CHARACTER ANALOGUES OF RAMANUJAN TYPE INTEGRALS

INVOLVING THE RIEMANN Ξ-FUNCTION

ATUL DIXIT
arXiv:1102.2680v1 [math.NT] 14 Feb 2011

Abstract. A new class of integrals involving the product of Ξ-functions associated

with primitive Dirichlet characters is considered. These integrals give rise to trans-
formation formulas of the type F (z, α, χ) = F (−z, β, χ) = F (−z, α, χ) = F (z, β, χ),
where αβ = 1. New character analogues of transformation formulas of Guinand and
Koshliakov as well as those of a formula of Ramanujan and its recent generalization
are shown as particular examples. Finally, character analogues of a conjecture of
Ramanujan, Hardy and Littlewood involving infinite series of Möbius functions are
derived.

1. Introduction
Modular transformations are ubiquitous in Ramanujan’s Notebooks [33] and his Lost
Notebook [35]. Ramanujan usually expressed them in a symmetric way and they were
valid under the conditions αβ = π or αβ = π 2 etc. In the same spirit, on page 220 in
one of the manuscripts of S. Ramanujan in the handwriting of G.N. Watson published
in [35], one finds the following beautiful claim.

Theorem 1.1. Define

1
λ(x) := ψ(x) + − log x,
2x
where
∞ 
Γ′ (x)

X 1 1
ψ(x) := = −γ − −
Γ(x) m=0
m+x m+1

is the logarithmic derivative of the Gamma function. Let the Riemann ξ-function be
defined by
1
ξ(s) := (s − 1)π − 2 s Γ(1 + 21 s)ζ(s),
and let
Ξ(t) := ξ( 21 + it)

2010 Mathematics Subject Classification. Primary 11M06; Secondary 11M35.

Keywords and phrases. Dirichlet character, Dirichlet L-function, modified Bessel function, Möbius
function, Mellin transform, Ramanujan, Hardy, Littlewood, Koshliakov, Guinand.
1
2 ATUL DIXIT

be the Riemann Ξ-function. If α and β are positive numbers such that αβ = 1, then

∞ ∞
√
   
γ − log(2πα) X p γ − log(2πβ) X
α + λ(kα) = β + λ(kβ)
2α 2β
k=1 k=1
2
cos 12 t log α
Z ∞   

1 1 −1 + it
= − 3/2 Ξ t Γ dt, (1.1)
π 0 2 4 1 + t2

where γ denotes Euler’s constant.

This identity is of a special kind since not only does it contain a modular trans-
formation, but also a beautiful integral involving the Riemann Ξ-function. In fact,
the invariance of the integral in (1.1) under the map α → β establishes the equality
of the first and the second expressions in (1.1). This idea is used in [6] to prove the
above claim of Ramanujan and later in [11, 12, 13] to obtain many transformation
formulas of the type F (α) = F (β) or F (z, α) = F (z, β), where αβ = 1, where an
integral involving the Riemann Ξ-function is always linked to them. This then gives
new identities involving infinite series of Hurwitz zeta function as well as extensions of
some well-known formulas of A.P. Guinand [16] and N.S. Koshliakov [25] (see [13] for
details). For example, we mention the following generalization of Theorem 1.1 found
in [11, 13].

Theorem 1.2. Let −1 < Re z < 1. Define ϕ(z, x) by

x−z 1 −z−1
ϕ(z, x) = ζ(z + 1, x) − − x ,
z 2

where ζ(z, x) denotes the Hurwitz zeta function. Then if α and β are any positive
numbers such that αβ = 1,

∞
! ∞
!
z+1
X ζ(z + 1) ζ(z) z+1
X ζ(z + 1) ζ(z)
α 2 ϕ(z, nα) − z+1
− =β 2 ϕ(z, nβ) − −
n=1
2α αz n=1
2β z+1 βz
z−3 Z ∞
t − iz cos 12 t log α
       

8(4π) 2 z − 1 + it z − 1 − it t + iz
= Γ Γ Ξ Ξ dt,
Γ(z + 1) 0 4 4 2 2 (z + 1)2 + t2
(1.2)

where Ξ(t) is the Riemann Ξ-function.

Another example of a transformation formula of the type F (z, α) = F (z, β) is the

extended version of Guinand’s formula [13, Theorem 1.4] given below.

Theorem 1.3. Let Kν (s) denote P the modified Bessel function of order ν, let γ denote
Euler’s constant and let σk (n) = d|n dk . Let −1 < Re z < 1. Then if α and β are
 CHARACTER ANALOGUES 3

positive numbers such that αβ = 1, we have

∞
!
√ z   
z −z z z −z X
α α 2 −1 π 2 Γ ζ(z) + α− 2 −1 π 2 Γ ζ(−z) − 4 σ−z (n)nz/2 K z2 (2nπα)
2 2 n=1
  ∞
!
p z −z
 z  z z −z X
= β β 2 −1 π 2 Γ ζ(z) + β − 2 −1 π 2 Γ ζ(−z) − 4 σ−z (n)nz/2 K z2 (2nπβ)
2 2 n=1
Z ∞  1


cos 2 t log α
  
32 t + iz t − iz
=− Ξ Ξ dt. (1.3)
π 0 2 2 (t + (z + 1)2 )(t2 + (z − 1)2 )
2

Letting z → 0 in (1.3) then gives the following extended version of Koshliakov’s

formula [12].
Theorem 1.4. Let d(n) denote the number of positive divisors of n and let K0 (n)
denote the modified Bessel function of order 0. If α and β are positive numbers such
that αβ = 1, then
∞
! ∞
!
√ γ − log(4πα) X p γ − log(4πβ) X
α −4 d(n)K0 (2πnα) = β −4 d(n)K0 (2πnβ)
α n=1
β n=1
Z ∞ t


2 1


32 Ξ 2 cos 2 t log α dt
=− .
π 0 (1 + t2 )2
By an ‘extended version’, we mean that the original identity known before is linked
to an integral involving the Riemann Ξ-function. It must be mentioned here that the
formulas of Guinand and Koshliakov were discovered earlier by Ramanujan (see [8]),
and are present in the Lost Notebook [35, p. 253–254].
After Ramanujan, Koshliakov was another person to do significant research in this
area. Apart from using contour integration, Mellin transforms and several summation
formulas that he developed, he frequently used a method similar to that developed
by Ramanujan in [32], [34, pp. 72–77] to obtain old and new transformation formulas
of the form F (α) = F (β), where αβ = k for some constant k. He also obtained
deep generalizations of many well-known formulas of Ramanujan and of G.H. Hardy
[17, Equation (2)], some of them being analogues in rational and number fields. See
[20, 21, 22, 23, 26]. In [24, 27], he used Fourier’s integral theorem to obtain expressions
for the Riemann Ξ-function, a method also enunciated by Ramanujan [32]. Around
the same time, W.L. Ferrar [14] also worked on transformation formulas of the above
kind.
As can be seen from (1.1), the general form of the integrals giving rise to formulas
of the type F (α) = F (β) where αβ = 1 is
Z ∞    
t t
f Ξ cos µt dt,
0 2 2
for µ real and f (t) = φ(it)φ(−it), where φ is analytic in t as a function of a real variable.
This integral is mentioned in [37, p. 35]. Similarly, from (1.2) and (1.3), it is clear that
the general form of the integrals giving rise to identities of the type F (z, α) = F (z, β)
4 ATUL DIXIT

where αβ = 1 is
∞      
t t + iz t − iz
Z
f z, Ξ Ξ cos µt dt, (1.4)
0 2 2 2
for µ real and f (z, t) = φ(z, it)φ(z, −it), where φ is analytic in t as a function of a real
variable as well as analytic in z in some complex domain. An integral of this kind was
first introduced by Ramanujan [32].
In this article, we find character analogues of all of the above-mentioned theorems.
The character analogue of Guinand’s, and hence Koshliakov’s formula given here differs
from the ones established in [7]. Throughout this article, we will be concerned with
the principal branch of the logarithm. Also, we work only with a primitive, non-
principal Dirichlet character χ modulo q. It is easy to see that its conjugate character
χ is also a primitive, non-principal character modulo q and χ is even (odd) if and
only if χ isPeven (resp. odd). Let L(s, χ) denote the Dirichlet L-function defined by
∞ s
L(s, χ) = n=1 χ(n)/n for Re s > 1. This series converges conditionally for 0 <
Re s < 1. Also, it can be analytically continued to an entire function of s. Let
G(χ) := G(1, χ), where G(n, χ) is the Gauss sum defined by
q
X
G(n, χ) := χ(m)e2πimn/q .
m=1

We know that [2, p. 168]

|G(χ)|2 = q (1.5)
and it is easy to see that
(
G(χ), for χ even,
G(χ) = (1.6)
−G(χ), for χ odd.
Define b as follows: (
0, χ(−1) = 1,
b= (1.7)
1, χ(−1) = −1.
Then the function ξ(s, χ) is defined by
 −(s+b)/2  
π s+b
ξ(s, χ) := Γ L(s, χ), (1.8)
q 2
and the analogue of the Riemann Ξ-function for Dirichlet characters is then defined as
 
1
Ξ(t, χ) := ξ + it, χ . (1.9)
2
L-functions satisfy the functional equation [2, p. 263]
q s−1 Γ(s) −πis/2
+ χ(−1)eπis/2 G(χ)L(s, χ),


L(1 − s, χ) = s
e (1.10)
(2π)
which can be rephrased in terms of ξ(s, χ) as [10]
ξ(1 − s, χ) = ǫ(χ)ξ(s, χ), (1.11)
 CHARACTER ANALOGUES 5

where ǫ(χ) = ib q 1/2 /G(χ). By (1.5), |ǫ(χ)| = 1. Next, we note Stirling’s formula in a
vertical strip α ≤ σ ≤ β, s = σ + it, namely,
  
1
σ−
1 1
− π|t| 1
|Γ(s)| = (2π) 2 |t| 2 e 2 1+O (1.12)
|t|
uniformly as |t| → ∞. Now using (1.10) and the fact [10, p. 82] that |L(s, χ)| = O(q|t|)
for Re s ≥ 1/2, we can easily see that for Re s ≥ −δ, δ > 0, we have
 3 3

+δ +δ
L(s, χ) = O q |t|
2 2 . (1.13)

We will subsequently use this result.

