Determining modular forms of half-integral

weight by central values of convolution

L-functions

Manish Kumar Pandey and B. Ramakrishnan

Abstract In this paper, we show that a Hecke eigenform g of half-integral weight on

Γ0 (4) which belongs to the Kohnen plus space is uniquely determined by the central
values of the family of convolution (Rankin-Selberg) L-functions L(s, f ⊗ g), where
f runs over an orthogonal basis of Hecke eigenforms of weight k + 1/2 on Γ0 (4)
which lie in the Kohnen plus space with k varying over an infinite set of integers.
This generalizes the work of Ganguly et. al [1] to the case of forms of half-integral
weight in the Kohnen plus space.

Key words: modular forms of half-integral weight, determination of modular forms,

Rankin-Selberg L-functions
2010 Mathematics Subject Classification 11F37 · 11F67 · 11F25

Manish Kumar Pandey

Department of Mathematics,
Indian Institute of Science Education and Research Bhopal,
Bhopal Bypass Road, Bhauri,
Bhopal 462 066, Madhya Pradesh, India,
e-mail: [email protected]
B. Ramakrishnan
(on lien) Harish-Chandra Research Institute, HBNI, Chhatnag Road, Jhunsi,
Prayagraj (Allahabad) 211 019, India.
Present Address:
Department of Statistics and Applied Mathematics,
Central University of Tamil Nadu,
Neelakudi, Thiruvarur 610 005, India,
e-mail: [email protected],[email protected]

1
2 Manish Kumar Pandey and B. Ramakrishnan

1 Introduction and Statement of Results

One of the interesting problems in number theory is to find criterions for the deter-
mination of Hecke eigenforms. It is known that a normalized Hecke eigenform of
integral weight is determined by its Fourier coefficients (which are nothing but the
eigenvalues) indexed by prime numbers. There are many authors who considered
this question of determination of a Hecke eigenform of integral weight by the special
values of its twists by a family of modular forms. Here we mention a few of them. In
[7] W. Luo and D. Ramakrishnan showed that if two normalized newforms f and 0
 f
are such that L(1/2, f , χd ) = L(1/2, f 0, χd ) for all quadratic characters χd = d· ,
then f and f 0 are equal. As an application, they proved that if g1 and g2 are newforms
in the Kohnen plus space of weight k + 1/2 on Γ0 (4N) (N is odd and square-free)
with Fourier coefficients b1 (n) and b2 (n) with the property that b21 (|D|) = b22 (|D|)
for almost all fundamental discriminants D with (−1)k D > 0, then g1 = ±g2 . This
question was posed by W. Kohnen in [5]. In 1999, Luo [6] proved the following
result. Let Hk (N) denote the orthogonal basis of normalized Hecke eigenforms of
weight k on Γ0 (N). Suppose that f and g are two normalized newforms of weight 2k
(resp. 2k 0) on Γ0 (N) (resp. Γ0 (N 0)). If there exist a positive integer ` and infinitely
many primes p such that for all forms h ∈ H2` (p), the central values of the Rankin-
Selberg L-functions are equal (i.e., L(1/2, f ⊗ h) = L(1/2, f 0 ⊗ h)), then k = k 0 and
N = N 0. This result of Luo can be viewed as the GL(2) analog of the result of Luo
and Ramakrishnan which is mentioned above. As a variant of Luo’s result, in [1],
S. Ganguly, J. Hoffstein and J. Sengupta considered twists by Hecke eigenforms of
fixed level and varying weight. More precisely, if g ∈ H` (1) and g 0 ∈ H`0 (1) are such
that L(1/2, f ⊗ g) = L(1/2, f ⊗ g 0), f ∈ Hk (1) for infinitely many k, then ` = ` 0 and
g = g 0. (Here k, `, ` 0 are all even positive integers.) There are other generalizations in
the case of eigenforms of integral weight (see for example [9, 12]). In this paper, we
generalize the work of Ganguly et. al in the case of forms of half-integral weight. We
consider Hecke eigenforms of half-integral weight on Γ0 (4) which lie in the Kohnen
plus space. The method is very similar to the one carried out by Ganguly et. al in
[1]. However, we need the development of Rankin-Selberg convolutions and their
approximate functional equations etc., for which we use a result of R. Munshi [8],
who has done the basics of this theory for the purpose of proving some simultaneous
non-vanishing of twisted L-functions of integral weight. We now state the main
result of this paper.
Main Theorem. Let g, g 0 be two Hecke eigenforms belonging to the Kohnen plus
space of weights ` + 1/2 and ` 0 + 1/2 on Γ0 (4), respectively. Suppose that

