This article was downloaded by:[Tričković, Slobodan B.

]
On: 2 June 2008
Access Details: [subscription number 793618343]
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954
Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Integral Transforms and Special

Functions
Publication details, including instructions for authors and subscription information:
http://www.informaworld.com/smpp/title~content=t713643686
On the summation of trigonometric series
Slobodan B. Tričković a; Mirjana V. Vidanović b; Miomir S. Stanković b
a
Department of Mathematics, Faculty of Civil Engineering, University of Ni , Ni ,
Serbia
b
Department of Mathematics, Faculty of Environmental Engineering, University of
Ni , Ni , Serbia

Online Publication Date: 01 January 2008

To cite this Article: Tričković, Slobodan B., Vidanović, Mirjana V. and Stanković,
Miomir S. (2008) 'On the summation of trigonometric series', Integral Transforms
and Special Functions, 19:6, 441 — 452
To link to this article: DOI: 10.1080/10652460801936689
URL: http://dx.doi.org/10.1080/10652460801936689

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf

This article maybe used for research, teaching and private study purposes. Any substantial or systematic reproduction,
re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expressly
forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents will be
complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should be
independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings,
demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or
arising out of the use of this material.
 Integral Transforms and Special Functions
Vol. 19, No. 6, June 2008, 441–452
Downloaded By: [Tričković, Slobodan B.] At: 08:27 2 June 2008

On the summation of trigonometric series

Slobodan B. Tričkovića *, Mirjana V. Vidanovićb and Miomir S. Stankovićb
a Department of Mathematics, Faculty of Civil Engineering, University of Niš, Niš, Serbia; b Department of
Mathematics, Faculty of Environmental Engineering, University of Niš, Niš, Serbia

(Received 28 September 2007; final version)

We deal with the series

∞
 (s)n−1 f ((an − b)x)
, α ∈ R+ ,
(an − b)α
n=1
and express it as a power series in terms of Riemann’s ζ or Catalan’s β function or Dirichlet functions η
and λ. Also, closed form cases as well as those when it is necessary to take limit have been thoroughly
analyzed. Some applications such as convergence acceleration are considered too.

Keywords: Riemann’s ζ and Catalan’s β function; Dirichlet η and λ functions

2000 Mathematics Subject Classification: Primary: 33C10; Secondary: 11M06, 65B10

1. Introduction and preliminaries

There have been many particular cases (see, for example, [2] and [7]) of the trigonometric series
of the type
∞
(s)n−1 f ((an − b)x)
, α ∈ R+ , (1)
n=1
(an − b) α

where s is 1 or −1, f = sin or f = cos and a = { 21 }b = { 01 }. We first derive a general formula

for finding the sum of Equation (1), meaning that we consider α to be a positive real parameter
with the exclusion of positive integers. Afterwards we regard α as a positive integer, whereupon
there appears a necessity to distinguish between the cases when infinite series (1) reduces to a
finite sum, and those when limiting value must be taken. Apart from all known particular cases,
the general formula yields new ones. We start with the following

LEMMA 1 Series (1) is uniformly convergent for α > 0. −z Convergence regions are given in
where ζ is Riemann’s zeta function ζ (z) =  ∞
Table 1, k=1 k , η and λ are Dirichlet functions
η(z) = ∞ k=1 (−1) k−1 −z
k = (1 −
∞2 1−z
)ζ (z), λ(z) = ∞
k=0 (2k + 1)−z = (1 − 2−z )ζ (z), and β is
−z
Catalan’s beta function β(z) = k=0 (−1) (2k + 1) .
k

*Corresponding author. Email: [email protected]

ISSN 1065-2469 print/ISSN 1476-8291 online

© 2008 Taylor & Francis
DOI: 10.1080/10652460801936689
http://www.informaworld.com
 442 S.B. Tričković et al.

Table 1. Parameters and convergence region

Downloaded By: [Tričković, Slobodan B.] At: 08:27 2 June 2008

a b s c F f δ r α∈ Convergence region

1 0 1 1 ζ sin 1 2m − 1 R+ \ 2N 0 < x < 2π

cos 0 2m R+ \ 2N − 1
−1 0 η sin 1 2m − 1 R+ \ 2N −π < x < π
cos 0 2m R+ \ 2N − 1
2 1 1 1/2 λ sin 1 2m − 1 R+ \ 2N 0<x<π
cos 0 2m R+ \ 2N − 1
−1 0 β sin 1 2m R+ \ 2N − 1 −π/2 < x < π/2
cos 0 2m − 1 R+ \ 2N

Remark 1 We note that ζ, η, λ are analytic in the whole complex

 planek except for z = 1, where
they have a pole, whereas Catalan’s beta function β(z) = ∞ k=0 (−1) (2k + 1) −z
satisfies the
functional equation β(z) = (π/2)z−1 (1 − z) cos(π z/2)β(1 − z) extending the beta function to
the left side of the complex plane Re z < 1.


