Bog̃aziçi University, Math 432 Complex Analysis II, C. Y.

Yıldırım

Notes on Math 432 (Chapters 17-18 of Bak & Newman), November 2022

• Definitions of the order of an entire function

In Exercise 16.12 of Bak & Newman, an entire function is said to have finite order ρ if
 f (z) 
ρ = inf k : lim |z|k = 0 .
z→∞ e

We here present give an alternative definition. Let f be an entire function, and M (r) :=
max |f (z)|. The order of f is defined as
|z|=r

ln ln M (r)
ρ := lim sup .
r→∞ ln r
If 0 < ρ < ∞, then the type of f is defined as

ln M (r)
σ := lim sup .
r→∞ rρ
∞
X
If f (z) = cn z n is the power series expansion of the entire function f (z), then the order ρ
n=0
and the type σ of this function can be determined from
n ln n 1 1 √

ρ = lim sup , (σeρ) ρ = lim sup n ρ n
cn .
n→∞ − ln |cn | n→∞

It may be a helpful exercise to establish the equivalences of these definitions (see also Conway’s
book (Vol. I), Chapter XI, §2, Exercise 5).
Z 1 ∞
zt2 2
X
• Determine the order of (a) f (z) = e dt ; (b) g(z) = e−n z n .
0 n=0
Z 1 Z 1
|z|t2 |z| 2 1 x
(a) |f (z)| ≤ e dt ≤ e , so that ρ ≤ 1. Now let x ∈ R . Then, f (x) > +
ext dt > e 4 ,
0 1
2
2
so that ρ ≥ 1. Hence ρ = 1.

(b) For any R, the series assumes its maximum value on |z| = R at z = R. So we may just
take into account the real positive values of z in finding the order of the function defined by
this series. For a given R (we are concerned with the behaviour of the function as R → ∞), let
us first determine around which n the maximum term of the series occurs. As n begins from 1,
the terms are increasing, and as n tends to ∞ the terms are decreasing. To find the maximum
d  −n2 n  log R
term we solve e R = 0. We see that the maximum term occurs when n ≈ with
dn 2
log2 R
value ≤ e 4 . Let us split the series as
∞
2 2 2
X X X
e−n Rn = e−n Rn + e−n Rn .
n=0 n≤2 log R n>2 log R
 We see that
X 2 log2 R
e−n Rn < 3e 4 log R.
n≤2 log R

n2 n2 n2
X
We view the other part as e− 2 e− 2 Rn . Now e− 2 Rn = 1, and the value of
n=2 log R
n>2 log R
2
− n2 n
e R decreases as n increases from 2 log R. Hence
X 2
X n2
X 2
e−n Rn ≤ e− 2 < e−n < .
n>2 log R n>2 log R n>2 log R
R2

2
log2 R log2 R
Therefore, for large R, we have M (R) < 3e 4< 4e log R +
4 log R, so that log M (R) <
R2
log2 R and log log M (R) < 2 log log R. Then, by the second formula above for the order ρ, we
see that the order of the function defined by the series in question is 0.
n ln n
If we use the last formula above for ρ to this problem, we immediately find ρ = lim sup 2
=
n→∞ −(−n )
0.

• Prove that the order of an entire function remains unchanged when the function is differentiated.
Z z
0
Let M (r) = max |f (z)|, M1 (r) = max |f (z)|. We have f (z) = f 0 (ζ) dζ + f (0). This gives
|z|=r |z|=r 0
M (r) ≤ rM1 (r) + |f (0)| which in turnZimplies ρ ≤ ρ1 , where ρ and ρ1 denote the orders of f
1 f (ζ)
and f 0 respectively. Next, f 0 (z) = dζ, where C(z, δ) is the circle of radius δ
2πi C(z,δ) (ζ − z)2
M (r + δ)
centered at z. Then, M1 (r) ≤ , so that ρ1 ≤ ρ. Thus ρ = ρ1 .
δ

• Find the product expansion of eaz − ebz , (a, b ∈ C).

We assume a 6= b. The case a = b is trivial. We apply the Weierstrass factorization theorem

az bz az (az)2
(as given with the questions of Homework IV). In our case f (z) = e − e = 1 + + +
1! 2!
2 2 2
bz (bz) a −b 2
z + · · · , so that m = 1. If eaz − ebz = 0,



··· − 1 + + + · · · = (a − b)z +
1! 2! 2
2πki
then e(a−b)z = 1, so that the zeros of the function are at the points , k ∈ Z. Each of these
a−b
X a−b
points is a simple zero (as f 0 doesn’t vanish at these points). Next, diverges, but
|2πki|
k∈Z\{0}
X (a − b)2
is convergent, so that we can take pk = 1 for all k. So we have
|2πki|2
k∈Z\{0}