Transformation formulas involving Dirichlet characters of the form
∞
X ∞
X
χ(n)f (n) = χ(n)g(n),
n=1 n=1

where
→∞  
2G(χ) 2πxt
 Z


 q cos f (t) dt, for χ(−1) = 1,
0 q
g(x) =
−2iG(χ) →∞
 
2πxt
Z


 sin f (t) dt, for χ(−1) = −1,
q 0 q
were considered by Guinand [15, Theorems 4–5], though he did not give any particular
examples. Here, we derive a character analogue of the integral in (1.4). Its general
form is Z ∞        
t t + iz t − iz 1
f z, Ξ ,χ Ξ , χ cos t log α dt, (1.14)
0 2 2 2 2
where f is an even function of both the variables z and t. These integrals give rise
to transformation formulas of the type F (z, α, χ) = F (−z, β, χ) = F (−z, α, χ) =
F (z, β, χ). Then via Fourier’s integral theorem, one may be able to obtain integral
representations for Ξ ((t + iz)/2, χ) Ξ ((t − iz)/2, χ) which are of independent interest.
The character analogue of Theorem 1.3 is as follows.
Theorem 1.5. Let −1 < Re z < 1 and let χ denote a primitive, non-principal charac-
ter modulo q. Let the number b be defined as in (1.7). Let Kν (z), d(n) and γ be defined
as before and let α and β be positive numbers such that αβ = 1. If
∞ X   
b+ 21
X
− z2 +b 2 z 2πnα
F (z, α, χ) := α χ(n)n χ (d)d K− 2z ,
n=1
q
d|n

then
F (z, α, χ) = F (−z, β, χ) = F (−z, α, χ) = F (z, β, χ)
Z ∞      
1 t + iz t − iz 1
= Ξ ,χ Ξ , χ cos t log α dt. (1.15)
8π 0 2 2 2
6 ATUL DIXIT

Define ψ(a, χ) by
∞
X χ(n)
ψ(a, χ) = − , (1.16)
n=1
n + a
where a ∈ C is a non-negative integer. For a real character χ, this agrees with the
character analogue of the psi function obtained by the logarithmic differentiation of the
following Weierstrass product form of the character analogue of the gamma function
for real characters derived by Berndt [4]:
∞ 
−aL(1,χ)
Y a −χ(n) aχ(n)/n
Γ(a, χ) = e 1+ e .
n=1
n
The character analogue of the Hurwitz zeta function ζ(z, a) is given by [3, Ex. 3.2]
∞
X χ(n)
L(z, a, χ) = z
, (1.17)
n=1
(n + a)
valid for Re z > 0, and provided a ∈ C is a non-negative integer. The above character
analogue of the Hurwitz zeta function can also be obtained as the special case when
x = 0 of the function L(z, x, a, χ) defined in [4] by
∞ ′
X
L(z, x, a, χ) := e2πinx/k χ(n)(n + a)−z ,
n=0
where the prime indicates that the term corresponding to n = −a is omitted if a
is a negative integer and χ(a) 6= 0. As shown in [4], L(z, x, a, χ) converges for Re
z > 0 if x is not an integer, or if x is an integer and gcd(x, k) > 1. If x is an integer
and gcd(x, k) = 1, the series converges for Re z > 1. For mean value properties of
L(z, a, χ) and asymptotic formulas, see the recent paper [29]. The character analogues
of Theorem 1.2 are given below.
Theorem 1.6. Let χ denote an even, primitive, non-principal character modulo q. Let
−1 < Re z < 1 and let L(z, a, χ) be defined as in (1.17). Define T (z, α, χ) by
αz/2 q z/2 Γ(z + 1)
T (z, α, χ) := , (1.18)
2z π z/2 G(χ)
and Ω(z, t) by
   
2 2 −z − 1 + it −z − 1 − it
Ω(z, t) := ((z + 1) + t )Γ Γ
4 4
   
2 2 z − 1 + it z − 1 − it
+ ((z − 1) + t )Γ Γ . (1.19)
4 4
Then if α and β are positive numbers such that αβ = 1,
∞ ∞
√
 X   X  
α T (z, α, χ) χ(n)L z + 1, nα, χ + T (−z, α, χ) χ(n)L − z + 1, nα, χ
n=1 n=1
 ∞
X   ∞
X  
p
= β T (−z, β, χ) χ(n)L − z + 1, nβ, χ + T (z, β, χ) χ(n)L z + 1, nβ, χ
n=1 n=1
 CHARACTER ANALOGUES 7

∞      
1 t + iz t − iz 1
Z
= Ω(z, t)Ξ ,χ Ξ , χ cos t log α dt. (1.20)
64π 3/2 q 0 2 2 2
Theorem 1.7. Let χ denote an odd, primitive, non-principal character modulo q. Let
−1 < Re z < 1 and let L(z, a, χ) be defined as in (1.17) and let T (z, α, χ) be defined
as in (1.18). Define Λ(z, t) by
       
z + 1 + it z + 1 − it −z + 1 + it −z + 1 − it
Λ(z, t) := Γ Γ +Γ Γ . (1.21)
4 4 4 4
Then if α and β are positive numbers such that αβ = 1,
∞ ∞
√
 X   X  
α T (z, α, χ) χ(n)L z + 1, nα, χ + T (−z, α, χ) χ(n)L − z + 1, nα, χ
n=1 n=1
 ∞
X   ∞
X  
p
= β T (−z, β, χ) χ(n)L − z + 1, nβ, χ + T (z, β, χ) χ(n)L z + 1, nβ, χ
n=1 n=1
∞      
1 t + iz t − iz 1
Z
= Λ(z, t)Ξ ,χ Ξ , χ cos t log α dt. (1.22)
4π 1/2 iq 2 0 2 2 2
In [18, p. 156, Section 2.5], Hardy and Littlewood discuss the following interesting
identity suggested to them by work of Ramanujan.

Theorem 1.8. Let µ(n) denote the Möbius function. P Let 1−ρ α and

′ β be
ρtwo positive
numbers such that αβ = 1. Assume that the series ρ Γ 2 /ζ (ρ) a converges,
where ρ denotes a non-trivial zero of the Riemann zeta function and a denotes a positive
real number, and that the non-trivial zeros of ζ(s) are simple. Then
∞ X Γ 1−ρ ρ


√ X µ(n) −πα2 /n2 1
α e − √ √ ′
2
π 2 αρ
n=1
n 4 π α ρ ζ (ρ)
∞ X Γ 1−ρ ρ


p X µ(n) −πβ 2 /n2 1
= β e − √ √ ′
2
π 2 β ρ. (1.23)
n=1
n 4 π β ρ
ζ (ρ)

The original formulation of the above identity is slightly different in [18] but can
readily be seen to be equivalent to (1.23). See also [5, p. 470], [31, p. 143] and [37,
p. 219, Section 9.8] for discussions on this identity. Based on certain assumptions, the
character analogues of (1.23) for even and odd primitive Dirichlet characters, which
furnish two examples of transformation formulas of the form F (α, χ) = F (β, χ), are
derived here and are as follows.

Theorem 1.9. Let χ be an odd, primitive character modulo q, and let α and β be
P ρ/2 ρ
two positive numbers such that αβ = 1. Assume that the series ρ π qαρ/2Γ((2−ρ)/2)
L′ (ρ,χ)
and
P πρ/2 β ρ Γ((2−ρ)/2)
ρ q ρ/2 L′ (ρ,χ)
converge, where ρ denotes a non-trivial zero of L(s, χ) and L(s, χ)
respectively, and that the non-trivial zeros of the associated Dirichlet L-functions are
8 ATUL DIXIT

simple. Then
ρ !
∞ X Γ 2−ρ  π  2


√ p X χ(n)µ(n) − πα 2
q
α α G(χ) 2
e qn2 − 2
2
′ (ρ, χ)
αρ
n=1
n 4πα ρ
L q
∞ ρ !
X Γ 2−ρ  π  2


p p X χ(n)µ(n) − πβ 2
q
= β β G(χ) 2
e qn2 − 2
2
′ (ρ, χ)
βρ . (1.24)
n=1
n 4πβ ρ
L q

Theorem 1.10. Let χ be an even, primitive character modulo q, and let α and β be
P ρ/2 ρ
two positive numbers such that αβ = 1. Assume that the series ρ π qαρ/2Γ((2−ρ)/2) L′ (ρ,χ)
and
P πρ/2 β ρ Γ((2−ρ)/2)
ρ q ρ/2 L′ (ρ,χ)
converge, where ρ denotes a non-trivial zero of L(s, χ) and L(s, χ)
respectively, and that the non-trivial zeros of the associated Dirichlet L-functions are
simple. Then
∞ √ X 1−ρ   ρ2

!
√ p X χ(n)µ(n) − πα2
q Γ π
α G(χ) e qn2 − √ 2
′ (ρ, χ)
αρ
n=1
n 4 πα ρ
L q
∞ √ X Γ 1−ρ  π  2

ρ !
p p X χ(n)µ(n) − πβ 2
q
= β G(χ) e qn2 − √ ′
2
βρ . (1.25)
n=1
n 4 πβ ρ
L (ρ, χ) q

This paper is organized as follows. In Section 2, we give a complex integral repre-

sentation of (1.14) that is used in subsequent sections. In Section 3, we prove Theorem
1.5. Then in Section 4, we compute the inverse Mellin transforms and asymptotic
expansions of certain functions which are subsequently used in Section 5. Section 5 is
devoted to proofs of Theorems 1.6 and 1.7. Character analogues of Ramanujan’s trans-
formation formula (Theorem 1.1) are derived as special cases of these theorems. We
conclude this section with some curious results on certain double series being always
real. In Section 6, we present proofs of Theorems 1.9 and 1.10. Finally we conclude
with some open problems in Section 7.

