Some applications of the Dirichlet integrals to the summation of series

and the evaluation of integrals involving the Riemann zeta function

Donal F. Connon

dconnon@btopenworld.com

1 December 2012

Abstract

Using the Dirichlet integrals, which are employed in the theory of Fourier series, this
paper develops a useful method for the summation of series and the evaluation of
integrals.

CONTENTS Page

1. Introduction 1
2. Application of Dirichlet’s integrals 5
3. Some applications of the Poisson summation formula:
3.1 Digamma function 11
3.2 Log gamma function 13
3.3 Logarithm function 14
3.4 An integral due to Ramanujan 16
3.5 Stieltjes constants 36
3.6 Connections with other summation formulae 40
3.7 Hurwitz zeta function 48
4. Some integrals involving cot x 62
5. Open access to our own work 68

1. Introduction

In two earlier papers [19] and [20] we considered the following identity (which is easily
verified by multiplying the numerator and the denominator by the complex conjugate
(1 − e −ix ) )
1 1 i sin x 1 i
(1.1) = − = + cot( x / 2)
1− e ix
2 2 1 − cos x 2 2

Using the basic geometric series identity

1 N
y N +1
= ∑ yn +
1 − y n =0 1− y
we obtain
 1 N
(1.1a) = ∑ einx + RN ( x)
1− e ix
n =0

1
i( N + ) x
i ( N +1) x
e 1 ie 2
where RN ( x) = = ei ( N +1) x [1 + i cot( x / 2)] =
1− e ix
2 2sin( x / 2)

Separating the real and imaginary parts of (1.1a) produces the following two identities
(the first of which is called Lagrange’s trigonometric identity and contains the Dirichlet
kernel DN ( x) which is employed in the theory of Fourier series [55, p.49])

1 N sin( N + 1 / 2) x
(1.2) = ∑ cos nx −
2 n =0 2 sin( x / 2)

1 N
cos( N + 1 / 2) x
(1.3) cot( x / 2) = ∑ sin nx +
2 n=0 2 sin( x / 2)

(and these equations may obviously be generalised by substituting αx in place of x .

We first multiply by p( x) and, in this section, p( x) is assumed to be twice continuously
differentiable on [a, b] . Integration then gives us

b b
1 N

2 ∫a
p ( x ) (1 + i cot( x / 2) ) dx = ∑ ∫
n =0 a
p ( x) einx dx + RN

and we have

i cos( N + 1/ 2) x − sin( N + 1/ 2) x
b b
1
RN = ∫ p ( x) RN ( x) dx = ∫ p ( x) dx
a
2a sin( x / 2)

1
b
⎧ cos( x / 2) ⎫
=−
2a∫ p ( x) ⎨cos Nx +
⎩ sin( x / 2)
sin Nx ⎬ dx
⎭

i
b
⎧ cos( x / 2) ⎫
+
2a∫ p ( x) ⎨− sin Nx +
⎩ sin( x / 2)
cos Nx ⎬ dx
⎭

Therefore, provided sin (x / 2 ) has no zero in [ a, b ], a weak version of the Riemann-

Lebesgue lemma (which was previously employed in Eq.(2.17) in [19] where we required
p( x) to be twice continuously differentiable) tells us that lim RN = 0.
N →∞

2
This will also be the case if sin (a / 2 ) = 0 , provided p(a) = 0 . From the above we can
therefore derive the following trigonometric integral identities:

b ∞ b
1
(1.4)
2 ∫a
p ( x ) dx = ∑ ∫ p( x) cos nx dx
n =0 a

b ∞ b
1
(1.4a)
2 ∫a
p ( x ) cot( x / 2) dx = ∑ ∫ p( x) sin nx dx
n =0 a

and more generally we have (by letting x → α x in (1.2) and (1.3) respectively)

b ∞ b
1
(1.5)
2 ∫a
p ( x ) dx = ∑ ∫ p( x) cos α nx dx
n =0 a

b ∞ b
1
(1.5a)
2 ∫a
p ( x ) cot(α x / 2) dx = ∑ ∫ p( x) sin α nx dx
n =0 a

Equations (1.5) and (1.5a) are valid provided (i) sin(α x / 2) ≠ 0 ∀ x ∈ [a, b] or,
alternatively, (ii) if sin(α a / 2) = 0 then p(a) = 0 also.

Similarly, using the identity

1 1 i sin x 1 i
(1.6) = − = − tan( x / 2)
1+ e ix
2 2 1 + cos x 2 2

we may easily prove that

b ∞ b
1
(1.7) ∫ p ( x)dx = ∑ ∫ p ( x)(−1) n cos nx dx
2a n =0 a

b ∞ b
1
(1.7a)
2a∫ p ( x ) tan( x / 2) dx = ∑ ∫
n =0 a
p ( x)(−1) n +1 sin nx dx

and more generally

b ∞ b
1
(1.8)
2a∫ p ( x ) dx = ∑ ∫
n =0 a
p ( x)(−1) n cos α nx dx

b ∞ b
1
(1.8a)
2a∫ p ( x ) tan(α x / 2) dx = ∑ ∫
n =0 a
p ( x)(−1) n +1 sin α nx dx

3
From (1.6) we note that the denominator is cos(x / 2) and hence (1.7) and (1.7a) are only
valid provided either (i) cos(x / 2) has no zero in [ a, b ] or (ii) if cos(a / 2) = 0 , then p(a)
must also be zero.

Equations (1.8) and (1.8a) are valid provided (i) cos(α x / 2) ≠ 0 ∀ x ∈ [a, b] or,
alternatively, (ii) if cos(α a / 2) = 0 then p(a) = 0 also.

The following simple trigonometric identities are easily proved

2
(1.9) cot( x / 2) + tan( x / 2) =
sin x

(1.10) cot( x / 2) − tan( x / 2) = 2 cot x

Therefore, combining (1.5) and (1.7) produces the following identity

b ∞ b ∞ b
p ( x)
(1.11) ∫a sin x ∑dx = ∫
n =0 a
p ( x ) sin nx dx − ∑ ∫
n =0 a
p( x)(−1) n sin nx dx

which simplifies to

b ∞ b
p ( x)
(1.12) ∫a sin x dx = 2 ∑ ∫ p( x) sin(2n + 1) x dx
n =0 a

Similarly, using (1.10) we obtain

b ∞ b ∞ b
(1.13) ∫ p( x) cot x dx = ∑ ∫ p( x) sin nx dx + ∑ ∫ (−1) p( x) sin nx dx
n

a n =0 a n =0 a

which simplifies to

b ∞ b
(1.14) ∫ p( x) cot x dx = 2∑ ∫ p( x) sin 2nx dx
a n =1 a

It should be noted that in the above formulae we require either (i) both sin( x / 2) and
cos(x / 2) have no zero in [ a, b ] or (ii) if either sin( a / 2) or cos(a / 2) is equal to zero
then p(a) must also be zero. Condition (i) is equivalent to the requirement that sin x has
no zero in [ a, b ].

Note that (1.14) is equivalent to (1.5a) with α = 2 .

More generally we have

4
 b ∞ b
p ( x)
(1.14a) ∫a sin α x dx = 2 ∑ ∫ p( x) sin [(2n + 1)α x ] dx
n =0 a

b ∞ b
(1.14b) ∫ p( x) cot α x dx = 2∑ ∫ p( x) sin 2α nx dx
a n =1 a

Equations (1.14a) and (1.14b) are valid provided (i) sin(α x) ≠ 0 ∀ x ∈ [a, b] or,
alternatively, (ii) if sin(α a ) = 0 then p(a) = 0 also.

We may also generalise Poisson’s integral [17, p.250] by applying the above analysis to
1 1
the quotient and useful results may also be obtained by using .
1 ± re ix
1 ± ireix

2. Application of Dirichlet’s integrals

In the previous section we required that p( x) was twice continuously differentiable on the
interval [a, b] ; with the assistance of Dirichlet we now show that this condition may be
significantly relaxed.

Dirichlet [26] considered the following integrals in his classical treatment of Fourier
series in 1829 (when he was then 23 years old)

sin( μt )
b
(2.1) lim ∫ f (t ) dt = 0
μ →∞
a
sin t

sin( μt ) π
b
(2.2) lim ∫ f (t ) dt = f (0+)
μ →∞
0
sin t 2

where 0 < a < b < π . Dirichlet showed that these limits are valid if f ( x) is continuous
on [0, b] and f ( x) only has a finite number of maxima and minima on [0,b] . Jordan [37]
subsequently showed in 1881 that the less restrictive condition that f ( x) be of bounded
variation on [0, b] is sufficient. Proofs may be found in [5, p.314] and [17, p.219].

We recall from [5, p.128] that if f ( x) is monotonic on [a, b] , then f ( x) is of bounded

variation on [a, b] . In addition, we also note that if f ( x) is continuous on [a, b] and if
f ′( x) exists and is bounded on the open interval, then f ( x) is of bounded variation on
[ a , b] .

Even though Jordan wanted to relax the smoothness requirement for the pointwise
convergence of Fourier series, it subsequently turned out that functions of bounded
variation are in fact fairly smooth anyway; Lebesgue ([50, p.356 ] and [7, p.192]) proved

5
that any monotonic function on a closed interval is differentiable almost everywhere, and
that its derivative is integrable.

With the substitution t = α x in (2.1) and (2.2) we see that

αb
sin( μα x)
(2.3) lim
μ →∞ ∫
αa
f (α x)
sin α x
dx = 0

or equivalently
b′
sin( μα x)
(2.4) lim ∫ g ( x) dx = 0
μ →∞
a′
sin α x

where 0 < a′ < b′ < π .

We may write (1.2) as

1 N sin(2 N + 1)α x
(2.5) = ∑ cos 2α nx −
2 n =0 2 sin α x

and, after multiplying this by a function f ( x) of bounded variation on [a, b] , we easily

see that
sin(2 N + 1)α x
b N b b
1
(2.6) ∫ f ( x) dx = ∑ ∫ f ( x) cos 2α nx dx − ∫ f ( x) dx
2a n =0 a a
2sin α x

Letting N → ∞ we obtain

b ∞ b
1
(2.7)
2 ∫a
f ( x ) dx = ∑ ∫ f ( x) cos 2α nx dx
n=0 a

provided 0 < α a < α b < π . This is a generalised version of (1.4) above and, in
particular, it should be noted that we no longer require f ( x) to be twice continuously
differentiable on [a, b] .

With α = π we may write (2.7) as

b ∞ b
1
(2.8)
2 ∫a
f ( x ) dx = ∑ ∫ f ( x) cos 2π nx dx
n =0 a

where 0 < a < b < 1.

We now consider the case where (2.2) applies. With the substitution t = α x in (2.2) we
see that

6
 αb
sin( μα x) π
(2.9) lim α ∫ f (α x) dx = f (0+)
μ →∞
0
sin α x 2

or equivalently
b′
sin( μα x) π
(2.10) lim ∫ g ( x) dx = g (0+)
μ →∞
0
sin α x 2α

where 0 < b′ < π .

Letting N → ∞ in (2.6) and using (2.10) we then obtain

1
b ∞ b
π
20∫ f ( x ) dx = ∑ ∫
n=0 0
f ( x) cos 2α nx dx −
4α
f (0+)

where 0 < α b < π .

With α = π we may write this as

b ∞ b
1
(2.11) f (0+) = ∫ f ( x) dx + 2∑ ∫ f ( x) cos 2π nx dx
2 0 n =1 0

where 0 < b < 1.

There is clearly a connection between (2.11) and the Poisson summation formula in the
form reported in Ramanujan’s Notebooks [10, Part II, p.252]. If f ( x) is a continuous
function of bounded variation on [a, b] , then

# b ∞ b
(2.12) ∑
a ≤ n ≤b
f (n) = ∫ f ( x) dx + 2∑ ∫ f ( x) cos 2π nx dx
n =1 a
a

where the # on the summation sign on the left-hand side indicates that if a or b is an
1 1
integer, then only f (a) or f (b) , respectively, is counted.
2 2

The equivalence is shown below.

As noted by Bromwich [15, p.492] we consider the integral

1
sin(2 N + 1)π x sin(2 N + 1)π x sin(2 N + 1)π x
1 2 1

∫
0
f ( x)
sin π x
dx = ∫
0
f ( x)
sin π x
dx + ∫ f ( x)
1 sin π x
dx
2

7
and make the substitution x = 1 − t in the second part. This immediately gives us

1
sin(2 N + 1)π x sin(2 N + 1)π x
1 2

∫
0
f ( x)
sin π x
dx = ∫ [ f ( x) + f (1 − x)]
0
sin π x
dx

and employing (2.10) we obtain

sin(2 N + 1)π x
1
1
lim ∫ f ( x) dx = [ f (0+) + f (1−)]
N →∞
0
sin π x 2

Therefore we have

1 ∞ 1
1
(2.13) [ f (0+) + f (1−)] = ∫ f ( x) dx + 2∑ ∫ f ( x) cos 2π nx dx
2 0 n =1 0

Similarly we have

sin(2 N + 1)π x sin(2 N + 1)π x sin(2 N + 1)π x

2 1 2

∫
0
f ( x)
sin π x
dx = ∫ f ( x)
0
sin π x
dx + ∫ f ( x)
1
sin π x
dx

and we write

3
sin(2 N + 1)π x sin(2 N + 1)π x sin(2 N + 1)π x
2 2 2

∫
1
f ( x)
sin π x
dx = ∫
1
f ( x)
sin π x
dx + ∫ f ( x)
3 sin π x
dx
2

The substitution x = 1 + t gives us

3 1
2
sin(2 N + 1)π x 2
sin(2 N + 1)π t
∫
1
f ( x)
sin π x
dx = ∫
0
f (1 + t )
sin π t
dt

and with x = 2 − t we obtain

1
sin(2 N + 1)π x sin(2 N + 1)π t
2 2

3
∫ f ( x)
sin π x
dx = ∫
0
f (2 − t )
sin π t
dt
2

Therefore, using (2.10) we see that

sin(2 N + 1)π x
2
1
lim ∫ f ( x) dx = [ f (0+) + f (1−) + f (1+ ) + f (2−)]
N →∞
0
sin π x 2

8
More generally we see that

sin(2 N + 1)π x 1 m −1
m
1
lim ∫ f ( x) dx = [ f (0+) + f (m−)] + ∑ [ f (n −) + f (n +)]
N →∞
0
sin π x 2 2 n =1

and, if f ( x) is continuous, this may be written as

sin(2 N + 1)π x m −1
m
1
lim ∫ f ( x) dx = [ f (0) + f (m)] + ∑ f (n)
N →∞
0
sin π x 2 n =1

1 m
= − [ f (0) + f (m)] + ∑ f (n)
2 n =0

Then using (2.6) we obtain a version of the Poisson summation formula

m ∞ m
m
1
(2.14) ∑ f ( n) = ∫ f ( x) dx + [ f (0) + f ( m) ] + 2∑ ∫ f ( x) cos 2nπ x dx
n=0 0
2 n =1 0

Letting m → ∞ and assuming that lim f (n) = 0 (which is of course required for the
n →∞

convergence of the infinite series) we obtain

∞ ∞∞ ∞
1
(2.15) f (0) + ∑ f (n) = ∫ f ( x) dx + 2∑ ∫ f ( x) cos 2π nx dx
2 n =1 0 n =1 0

as previously noted by Guinand [31].

Another form of the Poisson summation formula is shown below [31]

+∞ +∞ +∞

∑
n =−∞
f ( n) = ∑∫
n =−∞ −∞
f ( x) cos 2π nx dx

which may be formally obtained by letting n → − n in (2.14). An equivalent

representation is

+∞ +∞ +∞

∑
n =−∞
f ( n) = ∑∫
n =−∞ −∞
f ( x)e 2π inx dx

Various proofs exist for the Poisson formula: these include, inter alia, those given by
Apostol [5, p.332], Guinand [31], Ivić [36, p.490], Mordell [42] and Wilton [53].

