JOURNAL OF

COMPUTATIONAL AND
APPLIED MATHEMATICS

ELSEVIER Journal of Computational and Applied Mathematics 100 (1998) 191-199

Polygamma functions of negative order

Victor S. A d a m c h i k *
100 Trade Center Drive, Campaign, IL 61820, USA
Received 10 July 1998

Abstract

Liouville's fractional integration is used to define polygamma functions ~,(")(z) for negative integer n. It is shown that
such ~k(n)(z) can be represented in a closed form by means of the first derivatives of the Hurwitz Zeta function. Relations
to the Barnes G-function and generalized Glaisher's constants are also discussed. (~) 1998 Elsevier Science B.V. All rights
reserved.
Keywords." Polygamma functions; Hurwitz Zeta function; Barnes G-function

1. Introduction

The idea to define the polygamma function @(")(z) for every complex v via Liouville's fractional
integration operator is quite natural and was around for a while (see [10, 13]). However, for arbitrary
negative integer v the closed form of ~k(")(z) was not developed yet - the only two particular cases
v - - - - 2 and v - - - 3 have been studied (see [9]). It is the purpose of this note to consider

@(-n)(z)- (n - 1 2)~ ( z - t ) "-2 logF(t)dt, N(z) > 0 (1)

when n is an arbitrary positive integer, and present ~(-")(z) in terms of the Bemoulli numbers and
polynomials, the harmonic numbers and first derivatives of the Zeta function. Our approach is based
on the following series representation of log F(1 + z):
,
log F(1 + z) = (1 - 7)z - ~ \]-L--~_
z j + ~ log sin(ltz)
~ z2k+1
+ 2k +_l~_ (1 - ((2k + 1)). (2)
k=l
* E-maih [email protected].

0377-0427/98/S-see front matter (~) 1998 Elsevier Science B.V. All rights reserved.
Plh S 0 3 7 7 - 0 4 2 7 ( 9 8 ) 0 0 1 9 2 - 7
192 KS. Adamchik / Journal of Computational and Applied Mathematics 100 (1998) 191-199

Replacing log F(1 + z ) in (1) by (2), upon inverting the order of summation and integration, we thus
observe that the essential part of this approach depends on whether or not we are able to evaluate
series involving the Riemann Zeta function. We will propose here a specific technique (for more
details see [2]) dealing with Zeta series and show that generally the latter can be expressed in terms
of derivatives of the Hurwitz function ~'(s, a) with respect to its first argument. Furthermore, we
1 5,
will show that when s is negative odd and a is rational, a = -~, ¼, ~, l 5,2 ~3 and 5 then ~'(s,a) can
be always simplified to less transcendental functions, like the polygamma function and the Riemann
Zeta function. In case of negative s we will understand the Hurwitz function, usually defined by the
series

~(s, a) = ~ (n + a) s' R ( s ) > 1, E ( a ) > O (3)

n=O

as the analytic continuation, provided by the Fourier expansion (see [11]):

~(s,a)=2(2n) s IF(1--s)y~n~-lsin 2nna+2s R(s)<O; O<a~<l, (4)

n=l

2. Series involving the Zeta function

Let us consider the general quantity

(X3

S-- ~-'~f(k)~(k ÷ 1, a), (5)

k=l

where the function f ( k ) behaves at infinity like O((-1)k/k). Replacing the Zeta function in (5) by
the integral representation
1 fo~ t s - l e - a t
~(s,a) = F(s-~ J0 -1- --- -e- ~ dt, ~(s) > 1, ~(a) > 0 (6)

and interchanging the order of summation and integration, we obtain

S = fo X F(t) -1- -e---at

- Se dt, (7)

where the function F(t) is a generating function of f ( k )

tk
F(t) --- ~ f ( k ) k~".
k=l

Thus, the problem of summation has been reduced to integration. Though, integral (7) looks terribly
complicated and hopeless for symbolic integration, the point is that we do not want to evaluate
integral (7), but reduce it again to the integral representation (6). It is easy to see that if the
generating function F(t) is a combination of the power, exponential, trigonometric or hyperbolic
functions then integral (7) is a combination of Zeta functions and their derivatives, and thus so is
 V.S. Adamchik/Journal of Computational and Applied Mathematics 100 (1998) 191-199 193

the sum (5). In other words, with this approach we are staying in the same class of functions -
sums involving the Zeta function are expressible in Zeta functions.
Next, we will provide a couple of examples demonstrating this technique. Consider
2c k2
A,(x) = Z k ~ - I (~(k) - 1)xk"
k=2

