KARNATAKA STATE OPEN UNIVERSITY

Mukthagangothri, Mysuru – 570006

M.Sc. MATHEMATICS (CBCS)

(THIRD SEMESTER)

Course-MMDSC 3.3

Functional Analysis
M.Sc. MATHEMATICS (CBCS)
THIRD SEMESTER

Course: MMDSC 3.3

Functional Analysis
Programme Name: M.Sc. Mathematics (CBCS) Year/Semester: III Semester
Course Code: MMDSC 3.3 Course Name: Functional Analysis
Credit: 4 Unit Number : 1-16
COURSE DESIGN COMMITTEE
Dr. Sharanappa. V. Halse Chairman
Vice Chancellor
Karnataka State Open University
Mukthagangothri, Mysuru-570006
Prof. N. Lakshmi Member
Dean (Academic)
Karnataka State Open University
Mukthagangothri, Mysuru-570006
Dr. Pavithra. M Course coordinator
Assistant Professor
DoS in Mathematics, KSOU, Mukthagangothri, Mysuru-06
EDITORIAL COMMITTEE
1. Dr. K. Shivashankara Chairman
BOS Chairman(PG), DoS in Mathematics, KSOU
Associate Professor, Yuvaraja College,
University of Mysore, Mysuru-06

2. Mr. S. V. Niranjana Member & Convener

Coordinator, (DoS in Mathematics)
Assistant Professor, DoS in Physics,
KSOU, Mysuru-06

3. Dr. Pavithra. M Member

Assistant Professor
DoS in Mathematics, KSOU, Mysuru-06

4. Dr. Chandru Hegde Member

Assistant Professor,
DoS in Mathematics,
Mangalagangotri, Mangaluru.
 COURSE WRITER

Prof. H. S. Ramane Block 3.3 A to Block 3.3 D

Professor (Block I – IV)
Department of Studies in Mathematics (Unit 01 to Unit 16)
Karnatak Unuiversity, Dharwad

COURSE EDITOR

Dr. K. R. Vasuki Block 3.3 A to Block 3.3 D

Professor (Block I – IV)
DOS in Mathematics (Unit 01 to Unit 16)
University of Mysore, Mysuru

The Registrar
Karnataka State Open University
Mukthagangothri, Mysuru-570006

Developed by the Department of Studies in Mathematics under the guidance of Dean

(Academic), KSOU, Mysuru.
Karnataka State Open University, 2023.
All rights reserved. No part of this work may be reproduced in any form or any other means, without
permission in writing from the Karnataka State Open University.
Further information on the Karnataka State Open University Programmes may be obtained from the
University’s Office at Mukthagangothri, Mysuru – 570 006.
 KARNATAKA STATE OPEN UNIVERSITY

Mysuru - 570006, INDIA

MMDSC 3.3: Functional Analysis

Dr. H. S. Ramane
Department of Mathematics, Karnatak University, Dharwad - 580003
Preface
Functional analysis is the branch of Mathematical analysis formed by the study
of vector spaces and the linear functions defined on these spaces. It helps to study
and solve linear and nonlinear problems posed on the normed spaces.

This book covers the fundamental results of Functional analysis, which are
importatnt in view of applications.

This book is divided into four blocks, each block contains four units. The
first block discusses the metric space, mapping theorem, Baires theorem and Ascoli-
Arzela theorem. Second block covers linear spaces, normed spaces, operators and
Hahn-Banach theorem. Third block covers Banach space, open mapping theorem,
closed graph theorem and principle of uniform boundedness. Fourth block covers
Hilbert space, orthogonality and Riesz representation theorem.

Suggestions for improvement of the book will be thankfully received.

Dr. H. S. Ramane
Department of Mathematics
Karnatak University,
Dharwad - 580003

i
Contents

1 Unit 1: Metric completion 2

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3 Complete Metric Space . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

1.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

2 Unit 2: Banach’s Contraction Mapping Theorem and Applications 28

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

2.2 Banach Contraction Mapping Theorem . . . . . . . . . . . . . . . . . 28

2.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

2.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3 Unit 3: Baire’s Category Theorem 38

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3.2 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

ii
 3.3 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

4 Unit 4: Ascoli-Arzela Theorem 47

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4.2 Ascoli-Arzela Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

4.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

5 Unit 5: Linear Spaces and Linear Operators 63

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

5.2 Linear Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

5.3 Linear operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

5.4 Exercises: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

5.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

6 Unit 6: Norm of bounded operator 74

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

6.2 Normed Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

6.3 Bounded linear operator . . . . . . . . . . . . . . . . . . . . . . . . . 77

6.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

6.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

7 Unit 7: Hahn-Banach Extension Theorem 80

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

7.2 Hahn-Banach Theorems . . . . . . . . . . . . . . . . . . . . . . . . . 80

iii
 7.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

7.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

8 Unit 8: Stone-Weierstrass Theorem 93

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

8.2 Stone-Weierstrass Theorem . . . . . . . . . . . . . . . . . . . . . . . . 94

8.3 Exercises: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

8.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

9 Unit 9: Banach Spaces 103

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

9.2 Banach Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

9.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

9.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

10 Unit 10: Open Mapping Theorem 109

10.1 Introducation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

10.2 Open Mapping Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 109

10.3 Exercise: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

10.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

11 Unit 11: Closed Graph Theorem 118

11.1 Introducation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

11.2 Closed Graph Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 118

11.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

iv
 11.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

12 Unit 12: Banach - Steinhaus Principle of Uniform Boundedness 127

12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

12.2 Uniform Bounded Principle . . . . . . . . . . . . . . . . . . . . . . . 127

12.3 Properties of T ∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

12.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

12.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

13 Unit 13: Hilbert Spaces 137

13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

13.2 Hilbert space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

13.3 Exercises: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

13.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

14 Unit 14: Properties of Hilbert Spaces 142

14.1 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

14.2 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

14.3 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

15 Unit 15: Orthogonal projection and nearly orthogonal elements 146

15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

15.2 Orthogonal projection . . . . . . . . . . . . . . . . . . . . . . . . . . 146

15.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

15.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

v
16 Unit 16: Riesz Lemma and Riesz Representation Theorem 156

16.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

16.2 Riesz Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

16.3 Riesz representation theorem . . . . . . . . . . . . . . . . . . . . . . . 157

16.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

16.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160

vi
Block - I

1
Chapter 1

Unit 1: Metric completion

1.1 Introduction

In this unit, we introduce the concept of metric spaces. We give the

properties of metric spaces and some theorems to metric space. Finally
we expalin the metric completion theorem.

1.2 Metric Spaces

Definition 1.2.1. Let X be a nonempty set. Let d be a function

d : X × X −→ R satisfying the following.
(1) d(x, y) ≥ 0, and d(x, y) = 0 ⇐⇒ x = y ∀x, y ∈ X.
(2) d(x, y) = d(y, x), ∀x, y ∈ X (Symmetric property).
(3) d(x, y) ≤ d(x, z) + d(z, y) ∀x, y, z ∈ X. (Triangle inequality).
Then d is called a metric on X. (X, d) is called a metric space.

2
Remark 1.2.2. (1) A metric d is always non-negative. For x, y ∈ X
it follows that
d(x, y) + d(y, x) ≥ d(x, x)

That is 2d(x, y) ≥ 0 and hence d(x, y) ≥ 0.

(2) If x, y, x′ , y ′ ∈ X it should be noted that

|d(x, y) − d(x′ , y ′ )| ≤ d(x, x′ ) + d(y, y ′ ).

Example 1.2.3. (1) X = R1 , define d(x, y) = |x − y|, then it can be

verified that d is metric on R1 . This metric d is called usual metric
on R1 . Similarly on X = C1 , define d(z1 , z2 ) = |z1 − z2 |. It is easy to
verify d is a metric on C1 . This metric d is called usual metric on C1 .
(2) X be any nonemepty set.

 0, x = y;
d(x, y) =
̸ y.
 1, x =

It is easy to verify that d is a metric on X. d is called discrete metric

on X. This shows any nonempty set can be regarded as a metric space.

Definition 1.2.4. Let (X, d) be a metric space and let Y be a nonempty

subset of X, then (Y, d) is called a subspace of (X, d).

Definition 1.2.5. Open ball in a metric space (X, d) is defined by

Sr (x0 ) = {x ∈ X : d(x, x0 ) < r}.

Definition 1.2.6. A set G ⊆ X is sad to be an open set if x ∈ G ⇒

x ∈ Sr (x) ⊆ G.

3
Definition 1.2.7. Let (X, d) be a metric space. Let A ⊆ X. A point
x ∈ X is said to be a limit point of A if every neighbourhood (open
ball around x) of x contains atleast one point of A.

Definition 1.2.8. Let (X, d) be a metric space. Let A ⊆ X, then

closure of A denoted by A is nothing but set of all points x ∈ X such
that either x ∈ A or x is a limit point of A.

Definition 1.2.9. Let (X, d) be a metric space. A set F ⊆ X is said

to be closed if and only if F = F (F contains all its limit points).

Definition 1.2.10. Let (X,d) be a metric space. Let A ⊆ X is said

to be dense in X or everywhere dense in X if and only if A = X.

Definition 1.2.11. Let A ⊆ X. Then diameter of A is given by

d(A) = sup d(x, y), (d(A) ≤ ∞).

x,y∈A

Definition 1.2.12. A set A ⊆ X is said to be a bounded set if d(A) <

∞ (d(A) ̸= ∞).

Definition 1.2.13. Let A, B be the subsets of X. Then distance

between A and B is given by

d(A, B) = inf d(x, y).

x∈A,y∈B

Definition 1.2.14. Let (X, d1 ) and (Y, d2 ) be two metric spaces. A

mapping f : X −→ Y is said to be an isometry if

d1 (x1 , x2 ) ≥ d2 (f (x1 ), f (x2 )) ∀ x1 , x2 , ∈ X.

4
 If such an isometry exists then (X, d1 ) and (Y, d2 ) are said to be
isometric to each other.

Theorem 1.2.15. A set G is open in a metric space (X, d) if and

only if G is a union of open spheres.

Proof. Suppose G is open set in a metric space (X, d).

x ∈ G ⇒ ∃ rx > 0 ∋ x ∈ Srx (x) ⊆ G.

⇒ {x} ⊆ Srx (x) ⊆ G.

[ [
⇒ {x} ⊆ Srx (x) ⊆ G.
x∈G x∈G
[
⇒ G⊆ Srx (x) ⊆ G.
x∈G
[
⇒ G= Srx (x).
x∈G

S
Suppose G = Sα (arbitrary union of open sphere i.e., ∀ α, Sα
α∈λ
is an open sphere in G).
S
If x ∈ G ⇒ ∃ α ∈ λ ∋ x ∈ Sα and Sα ⊆ Sα = G.
α∈λ

⇒ x ∈ Sα ⊆ G. Therefore G is open.

Theorem 1.2.16. If A is a finite in a metric space (X, d), then A′

(or Ac or ∼ A) complement of A is open set.

Proof. Let A = {a1 , a2 , . . . , an }. Let x ∈ A′ , then x ∈

/ A or x ̸= ai ,

5
1 ≤ i ≤ n or d(x, ai ) > 0, 1 ≤ i ≤ n. Set r = min d(r, ai )
1≤i≤n

y ∈ Sr (x) ⇒ d(x, y) < r.

To show that y ̸= ai , i ≤ i ≤ n or d(y, ai ) > 0, 1 ≤ i ≤ n.

Consider d(x, ai ) ≤ d(x, y) + d(y, ai )

⇒ d(y, ai ) ≥ d(x, ai ) − d(x, y) > d(x, ai ) − r ≥ 0

⇒ d(y, ai ) > 0

/ A or y ∈ A′ ⇒ y ∈ Sr (x) ⊆ A′ .
Therefore y ̸= ai or y ∈

∴ A′ is open.

Theorem 1.2.17. A set F is closed set in (X, d) if and only if F ′ is

open.

Proof. Suppose F is a closed set in (X, d).

Let x ∈ F ′ , then x ∈
/ F.

⇒ x is not a limit point of F .

⇒ ∃ an open sphere Sr (x) such that Sr (x) ∩ F = ∅.

⇒ x ∈ Sr (x) ⊆ F ′ .

∴ F ′ is open.

Conversely, suppose F ′ is open.

x ∈ F ′ ⇒ x ∈ Sr (x) ⊆ F ′

⇒ Sr (x) ∩ F = ∅

6
 ⇒ x is not a limit point of F .

⇒ No points of F ′ is a limit point of F .

⇒ Every limit point of F is in F .

∴ F is closed.

Problem 1.2.18. (1) Show that A = {x ∈ X : d(x, A) = 0}

Solution: Let {x ∈ X : d(x, A) = 0} = B.

To show that A ⊆ B and B ⊆ A:

If x ∈ A ⇒ either x ∈ A or x is a limit point of A. If x ∈ A,

d(x, A) = d(x, x) = 0.

Suppose x is a limit point of A, ⇒ for every r > 0, Sr (x) ∩ A ̸= ∅.

∴ y ∈ Sr (x) ∩ A ⇒ y ∈ A and d(x, A) < r, ∀ r > 0.

Allow r → 0, hence d(x, A) = 0.

∴ x ∈ B or A ⊆ B.

If x ∈ B ⇒ d(x, A) = 0.

⇒ inf d(x, y) = 0.
y∈A

⇒ Given ϵ > 0, ∃ y ∈ A such that d(x, y) < ϵ.

⇒ Sϵ (x) ∩ A ̸= ∅, ∀ ϵ > 0.

⇒ x is a limit point of A.

7
 ⇒ x ∈ A or B ⊆ A.

Hence A = B.

(2) Prove that a set A in X is dense in X or A = X, if and only if the

closed superset of A is X.

Solution: Suppose A = X and B is a closed superset of A.

⇒ A = X, A ⊆ B and B = B.

A ⊆ B ⇒ A ⊆ B ⇒ X ⊆ B ⊆ X ⇒ B = X.

∴ X is the only closed superset of A.

Suppose the only closed suoerset of A is X.

By the definition, A is the largest closed set containing X or A

is a closed superset of A.

⇒ A = X.

∴ A is dense in X.

1.3 Complete Metric Space

Simplest example of complete metric space is R or real line. The

following three fundamental concepts of real analysis are to describe
the completeness of R1 .

Definition 1.3.1. A sequence {xn } of real numbers is said to be a

8
Cauchy sequence if given ϵ > 0, ∃ a positive integer N such that
m, n ≥ N
⇒ |xm − xn | < ϵ or dR (xm , xn ) < ϵ.

Definition 1.3.2. A sequence {xn } of real numbers is said to be a

convergent to x0 if given ϵ > 0, ∃ a positive integer N such that
n≥N
⇒ |xn − x0 | < ϵ or dR (xn , x0 ) < ϵ.

Note 1.3.3. Every convergent sequence is a Cauchy sequence. But

converse is not true in general.

Definition 1.3.4. A sequence {xn } in a metric space (X, d) is said to

be a Cauchy sequence if given ϵ > 0, ∃ a positive integer N such that
m, n ≥ N

⇒ d(xm , xn ) < ϵ.

Definition 1.3.5. A sequence {xn } in a metric space (X, d) is said to

be a convergent sequence with limit if given ϵ > 0, ∃ a positive integer
N such that n ≥ N , ⇒ d(xn , x0 ) < ϵ.

Definition 1.3.6. A metric space (X, d) is said to be a Complete

metric space if every Cauchy sequence is convergent in X.

Example 1.3.7. (1) X = Rn and

p
d((x1 , x2 , . . . , xn ), (y1 , y2 , . . . , yn )) = (x1 − y1 )2 + (x2 − y2 )2 + · · · + (xn − yn )2 .

Then (X, d) is a metric space.

9
 (n) (n) (n)
Put x = (x1 , x2 , . . . , xn ), {x(n) } = {(x1 , x2 , . . . , xn )} is a se-
quence in Rn .

Suppose {x(n) } is a Cauchy sequence

⇒ Given ϵ > 0, ∃ a positive integer N such that m, p ≥ N ⇒

d(x(m) , x(p) ) < ϵ
(m) (m) (m) (p) (p) (p)
⇒ d((x1 , x2 , . . . , xn ), (x1 , x2 , . . . , xn ) < ϵ
q
(m) (p) (m) (p) (m) (p)
⇒ (x1 − x1 )2 + (x2 − x2 )2 + · · · + (xn − xn )2 < ϵ.
(m) (p) (m) (p) (m) (p)
⇒ (x1 − x1 )2 + (x2 − x2 )2 + · · · + (xn − xn )2 < ϵ2 .
(m) (p)
⇒ |xi − xi | < ϵ, ∀ i = 1, 2, . . . , n.
(m)
⇒ {xi } is a Cauchy sequence in R1 .

We know that R1 is complete and hence for each i = 1, 2, . . . , n,

(m)
{xi } is convergent sequence.

∴ Rn is complete with Euclidean metric.

(2) Cn = {(z1 , z2 , . . . , zn ) : z1 , z2 , . . . , zn ∈ C}.

(m) (m) (m)
Put z = (z1 , z2 , . . . , zn ) ∈ Cn , {z (m) } = {(z1 , z2 , . . . , zn )} is
a sequence in Cn . Put w = (w1 , w2 , . . . , wn ) ∈ Cn ,
q
d(z, w) = (z1 − w1 )(z1 − w1 ) + (z2 − w2 )(z2 − w2 ) + · · · + (zn − wn )(zn − wn )
p
⇒ d(z, w) = |z1 − w1 |2 + |z2 − w2 |2 + · · · + |zn − wn |2 , ∵ |z|2 = zz.

(i) d(z, w) ≥ 0 (sum of sequences of absolute values of complex num-

10
bers are non-negative).

d(z, w) = 0 ⇐⇒ |zi − wi |2 = 0, ∀ 1 ≤ i ≤ n.

⇐⇒ |zi − wi | = 0, ∀ 1 ≤ i ≤ n.

⇐⇒ zi = wi , ∀ 1 ≤ i ≤ n.

⇐⇒ (z1 , z2 , . . . , zn ) = (w1 , w2 , . . . , wn )

⇐⇒ z = w.

(ii) d(z, w) = d(w, z)

|zi − wi |2 = (zi − wi )(zi − wi )

= (zi − wi )(zi − wi )

= {−(wi − zi )}{−(wi − zi )}

= (wi − zi )(wi − zi )

= |wi − zi |2 .

As a direct consequence we get the symmetry.

(iii) d(z, w) ≤ d(z, t) + d(t, w)

p
d(z, w) = |z1 − w1 |2 + |z2 − w2 |2 + · · · + |zn − wn |2
p
= |z1 − t1 + t1 − w1 |2 + · · · + |zn − tn + tn − wn |2
p
≤ |z1 − t1 |2 + |z2 − t2 |2 + · · · + |zn − tn |2
p
+ |t1 − w1 |2 + |t2 − w2 |2 + · · · + |tn − wn |2

(Cauchy Schwartz inequality)

d(z, w) ≤ d(z, t) + d(t, w).

11
 ∴ (Cn , d) is metric space.
(m) (m) (m)
Suppose {Z (m) } = {z1 , z2 , . . . , zn } is Cauchy sequence.
(m)
⇒ {zi } is Cauchy sequence in C and C is complete.

⇒ {z (m) } is also convergent.

Hence (Cn , d) is a Complete metric space.

Definition 1.3.8. (N, +, ·, F ) is said to be a Linear space over a field

F if N ̸= ∅ and if the following axioms are satisfied.

(i) x, y ∈ N ⇒ x + y ∈ N (Vector addition).

(ii) α ∈ F, x ∈ N ⇒ α · x ∈ N (Scalar multiplication)

(iii) x, y, z ∈ N ⇒ x + (y + z) = (x + y) + z (Associative law).

(iv) x, y ∈ N ⇒ x + y = y + x (Commutative law).

(v) ∃ 0N ∈ N such that x ∈ N ⇒ x+0N = x (0N is Identity element).

(vi) x ∈ N ⇒ ∃ − x ∈ N such that x + (−x) = 0N (Inverse element

law).

(vii) α ∈ F, x, y ∈ N ⇒ α · (x + y) = α · x + α · y (Scalar multiplication

is distributive with respect to Vector addition)

(viii) α, β ∈ F, x ∈ N ⇒ (α + β) · x = α · x + β · x.

(ix) α, β ∈ F, x ∈ N ⇒ α · (β · x) = (αβ) · x.

(x) x ∈ N ⇒ 1 · x = x (1 is multiplicative identity of the field).

12
Definition 1.3.9. Let (N, +, ·, F ) be a linear space where F = R or
C. Then a Norm on N is || · || : N → R+ satisfying the following
properties.
(i) ||x|| ≥ 0 and ||x|| = 0 ⇐⇒ x = 0N .
(ii) ||λx|| = |λ|||x||.
(iii) ||x + y|| ≤ ||x|| + ||y||, ∀ x, y ∈ N and λ ∈ F .

Definition 1.3.10. If (N, +, ·, F ) is a linear space over F = R or C

and if || · || : N → F is a norm on N , then (N, || · ||) is called Normed
linear space.

Definition 1.3.11. Let (X, dX ) be a metric space. A function f :

X → F where F = R or C is said to be bounded function if |f (x)| ≤ kf
where kf > 0 is a constant ∀ x ∈ X.

B ∗ (X) = {f : X → F | (X, dX ) is a metric space, F = R or C}

where f is a bounded scalar valued function on X.

Note 1.3.12. Every normed linear space is a metric space. Define

d(x, y) = ||x − y||, ∀ x, y ∈ N .
(i) d(x, y) ≥ 0 and d(x, y) = 0 ⇐⇒ ||x − y|| = 0 ⇐⇒ x − y = 0N ⇐⇒
x = y.

13
(ii)

d(x, y) = ||x − y||

= ||(−1)(y − x)||

= | − 1|||y − x||

= ||y − x|| = d(y, x)

(iii)

d(x, y) = ||x − y||

= ||(x − z) + (z − y)||

≤ ||x − z|| + ||z − y||

= d(x, z) + d(z, y)

∴ (N, d) is a metric space.

Theorem 1.3.13. Let (X, dX ) be a metric space and let F = R or C.

(i) (B ∗ (X), +, ·, F ) is a linear space,
where vector addition: (f + g)(x) = f (x) + g(x) and
scalar multiplication: (α · f )(x) = α · f (x).
(ii) Define ||f || = sup |f (x)|. Then (B ∗ (X), || · ||) is a normed linear
x∈X
space over F = R or C and hence (B ∗ (X), d) is a metric space where
d(f, g) = ||f − g||.
(iii) (B ∗ (X), d) is a Complete metric space.

14
Proof. (i) Consider

|(f + g)(x)| = |f (x) + g(x)|

≤ |f (x)| + |g(x)|

≤ kf + kg , (∵ f, g ∈ B ∗ (X)).

Define kf +g = kf + kg (sum of two positive constant is another con-

stant).

∴ |(f + g)(x)| ≤ kf +g , ∀ x ∈ X or f + g ∈ B ∗ (X).

Next, consider

|(λ · f )(x)| = |λ · f (x)|

= |λ||f (x)|

≤ |λ|kf .

Define kλf = |λ|kf (product of two positive constants is another con-

stant).
∴ |(λ · f )(x)| ≤ kλf , ∀ x ∈ X or λf ∈ B ∗ (X).

f (x), λf (x) ∈ F and F is vector space over itself ⇒ (N, +, ·, F )

is also a vector space.