L(1/2, f ⊗ g) = L(1/2, f ⊗ g 0) (1)

for any Hecke eigenform f of weight k + 1/2 on Γ0 (4) belonging to the Kohnen plus
space. Then we have ` = ` 0 and g = g 0.
As mentioned before, we follow the method of Ganguly et. al [1]. After recalling
the basic properties of Rankin-Selberg convolution L-functions and its approximate
Title Suppressed Due to Excessive Length 3

functional equation in sections 3 and 4, we first prove an auxiliary result in section

5 giving an asymptotic expression for the spectral average of central values of the
Rankin-Selberg convolution L-functions and use it to prove our main theorem. In
sections 6 and 7 we give estimates for the main and error terms appearing in the
auxiliary theorem. Finally in section 8 we demonstrate the method of proving our
main theorem. As a first reduction, we use the auxiliary theorem to show that the
normalized |D|-th Fourier coefficients of g and g 0 are equal and then use the explicit
Waldspurger theorem obtained by Kohnen [5] to deduce that the special values of
the twisted L-functions of the corresponding (via the Shimura-Kohnen lift) integral
weight eigenforms are equal upto some constant. We then use the result of Luo and
Ramakrishnan [7] to finally prove the main theorem. As mentioned above, in [1], the
authors consider the twist by level 1 Hecke eigenforms and they had mentioned that
it is possible to consider twists by higher level newforms also to get similar result and
in [12], Y. Zhang had carried out this task. It is likely that in the case of half-integral
weight also one may carry out similar twists by newforms in the Kohnen plus space
of higher (square-free) level and obtain analogous result as in our main theorem. We
plan to do it in our forthcoming work. The contents of this paper are part of the first
author’s thesis [10].

2 Notations and Preliminaries

Let k ≥ 2 be a natural number. We denote by Sk+1/2 + (Γ0 (4)), the Kohnen plus space
of the space of modular forms f of weight k + 1/2 on Γ0 (4) whose Fourier coeffi-
cients a f (n) satisfy the following property that a f (n) , 0 implies that (−1)k n ≡ 0, 1
(mod 4). Let m ∈ N such that (−1)k m ≡ 0, 1 (mod 4). Then the m-th Poincaré series
in the Kohnen plus space Sk+1/2+ (Γ0 (4)) is characterised by

+ Γ(k − 1/2)
h f , Pk+1/2,m i= a f (m),
6(4πm)k−1/2
+
for all f ∈ Sk+1/2 (Γ0 (4)). The factor 6 in the denominator comes from the index
+
of Γ0 (4) in SL2 (Z). The Fourier expansion of the Poincaré series Pk+1/2,m (z) in
Sk+1/2 (Γ0 (4)) is given by
Õ
Pk+1/2,m (z) = gk+1/2,m (n)q n, (2)
n≥1
(−1) k n≡0,1 (mod 4)

where
" √ #
2 [ k+1
√ Õ π mn
gk+1/2,m (n) = δm,n + (−1) 2 π 2(n/m)
] (k/2−1/4)
Hc (m, n)Jk−1/2 ( ) ,
3 c>1
c
(3)
and q = e2πiz , z ∈ H , the complex upper half-plane. In the above,
4 Manish Kumar Pandey and B. Ramakrishnan
      k+1/2
4
1 Õ 4c −4
Hc (m, n) = (1 − (−1) i) 1 + k
e4c (mδ + nδ−1 ),
c 4c δ δ
∗ δ(4c)

where δ−1 is an integer such that δδ−1 ≡ 1 (mod 4c) and Jk−1/2 (x) is the Bessel
function of order k − /12. Further, we have also
used the notation ec (x) = e2πix/c
for a complex x and an integer c. The symbol dc denotes the generalised quadratic
residue symbol as described in [11, 3]. Now, we have the following Petersson formula:

+
Õ a f (m)
Pk+1/2,m (z) = ωf f, (4)
+
f ∈ Fk+1/2 (4)
mk−1/2

+
where Fk+1/2 (4) denotes an orthogonal basis for the plus space Sk+1/2 + (Γ0 (4)) and
Γ(` − 1/2) 1
ωf = . By comparing the n-th Fourier coefficients of both the sides
6(4π)k−1/2 h f , f i
of the above equation, we get

" √ #
Õ a f (m)a f (n) 2 [ k+1
√ Õ π mn

ωf = δm,n + (−1) 2 π 2(n/m)
] (k/2−1/4)
Hc (m, n)Jk−1/2 ,
+
f ∈ Fk+1/2 (4)
mk−1/2 3 c>1
c
(5)
where ω f is defined after Eq.(4). For details on modular forms of half-integral weight
and the plus space, we refer to [3, 4, 5].

