MA 611: Homework #1

∗
Yingwei Wang
Department of Mathematics, Purdue University, West Lafayette, IN, USA

1 Finite dimensional normed linear space

Question: Let Y be a finite dimensional linear subspace of a normed linear space X, and
let x0 ∈ X, x0 ∈
/ Y . Then there exist a point y0 ∈ Y such that

inf kx0 − yk = kx0 − y0 k.

y∈Y

Proof: Define the distance between x0 and Y as

d = inf kx0 − yk.

y∈Y

So we can find a sequence {yn }∞

n=1 such that

1
kyn − x0 k ≤ d − , ∀n > N.
n
Since {yn }∞ ∞
n=1 is a bounded sequence, it has a convergent subsequence {ynk }k=1 , and
let
y0 = lim ynk .
k→∞

Since Y is a finite dimensional linear subspace, by Theorem 4.3.2 in Friedman’s book,

we know that Y is closed and y0 ∈ Y .
We need to show that kx0 − y0 k = d.
On one hand, it is obvious that

kx0 − y0 k ≥ d.
∗
E-mail address: [email protected]; Tel : 765 237 7149

I
Yingwei Wang Methods Of Applied Mathematics

On the other hand,

kx0 − y0 k ≤ ky0 − ynk k + kynk − x0 k → d, as k → ∞.

So y0 is just what we want to find.

2 Linear transformation
Question: Find the norm of the operator A ∈ B(X) given by (Af )(t) = tf (t), 0 ≤ t ≤ 1,
where
(a) X = C([0, 1])
(b) X = Lp ([0, 1]), 1 ≤ p < ∞.
Solution:
(a) On one hand,

kAf k = ktf (t)k ≤ ktkkf (t)k = kf (t)k, ∀f ∈ C([0, 1]),

so kAk ≤ 1.
On the other hand, let f ≡ 1, then kf k = 1 and

kAk ≥ ktf (t)k = 1.

So kAk = 1.
(b) On one hand,
  p1
1
kAf kp = ktf (t)kp ≤ ktkp kf (t)kp = kf (t)kp , ∀f ∈ Lp ([0, 1]),
p+1
  p1
1
so kAk ≤ p+1
.
On the other hand, let f ≡ 1, then kf kp = 1 and
  p1
1
kAk ≥ ktf (t)kp = .
p+1
  p1
1
So kAk = p+1
.

II
Yingwei Wang Methods Of Applied Mathematics

3 Linear transformation
Question: A linear operator from a normed linear space X into a normed linear space is
bounded if and only if it maps bounded sets onto bounded sets.
Proof: ⇒): Suppose T : X → Y is a bounded linear operator, then

kT xk ≤ kT kkxk,

which means that if x is bounded then T x should also bounded.

⇐): Choose the closed unit ball B = B1 (0) ⊂ X, then B is bounded. so

α = sup kT xk < ∞.
x∈B

x
Since ∀x ∈ X, kxk
∈ B, so
 

T x ≤α
kxk
⇒ kT xk ≤ αkxk.

It implies that T is bounded.

III
 MA 611: Homework #2
∗
Yingwei Wang
Department of Mathematics, Purdue University, West Lafayette, IN, USA

1 Problem 3.1.5
Question: Prove that the spaces l1 , l∞ , s, c, c0 and C[a, b] are complete metric spaces.
Proof:

1.1 l1
1 (n) (n) P∞ (n)
Let {xn }∞
n=1 ⊂ l be a Cauchy sequence, where xn = (ξ1 , · · · , ξk , · · · ), k=1 |ξk | < ∞.
∀ε > 0, ∃N such that ∀m, n > N,
∞
(n) (m)
X
kxn − xm kl1 = |ξk − ξk | < ε. (1.1)
k=1

(n) 1
It implies that for fixed k, {ξk }∞ n=1 is a Cauchy sequence in R . So we can denote
(n) (0) (0) (0)
lim ξk = ξk and let x0 = (ξ1 , · · · , ξk , · · · ).
n→∞
In Eq.(1.1), let m → ∞, we can get

kxn − x0 kl1 ≤ ε, for n > N, (1.2)

which means lim xn = x0 in the sense of l1 .

n→∞
Besides, since
kx0 kl1 ≤ kx0 − xn kl1 + kxn kl1 < ∞,
we know that x0 ∈ l1 and l1 is complete.
∗
E-mail address: [email protected]; Tel : 765 237 7149

I
Yingwei Wang Methods of Applied Mathematics

1.2 l∞
(n) (n) (n)
Let {xn }∞
n=1 ⊂ l
∞
be a Cauchy sequence, where xn = (ξ1 , · · · , ξk , · · · ), sup |ξk | < ∞.
k
∀ε > 0, ∃N such that ∀m, n > N,
(n) (m)
kxn − xm kl∞ = sup |ξk − ξk | < ε. (1.3)
k

(n) 1
It implies that for fixed k, {ξk }∞ n=1 is a Cauchy sequence in R . So we can denote
(n) (0) (0) (0)
lim ξk = ξk and let x0 = (ξ1 , · · · , ξk , · · · ).
n→∞
In Eq.(1.3), let m → ∞, we can get

kxn − x0 kl∞ ≤ ε, for n > N, (1.4)

which means lim xn = x0 in the sense of l∞ .

n→∞
Besides, since
kx0 kl∞ ≤ kx0 − xn kl∞ + kxn kl∞ < ∞,
we know that x0 ∈ l∞ and l∞ is complete.

1.3 s
(n) (n)
Let {xn }∞
n=1 ⊂ s be a Cauchy sequence, where xn = (ξ1 , · · · , ξk , · · · ) with

∞ (n)
X 1 |ξk |
k (n)
< ∞.
k=1
2 1 + |ξ k |

∀ε > 0,
∞ (n) (m)
X 1 |ξk − ξk |
kxn − xm ks = → 0, as n, m → ∞. (1.5)
k=1
2k 1 + |ξk(n) − ξk(m) |
(n) (m)
So |ξk − ξk | → 0 as n, m → ∞.
(n) 1
It implies that for fixed k, {ξk }∞ n=1 is a Cauchy sequence in R . So we can denote
(n) (0) (0) (0)
lim ξk = ξk and let x0 = (ξ1 , · · · , ξk , · · · ).
n→∞
On one hand, ∀ε > 0, ∃K such that
X 1 ε
k
<
k>K
2 2

II
Yingwei Wang Methods of Applied Mathematics

On the other hand, ∃N such that ∀n > N, ∀k ≤ K,

(n) (0) ε
|ξk − ξk | < .
2
Hence, we have
∞ (n) (0)
X 1 |ξk − ξk |
kx0 − xn ks =
k=1
2k 1 + |ξk(n) − ξk(0) |
K (n) (0) ∞ (n) (0)
X 1 |ξk − ξk | X 1 |ξk − ξk |
= +
2k 1 + |ξ (n) − ξ (0) | 2k 1 + |ξ (n) − ξ (0) |
k=1 k k k=K+1 k k
K ∞
X 1 (n) (0)
X 1
≤ k
|ξk − ξk | +
k=1
2 k=K+1
2k
K
εX 1 ε
< k
+
2 2 2
k=1
< ε,

which means lim xn = x0 in the sense of s.

n→∞
Besides, since
kx0 ks ≤ kx0 − xn ks + kxn ks < ∞,
we know that x0 ∈ s and s is complete.

