Royal University of Phnom Penh

Faculty of Science
Department of Mathematics

Lecture Notes

Introduction to Functional Analysis

MEAS Len

©2022

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Contents

Preface 1

1 Foundations 3
1.1 Lebesgue Integral 3
1.2 Metric Space 27
1.3 Banach Space 46
1.4 Bounded Linear Operators 54
1.5 Hilbert Space 65
1.6 Dual Space 72

2 Main Theorems 77
2.1 Baire’s Category Theorem 77
2.2 Uniform Boundedness Principle 79
2.3 Open Mapping Theorem 84
2.4 Closed Graph Theorem 85
2.5 Hahn-Banach Theorem 87

3 Compact Operators 97
3.1 Compact operators 97
3.2 Spectral theorem for self-adjoint compact linear operators 104

References 106

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

ii

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Preface

The aim of this book is to build a foundation of functional analysis for undergraduate and the
beginning graduate students in science, engineering and mathematics who are interested in the
research in analysis.

Functional analysis is one of the important fields of mathematics. It is an extension of linear

algebra which deals with linear spaces (spaces of functions, and usually infinite dimensional
spaces). It plays an important role in the applied sciences as well as mathematics. The impetus
come from applications: ordinary differential equations, partial differential equations, integral
equations, calculus of variations, numerical analysis, optimization, harmonic analysis, approx-
imation theory and so on.

The focus of the first part of this book is on the presentation of the fundamental concepts in
functional analysis: Banach space, linear operator, Hilbert space and dual space. The emphasis
in the second part is on the main results such as uniform boundedness principle, open mapping
theorem, closed graph theorem and Hahn-Banach theorem. The third part is devoted to the
concept of compact operators which play a crucial role in the theory of integral equations and
partial differential equations and in mathematical physics. The relation of compactness with
weak convergence and reflexivity, the spectral properties of self-adjoint compact linear oper-
ators are hightlighted. Specifically, the spectral theorem for self-adjoint comapct operator is
presented.

Without the wonders of LaTeX, this book would never have been prepared. I can only hope
that the errors are unintentional. For this reason, I am particularly interested in receiving cor-
rections, comments and suggestions via email: [email protected]

Phnom Penh, Cambodia

MEAS Len

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Chapter 1

Foundations

This chapter will focus on the fundamental concepts in functional analysis: Banach space,
linear operator and dual space. The concepts of continuity, connectedness and compactness
will also be covered. In addition, the distinction features of finite and infinite dimensional
normed spaces will also be discussed.

1.1 Lebesgue Integral

This section concerns the Lebesgue integral which a powerful extension of the Riemann inte-
gral. One of the most important examples of Banach space is the space of Lebesgue p integrable
functions. In particular, the space of the Lebesgue square integrable functions is a Hilbert space.

Lebesgue Outer Measure (generalization of volume or area)

In Rn , we define a closed interval to be a closed rectangle I where

I = {x = (x1 , x2 , . . . , xn ) ∈ Rn | ai ≤ xi ≤ bi for i = 1, 2, . . . , n}.

Example 1.1

I = {(x1 , x2 ) ∈ R2 | 1 ≤ x1 ≤ 3, −2 ≤ x2 ≤ 4}
is a closed interval in R2 , while

J = {x ∈ R1 | 2 ≤ x ≤ 6}

is a closed interval in R1 .

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

The volume of a closed interval in Rn is v(I) = ni=1 (bi − ai ), where the constants ai and bi are
Q
all finite, and that ai < bi for all i.
More precisely, let S = {Ik } be a countable (finite or countably infinite) collection of closed
intervals in Rn . We say that S is a covering of A by closed intervals if A ⊆ ∪Ik . We set
σ(S ) = v(Ik ). If the series v(Ik ) diverges, set σ(S ) = +∞.
P P

Definition 1.1

Let A ⊆ Rn . The Lebesgue outer measure of A is

m∗ (A) = inf σ(S ) | S is a covering of A by closed intervals .



By Definition 1, m∗ (A) is always greater or equal to 0. Also, it follows that if S is any covering
of A by closed intervals, then m∗ (A) ≤ σ(S ) ≤ +∞. In other words, for any set A ⊆ Rn , 0 ≤
m∗ (A) ≤ +∞.

Remark 1.1

An important feature of the definition of Lebesgue outer measure is that given any set A
and any given ϵ > 0, there is a covering S of A by closed intervals such that

σ(S ) ≤ m∗ (A) + ϵ.

Example 1.2

Compute the Lebesgue outer measure of A = {3}.

Let ϵ > 0. Set S = {[3 − ϵ, 3 + ϵ]}. Thus,

0 ≤ m∗ (A) ≤ σ(S ) = 2ϵ.

Since ϵ was arbitrary, it follows that m∗ (A) = 0.

Example 1.3

The Lebesgue outer measure of ∅ is 0. To see this, let ϵ > 0 be given. Then S = {[−ϵ, ϵ]}
is a covering of ∅ by closed intervals. Therefore,

m∗ (∅) ≤ σ(S ) = 2ϵ

Since ϵ was arbitrary, it follows that m∗ (∅) = 0.

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Property 1.1: Lebesgue Outer Measure

1. If A ⊆ B ⊆ Rn , then m∗ (A) ≤ m∗ (B).

2. For any two sets A and B,

m∗ (A ∪ B) ≤ m∗ (A) + m∗ (B).

3. For any countable collection of sets {An },

X
m∗ (∪An ) ≤ m∗ (An ).

4. If A ⊆ B ⊆ Rn , and m∗ (B) is finite, then

m∗ (B) − m∗ (A) = m∗ (B \ A).

Propoistion 1.1

For any closed interval I ⊆ Rn , m∗ (I) = v(I).

Example 1.4

Let A = [0, 1]. The Lebesgue outer measure of A is 1. (This is just the length of interval
[0, 1]).

Example 1.5

Compute the Lebesgue outer measure of B = [−1, 2] ∪ {3}.

We have
m∗ (B) ≤ m∗ ([−1, 2]) + m∗ ({3})
Since m∗ ([−1, 2]) = 2 − (−1) = 3 and m∗ ({3}) = 0. Thus

m∗ (B) ≤ 3.

On the other hand, [−1, 2] ⊆ B. Hence,

3 = m∗ ([−1, 2]) ≤ m∗ (B).

Consequently, m∗ (B) = 3.

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Definition 1.2: Lebesgue Measure

A set E ⊆ Rn is Lebesgue measurable if for every ϵ > 0 there is an open set G so that
E ⊆ G and
m∗ (G \ E) < ϵ.
In this case, we define the Lebesgue measure of E, denoted by m(E), to be

m(E) = m∗ (E).

Example 1.6

We will show that E = {3} is Lebesgue measurable.

Given ϵ > 0, let G = (3 − 3ϵ , 3 + 3ϵ ). Then we have
 ϵ   ϵ 
m (G \ E) = m 3 − , 3 ∪ 3, 3 +
∗ ∗
3 3
h ϵ ϵ i
≤ m 3 − ,3 +
∗
3 3
2ϵ
= < ϵ.
3

Example 1.7

If G is open set, then m∗ (G \ G) = m∗ (∅) = 0 < ϵ for every ϵ > 0. Consequently, every
open set in Rn is Lebesgue measurable.

Theorem 1.1

Let {Ek } be a countable collection of Lebesgue measurable sets. Then

[
E= Ek

is Lebesgue measurable and X

m(E) ≤ m(Ek ).

As a consequence:
One has any intervals are Lebesgue measurable and m(I) = m∗ (I) = v(I).

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Property 1.2: Lebesgue Measure

M
Let {In }n=1 be a finite collection of pairwise non-overlapping closed intervals. Then
M  M
[  X
m  In  =
  v(In ).
n=1 n=1

In general, every open set G ⊆ Rn can be written as a countably infinite union of non-overlapping
closed intervals G = ∞
S
k=1 Ik with
X∞
m(G) = v(Ik ).
k=1
n
What are Lebesgue measurable subsets of R ?
Lebesgue measurable subsets of Rn are

1. The set with zero outer measure

2. Open sets

3. Closed sets

4. Intervals

5. Finite union of these sets.

6. Let E ⊆ Rn . If E is Lebesgue measurable, then its complement E c is Lebesgue measur-

able.

7. Let {A j } be a countable collection of Lebesgue measurable subsets of Rn . Then the set

\
A= Aj

is Lebesgue measurable.

A nonmeasurable subset of R or Rn can be constructed as the following result.

Theorem 1.2: Vitali

There exist nonmeasurable sets.

For the proof see [8].

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Theorem 1.3

Let {Ek } be a countable collection of pairwise disjoint Lebesgue measurable subsets of Rn .

Then [  X
m Ek = m(Ek ).

It says that the measure of the union of disjoint Lebesgue measurable sets is the sum of mea-
sures.
Remark 1.2

Lebesgue measure on Rn is an extension of the notions about length, area or volume,

depending on the dimension n.

Definition 1.3: General measure theory

Let X be a nonempty set. A collection M of subsets of X is a σ-algebra of sets in X if

(i). X ∈ M.

(ii). A \ B ∈ M whenever A, B ∈ M.
∞
S
(iii). An ∈ M whenever (An )n≥1 ⊂ M.
n=1

Example 1.8

Let X be any set. Then the power set of X, P(X) is a σ-algebra.

Example 1.9

Let X = R. Borel σ-algebra of R, B is the smallest σ-algebra which contains all the open
subsets of R.
Borel sets: (a, b), [a, b], {a}, (a, b], [a, b)
∞ !
\ 1
(a, b) = a, b +
n=1
n
∞ !
\ 1 1
{a} = a − ,a +
n=1
n n

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

∞ !
\ 1 1
[a, b] = a − ,b + .
n=1
n n

Definition 1.4

Let M be a σ-algebra in X. A positive measure on M is a function µ : M → [0, ∞] which

satisfies

a µ(∅) = 0.

b µ is countably additive: If (An )n≥1 in M is a pairwise disjoint collection of sets, then

∞  ∞
[  X
µ  An  = µ(An ).
n=1 n=1

The triple (X, M, µ) is called a measure space on X.

Example 1.10: Lebesgue measure on real line

The triple ([a, b], L([a, b]), m) is a measure space m : L([a, b]) → [0, ∞) by

1. m(∅) = 0;
N
2. if A =
S
In disjoint uniform of intervals,
n=1

N  N
[  X
m  In  =
  l(In )
n=1 n=1

where l(In ) : length of In .

Notation: L([a, b]) ≡ collection of all subsets of [a, b], which are Lebesgue measur-
able.

Example 1.11

Let (X, M, µ) be a measure space with µ(X) = 1 (probability space, µ: probability mea-
sure).

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Theorem 1.4: Properties of Measure

Let M be a σ-algebra on a set X and let µ be a positive measure on M.

(a). Monotonicity: If A, B ∈ M with A ⊆ B, then µ(A) ≤ µ(B).

(b). Continuity for increasing union: Suppose (An )n≥1 ⊂ M

∞
with A1 ⊆ A2 ⊆ A3 ⊆ · · · and A =
S
An .
n=1
Then
µ(A) = lim µ(An ).
n→∞

(c). Continuity for decreasing intersection: Suppose (An )n≥1 ⊂ M

∞
with A1 ⊇ A2 ⊇ A3 ⊇ · · · and A =
T
An .
n=1
If µ(A) < ∞, then
µ(A) = lim µ(An ).
n→∞

(d). Countably subadditive: If (An )n≥1 ⊂ M then

∞  ∞
[  X
µ  An  ≤
  µ(An ).
n=1 n=1

Proof. (a). B can be decomposed as B = A ∩ (B \ A), a disjoint union. Thus the additivity of µ
yields
µ(B) = µ(A ∪ (B \ A)) = µ(A) + µ(B \ A) ≥ µ(A).

(b). Define (B j ) j≥1 by B1 = A1 and B j = A j \ A j−1 for j = 2, 3, 4, · · · . Notice that B j ∈ M and

∞ n
Bi ∩ B j , ∅ when i , j. A = B j , and An = B j . Applying the additivity of µ, we
S S
j=1 j=1
obtain
∞
X n
X
µ(A) = µ(B j ) and µ(An ) = µ(B j ).
j=1 j=1

∞
X n
X
µ(A) = µ(B j ) = lim µ(B j ) = lim µ(An ).
n→∞ n→∞
j=1 j=1

(c). Define (Cn )n≥1 by Cn = A1 \ An . Notice that C1 ⊆ C2 ⊆ C3 ⊆ · · · and

µ(Cn ) = µ(A1 \ An ) = µ(A1 ) − µ(An )

since A1 = An ∪ (A1 \ An ) and µ(A1 ) < ∞.

10

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Moreover,
∞ 
[ 
µ(A1 \ A) = µ  Cn  = lim µ(Cn )
n→∞
n=1
= lim (µ(A1 ) − µ(An ))
n→∞
µ(A1 ) − µ(A) = µ(A1 ) − lim µ(An )
n→∞

and µ(A) = lim µ(An ) since µ(A1 ) < ∞.

n→∞

(d). Let B1 = A1 , B2 = A1 ∪ A2 , · · · , Bn = A1 ∪ A2 ∪ · · · ∪ An .
∞
[ ∞
[
Bn = An .
n=1 n=1

Therefore,
∞ 
[ 
µ  Bn  = lim µ(Bn ) as (Bn )n≥1 is increasing
n→∞
n=1
 n 
[ 
= lim µ  Ai 
n→∞
i=1

Hence,
∞  n ∞
[  X X
µ  An  ≤ lim
  µ(Ai ) = µ(An ).
n→∞
n=1 i=1 n=1

Measurable Function
Definition 1.5: Measurable function

Let (X, M, µ) be a measure space and let f : X → [−∞, ∞]. We say that f is a measurable
function (or M-measurable) if for every a ∈ R, the set {x : f (x) > a} is a measurable set
(i.e., element of σ-algebra).

Example 1.12

1. Any continuous function f is measurable with measure space (R, L, m) since the
inverse image of any open set is open and L contains all open subsets of R.

11

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

2. Discontinuous function f defined by


 1x ,
 x,0
f (x) = 

0,
 x = 0.

is measurable with measure space (R, L, m). Indeed,





 (−∞, 1a ) ∪ [0, ∞), a<0
{x : f (x) > a} =  a=0

(0, ∞),



(0, 1 ), a > 0.


a

Each of these sets is on L.

3. Let f be the Dirichlet function


1,
 x∈Q
f (x) = 

0,
 x < Q.

In the measure space (R, L, m):





R, a<0
{x : f (x) > a} =  0≤a<1


Q,


∅
 a ≥ 1.

Since each of the sets R, Q and ∅ is Lebesgue measurable, it follows that the Dirichlet
function is Lebesgue measurable.

Definition 1.6

Let (X, M, µ) be a measure space and let E ⊆ X. The characteristic function of E denoted
by 1E , defined by

1, x∈E
1E = 


0, x < E.

A simple function s is a function whose range consists of finitely many values.

If c1 , c2 , . . . , cn are the distinct ranges values and C j = {x : s(x) = c j }, then s(x) =
n
c j 1C j (x) “it is called canonical representation”.
P
j=1

12

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Theorem 1.5

Let (X, M, µ) be a measure space. The function 1E is measurable function ⇔ E is a

measurable set. A simple function s is measurable ⇔ C j is measurable for j = 1, 2, . . . , n.

Proof. • Suppose E is measurable.

If a ≤ 0, then {x : 1E (x) < a} = ∅ a measurable set.
If a > 1 then {x : 1E (x) < a} = X a measurable set.
If 0 < a ≤ 1, {x : 1E (x) < a} = X \ E a measurable set.

• Suppose 1E is a measurable function. Then E = X \ {x : 1E (x) < 1/2} is a measurable

set.
□

Theorem 1.6

Let (X, M, µ) be a measure space and let f : X → [−∞, ∞]. The following are equivalent:

(a). {x : f (x) > a} is a measurable set for every a ∈ R.

(b). {x : f (x) ≥ a} is a measurable set for every a ∈ R.

(c). {x : f (x) < a} is a measurable set for every a ∈ R.

(d). {x : f (x) ≤ a} is a measurable set for every a ∈ R.

∞ n o
• (a) ⇔ (b) : {x : f (x) ≥ a} = x : f (x) > a − 1
T
Proof. k
∈ M.
k=1

∞ n o
• (b) ⇔ (a) : {x : f (x) > a} = x : f (x) ≥ a + 1
S
k
∈ M.
k=1

• (a) ⇔ (d)&(b) ⇔ (c): complement ∈ M.

□

Propoistion 1.2

Let (X, M, µ) be a measure space and let f, g : X → R be measurable functions, and let
c ∈ R. Then | f |, c f, f + g, f g and f /g (g , 0) are measurable.

13

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Proof. We have

X
 if a < 0
{x : | f (x)| > a} = 

{x : f (x) > a} ∪ {x : f (x) < −a} if a ≥ 0


Since X ∈ M and f is measurable it follows that {x : f (x) > a} ∈ M and {x : f (x) < −a} ∈ M;
hence
{x : f (x) > a} ∪ {x : f (x) < −a} ∈ M.
If c = 0 it is trivial. Assume c , 0. We have

{x : f (x) < a/c} if c < 0

{x : c f (x) > a} = 

{x : f (x) > a/c} if c > 0


Each of these set are in M since f is measurable. Hence c f is measurable.

To show that f + g is measurable, let a ∈ R. The inequality ( f + g)(x) > a ⇔ f (x) > a − g(x)
and we can find a rational number r with f (x) > r > a − g(x). Then
[
{x : ( f + g)(x) > a} = ({x : f (x) > r} ∩ {x : r > a − g(x)})
r∈Q
[
= ({x : f (x) > r} ∩ {x : g(x) > a − r})
r∈Q

Since f and g are measurable, the sets {x : f (x) > r} and {x : g(x) > a − r} are rational for
every rational r. Thus the intersection of these sets is measurable and the countable union is
also measurable. The function f 2 is measurable since

X
 if a < 0
{x : f (x) > a} = 
2

√ √
{x : f (x) < − a} ∪ {x : f (x) > a} if a ≥ 0


It follows that
1
f g = [( f + g)2 − f 2 − g2 ]
2
is measurable.
Finally, if g , 0




{x : g(x) < 1/a} if a > 0
{x : 1/g(x) > a} = {x : g(x) > 0} if a = 0




{x : g(x) > 0} ∪ {x : g(x) < 1/a} if a < 0



so 1/g is measurable and therefore f /g is measurable. □

Propoistion 1.3

Let f, g be measurable functions. Then max{ f, g}, min{ f, g} are measurable functions.

14

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Proof. It follows from

{x : max{ f, g}(x) > a} = {x : f (x) > a} ∪ {x : g(x) > a},

{x : min{ f, g}(x) > a} = {x : f (x) > a} ∩ {x : g(x) > a}.

Remark 1.3

As a consequence, the positive part of f , f+ (x) = max{ f (x), 0} is measurable and the
negative part of f is f− = max{− f (x), 0} is measurable. Note that f = f+ − f− and | f | =
f+ + f− .

We define
 
sup fn (x) : = sup{ fn (x)}.
n∈N n∈N
 
inf fn (x) : = inf { fn (x)}.
n∈N n∈N
(lim sup fn )(x) := lim sup{ fk (x) : k ≥ n} = inf sup{ fk (x) : k ≥ n} .



n→∞ n≥1
(lim inf fn )(x) := lim inf{ fk (x) : k ≥ n} = sup inf{ fk (x) : k ≥ n} .



n→∞ n≥1

Theorem 1.7

Let ( fn )n∈N be a sequence of measurable functions. Then supn∈N fn , inf n∈N fn , lim sup fn and
lim inf fn are measurable.

Proof. We have for every a ∈ R that

∞
[
x : (sup fn )(x) > a = {x : fn (x) > a},

n=1
\∞
{x : (inf fn )(x) > a} = {x : fn (x) > a},
n=1

so the supremum and infimum are measurable. Moreover, by definition of limsup and liminf,
we get that both are measurable. □

15

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Theorem 1.8

Let ( fn )n∈N be a sequence of measurable functions and fn → f pointwise as n → ∞. Then

f is measurable.

Proof. Since fn → f pointwise, then

f = lim sup fn = lim inf fn

which is measurable. □

Lebesgue Integral
Theorem 1.9: Fundamental Approximation

Let (X, M, µ) be a measure space and let f : X → [0, ∞] be measurable. Then there exists
a sequence of measurable, nonnegative, simple functions (sn ) with

(a). 0 ≤ s1 (x) ≤ s2 (x) ≤ · · · for every x ∈ X;

(b). lim sn (x) = f (x) for every x ∈ X.

n→∞

Corollary 1.1

If f : X → [−∞, ∞] is a measurable function, then there exists a sequence of measurable,

simple functions (sn ) with lim sn (x) = f (x) for every x ∈ X.
n→∞

Definition 1.7: Lebesgue Integral for Nonnegative Simple Function

Let (X, M, µ) be a measure space and s is a measurable, nonnegative, simple function

n
c j 1C j , C j = {x : s(x) = c j }.
X
s=
j=1

We define the Lebesgue integral of s over X by

Z n
X
sdµ = c j µ(C j ).
X j=1

16

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Moreover, if E ∈ M, then we define the Lebesgue integral of s over E by

Z n
X
sdµ = c j µ(C j ∩ E).
E j=1

Remark 1.4

Convention: 0 · ∞ = 0.

Example 1.13

Let µ be Lebesgue measure on R and let φ be defined by

if − 1 < x < 2




 1,
if 2 ≤ x < 4

2,


φ(x) = 





 3, if 4 ≤ x ≤ 8

0,
 otherwise.

Z
φ(x)dx = 1 · µ((−1, 2)) + 2 · µ([2, 4)) + 3 · µ([4, 8])
R
= 1 × 3 + 2 × 2 + 3 × 4 = 19.

Example 1.14

Let µ be Lebesgue measure on R and let φ : R → R be given by

φ(x) = 2χ[0,2] (x) + 3χ(2,3] (x).

That is, 



2, if 0 ≤ x ≤ 2
φ(x) =  if 2 < x ≤ 3


3,


0,
 otherwise

17

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Then
Z
φ(x)dx = 2.µ([0, 2]) + 3.µ((2, 3])
R
= 2(2 − 0) + 3(3 − 2)
= 7.

Example 1.15

Let µ be Lebesgue measure on R and let φ : R → R be given by

φ(x) = 2χ[−1,1] (x) + 3χ(3,7) (x) − 1χ[−4,−3) (x).

