On the Shimura Correspondence

Tommy Tang

Abstract
In which we define notions analagous to well-known ideas for modular forms of half-integer weight modular
forms that Shimura first described in his paper On Modular Forms of Half Integral Weight in 1973. We
will then motivate and prove his papers central theorem, giving a correspondence between modular forms
of half integral weight and modular forms of even weight, as well as a property of certain Hecke operators
for these forms. Furthermore, we will discuss some work of Niwa and Kohnen that extend Shimuras main
result. This paper does not contain original work, but rather attempts to merge the approaches by Koblitz
and Shimura into a more readily understood and lucidly motivated exposition.

0. Introduction
Since the 1800s there has been a rich, growing theory surrounding the modular forms of integer weight.
Mathematicians like Hardy and Hecke have long seen the value in a similar theory for modular forms of
half-integer weight but often thought such a theory was improbable. Hecke developed a theory of Euler
product for modular forms, but claimed that a similar theory for half-integer weight forms was impossible.
It was not until Shimuras paper in 1973 in which he defined a correcting automorphy factor did a rich theory
for half-integer weight modular forms begin to develop.

Theorem 0: M2k (0 (4)) consists exactly of homogeneous polynomials in 4 , F of degree k.

Theorem 0:Mk (0 (4)) = {f C[2 , F ] : f (z) =

ai bi F ci , bi /2 + ci = k/2}

I. Basic Definitions
For the remainder of these notes, we will use to denote SL2 (Z) and (N ) to denote the congruence
subgroup of level N , that is, the matrices congruent to the identity modulo N . In addition, we will take the
branch of the square root so that the argument is in (/2, /2].
k
It is tempting to try to simply define a modular form of weight for a congruence subgroup to be a function
2
f that satisfies
f (z) = (cz + d)k/2 f (z)
(1)
for all 0 (N ). Unfortunately,
 a bit of computation

shows that
 this is impossible for any congruence
1+N
N
1
0
subgroup of level N > 2. Let =
, =
(N ). Applying (1) to
N
1N
N 1
f (z) f (z)
f (z)
=

,
f (z)
f (z)
f (z)
for some f Mk/2 (0 ), we see that the following equation, raised to the k-th power, must hold for z H:
p
p

(N 2 2N )z + 1 N = N z/(N z + 1) + 1 N N z + 1
(2)
On the RHS, both expressions inside the radicals are expressions in the lower half-plane; thus the radical of
each is in the fourth quadrant and their product will also be in the lower half-plane. On the other hand, the
expression on the LHS is clearly in the upper-half plane, since N 2 2N > 0 whenever N > 2. Thus, for all
odd k the equation (2) will be off by a factor of 1.
Instead, we will adopt a new definition for our automorphy factor:
c

Definition 1.1:
j(, z) := (z)(z) =
1
cz + d,
(3)
d
d


where dc is the Legendre symbol, defined for positive odd d as the usual quadratic residue Legendre symbol,
c
then extended to all positive odd d by multiplicativity, and finally for negative odd d, when c > 0 as
,
d
c
when c < 0 as
and simply 1 whenever c = 0). Further, d = 1 when d 1 (mod 4) and d = i when
d
1

d 1 (mod 4). In other words, 2d = (1)(d1)/2 .

As with integer-weight modular forms, we would like to have some definition for []k/2 for arbitrary
GL+
2 (Q). But our automorphy factor above is only defined for 0 (4). So it is convenient to instead
choose
from a larger group G that specifies an automorphy factor. Hence we define
Definition 1.2:

cz + d
2
G := {(, (z)) : GL+
},
2 (Q), satisfies (z) =
det
a = (, 1 (z)), b = (, 2 (z)) G, ab = (, 1 (z)2 (z)).

(4)

It is easy to check that G is a group under the multiplication defined in (4). Now we can define, for
= (, (z)) G, define for a function f on H,
f (z)|[]k/2 := f (z)(z)k

Definition 1.3:

(5)

Analagously, for 0 0 (4) we have

0 = {(, j(, z), 0 }.

(6)

0
0 (4), for f acting on H,
As with (5), for
f (z)|[
]k/2 := f (z)j(, z)k

(7)

It will be useful to let G1 denote the subgroup of G that is the set of pairs (, (z)) with SL2 (Z).
0 (4). Thus
Remark: The map L that takes = (, j(, z)) is in fact an isomorphism from 0 (4)
(6) and (7) arise naturally from Definition 1.2.
Now let us discuss the cusps of 0 (4) for half-integer weight forms. Recall that there are exactly 3 equivalence
classes of cusps for 0 (4) : 0, , and 1
2 .
0 : s = s}
0s = {

Proposition 1.4: Let s = 1 be any cusp of 0 , and let = (, (z)) G. Then

can be described by


1 h
0 1

s = {(
, t)j }jZ,tT ={1,i} .
0 1


1 h
Moreover, the element (
, t) G1 depends only on the 0 equivalence class of s.
0 1


1 j
Proof: Note that
are precisely the matrices in that fix , and (z) must be a constant
0 1 jZ
0
0
function t T = {1, i}. Now any

 0 (4) must have finite index in , and so the matrices in that
1 jh
fix are precisely the matrices
, with perhaps a factor of 1 allowed. We fix h > 0 and see
0 1 jZ
that if 1 0 , then we must include a sign. Otherwise we do not.
0 = 1 {( 1 h , t)j }jZ,tT ={1,i} . The additional sign simply helps
It is easy to see, then that
s
0 1
us generalize regardless of whether 1 is in 0 .

To show that h and t depend only on the 0 -equivalence class of s requires a bit of computation. Suppose
0
is replaced by some

= (0 , (z)) G1 with s = 01 . Then, 01 G1 fixes . Therefore, it must be

1 l
of the form (
, t1 ). Now, we have
0 1


 ( 
 j ) 
 
1
l
1
h
1 l
1
0 0 01
0 1
0
1
01
s )( ) =
s
= ( ) (
, t1

,t
, t1 .
0 1
0 1
0 1




1 l
1 h
0s 01 =
0s 1 . Now, generalizing to any other
But (
, t1 ) commutes with (
, t) so 0
0 1
0 1
0s = 1
0s , we see with
cusp in the same equivalence class, s1 = 1 s, where 0 and noting that
1
0
1
0 1

similar computation that (

)s1 (
) = s as needed. 
With the proposition above, we can now describe more clearly what it means to be meromorphic, holomorphic, or vanishing at a particular cusp s = 1 of 0 . Suppose f is a meromorphic function
1
0 1 , we
on H that is invariant under [
]k/2 for all 0 . Let g = f |[ 1 ]k/2

 . Then

1 h
have g|[
1 ]k/2 = f |[
1 ]k/2 = g. Hence, g is invariant under [(
, t)]k/2 . In other words,
0 1
g(z) = tk g(z + h). But since t4 = 1, we see that e2izr/4h g(z), where r {0, 1, 2, 3}, is invariant under
the transformation z z + h. We can thus obtain a Fourier series expansion
X
g(z) =
an e(2iz(n+r/4))/h .
n

