M.Sc.

(MATHEMATICS)
SEMESTER - III

MATHEMATICS PAPER - II
FUNCTIONAL ANALYSIS
SUBJECT CODE : PSMT/PAMT 302
© UNIVERSITY OF MUMBAI
Prof. Suhas Pednekar
Vice-Chancellor,
University of Mumbai,

Prof. Ravindra D. Kulkarni Prof. Prakash Mahanwar

Pro Vice-Chancellor, Director,
University of Mumbai, IDOL, University of Mumbai,

Programme Co-ordinator : Shri Mandar Bhanushe

Head, Faculty of Science and Technology,
IDOL, University of Mumbai, Mumbai
Course Co-ordinator : Shri Sumit Dubey
Assistant Professor,
Department of Mathematics,
IDOL, University of Mumbai

Course Writers : Dr. Nitin Katake

Associate Professor,
Central University of Kashmir, Srinagar

: Dr Ashok Bhingi
Associate Professor,
St. Xavier’s College(Autonomous), Mumbai

: Shri Mandar Bhanushe

Assistant Professor,
Department of Mathematics,
IDOL, University of Mumbai

: Miss. Taqdis Pawle

Assistant Professor,
S. M. Shetty College, Powai, Mumbai

May 2022, Print - I, ISBN-978-93-95130-05-9

Published by : Director
Institute of Distance and Open Learning ,
University of Mumbai,
Vidyanagari, Mumbai - 400 098.
ipin Enterprises
DTP Composed and : Tantia Jogani
Mumbai Industrial
University Press Estate, Unit No. 2,
Printed by Ground Floor,
Vidyanagari, Sitaram
Santacruz Mill
(E), Compound,
Mumbai - 400098
J.R. Boricha Marg, Mumbai - 400 011
CONTENTS
Unit No. Title Page No.

1. Baire Spaces 01

2. Hilbert spaces 13

3. Normed Spaces 31

4. Banach Space 63

5. Bounded Linear Transformations and Dual Spaces 89

6. Four Pillars of Functional Analysis 103


PSMT 302 / PAMT 302 Functional Analysis

Course Outcomes:

1. Students will learn Hilbert spaces and Banach spaces.

2. Students will be able to understand the concept of dimension of a Hilbert space,

bounded linear transformations, norms, inner products, dual spaces and their differ-
ence from the finite dimensional cases.

3. Students should know about `p , Lp spaces, dual spaces and their properties;

4. Students should understand the fundamental theorems as mentioned in the syllabus.

Unit I: Baire spaces and Hilbert spaces (15 Lectures)

Baire spaces. Open subspace of a Baire space is a Baire space. Complete metric spaces are
Baire spaces and application to a sequence of continuous real valued functions converging
point-wise to a limit function on a complete metric space. Hilbert spaces, examples of
Hilbert spaces such as `2 , L2 [−π, π], L2 (Rn ) (with no proofs). Inner product induced
by norm, Bessel’s inequality. Equivalence of complete orthonormal set and maximal
orthonormal set, orthogonal decomposition, Parseval’s identity. Existence of a maximal
orthonormal set, separability of Hilbert space.

Unit II: Normed Linear Spaces (15 Lectures)

Normed Linear spaces, Banach spaces. Examples of Normed linear spaces, Arzela-Ascoli
theorem, `p (1 ≤ p ≤ ∞) spaces are Banach spaces, Lp (µ)(1 ≤ p ≤ ∞) spaces: Holder’s
inequality, Minkowski’s inequality, Lp (µ)(1 ≤ p ≤ ∞) are Banach spaces. Quotient space
of a normed linear space. Finite dimensional normed linear spaces, Equivalent norms,
Riesz Lemma and application to infinite dimension of a normed linear spaces.

Unit III: Bounded Linear Transformations (15 Lectures)

Bounded linear transformations, Equivalent characterizations. The space B(X, Y ). Com-
pleteness of B(X, Y ) when Y is complete. Dual space of a normed linear space, Dual
space of `p (1 ≤ p < ∞), Riesz Representation theorem for Hilbert spaces. Dual of
Lp (µ)(1 ≤ p < ∞) spaces: Riesz-Representation theorem for Lp (µ)(1 ≤ p < ∞) spaces.
Separable spaces, examples of separable spaces such as `p (1 ≤ p < ∞). If the dual space
X 0 of X is separable, then X is separable.

Unit IV: Basic Theorems (15 Lectures)

Hahn-Banach theorem (Extension and Separation), applications of Hahn-Banach theo-
rem. Open mapping theorem, Closed graph theorem, Uniform boundedness principle and
application.
Recommended Text Books:

1. Andrew Browder, Mathematical Analysis, An Introduction, Springer International

Edition, 1996.

2. E. Keryszig, Introductory Functional Analysis with Applications, Wiely India, 1978.

3. B. V. Limaye, Functional Analysis, New Age International, 1996.

4. J. R. Munkres, Topology, Prentice Hall, 2000.

5. M. T. Nair, Functional Analysis, Prentice Hall, India

6. H. L. Royden, Real Analysis, Pearson, 4th edition, 2017.

7. G. F. Simmons, Introduction to Topology and Modern Analysis, Tata McGraw-Hill,

2004.
Contents

1 Baire spaces 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Few definitions with examples . . . . . . . . . . . . . . . 2
1.4 Baire Category Theorem . . . . . . . . . . . . . . . . . 3
1.5 Theorems on Baire spaces . . . . . . . . . . . . . . . . 4
1.6 G− delta set (Gδ set) . . . . . . . . . . . . . . . . . . . . 5
1.7 Applications . . . . . . . . . . . . . . . . . . . . . . . . 8
1.8 LET US SUM UP . . . . . . . . . . . . . . . . . . . . . . 10
1.9 Chapter End Exercise . . . . . . . . . . . . . . . . . . . 11

2 Hilbert spaces 13
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Definition of Hilbert Space . . . . . . . . . . . . . . . . . 14
2.4 Examples of Hilbert spaces . . . . . . . . . . . . . . . . 14
2.5 Parallelogram equality . . . . . . . . . . . . . . . . . . . 15
2.6 Few more inequalities . . . . . . . . . . . . . . . . . . . . 17
2.7 Theorems on Hilbert spaces . . . . . . . . . . . . . . . . 18
2.8 Orthogonal complements . . . . . . . . . . . . . . . . . . 19
2.9 Orthonormal sets . . . . . . . . . . . . . . . . . . . . . . 22
2.10 Complete orthonormal set . . . . . . . . . . . . . . . . . 25
2.11 Separable Hilbert space . . . . . . . . . . . . . . . . . . . 26
2.12 LET US SUM UP . . . . . . . . . . . . . . . . . . . . . . 27
2.13 Chapter End Exercise . . . . . . . . . . . . . . . . . . . 28

3 Normed Spaces 31
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3 Few definitions and examples . . . . . . . . . . . . . . . 32
3.4 Convergent Sequence and Cauchy Sequence in a Normed
Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.5 LET US SUM UP . . . . . . . . . . . . . . . . . . . . . . 56
3.6 Chapter End Exercise . . . . . . . . . . . . . . . . . . . 59

i
FUNCTIONAL ANALYSIS

4 Banach Space 63
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3 Few definitions and examples . . . . . . . . . . . . . . . 64
4.4 Equivalent Norms and Finite-Dimensional Spaces . . . . 76
4.5 Arzela-Ascoli theorem . . . . . . . . . . . . . . . . . . . 80
4.6 LET US SUM UP . . . . . . . . . . . . . . . . . . . . . . 84
4.7 Chapter End Exercise . . . . . . . . . . . . . . . . . . . 86

5 BOUNDED LINEAR TRANSFORMATIONS AND DUAL

SPACES 89
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.3 Definitions, notations, theorems . . . . . . . . . . . . . . 90
5.4 Separable spaces . . . . . . . . . . . . . . . . . . . . . . 99
5.5 Let us Sum Up . . . . . . . . . . . . . . . . . . . . . . . 100
5.6 CHAPTER END EXERCISES . . . . . . . . . . . . . . 101

6 Four Pillars of Functional Analysis 103

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 103
6.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.3 Few Definitions and Notations . . . . . . . . . . . . . . . 104
6.4 Hahn-Banach Theorem . . . . . . . . . . . . . . . . . . . 106
6.5 Uniform Boundedness Principle . . . . . . . . . . . . . . 107
6.6 Open Mapping Theorem . . . . . . . . . . . . . . . . . . 110
6.7 Closed Graph Theorem . . . . . . . . . . . . . . . . . . . 113
6.8 Applications of Hahn-Banach theorem . . . . . . . . . . 115
6.9 Chapter End Exercises . . . . . . . . . . . . . . . . . . . 118

ii
Chapter 1

Baire spaces

Chapter Structure
1.1 Introduction
1.2 Objectives
1.3 Few definitions with examples
1.4 Baire Category Theorem
1.5 Theorems on Baire spaces
1.6 G− delta set (Gδ set)
1.7 Applications
1.8 Let Us Sum Up
1.9 Chapter End Exercises

1.1 Introduction
In this chapter, we shall introduce definition and various examples of
Baire spaces. We shall also introduce Baire category theorem which has
application in Open Mapping Theorem, Uniform Boundedness Princi-
pal and in later chapter of Banach spaces. Various applications of Baire
spaces are there is analysis and branch of topology called Dimension
theory. The term “Baire spaces ” were coined by Nicolas Bourbaki. Gδ
sets are also introduced in this chapter.

1.2 Objectives
After going through this chapter you will be able to:
• Define Baire spaces.
• Identify which spaces are Baire spaces.
• Learn that open subspace of Baire space is Baire.
• Learn about Baire category theorem.

1
FUNCTIONAL ANALYSIS

• Learn to prove Hausdroff spaces which are compact or locally com-

pact are Baire spaces.
• Learn about Gδ sets.
• Application to a sequence of continuous real valued functions con-
verging point-wise to a limit function on complete metric space.

1.3 Few definitions with examples

Definition 1.1. Let X be a topological space and A ⊆ X be any

subset. The interior A◦ of A in the space X is defined as the union of
all open subsets of X which are contained in A.

Examples

1) Interior of [0, 1] = ([0, 1])◦ = (0, 1).

2) R◦ = R.

3) Q◦ in R = ϕ.

4) Q◦ in Q = Q.

Definition 1.2. A subset K of a topological space X is dense if K̄ =

X.

Example

• Set of rational numbers Q is dense in R.

Remark: Let A be a subset of X. Then A◦ = ϕ if and only if

X \A is dense in X.

Definition 1.3. A topological space X is called a Baire space if given

any countable collection {An } of closed sets of X, each having empty
interior in X then the union ∪ An also has empty interior in X.
n∈N

Examples

1. The space of rational numbers Q is not a Baire space.

2. The space of integers Z is a Baire space.

3. The space of all irrational numbers R\Q is a Baire space.

4. Any open subset of Baire space is Baire space.

5. Any complete metric space is a Baire space.

2
CHAPTER 1. BAIRE SPACES

6. Any compact Hausdroff space X is a Baire space.

Note: All the above examples are proved later.
Example 1. The space of rational numbers Q is not a Baire space.
Solution: Consider the field Q of rational numbers, a metric space
with the metric d(x, y) = |x − y| for all x, y ∈ Q. Now for all q ∈ Q,
{q} is closed . Also {q}◦ = ϕ (since for a < b ∈ R, (a, b) ∩ Q is infinite,
therefore (a, b) ∩ Q ⊊ {q}). So {q}q∈Q is a countableScollection of closed
subsets of Q, each having empty interior in Q. Now q∈QS{q} = Q. Thus
( q∈Q {q})◦ = (Q)◦ . But (Q)◦ = Q ̸= ϕ. Therefore ( q∈Q {q})◦ ̸= ϕ.
S
Hence by definition of Baire spaces, Q is not a Baire space.
Example 2. The space of integers Z is a Baire space.
Solution: Let An = {n} ⊆ Z.
So An is open in Z ( since Z is discrete metric space).
So {n}◦ = {n}.
Also An is closed in Z (singletons are closed set ).
So there is no closed set in Z with empty interior except for empty set.
Therefore Z is Baire space vaciously.

1.4 Baire Category Theorem

Theorem 1.4.1. Any non-empty complete metric space is a Baire

space.
Proof. Suppose {An | n ∈ N } is countable collection of closed subsets
of the space X such that each An has empty interior in X.
Then we show that ∪∞ n=1 An has empty interior in X.
Assume the contrary, i.e. there exists a non- empty open subset U of
X contained in ∪∞ n=1 An . Put Un = X \ An for each n ∈ N. Since each
An is closed thus each Un is open. Also since each An ◦ = ϕ, so each Un
is dense, so the intersection of any nonempty open subset of X with
each Un is nonempty. Thus we get U ∩ Un ̸= ϕ, for all n ∈ N.
So there exists an open ball V1 = Bd (x1 , r1 ) in X with r1 < 1 such that
V̄1 ⊂ U ∩ U1 . Now since U2 is dense so V1 ∩ U2 ̸= ϕ. So there exist V2
= Bd (x2 , r2 ) in X with r2 < 1/2 such that V̄2 ⊂ V1 ∩ U2 .

Continuing this way inductively we get open set Vn = Bd (xn , rn ),

for n ∈ N, such that rn < 1/n and V̄n ⊂ Vn−1 ∩ Un , for n ∈ N.
Thus we have nested sequence of closed sets V̄1 ⊇ V̄2 · · · in X with
diameter (V̄n ) < 1 /n. Since X is complete so by Cantor’s intersection
theorem ∩∞ ∞
n=1 V̄n ̸= ϕ. Fix any p ∈ ∩n=1 V̄n . Then p ∈ U ( since p ∈ V̄1

3
FUNCTIONAL ANALYSIS

⊂ U ). Now p ∈ V̄n ⊂ Un for all n ∈ N, so p ∈

/ An for each n ∈ N. This
∞ ∞
/ ∪n=1 An . But p ∈ U ⊂ ∪n=1 An . This is not possible.
implies that p ∈
So our assumption was wrong. Thus ∪∞ n=1 An has empty interior in X.
Hence proved.

Theorem 1.4.2. Let U be any non-empty open subset of a compact

Hausdroff space X and x ∈ U . Then there exist an open neighbourhood
V of x ∈ X such that x ∈ V̄ ⊂ U .

Theorem 1.4.3. Any compact Hausdroff space is a Baire space.

Proof. Let X be a compact Hausdroff space. Suppose {An | n ∈ N } is

countable collection of closed subsets of the space X such that each An
has empty interior in X.
Then in order to show that X is Baire space we will prove that ∪∞ n=1 An
has empty interior in X.
Assume the contrary, i.e. (∪∞ ◦
n=1 An ) ̸= ϕ.
Let V be any proper non-empty open set in X such that V ⊂ ∪∞ n=1 An .
Put Un = X \ An , for n ∈ N. As each An is closed, hence each Un is
an open subset of X. Also since each A◦n = ϕ, thus each Un is dense,
hence intersection of any nonempty open subset of X with each Un is
nonempty. Thus V ∩ U1 ̸= ϕ. So by Theorem 1.4.2, there exists open
set V1 such that V̄1 ⊂ V ∩ U1 .
Since U2 is dense, thus V1 ∩ U2 ̸= ϕ, so by Theorem 1.4.2, there exists
open set V2 in X such that V̄2 ⊂ V1 ∩ U2 .
Continuing this way inductively, we get open set Vn of X, for n ∈ N
such that V̄n ⊂ Vn−1 ∩ Un and V̄1 ⊃ V̄2 ⊃ · · · ⊃ V̄n ⊃ · · ·
Thus {V̄n }n∈N has finite intersection property. Since X is compact,
hence we get ∩∞ n=1 V̄n ̸= ϕ.
∞
Let p ∈ ∩n=1 V̄n . Then p ∈ V (since p ∈ V̄1 ⊂ V ).
Also p ∈V̄n ⊆ Un , for all n ∈ N. Now since p ∈ Un , for all n ∈ N.
Thus p ∈ / An , for all n ∈ N. This implies p ∈ / ∪∞
n=1 An . But p ∈ V ⊆
∞
∪n=1 An . This is not possible. Hence our assumption was wrong. Thus
(∪∞ ◦
n=1 An ) = ϕ. Hence proved.

1.5 Theorems on Baire spaces

Theorem 1.5.1. Any open subset of a Baire Space is a Baire space.

Proof. Let U be any non-empty proper open subset of a Baire space

X. Suppose {An | n ∈ N} is countable collection of closed subsets
of U such that each An has empty interior. Then we will prove that
(∪∞ ◦
n=1 An ) = ϕ.
Let Ān denote the closure of An in the space X. Since An is closed in

4
CHAPTER 1. BAIRE SPACES

U , so An = U ∩ Ān for all n ∈ N.

