Problems and Solutions

Diamond and Shurman:

A First Course in Modular Forms
Author:
Alex Youcis
Last Revised:
March 12, 2013
1
1.1
Exercise 1.1.1. Prove that SL
2
(Z) is generated by A =
_
1 1
0 1
_
and B =
_
0 1
1 0
_
.
Solution This amounts to the Euclidean algorithm. Namely, lets begin with an element
_
a b
c d
_
SL
2
(Z). We note
then that
B
m
_
a b
c d
_
=
_
a +cm b +dm
c d
_
(1)
for every m Z and
A
_
a b
c d
_
=
_
c d
a b
_
(2)
So, now, by the Euclidean algorithm we can write a = cq +r and so we see that if we let m = q we see by (1) that
B
m
_
a b
c d
_
=
_
a cq b dq
c d
_
=
_
r b dq
c d
_
So, we see then
SB
m
_
a b
c d
_
=
_
c d
r b dq
_
So, by the Euclidean algorithm we see that there exists r
1
and q
1
so that c = rq
1
+r
1
and so taking m
1
= q
1
in (1)
we see that
B
m1
SB
m
_
a b
c d
_
=
_
c rq
1
d rq
1
r d rq
_
=
_
r
1
d rq
1
r b dq
_
Continuing in this process, and using the fact that ad bc = 1 so that (a, c) = 1, we see that eventually we will arrive
at a matrix of the form
_
1
0
_
or
_
0
1
_
Since S moves between matrices of these two forms, we may assume without loss of generality that we have a matrix
of the rst form. But, now since this matrix must have determinant 1 its easy to see that the bottom left entry must
be 1. Thus, we see that, in fact, we may continue the process of applying powers of B and S to
_
a b
c d
_
to get a
matrix of the form
_
1 x
0 1
_
Now, since x is an unknown integer, we may up to relabeling, multiply this matrix by I = S
2
to get the matrix
_
1 x
0 1
_
with x Z. But, since this matrix is precisely B
x
we see it lands in the subgroup generated by A and B. Thus, putting
this all together, we see that we can apply combinations of powers of A and B to the left side of
_
a b
c d
_
to get into
the subgroup generated by A and B, and so
_
a b
c d
_
is in the generated subgroup. Since
_
a b
c d
_
was arbitrary the fact
that the subgroup generated by A and B is SL
2
(Z) follows.
Exercise 1.1.2. Let =
_
a b
c d
_
SL
2
(Z).
1
i) Prove that Im((z)) =
Im(z)
|cz +d|
2
ii) Show that if

SL
2
(Z) then (

)(z) = (

(z)) for all z H.

iii) Show that
d
dz
=
1
(cz +d)
2
.
Solution i) We merely make the following set of calculations
(z) =
az +b
cz +d
=
(az +b)(cz +d)
|cz +d|
2
=
ac|z|
2
+adz +bcz +bd
|cz +d|
2
= (ad bc)
Im(z)i
|cz +d|
2
+
(ad +bc)Re(z) +ac|z|
2
+bd
|cz +d|
2
=
Im(z)i
|cz +d|
2
+
(ad +bc)Re(z) +ac|z|
2
+bd
|cz +d|
2
Since the second summand of the last line is real, it follows that Im((z)) =
Im(z)
|cz +d|
2
as desired.
ii) Suppose that

