ANALYSIS QUALIFYING EXAM SOLUTIONS

RAYMOND CHU

These solutions should contain a majority of the questions asked from Spring 2010 to Fall 2020. Some
of these solutions are inspired by the solutions from Adam Lott’s solutions which can be found here and
for Steven Truong’s solutions which can be found here.

I am very thankful for James Chapman, Steven Truong, Jianming Wang, James Leng, Erin George, and
Kyle Gettig for many useful discussions on the analysis qualification exam.

Contents
1. Spring 2010 1
2. Fall 2010 12
3. Spring 2011 18
4. Fall 2011 26
5. Spring 2012 32
6. Fall 2012 39
7. Spring 2013 46
8. Fall 2013 53
9. Spring 2014 61
10. Fall 2014 68
11. Spring 2015 75
12. Fall 2015 82
13. Spring 2016 92
14. Fall 2016 101
15. Spring 2017 109
16. Fall 2017 117
17. Spring 2018 125
18. Fall 2018 134
19. Spring 2019 142
20. Fall 2019 151
21. Spring 2020 157
22. Fall 2020 167

1. Spring 2010

Problem 1. Let 1 ≤ p < ∞. Show that if a sequence of real-valued functions {fn }n≥1 converges
in Lp (Rn ), then it contains a subsequence that converges almost everywhere.

Also give an example of a sequence of functions converging to zero in L2 (R) that does not converge
almost everywhere.

Proof. Assume fn → f in Lp (R). Then there exists a subsequence {fnk } such that
1
||fnk − f ||pLp (R) ≤ −k
2
Date: March 25, 2021.
1
2 RAYMOND CHU

Therefore,
∞ ˆ
X
|fnk − f |p dx ≤ 1
k=1 R

which by the monotone convergence theorem tells us that

X∞ ˆ ˆ X
∞
|fnk − fnk+1 |p dx = |fnk − f |p dx ≤ 1
k=1 R R k=1
P∞
Therefore, it follows that for a.e. x that k=1 |fnk (x) − f (x)|p < ∞ so fnk (x) − f (x) → 0 for a.e. x

Problem 2. Let {pi }ni=1 be distinct points in C and let U := C \ {p1 , .., pn }. Let A be the vector
space of real harmonic functions on U and let B ⊂ A be the subspace of real parts of complex
analytic functions on U . Find the dimension of the quotient vector space A/B, giver a basis of
this quotient space, and prove that it is a basis.

Proof. Fix ψ(z) ∈ A then define

g(z) := ∂x ψ(z) − i∂y ψ(z)
then observe that this function satisfies the Cauchy-Riemann Equations since ψ is harmonic. And as g(z)
is real differentiable, we see that is must be holomorphic. Now notice that we have isolated singularities
at {pi }ni=1 , so for each of these isolated singularities define cj as the residue of g at pi . Then notice
n
X cj
h(z) := g(z) −
j=1
z − pj

is a holomorphic function that integrates to zero over any closed curve γ ⊂ U thanks to the Residue
Theorem. Therefore, h(z) has a primitive which we denote by u(z). Now observe if we define
n
X
ũ(z) := ψ(z) − cj log |z − pj | ⇒ ∂x ũ(z) − i∂y ũ(z) = h(z)
j=1

And the Cauchy Riemann Equations tells us that

∂u
u0 (z) = = ∂x Re(u(z)) − i∂y Re(u(z)) = h(z)
∂x
i.e. ũ is the real part of u(z) up to some positive constant. So it follows that ũ is the real part of a
holomorphic function. Therefore, we have shown that the set of functions

{log |z − pj |}nj=1

spans A/B. And they are also linearly independent. Indeed, if

n
X
cj log |z − pj | = 0
j=1

So taking exponentials give

n
Y
|z − pj |cj = 1
j=1

Say cj 6= 0 then taking z = pj gives 0 = 1, so we must have cj = 0 for all j. Therefore, {log |z − pj |}nj=1
is a basis of A \ B, so this vector space has dimension n with the above basis.

3

Problem 3. For an f : R → R, f ∈ L1 (R) define the Hardy-Littlewood maximal functions as

ˆ x+h
1
(M f )(x) := sup |f (y)|dy
h>0 2h x−h
Prove that it has the following property: There is a constant A such that for any λ > 0,
A
m({x ∈ R : (M f )(x) > λ}) ≤ ||f ||L1
λ

Proof. Let m(A) denote the Lebesgue Measure of the measurable subset A. We first need the following
covering lemma:

Lemma: Given a finite collection of balls {Bi }N m

i=1 then there exists a collection of balls {Bij }j=1 that
are disjoint such that
[N Xm
m( Bi ) ≤ 3 m(Bij )
i=1 j=1
Proof of Lemma: Notice if B and B 0 are balls such that they intersect with the radius of B greater
than or equal to B 0 then B 0 is contained in the ball concentric with B with 3 times its radius. Now we
proceed with a greedy algorithim. Choose Bi1 such that m(Bi1 ) ≥ m(Bj ) for any j then consider all the
other balls that intersect Bi1 . Then those balls are contained in a concentric ball of Bi1 with 3 times the
radius denoted by B˜i1 . Now iterate this process of choosing the maximal remaining balls (maximal in
SN
the sense of volume) we eventialy stop to obtian a collection {B˜i1 } such that they cover i=1 Bi and are
disjoint such that
[N [m Xm m
X
m( Bi ) ≤ m( B˜ij ) = m(B˜ij ) = 3 m(Bij )
i=1 j=1 j=1 j=1
and the covering lemma is proven.

Now fix a compact subset λ > 0 and fix a comapct subset K ⊂ {x : (M f )(x) > λ}. Now by the
translation continuity of the lebegsue integral we know for any x ∈ K there exists an εx > 0 such that
Bεx (x) ⊂ {x : (M f )(x) > λ}. Therefore, by compactness we can find a finite subcover say Bεi for
i = 1, .., N . This implies that by our covering lemma we can find a collection B̃εij such that they cover
the original balls and the measure inequality holds. Therefore,
N n ˆ
X X 3 3
m(K) ≤ m(Bεi ) ≤ 3 m(B˜εij ) ≤ |f (x)| ≤ ||f ||L1
i=1 j=1
λ S ˜
Bεi λ
j

where the last inequality is due to the balls are disjoint. And the second last inequality is due to
ˆ
|f (x)| ≥ αm(B̃εij )
B̃εi
j

and all the sets are disjoint. Now we use the inner regularity of the Lebesgue measure as its a Radon
Measure to conclude.

Problem 4. Let f (z) be a continuous function ont he closed unit disk D such that f (z) is analytic
on the open disk D and f (0) 6= 0.
(1) Prove that if 0 < r < 1 and if inf |z|=r |f (z)| > 0, then
ˆ 2π
1
log |f (reiθ )|dθ ≥ log |f (0)|
2π 0
(2) Use (a) to prove that m({θ ∈ [0, 2π] : f (eiθ = 0}) = 0
4 RAYMOND CHU

Proof. Let f (z) : Ω → C be holomorphic then we claim log(|f (z)|) is subharmonic. That is it satisfies
the mean value inequality: if z ∈ C and Bε (z) ⊂ Ω then
ˆ 2π
1
log |f (z)| ≤ log |f (εeiθ + z)|dθ
2π θ=0
Indeed, if f (p) 6= 0 then on a small ball around p it’s non-zero so locally we can write log |f (z)| as the
real part of a holomorphic function, so log |f (z)| is harmonic and the inequality is an equality. Now
if f (p) = 0 then log |f (p)| = −∞ and the inequality is obvious. So this implies log |f (z)| is locally
sub-harmonic which is equivalent to being sub-harmonic. Now apply this claim to z = 0 to get a). An
alternative approach would be to derive Jensen’s formula.

Proof. Define the cut off function

gn (z) := max(log |f (z)|, −n)
and notice that is is continuous and gn (z) → log |f (z)| as n → ∞. Now observe by continuity and Fatou’s
lemma that
ˆ 2π ˆ 2π ˆ 2π
1 1 1
|gn (eiθ )|dθ = lim inf |gn (reiθ )| ≤ lim inf |gn (reiθ )|
2π 0 2π 0 r→1 r→1 2π 0

Also observe that gn (z) is still sub-harmonic since it is the maximum of two sub-harmonic functions, so
we have ˆ 2π
1
gn (reiθ )dθ ≥ gn (0)
2π 0
and as f ∈ C(D) we have that f is bounded. So we have that if f + , f − denotes the positive and negative
part of f then
ˆ 2π ˆ 2π ˆ 2π
1 + iθ − iθ 1 − iθ 1
gn (re ) − gn (re )dθ ≥ gn (0) ⇒ gn (re )dθ ≤ −gn (0) + gn+ (reiθ )dθ
2π 0 2π 0 2π 0
Notice that f bounded implies log |f (z)| is bounded above, so we have
ˆ 2π
1
gn− (reiθ ) ≤ −gn (0) + C = −f (0) + C
2π 0
for n large since f (0) 6= 0. So it follows that for n large that gn (reiθ ) ∈ L1 ([0, 2π]) with a L1 bound
independent of n or r when n is large i.e.
ˆ 2π
1
|gn (eiθ )|dθ ≤ C
2π 0
so now again using Fatou’s Lemam gives
ˆ 2π ˆ 2π ˆ 2π
1 iθ 1 iθ 1
| log |f (e )||dθ = lim inf |gn (e )|dθ ≤ lim inf |gn (eiθ )|dθ ≤ C
2π 0 2π 0 n→∞ n→∞ 2π 0

so log(f (eiθ )) ∈ L1 ([0, 2π) so {θ ∈ [0, 2π] : f (eiθ ) = 0} has zero measure.

Problem 5. For f ∈ L2 (R) and a sequence {xn } ⊂ R which converges to zero, define
fn (x) := f (x + xn )
show that {fn } converges to f int he L2 sense.

Let W ⊂ R be a Lebesgue measurable set of positive Lebesgue measure. Show that the set of
differences
W − W = {x − y : x, y ∈ W }
contains an open neighborhood of the origin.
5

Proof. Note that compactly supported continuous functions are dense in L2 (R). Therefore, for any ε > 0
there exists a g ∈ Cc (R) such that ||f − g||L2 ≤ ε. Say supp(g) = K then by uniform continuity we have
that for n sufficiently large we have supx∈K |g(x) − g(x + xn )| ≤ ε. Therefore,
ˆ ˆ
|f (x) − f (x + xn )|2 dx ≤ 4 |f (x) − g(x)|2 + |g(x) − f (x + xn )|2
R R
ˆ
2 2 2
≤ 16 |f (x) − g(x)| + |g(x) − g(x + xn )| + |g(x + xn ) − f (x + xn )|
R
√
≤ 16 ε + 2 εm(K) + ε → 0
For the second part, let WR := W ∩ BR (0) then notice
χWR ∈ L2
and for R sufficiently large we have
ˆ
0 < m(WR ) = χ2WR dx
R
now assume for the sake of contradiction that W − W does not have a neighborhood of zero then there
exists a sequence xn → 0 such that xn ∈/ W − W . Therefore,
ˆ ˆ
χW (x)χW (x + xn )dx = χW (x + xn )dx = 0
R W
since if w ∈ W and w + xn ∈ W then xn = (w + xn ) − (w) which would imply xn ∈ W − W so the
integrand is zero. But we also have from from arguing like in 5a) that
ˆ ˆ
0= χW (x + xn )dx → χWR > 0
W W
which is our contradiction.

Problem 6. Let µ be a finite, positive, regular Borel measure supported on a compact subset of
C and define the Newtonian potential of µ to be
ˆ
1
Uµ (z) := dµ(w)
C z − w
(1) Prove that Uµ exists lebesgue a.e. and that
ˆ ˆ
Uµ (z)dxdy < ∞
K
for every compact K ⊂ C. ´
(2) Prove that for almost every horizontal or vertical line L ⊂ C, µ(L) = 0 and K Uµ (z)ds <
∞ for every compact subset K ⊂ L where ds denotes Lebesgue linear measure on L.
(3) Define the Cauchy Potential of µ to be
ˆ
1
Sµ (z) := dµ(w)
C z−w
which trivially exists whenever Uµ (z) < ∞. Let R be a rectangle in C whose four sides
are contained in lines L having the conditions of (b). Prove that
ˆ
1
Sµ (z)dz = µ(R)
2πi ∂R

Proof. Define ˆ
1
Uµ (z) := dµ(w)
C |z − w|
then if K ⊂ C is compact then
ˆ ˆ ˆ
1
Uµ (z)dxdy = dxdydµ(w) thanks to Tonelli
K C K |z − w|
6 RAYMOND CHU

ˆ ˆ ˆ ˆ
1 1
= dxdydµ(w) + dxdydµ(w) = (I) + (II)
C K∩B1 (w) |z − w| C K∩|z−w|≥1 |z − w|
Then observe ˆ
1
dxdy ≤ C
B1 (w) |z − w|
because 1/|x| is integrable om B1 (0). So
ˆ
(I) ≤ Cdµ(w) ≤ K
C
since µ is of finite measure and
ˆ ˆ
(II) ≤ 1dxdydµ(w) = m(K)µ(C) = K2 < ∞
C K

Therefore, for any compact set K we have Uµ ∈ L1 (K, dxdy). Now this implies Uµ is finite a.e. i.e. exists
a.e.in K and let K = Br (0) and send r → ∞ to deduce Uµ is finite a.e. on C.

For the second part, assume for the sake of contradiction that there was the set of lines such that
µ(L) > 0 had positive (Lebesgue) measure. Then there must exist an n ∈ N such that there are infinitely
many disjoint lines {Lm }m∈N such that µ(Lm ) ≥ 1/n. Notice that
[ X X
µ(C) ≥ µ( Lm ) = µ(Lm ) ≥ 1/n = ∞
m∈N m∈N m∈N

which contradicts that µ is finite. Therefore, for almost every horizontal or vertical line we have µ(L) = 0.

Now fix R > 0 and n ∈ Z then [n, n + 1] × [−R, R] is a compact set of C so by part a) we have with
Fubini-Tonelli ˆ n+1 ˆ R
Uµ (x + iy)dxdy < ∞
n −R
this implies for a.e. y that
ˆ R
Uµ (x + iy)dx < ∞
−R
i.e. integrating along horizontal lines Uµ is finite for a.e. y ∈ [n, n + 1]. For each n ∈ Z define
ˆ R
n
YR := {y ∈ [n, n + 1] : Uµ (x + iy)dx < ∞}
−R
n
Ync n c
and each (YRn )c has measure zero, so YR is a set of
T S
define YR := n∈Z YR
and notice = n∈Z (YR )
full measure. And if y ∈ YR we have
ˆ R
Uµ (x + iy)dx < ∞
−R

i.e. if L = {(x, y) : x ∈ [−R, R]} then

ˆ
Uµ (x + iy)dx < ∞
L
T
Now finally define Y := R∈Z YR then arguing as before we see that Y is a set of full measure and if
y ∈ Y then we have for any R > 0
ˆ R
Uµ (x + iy)dx < ∞
−R
so it follows if we let L be the horizontal lines with y coordinates in Y then if K is compact then
ˆ
Uµ (x + iy)dx < ∞
K
as desired.
7

Finally for the third part. Observe that

ˆ ˆ ˆ
1 1 1
Sµ (z)dz = dzdµ(w)
2πi ∂R 2πi C ∂R z − w
where Fubini is justified since Sµ (z) ∈ L1 and Cauchy’s Theorem tells us that
ˆ (
1 2πi if w ∈ R
dz =
∂R z − w 0 else
so ˆ
= χR (w)dµ(w) = µ(R)
C

Problem 7. Let H be a Hilbert space and let E be a closed convex subset of H. prove that
there exists a unique element x ∈ E such that
||x|| = inf ||y||
y∈E

Proof. Let E be a closed subset of H. Then let

m := inf ||y||
y∈E

and let xn be a minimizing sequence that is xn ∈ E and

||xn || → m
It suffices to show {xn } is Cauchy since H is complete. Indeed, observe that as E is convex that
(xn + xm )/2 and (xn − xm )/2 ∈ E. Therefore, m ≤ ||(xn + xm )/2|| ≤ ||xn /2|| + ||xm /2|| → m so
||(xn + xm )/2|| → m. But then by the parallelogram law we have
||xn ||2 + ||xm |2 − 2||xn + xm ||2 = 2||xn − xm ||2
and the left hand side approaches to 0 as n, m → ∞. Therefore, the sequence is Cauchy and we are done.

Problem 8. Let F (z) be a non-constant meromorphic function on C such that for all z ∈ C
F (z + 1) = F (z) and F (z + i) = F (z)
Let Q be a square with vertices z, z+1, z+i, z+1+i such that F has no poles or zeros on ∂Q. Prove
that inside Q the function F has the same number of zeros and poles (counting multiplicities).

Proof. By the argument principle

ˆ
F 0 (z)
dz = number of zeros in Q − number of poles in Q
γ∂Q F (z)

counted with multiplicity. But as the above integral is zero due to the periodicity we are done.

Problem 9. Let X
A = {x ∈ `2 : n|xn |2 ≤ 1}
n≥1
(1) Show that A is compact in the `2 topology.
(2) Show that the mapping from A to R defined by
ˆ 2π X
dθ
x 7→ | xn einθ |
0 2π
n≥1

achieves its maximum on A.

8 RAYMOND CHU

Proof. Fix {xn } ⊂ A where we use the notation xn refers to the sequence xn = {xn,1 , xn,2 ...}. Then
notice for all n that {xn,1 }n∈N is a bounded sequence. So there is a subsequence n1j such that {xn1j ,1 }j∈N
converges to a number denoted by x1 . Now note that again {xn1j ,2 } is a bounded subsequence so there
exists a subsequence n2j ⊂ n1j with {xn2j ,2 } converging to a number x2 . Iterate this process for all n. Define
the sequence x = {x1 , x2 , .., } and consider the diagonal subsequence {xnn }. Then we claim xnn → x in
`2 . First note x ∈ A since for any N fixed we have
N
X N
X
2
1 ≥ lim n|xj j ,n | = n|xn |2
j→∞
n=1 n=1

due to pointwise convergence so taking limits gives

∞
X
1≥ n|xn |2
n=1

Now observe
∞ N ∞
X X N X
|xn − xj j ,n |2 = |xn − xj j ,n |2 + |xn − xj j ,n |2
n=1 n=1
N
n=N +1
N ∞
X 1 X
≤ |xn − xj j ,n |2 + n|xn − xj j ,n |2
n=1
N
n=N +1
N
X 2
≤ |xn − xj j ,n |2 +
n=1
N
now fix ε > 0 and choose N so that 2/N < ε/2 and now make j so large such that the first sum is < ε/2
which can be done thanks to pointwise convergence. Therefore, xj j → x in `2 so A is compact.

For the second part observe that since (einθ , eimθ ) = 2πδnm that
∞
X ∞
X ∞
X ∞
X
|| an einθ ||2L2 = ( an einθ , an einθ ) = 2π |an |2
n=1 n=1 n=1 n=1

so it follows that for

ˆ 2π X dθ
f (x) := xn einθ
0 2π
n≥1

that
ˆ 2π ˆ 2π
X X dθ 1 X
|f (x) − f (y)| = xn einθ − yn einθ ≤ (xn − yn )einθ dθ
0 2π 2π 0
n≥1 n≥1 n≥1
 2 1/2
ˆ
 !1/2
∞
1 2π X √ X √
≤√ (xn − yn )einθ   = 2π |xn − yn | 2
= 2π||x − y||`2
  

2π 0 n≥1 n=1

where the equality is due to {einθ /2π} being an orthonormal basis of L2 ([0, 2π]) which shows that f is
continuous so it attains its max over A since A is compact.

Problem 10. Let Ω ⊂ C be a connected open set, let z0 ∈ Ω, and let U be the set of positive
harmonic functions U on Ω such that U (z0 ) = 1. Prove for every compact set K ⊂ Ω there is a
finite constant M (depending on Ω, z0 , and K) such that
sup sup U (z) ≤ M
U ∈U z∈K
You may use Harnack’s inequality for the disk without proving it, provided you state it correctly.
9

Proof. Harnack’s Inequality on a ball states that if Br (w) is a ball of radius r and center w and A is the
set of positive harmonic functions on Br (w) then there exists a constant C(r) depending only on r such
that
sup u(z) ≤ C(r) inf u(z)
z∈Br (w) z∈Br (w)

for any u ∈ A. Now if K ⊂ Ω is compact then there exists a δ > 0 such that d(K, Ω) > 2δ. Then
the collection {Bδ (z)}z∈K is an open cover of K so there exists a finite sub-cover {Bδ (zk )}N
k=0 where we
adjoined Bδ/2 (z0 ). Therefore, by Harnack’s on each of these balls we have
sup u(z) ≤ C inf u(z)
z∈Bδ (zi ) z∈Bδ/2 (zi )

Now as Ω is connected in C it is path connected, so after a cyclic permutation on {zi }N i=1 let γ be a piece
wise line that connects zi to zi+1 for i = 0, .., n − 1 where Bδ/2 (zi ) ∩ Bδ/2 (zi+1 ) 6= ∅. It therefore, follows
that
sup u(z) ≤ C
z∈Bδ (z0 )
since u(z0 ) = 1, which implies as Bδ (z0 ) ∩ Bδ (z1 ) 6= ∅ that
sup u(z) ≤ C sup u(z) ≤ C 2
z∈Bδ (z1 ) Bδ (z0 )

then again as all the balls are not disjoint we can iterate to get
sup ≤ C i+1
z∈Bδ (zi )

Therefore, it follows that

sup u(z) ≤ max{C, C N +1 } := A
z∈ N
S
i=0 Bδ (zi )

and as K is contained in these balls we have

sup u(z) ≤ A
z∈K
as desired.

Problem 11. Let φ : R → R be in Cc (R). Prove that there is a constant A such that
||f ∗ φ||Lq ≤ A||f ||Lp for all 1 ≤ p ≤ q ≤ ∞ and f ∈ Lp

Problem 11. We first prove Young’s Convolution Inequality: If p, r, q ≥ 1 are such that
1 1 1
1+ = +
r p q
with f ∈ Lp and g ∈ Lq . Then we have
||f ∗ g||Lr ≤ ||f ||Lp ||g||Lq
Indeed, observe that one has
ˆ ˆ
|f ∗ g| ≤ |f (y)g(x − y)|dy = |f (y)|1−p/r |g(x − y)|1−q/r |f (y)|p/r |g(x − y)|q/r dy
ˆ
1/r
= (|f (y)|p |g(x − y)|q ) (|f (y)|(r−p) )1/r (|g(x − y)|(r−q) )1/r
Notice that
1 r−p r−q 1 1 1 1 1 1 1 1
+ + = + − + − = + − =1
r pr rq r p r q r p q r
so we can apply Holder’s Inequality to obtain
r−p r−q
|f ∗ g| ≤ (||f ||p/r q/r
p ||g||q )(||f ||p
r
)(||g||q r )
So now we have from the previous computation
ˆ ˆ ˆ r
|f ∗ g|r dx ≤ f (x − y)g(y)dy
10 RAYMOND CHU

ˆ ˆ r
p/r q/r 1−p/r 1−q/r
≤ |f (x − y)| |g(y)| |f (x − y)| |g(y)| dy
ˆ ˆ
≤ ||f ||r−p
p ||g||r−q
q |f (x − y)|p |g(y)|q dxdy
ˆ ˆ
r−p r−q q
≤ ||f ||p ||g||q |g(y)| |f (x − y)|p

= ||f ||rp ||g||rq

so we have
||f ∗ g||Lr ≤ ||f ||p ||g||q
Now back to the problem, define r ≥ 1 such that
1 1 1
:= 1 + −
r q p
then we can define
A := ||ϕ||Lr
to get the desired result.

Problem 12. Let F be a function from the open unit disk D to D such that whenever z1 , z2 and
z3 are distinct points of D there exists an analytic function fz1 ,z2 ,z3 from D to D such that
F (zj ) = fz1 ,z2 ,z3 (zj ) for j = 1, 2, 3
Prove that F is analytic at every point of D.

Proof. By Montel’s Theorem the family {fz1 ,z3 ,z3 }z1 ,z2 ,z3 ∈D is uniformly Lipschitz. Therefore, the differ-
ence quotient
F (zn ) − F (z) fz ,z,w (zn ) − fzn ,z,w (z)
= n
zn − z zn − z
is uniformly bounded where w 6= zn or z. So along a sub-sequence the difference quotients converge. Now
let zi(n) be a sub-sequence such that the difference quotient converges and zj(n) be another sub-sequence.
Then the family
{fzi(n) ,zj(n) ,z }n∈N
is pre-compact by Montel’s Theorem. So for δ > 0 so small such that Bδ (z) ⊂ D we have that along a
subsequence nk the functions fi(n),j(n),z := fzi(n) ,zj(n) ,z converge uniformly to a holomorphic function f
0
on Bδ (z). This implies fi(n),j(n),z → f 0 uniformly on Bδ (z) (denote the convergent sub-sequence as nk .
So as
F (zi(nk ) ) − F (z) fi(nk ),j(nk ),z (zi(nk ) ) − fi(nk ),j(nk ),z (z)
= → f 0 (z)
zi(nk ) − z zi(nk ) − z
and similarly
F (zj(nk ) ) − F (z) fi(nk ),j(nk ),z (zj(nk ) ) − fi(nk ),j(nk ),z (z)
= → f 0 (z)
zi(nk ) − z zi(nk ) − z
where the convergence is due to uniform convergence of this families derivative. Therefore, the limit on
any two sub-sequence is the same, so we know the limit exists and is f 0 (z). Therefore, F is holomorphic
on D.

Problem 13. Let X and Y be two Banach Spaces. Let A : X → Y be a compact operator.
Suppose X is reflexive and X ∗ is separable. Show that A is compact iff for every bounded
sequence {xn } there exists a sub-sequence {xnj } and a vector φ ∈ X such that xnj = φ + rnj and
Arnj → 0 in Y
11

Proof. Let A be compact and {xn } bounded. By Banach Alagou (since X is reflexive and X ∗ is separable
so we can upgrade weak* compactness to weak* subsequential compactness) the closed ball BM (0) ⊂ X
is weakly compact for M > 0. And as X ∗ is separable, we know that BM (0) is weakly subsequentially
compact, so there is a subsequence {xnj } such that xnj * φ ∈ BM (0). Define rnj := (xnj − φ) then
xnj = φ + rnj . As A is compact we can by looking at a furhter subsequence if necessary assume that
Axnj converges. So
lim A(xnj ) = lim A(φ + rnj ) = A(φ) + lim A(rnj )
j→∞ j→∞ j→∞
Thus it satisfies to show
lim A(xnj ) = A(φ)
j→∞
Indeed, let f ∈ Y ∗ and define y := limj→∞ A(xnj )
f (y) = lim f (A(xnj )) = lim (At f )(xnj ) = At f (φ) = f (A(φ))
j→∞ j→∞

which implies y = Aφ so this direction is proven.

The other direction implies A(xnj ) → A(φ) by linearity so A is compact.

12 RAYMOND CHU

2. Fall 2010

Problem 1. For this problem, consider just the Lebesgue measurable functions f : [0, 1] → R
together with the Lebesgue emasure.
(1) State Fatou’s Lemma (no proof is required).
(2) State and prove the Dominated Convergence ´Theorem.
(3) Given an example where fn (x) → 0 a.e., but fn (x)dx → 1

Problem 1a. Fatou’s Lemma states that if fn : [0, 1] → [0, ∞) are a sequence of Lebesgue measurable
functions then we have ˆ 1 ˆ 1
lim inf fn (x)dx ≤ lim inf fn (x)dx
0 n→∞ n→∞ 0
1
Problem 1b. DCT states that if fn ∈ L ([0, 1]) are such that fn → f pointwise a.e. and if there exists
a g ∈ L1 ([0, 1]) such that |fn | ≤ g then we have

fn → f in L1 ([0, 1])
To see the proof first observe that 2g − |f − fn | ≥ 0 so we have by Fatou’s Lemma
ˆ 1 ˆ 1 ˆ 1
2g(x)dx = lim inf {2g − |f − fn |} dx ≤ lim inf 2g − |f − fn |dx
0 0 n→∞ n→∞ 0
so we have
ˆ 1 ˆ 1 ˆ 1
0 ≤ lim inf −|f − fn |dx = − lim sup |f − fn | i.e. 0 ≥ lim sup |f − fn |
n→∞ 0 n→∞ 0 n→∞ 0

so fn → f in L1 ([0, 1]) .
Problem 1c. Take f (x) = nχ[0,1/n] (x)

Problem 2. Prove the following form of Jensen’s inequality: If f : [0, 1] → R is continuous then
ˆ 1 ˆ 1
f (x)
e dx ≥ exp{ f (x)dx}
0 0

Proof. As ex is convex we have that for any y, z ∈ R

ey ≥ ez + ez (y − z)
´1
Taking z = 0
f (x) and y = f (x) we have
´1 ´1
ˆ 1
f (x) f (x)dx f (x)dx
e ≥e 0 +e 0 (f (x) − f (x)dx)
0
so integrating this gives
ˆ 1 ˆ 1
f (x)
e dx ≥ exp f (x)dx
0 0
as desired .

Problem 3. Consider the following sequence of functions:

fn : [0, 1] → R by fn (x) = exp(sin(2πnx))
(1) Prove that fn converges weakly in L1 ([0, 1])
(2) Prove that fn converges weakly-* in L∞ ([0, 1]) viewed as the dual of L1 ([0, 1])
13

Proof. Observe that

ˆ 1 ˆ ˆ 1 ˆ 1
1 n 1
exp(sin(2πnx))dx = exp(sin(2πy))dy = n exp(sin(2πy))dy = f1 (x)dx
0 n 0 n 0 0

So it follows that L : (L1 ([0, 1]))∗ → R defined via

ˆ 1
L(f ) := f (x)dx
0
´1
satisfies L(fn ) → C := 0 f1 (x)dx so we make the guess that fn * C. Let L ∈ (L1 ([0, 1]))∗ then by
Risez-Representation Theorem there exists a g ∈ L∞ ([0, 1]) such that
ˆ 1
L(f ) = gf dx
0
for all f ∈ L1 ([0, 1]). Note that L∞ ([0, 1]) ⊂ L1 ([0, 1]), so it suffices to show the problem L1 ([0, 1])
functions. We will first prove the problem when g ∈ L∞ ([0, 1]) ∩ C([0, 1]) and use density to conclude (in
L1 norm). Indeed, observe
ˆ 1 ˆ n−1 ˆ
1 n x 1 X (k+1) x
L(fn ) = g(x) exp(sin(2πnx))dx = g( ) exp(sin(2πx))dx = g( ) exp(sin(2πx))dx
0 n 0 n n k n
k=0
so
n−1 n−1 ˆ
1 X k+1

CX k x k
L(fn ) − g( ) = g( ) − g( ) exp(sin(2πx))dx
n n n k n n
k=0 k=0
Therefore, by uniform continuity if ε > 0 there exists a δ > 0 such that if |x − y| < δ ⇒ |g(x) − g(y)| < ε.
Choose N so large such that if n ≥ N then 1/n < δ then
n−1 n−1 ˆ
CX k 1 X k+1
|L(fn ) − g( )| ≤ ε exp(sin(2πx))dx = Cε
n n n k
k=0 k=0
so as
n−1 ˆ 1
CX k
g( ) → C g(x)dx
n n 0
k=0
´1
it follows that L(fn ) → C 0 g(x)dx = L(C). Now the general case follows from density. Indeed, if
fn ∈ C([0, 1]) such that fn → f in L1 ([0, 1]) then we have
ˆ 1
|(fm − f ) exp(sin(2πnx))| . ||fm − f ||L∞
0
so
ˆ 1 ˆ 1 ˆ 1 ˆ 1
|f exp(sin(2πnx))−Cf |dx ≤ |(f −fm ) exp(sin(2πnx))|+ |fm exp(sin(2πnx))−Cfm |dx+ |Cf −Cfm | → 0
0 0 0 0
´1
so if f ∈ L1 ([0, 1]) then L(fn ) → 0
Cf (x)dx i.e. (sin(2πnx)) * C in L1 .

Note we proved the second part in the proof of the first part.

Problem 4. Let T be a linear transformation on Cc (R) such that

||f ||L1
||T f ||L∞ ≤ ||f ||L∞ and m({x ∈ R : |T f (x)| > λ}) ≤
λ
Prove that for all f ∈ Cc (R) ˆ ˆ
|T f (x)|2 . |f (x)|2
R R
14 RAYMOND CHU

Problem 4. Fix f ∈ Cc0 (R) and assume that f ≥ 0 then by the Layer Cake Decomposition we have
ˆ ˆ
.|T f (x)|2 dx = 2 s|x : |T f (x)| ≥ s|ds
R R
Now we decompose f (x) = g(x, s) + h(x, s) where g(x, s) := min{f (x), s/2} and h(x, s) = 0 if f (x) < s/2
and h(x, s) = f (x) − s/2 for f (x) ≥ s/2. Then observe that since |T f | ≤ |T g| + |T h| so if |T f | > s then
we must have |T g| > s/2 or |T h| > s/2 so
{x : |T f (x)| > s} ⊂ {x : |T g(x)| > s/2} ∪ {x : |T h(x)| > s/2}
But as ||T g||L∞ ≤ ||g||L∞ = s/2 we conclude the first set is a null set. Therefore,
ˆ ˆ ˆ
|T f (x)|2 dx ≤ 2 s|x : |T h(x)| > s/2|ds ≤ 4 ||h(x, s)||L1 (R) ds
R R R
ˆ ˆ ˆ ˆ
=4 (f (x) − s/2)dxds ≤ 4 f (x)dxds
R {x:f (x)≥s/2} R {x:f (x)≥s/2}
ˆ ˆ 2f (x) ˆ
=4 f (x)dsdx = 8 |f (x)|2 dx
R 0 R
For a general f decompose it into its positive and imaginary part

Problem 5. Let R \ Z denote the torus (whose elements we will write as cosets) and fix an
irrational number α > 0
(1) Show that
N −1 ˆ 1
1 X
lim f (nα + Z) = f (x + Z)dx
N →∞ N 0
n=0
for all f ∈ C(R \ Z)
(2) Show that the conclusion is also true when f is the characteristic function of a closed
interval.

Proof. By Stone-Weiestrass trigonometric polynomials are dense on R/Z ' [0, 1]. By linearity it suffices to
show that e2πikx for k ∈ Z satisfies the desired conclusion to have it hold for all trigonometric polynomials.
Then ˆ ˆ
1 1
e2πikx dx = cos(2πkx) + i sin(2πkx)dx = 0
0 0
and we have
N −1 N −1
1 X 1 X n 1 1 − exp(2πiknα)
exp(2πiknα) = (exp(2πikα)) =
N n=0 N n=0 N 1 − exp(2πikα)
now because α is irrational we have the denominator is never 0, so as the numerator is bounded by 2 we
have
N −1
1 X
exp(2πiknα) → 0
N n=0
so we have the desired conclusion for all trigonometric polynomials. Therefore, it follows for all continu-
ous functions by density. Since the sum operator is uniformly bounded for any N by the L∞ norm and
the integral is continuous w.r.t. uniform convergence.

For the second part, if f (x) := χ[a,b] (x) then it is obvious that there exists a sequence of functions
0 ≤ un (x) ≤ f (x) ≤ vn (x) ≤ 1 with un (x) and vn (x) pointwise converging to f (x) where un , vn are
continuous. Then we have
ˆ 1 N −1 N −1 N −1 ˆ 1
1 X 1 X 1 X
un (x)dx = lim un (nα+Z) ≤ lim f (nα+Z) ≤ lim vn (nα+Z) = vn (x)dx
0 N →∞ N N →∞ N N →∞ N 0
n=0 n=0 n=0
15

Then as all these functions are dominated by 1 DCT tells us that

ˆ 1 ˆ 1 ˆ 1
lim un (x) = lim vn (x) = f (x)
n→∞ 0 n→∞ 0 0
so it follows that
N −1 ˆ 1
1 X
lim f (nα + Z) = f (x)
N →∞ N 0
n=0

Problem 6. Consider the Hilbert space

X∞ ∞
X
H := {f : D → C : f (z) = fˆ(k)z k with ||f ||2 = (1 + |k|2 )|fˆ(k)|2 < ∞}
k=0 k=0
(1) Prove that the linear functional L : f 7→ f (1) is bounded.
(2) Find the element g ∈ H representing L.
(3) Sow that f 7→ ReL(f ) achieves its maximal value on the set
B := {f ∈ H : ||f || ≤ 1 and f (0) = 0}
that this maximum occurs at a unique point, and determine its maximal value.

Problem 6a. Observe that

∞
X
f (1) = fˆ(k)
k=0
and !1/2 !1/2
∞ p ∞ ∞
X
ˆ 1 + |k|2 X
ˆ 2 2
X 1
|f (1)| ≤ |f (k)| p ≤ |f (k)| (1 + |k| ) . ||f ||
1 + |k|2 1 + |k|2
k=0 k=0 k=0
where the second inequality is due to Holder and the last one is due to the sum on the right is a convergent
sum.
Problem 6b. Observe that the inner product is
∞
X
(f, g) = (1 + |k|2 )fˆ(k)ĝ(k)
k=0
1
so if ĝ(k) = 1+|k|2 then
(f, g) = f (1)
Therefore, define
∞
X zk
g(z) :=
1 + |k|2
k=0
1 1
absolutely and ||g||2 =
P P
which is a well defiend function on D since 1+|k|2 converges 1+|k|2 < ∞ so
g is the desired element representing L.
Problem 6c. Note on B that we have
∞
X zk
h(z) :=
1 + |k|2
k=1
is such that
(f, g) = f (1)
for f ∈ B by an identical argument as above since f (0) = 0. Now by Cauchy-Schwarz we have
|ReL(f )| = |Ref (1)| = |Re(h, f )| ≤ ||h||
since ||f || ≤ 1. Now taking f := h/||h|| shows that this maximal value occurs. This maximal is also
unique since equality in Cauchy-Schwarz inequality happens iff f = λh for some λ and by the constraint
||λh|| = 1 implies f = h/||h||.
16 RAYMOND CHU

Problem 7. Suppose f : C → C is continuous on C and holomorphic on C \ R. Prove that f is

entire.

Proof. This is a standard application of Morrera’s Theorem/Schwarz Reflection Principle.

Problem 8. Let A(D) be the C-vector space of all holomorphic functions on D and suppose that
L : A(D) → C is a multiplicative linear functional, that is
L(af + bg) = aL(f ) + bL(g) and L(f g) = L(f )L(g)
for all a, b ∈ C and all f, g ∈ A(D). If L is not identically zero, show that there is a z0 ∈ D so that
L(f ) = f (z0 ) for all f ∈ A(D).

Proof. Observe that for any f ∈ A(D) that

L(f ) = L(f · 1) = L(f )L(1) ⇒ L(1) = 1
So linearlity implies for any constant C that we have L(C) = 1. Define
z0 := L(z)
then observe
n
Y
L(z n ) = L( z) = (z0 )n
i=1
so linearity implies for any polynomial P we have
L(P (z)) = P (z0 )
1
Now we show z0 ∈ D. Assume z0 ∈
/ D then z−z0 ∈ A(D) so
z − z0 1
1 = L(1) = L( ) = L(z − z0 )L( )
z − z0 z − z0
but as L(z − z0 ) = 0 we arrive at a contradiction. Therefore, z0 ∈ D, so we have for any f ∈ A(D) that
L(f − f (z0 )) = L((z − z0 )g(z)) = 0
where g(z) ∈ A(D). So in particular linearity implies L(f ) = L(f (z0 )) = f (z0 ) for any f ∈ A(D).

P∞
Problem 9. Let f (z) = n=0 an z n be holomorphic in D. Show that if
∞
X
n|an | ≤ |a1 |
n=2
with a1 6= 0 then f is injective.

Proof. If f 0 (z0 ) 6= 0 for all z0 ∈ D then the inverse function theorem implies f (z) is locally injective on
D and the open mapping theorem implies f (z) is injective on D. So we compute
∞
X ∞
X ∞
X
|f 0 (z)| = nan z n−1 = a1 + nan z n−1 ≥ |a1 | − | nan z n−1 |
n=1 n=2 n=2
Now for any z ∈ D we have |z| < 1 so we have by the triangle inequality
∞
X ∞
X ∞
X
| nan z n−1 | ≤ n|an ||z| < n|an |
n=2 n=2 n=2
from which it follows that
∞
X
|f 0 (z)| > |a1 | − n|an | ≥ 0
n=2
so it follows that f 0 (z) 6= 0 on D, so f is injective.
17

Problem 10. Prove that the punctured disk {z ∈ C : 0 < |z| < 1} and the annulus given by
{z ∈ C : 1 < |z| < 2} are not conformally equivalent.

Proof. Assume for the sake of contradiction that there exists a conformal map ϕ from the punctured unit
disk to the annulus A1,2 := {z ∈ C : 1 < |z| < 2}. Then notice this implies ϕ is bounded, so by Riemann
Extension Theorem ϕ extends to a holomorphic map on D. Note by the mean value property we deduce
that 1 < |ϕ(0)| < 2 so ϕ is still a surjective map onto A1,2 . Now observe it is still injective since if
ϕ(0) = ϕ(w) for some w 6= 0 the open mapping principle tells us by taking small balls arounnd 0 and w
that ϕ is not injective on the punctured unit disk which is a contradiction. Therefore, we have found a
homeomorphism from the unit disk to the annulus A1,2 , which is a contradiction since homeomorphisms
preserve simply connectedness.

Problem 11. Let Ω ⊂ C be a non-empty open connected set. If f : Ω → C is harmonic such

that f 2 is also harmonic, show that either f or f is holomorphic on Ω.

Problem 11. Define the Wirtinger Derivatives

1 ∂ ∂
∂z := −i
2 ∂x ∂y

1 ∂ ∂
∂z := +i
2 ∂x ∂y
and the Cauchy Riemann equations imply f is holomorphic iff ∂z f = 0 and f is holomorphic iff ∂z f = 0.
A standard computation yields
∂2f
=4
∂z∂z
Now observe since ∆f = ∆f 2 = 0 we obtain
∂f ∂f
2 =0
∂z ∂z
so as f is C 2 in the real sense we conclude either ∂z f = 0 or ∂z f = 0 in Ω i.e. f or f is holomorphic.

Problem 12. Let F be the family of holomorphic functions on D with

ˆ
|f (x + iy)|2 dA(x, y) < 1
D
prove that for each compact subset K ⊂ D there is an A > 0 such that |f (z)| < A for all z ∈ K
and f ∈ F

Proof. Let r := dist(K, Ω)/2 where K is a compact subset of Ω. Then for any z ∈ K observe that
Br (z) ⊂ Ω so the mean value property tells us that
ˆ
1
u(z) = 2 u(z)dz
πr Br (z)
so we have ˆ
1
|u(z)| ≤ 2 |u(z)|dz
πr Br (z)
1 √ 1 1
≤ 2 πr||u||L2 (Br (z)) ≤ √ ||u||L2 (B1 (0)) = √
πr πr πr
and this bound is independent of z ∈ K so we have the desired conclusion.
18 RAYMOND CHU

3. Spring 2011

Problem 1. Define what it means to say that fn * f weakly in L2 ([0, 1]).

Suppose fn ∈ L2 ([0, 1]) converges weakly to f ∈ L2 ([0, 1]) and define the primitive functions
ˆ x ˆ x
Fn (x) := fn (t)dt and F (x) := f (t)dt
0 0
Show that Fn , F ∈ C([0, 1]) and that Fn → F uniformly on [0, 1].

Problem 1a. Let X := L2 ([0, 1]) then X 3 fn * f ∈ X means for any L ∈ X ∗ where X ∗ is the
topological dual of X then L(fn ) → L(f ). Fix L ∈ X then by Risez Representation Theorem we have
that there exists g ∈ X such that ∀f ∈ X
ˆ 1
L(f ) = f (x)g(x)dx
0
so it is equivalent that fn * f to mean for any g ∈ X that
ˆ 1 ˆ 1
fn (x)g(x)dx → f (x)g(x)dx
0 0

Problem 1b. Since fn * f it follows that for any L ∈ X ∗ that supn |L(fn )| < ∞ since it is a convergent
sequence, so by the uniform boundness principle we have supn ||fn ||L2 ([0,1]) < ∞. So as
ˆ 1
|Fn (x)| ≤ |fn (t)|dt ≤ ||fn ||L2 ([0,1]
0
due to Cauchy-Schwarz we have that the family is uniformly bounded. Now obserrve it is equicontinuous
since if 0 ≤ x ≤ y ≤ 1 we have
ˆ y ˆ 1 p p
|Fn (x) − Fn (y)| ≤ |fn (t)|dt = |fn (t)|χ[x,y] dt ≤ ||fn ||L2 |x − y| . |x − y|
x 0

where the final inequality is due to ||fn ||L2 being uniformly bounded, so the family is equicontinuous. So
now observe that as χ[0,x] (t) ∈ X ∗ = L2 ([0, 1]) we have
ˆ x ˆ 1 ˆ x
Fn (x) = fn (t)dt = fn (t)χ[0,x] (t)dt → f (t) = F (x)
0 0 0
so Fn → F pointwise. Then for any subsequence we have from Arzela-Ascoli a further subsequence which
uniformly converges to F (x), which implies the whole sequence uniformly converges to F (x) .

Problem 2. Let f ∈ L3 (R) and

(
sin(πx) : |x| ≤ 1
φ(x) :=
0 else
Show that ˆ
fn (x) := n f (x − y)φ(ny)dy → 0
R
Lebesgue almost everywhere.

Proof. Note that if we define φn (x) := φ(nx) we have that

ˆ ˆ 1/n
φn (x) = sin(nπx)dx = 0
R −1/n

so we have
ˆ ˆ 1/n
|fn (x)| ≤ n |f (x − y) − f (x)||φ(ny)|dy ≤ n |f (x − y) − f (x)|dy
R −1/n
19

and Holder’s Inequality implies f ∈ L1loc (R) so we conclude by the Lebesgue Differentiation Theorem
that ( ˆ )
1/n
lim sup n |f (x − y) − f (x)|dy = 0 a.e.
n→∞ −1/n

so fn (x) → 0 a.e.

´
Problem 3. Let µ be a Borel probability measure on R and define f (t) := eitx dµ(x). Suppose
also that
f (0) − f (t)
lim =0
t→0 t2
Show that µ is supported at {0}.

Proof. First observe that

ˆ
f (0) − f (t) 1 − eitx
2
= dµ(x)
t R t2
and by using Taylor Expansion, we see that
1 − eitx x2

lim Re 2
=
t→0 t 2
and we have from Fatou’s Lemma that (since Re(1 − eitx ) ≥ 0 that
ˆ ˆ ˆ 2
Re(1 − eitx ) Re(1 − eitx ) x
0 = lim inf 2
dµ(x) ≥ lim inf 2
dµ(x) = dµ(x)
t→0 R t R t→0 t R 2

So now suppose for the sake of contradiction that µ is not supported on {0}, this means we can find a
measurable set with 0 ∈
/ E such that µ(E) > 0. As µ is a Borel probability measure on the metric space
R we conclude that µ is regular, so we can find a compact set K ⊂ E such that µ(K) > 0. As K is
compact there exists a minimum m ∈ K such that |m| > 0 (since 0 ∈ / K). Therefore,
ˆ 2 ˆ 2 2
x x m
0= dµ(x) ≥ dµ(x) ≥ µ(K) > 0
R 2 K 2 2
which is our desired contradiction.

Problem 4. Let fn : [0, 1] → [0, ∞) be Borel functions with

ˆ 1
sup fn (x) log(2 + fn (x))dx < ∞
n 0
Suppose fn → f Lebesgue almost everywhere. Show that f ∈ L1 and fn → f in the L1 sense.

Proof. Assume that there exists a C independent of n such that

ˆ 1
sup fn (x) log(2 + fn (x))dx ≤ C < ∞
n∈N 0
and there exists an f such that fn → f and fn : R → R+ . First observe that as log(x) is an increasing
function that log(2 + fn (x)) ≤ log(2) so it follows that
ˆ 1 ˆ 1
sup fn (x) log(2)dx ≤ sup fn (x) log(2 + fn (x))dx ≤ C
n∈N 0 n∈N 0
so each fn ∈ L1 with a uniform bound of C. Then we have from Fatou’s Lemma that
ˆ 1 ˆ 1 ˆ 1
f (x) log(2)dx ≤ f (x) log(2 + f (x))dx ≤ lim inf fn (x) log(2 + fn (x))dx ≤ C
0 0 n→∞ 0
20 RAYMOND CHU

and we have that f ∈ L1 . Now fix ε > 0 and assume E is measurable with |E| < δ then for AM n := {x :
|fn (x)| ≤ M }
ˆ ˆ ˆ
log(2 + fn (x))
fn (x)dx = fn (x)dx + fn (x) dx = (I) + (II)
E E∩AM
n E∩(AM
n )
c log(2 + fn (x))
now observe
ˆ 1
C
(II) ≤ fn (x) log(2 + fn (x)) / log(2 + M ) ≤
0 log(2 + M )
and
(I) ≤ M δ
ε
so taking δ = 2M we conclude that if M is very large then
ˆ
fn (x)dx ≤ ε
E
´
so {fn } is uniformly integrable. Now choose a δ > 0 so small such that if |E| < δ ⇒ E fn (x)dx ≤ ε/2.
Now by Egorov’s theorem there exists a compact set K ⊂ [0, 1] such that |[0, 1] \ K| ≤ δ and fn → f on
K. So if n is large then supK |fn (x) − f (x)| ≤ ε/2 so
ˆ 1 ˆ ˆ
|fn (x) − f (x)|dx = |fn (x) − f (x)|dx + |fn (x) − f (x)|dx
0 K [0,1]\K
≤ ε/2 + ε/2 = ε
1
so fn → f in L .

Problem 5. Show that `∞ (Z) contains continnuum many functions xα : Z → R obeying

||xα ||`∞ = 1 and ||xα − xβ ||`∞ ≥ 1
Deduce (assuming the axiom of choice) that the Banach Space `∞ (Z) is not separable.

Deduce that `1 (Z) is not reflexive.

Proof. Consider the set of binary strings i.e. xα = (xα,1 , xα,2 , ..) where each xi = 0 or 1. This is
uncountably many distinct elements in `∞ (Z). Fix any two distinct binary strings xα and xβ that are
not identically zero then
||xα ||`∞ = 1
And since xα 6= xβ there is a j such that |xα,j − xβ,j | = 1. Therefore,
||xα − xβ ||`∞ ≥ 1

Let {ej }j∈N be a countable subset of `∞ (Z). Arguing for the sake of a contradiction, if {ej }j∈N is
dense, then for each j we can find an α(j) such that xα(j) is a binary string and
||ej − xα(j) ||`∞ ≤ 1/2
Then this implies if β 6= α(j) then the reverse triangle inequality implies
||ej − xβ || ≥ ||xα(j) − xβ || − ||ej − xα(j) || ≥ 1 − 1/2 = 1/2
Therefore, as {xα(j) }j∈N is countable and {xα } is uncountable, we may find a binary string xβ ∈
/
{xα(j) }j∈N . Therefore, for all j ∈ N then
||ej − xβ || ≥ 1/2
so ej cannot be dense. So we have arrived at a contradiction, so `∞ (Z) is not separable.
21

Note that the dual of `1 (Z) is `∞ (Z), so if `1 (Z) was reflexive then `1 (Z) ∼
= (`∞ (Z))∗ . Now we claim
∗
that if X is separable then this implies X is separable (where X is a normed linear space). This lemma
will then gives us `1 (Z) is isomorphic to a non-separable space, so this implies `1 (Z) is not separable, but
`1 (Z) is separable, which is our contradiction. So it suffices to prove the lemma.

Indeed, let {fn } ⊂ X ∗ be a dense countable set. Then for each fn if we define
||fn || := sup |fn (x)|
x∈X:||x||=1

then there is some xn ∈ X with ||xn || = 1 such that

fn (xn ) ≥ 1/2||fn ||
PN
thanks to linearity. Now let S be the set of finite rational combinations of xn i.e. x ∈ S if x = n=1 qn xn
for some qn ∈ Q and N ∈ N. We claim S = X, so assume for the sake of contradiction it is not. So fix
x ∈ X \ S with ||x|| = 1 then by Hanh-Banach there is some f ∈ X ∗ such that f (x) 6= 0 and f |S = 0.
But observe from the triangle inequality that
|fn (xn )| ≤ |f (xn ) − fn (xn )| + |f (xn )| = |f (xn ) − fn (xn )|
and by our choice of xn we have
1/2||fn || ≤ |fn (xn )| ≤ |f (xn ) − fn (xn )|
i.e.
||fn || ≤ 2||f − fn ||
and as fn is dense we can find a subsequence nk such that ||fnk − f || → 0. Thus in particular, ||fnk || →
0,from which it follows that ||f || = 0 i.e. f is the zero operator, which is our desired contradiction.

Problem 6. Suppose µ and ν are finite positive (regular) Borel measures on Rn . Prove the
existence of the Lebesgue decomposition: There is a unique pair of positive Borel measures µa
and µs so that
µ = µa + µs , µa ν, and µs ⊥ ν

Proof. As µ is finite we have for any Borel set E that µ(E) < ∞. Therefore, consider
ˆ ˆ
E := sup { f (x)dν(x) : f (x)dν(x) ≤ µ(E) for all E Borel }
f ∈L1 (dν),f ≥0 ν a.e. Rn E

Let fn be a maximizing sequence i.e.

ˆ
lim fn (x)dν(x) = E
n→∞ Rn

where fn ∈ L1 (dν). Observe that if gm := max{f1 , .., fm } satisfy the above constraints too. Indeed,
gm ≥ 0 trivially and if we define Ej to be the set where gm = fj we have
ˆ Xm ˆ m
X
gm dν(x) = fj dν(x) ≤ ν(E ∩ Ej ) = ν(E)
E j=1 E∩Ej j=1

where for the last equality we used Ej partitions E. So it follows that fn ≤ gn for all n and the monotone
convergence theorem shows that
ˆ
E= sup {fm (x)}dν(x) ≤ µ(Rn )
Rn m∈N

so if we define f := supm∈N fm (x) then it obtains this maximum. Now define

ˆ
µa (E) := f (x)dν(x)
E
22 RAYMOND CHU

and note µa ν µs := µ − µa . Suppose for the sake of contradiction that µs is not perpendicular to ν.
Then there is an ε > 0 and a borel set E with ν(E) > 0 such that
(µs − εν) ≥ 0 on E
i.e. for any Borel set F we have
µs (F ∩ E) − εν(F ∩ S) ≥ 0
ˆ
µ(F ∩ E) ≥ εν(F ∩ E) + f (x)dν(x)
F ∩E
this means if g(x) := f (x) + εχE (x) then we have found a strictly bigger maximizer, which is a contra-
diction to the definition of f (x). Therefore, µs ⊥ ν. So we have shown such a decomposition exists. Now
if there were two such decomposition’s denoted µa1 , µs1 , µa2 , µs2 then we have
µs1 − µs2 = µa2 − µa1
the left hand side is singular to ν and the right hand side is absolutely continuous to ν, so µs1 = µs2 and
µa1 = µa2 , so this decomposition is unique.

Problem 7. Prove Gorsat’s Theorem: if f : C → C is complex differentiable then for every

triangle ∆ ⊂ C ˆ
f (z)dz = 0
∂∆
where the line integral is over the three sides of the triangle.

Proof.

Problem 8. (1) Define upper-semicontinuous for functions f : C → [−∞, ∞).

(2) Define what it means for such an upper-semicontinuous to be subharmonic.
(3) Prove or refute each of the following
• The pointwise supremum of a bounded family of subharmonic function is subhar-
monic.
• The pointwise infimum of a family of subharmonic functions is subharmonic.
(4) Let A(z) be a 2 × 2 matrix-valued holomorphic function (i.e. the entires are holomorphic).
Show that
z 7→ log(||A(z)||) is subharmonic
where ||A(z)|| is the operator norm on C2 .

Problem 8a. An upper-semicontinuous function f : C → [−∞, ∞) is a function such that for any α ∈ R
{z ∈ C : f (z) < α} is open
Problem 8b. We say a upper-semicontinuous function u : C → [−∞, ∞) if for any r > 0 small enough
we have ˆ 2π
1
f (z) ≤ f (z + reiθ )dθ
2π θ=0
i.e. the local sub-mean value property.
Problem 8c. The first claim is true as long as the pointwise supremum is upper semi-continuous. Indeed,
observe if F is a family of subharmonic functions then for any z ∈ C and f ∈ F we have for r > 0 small
enough that we see for g(z) := supf ∈F f (z)
ˆ 2π ˆ 2π
1 1
f (z) ≤ f (z + reiθ )dθ ≤ g(z + reiθ )dθ
2π 0 2π 0
since f ≤ g. Now taking the supremum over f for the left hand side implies
ˆ 2π
1
g(z) ≤ g(z + reiθ )dθ
2π 0
23

so g(z) satisfies the local mean value inequality, so if it is upper semi-continuous then it is subharmonic.

The second claim is false. Indeed, consider f (x + iy) := x and g(x + iy) = −x then f and g are
harmonic so they are subharmonic. And min{f (x + iy), g(x + iy)} = −|x| and −|x| is concave, so it is
superharmonic. In particular −|x| is not subharmonic.
Problem 8d. Note that
||A(z)|| = sup ||A(z)w||2 = sup |hA(z)w, ξi|
w∈C2 :||w||=1 w,ξ∈C 2 :||w||=||ξ||=1

where || · ||2 refers to the Euclidan norm on C2 . So we have

log(||A(z)||) = sup log |hA(z)w, ξi|
w,ξ∈C 2 :||w||=||ξ||=1

and for each fixed w, ξ we have (A(z)w, ξ) is holomorphic in z so log |(A(z)w, ξ)| is subharmonic, so
log(||A(z)||) is subharmonic since it is the sup of a family of subharmonic functions.