We say that f is meromorphic at the cusp s if an = 0 for all but finitely many negative n and holomorphic
if an = 0 for all n < 0. If f is holomorphic, then we can define f (s) := limzi g(z). Proposition 1.4 above
shows that these notions depend only on the 0 equivalence class of s.
We now formalize the notions above with some definitions. Let k be any integer, and let 0 0 (4) be a
0.
subgroup of finite index. Let f be a meromorphic function on H that is invariant under [
]k/2 for all
Definition 1.5: f is a modular function of weight k/2 if it is meromorphic at every cusp of 0 .
Definition 1.6: A modular function f of weight k/2 is a modular form if it is holomorphic on H and at
0 is denoted
every cusp of 0 . The space of all modular forms of weight k/2 for a congruence subgroup
0

Mk/2 ( ).
Definition 1.7: A modular form of weight k/2 is a cusp form if it vanishes at every cusp. The space of all
0 is denoted Sk/2 (
0 ).
cusp forms of weight k/2 for a congruence subgroup
Recall that for k Z, we can formally define an important subspace Mk () Mk (0 (N )) by writing


a b
Mk (, N ) = {f Mk (1 (N ) : f |[]k = (d)f for =
0 (N )}
c d
As with modular forms of integer weight, there is a theory surrounding the modular forms of character for
half integral weight. In particular we denote


a b

Definition 1.8
Mk/2 (0 (N ), ) = {f Mk/2 (1 (N )) : =
0 (n), f |[
]k/2 = (d)f }.
(8)
c d
3

Proposition 1.9: Suppose f1 Mk1 /2 (0 , 1 ) and f2 Mk2 /2 (0 , 2 ). Then f1 f2 M(k1 +k2 )/2 (0 , 1 2 ).
Proof: We have from definitions that
f1 (1 z)j(1 , z)k1 /2 = 1 (d1 )f1 (z)
f2 (2 z)j(2 , z)k2 /2 = 2 (d2 )f2 (z).
Then, we have
f1 f2 |[
](k1 +k2 )/2 = f1 f2 (z)(j(z, z))(k1 +k2 )/2 = f1 (z)j(, z)k1 /2 f2 (z)j(, z)k2 /2
= 1 (d)f1 (z)2 (d)f2 (z) = 1 2 (d)f1 f2 (z).
That the condition at the cusps is preserved is clear, so we are done. 


 
1 0
Remark: Since =
, 1 G acts as an identity, so since f |[
]k/2 = f , whenever is odd, it is
0 1
0 (N ), ) = 0. Similarly, when = 1, we simply have Mk/2 (
0 (N ), ) = Mk/2 (
0 (n)).
easy to see that Mk/2 (

Our definitions above preserve the cusp conditions we expect for forms of integer weight. However, when
k/2 Z is odd, it is easy to see that [
]k/2 differs from [k/2 by 1 (d), where 1 (n) = (1)(n1)/2 . Hence,
Proposition 1.10: If 4|N , and k/2 Z, then
k/2

k/2

0 (N ), ) = Mk/2 (N, ).
Mk/2 (
1

0 (N ), ) = Sk/2 (N, ).
Sk/2 (
1

Armed now with our knowledge of Mk/2 from the discussion above, we now characterize the space of weight
k/2 modular forms, with the following theorem:
X
0 (4)) = {f C[, F ] : f (z) =
Theorem 1.11: Mk/2 (
ai bi F ci , bi /4 + ci = k/4}.
Remark: This is analagous to the theorem, regarding M2k (0 (4)), characterizing the space as homogeneous
polynomials in 4 , F . This theorem does much the same, but allows for a fractional power of 4 .
Proof: . This part will follow from our lemma above about multiplicativity in Mk/2 , since we have already
verified that and F are modular forms of weight 1/2 and 2, respectively.
0 (4)) for some odd
. Theorem 0 already gives us this result whenever k/2 Z. Now, suppose f Mk/2 (
k. Then f is a modular form of weight (k + 1)/2 Z. Thus, applying Theorem 0 and separating out the
pure F term we can write
f = aF (k+1)/4 + 2 P (, F ),
where a C is a constant that is zero if 4|k + 1 and P (, F ) is a polynomial that is a sum of terms ai bi F ci
such that bi /4 + ci = (k 1)/2. Then
0 (4)).
f P (, F ) = aF (k+1)/4 / Mk/2 (
But vanishes at 21 , whereas F does not, so this form is holomorphic at all cusps only if a = 0 = f =
P (, F ) as desired. 

 0 (4)) = 1 + [k/4].
Corollary: Dim Mk/2 (

II. Hecke Operators

While the action of Hecke operators Tn on a modular form f (z) is often defined at first as the sum of functions
of modular points n1 F corresponding to f , we are more interested in the definition of Hecke operators via
double cosets. We first quickly go through the definitions for integer weight Hecke operators, before moving
onto the analagous definitions for half-integral weight forms.
Suppose 1 , 2 are subgroups of some group G, and G. Then the double coset 1 2 is defined
Definition 2.1

1 2 := {1 2 : 1 1 , 2 2 }

In general, 1 2 is a union of right cosets of the form 1 2 . Now, we are primarily concerned with double
cosets of the form 0 0 , where 0 denotes, as before, a congruence subgroup of . For 0 and some
00
0
1 0
fixed GL+
. Denote d := [0 : 00 ] < , and write 0 = dj=1 00 j . It is
2 (Q), define =
0
0
d
0
easy to check that then = j=1 j is a disjoint union of cosets.
Keeping the notation above, we have
Definition 2.2:

f |[0 0 ]k :=

f |[j ]k ,

0 j 0 0

and subsequently we can define, for n Z

Definition 2.3

Tn f := n(k/2)1

f |[0 0 ]k ,

0 0 n
n
where the sum is over all double cosets 0 0 in n, where
 denotes the subgroup of integer-coefficient
1 0
matrices with determinant n that are congruent to
(mod N ).
0 n

Now, before we proceed to the analogous definitions for modular forms of half-integer weight, we must make
a few remarks on the above.
Proposition 2.4: If 0 1 0 = 0 2 0 , then f |[0 1 0 ]k = f |[0 2 0 ]k ; that is, Definition 2.2 depends only
on the double coset itself, and not our choice of from within the double coset. Furthermore, our sum in
Definition 2.2 does not depend on our choice of right coset representatives j .
Proof: We will first prove the assertion that the sum in Definition 2.2 does not depend on our choice
of representatives {j }. Suppose we replace these representatives with new representatives {j0 }, where
{j0 } are chosen from 0 . Then there is a set {j } that can be chosen from 00 such that for each j,
j0 = j j . Now, 00 1 0 , so we can write j = 1 j0 , where j0 0 . Then since f Mk (0 ),
f |[j0 ] = f |[j j ]k = f |[j ]k . Then since this shows we can choose any {aj } such that dj=1 0 j = 0 0 ,
it is clear that our choice of does not matter.
Proposition 2.5: If f (z) Mk (0 ), then f (z)|[0 0 ] Mk (0 ).

