# NOTES

Edited by:

John Duncan

Geometric Properties of the Gamma Function
Patrick Ahem and Walter Rudin
The fact that log r(x) is convex on the positive real axis is one of the crucial properties of f. Combined with the functional equation f(x + 1) xr(x) and the normalization f (1) 1, it uniquely characterizes the gamma function. This theo­ rem, due to Bohr and Mollerup, is the basis of Artin's elegant treatment of r(x) in [2]. We extend this convexity property into the complex plane, use this extension to obtain information about the argument of r(z) on vertical lines, and describe some features of the conformal mappings induced by f' /f and by log f. In spite of the apparently inexhaustible supply of formulas, identities, expansions, and integrals concerned with f, we have found no mention of such geometric proper­ ties in the literature. For brevity, we use the notation
= =

G(z)

=

log

f(z). z having Re z > a,

We let na denote the right half-plane consisting of all complex and denote its closure by na'

Theorem 1. (A) If x � � then Re G" (x + iy) > 0 for all real y. (B) If x < � then Re G" (x + iy) < 0 for all sufficiently large y.

Note that Re G" (x + iy) (a2 /ax2)1og Ir(x + iy)l. The theorem asserts there­ fore that log Ir(x + iy)1 is a convex function of x in n 1/ , but in no larger half-plane.
=

2

Proof: We begin with (B) since its proof gave us a hint that 1/2 might be the cut-off between (A) and (B). Define I/I(s) � - s on (0, 1], I/I(s + 1) I/I(s) for all real s. One way of writing Stirling's formula is [4; p. 151]
= =

log

f(z)

=

(z - �) log z - z + � log 27T +

1

00

for all z "" 0 that are not negative real numbers. Differentiate this twice, then perform an integration by parts. The result is

o s +z

--

I/I(s)

dx

(1)

G"(z)
where
678
cp (s)
=

=

-

z

1

+

-

1 cp( s) +6r ds ' J (s + Z)4 2Z2 o
00

(2)

JoI/I(t) dt. Since 1/1 has mean-value 0,
NOTES

cp

is periodic, hence bounded. [October

In fact, 0

::s:

�(s) ::s: 1/8. It follows that
::s:

. R e G" ( x + zy)

ds 3 (XJ 2x3 + x2 + ( 2 x - 1)y2 + -), 2 4 0 (s + X)2 + y2 2· 2 ( x2 + y2)

[

]

The last integral is O(y-3), as y /' 00. When 2 x - 1 =I=- 0, the dominant term, for fixed x and large y, is thus (x - )y-2, and this is negative when x < 1/2; (B) follows. Next, we apply log to the identity [2; p. 24]

f(z)f(l - z)
then differentiate twice, and obtain

=

sm( ",z)
.

",

'

G"(z) + G"(l - z)
Since sin

=

",2 sm2( ",z)
.

.

",(�

+ iy)

=

cos(",iy)

=

cosh(",y), it follows that

or

2 ReG" +iy) =

U

",2 > 0. COSh2( "'y)

(3)

Finally, (2) shows that G" is bounded in 0.8, for every 8 > o. Since bounded harmonic functions in a half-plane are the Poisson integrals of their boundary values, (A) follows from (3). The following monotonicity property was needed in [1], but only when b - a is an integer, and in that case a more direct proof exists.
Theorem 2. (i) If

� ::s: a

<

b then
arg

f(b f( a

+ iy) + iy)

is an increasing function of y on ( - 00, 00). (ii) The same conclusion holds if 0 < a <

� and b > 1 -a.
=

Proof: Let u and
Then
v
=

v be the real and imaginary parts of G log f, respectively. l!!.g f, the Cauchy-Riemann equations give Ux vy' and hence vxy uxx > 0 in 0.1/2, by Theorem 1. This means that v a + iy) < v b + iy), or

/

=

/

=

a - arg ay This proves <D.

f(a

+ iy) < - arg f(b + iy).
ay

a

1996]

NOTES

679

To deduce (iO, note that

f(b+iy) f(a+iy)

f(b+iy) f(1-a+iy) 1

f(l-a+iy) f(a -iy) f(a+iy) f(a -iy) f(b+iy) 7T . f(1 -a+iy) sin 7T(a -iy)

(4)