(ii) Define ||f || = sup |f (x)|

x∈X

15
(a) |f (x)| ≥ 0 ⇒ ||f || ≥ 0

||f || = 0 ⇐⇒ 0 < |f (x)| ≤ ||f || = 0.

⇐⇒ |f (x)| = 0, ∀ x ∈ X

⇐⇒ f (x) = 0 (number), ∀ x ∈ X

⇐⇒ f = O (function)

(b)

||λf || = sup |(λf )(x)|

x∈λ
= sup[|λ||f (x)|]
x∈λ
= |λ| sup |f (x)|
x∈λ
= |λ|||f ||

||f + g|| = sup[|(f + g)(x)|]

x∈λ
= sup[f (x) + g(x)]
x∈λ
≤ sup[|f (x)| + |g(x)|]
x∈λ
≤ sup |f (x)| + sup |g(x)|
x∈λ x∈λ
≤ ||f || + ||g||.

∴ (B ∗ (X), || · ||) is a normed linear space.

(iii) Suppose {fn } is a Cauchy sequence in B ∗ (X).

⇒ For every ϵ > 0, ∃ a positive number N , such that m, n ≥ N ,

16
 ⇒ d(fm , fn ) < ϵ

⇒ ||fm − fn || < ϵ

⇒ sup |(fm − fn )(x)| < ϵ

x∈λ
⇒ sup |fm (x) − fn (x)| < ϵ
x∈λ
⇒ |(fm − fn )(x)| < ϵ

∴ {fn (x)} is a Cauchy sequence in F = R or C.

Since, R and C are Complete metric spaces, {fn (x)} is convergent.
Let lim fn (x) = f (x), ∀ x ∈ X.
n→∞

Consider,

|f (x)| = | lim fn (x)|

n→∞

= lim |fn (x)|, (∵ fn ∈ B ∗ (X))

n→∞

≤ lim kfn
n→∞

≤ sup kfn
n
= kf < ∞.

∴ f ∈ B ∗ (X).
Hence B ∗ (X) is a Complete metric space.

Lemma 1.3.14. Suppose (X, d) is a metric space and {xn } is a con-

vergent sequence with infinitely many distinct points. Then the limit
of {xn } is same as limit point of S = {x1 , x2 , . . . , xn , . . .}.

17
Proof. Suppose if possible x = lim xn is not a limit point of S.
n→∞

⇒ there exists r > 0, Sr (x) ∩ (S \ {x}) = ∅

⇒ xn = x for infinitely many n.

⇒ S = {x1 , x2 , . . . , xn , x} for some n.

Which is contradiction.

Theorem 1.3.15. Let (X, d) be a Complete metric space and (Y, d)

be a subspace of (X, d). Then Y is Complete ⇐⇒ Y is closed.
OR
Every closed subspace of a Complete metric space is also Complete.

Proof. Suppose Y is Complete and suppose x ∈ X is a limit point of

Y.

⇒ for every n, S1/n (x) ∩ Y ̸= ∅.

⇒ ∃ yn ∈ Y such that d(yn , x) < n1 .

1
Given ϵ > 0, choose N , a positive integer such that N < ϵ.
1 1
Then n ≥ N ⇒ d(yn , x) < n ≤ N < ϵ.

⇒ {yn } is a convergent sequence in Y .

⇒ {yn } is a Cauchy sequence in Y . [∵ d(ym , yn ) ≤ d(ym , x) +

d(x, yn )].

Since Y is Complete and {yn } is a Cauchy sequence in Y .

18
 lim yn = x ∈ Y.
n→∞

∴ Y is closed.
Suppose Y is closed and suppose {yn } is a Cauchy sequence in Y .

To show that: lim yn = x ∈ Y .

n→∞

Since (Y, d) is a subsequence of (X, d), {yn } is also a Cauchy

sequence in X. But (X, d) is a Complete metric space. Hence {yn } is
convergent sequence in X. Let lim yn = x.
n→∞

Case I: S = {y1 , y2 , . . . , yn , . . .} = {y1 , y2 , . . . , yn , x} for some n is

finite. Since every yi ∈ Y, x ∈ Y .

Case II: S = {y1 , y2 , . . . , yn , . . .} has infinitely many distinct points

and lim yn = x
n→∞

⇒ x is a limit point of S (by Lemma 1.3.14)

⇒ x is also a limit point of Y .

But Y is closed and hence it contains all its limit points.

∴ x ∈ Y.
∴ Every Cauchy sequence in Y is convergent and hence Y is complete.

Definition 1.3.16. Let (X, d) be a metric space and let F = R or C.

A function f : X → F is said to be continuous at x0 , if given ϵ > 0,
∃ δ > 0 such that d(x, x0 ) < δ ⇒ |f (x) − f (x0 )| < ϵ.

19
Definition 1.3.17. A function f : X → F is said to be continuous
on X, if it is continuous for all x ∈ X.

Note 1.3.18. Isometry between two metric space is obviously contin-

uous.

Theorem 1.3.19. Let C ∗ (X) = {f : X → F | (X, d) is a metric

space, F = R or C}
where f is a continuous and bounded on X. Then C ∗ (X) ⊂ B ∗ (X)
and further
(i) C ∗ (X) is a subspace of B ∗ (X).
(ii) C ∗ (X) is a closed subspace of B ∗ (X) as a consequence, C ∗ (X) is
a complete metric space.

Proof. (i) Suppose f, g ∈ C ∗ (X) and α, β ∈ F .

To show that: (a) f + g ∈ C ∗ (X)

(b) αf ∈ C ∗ (X)
⇐⇒ αf + βg ∈ C ∗ (X).

Consider (f + g)(x) and x0 ∈ X. Since f, g ∈ C ∗ (X), given ϵ > 0,

∃ δ1 > 0, δ2 > 0 such that

ϵ
d(x, x0 ) < δ1 ⇒ |f (x) − f (x0 )| <
2
ϵ
d(x, x0 ) < δ2 ⇒ |g(x) − g(x0 )| < .
2

20
 Choose δ = min(δ1 , δ2 ). Then

d(x, x0 ) < δ ⇒ |(f + g)(x) − (f + g)(x0 )|

= |{f (x) + g(x)} − {f (x0 ) + g(x0 )}|

= |{f (x) − f (x0 )} + {g(x) − g(x0 )}|

≤ |f (x) − f (x0 )| + |g(x) − g(x0 )|

ϵ ϵ
< +
2 2
= ϵ

∴ f + g is continuous on X.

|f (x)| ≤ kf , |g(x)| ≤ kg

⇒ |(f + g)(x)| = |f (x) + g(x)| ≤ |f (x)| + |g(x)|

≤ kf + kg

= kf +g

∴ f + g is bounded on X.

Hence f + g ∈ C ∗ (X).
ϵ
Next, consider αf . Since f is continuous at x0 , for |α| > 0.

∃ δ = δ(ϵ, |α|) such that d(x, x0 ) < δ

ϵ
⇒ |f (x) + f (x0 )| <
|α|
ϵ
⇒ |α||f (x) + f (x0 )| < |α| =ϵ
|α|
⇒ |(αf )(x) − (αf )(x0 )| < ϵ

∴ αf is continuous on X.

21
 |f (x)| ≤ kf , ⇒ |(αf )(x)| = |α||f (x)| ≤ |α|kf = kαf

∴ f is bounded on X.

Hence αf ∈ C ∗ (X).

As a result, C ∗ (X) is a subspace of B ∗ (X).

(ii) To show that: C ∗ (X) is closed.

Suppose f ∈ B ∗ (X) and f ∈ C ∗ (X).

Claim: f ∈ C ∗ (X) or f is continuous on X, or f is continuous at x0

(at any point x0 ∈ X).

f ∈ C ∗ (X) ⇒ S 3ϵ (f ) ∩ C ∗ (X) ̸= ∅
ϵ
⇒ ∃ f0 ∈ C ∗ (X) such that ||f − f0 || <
3
ϵ
⇒ sup |f (x) − f0 (x)| <
x∈X 3
ϵ
⇒ |f (x) − f0 (x)| < ∀ x ∈ X.
3
Since f0 ∈ C ∗ (X), f0 is continuous at x0 .
⇒ Given ϵ > 0, ∃ δ > 0 such that d(x, x0 ) < δ ⇒ |f0 (x) − f0 (x0 )| < 3ϵ ,

Consider,

d(x, x0 ) < δ ⇒ |f (x) − f (x0 )|

⇒ |{f (x) − f0 (x)} + {f0 (x) − f0 (x0)} + {f0 (x0 ) − f (x0 )}|

≤ |f (x) − f0 (x)| + |f0 (x) − f0 (x0 )| + |f0 (x0 ) − f (x0 )|

ϵ ϵ ϵ
< + + = 3.
3 3 3
22
∴ C ∗ (X) is a closed subspace of B ∗ (X).
Since B ∗ (X) is a complete metric space, by Theorem 1.3.15, C ∗ (X) is
also a complete metric space.

Theorem 1.3.20 (Metric Completion Theorem). Any metric space

(X, d) is isometric to a dense subspace of a complet metric space
(X ∗ , d∗ ).

Proof. Let (X, d), be a given metric space. Using Theorem 1.3.19,
(C ∗ (X), d∗ ) is a complete metric space over real field in which the
metric
d∗ (f, g) = sup |f (x) − g(x)| = ||f − g||.
x∈X

Let z ∈ X, be any element arbitrarily fixed.

Define a mapping f : X → C ∗ (X) by x → f (x) where f (x) : X → R,
given by f (x)(ξ) = d(x, ξ) − d(ξ, z).
To show that f (x) is continuous on X:
It is enough, if we show that it is continuous at any point ξ0 ∈ X.

23
 Given ϵ > 0, choose δ = 2ϵ .

d(ξ, ξ0 ) < δ ⇒ |f (x)(ξ) − f (x)(ξ0 )|

= |{d(x, ξ) − d(ξ, z)} − {d(x, ξ0 ) − d(ξ0 , z)}|

(∵ d(x, y) ≤ d(x, z) + d(z, y) ⇒ d(x, y) − d(x, z) ≤ d(z, y))

= |{d(x, ξ) − d(x, ξ0 )} − {d(ξ, z) − d(ξ0 , z)}|

≤ |d(x, ξ) − d(x, ξ0 )| + |d(ξ, z) − d(ξ0 , z)|

≤ d(ξ, ξ0 ) + d(ξ, ξ0 ) = 2d(ξ, ξ0 )

ϵ
≤ 2 × = ϵ.
2

∴ f (x) continuous on X.

To show that f (x) is bounded on X:

|f (x)(ξ)| = |d(x, ξ) − d(ξ, z)|

= |d(x, ξ) − d(z, ξ)|

≤ d(x, z) = kf (x) < ∞, ∀ ξ ∈ X.

∴ f (x) is bounded on X and hence f (x) ∈ C ∗ (X) is well defined.

To show that f (x) is an isometry:

Consider,

d∗ (f (x), f (y)) = sup |f (x)(ξ) − f (y)(ξ)|

ξ∈X
= sup |{d(x, ξ) − d(ξ, z)} − {d(y, ξ) − d(ξ, z)}|
ξ∈X
= sup |{d(x, ξ) − d(y, ξ)}|
ξ∈X
≤ d(x, y).

24
∴ f : X → f (X) is isometric, one to one and onto or X ∼
= f (X).
We know that f (X) is dense in f (X).
f (X) is a closed subspace of a complete metric space C ∗ (X). Using
Theorem 1.3.15, f (X) = X ∗ is complete.
∴ (X, d) is isometric to (f (X), d∗ ) which is dense in a complete metric
space (X ∗ , d∗ ).

Definition 1.3.21. Let (X, d) be a given metric space and let f : X →

C ∗ (X) be an isometry. Then X ∗ = f (X), being a closed subspace of a
complete metric space (C ∗ (X), d∗ ), (X ∗ , d∗ ) is a complete metric space
and it is called Completion of (X, d).

Theorem 1.3.22 (Metric completion is unique upto isome-

try:). All completion of a given metric space are isometric.

Proof. Suppose (X ∗ , d∗ ) and (X ∗∗ , d∗∗ ) are any two metric completions

of a given metric space (X, d). Then by Theorem 1.3.20, ∃ f, g : X →
C ∗ (X) two isometries such that X ∗ = f (X) and X ∗∗ = g(X).
Without loss of generality, X(∼
= f (X) ∼
= g(X)) itself can be taken to
be dense in both X ∗ and X ∗∗ .

Claim: ∃ F : X ∗ → X ∗∗ such that d∗∗ (F (x∗ ), F (y ∗ )) ≤ d∗ (x∗ , y ∗ ).

Let x∗ ∈ X ∗ . Since X is dense in X ∗ or X = X ∗ , x∗ is a limit point
of X. ∃ a sequence {xn } in X such that lim xn = x∗ . So, {xn } is a
n→∞
convergent sequence in X. Hence {xn } is a Cauchy sequence in X.

Then, X ⊆ X = X ∗∗ ⇒ {xn } is a Cauchy sequence in X ∗∗ and

25
X ∗∗ is Complete. So, ∃ x∗∗ ∈ X ∗∗ such that lim xn = x∗∗ .
n→∞
x∗ → x∗∗ will be well defined provided {xn } is unique.
To prove that {xn } is unique:
Suppose, if possible there is another sequence {x′n } in X converging
to x∗ .
⇒ d(xn , x∗ ) → 0, as n → ∞ and d(x′n , x∗ ) → 0, as n → ∞

⇒ 0 ≤ d(xn , x′n ) ≤ d(xn , x∗ ) + d(x∗ , x′n )

= d(xn , x∗ ) + d(x′n , x∗ ) → 0, as n → ∞

⇒ d(xn , x′n ) = 0

⇒ xn = x′n .

Hence, for every x∗ ∈ X ∗ , there is well defined element x∗∗ ∈ X ∗∗ .

Define F : X ∗ → X ∗∗ by F (x∗ ) = x∗∗ .
Suppose x∗ , y ∗ ∈ X ∗ . Then ∃ {xn } and {yn } in X such that lim xn =
n→∞
∗ ∗
x , lim yn = y .
n→∞
Since X = X ∗∗ , lim xn = x∗∗ , lim yn = y ∗∗ .
n→∞ n→∞

Consider,

d∗∗ (F (x∗ ), F (y ∗ )) = d∗∗ (x∗∗ , y ∗∗ )

= lim d(xn , yn )
n→∞

= d∗ (x∗ , y ∗ ).

Hence F is an isometry and X ∗ ∼

= X ∗∗ .

26
1.4 Exercises

(1) Prove that a set A in X is dense in X or A = X.

(i) ⇐⇒ the only closed superset of A is X.
(ii) ⇐⇒ the only open set disjoint from A is ∅.
(iii) ⇐⇒ A intersects every open sphere.

(2) Show that the metric space of complex numbers with usual metric
is isometric to R2 with Euclidean metric.

1.5 References

1. G. F. Simmons - Introduction to Topology and Modern Analysis,

Tata McGraw-Hill, New Delhi, 2004.

2. A. E. Taylor - Introduction to Functional Analysis, Willey, New

York, 1958.

−−−−−−−−−−−−−−−−−−−−−

27
Chapter 2

Unit 2: Banach’s Contraction

Mapping Theorem and
Applications

2.1 Introduction

In this unit, we introduce the concept of Banach Contraction Mapping

Theorem and its Applications.

2.2 Banach Contraction Mapping Theorem

Definition 2.2.1. Let (X, d) be a metric space and let T : X → X be a

mapping from X into X. Then T is said to be a contraction mapping,
if there exists α ∈ R1 such that 0 < α < 1 and d(Tx , Ty ) ≤ αd(x, y),
∀ x, y ∈ X.

28
Definition 2.2.2. Let (X, d) be a metric space and let T : X → X be
a mapping from X into X. Then a point x ∈ X is said to be a fixed
point of T , if Tx = x.

Note 2.2.3. Every Contraction mapping in continuous.

Given ϵ > 0, select δ = ϵ.
Then d(x, x0 ) < δ ⇒ d(Tx , Tx0 ) ≤ αd(x, x0 ) < d(x, x0 ) < δ = ϵ,
(∵ α < 1).
∴ By definition, T is continuous at x0 , and hence it is continuous on
X.

Theorem 2.2.4 (Banach Fixed Point Theorem or Banach Con-

traction Mapping Theorem). Let (X, d) be a complete metric space
and let T : X → X be a Contraction Mapping. Then T has a unique
fixed point.

Proof. : Let x0 ∈ X. Define sequence {xn } in X as follows:

x1 = Tx0

x2 = Tx1 = T (Tx0 ) = Tx20

..
.

xn = Txn−1 = Txn0 .

To show that {xn } is a Cauchy sequence:

29
Let us observe that,

d(xm+1 , xm ) = d(Txm , Txm−1 )

≤ αd(xm , xm−1 )

≤ αd(Txm−1 , Txm−2 )

≤ α2 d(xm−1 , xm−2 )
..
.

≤ αm d(x1 , x0 ), (∵ d(x1 , x0 ) = d(xm−(m−1) , xm−m )) (∗)

Consider,

d(xm , xn ) ≤ d(xm , xm−1 ) + d(xm−1 , xm−2 ) + · · · + d(xn+1 , xn ), using (∗)

= αn [1 + α + · · · + αm−n−1 ]d(x1 , x0 )

< αn [1 + α + · · · + αm−n−1 + · · · + ∞]d(x1 , x0 )

αn
= d(x1 , x0 ), 0 < α < 1,
1−α
→ 0 as m, n → ∞.

∴ {xn } is a Cauchy sequence in X.

Since (X, d) is a complete metric space, {xn } converges to x in X or
lim xn = x.
n→∞
Next, to show that x = lim xn is a fixed point for T (or to show that
n→∞
Tx = x or to show that d(Tx , x) = 0).

30
Consider,

d(Tx , x) ≤ d(Tx , xn ) + d(xn , x)

= d(Tx , Txn−1 ) + d(xn , x)

≤ αd(x, xn−1 ) + d(xn , x)

= αd(xn−1 , x) + d(xn , x), ∀ n.

Allow n → ∞, 0 ≤ d(Tx , x) ≤ αd(xn−1 , x) + d(xn , x) → 0

∴ d(Tx , x) = 0.
Finally, let us show that the fixed point is unique.
Suppose, if possible Tx1 = x1 and Tx2 = x2 .
⇒ 0 ≤ d(x1 , x2 ) = d(Tx1 , Tx2 ) ≤ αd(x1 , x2 ) < d(x1 , x2 )
⇒ d(x1 , x2 ) < d(x1 , x2 ) which is meaningless.
∴ x1 = x2 .

Theorem 2.2.5. Let T be a mapping of a complete metric space X

into itself. Let T be a contraction on a closed ball

B r (x0 ) = {x ∈ X : d(x, x0 ) ≤ r}.

Suppose that d(x0 , Tx0 ) < (1−α)r. Then the iterative sequence defined
by xn = Txn0 = Txn−1 converges to x ∈ B r (x0 ) and this x is the unique
fixed point of T .

Proof. Setting m = 0 in the proof of the Theorem 2.2.4, we have

d(x0 , x1 )
d(x0 , xn ) = .
1−α
31
It follows that d(x0 , xn ) < r, (∵ d(x0 , Tx0 ) < (1 − α)r).
This shows that all xn ’s are in B r (x0 ). Since xn → x and B r (x0 ) is
closed, it follows that x ∈ B r (x0 ). The assertion of the theorem now
follows from the proof of the Theorem 2.2.4.

2.3 Applications

We shall now give some application of Banach Contraction Mapping

Theorem.

Theorem 2.3.1 (Picard’s Theorem). Let f : D → R1 be a con-

tinuous function of a open domain D ⊆ R2 . Let (x0 , y0 ) ∈ D. If f
satisfies the Lipschitz condition:

|f (x, y1 ) − f (x, y2 )| ≤ M |y1 − y2 |

for all (x, y1 ) and (x, y2 ) are in D, then there exists α > 0 such that
the initial value problem.

dy
= f (x, y), y(x0 ) = y0 ,
dx

has a unique solution on [x0 − α, x0 + α].

Proof. First, we note that the initial value problem

dy
= f (x, y), y(x0 ) = y0
dx

32
 Z x Z x
dy
⇐⇒ dx = f (t, y(t))dt
x0 dx
Zx0x
or y(x) − y(x0 ) =
f (t, y(t))dt
x0
Z x
or y(x) = y0 + f (t, y(t))dt.
x
Rx 0
Define T : X → X by Ty(x) = y0 + x0 f (t, y(t))dt, where X is the
space of all solutions of the initial value problem.

Now, the problem, if and only if finding a unique y such that

Ty = y or T has a unique fixed point.
To apply the Banach Contraction Mapping Theorem, we need to show
the following:
(1) X is a complete metric space.
(2) T : X → X is a Contraction Mapping.
To clearly define the space X:
Given f (x, y) satisfies the Lipschitz condition:

|f (x, y1 ) − f (x, y2 )| ≤ M |y1 − y2 |.

Also f (x, y) is continuous at (x0 , y0 ).

Given ϵ = 1, ∃ δ > 0 such that

(x, y) ∈ Sδ ((x0 , y0 )) ⇒ |f (x, y) − f (x0 , y0 )| < 1.

⇒ |f (x, y)| ≤ 1 + |f (x0 , y0 )| = k.

Choose, α > 0 such that αM < 1.

Define a rectangle in R2 .

R = {(x, y) : x ∈ [x0 − α, x0 + α], y ∈ [y0 − αk, y0 + αk]}.

33
Then R can hold all the solutions passing through (x0 , y0 ).
Define, X = {ϕ : [x0 − α, x0 + α] → [y0 − αk, y0 + αk]}, where ϕ is
continuous and ϕ(x0 ) = y0 .
To show that X is complete:
Clearly, X ⊆ C ∗ ([x0 −α, x0 +α]) = {ϕ : [x0 −α, x0 +α] → R1 | ϕ is continuous}
C ∗ ([x0 −α, x0 +α]) is a complete metric space, ψ ∈ C ∗ ([x0 −α, x0 +α])
and ψ ∈ X
⇒ ∃ {ϕn } in X such that ψ(x) = lim ϕn (x), ∀ x ∈ [x0 − α, x0 + α].
n→∞
For all n, y0 − αk ≤ ϕn (x) ≤ y0 + αk.
⇒ y0 − αk ≤ ψ(x) ≤ y0 + αk.
ψ(x0 ) = lim ϕn (x0 ) = lim y0 = y0 .
n→∞ n→∞
∴ ψ ∈ X or X is a closed subspace of C ∗ ([x0 − α, x0 + α]).
Hence, using the standard theorem, X is a complete metric space.
Rx
To show that T : X → X, by Ty = y0 + x0 f (t, y(t))dt is a
contraction mapping.
First, let us show that if ϕ(x) ∈ [y0 − αk, y0 + αk],
then Tϕ(x) ∈ [y0 − αk, y0 + αk].
Z x
|Tϕ(x) − y0 | = f (t, ϕ(t))dt
x0
Z x
≤ |f (t, ϕ(t))|dt
Zx0x
< kdt
x0
= k(x − x0 ) ≤ kα, (∵ x ∈ [x0 − α, x0 + α] ⇒ (x − x0 ) < α).

⇒ −kα ≤ Tϕ(x) − y0 ≤ kα, or Tϕ(x) ∈ [y0 − αk, y0 + αk].

34
Next, T is continuous:
 Z x1   Z x2 
|Tϕ(x1 ) − Tϕ(x2 ) | = y0 + f (t, ϕ(t))dt − y0 + f (t, ϕ(t))dt
Z x2 x0 x0

= − f (t, ϕ(t))dt
x1
Z x2
≤ |f (t, ϕ(t))|dt
x1
≤ k|x1 − x2 |.