Proof We shall make use of Dirichlet’s test saying ∞ that (see [3]) the series ∞n=0 an (x)bn (x) is
uniformly convergent in D, if the partial sums of n=0 an (x) are uniformly bounded in D, and
the sequence bn (x), being monotonic for every fixed x, uniformly converges to 0.
Let us suppose first that a = 1, b = 0. If s = 1, then there holds
 n 
  1 1
 f (kx) ≤ ≤
 sin(x/2) sin ε
k=1

for 0 < x < 2π, because sin(x/2) ≥ sin ε > 0 for each ε > 0 satisfying ε ≤ x/2 ≤ π − ε,
meaning that the partial sums are uniformly bounded with respect to x ∈ (0, 2π ). Similarly,
if s = −1, we would have
 
 n  1 1
 (−1) f (kx) ≤
k−1
≤
 cos(x/2) sin ε
k=1

for −π < x < π, because cos(x/2) ≥ sin ε > 0 for each ε > 0 satisfying −(π/2) + ε ≤ (x/2) ≤
(π/2) − ε. Here the partial sums are uniformly bounded with respect to x ∈ (−π, π ).
We now suppose a = 2, b = 1. If s = 1, then
 n 
  1 1
 f ((2k − 1)x) ≤ ≤
 sin x sin ε
k=1

for 0 < x < π, because sin x ≥ sin ε > 0 for each ε > 0 satisfying ε ≤ x ≤ π − ε. The partial
sums are uniformly bounded with respect to x ∈ (0, π ). Finally if s = −1, then
 n 
 
 (−1) k−1
f ((2k − 1)x) ≤ 1 ≤ 1
  cos x sin ε
k=1

for −(π/2) < x < (π/2), because cos x ≥ sin ε > 0 for each ε > 0 satisfying −(π/2) + ε ≤
x ≤ (π/2) − ε, and the partial sums are uniformly bounded with respect to x ∈ (−(π/2), (π/2)).
Since in each of the particular cases the sequence 1/(an − b)α tends to 0 for α > 0, by virtue of
Dirichlet’s test all the series in question uniformly converge, whereby the proof is complete. 
 Integral Transforms and Special Functions 443

2. The general formula

Downloaded By: [Tričković, Slobodan B.] At: 08:27 2 June 2008

THEOREM 1 For the values of α and the other relevant parameters from Table 1, there holds
∞
 ∞
 (−1)k F (α − 2k − δ)
(s)n−1 f ((an − b)x) cπ x α−1
= + x 2k+δ . (2)
n=1
(an − b)α 2 (α)f (π α/2) k=0 (2k + δ)!

Proof We make use of the polylogarithm Liα (z) defined by the series (see [4]), and with the
following integral representation

∞  ∞
zn 1 t α−1
Liα (z) = = dt,
n=1
nα (α) 0 et /z − 1

where the right-hand side integral converges for z ∈ C \ {z | z ∈ R, z ≥ 1}, and it is referred to
as Bose’s integral. So for α > 0, we have
∞
 ∞
sin nx i  e−inx − einx i  
= = Liα (e−ix ) − Liα (eix ) ,
n=1
nα 2 n=1 n α 2
∞ ∞
(3)
 cos nx 1  e−inx + einx 1 
= = Liα (e−ix ) + Liα (eix )
n=1
nα 2 n=1 n α 2

and
∞
 (−1)n−1 sin nx i  
= Liα (−eix ) − Liα (−e−ix ) ,
n=1
nα 2
∞
(4)
 (−1)n−1 cos nx 1 
= − Liα (−e−ix ) + Liα (−eix ) .
n=1
nα 2