=
_
a

_
. Then,
(

)(z) =
_
aa

+bc

ab

+bd

ca

+dc

cb

+dd

_
(z)
=
(aa

+bc

)z + (ab

+bd

)
(ca

+dc

) + (cb

+dd

)
and
(

(z) =
_
a

z +b

z +d

_
=
a
_
a

z +b

z +d

_
+b
c
_
a

z +b

z +d

_
+d
=
a(a

z +b

) +b(c

z +d

)
c(a

z +b

) +d(c

z +d

)
=
(aa

+bc

)z + (ab

+bd

)
(ca

+dc

)z + (cb

+dd

)
Since this was true for all z H it follows that

as maps, as desired.
iii) This is merely dierentiating z
az +b
cz +d
which gives the desired result by applying the quotient rule.
Exercise 1.1.3. i) Show that the set M
k
(SL
2
(Z)) is a vector space over C.
ii) If f and g are modular forms weight k and respectively then fg is a modular form of weight k + respectively.
iii) Show that S
k
(SL
2
(Z)) is a vector subspace of M
k
(SL
2
(Z)) and that S(SL
2
(Z)) is an ideal of M(SL
2
(Z)).
2
Solution i) Since a linear combination of holomorphic functions on H is holomorphic it suces to prove that a
linear combination of weakly modular functions of weight k are weakly modular of weight k, and that a linear
combination of bounded at i functions are bounded at i. The rst is clear for if f and g are weakly modular
of weight k, and if =
_
a b
c d
_
SL
2
(Z) then
(
1
f +
2
g)((z)) =
1
f((z)) +
2
g((z))
=
1
(cz +d)
k
f(z) +
2
(cz +d)
k
g(z)
= (cz +d)
k
(
1
f(z) +
2
g(z))
= (cz +d)
k
(
1
f +
2
g)(z)
To show the second claim let f and g be bounded at iwhich is equivalent to lim
Im(z)
f(z) and lim
Im(z)
g(z) exist-
ing. But, its easy to see that, in fact, lim
Im(z)
(
1
f +
2
g)(z) exists, and in particular, is equal to
1
lim
Im(z)
f(z)+

2
lim
Im(z)
g(z). Thus, with all of these properties veried it follows that M
k
(SL
2
(Z)) is a C-vector space as fol-
lows.
ii) Since the product of holomorphic functions is holomorphic, it suces to prove that fg is bounded at i and
that fg is weakly modular of weight k + . To see this rst fact it suces to note that, as is easy to verify,
lim
Im(z)
(fg)(z) =
_
lim
Im(z)
f(z)
__
lim
Im(z)
g(z)
_
, and in particular, that lim
Im(z)
(fg)(z) exists. The second
claim follows since if =
_
a b
c d
_
SL
2
(Z) then
(fg)((z) = f((z))g((z))
= (cz +d)
k
f(z)(cz +d)

g(z)
= (cz +d)
k+
f(z)g(z)
= (cz +d)
k+
(fg)(z)
for every z H from which the conclusion follows.
iii) Consider the map F = lim
Im(z)
: M
k
(SL
2
(Z)) C, dened in the obvious way. Now, noting that F |
M
k
(SL2(Z))
is a C-linear map and that S
k
(SL
2
(Z)) is the kernel of this map we see that S
k
(SL
2
(Z)) is a C-subspace of
M
k
(SL
2
(Z)). Now, since S(SL
2
(Z)) is the kernel of F, and F is easily seen to be a C-algebra map, it follows
that S(SL
2
(Z)) is an ideal in M(SL
2
(Z)). Moreover, since F is obviously surjective (being C-linear and nonzero)
it follows from the rst isomorphism theorem that M(SL
2
(Z))/S(SL
2
(Z))

= C and so, in fact, S(SL
2
(Z)) is a
maximal ideal of M(SL
2
(Z)).
Exercise 1.1.4. Let k 3 be an integer and let L

= Z
2
{(0, 0)}.
i) Show that the series

(c,d)L
(max{|c|, |d|})
k
converges by considering partial sums over expanding squares.
ii) Fix positive numbers A and B and let
= {z H : | Re(z)| A, Im(z) B}
Prove that there is a constant C > 0 such that |z +| > C max{1, ||} for all z and R.
iii) Use parts i) and ii) to prove that the series dening G
k
(z) converges absolutely and uniformly for z . Conclude
that G
k
is holomorphic on H.
iv) Show that for SL
2
(Z), right multiplication by denes a bijection L

.
3
v) Use the calculation from iii) to show that G
k
is bounded on . From the text and part iv), G
k
is weakly modular
so in particular G
k
(z + 1) = G
k
(z). Show that therefore G
k
is bounded as Im(z) .
Solution i) As intuition shows us by considering expanding squares about the origin dened by {(x, y) R
2
:
max{|x|, |y|} = m} we see that

(c,d)L

(max{|c|, |d|})
k
=

m=1
A
m
m
k
where A
m
= #{(x, y) Z
2
: max{|x|, |y|} = m}. That said, its easy to see that A
m
8(m + 1) (by considering
the 8 half edges of the square which each contain m+ 1 points ) so that

m=1
A
m
m
k
8

m=1
(m+ 1)
m
k
8

m=1
(m+ 1)
m
k
<
and so the conclusion follows.
ii) As the book suggests, we break this into four cases. Let z = x +iy
Case 1) Suppose that || < 1. Then,
|z +| =
_
B
2
+ ( A)
2
B = Bmax{1, ||}.
Case 2) Suppose next that 1 || 3A and y > A. Note then that
|z +| =
_
((x )
2
+y
2
A >
||
3
=
1
3
max{1, |delta|}.
Case 3) Now, if B > A then Im(z) B > A for all z and so we may assume without loss of generality
that B A. So, assume in particular that 1 | 3A and B Im(z) A. Then, since the region
[1, 3A] {z : B Im(z) A} is compact the mapping (z, )
|z +

