You are on page 1of 52

Chapter 2.

6, Page 64
Exercise 1: Prove that
_

0
sin(x
2
)dx =
_

0
cos(x
2
)dx =

2
4
.
These are the Fresnel integrals. Here,
_

0
is interpreted as lim
R
_
R
0
.
Solution. Let f(z) = e
iz
2
. We integrate f(z) around a circular sector of
radius R running from = 0 to

4
. Along the x axis the integral is
_
R
0
e
ix
2
dx.
Along the curved part we have z = Re
i
and the integral is
_
/4
0
e
iR
2
e
2i
iRe
i
d = iR
_
/4
0
e
R
2
sin(2)
e
i(+iR
2
cos(2))
d.
Finally, along the segment at angle

4
we have z = re
i/4
and the integral
is
_
0
R
e
r
2
e
i/4
dr. The total integral is zero since f is analytic everywhere.
As R , the integral over the third piece approaches
e
i/4
_

0
e
x
2
dx = e
i/4

2
=

2
4

2
4
i.
To estimate the integral over the curved piece, we used the fact that
sin(2)
4

for 0

4
; this follows from the concavity of sin(2).
Using this,

iR
_
/4
0
e
R
2
sin(2)
e
i(+iR
2
cos(2))
d

R
_
/4
0

e
R
2
sin(2)
e
i(+iR
2
cos(2))

d
= R
_
/4
0
e
R
2
sin(2)
d
R
_
/4
0
e
4R
2
/
d
=

4R
e
4R
2
/

/4
0
=
(1 e
R
2
)
4R
.
As R , this approaches zero and we are left with
_

0
e
ix
2
dx

2
4

2
4
i = 0.
Taking real and imaginary parts, we have
_

0
cos(x
2
)dx =
_

0
sin(x
2
)dx =

2
4
.

Exercise 2: Show that


_

0
sin x
x
dx =

2
.
1
2
Solution. We integrate f(z) =
e
iz
z
around an indented semicircular contour
bounded by circles of radius and R in the upper half plane. The integrals
along the two portions of the real axis add up to
_

R
cos(x) +i sin(x)
x
dx +
_
R

cos(x) +i sin(x)
x
dx = 2i
_
R

sin(x)
x
dx
because cosine is even and sine is odd. The integral around the arc of radius
R tends to zero as R , by the Jordan lemma; since this lemma isnt
mentioned in the book, heres a proof for this specic case: On this arc,
z = Re
i
so the integral is
_

0
e
iRe
i
Re
i
iRe
i
d = i
_

0
e
Rsin()
e
iRcos()
d.
The absolute value of this integral is at most
_

0
e
Rsin()
d = 2
_
/2
0
e
Rsin()
d
by symmetry. Now sin()
2

for 0

2
by the concavity of the sine
function, so this is at most
2
_
/2
0
e
2R/pi
d =
e
2R/pi
R

/2
0
=
(1 e
R
)
R
which tends to 0 as R .
Finally, the integral over the inner semicircle tends to i; this is an
immediate consequence of the fractional residue theorem, but since that
doesnt seem to be mentioned in this book either (gosh!), we can also see
it from the fact that
e
iz
z
=
1
z
+ O(1) as z 0, and since the length of the
semicircle is tending to zero, the integral over it approaches the integral of
1
z
over it, which is
_
0

1
e
i
ie
i
d =
_

0
1d = i.
Putting the pieces together and letting R and 0, we have
2i
_

0
sin(x)
x
dx i = 0
_

0
sin(x)
x
dx =

2
.

Exercise 3: Evaluate the integrals


_

0
e
ax
cos bxdx and
_

0
e
ax
sinbxdx, a > 0
by integrating e
Ax
, A =

A
2
+B
2
, over an appropriate sector with angle
, with cos =
a
A
.
Solution. As indicated, we integrate f(z) = e
Az
around a circular sector
of radius R with 0 , where = cos
1
(
a
A
) is strictly between 0 and
3

2
. (Here we assume b ,= 0 since otherwise the integrals are trivially equal
to
1
a
and 0 respectively). The integral along the x axis is
_
R
0
e
Ax
dx
_

0
e
Ax
dx =
1
A
as R . To estimate the integral over the curved part we use the fact
that cos() 1
2

for 0

2
, which follows from the concavity of the
cosine in the rst quadrant. Then we have

_

0
e
ARe
i
Re
i
d

_

0

e
ARe
i
Re
i

d
= R
_

0
e
ARcos()
d
R
_

0
e
AR
e
2AR/
d
= Re
AR

2AR
e
2AR/

0
=

2A
_
e
AR(1
2

)
e
AR
_
.
Since 1
2

is a positive constant, this tends to 0 as R . Finally, on


the segment with = , z = re
i
= r
a+bi
A
, so the integral is
_
0
R
e
Ar(a+ib)/A
a +ib
A
dr =
a +ib
A
_
0
R
e
ar
e
ibr
dr.
Putting the pieces together and letting R , we have
a +ib
A
_
0

e
ax
e
ibx
dx +
1
A
= 0
_

0
e
ax
e
ibx
dx =
1
a +ib
=
a ib
a
2
+b
2
.
Comparing the real and imaginary parts, we have
_

0
e
ax
cos(bx)dx =
a
a
2
+b
2
and
_

0
e
ax
sin(bx)dx =
b
a
2
+b
2
.

Exercise 4: Prove that for all C we have e

2
=
_

e
x
2
e
2ix
dx.
4
Solution. Let = a +bi with a, b R. Then
_

e
x
2
e
2ix
dx
_

e
x
2
e
2ix(a+bi)
dx
=
_

e
(x
2
2bx)
e
2iax
dx
= e
b
2
_

e
(xb)
2
e
2iax
dx
= e
b
2
e
2iab
_

e
(xb)
2
e
2ia(xb)
dx
= e
b
2
e
2iab
_

e
u
2
e
2iau
du
= e
b
2
e
2iab
e
a
2
= e
(a+bi)
2
= e

2
.

Exercise 5: Suppose f is continuously complex dierentiable on , and T


is a triangle whose interior is also contained in . Apply Greens theorem
to show that
_
T
f(z)dz = 0.
This provides a proof of Goursats theorem under the additional assumption
that f

is continuous.
Solution. Write f(z) as f(x, y) = u(x, y)+iv(x, y) where u, v are real-valued
and z = x +iy. Then dz = x +idy so
_
T
f(z)dz =
_
T
(u(x, y) +iv(x, y))(dx +idy)
=
_
T
udx v dy +i
_
T
v dx +udy
=
__ _
v
x
+
u
y
_
dxdy +i
__ _
u
x

v
y
_
dxdy
= 0
by the Cauchy-Riemann equations. (The double integrals are, of course,
taken over the interior of T.)
Exercise 6: Let be an open subset of C and let T be a triangle whose
interior is also contained in . Suppose that f is a function holomorphic
in except possibly at a point w inside T. Prove that if f is bounded near
w, then
_
T
f(z)dz = 0.
5
Solution. Let

be a circle of radius centered at w, where is suciently


small that

lies within the interior of T. Since f is holomorphic in the


region R between T and

,
_
R
f(z)dz =
_
T
f(z)dz
_

f(z)dz = 0.
Thus,
_
T
f(z)dz =
_

f(z)dz. But f is bounded near w and the length


of

goes to 0 as 0, so
_

f(z)dz 0 and therefore


_
T
f(z)dz = 0.
(Note: If were not allowed to use Cauchys theorem for a region bounded
by two curves, one can use a keyhole contour instead; the result is the
same.)
Exercise 7: Suppose f : D C is holomorphic. Show that the diameter
d = sup
x,wD
[f(z) f(w)[ of the image of f satises
2[f

(0)[ d.
Moreover, it can be shown that equality holds precisely when f is linear,
f(z) = a
0
+a
1
z.
Solution. By the Cauchy derivative formula,
f

(0) =
1
2i
_
Cr
f()

2
d
where C
r
is the circle of radius r centered at 0, 0 < r < 1. Substituting
for and adding the two equations yields
2f

(0) =
1
2i
_
Cr
f() f()

2
d.
Then
[2f

(0)[
1
2
_
[f() f()[
r
2
d
M
r
r

d
r
,
where
M
r
= sup
||=r
[f() f()[.
Letting r 1 yields the desired result.
Exercise 8: If f is a holomorphic function on the strip 1 < y < 1, x R
with
[f(z)[ A(1 +[z[)

, a xed real number


for all z in that strip, show that for each integer n 0 there exists A
n
0
so that
[f
(n)
(x)[ A
n
(1 +[x[)

, for all x R.
Solution. For any x, consider a circle C centered at x of radius
1
2
. Applying
the Cauchy estimates to this circle,
[f
(n)
(x)[
n!|f|
C
(1/2)
n
where |f|
C
= sup
zC
[f(z)[. Now for z C, 1 + [z[ 1 + [x[ + [z x[ =
3
2
+[x[ < 2(1 +[x[), so
[f(z)[ A(1 +[z[)

A2

(1 +[x[)

.
6
Hence, |f|
C
A2

(1 +[x[)

, so
[f
(n)
(x)[ n!2
n
A2

(1 +[x[)

= A
n
(1 +[x[)

with A
n
= n!2
n
A2

.
Exercise 9: Let be a bounded open subset of C, and : a holo-
morphic function. Prove that if there exists a point z
0
such that
(z
0
) = z
0
and

(z
0
) = 1
then is linear.
Solution. Let f(z) = (z +z
0
) z
0
for z z
0
. Then f(z) z
0
and
f is linear i is. Hence, we may assume WLOG that z
0
= 0. Expanding
in a power series around 0, we have (z) = z + a
2
z
2
+ . . . . Suppose a
n
is
the rst nonzero coecient with n > 1. Then (z) = z +a
n
z
n
+O(z
n+1
).
By induction this implies that
k
(z) = (z) = z +ka
n
z
n
+O(z
n+1
);
the base case k = 1 has been established, and if it is true for k it follows
that

k+1
(z) = (z+ka
n
z
n
+O(z
n+1
))+a
n
_
z +ka
n
z
n
+O(z
n+1
)
_
n
+O
_
_
z +ka
n
z
n
+O(z
n+1
)
_
n+1
_
= z+(k+1)a
n
z
n
+O(z
n+1
).
Now let r > 0 such that z for [z[ r. By the Cauchy estimates,

k
_
(n)
(0)


n!|
k
|
r
r
n
where |
k
|
r
= sup
|z|=r
[
k
(z)[. But
k
(z) which is bounded, so
|
k
|
r
M for some constant M independent of n and k. Now (
k
)
(n)
=
kn!a
n
, so we have
kn!a
n

Mn!
r
n
a
n

M
kr
n
for all k. Letting k , we have a
n
= 0. Thus, there can be no nonzero
terms of order n > 1 in the power series expansion of , so is linear.
Exercise 11: Let f be a holomorphic function on the disc D
R0
centered at
the origin and of radius R
0
.
(a) Prove that whenever 0 < R < R
0
and [z[ < R, then
f(z) =
1
2
_
2
0
f(Re
i
)Re
_
Re
i
+z
Re
i
z
_
d.
(b) Show that
Re
_
Re
i
+r
Re
i
r
_
=
R
2
r
2
R
2
2Rr cos +r
2
.
(Hint.)
Solution.
7
(a) Starting with the RHS,
1
2
_
2
0
f(Re
i
)Re
_
Re
i
+z
Re
i
z
_
d =
1
4
_
2
0
f(Re
i
)
_
Re
i
+z
Re
i
z
+
Re
i
+ z
Re
i
z
_
d
=
1
4
_
2
0
f(Re
i
)
_
2
Re
i
Re
i
z
1 + 1
2 z
z Re
i
_
d
=
1
2i
_
2
0
f(Re
i
)
iRe
i
d
Re
i
z
+
1
2
_
2
0
f(Re
i
)
z
z Re
i
d
=
1
2i
_
2
0
f(Re
i
)
iRe
i
d
Re
i
z
+
1
2
_
2i
0
f(Re
i
)
iRe
i
Re
i
R
2
/ z
d.
The rst integral is equal to f(z) by the Cauchy integral formula, and
the latter is equal to zero since
f()
R
2
/w
is analytic on and inside the
circle of radius R.
(b) By straightforward calculations,
Re
i
+r
Re
i
r
=
Rcos() +r +iRsin()
Rcos() r +iRsin()
=
(Rcos() +r +iRsin())(Rcos() r iRsin())
(Rcos() r)
2
+R
2
sin
2
()
=
R
2
cos
2
() r
2
+R
2
sin
2
() +i(stu)
R
2
2Rr cos() +r
2
=
R
2
r
2
R
2
2Rr cos() +r
2
+ (imaginary stu).

Exercise 12: Let u be a real-valued function dened on the unit disc D.