2. A complex integral representation of (1.14)

In this section, we give a formal way of transforming an integral involving a character
analogue of Riemann’s Ξ-function into an equivalent complex integral which allows us
to use residue calculus and Mellin transform techniques for its evaluation.

Theorem 2.1. Let

φ(z, it)φ(z, −it) + φ(−z, it)φ(−z, −it)
f (z, t) = , (2.1)
2
where φ is analytic in t as a function of a real variable and analytic in z in some
complex domain. Let y = eµ with µ real. Then, under the assumption that the integral
 CHARACTER ANALOGUES 9

on the left side below converges,

Z ∞    
iz iz
f (z, t)Ξ t + , χ Ξ t − , χ cos µt dt
0 2 2
Z 1 +i∞         
1 2 1 1 1 1
= √ φ z, s − φ z, − s + φ −z, s − φ −z, − s
4i y 12 −i∞ 2 2 2 2
 z   z 
× ξ s − , χ ξ s + , χ y s ds. (2.2)
2 2
Proof. Let
Z ∞    
iz iz
I(z, µ, χ) := f (z, t)Ξ t + , χ Ξ t − , χ cos µt dt.
0 2 2
Then
Z ∞    
1 iz iz
I(z, µ, χ) = f (z, t)Ξ t + , χ Ξ t − , χ y it dt
2 0 2 2
Z ∞     
iz iz −it
+ f (z, t)Ξ t + , χ Ξ t − , χ y dt
0 2 2
Z ∞    
1 iz iz
= f (z, t)Ξ t + , χ Ξ t − , χ y it dt
2 0 2 2
Z 0     
iz iz it
+ f (z, −t)Ξ −t + , χ Ξ −t − , χ y dt .
−∞ 2 2
(2.3)
However, using (1.11), we readily see that
       
iz 1 z 1 z iz
Ξ −t + , χ = ξ − it − , χ = ǫ(χ)ξ + it + , χ = ǫ(χ)Ξ t − , χ ,
2 2 2 2 2 2
       
iz 1 z −1 1 z −1 iz
Ξ −t − , χ = ξ − it + , χ = (ǫ(χ)) ξ + it − , χ = (ǫ(χ)) Ξ t + , χ ,
2 2 2 2 2 2
so that
       
iz iz iz iz
Ξ −t + , χ Ξ −t − , χ = Ξ t + , χ Ξ t − , χ . (2.4)
2 2 2 2
Thus from (2.3), (2.4) and the fact that f is an even function of t, we obtain
1 ∞
   
iz iz
Z
I(z, µ, χ) = f (z, t)Ξ t + , χ Ξ t − , χ y it dt
2 −∞ 2 2
Z 1 +i∞         
1 2 1 1 1 1
= √ φ z, s − φ z, − s + φ −z, s − φ −z, − s
4i y 12 −i∞ 2 2 2 2
 z   z 
× ξ s − , χ ξ s + , χ y s ds,
2 2
where in the penultimate line, we made the change of variable s = 21 + it. 
10 ATUL DIXIT

For our purpose here, we replace µ by 2µ in (2.2) and then t by t/2 on the left-hand
side of (2.2). Thus with y = e2µ , we find that
Z ∞      
t t + iz t − iz
f z, Ξ ,χ Ξ , χ cos µt dt
0 2 2 2
Z 1 +i∞         
1 2 1 1 1 1
= √ φ z, s − φ z, − s + φ −z, s − φ −z, − s
2i y 12 −i∞ 2 2 2 2
 z   z 
× ξ s − , χ ξ s + , χ y s ds. (2.5)
2 2
It is this equation with which we will be working throughout this paper.

3. Character analogues of the extended version of Guinand’s formula

We require the following lemma.
Lemma 3.1. For Re s > 1 and Re (s − η) > 1,
∞
X χ(n) X 2
L(s, χ)L(s − η, χ) = s
χ (d)dη . (3.1)
n=1
n
d|n

Proof. Since the Dirichlet series for both the L-functions converge absolutely under
the given hypotheses, using [2, Theorem 11.5], we see that
∞ ∞
X χ(n) X χ(k)
L(s, χ)L(s − η, χ) =
n=1
ns k=1 k s−η
∞
X 1 X
= s
χ(n)χ(k)k η
j=1
j nk=j
∞
X χ(j) X
= χ2 (k)k η ,
j=1
js
nk=j

since χ(k)χ(k) = 1. 
Proof of Theorem 1.5. First assume that χ is even. Let φ(z, s) ≡ 1. Then from (2.1),
we see that f (z, t) ≡ 1. Using (1.9), (1.8), (1.12) and (1.13), we find that the integral
Z ∞    
t + iz t − iz
M(z, µ, χ) := Ξ ,χ Ξ , χ cos µt dt
0 2 2
does converge. Using (2.5), we observe that
Z 1 +i∞ 
1 2 z   z 
M(z, µ, χ) = √ ξ s − , χ ξ s + , χ y s ds
i y 21 −i∞ 2 2
Z 1 +i∞   −s
1 2 s z  s z   z   z  π
= √ Γ − Γ + L s− ,χ L s+ ,χ ds.
i y 21 −i∞ 2 4 2 4 2 2 qy
(3.2)
 CHARACTER ANALOGUES 11

Since
Re s = 1/2 and −1 < Re z < 1, we have 0 < Re s − 2z < 1 and 0 < Re

s + z2
< 1. Now replace
s by s − z2 and let η = −z in Lemma 3.1. Then, for Re

s − z2 > 1 and Re s + z2 > 1,
∞
 z   z  X χ(n) X 2
L s− ,χ L s+ ,χ = s− z2
χ (d)d−z . (3.3)
2 2 n=1
n d|n

We wish to shift the line of integration from Re s = 1/2 to Re s = 3/2 in order to be

able to use (3.3) in (3.2). Consider a positively oriented rectangular contour formed
by [ 12 + iT, 21 − iT ], [ 21 − iT, 32 − iT ], [ 23 − iT, 32 + iT ] and [ 23 + iT, 21 + iT ], where T is any
positive real number. The integrand on the extreme right-hand side of (3.2) does not
have any pole inside the contour. Also as T → ∞, the integrals along the horizontal
segments [ 12 − iT, 32 − iT ] and [ 23 + iT, 12 + iT ] tend to zero, which can be seen by using
(1.12). Hence employing residue theorem, letting T → ∞, using (3.3) in (3.2), and
interchanging the order of summation and integration because of absolute convergence,
we observe that
∞ X  Z 3 +i∞   −s
1 X z/2 2 −z
2 s z   s z  nπ
M(z, µ, χ) = √ χ(n)n χ (d)d Γ − Γ + ds.
i y n=1 3
−i∞ 2 4 2 4 qy
d|n 2

(3.4)

But from [30, p. 115, formula 11.1], for c = Re s > ± Re ν,

c+i∞
1 ν   s ν  −s
s
Z
s−2 −s
2 w Γ − Γ + x ds = Kν (wx). (3.5)
2πi c−i∞ 2 2 2 2

Hence using (3.5) with c = 3/2, ν = z/2, w = 2 and x = nπ/qy in (3.4), we find that
∞ X   
8π X z/2 2 −z 2πn
M(z, µ, χ) = √ χ(n)n χ (d)d K 2z . (3.6)
y n=1 qy
d|n

Now let µ = 21 log α in (3.6) so that y = e2µ implies that y = α. Then using the fact
that αβ = 1, we deduce that
Z ∞      
1 t + iz t − iz 1
Ξ ,χ Ξ , χ cos t log α dt
8π 0 2 2 2
∞ X   
p X z/2 2 −z 2πnβ
= β χ(n)n χ (d)d K z2 . (3.7)
n=1
q
d|n

Next, observing that replacing α by β and/or replacing simultaneously χ by χ and z

by −z in (3.7) leaves the integral on the left-hand side invariant, we obtain (1.15).
Now consider the case when χ is odd. Again the convergence of the integral M(z, µ, χ)
can be seen from (1.12) and (1.13). Following similar steps above as in the case of even
12 ATUL DIXIT

χ, and using the definition of ξ(s, χ) from (1.8) for χ odd, we find that
∞ Z 3 +i∞      −s
q X z/2
X
2 −z
2 s z 1 s z 1 nπ
M(z, µ, χ) = √ χ(n)n χ (d)d Γ − + Γ + + ds.
iπ y n=1 3
−i∞ 2 4 2 2 4 2 qy
d|n 2

(3.8)
Now replacing s by s + 1 in (3.5), we find that for c = Re s > ± Re ν − 1,
Z c+i∞    
1 s−1 −s−1 s+1 ν s+1 ν
2 w Γ − Γ + x−s ds = xKν (wx). (3.9)
2πi c−i∞ 2 2 2 2
Then using (3.9) with c = 3/2, ν = 0, w = 2 and x = nπ/qy in (3.8), we see that
∞ X   
8π X z
+1 2 −z 2πn
M(z, µ, χ) = 3/2 χ(n)n 2 χ (d)d Kz/2 . (3.10)
y n=1 qy
d|n

Now let µ = 21 log α in (3.10) so that y = e2µ implies that y = α. Then using the fact
that αβ = 1, we deduce that
Z ∞      
1 t + iz t − iz 1
Ξ ,χ Ξ , χ cos t log α dt
8π 0 2 2 2
∞ X   
3/2
X z
+1 2 −z 2πnβ
=β χ(n)n 2 χ (d)d Kz/2 . (3.11)
n=1
q
d|n

Next, observing that replacing α by β and/or replacing simultaneously χ by χ and z

by −z in (3.11) leaves the integral on the left-hand side invariant, we obtain (1.15). 
Remark. Letting z → 0 in Theorem 1.5 gives a new character analogue of the ex-
tended version of Koshliakov’s formula, i.e., Theorem 1.4.