Also, we have
∞
  
sin(2n − 1)x i 1 −ix−2ix 1
= Liα (e ) − α Liα (e ) − Liα (eix ) − Liα (e2ix ) ,
n=1
(2n − 1) α 2 2 2α
∞   (5)
 cos(2n − 1)x 1 1 1
−ix −2ix
= Liα (e ) − α Liα (e ) + Liα (eix ) − Liα (e2ix )
n=1
(2n − 1) α 2 2 2α

and
∞
 (−1)n−1 sin(2n − 1)x 1 
= Liα (−ieix ) − Liα (−ie−ix ) − Liα (ieix ) + Liα (ie−ix ) ,
n=1
(2n − 1)α 4
∞
(6)
 (−1)n−1 cos(2n − 1)x i  
= Liα (−ieix ) + Liα (−ie−ix ) − Liα (ieix ) − Liα (ie−ix ) .
n=1
(2n − 1)α 4

Here, we note that in order to obtain the right-hand sides of Equation (4) and (6), we have taken
advantage of the representation (−1)n = cos nπ, and then, by means of elementary trigonometric
identities, we split up the product into the sum of two trigonometric functions.
 444 S.B. Tričković et al.

We shall now consider the Mellin transform of the polylogarithm in the form of Bose’s integral.
Downloaded By: [Tričković, Slobodan B.] At: 08:27 2 June 2008

The Mellin transform of a function f and the inverse transform of a function ϕ are (see [6])
 ∞  c+i∞
1
M(f (x)) = x u−1 f (x)dx, M −1 (ϕ(u)) = x −u ϕ(u)du.
0 2π i c−i∞
This integral transform is closely connected to the theory of Dirichlet series and is often used in
number theory and the theory of asymptotic expansions. Also, it is closely related to the Laplace
and Fourier transform as well as to the theory of the gamma function and allied special functions.
So we find
 ∞  ∞  ∞ α−1 u−1
1 t x
M(Liα (p e−x )) = x u−1 Liα (p e−x )dx = t+x /p − 1
dtdx.
0 (α) 0 0 e
The change of variables x = ab, t = a(1 − b) allows the integrals to be separated
 1  ∞
1 a α+u−1
M(Liα (p e−x )) = bu−1 (1 − b)α−1 db da = (u)Liα+u (p). (7)
(α) 0 0 ea /p − 1
For p = 1, because Liα+u (1) = ζ (α + u), through the inverse Mellin transform, we have
 c+i∞
1
Liα (e−x ) = (u)ζ (α + u)x −u du,
2πi c−i∞

where c is a constant to the right of the poles of the integrand. The path of integration may
be converted into a closed contour, and the poles of the integrand are those of (u) at u =
0, −1, −2, . . . , and of ζ (α + u) at u = 1 − α. Summing the residues yields a representation of
the polylogarithm as a power series
∞
 ζ (α − k)
Liα (eμ ) = (−μ)α−1 (1 − α) + μk , |μ| < 2π, α  = 1, 2, 3, . . . (8)
k=0
k!

about μ = 0. Further, following Equation (3), we have

 ζ (α − 2k − 1) ∞
i i
(Liα (e−μ ) − Liα (eμ )) = (μα−1 − (−μ)α−1 ) (1 − α) − i μ2k+1 ,
2 2 k=0
(2k + 1)!
∞
(9)
1 1  ζ (α − 2k)
(Liα (eμ ) + Liα (e−μ )) = ((−μ)α−1 + μα−1 ) (1 − α) + μ2k ,
2 2 k=0
(2k)!

where we replace μ with ix, 0 < x < 2π. For a positive real non-integer α, the gamma function
is finite, and so in view of
π π
(± i)α−1 = e± i(α−1)π/2 = cos (α − 1) ± i sin (α − 1), (10)
2 2
we calculate
i α−1 iπ x α−1 π x α−1
(μ − (−μ)α−1 ) (1 − α) = (i α−1 − (−i)α−1 ) = ,
2 2 (α) sin πα 2 (α) sin(π/2)α
1 πx α−1 π x α−1
((−μ)α−1 + μα−1 ) (1 − α) = ((−i)α−1 + i α−1 ) = ,
2 2 (α) sin πα 2 (α) cos(π/2)α
 Integral Transforms and Special Functions 445

which, in conjunction with Equation (9) gives the sums of the series in Equation (3), i.e.
Downloaded By: [Tričković, Slobodan B.] At: 08:27 2 June 2008

∞
 ∞
 (−1)k ζ (α − 2k − 1)
sin nx πx α−1
= + x 2k+1 ,
n=1
nα 2 (α) sin(π/2)α k=0 (2k + 1)!
∞
 ∞
 (−1)k ζ (α − 2k)
cos nx πx α−1
= + x 2k .
n=1
nα 2 (α) cos(π/2)α k=0 (2k)!