1.4 c0
(n) (n) (n)
Let {xn }∞
n=1 ⊂ c0 be a Cauchy sequence, where xn = (ξ1 , · · · , ξk , · · · ) with sup |ξk | < ∞
k
(n)
and lim ξk = 0.
k→∞
Since
(n) (m) (n) (m)
kxn − xm kl∞ = sup |ξk − ξk | > |ξk − ξk |, (1.6)
k
(n) 1 (n) (0)
for fixed k, {ξk }∞
n=1 is a Cauchy sequence in R . So we can denote lim ξk = ξk and let
n→∞
(0) (0)
x0 = (ξ1 , · · · , ξk , · · · ).
We need to prove x0 ∈ c0 .
∀ε > 0, ∃N such that
(n) (N ) ε
|ξk − ξk | < , ∀n > N.
2

III
Yingwei Wang Methods of Applied Mathematics

Let n → ∞, we can get

(0) (N ) ε
|ξk − ξk | ≤ .
2
(N ) (N )
Since lim ξk = 0, there exists K > 0 such that ∀k > K, |ξk | < 2ε .
k→∞
So for k > K,
ε ε
|ξk0| ≤ |ξk0 − ξkN | + |ξkN | <
+ < ε.
2 2
It implies that {ξk0 }∞
k=1 is a sequence convergent to 0, which means x0 ∈ c0 .
We also need to prove that lim xn = x0 .
n→∞
∀ε > 0, ∃N such that ∀m, n > N,
(n) (m)
kxn − xm kl∞ = sup |ξk − ξk | < ε. (1.7)
k

Let m → ∞, we can get

kxn − x0 kl∞ ≤ ε, for n > N, (1.8)
which means lim xn = x0 in the sense of l∞ .
n→∞

1.5 c
(n) (n) (n)
Let {xn }∞
n=1 ⊂ c be a Cauchy sequence, where xn = (ξ1 , · · · , ξk , · · · ) with sup |ξk | < ∞
k
and
(n)
lim ξk < ∞, ∀n ∈ N.
k→∞

Given ε > 0, ∃N, such that ∀m, n > N,

sup |ξkn − ξkm | < ε. (1.9)

k

(n) 1 (n) (0)

So for fixed k, {ξk }∞
n=1 is a Cauchy sequence in R . Then we can denote lim ξk = ξk
n→∞
(0) (0)
and let x0 = (ξ1 , · · · , ξk , · · · ).
Let m → ∞ in (1.9), we know that

sup |ξkn − ξk0 | < ε,

k

which means xn → x0 as n → ∞.
(0)
We need to prove x0 ∈ c, that is to say, lim ξk exists.
k→∞
Notice the fact that ∃K, such that ∀p, q > K,

|ξpn − ξqn | < ε.

IV
Yingwei Wang Methods of Applied Mathematics

Then for p, q > K and n > N, we have

kξp0 − ξq0 kl∞ ≤ kξp0 − ξpn kl∞ + kξpn − ξqn kl∞ + kξqn − ξq0 kl∞
≤ 3ε,

which means x0 ∈ c.

1.6 C[a, b]
Let {xn (t)}∞
n=1 be a Cauchy sequence in C[a, b]. Then ∀ε > 0, ∃N such that

max |xn (t) − xm (t)| < ε, ∀n, m > N. (1.10)

t∈[a,b]

For fixed t0 ∈ [a, b], {xn (t)}∞

n=1 is a Cauchy sequence in R. So we can define the func-
tion x0 (t) = lim xn (t). By (1.10) we know that the convergence of {xn (t)}∞
n=1 is uniformly
n→∞
convergent on [a, b].
Hence, ∀ε > 0, ∃N such that

|xN (t) − x0 (t)| < ε, ∀t ∈ [a, b]. (1.11)

Since xN (t) ∈ C[a, b], for t0 ∈ [a, b], there exists a δ > 0o such that ∀t ∈ (t0 − δ, t0 + δ), we
have
|xN (t) − xN (t0 )| < ε. (1.12)
By (1.11) and (1.12), we know that

|x0 (t) − x0 (t0 )| ≤ |x0 (t) − xN (t)| + |xN (t) − xN (t0 )| + |xN (t0 ) − x0 (t0 )|
≤ 3ε,

which means x0 (t) ∈ C[a, b].

2 Problem 3.16
Question: Prove that the spaces l1 , s, c, c0 are separable metric spaces.
Proof:

V
Yingwei Wang Methods of Applied Mathematics

2.1 l1
The set P = {y = (η1 , η2 , · · · , ηn , 0, 0, · · · )|ηk ∈ Q, 1 ≤ k ≤ n} is a countable subset of l1 .
Given ε > 0 and x = (ξ1 , · · · , ξk , · · · ) ∈ l1 , then ∃N such that
∞
X ε
|ξk | < .
k=N +1
2

Q is dense in R, so there is a y ∈ P such that

N
X ε
|ξk − ηk | < .
2
k=1

Hence,
N
X ∞
X
kx − yk =
l1 |ξk − ηk | + |ξk | < ε.
k=1 k=N +1

It implies that P = l1 .

2.2 s
Also consider the set P = {y = (η1 , η2 , · · · , ηn , 0, 0, · · · )|ηk ∈ Q, 1 ≤ k ≤ n} is a countable
subset of s.
t
For x = (ξ1 , · · · , ξk , · · · ) ∈ s, since Q is dense in R and the function ϕ(t) = 1+t is increasing,
∀ε > 0, there is a y ∈ P such that
N
X 1 |ξk − ηk | ε
k
< .
2 1 + |ξk − ηk | 2
k=1

Similarly as Section 2.1, we know that P = s.

2.3 c0
Also, since the set P = {y = (η1 , η2 , · · · , ηn , 0, 0, · · · )|ηk ∈ Q, 1 ≤ k ≤ n} is a countable subset
of c0 .
Given ε > 0 and x = (ξ1 , · · · , ξk , · · · ) ∈ l1 , then ∃N such that
ε
sup |ξk | < .
k≥N +1 2

VI
Yingwei Wang Methods of Applied Mathematics

Q is dense in R, so there is a y ∈ P such that

ε
sup |ξk − ηk | < .
1<k<N 2
Hence,
kx − ykl∞ ≤ sup |ξk − ηk | + sup |ξk | < ε.
1<k<N k≥N +1

It implies that P = c0 . Hence, c0 is separable.

2.4 c
Consider the set P 1 = {y = (η1 , η2 , · · · , ηn , ηn , ηn , · · · )|ηk ∈ Q, 1 ≤ k ≤ n}, which is countable.
Then ∀y = (η1 , η2 , · · · , ηn , ηn , ηn , · · · ) ∈ P 1, lim ηk = ηn , which means P 1 ⊂ c.
k→∞
Following the way in Section 2.3, it is easy to know that P 1 = c.

3 Problem 3.1.7
Question: Prove that the space l∞ is not separable.
Proof:
Consider the set of all sequences made of elements 0, 1 :
S = {x = (ξ1 , · · · , ξk , · · · ) ∈ l∞ |ξk = 0 or 1, for k = 1, 2, · · · }.
It is obvious that S ⊂ l∞ .
ξk
∀x ∈ S can be seen as a binary representation of of a number α = ∞
P
k=1 2k ∈ [0, 1]. Since
the interval [0, 1] is uncountable, S is also uncountable. Besides, if x, y ∈ S and x 6= y, then
kx − ykl∞ = 1, which means there are uncountable many sequences of zeros and ones.
Consider the set made of the balls centered at x ∈ S with radius 13 :
B = {B1/3 (x)|x ∈ S, B1/3 (x) ∩ B1/3 (y) = ∅, if x 6= y}.
It is obvious that B is also uncountable.
Let M be any dense subset of l∞ , then ∀B1/3 (x) ∈ B must contain a element of M. It
implies that B should be a uncountable set. Hence, l∞ is not separable.

4 Problem 3.1.8
Question: Construct, in each of the spaces l1 , l∞ , s, c, c0 , C[a, b], bounded closed set that are
not sequentially compact.
Solution:

VII
Yingwei Wang Methods of Applied Mathematics

4.1 l1, l∞ , c, c0
(n) (n)
Consider the sequence X = {xn }∞
n=1 . ∀xn = (ξ1 , · · · , ξk , · · · ) ∈ X is defined as

(n) 1, k = n,
ξk =
0, k 6= n.

It is obvious that
kxn kl1 = kxn kl∞ = 1.
So X is bounded.
(n)
Besides, for fixed n, lim ξk = 0. So X ⊂ c0 ⊂ c ⊂ l∞ . And it is obvious that X ⊂ l1 .
k→∞
However, X has no convergent subsequence.

4.2 s
Consider the set X = {xn = (n, 0, 0, · · · )}. Then ∀xn ∈ X,
1 n 1
kxn ks = · < ,
2 1+n 2
which means X is bounded.
However,
1
kxn+1 − xn ks = ,
4
which means X does not have convergent subsequence.

