You are on page 1of 51

Analysis Qual Problems

Sarah Constantin
August 23, 2011
1 Dec. 13, 2002
1. Suppose K is a closed subset of R
2
such that K is neither the empty set nor the whole
plane. Prove there exists a function f C
3
(R
2
) such that f = 0 on K, f > 0 everywhere
on K
c
.
Its possible to have a smooth function of compact support: for example, f(x) = e
1/(1x
2
)
when |x| 1 and 0 otherwise.
Given a point x not in K, take a closed disc C
x
around x, and a smooth function f
x
supported on C
x
, with supremum equal to 1. Because K
c
is open, it is the countable union
of open discs. We can choose a countable sequence {x
i
} and discs around them so that
K
c
= C
x
i
. Let our smooth function f be the sum

2
i
f
x
i
. Since each f
x
i
is bounded
above by 1, this converges to a smooth function. By construction, it is zero on K and
nonzero outside.
2. Consider the Volterra operator T : L
2
([0, 1]) L
2
([0, 1]),
Tf(x) =
_
x
0
f(y)dy
a. Show T is compact.
Tf(x) =
_
1
0

[0,x]
f(y)dy

[0,x]
is square-integrable, therefore T is compact.
b. Prove that T has no eigenvalues.
1
Suppose T had an eigenvalue:
Tf =
_
x
0
f(y)dy = f
f is continuous, so the integral is continuously dierentiable; this implies that f is continu-
ously dierentiable. Dierentiating we get
f(x) = f

(x)
f(0) = 0
The only possible solution to this dierential equation is f = 0, so K has no eigenval-
ues.
c. Compute the spectral radius r(T) = sup ||, where (T) is the spectrum of T. The
spectrum of T consists of those so that
(I T)
1
is not a bounded linear operator: Because T is compact, every nonzero element in the
spectrum is an eigenvalue. Since it has no eigenvalues, its spectrum must consist only of
the point zero, and thus its spectral radius must be zero.
d. Show that the range of T is contained in {u C([0, 1])|u(0) = 0}. First, Tf(0) =
0, because Tf(0) =
_
0
0
f(y)dy = 0. Second, we claim Tf is continuous; if x
n
0,
|
_
xn
0
f(y)dy| max
[0,1]
f(x)x
n
0 so Tf is continuous.
Let B be a Banach space and T : B B

a linear operator such that


< Tx, x > 0
for any x B. Show that T is continuous.
If A : X Y is a linear transformation, then the graph of A is closed i when x
n
0,
and Ax
n
y it must be that y is 0.
Suppose x
n
0 and Tx
n
y, some bounded linear functional on B. To show that T is
continuous, its enough to show that y is the zero functional. The graph of T is dened
as {x Tx}. We claim that the graph is closed. Suppose (x
n

1
Tx
n
) 0 y, that is,
|x
n
0|
B
+|Tx
n
y|
B
0.
|Tx
n
y|
B
= max
x
|Tx
n
(x) y(x)|
B
= max
x
|(Tx
n
y)(x)|
2
max
xn
|Tx
n
(x
n
) y(x
n
)|
since Tx
n
(x
n
) 0 by assumption. But x
n
0, so as n this goes to 0. So we have
y = 0 as was to be proven. So the graph is closed, and thus by the Closed Graph Theorem
T is continuous.
2 Dec. 15, 2003
1. Prove there is a subset E of R such that E is uncountable and E contains no rational
points.
The irrationals are uncountable. Indeed, the rational numbers have measure 0, because
theyre countable, and so the irrationals have positive measure, and thus cannot be a
countable set.
2. Let |F(z)| 1, F an analytic function, F(1/2) = F(1/2) = 0. Prove |F(0)|
1/2.
We apply Jensens Formula.
log F(0) =
_
2
0
log |F(e
i
)|d
n

k=1
log(
1
|a
k
|
)
where a
k
are the zeros of the function. The rst term on the right hand side equals 0,
since |F(z)| = 1 on the boundary of the circle by the maximum modulus principle. So we
have
log(F(0)) log 1/2 log 1/2 = log 1/4
so
|F(0)| 1/4
as desired.
3. Suppose f L
1
[0, 1] is positive, real valued, measurable. Show that
lim
p0
+
(
_
1
0
f(x)
p
dx)
1/p
= e
(
_
1
0
log f(x)dx)
3
The above expression equals
e
1/p log
R
1
0
f(x)
p
dx
= e
g
where
g
p
=
log
_
1
0
f(x)
p
p
We compute the limit as p 0. Applying LHopitals rule, we get
lim
p0
e
gp
= exp[
_
1
0
d
dp
e
p log x
dx
_
1
0
f(x)
p
dx
]
Setting p = 0, we get
e
R
1
0
log f(x)dx
as desired.
4. Let be a nite complex Borel measure on the unit circle such that
_
2
0
d = 0. Prove
that is the zero measure.
It is enough to show that (S) = 0 for all intervals [a, b] on the circle, because measurable
sets can be written as limits of unions of such intervals. By the Stone-Weierstrass Theorem,
functions of the form e
in
are dense in C(0, 2) because C(0, 2) is separable, contains the
identity, and is closed under complex conjugation. Thus, if we let the linear functional f
denote f(g) =
_
gd, f is zero on a dense set, so f is the zero functional. By the Riesz
Representation Theorem, then = 0.
5. Let be a -nite positive measure on a space A and let F(x, y) be AA-measurable.
Prove the Minkowski integral inequality
_
A
(
_
A
F(x, y)d(x))
p
d(y))
1/p

_
A
(
_
A
F(x, y)
p
d(y))
1/p
d(x)
Let H =
_
A
F(x, y)d(y). We claim
_
A
H
p
dx
_
(
_
F(x, y)
p
d(y))
1/p
d(x).
_
H
p
d(x) =
_
H H
p1
d(x)
(
_
|H|
p
)
1/p
(
_
|H|
p1p/p1
)
p1
p
4
observing that 1 1/p =
p1
p
. By Tonellis theorem,
_
A
_
A
F(x, y)H
p1
dxdy
_
A
(
_
A
F(x, y)
p
d(x))
1/p
(
_
A
F(x, y)
p1p/(p1)
)
p1
p
=
_
A
(
_
F(x, y)
p
d(x))
1/p
(
_
F(x, y)
p
)
p1
p

_
A
(
_
A
F(x, y)
p
dx
1/p
7. Let ||T||

= 1.
a. Let X be a Banach space, T a self-adjoint operator. Then (prove or disprove) there
exists y X such that ||y|| = ||T(y)|| = 1.
For any self-adjoint bounded operator
||T||

= sup{||Ty|| : y X, ||y|| = 1}.


Because X is a Banach space and therefore complete, argmax||Ty|| X.
Proof of the claim:
sup((Ty, y) : ||y|| = 1) ||Ty||
but if ||h|| = ||g|| = 1,
(T(h g), h g) = (Th, h) + (Tg, g) (Tg, h) (Th, g)
= (Th, h) (Th, g) (g, T h) + (Tg, g)
= (Th, h) 2Re(Th, g) + (Tg, g)
Subtract one of these expressions from the other:
4Re(Th, g) = (T(h +g), h +g)) (T(h g), h g)
Since
(Tf, f) M||f||
2
where M is the operator norm of T,
4Re(Th, g) M(||h +g||
2
+||h g||
2
)
= 2M(||h||
2
+||g||
2
)
5
= 4M
This proves the claim.
b. If T is compact,, is it true that there exists y X such that ||y|| = ||T(y)|| = 1?
Consider the Volterra operator Tf(x) =
_
x
0
f(y)dy. This is a compact operator. If ||y|| = 1
on L
2
([0, 1]),
_
1
0
|f|
2
= 1 i f = 1 a. e.
Tf(x) =
_
x
0
f(y)dy =
_
x
0
1dy = x
||Tf(x)|| =
_
1
0
|x|
2
dx < 1
for x < 1. Thus there is no such y with ||y|| = 1 and ||T(y)|| = 1.
9. Let g be a distribution on the Schwartz space whose Fourier transform is a function
F,
|F(x)| (1 +x
2
)
1
.
Is g a function such that g L
4
?
We dene the Fourier transform of a distribution g as g(

) for all Schwartz functions .


The Fourier inversion formula holds for distributions: < u, >=< u,

> and the Fourier
transform is an isomorphism from L
2
to L
2
if the distribution is an L
2
function:
|| u||
2
= ||u||
2
Thus, we observe that since
_
|F(x)|
2
dx
_
1
(1 +x
2
)
2
<
g is a function in L
2
. But
||g
4
||
1
||g
2
||
2
||g
2
||
2
by Cauchy-Schwarz

_
(
1
(1 +x
2
)
2
dx)
2
(
_
1
(1 +x
2
)
2
dx)
2
<
so _
|g|
4
dx <
and thus ||g||
4
< .
6
3 Dec. 11, 2003
1. Consider the closed ball B = {f L
2
(0, 1) :
_
f
2
1}
a. Prove that B is a closed subset of L
1
(0, 1).
First we prove that B is a subset of L
1
(0, 1). Because B is a compact set, L
2
L
1
,
because _
|f|dx =
_
1 fdx (
_
1dx)
1/2
(
_
|f|
2
dx)
1/2
< .
Next we show that B is closed as a subset of L
1
(0, 1). If
_
1
0
|f
n
f|dx 0 and
_
1
0
|f
n
|
2
dx
1, we claim
_
1
0
|f|
2
dx 1.
If ||f
n
f||
1
0, then
_
|f
n
f|dx 0
We claim a
k
(n) =

f
n
(k)

f(k) = a
k
. Indeed,
_
1
0
(f
n
(x) f(x))e
ikx
dx
_
1
0
|f
n
(x) f(x)|dx 0
So
||
N

k=1
a
k
(n)
N

k=1
a
k
||
2
,
which by Parsevals formula implies that
||f
n
f||
2
0
because we can allow N to go to innity since the tails where k > N are uniformly
bounded in L
2
for all n.
b. Prove that B has empty interior as a subset of L
1
(0, 1). T Given an arbitrary , we
show there is no f B such that for all g with
_
|f(x) g(x) < , we have g B.
For any > 0, we construct g as follows. g(x) = f(x) except on a set E

of small measure
where |g(x) f(x)| = N.
||g||
2
= ||g||
2
E

+||f||
2
[0,1]E

= N
2
+ 2N
_
E

Re(f)dx +||f||
2
N
2
2N

, because ||f|| 1.
7
Choose =
2
/9and N = 9/. Then
_
1
0
|f g|dx =
_
E

Ndx N =
but
_
1
0
|g|
2
dx N
2
2N

= 3 > 1
So g can be chosen arbitrarily close to f in L
1
-norm, but still have L
2
-norm greater than
1; this shows that there are no interior points in B.
2. Let
n
be a sequence of continuous functions on the torus T.
a.) Suppose that for every continuous function f on T, the sequence of convolutions
f
n
(x) =
_

n
(x y)f(y)dy
has the property that lim
n0
f
n
(0) exists. Prove that for some K,
_

|
n
(x)|dx
K.
Consider
n
f =
_

n
(x y)f(y)dy as a family of functionals on C(T).
f
n
(0) : f
_

n
(y)f(y)dy
The norm of the functional
n
is the L
1
norm of
n
,
_
|
n
(x)|dx. Indeed,
||||