assumes a minimum, call it

M. Then, by denition
|z +| M|| = M max{1, ||}
Case 4) Lastly, if | > 3A then
|z +| || A
2
3
| =
2
3
max{1, ||}
where we used the reverse triangle inequality.
So, if we take C =
1
2
min{
1
3
, M, B} the problem follows from these four cases.
iii) Note that on if we dene f
(c,d)
(z) = (cz +d)
k
then we have that
|f
(c,d)
(z)|
C
k
max{|c|, |d|}
k
where C is the constant found in i). Then, by applying a double series version of the Weierstrass M-test we
may conclude that G
k
(z) converges uniformly and absolutely on . The fact that G
k
(z) then converges to a
holomorphic function on is simple complex analysis. Indeed, take any z
0
H and choose a ball B

(z
0
) around
that point. Since B

(z
0
) is contained in some region we know that G
k
(z) converges uniformly and absolutely
on B

(z
0
), and moreover since G
k
(z) is extendable to B

(z
0
) which is compact we know that G
k
is bounded on
B

(z
0
). Then, to see that G
k
(z) is holomorphic on B

(z
0
) it suces to note that if T is any triangle in B

(z
0
)
then
_
T
G
k
(z) =

(c,d)L

_
T
1
(cz +d)
k
=

(c,d)L

0 = 0
where we could interchange the sum and series by uniform convergence, and the integral over the inside is zero
by Cauchys theorem (note that the roots of cz + d = 0 arent in H). Since T was arbitrary Moreras theorem
implies that G
k
(z) is holomorphic in B

(z
0
) and so, in particular, holomorphic at z
0
. Since z
0
was arbitrary the
conclusion follows.
4
iv) We merely note that since SL
2
(Z) GL
2
(Z) that is an invertible linear map Z
2
Z
2
and so, in particular,
is a bijection L

.
v) We know that if M denotes the constant C
k

(c,d)L

1
max{|c|, |d|}
k
found in iii) then we know from iii) that
|G
k
(z)| M for all z . Now, since G
k
(z) is 1-periodic that G
k
(z) being bounded as Im(z) with z ,
with being with respect to A = B = 1, implies that G
k
(z) is boundeded as Im(z) with z H.
Exercise 1.1.5. Establish the following formulae:
(1) cot(z) =
1
z
+

n=1
_
1
z n
+
1
z +n
_
(2) cot(z) = i 2i

n=0
e
2inz
Solution To prove (1) start with the well-known formula
sin(z) = z

n=1
_
1
z
n
n
2
_
Then, we see that taking the logarithmic derivative gives

cos(z)
sin(z)
=
1
z
+

n=1
2z
z
2
n
2
Now, by denition the left hand side of the above equation is cot(z) and since
2z
z
2
n
2
=
1
z n
+
1
z +n
the equation
(1) follows.
To establish (2) we make the following series of computations
cot(z) =
cos(z)
sin(z)
=
e
iz
+e
iz
2
e
iz
e
iz
2i
= i
e
iz
+e
iz
e
iz
e
iz
= i
e
2iz
+ 1
e
2iz
1
= i
_
1 +
2
e
2iz
1
_
= i
_
1
2
e
2iz
1
_
= i 2i
1
e
2iz
1
= i 2i

n=0
e
2inz
The last step where we expanded
1
1 e
2iz
as a series was valid for if z h then

e
2iz

= e
Re(2iz)
= e
2i Im(z)
< 1
since Im(z) > 0.
5
Exercise 1.1.6. This exercise obtains formula (1.2) without using the cotangent. Let f(z) =

dZ
1
(z +d)
k
for k 2
and z H. Since f is holomorphic (by the method of exercise 1.1.4) and Z-periodic and since lim
Im(z)
f(z) = 0 there
is a Fourier expansion f(z) =

m=1
a
m
q
m
= g(q) as in the section where q = e
2iz
and
a
m
=
1
2it
_

g(q)
q
m+1
dq
is a path integral once counterclockwise over a circle about 0 in the punctured disc D