Problem 9. Let E ⊂ [0, 1] be the Cantor Set. Embedding [0, 1] naturally into C, we may regrad
E ⊂ C. Suppose f : C \ E → C is holomorphic and (uniformly) bounded. Show that f is constant.

Proof. Note that E has measure zero. That is if ε > 0 there is a collection of balls B(rn , xn ) ⊂ C such
that [ X
E⊂ B(xn , rn ) and 2rn < ε
n∈N
´
Now let R ⊂ C be a rectangle. Then we will show ∂R f (z)dz = 0. This is trivially true if R ⊂ C \ E, so
SN
assume R ∩ E 6= ∅. Now as R ∩ E is compact, we can find an N ∈ N such that R ∩ E ⊂ i=1 B(rn , xn ).
Now observe that R = (R ∩ E) ∪ (R ∩ E c ) and on the second set, f is holomorphic so it integrates to zero
over any closed curve in (R ∩ E c ), so the only remaining parts of the integral is where the balls are at.
SN
Now let γi be a closed curve parametrization of the connected components of ∂( n=1 B(rn , xn )) where
i = 1, ..., m then
ˆ m ˆ
X
f (z)dz = f (z)dz
∂R i=1 γi

so we have
ˆ N
X N
X
f (z)dz ≤ M `(γi ) ≤ M `(∂B(rn , xn ))
∂R i=1 i=1
where M is the upper bound of f and ` means length.
N
X
≤ M `2πrn = O(ε)
n=1
´
sending ε → 0 implies that ∂R f (z)dz = 0, so it follows that f extends to a bounded entire function by
Morrera’s Theorem, so it is constant.

Problem 10. Let Ω = {z ∈ D : Im(z) > 0}. Evaluate

sup{Ref 0 (i/2) : f : Ω → D is holomorphic}

Proof. Consider the conformal map

z+i
ψ(z) :=
z−i
where ψ conformally maps Ω to D. Then its inverse
z+1
ψ −1 (z) = i
1−z
24 RAYMOND CHU

is a conformal map of D to Ω such that −1/3 gets mapped to i/2. Also consider the automorphisms of
the disk for α ∈ D
α−z
φα (z) :=
1 − αz
then φ maps 0 to α and α to 0. Therefore,
g(z) := φf (i/2) ◦ f ◦ ψ −1 ◦ φ−1/3
is a conformal map from D to D such that g(0) = 0. So by Schwarz Lemma we have
1 ≥ |g 0 (0)| = |φ0f (i/2) (f (i/2))f 0 (i/2)[(ψ −1 )0 (−1/3)]φ0−1/3 (0)|
and by computation
2i
φ0α (α) = 1/(1 − |α|2 ) and φ0α (0) = 1 − |α|2 and (ψ −1 )0 (z) =
(1 − z)2
i.e.
1
1≥ |f 0 (i/2)|
1 − |f (i/2)|2
so we obtain the bound
1 ≥ |f 0 (i/2)|
and by our computation above we see that this bound is obtained for any f : Ω → D such that f (i/2) = 0.
For instance take
h(z) := φψ(i/2) ◦ ψ(z)
and this function obtains the desired bound.

Problem 11. Consider the function defiend for s ∈ (1, ∞) by

ˆ ∞ s−1
x
f (s) := x−1
dx
0 e
Show that f has an analytic continuation to {s ∈ C : Res > 0, s 6= 1} with a simple pole at s = 1.
Compute the residue at s = 1.

Proof. Note that

ex − 1 = x + O(x2 ) as x → 0
so the integrand is of O(xs−2 ) near the origin, while far away it is of O(e−x ) so if Re(s) > 1 the integral
is well defined. But notice by integration by parts with u = x/(ex − 1), dv = xs−2 that in this region we
have
ˆ ∞ x
1 e (1 − x) − 1
f (s) = − xs−1 dx
s−1 0 (ex − 1)2
And note that
ex (1 − x) − 1

lim = −1/2
x→0 (ex − 1)2
so the integrand is of O(xs−1 ) near x = 0 and O(e−x ) as x → ∞. Therefore, the integral converges for
Re(s) > 0 with s 6= 1, so this is a meromorphic extension of f to {s ∈ C : Res > 0, s 6= 1}. It is also clear
from the form of f (s) that the pole is simple, so the residue at s = 1 is
ˆ ∞ x
e (1 − x) − 1
− dx = 1
0 (ex − 1)2

25

Problem 12. Let Ω := C \ (−∞, 0] and let log(z) be the branch of the complex logarithm on Ω
that is real on the positive real axis (and analytic throughout Ω). Show that for 0 < t < ∞, the
number of solutions z ∈ Ω to
t
log(z) =
z
is finite and independent of t.

Proof. Notie that if z = reiθ where θ ∈ (−π, π] then by our choice of log we have log(reiθ ) = log(r) + iθ
so if
t
log(r) + iθ = (cos(θ) − i sin(θ))
r
we obtain
t
θ = − sin(θ)
r
and as t/r > 0 we see that θ and sin(θ) have different signs, which means θ = 0 because θ ∈ (−π, π].
Therefore, we must have
t
log(r) =
r
i.e.
t = f (r) := r log(r) > 0 on r > 1
And observe that
df
= log(r) + 1 > 0 on r > 1
dr
so we have that f (r) is injective on (0, ∞) so there is at most one solution of log(r) = t/r. But it is clear
lim r log(r) = ∞
r→∞
so f (r) : (1, ∞) → (0, ∞) is surjective, so there is one solution for every t.

26 RAYMOND CHU

4. Fall 2011

Problem 1. Prove Egorov’s Theorem

Proof.

Problem 2a. Let dσ denote the surface measure on the unit sphere S2 ⊂ R3 .
(1) For ξ ∈ R3 compute ˆ
eix·ξ dσ(x)
S2

(2) Using this or otherwise show that the mapping

ˆ ˆ
f 7→ f (x + y)dσ(x)dσ(y)
S2 S2
∞
extends uniquely from f ∈ Cc (R3 ) to a bounded linear functional on L2 (R3 ).

Problem 2a. We claim that the integral only depends on |ξ|. Indeed for any orthogonal matrix we have
that Ax · ξ = x · A−1 ξ and A−1 = At . Then we have from integration by sub
ˆ ˆ ˆ
−1
eix·ξ dσ(x) = eiA x·ξ dσ(x) = eix·Aξ
S2 S2 S2
so wlog assume ξ = (0, 0, |ξ|) then
ˆ ˆ
eix·ξ dσ(x) = cos(z|ξ|) + i sin(z|ξ|)dσ(x)
S2 S2
ˆ 2π ˆ π
= [cos(cos(φ)|ξ|) + i sin(cos(φ)|ξ|)] sin(φ)dφdθ
θ=0 φ=0
ˆ 2π ˆ 1
= cos(u|ξ|) + i sin(u|ξ|)dudθ
θ=0 u=−1
ˆ |ξ|
2π
= cos(w) + i sin(w)dw
|ξ| −|ξ|
4π
= sin(|ξ|)
|ξ|
Problem 2b. Consider ˆ ˆ
L(f ) := f (x + y)dσ(x)dσ(y)
S2 S2
then for f ∈ Cc∞ (R3 ) we have
ˆ ˆ ˆ
L(f ) = fˆ(ξ)e−2πi(x+y)·ξ dξdσ(x)dσ(y)
S2 S2 Rd

and in particular by Fubini since f ∈ Cc∞ (R3 ) we have

ˆ ˆ ˆ
L(f ) = fˆ(ξ) e−2πix·ξ e−2πiy·ξ dσ(y)dσ(x)dξ
Rd S2 S2
ˆ 2
64π
= fˆ(ξ) 2 sin(|ξ|)2
Rd |ξ|
so in particular,
|L(f )| ≤ C||fˆ||L2 || sin(|ξ|)2 /|ξ|2 ||L2 ≤ K||fˆ||L2 = K||f ||L2
so L extends to a continuous operator for L2 functions.
27

Problem 3. Let 1 < p, q < ∞ such that 1/p + 1/q = 1. Fix f ∈ Lp (R3 ) and g ∈ Lq (R3 ).
(1) Show that ˆ
[f ∗ g](x) := f (x − y)g(y)dy
R3
defines a continuous function on R3 .
(2) Moreover, show that [f ∗ g](x) → 0 as x → ∞.

Problem 3a. Note that

ˆ
|[f ∗ g](x) − [f ∗ g](y)| ≤ |f (x − z) − f (y − z)||g(z)|dz
R3

≤ ||f (x − y) − f (y − z)||Lp (R3 ,dz) ||g(z)||Lq

and from the translation continuity of the integrals we deduce that this is a continuous map.
Problem 3b. Let fn , gn ∈ Cc∞ (R3 ) such that ||fn − f ||Lp < ε and ||gn − g||Lp < ε where ε > 0 is
arbitrary. Then
ˆ
|[f ∗ g](x) − [fn ∗ gn ](x)| ≤ |f (x − y)g(y) − fn (x − y)gn (y)|dy
R3
ˆ
≤ |f (x − y)g(y) − g(y)fn (x − y)| + |g(y)fn (x − y) − fn (x − y)gn (y)|dy
R3
≤ ||g||Lq ||f − fn ||Lp + ||fn ||Lp ||g − gn ||Lq
M 1 ε + M2 ε
where we used g ∈ L and fn converges in L so its bounded in Lp . As this bound was independent of x
q p

we conclude that
fn ∗ gn → f ∗ g uniformly
but as fn , gn are compactly supported we have for K1 := supp(gn ) and K2 := supp(fn ) that
ˆ
fn ∗ gn = fn (x − y)gn (y)dy
K
which implies supp(fn ∗ gn ) ⊂ K1 + K2 := {k1 + k2 : k1 ∈ K1 , k2 ∈ K2 } which implies limx→∞ fn ∗ gn = 0
for any n so uniform convergence implies limx→∞ f ∗ g = 0.

Problem 4. Let f ∈ C ∞ ([0, ∞) × [0, 1]) such that

ˆ ∞ˆ 1
|∂t f (x, t)|2 (1 + t2 )dxdt < ∞
0 0
Prove there is a function g such that f (t, ·) converges to g(·) in L2 ([0, 1]) as t → ∞.

Proof. Notice by the fundamental theorem of Calculus that we have for t1 < t2 that
ˆ t2
1 + s2
f (t2 , x) − f (t1 , x) = ∂ f (s, t)ds
2 t
t1 1 + s

so we have
ˆ t2 ˆ t2 1/2 ˆ t2 1/2
2 ∂t f (s, t) 2 2 2 1
|f (t2 , x) − f (t1 , x)| ≤ |1 + s | ds ≤ |∂t f (s, x)| |1 + s | ds ds
t1 |1 + s2 | t1 t1 |1 + s2 |2
so it follows that
ˆ 1 ˆ t2 ˆ ∞ ˆ 1
1
|f (t2 , x) − f (t1 , x)|2 dx ≤ 2 2
ds |∂t f (s, x)| 2
|1 + s2 2
| ds
x=0 t1 |1 + s | t=0 x=0
ˆ t2
1
≤C 2 2
ds
t1 |1 + s |
28 RAYMOND CHU

and as 1/(1+x2 )2 ∈ L1 ([1, ∞)) it follows that the integral term can be made arbitrarily small as t1 , t2 → 0.
So we have {f (t, x)}t∈[0,∞) is Cauchy as t → ∞ so it converges to some function g(x) ∈ L2 ([0, 1]) since
L2 is complete.

Problem 5. For a function f ∈ L1 (R), we define

ˆ x+h
1
(M f )(x) := sup |f (y)|dy
h>0 2h x−h
Prove that there is the following property: There is a constant A > 0 such that for any λ > 0
A
m({x ∈ R : M f (x) > λ}) ≤ ||f ||L1
λ
If you use a covering lemma, you should prove it.

Proof. Vitali Covering Lemma Let {Bi }N i=1 be a finite collection of balls. Then there exists a subcol-
lection of balls Bij that are disjoint such that
N
[ M
[
Bi ⊂ 3Bij
i=1 j=1

Indeed, let Bi be the ball with maximal radius of this finite collection. Then if Bj ∩ Bi 6= ∅ we also
remvoe Bj from this collection. Now we have a smaller subcollection and we repeat our algorithim of
choosing the balls with maximum radius. It’s clear from construction that the new balls {Bij }M j=1 are
disjoint and as if Bj ∩ Bk 6= ∅ with Bj being the circle with the biggest radius of the two balls then

Bk ⊂ 3Bj

so we have found such a subcollection.

Now fix λ and let m denote the Lebesgue measure then F (λ) := {x ∈ R : (M f ) > λ} now let K ⊂ F (λ)
be compact. Then we claim that F (λ) is open; indeed, if x ∈ F (λ) then there is an r > 0 such that
ˆ x+r
1
λ< |f (y)|dy := Ar (x)
2r x−r

Notice that Ar (x)is continuous in x since

ˆ
1
Ar (x) = |f (y)|χ[x−r,x+r] (y)dy
2r R

so DCT implies continuity. Therefore, by continuity there exists a small ball around x such that for any
y in this ball Ar (y) > λ. Therefore, for any x ∈ K there is a ball Bx ⊂ F (λ). Now compactness lets us
find a subcover say {B1 , .., BN } and the covering lemma lets us find a subcollection of balls {B̃1 , .., B̃M }
SN SM
that are disjoint and K ⊂ j=1 Bj ⊂ j=1 3B̃j . So in particular,
M M ˆ
X 3X 3
m(K) ≤ 3 m(B̃j ) ≤ |f | ≤ ||f ||L1 (R)
j=1
λ j=1 B̃j λ

where the last inequality is due to the balls are disjoint. Now we use that the lebesgue measure is a radon
measure so
m(F (λ)) = sup m(K)
K⊂F (λ)

to get the desired result .

29

Problem 6. Let (X, d) be a compact metric space. Let µn be a sequence of positive Borel
Measures on X that weak* converge to a finite positive Borel measure µ, that is,
ˆ ˆ
f dµn → f dµ for all f ∈ C(X)
X X
Show that if K is compact then
µ(K) ≥ lim sup µn (K)
n→∞

Proof. As K is compact, it is closed so χK is upper semi-continuous, so there is a sequence of functions

C(X) 3 fn (x) ≥ χK with fn (x) → χK pointwise. Now we have
ˆ ˆ
µn (K) = χK dµn ≤ fm dµn
X X

so we have ˆ
lim sup µn (K) ≤ fm dµ
n→∞ X
where we used weak* convergence. Now as µ is finite and f1 is bounded that f1 ∈ L1 (dµ) so by the
dominated convergence theorem we have
ˆ ˆ
lim fm dµ = χK dµ
m→∞ X X
so ˆ
lim sup µn (K) ≤ χK dµ = µ(K)
n→∞ X
as desired.

´∞ cos(x)
Problem 7. Compute 0 (1+x2 )2 dx. Justify all steps!

iz
e
Proof. Define f (z) := (1+z 2 )2 and notice that it has a pole of order 2 at z = i and z = −i. Then f (z) is

meromorphic with poles of order 2 at z = i, −i. Let γR := {Reiθ : 0 ≤ θ ≤ π} and γ := {−R(1 − t) + Rt :

t ∈ [0, 1]} then
ˆ ˆ π iθ
eiRe
f (z)dz = iReiθ dθ
γR θ=0 (1 + R2 e2iθ )2
so ˆ ˆ
π π
R R
|f (z)| ≤ e−R sin(θ) dθ ≤ e−CRθ dθ
(1 − R2 )2 θ=0 (1 − R2 )2 θ=0
so the integral of f (z) over γR converges to 0 as R → ∞. Therefore, by the residue theorem and
cos(x)
Re(f(z)) = (1+x 2 )2 is symmetric that

ˆ ∞
cos(x) π
2 2
= πiRes(f, i) =
0 (1 + x ) 2e

Problem 8. Determine the number of solutions of

z − 2 − e−z = 0
with z in the right half plane H = {z ∈ C : Re(z) > 0}.
30 RAYMOND CHU

Proof. Observe that we have for any solutions of z − 2 − e−z = 0 that

|z| ≤ 2 + |e−z | ≤ 3
since z is in the upper half plane. Therefore, it suffices to consider the region B3 (0) ∩ {x ≥ 0} := U to
find our zeros. Notice on ∂U we have
|z − 2| > |e−z |
since on the circle part we have
|z − 2| ≥ |z| − 2 = 1
and on {x = 0} we have |z − 2| = |iy − 2| = |y| + 2 > 1 so we have
|z − 2| > |e−z | on ∂U
so Rouche’s theorem implies they have the same number of zeros inside U which is exactly 1.

Problem 9. Let f be holomorphic on D∗ := D \ {0} such that f ∈ L2 (D, dxdy). Show that f
has a holomorphic extension to D.

Proof. Fix an 0 < ε 1 then on Aε,1/2 := {z : ε < |z| < 1/2} then for any z0 ∈ Aε,1/2 there is a δ > 0
such that Bδ (z) ⊂ D∗ . So fix a ρ < δ and z ∈ Aε,1/2 then we have from the Mean Value Theorem that
ˆ 2π
1
f (z) = f (z + ρeiθ )dθ
2π θ=0
so ˆ ˆ ˆ
δ δ 2π
ρf (z)dρ = ρ f (z + ρeiθ )dθdρ
ρ=0 ρ=0 θ=0
which gives ˆ
1
f (z) = 2
f (z)dλ(z)
δ π Bδ (z)
so in particular,
1
|f (z)| ≤
πδ 2
and from Holder’s we get
|f (z)| ≤ ||f ||L2 (Bδ (z))
so we have on Aε,1 that f (z) is bounded and this bound is uniform so we have |f (z)| ≤ C on B1/2 (0).
This allows us to use Riemann’s Theorem on removable singularities to conclude f (z) has a removable
singularity at zero.

Problem 10. Let Ω ⊂ C be simply connected with Ω 6= C and f : Ω → Ω is a holomorphic

mapping. Suppose there exists z1 6= z2 such that f (zi ) = zi for i = 1, 2. Show that f (z) = z for
all z ∈ Ω.

Proof. By Riemann’s Mapping Theorem there exists a conformal ψ : Ω → C and by composing with a
Mobius Transformation we can assume ψ(z1 ) = 0 so it follows that
g := ψ ◦ f ◦ ψ −1 : D → D
is conformal. Then we have g(0) = ψ ◦ f (z1 ) = ψ(z1 ) = 0 and g(ψ(z2 )) = ψ ◦ f (z2 ) = ψ(z2 ). Therefore,
by schwarz lemma as g(0) = 0 and there is a p 6= 0 such that |g(p)| = |p| we have g(z) = eiθ z for some θ
and from equality we conclude g(z) = z. This implies
f (z) = z
as desired.
31

Problem 11. Let f : C → C be holomorphic with f (z) 6= 0 for all z ∈ C. Define U = {z ∈ C :

|f (z)| < 1}. Show that all connected components of U are unbounded.

Proof. Assume that there existed a connected component of U denoted by Ω that is bounded. Then this
implies Ω is compact, so in particular we have on ∂Ω that |f (z)| = 1 and since f (z) is never zero, we
know 1/f is holomorphic from which we deduce from the maximum modulus principle |f (z)| = 1 in Ω
which contradicts the definition of U .

Problem 12. A holomorphic function f : C → C is of exponential type if there are c1 and c2 > 0
such that
|f (z)| ≤ c1 ec2 |z| for all z ∈ C
Show that f is of exponential type iff f 0 is of exponential type.

Proof. For any z ∈ C we have B1 (z) ∈ C so by Cauchy’s Estimate

|f 0 (z)| ≤ max |f (w)| ≤ max c1 exp(|c2 z + c2 eiθ |)
w∈B 1 (z) θ∈[0,2π]

≤ exp(|c2 |)c1 exp |c2 z|

0
so f is of exponential type.

Now observe ˆ
f (z) = f 0 (z) + f (0)
γ0→z
where γ0→z = {tz : t ∈ [0, 1]}
ˆ 1
= zf 0 (tz)dt + f (0)
t=0
so
|f (z)| ≤ |z||c1 exp(c2 |z|)) + |f (0)| ≤ |c1 | exp(c3 |z|) + |f (0)| ≤ |c4 | exp(c3 |z|)
so f is of exponential type
32 RAYMOND CHU

5. Spring 2012

Problem 1. Some of the following statements for functions fn in L3 ([0, 1]) are false. Indicate
these and provide an appropriate counter example.
(1) If fn converges a.e. to f then a subsequence of fn converges to f in L3 .
(2) If fn converges to f in L3 then a subsequence converges almost everywhere.
(3) If fn converges to f in measure then the sequence converges to f in L3 .
(4) If fn converges to f in L3 then the sequence converges to f in measure.

Proof. (1) is false since we can take fn := n1/3 χ[0,1/n] then fn → 0 everywhere except for x = 0. But
ˆ 1 ˆ 1/n
fn3 dx = ndx = 1
0 0
so fn does not converge along any subsequence to 0.

(2) is true. Indeed, as fn → f in L3 this means we can find a subsequence fnk such that we have
||fnk − f ||L3 ([0,1]) ≤ 2−k
for any k ∈ N. Now define
N
X
gN (x) := |f − fnk |3
k=0
then notice that due to the monotone convergence theorem that we have
ˆ 1 ˆ 1 ˆ 1X
∞
lim |gN (x)|dx = lim |gN (x)| = |f − fnk |3
N →∞ 0 0 N →∞ 0 k=0

and for each N we have the uniform upper bound

ˆ 1 ˆ N
1X
|gN (x)| ≤ 2−k ≤ 2
0 0 k=0

so it follows that
∞
X
L3 ([0, 1]) 3 g(x) := |f − fnk |3
k=0
so it must be finite a.e., which implies the sum converges a.e., so fnk − f → 0 as k → ∞ a.e. along this
subsequence.

(3) is false. Indeed, let m denote the Lebesgue Measure, then if ε > 0 we have for fn := n1/3 χ[0,1/n]
m({x : |fn (x)| > ε}) ≤ m({x : |fn (x)| > 0}) ≤ 1/n
since these functions are supported on [0, 1/n]. So we have fn converges to 0 in measure, but arguing as
in (1) there is no subsequence of fn that converges to 0 in L3 .

(4) is true. Notice if ε > 0 then

ˆ 1 ˆ
|fn (x) − f (x)|dx ≥ |fn (x) − f (x)|dx ≥ εm({x : |fn (x) − f (x)| > ε})
0 {x:|fn (x)−f (x)|>ε}

so we have ˆ
1 1
|fn (x) − f (x)| ≥ m({x : |fn (x) − f (x)| > ε})
ε 0
But by Holder’s Inequality we have
ˆ 1
|fn (x) − f (x)|dx ≤ ||fn (x) − f (x)||L3 ([0,1])
0
33

so by choosing n sufficiently large we have for δ > 0 that ||fn (x) − f (x)||L3 ([0,1]) ≤ δ so we have
δ
≥ m({x : |fn (x) − f (x)| > ε})
ε
and as ε is fixed we can let δ → 0 to conclude limn→∞ m({x : |fn (x) − f (x)| > ε}) = 0.

Problem 2. Let X and Y be topological spaces and let X × Y the Cartesian product endowed
with the product topology. B(X) denotes the Borel Sets in X and similarly, B(Y ) and B(X × Y ).
(1) Suppose f : X → Y is continuous. Prove that E ∈ B(Y ) implies f −1 (E) ∈ B(X).
(2) Suppose A ∈ B(X) and E ∈ B(Y ). Show that A × E ∈ B(X × Y ).

Proof. Let A := {E ∈ B(Y ) : f −1 (E) ∈ B(X)}. Then we claim that A is a σ-algebra that contains
the open subsets of Y . Indeed, it is clear that Y ∈ A since f −1S(Y ) = X and similarly forSthe empty
∞ ∞
set.
S∞ Now if {Xi }∞
i=1 ∈ A then as B(Y ) is a σ-algebra we have i=1 Xi ∈ B(Y ) with f
−1
( i=1 Xi ) =
−1
i=1 f (Xi ) ∈ B(X) since each f −1 (Xi ) is borel and borel sets are closed under countable unions.
Finally if E ∈ A then E c ∈ B(Y ) with
f −1 (E c ) = (f −1 (E))c ∈ B(X)
since borel sets are closed under complements. Therefore, A is a σ−algebra. And it contains the open
sets since f is continuous so if E ⊂ Y is open then f −1 (E) is open i.e. f −1 (E) ∈ B(X). So if we denote
the collection of open sets in Y as G then we have
G ⊂ A ⇒ σ(G) = B(Y ) ⊂ A
since the σ-algebra generated by the open sets is the borel sets and A is a σ-algebra. Therefore, we have
proven (1).

For (2) we know the canonical projection map πX : X × Y → X and πY : X × Y → Y are continuous
since we are in the product topology. Then if A ∈ B(X) and E ∈ B(Y ) then using part 1) gives
−1
B(X × Y ) 3 πX (A) = A × Y
B(X × Y ) 3 πY−1 (E) = X × E
so we deduce that
B(X × Y ) 3 πY−1 (E) ∩ πX
−1
(A) = A × E
as desired.

Problem 3. Given f : [0, 1] → R belonging to L1 (dx) and n ∈ {1, 2, 3, ..., } define

ˆ (k+1)/n
fn (x) := n f (y)dy for x ∈ [k/n, (k + 1)/n) and k = 0, .., n − 1
k/n
1
Prove fn → f in L (dx)

Proof. We first recall that compactly supported continuous functions are dense in L1 ([0, 1], dx). In
particular, if ε > 0 then there exists an g ∈ Cc ([0, 1]) such that ||f − g||L1 ([0,1]) ≤ ε. We will first prove
the theorem is true for this dense subclass then extend it to f ∈ L1 (dx). Indeed, observe that
ˆ 1 X ˆ (k+1)/n
n−1 ˆ (k+1)/n
||g − gn ||L1 = |g(x) − gn (x)|dx = |g(x) − n g(y)dy|dx
0 k=0 k/n k/n

X ˆ (k+1)/n
n−1 ˆ (k+1)/n X ˆ (k+1)/n
n−1 ˆ (k+1)/n
= n g(x) − g(y)dy dx ≤ n |g(x) − g(y)dy| dx
k=0 k/n k/n k=0 k/n k/n
34 RAYMOND CHU

So by uniform continuity if n is sufficiently large then we have |g(x) − g(y)| < ε for x, y ∈ [k/n, (k + 1)/n]
so if n is large enough
≤ n(1/n)(1/n)nε = ε
so we have gn → g in L1 (dx).

Now we also have by the traingle inequality

||f − fn ||L1 (dx) ≤ ||f − g||L1 (dx) + ||gn − fn ||L1 (dx) + ||gn − g||L1 (dx)
so if n is large
≤ 2ε + ||gn − fn ||L1 (dx)
Now we compute
ˆ 1 Xˆ
n−1 (k+1)/n X ˆ (k+1)/n
n−1 ˆ (k+1)/n
|gn − fn |dx = |gn (x) − fn (x)|dx = n|g(y) − f (y)|dydx
0 k=0 k/n k=0 k/n k/n

X ˆ (k+1)/n
n−1 ˆ 1
= |g(y) − f (y)|dy = |g(y) − f (y)|dy
k=0 k/n 0
so
≤ 3ε
so we are done .

´
Problem 4. Let S = {f ∈ L1 (R3 ) : f dx = 0}
(1) Show that S is closed in the L1 topology
(2) Show that S ∩ L2 (R3 ) is a dense subset of L2 (R3 )

Proof. For (1) observe that if fn ∈ S such that fn → f ∈ L1 (R3 ) where the convergence is in the L1
sense, then
ˆ ˆ ˆ
f = f − fn ≤ |f − fn |dx → 0 as n → ∞
R3 R3 R3
which gives us f ∈ S.

For (2) it suffices to show the problem for f ∈ ´Cc (R3 ) since this is a dense subclass of L2 (R3 ). Then
say f is supported on BR (0) then define I := R3 f (x)dx then for ε > 0 choose M (ε) > R such that
m(B(0, M ) \ B(0, R)) = 1/ε i.e. 4/3π(M (ε)3 − R3 ) = 1/ε
(
f (x) for x ∈ B(0, R)
fε (x) :=
−εI for x ∈ B(0, M (ε)) \ B(0, R)
´
then fε (x) = 0 and
ˆ ˆ
2 2 2
|f (x) − fε (x)| dx = ε I 1dx = εI 2 → 0
Rd B(0,M (ε))\B(0,R)

as desired.

Problem 5. State and prove the Risez Representation Theorem for linear functionals (on a
separable) Hilbert Space.
35

Proof. Let H be our Hilbert space with inner product (·, ·) and Λ ∈ H ∗ i.e. Λ is a continuous linear
functional. Then there is a unique y ∈ H such that
Λ(x) = (x, y)
Indeed, notice as Λ is continuous we must have Ker(Λ) be a closed subset of H, so as it is a closed
subspace we have the decomposition
M
H = Ker(Λ) (Ker(Λ))⊥
So as long as Λ is not the trivial functional i.e. Λ(x) = 0 from which the theorem follows trivially with
y = 0 there must exist an z ∈ (Ker(Λ))⊥ . Now notice that for any x ∈ H we have
zΛx − xΛz ∈ Ker(Λ)
so we have
Λz
(zΛx − xΛz, z) = 0 ⇒ Λx||z||2 − (xΛz, z) = 0 ⇒ Λx = (x, z)
||z||2
Λz
So by defining y := ||z||2 z then we have
Λx = (x, y)
for any y ∈ H. Uniqueness follows from if
(x, y) = (x, z) for all x ∈ H ⇒ (x, y − z) = 0 for all x ∈ H
which implies y = z.

Problem 6. Suppose f ∈ L2 (R) and that the Fourier transform obeys fˆ(ξ) > 0 for almost every
ξ. Show that the set of finite linear combinations of translates of f is dense in the Hilbert Space
L2 (R).

Proof. Define S L
as the closure of the set of finite linear combinations of translates of f then we know
that L2 (R) = S S ⊥ . So it suffices to show S ⊥ = ∅. Indeed, observe that if g ∈ S ⊥ then by Plancherel,
we have that ˆ
0 = (f (x − a), g) = (e−ita fˆ(t), ĝ(t)) = e−ita fˆ(t)ĝ(t)dt
R
and as fˆ, ĝ ∈ L2 we know fˆĝ ∈ L1 , so its Fourier Transform is well defined and we have
0 = F(fˆĝ)(a)
So as F(fˆĝ) = 0 ∈ L1 (R) and fˆĝ ∈ L1 we can apply the Fourier Inverse Formula to get
fˆĝ(a) = 0
for a.e. a. This implies from fˆ ≥ 0 that ĝ(a) = 0 a.e.

Problem 7. Let {un (z)} be a sequence of real-valued harmonic functions on D that obey
u1 (z) ≥ u2 (z) ≥ u3 (z) ≥ ... ≥ 0 for all z ∈ D
Prove that z 7→ inf n un (z) is a harmonic function on D.

Proof. Notice that if n ≥ m then we have un − um ≥ 0 is a harmonic function and Harnack’s inequality
implies
r + |z|
0 ≤ un (z) − um (z) ≤ (un (0) − um (0))
r − |z|
where z ∈ D(r, 0) ⊂ D(1, 0) and we know that {un (0)} is a cauchy sequence since it converges to
u(0) := inf n un (0). Therefore, {un } converges locally uniformly (i.e. on every compact subset of D). By
the mean value property equivalence, we see that the limiting function is harmonic. And as the sequence
36 RAYMOND CHU

is decreasing it must pointwise converge to u(z) := inf n un (z), which implies u(z) is harmonic on every
compact subset, so it is harmonic on D.

Problem 8. Let Ω := {x + iy : x > 0, y > 0, xy < 1}. Give an example of an unbounded

harmonic function on Ω that continuously extends to 0 on ∂Ω.

Proof. Notice that by squaring Ω that the domain becomes U := {x + iy : x ∈ R, 0 < y < 2}. Define
g(z) := Im(eπz ) then g(∂Ω) = 0 since the function becomes real valued. Also g(x + i/2) = eπx which is
2
unbounded, so the function Im(eπz ) works.

Problem 9. Prove Jordan’s Lemma: If f (z) : C → C is meromorphic, R > 0, and k > 0, then
ˆ
100
f (z)eikz dz ≤ sup |f (z)|
Γ k z∈Γ
where Γ is the quarter-circle z = Reiθ with 0 ≤ θ ≤ π/2.

Proof. Note that

ˆ ˆ π/2
iθ
f (z)eikz dz = f (Reiθ )eikRe iReiθ
Γ θ=0
so we have from Holder ˆ ˆ π/2
f (z)eikz dz ≤ sup |f (z)| Re−kR sin(θ) dθ
Γ z∈Γ θ=0
Now using that sin(x) > x/2 for x ∈ [0, π/2] we have
ˆ π/2 ˆ π/2 ˆ ˆ
−kR sin(θ) −kRθ/2 2 kRπ/4 −θ 2 ∞ −k 2
Re ≤ Re dθ = e dθ ≤ e =
θ=0 θ=0 k θ=0 k 0 k
so we conclude that ˆ
2 100
f (z)eikz dz ≤ sup |f (z)| ≤ sup |f (z)|
Γ z∈Γ k k z∈Γ

Problem 10. Let us define the Γ function via

ˆ ∞
Γ(z) := tz−1 e−t dt
0
at least when the integral is absolutely converges. Show that this function extends to a mero-
morphic function in the whole complex plane. You cannot use any particular properties of the Γ
function unless you derive it from this definition.

Proof. Let us show first that Γ(z) is holomorphic in the region U := {z : Re(z) > 0}. Indeed, observe
first ˆ ∞ ˆ ∞
z−1 −t
|Γ(z)| ≤ |t |e dt ≤ |t|Re(z)−1 e−t
0 0
so Γ(z) is absolutely convergent in the region Re(z) > 0 since the |t|1−ε for ε > 0 is integrable near the
origin and e−t gives enough decay factor at ∞. Also by the dominated convergence theorem, it follows
that Γ(z) is continuous in U .

In particular, continuity implies that if R ⊂ U is a rectangle then Γ ∈ L1 (R) so we have

ˆ ˆ ˆ ∞ ˆ ∞ˆ
Γ(z)dz = tz−1 e−t dt = tz−1 e−t dt = 0
R R 0 0 R
37

where in the last step we used the integrand is holomorphic in z. Therefore, Morrera’s theorem implies
Γ(z) is holomorphic on U .

Now we claim that Γ(z) has a simple pole at z = 0. Indeed, observe by integration by parts that
for z with Re(z) ≥ 1 that we have from integration by parts
ˆ ∞ ˆ
1 ∞ z −t
Γ(z) = tz−1 e−t dt = t e dt
0 z 0
´∞
so by analytic continuation Γ(z) = z1 0 tz e−t dt on U . Therefore, it follows that Γ(z) extends to a
meromorphic function on {z : Re(z) > −1} with a simple pole at z = 0. Now we can also keep iterating
this process infinitely many times to see Γ(z) extends to a meromorphic function on C with simple poles
at the negative integers, or we can use the identity
zΓ(z) = Γ(z + 1)
to define Γ(z) := z1 Γ(z + 1) for z ∈ {z : Re(z) > −2} then on {z : Re(z) > −3} and inductively to define
it on all of C. So it suffices to justify this formula. Indeed, observe on U we have
ˆ ∞ ˆ ∞
zΓ(z) = z tz−1 e−t dt = tz e−t = Γ(z + 1)
0 0

by integration by parts; therefore, by analytic continuation this property holds over {z : Re(z) > 0} then
we can use this as the definition of Γ(z) for {z : Re(z) ≤ 0}

Problem 11. Let P (z) be a polynomial. Show that there is an integer n and a second polynomial
Q(z) so that
P (z)Q(z) = z n |P (z)|2 whenever |z| = 1

Proof. Say P has degree m then observe that when P (z) 6= 0

z n |P (z)|/P (z) = z n P (z)
so when |z| = 1 and P (z) 6= 0 we get that this is equal to
m
X m
X
inθ −ikθ
e aj e = aj ei(n−k)θ
k=0 k=0

so taking n = m we get
m
|P (z)| X
zn = aj ei((m−k)θ) := Q(eiθ )
P (z)
k=0
Pm m−k
Observe z = k=0 aj z is a polynomial and when P (z) 6= 0 we have
P (z)Q(z) = z m |P (z)|2
and when P (z) = 0 the equality is trivial.

Problem 12. Show that the only entire function f (z) obeying both
n
|f 0 (z)| ≤ exp(|z|) and f ( p ) = 0 for all n ∈ Z
1 + |n|
is the zero function.

Proof. We first claim that f is an entire function of order 1. Indeed, by the fundamnetal theorem of
calculus we have ˆ
f (w) = f 0 (z)dz
γ
38 RAYMOND CHU

where γ = {tw : t ∈ [0, 1]} so

ˆ 1 ˆ 1 ˆ |w|
f (w) = wf 0 (tw)dt ⇒ |f (w)| ≤ |w| exp(|tw|)dw = exp(t)dt = exp(|w|) − 1 ≤ exp(|w|)
t=0 0 0
Therefore, we f (z) is an entire function of order 1. Therefore, by Jensen’s Formula unlesspf ≡ 0 then we
√
must the number of zeros in a circle of radius R must be of order C 0 R. However, as n/ 1 + |n| ∼ n
this implies there should roughly N 2 zeros in a circle of radius N ∈ N for large N . Therefore, f is the
zero function.
39

6. Fall 2012

Problem 1. Let 1 < p < ∞ and let fn : R3 → R be a sequence of functions such that
lim sup ||fn ||Lp < ∞. Show that if fn converges almost everywhere, then fn converges weakly
in Lp .

Proof. Note that as 1 < p < ∞ we have that Lp is reflexive, so weak-* convergence is the same as weak
convergence, so in particular as lim sup ||fn ||Lp < ∞ this is a bounded sequence in Lp so by Banach-
Alagou, there exists a sub-sequence fnk and a f such that fnk * f but as fnk → g a.e.

We then claim f = g. Indeed, fix any compact set K ⊂ R3 then by egorov for any ε > 0 there is a
compact set E ⊂ K such that m(K \ E) < ε and fn → g uniformly on E. Then for any ψ ∈ Lq where q
is the dual conjuagte of p we have
ˆ ˆ ˆ
f nk ψ = fnk ψ + f nk ψ
K E K\E
and observe that by uniform convergence we have for k sufficiently large we have
ˆ
|gψ − fnk ψ| ≤ ε
E
1
and as gψ is in L (K). We also observe
ˆ
|fnk ψ| ≤ ||fnk ||Lp ||ψχK\E ||Lq = o(ε)
K\E
and ˆ
|gψ| = o(ε)
K\E
so it follows that ˆ ˆ
lim fnk ψ = fψ
k→∞ K K
then this implies by DCT that ˆ ˆ
lim fnk ψ = fψ
k→∞ R3 K
which implies f = g by uniqueness of weak limits. So this implies every sub-sequence has a further
sub-sequence that converges to the same limit g; therefore, the whole sequence converges to g.

Problem 2. Suppose dµ is a probability measure on the unit circle in the complex plane such
that ˆ
lim z n dµ(z) = 0
n→∞ S1
1
For f ∈ L (dµ) show that ˆ
lim z n f (z)dµ(z) = 0
n→∞ S1

PM
Proof. By Stone Weiestrass we know that trigonometirc polynomials i.e. P (z) = n=−N an z n are dense
in S 1 . So one has for any fixed trigonometric polynomial that
ˆ M ˆ
X
z n P (z)dµ(z) = aj z j+n dµ(z) → 0
S1 j=−N S1

Therefore, for f ∈ L1 (dµ) and ε > 0 we can find a trignometric polynomial such that ||P (z)−f (z)||L∞ (S 1 ) <
ε then
ˆ ˆ ˆ ˆ
f (z)z n ≤ |f (z) − P (z)|dµ(z) + P (z)z n dµ(z) ≤ ε + P (z)z n dµ(z)
S1 S1 S1 S1
40 RAYMOND CHU

so this goes to 0 as n → ∞.

Problem 3. Let H be a Hilbert Space and let E be a closed convex subset of H. Prove that
there exists a unique element x ∈ E such that
||x|| = inf |y||
y∈E

Proof. Let α := inf y∈E ||y|| and let yn ∈ E be a minimizing sequence i.e. ||yn || → α. Notice by convexity
that yn +y
2
m
∈ E so by the Parallelogram Law
yn + ym 2 yn − ym 2 1 1
|| || + || || = ||yn ||2 + ||ym ||2
2 2 2 2
and we have
yn + ym 2
α ≤ || || ≤ ||yn ||/2 + ||ym ||/2 → α
2
from which we deduce that ||yn − ym || → 0 i.e. {yn } is a Cauchy sequence, so as we are on a Hilbert
space and E is closed we know there exists a y ∈ E such that yn → y. Now as the norm is continuous,
we know that ||y|| = α i.e. y is our minimizer.

Uniqueness arises since if α = ||y||, ||z|| for y, z ∈ E then we know

y+z
α ≤ || || ≤ ||y||/2 + ||z||/2 = α
2
so again by the Parallelogram law we deduce that ||y − z|| = 0 i.e. y = z.

Problem 4. Fix f ∈ C(T) where T = R/2πZ. Let sn denote the n-th partial sum of the Fourier
Series of f . Prove that
||sn ||L∞ (T)
lim =0
n→∞ log(n)

Proof. Recall that we have

N ˆ π ˆ N
1 π

X 1 X
sN = f (t)e−int dt einx = f (t) ein(x−t) dt
π −π π −π
n=−N n=−N

= (f ∗ DN )(t)
1
PN int
where DN (t) = π n=−N e and notice that its a geometric sum so we have
e−iN t − ei(N +1)t e−it(N +1/2) − ei(N +1/2)t sin((N + 1/2)t)
πDN (t) = = =
1−e it e−i/2t − ei/2t sin(t/2)
Therefore, we have the formula
ˆ
1 π sin((N + 1/2)t)
sN = f (x − t)
π −π sin(t/2)
Therefore, we have from Holder’s Inequality that
ˆ π
1 sin((N + 1/2)t
||sN ||L∞ ≤ ||f ||L∞ (T)
π −π sin(t/2)
and the inner integral can be approximated by using concavity of log to deduce on x ∈ [0, π] that
sin(t/2 + 0/2) ≥ t sin(1/2) := αt
and oddness of sin gives us the bound | sin(t/2)| ≥ α|t| for t ∈ [−π, π], so we conclude
ˆ 0 ˆ π ˆ
sin((N + 1/2)t | sin((N + 1/2)t)| 1 (N +1/2)π | sin(t)|
≤ dt =
−π sin(t/2) 0 αt α 0 t
41

N ˆ (k+1)π ˆ π N
1X | sin(t)| 1 | sin(t)| X 1 1
≤ ≤ +
α kπ t α 0 t α kπ
k=0 k=1
so we have since sin(t)/t is continuous on [0, π] that
N
!
||sN ||L∞ C 1X 1
≤ ||f ||L∞ + ≤ K||f ||L∞
log(N ) log(N ) α log(N )kπ
k=1

since the Harmonic Series grows like log(N ). Therefore, the family of linear operators
sn (f )
Λn (f ) :=
log(N )
is uniformly bounded.

Now by Stone Weiestrass we can find a trigonoemetirc polynomial P such that if ε > 0 then ||f −
P ||L∞ (T ) ≤ ε. Then notice that
ˆ π
P (x)einx = 0
−π
inx
for all but finitely many n since {e } are orthogonal. Therefore,
|Λn (f )| ≤ |Λn (f ) − Λn (P )| + |Λn (P )| ≤ K||f − P ||L∞ + |Λn (P )| ≤ ε + |Λn (P )|
and we know |Λn (P )| → 0 since we have only a uniform amount of finitely many terms are non-zero and
they are being scaled by 1/ log(N ). Therefore,
lim |Λn (f )| = 0
n→∞

Problem 5. Let fn : R3 → R be a sequence of functions such that supn ||fn ||L2 < ∞. Show that
if fn converges almost everywhere to a function f : R3 → R, then
ˆ
|fn |2 − |fn − f |2 − |f |2 dx → 0
R3

Proof. Notice that by expanding we have

|fn |2 − |fn − f |2 − |f |2 = fn2 − (fn2 − 2fn f + f 2 ) − f 2 = 2fn f − 2f 2 = 2f (fn − f )
Therefore, we have ˆ ˆ
2 2 2
|fn | − |fn − f | − |f | dx = |2f (f − fn )|
R3 R3
And notice by Fatou’s Lemma that
ˆ ˆ
|f |2 ≤ lim inf |fn |2 dx ≤ C
R3 n→∞ R3
´
Therefore, f ∈ L2 (R3 ), so there exists a compact set K such that on R3 \ K we have R3 \K |f |2 ≤ ε.
By Egorov’s Theoremm there exists a compact subset K1 ⊂ K with m(K \ K1 ) ≤ ε such that fn → f
uniformly on K1 . In particular,
ˆ ˆ ˆ ˆ
|2f (f − fn )| = |2f (f − fn )| + |2f (f − fn )| + |2f (f − fn )| := (I) + (II) + (III)
R3 K1 K\K1 R3 \K

Notice that Cauchy-Schwarz gives

(I) ≤ 2||f ||L2 (R3 ) ||f − fn ||L2 (K1 ) ≤ Kε
due to uniform convergence and our prior estimates. Also
(II) ≤ 2||f ||L2 (K\K1 ) ||f − fn ||L2 (R) = o(ε)
42 RAYMOND CHU

since f is uniformly integrable as its in L2 and the second term is bounded by a constant. Also
(III) ≤ 2||f ||L2 (R3 \K) ||f − fn ||L2 (R) ≤ K̃ε
Therefore, we have ˆ
|2f (f − fn )| → 0
R3
as desired

Problem 6. Let f ∈ L1 (R) and Mf denote its maximal function, that is,
ˆ r
1
(Mf )(x) = sup |f (x − y)|dy
0<r<∞ 2r −r
By the Hardy-Littlewood maximal function theorem,
|{x ∈ R : (Mf )(x) > λ}| < 3λ−1 ||f ||L1
Using this show that
ˆ r
1
lim sup |f (y) − f (x)|dy = 0 for a.e. x ∈ R
r→0 2r −r

Proof. First we recall that Cc (R) ⊂ L1 (R) is a dense subclass and is g ∈ Cc (R) then g is uniformly
continuous. So if ε > 0 then there exists a δ > 0 such that if |x − y| < δ implies |g(x) − g(y)| ≤ ε. Take
r < δ then ˆ r ˆ r
1 1
|g(y) − g(x)|dy ≤ ε=ε
2r −r 2r −r
so we have ˆ r
1
lim sup |g(y) − g(x)|dy = 0 for all g ∈ Cc (R)
r→0 2r −r
Now we will use the maximal inequality to extend this to f ∈ L1 (R). By density, there is a sequence
{gn } ⊂ Cc (R) such that ||f − g||L1 (R) < n1 . Therefore, we have
ˆ r ˆ r ˆ r ˆ r
1 1 1 1
|f (y) − f (x)|dy ≤ |f (y) − gn (y)|dy + |gn (y) − gn (x)|dy + |gn (x) − f (x)|dy
2r −r 2r −r 2r −r 2r −r
ˆ r
1 1
≤ |f (y) − gn (y)|dy + + |gn (x) − f (x)|
2r −r n
Since gn → f in L1 (R) there exists a sub sequence such that gn → f a.e., so by replacing an n in this sub-
sequence we can assume gn (x) → f (x) for a.e. x. Denote E as the set x such that along this subsequence
we have gn (x) → f (x). Then if x ∈ E and ε > 0 arbitrary then we have by taking n sufficiently large
ˆ r ˆ r
1 1
|f (y) − f (x)|dy ≤ δ + |f (y) − gn (y)|dy
2r −r 2r −r
ˆ r ˆ r
1 1
Rδ := {x ∈ E : lim sup |f (y) − f (x)|dy > 2δ} ⊂ {x ∈ E : |f (y) − gn (y)| > δ}
r→0 2r −r 2r −r
⊂ {x ∈ R : (M(f − gn ))(x) > δ}
so
|Rδ | ≤ 3δ −1 ||f − gn ||L1 ≤ 3δ −1 n−1 → 0 as n → ∞
Therefore, Rδ is a null set. But notice that
ˆ r
1
{x ∈ R : lim sup |f (y) − f (x)|dy > 2δ} ⊂ Rδ ∪ E c
r→0 2r −r

and E c is a null set, so it follows from

ˆ r ∞ ˆ r
1 [ 1
E := {x ∈ R : lim sup |f (y) − f (x)|dy > 0} = {x ∈ R : lim sup |f (y) − f (x)|dy > n−1 }
r→0 2r −r r→0 2r −r
n=1
43

that E is a null set. So this implies the problem statement.

Problem 7. Let f be a function holomorphic in C with f (0) = 0 and f (1) = 1 with f (D) ⊂ D.
Show that
(1) f 0 (1) ∈ R
(2) f 0 (1) ≥ 1

Proof. Assume for the sake of contradiction that f 0 (1) = a + ib where b 6= 0 and a, b ∈ R. Then by the
chain rule one has
f (1 + tv) − f (1) f (1 + tv) − 1
f 0 (1)v = lim = lim
t→0 t t→0 t
where we choose v such that 1 + tv ∈ D for small enough t. So one has
Re(f (1 + tv)) − 1
Re(f 0 (1)v) = lim ≤0
t→0 t
where we are using f (D) ⊂ D. But as f 0 (1) has an imaginary component this means we can rotate and
find a v (thanks to rotating) such that Re(f 0 (1)v) > 0. In particular, we can choose v = −ε − iδb where
0 < ε 1 and δ > 0 is a fixed constant chosen to ensure 1 + tv ∈ D for small enough t. And we have
f 0 (1)v = −aε + δ 2 b2 so if ε is sufficiently small compared to δ we have arrived at a contradiction.

Notice that for 0 < t < 1

1 − f (1 − t) 1 − Re(f (1 − t))
f 0 (1) = lim = lim
t→0 t t→0 t
and we see from Schwarz Lemma that
|f (1 − t)| ≤ 1 − t ⇒ −Re(f (1 − t)) ≥ t − 1
so we deduce that
f 0 (1) ≥ 1
as desired .

Problem 8. Let f : C → C be a non-constant holomorphic function such that every zero of f

has even multiplicity. Show that f has a holomorphic square root.

Proof. As the zeros of f (z) are isolated, we know there are only countably many of them. Enumerate
them as {zn }n∈N where zn 6= zm unless n = m and with multiplicity 2mn for mn ∈ N then define
Em (z) := exp(z + z 2 /2 + ... + z m /m)
Then either {zn } is finite or |zn | → ∞, so by the Weiestrass Factorization Theorem we can find an entire
function g(z) with zeros only at zn with mn multiplicity. Then we have f (z)/g 2 (z) is an entire function
with no zeros. Therefore, there is a entire function h(z) such that
f (z)/g 2 (z) = exp(h(z)) ⇒ f (z) = g 2 (z) exp(h(z)) = (g(x) exp(h(z)/2))2
so f has a holomorphic square root g(z) exp(h(z)/2).

Problem 9. Suppose f is analytic in the unit disk D and {xn } is a sequence of real numbers
satisfying 0 < xn+1 < xn < 1 for all n with limn→∞ xn = 0. Show that if f (x2n+1 ) = f (x2n ) for
all n ∈ N, then f is constant.
44 RAYMOND CHU

Proof. Assume by subtracting off by a constant that f (0) = 0. Now decompose f (z) = u(z) + iv(z)
where u and v are the real and imaginary parts of f . Now we restrict the domain of u and v to the real
(1)
axis. Therefore, as u(x2n+1 ) = u(x2n ) we can apply Rolle’s theorem to conclude there is an y2n such
(1) (1)
that x2n+1 ≤ y2n ≤ x2n for which u0 (x2n+1 ) = 0. This lets us find a sequence {y2n } that converges to
(1)
0 with u0 (y2n ) = 0 (we use prime to denote x derivatives since we are viewing u as a function on the
reals). So in particular continuity gives us u0 (0) = 0 so using f 0 = ux + ivx we get that f 0 (0) = 0 since
we can do an identical argument on v. Now we can again repeating the above argument of using Rolle’s
(2) (2)
we can find a decreasing sequence y2n for which u00 (y2n ) = 0 to get f 00 (0) = 0. We can keep iterating
(n)
this argument for all n to deduce that f (0) = 0 and f (0) = 0 for all n ∈ N, which means
A := {z : f (z) = 0, f (n) (z) = 0 for all n ∈ N}
is a non-empty closed and open subset of a connected subset D so it is the entire space i.e. f (z) ≡ 0 in
D.

Problem 10. Let {fn } be a sequence of holomorphic functions on D satisfying |fn (z)| ≤ 1 for
all z ∈ D and n ∈ N. Let A ⊂ D be the set of all z ∈ D for which limn→∞ fn (z) exists. Show that
if A has an accumulation point in D, then there exists a holomorphic function f on D such that
fn → f locally uniformly on D as n → ∞.

Proof. Fix a subsequence {fnk } and another subsequence {fmk }. By Montel’s theorem both of these
subsequence have a further subsequence that converges locally uniformly to some holomorphic functions
f and g respectively. As A has an accumulation point we deduce f = g on an accumulation point, so we
must have f ≡ g on D. Therefore, every subsequence has a further subsequence that converges locally
uniformly to f , which implies the entire sequence converges locally uniformly to f .

Problem 11. Find all holomorphic functions f : C → C satisfying f (z +1) = f (z) and f (z +i) =
e2π f (z).

Proof. Notice that f (z) := exp(−2πiz) satisfies f (z + 1) = f (z) and f (z + i) = e2π f (z). Let g be another
entire function satisfying the periodicity conditions, then h := g/f is an entire function since f never
vanishes.And h satisfies h(z+1) = h(z) and h(z+i) = h(z). In particular, let M := maxz∈[0,1]2 |h(z)| < ∞,
then from the periodicity condition, we see that this bounds h everywhere. So in particular, h(z) ≡ C
for some constant C by Liouville, so g = C exp(−2πiz) and this classifies all such functions.

Problem 12. Let M ∈ R and Ω ⊂ C be bounded open set, and u : Ω → R be a harmonic

function.
(1) Show that if
lim sup u(z) ≤ M
z→z0
for all z ∈ ∂Ω, then u(z) ≤ M for all z ∈ Ω
(2) Show that if u is bounded from above and there exists a finite set F ⊂ ∂Ω such that the
inequality in (1) is satisfied for all z0 ∈ ∂Ω \ F then the conclusion of (1) is still true.

Proof. If we fix an ε > 0 then due to the inequality, for any z0 ∈ ∂Ω there exists a r > 0 such that
on Br (z0 ) ∩ Ω we have u(z) ≤ M + ε. By compactness we can find a finite subcover of these balls say
SN
{Bri (zi )}N
i=1 then on Ωε := Ω \ i=1 Bri (zi ) then we know that we have u(z) ≤ M + ε on ∂Ωε , so the
max principle implies u(z) ≤ M + ε on Ωε . But on the union of these balls we also have this inequality,
so we deduce that u(z) ≤ M + ε on Ω and letting ε → 0 concludes (1).
45

Part (ii) will follow from the standard ε log trick. Indeed, let d := diam(Ω) and enumerate these finite
z−z0 z−z1 z−zN
points as {zi }N
i=1 and define h(z) := − log | d | − − log | d | − ... − − log | d | then we have h(z)
is harmonic on Ω since its locally the real part of a holomorphic function with h(zi ) := ∞. Then we
consider
u(z) − εh(z)
and thanks to the bounded above condition
lim sup u(z) − εh(z) = −∞ < M
z→zi

and if δ > 0 then for any w ∈

/ F we have
lim sup u(z) − εh(z) ≤ M + δ
z→w
since h(z) ≥ 0. This lets us conclude with part (1) that
u(z) − εh(z) ≤ M + δ on Ω
and letting ε → 0 since h(z) is finite on Ω we deduce that
u(z) ≤ M + δ
and finally letting δ → 0 gives the claim

46 RAYMOND CHU

7. Spring 2013

Problem 1. Suppose f : R → R is bounded, Lebesgue Measurable, and

ˆ 1
|f (x + h) − f (x)|
lim dx = 0
h→0 0 h
Show that f is a.e. constant on [0, 1].

´x
Proof. Observe that for F (x) := f (y)dy that we have
0
ˆ
F (x + h) − F (x) 1 x+h
= f (y)dy → f (x) a.e.
h h x
thanks to Lebesgue’s Differentiation Theorem since f bounded implies f ∈ L1l oc(R). We also have for
x<y
ˆ ˆ
F (x + h) − F (y + h) + F (y) − F (x) 1 y+h 1 y
= F (z)dz − F (z)dz
h h x+h h x
ˆ
1 y
= F (z + h) − F (z)dz
h x
so
ˆ ˆ
F (x + h) − F (y + h) + F (y) − F (x) 1 y 1 1
≤ |F (z + h) − F (z)|dz ≤ |F (z + h) − F (z)|dz
h h x h 0
So we have
F (x + h) − F (y + h) + F (y) − F (x)
lim =0
h→0 h
but by Lebesgue’s Differentiation Theorem we know that for a.e. x and u
F (x + h) − F (y + h) + F (y) − F (x)
lim = f (x) − f (y)
h→0 h
so we have
f (x) = f (y) a.e.
so f is constant a.e. on [0, 1].