Claim :(Ān )◦ = ϕ, for n ∈ N.
If we prove this claim then since X is Baire space so we will get
(∪∞ ◦ ∞ ◦
n=1 Ān ) = ϕ. Hence (∪n=1 An ) = ϕ. Hence the result.
Let us proof the claim. Suppose for some m ∈ N, ( Ām )◦ ̸= ϕ. Let W
be any non-empty open subset of X contained in Ām , i.e. W ⊂ Ām ,
so W ∩ U ⊂ Ām ∩ U = Am . Since W ⊂ Ām , so W ∩ Am ̸= ϕ. Hence
W ∩ U ̸= ϕ (since Am ⊂ U ). Thus we get W ∩ U a non-empty, open
subset of X contained in Am . Hence A◦m ̸= ϕ, which is a contradiction,
as Am was choosen to be closed subset with empty interior. Hence our
assumption that for some m ∈ N, (Ām )◦ ̸= ϕ was wrong. Thus (Ān )◦
= ϕ, for all n ∈ N. This completes the proof.
Theorem 1.5.2. A topological space X is a Baire space if and only if
any countable intersection of open, dense subsets of X is a dense subset
of X.
Proof. Given X is a Baire space.
Let {Un }n∈N be any collection of open dense subsets of X.
To show that ∩n∈N Un is dense.
Let An = X\ Un , for each n ∈ N.
Since each Un is open, therefore each An is closed.
Also since each Un is dense implies that A◦n = ϕ for each n ∈ N.
So {An }n∈N is a countable collection of closed sets with each having
empty interior in X.
Since X is Baire space so we get (∪n∈N An )◦ = ϕ.
Thus ∩n∈N Un is dense in X.
Conversely: Given that {Un }n∈N is a countable collection of open dense
subsets of X such that ∩n∈N Un is dense in X.
To show that X is Baire space .
Choose An = X\ Un .
Then each An is closed (since each Un is open) and A◦n = ϕ (since each
Un is dense). Hence we get {An }n∈N countable collection of closed sets
with empty interior. Also since ∩n∈N Un is dense implies (∪∞ ◦
n=1 An) =
ϕ. So by definition of Baire space, X is a Baire space.

1.6 G− delta set (Gδ set)

Definition 1.4. Gδ set: Let X be a topological space. A subset S of X

is called a Gδ subset of X if it can be written as a countable intersection
of open subsets of X.
Theorem 1.6.1. If Y is a dense Gδ set in X and if X is a Baire space,
then Y is a Baire space in the subspaace topology.

5
FUNCTIONAL ANALYSIS

Proof. Given Y is a Gδ set of X. So Y = ∩n∈N Gn , where each set Gn

is open in X. Now let {Vm }m∈N be a countable collection of open dense
subsets of Y . Inorder to show Y is Baire space we will prove ∩m∈N Vm
is dense in Y .
Since each Vm is open in Y hence there exist open set Wm of X such
that Vm = Y ∩ Wm .
Claim 1: Each Wm is dense in X.
Let U be any nonempty open subset of X. Then U ∩ Y ̸= ϕ (since Y
is dense in X).
Now U ∩ Y is open in Y , so Vm ∩ (U ∩ Y ) ̸= ϕ. . . (i).
Now Vm ∩ (U ∩ Y ) = (Y ∩ Wm ) ∩ (U ∩ Y ) = (Wm ∩ U ) ∩ Y ⊆
Wn ∩ U . . . (ii)
Hence Wn ∩ U ̸= ϕ (from (i)and (ii)). Since U was arbitrary open set
in X. Hence Wm is dense in X, for each m ∈ N.
So we get {Wm }m∈N a collection in which each Wm is nonempty open
dense set in X. Therefore (∩n∈N Gn ) ∩ Wm is dense in X, for each
m ∈ N. Hence ∩m∈N ((∩n∈N Gn ) ∩ Wm ) is dense in X. . . . (iii)
Now ∩m∈N Vm = ∩m∈N (Y ∩ Wm ) =∩m∈N ((∩n∈N Gn ) ∩ Wm ). . . (iv).
So ∩m∈N Vm is dense in X (from (iii),(iv)).
Claim 2: ∩m∈N Vm is dense in Y.
Let U1 be a nonempty open set of Y . Then there exist an open set U ′
of X such that U1 = U ′ ∩ Y .
Now ∩m∈N Vm ∩ U ′ ̸= ϕ (since ∩m∈N Vm is dense in X.)
Thus we get U1 ∩ (∩m∈N Vm ) ̸= ϕ.
Since U1 was arbitrary open set in Y . Hence ∩m∈N Vm is dense in Y.
Hence proved.
Theorem 1.6.2. The space of irrational numbers is a Baire space.
Proof. We know that space of rational Q = ∪q∈Q {q}.
So the space of irrationals i.e. Qc = ∩q∈Q (R \ {q}).
Since each {q} is closed hence each R \ {q} is open.
Also since { q}◦ = ϕ, so R \ {q} is dense. Therefore Qc is countable
intersection of open sets in R, hence a Gδ set . . . (i)
We know that Qc is dense in R. . . (ii)
Now R is complete metric space hence a Baire space (by Theorem
1.4.1). . . (iii).
Thus from (i),(ii),(iii) and by Theorem 1.6.1, we get that space of of
irrational numbers is a Baire space.
Theorem 1.6.3. Set of rational numbers is not a Gδ subset of R.
Proof. Suppose there exists a countable collection {Un }n∈N of open set
in R such that Q = ∩∞ n=1 Un .
Put An = R \ Un , for all n ∈ N. This implies that An is closed.
Since Q ⊆ Un , for each n ∈ N. Thus A◦n = ϕ.
So R = (∪∞n=1 An ) ∪ (∪q∈R {q}).

6
CHAPTER 1. BAIRE SPACES

i.e. R is expressed as union of closed subset each having empty interior.

We know that R is complete metric space so R is a Baire space (by
theorem 1.4.1), so by definition of Baire space, R◦ = ((∪∞
n=1 An ) ∪ (∪q∈R
{q}))◦ = ϕ. But R◦ = R ̸= ϕ. Hence a contradiction.
So our assumption was wrong. Hence Q is not a Gδ subset of R.
Theorem 1.6.4. f : R −→ R be a function then the set of points at
which f is continuous is a Gδ set in R.
Proof. Let f : R −→ R be a function.
Let Sf be the set of all points of continuity of f .
To show that Sf is Gδ set in R.
Let Bn = {V ⊆ R, where V is open in R and diam(f (V )) < 1/n}.
Also let Vn = ∪V ∈Bn V .
Claim 1: Vn is open for each n ∈ N.
Let x ∈ Vn , this implies x ∈ V for some V ∈ Bn . Since V is open in
R so for some r > 0, Bd (x, r) ⊂ V ⊆ Vn . Thus for x ∈ Vn , we get
Bd (x, r) ⊂ Vn . Hence Vn is open for each n ∈ N.
Claim 2: Sf = ∩n∈N Vn
First we show that Sf ⊆ ∩n∈N Vn
Let x ∈ Sf , i.e. f is continuous at x. Then for each n ∈ N , we have
δn > 0, such that |f (x) − f (y)| < 1/2n, whenever | x − y| < δn .
Thus diam(f (Bd (x, δn ))) < 1/2n < 1/n.
Therefore x ∈ Bd (x, δn ) ⊆ Bn , for each n ∈ N.
Hence x ∈ Vn , for each n ∈ N.
Therefore x ∈ ∩n∈N Vn .
Since x was arbitrary element of Sf , therefore Sf ⊆ ∩n∈N Vn .
Conversely, let x ∈ ∩n∈N Vn .
Choose ϵ > 0 such that for m ∈ N, 1/m < ϵ. Now since x ∈ Vm so
there is an open set V ∈ Bm such that x ∈ V .
Since V is open, hence for some δ > 0, Bd (x, δ) ⊆ V .
Now for any y ∈ Bd (x, δ) ,we get y ∈ V and since diam (f (V )) <
1/m < ϵ, therefore |f (x) − f (y)| < 1/m < ϵ. Hence f is continuous at
x. Therefore x ∈ Sf . Since x ∈ ∩n∈N Vn was arbitrary element, hence
∩n∈N Vn ⊆ Sf . Therefore Sf = ∩n∈N Vn .
Theorem 1.6.5. If A is countable dense subset of R, then there is no
continuous function f : R −→ R which is continuous precisely at the
points of A.
Proof. First we will claim that any countable dense subset of R is not
a Gδ set.
If possible, let A is a Gδ set with A = ∩n∈N Un , where each Un is open
subset in R.
Since A is dense in R. So we have R = Ā.
Since A ⊆ Un , for each n ∈ N. Therefore Ā ⊆ Ūn ⊆ R.
Hence R = Ā ⊆ Ūn ⊆ R. Therefore R = Ūn .

7
FUNCTIONAL ANALYSIS

Hence for each n ∈ N, Un is dense in R.

Thus for each n ∈ N, (R \ Un )◦ = ϕ. Also R \ Un is closed subset for
each n ∈ N ( Since Un is open).
So R = (∪d∈A {d}) ∪ (∪n∈N (R\ Un ) ).
i.e. R is countable union of closed set, each having empty interior.
Since R is complete metric space hence Baire space (by Theorem 1.4.1).
So (R)◦ = ((∪d∈A {d}) ∪ (∪(R\ Un ) ))◦ = ϕ.
But R◦ = R ̸= ϕ.
Hence a contradiction. So our assumption was wrong .
Thus A cannot be a Gδ set . . . (i)
Suppose there exists a function which is precisely continuous at the
points of A.
Then A is Gδ set (by Theorem 1.6.4). . . (ii)
From (i) and (ii), we get a contradiction. So our assumption was wrong.
Hence no such function exists which is continuous at precisely at the
points of A.

Example 3. Is it possible to find a function f : R −→ (Y, d) which is

continuous precisely at rational numbers.
Solution: Let us suppose that there exists a function which is contin-
uous at rational numbers. Then Sf ={ x ∈ R | f is continuous on x}
= Q. We have proved that Sf is Gδ set (by Theorem 1.6.4). Hence we
will get Q as Gδ set, which is a contradiction as we have proved that Q
is not a Gδ set (by Theorem 1.6.3). Hence no such function is possible
which is continuous precisely at rational numbers.

1.7 Applications

Theorem 1.7.1. Let X = (X, d) be a complete metric space.

Let {fα }α∈N be a family of continuous function from X to R such that
for each x ∈ X ∃ a constant Mx ∈ R such that |fα (x)| ≤ Mx , ∀ α ∈ ∧.
Then ∃ a constant M ∈ R and a non-empty open set Bd (x0 , r) in X
such that |fα (x)| ≤ M , ∀ α ∈ ∧, ∀ x ∈ Bd (x0 , r).
Proof. Let An = { x ∈ X | |fα (x)| ≤ n, for all α ∈ Λ}, for each n ∈ N.
Then An = ∩α∈Λ fα−1 [0, n].
Since [0, n] is closed and each fα is continuous, hence fα−1 [0, n] is closed.
Since arbitrary intersection of closed set is closed. Thus An is closed in
X.
As X is complete, so X is a Baire space (by Theorem 1.4.1). Also since
∪∞ ◦
n=1 An = X, so there exist m ∈ N such that Am ̸= ϕ.
So there exists Bd (x0 , r) ⊆ Am .

8
CHAPTER 1. BAIRE SPACES

Let x ∈ Bd (x0 , r) ⊆ Am . Therefore |fα (x)| ≤ m.

Let us take m = M . Hence we get |fα (x)| ≤ M , for all x ∈ Bd (x0 , r),
for all α ∈ Λ. Hence proved.
Theorem 1.7.2. fn : X −→ R for ( n ∈ N) be continuous function
defined on a Baire space X converging pointwise to a limit function
f : X −→ R then the points of continuity of f contains a dense subset
of X.
Proof. Let us fix ϵ > 0 and n ∈ N and take An = {x ∈ X| | fk (x) −
fl (x) |≤ ϵ for all k, l ≥ n } = ∩∞ −1
n=1 (∩k,l≥n |fk − fl | [0, ϵ] )
As [0, ϵ] is closed and fk ’s are continuous functions so |fk − fl |−1 [0, ϵ] is
closed for all (k, l) ∈ N × N.
Since arbitrary intersection of closed subsets in X is again closed, hence
An (ϵ) is closed in X.
Claim: ∪∞ n=1 An (ϵ) = X.
Since An ⊆ X, for all n ∈ N. Thus ∪∞ n=1 An ⊆ X.
Enough to show X ⊆ ∪∞ n=1 n A .
Let x0 ∈ X be arbitrary, since fn (x0 ) −→ f (x0 ) as n −→ ∞, so for
given ϵ > 0, there exists n0 = n0 (x, ϵ) ∈ N such that |fk (x0 )−fl (x0 )| ≤ ϵ
for all k, l ≥ n0 . Therefore x0 ∈ An0 (ϵ). Thus x0 ∈ ∪∞ n=1 An (ϵ).
∞
Therefore X ⊆ ∪n=1 An (ϵ) (as x0 was an arbitrary element of X ).
Hence ∪∞ n=1 An (ϵ) = X . . . (i)
Now from (i) and since X is Baire space, so not every An (ϵ) can have
empty interior. So there exists m ∈ N such that A◦m ̸= ϕ.
Let U (ϵ) = ∪∞ ◦
n=1 An (ϵ) . . . (ii)
Then U (ϵ) is nonempty open subset of X.
Claim: U (ϵ) is dense in X.
Let V be any nonempty open subset of X.
So we will prove that U (ϵ) ∩ V ̸= ϕ.
Since V is open in X. Hence V is Baire (by Theorem 1.5.1). . . (iii)
Since X = ∪∞ ∞
n=1 An (ϵ). Thus V = ∪n=1 (V ∩ An (ϵ)).
So V is countable union of closed subsets. ( . . . (iv)
From (iii),(iv) we get that there exists m ∈ N such that (V ∩Am (ϵ))◦ ̸= ϕ.
But (V ∩ Am (ϵ))◦ = V ◦ ∩ A◦m (ϵ) = V ∩ A◦m (ϵ) (as V ◦ = V ).
So V ∩ A◦m (ϵ) ̸= ϕ.
Also V ∩ A◦m (ϵ)⊆ V ∩ U (ϵ)
So V ∩ U (ϵ) ̸= ϕ. Hence the claim.
So for ϵ > 0, U (ϵ) is nonempty open dense subset of X.
Thus {U (1/n)}n∈N is a countable collection of open dense subset of the
Baire space X. Hence ∩n∈N U (1/n) is dense in X.
Inorder to prove the Theorem it is enough to show that f is continuous
at every x ∈ ∩∞ n=1 U (1/n).
Let C = U (1) ∩ U (1/2) . . . U (1/n) . . .
Also let x0 be arbitrary element in C.
Then we will show that f is continuous at x0 .

9
FUNCTIONAL ANALYSIS

i.e. To show that for given ϵ > 0, we will find a neighbourhood B of x0

such that d(f (x), f (x0 )) < ϵ, for all x ∈ B.
First lets choose k ∈ N such that 1/k < ϵ/3.
Since x0 ∈ C. So x0 ∈ U (1/k). So there exists some N ∈ N such that
x0 ∈ (AN (1/k))◦ (from ii).
Given that fN is continuous, so continuity of fN enables us to choose
a neighbourhood B of x0 contained in AN (1/k) such that
d(fN (x), fN (x0 )) < ϵ/3, for all x ∈ B . . . (v)
Now B ⊂ AN (1/k), so we get d(fn (x), fN (x)) ≤ 1/k < ϵ/3, for all
x ∈ B. Letting n −→ ∞, we obtain that d(f (x), fN (x)) ≤ 1/k < ϵ/3,
for all x ∈ B . . . (vi)
In particular, since x0 ∈ B, so we have d(f (x0 ), fN (x0 )) < ϵ/3 . . . (vii)
Applying triangle inequality and using (v),(vi) and (vii) we obtain
d(f (x), f (x0 )) < ϵ for all x ∈ B.
Hence f is continuous at x0 ∈ C
Since x0 was arbitrary chosen, therefore f is continuous at all points of
C. Hence proved.

1.8 LET US SUM UP

• Given a countable collection of closed sets with each having empty

interior in a topological space X, if Union of those sets also has
empty interior, then such a topological space is called Baire space.

• The space of integers Z, irrational numbers R \Q are Baire spaces.

• The space of rational numbers is not a Baire space.

• Any complete metric space is a Baire space.

