Math 372: Solutions to Homework

Steven Miller

October 21, 2013

Abstract
Below are detailed solutions to the homework problems from Math 372 Complex Analysis (Williams College,
Fall 2013, Professor Steven J. Miller, [email protected]). The course homepage is
http://www.williams.edu/Mathematics/sjmiller/public_html/372
and the textbook is Complex Analysis by Stein and Shakarchi (ISBN13: 978-0-691-11385-2). Note to students:
it’s nice to include the statement of the problems, but I leave that up to you. I am only skimming the solutions.
I will occasionally add some comments or mention alternate solutions. If you find an error in these notes,
let me know for extra credit.

Contents
1 Math 372: Homework #1: Yuzhong (Jeff) Meng and Liyang Zhang (2010) 2

2 Math 372: Homework #2: Solutions by Nick Arnosti and Thomas Crawford (2010) 5

3 Math 372: Homework #2: Solutions by Nick Arnosti and Thomas Crawford (2010) 6

4 Math 372: Homework #3: Carlos Dominguez, Carson Eisenach, David Gold 9

5 Math 372: Homework #4: Due Friday, October 4, 2013: Pham, Jensen, Koloğlu 14
5.1 Chapter 3, Exercise 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.2 Chapter 3, Exercise 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5.3 Chapter 3, Exercise 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.4 Chapter 3 Exercise 15d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.5 Chapter 3 Exercise 17a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.6 Additional Problem 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

6 Math 372: Homework #5: Due Friday October 25: Pegado, Vu 22

1
1 Math 372: Homework #1: Yuzhong (Jeff) Meng and Liyang Zhang (2010)
Due by 11am Friday, September 13: Chapter 1: Page 24: #1abcd, #3, #13.

Problem: Chapter 1: #1: Describe geometrically the sets of points z in the complex plane defined by the fol-
lowing relations: (a) |z − z1 | = |z − z2 | where z1 , z2 ∈ C; (b) 1/z = z; (c) Re(z) = 3; (d) Re(z) > c (resp., ≥ c)
where c ∈ R.

Solution: (a) When z1 6= z2 , this is the line that perpendicularly bisects the line segment from z1 to z2 . When
z1 = z2 , this is the entire complex plane.
(b)
1 z z
= = 2. (1.1)
z zz |z|
So
1 z
= z ⇔ 2 = z ⇔ |z| = 1. (1.2)
z |z|
This is the unit circle in C.
(c) This is the vertical line x = 3.
(d) This is the open half-plane to the right of the vertical line x = c (or the closed half-plane if it is ≥).

Problem: Chapter 1: #3: With ω = seiϕ , where s ≥ 0 and ϕ ∈ R, solve the equation z n = ω in C where n is a
natural number. How many solutions are there?

Solution: Notice that

ω = seiϕ = sei(ϕ+2πm) , m ∈ Z. (1.3)
It’s worth spending a moment or two thinking what is the best choice for our generic integer. Clearly n is a bad
choice as it is already used in the problem; as we often use t for the imaginary part, that is out too. The most natural
is to use m (though k would be another fine choice); at all costs do not use i!
Based on this relationship, we have
z n = sei(ϕ+2πm) . (1.4)
So,
i(ϕ+2πm)
z = s1/n e n . (1.5)
Thus, we will have n unique solutions since each choice of m ∈ {0, 1, . . . , n − 1} yields a different solution so
long as s 6= 0. Note that m = n yields the same solution as m = 0; in general, if two choices of m differ by n then
they yield the same solution, and thus it suffices to look at the n specified values of m. If s = 0, then we have only
1 solution.

Problem: Chapter 1: #13: Suppose that f is holomorphic in an open set Ω. Prove that in any one of the follow-
ing cases f must be constant:
(a) Re(f ) is constant;

2
(b) Im(f ) is constant;
(c) |f | is constant.

Solution: Let f (z) = f (x, y) = u(x, y) + iv(x, y), where z = x + iy.

(a) Since Re(f ) = constant,
∂u ∂u
= 0, = 0. (1.6)
∂x ∂y
By the Cauchy-Riemann equations,
∂v ∂u
=− = 0. (1.7)
∂x ∂y
Thus, in Ω,
∂f ∂u ∂v
f ′ (z) = = +i = 0 + 0 = 0. (1.8)
∂x ∂x ∂x
Thus f (z) is constant.

(b) Since Im(f ) = constant,

∂v ∂v
= 0, = 0. (1.9)
∂x ∂y
By the Cauchy-Riemann equations,
∂u ∂v
= = 0. (1.10)
∂x ∂y
Thus in Ω,
∂f ∂u ∂v
f ′ (z) = = +i = 0 + 0 = 0. (1.11)
∂x ∂x ∂x
Thus f is constant.

√ (c) We first give a mostly correct argument; the reader should pay attention to find the difficulty. Since |f | =
u2 + v 2 is constant,
2 2
0 = ∂(u∂x+v ) = 2u ∂u
(
∂v
∂x + 2v ∂x .
2 2 (1.12)
0 = ∂(u∂y+v ) = 2u ∂u ∂v
∂y + 2v ∂y .

Plug in the Cauchy-Riemann equations and we get

∂v ∂v
u +v = 0. (1.13)
∂y ∂x
∂v ∂v
−u +v = 0. (1.14)
∂x ∂y
∂v v ∂v
(1.14) ⇒ = . (1.15)
∂x u ∂y
Plug (1.15) into (1.13) and we get
u2 + v 2 ∂v
= 0. (1.16)
u ∂y

3
So u2 + v 2 = 0 or ∂v
∂y = 0.
If u2 + v 2 = 0, then, since u, v are real, u = v = 0, and thus f = 0 which is constant.
Thus we may assume u2 + v 2 equals a non-zero constant, and we may divide by it. We multiply both sides by
u and find ∂v ∂v ∂u
∂y = 0, then by (1.15), ∂x = 0, and by Cauchy-Riemann, ∂x = 0.

∂f ∂u ∂v
f′ = = +i = 0. (1.17)
∂x ∂x ∂x
Thus f is constant.

Why is the above only mostly a proof? The problem is we have a division by u, and need to make sure
everything is well-defined. Specifically, we need to know that u is never zero. We do have f ′ = 0 except at points
where u = 0, but we would need to investigate that a bit more.
Let’s return to 2 2
0 = ∂(u∂x+v ) = 2u ∂u
(
∂v
∂x + 2v ∂x .
2 2 (1.18)
0 = ∂(u∂y+v ) = 2u ∂u
∂y + 2v ∂v
∂y .
Plug in the Cauchy-Riemann equations and we get
∂v ∂v
u +v = 0
∂y ∂x
∂v ∂v
−u +v = 0. (1.19)
∂x ∂y
We multiply the first equation u and the second by v, and obtain
∂v ∂v
u2 + uv = 0
∂y ∂x
∂v ∂v
−uv + v2 = 0. (1.20)
∂x ∂y
Adding the two yields
∂v ∂v
u2 + v2 = 0, (1.21)
∂y ∂y
or equivalently
∂v
(u2 + v 2 ) = 0. (1.22)
∂y
We now argue in a similar manner as before, except now we don’t have the annoying u in the denominator. If
u2 + v 2 = 0 then u = v = 0, else we can divide by u2 + v 2 and find ∂v/∂y = 0. Arguing along these lines finishes
the proof. 2

One additional remark: we can trivially pass from results on partials with respect to v to those with respect to u
by noting that if f = u + iv has constant magnitude, so too does g = if = −v + iu, which essentially switches the
roles of u and v. Though this isn’t needed for this problem, arguments such as this can be very useful.

4
 The following is from Steven Miller. Let’s consider another proof. If |f | = 0 the problem is trivial as then
f = 0, so we assume |f | equals a non-zero constant. As |f | is constant, |f |2 = f f is constant. By the quotient
rule, the ratio of two holomorphic functions is holomorphic, assuming the denominator is non-zero. We thus find
f = |f |2 /f is holomorphic. Thus f and f are holomorphic, and satisfy the Cauchy-Riemann equations. Applying
these to f = u + iv yields
∂u ∂v ∂u ∂v
= , = − ,
∂x ∂y ∂y ∂x
while applying to f = u + i(−v) gives
∂u ∂(−v) ∂u ∂(−v)
= , = − .
∂x ∂y ∂y ∂x
Adding these equations together yields
∂u ∂u
2 = 0, 2 = 0.
∂x ∂y
Thus u is constant, and by part (a) this implies that f is constant. If we didn’t want to use part (a) we could subtract
rather than add, and similarly find that v is constant.

The following is from Craig Corsi, 2013 TA. The problem also follows from the polar form of the Cauchy-
Riemann equations.