Proof: Suppose that f Mk (0 ). Then, for 0 , we have (f |[0 0 ]k )|[]k =

since right multiplication by would just rearrange the right cosets 0 j .

f |[j ]k = f |[0 0 ]k ,

Recall that Hecke operators are typically defined for modular forms of the congruence subgroup 1 (N ).
n
Also recall that
 we let
 denote the subgroup of integer-coefficient matrices with determinant n that are
1 0
congruent to
(mod N ). We will now prove a useful propositions about the structure of this group
0 n
relative to 1 (N ).


S
a b
n
Proposition 2.6: Suppose n is relatively prime to N . Then =
1 (N )a
, where the union
0 d
disjoint
 1

a
0
is taken over all ad = n, (a, N ) = 1, for each a we define a is some fixed matrix congruent to
0
a
(mod N ), and for each selection of d we take b = 0, 1, ..., d 1.


S
a b
n
Proof: First note that it is clear
1 (N )a
(routine computation confirms the union is
0 d
disjoint
indeed disjoint).
 0



a b0
a b
n

can
be
written
in
the
form

, where 0 1 (N ),
Now, we show that any =
a
c0 d0
0 d
and a is chosenas inthe Proposition. Select g, h relative
prime
such that ga0 + hc0 = 0. Then complete


e f

the matrix =
. Now, is of the form
with determinant n. We can multiply by
g h
0




1 j
a b
if necessary, and thus assume WLOG that is in the form
with a > 0, ad = n, 0 b < d.
0 1
0 d


a b
Then = 1
. Taking this equation modulor N , we see that
0 d

  1


1
a

a b

,
0 n
0
a
0 d




a b
a b
1
0
0
0
that is, is in the form a for some 1 (N ). Thus = a
1 (N )a
as desired.
0 d
0 d
Now, recall that in the integer-weight case, we defined the Hecke operator Tp for prime p and used these as
building blocks for all Hecke operators, since Tmn = Tm Tn . With the following results, we will see that in
the half-integer weight case, we will want to only consider Tp2 as the building blocks for our maps.
We first elaborate on the isomorphism L mentioned in section I. Let GL+
2 (Q), G = (, ). Now, we
will construct a homomorphism that depends on . For 0 1 0 , we can write
L(1 ) = L() 1 (1, t ()),
defining t : 1 T = {1}. It can be checked with a bit of computation that with this construction, t in fact only depends on and not on our choice of for .
0 1
0 .
Proposition 2.7: For 0 , with our notation above, restrict 0 . Then L(Ker(t)) =
0
k
0
0 0

Furthermore, if L(Ker(t)) 6= (that is, if t is nontrivial), then for all f Mk/2 ( ), we have f |[ ]k/2 = 0.

 0 1
0 ,
Proof: The proof of our first assertion is fairly straightforward. It suffices to show L(Ker(t))
0
1
is of the form 2 for 2 0 .
since inclusion in the other direction is clear. Suppose that 1
1
Applying the projection (the inverse of L), we have 1 = 2 . Then L(1 1 ) = L(2 ), so t (1 )
satisfies L(2 ) = L(1 ) 1 (1, t (1 )) = 2 =
1 1 (1, t (1 )) = 1 = 1 (1, t (1 ) = t (1 ) = 1
as claimed.
Now, choose {j } such that
0 = j (0 1 0 )j
is a disjoint union of right cosets. Similarly, choose {s } to be a set of representaties for 0 1 0 modulo
Ker(t). Then, we have
0 = j L(s Ker(t)s )

j
0
1
)
s j .
= j,s (
0
0 = s,j
0 s j is a disjoint union. We have L(s 1 ) = L(s ) 1 (1, t (s )) = L(s ) =
Thus,
0
0 as a disjoint union to fit the form
L(s 1 ) (1, t (s )1 ). Thus, combining with our expression for
of Definition 2.2, we find that
X
X
0
0 ]k/2 = det()k/41 (
f |[
t (s )k )(
f |[ L(j )]k/2 )
s

which is 0 when tk is nontrivial.

Corollary 2.8: If we suppose m is not a perfect square, and naively attempt to use Definition 2.3 (with
0 = 1 (N )), then we obtain Tm f = 0.
Proof: As in Proposition 2.6, we can write m = n0 n21 , and letting n0 , n1 vary obtain our double coset
representaties. We will consider especially the case n0 = m, n1 = 
1; the general
case is computed analagously.

1 0
From Proposition 2.7, then, it suffices to show that with =
, our homomorphism t is indeed
0 m
nontrivial. But straightforward omputation shows that


 
d
a b
, =
1 1 (N ) 1 (N ),
t () =
c d
n
and thus is trivial only if m is a perfect square.
1 (N )),
Now, we will define, for prime p and modular form f Mk/2 (
def

1 (N )p2
1 (N )]k/2 , where p2 =
Tp2 f (z) = pk/22 f |[



1
0




0
,
p
.
p2
p2

1
= 1 (N )
0


0
1 (N )
p2

Note that this not quite the immediate analogue of Tp2 acting on f Mk (1 (N )), since
 1

p
0
1 (N )pp 1 (N ), where p is some fixed matrix congruent to
modulo N . As before, we will want
0
p
to be able to compute the coefficients of a modular form after a Hecke operator transformation.

Proposition 2.9: Suppose N is a multiple, a Dirichlet

N , and p relatively prime to N .
P character modulo
0 (N ), ), we have
Let k = 2 + 1 be an odd integer. Then, if f (z) = n=0 an q n Mk/2 (
Tp2 f (z) =

bn q n ,

n=0


where bn = ap2 n + (p)

(1) n
p

p1 an + (p2 )pk2 an/p2 , taking an/p2 = 0 if p2 does not divide n.

2

Proof: As shown in Proposition 2.6, we know that that p is a disjoint union of the right cosets of 1 (N ),
with three classes of representatives (namely, as we vary the entries in the main diagonal):


1 b
b =
, for 0 b < p2
0 p2



h
, for 0 < b < p
p
 2

p 0
= p2
.
0 1

h = p

p
0

In other words, we have


1
1 (N )
0

2

p[
1
p1
[
0

(N
)
=

(N
)

1 (N )h 1 (N ).
1
1
b
2
p

b=0

h=1

First, as in Proposition 2.7, we want to put each of these in a format

that

 resembles Definition 2.2. That
0
0 1
is, we want to write each of these representatives in the form
with , 0 1 (N ), and then
0 p2
compute



X

1 0
k/22
f (z)|[p2 ]k/2 , p2 =
, p .
Tp2 = p
0 p2

With some computation, we have:




1 0
1 b
0 p2
0 1



1
0
1 0
p
h = p
pN s 1
0 p2
N s
 2

 2
p t
1 0
p d
= p2
N d
0 p2
N
b =


1
0

0
1

(1)

h
r

t
1

(2)
(3)

For each h , we fix for each h integers r, s such that pr + N sh = 1, and similarly for we choose integers
t, d such that p2 d + N t = 1.
Remark: Our matrices for h , differ slightly from Shimuras approach. He searches for representatives of
0 (N )0 (N ) and later accounts in the lift for multiplication by a character. Our approach here will yield
the same result, but compute the contribution of p , p2 more directly.