.
i) Show that
a
m
=
_
1+iy
z=0+iy
f(z)e
2imz
dz =
_
+iy
z=+iy
z
k
e
2imz
dz
for all y > 0.
ii) Let g
m
(z) = z
k
e
2imz
, a meromorphic function on C with a singularity only at the origin. Show that
2i Res(g
m
(z), 0) =
(2i)
k
(k 1)!
m
k1
iii) Establish (1.2) by integrating g
m
(z) clockwise about a large rectangular path and applying the Residue Theorem.
Argue that the integral along the top side goes to a
m
and the integral along the other three sides go to 0.
iv) Let h : R C be a function such that the integral
_

|h(x)|dx is nite and the sum

dZ
h(z + d) converges
absolutely and uniformly on compact subsets and is innitely dierentiable. Then the Poisson summation formula
says that

dZ
h(x +d) =

mZ

h(m)e
2imx
where

h is the Fourier transform of h,

h(x) =
_

h(t)e
2ixt
dt
We will not prove this, but the idea is that the left side sum symmetries h to a function not period 1 and the right
side sum is the Fourier series of the left side since the mth Fourier coecient is
_
1
0

dZ
h(t +d)e
2imt
=

h(m).
Letting h(x) =
1
(x +iy)
k
where y > 0 , show that h meets the requirements for Poisson summation. Show that

h(m) = e
2my
a
m
with a
m
from above for m > 0 and that

h(m) = 0 for m 0. Establish formula (1.2) again,
this time as a special case of Poisson summation.
Solution i) We begin by noting by the homotopy version of Cauchys theorem that a
m
is equal no matter what
radius the circle is. So, parameterize the circle as (t) = re
2it
for some r (0, 1). Note then that by making
this parameterization we have that
a
m
=
1
2i
_

g(z)
z
m+1
dz
=
1
2i
_
1
0
g
_
re
2it
_
(re
2it
)
m+1
_
2ire
2it
_
dt
=
_
1
0
g
_
exp
_
2i
_
log(r)
2i
+t
___
r
m
e
2imt
dt
= r
m
_
1
0
f
_
t i
log(r)
2
_
e
2imt
dt
6
Now, note that this makes sense in terms of domains since Im
_
t i
log(r)
2
_
is
log(r)
2
> 0 since r < 1. Since
this was independent of r we have that
a
m
= lim
r1

_
r
m
_
1
0
f
_
t i
log(r)
2
_
e
2imt
dt
_
= lim
r1

_
1
0
f
_
t i
log(r)
2
_
e
2imt
dt
Now, noting that
_
t i
log(r)
2
: (t, r) [0, 1] [
1
2
, 1)
_
is contained in a bounded region of H we have by the
continuity of f on H that f is uniformly bounded on this region. Thus, we may apply the Bounded Convergence
Theorem to conclude that
a
m
= lim
r1

_
1
0
f
_
t i
log(r)
2
_
e
2imt
dt
=
_
1
0
lim
r1

f
_
t i
log(r)
2
_
e
2imt
dt
=
_
1
0
f
_
lim
r1

_
t i
log(r)
2
__
e
2imt
dt
=
_
1
0
f(t)e
2imt
dt
Now, making the substitution t t +iy for any y > 0 we see that
a
m
=
_
1
0
f (t +iy) e
2im(t+iy)
dt =
_
1+iy
iy
f(z)e
2imz
dz
where we used the fact that t t +iy was a parameterization of a path from iy to 1 +iy (the linear path). This
proves the rst equality. The second is achieved by making the following calculations
a
m
=
_
1+iy
iy
f(z)e
2imz
dz
=
_
1+iy
iy

dZ
e
2imz
(z +d)
k
dz
=

dZ
_
1+iy
iy
e
2imz
(z +d)
k
dz
where the interchanging of the limits was valid because the convergence of the series was uniform and f is bounded
as z i. Now, making the change of variables z z +d and using the 2i periodicity of exp gives us that
a
m
=

dZ
_
1+iy
iy
e
2imz
(z +d)
k
dz
=

d Z
_
d+1+iy
d+iy
e
2imz
z
k
dz
=
_
+iy
+iy
z
k
e
2imz
dz
as desired.
ii) To nd the residue at 0 we merely nd the Laurent series of g
m
(z) at 0. Indeed,
g
m
(z) = z
k
e
2imz
= z
k

n=0
(2im)
n
z
n
n!
=

n=0
(2im)
n
z
nk
n!
7
Thus, the residue of g
m
(z) at 0 is the coecient of z
1
which occurs when n k = 1 or n = k 1. Thus, we see
that the residue is
(2im)
k1
(k 1)!
. Thus, 2i Res(g
m
(z), 0) is
(2i)
k
(k 1)!
m
k1
as desired.
iii) Let
R,T
be the symmetric rectangle of width 2R and height 2T centered at the origin. Consider then that by the
Residue Theorem and ii) we have that
(2i)
k
(k 1)!
m
k1
=
_