When χ is real, Theorem 1.5 reduces to the following corollary.

Corollary 3.2. Let −1 < Re z < 1 and let χ denote a real, primitive, non-principal
character modulo q. Let the number b be defined as in (1.7). If
∞  
b+ 12
X
− z2 +b 2πnα
F (z, α, χ) = α χ(n)n σz (n)K− z2 ,
n=1
q

then
F (z, α, χ) = F (−z, β, χ) = F (−z, α, χ) = F (z, β, χ)
Z ∞      
1 t + iz t − iz 1
= Ξ ,χ Ξ , χ cos t log α dt.
8π 0 2 2 2
The above corollary (without the integrals) is equivalent to the special cases, when χ
is real, of the character analogues of Guinand’s formula established in [7] (see Theorems
3.1 and 4.1).
 CHARACTER ANALOGUES 13

4. Inverse Mellin transforms and asymptotic expansions of certain

functions
In this section, we evaluate inverse Mellin transforms of some functions and asymp-
totic expansions of certain other functions all of which are subsequently used in the
later sections.
Lemma 4.1. For a primitive, non-principal character χ, let ψ(a, χ) be defined as in
(1.16). Then for 0 < c = Re s < 1 and x ∈ R\Z<0 ,
Z c+i∞
1 L(1 − s, χ) −s 1
x ds = − ψ(x, χ). (4.1)
2πi c−i∞ sin πs π
Proof. We will prove the result for even characters only. The case when χ is odd
can be proved similarly. We first assume |x| < 1 and later extend it to any real
x ∈ R\Z<0 by analytic continuation. Let 0 < c < 1. Consider a positively oriented
rectangular contour formed by [c − iT, c + iT ], [c + iT, −M + iT ], [−M + iT, −M − iT ]
and [−M − iT, c − iT ], where T is any positive real number such that T > ln2π2 and
M = n − 1/2 where n is a positive integer. Let s = σ + it. Among the poles of the
function (L(1 − s, χ)/ sin (πs)) x−s , the only ones that contribute are the poles at the
non-positive integers. Let Ra denote the residue of the function (L(1 −s, χ)/ sin πs)x−s
at a. Then,
sL(1 − s, χ) −s 1
R0 = lim x = L(1, χ). (4.2)
s→0 sin πs π
and
(s + m)L(1 − s, χ) −s (−1)m
R−m = lim x = L(1 + m, χ)xm . (4.3)
s→−m sin πs π
From (4.2), (4.3) and the residue theorem, we have
Z c+iT Z −M +iT Z −M −iT Z c−iT 
L(1 − s, χ) −s
+ + + x ds
c−iT c+iT −M +iT −M −iT sin πs
!
1 X (−1)m
= 2πi L(1, χ) + L(1 + m, χ)xm . (4.4)
π 0<m<M
π

We first estimate the integrals along the upper and lower horizontal segments. Using
(1.13), one finds that for −M ≤ σ ≤ c,
L(1 − σ ± iT, χ) = O q c+1/2 T c+1/2 .


(4.5)
ln 2
Since T > 2π
, on the upper horizontal segment, we have
1 2eπis
= 2πis < 4e−πT . (4.6)
sin πs e −1
Similarly, on the lower horizontal segment,
1 2e−πis
= < 4e−πT .
sin πs 1 − e−2πis
14 ATUL DIXIT

Since |x| < 1, from (4.5) and (4.6),

Z −M +iT
L(1 − s, χ) −s
x ds ≤ K1 (c + M)|x|−c · 4e−πT q c+1/2 T c+1/2 ,
c+iT sin πs
where K1 is some constant. Therefore,
Z −M +i∞
L(1 − s, χ) −s
x ds = 0. (4.7)
c+i∞ sin πs
Similarly,
c−i∞
L(1 − s, χ) −s
Z
x ds = 0. (4.8)
−M −i∞ sin πs
Then from (4.4), (4.7) and (4.8), we find that
Z c+i∞ Z −M −i∞  !
L(1 − s, χ) −s 1 X (−1)m
+ x ds = 2πi L(1, χ) + L(1 + m, χ)xm .
c−i∞ −M +i∞ sin πs π 0<m<M
π
(4.9)
R −M −i∞ −s
It remains to examine −M +i∞ (L(1 − s, χ)/ sin πs)x ds. Since M = n − 1/2, we have
| sin π(−M + it)| = | cosh πt| ≥ 1 and L(1 + M ± it) = O(1) as 1 + M > 1. Thus,
Z −M +i∞ Z ∞
L(1 − s, χ) −s L(1 + M − it, χ) M −it
x ds = i x dt
−M −i∞ sin πs −∞ sin π(−M + it)
Z 1 Z ∞ Z −1
M M −π|t| M
O e−π|t| dt



= |x| O(1) dt + |x| O e dt + |x|
−1 1 −∞
M
= O(|x| ),
as |t| → ∞. Since |x| < 1,
−M −i∞
L(1 − s, χ) −s
Z
lim x ds = 0. (4.10)
M →∞ −M +i∞ sin πs
From (4.9) and (4.10), we see that
Z c+i∞ ∞
!
1 L(1 − s, χ) −s 1 X (−1)m
x ds = L(1, χ) + L(1 + m, χ)xm
2πi c−i∞ sin πs π m=1
π
∞ ∞  m !
−1 X χ(k) X −x
= −L(1, χ) −
π k=1
k m=1 k
∞  !
−1 X 1 1
= −L(1, χ) − χ(k) −
π x+k k
k=1
−1
= ψ(x, χ). (4.11)
π
Since both sides of (4.1) are analytic for any x ∈ R\Z<0 , the result follows by analytic
continuation. 
 CHARACTER ANALOGUES 15

Lemma 4.2. Let z ∈ C be fixed such that −1 < Re z < 1. For a primitive, non-
principal character χ, let L(z, a, χ) be defined as in (1.17). Then for − 21 Re z < c =
Re s < 12 Re z and x ∈ R\Z<0 ,
Z c+i∞ 
1 z  z  z 
Γ s+ Γ 1−s+ L 1 − s + , χ x−s ds = xz/2 Γ(z+1)L(z+1, x, χ).
2πi c−i∞ 2 2 2
(4.12)
Proof. We prove the result only for even characters. The case for odd characters
can be proved similarly. We first assume |x| < 1 and later extend it to any real
x ∈ R\Z<0 by analytic continuation. Let − 21 Re z < c = Re s < 12 Re z. Consider
a positively oriented rectangular contour formed by [c − iT, c + iT ], [c + iT, −M +
iT ], [−M + iT, −M − iT ] and [−M − iT, c − iT ], where T is some positive real number
and M = n − 1/2, where n is a positive integer. Let s = σ + it. Among the poles of the
function Γ(s + z/2)Γ(1 − s + z/2)L(1 − s + z/2, χ)x−s , the only ones that contribute
are the poles at s = −z/2 − m, m ≥ 0. Let Ra denote the residue of the function
Γ(s + z/2)Γ(1 − s + z/2)L(1 − s + z/2, χ)x−s at a. Then for m ≥ 0,
 z   z  z  z  −s
R− z2 −m = lim s + + m Γ s + Γ 1 − s + L 1 − s + ,χ x
s→− 2z −m 2 2 2 2
(−1)m
= Γ (1 + z + m) L (1 + z + m, χ) xz/2+m . (4.13)
m!
From (4.13) and the residue theorem, we have
Z c+iT Z −M +iT Z −M −iT Z c−iT  
z  z  z  −s
+ + + Γ s+ Γ 1−s+ L 1 − s + , χ x ds
c−iT c+iT −M +iT −M −iT 2 2 2
X (−1)m
= 2πixz/2 Γ (1 + z + m) L (1 + z + m, χ) xm . (4.14)
0≤m<M
m!
We now estimate the integral along the upper horizontal segment. From (4.5), we
easily see that for −M ≤ σ ≤ c, i.e., −M− Re z2 ≤ σ− Re z2 ≤ c− Re z2 ,
z c−Re z2 + 12
 
  z  z  c−Re z2 + 12

L 1 − σ − Re − i T − Im , χ = O q T − Im . (4.15)
2 2 2
By (1.12), we observe that
 z √ − π2 |T +Im z2 | z σ+Re z2 − 21
Γ s+ ∼ 2πe · T + Im , (4.16)
2 2
and  z √ z −σ+Re z2 + 21
∼ 2πe− 2 |T −Im 2 | · T − Im
π z
Γ 1−s+ , (4.17)
2 2
Since |x| < 1, from (4.15), (4.16) and (4.17), we deduce that
Z −M +iT
Γ(s + z/2)Γ(1 − s + z/2)L(1 − s + z/2, χ)x−s ds
c+iT
z σ+Re z2 − 21 z −σ+Re z2 + 21
−c − π2 (|T +Im z2 |+|T −Im z2 |)
≤ 2πK3 (c + M)|x| e T + Im T − Im ,
2 2
16 ATUL DIXIT

where K3 is a constant. Hence,

Z −M +i∞
Γ(s + z/2)Γ(1 − s + z/2)L(1 − s + z/2, χ)x−s ds = 0. (4.18)
c+i∞

Similarly for the integral along the lower horizontal segment, using (4.16), (4.17) and
the fact that
z c−Re z2 + 12
 
  z  z  c−Re z2 + 12

L 1 − σ − Re + i T + Im , χ = O q T + Im ,
2 2 2
we observe that
Z c−i∞
Γ(s + z/2)Γ(1 − s + z/2)L(1 − s + z/2, χ)x−s ds = 0. (4.19)
−M −i∞

