CONTENTS Kari Eifler

Solutions to Texas A&M’s Real Analysis Qual Courses

Kari Eifler

August 29, 2019

These are the solutions to the majority of the available past real qualifying exams for Texas A&M.
Incomplete/non-existent solutions are marked in red. If you find any errors/typos or have solutions
to the unsolved questions, please email me at [email protected].

Contents

1 August 2019 3

2 January 2019 5

3 August 2018 10

4 January 2018 17

5 August 2017 25

6 January 2017 32

7 August 2016 40

8 January 2016 46

9 August 2015 53

10 January 2015 59

1
CONTENTS Kari Eifler

11 August 2014 65

12 January 2014 71

13 August 2013 75

14 January 2013 81

15 August 2012 87

16 January 2012 92

17 August 2011 97

2
1 AUGUST 2019 Kari Eifler

1 August 2019

Problem 1. Let (X, M, µ) be a measure space and f a measurable non-negative function on X.

Define ν : M → [0, ∞] by

Z
ν(E) = f dµ.
E

(a) Prove that ν is a measure

R
Proof. Indeed, it’s clear that ν(E) = E f dµ ≥ 0 for all E since f is assumed to be non-
R
negative. It’s equally clear that ν(∅) = ∅ f dµ = 0.
We are only left to prove countable additivity. Take a countable collection {Ei } of pairwise dis-
joint sets in M, so we see for finitely many

Z Z N Z
X N Z
X N
X
ν ∪N


k=1 Ek = f dµ = χ∪N f dµ = χEk f dµ = f dµ = ν(Ek ).
k=1 Ek
∪N
k=1 Ek k=1 k=1 Ek k=1

By the monotone convergence theorem (the finite sums of characteristic functions form an in-
creasing sequence that converges to the infinite sum pointwise), then ν is countably additive.
Hence, ν is a measure.

(b) Prove that g ∈ L1 (ν) if and only if gf ∈ L1 (µ) and in that case
R R
X
gdν = X
gf dµ.

Proof. First we show that ν  µ. Indeed, if µ(E) = 0 then choose an increasing sequence of
simple functions fn such that fn → f . Then by monotone convergence theorem and the defini-
tion of integral for simple functions, we have
Z Z Z
ν(E) = f dµ = (lim fn )dµ = lim fn dµ = 0.
E E E

dν
Then we may apply Radon-Nikodym theorem to see that f = dµ and see that g ∈ L1 (ν) if
dν
R R R
and only if X |g|dν < ∞ which is equivalent to X |g|f dµ = X |g| dµ dµ < ∞. Since f is
R
non-negative, this is equivalent to having X |gf |dµ < ∞. Radon-Nikodym also tells us that
R R
X
gdν = X gf dµ.

Problem 2. (a) State Fatou’s lemma

Proof. For fn ∈ L+ then

Z Z
lim inf fn ≤ lim inf fn

(b) State the dominated convergence theorem

3
1 AUGUST 2019 Kari Eifler

Proof. Let g, gn ∈ L+ be measurable, |fn | ≤ gn µ-a.e., fn → f and gn → f µ-a.e. with

R
gn →
R R R R
g < ∞. Then fn → f . Moreover, |f − fn | → 0.

(c) Let fn , gn , hn , f, g, h be measurable functions on Rn satisfying fn ≤ gn ≤ hn , fn → f a.e.,

gn → g a.e., and hn → h a.e. Suppose moreover that f, h ∈ L1 and fn → f , hn → h.
R R R R

Prove that g ∈ L1 and gn → g.

R R

Proof. here

Problem 3. Let {Ak }∞ k=1 be measurable subsets of a measure space and define Bm to be the set of
all points which are contained in at least m of the sets {Ak }∞
k=1 . Prove that Bm is measurable and

∞
1 X
µ(Bm ) ≤ µ(Ak ).
m
k=1

Proof. Let C = {F ⊆ N | |F | = m} which is a countable infinite set. Then we may express

[ \
Bm = Ai .
F ∈C i∈F

Therefore, each Bm is measurable.

here

Problem 4. Let E be a subset of R which is not Lebesgue measurable. Prove that there exists an
η > 0 such that for any two Lebesgue measurable sets A, B satisfying A ⊆ E ⊆ B one has λ(B\A) >
η, where λ denotes Lebesgue measure.

Proof. here

Problem 5. Let {Ak }∞ n

k=1 be Lebesgue measurable sets in R equipped with Lebesgue measure λ.

(a) Prove that if Ak ⊆ Ak+1 for all k then λ(∪∞

k=1 Ak ) = limk→∞ λ(Ak )

P∞
Proof. We will assume that λ is subadditive, so λ(∪∞
1 Ak ) ≤ 1 λ(Ak ).Then by setting Ak = ∅,
we have

∞
X n
X
λ (∪∞
1 Ak ) = λ(Aj \Aj−1 ) = lim λ(Aj \Aj−1 ) = lim λ(An ).
n→∞ n→∞
1 1

(b) Prove that if Ak+1 ⊆ Ak for all k and λ(A1 ) < ∞ then λ(∩∞
k=1 Ak ) = limk→∞ λ(Ak )

Proof. Let Bj = A1 \Aj so B1 ⊂ B2 ⊂ . . ., and λ(A1 ) = λ(Bj ) + λ(Aj ), and ∪∞

1 Bj =
E1 \(∩∞
1 A j ). Then by part (a), we have

4
2 JANUARY 2019 Kari Eifler

∞
!
\
λ(A1 ) = λ Aj + lim λ(Bj ) = λ (∩∞
1 Aj ) + lim (λ(A1 ) − λ(Aj )).
j→∞ j→∞
1

Since λ(A1 ) < ∞, we may subtract it from both sides to yield the desired result.

(c) Give an example showing that without assuming λ(A1 ) < ∞ the conclusion of the previous part
does not hold.

Proof. Consider Aj = [j, ∞) so that for each j, λ(Aj ) = ∞ but ∩∞ ∞

1 Aj = ∅ so λ(∩1 Aj ) = 0.

Problem 6. Let X and Y be Banach spaces. Show that the linear space X ⊕ Y is a Banach space
under the norm k(x, y)k = kxk + kyk. Also determine (with justification) the dual (X ⊕ Y )∗ .

Proof. here
Pn
Problem 7. For each n ∈ N define on `∞ the linear functional ϕn (x) = n−1 k=1 x(k). Let ϕ be
the weak* cluster point of the sequence {ϕn }. Show that ϕ does not belong to the image of `1 under
the canonical embedding `1 ,→ (`∞ )∗ .

Proof. here

Problem 8. Let T : X → Y be a surjective linear map between Banach spaces and suppose that
there is a λ > 0 such that kT xk ≥ λkxk for all x ∈ X. Show that T is bounded.

Proof. here

Problem 9. Let X be a compact metric space and µ a regular Borel measure on X. Let f : X →
[0, ∞) be a continuous function and for each n ∈ N set fn (x) = f (x)1/n for all x ∈ X. Show that
R
fn dµ → µ(supp f ) as n → ∞ where supp f = {x ∈ X | f (x) > 0}.

Proof. here

Problem 10. Let X be a compact metric space and let x ∈ X. Suppose that the point mass δx is
the weak* limit of a sequence of atomless Radon measures on X (viewing all of these measures as
elements of C(X)∗ ). Show that every neighborhood of x is uncountable.

Proof. here

2 January 2019

Problem 1. True or false (prove or give a counter example)

(a) Let E ⊆ R be a Borel set, then {(x, y) ∈ R2 | x − y ∈ E} is a Borel set in R2 .

5
2 JANUARY 2019 Kari Eifler

Proof. TRUE.
Define f (x, y) = x − y : R2 → R. This is continuous. Let

A := {S ⊆ R | f −1 (S) is a Borel set of R2 }

Then A is a σ-algebra (easy to check). If S is open, then f −1 (S) is open in R2 , thus Borel. So
{open sets} ⊆ A and so the Borel algebra is a subset of A. In particular, E ∈ A.

(b) Let E ⊆ Q := [0, 1] × [0, 1]. Assume that for every x, y ∈ [0, 1] the sets Ex = {y ∈ [0, 1] | (x, y) ∈
E} and E y = {x ∈ [0, 1] | (x, y) ∈ E} are Borel. Then E is Borel.

Proof. FALSE.
Consider a non-Borel set A ⊂ [0, 1]. Set E = {(x, x) | x ∈ A}. Then each E y and Ex is a
singleton which is Borel, but E is not.

(c) A function f : R → R is called Lipschitz if there exits a M > 0 such that ∀x, y ∈ R, |f (x) −
f (y)| ≤ M |x − y|. If A ⊆ R is Lebesgue measureable and f is Lipschitz then f (A) is Lebesgue
measurable.

Proof. TRUE.
Since A is Lebesgue measurable, then we can write A = (∪j Kj ) ∪ N where each Kj is a compact
set and N has Lebesgue measure zero. Then f (A) = (∪j f (Kj )) ∪ f (N ). It’s clear that each
f (Kj ) is Lebesgue measurable, since f is Lipschitz. We are only left to see that f (N ) is also
Lebesgue measurable.
Indeed, for every  > 0 we can write N ⊆ ∪k Bk where each Bk is a ball of radius rk and
P
k m(Bk ) < . But then by Lipschitz continuity, f (Bk ) is contained in a ball of radius M rk
where M is the Lipschitz constant of f . Thus, m(f (Bk )) ≤ M m(Bk ) so that m(f (N )) ≤
P
M k m(Bk ) < M . Let  → 0 so f (N ) must have outer measure equal to zero, hence it is a
null set.

Problem 2. Let (X, F, µ) be a measure space. is it true that for every measurable essentially bounded
f : X → R we have limp→∞ kf kp = kf k∞ ? Give an answer both in the case that µ is finite and the
case that µ is σ-finite.

Proof. If µ is finite: By Hölder, we know that kf kp ≤ kf kq when p ≤ q. Also, kf kp ≤ kf k∞ for all p.

Therefore, kf kp %≤ kf k∞ and so limp kf kp ≤ kf k∞ .

On the other hand, for every  > 0, let E = {x | |f (x)| > kf k∞ − } and 0 < µ(E) ≤ 1 since kf k∞ =

p
esssup |f (x)| < ∞. Then kf kpp ≥ E |f |p > kf k∞ −  µ(E). Take p → ∞ so limp kf kp ≥ kf k∞ − ,
R

implying limp kf kp ≥ kf k∞ .
1
If µ is σ-finite: No, this is not true. Consider f (x) = x on [1, ∞). Then limp kf kp = 0 6= kf k∞ =
1.

Problem 3. Let f : R → R Lebesgue integrable and for n ∈ N define

Z
gn (x) = n f dλ.
(x,x+1/n)

6
2 JANUARY 2019 Kari Eifler

(a) Prove that limn→∞ gn = f λ-a.e.

Proof. This is Lebesgue Differentiation Theorem, with Er = (x, x + r).

R R
(b) Prove that for every n ∈ N, R
|gn |dλ ≤ R
|f |dλ.

Proof. here!
R R
(c) Prove limn→∞ R
|gn |dλ = R
|f |dλ.

Proof. Apply dominated convergence theorem with parts (a) and (b).

Problem 4. Let f ∈ L1 ((0, 1]2 , λ2 ) such that

R
(0,x]×(0,y]
f dλ2 = 0 for every x, y ∈ (0, 1]. Prove that
f = 0 λ2 -a.e.

Proof. First note that

[
(a, b) × (c, d) = ((0, b − 1/n] × (0, d − 1/n]) \ ((0, a] × (0, 1] ∪ (0, 1] × (0, b])
n

And since all open rectangles generate all Borel sets in R2 , then we have that for every Borel set
B ⊆ R2 , B f dλ2 = 0.
R

Since every Lebesgue set A is of the form A = B ∪ N where B is a Borel measurable set and N is a
R
set of measure zero. Hence, A f dλ2 = 0 for any Lebesgue measurable set A.

Now consider A+ = {x | f (x) > 0} and A− = {x | f (x) < 0}. Since both are measurable, then
R R
A+
f dλ2 = 0 = A− f dλ2 . Hence, f = 0 λ2 -a.e.

Problem 5. Let λ be the Lebesgue measure on R. Let E ⊆ R be Lebesgue measurable such that
0 < λ(E) < ∞. Prove that for all 0 ≤ γ < 1 there exists an open interval I ⊆ R such that

λ(E ∩ I) ≥ γλ(I).

Proof. Choose an open set O ⊃ E such that λ(E) ≥ γλ(O). We can write O = ∪i OI for open and
disjoint intervals Oi . Hence

[ [
E =E∩O =E∩ Oi = (E ∩ Oi )
i i

Suppose to the contrary that λ(E ∩ Oi ) < γλ(Oi ) for all i. Then

[  X X
λ(E) = λ (E ∩ Oi ) = λ(E ∩ Oi ) < γ λ(Oi ) = γλ(O)
i i

which is a contradiction with the fact that λ(E) ≥ γλ(O). Hence, it must be that for some k, λ(E ∩
Ok ) ≥ γλ(Ok ).

7
2 JANUARY 2019 Kari Eifler

Problem 6. Let X be a compact metrizable space and {µn } a sequence of Borel measures on X
with µn (X) = 1 for every n. Consider the linear map ϕ : C(X) → `∞ (N) defined by ϕ(f ) =
R

X
f dµn n . What conditions on the sequence {µn } are equivalent to ϕ being an isometry? Provide
justification.

Proof. We would require, for all f ∈ C(X)

Z  Z 
 
sup |f (x)| = kf k∞ =
f dµn = sup  f dµn 

n
x∈X

∞

here

Problem 7. Let X be a compact metric space and {fn } a sequence in C(X). Prove that {fn } con-
verges weakly in C(X) if and only if it converges pointwise and supn kfn k < ∞. Also, give an exam-
ple of an X and a sequence {fn } in C(X) which converges weakly but not uniformly.

Proof. By considering fn − f , we may assume without loss of generality that fn converges to 0.

⇒) We know C[0, 1]∗ = M[0, 1]. Then fn → 0 weakly implies fn dµ → 0 for all µ ∈ M[0, 1].
R

Choose µ = δt so

Z
fn dδt = fn (t) → 0 ∀t ∈ [0, 1]

(this follows from the fact that weak convergence implies uniformly bounded). Consider

χ : C[0, 1] → C[0, 1]∗∗ = M[0, 1]∗

χ(fn )(µ) = µ(fn )

Since µ(fn ) → 0 then χ(fn )(µ) → 0 for all µ ∈ M[0, 1]. Since convergent sequences are bounded,
then supn |χ(fn )(µ)| ≤ M .

By the uniform boundedness theorem, supn kχ(fn )k < ∞. By isometry, kfn k = kχ(fn )k so supn kfn k <
∞.

⇐) By Dominated Convergence Theorem, fn → 0 in L1 (µ). So therefore, | fn dµ| ≤ |fn |d|µ| → 0.

R R

So fn → 0 weakly.

Example: Take X = [0, 1] and consider the functions


nx

 x ∈ [0, 1/n]
fn (x) = −nx + 2 x ∈ (1/n, 2/n]

x ∈ (2/n, 1]

0

Then they converge to 0 weakly, but not strongly.

8
2 JANUARY 2019 Kari Eifler

Problem 8. Let X be a Banach space. Show that if X ∗∗ is separable then so is X. Also, give an
example, with justification, to show that the converse is false.

Proof. We will show the weaker result that states that if the dual X ∗ is separable, then so is X.

Let X ∗ be separable. Consider the unit sphere SX ∗ = {ϕ ∈ X ∗ | kϕk = 1}. Then SX ∗ is separable
and so we can let {ϕn } be a countable dense subset of SX ∗ .

For each n ∈ N, choose xn ∈ N with kxn k = 1 such that |ϕn (xn )| > 1/2. Let D = span{x1 , x2 , . . .}.

Then D is countable; ex. we can consider the following set countable and dense subset of D:

 
[ Xn 
(aj + ibj )xj | aj , bj ∈ Q
 
n∈N j=1

We want to show that D = X. Suppose it were not, then there ewould be some ϕ ∈ SX ∗ with
ϕ|D = 0. Since {ϕn } is dense, there exists some n such that kϕ − ϕn k < 1/4. Therefore,

1 1
≤ |ϕn (xn )| = |ϕn (xn ) − ϕ(xn )| ≤ kϕn − ϕkkxn k < .
2 4

This is a contradiction and hence, D = X.

example. c0 is separable, but `∞ = c∗∗

0 is not separable.

Problem 9. (a) Let X be a compact metrizable space. Describe the dual of C(X) according to the
Riesz representation theorem.

Proof. For every ϕ ∈ C(X)∗ , there exists a unique finite regular signed measure µ on the Borel
subsets of X such that
Z
ϕ(f ) = f dµ
X

for each f ∈ C(X). Moreover, kϕk = |µ|(X).

(b) Consider the spaces X = {1/n | n ∈ N} ∪ {0} and Y = [0, 1] with the topologies inherited from
R. Prove that there does not exist a bijective bounded linear map from C(X) to C(Y ).

Proof. By contradiction. Suppose there exists a bijective bounded linear map T : C(X) →
C(Y ). Then by the Open Mapping Theorem (or more accurately, the corollary that is the Bounded
Inverse Theorem), then T −1 is a bijective bounded linear map from C(Y ) to C(X). This says
that the two spaces are isomorphic.
Therefore, the duals of these two spaces should also be isomorphic, C(X)∗ ∼
= C(Y )∗ . But by the
Riesz-representation theorem, here!

Problem 10. Let X be a Banach space and Y a subspace of X. Show that kx + Y k = inf{kx + yk |
y ∈ Y } defines a norm on X/Y if and only if Y is closed.

9
3 AUGUST 2018 Kari Eifler

Proof. ⇐) Suppose Y is closed. It’s easy to see kx + Y k is well-defined and a semi-norm. Suppose
kx + Y k = 0. Then there exists yn ∈ Y such that kx − yn k → 0. Since Y is closed, then x ∈ Y .
Therefore, x + Y = Y = 0 + Y which is the zero vector in X/Y .

⇒) Suppose this is a norm. Take any convergent sequence yn in Y with yn → y 0 . Then inf y∈Y ky −
y 0 k ≤ kyn − y 0 k → 0 and so ky 0 + Y k = 0. Since this is a norm, then y 0 + Y = 0 + Y = Y and so
y 0 ∈ Y . Hence Y must be closed.

3 August 2018

(Solve any 10 of the following 12 problems)

Problem 1. Let µ and ν be positive measures on the same measurable space with ν finite and abso-
lutely continuous with respect to µ. Show that for every  > 0, there exists δ > 0 such that µ(E) < δ
implies ν(E) < .

Proof. Suppose for contradiction that ∃ > 0 such that µ(E) < δ then ν(E) ≥  for all δ > 0 adn
for some E. We’ll construct the set En to be some set with µ(En ) < 2−n . Let Fk = ∪∞n=k En so
µ(Fk ) < 2−k+1 .

Let F = ∩∞
k=1 Fk so µ(F ) = 0. Since ν  µ, then ν(F ) = 0.

However, since Fk is a decreasing sequence, we have

ν(F ) = lim ν (∩nk=1 Fk ) = lim ν(Fn ) ≥ .

n n

Contradiction!

Problem 2. Let µ be a positive measure. Suppose that (fn )∞ 1

n=1 is a Cauchy sequence in L (µ).
Show that for all  > 0 there exists a δ > 0 such that µ(E) < δ implies

Z 
 
∀n ≥ 1  fn dµ < .
 
E

You may use without proof the result of problem #1.

Proof. Let  > 0. Since {fn } is Cauchy in L1 (µ), there exists f ∈ L1 (mu) such that fn → f in
L1 (µ) as n → ∞, since L1 (µ) is a Banach space.
R 
Define ν(E) :=  E f dµ.
R 
Then by Problem 1, there exists some δ > 0 such that ν(E) =  E f dµ < /2 when µ(E) < δ, thenf
or large enough n (say n ≥ N ) we have

Z  Z  Z  Z 
 fn dµ =  (fn − f + f )dµ ≤  fn − f dµ +  f dµ <  +  = 
       
        2 2
E E E E

10
3 AUGUST 2018 Kari Eifler

for µ(E) < δ.

R 
For each i, N , we can find δi such that  E fi dµ  <  when  µ(E) < δi . By the same reasoning as
R
above, if we set δ̃ = min{δ1 , . . . , δN −1 , δ} then  E fn dµ <  whenever µ(E) < δ̃ for all n ∈ N.

Problem 3. Let f : [0, 1] → [0, ∞) be Lebesgue measurable. For n ∈ N define

fn
gn = .
1 + fn

R1
(a) Explain why 0
gn (t)dt exists and is finite for all n.

fn
R1 R1
Proof. Since gn = 1+f n ≤ 1 for all n, then 0
gn dx ≤ 0
1dx = 1 for all n.
R1
(b) Prove that limn 0 gn (t)dt exists and find an expression for it. Make sure to state which major
theorems you are using in your proof.

Proof. Define E1 = {x | 0 ≤ f (x) < 1}, E2 = {x | f (x) = 1} and E3 = {x | f (x) > 1}.
f n (x) R R
If x ∈ E1 then gn (x) = 1+f n (x) → 0. So by DCT, limn E gn dx = E 0dx = 0.
1 1

f n (x) 1
If x ∈ E2 then gn (x) = 1+f n (x) = 2 for all n and so
Z Z
1 1
lim gn dx = dx = m(E2 ).
n E2 E2 2 2
f n (x)
If x ∈ E3 then gn (x) = 1+f n (x) → 1 and so by DCT,
Z Z
lim gn dx = dx = m(E3 ).
n E3 E3

Thus,

Z 1 Z Z Z
1
lim gn dx = lim gn dx + gn dx + gn dx = m(E2 ) + m(E3 ).
n 0 n E1 E2 E3 2

Problem 4. Consider C([0, 1]) endowed with its usual uniform norm. Prove or disprove that there
is a bounded linear functional ϕ on C([0, 1]) such that for all polynomials p, we have ϕ(p) = p0 (0),
where p0 is the derivative of p.

Proof. DISPROVE.

Consider pn = 1 − (x − 1)n so then kpn k∞ = 1 but p0n (0) = n → ∞. If such a ϕ existed, then
n = |ϕ(pn )| = kϕ(pn )k ≤ ckpn k which cannot happen.

Problem 5. (a) Define the product topology on the Cartesian product Πα∈A Xα of a family of topo-
logical spaces (Xα )α∈A

Proof. here!

11
3 AUGUST 2018 Kari Eifler

(b) State Tychonoff ’s compactness theorem.

Proof. If {Xα } is a family of compact topological spaces then Πα∈A Xα is compact.

(c) State and prove the Banach-Alaoglu theorem (Hint: Use Tychonoff ’s theorem)

Proof. Thoerem: Let X be a normed vector space. The closed unit ball {f ∈ X ∗ | kf k ≤ 1} is
compact in the weak*-topology.
For all x ∈ X, let Dx := {ξ | |ξ| ≤ kxk} ⊆ C. Then Dx is compact, and by Tychonoff’s theorem,
D := Πx∈X Dx is comapct. Define complex function ϕ with ϕ(x) ≤ kxk.
We define B ∗ ⊆ D to consist of linear functions of D. We claim B ∗ is closed. Indeed, let {fα }
be a net in B ∗ that converges to f . Then

f (ax + by) = lim fα (ax + by) = lim(afα (x) + bfα (y)) = a lim fα (x) + b lim fα (y) = af (x) + bf (y).

So f ∈ B ∗ . Since closed subsets of comapct spaces are compact, then B ∗ is compact in the
weak*-topology.

Problem 6. Let (X, d) be a compact metric space.

(a) Show that X has a countable, dense set {xn | n ∈ N}.

Proof. If X is countable, we are done. So suppose X is uncountable. Since X is compact, for

all n ∈ N, X can be covered by finitely many balls of radius n1 . For each n, choose such a finite
cover with balls centered at the points {xnj }N n Nn
j=1 . Then the collection E := ∪n {xj }j=1 is count-
n

able.
For x ∈ X, for all n ∈ N, x ∈ B(1/n, xnj ) for some xnj ∈ E so E is dense.

(b) Let fn : X → [0, ∞) be fn (x) = d(x, xn ). Show that if x, y ∈ X and fn (x) = fn (y) for all n ∈ N,
then x = y.

Proof. We then have that d(x, xn ) = d(y, xn ) for all n. We know for all m ∈ N we can find xm
such that d(x, xm ) < 1/m so d(y, xm ) < 1/m. So we can find a sequence {xm }∞ m=1 such that
xm → x and xm → y as m → ∞. But X is a metric space and thus Hausdorff, so limits are
unique. Therefore, x = y.

Problem 7. Let K > 0 and let LipK be the set of functions f : R → R satisfying |f (x) − f (y)| ≤
K|x − y|.

(a) Prove that

∞
X
d(f1 , f2 ) = 2−j sup |f1 (x) − f2 (x)|
j=0 x∈[−j,j]

defines a metric on LipK

12
3 AUGUST 2018 Kari Eifler

P∞
Proof. First, suppose d(f1 , f2 ) = 0. Then j=0 2−j supx∈[−j,j] |f1 (x)−f2 (x)| = 0 so supx∈[−j,j] |f1 (x)−
f2 (x)| = 0 for all j. Thus, f1 (x) = f2 (x) for all x.
It’s trivial to see that d(f1 , f2 ) = d(f2 , f1 ).
Finally,we’ll show the triangle inequality. This again follows directly: |f1 (x) − f2 (x)| ≤ |f1 (x) −
f3 (x)| + |f3 (x) − f2 (x)| for all x. Taking sup on both sodies and multiplying by 2−j we get
d(f1 , f2 ) ≤ d(f1 , f3 ) + d(f1 , f2 ).

(b) Prove that LipK is a complete metric space

Proof. Suppose (fn ) is a Cauchy sequence in LipK . Then for every  > 0 there exists N ∈ N
P∞
such that d(f1 , fm ) = j=1 2−j supx∈[−j,j] |f1 (x) − f2 (x)| < . Then for each j and x ∈ [−j, j]
we have |fn (x) − fm (x)| < 0 .
Thus, {fn (ξ)} is Cauchy sequence on [−j, j] for each ξ. But we can find f (x) such that fn (x) →
f (x).
We want to show that d(fn , f ) → 0. Since fn (x) → f (x), then for all  > 0 we can find some
N ∈ N such that for all n ≥ N , |fn (x) − f (x)| < . THen

∞
X ∞
X
d(fn , f ) = 2−j sup |fn (x) − f (x)| < 2−j  = 
j=1 x∈[−j,j] j=1

So d(fn , f ) → 0. To see f ∈ LipK ,

 
|f (x) − f (y)| = lim fn (x) − lim fn (y) = lim |fn (x) − fn (y)| ≤ K lim |x − y| = K|x − y|.
 
n n n n

Problem 8. Let X, Y be topological spaces. A map f : X → Y is said to be proper if for every

compact subset K ⊆ Y , the inverse image f −1 (K) is compact.

(a) Suppose X is a compact space and Y is Hausdorff. Prove that every continuous map f : X → Y
is proper.

Proof. Let K ⊆ Y be compact. Since Y is Hausdorff, then K is closed. Since f is continuous,

and Y \K is open in Y then f −1 (Y \K) is open in X. So f −1 (K) = X\f −1 (Y \K) is closed.
Since X is compact, f −1 (K) is compact.

(b) Give an example of a continuous map which is not proper.

Proof. Consider the constant function 1 : R → R which sends x 7→ 1. So 1−1 ({1}) = R.

(c) Suppose f : Rm → Rn is a proper continuous map. Prove that f is a closed map, ie. f (C) is
closed in Rn whenever C is a closed subset of Rm .

Proof. Let {yn } ⊆ f (C) with yn → y. Define A = {y} ∪ {yn } (compact). Then f −1 (A) is
compact, so there exists xn ∈ f −1 (A) ∩ C such that f (xn ) = yn . Find a convergent subsequence
xnk with xnk → x for x ∈ C ∩ f −1 (A). By continuity of f , we have f (x) = y.

13
3 AUGUST 2018 Kari Eifler

Problem 9. Consider the interval [−π, π] equipped with Lebesgue measure µ. For n ∈ Z, consider
the functions fn ∈ C([−π, π]) given by fn (t) = eint .

(a) Prove that spanC {fn | n ∈ Z} is dense in the space

A := {f ∈ C([−π, π]) | f (−π) = f (π)}

with respect to the uniform norm.

Proof. Let B = spanC {fn } ⊆ A ⊆ C([−π, π]). Note that B separates points and is closed under
complex conjugates. By Stone-Weierstrass, B is dense in C[−π, π] hence also dense in B.
n o
(b) Show that √fn | n ∈ Z is an orthonormal basis for the Hilbert space L2 ([−π, π], µ).
2π

Proof. Note that

1/2
π √
Z
int −int
 
khfn , fn ik2 =  e e dt = 2π
−π

For n 6= m,

π
ei(n−m)t π
Z
hfn , fm i = eint e−int dt = | = 0.
−π n − m −π

So they are orthonormal.