Given ϵ > 0, choose δ = kϵ , then continuity follows.

Rx
Finally, to show that T : X → X, by Ty = y0 + x0 f (t, y(t))dt is
a contraction mapping:
For every, ϕ1 , ϕ2 ∈ X, consider,

d∗ (Tϕ1 , Tϕ2 ) = sup |Tϕ1 (x) , Tϕ2 (x) |

x∈[x0 −α,x0 +α]=I
 Z x   Z x 
= sup y0 + f (t, ϕ1 (t))dt − y0 + f (t, ϕ2 (t))dt
x∈I x0 x0
Z x
= sup {f (t, ϕ1 (t)) − f (t, ϕ2 (t))}dt
x∈I x0
Z x
≤ sup |f (t, ϕ1 (t)) − f (t, ϕ2 (t))|dt
x∈I x0
Z x
≤ sup M |ϕ1 (t) − ϕ2 (t)|dt
x∈I x0
≤ sup M |ϕ1 (t) − ϕ2 (t)||x − x0 |
x∈I
∗
≤ d (ϕ1 (t), ϕ2 (t))M α

∴ d∗ (Tϕ1 , Tϕ2 ) ≤ d∗ (ϕ1 (t), ϕ2 (t)), (∵ M α < 1).

Let us apply Banach Contraction Mapping Theorem for
Z x
Tϕ(x) = y0 + f (t, ϕ(t))dt.
x0

35
There is a unique y(x) ∈ X such that Ty(x) = y(x) or
Z x
y(x) = y0 + f (t, y(t))dt,
x0

has unique solution or

dy
= f (x, y), y(x0 ) = y0
dx

has a unique solution.

2.4 Exercises

(1) Let T : X → X, where X is a metric space, satisfy d(Tx , Ty ) <

d(x, y) where x ̸= y. Suppose that T has a fixed point. Show that the
fixed point is unique.
(2) Let X be a metric space. If T is a contraction on X, then show
that T n (n ∈ N) is a contraction. If T n is a contraction on X for an
n > 1, then show that T need not be a contraction.
(3) A sufficient condition for convergence of an iteration xn = g(xn−1 )
is that g be continuously differentiable and |g ′ (x)| ≤ α < 1. Verify
this by using Banach Contraction Mapping Theorem.

2.5 References

1. G. F. Simmons - Introduction to Topology and Modern Analysis,

Tata McGraw-Hill, New Delhi, 2004.

36
2. A. E. Taylor - Introduction to Functional Analysis, Willey, New
York, 1958.

−−−−−−−−−−−−−−−−−−−−−

37
Chapter 3

Unit 3: Baire’s Category Theorem

3.1 Introduction

In this unit, we introduce the concept of Baire’s Category Theorem.

We shall first define some basic definitions.

Definition 3.1.1. Let (X, d) be a metric space. If Y ⊆ X, then

Y = {x ∈ X : either x ∈ Y or x limit point of Y }

\
= F, where F is a closed set in X.
F ⊇Y

Y is the smallest closed set in X containing Y .

Definition 3.1.2. Let (X, d) be a metric space. If Y ⊆ X, then

interior of Y is denoted by Int(Y ) and defined by

Int(Y ) = the largest open set contained in Y .

[
= G, where G is an open subset of X.
G⊆Y

38
Definition 3.1.3. Let (X, d) be a metric space. A subset Y ⊆ X is
said to be everywhere dense, if Y = X.
OR Int(Y ) = X.

Definition 3.1.4. Let (X, d) be a metric space. A subset Y ⊆ X is

said to be nowhere dense, if Int(Y ) = ∅ (where Y need not be empty).

Example 3.1.5. (1) Y = {y} ⇒ Y = {y} =

̸ ∅.
Int(Y ) = ∅.
∴ Every singleton sets are nowhere dense.
⇒ Every finite set is also nowhere dense.
⇒ Every countably infinite set is also nowhere dense.
Z is nowhere dense in R1 .

(2) X = R1 , Y = Q,
Q = R1 , Q is everywhere dense in R1 .

Definition 3.1.6. Let (X, d) be a metric space. A subset Y ⊆ X is

∞
S
said to be of first category, if Y = Yn , where Yn is nowhere dense
n=1
in X.
OR Int(Yn ) = ∅, n = 1, 2, . . ..

Definition 3.1.7. Let (X, d) be a metric space. A subset Y ⊆ X is

said to be of second category, if it is not a set of first category.
∞
S
OR If Y = Yn , then for at least one n, Int(Yn ) ̸= ∅.
n=1

39
Theorem 3.1.8 (Baire’s Category Theorem). Every nonempty
complete metric space is of second category.

Proof. we need following Lemma.

Lemma: Suppose {Gn : n = 1, 2, . . .} is a sequence of everywhere

∞
T
dense subsets of a complete metric space (X, d). Then Gn is also
n=1
everywhere dense in X.

Proof of the Lemma: We know that Y is everywhere dense in X.

⇐⇒ Y intersects every open sphere in X.
i.e. x ∈ X and r > 0 ⇒ Y ∩ Sr (x) ̸= ∅.
∞
T
It is enough, we show that A = Gn intersects any open sphere in
n=1
X.
Let x0 ∈ X and r0 > 0 be arbitrary. Then it is enough to show that
A ∩ Sr0 (x0 ) ̸= ∅.
Given, G1 is everywhere dense in X.
⇒ Sr0 (x0 ) ∩ G1 ̸= ∅, we can find x1 ∈ Sr0 (x0 ) ∩ G1 , choose, 0 < r1 < 1.
Then Sr1 (x1 ) ⊆ Sr0 (x0 ) ∩ G1 and Sr0 (x0 ) ⊇ Sr1 (x1 ).
Next, G2 is also given to be everywhere dense in X.
⇒ Sr1 (x1 ) ∩ G2 ̸= ∅. So, we can find x2 ∈ Sr1 (x1 ) ∩ G2 , choose, 0 <
r2 < 21 . Then Sr2 (x2 ) ⊆ Sr1 (x1 ) ∩ G2 and Sr0 (x0 ) ⊇ Sr1 (x1 ) ⊇ Sr2 (x2 ).

Continuing this process for every n, we can find

1
xn ∈ Srn−1 (xn−1 ) ∩ Gn and we may choose 0 < rn < n

So, {xn } is a sequence in X and

40
Sr0 (x0 ) ⊇ Sr1 (x1 ) ⊇ Sr2 (x2 ) ⊇ · · · ⊇ Srn (xn ) ⊇ · · · .

Claim: {xn } is a Cauchy sequence in X.

1
xm ∈ Srn (xn ) ⇒ d(xm , xn ) < rn < n
1
xn ∈ Srm (xm ) ⇒ d(xm , xn ) < rm < m
1
⇒ d(xm , xn ) < min(m,n) → 0 as m, n → ∞.
∴ {xn } is a Cauchy sequence in X.
Since, X is complete. We have lim xn = x ∈ X.
n→∞
⇒ x ∈ Srn (xn ), ∀ n ≥ n0 .
⇒ x ∈ Srn (xn ), ∀ n
(∵ Sr0 (x0 ) ⊇ Sr1 (x1 ) ⊇ Sr2 (x2 ) ⊇ · · · ⊇ Srn (xn ) ⊇ · · · )

rn (xn ) ⊆Gn , x ∈ Gn , ∀ n
since, S
∞
T
⇒x∈ Gn = A and x ∈ Sr0 (x0 )
n=1
⇒ Sr0 (x0 ) ∩ A ̸= ∅.
∞
T
∴A= Gn everywhere dense in X.
n=1

Proof of the Theorem 3.1.8:

Suppose {Yn : n = 1, 2, . . .} is a sequence of nowhere dense sub-

sets of X.
∞
S
Claim: X ̸= Yn or ∃ x ∈ X such that x ∈
/ Yn , ∀ n.
n=1

41
 S
∅ = Int(Yn ) = Gn ( where Gn is an open subset of Yn ),
Gn ⊆Yn

⇒ X = ϕ′ = [Int(Yn )]′
 ′
[
=  Gn 
G ⊆Yn
n
\
= G′n (Gn is open ⇍ G′n is closed)
′
Yn ⊇G′n
′
= Yn
′
{Yn : n = 1, 2, . . .} is a sequence of everywhere dense subsets of X.
∞
T ′
By Lemma, Yn is also everywhere dense.
n=1
∞ 
T ′
⇒ Sr0 (x0 ) ∩ Yn ̸= ∅
n=1
∞
T ′
⇒ x ∈ X such that x ∈ Yn
n=1
′
⇒ x ∈ Yn , ∀ n.
′
⇒x∈
/ Yn , f orall n = 1, 2, . . .
⇒x∈
/ Yn , f orall n = 1, 2, . . ., (∵ Yn ⊆ Yn )
S∞
∴X∈
/ Yn .
n=1
Hence X is of second category.

Theorem 3.1.9 (Cantor’s Intersection Theorem). Let (X, d) be

a complete metric space. Let {In } be a sequence of nonempty closed
subsets of X such that
(1) F1 ⊇ F2 ⊇ F3 ⊇ · · · ⊇ Fn ⊇ · · ·
(2) d(Fn ) → 0 as n → ∞.
∞
T
Then Fn contains exactly one element.
n=1

42
 ∞
T
Proof. First, we shall show that Fn cannot have more than one
n=1
element.
Given that d(Fn ) → 0 as n → ∞,
⇒ given ϵ > 0, ∃ a positive integer N such that n ≥ N ⇒ d(Fn ) < ϵ.
⇒ As ϵ → 0, d(Fn ) = 0 for n = N, N + 1, . . .
∴ Fn cannot have more than one element for n = N, N + 1, . . .
∞
T
∴ Fn cannot have more than one element.
n=1
∞
T
Next, we shall show that Fn ̸= ∅ or to show that ∃x ∈ X such that
n=1
x ∈ Fn , ∀ n.
Given that Fn ̸= ∅ ⇒ ∃ xn ∈ Fn , ∀ n = 1, 2, . . ..
It is also given that F1 ⊇ F2 ⊇ · · · ⊇ Fn ⊇ · · · or F1 ⊇ F2 ⊇ · · · ⊇
Fn ⊇ · · · ⊇ Fm ⊇ · · ·
⇒ xn ∈ Fn , m < n will give us xn ∈ Fm or xm ∈ Fm , m > n will give
us xm ∈ Fn .
⇒ either d(xm , xn ) < d(Fm ) or d(xm , xn ) < d(Fn )
⇒ d(xm , xn ) < min(d(Fm ), d(Fn )) → 0 as n → ∞ (∵ d(Fn ) → 0 as
n → ∞)
So, {xn } is a Cauchy sequence in X and X being Complete, lim xn =
n→∞
x ∈ X.

Case I: Suppose {xn } = {x1 , x2 , . . . , xN −1 , x, x, . . . , x} and

S = {x1 , x2 , . . . , xN −1 , x} is finite.
⇒ x ∈ Fn , ∀ n = N, N + 1, . . ..

43
Since, F1 ⊇ F2 ⊇ · · · ⊇ FN −1 ⊇ FN , x ∈ Fn , n = 1, 2, . . . , N − 1
∞
T ∞
T
∴ x ∈ Fn , ∀ n or x ∈ Fn or Fn ̸= ∅.
n=1 n=1

Case II: Suppose S = {x1 , x2 , . . . , xn , . . .} is infinite.

Since {xn } converges to x, x is a limit point of S.
⇒ x is also a limit point of {xn , xn+1 , . . .} ⊆ Fn , n = 1, 2, . . ..
⇒ x is also a limit point of Fn and Fn is a closed set.
∞
T ∞
T
⇒ x ∈ Fn , for n = 1, 2, . . .. ∴ x ∈ Fn or Fn ̸= ∅
n=1 n=1
∞
T
Hence Fn has exactly one point.
n=1

Problem 3.1.10. Use of Cantor’s Intersection Theorem to prove Baire’s

Category Theorem.

Proof. If {An : n = 1, 2, . . .} is a sequence of nowhere dense subsets

of a Complete metric space (X, d). Then ∃ x ∈ X such that x ∈
/ An ,
∀ n = 1, 2, . . ..
Given that A1 is nowhere dense subset of X.
⇒ Choose, x1 ∈ X and r = 1 so that S1 (x1 ) ∩ A1 = ∅.
Put F1 = S1/2 (x1 ) so that F1 ∩ A1 = ∅.
Next, it is also given that A2 is a nowhere dense subset of X.
⇒ Choose x2 ∈ S1/2 (x1 ) and r = 1/2, so that S1/2 (x2 ) ∩ A2 = ∅.
Put F2 = S1/4 (x2 ) so that F1 ⊇ F2 and F2 ∩ A2 = ∅.
We may continue this process for every n = 1, 2, . . ., so that we may
follow the same procedure:
An is a nowhere dense subset of X.

44
⇒ Choose, xn ∈ S1/2n−1 (xn−1 ) and rn = 1/2n−1 , so that S1/2n−1 (xn ) ∩
An = ∅.
Put Fn = S1/2n (xn ) so that F1 ⊇ F2 ⊇ · · · ⊇ Fn and Fn ∩ An = ∅.
Also, d(Fn ) = diameter of the closed sphere with radius 1/2n =
2/2n → 0 as n → ∞.
Hence by Cantor’s Intersection Theorem,
∞
T
Fn = {x} for some x ∈ X.
n=1
⇒ x ∈ Fn , ∀ n = 1, 2, . . .
But Fn ∩ An = ∅.
∴x∈
/ An , ∀ n = 1, 2, . . ..

3.2 Exercises

(1) Show that

∞
( )
X
l1 = x = (x1 , x2 , . . . , xn , . . .) | |xj | < 1
j=1

is dense, but of the first category.

(2) Prove that

 
l∞ = x = (x1 , x2 , . . . , xn , . . .) | ||x||∞ = sup |xj | < ∞
j

is dense, but of the first category in S, where S denotes the set of all
sequences of real or complex numbers.

45
3.3 References

1. G. F. Simmons - Introduction to Topology and Modern Analysis,

Tata McGraw-Hill, New Delhi, 2004.

2. A. E. Taylor - Introduction to Functional Analysis, Willey, New

York, 1958.

−−−−−−−−−−−−−−−−−−−−−

46
Chapter 4

Unit 4: Ascoli-Arzela Theorem

4.1 Introduction

In this unit, we introduce the concept of Ascoli-Arzela Theorem.

4.2 Ascoli-Arzela Theorem

In order to prove Ascoli-Arzela Theorem, we shall first define some

basic definitions.

Definition 4.2.1. Let (X, d) be a metric space. A family of open set

S
{Gi } is said to be an open cover of X, if X = Gi .
i

Definition 4.2.2. Let (X, d) be a metric space. If {Gin } is a subfamily

S
of {Gi } and X = Gin , then {Gin } is called subcover of X.
in

Definition 4.2.3. Let (X, d) be a metric space. (X, d) is said to be a

compact metric space, if every open cover of X has a finite subcover.

47
Definition 4.2.4. Let (X, d) be a metric space. (X, d) is said to be
a sequentially compact, if every sequence {xn } in X has a convergent
subsequence {xnk }.

Definition 4.2.5. Let (X, d) be a metric space. (X, d) is said to have

Balzano - Weierstrass property, if every infinite subset A ⊆ X has a
limit point.

Theorem 4.2.6. Let (X, d) be a metric space. Then X is sequentially

compact if and only if X has Balzano - Weierstrass property.

Proof. Suppose X is sequentially compact and suppose A is any infi-

nite subset of X.
⇒ ∃ a sequence {xn } in A with distinct points.
⇒ {xn } is a sequence in X and X is sequentially compact.
⇒ ∃ a convergent subsequence {xnk } in A ⊆ X.
⇒ {xnk } is a convergent sequence with infinitely many distinct points
in A.
⇒ x = lim xnk is a limit point of A.
nk →∞
∴ A has Balzano - Weierstrass property.

Suppose X has Balzano - Weierstrass property and suppose {xn }

is any sequence in X.

Case I: Suppose {xn } = {x1 , x2 , . . . , xN −1 , x, xN +1 , . . . , xN +k , x, . . .}.

Then, xnk = x or {x, x, . . .} is a convergent subsequence.

Case II: Suppose {xn } has distinct elements.

48
⇒ S = {x1 , x2 , . . . , xn , . . .} will be an infinite set in X.
Since, X has Balzano-Weierstrass property, A has a limit point x ∈ X.
⇒ For every k > 0, ∃ xnk ∈ S such that xnk ∈ Srnk (x) ∩ S(̸= ∅).
⇒ {xnk } is a convergent subsequence of {xn }.
∴ X is also sequentially compact.

Theorem 4.2.7. Every compact metric space has Balzano-Weierstrass

property. Hence it is also sequentially compact.

Proof. Suppose X is a compact metric space and suppose A is any

infinite subset of X. Suppose if possible A has no limit point in X.
S = {a1 , a2 , . . . , an , . . .}, Sri (ai ) ∩ A = {ai }
Set Gi = Sri (ai ), i = 1, 2, . . ..
Then {Gi } is a collection of open subsets of X. We may make {Gi }
an open cover of X selecting ri suitably,
[
X= Gi .
i

Since X is compact, X = Gi1 ∪ Gi2 ∪ · · · Gin .

Since A is nothing but set of centers of these sphere, A = {ai1 , ai2 , . . . , ain }
is a finite subset of X which is a contradiction.

Definition 4.2.8. Let (X, d) be a metric space. Then X is said to be

totally bounded, if for every ϵ > 0, ∃ A = {a1 , a2 , . . . , aN } ⊆ X such
that
N
[
X= Sϵ (ak ).
k=1

49
Definition 4.2.9. Let Y be a subset of a metric space (X, d). Then Y
is said to be totally bounded, if for every ϵ > 0, ∃ A = {a1 , a2 , . . . , aN } ⊂
X such that
N
[
Y ⊆ Sϵ (ak ).
k=1

Note 4.2.10. (1) A is also called as ϵ-net of X.

(2) x ∈ X ⇒ ∃ at least one ak ∈ A such that d(x, ak ) < ϵ.

Observation 4.2.11. (1) Suppose Y1 ⊆ Y2 ⊆ X and (X, d) is a metric

space. Then Y2 is totally bounded ⇒ Y1 is also totally bounded
N
S
(∵ Y1 ⊆ Y2 ⊆ Sϵ (ak )).
k=1
(2) A subset Y of a metric space (X, d) is totally bounded if and only
if Y is totally bounded.
(3) A subset Y in a metric space (X, d) is totally bounded ⇒ Y is also
bounded.
(4) A bounded set in a metric space need not be totally bounded.
(5) In Rn , every bounded set is also totally bounded.

Theorem 4.2.12. Every sequentially compact metric space is both

complete and totally bounded.

Proof. First, we shall show that a sequentially compact metric space

(X, d) is complete.

Let {xn } be any Cauchy sequence in X. By definition, d(xm , xn ) →

0 as m, n → ∞. It is given that X is sequentially compact. By defi-
nition, every sequence in X has a convergent subsequence.

50
 So, for {xn } there is a convergent subsequence {xnk } in X with
lim xnk = x.
nk →∞

0 ≤ d(xn , x) ≤ d(xn , xnk ) + d(xnk , x) → 0 as n, nk → ∞.

∴ {xn } converges to x in X.
Hence X is complete.

Next, we shall show that X is also totally bounded.

Suppose if possible, X is not totally bounded.
∃ ϵ > 0, such that X has no ϵ-net or X has no covering by finitely
many open spheres with radii ϵ. We shall construct a sequence {xn }
such that it has no convergent subsequence. This means X is not
sequentially compact which will be a contradiction.

Let x1 ∈ X be any point in X. Since, X ̸= Sϵ (x)

∃ x2 ∈ X \ Sϵ (x1 ) (otherwise, X = Sϵ (x1 ) ⇒ X is totally bounded
which is against to the contra assumption).
⇒ d(x1 , x2 ) ≥ ϵ.
Suppose, we have constructed x1 , x2 , . . . , xn ∈ X such that d(xi , xj ) ≥
ϵ, j = 1, 2, . . . , n.
n
S
Since, X is not totally bounded, X ̸= Sϵ (xk )
k=1
n
S
⇒ ∃ xn+1 ∈ X \ Sϵ (xk ) and hence d(xn+1 , xi ) ≥ ϵ, ∀ i = 1, 2, . . . , n.
k=1
Thus, inductively we have constructed a sequence {xn } in X such that
d(xm , xn ) ≥ ϵ.
⇒ {xn } can not have a Cauchy sequence, ({xn } is convergent ⇒ {xn }

51
is Cauchy sequence if and only if {xn } is not Cauchy sequence ⇒ {xn }
is not convergent).
⇒ {xn } can not have a convergent subsequence.

Theorem 4.2.13. Let (X, d) be a complete and totally bounded metric

space. Then X is compact.

Proof. Suppose if possible, X is not compact.

⇒ an open cover G of X which has no finite subcover.

Let us construct a sequence {Sn } = {S1/2n (xn )} = {S(xn ; 1/2n )}

of open spheres in X in which no open sphere has a finite subcover of
G.
To construct S1 :
It is given that X is totally bounded.
(1) (1) (1)
⇒ for ϵ = 1/2, ∃ A1 = {a1 , a2 , . . . , aN1 } in X such that
N1
(1)
[
X= S(ak ; 1/2).
k=1

(1)
Since X is not compact, there is at least one k such that S(ak ; 1/2)
has no finite subcover of G.
(1) (1)
Set S1 = S(ak ; 1/2) = S(x1 ; 1/2), where x1 = ak .

52
Suppose we have constructed,

S1 = S(x1 ; 1/2),

S2 = S(x2 ; 1/22 ),
..
.

Sn = S(xn ; 1/2n ),

such that no Si , i = 1, 2, . . . , n can be covered finitely by G.

To construct Sn+1 :
Again, since X is totally bounded, for ϵ = 1/2n+1 ,
NS
n+1
(n+1) (n+1) (n+1) (n+1)
∃ An+1 = {a1 , a2 , . . . , aNn+1 } such that X = S(ak ; 1/2n+1 ).
k=1
(n+1)
Again, since X is not compact, ∃ at least one k, such that S(ak ; 1/2n+1 )
has no finite subcover of G.
(n+1) (n+1)
Set Sn+1 = S(ak ; 1/2n+1 ) = S(xn+1 ; 1/2n+1 ), where xn+1 = ak .

Claim: {xn } is a Cauchy sequence.

Let y ∈ Sn ∩ Sn+1 . Consider

d(xn , xn+1 ) ≤ d(xn , y) + d(y, xn+1 )

1 1
< n + n+1
2 2
1
= n+1 [2 + 1]
2
3 22
= n+1 < n+1 , ∀ n.
2 2

53
 Next,

d(xm , xn ) ≤ d(xm , xm+1 ) + d(xm+1 , xm+2 ) + · · · + d(xn−1 , xn )

1 1 1
< m−1 + m + · · · + n−2
2  2 2 
1 1 1
= m−1 1 + + · · · + n−m−1
2 2 2
n+1
 
1 − r
∵ GP = 1 + r + r2 + · · · + rn =
1−r
n−m
1 (1 − 1/2 )
= m−1
2 (1 − 1/2)
 
1 1
= 2 m−1 1 − n−m
2 2
 
1 1
= m−2 1 − n−m
2 2
1
< m−2 → 0 as m → ∞ and n → ∞.
2

Since, X is complete and {xn } is Cauchy sequence in X.

lim xn = x, (where x ∈ X),

n→∞

since x ∈ X and G is an open cover of X, x ∈ G for some G ∈ G.