• Any compact Haudroff space is a Baire space.

• Any open subset of a Baire space is Baire space.

• The set of rationals Q is not a Gδ set.

• If f : (X, d) to R is any function, then set of points of continuity

of f is a Gδ set.

• There is no function f : R to R which is precisely continuous on

rationals.

• If Y is a dense Gδ set in X and if X is a Baire space, then Y is

a Baire space in the subspaace topology.

10
CHAPTER 1. BAIRE SPACES

• If X is a nonempty Baire space and (fn ) be a sequence of con-

tinuous maps converging point-wise to a limit function f . Then
points of contiunity of f is dense in X.

1.9 Chapter End Exercise

1. Let A1 ⊃ A2 ⊃ . . . be a nested sequence of nonempty T

closed sets
in the complete metric space X. If diamAn → 0, then An ̸= ϕ.

2. See Theorem 1.4.2.

3. Show that every locally compact Hausdroff space is a Baire space.

4. Let X equal the countable union ∪n∈N Un . Show that if X is

a nonempty Baire space then at least one of the sets Ūn has a
nonempty interior.

5. Show that if every point x of X has a neighbourhood that is a

Baire space, then X is a Baire space.

11
FUNCTIONAL ANALYSIS

12
Chapter 2

Hilbert spaces

Unit Structure:
2.1 Introduction
2.2 Objectives
2.3 Definition of Hilbert Space
2.4 Examples of Hilbert spaces
2.5 Parallelogram equality
2.6 Few more inequalities
2.7 Theorems on Hilbert spaces
2.8 Orthogonal complements
2.9 Orthonormal sets
2.10 Complete orthonormal set
2.11 Separable Hilbert space
2.12 Let Us Sum Up
2.13 Chapter End Exercises

2.1 Introduction
This chapter introduces Hilbert spaces which are special type of
Banach spaces. Hilbert spaces have additional structure which enable
us to know when two vectors are orthogonal. The whole theory was
initiated by the work of D.Hilbert(1912) on integral equation. The cur-
rently used geometrical notations and terminology is analogous to that
of the Euclidean geometry. These are most useful spaces in practical
applications of functional analysis.

2.2 Objectives

13
FUNCTIONAL ANALYSIS

After going through this chapter you will be able to:

• Define Hilbert spaces.
• Identify which spaces are Hilbert spaces.
• Understand and apply Parallelogram equality.
• Understand and apply Schwarz inequality.
• Understand various properties of Hilbert spaces.
• Understand Complete othonormal sets.
• Understand Bessel’s inequality and Parseval’s Indentity.
• Define Separable Hilbert space.

2.3 Definition of Hilbert Space

Definition 2.1. Inner product Space:

An Inner product space is a vector space X with an inner product
defined on X.
Definition 2.2. Complete inner product Space:
An inner product space X is said to be complete if every Cauchy se-
quence in X has a limit that is also in X,(i.e. every Cauchy sequence
in X is convergent in X.)
Definition 2.3. Hilbert space:
A Hilbert space is a complete inner product space (complete in the
metric defined by the inner product), where inner product on X is a
mapping of X × X into the scaler field K of X.
i.e. with every pair of vectors x, y there is a associated scalar which is
written as < x, y > and is called the inner product of x and y such that
for all the vectors x, y, z and for scalar α ∈ K, we have
1) < x, x > ≥ 0 , < x, x > = 0 if and only if x = 0.
2) < x, y > = < y, x >.
3) < αx, y > = α < x, y >.
4) < x + y, z > = < x, z > + < y, z >.

2.4 Examples of Hilbert spaces

1) The Euclidean space Rn is a Hilbert space with inner product

defined by
< x, y > = < (x1 , . . . , xn ), (y1 , . . . , yn ) >
= x1 y1 + x2 y2 + . . . + xn yn .
Where x = (x1 , . . . , xn ), y = (y1 , . . . , yn ) ∈ Rn .

14
CHAPTER 2. HILBERT SPACES

2) The space Cn is Hilbert space with inner product defined by

< x, y > = < (x1 , . . . , xn ), (y1 , P
. . . , yn ) >
= x1 y¯1 + x2 y¯2 + . . . + xn y¯n = ni=1 xi ȳi .
Where x = (x1 , . . . , xn ), y = (y1 , . . . , yn ), each xi , yi ∈ Cn , for all
1≤ i ≤ n.

3) The space l 2 isP

a Hilbert space with inner product space defined
by < x, y > = ∞ i=1 xi ȳi .
Where x = (x1 , . . . , xn , . . . ), y = (y1 , . . . , yn , . . . ) ∈ l 2 .

4) The space L2 (Rn ) space is a Hilbert space with inner product

space defined by
a) If functions
R are assumed to be real valued then
< x, y > = x(t)y(t)dt.
b) If functions
R are assumed to be complex valued then
< x, y > = x(t)y(t)dt.

5) The space L2 [−π, π] space is a Hilbert space with inner product

space defined by
a) If functions
R π are assumed to be real valued then
< x, y > = −π x(t)y(t)dt.
b) If function
R πare assumed to be complex valued then
< x, y > = −π x(t)y(t)dt.

2.5 Parallelogram equality

An inner product on X defines a norm on X given by
√
||x|| = < x, x > (2.1)
and a metric on X given by
√
d(x, y) = ||x − y|| = < x − y, x − y >. (2.2)

Hence inner product spaces are normed spaces.

The norm generalizes the elementary concept of the length of the vector.
From the elementary geometry we know that the sum of the squares
of the sides of a parallelogram equals the sum of the squares of its
diagonals. In any Hilbert space we have an analogue to it. There is an
important equality given by

||x + y||2 + ||x − y||2 = 2(||x||2 + ||y||2 ). (2.3)

This is called parallelogram equality.

15
FUNCTIONAL ANALYSIS

Parallelogram with sides x and y in the plane.

If a norm does not satisfy parallelogram law then that norm cannot be
obtained from inner product (from equation (2.1)) .
Hence not all normed spaces are inner product spaces.

Example 4. The normed space lp with p ̸= 2 is not an inner product

space, hence not a hilbert space .

Solution: The norm on lp with p ̸= 2 cannot be obtained from inner

product we prove this by showing that norm does not satisfy parallel-
ogram equality.
Let x = (1, 1, 0, 0, . . . ) ∈ lp .
||x|| = (|1|p + |1|p + |0|p + . . . )1/p = (1 + 1 + 0 + 0 . . . )1/p = 21/p . . . (i)
y = (1, −1, 0, 0, . . . ) ∈ lp .
||y|| = (|1|p + | − 1|p + |0|p + . . . )1/p = (1 + 1 + 0 + 0 . . . )1/p = 21/p . . . (ii)
x + y = (2, 0, 0, . . . )
||x + y|| = (|2|p + 0 + 0, . . . )1/p = (2p )1/p = 2. . . (iii)
similarly ||x − y|| = 2 . . . (iv)
For p = 2, let us check the parallelogram equality.
||x + y||2 + ||x − y||2 = 22 + 22 = 8 (from (iii) and (iv) ). . . (v)
Now 2 (||x||2 + ||y||2 )= 2(2+2) = 8 (from (i),(ii)). . . (vi)
Thus from (v) and (vi) ||x + y||2 + ||x − y||2 = 2 (||x||2 + ||y||2 ).
i.e. parallelogram equality holds.
For p ̸= 2, let us check whether parallelogram equality holds.
||x + y||2 + ||x − y||2 = 22 + 22 = 8 (from (iii) and (iv)). . . (vii)
Now 2 (||x||2 +||y||2 )= 2(22/p + 22/p ) ̸= 8 (from (i)(ii) for p ̸= 2). . . (viii)
Thus from (vii) and (viii) ||x + y||2 + ||x − y||2 ̸= 2 (||x||2 + ||y||2 ).
i.e. for p ̸= 2 parallelogram equality does not hold.
Hence for p ̸= 2, lp is a normed linear space but is not a Hilbert space.

We know that to each inner product there corresponds a norm which

is given by equation (2.1).
One can rediscover inner product from the corresponding norm by some
straight forward calculations.
For a real Inner product space:

16
CHAPTER 2. HILBERT SPACES

< x, y > = 41 (||x + y||2 - ||x − y||2 )

Calculation: Consider 41 (||x + y||2 - ||x − y||2 )

1
= 4
(< x + y, x + y > - < x − y, x − y > )
1
= 4
(< x, x > + < x, y > + < y, x > + < y, y > - < x, x > +< x, y >
+ < y, x > - < y, y > )
4
= 4
< x, y > = < x, y >.

For Complex valued Inner product space:

Re< x, y > = 14 (||x + y||2 - ||x − y||2 ).

Im< x, y > = 41 (||x + iy||2 - ||x − iy||2 ).

The above is called polarization identity.

2.6 Few more inequalities

Theorem 2.6.1. If x, y are any two vectors in Hilbert space, then

| < x, y > | ≤ ||x|| ||y|| (Schwarz inequality).

It can be easily proved that the inner product in a Hilbert space is

continuous by using Schwarz inequality.

Theorem 2.6.2. (Continuity of inner product) If in an inner product

space sequence xn → x and yn → y as n → ∞. Then < xn , yn > →
< x, y > as n → ∞.

Proof. Consider | < xn , yn > − < x, y > |

= | < xn , yn > − < xn , y > + < xn , y > − < x, y > |
≤ | < xn , yn > − < xn , y > | + | < xn , y > − < x, y > |
= | < xn , yn − y > | + | < xn − x, y > |
≤ ||xn ||||yn − y|| + ||xn − x||||y|| (by Schwarz inequality). . . (i)
As n → ∞, xn → x and yn → y (given)
Thus limn→∞ | < xn , yn > − < x, y > | ≤ ||xn || limn→∞ ||yn − y|| +
limn→∞ ||xn − x||||y|| = 0.
Therefore < xn , yn > → < x, y > as n → ∞. Hence proved.

The norm also satisfies one more inequality

||x + y|| ≤ ||x|| + ||y|| (Triangle inequality).

17
FUNCTIONAL ANALYSIS

2.7 Theorems on Hilbert spaces

Theorem 2.7.1. Let Y be a finite dimensional subspace of a Hilbert

space H then Y is complete.

Proof. Given that Y is a finite dimensional subspace of a Hilbert space

H. Let dim Y = n and {e1 , . . . , en } is a basis for Y .
Let (ym )m∈N be a Cauchy sequence in Y .
Now ym = α1m e1 + α2m e2 + . . . +αnm en , for each m ∈ N, where αim are
scalers. Since (ym )m∈N is Cauchy sequence, so by definition, for given
ϵ > 0, there exist N ∈ N such that ||ym − yr || < ϵ, for all m, r ≥ N .
So we have ϵ > ||ym −P yr || = ||(α1m − α1r )e1 + (α2m − α2r )e2 + . . . +(αnm −
r
αn )en ||. Hence Pϵ > || ni=1 (αim − αir )ei || . . . (1)
Define ||y||1 = ni=1 |αi |, for all y ∈ Y , where y = ni=1 αi ei .
P
We know that any two norms on finite dimensional space are equivalent.
Thus there exist c > 0 such that ||y|| Pn > c||y|| 1 . . . (2)
(2) we get, ϵ > || i=1 (αi − αir )ei || ≥ c || ni=1 (αim −
m
P
So from (1)and
αir )ei ||1 ≥ c ni=1 |αim − αir |.
P
For fix i, |αim − αir | < ϵ/c , for all m, r > N and for each αi ∈ H.
Thus for each i, the sequence of scalers (αim ) is a Cauchy sequence in
H. Since H is Hilbert space so H is complete.
Thus αim → αi as m → ∞, for 1≤ i ≤ n.
Consider y = α1 e1 + α2 e2 + . . . + αn en ∈ Y .
Claim: ym → y as m →P ∞.
Consider ||ym − y|| = || ni=1 (αim − αi )ei ||
Take K = max ||ei ||. So we get ||ym − y||≤ K ni=1 |αim − αi |.
P
As m → ∞, αim → αi hence ym → y. Thus (ym )m∈N is convergent in
Y , therefore Y is complete. Hence Proved.

Theorem 2.7.2. Let Y be a subspace of a Hilbert space H, then Y is

complete if and only if Y is closed in H.

Theorem 2.7.3. A closed convex subset D of a Hilbert Space H con-

tains a unique vector of smallest norm.

Proof. Given that D is convex, so whenever u, v ∈ D, (u + v)/2 ∈ D.

Let d = inf {||u|| : u ∈ D }.
So there exist a sequence {un } of vectors in D such that ||un || → d as
n → ∞. i.e. there exist n0 ∈ N such that
limn→∞ ||un || = d, for all n ≥ n0 . . . (1)
By convexity of D, (um + un ) /2 ∈ D and ||(um + un )/2|| ≥ d.
So ||(um + un )|| ≥ 2d.
Using the parallelogram law we get, ||um − un ||2 = 2||um ||2 + 2||un ||2 -
||um + un ||2 ≤ 2||um ||2 + 2||un ||2 -4d2 .

18
CHAPTER 2. HILBERT SPACES

Now from (1) we get, 2||um ||2 + 2||un ||2 - 4d2 → 2d2 + 2d2 - 4d2 = 0,
for all n, m ≥ n0 .
Hence we get ||um − un || → 0, all n, m ≥ n0 . Therefore (un )n∈N is a
Cauchy sequence in D. Since H is complete and D is closed hence D
is complete by Theorem 2.7.2.
So there exist u ∈ D such that limn→∞ un = u.
Now ||u|| = || limn→∞ un || = limn→∞ ||un || = d. Hence u is the vector
in D with smallest norm.
Claim: u is unique.
Suppose that u′ is another vector in D other than u which also has
norm d. Then (u + u′ )/2 is also in D.
Again by using parallelogram law, we get ||(u + u′ )/2||2 = ||u||2 /2 +
||u′ ||2 /2 - ||(u − u′ )/2||2 < ||u||2 /2 + ||u′ ||2 /2 = d2 .
Hence we get ||(u+u′ )/2|| < d, which is a contradiction to the definition
of d. Thus our assumption was wrong. Therefore u is unique. Hence
proved.

2.8 Orthogonal complements

Definition 2.4. An element x of an inner product space X is said to

be orthogonal to an element y ∈ X if < x, y > = 0.
We write x is orthogonal to y as follows x ⊥ y.
Note: If x ⊥ y then y ⊥ x.

Pythagorean Theorem If x ⊥ y in an inner product space X.

Then
||x + y||2 = ||x||2 + ||y||2 .

Consider ||x + y||2 = < x + y, x + y >

= < x, x > + < x, y > + < y, x > + < y, y >
= < x, x > + 0 + 0 + < y, y > (since x ⊥ y so < x, y > = 0, also
< y, x >= 0)
= ||x||2 + ||y||2 . Hence proved.

19
FUNCTIONAL ANALYSIS

Definition 2.5. A vector x is said to be orthogonal to a non-empty

set A (written x ⊥ A) if x ⊥ y for every y ∈ A.

Definition 2.6. The Orthogonal complement of A is denoted by A⊥

is the set of all vectors orthogonal to A.

Theorem 2.8.1. If H is a Hilbert space then

1) {0}⊥ = H, H ⊥ = {0}.

2) A ∩ A⊥ ⊆ {0}.

Theorem 2.8.2. If A1 ⊆ A2 then A⊥ ⊥

2 ⊆ A1 .

Proof. Let y ∈ A⊥
2 . This implies that < x, y > = 0 for all x ∈ A2 .
Thus < x, y > = 0 for all x ∈ A1 ( since A1 ⊆ A2 ).
So y ∈ A⊥ ⊥
1 . Hence A2 ⊆ A1 .
⊥

Theorem 2.8.3. A⊥ is closed linear subspace.

Theorem 2.8.4. Let A be a closed linear subspace of a Hilbert space

H. Let x be a vector not in A and let d be the distance from x to A.
Then there exists a unique vector y0 in A such that ||x − y0 || = d.

Proof. Let C = x + A. Then C is closed (since A is closed).