Problem 2. Consider the Hilbert Space `2 (Z). Show that the Borel σ-algebra N on `2 (Z) asso-
ciated to the norm topology agrees with the Borel σ-algebra W on `2 (Z) associated to the weak
topology.

Proof. We first recall that the weak topology is the coarsest topology for which linear functionals are
continuous so we have every open subset of the weak topology of `2 (Z) is contained in the norm topology
of `2 (Z). This implies W ⊂ N , For the reverse direction recall that `2 (Z) is separable for instance take
finite rational linear combinations of {ei } where ei is 0 everywhere except for a 1 on the ith coordinate so
every open set is aScountable union of balls. So it suffices to prove if x ∈ `2 (Z) then Br (x) ∈ W. Observe
that y ∈ Br (x) = n∈N {y : |y| ≤ r − 1/n}. And notice
||x − y||2 ≤ r2 ⇐⇒ ||x||2 − 2Re(x, y) + ||y||2 ≤ r2
and recall
||y||2 = sup (y, z)
||z||=1
2
so y ∈ Br (x) iff for all z ∈ ` (Z) with ||z|| = 1 such that
||x||2 − 2Re(x, y) + (y, z) ≤ r2
47

As `2 (Z) is separable, there is a countable dense subset which we label as vn and we can assume 0 6= vn
for any n. Then by continuity of the inner product we have for all ||z|| = 1
vn
||x||2 − 2Re(x, y) + (y, z) ≤ r2 ⇐⇒ ||x||2 − 2Re(x, y) + (y, ) ≤ r2 for all n
||vn ||
So for each n ∈ N define `n (y) := ||x||2 − 2Re(x, y) + (y, ||vvnn || ). Then we have for any n ∈ N
[ [ \
Br (x) = {y : |y| ≤ r − 1/n} = `−1
m ([0, r − 1/n]) ∈ W
n∈N n∈N m∈N

where we used `−1

m ([0, r − 1/n]) ∈ W since linear functionals generate the topology. (Note implicitly the
sum over n ∈ N is only taken for r − 1/n ≥ 0). Therefore, W = N .

Problem 3. Given f : R2 → R, continuous, we define

ˆ π
1
[Ar f ](x, y) := f (x + r cos(θ), y + r sin(θ))dθ and [M f ](x, y) := sup [Ar f ](x, y)
2π −π 0<r<1
By a theorem of Bourgain, there is an absolute constant C so that
||M f ||L3 (R2 ) ≤ C||f ||L3 (R2 ) for all f ∈ Cc (R2 )
Use this to show the following: If K ⊂ R2 is compact, then [Ar χK ](x, y) → 1 as r → 0 at almost
every point (x, y) ∈ K (w.r.t to the Lebesgue measure).

Proof. We first extend the inequality to f = χK − ψε where ψε where will be an approximation of χK .

Indeed for 0 < ε ≤ 1 define
Kε := {x ∈ R2 : d(x, K) ≤ ε}
then Kε is compact. Then by Uroshyn’s lemma there is a continuous function ψε such that ψε ≡ 1 on
Kε/2 and ψε ≡ 0 on Kεc . Then we have for any 0 < ε ≤ 1 that ψε/2 ≤ ψε ∈ Cc (R2 ) so we have from
construction that χK − ψε ≤ ψε − ψε/2 , χK ≤ ψε , and ψδ ∈ Cc (R2 ) so we have
[Ar (χK − ψε )] ≤ [Ar (ψε − ψε/2 )] ⇒ [M (χK − ψε )] ≤ [M (ψε − ψε/2 )]

⇒ ||[M (χK − ψε )]||L3 ≤ ||[M (ψε − ψε/2 )]||L3 ≤ C||ψε − ψε/2 ||L3
then using that ψε ≤ χK1 ∈ L3 (R2 ) since it is a compact set, so we have from the dominated convergence
theorem that
||ψε − ψε/2 ||L3 → 0 as ε → 0
since by construction ψε → χK pointwise. Therefore, for any ε > 0 by Chebyshev inequality and our
above inequality we have
C
m({x : |M f (χK − ψε )| > α}) ≤ ||ψε − ψε/2 ||L3
α3
Now one repeats the proof of Lebesgue Differentiation Theorem to conclude the problem using the
approximation scheme ψε ∈ Cc (R2 ) since they converge in L3 and L1 to χK by DCT.

Problem 4. Let K be a non-empty compact subset of R3 . For any Borel probability measure µ
on K, define the Newtonian energy I(u) ∈ (0, +∞] by
ˆ ˆ
1
I(µ) := dµ(x)dµ(y)
K K |x − y|
and let RK be the infimum of I(µ) over all Borel probability measures µ on K. Show that there
exists a Borel probability measure µ such that I(µ) = RK .
48 RAYMOND CHU

Proof. Take a minimizing sequence, that is a sequence {µn } of borel probability measures such that
I(µn ) → RK
which exists thanks to the definition of inf. So now as this is a sequence of Borel probability measures
∗
that there exists a subsequence and a borel measure µ such that µn * µ i.e. for all f ∈ C(K) we have
ˆ ˆ
f dµn → f dµ
K K
note that as 1 is continuous we immediately obtain µ(K) = 1 so it is a probability measure on K. Now we
∗
claim that µn ⊗ µn * µ ⊗ µ. Indeed, by Stone Weiestrass we know that continuous functions of the form
f (x)g(y) are dense in C(K × K), from which we get the desired claim by a standard density argument.
But notice that 1/|x − y| is not continuous, but it is lower semi-continuous, so we can approximate it
from above by continuous functions to get that the functional is lower semi continuous with respect to
weak convergence
I(µ) = RK

Problem 5. Define the Hilbert Space

ˆ
H := {u : D → R : u is harmonic and |f (x, y)|2 dxdy < ∞}
´
D
with inner product (f, g) := D f gdxdy.
(1) Show that f 7→ ∂f
∂x (0, 0) is a bounded linear operator on D.
(2) Compute the norm of this operator.

Proof. Note by the linearity of ∆ that we have ∂x f is another harmonic function and by the mean value
theorem we have that for 0 < r < 1
ˆ ˆ
1 1 x
∂x f (0, 0) = 2 ∂x f dS = 2 f (x, y) dS
πr Br (0) πr ∂Br (0) r
where the last inequality is due to Green’s theorem since the first component of the normal if x/r = cos(θ).
We compute to see
ˆ 2π
1
= cos(θ)f (r cos(θ), r sin(θ))dθ
πr θ=0
Therefore, we have from Cauchy-Schwarz that
ˆ 2π
2 1 2
|∂x f (0, 0)| ≤ 2 |f (r cos(θ), r sin(θ))| dθ
πr θ=0

so multiplying both sides by πr3 gives

ˆ 2π
πr3 |∂x f (0, 0)|2 ≤ r|f (r cos(θ), r sin(θ))|2 dθ
θ=0
so integrating again in r from r = 0 to ρ < 1
ˆ ˆ
4 4
|∂x f (0, 0)|2 ≤ |f (x, y)|dxdy ≤ |f (x, y)|dxdy
πρ4 Bρ (0) πρ4 B1 (0)
for all 0 < ρ < 1 so we can take ρ → 1 to yield
ˆ
4
|∂x f (0, 0)|2 ≤ |f (x, y)|dxdy
π B1 (0)

i.e.
2
|∂x f (0, 0)| ≤ √ ||f ||L2 (B1 (0)
π
so it is a continuous linear operator.
49

For part (2) taking f (x, y) = x gives

ˆ ˆ 2π ˆ 1
π
x2 dxdy = r3 cos2 (θ)drdθ =
B1 (0) θ=0 r=0 4
so
2
√ ||f ||L2 (B1 (0)) = 1 = |∂x f (0, 0)|
π
√
so the sharp constant is 2/ π.

Problem 6. Let ˆ
X := {ξ 7→ eixξ f (x)dx|f (x) ∈ L1 (R)}
R
(1) X is a subset of C0 (R)
(2) X is a dense subset of C0 (R)
(3) X 6= C0 (R)

Proof. For (1) observe that if f (x) ∈ L1 (R) then

ˆ ˆ ˆ
π
g(ξ) := eixξ f (x)dx = − eiξ(x+ ξ ) f (x)dx = − eiξx f (x − π/ξ)
R R R
Hence, one has ˆ
1
g(ξ) = eiξx (f (x) − f (x − π/ξ))
2 R
so we have
|g(ξ)| ≤ ||f (x) − f (x − π/ξ)||L1 (dx)
which converges to 0 as ξ → ∞ due to translation continuity of the Lebesgue integral. And g is continuous
due to the translation continuity of the Lebesgue integral.
For (2) observe that if f ∈ S where S is in the Schwarz class then its fourier transform fˆ is also in the
Schwarz class. In particular, the fourier inverison formula holds for f ∈ S. Then for any f ∈ S one has
ˆ
f (ξ) = eixξ fˆ(x)dx
R

since fˆ ∈ L1 . Therefore, X contains the Schwarz class, which contains Cc∞ (R). And the Cc∞ (R) is dense
in C0 (R). So X is a dense subset of C0 (R).

Note that if we define for f ∈ L1 (R)

ˆ
Lf := eixξ f (x)dx
R
then this is operator is injective since due to Fourier Theory we have that
F(Lf ) = f
so if Lf = 0 then we deduce f = 0, which by linearity of the operator implies it is injective. So now if Lf
was surjective to C0 (R) then L : L1 (R) → C0 (R) is bijective so by the open mapping theorem we have
that its inverse is continuous. Hence, we have constants C1 , C2 > 0 such that
||Lf ||L∞ ≤ C1 ||f ||L1 and ||L−1 g||L1 ≤ C2 ||g||L∞
for any f ∈ L1 (R) and g ∈ C0 (R). Taking g = Lf gives that
C2 ||f ||L1 ≤ ||Lf ||L∞ ≤ C1 ||f ||L1
Taking fn = (1/n)χ[0,n] (x) − (1/n)χ[−n,0] gives for ξ 6= 0
ˆ n ˆ 0
Lfn (ξ) = (1/n) eixξ dx − (1/n) eixξ dx
0 −n
50 RAYMOND CHU

1 inξ
(e + e−inξ − 2)
=
inξ
which converges to 0 as ξ → 0. Notice that
eiξ + e−iξ − 2
|| ||L∞ (R) ≤ C
ξ
since the expression is bounded near the origin since it converges to 0 and the expression decays to 0 as
ξ → ∞. Hence, we have
||Lfn ||L∞ ≤ C/n → 0 as n → ∞
but we have
||fn ||L1 = 1 for all n
which contradicts the contiuity of the inverse.

Problem 7. Let f : C → C be an entire function such that log |f | is absolutely integrable with
respect to the planar Lebesgue measure

Proof. As f is holomorphic we know log |f | is subharmonic i.e. we have the mean value inequality for
any z ∈ C
ˆ 2π
1
log |f |(z) ≤ log |f |(z + reiθ )dθ
2π θ=0
so multiplying by r and integrating in r from 0 to R gives
ˆ
1 1
log |f (z)| ≤ 2
log |f (z)|dλ(z) ≤ ||(log |f |)||L1 (R)
πR BR (z) πR2
so we have for any z by taking R → ∞ that
log |f (z)| ≤ 0 ⇒ |f (z)| ≤ 1
so by Liouville’s Theorem we have that f is constant.

Problem 8. Let A and B be real positive definite n × n symmetric matrix with the property
||BA−1 x|| ≤ ||x|| for all x ∈ Rd
(1) Show that for each pair x, y ∈ Rn
z 7→ (y, B z A−z x)
admits an analytic continuation from 0 < z < 1 to the entire complex plane.
(2) Show that
||B θ A−θ x|| ≤ ||x||
for all 0 ≤ θ ≤ 1

Proof. As A and B are real positive definite n × n symmetric matrix, the Spectral Theorem tells us there
are λi > 0 and σi > 0 with orthogonal matrix S, V such that
A = S > diag(λ1 , ..., λn )S and B = V > diag(σ1 , ..., σn )V
Then using that A, B are symmetric, we have for any x, y ∈ Rn that
(y, B z A−z x) = (B z y, A−z x) = (S > diag(λz1 , ..., λzn )Sy, V > diag(σ1−z , ..., σn−z )V )
where az := exp(z log(a)) where we use the standard complex log with a branch cut on the negative real
axis. So in particular, (y, B z A−z x) is some polynomial combination of {λzi } and {σiz }, so z 7→ (y, B z A−z x)
is an entire function for any z ∈ C.
51

Now observe for any 0 ≤ θ ≤ 1 that

||B θ A−θ x|| = sup (y, B θ A−θ x)
||y||=1

Define the entire function for ||y|| ≤ 1

fx,y (z) := (y, B z A−z x)
for z ∈ {x + iy : 0 ≤ x ≤ 1, y ∈ R} and define
M (s) := sup |fx,y (s + it)|
t

Now observe that

M (1) ≤ sup ||B 1+it A−1−it x|| = ||BA−1 x|| ≤ ||x||
t
Now observe
M (0) ≤ sup ||B it A−it x|| ≤ (||B it ||)(||A−it ||)(||x||) = ||x||
t
since the eigenvalues have magnitude 1. Therefore, by Hadamard’s three lines theorem we have for any
θ ∈ [0, 1]
M (θ) = M (θ + (1 − θ)0) ≤ M (0)θ M (1)1−θ ≤ ||x||θ ||x||1−θ = ||x||
Hence, we have
||B θ A−θ x|| = sup fx,y (θ) ≤ sup M (θ) ≤ ||x||
||y||≤1 ||y||≤1

as desired.

Problem 9. Let P (z) be a non-constant complex polynomial, all of whose zeros lie in a half plane
{z ∈ C : Re(z) < σ}. Show that all the zeros of P 0 (z) also lie in the half plane {z ∈ C : Re(z) < σ}.

Proof. By the fundamental theorem of calculus we have that

P (z) = α(z − z1 )(z − z2 )...(z − zn )
where zi are the zeros of P and α ∈ C \ {0} and Re(zi ) < σ. Then it follows that the log derivative
n
P 0 (z) X 1
=
P (z) i=1
z − zi

We already know that if w is a repeated root of P then Re(w) < σ since P (w) = 0, so it suffices to
assume that w is a root of P 0 but not a root of P . So this implies P 0 (w) = 0 ⇐⇒ P 0 (w)/P (w) = 0 (since
P (w) 6= 0). And we obtain
n n
X w − zi n
X Re(w) − Re(zi )
X 1
0= = 2
⇒0=
i=1
w − zi i=1
|w − zi | i=1
|w − zi |2

as Re(zi ) < σ we obtain

n n
X Re(w) − Re(zi ) X Re(w) − σ
0= >
i=1
|w − zi |2 i=1
|w − zi |2
from which it follows that Re(w) < σ. Therefore, all the roots of P 0 (z) also lie in the half plane
{z ∈ C : Re(z) < σ}

Problem 10. Let f : C → C be a non-constant entire function. Without using either of the
Picard theorems, show that there exists arbitrarily large complex numbers z for which f (z) is
positive real.
52 RAYMOND CHU

Proof. Let f be an entire function such that for arbitrarily large complex numbers f (z) is not positive
real. That is there exists a M ∈ N such that
f |{z:|z|>M } : {z : |z| > M } → C \ {x : x ≥ 0}
But by compactness anf continuity we know that there is a K such that |Re(f )| ≤ K on BM (0). Therefore
f is an entire function that maps the simply connected domain Ω := C \ {x : x > K + 1}. Ω is simply
connected since it is star shaped, so by Riemann’s theorem there is a conformal map ϕ : Ω → D such
that
ϕ◦f :C→D
so it follows from Liouville’s theorem that ϕ ◦ f is constant. So it follows from ϕ being bijective that f
is constant.

Problem 11. Let f (z) := −πz cot(πz) be a meromorphic function on C.

(1) Locate all the poles of f and determine their residues.
(2) Show that for each n ≥ 1 the coefficient of z 2n in the Taylor expansion of f (z) about z = 0
coincides with
∞
X 2
an :=
k 2n
k=1

Proof. By using cot(πz) = cos(πz)/ sin(πz) its clear that the poles of f (z) are at j ∈ Z \ {0}. Then for
j ∈ Z \ {0} we have the pole is simple by taylor expansion of sin(πz) so
(z − j) cos(πz)
Res(f, j) = lim −πz = −j
z→j sin(πz)
where the last equality is due to L’hopitals rule (which extends to holomorphic functions thanks to the
taylor series expansion). So f (z) has a pole at each j ∈ Z \ {0} with residue −j.

1
P∞ 2z
Now we recall that π cot(πz) = z + n=1 z 2 −n2 so it follows that
∞
X 2z 2
f (z) = −1 −
n=1
z 2 − n2
Now if we define
∞
X 2z
g(z) :=
n=1
z − n2
then −1 − g(z 2 ) = f (z) so the z 2n taylor coefficient of f (z) is the nth taylor coefficient of −g(z). Now
we see
∞
X 2n2
g 0 (z) = −
n=1
(z − n2 )2
and in general
∞ ∞
X 2n2 n! g (n) (0) X 2
g (n) (z) = (−1)n 2 )n+1
⇒ = − 2n
n=1
(z − n n! n=1
n
which gives the desired result.

53

8. Fall 2013

Problem 1. Let U and V be open and connected sets in C and f : U → C be a holomorphic

function with f (U ) ⊂ V . Suppose that f is proper map from U into V i.e. f −1 (K) ⊂ U is
compact whenever K ⊂ V is compact. Show then that f is surjective.

Proof. Let us show that f (U ) is open and closed. Indeed, if z0 ∈ f (U ) then there exists a z such that
f (z) = z0 then as the proper condition implies that f (z) is non-constant, so the open mapping theorem
tells us that if ε > 0 is so small such that Bε (z) ⊂ U then f (U ) so the image is open.

Now let {yn }n∈N ∈ f (U ) such that yn → y ∈ V where f (zn ) = yn then K := {yn } ∪ {y} is compact so
f −1 (K) is compact by properness
so the sequence {zn } ⊂ f −1 (K), so it has a limit denoted z0 along a subsequence znj . Now we claim
f (z0 ) = y indeed as f (znj ) = ynj → y it follows from continuity that f (z0 ) = y. Therefore, f (U ) is a
non-empty open and closed subset of a connected subset V so f (U ) = V .

Problem 2. Show that there is no function f that is holomorphic near 0 ∈ C and satisfies
n2 − 1
f (1/n2 ) =
n5

Proof. Assume f (z) is holomorphic then f (z 2 ) is holomorphic. Notice that if we define g(z) = z 5/2 z1 − 1 =

z 3/2 − z 5/2 then g(1/n2 ) = f (1/n2 ) and that g(z 2 ) = z 3 − z 5 is holomorphic such that g(z 2 ) = f (z 2 ) on
an accumulation point so g(z 2 ) = f (z 2 ) for all z. But then observe we have the following contradiction
for z 6= 0
f (z 2 ) = f ((−z)2 ) = −z 3 + z 5 = −f (z 2 )
Therefore, f (z) cannot be holomorphic.

Problem 3. Does there exist a holomorphic function f : D → C such that

lim |f (zn )| = +∞
n→∞
for all sequence {zn } ⊂ D with limn→∞ |zn | = 1? Justify your answer.

Proof. No such functions exist. Indeed, as for any z0 ∈ ∂D we have lim supz→z0 |f (z)| = +∞, we can find
a ball B(z0 ) such that on D ∩ B(z0 ) we have |f (z)| > 1. By compactness we can find a finite sub-cover
denoted by B1 , .., BN that cover ∂D and on the compliment of these balls within D we know that f can
only have finitely many zeros, but they have no zeros in these balls. This implies that f has only finitely
many zeros on D, say {zi } with multiplicities mi then define
g(z) := f (z)/(z − zi )mi
which is a new holomorphic function that is non-zero everywhere. And as these zeros are all δ for some
δ > 0 distance away from the boundary, we know that we still have
lim sup |g(z)| = +∞
z→z0

Now fix an M > 0 large then for an z0 ∈ ∂D we can find a ball B(z0 ) with radius ≤ 1/M such that on
B(z0 ) we have |g(z)| ≥ M . By compactness we can find a finite collection of balls B1 , .., BN (with all of
SN
their radius ≤ 1/M ) that cover ∂D and on Bi we have |g(z)| ≥ M . Define ΩM := D \ ( i=1 Bi ) which
is open, and on ∂Ω we have |g(z)| ≥ M and since g never vanishes we can apply the Minimum Modulus
Principle (by looking at 1/g(z) which is holomorphic and applying the max-principle) to conclude that
|g(z)| ≥ M on ΩM . As M → ∞ we know that ΩM → D so we conclude that |g(z)| = +∞ everywhere
which is a contradiction, since g(z) then cannot be continuous, but it was holomorphic.
54 RAYMOND CHU

Problem 4. Let u be a non-negative continuous function on D \ {0} that is subharmonic on

D \ {0}. Suppose that u|∂D ≡ 0 and
ˆ
1
lim u(z)dλ(z) = 0
r→0+ r 2 log(1/r) {z∈C:0<|z|<r}

where integration is with respect to Lebesgue measure λ on C. Show that then u ≡ 0.

Proof. Observe that for ε > 0 that we have for sufficiently small r > 0 that
ˆ
1 1
u(z)dλ(z) ≤ ε log( )
πr2 Br (0) |r|
so we know that z ∈ Br/2 (0) we have Br/4 (z) ⊂ Br (0) \ {0} so we have from the mean value inequality
and non-negativity that
ˆ ˆ
16 16 1 1
u(z) ≤ 2 u(w)dλ(w) ≤ 2 u(w)dλ(w) ≤ 16ε log( ) ≤ 16ε log
πr Br/4 (z) πr Br (0) |r| |z|
1
so u(z) = o(log( |z| ).

1
Now fix any α > 0 then we know u(z) + α log( |z| ) → −∞ as z → 0 thanks to u(z) = o(log(1/|z|)). So
there exists an Rα = R > 0 such that on BRα (0) we have v(z) = u(z) + α log(1/|z|) ≤ 0. Therefore, on
the annulus A1,R = {|z| : R < |z| < 1} we have that v|∂A1,R ≤ 0 but as v(z) is sub-harmonic it follows
that the max is obtained on the boundary so we have v(z) ≤ 0 on A1,R which by letting R → 0 implies
v(z) ≤ 0 on D\{0}. Therefore, letting α → 0 gives u(z) ≤ 0 on D\{0}, which implies from non-negativity
that u(z) ≡ 0

Problem 5. Let {fn } be a sequence of holomorphic functions on D and suppose that

ˆ
|fn (z)|dλ(z) ≤ 1
D
for all n ∈ N where dλ denotes integration with respect to Lebesgue measure λ on C. Show that
then there exists a subsequence {fnk } that converges uniformly on all compact subsets of D.

Proof. Fix a compact subset K ⊂ D and let dist(K, ∂D) := 2δ then for any z ∈ K we have Bδ (z) ⊂ D.
So by the Mean Value Property we have for any z ∈ K and 0 < r < δ
ˆ 2π
1
fn (z) = f (z + reiθ )dθ
2π θ=0
so we have ˆ ˆ ˆ ˆ
δ δ 2π
1 1
rfn (z)dr = r f (z + reiθ )drdθ = fn (z)dλ(z)
r=0 r=0 2π θ=0 2π Bδ (z)
so we have ˆ
1
fn (z) =
fn (z)dλ(z)
πδ 2 Bδ (z)
so in particular, on K we have the uniform bound
1
sup |fn (z)| ≤ 2
z∈K πδ
This implies on any compact subset the family {fn } is uniformly bounded. Therefore, for any n ∈ N with
Ωn := B1− n1 (0) we can by Montel’s theorem find a uniformly convergent subsequence on Ωn . We will
now use a diagonal argument: On Ω1 we can find a subsequence {fn(1) } such that it convegres uniformly
k
(2) (1)
on Ω1 to a limiting function which we denote by f . Then on Ω2 we can a subsequence nk ⊂ nk for
(k)
which fn(2) converges uniformly on Ω2 to f . We repeat this for all n. Define the index nk := nk i.e.
k
55

the diagonal subsequence then we have fnk → f uniformly on Ωj for any j. Indeed, if ε > 0 then there
exists an K ∈ n such that if k 6= N then
||fn(j) − f ||L∞ (Ωj ) ≤ ε
k

(j)
and since nk is a subsequence of nk for k ≥ j we can take K1 := max{K, nk } and obtain
||fnk − f ||L∞ (Ωj ) ≤ ε
Therefore, fnk → f uniformly on Ωn for all n. Now if K is compact it must live in an Ωn for some n
which implies fnk converges uniformly to f on K .

Problem 6. Let U ⊂ C be a bounded open set with 0 ∈ U and f : U → C be a holomorphic

map with f (U ) ⊂ U and f (0) = 0. Show that |f 0 (0)| ≤ 1

Proof. We first observe by the chain rule that if f n := f ◦ f ◦ ... ◦ f where we do n compositions, then we
have
d n
(f (z))(0) = (f 0 (0))n
dz
And we have that for all n that f n (U ) ⊂ U and f n (0) = 0. Since U is bounded there exists an R > 0
such that U ⊂ BR (0). As U is open there exists an ε > 0 such that Bε (0) ⊂ U , so by Cauchy’s theorem
we have for gn := f n
1/n
0 1 n R 0 R
|gn (0)| ≤ ||gn ||L∞ ≤ ⇒ |f (0)| ≤
ε ε ε
where U ⊂ BR (0). Letting n → ∞ gives
|f 0 (0)| ≤ 1
as desired.

Problem 7. Show that there is a dense set of functions f ∈ L2 ([0, 1]) such that x 7→ x−1/2 f (x) ∈
´1
L1 ([0, 1]) and 0 x−1/2 f (x)dx = 0

Proof. As Cc ([0, 1]) is a dense subclass of L2 ([0, 1]) it will suffice to show the claim for f ∈ Cc ([0, 1]). So
´1
fix f ∈ Cc ([0, 1]) this implies there exists an δ = δ(f ) > 0 such f = 0 on [0, δ]. Define I := 0 x−1/2 f (x)dx
which is finite since x−1/2 ∈ L1 ([0, 1]) and f is bounded and for 0 < ε < δ define
I −1/2+ε
gε := − x
δ(ε)
where ˆ ε
δ(ε) = x−1+ε
0
and we write (
f (x) for x ∈ [ε, 1]
f ε (x) :=
gε (x) for x ∈ [0, ε]
so observe
ˆ 1 ˆ ε ˆ 1
−1/2 −1+ε
x gε dx = −I/δ(ε)x dx + x−1/2 f (x)
0 0 ε
ˆ 1
= −I + x−1/2 f (x) = 0
ε
And it is clear that x−1/2 f ε ∈ L1 ([0, 1]) and its clear f ε ∈ L2 ([0, 1]). Then observe that
ˆ 1 ˆ ε 2
I I 2 ε2ε
|f ε − f |2 = 2
x−1+2ε = 2
0 0 δ (ε) δ 2ε
56 RAYMOND CHU

and we observe
εε (εε )2
δ= ⇒ δ2 = 2
ε ε
so ˆ 1
|f ε − f |2 = O(ε)
0
so f ε → f and f so the desired class is dense.

Problem 8. Compute the following limits and justify your calculations!

´k
(1) limk→∞ 0 xn (1 − xk )k
´∞
(2) limk→∞ 0 (1 + xk )−k cos(x/k)

Proof. We first compute

ˆ ∞ ˆ ∞
x
n
lim x (1 − )k dx = xn exp(−x) = n!
0 k→∞ k 0
So we claim that
ˆ k
x k
lim ) = n! xn (1 −
k→∞ 0 k
which we will justify by swapping the limits with the integral. Indeed, observe
ˆ k ˆ ∞
x x
xn (1 − )k = xn (1 − )k χ[0,k]
0 k 0 k
and pointwise we have
x k
xn (1 −
) χ[0,k] → xn exp(−x)
k
Notice by the AMGM inequality we have that
x 1/(k+1) 1 + k(1 − xk ) 1+k−x x
1 · (1 − )k ≤ = =1−
k k+1 k+1 k+1
i.e.
x x k+1
(1 − )k ≤ (1 − )
k k+1
so the family is increasing and we can apply the Monotoe Convergence Theorem to interswap limits with
the derivative to get the integral is n!.

Notice pointwise we have

x −k
lim (1 + ) cos(x/k) → e−x
k→∞ k
so we should have ˆ ˆ
∞ ∞
x
lim (1 + )−k cos(x/k)dx = e−x = 1
k→∞ 0 k 0
Notice that for k ≥ 2
x x x x x
) ≥ (1 + ) ⇒ (1 + )−k ≤ (1 + )−k ≤ (1 + )−2 ∈ L1 ([0, ∞])
(1 +
k 2 k 2 2
so by the DCT we can swap limits since
x x
(1 + )−k | cos(x/k)| ≤ (1 + )−2 ∈ L1 ([0, ∞])
k 2
so ˆ ∞ ˆ ∞
x −k
lim (1 + ) cos(x/k)dx = e−x = 1
k→∞ 0 k 0

57

Problem 9. Let X be a Banach Space, Y a normed linear space, and B : X × Y → R be

a bilinear function. Suppose that for each x ∈ X there exists a constant Cx ≥ 0 such that
|B(x, y)| ≤ Cx ||y|| for all y ∈ Y , and for each y ∈ Y there exists a constant Cy ≥ 0 such that
|B(x, y)| ≤ Cy ||x|| for all x ∈ X.

Show that then there exists a constant C ≥ 0 such that |B(x, y)| ≤ C||x|| · ||y|| for all x ∈ X and
y∈Y

Proof. Fix a y ∈ Y and define the linear operator By : X → R and with By (x) := B(x, y) then by
assumption we have
|By (x)| ≤ Cy ||x||
so By ∈ X ∗ . So we consider the family F := {By }y∈Y with ||y|| = 1 of continuous linear operators.
Fixing an x ∈ X we see that
|By (x)| ≤ Cx ||y|| = Cx since ||y|| = 1
Therefore,
sup |By (x)| < ∞
y∈Y,||y||=1

so by the uniform boundness principle since X is a banach space, the family is uniformly bounded i.e.
there exists a C > 0 such that for any x
|By (x)| ≤ C||x|| for all ||y|| = 1
Now for y 6= 0 we have that
y y
B(x, y) = B(x, ||y|| ) = ||y||B(x, ) ≤ C||y|| · ||x||
||y|| ||y||
as desired .

´
Problem 10. Let f ∈ L2 (R) and define h(x) := R f (x − y)f (y)dy for x ∈ R. Show then that
there exists a function g ∈ L1 (R) such that
ˆ
h(ξ) = e−iξx g(x)dx
R
for ξ ∈ R i.e. h is the fourier transform of some function in L1 (R).

Conversely, show that if g ∈ L1 (R), then 2

´ there is a function f ∈ L (R) such that the fourier
transform of g is given by x 7→ h(x) := R f (x − y)f (y)dy

Proof. Let fn ⊂ L1 (R) ∩ L2 (R) be such that fn → f in L2 (R). Then define

ˆ
hn (x) := fn (x − y)fn (y)dy
R
´
´ then claim hn (x) → h(x) uniformly. Note that h(x) is continuous since h(x) = R f (x − y)f (y)dy =
we
R
f (x + y)f (−y) = (τ−x f (y), f (−y)) where τx f (y) := f (y − x). This implies continuity since
|h(x) − h(z)| = |(τ−x f (y) − τ−z f (y), f (−y))| ≤ ||f (y + x) − f (y + x)||L2 ||f ||L2
Now observe that
ˆ
|h(x) − hn (x)| ≤ |fn (x − y)fn (y) − fn (x − y)f (y)| + |fn (x − y)f (y) − f (x − y)f (y)|dy
R

≤ ||fn ||L2 ||fn − f ||L2 + ||f ||L2 ||fn − f ||L2 → 0

so we have uniform convergence.
58 RAYMOND CHU

´
Now for each fn we know that F −1 (hn ) = R
(hˆn )eitx dx by computation since fn ∈ L1 ∩ L2 . But again
since fn ∈ L1 we have
hˆn = (fˆn )2
´
which implies F −1 (hn ) = (fˆn )2 eitx d. So it follows that
ˆ ˆ
R

hn (t) = F(F −1 (hn ))(t) = gn (x)e−itx dx where gn (x) := (fˆn (t))2 eitx dt
R R

and notice that by Plancherel Theorem that fˆn ∈ L (R) so gn (x) ∈ L (R). Also notice that for g(x) :=
2 1
´
R
(fˆ)2 eitx dx
ˆ h i
|gn (x) − g(x)| = (fˆn (t) + fˆ(t))(fˆn (t) − fˆn (t)) e−itx ≤ ||fn + f ||L2 ||fn − f ||L2 → 0
R
so gn uniformly converges to g. So it follows that from taking n → ∞ on
ˆ
hn (t) = gn (t)e−itx
R
that ˆ
h(t) = g(t)e−itx dx
R
and its clear by Cauchy-Schwarz that g ∈ L1 (R) as desired.

Let us formally derive what f should be first. Indeed, observe if we have such an f then
ˆ ˆ ˆ
g(x)e−iξx = f (ξ − y)f (y)dy = F (x)e−iξx dx
R R R
where ˆ
F (x) := (fˆ(t))2 eitx
R
so we expect
ˆ
(fˆ(t))2 eitx dt = F −1 (fˆ2 ) ⇒ F(g(x)) = fˆ2 ⇒ f = F −1 ( F(g(x)))
p
g(x) =
R
where
p p F(g(x))|
F(g(x)) := |F(g(x))
|F(g(x))|
and 0 when F(g(x)) = 0.

Problem 11. Consider the space C([0, 1]) of real-valued continuous functions on the interval
´ 1/2
1
[0, 1]. We denote ||f ||∞ := supx∈[0,1] |f (x)| the supremium norm and by ||f ||2 := 0 |f (x)|2
the L2 norm of the function f ∈ C([0, 1]).

Let S be a subspace of C([0, 1]). Show that if there existed a constant K > 0 such that
||f ||∞ ≤ K||f ||2 for all f ∈ S, then S is finite dimensional.

Proof. Notice that if we endow S with || · ||2 then this is an equivalent norm, that the evaluation linear
functional (for x ∈ [0, 1])
Lx (f ) := f (x)
is a continuous linear functional since
|Lx (f )| ≤ ||f (x)||L∞ ≤ K||f ||L2
so it extends to a continuous linear operator on S. So by Risez Representation Theorem (the equivalent
norm implies S is a Hilbert Space), we can find a gx ∈ S such that
ˆ 1
Lx (f ) = (f, gx ) = f gx dy
0
59

Fix N orthonormal vectors {ei (x)}N

i=1 , then we compute from Bessel’s Inequality that
N
X N
X
|ei (x)|2 = |(ei , gx )|2 ≤ ||gx ||2
i=1 i=1

and we have
||gx ||22 = (gx , gx ) = gx (x) ≤ ||gx ||∞ ≤ K||gx ||2 ⇒ ||gx ||2 ≤ K
so we have
N
X N ˆ
X 1
|ei (x)|2 ≤ K 2 ⇒ |ei (x)|2 dx ≤ K 2
i=1 i=1 0

i.e.
N ≤ K2
so we can have at most K orthnormal vectors i.e. at most K linearly independent vectors since we can
preform Grahm-Schmit, so the space is fintie dimensional.

Problem 12. Let f : [0, 1] → R be a continuous function that is absolutely continuous on each
interval [ε, 1] for 0 < ε ≤ 1.
(1) Show that f is not necessarily absolutely continuous on [0, 1].

(2) Show that if f is of bounded variations on [0, 1], then f is absolutely continuous on [0, 1].

Proof. For (1) we take f (x) := x sin(1/x) with f (0) := 0 then f ∈ C([0, 1]) by the squeeze theorem and
f (x) ∈ C 1 ((0, 1)) since
cos(1/x)
f 0 (x) = sin(1/x) −
x
which is continuous on the open interval (0, 1). So in particular, if 1 ≥ ε > 0 then on [ε, 1] we have
||f 0 ||L∞ ≤ C(ε) so f (x) is lipschitz on [ε, 1] so it is absolutely continuous on [ε, 1]. But f is not absolutely
2
continuous on [0, 1] since it is not of bounded variation. Indeed, notice that at xn := π(2n+1) and xn → 0
n
as n∞ where n ≥ 1 and sin(1/xn ) = (−1) so we have
N N N N
X X X 2X 1
|f (xn ) − f (xn+1 )| = |(−1)n xn − (−1)n+1 xn+1 | = |xn + xn+1 | ≥ →∞
n=1 n=1 n=1
π n=1 2n + 1

Therefore, the total variation is unbounded over [0, 1] is unbounded since

N
X
Tf ([0, 1]) = sup{ |f (xn ) − f (xn+1 )| : 0 = x1 < ... < xN +1 = 1}
n=1

and we can always adjoin to the above sum xN +2 = 0 which only increases the sum size. So in particular,
f is not of bounded variation so it cannot be absolutely continuous.

For (2) we claim that the total variation is continuous i.e.

N
X
Tf ([x, y]) := sup{ |f (xn ) − f (xn+1 )| : x = x1 < ... < xN +1 = y}
n=1

for any f with the above conditions. Indeed, by uniform continuity if ε > 0 then there exists a δ > 0
such that if |x − y| < δ then |f (x) − f (y)| < ε. So notice we can find a partiton 0 = x1 < ... < xN +1 = 1
such that
N
X
Tf ([0, 1]) ≤ ε + |f (xi+1 ) − f (xi )|
i=1
60 RAYMOND CHU

Now fixing x ∈ (0, 1] we can assume that x = xM and |xM − xM +1 | < δ for some M since adding a point
to our partiton only increases the sum. So in particular,
M
X −1 N
X
Tf ([0, 1]) ≤ ε + |f (x) − f (xM +1 )| + |f (xi+1 ) − f (xi )| + |f (xi+1 ) − f (xi )|
i=1 M +1

≤ 2ε + Tf ([0, x]) + Tf ([xM +1 , 1])

Therefore, as
Tf ([0, 1]) = Tf ([0, x]) + Tf ([x, xM +1 ] + Tf ([xM +1 , 1]
we conclude that
Tf ([x, xM +1 ] ≤ 2ε
so in particular Tf is right continuous. But an identical argument using the if |x − xM −1 | < δ shows that
Tf is left continuous, so its continuous.

Now as Tf ([0, x]) is uniformly continuous since [0, 1] is compact we can find a δ > 0 such that if ε > 0
is given then
Tf ([0, x]) ≤ ε when |x| < δ
PN
Then on [δ/2, 1] we know that f is absolutely continuous so we can find a η > 0 such that if i=1 |xi −yi | <
PN
η then i=1 |f (xi ) − f (yi )| < ε where {xi }N N
i=1 and {yi }i=1 are in the interval [δ, 1]. Let δ̂ := min{δ, η}/4
then if we are given
X M
|xi − yi | < δ̂
i=1
where the xi , yi ∈ [0, 1] then we know that if xi ∈ [0, δ/2] then as
|xi − yi | < δ/4 ⇒ yi ∈ [0, δ/2]
So we relabel our sequence to {xi }M M
and {xi }N
i=1 , {yi }i=1
N
i=M +1 , {yi }i=M +1 where xi , yi ∈ [0, δ/2] for
1 ≤ i ≤ M and xi , yi ∈ [δ/2, 1] for M + 1 ≤ i ≤ N . Then we have
M
X N
X M
X
|f (xi ) − f (yi )| + |f (xi ) − f (yi )| ≤ |f (xi ) − f (yi )| + ε
i=1 i=M +1 i=1

where the second inequality is due to absolute continuity of [δ/2, 1] and observe
≤ Tf ([0, δ/2]) + ε ≤ 2ε
by uniform continuity of the total variation. So f is uniformly continuous.

61

9. Spring 2014

Problem 1. Consider a measure space (X, X ) with a σ-finite measure µ and, for each t ∈ R,
let et denote the characteristics function
´ of the interval (t, ∞). Prove that if f, g : X → R are
X -measurable, then ||f − g||L1 (X) = R ||et ◦ f − et ◦ g||L1 (X) dt

Proof. Notice that ˆ ˆ ˆ

||et ◦ f − et ◦ g||L1 (X) = |et ◦ f − et ◦ g|dµdt
R R X
so by Tonelli’s since the integrand is non-negative and µ is σ-finite we have
ˆ ˆ ˆ ˆ
= |et ◦ f − et ◦ g|dtdµ = |χf (x)≥t − χg(x)≥t |dtdµ
X R X R
Now we compute
ˆ ˆ max{f (x),g(x)}
|χf (x)≥t − χg(x)≥t |dt = 1dt = |f (x) − g(x)|
R min{f (x),g(x)}

so this gives us ˆ
||et ◦ f − et ◦ g||L1 (X) = ||f − g||L1 (X)
R

Problem 2. Let f ∈ L1 (R, dx) and β ∈ (0, 1). Prove that

ˆ
|f (x)|
dx < ∞
R |x − a|β
for (Lebesgue) a.e. a ∈ R.

Proof. Fix n ∈ N then by Tonelli since |f (x)|/|x − a|β ≥ 0 we can justify the following computation
ˆ n ˆ ˆ ˆ n ˆ ˆ n
|f (x)| |f (x)| 1
β
dxda = β
dadx = |f (x)| dadx
a=−n R |x − a| R a=−n |x − a| R a=−n |x − a|β
and observe that ˆ ˆ ˆ
n x+n n
1 1 1
da = da ≤ da ≤ C(β, n)
a=−n |x − a|β x−n |a|β −n |a|β
since 1/|x|β ∈ L1loc (R) so we have
ˆ n ˆ ˆ
|f (x)|
dxda ≤ C(β, n) |f (x)|dx < ∞
a=−n R |x − a|β R
´ |f (x)|
so we conclude that R |x−a|β
dx < ∞ a.e. on [−n, n], which by letting n → ∞ implies this is true for a.e.
a ∈ R.

Problem 3. Let [a, b] ⊂ R be a finite interval and f : [a, b] → R be a bounded Borel Measurable
function.
(1) Prove that {x ∈ [a, b] : f (x) is continuous at x } is Borel Measurable.
(2) Prove that f is Riemann Integrable if and only if its continuous almost everywhere.

Proof. Define A := {x ∈ [a, b] : f (x) is continuous at x } then we claim that A is a Gδ set. Indeed, define
the oscillation
ω(f, A) := sup |f (x) − f (y)|
x,y∈A
62 RAYMOND CHU

then notice that f is continuous at x iff

lim ω(f, Br (x)) := ω(f, x) = 0
r→0+

where the limit exists since its an inf. Now observe

∞ ∞
\ 1 \
A = {x : ω(f, x) = 0} = {x : ω(f, x) < } := An
n=1
n n=1

Now we claim that each An is an open set. Indeed, if ω(f, x) < 1/n then there is a ball Br (x) such that
ω(f, Br (x)) < 1/n
Then for any y ∈ Br/2 (x) we can find a δy > 0 such that Bδy (y) ⊂ Br (x) so it follows that
ω(f, Bδy (y)) ≤ ω(f, Br (x)) < 1/n
so Br/2 (x) ⊂ An so An is open and the set of continuity is a Gδ , so it is Borel Measurable.

Now we prove f is Riemann Integrable iff it is continuous a.e. Notice if B is the set where f (x) is
discontinuous on [a, b] then
∞ ∞
[ 1 [
B= {x ∈ [a, b] : ω(f, x) > } := Bn
n=1
n n=1

so it suffices to show m(Bn ) = 0 for all n.

For the second condition fix a partition P a = x0 < ... < xn = b. Then define the lower and upper
Riemann Sums
n
! n
X X
U (f, P ) := sup f (x) ∆xi and L(f, P ) := inf f (x) ∆xi
x∈[xi−1 ,xi ] x∈[xi−1 ,xi ]
i=1 i=1

where ∆xj = xj − xj−1 . Observe

Xn ˆ b n ˆ b
X
U (f, P ) = (sup f (x))χIi (y)dy and L(f, P ) = ( inf f (x))χIi (y)dy
a x∈Ii a x∈Ii
i=1 i=1

where Ii = [xi−1 , xi ]. Then

n ˆ
X b
U (f, P ) − L(f, P ) = ω(f, Ii )χIi (y)dy
i=1 a

So if f is continuous a.e., then m(A) = m({x : ω(f, x) = 0}) = b − a. So in particular, ω(f, Ii ) → 0 as

∆x → 0 where ∆x := max1≤j≤n ∆xj , so the dominated convergence theorem implies since the oscillation
is bounded since f is bounded that
U (f, P ) − L(f, P ) → 0 as ∆x → 0
i.e. f is Riemann Integrable.

Now for the reverse direction fix ε > 0, since f is Riemann Integrable there exists an δ > 0 such that
if ∆x < δ then
n
X
0 ≤ U (f, P ) − L(f, P ) = ω(f, Ii )m(Ii ) ≤ ε
i=1
Write Fn := {x : ω(f, x) > 1/n} Now write
I := {k ∈ 0, 1.., n : Fn ∩ Ik 6= ∅}
this implies
[
Fn ⊂ Ii ∪ {x0 , .., xn }
i∈I
63

P
so it suffices to show i∈I m(Ii ) = O(ε). Notice as Ii ∩ Fn 6= ∅ that ω(f, Ii ) > 1/n so we have
n
1X X X
m(Ii ) ≤ ω(f, Ii )m(Ii ) ≤ ω(f, Ii )m(Ii ) ≤ ε
n i=1
i∈I i∈I

so it follows that X
m(Ii ) ≤ εn
i∈I
this implies m(Fn ) = 0 so f is contiuous a.e.

Problem 4. Consider a sequence {an : n ≥ 1} ⊂ [0, 1]. For f ∈ C([0, 1]), define
X∞
ϕ(f ) := 2−n f (an )
n=1
1
´
Orive that there is no g ∈ L such that ϕ(f ) = f (x)g(x)dx is true for all f ∈ C([0, 1]).

(2) Each g ∈ L1 defines a continuous functional Tg ∈ L∞ via

ˆ
Tg (f ) = f (x)g(x)dx

Show there are continuous functionals on L∞ ([0, 1]) that are not of this form.

Proof. Define B([0, 1]) to be the space of bounded functions on [0, 1] endowed with the sup norm, now
we extend ϕ to a map on this space via
∞
X
ϕ(f ) := 2−n f (an )
n=1
So now we can define a measure
µ(E) := ϕ(E)
and it is easy to see µ is a measure and it is not absolutely continuous with respect to the Lebesgue
measure since µ({an }) = 2−n and {an } is a null set for the Lebesgue measure. Therefore, by Radon
Nikodyn there exists a mf m and λ ⊥ m with λ 6= 0 such that
µ = mf + λ
And notice that ˆ
ϕ(f ) = f dµ
due to linearity and equality holding for simple functions. So it follows that ϕ µ so ϕ is not absolutely
continuous with the Lebesgue measure.
For the second part, by Hanh-Banach as ϕ is a continuous linear functional on the subspace C([0, 1])
we can extend it to a continuous linear functional on L∞ ([0, 1]). And the desired result holds since if ϕ
is a functional of that form it implies its restriction is as well, which we proved in a) was not.

Problem 5. Recall that a metric space is separable if it contains a countable dense subset.
(1) Prove that `1 (N) and `2 (N) are separable Banach Spaces, but `∞ is not.
(2) Prove there exists no linear bounded surjective map T : `2 (N) → `1 (N)

Proof. Denote ei to be the vector that is 0 everywhere except for a 1 at the ith position. We claim that
finite rational combinations of ei are dense in `p for any 1 ≤ p < ∞. This immediately implies (1).
Indeed, if x = (x1 , x2 , ...) ∈ `p and ε > 0 there is an N ∈ N such that
∞
X
|xn |p ≤ ε
n=N
64 RAYMOND CHU

Then for each i = 1, .., N we can find a rational qi such that |xi − qi |p ≤ ε2−i . This implies for
PN
y := i=1 qi ei with y = (y1 , y2 , ...)
∞
X N
X ∞
X ∞
X
|xn − yn |p = |xn − qn |p + |xn |p ≤ ε2−n + ε = 2ε
n=1 n=1 n=N +1 n=1

Therefore, we have shown finite rational combinations of ei are dense in `p for 1 ≤ p < ∞ as desired.

To show `∞ is not separable we will show there exists an uncountable sequence {xα }α∈A such that
||xα ||`∞ = 1 and ||xα − xβ ||`∞ ≥ 1/2 whenever α 6= β. This implies `∞ is not separable since if {vi } is a
countable sub-sequence of `∞ then if
||xα − vi ||∞ < 1/4
then
||xβ − vi ||∞ ≥ ||xα − xβ || − ||xβ − vi || ≥ 1/2 − 1/4 = 1/4
so the vi cannot be dense since there are uncountably many xβ . Now we construct such xα . Notice that
there are uncountably many binary strings i.e. sequences where every entry is 0 or 1. Observe that if xα
and xβ are different binary strings then
||xα − xβ ||`∞ = 1
and ||xα ||`∞ = 1, so we are done.

For (2), if T is surjective then its adjoint T ∗ : `∞ (N) → `2 (N) is injective. So T ∗ is a linear isomorphism
from `∞ to a subset of `2 . This implies that T ∗ (`∞ (N)) is separable since it is a subset of `2 . But then
since (T ∗ )−1 : T ∗ (`∞ (N)) → `∞ is a homeomomorphism we see it preserves separability, which means
`∞ is separable, which contradicts the first part.

Problem 6. Given a Hilbert Space H, let {an }n≥1 ⊂ H be a sequence with ||an || = 1 for all
n ≥ 1. Recall that the convex hull of {an }ng1 is the closure of the set of all convex combinations
in {an }n
Pm
(1) Show that if {an } spans H linearly (i.e., any x ∈ H is of the form k=1 ck ank , for some
m and ck ∈ C), then H is finite dimensional.
(2) Show that if (an , ξ) → 0 for all ξ ∈ H, then 0 is in the closed convex hull of {an }n .

Proof. We argue by contradiction and assume that there are infinitely many an that are linearly inde-
pendent. Denote the largest subset of {an } such that every term is linearly independent as {bn }. This
automatically implies span{bn } = span{an } which exists thanks to Zorn’s Lemma. Now we go Gram-
Schmit on bn to obtain a new sequence {αn } that are orthonormal and span{αn } = span{an }. Note
there are infinitely many αn . So in particular define
∞
X 1
y := α ∈H
2 n
n=1
n
Pn
since it is the limits of yn := k=1 (1/n2 )αn (and H is complete). But as αn is a linear combination of an
it must still linearly span H, but our y is not any finite linear combination of αn . Thus we have arrived
at a contradiction. This implies H is finite dimensional.

For (2) notice that if v1 , .., vN ∈ H with ||vi || = 1 then

1 X 1
|| vi || = + sum of N inner products
N N
taking vi to be some ank we see that we can make their convex combination norm get arbitrarily close
to 0 for a correct subsequence thanks to (an , ξ) → 0.

65

Problem 7. Characterize all entire functions with |f (z)| > 0 for |z| large and
| log |f (z)||
lim sup <∞
z→∞ |z|

Proof. Notice that as zeros are isolated and f (z) is non-zero for large z we know that f has only finitely
many roots. So there is a polynomial P (z) with the exact same zeros as f . It follows that g(z) :=
f (z)/P (z) is an entire function with no zeros. Therefore, there is a h(z) entire such that
f (z)/P (z) = exp(h(z))
Then notice
| log |g(z)|| | log |f (z)| − log |P (z)|| | log |f (z)|| + | log |P (z)||
lim sup = lim sup ≤ lim sup <∞
z→∞ |z| z→∞ |z| z→∞ |z|
so it follows that from log |g(z)| being finite on every compact set since it is a continuous function that
there is a C > 0 such that
| log |g(z)|| ≤ C|z|
This implies by taking exponentials that
|g(z)| = | exp(h(z))| ≤ exp(C|z|)
i.e. g is an entire function of order 1 with no zeros, so by Hadamard’s Factorization Theorem we know
there is a linear polynomial Az + B such that
g(z) = exp(Az + B) ⇒ p(z) = P (z) exp(Az + B)
so these functions are polynomials multiplied by the exponential of a linear function.

Problem 8. Construct a non-constant entire function f (z) such that the zeros of f are simple
and coincide with the set of all (positive) natural numbers.

Proof. We mimic the proof of Weiestrass Factorization Theorem. Define the canonical factors
En := (1 − z) exp(z + z 2 /2 + .. + z n /n)
and define
∞
Y z
g(z) := Ej ( )
j=1
n
then this is an entire function with simple zeros with zeros at the positive natural numbers.

Problem 9. Prove Hurwitz’s Theorem. Let Ω ⊂ C be connected open set and fn , f : Ω → C be

holomorphic functions such that fn (z) converges uniformly on compact sets to f (z). Prove that
if fn (z) 6= 0 for all n then either f is identically zero or f (z) 6= 0 for any z ∈ Ω.

Proof. Define A := {z ∈ Ω : f (z) = 0}. Continuity implies A is closed, so we will show that A is open
which will imply either f ≡ 0 or f is never zero due to connectedness.

Indeed, fix zn ∈ A. Assume for the sake of contradiction that f is not identically 0 in a neighborhood of
zn . Then as all the zeros are isolated, this implies there exists an ε > 0 such that Bε (zn ) ⊂ Ω such that
on ∂Bε (zn ) we have f 6= 0. Now by the argument principle we have that if we define γ := εeiθ + zn for
θ ∈ [0, 2π) then ˆ 0
1 f
dz ≥ 1
2πi γ f
66 RAYMOND CHU

since we have at least one zero in our domain. But as fn → f uniformly on Bε (zn ) this implies by
Cauchy’s Integral Formula that fn0 → f 0 uniformly on Bε (zn ). Indeed, recall
ˆ ˆ 2π
0 1 fm (w) 1 fm (zn + εeiθ )
fm (z) = dw =
2πi γ (w − z)2 2π 0 εeiθ
So we do have fn0 → f 0 uniformly. Therefore, we have
ˆ 0 ˆ 0
1 fm 1 f
0= dz → dz ≥ 1
2πi γ fm 2πi γ f
which is a contradiction. This implies f is identically zero in a neighborhood of 0. So A is open and
closed, which implies by connectedness either f ≡ 0 or f is never zero.

Problem 10. Let α ∈ [0, 1] \ Q and let {an } ∈ `1 (N) with an 6= 0 for all n ≥ 1. Show that
X an
f (z) := ,z ∈ D
z − eiαn
n≥1

converges and defines a function that is analytic in D which does not admit an analytic continuation
to any domain larger than D

Pn
Proof. We will show fn := j=1 an /(z − eiαn ) converges locally uniformly to f . Indeed, fix a compact
set K ⊂ D then there is an 0 < r < 1 such that K ⊂ Br (0). Then we have
an |an |
≤
z − eiαn 1−r
this implies since an ∈ `1 (N) that f (z) is absolutely convergent on K. This implies fn → f uniformly
on K and as fn is a finite sum of holomorphic functions on D we see that fn is holomorphic, so since
we have uniform convergence this implies by Morrera’s theorem that f is holomorphic on K. Then this
implies by taking a compact exahusation of D that f is holomorphic on D.

Now to see why there is no analytic continuation onto a larger domain that contains D. If such an
extension existed then it must contain a circle arc of ∂D. So by density of eiαn on the Torus since α is
irrational we have that there is an eiαm in this arc. But then limr→1− |f (reiαm )| = +∞, so this function
cannot be continuous on ∂D. Therefore, no such analytic continuation exists.

Problem 11. For each p ∈ (−1, 1) compute the improper Riemann integral
ˆ ∞
xp
dx
0 x2 + 1

Proof. Fix ε > 0 and R > 0 and define γ1 := {Reiθ : θ ∈ [0, π]}, γ2 := {−R(1 − t) + εt : t ∈ [0, 1]},
γ3 := {εeiθ : θ ∈ [0, π]} (with γ3 having clock wise orientation) and γ4 := {ε(1 − t) + Rt : t ∈ [0, 1]}
P4
where all these curves except γ3 have counter clock wise orientation. Now define γ := i=1 γi and now
we compute ˆ
zp
2
dz
γ z +1
where z p := exp(p log(z)) where log(z) := log(|z|) + iarg(z) with arg(z) ∈ [−π/2, 3π/2] i.e. the log with
a branch cut on the negative imaginary axis. On the big circular arc γ1 we have
ˆ ˆ π
zp Rp epiθ
2
dz = 2 2iθ + 1
Rieiθ
γ1 z + 1 θ=0 R e
67

so ˆ ˆ π
zp R1+p R1+p
2
dz ≤ 2
=π
γ1 1 + z 0 1+R 1 + R2
since p ∈ (−1, 1) this goes to 0 as R → ∞.

Now let us see what happens on the small semi-circle of radius ε. Observe that on this circle circle
that |z| = ε so we have
ˆ ˆ π p ipθ
zp ε e
2+1
dz = 2 eiθ
iεiθ
γ3 z 0 1 + ε
so ˆ p
z ε1+p
2
dz ≤ π
γ3 z + 1 1 + ε2
which vanishes as ε → 0 since p ∈ [−1, 1] so the numerator is a positive power of ε. So now we compute
the residuals at z = i. to find the integral evaluation. Indeed,
zp ip exp(p( π2 i))
Res(f, i) = lim = =
z→i z + i 2i 2i
and using if x is real then (−x)p = exp(p log |x|) exp(piπ) = |x|p exp(ipπ) to conclude
ˆ ∞ ˆ
xp 1 zp 1
2
dx = lim 2
dz = π (exp(p(π/2i)))
0 1+x 1 + exp(ipπ) ε→0,R→∞ γ 1+z 1 + exp(ipπ)

Problem 12. Compute the number of zeros, including multiplicity, of f (z) := z 6 + iz 4 + 1 in the
upper half plane in C.