That is, 



2, if − 1 ≤ x ≤ 1
if 3 < x < 7

3,


φ(x) = 





−1, if − 4 ≤ x < −3

0,
 otherwise
Then
Z
φ(x)dx = 2.µ([−1, 1]) + 3.µ((3, 7)) − 1.µ([−4, −3))
R
= 2.(1 − (−1)) + 3.(7 − 3) + (−1).((−3) − (−4))
= 15.

Example 1.16

Lebesgue integral of f (x) = c1[a,b] (x).

Z Z
f (x)dx = 1[a,b] (x) f (x)dx
[a,b]
Z
= c 1[a,b] (x)dx

= cµ([a, b])
= c(b − a).

Example 1.17

Lebesgue integral of Dirichlet function on [0, 1].

18

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

A = Q ∩ [0, 1], A is a subset of Q, hence A is measurable and µ(A) = 0. Thus

Z
1A (x)dx = µ(A) = 0.

Remark 1.5

A = Q ∩ [0, 1] ⇒ A has Lebesgue measure zero.

Let (qi )i∈N be enumeration of the point in A.
ε ε ε

Given ε > 0, let Ri = qi − 2i+1 , qi + 2i+1 be an interval of length 2i
∋ qi . Then
∞
[ ∞
X
A⊂ Ri so 0 ≤ µ(A) ≤ µ(Ri ) < ε.
i=1 i=1

Hence µ(A) = 0 since ε > 0 is arbitrary.

In general, every countable set has Lebesgue measure zero. But the set of measure zero
maybe uncountable.
For example a Cantor set. A1 = [0, 1] \ ( 31 , 23 ) { removing the open middle third of each
remaining interval.

Remark 1.6

Let [a, b] be interval. Then

µ([a, b]) = µ(Q ∩ [a, b]) +µ(Qc ∩ [a, b])

| {z } | {z }
=b−a>0 =0

It follows that
µ(Qc ∩ [a, b]) > 0.

Example 1.18

For (R, L, m) and consider the Dirichlet function 1Q on [0, 1].

Z
1Q dm = 1 · |
m([0, 1] ∩ Q) +0 · m([0, 1] ∩ Qc )
[0,1]
{z }
=0
= 0.

19

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Property 1.3

If φ, ψ are simple, nonnegative functions and if c ≥ 0, then

Z Z
cφdµ = c φdµ,
Z Z Z
(φ + ψ)dµ = φdµ + ψdµ.

Theorem 1.10

Let (X, M, µ) be a measure space and let s be a measurable, nonnegative, simple function.
The set function φ : M → [0, ∞] defined by
Z
φ(E) = sdµ
E

is a positive measure on M.

Proof. 1. φ(∅) = 0 since µ(∅) = 0.

2. To show that φ is countably additive. Let (Ak )k≥1 ⊂ M pairwise disjoint sets in M and
∞ n
let A = Ak , s = c j 1C j .
S P
k=1 j=1

Then
Z n
X
φ(A) = sdµ = c j µ(C j ∩ A)
A j=1
n
∞  n ∞
X [  X X
= c j µ  (C j ∩ Ak ) = µ(C j ∩ Ak )

 cj
j=1 k=1 j=1 k=1
∞
XX n ∞
XZ ∞
X
= c j µ(C j ∩ Ak ) = sdµ = φ(Ak ).
k=1 j=1 k=1 Ak k=1
□

Definition 1.8: Lebesgue Integral for Nonnegative Functions

Let (X, M, µ) be a measure space. If f : X → [0, ∞] is measurable and E ∈ M, we define

the Lebesgue integral of f over E by
Z (Z )
f dµ = sup sdµ : s is measurable and simple with 0 ≤ s ≤ f on E .
E E

20

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

If this value is finite, we say that f is Lebesgue integrable.

Definition 1.9

Let (X, M, µ) be a measure space and f a nonnegative, measurable function. Let E ∈ M

(i.e., E is measurable). We define the integral of f over E with respect to µ to be
Z Z
f dµ = f.1E dµ.
E

This is the Lebesgue integral of f .

Property 1.4

If f and g are nonnegative, measurable functions and f ≤ g, then

Z Z
f dµ ≤ gdµ.

If f is a nonnegative, measurable function and E, F ∈ X and if E ⊆ F, then

Z Z
f dµ ≤ f dµ.
E F

Proof. If φ is a simple, nonnegative function such that 0 ≤ φ ≤ f , then it follows that 0 ≤ φ ≤ g.

Thus the first holds.
Since f.1E ≤ f.1F , the second assertion follows from the first. □

Definition 1.10: Lebesgue Integral for Measurable Functions

Let (X, M, µ) be a measure space. If f : X → [−∞, ∞] is measurable and E ∈ M, then we

define the Lebesgue integral of f over E by
Z Z Z
f dµ = f+ dµ − f− dµ
E E E
R R
provided either E f+ dµ or E f− dµ is finite.
If both integrals are finite we say that f is Lebesgue integrable.

Remark 1.7

f+ (x) = max{ f (x), 0} = | f (x)|+ f (x)

2
is the positive part of f and f− (x) = max{− f (x), 0} =

21

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

| f (x)|− f (x)
2
is the negative part of f . If E is a measurable set then we define
Z Z Z Z Z
f dµ = +
f dµ − f dµ =
−
f .1E dµ −
+
f − .1E dµ.
E E E X X

Propoistion 1.4

Let f and g be two functions that are integrable with respect to µ. If f (x) = g(x) a.e., then
Z Z
f dµ = gdµ.
X X

This tell us that sets of µ-measure 0 do not affect the integral.

Proof. We first prove this in the case where both f and g are nonnegative. Let Z = {x ∈
X| f (x) , g(x)}. Then µ(Z) = 0. Let φ be a simple function with 0 ≤ φ ≤ f , say
n
ai 1Ei (x),
X
φ(x) =
i=1

where ai ≥ 0. Set
n
ai 1Ei \Z (x).
X
ψ(x) =
i=1

Then 0 ≤ ψ ≤ g and
Z n
X n
X Z
ψdµ = ai µ(Ei \ Z) = ai µ(Ei ) = φdµ.
X i=1 i=1 X

Thus, Z Z
φdµ ≤ gdµ.
X X
Taking the supremum over all simple functions 0 ≤ φ ≤ f,
Z Z
f dµ ≤ gdµ.
X X

The reverse inequality follows the same manner. Therefore, in this case
Z Z
f dµ = gdµ.
X X

Finally, the general result follows by considering the positive and negative parts of f and g. □

22

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Property 1.5

Let f, g be Lebesgue integrable

R R
(i). If 0 ≤ f (x) ≤ g(x), then A
f dµ ≤ A
gdµ.
R R
(ii). A ⊆ B, 0 ≤ f (x) on B then A f dµ ≤ B f dµ.

(iii). f + g is integrable and A f + gdµ = A f dµ + A gdµ.

R R R

(iv). c f is integrable and A c f dµ = c A f dµ.

R R

(v). A, B, C ∈ M with A ∪ B = C and A ∩ B = ∅. If f is Lebesgue integrable, then

Z Z Z
f dµ = f dµ + f dµ.
C A B

Theorem 1.11: Lebesgue’s Monotone Convergence

Let (X, M, µ) be a measure space. If fn : X → [0, ∞] is a sequence of measurable functions

with 0 ≤ f1 (x) ≤ f2 (x) ≤ · · · for every x ∈ X, then the limit function lim fn (x) = f (x) is
n→∞
measurable with Z Z
lim fn dµ = f dµ.
n→∞ X X

Proof. Note that the limit function f is measurable. The monotonicity of the integrals yields
Z Z Z
fn dµ ≤ fn+1 dµ ≤ f dµ
X X X

for every n. By monotone convergence of sequences, the limit of the integrals exists as a value
in [0, ∞], call it α. We have that Z
α≤ f dµ.
X
If α = ∞, the equality holds. Now assume that α < ∞. Consider a simple function s with
0 ≤ s(x) ≤ f (x) on X and fix 0 < δ < 1. For each n, define a set

En = {x : fn (x) ≥ δs(x)} = {x : fn (x) − δs(x) ≥ 0}.

• each En is measurable since fn and s are measurable.

• E1 ⊆ E2 ⊆ E3 ⊆ · · · since fn is increasing sequence.

• X= ∞
S
n=1 E n . Why?

23

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

S∞
(i) n=1 En ⊆ X
(ii) Now we prove that X ⊆ ∞ n=1 E n . If x ∈ X and f (x) = 0, then s(x) = 0 and
S
x ∈ En for all n. If 0 < f (x) < ∞, then δs(x) ≤ δ f (x) < fn (x) for all large n. If
f (x) = ∞, fn (x) → ∞ so δs(x) < fn (x) for all large n.

Now we have the set function Z

φ(E) = sdµ
E
is a measure and thus Z ∞ 
[ 
sdµ = φ(X) = φ  En  = lim φ(En ).
X n→∞
n=1

Furthermore, by definition of the set En , we also see that

Z Z Z
1 1
φ(En ) = sdµ ≤ fn dµ ≤ fn dµ
En En δ δ X

Taking limit we get

α
Z Z
1
sdµ = lim φ(En ) ≤ lim fn dµ = .
X n→∞ n→∞ δ X δ
Since it holds for every δ ∈ (0, 1), it must be the case that
Z
sdµ ≤ α.
X

Moreover, since s is arbitrary simple function, taking the supremum gives us

Z
f dµ ≤ α.
X

Theorem 1.12: Fatou’s Lemma

If fn : X → [0, ∞] is a sequence of measurable functions, then

Z Z
(lim inf fn )dµ ≤ lim inf fn dµ.
X X

Proof. Define a sequence of functions (gn ) by gn (x) = inf k≥n { fk (x)}. Note that the functions
in this sequence is measurable and increasing by construction. The function lim inf fn is mea-
surable and we can write lim inf fn = limn→∞ gn . Applying Lebesgue’s monotone convergence
theorem 11, Z Z Z
lim inf fn dµ = lim gn dµ = lim gn dµ.
X X n→∞ n→∞ X

24

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

By the properties of infimum it follows that gn (x) ≤ fk (x) for every x ∈ X and all k ≥ n, from
which the monotonicity of the integral implies
Z Z
gn dµ ≤ inf fk dµ.
X k≥n X

Hence Z Z ! Z
lim inf fn dµ ≤ lim inf fk dµ = lim inf fn dµ.
X n→∞ k≥n X X

Is it true in general that

Z b Z b
lim fn (x)dx = lim fn (x)dx?
n→∞ a a n→∞

Example 1.19

For n = 1, 2, 3, . . . define 
n if 0 < x ≤
 1
fn (x) = 
 n
0 otherwise


In this case, we have

Z 1 Z 1
1 = lim fn (x)dx , lim fn (x)dx = 0.
n→∞ 0 0 n→∞

Theorem 1.13: Lebesgue’s Dominated Convergence

Suppose that fn : X → [−∞, ∞] is a sequence of measurable functions such that the limit
function lim fn (x) = f (x) exists for every x ∈ X.
n→∞
Further suppose there exists a Lebesgue integrable function g such that | fn (x)| ≤ g(x) for
every x ∈ X. Then

(a). f is Lebesgue integrable.

(b). lim X | fn − f |dµ = 0.

R
n→∞

fn dµ =
R R
(c). limn→∞ X X
f dµ.

25

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Proof. (a). Note that the function f is measurable. Since ( fn ) converges pointwise to f and
| fn (x)| ≤ g(x) on X, it must also be the case that | f (x)| ≤ g(x) on X. Then we have
Z Z
| f |dµ ≤ gdµ
X X

and f is integrable.

(b). We first observe that | fn − f | ≤ 2g. Then the sequence of functions 2g − | fn − f | is

a nonnegative sequence of measurable functions and we may apply Fatou’s Lemma 12.
Notice that limn→∞ | fn − f | = lim sup | fn − f | = lim inf | fn − f | = 0 on A from which it
follows that

lim inf(2g − | fn − f |) = 2g + lim inf(−| fn − f |) = 2g − lim sup(| fn − f |) = 2g.

Fatou’s Lemma 12 now yields

Z Z
2gdµ = lim inf(2g − | fn − f |)dµ
X X
Z !
≤ lim inf 2g − | fn − f |dµ
X
Z Z !
≤ 2gdµ + lim inf − | fn − f |dµ
X
Z Z X
= 2gdµ − lim sup | fn − f |dµ.
X X

Hence Z
lim sup | fn − f |dµ ≤ 0.
X
However, | fn − f | ≥ 0 and thus
Z
lim inf | fn − f |dµ ≥ 0.
X

Therefore, we get
Z Z
0 ≤ lim inf | fn − f |dµ ≤ lim sup | fn − f |dµ ≤ 0
X X

and this implies that Z

lim | fn − f |dµ = 0.
X

(c). Notice that the function | fn − f | is integrable and we have that

Z Z Z Z
0≤ fn dµ − f dµ = fn − f dµ ≤ | fn − f |dµ.
X X X X

26

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Applying the squeeze theorem, we obtain

Z Z
lim fn dµ = f dµ.
n→∞ X X

1.2 Metric Space

In this section, we will discuss on the concept of distance which has been generalized in the
abstract spaces known as the metric spaces. The notion of continuity, connectedness and com-
pactness will be covered.
Definition 1.11: Metric space

A metric space is a pair (X, d) consisting of a set X and a function d : X × X → R that

satisfies the following axioms:

(M1 ). d(x, y) ≥ 0, ∀x, y ∈ X, with equality if and only if x = y.

(M2 ). d(x, y) = d(y, x), ∀x, y ∈ X.

(M3 ). d(x, z) ≤ d(x, y) + d(y, z), ∀x, y, z ∈ X.

Remark 1.8

• d : X × X → R is a “distance function”.

• (M3 ): is called a triangle inequality.

Consequences:

1. d(x, z) ≥ d(x, y) − d(z, y) .

2. d(x1 , xn ) ≤ d(x1 , x2 ) + · · · + d(xn−1 , xn ).

Example 1.20

For any integer n ≥ 1, the function d : Rn × Rn → R defined by

n
X 1/2
d(x, y) = 2
|xi − yi |
i=1

is a metric on the set Rn .

27

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

An alternative metric on Rn is the function d1 : Rn × Rn → R defined by

n
X
d1 (x, y) = |xi − yi |.
i=1

Example 1.21

For any interger n > 1 and let 1 ≤ p ≤ ∞. The space (Rn , d p ) or (Cn , d p ) is a metric space,
where  1/p
n
if 1 ≤ p < ∞
 P p
|xi − yi |


d p (x, y) = 


i=1
max1≤i≤n |xi − yi | if p = ∞,




for x = (x1 , . . . , xn ) and y = (y1 , . . . , yn ).

Remark 1.9

The trianlge inequality for the distance function is called Minkowski’s inequality. It is
straightforward to verify if p = 1 or p = ∞, but it is not obvoius if 1 < p < ∞.

The proof of the triangle inequality uses the following sequence of inequalities.
Lemma 1.1

Let x, y > 0 and 0 < λ < 1. Then it holds

xλ y1−λ ≤ λx + (1 − λ)y.

Proof. Since the exponential is convex, we have

xλ y1−λ = exp[λ ln x + (1 − λ) ln y] ≤ λ exp(ln x) + (1 − λ) exp(ln y) = λx + (1 − λy).
□

Proposition 1.5: Young’s inequality

Let p ∈ (1, ∞) and define q ∈ (1, ∞) by 1p + 1q = 1. Then for any real numbers α, β > 0, it
holds
α p βq
αβ ≤ + .
p q

Proof. Take λ = 1p , x = α p and y = βq in Lemma 1. □

28

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Proposition 1.6: Hölder’s inequality

Let p, q ∈ (1, ∞) satisfying 1p + 1q = 1. For x = (x1 , x2 , . . . , xn ) and y = (y1 , y2 , . . . , yn ) in Rn

(or Cn ), it holds
X n n
X n
1/p  X 1/q
|xi yi | ≤ |xi | p
|yi |q
.
i=1 i=1 i=1

Note that when p = q = 2 this is the Cauchy-Schwarz inequality. Note also that it also holds
when p = 1 and q = ∞.

Proof. The inequality holds if x = 0 or y = 0 or both. Let x and y be nonzero elements of Rn

or Cn . For each i, 1 ≤ i ≤ n, let
|xi | |yi |
α=  1/p and β =  1/q
Pn p
Pn q
i=1 |xi | i=1 |yi |

Using the Young’s inequality 5 and sum over i, we get

Pn
i=1 |xi yi | 1 1
P 1/p  P 1/q ≤ p + q = 1.
n p n q
i=1 |xi | i=1 |yi |

Proposition 1.7: Minkowski’s inequality

Let 1 ≤ p < ∞. Then for any x = (x1 , x2 , . . . , xn ) and y = (y1 , y2 , . . . , yn ) in Rn (or Cn ), it

holds n n n
X 1/p  X 1/p  X 1/p
|xi + yi | p
≤ |xi | p
+ |yi | p
.
i=1 i=1 i=1

Proof. It is trivial when p = 1. Let consider 1 < p < ∞. We have

n
X n
X n
X
|xi + yi | ≤p
|xi ||xi + yi | p−1
+ |yi ||xi + yi | p−1 .
i=1 i=1 i=1

Apply Hölder’s inequality for each of the two sums on the right, we get
Xn n
X n
1/p  X 1/q
|xi ||xi + yi | ≤
p−1
|xi | p
|xi + yi |(p−1)q

i=1 i=1 i=1

n
X n
1/p  X 1/q
= |xi | p |xi + yi | p
i=1 i=1

29

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

and
n
X n
X n
1/p  X 1/q
|yi ||xi + yi | p−1
≤ |yi | p
|xi + yi | p
.
i=1 i=1 i=1

Thus,
n
X n
 X 1/p n
X n
1/p  X 1/q
|xi + yi | ≤
p
|xi | p
+ |yi | p
|xi + yi | p
.
i=1 i=1 i=1 i=1
P 1/q
If x + y , 0, we divide both sides by n
i=1 |xi + yi | p to obtain

n
X 1/p n
X 1/p n
X 1/p
|xi + yi | p
≤ |xi | p
+ |yi | p
.
i=1 i=1 i=1

since 1 − 1
q
= 1
p
and note that if x + y = 0 the desired inequality is trivial. □

Example 1.22

Let 1 ≤ p ≤ ∞. The set l p of all real or complex sequences (xk )k≥1 satisfying
∞
X
|xk | p < ∞ (1 ≤ p < ∞),
k=1
sup |xk | < ∞ (p = ∞)
k∈N

Then (l p , d p ) is a metric space where d p is defined by

 1/p
∞
if 1 ≤ p < ∞

|xk − yk | p

 P
d p (x, y) = 


k=1
if p = ∞,


sup |x − y |
k∈N k k


for x = (xk )k≥1 , y = (yk )k≥1 ∈ l p .

The nontrivial thing to verify is the triangle inequality.

Proposition 1.8: Minkowski’s inequality in l p

Let 1 ≤ p < ∞. Then for any x = (x1 , x2 , . . .) and y = (y1 , y2 , . . .) in l p it holds x + y ∈ l p

and ∞ ∞ ∞
X 1/p  X 1/p  X 1/p
|xi + yi | p
≤ |xi | p
+ |yi | p
.
i=1 i=1 i=1

30

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Proof. From Minkowski’s inequality in Proposition 7 we have for all n

n
X 1/p n
X 1/p n
X 1/p
|xi + yi | p ≤ |xi | p + |yi | p
i=1 i=1 i=1
∞
X 1/p ∞
X 1/p
≤ |xi | p
+ |yi | p
,
i=1 i=1

which is finite since by hypothesis x, y ∈ l p . We now take n → ∞ to conlcue that

∞
X 1/p ∞
X 1/p ∞
X 1/p
|xi + yi | p ≤ |xi | p + |yi | p .
i=1 i=1 i=1

Example 1.23

Let 1 ≤ p ≤ ∞. Define the set

 Z 1/p 
L (X) = f : f is measurable and
p
| f | p dµ <∞
X

and
L∞ (X) = f : f is measurable and ess sup | f | < ∞ .


Then (L p , d p ) is a metric space where

 R 1/p
 X | f − g| dµ ,
p
1 ≤ p < ∞,


d p ( f, g) = 

p = ∞.

ess sup | f − g|,


Note that the function in L p is defined a.e. In other words, for a measurable function f , the set
of all functions which are a.e. equal to f can be identified as

[ f ] = g : g is measurable with g = f a.e. .



31

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Proposition 1.9: Hölder’s inequality in L p

Let 1 ≤ p, q ≤ ∞ satisfying 1
p
+ 1
q
= 1. Then for f ∈ L p and g ∈ Lq it holds f g ∈ L1 and
Z Z1/p  Z 1/q
| f g|dµ ≤ | f | dµ qp
|g| dµ (1 < p < ∞),
ZX X Z X


| f g|dµ ≤ ess sup | f | |g|dµ.
X X

Proof. We will consider the case where 1 < p, q < ∞ as the case p = 1 or p = ∞ is easily ver-
R 1/p R 1/q
p
ified. We can assume that X | f | dµ > 0 and X |g| dµ
q
> 0. Using Young’s inequality
5 with
|f| |g|
α=  1/p and β =  R 1/q
R
| f | p dµ |g|q dµ
X X

and integrating over X to get

R
| f g|dµ 1 1
1/q ≤ p + q = 1.
X
R 1/p  R
X
| f | p dµ X
|g|q dµ

The case p = q = 2 is a classical inequality

Corollary 1.2: Cauchy-Schwarz’s inequality

Z Z 1/2  Z 1/2
| f g|dµ ≤ 2
| f | dµ |g|2 dµ .
X X X

Proposition 1.10: Minkowski’s inequality in L p

Let 1 ≤ p < ∞. For any f, g ∈ L p it holds f + g ∈ L p and

Z 1/p  Z 1/p  Z 1/p
| f + g| dµ
p
≤ p
| f | dµ + |g| p dµ
X X X

and for p = ∞,
ess sup | f + g| ≤ ess sup | f | + ess sup |g|.