(c) Is the following statement true or false?:

1
PN
“For every f ∈ A, f = limN →∞ 2π n=−N hf, fn ifn with respect to the uniform norm.”

Give a brief explanation why or why not.

Proof. TRUE.
P∞ P∞ P∞
Claim: −∞ hf, fn ifn exists.
P∞
By Pythagorean theorem, −∞ hf, fn ifn = −∞ khf, fn ifn k.
By Bessel’s inequality, −∞ hf, fn ifn is bounded so it exists.
P∞
Let g := f − −∞ hf, fn ifn so that

∞
X
hg, fm i = hf, fm i − hhf, fn ifn , fm i = hf, fm i − hf, fm i = 0.
−∞
P∞
By completeness of Hilbert spaces, g = 0. So f = −∞ hf, fn ifn .

Problem 10. Let (X, k · k) be a normed linear space and let (X ∗ , k · kX ∗ ) denote its dual Banach
space of bounded linear functionals. Recall that kϕkX ∗ = supkxk=1 |ϕ(x)| for ϕ ∈ X ∗

(a) Prove that for each x ∈ X, there exits ϕ ∈ X ∗ with kϕkX ∗ = 1 and kxk = ϕ(x).

14
3 AUGUST 2018 Kari Eifler

Proof. We will prove the more general case: let M be closed and x ∈ X\M . Then there exists
φ ∈ X ∗ such that φ(x) = inf y∈M kx − yk and kφk = 1 and φ|M = 0.
Restrict to the space M + Cx and define φ(y + λx) = λ inf y∈M kx − yk. Then φ(x) = inf y∈M kx −
yk and φ|M = 0.
kxk |φ(x)|
Since φ(x) = kxk, then 1 = kxk = kxk ≤ kφk and

|φ(y + λx)| ≤ |φ(y)| + |φ(λx)| = 0 + |λ||φ(x)| = |λ| inf kx − yk ≤ |λ|kx − λ−1 yk = kλx + yk.
y∈M

|φ(y+λx)|
Therefore, kφk = supy+λx kλx+yk ≤ 1 so kφk = 1.
Finally, if we define p(x) = kxk for x ∈ M + Cx then by Hahn-Banach, φ can be extended to ψ
on all x with ψ|M +Cx = φ. To prove the result, set M = {0}.

(b) Prove that the linear map ι : X → X ∗∗ given by

ι(x)(ϕ) = ϕ(x) x ∈ X, ϕ ∈ X ∗

is an isometry.

Proof. Fix x ∈ X, so

|ι(x)(φ)| |φ(x)|
kι(x)k = = sup
kφkX ∗ φ∈X ∗ kφkX ∗

By part (a), there exists φ ∈ X ∗ such that kφkX ∗ = 1 and φ(x) = kxk, which implies that
kxk ≤ kι(x)kX ∗ .
Also, for any φ ∈ X ∗ , |φ(x)| ≤ kφkX ∗ kxk and so

|φ(x)| kφkkxk
kι(x)k ≤ sup ≤ = kxk.
φ∈X ∗ kφkX ∗ kφk

So kι(x)k = kxk and so ι is an isometry.

(c) A Banach space X is called reflexive if ι(X) = X ∗∗ . Prove that the Banach space

X
`1 = {f ∈ N → C | kf k1 = |f (k)| < ∞}.
k

is not reflexive.
Hint: Consider a weak-∗ cluster point of the sequence (ι(fn ))n∈N ⊆ (`1 )∗∗ , where fn ∈ `2 is the
unit vector
(
1/n k≤n
fn (k) =
0 k>n

Proof. here!

15
3 AUGUST 2018 Kari Eifler

Problem 11. Let (gn )n∈N ⊆ C([0, 1]) be a sequence of non-negative continuous functions. Assume
that for each k = 0, 1, 2, . . . the limit

Z 1
lim xk gn (x)dx exists.
n→∞ 0

Prove that there exists a unique finite positive Radom measure µ on [0, 1] such that

Z 1 Z 1
f (x)dµ(x) = lim f (x)gn (x)dx for all f ∈ C([0, 1]).
0 n→∞ 0

R1
Proof. Define M := limn 0 gn (x)dx < ∞. Let A = span{xk | k ∈ N}. For each φ ∈ A, by linearity,
R1
limn 0 φ(x)gn (x)dx exists.
R1
By Stone-Weierstrass, A is dense in C[0, 1], so for every f ∈ C[0, 1], limn 0 f (x)gn (x)dx.
R1
Next, let φ : C[0, 1] → C be defined by φ(f ) = lim 0 f (x)gn (x)dx. Linearity is obvious. Moreover,
for every f ∈ C[0, 1],



Z 1 

Z 1 Z 1
|φ(f )| = lim f gn (x)dx ≤ lim |f (x)||gn (x)|dx ≤ kf kn lim gn (x)dx = M kf k∞
n 0 n 0 n 0

Hence, φ is a bounded linear functional on C[0, 1].

By Riesz-Representation, there exists a positive Radon measure µ such taht

Z 1 Z 1
lim f (x)gn (x)dx = φ(f ) = f (x)dµ(x) ∀f ∈ C[0, 1].
n 0 0

Problem 12. Let X be a locally compact Hausdorff space equipped with a Radon probability mea-
sure µ. Let E ⊆ L2 (X, µ) be a closed linear subspace and assume that E is contained in C0 (X). The
goal of this problem is to prove that dim(E) < ∞ by justifying the following steps:

(a) There exists a constant 1 ≤ K < ∞ such that

kf k2 ≤ kf ku ≤ Kkf k2 for all f ∈ E,

where k · ku denotes the uniform norm. Hint: us the closed graph theorem for one of the inequal-
ities.

(b) For each x ∈ X, there exists a unique gx ∈ E such that kgx k2 ≤ K and

f (x) = hf, gx i for all f ∈ E.

(c) Let (fi )i∈I be any orthonormal basis for E. Then

16
4 JANUARY 2018 Kari Eifler

X
|fi (x)|2 = kgx k22 ≤ K 2 for all x ∈ X.
i∈I

(d) dim(E) = |I| ≤ K 2 .

Proof. See January 2017, Problem #5 for a solution to a similar question.

4 January 2018

Problem 1. Suppose U1 , U2 , . . . are open subsets of [0, 1]. In each case, either prove the statement
or disprove it.

(a) If λ ∩∞n=1 Un ) = 0 then for some n ≥ 1, we have λ(Un ) < 1, where λ is Lebesgue measure and
Un is the closure of Un in the usual topology on [0, 1].

Proof. FALSE. Let rm be an enumeration of the rationals on [0, 1], and set an,m = 1/2n+m . Set

Un := ∪m (rm − an,m , rm + an,m )

These are open since they are a union of open intervals. Moreover, since Q ⊆ Un then λ(Un ) =
λ([0, 1]) = 1. But by upper continuity of the Lebesgue measure, then

λ (∩Un ) = lim λ (∪(rn − an,m , rn + an,m )) = 0.

m

(b) If ∩∞
n=1 Un = ∅, then for some n ≥ 1, the set [0, 1]\Un contains a non-empty open interval.

Proof. TRUE. Recall that the Baire Category Theorem states that under these assumptions, if
each Un is dense in [0, 1] then ∩∞ ∞
n=1 Un is also dense in [0, 1]. Then since we have that ∩n=1 Un
is not dense, then there must be some n such that [0, 1]\Un is not dense. This precisely means
that Un contains a non-empty open interval.

Problem 2. Let X be a separable compact metric space and show that C(X) is separable.

Proof. Remark: If X is a compact metric space, then X is separable. So the separable assumption
is superfluous.

Suppose d is the metric on X and (xn ) is a dense countable subset of X. For each n ∈ N, de-
fine the functional fn by fn (x) := d(x, xn ). Then each fn is a continuous functional. Consider
F = {1, f1 , f2 , . . .} and consider the subalgebra generated by the rational span of F , call it Q[F ]
(this is still countable, we can consider the span, then cosider the set where two elements of it are
multiplied together, then the set where three elements are multiplied together, etc). This is count-
able and dense in A := R[F ]. so it is sufficient to show that A is dense in C(X).

17
4 JANUARY 2018 Kari Eifler

We will attempt to use the Stone Weierstrass Theorem:

By definition, R[F ] contains the constant function 1. We are left to show it separates points. Take
two points x 6= y in X. Since {xn } is dense, then there must exist some m such that d(x, xm ) ≤
1
3 d(x, y) 6= 0. If d(y, xm ) = d(x, xm ) then

2
d(x, y) ≤ d(x, xm ) + d(y, xm ) = 2d(x, xm ) ≤ d(x, y).
3

This cannot be true under our assumption d(x, y) 6= 0. So then fm (y) = d(y, xm ) 6= d(x, xm ) =
fm (x). So fm separates x and y.

Therefore, by Stone-Weierstrass, A is dense in C(X). But Q[F ] is countable and dense in A, so

therefore C(X) is separable.

Problem 3. Let f : [0, 1] → R be a bounded Lebesgue measurable function such that

Z 1
f (t)ent dt = 0
0

for every n ∈ {0, 1, 2, . . .}. Prove that f (t) = 0 for almost every t ∈ [0, 1].

R1
Proof. Let f (t) = 0 on t = 0, 1. Using Stone-Weierstrass to show we can pass to the case 0
f (t)g(t) =
0 for all g ∈ C[0, 1].

By a standard density argument, we may pass to the case where g is a step function. We claim that
f = 0 a.e.

Assume not. WLOG there exists some E = {x ∈ [0, 1] | f (x) > 0} with m(E) > 0 (else consider
−f ).

Since f is bounded, then E∞ := {x ∈ [1, 2] | f (x) = ∞} is a null set. Define En := {x ∈ [0, 1] |

S
1/n < f (x) < n}. We can write E = ( n En ) ∪ E∞ . So there exists some N such that m(EN ) = a >
0.

We can write A as a finite disjoint union of open intervals, A = tm

i=1 Ii , such that m(EN 4A) < 
and A ⊆ EN .
Pm R2
Put g = i=1 χIi , then 1 g(x)f (x) = EN f (x)dx. Since
R

Z Z 
 

 f (x) − f (x) ≤ N m(EN 4A) < N 
EN A

R1
If we choose  small enough, we see the contradiction since 0
g(x)f (x) > 0.

Problem 4. (a) Prove that every compact subset of a Hausdorff space is closed.

Proof. Let A be a compact subset of the Hausdorff space X. To show A is closed, we’ll show
Ac = X\A is open. Take x ∈ X\A. Then for every y ∈ A, there are disjoint sets Uy and Vy with
x ∈ Vy and y ∈ Uy .

18
4 JANUARY 2018 Kari Eifler

The collection of open sets {Uy | y ∈ A} forms an open cover of A. Since A is compact, this
open cover has a finite subcover, Uy1 , Uy2 , . . . , Uyn . Let

U := ∪ni=1 Uyi V := ∩ni=1 Vyi

Since each Uyi and Vyi are disjoint, then U and V are disjoint. Also, A ⊆ U and x ∈ V .
Thus, for every point x ∈ X\A we have found an open set V containing x which is disjoint from
A. So X\A is open and A is closed.

(b) Let f : X → Y be a bijective continuous function between topological spaces. Suppose that X is
compact and Y is Hausdorff and prove that f is a homeomorphism.

Proof. Let g = f −1 . We need to show that g is continuous.

For every V ⊆ X, we have g −1 (V ) = f (V ). We want to show that if V is closed in X then
g −1 (V ) is closed in Y .
Suppose V is closed in X. Since X is compact, V is compact by part (a). So f (V ) is compact
since the continuous image of a compact space is compact.
Since Y is Hausdorff, f (V ) is closed by the fact that a compact subspace of Hausdorff space is
closed. But f (V ) = g −1 (V ) so g −1 (V ) is closed. So g is continuous and f is a homeomorphism.

(c) Prove or disprove that if X is a dense subset of a topological space Y and if X is Hausdorff in
the relative topology, then Y is also Hausdorff.

Proof. FALSE. Consider Y = {a, b} with discrete topology τ = {∅, {a, b}}. Let X = {a} with
relative topology τX = {∅, {a}}.
Then it’s easy to see X is dense in Y (since every open set containing an element of Y has nonempty
intersection with X, trivially). Since X has only a single element, it’s Hausdorff in the relative
topology trivially. But Y is not Hausdorff.

Problem 5. Prove that the following limit exists and compute its value:

n
!
n
(−1)k x2k
Z X
lim e−2x dx.
n→∞ 0 (2k)!
k=0

Proof. Solution from Sheagan John

Let us first note an important simplification of the integrand, by considering the Taylor series ex-
pansion of cos x around a neighbourhood of 0.

∞
X (−1)k
cos x = x2k
(2k)!
k=0

k
Letting f (x) := (cos x)e−2x and fk (x) := e−2x (−1)
(2k)! x
2k
then it’s clear that {fk } is a convergent
sequence which converges to f (x).

19
4 JANUARY 2018 Kari Eifler

Now, since | cos x| ≤ 1 we further have that for some positive constant c < 2

|fn (x)| ≤ c|f (x)| = c| cos xe−2x | = c| cos x|e−2x ≤ ce−2x

Since g(x) = ce−2x is integrable on the positive half line

∞ ∞
−c −2x 
Z
c
ce−2x dx = e  =2
0 2 0

the dominated convergence theorem can be applied to the original Lebesgue integral limit, with
fn −→ f , and with g as the dominating function.

∞
!
n n ∞ ∞
(−1)k
Z X Z Z Z
lim x 2k
e−2x dx = lim fn (x) dx = f (x) dx = e−2x cos x dx
n−→∞ 0 (2k)! n−→∞ 0 0 0
k=0

R∞
Recall that the Laplace transform of cos x is 0
e−st cos atdt = L{cos ax}(s) = s
s2 +a2 . Therefore,
the last integral is equal to 222+1 = 2/5.

Problem 6. Let X and Y be Banach spaces (over C)

(a) A linear map T : X → Y is called adjointable if T ∗ f ∈ X ∗ for every f ∈ Y ∗ . Prove that T is

adjointable if and only if T ∈ B(X, Y ).

Proof. ⇐) if T ∈ B(X, Y ) then by definition, for every f ∈ Y ∗ , we have T ∗ f ∈ X ∗

⇒) Suppose T ∗ f ∈ X ∗ for every f ∈ Y ∗ . We will use the Closed Graph Theorem. Suppose
xn → x in X and that T xn → y in Y . Then since T ∗ f ∈ X ∗ for every f ∈ Y ∗ we can apply this
to the convergence to see that

f (T xn ) = (T ∗ f )(xn ) → (T ∗ f )(x) = f (T x) ∀f ∈ Y ∗

By the Hahn-Banach theorem, Y ∗ separates points in Y so therefore, T xn → T x. Uniqueness

of limits implies T x = y and so the graph of T is closed. By the Closed Graph Theorem, T is
bounded.

(b) Suppose a bounded linear functional ψ : X ∗ → C is weak*- continuous. Show (from the defini-
tions) that there exists x ∈ X such that ψ(φ) = φ(x).

Proof. Define the functional

evx : X ∗ → C
f 7→ f (x)

20
4 JANUARY 2018 Kari Eifler

We want to show that every bounded, linear, weak*-continuous functional ψ : X ∗ → C is of this

form.
Indeed, since ψ is weak*-continuous, then it is weak* continuous at 0. Thus, the set {f ∈ X ∗ |
|ψ(f )| < 1} is weak* open and must containa neighborhood of 0. By definition of weak* topol-
ogy, there must exist x1 , . . . , xn ∈ X such that

V (x1 , . . . , xn ) := {f ∈ X ∗ | |f (xi )| ≤ 1, i = 1, . . . , n} ⊆ {f ∈ X ∗ | |ψ(f )| < 1}.

Then we will next show that ∩ni=1 ker(evxi ) ⊆ ker(ψ).

Indeed, let f ∈ ker(evxi ) so |f (xi )| = 0 for all i = 1, . . . , n. Take  > 0 and conser g = 1 f , so
|g(xi )| = 1 |f (xi )| = 0 for all i = 1, . . . , n. In particular, g ∈ V (x1 , . . . , xn ) and so then we have
if |ψ(g)| < 1 then |ψ(f )| < . But  is arbitrary so ψ(f ) = 0, i.e. f ∈ ker(ψ).
Now recall the linear algebra trick that says if for linear functionals ker(T ) ⊆ ker(S) then S is a
scalar multiple of T . In this case, we get that ψ is a linear combination of the evxi , i.e. is of the
form evx where x is a linear combination of the xi ’s.
Moreover, because the weak* topology is Hausdorff, x is necessarily unique.

(c) Let S ∈ B(Y ∗ , X ∗ ). Prove that S is weak*-weak*-continuous if and only if S = T ∗ for some
T ∈ B(X, Y ).

Proof. ⇐) If S = T ∗ then if fα → f is a weak* convergent net in Y ∗ then for any y ∈ Y ,

fα (y) → f (y). Therefore,

(Sfα − Sf )(x) = (T x) (fα − f ) → 0.

| {z } | {z }
∈X ∗ →0

So S is weak*-weak* continuous.
⇒) Suppose S : Y ∗ → X ∗ is weak*-weak* continuous. Then the evaluation function on x,
evx (S) is weak* continuous on Y ∗ (where evx (S) : Y ∗ → C, (evx (S))(f ) = (Sf )(x)).
By part (b), we know that evx (S) is of the form evT (x) for some unique T (x) ∈ Y . Since T (x) is
uniquely determined, it follows that T is linear.
We will now check that T is continuous by the closed graph theorem: if xn → x adn T xn → y
in norm then for each φ ∈ Y ∗ we have

hφ, yi = limhφ, T xn i = limhSφ, xn i = hSφ, xi = hφ, T xi.

And so y = T x as desired. So T is bounded and therefore, S = T ∗ is bounded as well.

Problem 7. Let (fn )∞

n=1 be a sequence of functions fn : [0, 1] → R.

NOTE: I think we also require continuous....

(a) What does it mean for {fn | n ≥ 1} to be equicontinuous?

Proof. {fn | n ≥ 1} is said to be equicontinuous if for every  > 0, there exists a δ > 0 such that
for all x, y ∈ [0, 1], if |x − y| < δ then |f (x) − f (y)| < .

21
4 JANUARY 2018 Kari Eifler

(b) Suppose that for every n, fn is differentiable and |fn0 (t)| ≤ 1 for all t. Prove that {fn | n ≥ 1} is
equicontinuous.

Proof. Since |fn0 (t)| ≤ 1 for all t, then for all n, we have by the mean value theorem that

|fn (x) − fn (y)|

≤ 1.
|x − y|

Hence, for any fixed  > 0, setting δ =  and for |x − y| < δ then

|fn (x) − fn (y)| ≤ |x − y| < δ = .

(c) Suppose the hypothesis of (b) holds and assume in addition that |fn (0)| ≤ 1 for every n ≥ 1.
Prove that there exists a continuous function f : [0, 1] → R and a subsequence (fn(k) )∞
k=1 con-
verging uniformly to f .

Proof. This is essentially the Arzela-Ascoli Theorem. Since |fn (0)| ≤ 1 for all n and since
|fn0 (t)| ≤ 1 for all t, then |fn (t)| ≤ 2 for all t ∈ [0, 1] and for all n. That is, {fn } is uniformly
bounded. It’s also equicontinuous by part (b). Therefore, Arzela-Ascoli theorem states that
there is a subsequence {fnk } which converges uniformly. Let f be the limit, and we finish by
recalling that the uniform convergence of continuous functions is also continuous.
Note: it might be good to know the Arzela-Ascoli Theorem.

(d) Show by example that the limit function f need not be differentiable.
p
Proof. Take fn (x) = x2 + 1/n so fn (0) = √1 ≤ 1 for all n and so {fn } is uniformly bounded.
n q
Next, we can see that fn0 (x) = √ 2x so that for x ∈ [0, 1] we have |fn0 (x)| ≤ n
n+1 ≤ 1 as
x +1/n
desired.
However, it’s also clear that the limit must be f = |x| which is not differentiable.

Problem 8. Let H be a complex Hilbert space. Given a non-empty set E ⊆ H and x ∈ H, put
dist(x, E) = inf{kx − yk | y ∈ E} and E ⊥ = {x ∈ H | hx, yi = 0 ∀y ∈ E}.

(a) Let H0 ⊆ H be a closed subspace and x ∈ H. Prove that there exists x0 ∈ H0 such that kx −
x0 k = dist(x, H0 ).

Proof. Let δ = dist(x, H0 ). Then there exists a sequence (yn ) ∈ H0 such that δn := kx − yn k →
δ. We will show that (yn ) is Cauchy. Indeed,

0 ≤ kyn − ym k2 = −kyn + ym − 2xk2 + 2(kyn − xk2 + kym − xk2 ) ≤ −4δ 2 + 2(δn2 + δm

2
) → 0.

where we use the fact that

22
4 JANUARY 2018 Kari Eifler

2


2
y n + y m
kyn + ym − 2xk = 4

≤ 4δ.
−x
| {z 2 }

∈H0

Thus, (yn ) is a Cauchy sequence and so because we are in a Hilbert space, (yn ) converges to
some point x0 ∈ H. Since H0 is closed and yn ∈ H0 for all n then we get that x0 ∈ H0 . Finally,
kx − x0 k = lim kx − yn k = lim δn = δ.
Exercise: it can be shown if H0 is convex, then the choice of x0 is unique!

(b) With x and x0 as above, prove that x − x0 is orthogonal to H0 .

Proof. Let y ∈ H0 be an arbitrary vector with kyk = 1, set α := hx − x0 , yi. Then since αhx −
x0 , yi = αα = |α|2 and αhy, x − x0 i = αα = |α|2 , we have

kx−(x0 +αy)k2 = kx−x0 −αyk2 = kx−x0 k2 −αhx−x0 , yi−αhy, x−x0 i+|α|2 = kx−x0 k2 −|α|2 .

So since x0 + αy ∈ H0 then kx − x0 − αyk ≥ kx − x0 k. Hence α = 0.

Therefore, for any nonzero y ∈ H0 we can write

hx − x0 , yi = kykhx − x0 , y/kyki = kyk0 = 0.

So hx − x0 , yi = 0 for all y ∈ H0 so x − x0 ⊥ H0 .

(c) Prove that H = H0 ⊕ H0⊥ (the algebraic direct sum)

Proof. This follows immediately from parts (a) and (b). Take some arbitrary x ∈ H. We can
find the appropriate x0 as above, so x = x0 + (x − x0 ) ∈ H0 ⊕ H0⊥ .
The fact that it is a direct sum follows from the fact that H0 ∩ H0⊥ = {0}.

(d) Let E ⊆ H be non-empty. Prove that (E ⊥ )⊥ = E if and only if E is a closed subspace.

Proof. If E is closed, then the above parts (a),(b), and (c) apply and prove that (E p erp)⊥ =
E. To see the converse, we will instead show that (E ⊥ )⊥ = E. The desired result will then
immediately follow.
⊥ ⊥ ⊥
Since E ⊆ E then E ⊆ E ⊥ and therefore, (E ⊥ )⊥ ⊆ (E )⊥ . Since E is closed, then (E )⊥ =
E so (E ⊥ )⊥ ⊆ E.
Conversely, since E ⊥ is closed for every E (independent of whether E is closed or not) then
(E ⊥ )⊥ is closed and so since E ⊆ (E ⊥ )⊥ , then by the monotonicity of topological closure we
have that E ⊆ (E ⊥ )⊥ = (E ⊥ )⊥ .
Therefore, (E ⊥ )⊥ = E.

Problem 9. Let V be a vector space over R or C. Recall that a Hamel basis for V is a linearly in-
dependent subset of V whose linear span equals V .

23
4 JANUARY 2018 Kari Eifler

(a) Let S ⊆ V and suppose the linear span of S equals V . Show that V has a Hamel basis that is a
subset of S.

Proof. Choose a Hamel basis B of S. Then it is easy to check that B is a Hamel basis of V .

(b) Suppose V has an infinite Hamel basis and show that all hamel bases of V have the same cardi-
nality.

Proof. Suppose that {vi }i∈I and {uj }j∈J are two infinite bases for V . For each i ∈ I, then vi is
in the linear span of {uj }j∈J . Therefore, there exists a finite subset Ji ⊆ J such that vi is in the
linear span of the vectors {uj }j∈Ji . Therefore, V = span({vi }i∈I ) ⊆ span{uj }j∈∪Ji . Since no
proper subset of {uj }j∈J can span V , it follows that J = ∪i∈I Ji . Therefore |J| ≤ |I|.
A symmetric argument shows that |I| ≤ |J|.

Problem 10. Suppose (X, M, ρ) is a finite measure space and A ⊆ M is an algebra of sets with a
finitely additive complex measure µ : A → C such that |µ(E)| ≤ ρ(E) for all E ∈ A. Show that there
exists a complex measure ν : M → C whose restriction to A is µ and such that |ν(E)| ≤ ρ(E) for all
E ∈ M.

Hint: you may want to consider the subspace V ⊆ L1 (ρ) that is spanned by the set of characteristic
functions χE for E ∈ A, and a certain linear functional on V .

Proof. Solution from Minh Kha.

Pn
For each subalgebra U of M, we define SU to be the set of all simple functions of the form χ
i=1 ci Ei
where ci ∈ R, Ei ∈ U. Then SA is a vector subspace of SM .

Now define p : SM → R such that

( n n
)
X X
p(f ) = sup |ci |ρ(Ei ) | f = ci χEi , Ei ∩ Ej = ∅ ∀i 6= j, Ei ∈ M, ci ∈ R ∀f ∈ SM
i=1 i=1

It’s not difficult to check that p satisfies p(f + g) ≤ p(f ) + p(g) for all f, g ∈ SM and p(tf ) = tp(f )
for all t ∈ R+ , f ∈ SM . Thus, p is a seminorm and is just an extension of the total variation of the
measure p when you apply to the function f = 1.

Define a linear map T : SA → R defined by

Z
T (f ) = f dµ ∀f ∈ SA
X

This is linear because of the finite additive property of µ. Then |T (f )| ≤ p(f ) for all f ∈ SA . By
Hahn-Banach, we get a linear extension of T on SM , which we denote by Te. Moreover, this exten-
sion Te : SM → R satisfies |Te(f )| ≤ p(f ) for all f ∈ SM .

Now, we define a finite additive measure ν on M by letting ν(E) = Te(χE ) for all E ∈ M. Thus,
ν|A = µ and |ν(E)| ≤ p(E) for all E ∈ M.

24
5 AUGUST 2017 Kari Eifler

To check the countably additive property of ν, consider any countable collection of disjoint measur-
able subsets Ei ∈ M and so χ∪i Ei = i χEi . Thus, ν(∪i Ei ) = i ν(Ei ) since the series i Te(Ei )
P P P

converges (use |Te(f )| ≤ p(f ) for all f ∈ SM and properties of the measure p).

For the complex case, repeat the trick by proving the complex version of the Hahn-Banach theorem
from the real version.

5 August 2017

Problem 1. Let (Ω, A, µ) be a measure space and let {fn } be a sequence of measurable functions
P
on X. Prove, directly from the definition of convergence almost everywhere, that if n µ[|fn | >
1/n] < ∞, then the sequence {fn } converges almost everywhere to zero. Deduce that every sequence
of measurable functions that converges in measure to zero has a subsequence that converges almost
everywhere to zero.

Proof. Let E = {x ∈ Ω | limn |fn (x)| = 0}. We want µ(E c ) = 0. Let

∞ [
\ ∞
M= {x ∈ Ω | |fn (x)| > 1/n}
m=1 n=m

Since

∞  ! ∞  
[ 1 X 1
µ x ∈ Ω | |fn (x)| > ≤ µ x ∈ Ω | |fn (x)| > → 0.
n=m
n n=m
n

Therefore, µ(M ) = 0 and

∞ \
[ ∞
Mc = {x ∈ Ω | |fn (x)| ≤ 1/n}.
m=1 n=m

Note: fn (x) → 0 if and only if ∀ > 0, ∃N s.t. ∀n > N , |fn (x)| < .
1 1
So for any x ∈ M c choose 1/N <  s.t. ∀n > N we have |fn (x)| ≤ n < N < .

Therefore M c ⊆ E, so E c ⊆ M , implying µ(E c ) = 0. So then {fn } converges almost everywhere to

zero.

Step 2: We will show that if fn → 0 in measure, then there exists a subsequence that converges to 0
pointwise almost everywhere.