= sup
||f||=1
_
f(x)(x)dx sup |f|||||
L
1 = ||||
L
1
Vice versa, there exists f so that
_
f(x)(x)dx = 1, because there exists a sequence
of continuous functions converging to /|| = sign() when is nonzero. Note that
_
|| =
_
sign(). Denote by s
n
the sequence of continuous functions converging to
sign(). Then
_
s
n

_
||dx
by the Dominated Convergence Theorem, so
_
s
n
|s
n
|||||

8
If lim
n
f
n
(0) exists, then
n
considered as elements of the dual of C( T) are pointwise
convergent, hence pointwise bounded. So by the Banach-Steinhaus theorem, the norms of
these functionals are uniformly bounded.
sup
n
||
n
|| <
, in other words,
_

|
n
(x)|dx < K
for some K.
b. Show that there is a continuous function f on T such that the Fourier series of f does
not converge at x = 0.
Consider
S
n
=

|k|n
e
ixk
a
k
(f)
If D
n
is the Dirichlet kernel

N
N
e
inx
then
S
n
(f)(0) =
_
D
n
(y)f(y)dy
If this always converged as n then the norms of D
n
in L
1
would have been bounded,
as in part a., but we know they are not, by comparing with harmonic series:
_
|
sin((n + 1/2)x)
sin(x/2)
|dx C
_
|
sin((n + 1/2)x)
x
dx
C log n
which diverges.
So for some f, sup
N
|
N
(f)| = , which means the Fourier series does not converge at
0.
3. For functions u C

0
(R), dene the principal value integral against 1/x by
< f, u >= lim
0
_
<|x|<1/
u(x)
x
dx
Prove that f extends to a tempered distribution.
If is in the Schwartz space, we claim that the limit
lim
0
_
<|x|<1/
(x)
x
dx
9
exists and is bounded by a continuous seminorm in S.
This integral is equal to
_

1/
(x)
x
dx +
_
1/

(x)
x
dx
=
_

1/
(x) (0)
x
dx +
_
1/

(x) (0)
x
+
_
<|x|<1/
(0)
x
dx
The last term is zero because 1/x is an odd function and the positive and negative parts
cancel. Because is dierentiable, the rst and second term converge as x 0. Let
(x) =
(x)(0)
x
for x = 0 and

(0) at x = 0. We express the above integral in terms of


.
_
1
1
(x)dx
_

(x)dx +
_
1|x|1/
(x)dx
The rst term is a constant, and the second term goes to 0 as 0. The last term is
_
1|x|<1/
(x) (0
x
dx =
_
1<|x|<1/
(x)
x
dx
If (x) is in the Schwartz space, the limit exists.
7. Let P be a polynomial in z with complex coecients. Prove that if all the roots of P
lie in the upper half plane then the roots of P

lie in the upper half-plane.


Proof. Write P(z) = (z a
1
) . . . (z a
n
). Then
P

(z)
P(z)
=
1
z a
1
+. . . +
1
z a
n
Suppose z
0
is a root of P

in the lower half-plane, z


0
= x
0
+iy
0
. Then
1
z
0
a
i
=
1
(x
0
x
i
) +i(y
0
y
i
)
=
(x
0
x
i
) +i(y
i
y
0
)
|z
0
a
k
|
2
This is a complex number with positive imaginary part, so P

(z
0
)/P(z
0
) is a complex
number with positive imaginary part. On the other hand, by construction it must be zero,
which is a contradiction. So all the roots of P

(z) must lie in the upper half-plane.


10
4 May 9, 2003
1. Let m
n
be Lebesgue measure on R
n
. Construct an m
2
-measurable subset of the plane
such that its orthogonal projection on any straight line is not m
1
-measurable.
Let our set X be the Vitali non-measurable set along the line y = 0 R
n
. That is, it is
chosen by taking a representative of each class, of the form E
r
[0, 1] = Q+r [0, 1] where
r is a real number. This set is non-measurable with respect to m
1
since its translates by
rational numbers are all contained in [1, 2] but are all disjoint, so if it had one-dimensional
Lebesgue measure, applying sigma-additivity results in a contradiction. Because X is a
subset of the unit interval, it has zero 2-dimensional Lebesgue measure (since it is a subset
of a zero-measure set.) The projection of X onto any straight line has measure m
2
(X cos )
where is the angle between the two lines. Since Lebesgue measure is invariant under
translations and rotations, and m
n
(X) = m
n
(X), we know that such projections must
also be non-measurable, except the projection onto the line perpendicular to X. So let
X

= X 2 X

where X

is the Vitali set along the y-axis. The projection of this


L-shaped set along any line contains one scaled Vitali set, and so is non-measurable.
2. Let f
n
be a sequence of increasing functions on [0, 1] and let f also be an increasing
function on [0, 1]. Assume that f
n
(0) = f(0) = 0 and f
n
(1) = f(1) = 1. Show that the
following are equivalent: 1. f
n
f a.e. on [0, 1] 2. C[0, 1],
_
1
0
(t)df
n
(t) converges
to
_
1
0
(t)df(t). Since f
n
are increasing, this integral can be given as
inf
t
i

i
_
1
0
(t)[f
n
(t
i
) f
n
(t
i1
)]dt
where t
i
are partitions of the unit interval. For a xed partition and a xed i,
_
1
0
(t)f
n
(t
i
) f
n
(t
i1
)dt =
_
1
0
(t)f
n
(t
i
)dt
_
1
0
(t)f
n
(t
i1
)dt
if f
n
f a.e., since all f
n
are dominated by 1, the Dominated Convergence Theorem
implies that the above integrals converge. Conversely, if the above integrals converge for
every C[0, 1], then one can construct a sequence of continuous test functions
k
such
that for each k
_

k
(t)f
n
(t
i
)dt
_
1
0

k
(t)f
n
(t
i1
)dt
_

k
(t)f(t
i
)dt
_
1
0

k
(t)f
n
(t
i1
)dt
If we consider the
k
as distributions, and let
k,s
= 1/k(1/k(t s)) then letting k 0,

t
if we choose to be a test function with integral 1; so for any choice of s,
_

k,s
(t)f
n
(t)dt
_

k,s
f(t)dt
11
implies f
n
(s) f(s).
4. Let L
2
a
(D) be the space of all analytic functions on the unit disc integrable with respect
to area. Let <> be the standard inner product in L
2
a
,
< f, g >=
_ _
f(z)g(z)dA
For each z
0
D nd g
z
0
L
2
a
such that for any f, < f, g
z
0
= f(z
0
). The monomials z
n
are basis elements for this set; by denition any analytic function can be approximated in
polynomials, and
_
z
n
z
m
dA =
_ _
r
nm
e
inim
rdrd = 0
for n = m. So the z
n
both span and are orthogonal. If we replace each z
n
with y
n
= Cz
n
,
normalized so that < y
n
, y
n
>= 1, then the y
n
form an orthonormal basis. Let (z) =

n=0
y
n
z
n
0
. Then the inner product with any f(z) = a
0
+a
1
y +a
2
y
2
+. . . is

a
0
+a
1
z
0
< y, y > +a
2
z
2
0
< y
2
, y
2
> +. . .
= f(z
0
).
Thus, the correct kernel to integrate against is

n=0
y
n
z
n
0
=
1
1zz
0
. We know this sum
converges because both z
0
and y have norm less than one; z
0
because its in the unit disc,
and y because its
1
sqrtn+1
z which has norm less than one since z is in the unit disc.
Construct a conformal map of the unit disc onto D [1, 0.5]. Map 1: inverse Cayley
transform, unit disc to plane, i
1+z
1z
Under this map, the interval [-1, -0.5] goes to the
vertical line [0, 1/3 i]. Map 2: z
1/2
yielding the plane minus the negative y-axis. Map
3: shift up by 1/3, yielding the plane minus [1/3i, i]. Map 4: z
2
, yielding the upper
half-plane minus the interval [0, 1/3 i] Map 5: Cayley transform,
iz
i+z
yields unit disc minus
[1, 0.5].
6. Let f be meromorphic in a neighborhood of the unit disk, and assume that |f(z)| = 1 if
|z| = 1. Prove that f is a rational function.
First, if f has no zeros in the disc, then f(z) = c, a constant function; indeed, g = 1/f
is holomorphic, so by the maximum modulus principle, g achieves its maximum on the
boundary |z| = 1. Since the maximum of g is the minimum of f, f achieves both its
12
maximum and minimum on |z| = 1; since |f(z)| = 1 for all |z| = 1, we conclude that |f(z)|
is constant so f(z) is constant.
Now suppose f has zeros in the disc, say a
1
. . . a
n
. (It couldnt be innitely many, since
the disc is compact, so there would be an accumulation point, and f would be identically
zero.) Consider
F(z) =
n

i=1
1 a
i
z
z a
i
the inverse of the Blaschke product of f. Each term in the product has norm 1. So fF is
holomorphic on D and |f(z)F(z)| = 1 for |z| = 1 and fF has no zeros in D. By the above
argument, that means fF is a constant. So f is the product of Blaschke factors times a
constant; so f is a rational function.
7. Construct a function on the unit square [0, 1] [0, 1] such that
g(y) =
_
1
0
f(x, y)dx
exists for every y,
h(x) =
_
1
0
f(x, y)dy
exists for every x, both g and h are integrable, but
_
1
0
g(t)dt =
_
1
0
h(t)dt
Divide the unit square into rectangles as follows. Let I
1
= [1/2, 1], I
2
= [1/4, 1/2] ,
I
k
= [2
k
, 2
(k1)
]. Partition the unit square into rectangles I
j
I
k
. Let g
k
be a hat
function on the interval I
k
, a piecewise-linear function such that g
k
(2
k
) = g
k
(2
(k1)
= 0,
g
k
(2
k
+ 2
k+1
) = 2
k+1
so that the integral equals 1. Then let f on I
j
I
j+1
equal
g
j
(x)g
j+1
(y), let f on I
j
I
j
= g
j
(x)g
j
(y) and let f equal zero elsewhere. Then for any y,
_
1
0
f(x, y)dy = g
n
(y)
_
I
j1
g
j1
(x)dx g
n
(y)
_
I
j
g
j
(x)dx
= g
n
(y) g
n
(y) = 0.
However, for any x,
_
1
0
f(x, y)dy = 0 except for x I
1
, when the integral is 1. Hence, by
integrating twice,
_
1
0
dx
_
1
0
dyf(x, y) =
_
I
1
g
1
(x)dx = 1
instead of 0 from integrating in the opposite order.
13
9. Let K(x, y) be a real-valued continuous function on the unit square and suppose
K(x, y) = K(y, x). Dene
Tf(x) =
_
1
0
K(x, y)f(y)dy
for f L
2
([0, 1]), dx).
a. Prove T is a bounded operator on L
2
||Tf||
2
= (
_
1
0
(
_
1
0
K(x, y)f(y)dy)
2
dx)
1/2