Proof. By Rouche’s Theorem we know that if we can show |iz 4 | < |z 6 +1| on the real axis, then z 6 +1+iz 4
has the same number of zeros on the upper half plane as z 6 + 1 which has 3 namely 3 roots of unity.
Indeed, it suffices to show for real x that
f (x) := x6 + 1 − x4 > 0
Observe as x → ±∞ that f (x) → +∞ so if f was non-positive somewhere, its minimum exist and is in
a compact set in R has to be non-positive. So differentiating, we have
f 0 (x) = 5x4 − 4x3 = x3 (5x − 4)
so the potential zeros are x = 0 or x = 4/5. And it is clear f (0), f (4/5) > 0 so f (x) > 0. So there are
exactly 3 roots in the upper half plane by Rouche’s Theorem
68 RAYMOND CHU

10. Fall 2014

Problem 1. Let ˆ
A := {f ∈ L3 (R) : |f |2 < ∞}
R
Show that A is a Borel subset of L3 (R).

´N
Proof. Note that f ∈ A if and only if there exists an M > 0 such that for all N we have −N
|f |2 ≤ M
i.e.
∞ \
[ ∞ ˆ n
3
A= {f ∈ L (R) : |f |2 ≤ m}
m=0 n=0 −n

so it suffices to show the linear functional

ˆ n
Λ(f ) := |f |2
−n
´n
is Borel measurable since {f ∈ L3 (R) : −n |f |2 ≤ m} = Λ−1 ([0, m)). In particular, it suffices to show it
is continuous. Indeed, fix f, g ∈ L3 (R) then
ˆ n ˆ n 23 ˆ n 1/3
2 3 3
|Λ(f − g)| = |f − g| dx ≤ |f − g| 1 = 21/3 n1/3 ||f − g||2L3 (R)
−n −n −n
´n
so in particular, Λ is continuous, so {f ∈ L3 (R) : −n
|f |2 ≤ m} is Borel, hence A is Borel.
´n
Alternative one can use Fatou’s Lemma to show that {f ∈ L3 (R) : −n
|f |2 ≤ m} is closed.

Problem 2. Construct an f ∈ L1 (R) so that f (x + y) does not converge a.e. to f (x) as y → 0.

Prove that your f has this property.

Proof. Let C be a fat cantor set, then m(C) > 0 and it has no open intervals. Define f (x) := χC then
this is not equal to 0 a.e. since m(C) > 0 and since C is measurable so is χC and χC is supported on a
set of finite measure so it is in L1 (R). Then notice for any x ∈ C as C has no open intervals around x,
we conclude there is a sequence yn ∈ K c such that yn → x to get
f (x + yn ) = 0 for all n
but f (x) = 1 and this is true for all x ∈ K so we do not have a.e. convergence of the translates pointwise.

Problem 3. Let (fn ) be a bounded sequence in L2 (R) and suppose that fn → 0 Lebesgue a.e.
Show that fn * 0 in the weak topology of L2 (R).

Proof. Same argument as Fall 2012 Number 1.

Problem 4. Given f ∈ L2 ([0, π]) we say f ∈ G if f admits a representation of the form

X∞ X∞
f (x) := cn cos(nx) and (1 + n2 )|cn |2 < ∞
n=0 n=0
Show that if f, g ∈ G then f g ∈ G.
69

Proof. Note that due to the summation condition we have that

∞ ∞ √
X X 1 + n2 p
|cn | = √ |cn | ≤ ||(1 + n2 )−1 ||`2 || 1 + n2 |cn ||`2 < ∞
n=0 n=0
1 + n2

where we used Cauchy-Schwarz. Therefore, the sum that defines f (x) converges uniformly. So in par-
ticular, there exists a representative of f ∈ L2 ([0, π]) that is continuous and we choose to look at this
representative. From now on we regard G as a subset of L2 ([0, π]) ∩ C([0, π]). We also may by taking an
even extension i.e.
f (−x) := f (x) for x ∈ [0, π]
may regard G as a subset of continuous even functions on [−π, π]. And we know that the basis of even
functions on C([−π, π]) in the L2 norm is {cos(nx)}. So now fix f, g ∈ G then f g is continuous and even.
Therefore, we have that the nth Fourier Coefficient in the L2 sense is
ˆ π X∞ ˆ π
πan := f (x)g(x) cos(nx) = bj cos(jx)cm cos(mx) cos(nx)
−π j,m=0 −π

∞
X
=π bj cn−j
j=0
P P
where f = bj cos(jx) and g = cj cos(jx) and we swapped sum and integrals thanks to uniform
convergence of each sum. So in particular, we have
X
an = bj cn−j

Therefore, we have
 2
X X Xp
(1 + n2 )|an |2 ≤  1 + n2 |bj ||cn−j |
n n j

Now we use that p p p

1 + n2 . 1 + (n − j)2 + 1 + j2
to get
Xp Xp p
1 + n2 |bj ||cn−j | . 1 + (n − j)2 |cn−j ||bj | + 1 + j 2 |bj ||cn−j |
j j

So it follows that (x + y)2 ≤ 2x2 + 2y 2 with taking x, y as the above sums that
 2  2
Xp Xp p 2
 1 + n2 |bj ||cn−j | .  1 + (n − j)|cn−j ||bj | + 1 + j 2 |bj ||cn−j |
j j

so by Young’s Convolution Inequality we have that

X √ p
(1 + n2 )|an |2 . ||bn ||`1 || 1 + n|cn |||`2 + || 1 + n2 |bn |||`2 ||cn ||`1 < ∞
n

so f g ∈ G .

Problem 5. Let φ : [0, 1] → [0, 1] be continuous and dµ be a Borel Probability measure on [0, 1].
Suppose µ(φ−1 (E)) = 0 for every Borel Set E ⊂ [0, 1] with µ(E) = 0. Show that there is a Borel
measurable function w : [0, 1] → [0, ∞) so that
ˆ ˆ
f ◦ φ(x)dµ(x) = f (y)w(y)dµ(y) for all continuous functions f : [0, 1] → R
70 RAYMOND CHU

Proof. Let E be Borel measurable subset of [0, 1] then define f = χE (x) then we have f ◦ φ(x) = χφ−1 (E)
ˆ 1 ˆ 1
f ◦ φ(x)dµ(x) = χφ−1 (E) (x)dµ = µ(φ−1 (E))
0 0
This implies for any continuous function f : [0, 1] → R that
ˆ 1 ˆ 1
f ◦ φ(x)dµ(x) = f dφ∗
0 0

where we define the push forward measure φ∗ (E) := µ(φ−1 (E)) by the definition of an integral. Now we
see from the condition of φ∗ (E) = 0 whenever µ(E) = 0 that there exists an w ∈ L1 (dµ) such that
dφ∗ = wdµ
i.e. for any continuous function f we have
ˆ 1 ˆ 1
f ◦ φ(x)dµ(x) = f (y)w(y)dµ(y)
0 0

and since φ∗ is a positive measure, we know w(y) ≥ 0 and its set where its +∞ is a µ null set, so we can
redefine it on this null set to make w(y) : [0, 1] → [0, ∞).

Problem 6. Let X be a Banach Space and let X ∗ its dual space. Suppose X ∗ is separable, show
that X is separable. (You may assume the axiom of choice).

Proof. Let {fn } be a countable dense subset of X ∗ . For each n ∈ N we can find an xn ∈ X such that
1
fn (xn ) ≥ ||fn ||
2
So define D to be the set of all finite linear combinations of xn and E to be the set of all finite rational
linear combinations of D. It suffices to show E is dense since D is a countable dense subset of E. Indeed,
assume for the sake of contradiction that D 6= X, so there is an x ∈ X \ D, so by Hanh Banach there is
a linear functional f ∈ X ∗ such that
f (x) 6= 0 and f |D = 0
Now observe that
||f || ≤ ||f − fn || + ||fn || ≤ ||f − fn || + 2fn (xn )
and as xn ∈ D we see that
= ||f − fn || + 2fn (xn ) − 2f (xn ) ≤ 3||f − fn ||
Therefore, as fn is dense we conclude that
||f || = 0
i.e. f is the zero function, but this contradicts the fact that f (x) 6= 0. Therefore, we must have D = X,
so it follows that E is a countable dense subset of X.

Problem 7. Find an explicit conformal map from the Upper Half Plane-Slit along the vertical
segment
{z ∈ C : Im(z) > 0} \ (0, 0 + ih]
for h > 0 to the unit disk D.
71

Problem 8. Let f : C → C be an entire function. Show that

|f (z)| ≤ Cea|z| , z∈C
for some constants C, a if and only if we have
|f (n) (0)| ≤ M n+1 , n = 0, 1, ..
for some constant M .

Proof. Note that as f is entire we have

∞
X f (n) (0) n
f (z) = z
n=0
n!

Now for the ⇐ direction observe that |f (n) (0)| ≤ M n+1 implies
∞ ∞
X M n+1 n X |M z|n
|f (z)| ≤ |z| = M = M exp(|M ||z|)
n=0
n! n=0
n!
so this direction has been shown with C = a = |M |.

For the reverse direction, note that Cauchy’s Estimate implies for any R > 0
|f (n) (0)| exp(aR)
≤ max |f (z)|/Rn ≤ C
n! z∈∂BR (0) Rn
By taking derivatives on the final expression, we see that it is minimized when R = n/a i.e.
an
an := |f (n) (0)|/n! ≤ C exp(n)
nn
now stirling’s approximation gives
√
nan an
an . . for n ≥ 1
n! (n − 1)!
Therefore, we have
∞ ∞
X X an
|f (z) − f (0)| ≤ an |z|n . |z|n
n=1 n=1
(n − 1)!
In particular,
|f (n) (0)| ≤ nan . (2a)n+1
as desired.

Problem 9. Let Ω ⊂ C be open and connected. Let (fn ) be a sequence of injective holomorphic
functions defined on Ω and suppose fn → f locally uniformly in Ω. Show that if f is not constant,
then f is injective in Ω.

Proof. Let w ∈ Ω and define A := {z ∈ Ω \ {w} : f (z) = f (w)}. Notice that A is closed since
A = f −1 (f (w)) and f is continuous. So it suffices to show A is open in Ω \ {w} to conclude the problem
since Ω \ {w} is still connected (since Ω is open). Indeed, let z0 ∈ A then we have for 0 < ε 1 that
Bε (z0 ) ⊂ Ω and w ∈/ Bε (z0 ) and the argument principle tells us
ˆ
f (z) − f (w)
dz ≥ 1
|z−z0 |=ε f 0 (z)
where we are assuming for the sake of contradiction that f (z) is not identically f (w) in a small ball
around z, so as zeros of holomorphic functions are isolated we can find a small enough ε > 0 such that
f (z) − f (w) 6= 0 on |z − z0 | = ε which allows us to apply the Argument Principle.
72 RAYMOND CHU

But notice that fn (z)−fn (w) → f (z)−f (w) (which by Cauchy’s Integral Formula implies fn0 (z) → f 0 (z)
uniformly on Bε (0)) uniformly on Bε (z0 ) and as fn is injective the argument principle tells us that
ˆ
fn (z) − fn (w)
dz = 0
|z−z0 |=ε fn0 (z)
so uniform convergence tells us
ˆ ˆ
fn (z) − fn (w) f (z) − f (w)
0 = lim dz = dz ≥ 1
n→∞ |z−z0 |=ε fn0 (z) |z−z0 |=ε f 0 (z)
which is our contradiction. Therefore, it follows that A is open and closed. Therefore, as Ω \ {w} is
connected we know that either f (z) ≡ f (w) on Ω \ {w} or f (z) 6= f (w) for any z ∈ Ω \ {w}. Therefore,
it follows that if f is not injective i.e. f (z0 ) = f (w) for some z0 6= w in Ω then f is identically f (w) in
Ω.

Problem 10. Let B be the vecor space defined as follows

ˆ ˆ
2 2
B := {u : C → C u holomorphic and |u(x + iy)|2 e−(x +y ) dxdy < ∞}
C
Show that B becomes complete when we introduce the norm
ˆ ˆ
2 2
||u||2 := |u(x + iy)|2 e−(x +y ) dxdy
C

Proof. Let 0 < r < R < ∞ then we claim that there is a C = C(r, R) such that for all entire functions f
that
||f ||L∞ (Br (0)) ≤ C||f ||
R+r
Indeed, notice that if δ := 2 then if z ∈ Br (0) then Bδ (z) ⊂ BR (0) and so we have by the mean value
property that
ˆ ˆ
1
u(z) = u(x + iy)dxdy
πδ 2 Bδ (z)
so we have ˆ ˆ ˆ ˆ
1 1
|u(z)| ≤ |u(x + iy)|dxdy ≤ |u(x + iy)|dxdy
πδ 2 Bδ (z) πδ 2 BR (0)
so Holder’s inequality gives
ˆ ˆ
R2
|u(z)|2 ≤ |u(x + iy)|2 dxdy
πδ 4 BR (0)

which implies since the right hand side is independent of z ∈ Br (0) that
ˆ ˆ ˆ ˆ
R2 R2 2
sup |u(z)|2 ≤ 4 |u(x + iy)|2 dxdy = 4 −R2 |u(x + iy)|2 e−R dxdy
z∈Br (0) πδ BR (0) πδ e BR (0)
ˆ ˆ
R2 2
+y 2 )
≤ |u(x + iy)|2 e−(x dxdy ≤ C(r, R)||u||2
πδ 4 e−R2 BR (0)
2
where C(r, R) = πδ4Re−R2 Now take R = 2r. This implies if {un } is Cauchy in B that we have local
uniform convergence since
sup |un (z) − um (z)| .r ||un − um || → 0 as n, m → ∞
z∈Br (0)

this implies there is an entire function u such that un → u locally uniformly. Then as {un } is cauchy, it
is a bounded sequence so Fatou’s lemma gives
ˆ ˆ ˆ ˆ
2 2 2 2
|u(x + iy)|2 e−(x +y ) dxdy ≤ lim inf |un (x + iy)|2 e−(x +y ) dxdy ≤ M < ∞
C n→∞ C
73

where M := supn ||un ||. That is u ∈ B. Now by local uniform convergence we know that along any
compact subset K ⊂ C that
ˆ ˆ
2 2
|u(x + iy) − un (x + iy)|2 e−(x +y ) dxdy → 0
K
And then observe that from the triangle inequality that
ˆ ˆ !1/2
2 −(x2 +y 2 )
|u(x + iy) − un (x + iy)| e dxdy ≤
C\K

ˆ ˆ !1/2 ˆ ˆ !1/2
2 −(x2 +y 2 ) 2 −(x2 +y 2 )
|um (x + iy) − un (x + iy)| e dxdy + |u(x + iy))| e dxdy
C\K C\K

The first term is small

√ due to the sequence being Cauchy and the last term is small when the compact
− x2 +y 2
set is big since ue ∈ L2 (C, dxdy). Therefore, ||u − un || → 0 and B is complete.
2 2
Alternatively It is also easy to see if we define dµ := e−(x +y ) dxdy to see that B is a closed subspace
of L2 (C, dµ) due to the L2 to interior L∞ estimate. And since {un } is cauchy in L2 (C, dµ) it converges
to some limit in L2 , which combined with pointwise convergence along a subsequence implies u is the
limit in L2 (dµ), which allows us to skip our estimates above to show ||u − un || → 0.

Problem 11. Let Ω ⊂ C be open, bounded, and simply connected and u a harmonic function on
Ω such that u ≥ 0. Show the following: for each compact set K ⊂ Ω there is a constant CK > 0
such that
sup u(z) ≤ Ck inf u(z)
z∈K z∈K

Proof. As Ω is simply connected open subset of C that is not all of C since Ω is bounded, we know by
the Riemann’s mapping theorem there is a conformal map ϕ : D → Ω. Therefore, v(z) := u ◦ ϕ is a
harmonic function on the disk. Hence, on any 0 < r < 1 we know by the Poisson Kernal Formula that
for z ∈ Br (0)
ˆ 2π 2 ˆ 2π
1 r − |z|2 iθ 1 r2 − |z|2
v(z) = v(re )dz = v(reiθ )dz
2π θ=0 |reiθ − z|2 2π θ=0 (reiθ − z)(re−iθ − z)
so the triangle inequality implies
ˆ 2π
1 (r − |z|)(r + |z|) r + |z|
v(z) = |v(z)| ≤ v(reiθ ) dθ = v(0)
2π θ=0 (r − |z|2 ) r − |z|
where we used the mean value property. And similarily, we also have
ˆ 2π
1 (r + |z|)(r − |z|) r − |z|
v(z) ≥ 2
v(reiθ )dθ = v(0)
2π 0 (r + |z|) r + |z|
i.e. for any z ∈ Br (0) we have
r − |z| r + |z|
v(0) ≤ v(z) ≤ v(0)
r + |z| r − |z|
so if |z| < r/2 this implies there are positive constants C1 = C1 (r) and C2 = C2 (r) such that
C1 v(0) ≤ v(z) ≤ C2 v(0)
Hence,
C1 v(0) ≤ inf v(z)
z∈Br/2 (0)

so it follows that
C2
sup v(z) ≤ inf v(z)
z∈Br/2 (0) C1 z∈Br/2 (0)
74 RAYMOND CHU

this immediately implies for any compact set K ⊂ D that

sup v(z) ≤ Ck inf v(z)
z∈K z∈K

Now as conformal maps map boundary to boundary imples if F ⊂ Ω is a compact set then there is an
1 > R > 0 such that F ⊂ ϕ(BR (0)) and taking K = BR (0) above gives
sup u ◦ ϕ(z) ≤ CR inf u ◦ ϕ(z)
z∈BR (0) z∈BR (0)

which implies since F ⊂ ϕ(BR (0)) that

sup u ≤ CF inf u
z∈F z∈F

as desired.

Problem 12. Let Ω := {z ∈ C : |z| > 1}. Suppose u : Ω → C is bounded and continuous on U
and is subharmonic on Ω. Prove the following: if u(z) ≤ 0 on |z| = 1 then u(z) ≤ 0 on Ω.

Proof. This is the standard -log trick. Indeed, fix ε > 0 and notice log |z| is Harmonic on Ω then define
uε (z) := u(z) − ε log |z|
and observe that as u is bounded that
lim uε (z) = −∞
|z|→∞
so we for R sufficiently large we have that the subharmonic function uε (z) ≤ 0 on ∂{1 ≤ |z| ≤ R} so
uε (z) ≤ 0 on {1 ≤ |z| ≤ R} by the maximum principle. So letting R → ∞ shows uε (z) ≤ 0 on Ω so
letting ε → 0 lets us conclude that u(z) ≤ 0 on Ω.

75

11. Spring 2015

Problem 1. Let f ∈ L1 (R). Show that

Xn2 ˆ (k+1)/n ˆ
lim f (x)dx = |f (x)|dx
n→∞ k/n R
k=−n2

Proof. We will show this problem first for the dense subclass Cc∞ (R). Indeed, if f ∈ Cc∞ (R) let K be a
compact set such that supp(f ) ⊂ K. Let Z := {x ∈ K : f (x) = 0} and notice that since f is continuous
SN
and K is compact that Z is compact. So there exists z1 , .., zN ∈ Z such that Z ⊂ j=1 B(1, zj ).

Now notice that if

ˆ (k+1)/n ˆ (k+1)/n
f (x)dx 6= |f (x)|dx
k/n k/n

then by continuity that f must change signs on [k/n, (k + 1)/n], so Z ∩ [k/n, (k + 1)/n] 6= ∅. Fix an n ∈ N
and let I be the index in −n2 ≤ k ≤ n2 where Z ∩ [k/n, (k + 1)/n] 6= ∅ and defining Ik := [k/n, (k + 1)/n].
Notice that this implies
[ N
[
Ik ⊂ B(2, zj )
k∈I j=1
so this means |I| ≤ 4N n since each B(2, zj ) covers at most 4n intervals of Ik since Ik has length 1/n and
there’s N of these balls. In particular, now we see

n2
X ˆ (k+1)/n ˆ n+1/n X ˆ ˆ
f (x)dx − |f (x)| = f (x)dx − |f (x)|dx
k=n2 k/n −n k∈I Ik Ik

so if ε > 0 is arbitrary, we can from uniform continuity find an N so large such that if n ≥ N then
|x − y| ≤ 1/n ⇒ |f (x) − f (y)| ≤ ε
Taking n to be sufficiently large, we see that for x ∈ Ik that |f (x)| ≤ ε. Therefore, we obtain since each
interval Ik is of length 1/n and height at most ε
X ε ε
≤ 2 = 2 |I| ≤ 8εN → 0
n n
k∈I
since ε is independent of N . So we have
Xn2 ˆ (k+1)/n ˆ n+1/n ˆ
lim f (x)dx = lim |f (x)|dx = |f (x)|dx
n→∞ k/n n→∞ −n R
k=−n2

Therefore, the problem is true for the dense subclass Cc∞ (R). Now by density we can find Cc∞ (R) 3 fn →
f in L1 (R). Then we have
n2
X ˆ (k+1)/n ˆ
f (x)dx − |f (x)|dx
k=−n2 k/n R

n
X
2 ˆ (k+1)/n n
X
2 ˆ (k+1)/n n2
X ˆ (k+1)/n ˆ
≤ fm (x)dx − f (x)dx + fm (x)dx − |fm (x)| + ...
k=−n2 k/n k=−n2 k/n k=−n2 k/n R
ˆ
+ |f (x)| − |fm (x)|
R
notice that the second term can be made small for large n and the third time is small for large m, so it
suffices to make the first term small. Indeed, observe the first term by the reverse triangle inequality
ˆ n+1/n ˆ
≤ |f (x) − fm (x)|dx ≤ |f (x) − fm (x)|
−n R
76 RAYMOND CHU

which can be made small for large m, which finishes the proof.

Remark: It is probably significantly easier to prove this statement first for the dense subclass of step
functions.

Problem 2. Let f ∈ L2loc (Rn ) and g ∈ L3loc (Rn ). Assume that for all real r ≥ 1, we have
ˆ ˆ
|f (x)|2 ≤ ra |g(x)|3 dx ≤ rb
r≤|x|≤2r r≤|x|≤2r

Here a, b ∈ R are such that 3a + 2b + n < 0. Show that f g ∈ L1 (Rn )

Proof. Notice that

ˆ ˆ ∞ ˆ
X
|f g|dx = |f g|dx + |f g|dx = (I) + (II)
Rn |x|≤1 m=0 2m ≤|x|≤2m+1

and Holder’s gives

∞
X
(II) ≤ ||f ||L2 (2m ≤|x|≤2m+1 ) ||g||L3 (2m ≤|x|≤2m+1 ) ||1||L6 (2m ≤|x|≤2m+1 )
m=0
∞
X
. 2m(a/2+b/3+n/6) < ∞
m=0
mn/6
where we used ||1|| .2
L6 (2m ≤|x|≤2m+1 ) since the volume of a sphere in Rn grows like rn and that
(3a + 2b + n)/6 < 0 to get the sum converges. And notice
ˆ
|f g|dx ≤ ||f ||L2 (|x|≤1) ||g||L3 (|x|≤1) ||1||L6 (|x|≤1) < ∞
|x|≤1

since f ∈ L2loc (Rn ) and g ∈ L3loc (Rn ), so it follows that

ˆ
|f g|dx < ∞
Rn

Problem 3. Let f ∈ L1loc (Rn ) and let

ˆ
1
M f (x) := sup |f (y)|dy
r>0 m(B(r, x)) B(r,x)
be the Hardy-Littlewood maximal function.
(1) Show that
ˆ
Cn
m({x : M f (x) > s}) ≤ |f (x)|dx, s>0
s |f |>s/2
where the constant Cn depends on n only. The Hardy-Littlewood maximal theorem may
be used.
(2) Prove that if ϕ ∈ C 1 (R), ϕ(0) = 0, and ϕ0 > 0 then
ˆ ˆ ˆ !
ϕ0 (t)
ϕ(M f (x))dx ≤ Cn |f (x)|
0<t<2|f (x)| t

Proof. Let us fix s > 0 and decompose

f (x) = f (x)χ|f (x)|≤s/2 + f (x)χ|f (x)|>s/2 := g(x) + h(x)
77

then notice that for any r > 0 we have

ˆ ˆ ˆ
1 1 1
|f (y)|dy ≤ |g(y)|dy + |h(y)|dy
m(B(r, x) B(r,x) m(B(r, x) B(r,x) m(B(r, x) B(r,x)
ˆ
s 1
≤ + |h(y)|dy
2 m(B(r, x) B(r,x)
since |g| > s/2. So if
ˆ
1
|f (y)|dy > s
m(B(r, x) B(r,x)

this implies that M h(x) > s/2, so in particular {x : M f (x) > s} ⊂ {x : M h(x) > s/2}. And by the
Hardy-Little wood theorem we have there is a Kn a constant that depends only on n such that
ˆ ˆ
2Kn Cn
m({x : M h(x) > s/2}) ≤ |h(x)|dx = |f (x)|dx
s R s |f |>s/2
by the definition of h where Cn = 2Kn . Thus we have
ˆ
Cn
m({x : M f (x) > s}) ≤ |f (x)|dx
s |f |>s/2

as desired.

For the second part, we have by the Fundamental Theorem of Calculus that and ϕ(0) = 0
ˆ M f (x)
ϕ(M f (x)) = ϕ0 (t)dt
0
so ˆ ˆ ˆ ˆ ˆ
0 0
ϕ(M f (x))dx ≤ ϕ (t)χ[0,M f (x)] (t)dtdx = ϕ (t) χ[0,M f (x)] dxdt By Tonelli

where Tonelli is justified by ϕ0 ≥ 0.

ˆ ∞ ˆ ∞ ˆ
0 ϕ0 (t)
= ϕ (t)m({x : M f (x) > t}) ≤ Cn |f (x)|dxdt
0 0 t |f |≥t/2

ˆ ∞ ˆ ˆ ∞ ˆ ˆ !
ϕ0 (t) ϕ0 (t) ϕ0 (t)
ˆ
= Cn |f (x)|χ|f |>t/2 dxdt = Cn |f (x)| χ|f |>t/2 dt dx = Cn |f (x)| dt dx
0 t 0 t 0<t<2|f (x)| t
as desired.

Problem 4. Let f ∈ L1loc (R) be 2π-periodic. Show that linear combinations of the translates
f (x − a), a ∈ R are dense in L1 ((0, 2π)) iff each Fourier coefficient of f is 6= 0.

Proof. Denote fˆ(n) as the nth Fourier coefficient of f .

⇒ Assume that {f (x − a)}a∈R is dense in L1 . Assume for the sake of contradiction that fˆ(n) = 0.
PN
For any {ai }N N
i=1 and {ci }i=1 let g(x) := i=1 ci f (x − ai ). Then observe by Parsavel’s identity that we
´ 2π P
have 0 u(x)v(x)dx = n∈N û(n)v̂(n) so
ˆ 2π
g(x)e−inx dx = ĝ(n) = 0
0

But this implies since functions of the form g(x) are dense in L1 that ||einx ||L2 ([0,2π]) = 0 but its not 0,
which is our contradiction.
78 RAYMOND CHU

⇐ Let S := span{f (x − a) : a ∈ R} then if S 6= L1 ((0, 2π)) we have by Hanh Banach the existence
of ` ∈ (L1 ((0, 2π))∗ ∼
= L∞ ((0, 2π)) such that `|S = 0 but ` is not the zero function. So in particular, we
have by Risez Representation Theorem a g ∈ L∞ ((0, 2π)) such that for any u(x) ∈ L1 ((0, 2π))
ˆ 2π
`(u) = u(x)g(x)dx
0
Then we have ˆ 2π
(f ∗ g)(a) = `(f (x − a)) = f (x − a)g(x)dx = 0
0
So we have that by taking the Fourier Series of (f ∗ g) and using uniqueness of the Fourier Series gives
fˆ(n)ĝ(n) = 0
and as fˆ(n) 6= 0 for any n we deduce that ĝ(n) = 0 so g ≡ 0 i.e. ` is the zero function, which is a
contradiction.

Problem 5. Let u ∈ L2 (R) and let us set

ˆ
2
U (x, ξ) := e−(x+iξ−y) /2 u(y)dy, x, ξ ∈ R

Show that U (x, ξ) is well defined on R2 and that there is a C > 0 such that for all u ∈ L2 (R), we
have ˆ ˆ ˆ
2
|U (x, ξ)|2 e−ξ dxdξ = C |u(y)|2 dy

Proof. Notice that by Cauchy-Schwarz that

ˆ 1/2
2
|U (x, ξ)| ≤ ||u||L2 (R) |e−(x+iξ−y) /2 2
| dy <∞
R
since the Gaussian is in L2 , so U is well defined. Now we will use Plancheral’s Theorem to prove the
second statement. Observe that
ˆ
−x2 /2−ixξ+ξ 2 /2 2
U (x, ξ) = e e−y /2+yx u(y)eiyξ dy
2
Define the Fx (y) := e−y /2+yx
u(y) then
2
/2−ixξ+ξ 2 /2
U (x, ξ) = e−x F̂x (−ξ)
so we have that ˆ ˆ ˆ ˆ
2 −ξ 2 2
|U (x, ξ)| e dxdξ = e−x |F̂x (−ξ)|2 dξdx
where we swapped integrals due to Tonelli since every term is non-negative. So Plancheral’s Theorem
gives ˆ ˆ ˆ ˆ
−x2 2 2
= e 2
Fx (y) dydx = e−x e−y +2yx |u(y)|2 dydx
ˆ ˆ
2
= |u(y)|2 e−(x−y) dxdy
ˆ
= C |u(y)|2 dy
´ 2 ´ 2 ´ 2
since e−(x−y) dx = e−(y) dy and C := e−y dy as desired.

Problem 6. When B1 and B2 are Banach spaces, we say that a linear operator T : B1 →
B2 is compact if for any bounded sequence (xn ) ∈ B1 , the sequence (T xn ) has a convergent
subseqeuence. Show that if T is compact then ImT has a dense countable subset.
79

Proof. First note that T is continuous. Indeed, if it was not then there is a sequence {xn } such that
||xn || = 1 and
||T (xn )|| ≥ n
but then {T (xn )} cannot converge along any sub-sequence, which is a contradiction to compactness. So
T is continuous.

Now we claim if Dr is a ball of radius r > 0 then T (Dr ) is pre-compact. Indeed, given any yn ∈ T (Dr )
we can find xn ∈ Dr such that ||xn || < r and T (xn ) = yn . By compactness, we can find a subsequence
xnk such that T (xnk ) converges to some limit y. So in particular, y ∈ T (Dr ), so T (Dr ) is pre-compact,
so T (Dr ) is totally bounded, which implies T (Dr ) is totally bounded, which implies its separable. So we
have [
T (B1 ) = T (Dn )
n∈N
and each T (Dn ) is separable, so the entire space is separable.

Problem 7. Let C+ := {z ∈ C : Im(z) > 0}. Suppose fn : D → C+ is a sequence of holomorphic

functions and fn (0) → 0 as n → ∞. Show that fn (z) → 0 uniformly on compact subsets of D.

Proof. Write fn = un + ivn where un and vn are the real and imaginary parts of fn respectively. Note
that vn ≥ 0 since fn maps to C+ . Therefore, by Harnack’s Inequality we have for any compact set the
existence of a constant C = C(K) that depends on the compact set K only such that

sup vn (z) ≤ C(K) inf vn (z) ≤ C(K)vn (0) → 0

z∈K z∈K

so it follows that as n → ∞, we have that vn (z) → 0 uniformly. Therefore, by Cauchy’s Estimate it

follows that ∇vn (z) → 0 uniformly on K, so by the Cauchy-Riemann equations we get that ∇un (z) → 0
uniformly on K. In particular, the Fundamental Theorem of Calculus then implies un → 0 uniformly on
K since limn→∞ un (0) = 0, so we have that fn → 0 uniformly on any compact subset as desired.

Problem 8. Let f : C → C be holomorphic and suppose

sup{|f (x)|2 + |f (ix)|2 } < ∞ and |f (z)| ≤ e|z| for all z ∈ C
x∈R

Deduce that f (z) is a constant.

Proof. This will follow from the Phragmen-Lindeolf method. We will prove on each of the four half
planes. Indeed, define R1 = {z ∈ C : Re(z) > 0, Im(z) > 0}. Then notice that for ε > 0 if we define
gε (z) := f (z)(exp(−ε(eiφ z)3/2 )) where z 3/2 is the branch with the negative real axis removed. Then
(eiφ z)3/2 ) = exp(3/2(log |z| + iArg(z + φ))) = |z|3/2 exp(3/2iArg(z + φ)). So we have
|(exp(−ε(eiφ z)3/2 ))| = exp(−ε|z|3/2 cos(3/2Arg(z + φ)))
Note that since z ∈ R1 we have
0 < Arg(z) < π/2 ⇒ 3/2φ < 3/2Arg(z + φ) < 3π/4 + 3φ/2
As we want 0 < cos(3/2Arg(z + φ)) we see we need
3/2φ > −π/2 and 3π/4 + 3φ/2 < π/2
For instance take φ = −π/4 gives the desired bound, so we have a δ > 0 such that δ < cos(3/2Arg(z −
π/4)). This implies
|gε (z)| ≤ |f (z)| exp(−ε|z|3/2 δ)
and we have that gε is bounded on ∂R1 and as |f (z)| ≤ e|z| it follows that |gε (z)| → 0 as |z| → ∞. So for
r > 0 large enough we have that |gε (z)| ≤ M := supx∈R {|f (x)|2 + |f (ix)|2 } < ∞ on ∂Br (0) ∩ R1 ∪ ∂R1 ,
80 RAYMOND CHU

so it follows from the maximum principle that on Br (0) ∩ R1 that |gε (z)| ≤ M and letting R → ∞ lets
us conclude
|gε (z)| ≤ M on R1
and letting ε → 0 using that gε (z) → f (z) lets us conclude that |f (z)| ≤ M on R1 . Repeating this
argument on the other 3 half planes lets us conclude that f is a bounded entire function, so it is constant.

Problem 9.

Problem 10. Determine ˆ ∞

dy
dy
−∞ (1 + y 2 )(1 + [x − y]2 )
for all x ∈ R. Justify all manipulations.

Proof. Note that this function has poles at z = ±i and z = x ± i. And that
1 1
Res(f, i) = lim =
z→i (z + i)(1 + [x − z]2 ) 2i(1 + [x − i]2 )
1 1
Res(f, x + i) = lim =
z→x+i (z + i)(z − i)(z − (x − i)) (x + 2i)(x)(2i)
Define γR := Reiθ for θ ∈ [0, π] and γ−R→R := −R(1 − t) + tR for t ∈ [0, 1] and γ(R) := γR + γ−R→R .
Then
ˆ ˆ π
dz Reiθ
= dθ =: I
γR (1 + z )(1 + [x − z]2 )
2
0 (1 + R2 e2iθ )(1+ [x − Reiθ ]2 )
and ˆ π
R
|I| ≤
0 (R2 − 1)(|1 − |x − Reiθ |2 )|
and notice that as R → ∞ that
1
→0
|1 − |x − Reiθ |2 |
since the denominator approahces ∞ as R → ∞ so we can find a C > 0 thanks to continuity to get
R
|I| ≤ 2πC →0
R2 −1
Therefore,
ˆ ˆ ∞
dz dy
lim =
R→∞ γ(R) (1 + z 2 )(1 + [x − z]2 ) −∞ (1 + y 2 )(1 + [x − y]2 )
and by the residue theorem we have that
ˆ
dz 1 1 1 1 2π
2 2
dz = π 2
+ = π( 2 + 2 )= 2
γ(R) (1 + z )(1 + [x − z] ) (1 + [x − i] ) (x + 2i)(x) x − 2ix x + 2ix x +4

Problem 11. Let Ω := D \ {0}. Prove that for every bounded harmonic function u : Ω → R
there is a harmonic function v : Ω → R obeying
∂u ∂v ∂u ∂v
= and =−
∂x ∂y ∂y ∂x
81

Proof. We will show that u is the real part of a holomorphic function. Define g(z) := ∂x u − i∂y u then
g is holomorphic on Ω since it is real differentiable in the real sense and we have that it satisfies the
Cauchy-Riemann equations. Indeed, we have
∂g 2 2 1 ∂g 2 2
= ∂xx u − i∂yx u and = −i∂yx u − ∂yy u
∂x i ∂y
and as u is harmonic we deduce that ∂x g = 1/i∂y g so g is holomorphic on Ω. Now let c := Res(g, 0) which
exists since g is holomorphic on the annulus {0 < |z| < 1}. Then we have from the Residue Theorem
that for any closed curve γ ⊂ D that ˆ
1
g(z) − c dz = 0
γ z
so it follows that h(z) := g(z) − c z1 has a holomorphic primitive f on D. In particular, it follows that
h(z) is holomorphic on D since h = f 0 (z). And notice that this implies
Re(f (z)) = u − c log |z|
so u−c log |z| is harmonic on D. So from the maximum modulus principle it follows that |u−c log |z|| ≤ C
where C is an upper bound for u since log |1| = 0 and this is a bound on ∂B1 (0). This implies c = 0 since
if c 6= 0 we have
lim |u − c log |z|| = ∞
z→0
since u is bounded. Therefore, u is the real part of the holomorphic function f (z), so it follows that it
has a harmonic conjugate, namely Im(f (z)).

Problem 12. Find all entire functions f : C → C that obey

f 0 (z)2 + f (z)2 = 1
Prove that your list is exhaustive.

Proof. Note that this implies

(f 0 + if )(f 0 − if ) = 1
so f 0 + if and f 0 − if omit the value 0, so there exists an entire function h(z) and g(z) such that
f 0 + if = exp(h(z)) and f 0 − if = exp(g(z)). From
(f 0 + if )(f 0 − if ) = 1 ⇒ g(z) = −h(z)
So we have that f 0 + if = exp(h(z)) and f 0 − if = exp(−h(z)). Therefore, we obtain that
exp(h(z)) − exp(−h(z))
f (z) = = sin(h(z)/i)
2i
plugging this into the ODE gives for w := h(z)/i that
− cos2 (w)(h0 (z))2 + sin2 (w) = 1 ⇒ − cos2 (w)(h0 (z))2 = cos2 (w) ⇒ (h0 (z))2 = −1
so we have
h0 (z) = ±i ⇒ h(z) = ±iz + c
where c ∈ C, so we have that
f (z) = sin(z + c) or sin(−z + c)
for any constant c ∈ C.
82 RAYMOND CHU

12. Fall 2015

Problem 1. Let gn be a sequence of measurable functions on Rd such that |gn (x)| ≤ 1 for all x
and assume that gn → 0 a.e. Let f ∈ L1 (Rd ). Show that the sequence
ˆ
f ∗ gn (x) := f (x − y)gn (y)dy → 0
Rd
d
uniformly on every compact subset of R .

Proof. We will show that when R > 0 is large enough that f ∗ gn uniformly converges to 0 on B R (0)
which will imply the claim. As f ∈ L1 (Rd ) we see that it is uniformly integrable. That is if ε > 0 then
there is a δ > 0 such that if m(E) < δ then
ˆ
|f (x)|dx ≤ ε
E
and by the translation invariance of the Lebesgue measure this implies
ˆ
|f (x − y)|dx ≤ ε
E

for all y. Now by Egorov’s theorm we can find a K ⊂ B R (0) with m(B R (0) \ K) < δ and gn → 0
uniformly on K. Then observe that if x ∈ B R (0) that we have
ˆ ˆ ˆ
|f ∗ gn |(x) ≤ |f (x − y)gn (y)|dy + |f (x − y)|dy + |f (x − y)|dy
K B R (0)\K Rd \B R (0)

As gn → 0 uniformly on K we can find an N such that if n ≥ N then ||gn ||L∞ (K) ≤ ε which implies
ˆ
≤ ε||f ||L1 (Rd ) + ε + |f (x − y)|dy
Rd \B R (0)
´ ´
and we have Rd \B R (0) |f (x − y)|dy = {|x−y|≥R} |f (y)|dy then if x ∈ Br (0) for 0 < r < R we have for
y ∈ {|x − y| ≥ R}
|y| ≥ |x − y| − |x| ≥ R − r = R − r
so we have {|x − y| ≥ R} ⊂ {|y| ≥ R − r}. Therefore,
ˆ ˆ
|f (y)|dy ≤ |f (y)|dy
{|x−y|≥R} |y|≥R−r

So as f ∈ L1 we can find an R > 0 such that if R > R then

ˆ
|f (x)|dx ≤ ε
|x|≥R

choosing such a large R implies for all x ∈ B r (0) we have

|f ∗ gn (x)| ≤ ε||f ||L1 + ε + ε = Cε
which implies uniform convergence on every compact subset.

Problem 2. Let f ∈ Lp (R), 1 < p < ∞, and let a ∈ R be such that a > 1 − 1/p. Show that the
series
X∞ ˆ n+n−a
|f (x + y)|dy
n=1 n
converges for a.e. x ∈ R

Proof 1
83

Proof. Notice that by Tonelli and Holder’s Inequality for sums that
∞ ˆ n+n−a ˆ 1 ∞ ˆ
∞ ∞
!1/q !1/p
X X X X 1
−a −a −aq −a p
|f (x+y)|dy = n |f (x+(n z+n))|dz ≤ n |f (x + n z + n| dz
n=1 n n=1 0 n=1 n=1 0

where 1/p + 1/q = 1. Notice as a > 1 − 1/p this implies −aq < −1 i.e. the sum to the left converges so
∞ ˆ 1
!1/p
X
−a p
=C |f (x + n z + n)| dz := g(x)
n=1 0

So now it suffices to show that if j ∈ Z that

ˆ j+1
|g(x)|p dx < ∞
j

Indeed, observe by Tonelli as all the integrand are positive that

ˆ j+1 ˆ j+1 X∞ ˆ 1 ∞ ˆ
X 1 ˆ j+1
−a
p
|g(x)| dx = K p
|f (x + n z + n)| dzdx = K |f (x + n−a z + n)|p dxdz
j j n=1 0 n=1 0 j

∞ ˆ 1
X ˆ j+1+n+n−a z ˆ ∞ ˆ j+1+n+n−a z
1X
=K |f (w)|p dwdz ≤ K |f (w)|p dwdz
n=1 0 j+n+n−a z 0 n=1 j+n+n−a z

Then notice that the intervals [j +n+n−a z, j+1+n+n−a z] and [j +n+1+(1+n)−a z, j+2+n+(1+n)−a z]
are disjoint, so we have
ˆ 1ˆ
≤K |f (w)|p = K||f ||pLp < ∞
0 R
so the sum is finite.
Proof 2
Proof. Fix k ∈ N then notice that it suffices to show that
ˆ k+1 X ∞ ˆ n+n−a
|f (x + y)|dy < ∞
k n=1 n

to get the desired claim. So Tonelli applies since all the theorems are non-negative to get
ˆ k+1 X∞ ˆ n+n−a ˆ k+1 X∞ ˆ
|f (x + y)|dydx = |f (x + y)|χ[n,n+n−a ] (y)dydx
k n=1 n k n=1 R
ˆ ∞ ˆ
k+1 X ˆ ∞ ˆ
k+1 X
= |f (z)|χ[n,n+n−a ] (z − x)dzdx = |f (z)|χ[n+x,n+x+n−a ] (z)dzdx
k n=1 R k n=1 R
ˆ ˆ ∞
k+1 X
= |f (z)| χ[n+x,n+x+n−a ] (z)dxdz
R k n=1
Now we claim that we have the bound
ˆ k+1 X∞
χ[n+x,n+x+n−a ] (z)dx . min(|z − k|−a , 1)
k n=1

Indeed, observe that

ˆ k+1 X∞ ˆ ∞
k+1 X ∞ ˆ
X k+1
χ[n+x,n+x+n−a ] (z)dx = χ[n+x,n+x+n−a ] (z)dx = χ[z−n−n−a ,z−n] (x)dx
k n=1 k n=1 n=1 k

Now we note that the integral is zero when k > z − n or k + 1 < z − n − n−a . So the region where the
intgeral is non-zero is contained in k + n ≤ z and (n + n−a ) ≥ z − (k + 1). Notice that n + n−a ≥ n so
the region where the integral is non-zero is contained in
n ∈ [z − (k + 1), z − k)]
84 RAYMOND CHU

so we have that
∞ ˆ
X k+1 z−k
X
χ[z−n−n−a ,z−n] (x)dx ≤ min{|n|−a , 1} ≤ min{|z − k|−a , 1}
n=1 k z−(k+1)

From which it follows that by Holder

ˆ ˆ k+1 X
∞
|f (z)| χ[n+x,n+x+n−a ] (x)dxdz ≤ ||f ||Lp || min{|z − k|−a , 1}||Lq
R k n=1

and aq > (1 − 1/p)q = (1/q)q = 1 where q is the Holder conjugate of p, so it follows that the right hand
side is finite.

Problem 3. Let f ∈ L1loc (Rd ) be such that for some 0 < p < 1, we have
ˆ
f (x)g(x)dx ≤ ||g||Lp

for all g ∈ C0 (Rd ). Show that f = 0 a.e.

Proof. The key is to notice that if a set E had measure m(E) = δ < 1 then m(E)1/p < m(E) so we
want to first plug in g to be the characteristic of a nice set say a cube and keep cutting the cube up into
smaller pieces which strengthens the bound to show f = 0 a.e.

Now let R be a rectangle unioned with its interior. So as R is closed, we know that χR is upper
semi-continuous, so it can be approximated from above by continuous functions. But as R is compact
we can make these approximations gn ∈ C0 (Rd ) and we can assume gn ≥ 0 with χR (x) = inf n≥1 gn (x).
Therefore, by the monotone convergence theorem since f1 ∈ L1 (Rd ) we have
ˆ ˆ
lim f (x)gn (x)dx = f (x)χR dx
n→∞

and for each n we have again by the monotone convergence theorem that
ˆ
f (x)gn (x)dx ≤ ||gn ||p → ||χR ||p

so for any rectangle R

ˆ
f (x)dx ≤ m(R)1/p
R
Now we decompose R into smaller rectangles. Indeed, fix an N ∈ N and cut R into equal 2N pieces with
each sub rectangle labeled Ri for 1 ≤ i ≤ 2N . Then we have
ˆ 2 ˆ
X
N
2N ˆ
X 2
X
N

f (x)dx = f (x)dx ≤ f (x)dx ≤ m(Ri )1/p = 2N ((m(R)/2N )1/p ) → 0

R i=1 Ri i=1 Ri i=1

as N → ∞ since 1/p > 1. Therefore, f integrates to zero on every rectangle, which implies since every
open subset of Rd is a countable union of rectangles that f integrates to zero on every open set. So in
particular, f ≡ 0 a.e.
Alternative Proof
Proof. As f ∈ L1loc (Rd ) we know that the lebesgue points of f are a set of full measure. Fix y ∈ E then
1
observe that if g(x) = m(B(r,y)) χB(r,y) (x) then there is a sequence of functions gn ∈ C0 (Rd ) such that
gn ≤ gn+1 ≤ g and gn → g pointwise (since g is lower semi continuous). Then
ˆ ˆ 1/p
f (x)gn (x)dx ≤ |gn (x)|p dx
Rd Rd
85

so the monotone convergence theorem gives

ˆ ˆ 1/p
p
lim f (x)gn (x)dx ≤ |g(x)| = (m(B(r, y))1/p−1
n→∞ Rd Rd

And note that |f (x)gn (x)| ≤ |f (x)g(x)| ∈ L1 (Rd ) so DCT gives us

ˆ ˆ
1
f (x)g(x)dx = f (x)dx ≤ (m(B(r, y))1/p−1
Rd m(B(r, y)) B(r,y)
Note that as 0 < p < 1 that 1/p − 1 > 0 so we obtain
ˆ
1
lim f (x)dx = 0
r→0 m(B(r, y)) B(r,y)

but as y ∈ E we obtain |f (y)| = 0, so f = 0 a.e.

Problem 4. Let H be a separable infinite-dimensional Hilbert space and assume that (en ) is
an orthonormal system in H. Let (fn ) be another orthnormal system that is complete, i.e. the
closure of the span of (fn ) is all of H.
P∞
(1) Show that if n=1 ||fn − enP ||2 < 1 then the orthonormal system (en ) is also complete.
∞
(2) Assume that we only have n=1 ||fn − en ||2 < ∞. Prove that it is still true that (en ) is
complete.

Proof. Let E := span(e1 , e2 , ...) then it suffices to show E ⊥ = {0} so fix any x ∈ E. Then as (fn ) is a
complete orthonormal system we know that
X∞
x= (x, fn )fn
n=1
P∞
so motivated by this, we define y := n=1 (x, fn )en . Now we know that x ⊥ y so

||x − y||2 = ||x||2 + ||y||2

but we also have
∞ ∞
!1/2 ∞
!1/2
X X X
2 2
||x − y|| ≤ ||(x, fn )(fn − en )|| ≤ |(x, fn )| ||fn − en ||
n=1 n=1 n=1
so in particular, we have
∞
!
X
||x − y||2 ≤ |(x, fn )|2 = ||x||2
n=1
where we used Cauchy-Schwarz and that (fn ) is a complete orthonormal system. In particular, this
implies ||y||2 = 0, so by Bessels’ Inequality we conclude |(x, fn )| = 0 for all n, which means x = 0 i.e.
E ⊥ = {0}.

For (2) we define EN := span{eN , eN +1 , ..., } and FN := span{fN , fN +1 , ..., }. We know that we have
⊥
H = EN ⊕ EN
⊥ ⊥
We also know that {e1 , ..., eN } ⊂ EN so it suffices to show there there is some N such that dim(EN )≤N
⊥
to conclude that {ei } are a complete orthonormal system. First we show that EN is finite dimensional
when N is large enough.

Indeed, for any closed subspace V ⊂ H define πV to be the orthogonal projection operator onto V .
Then we have
∞
X
||πEN (x) − πFN (x)|| = || (x, en )en − (x, fn )fn ||
n=N
86 RAYMOND CHU

∞
X ∞
X ∞
X
= || (x, fn )en − (x, fn )fn + (x, en − fn )en || ≤ |(x, fn )|(||en − fn ||) + || (x, en − fn )fn ||
n=N n=N n=N
∞
X X∞
≤ |(x, fn )|(||en − fn ||) + ( ||(x, en − fn )||2 )1/2
n=N n=N
∞
!1/2 ∞
!1/2 ∞
X X X
≤ |(x, fn )|2 ||en − fn ||2 + ||x||( ||en − fn ||2 )1/2 → 0
n=N n=N n=N
so we see that the operator norm of πEN − πFN converges to 0 as N → ∞. Note that in the above
inequalities we used Cauchy Schwarz and Pythagerous Theorem for sums.

Note that we also have

H = FN ⊕ FN⊥ and H = EN ⊕ EN ⊥

⊥
where dim(EN ) = N since {fn } is complete and as πFN⊥ = id − πFN and πEN + πEN⊥ = id which implies
we can choose an N so large such that
||πFN⊥ − πEN⊥ ||op = ||πFN − πEN ||op < 1/2
Now we claim this implies dim(FN⊥ ) ≤ dim(EN ⊥
) = N which lets us conclude. Indeed, observe that for
N +1 ⊥
PN +1 ⊥
any N + 1 vectors {xi }i=1 in FN there is some α1 , .., αN +1 such that πEN⊥ ( i=1 αi xi ) = 0 since EN
has dimension N . But then
N +1 N +1 N +1 N +1
1 X X X X
|| αi xi || ≥ ||πF⊥N ( ⊥
αi xi ) − πE ( αi x i )|| = || αi xi ||
2 i=1 i=1
N
i=1 i=1
PN +1
i.e. i=1 αi xi = 0 so FN⊥ has dimension at most N linearly independent vectors which lets us conclude
the problem since then {f1 , ..., fN } is a basis for FN⊥ and the closure of {fN +1 , ...} is a basis of FN . Hence,
the closure of {f1 , f2 , ...} is a basis of H so it is also complete.

Problem 5. Show that the Holder continuous functions form a set of first category (a meager
set) in C([0, 1]).

Proof. Use C α to denote the space of α Holder continuous functions. Then if β < α and f ∈ C α we have
|f (x) − f (y)| |f (x) − f (y)| |f (x) − f (y)|
= ≤C⇒ ≤ C|x − y|α−β < K
|x − y|α |x − y|β+(α−β) |x − y|β
since α − β > 0 and x, y ∈ [0, 1], so we have f ∈ C β . Therefore, the space of Holder continuous functions
can be written as
∞
[
C 1/n ([0, 1])
n=1
Then notice that
∞
[ ∞
[
1/n n
C ([0, 1]) = {f : ||f ||C 1/n ≤ M } := EM
M =0 M =0
|f (x)−f (y)| n
where ||f ||C α := ||f ||L∞ + supx6=y |x−y|α . And then we claim each EM is closed. Indeed,
|f (x) − f (y)| |f (x) − fn (x)| |fn (x) − fn (y)| |f (y) − fn (y)|
||f ||L∞ + ≤ ||fn − f ||L∞ + ||fn ||L∞ + + +
|x − y|α |x − y|α |x − y|α |x − y|α
n
which can be made arbitrarily smaller than M thanks to uniform convergence. So EM is closed. Therefore,
M 1/(2n) M
if f ∈ En we claim that f + ε|x| ∈
/ En . Indeed, observe at x = 0 that for any x > 0
f (x) − f (0) + ε|x|1/(2n)
→ ∞ as x → 0
|x|1/n
since |f (x) − f (0)|/|x|1/n is bounded and ε|x|−1/n → ∞. Therefore, EnM has empty interior, so the space
of holder continuous functions is meager.
87

Problem 6. Let u ∈ L2 (Rd ) and let us say u ∈ H 1/2 (Rd ) if

(1 + |ξ|1/2 )û(ξ) ∈ L2 (Rd )
Show that u ∈ H 1/2 (Rd ) iff
ˆ ˆ
|u(x + y) − u(y)|2
dydx < ∞
|y|d+1

Proof. Notice by Plancheral and Tonelli that that

ˆ ˆ ˆ ˆ
|u(x + y) − u(y)|2 2 |1 − e2πiξ·y |2
d+1
dydx = |û(ξ)| dydξ
|y| |y|d+1
So it suffices to show there are C1 , C2 > 0 such that
ˆ
|1 − e2πiξ·y |2
C1 (1 + |ξ|1/2 )2 ≤ dy ≤ C2 (1 + |ξ|1/2 )2
|y|d+1
But as u ∈ L2 (Rd ) we know that û ∈ L2 (Rd ), so it suffices to prove
ˆ
|1 − e2πiξ·y |2
C1 |ξ| ≤ dy ≤ C2 |ξ|
|y|d+1
For the upper bound observe that
ˆ ˆ ˆ
|1 − e2πiξ·y |2 |1 − e2πiξ·y |2 |1 − e2πiξ·y |2
dy = dy + dy
|y|d+1 |yξ|≤1 |y|d+1 |yξ|≥1 |y|d+1
´x
Using ex − 1 = 0 ex dx ⇒ |1 − ex | ≤ |x||ex | so
ˆ ˆ
4π 2 |ξ|2 |y|2 2|ξ|d+1
≤ +
|y·ξ|≤1 |y|d+1 |y·ξ|≥1 |y/ξ|
d+1

. |ξ| + |ξ| = |ξ|

Now for the lower bound observe that, for any fixed ξ that there is an orthogonal matrix A such that
A(ξ/|ξ|) = en = (0, 0, ..., 1). Also the FTC also tells us |1 − ex | ≥ |x|.So we have
ˆ ˆ ˆ
|1 − e2πiξ·y |2 4π 2 |ξ · y|2 4π 2 |ξ|2 |y · en |2
dy ≥ dy =
|y|d+1 |y|d+1 |y|d+1
ˆ ˆ
|ξ|2 |y · en |2 |ξ|2
≥ 4π 2 ≥ π 2
|ξ| 2
= Cn |ξ|
|yξ|≤1∩(|y·en |≥1/2||y||) |y|d+1 |yξ|≤1∩|y·en |≥1/2||y|| |y|
d−1

so we have the desired result

Problem 7. Assume that f (z) is analytic in {z : |z| < 1} and continuous on {z : |z| ≤ 1}. If
f (z) = f (1/z) when |z| = 1 , prove that f is constant.

Proof. Note that f (1/z) is analytic on {z : |z| ≥ 1} and extends continuously to f (z) on |z| = 1.
Therefore, by Morrera’s Theorem we conclude that
(
f (z) if z ∈ {|z| ≤ 1}
g(z) :=
f (1/z) else

is an entire function. However, as f extends continuously to {|z| = 1} f (z) is bounded on D, which

implies f (1/z) is bounded on C \ D. Therefore, g(z) is a bounded entire function, so by Liouville it is
constant which implies f is constant.

88 RAYMOND CHU

Problem 8. Assume that f (z) is an entire function that is 2π-periodic in the sese that f (z+2π) =
f (z) and
|f (x + iy)| ≤ Ceα|y|
for some C > 0 and 0 < α < 1. Prove that f is constant.

Proof. Note that as eα|x| > 1 this implies that

|f (x + iy)| ≤ Ceα|y| eα|x| = Ceα(|x|+|y|)
that is f is an entire function of order 0 < α < 1. In particular, we also have g(z) := f (z) − f (0) is an
entire function of order α that is 2π-periodic. So by periodicity we have that if we denote Z as the set of
zeros of g that
{2nπ} ⊂ Z
Therefore, the zeros of g grow at least linearly. However, we know that if f is not identically 0 then by
Jensen’s formula that for large enough R that the number of zeros of g on BR (0) should be bounded by
C|R|α where α < 1, but our zeros grow at least linearly, so which means g must be the zero function i.e.
f is constant.