In view of
1 r °~ t s le-I
((s) - 1 = ( ( s , 2 ) - - F(s) Jo e-7-- - ~ dt

upon inverting the order of summation and integration, which can be justified by the absolute
convergence of the series and the integral involved, we find that
e'
Al(x)= et - 1 t ~=2(k+l)F(k)"

The inner sum is a combination of power series of the exponential function

k2(xt) k 1 tx e,X( ---+1).

k=2 (k + 1)F(k) tx 2 + t2x2 tx1

Now we need to substitute this into the above integral and integrate the whole expression term by
term. Unfortunately, we cannot do that since each integral does not pass the convergency test at
t = 0. To avoid this obstacle we multiply the whole expression by t ; and then integrate each tenn.
We thus obtain

A,(x) = lim;.0 ( F ( 2 + 2)~(2 + 2, 2 - x)x 2 + F(2)((2, 2 - x)

XF(2 + 1)~(2 + 1,2) --

F(2- 1)((2- 1,2-x) -}-
F(2- 1 ) ~ ( 2 - 1,2)'~ .

X X

Evaluating the limit, we finally arrive at

Al(x) = 3 7x ¢ ' ( - 1) ¢'(-1,2 -x)
2 2 +((2'2-x)x2 - - + ('(0, 2 - X) x ' (8)

where 7 is the Euler-Mascheroni constant and (' denotes the derivative of ((s,z) with respect to the
first parameter. As we will see later, for some rational x the sum A~(x) can be further simplified.
For example, if x = ¼, then

A, = ~ + 2 5 ~216 87 + G- ('(-1) +log~64x/_~]

where G is Catalan's constant. If x = - ½ , then

A,- =]--~+~-~+g+log\ 64x/~ ,]+3-64 +4('(-1).

194 KS. Adamchikl Journal of Computational and Applied Mathematics 100 (1998) 191-199

All these bring us to another interesting topic: for what values of x the above expression (8) can
be simplified to less transcendental functions? It is well known that
=

~'(-1)= ~ - logA,

~'(0,x) = l o g \ x / ~ ) ,

where A is Glaisher's constant (see [7, 8]) (also known as the Glaisher-Kinkelin constant). But
what is ~ ' ( - 1 , x ) ? Or more general ( ' ( - 2 n - 1,x), n = 0 , 1,2 .... ?

3. Derivatives of the Hurwitz Zeta function

From Lerch's transformation formula (see [4]):

~O(z,s, v) = iz-"(27t)S-lF(1 - s)
×(e_l/2~iS~(e_2~ir, l - log(z) ~ eni((s/2)+2v)~ (e 2rtiv, 1 -- s, 1 log(z))
s, -T -i ] - 2rti ] )

putting v = 0, s = 1 - s and z = e 2~x it follows that

rfis/2 2~ s
~(s, 1 -- x) + e~i~(s,x)-- e ( ) Lil_s(e2~,~) '
r(s)
where we assume that 0 < x < 1 and s is real. Differentiating this functional equation with respect
to s, setting s to - n , where n is a positive integer, we obtain

Proposition 1. Let n be a positive integer and 0 < x < 1, then

~'(-n,x)+(-1)"~'(-n, 1 -x)=~i B"+l(x-----~)-t-e -nin/2 n] 1-. :e2rfixx (9)

n+ 1 ( ~ ) nLln÷It )'

where B,(x) are Bernoulli polynomials, and Li,(x) is the polylogarithm function.

Taking into account the multiplication property of the Zeta function

k--I

i=0
and Proposition 1, we easily derive the following representations:
log 12 rt ~9'(5)
(' -1, -- 144 12~ + 8v/-3~--~+ ('(-1),

( 6) 13 log2 710g3 rt 0 tit(5)

1 ~16
~' --3, -- 25 920 51 84-----6+ 144-----~ 384X/3~ 3 + ~'(--3),
 V.S. AdamchiklJournal of Computationaland Applied Mathematics 100 (1998) 191-199 195

(' -1, - 4re 8 ('(-1)'

( ~) log2 rc ~bttt(~)
l 78 fit(__3) '
(' -3, - 256~ + 256 2048rc 3 1
( ~ ) log3 rc Ip'(½)
(' - 1 , = 72 18v~ + 1 2 v ~ n ½('(-1),

( ~ ) = _ _log3 x _ _
ip'"(½) 13 , 3 ).
~' - 3 , 6480 + 162x/3 432x/~rc 3
~ 5,
Similar formulas can be obtained for ( ' ( - n , x ) when n is odd and x = g, ~, ~ 3,
~ 3,
2 4
3 and g.
5 For
additional formulas of this kind refer to papers [1, 12].