3 Rankin-Selberg L-functions

In this section, we shall obtain a functional equation satisfied by the Rankin-Selberg

L-function associated to forms of half-integral weight. Such a study for forms of
half-integral weight was first done by R. Munshi [8] and we refer the Í reader to this
work for more details. Let fi ∈ Sk+i +1/2 (Γ0 (4)), i = 1, 2, and fi (z) = n0 ≥1 a fi (n)e(nz)
be their Fourier expansions, where the sum 0 varies over all natural numbers n with
Í
(−1)ki n ≡ 0, 1 (mod 4). We also assume that k1 and k2 have  the  same parity, i.e.,
ab
k1 ≡ k2 (mod 2). Set H(z) = f1 (z) f2 (z). Then for any γ = ∈ Γ0 (4), we get
cd
k1 +1/2
H(γ(z)) = (cz + d) (cz + d)k2 +1/2 f1 (z) f2 (z).
 
10
Note that the group Γ0 (4) has three cusps ∞, 0, 1/2 and the matrices g∞ = ,
01
 −1  −1
0 2 1 2
g0 = , g1/2 = take the cusp ∞ to the corresponding cusps ω = ∞, 0, 1/2
20 20
respectively. Corresponding to each cusp ω = ∞, 0, 1/2, there is an integral weight
Title Suppressed Due to Excessive Length 5

Eisenstein series of level 4 (and with weight k 0) given by the following.

Õ 0
Eω (z, s; k 0) = −1
j(gω γ, z)k Im(gω
−1
γz)s . (6)
γ ∈Γω \Γ0 (4)

In the above, j(γ, z) = (cz + d)(cz + d)−1 and the stabilizer Γω of the cusps ω is
given by Γω = gω Γ∞ gω −1 , ω = ∞, 0, 1/2. It is known that these Eisenstein series

converge absolutely for Re(s) > 1 and that they have analytic continuations to the
whole of C. Further, they satisfy a functional equation. In particular, we have

E∞ (z, 1 − s; k 0) = φ∞ (s)E∞ (z, s; k 0) + φ0 (s)E0 (z, s; k 0) + φ1/2 (s)E1/2 (z, s; k 0), (7)
where
24s−3 ζ(2s)Γ(s + k 0)π −s
φ∞ (s) = (8)
(1 − 22s−2 ζ(2 − 2s)Γ(1 − s + k 0)π −(1−s) )
and
1
φ0 (s) = φ1/2 (s) = ( − 1)φ∞ (s). (9)
22s−1
 
ab
Also for any γ = ∈ Γ0 (4) we have
c d
−k 0 0
Eω (γz, s; k 0) = (cz + d) (cz + d)k Eω (z, s; k 0).
k 1 +k2 +1
So, the function H(z)Eω (z, s; k1 −k 2
2 )y
2 is invariant under Γ0 (4) and therefore,
one can consider the integral
∫
k1 − k2 k1 +k2 +1 dxdy
Rω = H(z)Eω (z, s; )y 2 . (10)
2 y2
Γ0 (4)\H

Now following the standard unfolding argument, we obtain the Rankin-Selberg L-

function as follows:
∞
Õ a f1 (n)a f2 (n)
L(s, f1 × f2 ) = . (11)
n=1
ns+(k1 +k2 −1)/2
In the following, we define the completed Rankin-Selberg L-functions associated to
f1 and f2 corresponding to each of the three cusps ∞, 0, 1/2.
∞
∗
Õ a f1 (n)a f2 (n)
Λ∞ (s, f1 ⊗ f2 ) = π −2s−k Γ(s + k 0)Γ(s + k ∗ )ζ(2s) ,
n=1
ns+k ∗
∞
∗
Õ a f1 (n)a f2 (n)
Λ0 (s, f1 ⊗ f2 ) = π −2s−k Γ(s + k 0)Γ(s + k ∗ )ζ(2s) ,
n=1
ns+k ∗ (12)
n≡0 (mod 4)

∞
∗
Õ a f1 (n)a f2 (n)
Λ∞ (s, f1 ⊗ f2 ) = π −2s−k Γ(s + k 0)Γ(s + k ∗ )ζ(2s) .
n=1
ns+k ∗
n≡(−1) k1 (mod 4)
6 Manish Kumar Pandey and B. Ramakrishnan

In the above, we made the following substitutions, k 0 = (k1 − k2 )/2 and k ∗ =

(k1 + k 2 − 1)/2. We also assume that k1 > k 2 . Note that our definition of completed
Rankin-Selberg L-function differs from the definition given by Munshi [8, p. 671] by
some power of 2. The functional equation we obtain below is the same as obtained
by Munshi. However, he combines both the cases of k 0 even and odd into a single
functional equation. Note that as k1 and k 2 have the same parity, k 0 is an integer.
These completed Rankin-Selberg L-functions together satisfy a functional equation,
which is given below.