We now consider Equation (4). Because of Equation (7), we find M(Liα (−e−x )) =
− (u)η(α + u), since Liα+u (−1) = −η(α + u). Through the inverse Mellin transform and
conversion of path of integration into a closed contour, after summing the residues at u =
0, −1, −2, . . . and u = 1 − α (which is 0), we have

∞
 η(α − k)
Liα (−eμ ) = − μk , |μ| < 2π, α  = 1, 2, 3, . . . ,
k=0
k!

whence we find
∞
 η(α − 2k − 1)
i
(Liα (−e−μ ) − Liα (−eμ )) = i μ2k+1 ,
2 k=0
(2k + 1)!
∞
(11)
1  η(α − 2k)
− (Liα (−eμ ) + Liα (−e−μ )) = μ2k ,
2 k=0
(2k)!

and for μ = ix (−π < x < π), we obtain the sums of the series in Equation (4), i.e.

∞
 ∞

(−1)n−1 sin nx (−1)k η(α − 2k − 1)
= x 2k+1 ,
n=1
nα k=0
(2k + 1)!
∞
 ∞

(−1)n−1 cos nx (−1)k η(α − 2k)
= x 2k .
n=1
nα k=0
(2k)!

As regards Equation (5), judging by its structure, we have to repeat the procedure as that for
obtaining Equation (8), i.e. once for Liα (e±μ ) (|μ| < 2π ), and once again for Liα (e±2μ ) (|μ| < π ),
so that we have
 
i 1 1
Liα (e−μ ) − α Liα (e−2μ ) − Liα (eμ ) − α Liα (e2μ )
2 2 2
∞
 λ(α − 2k − 1)
i α−1
= (μ − (−μ)α−1 ) (1 − α) − i μ2k+1 ,
4 k=0
(2k + 1)!
 
1 1 1
Liα (e−μ ) − α Liα (e−2μ ) + Liα (eμ ) − α Liα (e2μ )
2 2 2
 λ(α − 2k) ∞
1 α−1
= (μ + (−μ)α−1 ) (1 − α) + μ2k , |μ| < π, α = 1, 2, 3, . . . ,
4 k=0
(2k)!
(12)
 446 S.B. Tričković et al.

where we have taken into account λ(z) = (1 − 2−z )ζ (z). Thus, for μ = ix (0 < x < π ), we
Downloaded By: [Tričković, Slobodan B.] At: 08:27 2 June 2008

obtain the sums of the series in Equation (5), i.e.

∞
 ∞
 (−1)k λ(α − 2k − 1)
sin(2n − 1)x πx α−1
= + x 2k+1 ,
n=1
(2n − 1)α 4 (α) sin(π/2)α k=0 (2k + 1)!
∞
 ∞
 (−1)k λ(α − 2k)
cos(2n − 1)x πx α−1
= + x 2k .
n=1
(2n − 1)α 4 (α) cos(π/2)α k=0 (2k)!

Finally, in the case of Equation (6), we first calculate Liα+u (±i) = 2−(α+u) η(α + u) ± iβ(α +
u) and find M(Liα (±ie−x )) = (u) 2−(α+u) η(α + u) ± iβ(α + u) . Applying again the inverse
Mellin transform and converting the path of integration into closed contour, we evaluate the
residues at u = 0, −1, −2, . . . and u = 1 − α (which is 0), so that after their summation, a
rearrangement and simplification, we obtain
∞
 β(α − 2k − 1)
1 
Liα (−ieμ ) − Liα (−ie−μ ) − Liα (ieμ ) + Liα (ie−μ ) = −i μ2k+1 ,
4 k=0
(2k + 1)!
∞
(13)
i    β(α − 2k) 2k
Liα (−ieμ ) + Liα (−ie−μ ) − Liα (ieμ ) − Liα (ie−μ ) = μ ,
4 k=0
(2k)!

where |μ| < π/2, α  = 1, 2, 3, . . . . For μ = ix(−π/2 < x < π/2), we find the sums of the series
in Equation (6), i.e.
∞
 ∞

(−1)n−1 sin(2n − 1)x (−1)k β(α − 2k − 1)
= x 2k+1 ,
n=1
(2n − 1)α k=0
(2k + 1)!
∞
 ∞
 (−1)k β(α − 2k)
(−1)n−1 cos(2n − 1)x
= x 2k .
n=1
(2n − 1) α
k=0
(2k)!