We first show that C is a convex set.
Let u, v ∈ C. Then u = x + u′ and v = x + v ′ , where u′ , v ′ ∈ A.
To show that (1 − t)u + tv ∈ C , where 0 ≤ t ≤ 1.
Consider (1 − t)u + tv = (1 − t)(x + u′ ) + t(x + v ′ ) = (1 − t)x + (1 − t)u′
+ tx + tv ′ = x + (1 − t)u′ + tv ′ .
Since u′ , v ′ ∈ A, so (1 − t)u′ and tv ′ ∈ A (since A is a subspace).
Hence (1 − t)u′ + tv ′ ∈ A.
So x + (1 − t)u′ + tv ′ ∈ x + A = C.
Therefore (1 − t)u + tv ∈ C. Hence C is convex.
Thus C is closed and convex set.
Let d be the distance from the origin to C.
So by Theorem 2.7.3, there exists a unique vector z0 in C such that
||z0 || = d. Since z0 ∈ C so z0 = x + (−y0 ), where y0 ∈ A.
The vector y0 = x - z0 is in A and ||x − y0 || = ||z0 || = d.
Claim: y0 is unique.
Suppose y1 is a vector in A such that y1 ̸= y0 and ||x − y1 || = d then
z1 = x − y1 is a vector in C such that z1 ̸= z0 and ||z1 || = d which
is contradiction to the uniqueness of z0 . Hence our assumption was
wrong. Thus we get a unique vector y0 in A such that ||x − y0 || = d.
Hence proved.

Theorem 2.8.5. If A is a proper closed linear subspace of a Hilbert

space H, then there exist a non-zero vector z0 ∈ H such that z0 ⊥ A.

20
CHAPTER 2. HILBERT SPACES

Proof. Let x be a vector not in A. Let d be the distance of x from A.

Then by Theorem 2.8.4 there exists a unique vector y0 in A such that
||x − y0 || = d.
We define z0 as follows z0 = x − y0 . So ||z0 || = ||x − y0 || = d.
Since d > 0, z0 is a non-zero vector.
Claim: z0 ⊥ A. It is enough to show that z0 ⊥ y, where y is arbitrary
element of A.
For any scaler α, consider ||z0 − αy|| = ||(x − y0 ) − αy||
= ||x − (y0 + αy)|| ≥ d = ||z0 ||.
Therefore ||z0 − αy|| ≥ ||z0 ||.
Thus ||z0 − αy||2 - ||z0 ||2 ≥ 0.
So ||z0 ||2 - ᾱ < z0 , y > - α < z0 , y > + |α|2 ||y||2 - ||z0 ||2 ≥ 0.
Thus - ᾱ < z0 , y > - α < z0 , y > + |α|2 ||y||2 ≥ 0.. . . (1)
Put α = β < z0 , y > for any arbitrary real number β. . . (2)
From (1) and (2) we get - β< z0 , y > < z0 , y > -β < z0 , y > < z0 , y >
+ β 2 | < z0 , y > |2 ||y||2 ≥ 0.
Thus −2β | < z0 , y > |2 + β 2 | < z0 , y > |2 ||y||2 ≥ 0.
We will now put a = | < z0 , y > |2 and b = ||y||2 to obtain
−2βa + β 2 ab ≥ 0.
Thus βa (βb − 2) ≥ 0, for real β . . . (3).
However if a > 0 then (3) is false for sufficiently small positive β.
Thus a = 0, so | < z0 , y > |2 = 0 , hence < z0 , y > = 0.
Therefore z0 ⊥ y. Since y is arbitrary element of A.
Therefore z0 ⊥ A.
Theorem 2.8.6. If A and B are closed linear subspace of a Hilbert
space H such that A is perpendicular to B i.e. A ⊥ B. Then the linear
subspace A + B is also closed.
Proof. Let z be the limit point of A + B.
It suffices to show that z is in A + B.
Let {zn }n∈N be a Cauchy sequence in A+B such that zn → z as n → ∞.
It is given that A ⊥ B, so < x, y > = 0 for all x ∈ A and y ∈ B.
Also A ∩ B = {0} and so we get A and B are disjoint.
Now each zn can be written uniquely in the form zn = xn + yn , where
xn ∈ A and yn ∈ B.
Now ||zm − zn ||2 = ||(xm + ym ) − (xn + yn )||2
= ||(xm − xn ) + (ym − yn )||2 .
Thus by pythagorean theorem we get ||zm − zn ||2 = ||xm − xn ||2 +
||ym − yn ||2 .
As {zn }n∈N is a Cauchy sequence so by definition, there exists n0 ∈ N
such that ||zm − zn || → 0, for all n, m ≥ n0 . Therefore ||xm − xn || →
0, for all n, m ≥ n0 , and ||ym − yn || → 0, for all n, m ≥ n0 .
Hence {xn }n ∈ N, {yn }n ∈ N are Cauchy sequence in A and B respec-
tively. Now A and B are closed and therefore complete by Theorem
2.7.2. So there exists vectors x and y in A and B such that xn → x

21
FUNCTIONAL ANALYSIS

and yn → y as n → ∞ .
Now x + y is in A + B.
Also z = lim n→∞ (xn + yn ) = limn→∞ (xn ) + limn→∞ (yn ) = x + y.
Thus z ∈ A + B. Hence A + B is closed.

Definition 2.7. Direct sum:

A vector space X is said to be the direct sum of two subspaces A and
B of X, if each x ∈ X can be represented uniquely as x = a + b, where
a ∈ A and b ∈ B, and A ∩ B = {0}.
It is written as
X = A⊕B
where B is called complement of A in X and vice-versa.

Theorem 2.8.7. If A is closed linear subspace of a Hilbert space H,

then H = A ⊕ A⊥ .

Proof. Given that A is closed. Also A⊥ is closed (by Theorem 2.8.3.)

Thus A + A⊥ is closed linear subspace of H (by Theorem 2.8.6.)
Claim: A + A⊥ = H.
Suppose A + A⊥ ̸= H. Then A + A⊥ ⊆ H. So there exists a nonzero
vector z0 such that z0 ⊥ (A + A⊥ ) (by Theorem 2.8.5).
Thus z0 ∈ ((A + A⊥ ))⊥ = A⊥ ∩A⊥⊥ which is a contradiction as A⊥ ∩A⊥⊥
= {0}. Hence H = A + A⊥ . . . (1)
Also since A ∩ A⊥ = {0} . . . (2)
So, from (1),(2) and by definition of direct sum H = A ⊕ A⊥ .

2.9 Orthonormal sets

Definition 2.8. An Orthonormal set in an Hilbert space X is a subset

of X in which each element has unit norm and elements are orthogonal
to each other.

Theorem 2.9.1. Let {e1 , e2 , . . . , en } be a finite orthonormal set in a

n
|< x, ei >|2 ≤ ||x||2 .
P
Hilbert space H. If x is any vector in H then,
i=1
n
P
Further x − < x, ei > ei ⊥ej , for each j .
i=1

n
< x, ei > ei ||2
P
Proof. 0 ≤ ||x −
i=1
n
P n
P
= <x− < x, ei > ei , x − < x, ei > ei >
i=1 i=1
n
P n
P n P
P n
= < x, x > - < x, ei > < x, ei > - < x, ei > < x, ei > +
i=1 i=1 i=1 j=1

22
CHAPTER 2. HILBERT SPACES

< x, ei > < x, ej > < ei , ej >.

P n n
P
= < x, x > - < x, ei > < x, ei > - < x, ei > < x, ei >
i=1 i=1
(since < ei , ej > = 0 if i ̸= j and < ei , ej > = 1 if i = j).
n
Thus 0 ≤ ||x||2 - | < x, ei > |2
P
i=1
n
| < x, ei > |2 ≤ ||x||2 .
P
Hence
i=1
n
P
Claim: x − < x, ei > ei ⊥ej , for each j.
i=1
n
P
Consider < x − < x, ei > ei , ej > =
i=1
n
P
< x, ej > - < x, ei >< ei , ej >
i=1
= < x, ej > − < x, ej > (since < ei , ej > = 0 if i ̸= j and < ei , ej > =
1 if i = j)
n
P
= 0. Hence we get x − < x, ei > ei ⊥ej for each j.
i=1

Theorem 2.9.2. If {ei } is an orthonormal set in a Hilbert space H

and if x is any vector in H then the set S={ei , < x, ei > ̸= 0} is either
empty or countable.

Proof. If x = 0
then < x, ei > =0 for each ei .
Therefore S is empty.
If x ̸= 0. For n ∈ N, we define Sn = {ej ∈ S : | < x, ej > |2 > ||x||2 /n}.
Claim 1: |Sn | ≤ n − 1.
Pn ei1 , ei2 , . . . , 2ein ∈ Sn2.
Suppose there exists
Now we know, i=1 | < x, ei > | ≤ ||x|| (by Theorem 2.9.1) . . . (1)
Also for each eij ∈ Sn we have | < x, ej > |2 > ||x||2 /n.
n
Thus we get n(||x||2 /n) < i=1 | < x, ei > |2 ≤ ||x||2 .
P
Hence ||x||2 < ||x||2 , a contradiction. Hence |Sn | ≤ n − 1.
Claim 2: S = ∪n∈N Sn .
Since Sn ⊆ S for all n ∈ N. So ∪n∈N Sn ⊆ S.
Now to show S ⊆ ∪n∈N Sn
Let e ∈ S. We will show that e ∈ Sn for some n.
Enough to prove that there exists n ∈ N such that
| < x, e > |2 > ||x||2 /n.
Assume that such n does not exist then we get,
| < x, e > |2 ≤ ||x||2 /n. As n → ∞, ||x||2 /n = 0.
Thus | < x, e > |2 = 0. Hence < x, e > = 0.
Therefore e ∈ / S, a contradiction.
Hence there exists n ∈ N such that | < x, e > |2 > ||x||2 /n.
Therefore e ∈ Sn .
Since e was arbitrary element of S. Thus S ⊆ Sn .

23
FUNCTIONAL ANALYSIS

Hence S ⊆ ∪n∈N Sn .
Therefore S = ∪n∈N Sn .
Now since each Sn is countable so S is countable. Hence proved.
Theorem 2.9.3. (Bessel’s
P inequality) If {ei } is an orthonormal set in
a Hilbert space H, then |< x, ei >| ≤ ||x||2 for every vector x ∈ H.
2

Proof. Consider S = { ei : < x, ei > ̸= 0}.

Then S is either empty or countable( by Theorem 2.9.2).
If S is empty.P Then < x,2ei > = 0 , for all ei .
Therefore | < x, ei > P
| = 0.
Since 0 ≤ ||x||2 . Hence | < x, ei > |2 ≤ ||x||2 .
Now assume S is non-empty then S can be finite or countably infinite.
If S isP finite, i.e. S = {e1P, e2 , . . . , en }.
Then | < x, ei > | = ni=1 | < x, ei > |2 ≤ ||x||2 (by Theorem 2.9.1)
2

If S is countably infinite .
Let usParrange the elements of S as Pfollows S = {e1 , e2 , . . . , en , . . . }.
∞ 2 n 2
Now
Pn i=1 | < x, ei > | = limn→∞ i=1 | < x, ei > | and
2 2
i=1 | < x, ei > | ≤ ||x|| (by Theorem P2.9.1).
∞ 2
So by Theory of absolute P convergence i=1 | < x, ei > | is convergent.
∞
< x, ei > |2 = i=1 | < x, ei P > |2 .
P
We therefore define |P
∞
2
< x, ei > | = limn→∞ ni=1 | < x, ei > |2
2
P
Now | < x, ei > | = i=1 | P
≤ limn→∞ ||x||2 = ||x||2 . Thus | < x, ei > |2 ≤ ||x||2 .
Hence proved.
Theorem 2.9.4. If {ei } is an orthonormal setPin a Hilbert space H
and if x is an arbitrary vector in H then, x − < x, ei > ei ⊥ej for
each j.
Proof. Let S = {ei :< x, ei > ̸= 0}.
Then S is either empty or countable (by Theorem 2.9.2)
When S is empty, P we get < x, ei > = 0, for each ei .
Consider < x − P < x, ei > ei , ej >
= < x, ej >
P- < x, ei >< ei , ej > = 0.
Thus x − < x, ei > ei ⊥ej for each j.
When S is P finite say S = {e1 , e2 , .P
. . , en }.
We define < x, ei > ei to be ni=1 < x, ei > ei . Then the result
holds by Theorem 2.9.1.
When S is countably infinite .
Pn the vectors in S in definite order say S = {e1 , e2 , . . . , en , . . . }.
Let us arrange
Put Sn = i=1 < x, ei > ei . P
2 m 2
Pm m > n, ||Sm − S2 n || = || i=n+1 < x, ei > ei || =
For
i=n+1 | < x, ei > | .
From Bessel’s inequality ∞ 2
P
n=1 | < x, en > | converges.
So {Sn }n∈N is cauchy sequence in H.
Since H isPcomplete, So Sn → S ′ as n → ∞.
Let S ′ = ∞ n=1 < x, en > en .

24
CHAPTER 2. HILBERT SPACES

x, ei > ei to be ∞
P P
We define <P i=1 < x, ei >
Pe∞
i.
Consider < x − < x, ei > ei , ej > = < x − i=1 < x, ei > ei , ej > =
< x − S , ej > = < x, ej > - < S ′ , ej > = < x, ej > - < limn→∞ Sn , ej >
′

= < x, ej >-P limn→∞ < Sn , ej > = < x, ej > - < x, ej > = 0.

Thus < x − <P x, ei > ei , ej > = 0.
Hence we get x - < x, ei > ei ⊥ ej for each j.

2.10 Complete orthonormal set

Definition 2.9. Complete orthonormal set:

An orthonormal set {ei } in a Hilbert space H is said to be complete
if it is maximal in its partially ordered set. i.e. if it is impossible to
adjoint a vector e to {ei } in such a way that {e, ei } is an orthonormal
set which properly contains ei .

Definition 2.10. Maximal orthonormal set:

An orthonormal set A in a Hilbert space H is maximal if the only point
in H which is orthogonal to every x ∈ A is 0, i.e. A cannot be extended
to a larger orthonormal set.

Note: Maximal orthonormal set and Complete orthonormal set are

equivalent. This is proved in Theorem 2.10.2

Theorem 2.10.1. Every non-zero Hilbert space H contains a complete

orthonormal set.

Proof. It is given that Hilbert space H is non- zero, so there exist a

non-zero vector x1 ∈ H. Let e1 = x1 / ||x1 ||
Now by using Gram Schmith process we can get orthonormal elements
Let A = collection of orthonormal set .
i.e. A = {Ei — where i ∈ ∧ and Ei is orthonormal set in H }
Then A is non-empty as {e1 } ∈ A. Now the elements of A can be
ordered in inclusion. Thus each chain in A has an upper bound given
by union of all elements in that chain.
Therefore by Zorn’s lemma, A has a maximal element say E.
To show that E is complete orthonormal set
Let us assume that there exists say {e, E} an orthonormnal set in H.
Then {e, E} belongs to A and E ⊆ {e, E} but this is a contradiction as
E is maximal element of A. Thus our assumption was wrong. Hence
E has to be complete orthonormal set. Hence proved.

Theorem 2.10.2. let H be a Hilbert space and {ei } be an orthonormal

set in H. Then the following conditions are equivalent:

25
FUNCTIONAL ANALYSIS

1) {ei } is complete.
2) x⊥ { ei } =⇒ x= 0. P
3)If x is an arbitrary vector in H then x = <P
x, ei > ei .
4) If x is an arbitrary vector in H then ||x||2 = |< x, ei >|2 .
Proof. a) To prove 1) implies 2)
Let us assume that 2) is not true. Then there exists a vector x ̸= 0
x
such that x ⊥ {ei }. Define e = ||x|| then, {ei , e} is an orthonormal set
which properly contains {ei }. This is a contradiction as it is given that
{ei } is complete. Hence our assumption was wrong.
Therefore x⊥ { ei } =⇒ x = 0.
b) To prove 2) implies P 3)
We know that x − < x, ei > ei is orthogonal
P to {ei } (by Theorem
2.9.4) So by P assumption 2) we get x − < x, e i > ei = 0.
Thus x = < x, ei > ei . Hence proved.
c) To prove 3) impliesP 4)
||x||2 =P<P
P
x, x > = < < x, ei > ei , < x, ei > ei > (by assumption
3))P = < x, ei > < x, ej > < ei , ej >
= P < x, ei > < x, ei > (as {ei } is orthonormal )
= | < x, ei > |2 .
d) To prove 4) implies 1)
Suppose that {ei } is not complete.
i.e. it is a proper subset of an orthnormal set say {ei , e}
Since e is orthogonal to all e′i s so we get < e,P ei > = 0. . . (I).
Now using the assumption 4) we get ||e||2 = | < e, ei > |2 . . . (II)
From (I) and (II) we get ||e||2 = 0. Thus e = 0 , which is contradiction,
as e is a unit vector. Hence our assumption is wrong.
Thus {ei } is complete.
e) To prove 2) implies 1)
Let us suppose that {ei } is not complete so {ei } is contained in an
orthonormal set say {e, ei }. Now e is orthogonal to every element in
{ei }. So by assumption in 2) we get e = 0, which is a contradiction.
Hence our assumption was wrong. Therefore {ei } is complete.
The equation, ||x||2 = |< x, ei >|2 is called Parseval’s Identity.
P

2.11 Separable Hilbert space

Definition 2.11. Dense set:

A subset A of a metric space X is said to be dense in X if Ā = X
Definition 2.12. Separable space:
A metric space X is said to be seprable if it has a countable subset
which is dense in X.

