ANALYSIS QUALIFYING EXAM PROBLEMS

SOLUTIONS BY DAVE GAEBLER

Contents
Fall 2001 1
Winter 2002 8
Spring 2002 11
Fall 2002 15
Winter 2003 18
Fall 2003 22
Winter 2004 24
Fall 2004 26
Winter 2005 28
Fall 2005 29
Spring 2006 31
Fall 2006 35
Fall 2001
Problem R1: Consider real numbers a
n,m
for n = 1, 2, . . . and m = 1, 2, . . .
and assume that the inner and outer sums in the expressions
A =

n=1
_

m=1
a
n,m
_
B =

m=1
_

n=1
a
n,m
_
are absolutely convergent.
(a) Give an example that shows that we may have A ,= B.
(b) Under what reasonable additional assumptions on a
n,m
can we con-
clude A = B?
Solution.
(a) Let
a
n,m
=
sign(n m)
2
|nm|
.
Then the inner and outer sums are absolutely convergent, but one way
all the inner sums are positive, and the other way theyre negative.
(b) If all the terms are nonnegative, or if

[a
n,m
[ converges, Fubinis
theorem guarantees that the two iterated sums are equal.

1
2 ANALYSIS QUALS
Problem R2: Let n 1. Let O(n) denote the set of all real n n matrices
G which satisfy
G
T
G = I
where I is the identity matrix.
(a) Prove that O(n) is compact.
(b) Prove that O(n) is not connected.
Solution.
(a) The columns of an orthogonal matrix are orthonormal; in particular,
they each have norm 1. This implies that the Euclidean norm of an
orthogonal matrix is

n. Thus, O(n) is a bounded subset of R
n
2
, and
it is closed because its the inverse image under the continuous function
G G
T
G of the closed set I. By the Heine-Borel theorem, O(n) is
compact.
(b) If O(n) were connected, its image under the continuous function det :
O(n) R would be connected as well. But det(G) = 1 for all G
O(n), and there are matrices with determinant 1 (e.g. the identity) as
well as those with determinant 1 (e.g. negate the last column of the
identity). Hence O(n) is not connected.

Problem R3: Prove the mean value theorem: Let f : R R be a dieren-

tiable function. For every a < b there exists a < < b such that
f(b) f(a)
b a
= f

().
Solution. Let
g(x) = f(x) f(a)
f(b) f(a)
b a
(x a).
Then g is dierentiable and g(a) = g(b) = 0. Since g is continuous on the
compact set [a, b], it attains a maximum and a minimum. If these are both
at the endpoints, g is constant on [a, b], so g

(x) = 0 for all x (a, b).

On the other hand, if g attains either a max or a min on the interval,
it must occur at a point where g

(x) = 0. Either way, (a, b) with

g

() = 0 f

() =
f(b)f(a)
ba
.
Problem R4: The Fourier transform

f of a function f in L
1
(R) is dened as

f() =
_
R
e
2ix
f(x)dx.
(a) State how the Fourier transform of a function in L
2
(R) is dened. (You
do not need to prove claims which you use to state this. Do not use
the fact (b) below.)
(b) Let f be in L
2
(R). Dene
Mf() = sup
r>0

_
r
r
f(x)e
2ix
dx

.
A deep fact proved by Carleson and Hunt states that
|Mf|
2
C|f|
2
ANALYSIS QUALS 3
for some universal constant C. Use this theorem to prove

f() = lim
r
_
r
r
e
2ix
f(x)dx
for almost every R.
Solution.
(a) For f L
1
L
2
the Fourier transform is dened already. Since C
C
(R) is
dense in L
2
and is contained in L
1
, this denes the Fourier transform
on a dense subset of L
2
. It turns out that the Fourier transform
restricted to L
1
L
2
is an isometry, i.e. |

f|
2
= |f|
2
. An isometry on
a dense subspace can always be extended to an isometry on the whole
space, which denes the Fourier transform L
2
L
2
.
(b) Let
E

=
_
R : limsup
r

_
r
r
e
2ix
f(x)dx

f()

> 2
_
.
For any > 0, g L
1
L
2
with |f g|
2
< . Then
_
r
r
e
2ix
f(x)dx

f() =
__
r
r
e
2ix
g(x)dx g()
_
+
_
r
r
e
2ix
(f(x)g(x))dx+(

g f()).
Since
limsup
r

_
r
r
e
2ix
g(x)dx g()

= 0
for all , this implies E

+G

, where
F

= R : M(f g)() >

and
G

= R : [

f g()[ > .
Now by Chebyshevs inequality,
m( : M(f g)() > ) <
|M(f g)|
2
2

2
<
C
2

2
and
m( : [

f g()[ > ) <

|

f g|
2
2

2
=
|f g|
2
2

2
<

2

2
,
so
m(E

) <
(C
2
+ 1)
2

2
.
This is true for any , so m(E

) = 0. But this is true for any , so

m
_
:
_
r
r
e
2ix
f(x)dx ,

f()
_
= m
_

_
n=1
E
1/n
_

n=1
m(E
1/n
) = 0.

Problem R5: For n 1 let

1
(n),
2
(n),

(n) be the Banach space R

n
equipped with the norm
n

i=1
[x
i
[,
_
n

i=1
[x
i
[
2
_
1/2
, sup
i=1,...,n
[x
i
[
4 ANALYSIS QUALS
respectively. Answer the following questions and prove your assertions:
(a) Which of the Banach spaces
1
(2),
2
(2),

(2) are isometrically iso-

morphic?
(b) Which of the Banach spaces
1
(3),
2
(3),

(3) are isometrically iso-

morphic?
Solution. First,
2
is not isometrically isomorphic to either
1
or

in
either 2 or 3 dimensions (or, in fact, any dimension higher than 1). This is
because the parallelogram law
|a +b|
2
+|a b|
2
= 2|a|
2
+ 2|b|
2
holds in
2
, but not in
1
or

. (A simple counterexample for both

1
and

is a = (1, 0) or (1, 0, 0) and b = (0, 1) or (0, 0, 1).)

(a) In 2 dimensions,
1

. One isomorphism is given by

(a, b)
_
a +b
2
,
a b
2
_
.
One can easily check by breaking into cases that

a +b
2

a b
2

is [a[ if [a[ [b[, and [b[ if [b[ [a[. So this is an isometry.

(b) In 3 dimensions,
1
,

. To prove this we will count the number

of extreme points on the unit ball. An extreme point of a convex set
S is one that is not in the interior of any line segment lying in S. It
is easy to verify that the property of being an extreme point of the
unit ball is preserved by an isometric isomorphism, so if
1

then
the unit balls of
1
and

have the same number of extreme points.

However, the unit ball of
1
is an octahedron, which has six extreme
points, whereas the unit ball of

is a cube, which has eight.

Problem R6: Let V be the complex Banach space

(N), i.e. the space

of all sequences x = (x
n
)
n=1,2,...
with the norm |x| = sup
n
[x
n
[. Every
sequence f in
1
(N) gives rise to a linear functional
f
: V C by the
formula
f
(x) =

n=1
f
n
x
n
.
a) Prove that
f
is continuous for each f
1
(N).
b) Prove that there are elements in the dual space of V which are not of
the form
f
for any f
1
(N).
Solution.
ANALYSIS QUALS 5
a) Since
f
is a linear functional, its continuity follows from its bound-
edness. Since
[
f
(x)[ =

n=1
f
n
x
n

n=1
[f
n
[[x
n
[

n=1
[f
n
[|x|

= |x|

n=1
[f
n
[
= |f|
1
|x|

f
is bounded.
b) One can easily check that the set of convergent sequences forms a
subspace of

and that lim is a bounded linear functional on this

subspace. By the Hahn-Banach theorem there exists an extension to
a linear functional f :

C such that f(x) = lim(x) whenever x is

convergent. But for f
1
,
f
cannot agree with lim on all convergent
sequences, because changing a nite number of components of x will
not aect its limit but can aect
f
(x) (e.g. change x
i
for some i
where f
i
,= 0).

Problem C1: Find an explicit conformal mapping from the region

[z[ < 1 [0, 1)
onto the upper half plane Imz > 0.
Solution. The mapping = i
w+
w1
takes this region to the UHP minus the
slit from i to i. The mapping =
1

takes this to the UHP minus the

nite slit from 0 to i. Finally, the mapping z =
_

2
+ 1 (found by invert-
ing the formula =

z
2
1 from the Schwarz-Christoel transformation)
sends this to the UHP. Explicitly, =
2

w
w+1
. (Handys solution is nicer.)
Problem C2: Evaluate the integral
_

cos x dx
9 + 10x
2
+x
4
.
Solution. Let f(z) =
e
iz
9+10z
2
+z
4
. Since 9 + 10z
2
+ z
4
= (z
2
+ 9)(z
2
+ 1) =
(z + i)(z i)(z + 3i)(z 3i), the only singularities are simple poles at i
and 3i. The residues in the upper half-plane are
Res(f, i) =
e
ii
d
dz
(9 + 10z
2
+z
4
)[
z=i
=
e
1
16i
and
Res(f, 3i) =
e
i3i
d
dz
(9 + 10z
2
+z
4
)[
z=3i
=
e
3
48i
.
6 ANALYSIS QUALS
By the Residue Theorem applied to a semicircular contour from R to
R in the upper half plane,
2i
_
e
1
16i
+
e
3
48i
_
=
_

f(z)dz =
_
R
R
f(z)dz +
_
C
f(z)dz
where C is the curved part of the contour. Since [e
iz
[ 1 on the UHP,
[f(z)[
1
R
4
10R
2
9
for large R; since the length of C is R,
_
C
f(z)dz 0
as R by the ML estimate. So
_

f(z)dz = lim
R
_
R
R
f(z)dz =
(3e
1
e
3
)
24
.
Now
cos x
10+9x
2
+1
= Ref(z), so
_

cos x
10 + 9x
2
+ 1
dx = Re
_

f(z)dz =
(3e
1
e
3
)
24
.

