2.

7 Picard Theorems for Functions of Finite Order 31

Jl(z) has the same number of zeros in the circle asf(z). This follows from
Rouchts theorem because f(z) and f,,(z) -f(z) are analytic in and on a
closed contour C in a region D and If,(z) -f(z) I < If(z) I. Thus if
f(zO) = 0, then f,,(z) has exactly one zero in C for all n 2 N so that z0 is
a limit point of zeros off,(z). Also, iff(z,) # 0, thenf,(z,,) # 0 in C. 0

2.7 The next group of theorems although elementary in nature, examine

more carefully the relation between order and the number of roots of an
equation of the form f(z) = AP(z) where f(z) is entire, A is a constant,
and P(z) is a nonzero polynomial.
The only deficiency in the theorems is that they do not deal with functions
of infinite order, a ramification which will be dealt with later on in the
survey.

2.7.1 Picard Theorems. It will be useful to establish first of all, a few lem-
mas concerning inequalities for coefficients of power series.
Consider the entire functionf(z) = a, a,z + + +
. . . anzn . . Write + -
a,,= an + iBn
and
z = r(cos 6 + i sin 0), r 2 0,
Then
+ i,Bn)rn(cos8 + i sin 0 ) n .
m

f(z) = C (orn
n=o

If we write
f(z> = U(r, 0) + Wr, 0
then
+C
m

U(r, 0) = a0 (or, cos n0 - B,, sin nO)rn. (1)

n=1

Fixing r and integrating with respect to 0 from 0 to 2n, we have

Llg = -
1 J 2 n U(r, 0) do.
2n 0

To compute orpr p 2 1 multiply ( I ) by cosp0 and integrate with respect

to 0 from 0 to 2n. Thus

U(r, 0) cos p6 d8 = aprp

32 11. The Expansion of Functions and Picard Theorems

from which

a =-j
nrp
2n V ( r , e) COspe dB p = 1,2, ..

and similarly

Pp =-
nrp
j2nV(r,0 ) sinpBd0, p = 1, 2, . . . .

Also,

2a0 fuprp = -
n o
12n
U ( r , e)(i f cospe) dB

and

2a0
n o
j
Bprp = - 2n V ( r , e ) ( l f sinp0) do.

Note that the factors multiplying U(r, 0 ) are now nonnegative.

Let p ( r ) = maxosB52nU(r, 0 ) on the circle of radius r. Then

2u, f uprp 5 -
’(‘)
n o
(1 j2’ f cos pe) d0 = 2p (r).

2.7.2 Lemma. If U(r, 0 ) the real part of an entire function f ( z ) satisfies

U(r, 0 ) 5 p ( r ) 5 Crs, S > 0,

for all r > N , then f(z) is a polynomial of degree not exceeding n = [S],
i.e., the integer part of 6.
Proof. Since I up I and I /-II,5 2 { p ( r ) - uo}/rP,we have that
2 (Cr6 - uo) 2(Crs - uo)
1% I 5 and 5

-
,p IBPI rp

+
If p > [a], since p is an integer, p 2 [S] 1 > 6 hence u p , PP 0,
+
r - + o o , and up iBP = ap = 0 for p > [d] = n. 0
2.7 Picard Theorems for Functions of Finite Order 33

2.7.3 Lemma. I f f ( z ) is a transcendental entire function and P ( z ) and Q ( z )

+
are polynomials of degree m, n, respectively and if P ( z ) 0, then the order
+
el of P ( z ) f ( z ) Q ( z ) coincides with the order e off(z), i.e.,
el = e.
Proof. Writing
M ( r ) = max I f ( z )
I2153

we confine ourselves to values of the functions at points on the circle I z I = r.