32

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Proof. The case p = 1 is trivial. Let consider 1 < p < ∞. We have

Z Z Z Z
| f + g| dµ =
p
| f + g| | f + g|dµ ≤
p−1
| f || f + g| dµ +
p−1
|g|| f + g| p−1 dµ.
X X X X

Apply Hölder’s inequality for each of the two integrals on the right, we get
Z Z 1/p  Z 1/q
| f || f + g| dµ ≤
p−1 p
| f | dµ | f + g|(p−1)q
dµ
X X X
 Z 1/p  Z 1/q
= p
| f | dµ | f + g| dµ
p
X X

and Z Z 1/p  Z 1/q

|g|| f + g| p−1
dµ = p
|g| dµ | f + g| dµ .
p
X X X
Therefore,
Z  Z 1/p  Z 1/p  Z 1/q
| f + g| dµ =
p p
| f | dµ + p
|g| dµ | f + g| dµ .
p
X X X X
R 1/q
The desired inequality follows by dividing both sides by X
| f + g| dµ . Remark that if
p

R 1/q
X
| f + g| dµ
p
= 0 the result is obvious.
If p = ∞, we have | f | ≤ ess sup | f | a.e. in X and |g| ≤ ess sup |g| a.e. in X . Therefore,
| f + g| ≤ ess sup | f | + ess sup |g| a.e. in X, so that

ess sup | f + g| ≤ ess sup | f | + ess sup |g|.

Definition 1.12: Convergence

A sequence (xn ) in a metric space (X, d) converges to x ∈ X if for every ε > 0, there exists
N ∈ N such that
d(xn , x) < ε, for all n ≥ N.

Notation: lim xn = x or xn → x.
n→∞

Definition 1.13

A sequence (xn ) in (X, d) is a Cauchy sequence if, for every ε > 0, there exists N ∈ N such
that
d(xm , xn ) < ε, for all m, n ≥ N.

33

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Theorem 1.14

Suppose that (xn ) is a convergent sequence in a metric space (X, d). Then

1. the limit x = lim xn is unique

n→∞

2. any subsequence of (xn ) also converges to x

3. (xn ) is a Cauchy sequence.

Proof. 1. Suppose lim xn = l and lim xn = m. Let ε > 0. Then for n large enough
n→∞ n→∞

ε ε
d(l, m) ≤ d(l, xn ) + d(xn , m) < + = ε.
2 2
Thus 0 ≤ d(l, m) ≤ ε holds for any ε > 0. It follows that d(l, m) = 0, i.e, l = m.

2. Let ε > 0 be given. Since xn → l, there exists an Nε ∈ N such that xn ∈ Bε (l) for all
n ≥ Nε . If k ≥ Nε then nk ≥ k, so that nk ≥ Nε , and thus xnk ∈ Bε (l). This implies, by
definition, that xnk → l as k → ∞.

3. Let ε > 0 , since xn → l, ∃N such that ∀n ≥ N, d(xn , l) < ε/2. Then if m, n > N,
ε ε
d(xn , xm ) ≤ d(xn , l) + d(l, xm ) ≤ + = ε.
2 2
Thus (xn ) is a Cauchy sequence.
□

Propoistion 1.11

In a metric space, a sequence (xn ) can only converge to one limit.

Proof. Suppose xn → x and xn → y as n → ∞, with x , y. Then they can be separated by two

disjoint ball Br (x) and Br (y). But convergence means

∃N1 , n ≥ N1 ⇒ xn ∈ Br (x)

∃N2 , n ≥ N2 ⇒ xn ∈ Br (y).
For n ≥ max(N1 , N2 ) this would result in xn ∈ Br (x) ∩ Br (y) = ∅, a contradiction. □

34

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Propoistion 1.12

Distinct points of a metric space can be separated by disjoint balls,

x , y ⇒ ∃r > 0, Br (x) ∩ Br (y) = ∅.

Proof. If x , y then d(x, y) > 0. Let r := d(x, y)/2, then Br (x) is disjoint from Br (y) else we get
a contradiction: z ∈ Br (x) ∩ Br (y). Then
d(x, z) < r and d(y, z) < r.
Hence
d(x, y) ≤ d(x, z) + d(y, z) < 2r = d(x, y).
□

Example 1.24

1. In any metric space, xn → x ⇐⇒ d(xn , x) → 0 as n → ∞ (because xn ∈ Bε (x) ⇐⇒

d(xn , x) < ε).
n
2. In R, n+1 → 1 as n → ∞, since for any ε, there is an N such that 1
N
< ε (Archimedean
property of R), so n ≥ N implies
n 1 1
1− = < < ε.
n+1 n+1 N

Definition 1.14

A function f : X → Y between metric spaces is continuous when it preserves convergence,

xn → x in X ⇒ f (xn ) → f (x) in Y.

Remark 1.10

In this case, therefore, f ( lim xn ) = lim f (xn ).

n→∞ n→∞

Theorem 1.15

A function f : X → Y between metric spaces. TFAE

(i). f is continuous

35

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

(ii). ∀ε > 0, ∃δ > 0, ∀x ∈ X, dX (x, x′ ) < δ ⇒ dY ( f (x), f (x′ )) < ε,

(iii). For every open set V in Y, f −1 (V) is open in X.

(iv). For every closed subset F in Y, f −1 (F) is closed in X.

Proof. (i) ⇒ (ii): Suppose (ii) is false, then there is a point x ∈ X and an ε > 0 such that
arbitrarily small changes to x can lead to large variations in f (x), ∀δ > 0, ∃x′ , dX (x, x′ ) < δ and
dY ( f (x), f (x′ )) ≥ ε.
In particular, let δ = n1 , there is a sequence xn ∈ X satisfying dX (x, xn ) < n1 but dY ( f (x), f (xn )) ≥
ε. This means that xn → x, but f (xn ) ↛ f (x), contradicting statement (i).
(ii) ⇒ (iii): Note that (ii) can be written as
∀ε > 0, ∃δ > 0, ∀x ∈ X, x′ ∈ Bδ (x) ⇒ f (x′ ) ∈ Bε ( f (x))
or even as
∀ε > 0, ∃δ > 0, ∀x ∈ X, f (Bδ (x)) ⊆ Bε ( f (x)).
Let V be an open set in Y. To show that U := f −1 (V) is open in X, let x be any point of U, then
f (x) ∈ V, which is open. Hence,
f (x) ∈ Bε ( f (x)) ⊆ V,
and so
∃δ > 0, f (Bδ (x)) ⊆ Bε ( f (x)) ⊆ V.
In other words, x is an interior point of U, ∃δ > 0, Bδ (x) ⊆ f −1 (V) ⊆ U.
(iii) ⇒ (i): Let (xn ) be a sequence converging to x. Consider any open neighborhood Bε ( f (x))
of f (x). Then f −1 (Bε ( f (x)) contains x and is open set by (iii), so
∃δ > 0, x ∈ Bδ (x) ⊆ f −1 (Bε ( f (x)) ⇒ ∃δ > 0, f (Bδ (x)) ⊆ Bε ( f (x)).
But eventually all the points xn are inside Bδ (x),
∃N > 0, n > N ⇒ xn ∈ Bδ (x)
⇒ f (xn ) ∈ f (Bδ (x)) ⊆ Bε ( f (x))
⇒ dY ( f (xn ), f (x)) < ε.
This shows that f (xn ) → f (x) as n → ∞. □

Remark 1.11

1. Continuous functions preserve convergence is a central concept in metric space.

2. A homeomorphism between metric spaces X and Y is a mapping f : X → Y such

that f is bijective, f is continuous and f −1 is continuous.

3. A metric space X is said to be embedded in another space Y, when there is a subset

Z ⊆ Y such that X is homeomorphic to Z.

36

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

When X and Y are homeomorphic, they are not only the same as a sets (bijection) but
also with repsect to convergence

xn → x ⇐⇒ f (xn ) → f (x),

and
A is open in X ⇐⇒ f (A) is open in Y.
The most vivid picture is that of “deforming” one space continuously and reversibly form
the other. The classic example is that a ‘teacup’ is homeomorphic to a “doughnut”.

Definition 1.15: Complete Metric Space

A metric space is complete when every Cauchy sequence in it converges.

Theorem 1.16

The real number R is complete.

The proof of this theorem follows the Cauchy criterion in real analysis which states that a
sequence converges if and only if it is Cauchy.

Propoistion 1.13

(Q, d), d(x, y) = |x − y| is not complete.

√
Proof. Consider a sequence rationals xn converging to 2 in (R, d), specifically, we may as-
sume
√ 1
xn − 2 ≤
n
√
(This is possible since there are rationals number arbitrarily close to 2). Indeed, we can use
recurrence x1 = 1 and
xn−1 1
xn = + ,
2 xn−1
√
for n > 1. As (xn ) converges, then it is Cauchy. But its limits 2 < Q. □

Propoistion 1.14

Let X be a complete metric space. A subset F ⊆ X is complete ⇐⇒ F is closed in X.

37

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Proof. Let F ⊆ X be complete, i.e. any Cauchy sequence in F converges to a limit in F. Let
x ∈ F̄, with a sequence xn → x, xn ∈ F. Since convergent sequences are Cauchy and F is
complete, x must be in F. Thus F = F̄ is closed.
Conversely, let F be a closed in X and let (xn ) be a Cauchy sequence in F. Then (xn ) is a Cauchy
sequence in X, which is complete. Therefore, xn → x for some x ∈ X. In fact x ∈ F̄ = F. Thus
any Cauchy sequence of F converges in F. □

Remark 1.12

1. In fact, a complete subspace of any metric space is closed.

2. Two metric spaces may be homeomorphic yet one space be complete and the other
not. For example: R is homeomorphic to ]0, 1[. But ]0, 1[ is not closed in R, hence
not complete.

Note that a continuous function need not preserve completeness, or even Cauchy sequences; so
we need to strengthen continuity.

Definition 1.16

A function f : X → Y is said to be uniformly continuous when

∀ε > 0, ∃δ > 0, ∀x ∈ X, f (Bδ (x)) ⊆ Bε ( f (x)).

Remark 1.13

1. The difference from continuity is that, here, δ is independent of x.

2. Uniformly continuity implies continuity.

1
3. But not every continuously map is uniformly. For example, f (x) := x
on ]0, ∞[.

Propoistion 1.15

A uniformly continuous function maps any Cauchy sequence to a Cauchy sequence.

Proof. By definition f : X → Y is uniformly continuous when

∀ε > 0, ∃δ > 0, ∀x, x′ , dX (x, x′ ) < δ ⇒ dY ( f (x), f (x′ )) < ε.

38

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

In particular, for a Cauchy sequence (xn ) in X, with this δ,

∃N, m, n > N ⇒ dX (xn , xm ) < δ ⇒ dY ( f (xn ), f (xm )) < ε,
proving that ( f (xn )) is a Cauchy sequence in Y. □

Example 1.25

Lipschitz maps are uniformly continuous: since for any ε > 0, we can let δ = 2cε inde-
pendent of x to obtain d(x, x′ ) < δ ⇒ d( f (x),√f (x′ )) < cδ < ε. Not every uniformly
continuous function is Lipschitz. √For example,
√ x on [0, 1] is uniformly
√ continuous. If it
were Lipschitz, it would satisfy | x − 0| ≤ c|x − 0| which leads to x ≥ 1c .

Definition 1.17

• A function f : X → Y is Lipschitz map when

∃c, ∀x, x′ ∈ X, dY ( f (x), f (x′ )) ≤ cdX (x, x′ ).

• f is contraction when it is Lipschitz with c < 1.

• f is isometry, when f preserves distances, i.e.,

∀x, x′ ∈ X, dY ( f (x), f (x′ )) = dX (x, x′ ).

Theorem 1.17: Banach Fixed Point

Let X , ∅ be a complete metric space. Then every contraction map f : X → X has a

unique fixed point x = f (x), and the iteration xn+1 := f (xn ) converges to it for any x0 .

Proof. Consider the iteration xn+1 := f (xn ) starting with any x0 in X. Note that
d(xn+1 , xn ) = d( f (xn ), f (xn−1 )) ≤ cd(xn , xn−1 ).
Hence, by induction on n,
d(xn+1 , xn ) ≤ cn d(x1 , x0 ).
So (xn ) is Cauchy since c < 1. As X is complete, xn converges x and by continuity of f ,
 
f (x) = f lim xn = lim f (xn ) = lim xn+1 = x.
n→∞ n→∞ n→∞
cn
Moreover, the rate of convergence is given at least by d(x, xn ) ≤ 1−c
d(x1 , x0 ).
Suppose there are two fixed points x = f (x) and y = f (y), then
d(x, y) = d( f (x), f (y)) ≤ cd(x, y)
implying d(x, y) = 0 since c < 1. □

39

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Remark 1.14

The assumption complete metric space is important. Consider (0, 1), |.| is not complete
metric space (the sequence (1/n)n≥1 ⊂ (0, 1) and converges to 0, making it a Cauchy se-
quence in (0, 1) ⊂ R, but 0 < (0, 1)). Consider the function f : (0, 1) → (0, 1), f (x) = 21 x.
Then for all x, y ∈ (0, 1)
1 1 1 1
| f (x) − f (y)| = x − y = |x − y| ≤ |x − y|
2 2 2 2

so f is a contraction with c = 12 . The fixed point equation for f , i.e., f (x) = x has no
solutions in (0, 1) since 21 x = x if and only if x = 0 < (0, 1).

Definition 1.18: Connectedness

A subset C of a metric space is disconnected when it can be divided into (at least) two
disjoint non-empty subsets C = A ∪ B such that each subset is covered exclusively by an
open set, i.e.,

A ⊆ U, B ∩ U = ∅, U open
B ⊆ V, A ∩ V = ∅, V open.

Otherwise a set is called connected.

Theorem 1.18

The connected subsets of R are intervals.

Propoistion 1.16

Continuous functions map connected sets to connected sets. That is, f : X → Y continu-
ous, C ⊆ X connected, then f (C) is connected.

Proof. Let C be a subset of X, and suppose f (C) is disconnected into the non-empty disjoints
sets A and B convered exclusively by open sets U and V; that is,

f (C) = A ∪ B ⊆ U ∪ V, U ∩ B = ∅ = V ∩ A.

Then
C = f −1 (A) ∪ f −1 (B) ⊆ f −1 (U) ∪ f −1 (V),

40

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

f −1 (U) ∩ f −1 (B) = ∅ = f −1 (V) ∪ f −1 (A).

Moreover, f −1 (A) and f −1 (B) are non-empty and disjoint and f −1 (U) and f −1 (V) are open sets.
Hence f (C) disconnected implies C is disconnected. □

Theorem 1.19

Let C be a connected space and f : C → R a continuous function. For any c with

f (a) < c < f (b) there exists an x ∈ C such that f (x) = c.

Proof. f (C) is connected in R and so must be an interval. Then

f (a), f (b) ∈ f (C) ⇒ c ∈ f (C)

so c = f (x) for some x in C. □

Definition 1.19: Compactness

A set K is compact if any open cover has a finite subcover, i.e.,

[ iN
[
K⊆ Ai ⇒ ∃i1 , · · · , iN , K = Ai
i i=i1

(Ai open subset of X).

Remark 1.15

In metric space,
Ai = Bεi (ai ).

Propoistion 1.17

Compact sets are closed and bounded.

Proof. Let K be compact and x ∈ X \ K. To show x is exterior to K, we need to surround it by

a ball outside K. We know that x can be separated from any y ∈ K by disjoint open balls Brx
and Bry . Since y ∈ Bry , these balls cover K. But K is compact, so there is a finite of these balls
that still covers K,
K ⊆ Br1 (y1 ) ∪ · · · ∪ BrN (yN ).

41

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Now let r := min{r1 , · · · , rN }; then Br (x) ∩ K = ∅ since

z ∈ Br (x) ⇒ z ∈ Bri (x), i = 1, · · · , N
⇒ z < Br1 (y1 ) ∪ · · · ∪ BrN (yN ) ⊇ K.
Therefore, x ∈ Br (x) ⊆ X \ K.
To prove that K is bounded, we choose any point y ∈ K and let Bn (x) = x ∈ K : d(x, y) <

n , n = 1, 2, . . . . These open balls cover K. By compactness of K, a finite number of these balls
also covers K. The largest of these is a ball that contains K. □

Propoistion 1.18

(i). A closed subset of compact set is compact.

(ii). A finite union of compact set is compact.

Proof. (i). Let F be a closed subset of a compact set K, and let the open set Ai cover F, then
[
K ⊆ F ∪ (X \ F) ⊆ Ai ∪ (X \ F).
i

But K is compact and therefore a finite number of these open sets are enough to cover it
N
[ N
[
K⊆ Ai ∪ (X \ F), so F = Ai .
i=1 i=1

(ii). Let the open sets Ai cover the finite union of compact sets K1 ∪ · · · ∪ KN . Then they cover
each individual Kn and a finite number will then suffice in each case,
kn
[
Kn ⊆ Aik .
k=1

For n = 1, . . . , N, the collection of chosen Aik remains finite and together cover all Kn .
□

Remark 1.16

Finiteness of compact is important. For instance, every finite union of closed bounded
intervals is a compact subset of R. But
[
R= [n, n + 1] “countable union of compact sets”.
n∈Z

42

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

But R is not compact.

Theorem 1.20

Continuous functions map compact sets to compact sets. That is, f : K ⊂ X → Y is

continuous and K is compact, then f (K) is compact.

Proof. Let the sets Ai be an open cover for f (K)

[
f (K) ⊆ Ai .
i

From this can be deduced  

[  [
K ⊆ f −1  Ai  = f −1 (Ai ) .


i i

But f −1 (Ai ) are open sets since f is continuous. Therefore, RHS is an open cover of K. As K
is compact, a finite number of these open sets still cover it,
N
[
f −1 Aik .


K⊆
k=1

It follows that there is a finite subcover

N
[
f (K) ⊆ Aik
k=1

as required to show that f (K) compact. □

Theorem 1.21

Let (X, d) be a compact metric space and f : X → R be a continuous function. Then f

attains a maximum and minimum value on X.

Proof. By Theorem 20, since X is compact and f is continuous, f (X) is compact subset of R.
Hence f (X) is closed and bounded subset of R. Since f (X) is bounded it has both a supremum
and an infimum and it is closed, sup f (X) ∈ f (X) and inf f (X) ∈ f (X). Thus there are x0 , x1 ∈ X
such that f (x0 ) = inf f (X) and f (x1 ) = sup f (X), i.e., f attains its minimum and maximum
value at x0 and x1 , respectively. □

43

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Theorem 1.22

Let f : (X, dX ) → (Y, dY ) be a continuous function and (X, dX ) be a compact metric space.
Then f is uniformly continuous.

Proof. Since f : X → Y is continuous, given ε > 0, for every x ∈ X there exists δ x such that
ε
dY ( f (x), f (y)) < for all y ∈ B(x, δ x ). (∗)
2
The collection of balls B(x, δ x /2) x∈X is an open cover of X. Since X is compact, there are

points x1 , . . . , xn such that X = ∪nk=1 B(xk , δ xk /2). Choose 0 < δ < 21 min{δ x1 , . . . , δ xn } and let x, y
be points in X such that dX (x, y) < δ. Then x ∈ B(xi , δ xi /2) for some 1 ≤ i ≤ n and
δ xi δ xi δ xi
dX (y, xi ) ≤ dX (y, x) + dX (x, xi ) < δ + ≤ + = δ xi .
2 2 2
By (∗),
ε ε
dY ( f (x), f (y)) ≤ dY ( f (x), f (xi )) + dY ( f (xi ), f (y)) < + = ε,
2 2
which proves that f is uniformly continuous. □

Theorem 1.23: Heine-Borel

A set K ⊆ Rn is compact if and only if K is closed and bounded.

For the proof see [11].

Remark 1.17

Note that the statement of Heine-Borel Theorem is not true if we replace Rn by an arbitrary
metric space. For example, take X = (0, 1) with the usual metric d(x, y) = |x − y|. Let
K = X. The set K is closed in X. Also, K is bounded since d(x, y) < 1 for all x, y ∈ K.
However K is not compact.

Theorem 1.24: Characterization of compact sets

Let (X, d) be a metric space and let K ⊂ X. Then TFAE

1. K is compact

2. Every collection of closed subsets of X with finite intersection property has a non-
empty intersection.

3. K is sequentially compact (every sequence in K has a subsequence that converges in

44

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

K).

4. K is complete and totally bounded (it can be covered by a finite number of ϵ-balls).

For the proof of theorem see [5].

Which subsets of Banach space or topological space are compact?

✓ For the standard Euclidean space (Rn , ∥.∥2 ) the Heine-Borel Theorem asserts that a subset
of Rn is compact if and only if it is closed and bounded.

✓ This continues to hold for every finite dimensional normed vector space and conversely,
every normed vector space in which the closed unit ball is compact is finite-dimensional.

• How about infinite-dimensional Banach spaces?

✓ For the Banach space of continuous function on a compact metric space a characterization
of the compact subsets is given by Arzéla-Ascoli Theorem.

Remark 1.18

Let X = l p , 1 ≤ p ≤ ∞ and let en = (0, . . . , 0, 1, 0, . . . , ) ∈ B̄(0, 1) be the nth unit vectors

with the n th position equals to 1, i.e., the set of unit vectors (en ) in l p which closed and
 P∞ 1/p
bounded. But we remark that k=1 (enk − emk )
p
= 21/p if n , m, so that (en ) has no
converging subsequence. Hence the set of unit vectors (en ) in l p is not compact.

The space C(X, Y)

Let (X, dX ) and (Y, dY ) be metric spaces and assume that X is compact. Then the space

C(X, Y) := { f : X → Y | f is continuous}

of continuous function from X to Y is a metric space with distance

d( f, g) := sup dY ( f (x), g(x)), f, g ∈ C(X, Y)

x∈X

This is well-defined because the function

X→R
x 7→ dY ( f (x), g(x))

45

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

is continuous and hence is bounded as X is compact.

When X , ∅, the metric space (C(X, Y), d) is complete if and only if Y is complete. Why?
Indeed, let fn : X → Y be a Cauchy sequence. Then for every x ∈ X,
dY ( fn (x), fm (x)) ≤ d( fn , fm ) → 0 as n, m → ∞,
so ( fn (x)) is a Cauchy sequence in Y. When Y is complete, fn (x) converges to f (x). The
convergence is uniform as it is
d( fn , fm ) = sup dY ( fn (x), fm (x)) → 0.
x∈X

So given any ε > 0, ∃N such that dY ( fn (x), fm (x)) < 2ε for any n, m ≥ N and any x ∈ X. For
each x, we can choose m ≥ N, depend on x and large enough so that dY ( fm (x), f (x)) < 2ε and
this implies for any x ∈ X,
dY ( fn (x), f (x)) ≤ dY ( fn (x), fm (x)) + dY ( fm (x), f (x)) < ε
for any n ≥ N. Since this N is independent of x, it follows that d( fn , f ) → 0.
Remark 1.19

If fn → f uniformly and fn is continuous, then f is also continuous.

