arXiv:1409.2971v1 [math.

CV] 10 Sep 2014

A note on the zeros of the Digamma function

and the derivative of the log-Barnes function
Istvan Mezo 1
Department of Mathematics,
Nanjing University of Information Science and Technology,
No.219 Ningliu Rd, Pukou, Nanjing, Jiangsu, P. R. China

Abstract
Little is known about the zeros of the Digamma function. Establishing some Weierstrassian infinite product representations for a given regularization of the Digamma
function we find interesting sums of its zeros. In addition, we study the same questions for the zeros of the logarithmic derivative of the Barnes G function.
Key words: Euler Gamma function; Digamma function; Barnes G function; zeros;
Weierstrass Product Theorem
1991 MSC: 33B15, 30C15, 30D99

Introduction

The Euler gamma function is defined by the improper integral

(z) =

ez tz1 dt ((z) > 0).

The logarithmic derivative of is the Digamma function

(z)
(z) =
(z)

(z C \ {0, 1, 2, . . . }).

This function is analytic everywhere except the non positive integers, and it has first order
poles in these points. It is known that the Digamma function has only real and simple
zeros, and all of them except only one are negative.
Email address: istvanmezo81@gmail.com (Istv
an Mezo).
The research of Istv
an Mezo was supported by the Scientific Research Foundation of Nanjing
University of Information Science & Technology
1

The Barnes G function [2] is a higher order analogue of the function satisfying the
functional equation
G(z + 1) = (z)G(z)
with the normalization G(1) = 1.
The derivative of log G does not give us a very different function, because
G (z) =

G (z)
log(2) + 1
=
z + (z 1)(z) (z C \ {0, 1, 2, . . . }).
G(z)
2

This function, similarly to the Digamma function, has first order poles in the non positive
integers and has real zeros of multiplicity one, two of the zeros are positive.
In this paper we study a regularization of the and G functions which gives us the
possibility to find interesting closed form relations for some infinite sums of the zeros of
these functions.

Zeros of the Digamma function

As we mentioned, the function has only real roots, and has first order poles at the non
positive integers. The zeros are of multiplicity one. Canceling the poles somehow, we can
get a Weierstrass product representation which is so useful many times. Especially, we
can use it to find various infinite sums for the zeros. What we have found is contained in
the following theorem.
Theorem 1 For all z C we have that

Y
z
z
(z)
e k .
= e2z
1
(z)
k
k=0

(1)

Here is the Euler constant, and k is the kth zero of counted from the right. So, in
special, 0 = 1.461632 . . . is the unique positive zero of .
Proof. By the properties of the poles of and it comes that is an entire function and
the set of zeros of this function is the same as that of . Moreover, it is known due to
Hermite [1, p. 259] that
!
1
.
(2)
k = k + O
log k
Hence, considering the theory of entire functions [3] we can be sure that

Y
z
z
(z)
1
e k ,
= ceaz
(z)
k
k=0

for some constants a and c, see [3, p. 26]. Since (0)

= 1 we get that c = 1. Taking
(0)
the logarithmic derivative on both sides and substituting zero again it also comes that
a = 2.
Some consequences immediately follow. Taking logarithmic derivative on both sides of (1)
we get that

X
(z)
1
(z) = 2 z
.
2
(z)
k=0 k k z
With z = 1 this becomes

2
1
=

+
.
2
6
k=0 k k

Another consequence of (1) is that

Y
(z) (z)
z2
=
1 2 ,
(z) (z) k=0
k

so, comparing the coefficient of z 2 on both sides we have

2
1
2
=

+
.
2

2
k
k=0

The logarithmic derivative of the penultimate formula shows that

X
1
1
1 (z) 1 (z)
+ cot(z) +

= 2z
,
2
2
z
(z)
(z)
k=0 k z

so, in special,
1

2
=
+
1.
2
2 12
k=0 k 1

One more series identity we can deduce with small effort. Letting
F (z) =

(z) (z)
,
(z) (z)

it follows that
F (z)F (iz) =

k=0
4

(3)

z4
1 4 .
k
!

Comparing the coefficients of z we arrive at the following identity.

1
2 2 2 4
4
=

+
+
+ 4(3).
4
3
9
k=0 k

Worth to note that all of these series converge slowly.

Zeros of the G function

Since the zeros of the G function behave similarly to the zeros of the function (see the
next subsection) we can repeat all the above arguments to have that
!

Y
z
1
1
z
G (z)
1
e k ,
= e(2+ 2 log 2 2 )z
(z)
k
k=0

where k is the kth zero of G on the real line such that

0 = 2.55766 . . . ,

1 = 1.39147 . . . ,

2 = 0.3662934,

and so on.
It follows from the Weierstrass product that
1
9 2
1
1
=
+
+ (1 + + log 2) log 2 + log2 2.
2
4
2
2
4
k=0 k

Similarly as above (see (3)), we can define the FG function to calculate the sum of the
reciprocals of k4 . This is a bit cumbersome, but the result is

4 1 2  2
1
4
=
2(3)(L
+
1)
+

+
L
+
3
+ 2 3 (L + 1)+
+

4
9
6
k=0 k

i
1 h
24(3) + 3 + 4 2 (L + 1) + 3L(L2 L + 7) +
6
h
i
i
1 2
1 h 4
L 4L3 + 22L2 36L + 49 ,
9L2 + 6L + 4 2 + 21 +
6
16
where L = log 2.

3.1 The asymptotic behavior of the zeros of G

As we already noted
 (2)), Hermite knew that for the roots of the function behaves
 (see
1
like k = k + O log k . This can be strengthened taking into account that
(1 x) (x) =

,
tan(x)

substituting x = k and then taking the asymptotic approximation

1
1
+O 2
(x) = log x
2x
x


for large x. This argument results that

  .
k k + arctan
1

log k 2k + O k12

More terms from the approximation of give more precise expression. As k grows, the
argument of arctan is close to zero. Close to the origin arctan(x) x, therefore Hermites estimation follows. (We note that the idea comes from the WikiPedia article on
the Digamma function but there is no reference there).
Such a trick can be applied for G , mutatis mutandis. The functional equation for G is
G (1 x) + G (x) = log 2

x
(x).
tan(x)

Substituting x = k and supposing that (k ) 0 (or at least converges to zero as k

which is, in fact, true) we can apply the asymptotic [2, p. 40]
1
1
G (1 + x) = log 2 x + x log x + O
2
x

 

to have that
k+2
In Hermitean form:

(x )

1
  .
= k + arctan

log k 1 log2k2 + O k12

k+2 = k +

 .
log k 1 log2k2 + O k12
(We need to shift the indices of by 2 to match the real location, 0 and 1 are positive.)

This shows that the zeros of G converge a bit more slowly to the bounding left integer
than k does. For example,
10 9.702672541,

100 99.80953650,

1000 999.8641415,

11 9.622785495,

101 99.77177415,

1001 999.8444267.

while

References
[1] M. Abramowitz, I. A. Stegun, Handbook of Mathematical Functions with Formulas, Graphs,
and Mathematical Tables, Dover, 1972.
[2] H. M. Srivastava, J. Choi, Zeta and q-Zeta functions and associated series and integrals,
Elsevier, 2012.
[3] B. Ya. Levin, Lectures on Entire Functions, American Mathematical Society, 1996 reprint.