Gathering all these results, we conclude that Equation (2) holds. 

3. Closed form cases

We shall express now, for certain values of parameters, series (1) as polynomials, saying then that
we have brought the infinite series in closed form.

THEOREM 2 For the values of r and the other relevant parameters from Table 1, there holds
∞ [r/2]
 (s)n−1 f (an − b)x) (−1)[r/2] cπ x r−1  (−1)k F (r − 2k − δ) 2k+δ
= + x . (14)
n=1
(an − b)r 2(r − 1)! k=0
(2k + δ)!

Proof First of all, by setting z = 2n + 1 in the functional equation for the Riemann zeta function
(see [1])
2ζ (z) (z) zπ
ζ (1 − z) = z
cos ,
(2π ) 2
we find ζ (−2n) = 0, n ∈ N. Taking into account the relations of the Riemann zeta functions to
Dirichlet functions η and λ (see Lemma 1), we easily notice that they also vanish at even negative
 Integral Transforms and Special Functions 447

integers. Moreover, λ(0) = 0. It is the other way round with the Catalan function β, i.e. it vanishes
Downloaded By: [Tričković, Slobodan B.] At: 08:27 2 June 2008

at odd negative integers. In order to prove this, we need the Hurwitz zeta function ζ (z, a) initially
defined for σ > 1 (z = σ + iτ ) by the series
+∞
 1
ζ (z, a) = ,
k=0
(k + a)z

where a is a fixed real number, 0 < a ≤ 1. The function β(z) can now be represented as follows
+∞
   
(−1)k −z 1 3
β(z) = =4 ζ z, − ζ z, ,
k=0
(2k + 1) z 4 4

whence we have
  
1 3
β(−(2n − 1)) = 4 2n−1
ζ −(2n − 1), − ζ −(2n − 1), , n ∈ N.
4 4

Considering that for a non-negative integer n, there holds (see [1]) ζ (−n, a) = −Bn+1 (a)/(n + 1),
where Bn (x) are Bernoulli polynomials defined by relations

 +∞
text tn
= B n (x) , B0 (x) = 1,
et − 1 n=0
n!

we first replace x with 1 − x, and have

+∞
  +∞
tn te(1−x)t (−t)e(−t)x (−1)n t n
Bn (1 − x) = t = −t = Bn (x) ,
n=0
n! e −1 e −1 n=0
n!

whence we obtain Bn (1 − x) = (−1)n Bn (x) implying B2n (3/4) = B2n (1/4), so that we calculate

B2n (1/4) B2n (3/4)
β(−(2n − 1)) = 42n−1 − + = 0.
2n 2n

It is known that for r ∈ N, the value (1 − r) of the gamma function becomes infinite, and
we cannot place α = r immediately in Equation (9). In order to get a finite value we take, on the
right-hand side sum in Equation (9), the first m − 1 terms if r = 2m − 1, and the first m terms if
r = 2m. Then we take limits

i α−1  ζ (α − 2k − 1)
m−1
lim (μ − (−μ)α−1 ) (1 − α) − i μ2k+1
α→2m−1 2 k=0
(2k + 1)!

log(−μ) − log μ 2m−2  ζ (2m − 2k − 2)

m−1
= −i μ −i μ2k+1 ,
2(2m − 2)! k=0
(2k + 1)!

1  ζ (α − 2k) m
lim ((−μ)α−1 + μα−1 ) (1 − α) + μ2k
α→2m 2 k=0
(2k)!

log(−μ) − log μ 2m−1  ζ (2m − 2k) 2k

m
=− μ + μ .
2(2m − 1)! k=0
(2k)!
 448 S.B. Tričković et al.

By virtue of ζ (−2n) = 0 (n ∈ N), the remainders of the series in Equation (9) vanish, so placing
Downloaded By: [Tričković, Slobodan B.] At: 08:27 2 June 2008

μ = ix, 0 < x < 2π, we obtain

∞
 (−1)m−1 πx 2m−2  (−1)k ζ (2m − 2k − 2) 2k+1
m−1
sin nx
= + x ,
n=1
n2m−1 2(2m − 2)! k=0
(2k + 1)!
(15)
∞
 (−1)m πx 2m−1  (−1)k ζ (2m − 2k) 2k
m
cos nx
= + x (m ∈ N),
n=1
n2m 2(2m − 1)! k=0
(2k)!