9
We now illustrate the intimate connection with the Euler-Maclaurin summation formula.
Cast in its simplest form, the Euler- Maclaurin summation formula may be written as
[38, p.521]

⎛ 1⎞
m m
m
1
∑ f ( n) = ∫ f ( x) dx + [ f (0) + f ( m) ] + ∫ ⎜ x − [ x] − ⎟ f ′( x) dx
n =0 0
2 0⎝
2⎠

1 ∞
sin 2nπ x
and with P1 ( x) = x − [ x] − we have the familiar Fourier series P1 ( x) = −∑ .
2 n =1 nπ
Therefore we have

⎛ 1⎞ ∞
sin 2nπ x
m m

∫0 ⎜⎝ x − [ x ] − ⎟
2⎠
f ′( x ) dx = − ∫ ∑ nπ
f ′( x) dx
0 n =1

and with integration by parts this becomes

m
sin 2nπ x
∞ ∞ m
= −∑ f ( x) + 2∫ ∑ f ( x) cos 2nπ x dx
n =1 nπ 0 0 n =1

m ∞
= 2 ∫ ∑ f ( x) cos 2nπ x dx
0 n =1

Hence, assuming that interchanging the order of integration and summation is valid, we
obtain the version of the Poisson summation formula in (2.15) above

∞ ∞ ∞ ∞
1
f (0) + ∑ f (n) = ∫ f ( x) dx + 2∑ ∫ f ( x) cos 2π nx dx
2 n =1 0 n =1 0

Letting f ( x) → f ( x) cos π x in (2.15) results in

∞ ∞ ∞ ∞
1
f (0) + ∑ (−1) n f (n) = ∫ f ( x) cos π x dx + 2∑ ∫ f ( x) cos 2π nx cos π x dx
2 n =1 0 n =1 0

It is easily seen that

∞ ∞ ∞ ∞ ∞ ∞
2∑ ∫ f ( x) cos 2π nx cos π x dx = ∑ ∫ f ( x) cos(2n + 1)π x dx + ∑ ∫ f ( x) cos(2n − 1)π x dx
n =1 0 n =1 0 n =1 0

10
 ∞ ∞ ∞
= ∫ f ( x) cos π x dx + 2∑ ∫ f ( x) cos(2n + 1)π x dx
0 n =1 0

and hence we obtain

∞ ∞ ∞
1
(2.16) f (0) + ∑ (−1) n f (n) = 2∑ ∫ f ( x) cos(2n + 1)π x dx
2 n =1 n =0 0

where the summation on the right-hand side now starts at n = 0 .

An alternative derivation of (2.16) is shown below. With f ( x) → f (2 x) we have

∞ ∞ ∞ ∞
1
f (0) + ∑ f (2n) = ∫ f (2 x) dx + 2∑ ∫ f (2 x) cos 2π nx dx
2 n =1 0 n =1 0

∞ ∞ ∞
1
= ∫ f (u ) du + ∑ ∫ f (u ) cos π nu du
20 n =1 0

∞ ∞ ∞
Since, for suitably convergent series, 2∑ a2 n − ∑ an = ∑ (−1) n an we have
n =1 n =1 n =1

∞ ∞ ∞ ∞
1
f (0) + 2∑ f (2n) − ∑ f (n) = 2∑ ∫ f ( x)[cos π nx − cos 2π nx] dx
2 n =1 n =1 n =1 0

∞ ∞
= 2∑ ∫ f ( x) cos(2n + 1)π x dx
n =0 0

and hence we derive (2.16) again.

3. Applications of the Poisson summation formula

Previously in [20] we gave many applications of the basic summation formulae stated in
section 1 above. Some further applications of the more generalised versions are given
below.

3.1 Digamma function

We consider f ( x ) = ψ ( a + x) for Re (a ) > 0 where ψ ( x) is the digamma function, being

d
the logarithmic derivative of the gamma function, i.e. ψ ( x) = log Γ( x) .
dx

We have shown in [] (this corrects the entry in [30, p.652, 6.467 2]) that

11
 1
(3.1.1) ∫ψ ( x + a) cos 2nπ x dx = sin(2nπ a)si(2nπ a) + cos(2nπ a)Ci(2nπ a)
0

where si ( x) and Ci ( x) are the sine and cosine integrals defined [30, p.878] by

∞
sin t
(3.1.2) si( x) = − ∫ dt
x
t

and for x > 0

∞
cos t − 1
x
cos t
(3.1.3) Ci ( x) = − ∫ dt = γ + log x + ∫ dt
x
t 0
t

where γ is Euler’s constant.

Note that there is a slightly different sine integral Si ( x) which is defined in [30, p.878]
and also in [1, p.231] by
x
sin t
(3.1.4) Si ( x) = ∫ dt
0
t

We have
∞ x ∞
sin t sin t sin t
si ( x) = − ∫ dt = ∫ dt − ∫ dt
x
t 0
t 0
t

and using the well-known integral from Fourier series analysis [5, p.286]

∞
π sin t
(3.1.5) =∫ dt
2 0
t

we therefore see that the two sine integrals are intimately related by

π
(3.1.6) si ( x ) = Si ( x ) −
2

It is easily seen that

∫ψ ( x + a) dx = log Γ(1 + a) − log Γ(a)

0
= log a

12
Since ψ (1 + x) is monotonic on [0,1] , we see that ψ (1 + x) is therefore of bounded
variation on [0,1] and hence we may apply (2.13) to obtain

∞
1
[ψ (a ) +ψ (1 + a )] = log a + 2∑ [cos(2nπ a )Ci (2nπ a ) + sin(2nπ a ) si (2nπ a )]
2 n =1

1
Since ψ (1 + a ) = ψ (a) + this may be written as
a

∞
1
(3.1.7) ψ (a ) = log a − + 2∑ [cos(2nπ a )Ci (2nπ a ) + sin(2nπ a ) si (2nπ a )]
2a n =1

which appears in Nörlund’s book [44, p.108]. Letting a = 1 results in

∞
1
(3.1.8) − γ = 2∑ Ci (2nπ )
2 n =1

and this corrects the corresponding formula given by Nielsen [43, p.80].

3.2 Log gamma function

Here we let f ( x ) = log Γ (a + x ) so that (2.13) becomes

1 ∞ 1
1
[log Γ(a) + log Γ(1 + a)] = ∫ log Γ(a + x) dx + 2∑ ∫ log Γ(a + x) cos 2nx dx
2 0 n =1 0

We showed in [23] that for a > 0 and n ≥ 1

1
1
(3.2.1) ∫ log Γ( x + a) cos 2nπ x dx = − 2nπ [ − sin(2nπ a)Ci(2nπ a) + cos(2nπ a)si(2nπ a)]
0

(which corrects the entry reported in [30, p.650, 6.443.3]) and hence we obtain

1
1 1 ∞ 1
[log Γ(a) + log Γ(1 + a)] = ∫ log Γ(a + x) dx + ∑ [sin(2nπ a)Ci (2nπ a) − cos(2nπ a) si(2nπ a) ]
2 0
π n =1 n

We designate F (a ) by

1
F (a ) = ∫ log Γ(a + x) dx
0

and differentiation gives us

13
 1
F ′(a) = ∫ψ (a + x) dx = log Γ(a + 1) − log Γ(a ) = log a
0

Integration then results in

F (u ) − F (0) = u log u − u

and using Raabe’s integral [49, p.18]

1
1
F (0) = ∫ log Γ( x) dx = log(2π )
0
2

we see that
1
1
∫ log Γ(a + x) dx = 2 log(2π ) + a log a − a
0

as reported in [49, p.207].

Combining the above results in

(3.2.2)
1 ⎛ 1⎞ 1 ∞ 1
log Γ(a ) = log(2π ) + ⎜ a − ⎟ log a − a + ∑ [sin(2nπ a )Ci (2nπ a ) − cos(2nπ a ) si (2nπ a ) ]
2 ⎝ 2⎠ π n =1 n

We note that (3.1.7) may be obtained by differentiating (3.2.2).

With a = 1 we obtain

∞
si (2nπ ) 1
(3.2.3) ∑
n =1 nπ
= log(2π ) − 1
2

which was given by Nielsen [43, p.79]. With a = 1/ 2 we obtain

∞
(−1) n si (nπ ) 1
(3.2.4) ∑
n =1 nπ
= (1 − log 2)
2

3.3 Logarithm function

We note that f ( x) = x log x does not meet the conditions required for (1.5) on the
interval [0,1] ; however the conditions required for (2.13) are satisfied.

14
Integration by parts gives us
−Ci (ax) + ax log x sin ax + (1 + log x) cos ax
∫ x log x cos ax dx = a2

so that we have the definite integral

1 −Ci (2π n) + lim [Ci (ax) − cos(ax) log x ]

∫ x log x cos 2π nx dx =
0
x →0
(2π n) 2

Using the definition of Ci ( x) in (3.3) we see that

cos t − 1
ax
Ci (ax) − cos(ax) log x = γ + log ax − cos(ax) log x + ∫ dt
0
t

cos t − 1
ax
= γ + log a − log x [ cos(ax) − 1] + ∫ dt
0
t
We consider the limit

cos(ax) − 1
lim log x[cos(ax) − 1] = lim x log x
x →0 x →0 x

cos(ax) − 1
= lim x log x ⋅ lim
x →0 x →0 x

and using L’Hôpital’s rule we obtain

cos(ax) − 1
lim =0
x →0 x

which shows that

lim[cos(ax) − 1]log x = 0
x →0

Then taking the limit as x → 0 we obtain

lim [Ci (ax) − cos(ax) log x ] = γ + log a

x →0

Hence we have

15
 −Ci (2π n) + γ + log(2π n)
1

∫ x log x cos 2π nx dx =
0
(2π n) 2
1
and the integral ∫ x log x dx = −1/ 4 is elementary.
0

Therefore, with f ( x) = x log x in (2.13) we obtain

1 ∞
−Ci (2π n) + γ + log(2π n)
= 2∑
4 n =1 (2π n) 2

so that

∞
Ci (2nπ ) 1
(3.3.1) ∑
n =1 n 2
= ς (2)[γ + log(2π )] − ς ′(2) − π 2
2

3.4 An integral due to Ramanujan

1
With f (u ) = ψ (1 + u ) − log(1 + u ) + in (2.15) we have
2(1 + u )

(3.4.1)
∞
1 ⎡1 ⎤ ∞ ⎡ 1 ⎤ ⎡ 1 ⎤
− γ ⎥ + ∑ ⎢ψ (1 + n) − log(1 + n) + ⎥ = ∫ ⎢ψ (1 + u ) − log(1 + u ) +
2(1 + u ) ⎦⎥
⎢ du
2 ⎣2 ⎦ n =1 ⎣ 2(1 + n) ⎦ 0 ⎣

∞ ∞
⎡ 1 ⎤
+2∑ ∫ ⎢ψ (1 + u ) − log(1 + u ) + cos 2π nu du
n =1 0 ⎣ 2(1 + u ) ⎥⎦

The left-hand side of (3.4.1) may be written as

1 ⎡1 ⎤ ∞ ⎡ 1 ⎤ ∞ ⎡ 1 ⎤ 1 ⎡1 ⎤
⎢
2 ⎣2
− γ ⎥ + ∑ ⎢
⎦ n =1 ⎣
ψ (1 + n ) − log(1 + n ) + ⎥ = ∑ ⎢ψ (n) − log n + ⎥ − ⎢ − γ ⎥
2(1 + n) ⎦ n =1 ⎣ 2n ⎦ 2 ⎣ 2 ⎦

and we split the other integral on the right-hand side into three components as shown
below
∞
⎡ 1 ⎤
∫ ⎢⎣ψ (1 + u) − log(1 + u ) + 2(1 + u ) ⎥⎦ cos 2π nu du
0

∞ ∞
= ∫ [ψ (1 + u ) − log u ] cos 2π nu du + ∫ [ log u − log(1 + u ) ] cos 2π nu du
0 0

16
 ∞
1 cos 2π nu
2 ∫0 1 + u
+ du

Ramanujan [12] obtained the following integral

∞
1
(3.4.2) ∫ [ψ (1 + u ) − log u ] cos 2nπ u du = 2 [ψ (1 + n) − log n]
0

which was rediscovered by Guinand [33] in 1947.

As regards the next component, integration by parts readily gives us

a ∫ [ log u − log(1 + u )] cos au du = − sin a Ci [a (1 + u )] + cos a Si [a (1 + u )] − Si (au )

+ [ log u − log(1 + u ) ] sin au

With a = 2π n this simplifies to

2π n ∫ [ log u − log(1 + u )] cos 2π nu du = Si [2π n(1 + u )] − Si (2π nu ) + [ log u − log(1 + u )] sin 2π nu

and we obtain the definite integral

∞
Si (0) − Si (2π n)
∫ [log u − log(1 + u)] cos 2π nu du =
0
2π n

Si (2π n) si (2π n) 1
=− =− −
2π n 2π n 4n

The substitution v = 1 + u results in

cos au
∫ 1+ u
du = cos aCi [a(1 + u)] + sin a Si [a(1 + u)]

π
where Si (u ) = + si (u ) . Therefore we have the definite integral
2

∞
cos au
∫0 1 + u du = sin a Si(∞) − [cos aCi(a) + sin a Si(a)]

17
and with a = 2π n this becomes
∞
cos 2nπ au
∫
0
1+ u
du = − Ci(2nπ )

∞
⎡ 1 ⎤
With I = ∫ ⎢ψ (1 + u ) − log(1 + u ) + du we then have
0 ⎣
2(1 + u ) ⎥⎦

∞
⎡ 1 ⎤ 1 ⎡1 ⎤ ∞
⎡1 Si (2π n) 1 ⎤
∑ ⎢⎣ψ ( n ) − log n + ⎥ − ⎢ − γ ⎥ = I + 2 ∑ ⎢ [ψ (1 + n) − log n] − − Ci (2nπ ) ⎥
n =1 2n ⎦ 2 ⎣ 2 ⎦ n =1 ⎣ 2 2π n 2 ⎦

∞
⎡ 1 si (2π n) ⎤
= I + ∑ ⎢[ψ (1 + n) − log n ] − − − Ci (2nπ ) ⎥
n =1 ⎣ 2n πn ⎦

∞
⎡ 1 ⎤ ⎡1 ⎤ 1 ⎡1 ⎤
= I + ∑ ⎢ψ (n) − log n + ⎥ − ⎢ log(2π ) − 1⎥ − ⎢ − γ ⎥
n =1 ⎣ 2n ⎦ ⎣ 2 ⎦ 2 ⎣2 ⎦

where we have employed the series

∞
1 ⎡1 ⎤
∑ Ci(2π n) = 2 ⎢⎣ 2 − γ ⎥⎦
n =1

∞
si (2π n) 1
∑
n =1 πn
= log(2π ) − 1
2

which were derived in (3.1.8) and (3.2.1) respectively.

We therefore obtain
∞
⎡ 1 ⎤ 1
(3.4.3) ∫ ⎢⎣ψ (1 + u) − log(1 + u) + 2(1 + u) ⎥⎦ du = 2 log(2π ) − 1
0

We now give an alternative derivation of the integral (3.4.3). We recall Binet’s second
formula for log Γ(u ) (which is derived in [51, p.250] using the Abel-Plana summation
formula)

⎛ 1⎞ 1
∞
tan −1 ( x / u )
(3.4.4) log Γ(u ) = ⎜ u − ⎟ log u − u + log(2π ) + 2 ∫ 2π x
dx
⎝ 2⎠ 2 0
e − 1

Another proof of this is also reported in [22]. This formula was also derived by
Ramanujan [10, Part II, p.221] in the case where u is a positive integer.