4.3 C[a, b]
Without of generality, we can assume a = 0, b = 1, that is to say, we consider the space C[0, 1].
Construct the sequence {fn }∞ n=1 as

1 − nx, x ∈ [0, n1 ],

fn (x) =
0, x ∈ ( n1 , 1].

Then kfn k∞ = 1 so they are bounded. However,


1, x = 0,
lim fn (x) = f (x) =
n→∞ 0, x ∈ (0, 1].

It is obvious that f (x) ∈

/ C[0, 1].

VIII
 MA 611: Homework #3
∗
Yingwei Wang
Department of Mathematics, Purdue University, West Lafayette, IN, USA

1 Problem 3.5.4
Question: A subset F of a compact metric space is compact ⇔ it is closed.
Proof: ”⇒”
Let x ∈ F̄ , then there exists a sequence {xn } ⊂ F , such that xn → x. Since the subset F
is compact in a compact space, by Theorem 3.5.4 in Friedman’s book, F is also sequentially
compact. Hence, x ∈ F and F is closed,
”⇐” Let X be the whole compact metric space and F is the closed subsect of X. For
any open cover F ⊃ F , we can let F1 = F ∪ {X \ F } is an open cover of X. Since X is
compact, so the open cover of X, named F1 has finite subcover. It is easy to know that F
also has finite subcover.

2 Problem 3.5.5
Question: A subset Y of a metric space is totally bounded ⇔ its closure Ȳ is totally
bounded.
Proof: ”⇒”
Y is totally bounded means for ∀ε > 0, there exists a finite number of balls B(xn , ε)
with xn ∈ Y such that
[
N
B(xn , ε) ⊃ Y.
n=1

∗
E-mail address: [email protected]; Tel : 765 237 7149

I
Yingwei Wang Methods of Applied Mathematics

It is obvious that ∪N
n=1 B(xn , ε) ⊃ Ȳ and for each n, B(xn , 2ε) ⊃ B(xn , ε). Then

[
N [
N
B(xn , 2ε) ⊃ B(xn , ε) ⊃ Y.
n=1 n=1

S
So N n=1 B(xn , 2ε) is the cover of Ȳ and Ȳ is totally bounded.
”⇐”
S
For any ε > 0, suppose N n=1 B(xn , ε) be a cover of Ȳ . That is to say,

[
N
B(xn , ε) ⊃ Ȳ ⊃ Y.
n=1

For ∀xn ∈ Ȳ , ∃ yn ∈ Y , such that

d(xn , yn ) < ε.

Then, ∀x ∈ B(xn , ε),

d(x, yn ) ≤ d(x, xn ) + d(xn , yn ) < 2ε,

which means B(xn , ε) ⊂ B(yn , 2ε).

So
[N [
N
B(yn , 2ε) ⊃ B(xn , ε) ⊃ Ȳ ⊃ Y,
n=1 n=1

where yn ∈ Y .
S
Then N n=1 B(yn , 2ε) is a cover of Y and Y is totally bounded.

3 Problem 3.5.9
Question: Prove that a closed set K of points x = (x1 , x2 , · · · ) in lp (1 ≤ p < ∞) is compact
if
P
(i) ∞ p
n=1 |xn | ≤ C for all x ∈ K, where C P
is a constant,
(ii) for any ε > 0, there is an n0 such that ∞ p
n=n0 |xn | ≤ ε for all x ∈ K.
Proof: By Theorem 3.5.6 in Friedman’s book, we just need to show that the set K is
totally bounded if K satisfies (i) and (ii).
(i) means ∀x ∈ K, kxk2p < C, which implies that K is uniformly bounded.

II
Yingwei Wang Methods of Applied Mathematics

P
For any ε > 0, by (ii), we can find n0 such that ∞ p ε
n=n0 |xn | ≤ 2 for all x ∈ K.
Let
Yε = {y ∈ K : y = (y1 , y2 , · · · , yn0 , 0, 0, · · · )}.
(k) ε
Then it is easy to know that Yε is closed and compact so it has a finite cover ∪K
k=1 B(y , 2 ),
(k) (k) (k) (k)
where y = (y1 , y2 , · · · , yn0 , 0, 0, · · · )} ∈ Yε .
Let
(k)
B = ∪K k=1 B(y , ε),
we want to show that B is a cover of K.
For any x ∈ K, by the definition of Yε , there exists y ∈ Yε such that d(x, y) < 2ε . Besides,
there exists some y (k) such that y ∈ B(y (k), 2ε ), i.e. d(y, y (k)) < 2ε . So we have
d(x, y (k)) ≤ d(x, y) + d(y, y (k)) < ε,
which means x ∈ B(y (k) , ε).
Thus, B is a cover of K and K is totally bounded and compact.

4 Problem 4.8.1
Question: Let X be Pan normed linear space and let {xn } ⊂ X. A point y0 is the limit of
linear combination j=1 cj xj ⇔ x (y0 ) = 0 for all x∗ for which x∗ (xj ) = 0 for 1 ≤ j < ∞.
∗

Proof: ”⇒”
P
Let sn = nj=1 cj xj , then
y0 = lim sn .
n→∞

Then,
x∗ (xj ) = 0, ∀j,
Xn
x∗ (sn ) = cj x∗ (xj ) = 0, ∀n,
j=1
x (y) = x ( lim sn ) = lim x∗ (sn ) = 0.
∗ ∗
n→∞ n→∞

”⇐”
Let Y = span {x1 , x2 , · · · , xn , · · · } and y0 ∈ X satisfying that x∗P
(y0 ) = 0 for all x∗ for
which x ∗ (xj ) = 0 for 1 ≤ j < ∞. Then for any y ∈ Y , we have y = nj=1 cj xj and
X
n
x∗ (y) = cj x∗ (xj ) = 0, (4.1)
j=1

III
Yingwei Wang Methods of Applied Mathematics

where x ∗ (xj ) = 0 for 1 ≤ j < ∞.

We claim that d(y0 , Y ) = inf y∈Y ky − y0 k = 0. If not, d(y0 , Y ) > 0, we want to get a
contradiction.
By Theorem
Pn 4.8.3, there exists a point x∗ ∈ X ∗ such that x∗ (y0 ) = 1 and x∗ (y) = 0 for
all y = j=1 cj xj ∈ Y , which contradicts with (4.1).
P
Since we have known that d(y0, Y ) = 0, there should exist a linear combination nj=1 cj xj
whose limit is y0 .

5 Extreme points
Question: Let c0 = {x = (x1 , x2 , · · · ) : lim xn = 0} and kxk∞ = sup |xn |. Let E = {x ∈
n→∞ n
c0 , kxk∞ ≤ 1}, then E has no extreme points.
Proof:
Step 1. Claim that ∀x ∈ E, there exists y, z ∈ E and x 6= y 6= z such that x = 12 x + 21 y.
Let x = (x1 , x2 , · · · ) ∈ E. Since lim xn = 0, there exits N such that for any n > N,
n→∞
|xn | < 21 .
Choose y = (x1 , · · · , xN , 0, · · · ) and z = (x1 , · · · , xN , 2xN +1 , 2xN +2 , · · · ). It is easy to
know that kyk∞ ≤ 1 and kzk∞ ≤ 1, which means y, z ∈ E. Besides,
1 1
x = x + y.
2 2

Step 2. Suppose {x} is the extreme point of E, then x can not be written as x = 21 y + 12 z
where y, z ∈ E and x 6= y, x 6= z. It is contradicts with the claim in Step 1.
Hence, E has no extreme points.