Definition 1.20

A subset F ⊂ C(X, Y) is called equicontinuous if for every ε > 0, there exists a constant
δ > 0 such that for all x, x′ ∈ X and all f ∈ F ,

dX (x, x′ ) < δ ⇒ dY ( f (x), f (x′ )) < ε.

Theorem 1.25: Arzéla-Ascoli

Let (X, d) be a compact metric space and let F ⊂ C(X, Rn ). Then F is compact if and only
if it is closed, bounded and equicontinuous.

For the proof of theorem see [5].

1.3 Banach Space

Banach spaces are important spaces in functional analysis. This section concerns the properties
of Banach spaces and some classical example of Banach spaces.

46

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Definition 1.21: Normed space

A normed space X is a vector space over R or C with a function ∥.∥ : X → R+ (called the
norm) such that for any x, y ∈ X, λ ∈ R or C,

1. ∥x∥ = 0 ⇔ x = 0.

2. ∥λx∥ = |λ|∥x∥.

3. ∥x + y∥ ≤ ∥x∥ + ∥y∥.

Consequences:

1. ∥x − y∥ ≥ ∥x∥ − ∥y∥ .
2. ∥x1 + · · · + xn ∥ ≤ ∥x1 ∥ + · · · + ∥xn ∥.

Definition 1.22: Banach space

X is a Banach space if the metric space (X, d) is complete; i.e., if every Cauchy sequence
in X converges, where
d(x, y) := ∥x − y∥.

Remark 1.20

Let (X, ∥.∥) be a normed space. Then

d(x, y) = ∥x − y∥ for x, y ∈ X

defines a distance on X.

Example 1.26

For 1 ≤ p < ∞, (Rn , ∥.∥ p ) is a Banach space where

X p 1/p
 n 
∥x∥ p =  |xk |  , x = (x1 , . . . , xn ) ∈ Rn .
k=1

For p = 2, this is the Euclidean norm.

For p = ∞, (Rn , ∥.∥∞ ) is a Banach space with the norm

∥x∥∞ = max |x1 |, |x2 |, . . . , |xn | .



47

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Then (Rn , ∥.∥ p ) is a finite dimensional Banach space for 1 ≤ p ≤ ∞.

Thanks to the Minkowski’s inequality, the space (Rn , ∥.∥ p ) is normed space. It remains to prove
the completeness. We will prove more general result in the next example for the completeness.

Example 1.27

Every finite dimensional vector space X is a Banach space.

It suffices to prove completeness for the Euclidean norm. The general result follows from
Theorem 28.
Suppose e1 , e2 , . . . , en is a basis of X and let (xk ) be a Cauchy sequence in X. Then for each k,
we can write xk in terms of basis,
Xn
xk = cki ei .
i=1

It follows that
n
X
|cki − cli |2 ≤ |cki − cli |2 = ∥xk − xl ∥22 ,
i=1

This implies that (cki ) is a Cauchy sequence for each i. Let ci be its limit and let x =
Pn
i=1 ci ei .
Then we have
n
X
∥xk − x∥2 =
2
|cki − ci |2 −→ 0,
i=1

as k → ∞ and this implies that xk converges to x as needed.

Example 1.28

X = C([0, 1]) = f : [0, 1] → R | f is continuous with the supremum norm



∥ f ∥∞ := sup | f (x)|.
x∈[0,1]

is a Banach space.

Note that X is a vector space and that ∥ f ∥∞ = 0 implies f = 0. The other conditions of norm
are satisfied

∥λ f ∥∞ = sup |λ|| f (x)| = |λ| sup | f (x)| = |λ|∥ f ∥∞ ,

x∈[0,1] x∈[0,1]

∥ f + g∥∞ = sup | f (x) + g(x)| ≤ sup (| f (x) + g(x)|) ≤ ∥ f ∥∞ + ∥g∥∞ ,

x∈[0,1] x∈[0,1]

48

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

for all f, g ∈ X and α ∈ R, so X is a normed space.

Now we prove the completeness. Let ( fn ) be a Cauchy sequence in X; that is, for all ϵ > 0 there
is an integer N such that
| fn (x) − fm (x)| ≤ ∥ fn − fm ∥∞ < ϵ,
for all n, m ≥ N and x ∈ [0, 1]. It follows that ( fn (x))n∈N is a Cauchy sequence in R. Since R is
complete, there exists f (x) := limn→∞ fn (x) in R for each x ∈ [0, 1]. Let x ∈ [0, 1] and ϵ > 0 be
given. Then for all n ≥ N, we deduce that
| f (x) − fn (x)| = lim | fm (x) − fn (x)| ≤ lim sup ∥ fm − fn ∥∞ < ϵ.
m→∞ m→∞

Since this inequality holds for all x ∈ [0, 1], it follows that ∥ f − fn ∥∞ < ϵ for all n ≥ N. Hence
fn converges uniformly to f and so f is continuous and f ∈ X.
Example 1.29

X = C([0, 1]) with a norm Z 1

∥ f ∥1 := | f (x)|dx
0
is not a Banach space.

It is easy to check that ∥.∥1 is a norm on X. Now, consider the functions given by




0, 0 ≤ x ≤ 12 − 1n ,
fn (x) = nx − 2n + 1, 12 − n1 ≤ x ≤ 12 ,




 1
1,

2
≤ x ≤ 1,
for n ≥ 3.
We claim that ( fn ) is a Cauchy seuqence in X. To see this, let m ≥ n ≥ 3, we have
Z 21
1
∥ fn − fm ∥1 = | fn (x) − fm (x)|dx ≤ → 0,
1 1
2−n
n
as n → ∞, thus ( fn ) is a Cauchy sequence with respect to ∥.∥1 .
We claim that ( fn ) does not converge to some f ∈ X. Take any a ∈ 0, 12 and let n ∈ N with


1
2
− 1n ≥ a. We then have
Z a Z a
0≤ | f (x)|dx = | f (x) − fn (x)|dx ≤ ∥ fn − f ∥1 → 0,
0 0
Ra
as n → ∞. This implies that 0 | f (x)|dx = 0. The continuity of f implies that f = 0 on [0, a]
for every a < 12 and hence f 12 = 0. On the other hand,

Z 1 Z 1
0≤ | f (x) − 1|dx = | f (x) − fn (x)|dx ≤ ∥ f − fn ∥1 → 0,
1 1
2 2

as n → ∞. As a consequence, f = 1 on , 1 , which is a contradiction.

1 
2

49

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Example 1.30

For 1 ≤ p < ∞, (l p , ∥.∥ p ) is a Banach space where

 ∞

 X 
= < ,
 
l p :=  x (x ) |x | p
∞
 
 k k∈N k 

 
k=1

X p 1/p
∞ 
∥x∥ p :=  |xk |  , x ∈ lp.
k=1

For p = ∞, (l , ∥.∥∞ ) is a Banach space of bounded sequence where

∞

∥x∥∞ := sup |xk |, x = (xk )k∈N ∈ l∞ .

k

Then (l p , ∥.∥ p ) is an infinite dimensional Banach space for 1 ≤ p ≤ ∞.

The space (l p , ∥.∥ p ) is a normed space follows from the Minkowski’s inequality. It remains to
see why it is complete.
The case p = ∞. Let (xk ) be a Cauchy sequence in l∞ . Write xk = (xk1 , xk2 , . . .). We have to find
an element x ∈ l∞ such that (xk ) converges to x. Let ϵ > 0. Because (xk ) is a Cauchy sequence,
there exists an integer N > 0 such that

∥xk − xl ∥∞ < ϵ,

for all k, l ≥ N. Thus,

sup |xkm − xlm | < ϵ,
m≥1

for all k, l ≥ N. In particular, we have

|xkm − xlm | < ϵ,

for all m and all k, l ≥ N. This implies that for each m, the sequence

x1m , x2m , . . .

is a Cauchy sequence in R or C, which is complete, we thus have a limit, call it xm .

lim xkm = xm .
k→∞

Let x denote the sequence x = (x1 , x2 , . . .). We want to prove that xk converges to x. Choose an
ϵ > 0. By replacing ϵ with ϵ/2, there exists an integer N > 0 such that

|xkm − xlm | < ϵ/2,

50

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

for all m and all k, l ≥ N. Taking limits as l → ∞, we get

|xkm − xm | < ϵ/2,
for all m and all k ≥ N. Taking the supremum in m, we get
sup |xkm − xm | < ϵ/2,
m≥1

for all k ≥ N. In other words,

∥xk − x∥∞ ≤ ϵ/2 < ϵ,
for all k ≥ N. This implies that xk converges to x.
The case 1 ≤ p < ∞. Suppose (xk ) is a Cauchy sequence in l p and let ϵ > 0. Then there exists
an integer N such that
∞
X  ϵ p
p
|xkm − xlm | ≤ |xkm − xlm | p = ∥xk − xl ∥ pp < ,
m=1
2

for all k, l ≥ N. In particular, the sequence (xkm )∞

k=1 is Cauchy for each m. Let xm denote its
limit. Given any n ≥ 1, we then have
n
X  ϵ p n
X  ϵ p
|xkm − xlm | p < =⇒ |xkm − xm | p <
m=1
2 m=1
2

It follows that  ϵ p
∥xk − x∥ pp <
2
p
for all k ≥ N and xk converges to x and that x ∈ l .

Example 1.31

For 1 ≤ p < ∞, (L p (µ), ∥.∥ p ) is a Banach space where

Z !1/p
∥ f ∥p = p
| f | dµ ,
M

where f : M → R measurable.
For p = ∞, (L∞ (µ), ∥.∥∞ ) is a Banach space of bounded measurable functions with

∥ f ∥∞ := ess sup | f | = inf c ≥ 0 | | f | ≤ c a.e



Note that

1. The case X = Rn and µ equals Lebesgue measure. Then we have

Z !1/p
∥ f ∥p = p
| f (x)| dx .
Rn

51

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

2. The case X = N and µ equals to the counting measure. Then L p and l p are coincide.

The space (L p (µ), ∥.∥ p ) is a normed space follows from the Minkowski’s inequality. To prove
the completeness we distinguish two cases:
Case 1 ≤ p < ∞:
Step 1. The Cauchy sequence: Let ( fn ) be a Cauchy sequence in L p . It suffices to show that
( fn ) has a convergence subsequence with limit f in L p . By definition of Cauchy sequence, for
each k ∈ N, we can find Nk ∈ N so that ∥ fm − fn ∥ p < 21k whenever m, n ≥ Nk . In addition,
construct an increasing sequence of natural numbers by choosing n1 = N1 and for k ≥ 2, set
nk = max{nk−1 , Nk }+1. Then we have a subsequence ( fnk ) with ∥ fnk+1 − fnk ∥ < 21k for every k ∈ N.
We want to show that ( fnk ) converges to a function f ∈ L p . We simply refer to fnk as fk .
Step 2. Conversion to a convergent monotone sequence. Define a sequence of functions (gk )
by g1 (x) = | f1 (x)| and

gk (x) = | f1 (x)| + | f2 (x) − f1 (x)| + | f3 (x) − f2 (x)| + · · · + | fk (x) − fk−1 (x)|, k ≥ 2.

It follows that
0 ≤ g1 (x) ≤ g2 (x) ≤ g3 (x) ≤ · · ·
and

∥gk ∥ p = ∥| f1 | + | f2 − f1 | + | f3 − f2 | + · · · + | fk − fk−1 |∥ p
≤ ∥ f1 ∥ p + ∥ f2 − f1 ∥ p + ∥ f3 − f2 ∥ p + · · · + ∥ fk − fk−1 ∥ p
X k
= ∥ f ∥p + ∥ f j − f j−1 ∥ p
j=2

< ∥ f1 ∥ p + 1,

where we used
k ∞ ∞
X X X 1
∥ f j − f j−1 ∥ p < ∥ f j − f j−1 ∥ p < = 1.
j=2 j=1 j=1
2j

This shows that gk ∈ L p for every k ∈ N. Set g(x) = limk→∞ gk (x) and g p (x) = limk→∞ gk (x) p .
Moreover, the sequence (gk (x)) p is increasing for each x ∈ X and an application of monotone
convergence theorem yields
Z !1/p Z !1/p Z !1/p
∥g∥ p = p
|g| dµ = p
lim gk dµ = lim p
gk dµ
X X k→∞ k→∞ X
Z !1/p
= lim p
gk dµ = lim ∥gk ∥ p ≤ ∥ f1 ∥ p + 1.
k→∞ X k→∞

This shows that g ∈ L p since ∥ f1 ∥ p < ∞.

Step 3. Pointwise convergence of ( fk ). Since g is the pointwise limit of (gk ), then the sequence

52

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

(gk (x)) is a Cauchy in R for every x. That is, for all ϵ > 0, there exists N ∈ N such that
|gk (x) − gm (x)| = gk (x) − gm (x) < ϵ whenever k > m ≥ N and thus
k
X k
X
| fk (x) − fm (x)| = f j (x) − f j−1 (x) ≤ | f j (x) − f j−1 (x)| = gk (x) − gm (x) < ϵ,
j=m+1 j=m+1

whenever k > m ≥ N. Hence ( fk (x)) is Cauchy in R. So fk → f a.e. Since

k
X
| fk | ≤ | f1 | + | f j − f j−1 | ≤ gk ≤ g.
j=2

It follows that | f | ≤ g a.e. Hence | fk | p ≤ g p , | f | p ≤ g p and | f − fk | p ≤ 2g p and dominated

convergence theorem yields
Z Z
lim | f − fk | dµ =
p
lim | f − fk | p dµ = 0.
k→∞ X X k→∞

Case p = ∞: Let ( fn ) be a Cauchy sequence in L∞ (µ) and let Ak and Bm,n be the sets such that
| fk (x)| > ∥ fk ∥∞ and | fn (x) − fm (x)| > ∥ fn − fm ∥∞ , repsectively and let E be the union of these
sets for k, m, n = 1, 2, 3, . . . . Then µ(E) = 0 and on the complement of E the sequence ( fn )
converges uniformly to a bounded function f . Define f (x) = 0 for x ∈ E. Then f ∈ L∞ (µ) and
∥ fn − f ∥∞ → 0 as n → ∞.

Remark 1.21

The main ingredients in the proof consist of Minkowski’s inequality, Monotone conver-
gence theorem, Lebesgue dominated convergence theorem and the following fact.

Lemma 1.2

Let (X, d) be a metric space. Then X is complete if and only if every Cauchy sequence
has a convergent subsequence (with limit x ∈ X).

Proof. Let (xn ) be a Cauchy sequence in the metric space (X, d) and (xnk ) be its conver-
gence subsequence, say xnk → x. We claim that (xn ) is convergent with the limit x. Then
∀ϵ > 0, ∃N ∈ N such that for all n, m, k > N,
ϵ ϵ
d(xn , xm ) < and d(xnk , x) < .
2 2
Therefore, note that nk ≥ k, we have
ϵ ϵ
d(xn , x) ≤ d(xn , xnk ) + d(xnk , x) < + = ϵ.
2 2

53

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

which implies that xn → x. □

1.4 Bounded Linear Operators

The second fundamental concept in functional analysis after that of Banach space, is the notion
of a bounded linear operator.

Definition 1.23: Bounded Linear Operator

Let (X, ∥.∥X ) and (Y, ∥.∥Y ) be real normed spaces. A linear operator T : X → Y is bounded
if there exists a constant c ≥ 0 such that

∥T x∥Y ≤ c∥x∥X , for all x ∈ X.

If no such constant exists, we say that T is unbounded. If T : X → Y is a bounded linear

operator, then we define the operator norm ∥T ∥ of T by

∥T ∥ := ∥T ∥L(X,Y) = inf c | ∥T x∥Y ≤ c∥x∥X for all x ∈ X .



Example 1.32

The linear operator T : C([0, 1]) → R defined by

T ( f ) = f (0)

is bounded.

Indeed, we have
|T ( f )| = | f (0)| ≤ sup {| f (x)|} = ∥ f ∥.
x∈[0,1]

Hence T is bounded with c = 1.

Remark 1.22

If C([0, 1]) is equipped with the norm

Z 1
∥ f ∥1 = | f (x)|dx.
0

Then T is unbounded.

54

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Let consider a sequence of functions ( fn )n≥2 defined by

 2
− n2 x + n, 0 ≤ x ≤ 2n

fn (x) = 

2
0,

n
≤ x ≤ 1.
R1
Then ∥ fn ∥1 = 0
| f (x)|dx = 1 for all n and f (0) = n. Thus |T ( fn )| = n → ∞ as n → ∞. So T is
unbounded.

Example 1.33

Let X = (C([0, 1]), ∥.∥∞ ) and define a linear operator T on X by

Z 1
(T f )(x) = k(x, y) f (y)dy,
0

where k : [0, 1] × [0, 1] → R is continuous. Then T is a bounded operator.

Note that k is continuous on the compact set [0, 1] × [0, 1] hence it is bounded, that is, ∥k∥∞ =
max x,y∈[0,1] |k(x, y)| < ∞. Thus,
Z 1
∥T f ∥∞ = sup |(T f )(x)| = sup k(x, y) f (y)dy
x∈[0,1] x∈[0,1] 0
Z 1
≤ sup |k(x, y)|| f (y)|dy ≤ ∥ f ∥∞ ∥k∥∞
x∈[0,1] 0

Thus, T is bounded with c = ∥k∥∞ .

Example 1.34

Let (X, ∥.∥∞ ) be the normed space of all polynomials on [0, 1] with ∥ f ∥∞ = max x∈[0,1] | f (x)|.
A differential operator T is defined on X by (T f )(x) = f ′ (x). Then T is unbounded.

Remark that T is linear operator due to the linearity of differentiation. To see that T is un-
bounded, we consider fn (x) = xn then (T fn )(x) = nxn−1 , x ∈ [0, 1] and

∥ fn ∥∞ = 1 and (T fn )(x) = fn′ (x) = nxn−1

Thus
∥T fn ∥∞ = n and ∥T fn ∥∞ /∥ fn ∥∞ = n → ∞ as n → ∞.
So T is unbounded.

55

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Example 1.35

Let (ak ) ∈ l∞ . The linear operator T : l1 → R defined by

∞
X
T (xk ) = ak xk
k=1

is bounded.

T is well defined: Since (ak ) ∈ l∞ and (xk ) ∈ l1 . Then

∞
X ∞
X
|ak xk | ≤ sup{|ak |} |xk | = ∥a∥∞ ∥x∥1 < ∞.
k=1 k≥1 k=1

T is bounded:
∞
X ∞
X ∞
X
|T (xk )| = ak xk ≤ |ak xk | ≤ sup{|ak |} |xk | = ∥a∥∞ ∥x∥1 .
k=1 k=1 k≥1 k=1

Hence T is bounded with c = ∥a∥∞ .

Example 1.36

Let (ak ) ∈ l∞ . The linear operator T : l2 → l2 defined by

∞
X
T (xk ) = ak xk
k=1

is bounded.

T is well defined: Since (ak ) ∈ l∞ and (xk ) ∈ l2 . Then

∞
X ∞

2 X
|ak xk |2 ≤ sup{|ak |} |xk |2 = ∥a∥2∞ ∥x∥22 < ∞.
k=1 k≥1 k=1

It follows that (ak xk ) ∈ l2 .

T is bounded:
∞
X ∞

2 X
∥T (xk )∥22 = 2
|ak xk | ≤ sup{|ak |} |xk |2 = ∥a∥2∞ ∥x∥22 .
k=1 k≥1 k=1

Hence T is bounded with c = ∥a∥∞ .

Let take a look at another example called a multiplication operator as follows.

56

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Example 1.37

Consider the normed space (l p , ∥.∥ p ) for 1 ≤ p ≤ ∞. For m ∈ l∞ , we define T m : l p → l p by

T m (x) = (mx1 , mx2 , mx3 , . . .).

Then T m ∈ L(l p ) and ∥T m ∥L(l p ) = ∥m∥∞ .

Indeed, for 1 ≤ p < ∞, we have

T is well defined: Since (mk ) ∈ l∞ and (xk ) ∈ l p . Then
∞
X ∞

p X p
|mk xk | p ≤ sup{|mk |} |xk | = ∥m∥∞
p
∥x∥ pp < ∞.
k=1 k≥1 k=1
p
It follows that (mk xk ) ∈ l .
T is bounded:
∞
X ∞
X ∞

p X
∥T m (x)∥ pp = |mk xk | p = |mk | p |xk | p ≤ sup{|mk |} |xk | p = ∥m∥∞
p
∥x∥ pp .
k=1 k=1 k≥1 k=1

Hence T m is bounded with c = ∥m∥∞ . It follows that ∥T m ∥L(l p ) ≤ ∥m∥∞ . To see that the equality
holds, consider (ek ) ∈ l p , where ek = (0, . . . , 0, 1, 0, . . .) with a 1 at the k-th position. Then
∥ek ∥ p = 1 and ∥T m ek ∥ p = |mk |. Thus ∥T m ∥L(l p ) ≥ |mk | for all k ∈ N. Hence ∥T m ∥L(l p ) ≥ ∥m∥∞ . The
case p = ∞ is similar.