18
Differentiation of Binet’s formula results in [51, p.251]
∞
1 x
(3.4.5) log u −ψ (u ) = + 2∫ 2 dx
2u 0
(u + x )(e 2π x − 1)
2

and thus we have

∞
1 x
(3.4.6) ψ (t + u ) − log(t + u ) + = −2∫ dx
2(t + u ) 0
[(t + u ) + x 2 ](e2π x − 1)
2

Integration with respect to u gives us

∞ ∞∞
⎡ 1 ⎤ x
I (t ) = ∫ ⎢ψ (t + u ) − log(t + u ) + ⎥ du = − 2 ∫ ∫ dx du
0 ⎣
2(t + u ) ⎦ 0 0
[(t + u ) + x 2 ](e2π x − 1)
2

∞ ∞
xdx du
= −2∫ 2π x ∫
0
e − 1 0 (t + u ) 2 + x 2
We see that
∞ ∞
du dy
∫0 (t + u )2 + x2 = ∫t y 2 + x2
∞
1 ⎛ y⎞
= tan −1 ⎜ ⎟
x ⎝ x ⎠t

1 ⎡π ⎛ t ⎞⎤
= ⎢ − tan −1 ⎜ ⎟ ⎥
x⎣2 ⎝ x ⎠⎦
Since
⎛ a+b ⎞
tan −1 a + tan −1 b = tan −1 ⎜ ⎟
⎝ 1 − ab ⎠

we see that

π
tan −1 ( x / u ) = − tan −1 ( u / x ) for u / x > 0
2

and the integral may thus be written as

∞
du 1 ⎛ x⎞
∫ (t + u )
0
2
+x 2
=
x
tan −1 ⎜ ⎟
⎝t⎠

19
We therefore have

∞
tan −1 ( x / t )
I (t ) = −2 ∫ dx
0
e 2π x − 1

We then obtain from (3.4.4)

⎛ 1⎞ 1
I (t ) = ⎜ t − ⎟ log t − t + log(2π ) − log Γ(t )
⎝ 2⎠ 2
so that
∞
⎡ 1 ⎤ ⎛ 1⎞ 1
(3.4.7) ∫ ⎣⎢ψ (t + u) − log(t + u) + 2(t + u) ⎦⎥ du = ⎝⎜ t − 2 ⎠⎟ log t − t + 2 log(2π ) − log Γ(t )
0

Letting t = 1 gives us (3.4.3)

∞
⎡ 1 ⎤ 1
(3.4.8) ∫ ⎢⎣ψ (1 + u) − log(1 + u) + 2(1 + u) ⎥⎦ du = 2 log(2π ) − 1
0

It is easy to determine that

⎡ 1 ⎤
N

∫ ⎢⎣log u − log(1 + u) + 1 + u ⎥⎦ du = N log N − (1 + N ) log(1 + N ) + log(1 + N )

0

⎛ 1⎞
= − N log ⎜1 + ⎟
⎝ N⎠

⎛ 1⎞ 1 ⎛ 1 ⎞
and, since log ⎜1 + ⎟ = + O ⎜ 2 ⎟ , we then see that
⎝ N⎠ N ⎝N ⎠

⎡ 1 ⎤
N
lim ∫ ⎢log u − log(1 + u ) + du = −1
N →∞
0 ⎣
1 + u ⎥⎦

We note that
∞
⎡ 1 ⎤
∫ ⎢⎣ψ (1 + u) − log(1 + u ) + 2(1 + u ) ⎥⎦ du
0

∞
⎡ 1 1 ⎤
= ∫ ⎢ψ (1 + u ) − log u − + log u − log(1 + u ) +
1 + u ⎥⎦
du
0 ⎣
2(1 + u )

20
and thus we have
∞ ∞
⎡ 1 ⎤ ⎡ 1 ⎤
∫0 ⎢⎣ψ (1 + u ) − log(1 + u ) + ⎥
2(1 + u ) ⎦
du = −1 + ∫ ⎢
0 ⎣
ψ (1 + u ) − log u −
2(1 + u ) ⎥⎦
du

Therefore using (3.4.8) we obtain

∞
⎡ 1 ⎤ 1
(3.4.9) ∫ ⎢⎣ψ (1 + u) − log u − 2(1 + u ) ⎥⎦ du = 2 log(2π )
0

as previously obtained by Berndt and Dixit [12]. This integral may also be evaluated in
the following manner:

We see that

∞ ∞
⎡ 1 ⎤ 1
∫0 ⎢⎣ψ (1 + u) − log u − 2(u + 1) ⎥⎦ du = log Γ(1 + u) − u log u + u − 2 log(1 + u ) 0

⎡ 1 ⎤
= lim ⎢log Γ(1 + u ) − u log u + u − log(1 + u ) ⎥
u →∞
⎣ 2 ⎦

⎡ 1 1 ⎛ 1 ⎞⎤
= lim ⎢log Γ(1 + u ) − u log u + u − log u − log ⎜1 + ⎟ ⎥
u →∞
⎣ 2 2 ⎝ u ⎠⎦

⎡ ⎛ 1⎞ ⎤
= lim ⎢log Γ(1 + u ) − ⎜ u + ⎟ u log u + u ⎥
u →∞
⎣ ⎝ 2⎠ ⎦

The integral (3.4.9) may then be obtained upon using the asymptotic expression [49, p.8]

⎛ 1⎞ 1 1
log Γ(1 + u ) ≈ ⎜ u + ⎟ log u − u + log(2π ) +
⎝ 2⎠ 2 12u

A generalised version of the integral (3.4.2) appears in [30, p.652, 6.471.3]

∞
1
(3.4.10) ∫ [ψ (1 + u ) − log u ] cos 2π ut du = 2 [ψ (1 + t ) − log t ]
0

and integration results in

21
 ∞
1 ψ (1 + u ) − log u 1
(3.4.11) ∫ sin 2π uv du = [ log Γ(1 + v) − v log v + v ]
2π 0 u 2
□

Merkle and Merkle [40] have recently shown that

1 1 ∞
⎡ 2x −1 ⎤
(3.4.12) log Γ( x) = [log(2π ) − 1] − γ (2 x − 1) + ∑ ⎢ψ (1 + n) − log( x + n) +
2 2 n =0 ⎣ 2(1 + n) ⎥⎦

and commencing the summation at n = 1 gives us

(3.4.13)
1 ⎛ 1⎞ 1 ∞
⎡ 2x −1 ⎤
log Γ(1 + x) = [log(2π ) − 1] − γ ⎜ x + ⎟ + (2 x − 1) + ∑ ⎢ψ (1 + n) − log( x + n) +
2 ⎝ 2⎠ 2 n =1 ⎣ 2(1 + n) ⎥⎦

Differentiation gives us

∞
⎡ 1 1 ⎤
ψ (1 + x) = −γ + 1 + ∑ ⎢ − +
n =1 ⎣ x + n 1 + n ⎥⎦

∞
⎡ 1 1 1 1⎤
= −γ + 1 + ∑ ⎢ − + + − ⎥
n =1 ⎣ x + n n 1+ n n ⎦

∞
⎡ 1 1⎤ ∞ ⎡ 1 1⎤
= −γ + 1 + ∑ ⎢ − + ⎥ +∑⎢ − ⎥
n =1 ⎣ x + n n ⎦ n =1 ⎣1 + n n ⎦

Hence we obtain the well-known formula for the digamma function [49, p.14]

∞
⎡ 1 1⎤
(3.4.14) ψ (1 + x) = −γ − ∑ ⎢ − ⎥
n =1 ⎣ x + n n⎦

Letting x = 1 in (3.4.13) gives us

1 3 1 ∞ ⎡ 1 ⎤
0 = [log(2π ) − 1] − γ + + ∑ ⎢ψ (1 + n) − log(1 + n) +
2 2 2 n =1 ⎣ 2(1 + n) ⎥⎦

which may be written as

∞
1 3 ⎡ 1⎤ 1
0= log(2π ) − γ + ∑ ⎢ψ (n) − log n + ⎥ −ψ (1) −
2 2 n =1 ⎣ 2n ⎦ 2

Hence we have

22
 ∞
⎡ 1⎤ 1
(3.4.15) ∑ ⎢⎣ψ (n) − log n + 2n ⎥⎦ = 2 [1 + γ − log(2π )]
n =1

and we note that this series appears in (3.4.1).

With x = 0 in (3.4.13) we get

∞
⎡ 1 ⎤ 1 1
(3.4.15.1) ∑ ⎢ψ (1 + n) − log n − 2(1 + n) ⎥ = 1 + 2 γ − 2 log(2π )
n =1 ⎣ ⎦

The left-hand side of (3.4.15.1) may be written as

∞
⎡ 1 1 1 ⎤ 1 1
(3.4.15.2) ∑ ⎢ψ (n) − log n + 2n + 2n − 2(1 + n) ⎥ = 1 + 2 γ − 2 log(2π )
n =1 ⎣ ⎦

and we see that this is consistent with (3.4.15) because subtraction results in the
telescoping series

∞
⎡1 1 ⎤
∑ ⎢⎣ n − 1 + n ⎥⎦ = 1
n =1

Letting x = 1/ 2 in (3.4.13) gives us

∞
1
(3.4.16) [1 − log 2] = ∑ [ψ (1 + n) + log 2 − log(2n + 1) ]
2 n =0

The Weierstrass expression for the gamma function may be written as [49, p.1]

∞
⎡ x+a⎤
(3.4.17) log Γ( x + a ) = − log( x + a) − γ ( x + a) + ∑ ⎢log n − log ( n + a + x ) +
n =1 ⎣ n ⎥⎦
so that
∞
⎡ 1+ x ⎤
(3.4.18) log Γ(1 + x) = − log(1 + x) − γ (1 + x) + ∑ ⎢log n − log ( n + 1 + x ) +
n =1 ⎣ n ⎥⎦

Subtracting (3.4.18) from (3.4.13) results in

∞
⎡ 2x −1 1 + x ⎤
(3.4.19) ∑ ⎢ψ (1 + n) − log( x + n) − log n + log ( n + 1 + x ) + −
n =1 ⎣ 2(1 + n) n ⎥⎦

1 1
= 1 − log(2π ) − γ − x − log(1 + x)
2 2

23
 and with x = 0 we obtain

∞
⎡ 1 1⎤ 1 1
∑ ⎢ψ (1 + n) − 2 log n + log ( n + 1) − 2(1 + n) − n ⎥ = 1 − 2 γ − 2 log(2π )
n =1 ⎣ ⎦

which may be expressed as

∞
⎡ 1 ⎤ ∞ ⎡1 ⎛ 1 ⎞⎤ 1 1
∑ ⎢
n =1 ⎣
ψ (1 + n ) − log n − ⎥ − ∑ ⎢ − log ⎜1 + ⎟ ⎥ = 1 − γ − log(2π )
2(1 + n) ⎦ n =1 ⎣ n ⎝ n ⎠⎦ 2 2

and this concurs with the result obtained by letting x = 0 in (3.4.13).

□

Merkle and Merkle [40] also showed that

∞
⎡ x −1⎤
(3.4.20) ∑ ⎢⎣ψ ( x + n) −ψ (1 + n) − 1 + n ⎥⎦ = (1 − x)[ψ ( x) + γ − 1]
n =0

where a misprint in their formula has been corrected. Starting the summation at n = 1
gives us

∞
⎡ x − 1⎤
(3.4.21) ∑ ⎢⎣ψ ( x + n) −ψ (1 + n) − 1 + n ⎥⎦ = − x[ψ ( x) + γ ] + 2( x − 1)
n =1

Differentiation results in

∞
⎡ 1 ⎤
(3.4.22) ∑ ⎢⎣ψ ′( x + n) − 1 + n ⎥⎦ = − xψ ′( x) −ψ ( x) − γ + 2
n =1

so that

∞
⎡ 1 ⎤
∑ ⎢⎣ψ ′(1 + n) − 1 + n ⎥⎦ = −ψ ′(1) + 2
n =1

or equivalently

∞
⎡ 1⎤
(3.4.23) ∑ ⎢⎣ψ ′(n) − n ⎥⎦ = 1
n =1

which may also be expressed as

24
 ∞
⎡ 1⎤
∑ ⎢⎣ς (2, n) − n ⎥⎦ = 1
n =1

We note that (3.4.23) is consistent with the asymptotic formula found by Barnes in 1899
[49, p.23]

∑ψ ′(n) = log m + 1 + γ + O ( m )
m
−1
m→∞
n =1

Adding (3.4.15) and (3.4.23) gives us

∞
⎡ 1 ⎤ 1 1
(3.4.24) ∑ ⎢⎣ψ (n) − log n + 2 ψ ′(n) ⎥⎦ = 1 + 2 γ − 2 log(2π )
n =1

which, as stated by Srivastava and Choi [49, p.29], may also be derived from the
following representation of the Barnes double gamma function

1
⎡ 1 1 ⎤ ∞ Γ(n) ⎡ 1 ⎤
G (1 + x) = (2π ) 2 exp ⎢ − x( x + 1) − γ x 2 ⎥ ∏
x
exp ⎢ xψ (n) + x 2ψ ′(n) ⎥
⎣ 2 2 ⎦ n =1 Γ( x + n) ⎣ 2 ⎦

by taking logarithms of both sides and letting x = 1 in the resulting equation.

□
Differentiating (3.4.22) gives us

∑ψ ′′( x + n) = − xψ ′′( x) − 2ψ ′( x)
n =1

and we see that

∞
(3.4.25) ∑ψ
n =1
( p)
( x + n) = − xψ ( p ) ( x) − pψ ( p −1) ( x)

which was obtained in a different manner by Merkle and Merkle [40].

We need to reconcile this with the result previously found by Adamchik and Srivastava
([2] and [49, p.156]) which is valid for z ˂ 1

∞
(−1) p +1 p ! ψ ( p ) ( x + 1) + (−1) p p ! z 2 Φ ( z , p + 1, x + 1)
(3.4.26) ∑ψ ( p ) ( x + n) z n =
n =0 x p +1
+
1− z

where Φ ( z , p, x) is the Hurwitz-Lerch zeta function defined by

25
 ∞
zn
Φ ( z , p, x ) = ∑
n = 0 ( x + n)
p

It should be noted that (3.4.26) has been corrected to show that the summation starts at
n = 0 (this is because the summation in the second equation on page 157 of [49] should
also have been started at k = 0 ).

We may express (3.4.26) as

∞
(−1) p +1 p ! (−1)
p +1
p ! ⎡⎣ς ( p + 1, x + 1) − z 2 Φ ( z , p + 1, x + 1) ⎤⎦
∑ψ
n =0
( p)
( x + n) z =
n

x p +1
+
1− z

and, since lim ⎡⎣ς ( p + 1, x + 1) − z 2 Φ( z , p + 1, x + 1) ⎤⎦ = 0 , we may apply L’Hôpital’s rule to

z →1

obtain

∞
(−1) p +1 p !
∑ψ ( p ) ( x + n) =
n=0 x p +1
+ (−1) p +1 p !lim ⎡⎣ 2 zΦ ( z , p + 1, x + 1) + z 2 Φ′( z , p + 1, x + 1) ⎤⎦
z →1

(−1) p +1 p ! ⎡ ∞
nz n +1 ⎤
= p +1
+ (−1) p !lim ⎢ 2ς ( p + 1, x + 1) + ∑
p +1
p +1 ⎥
n = 0 ( x + 1 + n)
→
x z 1
⎣ ⎦

(−1) p +1 p ! ∞
2+n
=
x p +1
+ ( −1) p +1
p ! ∑
n = 0 ( x + 1 + n)
p +1

Since

∞
2+n ∞
x +1+ n ∞
1− x
∑
n = 0 ( x + 1 + n)
p +1
= ∑
n = 0 ( x + 1 + n)
p +1
+ ∑
n = 0 ( x + 1 + n)
p +1

we see that

∞
(−1) p +1 p !
∑ψ ( p ) ( x + n) =
n =0 x p +1
+ (−1) p +1 p ![ς ( p, x + 1) + (1 − x)ς ( p + 1, x + 1)]

(−1) p +1 p !
= p +1
− pψ ( p −1) ( x + 1) + (1 − x)ψ ( p ) ( x + 1)
x

1
Since ψ (1 + x) = ψ ( x) + this may be written as
x

26
 ∞

∑ψ n=0
( p)
( x + n) = (1 − x)ψ ( p ) ( x) − pψ ( p −1) ( x)

which concurs with (3.4.25).

Equation (3.4.13) may be written as

1 1 ∞
⎡ 2x −1 ⎤
log Γ(1 + x) = [log(2π ) − 1] + (1 − γ )(2 x − 1) − γ + ∑ ⎢ψ (1 + n) − log( x + n) +
2 2 n =1 ⎣ 2(1 + n) ⎥⎦

We multiply this by sin 2kπ x and integrate this to obtain

1 ∞ ⎡1 ⎤
1 1
∫ log Γ(1 + x) sin 2kπ x dx = − (1 − γ ) − ∑ ⎢ ∫ log( x + n) sin 2kπ x dx + ⎥
0
2 kπ n =1 ⎣ 0 2kπ (1 + n) ⎦

Using

1
2kπ ∫ log( x + n) sin 2kπ x dx = Ci [2kπ (n + 1)] − Ci [2kπ n] − log( n + 1) + log n
0

we have the finite sum

N 1 N
2kπ ∑ ∫ log( x + n) sin 2kπ x dx = ∑ [Ci [2kπ (n + 1)] − Ci [2kπ n] − log(n + 1) + log n ]
n =1 0 n =1

and this telescopes to

= Ci [2kπ ( N + 1)] − Ci [2kπ ] − log( N + 1)

We then have

N ⎡1 1 ⎤
2kπ ∑ ⎢ ∫ log( x + n) sin 2kπ x dx + ⎥ = Ci [2kπ ( N + 1)] − Ci [2kπ ] − log( N + 1) + H N + 2 − 1
n =1 ⎣ 0 1+ n ⎦

and thus we have the limit as N → ∞

∞ ⎡1
1 ⎤
2kπ ∑ ⎢ ∫ log( x + n) sin 2kπ x dx + ⎥ = −Ci (2kπ ) + γ − 1
n =1 ⎣ 0 1+ n ⎦

27
Hence we obtain the known integral

Ci (2kπ )
1

∫ log Γ(1 + x) sin 2kπ x dx =

0
2 kπ
□

Merkle and Merkle [40] also showed that

1 1
(3.4.27) log G ( x) = [log(2π ) − 1]( x − 1) − γ ( x − 1) 2
2 2

∞
⎡ ( x − 1) 2 ⎤
−∑ ⎢log Γ( x + n) − log Γ(1 + n) −ψ (1 + n)( x − 1) − ⎥
n =0 ⎣ 2(1 + n) ⎦

so that

1 1
(3.4.28) log G (1 + x) = [log(2π ) − 1]x − γ x 2
2 2

∞
⎡ x2 ⎤
−∑ ⎢log Γ(1 + x + n) − log Γ(1 + n) −ψ (1 + n) x − ⎥
n =0 ⎣ 2(1 + n) ⎦

It is well known that the Barnes double gamma function may be represented by [49, p.25]

∞ ⎡
1 1 1 1 ⎛ x ⎞⎤
(3.4.29) log G (1 + x) = x log(2π ) − x(1 + x) − γ x 2 + ∑ ⎢ x 2 − x + n log ⎜1 + ⎟ ⎥
2 2 2 n =1 ⎣ 2n ⎝ n ⎠⎦

and equating (3.4.28) with (3.4.29) gives us

1 ∞
⎡1 ⎛ x ⎞⎤ ∞
⎡ x2 ⎤
− x 2 + ∑ ⎢ x 2 − x + n log ⎜1 + ⎟ ⎥ = −∑ ⎢log Γ(1 + x + n) − log Γ(1 + n) −ψ (1 + n) x − ⎥
2 n =1 ⎣ 2n ⎝ n ⎠⎦ n =0 ⎣ 2(1 + n) ⎦

which may be written as

1 2 ∞ ⎡ ⎛ x⎞ ⎤
(3.4.30) x = ∑ ⎢log Γ( x + n) − log Γ(n) + n log ⎜1 + ⎟ −ψ (n) x − x ⎥
2 n =1 ⎣ ⎝ n⎠ ⎦

With x = 1 we get

1 ∞ ⎡ ⎛ 1⎞ ⎤
(3.4.31) = ∑ ⎢log n + n log ⎜1 + ⎟ −ψ (n) − 1⎥
2 n =1 ⎣ ⎝ n⎠ ⎦

28
Combining this with (3.4.15) results in

∞ ⎡ n
⎤
1 1 ⎛ 1⎞ 1
(3.4.32) 1 + γ − log(2π ) = ∑ ⎢log ⎜ 1 + ⎟ + − 1⎥
2 2 ⎣ ⎝ n ⎠ 2n ⎥⎦
n =1 ⎢

⎛ 1⎞
n
⎡ ⎛ 1 ⎞n 1 ⎤
Since lim log ⎜1 + ⎟ = log e = 1 , we note that lim ⎢log ⎜1 + ⎟ + − 1⎥ = 0 and hence,
n →∞
⎝ n⎠ n →∞
⎢⎣ ⎝ n ⎠ 2n ⎥⎦
as expected, the nth term of the series (3.4.32) approaches zero as n → ∞ .