_
1
0
(
_
1
0
K(x, y)
2
dy)
1/2
(
_
1
0
f(y)
2
dy)
1/2
dx C||f||
2
.
b. Prove or disprove the following statement: T
2
T Range(T) is nite. This is
true. If T
2
T = 0, then T satises a minimal polynomial, and thus can only have
two eigenvalues, at the roots of this polynomial; since T is compact and self-adjoint, its
spectrum consists only of its eigenvalues, so T =
1
P
1
+
2
P
2
where P
1
and P
2
are nite-
dimensional eigenspaces, by the Spectral Theorem.
10. Suppose f L
1
and

f(x) = F(x).
|f(z)|
e
|z|
1 +x
2
for all z. Prove that f has compact support. We have e

|z|F L
1
so the inverse Fourier
transform of e
|z|
F vanishes at innity, by the Riemann-Lebesgue lemma.
If a = A,

b = B, then (AB)

= A

. In our case, this means the inverse Fourier


transform of e
|z|
F is
f ((e
|z|
)

)
= f 2
_

0
e
(1+ix)
d
= f 2e

=
_

f(y )e
y
dy = g()
We know this vanishes as goes to innity. [Id like to conclude from this that f has
compact support but Im stumped.]
14
5 May 29, 2004
1.a. State Fatous Lemma.
If f
n
is a sequence of non-negative measurable functions, and if liminf f
n
(x) = f(x) point-
wise, then
_
f(x)d liminf
_
f
n
(x)d.
b. State the Monotone Convergence Theorem.
If f
n
is an increasing sequence of non-negative measurable functions and f
n
f pointwise,
for some non-negative measurable function f, then
_
f
n
(x)d
_
f(x)d
c. State the Dominated Convergence Theorem.
If f
n
is a sequence of non-negative measurable functions, |f
n
| < g for some integrable
function g, then
_
f
n
d
_
fd
d. Prove the Monotone Convergence Theorem with Fatous Lemma.
If f
n
is an increasing sequence of non-negative measurable functions converging to f, then
limsup
_
f
n

_
f. By Fatous lemma,
_
fd liminf
_
f
n
d
So
liminf
_
f
n
d = limsup
_
f
n
d = lim
_
f
n
d
_
f liminf f
n
limsup
_
f
n

_
f
=
_
fd
3. Let f be dierentiable at every point of [0, 1]. Prove there is a point x
0
[0, 1] such that
df
dx
is continuous at x
0
.
15
Let f
n
(x) =
f(x+1/n)f(x)
1/n
. These functions are continuous for each n, and converge point-
wise to
df(x)
dx
. The problem now reduces to Problem 1 of Dec. 2008; showing that the limit
of a pointwise convergent sequence of continuous functions on an interval must contain a
point of continuity.
4. Prove or disprove by counterexample the following statement. If T : H H is
a bounded self-adjoint operator on a separable complex Hilbert space H then T has a
nontrivial eigenvector, i.e. there is x = 0 and = 0 in R such that
Tx = x
Consider the multiplication operator on L
2
([0, 1]):
[A](t) = t(t)
This operator is bounded:
_
1
0
|A|
2
dt =
_
1
0
|t(t)|
2
dt
=
_
1
0
|t|
2
|(t)|
2
dt
_
1
0
|(t)|
2
dt
This operator is also self-adjoint:
< A, >=
_
1
0
t(t)

(t)dt
=
_
1
0
(t)

(t)dt
since t is real
=
_
1
0
(t)A(t)dt =< , A >
But this operator has no eigenvalues; if
A(t) = (t)
t(t) = (t)
t =
for all t, so if (t) is in L
2
andnot identically zero, there exists a set of positive measure
where (t) = 0, so t = on a set of positive measure, which is a contradiction. Thus the
operator A is a counterexample.
16
5. Let A be the space of all analytic functions F(z) on the open unit disc D such that
||F||
A
= (
_ _
D
|F(z)|
2
dxdy)
1/2
Prove: if {F
n
} A, and ||F
n
||
A
1, then there exists a subsequence {F
n
k
} and a limit F,
such that F
n
k
(z) F(z) uniformly on compact sets, and ||F||
A
1.
We apply Montels Theorem, which states that the following are equivalent:
1. F
n
has a subsequence converging uniformly on compact subsets.
2. F
n
is locally bounded: i.e. for any point x D, there exists a neighborhood N of x and
a bound B such that |F
n
(y)| B for all n, and all y N.
Because functions in the Bergman space are analytic,
f(a) =
1
r
2
_ _
B(a,r)
f.
Also,
|f(a)| =
1
r
2
|
_ _
B(a,r)
f 1|

1
r
2
[
_ _
B(a,r)
|f|
2
]
1/2
[
_ _
B(a,r)
1
2
]
1/2

1
r
2
||f||
2
r

=
1

r
||f||
2
.
So |F
n
||
N
C
N
||F
n
||
2
C
N
for all n, which proves that F
n
is locally bounded. Thus,
Montels Theorem is applicable, so there exists a subsequence F
n
k
converging uniformly on
compact sets.
6. Find with proof all extreme points of the unit ball of L

([0, 1], dx).


First, we claim that if f is an extreme point, we must have ||f||

= 1. If f(x) = 1 for
any x, then |f(x)| 1 for some > 0, and f can be written as the average of f(x) +/2
and f(x) /2, both of which are in the unit ball of L

([0, 1], dx).


Now we claim that the extreme points of the unit ball are {f L

[0, 1] : |f| = 1a.e.}.


Suppose there were an extreme point f such that |f| < 1 on a set E of positive measure.
Then E = E
n
where E
n
= {x : |f(x)| 1 1/n}. These are closed sets, and their union
is a set of positive measure, so some E
n
has positive measure. For this choice of n, dene
g(x) = f(x) +
1
2n
on x E
n
and g(x) = f(x) outside, and dene h(x) = f(x)
1
2n
on
17
x E
n
and h(x) = f(x) outside. Then f(x) is the average of g(x) and h(x), but both
functions dier from f on a set of positive measure, so f cannot be an extreme point.
On the other hand, if f(x) = 1 a.e. and f(x) = tg(x) + (1 t)h(x), then
1 = |tg(x) + (1 t)h(x)| t|g(x)| + (1 t)|h(x)|
and this is impossible if either |g(x)| < 1 and |h(x)| < 1, so |g(x)| = 1a.e. and |h(x)| = 1a.e.
proving the claim.
8. Suppose F is analytic on the open unit disc D and |F(z)| 1 for z D. Suppose
furthermore that F(1/2) = F(1/2) = 0. Prove |F(0) 1/4.
We apply Jensens Formula.
log |F(0)| =
n

k=1
log(
1
|a
k
|
+
1
2
_
2
0
log |F(e
i
)|d
where a
k
are the zeroes in the unit disc. Here, |F(e
i
| log |1|d = 0, because |F(z) 1,
so the second term above vanishes.
log |F(0)| = log |1/2| log |1/2|
n

k=2
log |1/a
k
|
log 1/4
so
|F(0)| 1/4
as desired.
6 May 11, 2005
1. Using calculus of residues, calculate
_

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

0
sin
2
(x)
x
2
dx
18
First integral:
log x
(x+1)(x
2
+1)
has poles at -1, i, and -i. Residue at -i is 3(i +1)/8, residue at i is (i 1)/8,
and residue at -1 is i/2. Integrate around a contour dened as a quarter-circle in the
bottom-right quadrant, with a semicircle of radius 1 cut out around i.
The integral around the semicircle is half a residue of -i, 3/16(i + 1). The integral along
the vertical piece is
_

2
log y +i3/2
(iy + 1)(y
2
+ 1)
dy
The integral along the quarter-circle arc is
_
0
/2
log R +i
(Re
i
+ 1)(R
2
e
2i
+ 1)
d 0
as R The integral along the horizontal piece is the desired integral which we denote
I. Now we do the same residue calculation for a quarter-circle in the top right quadrant,
with a semicircle of radius 1 cut out around i. The integral around this semicircle is half a
residue, or (i 1)/16. As before, the integral around the quarter-circle arc goes to zero
as R .
0 = I (i 1)/16
_

2
log iy
(iy + 1)(y
2
+ 1)
0 = I + 3(i + 1)/16 +
_

2
log(iy)
(iy + 1)(y
2
+ 1)
0 = 2 +
2i + 4
16
+ 0
I = i/16 /8
Second integral: Let f(z) =
1e
2iz
z
2
. Observe that on the real line 1 e
2ix
= 1 cos(2x)
isin(2x) = 2sin
2
xisin(2x). Consider the contour thats a semicircle with radius R in the
upper half-plane with a semicircle of radius cut out around zero. By Cauchys theorem,
the integral around this contour is 0. As R , note that |
1e
2iz
z
2
|
2
|z|
2
so the integral
around the semicircle of radius R goes to 0. Therefore
_
|x|
1 e
2ix
x
2
dx =
_
C
1 e
2iz
z
2
Now f(z) =
2iz
z
2
+E(z) where E is bounded as z 0, while on C, we have z = e
i
, dz =
ie
i
d so
_
C
1 e
2iz
z
2
dz
_
0

(2i
2
)d = 2
19
Taking real parts yields
_

2 sin
2
x
x
2
dx = 2
Since the integrand is an even function, we have
_

0
sin
2
x
x
2
dx =

2
5. Let S
n
denote the Schwartz space of functions in n variables. Consider linear subspaces
L
+
= {f S
n
:

f = f} and L

= {f S
n
:

f = f}.
Prove that L
+
L

coincides with the subspace of even functions in S


n
.
First, every function in L
+
or L

is even, so their sum is even.. Indeed, if f =



f, f

(x) =

f(x) = f(x) so f is even, and if f =

f, f(x) = f

(x) =

f(x) = f(x) so again


f is even. Here f

denotes the inverse Fourier transform. Conversely, suppose f is even.