26
CHAPTER 2. HILBERT SPACES

Example 5. The space of real numbers is seprable

Solution: The set Q of all rational numbers is countable and is dense

in R.

Theorem 2.11.1. If H is separable, every orthonormal set in H is

countable.

Proof. Given that H is seperable.

Let A be any dense set in H and N be any orthonormal set in H.
To show that N is countable.
Let x, y be two distinct element in N so < x, y > = 0 . . . (1)
Now ||x − y||2 = < x − y, x − y > = < x, x > - < x, y > - < y, x > +
< y, y > = < x, x > + < y, y > (from (1)). √
= ||x||2 + ||y||2 = 2 . Hence ||x − y||2 = 2. Thus ||x − y||p= 2 .
Let Bx and By be the neighbourhood of x, y with radius 2/3.
Since A is dense in H, so Bx ∩ A ̸= ϕ and By ∩ A ̸= ϕ . . . (2)
Also Bx ∩ By = ϕ . . . (3)
If N is uncountable we would have uncountably many such pairwise
disjoint neighbourhood hence A would become uncountable. Thus we
will not be able to get dense set which is countable in H, which is
contradiction to the seperability of H. Therefore our assumption that
N is uncountable is wrong. Thus N is countable. Hence proved.

Theorem 2.11.2. Let Y be a subspace of a separble Hilbert space H

then Y is also seperable.

2.12 LET US SUM UP

• A space which is complete with respect to metric defined by inner

product is called Hilbert space.

• The equation ||x + y||2 + ||x − y||2 = 2(||x||2 + ||y||2 ) is called

parallelogram equality.

• If a norm does not satisfy Parallelogram law then that norm can-
not be obtained from inner product. So all normed spaces are not
Hilbert spaces.

• For x, y be any two vectors in a Hilbert space

| < x, y > | ≤ ||x|| ||y|| is called Schwarz iequality.

• For x, y be any two vectors in a Hilbert space

||x + y|| ≤ ||x|| + ||y|| is called triangle inequality.

27
FUNCTIONAL ANALYSIS

• If Y be a finite dimensional subspace of a Hilbert space H then

Y is complete.

• If Y be a subspace of a Hilbert space H, then Y is complete if

and only if Y is closed in H.

• There exists a unique vector of smallest norm in a closed convex

subset C of a Hilbert Space H.

• A⊥ is closed linear subspace.

• If M is a closed linear subspace of a Hilbert space H, x is a vector

not in M such that d be the distance from x to M . Then there
exists a unique vector y0 in M such that ||x − y0 || = d.

• If M is a proper closed linear subspace of a Hilbert space H, then

there exists a non-zero vector z0 ∈ H such that z0 ⊥ M .

• If A is closed linear subspace of a Hilbert space H, then H =

A ⊕ A⊥ .

• If {ei } is an orthonormal set in a Hilbert space H and if x is any

vector in H then the set S={ei , < x, ei ≯= 0} is either empty or
countable.
P
• If {ei } is an orthonormal set in a Hilbert space H then |<
2 2
x, ei >| ≤ ||x|| ∀ every vector x ∈ H.. This is called Bessel’s
inequality.

• An orthonormal set {ei } in H is said to be complete if it is max-

imal in its partially ordered set.

• Every non-zero Hilbert space H contains a complete orthonormal

set.

• H be a Hilbert space, {ei } be an orthonormal

P set in H and x is
an arbitrary vector in H then x = < x, ei > ei . This is called
Parseval’s Identity.

• A metric space X is said to be seprable if it has a countable subset

which is dense in X.

2.13 Chapter End Exercise

1. Any two norms on finite dimensional space are equivalent.

2. Prove Theorem 2.7.2

28
CHAPTER 2. HILBERT SPACES

3. Prove Theorem 2.8.1

4. Prove Theorem 2.8.3

5. State and Prove Schwarz inequality.

6. Prove Theorem 2.11.2

29
FUNCTIONAL ANALYSIS

30
Chapter 3

Normed Spaces

Unit Structure :
3.1 Introduction
3.2 Objective
3.3 Few definitions and examples
3.4 Convergent Sequence and Cauchy Sequence in a Normed Space
3.5 LET US SUM UP
3.6 Chapter End Exercise

3.1 Introduction
In this chapter, you will be introduced to the notion of a norm on
a vector space. The concept of norm of a vector is a generalization of
the notion of length. The definition of a normed space (a vector space
equipped with a norm on it) was given (independently) by S. Banach,
H. Hahn and N. Wiener in 1922. In one section of this chapter, you
will study the concept of normed spaces which is fundamental to the
development of the theory of Banach spaces. You will come to know
the relation between a normed space and a metric space. In another
section of this chapter, you will learn about convergent sequences and
Cauchy sequences in a normed space.

3.2 Objectives

The main objective of this chapter is to learn the normed spaces and
Cauchy sequences in it.
After going through this chapter you will be able to:
• Define a norm on a vector space.

31
FUNCTIONAL ANALYSIS

• Define a normed space.

• Learn how to check a vector space is a normed space under the given
norm.
• Learn Hölder’s inequality and Minkowski’s inequality (for finite sums
and for integrals).
• Prove that every normed space is a metric space.
• Show that a metric space need not be a normed space.
• Define a convergent sequence and Cauchy sequence in a normed space.
• Prove that the quotient space is a normed space under the given norm.
• Prove that every convergent sequence in a normed space is a Cauchy
sequence.

3.3 Few definitions and examples

You have learnt the definition of a norm (and properties of a norm)
on an inner product space V in your B.Sc. So you can guess the defi-
nition of norm on a vector space and get convinced with the following
definition.
Definition 3.1. Let V be a vector space over the field F(=R or C). A
norm ∥ ∥ on V is a real valued function (i.e. ∥ ∥ : V −→ R), satisfying
the following 4 properties/axioms:
(N 1) ∥x∥ ⩾ 0 ∀x ∈ V
(N 2) ∥x∥ = 0 if and only if x = 0v ∀x ∈ V
(N 3) ∥x + y∥ ⩽ ∥x∥ + ∥y∥ ∀x, y ∈ V
(N 4) ∥αx∥ = |α| ∥x∥ ∀x ∈ V and ∀α ∈ F
Now, you will come to know when a vector space is called, a normed
space.
Definition 3.2. A normed space V is a vector space over the field
F(=R or C) with a norm ∥ ∥ defined on it. In such a case, we say, (V ,
∥ ∥) is a normed space over F. Here, if F = R then V is called a real
normed space and if F = C then V is called a complex normed space.
Now, you will notice that examples of normed spaces are in abun-
dance.
Example 6. Show that the vector space R = {x|x ∈ R} over R is a
normed space under the norm ∥x∥ = |x| = absolute value of x ∈ R.

Solution: Using the properties of absolute value of a real number, we

have

32
CHAPTER 3. NORMED SPACES

(1) ∀x ∈ R, ∥x∥ = |x| ⩾ 0 and hence property (N 1) of norm is

satisfied.

(2) ∀x ∈ R, ∥x∥ = 0 ⇐⇒ |x| = 0 ⇐⇒ x = 0 and hence property

(N 2) of norm is satisfied.

(3) ∀x, y ∈ R, ∥x + y∥ = |x + y| ⩽ |x| + |y| = ∥x∥ + ∥y∥ and hence

property (N 3) of norm is satisfied.

(4) ∀α ∈ R and x ∈ R, ∥αx∥ = |α x| = |α| |x| = |α| ∥x∥ and hence

property (N 4) of norm is satisfied.

Thus, R is a normed space under defined norm.

Can you recall the name of the following inequality?

n
X X n 1/2 X n 1/2
2 2
| xi yi | ⩽ | xi | | yi |
i=1 i=1 i=1

for any complex (or real) numbers x1 , · · · , xn ; y1 , · · · , yn

It’s the Cauchy-Schwarz inequality in an inner product space which

you have learnt in your B.Sc. You will revisit this inequality in further
example(s).
Now, recall the vector space Cn = {(x1 , · · · , xn ) | xi ∈ C} over C,
where the vector addition and scalar multiplication is defined as follows,
respectively: for every scalar α ∈ C and (x1 , · · · , xn ), (y1 , · · · , yn ) ∈
Cn
(x1 , · · · , xn ) + (y1 , · · · , yn ) = (x1 + y1 , · · · , xn + yn )
α (x1 , · · · , xn ) = (αx1 , · · · , αxn )
Also, (x1 , · · · , xn )=(y1 , · · · , yn ) ⇐⇒ xi = yi ∀ i = 1, · · · , n

Example 7. Show that the vector space Cn = {(x1 , · · · , xn ) | xi ∈ C}

X n 1/2
2
over C is a normed space under the norm ∥x∥ = |xi | .
i=1

Solution: Using the properties of modulus of a complex number, we

have
n
X 1/2
n 2
(1) ∀x = (x1 , · · · , xn ) ∈ C , each |xi | ⩾ 0, and hence |xi | =
i=1
∥x∥ ⩾ 0. So property (N 1) of norm is satisfied.

(2) ∀x = (x1 , · · · , xn ) ∈ Cn ,
n
X
2
∥x∥ = 0 ⇐⇒ ∥x∥ = 0 ⇐⇒ |xi |2 = 0 ⇐⇒ |xi | = 0(1 ⩽ i ⩽ n)
i=1
⇐⇒ xi = 0 (1 ⩽ i ⩽ n) ⇐⇒ x = (x1 , · · · , xn ) = (0, · · · , 0) = 0
and hence property (N 2) of norm is satisfied.

33
FUNCTIONAL ANALYSIS

(3) ∀x = (x1 , · · · , xn ), y = (y1 , · · · , yn ) ∈ Cn ,

consider
n
X
∥x + y∥2 = |xi + yi |2
i=1
n
X
= |xi + yi | |xi + yi |
i=1
n
X
⩽ |xi + yi | (|xi | + |yi |) by triangle inequality
i=1
Xn n
X
= |xi + yi | |xi | + |xi + yi | |yi |
i=1 i=1
n
X n
X
= |(xi + yi ) xi | + |(xi + yi ) yi |
i=1 i=1
⩽ ∥x + y∥ ∥x∥ + ∥x + y∥ ∥y∥
by Cauchy − Schwarz inequality

= ∥x + y∥ ∥x∥ + ∥y∥

If ∥x + y∥ ̸= 0 then dividing both sides by ∥x + y∥, we get,

∥x + y∥ ⩽ ∥x∥ + ∥y∥. If ∥x + y∥ = 0 then the inequality ∥x + y∥ ⩽
∥x∥ + ∥y∥ is trivial, since both sides reduce to zero and hence in
any case, property (N 3) of norm is satisfied.

(4) ∀α ∈ C and x = (x1 , · · · , xn ) ∈ Cn ,

Xn X n n
X
2 2 2 2
2
∥αx∥ = |α xi | = |α| |xi | = |α| 2
|xi | = |α|2 ∥x∥2
i=1 i=1 i=1
Taking positive square root on both sides, we get, ∥α x∥ = |α| ∥x∥
and hence property (N 4) of norm is satisfied.

Thus, Cn is a normed space under defined norm.

Note: This norm is referred as Euclidean

p norm in Cn and is de-
noted as ∥x∥2 . If n = 1 then ∥x∥ = x21 + x22 where x = x1 + i x2 .
Clearly, the notion of norm is actually a generalisation of the concept
of (Euclidean) length.

Recall the inequality between arithmetic mean and geometric mean.

p 1
It states that ∀ α, β ∈ R+ , αβ ⩽ (α + β)
2

The generalisation of inequality between arithmetic mean and geo-

metric mean is given in the following lemma.

34
CHAPTER 3. NORMED SPACES

Lemma 3.3.1. Let 0 < λ < 1. Then αλ β 1−λ ⩽ λα + (1 − λ)β holds

good for every pair of non-negative real numbers α and β.

Proof. If either α = 0 or β = 0 then we are done. So, assume that

α > 0 and β > 0. For every non-negative real number t, define a
function ϕ as ϕ(t) = (1 − λ) + λt − tλ . For extreme value of ϕ, we must
have, ϕ′ (t) = 0 which implies λ (1 − tλ−1 ) = 0 and hence tλ−1 = 1 as
λ ̸= 0. Thus t = 1. So, at t = 1, we have extreme value of ϕ. It is easy
to see that (
< 0 if t < 1
ϕ′ (t)
> 0 if t > 1
This implies that ϕ attains minimum at t = 1. Thus, ϕ(t) ⩾ ϕ(1) which
α
gives (1 − λ) + λt ⩾ tλ . In this last inequality, put t = where α and
β
λ
α α
β are non-negative real numbers. Then (1 − λ) + λ ⩾ and
β β
on multiplying by β throughout, we are done.

Definition 3.3. Let p and q be non-negative extended real numbers.

For p ⩾ 1, q is said to be conjugate of p if
1 1
+ = 1, f or 1 < p < ∞
p q
q = ∞, f or p = 1
q = 1, f or p = ∞

Let 1 ⩽ p < ∞. To show that Cn is a normed space under the

X n 1/p
p
norm ∥x∥ = |xi | , we need a special inequality called as
i=1
Minkowiski’s inequality for finite sums. To prove this Minkowiski’s
inequality, we need another special inequality called as Hölder’s in-
equality for finite sums. You will see these inqualities as following two
Lemma’s.
Lemma 3.3.2. Hölder’s inequality for finite sums:
1 1
Let 1 < p, q < ∞ and + = 1. For any complex (or real) numbers
p q
x1 , · · · , xn ; y1 , · · · , yn
n
X n
X 1/p n
X 1/q
p q
| xi y i | ⩽ | xi | | yi |
i=1 i=1 i=1

Proof. If xi and yi are all zero then the result is obvious. So let atleast
one xi ̸= 0 and atleast one yi ̸= 0.
By Lemma 3.3.1, for α ⩾ 0 and β ⩾ 0, we have αλ β 1−λ ⩽ λα+(1−λ)β.

35
FUNCTIONAL ANALYSIS
p
1 | xi |
In this inequality, take λ = , α = and β =
( ni=1 | xi |p )1/p
P
q p
| yi | 1
Pn . Then we get, 1 − λ = and
( i=1 | yi |q )1/q q

| xi | | yi | 1 | xi |p 1 | yi |q
⩽ +
( i=1 | xi |p )1/p ( ni=1 | yi |q )1/q
Pn
p ( ni=1 | xi |p ) q ( ni=1 | yi |q )
P P P

∀ i = 1, · · · , n.
Adding all these inequalities, we get
Pn Pn Pn
i=1 | xi |P | yi | 1 i=1 | xi |p 1 i=1 | yi |
q
⩽ Pn +
( ni=1 | xi |p )1/p ( ni=1 | yi |q )1/q p ( i=1 | xi |p ) q ( ni=1 | yi |q )
P P

n
X n
X 1/p n
X 1/q
p q
Thus, | xi y i | ⩽ | xi | | yi | .
i=1 i=1 i=1

Note: If we take p = 2 in Hölder’s inequality then q = 2 and we

get the Cauchy-Schwarz inequality. So, Cauchy-Schwarz inequality is a
special case of Hölder’s inequality.

Lemma 3.3.3. Minkowski’s inequality for finite sums: Let 1 ⩽

p < ∞. For any complex (or real) numbers x1 , · · · , xn ; y1 , · · · , yn

n
X 1/p n
X 1/p n
X 1/p
p p p
| xi + y i | ⩽ | xi | + | yi |
i=1 i=1 i=1

Proof. If p = 1 then
X n 1/p n
X n
X
p
| xi + y i | = | xi + y i | ⩽ (| xi | + | yi |)
i=1 i=1 i=1
n
X n
X n
X 1/p n
X 1/p
p p
= | xi | + | yi |= | xi | + | yi |
i=1 i=1 i=1 i=1
1 1
Therefore the inequality holds for p = 1. So let p > 1 and = 1 −
q p
p
so that q > 1. Then p = (p − 1)q and p − = 1
q

36
CHAPTER 3. NORMED SPACES

Now by Hölder’s inequality in Lemma 3.3.2 (for finite sums), we have

n
X n
X n
1/p X q 1/q
p−1 p p−1
| xi | | xi + y i | ⩽ | xi | | xi + y i |
i=1 i=1 i=1
Xn 1/p Xn 1/q
p q(p−1)
= | xi | | xi + y i |
i=1 i=1
Xn 1/p X n 1/q
p p
= | xi | | xi + y i |
i=1 i=1
n
X n
1/p X 1/p p/q
p p
= | xi | | xi + y i |
i=1 i=1
−→ (∗)
n
X n
X
N ow | xi + yi |p = | xi + yi | · | xi + yi |p−1
i=1 i=1
n
X
⩽ (| xi | + | yi |) | xi + yi |p−1
i=1
n
X n
X
p−1
= | xi | | xi + yi | + | yi | | xi + yi |p−1
i=1 i=1

On using (∗) to the two summation terms on RHS, we get,

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

Hence,
n
X 1/p n
X 1/p n
X 1/p
p p p
| xi + y i | ⩽ | xi | + | yi | .
i=1 i=1 i=1

Example 8. Show that the vector space Cn = {(x1 , · · · , xn ) | xi ∈ C}

n
X 1/p
p
over C is a normed space under the norm ∥x∥ = |xi | where
i=1

37
FUNCTIONAL ANALYSIS

1 ⩽ p < ∞.