⇒ x ∈ S(x; ϵ) ⊆ G.
Consider, Sk for any k and z ∈ Sk .
By construction, Sk can not be covered finitely by G.
But,

d(x, z) ≤ d(x, xk ) + d(xk , z)

1
< ϵ + k , (allow k → ∞)
2
< ϵ, for large value of k or z ∈ S(x; ϵ)

54
∴ For large value of k, Sk ⊆ S(x; ϵ) ⊆ G ∈ G
Sk is covered by a single member of G.
This is a contradiction.

Theorem 4.2.14. Let (X, d) be a metric space. Then the following

are equivalent:
(1) X is compact.
(2) X is sequentially compact.
(3) X is complete and totally bounded.

Proof. (1) ⇒ (2).

X is compact ⇒ X has Balzano - Weierstrass property.
⇐⇒ X is sequentially compact (using Theorem 4.2.7).

(2) ⇒ (3).
X is sequentially compact ⇒ X is complete and totally bounded (using
Theorem 4.2.12).

(3) ⇒ (1).
X is complete and totally bounded ⇒ X is compact (using Theorem
4.2.13).

Corollary 4.2.15. Let (X, d) be a complete metric space. Then a

closed subset Y of X is compact if and only if Y is totally bounded.

Proof. Suppose X is complete and Y is a closed subset of X.

⇒ Y is also complete.

55
By hypothesis, Y is totally bounded. Using Theorem 4.2.14, Y is
compact.
Suppose Y is compact ⇒ Y is totally bounded (again using Theorem
4.2.14).

Corollary 4.2.16. Let (X, d) be a complete metric space and let Y ⊆

X. Then Y is compact if and only if Y is totally bounded.

Proof. Suppose, Y is compact ⇒ Y is complete and totally bounded.

n
S
⇒ ϵ > 0, ∃ A = {a1 , a2 , . . . , an } ⊆ X such that Y ⊆ Sϵ (ak )
k=1
n
S
⇒Y ⊆Y ⊆ Sϵ (ak )
k=1
∴ Y is also totally bounded.
Suppose, Y is totally bounded ⇒ Y is also totally bounded (using (2)
of Observation 4.2.11) since Y is a closed subset of a complete space
X.
⇒ Y is also complete.
Since, Y is complete and totally bounded, Y is compact (using Theo-
rem 4.2.14).

Definition 4.2.17. Let (X, d) be a metric space. Let C(X) be the

complete metric space of all continuous, bounded, real or complex val-
ued functions on X with the metric is given by the supremum norm

||f || = sup |f (x)| and d(f, g) = ||f − g||.

x∈X

Then a family of functions F in C(X) is said to be equi-continuous,

56
if for every ϵ > 0, ∃ δ > 0 such that d(x, x′ ) < δ

⇒ |f (x) − f (x′ )| < ϵ, ∀ x, x′ ∈ X, f ∈ F.

Also, a family of functions F in C(X) is said to be uniformly bounded,

if ∃ M > 0 such that |f (x)| ≤ M , ∀ x ∈ X and f ∈ F.

Note 4.2.18. (1) f ∈ F and F is equi-continuous family

⇒ f is uniformly continuous on X.
(2) If K is the set of all uniformly bounded functions, then K is
a bounded set in C(X). The uniformly boundedness is nothing but
boundedness in C(X).

Theorem 4.2.19 (Ascoli-Arzela Theorem). Let (X, d) be a com-

pact metric space. Let C(X) be the complete metric space of all contin-
uous, bounded, real or complex valued functions on X with the metric
is given by supremum norm. Then a closed subset K of C(X) is com-
pact ⇐⇒ K is equi-continuous and uniformly bounded.

Proof. Let us apply the Theorem 4.2.14 to K. Then K is compact if

and only if K is complete and totally bounded. But C(X) is complete
and K is a closed subset of C(X). So, K is complete. As a result, we
obtain K is compact if and only if K is totally bounded.

Suppose K is compact. First we shall show that K is equi-

continuous.
K is compact ⇒ K is totally bounded.

57
⇒ K has an 3ϵ -net, A = {f1 , f2 , . . . , fN }
⇒ for every f ∈ K, |f (x) − fi (x)| < 3ϵ , i = 1, 2, . . . , N .
Since, fi ’s continuous and K is compact, fi is also uniformly continu-
ous on K. By definition of uniformly continuity, given ϵ > 0, ∃ δi > 0
such that d(x, x′ ) < δi

ϵ
⇒ |fi (x) − fi (x′ )| < , ∀ i = 1, 2, . . . , N.
3

Put δ = min1≤i≤N δi . Consider, d(x, x′ ) < δ

⇒ |f (x) − f (x′ )|

= |f (x) − fi (x) + fi (x) − fi (x′ ) + fi (x′ ) − f (x′ )|

≤ |f (x) − fi (x)| + |fi (x) − fi (x′ )| + |fi (x′ ) − f (x′ )|

= |f (x) − fi (x)| + |fi (x) − fi (x′ )| + |f (x′ ) − fi (x′ )|

ϵ ϵ ϵ
< + + = ϵ, ∀ x, x′ ∈ X, f ∈ K.
3 3 3

∴ K is equi-continuous.
Next, we shall show that K is uniformly bounded. Here, we shall make
use of definition of compactness. Let f0 ∈ K. Then G = {S(f0 ; n) :
n = 1, 2, . . .} is an open cover of K. Since K is compact, G has a finite
subcover. Let {S(f0 ; ni ) : i = 1, 2, . . . , N } be a finite subcover of K.
Put M = min ni . Then K ⊆ S(f0 , M ) or d(K) ≤ M
1≤i≤N
⇒ sup |f (x) − g(x)| ≤ M , ∀ f, g ∈ K.
x∈X
Take g = 0. Then |f (x)| ≤ M , ∀ x ∈ X and f ∈ K
∴ K is uniformly bounded.

58
 Suppose K is equi-continuous and uniformly bounded.
We know that C(X) is complete metric space and K is a closed subset
of C(X), we shall show that K is totally bounded.
K is equi-continuous ⇒ given ϵ > 0, ∃ δ > 0 such that |f (x)−f (x′ 0)| <
ϵ
4 whenever d(x, x′ ) < δ, ∀ x, x′ ∈ X, f ∈ K.
K is uniformly bounded ⇒ ∃ M > 0 such that |f (x)| ≤ M , ∀ x ∈
X, f ∈ K.
1
Select, a large integer m > 0 such that m < 4ϵ .
Divide the interval [−M, M ] into k = 2mM equal parts.
Then, we can partition [−M, M ] as follows:

−M = y0 < y1 < y2 < · · · < yk = M

So that
 
1 ϵ M − (−M ) 1 ϵ
|yi − yi+1 | = < , ∵ = < ,
m 4 2mM m 4

we can form n-tuples (y1j , y2j , . . . , ynj ) using {y0 , y1 , . . . , yk }. Since

X is compact, there is a δ-net for X, {x1 , x2 , . . . , xn }, i.e., X =
Sn
S(xk , δ).
k=1
For every such n-tuples, ∃ f ∈ K such that |f (xi ) − yij| < 4ϵ .
Set Ei = {e ∈ K : |e(xi ) − yij | < 4ϵ , i, j, = 1, 2, . . . , n}.
Choose, ei ∈ Ei , i = 1, 2, . . . , n. Set E = {e1 , e2 , . . . , en }.

59
Claim: E is an ϵ-net.

|f (x) − ei (x)| = |f (x) − f (xi ) + f (xi ) − yij + yij − ei (xi ) + ei (xi ) − ei (x)|

≤ |f (x) − f (xi )| + |f (xi ) − yij | + |e(xi ) − yij | + |ei (x) − ei (xi )|

ϵ ϵ ϵ ϵ
< + + + =ϵ
4 4 4 4
n
S
∴f ∈K⇒f ∈ S(ei , ϵ).
i=1
n
S
∴K⊆ S(ei , ϵ) or K is totally bounded. Hence K is compact.
i=1

4.3 Exercises

(1) A subset Y of a metric space (X, d) is totally bounded if and only

if Y is totally bounded.
(2) Prove that a subset Y in a metric space (X, d) is totally bounded
⇒ Y is also bounded.
(3) Prove that a bounded set in a metric space need not be totally
bounded.
(4) Show that in Rn , every bounded set is also totally bounded.

4.4 References

1. G. F. Simmons - Introduction to Topology and Modern Analysis,

Tata McGraw-Hill, New Delhi, 2004.

60
2. A. E. Taylor - Introduction to Functional Analysis, Willey, New
York, 1958.

−−−−−−−−−−−−−−−−−−−−−

61
Block - II

62
Chapter 5

Unit 5: Linear Spaces and Linear

Operators

5.1 Introduction

In this unit, we introduce the concept of linear spaces. We give the

properties of linear spaces and some theorems related to linear spaces.
Finally, we explain the linear operators and some theorems related to
linear operators.

5.2 Linear Spaces

Linear space or vector space is the most fundamental and important

structure in the functional analysis. This structure is built making use
of a nonempty set of vectors X, a field of scalars F , which is either
field of real numbers or field of complex numbers and two algebraic

63
operations in the form of linear expressions namely, vector addition
and scalar multiplication. Starting from any vectors x, y, ∈ X, the
vector addition enables to find the third vector z = x+y ∈ X. Starting
from a scalar α and a vector x, the scalar multiplication enables to
find the vectors αx ∈ X.

Definition 5.2.1. The Linear space X, written as (X, +,˙, F ) satsfies

the following properties.
(1) (X, +) is an abelian group.
(2) (a) α ∈ F, x, y ∈ X =⇒ α(x+y) = αx +αy.
(b) α,β ∈ F , x ∈ X =⇒ (α + β)x = αx + βx.
(c) α, β ∈ F, x ∈ X =⇒ α(βx) = (αβ)x.
(d) x ∈ X =⇒ 1.x = x.
The elements of X are called vectors and X is called Linear space.

Example 5.2.2. (1) The set R of real numbers, with ordinary addi-
tion and multiplication taken as the linear operations, is a real linear
space.
(2) The set Rn of all n-tuples of real numbers is a real linear space
under the following co-ordinatewise linear operations: if
x = (x1 , x2 , ..., xn ) and y = (y1 , y2 , ..., yn ),
then x + y = (x1 + y1 , x2 + y2 , ..., xn + yn ) and αx = (αx1 , αx2 , ..., αxn ).
(3) The set C n of all complex numbers is a complex linear space under
ordinary addition and multiplication.
(4) The set C n of all n-tuples of complex numbers is a complex linear

64
space with respect to the co-ordinatewise linear operations defined in
Example 2.
(5) Let P be set of all polynomials, with real coefficients, defined on
the closed unit interval [0, 1]. We specifically include all non-zero con-
stant polynomials (which have degree 0) and the polynomial which is
identically zero (this has no degree at all). If the linear operations are
taken to be the usual addition of two polynomials and the multiplica-
tion of a polynomial by a real number, then P is a real linear space.
(6) For a given positive integer n, let Pn be subset of P consisting of
the polynomial which is identically zero and all polynomials of degree
less than n. Pn is a real linear space with respect to the linear opera-
tions defined in P.
(7) A linear space may consist solely of the vector 0, with scalar mul-
tiplication defined by α.0 = 0 for all α. We refer to this as the zero
space, and we always denote it by {0}.

Definition 5.2.3. A nonempty subset Y ⊂ X is said to be a subspace

of X = (X, +, ., F ), if (Y, +, ., F ) is also a linear space.
⇐⇒ (i) y1 , y2 ∈ Y =⇒ y1 + y2 ∈ Y
(ii) α ∈ F, y ∈ Y =⇒ αy ∈ Y
⇐⇒ α, β ∈ F, y1 , y2 ∈ Y =⇒ αy1 + βy2 ∈ Y . As a consequence, if α
= 1, β = −1, y1 = y2 , then αy1 + βy2 = y1 − y2 = 0 ∈ Y.

Note 5.2.4. The necessary condition for Y to be a subspace of X is

0∈Y.

65
Definition 5.2.5. Let X be a linear space and Y ⊂ X be a subspace
of X. Then the quotient space of X with respect to Y is given by
X/Y = {x + Y : x ∈ X}.
Equality of two numbers : x1 + Y = x2 + Y ⇐⇒ x1 − x2 ∈ Y .
Vector addition: (x1 + Y ) + (x2 + Y ) = (x1 + x2 ) + Y .
Scalar multiplication: α(x + Y ) = αx + Y .
Zero element: (x + Y ) + (z + Y ) = (x + Y )
⇐⇒ (x + z) + Y = x + Y
⇐⇒ (x + z) − x ∈ Y
⇐⇒ z ∈ Y .
⇐⇒ z + Y = 0 + Y = Y (zero element).

Theorem 5.2.6. Let M be a subspace of a linear space L, and let the

coset of an element x in L be defined by x + M = {x + m : m ∈ M }.
Then the distinct cosets form a partition of L; and if addition and
scalar multiplication are defined by
(x + M ) + (y + M ) = (x + y) + M and α(x + M ) = αx + M ,
then these cosets constitute a linear space denoted by L/M and called
the quotient space of L with respect to M .

Proof. The proof of this theorem is routine.

Definition 5.2.7. If x1 , x2 , ..., n is a finite nonempty set of vectors in

linear space X, then the vectors
x = α1 x 1 + α2 x 2 + · · · + αn x n

66
is called a linear combination of x1 , x2 , . . . , xn .

Definition 5.2.8. If S is an arbitrary nonempty subset of linear space

X, then the set of all linear combinations of vectors in S is clearly
a subspace of X; we denote this subspace by [S], and we call it the
subspace spanned by S.

Definition 5.2.9. If M is a subspace of linear space X, then a nonempty

subset S of M is said to span M if [S] = M .

Definition 5.2.10. A subset M of linear space X, is called linearly

independent if for every nonempty finite subset {x1 , x2 , . . . , xn } of M ,
the relation

n
X
λi xi = 0 =⇒ λi = 0 for i = 1, 2, 3, . . . , n.
i=1
A subset M of linear space X which is not linearly independent is
called linearly dependent.

Definition 5.2.11. A linearly independent subset of linear space X,

which is maximal (with respect to set inclusion) is called a basis of X.

Suppose now that M and N are subspaces of linear space X, and

consider the set M + N of all sums of the form x + y, where x ∈ M
and y ∈ N . Since M and N are subspaces, it is easy to see that
M + N is the subspace spanned by all vectors in M and N together,
i.e., M + N = [M ∪ N ]. If it happens that M + N = X, then we say

67
that X is the sum of the subspaces M and N . This means that each
vector in X expressible as the sum of a vector in M and a vector in
N . The case in which even more is true namely, that each vector z
in X is expressible uniquely in the form z = x + y, with x ∈ M and
y ∈ N will be of particular importance for us. In this case we say that
X is the direct sum of the subspaces M and N , and we symbolize this
statement by writing X = M ⊕ N .

Theorem 5.2.12. Let a linear space X be the sum of two subspaces

M and N , so that X = M + N . Then X = M ⊕ N ⇔ M ∩ N = {0}.

Proof. We begin by assuming that X = M ⊕ N and we deduce a

contradiction from the further assumption that there is a nonzero
vector z in M ∩ N . It suffices to observe that z is expressible in
two different ways as the sum of a vector x in M and a vector y in N ,
for z = z + 0 (here x = z and y = 0) and z = 0 + z (here x = 0 and
y = z). This contradicts the uniqueness required by the assumption
that X = M ⊕ N .

We now assume that M ∩ N = {0} and we show that it follows

from X = M + N can be strengthened to X = M ⊕ N . Since X =
M + N , each z in X can be written in the form z = x + y with x ∈ M
and y ∈ N . We wish to show that this decomposition is unique. If we
have two such decompositions of z, so that z = x1 + y1 = x2 + y2 , then
x1 − x2 = y2 − y1 . The left side of this is in M , the right side is in

68
N and they are equal; it therefore follows from M ∩ N = 0 that both
sides are 0, that x1 = x2 and y1 = y2 and that the decomposition of z
is unique.

Theorem 5.2.13. If S is a linearly independent set of vectors in a

linear space X, then there exists a basis B for X such that S ⊂ B.

Proof. Consider the class P of all linearly independent subsets of X

which contain X. P is clearly a partially ordered set with respect to
set inclusion. It suffices to show that P contains a maximal set B,
for such a maximal set will automatically be a basis for X such that
S ⊆ B. By Zorn’s lemma ( i.e., If P is a partially ordered set in which
every chain has an upper bound, then P has a maximal element),
it suffices to show that every chain in P has an upper bound in P.
But this is evident from the fact that the union of all the sets in any
chain of linearly independent sets which contain S is itself a linearly
independent set which contains S.

5.3 Linear operators

In this section, we define the linear operators and a theorem related

to linear operators.

Definition 5.3.1. Let N and N ′ be two normed linear spaces over the
same field F (= R or C). A mapping T : N −→ N ′ is said to be a

69
linear operator from N into N ′ , if
T (αx + βy) = αT (x) + βT (y), ∀ x, y ∈ N and α, β ∈ F .

Definition 5.3.2. A linear operator T : N −→ N ′ is said to be con-

tinuous on N , if T (xn ) −→ T (x) as xn −→ x, ∀ x ∈ N .

Definition 5.3.3. A linear operator T : N −→ N ′ is said to be

bounded linear operator, if
||T (x)|| ≤ α||x|| for some α > 0, and ∀ x ∈ N .

Theorem 5.3.4. Let T : N −→ N ′ be a linear operator from a normed

linear space N into a normed linear space N ′ . Then the following are
equivalent:
(1) T is a continuous linear operator.
(2) T is continuous at x = 0.
(3) T is a bounded linear operator.
(4) If S = {x ∈ N : ||x|| ≤ 1} is the closed unit sphere in N , then
T (S) is bounded in N ′ .

Proof. (1) =⇒ (2):

T is a continuous linear operator
=⇒ T (xn ) → T (x) as xn → x
=⇒ T (xn ) → T (0) as xn → 0
=⇒ T (xn ) → 0 as xn → 0, (∵ T (0) = T (0v) = 0 · T (v) = 0)
=⇒ T is continuous at x = 0.
(2) =⇒ (1)

70
Given: T (xn ) → 0 as xn → 0
=⇒ T (xn − x) → 0 as xn → x → 0 or xn → x
=⇒ (T (xn ) − T (x)) → 0 as xn → x
=⇒ T (xn ) → T (x) as xn → x, ∀ x ∈ N .
=⇒ T is a continuous linear operator.
(2) =⇒ (3) or not(3) =⇒ not(2)
Suppose (3) is false =⇒ for every n > 0, ∃ xn ∈ N such that
||T (xn )|| > n||x|| .
xn xn 1
Set yn = n||xn || . Then ||yn || = || n||x n ||
|| = n||xn || ||xn || = n1 .
1
As n → ∞, yn →  0 (∵||yn || =⇒ n → 0 as n → ∞).
xn
But, T (yn ) = T n||x n ||
T (xn )
= n||x n ||
=⇒ ||T (yn )|| = ||Tn||x
(xn )||
n ||
> 1, ∀ n.
∴ ||T (yn )|| → 0 as n → ∞.
Therefore (2) is also false.
(3) =⇒ (2)
(3) =⇒ T is a bounded linear operator
=⇒ ∃ α > 0, such that ||T (x)|| ≤ α||x||, ∀ x ∈ N
=⇒ as xn → 0 or ||xn || → 0.
This will lead to 0 ≤ ||T (xn )|| ≤ α||xn || → 0
=⇒ T (xn ) → 0 as xn → 0
=⇒ T is continuous at x = 0 =⇒ (2).
(3) =⇒ (4)
(3) =⇒ T is a bounded linear operator.
=⇒ ∃ α > 0 such that ||T (x)|| ≤ α||x||, ∀ x ∈ N

71
=⇒ for s ∈ S = {x ∈ N : ||x|| ≤ 1}, ||T (s)|| ≤ α||s|| ≤ α
=⇒ ||T (s)|| ≤ α, ∀ s ∈ S
=⇒ d(T (s)) ≤ 2α < ∞, (d(A) = diameter of A = sup d(a, b))
a,b∈A
=⇒ T (s) is bounded in N ′ =⇒ (4).
(4) =⇒ (3)
(4) =⇒ If S = {x ∈ N : ||x|| ≤ 1}, then T (s) is bounded in N ′
=⇒ d(T (s)) = α < ∞
=⇒ ||T (s)|| ≤ α, for s ∈ S
x
=⇒ 0 ̸= x ∈ N , s = ||x|| , ||T (s)|| ≤ α
 
x
=⇒ ∃ α > 0 such that ||T ||x|| || ≤ α
=⇒ α > 0 such that ||T (x)|| ≤ α||x||, ∀ x ∈ N
=⇒ T is a bounded linear operator =⇒ (3)

5.4 Exercises:

(1) Each of the following conditions determines a subset of the real

linear space R3 of all triples x = (x1 , x2 , x3 ) of real numbers:
(a) x1 is an integer;
(b) x1 = 0 or x2 = 0;
(c) x1 + 2x2 = 0;
(d) x1 + 2x2 = 1.
Which of these subsets are subspaces of R3 ?

72
(2) Each of the following conditions determines a subset of the real
linear space C[−1, 1] of all bounded continuous real functions y = f (x)
defined on [−1, 1]:
(a) f is differentiable;
(b) f is polynomial of degree 3;
(c) f is an even function, in the sense that f (−x) = f (x) for all x;
(d) f is an odd function, in the sense that f (−x) = −f (x) for all x;
(e) f (0) = 0;
(f) f (0) = 1;
(g) f (x) ≥ 0 for all x.
Which of these subsets are subspaces of C[−1, 1]?

5.5 References

1. G. F. Simmons - Introduction to Topology and Modern Analysis,

Tata McGraw-Hill, New Delhi, 2004.

2. A. E. Taylor - Introduction to Functional Analysis, Willey, New

York, 1958.

−−−−−−−−−−−−−−−−−−−−−

73
Chapter 6

Unit 6: Norm of bounded operator

6.1 Introduction

In this unit, we introduce the concept of Normed spaces and bounded

operators.

6.2 Normed Spaces

We begin by restating the definition of Normed spaces.

Definition 6.2.1. Let X be a real or complex linear space and F be

a field of real or complex numbers. A norm on X is a real function
|| · || : X −→ R defined on X such that for any x, y ∈ X and for all
λ ∈ F,
(1) ||x|| ≥ 0
(2) ||x + y|| ≤ ||x|| + ||y||

74
(3) ||λx|| = |λ|||x||
(4) ||x|| = 0 implies x = 0.

Definition 6.2.2. A normed linear space is a linear space with a

norm.

Definition 6.2.3. A seminorm is defined by omitting property (4) in

the definition of a norm.

Example 6.2.4. (1) Rn is a normed linear space with the norm defined
by
n
X  21
||x|| = |xi |2
i=1

(2) C(S) (the set of real or complex valued continuous functions on S)

is a normed linear space with the norm
||f || = sup{|f (x)| : x ∈ S}.

Proposition 6.2.5. Every normed linear space X is metric space,

relative to the natural metric d defined by
d(x, y) = ||y − x||, ∀ x, y ∈ X.
Furthermore, for any x, y, z ∈ X and for all λ ∈ F , we have
||x|| = d(0, x) as well as
(a) d(x + z, y + z) = d(x, y), (i.e. translation invariance).
(b) d(λx, λy) = |λ|d(x, y).