It’s worth mentioning that (a) and (b) follow immediately from (c). For example, assume we know the real part
of f is constant. Consider
g(z) = exp(f (z)) = exp(u(x, y)) exp(iv(x, y)).
As |g(z)| = exp(u(x, y)), we see that the real part of f being constant implies the function g has constant magni-
tude. By part (c) this implies that g is constant, which then implies that f is constant.

2 Math 372: Homework #2: Solutions by Nick Arnosti and Thomas Crawford
(2010)
Due at the start of class by 11am Friday, September 20: Chapter 1: Page 24: #16abc, #24, #25ab. Chapter 2:
(#1) We proved Goursat’s theorem for triangles. Assume instead we know it holds for any rectangle; prove
it holds for any triangle. (#2) HLet γ be the closed curve that is the unit circle centered at the origin, oriented
counter-clockwise. Compute γ f (z)dz where f (z) is complex conjugation (so f (x + iy) = x − iy). Repeat
the problem for γ f (z)n dz for any integer n (positive or negative), and compare this answer to the results
H

for γ z n dz; is your answer surprising? (#3) Prove that the four triangles in the subdivision in the proof of
H

Goursat’s theorem are all similar to the original triangle. (#4) In the proof of Goursat’s theorem we assumed
that f was complex differentiable (ie, holomorphic). Would the result still hold if we only assumed f was
continuous? If not, where does our proof break down?

5
3 Math 372: Homework #2: Solutions by Nick Arnosti and Thomas Crawford
(2010)
Due at the start of class by 11am Friday, September 20: Chapter 1: Page 24: #16abc, #24, #25ab. Chapter 2:
(#1) We proved Goursat’s theorem for triangles. Assume instead we know it holds for any rectangle; prove
it holds for any triangle. (#2) HLet γ be the closed curve that is the unit circle centered at the origin, oriented
counter-clockwise. Compute γ f (z)dz where f (z) is complex conjugation (so f (x + iy) = x − iy). Repeat
the problem for γ f (z)n dz for any integer n (positive or negative), and compare this answer to the results
H

for γ z n dz; is your answer surprising? (#3) Prove that the four triangles in the subdivision in the proof of
H

Goursat’s theorem are all similar to the original triangle. (#4) In the proof of Goursat’s theorem we assumed
that f was complex differentiable (ie, holomorphic). Would the result still hold if we only assumed f was
continuous? If not, where does our proof break down?
R R
Problem: If γ is a curve in C, show that −γ f (z)dz = − γ f (z)dz.

Parameterize γ by z = g(t) for t in [a, b], and define w(t) = g(a + b − t). Then w(t) is a parameterization of
−γ on the interval [a, b] (note that w(a) = g(b), w(b) = g(a)). Additionally, w′ (t) = −g′ (a + b − t). It follows
that
Z Z b Z b
f (z)dz = f (w(t))w′ (t)dt = − f (g(a + b − t))g′ (a + b − t)dt.
−γ a a

Making the substitution u = a + b − t, we get that

Z b Z a
′
− f (g(a + b − t))g (a + b − t)dt = f (g(u))g′ (u)du
t=a u=b
Z b
= − f (g(u))g′ (u)du. (3.1)
u=a

But Z b Z
′
− f (g(u))g (u)du = − f (z)dz,
u=a γ

which proves the claim.

Problem: If γ is a circle centered at the origin, find γ z n dz.
R

We start by parameterizing γ by z = reiθ , 0 ≤ θ < 2π, so dz = ireiθ dθ. Then

Z Z 2π Z 2π
n n inθ iθ n+1
z dz = r e (ire )dθ = ir ei(n+1)θ dθ.
γ 0 0

6
 R 2π
If n = −1, this is ir 0 0 dθ = 2πi. Otherwise, we get
2π
r n+1 i(n+1) 2π
Z 
ir n+1 ei(n+1)θ dθ = e θ  = 0.
0 n+1 0

z n dz.
R
Problem: If γ is a circle not containing the origin, find γ

n+1
R If n 6= −1, the function f (z) = z n has a primitive (namely zn+1 ), so by Theorem 3.3 in Chapter 1 of our book,
γ f (z)dz = 0.
If n = −1, we parameterize γ by z = z0 + reiθ , 0 ≤ θ < 2π, so dz = ireiθ dθ.Then
2π 2π
1 ireiθ ir eiθ
Z Z Z
dz = dθ = dθ.
γ z 0 z0 + re iθ z0 0 1 + zr0 eiθ

Note that because our circle does not contain the origin, |z0 | > r, so | zr0 eiθ | < 1. Thus, we can write this
expression as a geometric series:
2π 2π ∞
ir eiθ ir −r
Z Z X
iθ
dθ = e ( eiθ )k dθ.
z0 0 1 + zr0 eiθ z0 0 z0
k=0

Interchanging the sum and the integral, we see that this is just
∞ ∞
−r k+1 2π i(k+1)θ −r k+1 ei(k+1)θ 2π
Z 
X X
−i ( ) e dθ = − ( ) dθ = 0.
z0 0 z0 k + 1 0
k=0 k=0

Why may we interchange? We can justify the interchange due to the fact that the sum of the absolute values
converges.

z̄ n dz.
R
Problem: If γ is the unit circle centered at the origin, find γ

We start by parameterizing γ by z = eiθ , 0 ≤ θ < 2π, so z̄ = e−iθ and dz = ieiθ dθ. Then
Z Z 2π Z 2π
n −inθ iθ
z̄ dz = e (ie )dθ = i e−i(n−1)θ dθ.
γ 0 0

R 2π
If n = 1, this is i 0 dθ = 2πi. Otherwise, we get
2π
ei(1−n)θ 2π
Z 
−i(n−1)θ
i e dθ = = 0.
0 1 − n 0

7
 Note that instead of doing the algebra, we could have observed that on the unit circle z̄ = z −1 , so γ z̄ n dz =
R
R −n
γz dz. Applying our work from Problem 3, we get the answer above.
Problem: Where in the proof of Goursat’s theorem do we use the fact that the function f is holomorphic? Is it
sufficient to know that f is continuous?

Start by recapping the main ideas behind the proof. We began by continually splitting our triangle T into smaller
triangles. These triangles converge to a point in the limit, and we called this point z0 . We then established the bound
Z Z
 f (z)dz  ≤ 4n 
   
f (z)dz .
T T (n)
Our goal was to show that this quantity tends to zero as z → z0 .
To do this, we Taylor expanded f (z) around the point z0 : f (z) = f (z0 ) + f ′ (z0 )(z − z0 ) + ψ(z)(z − z0 ).
Note that (z − z0 ) divides ψ(z), so ψ(z) → 0 as z → z0 .
Z  Z  Z
′
   
 (n) f (z)dz  ≤  (n) f (z0 ) + f (z0 )(z − z0 )dz  + (n) |ψ(z)(z − z0 )|dz
   
T T T

The first integrand in this sum has a primitive, so the value of this integral is zero. Let Mn = maxz on T (n) |ψ(z)|.
Then |ψ(z)| ≤ Mn , and z − z0 ≤ diam(T (n) ). Hence, the value of the second integral is at most perim(T (n) ) ·
diam(T (n) ) · Mn .
Since the perimeter and diameter of T (n) both decay at a rate of 2−n
R 
, we establish the bound that  T (n) f (z)dz  ≤
4−n CMn for some constant C. Hence, CMn is an upper-bound for  T f (z)dz , and since ψ(z) → 0 as z → z0 ,
 R 

Mn → 0 as desired.
Now let us see what happens if we don’t know that f is differentiable. Using only continuity, we can approxi-
mate f (z) by f (z0 ) + ψ(z)(z − z0 ). Defining Mn as before, we can still bound our integral by CMn . We want to
say that Mn tends to 0, but limz→z0 ψ(z) = limz→z0 f (z)−f (z0 )
z−z0 , which may not exist if f is not differentiable (and
certainly may not tend to zero). Thus, this approach fails.
We could also try the expression fR (z) = f (z0 ) + ψ(z), and then ψ(z) → 0 as z → z0 . Unfortunately, without
the factor of (z − z0 ), our bound on | T (n) f (z)dz| will simply be perim(T (n) ) · Mn = 2−n CMn . Thus, our bound
for | T (n) f (z)dz| is 4n 2−n CMn = 2n CMn . Even though Mn tends to 0, the factor of 2n may overwhelm it, so
R

this approach fails. From these attempts, it seems that knowing that f was differentiable was a fairly important step
in the proof.
Problem: Prove Goursat’s theorem for triangles using only the fact that it holds for rectangles.