Suppose for every  > 0, µ({x | |fn (x)| ≥ }) → 0. Choose a subsequence {fnk } such that if

Ej = {x | |fnj (x) − fnj+1 (x)| > 2−j }

25
5 AUGUST 2017 Kari Eifler

P∞
satisfies µ(Ej ) < 2−j . Let Fk = ∪∞
j=k Ej so µ(Fk ) ≤ j=k 2−j ≤ 21−k . Let F = ∩k Fk so µ(F ) = 0.

For x ∈
/ Fk and for i ≥ j ≥ k then

i−1
X i−1
X
|fni (x) − fnj (x)| ≤ |fn` (x) − fn`+1 (x)| ≤ 2` ≤ 2−j → 0 as k → ∞.
`=j `=j

So fnk is pointwise Cauchy on x ∈

/ F , so let

(
lim fnk (x) x ∈
/F
f (x) =
0 otherwise

So fnk → 0 almost everywhere and fn → f in measure since

µ({x | |fn (x) − f (x)| ≥ }) ≤ µ({x | |fn (x) − fn` (x)| ≥ /2}) + µ({x | |fn` (x) − f (x)| ≥ })
| {z } | {z }
→0 →0

and

µ({x | |f (x)| ≥ }) ≤ µ({x | |f (x) − fn (x)| ≥ /2} + µ({x | |fn (x)| ≥ /2})
| {z } | {z }
→0 →0

so f = 0 almost everywhere. Thus, {fnk } converges to 0 almost everywhere.

Problem 2. Show that there is a sequence of nonnegative functions {fn } in L1 (R) such that kfn kL1 (R) →
0, but for any x ∈ R, lim supn fn (x) = ∞.

Proof. We will explicitely construct such a sequence. Consider the following pattern:

To cover [−1, 1] let

√ √
f1 = 1χ[−1,0] , f2 = 1χ[0,1] .

so that kf1 kL1 (R) = 1 = kf2 kL1 (R) . To cover [−2, 2], next let

√ √ √
f3 = 2χ[−2,−1.5] , f4 = 2χ[−1.5−1] . . . , f10 = 2χ[1.5,2]

so then √1 = kf3 kL1 (R) = kf4 kL1 (R) = . . . = kf1 0kL1 (R) . Next, we cover [−3, 3] so that
2

√ √
f11 = 3χ[−3,−2.666] , . . . f28 = 3χ[2.666,3]

so that √13 = kf11 kL1 (R) = . . . = kf28 kL1 (R) . If we continue in this fashion, we get the desired
functions.

26
5 AUGUST 2017 Kari Eifler

PN −1
Explicitely, for n = i=1 2i2 + k = 13 (N − 1)(N )(2N − 1) + k where N ∈ N, 0 ≤ k < 2N 2 , then we
set

√
fn = N χ[−N +k/N,−N +(k+1)/N ]

so that kfn kL1 (R) = √1 but for every x ∈ R, it’s clear that lim supn fn (x) = ∞.
N

Problem 3. Construct a sequence of nonnegative Lebesgue measurable functions {fn } on [0, 1] such
that

(a) fn → 0 almost everywhere, and

(b) for any interval [a, b] ⊆ [0, 1],
Z b
lim fn (x)dx = b − a.
n→∞ a

(f (y) − f (x))dy = o( n1 ).
R
Proof. Claim. For any f ≥ 0 that is continuous on [0, 1], n [x,x+ n12 ]

1
R
For any  > 0, ∃N ∈ N such that 0 < x < N then f (x)−f (0) < ε. Hence for n > N , n [0, n12 ]
(f (x)−
f (0)) < n n12 .

(f (y) − f (x))dy = o( n1 ).
R
A clear extension to this is that 1
[x,x+ n ]
Pn−1
Let fn (x) := k=0 nχ[ k , k + 12 ] , and let ε, N be as above. Clearly fn is measurable and satisfies (a)
n n n
above. To prove (b), observe that, for n > N ,

Z k+1 
n
 
Ik :=  nf χ[ k , k + 12 ] − f 
 
 k n n n 
n
 Z k 1 k+1

n + n2
 Z n

= n f− f
 
 k k 
n n
 
1
=o .
n

R Pn−1
Hence | (f fn − f )| ≤ k=0 Ik = o(1).

This holds for all f ∈ C([0, 1]), so in particular, it will hold for f = χ[a,b] .

Problem 4. In this problem the measure is Lebesgue measure on [0, 1]. The norm on L∞ [0, 1] is
the essential supremum norm, which for a continuous function is the same as the supremum norm.

(a) Prove or disprove that L∞ [0, 1] is separable in the norm topology.

Proof. L∞ [0, 1] is not separable in the norm topology. Consider the collection of functions fr =
χ[−r,r] for real 1 ≥ r > 0. Since there are uncountably many such r and since kfr − fr0 k∞ = 1
for any r 6= r0 , it’s impossible to have a countable subset of L∞ [0, 1] that is dense in it.

27
5 AUGUST 2017 Kari Eifler

(b) Recall that L∞ [0, 1] = (L1 [0, 1])∗ . What is the weak* closure in L∞ [0, 1] of the unit ball of
C[0, 1]? Prove your assertion.

w∗
Proof. More general claim: For X an infinite dimensional Banach space, SX ∗ = BX ∗ .
We know for any x1 , x2 , . . . , xn ∈ X, there exists some x∗0
6= 0 such that x∗0 (xi )
= 0. Indeed,
if this were not true then otherwise, x∗0 (xi ) 6= 0 for some i, let ϕ : X ∗ → R be ϕ(x∗ ) = n

(x∗ (x1 ), . . . , x∗ (xn )) then ϕ is injective so dim(X ∗ ) ≤ dim(Rn ) = n. Contradiction, so true.

Now for any x∗ ∈ BX ∗ , consider it’s neighborhood (open under the w∗ -neighborhood)

V = ∩ni=1 {y ∗ ∈ X ∗ | |x̂i (x∗ − y ∗ )| = |x∗ (xi ) − y ∗ (xi )| < }

for each {xi }ni=1 choose such an x∗0 6= 0 from the claim.
Consider the line {x∗ + tx∗0 | t ∈ R} in X ∗ .
Since for any x̂i ,

x̂i (x∗ + tx∗0 − x∗ ) = tx̂i (x∗0 ) = tx∗0 (xi ) = 0 < .

Then {x∗ + tx∗0 | t ∈ R} ⊆ V . Since kx∗ + tx∗0 k is continuous about t, then we can find t0 such
that kx∗ + t0 x∗0 k = 1 ⇒ V ∩ SX ∗ 6= ∅.
Since any neighborhood of x∗ contains a neighborhood of the form V as above (i.e. these V ’s
w∗
are a neighborhood basis) then BX ∗ ⊆ SX ∗ .
On the other hand, for any x∗0
∈ BX ∗ , by Hahn-Banach separation Theorem, we know there
exists x ∈ X and c ∈ R such that x∗ (x) < c < x∗0 (x) for all x∗ ∈ BX ∗ .
Then for all {x∗n } ⊆ BX ∗ , x∗n (x) ≤ c < x∗0 (x). Therefore, x∗0 isn’t an accumulation point of BX ∗
w∗ w∗ w∗ w∗
which implies BX ∗ = BX ∗ . Thus, SX ∗ ⊆ BX ∗ = BX ∗ so BX ∗ = SX ∗ .

Problem 5. Prove that if a1 , a2 , . . . , aN are complex numbers, then

R 1 PN PN
(a) 0
| k=1 ak exp(2πikt)|p dt ≤ k=1 |ak |p , if 1 ≤ p ≤ 2, and
R 1 PN PN
(b) 0 | k=1 ak exp(2πikt)|p dt ≥ k=1 |ak |p , if 2 ≤ p < ∞.

Proof. Note first the following facts:

• {exp(2πikt)} is orthonormal in L2

• For a finite measure space and p ≤ q, then

kf kp ≤ µ(X)1/p−1/q kf kq

• For a discrete X and p ≤ q, kf kq ≤ kf kp .

28
5 AUGUST 2017 Kari Eifler

Since {exp(wπikt)} is orthonormal, then

2
XN N
X
ak exp(2πikt) = |ak |2 .



k=1 2 k=1

PN
Then if we let a = (a1 , . . . , aN ) and f = k=1 ak exp(2πikt), we see that for 1 ≤ p ≤ 2, we have

N p
Z 1 X 
ak exp(2πikt) dt = kf kpp ≤ kf kp2 = kakp2 ≤ kakpp .
 

0 
k=1


To see (b), then similarly for 2 ≤ p < ∞,

 p
Z 1 XN 
ak exp(2πikt) dt = kf kpp ≥ kf kp2 = kakp2 ≥ kakpp .
 

) 
k=1


Problem 6. Prove that if X is an infinite dimensional Banach space and X ∗ is separable in the
norm topology, then there is a sequence {xn } of norm one vectors in X such that {xn } converges
weakly to zero.

Proof. Suppose {x∗n } is a dense, countable subset of X ∗ .

Claim: For every n, then ∩nk=1 ker(x∗k ) is non-trivial.

Indeed, assume to the contrary that ∩nk=1 ker(x∗k ) = {0}. Then the map

F : X → Fn
x 7→ (x∗1 (x), . . . , x∗n (x))

is linear and injective. Let {e1 , . . . , em } be a basis for F (X). Choose yk ∈ F −1 ({ek }). For all x ∈
Pm P P P
X, we can write F (x) = i=1 ai ei . so F (x − ai yi ) = ai ei − ai ei = 0 so x is in the span and
then X must be finite dimensional, contradiction! So the claim holds.

Now, choose xn ∈ SX ∩ (∩nk=1 ker(xk )). Fix x∗ ∈ X ∗ ,  > 0, so ∃N ∈ N such that kx∗ − x∗N k < .
Then for all n ≥ N , xn ∈ ker(x∗N ) so

|x∗ (xn )| = |(x∗ − x∗n )(xn )| ≤ kx∗ − x∗N k < 

So then x∗ (xn ) → 0.

Problem 7. Prove or disprove each of the following statements.

(a) If {fn } is a sequence in C[0, 1] that converges weakly, then also {fn2 } converges weakly.

29
5 AUGUST 2017 Kari Eifler

Proof. YES. Recall that fn ∈ C[0, 1] converges weakly if and only if it converges pointwise and
is uniformly bounded.
Suppose fn → f weakly, let M := supn kfn k < ∞. Then fn → f pointwise so fn2 → f 2 pointwise
and supn kfn2 k = M 2 < ∞. So fn2 → f 2 weakly.

(b) If {fn } is a sequence in L2 [0, 1] that converges weakly, then also {fn2 } converges weakly. (Lebesgue
measure on [0, 1])

Proof. NO. Take fn (x) = x−1/3 χ[1/n,1] (x), so fn → f = x−1/3 in norm but fn2 (x) = x−2/3 χ[1/n,1] (x)
but
Z 1 Z 1
fn2 (x)x−1/3 dx = x−1 χ[1/n,1] = log(n) → ∞.
0 0

Problem 8. Let {fn } be a sequence of continuous functions on R that converges pointwise to a real
valued function f . Prove that for each a < b, the function f is continuous at some point of [a, b].

Hint: Let En,m,k = [|fn − fm | ≤ 1/k].

Proof. Fix some [a, b] ⊆ [0, 1]. By Egoroff’s Theorem, fn → f uniformly outside a set of measure
b−a
2 . Then f must be continuous outside of this set.

Note: Likely, the question was meant to prove Egoroff’s theorem, see Folland for that proof!

Problem 9. Let X and Y be compact Hausdorff spaces and let S be the set of all real functions on
X × Y of the form h(x, y) = f (x)g(y) with f in C(X) and g in C(Y ).

Prove or disprove that the linear span of S is dense in C(X × Y ).

Proof. We will use Stone-Weirstrass theorem here. Note that if h1 (x, y) = f1 (x)g1 (y) and h2 (x, y) =
f2 (x)g2 (y) are two functions in S, then

(h1 h2 )(x, y) = h1 (x, y)h2 (x, y) = f1 (x)g1 (y)f2 (x)g2 (y) = (f1 f2 )(x)(g1 g2 )(y)

where if f1 , f2 ∈ C(X) then so is f1 f2 (and similarly, g1 g2 ∈ C(Y )). So then S is an algebra. Thus,
it follows that span(S) is an algebra as well.

Next, S separates points. Indeed, suppose (x, y) 6= (x0 , y 0 ) in X × Y . If x 6= x0 then choose some
f ∈ C(X) that separates x and x0 . Take g ∈ C(Y ) to be the constant function g = 1. Then letting
h(x, y) := f (x)g(y) = f (x), h separates the two points. If x = x0 then y 6= y 0 so the same trick
works, setting f = 1 ∈ C(X) and choosing g to separate y and y 0 , letting h(x, y) := f (x)g(y) = g(y)
to then separate points.

Therefore, by the Stone-Weierstass theorem, span(S) is dense in C(X × Y ).

Problem 10. Let X be a Hilbert space and assume that {xn } is a sequence in X that converges
PN
weakly to zero. Prove that there is a subsequence {yk } of {xn } such that the sequence kN −1 k=1 yk k
converges to zero.

30
5 AUGUST 2017 Kari Eifler

Caution: the same statement is NOT true in all Banach spaces, not even in all reflexive Banach
spaces.

Proof. Note: This is the Banach-Saks Theorem

We shall successively choose the nk in the following manner. Beginning for definiteness with n1 = 1,
let n2 be the first index such that |hf1 , fn i| ≤ 1 (this choice is possible since hf1 , fn i → 0 as n →
∞). In general, after having chosen fn1 , fn2 , . . . , fnk , we choose nk+1 so that

1 1
|hfn1 , fnk+1 i| ≤ , . . . , |hfnk , fnk+1 i| ≤
k k

Since {fn } converges weakly, then it is bounded and so kfn k forms a bounded sequence, say kfn k ≤
M so by expanding the inner product, we get

fn1 + fn2 + . . . + fnk 2 kM 2 + 2 × 1 + 4 × 12 + . . . + 2(k − 1) × 1

M2 + 2

≤ k−1
<
k k2 k

which then implies

fn1 + fn2 + . . . + fnk 2

→ 0.
k

Problem 11. Let F ⊆ C([0, 1]) be a family of continuous functions such that

(a) the derivative f 0 (t) exists for all t ∈ (0, 1) and f ∈ F .

(b) supf ∈F |f (0)| < ∞ and supf ∈F supt∈(0,1) |f 0 (t)| < ∞.

Prove that F is precompact in the Banach space C([0, 1]) equipped with the norm kf k = supt∈[0,1] |f (t)|.

Proof. We will use the Arzela-Ascoli Theorem.

To see F is equicontinuous, fix some  > 0, and let δ = M 

where M = supf ∈F supt∈(0,1) |f 0 (t)| < ∞.
Then by the mean value theorem, for any a < b, there exists some c ∈ (a, b) such that f 0 (c) =
f (b)−f (a)
b−a so that |f (b) − f (a)| ≤ |f 0 (c)||b − a| ≤ M |b − a| < M δ = .

To see F is pointwise bounded, we see that for any b ∈ [0, 1], then for some c ∈ [0, b], we have f (b) =
f 0 (c)b + f (0), so that

|f (b)| ≤ M + sup |f (0)|.

f ∈F

That is, F is uniformly bounded!

Then by Arzela-Ascoli, F is compact.

31
6 JANUARY 2017 Kari Eifler

Problem 12. Let {xn } be a weakly Cauchy sequence in a normed linear space X. Prove that

(a) xn is norm bounded in X

Proof. Let c denote the space of convergent sequences, and consider the map

T : X∗ → c
x∗ 7→ (x∗ (xn ))

By the Uniform Boundedness Principle, T is closed. Then by the closed graph theorem, T is
continuous, so kT k < ∞. By Hahn-Banach Theorem, kT k = supn kxn k.

(b) There exists x∗∗ in X ∗∗ such that xn converges weak* to x∗∗ , and kx∗∗ k ≤ lim inf kxn k.

Proof. Since (xn ) is weakly Cauchy, then for every x∗ ∈ X ∗ the sequence (x∗ (xn )) is Cauchy,
hence convergent. We can define

x∗∗ : X ∗∗ → C
x∗ 7→ lim x∗ (xn )
n

Uniform boundedness shows that kxn k is bounded, hence x∗∗ is bounded. Finally,
  
|x∗∗ (x∗ )| = lim inf |x∗ (xn )| ≤ lim inf kx∗ kkxn k = lim inf kxn k kx∗ k .

So then kx∗∗ k ≤ lim inf kxn k.

6 January 2017

Problem 1. Let (Ω, A, µ) be a measure space. Prove directly from the definition of convergence
almost everywhere that if for all n, µ x ∈ Ω | |fn (x)| > n1 < n−3/2 , then fn → 0 µ-a.e.


Proof. Let E = {x ∈ Ω | limn |fn (x)| = 0}. We want µ(E c ) = 0. Let

∞ [
\ ∞
M= {x ∈ Ω | |fn (x)| > 1/n}
m=1 n=m

Since

∞ ∞ ∞
!  
[ 1 X 1 X
µ {x ∈ Ω | |fn (x)| > } ≤ µ x ∈ Ω | |fn (x)| > < n−3/2 → 0.
n=m
n n=m
n n=m

32
6 JANUARY 2017 Kari Eifler

Therefore, µ(M ) = 0 and

∞ \
[ ∞
Mc = {x ∈ Ω | |fn (x)| ≤ 1/n}.
m=1 n=m

Note: fn (x) → 0 if and only if ∀ > 0, ∃N s.t. ∀n > N , |fn (x)| < .
1 1
So for any x ∈ M c choose 1/N <  s.t. ∀n > N we have |fn (x)| ≤ n < N < .

Therefore M c ⊆ E, so E c ⊆ M , implying µ(E c ) = 0.

R2
Problem 2. Find all f in L1 (1, 2) such that for every natural number n we have 1
x2n f (x)dx = 0.
Give reasons for all assertions you make.

Proof. Let f (x) = 0 on x = 1, 2. We now consider f ∈ L1 [1, 2]. Using Stone-Weierstrass to show we
R2
can pass to the case 1 g(x)f (x) = 0 for all g ∈ C[1, 2].

By a standard density argument, we may pass to the case where g is a step function. We claim that
f = 0 a.e.

Assume not. WLOG there exists some E = {x ∈ [1, 2] | f (x) > 0} with m(E) > 0 (else consider
−f ).

Since f ∈ L1 [1, 2] then E∞ := {x ∈ [1, 2] | f (x) = ∞} is a null set. Define En := {x ∈ [1, 2] | 1/n <
S
f (x) < n}. We can write E = ( n En ) ∪ E∞ . So there exists some N such that m(EN ) = a > 0.

We can write A as a finite disjoint union of open intervals, A = tm

i=1 Ii , such that m(EN 4A) < 
and A ⊆ EN .
Pm R2
Put g = i=1 χIi , then 1 g(x)f (x) = EN f (x)dx. Since
R

Z Z 
 

 f (x) − f (x) ≤ N m(EN 4A) < N 
EN A

R2
If we choose  small enough, we see the contradiction since 1
g(x)f (x) > 0.

Problem 3. A. Prove that there exists a sequence of measurable functions gn on [0, 1] such that

(a) gn (x) ≥ 0 for any x ∈ [0, 1];

(b) limn gn (x) = 0 a.e.;
(c) For any continuous function f ∈ C[0, 1],
Z 1 Z 1
lim f (x)gn (x)dx = f (x)dx.
n→∞ 0 0

Proof. (Solution from Ting Lu, TeX-ed by John Weeks)

It suffices to assume f is non-negative. Any f ∈ C[0, 1] is uniformly continuous since [0, 1] is
compact. The following lemma will then come in handy:
Claim. With f as above, n [x,x+ 1 ] (f (y) − f (x))dy = o( n1 ).
R
n2

33
6 JANUARY 2017 Kari Eifler

For any  > 0, ∃N ∈ N such that 0 < x < N1 then f (x) − f (0) < ε. Hence for n > N ,
n [0, 1 ] (f (x) − f (0)) < n n12 .
R
n2

A clear extension to this is that [x,x+ 1 ] (f (y) − f (x))dy = o( n1 ).

R
n
Pn−1
Let gn (x) := k=0 nχ[ k , k + 12 ] , and let ε, N be as above. Clearly gn is measurable and satisfies
n n n
(a) and (b) above. To prove (c), observe that, for n > N ,

Z k+1 
n
 
Ik :=  nf χ[ k , k + 12 ] − f 
 
 k n n n 
n
 Z k 1 k+1

n + n2
 Z n 
= n f− f
 
 k k 
n n
 
1
=o .
n

R Pn−1
Hence | (f gn − f )| ≤ k=0 Ik = o(1).

B. If gn is a sequence of measurable functions on [0, 1] such that (a), (b), and (c) are satisfied,
R1
what can you say about 0 supn gn (x)dx?

Proof. here

Problem 4. We say that a sequence {an }∞

n=1 in [0, 1] is equidistributed (in [0, 1]) if and only if for
all intervals [c, d] ⊂ [0, 1],

|{a1 , . . . , an } ∩ [c, d]|

lim = d − c.
n→∞ n

(Here |A| denotes the number of elements in the set A.)

Let µN = N1 1≤n≤N δan with δan the point measure at an , that is, for any subset S ∈ [0, 1],
P
(
1 if an ∈ S
δan (S) = .
0 if an ∈/S

Show that {an } ⊂ [0, 1] is equidistributed if and only if

Z 1 Z 1
lim f dµN = f dm,
N →∞ 0 0

for all continuous functions on [0, 1], where m is Lebesgue measure.

Proof. Note that {an } is equidistributed if and only if

|{a1 , . . . , an } ∩ [c, d]|

lim =d−c
n n

34
6 JANUARY 2017 Kari Eifler

if and only if

Z 1 Z 1
lim f dµN = f dm for f simple functions (since we can take f = χ[c,d] )
n 0 0

⇒) It’s easy to see if {an } is equidistributed for f = χ[c,d] .

1 1
|{a1 , . . . , aN } ∩ [c, d]|
Z Z
lim f dµN = lim =d−c= f dm
N 0 N N 0

Thus, ”=” holds for step functions.

Using Darboux’s definition of integral for f ∈ C[0, 1], ∀ > 0 there exists step functions f1 , f2 such
R1
that f1 ≤ f ≤ f2 and 0 (f2 − f1 )dx <  where the lower sum is

Z 1 N N
1 X 1 X
f1 (x)dx = lim f1 (an ) ≤ lim inf f (an )
0 N N 1 N N 1

and the upper sum is

Z 1 N N
1 X 1 X
f2 (x)dx = lim f2 (an ) ≥ lim sup f (an )
0 N N 1 N N 1

Then

 
N N
 1 X 1 X 
lim sup f (an ) − lim inf f (an ) ≤ .
 
 N N 1 N N 1 

1
PN R1
Therefore limN f (an ) exists and by definition must be 0 f dµ.
N 1
R1 R1
⇐) If we know limK 0 gn dµK = 0 gn dµ for all gn ∈ C[0, 1]. Let f = χ[c,d] , choose gn → f in L1
and each gn & f positive, gn ∈ C[0, 1] with gn |[c,d] = 1 = f |[c,d] .
R1 R1
We want to show limK 0 f dµK = 0 f dm. Indeed,

35
6 JANUARY 2017 Kari Eifler

Z 1 Z 1  Z d Z d 

  
f dµK − f dm =  f dµK − f dm
  

0 0  c c 
Z 
 d Z d 
= gn dµK − f dm
 
 c c 
Z d Z d
= gn dµK − f dm
c c
Z 1 Z 1
≤ gn dµK − f dm
0 0
Z
 1 Z 1  Z
  1 Z 1 

≤  gn dµK − gn dm +  gn dm − f dm → 0.
0 0 0 0

Problem 5. Consider the space C([0, 1]) of real-valued continuous functions on the unit interval
[0, 1]. We denote by kf k∞ := supx∈[0,1] |f (x)| the supremum norm of f ∈ C([0, 1]) and by kf k2 :=
R1 1
( 0 |f (x)|2 dx) 2 the L2 -norm of f ∈ C([0, 1]). Let S be a closed linear subspace of (C([0, 1]), k · k∞ ).
Show that if S is complete in the norm k · k2 , then S is finite-dimensional.

Proof. Let T : (S, k · k2 ) → (S, k · k∞ ) by T (x) = x. Note that both spaces are complete.

Assume xn → x in k · k2 and T (xn ) → y in k · k∞ then

kT (xn ) − yk2 ≤ kT (xn ) − yk∞ → 0.

So

kx − yk2 ≤ kx − T (xn )k2 + kT (xn ) − yk2 ≤ kx − xn k2 + kT (xn ) − yk∞ → 0

so x = T (x) = y.

Therefore, by closed graph theorem, we know T is bounded. So there exists some C such that kf k∞ ≤
Ckf k2 .

Now let f1 , . . . , fn be an orthonormal family in S. Then for all fixed x ∈ [0, 1]

f1 (x)2 + · · · + fn (x)2 ≤ kf1 (x)f1 + · · · + fn (x)fn k∞ ≤ Ckf1 (x)f1 + · · · + fn (x)fn k2

So then because fn ’s are orthogonal and kfk k22 = 1,

(f1 (x)2 + · · · + fn (x)2 )2 ≤ C 2 f1 (x)2 kf1 k22 + · · · + fn (x)2 kfn k22 = C 2 (f1 (x)2 + · · · + fn (x)2 )

36
6 JANUARY 2017 Kari Eifler

Then f1 (x)2 + · · · + fn (x)2 ≤ C 2 . So

Z 1 Z 1
n= f1 (x)2 + · · · + fn (x)2 dx ≤ C 2 dx = C 2 ⇒ n ≤ C 2
0 0

Thus, the number of orthogonal family in S is at most C 2 . So S is finite dimensional.

Problem 6. Prove that if a function f : [0, 1] → R is Lipschitz, with

|f (x) − f (y)| ≤ M |x − y|

for all x, y ∈ [0, 1], then there is a sequence of continuously differentialbe functions fn : [0, 1] → R
such that

(i) |fn0 (x)| ≤ M for all x ∈ [0, 1];

(ii) fn (x) → f (x) for all x ∈ [0, 1].

Proof. It’s easy to prove f is aboslutely continuous ⇒ f is of bounded variable ⇒ f is differentiable

a.e. ⇒ f 0 exists a.e.

Also, when f 0 exists, |f 0 (x)| ≤ M .

Then there exists simple φ1 , φ2 , . . . such that 0 ≤ |φ1 | ≤ |φ2 | ≤ · · · ≤ |φn | ≤ · · · ≤ |f 0 | ≤ M and
φn → f 0 uniformly on [0, 1] where f 0 exists. Define

Z x Z x
fn (x) := φn (t)dt + f (0) f (x) := f 0 (t)dt + f (0)
0 0

Then |fn0 (x)| = |φn (x)| ≤ M and for all x ∈ [0, 1],

Z x
|fn (x) − f (x)| ≤ |φn (t) − f 0 (t)|dt → 0
0

since φn converges to f 0 uniformly.

Problem 7. Given f : R → R bounded and uniformly continuous and Kn with Kn ∈ L1 (R) for
n = 1, 2, 3, . . . such that

(i) kKn k1 ≤ M < ∞, n = 1, 2, 3, . . .

R∞
(ii) −∞ Kn (x)dx → 1 as n → ∞.
R
(iii) {x||x|>δ} |Kn (x)| → 0 as n → ∞ for all δ > 0.

Show that Kn ∗ f → f uniformly, where

37
6 JANUARY 2017 Kari Eifler

Z ∞
Kn ∗ f (x) = Kn (y)f (x − y)dy.
−∞

Proof. For all x ∈ R,

 Z ∞  Z ∞ 
   
|Kn ∗ f (x) − f (x)| ≤ Kn ∗ f (x) −
 Kn (y)f (x)dy  + 
  Kn (y)f (x)dy − f (x)
−∞ −∞
Z ∞ Z ∞ 
 
≤ |Kn (y)||f (x − y) − f (x)|dy + kf k∞ 
 Kn (y)dy − 1
−∞ −∞
Z ∞
≤ |Kn (y)||f (x − y) − f (x)|dy + c
Z−∞ Z
= |Kn (y)||f (x − y) − f (x)|dy + |Kn (y)||f (x − y) − f (x)|dy + c
B(0,δ) B(0,δ)c

R
For B(0,δ)
|Kn (y)||f (x − y) − f (x)|dy, by uniform continuity we ahve

Z Z
|Kn (y)||f (x − y) − f (x)|dy ≤  |Kn (y)|dy ≤ kKn k1 < M
B(0,δ) B(0,δ)

R
For B(0,δ)c
|Kn (y)||f (x − y) − f (x)|dy, by the third assumption we have

Z Z
|Kn (y)||f (x − y) − f (x)|dy ≤ 2kf k∞ |Kn (y)|dy ≤ 2C
B(0,δ)c B(0,δ)c

Let  → 0, so we’ve got it.

Problem 8. (a) Construct a Lebesgue measurable subset A of R so that for all reals a < b, 0 <
m(A ∩ [a, b]) < b − a where m is Lebesgue measure on R.