Suppose that u is twice continuously dierentiable and harmonic, that is,
u(x, y) = 0
for all x, y D.
(a) Prove that there exists a holomorphic function f on the unit disc such
that
Re(f) = u.
Also show that the imaginary part of f is uniquely dened up to an
additive (real) constant.
(b) Deduce from this result, and from Exercise 11, the Poisson integral
representation formula from the Cauchy integral formula: If u is har-
monic in the unit disc and continuous on its closure, then if z = re
i
one has
u(z) =
1
2
_
2
0
P
r
( )u()d
where P
r
() is the Poisson kernel for the unit disc given by
P
r
() =
1 r
2
1 2r cos +r
2
.
Solution.
8
(a) If we are allowed to import a little manifold theory, we can observe
that the 1-form u
y
dx +u
x
dy is closed, because its dierential is
(u
xy
dx u
yy
dy) dx + (u
xx
dx +u
xy
dy) dy = (u
yy
+u
xx
)dx dy = 0.
Since the unit disc is simply connected, every closed form is exact.
Hence there exists a C
2
function v such that
dv = u
y
dx +u
x
dy.
But by denition, dv = v
x
dx+v
y
dy. Hence v
x
= u
y
and v
y
= u
x
, so
the function f = u + iv satises the Cauchy-Riemann equations and
is therefore holomorphic.
However, it is not necessary to import external knowledge about man-
ifold theory here. By the equality of mixed partials,

z
=
1
4
_

2
x
2
+

2
y
2
_
so if we let g = 2
u
z
, then
g
z
= 2

z

z
u = 0. Hence g is holomorphic.
By Theorem 2.1, F with F

= g. Then
Re(F)
z
=
1
2
F
z
=
1
2
g =
u
z
by
Proposition 2.3 on page 12, so Re(F) diers from u by a constant u
0
.
Then if f(z) = F(z) u
0
, f is holomorphic and Re(f) = u.
(b) By Exercise 11,
u(z) +iv(z) =
1
2
_
2
0
_
u(e
i
) +iv(e
i
)
_
Re
_
e
i
+z
e
i
z
_
d
so
u(re
i
) =
1
2
_
2
0
u(e
i
)Re
_
e
i
+re
i
e
i
re
i
_
d
=
1
2
_
2
0
u(e
i
)Re
_
e
i()
+r
e
i()
r
_
d
=
1
2
_
2
0
u(e
i
)
1 r
2
1 2r cos( ) +r
2
=
1
2
_
2
0
P
r
( )u(e
i
)d.

Exercise 13: Suppose f is an analytic function dened everywhere in C and


such that for each z
0
C at least one coecient in the expansion
f(z) =

n=0
c
n
(z z
0
)
n
is equal to 0. Prove that f is a polynomial.
Solution. First, we prove the following lemma:
Lemma 1. Let S C be a subset of the plane with no accumulation points.
Then S is at most countable.
9
Proof. For each x S, since x is not an accumulation point of S, r
x
> 0
such that B
rx
S = x. Then B
rx/2
(x) : x S is a disjoint family
of open sets; since each contains a distinct rational point, it is at most
countable. But this set is bijective with S, so S is at most countable.
Now suppose that f is not a polynomial. Then none of its derivatives can
be identically zero, because if f
(n)
were identically zero, then f
(k)
would be
zero for k n and f would be a polynomial of degree n 1. Since the
derivatives of f are entire functions that are not everywhere zero, the set of
zeros of f
(n)
has no accumulation points, so it is at most countable by the
lemma. The set of zeros of any derivative of f must then be countable since
it is a countable union of countable sets. But by hypothesis, every point
z C is a zero of some derivative of f, since if f(z) =

c
n
(z z
0
)
n
and
c
k
= 0, then
d
k
dz
k
f(z)

z0
= 0. Since C is uncountable, this is a contradiction,
so f must be a polynomial.
Exercise 14: Suppose that f is holomorphic in an open set containing the
closed unit disc, except for a pole at z
0
on the unit circle. Show that if

n=0
a
n
z
n
denotes the power series expansion of f in the open unit disc, then
lim
n
a
n
a
n+1
= z
0
.
Solution. By replacing z with z/z
0
, we may assume WLOG that z
0
= 1.
Now let be an open set containing

D such that f is holomorphic on
except for a pole at 1. Then
g(z) = f(z)
N

j=1
a
j
(z 1)
j
is holomorphic on for some N and a
1
, . . . , a
N
, where N is the order
of the pole at 1. Next, we note that must contain some disk of radius
1 + with > 0: the set z : [z[ 2 is compact, so its image under
the map z [z[ is also compact and hence attains a lower bound, which
must be strictly greater than 1 since the unit circle is contained in . Now
since g converges on the disk [z[ < 1+, we can expand it in a power series

n=0
b
n
z
n
on this disk, and we must have b
n
0. (This follows from the
fact that limsup
bn+1
bn
< 1 when the radius of convergence is greater than
1.) Now for [z[ < 1, we have

n=0
a
n
z
n
=
N

j=1
a
j
(z 1)
j
+

n=0
b
n
z
n
.
Using the fact that
1
(z 1)
j
=
(1)
j
(j 1)!
d
j1
dz
j1
1
z 1
=
(1)
j
(j 1)!

s=0
(s +j 1)!
s!
z
s
for [z[ < 1,
10
we can write

n=0
a
n
z
n
=

s=0
_
_
N

j=1
(1)
j
a
j
(j 1)!
(s +j 1)!
s!
_
_
z
s
+

n=0
b
n
z
n
a
n
= P(n) +b
n
where P(n) is a polynomial in n of degree at most N1. Here the rearrange-
ments of the series are justied by the fact that all these series converge
uniformly on compact subsets of D. Since b
n
0,
an
an+1
lim
P(n)
P(n+1)
= 1.
(Every polynomial P has the property that
P(n)
P(n+1)
1 since if the leading
coecient is c
k
n
k
,
P(n)
P(n+1)

cn
k
c(n+1)
k
= (1
1
n+1
)
k
1.)
Exercise 15: Suppose f is a non-vanishing continuous function on

D that is
holomorphic in D. Prove that if
[f(z)[ = 1 whenever [z[ = 1,
then f is constant.
Solution. Dene
F(z) =
_
_
_
f(z) [z[ 1
1
f(
1
z
)
else.
Then F is obviously continuous for [z[ < 1 and [z[ > 1; for [z[ = 1 we
clearly have continuity from the inside, and if w z with [w[ > 1, then
1
w

1
z
= z and F(w) =
1

f(
1
w
)

1

f(z)
= f(z) = F(z). Hence F is
continuous everywhere. It is known to be holomorphic for [z[ < 1. For
[z[ > 1 we can compute
f
z
= 0; alternatively, if is any contour lying in
the region [z[ > 1, let

be the image of under the map w =


1
z
. Then

is a contour lying in the region [w[ < 1 and excluding the origin from its
interior (since the point at innity does not lie within ), so
_

F(z)dz =
_

1
f( w)
dw
w
2
= 0
since
1
w
2
f( w)
is analytic on and inside

. To show F is analytic at points on


the unit circle we follow the same procedure as with the Schwarz reection
principle, by subdividing a triangle which crosses the circle into triangles
which either have a vertex on the circle or an edge lying along the circle
(i.e. a chord of the circle). In the former case we may move the vertex by
to conclude that the integral around the triangle is zero. In the case where
a side of the triangle is a chord of the circle, we subdivide into smaller
triangles (take the midpoint of the circular arc spanned by the chord) until
the chord lies within of the circle and apply the same argument. The result
is that F is entire. But F is bounded since f(

D) is a compact set which


excludes 0 and hence excludes a neighborhood of zero, so
1
f
is bounded on
D. Since F is a bounded entire function, it is constant, so f is constant.
11
Chapter 2.7, Page 67
Problem 3: Moreras theorem states that if f is continuous on C, and
_
T
f(z)dz =
0 for all triangles T, then f is holomorphic in C. Naturally, we may ask if
the conclusion still holds if we replace triangles by other sets.
(a) Suppose that f is continuous on C, and
_
C
f(z)dz = 0
or every circle C. Prove that f is holomorphic.
(b) More generally, let be any toy contour, and T the collection of all
translates and dilates of . Show that if f is continuous on C, and
_

f(z)dz = 0 for all T


then f is holomorphic. In particular, Moreras theorem holds under
the weaker assumption that
_
T
f(z)dz = 0 for all equilateral triangles.
Chapter 3.8, Page 103
Exercise 1: Using Eulers formula
sinz =
e
iz
e
iz
2i
,
show that the complex zeros of sinz are exactly at the integers, and that
they are each of order 1.
Calculate the residue of 1/ sinz at z = n Z.
Solution. Let z = a +bi and suppose sin(z) = 0. Then
0 = 2i sin((a +bi)) = (e
b
e
b
) cos(a) +i(e
b
+e
b
) sin(a)
where we have used the formula e
i
= cos() + i sin() for real . Since
e
b
and e
b
are both strictly positive, the imaginary part can only be
zero if sin(a) = 0, which happens i a Z. For the real part to be zero,
either cos(a) = 0 or e
b
= e
b
. But cos(a) = 1 when a Z, so
e
b
= e
b
b = 0. Hence z = a + bi Z. To nd the order of the zero,
we note that
2i
d
dz

z=n
sin(z) = i(e
in
+e
in
) = 2ie
in
,= 0
so the zero is of order 1. Finally, the residue at n of
1
sin(z)
is
lim
zn
(z n)
1
sin(z)
=
1
cos(n)
=
(1)
n

by LHopitals Rule.
Exercise 2: Evaluate the integral
_

dx
1 +x
4
.
Where are the poles of
1
1+z
4
?
12
Solution. Since 1 + z
4
= (z )(z
3
)(z
5
)(z
7
) where =
1+i

2
is a primitive 8th root of unity, the poles of
1
1+z
4
are at ,
3
,
5
, and

7
, of which and
3
are in the upper half plane. The residues may be
calculated in various ways; the easiest is to note that if
1
f
has a simple
pole at z
0
, the residue is
1
f

(z0)
. This follows from the observation that
if f(z) = (z z
0
)h(z), then f

(z
0
) = h(z
0
). In this case, the residues of
interest are
Res(f; ) =
1
4
3
=

5
4
and
Res(f;
3
) =
1
4
=

7
4
which add up to

5
+
7
4
=

2
4
i.
Now if we integrate f(z) =
1
1+z
4
along a semicircular contour C
R
of radius
R, the Residue Theorem guarantees that
_
C
R
f(z)dz = 2i
_

2
4
i
_
=

2
.
Since [1 +z
4
[ [z[
4
1, the integral over the curved part is at most
R
R
4
1
,
which tends to zero as R . Hence
_

dx
1 +x
4
=

2
.

Exercise 3: Show that


_

cos x
x
2
+a
2
dx =
e
a
a
, for a > 0.
Solution. Let f(z) =
e
iz
z
2
+a
2
. We integrate f around a semicircular contour
of radius R in the upper half plane. Along the curved part of the semicircle,
we note that [e
iz
[ 1 for z in the UHP, so [f(z)[
1
|z
2
+a
2
|

1
R
2
a
2
. Hence
the integral over the curved part is at most
R
R
2
a
2
0 as R . Hence
_

f(z)dz = 2i

zUHP
Res(f; z).
The residues of f are z = ai, so the only pole in the UHP is at ai with
residue
lim
zai
(z ai)
e
iz
(z ai)(z +ai)
=
e
a
2ai
.
Hence
_

f(z)dz =
e
a
a
. Taking the real parts of both sides,
_

cos x
x
2
+a
2
dx =
e
a
a
.

13
Exercise 4: Show that
_

xsinx
x
2
+a
2
dx = e
a
, for all a > 0.
Solution. Im getting rather fed up with the fact that this book doesnt even
mention the Jordan lemma, so here it is, taken from page 216 of Gamelin
(in Gamelins lingo this is actually a corollary to Jordans lemma):
Lemma 2 (Jordans Lemma). If
R
is the semicircular contour z() =
Re
i
, 0 , in the upper half-plane, and P(z) and Q(z) are polynomials
with deg Q(z) deg P(z) + 1, then
lim
R
_

R
P(z)e
iz
Q(z)
dz = 0.
Proof. By concavity, sin
2

for 0

2
. Hence
[e
iRe
i
[ = e
Rsin
e
2/
for 0

2
, and
_

0
e
Rsin
d = 2
_
/2
0
e
Rsin
d 2
_
/2
0
e
2R/
d =

R
(1 e
R
) <

R
.
Since [dz[ = R, this implies
_

R
[e
iz
[[dz[ < . Since
|P(z)|
|Q(z)|
= O(
1
R
),
_

R
P(z)e
iz
Q(z)
O(
1
R
)

R
e
iz
dz

O(
1
R
) 0.

Now back to our problem. Let f(z) =


ze
iz
z
2
+a
2
and integrate f around a
semicircular contour of radius R. By Jordans lemma, the integral over the
curved part tends to 0 as R . Hence
_

f(z)dz = 2i

zUHP
Res(f; z).
The poles of f are at z = ai; the residue at ai is
lim
zai
(z ai)
e
iz
(z ai)(z +ai)
= aie
a
2ai =
e
a
2
.
Hence
_

f(z)dz = ie
a
; taking imaginary parts,
_

xsinx
x
2
+a
2
dx = e
a
.