∞ 
az bz g(z)
Y  a − b  (a−b)z g(z)
Y (a − b)2 2 
f (z) = e − e = ze 1− z e 2πik = ze 1+ z .
2πik k=1
4π 2 k 2
k∈Z\{0}

Now we need to determine the entire function g(z). Because eaz − ebz is of order 1, g(z) must
be a polynomial of degree ≤ 1, so that
∞ 
Y (a − b)2 2 
f (z) = eaz − ebz = zeuz+v 1+ z .
k=1
4π 2 k 2
 eaz − ebz
Since lim = a − b, and the last product has value 1 at z = 0, we have ev = a − b. To
z→0 z
f (z)
0
find u, we will examine fz(z) at z = 0. We easily find from the series expansion of f (z) that
f (z)
0
z
eaz −ebz 0 a2 −b2
z z a+b 2
lim f (z)
= lim eaz −ebz
= =
. On the other hand, from the product expression
z→0 z→0
z
a−b 2
z
for f (z) we have (calling the last product P for brevity)
 0
∞ (a−b)2 2
 f (z) 0 X 1 + 4π2 k2 z i
= (a − b)euz uP + P

  ,
z k=1
2
1 + (a−b) z 2 4π 2 k2

so that
 0
(a−b)2 2
1+ z i
4π 2 k2
(a − b)euz uP + P ∞
 P
k=1
 
f (z)
0 (a−b)2 2 ∞
1+ z
z 4π 2 k2
X 2(a − b)2 z
= =u+
f (z) (a − b)euz P k=1
4π 2 k 2 + (a − b)2 z 2
z

a+b
Now, letting z → 0, we find u = , and we conclude
2
∞ 
a+b
Y (a − b)2 2 
eaz − ebz = (a − b)ze 2
z
1+ 2k2
z .
k=1
4π

∞
X
• Suppose f (z) = an z n has radius of convergence 1 and assume that an analytic continuation
n=0
∞
X
of f has a pole at z = 1. Show that an z n diverges at every point on the unit circle. (Hint:
n=0
Show that if {an } → 0, then (1 − z)f (z) → 0 as z → 1− along the real axis). (Bak & Newman,
Ex. 18.2(a)).

In |z| < 1 we see that (1 − z)f (z) = a0 + (a1 − a0 )z + (a2 − a1 )z 2 + · · · . If an → 0, then putting
z = 1 in the very last series we have a0 +(a1 −a0 )+(a2 −a1 )+· · · and this converges to 0. Now we
X∞ X∞
recall Abel’s theorem: ”If bn = s, then bn z n → s as z → 1 along any path lying between
n=0 n=0
two chords of the unit circle which pass through z = 1”. We let b0 = a0 , bn = (an − an−1 ) for
n ≥ 1, and so s = 0, and Abel’s theorem gives us lim− (1 − z)f (z) = 0. But then f (z) along
z→1
with its analytic continuation cannot have a pole at z = 1, and this contradicts the assumption
in the problem. Hence an 6→ 0. This means for any z such that |z| = 1, the general term of the
∞
X
series an z n doesn’t tend to 0. Therefore the series can’t converge on any point of |z| = 1.
n=0
Putting z = Reiφ w when the pole is at Reiφ , Ex. 18.2(b) reduces to Ex. 18.2(a).
∞
X
Proof of Abel’s theorem: Let for p ≥ n, sn,p := bn + bn+1 + · · · + bp . Since bn was assumed to
n=0
converge, by Cauchy convergence criterion we have given any  > 0, |sn,p | <  for n0 ≤ n ≤ p.
Now
 m
X
bν z ν = sn,n z n + (sn,n+1 − sn,n )z n+1 + · · · + (sn,m − sn,m−1 )z m
ν=n
= sn,n (z n − z n+1 ) + · · · + sn,m−1 (z m−1 − z m ) + sn,m z m .

So, for n ≥ n0 ,
m
X m−1
X   ∞
X   |1 − z| 
ν ν ν+1 m
bν z | ≤  |z − z | + |z| ≤  |1 − z| |z|ν + 1 =  +1 .
ν=n ν=n ν=0
1 − |z|

|1 − z|
The result now follows provided that is bounded as z → 1 on the path taken. It is this
1 − |z|
that makes it necessary to restrict the path, because this function can be made large by taking
z near to 1, but still nearer to the unit circle.
You may also read Abel’s theorem presented with examples from :
https://kconrad.math.uconn.edu/blurbs/analysis/abelthm.pdf.
Z π
dt π
• Prove that =√ except when z ∈ R and z ≥ 1 or z ≤ −1.
0 1 − z cos t 1 − z2
t 2 du t u
Using the substitution u = tan (so that dt = 2
, sin = √ , we see
2 1+u 2 1 + u2
Z π Z π Z ∞
dt dt 2 du
= 2 t =