Proof. Enumerate all rational intervals I1 , I2 , . . . . For each In , construct a fat Cantor set Nn ⊆
In with positive measure.
Since Nn is nowhere dense, there exists some interval Ien ⊆ In and Ien ∩ Nn = ∅.
Construct another fat Cantor set Mn ⊆ Ien and define A := ∪Mn .
Now, for all I = [a, b] there exists some n such that Nn ⊆ In ⊆ I with Nn ∩ A = ∅ (can be done
by induction). We see m(A ∩ I) ≥ m(Mn ) > 0 and

m(A ∩ I) < m(I\Nn ) = m(I) − m(Nn ) < m(I) = b − a.

(b) Suppose A ⊆ R is a Lebesgue measurable set and assume that

b−a
m(A ∩ (a, b)) ≤
2

38
6 JANUARY 2017 Kari Eifler

for any a, b ∈ R, a < b. Prove that µ(A) = 0.

Proof. Consider an open set U ⊇ A with m(U \A) < . Then U = t∞

i=1 (ai , bi ) and measurable.
So

m(U ) = m(A ∩ U ) + m(U ∩ Ac ) < m(A ∩ U ) + 

Since

∞ ∞
X X bi − ai 1
m(A ∩ U ) = m (A ∩ (t∞
i=1 (ai , bi ))) = m(A ∩ (ai , bi )) ≤ = m(U )
i=1 i=1
2 2

then

1
m(U ) < m(U ) +  ⇒ m(U ) < 2 ⇒ m(A) ≤ m(U ) → 0
2

Problem 9. Prove or disprove that the unit ball of L7 (0, 1) is norm closed in L1 (0, 1).

Proof. Let

 Z 1 
B := f| |f |7 dx ≤ 1 .
0

R1
Let {fn } ⊆ B such that fn → f in L1 . We want to show f ∈ B ⇔ 0
|f |7 dx ≤ 1.

Since fn → f in L1 , then fn → f in measure. Thus, there exists a subsequence fnk such that fnk →
f a.e.

Therefore, |fnk |7 → |f |7 a.e.. By Fatou’s Lemma,

Z 1 Z 1
|f |7 dx ≤ lim inf |fnk |7 dx ≤ 1.
0 k 0

Problem 10. Let C be the Banach space of convergent sequences of real numbers under the supre-
mum norm. Compute the extreme points of the closed unit ball, B, of C and determine whether B
is the closed convex hull of its extreme points.

Proof. If |x(m)| < 1 for some m then there exists δ > 0 such that |x(m) − δ| ≤ 1, |x(m) + δ| ≤ 1.

Define y1 , y2 ∈ B such that

y1 (n) = x(n) for n 6= m and y1 (m) = x(m) + δ

y2 (n) = x(n) for n 6= m and y2 (m) = x(m) − δ

39
7 AUGUST 2016 Kari Eifler

Then y1 6= y2 and x = 12 (y1 + y2 ) so x is not an extreme point.

If |x(n)| = 1 for all n, if x = λy1 + (1 − λ)y2 for y1 6= y2 ∈ B then since |yi (n)| ≤ n,

|x(n)| = 1 = |λy1 (n) + (1 − λ)y2 (n)| ≤ λ|y1 (n)| + (1 − λ)|y2 (n)| ≤ 1

Equality holds only when y1 (n) = y2 (n) = ±1. So y1 = y2 so x is indeed an extreme point. Also, x
needs to be convergent so

Ext(B) = {x | |x(n)| = 1 ∃N s.t. x(n) = 1 or − 1 for all n > N }.

Problem in my proof of determining whether B is the closed convex hull of it’s extreme points.

Problem 11. Show that every convex continuous function defined on the convex unit ball of a re-
flexive Banach space achieves a minimum. (A convex function on a convex subset A of a normed
space is a real valued function, f , on A s.t. for every x, y ∈ A and every 0 < λ < 1 we have
f (λx + (1 − λ)y) ≤ λf (x) + (1 − λ)f (y).)

Proof. Recall the following classical result in convex analysis: f is lower-semicontinuous convex ⇔
f is weakly lower-semicontinuous convex ⇒ f can achieve minimum (since the unit ball is weak-
compact).

Then by Alaoglu, closed unit ball of a reflexive Banach space is weak-compact = weak*-compact.

here

7 August 2016

Problem 1. Let A be the set of all real valued functions on [0, 1] for which f (0) = 0 and |f (t) −
f (s)|1/2 ≤ t − s for all 0 ≤ s < t ≤ 1

(a) Prove that A is a compact subset of C[0, 1].

Proof. It should be clear to the reader that this question requires Arezela-Ascoli Theorem. To
√
see A is equicontinuous, fix x ∈ [0, 1] and  > 0. Then for y ∈ B( , x),

|f (x) − f (y)| ≤ |x − y|2 < 

For pointwise bounded, for x ∈ [0, 1] then |f (x)|1/2 = |f (x) − f (0)|1/2 ≤ x implies |f (x)| ≤ x2 .
To see A is closed, take a sequence {fn } ⊆ A such that fn → f (i.e. for all open U containing
f , there exists N such that for all n ≥ N , fn ∈ U ), then

|f (t) − f (s)| ≤ |f (t) − fn (t)| + |fn (t) − fn (s)| + |fn (s) − f (s)| < 2 + |t − s|2

This holds for all  > 0 so |f (t) − f (s)| ≤ |t − s|2 .

40
7 AUGUST 2016 Kari Eifler

Clearly, f (0) = 0 so f ∈ A. Thus A is closed so by Arzela-Ascoli, A is compact in C[0, 1].

(b) Prove that A is a compact subset of L1 [0, 1]

Proof. Consider the map

id : C[0, 1] → L1 [0, 1]
f 7→ f

R1
Since k id k1 = 0
|f |dx ≤ kf k∞ , id is a bounded map.
From (a), A is compact in C[0, 1] so id(A) = A ⊆ L1 [0, 1] is also compact.
Remark: A is also closed in L1 [0, 1] since all compact subsets of a metric space is closed.

Problem 2. (a) Let f (x) be a real valued function on the real line that is differentiable almost ev-
erywhere. Prove that f 0 (x) is a Lebesgue measurable function.

Proof. Let

f (x + 1/n) − f (x)
fn 9x) =
1/n

so fn → f 0 almost everywhere. Since f is differentiable almost everywhere, then f is continuous

almost everywhere.
Claim: f is Lebesgue measurable
Let D = {all discontinuities of f } so m(D) = 0 and D is measurable. Let E = Dc = {x |
f is continuous at x} so E is measurable too.

f −1 ((a, ∞)) = f −1 ((a, ∞) ∩ E) ∪ f −1 ((a, ∞) ∩ E c )

Since f |E is continuous, f −1 (a, ∞) ∩ E = f |−1

E (a, ∞) is open in E. So f
−1
(a, ∞) ∩ E = U ∩ E
−1
for some open set U ⊆ R. Then f (a, ∞) ∩ E is measurable.
Now f −1 (a, ∞) ∩ E c ⊆ E c , so completeness implies f −1 (a, ∞) ∩ E c is measurable. Thus, f is
Lebesgue measurable so the claim holds.
So each fn is measurable, thus f 0 = lim fn almost everywhere is also Lebesgue measurable.

(b) Prove that if f is a continuous real valued function on the real line, then the set of points at
which f is differentiable is measurable.

Proof. Let

f (x + h) − f (x)
F (x, h) =
h
which is continuous on R × (R\{0}). If x is a differentiable point of f , then for all  > 0, there
exists a δ > 0 and some Y such that for all h with |h| < δ, we have |F (x, h) − Y | < . i.e.

41
7 AUGUST 2016 Kari Eifler

\[[ \
D = {x | differentiable point of f } = {x | |F (x, h) − Y | < }
 δ Y |h|<

For fixed , δ, Y, h then {x | |F (x, h) − Y | < } is open, thus Borel.

By taking only rational , δ, Y, h we have D Borel measurable.

Problem 3. (a) Let f be a real valued function on the unit interval [0, 1]. Prove that the set of
points at which f is discontinuous is a countable union of closed subsets.

Proof. f is continuous at p if for all n, there exists an open U containing p such that |f (x) −
f (y)| < 1/n for all x, y ∈ U . Fix n and let

[ [
Vn = {p s.t. there exists an appropriate U } = {appropriate U }
p

Hence, Vn is open. Then

\
{points where f is continuous} = Vn
n

Vnc where Vnc is closed.

S
So {points where f is discontinuous} = n

(b) Prove that there does not exist a real valued function on [0, 1] that is continuous at all rational
points but discontinuous at all irrational points.

Proof. By (a), the irrational poitns would be a countable union of closed subsets. Note that be-
cause any open set in [0, 1] contains a rational point, then if Qc[0,1] = n Fn where Fn is closed
S

and Fn◦ = ∅. Then

  !
[ [
[0, 1] = Q[0,1] ∪ Qc[0,1] = {q} ∪ Fn
q∈Q n

So [0, 1] is a countable union of nowhere dense sets. This contradicts Baire-Category Theorem.

Problem 4. Let (Ω, A, µ) be a finite measure space and let (fn ) be a sequence of measurable func-
tions on X that converges pointwise to zero. Prove that (fn ) converges in measure to zero. Show
that the converse is false for [0, 1] with Lebesgue measure.

Proof. Fix  > 0. To show µ({x | |fn (x)| > }) → 0, we need ∀m ∃Nm such that ∀n ≥ Nm ,
µ({x | fn (x)| > }) < 1/m.

By Egoroff’s Theorem, there exists some E ⊆ X with µ(E) < 1/m and fn ⇒ 0 uniformly on E c .

Thus, ∃Nm such that for n ≥ Nm |fn (x)| <  for all x ∈ E c so

1
µ({x | |fn (x)| > }) ≤ µ(E) < ∀n ≥ Nm
m

42
7 AUGUST 2016 Kari Eifler

Thus, µ({x | |fn (x)| > }) → 0.

Counterexample: Let f1 = χ[0,1] , f2 = χ[0,1/2] , f3 = χ[1/2,1] , . . . , fn = χ[j/2k ,(j+1)/2k ] for n =

2k + j, 0 ≤ j < 2k .

So fn does not approach 0 pointwise, but fn → 0 in L1 , hence in measure.

R
Problem 5. If f is Lebesgue integrable on the real line, prove that limh→0 R
|f (x+h)−f (x)|dx = 0.

Proof. Recall: the set Cc (R) of continuous, compactly supported functions is dense in L1 (R).

Fix  > 0 and find g ∈ Cc (R) with kf − gk1 < . Since g is continuous, limn |g(x + 1/n) − g(x)| = 0
ofr all x.

Since g is compactly supported, then there exists some compact K such that supp(g) ⊆ K.

So there exists a compact K 0 such that supp(g) ∪ supp(g(x + 1/n)) ⊆ K 0 for all n (this follows from
1/n ≥ 1 for all n since we can take K 0 = {k + x | k ∈ K, x ∈ [0, 1]}).

Dini’s theorem implies that |g(x + 1/n) − g(x)| ⇒ 0 so

Z Z
R|g(x + 1/n) − g(x)|dx = |g(x + 1/n) − g(x)|dx → 0
K0

So then

Z
|f (x + 1/n) − f (x)|dx
R
Z Z Z
≤ |f (x + 1/n) − g(x + 1/n)|dx + |g(x + 1/n) − g(x)|dx + |g(x) − f (x)|dx
R R R
Z
< 2 + |g(x + 1/n) − g(x)|dx → 2
R

R
Since it holds for all  > 0 then limn R
|f (x + 1/n) − f (x)|dx = 0.

43
7 AUGUST 2016 Kari Eifler

Problem 6. Prove or disprove that there exists a sequence (Pn ) of polynomials such that (Pn (t))
R1
converges to one for every t ∈ [0, 1] but 0 Pn (t)dt converges to two as n → ∞.

Proof. Consider


2
n x

 x ∈ [0, 1/n]
fn (x) = 2
−n x + 2n + 1 x ∈ [1/n, 2/n]

x ∈ [2/n, 1]

1

(that is, fn linearly connects the points (0, 1), (1/n, n + 1), (2/n, 1), (1, 1).)
R1
So fn (x) → 0 for all x ∈ [0, 1] but 0 fn (x)dx = 2.

Then by Stone-Weierstrass, we can find polynomials Pn such that kfn − Pn k∞ ≤ 2−n . Then ∀x

|Pn (x) − 1| ≤ |Pn (x) − fn (x)| + |fn (x) − 1| → 0

R1 R1
and 0
|fn (x) − Pn (x)|dx → 0 so 0
Pn (x)dx → 2.

Problem 7. Let (fn ) be a uniformly bounded sequence of continuous functions on [0, 1] that con-
verges pointwise to zero. Prove that 0 is in the norm closure in C[0, 1] of the convex hull of (fn )
(the norm is of course the sup norm on C[0, 1]).

Proof. By the Geometrical version of the Hahn-Banach,

weak k·k
conv{fn } = conv{fn }

weak
We just need to show that 0 ∈ conv{fn } . By Riesz-Representation Theorem, C[0, 1]∗ = M[0, 1].
For all µ ∈ M [0, 1],

Z  Z
 
fn dµ ≤ |fn |d|µ| → 0
 

 [0,1]  [0,1]

by Dominated Convergence Theorem. Thus, fn → 0 weakly.

Problem 8. Assume that X is a reflexive Banach space and φ is a continuous linear functional on
X. Prove that φ achieves its norm; that is, prove that there is a norm one vector x in X such that
φ(x) = kxk. Show by example that there is a continuous linear functional on the Banach space `1
that does not achieve its norm.

Proof. Recall: X reflexive ⇒ BX is weak-compact ⇒ BX is weak-sequentially compact.

There exists a sequence {xn } ⊆ BX such that φ(xn ) % kφk.

Choose a weakly-convergent subsequence {xnk } that converges to x ∈ BX . Then for all ϕ ∈ X ∗ ,

ϕ(xnk ) → ϕ(x).

44
7 AUGUST 2016 Kari Eifler

In particular,

kφk = lim φ(xn ) = lim φ(xnk ) = φ(x).

n k

Alternative Proof. For all φ ∈ X ∗ , by Hahn-Banach Separation Theorem, there exists some x∗∗ ∈
X ∗∗ such that kx∗∗ kX ∗∗ = 1 and x∗∗ (φ) = kφkX ∗ .

Since X is reflexive, ∃x ∈ X such that x̂ = x∗∗ so

kφkX ∗ = x∗∗ (φ) = x̂(φ) = φ(x).

Counterexample: Choose y = (1 − 1/n)n ∈ `∞ . Then ∀x ∈ `1 ,

∞    ∞   ∞
X 1  X 1 X
y(x) =  1− x(n) ≤ 1− |x(n)| < |x(n)| = kxk1 = 1 = kyk∞ .
 

n=1
n 
n=1
n n=1

Problem 9. Suppose that X is a non separable Banach space. Prove that there is an uncountable
subset A of the unit ball of X such that for all x 6= u in X, kx − yk > 0.9.

Proof. By transfinite induction, construct (xα )α < ω1 ⊆ BX where ω1 is the uncountable ordinal.

Given α < ω1 , let Uα := span{xβ | β < α} which is separable.

Since X is not separable, Uα ( X.

By Riesz-Lemma, there exists kxα k = 1 such that d(xα , Uα ) ≥ 1 −  (put  > 0.1).

So (xα ) satisfies kxα − xβ k ≥ 0.9 and is uncountable.

Alternative Proof if it were not restricted to BX . Fix r > 0. Zornicate over all subsets A ⊆ X such
that ∀x 6= y, kx − yk > r.

Find a maximal subset Ar ⊆ X as above. If Ar is uncountable, by scaling of r, we’re done.

Suppose not, so each Ar is countable. Enumerate as {xrn }n . By maximality, for all x ∈ X, ∀ > 0 if
r > 1/ then there exists n ∈ N such taht kx − xm
n k < r <  (i.e. d(x, Ar ) < r, ∀x ∈ X).
S
Let A = q∈Q Aq so A is a countable dense subset of X. Contradiction!

Therefore, there exists q ∈ Q such that Aq is uncountable. Consider A0 = { 0.9

q x | x ∈ Aq } so for all
0 0
x , y ∈ A,

0 0
0.9 0.9 0.9
kx − y k =
x− y = kx − yk > 0.9
q q q

Thus, there eixsts an uncountable A ⊆ X such that for all x, y ∈ A, kx − yk > 0.9.

45
8 JANUARY 2016 Kari Eifler

Problem 10. If A is a Borel subset of the line, then E = {(x, y) | x − y ∈ A} is a Borel subset of
the plane. If the Lebesgue measure of A is 0, then the Lebesgue measure of E is 0.

Proof. Define f (x, y) = x − y : R2 → R. This is continuous. Let

A := {S ⊆ R | f −1 (S) is a Borel set of R2 }

Then A is a σ-algebra (easy to check). If S is open, then f −1 (S) is open in R2 , thus Borel. So {open sets} ⊆
A and so the Borel algebra is a subset of A. In particular, A ∈ A.

Let E = f −1 (A) which is a Borel set of R2 . If m(A) = 0, let

E y = {x ∈ R | (x, y) ∈ E} = y + A

m(E y )dm(y) = 0.
R
This is a null set since m(y + A) = m(A) = 0. Thus, (m × m)(E) =

8 January 2016

Problem 1. Let E be a measurable subset of [0, 1]. Suppose there exists α ∈ (0, 1) such that

m(E ∩ J) ≥ α · m(J)

for all subintervals J of [0, 1]. Prove that m(E) = 1.

Proof. It’s easy to see that m(E) ≤ 1.

For any open U ⊆ [0, 1], write U = t∞

i=1 Ii where each Ii is an open interval. Then

∞
X ∞
X
m(E ∩ U ) = m(E ∩ Ii ) ≥ αm(Ii ) = αm(U ).
i=1 i=1

Assume m(E) < 1, so m(E c ) := a > 0. We may find some open U ⊇ E c such that m(U ∩ E) =
m(U \E c ) < . So

 > m(U ∩ E) ≥ αm(U ) ≥ αm(E c ) = αa > 0.

Letting  → 0, this leads to a contradiction.

Problem 2. Let f, g ∈ L1 ([0, 1]). Suppose

Z 1 Z 1
xn f (x)dx = xn g(x)dx
0 0

46
8 JANUARY 2016 Kari Eifler

for all integers n ≥ 0. Prove that f (x) = g(x) a.e.

Proof. See # 2 from January 2017.

Problem 3. Let f, g ∈ L1 ([0, 1]). Assume for all functions ϕ ∈ C ∞ [0, 1] with ϕ(0) = ϕ(1) we have

Z 1 Z 1
0
f (t)ϕ (t)dt = − g(t)ϕ(t)dt.
0 0

Show that f is absolutely continuous and f 0 = g a.e.

Proof. Fix x ∈ [0, 1] and construct hn via



 nt t ∈ [0, 1/n]


1 t ∈ [1/n, x]
hn (t) =
1 − n(t − x) t ∈ [x, x + 1/n]



0 t ∈ [x + 1/n, 1]


(i.e. hn (t) linearly connects the points (0, 0), (1/n, 1), (x, 1), (x + 1/n, 0), and (1, 0).

Since C ∞ [0, 1] is dense in k · k∞ , we may use this example rather than some ϕ ∈ C ∞ [0, 1] (i.e. pass
to the continuous case). Then

Z 1 Z 1/n Z x+1/n
f (t)h0n (t)dt = f (t)ndt + 0 + f (t)(−n)dt + 0 → f (0) − f (x)
0 0 x

where the limit follows from Lebesgue Differentiation Theorem. Also,

Z 1 Z 1/n Z x Z x+1/n Z x+1/n

g(t)hn (t)dt = nt g(t) dt + g(t)dt + g(t)dt − n (t − x)g(t) dt + 0
0 0 |{z} 1/n x x | {z }
→0 →0 as t→x
Z x+1/n
→0+ g(t)dt − 0
1/n

where the limit again follows from Lebesgue Differentiation Theorem. Taking the limit as n → ∞
Rx R x+1/n
on both sides, we get 0 g(t)dt = limn 1/n g(t)dt. So

Z 1 Z 1 Z x
f (0) − f (x) = lim f (t)h0n (t) = lim − g(t)hn (t)dt = − g(t)dt
n 0 n 0 0

Rx
Implying f (x) = f (0) + 0
g(t)dt. Then

47
8 JANUARY 2016 Kari Eifler

R x+h Rx R x+h
0 f (x + h) − f (x) f (0) + 0
g(t)dt − f (0) − 0
g(t)dt x
g(t)dt
f (x) = lim = lim = lim = g(x)
h→0 h h→0 h h→0 h

and f is absolutely continuous.

Problem 4. Let {gn } be a sequence of measureable functions on [0, 1] such that

(i) |gn (x)| ≤ C, for a.e. x ∈ [0, 1]

Ra
(ii) and limn→∞ 0 gn (x)dx = 0 for every a ∈ [0, 1].

Prove that for each f ∈ L1 ([0, 1]), we have

Z 1
lim f (x)gn (x)dx = 0.
n→∞ 0

Proof. Let S = span{χ[0,a] | a ∈ [0, 1]}. Then S is dense in the space of step functions in L1 . Step
function space is dense in L1 so S is dense in L1 . Then for every f ∈ L1 [0, 1] there exists a sequence
PKm (m)
hm = i=1 Ki χ[0,ai ] → f in L1 .

For a fixed m,

Z 1 Km Z ai
(m)
X
lim hm gn dx = Ki lim gn (x)dx = 0
n 0 n 0
i=1

where the second equality follows from (ii). For every  > 0, we can choose some m such that khm −
f k1 < .
R 
 1
For that m, choose soem N such that  0 hm gn dx <  for all n > N . Then


Z
 1  Z
  1

 Z
  1 


 f (x)gn (x)dx ≤  f (x) − hm (x) gn (x)dx +  hm (x)gn (x)dx
0 0 0
≤ ckf − hm k1 + 
< (c + 1)

R1
Thus, 0
f (x)gn (x)dx = 0.

Problem 5. (a) Let X be a normed vector space and Y be a closed linear subspace of X. Assume
Y is a proper subspace, that is, Y 6= X. Show that, for all 0 <  < 1, there is an element x ∈ X
such that kxk = 1 and

inf kx − yk > 1 − 
y∈Y

48
8 JANUARY 2016 Kari Eifler

Proof. Fix some x0 ∈ X\Y , denote inf y∈Y kx0 − yk = d > 0. Now for every  > 0 choose some
d
δ > 0 such that d+δ > 1 − .
x0 −y0
Choose y0 ∈ Y such that kx0 − y0 k < d + δ. Let x = kx0 −y0 k so kxk = 1 and

x0 − y0 1 d
kx0 − y 0 k >

inf kx − yk = inf − y =
kx0 − y0 k yinf > 1 − .
y∈Y y∈Y kx0 − y0 k
0 ∈Y d+δ

(b) Use part (a) to prove that, if X is an infinite dimensional normed vector space, then the unit
ball of X is not compact.

Proof. If we construct a sequence {xn } such that there are no convergent subsequences, we are
done.
Assume we have chosen {x1 , x2 , . . . , xn−1 } ⊆ BX . Let Y = span{x1 , x2 , . . . , xn−1 }. By part (a),
there exists some xn ∈ B such that kxn k = 1 and inf y∈Y kxn − yk > 1/2.
Then we have a sequence {xn } ⊆ BX such that kxn − xm k > 1/2 for all n 6= m so no convergent
subsequence may exist.

Problem 6. Let {fk } be a sequence of increasing functions on [0, 1]. Suppose

∞
X
fk (x)
k=1

converges for all x ∈ [0, 1]. Denote the limit function by f , that is,

∞
X
f (x) = fk (x).
k=1

Prove that

∞
X
0
f (x) = fk0 (x), a.e. x ∈ [0, 1].
k=1

PN
Proof. It’s easy to see f is increasing, so it’s differentiable almost everywhere. Let FN = n=1 fn
so FN → f for all x ∈ [0, 1]. Choose an increasing sequence Nk such taht 0 ≤ f (1) − FNk (1) ≤ 2−k .
Then

∞
X ∞

X
f (1) − FNk (1) ≤ 2−k = 1.
k=1 k=1

P∞
P∞ P∞
Now, let g(x) := k=1 f (x) − FN −k (x) = k=1 n=Nk +1 fn (x).
P∞
Since n=Nk +1 fn (x) is increasing as x increases, then g is increasing.

49
8 JANUARY 2016 Kari Eifler

So 0 ≤ g(x) ≤ g(1) ≤ 1 and g is differentiable almost everywhere. Now,

∞
1
1X


g(x + h) − g(x) = f (x + h) − FNk (x + h) − f (x) − FNk (x) .
h h
k=1

P∞ P∞
So since f 0 (x)−FNk (x) = n=Nk +1 fn (x) is increasing, g 0 (x) ≥ k=1 f 0 (x)−FN0 k (x) ≥ 0. Therefore,
P∞ 0 0 0 0
P∞ 0
k=1 f (x) − FNk (x) converges. So limk FNk (x) = f (x), implying f (x) = k=1 fk (x) almost
everywhere.

Problem 7. Suppose f, g : [a, b] → R are both continuous and of bounded variation. Show that the
set

{(f (t), g(t)) ∈ R2 | t ∈ [a, b]}

cannot cover the entire unit square [0, 1] × [0, 1].

Proof. Define r(t) = (f (t), g(t)). Since R2 is finite dimensional, `1 ∼ `2 . Since f and g have bounded
variation, so does r. Thus, we know that whenever a = x0 < x1 < x2 < . . . < xn = b we have
Pn
i=1 kr(xi ) − r(xi−1 )k2 < M .

Now suppose [0, 1] × [0, 1] can be covered. Divide [0, 1] × [0, 1] into n2 small squares with center zj
and the length of each edge is 1/n. Then choose tj such that r(tj ) = zj .

Now relabel/reorder the tj in increasing order so that s1 < s2 < . . . < sn2 . Then since the distance
between two centers is at least 1/n,

2 2
nX −1 nX −1
n2 − 1
kr(sj+1 ) − r(sj )k2 ≥ 1/n = → ∞.
j=1 j=1
n

This is a contradiction!

Problem 8. Prove the following two statements:

(a) Suppose f is a measurable function on [0, 1], then

kf kL∞ = lim kf kLp

p→∞

Proof. In [0, 1], by Hölder, we know that kf kp ≤ kf kq when p ≤ q. Also, kf kp ≤ kf k∞ for all p.
Therefore, kf kp %≤ kf k∞ and so limp kf kp ≤ kf k∞ .
On the other hand, for every  > 0, let E = {x | |f (x)| > kf k∞ − } and 0 < µ(E) ≤ 1 since

p
kf k∞ = esssup |f (x)|. Then kf kpp ≥ E |f |p > kf k∞ −  µ(E). Take p → ∞ so limp kf kp ≥
R

kf k∞ − , implying limp kf kp ≥ kf k∞ .
R R
(b) If fn ≥ 0 and fn → f in measure, then f ≤ lim inf fn .

50
8 JANUARY 2016 Kari Eifler

R R
Proof. Choose a subsequence {fnk } such that limk fnk = lim inf fn . Since fn → f in mea-
sure, fnk → f in measure, so there exists a further subsequence {fnk` } → f a.e. Then by Fa-
tou’s Lemma,
Z Z Z Z
f= lim fnk` ≤ lim inf fnk` = lim inf fnk = lim inf fn .
` ` k n

Problem 9. Suppose {fn } is a sequence of functions in L2 ([0, 1]) such that kfn kL2 ≤ 1. If f is
measurable and fn → f in measure, then

(a) f ∈ L2 ([0, 1]);

Proof. fn → f in measure implies {fnk } → f almost everwhere which implies |fnk |2 → |f |2

almost everywhere. By Fatou’s Lemma,

Z 1 Z 1
|f |2 dx ≤ lim |fnk |2 dx ≤ 1.
0 n 0

So f ∈ L2 .

(b) fn → f weakly in L2 ;

Proof. Let g ∈ L2 . We want to show that fn g → f g in L1 . Now, fn → f in measure, then

fn g → f g in measure and thus is Cauchy in measure.
Define Am,n = {x ∈ [0, 1] | |fn g(x) − fm g(x)| ≥ }. Then

Z 1 Z Z Z
|fn g−fm g|dx = |fn g(x)−fm g(x)|dx+ |fn g(x)−fm g(x)|dx ≤ |fn g|+|fm g|dx+.
0 Am,n [0,1]\Am,n Am,n

We know for all  > 0 there exists some δ > 0 such that µ(Am,n ) < δ,

Z Z !1/2 Z !1/2 Z !1/2

2 2 2
|fn g| ≤ |fn | dx |g| dx ≤ |g| < .
Am,n Am,n Am,n Am,n

since g ∈ L2 . Then since {fn g} is Cauchy in measure, there exists some N such that for all
R1
m, n > N , µ(Am,n ) < δ. Then 0 |fn g − fm g|dx < 3 implies {fn g} is Cauchy in L1 .
Therefore, there exists some h ∈ L1 such taht fn g → h in L1 .
We know fn g → f g in measure, so fnk g → f g almost everywhere. Also, ∀ > 0, ∃δ > 0 such
R
that A |fnk g| <  for all A such that µ(A) < δ.
Therefore, {fnk } is uniformly integrable. By Viteli Convergence Theorem, fnk g → f g in L1 .
Thus, h = f g so fn g → f g in L1 . So fn → f weakly.
Note: We could also have used the uniqueness of limit in the measure.