Proof. It can be easily verified that the axioms for a metric hold good.
For example: d(x, z) ≤ d(x, y)+d(y, z) follows immediately by writing

75
z − x = (y − x) + (z − y)
so that ||z − x|| ≤ ||y − x|| + ||z − y||.
Proof of (a): d(x + z, y + z) = ||(y + z) − (x + z)|| = ||y − x|| = d(x, y).
Proof of (b): d(λx, λy) = ||λy − λx|| = ||λ(y − x)|| = |λ|||y − x|| =
|λ|d(x, y).

Theorem 6.2.6. A normed linear space X is complete if and only if

every absolutely convergent series in X is convergent.

Proof. Let X be complete. For each positive integer n let xn be an

element of X such that

∞
X
||xn || < ∞.
n=1
∞
P
Let yk = xn . Then,
n=1
k+p
X k+p
X
||yk+p − yk || = xn ≤ ||xn ||,
n=k+1 n=k+1

which tends to zero as k → ∞. Hence {yk }∞

k=1 is a Cauchy sequence

in X and since X is complete, there exists x ∈ X such that

k
X ∞
X
x = lim yk = lim xn = xn .
k→∞ k→∞
n=1 n=1
∞
P
Thus the series xn converges.
n=1

Conversely, Suppose every absolutely convergent series in X is

convergent. Let {xn } be a Cauchy sequence in X. For each positive

76
integer k, there is a positive integer nk such that
1
||xn − xm || < , ∀ n, m ≥ nk .
2k

Choose nk+1 > nk . Let y1 = xn1 and yk+1 = xnk+1 − xnk , k ≥ 1.

∞
P
Then ||yk || < ∞. Therefore, there exists y ∈ X such that
k=1
m
X
y = lim yk = lim xnm .
m→∞ m→∞
k=1

Since {xn } is Cauchy sequence, lim xn is also y and this completes

m→∞
the proof of the theorem.

Definition 6.2.7. Let T : N −→ N ′ be a bounded linear operator

from a normed linear space N into a normed linear space N ′ . Then,
∃ α > 0 such that ||T (x)|| ≤ α||x||, ∀ x ∈ N .
By Theorem 5.3.4, ||T (x)|| ≤ α, ∀ x ∈ S.
Define ||T || = sup{||T (x)|| : ||x|| ≤ 1} = sup ||T (x)||.
x∈S

6.3 Bounded linear operator

B(N, N ′ ) = {T : N −→ N ′ : T is a bounded linear operator}, where

N and N ′ are normed linear spaces.
Vector addition: (T1 + T2 )(x) = T1 (x) + T2 (x), ∀ x ∈ N .
Scalar multiplication: (λT )(x) = λT (x), ∀ x ∈ N , λ ∈ F (R or C).
Since N ′ is a linear space over F , B(N, N ′ ) is also a linear space over
F.

77
Theorem 6.3.1. The linear space B(N, N ′ ) over F is a normed linear
space with respect to
||T || = sup{||T (x)|| : ||x|| ≤ 1}.

Proof. To show that || · || is a norm on B(N, N ′ ):

(1) ||T || ≥ 0, by definition.

||T || = 0 ⇐⇒ sup{||T (x)|| : ||x|| ≤ 1} = 0

⇐⇒ if s ∈ S = sup{x ∈ N : ||x|| ≤ 1}, then ||T (s)|| = 0

 
x x
⇐⇒ if x ∈ N, then ∈ S and hence T = 0 (x ̸= 0)
||x|| ||x||
1
⇐⇒ T (x) = 0, ∀ x ∈ N (|| · ||N ′ is norm on N ′ )
||x||
⇐⇒ T = 0.

(2)

||λT || = sup{||(λT )(x)|| : ||x|| ≤ 1}

= sup{|λ|||T (x)|| : ||x|| ≤ 1}

= |λ| sup{||T (x)|| : ||x|| ≤ 1}

= |λ|||T ||

78
(3)

||T1 + T2 || = sup{||(T1 + T2 )(x)|| : ||x|| ≤ 1}

= sup{||T1 (x) + T2 (x)|| : ||x|| ≤ 1}

≤ sup{||T1 (x)|| + ||T2 (x)|| : ||x|| ≤ 1}

= sup{||T1 (x)|| : ||x|| ≤ 1} + sup{||T2 (x)|| : ||x|| ≤ 1}

= ||T1 || + ||T2 ||

B(N, N ′ ) is a normed linear space.

6.4 Exercises

(1) Prove that R* is a normed linear space with following norms

n
P
(a) ||x||1 = |xi |.
i=1
(b) ||x||∞ = max |xi |.
1≤i≤n
(2) Prove that C[a, b] is a normed linear space with ||f || = sup |f (x)|.
x∈[a,b]

6.5 References

1. G. F. Simmons - Introduction to Topology and Modern Analysis,

Tata McGraw-Hill, New Delhi, 2004.

2. A. E. Taylor - Introduction to Functional Analysis, Willey, New

York, 1958.

−−−−−−−−−−−−−−−−−−−−−

79
Chapter 7

Unit 7: Hahn-Banach Extension

Theorem

7.1 Introduction

In this unit, we shall prove the Hahn-Banach Extension Theorem.

This is one of the most fundamental theorems in functional analysis
and is due to Hahn and Banach. It yields the existence of nontrivial
continuous linear functionals on a normed linear space, a basic result
necessary for the development of a large portion of functional analysis.
There are several forms of this famous theorem. We shall first prove
the extended form of the Hahn-Banach Theorem.

7.2 Hahn-Banach Theorems

We shall define some definitions.

80
Definition 7.2.1. Let P be a nonempty set. A partial order relation in
P is a relation symbolized by ≤ and it satisfies the following properties:
(1) Reflexivity: x ≤ x, ∀ x ∈ P .
(2) Anti-symmetry: x ≤ y and y ≤ x =⇒ x = y, ∀ x, y ∈ P .
(3) Transitivity: x ≤ y, y ≤ z then x ≤ z, ∀ x, y, z ∈ P .
Then (P, ≤) is called a partially ordered set.

Definition 7.2.2. A partially ordered set (P, ≤) is said to be totally

ordered set. If x, y ∈ P, then either x ≤ y or y ≤ x.

Definition 7.2.3. Let (P, ≤) be a totally ordered set. Then y ∈ P is

called a maximal element, if x ≥ y ⇒ x = y.

Definition 7.2.4. Let (P, ≤) be a totally ordered set and A is a

nonempty subset of P , then y ∈ P is called an upper bound for A,
if x ≤ y, ∀ x ∈ A.

Definition 7.2.5. If P is a partially ordered set in which every chain

has an upper bound, then P has a maximal element
OR
If P is a totally ordered set in which every subset has an upper bound,
then P has a maximal element.

Definition 7.2.6. Let X be a real linear space. By a sublinear func-

tional on X, we mean p : X → R satisfying the following properties:
(1) p(x + y) ≤ p(x) + p(y), ∀ x, y ∈ X.
(2) p(αx) = αp(x), ∀ x ∈ X, α ≥ 0.

81
Definition 7.2.7. Let X be a complex linear space. By a seminorm
on X, we mean p : X → R satisfying the following properties:
(1) p(x + y) ≤ p(x) + p(y), ∀ x, y ∈ X.
(2) p(αx) = |α|p(x), ∀ x ∈ X, α ∈ C.

Note 7.2.8. (1) If we restrict C to only R, then every seminorm is

also a sublinear functional.
(2) Every norm, by definition, is also a seminorm. Any result proved
on seminorm will hold for norms.
(3) For any seminorm, p(x) ≥ 0.

Theorem 7.2.9 (Hahn-Banach Theorem for any real linear

space). Let X be a real linear space. Let S be a subspace of X. Let
p be a sublinear functional on X. Let f : S → R be a real linear
functional on S such that f (x) ≤ p(x), ∀ x ∈ S. Then there exists a
real linear functional F : X → R such that:
(1) F is an extension of f or F (x) = f (x), ∀ x ∈ S and
domain of F ⊇ domain of f .
(2) F (x) ≤ p(x), ∀ x ∈ X.

Proof. Let F denote the collection of all extensions of f

i.e. g ∈ F =⇒
(i) g(x) = f (x), ∀ x ∈ S and D(f ) ⊆ D(g), (where D(f ) = domain of
f ).
(ii) g(x) ≤ p(x), ∀ x ∈ D(g).

82
Since f ∈ F, F ̸= ∅.
Let us define a partial order “≤” in F as follows:
g1 ≤ g2 ⇐⇒ D(g1 ) ⊆ D(g2 ) and g1 (x) = g2 (x) on x ∈ D(g1 ).
(a) “≤” is reflexive:
g ≤ g because D(g) = D(g), g(x) = g(x) (obvious).
(b) “≤” is antisymmetric:
g1 ≤ g2 and g2 ≤ g1 =⇒ D(g1 ) ⊆ D(g2 ) ⊆ D(g1 ) or D(g1 ) = D(g2 ),
g1 (x) = g2 (x) on D(g1 ) = D(g2 ) =⇒ g1 = g2 .
(c) “≤” is transitive:
Suppose g1 ≤ g2 and g2 ≤ g3 =⇒ D(g1 ) ⊆ D(g2 ) ⊆ D(g3 ) or
D(g1 ) ⊆ D(g3 ), g1 (x) = g2 (x), ∀ x ∈ D(g1 ), or g2 (x) = g3 (x), ∀ x ∈
D(g3 ) =⇒ g1 (x) = g3 (x), ∀ x ∈ D(g1 )
∴ g1 ≤ g3 .
Hence (F, ≤) is a partially ordered set.
Let G be a subcollection of F such that (G, ≤) is a totally ordered set.
Claim G has an upper bound or there exists an extension h of f such
that g ≤ h, ∀ g ∈ G.
S
Set U = D(g). U is a subspace of X.
g∈G
Because, x, y ∈ U and α, β ∈ R.
=⇒ x ∈ D(g1 ) and y ∈ D(g2 ) for some g1 , g2 ∈ G.
=⇒ since G is totally ordered, either g1 ≤ g2 or g2 ≤ g1 .
Without loss of generality, g1 ≤ g2 .
=⇒ x ∈ D(g1 ) ⊆ D(g2 ) and y ∈ D(g2 )

83
=⇒ x, y ∈ D(g2 ), D(g2 ) is a linear subspace of X.
=⇒ αx + βy ∈ D(g2 ) ⊆ U .
=⇒ αx + βy ∈ U .
∴ U is a subspace of X.
Suppose x ∈ U , then x ∈ D(g) for some g ∈ G may be more than one
domain.
If x ∈ D(g1 ) and x ∈ D(g2 ) for some g1 , g2 ∈ G or x ∈ D(g1 ) ∩ D(g2 ),
then, again we can use the fact that G is totally ordered and assume
without loss of generality, g1 ≤ g2 or D(g1 ) ⊆ D(g2 ) which will lead to
x ∈ D(g1 ) ∩ D(g2 ) = D(g1 ).
Hence, every x ∈ U falls or lies exactly in a precise domain of a
members g1 ∈ G.
As a result, define h : U → R by h(x) = g(x) whenever x ∈ D(g) ⊆ U
(g is fixed according to the above discussion).
h is well defined. h is linear because

h(αx + βy) = g(αx + βy)

= αg(x) + βg(y)

= αh(x) + βh(y)

h is an extension of f because

h(x) = g(x), ∀ x ∈ D(g) ∩ S

= f (x).
S
g ≤ h, ∀ g ∈ G because D(g) ⊆ D(g) = U .
g∈G

84
Hence h is an upper bound of F or G.
We have proved that in the partially ordered set (F, ≤), every totally
ordered subset has an upper bound. By Zorn’s Lemma, F has a
maximal element. Let F : X → R be that maximal element i.e.,
(1) F is an extension of f .
(2) F (x) ≤ p(x), ∀ x ∈ D(F ).
The proof will be complete if, we show that D(F ) = X.
We shall show that, if D(F ) ̸= X, then F is not a maximal element.
If D(F ) ̸= X, then set D = Span{D(F )∪{x0 }}, where x0 ∈ X\ D(F ).
For every x ∈ D, we may write x = y + αx0 , where y ∈ D(F ), α∈ R.
∀ y, z ∈ D(F ),

F (z) − F (y) = F (z − y)

≤ p(z − y) = p(z + x0 − y − x0 )

≤ p(z + x0 ) + p(−y − x0 )

or − F (y) − p(−y − x0 ) ≤ p(z + x0 ) − F (z).

Then, {y ∈ D(F ) : −F (y) − p(−y − x0 )} is bounded above and {z ∈

D(F ) : p(z + x0 ) − F (z)} is bounded below.
Let,

b = sup {−F (y) − p(−y − x0 )}

y∈D(F )

and

a = sup {p(z + x0 ) − F (z)}.

z∈D(F )

85
Then a < b and choose c such that a < c < b. As a result
−F (y) − p(−y − x0 ) ≤ c ≤ p(z + x0 ) − F (z).
Now construct a real linear function g : D −→ R by g(x) = F (y) + αc,
(x = y + αx0 ), since F is well defined and linear, g is also well defined
and linear.
To show that g(x) ≤ p(x), ∀ x ∈ D.
Case 1: α = 0.
g(x) = g(y) = F (y) ≤ p(y) = p(x)
∴ g(x) ≤ p(x).
Case 2: α > 0.
c ≤ p(z + x0 ) − F (z), ∀ z ∈ D(F ).
Take z = αy ,

y y
c ≤ p( + x0 ) − F ( ), (α > 0)
α α
y y
αc ≤ αp( + x0 ) − αF ( )
α α
= p(y + αx0 ) − F (y)

= p(x) − F (y)

or F (y) + αc ≤ p(x)

∴ g(x) ≤ p(x).

Case 3: α < 0.
−α −F ( αy ) − p(− αy − x0 ) ≤ c(−α), [(−α) > 0]
 

=⇒ F (y) − p(αx0 + y) ≤ −cα

=⇒ F (y) + cα ≤ p(αx0 + y)

86
g(x) ≤ p(x).
This shows that there is one more maximal element superseding F or
F is not a maximal element. This is impossible.
∴ D(F ) = X. This completes the proof.

Theorem 7.2.10 (Hahn-Banach Theorem for any complex lin-

ear space). Let X be a complex linear space. Let S be a subspace of
X. Let p : X −→ R be a seminorm on X. Let f : S −→ C be a linear
functional on S such that ∥f (x)∥ ≤ p(x), ∀ x ∈ S. Then there exists
a complex linear functional F : X −→ C such that:
(1) F (x) = f (x), ∀ x ∈ S.
(2) |F (x)| ≤ p(x), ∀ x ∈ X.

Proof. First, we observe that f (x) = g(x) + ih(x), ∀ x ∈ S,

where g : S −→ R and h : S −→ R are two real linear functionals.
Next, we want to express h(x) in terms of g(x):
if (x) = ig(x) − h(x)
if (x) = f (ix),
( ∵ f : S −→ C is linear f (αx + βy) = αf (x) + βf (y))
= g(ix) + ih(ix),
we may equate ig(x) − h(x) = g(ix) + ih(ix) =⇒ −h(x) = g(ix) and
g(x) = h(ix)
h(x) = −g(ix) or f (x) = g(x) + ih(x) = g(x) − i(g(ix)),
we shall apply Hahn-Banach Theorem for any real linear space on g(x).
By restricting C or R, every seminorm is called a sublinear functional.

87
g(x) ≤ |f (x)| ≤ p(x), ∀ x ∈ S, (∵ z = Re(z) + iIm(z), Re(z) ≤ |z|).
So, by restricting C to R, X is also a real linear space. S is a subspace
of X. p : S −→ R is a sublinear functional. g : S −→ R is a linear
functional such that g(x) ≤ p(x).
By applying Hahn-Banach Theorem for any real linear space, ∃ a real
linear functional, G : X −→ R such that
(1) G(x) = g(x), ∀ x ∈ S.
(2) G(x) ≤ p(x), ∀ x ∈ X.
Define F : X −→ C by F (x) = G(x) − iG(ix), ∀ x ∈ X.
F is a linear functional:

F (x + y) = G(x + y) − iG(i(x + y))

= G(x + y) − iG(ix + iy)

= [G(x) + G(y)] − i [G(ix) + G(iy)]

= [G(x) − iG(ix)] + [G(y) − iG(iy)]

= F (x) + F (y).

88
 F ((α1 + iα2 )x) = F (α1 x + iα2 x)

= G(α1 x + iα2 x) − iG(i(α1 x + iα2 x))

= G((α1 + iα2 )x) − iG(iα1 x − α2 x)

= G(α1 x) + G(iα2 x) − iG(iα1 x − α2 x)

= G(α1 x) + G(α2 (ix)) − iG(α1 (ix)) − iG(−α2 x)

= α1 G(x) + α2 G(ix) + iα2 G(x) − iα1 G(ix)

= (α1 + iα2 )G(x) − i(α1 + iα2 )G(ix)

= (α1 + iα2 ) [G(x) − iG(ix)]

= (α1 + iα2 )F (x).

∴ F is linear.
∀ x ∈ S, F (x) = G(x) − iG(ix), (∵ G(x) = g(x), ∀ x ∈ S)
∴ F (x) = g(x) − ig(x) = f (x).
Finally, to show that |F (x)| ≤ p(x), ∀ x ∈ X, (∵ z = x+iy = |z| = eiθ )
Note that, F (x) = |F (x)|eiθ

or |F (x)| = F (x)e−iθ

= F (xe−iθ ), ∵ F is linear

= G(xe−iθ ), ∵ |F (x)| is real and positive.

=⇒ F (xe−iθ ) must be real.

89
=⇒ Im F (xe−iθ ) = 0.
 

=⇒ |F (x)| ≤ p(xe−iθ )

= p(x)|e−iθ |, ∵ p(αx) = |α|p(x)

= 1c · p(x), ∵ cos2 (θ) + sin2 (θ) = 1.

∴ |F (x) ≤ p(x), ∀ x ∈ X.
This completes the proof.

Theorem 7.2.11 (Hahn-Banach Theorem for normed linear

space). Let X be a normed linear space. Let S be a subspace of X.
If f : S −→ F (R or C) is a continuous linear functional on S, then
there exists F : X −→ F, a continuous linear functional on X such
that:
(1) F (x) = f (x), ∀ x ∈ S.
(2) ∥F ∥ = ∥f ∥.

Proof. First, we shall define a seminorm on X as follows:

p : X −→ C by p(x) = ∥F ∥∥f ∥, where f : S −→ F is given continuous
linear functional so that
 
x
∥f ∥ = sup : x ∈ S, x ̸= 0 < ∞.
∥x∥

90
(1)

p(x + y) = ∥f ∥∥x + y∥

≤ ∥f ∥{∥x∥ + ∥y∥}

= ∥f ∥∥x∥ + ∥f ∥∥y∥

= p(x) + p(y)

or p(x + y) ≤ p(x) + p(y), ∀ x, y ∈ X.

(2) p(αx) = ∥f ∥∥x∥ = |α|{∥f ∥∥x∥} = |α|p(x).

Also note that |f (x)| ≤ ∥f ∥∥x∥ = p(x), ∀ x ∈ S.
Now, we have, X is a complex linear space (setting F = C) and S is
a subspace of X. p : X −→ C is a seminorm on X, p : f −→ C is a
linear functional on S such that |f (x)| ≤ p(x), ∀ x ∈ S.
Applying Hahn-Banach Theorem for complex linear case, ∃ a complex
linear functional F : X −→ C such that
(i) F (x) = f (x).
(ii) |F (x) ≤ p(x), ∀ x ∈ X.
(ii) =⇒ |F (X)| ≤ ∥f ∥∥x∥, ∀ x ∈ X.
n o
|F (x)|
=⇒ ∥x∥ : x ∈ X ≤ ∥f ∥
n   o
x
=⇒ sup F ∥x∥ : x ∈ X ≤ ∥f ∥
∴ ∥F ∥ ≤ ∥f ∥
Finally, we shall show that ∥F ∥ ≥ ∥f ∥ :

91
      
x |F (x)|
∥F ∥ = sup F : x ∈ X = sup :x∈X
∥x∥ ∥x∥
 
|F (x)|
≥ sup :x∈S ∵ S is a subspace of X.
∥x∥
 
|f (x)|
= sup : x ∈ S , ∵ F (x) = f (x), ∀ x ∈ X
∥x∥
= ∥f ∥

∴ ∥F ∥ ≥ ∥f ∥. Hence ∥F ∥ = ∥f ∥.

7.3 Exercises

1. State and Prove Hahn-Banach Theorem for real linear space.

2. State and Prove Hahn-Banach Theorem for complex linear space.
3. State and Prove Hahn-Banach Theorem for normed linear space.

7.4 References

1. G. F. Simmons - Introduction to Topology and Modern Analysis,

Tata McGraw-Hill, New Delhi, 2004.

2. A. E. Taylor - Introduction to Functional Analysis, Willey, New

York, 1958.

−−−−−−−−−−−−−−−−−−−−−

92
Chapter 8

Unit 8: Stone-Weierstrass
Theorem

8.1 Introduction

The space of all polynomials on [a, b] is dense in C[a, b]. Since p(x) =
a0 + a1 x + · · · + an xn , the space of all polynomials form an algebra gen-
erated by {1, x}. Then, the Weierstrass theorem states that, algebra
of C[a, b] = algebra generated by {1, x}.

If X is topological space, C(X, R) is a lattice (a set P with a

partial order “ ≤ ” is called a lattice, if (P, ≤) is a partial order set
and for every pair (p, q) ∈ P , p ∨ q and p ∧ q, maximal element and
minimal element exists).
Because, f, g ∈ C(X, R) =⇒ (f ∨ q)(x) = max(f (x), g(x))
(f ∧ g)(x) = min(f (x), g(x)).

93
If A ⊆ C(X, R) and A is also a lattice, then we call A as sublattice of
C(X, R).

8.2 Stone-Weierstrass Theorem

In order to state the result we need the following definition.

Definition 8.2.1. C(X, R) is said to separate points of X, if x, y ∈ X,

x ̸= y, then ∃ f ∈ C(X, R) such that f (x) ̸= f (y).

Lemma 8.2.2. If X is a compact Hausdorff space, then C(X, R) sep-

arates points of X.

Proof. Let x ̸= y in X. Since X is Hausdorff space, {x} and {y}

are closed sets in X. X is compact Hausdorff space, so it is normal.
We can use Urysohn’s Lemma ( i.e. let X be a normal space and
let A and B be disjoint closed subspaces of X. Then there exists a
continuous real function f defined on X, all of whose values lie in the
closed unit interval [0, 1], such that f (A) = 0 and f (B) = 1), ∃ a
continuous function f on X such that f ({x}) = 0 and f ({y}) = 1 or
f (x) ̸= f (y).

Theorem 8.2.3 ( Stone-Weierstrass Theorem in real case). Let

X be a compact Hausdorff space and let C(X, R) denotes the space of
all real bounded continuous functions on X. If A is a closed subalge-
bra of C(X, R) which separates points of X and contains a non zero

94
constant function, then A = C(X, R).