R,T
z
k
e
2imz
dz
= f(R, T) +g
1
(R, T) +g
2
(R, T) +g
3
(R, T)
(1)
where
f(R, T) =
_
R+iT
R+iT
z
k
e
2iz
dz
is the integral over the top portion of
R,T
,
g
1
(R, T) = ie
2imR
_
T
T
(R +it)
k
e
2mt
dt
is the integration over the rightmost vertical part of
R,T
(after making the standard parameterization),
g
2
(R, T) = ie
2imR
_
T
T
(it R)
k
e
2mt
dt
is the integration over the leftmost vertical part of
R,T
(after making the standard parameterization),and
g
3
(R, T) = e
2mT
_
R
R
(t iT)
k
e
2imt
dt
is the integration over the bottom portion of
R,T
(after making the standard parameterization). Now, we make
the claim that lim
R
g
i
(R, T) = 0 for i = 1, 2 and lim
R
g
3
(R, T) is well-dened (non-innite) function of T. To see
that the rst of these assertions is true we merely note that
|g
1
(R, T)|
_
T
T
|R +it|
k
e
2mt
dt

1
R
k
_
T
T
e
2imt
dt
And since, for xed T, this last term approaches 0 as R it follows that lim
R
g
1
(R, T) = 0 for each xed
T as desired. A nearly identical calculation shows that lim
R
g
2
(R, T) = 0 for each xed T. Now, the fact that
lim
R
g
3
(R, T) is a well-dened function of T follows by the mere, and obvious fact, that
_

(t iT)
k
e
2imt
is convergent (because the exponential term is modulus 1). We last make the observation that by i) we have that,
independent of T, we have that lim
R
f(R, T) = a
m
. Thus, putting this all together and letting R tend to innity
inf (1) gives
(2i)
k
(k 1)!
m
k1
= a
m
+e
2mT
_

(t iT)
k
e
2imt
(2)
Now, since
8

e
2mT
_

(t iT)
k
e
2imt

e
2mT
_

|t iT|
k
dt
= e
2mT
_

1
(t
2
+T
2
)
k
2
dt
e
2mT
_

dt
(t
2
+ 1)
k
2
and because k is large enough the integral in the last term is convergent, and since e
2mT
0 as T we see
in particular that
lim
T
_
e
2mT
_

(t iT)
k
e
2mt
dt
_
= 0
So, letting T tend to innity in (2) gives us
(2i)
k
(k 1)!
m
k1
= a
m
as desired.
iv) Since |h(x)| =
1
(x
2
+y
2
)
k
2
and so |h(x)|
1
|x|
k
(asymptotically equivalent to) which converges because k 2.
Moreover, the fact that

dZ
h(z +d) converges uniformly and absolutely on compact subsets is clear from previous
exercise. Lastly,

dZ
h(z + d) is innitely dierentiable since it is holomorphic which is true precisely because it
converges uniformly on compact subsets of C. Thus, Poisson summation applies. Noting that

h(m) =
_

h(x +iy)e
2imx
dx
= e
2my
_

h(x +iy)e
2im(x+iy)
dx
= e
2my
_
+iy
+iy
h(z)e
2imz
dz
= e
2my
a
m
for m > 0 where the last step follows from i). Since

dZ
h(z +d) = f(z) we see that
f(z) =

dZ
a
m
e
2im(x+iy)
=

dZ
a
m
q
m
Exercise 1.1.7. The Bernoulli numbers B
k
are dened by the formal power series expansion
t
e
t
1
=

k=0
B
k
t
k
k!
Thus they are calculable in succession by matching coecients in the power series identity
t = (e
t
1)

k=0
B
k
t
k
k!
=

n=1
_
n1

k=0
_
n
k
_
B
k
_
t
n
n!
(i.e. the nth parenthesized sum is 1 if n = 1 and 0 otherwise) and they are rational. Since the expression
t
e
t
1
+
t
2
=
t
2

e
t
+ 1
e
t
1
9
is even it follows that B
1
=
1
2
and B
k
= 0 for all other k. The Bernoulli numbers will be motivated, discussed, and
generalized in Chapter 4.
i) Show that B
2
=
1
6
, B
4
=
1
30
, and B
6
=
1
42
.
ii) Use the expression for cot(z) from the section to show
1 2

k=1
(2k)z
2k
= cot(z) = iz +

k=0
(2iz)
k
k!
.
Use these to show that for k 2 even, the Riemann zeta function satises
2(k) =
(2i)
k
k!
B
k
so in particular (2) =