We now examine the effects of each of these classes on the expression for Tp2 . First let us consider the
contribution of the b to our sum. This contribution can be written via (1) as
k/22

2
pX
1

h=0


1
f (z)|
0

 


1
0
, p
0
p2

2

 
pX
1 
z + b k
b
k/22
( p)
,1
=p
f
1
p2
k/2

b=0

1
= 2
p
2
pX
1

2
pX
1

b=0


f

z+b
p2


.

e[2in](z+b)/p is 0 when p2 6 |n and simply equal to p2 e2inz/p when it does. Hence the contribution
b=0
P
of the h is n ap2 n e2inz and we see that this is the first term in our expression for bn inthe statement

p2 t
of the proposition. Now we consider the contribution of . Using (3) and writing 0 =
and
N d
 2

p d t
=
as in (3), we write this as
N 1

But







1 0
p
(,
j(,
z))
pk/22 f (z)|[
p2 0 p2 ]k/2 = pk/22 (p2 )f (z)| ( 0 , j( 0 , z))
,
0 p2
k/2
 2



p
0
= pk/22 (p2 )f (z)|
, 1/ p
0 1
k/2
X
k2
2
2inp2 z
=p
(p )
an e
,
n

hence we see that contributes the third term in our expression for bn .
Now we consider the h . With similar computation we have that the contribution from the h is equal to
pk/22 (p)

p1
X

f (z)|

h=1


p
0





 p1  
h
1 k X h
h
h
k/22
, 1
=
p
(p)
p
f (z + ).
p
p
p
p
p
p
k/2
h=1

Now we can compute the inner sum:

p1  
X
h
h=1

f (z +

p1  

 
X
X
h 2inh/p
h
X n
an e2inz
)=
e
= p p
an e2inz ,
p
p
p
n=0
n=0
h=1

where we have replaced nh by h and used the value p p for the Gauss sum. Thus we see that our sum was
equal to
pk/2 2(p)

1
p

 

 X
 
n
1
n
an e2inz = p1 (p)
an e2inz
p
p
p
n=0
n=0

X
k+1
p
p

and we see that the contribution of the h to the n-th coefficient in the q-series expansion of Tp2 f (z) is the
middle term, as needed. 

Using the proposition above, it is quite easy to prove the multiplicativity property and build the the general
Hecke operator Tn2 . However, for the purposes of understanding Shimuras lift, only squares of primes are
needed, so we will not have those proposition.

III. Main Theorem

We now exhibit the Shimuras main theorem on a correspondence between cusp forms of weight (2k + 1)/2
and modular forms of weight 2k as well as its accompanying result on pertinent operators.
Theorem (Shimuras Correspondence):
Let k = 2 + 1 3 be odd, 4|N , and a Dirichlet character
P
0 (N ), ) be an eigenform for every Hecke operator in the
modulo N . Suppose f (z) = n=1 an q n Sk/2 (
set {Tp2 } as p ranges over p|N not dividing the conductor of . Then define
g(z) =

bm q m ,

m=1

where the coefficients bm are encoded thus:

X
m=1

bm ms =

1
.
1 p ps + (p)2 pk22s

(1)

Then g is a modular form of weight k 1 of 2 for level N 0 divisible by the conductor of 2 . Furthermore,
g is a cusp form when k 5.
Proof: We will show that g necessarily satisfies the conditions of Weils Converse Theorem.
P
P
Lemma 1 (Weils Converse Theorem): Let F (z) = n=1 cn q n , D(s) = n=1 cn ns . Let be a set of prime
numbers that intersects non-trivially with every arithmetic progression {a + jb}
j=0 for (a, b) = 1. Then F
is a modular form of weight k of level M for the character if the following are satisfied:
(i) D(s) is absolutely convergent in some upper plane Re(s).
(ii) For every primitive
character mod r where r {1} and (r, M ) = 1, the function R(s, ) =
P
(2)s (s) n=1 (n)cn ns can be continued to a holomorphic function in the whole splane, and is
bounded in every vertical strip 1 < Res < 2 .
P
(iii) There exists a Dirichlet series D0 (s) = n=1 c0n ns , absolutely convergent in some half-plane
0
P Re(s)0 >s
2
s(k/2)
s
such that for all characters considered in , R(k s, ) = C (r M )
(2) (s) n=1 (n)cn n ,
where C = (M )(r)g()2 /r.
Furthermore, F is a cusp form if D(s) is absolutely convergent for some s, with Re(s) < k.
- End Lemma In order to prove that g satisfies conditions (i), (ii), (iii), we will require several key lemmas:
0
0 (4), then its coefficients
Lemma 2: If f (z) is a cusp form of weight k/2 for the congruence subgroup
k/4
satisfy an = O(n ).

10

Proof: Recall
that the intersection of 0 and and the stabilizer of infinity is essentially generated by the


1 h
matrix
for some h > 0. Thus f has an expansion in powers of e2iz/h - that is, we can write
0 1
P
f (z) = f (x + iy) = n=0 an q n/h . Then, observe that for each n and fixed y we have
Z
1 h
(2)
f (x + iy)e2in(x+iy)/h dx.
an =
h 0
Now, consider the function g(z) = y k/4 f (z), where z = x + iy. Since Im z = Im(z)|j(, z)|4 , g is 0 invariant. Moreover, at each cusp s = 1 , we have that G = f |[1 ]k/2 is holomorphic and satisfies
G 0 as q 0 (or equivalently, as y ). But g(1 (z)) = G(q)y k/4 , so from 0 -invariance g is bounded
over all x. Thus we have |f (x + iy)| B y k/4 for all y and some constant B independent of x. Applying
this bound to (2), we see that for each n,
|an | C 0 y k/4 e2ny/h .
Thus letting y = 1/n gives the bound we desire.
- End LemmaLemma 3 : Let N be a positive integer, and N 0 some positive divisor. Let


a b
SL2 (Z)| b 0 (mod N 0 ), c 0 (mod N )}
={
c d
and

1
= {
0


mN 0
|m Z}
1

be the subgroup of 1 (N ) that fixes . Furthermore, let WN be the set of duples {(c, d)}, c 0 such that
c and d are relatively prime, N |c, and d = 1 when c = 0. Then there is a 1-1 correspondence between the
right cosets of \ and WN .