which is formula (14) for s = 1, a = 1, b = 0, c = 1, F = ζ , f = cos

sin
, r = 2m−1
2m
.
When dealing with Equation (11), we do not have a problem with the singularity of the gamma
function any more, so that we may replace there α with r ∈ N, and considering that η(−2n) = 0
(n ∈ N), we obtain
∞
 
m−1
(−1)n−1 sin nx (−1)k η(2m − 2k − 2) 2k+1
= x ,
n=1
n2m−1 k=0
(2k + 1)!
(16)
∞
  (−1)k η(2m − 2k)
m
(−1)n−1 cos nx
= x 2k (m ∈ N),
n=1
n2m k=0
(2k)!

which is formula (14) for the choice of parameters s = −1, a = 1, b = 0, c = 0, F = η, f =

sin
cos
r = 2m−1
2m
.
In the case of Equation (12), there appears again the singularity of (1 − r) at r ∈ N, so quite
similarly as above we take limits, and after placing μ = ix, 0 < x < π, because λ(−2n) = 0
(n ∈ N), we obtain
∞
(−1)m−1 πx 2m−2  (−1)k λ(2m − 2k − 2) 2k+1
m−1
sin(2n − 1)x
= + x ,
n=1
(2n − 1)2m−1 4(2m − 2)! k=0
(2k + 1)!
(17)
∞
 (−1)m πx 2m−1  (−1)k λ(2m − 2k) 2k
m
cos(2n − 1)x
= + x (m ∈ N),
n=1
(2n − 1)2m 4(2m − 1)! k=0
(2k)!

which is formula (14) for the choice of s = 1, a = 2, b = 1, c = 1/2, F = λ, f = cos sin

r=
2m−1
2m
.
Finally, in Equation (13), similarly as for Equation (11), we do not have to deal with the
singularity of the gamma function and replacing α with r ∈ N, considering that β(−2n + 1) = 0
(n ∈ N), we obtain
∞
 
m
(−1)n−1 sin(2n − 1)x (−1)k β(2m − 2k − 1)
= x 2k+1 ,
n=1
(2n − 1)2m k=0
(2k + 1)!
(18)
∞
 
m−1
(−1)n−1 cos(2n − 1)x (−1)k β(2m − 2k) 2k
= x (m ∈ N),
n=1
(2n − 1)2m−1 k=0
(2k)!

which is formula (14) for the choice of parameters s = −1, a = 2, b = 1, c = 0, F = β, f =

sin
cos
r = 2m−1
2m
, wherby we complete the proof. 

We note that because of λ(0) = 0 the upper bounds in Equation (17) are actually m − 2 and
m − 1 and m − 1 for the first sum in Equation (18) since β(−1) = 0, but purely for the sake of
fitting them into a general formula (14), we retain m − 1 and m in Equation (17), and m for the
first sum in Equation (18), without formally changing values.
 Integral Transforms and Special Functions 449

3.1. Some applications

Downloaded By: [Tričković, Slobodan B.] At: 08:27 2 June 2008

Now we are going to present significant and important applications of our closed form
formula (14).

3.1.1. Integral transforms

Applying some of the integral transforms, one can obtain various series in closed form. For
instance, if we take f = cos, δ = 0 in Equation (14), then apply the Laplace transform, we have

∞ [r/2]
 (s)n−1 p  (−1)i F (r − 2i)
[r/2] cπ
= (−1) + .
n=1
(an − b)r (p 2 + (an − b)2 ) 2p r i=0
p 2i+1

Further, if we set s = 1, a = 1, b = 0, then F = ζ, c = 1 must be taken, and we obtain

∞ [r/2]
 p  (−1)i ζ (r − 2i)
[r/2] π
= (−1) + . (19)
n=1
nr (p 2 + n2 ) 2p r i=0
p 2i+1

Now we apply the inverse Mellin transform to this series, knowing that (see [6, p. 166, 2.16 and
p. 167, 2.25])

 ⎧
  ⎨ (log(1/x))
ν−1
z cos(n log x) x < 1 1 x<1
M −1 = M −1 = (ν)
z + n2
2 0 x > 1, zν ⎩
0 x > 1,

coming to the sum of a trigonometric series

∞
   (−1)i ζ (r − 2i) 
[r/2]
cos(n log x) (−1)[r/2] π 1 r−1
1 2i
= log + log , x < 1.
n=1
nr 2(r − 1)! x i=0
(2i)! x