4. Negapolygammas

In the second section dealing with zeta sums we mentioned Glaisher's constant A. First this
transcendent was studied by Glaisher (see [8]). He found the following integral representation
log(25rc6) 2 fl/2
logA - / log F(z) dz.
36 + 3 a0
Let us consider a more general integral

~q log F(z) dz (10)

and show that

(1 q)q q
foqlogF(z)dz + log(2x) - ( ' ( - 1 ) + ( ' ( - 1 q). (11)

The proof is based on the series representation (2). Integrating each term of it with respect to z and
taking into account the identity
1 - ~(2k + 1)
q2k+2 = (~ - 2)q2 ~ 2 ~ 1 ~~ ~ 1~ 2 q q~ ~ ~ l ~ 2 ~ q~
~=~ ( 2 k + 1 ) ( k + 1)
(that can be easily deduced by using the idea described in the second section), we prove (11).
Formula (11) first was obtained by B. Gosper [9]. Integral (10) can be envisaged from another
point of view. It is known that the polygamma function is defined by
~n+l
~(")(z) = 0 - ~ log F(z) (12)

for positive integer n. However, using Liouville's fractional integration and differentiation operator
we can extend the above definition for negative integer n. Thus, for n = - 1 and n = - 2 it follows
immediately that
¢(-1)(z) = log F(z)
196 V.S. Adamchik/ Journal of Computational and Applied Mathematics 100 (1998) 191-199

and

q,(-2)(z) =/'Jo z log F(t) dt,

respectively. This means that the integral (10) is actually a "negapolygamma" of the second order
(the term was proposed by B. Gosper [9]). Generally, if we agree on that the bottom limit of
integration is zero, we can define polygammas of the negative order as it follows:

1 (z - t) "-2 logF(t)dt, N(z) > 0. (13)

O(-")(z)- (n - 2)!

As a matter of fact, using the series representation (2) for log F( 1 +z), integral (13) can be evaluated
in a closed form.

Proposition 2. Let n be a positive integer and R(z) > O, then

n
n!~k~-")(z) -- ~ log(2n)z "-l - B,(z)H,_j + n~'(1 - n, z)

-- ~ ~ ' ( - - i ) ( n - i)zn-i-I"~ ~ B2iO2i_lzn 2i,

i=l i=1

where B, and B,(z) are Bernoul6 numbers and polynomials, and H, are harmonic numbers.

Here are some particular cases:

~:_2)(z ) _ (1 - z)z z
2 + ~ log(2rt) - ~ ' ( - 1 ) ÷ ~'(-1,z),

:
O(-3)(z)=-~-~(6z -9z+ 1)+ - ~'(-2)-z~'(-1)+ ~'(-2,z).

More formulas:

@(-3)(1) = logA + I log(2rt),

1~(--3)(½) = ½10gA + ~6 l°g(Zr0 - 7 ( ' ( - 2 ) ,

4/-3)(½) + q,( - 3 ) (~)

2
= logA + ~ log(2rt) - ~ ' ( - 2 ) .

If we integrate both sides of Eq. (14) with respect to z from 0 to z, we obtain the following
recurrence relation for ~'(-n,z).

Corollary 1. Let n be a positive inteyer and R(z) > O, then

n /o z ~'(1 - n , x ) d x = B"+ln ( n + 1) -('(-n)+('(-n,z). (15)

 V.S. Adamchik/Journal of Computationaland Applied Mathematics 100 (1998) 191-199 197

4.1. Integrals with polygamma funetions

From definition (12), using simple integration by parts, we can express the integral

f x"~(x) dx

in terms of negapolygammas. We have

~,{-Z)(z ) = z@{-l )(z ) - ]o"~x~k(x) dx,

-
Z2 +
1 fO z x2 (x)dx
and more generally,

fo=xn¢(x)dx=(-1)nn[z"~
~ .'--l. ~k'bl~-"-')(z)Zk
. . . k[" (16)
k=0

Thus, taking into account representation (14) of negapolygammas, we obtain

Proposition 3. Let n be a nonnegative integer and ~(z ) > O, then

--1 n
~Zxnl~(x)dx =(-1)n-l~'(-n) q- ~ l Bn+lHn

- ~--~(-1) k ~--~Bk+l(z)Hk + ~-~ ( - 1 ) k zn-k~'(-k,z). (17)

k=0 k=0

4.2. Barnes G-function

Choi et al. [6] considered a class of series involving the Zeta function that can be evaluated by
means of the double Gamma function G (see [3]) and their integrals. If we apply our technique
described in the second section to those sums we get results in terms of the Hurwitz functions. To
compare both approaches we need to establish a connection between the Barnes G-function and the
derivatives of the Hurwitz function. The G-function and ~' are related to each other by

log G(z + 1) - z log F(z) = [ ' ( - 1 ) - [ ' ( - 1 , z ) . (18)