Problem C3: Dene

z
=
1
2
_

x
i

y
_
and

z
=
1
2
_

x
+i

y
_
.
(a) For m and n positive integers, calculate

z
z
n
z
m
.
(b) Let P(x, y) be a polynomial in the two real variables x and y. Then
P has the form
P(x, y) =
N

n=0
N

m=0
a
n,m
z
n
z
m
but you do not need to prove it. Prove, howover, that if

z
P(x, y) = 0,
then a
n,m
= 0 for all m > 0.
Solution.
ANALYSIS QUALS 7
(a)

x
z
n
z
m
=

x
(x +iy)
n
(x iy)
m
= n(x +iy)
n1
(x iy)
m
+m(x iy)
m1
(x +iy)
n
= (x +iy)
n1
(x iy)
m1
_
n(x iy) +m(x +iy)
_

y
z
n
z
m
= i(x +iy)
n1
(x iy)
m1
_
n(x iy) m(x +iy)
_

z
z
n
z
m
=
1
2
_

x
+i

y
_
z
n
z
m
=
1
2
2m(x +iy)
n
(x iy)
m1
= mz
n
z
m1
.
(b) By linearity,

z
P(x, y) =

m
a
n,m
mz
n
z
m1
=

m
m z
m1
_

n
a
n,m
z
n
_
.
The m = 0 term is automatically zero, but for m > 0 this can only
be identically zero if

n
a
n,m
z
n
= 0. (We are assuming here that a
polynomial in x and y has a unique representation as a polynomial
in z and z, i.e. that terms with dierent powers of z and z cannot
somehow cancel each other; I take this to be part of what were allowed
to assume.) But a polynomial in z is zero only if each coecient is
zero, so a
n,m
= 0 for m > 0 and for all n.

Problem C4: Let u(z) be a harmonic function on the entire plane C such
that
_ _
C
[u(z)[
2
dxdy < .
Prove that u(z) = 0 for all z.
Solution. For any z, by the mean value property,
u(z) =
1
R
2
_ _
D
R
u(z)dz,
where D
R
is the disk of radius R centered at z. Then
[u(z)[
1
R
2
_ _
D
R
[u(z)[dxdy.
By the Cauchy-Schwarz inequality applied to L
2
(D
R
), this is at most
1
R
2

__ _
D
R
[u(z)[
2
dxdy
_
(R
2
)
1
R

_ _
C
[u(z)[
2
dxdy.
Letting R , we see that [u(z)[ = 0.
8 ANALYSIS QUALS
Winter 2002
Problem 1: Suppose f
n
is a sequence of continuous functions on [0, 1] which
converges to a continuous function f on [0, 1]. Does it follow that f
n
con-
verges uniformly? Give a proof or provide a counterexample.
Solution. No, it does not follow. Pick your favorite of the standard coun-
terexamples. For example, the triangular spikes
f
n
(x) =
_

_
2nx 0 x
1
2n
1 2n
_
x
1
2n
_
1
2n
x
1
n
0
1
n
x 1
converge to 0 on [0, 1], but not uniformly because they all have uniform
norm 1.
Problem 2: Let a
n
be a sequence of positive numbers converging to 0. Show
that given any x > 0 there exist non-negative integers k
1
, k
2
, . . . such that

k
n
a
n
= x.
Solution. The obvious greedy algorithm works, i.e. just pick each k
n
to
be as large as possible without making the sum exceed x. Formally, dene
S
n
=
n

j=1
k
j
a
j
and inductively dene
k
n+1
=
_
x S
n
a
n+1
_
.
Then
0
x S
n
a
n+1
k
n+1
< 1 0 x S
n+1
< a
n+1
.
Since a
n
0, we can take the limit of this inequality to conclude that
S
n
x.
Problem 3: Let f and g be Lebesgue integrable functions on the interval
[0, 1]. Set
F(x) =
_
x
0
f(y)dy, G(y) =
_
y
0
g(x)dx
where dy and dx denote Lebesgue measure. Assume h(x, y) = f(y)g(x) is
Lebesgue measurable on [0, 1] [0, 1]. Prove
_
1
0
F(x)g(x)dx = F(1)G(1)
_
1
0
f(y)G(y)dy.
Solution. By the denition of F,
_
1
0
F(x)g(x)dx =
_
1
0
g(x)
_
x
0
f(y)dydx =
_
[0,1]
2
f(y)g(x)
yx
dA =
_
[0,1]
2
f(y)g(x)(1
y>x
)dA.
ANALYSIS QUALS 9
Here we have used Fubinis theorem to turn the iterated integral into a
two-dimensional integral; this is legitimate because
_
[0,1]
2
[f(y)g(x)
yx
dA
_
[0,1]
2
[f(y)g(x)[ =
__
1
0
[f(y)[dy
___
1
0
[g(x)[dx
_
<
by Tonellis theorem. Returning to our series of integrals above, the last is
equal to
_
[0,1]
2
f(y)g(x)dA
_
[0,1]
2
f(y)g(x)
y>x
dA =
__
1
0
f(y)dy
___
1
0
g(x)dx
_

_
1
0
_
y
0
f(y)g(x)dx
= F(1)G(1)
_
1
0
f(y)G(y)dy.

Problem 4: Let F(x) be a bounded real function on R such that

lim
x
F(x) = 0
and such that for all > 0 there exists > 0 such that whenever (a
j
, b
j
),
1 j n is a nite family if pairwise disjoint open intervals,

(b
j
a
j
) <

[F(b
j
) F(a
j
)[ < .
Prove that there is f L
1
(R) such that for all x R,
F(x) =
_
x

f(t)dt.
Solution. This is the standard theorem on absolutely continuous functions;
see for example page 106 of Folland, or pp. 146-148 of Rudin. Depending
on how much youre allowed to assume, writing this up could be a real
pain (e.g. if you can assume that BV functions are dierentiable a.e. it
makes life easier, but if you want to prove that from scratch it could take
a while).
Problem 5: Let f : [0, 1] C be Lebesgue measurable. Assume fg
L
2
([0, 1]) for every g L
2
([0, 1]). Prove that f L

([0, 1]).
Solution. (From Handy.) Let
E
k
= x [0, 1] : 2
k
[f(x)[ < 2
k+1
,
and let m
k
= m(E
k
). Suppose f / L

; then m
k
> 0 for innitely many k.
Now let
g =

k=0
2
k
1

m
k

E
k
where the summand is understood to be zero if m
k
= 0. Then
_
1
0
g
2
(x)dx =

k=0
2
2k
<
10 ANALYSIS QUALS
so g L
2
. However,
f

2
k

E
k
fg

m
k
=0
1

m
k

E
k
f
2
g
2

m
k
=0
1
m
k

E
k

_
1
0
f
2
g
2

m
k
=0
1
which is innite because there are innitely many k with m
k
,= 0. Hence
fg / L
2
.
Problem 9: Let D be the domain in the complex plane C that is the in-
tersection of the two open disks centered at 1 whose boundary circles
pass through i. Find a conformal map f of D onto the open unit disk
= [w[ < 1 such that f(i) = 1 and f(i) = 1. (You may express f
as a composition of other specic maps.) What are the images of arcs of
circles passing through i under your map f? (Justify your answer.)
Solution. The map =
zi
z+i
sends the region D to z : < arg(z) < +2,
where = 2 arctan(

2 +1). To see this, consider that the fractional linear

mapping z
zi
z+i
maps i to 0 and i to , so circles through i and i
must go to circles through 0 and , i.e. rays. We can nd the specic rays
that are the images of our circles by plugging in z = (

2 1), resulting
in angles of 2 arctan(
1

21
) = 2 arctan(

2 1). (News ash: Turns out

=

4
.) To nd which of the two sectors bounded by these rays is our
image, we note that 0 maps to 1. We now compose this map with the
maps = e
i
z, which rotates this to the sector 0 < arg(z) < 2( ), and
=
/2()
which maps this to the upper half plane. We thus obtain a
map from D to the upper half plane. We now compose this with the map
w =
zi
z+i
, where is a unimodular constant to be chosen. This maps to
the unit disk; so far i 0 1 and i 1, so we choose = 1.
Circles through i are mapped to rays, which are mapped to other rays,
which are mapped to other rays; the nal transformation which maps into
sends these to circular arcs through 1.
Problem 11: Show by contour integration that
_
2
0
d
x + cos
=
2

x
2
1
, x > 1.
Determine for which complex values of z the integral
_
2
0
d
z + cos
exists and evaluate the integral. Justify your reasoning.
ANALYSIS QUALS 11
Solution. Let z = e
i
, so cos =
z+1/z
2
and dz = ie
i
d d =
dz
iz
. Then
_
2
0
d
x + cos
= 2i
_
dz
z
2
+ 2xz + 1
,
where the integral is taken around the unit circle. Of the poles x

x
2
1, only x +

x
2
1 lies inside the unit circle; it has residue
1
d
dz
(z
2
+ 2xz + 1)[
z=x+

x
2
1
=
1
2

x
2
1
,
so by the Residue Theorem,
_
2
0
d
x + cos
= 2i
_
2i
1
2

x
2
1
_
=
2

x
2
1
.
On the region C [1, 1], both
_
2
0
d
z+cos
and
2

z
2
1
are analytic func-
tions of z, and since we have shown them to be equal for z (1, ), the
uniqueness principle guarantees that they are equal on C [1, 1]. (We
use the principal branch of the square root, with the branch cut along the
negative real axis, and the square root of a positive real number taken to
be positive.) For z [1, 1] the integral diverges since the denominator
equals zero for at least one value of .
Problem 12: Let S be a sequence of points in the complex plane that con-
verges to 0. Let f(z) be dened and analytic on some disk centered at 0
except possibly at the point of S and at 0. Show that either f(z) extends to
be meromorphic in some disk containing 0, or else for any complex number
w there is a sequence
j
such that
j
0 and f(
j
) w.
Solution. Assume WLOG that f has no removable singularities, since oth-
erwise we may redene it appropriately at those points. Suppose that f is
meromorphic at 0 and is eventually meromorphic at the points z
j
of S, say
for j N. Then if r = min
j<N
[z
j
[, f is meromorphic on the disk [z[ < r.
Conversely, suppose that either 0 is an essential singularity, or innitely
many of the z
j
are essential singularities. If 0 is an essential singularity, the
Casorati-Weierstrass theorem immediately implies the result. Suppose now
that there is a subsequence z
j
k
of essential singularities. Assume WLOG
that [z
j
k
[ <
1
2
k
(since otherwise we may take a sub-subsequence). Then
for any w, the Casorati-Weierstrass theorem guarantees a point
k
with
[
k
z
j
k
[ <
1
2
k
and [f(
k
) w[ <
1
2
k
. Then [
k
[ <
1
2
k1
, so
k
0, and
f(
k
) w.
Spring 2002
Problem 1: Let V be a nite dimensional real vector space, and let ||
V
be
a norm on V . Let P be the set of one-dimensional linear subspaces of V
(P is called a real projective space). For W
1
, W
2
P dene
d(W
1
, W
2
) = inf|v
1
v
2
| : v
j
W
j
and |v
j
|
V
= 1.
Prove that d is a metric on P and that P is compact with respect to this
metric.
12 ANALYSIS QUALS
Solution. What the crap. This is so a geometry problem, not an analysis
problem. Not to mention that using inf instead of min is totally unnecessary
since each subspace only has two unit vectors. Whatever.
Problem 4: Let f
(
x) be a sequence of Borel measurable real-valued functions
on the interval [0, 1] such that f
n
(x) 0 for all x [0, 1]. Prove there is a
Borel set A [0, 1] such that
(i) m([0, 1] A