Solution: Using the properties of modulus of a complex number, we

have
n
X 1/p
n p
(1) ∀x = (x1 , · · · , xn ) ∈ C , each |xi | ⩾ 0, and hence |xi |
i=1
= ∥x∥ ⩾ 0. So property (N 1) of norm is satisfied.

(2) ∀x = (x1 , · · · , xn ) ∈ Cn ,
n
X 1/p
p
∥x∥ = 0 ⇐⇒ |xi | =0
i=1
n
X
⇐⇒ |xi |p = 0
i=1
⇐⇒ |xi | = 0 ∀ i = 1, · · · , n
⇐⇒ xi = 0 ∀ i = 1, · · · , n
⇐⇒ x = (x1 , · · · , xn ) = (0, · · · , 0) = 0

and hence property (N 2) of norm is satisfied.

(3) ∀x = (x1 , · · · , xn ), y = (y1 , · · · , yn ) ∈ Cn , by Minkowski’s in-

equality in Lemma 3.3.3 (for finite sums), we have,
X n 1/p X n 1/p X n 1/p
p p p
| xi + yi | ⩽ | xi | + | yi |
i=1 i=1 i=1

=⇒ ∥x + y∥ ⩽ ∥x∥ + ∥y∥ and hence property (N 3) of norm is

satisfied.

(4) ∀α ∈ C and x = (x1 , · · · , xn ) ∈ Cn ,

n
X 1/p
p
∥αx∥ = |α xi |
i=1
Xn 1/p
p
= (|α| |xi |)
i=1
n
X 1/p
p p
= |α| |xi |
i=1
n
X 1/p
p
= |α| |xi |
i=1
= |α| ∥x∥

and hence property (N 4) of norm is satisfied.

38
CHAPTER 3. NORMED SPACES

Thus, Cn is a normed space under defined norm.

Example 9. Show that the vector space Cn = {(x1 , · · · , xn ) | xi ∈ C}
over C is a normed space under the norm ∥x∥ = max{| x1 |, · · · , | xn |}.
(This norm is referred as ∥x∥∞ on Cn ).

Solution: Using the properties of of modulus of complex number, we

have
(1) ∀x = (x1 , · · · , xn ) ∈ Cn , each |xi | ⩾ 0, and hence max{| x1 |,
· · · , | xn |} = ∥x∥ ⩾ 0. So property (N 1) of norm is satisfied.
(2) ∀x = (x1 , · · · , xn ) ∈ Cn ,
∥x∥ = 0 ⇐⇒ max{| x1 |, | x2 |, · · · , | xn |} = 0
⇐⇒ |xi | = 0 ∀ i = 1, · · · , n
⇐⇒ xi = 0 ∀ i = 1, · · · , n
⇐⇒ x = (x1 , · · · , xn ) = (0, · · · , 0) = 0
and hence property (N 2) of norm is satisfied.
(3) ∀x = (x1 , · · · , xn ), y = (y1 , · · · , yn ) ∈ Cn ,
∥x + y∥ = max{| x1 + y1 |, · · · , | xn + yn |}
⩽ max{| x1 | + | y1 |, · · · , | xn | + | yn |}
⩽ max{| x1 |, · · · , | xn |} + max{| y1 |, · · · , | yn |}
= ∥x∥ + ∥y∥
and hence property (N 3) of norm is satisfied.
(4) ∀α ∈ C and x = (x1 , · · · , xn ) ∈ Cn ,

∥αx∥ = max{| αx1 |, · · · , | αxn |}

= max{| α || x1 |, · · · , | α || xn |}
=| α | max{| x1 |, · · · , | xn |}
= |α| ∥x∥
and hence property (N 4) of norm is satisfied.
Thus, Cn is a normed space under defined norm.
Remark 3.3.1. In view of norms defined on Cn in examples 8 and 9,
we have
 n 1/p
X
p
|xi | if 1 ⩽ p < ∞


∥x∥p = i=1

max{| x1 |, · · · , | xn |} if p = ∞


These norms are referred as p-norms on Cn .

39
FUNCTIONAL ANALYSIS

Remark 3.3.2. From examples 7, 8 and 9, it is clear that on a vector

space V, we can define more than one norm and accordingly different
normed spaces are obtained from same vector space V.

Recall a sequence space which is a vector space whose elements are

infinite sequences of real or complex numbers where the vector addition
and scalar multiplication, respectively are defined as follows:

{x1 , · · · , xn , · · · } + {y1 , · · · , yn , · · · } = {x1 + y1 , · · · , xn + yn , · · · }

α{x1 , · · · , xn , · · · } = {αx1 , · · · , αxn , · · · }
for every scalar α ∈ C or R
Also, {x1 , · · · , xn , · · · }={y1 , · · · , yn , · · · } if and only if xi = yi ∀i=
1, · · · , n, · · · .

Example 10. Let 1 ⩽ p < ∞. Show that the sequence space lp =

∞
X
p
{x1 , · · · , xn , · · · } | xi | < ∞ and xi ∈ C over C is a normed
i=1
∞
X 1/p
p
space under the norm ∥x∥ = |xi | .
i=1

Solution: Using the properties of modulus of a complex number, we

have
∞
X 1/p
p p
(1) ∀x = {x1 , · · · , xn , · · · } ∈ l , each |xi | ⩾ 0, and hence |xi |
i=1
= ∥x∥ ⩾ 0. So property (N 1) of norm is satisfied.

(2) ∀x = {x1 , · · · , xn , · · · } ∈ lp ,

∞
X 1/p
p
∥x∥ = 0 ⇐⇒ |xi | =0
i=1
∞
X
⇐⇒ |xi |p = 0
i=1
⇐⇒ |xi | = 0 ∀ i = 1, · · · , n, · · ·
⇐⇒ xi = 0 ∀ i = 1, · · · , n, · · ·
⇐⇒ x = {x1 , · · · , xn , · · · } = {0, · · · , 0, · · · } = 0

and hence property (N 2) of norm is satisfied.

(3) Let x = {x1 , · · · , xn , · · · }, y = {y1 , · · · , yn , · · · } ∈ lp .

By Minkowski’s inequality in Lemma 3.3.3 (for finite sums), we
have,

40
CHAPTER 3. NORMED SPACES

n
X 1/p n
X 1/p n
X 1/p
p p p
| xi + yi | ⩽ | xi | + | yi |
i=1 i=1 i=1
X∞ 1/p X∞ 1/p
p p
⩽ | xi | + | yi |
i=1 i=1
−→ (∗)

This is true for all n ∈ N. As the 2 series on RHS of (∗) converge,

it is clear that the series on LHS of (∗) must converge. Therefore,
X ∞ 1/p X ∞ 1/p X ∞ 1/p
p p p
| xi + y i | ⩽ | xi | + | yi |
i=1 i=1 i=1

=⇒ ∥x + y∥ ⩽ ∥x∥ + ∥y∥ and hence property (N 3) of norm is

satisfied.

(4) ∀α ∈ C and x = {x1 , · · · , xn , · · · } ∈ lp ,

∞
X 1/p
p
∥αx∥ = |α xi |
i=1
X∞ 1/p
p
= (|α| |xi |)
i=1
∞
X 1/p
p p
= |α| |xi |
i=1
∞
X 1/p
p
= |α| |xi |
i=1
= |α| ∥x∥

and hence property (N 4) of norm is satisfied.

Thus, lp is a normed space under defined norm.

p
Example 11. Show that the sequence space l = {x1 , · · · , xn , · · · } sup

{| x1 |, · · · , | xn |, · · · } < ∞ and xi ∈ C over C is a normed space
under the norm ∥x∥ = sup{| x1 |, · · · , | xn |, · · · }.
(This norm is denoted as ∥x∥∞ on lp ).

Solution: Using the properties of modulues of a complex number, we

have

41
FUNCTIONAL ANALYSIS

(1) ∀x = {x1 , · · · , xn , · · · } ∈ lp , each |xi | ⩾ 0, and hence sup{| x1 |

, · · · , | xn | · · · } = ∥x∥ ⩾ 0. So property (N 1) of norm is satisfied.

(2) ∀x = {x1 , · · · , xn , · · · } ∈ lp ,

∥x∥ = 0 ⇐⇒ sup{| x1 |, | x2 |, · · · , | xn |, · · · } = 0
⇐⇒ |xi | = 0 ∀ i = 1, · · · , n, · · ·
⇐⇒ xi = 0 ∀ i = 1, · · · , n, · · ·
⇐⇒ x = {x1 , · · · , xn · · · } = {0, · · · , 0, · · · } = 0

and hence property (N 2) of norm is satisfied.

(3) ∀x = {x1 , · · · , xn , · · · }, y = {y1 , · · · , yn , · · · } ∈ lp ,

∥x + y∥ = sup{| x1 + y1 |, · · · , | xn + yn |, · · · }
⩽ sup{| x1 | + | y1 |, · · · , | xn | + | yn |, · · · }
⩽ sup{| x1 |, · · · , | xn |, · · · } + sup{| y1 |, · · · , | yn |, · · · }
= ∥x∥ + ∥y∥

and hence property (N 3) of norm is satisfied.

(4) ∀α ∈ C and x = {x1 , · · · , xn , · · · } ∈ lp ,

∥αx∥ = sup{| αx1 |, · · · , | αxn |, · · · }

= sup{| α || x1 |, · · · , | α || xn |, · · · }
= |α| sup{| x1 |, · · · , | xn |, · · · }
= |α| ∥x∥

and hence property (N 4) of norm is satisfied.

Thus, lp is a normed space under defined norm.

Remark 3.3.3. In view of norms defined on lp in examples 10 and 11,
we have
 ∞ 1/p
X
p
|xi | if 1 ⩽ p < ∞


∥x∥p = i=1

sup{| x1 |, · · · , | xn |, · · · } if p = ∞


These norms are referred as p-norms on lp .

Recall the measure theory and Lebesgue integration that you learnt.
Consider a measure space (E, S, µ), where E is a measurable set, S is
p
Z ∞, let L (E, µ) =
a σ-algebra and µ is a measure on S. For 1 ⩽ p <
f : E −→ R f is a measurable function on E & | f (x) |p dx < ∞
E

42
CHAPTER 3. NORMED SPACES
Z Z
p
For convenience, (1) we denote | f (x) | dx < ∞ as | f |p < ∞.
E E
(2) we take E = [a, b] and µ as a Lebesgue measure. We write this
space as Lp (E).
It is easy to verify that Lp (E) is a vector space over R. i.e.,
f + g ∈ Lp (E) and αf ∈ Lp (E) ∀ f, g ∈ Lp (E), ∀α ∈ R. Note
that the elements of Lp (E) Zare equivalence classes of those functions,
where f is equivalent to g if | f − g |p = 0. i.e. the elements of Lp (E)
E
are equivalence classes of measurable functions which are equal almost
everywhere (a.e.).

To show that Lp (E) is a normed space under the norm ∥f ∥ =

Z 1/p
p
|f | where 1 ⩽ p < ∞, we need a special inequality called
E
as Minkowiski’s inequality for integrals. To prove this Minkowiski’s in-
equality, we need another special inequality called as Hölder’s inequality
for integrals. You will see these inqualities as following two Lemma’s.
Lemma 3.3.4. (Hölder’s inequality for integrals):
1 1
Let 1 < p, q < ∞ and + = 1. For f ∈ Lp (E) and g ∈ Lq (E)
p q
(where E is bounded closed interval in R),
Z Z 1/p Z 1/q
p q
|f g| ⩽ |f | |g|
E E E

Proof. The inequality is trivial ifZ either f = 0 a.e. or

Z g = 0 a.e. So let
f ̸= 0 a.e. and g ̸= 0 a.e. Then | f |p > 0 and | g |q > 0. By
E E
λ 1−λ
Lemma 3.3.1, for α ⩾ 0 and β ⩾ 0, we have α β ⩽λα + (1 − λ)β.
p
1 |f |
In this inequality, take λ = , α = R p )1/p
and β =
q
p ( E
| f |
|g| 1
R
q 1/q
. Then we get, 1 − λ = and
( E|g| ) q
|f | |g| 1 | f |p 1 | g |q
R R ⩽ R + R
( E | f |p )1/p ( E | g |q )1/q p ( E | f |p ) q ( E | g |q )
On integrating, we get
R R R
E
|f | |g| 1 E | f |p 1 E | g |q
R R ⩽ R + R
( E | f |p )1/p ( E | g |q )1/q p ( E | f |p ) q ( E | g |q )
Z Z 1/p Z 1/q
p q
Thus, |f g| ⩽ |f | |g|
E E E

Note: Recall that if f and g are Lebesgue measurableR over E then

the product f g is also Lebesgue measurable over E. If E | f |p < ∞

43
FUNCTIONAL ANALYSIS

R R
and E | g |q < ∞ then by Hölder’s inequality, E | f g |< ∞.
i.e. if f ∈ Lp and g ∈ Lq then f g ∈ L

Lemma 3.3.5. (Minkowski’s inequality for integrals):

Let 1 ⩽ p < ∞. If f ∈ Lp (E) and g ∈ Lp (E) (where E is bounded
closed interval in R), then
Z 1/p Z 1/p Z 1/p
p p p
|f +g | ⩽ |f | + |g|
E E E

Proof. It is easy to see that if f and g are Lebesgue measurable

Z over E
then f + g is also Lebesgue measurable over E. Also, if | f |p < ∞
Z Z E
p p
and | g | < ∞ then | f + g | < ∞.
E E
If p = 1 then as | f + g |⩽| f | + | g |, we are done. So let p > 1 and
1 1 p
= 1 − so that q > 1. Then p = (p − 1)q and p − = 1.
q Zp q
Zq
p−1
Clearly, |f +g | = | f + g |p < ∞.
ZE 1/q EZ 1/q
p−1 q p
Also, (| f + g | ) = |f +g | . −→ (1)
E E

Now by Hölder’s inequality in Lemma 3.3.4 (for integrals),

Z Z 1/p Z 1/q
p−1 p p−1 q
| f | | f +g | ⩽ |f | (| f +g | ) −→ (2)
E E E
and
Z Z 1/p Z 1/q
p−1 p p−1 q
| g | | f +g | ⩽ |g| (| f +g | ) −→ (3)
E E E

Z Now, consider
Z Z
p p−1
|f +g | = |f +g | | f + g |⩽ | f + g |p−1 (| f | + | g |)
E Z E Z EZ
p p−1
=⇒ |f +g | ⩽ |f | |f +g | + | g | | f + g |p−1
E E E

Using (1), (2), (3), we get,

Z Z 1/p Z 1/p Z 1/q
p p p p
| f +g | ⩽ |f | + |g| | f +g |
EZ E E ZE
If | f + g |p ̸= 0 then on dividing throughout by | f + g |p and
E E
1 1
using 1 − = , we get,
q p
Z 1/p Z 1/p Z 1/p
p p p
|f +g | ⩽ |f | + |g| .
Z E E E

If | f + g |p = 0 then there is nothing to prove and we are done.

E

44
CHAPTER 3. NORMED SPACES

Example 12. Let 1 ⩽ p < ∞ & E be a (bounded closed interval in R)

measurable set. Show that the vector space Lp (E) = {f : E −→ R|f
is Lebesgue measurable function on E and | f |p is Lebesgue integrable
Z 1/p
p
over E} over R is a normed space under norm ∥f ∥p = |f | .
E
(This norm is referred as p-norms on Lp (E)).

Solution: Using the properties of Lebesgue measurable and Lebesgue

integrable functions, we have

Z 1/p
p p
(1) ∀f ∈ L (E), ∥f ∥p = |f | ⩾ 0. So property (N 1) of
E
norm is satisfied.

