Summation rules for the values of the Riemann zeta–function and generalized

harmonic numbers obtained using Laurent developments of polygamma

functions and their products

Sergey K. Sekatskii

Laboratory of Biological Electron Microscopy, IPHYS, BSP 419, Ecole Polytechnique

Fédérale de Lausanne, and Dept. of Fundamental Biology, Faculty of Biology and Medicine,
University of Lausanne, CH1015 Lausanne-Dorigny, Switzerland.
E-mail : [email protected]

Following the Mellin and inverse Mellin transforms technique presented in our paper
arXiv:1606.02150 (NT), we have established close forms of Laurent series expansions
of products of bi- and trigamma functions ψ '( z )ψ (− z ) and ψ '( z )ψ '(− z ) . These series
were used to find summation rules which include generalized harmonic numbers of
first, second and third powers and values of the Riemann zeta-functions at integers /
∞
H k(2)
Bernoulli numbers, for example 2= ∑
k =1 k
3
6ς (2)ς (3) − 9ς (5) . Some of these rules

were tested numerically.

1
 Introduction
In a recent paper [1] we have established numerous integral representations of the
Riemann zeta-function and other related functions as Mellin transforms of appropriate
trigonometrical functions. Then, using inverse Mellin transform technique, we obtained
Laurent expansions of the trigonometrical functions involved, which in their turn were
used to establish summation rules for the Riemann zeta-functions and closely related
with them Bernoulli, Euler, or some related numbers. In the last Section of [1], not
trigonometrical functions but digamma functions and its derivatives were used for the
same purposes, and we sketched the program of further applications of this approach.
In this short Note we report the first realizations of such a program.

1. Summation rules
1.1 Product of trigamma functions
Let us start with the function ψ '( z )ψ '(− z ) which is an “even product” of two
trigamma functions (derivatives of the digamma functions, see e.g. [2] for necessary
definitions and basic properties). We know the Laurent series
1 ∞
ψ ( x) = − − γ + ∑ (−1) n−1ς (n + 1) x n (1),
x n =1

valid for | x |< 1 (see again [2]; these series were also obtained in [1]). Here ς ( s) is the
Riemann zeta-function (see e.g. [3] for its main properties) and γ is Euler –
Mascheroni constant. From (1) we have
1 ∞
ψ '( x) = + ∑ (−1)n (n + 1)ς (n + 2) x n
x 2 n =0
(2),
1 2ς (2)
and it is easy to see that the function F=
( z ) : ψ '( z )ψ '(− z) − 4
− 2 − ς 2 (2) − 6ς (4) is
z z
O(z2) at zero and O(1) at infinity (trigamma function is O(1/z) at infinity [2]), hence the
1 2ς (2)
contour integral ∫z
− s −1
(ψ '( z )ψ '(− z ) − 4
− 2 − ς 2 (2) − 6ς (4))dz exists in the strip
C z z
0<Res<2. Here the contour C consists of the line –iX, +iX and demi-circle connecting
these points and lying in the right complex semi-plane; X is real and tends to infinity.
(If a pole of the function under integral sign occurs on the contour, just change the
value of X a bit to avois this). In this strip, this integral is equal to (for consistence, we
repeat the material from [1]):
 1 2ς (2) 
∫z ψ '( z )ψ '(− z ) − 4 − 2 − ς (2) − 6ς (4) dz
− s −1 2

C  z z 
∞
 1 2ς (2) 
=−i (i − s −1 + (−i ) − s −1 ) ∫ x − s −1 ψ '(ix)ψ '(−ix) − 4 + 2 − ς 2 (2) − 6ς (4) dx
0  x x 

2
 ∞
π ( s + 1)  1 2ς (2) 
−2i cos(
= ) ∫ x − s −1 ψ '(ix)ψ '(−ix) − 4 + 2 − ς 2 (2) − 6ς (4) dx
2 0  x x 
∞
πs  1 2ς (2) 
2i sin( ) ∫ x − s −1 ψ '(ix)ψ '(−ix) − 4 + 2 − ς 2 (2) − 6ς (4) dx (3).
2 0  x x 
From the other side, this contour integral can be calculated using Cauchy residues
∞ ∞
theorem as equal to 2π i(∑ n − s −2ψ '(n)( s + 1) − ∑ψ ''(n)n − s −1 ) : in the vicinity of points z=1, 2,
=n 1=n 1

3… the function ψ (− z ) is given by [1, 2]