Note that it suffices to prove that the integral along any right triangle is zero, since any triangle can be divided
into two right triangles by dropping an altitude.
Given a right triangle ABC, by drawing a series of rectangles inside the triangle, we can reduce the desired
integral to the integral along a series of n congruent triangles similar to ABC, each of which border the original
hypotenuse (as shown in the figure).

8
 Since f is continuous on the original triangle ABC (a compact set) we know that f is uniformly continuous on
the region of interest.
Thus, given any ε > 0, there exists a δ > 0 such that for any two points x, y in ABC with |x − y| < δ,
|f (x) − f (y)| < ε. If h is the length of the hypotenuse of ABC, choose n large enough so that the diameter of
each small triangle, h/n, is less than δ. Then for any triangle Tk and any point zk on that triangle write f (z) =
f (zk ) + ψ(z), so that
Z Z Z Z
f (z)dz = f (zk ) + ψ(z)dz = f (zk )dz + ψ(z)dz
Tk Tk Tk Tk

Since f (zk ) is a constant, it has a primitive, so the first integral is zero. Meanwhile, since any point on triangle
TRk is within h/n of zk , and we chose n to be such that h/n < δ, we know that |ψ(z)| = |f (z) − f (zk )| < ε. Thus,
| Tk ψ(z)dz| < perim(Tk ) · ε. But perim(Tk ) < 3h/n, so the integral of f (z) along triangle Tk is at most 3hε/n.
Summing over all n triangles, we see that the integral of f (z) along the entire curve is at most 3hε. Since this
technique works for arbitrarily small ε, this implies that the integral of f along any right triangle is zero, proving
the claim.

4 Math 372: Homework #3: Carlos Dominguez, Carson Eisenach, David Gold
HW: Due at the start of class by 11am Friday, September 27: Chapter 2, Page 64: #1, #8. Also do: Chap-
ter 2: (Problems from me): (#1) In the proof of Liouville’s theorem we assumed f was bounded. Is it
possible to remove that assumption? In other words, is it enough to assume that |f (z)| < g(z) for some
real-valued, non-decreasing function g? If yes, how fast can we let f grow? (#2) a) Find all z where the
function f (z) = 1/(1 + z 4 ) is not holomorphic; b) Let a, b, c, and d be integers such that ad − bc = 1. Find all
z where the function g(z) = (az + b)/(cz + d) is not holomorphic. (#3) Compute the power series expansion
of f (z) = 1/(1 − z) about the point z = 1/2 (it might help to do the next problem first, or to write 1 − z as
1/2 − (z − 1/2)). (#4) Do Chapter 1, Page 29, #18.

9
Math 372: Complex Analysis

HW #3: Due at the start of class by 11am Friday, September 27: Chapter 2, Page 64: #1, #8. Also do:
Chapter 2: (Problems from me): (#1) In the proof of Liouville’s theorem we assumed f was bounded. Is
it possible to remove that assumption? In other words, is it enough to assume that |f (z)| < g(z) for some
real-valued, non-decreasing function g? If yes, how fast can we let f grow? (#2) a) Find all z where the
function f (z) = 1/(1 + z 4 ) is not holomorphic; b) Let a, b, c, and d be integers such that ad − bc = 1. Find all
z where the function g(z) = (az + b)/(cz + d) is not holomorphic. (#3) Compute the power series expansion
of f (z) = 1/(1 − z) about the point z = 1/2 (it might help to do the next problem first, or to write 1 − z as
1/2 − (z − 1/2)). (#4) Do Chapter 1, Page 29, #18.

π
1. Let γ1 denote the straight line along the real line from 0 to R, γ2 denote the eighth of a circle from R to Rei 4 ,
π
and γ3 denote the line from Rei 4 to 0. Then by Cauchy’s theorem,
Z
2
e−z dz = 0.
γ1 +γ2 +γ3

We can calculate
Z Z R
−z 2 iπ/4 t)2
− e dz = e−(e eiπ/4 dt
γ3 0
Z R
iπ/4 2
=e e−it dt
0
Z R
= eiπ/4 cos (−t2 )dt + i sin (−t2 ) dt
0
Z R
= eiπ/4 cos (t2 )dt − i sin (t2 ) dt
0
2
So we can calculate the Fresnel integrals by calculating γ1 +γ2 e−z dz, taking R → ∞, dividing by eiπ/4 ,
R

and looking at the real and negative imaginary parts. First we show the integral over γ2 goes to zero:
Z  Z π/4 

 −z 2  2
−R e 2iθ iθ
e dz  =  e iRe dθ
 
   
γ2 0 
Z π/4
2
≤R e−R cos 2θ dθ
0
Z π/4−1/R log R Z π/4
−R2 cos 2θ 2
=R e dθ + R e−R cos 2θ
dθ
0 π/4−1/R log R
 
π 1 1
 
π 2
−R2 cos − R log
≤R − e 2 R
+R·
4 R log R R log R
π −R2 sin 1
 
2
≤ Re R log R
+
4 log R

10
 The log1 R term goes to zero as R goes to infinity. So we need to show that the first term goes to zero. Note
d
that sin x ≥ x/2 for positive x sufficiently close to 0, since sin 0 = 0 and dx sin x ≥ 1/2 for sufficiently
small x. So for sufficiently large R the first term is less than or equal to
π −R2 · R log
1 π log R− logRR
Re R = e ,
4 4

R Ras R 2goes to √
which goes to zero infinity. So, as R → ∞, the contribution from γ2 goes to zero. And we know
that as R → ∞, 0 e−x dx = π/2. So, finally,
∞ √
π 1
Z
2 2
cos (t )dt − i sin (t ) dt = ·√ √
0 2 2/2 + i 2/2
√ √
2π 2π
= − i
4 4
as desired.

8. Since x ∈ R, f is holomorphic in an open circle of radius ǫ centered at x, 0 < ǫ < 1. And by Cauchy’s
inequality,
n!||f ||C
|f (n) (x)| ≤
Rn
Case 1: η ≥ 0. For some 0 < ǫ < 1,
|z| ≤ |x + ǫ|
thus,
|f (z)| ≤ A(1 + |x + ǫ|)η ≤ A(1 + ǫ + |x|)η
by both the given and the triangle inequality. And in Cauchy’s inequality R is just ǫ. So by combining results
from above
n!||f ||C
|f (n) (x)| ≤
ǫn
An!
≤ (1 + ǫ + |x|)η
ǫn
An!
≤ (1 + ǫ + |x| + ǫ|x|)η
ǫn
An!
≤ (1 + ǫ)η (1 + |x|)η . (4.1)
ǫn
Now let
A(n!)
An = (1 + ǫ)η
ǫn
thus,
|f (n) (x)| ≤ An (1 + |x|)η .

Case 2: η < 0. For some 0 < ǫ < 1,

ǫ ≥ |x − z| ≥ |x| − |z|

11
 by the reverse triangle inequality. When we rearrange the inequality we see that

|z| ≥ |x| − |ǫ| = |x| + ǫ

Since η is negative, our goal is to minimize (1+|z|) in order to get an upper bound. Now, by combining our
result above with the Cauchy inequality we get that:

n!||f ||C An!

|f (n) (x)| ≤ n
≤ n (1 − ǫ + |x|)η
ǫ ǫ
An!
≤ (1 − ǫ + |x| − ǫ|x|)η
ǫn
An!
≤ (1 − ǫ)η (1 + |x|)η . (4.2)
ǫn
Now let
A(n!)
An = (1 − ǫ)η
ǫn
thus,
|f (n) (x)| ≤ An (1 + |x|)η .
q.e.d.

1. In the proof of Liouville’s theorem, we had that

B
|f ′ (z0 )| ≤
R
where B was an upper bound for f . It only matters that B is an upper bound for f in a disc of radius R about
z0 , however. Let BR be the smallest upper bound for f in a disc of radius R about z0 . Liouville’s theorem
still holds if BR → ∞ as long as BR /R → 0 for every choice of z0 . Alternatively, we just need f to grow
slower than linear; say |f (z)| is less than C|z|1−ǫ or C|z|/ log |z| or anything like this (for those who have
seen little-oh notation, f (z) = o(z) suffices).

2. (a) f is holomorphic wherever its derivative exists:

4z 3
f ′ (z) = −
1 + z4
√ √ √ √
2 2 2 2
That
√
is, whenever
√
z 4 6= −1.
√ √
This gives z = eiπ/4 , e3iπ/4 , e5iπ/4 , and e7iπ/4 , or 2 + 2 i, − 2 + 2 i,
2 2 2 2
− 2 − 2 i, and 2 − 2 i.
(b) The ad − bc = 1 condition prevents g from being a mostly-constant function with an undefined value
at z = −d/c. (That is, if ad − bc = 0, then a/c = b/d, and so the function would simply collapse to
the value of a/c.) So
(cz + d)a − (az + b)c 1
g′ (z) = =
(cz + d)2 (cz + d)2
The function is then not holomorphic at z = −d/c.