The identity pops up immediately from Alexeiewsky's theorem (see [3]) and formula (11). Integrat-
ing both sides of (18) with respect to z, in view of formulas (14) and (15), we obtain the following
(presumably new) representation

~log 1 dx - Z ( 1 - 2z 2) z 2
G(x + )
f 12 + ~- log(2rc)
+ z ( [ ' ( - 1 ) + ~ ' ( - 1, z)) + [ ' ( - 2 ) - ~'(-2, z). (19)
198 KS. Adamchik l Journal of Computational and Applied Mathematics 100 (1998) 191-199

5. Generalized Glaisher's constants

In 1933 Bendersky [5] considered the limit

/
logAk= lim [
n ---+O~ \ m=l~mk
l°gm--p(n'k)) ' (20)

where
1"/k nk+J ( 1 )
p ( n , k ) = -~- logn + k--~--]- logn k+l
_ [ k 1
k n k JBj+l [l°gn + (1 - 6~_~) Z k i+ 1
+ k ! ~-, ( j ~_ ij~.-.-.-~--j) t
j=l i=1

and 6~ is the Kronecker symbol. He found that

logA0 = 1 log(2rt).

and

logA1 = ~ - if'(-1) = logA

and for the next three values he gave their numerical approximations. However, it turns out that all
A~ can be expressed in terms of derivatives of the Zeta function, by using the asymptotic expansion
of the Hurwitz Zeta function (see [11]):

c~l_z _~ m--1 B2~ F ( z + 2 j - 1 ) _2j_;+ 1 + O ( _2m .~-1) (21)

{(z, ~) - z - 1 + 2-- + ~j = l (2j)~ F(z)

when I~1---~c~ and [arg ~1 < n. Differentiating (21) with respect to z and setting z to -1 and -2,
for example, we have

1 ~2 (1 ~ ~) (~)
('(-1'~)-12 4 +logs 1~-~+ +O (22)

and

('(-2,a)- 12 9 +logc~ --~-+ +O . (23)

Now, taking into account the analytical property of the Hurwitz function, the sum in (20) is

~ '~m k l o g m = ~ ' ( - k , n + 1) - ~ ' ( - k ) .

m=l
 KS. Adamchik/Journal of Computational and Applied Mathematics 100 (1998) 191-199 199

Therefore, applying asymptotic expansions of the derivatives of the Hurwitz functions to (20), we
find that

logA2 = - ~ ' ( - 2 ) ,
11
logA3 - ~'(-3),
720
logA4 = - ~ ' ( - 4 ) .

Generally,

Proposition 4. Let k be a nonnegative &teger, then the 9eneralized Glaisher constants Ak are of
the form

Bk+,Hk ~'(-k), (24)

logAk -- k + 1

where B. are Bernoulli numbers and H. are harmonic numbers.

References
[1] V.S. Adamchik, A class of logarithmic integrals, Proc. ISSAC'97, 1997, pp. l-8.
[2] V.S. Adamchik, H.M. Srivastava, Some series of the zeta and related functions, Analysis 31 (1998) 131-144.
[3] E.W. Barnes, The theory of G-function, Quart. J. Math. 31 (1899) 264-314.
[4] H. Bateman, A. Erdelyi, Higher Transcendental Functions, vol. 1, McGraw-Hill, New York, 1953.
[5] L. Bendersky, Sur la function gamma gbnrralisre, Acta Math. 61 (1933) 263-322.
[6] J. Choi, H.M. Srivastava, J.R. Quine, Some series involving the zeta function, Bull. Austral. Math. Soc. 5l (1995)
383 -393.
[7] S. Finch, Glaisher-Kinkelin constant, in HTML essay at URL, www.mathsoft.com/asolve/zconstant/glshkn/glshkn.
html, 1996.
[8] J.W.L. Glaisher, On a numerical continued product, Messenger Math. 6 (1877) 71-76.
m/6
[9] R.W. Gosper, fl/4 log F(z)dz, In special functions, q-series and related topics, Amer. Math. Soc. 14 (1997).
[10] N. Grossman, Polygamma functions of arbitrary order, SIAM J. Math. Anal. 7 (1976) 366-372.
[11] W. Magnus, F. Oberhettinger, R.P. Soni, Formulas and Theorems for the Special Functions of Mathematical Physics,
Springer, Berlin, 1966.
[12] J. Miller, V.S. Adamchik, Derivatives of the Hurwitz Zeta function for rational arguments, J. Comput. Appl. Math.,
to appear.
[13] B. Ross, Problem 6002, Amer. Math. Monthly 81 (1974) 1121.