  0 0
a b
a b
Proof: Consider two matrices
,
. Then b b0 0 (mod N )0 and since ad bc = 1 =
c d
c d
b (c)1 (mod d), we also have b b0 (mod d). Then, since in general we have


 

1 mN 0
x y
x + cmN 0 y + dmN 0
=
,
0
1
c d
c
d
we must be able to select some m such that b + dmN 0 = b0 . Furthermore, from (c, d) = 1 we see that we
can similarly find a m satisfying the above to also satisfy a + cmN 0 = a0 . Thus simply mapping right coset
representatives to their bottom row (c, d) gives us our 1-1 correspondence as needed.
- End LemmaLemma 4 : Let be a nonnegative integer, A a positive integer, and a primitive character modulo A. Let
X
H (s, z, ) = s (s)y s
(n)(mAz + n) |mAz + n|2s ,
m,n

where the sum is taken over all (0, 0) 6= (m, n) Z2 , and z H. Suppose that either > 0 or A > 1
strictly. Then the series is absolutely convergent for Re(s) > (/2) + 1, and H can be continued to the
entire s-plane as an entire function satisfying

H ( + 1 s, z, ) = (1) G()A3s2 z H (s, 1/Az, ),

11

PA
where G() = k=1 (k)e2ik/A In addition, if f and g are a cusp form and modular form, respectively,
of weight k/2 for level B 0 (mod 4)A, and D is a fundamental domain for (B)\H, then, writing =
(k )/2 0, the integral
Z
f (z)
g (z)y /2 H (s, z, )y 2 dxdy

is absolutely convergent for any s, and as a function in s bounded in every vertical strip 1 < Re(s) < 2 .
Proof: Remark : This is a type of Eisenstein-Epstein function, whose behavior is well-known.
We will first prove the equality
Z Z
2
2
(uz + v) et|uz+v| /y e2i[ur+vs] dudv = (1) t1 (r sz) e|rsz| /yt ,

where r, s, t, u, v are real variables and t > 0. For = 0, this equality follows quickly. With this in mind,
integration by parts gives the result for > 0 as well. Now, substituting u + A1 p, v + A1 q for u and v
respectively and applying the Poisson summation formula gives
X
X
2
2
2
(mz + n) et|mz+n| /A y = (A) t1
(mz + n) e|mz+n| /yt e2i(qppn)/A .
(m,n)(p,q)

(m,z)Z2

(mod A)

Denote our function on the LHS as (t, z, (p, q)). Then our equation above gives
X
(t, z, (p, q)) = (A) t 1
e2i(qapb)/A (A2 t1 , z, (a, b)).
(a,b)

Now, put (t, z, (p, q)) =

apply (3) to obtain

PA

k=1

(k)(t, z, k(p, q)). Then, assuming (p, q) 6 (0, 0) (mod A) when A > 1 we
X

(t1 , z, (p, q)) = (A) t+1 G()

(a,b)

Furthermore, for any =


a
c

(3)

(mod A)

(qa
pb)(A2 t, z, (a, b)).

(4)

(mod A)


b
SL2 (Z), we have [1]
d
(t, (z), (p, q))(cz + d) = (t, z, (p, q)).

Since, as in the statement of the Lemma we have either > 0 or A > 1, the term (m, n) = (0, 0) does not
appear in (4). Then, at least formally, we have in some strip [2]:
Z
0

(t, z, (p, q))ts1 dt = A2s s y s (s)

A
X

(k)

k=1

X
(m,n)k(p,q)

(mz+n) |mz+n|2s .
(mod A)

The RHS of this equation is absolutely convergent for all Re(2s) > + 2. Thus, for fixed p, q, we from our
first equation relating the two sums that [3]
(
M ect
if t > 1
|(t, z, (pq))|
0
M 0 t1 ec /t if t < 1
with positive constants M, M 0 , c, c0 that depend only on z and (p, q). Therefore, we can split the integral
in (5) into two integrals over (0,1) and (1, ), and hence from the above we see that the integral i in fact
12

(5)

convergent for all of s. Thus the equation (5) holds whenever the sum on the RHS is convergent, and so
holds for all Re(2s) > + 2. Letting (p, q) = (0, 1) yields
Z
A2s H (s, z, ) =
(t, z, (0, 1))ts1 dt,
0

so substituting in + 1 s for s, we obtain

Z
2(+1s)
A
H ( + 1 s, z, ) =
(t, z, (0, 1))ts dt
0
Z
=
(t1 z, (0, 1))ts2 dt
0

= (A) G()
(a,b)

(m)

(A2 t, z, (a, b))ts1 dt

(mod A)

= (A) G() s y s (s)

(m)(mz
+ n) |mz + n|2s

(m,n)Z2

= (1) G()A+s z H (s, 1/Az, phi),

as claimed in the Lemma.
We now wish to prove the second assertion, that the integral over the fundamental domain as written in the
Lemma, is absolutely convergent over all s. We majorize the following function:
F (u, z) =

0
X

|mz + n| eu|mz+n| ,

where the sum is taken over all (m, n) 6= (0, 0) and 0 < u R. Pick to be the smallest integer /2.
Then the sum of all the terms with m = 0 gives
2

n en

n=1

k eku = eu P (eu )/(1 eu )+1

k=1

for some polynomial P . Thus, this sum is O(eu ) as u and O(u1 ) as u 0 (since (1 eu )/u 1
as u 0. Therefore, we see that

n en

M u1 eu/2

(6)

n=1

where the constant M depends only on . Next, for m 6= 0 and z = x + iy fixed, for each n there is some
positive integer k such that k 1 |mx + n| < k. Thus summing up the terms where m 6= 0, we have
2

X
X

|mz + n| eu|mz+n| 4

m=1 n=

(m2 y 2 + k 2 ) eu(m

2 2

y +(k1)2 )

m=1 k=1

 

X
X
2 2 X
2

(m2 y 2 )v eum y
k 2v eu(k1) .
v
m=1 v=0
k=1

13

Applying (6) to the last sum, we see that our expression above is bounded by
X
2
M0
y 2 (uy 2 )1 euy /2 (1 + 22v M2v uv1 eu/2 ),
+v=

where M 0 depends only on . Thus, for any c > 0 we have that when t > 1 and y > c > 0,
F (t/A2 y, z) Ly +1 ekt/y
where L, h are positive and depend only on , A, and c. Then it follow that
|(t, z, (p, q)) F (t/A2 y, z) Ly +1 eht/y
when t > 1, y > c. Similarly, from (4) we see that
0

|(t1 , z, (p, q)| L0 t+1 y +1 eh t/y

if t > 1, y > c and again for constants L0 , h0 depending

 only on , A, c. Now, select f, g, and D as in our
a j bj
Lemma. Then we can select a finite set {j =
} each in SL2 (Z) such that
cj dj
D j j (U ), U = {x + iy|y > c, 0 x B}.
Denote fj = f |[j ]k/2 , gj = g|[j ]. Then
Z
XZ
/2
2
|f gy H (s, z, )|y dxdy
|fj gj |y /2 |cj z + dj | |H (s, j (z))|y 2 dxdy.
D

Now, for =Re (s), we can write

A2 |cj z + dj | |H (s, j (z), )|

|(t, j (z), (0, 1))(cj z + dj ) |t1 dt

Z0
=

|(t, z, (0, 1)j )|t1 dt

|(t, z, (0, 1)j )|t1 dt +

=
1

+1 ht/y 1

dt + L

If 1, we can replace
is bounded by

with

R
0

|(t1 , z, (0, 1)j )|t1 dt

Z
R

y +1 eh t/y t dt.