12
 3. Just use the geometric series formula:
1 1
=
1−z 1/2 − (z − 1/2)
2
=
1 − 2(z − 1/2)
∞
X
= 2n+1 (z − 1/2)n .
n=0

P∞ n.
4. Let f (z) = n=0 an z Then
∞
X
f (z) = an (z0 + (z − z0 ))n
n=0
∞ n  
" #
X X n
= an (z − z0 )m z0n−m
m
n=0 m=0
∞ ∞   !
X X n n−m
= (z − z0 )m an z .
n=m
m 0
m=0

The inner sum converges by the root test:

s   s 
n n 1 n
lim sup an = lim n
n→∞ m R n→∞ m
1
=
R
whereq R is the radius of convergence of the original power series for f and second limit is evaluated by noting
n n
m/n and lim m/n = 1. Since the inner sum has the same radius of convergence as
1 ≤ m ≤ n n→∞ n
the original sum, z0 still lies in the disc of convergence in the inner sum; hence all the coefficients of z − z0
converge, and f has a power series expansion about z0 .

Homework 4: Due at the start of class by 11am Friday, October 11 (even if this is Mountain Day): Chapter
3, Page 103: #1, #2, #5 (this
P∞ is related to the Fourier
P∞ transform of the Cauchy density), #15d, #17a (hard).
n m
Additional: Let f (z) = n=−5 an z and g(z) = m=−2 bm z be the Laurent expansions for two functions
holomorphic everywhere except possibly at z = 0. a) Find the residues of f (z) and g(z) at z = 0; b) Find the
residue of f (z) + g(z) at z = 0; c) Find the residue of f (z)g(z) at z = 0; d) Find the residue of f (z)/g(z) at
z = 0.

13
5 Math 372: Homework #4: Due Friday, October 4, 2013: Pham, Jensen, Koloğlu
HW: Due at the start of class by 11am Friday, October 11 (even if this is Mountain Day): Chapter 3, Page
103: #1, #2, P
#5 (this is related to the Fourier
P∞ transform of the Cauchy density), #15d, #17a (hard). Additional:
∞ n m
Let f (z) = n=−5 an z and g(z) = m=−2 bm z be the Laurent expansions for two functions holomorphic
everywhere except possibly at z = 0. a) Find the residues of f (z) and g(z) at z = 0; b) Find the residue of
f (z) + g(z) at z = 0; c) Find the residue of f (z)g(z) at z = 0; d) Find the residue of f (z)/g(z) at z = 0.

5.1 Chapter 3, Exercise 1

iπz
Exercise 5.1. Using Euler’s formula sin πz = e −e
−iπz
2i , show that the complex zeros of sin πz are exactly the
integers, and that they are each of order 1. Calulate the residue of sin1πz at z = n ∈ Z.
eiπz0 −e−iπz0
Solution: To show that the complex zeros of sin πz are exactly the integers, we will show that 2i =0
if and only if z0 ∈ Z.
iπz
First prove the forward direction. We see that e 0 −e
−iπz0
2i = 0 gives

eiπz0 = e−iπz0 . (5.1)

Since z0 = x + iy with x, y ∈ R,
eiπx e−πy = e−iπx eπy . (5.2)
For complex numbers to be equivalent, their magnitudes must be the same. Thus,

e−πy = eπy . (5.3)

This implies that −πy = πy, so y = 0. The angles corresponding to Equation 5.2 must be congruent modulo 2π as
well. Thus,
πx ≡ −πx mod 2π, (5.4)
which means πx ≡ 0 or π. So we have
2πx mod 2π ≡ 0, (5.5)
which implies that x is an integer. Thus x ∈ Z. Since y = 0, we have z0 = x, implying z0 ∈ Z.
To prove the backward direction, consider z0 ∈ Z for z0 even,

eiπz − e−iπz
sin πz0 =
2i
1−1
= = 0. (5.6)
2i
Similarly for z0 odd,

eiπz − e−iπz
sin πz0 =
2i
−1 + 1
= = 0. (5.7)
2i
Thus sin πz0 = 0 if and only if z0 ∈ Z. So the zeros of sin πz are exactly the integers.
Next we must show that each zero has order 1. We refer to Theorem 1.1 in Stein and Shakarchi.

14
Theorem 5.2. Suppose that f is holomorphic in a connected open set Ω, has a zero at a point z0 ∈ Ω, and does not
vanish identically in Ω. Then there exists a neighborhood U ⊂ Ω of z0 , a non-vanishing holomorphic function g on
U, and a unique positive integer n such that f (z) = (z − z0 )n g(z) for all z ∈ U .
Since sin πz is analytic, take its Taylor series about z0 . We add zero to write z as z − z0 + z0 . Using properties
of the sine function, we claim

sin πz = sin π(z + z0 − z0 ) = sin π(z − z0 ) cos πz0 + cos π(z − z0 ) sin πz0 . (5.8)

Note this statement does require proof, but will follow from standard properties of the exponential function (or
from analytic continuation). The reason some work needs to be done is that z − z0 need not be real, but the relation
above does hold when z is real. What we are trying to do is understand the behavior of the function near z0 from
knowledge near 0 (as z − z0 is close to zero). This is a common trick, but of course what makes this tractable is
that we have the angle addition formula for sine.
When z0 is an integer, we always have sin πz0 = 0. If z0 is odd then cos πz0 is -1 while if z0 is even it is 1.
Thus for odd z0 ,
π π3 π5
sin πz = − (z − z0 )1 + (z − z0 )3 − (z − z0 )5 + · · · (5.9)
1! 3! 5!
and for even z0 ,
π π3 π5
sin πz = (z − z0 )1 − (z − z0 )3 + (z − z0 )5 − · · · . (5.10)
1! 3! 5!
We thus see that all zeros are simple.
We now turn to finding the residue at z = n for 1/ sin πz. From our Taylor expansion above, we have
1 1 1 1
= = . (5.11)
sin πz sin π(z − n) cos πn cos πn sin π(z − n)
The problem is now solved by using the Taylor expansion of sine and the geometric series. We have cos πn =
(−1)n , so
1 1
= (−1)n 1 3
sin πz π(z − z0 ) − − z0 )3 + · · ·
3! π (z
(−1)n 1
= 1


π(z − z0 ) 1 − 3! π (z − z0 )2 + · · ·
2
2 !
(−1)n
  
1 2 1
= 1+ π (z − z0 )2 + · · · + π 2 (z − z0 )2 + · · · + ··· .
π(z − z0 ) 3! 3!
(5.12)

Note that each term in parentheses in the last line is divisible by (z − z0 )2 , and thus none of these will contribute to
the residue, which is simply (−1)n /π.

5.2 Chapter 3, Exercise 2

Exercise 5.3. Evaluate the integral
∞
dx
Z
.
−∞ 1 + x4

15
 1
Solution: Consider the function f (z) = 1+z 4 . This function has poles at

1/f (z) = 0
1 + z4 = 0
π π
z = ei( 4 +n 2 ) . (5.13)

Consider the contour of the semicircle in the upper half plane of radius R, denoted γ. Denote the part of the
contour along the real line γ1 and the part along the arc γ2 . Note that two of the poles of f (z) lie inside this contour.
Thus by Cauchy’s residue theorem,
1
I
f dz = Resf (eiπ/4 ) + Resf (ei3π/4 ). (5.14)
2πi γ

To find the residues, write

    
1 1 1 1 1
f (z) = = π 3π 5π 7π .
1 + z4 z − ei 4 z − ei 4 z − ei 4 z − ei 4

Thus
   
iπ/4 1 1 1
Resf (e ) = π 3π π 5π π 7π
ei 4 −
e
i 4 i4
 e  −
e
i 4 i4
 e −e
i 4

3π 1 1 1
= e−i 4
1−i 2 1+i
1+i
= − √ (5.15)
4 2
and similarly
   
i 3π −i 9π 1 1 1
Resf (e 4 ) = e 4
1+i 1−i 2
1−i
= √ (5.16)
4 2
Thus we have
1 1+i 1−i
I
f dz = − √ + √
2πi γ 4 2 4 2
i
= − √
2 2
π
I
f dz = √ . (5.17)
γ 2

Now, note that I I Z Z

f dz = f dz = f dz + f dz. (5.18)
γ γ1 +γ2 γ1 γ2

16
 Observe that
R
1
Z Z
f dz = dx
γ1 −R 1 + x4
and that
R
1
Z Z
f dz = dx
γ2 −R 1 + z4
R
Z  Z 
   1 
 f dz  = 
4
dx 
−R 1 + z
  