Equivalent expressions for ∥T ∥ are:

∥T ∥ = inf c | ∥T x∥Y ≤ c∥x∥X for all x ∈ X ,

(1.4.1)
∥T x∥Y
∥T ∥ = sup , (1.4.2)
x,0 ∥x∥X
∥T ∥ = sup ∥T x∥Y , (1.4.3)
∥x∥X ≤1

∥T ∥ = sup ∥T x∥Y . (1.4.4)

∥x∥X =1

We have the estimate

sup ∥T x∥Y ≤ sup ∥T x∥Y
∥x∥X =1 ∥x∥X ≤1

from the definition of supremum since we are taking suprema of the same quantity over a larger
set. Since
∥T x∥Y  x 
= T
∥x∥X ∥x∥X
we have
∥T x∥Y
sup ≤ sup ∥T x∥Y .
x,0 ∥x∥X ∥x∥X =1

57

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

If ∥x∥X ≤ 1 we have
∥T x∥Y
∥T x∥Y ≤ .
∥x∥X
Then
∥T x∥Y
sup ∥T x∥Y ≤ sup
∥x∥X ≤1 x,0 ∥x∥X
and thus
∥T x∥Y
sup = sup ∥T x∥Y = sup ∥T x∥Y .
x,0 ∥x∥X ∥x∥X ≤1 ∥x∥X =1

Finally, we have
∥T x∥Y 

∥T x∥Y ≤ sup ∥x∥X
x,0 ∥x∥X

for all x ∈ X. Then ∥T ∥ ≤ sup x,0 ∥T∥x∥x∥XY . Moreover, by the definition of supremum there exists a
sequence (xn ) such that
∥T xn ∥Y ∥T x∥Y 1
≥ sup − for all n.
∥xn ∥X x,0 ∥x∥X n
By the definition of ∥T ∥ we have that for all n, ∥T ∥ ≥ ∥T∥xxnn∥∥XY . Therefore, ∥T ∥ = sup x,0 ∥T x∥Y
∥x∥X
.
The space of bounded linear operators from X to Y is denoted by
L(X, Y) := T : X → Y | T is linear and bounded .


Theorem 1.26

Let (X, ∥.∥X ) and (Y, ∥.∥Y ) be real normed spaces. Then (L(X, Y), ∥.∥L(X,Y) ) is a normed
space.

Proof. Let T, S ∈ L(X, Y) and λ ∈ R or C. If ∥T ∥L(X,Y) = 0 then T x = 0 for all x ∈ X, it follows

that T = 0. Moreover, we have
∥λT ∥L(X,Y) = sup ∥λT x∥Y = |λ| sup ∥T x∥Y = |λ|∥T ∥L(X,Y) .
∥x∥X ≤1 ∥x∥X ≤1

Finally,
∥T + S ∥L(X,Y) = sup ∥(T + S )x∥Y
∥x∥X ≤1

≤ sup (∥T x∥X + ∥S x∥Y )

∥x∥X ≤1

≤ sup ∥T x∥Y + sup ∥S x∥Y

∥x∥X ≤1 ∥x∥X ≤1

= ∥T ∥L(X,Y) + ∥S ∥L(X,Y) .
It follows that (L(X, Y), ∥.∥L(X,Y) ) is a normed space. □

58

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Theorem 1.27

Let (X, ∥.∥X ) and (Y, ∥.∥Y ) be real normed spaces and let T : X → Y be a linear operator.
TFAE

(1). T is bounded

(2). T is continuous

(3). T is continuous at x = 0.

Proof. (1) ⇒ (2). If T is bounded, then

∥T x − T x′ ∥Y = ∥T (x − x′ )∥Y ≤ ∥T ∥∥x − x′ ∥X

for all x, x′ ∈ X and so T is Lipschitz continuous. Since every Lipschitz continuous function is
continuous, this shows that (1) ⇒ (2).
(2) ⇒ (3): By definition of continuity.
(3) ⇒ (1): Assume that T is continuous at x = 0. Then it follows from ϵ − δ definition of
continuity with ϵ = 1 that there exists a constant δ > 0 such that for all x ∈ X,

∥x∥X ≤ δ ⇒ ∥T x∥Y ≤ 1.

This implies ∥T x∥Y ≤ 1 for every x ∈ X with ∥x∥X ≤ δ. Now let x ∈ X \ {0}. Then

X x∥X ≤ δ
∥δ∥x∥−1

and so  
T δ∥x∥−1
X x ≤1
Y

Hence
∥T x∥Y ≤ δ−1 ∥x∥X , for all x ∈ X.
□

The following results hold.

1. T ∈ L(X, Y) iff ∥T ∥ < ∞, in which case,

∥T x∥ ≤ ∥T ∥∥x∥.

2. L(X, Y) is complete if Y is.

59

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Indeed, let T n be a Cauchy sequence of operator in L(X, Y); that is, ∥T n −T m ∥ → 0 as n, m → ∞.

Then for each x ∈ X,
∥T n x − T m x∥ ≤ ∥T n − T m ∥∥x∥ → 0
implies that (T n x) is a Cauchy sequence in Y so that T n x converges to some vector denoted by
T x, if Y is complete. We now show that T is linear. As
T n (x + y) = T n x + T n y, T n (λx) = λT n x
then
T (x + y) = T x + T y, T (λx) = λT x
by continuity of addition and scalar multiplication. Finally, for any ϵ > 0 and any x ∈ X,
∥(T n − T )x∥ ≤ ∥T n − T m ∥∥x∥ + ∥T m x − T x∥
≤ ϵ∥x∥ + ϵ∥x∥
where m is chosen large enough, depending on x, to make ∥T m x − T x∥ ≤ ϵ∥x∥ and n, m ≥ N
large enough to make ∥T n − T m ∥ ≤ ϵ. Hence
∥T n − T ∥ ≤ 2ϵ
for n ≥ N. This shows that T n − T and so T are continuous and furthermore that T n → T .

Finite Dimensional Normed Spaces

The special features of a finite dimensional normed space are that every finite dimensional (real
or complex) normed space is a Banach space, that every linear operator on finite dimensional
space is bounded (continuous) and that all norms on a finite dimensional space are equivalent.

Definition 1.24

Let X be a real vector space. Two norms ∥.∥ and ∥.∥′ on X are equivalent if there is a
constant C ≥ 1 such that
1
∥x∥ ≤ ∥x∥′ ≤ C∥x∥ for all x ∈ X.
C

Remark 1.23

Definition 24 is equivalent to
1 ′
∥x∥ ≤ ∥x∥ ≤ C∥x∥′ for all x ∈ X.
C
An alternative condition for the norms equivalent is that there exist positive real numbers

60

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

c, d such that
c∥x∥ ≤ ∥x∥′ ≤ d∥x∥ for all x ∈ X.
and this condition is equivalent to the above by setting C = max{1/c, d}.

Theorem 1.28

Let X be a finite dimensional real vector space. Then any two norms on X are equivalent.

Proof. Choose a basis e1 , . . . , en in X and define

 n 1 n
X 2  2 X
∥x∥2 :=  |xi |  for x= xi ei , xi ∈ R.
i=1 i=1

This is a norm on X. We prove that every norm on X is equivalent to ∥.∥2 . Fix any norm function

X → R : x 7→ ∥x∥.

Step 1: There is a constant c > 0 such that ∥x∥ ≤ c∥x∥2 for all x ∈ X. Define
 n  12 n
X X
x= xi ei , xi ∈ R.

c :=  ∥ei ∥2  and let
i=1 i=1

Then by triangle inequality for ∥.∥ and the Cauchy-Schwarz inequality on Rn , we have
n
 n 1  n  12
X X 2  2 X
 ∥ei ∥  ≤ c∥x∥2 .
2

∥x∥ ≤ |xi |∥ei ∥ ≤  |xi | 
i=1 i=1 i=1

Step 2: There is a constant δ > 0 such that δ∥x∥2 ≤ ∥x∥ for all x ∈ X.
The set S := x ∈ X | ∥x∥2 = 1 is compact with respect to ∥.∥2 by the Heine-Borel Theorem

23 and the function S → R : x 7→ ∥x∥ is continuous. Hence by Theorem 21, there exists an
element x0 ∈ S such that ∥x0 ∥ ≤ ∥x∥ for all x ∈ S . Define δ := ∥x0 ∥ > 0. The every nonzero
2 x ∈ S , hence ∥x∥2 x ≥ δ and so ∥x∥ ≥ δ∥x∥2 .
vector x ∈ X satisfies ∥x∥−1 −1
□

Corollary 1.3

Every finite dimensional normed space is complete.

Proof. This holds for Rn for Euclidean norm. By Theorem 28 it holds for every norms on Rn .
Thus it holds for every finite dimensional normed spaces. □

61

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Corollary 1.4

Every finite dimensional linear subspace of normed space is closed.

Proof. Let Y ⊂ X be a finite dimensional linear subspace and denote by ∥.∥Y the restriction of
the norm on X to the subspace Y. Then (Y, ∥.∥Y ) is complete and hence Y is closed subset of
X. □

Corollary 1.5

Let (X, ∥.∥) be a finite dimensional normed space and let K ⊂ X. Then K is compact if and
only if K is closed and bounded.

Proof. This holds for Euclidean norm on Rn by Heine-Borel Theorem 23. Hence it holds for
every norms on Rn . Hence it holds for every finite dimensional normed spaces. □

Theorem 1.29

Let (X, ∥.∥X ) be a finite dimensional normed space and let (Y, ∥.∥Y ) be any normed space.
Then any linear operator T : X → Y is bounded.

Proof. Let such T be given, we define for every x ∈ X,

∥x∥T := ∥x∥X + ∥T x∥Y .
This is a norm on X. Since dim X is finite, ∥.∥T is equivalent to ∥.∥X by Theorem 28. In
particular, there exists a constant C such that
∥T x∥Y ≤ ∥x∥T ≤ C∥x∥X .
It follows that T is bounded. □

Theorem 1.30

Let (X, ∥.∥) be a normed space and denote the closed unit ball and the closed unit sphere in
X by
B := {x ∈ X | ∥x∥ ≤ 1}, S := {x ∈ X | ∥x∥ = 1}.
Then the following are equivalent

(1). dim X < ∞

(2). B is compact

62

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

(3). S is compact.

Proof. (1) ⇒ (2).

Let X be a normed space with dim X = n and B be its closed unit ball. Then X is homeomorphic
to Rn , so let T : X → Rn be the homeomorphism (by definition T and T −1 are continuous).
Since B is bounded set in X and T is bounded linear operator, then T (B) is bounded. So T (B)
is bounded and closed since T −1 is continuous (by the fact that the inverse images of open sets
under a continuous function are open and similarly for closed sets). Hence T (B) is a closed and
bounded subset of Rn and so T (B) is compact. Since T −1 is continuous then T −1 (T (B)) = B is
compact.
(2) ⇒ (3).
It follows from the fact that a closed subset of a compact set in a topological space is compact.
(3) ⇒ (1).
We argue indirectly and show that if X is infinite-dimensional then S is not compact. Thus
assume X is infinite-dimensional. We claim that there exists a sequence xi ∈ X such that
∥xi ∥ = 1, ∥xi − x j ∥ ≥ 12 for all i, j ∈ N with i , j (*). This is then a sequence in S that does not
have any convergent subsequence and so it follows that S is not compact.
To prove the existence of a sequence in X satisfying (*) we argue by induction. For i = 1,
choose any element xi ∈ S . If x1 , . . . , xk ∈ S satisfy ∥xi − x j ∥ ≥ 12 for i , j, consider the
subspace Y ⊂ X spanned by the vectors x1 , . . . , xk . This is a closed subspace of X by Corollary
4 and is not equal to X as dim X = ∞. Hence Riesz Lemma 3 asserts that there exists a vector
x = xk+1 ∈ S such that ∥x − y∥ ≥ 12 for all y ∈ Y and hence in particular, ∥xk+1 − xi ∥ ≥ 21 for
i = 1, . . . , k. This shows that there exists a sequence xi ∈ X that satisfies (*) for i , j. □

Lemma 1.3: Riesz

Let (X, ∥.∥) be a normed space and let Y ⊂ X be a closed linear subspace that is not equal
to X. Fix a constant 0 < δ < 1. Then there exists a vector x ∈ X such that

∥x∥ = 1, inf ∥x − y∥ ≥ 1 − δ.
y∈Y

Proof. Let x0 ∈ X \ Y. Then d := inf y∈Y ∥x0 − y∥ > 0 because Y is closed. Choose y0 ∈ Y such
d
that ∥x0 − y0 ∥ ≤ 1−δ and define x = ∥x0 − y0 ∥−1 (x0 − y0 ). Then ∥x∥ = 1 and

x0 − y0 − ∥x0 − y0 ∥y d
∥x − y∥ = ≥ ≥1−δ
∥x0 − y0 ∥ ∥x0 − y0 ∥

for all y ∈ Y. □

63

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Banach space of bounded linear operators

In Theorem 26, we have shown that the space of bounded linear operators mapping a normed
space (X, ∥.∥X ) into another normed space (Y, ∥.∥Y ) is a normed space. It is natural to ask when
L(X, Y) is a Banach space. Indeed, it is only the completeness of Y which matters.

Theorem 1.31

Let X be a normed space and let Y be a Banach space. Then L(X, Y), ∥.∥L(X,Y) is a Banach
space with respect to the operator norm.

Proof. Let (T n )n∈N be a Cauchy sequence in L(X, Y). Then

∥T n x − T m x∥Y = ∥(T n − T m )x∥Y ≤ ∥T n − T m ∥∥x∥X

for all x ∈ X and all m, n ∈ N. Hence (T n x)n∈N is a Cauchy sequence in Y for every x ∈ X. Since
Y is complete, this implies that the limit T x := limn→∞ T n x exists for all x ∈ X. This defines a
map T : X → Y.

• T is linear since

T (x + y) = lim T n (x + y) = lim (T n x + T n y)
n→∞ n→∞
= lim T n x + lim T n y = T x + T y
n→∞ n→∞

and

T (λx) = lim T n (λx) = lim λT n x = λ lim T n x = λT x.

n→∞ n→∞ n→∞

• T is bounded since
∥T x∥ = lim ∥T n x∥ ≤ C∥x∥.
n→∞

• We now prove that limn→∞ T n = T . Let ϵ > 0. Since (T n )n∈N is a Cauchy sequence with
repsect to the operator norm, there exists an N ∈ N such that when m, n ≥ N,
ϵ
∥T n − T m ∥ < .
2
Then for any x with ∥x∥X ≤ 1 and any m, n ≥ N,
ϵ
∥T n x − T m x∥Y ≤ ∥T n − T m ∥∥x∥X < .
2
Since T x = limn→∞ T n x, there exists an N1 ∈ N such that when m ≥ N1 ,
ϵ
∥T x − T m x∥Y < .
2
64

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Then when n ≥ N and n ≥ N1 ,

∥T x − T n x∥Y ≤ ∥T x − T m x∥Y + ∥T n x − T m x∥Y
ϵ ϵ
< + ∥x∥X
2 2
= ϵ.
Thus ∥T − T m ∥ < ϵ when n ≥ N and thus limn→∞ T n = T.

Remark 1.24

In particular,

• if X is a Banach space then the space L(X) := L(X, X) of bounded linear operators
from X to itself is a Banach space.

• for any normed space (X, ∥.∥X ) the dual space X ∗ = L(X, R) (or X ∗ = L(X, C) if X is
a complex vector space) is complete as both R and C are complete. (For more details
see Section 1.6).

1.5 Hilbert Space

A Hilbert space is a complete inner product space. An inner product space is generalization of
the n-dimensional Euclidean space to infinite dimensions. A Hilbert space is a Banach space
endowed with scalar product. The main feature of Hilbert space is that it admits both concepts
such as the concept of distance or norm and the concept of orthogonality (i.e., angle between
two vectors) while normed space does not admit the concept of angle between two vectors.
This section is concerned with inner product spaces and Hilbert spaces.
Definition 1.25: Inner product space

Let H be a complex vector space. A mapping

H × H → C : (x, y) 7→ ⟨x, y⟩

is called an inner product if for any x, y, z ∈ H and α, β ∈ C the following conditions are
satisfied

(i). ⟨x, y⟩ = ⟨y, x⟩

(ii). ⟨αx + βy, z⟩ = α⟨x, z⟩ + β⟨y, z⟩

65

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

(iii). ⟨x, x⟩ ≥ 0 with equality if and only if x = 0.

A vector space with an inner product is called an inner product space.

Remark 1.25

By Definition 25, the inner product ⟨x, y⟩ of two vectors is a complex number. By (i).
⟨x, x⟩ = ⟨x, x⟩, it follows that ⟨x, x⟩ is a real number for every x ∈ H. The condition (ii).
implies that

⟨x, αy + βz⟩ = ⟨αy + βz, x⟩ = α⟨y, x⟩ + β⟨z, x⟩ = ᾱ⟨x, y⟩ + β̄⟨x, z⟩.

In particular,
⟨αx, y⟩ = α⟨x, y⟩ and ⟨x, αy⟩ = ᾱ⟨x, y⟩.

Every inner product space is also a normed space with the norm defined by
p
∥x∥ = ⟨x, x⟩.

Indeed, this functional is well defined because ⟨x, x⟩ is always a nonnegative real number and
∥x∥ = 0 if and only if x = 0. In addition,
p q
∥λx∥ = ⟨λx, λx⟩ = λλ̄⟨x, x⟩ = |λ|∥x∥.

It is left with the tirangle inequality where we use the Cauchy-Schwarz inequality.

Lemma 1.4

The inner product and norm satisfy the Cauchy-Schwarz inequality

|⟨x, y⟩| ≤ ∥x∥∥y∥

and the triangle inequality

∥x + y∥ ≤ ∥x∥ + ∥y∥
for all x, y ∈ H. Thus ∥.∥ is a norm on H.

Proof. The Cauchy-Schwarz inequality is obvious when x = 0 or y = 0. Hence we assume

x , 0 and y , 0 and define
x y
ξ= , η=
∥x∥ ∥y∥
Then ∥ξ∥ = ∥η∥ = 1. Hence

0 ≤ ∥η − ⟨η, ξ⟩ξ∥2 = ⟨η, η − ⟨η, ξ⟩ξ⟩ = 1 − |⟨η, ξ⟩|2 .

66

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

This implies |⟨η, ξ⟩| ≤ 1 and hence

|⟨x, y⟩| ≤ ∥x∥∥y∥.
It follows from the Cauchy-Schwarz inequality that
∥x + y∥2 = ∥x∥2 + 2Re⟨x, y⟩ + ∥y∥2
≤ ∥x∥2 + 2|⟨x, y⟩| + ∥y∥2
≤ ∥x∥2 + 2∥x∥∥y∥ + ∥y∥2
≤ (∥x∥ + ∥y∥)2 .
Hence ∥.∥ is a norm. □

Definition 1.26: Hilbert space

A Hilbert space is an inner product space which is complete as a metric space.

Propoistion 1.19

The inner product is continuous.

Proof. Let xn → x and yn → y. Since a convergence sequence implies Cauchy sequence and
Cauchy sequence implies bounded sequence hence yn is bounded. We have
|⟨xn , yn ⟩ − ⟨x, y⟩| ≤ |⟨xn − x, yn ⟩| + |⟨x, yn − y⟩|
≤ ∥xn − x∥∥yn ∥ + ∥x∥∥yn − y∥ → 0
□

Example 1.38: Finite dimensional Hilbert space

• Rn is a Hilbert space with inner product defined by

n
X
⟨x, y⟩ = xi yi ,
i=1

where x = (x1 , . . . , xn ), y = (y1 , . . . , yn ) ∈ Rn .

• Cn is a Hilbert space with inner product defined by

n
X
⟨x, y⟩ = xi ȳi ,
i=1

where x = (x1 , . . . , xn ), y = (y1 , . . . , yn ) ∈ Cn .

67

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

• a finite dimensional inner product space is a Hilbert space.

Example 1.39: Infinite dimensional Hilbert space

Let (X, M, µ) be a measure space then L2 (µ) is a Hilbert space with the inner product
Z
⟨ f, g⟩ = f ḡdµ.
X

Indeed, the norm associated to the given inner product is the L2 -norm.
sZ
∥ f ∥ = ⟨ f, f ⟩ = | f |2 dµ = ∥ f ∥2
p
X

We have already proved that L2 (µ) with the L2 -norm is complete and hence L2 (µ) is a
Hilbert space.
In the case where X = N, it follows that l2 is a Hilbert space with the inner product
∞
X
⟨x, y⟩ = xn ȳn ,
n=1

where x = (x1 , x2 , . . .), y = (y1 , y2 , . . .) ∈ l2 .

Do all norms on vector spaces come from inner products and if not, which property character-
izes inner product spaces?

Proposition 1.20: Parallelogram law

A norm is induced from an inner product if and only if it satisfies, for all x, y,

∥x + y∥2 + ∥x − y∥2 = 2(∥x∥2 + ∥y∥2 ).

The statement asserts that the sum of the lengths squared of diagonals of parallelogram equals
that of the sides.

68

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Proof. (⇒) In any Hilbert space the parallelogram identity holds. It follows from adding the
identities
∥x + y∥2 = ∥x∥2 + 2Re⟨x, y⟩ + ∥y∥2
∥x − y∥2 = ∥x∥2 − 2Re⟨x, y⟩ + ∥y∥2 .
(⇐) If the parallelogram indentity holds in a Banach space then its norm comes from an inner
product which can be recovered in terms of the norm by the polarization identity: for complex
cases
1
⟨x, y⟩ = ∥x + y∥2 − ∥x − y∥2 + i∥x + iy∥2 − i∥x − iy∥2


4
and for real spaces
1
⟨x, y⟩ = ∥x + y∥2 − ∥x − y∥2 .


4
□

Theorem 1.32

Let H be a Hilbert space and let K ⊂ H be a nonempty closed convex subset of H. Then
there exists a unique element x0 ∈ K such that ∥x0 ∥ ≤ ∥x∥ for all x ∈ K.

Proof. Define δ := inf ∥x∥ | x ∈ K ≥ 0.


Existence:
Chose a sequence (xk ) ⊂ K with limk→∞ ∥xk ∥ = δ. We prove that (xk ) is a Cauchy sequence. Let
ε > 0. Then there exists an k0 ∈ N such that k ∈ N, k ≥ k0 , then
ε2
∥xk ∥2 < δ2 + .
4
Let k, l ∈ N such that k ≥ k0 and l ≥ k0 . Then 12 (xk + xl ) ∈ K since K is convex and hence
∥xk + xl ∥ ≥ 2δ. This implies
∥xk − xl ∥2 = 2∥xk ∥2 + 2∥xl ∥2 − ∥xk + xl ∥2
ε2
!
<4 δ +
2
− 4δ2 = ε2 .
4
Thus (xk ) is a Cauchy sequence. Since H is complete the limit x0 = limk→∞ xk exists. Moreover,
x0 ∈ K because K is closed and ∥x0 ∥ = δ because the norm is continuous.
Uniqueness:
Fix two elements x0 , x1 ∈ K with ∥x0 ∥ = ∥x1 ∥ = δ. Then 12 (x0 + x1 ) ∈ K as K is convex and so
∥x0 + x1 ∥ ≥ 2δ. Thus
∥x0 − x1 ∥2 = 2∥x0 ∥2 + 2∥x1 ∥2 − ∥x0 + x1 ∥2 = 4δ2 − ∥x0 + x1 ∥2 ≤ 0
and therefore x0 = x1 . □

69

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Theorem 1.33: Riesz Representation

Let H be a Hilbert space and let Λ : H → R be a bounded linear functional. Then there
exists a unique element y ∈ H such that

Λ(x) = ⟨y, x⟩ for all x ∈ H. (1.5.1)

This element y ∈ H satisfies

|⟨y, x⟩|
∥y∥ = sup = ∥Λ∥. (1.5.2)
0,x∈H ∥x∥

Thus the map H → H ∗ : y 7→ ⟨y, .⟩ is an isometry of normed spaces.