Problem 9. Let (fj ) be a sequence of entire functions such that

ˆ ˆ
2
|fj (z)|2 e−|z| dxdy ≤ C
C
for some constant C > 0. Show that there is a sub-sequence {fjj } and an entire function f such
that ˆ ˆ
2
|fj (z) − f (z)|2 e−2|z| dxdy → 0
C

Proof. We first claim that there is an entire function f such that fn → f uniformly on every compact
subset of C. Indeed, fix an R > 0 then we have for any z0 ∈ BR (0) that for any 0 < r < R
ˆ 2π
1
fj (z0 ) = fj (z0 + reiθ )dθ
2π θ=0
so we have ˆ R ˆ R ˆ 2π
1
rfj (z)dr = fj (z0 + reiθ )rdθdr
2π
r=0
ˆr=0ˆ θ=0
1
fj (z0 ) = fj (z)dxdy
πR2 BR (z0 )
so Holders shows
ˆ ˆ ˆ ˆ ˆ ˆ
1 1 C(R) 2
|fj (z)| ≤ √ √ |fj (z)|2 ≤ √ √ |fj (z)|2 ≤ √ √ |fj (z)|2 e−|z|
π R BR (z0 ) π R B2R (0) π R C

which implies by the given assumptions that {fj (z)} is a uniformly bounded family on every compact
subset of C. Which implies by Montel’s theorem that on every compact subset we have a uniformly
convergent sub-sequence. By taking a diagonal sub-sequence we can find a sub-sequence {fjk (z)} that
uniformly converges to some function f on every Bn (0) where n ∈ N. This implies thanks to the Morrera’s
Theorem that f is holomorphic on every Bn (0) so it is an entire function. And uniform convergence on
every compact subset implies that
ˆ ˆ ˆ ˆ
2 −|z|2 2
C≥ |fjk (z)| e χBn (0) dxdy → |f (z)|2 e−|z| χBn (0) dxdy
C C
so it follows from Monotone Convergence Theorem that
ˆ ˆ ˆ ˆ
2 2
C ≥ lim |f (z)|2 e−|z| χBn (0) dxdy = |f (z)|2 e−|z| dxdy
n→∞ C C
89

Then we have for any R > 0 that

ˆ ˆ ˆ ˆ ˆ ˆ
2 −2|z|2 2 −2|z|2 2
|fjk (z)−f (z)| e dxdy = |fjk (z)−f (z)| e dxdy+ |fjk (z)−f (z)|2 e−2|z| dxdy
C BR (0) C\BR (0)

and
ˆ ˆ ˆ ˆ
2 2
|fjk (z) − f (z)|2 e−2|z| dxdy ≤ exp(−R2 ) |fjk (z) − f (z)|2 e−|z| dxdy ≤ K exp(−R2 )
C\BR (0) C\BR (0)
´´
thanks to our previous computation. Therefore, if ε > 0 we can find an R so large such that C\BR (0)
|fjk (z)−
2 −2|z|2
f (z)| e dxdy ≤ ε/2. And by uniform convergence on BR (0) we can choose a K so large such that
for any k ≥ K we have
ˆ ˆ
2
|fjk (z) − f (z)|2 e−2|z| dxdy ≤ ε/2
C\BR (0)

This means for any k ≥ K we have

ˆ ˆ
2
|fjk (z) − f (z)|2 e−2|z| dxdy ≤ ε
C

so we have the desired result.

Problem 10. Use the residue theorem to prove that

ˆ ∞
dx π
ecos(x) sin(sin(x)) = (e − 1)
0 x 2
Use a large semi-circle as part of the contour.

Proof. Fix R > 0 large and r > 0 small. Let γR be the semi-circle centered at the origin of radius R
i.e. γR = Reiθ where θ ∈ [0, π] and γ−R→−r be the line from z = −R to z = −r with counterclock wise
orientation, and similarly for γr→R . Let γr be the semi circle of radius r starting at −r and ending at r
ix
i.e. γr = reiθ for θ ∈ [π, 2π]. Also notice that exp(cos(x)) sin(sin(x)) = Im(ee ) so
ˆ ∞ ˆ ∞
ix dx ix dx
exp(cos(x)) sin(sin(x)) = Im(ee ) = Im exp(cos(x)) sin(sin(x)) = Im(ee )
0 x 0 x
iz
Let γ := γR +−R→−r +γr + γr→R then we know by the Residue Theorem since ee /z has a residue of e
at z = 0 that
ˆ
iz dz
ee = 2πie
γ z
and
ˆ ˆ π
iz dz
ee = i exp(exp(−R sin(θ))[cos(R cos(θ)) + i sin(R cos(θ))]dθ → iπ
γR z θ=0

where the last convergence is due to the dominated convergence theorem. And we also have
ˆ
iz dz
ee → iπe
γr z
and notice on the real line our integrand is even, so we obtain by the Residue theorem that
ˆ ∞ ˆ ∞
dx dx π
2 ecos(x) sin(sin(x)) + πe + π = 2πe ⇒ ecos(x) sin(sin(x)) = (e − 1)
0 x 0 x 2

90 RAYMOND CHU

Problem 11. Let Ω := {(x, y) ∈ R2 : x, y > 0} and let u be subharmonic on Ω, continuous on Ω

such that
u(x, y) ≤ |x + iy|
for large (x, y) ∈ Ω. Assume that
u(x, 0) ≤ ax u(0, y) ≤ by x, y ≥ 0
for some a, b > 0. Show that
u(x, y) ≤ ax + by for (x, y) ∈ Ω

Proof. This is a standard application of the Phragmén–Lindelöf method. In general, in the sector {z :
π
α ≤ Arg(z) ≤ β} we should have for any 0 < k < β−α that |z|k is a barrier function. Note for our domain
α = 0, β = π/2 so we should have for any 0 < k < 2 that |z|k is a barrier. We want our barrier to grow
much faster than u at infinity, so we want 1 < k < 2 say k = 3/2. Now we notice that for any φ ∈ [0, 2π]
that
3
Re(eiφ z 3/2 ) = |z|3/2 cos( Arg(z) + φ)
2
now we want to choose φ such that − π2 < 23 Arg(z) + φ < π2 to make the phase term bounded above and
below by a positive constant and recall 0 ≤ Arg(z) ≤ π2 so we take φ = − 3π 8 then we have the desired
3/2 3 3π
bounds. Therefore, φ(z) := |z| cos( 2 Arg(z) − 8 ) is harmonic so we have that for any ε > 0
v(x, y) := u(x, y) − ax − by − εφ(z)
is subharmonic and we have
3 3π
v(x, 0) ≤ −by − ε|x|3/2 cos( − )≤0
2 8
and
3π 3π
v(0, y) ≤ −ax − ε|y|3/2 cos(
− )≤0
4 8
Then there exists an R(ε) > 0 such that for any r ≥ R(ε) that
v(x, y) ≤ 0 on Ω ∩ ∂Br (0)
since u − ax − by grows at most linearly and our barrier function is super linear. This implies by the
maximum principle that on Ω ∩ Br (0) that we have
v(x, y) ≤ 0 on Br (0) ∩ Ω
and let r → ∞ to conclude for any ε > 0 that
u(x, y) − ax − by − εφ(z) ≤ 0 on Ω
let ε → 0 to conclude
u(x, y) ≤ ax + by

Problem 12. Find a function u(x, y) harmonic in the region between the circles |z| = 2 and
|z − 1| = 1 which equals 1 on the outer circle and 0 on the inner circle (except at the point where
the two circles are tangent to one another).

Proof. Note that the two circles are tangent at z = 2. We want to map this conformally onto a strip
and solve the problem there then invert back. We recall that Mobius Transformations map circles to
generalized circles i.e. circles and lines, so we choose a mobius transformation such that 2 is sent to infinity,
1
to make the circles become lines. Indeed, consider φ(z) := z−2 then this is a mobius transformation and
1 1 1 i
φ(eiθ ) = ⇒ φ(eiπ ) = − and φ(eiπ/2 ) = − −
2eiθ − 2 4 4 4
91

so |z| = 2 gets mapped to the line Re(z) = −1/4. And similarly φ(0) = −1/2 and φ(1 + i) = −1/2 − i/2
so |z − 1| gets mapped to the line Re(z) = −1/2. And as φ is a continuous map on the interior,
we know that it maps Ω to a connected set with boundary Re(z) = −1/4 and Re(z) = −1/2, so
φ(Ω) = {z : −1/2 < Re(z) < −1/4}. Then we want to solve the Problem

∆u = 0 on φ(Ω)

u = 1 on Re(z) = −1/2

u = 0 on Re(z) = −1/4


so we make the guess u = ax + by + c for constants a, b, c. Then the PDE becomes solving a 2 × 2 matrix,
which implies that a = −4, c = −1, b = 0 so u(x + iy) = −4x − 1 = Re(−4z − 1) solves that PDE. So
u ◦ φ = Re(−4/(z − 2) − 1) is the desired harmonic function.

92 RAYMOND CHU

13. Spring 2016

Problem 1. Let 2
Kt (x) = (4πt)−3/2 e−|x| /4t
, x ∈ R3 , t > 0
where |x| is the Euclidan norm of x ∈ R3 .
(1) Show that the linear map
L3 (R3 ) 3 f 7→ t1/2 Kt ∗ f ∈ L∞ (R3 )
is bounded, uniformly in t > 0.
(2) Prove that t1/2 ||Kt ∗ f ||L∞ → 0 as t → 0 for f ∈ L3 (R3 ).

Proof. Notice that by Holder’s Inequality that

|Kt ∗ f | ≤ ||f ||L3 (R3 ) ||Kt ||L3/2 (R3 )
and ˆ ˆ ˆ
2
|Kt |3/2 dx = C t−9/4 exp(−3|x|2 /8)dx = C̃ t−3/4 e−|u| du = M t−3/4
R3 R3 R3
by the change of coordinates. So we have
||Kt ||L3/2 = M̃ t−1/2
So we have
t1/2 |Kt ∗ f | ≤ M̃ ||f ||L3
i.e. this functional is uniformly bounded in t.

For the second part, it suffices by the uniform boundness of the operator in t to show that the result
is true for a dense subclass of L3 . In particular, it suffices to show it for simple functions and as the
operator is linear, it suffices to show it for characteristics of measurable sets. Indeed, observe that
ˆ ˆ
1 −|x−y|2 /4t 1 2
t1/2 (Kt ∗ χE ) = e dy ≤ e−|x| /4t dx
(4π)3/2 t E (4π)3/2 t R3
ˆ
t1/2 2
= e−|x| /4 dx = Ct1/2 → 0 as t → 0
(4π)3/2 R3
so t1/2 (Kt ∗ f ) → 0 is true for the dense subclass of simple functions since (Kt ∗ χE ) ≥ 0, so it is true by
uniform boundness in L3 for functions in L3 .

Alternative Proof Of Second Part: Heat Kernal Approach For the second part, notice that it
suffices to prove the claim for Cc∞ (R3 ) since its a dense subclass of L3 because we have if g ∈ Cc∞ (R3 )
and f ∈ L3 then
t1/2 ||Kt ∗ f ||L∞ ≤ t1/2 ||Kt ∗ (f − g)||L∞ + t1/2 ||Kt ∗ g||L∞
≤ C||f − g||L3 + t1/2 ||Kt ∗ g||L∞
and the first term can be made small by using density of test functions on L3 . Now we claim the following
lemma: If g ∈ Cc∞ (R3 ) then as t → 0 we have Kt ∗ g(x) → g(x) uniformly. This implies the problem
since then we have t1/2 ||Kt ∗ g||L∞ → 0 due to uniform convergence. Now observe
ˆ
Kt (x)dx = 1
R3
then ˆ ˆ
|Kt ∗ g(x) − g(x)| ≤ Kt (x − y)|g(x) − g(y)|dy + Kt (x − y)|g(x) − g(y)|
Bδ (x) |x−y|≥δ
by uniform continuity of g we can choose δ > 0 so small such that if ε > 0 is given then
ˆ
|g(x) − g(y)| ≤ ε ⇒ Kt (x − y)|g(x) − g(y)|dy ≤ ε
Bδ (x)
93

since Kt ≥ 0 and has mass 1. Now for the second integral, we have if M = ||g||L∞ then
ˆ ˆ
Kt (x − y)|g(x) − g(y)| ≤ 2M Kt (x)dx
|x−y|≥δ |x|≥δ
1 3
´
As Kt ∈ L (R ) we can find a compact set K such that x∈K
/
Kt (x) < ε so
ˆ ˆ
2M Kt (x)dx ≤ 2M Kt (x) + 2M ε
|x|≥δ |x|≥δ∩K

Then observe on |x| ≥ δ we have

2
Kt (x) ≤ Ct−3/2 e−δ /4t

so ˆ
2
Kt (x)dx ≤ Ct−3/2 e−δ /4t
m(K)
|x|≥δ
and we know that as t → 0 this term goes to 0 since exponential decays much faster than polynomials
grow, so we obtain if 0 < t 1
|Kt ∗ g(x) − g(x)| ≤ ε + ε + 2M ε
and this bound is independent of x, so we conclude Kt ∗ g → g uniformly as desired.

Problem 2. Let f ∈ L1 (R). Show that the series

∞
X 1 √
√ f (x − n)
n=1
n
converges absolutely for almost all x.

Proof. Define F (x) := f (−x) then it suffices to show that the series
∞ ∞
X 1 √ X 1 √
√ F (x + n) = √ f (−x − n)
n=1
n n=1
n
converges for a.e. x. Indeed, fix a k ∈ N then WLOG by replacing F with |F | if necessary we can assume
that F ≥ 0 to see
ˆ k+1 X∞ ∞ ˆ k+1 ∞ ˆ k+1+ n
√
1 √ X 1 √ X 1
√ F (x + n) = √ F (x + n) = √
√ F (x)dx
x=k n=1 n n=1
n x=k n=1 k+ n
n
where the interswap of derivative is justified by Tonellis since F ≥ 0. Now observe
∞ j2 ˆ k+1+√n
X X 1
= √
√ F (x)dx
j=2 2 k+ n n
n=(j−1)

and
j2 ˆ √
k+1+ n j2 ˆ k+1+j ˆ k+1+j ˆ k+1+j
X 1 X F (x) (2j − 1)
√
√ F (x)dx ≤ dx = F (x) ≤ C F (x)dx
k+ n n k+j−1 j−1 j−1 k+j−1 k+j−1
n=(j−1)2 n=(j−1)2

since (2x − 1)/(x − 1) is bounded on [2, ∞). Therefore,

∞
X ˆ k+1+j ˆ
≤ C F (x)dx ≤ 2C F (x)dx
j=2 k+j−1 R
P∞
since each interval overlaps at most twice. Therefore, for a.e. x ∈ [k, k + 1] we know that n=1 √1n F (x +
√
n) < ∞. Then as these sets are a countable partiton of R, it follows that for a.e. x ∈ R the sum is
absolutely convergent.

94 RAYMOND CHU

Problem 3. Let f ∈ L1loc (R) be real valued and assume for all integers n > 0, we have that
f (x + 1/n) ≥ f (x)
for a.e. x ∈ R. Show that for each real number a ≥ 0 we have
f (x + a) ≥ f (x)
for almost all x ∈ R

Proof. For m, n ∈ N let

Xm,n := {x ∈ R : f (x + m/n) < f (x + (m − 1)/n)}
c
S T
then m,n∈N Xm,n is a null set by the give assumptions. Therefore, X := m,n∈N Xm,n is a set of full
measure i.e. for almost every x ∈ R we have
f (x + m/n) ≥ f (x)
Then write E to be the Lebesgue points of f , and we have m(E c ) = 0 and let Y := E ∩ X then this is
also a set of full measure. Finally if g(x) := f (x + a) let E2 be the Lebesgue points of g then this also a
set of full measure, so W := E ∩ X ∩ E2 is of full measure. And if x ∈ W then we have that
ˆ x+a+r ˆ x+r
1 1
f (x + a) = lim f (y)dy and f (x) = lim f (y)dy
r→0 2r x+a−r r→0 2r x−r

and as x ∈ W we know for any positive rational q that

f (x + q) ≥ f (x)
Now fix ε > 0 and a positive rational such that 0 < a − q < δ [we can assume a 6= 0 otherwise this is
trivial] then we have
ˆ x+a+r ˆ x+q+r ˆ x+a−r ˆ x+a+r
1 1 1 1
f (y)dy − f (y)dy = − f (y)dy + f (y)dy
2r x+a−r 2r x+q−r 2r x+q−r 2r x+q+r
Then as f ∈ L1loc it is locally uniformly integrable, so we can make the above two integrals smaller than
ε if δ > 0 is sufficiently small i.e. choose δ > 0 so small such that
ˆ x+a+r ˆ x+q+r
1 1
f (y)dy − f (y)dy ≤ ε
2r x+a−r 2r x+q−r
Therefore, as
ˆ x+q+r ˆ x+r ˆ x+r
1 1 1
f (y)dy − f (y)dy = f (y + q) − f (y)dy ≥ 0
2r x+q−r 2r x−r 2r x−r

where for the above inequality we used that the set where f (y + q) ≥ f (y) is of full measure. So we have
ˆ x+a+r ˆ x+r
1 1
f (y)dy − f (y) ≥ −ε
2r x+a−r 2r x−r
and letting ε → 0 and r → 0 using these points are Lebesgue points yields
f (x + a) ≥ f (x)
and as x ∈ W is of full measure we are done.

Problem 4. Let V1 be a finite-dimensional subspace of a Banach Space V . Show that there

exists a continuous projection map P : V → V1 .
95

Proof. Let {e1 , .., en } be a basis of V1 then define on S := span(e1 , ..., en )

n
X
fj (x) = fj ( ai ei ) = aj for 1 ≤ j ≤ n
i=1
And as fj is a linear functional on a finite dimensional vector space, it is automatically continuous. By
Hanh-Banach, we can extend fj to a continuous linear functional on V . Then write
n
X
P (x) := fj (x)ej
i=1
then we have Im(P ) ⊂ V1 . Also we have
n
X n
X
P (x) = αj ej ⇒ P 2 (x) = αj ej
i=1 i=1
where αj = fj (x). We also have Im(P ) = V1 by using P (x) is the identity on V1 . So P (x) is a projection
map.

Problem 5. For f ∈ C0∞ (R2 ) define u(x, t) by

ˆ
sin(t|ξ|)
u(x, t) = eix·ξ f (ξ)dξ, x ∈ R2 , t > 0
R 2 |ξ|
Show that limt→∞ ||u(·, t)||L2 = ∞ for a set of f that is dense in L2 (R).

Proof. Define ft (x) := sin(t|x|)/|x|f (x) then

ˆ
u(x, t) = fˆt (x) ⇒ ||u||L2 = ||fˆt (x)||L2 = ||ft ||L2
ˆ
where we used Plancheral’s and that fˆ(x) = f (−x). So we have
ˆ 2
2 sin(t|ξ|)
||u||L2 = |f (ξ)|2 dξ
R2 |ξ|
Notice that as
| sin(|x|)|
lim =1
(x,y)→(0,0) |x|
that we can find a δ > 0 such that on Bδ (0) we have
1
| sin(|x|)| ≥ |x|
2
this implies on Bδ/t (0) that
| sin(t|x|)| ≥ t/2|x|
so this gives us
ˆ 2 ˆ
sin(t|ξ|) t2
|f (ξ)|2 dξ ≥ |f (ξ)|2 & essinfξ∈Bδ/t (0) |f (ξ)|2
R2 |ξ| Bδ/t (0) 4

Therefore, if lim inf ε→0 essinfξ∈Bε (0) |f (ξ)| = ∞ then we will have
lim ||u(·, t)||L2 = ∞
t→∞

and notice that such f ∈ L2 is dense since for any g ∈ L2 such that lim inf ε→0 essinfξ∈Bε (0) |g(ξ)| is finite
we can consider g + ε|x|−1/2 χB1 (0) ∈ L2 .

Problem 6. Suppose that {φn } is an orthonormal system of continuous functions in L2 ([0, 1])
and let S be the closure of the span of {φn }. If supf ∈S\{0} ||f ||∞ /||f ||2 is finite show that S is
finite dimensional.
96 RAYMOND CHU

Proof. Let K > 0 be such that

||f ||∞ ≤ K||f ||2
for f ∈ S. Then notice that S ⊂ C([0, 1]) where f is endowed with the L2 norm since if fn → f in L2 we
have
||fn − f ||∞ ≤ K||fn − f ||2 → 0
and each fn ∈ S can be approximated arbitrarily well with finite linear combinations of φn ∈ C([0, 1])
which implies that f is continuous due to uniform convergence.

Now observe that this implies the evaluation functional for y ∈ [0, 1]
Ly (f ) = f (y)
is continuous in this norm since
|Ly (f )| ≤ ||f ||∞ ≤ K||f ||2
so by Risez Representation Theorem since S is a Hilbert Space, we know that there is a gx such that
ˆ 1
(f, gx ) = f (y)gx (y)dy = f (x)
0
Now observe that
N
X N
X
|φn (y)|2 = |(φn , gy )|2 ≤ ||gy ||L2
n=1 n=1
where the last inequality is due to Bessel’s Inequality since φn are orthonormal. But we have that
||gy ||2L2 = (gy , gy ) = g(y) ≤ ||f ||∞ ≤ K||fy ||L2
i.e.
||gy ||L2 ≤ K
so we obtain
N
X
|φn (y)|2 ≤ K 2
n=1
so integrating and using φn are orthonormal implies
N ≤ K2
therefore, we can have at most bK 2 c orthonormal vectors i.e. S is finite dimensional.

Problem 7. Determine ˆ ∞
xa−1
dx
0 x+z
for 0 < a < 1 and Rez > 0. Justify all computation.

Problem 8. Let C+ := {z ∈ C; Imz > 0} and let fn : C+ → C+ be a sequence of holomorphic

functions. Show that unless |fn | → ∞ uniformly on compact subsets of C+ , there exists a
subsequence converging uniformly on compact subsets of C+ .

Proof. First proof via Conformal Maps and Montels Define ψ(z) := (z +i)/(z −i) then ψ : C+ → D
is a conformal map. Then notice that gn := ψ ◦ fn : C+ → D, so the family is uniformly bounded, so by
Montel’s Theorem there exists a subsequence that converges locally uniformly to another holomorphic
function g. We still denote this subsequence as gn .

Case 1 If g(z) 6= −1 for any z ∈ C+ then we claim fn → ψ −1 ◦ g locally uniformly. Indeed, fix a
compact set K ⊂ C+ and notice ψ −1 (z) = i(1 + z)/(z − 1), so as g(z) 6= 1 on K and g(K) is compact,
97

there is a δ > 0 such that |g(z) + 1| > 2δ. By uniform convergence of gn to g on K, we may by assuming
n is large enough that |gn + 1| > δ. Then
|fn − ψ −1 g| = |ψ −1 ◦ ψ ◦ fn − ψ −1 g| ≤ C(δ)|ψ ◦ fn − g| ≤ C(δ)||ψ ◦ fn − g||L∞ (K)
where we used that ψ −1 is Lipschitz whenever we are a finite distance away from −1 since it is holomor-
phic. Therefore, fn → ψ −1 g uniformly on K, so we have fn → ψ −1 g locally uniformly on K.

Case 2 By the maximum principle, if g(z) = −1 for any z ∈ C+ then g(z) ≡ −1 on C+ . Now notice
we can reapply Montels theorem to any subsequence of {gn } to find a further subsequence that locally
uniformly converges. If every subsequence has a further subsequence that locally uniformly converges to
−1 then we have that gn locally uniformly converges to −1, so |fn | → ∞ locally uniformly. But if there
exists a subsequence of gn that has a further subsequence that does not converge locally uniformly to
−1, then we may reapply Case 1 to deduce that along a subsequence the subsequence locally uniformly
converges. Hence, we are done.

Alternative Proof via Harmonic Function Theory Notice that if for all w ∈ C+ we have
lim Im(fn (w)) = ∞
n→∞
then Harnack’s Inequality implies since vn := Im(fn ) ≥ 0 that for any compact set K
sup vn (z) ≤ C(K) inf vn (z)
z∈K z∈K

so we have local uniform convergence of the imaginary part to ∞. Now if there exists a point w such that
lim inf vn (w) = M < ∞
n→∞
then we claim we have local uniform convergence to a harmonic function along a subsequence.

WLOG by looking at the subsequence Im(fnk )(w) → Im(f )(w) = M < ∞, we can assume Im(fn )(w) →
M . By Harnack’s Inequality we obtain for any compact set K ⊂ C+ with w ∈ K that
sup vn (z) ≤ C(K)vn (w) ≤ M (K)
z∈K
i.e. the family of harmonic functions is uniformly bounded. Then observe that for z ∈ K there is a δ > 0
such that d(K, ∂C+ ) = 2δ then for any z ∈ K we have Bδ (z) ⊂ C+ so
ˆ
1
∂xi vn (z) = 2 ∂x vn (z)dA(z)
πδ Bδ (z) i
where dA is the lebesgue area measure. So by the divergence theorem we know
ˆ
1
= 2 vn (z)ni dσ(z)
πδ ∂Bδ (z)
where ni is the ith component of the normal and σ is the surface area measure. So in particular, we have
that
|∇vn (z)| ≤ C(K) sup |vn (z)| ≤ C̃(K)
z∈K
where the last constant does not depend on n thanks to our earlier remarks. In particular, this implies
the family is uniformly Lipschitz and Bounded. So by Arzela-Ascoli there is a uniformly convergent
sub-sequence, which we denote by vnk → v uniformly on K. By taking a compact exhausation of K,
and diagonal subsequence, we can find a subsequence which we denote by m such that vm → v locally
uniformly on K for any compact set K ⊂ C+ .

Now this implies that by the mean value equivalence of harmonic functions, that v is harmonic on C+
which is simply connected, so we can find a u such that u + iv is holomorphic on C+ . Now we claim
that fm → u + C + iv := f where C ∈ R is some constant locally uniformly. Indeed, notice that on any
compact set K ⊂ C+ we have
sup |Im(fm (z)) − Im(f (z))| → 0
z∈K
98 RAYMOND CHU

so the imaginary part of the holomorphic function f − fm converges to 0. In particular, by the Cauchy
Estimates we deduce that the gradients of the imaginary parts also go to zero uniformly, which imply
that the gradients of the real parts of f − fm go to 0 uniformly. This means f − fm converges locally
uniformly to a constant, and we choose C to make this constant 0 i.e. fm → f locally uniformly on C+ .

Therefore, we have shown if there is some z0 such that

lim inf vn (z0 ) = M < ∞
n→∞

then fn converges locally uniformly along a sub-sequence to a holomorphic function on C+ . But if the
alternative does not hold i.e.
lim vn (z0 ) = ∞
n→∞
for every z0 then |fn | → ∞ locally uniformly thanks to Harnack’s Inequality.

Problem 9. Let f : C → C be entire and assume |f (z)| = 1 when |z| = 1. Show that f = Cz m
for some m ∈ N and some C ∈ C with |C| = 1.

Proof. As f is entire such that |f (z)| = 1 on D, there are only finitely many zeros on D and as each zero
is isolated, there is some ε > 0 such that there are no zeros on {1 − ε < |z| < 1 + ε}. Enumerate the
zeros as {zn }Nn=1 then define the Blaschke products
z − zn
ψn (z) :=
1 − zn z
then |ψn (z)| = 1 with poles on 1/zn on ∂D and ψn (z) = 0 iff z = zn . Then define
N
Y
g(z) := f (z)/ ψn (z)
n=1

then g(z) is a holomorphic function on the disk such that |g(z)| = 1 on ∂D. Then observe that for any
|z| = 1 we have
1
z = ⇒ g(z) = 1/(g(1/z)) := h(z) on ∂D
z
and g and h extends to be continuous on ∂D since f is entire and the blashcke factors do not have poles
on {1 − ε < |z| < 1 + ε}. And similarly AS h(z) is holomorphic on C \ D and extends continuously to
∂D. Therefore, by Schwarz Reflection Principle, we know that g(z) extends to be entire with
(
g(z) for z ∈ D
g(z) :=
h(z) for z ∈ C \ D
Notice that g(z) is a bounded entire function since g and h are bounded, so
N
Y
g(z) = C ⇒ f (z) = C ψn (z)
n=1
QN
where |C| = 1 where the equality is due to analytic continuation since f (z) = C n=1 ψn (z) on D.
However, notice that ψn (z) is entire if and only if zn = 0 i.e. ψn (z) = z, so f is entire iff
f (z) = Cz m
where m is the multiplicity of the zeros at the origin.

Problem 10. Does there exist a function f (z) holomorphic on D such that lim|z|→1 |f (z)| = ∞.
Either find one or prove that it does not exist.
99

Proof. No. Suppose for the sake of contradiction that such a function existed, then for each w ∈ ∂D we
have
lim |f (z)| = ∞
z→w
so for any M ∈ N we can find an ε(w) > 0 such that on z ∈ Bε(w) (w) ∩ ∂D we have |f (z)| ≥ M . By
compactness, there are {wi }N
i=1 ⊂ ∂D such that
N
[
∂D ⊂ Bε(w) (w) ∩ ∂D := Γ1
i=1

Define Ω1 := D \ Γ1 then we the only zeros of f are in Ω1 by construction, so as each zero is isolated there
are only finitely many zeros on Ω1 . Enumerate the zeros as {zi }Ni=1 and define the Blaschke products
z − zn
ψn (z) :=
1 − zn z
then ψn (z) = 0 iff z = zn and |ψn (z)| = 1 on ∂D. In particular, define
N
Y
g(z) := f (z)/ ψn (z)
n=1

then g is a holomorphic function on D with no zeros on D such that lim|z|→1 |g(z)| = ∞ since the g extend
continuously to ∂D. Then by repeating the above argument, we can find for any M ∈ N a open and
connected subset ΩM of D such that on ∂ΩM we have |g(z)| ≥ M . Then as 1/g(z) is holomorphic, we
see that |1/g(z)| ≤ 1/M on ∂ΩM so we see this implies 1/|g(z)| ≤ 1/M on ΩM thanks to the maximum
principle. Notice also by construction ΩM → D as M → ∞, so we conclude that
1/|g(z)| = 0
i.e. |g(z)| = ∞ for all z, which implies |f (z)| = ∞ everywhere which is our contradiction.

Problem 11. Assume that f (z) is holomorphic on |z| < 2. Show that
1
max f (z) − ≥1
|z|=1 z

Proof. Notice that the residue of f (z) − 1/z at z = 0 is −1. So in particular, as f (z) − 1/z is holomorphic
on B2 (0)\{0} and the curve γ := eiθ where θ ∈ [0, 2π) with counterclock wise orientation is in the interior
of B2 (0) \ {0} , we can apply the Residue Theorem. This gives
ˆ
2πi = f (z) − 1/zdz
γ
but we also have the estimate ˆ
f (z) − 1/zdz ≤ 2πM
γ
where M := max|z|=1 |f (z) − 1/z| ≥ 1 which exist since both functions are continuous on the compact
set {|z| = 1}. So combining this we conclude
M ≥1
as desired.

Problem 12. Find a real valued harmonic function v defined on the disk D such that v(z) > 0
and limz→1 v(z) = ∞.

Let u be a real valued harmonic function on the disk D such that u(z) < M ≤ ∞ and
lim supr→1 u(reiθ ) ≤ 0 for all θ ∈ (0, 2π). Show that u(z) ≤ 0.
100 RAYMOND CHU

Proof. Notice that as D is simply connected and f (z) := z − 1 is a holomorphic function on D that
vanishes nowhere, we can find a branch of the log such that g(z) := log(z − 1) is holomorphic on D. This
then implies that the real pet i.e. log |z − 1| is a harmonic function on D. Then observe that
2
h(z) := log(2) − log |z − 1| = log( )≥0
|z − 1|
and is harmonic such that limz→1 h(z) = ∞.

Observe that the usual ε-log trick does not work here since we only have radial limits, so we instead
proceed via Poisson Integral Formula. Indeed, observe if 0 < t < 1 and ϕ ∈ [0, 2π) then for any t < r < 1
ˆ 2π
iϕ 1 r 2 − t2
u(te ) = u(reiθ )dθ
2π θ=0 |reiθ − teiϕ |2
and observe that
r 2 − t2 r 2 − t2
iθ iϕ 2
u(reiθ ) ≤ M iθ := A(r, θ)
|re − te | |re − teiϕ |2
so by Fatou’s Lemma we have
ˆ 2π ˆ 2π
r2 − t2 r 2 − t2

iθ iθ
lim inf A(r, θ) − u(re )dθ ≤ lim inf A(r, θ) − u(re )dθ
θ=0 r→1 |reiθ − teiϕ |2 r→1 θ=0 |reiθ − teiϕ |2
and by DCT we have ˆ ˆ
2π 2π
lim A(r, θ)dθ = A(1, θ)dθ
r→1 θ=0 θ=0
so ˆ ˆ
2π 2π
r2 − t2 1 − t2
u(teiϕ ) = lim sup iθ iϕ 2
u(reiθ )dθ ≤ lim sup(u(reiθ ))dθ ≤ 0
r→1 θ=0 |re − te | θ=0 |eiθ − teiϕ |2 r→1
where we used lim supr→1 u(reiθ ) ≤ 0 and 1 − t2 > 0.

101

14. Fall 2016

Problem 1. We consider the space L1 (µ) of integrable functions on a measure space (X, M, µ).
Suppose that f and fn are functions in L1 (µ) such that
(1) fn (x) → f (x) for µ almost every x ∈ X and
(2) ||fn ||1 → ||f ||1
Show that then ||fn − f ||1 → 0.

Proof. Observe by the triangle inequality that |f | + |fn | − |f − fn | ≥ 0 so we may apply Fatou’s Lemma
to see ˆ ˆ
lim inf |f | + |fn | − |f − fn |dµ ≤ lim inf |f | + |fn | − |f − fn |
R n→∞ n→∞
´ R
by pointwise convergence the left hand side converges ´ to R 2|f | while the´ right hand side by norm
convergence and linearity of the integral becomes R 2|f |dµ − lim supn→∞ R |f − fn |dµ which lets us
conclude that since f ∈ L1 (µ) that ˆ
0 ≥ lim sup |f − fn |dµ
n→∞ R
i.e. ||fn − f ||1 → 0 as desired.

Problem 2. Let µ be a finite positive Borel measure on R that is singular to the Lebesgue
measure. Show that
µ([x − r, x + r])
lim = +∞
r→0+ 2r
for µ a.e. x ∈ R.

Proof. Write the Lebesgue measure as m then notice that 2r = m([x − r, x + r]). As µ ⊥ m there exists
a set A such that µ(Ac ) = m(A) = 0. So now define for k ∈ N
µ([x − r, x + r])
Fk := {x ∈ A : lim sup < k}
r→0+ m([x − r, x + r])
then we claim that µ(Fk ) = 0 for any k, which implies that as
µ([x − r, x + r]) [
F := {x ∈ A : lim sup < ∞} = Fk
r→0+ m([x − r, x + r])
k∈N

that µ(F ) = 0 since Fk ⊂ Fk+1 and limk→∞ µ(Fk ) = µ( k∈N Fk ) = µ(F ). Therefore, F ∪ Ac is a null set
S
for µ, which implies the desired claim.

Fix k ∈ N and ε > 0 then by outer regularity of the Lebesgue measure there exists an open set Uε
such that A ⊂ Uε and m(Uε ) < ε. Then for any x ∈ F we have an rx > 0 such that
µ([x − rx , x + rx ]) < km([x − rx , x + rx ])
Now this implies Fk is covered by a collection of balls, so by Vitali’s covering lemma, we can find countably
many balls with radii xi and center ri such that xi ∈ Fk and the above inequality is true and the sub
collection of balls is disjoint and [
Fk ⊂ B5ri (xi )
k∈N
and by choosing the radii even smaller if necessary we can assume each Bri (xi ) ⊂ Uε so in particular,
∞
X ∞
X ∞
X
µ(Fk ) ≤ µ(B5ri (xi )) ≤ k m([xi − 5ri , xi + 5ri ]) = 5k m([xi − ri , x + i + ri ])
k=1 k=1 k=1

≤ 5km(Uε ) ≤ 5kε → 0
102 RAYMOND CHU

where for the final inequality we used the balls are disjoint, so we conclude µ(Fk ) = 0 for all k. Therefore,
the desired claim holds.

Problem 3. If X is a compact metric space, we denote by P(X) to be the set of postivie Borel
measures µ on X such that µ(X) = 1.
(1) Let ϕ : X → [0, ∞] be a lower-semi continuous function on a compact metric space X.
Show that if µ and µn are in P(X) and un → µ with respect to the weak-star topology
on P(X), then ˆ ˆ
ϕdµ ≤ lim inf ϕdµn
n→∞

(2) Let K ⊂ Rd be a compact set. For µ ∈ P(K), we define

ˆ ˆ
1
E(µ) = dµ(x)dµ(y)
K K |x − y|
Show that E : P(K) → [0, ∞] attains its minimum on P(K) (which could possibly be ∞).

Proof. By Risez-Representation Theorem, we know that the dual space of the C(X) (since X is compact
we know all continuous functions are bounded) is the space of Radon Measures. So µn → µ in the weak
star topology iff for every f ∈ C(X) we have
ˆ ˆ
f dµn → f dµ

As ϕ is lower semi-continuous and bounded from below there exists a sequence of ϕn ∈ C(X) such that
ϕn ≤ ϕ and ϕn → ϕ. We observe that as ϕ − ϕn ≥ 0 that
ˆ ˆ
ϕdµn ≥ ϕm dµn

so we have from weak* convergence

ˆ ˆ ˆ
lim inf ϕdµn ≥ lim inf ϕm dµn = ϕm dµ
n→∞ n→∞

and this holds for all m and from the Monotone Convergence Theorem (since ϕ1 is bounded below since
it is continuous and we are on a compact space) we have that
ˆ ˆ
ϕdµ = lim ϕm dµ
m→∞

so it follows that ˆ ˆ
lim inf ϕdµn ≥ ϕdµ
n→∞

For the second part assume the minimum isn’t +∞, otherwise every measure attains the minimum.
Let {µn } be a minimizing sequence of E(µ) i.e. E(µn ) → inf µ∈P(K) E(µ). By Banach-Alagou combined
with C(K) being separable , as µn are probability measures, we know that there is a weak* convergent
subsequence say µnk that converges to µ (since the weak* topology on C(K) is metrizable sicne C(K)
is separable). µ is a Borel Probability Measure since 1 is continuous so 1 = µn (X) → µ(X). By Stone-
Weiestrass, functions of the form f (x)g(y) are dense in C(X × X) f ∈ C(X) and g ∈ C(X). This implies
that µnk ⊗ µnk → µ ⊗ µ in the weak* topology [Fubini is justified since µ is a probability measure and
f ∈ C(X) is bounded so f ∈ L1 (X)]. Notice that 1/|x − y| is lower semi-continuous, so by part (1) we
know that ˆ
E(µ) ≤ lim inf E(µn ) = inf E(ν)
n→∞ ν∈P(K)

and since µ ∈ P(K) we obtain that it is a minimizer.

103

Problem 4. Let L1 = L1 ([0, 1]) and L2 = L2 ([0, 1]). Show that L2 is a meager subset of L1 .

Proof. First observe that

∞
[ ˆ 1 ˆ 1 ∞
[
L2 = {f ∈ L2 ∩ L1 : |f (x)|dx ≤ n and |f (x)|2 dx ≤ n} = Fn
n=1 0 0 n=1

We claim that Fn is closed in L1 . Indeed, if fn ∈ Fn converge to f ∈ L1 ([0, 1]) then we have that along
a subsequence we have fnk → f pointwise a.e. so by Fatou’s Lemma
ˆ 1 ˆ 1 ˆ
|f (x)|2 = lim inf |fnk (x)|2 ≤ lim inf |fnk (x)|2 ≤ n
0 0 k→∞ k→∞

so f ∈ Fn i.e. Fn is closed.

Now we claim Fn has empty interior, indeed fix f ∈ Fn . Then for ε > 0 consider
ε
gε (x) := f (x) − √
2 x
then we have that ˆ ˆ
1 1
ε
|f (x) − gε (x)|dx = x−1/2 dx = ε
0 2 0
/ Fn since |gε (x)|2 = |f |2 − εx−1/2 + εf 2 /4x−1 and the first term is in L1 by definition of Fn ,
and gε (x) ∈
the second term is in L1 by Holder’s Inequality, while the second term is not in L1 . In particular,
ˆ 1
|gε (x)|2 dx = +∞
0

since the first two terms are bounded and the last term is unbounded in L1 . Therefore, Fn has empty
interior in L1 since for any f ∈ Fn we constructed a sequence g1/n → f in L1 and g1/n ∈ Fnc , so L2 is
meager.

Problem 5. Let X = C([0, 1]) be equipped with the norm ||f || = maxx∈[0,1] |f (x)|. Let A be the
borel σ-algebra on X. Show that A is the smallest σ-algebra on X that contains sets of the form
S(t, B) = {f ∈ X : f (t) ∈ B}
where t ∈ [0, 1] and B is a borel subset of R.

Proof. Define the evaluation map for t ∈ [0, 1] via

φt (f ) = f (t)
then φt is continuous linear function on X since
|φt (f )| = |f (t)| ≤ ||f ||
−1
and observe S(t, B) = φ (B) so by continuity, S(t, B) is a Borel Subset of X i.e. S ⊂ A where S is the
σ-algebra generated by S(t, B).

For the reverse inclusion observe that if r > 0 and f ∈ X then

\
{g : ||f − g|| ≤ r} = φ−1
q ([f (t) − r, f (t) + r]) ∈ S
q∈Q∩[0,1]
−1
T
since g ∈ q∈Q∩[0,1] φq ((f (t) − r, f (t) + r)) iff for every q rational |f (q) − g(q)| ≤ r, which by continuity
implies ||f − g|| ≤ r. Then observe
[ \
Br (f ) = φ−1
q ([f (t) − m, f (t) + m]) ∈ S
m∈Q:m<r q∈Q∩[0,1]
104 RAYMOND CHU

since g ∈ Br (f ) iff ||g − f || < r so there is some rational q such that ||g − f || ≤ q < r. So in particular,
as A is the smallest σ-algebra that contains open sets of X we see that A ⊂ S so combining all of this
we obtain S = A as desired.

Problem 6. Show that there is no sequence {un } ∈ `1 such that (i) ||un ||1 ≥ 1 for all n ∈ N and
(ii) hun , vi → 0 for all v ∈ `∞ .

Also show that every weakly convergent sequence in {un } in `1 converges in the norm topology of
`1 .

Proof. First Part Missing

For the second part, assume un * u. Assume for the sake of contradiction that there existed a
sub-sequence and ε > 0 such that
||unk − u||1 > ε
then we know that
1
h unk − u, vi → 0
ε
since un weakly converges to 0, but ||2/ε(unk − u)|| ≥ 1 for large enough k, which contradicts the first
part, so ||un − v|| → 0 i.e. we have strong convergence to zero.

Problem 7. Let H be the space of holomorphic functions on the unit disk that are in L2 (D) with
respect to the Lebesgue measure on D. Endow H with the inner product
ˆ
(f, g) = f (z)g(z)dA
D
Fix z0 ∈ D and define Lz0 (f ) = f (z0 ) for f ∈ H.
(1) Show that Lz0 : H → C is a bounded linear functional on H.
(2) Find an explicit gz0 ∈ H such that
Lz0 (f ) = f (z0 ) = hf, gz0 i
for all f ∈ H.

Proof. As z0 ∈ D there is a δ > 0 such that Bδ (z0 ) ⊂ D. Then for any 0 < r < δ notice that by the mean
value property that
ˆ 2π
1
f (z0 ) = f (z0 + reiθ )dθ
2π 0
so ˆ δ ˆ δ ˆ 2π
1
rf (z0 ) = r f (z0 + reiθ )dθ
r=0 r=0 2π 0
ˆ δ ˆ 2π ˆ
1 1
= rf (z0 + reiθ )dθ = f (z)dA(z)
2π r=0 θ=0 2π Bδ (z0 )
where all the computation is justified since f ∈ L2 (D) which allows us to use Fubini since D is a finite
measure space (i.e. L2 (D) ⊂ L1 (D)). So in particular, we conclude that
ˆ
1
f (z0 ) = 2 f (z)dA(z)
πδ Bδ (z0 )
so we have
ˆ ˆ 1/2
1 1
|Lz0 (f )| = |f (z0 )| ≤ |f (z)|dA(z) ≤ √ |f (z)|2 dA(z)
πδ 2 D πδ 2 D
105

where the last inequality is due to Holder. So in particular, we conclude that Lz0 (f ) is a bounded linear
functional on H.
We know such a function exists by Risez Representation Theorem. Note that by our previous computation
we have that if z0 = 0 that for any 0 < δ < 1 that
ˆ
1
L0 (f ) = f (0) = 2 f (z)dA(z)
πδ Bδ (0)
and since 1δ f χBδ (0) → f χB1 (0) as δ → 0 and is dominated by 2f χBδ (0) for 1/2 < δ < 1 we may apply the
DCT to get that ˆ
1
L0 (f ) = f (z)dA(z) = (f, 1/π)
π D
Now we observe that if we define the automorphism of the disk
z0 − z
φz0 (z) :=
1 − z0 z
then φz0 (0) = z0 and φz0 (z0 ) = 0 so we have
f ◦ φz0 (0) = f (z0 )
so we have ˆ
1
Lz0 (f ) = L0 (f ◦ φz0 ) = f ◦ φz0 (z)dA(z)
π D
So we know that by integration by substitution
ˆ
1
= f (z)|(φ−1 0 2
z0 ) | dA(z) = (f ψ, ψ/π)
π D
where ψ = (φ−1 0
z0 ) . Therefore, we have shown that

f (z0 )ψ(z0 ) = (f ψ, ψ(z0 )ψ/π)

And as ψ is the derivative of a conformal map we know that ψ 6= 0 anywhere, so we have for any F ∈ H
that
ψ(z0 )
F (z0 ) = F (z0 )/ψ(z0 )ψ(z0 ) = ((F/ψ)ψ, ψ(z0 )/πψ) = (F, ψ)
π
ψ(z0 )
and π ψ ∈ H, so this is the desired function.

Problem 8. Let f be a continuous complex-valued function on the closed unit disk D and f
holomorphic on the open unit disk and f (0) 6= 0.
(1) Show that if 0 < r < 1 and if inf |z|=r |f (z)| > 0, then
ˆ 2π
1
log |f (reiθ )|dθ ≥ log |f (0)|
2π 0
(2) Show that m({θ ∈ [0, 2π] : f (eiθ ) = 0)}) = 0.

Proof. Note that as f (z) is holomorphic then whenever f (z) 6= 0 then in a small ball we know that
f (z) 6= 0 since 0s are isolated, so whenever f (z) 6= 0 we can define a complex log in a small neighborhood
of z to get that log |f (z)| is locally the real part of the holomorphic function log(f (z)) whenever z 6= 0.
Then if f (z) = 0 then log |f (reiθ )| = −∞, so it follows that log |f (z)| is sub-harmonic since it is upper
semi-continuous and locally satisfies the mean value inequality. So in particular,
ˆ 2π
1
log |f (reiθ )|dθ ≥ log |f (0)|
2π 0
as desired.
106 RAYMOND CHU

For the second part, fix an n ∈ N and define gn (z) := max(log |f (z)|, −n). Then as f (z) is continuous
on D we see that gn (z) is continuous on D since
(
log |f (z)| for z ∈ |f |−1 (e−n , ∞)
gn (z) =
−n else

and log |f (z)| is continuous on |f |−1 (e−n , ∞) since f is and they obtain the same value on the boundary.
Therefore, from continuity and Fatou’s Lemma we have
ˆ 2π ˆ 2π
1 1
|gn (eiθ )|dθ = lim inf |gn (reiθ ))|
2π 0 2π 0 r→1
ˆ 2π
1
≤ lim inf |gn (reiθ )|dθ
r→1 2π 0

where we used gn (eiθ ) = lim inf r→1 gn (reiθ ) by continuity. Now recall that from part 1 that
ˆ 2π
1
log |f (reiθ )|dθ ≥ log |f (0)|
2π 0
and since f is continuous then f is bounded, so log |f | is bounded from above, so this implies
ˆ 2π
1
| log |f (reiθ )||dθ ≤ C
2π 0
where C is independent of r. And observe that |gn (reiθ )| ≤ | log |f (reiθ )|| so we conclude that
ˆ 2π
1
|gn (eiθ )|dθ ≤ C
2π 0
and since |gn (eıθ )| increases to | log |f (eiθ )|| we conclude from the monotone convergence theorem that
ˆ 2π
1
| log |(f (eiθ ))|| ≤ C
2π 0
so log |(f (eiθ ))| ∈ L1 ([0, 2π]) which implies the problem.

Remark: These inequalities can also be proved using that | log |f (z)|| is lower semi-continuous com-
bined with Fatou’s Lemma.

Problem 9. Let µ be a positive Borel measure on [0, 1] with µ([0, 1]) = 1.

(1) Show that the function f defined as
ˆ
f (z) = eizt dµ(t)
[0,1]
for z ∈ C is holomorphic on C.
(2) Suppose that there is an n ∈ N such that
lim sup |f (z)|/|z|n < ∞
|z|→∞

Show that then µ is equal to the Dirac measure δ0 at 0.

Proof. Note that on any compact set K ⊂ C that on K × [0, 1] we have that |eizt | is bounded due to
continuity. Say on K × [0, 1] we have that |eizt | ≤ M then by Tonelli since all the terms are non-negative
we have ˆ ˆ ˆ ˆ
|f (z)|dA(z) ≤ |eizt |dµ(t)dA(z) = |eizt |dµ(t) ⊗ dA(z)
K K [0,1] K×[0,1]
ˆ
≤ M dµ(t) ⊗ dA(z) = M A(K) < ∞
K×[0,1]
107

where dA is the Lebesgue Area Measure on R2 ∼ = C and A(K) is the Lebesgue measure of K. So
f (z) ∈ L1loc (C, dA). So if R is a rectangle, then we know f (z) ∈ L1 (R) so we have by Fubini that
ˆ ˆ ˆ
f (z)dz = eizt dzdµ(t) = 0
∂R [0,1] ∂R
izt
since for any fixed t we have e is holomorphic on the interior of R and its boundary. Also f (z) is
continuous due to DCT, so by Morrera’s Theorem, we deduce that f (z) is entire.

For any m ≥ n, we know by Cauchy’s Estimates that for large enough m that
m!
|f m (0)| ≤ sup |f (z)| ≤ m!Rn−M → 0
Rm z∈BR (0)

so it follows that f (z) is a polynomial of degree at most n. Now we assume for the sake of contradiction
that µ is not any scalar multiple of the Dirac measure at z = 0. Indeed, if not then there is a set
0 ∈/ A and that µ(A) > 0. As [0, 1] is compact and µ is finite, we know that it is inner regular, so
µ(A) = supK⊂A µ(K) where K is compact, so we can find a compact K such that 0 ∈ / K and µ(K) > 0.
As K is compact there exists a min 0 < s ∈ K. We will show that f (z) blows up on the negative
imaginary axis exponentially which contradicts that f (z) is a polynomial. . Indeed, observe that for any
x ∈ (0, ∞)
ˆ ˆ
f (−ix) = ext dµ(t) ≥ ext dµ(t) ≥ exs µ(K)
[0,1] K

and µ(K) 6= 0, so this implies on the negative real axis, that t blows up exponentially i.e. f cannot be a
polynomial. Therefore, this is a contradiction so µ is a scalar multiple of the dirac measure δ0 but from
µ([0, 1]) = 1 we see the scalar must be one.

Problem 10. Consider the quadartic polynomial f (z) = z 2 − 1 on C. We are interested in the
iterates f n of f defined to be f 0 = idC for n = 0 and
f n = f ◦ .... ◦ f n times
(1) Find an explicit M > 0 such that the following dichotomy holds for each z ∈ C: either
(i) |f n (z)| → ∞ as n → ∞ or (ii) |f n (z)| ≤ M for all n ∈ N0 .

(2) Let U be the set of all z ∈ C for which the first alternative (i) holds and K be the set of
all z ∈ C for which the second alternative (ii) holds.
Show that U is an open set and K is a compact set without holes, i.e. C \ K has no
bounded connected components.

Proof. For the first part, observe that

|f (z)| ≥ |z|2 − 1
so now we claim if |f (z)| ≥ M := 10 then |f n (z)| ≥ 10 for all n. Indeed, |f 2 (z)| ≥ |f (z)|2 − 1 ≥ 102 − 1 ≥
99. Now by induction, assume it holds for m then |f m+1 (z)| ≥ |f m (z)|2 − 1 ≥ 102 − 1 = 99 as desired.
It therefore,follows that for any m if |f m (z)| ≥ 10 then
|f m−1 (z)|2
|f m (z)| ≥ |f m−1 (z)|2 − 1 ≥
2
so now we claim that if |f m (z)| ≥ 10 then
n
m+n |f m−1 (z)|2
|f (z)| ≥
22n −1
Indeed, by induction, the base case is true so
|f m+n |2 1 n+1 n+1 n+1 n+1
|f m+n+1 (z)| ≥ ≥ |f m−1 (z)|(2 ) /22 −2 = |f m−1 (z)|2 /22 −1
2 2
108 RAYMOND CHU

as desired. So if |f m (z)| ≥ 10 for some m then

n
1 (10)2 1 n
|f m+n (z)| ≥ 2n = 52 → ∞
2 2 2
so we have found such an M .

For the second part, notice that U is open since

[
U= {z : |f m (z)| > M }
m∈N0

and each set in its union is the preimage of a continuous function on an open set, so U is open. Notice
that K is compact since \
K= {z : |f m (z)| ≤ M }
m∈N0
and each set is closed while the m = 0 set is bounded and closed, so the entire set is bounded and closed
i.e. compact.

Assume for the sake of contradiction that C \ K has a bounded connected component. Write this
component as S.Then we from C = K ∪ (C \ K) and S is a bounded connected component of C \ K that
on ∂S we have for any m that |f m (z)| ≤ M , so by the maximum modulus principle we have |f m (z)| ≤ M
on S. However, as S ⊂ U we have |f m (z)| → ∞, which is our desired contradiction.

Problem 11. Suppose f : C → C is holomorphic function such that z 7→ g(z) := f (z)f (1/z) is
bounded on C \ {0}.
(1) Show that if f (0) 6= 0, then f is constant.
(2) Show that if f (0) = 0, then there exists an n ∈ N and a ∈ C such that f (z) = az n for all
z ∈ C.

Proof. Note that g(z) is holomorphic on C \ {0} and f (z) is entire and f (1/z) is holomorphic on C \ {0}.
As g(z) is bounded near 0, we know by Riemann’s Removable Singularity theorem that g(z) extends to
be an entire function on C. It therefore, follows that g(z) is a constant by Liouville theorem since g(z)
is a bounded entire function. So we have an a ∈ C such that
f (z)f (1/z) = a
As f (0) 6= 0 that
lim f (1/z) = a/f (0)
z→0
so it follows that the singularity of f (z) at ∞ is removable. Therefore, f (z) has to be a constant since
non-polynomials have an essential singularity at ∞ by just looking at the f (1/z) Taylor Expansion at
z = 0 and all non-constant polynomials have a pole at ∞.

Let n ∈ N be the multiplicity of the zero at z = 0 of f (z). Then f (z) = z n g(z) where g and g(0) 6= 0
is entire. Therefore,
f (z)f (1/z) = g(z)g(1/z)
and g(z)g(1/z) is bounded on C \ {0} with g(0) 6= 0, so we may reapply the first part to deduce g(z) = a
for some a ∈ N i.e. f (z) = az n .
109

15. Spring 2017

Problem 1. Let K ⊂ R be a compact set of finite measure and let f ∈ L∞ (R). Show that the
function ˆ
1
F (x) := f (x + t)dt
m(K) K
is uniformly continuous on R. Here m(K) denotes the Lebesgue measure of K.

Proof. Fix h ∈ R then observe

ˆ ˆ
1 1
F (x + h) = f (x + h + y)dy = f (x + t)dt
m(K) K m(K) K+h

where K + h := {x + h : x ∈ K}. Therefore,

ˆ
1
F (x + h) − F (x) = f (x + y) (χK+h − χK )
m(K) R
so we have from the triangle inequality that
ˆ
||f ||L∞
|F (x + h) − F (x)| ≤ |χK+h − χK |dx
m(K) R
Now observe that χK+h → χK pointwise a.e. as h → 0. So as K is compact there is an R > 0 such that
K ⊂ BR (0), so now observe for h small enough that χK+h ≤ χ2BR (0) so by the dominated convergence
theorem ˆ
|χK+h − χK |dx = m((K + h)∆K) → 0 as h → 0
R
so we deduce that F (x) is uniformly continuous since our above bounds are independent of x.

Problem 2. Let fn : [0, 1] → [0, ∞) be a sequence of functions, each of which is non-decreasing

on [0, 1]. Suppose that fn is uniformly bounded in L2 (R). Show that there exists a subsequence
that converges in L2 (R).

Proof. Observe that by assumpution there is an M > 0 such that

ˆ 1 ˆ 1
2 2
|fn | ≤ M ⇒ (1 − x)|fn (x)| ≤ |fn |2 ≤ M
0 x
where the last implication is due to fn being non-decreasing. So in particular, we deduce
M
|fn (x)| ≤ √ ∈ L1 ([0, 1])
1−x
so by the dominated convergence theorem it suffices to show that along a sub-sequence fn converges
pointwise a.e. Now observe for any ε > 0 that for x ∈ [0, 1 − ε] that the sequence |fn (x)| is uniformly
bounded. Therefore, by doing a diagonal subsequence argument we know that there is a subsequence
which we denote by nk such that for q ∈ Q ∩ [0, 1) that
fnk (q) → f (q)
0 0
Observe that if q < q then f (q) ≤ f (q ) since fnk are non-decreasing.