γ2
 
 1 
≤ max   πR
z∈γ2 1 + z 4 
1
= πR. (5.19)
R4 − 1
Thus Z 
  πR
lim  f dz  ≤ lim 4 = 0. (5.20)
R→∞ γ2R→∞ R − 1
R
Hence, as R → ∞, γ2 f dz → 0. Therefore as R → ∞ we get our final result;
R
1 π
Z Z
lim dx + lim f dz = √
R→∞ −R 1 + x4 R→∞ γ2 2
Z ∞
1 π
4
dx = √ . (5.21)
−∞ 1 + x 2

5.3 Chapter 3, Exercise 5

R +∞e−2πixξ
Exercise 5.4. Use contour integration to show that −∞ (1+x2 )2 dx = π2 (1 + 2π|ξ|)e−2π|ξ| for all ξ real.
e−2πizξ e−2πizξ
Solution: Let f (z) = (1+z 2 )2
= (z+i)2 (z−i)2
. We see that f (z) has poles of order 2 at z = ±i. Thus

d
resz0 f (z) = lim (z − z0 )2 f (z). (5.22)
z→z0 dz
Alternatively, we could write our function as
g(z)
f (z) = , (5.23)
(z − z0 )2
and then we need only compute the coefficient of the z − z0 term of g.
Now consider the residue at z0 = i:
d −2πizξ
resz0 =i f (z) = lim (e (z + i)−2 )
z→i dz
= lim (−2πiξe−2πizξ (z + i)−2 − 2e−2πizξ (z + i)−3 )
z→i
1 1
= πiξe2πξ − ie2πξ . (5.24)
2 4

17
For z0 = −i, we have:
d −2πizξ
resz0 =−i f (z) = lim (e (z − i)−2 )
z→i dz
= lim (−2πiξe−2πizξ (z − i)−2 − 2e−2πizξ (z − i)−3 )
z→−i
1 1
= πiξe−2πξ + ie−2πξ . (5.25)
2 4
Now let us first consider the case when ξ < 0. We will use the contour γ of a semicircle oriented counterclockwise
in the upper half-plane with radius R. Call the portion of γ along the real line γ1 and the arc portion γ2 . Note that
there is a pole inside γ at z0 = i. By the residue formula, we have that
 
1 1 2πξ 1
Z
2πξ
f (z)dz = 2πi πiξe − ie = −π 2 ξe2πξ + πe2πξ . (5.26)
γ 2 4 2

We also know that Z +∞ Z

f (x)dx = lim f (z)dz. (5.27)
−∞ R→∞ γ1

Along γ2 , z = Reiθ and dz = iReiθ dθ, where z = R cos θ + iR sin θ. Thus

π iθ
e−2πiξRe iReiθ
Z Z
f (z)dz = dθ. (5.28)
γ2 0 (1 − R2 ei2θ )2

Then it follows that

Z π  −2πiξR cos θ 2πξR sin θ
iReiθ 
Z  
  e e
 f (z)dz  ≤
 
 dθ

γ2 0
 (1 − R2 ei2θ )2
Z π  −2π|ξ|R sin θ 
 Re
≤ dθ


(1 − R 2 )2 
0 
Z π
R πR
≤ 2 )2
dθ = 2 − 1)2
. (5.29)
0 (1 − R (R
Taking the limit as R goes to infinity, we have
Z 
  πR
lim  f (z)dz  ≤ lim = 0. (5.30)
R→∞ γ2 R→∞ (R2 − 1)2

Thus Z
lim f (z)dz = 0. (5.31)
R→∞ γ2
R R
So limR→∞ γ f (z) = limR→∞ γ1 f (z). It thus follows from Equation 5.26 that
+∞
e−2πixξ 1
Z
2 2
dx = −π 2 ξe2πξ + πe2πξ
−∞ (1 + x ) 2
π
= (1 + 2π|ξ|) e−2π|ξ| (5.32)
2

18
 Now consider ξ ≥ 0. We will use the contour γ of a semicircle oriented counterclockwise in the lower half-
plane with radius R. Call the portion of γ along the real line γ1 and the arc portion γ2 . Note that there is a pole
inside γ at z0 = −i. By the residue formula, we have that
 
1 1 −2πξ 1
Z
−2πξ
f (z)dz = 2πi πiξe + ie = −π 2 ξe−2πξ − πe−2πξ . (5.33)
γ 2 4 2

Also note that, Z +∞ Z

f (x)dx = − lim f (z)dz. (5.34)
−∞ R→∞ γ1

Along γ2 , z = Reiθ and dz = iReiθ dθ, where z = R cos θ + iR sin θ. Thus,

Z 0 −2πiξR cos θ 2πξR sin θ
e e iReiθ
Z
f (z)dz = dθ. (5.35)
γ2 −π (1 − R2 ei2θ )2

Accordingly,
Z  Z
 

 f (z)dz  ≤
 |f (z)|dz
γ2 γ2
Z 0  2π|ξ|R sin θ 

Re
≤ dθ
 

(1 − R 2 ei2θ )2 
−π 
Z 0  
 R 
≤ 
2 2
 dθ
−π (1 − R )

πR
= (5.36)
(1 − R2 )2

Taking the limit as R goes to infinity, we have

Z 
  πR
lim  f (z)dz  ≤ lim = 0. (5.37)
R→∞ γ2 R→∞ (R2 − 1)2

And thus, Z
lim f (z)dz = 0. (5.38)
R→∞ γ2
R R
So limR→∞ γ f (z) = limR→∞ γ1 f (z). It thus follows from Equation 5.33 that

+∞
e−2πixξ
 
1 −2πξ
Z
2 −2πξ
dx = − −π ξe − πe
−∞ (1 + x2 )2 2
π
= (1 + 2π|ξ|) e−2π|ξ| (5.39)
2
Thus for all ξ real,
+∞
e−2πixξ π
Z
2 2
dx = (1 + 2π|ξ|) e−2π|ξ| (5.40)
−∞ (1 + x ) 2

19
5.4 Chapter 3 Exercise 15d
1 For any entire function f , let’s consider the function ef (x) . It is an entire function and furthermore we have the
real part of f is bounded so:
|ef | = |eu+iv | = |eu | ≤ ∞
Hence ef is bounded and therefore, by Louisville’s Theorem, ef is constant. It then follows that f is constant .
Alternatively, we could argue as follows. We are told the real part of f is bounded. Let’s assume that the real
part is always at most B − 1 in absolute value. Then if we consider g(z) = 1/(B − f (z)) we have |g(z)| ≤ 1.
To see this, note the real part of B − f (z) is at least 1. We again have constructed a bounded, entire function, and
again by Liouville’s theorem we can conclude g (and hence f ) is constant.

5.5 Chapter 3 Exercise 17a

Exercise 5.5. Let f be non-constant and holomorphic in an open set containing the closed unit disc. Show that if
|f (z)| = 1 whenever |z| = 1, then the image of f contains the unit disc.

Solution: Suppose f (z) does not have a zero in the unit disc, D. Then 1/f (z) is holomorphic in D. Note that
since |f (z)| = 1 whenever |z| = 1, |1/f (z)| = 1/|f (z)| = 1 whenever |z| = 1 as well. But f (z) is holomorphic
in D, implying |f (z)| ≤ 1 in D by the maximum modulus principle since |f (z)| = 1 on the boundary of D. We
find 1 ≤ |f (z)| ≤ 1 in the unit disk, which implies that our function is constant as its modulus is constant (we
would like to use Exercise 15d, but that requires our function to be entire; fortunately we can obtain constancy by
the Open Mapping Theorem), contradicting the assumption that f is not constant!
Let w0 ∈ D. Consider the constant function g(z) = −w0 . On the unit circle, |f (z)| = 1 > |w0 | = |g(z)| for
all |z| = 1. Thus by Rouché’s theorem, f (z) and f (z) + g(z) have the same number of zeroes inside the unit circle
(ie, in D). But we have shown that f (z) has at least one zero, thus for some zw , 0 = f (zw ) + g(zw ) = f (zw ) − w0 .
Thus for all w0 ∈ D, there exists zw such that f (zw ) = w0 . Thus the image of f (z) contains the unit disc. 

5.6 Additional Problem 1

2 Let:
∞
X ∞
X
n
f (z) = an z g(z) = bm z m
n=−5 m=−2

1. We have:
res0 f = a−1 res0 g = b−1

2. We have
−3
X ∞
X
n
f (z) + g(z) = an z + (an + bn )z n
n=−5 n=−2

So res0 (f + g) = a−1 + b−1 .