IV
 MA 611: Homework #4
∗
Yingwei Wang
Department of Mathematics, Purdue University, West Lafayette, IN, USA

1 Problem 4.5.2
Question: Show that
2π  sin(n + 12 )x 
Z  
 sin( 1 x)  dx → ∞, as n → ∞ (1.1)
 
0 2

Proof:
2π sin(n + 12 )x 
Z  
 sin( 1 x)  dx
 
0 2
Z 2π  
sin(n + 12 )x
= dx
0 sin( 12 x)
Z 2π  
sin(n + 12 )x
≥ 1 dx
0 2
x
Z (2n+1)π  
| sin y| 1
= 2 dy, where y = n + x
0 y 2
2n R π
X
0
sin ydy
≥ 2
k=0
k+1
2n
X 1
= 2 .
k=1
k
P∞ 1
Since k=1 k = ∞, we can get (1.1).
∗
E-mail address: [email protected]; Tel : 765 237 7149

I
Yingwei Wang Methods of Applied Mathematics

2 Problem 4.5.3
Question: The Fourier series of an integrable function f (x) defined on (0, 2π) is the series
∞
X
s(x) = am eimx ,
m=−∞
1
R 2π
where am = 2π 0
f (ξ)e−imξ dξ.
Set sn (x) = nm=n am eimx , f (x) = f (x + 2π)(0 < x < 2π). Prove that
P
Z 2π
1
sn (y) = f (y + x)Dn (x)dx,
2π 0
where
sin(n + 21 )x
Dn (x) = .
sin( 12 x)
Proof: It is easy to know that
sin(n + 12 )x
Dn (x) =
sin( 12 x)
N
X
=1+2 cos(kx)
k=1
N
X
= eikx . (2.1)
k=−N

By the definition, we know that

Xn
sn (y) = am eimy
m=−n
n Z 2π 
X 1 −imξ
= f (ξ)e dξ eimy
m=−n
2π 0
Z 2π n
!
1 X
im(y−ξ)
= f (ξ) e dξ
2π 0 m=−n
Z 2π n
!
1 X
= f (y + x) eimx dx, where x = y − ξ,
2π 0 m=−n
Z 2π
1
= f (y + x)Dn (x)dx.
2π 0

II
Yingwei Wang Methods of Applied Mathematics

3 Problem 4.5.4
Question: Denote by X the Banach space of all continuous functions f (x) on 0 ≤ x ≤ 2π,
with f (0) = f (2π), provided with the uniform norm. Denote by Y the Euclidean space R1 .
Prove that the linear operator
Z 2π
1
Tn (f ) = f (x)Dn (x)dx
2π 0
from X into Y is bounded, and
2π
1
Z
kTn k = |Dn (x)|dx. (3.1)
2π 0

Proof: On one hand, for each f ∈ X,

Z 2π
1
|Tn (f )| ≤ |f (x)||Dn (x)|dx
2π 0
Z 2π
1
≤ max |f (x)| |Dn (x)|dx
x∈[0,2π] 2π 0
Z 2π
1
= kf k |Dn (x)|dx,
2π 0
which means
2π
1
Z
kTn k ≤ |Dn (x)|dx. (3.2)
2π 0
1
R 2π
On the other hand, we want to prove that kTn k ≥ 2π 0
|Dn (x)|dx.
It is easy to know that 
= 0, x = δn,k ,
Dn (x)
6= 0, otherwise,
kπ
where δn,k = n+ 12
.
Given ε > 0, for each n, construct a continuous function f˜n ∈ X as

sgn(Dn (x)), x = [0, 2π]\,
f˜n (x) =
linear functions, x ∈ ∆,

where ∆ = (∪(δn,k − ε, δn,k + ε)) ∩ [0, 2π].

It is easy to verify that: (1) f˜n (x)Dn (x) ≥ 0, for ∀x ∈ [0, 2π]; (2) |Dn (x)| ≤ ε for ∀x ∈ ∆;
(3) kf˜n k = 1.

III
Yingwei Wang Methods of Applied Mathematics

So we have
Z 2π
1
|Tn (f˜n )| = f˜n (x)Dn (x)dx
2π 0
1 1
Z Z
≥ |Dn (x)|dx − |Dn (x)|dx
2π [0,2π]\∆ 2π ∆
Z 1
1 1
Z
≥ |Dn (x)|dx − 2 |Dn (x)|dx
2π 0 2π ∆
Z 1
1
≥ |Dn (x)|dx − 2ε.
2π 0

Let ε → 0, we know that

2π
1
Z
kTn k ≥ |Dn (x)|dx. (3.3)
2π 0

1
R 2π
By (3.2) and (3.3), we can conclude that Tn (f ) = 2π 0
|Dn (x)|dx.

4 Problem 4.5.5
Question: There exists a continuous function f (x) on 0 ≤≤ x ≤ 2π, with f (0) = f (2π),
such that its Fourier series diverges at x = 0.
Proof: By (1.1) and (3.1), we know that

kTn k → ∞, as n → ∞,

which means {kTn k} is not bounded. By Banach-Steinhaus theorem, there must exist f¯ ∈ X
such that the set {Tn (f)}¯ is not bounded.
2π
1
f¯(x)Dn (x)dx = Tn (f)
¯ → ∞ as n → ∞. That is to say, the Fourier
R
Then sn (0) = 2π 0
series of f¯ diverges at n → ∞.

IV
 MA 611: Homework #5
∗
Yingwei Wang
Department of Mathematics, Purdue University, West Lafayette, IN, USA

1 Differential operator
Question: Let DT = {u(x) ∈ C 1 [−1, 1] : kuk1,2 < ∞} where
Z 1 Z 1 1/2
2 ′ 2
kuk1,2 = u dx + (u ) dx . (1.1)
−1 −1

d
Let T = dx , DT ⊂ L2 [−1, 1], T : DT → L2 [−1, 1].
(a) Prove that T is not bounded.
(b) Prove that DT̄ ( L2 [−1, 1].
Proof:
(a)
Choose fn (x) = sin(nt), since
Z 1 Z 1 1/2
2
kfn (x)k1,2 = |fn | dx + |fn′ |2 dx
−1 −1
Z 1 Z 1 1/2
2 2
= (sin(nt)) dx + (n cos(nt)) dx < ∞, ∀n ∈ N.
−1 −1

we know that fn (x) ∈ DT , for each n.

∗
E-mail address: [email protected]; Tel : 765 237 7149

I
Yingwei Wang Methods of Applied Mathematics

When we consider the operator T : DT → L2 [−1, 1], the norm in DT should be L2 -norm.
2 kT (fn )k2L2
kT k ≥
kfn k2 2
R 1 ′L 2
|f (x)| dx
−1 n
= R1
|f (x)|2 dx
−1 n
R1
−1
(n cos(nt))2 dx
≥ R1
−1
(sin(nt))2 dx
1
n2 −1 (1+cos(2nt)
R
2
dx
≥ R1
−1
1dx
n2 + 12 n sin(2n)
≥ → ∞, as n → ∞.
2
It implies that T can not be bounded.
(b) If DT̄ = L2 [−1, 1], then T should be bounded, which contradicts with the conclusion
in (a).
Actually, DT̄ = Ĉ m [−1, 1] = H 1 [−1, 1] ( L2 [−1, 1].

2 Problem 4.6.5
Question: Let X and Y be Banach spaces and let T be a bounded linear map from X into Y .
If T (X) is of the second category (in Y ), then T (X) = Y .
Proof: Since T (X) is of the second category in Y , without of generality, we can assume
that
∃δ > 0, such that B(0, δ) ⊂ T (B(0, 1)). (2.1)
δ1 y
It is obvious that 0Y ∈ T (X). Let 0 6= y ∈ Y , for ∀δ1 ∈ (0, δ), since kyk
∈ B(0, δ), by (2.1),
we know that
δ1 y
∈ T (B(0, δ)),
kyk
δ1 y
⇒ ∃x ∈ B(0, 1), such that T x = ,
kyk
kyk
⇒ y = T( x),
δ1
⇒ y ∈ T (X).
Thus, T (X) = Y .

II
Yingwei Wang Methods of Applied Mathematics

3 Problem 4.6.6
Question: Let X and Y be a Banach spaces and let T be a linear map from a linear subspace
DT of X into Y . If DT in X and the graph of T in X × Y are closed, then T is bounded, that
is
kT xk ≤ Kkxk, for all x ∈ DT , K constant.
Proof: Since DT is a linear closed subspace in X and X is a Banach space, we know that
DT is also a Banach space.
By the Closed-Graph Theorem (Theorem 4.6.4), if T is a closed linear operator from a
Banach space DT to another Banach space Y , then T should be bounded.