Proof. Step 1: To show that A is a closed sublattice of C(X, R). First,

we observe that if f ∈ C(X, R), |f | ∈ C(X, R).
Here, |f |(x) = |f (x)|, ∀ x ∈ X.
Next, we note that

(f ∨ g)(x) = max{f (x), g(x)}

(f + g) + |f − g|
=
2

(f ∧ g)(x) = min{f (x), g(x)}

(f + g) − |f − g|
= .
2

Because,
Case 1: max{f (x), g(x)} = f (x)

=⇒ |f − g| = f − g
(f + g) + (f − g)
=⇒ f = (f ∨ g)(x) = = f.
2
(f + g) − (f − g)
Also, =⇒ g = (f ∧ g)(x) = = g.
2

Case 2: max{f (x), g(x)} = g(x)

=⇒ |f − g| = g − f
(f + g) + (g − f )
=⇒ g = (f ∨ g)(x) = = g.
2
(f + g) − (g − f )
Also, =⇒ f = (f ∧ g)(x) = = f.
2
95
 So, we want to show that a closed subalgebra is also a closed
sublattice of C(X, R). i.e., maximal and minimal elements exists in
the subalgebra.

This reduces to show that if f ∈ A, a closed subalgebra of

C(X, R), then |f | ∈ A.

Let ϵ > 0, we know that |t| is a continuous function on [a, b]. By

Weierstrass theorem (i.e., let f : [a, b] → R be a continuous function
and let ϵ > 0 be given. Then ∃ a polynomial p(x) ∈ R[x], such that
|f (x) − p(x)| < ϵ), there exist a polynomial p(t) such that
||t| − p(t)| < ϵ.
Take [a, b] = [−||f ||, ||f || ], since A is a subalgebra of C(X, R),
p(f ) ∈ A ⊆ C(X, R).
Then, ||f (x)| − p(f (x))| < ϵ, ∀ x ∈ X.
=⇒ || |f (x)| − p(f (x)) || < ϵ.
Hence |f (x)| ∈ A = A.
Hence A is a closed sublattice of C(X, R).

Step 2: To show that, if L is any closed sublattice of C(X, R) satis-

fying the property: if x, y are two distinct points of X and if a, b are
two real numbers, then ∃ f ∈ L such that f (x) = a and f (y) = b,
then L = C[a, b].
Since L is a closed sublattice of C(X, R), L ⊆ C(X, R).
To show that C(X, R) ⊆ L or h ∈ C(X, R) =⇒ h ∈ L.

96
Since L is closed, it is enough to show that for every ϵ > 0, ∃ gϵ ∈ L
such that |gϵ − h| < ϵ.
Let x, y ∈ X and x ̸= y. Taking, a = h(x) and b = h(y), then ∃ f ∈ L
such that
f (x) = a = h(x) and f (y) = b = h(y).

Consider the open set Gf = {z : f (z) < h(z) + ϵ}.

Keep x fixed and vary y in X. Then the corresponding f is denoted
by fy and the corresponding Gf is denoted by Gfy or simply Gy .
Then, Gy = {z ∈ X : fy (z) < h(z) + ϵ} is an open set, ∀ y ∈ X.
S
=⇒ X = Gy. But X is compact, ∃, y1 , y2 , . . . , yn ∈ X such that
y
n
S
X= Gyi .
i=1
Correspondingly, in L, ∃ f1 , f2 , . . . , fn such that
fi (z) < h(z) + ϵ, ∀ i = 1, 2, . . ..
Let gx = f1 ∧f2 ∧· · · wedgefn . Infact gx (x) = h(x) and gx (z) < f (z)+ϵ,
∀ z ∈ X.
Next, consider for every x, the set Hx = {z ∈ X : gx (z) > f (z) − ϵ}
(hence, f (z) − ϵ < gx (z) < f (z) + ϵ, ∀ z ∈ X). Hx is an open set in
X.
[
∴X= Hx .
x
m
S
X is compact, hence ∃ x1 , x2 , . . . , xm ∈ X such that X = Hxi .
i=1
Correspondingly, ∃ g1 , g2 , . . . , gm ∈ L such that
f (z) − ϵ < gi (z) < f (z) + ϵ, ∀ i = 1, 2, . . . , m.

97
Let gϵ = g1 ∨ g2 ∨ · · · ∨ gm . Then gϵ ∈ L such that
h(z) − ϵ < gϵ (z) < h(z) + ϵ or |h(z) − gϵ (z)| < ϵ.

Step 3: To show that A = C(X, R).

Case 1: Suppose X has only one point, then C(X, R) contains only
constant functions. Since A also contains a non zero constant function
and A is a subalgebra of C(X, R). Hence A contains all the constant
functions or A = C(X, R).
Case 2: Suppose X has more than one point, then A is a closed
sublattice (by Step 1), containing a non zero constant function and
separates point of X. If x ̸= y in X, ∃ g ∈ A such that g(x) ̸= g(y) or
g(x) − g(y) ̸= 0.
Let a, b be any two real numbers. Set
g(z) − g(y) g(z) − g(x)
f (z) = a +b
g(x) − g(y) g(y) − g(x)
g ∈ L =⇒ f ∈ L,
g(x) − g(y) g(x) − g(x)
f (x) = a +b = a,
g(x) − g(y) g(y) − g(x)
g(y) − g(y) g(y) − g(x)
f (y) = a +b = b,
g(x) − g(y) g(y) − g(x)
using Step 2, A = C(X, R).
This completes the proof.

Theorem 8.2.4 (Stone-Weierstrass Theorem for complex case).

Let X be a compact Hausdorff space and A be closed subalgebra of
C(X, C), which separates the points of X, contains non zero constant

98
function and contains conjugates of each of its functions. Then A
equal to C(X, C).

Proof. First, we note that the real functions in A forms a subalgebra

B of C(X, R). B is also subalgebra of A. If f ∈ C(X, C) is an arbitrary
element, then f = Re (f )+iIm (f ), so that Re (f ) and Im (f ) ∈ C(X, R).
If we show that B = C(X, R), then A = C(X, C).
B is a closed subalgebra of C(X, R). B separates points of X. Because,
it is given that A separates points of X.
=⇒ ∀ x ̸= y ∈ X, ∃ f ∈ A such that f (x) ̸= f (y)
=⇒ Re(f (x)) + iIm(f (x)) ̸= Re(f (y)) + iIm(f (y))
=⇒ Re(f )(x) ̸= Re(f )(y) and Im(f )(x) ̸= Im(f )(y)
=⇒ Re(f ) and Im(f ) are in B and separates points of X.
It is given that A has a nonzero constant function
=⇒ fc ∈ A
=⇒ fc ∈ A, ∵ f ∈ A =⇒ f ∈ A
=⇒ fc fc = |fc |2 is a real function in A.
=⇒ |fc | is a non zero constant function in B.
B satisfies all the hypothesis of Stone-Weierstrass theorem in real case.
Hence, B = C(X, R), as a consequence A = C(X, C).
This completes the proof.

Corollary 8.2.5. Weierstrass theorem on C[a, b] is a corollary to

Stone-Weierstrass theorem in real case.

99
Proof. X = [a, b] =⇒ C[a, b] = C(X, R)
P [a, b] = {p(x) ∈ C[a, b] : p(x) = a0 + a1 x + · · · + an xn , a1 , a2 , . . . , an ∈
R}.
The nonzero constant function 1 ∈ P [a, b]. The function f (x) = x ∈
P [a, b], obviously separates points of X = [a, b].
A = P [a, b] is a closed subalgebra of C([a, b], R) containing nonzero
constant function 1 and f (x) = x separates points of [a, b].
∴ C[a, b] = A = P [a, b]. This completes the proof.

Note 8.2.6. P (A, B) is dense in C[a, b]

=⇒ given ϵ > 0 and f ∈ C[a, b], ∃ pϵ ∈ P [A, B] such that
|f − pϵ | < ϵ.
This is called Classical Weierstrass Approximation theorem.

Corollary 8.2.7. Weierstrass theorem is true in several variables.

Proof. X = Sr (x0 ) = {x ∈ Rn : ||x − xo|| ≤ r}

C(X, R), A = P (X, R), closures of the space of all polynomials in n-
variables with real coefficients.
A = C(X, R).

8.3 Exercises:

(1) Prove the two-variable Weierstrass approximation theorem: if

f (x, y) is a real function defined and continuous on the closed rectangle

100
X = [a, b] × [c, d] in the Euclidean plane R2 , then f can be uniformly
approximated on X by polynomials in x and y with real coefficients.

(2) Let X be the closed unit disc in the complex plane, then show
that any function f in C(X, C) can be uniformly approximated on X
by polynomials in z and z with complex coefficients.

(3) Let X and Y be compact Hausdorff spaces and f a function in

C(X ×Y, C). Show that f can be uniformly approximated by functions
n
P
of the form fi gi where the fi ’s are in C(X, C) and the gi ’s are in
i=1
C(Y, C).

8.4 References

1. G. F. Simmons - Introduction to Topology and Modern Analysis,

Tata McGraw-Hill, New Delhi, 2004.

2. A. E. Taylor - Introduction to Functional Analysis, Willey, New

York, 1958.

−−−−−−−−−−−−−−−−−−−−−

101
Block-III

102
Chapter 9

Unit 9: Banach Spaces

9.1 Introduction

We introduce the concept of Banach spaces and give the properties of

Banach spaces.

9.2 Banach Spaces

Banach space is a linear space which is also, in a special way, com-

plete metric space. This combination of algebraic and metric structure
opens up the possibility of studying linear transformations of one Ba-
nach space into another which have the additional property of being
continuous.

Definition 9.2.1. A Banch space is a complete normed linear space.

Example 9.2.2. (1) X = R or C, F = R or C is a Banach space

103
with the norm is ∥x∥ = |x|.
(2) X = Rn or Cn = F n is a Banach space with the norm is
p
x = (x1 , x2 , . . . , xn ) =⇒ ∥x∥ = |x1 |2 + |x2 |2 + · · · + |xn |2 .
 !1/p 
 x
X 
p
(3) X = lp = x = (x1 , x2 , . . . , xn ) : |xj | < ∞, xj ∈ F = R or C
 
j=1

is a Banach space with the norm is

∞
!1/p
X
∥ x ∥p = |xj |p .
j=1
a p bq
Lemma 9.2.3. If p > 1, q > 1, p1 + 1q = 1 and a, b > 0 then ab < p+q.

Proof.
a
ap
Z
A1 = xp−1 dx = .
0 p

y = xp−1 ⇐⇒ x = y 1/p−1 = y q−1 .

b
bq
Z
A2 = y q−1 dy = .
0 q
ap bq
∴ ab < A1 + A2 = + .
p q

Theorem 9.2.4 (Holder’s Inequality). Let (x1 , x2 , . . . , xn ) and (y1 , y2 , . . . , yn )

be n-tuples. Then
n n
!1/p n
!1/q
X X X
|xj yj | ≤ |xj |P |yj |q
j=1 j=1 j=1
1 1
where p > 1, q > 1 and p + q = 1.

104
Proof.
n
!1/p n
!1/q
X X
P ut α = |xj |P and β = |yj |q .
j=1 j=1

xj yj
Put x′j = α and yj′ = β,

n
!1/p n
!1/p n
!1/p
X 1 X 1 X 1
|x′j |P = p
|xj |P = |xj |P = α = 1.
j=1
α j=1 α j=1
α

n
!1/q
X
Similarly |yj′ |q = 1.
j=1

Consider,
n n  ′ p ′ q
X X |y
|xj | j |
|x′j yj′ | ≤ + , using Lemma 9.2.3
j=1 j=1
p q
n
! n
!
1 X 1 X
= |x′j |P + |yj′ |q
p j=1 q j=1
1 1
= + = 1.
p q

n
X x j yj
⇒ ≤1
j=1
α β
n
1 X
⇒ |xj yj | ≤ 1
αβ j=1
n n
!1/p n
!1/q
X X X
⇒ |xj yj | ≤ αβ = |xj |p |yj |q
j=1 j=1 j=1

This completes the proof.

105
Problem 9.2.5. Let C ∗ (X) = {f : X → F : F = R or C}, where f
is bounded continuous function on X, with
vector addition: (f + g)(x) = f (x) + g(x), ∀x ∈ X.
scalar multiplication: (αf )(x) = αf (x), ∀ x ∈ X
and norm
∥ f ∥= sup |f |.
x∈X

Since, F is linear space on itself, C ∗ (X) is also a linear space on F .

Show that C ∗ (X) is Banach space.

Solution (i) ∥ f ∥≥ 0,

∥ f ∥= 0 ⇐⇒ sup |f (x)| = 0.
x∈X
⇐⇒ |f (x)| = 0, ∀ x ∈ X.

⇐⇒ f (x) = 0, ∀ x ∈ X.

⇐⇒ f = 0, ∀ x ∈ X.

(ii)
∥ λf ∥= sup |λf (x)| = λ sup |f (x)| = λ ∥ f ∥ .
x∈X x∈X

(iii)

∥ f + g ∥ = sup |(f + g)(x)

x∈X
= sup |f (x) + g(x)|
x∈X
≤ sup |f (x)| + sup |g(x)|
x∈X x∈X
= ∥f ∥+∥g∥.

106
 C ∗ (X) is complete.
Let {fn } be a Cauchy sequence in C ∗ (X).

∥ fm − fn ∥ = sup {|(fm − fn )(x)|}

x∈X
= sup {|fm (x) − fn (x)|}.
x∈X

=⇒ {fn (x)} is a Cauchy sequence in F and F is complete.

Hence {fn (x)} → f (x) as n → ∞, ∀ x ∈ X.
∥ fm − fn ∥≥ |(fm (x) − fn (x))|
or |(fm (x) − fn (x))| ≤∥ fm − fn ∥→ 0 as m, n → ∞.
=⇒ {fn } converges uniformly to f .
=⇒ Being uniform limit of a sequence of continuous bounded func-
tions, f is also continuous bounded function.
=⇒ f ∈ C ∗ (X).
Hence C ∗ (X) is complete and it is a Banach space.

Problem 9.2.6. Show that l∞ is a Banach space.

Solution. We know that

l∞ = |{x = (x1 , x2 , . . . , xn , . . .) : xi ∈ R or C, sup |xi | < ∞}.

1≤i≤∞

Set X = {1, 2, . . . , n, . . .}, F = R or C.

Then a sequence in F is nothing but a function f : X → F , f (n) = xn .
Obviously, f is continuous function on X.
Also, sup |xi | < ∞
1≤i≤∞
=⇒ f is also bounded.
∴ l∞ = C ∗ (X), X = {1, 2, . . . , n, . . .} is also Banach space.

107
9.3 Exercises

(1) Let N be a nonzero normed linear space and prove that N is a

Banach space {x : ||x|| = 1} is complete.
(2) Let p > 1, then prove that Minkowski’s inequality:

∥x + y∥p ≤ ∥x∥p + ∥y∥p .

9.4 References

1. G. F. Simmons - Introduction to Topology and Modern Analysis,

Tata McGraw-Hill, New Delhi, 2004.

2. A. E. Taylor - Introduction to Functional Analysis, Willey, New

York, 1958.

−−−−−−−−−−−−−−−−−−−−−

108
Chapter 10

Unit 10: Open Mapping Theorem

10.1 Introducation

In this unit, we introduce the concept of Open mapping theorem.

10.2 Open Mapping Theorem

In order to state the the result we need the following definition.

Definition 10.2.1. A mapping T : X −→ Y from topological space X

into topological space Y is said to be continuous ⇐⇒ T −1 is open in
X whenever G is open in Y .

Definition 10.2.2. A mapping T : X −→ Y from topological space X

into topological space Y is said to be open mapping, if G is open in
X, then T (G) is open in Y .

Lemma 10.2.3. If X and Y are two Banach spaces and if

109
T : X −→ Y is continuous linear map from X onto Y , then the image
of every open sphere centered on the origin of X contains an open
sphere centered on the origin of Y .

Proof. Let Sr = {x ∈ X : ∥x − 0∥x < r} be an open sphere in X,

centered on origin of X.
Let Sr′ = {y ∈ Y : ∥y − 0′ ∥Y < r} be an open sphere in Y , centered
on origin of Y .
Consider,

Sr = {x ∈ X : ∥x∥ < r}
∥x∥
= {x ∈ X : < 1}
r
x
= {x ∈ X : < 1}.
r

x
Put r = x1

Sr = {rx1 ∈ X : ∥x1 ∥ < 1}

= rS1

Therefore, Sr = rS1 , ∀ r > 0.

From this, we note that it is enough to prove the Lemma for S1 .
′
i.e. to prove that, T (S1 ) ⊇ Sr1 , for some r1 > 0.

110
Next, let us note that
∞
[
X = Sn .
n=1
∞
!
[
=⇒ T (x) = T Sn
n=1
∞
[
= T (Sn ).
n=1

It is given that T is onto, hence T (x) = Y .

∞
S
∴Y = T (Sn ).
n=1
Since Y is complete, we can apply Baire’s Category Theorem, i.e. Y
is of second category or Y is not a union of nowhere dense sets.
 
=⇒ ∃ T (Sn0 ) such that Int T (Sn0 ) ̸= ∅.
 
Let y0 ∈ Int T (Sn0 ) or y0 ∈ T (Sn0 ).
Next, consider z ∈ T (Sn0 ) − y0 .
=⇒ z = Tx − y0 , where x ∈ Sn0
=⇒ z = Tx − Tx0 , ∴ y0 ∈ T (Sn0 ) =⇒ ∃ x0 ∈ Sn0 such that
T (x0 ) = y0 .
=⇒ z = T (x − x0 ).
∥x − x0 ∥ ≤ ∥x∥ + ∥x0 ∥ < n0 + n0 = 2n0 , ∴ x − x0 ∈ S2n0 .
∴ z ∈ T (S2n0 )
and hence,
T (Sn0 ) − y0 ⊆ T (S2n0 ).
We shall observe that origin of Y is an interior point of T (Sn0 ) − y0 .
The map y −→ y − y0 is a homeomorphism from Y onto Y . Since

111
y0 is an interior point of T (Sn0 ), 0′ = y0 − y0 is an interior point of
T (x0 ) − y0 . Hence origin of Y is an interior point of T (S2n0 ).
∴ 0′ is an interior point of T (S2n0 ),
because, T (Sn0 − y0 ) = T (Sn0 ) − y0 ⊆ T (S2n0 ).
But, we know that T (S2n0 ) = T (2n0 S1 ) = 2n0 T (S1 ), ∵ Sr = rS1 .
So that, T (S2n0 ) = 2n0 T (S1 ),
we can have one more homeomorphism y −→ 2n0 y from Y onto Y .
Hence origin 0′ is an interior point of T (S1 ).
By definition, ∃ ϵ > 0 such that 0′ ∈ Sϵ′ ⊂ T (S1 ) using this, we shall
show that
Sϵ′ ⊆ T (S3 ) ⇐⇒ S ′ϵ ⊆ T (S1 ),
3

which will prove the lemma.

We shall work out inductively two sequences {yn } in Y and {xn }

in X as follows:
Let y ∈ Sϵ′ , then y ∈ T (S1 ) or y is a limit point of T (S1 )) (if y ∈
T (S1 ) ⊆ T (S3 ) there is nothing to prove).
ϵ
=⇒ 2, ∃ x1 ∈ S1 such that y1 = T (x1 ) and ∥y − y1 ∥ < 2ϵ ,
∵ x1 ∈ S1 ⇐⇒ ∥x1 ∥ < 1.
=⇒ ∥(y − y1 ) − 0′ ∥ < ϵ
2 or y − y1 ∈ S ′ϵ using 0′ ∈ Sϵ′ ⊆ T (S1 ), we get
2

Sϵ′ ⊆ T (S1 ).
Sϵ′ T (S1 )
=⇒ S ′ϵ = 2 ⊆ 2 = T (S 12 ), ∵ T is linear.
2

=⇒ y − y1 ∈ T (S 12 ) or y − y1 is a limit point of T (S 12 ).
ϵ 1
=⇒ 22 , ∃ x2 ∈ S 21 or ∥x∥ < 2 such that y2 = T (x2 ) and ∥y −y1 −y2 ∥ <

112
ϵ
2.

=⇒ y − y1 − y2 ∈ S ′ϵ2 .
2
′
Again using 0 ∈ Sϵ′ ⊆ T (S1 ), we get
Sϵ′ T (S1 )
Sϵ′ ⊆ T (S1 ) ⇔ S ′ϵ2 = 22 ⊆ 22 = T (S 12 ).
2 2

Continuing this way we arrive at two sequences: {xn } is a sequence in

X such that
1
∥x∥ < , n = 1, 2, . . . .
2n−1
ϵ
and yn is a sequence in Y such that ∥y − y1 − y2 − · · · − yn ∥ < 2n .

=⇒ y = lim (y1 + y2 + · · · + yn )
n→∞

= lim (T (x1 ) + T (x2 ) + · · · + T (xn ))

n→∞

= lim (T (x1 + x2 + · · · + xn ))
n→∞

= lim (T (Sn )),

n→∞

where Sn = x1 + x2 + · · · + xn .

n
X
∥Sn ∥ = xi
i=1
n
X
≤ ∥xi ∥
i=1

∵ |x1 | < 1, |x2 | < 21 , . . ., |xn | < 1

2n .

113
 1 1
∥Sn ∥ < 1 + + · · · + n−1
2 2
1 − 21n
=
1 − 12
< 2.

To show that {Sn } is a Cauchy sequence in X:

n
X
∥Sn − Sm ∥ = xi
i=m+1
Xn
≤ ∥xi ∥
i=m+1
1 1 1
< + + · · · +
2m  2m+1  2n−1
1
1 1 − 2n−m
= m
2 1 − 12
 
1 1
= m−1 1 − n−m
2 2
1
= m−1 −→ 0 as n, m −→ ∞
2
∴ {Sn } is a Cauchy sequence in X and X is complete.
Let x = lim Sn .
n→∞
Since T is continuous, T (x) = limn→∞ T (Sn ) = y.
∥x∥ = lim ∥Sn ∥ ≤ 2 < 3.
n→∞
=⇒ x ∈ S3
=⇒ y = T (x) ∈ T (S3 ) or S ′ (ϵ) ⊆ T (S3 ) or S ′ (ϵ/3) ⊆ T (S1 ).
This completes the proof.

Theorem 10.2.4 (Open mapping theorem). If X and Y are two

114
Banach spaces and if T : X −→ Y is a continuous linear map from X
onto Y , then T is an open mapping.

Proof. y ∈ T (G) =⇒ y = T (x) for some x ∈ G.

G is open in X =⇒ ∃ r > 0 such that x ∈ S(x, r) ⊆ G.
But we know that S(x, r) = x + Sr .
By the Lemma 10.2.3, ∃ r1 > 0 such that Sr′ 1 ⊆ T (Sr ).
S ′ (y, r1 ) = y + Sr′ 1 ⊆ y + T (Sr ) = T (x + Sr ) = T (S(x, r)).
∴ y ∈ S ′ (y, r1 ) ⊆ T (S(x, r)) ⊆ T (G)
∴ T (G) is open.

Theorem 10.2.5. If T : X −→ Y is a one to one, continuous linear

transformation (or bounded linear operator) from X onto Y , then T
is a homeomorphism from a Banach space X onto Banach space Y .

Proof. It is given that T is a continuous linear transformation, one to

one and onto from a Banach space X to a Banach space Y .

By Open Mapping Theorem, G ⊆ X is open =⇒ T (G) ⊆ Y is

also open.

=⇒ T (G) ⊆ Y is open =⇒ T −1 (T (G)) = G ⊆ X is open.

=⇒ T −1 is continuous.

T −1 (y) = T −1 (T (x)) = x.