Hence from (4.14), (4.18) and (4.19), it is clear that

Z c+i∞ Z −M −i∞ 
+ Γ(s + z/2)Γ(1 − s + z/2)L(1 − s + z/2, χ)x−s ds
c−i∞ −M +i∞
X (−1)m
= 2πixz/2 Γ (1 + z + m) L (1 + z + m, χ) xm . (4.20)
0≤m<M
m!
R −M −i∞
It remains to evaluate −M +i∞ Γ(s + z/2)Γ(1 − s + z/2)L(1 − s + z/2, χ)x−s ds. In [9],
we find that as |t| → ∞,
Γ (−M + it) = O |t|−M −1/2 e−π|t|/2 .

Hence as T → ∞,
−M +Re z2 − 21
 
 z z − π2 |T +Im z2 |
Γ −M + it + = O T + Im e . (4.21)
2 2
Again by (1.12), as T → ∞,
!!
 z √ z M +Re z2 + 12 1
= 2πe− 2 |T −Im 2 | · T − Im
π z
Γ 1 + M − it + 1+O .
2 2 T − Im z2
(4.22)
Also, L 1 + M − it + 2z , χ is bounded as Re 1 + M − it + z2 > 1. Hence,

Z −M −i∞ 
z  z  z 
Γ s+ Γ 1−s+ L 1 − s + , χ x−s ds
−M +i∞ 2 2 2
Z ∞ 
z   z   z  M −it
= i Γ −M + it + Γ 1 + M − it + L 1 + M − it + , χ x dt
−∞ 2 2 2
Z 1 Z ±∞ 
z −M +Re z2 − 12 z M +Re z2 + 12 − π2 (|T +Im z2 |+|T −Im z2 |)

M M
= |x| O(1) dt + |x| O T + Im T − Im e dt
−1 1 2 2
= O |x|M ,


 CHARACTER ANALOGUES 17

as T → ∞. Since |x| < 1,

Z −M −i∞ 
z  z  z 
lim Γ s+ Γ 1−s+ L 1 − s + , χ x−s ds = 0. (4.23)
M →∞ −M +i∞ 2 2 2
From (4.20), (4.23), we finally deduce that
Z c+i∞ 
1 z  z  z  −s
Γ s+ Γ 1−s+ L 1 − s + , χ x ds
2πi c−i∞ 2 2 2
∞
X (−1)m
= xz/2 Γ (1 + z + m) L (1 + z + m, χ) xz/2+m
m=0
m!
∞ ∞
X (−1)m Γ (1 + z + m) X χ(k) m
= xz/2 Γ(z + 1) z+m+1
x
m=0
m! Γ(1 + z) k=1
k
∞ ∞  m
z/2
X χ(k) X Γ(1 + z + m) −x
= x Γ(z + 1)
k=1
k z+1 m=0 m!Γ(1 + z) k
∞
z/2
X χ(k)  x −z−1
= x Γ(z + 1) 1 +
k z+1 k
k=1
∞
X χ(k)
= xz/2 Γ(z + 1)
(k + x)z+1
k=1
z/2
=x Γ(z + 1)L(z + 1, x, χ),
where in the fourth step above, we have utilized binomial theorem since |x| < 1. Since
both sides of (4.12) are analytic for any x ∈ R\Z<0 , the result follows by analytic
continuation. 
For j ≥ 1, the generalized Bernoulli numbers Bj (χ) are given by [4, p. 426]
2(−1)j−1 G(χ)(2j)!
B2j (χ) = L(2j, χ),
k(2π/k)2j
for χ even, and by
2(−1)j−1 iG(χ)(2j − 1)!
B2j−1 (χ) = L(2j − 1, χ),
k(2π/k)2j−1
for χ odd. Also it is known [4, p. 423, Corollary 3.4] that B2j−1 (χ) = 0 when χ is even
and B2j (χ) = 0 when χ is odd. Next, we give an asymptotic expansion of ψ(a, χ) as
|a| → ∞.
Lemma 4.3. For −π < arg a < π, as |a| → ∞,
∞
L(0, χ) X Bj (χ)
ψ(a, χ) ∼ − − χ(−1) .
a j=2
jaj

Proof. The proof is very much identical to Proposition 5.3 in [4, p. 435] and hence is
omitted. 
The asymptotic expansion of L(z, a, χ) as |a| → ∞ is given below.
18 ATUL DIXIT

Lemma 4.4. For Re z > 0 and −π < arg a < π, as |a| → ∞,

∞ Qj−2
X Bj (χ) m=0 (z + m)
L(z, a, χ) ∼ χ(−1) z+j−1
.
j=1
j!a

Proof. To derive this, one takes (4.3) and (4.4) in [4, p. 424] valid for χ even and odd
respectively, substitutes A = 0, B = N, r = 1 and f (u) = (u + a)−z , lets N → ∞ and
then performs repeated integration by parts on the prevalent integral. 

5. Character analogues of Theorem 1.2

In this section, we prove analogues of Theorem 1.2 for even and odd primitive charac-
ters. Then we give character analogues of Ramanujan’s transformation formula (The-
orem 1.1) as special cases.
Proof of Theorem 1.6. Using Lemma 4.4, one sees that the series involving the func-
−z−1 s


tions L(z, a, χ) in the theorem are convergent. Let φ(z, s) = (z + 1 + 2s)Γ 4
+ 2
.
t

1
Then from (2.1) and (1.19), we find that f z, 2 = 2 Ω(z, t). From (2.5), we have
Z ∞    
t + iz t − iz 1
Ω(z, t)Ξ ,χ Ξ , χ cos µt dt = √ (J(z, y, χ) + J(−z, y, χ)) ,
0 2 2 i y
(5.1)
where
1
Z
2
+i∞
J(z, y, χ) := U(z, s, y, χ) ds (5.2)
1
2
−i∞

with
  
z s 1 z s  z   z  s
U(z, s, y, χ) := (−z+2s)(−z+2−2s)Γ + − Γ − ξ s− ,χ ξ s+ ,χ y .
4 2 2 4 2 2 2
Using (1.12) and (1.13), one sees that indeed the integral on the left side of (5.1)
converges. We first simplify the integrand in (5.2). Using (1.8) with b = 0, and
then duplication formula [36, p. 46, Equation (3.4)] and reflection formula [36, p. 46,
Equation (3.5)] for Gamma function in the second equality below, we have
 −s     
π z s+1 z s n z s   s z o
U(z, s, y, χ) = 16 Γ + Γ + Γ − +1 Γ −
qy 4 2 4 2 4 2 2 4
 z   z 
×L s− ,χ L s+ ,χ
2 2
 −s √
π π  z π  z   z 
= 16 · s+ z −1 Γ s + · · L s − , χ L s + ,χ .
sin π 2s − z4



qy 2 2 2 2 2
(5.3)
Substituting (1.10) in the form
z
 z  (2π)s− 2 L(1 − s + 2z , χ)
L s − , χ = s− z −1
2q 2 G(χ)Γ s − z2 cos π2 s − 2z




2
 CHARACTER ANALOGUES 19

in (5.3) and then simplifying, we find that

32y s 2−z π (1−z)/2  z  z  z   z 

U(z, s, y, χ) = z Γ 1 − s + Γ s + L 1 − s + , χ L s + ,χ .
q − 2 −1 G(χ) 2 2 2 2
(5.4)
We wish to shift the line of integration from Re s = 1/2 to Re s = 3/2 in order
to evaluate (5.2), since then −1 < Re z < 1 implies
that Re (s + z/2) > 1, which
allows us to use the series representation of L s + z2 , χ . Consider a positively oriented
rectangular contour formed by [ 12 + iT, 21 − iT ], [ 21 − iT, 32 − iT ], [ 23 − iT, 23 + iT ] and
[ 32 + iT, 21 + iT ], where T is any positive real number. The integrand in (5.2) does
z


not have any pole inside the contour since
the pole of Γ 1 − s + 2 at s = 1 + z/2 is
z
cancelled by the zero of L 1 − s + 2 , χ there. Also as T → ∞, the integrals along
the horizontal segments [ 12 − iT, 32 − iT ] and [ 32 + iT, 12 + iT ] tend to zero, which can
be seen using (1.12). Employing residue theorem, letting T → ∞ and using (5.4), we
find that
3
+i∞
32 · 2−z π (1−z)/2 z  z  z   z 
Z
2