Exercise 5: Use contour integration to show that


_

e
2ix
(1 +x
2
)
2
dx =

2
(1 + 2[[)e
2||
for all real.
14
Solution. First, we note that if > 0, substituting u = x reveals that
_

e
2ix
(1 +x
2
)
2
dx =
_

e
2i(u)
(1 +u
2
)
2
(du) =
_

e
2iu()
(1 +u
2
)
2
du
so that we may assume WLOG that 0. We integrate f(z) =
e
2iz
(1+z
2
)
2
along a semicircular contour of radius R. Since [e
2iz
[ 1 for z UHP
and 0, the integral over the curved part is at most
2R
(R
2
1)
2
0. Hence
_

f(z)dz = 2i

zUHP
Res(f; z).
The only pole in the UHP is at z = i; because this is a double pole, the
residue is
lim
zi
d
dz
(z i)
2
f(z) = lim
zi
(2i(z +i) 2)e
2iz
(z +i)
3
=
(2 4)e
2
8i
.
Hence
_

e
2ix
(1 +x
2
)
2
dx = 2i
(2 4)e
2
8i
=

2
(1 + 2[[) e
2||
.

Exercise 6: Show that


_

dx
(1 +x
2
)
n
+ 1
=
1 3 5 (2n 1)
2 4 6 (2n)
.
Solution. Let f(z) =
1
(z
2
+1)
n+1
. Then the pole of z in the UHP is at z = i;
because this is an (n + 1)-fold pole, its residue is
1
n!
lim
zi
d
n
dz
n
1
(z +i)
n+1
=
1
n!
lim
zi
(n 1)(n 2) . . . (2n)
1
(z +i)
2n+1
=
1
n!
lim
zi
(1)
n
(2n)!
n!
1
z +i
2n+1
=
(1)
n
(2n)!
(n!)
2
1
(2i)
2n+1
=
(2n)!
(2
n
n!)
2
1
2i
=
1 3 5 (2n 1)
2 4 6 (2n)
1
2i
.
Integrating f around a semicircular contour of radius R, the integral over
the curved part is bounded by
R
(R
2
1)
n+1
0, so
_

f(z)dz = 2iRes(f; i) =
1 3 5 (2n 1)
2 4 6 (2n)
.

Exercise 7: Prove that


_
2
0
d
(a + cos )
2
=
2a
(a
2
1)
3/2
, whenever a > 1.
15
Solution. Letting z = e
i
turns this into a contour integral around the unit
circle C:
_
2
0
d
(a + cos )
2
=
_
C
dz/iz
_
a +
_
z+1/z
2
__
2
=
_
C
4zdz
i(z
2
+ 2az + 1)
2
= 2iRes
_
4z
i(z
2
+ 2az + 1)
2
; a +
_
a
2
1
_
= 2i lim
za+

a
2
1
d
dz
4z
i(z +a +

a
2
1)
2
= 2i lim
za+

a
2
1
4(a +

a
2
1 z)
i(z +a +

a
2
1)
3
= 2i
8a
i(2

a
2
1)
3
= 2a(a
2
1)
3/2
.

Exercise 8: Prove that


_
2
0
d
a +b cos
=
2

a
2
b
2
if a > [b[ and a, b R.
Solution. Again, we convert into a contour integral around the unit circle
by substituting z = e
i
:
_
2
0
d
a +b cos
=
_
C
dz/iz
a +b
_
z+1/z
2
_
2
=
_
C
2dz
i(bz
2
+ 2az +b)
= 4Res
_
1
bz
2
+ 2az +b
;
a +

a
2
b
2
b
_
= 2
1
2bz + 2a

z=
a+

a
2
b
2
b
= 2
1
2

a
2
b
2
=

a
2
b
2
.

Exercise 9: Show that


_
1
0
log(sinx)dx = log 2.
16
Solution. Consider the function f(z) = 1 e
2iz
= 2ie
iz
sin(z). (This
clever trick comes from Ahlfors, page 160.) We will integrate log f(z) along
the rectangular contour bounded by 0, 1, iY , and 1 + iY , with quarter-
circle indentations of radius at 0 and 1. On this region, e
iz
is in the
upper half plane, and sin(z) is in the right half plane. Hence, if we take
the principal branch Arg of the argument function, 0 Arg(e
iz
) < and

2
Arg(sin(z)) <

2
. Then
Arg(2i) +Arg(e
iz
) +Arg(sin(z)))
Arg(f(z)) = Arg(2i) +Arg(e
iz
) +Log(sin(z))
Log(f(z)) = Log(2i) +Log(e
iz
) +Log(sin(z))
= log 2

2
i +iz +Log(sin(z))
where Log is the principal branch of the logarithm function, corresponding
to the principal branch Arg of the argument function.
Returning to our contour integral, we note that the two vertical pieces
cancel out by the periodicity of the sine function. Moreover, the integral
over the upper half of the rectangle approaches zero as Y since for
z = t + iY with 0 t 1, e
2iz
0 f(z) 1 log f(z) 0 as Y
. Whats more, the integrals over the quarter-circle indentations also
approach zero: Since f(z) z for z near 0, log f(z) log z so the integral
over the quarter-circle near zero is O([ log )[) 0. Similar analysis holds
near . Since Logf(z) is analytic on and inside our contour, we are left
with
_
1
0
Logf(z)dz = 0.
By our above comments concerning the branches of the logarithm, this
implies
0 =
_
1
0
_
log 2

2
i +iz +Log(sin(z))
_
dz = log 2 +
_
1
0
log sin(x)dx.
Hence
_
1
0
log sin(x)dx = log 2.

Exercise 10: Show that if a > 0, then


_

0
log x
x
2
+a
2
dx =

2a
log a.
Solution. We use the branch of the logarithm function with a branch cut
along the negative imaginary axis, corresponding to a branch of the argu-
ment function such that

2
arg(z) <
3
2
. We integrate f(z) =
log z
z
2
+a
2
around an indented semicircle of radius R and indentation radius . The
integral around the outer curved part is O(
Rlog R
R
2
) 0 and the integral
17
around the indentation is O([ log [) 0. The integral along the axis is
_

R
log x
x
2
+a
2
dx +
_
R

log x
x
2
+a
2
dx
=
_
R

_
log x
x
2
+a
2
+
log(x)
x
2
+a
2
_
dx
= 2
_
R

log x
x
2
+a
2
dx +i
_
R

1
x
2
+a
2
dx
Letting R and 0 and using the fact that
_

0
dx
x
2
+a
2
=
1
a
arctan(x/a)[

0
=

2a
,
_

0
log x
x
2
+a
2
=
1
2
_
2i

zUHP
Res(f; z)
i
2
2a
_
.
The only pole in the UHP is at z = ai with residue
lim
zai
log z
z +ai
=
log(ai)
2ai
=

4a
+
log a
2ai
.
Hence
_

0
log x
x
2
+a
2
dx =
1
2
_
2i
_

4a
+
log a
2ai
_

i
2
2a
_
=
log a
2a
.

Exercise 11: Show that if [a[ < 1, then


_
2
0
log [1 ae
i
[d = 0.
Then, prove that the above result remains true if we assume only that
[a[ 1.
Solution. For [a[ < 1, 1 ae
iz
is in the right half-plane for z = +yi, 0
2, y 0. Hence the principal branch of log(1ae
i
) is analytic on this
region. We integrate log(1 ae
iz
) around a rectangular contour bounded
by 0, 2, 2 + Y i, and Y i. The integrals over the vertical portions cancel
out, and the integral over the top tends to zero because 1 ae
i(+Y i)
=
1 ae
Y
e
i
1 uniformly so log(1 ae
iz
) 0 uniformly in as Y .
Hence
0 =
_
2
0
log(1 ae
i
)d =
_
2
0
log [1 ae
i
[d +i
_
2
0
arg(1 ae
i
)d.
But the integral of the argument is 0 by symmetry, since arg(1ae
i(+)
) =
arg(1 ae
i
). Hence
_
2
0
log [1 ae
i
[d = 0
for [a[ < 1. To get the same result for [a[ = 1, I suppose some kind of
continuity argument is needed, but I dont see it.
18
Exercise 12: Suppose u is not an integer. Prove that

n=
1
(u +n)
2
=

2
(sinu)
2
by integrating
f(z) =
cot z
(u +z)
2
over the circle [z[ = R
N
= N + 1/2 (N [u[), adding the residues of f
inside the circle, and letting N tend to innity.
Solution. Just to be contrary, Ill use a square rather than a circle. (Actu-
ally, the reason is that my solution is ripped o from Schaums Outline of
Complex Variables, page 188.) Specically, let S
N
be the boundary of the
square lying inside the lines [x[ = N +
1
2
and [y[ = N +
1
2
in the complex
plane z = x +iy. When [y[ 1 we have
[ cot (x +iy)[ =
[e
ixy
+e
ix+y
[
[e
ixy
e
ix+y
[

[e
ixy
[ +[e
ix+y
[
[[e
ixy
[ [e
ix+y
[[
=
1 +e
2y
1 e
2|y|

1 +e
2
1 e
2
= C
1
.
Moreover, when [y[ 1 then x = (N +
1
2
) and
[ cot((x +iy))[ = [ cot
_
N +
1
2
+iy
_
[ = [ tanh(y)[ tanh
_

2
_
= C
2
.
Hence, [ cot z[ C on all such squares, where C = max(C
1
, C
2
) is a
universal constant. Since [
1
(z+u)
2
[ = O(
1
N
2
), and the length of the contour
is 8N + 4,
lim
N
_
S
N
cot(z)
(z +u)
2
dz = 0
by the ML estimate. The poles of f inside S
N
are at n, . . . , n with
residues
lim
zn
(z n) cos(z)
sin(z)(z +u)
2
=
cos(n)
(n +u)
2
lim
zn
z n
sin(z)
=
1
(n +u)
2
by LHopitals Rule, as well as the pole at u with residue
lim
zu
d
dz
(z +u)
2
cot(z)
(z +u)
2
= lim
zu

2
csc
2
(z) =

2
sin
2
(z)
.
By the Residue Theorem,
_
S
R
f(z)dz =
N

n=N
1
(z +u)
2


2
sin
2
(z)
.
19
As N , the LHS approaches zero and we have
N

n=N
1
(z +u)
2
=

2
sin
2
(z)
.

Exercise 13: Suppose f(z) is holomorphic in a punctured disc D


r
(z
0
) z
0
.
Suppose also that
[f(z)[ A[z z
0
[
1+
for some > 0, and all z near z
0
. Show that the singularity of f at z
0
is
removable.
Solution. Let g(z) = (z z
0
)f(z). Then g is analytic in the punctured
disc and g(z) 0 as z z
0
; since g is bounded, the singularity at z
0
is
removable, so we can take g to be analytic at z
0
. Since g is a holomorphic
function with g(z
0
) = 0, g(z) = (z z
0
)h(z) in some neighborhood of z
0
,
where h is holomorphic at z
0
. Then f(z) = h(z) in a deleted neighborhood
of z
0
, so by dening f(z
0
) = h(z
0
) we can extend f to a holomorphic
function at z
0
. Thus the singularity is removable.
Exercise 14: Prove that all entire functions that are also injective take the
form f(z) = az +b with a, b C and a ,= 0.
Solution. Dene g : C0 C by g(z) = f(
1
z
). Then g is holomorphic on
the punctured plane. It cannot have a removable singularity at 0, because
this would imply that f is bounded as [z[ and therefore constant
(so certainly not injective!) by Liouvilles theorem. Moreover, g cannot
have an essential singularity at 0: Suppose it did. Let z C 0 and
let r < [z[ and r

< [z[ r. By the Open Mapping Theorem, g(B


r
(z))
contains a neighborhood of g(z). But by the Casorati-Weierestrass theorem,
g(B
r
(0) 0) is dense and hence contains a point in said neighborhood of
g(z). This implies that g, and therefore f, is not injective. Thus, g must
have a pole at 0, so
g(z) =
a
m
z
m
+ +
a
1
z
+h(z)
where h is analytic at zero. But then
f(z) = a
m
z
m
+ +a
1
z +h(
1
z
)
cannot be analytic unless h is constant. So f is a polynomial. The only
polynomials without multiple distinct roots are powers, so f(z) = a(zz
0
)
m
for some a, z
0
C and m N. If m > 1 then f(z
0
+1) = f(z
0
+e
2i/m
) so
f is not injective. Hence m = 1 and f = a(z z
0
). Note that we cannot
have a = 0 since then f is constant and obviously not injective.
Exercise 15: Use the Cauchy inequalities or the maximum modulus principle
to solve the following problems:
20
(a) Prove that if f is an entire function that satises
sup
|z|=R
[f(z)[ AR
k
+B
for all R > 0, and for some integer k 0 and some constant A, B > 0,
then f is a polonymial of degree k.
(b) Show that if f is holomorphic in the unit disc, is bounded, and con-
verges uniformly to zero in the sector < arg z < as [z[ 1, then
f = 0.
(c) Let w
1
, . . . , w
n
be points on the unit circle in the complex plane. Prove
that there exists a point z on the unit circle such that the product of
the distances from z to the points w
j
, 1 j n, is exactly equal to 1.
(d) Show that if the real part of an entire function f is bounded, then f
is a constant.
Solution.
(a) By the Cauchy inequalities,
[f
(n)
(0)[
n!(AR
k
+B)
R
n
.
For n > k, taking the limit as R implies f
(n)
(0) = 0. Since f
is entire and all derivatives higher than k vanish, it is a polynomial of
degree at most k.
(b)
(c) Let f(z) = (z w
1
) . . . (z w
n
). Then [f(z)[ is the product of the
distances from z to the points w
1
, . . . , w
n
. Now [f(0)[ = 1 because
[w
i
[ = 1 for all i. Because f is analytic, the maximum modulus prin-
ciple guarantees that [f(z
0
)[ 1 for some z
0
with [z
0
[ = 1. Now
consider the restriction of f to the unit circle. Then [f(e
i
)[ is a con-
tinuous function of which takes on the values 0 (at each of the w
i
)
and a value at least equal to 1 (at z
0
). By the intermediate value
theorem, it is exactly equal to 1 somewhere on the unit circle.
(d) In fact, it is sucient for the real part of f to be bounded from one side
(WLOG above). Suppose Re(f) M everywhere. Let g(z) = e
f(z)
.
Then g is entire, and [g[ = e
Re(f)
e
M
. By Liouvilles theorem,
g(z) = c for some constant c. Then f(z) = log c everywhere. Although
there are multiple branches of the log function, they dier by 2i and
f is continuous, so f must be constant.