115
 =⇒ T −1 (αy1 + βy2 ) = T −1 (αT (x1 ) + βT (x2 ))

= T −1 (T (αx1 + βx2 ))

= αx1 + βx2

= αT −1 (y1 ) + βT −1 (y2 )

T is one to one and onto =⇒ T −1 is also one to one and onto.

∴ T is homeomorphism.

Problem 10.2.6. If T : X −→ Y is a one to one continuous linear

transformation from a Banach space X onto Banach space Y , then
T −1 : Y −→ X is also continuous linear transformation.

Solution: To show that T −1 is continuous.

By Open mapping theorem, G ⊆ X is open =⇒ T (G) ⊆ Y is also
open.
=⇒ T (G) ⊆ Y is open =⇒ T −1 (T (G)) = G ⊆ X is open.
To show that T −1 is linear:

T −1 (αy1 + βy2 ) = T −1 (αT (x1 ) + βT (x2 ))

= T −1 (T (αx1 + βx2 ))

= αx1 + βx2

= αT −1 (y1 ) + βT −1 (y2 ).

116
10.3 Exercise:

Let T : X −→ Y be the continuous linear map from the Banach space

X to Banach space Y . Then prove that T (G) is open in Y if G is open
in X.

10.4 References

1. G. F. Simmons - Introduction to Topology and Modern Analysis,

Tata McGraw-Hill, New Delhi, 2004.

2. A. E. Taylor - Introduction to Functional Analysis, Willey, New

York, 1958.

−−−−−−−−−−−−−−−−−−−−−

117
Chapter 11

Unit 11: Closed Graph Theorem

11.1 Introducation

In this unit, we introduce the concept of Closed graph theorem.

11.2 Closed Graph Theorem

First, we define some definition and then go to the theorem.

Definition 11.2.1. Let T : X −→ Y be a linear transformation from

a normed linear space X into a normed linear space Y . Then T is
said to be a closed linear operator if
xn −→ x in X and Txn −→ y in Y =⇒ y = Tx .

Definition 11.2.2. Let T : X −→ Y be a linear transformation from

a normed linear space X into a normed linear space Y . Then T is
said to be continuous if xn −→ x in X and Txn −→ Tx .

118
Note 11.2.3. Every continuous linear operator is also a closed linear
operator. But the converse is not true.

Counter example
Let

X = {x : [0, 1] −→ R : ∥x∥ = sup |x(t)|, x(t) and x′ (t) are continuous.}

0≤t≤1

Y = {y : [0, 1] −→ R : ∥y∥ = sup |y(t)|, y(t) is continuous.}

0≤t≤1

T : X −→ Y , by Tx (t) = dx
dt = x′ (t).
T is linear:
d
T (αx1 (t) + βx2 (t)) = [αx1 (t) + βx2 (t)]
dt
dx1 dx2
= α +β
dt dt
= αTx1 + βTx2 .

T is closed:
Suppose xn −→ x in X and Txn −→ y in Y .
i.e. xn (t) −→ x(t) in X and x′n (t) −→ y(t) ∈ Y.
=⇒ ∥xn − x∥ −→ 0 and ∥x′n − y∥ −→ 0 as n −→ ∞.
=⇒ xn (t) −→ x(t) uniformly and x′n (t) −→ y(t) uniformly on [0, 1].
By the theorem on uniform convergence and differentiation,
x′n −→ x(t) uniformly and hence y = x′ (t).
∴ y = Tx or T is a closed linear operator.
But T is not continuous.
( T is continuous ⇐⇒ ∃ α > 0 such that ∥Tx ∥ ≤ α∥x∥ ⇐⇒ x′ (t) is

119
uniformly continuous.)
Consider, xn (t) = tn .

∥xn ∥ = sup tn = 1,
0≤t≤1

then

dxn
T xn = = ntn−1
dt
∥Txn ∥ = sup ntn−1 = n.
0≤t≤1

As n −→ ∞, ∥xn ∥ −→ 1.
But ∥Txn ∥ −→ ∞
∴ ∥Txn ∥ ≰ α∥xn ∥.
∴ T is not continuous.

Definition 11.2.4. Cartesian product of two normed linear space

X, Y is defined as
X × Y = {(x, y) : x ∈ X, y ∈ Y }.
X × Y is also called as normed linear space with the following opera-
tions:
vector addition: (x1 , y1 ) + (x2 , y2 ) = (x1 + x2 , y1 + y2 ).
scalar multiplication: α(x, y) = (αx, αy)
and norm ∥x, y∥ = ∥x∥x + ∥y∥y .
The zero element of X × Y is (0x , 0y ).

Definition 11.2.5. Let T : X −→ Y be a linear operator from a

normed linear space X into a normed linear space Y . Then graph of

120
T is defined by

Γ(T ) = {(x, T (x)) : x ∈ X} ⊆ X × Y.

Note 11.2.6. Γ(T ) is a subspace of X × Y :

α(x1 , T (x1 )) + β(x2 , T (x2 )) = (αx1 , αT (x1 )) + (βx2 , βT (x2 ))

= (αx1 + βx2 , αT (x1 ) + βT (x2 )).

But T is linear =⇒

T (αx1 + βx2 ) = αT (x1 ) + βT (x2 )

= (αx1 + βx2 , T (αx1 + βx2 )) ∈ Γ(T ),

∀ α, β ∈ F , and (x1 , T (x1 )), (x2 , T (x2 )) ∈ Γ(T ).

Problem 11.2.7. Show that if T : X −→ Y is a closed linear operator

from a normed space X into a normed linear space Y , then Γ(T ) is a
closed subspace of X × Y .

Solution: If (x, y) ∈ Γ(T )

=⇒ ∃ {(xn , T (xn ))} ∈ Γ(T ) such that lim (xn , T (xn )) = (x, y).
n→∞
=⇒ ∥(xn − x, T (xn ) − y)∥ −→ 0 as n −→ ∞.
=⇒ ∥xn − x∥ + ∥T (xn ) − y∥ −→ 0 as n −→ ∞.
=⇒ ∥xn − x∥ −→ 0 and ∥T (xn ) − y)∥ −→ 0 as n −→ ∞.
=⇒ xn −→ x in X and T (xn ) −→ y in Y .
Since T : X −→ Y is a closed linear operator y = T (x).
∴ (x, y) = (x, T (x)) ∈ Γ(T )
∴ Γ(T ) = Γ(T ) or Γ(T ) is closed subspace of X × Y .

121
Theorem 11.2.8 (Closed Graph Theorem). Let X and Y be two
Banach spaces and let T : X −→ Y is a linear transformation from
X into Y . Then T is continuous ⇐⇒ Γ(T ) is closed.

Proof. Suppose T is continuous. Let (x, y) be a limit point of Γ(T ).

Then by definition of limit point, ∃ a sequence {(xn , T (xn ))} in Γ(T )
such that lim (xn , T (xn )) = (x, y).
n→∞
=⇒ ∥(xn , T (xn )) − (x, y)∥ −→ 0 as n −→ ∞.
=⇒ ∥(xn − x, T (xn ) − y)∥ −→ 0 as n −→ ∞.
=⇒ ∥xn − x∥ + ∥T (xn ) − y∥ −→ 0 as n −→ ∞.
=⇒ ∥xn − x∥ −→ 0 and ∥T (xn ) − y∥ −→ 0 as n −→ ∞.
(T is continuous ⇐⇒ xn −→ x =⇒ T (xn ) −→ T (x))
=⇒ As xn −→ x, T (xn ) −→ y and T (xn ) −→ T (x).
=⇒ y = T (x).
∴ (x, y) = (x, T (x)) ∈ Γ(T ).
Γ(T ) has all its limit points or Γ(T ) = Γ(T ).
∴ Γ(T ) is closed.
Suppose Γ(T ) is closed subspace of X × Y . It is given that X and Y
are Banach spaces.
To show that X × Y is also a Banach space:
X × Y = {(x, y) : x ∈ X, y ∈ Y } is a vector space on common field F
under the following operations:
vector addition: (x1 , y1 ) + (x2 , y2 ) = (x1 + x2 , y1 + y2 ).
scalar multiplication: α(x, y) = (αx, αy).

122
X × Y is also a normed linear space under the norm

∥(x, y)∥ = ∥x∥ + ∥y∥.

To show that X × Y is also complete.

Let {(xn , yn )} be a Cauchy sequence in X × Y .
=⇒ ∥(xm , ym ) − (xn − yn )∥ −→ 0 as m, n −→ ∞
=⇒ ∥(xm − xn , ym − yn )∥ −→ 0 as m, n −→ ∞
=⇒ ∥xm − xn ∥X −→ 0 and ∥ym − yn ∥Y −→ 0 as m, n −→ ∞.
=⇒ {xn } and {yn } are Cauchy sequences in X and Y respectively.
Also, X and Y are complete.
=⇒ ∃ x ∈ X, y ∈ Y such that lim xn −→ x and lim yn −→ y.
n→∞ n→∞
=⇒ lim (xn , yn ) = (x, y) ∈ X × Y .
n→∞
∴ X × Y is a Banach space.
Next, X × Y is complete and Γ(T ) is a closed subspace of X × Y .
=⇒ Γ(T ) is also complete.
=⇒ Γ(T ) is also a Banach space.
Define a map, ϕ : Γ(T ) −→ X by ϕ(x, T (x)) = x.
To show that ϕ is linear.

ϕ(α(x1 , T (x1 )) + β(x2 , T (x2 ))) = ϕ((αx1 + βx2 , αT (x1 ) + βT (x2 )))

= ϕ((αx1 + βx2 , T (αx1 + βx2 )))

∵ T is linear.

= αx1 + βx2

= αϕ((x1 , T (x1 ))) + βϕ((x2 , T (x2 ))).

123
∴ ϕ is linear.
To show that ϕ is continuous:
∥ϕ((x, T (x)))∥ = ∥x∥ ≤ ∥x∥ + ∥T (x)∥ = ∥(x, T (x))∥.
∴ ϕ is continuous. By definition, ϕ is one to one and onto.
∴ ϕ is one to one continuous linear mapping from the Banach space
Γ(T ) onto the Banach space X.
=⇒ By Theorem 10.2.5., ϕ−1 is also a one to one continuous linear
mapping from X onto Γ(T ).
=⇒ ∥ϕ−1 ∥ = α > 0, is well defined.
Consider,

∥T (x)∥ ≤ ∥x∥ + ∥T (x)∥

= ∥(x, T (x))∥

= ∥ϕ−1 (x)∥, ∵ ϕ((x, T (x))) = x =⇒ (x, T (x)) = ϕ−1 (x).

≤ ∥ϕ−1 ∥∥x∥

= α∥x∥.

∴ ∥T (x)∥ ≤ α∥x∥ or T is continuous. This completes the proof.

Problem 11.2.9. To prove closed graph theorem we have used one of

the consequence of open mapping theorem, namely, if ϕ : X −→ Y is
a one to one continuous linear mapping from a Banach space X onto
a Banach space Y , then ϕ−1 : Y −→ X is also one to one continuous
linear mapping from Y onto X.

124
Solution: Let X and Y be Banach spaces.
Given the Closed graph theorem is true. i.e if T : X −→ Y is a linear
transformation, then T is continuous ⇐⇒ Γ(T ) is closed.
In the theorem, ϕ : X −→ Y is a one to one, continuous linear mapping
from X onto Y .
To prove that ϕ−1 is one to one, linear, continuous and onto:
ϕ−1 is continuous ⇐⇒ Γ(ϕ−1 ) is closed.
Γ(ϕ−1 ) = {(y, ϕ−1 y) : y ∈ Y }.
Let (y, x) be a limit point of T (ϕ−1 )
=⇒ ∃ (yn , ϕ−1 (yn )) in Γ(ϕ−1 ) such that lim (yn , ϕ−1 (yn )) = (y, x)
n→∞
−1
=⇒ ∥(yn , ϕ (yn )) − (y, x)∥ −→ 0 as n −→ ∞.
=⇒ ∥(yn − y, ϕ−1 (yn ) − x)∥ −→ 0 as n −→ ∞
=⇒ ∥yn − y∥ −→ 0 and ∥ϕ−1 (yn ) − x∥ −→ 0 as n −→ ∞.
Since ϕ is onto, yn = ϕ(xn ).
=⇒ ∥ϕ(xn ) − y∥ −→ 0 and ∥ϕ−1 (ϕ(xn )) − x∥ −→ 0 as n −→ ∞.
=⇒ ∥xn − x∥ −→ 0 and ∥ϕ(xn ) − y∥ −→ 0 as n −→ ∞.
Since ϕ is continuous, ϕ is closed. ϕ(xn ) −→ ϕ(x) and y = ϕ(x) or
x = ϕ−1 y
(ϕ : X −→ Y is closed if xn −→ x and ϕ(xn ) −→ y, then y = ϕ(x)).
∴ ϕ−1 (yn ) −→ x = ϕ−1 (y) as yn −→ y.
∴ (y, x) = (y, ϕ−1 y) ∈ Γ(ϕ−1 )
∴ Γ(ϕ−1 ) is closed.
By Closed graph theorem, ϕ−1 is continuous.

125
ϕ is linear, one to one and onto =⇒ ϕ−1 is linear, one to one and onto.
This completes the proof.

11.3 Exercises

If T : X −→ Y is a closed linear operator from a normed space X into

a normed linear space Y , then show that Γ(T ) is a closed subspace of
X ×Y.

11.4 References

1. G. F. Simmons - Introduction to Topology and Modern Analysis,

Tata McGraw-Hill, New Delhi, 2004.

2. A. E. Taylor - Introduction to Functional Analysis, Willey, New

York, 1958.

−−−−−−−−−−−−−−−−−−−−−

126
Chapter 12

Unit 12: Banach - Steinhaus

Principle of Uniform Boundedness

12.1 Introduction

In this unit, we introduce te concept of Banach - Steinhaus prin-

ciple of uniform boundedness.

12.2 Uniform Bounded Principle

Theorem 12.2.1 (The Banach - Steinhaus Theorem (Uniform

Bounded Principle)). Let B be a Banach space and let N be normed
linear space. Let {Ti } be a non empty set of continuous linear trans-
formations from B into N with the property that {Ti (x)} is bounded
set of N for every x ∈B. Then {∥Ti ∥} is a bounded set of numbers i.e.,
{Ti } is bounded in B(B, N ).

127
Proof. For each positive number n, define

Fn = {x ∈ B : ∥Ti (x)∥ ≤ n, ∀ i}
\
= Ti−1 (S(0, n)).
i

Because, x ∈ Fn ⇐⇒ Ti (x) ∈ S(0, n) in N.

⇐⇒ x ∈ Ti−1 (S(0, n)) in B, ∀ i.

Since, each Ti is continuous.

Ti−1 (S(0, n)) is also closed and hence each Fn is closed, being arbitrary
intersection of closed sets.
∞
S ∞
S
Now, B = Fn . Because, if B ̸= Fn , then ∃ x ∈ B such that
n=1 n=1
x∈
/ Fn , ∀ n. So that ∥Ti (x)∥ > n, ∀ n.
This means, there is one x such that {Ti (x)} is not bounded set of
N which is against to hypothesis. Since, B is a Banach space, B is
complete.
By Baire’s Category Theorem, B is of second category. So, ∃ n0 such
that
Int(Fn0 ) = Int(Fn0 ) ̸= ∅.

Let x0 ∈ Int(Fn0 ). This implies ∃ r0 > 0 such that

x0 ∈ S(x0 , r0 ) ⊆ Int(Fn0 )
=⇒ x0 ∈ S(x0 , r0 ) ⊆ Fn0 .

128
Next, let y ∈ B and ∥y∥ ≤ 1,
 
z
=⇒ ∥Ti (y)∥ = Ti , where z = r0 y =⇒ ∥z∥ < r0 .
r0
1
= ∥Ti (z)∥
r0
1
= ∥Ti ((z + x0 ) − x0 )∥
r0
1
≤ [∥Ti (z + x0 )∥ + ∥Ti (x0 )∥].
r0
Since, ∥(z + x0 ) − x0 ∥ = ∥z∥ = ∥r0 y∥ = r0 ∥y∥ ≤ r0 ,
z + x0 ∈ S(x0 , r0 ).
1
∥Ti (y)∥ ≤ [∥Ti (z + x0 )∥ + ∥Ti (x0 )∥]
r0
1 2n0
≤ [n0 + n0 ] = .
r0 r0
=⇒ ∥Ti ∥ = sup{∥Ti (y)∥ : ∥y∥ ≤ 1}
2n0
≤ , ∀ i.
r0
=⇒ {∥Ti ∥} is bounded in B(B, N ).
This completes the proof.

Problem 12.2.2. Let B be a Banach space and N be a normed linear

space. Let {Tn } be a sequence in B(B, N ) such that lim Tn (x) = T (x)
n→∞
exists for all x ∈ B. Show that T ∈ B(B, N ).

Solution: Given:

1. B is a Banach space.

2. N is a normed linear space.

129
 3. {Tn } is a sequence of continuous linear transformation from B to
N.

4. lim Tn (x) = T (x) exists for all x ∈ B.

n→∞

From (4), ∥Tn (x) − T (x)∥ −→ 0asn −→ ∞, ∀x ∈ B.

=⇒ ∥(Tn − T )(x)∥ −→ 0 as n −→ ∞, ∀ x ∈ B.
=⇒ {(Tn − T )(x)} is a bounded sequence of N for all x ∈ B.
We can apply principle of uniform boundedness on {(Tn − T )(x)} to
get {∥Tn − T ∥} is a bounded set of numbers.
=⇒ ∥Tn − T ∥ ≤ α, ∀ x ∈ B.
=⇒ ∥(Tn − T )(x)∥ ≤ α, ∀ x ∈ B
=⇒ ∥Tn (x) − T (x)∥ ≤ α∥x∥, ∵ |a − b| ≥ |a| − |b|.
=⇒ ∥T (x) − Tn (x)∥ ≤ α∥x∥
=⇒ ∥T (x)∥ − ∥Tn (x)∥ ≤ α∥x∥
=⇒ ∥T (x)∥ ≤ ∥Tn (x)∥ + α∥x∥
Since Tn is continuous ∃ αn > 0 such that ∥Tn (x)∥ ≤ αn ∥x∥.
=⇒ ∥T (x)∥ ≤ (αn + α)∥x∥.
∴ T is continuous or T ∈ B(B, N ).

Theorem 12.2.3. Let N be a normed linear space. A non empty

subset X of N is bounded ⇐⇒ f (X) is a bounded set of numbers for
every f ∈ N ∗ , where N ∗ is the conjugate of N .

Proof. Suppose X is bounded =⇒ diam(X) < ∞.

=⇒ ∀ x ∈ X, ∥x∥ ≤ diam(X) < ∞.

130
Given: f ∈ N ∗ =⇒ f : N −→ F is a continuous linear transforma-
tion.
=⇒ f : N −→ F is a bounded linear transformation.
=⇒ ∥f ∥ < ∞.
Consider,
diam(f (X)) = sup |f (x) − f ′ (x)|
x,x′ ∈X

≤ 2 sup |f (x)| < ∞

x∈X

provided |f (x)| < ∞, ∀ x ∈ X.

|f (x)| ≤ ∥f ∥∥x∥ ≤ ∥f ∥diam(X) < ∞
=⇒ diam(f (x)) < ∞.
∴ f (X) is bounded set of numbers ∀ f ∈ N ∗ .

Suppose f (X) is a bounded set of numbers ∀ f ∈ N ∗ .

To prove X is bounded:
Let us write X = {xi } (i need not run over countably infinite) for our
convenience.

Next, consider the natural embedding xi −→ Fxi of N into N ∗∗ .

Since, f (X) = f (xi ) is bounded set of numbers ∀ f ∈ N ∗ , the corre-
sponding set {f (xi )} or {Fxi (f )} is bounded set of N ∗∗ , ∀ f ∈ N ∗ .

Let us recall the principle of uniform boundedness. Let B be

a Banach space and let N1 be normed linear space. Let {Ti } be a
nonempty set of continuous linear transformation B into N1 with the

131
property that {Ti (x)} is bounded set of N1 for every x ∈ B. Then
{∥Ti ∥} is bounded set of numbers i.e. {Ti } is bounded in B(B, N ).
Set B = N ∗ , N1 = N ∗∗ and {Ti } = {Fxi }.

Then by principle of uniform boundedness, {Fxi } is bounded set

of numbers. But by the constructions, ∥Fxi ∥ = ∥xi ∥.
Hence {∥xi ∥} is a bounded set of numbers or X is bounded in N .
This completes the proof.

Theorem 12.2.4. Let N be a normed linear space. Let T ∈ B(N )

i.e. T : N −→ N is a continuous linear transformation on N. Define
T ∗ : N ∗ −→ N ∗ by (T ∗ (f ))(x) = f (T (x)). Then T ∗ is also a continu-
ous linear transformation from N to F or T ∗ (f ) ∈ N ∗ , ∀ f ∈ N ∗ .
Further, T −→ T ∗ is an isometric isomorphism from B(N ) into B(N ∗ ).

Proof. To show that T ∗ is linear:

[T ∗ (αf + βg)](x) = (αf + βg)(T (x))

= αf (T (x)) + βg(T (x))

= α[T ∗ (f )](x) + β[T ∗ (g)](x)

= [αT ∗ (f ) + βT ∗ (g)](x), ∀ x ∈ N.

132
 To show that T ∗ is continuous:

∥T ∗ (f )∥ = sup{|[T ∗ (f )](x)| : ∥x∥ ≤ 1}

= sup{|f (T (x))| : ∥x∥ ≤ 1}

= ∥f T ∥

≤ ∥f ∥∥T ∥, ∵ ∥T1 T2 ∥ ≤ ∥T1 ∥∥T2 ∥.

Since T ∈ B(N ), take α = ∥T ∥ < ∞.

∴ ∥T ∗ (f )∥ ≤ α∥f ∥
It remains to show that ∥T ∗ ∥ = ∥T ∥:

∥T ∗ ∥ = sup{∥T ∗ (f )∥ : ∥f ∥ < 1}

= ∥T ∥, (∥T ∗ (f )∥ ≤ α∥f ∥).

 
∥T (x)∥
∥T ∥ = sup : x ̸= 0 in N
∥x∥
 
|f (T (x))|
= sup : x ̸= 0 in N, ∥f ∥ = 1 ,
∥x∥
∵ ∥f ∥ = 1 =⇒ ∥T (x)∥ = |f (T (x))|
 ∗ 
|T (f )|
= sup : x ̸= 0 in N, ∥f ∥ = 1
∥x∥
 ∗ 
∥T (f )∥∥x∥
≤ sup : x ̸= 0 in N, ∥f ∥ = 1
∥x∥
= ∥T ∗ ∥.

133
12.3 Properties of T ∗

Proposition 12.3.1. (1) T ∗ is also a continuous linear transforma-

tion from N into F

(2) ∥T ∗ ∥ = ∥T ∥ (Isometry)

(3) (T1 T2 )∗ = T2∗ T1∗ (reversing the product property).

(4) I ∗ = I (Preserving the identity).

Proof. Proof of (1) and (2) follows from the Theorem 12.2.4.
Proof of (3):

[(T1 T2 )∗ (f )](x) = f ((T1 T2 )(x))

= f (T1 (T2 (x))), ∵ f (T (x)) = [T ∗ (f )](x).

= T1∗ (f (T2 (x)))

= T2∗ [T1∗ (f )(x)]

= [(T2∗ T1∗ )(f )](x), ∀ x ∈ N, f ∈ N ∗ .

∴ implies(T1 T2 )∗ = T2∗ T1∗ .