) < ;
(ii) f
n
(x) 0 uniformly on A.
Solution. This is just Egorovs theorem, of course, which you can look up
in any real analysis book; in case youre lazy, heres a proof: Let
E
N,m
=

n=N
x : [f
n
(x)[ <
1
m
.
This is the set for which f
n
(x) stays less than
1
m
. Since
N
E
N,m
is the
set of points where f
n
0, which is all of [0, 1] by hypothesis, and since
E
N
E
N+1
, lim
N
m(E
N,m
) = 1 for all m. Choose K with >
1
2
K
and
N
1
, N
2
, . . . such that
m(E
Nm,m
) > 1
1
2
K+m
, m = 1, 2, . . . .
Let A =
m
E
Nm,m
. Then
m(A

) = m(E
Nm,m
)

= m
_
E

Nm,m
_

m(E
Nm,m
)

1
2
K+m
=
1
2
K
< .
Moreover, for any > 0, choose m such that
1
m
< ; then because A
E
Nm,m
, [f
n
(x)[ <
1
m
< for n > N
m
, for all x A. Hence f
n
0
uniformly on A.
Problem 6: On the two-point space 0, 1 let be the measure such that
(0) = (1) =
1
2
. Let X be the innite product space
X =

j=1
X
j
where each X
j
= 0, 1, and let be the product measure on X (dened by
(
n
j=1
x : x
j
= a
j
) = 2
n
). Find an explicit mapping f : X [0, 1] such
that there exists X

X, with (X

) = 0, and Y

[0, 1] with Lebesgue

measure zero such that
f : X X

Y Y

is one-to-one and measure preserving, i.e. m(F(E)) = (E) where m de-

notes Lebesgue measure.
Solution. This is one of those problems where its clear whats going on (a
point in X corresponds to a binary expansion of a point in [0, 1]), but its
a headache to write up. The map is
f(x
n
) =

n=1
x
n
2
n
ANALYSIS QUALS 13
and the sets are X = eventually constant sequences and Y = dyadic rational points q/2
k
.
The latter has measure zero because its countable; the former has measure
zero because it can be explicitly written as

_
N=1

n=N
x : x
n
= 0
plus a corresponding set where x
n
is eventually 1. Now
N+k
n=N
x
n
= 0 is
a union of 2
N1
sets where the rst N + k terms are specied, which has
measure 2
Nk
by denition; this measure approaches zero as k .
Finally, we need to show that f is one-to-one and measure preserving
once the appropriate sets are excised. Letx
n
and y
n
X, and let m
be the rst point at which they dier; suppose WLOG x
m
= 1 and y
m
= 0
Then
f(x
n
) f(y
n
) =
1
2
m
+

j=m+1
x
j
y
j
2
j
> 0
because x
j
y
j
1, 0, 1 and cannot always be 1 if x and y are not
eventually constant. Thus f is one-to-one. Its measure-preserving because
each of the generating sets
n
j=1
x : x
j
= a
j
is mapped to a dyadic interval
with the same measure.
Problem 9: Let f be any conformal mapping from the strip S = z C :
1 < 1z < 1 onto the unit disc D = z : [z[ < 1 such that uniformly in
y (1, 1),
lim
x
f(x +iy) = 1, and lim
x
f(x +iy) = 1.
Find the images in D of the set of horizontal lines in S and the set of vertical
line segments in S.
Solution. The conformal map = e
/2(z+i)
maps S to the UHP; composing
this with the map w =
zi
z+i
maps to D. Any other conformal map must
come from composing with a conformal self-map of the unit disk, which has
the form w

=
wa
1 aw
for some [[ = 1 and [a[ < 1. One can easily check
that our initial map has the desired uniform property; the self-maps of the
unit disk are all uniformly continuous because they are continuous on the
closed disk, so we need to nd which ones x 1. The equations

1 a
1 a
a = 1 a
and

1 a
1 + a
= 1 a = 1 a
together imply = 1 and =
a
a
, so w

=
za
1az
for some real a < 1.
The exponential map takes horizontal lines to rays through the origin;
these in turn are mapped to circular arcs through 1, since they are cir-
cles through 0 and . The subsequent conformal self-map of the disk
sends them to other circles through 1 since these points are xed. The
vertical line segments are sent to orthogonal trajectories of such arcs, which
are circular arcs through i.
14 ANALYSIS QUALS
Problem 11: Let u(z) be a harmonic function on the complex plane C such
that
_ _
C
[u(z)[
2
dxdy < .
Prove u(z) = 0 for all z C.
Solution. This is identical to Fall 2001 Problem 4, above.
Problem 12: Evaluate
_

0
x
2
dx
x
4
+ 6x
2
+ 13
.
Solution. Let f(z) =
z
2
z
4
+6z
2
+13
=
z
2
(z
2
+3)
2
+4
. Then f has simple poles at

3 2i. Of these,

3 + 2i and

3 2i are in the upper half plane.
To compute the residues, we use the fact that
f
g
has residue
f(z0)
g

(z0)
at a
simple pole z
0
. So
Res(f;

3 + 2i) =
3 + 2i
4(2i)

3 + 2i
=
1
8i

3 + 2i
and
Res(f;

3 2i) =
3 2i
4(2i)

3 2i
=
1
8i

3 2i.
Hence
_

f(z)dz = 2i

Res(f; z) = 2i
1
8i
_

3 + 2i

3 2i
_
=

4

_
2(

13 3)
4
where is a semicircular contour in the upper half-plane with radius R.
(Here the square roots are both taken to be in the upper half plane.) Since
f(z)
1
R
2
on the curved part of the contour, the ML estimate guarantees
that the integral over the curved part goes to zero. Hence
_

0
f(x)dx =
1
2
_

f(x)dx =

4

13 3

13
2
.

Problem 14: Let S = [0, 1] [0, 1] be the unit square in C and let f : S C
be a continuous map such that f(z) ,= 0 for all z S. Prove that there
exists continuous g : S C such that
f(z) = e
g(z)
for all z S.
Solution. Write f(x, y) = r(x, y)e
i(x,y)
where r and are continuous real
functions of x and y. Then since r ,= 0 one can dene g(x, y) = log r(x, y)+
i(x, y). The problem thus reduces to showing that r and can be dened
in a continuous fashion. For r this is clear since r = [f[. For , we note
that f : S T, where (z) =
z
|z|
and T is the unit circle, is a continuous
map. Since S is simply connected, and is path connected and locally path
connected, the Lifting Criterion (Prop 1.33 in Hatcher) guarantees the ex-
istence of a lift

f : S R. (Alternatively, simple connectedness implies
ANALYSIS QUALS 15
that the closed dierential d(arg(f(z))) (which can be dened locally) is
exact.)
Fall 2002
Problem 1: Let f, g be two absolutely continuous functions on the interval
[0, 1] which are everywhere positive. Show that the pointwise quotient f/g
is also absolutely continuous on [0, 1].
Solution. Since f, g are continuous on a compact domain, they are bounded,
so M such that [f(x)[ M and [g(x)[ M for all x [0, 1]. Moreover,
g attains a minimum m which is positive by hypothesis. Now given > 0,
choose > 0 such that
k

n=1
[b
n
a
n
[ <
k

n=1
[f(b
n
) f(a
n
)[ <
m
2
M
and
k

n=1
[g(b
n
) g(a
n
)[ <
m
2
2M
.
Now

f(b
n
)
g(b
n
)

f(a
n
)
g(a
n
)

=
[f(b
n
)g(a
n
) f(a
n
)g(b
n
)[
[g(b
n
)g(a
n
)[

[f(b
n
)g(a
n
) f(a
n
)g(b
n
)[
m
2
=
[f(b
n
)(g(a
n
) g(b
n
)) +g(b
n
)(f(b
n
) f(a
n
))[
m
2

[f(b
n
)[[g(a
n
) g(b
n
)[ +[g(b
n
)[[f(b
n
) f(a
n
)[
m
2

M
m
2
([g(a
n
) g(b
n
)[ +[f(b
n
) f(a
n
)[) .
Summing over n, this implies
k

n=1

f(b
n
)
g(b
n
)

f(a
n
)
g(a
n
)

M
m
2
k

n=1
[f(b
n
)f(a
n
)[+
M
m
2
k

n=1
[g(b
n
)g(a
n
)[ <

2
+

2
= .
Hence f/g is absolutely continuous.
Problem 10: Let (X, /, ) be a measure space, and let f
1
, f
2
, . . . be a se-
quence of complex-valued functions in L
1
(X, /, ) such that
|f
n
|
L
1
(X,M,)
2
n
and
|f
n
|
L

(X,M,)

1
2
for all n = 1, 2, . . . . Show that the innite product

n=1
(1 +f
n
(x))
is convergent for -a.e. x, and that the product is a measurable function.
Solution. Since
_

[f
n
(x)[d =

_
[f
n
(x)[d

2
n
= 1,
16 ANALYSIS QUALS
where we have used the monotone convergence theorem to interchange the
sum and the integral, we see that

[f
n
(x)[ L
1
, so it must be nite a.e.
Since the product

(1 + f
n
) converges absolutely exactly where

[f
n
[
does, this product is absolutely convergente (hence convergent) a.e. Now
the product of nitely many measurable functions is measurable (this is
easily proved by induction), and

n=1
(1 +f
n
(x)) = lim
N
N

n=1
(1 +f
n
(x))
is a pointwise limit of measurable functions, and therefore measurable.
(The condition [f
j
[
1
2
comes in handy if you want to prove from
scratch that the convergence of the sum is equivalent to the convergence of
the product. Since
[ log(1 +z)[ =

z
z
2
2
+. . .

[z[ +

z
2
2

z
3
3
+. . .

