Professional Documents
Culture Documents
Dusan V. Slavic
for the logarithm of the gamma function, where z=x+iy, a is a real number.
Bk (a) are BERNOULLI'S polynomials and
This relation differs from the known ones, since it contains the additional
term -log (l-bei21'tc(z+a», without which one must suppose that larg (z + a)l::;;;
::£;7t-a, o<a::;;;7t. In the neighbourhood of singular points, this additional term
is dominant.
We also give the corresponding additional terms for the asymptotic
relations for r (z + a), tjJ(z + a) and tjJ' (z + a).
as the definition of the gamma function. For Re z::;;;0, we take the values
which are obtained by analytical continuation of the gamma function using
the functional equation
(1) T(z+ 1)=zT(z)
except in simple poles for z = p (p = 0, - 1, - 2, ...).
. Presented October 19, 1971 by D. D. ADAMOVIC.
73
74 Supplementing the asymptotic expansion of the gamma function. . .
I
" 1 k+- 2
k+-
~ exp( -~-+ 7
(2) k!,,42r.
[ - +.. .
e J
24(k +
l ~)
2880( k +
~ r J
see [1], p. 180, as well as [2], p. 341.
Later on, DE MOIVREgave the relation
+00
1 B 2n
(3) 10gF(z)'" (
z--
2 ) logz-z+logV27t+ 2:
n~12n (2n-l) z2n-t
defined by
+00
z Bn n
--L..,-z, - "" Izl<27t.
ez-l n~1 n!
~(s)= y
n=1
{ nS
~- T(n-S)
T(n) }
(Res>O),
see [4], p. 56, it can be shown that the modulus of the general term in (3)
for n- + 00, behaves as
IB2 n I -,...",(~ - log (1t e2 n» 2n-1 1
2n(2n-I)lzI2n-1 \1te 41te ) [z!2n-1 '
i. e., that it tends to infinity for a fixed z, which implies the divergence of (3)
for all z. Naturally, this does not diminish the importance of (3), but it requ-
ests further specifications. In virtue of (4) it is easy to show that the modu-
lus of the (n + 1)-th term of (3) is less than the modulus of the n-th term if
n-l 1
Izl>-+ .
1t" "-
V30
D. V. Slavic 75
(5) LOgr(z)=
{(
z-~
2 ) 10gZ-Z+IOgV27t+ ~
n=12n
B2n
(2n-l) z2n-1 }
+RN(Z),
(o~largzl~:)
7t 7t
For - - ~ arg z ~ - the modulus of the remainder is less than the mo-
4 4
dulus of the fLst omitted term
largz!~7t-a, o<a~7t.
76 Supplementing the asymptotic expansion of the gamma function. .
"
and (6) implies that RN (z) does not tend to zero when z~ 00.
The expression in the curved brackets of the relation (5) is a bounded
function for finite z. Since LogF(z) for z=p (p=O, -1, -2, ...) has loga-
rithmic singularities, it is clear that RN(Z) does not tend to zero for z~oo;
consult [9], p. 29.
(-l)p
Since the gamma function has for z = p simple poles with residues
p!
(see, for example, [10], p. 18), the real part of the logari~hm of the gamma
functions in the neighbourhood of z = p has the following behaviour
(11) Re Log F(z)"" -log 1
z - pl.
Th:::refore, RN (z) must take over th::: role of th~ term which expresses the
logarithmic singulariLies. We shall now determine the asymptotic expression of
function F (z) such th:1t RN (z) = F (z) + 0 (Z-2N-1) since only after its compu-
tation we obtain the relation
N
1
(12) LogF(z)=
( 2) z-- logz-z+logV27t - +F(z) + L B 2n
z2n-l
+O(Z-2N-1)
n=12n(2n-l)
(16)
D. V. Slavic 77
The eventual additional term F (z) cannot be, therefore, determined from (16).
This implies
(17) F(2Z)=F(z)+F(z+ ~)+0(Z-2N-l)
compare [12], pp. 233-234.
Combining (1), (6), (7) and (16) we obtain
N (l_2-(2n-I»)B
(18) Logr
( )
z+-
1
2
=zlogz-z+log V-27t- 2:
n=12n(2n-1)z2n-1
2n
+F(Z+ ~)+O(Z-2N-l),
where F(Z+ is up to now unknown term which satisfies (13), (14), (15)
~)
and (17).
Expansion (18), together with (13), for z = k + J... contains STIRLING'S
2
formula (2).
We shaH now evaluate the asymtotic value of F (z + for Rez<O,
~)
1m Z 6; 0, when z-+ 00. Replacing z by - z from (18) it follows
(19) -Logr(~ -z)=zlogz+7tZi-z-1ogV27t -F(~ -z)
N (1 2-(2n-I»)B
- 2n 1t
-<argz~7t.
- 2: 2n (2n-1)z2n-1 + 0 (Z-2N-l), 2
n~1
Equality 2 cos 7tZ = einz + e-inz implies
The obtained asymptotic values (23) are in accordance with (14), (15)
and (17).
We therefore have for z_oo
N B
+ 2: 'n~ + 0 (Z-ZN-l),
n=12n(2n-l)z2f1-1
where z and c have the SE-me meaning, as b = 1 (x< -}) and b = 0 x ~ +).
(
E. W. BARNESgave i:l 1899 (see, for txemple, [13], p. 63) the following
asymptotic relation, which is more general than (10)
S=-I)~EkC~
(26)
Logr(z+
(
a)= z+ a- --'-
2 )log z- z+ log V27t + ~
k=2k(k-l)zk-l
+ 0 (z-m),
larg(z+a)I~7t-a, 0<a~7t.
where Bk(a) BERNOULLI'S polynomials
zeaz + 00 zk
--=
eZ-1
2: -Bda),
k~ok!
Izl<2r..
(27) + ~ (-I):!k~
k=2k(k-l)zk-'
+ 0 (z-m),
For a = 0 relation (27) contains (25), since B2n+l (0) = 0, B2n (0) = B2n.
Using the known relations B2n+l(i)=0, B2n(~)= _(1-2-(2n-l))B2nwe
I.
prove t h at (27) for a = - contz.lllS (24).
2
Notice that the expressions on the regiht hand sides of asymptotic rela-
tions (26) and (27) c::~nhave arbitrary imaginary part, whose modulus is gre-
ater than 7t. It is therefore justifia.ble to write the left hand sides of those
relations as Log r (z + a), or to add to the right hand sides 2/7t i (/ a conve-
nient integer). One should have in mind that formulas (26) and (27) give
asymptotic valaes of the logarithm of the gamma function Log r (z + a) on
one sheet of the RIEMANN surf,ce of that function, actually on the one for
which arg (z + a) = 0 ha.s 1m Log r(z + a) = O.
The additional term for x< ~ and Iy 1-+ + 00 has the asymptotic btha-
2
vlOur
-log (1 - bei2ncz), ,e-2nIY I(cos 27tx + i sin 27tx),
(28)
Iarg z I ;:;;; 7t - ~,
*
D. S. MITRINOVIC,P. M. VASIC, D. D. ADAMOVIC,J. D. KECKIC, R. E. SHAFER,
M. S. KLAMKIN,J. FIELDSand Y. L. LUKE have read this article in manuscript and have
made some valuable remalks and sugg~stions.
REFERENCES
Katedra za matematiku
E1ektrotehnicki fakultet
11000 Beograd, Jugoslavija