4 Problem 4.6.7
Question: Let X be a normed linear space with any one of two norms, k k1 , k k2 . If kxn k2 → 0
implies kxn k1 → 0, then there is a constant K such that

kxk1 ≤ Kkxk2 , ∀x ∈ X. (4.1)

Proof: Assume that (4.1) is not true, we want to get a contradiction.

Assume that for each n, there exists xn ∈ X such that

kxn k1 ≥ nkxn k2 ,
1 kxn k2
⇒ ≥ .
n kxn k1
xn
Let yn = kxn k1
∈ X, then we have

kyn k2 → 0, but kyn k1 = 1. (4.2)

It contradicts with the fact that ”kxn k2 → 0 implies kxn k1 → 0”.

III
 MA 611: Homework #6
∗
Yingwei Wang
Department of Mathematics, Purdue University, West Lafayette, IN, USA

1 Problem 4.10.1
Question: A sequence {xn } can not have two distinct weak limits.
Proof: Suppose {xn } have two weak limits, x and y, then

x∗ (x) = lim x∗ (xn ) = x∗ (y)

n→∞
⇒ x (x − y) = 0,
∗
∀x∗ ∈ X ∗ .

By Hahn-Banach theorem, we know that x − y ≡ 0. It implies that the weak limit should
be unique.

2 Problem 4.10.2
Question: Let X be a normed linear space and let B be a dense subset of X ∗ . If a sequence
{xn } in X is bounded, and if limn→∞ x∗ (xn ) exists for each x∗ ∈ B, then limn→∞ x∗ (xn )
exists for all x∗ ∈ X ∗ .
Proof: That {xn } is bounded means ∃K > 0, such that

kxn k ≤ K, ∀n.

That B is dense in X ∗ means ∀f ∈ X ∗ , ∀ε > 0, ∃fε ∈ B such that

kf − fε k < ε.
∗
E-mail address: [email protected]; Tel : 765 237 7149

I
Yingwei Wang Methods of Applied Mathematics

That limn→∞ fε (xn ) exists means ∃N such that ∀n, m > N,

|fε (xn ) − fε (xm )| < ε.

Thus, we have

|f (xn ) − f (xm )|
≤ |f (xn ) − fε (xn )| + |fε (xn ) − fε (xm )| + |fε (xm ) − f (xm )|
≤ kf − fε k kxn k + |fε (xn ) − fε (xm )| + kfε − f k kxm k
≤ (2K + 1)ε.

It implies that limn→∞ f (xn ) exists for all f in X ∗ .

II
 MA 611: Homework #7
∗
Yingwei Wang
Department of Mathematics, Purdue University, West Lafayette, IN, USA

1 Problem 4.10.5
Question: A sequentially compact set is weakly sequentially compact, but not con-
versely.
Proof: Suppose A be a sequentially compact set, then any sequence {xn } ⊂ A has
a subsequence {xnk } that (strongly) converges to a point x ∈ A. Then
|f (xnk ) − f (x)| = |f (xnk − x)| ≤ kf kkxnk − xk → 0, ∀f ∈ A∗ .
It implies that {xnk } is weakly convergent.
However, a weakly sequentially compact set may not be sequentially, particularly in
an infinite-dimensional reflexive space. Consider the unit ball in l2 space.
Let B be the closed unit ball in l2 . and xn = {ξn1, ξn2 , · · · , ξnj , · · · } ∈ B, then
∞
X
2
kxn k = |ξnj |2 ≤ 1
j=1
2
⇒ |ξnj | ≤ 1, n, j = 1, 2, · · · .
We can choose a subsequence xnk = {ξnk 1 , ξnk 2 , · · · , ξnk j , · · · } such that
lim ξnk j = ξj , j = 1, 2, · · · .
k→∞

For any N > 0, we have

N
X
|ξnk j |2 ≤ kxnk k ≤ 1.
j=1

∗
E-mail address: [email protected]; Tel : 765 237 7149

I
Yingwei Wang Methods of Applied Mathematics

Let k → ∞, we have
N
X
|ξj |2 ≤ 1
j=1
∞
X
⇒ |ξj |2 ≤ 1
j=1
⇒ x0 = {ξ1 , ξ2, · · · , ξj , · · · , } ∈ B.

Since (l2 )∗ = l2 , for any y = {η1 , η2 , · · · , ηj , · · · } ∈ (l2 )∗ ,

|y(xnk ) − y(x0 )|
= |(xnk − x0 , y)|
 
X ∞ 
=  (ξnk j − ξj )ηj 
 
 
j=1
     
X N   X∞   X ∞ 
≤  (ξnk j − ξj )ηj  +  ξnk j ηj  +  ξj ηj 
     
     
j=1 j=N +1 j=N +1
N ∞ ∞
! ∞
X X X X
2 1/2 2 1/2
≤ |ξnk j − ξj ||ηj | + ( |ξnk j | ) +( |ξj | ) ( |ηj |2 )1/2
j=1 j=N +1 j=N +1 j=N +1
N ∞
!1/2
X X
≤ |ξnk j − ξj ||ηj | + 2 |ηj |2 → 0, as j → ∞ and N sufficient large.
j=1 j=N +1

So {xnk } is weakly convergent.

But for the sequence {en }∞
n=1 ⊂ B where en = (0, · · · , 0, 1, 0, · · · ), 1 in the nth place,
there is no any strongly convergent subsequence.

II
Yingwei Wang Methods of Applied Mathematics

2 Problem 4.10.9
Question: Let X and Y be Banach spaces and let {Tα } be a family of bounded linear
maps from X to Y . If sup |y ∗ (Tα x)| < ∞ for any x ∈ X, then sup kTα k < ∞.
α α
∗ ∗ ∗
Proof: For any fixed y ∈ Y , |y (Tα x)| is bounded. Then, on one hand, if we view
Tα x as an element in the space Y ∗∗ and define Tα x(y ∗ ) = y ∗ (Tα x), by the principle of
uniform boundedness, we know that
sup kTα xkY ∗∗ < ∞. (2.1)
α

Furthermore, since kTα xkY ∗∗ = kTα xkY , we have

sup kTα xkY < ∞. (2.2)
α

On the other hand, by (2.2), we know that {Tα } are pointwise bounded in the space
B(X, Y ), so by the principle of uniform boundedness, we can know that the set {kTα k}
is bounded.

3 Problem 4.10.10
Question: Let X be a normed linearPspace and let {xn } ⊂ X. If {xn } converges weakly
to y, then there exists a sequence { m n
i=1 λi,n xi } that converges to y.
Proof: By the theorem 4.10.7(b), since y is the weak limit of {xn }, then y ∈ Ȳ
where Y = span {x1 , x2 , · · · , xn , · · · }.
It implies that there exits a sequence { m
P n
i=1 λi,n xi } that converges to y.

4 Problem 4.10.11
Question: Let T be a bounded linear operator from a normed linear space X to a
normed linear space Y . If {xn } is a sequence in X that is weakly converges to x0 , then
{T xn } (in Y ) is weakly convergent to T x0 .
Proof: Construct a operator g = y ∗ ◦ T : X → Y → R. Since T ∈ B(X, Y ) and
y ∈ Y ∗ , it is easy to know that g ∈ X ∗ .
∗

Assume that {T xn } (in Y ) is not weakly convergent to T x0 , we want to get a

contradiction.

III
Yingwei Wang Methods of Applied Mathematics

If there exists y ∗ ∈ Y ∗ such that

y ∗ (T xn ) 9 y ∗(T x0 ),
then there should also exist a operator g = y ∗ ◦ T ∈ X ∗ such that
g(xn ) 9 g(x0 ),
which contradicts to the fact that xn ⇀ x0 .
Thus, T (xn ) ⇀ T (x0 ) in Y .