1 ∞
ψ (−n − x) = + H n − γ + ∑ ((−1) k H n( k +1) − ς (k + 1)) x k (4),
x k =1

whence its derivative has here poles of the second order having Laurent expansion
1 n
1
− 2
+ O (1) . Here H n = ∑ k is generalized harmonic number, H 0( k ) = 0 by definition.
(k )

x l =1 l
n
1
Below we will omit superscript 1 for H n(1) = ∑ writing it simply as H n .
l =1 l

Now we need to introduce the functions

∞
χ k ( s ) := ∑ψ ( k ) (n)n − s (5),
n =1

d kψ
where ψ ( k ) = (similar functions were introduced and considered earlier, see e.g.
dz k
[4]) to write in a compact form
∞
 1 2ς (2)  π
∫x ψ '(ix)ψ '(−ix) − 4 + 2 − ς (2) − 6ς (4) dx = (( s + 1) χ1 ( s + 2) − χ 2 ( s + 1)) (6).
− s −1 2

 x x  πs
0 sin( )
2
This is a Mellin transform. Inverse Mellin transform
c + i∞
1  ( s + 1) χ1 ( s + 2) − χ 2 ( s + 1) 
F (ix) = ∫ π xs  ds (it is easy to see, that all conditions to
2π i c −i∞  sin( sπ / 2) 
apply the corresponding theorem hold, see [1], [5], we take 0<c<2), by virtue of
Cauchy residues theorem (simple poles are located at s=2n, n=1, 2, 3… with the
residues of 2 / π and alternating signs) then easily supplies the following Laurent series
development valid for | x |< 1 :
1 2ς (2) ∞
ψ '(iz )ψ '(−iz ) −
z 4
+
z 2
− ς 2
(2) − 6ς (4)
= 2 ∑
n =1
(−1) n ((2n + 1) χ1 (2n + 2) − χ 2 (2n + 1)) z 2 n and thus

1 2ς (2) ∞
ψ '( z )ψ '(− z ) − 4 − 2 − ς 2 (2) − 6= ς (4) 2∑ ((2n + 1) χ1 (2n + 2) − χ 2 (2n + 1)) z 2 n (7).
z z n =1

3
 To obtain the summation rules, this should be compared with the product of
1 ∞
ψ '( x) , eq. (2), and ψ '(− x) = 2
+ ∑ (n + 1)ς (n + 2) x n . This gives for the coefficients
x n =0
standing in front of x2n, n=1, 2, 3…:
2n
2(2n + 3)ς (2n + 4) + ∑ (−1)l (l + 1)(2n − l + 1)ς (l + 2)ς (2n − l + 2)
= 2(2n + 1) χ1 (2n + 2) − 2 χ 2 (2n + 1)
l =0

(8).
2 3
∞
1  ∞
1 
Now, we remind that ψ '( z ) = ∑   and ψ ''( z ) = −2∑   . Using harmonic
l =0  l + z  l =0  l + z 
1
numbers notation, we can write ψ '(=
n) ς (2) − H n(2) −1 = ς (2) − H n + 2 ,
(2)

n
(2)
2 ∞
H
ψ ''(n) = −2ς (3) + 2 H n(3)−1 = ( s ) ς (2)ς ( s ) − ∑ ns + ς ( s + 2) and
−2ς (3) + 2 H n(3) − 3 and obtain χ1=
n n =1 n
(3)
∞
H
χ 2 (s) = −2(ς (3)ς ( s ) − ∑ ns + ς ( s + 3)) , that is
n =1 n
∞
H (2)
χ1 (2n= + 2) ς (2)ς (2n + 2) − ∑ 2 nk+ 2 + ς (2n + 4)
л =1 k
∞
H (3)
χ 2 (2n + 1) = −2ς (3)ς (2n + 1) + 2∑ 2 nk +1 − 2ς (2n + 4)) . Thus we can recast eq. (8) as the
k =1 k

following Proposition:

Proposition 1. For n=1, 2, 3…, we have

2 n −1
H k(2) ∞
H k(3)
∞

∑ (−1) (l + 1)(2n − l + 1)ς (l + 2)ς (2n=

=l 1
l
− l + 2) 4ς (3)ς (2n + 1) − 2(2n + 1)∑ 2 n + 2 − 4∑ 2 n +1 (10).
= k
k 1= k 1 k

For n=2, 3… this can be written as:

2n−2
H k(2) ∞ ∞
H k(3)
=l 2
∑ (−1)l (l + 1)(2n − l + 1)ς (l + 2)ς (2n − l + 2)
= 4(2n + 1)ς (3)ς (2n + 1) − 2(2n + 1)∑
= k 2n+2
k 1=
− 4 ∑
k 1 k
2 n +1

(10a).
The first two examples of (10), corresponding to respectively n=1 and n=2, are
H k(2) ∞ ∞
H k(3)
4
3∑+ 2 ∑ 3
4ς 2 (3)
= (11)
=k 1=k k 1 k

and
H k(2)
∞ ∞
H k(3)
10∑
6
+ 4 ∑ 5
= 20ς (3)ς (5) − 9ς 2 (4) (12).
=k 1=k k 1 k
Equalities (11), (12) have been tested numerically.