Exercise 16: Suppose f and g are holomorphic in a region containing the


disc [z[ 1. Suppose that f has a simple zero at z = 0 and vanishes
nowhere else in [z[ 1. Let
f

(z) = f(z) +g(z).


Show that if is suciently small, then
(a) f

(z) has a unique zero in [z[ 1, and


(b) if z

is this zero, the mapping z

is continuous.
Solution.
21
(a) Since the closed unit disc is compact and [g[ is continuous, it attains a
maximum M on it. Similarly, [f[ is continuous and nonzero on the unit
circle, so it has a minimum value m. Then for <
m
M
, [f(z)[ > [g(z)[
on the unit circle, so f and f

have the same number of zeros inside


the circle by Rouches theorem.
(b) Fix and z

. For r > 0, let N


r
be the open neighborhood of z

of radius r. Then [f

[
r
on D N
r
for some
r
> 0, because
D N
r
is compact and [f

[ > 0 on D z

. Then for [

[ <
r
M
,
[f

(z) f

(z)[ = [

[[g(z)[ <
r
so f

(z) ,= 0 for z D N
r
. Hence
z

N
r
provided [

[ <
r
M
.

Exercise 17: Let f be non-constant and holomorphic in an open set contain-


ing the closed unit disc.
(a) Show that if [f(z)[ = 1 whenever [z[ = 1, then the image of f contains
the unit disc. (Hint.)
(b) If [f(z)[ 1 whenever [z[ = 1 and there exists a point z
0
D such
that [f(z
0
)[ < 1, then the image of f contains the unit disc.
Solution.
(a) By Rouches theorem, f(z) and f(z) w
0
have the same number of
zeros inside the unit circle provided [w
0
[ < 1. Hence, if f has a zero,
its image includes the unit disc. If f is nonzero, then
1
f
is holomorphic,
so

1
f(z)

1 for z

D by the maximum modulus principle. But then
[f(z)[ 1 for z

D, which contradicts the open mapping theorem.
(Pick any z with [z[ = 1; then f(

D) contains a neighborhood of f(z),
which includes points w with [w[ < 1 since [f(z)[ = 1.)
(b) Let w
0
= f(z
0
) where [z
0
[ < 1 and [w
0
[ < 1. By Rouches theorem
again, f(z) and f(z) w have the same number of zeros for all w with
[w[ < 1. Since there exists a w (namely w
0
) for which f(z) w has a
zero, it has a zero for all w D. So the image of f contains D.

Chapter 3.9, Page 108


Problem 3: If f(z) is holomorphic in the deleted neighborhood 0 < [z
z
0
[ < r and has a pole of order k at z
0
, then we can write
f(z) =
a
k
(z z
0
)
k
+ +
a
1
(z z
0
)
+g(z)
where g is holomorphic in the disc [z z
0
[ < r. There is a generalization
of this expansion that holds even if z
0
is an essential singularity. This is
a special case of the Laurent series expansion, which is valid even in a
more general setting.
Let f be holomorphic in a region containing the annulus z : r
1

[z z
0
[ r
2
where 0 < r
1
< r
2
. Then,
f(z) =

n=
a
n
(z z
0
)
n
22
where the series converges absolutely in the interior of the annulus. To
prove this, it suces to write
f(z) =
1
2i
_
Cr
2
f()
z
d
1
2i
_
Cr
1
f()
z
d
when r
1
< [z z
0
[ < r
2
, and argue as in the proof of Theorem 4.4, Chapter
2. Here C
r1
and C
r2
are the circles bounding the annulus.
Chapter 5.6, Page 153
Exercise 1: Give another proof of Jensens formula in the unit disc using the
functions (called Blaschke factors)

(z) =
z
1 z
.
Solution. We use the same proof as before to establish Jensens formula for
functions with no zeros in the unit disc. Now suppose f is analytic on the
unit disc and has zeros at z
1
, . . . , z
N
, counted with multiplicity. Then the
function
g(z) =
f(z)

1
(z)
N
(z)
is analytic on the unit disc and has no zeros, where we use the notation

k
(z) =
z
k
(z). Hence
log

f(0)

1
(0)
N
(0)

=
1
2
_
2
0
log

f(e
i
)

1
(e
i
)
N
(e
i
)

d
log [f(0)[
N

k=1
log [
k
(0)[ =
1
2
_
2
0
log [f(e
i
)[d
1
2
N

k=1
_
2
0
log [
k
(e
i
)[d.
Since
k
(0) = z
k
and
[
k
(e
i
)[ =

z
k
e
i
e
i
(z
k
e
i
)

= 1,
this becomes
log [f(0)[ =
N

k=1
log [z
k
[ +
1
2
_
2
0
log [f(e
i
)[
as desired.
Exercise 3: Show that if is xed with Im() > 0, then the Jacobi theta
function
(z[) =

n=
e
in
2

e
2inz
is of order 2 as a function of z.
23
Solution. Let = s +it with s, t R; by hypothesis, t > 0. Then
[(z[)[

n=
[e
in
2

[[e
2inz
[

n=
e
n
2
t
e
2|n||z|
= 1 + 2

n=1
e
n
2
t
e
2n|z|
= 1 + 2
4|z|/t

n=0
e
n
2
t+2n|z|
+ 2

n=4|z|/t+1
e
n
2
t+2n|z|
.
For n > 4[z[/t, n
2
t + 2n[z[ < n
2
t/2, so the second sum is less than

n=4|z|/t+1
e
n
2
t/2

n=1
e
n
2
t/2
= C

.
In the rst sum, there are at most 4[z[/t terms, each of which is at most
e
24|z|/t|z|
= e
8|z|
2
/t
, so we have
[(z[)[ C
1
+C
2
[z[e
C3|z|
2
for constants C
1
, C
2
, C
3
. This implies [(z[)[ C

e
B|z|
2+
for any > 0,
so has order at most 2.
Exercise 4: Let t > 0 be given and xed, and dene F(z) by
F(z) =

n=1
(1 e
2nt
e
2iz
).
Note that the product denes an entire function of z.
(a) Show that [F(z)[ Ae
a|z|
2
, hence F is of order 2.
(b) F vanishes exactly when z = int + m for n 1 and n, m integers.
Thus, if z
n
is an enumoration of these zeros we have

1
[z
n
[
2
= but

1
[z
n
[
2+
< .
Solution.
(a) Given z, let c >
1
t
and let N = c[z[|. Let
F
1
(z) =
N

n=1
(1 e
2nt
e
2iz
), F
2
(z) =
N

n=N+1
(1 e
2nt
e
2iz
).
24
Since e
2(N+1)t
e
2|z|

1
2
by our choice of N (assuming [z[ suciently
large), and since [ log(1 +y)[ 2[y[ for [y[
1
2
, we have
[ log F
2
(z)[ =

n=N+1
log(1 e
2nt
e
2iz
)

n=N+1
[ log(1 e
2nt
e
2iz
)[

n=N+1
2[e
2nt
e
2iz
[
2

n=N+1
e
2nt
e
2|z|
= 2e
2(N+1)t+2|z|

k=0
e
2kt

k=0
e
2kt
= C
so [F
2
(z)[ is bounded by a constant. Moreover, for any n,
[1 e
2nt
e
2iz
[ 1 +e
2|z|
2e
2|z|
.
Since F
1
is the product of N such terms, we have
[F
1
(z)[ (2e
2|z|
)
N
= 2
N
e
2N|z|
2
c|z|
e
2c|z|
2
e
c

|z|
2
for an appropriate choice of c

. Hence F
1
(and therefore F because
F = F
1
F
2
and F
2
is bounded) is of order at most 2.
(b) By Proposition 3.1 the product F(z) is zero exactly when, for some
integer n 1,
e
2(nt+iz)
= 1 z = int +m
for some integer m. Thus, if z
k
is an enumeration of these zeros,

1
[z
k
[
2+
<
25
by Theorem 2.1. However,

1
[z
k
[
2
=

m=

n=1
1
m
2
+t
2
n
2
=

2
6t
2
+ 2

m,n=1
1
m
2
+n
2
t
2
2

m,n=1
1
m
2
+n
2
t
2
2
_

1
_

1
1
x
2
+t
2
y
2
dxdy
=
2
t
2
_

1
_

1
1
x
2
+u
2
dxdu
=
2
t
2
_
2
0
_

1
1
r
2
rdrd
=
4
t
2
_

1
dr
r
= ,
where the rearrangement of the double sum is allowed because all
terms are nonnegative, and the sum-integral relation follows from the
fact that
1
x
2
+y
2
is decreasing in both x and y.

Exercise 8: Prove that for every z the product below converges, and
cos(z/2) cos(z/4) cos(z/8) =

k=1
cos(z/2
k
) =
sinz
z
.
Solution. First, to show that the product converges, we note that [1
cos(t)[ t
2
for suciently small [t[; this follows from the Taylor series
expansion for cosine. Thus, for large enough k we have [1 cos(z/2
k
)[
(z/2
k
)
2
= z
2
/4
k
, and since

1/4
k
converges, Proposition 3.1 guarantees
that

cos(z/2
k
) converges as well. Now let F(z) =

k=1
cos(z/2
k
). Using
the trigonometric identity sinz = 2 cos(z/2) sin(z/2) repeatedly, we have
sinz = 2 cos(z/2) sin(z/2)
= 4 cos(z/2) cos(z/4) sin(z/4)
= . . .
= 2
N
sin(z/2
N
)
N

k=1
cos(z/2
N
)
for each N = 1, 2, . . . . If we take the limit as N , the product ap-
proaches F(z), and since sin(t) t for small [t[, we end up with
sinz = zF(z) F(z) =
sinz
z
.

26
Exercise 9: Prove that if [z[ < 1, then
(1 +z)(1 +z
2
)(1 +z
4
) =

k=0
(1 +z
2
k
) =
1
1 z
.
Solution. We rst note that the product is convergent; indeed, it converges
uniformly on the compact subdisk [z[ R < 1 by Proposition 3.2 since

[z[
2
k

R
2
k
<

R
k
=
1
1 R
< .
We next prove by induction that
(1)
N

k=0
(1 +z
2
k
) =
2
N+1
1

j=0
z
j
.
The base case N = 0 is obvious. The inductive step is
N+1

k=0
(1 +z
2
k
) = (1 +z
2
N+1
)
N

k=0
(1 +z
2
k
)
= (1 +z
2
N+1
)
2
N+1
1

j=0
z
j
=
2
N+1
1

j=0
z
j
+z
2
N+1
2
N+1
1

j=0
z
j
=
2
N+1
1

j=0
z
j
+
2
N+2
1

j=2
N+1
z
j
=
2
N+2
1

j=0
z
j
.
Since a subsequence of a convergent sequence converges to the same limit,
lim
N
2
N+2
1

j=0
z
j
= lim
M
M

j=0
z
j
=
1
1 z
.
Thus, taking the limit of both sides of (1),

k=0
(1 +z
2
k
) = lim
N
N

k=0
(1 +z
2
k
) = lim
N
2
N+1
1

j=0
z
j
=
1
1 z
.