1/p
p
R p
(2) Let f ∈ L (E). If f = 0 a.e. then ∥f ∥p = E
|0| = 0.
Conversely,

Z 1/p
p
∥f ∥p = 0 =⇒ |f | =0
E
p
=⇒ |f | = 0 a.e.
=⇒ |f | = 0 a.e.
=⇒ f = 0 a.e.

(note that the condition ∥f ∥ = 0 ⇐⇒ f = 0 is not satisfied)

If we do not distinguish between equivalent functions then the
property (N 2) of norm is satisfied.

(3) Let f, g ∈ Lp (E).

By Minkowski’s inequality 3.3.5 (for integrals), we have,

Z 1/p Z 1/p Z 1/p

p p p
|f +g | ⩽ |f | + |g|
E E E

Therefore, ∥f + g∥p ⩽ ∥f ∥p + ∥g∥p and hence property (N 3) of

norm is satisfied.

45
FUNCTIONAL ANALYSIS

(4) ∀α ∈ R and f ∈ Lp (E),

Z 1/p
p
∥αf ∥p = |α f |
E
Z 1/p
p
= (|α| |f |)
E
Z 1/p
p p
= |α| |f |
E
Z 1/p
p
= |α| |f |
E
= |α| ∥f ∥p

and hence property (N 4) of norm is satisfied.

Thus, Lp (E) is a normed space under defined norm.
Definition 3.4. Let E be a (bounded closed interval in R) measurable
set. A measurable function f : E −→ R is said to be essentially bounded
on E if there exists a finite real number m > 0 such that |f (x)| ⩽ m
a.e. on E. Here m is called essential (upper) bound for f . (i.e.f is
bounded except possibly on a set of measure zero)
If f has an essential upper bound then least upper bound exists.
The least such bound is denoted by ess sup |f |. If f does not have any
essential bound, then its essential supremum is defined to be ∞.

We define L∞ (E) = the class of all those measurable functions de-

fined on E which are essentially bounded on E. The elements of L∞ (E)
are equivalence classes of f . It is easy to verify that L∞ (E) is a vector
space over R.
Example 13. Let E be a (bounded closed interval in R) measurable
set. Show that the vector space L∞ (E) = {f : E −→ R|f is a mea-
surbale function on E and ess sup|f | < ∞} over R is a normed space
under norm ∥f ∥∞ = ess sup |f (x)| = inf {m > 0 | |f (x)| ⩽ m a.e. on
E
E}.
Hint.(Check)
(1) ∀f ∈ L∞ (E), ∥f ∥p ⩾ 0.

(2) ∀f ∈ L∞ (E), ∥f ∥∞ = 0 if and only if f = 0 a.e.

(3) Let f, g ∈ L∞ (E). Clearly, |f | ⩽ ∥f ∥∞ a.e. and |g| ⩽ ∥g∥∞ a.e.

As |f + g| ⩽ |f | + |g| ⩽ ∥f ∥∞ + ∥g∥∞ a.e., it follows that,
∥f + g∥∞ ⩽ ∥f ∥∞ + ∥g∥∞

(4) ∀α ∈ R and f ∈ L∞ (E), ∥αf ∥∞ = |α| ∥f ∥∞

46
CHAPTER 3. NORMED SPACES

Remark 3.3.4. In view of norms defined on Lp (E) in examples 12 and

13, we have
 Z 1/p
 p
|f | if 1 ⩽ p < ∞

∥f ∥p = E
ess sup|f |

if p = ∞
Recall that the set C[0, 1] = {f : [0, 1] −→ R(or C)| f is a con-
tinuous function} is a vector space over R or C under the opera-
tions (f + g)(x) = f (x) + g(x) ∀ x ∈ [0, 1]

(α f )(x) = α f (x) ∀ x ∈ [0, 1]

Example 14. Show that the vector space C[0, 1] = {f : [0, 1] −→ R|

f is a Zcontinuous function} over R is a normed space under norm
1
∥f ∥ = |f (t)|dt.
0

Solution: Using the properties of Riemann integration and absolute

value of a real number , we have
Z 1
(1) Let f ∈ C[0, 1]. As |f (t)| ⩾ 0 ∀ t ∈ [0, 1], we have |f (t)|dt ⩾ 0
0
and hence ∥f ∥ ⩾ 0. So property (N 1) of norm is satisfied.
(2) Let f ∈ C[0, 1].
Z 1
∥f ∥ = 0 ⇐⇒ |f (t)|dt = 0
0
⇐⇒ |f (t)| = 0 ∀ t ∈ [0, 1]
⇐⇒ f (t) = 0 ∀ t ∈ [0, 1]
⇐⇒ f = 0 (zero f unction)
and hence property (N 2) of norm is satisfied.
(3) Let f, g ∈ C[0, 1].
Z 1
∥f + g∥ = |(f + g)(t)|dt
0
Z 1
= |f (t) + g(t)|dt
0
Z 1 Z 1
⩽ |f (t)|dt + |g(t)|dt
0 0
= ∥f ∥ + ∥g∥

Therefore, ∥f + g∥ ⩽ ∥f ∥ + ∥g∥ and hence property (N 3) of

norm is satisfied.

47
FUNCTIONAL ANALYSIS

(4) ∀α ∈ R and f ∈ C[0, 1],

Z 1
∥αf ∥ = |(αf )(t)|dt
0
Z 1
= |α f (t)|dt
0
Z 1
= |α| |f (t)|dt
0
Z 1
= |α| |f (t)|dt
0
= |α| ∥f ∥
and hence property (N 4) of norm is satisfied.
Thus, C[0, 1] is a normed space under defined norm.
Example 15. Show that the vector space C[0, 1] = {f : [0, 1] −→
R(or C)| f is a continuous function} over R(or C) is a normed space
under norm ∥f ∥ = sup {|f (x)|}.
x∈[0,1]
(This norm is referred as ∥f ∥∞ or sup norm on C[0, 1]).

Solution: Using the properties of supremum and absolute value (or

modulus in C) of a real number , we have
(1) Let f ∈ C[0, 1]. As |f (x)| ⩾ 0 ∀ x ∈ [0, 1], we have sup {|f (x)|} ⩾
x∈[0,1]
0 and hence ∥f ∥ ⩾ 0. So property (N 1) of norm is satisfied.
(2) Let f ∈ C[0, 1].
∥f ∥ = 0 ⇐⇒ sup {|f (x)|} = 0
x∈[0,1]

⇐⇒ |f (x)| = 0 ∀ x ∈ [0, 1]
⇐⇒ f (x) = 0 ∀ x ∈ [0, 1]
⇐⇒ f = 0 (zero f unction)
and hence property (N 2) of norm is satisfied.
(3) Let f, g ∈ C[0, 1].
∥f + g∥ = sup {|(f + g)(x)|}
x∈[0,1]

= sup {|(f (x) + g(x)|}

x∈[0,1]

⩽ sup {|(f (x)| + |g(x)|}

x∈[0,1]

= sup {|f (x)| + sup {|g(x)|}

x∈[0,1] x∈[0,1]

= ∥f ∥ + ∥g∥

48
CHAPTER 3. NORMED SPACES

Therefore, ∥f + g∥ ⩽ ∥f ∥ + ∥g∥ and hence property (N 3) of

norm is satisfied.

(4) ∀α ∈ R(or C) and f ∈ C[0, 1],

∥αf ∥ = sup {|(αf )(x)|}

x∈[0,1]

= sup {|α f (x)|}

x∈[0,1]

= sup {|α| |f (x)|}

x∈[0,1]

= |α| sup {|f (x)|}

x∈[0,1]

= |α| ∥f ∥

and hence property (N 4) of norm is satisfied.

Thus, C[0, 1] is a normed space under defined norm.

Remark 3.3.5. 1. You can mimic Examples 14 and 15 for the vec-
tor space C[a, b] over R or C.

2. In view of norms on C[a, b] defined in examples 14 and 15, we have

 Z 1/p
b
p


 |f (t)| dt if 1 ⩽ p < ∞
∥f ∥p = a
 sup {|f (t)|} if p = ∞


t∈[a,b]

3. Note that the vector space C[a, b] is a particular case of the vector
space C(X) where X is a compact space. So you can have similar
versions of Examples 14 and 15 for the vector space C(X) over R
or C.
Definition 3.5. A normed space is called finite dimensional if the un-
derlying vector space is finite dimensional, otherwise it is called infinite
dimensional.
Remark 3.3.6. The normed spaces in examples 6, 7, 8, 9 are finite
dimensional and the normed spaces in examples 10, 11, 12, 14 are
infinite dimensional.
In a given normed space (V, ∥ ∥), for x, y ∈ V , x − y ∈ V . (Why?).
This suggests us to give following definition.
Definition 3.6. In a given normed space (V, ∥ ∥), a function d : V ×
V −→ R+ defined as d(x, y) = ∥x − y∥ is called the distance from x to
y, where x, y ∈ V . Here d is referred as distance function on V .

49
FUNCTIONAL ANALYSIS

Clearly, ∥x∥ is the distance from the zero vector to vector x ∈ V .

Recall the definition of a metric space. In the next results, you will see
the relation between normed space and metric space.

Theorem 3.3.1. Every normed space is a metric space with respect to

the distance function

Proof. Let (V, ∥ ∥) be a normed space and d be distance function on

V . So for x, y ∈ V , d(x, y) = ∥x − y∥. Let x, y, z ∈ V

(1) By property (N 1) of norm, ∥x − y∥ ⩾ 0 and hence d(x, y) ⩾ 0.

(2) By property (N 2) of norm,

∥x − y∥ = 0 ⇐⇒ x − y = 0
⇐⇒ x = y
T hus, d(x, y) = 0 ⇐⇒ x = y

(3) By property (N 4) of norm, ∥x−y∥ = ∥−(y−x)∥ = |−1| ∥y−x∥ =

∥y − x∥. Thus, d(x, y) = d(y, x).

(4) By property (N 3) of norm,

∥x − y∥ = ∥(x − z) + (z − y)∥ ⩽ ∥x − z∥ + ∥z − y∥ and thus
d(x, y) ⩽ d(x, z) + d(z, y)

Therefore, all the conditions of the metric are satisfied by d. So d is a

metric on V and hence (V, d) is a metric space.

The metric defined in this way is called as the natural metric induced
by the norm.

Remark 3.3.7. You will come to know from the following example
that the converse of Theorem 3.3.1 need not be true.

Example 16. Let 0 < p < 1. Consider the sequence space lp =

∞
X
p
x = {x1 , · · · , xn , · · · } xi ∈ R and |xi | < ∞ over R. For
i=1
x = {x1 , · · · , xn , · · · } and y = {y1 , · · · , yn , · · · } in lp , define a function
X∞
p p +
d : l × l −→ R as d(x, y) = |xk − yk |p . Then (lp , d) is a metric
k=1
space (Check!) but not a normed space as property (N 4) of norm is
not satisfied as shown below:
for z = {0, 1, 0, · · · } ∈ lp and α = 2 ∈ R, we have,
∥αz∥ = ∥αz−0∥ = d(αz, 0) = |0−0|p +|α−0|p +|0−0|p +· · · = |α|p = αp
& |α|∥z∥ = α∥z−0∥ = α d(z, 0) = α (|0−0|p +|1−0|p +|0−0|p +· · · ) = α

50
CHAPTER 3. NORMED SPACES

3.4 Convergent Sequence and Cauchy

Sequence in a Normed Space
You are already familiar with the definition of convergence of se-
quences in a set of points and its related results. Now, you will learn
the concepts of convergent sequences and Cauchy sequences in a normed
space to obtain analogous results, which you learnt in your B.Sc.
Definition 3.7. Let (V, ∥ ∥) be a normed space. A sequence {xn } of
vectors in V is convergent to x ∈ V if ∀ ϵ > 0 ∃ n0 ∈ N such that
∀ n ⩾ no we have ∥xn − x∥ < ϵ. (or equivalently, if ∥xn − x∥ converges
to 0 as n −→ ∞). Here, we say, x is the limit of the sequence {xn }
or {xn } is a convergent sequence, converging to x. In this case, we
write, lim xn = x or (xn −→ x as n −→ ∞) or lim ∥xn − x∥ = 0 or
n−→∞ n−→∞
(∥xn − x∥ −→ 0 as n −→ ∞)
In the following result, you will notice that, in a normed space, a
sequence can have at most one limit.
Theorem 3.4.1. The limit of a convergent sequence in a normed space
is unique.

Proof. Consider a convergent sequence {xn } in normed space (V, ∥ ∥).

Assume that {xn } converges to x ∈ V and {xn } converges to y ∈ V .
Then as n −→ ∞, we have, ∥xn −x∥ −→ 0 and ∥xn −y∥ −→ 0. Clearly,
∥xn − x∥ = ∥x − xn ∥.
Consider

∥x − y∥ = ∥(x − xn ) + (xn − y)∥

⩽ ∥x − xn ∥ + ∥xn − y∥ by property (N 3) of norm
⩽ 0 as n −→ ∞
T hus, ∥x − y∥ = 0 by property (N 1) of norm
∴x−y =0 by property (N 2) of norm

∴ x = y and we are done.

Example 17. In a normed space (V, ∥ ∥), show that ∥x∥ − ∥y∥ ⩽
∥x − y∥ ∀ x, y ∈ V .

Solution: Using property (N 3) of norm, we have, ∥x∥ = ∥(x−y)+y∥ ⩽

∥x − y∥ + ∥y∥. So, ∥x∥ − ∥y∥ ⩽ ∥x − y∥ −→ (1)
Similarly, ∥y∥ − ∥x∥ ⩽ ∥x − y∥ as ∥y − x∥ = ∥x − y∥

51
FUNCTIONAL ANALYSIS

So, −(∥x∥ − ∥y∥) ⩽ ∥x − y∥ −→ (2)

From (1) and (2), we get, ∥x∥ − ∥y∥ ⩽ ∥x − y∥

Now, you will learn the definition of a continuous function on normed

spaces and then prove that norm is a continuous function on R.

Definition 3.8. Consider the normed spaces (V, ∥ ∥V ) and (W, ∥ ∥W ).

The function f : V −→ W is continuous at x0 ∈ V if ∀ ϵ > 0 ∃ δ > 0
such that ∥x − x0 ∥V < δ =⇒ ∥f (x) − f (x0 )∥W < ϵ
Equivalently, we write, as n −→ ∞, xn −→ xo =⇒ f (xn ) −→ f (xo )
i.e. for every sequence {xn } in V converging to x0 ∈ V , the sequence
{f (xn )} in W converges to f (x0 ) ∈ W .
Here, the notations ∥ ∥V and ∥ ∥W mean the norms in normed spaces
V and W , respectively.

Theorem 3.4.2. Let (V, ∥ ∥) be a normed space. Define a function

f : V −→ R as f (x) = ∥x∥. Then the norm ∥ ∥ on V is a real valued
continuous function.

Proof. Let x0 ∈ V . Consider a sequence {xn } in V such that as n −→

∞, xn −→ xo . i.e. as n −→ ∞, ∥xn − x0 ∥ −→ 0. Then using Example
17, we get, f (xn ) − f (x0 ) = ∥xn ∥ − ∥x0 ∥ ⩽ ∥xn − x0 ∥ −→ 0 as
n −→ ∞ and thus as n −→ ∞, we have f (xn ) −→ f (x0 ). Therefore, for
every sequence {xn } in V converging to x0 ∈ V , the sequence {f (xn )}
in R converges to f (x0 ) ∈ R. Hence, the norm ∥ ∥ on V is a real valued
continuous function.

Definition 3.9. A subspace M of a normed space V is a subspace of

V considered as a vector space, with the norm obtained by restricting
the norm on V to the subset M . This norm on M is said to be induced
by the norm on V .

Note that, if M is closed in a normed space V , then M is called a

closed subspace of V .

Recall the quotient space (or factor space). Let M be a subspace of

a vector space V . The coset of an element x ∈ V with respect to M is
x + M = {x + m | m ∈ M }. Under the following algebraic operations
(x + M ) + (x′ + M ) = (x + x′ ) + M and α (x + M ) = αx + M , the
quotient space of V by M , denoted by V /M is a vector space. Note
that, x + M = M if and only if x ∈ M .

Given a normed space, you will learn how to form a new normed
space.

52
CHAPTER 3. NORMED SPACES

Theorem 3.4.3. Let M be a closed subspace of a normed space (V, ∥ ∥).