, and t1 with 1 if < 1. Thus the first integral of our last expression
Lh ()y +1+ if 1

and by
(L/h)y +2 if < 1.
Similarly, depending on whether or > , we have that the second integral is bounded by
L0 h01 ( + 1)y ++2
or by
(L0 /h0 )y +2 .
14

Thus, in every vertical strip 1 < Re(s) < 2 , we have

Z
XZ
|f gy /2 H (s, z, )|y 2 dxdy L00
|fj gj y |y 2 dxdy
D

with positive constants L00 , that depend on 1 , 2 . Now fj gj is a cusp form of level B, so |fj (z)gj (z)| =
O(e2i/B ) as y . Thus the last two integrals are convergent, as needed.
- End LemmaLemma 5 : Let
g Gk/2
 (1 (N )), and let
 N = M K be a positive integer. For every v, write gv = g|[v ]k/2 ,
1
0
where v =
, (M vz + 1)1/2 . Now suppose that either M or K is divisible by 4. Then gv (z +
Mv 1
P
n/k
K) = gV (z), and hence gv must have a q-series expansion of the form gv (z) =
. If we
n=0 v,n q
furthermore
impose
the
condition
that
g
is
a
modular
form
of
character,
that
is,
g

M
(N,
),
then
for
k/2


a
b
=
0 (M ), d positive and relatively prime to N , then
Mc d


M c k
k/2
d gw (z),
(M cz + d
)g(z) = (d)
d
where w is some integer with dw c (mod K).


1 Nv
K
Proof: Write =
. We claim that 1 (N ), and we have
M N v 2 1 + N v


1 K
v L
= v .
0 1


K
Note that the automorphy factor of these two sides will differ by 1
1+N v 1+N v . Now, 4|N implies that our


K
1+N v = 1. Then, either 4|M implies 4K|N v = N v + 1 1 (mod 4K) = 1
= 1, or
1+N v 1+N v
 



0
4
K
K
= 1+N
where 4K 0 |N v, so again this is
similarly if 4|K, then by multiplicity we have 1+N
v
v
1+N v
simply 1 and we have our equality as claimed. Then, applying our equality to g, we see that gv (z+K) = gv (z)
as in the Lemma.
For our second assertion, write dv = c Kc0 , a0 = a bM w. Then

  0


a
b
a
b
1
0
=
.
Mc d
N c0 d
Mw 1
 0




0
But Ndc = M Kc
= Md c since d > 0, Kc0 c (mod d), so using this to compute the automorphy
d
factor of the lift and applying the first assertion, we are done.
- End LemmaLemma 6 : Let m be an integer, r an odd prime, and a primitive character modulo r. Then
X

r X
r  
X
u

u,v (mod r) t=1 k=1

4uv1 (mod r)

(t)e2i[(m

vkmkt+uk2 )/r]

15

= r r3/2 (2m).

(7)

Proof: For fixed m, r, t, u, let x be an integer such that m + t 2ux (mod r). Such an x must exist since r
is prime and the condition above ensures that u 6 0 (mod r). Then we have
r
X

e2i[(uk

kmkt)/r]

k=1

r
X

r
X

e2i[k(uk2ux)/r] =

k=1

r
X

e2i[u(k

2kx+x2 )/r] 2i[ux2 ]

k=1

e2i[u(kx)

/r] 2i[ux2 /r]

u
r

k=1

where the last equality comes from the well-known relation

Pp

 
k
p

k=1

r1/2 r e2i[ux

/r]

e2i[k/p] = p p1/2 [Davenport, p. 13].

Thus the LHS of (7) is equal to

r
XX

(t)r1/2 r e2i[(m

vux2 )/r]

u,v t=1

Now, ux2 4uv ux2 v(2ux)2 v (m + t)2 (mod r), so by letting v vary from 1 to r 1, and simply
summing over all u satisfying 4uv 1 (mod r), this becomes
r X
r1
X

r1/2 r

(t)e2i[v(m

(m+t)2 )/r]

v=1 t=1

Moreover, (r) = 0, so we can extend the upper limit of the inner sum to range from r 1 to r. Now, nonvanishing terms only occur for m2 (m + t)2 0 (mod r) m2 = (m + t)2 (mod r) t 0 or t 2m
as needed.
- End Lemma Now, from the Lemma 2 we have a(n) = O(nk/4 ). Then
(k + 2)/2. Now, we can write our equation (1) as

X
m=1

bm m

am2 ms is absolutely convergent for Re(s) >

1 1s X
)n
)(
an2 ns ),
=(
(n)(
n
n=1
n=1

(8)

P
P
1s
the L-series n=1 (n)( 1
is absolutely convergent for Res > , so
bm ms is absolutely conn )n
vergent for Re(s) > (k + 2)/2. Thus the first condition is satisfied. Note further that if (ii) and (iii) are
satisfied, then whenever k 5 we have (k + 2)/2 < k 1 so from the last statement g is in fact a cusp form.
We now proceed onto the proof of properties (ii) and (iii).


P
n
, where = (k 1)/2.
Suppose f (z) =
Sk/2 (N, ), where 4|N . Define 0 = (m) 1
n=1 an q

Furthermore let r be either 1 or some odd prime non-divisor of N , and some primitive character modulo
r.
Now, as per the rewrite of our generating series for bm in (8), we have
D(s, ) =

X
n=1

(m)bm ms = (

(n)0 (n)n1s )(

n=1

)(n)an2 ns .

n=1

We will first show that R(s, ) constructed from D satisfies conditions (ii).

16

(9)

 =
Write h(z, )

1
2

m=

v 2im2 z

(m)m
e
, where v is chosen from {0, 1} such that (1) = (1)v . Then

f (z)h(z)
= 21

an (m)mv e[2i](nm

)x 2(n+m2 )y

m,n

Writing z = x + iy, we compute

Z

f (z)h(z)dx
= 21

v 2(n+m2 )y

an (m)m e

e[2i](nm

)x

dx,

n,m

where the terms being summed up are nonzero only when n m2 = 0.

X
2
am2 (m)mv e4m y , so for certain values of s, we can write

Thus our sum is equal to

m=1

Z
0

Z

f hdx y s1 dy = (4)s (s)

(m)am2 mv2s .