Exercise 11: Show that if f is an entire function of nite order that omits
two values, then f is constant. This result remains true for any entire
function and is known as Picards little theorem.
Solution. Suppose f is never equal to a. Then f(z)a is an entire function
which is nowhere zero; by Theorem 6.2 of Chapter 3, this implies f(z)a =
e
g(z)
for an entire function g. If f has nite order , then [g(z)[ [z[

,
which implies g is a polynomial. Every nonconstant polynomial takes on
27
all complex values, so either g is constant (in which case f is as well) or
it takes on every complex value, which implies e
g
takes on every nonzero
value and f takes on every value other than a.
(Note: In case its not obvious that [g(z)[ [z[

implies g is a polynomial,
write g(z) = p(z) +z
k
h(z) for a polynomial p, entire function h, and power
k > . Then h is entire and bounded, hence constant.)
Exercise 13: Show that the equation e
z
z = 0 has innitely many solutions
in C.
Solution. Suppose the equation has nitely many solutions a
1
, . . . , a
N
, where
we allow the possibility N = 0. Since e
z
z has order 1, Hadamards the-
orem tells us that e
z
z = p(z)e
az+b
for some constants a, b, where
p(z) =
N

n=1
(1 z/a
n
).
Now for the equation e
z
z = p(z)e
az+b
to be true for large real values
of z, we must have a = 1, and since e
b
is a constant we can rewrite this
as e
z
z = p(z)e
z
p(z) = 1 ze
z
, a contradiction. Hence e
z
z has
innitely many zeros.
Just for fun, lets also give a semi-constructive, real-variables proof. Let
z = x +iy; then the equation e
z
= z becomes
e
x+iy
= x +iy e
x
cos y +ie
x
siny = x +iy e
x
cos y = x and e
x
siny = y.
We will narrow our search to solutions with x, y > 0. In this case the above
equations are equivalent to the system of equations x
2
+y
2
= e
2x
,
y
x
= tany.
From this we see that y =

e
2x
x
2
and tan

e
2x
x
2
=
_
e
2x
/x
2
1.
This in turn implies
(2)
_
e
2x
x
2
= arctan
_
e
2x
x
2
1 +k
for some integer k. Now

e
2x
x
2
is equal to 1 at x = 0 and tends to
innity as x . On the other hand, for k 0 the right hand side of (2)
is greater than 1 at x = 0, but is bounded. Hence, the intermediate value
theorem guarantees at least one solution of (2) for each k = 0, 1, 2, . . . .
Given any such solution x, we can let y =

e
2x
x
2
, and x +iy will be a
solution of e
z
= z.
Exercise 14: Deduce from Hadamards theorem that if F is entire and of
growth order that is non-integral, then F has innitely many zeros.
Solution. If F is entire and has nitely many zeros, Hadamards theorem
implies that F(z) = P
1
(z)e
P2(z)
for some polynomials P
1
and P
2
. But then
F would have order deg P
2
, an integer, because [P
1
(z)[ C

e
|z|

for any
> 0. Thus, if the order of F is not an integer, F must have innitely
many zeros.
Exercise 15: Prove that every meromorphic function in C is the quotient
of two entire functions. Also, if a
n
and b
n
are two disjoint sequences
having no nite limit points, then there exists a meromorphic function in
28
the whole complex plane that vanishes exactly at a
n
and has poles exactly
at b
n
.
Solution. Let f be a meromorphic function on C. Let a
n
be the poles of
f counted with multiplicity. By the Weierstrass product theorem, there
exists an entire function g with zeros exactly at a
n
. Then the product fg
is an entire function h, so f = h/g where h and g are both entire. Now
let a
n
and b
n
be two sequences with no nite limit points. Let F and G
be entire functions with zeros precisely at the a
n
and the b
n
, respectively;
such functions exist by the Weierstrass product theorem. Then the quotient
F/G has zeros exactly at the a
n
and poles exactly at the b
n
.
Exercise 16: Suppose that
Q
n
(z) =
Nn

k=1
c
n
k
z
k
are given polynomials for n = 1, 2, . . . . Suppose also that we are given a
sequence of complex numbers a
n
without limit points. Prove that there
exists a meromorphic function f(z) whose only poles are at a
n
, and so
that for each n, the dierence
f(z) Q
n
_
1
z a
n
_
is holomorphic near a
n
. In other words, f has prescribed poles and principal
parts at each of these poles. This result is due to Mittag-Leer.
Solution. (Solution adapted from Gamelin, Complex Analysis, p. 348.) Let
K
m
= B
m
(0) = z C : [z[ m. Let
f
m
(z) =

a
k
Km+1\Km
Q
k
_
1
z a
k
_
.
This is a nite sum, so f
m
is well-dened for z ,= a
1
, a
2
, . . . . By the Runge
approximation theorem (Theorem 5.7 of Chapter 2), there exist polynomials
g
m
(z) such that [f
m
(z) g
m
(z)[
1
2
m
on K
m
. Let
f(z) =

m=1
_
f
m
(z) g
m
(z)
_
.
This sum converges uniformly on compact subsets of C, by the Weierstrass
M-test. Hence f is well-dened on C and is meromorphic. Moreover, on
K
m
the tail

j=m
(f
j
(z) g
j
(z)) is analytic, whereas f
k
(z) g
k
(z) for
k < m has poles precisely at those a
j
in K
k+1
K
k
, with the prescribed
principal parts; thus, the poles of f lying in K
m
are precisely the a
k
within
K
m
, with the correct principal parts, for each m. Hence f has the desired
properties.
29
Chapter 5.7, Page 156
Problem 1: Prove that if f is holomorphic in the unit disc, bounded, and
not identically zero, and z
1
, z
2
, . . . are its zeros ([z
k
[ < 1), then

n
(1 [z
n
[) < .
Solution. By Jensens formula, we have for each R < 1

|z
k
|<R
log

R
z
k

=
_
2
0
log [f(Re
i
)[
d
2
log [f(0)[.
Because f is bounded, the right-hand side is bounded above by some con-
stant M as R varies. Suppose now we x R and let R

> R be variable.
We get
(3)

|z
k
|<R
log

z
k

|z
k
|<R

log

z
k

< M
since the rst sum is a partial sum of the second and all terms are positive.
Since the rst sum in (3) is nite, we can let R

1 and get

|z
k
|<R
log

1
z
k

M.
This is true for all R < 1, so letting R 1 we have

k
log

1
z
k

M.
(If all the partial sums are at most M, the innite sum is as well.) Now
1 x log x for all real x > 0, so

k
1 [z
k
[

k
log

1
z
k

M < .

Chapter 6.3, Page 174


Exercise 1: Prove that
(s) = lim
n
n
s
n!
s(s + 1) (s +n)
whenever s ,= 0, 1, 2, . . . .
30
Solution.
1
(s)
= se
s

n=1
_
1 +
s
n
_
e
s/n
= s
_
lim
N
e
s(

N
n=1
1/nlog N)
_
_
lim
N
N

n=1
_
1 +
s
N
_
e
s/n
_
= lim
N
sN
s
N

n=1
e
s/n
N

n=1
_
n +s
n
_
e
s/n
= lim
N
sN
s
N

n=1
_
n +s
n
_
= lim
N
s(s + 1) (s +N)
N
S
N!
.

Exercise 3: Show that Walliss product formula can be written as


_

2
= lim
n
2
2n
(n!)
2
(2n + 1)!
(2n + 1)
1/2
.
As a result, prove the following identity:
(s)(s + 1/2) =

2
12s
(2s).
Solution. Wallis product formula says

2
= lim
n
n

k=1
(2k)
2
(2k + 1)(2k 1)
.
31
Now
n

k=1
(2k 1)(2k + 1) =
n

k=1
(2k 1)
n

k=1
(2k + 1)
=
_
_
_
_
2n1

j=1
j
j odd
_
_
_
_
_
_
_
_
2n+1

j=1
j
j odd
_
_
_
_
= (2n + 1)
_
_
_
_
2n1

j=1
j odd
j
_
_
_
_
2
= (2n + 1)
_
_
_
_
(2n)!

2n
j=1
j
j even
_
_
_
_
2
= (2n + 1)
_
(2n)!

n
k=1
2k
_
2
= (2n + 1)
_
(2n)!
n!2
n
_
2
whereas
n

k=1
(2k)
2
= 2
2n
n

k=1
k
2
= 2
2n
(n!)
2
.
Hence

2
= lim
n
2
2n
(n!)
2
(2n + 1)
((2n)!)
2
(n!)
2
2
2n
= lim
n
2
4n
(n!)
4
(2n + 1)((2n)!)
2
= lim
n
2
4n
(n!)
4
(2n + 1)
((2n + 1)!)
2
and the result follows by taking square roots of both sides.
Using the result of Problem 1,
(2s)
(s)(s + 1/2)
= lim
s(s + 1) (s +n)
n
s
n!
lim
(s + 1/2) (s + 1/2 +n)
n
s+1/2
n!
lim
(2n + 1)
2s
(2n + 1)!
(2s)(2s + 1) (2s + 2n + 1)
= lim
s(s + 1/2)(s + 1)(s + 3/2) (s +n + 1/2)(2n + 1)
2s
(2n + 1)!
n
2s+1/2
(n!)
2
(2s)(2s + 1) (2s + 2n + 1)
= lim
2s(2s + 1) (2s + 2n + 1)
2
2n+2
(2n + 1)
2s
(2n + 1)!
n
2s+1/2
(n!)
2
2s(2s + 1) (2s + 2n + 1)
= lim
_
2n + 1
n
_
2s
_
2n + 1
n
(2n + 1)!
2
2n+2
(n!)
2

2n + 1
= 2
2s

2
1
4
_

2
= 2
2s1

32
Exercise 5: Use the fact that (s)(1 s) = / sins to prove that
[(1/2 +it)[ =
_
2
e
t
+e
t
, whenever t R.
Solution. Using the trigonometric identity sin( +/2) = cos(),
(1/2 +it)(1/2 it) =

sin(1/2 +it)
=

cos(it)
=

cosht
=
2
e
t
+e
t
.
Using the fact that (z) = ( z), which follows from the meromorphicity of
,
[(1/2 +it)[
2
= (1/2 +it)(1/2 it) =
2
e
t
+e
t
and the result follows by taking square roots of both sides.
Exercise 7: The Beta function is dened for Re() > 0 and Re() > 0 by
B(, ) =
_
1
0
(1 t)
1
t
1
dt.
(a) Prove that B(, ) =
()()
(+)
.
(b) Show that B(, ) =
_

0
u
1
(1+u)
+
du.
Solution.
(a)
()() =
__

0
x
1
e
x
dx
___

0
y
1
e
y
dy
_
=
_

0
_

0
x
1
y
1
e
(x+y)
dxdy.
Making the change of variables u = x + y, v =
x
x+y
, we have x = uv
and y = u(1 v), so
(x, y)
(u, v)
=

v u
1 v u

= u,
so
()() =
_

0
_
1
0
(uv)
1
(u(1 v))
1
e
u
udvdu
=
__

0
u
+1
e
u
du
___
1
0
v
1
(1 v)
1
dv
_
= ( +)B(, ).
(b) We make the change of variables u =
1
t
1, so t =
1
u+1
, 1 t =
u
u+1
,
and dt =
du
(u+1)
2
. Then
B(, ) =
_
1
0
(1 t)
1
t
1
dt
=
_
0

_
u
u + 1
_
1
_
1
u + 1
_
1
du
(u + 1)
2
=
_

0
u
1
u
+
du.