Then write g =
f+

f
2
and h =
f

f
2
. f = h + g. But g = 1/2(

f +

f)) and if f is even


_
f(x)e
ix
dx = f

() =

f(x) so its equal to its double Fourier transform; thus g L
+
.
Likewise,

h = 1/2(

f

f) = h so h L

. This proves the claim.


6.a. Find all distributions on R such that

= 0 where

denotes the rst derivative of


.
The only such distributions are the constant functions (f) = c
_
fd. Indeed, suppose
: C

c
R is a continuous linear functional.

(f) is dened as (f

). Let F

= f.
Then F C

c
if f is.
(f) = (F

) =

(F) = 0
Let g C

c
; we want to show (g) = c
_
gd. Let h C

c
such that
_
h = 1. Then
_
(g h
_
g) =
_
g
_
h
_
g = 0.
This implies
(g h
_
g) = 0
so
(g) = (h)
_
g = c
_
g.
20
b. Show that a translation-invariant distribution is a multiple of Lebesgue measure. A
translation-invariant distribution has the property that
(x +) (x)

= 0/ = 0
for all . We want to show that given a sequence of n = 1/ , there exists a
subsequence such that lim
n
(f(x+1/n))(f(x))
1/n
0 uniformly. This will show that
the rst derivative is zero, and therefore by part a. that is a constant times Lebesgue
measure.
We wish to apply the Arzela-Ascoli Theorem; we will have to prove that the functions
g
n
=
(f(x+1/n))(f(x))
1/n
are equicontinuous. By the Mean Value Theorem, there exists
some x
0
such that g
n
(x) = f

(x
0
). g
n
(x)| = |f

(x
0
)| sup |f

| so the g
n
are uniformly
bounded. Note that
|g
n
(x) g
n
(x
0
)| |g(x) f

(x)| +|f

(x) f

(x
0
)| +|f

(x
0
) g
n
(x
0
)|
Since all these terms go to zero uniformly as n , we can conclude that the g
n
are
equicontinuous, so there exists a subsequence of the g
n
that converges. By applying the
same procedure again, there is a subsequence of that subsequence of the g
n
whose deriva-
tives converge; continuing in this fashion we conclude that the limit in the C

c
norm of the
dierence quotients is 0.
7. Let K [0, 1] R
n
be compact and have Lebesgue measure zero. Show that there is a
real-valued continuous function F dened on R
n
K with the properties
_
[0,1]
n
e
f(x)
dx <
lim
xy
f(x) = +
for all y K
First we look for F of the form
F(x) = log h(d(x, K))
with h strictly increasing, dierentiable, lim
t0
h(t) = .
Let D be the diameter of the unit cube and = h(D). If F is of the above form,
lim
zK
F(x) = , so we just want to show that h is integrable.
21
Consider the sets K
t
in [0, 1]
n
of points of distance at most t from K. Let (t) denote the
measure of K
t
. This is a continuous nondecreasing function because K
t
K
s
if s > t, and
(0) = 0 because (K) = 0.
We can write
_
[0,1]
n
h(d(x, K))dx =
_

{x : h(d(x, K) }d
Because h is decreasing, h(d) is the same as d h
1
(), so we have
_

{x : d(x, k) h
1
()}d
=
_
(h
1
())d
Let t = h
1
(), so h(t) = , h

(t) = d, so
_
[0,1]
n
h(d(x, K)) =
_
D
0
(t)h

(t)dt
We want to nd h such that
g(t) := (t)h

(t)
is integrable on [0, D] with h(0) = . in other words
h(t) =
_
D
t
g(s)
(s)
ds
and
_
D
0
g(t)
(t)
dt =
for a given vanishing at t = 0.
For some g
k
, let g(t) = D
1
2
k+1
g
k
on the interval (D2
k1
, D2
k
) Then
_
D
0
g(t)
(t)

k=0
g
k
(2
k
because is increasing. As k , (2
k
) 0. Choose g
k
j
to be zero unless (2
k
j
)
2
k
, which is possible because (2
k
) 0. Otherwise, g
k
j
= 2
j
. Then
_
D
0
g(t)dt =

j
2
j
<
22
and
_
g(t)
(t)
dt

j
2
j
2
j
=
so this g works.
10. Let T be bounded, self-adjoint, compact operator on a separable complex Hilbert space
H. Let T
4
3T
2
+T = 0. Prove or disprove: T has nite rank.
Because T is compact and self-adjoint, the spectral theorem applies. Since T is the linear
combination of projections onto eigenspaces,

n
P
n
each eigenvalue satises

4
n
3
3
n

n
= 0
So there can be only four nonzero eigenvalues. Because every eigenspace is nite-dimensional,
T is therefore nite rank.
7 August 25, 2005
1. Let H be the space of all analytic functions F on the open unit disk, D, with norm
||F||
H
= (
_ _
D
|F(z)|
2
dxdy)
1/2
<
a. Prove that H is complete in this norm.
L
2
is a Hilbert space, and H is a subset of a Hilbert space with the induced norm. Because
a closed subspace of a Hilbert space is a Hilbert space, its enough to show that H is
closed in L
2
. Suppose ||f
n
f||
2
0. We claim that f H. In other words, that f is
analytic
First we note that for any g H,
g(z) =
1
R
2
_ _
D
R
fdxdy
by the mean value property, where D
R
is a disc of radius R around z, contained in the unit
disc. So
|g(z)|
1
R
2
_ _
|f| 1
1
R
2
||f||
2
(R
2
)
1/2
23
=
1

R
||f||
2
We use this to show that the family of functions f
n
is uniformly bounded on compact sets.
f
n
is a Cauchy sequence with respect to the L
2
norm. So f
n
is bounded in L
2
. There exists
M such that ||f
n
||
2
M.
Let K be a compact subset of the disc. Consider sup
zK
|f(z)|.
|f(z)|
1

R
||f
n
||
2

1

R
M
So f
n
is uniformly bounded on compact sets. Applying Montels Theorem gives us that
there is a subsequence of f
n
that converges uniformly on compact sets; the uniform limit
of analytic functions is analytic, so H is closed in L
2
.
b. Exhibit a complete orthonormal basis for H.
Polynomials are orthogonal in H:
_
2
0
_
1
0
(re
in
)(re
im
)drd
=
_
2
0
_
1
0
r
2
e
i(nm)
drd
= 0
Normalized polynomials z
n
/|z| form an orthonormal basis, because polynomials are dense
in the analytic functions each analytic function can be expressed as a polynomial expan-
sion.
2. Let h : mathbbR R be continuous and strictly increasing with h(0) = 0. Construct a
subset K of the unit disk in the plane having the following properties:
a.) m(K) > 0
b.) K is nowhere dense
c.)
_ _
D(0,2)K
1
h(d(x,y),k)
dxdy = .
Let f(t) =
1
h(t)
a continuous function, lim
t0
f(t) = . We want
_ _
D(0,2)K
f(d(z, k)dz =
Let K be a thickened Cantor set: it will be compact, nowhere dense, and have positive
measure. In particular, we dene K to be the intersection of K
n
, where K
1
is the unit
24
square, and each successive K
n
removes four rectangles from each rectangle in K
n1
, one
from the center of each side, a set of measure 4
n1
(2
n

2
n
) := 4
n
c
n
where c
n
is the area
of each removed rectangle. Let U be the set of removed rectangles.
_ _
U
f(d(z, k))dz
_ _
f(c
n
/2)4
n
c
n
We want this integral to be innite while still allowing K to have positive measure (in
other words

4
n
c
n
< 1.
Choose n(k) :
1
2
k
f(
1
2
k
/2
n
) > 1. Then c
n
=
1
2
k
1
4
n
satises the condition.
3. Find with proof the number of zeros of the function 3z
4
+e
z
on the closed unit disk in
the complex plane.
We apply Rouches theorem. |3z
4
| > |e
z
| within the closed unit disc, so the number of
zeros equals the number of zeros of 3z
4
, which is 4.
4. Describe all extreme points of the unit ball of L
1
([0, 1], dx).
This set has no extreme points. Any extreme points on the unit L
1
ball would be on the
boundary. Let ||f||
1
= 1; we claim that it is not an extreme point.
There exists a such that
_
a
0
|f|d = 1/2 and
_
1
a
|f|d = 1/2 by the intermediate value
theorem. Set f
1
(x) = 2f(x) for x [0, a), and f
1
(x) = 0 elsewhere. Set f
2
(x) = 2f(x) for
x [a, 1] and 0 elsewhere. Then f = 1/2f
1
+ 1/2f
2
. So f is not an extreme point because
it is a linear combination of two other points on the boundary of the ball.
5. Show how to conformally map the region R to the unit disk, where R is the upper
half-plane minus the slit from 0 to i. You may express your answer as a composition of
maps.
Map 1: clockwise rotation by /2, f
1
(x) = x e
i/2
, yields the right half-plane minus a
slit from 0 to 1.
Map 2: f
2
(x) = x
2
, yields the plane minus the slit (, 1)
Map 3: f
3
(x) = x 1, yields the plane minus the slit (, 0).
25
Map 4: f
4
(x) =

x, with the standard branch of the logarithm, yields the right half-
plane.
Map 5: f
5
(x) = xe
i/2
, rotates to yield the upper half-plane.
6. Integrate
_

0
1
(1+x
2
)
2
log xdx. We integrate over the semicircular contour of radius R,
with an indented semicircle of radius avoiding the origin. We use the branch of the
logarithm that omits the negative imaginary axis.
The poles of the function are at i, -i, and 0. The pole at i is a double pole, so its residue
is zero.
0 =
_