.
0 1 − z cos t 0 1 − z(1 − 2 sin 2 ) 1 − z 0 1 + u2 1+z
1−z

1 + z
12 1+z


1 1+z
log 1−z
Let now w = u = ue 2 . The map z 7→ takes (−∞, −1) ∪ (1, ∞) to
1−z 1−z
(−∞, 0), so that we take the principal branch of the logarithm (with the cut on the negative
real axis). Then our integral equals

1 − z
21 ∞ dw 1 − z
12 π
Z
2 2 π
2
= =√ .
1−z 1+z 0 1+w 1−z 1+z 2 1 − z2

∞
Γ0 (z) 1 X z
• (a) Show that ψ(z) := = −γ − + , (z 6= 0, −1, −2, . . .) where the con-
Γ(z) z n=1
n(n + z)
vergence
Z ∞is uniform on compacta of C \ {0, −1, −2, . . .}. (b) Evaluate Γ0 (1). (c) Show that
γ=− e−x log x dx.
0
∞
1 Y z
z
(a) We know the formula = ze γz
1 + e− k . We obtain the asserted formula for ψ(z)
Γ(z) k=1
k
by logarithmic differentiation of this product representation. The digamma function ψ(z) has
poles at {0, −1, −2, · · · }. Hence for uniform convergence of the series z has to stay away by a
fixed distance from these poles. When this condition is satisfied by z it is easy to show uniform
convergence by bounding the tail of the relevant series. (b) By putting z = 1 in the formula
0
Z immediately obtain ψ(1) = −γ. Then, since Γ(1) = 1, we have Γ (1) = −γ. (c)
of (a) we
∞
Γ(z) = e−t tz−1 dt for < z > 0, so that (since the integral converges uniformly in < z ≥ δ > 0
0
for any such fixed δ we can differentiate under the integral Zsign – see the theorem on p.292 of
∞
0
R. C. Buck’s Advanced Calculus (3rd ed.)). Then Γ (z) = e−t tz−1 log t dt. With z = 1 we
0
obtain the assertion.
• Show that ζ(σ) < 0 for σ ∈ (0, 1). One may use the following analytic continuation (or rather
meromorphic continuation because there is a pole at s = 1) of the Riemann zeta-function to
the right half-plane: For s ∈ C, < s > 1 (< and = denote the real part and the imaginary part
respectively; a real number x will be expressed below as the sum of its integer part and its
fractional part: x = bxc + {x}),
∞ ∞ ∞ Z n+1 ∞ Z n+1
X 1 X
−s −s
X
−s−1
X
ζ(s) = s
= n[n − (n + 1) ] = s n x dx = s bxcx−s−1 dx
n=1
n n=1 n=1 n n=1 n
Z ∞ Z ∞ Z ∞ Z ∞
−s−1 −s−1 −s
=s bxcx dx = s (x − {x})x dx = s x dx − s {x}x−s−1 dx
1 1 1 1
Z ∞
s
= −s {x}x−s−1 dx.
s−1 1

The very last integral converges for <s > 0, and thus the last line of the displayed formulas
gives the analytic continuation of ζ(s) to the right half-plane. The pole of ζ(s) at s = 1 is
s
present in the term and we also see there are no other poles in the right half-plane. (In
s−1
fact, there are analytic continuations of ζ(s) the whole complex plane, and s = 1 is the only
pole).
Z ∞
s
ζ(s) = −s {x}x−s−1 dx for < z > 0, so first of all ζ(σ) ∈ R for σ ∈ (0, 1). If ζ(σ) = 0
s−1 1 Z ∞
1 {x}
for σ ∈ (0, 1), then = dx. But, for 0 < σ < 1, the left-hand member of the
σ−1 0 xσ+1
last equation is negative, while the right-hand member is positive. This contradiction implies
ζ(σ) 6= 0 for 0 < σ < 1. It is also clear from the formula for the analytic continuation to
< z > 0 that ζ(σ) → −∞ as σ → 1− . Thus ζ(σ) < 0 for 0 < σ < 1.
∞
X
• Prove that f (z) = z n! can not be continued beyond its circle of convergence.
n=0
P∞
n=0 z n! is convergent for |z| < 1, and divergent for |z| > 1. By the theorem on lacunary series
(n + 1)!
(Theorem 18.5 of Bak & Newman), since lim = ∞, the circle of convergence |z| = 1
n→∞ n!
is a natural boundary for f (z).
∞
Y 1
1
• Express 1− e 2n in terms of the Γ-function.
n=1
2n
∞ γ ∞
1 −γz
Y z
−z 1 1 e− 2 Y 1
1
We use = ze 1 + e with z = − , so that
k
1 = − 1 − e 2n .
Γ(z) n=1
k 2 Γ(− 2
) 2 n=1
2n
1 1 1 √
Also, zΓ(z) = Γ(z + 1) with z = − gives Γ(− ) = −2Γ( ) = −2 π, so that
2 2 2
∞ r
Y 1
1 eγ
1− e 2n = .
n=1
2n π