5 Problem 4.10.12
Question: A sequence {xn } in a normed linear space X is weakly convergent to x0 if
and only if the following conditions holds:
(i) the sequence {kxn k} is bounded,
(ii) x∗ (xn ) → x∗ (x0 ) as n → ∞, for any x∗ in a set B dense in X ∗ .
Proof: ⇒:
It is obvious by the theorem 4.10.7.
⇐:
By (i), we know that ∃K > 0 such that
kxn k ≤ K, n = 0, 1, · · · . (5.1)
Since B is dense in X ∗ , we know that ∀x∗ ∈ X ∗ , ∀ε > 0, ∃x∗ε ∈ B such that
kx∗ − x∗ε k < ε. (5.2)
Besides, by (ii), we know that ∃N > 0 such that for ∀n > N,
|x∗ε (xn ) − x∗ε (x0 )| < ε. (5.3)
By Eqs.(5.1)-(5.3), we have
|x∗ (xn ) − x∗ (x0 )|
≤ |x∗ (xn ) − x∗ε (xn )| + |x∗ε (xn ) − x∗ε (x0 )| + |x∗ε (x0 ) − x∗ (x0 )|
≤ kx∗ − x∗ε kkxn k + |x∗ε (xn ) − x∗ε (x0 )| + kx∗ − x∗ε kkx0 k
≤ (2K + 1)ε, ∀n > N.
So xn ⇀ x0 .

IV
Yingwei Wang Methods of Applied Mathematics

6 Problem 4.10.13
Question: Let K be a weakly closed set in a reflexive Banach space X. Then there
exists an element x̄ ∈ K such that

inf kxk = kx̄k.

x∈K

Thus the function f (x) = kxk takes a minimum on K.

Proof: Let m = inf x∈K kxk. Then ∃{xn }∞ n=1 ⊂ K such that

lim kxn k = m.
n→∞

By Theorem 4.10.9, ∃ a subsequence {xnj }∞

j=1 that converges weakly to some element
x̄. Since K is weakly closed, x̄ ∈ K.
Let x∗ ∈ X ∗ , kx∗ k ≤ 1.

x∗ (x̄) = lim x∗ (xnj )

j→∞
⇒ ∗
|x (x̄)| ≤ lim kx∗ kkx̄ − xnj k ≤ m.
j→∞

Since X is a reflexive Banach space, we can view x̄ as an element in X ∗∗ , then

kx̄k = sup |x∗ (x̄)| ≤ m.

kxk≤1

But on the other hand, by the definition of m, kx̄k ≥ m. Thus,

inf kxk = kx̄k.

x∈K

7 Problem 4.10.14
Question: (a) Prove that the unit ball K of a normed linear space X is weakly closed.
(b) Prove also that if the norm is strictly convex, then the minimum of f (x) = kxk
on K is attained at one point only.

Definition 7.1. A norm k k is said to be strictly convex if kxk = 1, kyk = 1, kx+yk = 2

imply that x = y.

V
Yingwei Wang Methods of Applied Mathematics

Proof:
(a) Let the sequence {xn } ∈ K and xn ⇀ x0 , then by Theorem 4.10.7,

kx0 k ≤ lim kxn k ≤ 1.

n→∞

So kx0 k ∈ K. It implies that any weakly convergent sequence in K has its weak limit
also in K, which means K is weakly closed.
(b) Let m = inf x∈K kxk. According to Section 6, there exists at least one point
x1 ∈ K such that kx1 k = m. If there were another element x2 ∈ K such that kx2 k = m,
we need to prove that x2 ≡ x1 .
Let x0 = x1 +x
2
2
∈ K, then

kx1 k kx2 k
m ≤ kx0 k ≤ + = m,
2 2
⇒ kx0 k = m,
⇒ kx1 + x2 k = k2x0 k = 2kx0 k = 2m. (7.1)

Without loss of generality, we can assume that m > 0. By (7.1), we can get
x
1 x2

+ = 1. (7.2)
m m
Besides,
x
1
= 1, (7.3)
m
x
1
= 1. (7.4)
m
By the definition 7.1 and (7.2)-(7.4), we can conclude that x1 = x2 .

VI
 MA 611: Homework #8
∗
Yingwei Wang
Department of Mathematics, Purdue University, West Lafayette, IN, USA

1 Norm of strictly convex

Definition 1.1. A norm k k is said to be strictly convex if kxk ≤ 1, kyk ≤ 1, x 6= y
imply that k x+y
2
k < 1.
Remark 1.1. Actually, if kxk < 1, kyk < 1, then it is sure that k x+y 2
k < 1. Besides,
if kxk = 1, kyk = 1, then it is sure that k x+y
2
k ≤ 1
2
kxk + 1
2
kyk ≤ 1.
So in order to prove a norm k k is strictly convex, we just need to show that
x+y x+y
k k=
6 1, (or k k < 1), for kxk = 1, kyk = 1.
2 2

1.1 (a)
Question: Give an example of a norm on R2 that is not strictly convex.
Solution: Consider the l1 norm in the space R2 . ∀x = (x1 , x2 ) ∈ R2 , the l1 norm of
x is
kxk = |x1 | + |x2 |.
x+y
Choose x = (1, 0) and y = (0, 1), then z = 2
= ( 12 , 12 ). Then we have

kxk = 1, (1.1)
kyk = 1, (1.2)

x + y
kzk =
2 = 1,
(1.3)

∗
E-mail address: [email protected]; Tel : 765 237 7149

I
Yingwei Wang Methods of Applied Mathematics

which contradicts with the definition 1.1.

So the l1 norm in R2 is not strictly convex.

1.2 (b)
Question: Prove X is strictly convex if and only if for each f ∈ X ′ , there exists at most
one x with kxk = 1 so that f (x) = kf k.
Proof: ⇒): Suppose X is strictly convex but ∃f ∈ X ′ such that there are x 6= y ∈ X
satisfying kxk = 1, kyk = 1 and f (x) = kf k, f (y) = kf k. Then, on one hand,
 
x+y 1 1
f = f (x) + f (y) = kf k. (1.4)
2 2 2

On the other hand,   


f x + y  x + y
 ≤ kf k
2 . (1.5)

 2 

Combining (1.4) and (1.5), we know that

x + y
2 ≥ 1,

which contradicts with the definition (1.1).

Thus, such f does not exists.
⇐): Assume that for each f ∈ X ′ , there exists at most one x with kxk = 1 so that
f (x) = kf k.
Let kxk = 1, kyk = 1, x 6= y and z = x+y 2
. In order to prove the norm on X is
strictly convex, we need to show that

kzk < 1. (1.6)

By Corollary 4.8.4 in Friedman’s book, for z, there exists z ∗ ∈ X ′ such that

z ∗ (z) = kzk, kz ∗ k = 1.
z z
Let z1 = kzk and z2 = − kzk , then it is easy to show that kz1 k = kz2 k = 1 and
∗ ∗ ∗
|z (z1 )| = |z (z2 )| = 1 = kz k.

II
Yingwei Wang Methods of Applied Mathematics

Consider ∀w ∈ X satisfying kwk = 1 and w 6= z1 , w 6= z2 . On on hand, by the

assumption, we know that
|z ∗ (w)| =
6 1. (1.7)
On the other hand,

1 = kz ∗ k = sup |z ∗ (w)| = 1 ⇒ |z ∗ (w)| ≤ 1. (1.8)

kwk=1

By (1.7) and (1.8), we know that

|z ∗ (w)| < 1. (1.9)

Since x 6= y, then x, y 6= z1 , z2 , by (1.9),

|z ∗ (x)| < 1, |z ∗ (y)| < 1. (1.10)

So we have
1
kzk = |z ∗ (z)| = |z ∗ (x/2) + z ∗ (y/2)| ≤ (|z ∗ (x)| + |z ∗ (y)|) < 1.
2

1.3 (c)
Question: Prove that X is strictly convex if and only if for each z ∈ X, and closed
nonempty convex subset C, there exists at most one x ∈ C such that

kx − zk = inf ky − zk. (1.11)

y∈C

Proof: ⇒): Suppose X is strictly convex.

Let m = inf y∈C ky − zk and x1 , x2 ∈ C such that

kx1 − zk = kx2 − zk = m.
x1 +x2
Let x0 = 2
. Since C is convex, then x0 ∈ C.

kx1 − zk kx2 − zk
m ≤ kx0 − zk = + =m
2 2
⇒ kx0 − zk = m. (1.12)

III
Yingwei Wang Methods of Applied Mathematics

Without loss of generality, assume m > 0. Let y1 = x1m−z , y2 = x2m−z , then ky1k = 1,
ky2k = 1. Besides, if x1 6= x2 , then y1 6= y2 . We want to get a contradiction.
On one hand, by Eq.(1.12), we know that

y1 + y2 x0 − z
2 = m = 1. (1.13)

On the other hand, since X is strictly convex, we know that

y1 + y2
2 < 1. (1.14)

It is obviously that Eq.(1.13) contradicts with Eq.(1.14).