We showed in [21] that

∞ ⎡
⎛ 1⎞ ⎤
n
1 1 ⎛ 1⎞
1 − log(2π ) = ∑ ⎢ log ⎜1 + ⎟ − 1 + log ⎜1 + ⎟ ⎥
2 ⎢2
n =1 ⎣ ⎝ n⎠ ⎝ n ⎠ ⎦⎥

and subtraction of (3.4.32) gives us the well known result [47]

∞
⎡1 ⎛ 1 ⎞⎤
γ = ∑ ⎢ − log ⎜1 + ⎟ ⎥
n
n =1 ⎣ n⎝ ⎠⎦
□

Letting x = 1/ 2 in (3.4.21) gives us

∞
⎡ ⎛ 1⎞ 1 ⎤ 1⎡ ⎛1⎞ ⎤
∑ ⎢ψ ⎝⎜ n + 2 ⎠⎟ −ψ (1 + n) + 2(1 + n) ⎥ = − 2 ⎢ψ ⎝⎜ 2 ⎠⎟ + γ ⎥ − 1
n =1 ⎣ ⎦ ⎣ ⎦

= log 2 − 1
It is easily seen that

∞
⎡ ⎛ 1⎞ ⎤ ∞ ⎡ ⎛ 1⎞ 1 ⎤ ∞ ⎡ 1 ⎤
∑ ψ
⎢ ⎜
n =1 ⎣ ⎝
n + ⎟
2⎠
− log n ⎥ = ∑ ⎢ψ ⎜ n + ⎟
2⎠
− ψ (1 + n ) + ⎥ + ∑ ⎢ψ (1 + n) − log n −
2(1 + n) ⎦ n =1 ⎣ 2(1 + n) ⎥⎦
⎦ n =1 ⎣ ⎝

and substituting (3.4.15.1) we get

∞
⎡ ⎛ 1⎞ ⎤ 1 1 2
(3.4.33) ∑ ⎢ψ ⎜⎝ n + 2 ⎟⎠ − log n ⎥ = 2 γ + 2 log π
n =1 ⎣ ⎦

This was obtained in a different way in [25].

Another derivation of (3.4.33) is shown below. Differentiating (3.4.30) gives us

29
 ∞
⎡ x ⎤
(3.4.34) x = ∑ ⎢ψ ( x + n) −ψ (n) −
n =1 ⎣ n + x ⎥⎦

and with x = 1/ 2 we have

1 ∞ ⎡ ⎛ 1⎞ 1 ⎤
= ∑ ⎢ψ ⎜ n + ⎟ −ψ (n) −
2 n =1 ⎣ ⎝ 2⎠ 2n + 1 ⎥⎦

∞
⎡ ⎛ 1⎞ 1 ⎤
= ∑ ⎢ψ ⎜ n + ⎟ − log n + log n −ψ (n) −
n =1 ⎣ ⎝ 2⎠ 2n + 1 ⎥⎦

∞
⎡ ⎛ 1⎞ ⎤ ∞ ⎡ 1 ⎤
= ∑ ⎢ψ ⎜ n + ⎟ − log n ⎥ − ∑ ⎢ψ (n) − log n +
n =1 ⎣ ⎝ 2⎠ ⎦ n =1 ⎣ 2n + 1 ⎥⎦

∞
⎡ ⎛ 1⎞ ⎤ ∞ ⎡ 1 1 ⎤
= ∑ ⎢ψ ⎜ n + ⎟ − log n ⎥ − ∑ ⎢ψ (n + 1) − log n − +
n =1 ⎣ ⎝ 2⎠ ⎦ n =1 ⎣ n 2n + 1 ⎥⎦

∞
⎡ ⎛ 1⎞ ⎤ ∞ ⎡ 1 1 1 1 ⎤
= ∑ ⎢ψ ⎜ n + ⎟ − log n ⎥ − ∑ ⎢ψ (n + 1) − log n − + − +
n =1 ⎣ ⎝ 2⎠ ⎦ n =1 ⎣ 2(1 + n) 2(1 + n) n 2n + 1 ⎥⎦

∞
⎡ ⎛ 1⎞ ⎤ ∞ ⎡ 1 ⎤
= ∑ ⎢ψ ⎜ n + ⎟ − log n ⎥ − ∑ ⎢ψ (n + 1) − log n −
n =1 ⎣ ⎝ 2⎠ ⎦ n =1 ⎣ 2(1 + n) ⎥⎦

∞
⎡ 1 1 1 ⎤
−∑ ⎢ − +
n =1 ⎣ 2(1 + n ) n 2n + 1 ⎥⎦

Using (3.4.15.1) we obtain

1 ∞ ⎡ ⎛ 1⎞ ⎤ 1 1 ∞
⎡ 1 1 1 ⎤
= ∑ ⎢ψ ⎜ n + ⎟ − log n ⎥ − 1 − γ + log(2π ) − ∑ ⎢ − +
2 n =1 ⎣ ⎝ 2⎠ ⎦ 2 2 n =1 ⎣ 2(1 + n) n 2n + 1 ⎥⎦

We see that

∞
⎡ 1 1 1 ⎤ ∞ ⎡ 1 1⎤ ∞ ⎡ 1 1 ⎤
∑ ⎢ 2(1 + n) − n + 2n + 1 ⎥ = ∑ ⎢ 2n + 2 − 2n ⎥ + ∑ ⎢ 2n + 1 − 2n ⎥
n =1 ⎣ ⎦ n =1 ⎣ ⎦ n =1 ⎣ ⎦

1 ∞ ⎡ 1 1⎤ 1 ∞ ⎡ 1 1⎤
= ∑⎢ − ⎥ + ∑⎢ − ⎥
2 n =1 ⎣ n + 1 n ⎦ 2 n =1 ⎢⎣ n + 2 n ⎥⎦
1

and using (3.4.14)

30
 ∞
⎡ 1 1⎤
ψ (1 + x) = −γ − ∑ ⎢ − ⎥
n =1 ⎣ x + n n⎦
we obtain

∞
⎡ 1 1 1 ⎤ 1 1
∑ ⎢ 2(1 + n) − n + 2n + 1⎥ = − 2 [ψ (2) + γ ] − 2 [ψ (3 / 2) + γ ]
n =1 ⎣ ⎦

1
= − + log 2 − 1
2
and then deduce (3.4.33).
□

In passing, we note that Guinand [33] has shown that for arg z ˂ π

∞
⎡ 1 ⎤ 1
(3.4.35) ∑ ⎢⎣ψ (1 + nz ) − log(nz ) − 2nz ⎥⎦ + 2 z [γ − log(2π z )]
n =1

1 ∞ ⎡ ⎛ n⎞ ⎛n⎞ z ⎤ 1⎡ ⎛ 2π ⎞⎤
= ∑ ⎢ψ ⎜1 + ⎟ − log ⎜ ⎟ − ⎥ + ⎢γ − log ⎜ ⎟⎥
z n =1 ⎣ ⎝ z ⎠ ⎝ z ⎠ 2n ⎦ 2 ⎣ ⎝ z ⎠⎦

and with z = 2 we have

∞
⎡ 1⎤ 1
∑ ⎢⎣ψ (1 + 2n) − log(2n) − 4n ⎥⎦ + 4 [γ − log(4π )]
n =1

1 ∞ ⎡ ⎛ n⎞ ⎛ n ⎞ 1⎤ 1
= ∑ ⎢ψ ⎜1 + ⎟ − log ⎜ ⎟ − ⎥ + [γ − log π ]
2 n =1 ⎣ ⎝ 2 ⎠ ⎝ 2 ⎠ n⎦ 2

We recall (3.4.29)

1 1 1 ∞
⎡1 ⎛ x ⎞⎤
log G (1 + x) = x log(2π ) − x(1 + x) − γ x 2 + ∑ ⎢ x 2 − x + n log ⎜1 + ⎟ ⎥
2 2 2 n =1 ⎣ 2n ⎝ n ⎠⎦

Batir [8] noted that

⎛ x⎞ ⎛1 1 ⎞
x
1 2
x − x + n log ⎜1 + ⎟ = ∫ ⎜ − ⎟ t dt
2n ⎝ n⎠ 0⎝k t+k ⎠

31
and we have the summation

∞
⎡1 2 ⎛ x ⎞⎤ ∞ ⎛ 1 1 ⎞
x

∑ ⎢ 2n
n =1 ⎣
x − x + n log ⎜1 + ⎟ ⎥ = ∑ ∫ ⎜ − ⎟ t dt
⎝ n ⎠ ⎦ n =1 0 ⎝ k t + k ⎠

x ∞
⎛1 1 ⎞
= ∫ ∑⎜ − ⎟ tdt
0 n =1 ⎝
k t+k ⎠

x
= ∫ [ψ (1 + t ) + γ ] t dt
0

where we have used (3.4.14).

Therefore having regard to (3.4.29) we see that

x
1 1 1
log G (1 + x) = x log(2π ) − x(1 + x) − γ x 2 + ∫ [ψ (1 + t ) + γ ] t dt
2 2 2 0

and integration by parts results in an easy derivation of Alexeiewsky’s theorem [49, p.32]

x
1 1
(3.4.36) log G (1 + x) = x log(2π ) − x(1 + x) + x log Γ(1 + x) − ∫ log Γ(1 + t ) dt
2 2 0

sin ax
Letting f ( x) = in (2.15) gives us
x

∞ ∞
1 ∞
sin na sin ax ∞
sin ax cos 2π nx
a+∑ =∫ dx + 2∑ ∫ dx
2 n =1 n 0
x n =1 0 x

where we have used f (0) = a .

It is well known that

∞ ∞
sin ax sin v π
∫0 x dx = ∫0 v dv = 2

and we have

32
 sin ax cos 2π nx 1 sin(2π n + a) x − sin(2π n − a ) x
N N

∫0 x
dx = ∫
20 x
dx

N
1
= [ Si (2π n + a) x − Si (2π n − a) x]
2 0

1
= [ Si (2π n + a) N − Si (2π n − a ) N ]
2

(2π n + a ) N (2π n − a ) N
1⎡ sin x sin ax ⎤
= ⎢
2 ⎢⎣ ∫
0
x
dx − ∫
0
x
dx ⎥
⎥⎦

(2π n + a ) N
1 sin x
= ∫
2 (2π n − a ) N x
dx

which gives us the limit

∞
sin ax cos 2π nx
∫
0
x
dx = 0

We therefore obtain the well known Fourier series

∞
sin na 1
∑
n =1 n
= (π − a)
2
□

cos ax − cos π x
Letting f ( x) = in (2.15) gives us
x

∞ ∞
∞
cos na − (−1) n cos ax − cos π x ∞
[cos ax − cos π x]cos 2π nx
∑
n =1 n
=∫
x
dx + 2∑ ∫
n =1 0 x
dx
0

where we have used f (0) = 0 .

μ μ
cos ax − cos bx cos ax − 1 + 1 − cos bx
∫0 x
dx = ∫
0
x
dx

μ μ
cos ax − 1 cos bx − 1
=∫ dx − ∫ dx
0
x 0
x

33
 aμ bμ
cos x − 1 cos x − 1
= ∫
0
x
dx − ∫0
x
dx

b
= Ci (a μ ) − Ci (bμ ) + log
a

where we have used (3.1.3). We therefore obtain the limit

∞
cos ax − cos bx b
∫
0
x
dx = log
a

as reported in [5, p.301] and [48, p.282]. In particular we have

∞
cos ax − cos π x π
∫
0
x
dx = log
a

We have

∞ ∞
[cos ax − cos π x]cos 2π nx 1 cos(2π n + a ) x + cos(2π n − a ) x
∫0 x
dx = ∫
20 x
dx

∞
1 cos(2n + 1)π x + cos(2n − 1)π x
− ∫ dx
20 x

∞ ∞
1 cos(2π n + a ) x − cos(2n + 1)π x 1 cos(2π n − a ) x − cos(2n − 1)π x
= ∫ dx + ∫ dx
20 x 20 x

1 (2n + 1)π 1 (2n − 1)π

= log + log
2 (2π n + a ) 2 (2π n − a)

We see that

∞
cos na − (−1) n ∞
cos na
∑
n =1 n
= ∑
n =1 n
+ log 2

and using the familiar trigonometric series shown in Carslaw’s book [17, p.241]

∞
cos na
log ⎡⎣ 2sin ( a / 2 ) ⎤⎦ = −∑ (0 < a < 2 π )
n =1 n
we obtain

34
 ∞
4n 2 − a 2
log sin(aπ / 2) = log a + ∑ log
n =1 4n 2 − 1

∞
4n 2 − a 2 ∞ 4n 2 − 1
= log a + ∑ log − ∑ log
n =1 4n 2 n =1 4n 2

Using the Wallis product

π N
2n 2n
= lim ∏ ⋅
2 N →∞
n =1 2n − 1 2n + 1

we see that
π ∞
4n 2 − 1
log = −∑ log
2 n =1 4n 2

and we obtain

4n 2 − a 2
∞
log sin(π a / 2) = log(π a / 2) + ∑ log
n =1 4n 2

or equivalently

n2 − a 2
∞
log sin(π a ) = log(π a ) + ∑ log
n =1 n2

which corresponds with the Euler product formula for the sine function

∞
⎛ a2 ⎞
sin π a = π a∏ ⎜1 − 2 ⎟
n =1 ⎝ n ⎠
□

Integration of (3.4.7) with respect to t results in

∞
⎡ 1 1 ⎤
(3.4.7) ∫ ⎢log Γ( x + u ) − log Γ(u ) − ( x + u ) log( x + u ) + u log u + x + log( x + u ) − log u ⎥ du
0 ⎣ ⎦
2 2

x
= [− x + 2( x − 1) log x] + log G (1 + x) − x log Γ( x)
4

where we have used (3.4.36).

35
Integrating (3.4.20) gives us

∞
⎡ 1 ⎛1 2 ⎞⎤ ⎛1 2 ⎞
x
1 2
∑ ⎢ log Γ( x + n) − log Γ(n) − xψ (1 + n) − 1 + n ⎜ 2 x − x ⎟ ⎥ = − 2 x γ + 2 ⎜ 2 x − x ⎟ − ∫ uψ (u ) du
n =1 ⎣ ⎝ ⎠⎦ ⎝ ⎠ 0

and we have

x x

∫ uψ (u )du = x log Γ( x) − ∫ log Γ(u ) du

0 0

This gives us

∞
⎡ 1 ⎛1 ⎞⎤ 1 1
∑ ⎢log Γ( x + n) − log Γ(n) − xψ (1 + n) − 1 + n ⎜⎝ 2 x 2
− x ⎟⎥ = − γ x 2 + ( x 2 − 3x )
n =1 ⎣ ⎠⎦ 2 2

1
− log G (1 + x) + x log(2π )
2

It is easily seen that this concurs with (3.4.36).