(10)

m=1

is a cusp form of weight (k + 2v + 1)/2 [4], so we have that |f h| =

To see where this holds, note that f h
2y
O(e
) when y and |f h| = O(y (k+2v+1)/4 ) when y 0. Thus (10) holds for Re(s) > 1 + (k + 2v +
1)/4. Put
 


1 n
= 0 (N r2 ), =
|n Z ,
0 1
Y = {x + iy|0 < x 1, 0 < y}
s+1

B(z) = B(z, s) = f (z)h(z)y

,

d0 z = y 2 dxdy.
Now, Y is a fundamental domain for \H, so the integral in (10) is equal to
Z
Z X
B(z)d0 z =
B((z))d0 z,
(11)
Y

where R is a set of representatives

\ and is a fundamental domain for \H. Since f Sk/2 (N, ), h


1
2
M(2k+1)/2 (4r , (d) d ) , we can write
B((z)) = (d)B(z)(cz + d)v |cz + d|2v12d


, where is a charcter modulo N r2 defined (d) = (d)(d) 1
. We must have (1) = 1 (because
d
otherwise we have f = 0 since it is a modular form of character), so we have (1) = (1)v . Thus, by
Lemma 3, (10) becomes
Z
X
B(z)[
(d)(cz + d)v |cz + d|2v12s ]d0 z

c,d

where the sum is taken over all relatively prime c, d such that c 0 (mod N r2 ), c > 0 if c 6= 0, d = 1 if c = 0.
Write

X
L(s, ) =
(n)ns ,
n=1

17

C(z, s) =

(n)(N r2 mz + n)v |N r2 mz + n|2v2s .

m= n=

Then we obtain from our relation (1) = (1)v that

2 (4)s (s)D(2s v, ) = 2 L(2s v + 1, )(4)s (s)

(m)am2 mv2s

m=1

Z
=

s+1 C(z, s)d0 z.

f hy

For the sake of simplifying notation, let = v, = (2v + 1)/2, 2t = 2s 2v + 1. Thus we have
s + 1 = t + . Now let 0 denote the primitive character associated with , 0 the primitive character


, and M the conductor of 0 . Let N = M K. Let denote the Mobius
associated with 1 (m) = (m) 1
m
function, defined

if n is square-free with an even number of prime factors

1
(n) = 1 if n is square-free with an odd number of prime factors

0
if n is a square
Let E be the product of all prime factors of N not dividing M . Then E|K and (n) is either equal to 0 (n)
or 0 depending on whether n, E are relatively prime or not. Thus,
X
X
C(z, s) =
0 (n)(N r2 mz + n) |N rm z + n|2t [
(k)]
m,n

k|GCD(n,E)

(k)0 (k)k 2t

0 (n)(M (K/k)r2 mz + n) |M (K/k)r2 mz + n|2t

m,n

k|E

where the sum runs over all (m, n) 6= (0, 0), and k over all positive divisors of E. Writing
C (z, t) = t (t)y t C(z, s),
X
H (t, z, 0 ) = t (t)y t
0 (n)(M rmz + n) |M rmz + n|2t ,
m,n

We have
C (z, t) = (Kr)t

(k)0 (k)k t H (t, Krz/k, 0 ),

k|E
s

2 (4)

(s)

Z
(t)D(2s v, ) =

C (z, t)d0 z.
f hy

(12)

But since v = 0 or 1, we have (s)(t) = 2 1/2 2v2s (2s v), and thus the LHS of (12) is equal to
222s (2)v2s (2s v)D(2s v, ).

(13)

Recall that in the statement of our Theorem we assumed k 3. Then 0 with equality only when
k = 3, v = 1. If k = 3, then 1 (1) = 1 = M > 1. Therefore, by the second assertion of Lemma 4 we
see that the function in (13) can be extended to a holomorphic function in the whole s-plane, bounded in
every vertical strip 1 < Re(s) < 2 . Thus D(s, ) has property (ii) required in Lemma 1.

18

Now we prove the most difficult condition (iii). Let (s) denote the function in (13). Substituting + v s
for s and + 1 t for t, we have
Z
C (c, + 1 t)d0 z.
( + v s =
f hy
(14)

Applying Lemma 4, we obtain

C (z, + 1 t) = (1)G(0 )z (M r)3t2 (Kr)t1

(k)0 (k)k t1 H (t, k/N r2 z, 0 ).

k|E

We will now make the transformation z 7 1/N

r2 z in the integral of (14). Let g(z) = f (1/N z)N k/4 (iz)k/2 .


N
0
0
Then g Sk/2 (N, ) with (d) = (d)

d , so we have
f (1/N r2 z) = g(r2 z)N k/4 (ir2 z)k/2 .
From a well-known theta series transformation formula, letting N = 4P , we have

h(1/N r2 z) = (i)v r1/2 G()(2rP

iz) h (P z),
so
(1/N r2 z) = T z g(r2 z)h
(P z)y ,
(f hy
Note that the transformation z 7 1/N r2 z takes a fundamental domain
where T = 2 r+ P k/4 iv g().
for \H to another fundamental domain, so we obtain
X
( + v s) = S
(k)0 (k)k t1 Ik (t),
k|E

R
(P z)y H (t, kz, 0 )d0 z and S = T G(0 )r4t33 M 3t22 K t1 . We want to
with Ik (t) = g(rz )h
transform the integral on as in (10). But this will prove rather difficult. First let

B 0 (z) = g(r2 z)h(P

z)y s+1 ,

a

2t

J(, z) = 0 (d)(cz + d) |cz + d|

, for
c


b
0 (M r).
d

Then B 0 ((z)) = B 0 (z)J(, z) for every 0 (N r2 ).

The case for which r = 1, N = M follows quickly from simple computation. However this approach does not
work if either r 6= 1 or N 6= M . We will deal with this difficulty by instead proving our result for the case
when all primes of N divide M , or equivalently, E = 1. The general case follows then from using the fact
that f is an eigenform to construct a desired primitive character.
P
Let R0 be a set of representatives for \0 (M r), and write (z) = R0 J(, z). From our assumption
we have
Z
( + v s) = 2S t (t)L(2t , 0 )
B 0 (z)(z)d0 z.