Now for any x ∈ [0, 1) define

Lx := inf f (q), Rx := sup f (q), q ∈ Q ∩ [0, 1)
q≤x q≥x

Then observe that for q1 ≤ x ≤ q2 where qi ∈ Q ∩ [0, 1) that

fnk (q1 ) ≤ fnk (x) ≤ fnk (q2 )
110 RAYMOND CHU

so if Lx = Rx we deduce from squeeze theorem that f (x) := limk→∞ fnk (x) is well defined. So if Lx 6= Rx
we see that from non-decreasing of fn that we have an interval (Lx , Rx ) and each such interval is disjoint
thanks to non-decreasing of fn . There can only be countably many of these intervals since each of these
intervals contain a rational point, therefore Lx = Rx a.e., so fnk (x) → f (x) := Lx a.e., so we can find a
pointwise sub-sequence on [0, 1) by a diagonalization trick. Then by applying DCT on this subsequence
on [0, 1 − ε] for all ε > 0 shows the desired result.

Problem 3. Let C([0, 1]) denote the Banach Space of Continuous Functions on the interval [0, 1]
endowed with the sup-norm. Let F be a σ-algebra on C([0, 1]) such that the map defined by
Lx (f ) := f (x)
is F-measurable. Show that F contains all the open sets.

Proof. Observe that if ε > 0 and f ∈ C([0, 1]) then

[
Bε (f ) = {g ∈ C([0, 1]) : ||f − g||∞ ≤ q}
q∈Q:q<ε

and that \
{g ∈ C([0, 1]) : ||f − g||∞ ≤ δ} = Lq ([f (q) − δ, f (q) + δ])
q∈Q
T
Indeed, observe that g ∈ q∈Q Lq ((f (q) − δ, f (q) + δ)) iff for all q ∈ Q we have
|g(q) − f (q)| ≤ δ ⇒ |g(x) − f (x)| ≤ δ
by density of rationals and continuity. Therefore, by definition as Lq ((f (q) − δ, f (q) + δ)) is F-measurable
since closed intervals are borel in R so we deduce that {g ∈ C([0, 1]) : ||f − g||∞ ≤ δ} := A(δ, f ) ∈ F .
So in particular, we have
[
Bε (f ) = A(q, f ) ∈ F
q∈Q:q<ε

Now as C([0, 1]) is separable (take polynomials by Stone Weiestrass), we deduce that every open set can
be written as a countable union of balls, so every open set is in F.

Problem 4. For n ≥ 1, let an : [0, 1) → {0, 1} denote the nth digit in the binary expansion of x,
so that X
x= an (x)2−n for all x ∈ [0, 1)
n≥1
(We remove any ambiguity from this definition by requiring that lim inf an (x) = 0 for all x ∈ [0, 1).)
Let M ([0, 1)) denote the Banach space of finite complex Borel measures on [0, 1) and define linear
functionals Ln on M ([0, 1]) via
ˆ 1
Ln (µ) := an (x)dµ(x)
0
Show that no subsequence of Ln converges in the weak* topology on M ([0, 1))∗

Proof. Fix a subsequence nk . Then define

X
x := 2−nk ∈ [0, 1)
k even

so we have X
x= an (x)2−n
n≥1
111

where an (x) = 0 for n 6= nk for some k where k is even. Therefore, lim inf n→∞ an (x) = 0 for instance
take k odd. So this is the digit binary expansion of x then define the Dirac Delta Measure at x δx i.e.
δx (E) = 1 if x ∈ E and 0 else. Then
(
1 if k is even
Lnk (δx ) = ank (x) =
0 if k is odd

Therefore, Lnk does not converge weak* in this subsequence and as this subsequence was arbitrary, we
are done.

Problem 5. Let dµ be a finite complex Borel measure on [0, 1] such that

ˆ 1
µ̂(n) := e2πinx dµ(x) → 0 as n → ∞
0
Let dν be a finite complex Borel measure on [0, 1] that is absolutely continuous w.r.t. dµ. Show
that
ν̂(n) → 0 as n → ∞

Proof. Notice by absolute continuity there is a f ∈ L1 (dµ) such that

ˆ 1 ˆ 1
2πinx
ν̂(n) = e dν(x) = e2πinx f (x)dµ(x)
0 0

Now observe that

ˆ 1
µ̂(−n) = cos(2πnx) − i sin(2πnx)dµ(x) → 0
0

where the convergence is due to µ̂(n) converges to 0 so its real and imaginary parts converge to 0 i.e.
´1 ´1
0
cos(2πnx) and 0 sin(2πnx) converge to 0 as n → ∞.

Now by Stone Weiestrass, we know that trigonometric polynomials are dense in C([0, 1]) with the sup
topology. Now observe that for m ∈ Z
ˆ 1
e2πimx e2πinx dµ(x) → 0 as n → ∞
0

so by linearity we deduce that for each trignometirc polynomial P (x) that

ˆ 1
P (x)e2πinx dµ(x) → 0
0

Then as µ is a finite complex Borel measure on a compact set we know that it is regular, therefore, by
Lusin’s theorem C([0, 1]) is a dense subclass of L1 ([0, 1], dµ) in the L1 norm. So we have that we can find
a trig polynomial P (x) such that ||P − f ||L∞ ([0,1]) ≤ ε where ε > 0 is a fixed number. Then
ˆ 1 ˆ 1
2πinx
|ν̂(n)| ≤ (f (x) − P (x))e dµ(x) + P (x)e2πinx dµ(x)
0 0

ˆ 1 ˆ 1 ˆ 1
≤ |f (x) − P (x)|dµ(x) + P (x)e2πinx dµ(x) ≤ εµ([0, 1]) + P (x)e2πinx dµ(x)
0 0 0
so we deduce as ε > 0 was arbitrary and µ is a finite measure that
|ν̂(n)| → 0 as n → ∞
as desired.
112 RAYMOND CHU

Problem 6. Let D be the closed unit disc in the complex plane, let {pn } be distinct points in
the open disc D and let rn > 0 be such that Dn = {z : |z − pn | ≤ rn } satisfy
(1) Dn ⊂ D;
Dn ∩ Dm = if n 6= m
(2) P
(3) rn < ∞
S
Prove X = D \ n Dn has positive area.

Hint: For −1 < x < 1 consider #{n : Dn ∩ {Re(z) = x}}

Proof.
P Let π(x, y) = x i.e. the projection map onto the x coordinate. Then we have that for f (x) :=
χπ(Dn)
(x) that
n
ˆ 1 Xˆ 1 X
f (x)dx = χπ (Dn )dx = 2rn < ∞
−1 n −1 n
where the second equality is due to the monotone convergence theorem. Now this implies f (x) < ∞ a.e.
so if f (x) < ∞ then we must have
#{n : Dn ∩ Re(z) = x} < ∞
S
Then observe that n (Dn ∩ {Re(z) = x}) is closed since this is a finite union of closed sets due to our
earlier remark, so it cannot be all of D ∩ {Re(z) = x}. Indeed this follows from D ∩ {Re(z) = x} being
open. Now observe that for [
Ex := (D \ Dn ) ∩ {Re(z) = x}}
n
we have that [
Ex ∪ (Dn ∩ {Re(z) = x}) = D ∩ {Re(z) = x}
n
where the union is disjoint because each Dn is disjoint, so by countable additivity of the one dimensional
Lebesgue measure m1 we deduce m1 (Ex ) > 0 for a.e. x since the second set is closed and their union is
open. Now observe if m2 is the two dimensional Lebesgue meausre then
ˆ ˆ 1 ˆ √1+x2
m2 (X) = χX (x, y)dA(x, y) = √ χD\(S Dn )∩Re(z)=x (y)dydx
n
D x=−1 y=− 1+x2
ˆ 1 ˆ √
1+x2 ˆ 1
= √ χEx (y)dydx = m1 (Ex )dx > 0
x=−1 y=− 1+x2 x=−1
since the measure is > 0 a.e.

Problem 7. Let f (z) be a one-to-one continuous mapping from the closed annulus
{1 ≤ |z| ≤ R}
onto the closed annulus
{1 ≤ |z| ≤ S}
such that f is analytic on the open annulus {1 < |z| < r}. Prove that S = R.

Proof. Let f be such an analytic function. By the open mapping theorem and continuity of f , it follows
that f maps boundary to boundary i.e. |f (eiθ )| = 1 or S and by continuity the modulus must be constant
on ∂D. If necessary, by considering f (z/R) we can assume that |f (eiθ )| = 1 so that f maps ∂D to ∂D.
Now consider
g(z) := 1/(f (1/z))
which is holomorphic since we took two conjugates and 1/f has no zeros. Then observe that 1/eiθ = eiθ
|g(eiθ )| = 1/f (eiθ ) = f (eiθ )
113

where for the last equality we used |f (eiθ )| = 1. Therefore, by Schwarz Lemma, this implies that
(
f (z) for z ∈ {1 ≤ |z| ≤ R}
f˜(z) :=
g(z) for z ∈ { R1 ≤ |z| ≤ 1}

is holomorphic such that f˜(z) : {1/R ≤ |z| ≤ R} → {1/S ≤ |z| ≤ R} such that |f˜(1/Reiθ )| = 1/S. This
6 1/|f |. We rewrite f˜ as f with this
map still is injective since |f (z)| > 1 for any 1 < |z| ≤ R so |f | =
extension.

Now as |f (1/Reiθ )| = 1/S we see by a similar agument that

f (z) = 1/S 2 f (1/(R2 z) on |z| = 1/R

so in particular, again f extends to a conformal map from {1/R2 ≤ |z| ≤ R} → {1/S 2 ≤ |z| ≤ S} since as
f is non-zero its maximum and minimum modulus must occur on the boundary and we have |f (eiθ )| = 1.
This map also has the property |f (1/R2 eiθ )| = 1/S 2

Now by iterating this argument of Schwarz reflecting off on ∂B1/R (0) we can extend f to a holomorphic
map on {0 < |z| ≤ R} to {0 < |z| ≤ S} such that |f (eiθ /Rn )| = 1/S n , but then the singularity at 0 is
removable since f is bounded by S. Then observe that by construction we have for any n ∈ N
|f (eiθ /Rn )| = 1/S n → 0
so f (0) = 0. Therefore, f extends to a conformal map from BR (0) to BS (0). So as f (z) is differentiable
at z = 0 it is locally injective in a small neighborhood of 0, but we have
|f (eiθ /Rn ) − f (0)| = 1/S n ≤ K1 /Rn
for some K1 > 0 being the Lipschitz constant. But by considering the inverse of f and that f −1 (0) = 0
we have that for some constant K2 that
|f −1 (eiθ /S n ) − f −1 (0)| = 1/Rn ≤ K2 /S n
so by letting K := max{K1 , K2 } we have that
(R/S)n ≤ K and (S/R)n ≤ K
so if S/R 6= 1 one of these two sides will approach ∞ as n → ∞. Therefore, S = R as desired.

Problem 8. Let a1 , a2 , ..., an be n ≥ 1 points in D, so that

n
Y z − aj
B(z) :=
j=1
1 − aj z

has n zeros in D. Prove that B 0 (z) has n − 1 zeros in D.

Problem 9. Let f (z) be an analytic function in the entire complex plane C and assume f (0) 6= 0.
Let {an } be the zeros of f , repeated according to their multiplicities.
(1) Let R > 0 such that |f (z)| > 0 on |z| = R. Prove
ˆ 2π
1 X R
log |f (Reiθ )|dθ = log |f (0)| + log
2π 0 |an |
|an |<R
λ
(2) Prove that if there are constants C and λ such that |f (z)| ≤ Ce|z| for all z, then
X 1 λ+ε
<∞
|an |
114 RAYMOND CHU

Proof. The first part is known as Jensen’s formula. We will first prove the formula on D. As each zero
of f (z) is discrete and f is entire there are only finitely many zeros on D. Then define
Y z − an Y
B(z) := = Ban (z)
1 − an z
|an |<1 |an |<1

which is holomorphic on D since there are only finitely many zeros and the poles are at 1/an which has
magnitude bigger than 1. Now we see that
log |Ban (z)| = log | − an | = log |an |
and ˆ 2π
1
log |Ban (eiθ )|dθ = 0
2π 0
since Ban is an automorphism of the unit disk so that |Ban (eiθ )| = 1. And the only zero of Ban is at an
so Jensen’s formula is satisfied for each Ban . Then notice that g = f /B is holomorphic on the unit disc
such that g has no zeros. Therefore, there is a branch of the log such that log(g) is holomorphic since
D is simply connected. There, log |g| is holomorphic, so Jensen’s Formula holds for g due to the Mean
Value Property. Now notice that Jensen’s formula is multiplicative since log |xy| = log |x| + log |y| so we
have that f (z) = g(z)B(z) satisfies Jensen’s formula since g and B satisfies the formula.

Now observe that ˆ R

1
dx = log(R/|an |)
|an | x
so
X R X ˆ R 1 ˆ R X
1
log = dx = (χx≥|an | ) dx
|an | |an x 0 x
|an |<R |an |<R |an |<R
P
and observe |an |<R (χx≥|an | (r)) = n(r) where n(r) represents the number of zeros of f in Br (0). So
ˆ R ˆ 2π
n(r) 1
dr = log |f (Reiθ )|dθ − log |f (0)|
0 r 2π 0
So we have ˆ ˆ 2R ˆ 2π
2R
n(r) n(r) 1
≤ dr = log |f (2Reiθ )|dθ − log |f (0)|
R r 0 r 2π 0
so as n(r) is decreasign we deduce
ˆ 2R ˆ 2R
1 n(r)
n(R) log(2) = n(R) dr ≤ dr
R r R r
so we have ˆ 2π
1
n(R) log(2) ≤ log |f (2Reiθ )|dθ − log |f (0)|
2π 0
ˆ 2π
1 λ
≤ log |Ce(2R) |dθ − log |f (0)| ≤ log |C| + 2λ Rλ − log |f (0)| .f Rλ
2π 0
so we have that
n(R) = O(Rλ )
for R large.

Now observe that we only need to control the tail of the sum to show convergence so we can take n so
large such that if n ≥ N then |an | ≥ R where R is chosen to be so large such that n(r) = O(rλ ) for any
r ≥ R. Now observe that if 2M ≤ R ≤ 2M +1

X 1 λ+ε ∞ λ+ε
X X 1
≤
|an | |an |
n≥N n=M 2n ≤|am |≤2n+1
∞
X ∞
X ∞
X
≤ 2−n(λ+ε) n(2n+1 ) . 2−n(λ+ε) (2λn+λ ) = 2−εn+λ < ∞
n=M n=M n=M
115

since −ε < 0.

Problem 10. Let a1 , ..., an be n ≥ 1 distinct points in C and let Ω := C \ {a1 , ..., an }. Let H(Ω)
be the vector space of real-valued harmonic functions on Ω and let R(Ω) ⊂ H(Ω) be the space of
real parts of analytic functions on Ω. Prove the quotient space H(Ω)
R(Ω) has dimension n, and find a
basis for this space, and prove it is a basis.

Proof. Let f ∈ H(Ω) then by using the Cauchy Riemann equations we see that g := fx − ify is holomor-
phic. Now g is an analytic function except for isolated singularities at z = a1 , ..., an . So for 1 ≤ j ≤ n
define cj := Res(g, aj ). Now define
n
X cj
h(z) := g(z) −
j=1
z − aj
so by the residue theorem over any closed curve γ in Ω this function integrates to 0. So this implies that
h(z) has a primitive, which we denote by u(z). And the Cauchy Riemann Equation shows that w(z) is
the real part of u(z) up to constants iff w is differentiable in the real sense and u0 (z) = wx − iwy and if
we define
n
X
w̃(z) := f (z) − cj log |z − aj | ⇒ w̃x − iw̃y = h(z)
j=1
Therefore, H(Ω) 3 w̃ = Re(u) up to constants, so in particular w̃ ∈ H(Ω) ∩ R(Ω). So we have shown
{log |z − aj |}nj=1 span H(Ω)/R(Ω). Since the aj are distinct we can find an ε > 0 such that Bε (aj ) ∩
Bε (ak ) = ∅ for j 6= k so if
Xn ˆ Xn
αj log |z − aj | = 0 ⇒ 0 = αj log |z − aj | = 2πiαj ⇒ αj = 0
j=1 ∂Bε (ak ) j=1

so these functions are linearly independent and hence a basis of H(Ω)/R(Ω), so this vector space has
dimension n.

Problem 11. Let 1 ≤ p < ∞ and let U (z) be a harmonic function on the complex plane such
that ˆ ˆ
|U (x, y)|p dxdy < ∞
C
Prove U (z) = 0 for all z = x + iy ∈ C.

Proof. For any r > 0 and z0 ∈ C the mean value property tells us that
ˆ
1
U (z0 ) = 2 U (x, y)dxdy
πr Br (z0 )
so we have from Holder’s inequality that
1
|U (z0 )| ≤ 2 ||U ||Lp (Br (z0 ),dxdy) (πr2 )1−1/p . r−2/p ||U ||Lp (C,dxdy) → 0 as r → ∞
πr
since U ∈ Lp (C, dxdy), so it follows that
|U (z0 )| = 0
so U (z) = 0 for all z.

Problem 12. Let 0 < α < 1 and f (z) be an analytic function such that f ∈ C α (D). Show there
is a constant A such that
|f 0 (z)| ≤ A(1 − |z|)α−1
116 RAYMOND CHU

Proof. Notice by Cauchy’s Theorem and the Residue Theorem that if |z| = 1 − 2δ then for any 0 < ε < 2δ
ˆ ˆ
0 1 f (w) 1 f (w) − f (z)
f (z) = 2
dw = dw
2πi |z−w|=ε (w − z) 2πi |z−w|=ε (w − z)2
ˆ 2π
1 f (z + εeiθ ) − f (z) iθ
= εe
2π 0 ε2 e2iθ
so we have
|f 0 (z)| ≤ sup |f (z + εeiθ ) − f (z)|/ε ≤ Cεα−1
θ
Now observe
1 − |z| = 2δ
so we have by sending ε → 2δ we see that
|f 0 (z)| ≤ C(2δ)α−1 = C(1 − |z|)α−1
as desired .

117

16. Fall 2017

Problem 1. Suppose f : R → R is non-decreasing. Show that if A ⊂ R is a Borel set, then so is

f (A).

Proof. Define the set A := {A ⊂ R is borel : f (A) is Borel }. We claim that A is a σ-algebra and that A
contains the open sets, which implies the desired claim.

First we will show that A contains the open set. As f (x) is monotone, it has only countably many
−
discontinuities say {di }∞
i=1 and on each di we have that f (di ) := limx→d− f (x) < limx→d+ f (x) := f (d+i )
i i
i.e. an upward jump discontinuity. Then for any open interval (a, b) we have that f (a, b) is a countable
union of the form [f (a− ), f (qi− )) ∪ [f (qi+ ), f (qj− )) ∪ ...[f (qk+ ), f (b+ )] where the qj are in [a, b] and the end
points at f (a) and f (b) may be included or not. In any case, it is always a countable union of a mixture
of half open intervals, open intervals, or closed intervals, so f (a, b) is Borel. Therefore, this implies A
contains the open sets.

In particular, this implies f (R) is Borel since R is open. Now if {Ai }∞

i=1 ∈ A then
∞
[ ∞
[
f( Ai ) = f (Ai )
i=1 i=1
so A is closed under countable union. Then for any A ∈ A we have that
f (R) = f (A) ∪ f (Ac )
and due to monotocity f (A) and f (Ac ) have at most a countable number of points common, so f (Ac ) =
f (A) \ f (R)∪ countably many points. Indeed, if x ∈ A and y ∈ Ac are such that f (x) = f (y) then this
implies there is an interval containing x and y such that f is constant on this interval. Then the collection
of all intervals where f is constant is countable since on each of these intervals we can find a rational
not in any of the other intervals since x 6= y. So this f (A) and f (Ac ) differ by at most countably many
points, so (f (A))c differs from f (Ac ) by at most countably many points, so f (Ac ) is Borel Measurable.
Thus A contains the borel σ-algebra

Problem 2. Let {fn } denote a bounded sequence in L2 ([0, 1]). Suppose that {fn } converges
almost everywhere. Show that then {fn } converges in the weak topology on L2 ([0, 1])

Proof. See Fall 2012 Problem 1.

Problem 3. Let {µn } denote a sequence of Borel probability measures on R. For n ∈ N and
x ∈ R we define
Fn (x) := µn ((−∞, x])
Suppose the sequence {Fn } converges uniformly on R. Show that for every bounded continuous
function f : R → R the numbers ˆ
f (x)dµn (x)
R
converges as n → ∞.

Proof. Note that C(R) is not separable, so we are not guaranteed that weak* compactness is equivalent
to weak* subsequential compactness. Indeed, let µn = δn then µn does not weakly converge on any
sub-sequence, but by Banach Alagou we know that {µn } is weak* compact. Therefore, we instead ap-
proximate µ by its restrictions on compact sets.
118 RAYMOND CHU

Fix an R > 0 then consider νn := {µn |[−R,R] } i.e. the measure restricted to [−R, R]. Then C(KR )
is separable so by Banach Alagou, there is a subsequence such that νn weak* converges ´ to ν (we still
denote this subsequence as νn ). Then as Fn → F uniformly, we know that Gn (x) := (−∞,x] dνn uni-
formly converges. Then
´ by approximating intervals of the form (a, b] with linear functions it is easy to
see Gn (x) → G(x) := (−∞,x] dν uniformly. Therefore, this implies that the entire sequence νn → ν.

Then let ε > 0 then by uniform convergence of Fn we can find an N ∈ N such that if n, m ≥ N then
sup |Fn (x) − Fm (x)| ≤ ε
x∈R

Now as µN is a finite measure there is an R > 0 such that for KR = (−R, R] that (1−(FN (R)−FN (−R)) =
c
µN ((KR ) < ε, then as
|Fn (x)| ≤ |Fn (x) − FN (x)| + |FN (x)| ≤ ε + |FN (x)|
c
we deduce that µn (KR ) < 4ε for n ≥ N .

Now for any bounded f we have

ˆ ˆ ˆ
f (x)d(µn − µm ) = f (x)d(µn − µm ) + |f (x)|d|µn − µm |
R KR c
KR

Notice the first term converges by our previous argument since KR ⊂ [−R, R] and we showed that
µn |[−R,R] weak* converges, so it is small for large n, m. And the second
´ term is bounded by 8 supx∈R |f (x)|ε
since |µn (KR ) − µm (KR )| ≤ mun (KR ) + µm (KR ) = 8ε. Hence, R f (x)dµn (x) converges for all f ∈ C(R)
that is bounded.

Problem 4. Consider the Banach Space V = C([−1, 1]) of all real-valued continuous functions
on [−1, 1] equipped with the sup norm. Let B be the closed unit ball in V .

Show that there exists a bounded linear functional Λ : V → R such that Λ(B) is an open
subset of R.

Proof. Define
ˆ 1 ˆ 0
Λ(f ) := f (x)dx − f (x)dx
0 −1

then we claim Λ(V ) = (−2, 2). Indeed, fix ε > 0 and define

1 on [1/n, 1]

fn (x) := nx

−1 on [−1/n, −1]


then
ˆ 1/n ˆ 0
Λ(f ) = 2(1 − 1/n) + nx − nx = 2 − 1/n
0 −1/n
´1 ´0
so (−2, 2) ⊂ Λ(f ). To see that it is equality, notice that 0 f (x) and −1 f (x)dx are both bounded by
1 so to get 2 = Λ(f ), we need both integrals the first integral to be 1 and the second to be −1, which
means f = 1 a.e. on [0, 1] and f = −1 a.e. on [−1, 0] which contradicts continuity. And it’s clear Λ is
linear and we have
|Λ(f )| ≤ 2||f ||V
so it is bounded.

119

Problem 5. Suppose f : R → R is a bounded measurable function satisfying f (x + 1) = f (x)

and f (2x) = f (x) for almost every x ∈ R. Show that then there is a c ∈ R such that f (x) = c
almost everywhere.

Proof. Fix a representation of f such that f (x + 1) = f (x) and f (2x) = f (x) everywhere. Then notice
that f (x) ∈ L1loc (R), and that as f (x + 1) = f (x) it suffices to show the theorem on [0, 1]. Then denote
the Lebesgue Points of f as E then m(E ∩ [0, 1]) = 1. Fix x0 , y0 ∈ E ∩ [0, 1]
ˆ y0 +r ˆ x0 +r
1 1
f (y0 ) = lim f (x)dx and f (x0 ) = lim f (x)dx
r→0 2r y −r r→0 2r x −r
0 0

Now notice for dyadic rationals of the form m/2n where m, n ∈ Z that
f (x + m/2n ) = f (2n (x + m/2n )) = f (2n x + m) = f (2n x) = f (x)
and we recall that dyadic rationals are dense in R. So in particular, there exists a dyadic rational m/2n
such that if ε > 0 then x0 ≤ y0 + m/2n < x0 + ε then
ˆ y0 +m/2n +r
1 n
f (y0 ) = f (y0 + m/2 ) = lim f (x)dx
r→0 2r y0 +m/2n −r
so
ˆ y0 +m/2n +r
! ˆ x0 +r ˆ y0 +m/2n −r ˆ y0 +m/2n +r
1 1 1 1
f (x) − f (x) ≤ |f (y)| + |f (y)|
2r y0 +m/2n −r 2r x0 −r 2r x0 −r 2r x0 +r
ε
≤ M
r
where M is a bound for f so by choosing ε = r2 since ε was independent of r so we obtain the bound
|f (x0 ) − f (y0 )| ≤ lim rM = 0
r→0

Therefore, f (x0 ) = f (y0 ).

Problem 6. For f ∈ L2 (C). For z ∈ C define

ˆ
|f (w)|
g(z) := dA(w)
B(1,z) |w − z|
2
Show that then |g| is finite a.e. and g ∈ L (C)

Proof. Observe that by Cauchy Schwarz that

ˆ !2 ˆ ! ˆ !
2 |f (w)| 1 |f |2 1
|g(z)| = 1/2 |w − z|1/2
dA(w) ≤ dA(w) dA(w)
B(1,z) |w − z| B(1,z) |w − z| B(1,z) |w − z|

and 1/|z| ∈ L (B1 (0)) since C ∼

1
= R2 , so
|f (w)|2
ˆ
2
|g(z)| ≤ C χB1 (z) (w) dA(w)
C |w − z|
so we have ˆ ˆ ˆ
|f (w)|2
|g(z)|2 ≤ C χB1 (z) (w) dA(w)dz
C C C |w − z|
So by Tonelli since the integrand is non-negative
ˆ ˆ ˆ ˆ
1 1
= |f (w)|2 χB1 (z) (w) dA(z)dA(w) = |f (w)|2 dA(z)dA(w)
C C |w − z| C B1 (w) |w − z|
ˆ ˆ ˆ
1
= |f (w)|2 dA(z)dA(w) = C |f (w)|2 dA(w) < ∞
C B1 (0) |z| C
120 RAYMOND CHU

´ 1
´ 1
where we used B1 (w) |w−z|
dA(w) = B1 (0) |w|
dA(w) < ∞, so g ∈ L2

Problem 7. Prove that there exists a meromorphic function f on C with the following three
properties:
(1) f (z) = 0 iff z ∈ Z
(2) f (z) = ∞ iff z − 1/3 ∈ Z
(3) |f (x + iy)| ≤ 1 for all x ∈ R with |y| ≥ 1

Proof. Define
sin(πz)
g(z) :=
sin(π(z + 1/3))
then this a meromorphic function with the first 2 properties. Now we will show there is a C > 0 such
that
lim sup |g(z)| ≤ C
z→∞ with |Im(z)|≥1

We will deal with the case that Im(z) ≥ 1 as the other case is a similar argument. Indeed, write z = x+iy
with y ≥ 1 then by the triangle and reverse triangle inequalities

eiπz − e−iπz |eiπz | + |e−iπz | eπy + e−πy

|g(z)| = ≤ =
eiπ(z+1/3)−e−iπ(z+1/3) ||eiπz | − |e−iπz || eπy − e−πy
1 + e−2πy 2
= ≤ := C1 < ∞
1 − e−2πy 1 − e−2π
and similarily one shows that when y ≤ 1 |g(x+iy)| is bounded. So by continuity on Ω := C\{|Im(z)| ≤ 1},
we deduce that |g(z)| is bounded on Ω. Say this bound is M then
f (z) := g(z)/M
is the desired function .

Problem 8. Show that a harmonic function u : D → R is uniformly continuous iff it admtis the
representation ˆ 2π iθ
1 e +z
u(z) = Re iθ f (eiθ )dθ
2π 0 e −z
where f : ∂D → R is continuous.

Proof. It is a basic fact that u is uniformly continuous on D iff u is continuous on D. So define

ˆ 2π iθ
1 e +z
v(z) := u(eiθ )dθ
2π 0 eiθ − z
Then observe that v(z) is holomorphic on D since if R ⊂ D is a rectangle then by Fubini’s which can be
applied since v(z) ∈ L1loc (D, dA(z)) (because the integrand is continuous on any compact subset of D) we
have that ˆ ˆ 2π ˆ iθ
1 e +z
v(z)dz = iθ − z
u(eiθ )dzdθ = 0
R 2π θ=0 R e
iθ
where the last equality is due to P (z, θ) := eeiθ −z
+z
harmonic on D. Then as v(z) is continuous since
u(eiθ ) and P (z, θ) are, we deduce from Morrera’s that v is holomorphic. Therefore, its real part is
harmonic. Observe that
ˆ 2π iθ
1 e +z
w(z) := Re(v(z)) = Re iθ u(eiθ )dθ
2π 0 e −z
121

so by the Maximum Principle it suffices to show w(eiθ ) = u(eiθ ) to obtain u = w.

´ 2π
Note that 0
P (z, θ)dθ = 2π which follows from the Fourier Series of P . By computation we have for
z ∈ D that
1 − |z|2 1 − |z|2
(16.1) Re(P (z, θ)) = = ≥0
1 − 2Re(ze−iθ ) + |z|2 |eiθ − z|2
Therefore, we have
ˆ 2π
1
|w(reiψ ) − u(eiψ )| = Re(P (reiψ , θ)) u(eiθ ) − u(eiψ ) dθ

2π 0
ˆ ˆ
1
Re(P (reiψ , θ)) u(eiθ ) − u(eiψ ) dθ + Re(P (reiψ , θ)) u(eiθ ) − u(eiψ ) dθ

≤
2π |θ−ψ|<δ∩θ∈[0,2π] |θ−ψ|≥δ∩θ∈[0,2π]

= (I) + (II)
Note that we can bound (I) due to continuity continuity of u and and that Re(P (z, θ)) has mass 1 to
make (I) arbitrarily small for small δ. And notice that
1 − r2
Re(P (reiψ , θ)) = → 0 uniformly on |θ − ψ| ≥ δ as r → 1
|eiθ − reiψ |2
this follows since the denominator is uniformly bounded on this set, while the numerator goes to zero.
Therefore,

lim w(reiψ ) = u(eiψ )

r→1
so it follows from continuity of w that w = u on ∂D, so by the maximum principle since both functions
are harmonic we deduce that w = u on D.

For the converse, note by the first part of our proof we showed that
ˆ 2π iθ
1 e +z
u(z) := Re iθ f (eiθ )dθ
2π 0 e −z
is harmonic due to Morrera’s and Fubini’s Theorem. Also arguing like in the previous proof, we see that
u(eiθ ) = f (eiθ )
Therefore, u extends to a continuous function on D (since its clearly continuous in the interior since the
integrand is continuous), so u is a uniformly continuous harmonic function.

Problem 9. Consider a map F : C × C → C with the following properties:

(1) For each fixed z ∈ C the map w 7→ F (z, w) is injective
(2) For each fixed w ∈ C the map z 7→ F (z, w) is holomorphic
(3) F (0, w) = w for w ∈ C
Show that then
F (z, w) = a(z)w + b(z)
for z, w ∈ C, where a and b are entire functions with a(0) = 1, b(0) = 0, and a(z) 6= 0 for z ∈ C

Hint: Consider
F (z, w) − F (z, 0)
F (z, 1) − F (z, 0)

Proof. Fix a w ∈ C such that w 6= 0 or 1 then define

F (z, w) − F (z, 0)
fw (z) :=
F (z, 1) − F (z, 0)
122 RAYMOND CHU

Then from condition (i) we see that the numerator is never 0 since as w 6= 0 we have F (z, w) 6= F (z, 0) by
injectivity. Also as w 6= 1 we deduce from injectivity in w that F (z, w) 6= F (z, 1) so we have fw (z) 6= 1.
Therefore, since fw (z) is holomorphic that by Little Picard’s Theorem, we deduce that fw (z) is constant
for w ∈ C \ ({0} ∪ {1}) but we have f1 (z) = 1 and f0 (z) = 0, so we deduce that fw (z) is constant for any
w.

In particular, this implies that for all w ∈ C that

fw (z) = g(w)
but condition 3 implies that
F (0, w) − F (0, 0)
g(w) = fw (0) = =w
F (0, 1) − F (0, 0)
we conclude that
fw (z) = w
so we have
F (z, w) = (F (z, 1) − F (z, 0))w + F (z, 0) := a(z)w + b(z)
and note that by the given assumptions we have all the desired properties for a and b.

Problem 10. Let {fn } be a sequence of holomorphic functions on D with the property that
X∞
F (z) := |fn (z)|2 ≤ 1
n=1
for all z ∈ D. Show that the series defining F (z) converges uniformly on compact subsets and
that F is subharmonic.

Proof. We claim that if f is holomorphic then |f (z)|2 is subharmonic. Indeed, if f (z) = u + iv where u
and v are the real and imaginary parts then
|f (z)|2 = |u(z)|2 + |v(z)|2
so it suffices to show if u is a real valued harmonic function then (u(z))2 is subharmonic. Indeed, observe
∆(u2 ) = 2u∆u + 2∇u · ∇u = 2|∇u|2 ≥ 0
so u2 is subharmonic. Therefore, each |fn (z)|2 is subharmonic. So in particular, as
m
X
F (z) = sup |fn (z)|2
m
n=1

i.e. F is the supremium of subharmonic functions, we have that F is subharmonic as long as F (z) is
upper semi continuous and not identically ∞ (the second part follows from F (z) ≤ 1). We will show the
last part by showing the sum converges locally uniformly to deduce that F (z) is continuous, from which
we get that its subharmonic.

Now we use Harnack’s inequality sicne each |fn |2 ≥ 0 is a non-negative subharmonic functions to for
any 0 < R < 1/2 that there is a constant C = C(R) that depends only on R such that
|fn (z)|2 ≤ C|fn (0)|2 for z ∈ B2R (0)
so we have for z ∈ BR (0) that
∞
X ∞
X
|fn (z)|2 ≤ C |fn (0)|2 → 0
n=N n=N
where the last implication is due to the sum at z = 0 being summable. Therefore, the series defining F (z)
is uniformly summable on any compact set, so F (z) is the uniform limit of the partial sums, so F (z) is
continuous and hence by our earlier remarks we have F (z) is subharmonic.
123

Problem 11. Let f : D → C be an injective holomorphic function with f (0) = 0 and f 0 (0) = 1.
Show that then
inf{|w| : w ∈
/ f (D)} ≤ 1
with equality iff f (z) = z for all z ∈ D.

Proof. First assume for the sake of contradiction that

inf{|w| : w ∈
/ f (D)} > 1
Therefore, as f (D) is an open map since injective functions are non-constant and f (0) = 0 ∈ D, we
deduce that there is an δ > 0 such that B1+δ (0) ⊂ f (D). Let U ⊂ D be open such that f (U ) = D. Note
this implies U ⊂ D since B1+δ (0) ⊂ f (U ) since conformal maps map boundary to boundary. So there is
some ε > 0 such that U ⊂ B1−ε (0).

Now define
g := f −1 |D
then g : D → U and
g(z)
h(z) := : D → D with h(0) = 0
1−ε
so by Schwarz Lemma, we deduce that
1
|g 0 (0)| ≤ 1 − ε ⇒ ≤1−ε⇒1≤1−ε
|f 0 (0)|
which is our contradiction. So we must have inf{|w| : w ∈
/ f (D)} ≤ 1 for any injective holomorphic
function with f (0) = 0 and f 0 (0) = 1.

Now we deal with the equality. Note that f (z) = z trivially obeys the given bound and if
inf{|w| : w ∈
/ f (D)} = 1
implies D ⊂ f (D). Indeed, notice
/ f (D)} = 1 ⇒ for all w ∈ f (D)c , |w| ≥ 1
inf{|w| : w ∈
so f (D)c ⊂ Dc , so by D ⊂ f (D). So it follows that f −1 (D) ⊂ D with f −1 (0) = 0 and (f −1 (0))0 = 1, and
this is a map from the unit disk to the unit disk, so by Schwarz Lemma we deduce that there is some θ
such that
f −1 (z) = eiθ z ⇒ f (z) = e−iθ z
so f 0 (0) = 1 implies θ = 0 so f (z) = z.

Problem 12. Let f, g, h be complex valued functions on C with

f =g◦h
Show that if h is continuous and both f, g are non-constant holomorphic functions then h is
holomorphic too.

Proof. As g is non-constant, then Z := {z ∈ C : g 0 (z) = 0} is discrete. So for each z0 ∈ C \ Z there is

an ε > 0 such that Bε (z0 ) ⊂ C \ Z. So by the inverse mapping theorem by taking ε smaller if necessary,
we can find an inverse of g on Bε (z0 ) and g −1 on this neighborhood is holomorphic since g 0 (z) 6= 0 on
/ h−1 (Z) we can locally invert thanks to continuity which allows
Bε (z0 ) . Therefore, we have for any z0 ∈
us to avoid all the other zeros since the zeros are discrete
g −1 ◦ f (z) = h(z) for z ∈ Bδ (z0 )
for δ > 0 small enough that h(z) ∈
/ Z and small enough to apply Inverse Function Theorem. In particular,
h(z) is holomorphic on C \ Z as it is the composition of two holomorphic functions.
124 RAYMOND CHU

Now fix R > 0 then BR (0) ∩ Z = {z1 , .., zN } for some finite collection zi since the zeros are isolated.
Now it suffices by Riemann’s Removable Singularities theorem to show that h is bounded on BR (0) to
show that h extends to a holomorphic function on BR (0). But this follows from h being continuous.
Therefore, h is holomorphic on BR (0) for any R > 0 so its holomorphic on C.
125

17. Spring 2018

Problem 1. Suppose f ∈ L1 (R) satisfies

ˆ
f (x + h) − f (x)
lim sup dx = 0
h→0 R h
Show that f = 0 a.e.

Proof. Define
ˆ x
F (x) := f (y)dy
0
Let h > 0 and x ∈ E where E is the lebesgue points of f then
ˆ x+h !
F (x + h) − F (x) 1
= f (y)dy → f (x)
h h x

and now observe that if x, y ∈ E where y < x then

ˆ x+h ˆ x ˆ x
F (x + h) − F (y + h) + F (y) − F (x) = F (z)dz − F (z)dz = F (z + h) − F (z)dz
y+h y y

so it follows by the given assumputions that

ˆ
|F (x + h) − F (y + h) + F (y) − F (x)| 1
≤ |F (z + h) − F (z)|dz → 0
h h R

but as x and y are Lebesgue points |F (x+h)−F (y+h)+Fh

(y)−F (x)|
→ |f (x) − f (y)|. Therefore, we conclude
if x, y ∈ E then f (x) = f (y) i.e. f is constant a.e. and the only constant in L1 is 0 so f = 0 a.e.

Problem 2. Given f ∈ L2 (R) and h > 0 define

ˆ
2
h Q(f, h) := (2f (x) − f (x + h) − f (x − h))f (x)dx
R
(1) Show that
Q(f, h) ≥ 0 for all f ∈ L2 (R) and all h > 0
(2) Show that the set
E := {f ∈ L2 (R) : lim sup Q(f, h) ≤ 1}
h→0

is closed in L2 (R)

Proof. For thje first part observe that

h2 Q(f, h) = 2(f, f ) − (fh , f ) − (f−h , f ) ≥ 2||f ||2L2 − ||f ||2L2 − ||f ||2L2 = 0
where (·, ·) is the L2 inner product, fh (x) := f (x + h), and we used Cauchy-Schwarz.

For the second part observe that formally we expect that

ˆ ˆ
00
lim sup Q(f, h) = −f (x)f (x)dx = |f 0 (x)|2
h→0 R R

where the last equality is due to integration by parts, but we do not have that f is smooth or that the
operator is uniformly bounded in h, so we remedy this by recalling that the Fourier Transform converts
differentiation into multiplication by polynomials. So by taking the fourier transform, we expect that E
126 RAYMOND CHU

will be a space where (fˆ(ξ) · ξ) ∈ L2 (R) with ||fˆ(ξ) · ξ||L2 (R) ≤ 1 which will be a closed subspace. Hence,
we are motivated to use Plancheral to rewrite Q(f, h) as
ˆ
2 − e2πihξ − e−2πihξ
Q(f, h) = |fˆ(ξ)|2 ( )
R h2
so if f ∈ E we have by Fatou’s Lemma that
ˆ ˆ
2 − e2πihξ − e−2πihξ
|fˆ(ξ)|2 4π 2 ξ 2 dξ ≤ lim inf |fˆ(ξ)|2 ( )
R n→∞ R h2
ˆ
2 − e2πihξ − e−2πihξ
≤ lim sup |fˆ(ξ)|2 ( )≤1
n→∞ R h2
where the first inequality is due to recognizing the exponential terms as a second finite difference scheme
for its negative second derivative.

Now let fn ∈ E such that fn → f in L2 then by passing along subsequences we can assume fn → f
and fˆn → fˆ. Therefore, by Fatou’s Lemma we have that
ˆ ˆ
ˆ 2 2 2
|f (ξ)| 4π ξ ≤ lim inf |fˆn (ξ)|2 4π 2 ξ 2 ≤ 1
R n→∞ R
So now it suffices to show
ˆ ˆ
ˆ 2 − e2πihξ − e−2πihξ
2 2 2
|f (ξ)| 4π ξ dξ = lim |fˆ(ξ)|2 ( )dξ
R h→0 R h2
to deduce that f ∈ E. But this follows from the Dominated Convergence Theorem since
2 − e2πihξ − e−2πihξ
( ) . ξ2
h2
for h small enough where the implies constant does not depend on h.

Problem 3. Suppose f ∈ L1 (R) satisfies

ˆ ˆ
|f (x)f (y)|
lim sup dxdy < ∞
ε→0 R R |x − y|2 + ε2
Show that f = 0 almost everywhere.

Proof. By the monotone convergence theorem, we deduce that

ˆ ˆ
|f (x)f (y)|
2
<∞
R R |x − y|

Now we assume for the sake of contradiction that f 6= 0 a.e., so there is a δ > 0 such that m({x : |f (x)| >
δ}) 6= 0 where m is the Lebesgue measure on R1 . Write A(δ) := {x : |f (x)| > δ} then as χA(δ) ∈ L1loc (R)
we deduce by Lebesgue differentiation theorem that there is a z ∈ A(δ) such that for any ε > 0 we can
find a η > 0 such that
m(B(η, z) ∩ A(δ)) ≥ (1 − ε)m(B(η, z)) = 2(1 − ε)η
Now observe that
∞
[ ∞
[
B(η, z) = B(η2−n , z) \ B(η2−n−1 , z) := B(n)
n=0 n=0
so we have ˆ ˆ ˆ ˆ
|f (x)f (y)| |f (x)f (y)|
2
dxdy ≥ dxdy
R R |x − y| A(δ)∩B(η,z) A(δ)∩B(η,z) |x − y|2
∞ ˆ ˆ ∞
X 1 X (m(A(δ) ∩ B(n)))2
≥δ ≥ δ
m=0 A(δ)∩B(n) A(δ)∩B(n) |x − y|2 m=0
8η 2 2−2(n+1)
127

Now notice that most of the measure of this annular partition of B(η, z) lives in the first few annuli, so
this implies there is an N (ε) such that if n ≤ N (ε) then m(A(δ) ∩ B(n)) ≥ 1/2m(B(n)) which follows
from m(B(η, z) ∩ A(δ)) ≥ (1 − ε)m(B(η, z)) and N (ε) → ∞ as ε → 0. Therefore,
N (ε) N (ε)
δ X η 2 2−2(n+1) δ X 1
≥ = →∞
2 m=0 8η 2 2−2(n+1) 2 m=0 8
as ε → 0, so we have arrived at a contradiction.

Alternative Proof via Lebesgue Differentiation Theorem As f ∈ L1 (R) we have that the set
of Lebesgue Points is a set of full measure. Let x0 be a Lebesgue point of f then
ˆ h
1
f (x0 ) = lim f (x + x0 )dx
h→0 2h −h

and we have for any h > 0

ˆ !2 ˆ ! ˆ ! ˆ ˆ
1 x0 +h x0 +h
|f (x)| x0 +h
|f (y)| x0 +h x0 +h
|f (x)||f (y)|
|f (x)|dx = dx dy = dxdy
h x0 −h x0 −h h x0 −h h x0 −h x0 −h h2
where the interswap of integrand is justified by Fubini since the integrand is non-negative.
ˆ x0 +h ˆ x0 +h
|f (x)||f (y)|
≤4 dxdy → 0
x0 −h x0 −h |x − y|2
where the convergence to zero as h → 0 is due to the itegrand being in L1 thanks to the give assumption
combined with the monotone convergence theorem. Hence, for every lebesgue point of f we have f = 0,
so f = 0 a.e.

Problem 4. Fix 1 < p < ∞. Show that

ˆ r ˆ ρ
1
f→
7 [M f ](x, y) := sup f (x + h, y + `)dhd`
r>0,ρ>0 4rρ −r −ρ

is bounded on Lp (R2 ).

Show that ˆ ˆ
r r2
1
[Ar f ](x, y) := f (x + h, y + `)dh`
4r3 −r −r 2
converges to f a.e. as r → 0.

Proof. For the first part, we need the following result: If 1 < p < ∞ then for
ˆ r
1
T f (x) := sup |f (y + x)|dx
r>0 −r 2r
we have ˆ ˆ
p
|T f (x)| . |f (x)|p
R R
So observe that by Layer Cake Decomposition that
ˆ ˆ ∞
p
|T f (x)| dx = psp−1 m({x : |T f (x)| ≥ s})ds
R 0
So as we have
f (x) = f (x)χ|f (x)|≤s/2 + f (x)χ|f (x)|≥s/2 := g(x) + h(x)
we deduce that from ||T f ||L∞ ≤ ||f ||L∞ if
s ≤ |T f (x)| = |T (g(x) + h(x))| ≤ |T g(x)| + |T h(x)| ≤ s/2 + |T g(x)|
i.e.
m({x : |T f (x)| ≥ s}) ≤ m({x : |T h(x)| ≥ s/2})
128 RAYMOND CHU

so in particular, we have that

ˆ ˆ ∞ ˆ ∞ ˆ
p p−1 p−2
|T f (x)| ≤ ps m({x : |T h(x)| ≥ s/2}) . s ( |f (x)|χ|f (x)|≥s/2 )
R 0 0 R
where the last inequality is due to Hardy Littlewood Maximal Inequality. Then Tonelli gives
ˆ ˆ 2|f (x)| ˆ
≤ |f (x)| sp−2 dsdx . |f (x)|p
R 0 R
as desired.

Now we use this Lemma on our problem. Observe for fx (y) := f (x, y) that
ˆ r ˆ ρ ˆ r ˆ ρ
1 1 1
|f (x + h, y + `)|dhd` ≤ sup sup |fy+` (x + h)|dhd`
4rρ −r −ρ r>0 2r −r ρ>0 2ρ −ρ
1
´ ρ
Then observe supρ>0 2ρ |f (x + h)|dhd` = T (fy+` (x)) where T was defined earlier as the Hardy
−ρ y+`
Littlewood Maximal Operator. So we have
|M f (x, y)| ≤ T (T (fy+` (x)))
ˆ ˆ ˆ ˆ ˆ ˆ
|M f (x, y)|p dydx . |T (fy+` (x))|p dydx = |T (fy+` (x))|p dxdy
R R
ˆ ˆ R R R R

. |f (x, y + `)|p dxdy = ||f ||pLp (R)

R R
so M f is a bounded linear operator from Lp to Lp .

For the second part, it’ll be identical to the usual proof of Hardy Little Maximal Inequality implies
Lebesgue Differentiation Theorem since thanks to the first part lets us obtain a similar estimate as the
Hardy Little Maximal Inequality via Chebyshev’s inequality.

Problem 5. Let µ be real valued Borel measure on [0, 1] such that

ˆ 1
1
dµ(t) = 0 for all x > 1
0 x+t
Show that µ = 0.

Proof. We will show that if f (t) is continuous then

ˆ 1
f (t)dµ(t) = 0
0
which will imply the claim. Observe that µ is finite since it is real valued everywhere and µ([0, 1]) ∈ R.
Observe that for h > 0 we have
ˆ ˆ 1
1 1 1 1 1
0= − dµ(t) = dµ(t)
h 0 x+t x+h+t 0 (x + t)(x + h + t)
So letting h → 0 with the DCT shows that (since 1/(x + t)(x + h + t) ≤ 1/x2 ≤ 1 ∈ L1 (dµ, [0, 1])
ˆ 1
1
dµ(t) = 0
0 (x + t)2
and repeating this argument n times shows for any n ∈ N
ˆ 1
1
dµ(t) = 0
0 (x + t)n
so in particular
ˆ N
1X
an
dµ(t) = 0
0 n=1
(x + t)n
129

where an ∈ N. Observe that A defined to be the set of finite linear combinations of 1/(2 + t), 1/(2 +
t)2 , ..., 1/(2 + t)n forms an algebra. It vanishes nowhere since 1/(2 + t) 6= 0 for t ∈ [0, 1] and it separates
points since if t1 6= t2 then 1/(2 + t1 ) 6= 1/(2 + t2 ) since 1/(2 + t) is injective on t ∈ [0, 1]. Therefore, by
Stone Weiestrass, A is dense in C([0, 1]) with the sup-norm, so it follows by uniform convergence that if
f (t) is continuous then
ˆ 1
f (t)dµ(t) = 0
0
which implies µ = 0

Problem 6. Let T denote the unit circle in C and let P(T) denote the space of Borel Probability
Measures on T and P(T × T) denote the space of Borel Probability Measures on T × T. Fix
µ, ν ∈ P(T) and define
ˆ ˆ ˆ ˆ
M := {γ ∈ P(T × T) : f (x)g(y)dγ(x, y) = f (x)dµ(x) · g(y)dµ(y) for all f, g ∈ C(T)}
T×T T T
Show that F : M → R defined by
ˆ ˆ
θ−φ
F (γ) := sin2 ( )dγ(eiθ , eiφ )
T×T 2
achieves its minimum on M.

Proof. Let {γn } ⊂ P(T × T) be a minimizing sequence i.e. F (γn ) → inf γ F (γ) then by Banach Alagou
and Risez Representation Theorem (since C(T) is separable, we can upgrade weak* compactness to weak*
subsequential compactness on bounded subsets) we can find a subsequence which we still denote by {γn }
such that γn weak star converges to γ. That is if f (x, y) ∈ C(T × T) then
ˆ ˆ ˆ ˆ
f (x, y)dγn → f (x, y)dγ

Taking f ≡ 1 implies γ ∈ P(T × T). Also observe that if f, g ∈ C(T) ⇒ f (x)g(y) ∈ C(T × T) so we have
ˆ ˆ ˆ ˆ ˆ ˆ
f (x)g(y)dγ(x, y) = lim f (x)g(y)dγn (x, y) = f (x)dµ(x) · g(y)dµ(y)
T×T n→∞ T×T T T

so γ ∈ M. Also observe that sin2 ( θ−φ

2 ) ∈ C(T × T) so weak* convergence implies
F (γn ) → F (γ)
but as γn is a minimizing sequence this implies γ is a minimizer.

Problem 7. Let F : C × C → C be (jointly) continuous and holomorphic in each variable. Show

that z 7→ F (z, z) is holomorphic.

Proof. Fix an R > 0 then if |w| < R then we have from the Residue Theorem that
ˆ
1 F (ξ, w)
F (z, w) = dξ
2πi |ξ|=R (ξ − z)
Now using the residue formula again gives
ˆ
1 F (ξ, η)
F (ξ, w) = dη
2πi |η|=R (η − w)
so it follows ˆ ˆ
1 F (ξ, η)
F (z, w) = − dξdη
4π 2 |ξ|=R |η|=R (ξ − z)(η − w)
130 RAYMOND CHU

i.e. ˆ ˆ
1 F (ξ, η)
F (z, z) = − dξdη
4π 2 |ξ|=R |η|=R (ξ − z)(η − z)
Now let R ⊂ BR (0) be a reactangle, then
ˆ ˆ ˆ 2π ˆ 2π
1 F (Reiϕ , Reiθ )
F (z, z)dz = 2 iϕ − z)(Reiθ − z)
R2 ei(ϕ+θ) dϕdθdz
∂R 4π ∂R θ=0 ϕ=0 (Re
ˆ 2π ˆ 2π ˆ
1 F (Reiϕ , Reiθ )
= R2 ei(ϕ+θ) dϕdθdz = 0
4π θ=0 ϕ=0 ∂R (Reiϕ − z)(Reiθ − z)
2

since the integrand is holomorphic in z and the swap of integrals is justified by continuity of the integrand
which on z ∈ BR (0) (just parametrize ∂R and convert the contour integral into a regular integral then
undo the parametrization on the last step for the rectangle). And as F (z, z) is continuous, we deduce
from Morrera’s that F (z, z) is holomorphic on BR (0). And as R was arbitrary we conclude F (z, z) is
entire.

Problem 8. Determine the supremum of

∂u
| (0, 0)|
∂x
among all harmonic functions u : D → [0, 1]. Prove that your answer is correct.

Proof. Fix 0 < R < 1 and define uR (z) := u(Rz) ∈ C(D) and is harmonic on D. So by Poisson Integral
Formula we have for 0 < r < 1 that
ˆ 2π ˆ 2π
iϕ 1 1 − r2 iθ 1 1 − r2
uR (re ) = u(Re )dθ = u(Reiθ )dθ
2π 0 |eiθ − reiϕ |2 2π 0 1 − 2r cos(θ − ϕ) + r2
so now we have by identifying reiϕ = r cos(ϕ) + ir sin(ϕ) = x + iy that for x > 0
ˆ 2π
1 1 − x2
uR (x, 0) = u(Reiθ )dθ
2π 0 1 − 2x cos(θ) + x2
Hence ˆ 2π
∂uR 1 −2x(1 − 2x cos(θ) + x2 ) − (−2 cos(θ) + 2x)(1 − x2 )
(x, 0) = u(Reiθ )dθ
∂x 2π 0 (1 − 2x cos(θ) + x2 )2
where differentiating under the integral sign is fine since R < 1 so the Poisson Kernal is smooth in BR (0).
In particular,
ˆ 2π
∂uR 1
(0, 0) = 2 cos(θ)uR (Reiθ )dθ
∂x 2π 0
so we have from 0 ≤ u ≤ 1 that
ˆ π/2
∂uR 1 2
(0, 0) ≤ 2 cos(θ)dθ =
∂x 2π −π/2 π
and
∂uR ∂u ∂u 2
=R ⇒ (0, 0) ≤
∂x ∂x ∂x π
so define f : S 1 → [0, 1] via (
1 on θ ∈ [−π/2, π/2]
f=
0 else
then ˆ 2π
1 1 − |z|2
u(z) := f (eiθ )dθ
2π 0 |eiθ − z|
is the desired harmonic function with these properties. Note that the usual proof of if f ∈ C(∂D) then
the usual proof of
ˆ 2π
1 1 − |z|2
lim u(reiθ ) := f (eiθ )dθ = f (eiθ )
r→1 2π 0 |eiθ − z|
131

extends to bounded functions at every point of continuity of f . Therefore, u(reiθ ) → χ[−π/2,π/2] every-
where except at θ = ±π/2 i.e. u obtains the boundary data f a.e., so by our above computation our
function u obtains the boundary data.
Alternative Proof By the mean value property we have for any 0 < r < 1
ˆ ˆ
∂u 1 ∂u 1
(0, 0) = 2 (z)dA(z) = 2 un1 dσ
∂x πr Br (0) ∂x πr ∂Br (0)
where n1 is the first component of the unit normal on ∂Br (0) i.e. x/r.
ˆ 2π ˆ 2π
1 1
= 2 ru(reiθ ) cos(θ)dθ = u(reiθ ) cos(θ)dθ
πr θ=0 πr θ=0
Since 0 ≤ u ≤ 1 we deduce
ˆ π/2
∂u 1 2
(0, 0) ≤ cos(θ)dθ =
∂x πr θ=−π/2 πr
So sending r → 1 we deduce that
∂u 2
(0, 0) ≤
∂x π
And from construction we see that the max must be 1 on [−π/2, π/2] and 0 else, so we use the same
function as before.

Problem 9. Consider the formal product

∞
Y 1 z
(1 + )z (1 − )
n=1
n n
(1) Show that the product converges for any z ∈ (−∞, 0)

(2) Show that this resulting function extends to an entire function.