Λ∞ (1−s, f1 ⊗ f2 ) = ψ∞ (s)Λ∞ (s, f1 ⊗ f2 )+ψ0 (s)Λ0 (s, f1 ⊗ f2 )+ψ1/2 (s)Λ1/2 (s, f1 ⊗ f2 ),

(13)
where 0
1 (−1)k (1 − 22s−1 )
ψ∞ (s) = , ψ0 (s) = ψ1/2 (s) = . (14)
2(1 − 22s−2 ) 2(1 − 22s−2 )
Note that ψ∞ (1/2) = 1. Since both f1 and f2 belong to the Kohnen plus space, we
see that
Λ∞ (s, f1 ⊗ f2 ) = Λ0 (s, f1 ⊗ f2 ) + Λ1/2 (s, f1 ⊗ f2 ).
So, with this observation we have the following.
(
Λ∞ (s, f1 ⊗ f2 ), if k 0 is even,
Λ∞ (1 − s, f1 ⊗ f2 ) = (15)
(2ψ∞ (s) − 1)Λ∞ (s, f1 ⊗ f2 ), if k 0 is odd.

4 Approximate functional equation

In this section, we determine approximate functional equation for the completed

Rankin-Selberg L-function and use it to get an expression for the central value
of the Rankin-Selberg L-function. The methods used in this section are standard.
For details we refer to [2, 1]. As in §3, we assume that fi ’s are modular forms
in Sk+i +1/2 (Γ0 (4)) and also we use the same notation for k 0 and k ∗ as given in §3.
Let G(u) be a holomorphic function on an open set containing |Re(u)| 6 3/2 and
bounded therein. We also choose the function G such that G(0) = 1, G(−u) = G(u)
2
(later we will be taking G(u) = eu ). For X > 0, we consider the integral
∫
1 G(u)
I(X, s) = X u Λ∞ (s + u, f1 ⊗ f2 ) du.
2πi (3/2) u

We now move the line of integration from 3/2 to −3/2, which will pick up the residue
at u = 0 and so we get
∫
1 G(u)
I(X, s) = X u Λ∞ (s + u, f1 ⊗ f2 ) du + Λ∞ (s, f1 ⊗ f2 ).
2πi (−3/2) u
Title Suppressed Due to Excessive Length 7

Therefore,
∫
1 G(u)
Λ∞ (s, f1 ⊗ f2 ) = I(X, s) − X u Λ∞ (s + u, f1 ⊗ f2 ) du.
2πi (−3/2) u

Using the functional equation given by (15), the above expression is given by

 I(X, s) + I(X , 1 − s),


 −1 if k 0 is even
Λ∞ (s, f1 ⊗ f2 ) =

I(X −1, 1 − s) (16)
 I(X, s) +
 , if k 0 is odd.
 2ψ∞ (s) − 1

(Recall that k 0 = (k1 − k2 )/2.) We now define

L(s, f1 ⊗ f2 ) := ζ(2s)L(s, f1 × f2 ), (17)

where L(s, f1 × f2 ) is defined by (11). We write the Dirichlet series corresponding to

ζ(2s)L(s, f1 × f2 ) as follows.
∞
Õ
L(s, f1 ⊗ f2 ) = ζ(2s)L(s, f1 × f2 ) = b f1 ⊗ f2 (n)n−s, (18)
n=1

where the coefficients are given by

Õ a f (m)a f (m)
b f1 ⊗ f2 (n) = 1 2
. (19)
2
mk ∗
n=mt

(Note: k ∗ = (k1 + k 2 − 1)/2.) Substituting the above notation, the integral I(X, s)
becomes
∞
b f1 ⊗ f2 (n) G(u)
∫
1 u −2s−2u−k ∗
Õ
I(X, s) = X π Γ(s + u + k )Γ(s + u + k )
∗ 0
du.
2πi (3/2) n=1
ns+u u

Now, interchanging the order of integration and summation, we get

∞
−2s−k ∗
Õ b f1 ⊗ f2 (n) π2 n
I(X, s) = π Γ(s + k )Γ(s + k )
∗ 0
Vs ( ), (20)
n=1
ns X

where
π2 n Xu
∫
1 G(u)
Vs ( )= γ(s, u) du,
X 2πi (3/2) (π 2 n)u u
(21)
Γ(s + u + k ∗ )Γ(s + u + k 0)
γ(s, u) = .
Γ(s + k ∗ )Γ(s + k 0)

Substituting X by X −1 and s by 1 − s in (20), we get

8 Manish Kumar Pandey and B. Ramakrishnan
∞
∗
Õ b f1 ⊗ f2 (n) π2 n
I(X −1, 1 − s) = π −2+2s−k Γ(1 − s + k ∗ )Γ(1 − s + k 0) V1−s ( ).
n=1
n1−s X −1