J(z, y, χ) = z Γ s+ Γ 1−s+ L 1 − s + , χ L s + , χ y s ds
q − 2 −1 G(χ) 3
2
−i∞ 2 2 2 2
∞ Z 3  −s
32 · 2−z π (1−z)/2 X χ(n) 2 +i∞  z  z  z  n
= z Γ s+ Γ 1−s+ L 1 − s+ ,χ ds.
q − 2 −1 G(χ) n=1
nz/2 32 −i∞ 2 2 2 y
(5.5)

Now, in order to use Lemma 4.2 to evaluate the integral in (5.5), we again have to shift
the line of integration from Re s > 3/2 to Re s = d, where − 21 Re z < d < 21 Re z.
Again, we do not encounter any pole in this process. Hence,
∞ ∞
64i2−z y −z/2π (3−z)/2 Γ(z + 1) X X χ(n)χ(k)
J(z, y, χ) = z . (5.6)
q − 2 −1 G(χ) n=1 k=1
(k + n/y) z+1

Since −1 < Re(z) < 1, the other integral, namely J(−z, y, χ), can be evaluated by
simply replacing z by −z and χ by χ in (5.6). Now (5.1), (5.6), (1.18) and the discussion
in the previous line give
Z ∞    
t + iz t − iz
Ω(z, t)Ξ ,χ Ξ , χ cos µt dt
0 2 2
∞ X ∞ ∞ X ∞
!
64π 3/2 q −1
X χ(n)χ(k) −1
X χ(n)χ(k)
= √ T (z, y , χ) + T (−z, y , χ) ,
y n=1 k=1
(k + n/y)z+1 n=1 k=1
(k + n/y)−z+1
(5.7)

where it is easy to see from the fact that −1 < Re z < 1, from the discussion just
preceding the statement of Theorem 1.6 and from Lemma 4.4 that both the double
series on the right-hand side of (5.7) converge.
20 ATUL DIXIT

Now let µ = 12 log α in (5.7) so that y = e2µ implies that y = α. Then using the fact
that αβ = 1 and using (1.17) in the second equality below, we deduce that
Z ∞    
t + iz t − iz
, χ cos 21 t log α dt


Ω(z, t)Ξ ,χ Ξ
0 2 2
∞ ∞ ∞ X ∞
!
p X X χ(n)χ(k) X χ(n)χ(k)
= 64π 3/2 q β T (z, β, χ) + T (−z, β, χ)
n=1 k=1
(k + nβ)z+1 n=1 k=1
(k + nβ)−z+1
∞ ∞
!
p X X
= 64π 3/2 q β T (z, β, χ) χ(n)L(z + 1, nβ, χ) + T (−z, β, χ) χ(n)L(−z + 1, nβ, χ) .
n=1 n=1

The integral on the extreme left-hand side above is invariant under the transformation
α → β or under the simultaneous application of the transformations α → β, χ → χ
and z → −z. Thus we obtain (1.20). 
Next we give an analogue of Ramanujan’s transformation formula (Theorem 1.1) for
even characters.
Corollary 5.1. For an even character χ, define P (α, χ) by
∞ X ∞
! ∞
!
√ X χ(n)χ(k) √ X
P (α, χ) = α Re G(χ) = − α Re G(χ) χ(n)ψ (nα, χ) ,
n=1 k=1
k + nα n=1

where ψ (a, χ) is defined in (1.16). Then we have

P (α, χ) = P (β, χ) = P (α, χ) = P (β, χ)
Z ∞          
1 2 −1 + it −1 − it t t 1
= (1 + t )Γ Γ Ξ ,χ Ξ , χ cos t log α dt.
64π 3/2 0 4 4 2 2 2
(5.8)
Proof. Using Lemma 4.3, we readily see that the double series in the definition of
P (α, χ) converges. Let z → 0 in (1.20). Then multiplying both sides by q and using
(1.6), we have
∞ X ∞ ∞ X ∞
!
√ X χ(n)χ(k) X χ(n)χ(k)
α G(χ) + G(χ)
n=1 k=1
k + nα n=1 k=1
k + nα
∞ X ∞ ∞ X ∞
!
p X χ(n)χ(k) X χ(n)χ(k)
= β G(χ) + G(χ)
n=1 k=1
k + nβ n=1 k=1
k + nβ
Z ∞          
1 2 −1 + it −1 − it t t 1
= (1 + t )Γ Γ Ξ ,χ Ξ , χ cos t log α dt.
32π 3/2 0 4 4 2 2 2
(5.9)
Each of the first two expressions in (5.9) can be written in two different ways as real
parts of a double series. Thus,
∞ X ∞
! ∞ X ∞
!
√ X χ(n)χ(k) √ X χ(n)χ(k)
α Re G(χ) = α Re G(χ)
n=1 k=1
k + nα n=1 k=1
k + nα
 CHARACTER ANALOGUES 21

∞ X ∞ ∞ X ∞
! !
p X χ(n)χ(k) p X χ(n)χ(k)
= β Re G(χ) = β Re G(χ)
n=1 k=1
k + nβ n=1 k=1
k + nβ
Z ∞          
1 2 −1 + it −1 − it t t 1
= (1 + t )Γ Γ Ξ ,χ Ξ , χ cos t log α dt.
64π 3/2 0 4 4 2 2 2
This implies (5.8). 
Moreover, if we start with the integral in Corollary 5.1, evaluate it using (2.5) with
z = 0 and make use of Lemma 4.1 when χ is even, we obtain the same result as in
Corollary 5.1, except that the function P (α, χ) is replaced by the function F (α, χ)
defined by
∞ X
∞ ∞
√ X χ(n)χ(k) √ X
F (α, χ) := αG(χ) = − αG(χ) χ(n)ψ (nα, χ) . (5.10)
n=1 k=1
k + nα n=1

It is then trivial to see that F (α, χ) = P (α, χ).

Theorem 1.7 can be analogously proved using Lemma 4.2
for χ odd. We just note
that there we have to take care of the pole of Γ 1 − s + 12 z in the integrands of two
separate integrals. However, in the calculations that follow later, the two residues turn
out to be additive inverses of each other and hence do not contribute anything.
The following is an analogue of Ramanujan’s transformation formula (Theorem 1.1)
for odd characters.
Corollary 5.2. For an odd character χ, define Q(α, χ) by
∞ X ∞
! ∞
!
√ X χ(n)χ(k) √ X
Q(α, χ) = α Im G(χ) = − α Im G(χ) χ(n)ψ (nα, χ) ,
n=1 k=1
k + nα n=1

where ψ (a, χ) is defined in (1.16). Then we have

Q(α, χ) = Q(β, χ) = Q(α, χ) = Q(β, χ)
Z ∞          
1 1 + it 1 − it t t 1
= 1/2 Γ Γ Ξ ,χ Ξ , χ cos t log α dt. (5.11)
4π q 0 4 4 2 2 2
Proof. Using Lemma 4.3, we find that the double series in the definition of Q(α, χ)
converges. Let z → 0 in Theorem 1.7. Multiplying both sides by −q and using (1.5)
and (1.6), we observe that
∞ X ∞ ∞ X ∞
!
√ X χ(n)χ(k) X χ(n)χ(k)
α G(χ) + G(χ)
n=1 k=1
k + nα n=1 k=1
k + nα
∞ X ∞ ∞ X ∞
!
p X χ(n)χ(k) X χ(n)χ(k)
= β G(χ) + G(χ)
n=1 k=1
k + nβ n=1 k=1
k + nβ
Z ∞          
i 1 + it 1 − it t t 1
= 1/2 Γ Γ Ξ ,χ Ξ , χ cos t log α dt. (5.12)
2π q 0 4 4 2 2 2
22 ATUL DIXIT

Now using (1.6) for odd characters to simplify (5.12), we see that
∞ X ∞
! ∞ X ∞
!
√ X χ(n)χ(k) √ X χ(n)χ(k)
2i α Im G(χ) = 2i α Im G(χ)
n=1 k=1
k + nα n=1 k=1
k + nα
∞ X ∞
! ∞ X ∞
!
p X χ(n)χ(k) p X χ(n)χ(k)
= 2i β Im G(χ) = 2i β Im G(χ)
n=1 k=1
k + nβ n=1 k=1
k + nβ
Z ∞          
i 1 + it 1 − it t t 1
= 1/2 Γ Γ Ξ ,χ Ξ , χ cos t log α dt.
2π q 0 4 4 2 2 2
This implies (5.11). 
If we now start with the integral in Corollary 5.2, evaluate it using (2.5) with z = 0 and
make use of Lemma 4.1 when χ is odd, we obtain the same result as in Corollary 5.2,
except that the function Q(α, χ) is replaced by −iF (α, χ), where F (α, χ) is defined in
(5.10). It is then trivial to see that F (α, χ) = iQ(α, χ).
We separately record the following theorem resulting from the discussion on the
previous line and the one succeeding Corollary 5.1.
Theorem 5.3. The sum F (α, χ) defined in (5.10) is real if χ is even and purely
imaginary if χ is odd.