2
6
, (4) =

4
90
, and (6) =

6
945
. Also, this shows that the normalized Eisenstein series of
weight K
E
k
(z) =
G
k
(z)
2(k)
= 1
2k
B
k

n=1

k1
(n)q
n
has rational coecients with a common denominator.
iii) Equation coecients in the relation E
8
(z) = E
4
(z)
2
to establish formula (1.3).
iv) Show that a
0
= 0 and a
1
= (2)
12
in the Fourier expansion of the discriminant function from the text.
Solution i) We know that B
0
= 1, B
1
=
1
2
, and B
3
= 0. Then, since
0 =
2

k=0
_
3
k
_
B
k
= B
0
+ 3B
1
+ 3B
2
we get that B
2
=
1
6
. The other values of B
k
are found using the exact same recursion.
ii)
Exercise 1.1.8. Recall that
3
denotes the complex cube root of unity e
2i
3
. Show that
_
0 1
1 0
_
(
3
) =
3
+ 1
so that by periodicity g
2
__
0 1
1 0
_
(
3
)
_
= g
2
(
3
). Show that by modularity g
2
__
0 1
1 0
_
(
3
)
_
=
4
3
g
2
(
3
) and
therefore g
2
(
3
) = 0. Conclude that g
3
(
3
) = 0 and j(
3
) = 0. Argue similarly to show that g
3
(i) = 0, g
2
(i) = 0, and
j(i) = 1728.
1.2
Exercise 1.2.1. f
Exercise 1.2.2. f
Exercise 1.2.3. i) Let p be a prime and let e be a positive integer. Show that | SL
2
(Z/p
e
Z)| = p
3e
p
3e2
.
ii) Show that | SL
2
(Z/NZ)| = N
3

p|N
_
1
1
p
2
_
, so this is also the index [SL
2
(Z) : (N)].
10
iii) Show that the map
1
(N) Z/NZ given by
_
a b
c d
_
b mod N surjects and has kernel (N).
iv) Show that the map
0
(N) (Z/NZ)

given by
_
a b
c d
_
d mod N surjects and has kernel
1
(N).
v) Show that [SL
2
(Z) :
0
(N)] = N

p|N
_
1 +
1
p
_
.
Solution i) We induct on e. So, to begin we nd | SL
2
(Z/pZ)| by brute force. Indeed, we are looking for
_
a b
c d
_

Mat
2
(Z/pZ) such that ad bc = 1. Suppose rst that d = 0, then we are looking for (a, b, c) (Z/pZ)
3
such that
bc = 1, clearly there are p(p 1) such pairs since a can be arbitrary, c is determined by b and b can be any unit.
Now, if d = 0 then d is a unit and so there are p 1 choices for d. Moreover, once we have found b and c we know
that a is determined. Moreover b and c can be anything for then a is just d
1
(bc + 1). Thus, there are (p 1)p
2
choices. Thus, overall there are (p 1)p
2
+p(p 1) = p
3
p choices as desired.
Now, suppose that we have proven that | SL
2
(Z/p
e
Z)| = p
3e
p
3e2
and consider SL
2
(Z/p
e+1
Z). There is a natural
group map SL
2
(Z/p
e+1
Z) SL
2
(Z/p
e
Z) given by reducing entries modulo p
e
. Now, something in the kernel of
this map is of the form
_
mp
e
+ 1 p
e
kp
e
np
e
+ 1
_
. Moreover, we know that the determinant of this matrix must be 1
modulo p
e
but the determinant modulo p
e
is just p
e
(m+ n) + 1. Thus, we need that p | m+ n and that k and
can be arbitrary. Now, since we are working modulo p
e+1
the p
e
in front of m, n, and k means that we are really
only looking for m, n, and k modulo p. So, the fact that k and can be arbitrary gives us p
2
choices (p for each).
To nd the number of m and n is equivalent to nding the number of m and n in Z/pZ that sum to 0 in Z/pZ.
To nd this note that we have the obvious group map (Z/pZ)
2
Z/pZ Z/pZ : (m, n) m + n for which we
are trying to nd the cardinality of the kernel. But, this map is obviously surjective and so the kernel has size
p
2
p
= p. Thus, we see that there are p choices for (m, n). Thus, overall there are p
2
p = p
3
choices. Thus, the
kernel of the map SL
2
(Z/p
e+1
Z) SL
2
(Z/p
e
Z) has cardinality p
3
and so the rst isomorphism theorem implies
that
| SL
2
(Z/p
e+1
Z)| = p
3
| SL
2
(Z/p
e
Z)| = p
3
_
p
3e
p
3e2
_
= p
3(e+1)
p
3(e+1)2
and so the induction is complete.
ii) Lets rst recall that the formation of SL
2
(R) for a ring R is functorial, so that if R

= S as rings then SL
2
(R)

=
SL
2
(S). Next, note that
_
(a
1
, a
2
) (b
1
, b
2
)
(c
1
, c
2
) (d
1
, d
2
)
_