[z[ +

z
2
2

z
3
3

+. . .
[z[ +
[z[
2
2
+
[z[
3
2
+
[z[
4
2
+. . .
= [z[ +
[z[
2
1 [z[
2[z[
for [z[
1
2
. Hence

[ log(1 +z)[ converges if

[z[ converges.
Problem 11: Find the linear fractional transformation f : C C such that
f(0) = 1, f(1) = 0, and f() = i. What is the image of the line z :
Re(z) = 1 under f?
Solution. Writing the transformation in the form
z+a
bz+c
and applying the
three conditions, we succesively discover that c = a, a = 1, and b = i.
So
z
z 1
iz 1
is the desired transformation. To nd the image of the given line, we note
that it must be a line or circle through the images of 1 and , which are
0 and i. Checking a third point, we nd that 1 + i 1, so the image is
the circle through these points (centered at
1
2
+
1
2
i, radius
1

2
.)
Problem 12: Suppose the function f(z) is continuous on the closed unit disk
[z[ 1 and analytic on the open disk [z[ < 1. Assume f(z) = 0 for all
z in the semi-circle z : [z[ = 1, Im(z) > 0. Prove that f(z) = 0 on the
closed disk.
Problem 13: Let U
n
(z) be a sequence of positive harmonic functions on a
connected open set containing the origin. Show that if
lim
n
U
n
(0) = 0,
ANALYSIS QUALS 17
then
lim
n
|U
n
|
L

(K)
= 0
for all compact subset K .
Solution. Harnacks inequality says that for a positive harmonic function u
on a domain D,
R r
R +r
u(z
0
) u(z)
R +r
R r
u(z
0
)
for z in the compact subdisk [z z
0
[ r of the disk [z z
0
[ < R D.
(This follows from the positivity of u and a straightforward estimate of the
Poisson kernel.) Now let G = z D : U
n
(z) 0. Then Harnacks
inequality implies that G is open, because around any z
0
G there exists
a disk contained in D, on which U
n
(z) 0 uniformly. Similarly, D G is
open, because if z
0
D G, there exists > 0 and a subsequence U
nj
with
U
nj
(z
0
) > ; then there is a disk on which u(z)
Rr
R+r
> 0 so u
n
(z) , 0
on this disk. Since both G and D G are open and D is connected, one of
them is empty; but 0 G, so D = G. Thus, u
n
(z) 0 for all z D. To
prove that this convergence is uniform on compact subsets of D, cover a
compact subset K D by nitely many open disks D
1
, . . . , D
m
D with
centers z
1
, . . . , z
m
; then u
n
(z) 0 uniformly on each disk, and since there
are nitely many, u
n
(z) 0 uniformly on all of K.
Problem 14: Evaluate the integral
lim
R
_
R
0
cos(x
2
)dx
and justify your reasoning.
Solution. We use the eighth-of-a-circle contour bounding the circular sector
of radius R from = 0 to

4
. Then
_

e
iz
2
dz = 0
since e
iz
2
is entire. Now the integral over the part with =

4
is
_
0
R
e
r
2
e
i/4
dr = e
i/4
_
R
0
e
r
2
dr
and the integral along the axis is
_
R
0
e
ix
2
dx. On the curved part,
z = Re
i
, dz = Rie
i
d,
_
e
iz
2
dz =
_
/4
0
Rie
i
e
iR
2
e
2i
d.
Now
[Rie
i
e
iR
2
e
2i
[ = R[e
iR
2
e
2i
[ = Re
R
2
sin 2
.
Since sin
2

for 0
2

, the integral over the curved part is at most

_
/4
0
Re
4R
2
/
d =

4R
(1 e
R
2
).
18 ANALYSIS QUALS
As R , this goes to zero. Hence
_

0
e
ix
2
dx = e
i/4
_

0
e
r
2
dr =

2
e
i/4
.
Taking real parts,
_

0
cos(x
2
)dx =
_

8
.

Problem 15: Let 1 p < and U(z) be a harmonic function on the entire
complex plane such that
_ _
C
[u(z)[
p
< .
Prove that u(z) = 0 for all z.
Solution. This is a slightly more general form of the familiar problem (Fall
2001 Problem C4 and Spring 2002 Problem 11) that uses the L
2
version of
this statement. The only dierence is that now we use Holders inequality
instead of Cauchy-Schwarz. For 1 < p < , let q be the conjugate exponent
p
p1
; then
[u(z
0
)[ =
1
R
2

_ _
D
R
(z0)
u(z)dxdx

1
R
2
_ _
D
R
(z0)
[u(z)[dxdy
=
1
R
2
|u
D
R
(z0)
|
1

1
R
2
|u|
p
|
D
R
(z0)
|
q
= (R
2
)
1/q1
|u|
p
= (R
2
)
1/p
|u|
p
which tends to 0 as R . (Here D
R
(z
0
) is the disk of radius R centered
at z
0
.) In the case p = 1, we simply have
[u(z
0
)[
1
R
2
_ _
D
R
[u(z)[dxdy
1
R
2
_ _
C
[u(z)[dxdy
which tends to zero as R .
Winter 2003
Problem 1: Let be a nite, positive, regular Borel measure on R
2
, and let
( be the family of nite unions of squares of the form
S = j2
n
x (j + 1)2
n
; k2
n
y (k + 1)2
}
,
where j, k, and n are integers. Prove that the set of linear combinations of
characteristic functions of elements of ( are dense in L
1
().
ANALYSIS QUALS 19
Solution. First, we note that this problem is easy if we are allowed to replace
either of the s with a strict inequality. (On the qual I might just do that.)
In that case every open set is a countable disjoint union of squares in S; one
can see this by taking the collection of squares in S which are contained in
the open set G, but which are not contained in any larger square contained
in G. This collection can be straightforwardly shown to be disjoint and
cover G. Since is regular, any measurable set E can be enclosed in an
open set G such that (G E) < . Then
G
=

Sn
approximates

E
within . Once we can approximate characteristic functions, we can
of course approximate simple functions, then nonnegative functions, then
complex functions.
If neither inequality is allowed to be strict, the same technique will work
if is absolutely continuous with respect to Lebesgue measure: any open
set can be written as an almost disjoint union of closed rectangles, and the
corresponding sum of characteristic functions will be equal a.e. to
G
. If
has a singular part, I think the theorem is still true, but I dont know how
to prove it.
Problem 4: Prove or disprove: If F is a strictly increasing continuous map
from the real line R onto itself and if A R is Lebesgue measurable, then
f
1
(A) is Lebesgue measurable.
Problem 5: Is the Banach space

of bounded complex sequences a =

a
n

n=1
with the supremum norm |a|

= sup[a
n
[ separable? Prove your
answer is correct.
Solution. No. There are uncountably many sequences all of whose terms
are 0 or 1, but they are all a distance 1 apart.
Problem 6: Let x R have decimal expansion
x = a
0
.a
1
a
2
a
3
. . .
and assume this expansion is unique (it is unique for almost all x). Dene
A
n
(x) =
_
1, if a
n
is even
1, if a
n
is odd.
Let f L
1
(R). Prove
lim
n
_
R
f(x)A
n
(x)dx = 0.
Solution. Note that A
n
(x) = A
n
(x +
1
10
n
). So
_
f(x)A
n
(x)dx =
_
f(x
1
10
n
)A
n
(x)dx.
Suppose f C
C
(R). Let supp(f) [M, M]. Then f is uniformly contin-
uous, so given > 0 there exists > 0 such that [f(x) f(y)[ <

2M+2
if
[x y[ < . Let
=
_ _
f(x
1
10
n
) f(x)
_
A
n
(x)dx.
20 ANALYSIS QUALS
Now for suciently large n,
1
10
n
< so the integrand is bounded by

M+2
and vanishes outside [M 1, M + 1], so [[ < . Now

_
f(x)A
n
(x)dx =
_
f(x
1
10
n
)dx
=
__ _
f(x
1
10
n
) f(x)
_
A
n
(x)dx
_
+
_
f(x)A
n
(x)dx
= +
_
f(x)A
n
(x)dx

_
f(x)A
n
(x)dx =

2
.
But [[ < , so
_
f(x)A
n
(x)dx 0. Now let f L
1
(R). Given > 0,
choose g C
C
(R) with |f g|
1
< . Then

_
f(x)A
n
(x)dx

_
(f(x) g(x))A
n
(x)dx +
_
g(x)A
n
(x)dx

_
g(x)A
n
(x)dx

+
_
[f(x)g(x)[dx
since [A
n
(x)[ = 1. This is eventually less than 2. Since this is true for any
,
_
f(x)A
n
(x)dx 0.
Problem 9: Find the Laurent series expansion for
1
z(z + 1)
valid in 1 < [z 1[ < 2.
Solution. Note that
1
z(z+1)
=
1
z

1
z+1
. Since
1
z
=
1
z 1
1
1 +
1
z1
=
1
z 1
_
1
1
z 1
+
1
(z 1)
2
. . .
_
and
1
z + 1
=
1
2
1
1 +
z1
2
=
1
2
_
1
1
2
(z 1) +
1
4
(z 1)
2
. . .
_
are both convergent in this annulus, the Laurent series expansion is
1
z(z + 1)
=

n=0
(1)
n
2
n+1
(z 1)
n
+
1

n=
(1)
n1
(z 1)
n
.

Problem 10: Prove, by using the residue calculus, that

_

0

x
1 +x
2
dx =

2.
Solution. We use the keyhole contour. By the ML estimate, the integral
over the small circle is at most 2

1+
2
0 and the integral over the large
circle is at most 2R

R
1+R
2
0. THe integral on the bottom part of the
contour is I and on the top is I, where I =
_

x
1+x
2
dx. The residues are
Res(f; i) =

i
2i
=
1 i
2

2
ANALYSIS QUALS 21
and
Res(f; i) =

i
2i
=
1 i
2

2
so
I = i

Res(f; z) =

2
.

Problem 11: Let f and g be continuous functions on the real line related by
the Fourier transform,
f(x) =
1

2
_

g(x)e
ikx
dx, g(x) =
1

2
_

f(x)e
ikx
dx.
Prove that both f and g cannot be compactly supported (i.e. vanish outside
some nite interval of the real line).
Solution. Actually they can be, if theyre both identically 0. :-) Otherwise,
suppose f is compactly supported. Then
h(k) =
1

2
_

f(x)e
ikx
dx
denes an analytic function of k on the entire complex plane. (It is easy to
show the integral converges everywhere, and one may dierentiate under
the integral sign to obtain a derivative.) This function is equal to g on the
real line. But if g is compactly supported, then it vanishes for real x of
large magnitude. Thus the set of zeros of h has an accumulation point,
so h must be everywhere zero, which then implies that both g and f are
everywhere zero.
Problem 12: Let f(z) be an entire function which is not a constant. Prove
that e
f(z)
has an essential singularity at z = .
Solution. Liouvilles Theorem for harmonic functions states that a har-
monic function on the whole plane which is bounded above or below must
be constant. (The proof of this follows easily from Harnacks inequality;
see Spring 2006 Problem 7.) The Cauchy-Riemann equations imply that
if Re(f) is constant, then f is constant. Hence Re(f) must be unbounded
in both directions; in particular, there are values of z with arbitrarily large
modulus for which Re(f(z)) < 0. This implies [e
f(z)
[ < 1 at these points,
so [e
f(z)
[ , as z . Hence e
f(z)
does not have a pole at . On
the other hand, the unboundedness of Re(f) guarantees that there exists
a sequence z
n
with Re(z
n
) and therefore with [e
f
(z
n
)[ ,
so e
f(z)
cannot have a removable singularity at (or be analytic there).
Hence e
f(z)
has an essential singularity at .
Problem 13: Explain why there is a unique analytic function g in the domain
D roughly sketched below (shaded region) with the property that e
g(z)
= z
for all z D and g(1) = 2i. Compute g(i + 1).
22 ANALYSIS QUALS
Solution. Let be any closed curve lying in the given domain. By the
argument principle,
1
2i
_