33
Exercise 9: The hypergeometric series F(, , ; z) was dened in Exercise
16 of Chapter 1. Show that
F(, , ; z) =
()
()( )
_
1
0
t
1
(1 t)
1
(1 zt)

dt.
Solution.
()
()( )
_
1
0
t
1
(1 t)
1
(1 zt)

dt
=
()
()( )
_
1
0
t
1
(1 t)
1
_
1 +

n=1
()( 1) ( n 1)
n!
(zt)
n
_
dt
=
()
()( )
_
B(, ) +

n=1
_
1
0
t
1
(1 t)
1
( + 1) ( +n 1)
n!
z
n
t
n
dt
_
= 1 +

n=1
()
()( )
B(n +, )
( + 1) ( +n 1)
n!
z
n
= 1 +

n=1
()(n +)
()(n +)
( + 1) ( +n 1)
n!
z
n
= 1 +

n=1
()( + 1) ( +n 1)()
()( + 1) ( +n 1)()
( + 1) ( +n 1)
n!
z
n
= 1 +

n=1
( + 1) ( +n 1)( + 1) ( +n 1)
n!( + 1) ( +n 1)
z
n
= F(, , ; z)
where the sum and integral can be interchanged because all terms are non-
negative; here we have used the identity (s + 1) = s(s) as well as the
properties of the beta function derived in Exercise 7. Since (1 w)

is
holomorphic for w in the plane slit along the ray [1, ), Theorem 5.4 of
Chapter 2 guarantees that the integral representation above is holomor-
phic for z in the same slit plane, yielding an analytic continuation of the
hypergeometric function.
Exercise 10: An integral of the form
F(z) =
_

0
f(t)t
z1
dt
is called a Mellin transform, and we shall write /(f)(z) = F(z). For
example, the gamma function is the Mellin transform of the function e
t
.
(a) Prove that
/(cos)(z) =
_

0
cos(t)t
z1
dt = (z) cos
_

z
2
_
for 0, Re(z) < 1,
and
/(sin)(z) =
_

0
sin(t)t
z1
dt = (z) sin
_

z
2
_
for 0 < Re(z) < 1.
34
(b) Show that the second of the above identities is valid in the larger strip
1 < Re(z) < 1, and that as a consequence, one has
_

0
sinx
x
dx =

2
and
_

0
sinx
x
3/2
dx =

2.
Solution.
(a) Let C
R
be the contour in the rst quadrant bounded by the quarter-
circles of radius R and 1/R and the axes. Let f(w) = e
w
w
z1
. In
evaluating this integral we will use the fact that [w
z1
[ C[[w[
z1
[;
this follows from writing z 1 = x + iy and w = Re
i
, from which
[w
z1
[ = R
x
e
y
CR
x
= C[[w[
z1
[ where C = e
/2|Im(z)|
. The
integral of f around the quarter-circle of radius 1/R tends to 0 since
[f(w)[ e
Re(w)
C[w[
Re(z)1
Ce
1/R
R
1
on this segment, where
= Re(z). The length of the segment is

2R
, so by the ML estimate the
integral is at most
Ce
1/R
2R

, which tends to 0 as R . The integral


over the outer quarter-circle also tends to zero by the Jordan lemma.
Since Stein and Shakarchi, annoyingly enough, never mention the Jor-
dan lemma, Ill prove it from scratch all over again, just like I did a
half-dozen times last quarter in the section on contour integrals. On
the outer quarter circle, [f(w)[ CR
1
[e
Re
i
[ = CR
1
e
Rcos
.
Now cos
2

1 for 0

2
, so

_
/2
0
f(Re
i
)Re
i
d

_
/2
0
[f(Re
i
)Re
i
[d
CR
1
_
/2
0
e
Rcos
Rd
CR

_
/2
0
e
R(2/1)
d
= CR

e
R
e
2R/
2R

/2
0
= CR


2R
_
1 e
R
_
which tends to 0 as R since < 1. Now f is analytic on and
inside the contour, so were left with
0 =
_

0
e
t
t
z1
dt +
_
0

e
iu
(iu)
z1
idu
(z) =
_

0
(cos u i sinu)u
z1
i
z
du
i
z
(z) =
_

0
(cos u i sinu)u
z1
du

_
cos
_
z
2
_
i sin
_
z
2
__
(z) = /(cos)(z) i/(sin)(z).
For real z, we can compare real and imaginary parts to conclude that
/(cos)(z) = (z) cos
_
z
2
_
and /(sin)(z) = (z) sin
_
z
2
_
.
35
By analytic continuation, these relations both hold for z in the strip
0 < Re(z) < 1.
(b) The right-hand side of the above equation for /(sin) is analytic for
1 < Re(z) < 1 because the zero of sine cancels the pole of at the
origin. The left-hand side is also analytic on this strip by Theorem
5.4 of chapter 2; the integral converges near 0 because sint t and
converges at because Re(z) < 1. Hence, by analytic continuation,
/(sin)(z) = (z) sin
_
z
2
_
for 1 < Re(z) < 1.
Letting z = 0 we get
/(sin)(0) =
_

0
sint
t
dt = (0) sin
0
2
.
In order to evaluate the right-hand side we rearrange the functional
equation of to read

(1 s)
= (s) sin(s) = 2(s) sin(s/2) cos(s/2) (s) sin(s/2) =

2(1 s) cos(s/2)
.
This equals

2
when s = 0, so
_

0
sin t
t
dt =

2
.
Letting z = 1/2 in the Mellin transform,
_

0
sint
t
3/2
dt =
_
1
2
_
sin
_

4
_
= (2

)
_

2
_
=

2.

Exercise 13: Prove that


d
2
log (s)
ds
2
=

n=0
1
(s +n)
2
whenever s is a positive number. Show that if the left-hand side is inter-
preted as (

/)

, then the above formula also holds for all complex numbers
s with s ,= 0, 1, 2, . . . .
Solution. For positive s we can take the logarithm of the Hadamard fac-
torization
1
(s)
= se
s

n=1
_
1 +
s
n
_
e
s/n
to obtain
log (s) = log s +s +

n=1
_
log
_
1 +
s
n
_

s
n
_
.
Dierentiating,

d log (s)
ds
=
1
s
+ +

n=1
_
1
n +s

1
n
_
where the termwise dierentiation is justied because the dierentiated sum
converges uniformly on compact intervals for s. Dierentiating again,

d
2
log (s)
ds
2
=
1
s
2
+

n=1
1
(n +s)
2
=

n=0
1
(n +s)
2
.
36
The right-hand side denes an analytic function of s on the region s ,=
0, 1, 2, . . . because the sum converges uniformly on compact subsets of
this region. Moreover, for positive s the second derivative on the left is
equal to (

/)

, which is analytic on the same region since is analytic


and nonzero there. Hence the above relation holds throughout this region
by analytic continuation.
Exercise 14: This exercise gives an asymptotic formula for log n!. A more
rened formula for (s) as s (Stirlings formula) is given in Appendix
A.
(a) Show that
d
dx
_
x+1
x
log (t)dt = log x, for x > 0,
and as a result
_
x+1
x
log (t)dt = xlog x x +c.
(b) Show as a consequence that log (n) nlog n as n . In fact,
prove that log (n) nlog n +O(n) as n .
Solution.
(a) By the Fundamental Theorem of Calculus,
d
dx
_
x+1
x
log (t)dt = log (x + 1) log (x) = log
(x + 1)
(x)
= log x.
Integrating both sides with respect to x,
_
x+1
x
log (t)dt = xlog x x +c.
(b) Since (t) is increasing for t 2,
log (n)
_
n+1
n
log (t)dt log (n + 1) = log n + log (n)
(n 1) log n n +c log (n) nlog n n +c.
Thus, log (n) = nlog n n +o(log n).

Exercise 15: Prove that for Re(s) > 1,


(s) =
1
(s)
_

0
x
s1
e
x
1
dx.
37
Solution. Using the geometric series
1
e
x
1
= e
x 1
1e
x
=

n=1
e
nx
, we
have
_

0
x
s1
e
x
1
dx =
_

0
x
s1

n=1
e
nx
dx
=

n=1
_

0
x
s1
e
nx
dx
=

n=1
_

0
_
t
n
_
s1
e
t
dt
n
=

n=1
1
n
s
_

0
t
s1
e
t
dt
= (s)(s).
Here the sum and integral can be interchanged because all terms are non-
negative.
Chapter 7.3, Page 199
Exercise 1: Suppose that a
n

n=1
is a sequence of real numbers such that
the partial sums
A
n
= a
1
+ +a
n
are bounded. Prove that the Dirichlet series

n=1
a
n
n
s
converges for Re(s) > 0 and denes a holomorphic function in this half-
plane.
Solution. Using summation by parts,
N

n=1
a
n
n
s
=
A
N
N
s

N

n=1
A
n
_
1
(n + 1)
s

1
n
s
_
.
Taking the limit as N , the rst term on the right vanishes and

n=1
a
n
n
s
=

n=1
A
n
_
1
(n + 1)
s

1
n
s
_
.
To prove that this converges, we note that by hypothesis [A
n
[ M for
some constant M. Now

1
(n + 1)
s

1
n
s

_
n+1
n
1
st
s+1
dt

[(n + 1) n[ max
x(n,n+1)

1
st
s+1

1
n
+1
38
where = Re(s). Thus, the tails of the series are dominated by

n=N
M
n
+1
which converges, so the series converges. In addition, this shows that the
convergence is uniform on closed half-planes
0
> 0, so the series
denes a holomorphic function on the right half-plane.
Exercise 2: The following links the multiplication of Dirichlet series with the
divisibility properties of their coecients.
(a) Show that if a
m
and b
k
are two bounded sequences of complex
numbers, then
_

m=1
a
m
m
s
__

k=1
b
k
k
s
_
=

n=1
c
n
n
s
where c
n
=

mk=n
a
m
b
k
.
The above series converges absolutely when Re(s) > 1.
(b) Prove as a consequence that one has
((s))
2
=

n=1
d(n)
n
s
and (s)(s a) =

n=1

a
(n)
n
s
for Re(s) > 1 and Re(s a) > 1, respectively. Here d(n) equals the
number of divisors of n, and
a
(n) is the sum of the ath powers of
divisors of n. In particular, one has
0
(n) = d(n).
Solution.
(a) For = Re(s) > 1, the two sums on the left both converge absolutely
since

m=1

a
m
m
s

m=1
A
m

<
where A is a bound for [a
m
[. Thus, Fubinis theorem allows us to write
the product as the double sum

m,k
a
m
m
s
b
k
k
s
=

m,k
a
m
b
k
(mk)
s
where the terms can be summed in any order. In particular, we can
group them according to the product mk to obtain

n=1

m,k
mk=n
a
m
b
k
n
s
=

n=1
c
n
n
s
.
(b) If we let a
m
= b
k
= 1 for all k, m above, then c
n,k
=

mk=n
1 = d(n)
and we have
((s))
2
=
_

m=1
1
m
s
__

k=1
1
k
s
_
=

n=1
d(n)
n
s
.
Assuming Re(a) > 0 (which presumably was intended in the problem
statement) and Re(s a) > 1, we can replace s with s a in part (a)
39
and let a
m
= m
a
, b
k
= 1. The result is
(s)(s a) =
_

m=1
m
a
m
sa
__

k=1
1
k
sa
_
=

n=1
c
n
n
sa
=

n=1
n
a
c
n
n
s
where c
n
=

mk=n
m
a
=
1
n
a

mk=n
k
a
=
a
(n).

Exercise 3: In line with the previous exercise, we consider the Dirichlet series
for 1/.
(a) Prove that for Re(s) > 1,
1
(s)
=

n=1
(n)
n
s
,
where (n) is the Mobius function dened by
(n) =
_

_
1 if n = 1,
(1)
k
if n = p
1
p
k
, and the p
j
are distinct primes,
0 otherwise.
Note that (nm) = (n)(m) whenever n and m are relatively prime.
(b) Show that

k|n
(k) =
_
1 if n = 1,
0 otherwise.
Solution.
(a) Consider the nite product
N

n=1
_
1
1
p
s
n
_
.
Applying the distributive law, this is equal to

n=1

N
(n)
n
s
where
N
(n) =
_

_
1 n = 1
(1)
k
n = p
1
p
k
and 1, 2, . . . , k N
0 else.
Note that
N
(n) = (n)
N
(n) where
N
(n) = 1 if n has no prime fac-
tors larger than p
N
, and 0 otherwise. Hence, this sum is a (rearranged)
partial sum of

n=1
(n)
n
s
. Because this latter sum is absolutely con-
vergent for > 1, we can take the limit as N to obtain
1
(s)
=

n=1
_
1
1
p
s
n
_
=

n=1
(n)
n
s
.
An alternate proof uses the fact that

d|n
(d) = 0 for n > 1. One
way to establish this is as follows: Let p
1
, . . . , p
k
be the distinct prime
factors of n. For each j = 1, ,k, there will be
_
k
j
_
squarefree divisors of
40
n which have j distinct prime factors, and (d) = (1)
j
if d is any of
these divisors. Hence the sum of over all divisors is
k

j=0
_
k
j
_
(1)
j
= (1 1)
k
= 0
by the binomial theorem. Given this, we can use the previous exercise
to write
_

m=1
1
m
s
__

k=1
(k)
k
s
_
=

n=1
c
n
n
s
where
c
n
=

mk=n
(k) =
_
1 n = 1
0 else.
Thus,
(s)

n=1
(n)
n
s
= 1

n=1
(n)
n
s
=
1
(s)
.
(b) This was already proved in part (a) using the binomial theorem. Al-
ternatively, if (a) was established using the Euler product, then mul-
tiplying the series for and 1/ using Exercise 2 yields part (b) as a
result. Thus, one can use (b) (proved using the binomial theorem) and
Exercise 2 to establish (a), or one can use (a) (proved using the Euler
product) and Exercise 2 to establish (b).