0
log(Re
i
)
1 + (Re
2i
)
2
iRe
i
d+
_

0
log(e
i
)
1 + (e
2i
)
2
ie
i
d+
_
R

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

R
log x
(1 +x
2
)
2
dx
The rst term goes to 0 as R goes to innity, being o(1/R
2
). The second term also goes
to zero as 0, by a LHopital argument. What remains:
_
R

log x +i
(1 +x
2
)
2
= I +i
_
R

dx
(1 +x
2
)
2
= I +iA
where A refers to
_
(1 +x
2
)
2
dx integrated over [0, ).
So
I = iA/2.
7. Exhibit with proof a Lebesgue measurable function in the plane R
2
such that
a.)
_
R
|F(x, y)|dx < for all y,
_
R
|F(x, y)|dy < for all x.
b.)
_
R
_
R
F(x, y)dxdy =
_
R
_
R
F(x, y)dydx
Let F(x, y) = 0 outside the unit square [0, 1] [0, 1]. On the unit square, let
F(x, y) =
x y
(x +y)
3
Then
_
1
0
x y
(x +y)
3
dx =
_
1
0
2y
(x +y)
3
dx +
_
1
0
x +y
(x +y)
3
dx
26
=
_
2y
2
(x +y)
2
dx +
_
1
(x +y)
2
dx
= (y(x +y)
2
(x +y)
1
)|
1
0
=
y
(1 y)
2

1
1 +y
+
y
y
2

1
y
=
y
(1 +y)
2

1
1 +y
_
1
0
y
(1 +y)
2

1
1 +y
dy
= [log(1 +y) + log(1 +y) +
1
1 +y
]
1
0
= 1/2
However, by symmetry, integrating in y rst and then x gives a result of 1/2. So Fubinis
theorem doesnt hold here.
8. a. State Fatous Lemma. If f
n
is a sequence of measurable nonnegative functions and
f = liminf f
n
, then
_
fd liminf f
n
b. State the Monotone Convergence Theorem If f
n
is an increasing sequence of measurable
nonnegative functions, and f
n
f pointwise, then
_
f
n
=
_
f
c. State the Dominated Convergence Theorem If f
n
is a sequence of nonnegative measurable
functions, f
n
f pointwise, and f
n
< g for some measurable g and all n, then
_
f
n

_
f.
c. Use Fatous lemma and/or the Monotone Convergence Theorem to prove the Dominated
Convergence Theorem.
Consider the sequences g+f
n
and gf
n
. By Fatous lemma,
_
(g+liminf f
n
) liminf
_
(g+
f
n
). Also,
_
(g + liminf(fn) =
_
(g limsup f
n
)
_
g limsup
_
f
n
. But f
n
f, so
these two conditions say
_
g +
_
f
_
g + liminf
_
f
n
_
g
_
f
_
g limsup
_
f
n
27
This implies that
limsup
_
f
n
liminf
_
f
n
so lim
_
f
n
=
_
f.
9. Suppose F is a twice continuously dierentiable function on the real line, and suppose
F, F

L
2
(R, dx).
a.) Prove F

L
2
(R, dx).
By Plancherels Theorem, F

L
2
i

L
2
.

= 2i

F = 1/(2i)

||

||
2
2
=
_

()|
2
d
=
_

4
2

2
|

F
2
|d
We know that both
_

F
2
|d <
and
_

16
4

2
|

F
2
|d <
By Holders inequality,
_

2

F
2
= ||x
2

F||
1
||x
2

F||
2
||

F||
2
<
This proves the claim.
b. Prove or disprove by an example that F

must be in L
4
(R, dx).
_
|

|
4
d =
_
16
4

4
|

F|
4
It suces to prove that F

, because then F

L
2
L

L
4
. By the Riemann-
Lebesgue Lemma, the Fourier transform of an L
1
function vanishes at innity. So we must
prove that

L
1
. In other words,
_
|2i

F|d <
28
_
|

F|d =
_
1 +
2
1 +
2
|

F|d
||

1 +
2
||
2
||(1 +
2
)

F||
2
by Holders inequality. Both terms in the product are nite, the second because F

L
2
.
8 May 15, 2006
1. Find the Fourier transform of (x
2
+ 2x + 3)e
x
2
/2
.
_

(x
2
+ 2x + 3)e
x
2
/2
e
2ix
dx
=
_

(x
2
+ 2x + 3)e
1/2(x2i)
2
e
4
2

2
dx
= e
4
2

2
[
_
x
2
e
1/2(x2i)
2
dx +
_
3e
1/2(x2i)
2
dx
= e
4
2

2
[

2 + 3

2] = e
4
2

2
[4

2]
2. Give an example of a compact non-self-adjoint operator in L
2
[0, 1].
Consider the Volterra operator V f(x) =
_
x
0
f(y)dy.
If ||f||
2
1,
||
_
x
0
f(y)||
2
=
_
1
0
|
_
x
0
f(y)dy|
2
dx
_
1
0
_
x
0
|f(y)|
2
dydx

_
1
0
xdx = 1/2
so this is a bounded operator. In general, kernel operators of the form
Kf(x) =
_
k(x, y)f(y)dy
29
if k L
2
.
V f(x) =
_
x
0
f(y)dy =
_
1
0

[
0, x]f(y)dy
||
[
0, x]||
2
= x < 1
so V must be compact. On the other hand, V is not self-adjoint.
< V f, g >=< f, Wg >
_
1
0
(
_
x
0
f(y)dy)g(y)dx =
_
1
0
f(x)
_
1
x
g(y)dydx
_
1
0
_
1
0
f(y)
[0,x]
dxg(x)dx
=
_
1
0

[0,x]
_
1
0
f(y)dyg(x)dx
which shows that V is not equal to its adjoint.
3. Let {x} = x [x] be the fractional part of x R. {x/2} is 2-periodic. Find its
Fourier series.
1
2
_
2
0
f(x)e
inx
dx =
1
2
_
2
0
(x/2 [x/2])e
inx
On [0, 2], [x/2] = 0, so the integral is
1
2
_
2
0
x
2
e
inx
dx
=
1
in

_
2
0
1
2
ine
inx
=
1
in
4. Dene
(,
1
z
2
) = lim
0
_
|z|>

(z)
1
z
2
dxdy
is C

on C compact support.
30
a. Prove this limit exists.
_
<|z|<K

(z)
1
z
2
dxdy Because this integral has compact support, |

(z)| C for some con-


stant. So this integral is less than
lim
0
_
|z|>
C
z
2
dxdy
????
b. Let
partial
partial z
=

x
+i

y
. Find

z
(
1
z
2
as a distribution.
lim
0
_
|z|

(z)wdxdy

z
1
(x +iy)
2
=
(2x + 2iy)
(x
2
y
2
+ 2ixy)
2
+
2x + 2iy
(x
2
y
2
+ 2ixy)
2
= 0
=
_
R
f

dx
5. Let ||q|| < 1, q R. Consider the Fourier series

nZ
q
n
2
e
2inx
Is the function dened by this series dierentiable and if so, how many terms?
This function is innitely dierentiable. To show this, we prove that

a
n
e
inx
and all its
derivatives converges uniformly. For any p,
|q|
n
2
n
p
=
e
n
2
log |q|
n
p
lim
n
e
n
2
log |q|
n
p
= lim
n
log |q|2ne
n
2
e
n
2
log |q|
pn
p1
lim
n
Ce
n
2
log q

31
9 December 18, 2007
1. Prove there exists a measurable set E [0, 1] such that E is closed, nowhere dense,
has Lebesgue measure 1/2, and for any y [0, 1] there exists x E such that |x y| >
1/10.
Let E be a nite union of thickened Cantor sets of positive measure. For example, let E
1
be the thickened Cantor set such that the rst removed interval has length 2/9, the union
of the second removed intervals has length 2/27, and so on, so (E
1
) = 12/3 = 1/3. Then
let E
2
be the thickened Cantor set of measure 1/6, placed inside the central removed
interval of length 2/9. Let E = E
1
E
2
. By construction, (E) = 1/2. No excised interval
is greater than 1/10, so no point in [0, 1] is more than 1/10 away from a point in E.
2. Let f be a Lebesgue-measurable function on [0, 1]. Dene

f
(t) = {x [0, 1] : f(x) > t}
Prove that
_
1
0
|f(x)|
p
dx = p
_

0
t
p1

f
(t)dt
p
_

0
t
p1

f
(t)dt => p
_

0
t
p1
_
1
0

x:|f(x)|>t
d(x)dt
=
_
1
0
_
|f(x)
0
pt
p1
dtd(x)
=
_
1
0
|f(x)|
p
d(x)
Prove that
_
/8 = lim
R
_
R
0
cost
2
dt
This is done with a contour integral of e
iz
2
over the sector of radius R between 0 and
/4.
_
/4
0
e
(Re
i
)
2
d 0
32
as R . The integral along the real line is
_
R
0
e
ix
2
dx
The real part of this is the desired integral. For the integral along the ray at = /4 write
z(t) = e
i(/4)t 1+i

2
t and dz =
1+i

2
dt.
_
R
0
e
iz
2
dz
=
1 +i

2
_
R
0
e
i(e
i/4t
2
)
2
dt
=
1 +i

2
_
R
0
e
t
2
dt
= 1/2
_
/2 +i/2
_
/2
Taking real parts of both sides and using Cauchys theorem to set them equal, we get
that
lim
R
_
R
0
cost
2
dt =
_

8
5. Let A = {z C : r
1
|z| r
2
} be an annulus. Show that there is a conformal map
between two annuli i the ratio of their boundary circles is equal.
The forward direction is clear; a translation and a dilation takes one annulus to the other
if the ratio of their boundary circles is equal.
Conversely, suppose there is a conformal map between the two annuli. A conformal map
sends a sub-circle in A
1
to a closed curve in A
2
with radius between s
1
and s
2
. By
the Schwartz reection principle, the outer sub-annulus gets sent either to the inner or the
outer sub-annulus. By the Schwartz reection principle, f also continues analytically across
the outer boundary, extending to the plane C and across the inner boundary, extending to
D{0}, so the conformal map extends to be an automorphism of the punctured plane.
In fact, the map must x the origin (consider a sequence of concentric circles with radius
converging to zero; the Schwartz reection principle shows that the region interior to each
circle must be sent to the region interior to the image of that circle, and in the limit this
means the origin must be xed.
Therefore f is a linear function. This implies that r
1
/r
2
= s
1
/s
2
, as desired.
6. Prove that the dual space of L

is not L
1
. By a previous problem The unit ball in L
1
has no extreme points. By the Krein-Milman theorem, B = co(ext(B)) if B is a compact
33
convex subset of a topological vector space. By the Banach-Alaoglu theorem, the unit ball
in X

is weakly compact if X

is the dual of a topological space. The unit ball is always


convex. So if L
1
were the dual of some topological space, the unit ball in L
1
would be
empty, which is a contradiction.
7. Let X, Y, Z be Banach spaces and B : X Y Z a bilinear map.
a. B is continuous i there exists c such that ||B(x, y)|| c||x||||y||.
Suppose the condition holds. Then if x
n
0, B(x
n
, y) c||x
n
||y|| 0. The same is true
if y
n
0.
Conversely, if B is continuous,
B
1
({x, y) : |x| < 1, |y| < 1}
contains an open ball around zero. There exists such that |x| < , |y| < implies
B(x, y) < 1.
1 > |B(
x
||x|| +
,
y
||y|| +
)| =