6. Character analogues of the Ramanujan-Hardy-Littlewood

conjecture
In this section, we prove Theorems 1.9 and 1.10. We require Lemma 3.1 from [1]
which states that if χ is a primitive character of conductor N and k ≥ 2 is an integer
such that χ(−1) = (−1)k , then
(k − 2)!N k−2 G(χ)
L(k − 1, χ) = L′ (2 − k, χ). (6.1)
2k−1 π k−2ik−2
Proof of Theorem 1.9. From [28], we have for Re s > 1,
∞
X χ(n)µ(n) 1
= . (6.2)
n=1
ns L(s, χ)
Also, since for −1 < c = Re s < 0,
c+i∞
1
Z
−x
(1 − e ) = − Γ(s)x−s ds, (6.3)
2πi c−i∞

replacing s by s + 1, we find that for −2 < c < −1,

Z c+i∞
−x 1
(1 − e ) = − Γ(s + 1)x−s−1 ds. (6.4)
2πi c−i∞
Using (6.2) and (6.4), we observe that
∞ ∞
X χ(n)µ(n) − πα
2
1 X χ(n)µ(n) − πα
2

e 2n q = − (1 − e n2 q )

n=1
n2 L(2, χ) n=1 n2
 CHARACTER ANALOGUES 23

∞
c+i∞ X  2 −s
1 q χ(n)µ(n) πα
Z
= + 2 2 −2s
Γ(s + 1) ds
L(2, χ) 2π iα c−i∞ n=1 n q
Z c+i∞ −s
Γ(s + 1) πα2

1 q
= + ds, (6.5)
L(2, χ) 2π 2 iα2 c−i∞ L(−2s, χ) q
where in the second step above, we interchanged the order of summation and integration
because of absolute convergence. For χ odd, (1.10) can be put in the form
 −(2−s)/2   −(s+1)/2 
iq 1/2 π
 
π 2−s s+1
Γ L(1 − s, χ) = Γ L(s, χ).
q 2 G(χ) q 2
Hence,
 2s+ 1
Γ 12 − s


Γ(s + 1) G(χ) π 2
= 1/2 . (6.6)
L(−2s, χ) iq q L(2s + 1, χ)
Substituting (6.6) in (6.5), we observe that
∞ 2 −s
Γ 21 − s
Z c+i∞
 
X χ(n)µ(n) − πα 2
1 G(χ) qα
2
e nq =
2
− 3/2 2 ds. (6.7)
n=1
n L(2, χ) 2π α c−i∞ L(2s + 1, χ) π
We wish to shift the line of integration from Re s = c, −2 < c < −1, to Re s = λ,
where 12 < λ < 23 . Consider a positively oriented rectangular contour formed by
[c − iT, λ − iT ], [λ − iT, λ + iT ], [λ + iT, c + iT ] and [c + iT, c − iT ], where T is any
positive real number. Let ρ = δ + iγ denote a non-trivial zero of L(s, χ). Let T → ∞
through values such that |T − γ| > exp (−A1 γ/ log γ) for every ordinate γ of a zero of
L(s, χ). It is known [10, p. 102] that for t not coinciding with the ordinate γ of a zero,
and −1 ≤ σ ≤ 2,
′
L (s, χ) X 1
= + O (log (q(|t| + 2))) ,
L(s, χ) s−ρ
|t−γ|≤1

from which we can conclude that

X
log L(s, χ) = log(s − ρ) + O (log (q(|t| + 2))) . (6.8)
|t−γ|≤1

Taking real parts in (6.8) gives

X
log |L(s, χ)| = log |s − ρ| + O (log (q(|t| + 2)))
|t−γ|≤1
X
≥ log |t − γ| + O (log (q(|t| + 2))) .
|t−γ|≤1

(6.9)
Hence from (6.9), we have
X
log |L(σ + iT, χ)| ≥ − A1 γ/ log γ + O (log (q(|T | + 2)))
|T −γ|≤1

> −A2 T, (6.10)

24 ATUL DIXIT

where A2 < π/4 if A1 is small enough and T > T0 for some fixed T0 . From (6.10), we
see that
1
< eA3 T , (6.11)
L(2s + 1, χ)
where A3 < π/2. Using (1.12) and (6.11), we observe that as T → ∞ through the above
values, the integrals along the horizontal segments tend to zero. Now let ρ−1 2
:= δ + iγ
denote a non-trivial zero of L(2s + 1, χ). Let R f (a) denote the residue at a of the
1


2
−s
Γ 2 −s qα
function f (s) := . The non-trivial zeros of L(2s + 1, χ) lie in the
L(2s + 1, χ) π
critical strip − 21 < Re s < 0, whereas the trivial zeros are at −1, −2, −3, · · · . Also,
Γ 2 − s has poles at 21 , 23 , 52 , · · · . Then the residue theorem yields
1


−s −s
Γ 21 − s Γ 21 − s
Z c+i∞
 Z λ+i∞

qα2 qα2
ds = ds
c−i∞ L(2s + 1, χ) π λ−i∞ L(2s + 1, χ) π
   !
X ρ−1 1
− 2πi Rf (−1) + Rf + Rf ,
ρ
2 2
(6.12)
where
−s
Γ 21 − s


qα2 α2 q
Rf (−1) = lim (s + 1) = √ ′ ,
s→−1 L(2s + 1, χ) π 4 πL (−1, χ)
−s  ρ−1
Γ 21 − s Γ 2−ρ



qα2
   
ρ−1 ρ−1 2 π 2
Rf = lim s − = ,
2 s→ ρ−1
2
2 L(2s + 1, χ) π 2L′ (ρ, χ) qα2
√
π
Rf (1/2) = − √ . (6.13)
α qL(2, χ)
Of course,Phere we have
assumed that the non-trivial zeros of L(2s + 1, χ) are all simple
ρ−1
and that ρ Rf 2 converges, since the afore-mentioned discussion regarding the
integrals along the horizontal segments tending
P to zero
as T → ∞ through the chosen
ρ−1
sequence does not imply the convergence of ρ Rf 2 in the ordinary sense. It only
means that the series converges only when we bracket the terms in such a way that
the two terms for which
|γ − γ ′ | < exp (−A1 γ/ log 2γ) + exp (−A1 γ ′ / log 2γ ′ )
are included in the same bracket. Using (6.2) and interchanging the order of summation
and integration because of absolute convergence, we obtain
−s ∞   2 2 −s
Γ 12 − s
Z λ+i∞
χ(n)µ(n) λ+i∞


qα2

1 qα n
X Z
ds = Γ −s ds
λ−i∞ L(2s + 1, χ) π n=1
n λ−i∞ 2 π
√ X ∞ −s
χ(n)µ(n) d+i∞

π π
Z
= √ Γ(s) ds,
α q n=1 n2 d−i∞ α2 n2 q
(6.14)
 CHARACTER ANALOGUES 25

where in the penultimate line, we have made the change of variable s → 21 − s so that
−1 < d < 0. Thus, (6.12), (6.13), (6.14) and (6.3) imply
−s
Γ 21 − s
Z c+i∞

qα2
ds
c−i∞ L(2s + 1, χ) π
∞
2π 3/2 i X χ(n)µ(n)  − 2π 2

=− √ 1 − e α n q
α q n=1 n2
X Γ 2−ρ  π  2

ρ−1 √ !
α2 q 2 π
− 2πi √ + − √ . (6.15)
4 πL′ (−1, χ) ρ
2L ′ (ρ, χ) qα 2 α qL(2, χ)
p √
From (6.7), (6.15) and the facts that αβ = 1 and G(χ)G(χ) = i q, we find that
∞
√ p X χ(n)µ(n) − πα 2

α α G(χ) e n2 q

n=1
n2
√ p √ p ∞
α α G(χ) β β G(χ) p p X χ(n)µ(n) − πβ 2

= − + β β G(χ) e n2 q
L(2, χ) L(2, χ) n=1
n2
√
 ρ/2 √ p
α αq 3/2 G(χ) q G(χ) X Γ 2−ρ
p p
2 π ρ β β G(χ)
− − √ β + . (6.16)
4πL′ (−1, χ) 2π β ρ L′ (ρ, χ) q L(2, χ)
Applying (6.1) with N = q and k = 3 and replacing χ by χ gives
1 4πi
′
= . (6.17)
L (−1, χ) qG(χ)L(2, χ)
Thus (6.16) and (6.17) yield
∞ ∞
√ p X χ(n)µ(n) − πα 2 p p X χ(n)µ(n) − πβ 2

α α G(χ) e n2 q − β β G(χ) e n2 q

n=1
n2 n=1
n2

 ρ/2
q G(χ) X Γ 2−ρ
p
π
=− √ ′
2
β ρ. (6.18)
2π β ρ L (ρ, χ) q
Switching the roles of α and β and those of χ and χ gives
q G(χ) X Γ 2−ρ

 ρ/2
X Γ 2−ρ  π ρ/2
p p

π q G(χ)
√ 2
αρ + √ 2
β ρ = 0. (6.19)
2π α ρ L′ (ρ, χ) q 2π β ρ L′ (ρ, χ) q
Finally (6.18) and (6.19) give (1.24) upon simplification. 
Remark. The approach used above for proving that the integrals along the horizontal
segments tend to zero as T → ∞ through the chosen sequence is adapted from [37,
p. 219].
To prove Theorem 1.10, we require the following lemma.
Lemma 6.1. ∞
X χ(n)µ(n) 1
= .
n=1
n L(1, χ)
26 ATUL DIXIT

Proof. Dividing n into its residue classes mod q by letting n = qr + b, 0 ≤ r < ∞,

0 ≤ b ≤ q − 1, we find that since χ has period q,
∞ q−1
∞ X q−1 ∞
X χ(n)µ(n) X χ(b)µ(qr + b) X X µ(qr + b)
= = χ(b) . (6.20)
n=1
n r=0 b=0
qr + b b=0 r=0
qr + b
The series ∞
P
r=0 µ(qr + b)/(qr + b) was first studied by J.C. Kluyver [19] and its conver-
gence was proved by E. Landau [28]. In fact, Landau gave an explicit representation
for this series in
Pterms of a finite sum consisting of L-functions. Thus (6.20) implies
convergence of ∞ n=1 χ(n)µ(n)/n. Then using (6.2) and an analogue of Abel’s theorem
for power series, we see that
∞ ∞
X χ(n)µ(n) X χ(n)µ(n) 1 1
= lim s
= lim = .
n=1
n s→1
n=1
n s→1 L(s, χ) L(1, χ)

Proof of Theorem 1.10. The proof is very similar to that of Theorem 1.9 and hence
we omit the details. However we note that Lemma 6.1, (1.10) in the form [10, p. 69]
q 1/2 −s/2 s/2  s 
 
−(1−s)/2 (1−s)/2 1−s
π q Γ L(1 − s, χ) = π q Γ L(s, χ)
2 G(χ) 2
and (6.1) with N = q and k = 2 are used in the proof. 