Proof. Existence:
If Λ = 0 then the vector y = 0 satisfies (1.5.1). Hence assume Λ , 0 and define

K := {x ∈ H | Λ(x) = 1}.

Then K , ∅ because there exists an element ξ ∈ H such that Λ(ξ) , 0 and hence x := Λ−1 (ξ)ξ ∈
K. The set K is closed because Λ : H → R is continuous and it is convex because Λ is linear.
Hence Theorem 32 asserts that there exists an element x0 ∈ K such that ∥x0 ∥ ≤ ∥x∥ for all
x ∈ K. We claim that
x ∈ H, Λ(x) = 0 ⇒ ⟨x0 , x⟩ = 0.
Indeed, let x ∈ H such that Λ(x) = 0. Then x0 + tx ∈ K for all t ∈ R. This implies for all t ∈ R,

∥x0 ∥2 ≤ ∥x0 + tx∥2 = ∥x0 ∥2 + 2t⟨x0 , x⟩ + t2 ∥x∥2

Thus the differentiable function t 7→ ∥x0 + tx∥2 attains its minimum at t = 0 and so its derivative
vanishes at t = 0. Hence
d
0= ∥x0 + tx∥2 = 2⟨x0 , x⟩.
dt t=0
This proves that ⟨x0 , x⟩ = 0.
Now define y = ∥x0 ∥−2 x0 . Let x ∈ H and define λ = Λ(x). Then

Λ(x − λx0 ) = Λ(x) − λ = 0.

Hence it follows that

0 = ⟨x0 , x − λx0 ⟩ = ⟨x0 , x⟩ − λ∥x0 ∥2 .
This implies
⟨y, x⟩ = ∥x0 ∥−2 ⟨x0 , x⟩ = λ = Λ(x).

70

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Thus y satisfies (1.5.1).

We prove (1.5.2). Assume that y ∈ H satisfies (1.5.1). If y = 0 then Λ = 0 and so ∥y∥ = 0 = ∥Λ∥.
Hence assume y , 0. Then
∥y∥2 Λ(y) |Λ(x)|
∥y∥ = = ≤ sup = ∥Λ∥.
∥y∥ ∥y∥ 0,x∈H ∥x∥

Conversely, it follows from the Cauchy-Schwarz inequality that

|Λ(x)| = |⟨y, x⟩| ≤ ∥y∥∥x∥ for all x ∈ H
and hence ∥Λ∥ ≤ ∥y∥.
Uniqueness:
Assume y, z ∈ H satisfy ⟨y, x⟩ = ⟨z, x⟩ = Λ(x) for all x ∈ H. Then
⟨y − z, x⟩ = 0 for all x ∈ H.
Take x = y − z to obtain
∥y − z∥2 = ⟨y − z, y − z⟩ = 0
and so y − z = 0. □

Corollary 1.6

Let H be a Hilbert space and let E ⊂ H be a closed subspace. Then H = E ⊕ E ⊥ .

The orthogonal space E of H is defined as

E ⊥ = {y ∈ H | ⟨x, y⟩ for all x ∈ S }.
It follows that E ⊥ is a closed subspace of H since by Cauchy-Schwarz inequality every ⟨x, .⟩ is
continuous on H and \
E⊥ = ker⟨x, .⟩
x∈S
which is an intersection of closed subspaces.

Proof. If x ∈ E ⊕ E ⊥ then ∥x∥2 = ⟨x, x⟩ = 0 and hence x = 0. If x ∈ H, the set K = x + E =

{x + ξ | ξ ∈ E} is closed and convex subset of H. Hence there exists an element ξ ∈ E such that
∥x − ξ∥ ≤ ∥x − η∥ for all η ∈ E. Since x − ξ + tη ∈ E then
∥x − ξ∥2 ≤ ∥x − ξ + tη∥2 = ∥x − ξ∥2 + 2t⟨x − ξ, η⟩ + t2 ∥η∥2
Thus the differentiable function t 7→ ∥x − ξ + tη∥2 attains its minimum at t = 0 and so its
derivative vanishes at t = 0. Hence
d
0= ∥x − ξ + tη∥2 = 2⟨x − ξ, η⟩.
dt t=0
This proves that ⟨x − ξ, η⟩ = 0. Thus x − ξ ∈ E ⊥ ans so x ∈ E ⊕ E ⊥ . □

71

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

1.6 Dual Space

In this section we study further the linear functionals and dual spaces. We begin with finite
dimensional spaces then the dual space of a general Hilbert space and particular Banach spaces.

Definition 1.27

Let X be a normed space over R (or C). The space L(X, R) (or L(X, C)) is called the dual
space of X and is denoted by X ∗ .

Remark 1.26

1. X ∗ = L(X, R) ≡ space of bounded (continuous) linear functionals on X.

2. By Theorem 26, X ∗ = L(X, R) is a Banach space for every normed space X.

Dual space of Rn
Theorem 1.34

∗
There is an isometric isomorphism map T : Rn → Rn that sends each vector α to the
bounded linear operator T α defined by
n
X
T α (x) = αi xi .
i=1

The dual space of Rn can be identified with itself.

Proof. The operator T α is linear for each α ∈ Rn . To see that T α is bounded, we apply the
Cauchy-Schwarz inequality
n
X n
X n
1/2  X 1/2
|T α (x)| ≤ |αi ||xi | ≤ α2i xi2 = ∥α∥2 ∥x∥2 .
i=1 i=1 i=1

This implies that ∥T α ∥ ≤ ∥α∥2 for each α ∈ Rn . It follows that T α is both bounded and linear,

∗
so it is an element of the dual Rn . Since the equality is attained for x = (α1 , . . . , αn ), we

∗
have ∥T α ∥ = ∥α∥2 . To see that T is an isomorphism, consider a map T : Rn → Rn defined
by T (α) = T α . First we show that it is injective. Suppose that T α = T β for α, β ∈ Rn . Then
T α (ei ) = T β (ei ) for each i where (e1 , . . . , en ) is the standard basis of Rn . Thus αi = βi for each i.

72

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

It follows that α = β. Then we prove that it is surjective. Suppose S is an element of the dual

∗
Rn and consider the vector

α = S (e1 ), . . . , S (en ) ∈ Rn .

For any x ∈ Rn , we can write x =

Pn
i=1 xi ei so that
n
X n
X
S (x) = xi S (ei ) = αi xi = T α (x).
i=1 i=1

This implies that S = T α . □

Dual space of Hilbert space

Let H be a Hilbert space, i.e., H is a Banach space
√ and the norm on H arises from an inner
product H × H → R : (x, y) 7→ ⟨x, y⟩ via ∥x∥ = ⟨x, x⟩.

Theorem 1.35

Let H be a Hilbert space. Let y ∈ H define a linear functional Λy : H → R by

Λy (x) = ⟨x, y⟩ for x ∈ H.

Then Λy ∈ H ∗ and
∥Λy ∥ = ∥y∥.
The map

Λ : H → H∗
y 7→ Λy

is an isometric isomorphism.

Theorem 35 tells us that

Every Hilbert spaces is isomorphic to its own dual space.

The main tools to prove theTheorem 35:

• Λy is bounded by the Cauchy-Schwarz inequality.

• Riesz Representation Theorem 33 asserts that the map H → H ∗ : y 7→ Λy is an isometric

isomorphism.

73

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Dual space of L p (µ)

Theorem 1.36

Let (M, F , µ) be a measure space and let 1 < p < ∞. Define the number 1 < q < ∞ by
1 1
+ = 1.
p q
Let g ∈ Lq (µ) and define Λg : L p (µ) → R by
Z
Λg ( f ) = f gdµ for f ∈ L p (µ).
M

Then Λg ∈ L p (µ)∗ and

∥Λg ∥ = ∥g∥q .
The map

Λ : Lq (µ) → L p (µ)∗
g 7→ Λg

is an isometric isomorphism.

The key idea of the proof: The Hölder’s inequality asserts that f ∈ L p (µ), g ∈ Lq (µ) then
f g ∈ L1 (µ) and Z
|Λg ( f )| = f gdµ ≤ ∥ f ∥ p ∥g∥q .
M
It turns out that Λ ∈ L p (µ)∗ and
∥Λg ∥ ≤ ∥g∥q
for all g ∈ Lq (µ). On the other hand, the function f1 = |g|q−2 g satisfies
Z Z Z
| f1 | dµ =
p
|g|(q−1)p
dµ = |g|q dµ < ∞
M M M

Hence f1 ∈ L p (µ). We have

|Λg f1 | ≤ ∥Λg ∥∥ f1 ∥ p
It follows that Z Z  1p
q
|g| dµ ≤ ∥Λg ∥ |g| dµ ,
p
M M
which means
∥g∥q ≤ ∥Λg ∥.
For more details see [6].

74

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Remark 1.27

If (M, F , µ) be σ-finite. The dual of L1 (µ) is L∞ (µ). This may not hold when the measure
is not σ-finite. On the other hand, the dual space of L∞ (µ) is general larger than L1 (µ).

Dual space of l p
Theorem 1.37

Let 1 < p < ∞ and 1

p
+ 1
q
= 1. Let y = (yk ) ∈ lq and define
∞
X
T y (x) = xk yk
k=1

where x = (xk ) ∈ l p . Then T y ∈ (l p )∗ and

∥T y ∥ = ∥y∥q .

The map T : lq → (l p )∗ defined by T (y) = T y is an isometric isomorphism.

The key idea of the proof: Let y ∈ lq and x ∈ l p , we have

∞
X ∞
X
|T y (x)| = xk yk ≤ |xk ||yk | ≤ ∥y∥q ∥x∥ p
k=1 k=1

by Hölder’s inequality. It follows that T y : l p → R is a bounded linear operator, so it is an

element of the dual (l p )∗ and ∥T y ∥ ≤ ∥y∥q . Consider a map T : lq → (l p )∗ defined by T (y) = T y
for each y ∈ lq . On the other hand, if we take xk = |yk |q (yk )−1 then x = (xk ) ∈ l p and
∞
X 1/p ∞
X 1/p ∞
X 1/p
∥x∥ p = |xk | p
= |yk |qp−p
= q
|yk | = ∥y∥q/p
q
k=1 k=1 k=1

Now ∞ ∞
X X
T y (x) = xk yk = |yk |q = ∥y∥qq .
k=1 k=1

Thus
|T y (x)| ∥y∥qq
∥T y ∥ ≥ = q/p
= ∥y∥q−q/p
q = ∥y∥q .
∥x∥q ∥y∥q
For more details see [2].

75

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

76

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Chapter 2

Main Theorems

In this chapter, we will discuss some fundamental results in functional analysis:

• Uniform boundedness principle: every pointwise bounded family of bounded linear op-
erators on a Banach space is bounded.

• Open mapping theorem: every surjective bounded linear operator between two Banach
spaces is open.

• Closed graph theorem: a linear operator between two Banach spaces is bounded if and
only if its graph is a closed linear subspace of the product space.

• Hahn-Banach theorem: every bounded linear functional on a linear subspace of a normed

space extends to a bounded linear functional on the entire normed space.

2.1 Baire’s Category Theorem

The Baire’s Category Theorem implies two most important theorems which make a Banach
spaces useful tools in analysis, the Banach-Steinhaus theorem and the open mapping theorem.

Theorem 2.1: Baire’s Category Theorem

Let X be a nonempty complete metric space and let

[
X= Ak ,
k∈N

with closed set Ak ⊂ X. Then there exists a k0 ∈ N with Åk0 , ∅.

77

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Proof. Assume that Åk = ∅ for all k. Then for U ⊂ Ak open, nonempty, U \ Ak is open,
nonempty for any k ∈ N. Then there exists a ball Bε (x) ⊂ U \ Ak with ε ≤ 1k . Hence we can
inductively choose balls Bεk (xk ) such that
1
Bεk (xk ) ⊂ Bεk−1 (xk−1 ) \ Ak and εk ≤ .
k
Consequently, we see that xl ∈ Bεk (xk ) for l ≥ k and εk → 0 as k → ∞ and the balls Bεk (xk ) are
nested and we conclude that (xl )l∈N is a Cauchy sequence. Hence there exists the limit

x = lim xl ∈ X
l→∞

and x ∈ Bεk (xk ) for all k. Since Bεk (xk ) ∩ Ak = ∅, we have that x < Ak = X, a contradiction.
S
k∈N
□

This is just one of a couple forms of Baire’s Category theorem. Another version is that the
intersection of countably many open dense sets is dense.

Definition 2.1

• A subset A ⊂ X is nowhere dense if Ā˚ = ∅, i.e., A is nowhere dense if the closure of

A has an empty interior.

• A set A is called first category or meager if it can be written as a countable union of

nowhere dense sets; otherwise it is called second category or nonmeager.

Theorem 2.2

Let (X, d) be a complete metric space. Then the intersection of every countable collection
of dense open subsets of X is dense in X.

Proof. The basic idea of the proof: let V1 , V2 , . . . be dense and open in X and let W be any open
T
set in X. We have to show that ( Vk ) ∩ W , ∅.
Since V1 is dense, W ∩ V1 is a nonempty set, and we can therefore find x1 and r1 such that

B̄(x1 , r1 ) ⊂ W ∩ V1 and 0 < r1 < 1.

If k ≥ 2 and xk−1 and rk−1 are chosen, the denseness of Vk shows that Vk ∩ B(xk−1 , rk−1 ) is not
empty and we can therefore find xk and rk such that
1
B̄(xk , rk ) ⊂ Vk ∩ B(xk−1 , rk−1 ) and 0 < rk < .
k
By induction, this process produces a sequence (xk ) in X. If i > k and j > k, the construction
shows that xi and x j both lie in B(xk , rk ) so that d(xi , x j ) < 2rk < 2k , and hence (xk ) is a Cauchy

78

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

sequence. Since X is complete, there is a point x ∈ X such that x = lim xn .

Since x j lies in the closed set B̄(xk , rk ) if j > k, it follows that x lies in each B̄(xk , rk ) and
therefore x lies in each Vn and W. □

Theorem 2 is equivalent to the statement that no complete metric space is of the first category.
To see this, just take complements in the statement of Theorem 2.

2.2 Uniform Boundedness Principle

In this section we present the main consequences of completeness for Banach spaces. Our first
result is the uniform boundedness principle and the Banach-Steinhaus theorem.

Theorem 2.3: Uniform Boundedness Principle

Let X be a nonempty complete metric space and let Y be a normed space. Let F ⊂ C(X, Y)
be a set of functions with

sup ∥ f (x)∥Y < ∞ for every x ∈ X. (∗)

f ∈F

Then there exists an x0 ∈ X and an ε0 > 0, such that

sup sup ∥ f (x)∥Y < ∞.

x∈ Bε0 (x0 ) f ∈F


Proof. For f ∈ F and k ∈ N, it holds that x ∈ X | ∥ f (x)∥Y ≤ k is a closed set (as f is
continuous). Hence the sets
\
Ak = x ∈ X | ∥ f (x)∥Y ≤ k
f ∈F

= x ∈ X | ∥ f (x)∥Y ≤ k for all f ∈ F



being intersection of closed sets, are closed and it follows from (∗) that they form a cover of
X. Then Baire’s Category Theorem 1 yields that Åk , ∅ for some k0 and hence there exists a
Bε0 (x0 ) ⊂ Ak0 . Noting that
sup sup ∥ f (x)∥Y < k0
x∈Ak0 f ∈F

yields the desired result. □

Now we present the Banach-Steinhaus theorem which states that every pointwise bounded
family of bounded linear operators on a Banach space is bounded.

79

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Theorem 2.4: Banach-Steinhaus

Let X be a Banach space and let Y be a normed space. Suppose T ⊂ L(X, Y) with

sup ∥T x∥Y < ∞ for every x ∈ X.

T ∈T

Then T is bounded set in L(X, Y), i.e.,

sup ∥T ∥L(X,Y) < ∞.

T ∈T

Proof. Setting fT (x) = ∥T x∥Y for T ∈ T , x ∈ X defines functions fT ∈ C(X, R) and F =

{ fT | T ∈ T } satisfies the assumptions in Theorem 3. Hence there exists an x0 ∈ X, an ε0 > 0
and a constant C < ∞ with

∥T x∥Y ≤ C for T ∈ T and ∥x − x0 ∥X ≤ ε0 .

Then it follows that for all T ∈ T and all x , 0 that

∥x∥X  x  ∥x∥X
∥T x∥Y = T x0 + ε0 − T (x0 ) ≤ (2C)
ε0 ∥x∥X Y ε0
Hence,
2C
∥T ∥L(X,Y) ≤ .
ε0
□

We remark that there is a notion of convergence in the space L(X, Y) of bounded linear
operators from X to Y in operator norm. A sequence (T n ) in L(X, Y) is uniformly convergent to
T ∈ L(X, Y) if ∥T n − T ∥ → 0 as n → ∞. (In other words, uniform convergence of a sequence
of operators is simply convergence in the operator norm).
A weaker notion of convergence for a sequence of operators is given by the following
definition.
Definition 2.2

A sequence (T n ) in L(X, Y) is strongly convergent if for any x ∈ X the sequence (T n x)

converges in Y. If there is a T ∈ L(X, Y) with limn→∞ T n x = T x for all x ∈ X, then (T n ) is
strongly convergent to T .

Corollary 2.1

Let X be a Banach space and Y be any normed space. If a sequence (T n ) in L(X, Y) is

80

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

strongly convergent, then there exists an operator T ∈ L(X, Y) such that (T n ) is strongly
convergent to T .

Proof. By assumption a sequence (T n ) in L(X, Y) is strongly convergent then for each x ∈ X,

the sequence (T n x) is convergent and hence bounded. By Theorem 4, there is a constant K such
that ∥T n ∥ ≤ K for all n. Hence

∥T n x∥ ≤ K∥x∥ for all x ∈ X. (∗)

Now we define T : X → Y by T x = limn→∞ T n x for all x ∈ X. Then T is linear and by (*)

we have ∥T x∥ ≤ K∥x∥ for all x ∈ X, so T is bounded. The construction of T means that (T n )
converges strongly to T . □

Now we gives an application of Theorem 4 to classical Fourier analysis on T.

Theorem 2.5

There exists a continuous function f ∈ C(T) whose Fourier series diverges at x = 0.

This theorem states that the nth partial sum of Fourier series
n
X
(S n f )(x) = fˆ(k)e2πikx
k=−n

form an unbounded sequence at x = 0. We have S n f is the convolution of f with Dirichlet

kernel Z
(S n f )(x) = Dn (x − s) f (s)ds,
T
where n
X
Dn (x) = e2πix .
k=−n

Indeed,
n Z
X 
(S n f )(x) = f (s)e−2πiks ds e2πikx
k=−n T
n
Z X 
= e2πi(x−s)
f (s)ds
T k=−n
Z
= Dn (x − s) f (s)ds.
T

81

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Lemma 2.1: Dirichlet kernel

The Dirichlet kernel Dn is real valued and can be expressed in the form

1. n
X
Dn (x) = 1 + 2 cos(2πkx).
k=1

2. 
2n + 1,

if x = 0


Dn (x) = 

n+ 21 2πx
 sin sin(πx)

 , if x , 0.

and satisfies Z
Dn (x)dx = 1.
T

In particular, we have
Z Z
(S n f )(0) = f (s)Dn (−s)ds = f (s)Dn (s)ds.
T T

Lemma 2.2

The linear functional T n : C(T) → R defined by

Z
Tn f = f (s)Dn (s)ds
T

is bounded with Z
∥T n ∥ = |Dn (s)|ds.
T

Proof. Let f ∈ C(T) then we have

Z Z Z
|T n f | = f (s)Dn (s)ds ≤ | fn (s)||Dn (s)|ds ≤ ∥ f ∥∞ |Dn (s)|ds,
T T T

Thus, Z
∥T n ∥ ≤ |Dn (s)|ds.
T
Fix δ > 0. Since Dn is analytic it can only have finitely many sign changes in [0, 1]. Therefore,
we can find a continuous function fn with ∥ fn ∥∞ ≤ 1 that differes from sign(Dn (s)) only a finite

82

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

union of intervals whose total length is less than ∥Dδn ∥∞ . The triangle inequality for intergrals
yields Z Z
fn (s)Dn (s) > |Dn (s)|ds − 2δ,
T T
as δ > 0 was arbitrary, this completes the proof. □

Lemma 2.3

The Dirichlet kernel Dn satisfies

sin n + 21 2πs
Z Z


|Dn (s)|ds = ds → ∞
T T sin(πs)
as n → ∞.

Proof. Since | sin x| ≤ |x| for all x ∈ R, it follows that

sin n + 12 2πs
Z

Z 1
1
ds ≥ | sin((2n + 1)πs)|ds.
T sin(πs) 0 πx

Now | sin x| ≥ 1
2
for all x ∈ πZ + [ π6 , 5π
6
]. In particular, it follows that if
2n h
[
i
(2n + 1)πx ∈ k + 1/6 π, k + 5/6 π


k=0

then
1
| sin((2n + 1)πx)| ≥ .
2
Thus
sin n + 12 2πs 2n Z (k+ 5 )π/(2n+1)
Z

X 6 1 1
ds ≥ ds
T sin(πs) 1
k=0 (k+ 6 )π/(2n+1)
π(k + 6 )/(2n + 1) 2
5

1 X 2n + 1 6 π
2n 4
=
2π k=0 k + 56 2n + 1
2n
1X 1
= →∞
3 k=0 k + 5
6
as n → ∞. □

Proof of Theorem 5. By Lemma 2 we have T n f = (S n f )(0) for all f ∈ C(T). Moreover, for
f ∈ C(T), if the Fourier series of f converges at 0, then the family {T n f }n≥1 is bounded. Thus if
the Fourier series of f converges at 0 for all f ∈ C(T), then for each f ∈ C(T) the set {T n f }n≥1
is bounded. By Theorem 4, this implies that the set {∥T n ∥}n≥1 is bounded, which contradicts
Lemma 2 and 3. It follows that there must be some f ∈ C(T) whose Fourier series does not
converge at 0 (and indeed the partial sums must be unbounded). □

83

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

2.3 Open Mapping Theorem

• Let X and Y be topological spaces. Then f : X → Y is called open if U is open in X
implies f (U) is open in Y. If f is bijective then f is open if and only if f −1 is continuous.