1
z
dz = N
0
N

,
where N
0
is the number of poles of f(z) = z enclosed by , and N

is
the number of poles. But f has no poles anywhere and no zeros in D, so
_
dz
z
= 0 around any closed curve in D. Since D is simply connected, every
closed form is exact, so
dz
z
= dg for some analytic function g on D. In
fact there is a family of such g, diering by an additive constant. Now let
h(z) = e
g(z)
. Then h

(z) = h(z)g

(z) =
h(z)
z
which implies
d
dz
h(z)
z
= 0, so
h(z) = cz for some nonzero constant c. By adding an appropriate constant
to our choice of g (namely any value of log(c)), we may nd a g such that
e
g(z)
= z. This proves existence. Now suppose there are two such analytic
functions g and g

. Then e
gg

=
e
g
e
g
=
z
z
= 1 so g g

= 2in for some

integer n (since gg

is continuous and D is connected, the value of n is the

same throughout D). Specifying the value of g at one point thus uniquely
determines it throughout D. To compute g(1 +i), we use
g(1 +i) g(1) =
_
1+i
1
dg =
_
1+i
1
dz
z
.
This integral is taken along a path lying in D, but since
dz
z
is closed on the
plane minus the negative real axis, the integral is path independent and the
answer is just Log(1 +i) =
1
2
log 2 +
i
4
. Thus
g(1 +i) =
1
2
log 2 +
9i
4
.

Fall 2003
Problem 9: Let A = z : Im(z) > 0 z : Re(z) = 0 and 0 Im(z) 1.
Find a conformal map that maps A one-to-one onto the upper half plane
z : Im(z) > 0, or show that no such map exists.
Solution. The map is w =

z
2
+ 1. One can get this from the Schwarz-
Christoel transformation with angles of

2
, 2, and

2
; it is also easy to
verify directly that z z
2
+ 1 maps A onto the plane minus the real axis,
so taking the (principal branch of the) square root maps to the UHP.
Problem 10: Use a contour integral to evaluate
_

0
1
1 +x
2n
dx, n 1.
Solution. We use the circular sector of radius R with 0

n
. Let I be
the desired integral. Then the integral along the axis is I and the integral
along the other radius is
z = re
i/n
, dz = e
i/n
dr,
_
f(z)dz =
_
0
R
1
1 +r
2n
e
i/n
dr = e
i/n
I.
ANALYSIS QUALS 23
The integral along the curved part is bounded by
2R
n(1+R
2n
)
0 by the ML
estimate. The only singularity within the countour is at e
i/2n
with residue
1
d
dz
(1 +z
2n
)[
z0
=
1
2ne
i/2n
.
Hence
(1 e
i/n
)I = 2i
1
2ne
i/2n
I =

2nsin
_

2n
_.

Problem 11: Let p(z) = z

n
+a
n1
z
n1
+ +a
0
be a complex polynomial.
Show that there must be at least one point with [z[ = 1 and [p(z)[ 1.
Solution. Suppose to the contrary that [p(z)[ < 1 = [z
n
[ on the unit circle.
By Rouches Theorem, z
n
and z
n
p(z) have the same number of zeros
(counting multiplicity) inside the unit circle. But z
n
has n zeros inside the
unit circle (an n-fold zero at the origin), whereas z
n
p(z) is a degree n1
polynomial and hence can have at most n 1.
Problem 12: Let D be the open unit disk in the complex plane. Endow D
with the Lebesgue measure . Let A L
2
(D, ) be the subspace consisting
of those L
2
functions which are analytic on the disk.
(a) Show that A is innite-dimensional.
(b) Show that A is a closed subspace of L
2
(D, ).
Solution.
(a) The analytic functions 1, z, z
2
, . . . are all linearly independent because
they are orthogonal:
int
D
z
n
z
m
dxdy =
_
D
r
n+m
e
i(nm)
rdrd = 0
since
_
2
0
e
i(nm)
d = 0 for n ,= m. Since A contains innitely many
linearly independent functions, it is innite-dimensional.
(b) Let f A. Let z D; then for any r < d(z, D
c
),
f(z)
2
=
1
r
2
_
B(z,r)
f(z)
2
dxdy
[f(z)[
2
=
1
r
2

_
B(z,r)
f(z)
2
dxdy

1
r
2
_
B(z,r)
[f(z)[
2
dxdy

1
r
2
|f|
2
.
Hence [f(z)[
1

d(z,D
c
)
|f|
2
. Then for any compact subset K D,
d(K, D
c
) > 0; this implies that if f
n
is Cauchy, then f
n
converges
uniformly on K, since
[f
n
(z) f
m
(z)[
1

d(K, D
c
)
|f
n
f
m
|
2
24 ANALYSIS QUALS
for all z K. Now any z D is contained in a compact subset of D,
and since the uniform limit of analytic functions is analytic, limf
n
is
analytic at z.

Winter 2004
Problem CA1: Let f(z) = u(x, y) +iv(x, y) denote a non-constant analytic
function on some open domain D C. Show that at each point of D, the
level curves u(x, y) =constant and v(x, y) =constant intersect at right
angles.
Solution. This is actually false as stated; the level curves intersect at right
angles at any point z for which f

(z) ,= 0. A counterexample that violates

this condition is f(z) = z
2
; at the origin the real axis is a level curve of
both u and v, as is the line y = x.
With this additional assumption, we note that u = (u
x
, u
y
) and v =
(v
x
, v
y
) are orthogonal to the curves (it is a general fact that the gradient is
orthogonal to the level sets); by the Cauchy-Riemann equations, u v =
u
x
v
x
+ u
y
v
y
= u
x
(u
y
) + u
y
(u
x
) = 0. Since the (nonzero) normal vectors
of the curves are orthogonal, the curves are orthogonal.
Problem CA2: Let K(z) denote a real-valued function of the complex vari-
able z dene in some open domain D C. Then K(z) is said to be strictly
subharmonic if the inequality
K(z
0
) <
1
2
_
2
0
K(z
0
+e
i
d
holds for all which are smaller than the distance from z
0
to the boundary
of D. Let f(z) denote a non-constant analytic function on D. Show that
f(z) is strictly subharmonic.
Proof. This is a simple application of the triangle inequality for integrals:
[f(z)[ =
1
2

_
2
0
f(z +e
i
)d

1
2
_
2
0
[f(z +e
i
)d.
The equality condition is that the integrand have constant argument. But
for an analytic function this implies the function itself is constant: WLOG
we assume f is real on the circle (else multiply it by an appropriate e
i
);
then by the maximum-modulus principle for harmonic functions, Im(f)
is zero on the disc, which by the Cauchy-Riemann equations implies f is
constant on the disc, which implies f is constant on the domain. Hence the
inequality is strict.
Problem CA3: Using contour (or any other) methods, compute the integral
_

0
xdx
1 +x
3
.
Solution. A partial fraction decomposition yields
x
1 +x
3
=
1/3
x + 1
+
1/3(x + 1)
x
2
x + 1
ANALYSIS QUALS 25
and
_
x
1 +x
3
dx =
1
6
ln
_
x
2
x + 1
(x + 1)
2
_
+
1

3
tan
1
_
2x 1

3
_
+C.
Evaluating between 0 and yields the answer
2
3

3
.
Alternatively, we use a third-of-a-circle contour. The integral along the
axis is
_
R
0
x
1+x
3
dx. The integral along the curved part goes to 0 by the ML
estimate. The integral along the other radius is
z = re
2i/3
, dz = e
2i/3
dr,
_
f(z)dz = e
4i/3
_
R
0
r
1 +r
3
dr = e
4i/3
_
R
0
r
1 +r
3
dr.
The sum of the integrals is thus
(1 e
4i/3
)
_

0
xdx
1 +x
3
=

3ie
2i/3
I
where I is the desired integral. The residue at e
i/3
(the only singularity
within the contour, a simple pole) is
z
d
dz
(1 +z
3
)
=
1
3z
=
1
3
e
i/3
so
2i
e
i/3
3
=

3ie
2i/3
I I =
2
3

3
.

Problem CA6: Let f(z) denote a function which is analytic at all points on
a simple closed curve and everywhere inside except for the (isolated)
points b
1
, . . . , b
r
where it has poles of order
1
, . . . ,
r
. Furthermore, f
does not vanish on but does have zeros at the points a
1
, . . . , a
s
whiche
are inside ande these zeros are of multiplicity
1
, . . . ,
s
. Then, according
to a well known formula,
1
2i
_

(z)
f(z)
dz =
s

k=1

k

r

k=1

k
.
Now suppose that g(z) is analytic inside and on . What is the general-
ization of the above formula when the integrand on the left hand side is
replaced by g(z)
f

(z)
f(z)
? Provide justication for your answer.
Solution. In a neighborhood of a
k
, f(z) = (z a
k
)

k
h(z), where h(z) is
analytic and nonzero at a
k
. Then f

(z) = h

(z)(z a
k
)

k
+
k
h(z)(z
a
k
)

k
1
so
g(z)
f

(z)
f(z)
= g(z)
_
h

(z)
h(z)
+
k
1
z a
k
_
in a neighborhood of a
k
. Thus, the residue of g(z)
f

(z)
f(z)
at a
k
is
k
g(a
k
).
Similarly, the residue at b
k
is
k
g(b
k
), so by the Residue Theorem,
1
2i
_

g(z)
f

(z)
f(z)
dz =
s

k=1

k
g(a
k
)
r

k=1

k
g(b
k
).

26 ANALYSIS QUALS
Fall 2004
Problem 8: Evaluate the integral
I =
_

cos x
(1 +x
2
)
2
dx.
You will need to briey justify any limits you take.
Solution. We use a semicircular contour of radius R to evaluate
_

e
iz
(1 +z
2
)
2
dz.
Since [e
iz
[ 1 on the upper half plane, the integral over the curved part
tends to 0 as R by the ML estimate. The only singularity inside the
contour is at i; since it is a pole of order 2, the residue is
lim
zi
d
dz
_
(z i)
2
e
iz
(1 +z
2
)
2
_
=
i
2e
.
By the Residue Theorem,
_

e
iz
(1 +z
2
)
2
dz = 2i
i
e
=

e
.
Taking real parts,
_

cos x
(1 +x
2
)
2
dx =

e
.