3.5 Stieltjes constants

The Stieltjes constants γ p ( x) are the coefficients of the Laurent expansion of the Hurwitz
zeta function ς ( s, x) about s = 1

∞
1 1 ∞
(−1) p
(3.5.1) ς ( s, x ) = ∑ = + ∑ γ p ( x)( s − 1) p
n =0 ( n + x ) s
s − 1 p =0 p !

and we have [54]

(3.5.2) γ 0 ( x) = −ψ ( x)

With x = 1 equation (3.5.1) reduces to the Riemann zeta function

1 ∞
(−1) p
ς (s) = +∑ γ p ( s − 1) p
s − 1 p =0 p !

where γ p (1) = γ p .

We write (2.15) in the form

36
 1 ⎡N N
⎤ ∞ ∞
(3.5.2.1) f (0) + lim ⎢ ∑ f (n) − ∫ f ( x) dx ⎥ = 2∑ ∫ f ( x) cos 2π nx dx
2 N →∞
⎣ n =1 0 ⎦ n =1 0

log p (1 + x)
and, with the function f ( x) = , this results in
1+ x

∞ ∞
⎡ N log p (1 + n) log p +1 (1 + N ) ⎤ log p (1 + x)
lim ⎢ ∑ − ⎥ = 2 ∑ ∫ 1 + x cos 2π nx dx
N →∞
⎣ n =1 1 + n 1+ p ⎦ n =1 0

since f (0) = 0 for p ≥ 1 .

Simple algebra gives us

N
log p (1 + n) log p +1 (1 + N ) N log p n log p (1 + N ) log p +1 (1 + N )
∑
n =1 1+ n
−
1+ p
=∑
n =1 n
+
1+ N
−
1+ p

N
log p n log p +1 N ⎡ log p +1 N log p +1 (1 + N ) ⎤ log p (1 + N )
=∑ − +⎢ − ⎥+
n =1 n 1+ p ⎣ 1+ p 1+ p ⎦ 1+ N

and it is easily shown that the expression in parentheses vanishes as N → ∞ . Successive

log p (1 + N )
applications of L’Hôpital’s rule shows that lim = 0 and hence we have
N →∞ 1+ N

⎡ N log p (1 + n) log p +1 (1 + N ) ⎤ ⎡ N log p n log p +1 N ⎤

(3.5.3) lim ⎢ ∑ − ⎥ N →∞ ⎢ ∑
= lim − ⎥
N →∞
⎣ n =1 1 + n 1+ p ⎦ ⎣ n =1 n 1+ p ⎦

It is well known that Stieltjes [35] proved in 1885 that the Stieltjes constants γ p may be
represented by

⎡ N log p n log p +1 N ⎤
(3.5.4) lim ⎢ ∑ − ⎥ =γp
N →∞
⎣ n =1 n 1+ p ⎦

We see that

∞ ∞
log p (1 + x) log p (1 + x)
∫ cos 2π nx
0
1+ x
dx = ∫ cos 2π n(1 + x)
0
1 + x
dx

and an elementary substitution gives us

37
 ∞ ∞
log p (1 + x) log p u
∫ cos 2π n(1 + x)
0
1+ x
dx = ∫ cos 2π nu
1
u
du

Hence we obtain for p ≥ 1

∞ ∞
log p x
(3.5.5) γ p = 2∑ ∫ cos 2π nx dx
n =1 1 x

which was originally derived by Briggs [14] in 1955.

It should be noted that this only applies for p ≥ 1 . From Knopp’s book [38, p.521] the
Euler-Maclaurin summation formula gives us

∞ ∞
1 sin 2π nx
(3.5.6) γ0 = γ = + ∑∫ dx
2 n =1 1 nπ x 2

and integration by parts gives us

∞ ∞ ∞ ∞
cos 2π nx sin 2π nx sin 2π nx sin 2π nx
∫1 x dx = 2π nx 1 + ∫1 2π nx 2 dx = ∫1 2π nx 2 dx
Hence we have an additional factor of 1/2 in the case where p = 0

∞ ∞
1 cos 2π nx
(3.5.7) γ 0 = γ = + 2∑ ∫ dx
2 n =1 1 x

This factor arises because f (0) = 1 in (3.5.2.1) when p = 0 .

We note from (3.5.5) that

∞ ∞
dp cos 2π nx
γ p = 2(−1) p
ds p
∑∫
n =1 1 xs
dx
s =1
□

log p (a + x)
The above analysis may be generalised by considering the function f ( x) = .
a+x
Proceeding as before results in

∞ ∞
1 log p a ⎡ N log p (a + n) log p +1 (a + N ) − log p +1 a ⎤ log p (a + x)
+ lim ⎢ ∑ − ⎥ = 2 ∑ ∫ a + x cos 2π nx dx
2 a N →∞
⎣ n =1 a+n p +1 ⎦ n =1 0

38
which may be written as

∞ ∞
1 log p a log p +1 a ⎡ N log p (a + n) log p +1 (a + N ) ⎤ log p (a + x)
− + + lim ⎢ ∑ − ⎥ = 2 ∑ ∫ a + x cos 2π nx dx
2 a p +1 N →∞
⎣ n =0 a+n p +1 ⎦ n =1 0

We will see below in (3.7.17) that for p ≥ 1

∞
1 log p a log p +1 a ∞
log p (a + x)
γ p (a) = − + 2∑ ∫ cos 2π nx dx
2 a p +1 n =1 0 a+x

and hence we have

⎡ N
log p (a + n) log p +1 (a + N ) ⎤
(3.5.8) γ p (a) = lim ⎢ ∑ − ⎥
N →∞
⎣ n =0 a+n p +1 ⎦

as previously shown by Berndt [9] in 1972. With a = 1 in (3.5.8) and using (3.5.3) we see
that Stieltjes’s formula (3.5.4) immediately follows.
□

We see that

∞
1 log p (1 + a ) log p +1 (1 + a ) ∞
log p (a + 1 + x)
γ p (1 + a) = − + 2∑ ∫ cos 2π nx dx
2 1+ a p +1 n =1 0 a +1+ x

∞
1 log p (1 + a ) log p +1 (1 + a ) ∞
log p (a + 1 + x)
= − + 2∑ ∫ cos 2π n(1 + x) dx
2 1+ a p +1 n =1 0 a +1+ x

∞
1 log p (1 + a ) log p +1 (1 + a) ∞
log p (a + x)
= − + 2∑ ∫ cos 2π nx dx
2 1+ a p +1 n =1 1 a+x

Therefore we have

γ p (1 + a) − γ p (a)

1
1 log p (1 + a ) 1 log p a log p +1 (1 + a ) log p +1 a ∞
log p (a + x)
= − − + − 2∑ ∫ cos 2π nx dx
2 1+ a 2 a p +1 p +1 n =1 0 a+x

We see from (2.13) that

39
 ∞ 1
1 ⎡ log p a log p (a + 1) ⎤
1
log p (a + x) log p (a + x)
2∑ ∫
a + 1 ⎦ ∫0
cos 2π nx dx = ⎢ + ⎥ − dx
n =1 0 a+x 2⎣ a a+x

1 ⎡ log p a log p (a + 1) ⎤ ⎡ log p +1 (a + 1) log p +1 a ⎤

= ⎢ + ⎥−⎢ − ⎥
2⎣ a a +1 ⎦ ⎣ p +1 p +1 ⎦

and so we obtain

log p a
(3.5.9) γ p (1 + a) − γ p (a) = −
a

This identity may be obtained more directly as follows:

1
Since ς ( s,1 + a ) − ς ( s, a) = − we have
as

log p a
ς ( p ) ( s,1 + a) − ς ( p ) ( s, a ) = (−1) p +1
as

and therefore using

∂p
γ p ( x) − γ p ( y ) = lim(−1) p [ς ( s, x) − ς ( s, y )]
s →1 ∂s p

we obtain
log p a
γ p (1 + a) − γ p (a) = −
a

3.6 Connections with other summation formulae

The following integral formula was originally obtained by Coffey in 2007 for the Stieltjes
constants

∞
1 log p a log p +1 a (a − ix) log p (a + ix) − (a + ix ) log p (a − ix )
(3.6.1) γ p (a) = − + i∫ 2π x
dx
2 a p +1 0
( a 2
+ x 2
)( e − 1)

and a different derivation appears in [24].

It should be noted that

∞
(a − ix) log p (a + ix) − ( a + ix) log p ( a − ix)
∫0 (a 2 + x 2 )(e 2π x − 1)
dx

40
 ∞
⎡ log p (a + ix) log p (a − ix) ⎤ dx
= ∫⎢ − ⎥
0 ⎣
a + ix a − ix ⎦ e 2π x − 1

The structure of the integral in (3.6.1) therefore indicates the close connection with the
Abel-Plana summation formula [49, p.90]
∞ ∞
1 ∞
f (ix) − f (−ix)
(3.6.2) f (0) + ∑ f (n) = ∫ f ( x) dx + i ∫ 2π x
dx
2 n =1 0 0
e − 1

which applies to functions which are analytic in the right-hand plane and satisfy the
convergence condition lim e −2π y f ( x + iy ) = 0 uniformly on any finite interval of x .
y →∞

Derivations of the Abel-Plana summation formula are given in [6], [34, p.339], [52,
p.145] and [51, p.118].

Adamchik [4] noted that the Hermite integral for the Hurwitz zeta function may be
derived from the Abel-Plana summation formula.
□

With f ( x) → f (2 x) in (3.6.2) we have

∞ ∞
1 ∞
f (2ix) − f (−2ix)
(3.6.3) f (0) + ∑ f (2n) = ∫ f (2 x) dx + i ∫ 2π x
dx
2 n =1 0 0
e − 1

∞ ∞
1 1 f (iu ) − f (−iu )
=
20∫ f (u ) du + i ∫
2 0 eπ u − 1
du

Subtraction results in
∞ ∞
1 ∞ ∞
f (ix) − f (−ix) f (ix) − f (−ix)
f (0) + 2∑ f (2n) −∑ f (n) = i ∫ πx
dx − i ∫ dx
2 n =1 n =1 0
e −1 0
e 2π x − 1

∞ ∞ ∞
and, since for suitably convergent series 2∑ a2 n − ∑ an = ∑ (−1) n an , we have
n =1 n =1 n =1

∞
1 ∞
f (ix) − f (−ix)
(3.6.4) f (0) + ∑ (−1) n f (n) = i ∫ dx
2 n =1 0
2 sinh π x

which is the alternating form of the Abel-Plana summation formula originally derived by
Abel [16]. This may be compared with (2.16).

41
For completeness, we note that Saharian [46] reports the summation
∞ ∞
∞
⎛ 1⎞ f (ix) − f (−ix)
(3.6.5) ∑
n =0
f ⎜ n + ⎟ = ∫ f ( x) dx − i ∫
⎝ 2⎠ 0 e 2π x + 1
dx
0

Letting f ( x) → f ( x) cos π x in (3.6.2) results in

∞ ∞
1 ∞
f (ix) − f (−ix)
f (0) + ∑ (−1) n f (n) = ∫ f ( x) cos π x dx + i ∫ 2π x
cos iπ x dx
2 n =1 0 0
e − 1

and this may be written as

∞ ∞
1 ∞
f (ix) − f (−ix)
(3.6.6) f (0) + ∑ (−1)n f (n) = ∫ f ( x) cos π x dx + i ∫ cosh π x dx
2 n =1 0 0
e 2π x − 1

and equating this with (3.6.4) gives us

∞ ∞ ∞
f (ix) − f (−ix) f (ix) − f (−ix)
i∫ dx = ∫ f ( x) cos π x dx + i ∫ cosh π x dx
0
2sinh π x 0 0
e 2π x
− 1

which simplifies to
∞ ∞
1 [ f (ix) − f (−ix)]e−π x
∫0
f ( x) cos π x dx = − i ∫
2 0 e 2π x − 1
dx

Comparing (3.6.2) and (2.15) we see that

∞ ∞
f (ix) − f (−ix) ∞
(3.6.7) i∫
e 2π x − 1
dx = 2 ∑ ∫ f ( x) cos 2π nx dx
n =1 0
0

for functions which satisfy the conditions relevant to both the Abel-Plana summation
formula and the Poisson summation formula. An application of (3.6.7) is shown below.

With f ( x) = e − ax we have

∞ ∞
f (ix) − f (−ix) sin(ax)
i∫ 2π x
dx = 2∫ 2π x dx
0
e −1 0
e −1

42
Therefore (3.6.7) gives us
∞ ∞ ∞
sin(ax)
dx = ∑
∫0 e2π x − 1 n=1 ∫0 e cos 2π nx dx
− ax

We easily obtain the indefinite integral

e − ax [2π n sin 2π n − a cos 2π nx]

∫ e cos 2π nx dx =
− ax

a 2 + 4π 2 n 2
so that
∞
a
∫e cos 2π nx dx =
− ax
(3.6.8)
0
a + 4π 2 n 2
2

We then have
∞ ∞
sin(ax) a
∫0 e2π x − 1 ∑
dx =
n =1 a + 4π n
2 2 2

We may then use the well known identity ([15, p.296], [38, p.378])

∞
a 1 1 1
(3.6.9) 2∑ = a − +
n =1 a + 4π n e −1 a 2
2 2 2

to determine Legendre’s relation [52, p.119]

∞
sin(ax) 1 1 1 1 a 1
(3.6.10) 2∫ 2π x
dx = a − + = coth −
0
e −1 e −1 a 2 2 2 a

A rigorous derivation of this result is shown in Bromwich’s book [15, p.501]. It is

interesting to note that Apostol [5, p.334] employed Poisson’s formula to derive the
corresponding partial fraction decomposition formula for coth x . A slightly simplified
version of this proof is set out below.

With f ( x) = e − ax in (2.15) we have

∞ ∞
1 ∞ − nx ∞
+ ∑ e = ∫ e − ax dx + 2∑ ∫ e − ax cos 2π nx dx
2 n =1 0 n =1 0

so that upon using (3.6.8) we immediately obtain (3.6.9).

43
Various applications of the Abel-Plana summation formula are contained in Ramanujan’s
Notebooks, Part V [11].

As noted by Ivić [36, p.6] the identity (3.6.9) may be readily obtained from Euler’s
infinite product representation of the sin x function

∞
⎛ x2 ⎞
sin x = x∏ ⎜1 − 2 2 ⎟
n =1 ⎝ nπ ⎠

which was derived earlier in this paper using the Poisson summation formula.

1
The substitution x = − iu gives us
2

⎛1 ⎞ 1 ∞ ⎛ u2 ⎞
sin ⎜ iu ⎟ = iu ∏ ⎜1 + 2 ⎟
⎝ 2 ⎠ 2 n =1 ⎝ (2π n) ⎠
so that
⎛ 1 ⎞ 1 ∞ ⎛ (2π n) + u ⎞
2 2
sinh ⎜ u ⎟ = u ∏ ⎜ ⎟
⎝ 2 ⎠ 2 n =1 ⎝ (2π n)
2
⎠

Logarithmic differentiation then results in

∞
1 ⎛1 ⎞ 1 u
coth ⎜ u ⎟ = + 2∑ 2
n =1 u + 4π n
2 2
2 ⎝2 ⎠ u

which is equivalent to (3.6.9).

□

1
Letting f ( x) = in (3.6.7) gives us
u+x
∞ ∞
x ∞
cos 2π nx
∫0 (u 2 + x 2 )(e2π x − 1) ∑
dx = ∫
n =1 0 u+x
dx

∞
= ∑ [cos(2nπ u )Ci (2nπ u ) + sin(2nπ u ) si (2nπ u )]
n =1

and using (3.1.7)

∞
1
ψ (u ) = log u − + 2∑ [cos(2nπ u )Ci (2nπ u ) + sin(2nπ u ) si (2nπ u )]
2u n =1

44
we obtain Binet’s formula (3.4.5)
∞
1 x
log u −ψ (u ) = + 2∫ 2 dx
2u 0
(u + x )(e 2π x − 1)
2

We also note the recent paper by Butzer et al. [16] “The Summation Formulae of Euler–
Maclaurin, Abel–Plana, Poisson, and their interconnections with the Approximate
Sampling Formula of Signal Analysis” where it is shown that these four fundamental
formulae are all equivalent, in the sense that each is a corollary of any of the others.