1
Hint from Professor Miller
2
Hint from Professor Miller

20
 3. We have −1 = −5 + 4 = −4 + 3 = −3 + 2 = −2 + 1 = −1 + 0 = 0 − 1 = 1 − 2 so:
res0 (f g) = a−5 b4 + a−4 b3 + a−3 b2 + a−2 b1 + a−1 b0 + a0 b−1 + a1 b−2

4. We have (assuming b2 6= 0):

P∞ n
f (z) n=−5 an z
= P∞ m
g(z) m=−2 bm z
P∞ n
1 n=−2 an−3 z
= ∞
z3 m
P
m=−2 bm z
P∞ n
1 n=−2 an−3 z
= ∞ . (5.41)
b−2 z 1 − (− b 1 m
P
−2 m=1 bm−2 z )

As z → 0 the final quantity in parentheses tends to zero, and thus we can expand using the geometric series
formula. We only care about the constant term of this fraction, as it is multiplied by 1/b−2 z and thus only the
constant term contributes to the pole. This is a very useful observation. It means that, when we expand with
the geometric series, we can drop many terms, as we only need to keep terms that contribute to the constant
term. Remember, we are not trying to find the Taylor expansion of this function, but rather just one particular
term. We can thus write:
∞ ∞ ∞
!k
f (z) 1 X
n
X 1 X m
= ( an−3 z ) − bm−2 z )
g(z) b−2 z b−2
n=−2 k=0 m=1
  
1 −2 −1 2 1 2 2
= (a−5 z ) (b0 z + · · · ) + 2 (b−1 z + · · · ) + · · ·
b−2 z b−2 b
 −2 
−1 −1 1 0
+(a−4 z ) (b−1 z + · · · ) + · · · + (a−3 z ) (1 + · · · ) + · · · .
b−2
(5.42)
So:  
f 1 b0 b−1 b−1
res0 ( ) = a−5 (− + 2 ) + a−4 (− ) + a−3
g b−2 b−2 b−2 b−2
.

Homework due Friday October 18 (though you are strongly encouraged to hand it in on Friday, October 18, you
may hand it in by 10am on Monday October 21, but 10am does not mean 10:05am!!!): The Midterm!

HW: Due at the start of class by 11am Friday, October 25: Chapter 5: Page 155: #6, #7, #9 (extra credit:
what is the combinatorial significance of this problem?). Chapter 3: Page 104: #10. Additional Problems:
(1) Find all poles of the function f (z) = 1/(1 − z 2 )4 and find the residues at the poles. (2) Consider the map
f (z) = (z − i)/(z + i). Show that this is a 1-to-1 and onto map from the upper half plane (all z = x + iy
with y > 0) to the unit disk. (3) Calculate the Weierstrass product for cos(πz) (this is also problem #10b in
Chapter 5, and the answer is listed there), and for tan(πz).

21
6 Math 372: Homework #5: Due Friday October 25: Pegado, Vu
HW: Due at the start of class by 11am Friday, October 25: Chapter 5: Page 155: #6, #7, #9 (extra credit:
what is the combinatorial significance of this problem?). Chapter 3: Page 104: #10. Additional Problems:
(1) Find all poles of the function f (z) = 1/(1 − z 2 )4 and find the residues at the poles. (2) Consider the map
f (z) = (z − i)/(z + i). Show that this is a 1-to-1 and onto map from the upper half plane (all z = x + iy
with y > 0) to the unit disk. (3) Calculate the Weierstrass product for cos(πz) (this is also problem #10b in
Chapter 5, and the answer is listed there), and for tan(πz).
6. Prove Wallis’s product formula
π 2·2 4·4 2m · 2m
= · ... ....
2 1·3 3·5 (2m − 1) · (2m + 1)

[Hint: Use the product formula for sin z at z = π/2.]

6. We know (from p. 142) the product formula for the sine function is
∞ 
z2

sin(πz) Y
=z 1− 2 .
π n
n=1

Let z = 1/2. Then,

∞ 
(1/2)2

sin(π/2) 1Y
= 1− .
π 2 n2
n=1

Using sin(π/2) = 1, we simplify this equation:

∞  ∞ 
z2
 
1 Y 1Y 1
= 1− 2 = 1−
π n=1
n 2 n=1 (2n)2
∞  ∞ 
(2n)2 − 1
 
2 Y Y (2n + 1)(2n − 1)
= = . (6.1)
π (2n)2 (2n)2
n=1 n=1

But this implies that

∞ 
(2n)2

π Y
= ,
2 n=1 (2n + 1)(2n − 1)
proving the identity.

7. Establish the following properties of infinite products.

(a) Show that if Σ|an |2 converges, and an 6= −1, then the product (1 + an ) converges
Q
to a non-zero limit if and only if Σan converges.
(b) FindQan example of a sequence of complex numbers {an } such that Σan converges
but (1 + an ) diverges. Q
(c) Also find an example such that (1 + an ) converges and Σan diverges.

22
 7. a) Let |an |2 converge
P
P with a1 6= −1.
(⇐) First assume an converges to a nonzero limit. Without loss of generality we may assume that each
an satisfies |an | ≤ 1/2; this is clearly true in the limit (as the sum converges, the
Q summands must tend to zero).
We assume
Q this
to facilitate
P expanding with logarithms. Consider the product (1 + an ). Taking logs, we see
log (1 + an ) = log(1 + an ). Setting x = −an and using the Taylor expansion

x2 x3 x4
log(1 + x) = −x + − + − ··· ,
2 3 4
we see that
Y
X a2 a3
an − n + n − · · · .


log (1 + an ) =
2 3
In general, notice that
∞  2 3
 X ∞
X
k
 x x
|x|k ,

−|x| ≤ − + − · · ·  ≤
2 3
k=2 k=2
or  2 3

2 2
 x x
− · · ·  ≤ |x|2 (1 + |x| + |x|2 + . . . ).

−|x| (1 + |x| + |x| + . . . ) ≤ − +
2 3
P
If a sum x converges to a nonzero limit, we know that |x| converges to zero; thus we may assume (without
changing convergence) that |x| ≤ 12 . Thus using the geometric expansion, we see that 1 + |x| + |x|2 + · · · = 1−|x|
1
.
1 1
Because |x| ≤ 2 , we have that 1−|x| ≤ 2. Hence we have that
 2
x3

2
 x
− · · ·  ≤ 2|x|2 .

−2|x| ≤ − +

2 3
2 3
an − a2n + a3n − · · · . Since an converges, we know
Q
P
P
Recall that we were looking at log (1 + an ) =
eventually we must have |an | < 1/2, so we can assume |an | < 1/2 without changing convergence, and thus use
the simplification involving the geometric series expansion developed in the previous paragraph. Thus we write
Y X a2 a3
an − n + n − · · ·



log (1 + an ) =
X 2 3 X X
2
|an |2 .


≤ an + 2|an | = an + 2
(6.2)

A QUICK WORD OF WARNING. THE ABOVE EQUATION, AND THE ONES BELOW, ARE A LIT-
TLE ODD. REMEMBER THAT OUR SEQUENCE NEED NOT BE JUST REAL NUMBERS. AS SUCH,
WE MUST BE CAREFUL WITH THE DEFINITION OF ABSOLUTE VALUE. WE ABUSE NOTATION
A BIT – WHEN WE WRITE a ≤ b + c, THIS MEANS THE DESIRED RELATION IS TRUE UP TO A
LINEAR RESCALING. REALLY WHAT WE MEAN IS a = b UP TO AN ERROR AT MOST |c|. WE
REALLY SHOULD WRITE THINGS LIKE |a − b| ≤ c, BUT IN A HOPEFULLY OBVIOUS ABUSE OF
NOTATION....
an and |an |2 converge, we must have that an + 2 |an |2 is finite, call it L.
P P P P
Since by
Qassumption
both Q
Thus log (1 + an ) ≤ L, so (1 + an ) ≤ eL , which is again finite. Thus the product converges.

23
 Q Q
(⇒) Next assume (1+an ) converges to a nonzero limit. Since (1+an ) is converging to a nonzero limit, the
terms in the product must be converging to 1, so we must have |an | approaching zero and we can assume |an | < 1/2
without affecting convergence. We now write:
Y X a2 a3
an − n + n − · · ·



log (1 + an ) =
2 3
X |an |2 |an |3 X
an − |an |2 − |an |3 − · · · .



≥ an − − − ··· ≥
2 3
(6.3)

As before, we substitute in using the geometric series expansion:

Y X
an − |an |2 − |an |3 − · · ·



log (1 + an ) ≥
X
an − |an |2 (1 + |an | + |an |2 + · · · )


=
X X X
an − 2|an |2 = |an |2 .