(c) fn → f with respect to norm in Lp for 1 ≤ p < 2.

51
8 JANUARY 2016 Kari Eifler

Proof. Define En = {x | |fn (x) − f (x)| ≥ }. From problem 8 on this exam, we know kfn kp ≤
kfn k2 ≤ 1 and kf kp ≤ kf k2 < ∞. Then
Z Z Z Z
p p p p−1
|fn − f | = |fn − f | + |fn − f | dx ≤ 2 |fn |p + |f |p + 
En c
En En

where the inequality follows from the fact that |a − b|p ≤ 2p−1 (|a|p + |b|p ).
Since fn → f in measure and m(En ) → 0 as n → ∞, so since f ∈ Lp then |f |p dx → 0 as
R
En
n → ∞.
For some A ⊆ [0, 1], we have

Z Z 1
2 2
|fn |p = |fn |p χA ≤ k|fn |p k2/p kχA k2/2−p = kfn kp2 m(A) 2−p ≤ m(A) 2−p .
A 0

|fn |p dx <  for

R
So similar to the previous case, we can take m(En ) small enough such that En
any fixed 1 ≤ p < 2.

There are a few hints in the qual

99
Problem 10. Suppose E is a measurable subset of [0, 1] with Lebesgue measure m(E) = 100 .

Show that there exists a number x ∈ [0, 1] such that for all r ∈ (0, 1),

r
m(E ∩ (x − r, x + r)) ≥ .
4

Hint: Use the Hardy-Littlewood maximal inequality

3
m({x ∈ R | M f (x) ≥ α}) ≤ kf k1
α

for all f ∈ L1 (R). Here M f denotes the Hardy-Littlewood Maximal function of f .

Proof. The Hardy-Littlewood Maximal function of χA is

Z x+r
1 1
M χA = sup χA (x)dx = sup m(A ∩ (x − r, x + r)).
r>0 2r x−r r>0 2r

Assume the result is not true. Then ∀x ∈ [0, 1], ∃rx ∈ (0, 1) such that m(E ∩ (x − xr , x + xr )) < r4x .
This happens if and only if 2r1x m(E ∩ (x − rx , x + rx )) < 1/8 which is equivalent to 2r1x m(E c ∩ (x −
rx , x + rx )) ≥ 87 .

Now set A = E c so M χA (x) ≥ 78 . However,

7
8 24 1 24
m {x ∈ [0, 1] | M χA (x) ≥ } ≤ 3 kχA k = = .
8 7 7 100 100

But we need it to be equal to 1. Contradiction!

52
9 AUGUST 2015 Kari Eifler

9 August 2015

Problem 1. Let f : R → R be a Borel measurable function. For each t ∈ R define

ft (x) = f (t + x), x ∈ R.

Prove that ft (x) is a Borel measurable function (in x) for each fixed t ∈ R.

Proof. We see that

ft−1 (−∞, a) = {x | f (x + t) ∈ (−∞, a)} = {x | x + t ∈ f −1 (−∞, a)} = f −1 ((−∞, a)) − t = B − t.

Since Tt (x) = x + t is continuous, then Tt−1 (B) = B − t is Borel.

Problem 2. Justify the statement that

1 1 1 1
(x − y) sin(xy) (x − y) sin(xy)
Z Z Z Z
dx dy = dy dx.
0 0 x2 + y 2 0 0 x2 + y 2

R 1 R 1  (x−y) sin(xy) 
Proof. We just need to show that 0 0  x2 +y 2  dxdy < ∞. But

1 1   Z π/2 Z √2   Z π/2 Z √2 √
 (x − y) sin(xy)   r cos θ − r sin θ 
Z Z
  dxdy =   |r|drdθ ≤ 2 drdθ = 2π < ∞.
 2
x +y 2   r 2 
0 0 0 0 0 0

So the function is in L1 and Fubini gives us the desired result.

Problem 3. Assume that (fn ) is a sequence in C[0, 1].

(a) Show that (fn ) converges weakly to 0 if and only if (fn ) is bounded in C[0, 1] and limn→∞ fn (t) =
0 for all t ∈ [0, 1].

Proof. ⇒) We know C[0, 1]∗ = M[0, 1]. Then fn → 0 weakly implies

R
fn dµ → 0 for all µ ∈
M[0, 1]. Choose µ = δt so
Z
fn dδt = fn (t) → 0 ∀t ∈ [0, 1]

(this follows from the fact that weak convergence implies uniformly bounded). Consider

χ : C[0, 1] → C[0, 1]∗∗ = M[0, 1]∗

χ(fn )(µ) = µ(fn )

53
9 AUGUST 2015 Kari Eifler

Since µ(fn ) → 0 then χ(fn )(µ) → 0 for all µ ∈ M[0, 1]. Since convergent sequences are
bounded, then supn |χ(fn )(µ)| ≤ M .
By the uniform boundedness theorem, supn kχ(fn )k < ∞. By isometry, kfn k = kχ(fn )k so
supn kfn k < ∞.
⇐) By Dominated Convergence Theorem, fn → 0 in L1 (µ). So therefore, | fn dµ| ≤ |fn |d|µ| →
R R

0. So fn → 0 weakly.

(b) Show that if (fn ) converges weakly in C[0, 1], then it converges in norm in Lp [0, 1] for all 1 ≤
p < ∞.

Proof. WLOG fn → 0 weakly. By (a) we know fn (t) → 0 and kfn k∞ is bounded. Thus,
|fn (t)|p → 0 pointwise and k|fn |k∞ is bounded.
By the Dominated Convergence Theorem, we have kfn kp → 0.

Problem 4. Let A be a Lebesgue null set in R. Prove that

B := {ex | x ∈ A}

is also a null set.

Proof. First, assume A ⊆ [0, 1]. Then f (x) = ex is Lipschitz-continuous (i.e. |f (x)−f (y)| ≤ M |x−y|
for some M ). Since m(A) = 0, we can find ∪∞ k=1 Bk where Bk are open intervals such that A ⊆
∞ ∞
∪k=1 Bk and m (∪k=1 Bk ) < . Then

∞
X
m(f (A)) ≤ m (f (∪∞
k=1 Bk )) ≤ M m(Bk ) < M .
k=1

P∞
So m(f (A)) = 0. Now we can write A = ∪∞
n=−∞ A ∩ [n, n + 1] so m(f (A)) = −∞ m(f (A ∩ [n, n +
1])) = 0.

Problem 5. (a) Define absolute continuity of a function f : R → R and of a function f : [a, b] →

R.

Proof. The function f : [a, b] → R is absolutely continuous if ∀ > 0, ∃δ > 0 such that when-
PN
ever a finite sequence of disjoint subintervals (xk , yk ) ⊆ I satisfies k=1 (yk − xk ) < δ then
P∞
k=1 |f (yk ) − f (xk )| < .

(b) Show that if f and g are absolutely continuous on [a, b], a, b ∈ R, a < b, then f · g is absolutely
continuous on [a, b].

Proof. Since f and g are continuous on [a, b], then they achieve a maximum so we can let Mf =
sup{f (x) | a ≤ x ≤ b} < ∞, Mg = sup{g(x) | a ≤ x ≤ b}.
Fix  > 0. Then there exists some δf , δg > 0 such that

54
9 AUGUST 2015 Kari Eifler

X X 
(yk − xk ) < δf ⇒ |f (yk ) − f (xk )| <
2Mg
X X 
(yk − xk ) < δg ⇒ |f (yk ) − f (xk )| <
2Mf
P
Choose finite and disjoint such that yk − xk < min(δf , δg ). Then

X X X
|f (yk )g(yk ) − f (xk )g(xk )| ≤ |f (yk )g(yk ) − f (yk )g(xk )| + |f (yk )g(xk ) − f (xk )g(xk )|
X X
≤ |f (yk )||g(yk ) − g(xk )| + |g(xk )||f (yk ) − f (xk )|
X X
≤ Mf |g(yk ) − g(xk )| + Mg |f (yk ) − f (xk )|
 
≤ Mf + Mg
2Mf 2Mg
=

This is what we wanted.

(c) Give an example to show that (b) is false if [a, b] is replaced by R.

Proof. Take f (x) = g(x) = x so f g = x2 . Then

|(x + δ)2 − x2 | = |x2 + 2δx + δ 2 − x2 | = |2δx + δ 2 | → ∞ as x → ∞.

So there does not exist any δ such that |f g(y) − f g(x)| <  (even for just one interval!)

Problem 6. Let X and Y be Banach spaces and T : X → Y be a one-to-one, bounded and linear
operator for which the range T (X) is closed in Y . Show that for each continuous linear functional φ
on X there is a continuous linear functional ψ on Y , so that φ = ψ ◦ T .

Proof. Since T : X → T (X) is bijective, by teh open mapping theorem, T −1 is bounded so φ ◦ T −1 ∈

T (X)∗ .

Then by the Hahn-Banach, there exists some ψ ∈ Y ∗ such that ψ(y) = (φ ◦ T −1 )(y) for all y ∈ Y .

For any x ∈ X, T (x) = y ∈ Y and we have

φ(x) = φ T −1 (T x) = ψ(T x) = (ψ ◦ T )(x).

Since this is true for all x ∈ X, φ = ψ ◦ T .

Problem 7. State the Open Mapping Theorema nd the Closed Graph Theorem for Banach spaces.
Derive the Open Mapping Theorem from the Closed Graph Theorem.

Proof. Assume T : X → Y is surjective, linear, and bounded. WLOG we want to show B(0, δ) ⊆
T (B(0, 1)) for some δ > 0. Define

55
9 AUGUST 2015 Kari Eifler

G : Y → X/ ker(T )
y 7→ [x] = x + ker(T ) where y = T x

Then G is well-defined, because T is surjective.

Claim: G is closed.

Assume yn → y in Y and G(yn ) → [x] in X/ ker(T ). WTS G(y) = [x] ⇔ T x = y.

We have T xn = yn so since [xn ] → [x] then k[xn ] − [x]k = inf z∈ker T kxn − x − zk → 0. Then take
(zn ) ⊆ ker(T ) such that kxn − x − zn k < 1/n. So xn − zn → x. Then

kT (xn − zn ) − T (x)k ≤ kT kkxn − x − zn k → 0.

Thus, T (xn − zn ) = T (xn ) → T (x). And also T (xn − zn ) = T (xn ) = yn → y. Together, these imply
T (x) = y. So G is closed, and the claim holds.

By the closed graph theorem, G is bounded so there exists some δ > 0 such that G(B(0, δ)) ⊆
B(0, 1) in X/ ker(T ). Now, let y ∈ B(0, δ) so then [x] = G(y) ∈ B(0, 1). Thus, if inf z∈ker(T ) kx−zk <
1, then there exists some z0 ∈ ker(T ) such that kx − z0 k < 1. This implies y = T x = T (x − z0 ) ∈
T (B(0, 1)) so B(0, δ) ⊆ T (B(0, 1)).

Problem 8. Let Y be a closed subspace of a Banach space X, with norm k · k. Let k · k1 be a norm
on Y which is equivalent to k · k, meaning that there is a C ≥ 1 so that

1
kyk1 ≤ kyk ≤ Ckyk1 for all y ∈ Y.
C

Let S be the set of all linear functions φ : X → R, so that

(i) |φ(y)| ≤ kyk1 for all y ∈ Y , and

(ii) |φ(x)| ≤ Ckxk for all x ∈ X.

Prove the following statements

(a) kxk2 := supφ∈S |φ(x)|, x ∈ X, defines a norm on X.

Proof. Easy to check.

(b) kyk2 = kyk1 for y ∈ Y .

Proof. Since |φ(y)| ≤ kyk1 then kyk2 ≤ kyk1 .

On the other hand, from the Hahn-Banach separation theorem, for all y 6= 0, there exists some
φ ∈ X ∗ such that kφk = 1 and φ(y) = kyk1 so kyk2 ≥ kyk1 .
To check that φ ∈ S: |φ(y)| = kyk1 and |φ(x)| ≤ kφkkxk = kxk.

56
9 AUGUST 2015 Kari Eifler

(c) The norms k · k2 and k · k are equivalent on X.

Proof. We just need to consider this on X\Y . For x ∈ X\Y , we have

kxk2 = sup |φ(x)| ≤ Ckxk.

φ∈S

Again by Hahn-Banach, for x̃ 6= 0, there exists some φ ∈ X ∗ such that φ(x̃) = kx̃k and kφk = 1.
1 1
Define ψ = Cφ so kψk = C.
Then to see ψ ∈ S:
1
• |ψ(x)| ≤ C kxk ≤ Ckxk for all x ∈ X
1
• |ψ(y)| ≤ C kyk ≤ kyk1 for all y ∈ Y

1 1
So ψ ∈ S and kx̃k ≥ |ψ(x̃)| = C kx̃k so C kxk ≤ kxk2 ≤ Ckxk.

Problem 9. Let f be increasing on [0, 1] and let

f (x + h) − f (x − h)
g(x) = lim sup , for 0 < x < 1.
h→0 2h

Prove that if A = {x ∈ (0, 1) | g(x) > 1} then

f (1) − f (0) ≥ m∗ (A).

Proof. For x ∈ A,

f (x + h) − f (x − h)
lim sup >1
h→0 2h

f (x+h)−f (x−h)
so for all  > 0, there exists some h > 0 such that 2h <  and 2h > 1 if and only if
f (x + h) − f (x − h) > 2h.

Let I = {(x − h, x + h) | x ∈ A, 2h < , (x − h, x + h) ⊆ [0, 1]}. Then I covers A in the sense

of Vitali. By Vitali’s Lemma, for every  > 0, there exists I1 , I2 , . . . , In disjoint from I such that
m∗ (A\ ∪ni=1 Ii ) < .

Since m∗ (A) = m∗ (A\ ∪ni=1 Ii ) + m∗ (∪ni=1 Ii ) for all Ii then write Ii = (xi − hi , xi + hi ) and
x1 − h1 < x1 + h1 < x2 − h2 < . . . < xn + hn . Then

n
X n
X
m∗ (A) <  + 2hi <  + |f (xi + hi ) − f (xi − hi )|.
i=1 i=1

Pn

Since f is increasing, i=1 f (xi + hi ) − f (xi − hi ) ≤ f (1) − f (0).

So m∗ (A) <  + f (1) − f (0) so m∗ (A) ≤ f (1) − f (0).

57
9 AUGUST 2015 Kari Eifler

Problem 10. (a) State a version of the Stone-Weierstrass Theorem.

Proof. See textbook.

(b) Let A be a uniformly dense subspace of C[0, 1] and let

 Z x 
B= F (x) | F (x) = f (t)dt, 0 ≤ x ≤ 1, f ∈ A .
0

Prove that B is uniformly dense in

C0 [0, 1] := {g ∈ C[0, 1] | g(0) = 0}.

Rx
Proof. Define B 0 = {F (x) | F (x) = 0 f (t)dt, 0 ≤ x ≤ 1, f ∈ C[0, 1]}. First show B is dense in
B0.
Rx Rx
For every F ∈ B 0 , G = B, F (x) = 0 f (t)dt and G(x) = 0 g(t)dt. Then

Z 1
kF (x) − G(x)k∞ ≤ |f − g|dt ≤ kf − gk∞ .
0

Since A is dense in C[0, 1], then kf − gk∞ <  so kF − Gk∞ < . So B is indeed dense in B 0 .
Then we will show B 0 is an algebra (in order to use Stone-Weierstrass). Let F, G ∈ B 0 so

Z x Z x Z x Z x Z x Z x Z t
F (x)G(x) = f (t)dt g(s)ds = f (t)g(s)dsdt = F (t)g(t)+G(t)f (t)dt = f (s)g(t)+g(s)f (t)dsdt.
0 0 0 0 0 0 0

Since F (t)g(t) + G(t)f (t) ∈ C[0, 1] then F G ∈ B 0 .

R1
Also, x = 0 1dt ∈ B 0 so B 0 separates points.
By Stone-Weierstrass, B 0 is dense in C0 [0, 1] since any function F ∈ B 0 , F (0) = 0. So B is dense
in C0 [0, 1].

(c) Prove that the span of {sin(nx) | n ∈ N} is dense in C0 [0, 1].

Rx
Proof. sin(nx) = 0
n cos(nx)dt. From part (b), it is sufficient to show

A = span{n cos(nx)} = span{cos(nt)}

is dense in C[0, 1]. A is an algebra:

1


• cos(nt) cos(mt) = 2 cos((m + n)t) + cos((m − n)t) ∈ A
• cos(t) separates [0, 1] (since 1 < π/2) so A is dense in C[0, 1].

58
10 JANUARY 2015 Kari Eifler

10 January 2015

Problem 1. Let f ∈ L1 (R). If

Z b
f (x)dx = 0
a

for all rational numbers a < b, prove that f (x) = 0 for almost all x ∈ R.

Proof. Let E + := {x | f (x) > 0}. Assume m(E + ) > 0 (the same argument will show E − := {x |
f (x) < 0} has measure zero).

There exists some n such that E + ∩ [n, n + 1] has positive measure. Consider F closed in R and
F ⊆ E + ∩[n, n+1] with m(F ) > 0. Then [n, n+1]\F is open in [n, n+1]. Thus, [n, n+1]\F = ∪∞ n=1 In
for In being disjoint open intervals in [n, n + 1].

For all In = (an , bn ), there exists some (ani )i , (bni )i ⊆ Q such that ani → an and bni → bn . Since

Z bn Z
f (x)dx = f (x)χ[an ,bn ] dx lim f (x)χ[ani ,bni ] = f (x)χ[an ,bn ]
an R i

R bn
then |f (x)χ[ani ,bni ] ≤ |f (x)| ∈ L1 so by Dominated Convergence Theorem, an
f (x)dx = 0.
R R n+1 R
Since F
f (x)dx > 0, by condition we know n
f (x)dx = 0 for all n. So then [n,n+1]\F
f (x)dx <
0.
R
So there exists some Im = (am , bm ) such that Im
f (x)dx < 0. Contradiction!

Proof #2 as in Problem 3 from August 2014, not restricted to rationals with f ∈ L1 .

For every open U , write U = ∪∞

n=1 (an , bn ) for disjoint open intervals, so

Z ∞ Z
X bn
f (x)dx = f (x)dx = 0.
U n=1 an

For every compact K ⊆ (a, b) then (a, b)\K is open in R and

Z Z b Z
f (x)dx = f (x)dx − f (x)dx
K a (a,b)\K

(because each is finite). Suppose E + = {x | f (x) > 0} has positive measure. Since E + = ∪n En
where En = {x | f (x) > 1/n} so there must exist some n such that m(En ) > 0.

By inner regularity, there exists some K ⊆ En with m(K) > 0. Then

Z Z
1 1
0= f (x)dx > dx = m(K) > 0
K K n n

59
10 JANUARY 2015 Kari Eifler

Contradiction!

Problem 2. Let {gn }∞ 1

n=1 and g be in L (R) and satisfy

lim kgn − gk1 = 0.

n→∞

Prove that there is a subsequence of {gn }∞

n=1 that converges pointwise almost everywhere to g.

Proof. Step 1: Suppose gn → g in L1 . Let En, = {x | |fn (x) − f (x)| ≥ }. Then

Z Z
|fn − f | ≥ |fn − f | ≥ µ(En, )
En,

1
R
So then µ(En, ) ≤  |fn − f | → 0.

Step 2: We will show that if gn → g in measure, then there exists a subsequence that converges to g
pointwise almost everywhere.

Suppose for every  > 0, µ({x | |fn (x) − f (x)| ≥ }) → 0. Choose a subsequence {gnk } such that if

Ej = {x | |gnj (x) − gnj+1 (x)| > 2−j }

P∞
satisfies µ(Ej ) < 2−j . Let Fk = ∪∞
j=k Ej so µ(Fk ) ≤ j=k 2−j ≤ 21−k . Let F = ∩k Fk so µ(F ) = 0.

For x ∈
/ Fk and for i ≥ j ≥ k then

i−1
X i−1
X
|gni (x) − gnj (x)| ≤ |gn` (x) − gn`+1 (x)| ≤ 2` ≤ 2−j → 0 as k → ∞.
`=j `=j

So gnk is pointwise Cauchy on x ∈

/ F , so let

(
lim gnk (x) x ∈
/F
f (x) =
0 otherwise

So gnk → f almost everywhere and gn → f in measure since

µ({x | |gn (x) − f (x)| ≥ }) ≤ µ({x | |gn (x) − gn` (x)| ≥ /2}) + µ({x | |gn` (x) − f (x)| ≥ })
| {z } | {z }
→0 →0

and

µ({x | |f (x) − g(x)| ≥ }) ≤ µ({x | |f (x) − gn (x)| ≥ /2} + µ({x | |gn (x) − g(x)| ≥ /2})
| {z } | {z }
→0 →0

60
10 JANUARY 2015 Kari Eifler

so f = g almost everywhere. Thus, {gnk } converges to g almost everywhere.

Problem 3. Let (X, M, µ) be a measure space with µ(X) < ∞. Let N ⊆ M be a σ-algebra. If f ≥
0 is M-measurable and µ-integrable then prove that there exists an N -measurable and µ-integrable
function g ≥ 0 so that

Z Z
gdµ = f dµ, E ∈ N.
E E

R
Proof. Define ν(E) = E
f dµ a finite positive measure on (X, N , µ). Then since µ(E) = 0, ν(E) = 0
so ν  µ.

Then by Radon-Nikodym Theorem, there exists some g : X → [0, ∞) and N -measurable and g ∈
L1 (µ) such that ν(E) = E gdµ. Then
R

Z Z
ν(E) = f dµ = gdµ ∀E ∈ N .
E E

Note: Folland doesn’t mention positive but there are other versions that give positive.

Problem 4. (a) State the closed graph theorem.

Proof. See wikipedia.

(b) If H is a Hilbert space and T : H → H is a linear operator satisfying

hT x, yi = hx, T yi, x, y ∈ H,

then prove that T is bounded.

Proof. Let xn → x and T xn → y. We want to show T x = y.

hT xn , zi = hxn , T zi → hx, T zi = hT x, zi.

| {z }
→hy,zi

So hT x − y, zi = 0 for all z ∈ H so then T x − y = 0 and so T x = y.

Problem 5. Let f, g ∈ L1 (R). Prove that h ∈ L1 (R), where h(x) is defined by

Z
h(x) = f (y)g(x − y)dy
R

whenever this integral is finite.

R R R 
Proof. We want to show that R
|h(x)|dx =  f (y)g(x − y)dy  dx < ∞. Indeed,
R R

61
10 JANUARY 2015 Kari Eifler

Z Z  Z Z
 
 f (y)g(x − y)dy  dx ≤ |f (y)||g(x − y)|dydx
 
R R R R
Z Z 
= |f (y)| |g(x − y)|dx dy
ZR R

= |f (y)|kgk1 dy
R
Z
= kgk1 |f (y)|dy
R
= kgk1 kf k1 < ∞.

Problem 6. Let f, g ∈ C[0, 1] with f (x) < g(x) for all x ∈ [0, 1].

(a) Prove that there is a polynomial p(x) so that

f (x) < p(x) < g(x), x ∈ [0, 1].

Proof. Let  = inf{g(x) − f (x) | x ∈ [0, 1]}. Since h(x) = g(x) − f (x) > 0 on [0, 1] and attains a
minimum on the compact set [0, 1] then the inf is attained and thus is positive. So  > 0.
By Stone-Weierstrass,
  polynomials are dense in C[0, 1] so there exists a polynomial p(x) such
that p − f +g < /2. Then

2
∞

f (x) + g(x)  1
1
p(x) < + ≤ f (x) + g(x) + (g(x) − f (x)) = (2g(x)) = g(x).
2 2 2 2

f (x) + g(x)  1
1
p(x) > − > f (x) + g(x) − (g(x) − f (x)) = (2f (x)) = f (x).
2 2 2 2
So f (x) < p(x) < g(x).
Remark: Let M = max{g(x) − f (x)} then
      
 f +g   f +g  M
|g(x) − f (x)| ≤ g(x) −
 (x) + 
  (x) − p(x) < + .
2 2 2

Alternative Proof. Let M := minx∈[0,1] g(x) − f (x). By Stone-Weierstrass, there exists some p̃(x)
polynomial such that kp̃(x) − g(x)k∞ < M/3. Let p(x) = p̃(x) − M/2. Then

M −M M M
g(x) − p(x) = g(x) − p̃(x) + > + = > 0.
2 3 2 6

M
−M M M
p(x)−f (x) = p(x)−g(x)+g(x)−f (x) = p̃(x)−g(x)− + g(x)−f (x) > − +M = > 0.
2 3 2 6

So f (x) < p(x) < g(x).

62
10 JANUARY 2015 Kari Eifler

(b) Prove that there is an increasing sequence of polynomial {pn (x)}∞

n=1 so that

f (x) < pn (x) < g(x), x ∈ [0, 1],

and pn → g uniformly on [0, 1].

g+g−1
1

1
Proof. Find p1 such taht g − 2 < p1 < g with p1 − 2
<
4. Then |g(x) − p1 (x)| <
1 1 1
4 + 4 = 2.
 
Recursively find a polynomial pn such that pn−1 < pn < g with pn − g+p2n−1 < 1
2n−1 ,

implying

Mn−1 1 1 1 1
|g(x) − pn (x)| < + n+1 = n+1 + n+1 = n .
2 2 2 2 2
1 1
So Mn < 2n . Then for every  > 0 choose N such that 2N
< , so for all n > N

1
|pn (x) − g(x)| < < ∀x ∈ [0, 1].
2N
Alternative Proof. From part (a) we can find f (x) < p1 (x) < g(x). Repeating, we can find
p1 (x) < p2 (x) < g(x). By requiring n instead of M , in kp̃(x) − g(x)k∞ <  and letting n → 0,
we get

n n 5
kpn (x) − g(x)k∞ ≤ kpn (x) − p˜n (x)k∞ + kp˜n (x) − gk∞ < + = n → 0.
2 3 6

Problem 7. If f ∈ L2 (R), g ∈ L3 (R), and h ∈ L6 (R) then prove that the product f gh is in L1 (R).

1/p
1p
|f |kp dx kp = kf kpkp .

Proof. Note: |f |k p = |f |kp dx
R R
=

Then it follows that

1/2
1/3
1/3

kf ghk1 ≤ kf k2 kghk2 ≤ kf k2 k|g|2 |h|2 k1 ≤ kf k2 k|g|2 kp=3/2 k|h|2 kq=3 ≤ kf k2 kgk3 khk6 < ∞.

1 1 2 1
Where we use p = 3/2, q = 3 so p + q = 3 + 3 = 1.

Problem 8. (a) A point y in a metric space Y is isolated if the set {y} is both open and closed in
Y . Prove taht y ∈ Y is isolated if and only if the complement {y}C is not dense in Y .

Proof. ⇒) If y is isolated, then {y} is open. But {y}c ∩ {y} = ∅ so {y}c is not dense.
⇐) Trivially, {y} is closed since we’re in a metric space. Suppose {y}c is not dense in Y . Then
there exists an open U 6= ∅ such that U ∩ {y}c = ∅ (since A is dense in Y ⇔ for all open U 6= ∅,
U ∩ A 6= ∅).
But if U ∩ {y}c = ∅ then U ⊆ {y}cc = {y} so U = {y} is open.

63
10 JANUARY 2015 Kari Eifler

(b) Let X be a countable nonempty complete metric space. Prove that the set of isolated points is
dense in X.

Proof. Let Y ⊆ X be the set of isolated points. Let X\Y = {zj }∞ n

j=1 (or {zj }j=1 ).

Since the singleton {zk } is not an isolated point, by (a) we know {zk }c is dense in X, so {zk }c =
X. So each {zk }c is open and dense in X.
By Baire-Category, Y = ∩∞ c n c
j=1 {zj } (or ∩j=1 {zj } ) is also dense in Y .