Let a,z*, b,zn denote, respectively, the leading terms in the polynomials
P(z) and Q(z). Taking E = & in the inequalities for polynomials previously
evaluated, we assert for I z I = r > r o , that

However, at z1 (on the same circle) at which 1 P(z)f(z) + Q(z) I attains

its maximum M l ( r ) , we have

Mi(r) = I P(zilf(zi) + Q(zi) I I 8 I a, I rrnM(r)+ 8 I bn I rn*

Thus

Sincef(z) is a transcendental entire function it has been proved that M ( r )

for r sufficiently large, the right-hand

-
increases faster than the maximum modulus of any polynomial and hence
faster than any power of r. Therefore each { . } + 1 as r 00 and hence
< 2 and the left hand
{ - . a }

{ } > Q , i.e.,
i I a, I rmM(r)L MI@)5 3 I a, I rnM(r).
34 11. The Expansion of Functions and Picard Theorems

We now require
- log log M(r)
lim Y
r+m log r

i.e.,

and each {. . . } -+ 1 as r + 00. Hence the right-hand { .} < 2 and the

---
9

left-hand { } > 4, and

Q log M ( r ) 5 log M,(r) I2 log M(r).

Taking logarithms again,

- log log M(r)
lim 5-
lim
log log M l ( r ) ~ lim log log M ( r )
,-+m logr ,+m logr ,-+- logr '
consequently e = el and the orders off(z) and P(z)f(z) + Q(z) are equal
[P(z)fOl. 0
2.7.4 Lemma. The order of an entire function

where P ( z ) , Q(z), and g(z) are polynomials and P(z) $0, is equal to the
degree of g(z).
Proof. From the previous lemma, the order e off(z) coincides with the or-
der of $(.) = e@). We need to show that

- log log Ml(r)

lim =n
r+.n logr

where max,,,,, 1 $(z) I = M,(r) and n is the degree of g(z). Set

and
z = r(cos 0 + i sin 0).
2.7 Picard Theorems for Functions of Finite Order 35

By hypothesis, en = 1 c, I # 0, thus

g(z) = goekrk(cos
n
ak + i sin Cr,)(cos kO + i sin kO)
=
n

Z O
Qkrk(COS(ai, + ke) -/- i S i l l ( a k $- k0))

and

and

We require now log rnaxlzI,,I $(z) I. For fixed r and 0 5 O 5 2n

hence for E in (0, 1) and r > r(E),

and

Let zo be a point on I z I = r such that cos(a,, + no) = 1 (actually there

are n such points). Then

1% Mi(r) 2 1% I d(zo) I = @ J n +
n-1
@krk cos(ffk + ke).
zo
Z O

n-1
2 enrn - @krk

1 +
- ...
en r en

> enrn(l - E ) for r > r(E).

Thus
ep(l - E) < log M l ( r ) < enrn(l + E )
and
- log log M , ( r )
lirn = n. 0
r+m logr
3.7 Theorem of Bore1 and Carathdodory 53

equality holding when g(z) is a constant Me'v/R. Therefore

I f ( r e f 0 )I 5 r M / R for r 5 R,

equality holding only when f(z) = zMeiY/R. 0

3.7 A THEOREM OF BOREL AND CARATmODORY

The following result enables us to deduce an upper bound for the modulus
of a function on I z 1 = r , from bounds for its real or imaginary parts on
a larger concentric circle I z I = R.

3.7.1 Theorem. Let f(z) be analytic for I z 1 5 R and let M ( r ) and A ( r )

denote the max I f(z) 1 and max 9{ f(z) 1 on I z I = r . Then for 0 < r < R ,

Proof: The result is clearly true forf(z) = constant. For f(z) nonconstant
suppose f(0)= 0. Then A(R) > A ( 0 ) = 0. Since by the Poisson-Jensen
formula

and if we let A(R) = maxO U(R, #), then U(R, 4) 5 A ( R ) , V4 and

Thus U ( 0 ) = A ( R ) implies U(R, 4) = A(R), V$ and A ( 0 ) 5 U ( 0 ) IA(R)

unless the function is a constant.
Let

Then #(z) is analytic for J z 1 5 R since the real part of the denominator
does not vanish. Now $(O) = 0 and iff(z) = u iv,+
54 111. Theorems Concerning the Modulus of a Function and its Zeros