As mentioned in [16], the Abel-Plana summation formula may be expressed as

⎡N N+
1
⎤ ∞
1 ⎢
2
⎥ f (ix) − f (−ix)
(3.6.11) f (0) + lim ⎢ ∑ f (n) − ∫ f ( x) dx ⎥ = i ∫ 2π x
dx
2 N →∞
n =1 e − 1
⎢⎣ 0
⎥⎦ 0

and with f ( x) = x log(1 + x) we have

∞
f (ix) − f (−ix) x log (1 + x
∞ 2
) dx
i∫ dx = − ∫0 e2π x − 1
0
e 2π x − 1

This gives us

⎡ N + 12 ⎤ ∞
⎢
N
⎥ x log (1 + x 2 )
x log(1 + x) dx − ∑ n log(1 + n) ⎥ = ∫
N →∞ ⎢ ∫
(3.6.12) lim dx
n =1 e 2π x − 1
⎢⎣ 0
⎥⎦ 0

We note that

N N +1

∑ n log(1 + n) = ∑ (m − 1) log m
n =1 m=2

N +1
= ∑ (m − 1) log m
m =1

N
= ∑ (n − 1) log n + N log( N + 1)
n =1

N
⎛ 1⎞
= ∑ (n − 1) log n + N log N + N log ⎜1 + ⎟
n =1 ⎝ N⎠

45
and we have

1
N+
2
1⎛ 3⎞ ⎛ 3⎞ 1⎛ 1⎞ 1⎛ 1⎞
∫
0
x log(1 + x) dx = ⎜ N 2 + N − ⎟ log ⎜ N + ⎟ − ⎜ N 2 + N + ⎟ + ⎜ N + ⎟
2⎝ 4⎠ ⎝ 2⎠ 4⎝ 4⎠ 2⎝ 2⎠

1⎛ 3⎞ 1⎛ 3⎞ ⎛ 3 ⎞ 1⎛ 2 1⎞ 1⎛ 1⎞
= ⎜ N 2 + N − ⎟ log N + ⎜ N 2 + N − ⎟ log ⎜1 + ⎟− ⎜N + N + ⎟+ ⎜N + ⎟
2⎝ 4⎠ 2⎝ 4⎠ ⎝ 2N ⎠ 4 ⎝ 4⎠ 2⎝ 2⎠

1
N+
2 N
Designating I N = ∫ x log(1 + x) dx − ∑ n log(1 + n) we then have
0 n =1

1
N+
2 N N
⎛ 1⎞
IN = ∫ x log(1 + x) dx − ∑ n log(1 + n) = −∑ (n − 1) log n − N log ⎜1 + ⎟
0 n =1 n =1 ⎝ N⎠

1⎛ 3⎞ 1⎛ 3⎞ ⎛ 3 ⎞ 1⎛ 2 1⎞ 1⎛ 1⎞
+ ⎜ N 2 − N − ⎟ log N + ⎜ N 2 + N − ⎟ log ⎜1 + ⎟− ⎜N + N + ⎟+ ⎜N + ⎟
2⎝ 4⎠ 2⎝ 4⎠ ⎝ 2N ⎠ 4 ⎝ 4⎠ 2⎝ 2⎠

⎛ u⎞ u u2
Since log ⎜ 1 + ⎟ = − 2
+ O ( N −3 ) we have
⎝ N ⎠ N 2 N

1⎛ 2 3⎞ ⎛ 3 ⎞ 1⎛ 2 3⎞⎡ 3 9 ⎤
⎜ N + N − ⎟ log ⎜1 + ⎟ = ⎜N + N − ⎟⎢ − 2
+ O ( N −3 ) ⎥
2⎝ 4⎠ ⎝ 2N ⎠ 2 ⎝ 4 ⎠ ⎣ 2N 8N ⎦

3 9 3
= N − + + O ( N −1 )
4 16 4

We then have

⎡ N + 12 ⎤
⎢ ⎥
N
x log(1 + x) dx − ∑ n log(1 + n) ⎥
N →∞ ⎢ ∫
(3.6.13) lim
n =1
⎢⎣ 0 ⎥⎦

⎡ N 1⎛ 3⎞ ⎤ 1 5
= lim ⎢ −∑ (n − 1) log n + ⎜ N 2 − N − ⎟ log N ⎥ + N − N 2 −
N →∞
⎣ n =1 2⎝ 4⎠ ⎦ 4 8

Using the Euler-Maclaurin summation formula, Hardy [34, p.333] showed that the
Riemann zeta function could be expressed as follows:

46
 ⎡ n
1 n1− s 1 − s ⎤
ς ( s ) = lim ⎢ ∑ − − n ⎥ Re( s ) > −1
⎣ k =1 k 1 − s 2
n →∞ s
⎦

⎡ n
1 n1− s 1 − s 1 − s −1 ⎤
ς ( s) = lim ⎢ ∑ − − n + sn ⎥ Re( s ) > −3
⎣ k =1 k 1 − s 2 12
n →∞ s
⎦

Differentiating the first identity results in for Re( s ) > −1

⎡ log k n1− s (1 − s ) log n − n1− s 1 − s

n
⎤
ς ′( s ) = lim ⎢ −∑ + + n log n ⎥
n →∞
⎣ k =1 k
s
(1 − s ) 2
2 ⎦

and with s = 0 we obtain

⎡ n
⎛ 1⎞ ⎤
ς ′(0) = lim ⎢ −∑ log k + ⎜ n + ⎟ log n − n ⎥
⎣
n →∞
k =1 2 ⎝ ⎠ ⎦

1
Hence, using the Stirling approximation we see that ς ′(0) = − log(2π ).
2

For Re( s ) > −3 we have

⎡ n log k n1− s (1 − s ) log n − n1− s 1 − s 1 1 ⎤

ς ′( s ) = lim ⎢ −∑ s + + n log n − sn − s −1 log n + n − s −1 ⎥
n →∞
⎣ k =1 k (1 − s ) 2
2 12 12 ⎦

and with s = −1 we get (cf [49, p.25] re the definition of the Glaisher-Kinkelin constant)

⎡ n ⎛ n2 n 1 ⎞ n2 1 ⎤
ς ′(−1) = lim ⎢ −∑ k log k + ⎜ + + ⎟ log n − + ⎥
n →∞
⎣ k =1 ⎝ 2 2 12 ⎠ 4 12 ⎦

We then have

3 1 ⎡ n ⎛ n2 n 5 ⎞ n2 2 ⎤
(3.6.14) ς ′(−1) − + log(2π ) = lim ⎢ −∑ (k − 1) log k + ⎜ − − ⎟ log n + n − + ⎥
4 2 n →∞
⎣ k =1 ⎝ 2 2 12 ⎠ 4 3⎦

Adamchik [3] has shown that

∞ x log 1 + x 2( ) dx = ς ′(−1) − 3 + 1 log(2π )

(3.6.15) ∫
0
e 2π x
−1 4 2

47
It is clear that further work is required to resolve the two discrepancies between (3.6.13)
and (3.6.14).

3.7 Hurwitz zeta function

1
We now consider the function f ( x) = , where Re ( s) > 1, and using (2.15) we
(a + x) s
obtain
∞ ∞
1 1 ∞ 1 dx ∞
cos 2π nx
+∑ =∫ + 2∑ ∫ dx
2 a n =1 (a + n)
s s
0
(a + x) s
n =1 0 ( a + x )
s

which may be written as

∞
∞
1 1 1 1 ∞
cos 2π nx
∑
n = 0 ( a + n)
s
= +
2 a ( s − 1)a
s s −1
+ 2 ∑ ∫
n =1 0 ( a + x )
s
dx

The Hurwitz zeta function is defined for Re ( s ) > 1 as

∞
1
ς ( s, a ) = ∑
n = 0 ( a + n)
s

and we therefore have

∞
1 1 1 ∞
cos 2π nx
(3.7.1) ς ( s, a ) = +
2 a ( s − 1)a
s s −1
+ 2 ∑ ∫
n =1 0 ( a + x )
s
dx

which corrects a misprint in Mordell’s paper [41].

Using (2.13) we obtain

1⎡ 1 1 ⎤ cos 2π nx
1 ∞ 1
dx
⎢ + s ⎥
=∫ + 2∑ ∫ dx
2 ⎣ a (a + 1) ⎦ 0 (a + x)
s s
n =1 0 ( a + x )
s

and hence we have

1⎡ 1 1 ⎤ 1 ⎡ 1 1 ⎤ ∞ 1
cos 2π nx
(3.7.2) ⎢ + s ⎥
=− ⎢ s −1
− s −1 ⎥ + 2∑ ∫ dx
2 ⎣ a (a + 1) ⎦
s
s − 1 ⎣ (a + 1) a ⎦ n =1 0 ( a + x )
s

Subtracting (3.7.2) from (3.7.1) gives us

∞
1 1 1 1 ∞
cos 2π nx
(3.7.3) ς ( s, a ) = + +
a 2 (a + 1) ( s − 1)(a + 1)
s s s −1
+ 2 ∑ ∫
n =1 1 ( a + x )
s
dx

48
Letting s = 0 in (3.7.1) results in the well-known formula

1
ς (0, a) = − a
2

and we are therefore lead to suspect that (3.7.1) is valid for a larger region than just
Re ( s ) > 1; as demonstrated by Mordell [41], by using integration by parts it is easily
shown that ς ( s, a) exists over all the s -plane except for a simple pole at s = 1 .

In particular, (3.7.1) is valid for Re ( s ) < 0 and, in view of this, Mordell [41] was able to
show how the representation (3.7.1) may be used to derive Hurwitz’s Fourier series for
ς ( s, a) ; for ease of reference, his proof is outlined below (with the correction of some
misprints).

Assuming that Re ( s) < 0, in the case where 1 ≥ a > 0 and the interval of integration is
[−a, 0] , then the left-hand side of (2.12) gives us

1 1
(a − a ) − s + (a + 0) − s = s
2 2a

so that (2.12) results in

cos 2π nx
0 ∞ 0
1 dx
2a s
= ∫ (a + x) s
+ 2 ∑ ∫
n =1 − a ( a + x )
s
dx
−a

cos 2π nx
0
a1− s ∞
=− + 2∑ ∫ dx
s −1 n =1 − a ( a + x )
s

Then, upon adding (3.7.1) to this, we obtain for Re ( s ) < 0

∞
∞
cos 2π nx
(3.7.4) ς ( s , a ) = 2∑ ∫ dx
n =1 − a ( a + x )
s

and using integration by parts this becomes

N ∞
∞
sin 2π nx ∞
sin 2π nx
ς ( s, a) = lim ∑ + 2s ∑ ∫ dx
n =1 π n( a + x ) n =1 − a 2nπ ( a + x )
N →∞ s s +1
−a

∞
∞
sin 2π nx
= 2s ∑ ∫ dx
n =1 − a 2nπ ( a + x )
s +1

49
 ∞ ∞
sin 2π n(u − a )
= 2s∑ ∫ du
n =1 0 2nπ u s +1
and we obtain

∞ ⎡ cos 2π na ∞ sin 2π nu sin 2π na cos 2π nu ⎤

∞
ς ( s, a ) = 2s ∑ ⎢ ∫ 2nπ ∫0 u s +1
(3.7.5) du − du ⎥
n =1 ⎣ 2nπ 0
u s +1 ⎦

Using contour integration we have ([50, p.107], [18, p.91]) for a > 0 and 1 > p > 0

∞
Γ( p ) cos(π p / 2)
∫u
p −1
(3.7.6) cos bu du =
0
bp

and with p = − s we see that for 0 > s > −1

∞
cos bu Γ(− s ) cos(π s / 2)
∫
0
u s +1
du =
b− s

Γ(1 − s ) cos(π s / 2)
=−
sb − s

Similarly, we have

∞
Γ( p ) sin(π p / 2)
∫u
p −1
(3.7.7) sin bu du =
0
bp

so that
∞
sin bu Γ(1 − s ) sin(π s / 2)
∫
0
u s +1
du =
sb − s

We therefore obtain Hurwitz’s formula

⎡ ⎛ π s ⎞ ∞ cos 2nπ a ⎛ π s ⎞ ∞ sin 2nπ a ⎤

(3.7.8) ς ( s, a) = 2Γ(1 − s ) ⎢sin ⎜ ⎟∑ + cos ⎜ ⎟∑ 1− s ⎥
⎝ 2 ⎠ n =1 (2π n) ⎝ 2 ⎠ n =1 (2π n) ⎦
1− s
⎣

where Re ( s ) < 0 and 0 < a ≤ 1 . In 2000, Boudjelkha [13] showed that this formula also
applies in the region Re ( s ) < 1.

With a = 1 this reduces to the functional equation for the Riemann zeta function

50
(3.7.9) ς (1 − s) = 2(2π ) − s Γ( s) cos(π s / 2)ς ( s)

cos π x
Employing the same procedure with f ( x) = , it should be possible to derive
(a + x) s
Boudjelkha’s formula [13] for the alternating Hurwitz zeta function ς a ( s, a ) which is
defined by
∞
(−1) n
ς a ( s, a ) = ∑
n = 0 ( a + n)
s

Boudjelkha’s formula is

⎡ ⎛ π s ⎞ ∞ cos(2n + 1)π a ⎛ π s ⎞ ∞ sin(2n + 1)π a ⎤

(3.7.10) ς a ( s, a) = 2Γ(1 − s )π s −1 ⎢sin ⎜ ⎟ ∑ 1− s
+ cos ⎜ ⎟∑ 1− s ⎥
⎣ ⎝ 2 ⎠ n =0 (2n + 1) ⎝ 2 ⎠ n =0 (2n + 1) ⎦

and holds under the same conditions as (3.7.8) above, namely:

( σ < 0, 0 < a ≤ 1 ; 0 < σ , a < 1)

We also note that Oberhettinger [45] used the Poisson summation formula to derive a
∞
zn
corresponding formula for the Lerch zeta function Φ ( z , s, a ) = ∑ ⋅
n =0 ( a + n)
s

It appears that Oberhettinger [45] had some reservations about the rigour employed by
Mordell [41].
□

Letting a = 1 in (3.7.5) results in [36, p.10]

∞
s ∞
1 sin 2π nu
(3.7.11) ς ( s) = ∑
π n =1 n ∫0 u s +1
du

and with a = 1/ 2 in (3.7.5) we obtain

∞
⎛ 1⎞ s ∞
(−1) n sin 2π nu
(3.7.12) ς ⎜ s, ⎟ =
⎝ 2⎠ π
∑
n =1 n ∫0 u s +1
du

Differentiating (3.7.11) gives us

∞
m1 sin 2π nu
∞
ς ( m)
(0) = (−1) ∑
π n =1 n ∫0
m

u
log m u du

51
 ∞
⎛ 1⎞ m ∞
(−1) n sin 2π nu
ς ( m ) ⎜ 0, ⎟ = (−1) m
⎝ ⎠2 π
∑
n =1 n ∫ u
log m u du
0

Formal derivations of (3.7.6) and (3.7.7) are set out below. Using the definition of the
gamma function we have

∞
Γ( s )
∫e
− zy
y s −1 dy =
0
zs

and with z = u ± ix we obtain

∫e
− uy
y s −1[cos( xy ) − i sin( xy )] dy = (u + ix) − s Γ( s )
0

∫e
− uy
y s −1[cos( xy ) + i sin( xy )] dy = (u − ix) − s Γ( s )
0

This gives us

∞
i ∫ e −uy y s −1 sin( xy ) dy = − ⎣⎡ (u + ix) − s − (u − ix) − s ⎦⎤ Γ( s )
0

∫e y s −1 cos( xy ) dy = ⎣⎡(u + ix) − s + (u − ix) − s ⎦⎤ Γ( s )

− uy

Then, noting that

(u + ix) − s − (u − ix) − s = (reiθ ) − s − (re −iθ ) − s

= r − s [e − isθ − eisθ ]

we have
2
(u + ix) − s − (u − ix) − s = sin( s tan −1 ( x / u ))
i (u + x )
2 2 s/2

and we obtain

∞
sin[ s tan −1 ( x / u )]
∫ e y sin( xy) dy = Γ(s)
− uy s −1

0
(u 2 + x 2 ) s / 2

52
and
∞
cos[ s tan −1 ( x / u )]
∫ e y cos( xy) dy = Γ(s)
− uy s −1

0
(u 2 + x 2 ) s / 2

Letting u = 0 gives us

∞
Γ( s)sin(π s / 2)
∫y
s −1
sin( xy ) dy =
0
xs
and
∞
Γ( s ) cos(π s / 2)
∫y
s −1
cos( xy ) dy =
0
xs

which correspond with (3.7.6) and (3.7.7) above.