⇐): Assume that for each z ∈ X, and closed nonempty convex subset C, there
exists at most one x ∈ C such that (1.11) holds.
Choose C as the closed unit ball in X. ∀x 6= y, kxk = 1, kyk = 1, if x+y

= 1, we
2
want to get a contradiction.
Let z = x + y, then kzk = 2. ∀w ∈ C, kwk ≤ 1, we know that

kz − wk ≥ |kzk − kwk| ≥ 1,

which implies
inf kz − wk ≥ 1. (1.15)
w∈C

However, for x, y ∈ C,

kz − xk = kyk = 1, (1.16)
kz − yk = kxk = 1. (1.17)

By (1.15), (1.16) and (1.17), there exist at least two distinguished point x, y in C,
such that
kz − xk = kz − yk = inf kz − wk = 1. (1.18)
w∈C

This contradicts with the assumption.

IV
Yingwei Wang Methods of Applied Mathematics

2 Weak * sequentially compactness

Question: Prove that BX ′ is weak * sequentially compact if X is separable.
Proof:

2.1 Introduce a lemma

Lemma 2.1. Let X be a normed linear space and G be a dense subset of X. If the
sequence {fn }∞ ∗
n=1 ⊂ X satisfy these two conditions:
(i) {kfn k}∞ n=1 is bounded;
(ii) fn (x) → f0 (x), as n → ∞, for any x in the set G (dense in X), where f0 ∈ X ∗ ;
then {fn }∞ n=1 is weakly * convergent to f0 .

I want to use this lemma to prove this problem. And a short proof of this lemma
will be given in the end.

2.2 Proof of this problem

Let {fn }∞
n=1 ⊂ BX ′ , which means ∃M > 0 such that

kfn k ≤ M, ∀n.

Since X is separable, we can find a dense subset of X, that is {xk }∞

k=1 .
Now we can use the so called ”Cantor-Hilbert diagonal method.”
Since |fn (x1 )| ≤ kfn kkx1 k ≤ Mkx1 k, we know that the sequence {fn (x1 )}∞
n=1 is
∞
bounded, so we can find a convergent subsequence from {fn (x1 )}n=1, which is
(1) (1)
f1 (x1 ), f2 (x1 ), · · · , fn(1) (x1 ), · · · .
(1)
Then consider the sequence {fn (x2 )}∞
n=1 , then similarly, we can also find a conver-
gent subsequence, which is
(2) (2)
f1 (x2 ), f2 (x2 ), · · · , fn(2) (x2 ), · · · .

V
Yingwei Wang Methods of Applied Mathematics

Keep doing this procedure, we can find this matrix of functionals on X,

(1) (1) (1)
f1 , f2 , f3 , · · · , fn(1) , · · ·
(2) (2) (2)
f1 , f2 , f3 , · · · , fn(2) , · · ·
······
(n) (n) (n)
f1 , f2 , f3 , · · · , fn(n) , · · ·
······
(j) (j−1) ∞
in which {fn }∞
n=1 is the subsequence of {fn }n=1 and for each k, the sequence
(k) (k)
f1 (xk ), f2 (xk ), · · · , fn(k) (xk ), · · · .

is convergent.
Take the diagonal of the matrix, we can get
(1) (2) (3)
f1 , f2 , f3 , · · · , fn(n) , · · ·
(n)
It is easy to know that for each k, {fn (xk )}∞
n=1 is convergent. That is to say,
(n) ∞
{fn }n=1 is bounded as well as convergent on a dense subset of X, so it should have a
weak * limit in X ∗ according to the Lemma 2.1.
Thus, BX ∗ is weak * sequentially compact.

2.3 Short proof of Lemma 2.1

By the assumption, we know that ∃M = supn {kfn k} > 0; for ∀ε > 0, ∀x ∈ X, ∃x0 ∈ G
ε
such that kx − x0 k < 4M ; ∃N such that ∀n > N, |fn (x0 ) − f0 (x0 )| < 2ε . So we have

|fn (x) − f0 (x)|

≤ |fn (x) − fn (x0 )| + |fn (x0 ) − f0 (x0 )| + |f0 (x0 ) − f0 (x)|
≤ kfn kkx − x0 k + |fn (x0 ) − f0 (x0 )| + kf0 kkx − x0 k
ε ε ε
≤ M + +M
4M 2 4M
= ε.

VI
 MA 611: Homework #9
∗
Yingwei Wang
Department of Mathematics, Purdue University, West Lafayette, IN, USA

1 Ex. 4.13.1
Question: Chose a basis {e1 , e2 , · · · , en } in Rn and choose a basis {e∗1 , · · · , e∗n } in
Rn∗ . To n
Pnevery linear operator T in B(R , R ) corresponds
n
a matrix (aij ), where
n∗ n∗
T ej = i=1 aij ei . Similarly,
Pn ∗ ∗ to every operator S ∈ B(R , R ) corresponds a matrix
∗ ∗
(aij ), where Sej = i=1 aij ei . Find the relation between (aij ) and (a∗ij ) if S = T ∗ .
Solution: Let A = (aij )n×n , A∗ = (a∗ij )n×n .
P
∀x = (x1 , · · · , xn )T ∈ Rn , x = ni=1 xi ei , then T x = Ax.
P
∀y ∗ = (y1 , · · · , yn )T ∈ Rn∗ , y ∗ = ni=1 yi e∗i , then Sy ∗ = A∗ y ∗.
On one hand,
X
n X
n X
n
Sy ∗ (x) = T ∗ y ∗ (x) = y ∗ (T x) =< y ∗ , Ax >= yi aij xj = aij xj yi .
i=1 j=1 i,j=1

On the other hand,

X
n X
n X
n
∗ ∗ ∗ ∗ ∗
Sy (x) = A y (x) =< A y , x >= xi a∗ij yj = a∗ij xi yj .
i=1 j=1 i,j=1

It is easy to know that

aij = a∗ji ,
which means A∗ = AT .
∗
E-mail address: [email protected]; Tel : 765 237 7149

I
Yingwei Wang Methods of Applied Mathematics

2 Ex. 4.13.3
Question: Let X and Y be Banach space and T ∈ B(X, Y ). Prove
(a) If T has a continuous inverse, then RT is closed.
(b) T ∗ is one-to-one ⇔ if RT is dense in Y .
(c) If T maps X onto Y , then T ∗ has a bounded inverse with domain RT ∗ .
Proof:
(a) Let {yn }∞n=1 ⊂ RT be a Cauchy sequence in Y . Since Y is a Banach space,
then ∃y ∈ Y such that
y = lim yn .
n→∞

In order to prove that RT is closed in Y , we need to show that y ∈ RT . That is

to say, to find x ∈ X such that y = T x.
Since T −1 ∈ B(X, Y ), we know that xn = T −1 (yn ) ∈ X, then

kxn − xm k = kT −1 (yn − ym )k ≤ kT −1kkyn − ym k,

which means {xn }∞

n=1 is a Cauchy sequence in X. Since X is a Banach space, we can
find x ∈ X such that
x = lim xn .
n→∞

Furthermore, since T ∈ B(X, Y ),

T x = lim T xn = lim yn = y.
n→∞ n→∞

We are done.
(b) ⇐): By Theorem 4.13.5,

NT⊥∗ = RT = Y,

which means
∀y ∈ Y, ∀y ∗ ∈ NT ∗ , y ∗(y) = 0. (2.1)
Assume the T ∗ is not one-to-one, then ∃y0∗ 6= 0 such that y0∗ ∈ NT ∗ .
So ∃y ∈ Y such that y0∗ (y) 6= 0, which contradicts with (2.1).
⇒): Assume RT 6= Y , then by Theorem 4.13.5, NT⊥∗ 6= Y .