≥ an − 2
(6.4)
2 ≥ |an |2 converge, we must
Q
P P Q P
Thus we see that log (1 + an ) + 2 |an | an . Since (1 + an ) and
(1 + an ) + 2 |an |2 are both
Q
P P
have that log P finite. Thus our sum an is bounded by finite terms, and so the
sum must also be finite itself. Hence the sum an must converge to a finite limit.

b) Let {an } = { √i1 , √

−1 √i √ −i P
1
, 2
, 2
, . . . }. The sum an converges by the alternating series test, since the absolute
value of the terms approaches zero (one can show this by showing that first the odd terms tend to zero in absolute
value and then that the even terms
Q do as well).

Consider now the product 1 + an . For an arbitrary integer N , look at the 2N -th partial product:
2N        
Y i i i i
1 + an = 1+ √ 1− √ ··· 1 + √ 1− √
n=1
1 1 2N 2N

i2  
i2 
= 1− √ ··· 1 − p
12 (2N )2
       
1 1 2 2N + 1
= 1+ ··· 1 + = ··· = 2N + 1.
1 2N 1 2N
(6.5)

Thus when we evaluate at an even term 2N , we see that

2N
Y
lim (1 + an ) = lim (2N + 1) = ∞,
2N →∞ 2N →∞
n=1

so the product diverges. Hence the product diverges at even terms and thus cannot converge in general.

24
 P
c) For a trivial example, let {an } = {1, −1, 1, −1, . . . }. The sum an does not converge because the limit of
the N th partial sum as N tends to infinity does not converge; it alternates between 0 and 1. However, the product
will clearly converge:
Y
an = (1 + 1)(1 − 1)(1 + 1)(1 − 1) · · · = (1)(0)(1)(0) · · · = 0.

For an example in which the sum diverges but the product converges to a nonzero limit, consider the sequence
√ √
{an |a2n−1 = 1/ n, a2n = −1/(1 + n)}∞ n=1 . Grouping the pairs 2n and 2n − 1 together, we see that

∞ ∞   ∞
X X 1 1 X 1
am = √ − √ = √ .
m=1 n=1
n 1 + n n=1
n + n

We’ll show that this series diverges. Notice that for every n,
∞ ∞
X 1 X 1
√ ≥
n+ n 2n
n=1 n=1
P
and since the series on the RHS diverges, by comparison test, so does the series on the LHS. So an diverges.
However, grouping again the even and odd pair terms, for even N , we have

N N/2
Y Y 1 1
(1 + am ) = (1 + √ )(1 − √ )
m=1 n=1
n n + 1
N/2
Y 1 1 1
= (1 + √ − √ −√ )
n n+1 n+n
n=1
N/2 √ √ N/2
Y − n+ n+1−1 Y
= (1 − √ ) = 1=1
n=1
n+n n=1

and for odd N,

N
Y 1
(1 + am ) = (1 + √ )
m=1
N
which converges to 1 as N → ∞. Thus,
∞
Y
(1 + an ) = 1.
n=1

Hence {an } is the desired sequence.

9. Prove that if |z| < 1, then
∞ 
2 4 8
Y k
 1
(1 + z)(1 + z )(1 + z )(1 + z ) · · · = 1 + z2 = .
1−z
k=0

9. Consider the product (1 + z)(1 + z 2 )(1 + z 4 )(1 + z 8 ) · · · . Suppose we tried to multiply this product out:
to get one term, we would need to choose either the 1 or the power of z in each term to multiply by. For example,

25
one term we could get out is simply z, where we would choose the z in the first term and the 1 in every succeeding
term; another way to say this is to write z = z × 1 × 1 × · · · . To write out the entire product, we would have to
make sure we evaluated every possible choice of ones and powers of z.
But this isn’t so bad if we think of choosing terms as counting in binary. In binary counting, a number is written
entirely in terms of 0s and 1s. For any given number, each digit represented a choice between the digit 0 and the
digit 1. If we think of selecting the power of z in a term as picking 1 for a given digit in binary counting, and
selecting the 1 in a term as picking 0 for a given digit in binary, we can identify a bijective correspondence between
integers written in binary and products from our term (with the exception that 000000000 · · · = 1 in our product).
For example, the binary number 101 = · · · 000101 = 22 × 1 + 21 × 0 + 20 × 1 = 5, and if choose the terms
(z)(1)(z 4 )(1)(1) · · · , we see that we get the product z 5 .
To evaluate our product we must sum over all such possible choices. Since all possible binary numbers together
yield precisely the nonnegative integers, this bijective correspondence importantly tells us that the sum over all such
products will be the sum over all nonnegative powers of z, or 1 + z + z 2 + z 3 + . . . . Thus we have (1 + z)(1 +
z 2 )(1 + z 4 )(1 + z 8 ) · · · = 1 + z + z 2 + z 3 + . . . . Since |z| < 1, we can use the geometric expansion of z to write
1
(1 + z)(1 + z 2 )(1 + z 4 )(1 + z 8 ) · · · = 1−z , as desired.
Significance for combinatorics: notice the way in which our solution invokes combinatorics (such as seeing
how many ways we can choose our terms to make a product).
Alternatively, we can truncate the product and multiply by 1 − z. Note that (1 − z)(1 + z) = (1 − z 2 ), then
(1 − z 2 )(1 + z 2 ) = (1 − z 4 ), and so
k k+1
(1 − z)(1 + z)(1 + z 2 )(1 + z 4 ) · · · (1 + z 2 ) = 1 − z 2 ;

as |z| < 1 the latter tends to 1, and thus

k+1
k 1 z2 1
(1 + z)(1 + z 2 )(1 + z 4 ) · · · (1 + z 2 ) = − → .
1−z z−1 1−z
Chapter 3

10. Show that if a > 0, then

∞
log x π
Z
dx = log a.
0 x2 +a 2 2a
[Hint: Use the contour in Figure 10.]

10. We will first find the residue at ia and then integrate over the given contour. Let f (z) = zlog z
2 +a2 , where we
take the branch cut of the logarithm along −ib for all b ∈ [0, ∞). Furthermore, ia is a zero of order 1. Finding the
residue at ia, we have
log z log z log ia log a π
resia f = lim (z − ia) = lim ( ) = = +
z→ia z 2 + a2 z→ia z + ia 2ia 2ia 2a
Label the contours from the portion on the positive real axis γ1 , the larger arc γ2 , the portion on the negative real
axis γ3 , and the smaller arc γ4 . Choose ǫ < min{a, 1}a, R > max{a, 1}. Parametrize γ1 with z(t) = t from ǫ

26
to R, γ2 with z(t) = Reit from 0 to π, γ3 with z(t) = t from −R to −ǫ, and γ4 with z(t) = ǫeit from π to 0.
Integrating over the γ2 and taking absolute values, we have
 Z π
log Reit
Z 
 log z   it 


2 2
dz  = 
it 2 2
Rie dt
γ2 z + a 0 (Re ) + a
   
Z π
 log Reit

it 

≤ 
 R2 e2it + a2 Rie  dt
0
Z π
 log Reit 

=  R2 e2it + a2  Rdt
 
0
Z π 
 log R + it 
=  R2 e2it + a2  Rdt
 
0
Z π
log R + |it|
≤ 2 e2it | + |a2 |
Rdt
0 |R
Z π Z π
log R + t log R + t log R + π
= |a2 | dt ≤ |a2 | dt ≤ π 2
0 |Re2it | +
R
0 R+
R R + |aR |

|a2 |
since t, log R > 0. Since R → ∞, log R + π, R + R → ∞, by L’Hopital,

log R + π 1/R 1
lim |a2 |
= lim |a2 |
= lim |a2 |
= 0.
R→∞ R+ R→∞ 1− R→∞ R−
R R2 R

Thus, as R → ∞, the contribution along γ2 vanishes to 0. Similarly, for γ4 , we have

Z  Z 0 it

 log z   log ǫe it


2 2
dz  =  it 2 2
ǫe dt
γ4 z + a π (ǫe ) + a

Z 0
 log ǫeit

it

≤  ǫ2 e2it + a2 ǫe  dt
 
π
Z 0
 log ǫeit 

=  2 2it + a2  ǫdt
 
π ǫ e
Z 0 
 − log ǫ + it 
=  ǫ2 e2it + a2  ǫdt
 
π
Z 0
− log ǫ + |it|
≤ 2 2it 2
ǫdt
π |ǫ e | + |a |
Z 0 Z 0
− log ǫ + t − log ǫ + t − log ǫ + π
≤ |a2 |
dt ≤ |a2 |
dt ≤ π 2 dt
2it
π |ǫe | +
ǫ
π ǫ+ ǫ ǫ + |aǫ |

|a2 |
since t, − log ǫ > 0. Since ǫ → 0, − log ǫ + π, ǫ + ǫ → ∞, by L’Hopital,

− log ǫ + π −1/ǫ −1
lim |a2 |
= lim |a2 |
= lim |a2 |
= 0.
ǫ→0 ǫ+ ǫ→0 1− ǫ→0 ǫ−
ǫ ǫ2 ǫ