2
(||x|| +)(||y|| +)
B(x, y)
B(x, y) <
1

2
(||x|| +)(||y|| +)
as 0, we have the desired claim.
b. B is continuous i B is separately continuous. Note that (x
n
, y
n
) 0 i x
n
0 and
y
n
0 in X Y . If B(x
n
, y) 0 when x
n
0 and B(x, y
n
) 0 when y
n
0, then
B(x
n
, y
n
) 0.
NOTE: got confused by the converse.
9. Let be in the Schwartz class.
F
t
() = lim
0
+
_
R
(t)
t +i
dt
denes a tempered distribution.
Because is in the Schwartz class, |(t)|
C
1+|t|

for some > 1, so


(t)
t+i
has bounded
integral for |t| > 0.
The integral is
lim
0
+
_
|t|>
(t)
t +i
+
_

(t)
t +i
dt
34
(t)
t +i
=
(t) (i)
t +i
.
As 0, this goes to
(t) (0)
t
+
(0)
t
.
The latter term is a constant divided by an odd function, so it integrates to zero; the former
approaches

(0) as 0, which shows that the integral over


_

(t)
t+i
is nite, since a
function in the Schwartz class is dierentiable.
10. Consider functions in the plane with the norm
_ _
|f(x +iy)|
2
e

x
2
+y
2
dxdy
Inner product:
< f(x, y), g(x, y) >=
_ _
f(x, y)g(x, y)e

x
2
+y
2
dxdy
The space H of functions with nite H-norm is complete with respect to this norm; If f
n
is a Cauchy sequence, then suppose |f
n
f
m
| < for all n, m greater than N. Then
_
|f
n
(x +iy) f
m
(x +iy)|e

x
2
+y
2
dxdy <
_ _
e

x
2
+y
2
dxdy
since the Gaussian has a nite integral, as 0, ||f
n
f
m
||
H
0. If we let ||f
n
k1

f
n
k
||
H
< 2
k
, then let
g = |f
n
1
(x, y)| +

k=1
|f
n
k+1
(x, y) f
n
k
(x, y)|
Then
||g||
H
||f
n
1
|| +

||f
n
k+1
f
n
k
|| ||f
n
|| +

2
k
This shows that H is complete in its norm.
10 December 8, 2008
1. Prove or disprove the following statement: If {f
n
} is a sequence of continuous functions
on [0, 1] which converge pointwise to a function f then there is a point x
0
[0, 1] such that
f is continuous at x
0
.
35
Proof of statement:
Fix
1
> 0. Then

N

n,mN
{x : |f
n
(x) f
m
(x)|
1
/3} = [0, 1]
That is, for any x in [0, 1] there exists an integer N such that for any n, m N, |f
n
(x)
f
m
(x)|
1
/3.
Let G
n
(
1
) =
n,mN
{x||f
n
(x) f
m
(x)|
1
/3}. Because f
n
, f
m
continuous, the sets
G
N
(
1
) are closed.
If x G
N
(
1
) then |f
n
(x) f
(
x)|
1
/3n N because we may let m .
By the Baire Category theorem, because the union of G
N
(
1
) is a closed interval, there
exists some N
1
such that G
N
has nonempty interior. Then let I
N
1
be a nontrivial closed
interval I
N
1
= [a
N
1
, b
N
1
]. If x, y I
N
1
then |f
N
1
(x) f(x)| /3, |f
N
1
(y) f(y)| /3,
and since N
1
is xed, f
N
1
is uniformly continuous. So there exists
N
1
, perhaps smaller
than b
N
1
a
N
1
, such that x, y [0, 1], |f
N
1
(x) f
N
1
(y)| /3. By choosing J
N
1
I
N
1
of
length
N
1
we have on J
N
1
that |f(x) f(y)| by the triangle inequality.
Now, we take a sequence {
k
} such that
1
is as chosen above, and
k
0. Repeating the
proof above with J
N
1
in place of [0, 1], we have

N

n,mN
{x : |f
n
(x) f
m
(x)|
1
/3} = J
N
1
. Applying the Baire Category Theory again, we get N
2
such that theres an interval
J
N
2
J
N
1
where |f(x) f(y)|
2
for x, y J
N
2
. Inductively, we get
J
N
1
J
N
2
. . .
a sequence of closed nontrivial intervals where |f(x) f(y)|
k
for |x, y J
N
k
. Let x
0
be a point in the intersection of all J
N
k
. Then f is continuous at x
0
indeed, for any
> 0 there exists an interval J
N
k
around x
0
(where
k
< ) such that |f(x) f(y)| < for
x, y J
N
k
.
2. a. Let F
n
L

([0, 1]). ||F


n
||

< 1. Prove or disprove the following statement:


There is a subsequence {F
n
k
} such that for all G L
1
([0, 1])
lim
k
_
1
0
F
n
k
(x)G(x)dx
exists. By the Principle of Uniform Boundedness, its enough to show that ||F
n
k
||

< C,
but this i true by assumption. Because L

is the dual of L
1
, the norm of F
n
k
as an operator
36
on L
1
is the L

norm of F
n
k
. So because {F
n
} is counded in L

, for any functional in


this set sup
a
|F
n
(a)| < which proves our claim.
b. Suppose F
n
is a sequence of functions in L
1
([0, 1]) with norm bounded by 1, ||F
n
||
1
1.
Prove or disprove the following statement:
There is a subsequence {F
n
k
} such that for all G L

([0, 1])
lim
k
_
1
0
F
n
k
(x)G(x)dx
exists.
This is false. Let Let F
n
be a sequence of constant functions with support an interval of
length 2
n
, F
n
= 2
n
where F
n
= 0. Given any subsequence n
k
, choose G =

Fn
2k
. Then
F
n
L
1
, ||F
n
||
1
= 1 for all n, but
_
F
n
G
is 1 for k even, and 0 for k odd. This is a counterexample to the claim.
3. Let P be a polynomial in z with complex coecients. Prove that if all the roots of P
lie in the upper half plane then the roots of P

lie in the upper half-plane.


Proof. Write P(z) = (z a
1
) . . . (z a
n
). Then
P

(z)
P(z)
=
1
z a
1
+. . . +
1
z a
n
Suppose z
0
is a root of P

in the lower half-plane, z


0
= x
0
+iy
0
. Then
1
z
0
a
i
=
1
(x
0
x
i
) +i(y
0
y
i
)
=
(x
0
x
i
) +i(y
i
y
0
)
|z
0
a
k
|
2
This is a complex number with positive imaginary part, so P

(z
0
)/P(z
0
) is a complex
number with positive imaginary part. On the other hand, by construction it must be zero,
which is a contradiction. So all the roots of P

(z) must lie in the upper half-plane.


4. State and sketch the proof of the Riemann Mapping Theorem.
Statement: Any simply connected bounded region in the complex plane is conformally
equivalent to the disc via a map F such that F(z
0
) = 0 and F

(z
0
) > 0.
37
Step 1 of proof: is conformally equivalent to a subset of the disc that contains the
origin.
Indeed, if / , let f(z) = log(z ). Then f is injective, and f(z) = f(w) + 2i if w is
a point in such that z = w, so the function
F(z) =
1
f(z) f(w) 2i
is bounded. By translating and rescaling, we have a holomorphic function from into the
disc that includes the origin.
Step 2:
Let F be the family of holomorphic functions f : D which are injective and have
f(0) = 0. This family is not empty, because it contains the identity. Its uniformly
bounded by construction. also, |f

(0)| is uniformly bounded by the Cauchy inequality,


because |f| is uniformly bounded for all f. Let s = sup |f

(0)|. Choose a sequence f


n
such
that |f

n
(0)| s. By Montels Theorem, the family F is normal, which means f
n
has a
subsequence converging uniformly on compact sets to some holomorphic function f.
Lemma: if f
n
is a sequence of injective holomorphic functions on a connected open set,
converging uniformly on every compact subset, then f is either injective or constant.
Since s > 0, f is injective. Since |f(z)| < 1, we conclude f F.
Step 3:
We claim f is surjective. (Since we already know f is injective into the disc and f(0) = 0,
this will complete the proof.)
Suppose f were not surjective. Then we could construct a function in F with derivative
greater than s at 0. Indeed, suppose D, f(z) = for any z. Then let

(z) =
z
1 z
So 0 / (

f)(). Thus g(w) = e


1/2 log w
is dened on

f)().
F =
g()
g

f
is in F because its injective, has positive derivative, and is 0 at 0. We claim |F

(0)| >
|f

(0)|. Indeed,
f =
1

h
g()
F := F
where h(w) = w
2
f

(0) =

(0)F

(0)
38
is not injective, so |

(0)| < 1, because rotations are the only holomorphic functions on


the disc with derivative 1 at zero. So |F

(0)| > |f

(0)|. By the construction of f, this is a


contradiction. So f cannot avoid a point in the disc, so f must be surjective.
5. Let C [0, 1/2] be a closed subset of Lebesgue measure zero. Suppose that f(z) is a
bounded holomorphic function on D C where D = {z C : |z < 1}. Prove that f can be
extended to a holomorphic function on D.
Since C has measure zero, the set [0, 1/2] D/C is dense in [0, 1/2], otherwise C would
contain an interval. So for every z C, and every radius r > 0, there is a circle D/C
centered at z with radius r. Claim that f