Since now 1 -a > 1/2, (iO follows from (i) and the fact that the argument of the last factor in (4) is which is an increasing function of y when 0 < a < 1/2. The proofs of the next two theorems will use the following sufficiency criterion for univalence. arctan [ cot( 7Ta) tanh( 7TY) ] ,

Suppose: (a) n is an open half-plane that does not contain the origin, (b) f is holomorphic in a convex region 0, and (c) f '(z) E n for every z E O. Then f is univalent in O.
When n = 00 this is proved on p. 47 of [3]. It is clear that the criterion is rotation-invariant.
Theorem 3. (i) f'If is univalent in 00, but in no larger half-plane. (iO Re(f'/f)(x+iy) � (f'/f)(x) in 00, (iii) Im(f' If) is bounded in Oil' for every {) > O. (iv) IIm(f'/f)(z)1 < 7T/2 in 01/2'

Proof: Since (fl If)' = Gil, it follows from Theorem 1 and the sufficiency criterion that f'If is univalent in 01/2, This proves only part of (0, but at least it points in

the right direction. Differentiation of (f'/f)(x+iy) = (u+iv)(x+iy) (where u and v are now the real and imaginary parts of f'/f) with respect to y gives

iG"(x+iy) = (u y+ivy)(x+iy). 7T2 vit+iy) = h2( 7Ty) . 2 cos v(t +iy) = '2 tanh( 7TY),

Thus Vy = Re Gil, hence (3) becomes

If we integrate this we obtain

7T

and this implies (iv). Differentiation of the logarithm of

[4; p. 150], where

1 __ =ze'YZ f(z)
'Y -

n ( 1+ ':' ) e-z/II n
1

'Y

u(x+iy) = v(x+iy) = 680

-

L
o

00

+ x2+y2 y (n+x)2+y 2'

x

E
1

(2. n

n+x ' (n+X)2+y2

)

(5) (6)

NOTES

[October

and

00 =E G"(z)
o

' (n +z)2

1

(7)

The right side of (5) is an increasing function of y 2, if x > O. For fixed x it therefore attains its minimum when y = O. This proves (ii), and (iii) follows from (6), because

v(x

+

iy) <

+E�--::, 1 n 2+ y 2 x 2+ y 2

y

00

y

< -+ (

1 2x

1 7T dt =- +- ' 2 2x 10 t 2+ y2 y

00

For the proof of (i), let rr+ = {y > O} and rr- = {y < O} be the upper and lower half-planes, respectively. If z E no n rr+ then (n + Z)2 E rr+ for all n � 0, hence (n + Z)-2 E rr-, so that G"(z) E rr-. Likewise, G"(z) E rr+ if z E no n rr-. It follows that f'/r is univalent in each of the quadrants n o n rr+ and no rr-. By (6), f'/r maps no n rr+ into rr+ and no n rr- into rr-; since f'/r is strictly increasing on the positive real axis, we conclude that f'/r is univalent in all of no . Finally, f'/r has a pole at z = 0, hence maps each neighborhood of 0 onto a neighborhood of 00. Since (f'/n( x ) /' 00 as x /' 00, it follows that f'/r is not univalent in nil if {) < O. In the next theorem, Xo is the unique positive number at which f'(x o) = O. Since r(n = r(2), 1 < Xo < 2.
Theorem 4. log r is univalent in

nxo' but in no larger half-plane.

Proo!' Recall that (log n' = f'Jr. If x > xo, it follows from part (iO of Theorem 3
that Re( log r)'(x + iy)

(f'/r)(x) > (f'/r)(xo) = o.

Thus log r is univalent in nxo' Clearly, Xo has no neighborhood in which log r is univalent, since (log n'(xo) = o.
REFERENCES 1. 2. 3. 4. P. Ahem, M. Flores, and W. Rudin, An invariant mean-value property, 1.

Functional Anal.

1993, 380-397. E. Artin, E. P. L. Duren, R.

111,

5.

The Gamma Function, Holt, Rinehart and Winston, 1964. Univalent Functions, Springer-Verlag, 1983. C. Titchmarsh, The Theory of Functions, 2nd ed., Oxford Univ. Press, 1939. Remmert, Wie1andt's theorem about the f-function, Amer. Math. Monthly,

103, 1996, 214-220.