□

We have the gamma function

∞
Γ( p ) = ∫ t p −1e − t dt
0

and with the substitution t = ux this becomes

∞
1 1
x p
= ∫
Γ( p ) 0
u p −1e − xu du

Therefore we obtain upon integration

∞ ∞∞
cos 2π nx 1
∫0 x p dx = Γ( p) ∫0 ∫0 u e cos 2π nx du dx
p −1 − xu

∞ ∞
1
= ∫
Γ( p ) 0
u p −1du ∫ e − xu cos 2π nx dx
0

and using
∞
u
∫e cos 2π nx dx =
− ux

0
u + 4π 2 n 2
2

gives us
∞ ∞
cos 2π nx 1 up
∫0 x p dx =
Γ( p ) ∫0 u 2 + 4π 2 n 2
du

We have the summation

53
 ∞ ∞ ∞
cos 2π nx 1 ∞ up
(3.7.13) ∑ ∫ xp
n =1 0
dx = ∑
Γ( p ) n =1 ∫0 u 2 + 4π 2 n 2
du

∞ ∞
1 up
= ∑
Γ( p ) ∫0 n =1 u 2 + 4π 2 n 2
du

assuming that the interchange of integration and summation is valid.

We then use the identity (3.6.9)

∞
y 1 1 1
2∑ = y − +
n =1 y + (2nπ )
2 2
e −1 y 2

to obtain

∞ ∞ ∞
cos 2π nx 1 ⎡ 1 1 1⎤
∑ ∫
n =1 0 x p
dx = ∫ ⎢ − + ⎥ u p −1du
2Γ ( p ) 0 ⎣ e − 1 u 2 ⎦
u

For −1 < σ < 0 we have [36, p.24]

∞
1 ⎡ 1 1 1⎤
ς (s) = ∫ ⎢ − + ⎥ u p −1du
Γ( p ) 0 ⎣ e − 1 u 2 ⎦
u

and we therefore have

∞ ∞
cos 2π nx 1
(3.7.14) ∑∫
n =1 0 x s
dx = ς ( s )
2

From (3.7.4) we see that

∞ ∞
cos 2π nx
ς ( s,1) = 2∑ ∫ dx
n =1 −1 (1 + x )
s

∞ ∞
cos 2π n(u − 1)
= 2∑ ∫ du
n =1 0 us

which corresponds with (3.7.14).

□

Integration by parts results in

54
 N
∞ ∞ ∞
cos 2π nx ∞
sin 2π nx ∞
sin 2π nx
∑ ∫
n =1 0 x s
dx = lim ∑
N →∞
n =1 2π n x
s
+ s∑ ∫
n =1 0 2π n x
s +1
dx
0

∞ ∞
sin 2π nx
= s∑ ∫ dx
n =1 0 2π n x
s +1

1
and Titchmarsh [50, p.15] tells us that this is equal to ς ( s )
2
□

Alternatively, we may consider the integral

∞ ∞
up 1 up
I =∫
t 2 ∫0 (u / t ) 2 + 1
du = du
0
u2 + t 2

Letting v = (u / t ) 2 we get

∞
t p −1 v ( p −1) / 2
2 ∫0 1 + v
I= dv

1 π
=
2t sin[π ( p + 1) / 2]
1− p

1 π
=
2t cos(π p / 2)
1− p

where we have used the well known integral

∞
v s −1 π
∫0 1 + v dv = sin π s .

Substituting this in (3.7.13) results in (cf [50, p.24])

∞ ∞
cos 2π nx 1 ∞ 1 π
∑ ∫
n =1 0 x p
dx = ∑
Γ( p ) n =1 2(2π n) cos(π p / 2)
1− p

π
= ς (1 − p)
2Γ( p )(2π ) 1− p
cos(π p / 2)

55
 1
= ς ( p)
2

where we have employed the functional equation for the Riemann zeta function. This
gives us (3.7.14).

We have from (3.7.6)

∞
cos u π
∫
0
u s
du =
2Γ( s ) cos(π s / 2)

and the substitution u = 2π nx gives us

∞
cos 2π nx π
∫
0
x s
dx =
2(2π ) n Γ( s ) cos(π s / 2)
1− s 1− s

and we have the summation

∞ ∞
cos 2π nx πς (1 − s)
∑∫
n =1 0 x s
dx =
2(2π ) Γ( s ) cos(π s / 2)
1− s

Using the functional equation for the Riemann zeta function

ς (1 − s) = 2(2π ) − s Γ( s) cos(π s / 2)ς ( s)

we obtain (3.7.14) again

∞ ∞
cos 2π nx 1
∑∫
n =1 0 x s
dx = ς ( s )
2

We see that

∞ ∞
cos 2π nx cos 2π n(u − 1)
∫0 (1 + x)s dx = ∫1 us
du

∞
cos 2π nu
=∫ du
1
us

56
and we therefore have

∞ ∞ ∞
cos 2π nx cos 2π nx cos 2π nx
∞ ∞ 1

∑ ∫
n =1 0 (1 + x )
s
dx = ∑ ∫
n =1 0 x s
dx − ∑ ∫
n =1 0 xs
dx

cos 2π nx
∞ 1
1
= ς ( s) − ∑ ∫ dx
2 n =1 0 xs

Letting a = 1 in (3.7.1) gives us

∞
1 1 ∞
cos 2π nx
ς ( s) = + + 2∑ ∫ dx
2 s −1 n =1 0 (1 + x )
s

which appears to suggest that

cos 2π nx
∞ 1
1 1
(3.7.15) 2∑ ∫ dx = + + ς ( s)
n =1 0 x s
2 s −1
which does not agree with equation (8.16) in Berndt’s book [10, Part II, p.317]

∞ 1
cos 2π nx
ς (1/ 2) = lim 2∑ ∫ dx
ε →0 + x
n =1 ε

Differentiating (3.7.15) gives us

∞ 1
log x cos 2π nx 1
−2 ∑ ∫ dx = − + ς ′( s )
n =1 0 x s
( s − 1) 2

and with s = 0 we have

∞ 1
ς ′(0) − 1 = −2∑ ∫ log x ⋅ cos 2π kx dx
k =1 0

It is easily seen that

1
sin 2π nx sin 2π nx
1 1

∫0 log x ⋅ cos 2π nx dx = log x ⋅ 2π n 0 − ∫0 2π nx dx

sin 2π kx
1
= −∫ dx
0
2π kx

57
 Si (2π n)
1

∫ log x.cos 2π nx dx = −
0
2π n

This therefore suggests that

∞
Si (2π n)
ς ′(0) − 1 = ∑
n =1 πn

but this does not agree with (3.2.3)

∞
si (2nπ ) π
∑
n =1 n
= log(2π ) − π
2
□

We write (3.7.1) as

∞
1 1 1 a1− s − 1 ∞
cos 2π nx
(3.7.16) ς ( s, a ) − =
s −1 2 a s
+
s −1
+ 2 ∑ ∫
n =1 0 ( a + x )
s
dx

It may be seen from (3.5.1) that

dp ⎡ 1 ⎤
γ p (a) = (−1) lim p
p
⎢ς ( s, a ) − s − 1 ⎥
s →1 ds
⎣ ⎦

and differentiating (3.7.16) p times gives us

∞
dp ⎡ 1 ⎤ a − s (−1) p log p a ∞
log p (a + x) cos 2π nx
⎢⎣ς ( s, a) − s − 1 ⎥⎦ = + f ( s ) + 2(−1) ∑ ∫
( p) p
dx
ds p 2 n =1 0 (a + x) s

where we have denoted f ( s ) as

a1− s − 1
f ( s) =
s −1

We can represent f ( s ) by the integral

a1− s − 1
a
f ( s) = = − ∫ x − s dx
s −1 1
so that

58
 a
f ( p ) ( s ) = −(−1) p ∫ x − s log p x dx
1
and thus
a p +1
log p x p +1 log a
f ( p)
(1) = −(−1) ∫ p
dx = (−1)
1
x p +1

Therefore, upon taking the limit s → 1 , we obtain for p ≥ 1

∞
1 log p a log p +1 a ∞
log p (a + x)
(3.7.17) γ p (a) = − + 2∑ ∫ cos 2π nx dx
2 a p +1 n =1 0 a+x

as previously shown in an equivalent form by Zhang and Williams [54, Eq(6.1)] in 1994.

This may be written as

∞
1 log p a log p +1 a ∞
log p u
γ p (a) = − + 2∑ ∫ cos 2π n(u − a ) du
2 a p +1 n =1 a u

and with a = 1 we obtain (3.5.5).

Letting s → 1 in (3.7.16) gives us

∞
1 ∞
cos 2π nx
γ 0 (a) = − log a + 2∑ ∫ dx
2a n =1 0 a+x
so that
∞
1
γ 0 (a) = − log a − 2∑ [ cos(2nπ a) si (2nπ a) + sin(2nπ a) Ci (2nπ a) ]
2a n =1

and, since γ 0 (a) = −ψ (a ) , this is equivalent to (3.7) as reported by Nörlund [44].

Differentiation of (3.7.16) with respect to s gives us

∞
1 log a ( s − 1) log a + 1 ∞
cos 2π nx log(a + x)
ς ′( s, a) = −
2 a s
− 2 s −1
( s − 1) a
− 2 ∑ ∫
n =1 0 (a + x) s
dx

so that
∞ ∞
⎛ 1⎞
(3.7.18) ς ′(0, a) = ⎜ a − ⎟ log a − a − 2∑ ∫ cos 2π nx log(a + x) dx
2⎝ ⎠ n =1 0

59
Integration by parts gives us

sin 2π nx 1 sin 2π nx
N N N

∫0 cos 2π nx log(a + x) dx = 2π n log(a + x) 0 − 2π n ∫0 a + x dx

sin 2π nx
N
1
2π n ∫
=− dx
0
a+x

and we therefore have

∞ ∞
1 sin 2π nx
∫ cos 2π nx log(a + x) dx = −
0
2π n ∫
0
a+x
dx

Using Hurwitz’s formula

1
ς ′(0, a) = log Γ(a) − log(2π )
2
we obtain
∞
1 ⎛ 1⎞ 1 ∞ sin(2nπ x)
(3.7.19) log Γ(a) = log(2π ) + ⎜ a − ⎟ log a − a + ∑ ∫ dx
2 ⎝ 2⎠ π n =1 0 n( x + a )

which is one of the exercises posed by Whittaker & Watson [51, p.261]. This result was
attributed by Stieltjes to Bourguet. Equation (3.7.19) may also be derived using the Euler-
Maclaurin summation formula (see for example Knopp’s book [38, p.530]).

With regard to (3.7.18) we have

∫ cos 2π nx log(a + x) dx
0
N
1
= ⎡sin 2nπ a Ci ⎡⎣ 2nπ ( a + x ) ⎤⎦ − cos 2nπ a Si ⎡⎣ 2nπ ( a + x ) ⎤⎦ + sin 2nπ x log(a + x) ⎤
2nπ ⎣ ⎦
0

N
1
= ⎡sin 2nπ a Ci ⎡⎣ 2nπ ( a + x ) ⎤⎦ − cos 2nπ a Si ⎡⎣ 2nπ ( a + x ) ⎤⎦ ⎤
2nπ ⎣ ⎦
0

∞
sin t π
Since lim Si ( x) = ∫ dt = and lim Ci ( x) = 0 we determine that
x →∞
0
t 2 x →∞

60
 ∞
1 ⎡ ⎧π ⎫⎤
∫ cos 2π nx log(a + x) dx = 2nπ ⎢⎣ − sin 2nπ a Ci(2nπ a) − cos 2nπ a ⎨⎩ 2 − Si(2nπ a) ⎬⎭⎥⎦
0

1
= [ cos(2nπ a)si(2nπ a) − sin(2nπ a) Ci(2nπ a)]
2nπ
π
since si ( x) = Si ( x) − .
2

Therefore we have

∞
⎛ 1⎞ 1
ς ′(0, a) = ⎜ a − ⎟ log a − a − ∑ [ cos 2nπ asi (2nπ a) − sin 2nπ a Ci(2nπ a)]
⎝ 2 ⎠ nπ n =1

which implies that

log Γ(a ) =

1 ⎛ 1⎞ 1 ∞ 1
log(2π ) + ⎜ a − ⎟ log a − a + ∑ [sin(2nπ a )Ci (2nπ a ) − cos(2nπ a ) si (2nπ a)]
2 ⎝ 2⎠ π n =1 n

which we saw earlier in (3.10).

We have the second derivative of (3.7.16)

∞
1 log 2 a 2( s − 1) log a + ( s − 1)2 log 2 a + 2 ∞
cos 2π nx log 2 (a + x)
ς ′′( s, a) =
2 as
+
( s − 1)3 a s −1
+ 2 ∑ ∫
n =1 0 ( a + x) s
dx

so that
∞ ∞
⎛ 1⎞ 2
ς ′′(0, a) = − ⎜ a − ⎟ log a + 2a log a − 2 + 2∑ ∫ cos 2π nx log 2 (a + x) dx
⎝ 2⎠ n =1 0

Integration by parts gives us

sin 2π nx 1 sin 2π nx log(a + x)

N N N

∫0 cos 2π nx log (a + x) dx = 2π n log (a + x) 0 − π n ∫0

2 2
dx
a+x

1 sin 2π nx log(a + x)
N

π n ∫0
=− dx
a+x

61
 N
1 sin 2π nx log(a + x) cos 2π nx log(a + x) cos 2π nx [1 − log(a + x)]
N N
1
− ∫
πn 0 a+x
dx = +
2(π n) (a + x) 0 2(π n) 2
2 ∫
0
(a + x)2
dx

log(a + N ) cos 2π nx [1 − log(a + x)]

N
log a 1
= − +
2(π n) (a + N ) 2(π n) a 2(π n) 2
2 2 ∫
0
(a + x) 2
dx

log(a + N )
and, since lim = 0 , we have
N →∞ ( a + N )

∞ ∞
⎛ 1⎞ ς (2) log a 1 cos 2π nx [1 − log(a + x)]
ς ′′(0, a) = − ⎜ a − ⎟ log 2 a + 2a log a − 2 −
⎝ 2⎠ π 2a
+ 2
π
∑∫
n =1 0 n 2 ( a + x) 2
dx

We note that Mathematica evaluates ∫ cos 2π nx log 2 (1 + x) dx in terms of the generalised

hypergeometric function 3 F3 and the incomplete gamma function.

4. Some integrals involving cot x

In this section we shall have need of Bonnet’s second mean-value theorem which states
[17, p.110]:

(i) Let φ ( x) be bounded, monotonic decreasing and never negative in [a, b] ; and let
ψ ( x) be bounded and integrable in [a, b] . Then we have

b ξ

∫ φ ( x)ψ ( x)dx = φ (a)∫ψ ( x)dx

a a

where ξ is some definite value of x in a ≤ ξ ≤ b .

(ii) Alternatively, subject to the same conditions as above but with φ ( x) monotonic
increasing, we then have
b b

∫ φ ( x)ψ ( x)dx = φ (b)∫ξ ψ ( x)dx

a

We now use Bonnet’s second mean-value theorem to prove the following proposition.

Proposition

62
We assume that f ( x) is continuous on [0, b] and that f (0) = 0 . Let us also assume that
f ( x) is non-negative and monotonic increasing on [0, b] .

Then we have

b ∞ b
(4.1) ∫ f ( x) cot( x / 2) dx = 2∑ ∫ f ( x) sin 2nx dx
0 n =1 0

Proof

We recall (1.3)

1 N
cos( N + 1 / 2) x
cot( x / 2) = ∑ sin nx +
2 n=0 2 sin( x / 2)

N
= ∑ sin nx + cot( x / 2) cos Nx − sin Nx
n =1

We now multiply this by f ( x) and integrate over the interval [π / N , b] where b ≤ π .