Proof of (4):

[I ∗ (f )](x) = f (I(x))

= f (x)

= [I(f )](x)

∴ I ∗ = I.

134
12.4 Exercises

1. Let B be a Banach space and N a normed linear space. If {Tn }

is a sequence of in B(B, N ) such that T (x) = lim Tn (x) exists
n→∞
for each x in B, then prove that T is a continuous linear trans-
formation.

2. Let T be an operator on a normed linear space N . If N is consid-

ered to be part of N ∗∗ by means of the natural embedding, then
show that T ∗∗ is an extension of T . Further show that if N is
reflexive, then T ∗∗ = T .

12.5 References

1. G. F. Simmons - Introduction to Topology and Modern Analysis,

Tata McGraw-Hill, New Delhi, 2004.

2. A. E. Taylor - Introduction to Functional Analysis, Willey, New

York, 1958.

−−−−−−−−−−−−−−−−−−−−−

135
Block-IV

136
Chapter 13

Unit 13: Hilbert Spaces

13.1 Introduction

In this unit, we introduce the concepts of Hilbert spaces.

Definition 13.1.1. Let H be a linear space or vector space over the

field F = C. The inner product on H is a mapping (, ) : H × H −→ F
satisfying the following properties:

1. (x, x) ≥ 0, (x, x) = 0 if and only if x = 0, for all x ∈ H.

2. (x, y) = (y, x), for all x, y ∈ H.

3. (αx + βy, z) = α(x, z) + β(y, z), for all α, β ∈ F, x, y, z ∈ H.

We define (H, (, )) as inner product space.

Theorem 13.1.2. If (H, (, )) is an inner product space, then H is a

normed linear space where
p
∥x∥ = (x, x), for all x ∈ H.

137
Proof. We verify the following three properties of the norm:

p
1. ∥x∥ = (x, x) ≥ 0 (∵ (x, x) ≥ 0 ) for all x ∈ H, and
∥x∥ = 0 ⇐⇒ ∥x2 ∥ = 0
⇐⇒ (x, x) = 0
⇐⇒ x = 0, (∵ (x, x) = 0 ⇐⇒ x = 0, ∀ x ∈ H).

2.
p
∥λx∥ = (λx, λx)
q
= λλ(x, x)
p
= |λ2 |∥x∥2

= |λ|∥x∥, ∀ λ ∈ F, x ∈ H.

3. The Schwart’s inequality is |(x, y)| ≤ ∥x∥∥y∥.

Consider,

∥x + y∥2 = (x + y, x + y)

= (x, x) + (x, y) + (y, x) + (y, y)

h i
2
= ∥x∥ + (x, y) + (x, y) + ∥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 ≤ (∥x∥ + ∥y∥)2

=⇒ ∥x + y∥ ≤ ∥x∥ + ∥y∥ ∀ x, y ∈ H.

138
Hence H is a normed linear space.

13.2 Hilbert space

Definition 13.2.1. A Banach space (or a complete normed linear

space) is said to be Hilbert space, if the norm is defined by the inner
product. In other words, a complete inner product space defining a
Banach space is called a Hilbert space.

Example 13.2.2.
∞
( )
X
l2 = x = (x1 , x2 , . . . , xn , . . .) : |xn |2 < ∞, xn ∈ C, n = 1, 2, . . . .
n=1

l2 is vector space over C under co-ordinatewise vector addition and

co-ordinatewise scalar multiplication. l2 is a Banach space under the
norm
∞
!1/2
X
∥x∥ = |xn |2 .
n=1

We show that the above norm is given by the following inner product:
∞
X
(x, y) = xn yn .
n=1

139
The series on the right hand side is absolutely convergent by Holder’s
inequality
∞
X ∞
X
xn yn = |xn yn |
n=1 n=1
∞
!1/2 ∞
!1/2
X X
≤ |xn |2 |yn |2
n=1 n=1
< ∞.

We can verify that (x, y) is an inner product.

Finally,
∞
!1/2
X p
2
∥x∥ = |xn | = (x, x) for all x ∈ H.
n=1

Hence l2 is a Hilbert space.

13.3 Exercises:
( 1/2 )
∞

|xn |2
P
1. Show that l2 = x = (x1 , x2 , . . . , xn , . . .) : < ∞, xi ∈ F .
n=1
is a Hilbert space.
Rb
2. Show that L2 (I) = {f : I −→ R or C : a |f |2 dx < ∞} is a
Hilbert space.

13.4 References

1. G. F. Simmons - Introduction to Topology and Modern Analysis,

Tata McGraw-Hill, New Delhi, 2004.

140
2. A. E. Taylor - Introduction to Functional Analysis, Willey, New
York, 1958.

−−−−−−−−−−−−−−−−−−−−−

141
Chapter 14

Unit 14: Properties of Hilbert

Spaces

14.1 Properties

In this section, we shall give the some properties of the Hilbert spaces.

Theorem 14.1.1. In a Hilbert space, the parallelogram law holds:

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

Proof. Consider,

∥x + y∥2 + ∥x − y∥2 = (x + y, x + y) + (x − y, x − y), using ∥x∥2 = (x, x)

= (x, x) + (x, y) + (y, x) + (y, y)

+(x, x) − (x, y) − (y, x) + (y, y)

= 2(x, x) + 2(y, y)

= 2∥x∥2 + 2∥y∥2 = 2 ∥x∥2 + ∥y∥2 .

142
Theorem 14.1.2. In a Hilbert space H, the inner product is contin-
uous. i.e. if xn −→ x, yn −→ y as n −→ ∞, then (xn , yn ) −→ (x, y)
as n −→ ∞.

Proof.

|(xn , yn ) − (x, y)| = |(xn , yn ) − (xn , y) + (xn , y) + (x, y)|

= |(xn , yn − y) + (xn − x, yn )|

≤ |(xn , yn − y)| + |(xn − x, yn )|

≤ ∥xn ∥∥yn − y∥ + ∥xn − x∥∥yn ∥,

(using Schwarz’ inequality)

= ∥xn ∥ × 0 + 0 × ∥y∥ = 0 as n −→ ∞.

This completes the proof.

 
Note 14.1.3. (1) lim (xn , y) = lim xn , y = (x, y).
n→∞ n→∞
 
(2) lim (x, yn ) = x, lim yn = (x, y).
n→∞ n→∞

(3) lim (xn , yn ) = (x, y).

n→∞

Theorem 14.1.4. In a Hilbert space H,

4(x, y) = ∥x + y∥2 − ∥x − y∥2 + i∥x + iy∥2 − i∥x − iy∥2 ,

√
where i = −1.

143
Proof. Consider,

∥x + y∥2 − ∥x − y∥2 + i∥x + iy∥2 − i∥x − iy∥2

= (x + y, x + y) − (x − y, x − y) + i(x + iy, x + iy) − i(x − iy, x − iy)

= {(x, x) + (x, y) + (y, x) + (y, y)} − {(x, x) − (x, y) − (y, x) + (y, y)}

+ i {(x, x) + (x, iy) + (iy, x) + (iy, iy)}

− i {(x, x) − (x, iy) − (iy, x) + (iy, iy)}

= 2(x, y) + 2(y, x) + 2i(x, iy) + 2i(iy, x)

= 2(x, y) + 2(y, x) + 2i(i)(x, y) + 2i(i)(y, x)

= 2(x, y) + 2(y, x) + 2i(−i)(x, y) + 2i(i)(y, x)

= 2(x, y) + 2(y, x) + 2(x, y) − 2(y, x)

= 4(x, y).

14.2 Exercise

Show that in a Hilbert space H, the inner product is continuous.

14.3 References

1. G. F. Simmons - Introduction to Topology and Modern Analysis,

Tata McGraw-Hill, New Delhi, 2004.

144
2. A. E. Taylor - Introduction to Functional Analysis, Willey, New
York, 1958.

−−−−−−−−−−−−−−−−−−−−−

145
Chapter 15

Unit 15: Orthogonal projection

and nearly orthogonal elements

15.1 Introduction

In this unit, we introduce the concept of orthogonal projection and

give the properties of orthogonality.

15.2 Orthogonal projection

We discuss here some basic concepts in the theory of orthogonality

and orthogonal complements.

Definition 15.2.1. Let H be a Hilbert space. Then two elements x, y ∈

H are said to be orthogonal or x ⊥ y ⇐⇒ (x, y) = 0.

146
Theorem 15.2.2. In a Hilbert space H, if x ⊥ y, then

∥x + y∥2 = ∥x∥2 + ∥y∥2 = ∥x − y∥2

Proof.

∥x + y∥2 = (x + y, x + y)

= (x, x) + (x, y) + (y, x) + (y, y)

= (x, x) + (x, y) − (y, x) + (y, y)

= ∥x∥2 + ∥y∥2 .

Similarly,

∥x − y∥2 = (x − y, x − y)

= (x, x) − (x, y) − (y, x) + (y, y)

= (x, x) − (x, y) + (y, x) + (y, y)

= ∥x∥2 + ∥y∥2 .

Definition 15.2.3. Let H be a Hilbert space and S ⊆ H. Orthogonal

complement of S, denoted by S ⊥ is defined by

S ⊥ = {x ∈ H : (x, y) = 0, ∀ y ∈ S}

147
Note 15.2.4. (1) If S = H, then S ⊥ = {0}.
(2) If S = {0}, then S ⊥ = H.

Theorem 15.2.5. Let H be a Hilbert space and S ⊆ H. Then S ⊥

satisfies the following properties:
(1) S ∩ S ⊥ = {0}.
(2) S ⊥ is a closed subspace of H.
(3) S1 ⊆ S2 =⇒ S2⊥ ⊆ S1⊥ .
(4) S ⊆ S ⊥⊥ .

Proof. (1)

x ∈ S ∩ S ⊥ ⇐⇒ x ∈ S and x ∈ S ⊥

⇐⇒ (x, x) = 0

⇐⇒ x = 0

⇐⇒ x ∈ {0}

∴ S ∩ S ⊥ = {0}

(2) To show that S ⊥ is a subspace of H :

Let α, β ∈ F , and x1 , x2 ∈ S ⊥
=⇒ α, β ∈ F , (x1 , y) = 0 = (x2 , y) for all y ∈ S.
=⇒ (αx1 + βx2 , y) = α(x1 , y) + β(x2 , y) = α × 0 + β × 0 = 0.
=⇒ αx1 + βx2 ∈ S ⊥ .
∴ S ⊥ is a subspace of H.

148
 To show that S ⊥ is a closed subspace of H :
Let x be a limit point of S ⊥ . Then ∃ {xn } in S ⊥ such that lim xn = x.
n→∞
Consider, for all y ∈ S,
 
(x, y) = lim xn , y = lim (xn , y) = 0.
n→∞ n→∞

∴ x ∈ S ⊥.
Hence S ⊥ has all of its limit points or S ⊥ is a closed subspace of H.

(3) Suppose, S1 ⊆ S2 . To show that S2⊥ ⊆ S1⊥ .

Let x ∈ S2⊥ . Then (x, y) = 0, ∀ y ∈ S2 .
=⇒ (x, y) = 0, ∀ y ∈ S2 ⊇ S1 .
=⇒ (x, y) = 0, ∀ y ∈ S1 .
∴ x ∈ S1⊥ .
Hence S2⊥ ⊆ S1⊥ .

(4) To show that S ⊆ S ⊥⊥ :

⇐⇒ If x ∈ S, then x ∈ S ⊥⊥ ,
/ S ⊥⊥ , then x ∈
⇐⇒ If x ∈ / S.
/ S ⊥⊥
Suppose, x ∈
=⇒ (x, y) ̸= 0, for some y ∈ S ⊥
/ S, because x ∈ S and y ∈ S ⊥ will lead to (x, y) = 0, which
=⇒ x ∈
is not possible.

This completes the proof.

149
Theorem 15.2.6. Let M be a closed linear subspace of a Hilbert space
H. Then H = M ⊕ M ⊥ .

Proof. Step - 1: First, let us prove that if M, N are two closed sub-
spaces of a Hilbert space H and M ⊥ N , then M + N is also a closed
suspace of H.
Let α, β ∈ F, z1 , z2 ∈ M + N
=⇒ αz1 + βz2 = α(x1 + y1 ) + β(x2 + y2 ) = (αx1 + βx2 ) + (αy1 + βy2 )
(where z1 = x1 + y1 and z2 = x2 + y2 , x1 , x2 ∈ M , y1 , y2 ∈ N ).
Since αx1 + βx2 ∈ M , being a subspace of H and αy1 + βy2 ∈ N ,
being a subspace of H. We must have αz1 + βz2 ∈ M + N .
Hence M + N is a subspace of H.
Suppose z is a limit point of M + N .
=⇒ ∃ a sequence {zn } in M + N such that lim zn = z.
n→∞
Since M ⊥ N , M ∩N = {0}. As a result zn ∈ M +N =⇒ zn = xn +yn
written uniquely. Next consider
zn − zm = (xn − xm ) + (yn − ym ), where (xn − xm ) ⊥ (yn − ym ). By
Pythagoras theorem,

∥zn − zm ∥2 = ∥xn − xm ∥2 + ∥yn − ym ∥2 .

Since, {zn } is convergent in M + N , it is Cauchy sequence and

hence
∥zn − zm ∥2 −→ 0 as m, n −→ ∞

=⇒ ∥xn − xm ∥ −→ 0, ∥yn − ym ∥ −→ 0, as m, n −→ ∞.

150
=⇒ {xn } is a Cauchy sequence in M and {yn } is Cauchy sequence in
N.
Every closed subspace of a complete space is complete =⇒ M, N are
complete.
=⇒ lim xn = x, lim yn = y, where x + y ∈ M + N .
n→∞ n→∞
∴ M + N is also a closed subspace of H.

Now, it is given M is a closed subspace of H and M ⊥ is always

a closed subspace of H. Hence M + M ⊥ is a closed subspace of H.
Further, M ∩ M ⊥ = {0}. As a result, M + M ⊥ = M ⊕ M ⊥ .

Step 2: To show that H = M + M ⊥ .

Suppose H ̸= M +M ⊥ . Then M +M ⊥ will be a proper closed subspace
of H. Then there exist 0 ̸= z0 ∈ H such that z0 ⊥ M + M ⊥ . Then
z0 ⊥ M and z0 ⊥ M ⊥⊥ orz0 ∈ M ⊥ ∩ M ⊥⊥ = {0} or 0 ̸= z0 = 0, which
is contradiction.
This completes the proof.

Definition 15.2.7. Let H be a Hilbert space. A non empty set {ei }

is said to be orthonormal set in H, if
(i) ei ⊥ ej ∀ i ̸= j.
(ii) ∥ei ∥ = 1, ∀ i.

Example 15.2.8. In l2 , {ei } = {e1 , e2 , . . . , en , . . .}, where e1 = (1, 0, . . . , 0, . . .),

e2 = (0, 1, 0, . . . , 0, . . .), . . ., en = (0, 0, . . . , 0, 1, 0, . . .), . . ., is an or-

151
thonormal set.

Theorem 15.2.9 (Bessel’s inequality). If {ei } is an orthonormal

set in Hilbert space H and if x ∈ H, then
n
X
|(x, ei )|2 ≤ ∥x∥2 .
i=1

Proof. First, let us observe that,

n 2
X
x− (x, ei )ei ≥0
i=1
n n
!
X X
=⇒ x− (x, ei )ei , x − (x, ei )ei ≥0
i=1 i=1
n
! n
!
X X
=⇒ (x, x) − x, (x, ei )ei − (x, ei )ei , x
i=1
n n
!i=1
X X
+ (x, ei )ei , (x, ei )ei ≥0
i=1 i=1
X n n
X n
X
2
=⇒ ∥x∥ − (x, ei )(x, ei ) − (x, ei )(ei , x) + (x, ei )(x, ei ) ≥ 0
i=1 i=1 i=1
Xn n
X n
X
=⇒ ∥x∥2 − |(x, ei )|2 + |(x, ei )|2 + |(x, ei )|2 ≥ 0.
i=1 i=1 i=1

n
X
∴ |(x, ei )|2 ≤ ∥x∥2 , if {ei } is finite.
i=1

If {ei } is countable, allow n −→ ∞, we get

n
X
|(x, ei )|2 ≤ ∥x∥2 .
i=1

152
Definition 15.2.10. Let H be a Hilbert space. Let P denote the class
of all orthonormal sets in H. Under set inclusion ⊆, (P, ⊆) is a
partially ordered set. By Zorn’s lemma, there exists a maximal element
for the chain of orthonormal sets in H. This set or the maximal
orthonormal set is called complete orthonormal set.

Theorem 15.2.11. Let {ei } be an orthonormal set in Hilbert space

H. Then the following are equivalent:
(1) {ei } is complete;
(2) x ⊥ {ei } =⇒ x = 0;
P
(3) x ∈ H =⇒ x = (x, ei )ei (called Fourier expansion of x);
(4) x ∈ H =⇒ ∥x∥2 = |(x, ei )|2 (called Parsval’s identity).
P

Proof. (1) =⇒ (2):

Suppose if possible, x ⊥ {ei } and x ̸= 0.
x
Put e = ∥x∥ . Then {ei , e} is also an orthonormal set containing {ei }.
=⇒ {ei } is not complete, a contradiction.
∴ x = 0.

(2) =⇒ (3)
We have, x ⊥ {ei } =⇒ x = 0.

153
Consider,
 X  X
x− (x, ei )ei , ej = (x, ei ) − (x, ei )(ei , ej )

= (x, ei ) − (x, ei )

= 0, ∀ ej .
P
=⇒ x − (x, ei )ei ⊥ {ei }.
P P
∴ x − (x, ei )ei = 0 or x = (x, ei )ei .

(3) =⇒ (4):
P
We have x ∈ H =⇒ x = (x, ei )ei

∥x∥2 = (x, x)
X X 
= (x, ei )ei , (x, ej )ej
XX
= (x, ei )(x, ej )(ei , ej ), ∵ (ei , ej ) = 0, (ei , ei ) = 1.
i j
X X
= (x, ei )(x, ei ) = |(x, ei )|2 .

(4) =⇒ (1)
Suppose (1) is false,
=⇒ ∃ e ∈ H such that e ⊥ {ei } and ∥e∥ = 1.
=⇒ ∥e∥2 = |(e, ei )|2 .
P

=⇒ 1 = 0, a contradiction.
This completes the proof.

154
15.3 Exercises

1. Show that the parallelogram law is not true in l1n (n > 1).

2. If S is a nonempty subset of a Hilbert space, then show that

S ⊥ = S ⊥⊥⊥ .

15.4 References

1. G. F. Simmons - Introduction to Topology and Modern Analysis,

Tata McGraw-Hill, New Delhi, 2004.

2. A. E. Taylor - Introduction to Functional Analysis, Willey, New

York, 1958.

−−−−−−−−−−−−−−−−−−−−−

155
Chapter 16

Unit 16: Riesz Lemma and Riesz

Representation Theorem

16.1 Introduction

In this unit, we introduce the Riesz Lemma and Riesz representation

theorem.

16.2 Riesz Lemma

Lemma 16.2.1 (Riesz Lemma). If M is a closed proper subspace

of a normed linear space N and ϵ > 0, then there is a point xϵ on the
unit sphere such that

inf{||x0 − xϵ || : xo ∈ M } > 1 − ϵ.

Proof. Let x ∈ N − M and let us define d = inf{||x − x0 || : x0 ∈ M }.

156
 Since M is closed, d must be strictly greater than zero. For oth-
erwise d = 0 implies x ∈ M from the property closed sets, contracting
the choice of x.

From the definition of infimum, choose a y0 ∈ M such that

d < ||y0 − x|| ≤ d(1 + ϵ). (16.1)

Now let us define

(y0 − x)
xϵ = −
||y0 − x||
Since x ∈ N − M and y0 ∈ M , y0 − x ̸= 0 and also ||xϵ || = 1. If
x0 ∈ M , then we get
y0 − x
||x0 − xϵ || = x0 +
||y0 − x||
1
= [ ||(y0 + ||y0 − x||x0 ) − x|| ]. (16.2)
||y0 − x||
Now (y0 + ||y0 − x||x0 ) ∈ M and x ∈ N . So by using the definition of
d and (16.1) in (16.2),
d d ϵ
||x0 − xϵ || > > =1− > 1 − ϵ.
||y0 − x|| d(1 + ϵ) 1+ϵ
Thus we have proved that ||xϵ || = 1 and ||x0 − xϵ || > 1 − ϵ. This
completes the proof of lemma.

16.3 Riesz representation theorem

Definition 16.3.1. Let H be a Hilbert space. Then

H ∗ = {f : H −→ F : f is a continuous linear functional}.

157
Theorem 16.3.2 (Riesz representation theorem). Let H be a
Hilbert space and let f be an arbitrary functional in H ∗ . Then there
exists a unique vector y ∈ H such that f (x) = (x, y) for every x in H.

Proof. It is easy to see that if such a y exists, then it is necessarily

unique. For if we also have f (x) = (x, y ′ ) for all x, then (x, y ′ ) = (x, y)
and (x, y ′ − y) = 0 for all x; and since 0 is the only vector orthogonal
to every vector, this implies that y ′ − y = 0 or y ′ = y.

We now turn to the problem of showing that y does exist. If

f = 0, then it clearly suffices to choose y = 0. We may therefore
assume that f ̸= 0. The null space M of f is thus a proper closed
linear subspace of H, then there exists a non-zero vector y0 which
is orthogonal to M . We show that if α is a suitable chosen scalar,
then the vector α may be, f (x) = (x, y) is true for every x in M ; for
f (x) = 0 for such an x and since x is orthogonal to y0 , we also have
(x, y) = 0. This allows us to focus our attention on choosing α in such
a way that f (x) = (x, y) is true for x = y0 . The condition this imposes
on α is that
f (y0 ) = (y0 , αy0 ) = α∥y0 ∥2 .

f (y0 )
We therefore choose α to be ∥y0 ∥2 and it follows that f (x) = (x, y)
is true for every x in M and for x = y0 . It is easily seen that each
x in M can be written in the form x = m + βy0 with m in M : all
that is necessary is to choose β in such a way that f (x − βy0 ) =

158
 f (x)
f (x) − βf (y0 ) = 0 and this is accomplished by putting β = f (y0 ) . Our
conclusion that f (x) = (x, y) is true for every x in H now follows at
once from

f (x) = f (m + βy0 )

= f (m) + βf (y0 )

= (m, y) + β(y0 , y)

= (m + βy, y)

= (x, y).

This completes the proof.

16.4 Exercises

1. Let H be a Hilbert space. Then show that H ∗ is also a Hilbert

space with respect to the inner product defined by (fx , fy ) =
(y, x). In just the same way, the fact that H ∗ is a Hilbert space
implies that H ∗∗ is a Hilbert space whose inner product is given
by (Ff , Fg ) = (f, g).

2. Let H be a Hilbert space. We have two natural mappings of H

into H ∗∗ , the second of which is onto: the Banach space natural
imbedding x −→ Fx , where Fx (f ) = f (x) and the product map-
ping x −→ fx −→ Ffx , where fx (y) = (y, x) and Ffx = (f, fx ).

159
 Such that these mappings are equal and conclude that H is re-
flexive. Show also that (Fx , Fy ) = (x, y).

16.5 References

1. G. F. Simmons - Introduction to Topology and Modern Analysis,

Tata McGraw-Hill, New Delhi, 2004.

2. A. E. Taylor - Introduction to Functional Analysis, Willey, New

York, 1958.

−−−−−−−−−−−−−−−−−−−−−

160