7. Open problems
The following are some open problems with which we will conclude.
1. We have indirectly given the proof of the fact that function F (α, χ) defined in
(5.10) is real (purely imaginary) respectively when χ is even (odd). Prove this directly,
i.e., without using Corollaries 5.1 and 5.2 and the integrals in those corollaries.
2. Since (1.23) is of the form F (α) = F (β), where αβ = 1, it is natural to ask if
there exists an integral representation involving the Riemann Ξ-function equal to the
two expressions in (1.23). Finding an P integral representation for either side of (1.23)
′
may throw light on the convergence of ρ (Γ ((1 − ρ)/2) aρ ) /ζ (ρ) provided, of course,
that the integral converges in the first place. It should be remarked here that Hardy
and Littlewood [18, p. 161] have shown that the relation
 1 
P (y) = O y − 4 +δ , (7.1)

where P (y) = ∞ n
P
n=1 (−y) /(n!ζ(2n+1)), can be derived from (1.23) provided we assume
P ′
the Riemann hypothesis and the absolute convergence of ρ Γ ((1 − ρ)/2) /ζ (ρ). They
have further shown that (7.1) is a necessary and sufficient condition for the Riemann
hypothesis to be true.
Similarly, it is natural to ask
if the expressions in (1.24) and (1.25) have integral
representations involving Ξ 2t , χ .

3. Does there exist a generalization of Theorem 1.8 admitting representations of the

form F (z, α) = F (z, β)? Similarly, do there exist generalizations of Theorems 1.9 and
 CHARACTER ANALOGUES 27

1.10 admitting representations of the form F (z, α, χ) = F (−z, β, χ) = F (−z, α, χ) =

F (z, β, χ)?
Acknowledgements
The author wishes to express his sincere thanks to Professor Bruce C. Berndt for his
constant support, careful reading of this manuscript and for several helpful comments,
and to Professors Alexandru Zaharescu, Kevin Ford and Adolf Hildebrand for their
help. He would also like to thank Becky S. Burner, Timothy W. Cole and Margaret
A. Lewis from the Mathematics library at University of Illinois at Urbana-Champaign,
and Anton Lukyanenko for their great help in locating the Russian references and
citations. Last but not the least, he would like to thank Reetesh Ranjan for numerically
verifying some of the identities by programming in Fortran.

References
[1] S. Ahlgren, B.C. Berndt, A.J. Yee and A. Zaharescu, Integrals of Eisenstein series and derivatives
of L-functions, Internat. Math. Res. Not., (2002), No. 32, 1723–1738.
[2] T.M. Apostol, Introduction to Analytic Number Theory, 5th ed., Springer, New York, 1998.
[3] B.C. Berndt, Identities involving the coefficients of a class of Dirichlet series. IV, Trans. Amer.
Math. Soc 149 (1970), 179–185.
[4] B.C. Berndt, Character analogues of the Poisson and Euler-MacLaurin summation formulas with
applications, J. Number Theory 7 (1975), 413–445.
[5] B.C. Berndt, Ramanujan’s Notebooks, Part V, Springer-Verlag, New York, 1998.
[6] B.C. Berndt and A. Dixit, A transformation formula involving the Gamma and Riemann zeta
functions in Ramanujan’s Lost Notebook, The legacy of Alladi Ramakrishnan in the mathematical
sciences, K. Alladi, J. Klauder, C. R. Rao, Eds, Springer, New York, 2010, pp. 199–210.
[7] B.C. Berndt, A. Dixit and J. Sohn, Character analogues of theorems of Ramanujan, Koshliakov
and Guinand, to appear in Advances in Applied Mathematics, (special issue in honor of Dennis
W. Stanton).
[8] B.C. Berndt, Y. Lee, and J. Sohn, The formulas of Koshliakov and Guinand in Ramanujan’s lost
notebook, Surveys in Number Theory, Series: Developments in Mathematics, vol. 17, K. Alladi,
ed., Springer, New York, 2008, pp. 21–42.
[9] B.C. Berndt and J. Sohn, Asymptotic formulas for two continued fractions in Ramanujan’s Lost
Notebook, J. London Math. Soc., (2) 65 (2002), 271–284.
[10] H. Davenport, Multiplicative Number Theory, 3rd ed., Springer–Verlag, New York, 2000.
[11] A. Dixit, Analogues of a transformation formula of Ramanujan, to appear in International Journal
of Number Theory.
[12] A. Dixit, Series transformations and integrals involving the Riemann Ξ-function, J. Math. Anal.
Appl. 368 (2010), 358–373.
[13] A. Dixit, Transformation formulas associated with integrals involving the Riemann Ξ-function,
to appear in Monatshefte für Mathematik.
[14] W.L. Ferrar, Some solutions of the equation F (t) = F (t−1 ), J. London Math. Soc. 11 (1936),
99–103.
[15] A.P. Guinand, On Poisson’s summation formula, Ann. Math. (2) 42 (1941), 591–603.
[16] A.P. Guinand, Some rapidly convergent series for the Riemann ξ-function, Quart. J. Math. (Ox-
ford) 6 (1955), 156–160.
[17] G.H. Hardy, Note by Mr. G.H. Hardy on the preceding paper, Quart. J. Math. 46 (1915), 260–261.
[18] G.H. Hardy and J.E. Littlewood, Contributions to the Theory of the Riemann Zeta-Function and
the Theory of the Distribution of Primes, Acta Math., 41(1916), 119-196.
P µ(m)
[19] J.C. Kluyver, Series derived from the series m , Koningl. Akad. Wetensch. Amsterdam Proc.
Sect. Sci. 6 (1904), 305–312.
28 ATUL DIXIT

[20] N.S. Koshliakov, Investigation of some questions of the analytic theory of a rational and quadratic
field, I, II and III (in Russian), Izv. Akad. Nauk SSSR Ser. Mat. 18 No. 2, 113–144; No. 3, 213–260;
No. 4, 307–326 (1954).
[21] N.S. Koshliakov (under the name N.S. Sergeev), Issledovanie odnogo klassa transtsendentnykh
funktsii, opredelyaemykh obobshchennym yravneniem Rimana (A study of a class of transcen-
dental functions defined by the generalized Riemann equation) (in Russian), Trudy Mat. Inst.
Steklov, Moscow, 1949.
[22] N.S. Koshliakov, On an extension of some formulae of Ramanujan, Proc. London Math. Soc. 12
(1936), 26–32.
[23] N.S. Koshliakov, On a general summation formula and its applications (in Russian),
Comp. Rend. (Doklady) Acad. Sci. URSS 4 (1934), 187–191.
[24] N.S. Koshliakov, On a transformation of definite integrals and its application to the theory of
Riemann’s function ζ(s), Comp. Rend. (Doklady) Acad. Sci. URSS 15 (1937), 3–8.
[25] N.S. Koshliakov, On Voronoı̈’s sum-formula, Mess. Math. 58 (1929), 30–32.
[26] N.S. Koshliakov, Some identities in quadratic fields (in Russian), Dokl. Akad. Nauk. 2 No. 9
(1934), 527–531.
[27] N.S. Koshliakov, Some integral representations of the square of Riemann’s function Ξ(t), Dokl.
Akad. Nauk. 2 (1934), 401–405.
[28] E. Landau, Remarks on the paper of Mr. Kluyver on page 305 of Vol. VI: Series derived from the
P µ(m)
series m , Koningl. Akad. Wetensch. Amsterdam Proc. Sect. Sci. 7 (1905), 66–77.
[29] R. Ma, Y. Yi and Y. Zhang, On the mean value of the generalized Dirichlet L-functions, Czechoslo-
vak Math. J. 60 (135) (2010), 597–620.
[30] F. Oberhettinger, Tables of Mellin Transforms, Springer-Verlag, New York, 1974.
[31] R.B. Paris and D. Kaminski, Asymptotics and Mellin-Barnes Integrals, Encyclopedia of Mathe-
matics and its Applications, 85. Cambridge University Press, Cambridge, 2001.
[32] S. Ramanujan, New expressions for Riemann’s functions ξ(s) and Ξ(t), Quart. J. Math. 46 (1915),
253–260.
[33] S. Ramanujan, Notebooks (2 volumes), Tata Institute of Fundamental Research, Bombay, 1957.
[34] S. Ramanujan, Collected Papers, Cambridge University Press, Cambridge, 1927; reprinted by
Chelsea, New York, 1962; reprinted by the American Mathematical Society, Providence, RI,
2000.
[35] S. Ramanujan, The Lost Notebook and Other Unpublished Papers, Narosa, New Delhi, 1988.
[36] N.M. Temme, Special functions: An Introduction to the Classical Functions of Mathematical
Physics, Wiley-Interscience Publication, New York, 1996.
[37] E.C. Titchmarsh, The Theory of the Riemann Zeta Function, Clarendon Press, Oxford, 1986.

Department of Mathematics, University of Illinois, 1409 West Green Street, Ur-

bana, IL 61801, USA
E-mail address: aadixit2@illinois.edu