(z) 0 as the radius of goes to 0. Let


k
be
a sequence of circles centered at z with radius r
k
0.
f

(z) =
1
2i
_

f()
z
d
f

k
(z) =
1
2
_
2
0
f(z +r
k
e
i
)
r
k
d
Because f is holomorphic,
|f(z +r
k
e
i
) f(z +r
j
e
i
)|
|r
k
e
i
r
j
e
i
|
0
as j, k for any xed = 0. So f
r
k
(z) is a Cauchy sequence and thus converges.
Dene
f(z) = lim
r
k
0
1
2i
_
r
k
f()
z
d
By the uniqueness of analytic continuation and Cauchys theorem, this is the only possible
analytic continuation on D.
6. Prove that

n=1
1
n
2
=
2
/6
Proof: The sum formula for the cotangent is

m=
1
(m+)
= cotx
39
Dierentiating, this yields

m=
1
(m+)
2
=

2
sin
2
()
Plugging in = 1/2,

m=
1
(m+ 1/2)
2
=
2
/1
We observe that (m+ 1/2)
2
= ((m+ 1) + 1/2)
2
, so this sum can be rewritten as
2 (2
2
+ (2/3)
2
+ (2/5)
2
+. . .) =

m1,odd
8
m
2
So

m=0
1
2m+ 1)
2
+ 1/4

m=1
1
m
2
=

m=1
1
m
2

2
/8 + 1/4

m=1
1
m
2
So

m=1
1
m
2
=
2
/6
7. Find with proof the number of zeros of the function 9z
6
+e
2z
on the closed unit disc in
the complex plane.
max
zD
|9z
6
| = 9, and max
zD
|e
2z
| = e
2
< 9, so we can apply Rouches theorem to see
that 9z
6
+e
2z
has the same number of zeros in the unit disc as 9z
6
, so the function has 6
zeros in the unit disc.
11 December 16, 2010
1.a. Let (X, A, ) be a nite measure space and f a real-valued function on X. Suppose
f L
p
(X, ) for all 1 p < . Show that lim
p
||f||
p
= ||f||

.
40
||f||
p
p
=
_
|f|
p
= ||f
p
||
1
||1||
1
||f
p
||

= (X)||f
p
||

= (X)||f||
p

||f||
p
(X)
1/p
||f||

||f||

.
Conversely, consider an increasing sequence a
n
such that lima
n
= ||f||

. This means that


on a set E
n
of positive measure, a
n
< |f|
(
_
En
|f|
p
)
1/p
(E
n
)
1/p
a
n
taking the limit as n
(
_
X
|f|
p
)
1/p
(X)
1/p
||f||

as desired.
b. Let f : R R and g : R R, f L
p
, g L
q
where 1 < p, q < such that
1/p + 1/q = 1. Show
lim
t
_
f(x +t)g(x)d(x) = 0
Since f L
p
, f(x) 0 as x . For any , there exists N such that |f(x)| < for
|x| > N. Likewise, for any

there exists N such that g(x) <

for |x| > N. For t


suciently large, whenever f(x + t) > , g(x) >

and whenever g(x) >

, f(x + t) < .
This means there are two sets, A and B, where one of the functions is greater than or

:
_
A
f(x +t)g(x)dx
_
A
f(x +t)

dx and
_
B
f(x +t)g(x)dx
_
b
g(x)dx.
5. Estimate the following integrals: i.
_

0
ln sin(x)dx
6. Find a conformal map that maps the open region {z C : |z + i| < 1} {z C :
|z 1| < 1} o the unit disc.
Map 1: translation. z z i. Result: the intersection of the two circles |z| < 1 and
|z i 1| < 1 Map 2: inverse Cayley transform i
z1
z+1
. Takes the boundary of the unit circle
centered at the origin to the real line; takes 1 to 0 and i to 1. The boundary of the other
circle becomes a line that intersects the real line. The intersection becomes the wedge
41
area between them. Map 3: z
/
where is the angle between the two lines; this takes
the wedge to a half-plane. Because conformal maps are angle-preserving, this map is z
3
.
Map 4: rotation, takes the half-plane to the upper half-plane. Map 5: Cayley transform
z iz +i takes the upper half-plane to the disc.
7. Let f : C C be a holomorphic function. Show that if ln |f| L
1
(C), then f(z) = e
i
for some R.
_
| ln |f(z)|dz <
If we write |f(z)| = e
g(z)
then this condition is equivalent to
_
|g(z)|dz < . We use
Jensens formula.
log |f(0)| =

log
1
a
k
+
1
2
_
2
0
log |f(e
i
)|d
where a
k
are the zeros of f in the unit disc around 0. We will show that log |f(0)| 0. We
know from Jensens that log |f(0)|
1
2
_
2
0
log |f(re
i
)|d. The right hand side is dened
because log f is integrable. We can conclude that log |f(0)| 0 so |f(0)| 1. Since 0
was arbitrarily chosen, in general |f(z)| 1. By Liouvilles theorem, an entire bounded
function is constant, so |f(0)| = C. In order for log |f(z)| to be integrable, we must have
|f(0)| = 1 otherwise were integrating a positive constant over the complex numbers, which
diverges. So f must indeed be a rotation of the form f(z) = e
i
.
8. Let H be a Hilbert space and T : H H a linear operator dened on all of H. If T is
self-adjoint, prove that T is continuous. The closed graph theorem says that if (x, Tx) is
closed, then T is continuous. T

is a closed operator: if x
n
x, then ||T

x
n
T

x|| =<
T

(x
n
x), T

(x
n
x) >=< (x
n
x), TT

(x
n
x) > 0. So, since T is self-adjoint, T is
also a closed operator, so (x, Tx) is closed, so T is continuous.
9. Let T = R/2Z. For each k L
(
T), dene an operator W
k
by
W
k
f(x) =
_
T
k(x y)f(y)dy
a. Show that W
k
: L
2
L
2
is a compact operator for each k L
1
. When is it self-
adjoint?
42
When k(x y) is real,
< W
k
f, g >=
_ _
k(x y)f(y)dyg(x)dx
< f, W
k
g >=
_
f(x)
_
k(x y)g(y)dydx
After applying Fubinis Theorem, the two integrals are equal if and only if k(x y) =
k(x y).
Observe that f(y) =

f(n)e
inx
. So if ||f(y)||
2
1,

f(n)| 1 by Parsevals Theo-


rem.
W
k
f(x) =
_
T
k(x y)

f(n)e
iny
dy
We claim that if there is a sequence f
m
of functions in the L
2
unit ball, then W
k
f
m
contains
a convergent subsequence. In the sequence of Fourier sums

f
m
(n)e
iny
there must be a
subsequence such that for any N there exists an M, for which

N
n=1
|f
m
(n) f
m
(n)| = 0
for m, m

> M. Thus ||f


m
f
m
||
2
2
=

n=N
|f
m
(n) f
m
(n)|
2
. and, more explicitly,
f
m
f
m
=

n=N
(f
m
(n) f
m
(n))e
iny
.
Now,
W
k
(f
m
f
m
) =
_
T
k(x y)

n=N
(f
m
(n) f

m
(n))e
iny
dy
_
T
|
_
T
k(x y)

n=N
(f
m
(n) f

m
(n))e
iny
dy|
2
dx
The inner integral can be written as the sum

n=N
_
T
k(x y)(f
m
(n) f

m
(n))e
iny
dy
Because k is integrable, and because (f
m
(n)) are square-summable by Parsevals Theorem,
this sum goes to 0 as N , and so does the entire double integral.
b. Fix k L
(
T) determine the spectrum for W
k
. is in the spectrum of W
k
if W
k
I
does not have a bounded inverse. Eigenvalues: If is an eigenvalue for f, then
f(x) =
_
T
k(x y)f(y)dy = k f
For each Fourier coecient

f(n), we must then have

f(n) =

k(n)

f(n)
43
which means that the Fourier coecients must be constant, which means that must be
0. So k has no nonzero eigenvalues.
10. Let L
2
(D) be the Hilbert space of holomorphic functions f on the open unit disk D C
such that
_
D
|f(x +iy)|
2
dxdy < . Fix z
0
in D. Let L be a bounded linear functional on
L
2
(D such that L(f) = f(z
0
) for all f L
2
(D). Prove or disprove that
L(f) =
1

_
D
f(x +iy)dxdy
(1 z
0
(x iy))
2
First we note that z
n
form an orthogonal basis of L
2
(D).
_
D
z
n
z
m
dz =
_
2
0
_
1
0
r
n
r
m
e
i(nm)
rdrd
= 0
Let
n
denote cz
n
, scaled so that all
n
have norm 1. Here, c =
_
n+1

.
So, suppose there exists a kernel K(z
0
, z) such that
f(z
0
) =
_
D
f(z)K(z
0
, z)dz
f(z
0
) =
_
D
f(z)(

n
(z
0
)
n
(z))dz
because f has an expansion in
n
itself, f(z) =

c
m

m
(z) and the only nonzero terms
in the integrand are

c
m

m
(z
0
) <
m
,
n
>= f(z
0
). Now, we compute this kernel
explicitly.

n
(z
0
)
n
(z)
=

n
_
n + 1

z
n
0
_
n + 1

z
n
=
1

n
(n + 1)(z
0
z)
n
Summing the geometric series, this gives us
1

1
(1 z
0
z)
2
44
12 May 4, 2009
1. a. State the Dominated Convergence Theorem.
If f
n
is a sequence of nonnegative measurable functions, and f
n
g for all n, where g is a
nonnegative measurable function, and if f
n
(x) f(x) pointwise for all x., then
_
f
n
d
_
fd.
b. Prove the Dominated Convergence Theorem. Let h
n
= g + f
n
. This is a sequence of
non-negative measurable functions so that g(x) +f(x) = lim
n
g(x) +f
n
(x) for all x. So by
Fatous Lemma _
g +fd liminf
_
g +f
n
d
so _
fd liminf
_
f
n
d.
Similarly, if j
n
= g f
n
, we have
_
g fd liminf
_
g f
n
d
so
limsup
_
f
n

_
fd
This proves
_
fd = lim
_
f
n
d.
2. Let H be a Hilbert space and let I be the identity operator on H. Let T : H H be a
linear operator with operator norm 1/2. Prove of disprove the following statement: The
operator I +T must have an inverse that is bounded on H.
First, we show that if I + T has an inverse, then it is bounded. Suppose G is a linear
operator and
G(I +T)(x) = x.
G(x) +GT(x) = x
||x|| = ||G(x) +GT(x)||
||x|| ||G(x)|| ||GT(x)||
45
||x|| ||G(x)|| 1/2||G(x)|| = 1/2||G(x)||
||G(x)|| 2||x||
so G(x) must be bounded.
Now we show that I +T has an inverse. We claim If (I +T)x = 0 then x = 0. But
||(I +T)x|| = ||x +Tx|| ||x|| ||Tx|| 1/2||x||
so if this equals 0 then ||x|| = 0.
3. Let D be the open unit disc in the complex plane. For a Lebesgue measurable function
F on D dene
TF(w) =
_
D
F(z) log(|z w|)dxdy
a. Prove T is a bounded operator from L
2
(D) to itself. T is self-adjoint:
_
D
(
_
D
F(z) log(|z w|)dz)G(w)dw =
_
D
F(z)
_
D
log(|z w|)G(w)dwdz
A self-adjoint everywhere dened operator on a Hilbert space is bounded. So if TF(w) is
dened for all square-integrable F, T is bounded. Since
_
1
0
(log x)
2
dx = (x(logx)
2
2xlogx + 2x)|
1
0
= 2 lim
x0
xlog
2
(x) 2xlog x
= 2
b. Prove or disprove the following statement: T is a compact operator from L
2
to it-
self.
T is an integral operator of the form
_
f(x)K(x, y)dx, where K is square-integrable. There-
fore, K is Hilbert-Schmidt, since
_
|f(x)K(x, y)|
2
dx for all square-integrable f implies
that