4
 1.2 Function ψ ( x)ψ '(− x)
Next let us analyse the case of ψ ( x)ψ '(− x) ± ψ '( x)ψ (− x) . We trivially have, from
1 π3 ∞
earlier established [1] ψ ( x)ψ (− x) =−
x 2
+ γ 2
+
3
− 2 ∑
n =1
χ 0 (2n + 1) x 2 n that

2 ∞
ψ '( x)ψ (− x) − ψ ( x)ψ '(− x) =
x 3
− 4 ∑
n =1
n χ 0 (2n + 1) x 2 n −1 (13),
∞
here χ 0 ( s) := ∑ψ (n)n − s .
n =1

To find Laurent series for ψ ( x)ψ '(− x) + ψ '( x)ψ (− x) , let us introduce the function
2γ
( x) : ψ ( x)ψ '(− x) + ψ (− x)ψ '( x) +
F1= + 2γς (2) + 6ς (3) . We have that in the vicinity of zero
x2
F1 ( x) = O( x 2 ) , so that the contour integral ∫z
− s −1
F1 ( z )dz exists in the strip 0<Res<2.
C

Using the same technique as above, we get

∞
π ∞

∫ x F1 (ix)dx
= − s −1
∑
sin(π s / 2) n =1
(2ψ '(n)n − s −1 − ( s + 1)ψ (n)n − s + 2 ) , and then using again the inverse
0

Mellin transform theorem, we obtain Taylor expansion

2γ ∞
ψ '(ix)ψ (−ix) + ψ (ix)ψ '(−ix) =
x 2
− 2γς (2) − 6ς (3) + 2 ∑
n =1
(−1) n (2 χ1 (2n + 1) − (2n + 1) χ 0 (2n + 2)) x 2 n

and thus
2γ ∞
ψ '( x)ψ (− x) + ψ ( x)ψ '(− x) =−
x 2
− 2γς (2) − 6ς (3) − 2 ∑
n =1
(2 χ1 (2n + 1) − (2n + 1) χ 0 (2n + 2)) x 2 n

(14).
Together with (13), we have
1 γ ∞ ∞
ψ '( x)ψ (− x) = −
x3 x 2
− γς (2) − 3ς (3) − ∑
n =1
(2 χ1 (2 n + 1) − (2 n + 1) χ 0 (2 n + 2)) x 2n
− ∑
n =1
2n χ 0 (2n + 1) x 2 n −1

(15)
and
1 γ ∞
ψ ( x)ψ '(− x) =−
x 3
−
x 2
− γς (2) − 3ς (3) − ∑
n =1
(2 χ1 (2n + 1) − (2n + 1) χ 0 (2n + 2)) x 2 n
∞
+ ∑ 2n χ 0 (2n + 1) x 2 n −1 (16).
n =1

Summation rules are obtained comparing (15) with the product of (1) and (2),
1 ∞
the latter is taken as ψ '(− x) = 2
+ ∑ (n + 1)ς (n + 2) x n .
x n =0

5
 To simplify the logistics, we put here ς (1) = γ (formally, of course, but it is
1
known that this is Cauchy principal value
= γ limε →0 (ς (1 + ε ) + ς (1 − ε )) [3]), and rewrite
2
1 ∞
(1) as ψ ( x) =− + ∑ (−1)n −1 ς (n + 1) x n . We also remind
x n =0
∞ ∞ ∞
χ0 (s)
=
= n 1=
∑=
ψ ( n) n − s
n 1
∑ H n−1n− s − γς (s) = ∑ H n n −s − γς (s) − ς (s + 1)
n =1
(17)
∞
H
whence χ (2n=
0 + 1)
k =1
k
2 n +1 ∑k − γς (2n + 1) − ς (2n + 2) .