27
Thus, as ǫ → 0, the contribution along γ4 also vanishes to 0. For the integral over γ1 , γ3 , we have
R −ǫ
log z log t log s
Z Z Z
dz = dt + ds.
γ1 +γ3 z 2 + a2 ǫ t2 + a2 −R s2 + a2

Letting s = −t, we have

R ǫ
log z log t log −t
Z Z Z
dz = dt + (−1)dt
γ1 +γ3 z + a2
2
ǫ t + a2
2
R (−t)2 + a2
R Z R
log t log −t
Z
= dt + dt
ǫ t2 + a2 ǫ t2 + a2
R Z R
log t log t + iπ
Z
= dt + dt
ǫ t2 + a2 ǫ t2 + a2
R
t R
Z R Z R 
log t 1 log t iπ
Z
=2 dt + iπ dt = 2 dt + arctan 
ǫ t2 + a2 2
ǫ t +a
2 2
ǫ t +a
2 a a ǫ
log a π
Thus we have, as R → ∞, ǫ → 0 and as resia f = 2ia + 2a , we have
 !
R
t R
 
log t iπ log a π
Z
lim 2 2 2
dt + arctan  = 2πi +
R→∞,ǫ→0 ǫ t +a a a ǫ 2ia 2a
 !
t R
 Z R
π log a iπ 2

log t iπ
lim 2 2 2
dt + lim arctan  = +
R→∞,ǫ→0 ǫ t +a R→∞,ǫ→0 a a ǫ a a
 Z R
iπ 2 π log a iπ 2

log t
lim 2 2 2
dt + = +
R→∞,ǫ→0 ǫ t +a a a a
 Z R 
log t π log a
lim 2 2 2
dt =
R→∞,ǫ→0 ǫ t +a a
∞
log t π log a
Z
dt =
0 t + a2
2 2a

as desired.

Additional Problems

1. Find all poles of the function f (z) = 1/(1 − z 2 )4 and find the residues at the poles.

Let g(x) = 1/f (z) = (1 − z 2 )4 = ((1 + z)(1 − z))4 . We see that the zeros of g are ±1, each with order 4.

28
Hence, the residues are
 4−1
1 d 1
res1 (f ) = lim (z − 1)4
z→1 (4 − 1)! dz (1 − z 2 )4
 3
1 d 1
= lim
z→1 6 dz (1 + z)4
1 1
= lim (−4)(−5)(−6)
z→1 6 (1 + z)7
−20 −20 −5
= lim 7
= 7
=
z→1 (1 + z) 2 32
and
 4−1
1 d 1
res−1 (f ) = lim (z + 1)4
z→−1 (4 − 1)! dz (1 − z 2 )4
 3
1 d 1
= lim
z→−1 6 dz (z − 1)4
1 −1
= lim (4)(5)(6)
z→−1 6 (z − 1)7
−20 −20 5
= lim 7
= 7
=
z→−1 (z − 1) −2 32
Thus we have found the desired residues.
We sketch an alternative proof. We have
1 1
f (z) =
(z − 1) (z + 1)4
4

1 1
=
(z − 1) (z − 1 + 2)4
4

1 1 1
=
(z − 1) 2 (1 + z−1
4 4
2 )
4
4
z − 1 (z − 1)2 (z − 1)3

1 1
= 1− + − + ··· . (6.6)
(z − 1)4 16 2 4 8
The difficulty is we have to expand the factor to the fourth power well enough to identify the coefficient of (z − 1)3 .
A little algebra shows it is − 52 (z − 1)3 , and thus (remembering the factor 1/16) the residue is just −5/32.

2. Consider the map f (z) = (z − i)/(z + i). Show that this is a one-to-one and onto map from the upper half
plane (all z = x + iy with y > 0) to the unit disk.

2. First we’ll show that the range of f is the unit disk. Writing z = x + iy where x, y ∈ R, y > 0, then we have
p
x2 + (y − 1)2
 
 x + (y − 1)i 
|f (x + iy)| =   = p
x + (y + 1)i  x2 + (y + 1)2

29
 p p
and since y > 0, x2 + (y − 1)2 < x2 + (y + 1)2 , f (x + iy) < 1, so the range of f is the unit disk.
Now we’ll show that f is injective. Suppose for z1 , z2 with imaginary part positive, f (z1 ) = f (z2 ). Then
z1 − i z2 − i
=
z1 + i z2 + i
(z1 − i)(z2 + i) = (z2 − i)(z1 + i)
z1 Z2 + z1 i − Z2 i + 1 = z1 z2 − z1 i + z2 i + 1
2i(z1 − z2 ) = 0
z1 = z2 . (6.7)
Here’s another, faster way to do the algebra. We add zero:
z1 − i z2 − i
=
z1 + i z2 + i
z1 + i − 2i z2 + i − 2i
=
z1 + i z2 + i
2i 2i
1− = 1− ; (6.8)
z1 + i z2 + 1
it is clear that the only solution is when z1 = z2 .

Now we’ll show that f is surjective. Given any w ∈ D, setting z = (w + 1)i/(1 − w), we see that
(w+1)i
(1−w) −i (w + 1)i − (1 − w)i
f (z) = (w+1)i
= = w.
+i (w + 1)i + (1 − w)i
(1−w)

Now we’ll show that z has positive imaginary part. Writing w = x + iy with x, y ∈ R, x2 + y 2 < 1, we have
(x + 1) + iy −2y + i(1 − y 2 − x2 )
z=i = .
(1 − x) − iy (1 − x)2 + y 2
So the imaginary part is 1 − (x2 + y 2 ) > 0, so z has positive imaginary part.

3. Calculate the Weierstrass product for cos(πz) (this is also problem 10b in Chapter 5, and the answer is listed
there) and for tan(πz).

3. By the Euler formulas for sine and cosine, we see that

eiπz + e−iπz
cos(πz) =
2
π
ei 2 (eiπz + e−iπz )
=
2i
iπ(z+ 12 ) 1
(e + e−iπ(z− 2 ) )
=
2i
iπ( 12 −z) 1
e − e−iπ( 2 −z) 1
= = sin(π( − z))
2i 2

30
and since the zeros of sin πz occur only at the integers, the zeros of cos πz occur at m + 12 for all m ∈ Z. Thus,
define the sequence {a2n−1 = n + 12 , a2n = −(n + 12 )}∞ n=1 , which are precisely the zeros of cos πz. Furthermore,
j
since the zeros of sine are of order 1, the zeros of cosine are also of order one. Thus we have, for hk (z) = kj=1 zj ,
P

grouping together the pairs 2n and 2n − 1, the Weierstrauss product of cos πz is, up to a factor of eh(z) for some
entire function h,
∞ ∞ ∞
Y z hm (z) Y z z Y
(1 − )e = (1 − 1 )(1 − 1 ) ehm (z)
am n + 2 −(n + 2 )
m=0 n=0 m=1
∞
Y z2 P∞
m=1 hm (z)
= (1 − 1 2 )e
n=0
(n + 2 )
∞
Y 4z 2 P∞
= (1 − 2
)e m=1 hm (z) .
(2n + 1)
n=0
Q∞ 4z 2
Considering n=0 (1 − (2n+1)2
), we’ll show this product converges. Note that

∞ ∞ ∞
X 1 X 1 X 1
2
= 2
+
n=1
n n=1
(2n) n=0
(2n + 1)2
∞ ∞
3X 1 X 1
2
=
4 n (2n + 1)2
n=1 n=0

so since the sum on the RHS is bounded ∞ 1

P
n=1 n2 , a convergent series, the series on the RHS converges as well,
and as the convergence is absolute,
Q the product4zconverges. Thus (up to the exponential of an entire function) the
2
Weierstrauss product of cos πz is ∞ n=0 (1 − (2n+1)2
).

Next, notice that tan(πz), has poles at odd integer multiples of π2 , and so by definition does not have a Weier-
strass product.

R∞
HW: Due Friday, November 1: (1) Evaluate −∞ cos(4x)dx/(x4 + 1). (2) Let U be conformally equivalent to
V and V conformally equivalent to W with functions f : U → V and g : V → U . Prove g ◦ f (g composed
with f ) is a bijection. (3) The Riemann mapping theorem asserts that if U and V are simply connected
proper open subsets of the complex plane then they are conformally equivalent. Show that simply connected
is essential: find a bounded open set U that is not simply connected and prove that it cannot be conformally
equivalent to the unit disk. (4) Chapter 8, Page 248: #4. (5) Chapter 8: Page 248: #5. (6) Chapter 8: Page
251: #14.

31