Proof. Notice if an := (1 + n1 )z (1 − nz ) then an → 1 as n → ∞, so by looking at the tail of the product if

necessary, we can assume an ∈ B1/2 (1). This means we can define the standard complex logarithim i.e.
log(z) where θ ∈ (−π, π] and log(an ) is well defined. So now by taking logs we see that by taking limits
∞
Y ∞
X ∞
X ∞
X
log( an ) = log(an ) = log(an − 1 + 1) = (an − 1) + O((an − 1)2 )
n=1 n=1 n=1 n=1
so the product converges iff
∞
X
(an − 1)
n=1
converges. So we need to show the following sum converges
∞
X 1 z z
|(1 + ) (1 − ) − 1|
n=1
n n
Observe that
∂
(1 + x)z = z(1 + x)z−1
∂x
so Taylor Expansion at x = 0 gives
(1 + x)z = 1 + zx + O(x2 )
so we have
1 z z z z z2
(1 + ) (1 − ) = (1 − )(1 + ) + O|z| (1/n2 ) = 1 − 2 + O|z| (1/n2 )
n n n n n
Therefore,
∞ ∞
X 1 z z X |z|2
|(1 + ) (1 − − 1)| = + O|z| (1/n2 )
n=1
n n n=1
n2
132 RAYMOND CHU

which converges.

For the second part, notice that for any z ∈ C we can define the complex log for an so we have
(1 + 1/n)z = exp(zlog(1 + 1/n)) is a well defined holomorphic function (where we are using the standard
log branch). And by our earlier computations the sum converges locally uniformly on compact subsets
of C so
M M M
Y X X |z|
an = exp( (an − 1) + O((an − 1)2 )) = exp( O( 2 ))
n
n=N n=N n=N
so the product converges locally uniformly, hence the limit is holomorphic.

Problem 10. Let C∗ = C ∪ {∞} be the Riemann Sphere and Ω := C∗ \ {0, 1}. Let f : Ω → Ω
be a holomorphic function.
(1) Prove that if f is injective then f (Ω) = Ω

(2) Make a list of all such injective functions.

Proof. Let
z
ϕ(z) :=
z−1
then ϕ is a Mobius Transformation such that ϕ(0) = 0, ϕ(1) = ∞, and ϕ(∞) = 1. In particular, if
U := C \ {0} then ϕ(U ) = Ω so
g(z) := ϕ−1 ◦ f ◦ ϕ : U → U
And since ϕ is conformal it suffices to show that if g is injective then it is surjective.

Now g is a holomorphic function on C \ {0}. By Great Picard Theorem the singularity at 0 is either
removable or a pole.

Removable Singularity Case If g(z) has a removable singularity at 0 then g extends to be an in-
jective entire function. Indeed, if g(0) = g(w) for some w 6= 0, the open mapping theorem implies that g
is not injective on C \ {0} (since a small ball around 0 and around w maps to a ball around g(0) = g(w)).
Therefore, g(z) is an injective entire function so g is a linear function i.e. g(z) = az + b, so it follows that
g is surjective. And as g(z) 6= 0 for z 6= 0 it follows that b = 0 so g(z) = az.

Pole Case If g(z) has a pole at z = 0, then 1/g has a removable singularity at z = 0 and since g
does not map to 0, so we see 1/g is an injective holomorphic function on C \ {0} that has a removable
1
singularity at z = 0. By applying our previous case, we deduce that 1/g(z) = az + b ⇒ g(z) = az+b . As
g(z) has a pole at z = 0 we must have b = 0 and a 6= 0, so g(z) = 1/(az) which is a Mobius Transforma-
tion and g(0) = ∞ with g(∞) = 0, so it follows from Mobius Transformations being conformal that g is
surjective on U .

By our previous considerations we must have

g(z) = 1/(az) or az
for some a ∈ C that is not 0. So by undoing our mobius transformations we deduce that
f (z) = ϕ ◦ (1/az) ◦ ϕ−1 or ϕ ◦ (az) ◦ ϕ−1

Problem 11. For R > 1 let AR be the annulus {1 < |z| < R} . Assume there is a conformal
map F from AR1 onto AR2 . Prove that R1 = R2 .
133

Proof. See Spring 2017 Number 7.

Problem 12. Let f (z) be a bounded and holomorphic on D. Prove that for any w ∈ D that we
have ˆ
1 f (z)
f (z) = dz
π D (1 − zw)2

Proof. Let w ∈ C then define

w−z
ψw (z) :=
1 − zw
then ψw is an auotmorphism of the disk such that ψw (w) = 0 and ψw (0) = w. Therefore, notice if
f ◦ ψw : D → C so the mean value property tells us that for 0 < r < 1
ˆ
1
f (w) = f (ψw (0)) = 2 f ◦ ψw
πr Br (0)
and as f is bounded we have f ∈ L1 (D) so we can take limits to get
ˆ ˆ
1 1 −1
f (ψw (0)) = f ◦ ψw dz = f (z)|(ψw (z))0 |2 dz
π D π D
where the last equality is due to change of variables and that ψw is an automorphism so its inverse
derivative is well defined by the inverse function theorem. Note that
−1 z−w −1 1 − |w|2
ψw (z) = ⇒ ∂z (ψw (z)) =
1 − wz (1 − wz)2
so in particular,
ˆ
1 1 − |w|2 1 − |w|2 1 0 0
f (w) = f (z)( 2
)( 2
)dz = (f (z)ψw (z), ψw (z))
π D (1 − wz) (1 − wz) π
0
so as ψw6= 0 anywhere, we have that if F is holomorphic then
1 0 1 0 F 0 0
(F (z), ψw (z)) = (ψw 0
, ψw ) = F (w)/ψw (w)
π π ψw
It therefore, follows that
1 0 (w)ψ 0 (z)) = F (w)
(F (z), ψw w
π
and
0 1 − |w|2 1
ψw (w) = =
(1 − |w|2 )2 1 − |w|2
i.e. ˆ
1 1
f (w) = f (z) dz
π D (1 − zw)2
as desired.
134 RAYMOND CHU

18. Fall 2018

Problem 1. Let {fn } be a sequence of real-valued Lebesgue measurable functions on R, and let
f be another such function. Assume that
(1) ´fn → f Lebesgue a.e.
(2) ´R |x||fn |dx ≤ 100, for all n, and
(3) R |fn (x)|2 dx ≤ 100
Prove that fn ∈ L1 for all n, that f ∈ L1 , and that ||fn − f ||L1 → 0. Also show that neither
assumputions (2) nor (3) can be omitted while making these deductions.

Proof. Notice that for any n we have

ˆ ˆ ˆ √ ˆ !1/2 ˆ
2
|fn |dx = |fn |dx + |fn |dx ≤ 2 |fn | + |x||fn |
R |x|≤1 |x|≥1 |x|≤1 |x|>1
√ √
≤ 2||fn ||L2 (R) + 100 ≤ 10 2 + 100 =: C
where we used Holder’s inequality and the given inequalities. Therefore, fn ∈ L1 (R) and by Fatou’s
Lemma ˆ ˆ ˆ
|f |dx = lim inf |fn |dx ≤ lim inf |fn |dx ≤ C
R R n→∞ n→∞ R
1
so f ∈ L (R) and similarly Fatou implies that
ˆ ˆ
|x||fn | ≤ 100 and |fn |2 dx ≤ 100
R R
Now observe that for any M > 0 that
ˆ ˆ ˆ
|x|
|f − fn |dx = |f − fn |dx + |f − fn |dx
R |x|≤M |x|≥M |x|
ˆ ˆ
1
≤ |f − fn |dx + |x||f − fn |dx
|x|≤M M |x|≥m
ˆ
200
≤ |f − fn |dx +
|x|≤M M
so if ε > 0 choose M so large such that 200/M < ε/3/ Also as BM (0) is a set of finite measure, we know
by Egorov’s Theorem that for any δ > 0 we can find a set K ⊂ BM (0) such that fn → f uniformly on
K. So we choose N ∈ N so large such that if n ≥ N then supx∈K |f (x) − fn (x)| ≤ ε/(6M ) then
ˆ ˆ
≤ |f − fn |dx + |f − fn | + ε/3
K BM (0)\K

ˆ !1/2
2
≤ ε/3 + |f − fn | δ 1/2 + ε/3
BM (0)\K

≤ 2ε/3 + 200δ 1/2

so choosing δ > 0 such that 200δ 1/2 < ε/3 gives
≤ε
1
so fn → f in L .

Now if (2) is satisfied but (3) this means we do not necessarily have control of the function near the
origin. Indeed, consider
fn (x) := nχ[0,1/n] (x) ⇒ fn → 0 a.e.
with ˆ ˆ ˆ ˆ
|fn |dx = 1, |fn |2 = n → ∞, |x||fn | ≤ |fn | = 1
R R R R
135

and fn clearly does not converge to 0 in L1 since each fn has mass 1.

Now if (3) is satisfied but (2) is not, then we do not have control of the tail. Indeed consider
ˆ ˆ ˆ n
1 1
fn (x) := χ[0,n] (x) ⇒ |fn |2 = , |x||fn | = x/n = n/2 → ∞
n R n R 0

and fn pointwise goes to 0 but clearly does not go to 0 in L1 since each fn has mass 1.

Problem 2. Let (X, ρ) be a compact metric space which has at least two points, and let C(X)
be the space of continuous functions on X with the sup norm. Let D be a dense subset of X and
for each y ∈ D define fy ∈ C(X) by
fy (x) = ρ(x, y)
Let A be the sub-algebra generated by the collection fy (with pointwise addition and multiplication
of functions).
(1) Prove that A is dense in C(X) under the uniform norm.
(2) Prove that C(X) is separable.

Proof. For the first part, we will use Stone-Weiestrass. First notice that since X has at least two points
so does D since the two points are a non-zero distance away from one another. We claim A seprates
points. Indeed, let x, y ∈ X with x 6= y then d(x, y) = δ > 0. As D is dense there is an xn ∈ D such that
d(x, xn ) < δ/2 then we have
0 ≤ fxn (x) < δ/2
and we also have from the triangle inequality that
fxn (y) = d(xn , y) ≥ d(x, y) − d(xn , y) ≥ δ − δ/2 = δ/2 > fxn (x)
Therefore, fxn (x) 6= fxn (y), so A separates points.

Now it vanishes nowhere since if x ∈ X then as X has at least two points there is a y ∈ X such that
x 6= y ⇐⇒ d(x, y) = δ > 0. then there is a yn ∈ D such that d(y, yn ) < δ/2 which implies
fyn (x) = d(x, yn ) ≥ d(x, y) − d(y, yn ) ≥ δ/2 > 0
so as X is a compact metric space, Stone Weiestrass implies that A is dense in C(X).

For the second part, we know that as X is compact, it is totally bounded. Therefore, X is separable
so we can find a countable dense set E and we can similarly define AE to be the sub-algebra generated
by fy for y ∈ E. Then AE = C(X) and let
S := { finite rational polynomial combinations of fy }
PM QN (j)
i.e. f ∈ S if f = j=1 cj i=1 fyi where cj ∈ Q then S is countable and S = AE , so S is a countable
dense set in C(X) so C(X) is separable.

Problem 3. Let (X, ρ) be a compact metric space and let P (X) be the set of all positive prob-
ability measures on the Borel σ-algebra. Assume that µn weakly converges to µ. Prove that
µn (E) → µ(E) for all E with µ(E) = µ(Int(E)).

Proof. If F is a closed set then we know that χF (x) is upper semi-continuous so there exists a sequence
of fn ∈ C(X) such that fn ≥ χF and fn → χF pointwise. Then we have for any m
ˆ ˆ ˆ
lim sup χF dµn ≤ lim sup fm dµn = fm dµ
n→∞ X n→∞ X X
136 RAYMOND CHU

so then as fm → χF and f1 ≥ fm for any m with f1 ∈ L1 (X, dµ) (this

´ follows from
´ X is compact so f1
is bounded and µ being a probability measure), so by DCT we know X fm dµ → X χF dµ i.e.
ˆ
lim sup χF dµn ≤ µ(F )
n→∞ X

Now if G is an open set, we know that χG is lower semi continuous so we can find fn ∈ C(X) such
that χG ≥ fn and fn → χG . Therefore,
ˆ ˆ ˆ
lim inf χG dµn ≥ lim inf fm dµn = fm dµ
n→∞ X n→∞ X X

so applying the monotone convergence theorem gives

ˆ ˆ
lim inf χG dµn ≥ χG dµ = µ(G)
n→∞ X X

Now by applying these two lemmas we see that for any set E with the given conditions that
lim sup µn (E) ≤ lim sup µn (E) ≤ µ(E) = µ(E) = µ(Int(E)) ≤ lim inf µn (Int(E)) ≤ lim inf µn (E)
n→∞ n→∞ n→∞ n→∞

which implies
µ(E) = lim µn (E)
n→∞
as desired.

Problem 4. Let T be the unit circle in C and for each α ∈ T define the rotation map Rα : T → T
by Rα (z) = αz. A Borel probabiltiy measure µ on T is called α-invariant if µ(Rα (E)) = µ(E) for
any Borel set E ⊂ T.
(1) Let m be the Lebesgue measure defined on T (defined, for instance, by identifying T with
[0, 1) through the exponential map). Show that for each α ∈ T that m is also α-invariant.
(2) Prove that if α is not a root of unity, then the set of powers {αn : n ∈ Z} is dense in T.
(3) Prove that if we fix α ∈ T that is not a root of unity then the only α-invariant Borel
probability measures is the Lebesgue measure.

Proof. For the first part if α ∈ T then α = eiψ for some ψ ∈ [0, 2π). So if E = {eiθ : θ ∈ [θ0 , θ1 ]} then we
have Rα (E) = {ei(θ+ψ) : θ ∈ [θ0 , θ1 ]} ∼
= [θ0 + ψ, θ1 + ψ] so
m(Rα (E)) = m([θ0 + ψ, θ1 + ψ]) = m([θ0 , θ1 ]) = m(E)
where in the last inequality we used that the Lebesgue measure is translation invariant for intervals.
Then as E generates the Borel σ-algebra on T, we see that for any Borel Set E that m(Rα (E)) = m(E).

For the second part, let α be a non root of unity, then as T is compact in C there is a subsequence which
we still denote by αn such that αn is convergent in T. We define ||eiθ ||T := inf{ψ ∈ [0, 2π) : eiψ = eiθ } i.e.
we want the angle in [0, 2π). Let the limit point be denote by β. Now if ε > 0 we know that evenetually
for large enough n, m with m > n that
0 < ||αn − αm ||T < ε
where we have the lower bound since α is not a root of unity. This means αm−n corresponds to a non
trivial rotation that rotates us at most by ε degrees. Since ε > 0 is arbitrarily small and the rotation
is non-trivial, we can rotate our fixed point β to be arbitrarily close to any other γ ∈ T with αn . This
means {αn } is dense in T.

For the third part, we claim that the Lebesgue measure m is the unique borel measure on R such that
m([0, 1]) = 1 and is translation invariant. Recall by Stone Weiestrass that trignometric polynomials are
137

dense in C([0, 1]) under the sup-norm. So we first show that if µ is another translation invariant measure
such that µ([0, 1]) = 1 then
ˆ 1 (
2πikx 1 if k = 0
e dµ(x) =
0 0 else
Note that the k = 0 case is trivial since we imposed µ([0, 1]) = 1. Indeed observe that
ˆ 1 ˆ 1/2 ˆ 1 ˆ 1/2 ˆ 1/2
cos(2πx)dµ(x) = cos(2πx)dµ(x)+ cos(2πx)dµ(x) = cos(2πx)dµ(x)− cos(2πx)dµ(x) = 0
0 0 1/2 0 0

where for the third equality we used the translation invariance of µ to change the integration to 1/2 to 1
combined with cos(2π(x + 1/2)) = − cos(2πx). The same trick works for e2πikx of splitting the integral
to where cos(2πkx) or sin(2πkx) changes sign and using translation invariance lets us deduce that
ˆ 1 (
2πikx 1 if k = 0
e dµ(x) =
0 0 else
Therefore, we see that for any trignometric polynomial P that
ˆ 1 ˆ 1
P (x)dµ(x) = P (x)dm(x)
0 0

so Stone Weiestrass implies that for all f ∈ C([0, 1]) that

ˆ 1 ˆ 1
f (x)dµ(x) = f (x)dm(x)
0 0

Therefore, µ = m.

So by this lemma it suffices to show that if µ is an α−invariant Borel probability measure on [0, 1]
then it is translation invariant. Indeed, observe if E ⊂ T is Borel then observe for all n ∈ N
n n−1
µ(Rα (E)) = µ(Rα (E)) = ... = µ(E)
where the power to n indicates we do do the rotation n times. Then by part b the set {αn } is dense in
T, so if β ∈ S 1 we can find a sequence nk such that αnk → β. And we have
nk
µ(Rα E) = µ(E) for all k
Then to take the limit observe that
ˆ
nk
µ(Rα E) = χαnk E (x)dµ(x)
T

and this is dominated by χT (x) lets us use the DCT combined with χαnk E (x) → χβE (x) to conclude
nk
µ(Rβ E) = lim µ(Rα E) = µ(E)
k→∞

so µ is a translation invariant measure such that µ([0, 1]) = 1 so it follows that µ is the Lebesgue measure.

Problem 5. Let {fn } be a sequence of continuous real valued functions such that fn → f point-
wise everywhere on [0, 1].
(1) Prove that for every ε > 0 there is a dense set Dε such that if x ∈ Dε then there is an
open interval I 3 x and a positive integer Nx such that if n ≥ Nx then
sup |fn (y) − f (y)| < ε
y∈I

(2) Prove that f cannot be the characteristic function of χQ∩[0,1]

138 RAYMOND CHU

Proof. For each N ∈ N and ε > 0 define

\
FN,ε := {x ∈ [0, 1] : |fn (x) − fm (x)| ≤ ε}
n,m≥N

then FN,ε is closed since each fn is continuous. And pointwise convergence everywhere implies that
∞
[
[0, 1] = FN,ε
N =1
so by Baire Categroy Theorem, we know that
∞
[
Dε := Int(FN,ε )
N =1

is open and non-empty. Now we claim Dε is dense; indeed, if (a, b) is an open interval then
∞
[
[a, b] = [a, b] ∩ Fn,ε
n=1

so again by Baire we now that there is some M such that Int([a, b] ∩ FM,ε ) 6= ∅ this implies that
(a, b) ∩ FM,ε 6= ∅
so Dε is dense. And for each x ∈ Dε by definition there is some Nx such that x ∈ Int(FNx ,ε ) so there is
some δ > 0 such that Bδ (x) ∈ Dε and
sup |fn (y) − fm (y)| ≤ ε for all n, m ≥ Nx
y∈Bδ (x)

using pointwise convergence implies that

sup |fn (y) − f (y)| ≤ ε for all n, m ≥ Nx
y∈Bδ (x)

as desired.

For the second part, suppose C([0, 1]) 3 fn → χQ∩[0,1] everywhere. Then let x ∈ [0, 1], so by part 1
there is some interval I and integer Nx such that if n ≥ Nx then
sup |fn (y) − f (y)| ≤ 1/3
y∈I

so by density of Q there is some r ∈ R \ Q ∩ I so f (q) = 0 this implies

sup |fn (r)| ≤ 1/3
q∈I∩Q

so by density of R \ Q in I and fn being continuous we deduce that

sup |fn (r)| ≤ 1/3
y∈I

so if q ∈ Q ∩ I we have
|fn (q) − f (q)| ≥ |f (q)| − |fn (q)| ≥ 1 − 1/3 = 2/3
so we cannot have fn → f everywhere.

Problem 6. Let f in L2 (R) and assume that the fourier transform satisfies |fˆ(ξ)| > 0 a.e. Prove
that the set of finite linear combinations of translates fa (x) := f (x − a) is norm dense in L2 (R)

Proof. See Spring 2012 Problem 6.

Problem 7. Let f (z) be an analytic function on the entire plane such that U (z) := log |f (z)| is
in L1 (C, dA) . Prove that f is constant.
139

Proof. See Spring 2013 Problem 7.

Problem 8. Let D be the space of analytic functions on D such that f (0) = 0 and f 0 ∈ L2 (D, dA)
(1) Prove that D is complete in the norm
ˆ 1/2
||f || := |f 0 (z)|2
D
(2) Give a necessary and sufficient condition on the coefficients an for the function
∞
X
f (z) := an z n
n≥1

to belong to D.

Proof. Let fn ⊂ D be a Cauchy Sequence in || · ||. Now we use that for any compact set K ⊂ D that
||f ||L∞ (K) ≤ C||f ||L2 (D)
where C = C(K) is a constant that depends only on K and not on f . This is shown by Mean Value
Property combined with integration (for instance see Fall 2013 Number 5 for a proof of this lemma).
Therefore, we have for f ∈ D that
||f 0 ||L∞ (K) ≤ C||f ||
so we see that
||fn0 − fm
0
||L∞ (K) ≤ C(K)||fn − fm ||
so as the space C(K) is complete we see there exists a function g ∈ C(D) such that fn0 → g locally
uniformly on D. In particular, g is holomorphic. And as {fn0 } is a Cauchy Sequence in L2 , which is
complete we know that there is a limit h ∈ L2 , so it follows from uniqueness of limits that g = h. So in
particular, g ∈ L2 (D) and is holomorphic on D.

As D is simply connected by Cauchy’s Theorem, there exists a primitive of g which we denote by f

such that f (0) = 0 (since primitives are determined uniquely up to constants). In particular as f 0 = g
we deduce that f is holomorphic with f (0) = 0 and f 0 ∈ L2 (D, dA) so f ∈ D. And we have
||fn − f || = ||fn0 − g||L2 (D) → 0
where the last line is due to our definition of g and f ∈ D. Therefore, D is complete in the norm || · ||.

For the second part, notice f (reiθ ) = n inθ

P
n≥1 an r e and we have f (0) = 0 so it suffices to find
equivalent conditions for the derivative of the sum being in L2 (D). By local uniform convergence, we
know that we can differentiate term by term to see
X
f 0 (reiθ ) = nan rn−1 ei(n−1)θ
n≥1
0
so we have by local uniform convergence and f ∈ L2 (C, dA) that
ˆ ˆ 1 ˆ 2π X ˆ 1 ˆ 2π
0 0
2
|f (z)| dA(z) = iθ 2
r|f (re )| dθdr = r(nan rn−1 ei(n−1)θ )(mam rn−1 e−i(n−1)θ )
D r=0 θ=0 n,m≥1 r=0 θ=0

Xˆ 1
= 2π r2n−1 n2 |an |2 dr
n≥1 r=0
inθ
where for the final equality we used e is an orthogonal family in [0, 2π]. So
X
||f 0 ||2L2 (D) = π n|an |2
n≥1

so we have X
f 0 ∈ L2 (D) ⇐⇒ n|an |2 < ∞
n≥1
140 RAYMOND CHU

which is the desired equivalence

Problem 9. Consider the meromoprhic function g(z) = −πz cot(πz) on the entire plane C.
(1) Find all poles of g and determine the
Presidues of g at each pole.
∞
(2) In the Taylor series representation k=0 ak z k of g(z) about z = 0, show that for each
k≥1
X 2
a2k =
n2k
n≥1

Proof. The poles are at n ∈ Z \ {0} and are of order 1. Now fix n ∈ Z \ {0} then we have
cos(πz)
Res(g, n) = lim −π(z − n)z = −n
z→n sin(πz)
For the second part, recall that
1 X 2z X 2z 2
π cot(πz) = + 2 2
⇒ g(z) = −1 −
z z −k z 2 − k2
k≥1 k≥1

2z 2
P
so if we define h(z) := −1 − k≥1 z−k 2 then h(z ) = g(z) and h is holomorphic since the sum converges

locally uniformly away from n ∈ Z \ {0}. In particular, observe that the 2kth taylor coefficient of g is the
kth taylor coefficient of h(z). Now observe
n
X 2 (n)
X n dj dn−j
h(z) = −1 − z 2 2
⇒ h (z) = − ( j (z)) n−j f (z)
z −k j=0
x dz dz
k≥1
2
P
where f (z) := k≥1 z 2 −k2 . So in particular,

dn−1
h(n) (0) = −2n f (z)
dz n−1
since the sum for f (z) converges locally uniformly we can differentiate term by term to see that
X 2
f (n) (z) = (−1)n n!
(z − k 2 )n+1
k≥1

so
X 2
h(n) (0) = n!
k 2n
k≥1

so we see that
X 2
a2k =
k 2n
k≥1

as desired.

Problem 10. For −1 < β < 1 evaluate

ˆ ∞
xβ
dx
0 1 + x2

Proof. See Spring 2014 Number 11.

141

Problem 11. An analytic Jordan Curve is a set of the form

Γ = f ({|z| = 1})
where f is analytic and 1 − 1 on an annulus {r < |z| < 1/r}, 0 < r < 1.

Let C∗ = C ∪ {∞} be the Riemann sphere, let N < ∞, and let Ω ⊂ C∗ be a domain for which
∂Ω has N connected components, none of which are single points. Prove there is a conformal
mapping from Ω onto a domain bounded by N pairwise disjoint analytic Jordan Curves.

Proof.

Problem 12. If α ∈ C satisfies 0 < |α| < 1 and if n ∈ N show that the equation
ez (z − 1)n = α
has exactly n simple roots in the half plane {z : Rez > 0}.

Proof. Let us first show all the roots are simple of

f (z) := ez (z − 1)n − α
So if z0 is a repeated root we know that f (z0 ) = f 0 (z0 ) = 0, so by differentiation,

ez0 (z0 − 1)n + ez0 (n)(z0 − 1)n−1 = 0

and as 0 < |α| we know that z0 − 1 6= 0 so
(z0 − 1) + n = 0 ⇒ z0 = 1 − n
Now we claim that
|ez0 (z0 − 1)n | ≥ 1
which will imply all the roots are simple since 0 < |α| < 1. Indeed, observe that
n n 1
|e1−n (−n)n | ≥ 1 ⇐⇒ ≥
e e
we note the equality is true for n = 1 and by differentiation in n we see the left hand inequality is
increasing in n, so we have the desired inequality. Therefore, every root is simple since f (z0 ) 6= 0.

Now it suffices to show there are n roots in the half plane {z : Re(z) > 0}. To do this we parameterize
γR as a square of length R with center R for R > 1. It is easy to see that on the boundary of this contour
that |ez (z − 1)n | ≥ 1 > |α| so by Rouche’s theorem,
ez (z − 1)n − α
has the same number of zeros as ez (z − 1) in this square. Therefore, by taking R → ∞ we see in this half
plane there are n roots in this half plane.

142 RAYMOND CHU

19. Spring 2019

Problem 1. Let f ∈ C 2 (R) be a real valued function that is uniformly bounded on R. Prove
that there exists a point c ∈ R such that f 00 (c) = 0.

Proof. Assume for the sake of contradiction that f 00 (x) 6= 0 for any x ∈ R, so as f ∈ C 2 we know that
either f 00 (x) > 0 or f 00 (x) < 0 for all x ∈ R. Without loss of generarlity, by looking at −f if necessary we
can assume that f 00 (x) > 0 for all x ∈ R. Then f (x) is convex and as f 00 (x) > 0 we know that f 0 (x) is
strictly increasing, so there is some x0 ∈ R such that f 0 (x0 ) 6= 0. Then by a first order taylor series with
remainder term expansion we see that
f (x) ≥ f (x0 ) + f 0 (x0 )(x − x0 ) for all x ∈ R
i.e. convex functions lie above their tangent lines. Now if f 0 (x0 ) > 0 then taking x → ∞ gives
limx→∞ f (x) ≥ ∞ so f is not uniformly bounded. And if f 0 (x0 ) < 0 then taking x → −∞ implies
limx→−∞ f (x) ≥ ∞ so f is unbounded in either case, so we have arrived at a contradiction. So f 00 (c) = 0
for some c ∈ R.

Problem 2. Let µ be a Borel probability measure on [0, 1] that has no atoms. Let also µ1 , µ2 , ..
be Borel probability measures on [0, 1] such that µn weak* converges to µ. Denote F (t) := µ([0, t])
and Fn (t) := µn ([0, t]) for each n ≥ 1 and t ∈ [0, 1]. Prove that Fn converges uniformly to F .

Proof. Note that for each fixed t ∈ [0, 1] that we can find a sequence of continuous functions fn , gn ∈
C([0, 1]) such that
fn (x) ≤ χ[0,t] (x) ≤ gn (x)
with fn → χ[0,t] (x) and gn → χ[0,t] (x) pointwise, so in particular, we have
ˆ 1 ˆ 1
fn (x)dµm (x) ≤ Fm (t) ≤ gn (x)dµm
0 0
so taking m → ∞ along with µn weak* converges to µ gives
( ´1
lim supn→∞ Fn (t) ≤ 0 gn (x)dµ
´1
f (x)dµ ≤ lim inf n→∞ Fn (t)
0 n
so now using DCT and MCT we see that
ˆ 1 ˆ 1
lim fn (x)dµ = lim gn (x)dµ = F (t)
n→∞ 0 n→∞ 0
so we deduce that
lim Fn (t) = F (t)
n→∞
so we have pointwise convergence everywhere. A small modification of the above proof shows that
µn ([a, b]) → µ([a, b]) for all a ≤ b.

Notice each Fn (t) is a monotone function that converges to a continuous function F (t) (it is continuous
since it has no atoms), so we have Fn → F uniformly since [0, 1] is compact. Indeed, as F is continuous
on [0, 1] if ε > 0 there is a δ > 0 such that if x, y ∈ [0, 1] and |x − y| < δ then |F (x) − F (y)| < ε/2. Now
as fix q ∈ Q ∩ [0, 1] then by pointwise convergence we can find an N (q) such that if n ≥ N (q) then
|fn (q) − F (q)| ≤ ε/2
S∞
and as Q ∩ [0, 1] is dense in [0, 1] we know that [0, 1] ⊂ n=1 Bδ/2 (qn ) where qn is an enumeration of
SN
Q. Therefore, we may find a finite subcover n=1 Bδ/2 (qn ) and let M := max1≤j≤N N (qn ). Then for
m ≥ M , if x ∈ Bδ/2 (qn ) such that if x ≤ qn we have from Fn being monotone increasing that (where we
reorder qn such that qn < qn+1 )
Fm (x) − F (qn ) ≤ Fm (qn ) − F (qn ) ≤ ε/2
143

but we also have

Fm (x) − F (qn ) ≥ Fm (qn−1 ) − F (qn ) ≥ −ε/2
so it follows that
|Fm (x) − F (qn )| ≤ ε/2
for all m ≥ M . Now observe if n ≥ M that

|Fn (x) − F (x)| ≤ |Fn (x) − F (qm )| + |F (qm ) − F (x)| ≤ ε/2 + ε/2 = ε

where x ∈ Bδ (qm ). The case of x ≥ qn is also similar. Therefore, we have shown Fn → F uniformly.

Problem 3. Let f (t) be a positive continuous function such that lim|t|→∞ f (t) = 0. Show that
the set {hf : h ∈ L1 (R, m), ||h||1 ≤ K} is a closed nowhere dense set in L1 (R, m), for any K ≥ 1.

Let {fn } be a sequence of positive continuous function on R such that for each n we have
lim|t|→∞ f (t) = 0. Show that there exists g ∈ L1 (R, m) such that g/fn ∈/ L1 (R, m) for all
n.

Proof. Denote E := {hf : h ∈ L1 (R, m), ||h||1 ≤ K} then if E 3 fn → f in L1 then we have for each fn
that fn = gn h where gn ∈ L1 (R, m) with ||gn ||1 ≤ K. So by looking at a subsequence if necessary, we
have that gn → f /h pointwise, so Fatou’s Lemma gives
ˆ ˆ ˆ
|f /h|dx = lim inf |gn |dx ≤ lim inf |gn |dx ≤ K
R R n→∞ n→∞ R

so it follows that f ∈ E. To see that it has empty interior, observe that if f ∈ E and ε > 0 that for any
N ∈ N we have that for gM (x) := f (x) + (ε/2)χ[M,M +1] (x)
ˆ M +1
||f (x) − gM (x)||1 = ε/2dx = ε/2
M

i.e. gM (x) ∈ Bε (f ). Now observe that

ˆ ˆ M +1
ε
|gM (x)/h(x)|dx = dx
R M 2|h(x)|
so now as lim|t|→∞ f (t) = 0, we can find an M ∈ N such that if δ > 0 then on [M, M + 1] we have
|f (t)| < δ. So in particular, we conclude
ˆ
ε
|gM (x)/h(x)|dx ≥
R 2δ
now choose δ so small such that ε/(2δ) > K + 1. Thus it follows that gM (x) ∈
/ E. So E has empty
interior in L1 .

For the second part define En,M := {hfn : h ∈ L1 (R, m), ||h||1 ≤ M } where n, M ∈ N. Then En,M is
a closed set with empty interior. Then as L1 (R, m) is complete we know that L1 is not meager i.e.
[
L1 (R, m) 6= En,M
n,M ∈N

so in particular there is some g ∈ L1 (R, m) such that G ∈ c

T
m,M ∈N En,M i.e. for all n, M ∈ N ||g/fn || ≥ M
/ L1 (R, m) for any n ∈ N.
i.e. g/fn ∈

144 RAYMOND CHU

Problem 4. Let V be the subspace of L∞ ([0, 1], µ) (where µ is the Lebesgue measure on [0, 1])
defined by ˆ
V := {f ∈ L∞ ([0, 1], µ) : lim n f (x)dµ exists }
n→∞ [0,1/n]
Prove that there
´ is a continuous linear functional ϕ ∈ L∞ ([0, 1], µ)∗ such that
ϕ(f ) = limn→∞ [0,1/n] f (x)dµ for every f ∈ V.

Show that, given any ϕ ∈ L∞ ([0, 1], µ)∗ satisfying the condition above that there is no g ∈
´1
L1 ([0, 1], µ) such that ϕ(f ) = 0 f (x)g(x)dµ for all f ∈ L∞ ([0, 1], µ)

Proof. Define the linear functional ϕ : V → R via

ˆ
ϕ(f ) = lim n f (x)dµ
n→∞ [0,1/n]

and this is linear and well defined since f ∈ V. It is also continuous since
ˆ ˆ
n f (x)dµ ≤ n |f (x)|dµ ≤ ||f ||L∞
[0,1/n] [0,1/n]

hence by taking limits

|ϕ(f )| ≤ ||f ||L∞
so ϕ ∈ V . So by Hanh-Banach we can extend ϕ to a linear functional φ on L∞ ([0, 1], µ) such that the
∗

||φ|| = ||ϕ|| where the operator norm on the left is taken over L∞ ([0, 1], µ) and the right is over V. So
this extension is continuous and this proves the first part.

For the second part note that over C([0, 1]) ⊂ L∞ ([0, 1], µ) that ϕ|C([0,1]) = δ0 where δ0 means the
dirac delta at zero since if f ∈ C([0, 1])
ˆ
f (0) = lim f (x)dµ
n→∞ [0,1/n]

So now if we assume for the sake of contradiction that there is some g ∈ L1 ([0, 1], µ) such that ϕ(f ) =
´1
0
f (x)g(x)dµ. Then for any y ∈ (0, 1) we have for h > 0 small enough that (y − h, y + h) ⊂ (0, 1) and we
can find a continuous function fn ≥ χ[y−h,y+h] such that fn → χ[y−h,y+h] pointwise with fn ∈ C([0, 1])
with ||fn ||L∞ ≤ 2 so we have from the dominated convergence theorem
ˆ 1 ˆ y+h
fn (0) = fn (x)g(x)dx → g(x)dx
0 y−h

where the first equality is due to ϕ(f ) = f (0) for all f ∈ C([0, 1]). In particular, we have
ˆ y+h ˆ y+h
1
0= g(x)dx ⇒ lim g(x)dx = 0
y−h h→0 2h y−h

so by Lebesgue differentiation Theorem we have g(x) = 0 a.e., but

ˆ 1
1 = ϕ(1) = g(x)dx = 0
0
which is our contradiction.

Problem 5. Prove that Lp ([0, 1], µ) are separable Banach spaces for 1 ≤ p < ∞ but L∞ ([0, 1], µ)
is not (µ is the Lebesgue measure on [0, 1]).

Also prove that there is no linear bounded surjective map T : Lp ([0, 1], µ) → L1 ([0, 1], µ), if p > 1.
145

Proof. We claim that for 1 ≤ p < ∞ that step functions with rational coefficients over rational intervals
are dense in Lp ([0, 1], µ) for 1 ≤ p < ∞. Indeed, first fix E ⊂ [0, 1] µ-measurable, then as µ is outer
regular, we know
ˆ 1
χE dµ(x) = µ(E) = inf{µ(G) : E ⊂ G and G is open}
0
S∞
Now let E ⊂ G where G is open. Then since [0, 1] is separable,
Pn we know that G = i=1 Ii where Ii are
disjoint open intervals. So then we have µ(G) = limn→∞ j=1 µ(Ij ), so if ε > 0 we can find some N ∈ N
Pn
such that if n ≥ N |µ(G) − j=1 µ(Ij )| < ε . Now each Ii we can thanks to density of rationals on [0, 1]
a rational interval Jj such that Ij ⊂ Jj and µ(Jj \ Ij ) ≤ 2εj , so it follows that And
ˆ 1 Xn
!1/p ˆ 1X n
!1/p ˆ 1 X∞
!1/p
p p p
|χG (x) − χJi | dµ(x) ≤ |χIi − χJi | + |χIi |
0 i=1 0 i=1 0 i=n+1

n
!1/p ∞
!1/p
X X
≤ µ(Ji \ Ii ) + µ(Ii )
i=1 i=n

≤ 2ε1/p
So now it follows that we can approximate simple functions arbitrarily well with finite rational combina-
tions of step functions on rational intervals in Lp for p ∈ [1, ∞) and as this set of functions is countable,
we deduce Lp ([0, 1], µ) is separable for p ∈ [1, ∞).

L∞ ([0, 1]) is not separable since for every irrational p ∈ [1/4, 3/4] \ Q we can define for 0 < ε < 1/8
fp (x) := χ[p−ε,p+ε] (x). Then we have for p, q distinct irrationals in [1/4, 3/4] that
||fp (x) − fq (x)||L∞ ([0,1]) = 1
But there are uncountably many such functions, which implies L∞ ([0, 1]) is uncountable.

For the second part, assume for the sake of contradiction that such a T existed, then we define its
adjoint T ∗ : (L1 ([0, 1], µ))∗ ∼
= L∞ ([0, 1], µ) → (Lp ([0, 1], µ))∗ ∼
= Lq ([0, 1], µ) where 1/p + 1/q = 1 via
T ∗ (f ) := f (T ) : Lp ([0, 1], µ) → R
Now as T is surjective we claim that T ∗ is injective. Indeed, as the adjoint is linear it suffices to show
if T ∗ (g) = 0 then g = 0. Indeed, observe T ∗ (g) = g(T ) so as T is surjective, we know this means for
all f ∈ L1 ([0, 1], µ) that g(f ) = 0 so by Hanh-Banach, we must have that g ∈ (L1 ([0, 1], µ))∗ is the zero
functional. So T ∗ is injective. So we have that L∞ ([0, 1], µ) is a isomorphic to a subset of Lq ([0, 1], µ) i.e.
T ∗ (L∞ ([0, 1], µ)) (note T ∗ is bounded). Notice that the first space is separable, while the second is not.
This contradicts the isomorphism, so no such maps exist.

Problem 6. Let H be a Hilbert space and {ξn }n a sequence of vectors in H such that ||ξn || = 1
for all n.

Show that if {ξn } converges weakly to a vector ξ ∈ H with ||ξ|| = 1, then limn→∞ ||ξn − ξ|| = 0.

Show that if limn,m→∞ ||ξn + ξm || = 2,then there exists a vector ξ ∈ H such that
lim ||ξn − ξ|| = 0
n→∞

Proof. For the first part, by the Parallelogram law, we have that
||ξn + ξ||2 + ||ξn − ξ||2 = 2||ξn ||2 + 2||ξ||2 = 4
Now observe that
||ξn + ξ||2 = (ξn + ξ, ξn + ξ) = ||ξn ||2 + ||ξ||2 + (ξn , ξ) + (ξ, ξn ) = 2 + (ξn , ξ) + (ξ, ξn )
146 RAYMOND CHU

Now as ξn * ξ we know that for any g ∈ H that (ξn , g) → (ξ, g). Taking g = ξ gives that
||ξn + ξ||2 → 4 as n → ∞
that is
||ξn − ξ||2 → 0 as n → ∞
For the second part, notice again by Parallelogram Law that
||ξn + ξm ||2 + ||ξn − ξ||2 = 2||ξn ||2 + 2||ξm ||2 = 4
so
||ξn − ξm ||2 = lim 4 − ||ξn + ξm ||2 = 0
n,m→∞

where the last equality is by the given assumpution. Now that H is complete we conclude since {ξn }
forms a Cauchy Sequence.

Problem 7. Let f : C → C be entire non-constant, and let us set

ˆ 2π
1
T (r) := log+ |f (reiϕ )|dϕ
2π 0
Here log+ (s) = max(log(s), 0). Show that T (r) → ∞ as r → ∞

Proof. By Jensen’s formula if we use ρ to denote the zeros of f and 0 is not a root of f we have
ˆ 2π
X r 1
log |f (0)| + log | | = log |f (reiϕ )|dϕ ≤ T (r)
ρ 2π 0
|ρ|≤r

So in particular, if f has at least one zero denoted by ξ then for r large enough we have
r
T (r) ≥ log |f (0)| + log → ∞ as r → ∞
ξ
so this proves the theorem when f (0) 6= 0 and f has at least one zero.

Now if f has no zeros, we know there is an entire function g(z) such that f (z) = exp(g(z)). Then
computation gives log |f (z)| = Re(g(z)) which is a harmonic function. So it suffices to show if u : C → R
is harmonic then ˆ 2π
1
u+ (reiϕ )dϕ → ∞ as r → ∞
2π 0
WLOG u(0) 6= 0 by adding a constant if necessary. Then by the mean value theorem we know that for
r>0 ˆ
1
|u(0)| ≤ 2 |u(z)|dA(z)
πr Br (0)
so it follows that if 0 < ε 1 that for R large that
ˆ
|u(z)|dA(z) ≥ R2−ε
BR (0)

otherwise we could find a sequence {Rn } such that Rn → ∞ and

1
|u(0)| ≤ →0
πRnε
´
So now we have two cases either R large or there is an ε > 0 such that BR (0) |u(z)|dA(z) & R2+ε for R
´
large or no such ε exist i.e. Br (0) |u(z)|dA(z) . R2 . For the first case, observe that
ˆ ˆ 2π ˆ R ˆ R
u+ (z)dA(z) = r(u+ (reiθ ))drdθ = rT (r)dr & R2+ε
BR (0) θ=0 r=0 r=0
147

where for the last equality follows from

ˆ
u(0) = (u+ − u− )dA(z)
BR (0)

where u− := − min{u, 0} and |u| = u+ + u− to deduce the above inequality. From which it follows that
T (r) → ∞ as R → ∞ since if T (r) is bounded on some subsequence Rn → ∞ the above inequality will
be invalid. Now if ˆ
|u(z)|dA(z) . R2
BR (0)

this implies for any 0 < r < R that

|u(z)| .r R2
where z ∈ Br (0), so it follows that u is at most a polynomial of degree 2 at most, from which computation
implies the result.

Now if f (z) has a zero at the origin, we know that f (z)/z m for some m is entire with no-zeros at the
origin (take m to be the order of the zero). Then we can repeat the computation above to get
ˆ m
1
log+ |f (reiθ )/rm |dϕ → ∞
2π 0
so in particular, as log |f (rei )/rm | = log |f (reiθ )| − log |rm | ≤ log |f (reiθ )| for r ≥ 1 so it follows that
ˆ m
1
log+ |f (reiθ )|dϕ → ∞
2π 0

Problem 8. Show that

∞
z3 Y z2

sin(z) − z cos(z) = 1 − 2 ,z ∈ C
3 n=1 λn
where λn is a sequence in C such that λn 6= 0 or all n and
X
|λn |−2 < ∞
n∈N

Proof. Observe that f (z) := sin(z) − z cos(z) is an entire function of order 1, so by Hadamard’s factor-
ization theorem, we know if {λn } are the zeros of f (z) with λn 6= 0 that
∞
Y z z/λn
f (z) = exp(az + b)z 3 (1 − )e
n=1
λn

since f is an entire function of order 1 with a zero of order 3 at z = 0 where the product converges locally
uniformly on C. We also know there are infinitely many roots since if f (z) had finitely many roots say
QM
z1 , .., zM then f (z) = j=1 (z − zi ) exp(g(z)) where g(z) is an entire function and f (z) is not of this form.
Now that f (z) = −f (−z) combined with λn being the zeros of f (z) (with λn 6= 0) shows there is an m
such that λn = −λm , so using the local uniform convergence let us rearrange terms in the sum to get
∞
Y z2
f (z) = exp(az + b)z 3 (1 − )
n=1
λ2n
z/λn z/λm 2
since (1 − z/λn )e (1 − z/λm )e = (1 − z /λ2n ) because λm = −λn . Now using oddness of f again
gives
∞ ∞
3 z2
Y
3
Y z2
exp(az + b)z (1 − 2 ) = exp(−az + b)z (1 − 2 )
n=1
λn n=1
λn
148 RAYMOND CHU

so it followsPthat a = 0. So now it suffices to show that b = log(1/3) since Jensen’s formula gives for any
ε > 0 that n∈N |λn |−1−ε < ∞ and specifying ε = 1 gives the desired sum bound. Now we differentiate
f (z) to see that
f 000 (z) = 6 exp(b) + zh(z)
where h(z) is a holomorphic function. Now using f 000 (0) = 2 gives exp(b) = 1/3 i.e.
∞
Y z2
f (z) = z 3 /3 (1 − )
n=1
λ2n
as desired and the sum properties hold.

Problem 9. Show that if A(D) is the space of holomorphic functions on D and

U := {f ∈ A(D) : |f | = 1 on ∂D}
then show f ∈ U if and only if
N
Y z − aj
f (z) = λ
j=1
1 − aj z
for some aj ∈ D and 1 ≤ j ≤ N < ∞ and |λ| = 1

Proof. Let aj ∈ D then for |z| = 1 then z = eiθ for some θ so

aj
|z − aj | = |z||1 − | = |1 − zaj | = |1 − zaj |
z
i.e. |z − aj |/|1 − aj z| = 1 on ∂D and this is holomorphic since 1/aj ∈ / D because 1/|aj | > 1. Therefore,
N
Y z − aj
λ
j=1
1 − aj z

with ak ∈ D and |λ| = 1 with finite N is in U .

For the converse, let f ∈ U , then notice as f is holomorphic its zeros are isolated and it has no
zeros on ∂D, so there are only finitely many zeros of f on D. Enumerate them as {aj }N j=1 then g(z) :=
QN z−aj
f (z)/ j=1 1−aj z is holomorphic on D with no zeros and is norm 1 on the boundary of the disk. Therefore,
by applying the maximum modulus principle on 1/g combined with the maximum modulus principle on
g we see that |g| = 1 everywhere on D. In particular, the Cauchy Riemann Equations then imply g is
constant so g = λ where || = 1 i.e.
N
Y z − aj
f (z) = λ
j=1
1 − aj z
for some aj ∈ D with 1 ≤ j ≤ N < ∞ and |λ| = 1.

Problem 10. For a > 0, b > 0, evaluate the integral

ˆ ∞
log x
dx
0 (x + a)2 + b2

Problem 11. Let u ∈ C ∞ (R) be a smooth 2π-periodic function. Show that there exists a
bounded holomorphic function f+ in the upper half-plane Imz > 0 and a bounded holomorphic
function f− in the lower half plane Imz < 0, such that
u(x) = lim (f+ (x + iε) − f− (x − iε))
ε→0+
149

Proof. As u is C ∞ and 2 π-periodic we know that

X ˆ 1
u(x) = an einx where an = u(x)e−inx
n∈Z 0

where the sum converges absolutely i.e. the Fourier Series of u agrees with u. Therefore, we can define
1X 1X
f+ (z) := an einz and f− (z) := − an e−inz
2 2
n∈Z n∈Z

then f+ is a bounded holomorphic function in the upper half plane and f− is a bounded function on
the lower half plane. (Note they are holomorphic since the sums are locally uniformly convergent in the
upper and lower half plane respectively). Then observe that
1X 1X
f+ (x + iε) − f− (x − iε) = an (ein(x+iε) + ein(x−iε) ) = an einx (e−nε + enε )
2 2
n∈Z n∈Z

Now we recall that the sum is absolutely convergent on R so f+ converges absolutely on 0 < Im(z) < M
for any M > 0 and similarily for f− , so we can take the limit ε → 0 inside the sum to get
lim f+ (x + iε) − f− (x − iε) = f (x)
ε→0
as desired.

Problem 12. Let H be the vector space of entire functions f : C → C such that
ˆ
|f (z)|2 dµ(z) < ∞
C
2
Here dµ(z) = e−|z| dλ(z) where λ is the Lebesgue measure on C.
(1) Show that H is a closed subspace of L2 (C, dµ).
(2) Show that for all f ∈ H we have
ˆ
1
f (z) = f (w)ezw dµ(w)
π C

Proof. Refer to Fall 2014 Number 10 for a solution of (1).

Observe that if (f (z), g(z)) = f (w) where (, ) is the inner product on L2 (dµ) then
∞
X
(f (z), g(z)) = an wn
n=0
n
P
where an z is P
the Taylor Expansion of f (z) (which is possible since f is entire). As g(z) is entire we
can write g(z) = bn z n then we have
ˆ R ˆ 2π X X 2
(f, g) = lim r( an rn einθ )( bn rn e−inθ )e−|r| dθdr
R→∞ r=0 θ=0
and by uniform convergence of the sums we have
ˆ R ˆ 2π X ∞ ˆ R ˆ 2π
X 2 X 2
r( an rn einθ )( bn rn e−inθ )e−|r| dθdr = r2n+1 an bm ei(n−m)θ e−|r| dθdr
r=0 θ=0 n,m=0 r=0 θ=0

now orthogonality of einθ tells us

∞ ˆ R ∞ ˆ R2 ∞
X 2 X X
= 2π r2n+1 an bn e−|r| dr = πan bn rn e−r dr → πan bn n!
n=0 r=0 n=0 r=0 n=0
so
∞
X ∞
X
n
an w = (f, g) = πan bn n!
n=0 n=0
150 RAYMOND CHU

1 wn
so bn = π n! i.e.
∞
X 1 wn n 1
g(z) = z = ezw
n=0
π n! π
as desired. Note that all of the above formal computation is justified since our g(z) is entire so the
convergence of the sums is uniform.

151

20. Fall 2019

Problem 1. Given σ -finite measures µ1 , µ2 , ν1 , ν2 on a measurable space (X, X ), suppose that

µi νi for i = 1, 2. Prove that the product measure µ1 ⊗ µ2 and ν1 ⊗ ν2 on (X × X, X ⊗ X )
satisfy µ1 ⊗ µ2 ν1 ⊗ ν2 and the Random-Nikodym derivatives obey
d(µ1 ⊗ µ2 ) dµ1 dµ2
(x, y) = (x) (y)
d(ν1 ⊗ ν2 ) dν1 dν2
for ν1 ⊗ ν2 a.e. (x, y) ∈ X × X.

Proof. By Radom-Nikodym as µi νi and all the measures are σ-finite that there exists fi ∈ L1 (X, dνi )
such that for any E ∈ X we have for i = 1, 2
ˆ
µi (E) = fi (x)dνi (x)
E
So it follows that if Ei ∈ X that by the definition of the product measure that
ˆ ˆ
µ1 ⊗ µ2 (E1 × E2 ) = µ1 (E1 ) · µ2 (E2 ) = f1 (x)dν1 (x) f2 (y)dν2 (y)
E1 E2
ˆ
= f1 (x)f2 (y)d(ν1 ⊗ ν2 )(x, y) = (f1 (x)f2 (y))d(ν1 ⊗ ν2 )(E1 × E2 )
E1 ×E2
where the last equality is justified by Fubini since f1 (x)f2 (y) ∈ L1 (d(ν1 ⊗ ν2 )) and the measures being
σ-finite. So as µ1 ⊗ µ2 = (f1 (x)f2 (y))d(ν1 ⊗ ν2 ) for rectangles and rectangles generate the X ⊗ X it
follows that
d(µ1 ⊗ µ2 ) = (f1 (x)f2 (y))d(ν1 ⊗ ν2 )
so in particular, we see that µ1 ⊗ µ2 ν1 ⊗ ν2 and the Random-Nikodym derivative of µ1 ⊗ µ2 is
f1 (x)f2 (y) for ν1 ⊗ ν2 a.e. in (x, y) ∈ X × X as desired.

´ 2. Let µ be a fintie Borel measure on R with µ({x}) = 0 for all x ∈ R and let
Problem
ϕ(t) = R eitx dµ(x). Prove that
ˆ T
1
lim |ϕ(t)|2 dt = 0
T →∞ 2T −T

Proof. Observe that

ˆ T ˆ T ˆ ˆ ˆ T ˆ ˆ
2 itx −ity
|ϕ(t)| dt = e dµ(x) e dµ(y) dt = eit(x−y) dµ(x)dµ(y)dt
−T −T R R −T R R
where the last equality is due to Fubini’s Theorem which can be applied since µ is a finite measure so
eit(x−y) ∈ L1 (dµ(x) ⊗ dµ(y) ⊗ dt, R × R × [−T, T ]) since the integrand is bounded. Applying Fubini again
gives
ˆ ˆ ˆ T ˆ ˆ
sin(T (x − y))
= eit(x−y) dtdµ(x)dµ(y) = 2 dµ(x)dµ(y)
R R −T R R (x − y)
so we have ˆ T ˆ ˆ
1 sin(T (x − y))
|ϕ(t)|2 dt = dµ(x)dµ(y)
2T −T R R T (x − y)
where we define
sin(T (x − y))
:= 1 for x = y
T (x − y)
and this definition makes the integrand continuous. So as the integrand is bounded by a constant and µ
is finite we may apply DCT to see that
ˆ T ˆ ˆ ˆ
1
lim |ϕ(t)|2 dt = χ{x=y} dµ(x)dµ(y) = µ({y})dµ(y) = 0
T →∞ 2T −T R R R
152 RAYMOND CHU

where the last equality is due µ having no atoms.

Problem 3. Consider a measure space (X, X ) with σ-finite measure µ and let p ∈ (1, ∞). Let
Lp,∞ be the set of measurable f : X → R with [f ]p := supt>0 tµ(|f | > t)1/p finite. Let
ˆ
1
||f ||p,∞ := sup 1−1/p
|f |dµ
E∈X ,µ(E)∈(0,∞) µ(E) E

prove that there exists c1 , c2 ∈ (0, ∞)- which may depend on p and µ - such that
∀f ∈ Lp,∞ : c1 [f ]p ≤ ||f ||p,∞ ≤ c2 [f ]p

Proof. Observe that if µ(|f | > t) > 0 then

ˆ
1
||f ||p,∞ ≥ |f |dµ ≥ tµ(|f | > t)1/p
µ({|f | > t})1−1/p {|f |>t}

so taking the supremium implies

||f ||p,∞ ≥ [f ]p
For the reverse inequality fix a E ∈ X with µ(E) ∈ (0, ∞). Then from the Layer Cake Decomposition
ˆ ˆ ∞ ˆ ∞
1 1 1
|f |dµ = µ({x ∈ E : |f (x)| > t})dt ≤ min{µ(E), µ({|f | > t})}
µ(E)1−1/p E µ(E)1−1/p t=0 µ(E)1−1/p t=0
ˆ ∞ ˆ α ˆ ∞
1 p p 1 1
≤ min{µ(E), [f ] p /t } = µ(E)dt + [f ]pp /tp dt
µ(E)1−1/p t=0 µ(E)1−1/p t=0 µ(E)1−1/p α
where α = [f ]p /µ(E)1/p so
1
= [f ]p + [f ]p = Cp ([f ]p )
p−1
as desired

Problem 4. Let A ⊂ R be measurable with positive Lebesgue measure. Prove that the set
´A − A := {z − y : z, y ∈ A} has non-empty interior. Hint: Consider the function ϕ(x) =
χA (x + y)χA (y).

Proof. Assume for the sake of contradiction that A − A has empty interior. Notice that 0 ∈ A − A, so
this means there exists a sequence xn → 0 such that xn ∈/ A − A. Notice by translation continuity of the
Lebesgue Integral that
ˆ ˆ
lim χA (y + xn )χA (y)dy = |χA (y)|2 dy = m(A) > 0
n→∞

where m is the Lebesgue measure. However, observe

ˆ ˆ
χA (y + xn )χA (y)dy = χA (y + xn )dy
A

and y + xn ∈
/ A for any y ∈ A. Indeed if y + xn ∈ A then xn = (y + xn ) − y so xn ∈ A − A. Therefore,
ˆ
χA (y + xn )dy = 0
A

which is a contradiction. Therefore, A − A has non-empty interior.

153

Problem 5. Prove the following claim: LetH be a Hilbert space with he scalar product of x and
y denoted by (x, y) and let A, B : H → H be (everywhere-defined) linear operators with
∀x, y ∈ H : (Bx, y) = (x, Ay)
Then A and B are both bounded (thus continuous).

Proof. Consider the family of

Lx (y) := (Bx, y)
where ||x|| ≤ 1 this is continuous since
|Lx (y)| ≤ ||Bx||(||y||) = C(x)||y||
and by assumption we have
|Lx (y)| = |(x, Ay)| ≤ (||x||)(||Ay||) ≤ ||Ay||
i.e. the family is pointwise bounded. So by Uniform Boundness Principle, we deduce that
sup ||Lx || ≤ C
||x||≤1

for some C. Now observe that if ||x|| ≤ 1

||Bx|| = sup (Bx, y) = sup Lx (y) ≤ C
||y||≤1 ||y||≤1

Therefore, B is continuous and an identical argument shows that A is continuous.