We have the following summation rule equating the coefficients standing in

front of the x2n-1, n=1, 2, 3…:
Proposition 2. For n=1, 2, 3…, we have
Hk ∞2 n −1
2n ∑
=
2 n +1 ∑ (−1)l −1 (2n − l )ς (l + 1)ς (2n − l + 1) (18).
= k
k 1= l 1

Actually, already eq. (13) is sufficient to establish (18). This is not an Euler
equality
m−2 ∞
Hn
∑ ς (k + 1)ς (m − k ) = (m + 2)ς (m + 1) − 2∑ n
k =1 n =1
m
(19),
(also obtained by us in [1] working with the expansion
1 π 2 ∞
ψ ( x)ψ (− x) +
x 2
− γ 2
−
3
= 2 ∑
n =1
χ (2n + 1) x 2 n ) but can be shown to be equivalent to it when

transformed applying known summation rules for Riemann zeta-functions. Moreover,

the result of Proposition 2 can be considered as a proof of the following summation
rule:
Proposition 3. For n=1, 2, 3…, we have
1 2 n −1 2 n −1

∑ (−1)l (2n − l )ς (l + 1)ς (2n − l=

n l =1
+ 1) ∑ ς (k + 1)ς (2n + 1 − k ) − (2n + 3)ς (2n + 2)
k =1
(20)

For example, the first two rules, following from (18) for n=1 and n=2
H n ς 2 (2) π 4
∞ ∞
Hn π6
respectively, are ∑ =
n =1 n 3
=
2 72
and 2 ∑
n =1 n 5
=2ς (2) ς (4) − ς 2
(3) =
270
− ς 2 (3) , while
∞
Hn 5 1 2 π4
from Euler summation rules (19) we have ∑
n =1 n
3
= ς (4) − ς (2) =
2 2 72
and
∞
H π6
2∑ 5n =7ς (6) − 2ς (2)ς (4) − ς 2 (3) = − ς 2 (3) , etc.
n =1 n 270
A bit more work is required for even powers of (15) or (16). We obtain first

6
 2n

∑ (−1)
l =0
l −1
(2n − l + 1)ς (l + 1)ς (2n − l + 2) − (2n + 3)ς (2n + 3) =−(2 χ1 (2n + 1) − (2n + 1) χ 0 (2n + 2))

(21),
∞
Hk
and then we have χ 0 (2n=
+ 2) ∑k
k =1
2n+2
− ς (1)ς (2n + 2) − ς (2n + 3)

H k(2)
∞
n + 1) ς (2)ς (2n + 1) − ∑ 2 n +1 + ς (2n + 3) ; see above. Thus we have proven the
and χ1 (2=
k =1 k

following Proposition.
Proposition 4. For n=1, 2, 3…, we have
2 n −1
Hk ∞ ∞
H k(2)
∑ (−1)l −1 (2n − l + 1)ς (l + 1)ς (2n − l + 2)= (2n + 1)∑
=l 2 = k 2n+2
k 1=
+ 2 ∑
k 1 k
2 n +1
− (2n + 1)ς (2)ς (2n + 1)

(22).
The first two summation rules following from (22) for n=1 and n=2 respectively
are
Hk ∞ ∞
H k(2)
3∑
+ 2 ∑ 3ς (2)ς (3)
= (23)
= k4
k 1= k 1 k
3

and
Hk ∞ ∞
H k(2)
5∑
+ 2 ∑ = 5ς (2)ς (5) − ς (3)ς (4) (24).
= k6
k 1= k 1 k
5

Equalities (23) and (24) have been tested numerically.

∞
H k(2)
Remark. Terms ∑ 2 n +1 can be isolated from (22) and then the values for the
k =1 k
∞
H
terms ∑ 2 nk+ 2 can be substituted using e.g. Euler summation rule to give an expression
k =1 k
n
1
for the sum involving H n(2) = ∑ 2 . For example, from (23) we have
l =1 l
(2)
∞
Hk ∞
Hk ∞
Hk
2=∑ k3
k 1=
3ς (2)ς (3) − 3∑k 1 k
4
, and thus using ∑ =
k =1 k
4
3ς (5) − ς (2)ς (3) :
∞
H k(2)
2=∑
k =1 k
3
6ς (2)ς (3) − 9ς (5) (25).