Therefore, using (16), when k 0 is even, we get

∞
∗
Õ b f1 ⊗ f2 (n) π2 n
Λ∞ (s, f1 ⊗ f2 ) = π −2+2s−k Γ(1 − s + k ∗ )Γ(1 − s + k 0) V1−s ( )
n=1
n1−s X −1
∞
∗
Õ b f1 ⊗ f2 (n) π2 n
+ π −2s−k Γ(s + k ∗ )Γ(s + k 0) Vs ( )
n=1
ns X

and when k 0 is odd, it follows that

∞
∗
Õ b f1 ⊗ f2 (n) π2 n
Λ∞ (s, f1 ⊗ f2 ) = π −2s−k Γ(s + k ∗ )Γ(s + k 0) Vs ( )
n=1
ns X
∗ ∞
π −2+2s−k Γ(1 − s + k ∗ )Γ(1 − s + k 0) Õ b f1 ⊗ f2 (n) π2 n
+ 1−s
V1−s ( −1 ).
2ψ∞ (s) − 1 n=1
n X

Observe that at the point s = 1/2, both sides of the above expressions have the same
gamma factor and the same power of π, and so after cancellation of these terms, we
get
∞
Õ b f1 ⊗ f2 (n) π2 n π2 n
L(1/2, f1 ⊗ f2 ) = (V1/2 ( ) + V1/2 ( −1 )).
n=1
n1/2 X X
Now, substituting X = 1 in the above, we have the following expression for the
central value:
∞
Õ b f1 ⊗ f2 (n)
L(1/2, f1 ⊗ f2 ) = 2 1/2
V1/2 (π 2 n). (22)
n=1
n

5 An Auxiliary Theorem

+
Let g ∈ S`+1/2 (Γ0 (4)) be Hecke eigenform with Fourier coefficients ag (n). For a
fixed fundamental discriminat D with (−1)k D > 0, the following average simplifies
using (22) as follows.
∞
Õ a f (|D|) Õ Õ b f ⊗g (n) a f (|D|)
ω f L(1/2, f ⊗g) =2 ωf V1/2 (π 2 n).
f ∈ Fk+1/2 (4)
|D| k/2−1/4 f ∈ Fk+1/2 (4) n=1
n1/2 |D| k/2−1/4
(23)
From now onwards, we use the following notation κ and κ ∗ (instead of k 0 and k ∗ ):
κ = (k − `)/2 and κ ∗ = (k + ` − 1)/2. Now substituting for b f ⊗g (n) from Eq.(19),
Title Suppressed Due to Excessive Length 9

we get,
∞
!
Õ a f (|D|) Õ Õ Õ a f (m)ag (m) a f (|D|) V1/2 (π 2 n)
ω f L(1/2, f ⊗ g) =2 ωf
|D| k/2−1/4 mκ ∗ |D| k/2−1/4 n1/2
f ∈ Fk+1/2 (4) f ∈ Fk+1/2 (4) n=1 n=mt 2

∞
Õ V1/2 (π 2 n) Õ ag (m) Õ a f (m)a f (|D|)
=2 ωf .
n=1
n1/2 m`/2−1/4 (m|D|)k/2−1/4
n=mt 2 f ∈ Fk+1/2 (4)
(24)

Using the Petersson formula (Eq.(5)), the above becomes

 
Õ a f (|D|) 4 ag (|D|)
ω f L(1/2, f ⊗ g) k/2−1/4 = M |D | (k, `) + Eg, |D | (k, `) ,
|D| 3 |D| `/2−1/4
f ∈F (4)
k+1/2
(25)
where M |D | (k, `) is the main term given by
∞
Õ V1/2 (π 2 |D|t 2 )
M |D | (k, `) = |D| −1/2 (26)
t=1
t

and Eg, |D | (k, `) is the error term given by

∞
!
π m|D|
p
Õ V1/2 (π 2 n) Õ ag (m) [ k+1
]
√ Õ
Eg, |D | (k, `) = (−1) 2 π 2 Hc (|D|, m)Jk−1/2 ,
n=1
n1/2 m`/2−1/4 c
n=mt 2 c>1
(27)
with V1/2 (π 2 x) is given by (21).
Thus, we obtained the following (auxiliary) theorem to prove our main result.
Theorem A: Let g be a cusp form in the Kohnen plus space S`+1/2 + (Γ0 (4)) and
+
Fk+1/2 (4) be an orthogonal basis for the space Sk+1/2 (Γ0 (4)). Then we have the
following formula for the spectral average of the central value of the Rankin-Selberg
(convolution) L-functions.
 