• If X and Y are normed spaces and T : X → Y is linear then T is open if and only if there
exists a δ > 0 with
Bδ (0) ⊂ T (B1 (0))
i.e.,
y ∈ Y | ∥y∥Y < δ ⊂ T x | x ∈ X, ∥x∥X < 1 .
 

Indeed, let U be open and let x ∈ U. Choose an ε > 0 with Bε (x) ⊂ U. Now, Bδ (0) ⊂
T (B1 (0)) implies Bεδ (T x) ⊂ T (Bε (x)) ⊂ T (U) and hence T (U) is open.

Theorem 2.6: Open Mapping

Let X and Y be Banach spaces. Then it holds for every operator T ∈ L(X, Y) that

T is surjective ⇐⇒ T is open.

That is, every surjective bounded linear operator between two Banach spaces is open.

Proof. (⇒) Since T is surjective, [

Y= T (Bk (0)).
k∈N

It follows from the Baire’s Category Theorem 1 that there exists a k0 and a ball Bε0 (y0 ) in Y
with
Bε0 (y0 ) ⊂ T (Bk (0)).
This means that for y ∈ Bε0 (0) there exists points xi ∈ Bk0 (0) with T xi → y0 + y as i → ∞. On
choosing an x0 ∈ X with T x0 = y0 , this implies that
!
xi − x0 y xi − x0
T → and < 1,
k0 + ∥x0 ∥ k0 + ∥x0 ∥ k0 + ∥x0 ∥
which proves that
ε0
Bδ (0) ⊂ T (B1 (0)) with δ = . (∗).
k0 + ∥x0 ∥
However, we want to show that such an inclusion holds without the closure of the set on the
right hand side of (∗), for a smaller δ if necessary. We note that (∗) implies that y ∈ Bδ (0)
implies there exists an x ∈ B1 (0) with y − T x ∈ Bδ/2 (0). Then 2(y − T x) ∈ Bδ (0). Hence for
y ∈ Bδ (0) we can inductively choose points yk ∈ Bδ (0) and xk ∈ B1 (0) such that

y0 = y and yk+1 = 2(yk − T xk ).

84

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Then
2−k−1 yk+1 = 2−k yk − T (2−k xk ),
and so  m 
X −k 
T  2 xk  = y − 2−m−1 ym+1 → y as m → ∞.
k=0
Since m m
X X
2 xk <
−k
2−k ≤ 2 < ∞,
k=0 k=0
Pm 
We have that k=0 2−k xk is a Cauchy sequence in X. Because X is complete, there exists
m∈N
x = k=0 2 xk in X with ∥x∥ < 2. The continuity of T then yields that
P∞ −k
 m 
X −k 
T x = lim T  2 xk  = y.
m→∞
k=0

This shows that Bδ (0) ⊂ T (B2 (0)) or equivalently Bδ/2 ⊂ T (B1 (0)). Hence T is open.
(⇐) The fact that Bδ (0) ⊂ T (B1 (0)) for some δ > 0 implies that BR (0) ⊂ T (BR/δ (0)) for all
R > 0. □

Theorem 2.7: Inverse Mapping

If X and Y are Banach spaces and if T ∈ L(X, Y) then

T is bijective ⇐⇒ T −1 ∈ L(X, Y).

That is, every bounded linear operator between two Banach spaces has a bounded inverse.

Proof. T −1 is linear. It follows from Theorem 6 that T is open and T −1 is continuous. □

2.4 Closed Graph Theorem

In this section, we discuss the question: under what conditions will a closed linear operator
be bounded? An answer is given by the followin theorem which is known as the closed graph
theorem.
Definition 2.3

The graph of an operator T is denoted by G(T ) and defined by

G(T ) = (x, y) | x ∈ D(T ), y = T x .



85

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Definition 2.4

Let X and Y be normed spaces and T : D(T ) → Y a linear operator with domain D(T ) ⊂ X.
Then T is called a closed linear operator if its graph

G(T ) = (x, y) | x ∈ D(T ), y = T x



is closed in the normed space X × Y, where the norm on X × Y is defined by

∥(x, y)∥ = ∥x∥ + ∥y∥.

When X and Y are Banach spaces, X × Y is a Banach space with the norm ∥(x, y)∥. Indeed, let
(zn ) be Cauchy in X × Y, where zn = (xn , yn ). Then for every ε > 0 there is an N such that for
all m, n ≥ N,
∥zn − zm ∥ = ∥xn − xm ∥ + ∥yn − ym ∥ < ε, (∗)
Hence (xn ) and (yn ) are Cauchy in X and Y, respectively and converge, say, xn → x and yn → y
since X and Y are complete. This implies that zn → z = (x, y) since from (∗) with m → ∞ we
have ∥zn − z∥ < ε for n > N. Since the Cauchy sequence (zn ) was arbitrary, X × Y is complete.

Theorem 2.8: Closed Graph

Let X and Y be Banach spaces and T : D(T ) → Y a closed linear operator, where D(T ) ⊂
X. Then if D(T ) is closed in X, the operator T is bounded.

Proof. By assumption, G(T ) is closed in X × Y and D(T ) is closed in X. Hence G(T ) and D(T )
are complete. We now consider the mapping

P : G(T ) → D(T )
(x, T x) 7→ x.

• P is linear since for x1 , x2 ∈ X,

P[(x1 , T x1 ) + (x2 , T x2 )] = P[(x1 + x2 , T (x1 + x2 ))] = x1 + x2 = P(x1 , T x1 ) + P(x2 , T x2 ).

• P is bounded since

∥P(x, T x)∥ = ∥x∥ ≤ ∥x∥ + ∥T x∥ = ∥(x, T x)∥.

• P is bijective; in fact the inverse mapping is

P−1 : D(T ) → G(T )

x 7→ (x, T x).

86

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Since G(T ) and D(T ) are complete, we can apply the inverse mapping theorem and conclude
that P−1 is bounded, say, ∥(x, T x)∥ ≤ b∥x∥ for some b and all x ∈ D(T ). Hence T is bounded
because
∥T x∥ ≤ ∥T x∥ + ∥x∥ = ∥(x, T x)∥ ≤ b∥x∥
for all x ∈ D(T ). □

2.5 Hahn-Banach Theorem

The Hahn-Banach theorem is the most important theorem about the structure of linear bounded
functionals on normed spaces. One of the most important questions one may ask of X ∗ is the
following: are there enough elements in X ∗ ? An answer is given by the following Hahn-Banach
theorem.

Theorem 2.9: Hahn-Banach, R version

Let X be a vector space over R. Let p : X → R be a mapping such that

p(αx) = αp(x)
p(x + y) ≤ p(x) + p(y)

for all x, y ∈ X and for all scalars α ≥ 0. Let Y be a subspace of X and let f : Y → R be a
linear map such that
f (x) ≤ p(x) for all x ∈ Y.
Then there exists a linear extension f˜ : X → R of f (i.e. f˜(x) = f (x) for all x ∈ Y) such
that
f˜(x) ≤ p(x) for all x ∈ X.

The proof of Theorem 9 requires the following Zorn’s lemma.

Lemma 2.4: Zorn

Let (M, ≤) be a nonempty partially ordered set (i.e., if m1 ≤ m2 and m2 ≤ m3 , then m1 ≤ m3

and m ≤ m for all m ∈ M) such that every totally ordered subset N (i.e., for all n1 , n2 ∈ N
it holds that n1 ≤ n2 or n2 ≤ n1 ) has an upper bound (i.e., there exists an m ∈ M with
n ≤ m for all n ∈ N). Then M contains a maximal element (i.e., there exists an m0 ∈ M
such that for all m ∈ M it holds that m0 ≤ m ⇒ m ≤ m0 ).

Proof of Theorem 9. Let

n o
M := (Yα , gα ) | Yα subspace, Y ⊂ Yα ⊂ X, gα : Yα → R linear, gα = f on Y, g ≤ p on Yα .

87

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

We make M into a partially ordered set by defining an order

(Yα , gα ) ≤ (Yβ , gβ ) ⇐⇒ Yα ⊆ Yβ and gα = gβ on Yα .

Let N ⊂ M be totally ordered subset and define

[
Y ′ := Yα ,
(Yα ,gα )∈N

g (y) := gα (y), if y ∈ Yα and (Yα , gα ) ∈ N.

′

We need to show that (Y ′ , g′ ) ∈ M. Now Y ⊂ Y ′ ⊂ X and g′ is well defined function as

y ∈ Yα ∩ Yβ , (Yα , gα ) ∈ N, (Yβ , gβ ) ∈ N

Then
(Yα , gα ) ≤ (Yβ , gβ ) or (Yβ , gβ ) ≤ (Yα , gα )
and in the first case,
Yα ⊂ Yβ and gα = gβ on Yα
Thus
gα (y) = gβ (y).
It holds true that g = f on Y and g ≤ p on Y ′ . We can show that Y ′ is a subspace and g′ is
′ ′

linear as follows: for all x, y ∈ Y ′ , λ ∈ R. Then there exists (Yα , gα ) ∈ N, (Yβ , gβ ) ∈ N with
x ∈ Yα and y ∈ Yβ . It follows that

(Yα , gα ) ≤ (Yβ , gβ ) or (Yβ , gβ ) ≤ (Yα , gα ).

This implies x, y ∈ Yξ with ξ = β in the first and ξ = α in the second case, hence also x + λy ∈
Yξ ⊂ Y ′ and
g′ (x + λy) = gξ (x + λy) = gξ (x) + λgξ (y) = g′ (x) + λg′ (y).
Applying Zorn’s Lemma: Indeed, by Zorn’s lemma there is a maximal element (Y0 , g0 ) in M.
We claim that Y0 = X (and we may take f˜ to be g0 ).
Extending by one dimension: Otherwise assume by contradiction that x ∈ X \ Y0 and let Y1 =
span{Y0 , x}. Each element z ∈ Y1 may be expressed uniquely in the form z = y + λx with y ∈ Y0
and λ ∈ R (as x is assumed not to be in the subspace Y0 ). Define a linear function g1 on Y1 by
g1 (y + λx) = g0 (y) + λc, where the constant c will be chosen to guarantee that g1 is bounded by
p. Now if y1 , y2 ∈ Y0 it follows that

g0 (y1 ) − g0 (y2 ) = g0 (y1 − y2 ) ≤ p(y1 − y2 ) ≤ p(y1 + x) + p(−x − y2 ),

Thus
−p(−x − y2 ) − g0 (y2 ) ≤ p(y1 + x) − g0 (y1 ).
This implies

A = sup − p(−x − y) − g0 (y) ≤ inf p(y + x) − g0 (y) = B.

 
y∈Y0 y∈Y0

88

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Choose c to be any number in the interval [A, B]. Then, by construction of A and B,

c ≤ p(y + x) − g0 (y) (2.5.1)

and
−p(−x − y) − g0 (y) ≤ c (2.5.2)
for all y ∈ Y0 . To prove the required bound for g1 we consider different cases as follows:

• For λ > 0, multiply (2.5.1) by λ and substitute y by λ1 y to get

λc ≤ p(y + λx) − g0 (y) (2.5.3)

by the homogeneity property.

• For λ < 0, multiply (2.5.2) by λ and substitute y by λ1 y to obtain

 1  1 
λc ≤ |λ|p − x − y − λg0 y
λ λ
Using the homogeneity assumption on p for |λ| = −λ we get (2.5.3) again.

• For λ = 0, the assumptions of g also give (2.5.3).

Therefore, we get
g1 (y + λx) = g0 (y) + λc ≤ p(y + λx)
for all λ ∈ R and y ∈ Y0 .
A contradiction: Thus we have found (Y1 , g1 ) ∈ M with (Y0 , g0 ) ≤ (Y1 , g1 ) and Y0 , Y1 . This
contradicts the maximality of (Y0 , g0 ) and hence f˜ = g0 is defined on all of X and satisfies the
conclusion of the theorem. □

The functional p is the statement of the Hahn-Banach Theorem 9 is called a sublinear func-
tional.
Definition 2.5

A function p : X → R which satisfies

p(x + y) ≤ p(x) + p(y)

and
p(αx) = αp(x)
for all x, y ∈ X and all α ≥ 0, is called a sublinear functional.

Next we present a version of the Hahn-Banach theorem for vector spaces over C.

89

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Theorem 2.10: Hahn-Banach, C version

Let X be a vector space over C. Let p : X → R be a mapping such that

p(αx) = |α|p(x)
p(x + y) ≤ p(x) + p(y)

for all x, y ∈ X and for all scalars α. Let Y be a subspace of X and let f : Y → C be a linear
map such that
| f (x)| ≤ p(x) for all x ∈ Y.
Then there exists a linear extension f˜ : X → C of f (i.e. f˜(x) = f (x) for all x ∈ Y) such
that
| f˜(x)| ≤ p(x) for all x ∈ X.

Proof. Let X be a vector space over C. We split a linear map f : Y → C into real and imaginary
parts
f (x) = Re f (x) + iIm f (x)
where Re f (x) and Im f (x) are R-linear. By the linearity of f we have
0 = i f (x) − f (ix) = iRe f (x) − Im f (x) − Re f (ix) − iIm f (ix)
= −(Re f (ix) + Im f (x)) + i(Re f (x) − Im f (ix))
for all x ∈ Y. In particular, Im f (x) = −Re f (ix) for all x ∈ Y. Therefore,
f (x) = Re f (x) − iRe f (ix), (∗)
ofr all x ∈ Y. We now consider X as a vector space over R. We denote that vector space by XR .
Let YR be a subspace of XR . Now Re f (x) is a R-linear functional and by assumption
Re f (x) ≤ | f (x)| ≤ p(x)
for all x ∈ YR . By Theorem 9 there exists a linear extension g̃ : XR → R such that g̃|YR = Re f
and g̃(x) ≤ p(x) for all x ∈ XR . We set
f˜(x) = g̃(x) − ig̃(ix)
for all x ∈ XR so that f˜|Y = f by (∗). To show that f˜ is linear from X → C we only need to look
at multiplication by i as f˜ is linear over R. We have
f˜(ix) = g̃(ix) − ig̃(2 x) = g̃(ix) − ig̃(−x) = g̃(ix) + ig̃(x) = i(g̃(x) − ig̃(ix)) = i f˜(x)
for all x ∈ X. It remains to show that | f˜(x)| ≤ p(x). Let x ∈ X and a real number θ such that
| f˜(x)| = eiθ f˜(x). Then since | f˜(x)| ∈ R and by the definition of f˜, we get
| f˜(x)| = eiθ f˜(x) = f˜ eiθ x = g̃ eiθ x ≤ p eiθ x = eiθ p(x) = p(x).

90

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Before stating the version of the Hahn-Banach theorem for linear functionals, we first indi-
cate how to pass from a complex normed space to a real normed space and vice versa.

Lemma 2.5

Let X be a vector space over C. Let f : X → C and u : X → R be functions. Then f is

a C-linear functional on X and u(x) = Re f (x) for all x ∈ X if and only if u is a R-linear
functional on X and f (x) = u(x) − iu(ix) for all x ∈ X.
If, in addition, X is a normed space, then f is a bounded C-linear functional on X if and
only if u is a bounded R-linear functional on X and satisfies ∥ f ∥X∗ = ∥u∥X∗ .

Proof. Suppose f is a C-linear and u(x) = Re f (x) for x ∈ X. Since

1
u(x) = ( f (x) + f (x)),
2
satisfies
u(x1 + x2 ) = u(x1 ) + u(x2 ) and u(αx) = αu(x)
for all x1 , x2 , x ∈ X and α ∈ R then u is R-linear. For all x ∈ X,

f (x) = Re f (x) + iIm f (x) = u(x) − iRei f (x) = u(x) − iRe f (ix) = u(x) − iu(ix).

Conversely, suppose u is R-linear and f (x) = u(x)−iu(ix) for all x ∈ X. Then Re f (x) = u(x)
for all x ∈ X and it follows that f is R-linear. Since

f (ix) = u(ix) − iu(−x) = u(ix) + iu(x) = i[u(x) − iu(ix)] = i f (x)

for all x ∈ X, it follows that f is indeed a C-linear.

Let X be a normed space. Let f : X → C be a C-linear and bounded. The for every unit
vector x ∈ X we have
1 
|u(x)| ≤ | f (x)| + | f (x)| = | f (x)|,
2
which shows that u is bounded and that ∥u∥X∗ ≤ ∥ f ∥X∗ . On the other hand, given a unit vector
x ∈ X, there is a real number θ such that | f (x)| = eiθ f (x). Thus, with y = eiθ x ∈ X, we have

| f (x)| = eiθ f (x) = f eiθ x = Re f (eiθ x) = u(y)

which shows that

∥ f ∥X∗ = sup | f (x)| ≤ sup |u(y)| = ∥u∥X∗ .
x∈X y∈X
∥x∥=1 ∥y∥=1

Hence ∥u∥X∗ = ∥ f ∥X∗ . Similarly, if u is bounded, then f is bounded and ∥ f ∥X∗ = ∥u∥X∗ . □

Now we are ready to state and prove the Hahn-Banach theorem for normed spaces as fol-
lows:

91

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Theorem 2.11: Hahn-Banach (for linear functionals)

Let X be a real or complex normed space and Y a linear subspace. Let f be bounded
linear functional on a subspace Y of a normed space X. There there exists a bounded linear
functional f˜ on X such that f˜(y) = f (y) for all y ∈ Y and

∥ f˜∥X∗ = ∥ f ∥Y ∗

where
∥ f˜∥X∗ = sup | f˜(x)| and ∥ f ∥Y ∗ = sup | f (x)|.
x∈X x∈Y
∥x∥=1 ∥x∥=1

Theorem 11 states that every bounded (continuous) linear functional on a linear space of a
normed space extends to a bounded (continuous) linear functional on the entire normed space.

Proof. The Real Case: Assume first that X is a real normed space. Define p(x) = ∥ f ∥Y ∗ ∥x∥ for
all x ∈ X. We see that p satisfies

p(αx) = ∥ f ∥Y ∗ ∥αx∥ = |α|∥ f ∥Y ∗ ∥x∥ = αp(x)

and
p(x + y) = ∥ f ∥Y ∗ ∥x + y∥ ≤ ∥ f ∥Y ∗ (∥x∥ + ∥y∥) = p(x) + p(y).
Hence we can apply the Theorem 9 and conclude that there exists a linear functional f˜ on X
such that f˜(y) = f (y) for all y ∈ Y and satisfies

f˜(x) ≤ p(x) = ∥ f ∥Y ∗ ∥x∥, x∈X

Further, since
− f˜(x) = f˜(−x) ≤ p(−x) = ∥ f ∥Y ∗ ∥ − x∥ = ∥ f ∥Y ∗ ∥x∥ = p(x)
for all x ∈ X, we see that | f˜(x)| ≤ ∥ f ∥Y ∗ ∥x∥ for all x ∈ X. Hence f˜ is bounded on X and

∥ f˜∥X∗ = sup | f˜(x)| ≤ ∥ f ∥Y ∗ .

x∈X
∥x∥=1

Since f˜(y) = f (y) for all y ∈ Y, it follows that ∥ f˜∥X∗ ≥ ∥ f ∥Y ∗ . Hence, we can conclude that

∥ f˜∥X∗ = ∥ f ∥Y ∗ .

The Complex Case: Now let X be a complex normed space, let Y ⊆ X be a complex linear
subspace. Let f ∈ Y ∗ and define a real linear functional u by u(y) = Re f (y) for y ∈ Y. Let
ũ : X → R be an extension of u with ∥u∥Y ∗ = ∥ũ∥X∗ = ∥ f ∥Y ∗ (by real case above and Lemma 5).
Define f˜ : X → C by f˜(x) = ũ(x) − iũ(ix) for all x ∈ X. By Lemma 5, f is bounded C-linear
functional on X satisfies

f˜(y) = ũ(y) − iũ(iy) = u(y) − iu(iy) = f (y)

92

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

for all y ∈ Y and

∥ f˜∥X∗ = ∥ũ∥X∗ = ∥u∥Y ∗ = ∥ f ∥Y ∗ .
□

Now let see the consequences of the Hahn-Banach theorem. One of the main consequences
is that the dual of a normed linear space is well equipped with functionals as in the following
proposition.

Propoistion 2.1

Let X be a normed space and x ∈ X a non-zero vector. Then there exists f ∈ X ∗ such that
∥ f ∥X∗ = 1 and f (x) = ∥x∥.

Proof. Let Y be the one-dimensional spanned by x. Define g(αx) = α∥x∥ for all scalars α.
Setting p(y) = ∥y∥ we have

g(αx) = |α|∥x∥ = ∥αx∥ = p(αx)

for all scalars α. Hence there exists f ∈ X ∗ such that f (αx) = g(αx) for all α and | f (y)| ≤ p(y) =
∥y∥ for all y in X. Hence f ∈ X ∗ and ∥ f ∥X∗ ≤ 1. Since | f (x)| = ∥x∥ be definition of f we have
∥ f ∥X∗ = 1 and f (x) = g(x) = ∥x∥. □

Propoistion 2.2

Let X be a normed space and Y ⊂ X a proper closed subspace. Then for every x0 ∈ X \ Y
there exists f ∈ X ∗ such that f|Y = 0 and f (x0 ) = 1.