7
 1.3 Function ψ 2 ( z )
Just for completeness, let us now briefly present the easiest example of the
1 2( H n − γ )
function ψ 2 ( z ) . By squaring of eq. (4), we obtain ψ 2 (−(n + x)) = + + O(1) .
x2 x
Thus for this function we have the poles of the second order, of the type 1/z2, and the
poles of the first order with the residues 2( H n − γ ) at points z=0, -1, -2… The contour
 2 2 
integral ∫z
− s −1
ψ ( z ) + ψ (− z ) − 2 − 2γ + 2ς (2) dz exists in the strip 0 < Re s < 2 and is
2 2

C  z 
∞
s 2
equal to 2i sin( π ) ∫ x − s −1 ψ 2 (ix) + ψ 2 (−ix) + 2 − 2γ 2 + 2ς (2) dx . From other side, by the
2 0  x 
Cauchy residues law, it is given by
∞
s +1 Hn − γ ∞
Hn
2π i ∑ ( s+2
− 2 s +1
=) 2π i (( s + 1)ς ( s + 2) + 2γς ( s + 1) − 2 ∑ s +1
) . Repeating our standard
n n
1= n n 1 n

consideration with inverse Mellin transform, we get

2 ∞ ∞
H k 2n
ψ 2 (ix) + ψ 2 (−ix) +
x
=
2
− 2γ 2
+ 2ς (2)
= 2 ∑ (
n 1=
−1) n
((2 n + 1)ς (2 n + 2) + 2γς (2 n + 1) − 2 ∑
k 1 k
2 n +1
)x

and thus
2 ∞ ∞
H k 2n
ψ 2 ( x) + ψ 2 (− x) −
= x 2
− 2γ 2
+ 2
=ς (2) 2 ∑ ((2
n 1=
n + 1)ς (2 n + 2) + 2γς (2 n + 1) − 2 ∑
k 1 k
2 n +1
)x .

4γ ∞ ∞
H
Similarly, ψ 2 ( x) − ψ 2 (− x) + =−2∑ (2nς (2n + 1) + 2γς (2n) − 2∑ 2kn )x 2 n −1 , and finally
= x n 1= k 1 k

1 2γ ∞ ∞
H
ψ 2 ( x) − 2 − γ 2 + ς (2) + = ∑ (−1) n ((n + 1)ς (n + 2) + 2γς (n + 1) − 2∑ n +k1 )x n (26).
= x x n 1= k 1 k

Comparing (26) with the square of (1) we get the following summation rule:

Proposition 5. For n= 2, 3…, we have

Hk ∞ n−2
= ( n + 2)ς ( n2∑
+ 1) − ∑ ς (n − k )ς (k + 1) (27).
= kn
k 1= k 1

Not surprisingly, here is nothing new: these are indeed Euler summation rules
(19), see Proposition 2. The first of them (not appearing in our study above) is Euler
∞
Hk
equality ∑k
k =1
2
= 2ς (3) .

2. Conclusion
Following the program outlined in our publication [1], we established numerous
summation rules involving the values of the Riemann zeta-function at integers and
generalized harmonic numbers. Certainly, in all these rules the values of the Riemann

8
 (−1) n−1 2(2n)!
B2 n = ς (2n) [2].
(2π ) 2 n
This line of research can be continued further by considering polygamma functions
of larger order and their products, such as, say, [ψ '( x)]2 ± [ψ '(− x)]2 , or
ψ ( x)ψ ''( x) ± ψ (− x)ψ ''(− x) and so forth. This leads to the summation rules involving larger
powers of harmonic numbers.
The present author indeed undertook some efforts in this direction, but obtained
formulae are complicated and do not look very useful, so we do not present them here.
Similarly, we continued also by considering e.g. the functions of the type
ψ (n + x) ± ψ (n − x) or ψ (n + x)ψ (n − x) - but again, obtained formulae are cumbersome
and do not look truly interesting.

References
[1] S. K. Sekatskii, Novel integral representations of the Riemann zeta-function and
Dirichlet eta-function, close expressions for Laurent series expansions of powers of
trigonometric functions and digamma function, and summation rules,
arXiv:1606.02150 (NT).
[2] M. Abramovitz and I. A. Stegun. Handbook of mathematical functions.
Washington D.C.; New York: United States Department of Commerce, National
Bureau of Standards; Dover Publications, 1983.
[3] E. C. Titchmarsh and E. R. Heath-Brown. The theory of the Riemann Zeta-
function. Clarendon Press, Oxford, 1988.
[4] T. M. Apostol and T. H. Vu, Dirichlet series related to the Riemann zeta
function. J. Number Theory, 19, 85-102 (1984).
[5] O. I. Marichev. Handbook of integral transformations of higher transcendental
functions: Theory and Algorithmic Tables. Ellis Horwood Ltd., Chichester, UK,
1983.