Problem 9. Suppose that f ∈ Lp (R) for all p ∈ (1, 2) and that supp∈(1,2) kf kp < ∞. Prove that
f ∈ L2 (R) and that

lim kf kp = kf k2 .
p→2−

Proof. Let A = {x | |f (x)| ≥ 1}, B = {x | |f (x)| < 1}. Then by Monotone Convergence Theorem,
|f |p dx % A |f |2 dx.
R R
A

Let pn ↑ 2. WLOG assume p1 = 3/2. We know on B, |f |p ≤ |f |3/2 ∈ L1 (B). By Dominated

Convergence Theorem, B |f |p dx → B |f |2 dx which implies R |f |p dx → R |f |2 dx.
R R R R

p/2
Therefore, kf kpp → kf k22 so kf kp → kf k2 as p → 2.
p/2−1
Also, since M = supp∈(1,2) kf kp < ∞, then kf kp ≤ M p/2−1 for all p ∈ (1, 2).
p/2
Then M p/2−1 → 1 as p → 2 which implies kf kp − kf kp → 0 as p → 2.

So then, kf kp → kf k2 as p → 2 and kf k2 < ∞ since M < ∞.

Problem 10. Let (X, k · k) be a normed vector space with a subspace Y and let k · k1 be another
norm on Y that satisfies

1
kyk1 ≤ kyk ≤ Kkyk1 , y ∈ Y,
K

where K > 1 is a fixed constant. Define S to be the set of linear functionals φ : X → R satisfying

(i) |φ(y)| ≤ kyk1 , y ∈ Y ,

(ii) |φ(x)| ≤ Kkxk, x ∈ X.

Prove the following statements:

(a) kxk2 := sup{|φ(x)| | φ ∈ S} defines a norm on X.

Proof. See August 2015

(b) For y ∈ Y , kyk1 = kyk2 .

Proof. See August 2015

64
11 AUGUST 2014 Kari Eifler

(c) The norms k · k and k · k2 are equivalent on X.

Proof. See August 2015

11 August 2014

Problem 1. For n ∈ N, let fn : [0, 1] → R be continuous, and assume that for every x ∈ [0, 1]
the sequence (fn (x)) is decreasing. Suppose that fn converges pointwise to a continuous function f .
Show that this convergence is uniform.

Proof. WLOG: by replacing fn by fn (x) − f (x), these are still continuous and decreasing pointwise.
So we want to prove fn ⇒ 0.

This is precisely Dini’s Theorem (aka freebie question).

Fix  > 0 and let Un = fn−1 ((−1, )) = {x ∈ X | gn (x) < } which is open. Then for all x, fn (x) & 0
so there exists N such that for all n ≥ N , |fn (x)| <  which implies x ∈ Un .

So [0, 1] = ∪n Un . By compactness of [0, 1], there exists a finite subcover Un1 , Un2 , . . . , Unk for n1 <
n2 < . . . < nk but since Un ⊆ Un+1 then Un1 ⊆ Un2 ⊆ . . . ⊆ Unk .
−1
Therefore, [0, 1] ⊆ Unk =: UN so for all x ∈ [0, 1], then x ∈ fN ((−1, )) ⇔ |fN (x)| < .

Decreasing fn implies that for all n ≥ N , |fn (x)| <  for all x ∈ [0, 1].

Problem 2. Let f ∈ L1 (0, ∞). For x > 0, define

Z ∞
g(x) = f (t)e−tx dt.
0

Prove that g(x) is differentiable for x > 0 with derivative

Z ∞
0
g (x) = −tf (t)e−tx dt.
0

Proof. Since

Z ∞ Z x Z ∞ Z x  Z ∞ Z ∞ Z ∞
−ty −ty
|tf (t)e |dydt = t|f (t)| e dy dt = −e−tx |f (t)|dt+ |f (t)|dt ≤ 2 |f (t)|dt < ∞.
0 0 0 0 0 0 0
| {z }
≤1

R∞Rx R∞
By Fubini, h(x) = 0 0
−tf (t)e−ty dtdy = 0
f (t)e−tx dt + c.

So h(x) = g(x) + c.

From the definition of h, we know h0 (x) = g 0 (x) and thus g(x) is differentiable. And h is differen-
tiable since it’s absolutely continuous.

65
11 AUGUST 2014 Kari Eifler

Problem 3. Let f : R → R be a Lebesgue integrable function such that

Z b
f (x)dx = 0 for every a < b.
a

Show that f (x) = 0 for almost every x ∈ R.

Proof. See question 1 from January 2015.

Problem 4. Let f be Lebesgue measurable on [0, 1] with f (x) > 0 a.e. Suppose (Ek ) is a sequence
R
of measurable sets in [0, 1] with the property that Ek f (x)dx → 0 as k → ∞.

Prove that m(Ek ) → 0 as k → ∞.

Proof. Let Fm = {x | f (x) ≥ 1/m} so Fn ⊆ Fn+1 .

Since f (x) > 0 almost everywhere, then

m (∪∞
n=1 Fn ) = lim m(Fn ) = 1.
n

Fix  > 0, so there exists N such that m(Fnc ) < /2 for n ≥ N . Now,

Z Z
1
m(Ek ∩ FN ) ≤ f (x)dx ≤ f (x)dx → 0 as k → ∞.
N Ek ∩FN Ek

So there exists some K such that m(Ek ∩ FN ) < /2 for all k ≥ K. Thus,

 
m(Ek ) = m(EK ∩ FN ) + m(Ek ∩ FNc ) < + = ∀k ≥ K.
2 2

Problem 5. Let (fn ) be a sequence of continuous functions on [0, 1] such that for each x ∈ [0, 1]
there is an N = Nx so that

fn (x) ≥ 0 for all n ≥ Nx .

Show that there is an open nonempty set U ⊂ [0, 1] and an N ∈ N, so that fn (x) ≥ 0 for all n ≥ N
and all x ∈ U .

Proof. Let

∞
\
En := {x | fm (x) ≥ 0 ∀m ≥ n} = {x | fn (x) ≥ 0}
n=m

66
11 AUGUST 2014 Kari Eifler

so En is closed and En ⊇ En+1 . For all x ∈ [0, 1] there exists N = Nx such taht fm (x) ≥ 0 for all
m ≥ N . Thus, x ∈ EN .

Then, [0, 1] = ∪∞
n=1 En . Since [0, 1] is compact, by Baire-Category we know there exists N such that
◦ ◦
EN 6= ∅ (i.e. EN 6= ∅).
◦
Let U = EN be open, non-empty so for all x ∈ U , fn (x) ≥ 0 for all n ≥ N .

Problem 6. (a) Define the w∗ -topology on the dual X ∗ of a Banach space X.

Proof. See wikipedia!

(b) Let X be an infinite dimensional Banach space. What is the w∗ -closure of

SX ∗ = {x∗ ∈ X ∗ | kx∗ k = 1}?

(as usual, prove your answer.)

w∗
Proof. Claim: SX ∗ = BX ∗ .
We know for any x1 , x2 , . . . , xn ∈ X, there exists some x∗0 6= 0 such that x∗0 (xi ) = 0. Indeed,
if this were not true then otherwise, x∗0 (xi ) 6= 0 for some i, let ϕ : X ∗ → Rn be ϕ(x∗ ) =
(x∗ (x1 ), . . . , x∗ (xn )) then ϕ is injective so dim(X ∗ ) ≤ dim(Rn ) = n. Contradiction, so true.
Now for any x∗ ∈ BX ∗ , consider it’s neighborhood (open under the w∗ -neighborhood)

V = ∩ni=1 {y ∗ ∈ X ∗ | |x̂i (x∗ − y ∗ )| = |x∗ (xi ) − y ∗ (xi )| < }

for each {xi }ni=1 choose such an x∗0 6= 0 from the claim.
Consider the line {x∗ + tx∗0 | t ∈ R} in X ∗ .
Since for any x̂i ,

x̂i (x∗ + tx∗0 − x∗ ) = tx̂i (x∗0 ) = tx∗0 (xi ) = 0 < .

Then {x∗ + tx∗0 | t ∈ R} ⊆ V . Since kx∗ + tx∗0 k is continuous about t, then we can find t0 such
that kx∗ + t0 x∗0 k = 1 ⇒ V ∩ SX ∗ 6= ∅.
Since any neighborhood of x∗ contains a neighborhood of the form V as above (i.e. these V ’s
w∗
are a neighborhood basis) then BX ∗ ⊆ SX ∗ .
On the other hand, for any x∗0
∈ BX ∗ , by Hahn-Banach separation Theorem, we know there
exists x ∈ X and c ∈ R such that x∗ (x) < c < x∗0 (x) for all x∗ ∈ BX ∗ .
Then for all {x∗n } ⊆ BX ∗ , x∗n (x) ≤ c < x∗0 (x). Therefore, x∗0 isn’t an accumulation point of BX ∗
w∗ w∗ w∗ w∗
which implies BX ∗ = BX ∗ . Thus, SX ∗ ⊆ BX ∗ = BX ∗ so BX ∗ = SX ∗ .

Problem 7. (a) State the Riesz Representation Theorem for the dual L∗p (µ) of Lp (µ), 1 ≤ p < ∞.

Proof. See Wikipedia!

67
11 AUGUST 2014 Kari Eifler

(b) Let µ be a finite measure on the measurable space (Ω, Σ). Prove the following part of the above
theorem:
If F ∈ L∗p (µ), then there exists an h ∈ L1 (µ) so that
Z
F (χA ) = hdµ for all A ∈ Σ.
A

Proof. Let ν(A) = F (χA ). The goal is to show ν is a σ-finite signed measure.

(a) ν(∅) = F (χ∅ ) = F (0) = 0

(b) Let {Ei } be disjoint, let E = ∪∞
i=1 Ei . Then

n n
!
X X
ν(E) − ν(Ei ) = F (χE ) − F χi
i=1 i=1

Xn
≤ kF kL∗p χE − χi



i=1 p

∞
X
≤ kF kL∗p
χi


i=n+1 p

1/p
= kF kL∗p µ ∪∞
i=n+1 Ei →0 as n → ∞.
P∞
Therefore, ν(E) = i=1 ν(Ei ).

When µ(A) = 0, then

ν(A) = F (χA ) ≤ kF kL∗p kχA kp = kF kL∗p µ(A)1/p = 0.

So ν  µ.
Then from the Radon-Nikodyn Theorem, there exists some h ∈ L1 (µ) such that ν(A) =
R
A
hdµ.
So
Z
F (χA ) = ν(A) = hdµ.
A

Problem 8. Assume that (xn ) is a weakly converging sequence in a Hilbert space H. Show that
there is a subsequence (yn ) of (xn ) so that

n
1X
yj
n j=1

converges in norm.

68
11 AUGUST 2014 Kari Eifler

Proof. WLOG xn → 0 weakly (hxn , yi → h0, yi for all y ∈ H) and we know kxn k is bounded,
supn kxn k ≤ C. For n > m,

2 * +
m n m n m n
1 X 1 X 1 X 1X 1 X 1X

m yj − yj =
yj − yj , yj − yj
j=1 n j=1 m j=1 n j=1 m j=1 n j=1
*  m n   m n
+
1 1 X 1 X 1 1 X 1 X
= − yj − yj , − yj − yj
m n j=1 n j=m+1 m n j=1 n j=m+1
2 * +   2
m m n
n
 2 X   X 2 X
1 1  1 1 1 X  1
≤ − y j
+ 2 − y j , yj
+ yj
.
m n j=1 
 m n j=1 n j=m+1 
 n j=m+1

Pm
Now by induction, we can choose yj such that |hyj , n=1 yn i| < 2−j for all m ≤ j − 1. Pick y1
randomly.

Since hxn , yi → h0, yi for all y ∈ H, then we can find y2 such that hy2 , y1 i < 2−2 .

Similarly, find y3 such that hy3 , y1 + y2 i < 2−3 and hy3 , y1 i < 2−3 , etc. Then

2 * +
m m m m−1 m m m
X X X X X X X
2 −j

yj
=
yj , yj = hy m , ym i+hy m , y j i+. . .+hy1 , y1 i ≤ ky j k + 2 < kyj k2 +2.
j=1 j=1 j=1 j=1 j=1 j=1 j=1

Therefore,

2  
 2 X m m
1 1 1  X 1
kyj k2 + 2 < 2 (mc + 2) → 0 as m → ∞.

−
≤ m2
yj
m n j=1 j=1
m

P 2
1 2 n


Similar argument holds for j=m+1 yj .

n

Finally,

* + * +
  m
X n   Xn m
 1 1 1 X  1 1 1 X
 − yj , yj  ≤ − yj , yk
 m n j=1 n j=m+1  n m n j=m+1
 k=1
  Xn
1 1 1
≤ − 2−(m+1)
n m n j=m+1
 
1 1 1
≤ − →0
n m n

69
11 AUGUST 2014 Kari Eifler

P 2
1 m 1
Pn
So then m j=1 yj − yj → 0 as n, m → ∞ so it’s Cauchy and therefore converges.

n j=1

Problem 9. Show that a linear functional φ on a Banach space X is continuous if and only if {x ∈
X | φ(2x) = 3} is norm closed.

Proof. ⇒) A = {x | φ(2x) = 3)} = {x | 2x ∈ φ−1 ({3})}. Let ψ(x) = φ(2x) so A = ψ −1 (φ−1 ({3})).

⇐) We want to show ker(φ) is closed. Note that {x ∈ X | φ(2x) = 3} = {x ∈ X | φ(x) = 3/2}.

Pick some a ∈ X such that φ(a) = 3/2. Then clearly

a + ker(φ) ⊆ {x ∈ X | φ(x) = 3/2}

and if φ(x) = 3/2 then φ(x − a) = 0 so x = a + (x − a) ∈ a + ker(φ).

Thus, a + ker(φ) = {x ∈ X | φ(2x) = 3}. Therefore ker(φ) = {x ∈ X | φ(2x) = 3} − a which is closed.

Then

φ0 : X/ ker(φ) → R
x + ker(φ) 7→ φ(x)

is an isomorphism. Let π : X → X/ ker φ which is also continuous, so φ = φ0 ◦ π is continuous.

Problem 10. Let C 1 [0, 1] be the space of functions f ∈ C[0, 1] such taht f 0 exists and is continuous
in [0, 1]. The space C 1 [0, 1] is given the supremum norm. Define T : C 1 [0, 1] → C[0, 1] by T f =
f 0 for f ∈ C 1 [0, 1]. Show that T has a closed graph and that T is not bounded. Decide if C 1 [0, 1]
(together with the supremum norm) is a Banach space or not. (Explain your answer).

Proof. Let fn → f and T fn → fn0 → g in k · k∞ .

Z x Z x
fn (x) = fn0 (t)dt + fn (0) f (x) = f 0 (t)dt + f (0)
0 0

Rx
Since fn → f then fn (0) → f (0). Let G = 0
g(t)dt + f (0). Then

Z x
kf − Gk ≤ kf − fn k + kfn − Gk ≤ kf − fn k + kfn − gk∞ ≤ kf − fn k + kfn − gk∞ → 0.
0

So f 0 = g meaning T has a closed graph.

To see T is not bounded, consider fn = xn so kfn k∞ = 1 but kT fn k = knxn−1 k∞ = n → ∞.

Thus, by the closed graph theorem, C 1 [0, 1] is not a Banach space.

70
12 JANUARY 2014 Kari Eifler

12 January 2014

Problem 1. Let (X, M, µ) be a non atomic masure space with µ(X) > 0. Show that there is a
measurable f : X → [0, ∞), for which

Z
f (x)dµ(x) = ∞.

Proof. Take X = E1 ⊇ E2 ⊇ E3 ⊇ . . . such that µ(E1 ) > µ(E2 ) > . . . > 0. Define

(
µ(En \En+1 )−1 if x ∈ En \En+1
f (x) =
0 if x ∈ ∩∞
n=1 En

R P∞
Then f (x)dx = n=1 1 = ∞.

Problem 2. Assume that µ is a finite measure on Rn . Prove that there is a closed set A ⊂ Rn with
the property that for each closed B ( A it follows that µ(A\B) 6= 0.

Proof. Since Rn is a locally compact Hausdorff space and µ is finite, µ is Radon (and regular). If
µ(Rn ) = a (so V 6= Rn ) then we can set

n ao
Mn := U | U is open and µ(U ) <
n

and V := ∪∞ c
n=1 {U | U ∈ Mn } so V is open. Let A = V is closed.

For any B ( A, A\A ∩ B c . Assume µ(A\B) = 0 then

µ(A\B) = inf{µ(U ) | A\B ⊆ U, U open} = 0.

Then there exists U ⊆ V such that A\B ⊆ U . Then A\B ∩ U ⊆ A ∩ U ⊆ A ∩ V = ∅, so A\B = ∅.

Contradiction!

Problem 3. For a nonnegative function f ∈ L1 ([0, 1]), prove that

Z 1 p
n
lim f (x)dx = m({x | f (x) > 0}).
n→∞ 0

Proof. Let

E1 = {x | f (x) ≥ 1}
E2 = {x | 0 < f (x) < 1}
E3 = {x | f (x) = 0}

71
12 JANUARY 2014 Kari Eifler

Then

Z 1 Z Z Z
f (x)1/n dx = f (x)1/n dx + f (x)1/n dx + f (x)1/n dx
0 E1 E2 E3

For the first integral on E1 , limn f (x)1/n dx = 1 and |f (x)1/n | ≤ |f (x)| ∈ L1 , so by DCT, f (x)1/n dx =
R
R E1
E1
dx = m(E1 ).

For the second integral on E2 , limn f (x)1/n = 1 and |f (x)1/n | ≤ 1 ∈ L1 so again by DCT, f (x)1/n dx =
R
R E2
E2
dx = m(E2 ). Therefore,

Z 1
f (x)1/n dx = m(E1 ) + m(E2 ) = m({x | f (x) > 0}).
0

Problem 4. Let f be Lebesgue integrable on (0, 1). For 0 < x < 1 define

Z 1
g(x) = t−1 f (t)dt.
x

Prove that g is Lebesgue integrable on (0, 1) and that

Z 1 Z 1
g(x)dx = f (x)dx.
0 0

Proof. Notice that

Z 1 Z t Z 1
−1
t f (t)dtdx = |f (t)|dt < ∞
0 0 0

since f ∈ L1 (0, 1) so then by Fubini, we have that

Z 1 Z 1 Z 1 Z t
t−1 f (t)dtdx = t−1 f (t)dxdt.
0 x 0 0

So we have

Z 1 Z 1 Z 1 Z 1 Z t Z 1
−1 −1
g(x)dx = t f (t)dtdx = t f (t)dxdt = f (t)dt.
0 0 x 0 0 0

Problem 5. Assume that ν and µ are two finite measures on a measurable space (X, M). Prove
that

72
12 JANUARY 2014 Kari Eifler

ν << µ ⇔ lim (ν − nµ)+ = 0.

n→∞

Proof. ⇐) Let µ(E) = 0. Then for all  > 0, there exists some N such that for all n ≥ N ,

 > (ν − nµ)+ (E) ≥ (ν − nµ)(E) = ν(E) − nµ(E) ≥ ν(E).

Letting  approach 0, we have that ν(E) = 0 so that ν << µ.

Problem 6. Let (pn ) be a sequence of polynomials which converges uniformly on [0, 1] to some
function f , and assume that f is not a polynomial. Prove the limn→∞ deg(pn ) = ∞, where deg(p)
denotes the degree of a polynomial p.

Proof. Assume to the contrary and consider the space P = span{1, x, x2 , . . . , xm } with (pn ) ⊆ P.
Since {1, x, . . . , xm } are basis elements and P is finite dimensional, then any two norms are equiva-
P∞
lent on P and so if P = k=0 ak xk , we can consider the two norms defined by

kP k1 := sup |ak | kP k2 := sup |P (x)|

x∈[0,1]

Pm
Since kpnk − pn` k2 → 0, then kpnk − pn` k1 → 0 so {ank } is Cauchy. Hense, P = `=0 a` x` where
a` = limk a`nk is a polynomial of degree at most m and pn converges uniformly to p. So therefore,
p = f . Contradiction!

Alternative proof. Assume not, so there exists some M such that for all N ∈ N, there exists some
n ≥ N such that deg(pn ) ≤ M . For N = 1, find n1 such that deg(pn1 ) ≤ M . For N = n1 , find n2
such that deg(pn2 ) ≤ M , etc.

Get a subsequence {pnk } such that deg(pnk ) ≤ M .

Since pn converges to f uniformly on [0, 1] then pnk converges to f uniformly on [0, 1].
Pm
We write pni := j=1 aij xj , then

m
X m
X X
f = lim aij xj = lim aij xj = xj where bj = lim aij .
i→∞ i→∞ i→∞
j=i j=1 j=1

To see limi→∞ aij exists: let X = Pp | p polynomial with degree ≤ M } is a finite dimensional
subspace of C[0, 1] hense closed so f ∈ X so f is a polynomial.

Problem 7. Let (fn ) be sequence of non zero bounded linear functionals on a Banach space X.
Show that there is an x ∈ X so that fn (x) 6= 0, for all n ∈ N.

Proof. Let En = {x | fn (x)} which is closed in X. Assume the result is not true, so for every x ∈ X,
there exists some n such that fn (x) = 0 implies x ∈ En , that is, X = ∪n En .

73
12 JANUARY 2014 Kari Eifler

Since X is a Banach space, then by Baire Category Theorem, there exists some n such that ∅ 6=
En◦ = En◦ .

Thus, there exists some r > 0, x ∈ X such that B(r, x) ⊆ En .Then for all y ∈ X,

y
r + x ∈ x + B(r, 0) = B(r, x)
kyk

y r
so then if fn (r kyk + x) = 0, then kyk fn (y) = −fn (x) = 0 so fn (y) = 0 so fn = 0.

Thus, by Baire Category, ∪Cn 6= X. Contradiction!

Problem 8. Assume that T : `1 → `2 is bounded, linear and one-to-one. Prove that T (`1 ) is not
closed in `2 .

Proof. If T (`1 ) is closed, then T (`1 ) is a hilbert space. Since T : `1 → T (`1 ) is bijective, by the open
mapping theorem. T is open and T −1 is bounded so T is an isomorphism. Then `1 ∼ = T (`1 ). But `1
is not reflexive and T (`1 ) is reflexive, so contradiction.

Alternative Proof. T (`1 ) is closed, hence complete. So T : `1 → T (`1 ) is bijective, so by the open
mapping theorem, T is an open map so T (`1 ) is both open and closed in `2 . Hence, T (`1 ) ∼ = `2 so
1 ∼ 2
then ` = ` . Contradiction!

Problem 9. For a uniformly bounded sequence (fn ) in C[0, 1] (i.e. supn∈N supξ∈[0,1] |fn (ξ)| < ∞)
show that fn converges weakly to 0 ⇔ limn→∞ fn (ξ) = 0 for all ξ ∈ [0, 1].

Is the equivalence true if we do not assume that (fn ) is uniformly bounded, explain?

Proof. This question is the same as 3 from August 2015.

⇒) C([0, 1])∗ = M[0, 1] for all ξ ∈ [0, 1], δξ ∈ M[0, 1]. So 0 = limn fn dδξ = limn fn (ξ) so then
R

limn fn (ξ) = 0 (note that this does not require uniformly boundedness!)
R
⇐) Fix µ ∈ M[0, 1], we want to show that fn dµ → 0. Since |fn (x)| ≤ M for all x and all n, then
R
by dominated convergence theorem, fn dµ → 0.

Finally, consider hn given by connecting (0, 0), (1/n, n), (2/n, 0) and (1, 0). So hn (ξ) → for all xı ∈
R
[0, 1]. But by taking Lebesgue measure, hn (x)dµ(x) = 1 so fn 9 0 weakly.

Problem 10. Assume that f is measurable and non negative function on [0, 1]2 and that 1 ≤ r <
p < ∞. Show that

p/r !1/p r/p !1/r

Z 1 Z 1 Z 1 Z 1
f r (x, y)dy dx ≤ f p (x, y)dx dy .
0 0 0 0

Hint: Let s = p/r, let 1 < s0 < ∞ be the conjugate of s and let

Z 1
F : [0, 1] → R+
0, x 7→ f r (x, y)dy.
0

74
13 AUGUST 2013 Kari Eifler

Then consider for an appropriate function h ∈ Ls0 [0, 1] the product hF .

R1 0
Proof. Let F (x) := 0
f r (x, y)dy. Let h ∈ Ls [0, 1] with khks0 = 1 and h ≥ 0. Then by Tonelli (since
F h ≥ 0), we have

Z 1 Z 1 Z 1
F (x)h(x)dx = f r (x, y)h(x)dydx
0 0 0
Z 1 Z 1
= f r (x, y)h(x)dxdy
0 0
Z 1
≤ kf r (·, y)ks khks0 dy
0
Z 1 Z 1 r/p
= f p (x, y)dx dy
0 0

R1 R 1 R 1 r/p
So then 0
F (x)h(x)dx ≤ 0 0
f p (x, y)dx dy for all khks0 = 1, h ≥ 0.

Notice that F ≥ 0 so when kh̃ks0 = 1, we have

Z 1 Z 1
sup F (x)h(x)dx = sup F (x)h(x)dx
kh̃ks0 =1 0 khks0 =1,h≥0 0

Therefore,

Z 1 Z 1 Z 1 p/r !r/p Z 1 Z 1 r/p

r
sup F (x)h(x)dx = kF ks = f (x, y)dy ≤ f p (x, y)dx dy
khks0 =1,h≥0 0 0 0 0 0

So then,

p/r !1/p r/p !1/r

Z 1 Z 1 Z 1 Z 1
r p
f (x, y)dy dx = f (x, y)dx dy .
0 0 0 0

13 August 2013

Problem 1. Let 1 ≤ p ≤ ∞ and let f ∈ Lp (R). For t ∈ R, let ft (x) = f (x − t) and consider
the mapping G : R → Lp (R) given by G(t) = ft . The space Lp (R) is equipped with the usual norm
topology.

(a) Show that G is continuous if 1 ≤ p < ∞.

75
13 AUGUST 2013 Kari Eifler

Proof. Since Cc∞ (R) is dense in Lp (R) for 1 ≤ p < ∞, we can choose g ∈ Cc∞ (R) such that
kg − f kp < . Then

kftn − ft kp ≤ kftn − gtn kp + kgtn − gt kp + kgt − ft kp

It’s easy to see kftn − gtn kp and kgt − ft kp are small since kg − f kp < . For kgtn − gt kp , then

Z 1/p Z 1/p
kgtn − gt kp = |g(x − tn ) − g(x − t)|p dx ≤ p = µ(A)
R A

where for each fixed tn → t, we can find a bounded set A ⊆ R such that ∪n supp gtn ∪ supp gt ⊆
A.

(b) Find an f for which the mapping G is not continuous when p = ∞ (and justify your answer)

Proof. We will take f = χ[0,1] , so

kχ[0,1] (tn ) − χ[0,1] (t)k∞ = kχ[tn ,tn +1) − χ[t,t+1) k∞ = 1 ∀n

although tn → t, we have k · k∞ 9 0.

(c) Let 1 ≤ p, q ≤ ∞ be conjugate exponents (i.e. satisfying p1 + 1q = 1). Let f ∈ Lp (R) and
g ∈ Lq (R) and show that their convolution h = f ∗ g is continuous. Recall
Z ∞
h(t) = f (x)g(t − x)dx.
−∞

Proof. Letting gt (x) = g(t − x), then we have (by Hölder)

Z
|h(t) − h(tn )| ≤ |f (x)||g(t − x) − g(tn − x)|dx ≤ kf kp kgt − gtn kq
R

This goes to zero when 1 < p, q < ∞ from part (a).

Also notice that
Z ∞ Z ∞
h(t) = f (x)g(t − x)dx = f (t − y)g(y)dy = g ∗ f.
−∞ −∞

So when p = 1, q = ∞ the same is true.

Problem 2. (a) For f ∈ CR ([0, 1]), show that f ≥ 0 if and only if kλ − f ku ≤ λ for all λ ≥ kf ku ,
where k · ku denotes the uniform (supremum) norm.

Proof. ⇒) Assume there exists some λ ≥ kf ku such that kλ − f k∞ > λ. Then there exists some
x ∈ [0, 1] such that |λ − f (x)| = λ − f (x) > λ so then f (x) < 0. Contradiction!
⇐) If there exists some x such that f (x) < 0, then if λ ≥ kf k∞ so λ > 0. Then kλ − f k∞ ≥
λ − f (x) > λ. Contradiction!

76
13 AUGUST 2013 Kari Eifler

(b) Suppose E ⊆ CR ([0, 1]) is a closed subspace containing the constant function 1. For φ ∈ E ∗ ,
we define φ ≥ 0 to mean φ(f ) ≥ 0 whenever f ∈ E and f ≥ 0. Show φ ≥ 0 if and only if
kφk = φ(1).