__
a
1
b
1
c
1
d
1
_
,
_
a
2
b
2
c
2
d
2
__
is a group isomorphism SL
2
(R S) SL
2
(R) SL
2
(S). Combining these results and the Chinese Remainder
Theorem we know that if N = p
e1
1
p
en
n
then
SL
2
(Z/NZ)

=
n

i=1
SL
2
(Z/p
ei
i
Z)
Taking orders of both sides gives
| SL
2
(Z/NZ)| =
n

i=1
p
3ei
i
_
1
1
p
2
i
_
= (p
e1
1
p
en
n
)
3

p|N
_
1
1
p
2
_
= N
3

p|N
_
1
1
p
2
_
as desired.
iii) To see that this is a group map we merely note that if f is the described map and
_
a b
c d
_
,
_
a

_

1
(N)
then
11
f
__
a b
c d
__
a

__
= ab

+bc

mod N b +b

mod N
since c c

1 mod N. But, its obvious that

b +b

mod N = f
__
a b
c d
__
+f
__
a

__
and so f really is a homomorphism. Its surjective for if b {0, , N 1} then
_
1 b
0 1
_
is in (n) and the image
is precisely b. Finally, the kernel of the map is
__
a b
c d
_

1
(N) : b 0 mod N
_
But, by denition, since
_
a b
c d
_

1
(N) we know that a d 1 mod N and c 0 mod N. Thus, with
b 0 mod N we have that
_
a b
c d
_

_
1 0
0 1
_
mod N
and so
_
a b
c d
_
(N) and the reverse inclusion is just as trivial.
iv) To see that the map
0
(N) (Z/NZ)

, call this map g, is a group map we merely note that if

_
a b
c d
_
,
_
a

0
(N) then
g
__
a b
c d
__
a

__
= cb

+dd mod N = dd

since c 0 mod N by the denition that

_
a b
c d
_

0
(N). But, clearly dd

is precisely
g
__
a b
c d
__
g
__
a

__
and so it follows that g really is a group map. This map surjects since if d (Z/NZ)

then (d, N) = 1 and so

there exists ba such that ad +bN = 1. We see then that =
_
a b
N d
_

0
(N) and g () = d as desired.
v) We have the following chain of subgroups (N)
1
(N)
0
(N) SL
2
(Z). Now, from Lagranges theorem we
know that
[SL
2
(Z) :
0
(N)] =
[SL
2
(Z) : (N)]
[
0
(N) :
1
(N)][
1
(N) : (N)]
Now, from iii) we know that [
1
(N) : (N)] = N, and from iv) we know that if N = p
e1
1
p
en
n
= (N)
= N
(N)
N
= N
N

i=1
p
ei1
(p 1)
n

i=1
p
ei
i
= N
n

i=1
_
1
1
p
i
_
12
Thus,
[ SL
2
(Z) :
0
(N)] =
[SL
2
(Z) : (N)]
[
0
(N) :
1
(N)][
1
(N) : (N)]
=
N
3
n

i=1
_
1
1
p
2
i
_
N
2
n

i=1
_
1
1
p
i
_
= N
n

i=1
_
1 +
1
p
i
_
= N

p|N
_
1 +
1
p
_
as desired.
Exercise 1.2.4.
Exercise 1.2.5. If SL
2
(Z) is a congruence subgroup and SL
2
(Z), then
1
is a congruence subgroup of
SL
2
(Z).
Solution We merely note that since is a congruence subgroup there exists some N Z such that (N) .
Then, (N)
1

1
, but (N) SL
2
(Z) and so (N)
1
= (N). Thus, (N)
1
and so
1
is a
congruence subgroup as desired.
Exercise 1.2.6. This exercise proves Proposition 1.2.4. Let be a congruence subgroup of SL
2
(Z), thus containing
(N) for some N, and suppose that the function f : H C is holomorphic and weight-k invariant under . Suppose
also that in the Fourier expansion f(z) =

n=0
a
n
q
n
N
the coecients for n > 0 satisfy |a
n
| Cn
r
for some positive
constants C and r.
i) Show that for any z = x +iy H,
|f(z)| |a
0
| +C

n=1
n
r
e
2ny
N
(1)
Changing n to a nonnegative real variable t, show that the continuous version g(t) = t
r
e
2ty
N
of the summand
increases monotonically on the interval
_
0,
rN
2y
_
and then decreases monotonically on
_
rN
2y
,
_
. Using this,
represent all but two terms of the sum in (1) as unit-wide boxes under the graph of g and consider the missing
terms individually to establish the estimate (where in this exercise C
0
and C can denote dierent constants in
dierent places)
|f(z)| C
0
+C
__

