplying (iii) by tr and summing over all r ≥ 2 we see that (ii) is equivalent
to
∞
X ∞
X ∞
X
r r k−1 2
a (f )t = ap (f )t
pr a (f )t − p
pr χ(p)t apr (f )tr
r=2 r=1 r=0
!
X
⇔ apr (f )tr (1 − ap (f )t + χ(p)pk−1 t2 ) = a1 (f ) + ap (f )t(1 − a1 (f )).
r≥0

Since a1 (f ) = 1 by assumption, we get - by substituting t = p−s - the equality

∞
X
−1
apr (f )p−rs = 1 − ap (f )p−s + χ(p)pk−1−2s . (3.7)
r=0

Conversely, if this equality holds, then letting s → ∞ we get a1 (f ) = 1, and the other
implications can also be reversed to show that (3.7) is equivalent to conditions (i) and
(iii).
The Fundamental Theorem of Arithmetic implies that if g is any function of prime
powers, then
YX ∞ ∞ Y
X
r
g(p ) = g(pr ).
p r=0 n=1 pr ||n

Using this fact, it is easy to see that (3.7) and condition (ii) are equivalent to the
existence of the Euler product.

74
3.6.2 L-functions of cusp forms
We now focus on cusp forms. We will prove that for f ∈ Sk (Γ1 (N )), L(f, s) satisfies a
very important symmetry property, the so-called functional equation.
Definition 3.6.4. Define the Gamma-function
Z ∞
dt
Γ(s) = ts e−t .
0 t
Note 3.6.5. We have Γ(n) = n! for all n ≥ 1.
Definition 3.6.6. The completed L-function of f ∈ Sk (Γ1 (N )) is defined as

Λ(f, s) = (2π)−s Γ(s)L(f, s)

k
for <(s) > 2
+ 1.
Proposition 3.6.7. We have
Z ∞
Λ(f, s) = f (it)ts−1 dt.
0

Proof. First note that the integral converges, since

Z ∞  Z ∞
t−k/2+s−1 dt,
 
s−1 

 f (it)t dt 
0 0

k
which converges for <(s) > 2
+ 1. Now we compute
Z ∞ X
−1 s−1 −t
Λ(f, s) = (2π) t e dt an n−s
0 n≥1
Z ∞  s
X t dt
= an e−t .
n≥1 0 2πn t

t
Via the change of variables t 7→ 2πn , we get
X Z ∞  t s X Z ∞
−t dt
an e = an ts−1 2−2πnt dt
n≥1 0 2πn t n≥1 0
Z ∞ X !
= an e−2πnt ts−1 dt
0 n≥1
Z ∞
= f (it)ts−1 dt
0

as required.
Definition 3.6.8. Λ(f, s) is called the Mellin transform of f .

75
Definition 3.6.9. Define the operator WN = ik N 1−k/2 wN .

Lemma 3.6.10. WN is a self-adjoint operator, and it is idempotent: we have WN2 = id.

Proof. Easy check (exercise).

Lemma 3.6.11. Define

Sk (Γ1 (N ))± = {f ∈ Sk (Γ1 (N )) : WN f = ±f } .

We then have an orthogonal decomposition

Sk (Γ1 (N )) = Sk (Γ1 (N ))+ ⊕ Sk (Γ1 (N ))− .

Proof. If f ∈ Sk (Γ1 (N )), then we can write

1 1
f = (f + WN .f ) + (f − WN .f ),
2 2
and it is clear that 12 (f ± WN .f ) ∈ Sk (Γ1 (N ))± . Moreover, if f ∈ Sk (Γ1 (N ))+ , g ∈
Sk (Γ1 (N ))− , then

hf, gi = hWN .f, gi = hf, WN .gi = −hf, −gi,

so hf, gi = 0.

Theorem 3.6.12. Let f ∈ Sk (Γ1 (N ))± . Then the function Λ(f, s) extends to entire
function on C, which satisfies the functional equation
k
Λ(f, s) = ±N s− 2 Λ(f, k − s).
s
Proof. Let ΛN (s) = N 2 Λ(f, s), so we need to show that ΛN (s) = ±ΛN (k − s). By
making the change of variables t 7→ √tN , we get
Z ∞ Z ∞  
s
s−1 it
ΛN (s) = N 2 f (it)t dt = f √ ts−1 dt.
0 0 N
We now break the integral at 1:
Z ∞   Z 1   Z ∞  
it s−1 it s−1 it
f √ t dt = f √ t dt + f √ ts−1 dt.
0 N 0 N 1 N
R∞  
Note that 1 f √itN ts−1 dt converges to an entire function of s, since

√
 
it
f √ = O(e−2πt/ N
)
N

76
as t → ∞. For the other part, compute that
   
i k it
(WN .f ) √ =t f √
Ny N
Z 1   Z 1  
it s−1 i
⇒ f √ t dt = (WN .f ) √ ts−k−1 dt
0 N Nt
Z0 ∞  
it
= (WN .f ) √ tk−s−1 dt,
1 N
where the second equality follows from the change of variables t 7→ t−1 . Since WN .f =
±f , this integral converges to en antire function.
To obtain the functional equation, note that in total the integral is
Z ∞     
it s it k−s dt
ΛN (s) = f √ t + (WN .f ) √ t .
1 N N t
Since WN .f = ±f , this is ±ΛN (s), completing the proof of the functional equation.

3.6.3 Relation to elliptic curves

Let E/Q be an elliptic curve, which is an algebraic curve defined by an equation of the
form
E : y 2 = x3 + ax + b
with a, b ∈ Z satisfying ∆E = 4a3 − 27b2 6= 0. The conductor NE of E is an integer
whose prime divisors are the same as the prime divisors of ∆E (even though in general
we have ∆E 6= NE .)
One can then attach an L-function attached to E, which - up to finitely non-zero
factors - is defined as
Y
L(E, s) = (1 − ap (E)p−s + p1−2s )−1 .
p-NE

Here, ap (E) = 1+p−|E(Fp )|, where E(Fp ) denotes the number of points of the reduction
of E (mod p) over the finite field Fp . (We include here the point at ∞.)
Theorem 3.6.13 (Eichler–Shimura). Let f ∈ Sk (Γ1 (N )) be a normalized eigenform
whose Fourier coefficients are all integers. Then there exists an elliptic curve Ef defined
over Q such that
L(Ef , s) = L(f, s).
So does the converse of this theorem hold? In other words, given an elliptic curve over
Q of conductor NE , can we find a cusp form of level NE having the same L-function as
E?
Definition 3.6.14. An elliptic curve E is modular if there is a newform f ∈ Sk (Γ1 (NE ))
with ap (E) = ap (f ) for all p, i.e. L(E, s) = L(f, s).

77
 The following theorem, which relies on the work of Andrew Wiles on Fermet’s Last
Theorem, gives a postive answer to this question.

Theorem 3.6.15 (Wiles, Taylor–Wiles, Breuil–Conrad–Diamond–Taylor). All elliptic

curves over Q are modular.

Corollary 3.6.16. Let E/Q be an elliptic curve. Then L(E, s) has analytic continuation
to C, and it satisfies a functional equation relating L(E, s) and L(E, 2 − s).

78