However, if we want to apply the Bessel instead of Mellin transform to series (19), we first refer
to (see [5, p. 36, 4.23 and p. 33, 4.6])

x ν+1/2 a ν−μ+1 y μ−1/2
B = Kν−μ+1 (ay)
(a 2 + x 2 )μ 2μ−1 (μ)

and
2μ+1/2 ((ν/2) + (μ/2) + (3/4))
B(x μ ) = ,
y μ+1 ((ν/2) − (μ/2) + (1/4))

where Kν is Hankel’s function. Thus we obtain the sum of one of the Schlömilch series

∞ [r/2]
 K1/2 (ny) (−1)[r/2] πy r−(3/2) (1 − (r/2))  (−1)i ζ (r − 2i) ((1/2) − i) 2i−1/2
= + y .
n=1
nr−(1/2) 2r+1/2 ((r + 1)/2) i=0
i! 22i+1/2
 450 S.B. Tričković et al.

3.1.2. Convergence acceleration

Downloaded By: [Tričković, Slobodan B.] At: 08:27 2 June 2008

On the basis of Equation (14), the convergence acceleration of trigonometric series is obtained.
As an example, we consider the series
∞
 n
T = sin nx. (20)
n=1
n2 + 1

For each natural number M we prove, by the method of mathematical induction, that there holds

n n 1 1 1 1
= · = 1 − 2 + 4 − ···
n2 + 1 n2 1 + (1/n2 ) n n n
(21)
M
(−1)m−1 (−1)M
= + 2M−1 2 .
m=1
n2m−1 n (n + 1)

Replacing Equation (21) in Equation (20), we have


M ∞
(−1)M sin nx
T = sin
(−1)m−1 T2m−1 + ,
m=1 n=1
(n2 + 1)n2M−1

sin
where T2m−1 denotes first of the closed form formulas (15), i.e.
∞
 (−1)m−1 πx 2m−2  (−1)k ζ (2m − 2k − 2) 2k+1
m−1
sin nx
sin
T2m−1 = = + x .
n=1
n2m−1 2(2m − 2)! k=0
(2k + 1)!

So, the greater M, the faster convergence of the remaining series, and accordingly we have the
faster convergence of the series T . Here is shown how many terms of the remaining series ought
to be taken for the given M and accuracy ε

ε 10−1 10−2 10−5 10−8

M=1 3 8 224 7072

M=3 2 2 6 16
M = 10 1 2 2 3

4. Limiting values

THEOREM 3 For the values of α complementary to those in Table 1, there holds

∞
 ∞

(s)n−1 f ((an − b)x) (−1)k F (α − 2k − δ) 2k+δ
= α (x) + x , (22)
n=1
(an − b)α k=[(α−1)/2]+1
(2k + δ)!

where
[(α−1)/2]
(−1)[(α−1)/2] c  (−1)[(α−1)/2]−k F (2k + 1) α−2k−1
α (x) = (ψ(α) + γ − log x) x α−1 + x
(α − 1)! k=1
(α − 2k − 1)!

and if c = 1, then F = ζ or if c = 1/2, then F = λ.

 Integral Transforms and Special Functions 451

Proof We have seen that if we let α tend to 2m − 1 and to 2m, respectively, in Equation (9), the
Downloaded By: [Tričković, Slobodan B.] At: 08:27 2 June 2008

right-hand side sums truncate because the zeta function vanish at even negative integers, and as
limiting values, we obtain polynomials (see Equation (15)). However, if we let α tend to 2m and
to 2m − 1, respectively in Equation (9), we find

i α−1  ζ (α − 2k − 1)
m−1
lim (μ − (−μ)α−1 ) (1 − α) − i μ2k+1
α→2m 2 k=0
(2k + 1)!
 
m−1
i log(−μ) + log μ ζ (2k + 1)
= ψ(2m) + γ − μ2m−1 − i μ2m−2k−1
(2m − 1)! 2 k=1
(2m − 2k − 1)!

1 
m−1
ζ (α − 2k) 2k
lim ((−μ)α−1 + μα−1 ) (1 − α) + μ
α→2m−1 2 k=0
(2k)!

log(−μ) + log μ 2m−2  ζ (2k + 1)
m−1
1
= ψ(2m − 1) + γ − μ + μ2m−2k−2 ,
(2m − 2)! 2 k=1
(2m − 2k − 2)!

where ψ is the digamma function. Setting μ = ix, 0 < x < 2π , we obtain

∞
 sin nx (−1)m−1
= (ψ(2m) + γ − log x) x 2m−1
n=1
n2m (2m − 1)!