Õ a f (|D|) 4 ag (|D|)
ω f L(1/2, f ⊗ g) k/2−1/4 = M |D | (k, `) + Eg, |D | (k, `) ,
|D| 3 |D| `/2−1/4
f ∈F (4)
k+1/2
(28)
where M |D | (k, `) and Eg, |D | (k, `) are given by Eqs.(26), (27) (ω f and L(1/2, f ⊗ g)
are defined in §2 and §3 respectively).
In the next two sections, we shall give estimates for these main and error terms in
order to get our main result.
10 Manish Kumar Pandey and B. Ramakrishnan

6 Estimation of the Main Term M|D| (k, `)

We have
∞
Õ V1/2 (π 2 |D|t 2 )
M |D | (k, `) = |D| −1/2
t=1
t
∞ −u
(π 2 |D|t 2 ) Γ(u + a)Γ(u + b)G(u)
∫
Õ 1 1
= |D| −1/2 du (by using (21)).
t=1
t 2πi (3/2) Γ(a)Γ(b)u

Here we have put a = κ ∗ + 1/2 and b = κ + 1/2, where κ, κ ∗ are defined as in §5. So
we have,
−u
|D| −1/2 (π 2 |D|) Γ(u + a)Γ(u + b)G(u)
∫
M |D | (k, `) = ζ(2u + 1)du.
2πi (3/2) Γ(a)Γ(b)u

Now by moving the line of integration to Re(u) = −1/2 and noticing that the
integrand has a double pole at u = 0, with the residue at u = 0 given by
0 0
Γ Γ
(a) + (b) + 2γ0 − log(π 2 |D|),
Γ Γ
where γ0 is the Euler’s constant. Therefore, we have
0 0
Γ Γ
(a) + (b) + 2γ0 − log(π 2 |D|) + I,
|D| 1/2 M |D | (k, `) =
Γ Γ
where I denotes the following integral along the line (−1/2).
−u
(π 2 |D|) Γ(u + a)Γ(u + b)G(u)
∫
1
I= ζ(2u + 1)du.
2πi (−1/2) Γ(a)Γ(b)u

For estimating the integral I, we need the following known estimate for the ratio of
Γ-functions, using the Stirling bound. See for example [1, Lemma 2]. For A(> 0), c
real with |c| < A/2, one has

Γ(A + c + it)
 | A + it| c, (29)
Γ(A + it)
where the implied constant depend on c. Using this estimate along with the fact that
|Γ(x + iy)| 6 |Γ(x)|, we get the following estimate of the integral I along the line
Re(u) = −1/2.
|D| 1/2 ∞ G(−1/2 + iv)
∫
I6 | ||ζ(2iv)|dv.
2κ −∞ (−1/2 + iv)
1/2
Further, using the fact that ζ(it)  |t| 1/2 , it follows that I  |Dκ| . Finally, combin-
ing everything, we have the following estimate for the main term:
Title Suppressed Due to Excessive Length 11
 0 0 
Γ Γ
M |D | (k, `) = |D| 1/2 (a) + (b) + 2γ0 − log(π 2 |D|) + O(1/k). (30)
Γ Γ

7 Estimation of the error term E g,|D| (k, `)

Before we proceed to estimate the error term, we obtain some preliminary results on
certain Dirichlet series in the following section.

7.1 Preliminary results

In this section we prove the functional equation of certain Dirichlet series associated
to modular forms of half-integral weight in the Kohnen plus space. Let g(z) =
Í ∞ 2πinz ∈ S +
n=1 ag (n)e `+1/2
(4). We consider the following Dirichlet series associated
to g, defined by
∞ a (n)e( αn )
α Õ g β
Lg (s, ) = , (31)
β n=1
n `/2−1/4+s

where α, β are positive integers with (α, β) = 1. In the above, we have used the
following notation: e(x) = e2πix . Next, we prove the functional equation satisfied by
the above Dirichlet series. Suppose that t ∈ R+ . Since g is invariant under the action
of Γ0 (4), we get
 c   −4  −`−1/2
g(γz) = (cz + d)`+1/2 g(z),
d d
 
ab
where γ = ∈ Γ0 (4). Now, taking z = −d
c + it
c, we get γz = a
c + i
ct , so that
cd

−d it  c   −4  `+1/2 a i
g( + )= (cz + d)−`−1/2 g( + ). (32)
c c d d c ct

The Fourier expansions of g(z) and g(γz) (for the above value of z) are given by
∞
−d it Õ −nd −2π t n
g( + )= ag (n)e( )e c ,
c c n=1
c
∞
a i Õ na −2π n
g( + ) = ag (n)e( )e c t .
c ct n=1
c