Proof. ⇒) We have

kφk = sup |φ(f )| ≥ φ(1).

kf ku =1

Also for kf ku = 1, we have 1 − f ≥ 0 so φ(1 − f ) ≥ 0 implies φ(1) ≥ φ(f ). Moreover, φ(1 + f ) =

φ(1) + φ(f ) ≥ 0 and so φ(1) ≥ −φ(f ) so then φ(1) ≥ |φ(f )|. Therefore, φ(1) = kφk
⇐) φ(1) = kφk ≥ |φ(f )| for all kf ku ≤ 1. Assume there exists some f ≥ 0 but φ(f ) < 0.
By rescaling we can assume kf ku < 1, so then
 
1−f 1
kφk ≥ φ = (φ(1) − φ(f )) ≥ φ(1) − φ(f ) > φ(1) = kφk
k1 − f k k1 − f k

which contradicts!

(c) If φ ∈ E ∗ and φ ≥ 0, show that there is a bounded linear functional ψ on CR ([0, 1]) so that
ψ ≥ 0 and the restriction of ψ to E is φ.

Proof. By Hahn-Banach, there exists some ψ which is an extension of φ such that kψk = kφk =
φ(1) = ψ(1). So ψ ≥ 0 follows from (b).

Problem 3. (a) Let µ and λ be mutually singular complex measures defined on the same measur-
able space (X, M) and let ν = µ + λ. Show |ν| = |µ| + |λ|.

Proof. Let X = E t F be a disjoint union such that λ(F ) = mu(E) = 0. Let P2 t N2 = E be

the Hahn-decomposition for λ. Let P3 t N3 = F be the Hahn-decomposition for µ.
Then P1 = P2 t P3 , N1 = N2 t N3 will be the Hahn-decomposition for ν = λ + µ. Then

ν + (A) = ν(A ∩ P2 ) + ν(A ∩ P3 ) = λ(A ∩ P2 ) + µ(A ∩ P3 )

ν − (A) = ν(A ∩ N2 ) + ν(A ∩ N3 ) = λ(A ∩ N2 ) + µ(A ∩ N3 )

So then

|ν|(A) = λ(A ∩ P2 ) + µ(A ∩ P3 ) + λ(A ∩ N2 ) + µ(A ∩ N3 ) = |µ|(A) + |λ|(A)

Therefore, |ν| = |µ| + |λ|.

(b) Construct a nonzero, atomless Borel measure on [0, 1] that is mutually singular with respect to
Lebesgue measure.

Proof. here. Maybe Cantor-Lebesgue?

77
13 AUGUST 2013 Kari Eifler

Problem 4. Let (fn )∞ n=1 be a sequence of continuous functions on [0, 1] and suppose that for all
x ∈ [0, 1], fn (x) is eventually nonnegative. Show that there is an open interval I ⊆ [0, 1] such that
for all n large enough, fn is nonnegative everywhere on I.

Proof. Let UN = ∩∞ −1
n=N fn [0, ∞) = {x | fn (x) ≥ 0 ∀n ≥ N }. This is closed. For every x ∈ [0, 1],
fn (x) is eventually non-negative so [0, 1] = ∪N UN .
◦ ◦
By Baire-Category, there exists some N such that ∅ =
6 UN = UN . So there exists some open I ⊆
◦
UN ⊆ [0, 1] and then for all n ≥ N , fn ≥ 0 on I.

Problem 5. Let µ be a nonatomic signed measure on a measure space (X, Ω), with µ(X) = 1.
Show that there is a measureable subset E ⊂ X with µ(E) = 1/2.

Proof. First notice that for every  > 0, there exists some E ⊆ X with µ(E) < . This is because we
can recursively divide our set into two non-trivial sets and chose the smaller one.

Therefore, for all n, we can find a set En such taht 0 < µ(En ) < 2−n . Let S = {E ⊆ X | µ(E) ≤ 21 }
ordered by inclusion.

Zorn’s Lemma implies that there exists a maximal element E. If µ(E) < 21 then we can find some
F ⊆ E c with 0 < µ(F ) < 21 − µ(E) but then µ(F ∪ E) ≤ µ(F ) + µ(E) ≤ 21 which contradicts
maximality.

Problem 6. Compute

Z ∞
n sin(x/n)
lim dx
n→∞ 0 x(1 + x2 )

and justify your computation.

Proof. Let fn (x) = nx(1+x

sin(x/n)
2 ) . Recall that limt→0
sin t
t = 1, so limn n sin(x/n)
x(1+x2 ) = 1
1+x2 and since
| sin(x/n)| ≤ x/n for x, n positive,

   
 n sin(x/n)   n x 1  1 1
 x(1 + x2 )   x n 1 + x2  = 1 + x2 ∈ L [0, ∞).
 ≤ 

Then by DCT,

Z ∞ Z ∞
n sin(x/n) n sin(x/n) π
lim = lim = arctan |∞
0 = .
n 0 x(1 + x2 ) 0 n 2
x(1 + x ) 4

Problem 7. Prove or disprove: for every real-valued continuous function f on [0, 1] such that f (0) =
0 and every  > 0, there is a real polynomial p having only odd powers of x, i.e. p is of the form

p(x) = a1 x + a3 x3 + a5 x5 + · · · + a2n+1 x2n+1 ,

78
13 AUGUST 2013 Kari Eifler

such that supx∈[0,1] |f (x) − p(x)| < .

Proof. Let

A = { polynomial with even power}

so A is an algebra that separates points. Stone-Weierstrass implies that A is dense in C[0, 1]. We
can let

Z x
B = F (x) | F (x) = f (t)dt 0 ≤ x ≤ 1, f ∈ A} = { all polynomials with odd powers}
0

From problem 10 of August 2015, B = C0 [0, 1].

Problem 8. Let f ∈ L1loc (R).

(a) What (by definition) are the Hardy-Littlewood maximal function Hf and the Lebesgue set Lf of
f?

Proof. Z
1
Hf (x) = sup |f (y)|dy .
r>0 m(B(r, x)) B(r,x)
| {z }
=:Ar f (x)

( R )
B(r,x)
|f (y) − f (x)|dy
Lf = x | lim =0
r→0+ m(B(r, x))

(b) State the Hardy-Littlewood Maximal Theorem.

Proof. kHf (x)kp ≤ kf kp .

(c) In each case, either construct concretely an example of f with the required property, or explain
why no such example exists (you may use theorems from Folland about the Lebesgue set, if you
state them).

(i) Lf = R
(ii) the complement of Lf is uncountable
(iii) Lf ⊆ (−∞, 0] ∪ [1, ∞).

Proof. here

Problem 9. Let X be a separable Banach space, let {xn | n ≥ 1} be a countable, dense subset of the
unit ball of X and let B be the closed unit ball in the dual Banach space X ∗ of X. For φ, ψ ∈ B, let

79
13 AUGUST 2013 Kari Eifler

∞
X
d(φ, ψ) = 2−n |φ(xn ) − ψ(xn )|.
n=1

Show that d is a metric on B whose topology agrees with the weak*-topology of X ∗ restricted to B.

Proof. We first check that d is a metric on B:

• d(φ, ψ) ≥ 0 clear
If d(φ, ψ) = 0 then φ = ψ on {xn } so φ = ψ by continuity / density

• triangle inequality follows as well

the weak*-topology is {z ∗ ∈ X ∗ | |(x∗ − z ∗ )(x)| < } for fixed x∗ ∈ X ∗ , x ∈ X,  > 0.

For fixed x∗ ∈ X ∗ consider the -ball in the metric d which is

∞
X
∗ ∗ ∗
ψ ∈ X | d(x , ψ) < } = {ψ ∈ X | 2−n |x∗ (xn ) − ψ(xn )| < }
n=1

We want to show that |(x ∗ −ψ)(x)| < 0 for any x ∈ BX and some 0 > 0. Since xn is dense in BX
then there exists a xnk → x. Then

|(x∗ − ψ)(x)| ≤ |(x∗ − ψ)(x − xnk )| + |(x∗ − ψ)(xnk )| < 0

P∞
On the other hand, if ψ is in a weak* neighborhood of x∗, we want to show n=1 2−n |x∗ (xn ) −
ψ(xn )| < . Let |(ψ − x∗ )(xn )| < 0 for all n, then

∞
X ∞
X
2−n |x∗ (xn ) − ψ(xn )| < 2−n  = .
n=1 n=1

Alternative Proof. We first check that d is a metric on B:

• d(φ, ψ) ≥ 0 clear
If d(φ, ψ) = 0 then φ = ψ on {xn } so φ = ψ by continuity / density

• triangle inequality follows as well

To see that the topologies agree:

Consider B(r, ϕ) under the metric. We need to show it contains an open U under the weak*-topology.
P∞
Say d(φk , ψ) → 0. Then n=1 2−n |φk (xn ) − ψ(xn )| → 0. So under the weak* topology, we need to
show for all x ∈ BX , |φk (x) − ψ(x)| → 0.

Indeed, this follows by density of {xn }. For large k, kφk k = supn |φk (xn )| ≤ M and |φk (xn )| ∼
|ψ(xn )|.

80
14 JANUARY 2013 Kari Eifler

Then for every  > 0, there exists some n such that kxn − xk <  so

|φk (x) − ψ(x)| ≤ |φk (x) − φk (xn )| + |φk (xn ) − ψ(xn )| + |ψ(xn ) − ψ(x)|

P∞
If |φk (x) − ψ(x)| → 0 for all x, then for all , choose N such that n=N 2−n < , so d(φk , ψ) =
P∞ −n
n=1 2 |φk (xn ) − ψ(xn )|.

Problem 10. Let T : X → Y be a linear map between Banach spaces that is surjective and satisfies
kT xk ≥ kxk for some  > 0 and all x ∈ X. Show that T is bounded.

Proof. Is Γ(T ) = {(x, T x) | x ∈ X} closed in X × Y ?

If xn → x and T xn → y, we want to show y = T x. T is surjective, so y = T x0 . Then for all ˜ > 0,

there exists N such that for all n ≥ N ,

˜ > kT xn − T x0 k ≥ kxn − x0 k.

So xn → x0 , and T xn → T x0 .

The closed graph theorem implies T is bounded.

14 January 2013

Problem 1. Let f be a Lebesgue integrable, real-valued function on (0, 1) and for x ∈ (0, 1) define

Z 1
g(x) = t−1 f (t)dt.
x

R1 R1
Show that g is Lebesgue integrable on (0, 1) and that 0
g(x)dx = 0
f (x)dx.

Proof. See January 2014, # 4

Problem 2. Let fn ∈ C[0, 1]. Show that fn → 0 weakly if and only if the sequence (kfn k)∞
n=1 is
bounded and fn converges pointwise to 0.

Proof. See August 2015, # 3

Problem 3. Let (X, µ) be a measure space with 0 < µ(X) ≤ 1 and let f : X → R be measurable.
State the definition of kf kp for p ∈ [1, ∞]. Show that kf kp is a monotone increasing function of
p ∈ [1, ∞) and that limp→∞ kf kp = kf k∞ .

Proof. See January 2016, # 8

81
14 JANUARY 2013 Kari Eifler

Problem 4. (a) Is there a signed Borel measure µ on [0, 1] such that

Z 1
p0 (0) = p(x)dµ(x)
0

for all real polynomials p of degree at most 19?

Proof. We first define the linear functional I(p) = p0 (0).

Write P = span{1, x, x2 , . . . , x19 }, which is a finite dimensional space. Thus, all norms are
equivalent. We take, in particular, the norms k · km = maxi=1,...,19 |ai | and k · k∞ . Then there
must exist some C such that if kpk∞ = 1 then kpkm ≤ C so |a1 | ≤ C which implies that I is
bounded.
By Hahn-Banach, there exists some I˜ ∈ C[0, 1]∗ such that I(p) ˜ = I(p) for all p ∈ P. By Riesz,
0
R1
there exists some µ such that Ẽ(p) = p (0) = 0 p(x)dµ.

(b) Is there a signed Borel measure µ on [0, 1] such that

Z 1
0
p (0) = p(x)dµ(x)
0

for all real polynomials p?

R1
Proof. Suppose there did exist such a measure µ on [0, 1]. Then since µ([0, 1]) = 0 1dµ = 0, we
R1
have that |µ|([0, 1]) < ∞. Therefore, the mapping T : p →
7 0
p(x)dµ(x) can extend continuously
to C[0, 1].
Consider fn (x) defined by nx for x ∈ [1/n, 2/n], fn (x) = 1 for x ∈ [2/n, 1] and smooth on the
interval [1/n, 2/n] but bounded above by 2 (it’s always possible to construct such an fn ). Then

Z 1
fn0 (0) = fn (x)dµ ≤ kfn k∞ kµk ≤ kµk < ∞
0

But limn |fn0 (0)| = limn = ∞. Contradiction!

Problem 5. Let F be the set of all real-valued functions on [0, 1] of the form

1
f (t) =
Πnj=1 (t − cj )

for natural numbers n and for real numbers cj ∈ / [0, 1]. Prove or disprove: for all continuous, real-
valued functions g and h on [0, 1] such that g(t) < h(t) for all t ∈ [0, 1], there is a function a ∈
span F such that g(t) < a(t) < h(t) for all t ∈ [0, 1].

1
Proof. Let A = span F. It’s easy to see this is an algebra since cj ∈
/ [0, 1]. Also t+1 separates
points, so Stone-Weierstrass theorem implies A = C[0, 1].

Let M = mint∈[0,1] |h(t) − g(t)|, so we can choose some a ∈ A such that a − h+g < M6 . Then

2
∞
−M h+g M
6 <a− 2 < 6 and since h − g ≥ M , then

82
14 JANUARY 2013 Kari Eifler

h h h g M h+g M M M M
h−a= −a+ ≥ −a+ + = −a+ > − = >0
2 2 2 2 2 2 2 2 6 3

g+g g+h M −M M M
a−g =a− ≥a− + > + = >0
2 2 2 6 2 3

So then g < a < h.

Problem 6. Let k : [0, 1] × [0, 1] → R be continuous and let 1 < p < ∞. For f ∈ Lp [0, 1], let T f be
the function on [0, 1] defined by

Z 1
(T f )(x) = k(x, y)f (y)dy.
0

Show that T f is a continuous function on [0, 1] and that the image under T of the unit ball in Lp [0, 1]
has compact closure in C[0, 1].

Proof. Note that

p
|T f (x) − T f (y)| ≤ int10 |k(x, z) − k(y, z)||f (z)|dz ≤ kk(x, ·) − k(y, ·)kq kf kp for q =
p−1

Since k is continuous on [0, 1]2 , then for every  > 0 there exists some δ > 0 such that if |x − y| < δ,
then

Z 1 Z 1
kk(x, ·) − k(y, ·)kqq = q
|k(x, z) − k(y, z)| dz < p dz = p .
0 0

Therefore, T f is continuous.

Now consider F = {T f | kf kp ≤ 1} ⊆ C[0, 1]. We’ll use Arzela-Ascoli:

• equicontinuous
follows from above

• pointwise bounded

Z 1 1/q
|T f (x)| ≤ kK(x, ·)kq kf kp ≤ kK(x, ·)kq ≤ M q dz =M
0

so it’s actually uniformly bounded

Therefore, by Arzela-Ascoli, F is compact in C[0, 1].

83
14 JANUARY 2013 Kari Eifler

Problem 7. (a) Define the total variation of a function f : [0, 1] → R and absolute continuity of f .

Proof. here

(b) Suppose f : [0, 1] → R is absolutely continuous and defines g ∈ C[0, 1] by

Z 1
g(x) = f (xy)dy.
0

Show that g is absolutely continuous.

Pn
Proof. Since f is absolutely continuous, there exists some δ 0 > 0 such that i=1 |bi − ai | < δ 0
Pn
implies i=1 |f (bi ) − f (ai )| < . Fix some y ∈ [0, 1] so that

n
X n
X
|bi y − ai y| ≤ |bi − ai | < δ 0
i=1 i=1
Pn
This implies then that i=1 |f (bi y) − f (ai y)| < . Therefore,

n
X Z n
1X Z 1
|g(bi ) − g(ai )| ≤ |f (bi ) − f (ai )| ≤ dx = .
i=1 0 i=1 0

So g is absolutely continuous.

Problem 8. (a) State the definition of absolute continuity, v << µ, for positive measures µ and
ν, and state the Radon-Nikodym Theorem, (or the Lebesgue-Radon-NIkodym Theorem, if you
prefer.)

Proof. here

(b) Suppose that we have ν1 << µ1 and ν2 << µ2 for positive measures νi and µi on measurable
spaces (Xi , Mi ) for i = 1, 2. Show that we have ν1 × ν2 << µ1 × µ2 , and

d(ν1 × ν2 ) dν1 dν2

(x, y) = (x) (y).
d(µ1 × µ2 ) dµ1 dµ2

Proof. Assume E ∈ M1 ⊗ M2 and µ1 × µ2 (E) = 0. Define

Ex = {y ∈ X2 | (x, y) ∈ E} E y = {x ∈ X | (x, y) ∈ E}

Then Ex ∈ M1 and E y ∈ M2 for all x ∈ X1 , y ∈ X2 . Since µ1 and µ2 are positive, then

0 = (µ1 × µ2 )(E) = µ1 (E y )dµ2 (y) then µ1 (E y ) = 0 µ2 -almsot everywhere and so then
R

ν1 (E y ) = 0 µ2 -almost everywhere.
Thus, µ2 ({y ∈ X2 | ν1 (E y ) > 0}) = 0 so then ν2 ({y ∈ X2 | ν1 (E y ) > 0}) = 0. Thus, ν1 (E y ) = 0
for ν2 -almost everywhere and therefore, (ν1 × ν2 )(E) = ν1 (E y )dν2 (y) = 0.
R

Thus, ν1 × ν2 << µ1 × µ2 . By Radon-Nikodym theorem,

84
14 JANUARY 2013 Kari Eifler

d(ν1 × ν2 )
Z
ν1 × ν2 (E) = (x, y)d(µ1 × µ2 ) for E ∈ M1 ⊗ M2
E d(µ1 × µ2 )

Since ν1 << µ1 , by Proposition 3.9(a) in Folland,

Z
(ν1 × ν2 )(E) = ν2 (Ex )dν1 (x)
Z
dν1
= ν2 (Ex ) (x)dµ1 (x)
dµ1
Z Z 
dν2 dν1
= (y)dµ2 (y) (x)dµ1 (x)
Ex dµ 2 dµ 1
Z
dν2 dν1
= (y) (x)d(µ1 × µ2 )(x, y)
E dµ 2 dµ 1

By the uniqueness of Radon-Nikodym derivative, we have

d(ν1 × ν2 ) dν1 dν2

(x, y) = (x) (y).
d(µ1 × µ2 ) dµ1 dµ2

Problem 9. (a) Let E be a nonzero Banach space and show that for every x ∈ E, there is φ ∈ E ∗
such that kφk = 1 and |φ(x)| = kxk.

Proof. This is the Hahn-Banach separation Theorem.

(b) Let E and F be Banach spaces, let π : E → F be a bounded linear map and let π ∗ : F ∗ → E ∗ be
the induced map on dual spaces. Show that kπ ∗ k = kπk.

Proof. We have π ∗ (y ∗ )(x) = y ∗ (π(x)) for all y ∗ ∈ F ∗ and x ∈ E. Then kπ ∗ (y ∗ )(x)k ≤

ky ∗ kkπkkxk so then kπ ∗ k ≤ kπk.
On the other hand, by part (a), for each x ∈ E such that kxk ≤ 1, π(x) ∈ F , we can find
y ∗ ∈ F ∗ such that |y ∗ (π(x))| = kπ(x)k and ky ∗ k = 1. Then

kπ ∗ k ≥ kπ ∗ (y ∗ )k ≥ |π ∗ (y)(x)| = |y ∗ (π(x))| = kπ(x)k ∀kxk ≤ 1

So kπ ∗ k ≥ kπk. THus, kπk = kπ ∗ k.

Problem 10. Let X be a real Banach space and suppose C is a closed subset of X such that

(i) x1 + x2 ∈ C for all x1 , x2 ∈ C,

(ii) λx ∈ C for all x ∈ C and λ > 0,

(iii) for all x ∈ X there exists x1 , x2 ∈ C such that x = x1 − x2 .

85
14 JANUARY 2013 Kari Eifler

Prove that, for some M > 0, the unit ball of X is contained in the closure of

{x1 − x2 | xi ∈ C, kxi k ≤ M }.

Deduce that every x ∈ X can be written x = x1 − x2 , with xi ∈ C and kxi k ≤ 2M kxk.

Proof. Define

Cn = {x1 − x2 | xi ∈ C, kxi k ≤ n}

◦
By (iii), we know that X = ∪Cn . By Baire Category, there exists some M such that ∅ =
6 CM =
◦
CM . Thus, there exists an open ball B ⊆ CM , B = B(x0 , 2r).

For any x ∈ BX , x0 + rx ∈ B ⊆ CM . From (i), we know that CM − CM ⊆ C2M so then rx =

(x0 + rx) − x0 ∈ CM − CM ⊆ C2M . From (ii), we know x ∈ C2M/r , so BX ⊆ C2M/r . Let M 0 = 2M
r .

For any x ∈ X, x ∈ CM 0 kxk . So we can find z1 , y1 ∈ C such that kz1 k, ky1 k ≤ M kxk and kx − (z1 −
y1 )k < 21 kxk. Therefore,

2(x − (z1 − y1 ))
∈ CM ⇒ x − (z1 − y1 ) ∈ CM kxk/2
kxk

M
So we can find z2 , y2 ∈ C such that kz2 k, ky2 k ≤ 2 kxk and

2

X 1
x − (zi − yi ) < 2 kxk.

2
i=1

M
Inductively, we can find {zn }, {yn } ⊆ C such that kzk k, kyk k ≤ 2k−1
kxk and

k

X 1
x − (zi − yi ) < k kxk.

2
i=1

Then,

∞ ∞
X X 1
kzk k ≤ M kxk < 2M kxk < ∞
2k
k=1 k=1

P∞ P∞
so k=1 zk converges to some x1 in C and similarly k=1 yk converges to some x2 in C (since C is
closed). Moreover,

n
n n
!
X X X
lim x − (zi − yi ) = lim x − zi − yi = 0.

n n
i=1 i=1 i=1

86
15 AUGUST 2012 Kari Eifler

P∞
So then x = i=1 (zi − yi ) = x1 − x2 .

15 August 2012

Problem 1. Let (X, M, µ) be a measure space. Prove that the normed vector space L1 (X, µ) is
complete. You may use any results except the convergence of function series.

Proof. See class notes.

Problem 2. Fix two measure spaces (X, M, µ) and (Y, N , ν) with µ(X), ν(Y ) > 0. Let f : X → C,
g : Y → C be measurable. Suppose f (x) = g(y), (µ × ν)-a.e. Show that there is a constant a ∈ C
such that f (x) = a µ-a.e. and g(y) = a ν-a.e.

Proof. Let E := {(x, y) ∈ X × Y | f (x) = g(y)}, so (µ ⊗ ν)(E c ) = 0. Then for every a ∈ C, by

Fubini-Tonelli,

0 = (µ ⊗ ν) ({(x, y) ∈ X × Y | f (x) = a, g(y) 6= a}) = µ ({x ∈ X | f (x) = a}) ν ({y ∈ Y | g(y) 6= a}) .

Assume µ({x ∈ X | f (x) = a}) = 0 for all a ∈ C. Then

0 < (µ ⊗ ν)(X × Y )
Z
= (µ ⊗ ν)(E) = χ{(x,y)|f (x)=g(y)} dµ(x)dν(y)
X×Y
Z Z 
= χ{x|f (x)=g(y)} dµ(x) dν(y)
ZY X

= 0dν(y) = 0.
Y

This is a contradiction so there must exist some a ∈ C with µ({x ∈ X | f (x) = a}) > 0. Then
ν({y ∈ Y | g(y) 6= 0}) = 0 so g(y) = a ν-a.e.

Similarly, we have (µ⊗ν)({(x, y) | f (x) 6= a, g(y) = a}) = 0. Since ν({y ∈ Y | g(y) = a}) = ν(Y ) 6= 0,
then µ({x | f (x) 6= a}) = 0 so f (x) = a µ-a.e.

Problem 3. Let f : R3 → R be a Borel measurable function. Suppose for every ball B, f is

R
Lebesgue integrable on B and B f (x)dx = 0. What can you deduce about f ? Justify your answer
carefully.

Proof. Since f ∈ L1loc (R2 ), by Lebesgue Differentiation Theorem, for a.e. x0 ∈ R2 ,

Z
1
lim f (x)dx = f (x0 )
r→0 |B(r, x0 )| B(r,x0 )

87
15 AUGUST 2012 Kari Eifler

This implies f (x0 ) = 0 so f = 0 almost everywhere.

Problem 4. Let X be a locally compact Hausdorff space. Denote by C0 (X) the space of complex-
valued continuous functions on X which vanish at infinity, and by Cc (X) the subset of compactly
supported functions. Use an approximate version of the Stone-Weierstrass theorem to prove that
Cc (X) is dense in C0 (X).

Proof. For any f, g ∈ Cc (X), the complex conjugation of f is also in Cc (X).

By complex-LCH-Stone-Weierstrass, we only need to show that Cc (X) separates points.

For every x 6= y, we can find open U, V with x ∈ U, y ∈ V with U ∩ V = ∅. Since X is LCH, we can
require U to be compact.

Now {x} ⊆ U ⊆ U ⊆ X\V ⊆ X\{y}. Then by Urysohn’s Lemma for LCH, we can find a continuous
function f : X → [0, 1] such taht f |U = 1 and f (x) = 0 outside a compact subset of X\{y}. So
f (x) = 1, f (y) = 0, and f ∈ Cc (X).

So Cc (X) separates points. Also, there does not exist any x0 ∈ X such that f (x0 ) = 0 for all f ∈
Cc (X).

Therefore, by Stone-Weierstrass, Cc (X) = C0 (X).

Problem 5. Give an example of each of the following. Justify your answers

(a) A nowhere dense subset of R of positive Lebesgue measure

Proof. Take a fat Cantor set.

(b) A closed, convex subset of a Banach space with multiple points of minimal norm.
R1
Proof. Let X = L1 [0, 1], C = {f ∈ X | 0 f (t)dt = 0}. It’s easy to see that C is closed and
convex. The minimum norm of elements in C is 1 because

Z 1 Z
 1 

kf k1 = |f (t)|dt ≥  f (t)dt = 1.
0 0

But every element of {aχ[0,1/2] + (2 − a)χ[1/2,1] }0≤a≤2 in C has norm 1.

Problem 6. Let

 
1
S= f ∈ L∞ (R) | |f (x)| ≤ a.e. .
1 + x2

Which of the following statements are true? Prove your answers.

(a) The closure of S is compact in the norm topology

88
15 AUGUST 2012 Kari Eifler

Proof. NO. Let

1
fn (x) := χ 1 1 (x)
x2 + 1 [ n+1 , n ]
in S. So there are no subsequences of (fn ) which are Cauchy in L∞ since kfn − fm k∞ ≥ 1 for
n 6= m.

(b) S is closed in the norm topology.

Proof. YES. Suppose (fn ) ⊆ S, fn → f in L∞ . Then

1 1
|f (x)| ≤ |fn (x)| + |fn (x) − f (x)| < + kfn − f k∞ < +  a.e.
1 + x2 1 + x2

Letting  → 0, we have |f (x)| ≤ 1

1+x2 a.e. and f ∈ L∞ so f ∈ S.

(c) The closure of S is compact in the weak* topology

Proof. YES. The unit ball in L∞ (R) is weak*-compact by Alaoglu. Since 1+x
1
2 ≤ 1 for all x ∈
∞ w∗
R, then S is a subset of the unit ball in L . Therefore, S is weak* compact.

Problem 7. Let T be a bounded operator on a Hilbert space H. Prove taht kT ∗ T k = kT k2 . State

the results you are using.

Proof. Clearly, kT ∗ T k ≤ kT ∗ kkT k = kT k2 . On the other hand,

kT k2 = sup |hT x, T xi| = sup |hT ∗ T x, xi|.

kxk=1 kxk=1

Since for kxk = 1,

|hT ∗ T x, xi| ≤ kT ∗ T xkkxk ≤ kT ∗ T kkxk2 ≤ kT ∗ T k

then kT k2 ≤ kT ∗ T k.

Problem 8. (a) Let g be an integrable function on [0, 1]. Does there exist a bounded measurable
R1
function f such that kf k∞ 6= 0 and 0 f gdx = kgk1 kf k∞ ? Give a construction or a counterex-
ample.

Proof. YES. For any g ∈ L1 , let f = sgn(g) where g(x) 6= 0, and 1 where g(x) = 0. Then
kf k∞ = 1 and

Z 1 Z 1
fg = |g(x)|dx = kgk1 = kgk1 kf k∞ .
0 0

89
15 AUGUST 2012 Kari Eifler

(b) Let g be a bounded measurable function on [0, 1]. Does there exist an integrable function f such
R1
that kf k1 6= 0 and 0 f gdx = kgk∞ kf k1 ? Give a construction or a counterexample.