Problem 9: Find the number of zeros of the polynomial z

6
+12z
4
+z
3
+2z+6
in the rst quadrant inside the unit circle (the domain z = x + iy : x >
0, y > 0, [z[ < 1), and also in the rst quadrant outside the unit circle (the
domain z = x +iy : x > 0, y > 0, [z[ > 1).
Solution. First, p(z) has no real zeros. It is obvious that p(x) > 0 for x > 0;
for x < 0 this follows from the fact that 6+2x+x
3
> 0 for 1 x 0 and
12x
4
+ 2x + x
3
> 0 for x 1. Now consider a quarter-disk D
R
of large
radius R lying in the rst quadrant. As z goes from 0 to R along the real
axis, arg p(z) doesnt change because p(z) is a positive real number. As z
goes from R to Ri along the curved part, p(z) z
6
so arg p(z) changes by
about 6

2
= 3. Finally, consider what happens as z goes from Ri to 0
down the imaginary axis. Here z = iy as y goes from R to 0 in the reals;
then p(z) = (y
6
+ 12y
4
+ 6) + i(2y y
3
). At y = R this has argument
slightly greater than . It stays in the lower half plane until it intersects
the real axis at y =

2, at which point p(z) = 46 and the argument has
increased by . Finally, p(iy) stays in the upper half plane from y =

2 to
y = 0, where p(z) = 6, so the argument does not change in this segment.
Thus, arg p(z) changes by a total of about 4 (and hence exactly 4) around
the disk, so there are two zeros inside it, i.e. p has two zeros in the rst
quadrant. Because the coecients of p are real, the roots occur in conjugate
pairs. Hence there are two zeros in the fourth quadrant, and therefore one
each in the second and third.
ANALYSIS QUALS 27
(At this point we might note that Rouches theorem guarantees that four
of these zeros lie inside the unit circle, but we dont know which four.)
We can repeat the process for a quarter-circle of radius one. Once again
there is no change in argument along the positive real axis. Along the
imaginary axis, p(z) stays in the upper half plane as noted above (since
0 y

2), and goes from 17+i to 6, a small negative change in argument.

Finally, the change in argument along the curved sector is 2k , where
= arctan(1/17). To nd k, we note that [12z
4
[ > [z
6
+z
3
+2z +6[ on the
unit circle. In general, if [g[ < [f[, then arg(f)

2
< arg(f+g) < arg(f)+

2
,
as may be seen by drawing a circle of radius [g[ centered at f. This implies
that the change in arg(f +g) is within

2
of the change in arg(f). Now the
change in arg(z
4
) along the sector is 2, so the change in arg p(z) must be
within

2
of 2. Hence k = 1 and the total change in argument along the
quarter of the unit circle is 2, so there is one zero inside the rst quadrant
inside the unit circle.
Problem 10: Let D = z = x +iy : y > 0 [i, i] be the domain obtained
from the open upper half-plane by excising the vertical ray from i to i.
Find a conformal map w = f(z) of D onto the open unit disk [w[ < 1.
You map represent the map f(z) as a composition of other maps. Include
a sketch or diagram with your solution.
Solution. First, note that z
1
z
maps this to the UHP minus the closed
slit from 0 to i. Then z

z
2
+ 1 maps this to the UHP (see Fall 2003
Problem 9).
Problem 12: Let D be a domain, i.e. a connected open set. Let f : D R
be a continuous function with the property that whenever B(z
0
, r) is a
closed disk contained in D, and h : B(z
0
, r) R is a harmonic function
such that f(z) h(z) for all z B(z
0
, r), f(z
0
) h(z
0
). (Such functions
are called subharmonic.)
Show that f cannot attain its maximum at any point in D unless it is a
constant function.
Solution. First we show that f has the property that
f(z
0
)
1
2
_
2
0
f(z
0
+re
i
)d
for any r such that B(z
0
, r) D; this is often used instead as the denition
of subharmonic. Given such a ball, let u be a harmonic function which
equals f on B(z
0
, r); such a u may be explicitly found from the Poisson
kernel, viz.
u(z
0
+e
i
) =
1
2
_
2
0
1
2
1 2 cos( ) +
2
f(z
0
+re
i
)d
for 0 < r. Then
f(z
0
) u(z
0
) =
1
2
_
2
0
u(z
0
+re
i
)d =
1
2
_
2
0
f(z
0
+re
i
)d.
Now suppose f attains a maximum value M on D at some point z
0
. Let
G = z D : f(z) = M. We will show that G = D. Clearly G is closed
28 ANALYSIS QUALS
in D since it is the inverse image under a continuous function of the closed
set M R. Now let z G and choose any r > 0 such that B(z, r) D.
Then for any < r,
M = f(z)
1
2
_
2
0
f(z +e
i
)d.
But the integrand is at most M everywhere, so this average is at most
M with equality i the integrand is M everywhere (since the integrand is
continuous). Hence f() = M for [ z[ < r. Thus, G is open as well as
closed; since D is connected, G = D or , but z
0
G, so G = D.
Winter 2005
Problem 2: Let be a connected open subset of the plane and let z
0
, z

.
Prove there is a constant M = M(z
0
, z
1
) > 0 such that whenever U(z) is a
positive harmonic function on ,
1
M

U(z
1
)
U(z
0
)
M.
Solution. By Harnacks inequality, such a constant exists provided z
0
and
z
1
are contained in a disk contained in . Since is connected, it is path
connected. Let be a path from z
0
to z
1
. Since is a compact subset of
, it is covered by nitely many open disks. This allows us to nd an M
for z
0
and z
1
.
Problem 4: Let 1 < a < 0 < b < 1 and let be the domain
z : [z[ < 1 ((1, a] [b, 1)) .
Find a conformal map from onto the open unit disc. You do not need to
write the full expression; just tell us the sequence of maps you are taking.
Solution. First, the map z i
1+z
1z
maps to the UHP minus the slit from
0 to
1+a
1a
i and the innite slit from
1+b
1b
i to i. Scaling via z
1a
1+a
z
maps this to the UHP minus the slits from 0 to i and from i to i, where
=
(1+b)(a1)
(1b)(1+a)
> 1. Then z

z
2
+ 1 maps this to the UHP minus the
slit from

2
1i to i. Then z
1
z/

2
1
takes this to the UHP
minus the slit from 0 to i; the map z

z
2
+ 1 takes this to the UHP,
and the map z
zi
z+i
takes this to the open unit disc.
Problem 5: Prove that the limit
lim
N
_
N
N
x
2
sin(x)
x
3
1
dx
exists and compute its value.
Solution. We use an indented semicircular contour to evaluate p.v.
_

z
2
e
iz
z
3
1
dz.
The integral of the curved part goes to zero by Jordans lemma. The
residues at simple poles are Res(f; z
0
) =
z
2
0
e
iz
0
3z
2
0
=
1
3
e
iz0
, so
Res(f; 1) =
1
3
, Res(f; e
2i/3
) = e
i(1/2+

3/2i)
/3 =
ie

3/2
3
.
ANALYSIS QUALS 29
By the fractional residue theorem, the integral around the indentation ap-
proaches
i
_
1
3
_
=
i
3
so by the Residue Theorem,
p.v.
_

z
2
e
iz
z
3
1
=
i
3
+ 2i
ie

3/2
3
=
2
3
e

3/2

i
3
.
Comparing imaginary parts,
p.v.
_

x
2
sin(x)
x
3
1
=

3
.

Problem 6: Let P(z) and Q(z) be polynomials and consider the function
f(z) = P
_
1
z
_
+Q
_
1
z 1
_
on the domain = C 0, 1. Give conditions on P and Q necessary and
sucient for the dierential equation
F

(z) = f(z)
to have a single valued solution on .
Solution. The condition is that the linear terms of both P and Q are zero.
Suppose this is the case. Then by the Residue Theorem,
_

f(z)dz = 0 for
any closed curve Omega, so the integral
_
z
z0
f(z)dz is path independent
and provides an antiderivative for f on . Conversely, if a single-valued
antiderivative F exists, then
_

f(z)dz =
_

dF = 0 around any closed

curve, which implies that the residues at 0 and 1 are both zero.
Fall 2005
Problem C1: Suppose that f(z) is analytic and non-constant on a connected
open set G in the complex plane. Prove that f(G) is an open subset of the
complex plane.
Solution. Let w
0
f(G), so w
0
= f(z
0
) for some z
0
G. Suppose f(z)w
0
has a zero of order m > 0 at z
0
. Because the zeros of the nonconstant
analytic function f(z) w
0
are isolated, there exists > 0 such that
B(z
0
; ) G and f(z
0
+ re
i
) ,= w
0
for 0 < r . Now because the
circle z : [z z
0
[ = is compact, its image under f is compact and hence
is a positive distance > 0 from w
0
. Now consider the integral
N(w) =
1
2i
_
|zz0|=
f

(z)
f(z) w
dz
on the region [ww
0
[ < . Then the denominator is nonzero on the contour
of integration, so N(w) is an analytic function of w (in particular, we can
nd a derivative for it by dierentiating under the integral sign). But this
logarithmic integral is integer-valued, so it must be constant on the region
[w w
0
[ < . Since f(z) w
0
has a zero of order m at z
0
and no other
30 ANALYSIS QUALS
zeros or poles on or inside the contour (i.e. for [z z
0
[ ), N(w
0
) = m,
so N(w) = m throughout the region [ww
0
[ < . Since f(z) w never has
any poles, this means that f(z) w must have m zeros inside the contour,
the interior of which lies in G. In particular, w f(G). This is true for
any w
0
f(G), so f(G) is open.
Problem C2: Find an explicit conformal mapping from the upper half-plane
slit along the vertical segment
z C : Im(z) > 0 (0, i]
to the unit disk z C : [z[ < 1.
Solution. The map z

z
2
+ 1 takes the given region to the UHP (cf. Fall
2003 Problem 9); the map z
zi
z+i
then takes this to the unit disk.
Problem C3: Consider the meromorphic function
f(z) =
1 z
2
2i(z
2
(a +
1
a
)z + 1)
, [a[ < 1.
Find the Laurent series expansion for f(z) valid in a neighborhood of the
unit circle [z[ = 1.
Solution. Since
f(z) =
1 z
2
2i(z a)(z
1
a
)
=
1
2ia
1
z a

1
2ia
1
z
1
a
,
and
1
z a
=
1
z
1
1
a
z
=
1
z
_
1 +
a
z
+
a
2
z
2
+. . .
_
and
1
z
1
a
=
1
1
a
1
1 az
=
1
a
_
1 +az +a
2
z
2
+. . .
_
are convergent for [a[ < [z[ <
1
|a|
, the Laurent expansion on this annulus is
f(z) =

n=0
ia
n
2
z
n
+

n=1
ia
n2
2
1
z
n
.