Exercise 5: Consider the following function

(s) = 1
1
2
s
+
1
3
s
=

n=1
(1)
n+1
n
s
.
(a) Prove that the series dening

(s) converges for Re(s) > 0 and denes
a holomorphic function in that half-plane.
(b) Show that for s > 1 one has

(s) = (1 2
1s
)(s).
(c) Conclude, since

is given as an alternating series, that has no zeros
on the segment 0 < < 1. Extend this last assertion to = 0 by
using the functional equation.
Solution.
(a) We rewrite the series as

(s) =

n=1
_
1
(2n 1)
s

1
(2n)
s
_
.
41
Now

1
(2n 1)
s

1
(2n)
s

_
2n
2n1
d
dt
t
s
dt

_
2n
2n1

(s)t
s1

dt
= [s[
_
2n
2n1
t
1
dt
[s[(2n 1)
1
.
Since

n=1
[s[(2n 1)
1
converges for > 0 and converges uniformly on closed sub-half-planes

0
> 0, the series dening

denes a holomorphic function on
the right half-plane. (To complete the proof we should also note that
the individual terms
(1)
n+1
n
s
tend to 0.)
(b) For s > 1,
2
1s
(s) = 2

n=1
2
s
n
s
= 2

n even
1
n
s
and the absolute converge of the series for on closed half-planes

0
> 1 allows us to rearrange terms and obtain
(1 2
1s
)(s) =

n
1
n
s
2

n even
1
n
s
=

n
(1)
n+1
n
s
=

(s).
(c) I assume the problem statement intends to say 0 < s < 1 (i.e. s
real), since otherwise the zeros of in the critical strip would create a
contradiction. For real s the terms
_
1
(2n 1)
s

1
(2n)
s
_
are all strictly positive, so the sum cannot be zero. Taking the limit
of both sides of the equation

(s) = (1 2
1s
) as s 0, the left-hand
side becomes the alternating series

(0) = 1
1
2
+
1
3
= log 2,
so the right-hand side cannot be zero.

Exercise 10: In the theory of primes, a better approximation to (x) (instead


of x/ log x) turns out to be Li(x) dened by
Li(x) =
_
x
2
dt
log t
.
(a) Prove that
Li(x) =
x
log x
+O
_
x
(log x)
2
_
as x ,
42
and that as a consequence
(x) Li(x) as x .
(b) Rene the previous analysis by showing that for every integer N > 0
one has the following asymptotic expansion
Li(x) =
x
log x
+
x
(log x)
2
+ 2
x
(log x)
3
+ + (N 1)!
x
(log x)
N
+O
_
x
(log x)
N+1
_
Solution.
(a) Substituting u =
1
log t
and v = t in the denition of Li(x) and inte-
grating by parts,
Li(x) =
_
x
2
dt
log t
=
t
log t

x
2
+
_
x
2
dt
(log t)
2
=
x
log x

2
log 2
+
_
x
2
dt
(log t)
2
.
We will estimate the latter integral in two pieces, one from 2 to

x
and one from

x to x. The graphs of the functions f(t) = (log t)
2
and g(t) =
(log

x)
2

x1
(t 1) both pass through the points (1, 0) and
(

x, (log

x)
2
; by concavity, f g for 1 t

x. Taking recipro-
cals,
1
(log t)
2

(

x 1)
(log

x)
2
1
t 1

_
x
2
1
(log t)
2
dt
_
x
2
(

x 1)
(log

x)
2
1
t 1
dt =
(

x 1) log(

x 1)
(log

x)
2

x
log

x
= O
_
x
(log x)
2
_
.
For the integral from

x to x it suces to approximate the integrand
by a constant:
_
x

x
dt
(log t)
2
(x

x)
1
(log

x)
2

x
_
1
2
log x
_
2
= O
_
x
(log x)
2
_
.
Putting the pieces together, we have
_
x
2
dt
log t
=
x
log x

2
log 2
+O
_
x
(log x)
2
_
=
x
log x
+O
_
x
(log x)
2
_
.
(b) Integrating by parts with the same substitution as above, we have
more generally
_
x
2
dt
(log t)
2
=
t
(log t)
k

x
2
+k
_
x
2
dt
(log t)
k+1
.
This plus an easy induction yields
Li(x) =
N

k=1
(k 1)!
t
(log t)
k

x
2
+
_
x
2
dt
(log t)
N+1
for each N = 1, 2, . . . . Evaluation at the lower terms yields a constant,
so we can write this as
Li(x) = C
N
+
N

k=1
(k 1)!
x
(log x)
k
+
_
x
2
dt
(log t)
N+1
.
43
We can estimate this integral in the same manner as before. By con-
cavity,
(log t)
N+1

(log

x)
N+1

x 1
(t 1) for 1 t

1
(log t)
N+1

x 1
(log

x)
N+1
1
t 1
for 1 t

_

x
2
dt
(log t)
N+1

x 1
(log

x)
N+1
_

x
2
dt
t 1
=
(

x 1) log(

x 1)
(log

x)
N+1
= O
_
x
(log x)
N+1
_
.
Also,
_
x

x
dt
(log t)
N+1
(x

x)
1
(log

x)
N+1
= O
_
x
(log x)
N+1
_
so that
_
x
2
dt
(log t)
N+1
= O
_
x
(log x)
N+1
_
Li(x) =
N

k=1
(k1)!
x
(log x)
k
+O
_
x
(log x)
N+1
_
.

Chapter 8.5, Page 248


Exercise 5: Prove that f(z) =
1
2
(z + 1/z) is a conformal map from the
half-disc z = x +iy : [z[ < 1, y > 0 to the upper half-plane.
Solution. Clearly f is holomorphic on the upper half-disc U. To show it is
injective, consider that if f(z) = f(z

) = w, then z and z

are both roots of


the equation t
2
+ 2tw + 1 = 0; the product of the roots of this equation is
1, so only one of the roots can have norm less than 1. So f is injective on
U. It maps U into the upper half-plane H because if z U, then
Im((z + 1/z)) = Im(z) Im(1/z) = Im(z)
1
[z
Im(z) =
_
1
[z[
1
_
Im(z),
and since [z[ < 1 this is positive. Finally, f is surjective from U onto H
because for any w H, the equation w =
z+1/z
2
, equivalent to z
2
+2zw+1,
has two roots with product 1, so one is inside the disc and one outside.
(They cannot both be on the disc because z + 1/z is real for [z[ = 1.) Let
z
0
be the root inside the disc. Then
Im(w) =
_
1
[z
0
[
1
_
Im(z
0
) > 0 Im(z
0
) > 0
so z
0
is in U.
Exercise 8: Find a harmonic function u in the open rst quadrant that ex-
tends continuously up to the boundary except at the points 0 and 1, and
that takes on the following boundary values: u(x, y) = 1 on the half-lines
y = 0, x > 1 and x = 0, y > 0, and u(x, y) = 0 on the segment
0 < x < 1, y = 0.
44
Solution. We follow the outline given in Figure 11. Dene F
1
: FQ
UHD, where FQ is the (open) rst quadrant and UHD the upper half-
disc, by F
1
(z) =
z1
z+1
. This is the inverse of the map in Example 3 on
page 210. The boundary is mapped as follows: y = 0, x > 1 is mapped
to the positive real axis, 0 < x < 1, y = 0 to the negative real axis,
and x = 0, y > 0 to the semicircular part of the boundary. Next, dene
F
2
: UHD LHS, where LHS is the left half-strip x < 0, 0 < y <
by F
2
(z) = log z, using the principal branch. Here the positive real axis is
mapped to the negative real axis, the negative real axis to the line x <
0, y = i, and the semicircular part to the segment of the imaginary axis
from 0 to i. Next, we take the map F
3
: LHS UHS, where UHS
is the upper half-strip /2 < x < /2, y > 0, where F
3
(z) =
z
i


2
.
This takes the imaginary segment to the real segment, upper and lower
boundaries of the strip to the left and right boundaries respectively. Next,
dene F
4
: UHS U, where U is the upper half-plane, by F
4
(z) =
1+sin(z)
2
,
which is conformal on this domain by Example 8 on page 212. This takes
the real segment to the segment between 0 and 1, and the left and right
boundaries to the rays x < 0, y = 0 and x > 1, y = 0 respectively.
Finally, let F
5
: U U be dened by F
5
(z) = z 1. Now since
1

arg(z) is
harmonic on U and equals 0 on the positive real axis and 1 on the negative
real axis, the composition z
1

arg(F
5
(F
4
(F
3
(F
2
(F
1
(z)))))) is harmonic
on the rst quadrant and has the desired boundary values.
Exercise 9: Prove that the function u dened by
u(x, y) = Re
_
i +z
i z
_
and u(0, 1) = 0
is harmonic in the unit disc and vanishes on its boundary. Note that u is
not bounded in D.
Solution. The real part of an analytic function is harmonic, and
i+z
iz
is
analytic on the open unit disc, so u is harmonic in D. Moreover, on the
boundary points other than (0, 1), write z = cos +i sin; then
i +z
i z
=
cos +i(1 + sin)
cos +i(1 sin)

cos +i(sin 1)
cos +i(sin 1)
=
2i cos
cos
2
+ (1 sin)
2
is pure imaginary, so its real part is zero.
Exercise 10: Let F : H C be a holomorphic function that satises
[F(z)[ 1 and F(i) = 0.
Prove that
[F(z)[

z i
z +i

for all z H.
Solution. Dene G : D D by
G(w) = F
_
i
1 w
1 +w
_
.
45
Then G is holomorphic and G(0) = F(i) = 0. By the Schwarz lemma,
[G(w)[ [w[ for all w D. Then for any z H,
[F(z)[ =

G
_
i z
i +z
_

z i
z +i

Exercise 11: Show that if f : D(0, R) C is holomorphic, with [f(z)[ M


for some M > 0, then

f(z) f(0)
M
2
f(0)f(z)

[z[
MR
.
Solution. For z D, let g(z) =
f(Rz)
M
. Since Rz D(0, R), [f(Rz)[ M
so g(z) D. Thus g : D D and is holomorphic. Let = g(0) =
f(0)
M
.
Then

g : D D satises

(g(0)) = 0, where

(w) =
w
1 w
. By the
Schwarz lemma,
[

g()[ [[ ( D)

g()
1 g()

[[

f(0)
M

f(R)
M
1
f(0)
M
f(R)
M

[[
M

f(0) f(R)
M
2
f(0)f(R)

[[

f(0) f(z)
M
2
f(0)f(z)

[z[
MR
where z = R D(0, R).
Exercise 13: The pseudo-hyperbolic distance between two points z, w
D is dened by
(z, w) =

z w
1 wz

.
(a) Prove that if f : D D is holomorphic, then
(f(z), f(w)) (z, w) for all z, w D.
Moreover, prove that if f is an automorphism of D then f preserves
the pseudo-hyperbolic distance
(f(z), f(w)) = (z, w) for all z, w D.
(Hint.)
(b) Prove that
[f

(z)[
1 [f(z)[
2

1
1 [z[
2
for all z D.
Solution.
46
(a) Let

(z) =
z
1 z
.
Then it is easy to check that

is an automorphism of D; in fact, its


inverse can be explicitly computed as

(z) =
z +
1 + z
.
Now let g =
f(w)
f
1
w
. Then
g(0) =
f(w)
_
f
_

1
w
(0)
__
=
f(w)
(f (w)) = 0
and since g is the composition of three functions which map D into D,
so it also maps D into D. By the Schwarz Lemma, [g(y)[ [y[ for all
y D. In particular, if y =
w
(z) we have
[g(y)[ [y[

f(w)
(f(
1
w
(y)))

[
w
(z)[

f(w)
(f(z))

z w
1 wz

f(z) f(w)
1 f(w)f(z)

z w
1 wz

(f(z), f(w)) (z, w)


as desired. Moreover, if f is an automorphism of D then by Theorem
2.2,
f(z) = e
i
z
1 z
for some R and D. Then
(f(z), f(w)) =

f(z) f(w)
1 f(w)f(z)

e
i
_
z
1 z

w
1 w
_

1 e
i
w
1 w
e
i
z
1 z

(zw)(1||
2
)
(1 z)(1 w)

(1||
2
)(1 wz)
(1 w)(1 z)

=
[z w[
[1 wz[
= (z, w)
so f preserves pseudo-hyperbolic distance.
47
(b) By a simple rearrangement,
(f(w), f(z)) (w, z)

[f(w) f(z)[
[1 f(w)f(z)

[w z[
1 wz

f(w) f(z)
w z

[1 f(w)f(z)[
1 wz[
.
Taking the limit as w z, we have
[f

(z)[
1 [f(z)[
2
1 [z[
2

[f

(z)[
1 [f(z)[
2

1
1 [z[
2
.