II
Yingwei Wang Methods of Applied Mathematics

Then ∃y0 ∈ Y such that y0 ∈ / NT⊥∗ . So ∃y0∗ ∈ NT ∗ such that y0∗(y0 ) 6= 0. It implies
that NT ∗ 6= {0}, which means T ∗ is not one-to-one.
(c) If RT = Y , by (b), we know that T ∗ is one-to-one, then it has inverse (T ∗ )−1
with domain RT ∗ . It is easy to know that (T ∗ )−1 is also bounded.

3 Ex. 4.13.4
Question: Let X and Y be normed linear space and let T ∈ B(X, Y ). Denote d(x, L)
be the distance (in X) from x to a set L. Prove that
d(x, NT ) = max{x∗ (x) : kx∗ k ≤ 1 and x∗ ∈ NT⊥ }. (3.1)

Proof: On one hand, ∀y ∈ NT , ∀x∗ ∈ NT⊥ and kx∗ k ≤ 1,

x∗ (x) = x∗ (x) − x∗ (y) = x∗ (x − y) ≤ kx∗ kkx − yk ≤ kx − yk,
⇒ max{x∗ (x) : kx∗ k ≤ 1 and x∗ ∈ NT⊥ } ≤ d(x, NT ). (3.2)

On the other hand, by the Theorem, 4.8.3, for given x ∈ X, we can find a x∗ such
that
x∗ (x) = d(x, NT ), kx∗ k = 1, and x∗ |NT = 0. (3.3)
By (3.2) and (3.2), we can get (3.1).

4 Ex. 4.13.5
Question: Let X and Y be Banach space and T ∈ B(X, Y ). If
RT ∗ = NT⊥ (4.1)
holds, then RT is closed.
Proof: Step One: Claim that ∃K > 0 such that
d(x, NT ) ≤ KkT xk. (4.2)

For ∀x∗ ∈ NT⊥ and kx∗ k ≤ 1, by (4.1), we know that there exists a y ∗ and K > 0
such that T ∗ y ∗ = x∗ , and
ky ∗k ≤ Kkx∗ k ≤ K.

III
Yingwei Wang Methods of Applied Mathematics

So for ∀x ∈ X,

x∗ (x) = T ∗ y ∗ (x) = y ∗ (T x) ≤ ky ∗kkT xk ≤ KkT xk

⇒ d(x, NT ) ≤ KkT xk.

Step Two: By (4.2), we can show that RT is closed.

Let {yn }∞
n=1 ⊂ RT be a Cauchy sequence in Y . Since Y is a Banach space, then
∃y ∈ Y such that
y = lim yn .
n→∞

In order to prove that RT is closed in Y , we need to show that y ∈ RT . That is

to say, to find x ∈ X such that y = T x.
For each yn ∈ RT , we can find xn ∈ X such that T xn = yn . By (4.2), we know
that
d(xm − xn , NT ) ≤ KkT xn − T xm k = Kkyn − ym k.
Since {yn }∞ ∞
n=1 is a Cauchy sequence in Y , we know that {xn }n=1 has a convergent
subsequence in X; without loss of generality, we also denote this subsequence as
{xn }∞
n=1 . Besides, X is a Banach space, we can find x ∈ X such that

x = lim xn .
n→∞

Furthermore, since T ∈ B(X, Y ), we can get

T x = lim T xn = lim yn = y.
n→∞ n→∞

It implies that RT is closed.

IV
 MA 611: Homework #10
∗
Yingwei Wang
Department of Mathematics, Purdue University, West Lafayette, IN, USA

1 Ex. 5.1.3
Question: Let (X, µ) and (Y, ν) be measure spaces and consider the operator A
defined by: Z
(Af )(x) = K(x, y)f (y)dν(y) (1.1)

from Lr (Y, ν) into Lp (X, µ) where 1/p + 1/r = 1, 1 < p < ∞.

Assume that X and Y are compact metric spaces, that open sets are measurable
in X and Y , that µ, ν are finite measures and that K(x, y) is continuous on X × Y .
Prove that A is compact linear operator from Lr (Y, ν) into Lp (X, µ).
Proof: Let B1 be the unit ball in the space Lr (Y, ν). We just need to show that
A(B1 ) is compact.
∀f ∈ B1 , kf kLr ≤ 1, Af ∈ A(B1).
By assumption, ∃M1 , M2 > 0 such that

M1 = max |K(x, y)|, M2 = µ(X)ν(Y ).

x,y∈X×Y

So we have

|Af (x)| ≤ max |K(x, y)|µ(X)ν(Y )kf kLr .

x,y∈X×Y
⇒ |Af (x)| ≤ M1 M2 (1.2)
∗
E-mail address: [email protected]; Tel : 765 237 7149

I
Yingwei Wang Methods of Applied Mathematics

For ∀ε > 0, ∃δ > 0, such that ∀y ∈ Y , |x − x′ | < δ,

|K(x, y) − K(x′ , y)| < ε,
Z
⇒ |Af (x) − Af (x )| ≤ kK(x, y) − K(x′ , y)k|f (y)|dy < εM2 kf kLr . (1.3)
′

Now we know that the family of functions {Af } : f ∈ B1 is uniformly bounded

(1.2) and equicontinuous (1.3) in Lp (X, µ), so by Arzela-Ascoli Theorem (Theorem
3.6.4), any sequence in this family has a subsequence that is uniformly convergent.
It implies that A(B1 ) is compact.

2 Ex. 5.1.4
Question: Let Ω1 , Ω2 , be bounded closed sets in Rn and X = Ω1 , Y = Ω2 . Denote
by µ the Lebesgue measure. If K(x, y) ∈ Lp (X × Y, µ × µ), then A, defined in
(1.1), is compact.
Proof: Let A be the operator defined in (1.1) where K(x, y) ∈ Lp (X × Y, µ × µ).
Choose a sequence {Kn (x, y)}∞ n=1 such that

kKn (x, y) − K(x, y)kLp (X×Y ) → 0, as n → ∞. (2.1)

Define a sequence {An }∞

n=1 as
Z
(An f )(x) = Kn (x, y)f (y)dy. (2.2)

By the conclusion of last problem, we know that each An is compact.

Besides,
Z
|An f − Af | = | (Kn (x, y) − K(x, y))f (y)dy| ≤ kK − Kn kLp (Y ) kf (y)kLr ,
⇒ kAn − AkLp ≤ kKn − KkLp (X×Y ) kf kLr ,
kAn − Af kLp
⇒ kAn − Ak = sup ≤ kKn − KkLp → 0, as n → ∞.
kf kLr
It implies that {An } is uniformly convergent to A. By Theorem 5.1.2, we know
that A is also compact.

II
Yingwei Wang Methods of Applied Mathematics

3 Ex. 5.2.1
Question: Let X be a compact set in Rn and let µ denote the Lebesgue measure.
Let K(x, y) ∈ L2 (X × Y, µ × µ). For any given g ∈ L2 (X, µ), consider the equation,
in L2 (X, µ), Z
f (x) = g(x) + λ K(x, y)f (y)dy. (3.1)

Prove that if g = 0 implies f = 0, then there exists a unique solution of (3.1) for
any g ∈ L2 (X, µ).
Proof: Step 1: Existence.
R
Define the operator T : f (y) 7→ K(x, y)f (y)dy, then

g = (I − λT )f,

where I is the identify operator.

It is easy to know that for λ 6= 0,
1
Nλ = {f : λf − T f = 0} = {f : f − λT f = (λf − T f ) = 0},
λ
and
1
Rλ = {g : g = λf − T f } = {g : g = f − λT f = (λf − T f )},
λ
if Rλ = L2 .
The assumption that g = 0 implies f = 0 tells us that Nλ = 0. By Theorem
5.2.7, Rλ = L2 , which means ∀g ∈ L2 , there exists at least f ∈ L2 such that
g = (I − λT )f .
Step 2: Uniqueness.
Suppose for g ∈ L2 , ∃f1 , f2 ∈ L2 such that

g = (I − λT )f1 = (I − λT )f2 ,
⇒ (f1 − f2 ) − λT (f1 − f2 ) = 0,
⇒ f1 − f2 = 0.

III