Problem C4: Using the residue calculus, evaluate the integral

_

0
log x
(x
2
+ 1)
2
dx.
Solution. We use the principal branch of log z, with the branch cut along
the positive real axis. Integrating along a semicircular contour indented at
the origin, the integral over the large curved part is bounded by
2 log R
(R
2
+1)
2
(R)
0 since [ log(Re
i
)[ 2 log R for R large. Similarly, the integral over the
small curved part is bounded by
2| log |
(
2
+1)
2
() 0 since [ log(e
i
)[ 2[ log [
for small. The integral from R to is
_

R
log z
(z
2
+ 1)
2
dz =
_
R

i + log x
(x
2
+ 1)
2
dx
ANALYSIS QUALS 31
so
2
_

0
log x
(x
2
+ 1)
2
dx +i
_

0
1
(x
2
+ 1)
2
dx = 2iRes(f; i).
The residue can be computed as
lim
zi
d
dz
(z i)
2
f(z) = lim
zi
d
dz
log z
(z +i)
2
= lim
zi
z +i 2z log z
z(z +i)
3
=
+ 2i
8
.
Comparing real and imaginary parts,
_

0
1
(x
2
+ 1)
2
dx =

4
and, what we are interested in,
_

0
log x
(x
2
+ 1)
2
dx =

4
.

Spring 2006
Problem 1: Let and
n
, n N, be nite positive Borel measures on R
such that
_
f(x)d
n
(x)
n
_
f(x)d(x)
for all continuous functions f with compact support.
(1) Show that then for each compact set K R,
limsup
n

n
(K) (K).
(2) Give an example that shows (R) < limsup
n

n
(R) is possible.
Solution.
(1) Let > 0. Since R has the property that every open set is -compact,
and since is nite on compact sets, is regular (Theorem 2.18 from
Rudin). Hence G K open such that (G K) < . G can also
be chosen to be bounded and hence have compact closure (if K
[M, M], replace G with G(M1, M +1)). By Urysohns lemma,
f

continuous such that 0 f 1, f = 1 on K and f = 0 outside G.

Then

n
(K)
_
f

d
n

_
f

d (K) +,
so
n
(K) is eventually less than, say, (K) + 2. This is true for any
, so limsup
n
(K) (K).
(2) Let A
n
= [0, 1] [n, n+1]. Dene
n
(E) = m(EA
n
) for all Lebesgue
measurable E, where m indicates Lebesgue measure. Let (E) =
m(E [0, 1]). Then if f is compactly supported, f = 0 on [n, n + 1]
for suciently large n, so
_
fd
n
=
_
1
0
f(x)dx =
_
fd. However,

n
(R) = 2 for all n whereas (R) = 1.

32 ANALYSIS QUALS
Problem 2: Let f
n
be a sequence of L
(
R) functions such that
lim
n
_
R
f
n
(x)g(x) = g(0),
for each g C
0
(R), that is, continuous functions vanishing at innity. Show
that |f
n
|
L
1 is uniformly bounded but that f
n
is not Cauchy in L
1
.
Solution. In functional lingo, were given that the absolutely continuous
measures
n
corresponding to f
n
are weakly convergent to the Dirac delta
measure in the space M(R) = C
0
(R)

. The fact that the sequence is uni-

formly bounded is then an immediate consequence of, guess what, the uni-
form boundedness principle. However, the sequence cannot be Cauchy be-
cause L
1
is complete, so if
n
were Cauchy it would have a limit in the
norm topology and this limit would be absolutely continuous. But norm
convergence implies weak convergence, so this norm limit would have to be
the same as the weak limit, which is not absolutely continuous.
Problem 3: Consider the space L

([0, 1], ) where denotes the Lebesgue

measure. Let
d(f, g) = inf
>0
_
+(x : [f(x) g(x)[ > )
_
.
Prove that d is a metric and that f
n
f if and only if f
n
f in measure.
Solution. Nonnegativity and symmetry are obvious. To prove positivity,
suppose f ,= g. Then there must be a set E of positive measure and a > 0
such that [f g[ > on E. Then for any , x : [f(x) g(x)[ >
(E), which implies d(f, g) min(, (E)) > 0. For the Triangle Inequality,
let > 0. Choose
fh
such that
fh
+ (x : [f(x) h(x)[ >
fh
) <
d(f, h) + and similarly for
hg
. Now at least one of [f h[ >
fh
and
[hg[ >
hg
must be true if [f g[ >
fh
+
hg
, so that (x : [f(x)g(x)[ >

fh
+
hg
) (x : [f(x) h(x)[ >
fh
) + (x : [h(x) g(x)[ >
hg
).
Thus, if we let =
fh
+
hg
, we have
+(x : [f(x) g(x)[ > ) d(f, h) +d(h, g) + 2.
Taking the inmum over can only decrease the expression on the left, so
d(f, g) d(f, h) + d(h, g) + 2. This is true for all > 0, so d(f, g)
d(f, h) +d(h, g). Hence d is a metric.
Suppose f
n
f in measure. Let > 0. Then (x : [f
n
(x) f(x)[ >

2
) <

2
for suciently large n. This implies d(f
n
, f) for suciently
large n. This is true for any > 0, so d(f
n
, f) 0.
Conversely, suppose f
n
, f in measure. Then there exists > 0 and
a number > 0 such that (x : [f
n
(x) f(x)[ > ) > for innitely
many n. The same computation used above to show positivity shows that
d(f
n
, f) min(, ) > 0 for these values of n, so that d(f
n
, f) , 0.
Problem 4: Prove (one direction of) the Arzela-Ascoli Theorem: Suppose
f
n
: [0, 1] R is a sequence of functions such that
(1) M : x [0, 1], n 1, [f
n
(x)[ M
(2) > 0, > 0 : n 1, [x y[ < [f
n
(x) f
n
(y)[ < .
Then f
n
has a subsequence which converges uniformly.
ANALYSIS QUALS 33
Solution. First, note that a subsequence of the f
n
must converge at every
rational point in [0, 1]. To see this, let r
1
, r
2
, . . . be an enumeration of the
rationals. Then for each n, the real number f
n
(r
1
) lies in the compact set
[M, M], so a subsequence of these numbers converges. Take the corre-
sponding subsequence of the functions f
n
and consider their values at r
2
.
These values also lie in the compact set [M, M], so a subsequence must
converge. Continuing in this manner, we then take the diagonal subse-
quence containing the rst term of the rst subsequence, the second term
of the second, and so on. This sequence will be convergent at every rational
point.
Now let f
j
k
denote a subsequence of the f
n
which converges at every
rational point. We will prove that this subsequence is uniformly Cauchy.
Let > 0 and choose such that [x y[ < [f
j
(x) f
j
(y)[ < /3 for all
j. Let L be a positive integer with
1
L
< . Since f
j
k
(x) is Cauchy at each
of the rational points 0,
1
L
,
2
L
, . . . , 1, N such that [f
jm
(
p
L
) f
jn
(
p
L
)[ < /3
for p = 0, . . . , L. Then for any x [0, 1], choose p such that [x
p
L
[ < /3.
For this p and for m, n > N,
[f
jm
(x)f
jn
(x)[ [f
jm
(x)f
jm
(p/L)[+[f
jm
(p/L)f
jn
(p/L)[+[f
jn
(p/L)f
jn
(x)[ < /3+/3+/3 = .

Problem 5: Suppose f L
p
(R
n
, dx) and g L
q
(R
n
, dx) where
1
p
+
1
q
= 1
and 1 < p < . Show that
f g(x) =
_
R
n
f(x y)g(y)dy
is a continuous function with
lim
|x|
f g(x) = 0.
Solution. (Folland page 241). It follows from Holders inequality that |h
1

h
2
|

|h
1
|
p
|h
2
|
q
. Let
y
f(x) = f(x y) denote the translates of f.
Using the L
p
-continuity of translation, we have
|
y
(f g) f g|
u
= |(
y
f f) g|

|
y
f f|
p
|g|
q
0
as y 0. Thus f g is continuous (in fact, uniformly continuous). Now
let f
n
and g
n
be continuous functions with compact support such that
|f
n
f|
p
0 and |g
n
g|
q
0. The convolution of two C
C
functions is
C
C
, so f g C
C
(R). Now
|f
n
g
n
|
u
|f
n
f|
p
|g
n
|
q
+|f|
p
|g
n
g|
q
0
so that f
n
g
n
f g uniformly. But the uniform limit of C
C
functions is
a C
0
function, so f g vanishes at innity.
Problem 7: Prove that any nonnegative harmonic function on R
2
is constant.
Solution. Let z = re
i
C. By Harnacks inequality,
R r
R +r
u(0) u(re
i
)
R +r
R r
u(0)
34 ANALYSIS QUALS
for any R. Letting R , u(re
i
) = u(0).
(Since Harnacks inequality almost makes this problem trivial, it might
be well to at least sketch a proof of it; simply express the harmonic function
in question in terms of its Poisson kernel and bound the Poisson kernel by
Rr
R+r
from below and
R+r
Rr
from above.)
Problem 8: Compute the limit
lim
R
_
R
0
e
ix
2
dx.
You may use the fact that
lim
R
_
R
0
e
x
2
dx =
1
2

without proof.
Solution. See Fall 2002 Problem 14. The answer is
_

8
+i
_

8
.

Problem 11: Let =

1
100
and let U be an -neighborhood of the spiral e
i
:
0 4 and let O be an -neighborhood of the spiral 2e
i
: 0
2. Let f : U C and g : O C be the corresponding analytic
continuations of
z log
_
cos(z)
1 z
2
_
from the -neighborhood of the origin on the real axis such that f(0) = 0 =
g(0). Find the imaginary part of f(4) g(4).
Problem 12: Prove that the innite product

n=0
(1 +z
2
n
)
converges and equals (1 z)
1
for all z in the open unit disc.
Solution. Since

[z[
2
n

[z[
n
converges in the unit disc (the inequality comes from the fact that its a
subsum), the innite product is absolutely convergent for [z[ < 1. Then

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

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

j=0
z
j
=

n=0
z
n
=
1
1 z
.

ANALYSIS QUALS 35
Fall 2006
Problem 1: Let [0, 1] be an irrational number. Let X by the unit circle
e
2it
: t [0, 1] endowed with the arclength measure, and let : X X
be the rotation by 2, given by (e
2t
) = e
2i(t+)
.
(1) Show that if f L
2
(X) and f = f, then f is a.e. constant.
(2) Use part 1 to show that if Y X is a Lebesgue-measurable subset so
that (Y ) = Y , then either Y has measure zero, or XY has measure
zero.
Solution.
(1) By denition of the Fourier transform,

f(n) =
_
1
0
f(e
2ix
)e
2inx
dx
and

f (n) =
_
1
0
f(e
2i(x+)
)e
2inx
dx
= e
2in
_
1
0
f(e
2i(x+)
)e
2in(x+)
dx
= e
2in

f(n)
by the periodicity of f. Since f = f, this implies e
2in
f(n) =

f(n).
But since is irrational, e
2in
,= 1 for n ,= 0, so all the Fourier coef-
cients of f are zero except the constant term. Since f has the same
Fourier transform as a constant, it must be constant a.e.
(Im not sure whats the most immediate way to see that Fourier in-
version must apply; one way would seem to be that L
2
L
1
for spaces
of nite measure, and inversion always works on L
1
L
2
.)
(2) Applying part (1) to
Y
(which is obviously in L
2
), we see that
Y
is
a.e. constant. If its 0 a.e, m(Y ) = 0; if
Y
= 1 a.e., m(X Y ) = 0.