Problem 6. Prove that there exists a continuous linear functional functional φ on `∞ (N) such
that
φ(x) := lim xn
n→∞
whenever the limit exists.

Also show that φ is not unique.

Proof. Let S := {x ∈ `∞ (N) : limn→∞ xn exists } then S is a linear subspace of `∞ (N). So define
φ : S → R via
φ(x) = lim xn
n→∞
Observe also
|φ(x)| ≤ ||xn ||`∞
so by Hanh Banach we can find an extension of φ from S to `∞ such that ||φ|| ≤ 1. This proves that
such a functional exists.

For non-uniqueness, observe that S is closed, so fix any x ∈

/ S then there is a δ > 0 such that
d(x, S) = δ > 0 then define for any y ∈ S
ψα (y + λx) = λα||x||
where α is a constant we’ll choose later. Observe
||y + λx|| = λ||y/λ + x|| ≥ λδ ≥ λ|α|(||x||) = |ψα (y + λx)|
where α is chosen so that |α|||x|| ≤ δ. Then by Hanh Banach this function extends to a continuous
functional on `∞ . Notice ψα = 0 on S, so if we choose α, β > 0 with α 6= β such that |α|||x|| ≤ δ and
|β||x|| ≤ δ then
φ(x) + ψα (x) and φ(x) + ψβ (x)
are two such extensions and they are not equal since α 6= β.

154 RAYMOND CHU

Problem 7. Let J ⊂ R be a compact interval, and let µ be a finite Borel measure whose support
lies in J. For z ∈ C \ J define ˆ
1
Fµ (z) := dµ(t)
R z − t
Prove that the mapping µ 7→ Fµ is one-to-one

Proof. Assume that J = [a, b] where a 6= b and that Fµ1 (z) = Fµ2 (z) then define µ := µ1 − µ2 then we
want to show µ = 0. We have that for z ∈ C \ J
ˆ ˆ b
1 1
Fµ (z) = dµ(t) = dµ(t) = 0
R z − t a z − t
Now we have for any 0 < h < 1/2
ˆ b
Fµ (b + 1 + h) − Fµ 1 1
=− dµ(t)
h h a (b + 1 − t)(b + 1 + h − t)
Notice that the integrand is bounded by 1/(1 + h) ≤ 1 and µ is finite, so by the Dominated Convergence
Theorem we have that
ˆ b
−1
0= 2
dµ(t)
a (b + 1 − t)
and similarily by differentiating again with an identical argument gives for any n ∈ N
ˆ b
1
0= dµ(t)
a (b + 1 − t)n
so let A be the sub-algebra generated by {1/(b + 1 − t), 1/(b + 1 − t)2 , ...} then this family of functions
vanish nowhere [a, b] since 1/(b + 1 − t) 6= 0 on [a, b]. It also separates points since 1/(b + 1 − t) is
injective. Therefore, as [a, b] is compact, Stone Weiestrass tells us A is dense, from which it follows that
if f ∈ C([a, b]) then
ˆ b
f (t)dµ(t) = 0
a
i.e. µ = 0 by Risez Representation Theorem, so the mapping µ 7→ Fµ

Problem 8. A function f : C → C is entire and has the property that |f (z)| = 1 when |z| = 1.
Prove that f (z) = az n for some integer n ≥ 0 and some a ∈ C with |a| = 1.

Proof. See Spring 2016 Number 9.

Problem 9. Determine the number of zeros of the polynomial

P (z) = z 6 − 6z 2 + 10z + 2
in the annulus {z ∈ C : 1 < |z| < 2}. Prove your claim.

Proof. Observe that if A := {z ∈ C : 1 < |z| < 2} then |z 6 + 10z| > |2 − 6z 2 | on ∂A, so by Rouche P (z)
has the same number of zeros on A as z 6 + 10z. But observe if 1 < |z| < 2 that z 6 + 10z = z(z 5 + 10)
so the only roots on A have to be when z 5 + 10 = 0. Let ξi be a root of unit of order 5 i.e. ξi5 = 1 then
−(10)1/5 ξi is a root of z 5 + 10 and there are 5 of them. And as 1 < (10)1/5 < 2 it follows that these roots
are in A, so P (z) has 5 roots on the annulus.

155

Problem 10. Evaluate ˆ x

lim sin(t2 )dt
x→∞ 0
Justify all steps.

2
Proof. Recall that sin(t2 ) = Im(eit ). Let γa→b be the straight line segment from a to b and γR := Reiθ
for θ ∈ [0, π/4] then define
γ := γ0→R + γR − γ0→Reiπ/4
Notice that ˆ ˆ π/4
2
eiz dz = exp(R2 (− sin(2θ) + i cos(2θ)))dθ
γR 0
Notice that
sin(2θ)
=1 lim
2θ θ→0
and so on θ ∈ [0, π/4] we have sin(2θ)/(2θ) ≥ δ > 0 for some δ. Therefore,
ˆ ˆ π/4
iz 2
e dz ≤ | exp(−R2 δ)|dθ = π/4 exp(−R2 δ) → 0 as R → ∞
γR 0
And we also have
ˆ ˆ 1 ˆ 1 ˆ R
iz 2 iπ/4 2 2 iπ/2 iπ/4 2 2 iπ/4
e dz = Re exp(i(t R e )) = Re exp(−t R ) = e exp(−t2 )dt
γ0→Reiπ/4 t=0 0 0
so ˆ √ √
2 π π
lim eiz dz = eiπ/4 = (1 + i) √
R→∞ γ0→Reiπ/4 2 2 2
So it follows from Cauchy Theorem that
ˆ R √
2 π
eit dt = (1 + i) √
lim
R→∞ 0 2 2
so ˆ ∞ ˆ ∞ √
it2
√
Im(e )dt = sin(t2 )dt = π/(2 2)
0 0

Problem 11. Find a conformap map of the domain

√ √
D = {z ∈ C : |z − 1| < 2, |z + 1| < 2}
onto the open unit disc centered at the origin. It suffices to write this map as a composition of
explicit conformal maps.

Proof. Note that the two circle B√2 (1) and B√2 (−1) intersect at z = ±i. So consider the conformal map
√
1− 2−i z+i
ϕ(z) := √ ( )
1− 2+i z−i
√
this is a Mobius Transformation that sends i 7→ ∞, −i 7→ 0, 1 − 2 7→ 1. Using that Mobius Transforma-
tions maps circles to circles and lines, we deduce that ϕ maps D to
H := {z ∈ C : Re(z) > 0 and Im(z) < 0}
√
since ϕ(1+ 2) = −i. Now we apply ϕ2 := iz to map this into the region {z ∈ C : Re(z) > 0 and Im(z) >
0} and use the map z 2 to map it into {z ∈ C : Im(z) > 0} and finally the Cayley Transform
z−i
φ=
z+i
to map this into the unit disc. Composing all of our maps give the desired conformal map.

156 RAYMOND CHU

Problem 12. Show that ˆ ∞

tz
F (z) := √ dt
1 1 + t3
is well defined (by the integral) and analytic in {z ∈ C : Re(z) < 1/2}, and admits a meromorphic
continuation to the region {z ∈ C : Re(z) < 3/2}

Proof. We define tz := exp(z log(t)) where log(t) is the standard branch of the logarithim, so that tz is
holomorphic since t ∈ [1, ∞). Write z = x + iy and then
|tx+iy | = | exp(x log(t) + iy log(t)| = | exp(x log(t)| = tx
Therefore, ˆ ˆ ∞ x
∞
tx t
|F (x + iy)| ≤ √ dt . 3/2
dt
1 1 + t 3
1 t
so if Re(z) = x < 1/2 then x − 3/2 < −1 so that the integral defining F (z) is absolutely integrable i.e.
it is well defined. Also if R ⊂ {z ∈ C : Re(z) < 1/2} is a rectangle then as the integral defining F is
absolutely integrable we may apply Fubini to see
ˆ ˆ ˆ ∞ ˆ ∞ˆ
tz tz
F (z) = √ dt = √ dt = 0
∂R ∂R 1 1 + t3 1 ∂R 1 + t3
√
where for the last equality we used tz / 1 + t3 is holomorphic for t ∈ [1, ∞). Therefore, by Morrera’s
Theorem since F (z) is continuous we see that F (z) (by the DCT) is holomorphic in {z ∈ C : Re(z) < 1/2}.

Intuitively from our earlier bound the pole should be at z = 1/2. So we rewrite
ˆ ∞ ˆ ∞
tz t3/2
F (z) = √ dt = tz−3/2 √ dt
1+t 3 1 + t3
1 1
since we will want to integrate by parts to pick up a 1/(z − 1/2) factor. This leads to by integration by
parts ˆ ∞
1 3 tz
F (z) = − dt
z − 1/2 2(z − 1/2) 1 (1 + t3 )3/2
so this integral converges by an identical computation as above if Re(z) − 9/2 < −1 ⇒ Re(z) < 7/2.
Therefore, F (z) extends meromorphically onto {z ∈ C : Re(z) < 3/2} with a pole at 1/2.
157

21. Spring 2020

Problem 1. Assume f ∈ Cc∞ (R) satisfies

ˆ
2
e−tx f (x)dx = 0 for any t ≥ 0
R
show that f (x) = −f (−x) for any x ∈ R

Proof. Observe that we have for any h > 0

ˆ 2 2
(e−(t+h)x − e−tx )
f (x)dx = 0
R h
so by the Mean Value Theorem and Dominated Convergence Theorem we can tkae h → 0 inside the
integral to conclude
ˆ
2
x2 e−tx f (x)dx = 0 for any t ≥ 0
R
2
then we can keep iterating this process since p(x)e−x for any polynomial is in L1 to conclude that for
any finite sum that
X ˆ 2 X ˆ
an xn e−tx f (x)dx = 0 for any t ≥ 0 ⇒ an xn f (x)dx = 0
n even R n even R

Now let f be supported on [−M, M ] then we have

X ˆ M
an xn f (x)dx = 0
n even −M

By Stone Weiestrass, polynomials are dense in C([−M, M ]) under the sup-norm. Then there is a poly-
PN
nomial P (x) = n=1 bn xn such that ||f (x) − P (x)||L∞ ([−M,M ] < ε/(2M ) where ε > 0 is given. Now
decompose f (x) = f (x)+f
2
(−x)
+ f (x)−f2
(−x)
= feven (x) + fodd (x) i.e. the even and odd decomposition of
f and observe that since the integral is symmetirc we have
X ˆ M X ˆ M
n
an x f (x)dx = 0 ⇒ an xn feven (x)dx = 0
n even −M n even −M

Therefore, since feven is even and Stone-Weiestrass we conclude that f can be uniformly approxiamted in
[−M, M ] by even polynomials, but then choosing an to be these polynomials coefficients lets us conclude
that
ˆ M
|feven (x)|2 dx = 0
−M

so feven = 0 i.e. f is an odd function as desired.

Problem 2. Assume fn : R → R is a sequence of differentiable functions satisfying

ˆ ˆ
|fn (x)|dx ≤ 1 and |fn0 (x)|dx ≤ 1
R R
Assume also that for any ε > 0 there is an R(ε) > 0 such that
ˆ
sup |fn (x)|dx ≤ ε
n |x|≥R(ε)

Show that there exists a subsequence of {fn } that converges in L1 (R).

158 RAYMOND CHU

Proof. Proof One: Frechet Kolmogorov Theorem We will prove the family is first equicontinuous
in the Lp norm. Indeed, observe for any h > 0 that by the Fundamental Theorem of Calculus that
ˆ x+h
|fn (x + h) − fn (x)| ≤ |fn0 (y)|dy
x

so we have
ˆ ˆ ˆ ˆ ˆ ˆ
0
|fn (x+h)−fn (x)|dx ≤ χ[x,x+h] (y)|fn (y)|dydx = 0
χ[y−h,y] (x)|fn (y)|dxdy = h |fn0 (x)|dx ≤ h
R R R R R R

where the interchange of integration is justified by Tonelli since the integrand is non-negative.

Now we claim this implies combined with the second given condition implies that {fn } is relatively
compact in L1 (R). By completeness of L1 (R) it suffices to show that {fn } is totally bounded. We
will do this with an approximation argument using Arzela-Ascoli. Indeed, fix any ρ ∈ Cc∞ (B1 (0)) with
||ρ||L1 (R) = 1 and define ρn := nρ(x/n) also has mass 1. Then define for any n, m ∈ N
ˆ
gm,n (x) := (ρm ? fn )(x) = ρm (x − y)fn (y)dy
R

Then it is well known gm,n (x) ∈ C ∞ (R) and we have the bounds
||gm,n ||L∞ ≤ ||fn ||L1 ≤ ||ρm ||L∞ := C(m)
so the family {gm,n }n∈N is uniformly bounded in the sup norm. Also
ˆ
|gm,n (x) − gm,n (x + h)| ≤ |ρm (y)||fn (x + h − y) − fn (x − y)|dy
R

≤ ||ρm ||L∞ ||f (x + h) − f (x)||L1 (R) ≤ h||ρm ||L∞

so the family {gm,n }n∈N is totally bounded in C(|x| ≤ R(ε)). Therefore, if ε > 0 we can find finitely
many N such that
N
[
{gm,n }n∈N ⊂ Bε/(2R(ε)) (gm,n )
n=1
where the ball is with respect to the sup norm. Now observe that
ˆ ˆ ˆ ˆ ˆ
1
|fn (x)−gm,n (x)|dx ≤ |ρm (y)||fn (x)−fn (x−y)|dxdy = |ρm (y)| |fn (x)−fn (x−y)|dxdy ≤ <ε
R R R B1/m (0) R m
for m large enough. Notice that this bound is independent of n. Hence, now we claim that
N
[
{fn } ⊂ B5ε,L1 (R) (fj )
i=1

where these balls are taken over L1 (R) metric. Indeed, observe if k ∈ N then we can find an 1 ≤ n ≤ N
such that ||gm,n − gm,k ||L∞ < ε then
||fn − fk ||L1 (|x|≤R(ε)) ≤ ||fn − gm,n ||L1 (|x|≤R(ε) + ||gm,k − gm,n ||L1 (|x|≤R(ε) + ||fk − gn,k ||L1 (|x|≤R(ε)

≤ 3ε
And we know that
||fn − fk ||L1 (|x|>R(ε) ≤ ||fn ||L1 (|x|>R(ε) + ||fk ||L1 (|x|>R(ε) ≤ 2ε
Hence, we conclude
||fn − fk ||L1 (R) ≤ 5ε
so {fn } is totally bounded in L1 (R); therefore, it is precompact since L1 (R) is complete. So there exists
along a sub-sequence denoted nk and an f ∈ L1 (R) such that ||fnk − f ||L1 (R) → 0

Proof Two: Helly’s Selection Theorem

159

Proof. We use the following theorem:

Helly’s Selection Theorem Given {fn } a sequence of monotone functions such that the family is
uniformly bounded, there exists a sub-sequence that converges everywhere.

The proof is a standard application of Bolzano-Weiestrass, a diagonalization argument along Q, and

using the limiting function has to be monotone to define f (x) := limq→x f (q) where f (q) is the limit of
the subsequence fnk along rationals.

Notice that this automatically extends into BV functions since they are the difference of two monotone
functions. And observe that by the fundamental theorem of calculus that for all x ∈ R
ˆ
|fn (x) − fn (0)| ≤ |fn0 (x)|dx ≤ 1
R
and from ˆ
|fn (x)|dx = 1
R
we conclude that |fn (x)| ≤ 1 a.e., which by continuity implies |fn (x)| ≤ 1 everywhere. And as the total
variation of differentiable functions is just the L1 norm of the derivative, we see that {fn } is a family of
bounded variations that is uniformly bounded. So by Helly’s Selection Theorem, we deduce that along a
subsequence which we still denote by n that fn (x) → f (x) pointwise everywhere. By Fatou’s lemma we
deduce ˆ ˆ
|f (x)|dx ≤ lim inf |fn (x)|dx ≤ 1
R n→∞ R
and for ε > 0 ˆ ˆ
|f (x)|dx ≤ lim inf |fn (x)| ≤ ε
|x|≥R(ε) n→∞ |x|≥R(ε)

So now let K := BR(ε) (0) which is compact so now fix ε > 0 and by Egorov’s theorem we have a set
E ⊂ K such that m(K \ E) < ε such that fn → f uniformly on K. Then
ˆ ˆ ˆ
|fn (x) − f (x)|dx ≤ |fn (x) − f (x)|dx + |fn (x)| + |f (x)|dx
K E K\E

And as ||fn ||L∞ ≤ 1 we see that ||f ||L∞ ≤ 1 so we see for n sufficiently large thanks to uniform convergence
on E which is of finite measure that we have
ˆ
|fn (x) − f (x)|dx ≤ 3ε
K
so for large enough n ˆ
|fn (x) − f (x)|dx ≤ 5ε
R
Therefore, we have ||fn − f ||L1 → 0.

Problem 3. Prove that L∞ (Rn ) ∩ L3 (Rn ) is a Borel subset of L3 (Rn )

Proof. Observe that

∞
[ ∞
[
L∞ (Rn ) ∩ L3 (Rn ) = {f ∈ L3 (Rn ) : ||f ||L∞ (Rn ) ≤ n} := An
n=1 n=1

and we claim each An is closed in L3 (Rn ). Indeed, if fn ∈ A converges to f ∈ A in the L3 sense we

conclude that f ∈ L3 (Rn ). Also along a subsequence which we denote by nk we have fnk → f pointwise
a.e., so we conclude that ||f ||L∞ (Rn ) ≤ n since each ||fnk ||L∞ ≤ n. Therefore, f ∈ An so we have written
L∞ (Rn ) ∩ L3 (Rn ) as a countable union of closed sets in L3 (Rn ), so it is Borel subset of L3 (Rn ).

160 RAYMOND CHU

Problem 4. Fix f ∈ L1 (R). Show that

ˆ 2
lim f (x) sin(xn )dx = 0
n→∞ 0

Proof. We will first show the statement for step functions. By linearity of the integral it suffices to show
that it is true for characteristic functions of intervals to show its true for all step functions. First observe
that if 0 ≤ a < b < 1 then
ˆ b
sin(xn )dx → 0
a
since sin(xn ) → 0 as 0 < a, b < 1 and by applying DCT with f (x) = 1 to get the above convergence.

Now we show that if 1 < a < b < 2 then

ˆ b
lim sin(xn )dx = 0
n→∞ a
via the method of non-stationary phase. Indeed observe that
ˆ b ˆ b !
n ixn
sin(x )dx = Im e dx
a a

And we have
ˆ b ˆ b
ixn 1 d ixn
e dx = e dx
a a inxn−1 dx
so integration by parts gives
n n ˆ b
eib eia

1 ixn 1−n
= − − e
in bn−1 an−1 a inxn
so it follows that
ˆ b
ixn 1 1 1 (1 − n)
e dx ≤ + + (b − a) →0
a n bn−1 an−1 nan
since a, b > 1

This implies that if 0 ≤ a < b < 2 that

ˆ b
sin(xn )dx → 0
a
´b
Therefore, if f (x) is a step function we have a f (x) sin(xn )dx → 0. Now if f ∈ L1 ([0, 2]) we can find a
sequence of step functions fn such that fn → f in L1 ([0, 2]) so this gives
ˆ 2 ˆ 2 ˆ 2
f (x) sin(xn ) ≤ |fm (x) − f (x)|dx + fm (x) sin(xn )dx → 0
0 0 0
1
where the first term is small due to L convergence and the second is true by our earlier computation.

Problem 5. Rigorously determine the infimum of

ˆ 1
|P (x) − |x||2 dx
−1
over all choices of polynomials P (x) ∈ R[x] of degree not exceeding three.
161

Proof. Consider the subspace V of at most three degree polynomials in L2 ([−1, 1]). It is clear that V is
a subspace of L2 (R). Observe then that if f (x) ∈ L2 ([−1, 1])
ˆ 1
inf ||f (x) − P (x)||2 = inf |P (x) − |x||2 dx
P (x)∈V P (x)∈V −1
so this reduces to an orthogonal projection question. Indeed, if f⊥ is the orthogonal project of f onto V
then for any P (x) ∈ V
||P − f ||2 = ||P − f⊥ + f⊥ − f ||2 = ||V − f⊥ ||2 + ||f⊥ − f ||2
where in the third equality we used Pythagerous Theorem since V ∈ (P − f⊥ ) ⊥ (f⊥ − f ), so it follows
that fp
⊥ is the infimum.pTo find f⊥ we do p Grahm Schmit on {1, x, x2 , x3 }. Doing this gives us that
3 3
{1/2, 3/2x, (x − 1/3) 8/45, (x − 3/5x) 175/8} := {v1 , v2 , v3 , v4 } is an orthonormal basis of V .
Therefore, noting that
X4
f⊥ = (f, vi )vi
i=1
P4
gives us the minimizer if i=1 (|x|, vi )vi as desired.

Problem 6. Let us define a sequence of linear functionals on L∞ (R) as follows:

ˆ
1 ∞ n −x
Ln (f ) := x e f (x)dx
n! 0
(1) Prove that no subsequence of this sequences converges weak-*.
(2) Explain why this does not contradict Banach-Alaoglu Theorem.

Proof. For the first part observe that Ln (1) is just Γ(n + 1) so Ln (1) = n!. Now we need the following
lemma

Lemma: For a fixed ε > 0 there exists a sequence of intervals {[an , bn ]} for n large where an , bn → ∞
as n → ∞ such that
ˆ ˆ
1 bn n −x 1 ∞ n −x
x e dx − x e dx ≤ ε
n! an n! 0
i.e.most of the mass of Ln (1) is concentrated on the intervals [an , bn ].
√ n
Proof of Lemma Recall stirling’s approximation i.e. n! ∼ 2πn nen so we see that
ˆ ˆ n/3
1 n/3 n −x en nn en e 1
x e dx ∼ √ xn e−x dx ≤ √ = ( )n √ √ → 0
n! 0 n
n 2πn 0 n
2π3 n n+1/2 3 2π n
and as exponential growth beats polynomials decay we can find such a bn . Hence, we have proved for n
large that most of the mass lives in [n/3, bn ] for some bn that tends to ∞ as n → ∞.

Now fix a subsequence nk and by choosing a further subsequence if necessary we can assume that
[nk /3, 3nk ] are disjoint. Then define
X X
f := χ[nk /3,3nk ] (x) − χ[nk /3,3nk ] (x)
k even k odd
which is well defined since all the intervals are disjoint. Then for n large enough, thanks to our lemma
we know that Lnk (f ) < 1/2 if k is odd and Lnk (f ) > 1/2 if k is even. Therefore, this subsequence does
not converge weak* and as this subsequence was arbitrary we conclude that Ln does not converge weak*
on any subsequence.

For the second part, this does not contradict Banach Alagou since the weak* topology is not metrizable,
so compactness is not equivalent to subsequential compactness.
162 RAYMOND CHU

Problem 7. Let FM denote the set of functions holomorphic on D and continuous on D that
satisfy ˆ 2π
|f (eiθ )|dθ ≤ M < ∞
0
Show that every {fn } contains a subsequence that converges uniformly locally on D.

Proof. Fix a compact set K ⊂ D then there is a δ > 0 such that d(K, ∂D) = δ. Then we have by Cauchy’s
Integral Formula that for w ∈ K and any f ∈ FM
ˆ ˆ 2π
1 f (z) 1 f (eiθ ) iθ
f (w) = dz = e
2πi |z|=1 z − w 2π θ=0 eiθ − w
so
ˆ 2π
1 |f (eiθ )| M
|f (w)| ≤ dθ ≤
2π θ=0 δ 2πδ
hence for any subsequence {fn } ⊂ FM we have that it is uniformly bounded on every compact subset,
so it follows by Montel’s theorem that along a subsequence fn converges uniformly locally on D.

Problem 8. For each z ∈ C define

∞
X (z/2)2n
F (z) := (−1)n
n=0
(n!)2
(1) Show that the resulting function is entire and |F (z)| ≤ e|z|
(2) Show that there is an infinite sequence an ∈ C so that
∞
Y z2
F (z) = (1 − 2 )
n=1
an
and the product converges locally uniformly on C.

Proof. For the first part it suffices to show the sum defining F (z) converges locally uniformly since each
term in the sum is holomorphic. Observe that since
∞
X |z|n
e|z| =
n=0
n!

converges for all z ∈ C that if K ⊂ C is compact then there is an R > 0 such that K ⊂ BR (0) so
∞ ∞ ∞
X |z|2n X |R|n |R|n X |Rn |
|F (z)| ≤ 2n 2
≤ ( ) ≤ C(R) = C(R)eR < ∞
n=0
2 (n!) n=0
(n!) n! n=0
n!

Therefore, by the Weiestrass M-test the series defining F (z) converges locally uniformly on C so F (z) is
an entire function since each term in the series is entire.

For the bound, observe that

∞
X |z|2n
|F (z)| ≤
n=0
22n (n!)2
and
∞ ∞
|z|
X |z|n X |z|2n
e ≥ ≥
n=0
n! n=0
(2n)!
163

since we only omitted the odd terms in the series expansion. So it suffices to show that (2n)! ≤ 22n (n!)2 .
The case of n = 1 is obvious , so assume it is true for n and then we will show its true for n + 1. Indeed,
observe if
(2n)! ≤ 22n (n!)2 ⇒ (2n + 1)! ≤ (2n + 1)22n (n!)2
and it’s clear that (2n + 1) ≤ 4(n + 1)2 so we conclude that (2n + 1)! ≤ 22n+2 (n + 1)!2 so we obtain that
|F (z)| ≤ e|z|
as desired.

For the second part, recall from part a) that F (z) is an entire function of order one, so by Hadamard’s
factorization theorem we can write
∞
Y z z/an
F (z) = eaz+b z m (1 − )e
n=1
a n

for some a, b ∈ C, m is the multiplicity of the zero of F (z) at z = 0 and an are the zeros of F (z) and
the product converges locally uniformly on C. Notice that F (0) = 1 so we conclude m = 0. Also as
F (z) = F (−z) we conclude for every n ∈ N we can find an m ∈ N such that an = −am . Now observe
z z/an z z/am z2
(1 − )e (1 − )e = (1 − 2 )
an am an
since am = −an , so by rearranging terms in the product which is allowed since the product converges
locally uniformly we have that
∞
Y z2
F (z) = eaz+b (1 − 2 )
n=1
an
Now using F (z) = F (−z) we conclude that a = 0. Using F (0) = 1 gives us 1 = eb so
∞
Y z2
F (z) = (1 − )
n=1
a2n
so it suffices to show that there are infinitely many zeros of F (z).

Qnfor the sake of contradiction that there are only finitely many roots of F (z) say {z1 , .., zn }.
Now assume
Then F (z)/ j=1 (z − zj ) is a entire function with no zeros so there is an entire function g(z) such that
n
Y
F (z) = (z − zj )eg(z)
j=1

but as |F (z)| ≤ e|z| we see that |g(z)| ≤ Q

C|z|, which by Cauchy’s inequalities imply g(z) = az + b for some
n
a, b ∈ C. Then observe as F is even that j=1 (z−zj ) is an even function since if F (zj ) = 0 ⇒ F (−zj ) = 0.
Qn
Therefore, we see using F (z) = F (−z) that a = 0 so F (z) = j=1 (z − zj )eb which means F is a
polynomial. But clearly F is not a polynomial, which is our contradiction. Therefore, F (z) has infinitely
many zeros i.e. there are infinitely many an which lets us conclude.

Problem 9. Let f (z) ∈ L1 (C) ∩ C 1 (C). Show that the integral

ˆ ˆ
1 f (ξ)
u(z) = − dλ(ξ)
2π C ξ−z
defines a C 1 function on the entire plane such that

∂ ∂
+i u(x + iy) = f (x + iy)
∂x ∂y
In this problem dλ represents the Lebesgue measure on C and C 1 is meant in the real variable
sense.
164 RAYMOND CHU

Proof. Note that the integral is well defined since 1/z ∈ L1 (B1 (0), dλ(z)) and f ∈ L1 (C). In particular,
DCT implies that u(z) is continuous. In addition, by a change of variables we have that
ˆ
1 f (ξ + z)
u(z) = − dλ(ξ)
2π C ξ
so we have that for h real that
ˆ
u(z + h) − u(z) 1 f (ξ + z + h) − f (ξ + z)
=−
h 2π C hξ
By Taylor Expansion of f we know that since h is real
∂f
f (ξ + z + h) = f (ξ + z) + (ξ + z)h + o(h)
∂x
so ˆ
u(z + h) − u(z) 1 ∂f 1
lim =− (ξ + z) dλ(ξ)
h→0 h 2π C ∂x ξ
And similarily
ˆ
∂u i ∂y f (ξ + z)
(z) = − dλ(ξ)
∂y 2π C ξ
so we have ˆ
∂u ∂u 1 1 ∂f ∂f
+ i )(z) = −
( (ξ + z) + (ξ + z)
∂x ∂y 2π C ξ ∂x ∂y
ˆ ˆ
1 1 ∂f ∂f 1 1 ∂
= lim − (ξ + z) + (ξ + z) dλ(z) = − f (ξ + z)dλ(z)
ε→0,R→∞ 2π ε≤|z|≤R ξ ∂x ∂y 2π ε≤|z|≤R ξ ∂z
so by the Generalized Stokes Theorem we have that
ˆ ˆ
1 1 1 1
= f (ξ + z)dz − f (ξ + z)dz)
2πi |z|=R ξ 2πi |z|=ε ξ
´ 1
Note the first integral goes to zero as R → ∞ since f (ξ)/ξ ∈ L1 (C). Now using that |ξ|=ε ξ
dξ = 2πi we
have that ˆ ˆ
1 1 1 f (z) − f (ξ + z)
f (z) − f (ξ + z)dz =
2πi |ξ|=ε ξ 2πi |ξ|=ε ξ
ˆ 2π
1 f (z) − f (εeiθ + z) iθ
= εe dθ → 0
2π θ=0 εeiθ
where the convergence is due to uniform continuity of f on B1 (z). Therefore,
∂x + i∂y u = f
so it solves the PDE.

Problem 10. Evaluate the improper Reimann Integral

ˆ ∞ 2
x − 1 sin(x)
dx
0 x2 + 1 x
Justify all manipulations

Proof.

Problem 11. Let K ⊂ T be a compact proper subset.

(1) Show there is a sequence of polynomials Pn (z) → z uniformly on K.
(2) Show there is no sequence of polynomials Pn (z) that uniformly converges to z on T
165

Proof. For the first part this is just simply a consequence of Runge’s theorem since z = 1/z on T. Recall
that Runge’s Theorem implies that if K is compact such that C \ K is connected, then any holomorphic
function on a neighborhood of K can be uniformly approximated by polynomials in K. Indeed, 1/z is
holomorphic on C \ {0} and since K is a proper subset of T we have C \ K is connected.

Indeed, recall that Runge’s Theorem shows by using the Cauchy Integral Formula and a Riemann Sum
Approximation that if f is holomorphic on a neighborhood of a compact set K, then there is a sequence of
rational functions {Rn } with poles outside of K such that Rn uniformly converges to f . It suffices to show
that the rational function 1/(z − z0 ) where z0 ∈/ K can be uniformly approximated in K by polynomi-
als when C \ K is connected since every Rn can be written as a polynomial combination of such functions.

Indeed, fix z0 ∈
/ K, then choose a z1 ∈ / K far away from z0 . Then as C \ K is open and connected, we
can find a curve γ : [0, 1] → K c such that γ(0) = z1 and γ(1) = z0 . Then we have
∞
1 1 1 1 X z n
=− =− ( )
z − z1 z1 1 − z/z1 z1 n=0 z1

where the sum converges when we choose z1 such that |z/z1 | < 1 for all z ∈ K. Choosing such a z1 we
see that 1/(z − z1 ) can be uniformly approximated in K by its partial sums due to Weiestrass M-Test,
so 1/(z − z1 ) can be uniformly approximated by polynomials in K. Now if we let ρ := 1/2d(K, γ) then
we can choose points {wi }N i=0 such that wi ∈ γ with |wi − wi+1 | < ρ and w0 = z1 and wN = z0 . Then we
claim we can uniformly approximate 1/(z − wi+1 ) by polynomials in 1/(z − wi ). Indeed, observe
∞ n
1 1 1 X wi+1 − wi
= −wi =
z − wi+1 (z − wi )(1 − wi+1
z−wi )
z − wi n=0 z − wi
−wi
which converges since | wi+1
z−wi | < 1/2. Then this means we can uniformly approximate 1/(z − w0 ) in
K by polynomials since we can do it for 1/(z − w1 ) combined with our earlier observation that we can
uniformly approximate 1/(z − w0 ) by polynomials of 1/(z − w1 ). Then by iterating this process N we
deduce that 1/(z − z0 ) can be uniformly approximated by polynomials in K and hence so can every
rational function with poles outside of K. This proves Runges Theorem.

Assume for the sake of contradiction that 1/z can be uniformly approximated by polynomials Pn on
T. Then we have from uniform convergence
ˆ ˆ ˆ
1
2πi = dz = lim Pn (z)dz = lim Pn (z) = 0
|z|=1 z |z|=1 n→∞ n→∞ |z|=1

which is our contradiction.

Alternative Proof: Assume that Pn (z) a sequence of polynomials uniformly converge to z = 1/z on
T. Then the uniform limit of these polynomials denoted by f (z) extends to be a holomorphic function
on D such that f (z) = 1/z on T. Then we have that for large enough n that

sup |Pn (z) − 1/z| < 1 ⇒ sup |zPn (z) − 1| < 1

|z|=1 |z|=1

so by the maximum modulus principle |zPn (z) − 1| < 1 on D but taking z = 0 gives a contradiction.

Problem 12. Let u be a continuous subharmonic function on C that satisfies

u(z)
lim sup ≤0
|z|→∞ log |z|

Show that u is constant.

166 RAYMOND CHU

Proof. This given condition implies if ε > 0 that u(z) − ε log |z| → −∞ as |z| → ∞. Now consider
the annular domain ΩR,1 := {1 ≤ |z| ≤ R} for R > 1. As log |z| is harmonic on ΩR,1 we have that
u(z) − ε log |z| is subharmonic on ΩR,1 , so by the maximum principle
sup u(z) − ε log(z) = max u(z) − ε log(z)
z∈ΩR,1 z∈∂ΩR,1

But from the given conditions of u(z) − ε log |z| → −∞ we see by taking R → ∞ that
sup u(z) − ε log(z) = max u(z)
z∈|z|>1 |z|=1

where we used log(1)=0 and continuity of u to deduce theres a max over {|z| = 1}. Now by letting ε → 0
we see that
sup u(z) = max u(z)
z∈|z|>1 |z|=1

but then this implies that u(z) has an interior maximum on D ⊂ C. Therefore, u(z) is constant (subhar-
monic functions with an interior maximum are constant thanks to the submean value inequality).

167

22. Fall 2020

Problem 1. Suppose f : [0, 1] × [0, ∞) → [0, 1] is continuous. Prove that

F (x) := lim sup f (x, y)
y→∞

is Borel Measurable.

Also show that for any borel set E ⊂ [0, 1] there is a choice of continuous function f : [0, 1] × R →
[0, 1] so that F agrees with the indicator function almost everywhere.

Proof. Note that intervals of the form [a, b] for 0 < a < b < 1 generate the borel σ-algebra on [0, 1], so it
suffices to show F −1 (a, b) is borel. Observe
∞
[ \
F −1 ([a, b]) = {x : b ≥ f (x, q) ≥ a}
M =0 q∈Q:q≥M

i.e. there is an M ∈ N such that for any rational q ≥ M we have b ≥ f (x, q) ≥ a. Note that we are
using continuity to conclude that if for all Q 3 q ≥ M we have b ≥ f (x, q) ≥ a then for any y ≥ M that
b ≥ f (x, y) ≥ a i.e. b ≥ F (x) ≥ a.

Define fq := f (x, q) then as f is continuous so is fq so we know that {x : b > f (x, q) > a} = fq−1 ((a, b))
so this set is open. Therefore, F −1 ([a, b]) is Borel Measurable. It is also easy to verify that {E Borel :
F −1 (E) is Borel } is a σ-algebra and we showed it contains the closed set so it contains the borel σ-
algebra, so F is Borel Measurable.

For the second part, let A denote the set of Borel Subsets of [0, 1] such that there is a continuous
function f so that F agrees a.e. with the indicator function of that set. We claim A is a σ-algebra.
Indeed, observe that [0, 1] ∈ A and ∅ ∈ A by defining f (x, y) = 1 and f (x, y) = 0 respectively.

Now if E ∈ A then there is a continuous f such that the corresponding F agrees a.e. with the indicator
function of E. Then notice 1 − F then agrees almost everywhere with the indicator function of E c ∩ [0, 1],
which corresponds to the function 1 − f . So this set is closed under compliments.

Now we claim that A is closed under finite intersections. Indeed, if E1 , E2 ∈ A then there is a corre-
sponding f1 and f2 continuous such that Fi agrees a.e. with the indicator function of Ei . Then observe
χA∩B = χA χB so we can take g := f1 f2 then lim supy→∞ g(x, y) will agree with χE1 ∩E2 a.e.

Now let {Ei }∞i=1 ⊂ A. As A is closed under finite intersections and compliments, we may assume
WLOG P∞ that E i are disjoint. Then if {fi } are the corresponding continuous functions we can define
f := i=1 fi (x, y) which is well defined when y is large since Ei are disjoint. Therefore, this set is closed
under countable union.

So now it suffices to show that it A contains intervals of the form (a, b). So define for y ≥ 1



 0 for x ≤ a − 1/y
y(x − (a − 1/y)) for a − 1/y ≤ x ≤ a



f (x, y) := 1 for a ≤ x ≤ b

−y(x − (b + 1/y)) + 1 for b ≤ x ≤ b + 1/y





0 else

i.e. we are adjoining a small line to the end of χ[a,b] that dissapears as y → ∞. This f has all the desired
properties for y ≥ 1 and for y < 1 just make the function equal to f (x, 1). Therefore, A contains the
borel σ-algebra, which implies the problem.

168 RAYMOND CHU

Problem 2. Show that there is a constant c ∈ R so that

ˆ 1 ˆ 1
lim f (x) cos(sin(2πnx))dx = c f (x)dx
n→∞ 0 0
for every f ∈ L1 ([0, 1]) where n is taken over N.

Proof. Observe that

ˆ 1 n−1 ˆ ˆ 1
1 X j+1
cos(sin(πnx))dx = cos(sin(πx))dx = cos(sin(πx))dx =: c
0 n j=0 j 0

Now fix f ∈ C([0, 1]) then observe that

ˆ 1 n−1 ˆ
1 X j+1 x
f (x) cos(sin(πnx))dx = f ( ) cos(sin(πx))dx
0 n j=0 j n

So by uniform continuity of f if ε > 0 we can find a δ > 0 such that if x, y ∈ [0, 1] and |x − y| < δ then
|f (x) − f (y)| < ε. Choose n so large such that 1/n < δ then

n−1 ˆ n−1 ˆ n−1 ˆ

1 X j+1 x 1 X j+1 j 1 X j+1
f ( ) cos(sin(πx))dx − f ( ) cos(sin(πx)) ≤ |f (x/n) − f (j/n)| ≤ ε
n j=0 j n n j=0 j n n j=0 j

And observe that

n−1 ˆ n−1 ˆ 1
1 X j+1 j 1X j
f ( ) cos(sin(πx)) = c f( ) → c f (x)dx
n j=0 j n n j=0 n 0

where the last convergence is due to f is continuous so the Riemann Sums converge to the integral of f .
Therefore, we have
n−1 ˆ ˆ 1 ˆ 1
1 X j+1 x
lim f ( ) cos(sin(πx))dx = lim f (x) cos(sin(πnx))dx = c f (x)dx
n→∞ n n n→∞ 0
j=0 j 0

so this shows the result for f ∈ C([0, 1]), so now by density if f ∈ L1 ([0, 1]) then there is a sequence
fn ∈ C([0, 1]) such that ||fn − f ||L1 ([0,1]) → 0 so
ˆ 1 ˆ 1 ˆ 1 ˆ 1 ˆ 1
f (x) cos(sin(2πnx))dx − c f (x)dx ≤ |f (x)−fm (x)|dx+ fm (x) cos(sin(2πnx))dx − c fm (x)dx
0 0 0 0 0

ˆ 1
+c |f (x) − fm (x)|dx
0

so all three terms converge to 0 in the limit. Therefore,

ˆ 1 ˆ 1
lim f (x) cos(sin(2πnx))dx = c f (x)dx
n→∞ 0 0
´1
for all f ∈ L1 ([0, 1]) for c := 0
cos(sin(πx))dx.

169

Problem 3. Let dµn be a sequence of probability measures on [0, 1] so that

ˆ 1
f (x)dµn (x)
0
converges for every f ∈ C([0, 1]).
(1) Show that ˆ ˆ
g(x, y)dµn (x)dµn (y)
[0,1]2
2
converges for every g ∈ C([0, 1] ).
(2) Show by example that under hypothesis, it is possible that
ˆ ˆ
dµn (x)dµn (y)
0≤x≤y≤1
does not converge.

Proof. As [0, 1] is compact and µn is a sequence of Borel Probability Measures, we know that they are
Radon. Therefore, by Risez Representation Theorem and Banach Alagou there is a subsequence such
that µnk * µ i.e. for any f ∈ C([0, 1]) we have
ˆ 1 ˆ 1
f (x)dµnk (x) → f (x)dµ
0 0
´1 ´1
And by the given assumption we know that 0 f (x)dµn converges so it must converge to 0 f (x)dµ. Now
if we fix two sub-sequence, then each by Banach Alaogu has a further weakly convergent sub-sequence
and if we test any f ∈ C([0, 1]) along the limit measures we obtain the same value. Therefore, by Risez
Representation Theorem since measures are uniquely determined by their action on C([0, 1]) we conclude
that the two weak limits are the same, so as every subsequence has a further subsbequence that converges
to the same limit we know that µn * µ.

Now we claim that µn ⊗µn * µ⊗µ which will prove the claim. Indeed let A ⊂ C([0, 1]2 ) be the algebra
generated by functions of the form f (x)g(y) where f, g ∈ C([0, 1]). Then constants are in this space, so
A vanishes nowhere and if (s1 , t1 ) 6= (s2 , t2 ) then WLOG x 6= s, then byd efining A 3 g(x, y) = x then
g(s1 , t1 ) 6= g(s2 , t2 ), so this family vanishes nowhere. Therefore, as [0, 1]2 is compact, we know by Stone
Weiestrass that A is dense in C([0, 1]2 ).

So there is a sequence {fn (x)hn (y)} that converges to g(x, y) in the sup norm. Then
ˆ ˆ ˆ 1 ˆ 1 ˆ 1 ˆ 1
fn (x)hn (y)dµm (x)dµm (y) = fn (x)dµm (x) hn (y)dµm (y) → fn (x)dµ(x) hn (y)dµ(y)
[0,1]2 0 0 0 0

ˆ
= fn (x)hn (y)dµ(x)dµ(y)
[0,1]2

where the interchange in derivatives is justified since fn hn ∈ L1 ([0, 1]2 , µm ⊗ µm ) for any m because fn hn
is bounded and µm ⊗ µm is a probability measure.

So uniform convergence implies

ˆ ˆ ˆ ˆ
lim g(x, y)dµm (x)dµm (y) = g(x, y)dµ(x)dµ(y)
m→∞ [0,1]2 [0,1]2

so the limit exists.

170 RAYMOND CHU

Problem 4. Let X be a separable Banach space over R and let F : X → R be norm-continuous

and convex. Suppose xn * x show that
F (x) ≤ sup F (xn )
n

Proof. We claim that F (x) is weakly lower semi continuous which implies the claim. Indeed, it suffices
to show if α ∈ R then A := {x : F (x) ≤ α} is weakly closed. Indeed, notice that A is convex since F is
convex and is norm-closed. So now let y ∈ Ac , then {y} is compact and convex since it is a singleton, so
by Hanh Banach there exists a linear functional ` and β ∈ R such that
A ⊂ {x : `(x) < β} and y ∈ {x : `(x) > β}
Notice as linear functionals generate the weak topology that {x : `(x) > β} is open in the weakly
open. Therefore, A is weakly closed, so F is weakly lower semi-continuous. So fix 0 < ε 1 then as
{y : F (y) > F (x) − ε} is weakly open and x is i this set, we see that
lim inf F (xn ) > F (x) − ε
n→∞

since xn * x so letting ε → 0 gives

F (x) ≤ lim inf F (xn ) ≤ sup F (xn )
n→∞ n

as desired.

Problem 5. Suppose f ∈ L1 ([0, 1]) has the property that

ˆ
|f (x)|dx ≤ m(E)1/2
E
for any Borel Set E ⊂ [0, 1]. Here m represents the Lebesgue measure on [0, 1].
(1) Show that if 1 ≤ p < 2 then f ∈ Lp ([0, 1]).

(2) Show that there is a function f satisfying the above bounds and is in L1 ([0, 1]) but is not
in L2 ([0, 1]).

Proof. Note that the original question has a typo and it was supposed to be 1 ≤ p < 2.

Observe by the Layer Cake Decomposition that

ˆ 1 ˆ ∞ ˆ ∞ ˆ !
p p−1 p−2
|f (x)| dx = p t m({x : |f (x)| ≥ t})dt ≤ p t |f (x)|dx dt
0 t=0 t=0 {|f |≥t}

where the second inequality is due to Chebyshev’s Inequality. Therefore, combining this with m({x :
|f (x)| ≥ t}) ≤ 1 gives
ˆ 1 ˆ ∞ ˆ !
p p−2
|f (x)| ≤ p t min(1, |f (x)|dx )
0 t=0 {|f |≥t}
ˆ 1 ˆ ∞ ˆ !
p−2 p−2
≤p t dt + t |f (x)|dx
t=0 t=1 {|f |≥t}

and as 1 < p < 2 notice the first term is bounded. So we focus on bounding the second term. Using the
given inequality gives the second term is bounded by
ˆ ∞ ˆ ˆ !1/2
∞
≤p tp−2 (m({x : |f (x)| ≥ t}))1/2 dt ≤ p tp−2−1/2 |f (x)|dx dt
t=1 t=1 {|f |≥t}
171

and by reiterating the above argument n times we deduce that

ˆ ˆ ∞
! ˆ ∞ Pn
ˆ !1/2n
p−2 p−2− 1/2j
t |f (x)|dx ≤p t j=1 |f (x)|dx dt
t=1 {|f |≥t} t=1 |f |≥t

´ 1/2n
And for any fixed n we can bound |f |≥t
|f (x)|dx since f ∈ L1 ([0, 1]) and this integral converges
as long as
n
X
p−2− 1/2j < −1
j=1

i.e.
n
X
p<1+ 1/2k
j=1

and as n → ∞ the sum converges to 1, so we conclude that f ∈ Lp ([0, 1]) for 1 ≤ p < 2 from our
inequalities.

For the counter example take f (x) := 1/2x−1/2 which is non-negative. Then notice for any 0 < a <
b < 1 that
ˆ b √ √ √
f (x)dx = b − a ≤ b − a
a
where the inequality can be seen by squaring both sides. Then observe that
ˆ
{E Borel : f (x)dx ≤ m(E)1/2 }
E

/ L2 since
forms a σ-algebra, which contains the open intervals, so this is true for any Borel set and f ∈
1
1/|x| ∈
/ L ([0, 1]).

Problem 6. Suppose f : [−1, 1] → R is C 1 and odd. Show that

ˆ 1 ˆ 1
|f (x)|dx ≤ |f 0 (x)|dx
−1 −1

Proof. First observe that as f is odd that f 0 is even, so it suffices to show the inequality on [0, 1]. Now
as f (x) is odd we know that f (0) = 0 so by the fundamental theorem of calculus if 0 < x < 1 we have
ˆ x ˆ 1 ˆ 1ˆ x
f (x) = f 0 (y)dy ⇒ |f (x)|dx ≤ |f 0 (y)|dydx
0 0 0 0
ˆ 1 ˆ 1 ˆ 1 ˆ 1 ˆ 1 ˆ 1
= χ[0,x] (y)|f 0 (y)|dydx = |f 0 (y)| χ[0,x] (y)dx = y|f 0 (y)|dy ≤ |f 0 (y)|dy
0 0 0 0 0 0
where the interchange in integrals is justified by Fubini since the integrand is non-negative. In particular,
we have shown
ˆ 1 ˆ 1
|f (x)|dx ≤ |f 0 (y)|dy
0 0
which implies from the odd condition on f that
ˆ 1 ˆ 1
|f (x)|dx ≤ |f 0 (y)|dy
−1 −1

172 RAYMOND CHU

Problem 7. Let ∆j = {z : |z − aj | ≤ rj } where S 1 ≤ j ≤ n be a collection of closed disks with

radii rj ≥ 0, all contained in D. Let Ω := D \ ( j ∆j ) and let u : Ω → R be harmonic. Prove that
there exists real numbers c1 , .., cn such that
n
X
u(z) − ci log |z − ai |
i=1
is the real part of a holomorphic function on Ω. Show also that c1 , ..., cn is unique.

Proof. By considering the Cauchy Riemann equations we define g(z) := ∂x u(z) − i∂y u(z) which is holom-
rophic in Ω since it is C 1 in the real sense and satisfies the Cauchy Riemann Equation since u is harmonic.
Now as u(z) is holomorphic in a small ball that contains ∆j that is disjoint from the other ∆i , we con-
clude that u(z) admits a Laurent Series Expansion near each ∆j . So let cj be the residue of g at aj then
consider
n
X cj
h(z) := g(z) −
i=1
z − aj
then by the Residue Theorem we know that h(z) integrates to zero along any closed curve. Therefore,
h(z) admits a primitive which we call v(z). We claim that up to a constant that
Xn
w(z) := u(z) − cj log |z − aj |
j=1

is the real part of v(z). Indeed, observe that by computation one has
∂x w − i∂y w = g(z) = ∂x Re(v(z)) − i∂y Re(v(z))
so one has w = Re(v(z)) + C for some constant. Therefore, we have that w(z) is the real part of a
holomorphic functions. So now it remains to verify uniqueness of ci .

So if
n
X
w(z) := u(z) − cj log |z − aj |
j=1
is the real part of a holomorphic function on Ω of say f then by the Cauchy Riemann equations (f 0 =
wx − iwy )
n
X cj
f 0 (z) = g(z) −
i=1
z − aj
and as this function must integrate to zero along any closed curve, we see that cj must be the residue of
g(z) since we can take a small circular curve around each ∆j and apply the Residue Theorem.

Problem 8. Let f : D → D be holomorphic and satisfy f (1/2) = f (−1/2) = 0. Show that

|f (0)| ≤ 1/4

Proof. Consider the automorphism of the disk

z − (1/2)
ψ(z) :=
1 − (1/2)z
then g := f /ψ is a holomorphic function since there is a removable singularity at z = 1/2 and g(−1/2) = 0.
It is still a map to the disk because of the maximum principle since |ψ1 | = 1 on ∂D. Then as g ◦ ψ1 (0) = 0
we have from the Schwarz lemma that
|g ◦ ψ| ≤ |z| ⇒ |g(0)| ≤ |ψ −1 (0)| = 1/2
But observe that g(0) = −2f (0) so we conclude that
|f (0)| ≤ 1/4
173

as desired.

Problem 9. Consider the following region

Ω := {x + iy : 0 < x < ∞ and 0 < y < 1/x}
Exhibit an explicit conformal map from Ω to D.

Proof. Observe ∂Ω = {(x, 0) : x ≥ 0} ∪ {(x, 1/x) : x > 0} ∪ (0, y) : y ≥ 0 then

z 2 (∂Ω) = {(x, 0) : x ∈ R} ∪ {(x, 2i) : x ∈ R}
so we know that z 2 (Ω) is either contained in {(x, y) : x ∈ R : 0 < y < 2} or its complement since z 2 is
continuous and Ω is connected. But by plugging in a point we see that z 2 (Ω) = {(x, y) : x ∈ R : 0 < y <
2} := Ω1 since z 2 is conformal on Ω. Now consider π2 z(Ω1 ) = {(x, y) : x ∈ R, 0 < y < π} := Ω2 . Then
exp(Ω2 ) = {(x, y) : x ∈ R, y > 0} so then we take the Cayley Transform
z−i
φ(z) :=
z+i
then this is a conformal map from the upper half plane to the disk. Then ϕ(z) := φ ◦ exp ◦(π/2z) ◦ z 2 is
our desired map.

Problem 10. Let K ⊂ C be a compact set of positive area but empty interior and define a
function F : C → C via ˆ
1
F (z) := dµ(w)
K w−z
where dµ denotes the planar measure on C.
(1) Prove that F (z) is bounded and continuous on C and analytic on C \ K.
(2) Prove that F (K) = F (C).

Proof. As K is compact there is an R > 0 such that K ⊂ BR (0) then

ˆ ˆ
1 1
|F (z)| ≤ dµ(w) ≤ dµ(w) := C < ∞
BR (0) |w − z| BR (0) |w|
where we are using 1/|x| ∈ L1 (BR (0), dµ(w)) since C ∼= R2 . This implies F (z) is bounded on C. Note
1
that F (z) is continuous since 1/w ∈ L (BR (0), dµ(w)) due to the continuity of the Lebesgue integral with
respect to translation.

Now let R ⊂ C \ K be a rectangle, then as F is bounded we have that it is locally integrable, so

ˆ ˆ ˆ ˆ ˆ
1 1
F (z)dz = dµ(w)dz = dzdµ(w) = 0
R R K w − z K R w − z
where the interchange in derivative is justified by Fubini and the last expression is 0 since w ∈ K and
R ⊂ C \ K so 1/(w − z) is holomorphic on R. Therefore, as F (z) is continuous and integrates to zero
over any rectangle C \ K, Morrera’s theorem tells us that F (z) is analytic on C \ K

Part 2: Missing

Problem 11. Let fn be a sequence of analytic functions on a connected domain Ω such that
|fn | ≤ 1 for all n ∈ N and z ∈ Ω. Suppose the sequence {fn (z)} converges for infinitely many z in
a compact set K in Ω. Prove that {fn (z)} converges for all z ∈ Ω.
174 RAYMOND CHU

Proof. Let K ⊂ Ω be compact then there is a δ > 0 such that d(K, ∂Ω) = 2δ then for any z ∈ K we have
Bδ (z) ⊂ Ω. Therefore, by Cauchy’s estimate we have for z ∈ Ω
|fn0 (z)| ≤ 1/δ
so the family {fn (z)} is uniformly bounded and equicontinuous so by Arzela-Ascoli there is a sub-sequence
such that fn converges uniformly on K. Then by a standard diagonlization argument, we can find a sub-
sequence that converges locally uniformly on Ω and by Cauchy’s Integral Theorem we see local uniform
convergence implies the limit is holomorphic.

Let {zj } be infinitely many points in K such that {fn (zj )} converges. As K is compact there is a
limit point z ∈ K and by looking at a subsequence if necessary assume that zn → z. Then given two
susbequence of fn , by our previous argument we can find further subsequences where both converge
locally uniformly to a holomorphic function f (z) and g(z) respectively. But we know that f (zn ) = g(zn )
for all n ∈ N and f (z) = g(z) since {fn (zj )} converges. Therefore, have f = g since holomorphic functions
are determined by their values on any infinite set with an accumulation point. So every sub-sequence has
a further sub-sequence that converges where the limit is the same, so the whole sequence converges.

Problem 12. Let Ω := {z ∈ C : −2 < z < 2}. Show that there is a finite C > 0 such that
ˆ ∞
|f (0)|2 ≤ C [|f (x − i)|2 + |f (x + i)|2 ]dx
−∞
for every holomorphic function f : Ω → D for which the right hand side is finite.

Proof. We will prove this by using Cauchy’s Integral Formula over a large rectangle. Indeed, let SR :=
{(x, y) : −R ≤ x ≤ R, −1 ≤ y ≤ 1} with counter clockwise orientation
ˆ ˆ 1 ˆ R ˆ 1 ˆ R
f (w) f (R + it) f (t + i) f (−R + it) f (t − i)
f (0) = dw = i dt − dt − i dt + dt
∂SR w t=−1 R + it −R t + i t=−1 −R + it −R t − i
so by the triangle inequality combined with |f | ≤ 1 gives
ˆ 2 ˆ R ˆ R
1 |f (t + i)| |f (t − i)|
|f (0)| ≤ dt + dt + dt
t=−1 |R + it| −R |t + i| −R |t − i|
and taking R → ∞ ˆ ∞
|f (t + i)| |f (t − i)|
|f (0)| ≤ + dt
−∞ |t + i| |t − i|
so using Cauchy-Schwarz gives
|f (0)| ≤ ||f (t + i)||L2 (R,dt) ||1/(t + i)||L2 (R,dt) + ||f (t − i)||L2 (R,dt) ||1/(t − i)||L2 (R,dt)
≤ C(||f (t + i)||L2 (R,dt) + ||f (t − i)||L2 (R,dt) )
where we used 1/(t + i) ∈ L (R, dt) since 1/|t + i|2 decays fast enough at ∞ and thanks to the i factor
2

it is integrable near the origin and similarily for 1/|t − i|. Now using (|a| + |b|)2 ≤ 4(|a|2 + |b|2 ) gives
ˆ
|f (0)|2 ≤ 4C |f (x + i)|2 + |f (x − i)|2 dx
R
as desired.