 ∞
(−1)m−1−k ζ (2k + 1) 2m−2k−1  (−1)k ζ (2m − 2k − 1) 2k+1
m−1
+ x + x ,
k=1
(2m − 2k − 1)! k=m
(2k + 1)!
∞
(23)
 cos nx (−1)m−1
= (ψ(2m − 1) + γ − log x) x 2m−2
n=1
n 2m−1 (2m − 2)!

 ∞
(−1)m−1−k ζ (2k + 1) 2m−2k−2  (−1)k ζ (2m − 2k − 1) 2k
m−1
+ x + x ,
k=1
(2m − 2k − 2)! k=m
(2k)!

which is Equation (22) for s = 1, a = 1, b = 0, c = 1, F = ζ , f = cos

sin
α = 2m−1
2m
δ = 01 .
In the case of Equation (12), bearing in mind λ(z) = (1 − 2−z )ζ (z), and applying the same
procedure as above yields
∞
 sin(2n − 1)x (−1)m−1
= (ψ(2m) + γ − log x) x 2m−1
n=1
(2n − 1)2m 2(2m − 1)!

 ∞
(−1)m−1−k λ(2k + 1) 2m−2k−1  (−1)k λ(2m − 2k − 1) 2k+1
m−1
+ x + x ,
k=1
(2m − 2k − 1)! k=m
(2k + 1)!
∞
 cos(2n − 1)x (−1)m−1
= (ψ(2m − 1) + γ − log x) x 2m−2
n=1
(2n − 1)2m−1 2(2m − 2)!

 ∞
(−1)m−1−k λ(2k + 1) 2m−2k−2  (−1)k λ(2m − 2k − 1) 2k
m−1
+ x + x ,
k=1
(2m − 2k − 2)! k=m
(2k)!

which is Equation (22) for s = 1, a = 2, b = 1, c = 1/2, F = λ, f = sin

cos
α= 2m
2m−1
δ= 1
0
.

 452 S.B. Tričković et al.

Remark 2 We note that formulas (23) can be obtained by letting in Equation (2) α tend to 2m if
Downloaded By: [Tričković, Slobodan B.] At: 08:27 2 June 2008

f = sin or to 2m − 1 if f = cos (m ∈ N), since we encounter division by zero. Apart from this,
the left-hand side series are known as Clausen functions defined by (see [4])
∞
 ∞

sin nx cos nx
Cl2ν (x) = , Cl2ν−1 (x) = , ν ∈ N,
n=1
n2ν n=1
n2ν−1

and, as a by-product, we have managed to express each of them as the sum of a polynomial and
a power series in terms of Riemann’s ζ function, that is

(−1)ν−1  (−1)ν−1−k ζ (2k + 1)

ν−1
Cl2ν (x) = (ψ(2ν) + γ − log x) x 2ν−1 + x 2ν−2k−1
(2ν − 1)! k=1
(2ν − 2k − 1)!
∞
 (−1)k ζ (2ν − 2k − 1)
+ x 2k+1 , ν ∈ N,
k=ν
(2k + 1)!

(−1)ν−1  (−1)ν−1−k ζ (2k + 1)

ν−1
Cl2ν−1 (x) = (ψ(2ν − 1) + γ − log x) x 2ν−2 + x 2ν−2k−2
(2ν − 2)! k=1
(2ν − 2k − 2)!
∞
 (−1)k ζ (2ν − 2k − 1)
+ x 2k , ν ∈ N.
k=ν
(2k)!

References

[1] T.M. Apostol, Introduction to Analytic Number Theory, Springer-Verlag, New York, 1976.
[2] I.S. Gradshteyn and I.M. Ryzhik, Tables of Integrals, Series and Products, A. Jeffrey and D. Zwillinger, eds., Academic
Press, New York, 2000.
[3] K. Knopp, Theory and Application of Infinite Series, Dover, New York, 1990.
[4] L. Lewin, Polylogarithms and Associated Functions, North Holland, New York, 1975.
[5] F. Oberhettinger, Tables of Bessel Transform, Springer-Verlag, 1972.
[6] ———, Tables of Mellin Transform, Springer-Verlag/Academic Press, 1974.
[7] A.P. Prudnikov, Y.A. Brychkov, and O.I. Marichev, Integrals and Series, Elementary Functions, Vol. 1 Gordon and
Breach Science Publishers, New York, 1986.