Proof. NO. Let g(x) = x on [0, 1] so kgk∞ = 1, implying g ∈ L∞ [0, 1]. Assume such an f ∈ L1
exists, so

Z 1 Z 1
kf k1 = kf k1 kgk∞ = f gdx = xf (x)dx
0 0
R1 R1 R1
and also kf k1 = 0
|f |dx so then 0
f (x)xdx = 0
|f |dx. Therefore,

Z 1 Z 1 Z 1   Z 1−1/n Z 1
1
|f (x)|dx = xf (x)dx ≤ x|f (x)|dx ≤ 1 − |f (x)|dx + |f (x)|dx
0 0 0 n 0 1−1/n

So then

Z 1−1/n Z 1  Z 1−1/n Z 1
1
|f (x)|dx + |f (x)|dx ≤ 1− |f (x)|dx + |f (x)|dx
0 1−1/n n 0 1−1/n

R 1−1/n
Thus, 0 |f (x)|dx = 0 for all n ∈ N. Letting fn (x) = χ[0,1−1/n] |f (x)| % |f (x)| then by
R R
monotone convergence theorem, |f (x)dx = limn fn (x) = 0 so kf k1 = 0.

Problem 9. Let F : R → C be a bounded continuous function, µ the Lebesgue measure, and f, g ∈

L1 (µ). Let

Z Z
f˜(x) = F (xy)f (y)dµ(y), g̃(x) = F (xy)g(y)dµ(y).

Show that f˜ and g̃ are bounded continuous functions which satisfy

Z Z
f g̃dµ = f˜gdµ.

Proof. We have kf˜k| inf ty ≤ kF k∞ kf k1 < ∞ and kg̃k ≤ kF k∞ kgk1 < ∞ so f˜, g̃ ∈ L∞ . By
R
dominated convergence theorem, we know that limn [−n,n] |f (x)dµ = kf k1 . So then for every  > 0,
R
there exists some N such that R\[−n,n] |f (x)|dµ < . Then

Z
|f˜(x1 ) − f˜(x2 )| ≤ |F (x1 y) − F (x2 y)||f (y)|dµ(y)
Z Z
= |F (x1 y) − F (x2 y)||f (y)|dµ(y) + |F (x1 y) − F (x2 y)||f (y)|dµ(y)
[−n,n] R\[−n,n]

≤ sup |F (x1 y) − F (x2 y)|kf k1 + 2kF k∞ 

y∈[−n,n]

90
15 AUGUST 2012 Kari Eifler

δ
Since F is continuous, let |x1 − x2 | < n such that |x1 y − x2 y| < δ imples |F (x1 y) − F (x2 y)| < . So
|f˜(x1 ) − f˜(x2 )| → 0 as |x1 − x2 | → 0.

A similar argument will show that g̃ is continuous. Since f g̃ ∈ L1 , by Fubini,

Z Z Z
f g̃dµ = f (x)F (xy)g(y)dµ(y)dµ(x)
Z Z 
= g(y) f (x)F (xy)dµ(x) dµ(y)
Z
= g(y)f˜(y)dµ(y)
Z
= f˜gdµ.

Problem 10. Let µ, {µn | n ∈ N} be finite Borel measures on [0, 1]. µn → µ vaguely if it converges
in the weak* topology on M [0, 1] = (C[0, 1])∗ . µn → µ in moments if for each k ∈ {0} ∪ N,

Z Z
xk dµn (x) → xk dµ(x).
[0,1] [0,1]

Show that µn → µ vaguely if and only if µn → µ in moments.

Proof. ⇒) trivial by the definitions

R R
⇐) We want to show that for all f ∈ C[0, 1], f dµn → f dµ. By Stone-Weierstrass, we can find
pn to be a sequence of polynomials which converge uniformly to f on [0, 1]

Z Z  Z Z  Z Z  Z Z 
       
 f (x)dµ − f (x)dµn  ≤  f dµ − pm dµ +  pm dµ − pm dµn  +  pm dµn − f dµn 
       

R R R R 
For the first part, | f dµ− pm dµ| ≤ kf −pm k∞ µ(X) → 0 as m → ∞. Similarly,  pm dµn − f dµn  ≤
kf − pm k∞ µn (X) → 0 for all n.

Next, find a polynomial qmj with  that kqmj − pm k∞ → 0 as j → ∞. Then

R degree at Rmost j such
since µn → µ in moments, then qmj dµ − qmj dµn  → 0 for all j. Thus,


Z Z  Z Z  Z Z  Z Z 
       
 pm dµ − pm dµn  ≤  pn dµ − qmj dµ +  qmj dµ − qmj dµn  +  qmj dµn − pm dµn 
       
Z Z 
 
≤ kpm − qmj k∞ (µ(X) + µn (X)) +  qmj dµ − qmj dµn  → 0.

91
16 JANUARY 2012 Kari Eifler

16 January 2012

Problem 1. Let A be the subset of [0, 1] consisting of numbers whose decimal expansions contain
no sevens. Show that A is Lebesgue measurable, and find its measure. Why does non-uniqueness of
decimal expansions not cause any problems?

Proof. Let Ai be the subset of [0, 1] consisting of numbers whose first i digits are not 7. Then An+1 ⊆
An and A = ∩n An ,

A1 = [0, 0.7] ∪ [0.8, 1]

A2 = [0, 0.07] ∪ [0.08, 0.17] ∪ . . . ∪ [0.98, 1]

So An is the union of some Borel intervals in [0, 1], so An is Lebesgue measurable. Therefore, A is
Lebesgue measurable.

Now for 0 ≤ i ≤ 9, let Ain be the subset of An such that the (n + 1)th digit is i. Then we can write
An = t9i=0 Ain .

Also, m(Ain ) = m(Ajn ), so m(An ) = 10m(Ain ) and An+1 = ti6=7 Ain so m(An+1 ) = 9m(Ain ). There-
9
fore, m(An+1 ) = 10 m(An ). Then

 n−1
9
m(A) = m (∩∞
n=1 An ) = lim m(An ) = lim m(A1 ) = 0
n n 10

The only numbers with non-unique decimal representation are 0.a1 a2 . . . an = 0.a1 a2 . . . an−1 999 . . ..
However ∀n there are only finitely many, so non-unique = ∪n {0.a1 . . . an } which is countable, hence
null, hence Lebesgue.

Problem 2. Let the functions fα be defined by

(
xα cos(1/x) x>0
fα (x) =
0 x=0

Find all values of α ≥ 0 such that

(a) fα is continuous

Proof. When a > 0, xa cos(1/x) ≤ xa → 0 as x → 0 so fa is continuous. If a = 0, we know

cos(1/x) isn’t continuous at 0.

(b) fα is of bounded variation on [0, 1]

Proof. First, 0 < a ≤ 1, put partitions

92
16 JANUARY 2012 Kari Eifler

 
1 1 1
Pm = 0, , ,..., ,1
2πm π(2m − 1) π

Then
 
1 −1 −1
fa (Pm ) = 0, , , . . . , a , cos(1)
(π2m)a (π(2n − 1))a π

So

      X2m
 1   −1 1  cos(1) − −1  ≈
  c
Tfα (Pm ) =  − 0+ − + . . . + →∞
(π(2m))a   (π(2m − 1))a (π2m)a   π a  i=1 (πi)a

when 0 < a ≤ 1 as m → ∞.
So when 0 < a ≤ 1, fa is not of bounded variation when 0 < a ≤ 1. For a > 1, let’s look at
(c).

(c) fα is absolutely continuous on [0, 1]

Proof. When a > 1, we see fa0 (0) = 0 and fa0 is integrable because fa0 (x) = axa−1 cos(1/x) +
Rx
xa−2 sin(1/x) so then fa (x) = 0 fa0 (t)dt. Thus, f is absolutely continuous.
So in (b) we have fa is of bounded variation for a > 1. Since fa isn’t bounded variation when
0 < a ≤ 1, so fa isn’t absolutely continuous either when 0 < a ≤ 1.

Problem 3. Let F denote the family of functions on [0, 1] of the form

∞
X
f (x) = an sin(nx)
n=1

where an are real and |an | ≤ 1/n3 . State a general theorem and use that theorem to prove taht any
sequences in F has a subsequence that converges uniformly on [0, 1].

Proof. We’ll use Arzela-Ascoli.

For all f ∈ F,

∞  ∞ ∞
X  X X
|f (x)| =  an sin(nx) ≤ |an | ≤ n−3 < ∞
 
 
n=1 n=1 n=1

so uniformly bounded. Also, for all f ∈ F,

∞ ∞ ∞
π2
   X
cos nx + ny  sin nx + ny  ≤
X X
−3 −2
  
|f (x)−f (y)| ≤ |an || sin(nx)−sin(ny)| ≤ 2n n |x−y| = |x−y|
n=1 n=1
 2  2 
n=1
6

93
16 JANUARY 2012 Kari Eifler

So F is equicontinuous.

Then F is compact, hence sequentially compact. So F has a subsequence that converges in the uni-
form norm.

Problem 4. Let H be a Hilbert space and W ⊂ H a subspace. Show that H = W ⊕ W ⊥ where W is

the closure of W .

Note: Do not just state this as a consequence of a standard result, prove the result.

Proof. here!

Problem 5. Suppose A is a bounded linear operator on a Hilbert space H with the property that

kp(A)k ≤ C sup{|p(z)| | z ∈ C, |z| = 1}

for all polynomials p with complex coefficients, and a fixed constant C. Show that to each pair x, y ∈
H there corresponds a complex Borel measure µ on the circle S 1 = {z ∈ C | |z| = 1} such that

Z
n
hA x, yi = z n dµ(z) n = 0, 1, 2, . . .

Proof. Consider

Tx,y : P (S 1 ) → C
p 7→ hP (A)x, yi

Then
|hP (A)x, yi| ≤ kP (A)kkxkkyk ≤ CkP k∞ kxkkyk

Thus, |Tx,y (P )| ≤ CkxkkykkP k∞ = f (P ) which is obviously a seminorm. By Hahn-Banach, Tx,y

can be extended to C(S 1 ).

Then apply Riesz-Representation Theorem, there exists a complex Borel measure µ on S 1 such that

Z
Tx,y (P ) = hP (A)x, yi = P (z)dµ(z)
S1

Take P (z) = z n so hAn x, yi = z n dµ(z).

R
S1

Problem 6. Let φ be the linear functional

Z 1
φ(f ) = f (0) − f (t)dt
−1

(a) Compute the norm of φ as a functional on the Banach space C[−1, 1] with uniform norm

94
16 JANUARY 2012 Kari Eifler

Proof. Z 1 Z 1
|φ(f )| ≤ |f (0)| + |f (t)|dt ≤ kf k∞ + kf k∞ dt = 3kf k∞
−1 −1

So kφk ≤ 3. On the other hand, let fn be piecewise linear functional such that fn = −1 on
[−1, −1/n] and [1/n, 1] and fn (0) = 1. Then

Z 1
2 4
fn (t)dt = −2(1 − 1/n) + = −2 + → −2
−1 n n

So sup |φ(fn )| ≥ 3 so kφk = 3.

(b) Comptue the norm of φ as a functional on the normed vector space LC[−1, 1] which is C[−1, 1]
with the L1 norm.

Proof. R1
|f (0) − −1
f (t)dt| |1 − 1/n|
kφk = sup ≥ lim =∞
f ∈LC[−1,1] kf k1 n (1/n)

Problem 7. Let X be a normed space and A ⊂ X be a subset. Show that A is bounded (as a set) if
and only if it is weakly bounded (that is, f (A) ⊂ C is bounded for each f ∈ X ∗ ).

Proof. ⇒) for all x ∈ A, for all f ∈ X ∗ , |f (x)| ≤ kf kkxk < ∞ so A is weakly bounded

⇐) on the other hand, consider A∗∗ = {a∗∗ | a ∈ A} by a∗∗ (f ) = f (a) for all f ∈ X ∗ . Since X ∗ is
Banach, and we know

sup ka∗∗ (f )k = sup |f (a)| < ∞ ∀f ∈ X ∗

a∗∗ ∈A∗∗ a∈A

By the uniform boundedness principle, supa∈A kak = supa∗∗ ∈A∗∗ ka∗∗ k < ∞.

Problem 8. Let X be a topological vector space.

(a) Define what this means.

Proof. Let X be a vector space, T a topology on X. Then (X, T ) is a topological vector space
provided

• + : X × X → X is continuous
• · : R × X → X is continuous

(b) Let A ⊂ X be compact and B ⊂ X be closed. Show that A + B ⊂ X is closed.

95
16 JANUARY 2012 Kari Eifler

Proof. Fix z ∈ (A + B)c . For x ∈ A, z − x ∈ B c so there exists an open neighborhood Vx 3

0 in X such that (z − x + Vx ) ∩ B = ∅. Since addition is continuous, there exists U1x , U2x
neighborhoods of 0 such that U1x + U2x ⊆ Vx .
Ux = U1x ∩ U2x ∩ (−U1x ) ∩ (−U2x ) so Ux = −Ux . Then {x + Ux }x∈A is an open cover of A. Since
A is compact, there exists a finite subcover x1 , . . . , xn ∈ A such that

A ⊆ ∪ni=1 xi + Uxi

Put U = ∩ni=1 Uxi . Then z + U is an open neighborhood of z. If there exists x ∈ A y ∈ B such

that x + y ∈ z + U then x ∈ xi + Uxi for some i and y ∈ z − x + U ⊆ z − xi + Uxi ⊆ z − xi + Vxi
but (z − xi + Vxi ) ∩ B = ∅. Contradiction!
So (z + U ) ∩ (A + B) = ∅.

(c) Give an example indicating that the condition ‘A closed’ is insufficient for the conclusion.

Proof. X = R2 , A = {(x, 0) | x ∈ R} and B = {(x, 1/x) | x > 0}. Then A + B = {(x, y) | y >
0}.

Problem 9. Let (X, M, µ) be a finite measure space. Let f, fn ∈ L3 (X, dµ) for n ∈ N be functions
such that fn → f µ-a.e. and |fn | ≤ M for all n. Let g ∈ L3/2 (X, dµ). Show that

Z Z
lim fn gdµ = f gdµ.
n

Proof. |fn g| ≤ M |g|. Since µ is a finite measure, M 1 ∈ L3 (µ). By Holder, M |g| ∈ L1 (µ). The result
follows from Dominated Convergence Theorem.

Problem 10. Let X be a σ-finite measure space, and fn : X → R a sequence of measurable func-
tions on it. Suppose fn → 0 in L2 and L4 .

(a) Does fn → 0 in L1 ?

Proof. NO.
Let X = R, µ=Lebesgue measure. fn = n−1 χ[0,n] so kfn k1 = 1 does not converge to 0, but
kfn k2 = n−1/2 → 0 and kfn k4 = n−3/4 → 0.

(b) Does fn → 0 in L5 ?

Proof. YES.
Since 0 < 2 < 3 < 4 < ∞, L2 ∩ L4 ⊆ L3 and kf k3 ≤ kf kλ2 kf k1−λ
4 where 1
3 = λ
2 + 1−λ
4 implies
λ = 13 . So

1/3 2/3
kfn k3 ≤ kfn k2 kfn k4 →0

(c) Does fn → 0 in L5 ?

96
17 AUGUST 2011 Kari Eifler

Proof. NO.
X = [0, 1], µ: Lebesgue measure. Let fn = nχ[0,n−5 ] . Then kfn k5 = 1 but kfn k2 = n−3/2 → 0
and kfn k4 = n−1/4 → 0.

17 August 2011

Problem 1. Let (X, M, µ) be a measure space.

(a) Give the definitions of convergence a.e. and convergence in measure for a sequence of measur-
able functions on X.

Proof. We say a sequence of measurable functions fn converge to f almost everywhere if µ({x |

limn fn (x) 6= f (x)}) = 0.
We say that fn converges to f in measure if ∀ > 0, limn µ({x | |f (x) − fn (x)| > }) = 0.

(b) Show that every sequence of measurable functions on X which converges in measure to 0 has a
subsequence which converges a.e. to 0.

Proof. Suppose for every  > 0, µ({x | |fn (x)| ≥ }) → 0. Choose a subsequence {fnk } such that
if

Ej = {x | |fnj (x) − fnj+1 (x)| > 2−j }

P∞
satisfies µ(Ej ) < 2−j . Let Fk = ∪∞
j=k Ej so µ(Fk ) ≤ j=k 2−j ≤ 21−k . Let F = ∩k Fk so
µ(F ) = 0.
For x ∈
/ Fk and for i ≥ j ≥ k then

i−1
X i−1
X
|fni (x) − fnj (x)| ≤ |fn` (x) − fn`+1 (x)| ≤ 2` ≤ 2−j → 0 as k → ∞.
`=j `=j

So fnk is pointwise Cauchy on x ∈

/ F , so let
(
lim fnk (x) x ∈
/F
f (x) =
0 otherwise

So fnk → 0 almost everywhere and fn → f in measure since

µ({x | |fn (x) − f (x)| ≥ }) ≤ µ({x | |fn (x) − fn` (x)| ≥ /2}) + µ({x | |fn` (x) − f (x)| ≥ })
| {z } | {z }
→0 →0

and

97
17 AUGUST 2011 Kari Eifler

µ({x | |f (x)| ≥ }) ≤ µ({x | |f (x) − fn (x)| ≥ /2} + µ({x | |fn (x)| ≥ /2})
| {z } | {z }
→0 →0

so f = 0 almost everywhere. Thus, {fnk } converges to 0 almost everywhere.

Problem 2. Let X be a separable Banach space. Show that there exists an isometric linear map
from X into `∞ . Also, show that this is false in general if `∞ is replaced by `2 .

Proof. Let (xn ) be a dense sequence in BX . For each n, use Hahn-Banach Theorem to find a norm-
one functional fn ∈ X ∗ with fn (xn ) = 1.

Define φ : X → `∞ via φ(x) = fn (x) . Suppose x ∈ X has norm one and let 1 >  > 0. Choose n

so that kxn − xk < . Then

 > |fn (xn − x)| = |fn (x)|

So kφ(x)k = supn |fn (x)| ≥ 1. For every n, |fn (x)| ≤ kfn kkxk = 1 so kφ(x)k ≤ 1. So kφ(x)k = 1
whenever kxk = 1. Then for all non-zero x, kφ(x)k = kxk supn |fn (x/kxk)| = kxk. So φ is an
isometry.

Why FALSE for `2 ?

Problem 3. Let X be a locally compact metric space and let {xk } be a sequence in X which has no
Pn
convergent subsequence. Show that {n−1 k=1 δxk } converges to 0 in the weak* topology on C0 (X)∗ ,
where δxk denotes the point mass at xk .

Proof. here

Problem 4. Let P be the set of all polynomials f on [0, 1] such that f (0) = f 0 (0) = 0. Determine,
with proof, the values of p with 1 ≤ p ≤ ∞ such that P is dense in Lp [0, 1].

Proof. All 1 ≤ p < ∞. Clearly, P is an algebra which separates points (ex. x2 ). Stone-Weierstrass
implies P = {f ∈ C[0, 1] | f (0) = 0}. Now for any f ∈ Lp , for all  > 0, there exists some N such
that

f − f χ[−N ≤f ≤N ] ≤ 

p 2

Define fN = f χ[−N ≤f ≤N ] . By Lusin’s theorem, there exists a closed set F such that m([0, 1]\F ) ≤
1 p 1  p


2p (2N )p = 2p 2N . and fN |F continuous.

Tietze extension theorem applied to fN and F implies the extension g is still bounded by N . Then

p
Z
kfN − gkpp = |fN − g|p ≤ (2N )p m([0, 1]\F ) ≤
[0,1]\F 2p

So then

98
17 AUGUST 2011 Kari Eifler

p 
kf − gkp ≤ kf − fN kp + kfN − gkp ≤ p
+ ≤
2 2

NOT L∞ [0, 1] since P = {f ∈ C[0, 1] | f (0) = 0}. If f ∈ L∞ [0, 1] with f (0) = a 6= 0, then ∀g ∈ P,
kf − gk∞ = a.

Problem 5. Let 1 < p < ∞ and let {xk }∞ p

k=1 be a sequence in ` (N) such that limk xk (n) = 0 for all
n ∈ N. Show that if there is an M > 0 such that kxk k ≤ M for all k ∈ N then xk → 0 weakly.

Also, show that if no such M exists, then {xk } can fail to converge weakly.

Proof. Note: Similar to August 2015, #3, just in a different space now.

Fix some y ∈ `q where p1 + 1q = 1. We want to show that n xk (n)y(n) → 0 as k → ∞. Fix  > 0.

P

Then we may choose a finite A ⊆ N such that Ac |y(n)|q < q . Since A is finite, choose some K
P
p

such that for all k ≥ K we have |xk (n)|p < |A| . Then for all k ≥ K, by using Holder, we have

X
|y(xk )| ≤ |xk (n)||y(n)|
n∈N
X X
= |xk (n)||y(n)| + |xk (n)||y(n)|
n∈A n∈Ac
!1/p !1/q !1/p !1/q
X X X X
p q p q
≤ |xk (n)| |y(n)| + |xk (n)| |y(n)|
n∈A n∈A n∈Ac n∈Ac
1/p
p

≤ |A| kykq + M 
|A|
=  (kykq + M )

By making  small enough, we see that |y(xk )| → 0 as k → ∞.

To see why we require (xk ) to be bounded, consider p = q = 2. Take

 
1 1 1
xk = (0, 0, . . . , 0, 2k , 0, . . .) = 2k ek y= , 2,..., n,...
2 2 2

where xk is all zeros except in the kth spot. Then we can see that limk xk (n) = 0 for all n, but that
for all k,

X 1
y(xk ) = xk (n)y(n) = 2k =1
n
2k

Problem 6. Let f ∈ C0 (R) and for every t ∈ R define ft ∈ C0 (R) by ft (x) = f (x + t) for all x ∈ R.

(a) Prove that {ft | t ∈ [0, 1]} is compact in the norm topology.

99
17 AUGUST 2011 Kari Eifler

Proof. Similar to August 2013 #1

Since Cc∞ (R) is dense in C0 (R), we can choose g ∈ Cc∞ (R) such that kg − f k∞ < . Then

kftn − ft k∞ ≤ kftn − gtn k∞ + kgtn − gt k∞ + kgt − ft k∞

It’s easy to see kftn − gtn k∞ and kgt − ft k∞ are small since kg − f k∞ < . For kgtn − gt k∞ , then

kgtn − gt k∞ = sup |gtn (x) − gt (x)| = sup |g(x + tn ) − g(x + t)|

x∈R x∈R

where for each fixed tn → t, since g is compactly supported and continuous then can be suffi-
ciently small for large enough n.
Therefore, the map G : R → C0 (R) given by G(t) = ft is continuous. Since {ft | t ∈ [0, 1]} =
G([0, 1]) and continuous maps preserve compactness, then the set is compact in the norm topol-
ogy.

(b) Prove that {ft | t ∈ R} is relatively compact in the weak topology.

Proof. here

Problem 7. Let f be an arbitrary real valued function on [0, 1]. Show that the set of points at
which f is continuous is a Lebesgue measurable set.

Proof. Similar to August 2016, #3.

In fact, we will prove that the set of points at which f is discontinuous is a countable union of closed
subsets.

f is continuous at p if for all n, there exists an open U containing p such that |f (x) − f (y)| < 1/n
for all x, y ∈ U . Fix n and let

[ [
Vn = {p s.t. there exists an appropriate U } = {appropriate U }
p

Hence, Vn is open. Then

\
{points where f is continuous} = Vn
n

Vnc where Vnc is closed.

S
So {points where f is discontinuous} = n

Problem 8. Show that not every nonempty bounded closed subset of `2 has a point of minimal
norm, but that every nonempty bounded closed convex subset of `2 has a point of minimal norm.

Proof. Let C be the bounded, closed, convex subset of `2 . Consider the set {y ∈ R | y = kxk, x ∈ C}
and since this set is bounded below, there exists an infimum of the set, say s. Then we can find a
sequence xn ∈ C such that s ≤ kxn k ≤ s + n1 .

100
17 AUGUST 2011 Kari Eifler

I claim that (xn ) is a Cauchy sequence. Indeed, for any  > 0, choose r to be the positive root of
2
the equation r2 + 2rs − 4 = 0.

Since kxn k → s then there is an N such that s ≤ kxn k < s + r for all n ≥ N . If n, m ≥ N , then

xm − xn 2 xm + xn 2 2 2 2
x 2 x 2
= 2 m n
+ 2 − < (s + r) + (s + r) − s2 = 2sr + r2 =  .
2 2 2 2 2 2 4

So (xn ) is a Cauchy sequence, which means it converges to some x. Since C is closed, x ∈ C and
obtains minimal norm.

Note: This choice of x is unique! If there were two points of minimal

norm, say x1 and x2 then
1 1
2 (x1 + x2 ) ∈ C by the convexity of C. So s ≤

2 (x1 + x2 )
1
≤ 2 kx1 k + 12 kx2 k = s. Hence,
kx1 + x2 k = 2s. By the parallelogram law,

kx1 + x2 k2 + kx1 − x2 k2 = 2kx1 k2 + 2kx2 k2

And so kx1 − x2 k2 = 4s2 − 4s2 = 0 so x1 = x2 , proving uniqueness.

= n+1

Counterexample: Consider M √ n en | n ∈ N . M is closed since the distance between any two
of its elements is greater than 2 (and thus the only convergent sequences from M are those that
are eventually constant). M is clearly non-empty and has no element of minimal norm.

Problem 9. Show that there is a sequence {fn } of continuous functions on [0, 1] such that

(a) |fn (t)| = 1 for all n and all t ∈ [0, 1] and

R1
(b) for all g ∈ L1 [0, 1] one has 0 fn (t)g(t)dt → 0 as n → ∞

Proof. NOT POSSIBLE??

If fn is continuous on [0, 1] and |fn (t)| = 1 then fn (t) = ±1. Since fn is continous, each fn is the
constant function at either 1 or −1. Write it as fn (x) = (−1)kn where kn is even or odd depending
on n.

Then if we take g to be the constant function 1, we get

Z 1 Z 1
fn (t)g(t)dt = (−1)kn dt = (−1)kn
0 0

which does not have to converge to 0 as n → ∞.

right? obvious? I don’t get it...

Problem 10. (a) Define what it means for a real valued function on [0, 1] to be absolutely continu-
ous.

101
17 AUGUST 2011 Kari Eifler

Proof. The function f : [0, 1] → R is absolutely continuous if for every  > 0 there exists δ >
0 such that whenever a finite sequence of pairwise disjoint sub-intervals (xk , yk ) of [0, 1] with
P P
xk , yk ∈ [0, 1] satisfy k (yk − xk ) < δ then k |f (yk ) − f (xk )| < .
Equivalently, f has a derivative f 0 almost everywhere and the derviative is Lebesgue integrable
and for all x ∈ [0, 1],
Z x
f (x) = f (0) + f 0 (t)dt.
0

(b) Prove that if f and g are absolutely continuous strictly positive functions on [0, 1] then f /g is
absolutely continuous on [0, 1].

Proof. Step 1: If f is absolutely continuous, then so is 1/f .

1
Since f > 0 is continuous on a compact space, there exists some M ∈ N such that M ≤ |f (x)| ≤
M for all x ∈ [0, 1].
Indeed, since g is absolutely continuous then for every  > 0, there exists a δ > 0 such that
whenever a finite sequence of pairwise disjoint sub-intervals (xk , yk ) of [0, 1] with xk , yk ∈ [0, 1]
satisfy k (yk − xk ) < δ then k |f (yk ) − f (xk )| < M 2 . Then for such intervals, we have
P P

X  1  
1  X  f (xk ) − f (yk ) 

 f (yk ) − f (xk  =

 f (yk )f (xk ) 
  
X 1  1 
≤  f (yk )   f (xk )  |f (yk ) − f (xk )|
   
X
≤ M2 |f (yk ) − f (xk )|

= M 2 2 = .
M

Step 2: If f and g are both absolutely continuous, then so is f g.

Find M ∈ N such that |f (x)|, |g(x)| ≤ M for all x ∈ [0, 1].
P P
Take δ1 such that if yk − xk < δ1 then |f (yk ) − f (xk )| < /2M . Similarly, take δ2 such that
P P
if yk − xk < δ2 then |g(yk ) − g(xk )| < /2M . Let δ = min(δ1 , δ2 ). Now

X X
|(f g)(yk ) − (f g)(xk )| = |f (yk )g(yk ) − f (xn )g(xn )|
X
≤ |f (yk )g(yk ) − f (yk )g(xk )| + |f (yk )g(xk ) − f (xk )g(xk )|
X X
≤ |f (yk )||g(yk ) − g(xk )| + |g(xn )||f (yk ) − f (xk )|
X X
≤M |g(yk ) − g(xk )| + M |f (yk ) − g(xk )|
 
≤M +M
2M 2M
=

Combining the two steps, we see immediately that f /g is absolutely continuous.

102