Problem 2: Let E, F be two Lebesgue-measurable subsets of R, and let

E
,
F
be their respective characteristic functions.
(1) Show that the convolution
E

F
dened by

E

F
(x) =
_
R

E
(y)
F
(x y)dy
is a continuous function.
(2) Show that
n(
E

[0,1/n]
)
E
as n pointwise a.e.
Solution.
(1) For this part I will assume that at least one of E and F has nite
measure, since otherwise the convolution might not be nite-valued.
36 ANALYSIS QUALS
WLOG, m(F) < , so
F
L
1
. We have
[
E

F
(x +)
E

F
(x)[ =

_
R

E
(y)
F
(x + y)dx
_
R

E
(y)
F
(x y)dx

_
R
[
E
(y) (
F
(x + y)
F
(x y))[ dy

_
R
[
F
(x + y)
F
(x y)[ dy
= |
F
(x +)
F
(x)|
L
1
by translation and reection invariance. Since translation acts contin-
uously on L
1
(as is easily proved by approximating L
1
functions with
continuous functions of compact support), this tends to 0 as 0.
(2) Using the fact that f g = g f (which follows from Fubinis theorem),
n(
E

[0,1/n]
)(x) = n
_
R

[0,1/n]
(y)
E
(x y)dy
=
1
1/n
_
1/n
0

E
(x y)dy
By the denition of Lebesgue points, this will tend to
E
(x) for every x
which is a Lebesgue point of
E
. But almost every point is a Lebesgue
point, so the result follows.
(Note: The fact that almost every point is a Lebesgue point is generally
true only for L
1
functions, and
E
may not be in L
1
because we
dont know that m(E) < . However, we can write E as a countable
union of sets with nite measure (say, its intersection with the intervals
[n, n+1)) and apply the result to them, then take a countable union.)

Problem 3: Let (X, /, ) be a measure space, (X) = 1. Let f L

(X, /, ).
Prove that
lim
p
__
[f[
p
d
_
1/p
= |f|

.
Solution. Let M = |f|

. For each p < , [f[

p
M
p
so f L
p
and
|f|
p
M. Now for any n, let E
n
= x : [f(x)[ M
1
n
. Dene

n
= (E
n
) > 0 since otherwise |f|

would be less than M. Now f(x)

(M
1
n
)
En
so
|f|
p

__
En
(M
1
n
)d
_
1/p
=
_
M
1
n
_

1/p
n
.
Since 0 <
n
1,
1/p
n
1 as p . Thus,
|f|
p
M
2
n
for suciently large p. Since this is true for any n, and since we already
know |f|
p
M for all p, we have |f|
p
|f|

.
ANALYSIS QUALS 37
Problem 4: Assume that f is a continuously dierentiable 2 periodic func-
tion on R. Show that the Fourier series

n=

f(n)e
int
, t R
is absolutely convergent, where

f(n) =
1
2
_
2
0
f(t)e
int
dt.
Solution. Since f

is continuous on the compact set R/2Z, it is in L

p
for
every p [1, ]; in particular, it is in L
2
, so

[

(n)[
2
< by Parsevals
theorem. Now

f

(n) = in

f(n) (integrate by parts), so

f(n)[ =

(n)[
1
n

_

(n)[
2
_

1
n
2
<
by the Cauchy-Schwarz inequality.
Problem 5: Let
2
be the space of all square-summable sequences of complex
numbers, and let T :
2

2
be a linear operator. Let e
n
be the sequence
e
n
= (00 . . . 010 . . . ),
where 1 is in the nth position. Let a
nm
= Te
m
, e
n
) be the matrix coe-
cients of T.
(1) Assume that

n,m=1
[a
nm
[
2
< . Show that T is a bounded operator.
(2) Assume instead that sup[a
nm
[ : 1 n, m < is nite. Must T be
bounded?
Solution.
(1) Let v =

n
c
n
e
n
, so |v|
2
=

n
[c
n
[
2
. Then
|Tv|
2
=
_
_
_
_
_

m,n
c
n
a
mn
e
n
_
_
_
_
_
2
=

m,n
[c
n
a
mn
[
2
=

n
[c
n
[
2
_

m
[a
mn
[
2
_
< M

n
[c
n
[
2
= M|v|
2
where M =

m,n
[a
mn
[
2
< .
(2) In this case T need not be bounded. Let
a
nm
=
_
1 m n
0 else
.
I.e. T(e
n
) = e
1
+ e
2
+ + e
n
. In this case sup[a
mn
[ = 1 < , but
|T(e
n
)| =

n is not bounded.

Problem 6: Let D = z : [z[ < 1 be the unit disk in the complex plane,
endowed with the usual Lebesgue measure. Let H = L
2
(D) be the space of
square-integrable complex-valued functions on D.
(1) Show that the functions z
n
: n 0 are orthogonal in H.
(2) Do the functions |z
n
|
1
L
2
(D)
: n 0 form an orthonormal basis for
H?
Solution.
38 ANALYSIS QUALS
(1) In polar form, z
n
= r
n
e
in
. Then
z
n
, z
m
) =
__
D
(r
n
e
in
)(r
m
e
im
)rdrd =
__
1
0
r
m+n
dr
___
2
0
e
i(nm)
d
_
.
For n ,= m the latter integral is zero. Hence z
n
and z
m
are orthogonal
for m ,= n.
(2) No. Consider the function f(re
i
) = r. Then f, z
n
) = 0 for n > 0,
but f is not equal a.e. to a constant function.

Problem 7: Let f : C C be entire and injective (i.e. univalent). Prove

that f is linear: f(z) = az +b.
Solution. If f has innitely many nonzero terms in its power series expan-
sion, then it has an essential singularity at ; by Picards theorem, f then
takes on all values (except possibly one) in every neighborhood of . But
then it cannot be injective (take any neighborhood of , choose any two
points outside the neighborhood, and at least one of their images must be
hit by something in the neighborhood as well). Hence f has only nitely
many nonzero terms in its Taylor series, so f is a polynomial. If f has
degree n > 2, f has n roots; unless theyre all the same, this contradicts
injectivity. If they are all the same, f(z) = c(z z
0
)
n
for some c and z
0
.
But then f(z
0
+1) = f(z
0
+e
2i/n
) so f is not injective. Hence the degree
of f must be at most 1. It cannot be zero because constant functions are
not injective, so f(z) = az +b with a ,= 0.
Problem 8: Suppose that f is holomorphic on the open unit disk D = z :
[z[ < 1. Suppose also that for z D, Re(f(z)) > 0 and that f(0) = 1.
Prove that for all z D,
[f(z)[
1 +[z[
1 [z[
.
Solution. Dene
g(z) =
f(z) 1
f(z) + 1
.
Then since Re(f) > 0, g is analytic on D and [g(z)[ 1 on D. Also note
that g(0) = 0. By the Schwarz lemma, [g(z)[ [z[ on D. Then
[f(z)[ 1 [f(z) 1[ [z[[1 +f(z)[ [z[ +[z[[f(z)[ [f(z)[
1 +[z[
1 [z[
by the Triangle Inequality.
Problem 9: Evaluate
_

0
d
2 + cos
using the residue theorem.
Solution. By the periodicity of the cosine, this is half of
_
2
0
d
2 + cos
.
ANALYSIS QUALS 39
To evaluate this integral, let z = e
i
. Then d =
dz
iz
and the integral
becomes a contour integral around the unit circle, namely
_
dz
iz
1
2 +
z+1/z
2
= 2i
_
dz
(z + 2)
2
3
.
The singularity inside the unit circle is at z = 2 +

3, with residue
1
(z +

3 + 2)
=
1
2

3
,
so by the Residue Theorem,
_

0
d
2 + cos
=
1
2
(2i)
_
2i
1
2

3
_
=

3
.

Problem 10: Find an explicit conformal mapping from the slit disk
S = z : [z[ < 1 and z / [
1
2
, 1)
onto the disk D = z : [z[ < 1. (You may express your answer as a
composition of explicit maps.)
Solution. The map z i
1+z
1z
takes this region to the UHP minus the slit
from 3i to i; then z
1
z/3
takes this to the UHP minus the slit from 0
to i. Then z

z
2
+ 1 takes this to the UHP, and z
zi
z+i
takes this to
the unit disc.
Problem 11: Show that if C satises 0 < [[ < 1 and n Z
+
=
1, 2, 3, . . . then the equation
e
z
(z 1)
n
=
has exactly n simple roots in the right half-plane z : Re(z) > 0.
Solution. First, we note that
d
dz
e
z
(z 1)
n
= e
z
(z 1)
n1
(z + n 1) ,= 0
in the right half-plane, so all roots must be simple. Next, for z in the
RHP but on or outside the circle [z 1[ = 1, [e
z
(z 1)
n
[ = [e
z
[[z 1[
n
=
e
Re(z)
[z 1[
n
> 1 > [[ so there can be no roots in this region. Finally, on
the circle [z 1[ = 1, since [e
z
[ 1, [e
z
(z 1)
n
[ 1, so Rouches theorem
guarantees that e
z
(z 1)
n
and e
z
(z 1)
n
have the same number of
zeros inside the circle; the former clearly has a zero of order n at 1 and no
other zeros, so e
z
(z 1)
n
has n simple zeros inside the circle [z 1[ = 1
and no others in the RHP.
Problem 12: For a
n
= 1
1
n
2
let
f(z) =

n=1
a
n
z
1 a
n
z
.
(1) Show that f denes a holomorphic function on D = z : [z[ < 1.
(2) Prove that f does not have an analytic continuation to any larger disk
z : [z[ < r where r > 1.
Solution.
40 ANALYSIS QUALS
(1) A little algebra reveals that
a
n
z
1 a
n
z
= 1
1 +z
(1 z)n
2
+z
.
Now

n=1
[1 +z[
[(1 z)n
2
+z[
converges (one way to see this is that the denominator is eventually
more than [1 z[n
2
1, and then applying the integral test), so the
product converges absolutely.
(2) Note that f(a
n
) = 0 and a
n
1. If f extends to an analytic function
on a region including 1, the set of zeros has an accumulation point,
so the function would have to be constant. But f(0) = 1 ,= 0, a
contradiction.