Using (32), the Mellin transform of g becomes

12 Manish Kumar Pandey and B. Ramakrishnan
∫ ∞
−d it dt  c   −4  `+1/2
∫ ∞
a i dt
g( + )t s+`/2−1/4 = (it)−`−1/2 g( + )t s+`/2−1/4 .
0 c c t 0 d d c ct t
(33)
Now substituting the Fourier expansion of g(z) and g(γz) as given above, we get

∞
Õ ag (n)e( −nd )
c
(c/2π)s+`/2−1/4 Γ(s + `/2 − 1/4)
n=1
ns+`/2−1/4
 c   −4  `+1/2 ∞
ag (n)e( na
c )
= (c/2π)`/2+3/4−s Γ(`/2 + 3/4 − s)
Õ
ι−`−1/2
d d n=1
n`/2+3/4−s
(34)

7.2 Error term estimation

For simplicity, we write the error term as E and it is given by

∞
!
π m|D|
p
Õ V1/2 (π 2 n) Õ ag (m) [ k+1
]
√ Õ
(−1) 2 π 2 Hc (|D|, m)Jk−1/2 ,
n=1
n1/2 m`/2−1/4 c
n=mt 2 c>1

So, we write it as
∞ ∞
!
π
2 mt 2 ) Õ
p
k+1 √ Õ a g (m) Õ V1/2 (π m|D|
E = (−1)[ 2 ]π 2 Hc (|D|, m)Jk−1/2 .
m=1
m`/2+1/4 t=1 t c>1
c

Proceeding as done in [1, §4], (by using the inverse Melling transform of the J-Bessel
function), we can write E as
k+1 √
(−1)[ 2 ] π 2
E=
2(2πi)2
G(u) Γ( 2 − 4 + 2 ) Γ( 2 + u + κ) Γ( 2 + u + κ ∗ ) s −2u−s −s/2
∫ ∫ k 1 s 1 1
× ζ(2u + 1) 2 π |D| Su duds,
u Γ( k2 + 34 − 2s ) Γ( 12 + κ) Γ( 12 + κ ∗ )
(3/2) (α)

where
∞ Õ
Õ ag (m) Hc (|D|, m)
Su = .
m=1 c>1
m`/2+1/4+u+s/2 c−s

Using the Weil bound for the Kloosterman sum Hc (|D|, m), the series converges
absolutely and so we can change the order of summation in Su to get
Title Suppressed Due to Excessive Length 13
Õ  4c   −4  k+1/2 Õ∞
|D|a + md
  s−1  
Õ 4 c ag (m)
Su = (1 − (−1) i)(1 +
k
) e
c>1
c 4 ∗
a
a (mod 4c)
a m=1
m`/2+1/4+u+s/2 4c
Õ  
4 Õ  4c   −4  k+1/2 e( |D |a ) d
4c
= (1 − (−1)k i)(1 + ) 1−s
Lg (1/2 + u + s/2, ),
c a a c 4c
c>1 a (mod 4c)∗

where d is an integer such that ad ≡ 1 (mod 4c) and we have denoted the sum
over m by the Dirichlet series Lg (s, d/4c) as defined in §7.1. Now by applying the
functional equation for the Dirichlet series given by (34), we get

i `+1/2 Γ( ` + 1
− u − 2s ) Õ Õ  −4  k−`
 
4 −1−2u
Su = (1 − (−1) i)k
(π/2)2u+s `2 4
(1 + )c
4 Γ( 2 + 1
4 + u + 2s ) c>1 c
a (mod 4c)∗
a
−a
|D|a
× Lg 1/2 − u − s/2, e( ).
4c 4c
Now we move the line of integration in the s variable to Re(s) = α = −7. Since
Re(−u − s/2 + 1/2) = 5/2, the Dirichlet series Lg (1/2 − u − s/2, −a/4c) is absolutely
convergent. Therefore, we can write
  k−`
Õ −4 |D|a −a

e( )Lg 1/2 − u − s/2,
a 4c 4c
a (mod 4c)∗
∞   k−`
Õ ag (m) Õ −4 (|D| − m)a

= e
m=1
m`/2+1/4−u−s/2 a (mod 4c)∗
a 4c

Since we are interested in the case where k and ` having the same parity, the sum
over a reduces to the Ramanujan sum and so we have the following estimate for Su :

Γ(`/2 + 1/4 − u − s/2)

Su  (π/2)2u+s .
Γ(`/2 + 1/4 + u + s/2)
Thus, the estimate of the error term E simplifies as

G(u) Γ(k/2 − 1/4 + s/2) Γ(1/2 + u + κ)

∫ ∫
E (π/2)2u+s ζ(2u + 1)
u Γ(k/2 + 3/4 − s/2) Γ(1/2 + κ)
(3/2) (−7) (35)
Γ(1/2 + u + κ ∗ ) Γ(`/2 + 1/4 − u − s/2)
× duds.
Γ(1/2 + κ ∗ ) Γ(`/2 + 1/4 + u + s/2)