Provided the relevant integrals exist, we easily see that

b b b b
1 N
f ( x) cot( x / 2) dx = ∑ ∫ f ( x) sin nx dx + ∫ f ( x) cot( x / 2) cos Nx dx − ∫ f ( x) sin Nx dx
2 π∫ n =1 π π π
N N N N

N
= ∑ J n + I1 ( N ) + I 2 ( N )
n =1

and, first of all, we employ the approach adopted by Zygmund [55, p.59] for the integral
b
I1 ( N ) =
π
∫ f ( x) cot( x / 2) cos Nx dx .
N

Since we assumed that f ( x) is continuous and that f (0) = 0 , given ε > 0, we may
choose η > 0 such that f (η ) < ε and we write

b η b

∫
π
f ( x) cot( x / 2) cos Nx dx = ∫
π
f ( x) cot( x / 2) cos Nx dx + ∫ f ( x) cot( x / 2) cos Nx dx
η
N N

and we have the inequality

63
 b η b

∫
π
f ( x) cot( x / 2) cos Nx dx ≤
π
∫ f ( x) cot( x / 2) cos Nx dx + ∫η f ( x) cot( x / 2) cos Nx dx
N N

We note that cot( x / 2) is monotonic decreasing on [π / N , π ] and never negative in

[π / N , π ] (which is the reason why we specified that b ≤ π ) and hence the second mean-
value theorem (i) gives us for the first part

η η′
(4.2)
π
∫ cot( x / 2) f ( x) cos Nx dx = cot(π / 2 N ) ∫ f ( x) cos Nx dx
π
N N

Since we assumed that f ( x) is non-negative and monotonic increasing on [0, b] , we may

apply the second limb of Bonnet’s second mean-value theorem (ii) to the integral on the
right-hand side of (4.2)

η′ η′

π
∫ f ( x) cos Nx dx = f (η ′) ∫ cos Nx dx
η ′′
N

and we obtain

η η′

π
∫ f ( x) cot( x / 2) cos Nx dx = cot(π / 2 N ) f (η ′) ∫ cos Nx dx
η ′′
N

[sin Nη ′ − sin Nη ′′]

= cot(π / 2 N ) f (η ′)
N

We recall the elementary inequality sin x < x < tan x for 0 < x < π / 2 which yields

1
0 < cot x <
x

and we then have

η
2N 2
π
∫ f ( x) cot( x / 2) cos Nx dx <
π
ε
N
N

Since η > 0, the strong version of the Riemann-Lebesgue lemma [5, p.313] tells us that

64
 b
lim ∫ f ( x) cot( x / 2) cos Nx dx = 0
N →∞
η

and similarly we also have lim I 2 ( N ) = 0 .Therefore as N → ∞ we have the limit

N →∞

b ∞ b

∫ f ( x) cot( x / 2) dx = 2∑ ∫ f ( x) sin 2nx dx

0 n =1 0

where, as stated above, we require that f (0) = 0 . We may also show that

b ∞ b
(4.3) ∫ f ( x) cot(α x / 2) dx = 2∑ ∫ f ( x) sin 2nα x dx
0 n =1 0

It is easily seen that this may be generalised to

b ∞ b
(4.4) ∫ f ( x) cot(α x / 2) dx = 2∑ ∫ f ( x) sin 2nα x dx
a n =1 a

We therefore have a version of (1.4a) which now encompasses a larger class of eligible
functions f ( x) .
□

Various applications of (1.4a) were considered in [] and these included the evaluation of
the following integrals:

π
6
1 π2 3 ⎡ 4 1 ⎧ ⎛ 1⎞ ⎛ 1 ⎞ ⎫⎤
∫ x 2 cot x dx = − ς (3) + log [ 2 sin ( π / 6 )] + π ⎢ − 9 ς (2) + 36 ⎨ς ⎜⎝ 2, 6 ⎟⎠ + ς ⎜⎝ 2, 3 ⎟⎠ ⎬⎥
0
3 36 6 ⎣ ⎩ ⎭⎦

π
π 1 ⎡ ⎤
( )
8
1 ⎛ 1⎞
∫ x cot x dx = 16 log ⎡⎣2 −
0
2 ⎤⎦ + ⎡⎣1 − 2 ⎤⎦ G + ⎢ 2ς ⎜ 2, ⎟ − 2
8 64 ⎣ ⎝ 8⎠
2 +1 π 2 ⎥
⎦

1
1
20∫ ( x 2 log x ) cot(π x / 2)dx =

3 2 ∞
Ci (nπ ) 2 2
3 [ a
− ς (3) + ς (3) ] − ∑ + 3 ς (3) [γ + logπ ] − 3 ς ′(3)
π π n =1 n
3 3
π π

65
 Ci (2nπ )
1 ∞ 1 ∞
1
∫ log Γ( x + 1) cot π x dx = 2∑ ∫ log Γ( x + 1) sin 2nπ x dx =
0 n =1 0 π
∑ n =1 n

Since ψ ( x) −ψ (1 − x) = −π cot π x we see that

1 ∞ 1
(4.5) ∫ f ( x)[ψ (1 − x) −ψ ( x)] dx = 2π ∑ ∫ f ( x) sin 2π nx dx
0 n =1 0

Part of the following is based on an observation made by Glasser [29] in 1966. Let us
consider the integral

1
I = ∫ f ( x)ψ ( x)dx
0

where f ( x) = − f (1 − x) and f ( x) is selected so that the integral converges. We then

have
1 1
I = ∫ f (1 − t )ψ (1 − t )dt = − ∫ f (t )ψ (1 − t )dt
0 0

and hence we see that

1
2 I = ∫ f ( x)[ψ ( x) −ψ (1 − x)] dx
0

Therefore, since ψ ( x) −ψ (1 − x) = −π cot π x , we have

1
1
π 1 2
(4.6) ∫0
f ( x)ψ ( x)dx = −
2 ∫0
f ( x) cot π x dx = −π ∫
0
f ( x) cot π x dx

Integration by parts formally gives us

1 1
I = ∫ f ( x)ψ ( x)dx = f ( x) log Γ( x) 0 − ∫ f ′( x) log Γ( x) dx
1

0 0

and, for suitably behaved functions, we have

66
 1 1

∫ 0
f ( x)ψ ( x)dx = − ∫ f ′( x) log Γ( x) dx
0

Hence we see that

1
π 1 2 1

2 ∫0 ∫
(4.7) f ( x) cot π x dx = π f ( x) cot π x dx = ∫ f ′( x) log Γ( x) dx
0 0

where f ( x) = − f (1 − x) and f ( x) is selected so that the integral converges.

An example of this is

1 1

∫ B2n+1 ( x)ψ ( x) dx = B2n+1 ( x) log Γ( x) 0 − (2n + 1)∫ B2n ( x) log Γ( x) dx

1

0 0

1
= −(2n + 1) ∫ B2 n ( x) log Γ( x) dx
0

Glasser’s formula (4.6) gives us

1
π 1
(4.8) ∫ B2n+1 ( x)ψ ( x) dx = −
0
2 ∫0
B2 n +1 ( x) cot π x dx

and hence we have

1
π 1
(4.9) ∫B
0
2n ( x) log Γ( x) dx =
2(2n + 1) ∫0
B2 n +1 ( x) cot π x dx

The following integral appears in Abramowitz and Stegun [1, p.807]

(−1) n +1 2(2n + 1)!ς (2n + 1)

1
(4.10) ∫0 B2n+1 ( x) cot π x dx = (2π ) 2 n +1

and there are many derivations of this; for example, see the recent one by Dwilewicz and
Mináč [27]. A similar identity was also derived by Espinosa and Moll [28] in the form

(2n)!ς (2n + 1)
1
(4.11) ∫B
0
2n ( x) log sin π x dx = (−1) n
(2π ) 2 n

67
and it is easily shown that equation (4.11) above is equivalent to (4.10) following a
simple integration by parts.

Hence we obtain

(−1) n +1 (2n)!ς (2n + 1)

1
(4.12) ∫ B2n ( x) log Γ( x) dx =
0
2(2π ) 2 n +1

in agreement with [28].

5. Open access to our own work

This paper contains references to various other papers and, rather surprisingly, most of
them are currently freely available on the internet. Surely now is the time that all of our
work should be freely accessible by all. The mathematics community should lead the way
on this by publishing everything on arXiv, or in an equivalent open access repository. We
think it, we write it, so why hide it? You know it makes sense.

REFERENCES

[1] M. Abramowitz and I.A. Stegun (Eds.), Handbook of Mathematical Functions

with Formulas, Graphs and Mathematical Tables. Dover, New York, 1970.
http://www.math.sfu.ca/~cbm/aands/

[2] V.S. Adamchik and H.M. Srivastava, Some Series of the Zeta and Related
Functions. Analysis 18, 131-144, 1998.
http://www-2.cs.cmu.edu/~adamchik/articles/sums.htm

[3] V.S. Adamchik, Symbolic and numeric computations of the Barnes function.
Computer Physics Comms., 157 (2004) 181-190.
http://www.cs.cmu.edu/~adamchik/articles/PhysCom.pdf

[4] V.S. Adamchik, On the Hurwitz function for rational arguments.

Applied Mathematics and Computation, 187 (2007) 3-12.
http://www.cs.cmu.edu/~adamchik/articles/AMC.pdf

[5] T.M. Apostol, Mathematical Analysis, Second Ed., Addison-Wesley Publishing

Company, Menlo Park (California), London and Don Mills (Ontario), 1974.

[6] H. Azuma, Application of Abel-Plana formula for collapse and revival of Rabi
oscillations in Jaynes-Cummings model.
Int. J. Mod. Phys. C 21, 1021-1049 (2010) arXiv:0901.4357 [pdf, ps, other]

[7] G. Bachman, L. Narici and E. Beckenstein, Fourier and Wavelet Analysis.

Springer-Verlag, New York, 2000.

68
[8] N. Batir, Inequalities for the double gamma function. 2012.
http://ajmaa.org/RGMIA/papers/v11n4/JMAA_BARNES.pdf

[9] B.C. Berndt, On the Hurwitz zeta function.

Rocky Mtn. J. Math. 2, 151-157, (1972).

[10] B.C. Berndt, Ramanujan’s Notebooks. Part II, Springer-Verlag, 1989.

[11] B.C. Berndt, Ramanujan’s Notebooks. Part V, Springer-Verlag, 1998.

[12] B.C. Berndt and A. Dixit, A transformation formula involving the Gamma and
Riemann zeta functions in Ramanujan's Lost Notebook.
arXiv:0904.1053 [pdf, ps, other], 2009.

[13] M.T. Boudjelkha, A proof that extends Hurwitz formula into the critical strip.
Applied Mathematics Letters, 14 (2001) 309-403.

[14] W.E. Briggs, Some constants associated with the Riemann zeta-function.
(1955-1956), Michigan Math. J. 3, 117-121.

[15] T.J.I’a Bromwich, Introduction to the theory of infinite series. Third edition.
AMS Chelsea Publishing, 1991.

[16] P. L. Butzer, P. J. S. G. Ferreira, G. Schmeisser and R. L. Stens,

The Summation Formulae of Euler–Maclaurin, Abel–Plana, Poisson, and their
interconnections with the Approximate Sampling Formula of Signal Analysis.
Results. Math., 59, No.3-4, 359-400, 2011, DOI 10.1007/s00025-010-0083-8
http://www.ieeta.pt/~pjf/PDF/Ferreira2011b.pdf

[17] H.S. Carslaw, Introduction to the theory of Fourier Series and Integrals.
Third Ed. Dover Publications Inc, 1930.

[18] H. Cohen, Number Theory. Volume II: Analytic and modern tools.
Springer Science, 2007.

[19] D.F. Connon, Some series and integrals involving the Riemann zeta function,
binomial coefficients and the harmonic numbers. Volume I, 2007.
arXiv:0710.4022 [pdf]

[20] D.F. Connon, Some series and integrals involving the Riemann zeta function,
binomial coefficients and the harmonic numbers. Volume V, 2007.
arXiv:0710.4047 [pdf]

[21] D.F. Connon, Some series and integrals involving the Riemann zeta function,
binomial coefficients and the harmonic numbers. Volume VI, 2007.
arXiv:0710.4032 [pdf]

69
[22] D.F. Connon, Some applications of the Stieltjes constants.
arXiv:0901.2083 [pdf], 2009.

[23] D.F. Connon, Some trigonometric integrals involving log Γ( x) and the digamma
function. arXiv:1005.3469 [pdf], 2010.

[24] D.F. Connon, Some integrals involving the Stieltjes constants: Part II.
arXiv:1104.1911 [pdf], 2011.

[25] D.F. Connon, Some applications of the sine and cosine integrals. 2012.

[26] P.G.L. Dirichlet, Sur la convergence des séries trigonométriques qui servent à
représenter une fonction arbitraire entre des limites données.
Journal für die reine und angewandte Mathematik, 4, (1829), 157-169.
http://www.digizeitschriften.de/zeitschriften/open-access/

[27] R. Dwilewicz and J. Mináč, An introduction to relations between the

values of ς ( s ) in terms of holomorphic functions of two variables.
Proceedings of the Hayama Symposium on Several Complex Variables,
Japan, Dec. 2000. Pages 28-38 (2001).
(see also http://www.mat.uab.es/matmat/PDFv2009/v2009n06.pdf)

[28] O. Espinosa and V.H. Moll, On some integrals involving the Hurwitz zeta
function: Part I. The Ramanujan Journal, 6,150-188, 2002.
http://arxiv.org/abs/math.CA/0012078

[29] M.L. Glasser, Evaluation of some integrals involving the ψ - function.

Math. of Comp., Vol.20, No.94, 332-333, 1966.

[30] I.S. Gradshteyn and I.M. Ryzhik, Tables of Integrals, Series and Products.
Sixth Ed., Academic Press, 2000.
Errata for Sixth Edition http://www.mathtable.com/errata/gr6_errata.pdf

[31] A.P. Guinand, On Poisson’s summation formula.

Annals of Mathematics, Vol. 42, No. 3 (Jul., 1941), pp. 591-603.

[32] A.P. Guinand, Some formulae for the Riemann zeta-function.

J. London Math. Soc. 22 (1947), 14–18.

[33] A.P. Guinand, Some finite identities connected with Poisson’s summation formula.
Proc. Edinburgh Math. Soc. (2) 12 (1960), 17–25.

[34] G.H. Hardy, Divergent Series. Chelsea Publishing Company, New York, 1991.

[35] C. Hermite, Correspondance d'Hermite et de Stieltjes. Gauthier-Villars, Paris,

70
 1905. http://ebooks.library.cornell.edu/m/math/browse/author/a.php

[36] A. Ivić, The Riemann Zeta- Function: Theory and Applications.

Dover Publications Inc, 2003.

[37] C. Jordan, Sur la série de Fourier.

Comptes Rendus Acad. Sciences, 92, 228-230, 1881.
http://gallica.bnf.fr/ark:/12148/bpt6k7351t/f227.image.r=camille+jordan.langFR

[38] K. Knopp, Theory and Application of Infinite Series.

Second English Ed. Dover Publications Inc, New York, 1990.

∞
[39] E.E. Kummer, Beitrag zur Theorie der Function Γ( x) = ∫ e − v v x −1dv .
0

J. Reine Angew. Math., 35, 1-4, 1847.

http://www.digizeitschriften.de/

[40] M. Merkle and M. M. R. Merkle, Krull's theory for the double gamma function.
Appl. Math. Comput., 218 (2011), 935-943, doi:10.1016/j.amc.2011.01.090.
http://www.milanmerkle.com/documents/radovi/KRULLcs.pdf

[41] L.J. Mordell, Some applications of Fourier series in the analytic theory
of numbers, Proc. Cambridge Phil. Soc., 24 (1928), 585-596.

[42] L.J. Mordell, Poisson's Summation Formula and the Riemann Zeta Function.
J. London Math. Soc. (1929) s1-4 (4): 285-291.

[43] N. Nielsen, Die Gammafunktion. Chelsea Publishing Company, Bronx and

New York, 1965.

[44] N.E. Nörlund, Vorlesungen über Differenzenrechnung.Chelsea, 1954.

http://math-doc.ujf-grenoble.fr/cgi-bin/linum?aun=001355

[45] F. Oberhettinger, Note on the Lerch zeta function.

Pacific J. Math., Volume 6, Number 1 (1956), 117-120.
http://projecteuclid.org/DPubS?service=UI&version=1.0&verb=Display&handle=euclid.pjm/1103044247

[46] A.A. Saharian, The generalized Abel-Plana formula with applications to Bessel
functions and Casimir effect. arXiv:0708.1187 [pdf, ps, other], 2008.

4
[47] J. Sondow, Double Integrals for Euler's Constant and log and an Analog
π
of Hadjicostas's Formula. Amer. Math. Monthly, 112, 61-65, 2005.
arXiv:math/0211148 [pdf]

[48] M.R. Spiegel, Schaum's Outline Series of Theory and Problems of Advanced

71
 Calculus. McGraw Hill Book Company, 1963.

[49] H.M. Srivastava and J. Choi, Series Associated with the Zeta and Related
Functions. Kluwer Academic Publishers, Dordrecht, the Netherlands, 2001.

[50] E.C. Titchmarsh, The Theory of Functions.

2nd Ed., Oxford University Press, 1932.

[51] E.T. Whittaker and G.N. Watson, A Course of Modern Analysis: An

Introduction to the General Theory of Infinite Processes and of Analytic
Functions; With an Account of the Principal Transcendental Functions. Fourth
Ed., Cambridge University Press, Cambridge, London and New York, 1963.

[52] Z.X. Wang and D.R. Guo, Special Functions.

World Scientific Publishing Co Pte Ltd, Singapore, 1989.

[53] J.R. Wilton, A Proof of Poisson's Summation Formula.

J. London Math. Soc. (1930) s1-5(4): 276-279.

[54] N.-Y. Zhang and K.S. Williams, Some results on the generalized Stieltjes
constants. Analysis 14, 147-162 (1994).
http://people.math.carleton.ca/~williams/papers/pdf/187.pdf

[55] A. Zygmund, Trigonometric Sums. Cambridge Mathematical Library, 2002.

Wessex House,
Devizes Road,
Upavon,
Pewsey,
Wiltshire SN9 6DL

72