Also, Schwarz's lemma gives

I #(.I I I
rlR

since I 4(z) I is bounded by 1. Thus

+
(use I R R# I 2 R - R I # I). Thus we obtain the result for f(0) = 0.
If f(0) # 0, apply the result to f ( z ) -f(O) and

since f(0) may be negative. Adding I f( 0 ) I to both sides, we have that

We also show that

since

where C is the circle

which ensures C is inside 1 z 1=R (Fig. 4). The previous theorem gives
4.1 Weierstrass Factorization Theorem 57

We must construct an entire function which vanishes at these points and

nowhere alse {n:=l(l- (z/z,)) unfortunately may diverge}. This is, in
effect, accomplished by Weierstrass's factorization theorem, which dem-
onstrates the construction and existence of such a function. We prove a
form of the theorem stated as follows:

4.1.1 Theorem. If ( z , ~ is
} an arbitrary sequence of complex numbers dif-
ferent from zero and whose sole limit point is 00 and if m is a nonnegative
integer, then 3 an entire function G ( z ) having roots at the points zl, z2 ,
. . . (and these points only) and a root of multiplicity m at the point zero.
Further, G ( z ) can be defined by the absolutely uniformly convergent prod-
uct
n
G ( z ) = eg(z)Zm
n-1
m
eQu(z/zn), where eg(z)

is an arbitrary entire function and Q,(z) is a polynomial such that

QU(z)= z 4--
2
+
22
* * a +y . ZU

The nonnegative integer v has the property that the series I / I z, lufl
C:=E"=I
converges uniformly in the whole plane.
Proof. Consider (1 - (z/zn))eQv(Z/Zn)
where Q J z ) is a polynomial of degree v.
This is an entire function which vanishes for z = z,~.Since

e)
where I z/z, I < 1,

(1 -
i Z
exp ~ , , ( z / z , )= exp - - - --
z, 2Zn2
Z'!
. . . + Q.($))

We are required to determine v such that nr=l(l is ab-

- (z/z,))eQv(*/Zn)
solutely and uniformly convergent for I z I < R and arbitrarily large R .
Choose R > 1 and u such that 0 < a < I . Then 3 a positive integer q
such that I zq I 5 R / a and I z q f l I > R/u. Thus we see that the partial
60 IV. Infinite Product Representation Order and Type

while M ( r ) > exp re-e for an infinite number of r + -+ 00. Thus

<e +E
log log M ( r ) log log M ( r )
and > @ - E
log r log r

for an infinite number of r + -t 00, hence

- log log M ( r )
e = lim
-
7+M logr

Alternatively, f(z) is of finite order, if 3 A > 0 such that as 1z I =r 00

If(z) I = 0(exP r A > .

Thusf(z) is of order e if

for every E > 0 but not for any negative E . The constant implied in 0 de-
pends in general on E , otherwise E could be replaced by zero in the formula.
A few theorems will now be proved using the above expression for the order.

4.2.1 Theorem. If f i ( z ) and J2(z)are entire functions of order el and p 2 ,

respectively, and if el < e2, then the order of F(z) = fl(z) fi(z) is equal +
to ez.
Proof: We suppose e2 is finite. Since

Wr; h +J2)IW r ; fd + M ( r ; fi)

< exp re^+^ f exp r e a t e
-
< 2 exp rezfs for r > r,,(E),

we have that e I p2 +
E and hence e 5 ez. On the other hand 3 a sequence
of numbers r, -+ 00 such that M(r,?;fz)> exp Thus

M ( r n ;fi + f 2 ) 2 exp r2a-E - exp ril+E = exp r p { I - exp ( r i ~ +-~r,$?-E)}

> 4 exp
+
provided E is so small that el E < 0%- E and n is sufficiently large.
Thus e 2 ez and the order of the sum, fi f2, equals e2. 0 +
NOTE.The proof also applies without significant changes when p2 = 00.
It should be noted that the theorem is sometimes false when el = e2.