K(x, y)e
n
< where e
n
are an orthonormal basis.
4. Construct a Lebesgue measurable set A [0, 1] with the following property; if J is a
nonempty open interval in [0, 1] then both m(A J) > 0 and m(A
c
J) > 0.
Let A
1
be a thickened Cantor set constructed to have Lebesgue measure 1/4. Let A
2
be
the union of A
1
with thickened Cantor sets placed in every excised interval of length at
46
least 1/3, so that each thickened Cantor set has the same measure and the total measure
of A
2
is 3/8. Let A
3
be the union of A
2
with thickened Cantor sets placed in every excised
interval of at least 1/9, so that the total measure of A
3
is 7/16. Continue in the same
fashion. Let A = A
n
. Then m(A) = 1/2, and A has a thickened Cantor set in every open
interval of [0, 1]. By construction, every open interval contains a Cantor set of positive
measure, so m(A J) > 0. But each Cantor set has measure less than 1/2 the measure of
the interval its contained in, so m(A
c
J) > 0.
5. Let S be the innite half-strip in the complex plane dened by S = {z|0 < x < , 0 <
y < 1}. Show how to construct a conformal mapping from S to the unit disc.
Map 1: e
i
z, rotation by so that the half-strip lies along the negative real axis. Map 2:
e
z
takes the half-strip into the half-disc; since y is between 0 and 1, the angle is between 0
and , and x < 0 so |e
z
| < 1. Map 3: (
z+1
z1
)
2
takes the semicircle to the upper half-plane.
Map 4:
zi
z+i
the Cayley transform takes the upper half-plane to the unit disc.
6. Use the method of contour integrals to evaluate
F(x) =
_

0
x
1/2
(1 +x
2
)
1
dx
Here use the branch of the logarithm that avoids the negative y-axis. log z = log |z| +i( +
/2). Theres a pole at i with residue
e
1/2 log i
i+i
=
e
i/2
2i
=
1
2
. We consider the semicircle
in the upper-half plane with radius R and center at the origin. The integral around the
outside is
_

0
(Re
i
)
1/2
/(1 + (Re
i
)
2
)
which goes to 0 as R goes to innity. That leaves
1/2 =
_
R
0
x
1/2
(1 +x
2
)
1
dx +
_
0
R
x
1/2
(1 +x
2
)
1
dx
Now e
1/2 log x
= e
1/2(log |x|+i/2)
= |x|
1/2
(1/

2+i/

2) for x > 0 and e


1/2 log x
= e
1/2(log |x|+i3/2)
=
|x|
1/2
(1/

2 +i/

2) for x < 0. So if we let I denote the integral in question,


I +
1/

2 +i/

2
1/

2 +i/

2
I = 1/2
I = 1/2
1/

2 +i/sqrt2
2i/sqrt2
47
7. Let be a nite, complex Borel measure on the unit circle and suppose that for integers
n,
_
2
0
e
in
d() = 0.
Prove that is the zero measure.
We will show that for any measurable function f,
_
2
0
fd = 0.
Any measurable function can be approximated by a continuous function except on a set
of small measure; there exists h continuous such that |f(x) h(x)| < except on a set of
measure less than .
_
2
0
h
n
d
_
2
0
fd
if we let h
n
be a continuous function less than
n
away from f except on a set of measure

n
, where
n
0. Indeed,
_
En
|f(x) h(x)|d ,where E
n
is the bad set, is at most
(E
n
) max
En
|f(x)| + |h(x)| max
S
1 |f(x)| + |h(x)| 0 as 0. Now, continuous
functions on the unit circle can be approximated by trigonometric polynomials of the form
S
N
=

N
n=1
a
n
e
in
. Since for each term
_
2
0
a
n
e
in
d = 0
, lim
N
_
S
N
d = 0, we must have that
_
h
n
d = 0, which thus implies that
_
fd = 0.
Thus, must be the zero measure.
8. Prove or disprove the following statement: if F(z) is a bounded holomorphic function
on the open unit disk D and there is a sequence of points z
n
in D such that z
n
1 for all
n, F(z
n
) = 0, then F(z) = 0 for all z D.
This is false. Consider F(z) = sin(
1
1z
). This is an analytic function on the open unit
disc. F(z) = 0 when
1
1z
= n or z = 1
1
n
. This is a sequence of zeros converging to 1.
However F(z) is not identically zero. This is a contradiction.
9. Let H be the normed linear space consisting of all holomorphic functions F on the open
unit disk D such that (
_
D
|F(z)|
2
dz)
1/2
= ||F|| < . Construct with proof an orthonormal
basis for H.
48
The monomials z
n
are orthogonal in D:
_
D
z
n
z
m
dz =
_
D
re
in
re
in
ddr
=
_
D
r
2
e
i(mn)
ddr = 0
for m = m. Now, the norm ||z
n
|| is
(
_
1
0
_
2
0
|re
in
|
2
ddr)
1/2
=

n + 1
so the orthonormal basis is
1

n+1
z
n
.
10. Prove or disprove the following statement; there is a Lebesgue measurable function
F(x, y) on the unit square such that for all y [0, 1) F(x, y) is square-integrable and
_
1
0
F(x, y)dx = 0
but for all x [0, 1) F(x , y) is square-integrable and
_
1
0
F(x, y)dy = 1.
Divide the unit square into rectangles as follows: the x-interval [0, 1] is divided into intervals
of the form [1/2
n
, 1/2
n1
] from n = 1 to . Likewise the y-interval [0, 1] is divided into
intervals of the form [1/2
n
, 1/2
n1
] from n = 1 to . The direct product of these intervals
forms a partition of [0, 1] [0, 1] into rectangles. For each point (x, y) let A
m,n
be the
rectangle that contains that point, of the form [1/2
m
, 1/2
m1
] [1/2
n
, 1/2
n1
]. Now let
F(x, y) = 0 for all x, y not in A
n,n
or A
n,n1
. Let F(A
1,1
) = 2, and F(A
2,1
) = 4. Then for
all x in [1/2, 1],
_
1
0
F(x, y)dy = 21/2 and for all y [1/2, 1],
_
1
0
F(x, y)dx = 1/44+1/2
2 = 0. Now let F(A
2,2
) = 12. Then for x [1/4, 1/2],
_
1
0
F(x, y)dy = 1/24+1/412 = 1.
Now let F(A
3,2
) = 24. Then for y [1/4, 1/2],
_
1
0
F(x, y)dy = 1/8 24 + 1/4 12 =
0. Solving linear equations in this fashion for each rectangle, we obtain in the limit a
measurable function satisfying the above conditions for all x and y in [0, 1] [0, 1].
49
13 Problems from Functional Analysis
Let X be a Banach space, A and B closed linear subspaces. Let inf{||x y|| : x A, y
B, ||x|| = ||y|| = 1} > 0.
a. Show that A+B is closed in B.
In other words, if a
n
A, b
n
B, a
n
+ b
n
x we claim that x = a + b where a A, b
B.
Let a = inf{||x y|| : x A, y B, ||x|| = ||y|| = 1}.
First we claim that if ||x|| > c, ||y|| > c, then inf{||x y||} > 0.
Now we prove the claim. Let D
c
= inf{||x y|| : x A, y B, ||x|| = ||y|| = c}. This is
greater than 0, since ||cx cy|| = c||x y||. If ||x|| > c, ||y|| > c then
inf{||x y||} inf{||x

||}
where x, y are scaled so that ||x

|| = ||y

|| = c, so ||x y|| > D


c
, which proves the
claim.
WLOG, assume ||x|| ||y||. Then
2||y x|| a||x||
2||y x|| = 2||x|| ||(y/||x||) x/||x||
||x y|| ||x||a ||x||(||y/||x|| y/||y||||)
||x y|| ||x||a (||x|| ||y||)
2||x y|| a(max ||x||, ||y||).
We claim that either a
n
or b
n
is Cauchy. For suppose ||a
n
a
m
|| > , ||b
n
b
m
|| > for all
n, m. Then
||(a
n
a
m
) + (b
n
b
m
)|| >
||a
n
+b
n
(a
m
+b
m
)|| >
which contradicts the fact that a
n
+ b
n
must be Cauchy (because it converges.) Now, say
a
n
is Cauchy. Then it converges to some a A, because A is closed. So a
n
+ b
n
a
converges to limb
n
B, so a
n
+b
n
a +b.
b. Assume A+B = V , A B = {0}. Show that inf{||x y|| : x A, y B, ||x|| = ||y|| =
1} > 0.
Because A and B are disjoint, consider the projection P
A
of V onto A. Because A and B are
disjoint, we can write V = AB and let the projection take P
A
(a, b) = a. This is surjective
50
and continuous, so by the open mapping theorem P
A
is open. Consider P
A
(x
n
y
n
) = x
n
.
lim||x
n
|| = 1 because all ||x
n
|| = 1. Because the graph of P
A
is closed, lim||x
n
|| = 0, which
is a contradiction. So inf{||x y||} > 0.
3. Let 1, z
2
. . . be the monomials on the unit disc. Let e
n
= ||z
n
|
1
z
n
. Show that they
form a basis for L
2
on the disc.
Clearly the e
n
are orthogonal, because
_
1
0
_
2
0
re
in
re
in

rdrd =
_
0r
3
dr = 0.
By the Stone-Weierstrass theorem, the polynomials are dense in the space of continuous
functions. The continuous functions are dense in L
2
. Indeed, suppose not. Then there
exists an L
2
function h so that
_
fhd = 0 for every continuous f. Since simple functions
can be approximated by continuous functions, we have
_
K
fd = 0 for all Borel sets K,
which means that f must be zero.
51

You might also like