Now writing u = 3/2 + iv and s = −7 + it and integrating with respect to the t

variable we get,
2
∞
e−v Γ(2 + iv + κ ∗ )Γ(2 + iv + κ)
∫
E k −7 (k 2 + v 2 )1/2 | |dv.
−∞ (9/4 + v 2 )1/2 Γ(1/2 + κ ∗ )Γ(1/2 + κ)
14 Manish Kumar Pandey and B. Ramakrishnan

Finally, using the bound for the ratio of Γ-functions and the estimate |Γ(x + iy)| 6
|Γ(x)|, we have
∞ 2
e−v
∫
E  k −7 (v 2 + k 2 )7/2 dv  1.
−∞ (9/4 + v 2 )1/2

8 Proof of the Main Theorem

Using our auxiliary result (Theorem A), we shall prove the main result in this
+
section. By assumption, the functions g ∈ S`+1/2 (Γ0 (4)) and g 0 ∈ S`+0 +1/2 (Γ0 (4))
are Hecke eigenforms such that L(1/2, f ⊗ g) = L(1/2, f ⊗ g 0), for all Hecke
+
eigenforms f ∈ Sk+1/2 (Γ0 (4)). Therefore, Theorem A implies that for all fundamental
discriminants D with (−1)k D > 0,

ag (|D|) ag0 (|D|)

M |D | (k, `) + Eg, |D | (k, `) = M |D | (k, ` 0) + Eg0, |D | (k, ` 0). (36)
|D| `/2−1/4 |D| `0 /2−1/4

Using Stirling’s formula for the derivatives of Γ(s), it follows that for k large,
M |D | (k, `) = log k + O(1). Also the error terms are bounded for large k. Using these
two observations in (36) we get

ag (|D|) ag0 (|D|)

`/2−1/4
= , (37)
|D| |D| `0 /2−1/4

for all fundamental discriminants D with (−1)k D > 0. Let F and F 0 be the normalised
Hecke eigenforms of weights 2` and 2` 0 on SL2 (Z), corresponding to the Hecke
eigenforms g and g 0 (via the Shimura-Kohnen maps). Using the corresponding
Waldspurger’s formula for g and g 0, obtained by Kohnen [5, Corollary 1] and using
(37), we see that
L(F, χD, `) = C L(F 0, χD, ` 0), (38)
for all fundamental discriminants with (−1)k D > 0 and C > 0 is a constant. (Here
L(F, χD, `) denotes the usual
L-function associated to the modular form F twisted
with the character χD = D· .) Using Theorem B of Luo-Ramakrishnan [7], this
implies that ` = ` 0 and F = F 0. Our main theorem now follows using the ‘multiplicity
+
1’ result in S`+1/2 (Γ0 (4)).

References

1. S. Ganguly, J. Hoffstein and J. Sengupta, Determining modular forms on SL(2, Z) by central

values of convolution L-functions, Math. Ann. 345 (2009), 843–857.
Title Suppressed Due to Excessive Length 15

2. H. Iwaniec and E. Kowalski, Analytic Number Theory, Amer. Math. Soc. Colloq. Publ., vol.
53, American Mathematical Society, Providence, RI, 2004.
3. N. Koblitz, Introduction to elliptic curves and modular forms, Graduate studies in Mathematics
97, Springer–Verlag, 1993.
4. W. Kohnen, Modular forms of half-integral weight on Γ0 (4), Math. Ann. 248 (1980), no. 3,
249–266.
5. W. Kohnen, Fourier coefficients of modular forms of half-integral weight, Math. Ann. 271
(1985), 237–268.
6. W. Luo, Special L-values of Rankin-Selberg convolutions, Math. Ann. 314 (1999), no. 3,
591–600.
7. W. Luo and D. Ramakrishnan, Determination of modular forms by twists of critical L-values,
Invent. Math. 130 (1997), no. 2, 371–398.
8. R. Munshi, A note on simultaneous nonvanishing twists, J. Number Theory 132 (2012), 666–
674.
9. Q. Pi, Determining cusp forms by central values of Rankin-Selberg L-functions, J. Number
Theory 130 (2010), 2283–2292.
10. Manish Kumar Pandey, Explicit Shimura Liftings of Certain Class of Forms and Some Prob-
lems Involving Modular L-Functions, Ph.D Thesis, Homi Bhabha National Institute (HRI,
Prayagraj), August 2019.
11. G. Shimura, On modular forms of half integral weight, Ann. Math (2) 97 (1973), 440–481.
12. Y. Zhang, Determining modular forms of general level by central values of convolution L-
functions, Acta Arith. 150 (2011), 93–103.