Exercise 14: Prove that all conformal mappings from the upper half-plane
H to the unit disc D take the form
e
i
z
z

, R and H.
Solution. Let g : H D be a conformal mapping. Let : D H be the
conformal mapping dened by
z = (w) = i
1 w
1 +w
.
As shown in section 1.1, is a conformal mapping with inverse w = (z) =
iz
i+z
. Then g : D D is a conformal automorphism of the disc, so by
Theorem 2.2 there exist R and D such that
g(z) = g
_
i
1 w
1 +w
_
= e
i
w
1 w
= e
i

iz
i+z
1
iz
i+z
= e
i
z(1 +) i(1 )
z(1 +) i(1 )
= e
i
z i
1
1+
z
1+
1+
+i
1
1+
= e
i
z
e
i
(z

)
= e
i
z
z

where = R, e
i
=
1+
1+
has unit length because conjugation
preserves norm, and = i
1
1+
H because = () and D.
Exercise 15: Here are two properties enjoyed by automorpisms of the upper
half-plane.
(a) Suppose is an automorphism of H that xes three distinct points on
the real axis. Then is the identity.
48
(b) Suppose (x
1
, x
2
, x
3
) and (y
1
, y
2
, y
3
) are two pairs of three distinct
points on the real axis with
x
1
< x
2
< x
3
and y
1
< y
2
< y
3
.
Prove that there exists (a unique) automorphism of H so that
(x
j
) = y
j
, j = 1, 2, 3. The same conclusion holds if y
3
< y
1
< y
2
or
y
2
< y
3
< y
1
.
Solution.
(a) By Theorem 2.4, there exist a, b, c, d R with ad bc = 1 and
(z) =
az +b
cz +d
.
Suppose xes x R. Then
x =
ax +b
cx +d
cx
2
+ (d a)x b = 0.
If c ,= 0 this equation has at most 2 distinct solutions; if c = 0 but
a ,= d it has only one. For it to have three or more both of these
conditions must fail, so c = 0 and a = d; the equation then becomes
b = 0, and the condition ad bc = 1 then implies a = d = 1, so
(z) =
z
1
= z.
(b) Let x
i
and y
i
be so chosen. The system of equations
ax
i
+b
cx
i
+d
= y
i
ax
i
+b = cx
i
y
i
+dy
i
, i = 1, 2, 3
can be written as the vector equation
ax c xy dy = b
_
_
1
1
1
_
_
,
where xy =
_
_
x
1
y
1
x
2
y
2
x
3
y
3
_
_
. We want to show that this equation has a unique
solution, up to multiplying a, b, c, d by a common factor. Consider
three cases:
(i) x, y, and xy are linearly independent. In this case,
_
_
1
1
1
_
_
can
be written as a unique linear combination of them, which yields
our solution for a, b, c, d and hence our automorphism of H.
(ii) x and y are linearly dependent, say y = x, where ,= 0. Then

1 1 1
x
1
x
2
x
3
x
1
y
1
x
2
y
2
x
3
y
3

1 1 1
x
1
x
2
x
3
x
2
1
x
2
2
x
2
3

is a Vandermonde determinant and hence nonzero for distinct


x
1
, x
2
, x
3
. This implies that span(x, y, xy) is 2-dimensional and
that
_
_
1
1
1
_
_
does not lie in it, so that the only solution to our
49
equation is b = 0 and a, b, c determined by the (projectively)
unique dependence relation of x, y, xy.
(iii) x and y are linearly dependent, and xy is a linear combination
of them. I havent gured out how to do this case; its worth
noting that this must be where the hypotheses x
1
< x
2
< x
3
and y
1
< y
2
< y
3
come into play, since so far Ive only used that
theyre distinct without using their cyclic order.

Exercise 16: Let


f(z) =
i z
i +z
and f
1
(w) = i
1 w
1 +w
.
(a) Given R, nd real numbers a, b, c, d such that ad bc = 1, and so
that for any z H
az +b
cz +d
= f
1
_
e
i
f(z)
_
.
(b) Given D nd real numbers a, b, c, d so that ad bc = 1, and so
that for any z H
az +b
cz +d
= f
1
(

(f(z))),
with

dened in Section 2.1.


(c) Prove that if g is an automorphism of the unit disc, then there exist
real numbers a, b, c, d such that ad bc = 1 and so that for any z H
az +b
cz +d
= f
1
g f(z).
Solution.
(a)
f
1
(e
i
f(z)) = f
1
_
e
i
i z
i +z
_
= i
1 e
i iz
i+z
1 +e
i
iz
i+z
= i
i +z ie
i
+e
i
z
i +z +ie
i
e
i
z
=
i(1 +e
i
)z +i(1 e
i
)
(1 e
i
)z +i(1 +e
i
)
=
2ie
i/2 e
i/2
+e
i/2
2
z + 2e
i/2 e
i/2
e
i/2
2i
2ie
i/2
e
i/2
e
i/2
2i
z + 2ie
i
e
i/2
+e
i/2
2
=
i cos(/2)z + sin(/2)
i sin(/2)z +i cos(/2)
.
The determinant here is adbc = cos
2
(/2)+i sin
2
(/2) ,= 0, so they
can all be scaled by an appropriate amount to make the determinant
equal 1.
50
(b)
f
1
(

(f(z))) = f
1
_

_
i z
i +z
__
= f
1
_

iz
i+z
1
iz
i+z
_
= f
1
_
i +z i +z
i +z i + z
_
= f
1
_
(1 +)z +i( 1)
(1 + )z +i(1 )
_
= i
1
(1+)z+i(1)
(1+ )z+i(1 )
1 +
(1+)z+i(1)
(1+ )z+i(1 )
= i
(1 + )z +i(1 ) (1 +)z +i(1 )
(1 + )z +i(1 ) + (1 +)z +i( 1)
= i
( )z +i(2 )
(2 + + )z +i( )
=
bz + (a 1)
(a + 1)z b
where = a + bi. The determinant is b
2
(a
2
1) = 1 [[
2
,= 0,
so it can be made 1 by an appropriate scaling.
(c) Let R

(z) = e
i
z. Then g = R

for some R and D, so


f
1
g f = f
1
R

f
= f
1
R

f f
1

f
= (f
1
R

f) (f
1

f)
is the composition of two Mobius transformations of determinant 1,
by parts (a) and (b); this is another Mobius transformation of deter-
minant 1, so were done.

Chapter 8.6, Page 254


Problem 2: Prove that a real-dierentiable function f : C with J
f
(z
0
) ,=
0 is holomorphic with f

(z
0
) ,= 0 i f preserves angles at z
0
.
Solution. If : [0, 1] C and : [0, 1] C are two curves passing through
z
0
with tangent vectors

and

, then the tangents to f and f at z


0
are J
f

and J
f

by the Chain Rule. For these always to have the same


angle as

and

, it is necessary and sucient that J


f
be a nonzero multiple
of a unitary matrix (this is a standard theorem from linear algebra), i.e.
J
T
f
J
f
= C Id where C ,= 0 and Id is the 2 2 identity matrix. Now
J
T
f
J
f
=
_
u
x
v
x
u
y
v
y
__
u
x
u
y
v
x
v
y
_
=
_
u
2
x
+u
2
y
u
x
v
x
+u
y
v
y
u
x
v
x
+u
y
v
y
v
2
x
+v
2
y
_
51
so the condition for f to preserve angles is that u
2
x
+ u
2
y
= v
2
x
+ v
2
y
and
u
x
v
x
+ u
y
v
y
= 0. We will show that these are equivalent to the Cauchy-
Riemann equations. Clearly if u
x
= v
y
and u
y
= v
x
then u
2
x
+u
2
y
= v
2
x
+v
2
y
and u
x
v
x
+u
y
v
y
= 0. Conversely, if the latter two equations are true, and
f

,= 0, choose some nonzero component of u or v; WLOG u


x
,= 0. Then
v
x
=
uy
ux
v
y
so
v
2
y
_
1 +
u
2
y
u
2
x
_
= v
2
x
+v
2
y
= u
2
x
+u
2
y
= u
2
x
_
1 +
u
2
y
u
2
x
_
v
y
= u
x
.
If v
y
= u
x
then we immediately have v
x
= u
y
as well, so the Cauchy-
Riemann equations are satised. The possibility v
y
= u
x
is impossible
because it would imply that det(J
f
) < 0 which is impossible for a multiple
of a unitary matrix (in an even number of dimensions). Hence, f preserving
angles is equivalent to f being analytic for J
f
,= 0.
Problem 7: Applying ideas of Caratheodory, Koebe gave a proof of the Rie-
mann mapping theorem by constructing (more explicitly) a sequence of
functions that converges to the desired conformal map.
Starting with a Koebe domain, that is, a simply connected domain /
0

D that is not all of D, and which contains the origin, the strategy is to nd
an injective function f
0
such that f
0
(/
0
) = /
1
is a Koebe domain larger
than /
0
. Then, one iterates this process, nally obtaining functions F
n
=
f
n
f
0
: /
0
D such that F
n
(/
0
) = /
n+1
and limF
n
= F is a
conformal map from /
0
to D.
The inner radius of a region / D that contains the origin is dened by
r
K
= sup 0 : D(0, ) /. Also, a holomorphic injection f : / D
is said to be an expansion if f(0) = 0 and [f(z)[ > [z[ for all z /0.
(a) Prove that if f is an expansion, then r
f(K)
r
K
and [f

(0)[ > 1.
Suppose we begin with a Koebe domain /
0
and a sequence of expansions
f
0
, f
1
, . . . , f
n
, . . . , so that /
n+1
= f
n
(/
n
) are also Koebe domains. We
then dene holomorphic maps F
n
: /
0
D by F
n
= f
n
f
0
.
(b) Prove that for each n, the function F
n
is an expansion. Moreover,
F

n
(0) =

n
k=0
f

k
(0), and conclude that lim
n
[f

n
(0)[ = 1.
(c) Show that if the sequence is osculating, that is, r
Kn
1 as n ,
then a subsequence of F
n
converges uniformly on compact subsets
of /
0
to a conformal map F : /
0
D.
To construct the desired osculating sequence we shall use the automor-
phisms

= ( z)/(1 z).
(d) Given a Koebe domain /, choose a point D on the boundary
of / such that [[ = r
K
, and also choose D such that
2
= .
Let S denote the square root of

on / such that S(0) = 0. Why


is such a function well dened? Prove that the function f : / D
dened by f(z) =

is an expansion. Moreover, show that


[f

(0)[ = (1 +r
K
)2

r
K
.
(e) Use part (d) to construct the desired sequence.
Solution.
(a) Since f is a holomorphic injection, it is a homeomorphism, so it maps
/
0
to a simply connected domain. Let < r
K
and consider the image
52
under f of C(0, ). This curve is mapped to another curve all of
whose points are at least away from the origin; since the image
under f of /
0
is simply connected and includes the origin, it includes
the interior of this curve. In particular, it contains D(0, ). Thus,
D(0, ) / D(0, ) f(/). Taking suprema, r
f(K)
r
K
.
Since f is holomorphic on /
0
and f(0) = 0, we can write f(z) = zg(z)
on a neighborhood of 0, where g is holomorphic. Then f

(0) = g(0).
Now g ,= 0 since [g(z)[ =
|f(z)|
|z|
> 1 for z ,= 0 and g is continuous.
Thus, by the Minimum Modulus Principle applied to a small circle
near the origin, since [g[ > 1 on the circle, [g(0)[ > 1. (The Minimum
Modulus Principle is just the Maximum Modulus Principle applied to
1
g
, which is valid since g ,= 0 in the region under consideration.)
(b) Since each f
i
xes the origin, so does F
n
. Moreover, it is easy to
see by induction that [F
n
(z)[ > [z[ since [F
n
(z)[ = [f
n
(F
n1
(z))[ >
[F
n1
(z)[ > [z[ by the induction hypothesis. By the Chain Rule,
F

n
(0) = (f
n
F
n1
)

(0) = f

n
(F
n1
(0))F

n1
(0) = f

n
(0)F

n1
(0)
so by another easy induction we have F

n
(0) =

n
k=1
f

k
(0). However,
if we let = r
K0
/2, then D(0, ) /
n
for all n, so we can dene
G
n
(z) : D D by G
n
(z) = F
n
(z). Then by the Schwarz Lemma,
[F

n
(0)[ = [G

n
(0)[ < 1 [F

n
(0)[ <
1

.
Thus, the sequence [F

n
(0)[ is bounded above. This implies that
[f

k
(0)[ 1 as otherwise the product would be innite.
(c) Since each F
n
maps into D, the sequence F
n
is uniformly bounded.
By Montels theorem, there is a subsequence F
n
k
that converges uni-
formly on all compact subsets K /
0
. The limit function F : /
0
D
must be holomorphic because the uniform limit of holomorphic func-
tions is holomorphic (and /
0
is the union of its compact subsets).
Then F is injective by Proposition 3.5 (it cannot be constant because
then it would be everywhere zero, and for z / 0, [f
n
k
(z)[ > [z[
for all n [F(z)[ = lim[f
n
k
(z)[ [z[ > 0). Thus F is an injective
holomorphic function, hence a homeomorphism, so F(/
0
) is simply
connected. I claim that r
F(K0)
r
F
N
(K0)
for all N. To see this,
let < r
F
N
(K0)
. For n
k
N, if w C(0, ), then [F
n
k
(w)[ > .
This implies [F(w)[ . Hence F maps C(0, ) to a smooth closed
curve whose points are all at least from 0; since the image of F is
simply connected, it contains the image of this curve, so it contains
D(0, ). Taking suprema yields r
F(K0)
r
F
N
(K0)
. Since r
F
N
(K0)
1
as N , this implies r
F(K0)
1. Hence D(0, ) F(/
0
) for all
< 1, so F is surjective. So F is a conformal map from /
0
to D.

You might also like