Now, E = 1 = every prime factor of N divides M , so we have the coset decomposition

0 (M r) =

Kr
[

0 (N r )u , u =

u=1

19


1
0
.
M ru 1

Now let be any fundamental domain for 0 (M r). Then considering as the disjoint union
can write
Z
Z X
Kr
B 0 (z)(z)d0 z =
B 0 (u (z))(u (z)d0 z

u (), we

u=1

Z
=

W (z)(z)d0 z,

(15)

PKr
where W (z) = i=1 B 0 (u (z))J(u , z)1 . Thus it is easy to see that for every 0 (M r), W ((z)) =
W (z)J(, z), so (15) is equal to
Z
XZ
W ((z))d0 z =
W (z)d0 z.
R0

Now we can decompose W (z) = W1 (z) + w2 (z), where

X
W1 (z) =
B 0 (u (z))J(u , z)1 ,
1uKr,(u,r)=1

W2 (z) =

K
X

B 0 (ru (z))J(ru , z)1 .

u=1

Next, we will use Lemma 5 to obtain Fourier expansions for W1 , W2 . For an integer v, put


1
0

v =
, (M vz + 1)1/2 ,
Mv 1
gv = g|[v ]k/2 .
Hence g depends only on v (mod K). Since we defined M to be conductor of 1 , by our assumption that
every P
prime factor of N divides M , we see that M is divisible by 4. Thus, from Lemma 5, we have that

gv (z) n=1 cv,n e2inz/K where cv,n are complex coefficients. For an integer u relatively prime to r, we can
choose integers p, q with p positive such that rp + N uq = 1. Then

  1



1
0
r
0
r
Kq
1 Kq/r
=
.
M ru 1
0
r
Mu
p
0
1
Thus, by Lemma 5 we have




 2 
r(z + Kq/r) Kq
M u k
r z
0
=g
= (p)
p (M uz + r1 )k/2 gv (z + Kq/r)
g
M ruz
M u(z + Kq/r) + p
p
0
with v satisfying pv 0 (mod
0 (r) and p = r . Furthermore, pr 1
 K).
 But pr
1 (mod N ), so (p) =
Mu
Mu
= r , and v ru (mod K). Thus, ince r and u are relatively prime,
(mod M Ku), and 4|M , so
p
we have


M u k k/2
g(r2 z/(M ruz + 1)) =
0 (r)
r r
(M ruz + 1)k/2 gru (z + Kq/r).
(16)
r

Further observe that

N uq 1

(mod r).
20

(17)

If u = rw, we have
g(r2 z/(M ruz + 1)) = g(r2 z/(M wrz + 1)) = gw (r2 z)(M ruz + 1)k/2 .

(18)

Now, to compute the transformation of h (P z) by u , write M = 4Q. Then by the theta trasnformation
formula we have
h (P z/(M ruz + 1)) = h (K Qz/(4ru Qz + 1))

= (1)v K 1v r1 (M ruz + 1) 21

mv (m)e2im

Qz/K

(19)

m=

where (m) =

Pr

j=1

PKr

k=1

(j)e2i[km+Kjkuk

W1 (z) = V y

Kr X

s+1

u=1 n=1

where V = (r)

K
r

Z
W (z)d0 z =

The integral

]/Kr]

. Combining this with (13), we have

u
X
2

cru,n e2[inq/r+nz/Km Qz/K] mv (m),

r
m=

r1k/2 (1)v K 1v k
r . Since W (z + 1) = W (z), we have
Z

RK

[K 1

(W1 (z) + W2 (z)dx]y 2 dy.

W1 (z)dx only has nonvanishing terms when n = m2 Q. So we can rewrite as

K

W1 (z)dx = V y s+1 21

(m)mv e2m

M y/K

m=

PKr Pr PKr
2im2 qQ/r 2i[uk2 kmKjk]/Kr

e
. Under the assumption r > 1,
where = u=1 j=1 k=1 cru,m2 Q ur (j)e
2
2 2
2 2
let k = Ka + rb. Then k K a + r b (mod Kr). Further put Ku v (mod r), u w (mod K).Then

PK
Pr Pr Pr PK v

2i[m2 qQamKajva2 ]/r 2i[rwb2 mb]/K

(m) = K
e
.
v=1
j=1
a=1
b=1 r (j)e
w=1 crw,m2 Q cw , where cw =
r
By (17), we have M qv 1 (mod r). So if 4vv 0 1 (mod r), we then have m2 qQ = m2 v 0 (mod r), so our
equation above for cw then becomes
cw =

K
X

e2i[rwb

mb]/K

r X
r X 
X
v
j=1 a=1 v,v 0

n=1

2i[m2 v 0 amKaj+va2 ]/r

(j)e
,

where the innermost sum is taken over all v, v 0 (mod r) satisfying 4vv 0 1 (mod r). Now, applying Lemma
6 to the last sum, we obtain
 
K
0

(m) =
(K)r r3/2 (2m)
(m)
(20)
,
r
where 0 (m) =
have

PK PK
b=1

w=1 cw,m2 Q e

K 1

Z
0

2i[wb2 mb]/K

. Now, notice that 0 (m) = 0 (m) and 0 (0) = 0, so we

W1 (z)dx = W y s+1

X
m=1

21

2
v 0

(m)m
(m)em M y/K ,

 v1

where W = (1)v 1 (r)(K)(2)r

K
. Then we obtain

W1 (z)y

dxdy = W

K (s)

(m)
(m)mv2s .

(21)

m=1

Now let us consider W2 for r > 1. By (18) and (19), we have that
K

W2 (z)dx = C y s+1 21

(m)mv er

m2 M y/K

m=

Pr PKr
PK
2i[rvk2 krmKjk]/Kr
.
where C is some constant not depending on m, and = v=1 cv,m2 Q j=1 k=1 (j)e
Now, if we let k a (mod r) and k b (mod K), we obtain
(m) =

K
X

cv,m2 Q

v=1

Pr

r X
r X
K
X

2i[vb2 mb]/K 2iaj/r

(j)e
e
.

j=1 a=1 b=1

Pr

2aj/r

But a=1 j=1 (j)e

= 0, so our expressions above are in fact zero. Therefore (15) in fact coincides
with (21), so we get

( + v s) = U L(2t , 0 )

(m)
(m)mv2s

(22)

m=1

where U = 2SW M s K s st (s)(t). To compute U , notice

G(0 ) = (M )0 (r)G(0 )G(),
= (1)G().
G()
Then, from the same relation we used to obtain (13), we have
U = U0 (r)2 (N/2)G()2 r1 r4s22v N 2s (2)v2s (2s v)
with constant U0 not depending on r and . Recall that 0 (m) does not depend on r and . Then, since
we also have 0 = 0 , we can rewrite (22) as
( + v s) = U

v2s

(n)A
nn

n=1

with coefficients An not depending on r and . Now put

Cp si = (N/2)(r)2 G()2 /r
s

R(s, ) = (2)

(s)

(n)bn ns .

n=1

Then R(s, ) = 2s+v2 ((s + v/2), so

R(2 s, ) = 22s+v2 ( + v (s + v)/2) = V C (r2 N/r)s (2)s (s)

X
n=1

22

(n)A
nn

where V is a constant again independent of r and . This shows that our Dirichlet series
satisfies (iii) of Weils Converse Theorem with N/2 and 2 as M and .

n=1 bn n

4. Bibliography
Neal Koblitz, Introduction to Elliptic Curves and Modular Forms, Springer-Verlag New York, Inc., NY, 1993
Goro Shimura, Introduction to the Arithmetic Theory of Automorphic Functions, Princeton University Press,
1971.
G. Shimura, On modular Forms of half-integral weight, Annals of Math. 97 (1973), 440-481.

A. Weil, Uber
die Bestimmung Dirichletscher Reihen durch Funktionalgleichungen, Math. Ann. 168 (1967),
149-156

23