Proof. Let x0 ∈ X \ Y and let Y1 := Y ⊕ span{x0 }. For y + αx0 ∈ Y1 define

g(y + αx0 ) = α,

for all scalars α and y ∈ Y. Then g(x0 ) = 1 and g(y) = 0 for all y ∈ Y. Moreover, since Y is
closed d := dist(x0 , Y) > 0 and

inf ∥y + αx0 ∥ = |α| inf ∥α−1 y − x0 ∥ = |α| inf ∥y − x0 ∥ = d|α|.

y∈Y y∈Y y∈Y

Hence
|g(y + αx0 )| = |α| ≤ d−1 ∥y + αx0 ∥
for all y ∈ Y and all scalars α. Hence g ∈ Y1∗ . Now by Theorem 11 we can conclude that there
exists f ∈ X ∗ such that f|Y = 0 and f (x0 ) = 1. □

93

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Propoistion 2.3

Let X be a normed space. Let x ∈ X. Then

∥x∥ = sup | f (x)| = max∗ | f (x)|.

f ∈X ∗ f ∈X
∥ f ∥X ∗ ≤1 ∥ f ∥X ∗ ≤1

Proof. We observe that | f (x)| ≤ ∥ f ∥X∗ ∥x∥ ≤ ∥x∥ when ∥ f ∥X∗ ≤ 1. On the other hand, by
Proposition 1, there exists f ∈ X ∗ such that ∥ f ∥X∗ = 1 and f (x) = ∥x∥ when x is non-zero. So
the result holds for non-zero vectors and is trivially true for zero vector. □

The Proposition 3 provides a nice property of Banach spaces callled ”reflexivity”.

Let x ∈ X and define
J x ( f ) = f (x),
for f ∈ X ∗ . Then by Proposition 3, it follows that J x ∈ (X ∗ )∗ = X ∗∗ and that in fact

∥J x ∥X∗∗ = ∥x∥X .

Thus J : X → X ∗∗ defined by x 7→ J x is a norm preserving linear operator.

Such a map is called an ”isometry”. The map J is clearly injective and maps X isometrically
onto a subspace of X ∗∗ .

Definition 2.6

A Banach space X is said to be reflexive if the canonical embedding J : X → X ∗∗ is

surjective.

Thus if X is reflexive we can identify the spaces X and X ∗∗ by the isometry J.

Remark 2.1

1. Because X ∗∗ , being the dual space, is always complete, the notion of reflexivity
makes sense only for Banach spaces.

2. Since dual spaces are always complete, every reflexive space is Banach space.

3. Not every Banach space is reflexive. Example of non-reflexive spaces are l1 , l∞ , L1 (µ)
and L∞ (µ).

94

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Example 2.1

Every finite dimensional space is reflexive.

Indeed, the map J, being an isometry, is injective. If X is finite dimensional, then

dim X = dim X ∗ = dim X ∗∗ ,
and so J is also surjective.

Example 2.2

• l p , 1 < p < ∞ is reflexive.

• L p (µ), 1 < p < ∞ is reflexive.

• l2 and L2 (µ) is reflexive.

• Every Hilbert space is reflexive.

We present properties of reflexive spaces. [See [3]]

Theorem 2.12: Kakutani

Let X be a Banach space. Then X is reflexive if and only if the unit ball

B = {x ∈ X | ∥x∥ ≤ 1}

is weakly compact.

Theorem 2.13

Assume that X is a reflexive Banach space and let (xn ) be a bounded sequence in X. Then
there exists a subsequence (xnk ) that converges weakly.

The converse is also true as follows:

Theorem 2.14: Eberlein-Šmulian

Assume that X is Banach space such that every bounded sequence in X admits a weakly
convergent subsequence. Then X is reflexive.

95

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Propoistion 2.4

Assume that X is a reflexive Banach space and let Y ⊂ X be a closed linear subspace of X.
Then Y is reflexive.

Corollary 2.2

A Banach space X is reflexive if and only if its dual X ∗ is reflexive.

Propoistion 2.5

Let X be a Banach space. Then X is reflexive and separable ⇐⇒ X ∗ is reflexive and

separable.

Theorem 2.15: Milman-Pettis

Every uniformly convex Banach space is reflexive.

Example 2.3

1. L p (µ) spaces are uniformly convex for 1 < p < ∞.

2. Hilbert spaces are uniformly convex.

96

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Chapter 3

Compact Operators

3.1 Compact operators

This section focuses on the concept of compact linear operator. Compact operators are a gen-
eralization of operators on finite dimensional spaces. The relation of compactness with weak
convergence and reflexivity, the spectral properties of self-adjoint compact linear operators are
hightlighted. Compact linear operators play a crucial role in the theory of integral equations
and in mathematical physics.

Definition 3.1

Let X and Y be Banach spaces over R or C. Then a linear map T : X → Y is called

a compact operator if for every bounded sequence (xn )n∈N in X the sequence (T xn )n∈N
contains a subsequence that is convergent in Y.

The following characterizations of compact operators are equivalent

1. T is a compact operator.

2. T (B1 (0)) ⊂ Y is compact.

3. T (B1 (0)) ⊂ Y is relatively compact.

4. A ⊂ X is bounded implies T (A) ⊂ Y is relatively compact.

Theorem 3.1

A compact operator is bounded.

97

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Proof. Let T be a compact operator and suppose that T is not bounded. Then for each integer
n ≥ 1 there exists a unit vector xn such that ∥T xn ∥ ≥ n. Since the sequence (xn ) is bounded,
there exists a subsequence (T xnk ) which converges. But this contradicts ∥T xnk ∥ ≥ nk . Hence T
must be bounded. □

Theorem 3.2

T is compact ⇐⇒ for every bounded subset A ⊂ X, the set T (A) ⊂ Y is relatively compact.

Proof. (⇒): Suppose that T is compact. Let A ⊂ X be bounded and let (yn ) be sequence in
T (A). Then for each n ∈ N, there exists xn ∈ A such that ∥yn − T xn ∥ < 1n and the sequence (xn )
is bounded since A is bounded. Thus the sequence (T xn ) contains a convergent subsequence
because T is compact. Hence (yn ) contains a convergent subsequence with limit in T (A). This
shows that T (A) is compact.
(⇐): Suppose that for every bounded subset A ⊂ X the set T (A) ⊂ Y is relatively compact.
Then for any bounded sequence (xn ) in X the sequence (T xn ) lies in a compact set, and hence
contains a convergent subsequence. So T is compact. □

Theorem 3.3

T is compact ⇐⇒ for any sequence of vectors (xn ) in a closed unit ball B1 (0) ⊂ X, the
sequence (T xn ) has a convergent subsequence.

Proof. (⇒): It is trivial by the definition of compact operator.

(⇐): Let (yn ) ⊂ X be a bounded sequence. Then there is an M > 0 such that ∥yn ∥ ≤ M for all
n ∈ N. Thus the sequence M −1 yn lies in the closed unit ball B1 (0) and by the assumption the


sequence T M −1 yn = M −1 T yn has a convergent subsequence and hence (T yn ) must have a



convergent subsequence. It follows that T is compact. □

Theorem 3.4

Let X, Y be normed spaces and T ∈ L(X, Y).

1. If T has a finite rank then T is compact.

2. If either dim X or dim Y is finite then T is compact.

Proof. 1. As T has finite rank, the space Im T is finite-dimensional normed space. More-
over, for any bounded sequence (xn ) in X, the sequence (T xn ) is bounded in Im T , hence
by the Bolzano-Weierstrass theorem this sequence must contain a covergent subsequence.
Thus T is compact.

98

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

2. If dim X is finite then the rank of T , r(T ) ≤ dim X, so r(T ) is finite, while if dim Y is finite
then the dimension of Im T ⊂ Y must be finite. Thus, in either case the result follows
from the first part.
□

Theorem 3.5

Let (T n )n≥1 be a sequence of compact linear operators from a normed space X into a Banach
spaces Y. If ∥T n − T ∥ → 0 then the limit operator T is compact.

Proof. We prove this theorem using a diagonal argument. Since T 1 is a compact operator, then
the sequence (T 1 (xn )) has a convergent (hence Cauchy) subsequence (T 1 (x1,m )) where x1,m is a
subsequence of the sequence (xn ). The subsequence (x1,m ) is bounded, so we can repeat the
argument with T 2 to produce a subsequence (x2,m ) of (x1,m ) with the property that (T 2 (x2,m ))
converges. We continue the process and then define a sequence (ym ) = (xm,m ). Note that (ym )
is a subsequence of (xn ) so it bounded, say by ∥ym ∥ ≤ c, and it has the property that for every
fixed n, the sequence (T n (ym )) is convergent and hence Cauchy.
We claim that (T (ym )) is a Cauchy sequence in Y. Let ϵ > 0. Since ∥T n − T ∥ → 0 there is
some l ∈ N such that ∥T l − T ∥ < 3cϵ . Moreover, since (T l (ym )) is Cauchy, there is some N > 0
such that ∥T l (y j ) − T l (yk )∥ < 3ϵ whenever j, k > N. Therefore, for j, k > N, we have
∥T (y j ) − T (yk )∥ ≤ ∥T (y j ) − T l (y j ) + T l (y j ) − T l (yk ) + T l (yk ) − T (yk )∥
≤ ∥T (y j ) − T l (y j )∥ + ∥T l (y j ) − T l (yk )∥ + ∥T l (yk ) − T (yk )∥
ϵ
< ∥T − T l ∥∥y j ∥ + + ∥T − T l ∥∥yk ∥
3
ϵ ϵ ϵ
< c + + c = ϵ,
3c 3 3c
this implies that (T (ym )) is Cauchy. Since Y is a Banach space, then (T (ym )) is convergent. Thus
we have proved that for any bounded sequence (xn ) ⊂ X, the sequence (T xn ) has a convergent
subsequence; hence T is compact. □

Example 3.1

A linear operator T N : l2 → l2 defined by

T N (x) = (x1 , x2 , . . . , xN , 0, 0, . . .).

Then T N is compact.

Indeed, for any bounded sequence (xn ) ⊂ l2 , the sequence (T N xn ) is a bounded sequence in
RN or CN ⊂ l2 . Since every bounded sequence in RN or CN has a convergent subsequence, it
follows that T N is compact.

99

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Example 3.2

Let T : l2 → l2 be an operator defined by

T x = (λn xn )n≥1 for a sequence λn → 0.

Then T is compact.

Indeed, we approximate T by compact operators T N such that

T N (x) = (λ1 x1 , . . . , λN xN , 0, . . .).

For x ∈ l2 ,
X 1/2
∥T N x − T x∥2 = |λn xn |2
n>N
  X 1/2
≤ sup |λn | |xn |2
n>N n>N
 
≤ sup |λn | ∥x∥2 .
n>N

It follows that ∥T N − T ∥ ≤ supn>N |λn |. Since λn → 0, it follows that ∥T N − T ∥ → 0. Hence T is

compact by Theorem 5.

Theorem 3.6

If dim(X) = ∞ then the identity operator I on X is not compact.

Proof. Since dim(X) = ∞, there exists a sequence of unit vectors (xn ) in X which does not
have any convergent subsequence. Hence the sequence (I xn ) = (xn ) cannot have a convergent
subsequence and thus the operator I is not compact. □

Corollary 3.1

If X is an infinite-dimensional normed space and T is compact on X then T is not invertible.

Proof. Assume that T is invertible. Then the identity operator I = T −1 T on X must be compact.
But since X is infinite-diemensional this contradicts Theorem 6. □

We present further properties of compact operators. [For the proof see [1]]

100

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Theorem 3.7

If X is a normed space, Y is a Banach space and (T n ) be a sequence of bounded, finite rank

operators which converges to T ∈ L(X, Y), then T is compact.

The converse is not true, in general, when Y is Banach space, but it holds when Y is a Hilbert
space.

Theorem 3.8

If X is a normed space, H is a Hilbert space and T is compact operator from X to H, then

there is a sequence of finite rank operators (T n ) which converges to T in L(X, H).

Example 3.3

Every operator on a finite dimensional Hilbert space is compact.

Example 3.4

The identity operator I : l2 → l2 is not compact.

Consider a bounded sequence (en ), where en = (0, · · · , 0, 1, 0, · · · ) with the n th position equals
to 1. Thus for any n , m, √
∥Ien − Iem ∥ = ∥en − em ∥ = 2.
This implies that any subsequence of (Ien ) cannot be Cauchy and hence cannot converge. So I
is not compact.

Example 3.5

Let X = C([0, 1]), ∥.∥∞ and let

K : [0, 1] × [0, 1] → R
(x, y) 7→ K(x, y)

be continuous. Define T : X → X by
Z 1
T ( f )(x) = K(x, y) f (y)dy.
0

101

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Then T is compact.

Note that T is linear by the linearity of integral. For any f ∈ X,

Z 1
|T ( f )(x)| = K(x, y) f (y)dy
0
Z 1
≤ |K(x, y)|| f (y)|dy
0
Z 1
≤ sup | f (x)| |K(x, y)|dy
x∈[0,1] 0
Z 1
≤ ∥ f ∥∞ |K(x, y)|dy.
0

Hence T is bounded.
To prove that T is compact, let A ⊂ X be bounded. Let us prove that T (A) is relatively compact.
First we prove that T (A) is equicontinuous. Let ϵ > 0. Since K is continuous on comapct
metric space then K is uniformly continuous so ∃δ > 0 such that ∀x1 , x2 ∈ [0, 1], |x1 − x2 | < δ
implies |K(x1 , y) − K(x2 , y)| < ϵ. For any f ∈ A,
Z 1
|T ( f )(x1 ) − T ( f )(x2 )| = (K(x1 , y) − K(x2 , y)) f (y)dy
0
Z 1
≤ ∥ f ∥∞ |K(x1 , y) − K(x2 , y)|dy
0
≤ ∥ f ∥∞ ϵ ≤ sup ∥ f ∥∞ ϵ < (1 + sup ∥ f ∥∞ )ϵ.
f ∈A f ∈A

Thus T (A) is equicontinuous. For any x ∈ [0, 1], {T ( f )(x); f ∈ A} is bounded in R so it is rela-
tively compact. Hence Arzéla-Ascoli theorem implies that T (A) is relatively compact.

Definition 3.2

Let (xn ), x ∈ X, we say that (xn ) converges to x weakly if for any φ ∈ X ∗ we have that
w
limn→∞ φ(xn ) = φ(x). The point x is called the weak limit of (xn ), written xn → x or
xn ⇀ x.

Remark 3.1

In particular, if X = (H, ⟨., .⟩) is a Hilbert space then xn ⇀ x if and only if ⟨y, xn ⟩ →
⟨y, x⟩, ∀y ∈ H.

As a consequence of Theorem Kakutani 12, in a reflexive space any bounded sequence has a
weakly convergent subsequence.

102

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Properties:

1. Weak limits are unique.

2. Strong convergence implies weak convergence.

xn → x then xn ⇀ x.

3. Weak convergence and convergence of the norm implies strong convergence.

If xn ⇀ x and ∥xn ∥ → ∥x∥ then xn → x.

4. If xn ⇀ x in X and φn → φ in X ∗ then φn (xn ) → φ(x).

5. If xn ⇀ x then (xn ) is bounded and ∥x∥ ≤ lim inf n ∥xn ∥.

We now present the relation of compactness and weakly convergence as follows:

Theorem 3.9

Let T be a compact operator from a normed space X to a normed space Y. Then T maps
any weakly convergent sequence into a strongly convergent sequence (a sequence that
converges in the norm of X).

Proof. Let x0 , (xn ) ∈ X such that xn ⇀ x. Then T xn ⇀ T x because for φ ∈ Y ∗ ,

|φ(T xn ) − φ(T x0 )| = |(φT )xn − (φT )x0 | → 0,

since φT ∈ X ∗ . Now suppose that T xn does not converge in the norm to T x0 . Then there exists
some ϵ > 0 and a subsequence xnk such that ∥T xnk − T x0 ∥ ≥ ϵ. Since xnk ⇀ x, we know that
(xnk ) is bounded. By the compactness of T , (T xnk ) has a convergent subsequence (T xn′k ) such
that T xn′k → ȳ ∈ Y. By the uniqueness of the weak limit, it follows that ȳ = T x0 , but this
contradicts the fact that ∥T xnk − T x0 ∥ ≥ ϵ. □

Remark 3.2

Let X, Y be reflexive spaces and let T : X → Y be a compact operator. If (xn ) ⊂ X

is a weakly convergent sequence in X and xn ⇀ x then (T xn ) ⊂ Y has a convergence
susequence (strongly) in Y,
T (xnk ) → T (x) in Y
In particular, if xn ⇀ x in H 1 (Ω) then there is a subsequence

xnk → x in L2 (Ω).

103

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

That is, weak convergence in H 1 (Ω) implies the existence of a subsequence that is strongly
convergent in L2 (Ω).

As an application of this result, we present Poincaré inequalities as follows:

Theorem 3.10

Let Ω ⊂ Rn be a bounded domain with boundary ∂Ω partitioned as ∂Ω = Γ∪Γ′ and |Γ| > 0.
Let H = u ∈ H 1 (Ω) | u|Γ = 0 . Then there exists a constant C(Ω) such that


Z Z
2
u dx ≤ C(Ω) |∇u|2 dx, ∀u ∈ H.
Ω Ω

3.2 Spectral theorem for self-adjoint compact linear opera-

tors
In this section we consider H a complex Hilbert space and T : H → H a compact operator.
We will discuss about the spectral theory of T ; that is, to describe the structure of the spectrum
σ(T ).
Spectral theory is a generalization of the theory of eigenvalues and eigenvectors of matri-
ces to operators on infinite dimensional spaces. If H is finite dimensional then we know that
the spectrum σ(T ) consists of a non-empty, finite collection of eigenvalues, each having finite
multiplicity. One of the most important results is that for every self-adjoint linear map T on
finite-dimensional space, there exists a basis consisting of eigenvectors. In particular, with rep-
sect to this basis the opeartor T can be represented by a diagonal matrix. The situation turns
out to more complicated for operators in infinite dimensional spaces.
For general operators in infinite-dimensional spaces the spectrum can be very different,
but for compact operators the spectrum has many similarities with the finite-dimensional case.
Compact operators appear frequently in the theory of partial differential equations and else-
where. Remarkably, every self-adjoint compact operator on Hilbert space can be diagonalized
with respect to suitable orthonormal set.

Definition 3.3

Let H be a Hilbert space and let T ∈ L(H). The resolvent of T , denoted by ρ(T ), is the set
of complex numbers λ such that (T − λI) : H → H is one-to-one and onto. The spectrum
of T , denoted by σ(T ), is the complement of the resolvent in C,

σ(T ) = C \ ρ(T ).

104

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 Introduction to Functional Analysis MEAS Len

Remark 3.3

If T − λI is one-to-one and onto then the open mapping theorem implies that (T − λI)−1 is
bounded. Hence, when λ ∈ ρ(T ) both T − λI and (T − λI)−1 are bounded.

Now we introduce the adjoint operators, which is guaranteed to exist by Riesz representation
theorem 33.
Theorem 3.11

Let T ∈ L(H) be a bounded operator on H. Then there exists a unique operator T ∗ ∈ L(H)
such that
⟨T x, y⟩ = ⟨x, T ∗ y⟩ for all x, y ∈ H.
The operator T ∗ is called the adjoint of T .

For the existence, let y ∈ H, and define a map

φ(x) = ⟨T x, y⟩, x ∈ H.
This is a bounded linear functional on H, as it is easily seen to be linear and
|φ(x)| = |⟨T x, y⟩| ≤ ∥T x∥∥y∥ ≤ ∥T ∥∥y∥∥x∥.
By Riesz representation theorem 33 it follows that there exists a unique element y∗ ∈ H such
that
φ(x) = ⟨T x, y⟩ = ⟨x, y∗ ⟩, x ∈ H.
We thus define T ∗ y = y∗ so by definition T ∗ satisfies ⟨T x, y⟩ = ⟨x, T ∗ y⟩.

Definition 3.4: Self-adjoint

A bounded linear operator T : H → H is said to be self-adjoint if T = T ∗ .

Theorem 3.12: Spectral Theorem-Compact Operators

Let H be a Hilbert space and let T : H → H be a compact self-adjoint operator. Then there
exists an orthornormal set (en )n≥1 consisting of eigenvectors of T with eigenvalues λn such
that ∞
X
Tx = λn ⟨x, en ⟩en , x ∈ H. (3.2.1)
n=1

To prove this theorem we need the following Lemmas. For the details proof see [10].

105

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 MEAS Len Introduction to Functional Analysis

Lemma 3.1

Let T : H → H be bounded and self-adjoint. Then

• all eigenvalues are real.

• eigenvectors belonging to distinct eigenvalues are orthogonal.

Lemma 3.2

Suppose T is compact and λ , 0. Then

• dim ker(T − λI) is finite.

• the eigenvalues of T form an at most countable set λ1 , . . . , λk , . . . with λk → 0 as

k → ∞.

• for each µ > 0 the space spanned by the eigenvectors corresponding to all λk with
|λk | > µ is finite dimensional.

The expression for the norm of self-adjoint operator is ∥T ∥ = sup∥x∥=1 |⟨x, T x⟩|. Hence we either
have
∥T ∥ = sup |⟨x, T x⟩| or ∥T ∥ = − inf |⟨x, T x⟩|.
∥x∥=1 ∥x∥=1

Lemma 3.3

Suppose T , 0 be compact and self-adjoint. Then either ∥T ∥ or −∥T ∥ is an eigenvalue of

T.

Therefore, the representation (3.2.1) of the compact self-adjoint operator T is an infinite dimen-
sional version of the well-known result in finite dimensional linear algebra that a self-adjoint
matrix can be diagonalized by a basis consisting of eigenvectors of the matrix.

106

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44
 References

[1] B. P. Rynne, M. A. Youngson, Linear Functional Analysis, 2nd edition, Springer (2008).

[2] E. Kreyszig, Introductory Functional Analysis with Applications, Wiley (1989).

[3] H. Brezis, Functional Analysis, Sobolev Spaces and Partial Differential Equations,
Springer (2011).

[4] H. W. Alt, Linear Functional Analysis: An Application-Oriented Introduction, Springer

(2012).

[5] J. Muscat, Functional Analysis: An Introduction to Metric Spaces, Hilbert Spaces, and
Banach Algebras, Springer (2014).

[6] L. F. Richardson, Measure and Integration: A Concise Introduction to Real Analysis,

Wiley (2009).

[7] M. Einsiedler, T. Ward, Functional Analysis, Spectral Theory, and Applications, Springer
(2017).

[8] R. L. Wheeden, A. Zygmund, Measure and Integral: An Introduction to Real Analysis,

2nd edition, CRC Press (2015).

[9] S. Salsa, Partial Differential Equations in Action: From Modelling to Theory, Springer
(2016).

[10] V. L. Hansan, Functional Analysis: Entering Hilbert Space, WSPC (2006).

[11] W. Rudin, Principles of Mathematical Analysis, McGraw Hill, 3rd edition (1976).

107

AMS Open Math Notes: Works in Progress; Reference # OMN:202208.111337; Last Revised: 2022-08-28 11:06:44