0
g(t) dt +
1
y
r
_
After a change of variables the integral takes the form
C
y
r+1
_

0
t
r
e
t
dt. This last integral is a gamma function
integral, to be introduced formally in Section 4.4, but at any rate it converges at both ends and is independent of
y. In sum,
13
|f(z)| C
0
+
C
y
r
as y .
ii) For every SL
2
(Z), the transformed function (f[]
k
)(z) is holomorphic and weight-k invariant under
1

and therefore has a Laurent expansion

(f[]
k
)(z) =

nZ
a

n
q
n
N
To show that the Laurent series truncates from the left to a power series it suces to show that
lim
q
N
0
(f[]
k
)(z)q
N
= 0
If xes then this is immediate from the Fourier series of f itself. Otherwise show that the transformed
function (f[]
k
)(z) = (cz +d)
k
f((z)) satises
lim
q
N
0
|(f[]
k
])(z)q
N
| C lim
q
N
0
(y
rk
|q
N
|)
Recalling that q
N
= e
2i(x+y)
N
, show that y = C log(|q
N
|), and use the fact that polynomials dominate logarithms
to complete the proof.
Solution i) The equation (1) follows immediately by applying the triangle inequality, the fact that |a
n
| Cn
r
, and
that

exp
_
2inz
N
_

= exp
_
Re
_
2inz
N
__
= exp
_
2ny
N
_
.
To establish the claim about g(t) we dierentiate to nd that g

(t) =
2yt
r
N
e
2ty
N
+ rt
r1
e
2ty
N
, or g

(t) =
e
2ty
N
t
r1
_
2yt
N
+r
_
. Since the multiplier is always positive, this is negative precisely when r
2yt
N
or
Nr
2y
t.
Thus, we see that g is decreasing precisely when t
_
Nr
2y
,
_
, and thus g must be increasing on
_
0,
Nr
2y
_
as
desired.
Using the fact that the sign change occurs at
Nr
2y
we may use Riemann sums to estimate that

n=1
g(n) =

n=0
g(n)
_

0
g(t) dt +
N
r
(2ye)
r
1
y
r
(2)
(the rst equality being true because g(0) = 0). In particular, choosing the partition {[n1, n] : n N} and using
Riemann left hand Riemann sums for intervals lying in
_
0,
Nr
2y
_
, using right hand Riemann sums on intervals lying
inside
_
Nr
2y
,
_
, and bounding the sum on the partition containing
Nr
2y
by g
_
Nr
2y
_
=
N
r
(2ye)
r
1
y
r
(since this is
the maximum g takes, and the interval is measure 1) we get (2). Thus, we see that

n=1
g(n) C

__

0
g(t) +
1
y
r
_
where C

= max
_
1,
N
r
(2ye)
r
_
. Thus, we see that
|f(z)| |a
0
| +C

n=1
g(n) |a
0
| +CC

__

0
g(t) dt +
1
y
r
_
Thus, taking C
0
= |a
0
| and C = CC

(where, as was stated, the C means dierent things on the right and left
hand sides) we get our desired inequality.
14
ii) We have the estimation
|f[]
k
(z)q
N
| = |j(, z)|
k
|f((z))q
N
|j(, z)|
k
_
C
0
+
C
y
r
|j(, z)
2r
|
_
q
N
where we used the previous part and made use of the fact that Im((z)) =
Im(z)
|j(, z)|
2
. Now, letting q
N
0 is
equivalent to letting y . Now, from this its easy to see that |j(, z)| is asymptotically C

y for some constant

C

(really its the c entry of ). Thus, we see that

lim
q
N
0
|f[]
k
(z)q
N
| |j(, z)|
k
_
C
0
+
C
y
r
|j(, z)|
2r
_
q
N
= lim
q
N
0
C
y
r
|j(, z)|
2rk
q
N
= lim
q
N
0
Dy
rk
q
N
where the rst equality is via the (obvious) fact that lim
q
N
0
|j(, z)|
k
C
0
q
N
= 0, and the second fact follows
from the asymptotic for |j(, z)| with D = CC

. Now, by denition we have that q

N
= exp
_
2i(x+iy)
N
_
and so
|q
N
| = e
2iy
N
and so |y| = Alog |q
N
| where A is a constant. Thus,
| lim
q
N
0
Dy
rk
q
N
| = lim
q
N
0
|y|
rk
|q
N
| = A lim
q
N
0
log |q
N
|
rk
|q
N
| = 0
because |q
N
dominates log |q
N
|
rk
.
15