For each coset x+M in quotient space V /M , define ∥x+M ∥ = inf {∥x+
m∥ | m ∈ M }. Then V /M is a normed space under the norm ∥x + M ∥.
Proof. Using the properties of infimum and properties in normed space
(V, ∥ ∥), we have
(1) As ∥x + m∥ is a non-negative real number and every set of non-
negative real numbers is bounded below, it follows that inf {∥x +
m∥ | m ∈ M } exists and is non-negative. i.e.∥x + M ∥ ⩾ 0
∀ x + M ∈ V /M . So property (N 1) of norm is satisfied.
(2) Let x + M = M (zero element of V /M ). Then x ∈ M . So

∥x + M ∥ =inf {∥x + m∥ | m ∈ M }
=inf {∥x + m∥ | m ∈ M, x ∈ M }
=inf {∥y∥ | y ∈ M } where y = x + m
=0
Thus, x + M = M (zero element of V /M ) =⇒ ∥x + M ∥ = 0.
Conversely, let ∥x + M ∥ = 0
=⇒ inf {∥x + m∥ | m ∈ M } = 0
=⇒ ∃ a sequence {mk } in M such that ∥x+mk ∥ −→ 0 as k −→ ∞
=⇒ lim mk = −x
k−→∞
=⇒ −x ∈ M as M is closed
=⇒ x ∈ M as M is subspace
=⇒ x + M = M
Thus, ∥x + M ∥ = 0 =⇒ x + M = M (zero element of V /M )
∴ ∥x + M ∥ = 0 ⇐⇒ x + M = M (zero element of V /M ) and
hence property (N 2) of norm is satisfied.
(3) Let x + M, y + M ∈ V /M .
∥(x + M ) + (y + M )∥ = ∥(x + y) + M ∥
= inf {∥x + y + m∥ | m ∈ M }
= inf {∥x + y + m1 + m2 ∥ | m1 , m2 ∈ M }
= inf {∥(x + m1 ) + (y + m2 )∥|m1 , m2 ∈ M }
⩽ inf {∥x + m1 ∥ + ∥y + m2 ∥ | m1 , m2 ∈ M }
= inf {∥x + m1 ∥ | m1 ∈ M }
+ inf {∥x + m2 ∥ | m2 ∈ M }
= ∥x + M ∥ + ∥y + M ∥

Therefore, ∥(x + M ) + (y + M )∥ ⩽ ∥x + M ∥ + ∥y + M ∥ and

hence property (N 3) of norm is satisfied.

53
FUNCTIONAL ANALYSIS

(4) If α = 0 then obviously ∥α(x + M )∥ = |α| ∥x + M ∥.

So, let α ̸= 0. Then

∥α(x + M )∥ = ∥αx + M ∥
= inf {∥αx + m∥ | m ∈ M }
= inf {∥α(x + m′ )∥ | m/α = m′ ∈ M }
= inf {|α| ∥x + m′ ∥ | m′ ∈ M }
= |α| inf {∥x + m′ ∥ | m′ ∈ M }
= |α| ∥x + M ∥

and hence property (N 4) of norm is satisfied.

Thus, V /M is a normed space under defined norm.
As with metric spaces, you can understand the concept of norms
from a geometrical point of view. In a normed space (V, ∥ ∥), the open
ball, centered at a with radius ϵ is defined by the set B(a; ϵ) = {x ∈
V | ∥x − a∥ < ϵ} and the open unit ball is given by B(0; 1) = {x ∈
V | ∥x∥ < 1}. Also, the closed ball, centered at a with radius ϵ is de-
fined by the set B[a; ϵ] = {x ∈ V | ∥x − a∥ ⩽ ϵ} and the closed unit ball
is given by B[0; 1] = {x ∈ V | ∥x∥ ⩽ 1}.
Recall that M =set of all limit points of M . For each x ∈ M and
for each ϵ > 0, the open ball {y| ∥y − x∥ < ϵ} must contain a point of
M . Hence, for each element to be in M , it suffices to show that for any
ϵ > 0, ∃ some element of M which is within ϵ distance from it.

Theorem 3.4.4. If M is a subspace of a normed space (V, ∥ ∥) then

M is a closed subspace of V .
Proof. Initially, we show that, M is a subspace of V . Let ϵ > 0. Con-
sider x, y ∈ M and non-zero scalars α, β. Then ∃ x1 , y1 ∈ M such
ϵ ϵ
that ∥x − x1 ∥ < and ∥y − y1 ∥ < .
2 |α| 2 |β|
Using the properties (N 3) and (N 4) of norm, we get,

∥(αx + βy) − (αx1 + βy1 )∥ = ∥α(x − x1 ) + β(y − y1 )∥

⩽ ∥α(x − x1 )∥ + ∥β(y − y1 )∥
= |α| ∥x − x1 ∥ + |β| ∥y − y1 ∥
ϵ ϵ
< |α| + |β|
2 |α| 2 |β|
ϵ ϵ
= +
2 2
=ϵ

So ∃ (α x1 + β y1 ) ∈ M (being subspace) such that ∥(αx + βy) − (αx1 +

βy1 )∥ < ϵ. Thus (αx + βy) ∈ M if x, y ∈ M and α, β are non-zero

54
CHAPTER 3. NORMED SPACES

scalars.(if α = 0 = β then αx + βy = 0 ∈ M ). Hence, M is a subspace

of V . Further, as M is closed, M is a closed subspace of V .
You have learnt the definition of a Cauchy sequence in Rn at your
B.Sc. So you can guess the definition of a Cauchy sequence in a normed
space and get convinced with the following definition.
Definition 3.10. Let (V, ∥ ∥) be a normed space. A sequence {xn }
of vectors in V is said to be a Cauchy sequence if ∀ ϵ > 0 ∃ n0 ∈ N
such that ∀ m, n ⩾ no we have ∥xn − xm ∥ < ϵ. (or equivalently, if
∥xn − xm ∥ −→ 0 as m, n −→ ∞).
You will come to know that the relation between convergent se-
quences and Cauchy sequences in normed spaces.
Theorem 3.4.5. Every convergent sequence in a normed space is a
Cauchy sequence.
Proof. In normed space (V, ∥ ∥), consider a convergent sequence {xn },
converging to x ∈ V . Then as n −→ ∞, we have, ∥xn − x∥ −→ 0.
Clearly, ∥xm − x∥ = ∥x − xm ∥.
Consider

∥xn − xm ∥ = ∥(xn − x) + (x − xm )∥
⩽ ∥xn − x∥ + ∥xm − x∥ by property (N 3) of norm
−→ 0 as m, n −→ ∞

Hence, {xn } is a Cauchy sequence.

Remark 3.4.1. The converse of Theorem 3.4.5 need not be true. i.e. A
Cauchy sequence in a normed space need not be a convergent sequence.
You will see its counter example in next chapter. In the next chapter,
you will come to know that Cauchy sequences play a vital role in the
theory of normed spaces.
Recall the concept of subsequences in R that you learnt at your
B.Sc. and think of it in a normed space (V, ∥ ∥). Can you guess, what
happens to a Cauchy sequence, having a convergent subsequence?. The
next result is related to it.
Theorem 3.4.6. In a normed space, every Cauchy sequence having a
convergent subsequence is convergent.
Proof. In normed space (V, ∥ ∥), consider a Cauchy sequence {xn },
having a convergent subsequence {xnk }. Let {xnk } converge to x ∈ V .
Then as nk −→ ∞, we have, ∥xnk − x∥ −→ 0. Also, since {xn } is a
Cauchy sequence, as n, nk −→ ∞, we have, ∥xnk − xn ∥ −→ 0.
As n −→ ∞, it follows that,
∥xn − x∥ = ∥xn − xnk + xnk − x∥ ⩽ ∥xn − xnk ∥ + ∥xnk − x∥ −→ 0.
Hence, the Cauchy sequence {xn } is convergent.

55
FUNCTIONAL ANALYSIS

3.5 LET US SUM UP

1. Let V be a vector space over the field F(=R or C). A norm ∥ ∥

on V is a real valued function (i.e. ∥ ∥ : V −→ R), satisfying the
following 4 properties/axioms:

(N 1) ∥x∥ ⩾ 0 ∀x ∈ V
(N 2) ∥x∥ = 0 if and only if x = 0v ∀x ∈ V
(N 3) ∥x + y∥ ⩽ ∥x∥ + ∥y∥ ∀x, y ∈ V
(N 4) ∥αx∥ = |α| ∥x∥ ∀x ∈ V and ∀α ∈ F

2. A normed space V is a vector space over the field F(=R or C)

with a norm ∥ ∥ defined on it. In such a case, we say, (V , ∥ ∥)
is a normed space over F. Here, if F = R then V is called a real
normed space and if F = C then V is called a complex normed
space.

3. A normed space is called finite dimensional if the underlying vec-

tor space is finite dimensional, otherwise a normed space is called
infinite dimensional.

4. The vector space Cn = {(x1 , · · · , xn ) | xi ∈ C} over C is a normed

X n 1/2
2
space under the Euclidean norm ∥x∥ = |xi | .
i=1

5. Let p and q be non-negative extended real numbers. For p ⩾ 1, q

is said to be conjugate of p if
1 1
+ = 1, f or 1 < p < ∞
p q
q = ∞, f or p = 1
q = 1, f or p = ∞

6. Hölder’s inequality for finite sums:

1 1
Let 1 < p, q < ∞ and + = 1. For any complex (or real)
p q
numbers x1 , · · · , xn ; y1 , · · · , yn
Xn X n n
1/p X 1/q
p q
| xi y i | ⩽ | xi | | yi |
i=1 i=1 i=1

(Cauchy-Schwarz inequality is a special case of Hölder’s inequality

for p = 2 = q)

56
CHAPTER 3. NORMED SPACES

7. Minkowski’s inequality for finite sums: Let 1 ⩽ p < ∞. For any

complex (or real) numbers x1 , · · · , xn ; y1 , · · · , yn

n
X 1/p n
X 1/p n
X 1/p
p p p
| xi + y i | ⩽ | xi | + | yi |
i=1 i=1 i=1

8. The vector space Cn = {(x1 , · · · , xn ) | xi ∈ C} over C is a normed

space under the norms
 n 1/p
X
p
|x | if 1 ⩽ p < ∞


i
∥x∥p = i=1

max{| x1 |, · · · , | xn |} if p = ∞


These norms are referred as p-norms on Cn .

9. On a vector space V, one can define more than one norm and ac-
cordingly different normed spaces are obtained from same vector
space V.
X∞
p
10. The sequence space l = {x1 , · · · , xn , · · · } | xi |p < ∞ and
i=1

xi ∈ C over C is a normed space under the norms

 ∞ 1/p
X
p
|x | if 1 ⩽ p < ∞


i
∥x∥p = i=1

sup{| x1 |, · · · , | xn |, · · · } if p = ∞


These norms are referred as p-norms on lp .

11. Hölder’s inequality for integrals:

1 1
Let 1 < p, q < ∞ and + = 1. For f ∈ Lp (E) and g ∈ Lq (E)
p q
(where E is bounded closed interval in R),

Z Z 1/p Z 1/q
p q
|f g| ⩽ |f | |g|
E E E

12. Minkowski’s inequality for integrals:

Let 1 ⩽ p < ∞. If f ∈ Lp (E) and g ∈ Lp (E) (where E is bounded
closed interval in R), then

Z 1/p Z 1/p Z 1/p

p p p
| f +g | ⩽ |f | + |g|
E E E

57
FUNCTIONAL ANALYSIS

13. The vector space Lp (E) = {f : E −→ R|f is Lebesgue measur-

able function on E and | f |p is Lebesgue integrable over E} over
R is a normed space under norms
 Z 1/p
 p
|f | if 1 ⩽ p < ∞

∥f ∥p = E
ess sup|f |

if p = ∞
where ess sup |f (x)| = inf {m > 0 | |f (x)| ⩽ m a.e. on E} and
E
L∞ (E) = the class of all those measurable functions f defined on
E which are essentially bounded on E with ess sup|f | < ∞.
14. The vector space C(X) = {f : X −→ R(or C)| f is bounded
continuous function on X} over R(or C) is a normed space under
norm ∥f ∥ = sup{|f (x)|}.
x∈X

15. The vector space C[0, 1] = {f : [0, 1] −→ R| f is aZcontinuous

1
function} over R is a normed space under norm ∥f ∥ = |f (t)|dt.
0
Note that C[0,1] is a particular case of C(X).
16. In a given normed space (V, ∥ ∥), a function d : V × V −→ R+
defined as d(x, y) = ∥x − y∥ is called the distance from x to y,
where x, y ∈ V . Here d is referred as distance function on V .
17. Every normed space is a metric space (with respect to the distance
function).
18. Every metric space need not be a normed space as shown in fol-
lowing example:-
p
Let 0 < p < 1. Consider the sequence space l = x = {x1 , · · · , xn ,
∞
X
p
· · · } xi ∈ R and |xi | < ∞ over R. For x = {x1 , · · · , xn , · · · }
k=1
and y = {y1 , · · · , yn , · · · } in lp , define a function d : lp ×lp −→ R+
X ∞
as d(x, y) = |xk − yk |p . Then (lp , d) is a metric space but not
k=1
a normed space.
19. Let (V, ∥ ∥) be a normed space. A sequence {xn } of vectors in V
is convergent to x ∈ V if ∀ ϵ > 0 ∃ n0 ∈ N such that ∀ n ⩾ no we
have ∥xn − x∥ < ϵ. (or equivalently, if ∥xn − x∥ converges to 0 as
n −→ ∞). Here, we say, x is the limit of the sequence {xn } or
{xn } is a convergent sequence, converging to x. In this case, we
write, lim xn = x or (xn −→ x as n −→ ∞) or lim ∥xn − x∥ =
n−→∞ n−→∞
0 or (∥xn − x∥ −→ 0 as n −→ ∞).

58
CHAPTER 3. NORMED SPACES

20. The limit of a convergent sequence in a normed space is unique.

21. In a normed space (V, ∥ ∥), ∥x∥ − ∥y∥ ⩽ ∥x − y∥ ∀ x, y ∈ V .

22. Consider the normed spaces (V, ∥ ∥V ) and (W, ∥ ∥W ). The func-
tion f : V −→ W is continuous at x0 ∈ V if ∀ ϵ > 0 ∃ δ > 0 such
that ∥x − x0 ∥V < δ =⇒ ∥f (x) − f (x0 )∥W < ϵ
Equivalently, we write, as n −→ ∞, xn −→ xo =⇒ f (xn ) −→
f (xo ). i.e. for every sequence {xn } in V converging to x0 ∈ V ,
the sequence {f (xn )} in W converges to f (x0 ) ∈ W .

23. Let (V, ∥ ∥) be a normed space. Define a function f : V −→ R as

f (x) = ∥x∥. Then the norm ∥ ∥ on V is a real valued continuous
function.

24. Let M be a closed subspace of a normed space (N, ∥ ∥). For

each coset x + M in quotient space N/M , define ∥x + M ∥ =
inf {∥x + m∥ | m ∈ M }. Then N/M is a normed space under the
norm ∥x + M ∥.

25. If M is a subspace of a normed space (V, ∥ ∥) then M is a closed

subspace of V .

26. Let (V, ∥ ∥) be a normed space. A sequence {xn } of vectors in

V is said to be a Cauchy sequence if ∀ ϵ > 0 ∃ n0 ∈ N such
that ∀ m, n ⩾ no we have ∥xn − xm ∥ < ϵ. (or equivalently, if
∥xn − xm ∥ −→ 0 as m, n −→ ∞).

27. Every convergent sequence in a normed space is a Cauchy se-

quence.

3.6 Chapter End Exercise

1. Prove that every inner product space V is a normed space with

√
respect to the norm ∥x∥ = < x, x > ∀ x ∈ V where < x, x >
denotes the inner product of vector x with itself.

2. Show that the vector space C = {z|z ∈ C} over C is a normed

space under the norm ∥z∥ = |z|=absolute value of z ∈ C.

3. Show that the vector space Rn = {(x1 , · · · , xn ) | xi ∈ R} over R is

X n 1/2
2
a normed space under the (Euclidean) norm ∥x∥ = |xi | .
i=1

59
FUNCTIONAL ANALYSIS

4. Show that the vector space Rn = {(x1 , · · · , xn ) | xi ∈ R} over R

X n 1/p
p
is a normed space under the norm ∥x∥ = |xi | where
i=1
1 ⩽ p < ∞.

5. Show that the vector space Rn = {(x1 , · · · , xn ) | xi ∈ R} over R

is a normed space under the norm ∥x∥ = max{| x1 |, · · · , | xn |}.

6. Show that the vector space Lp ([a, b]) = {f : [a, b] −→ R|f is

Lebesgue measurable function on [a, b] and | f |p is Lebesgue
integrable over [a, b]} over R is a normed space under the norm
Z b 1/2
2
∥f ∥ = |f |
a

7. Show that the vector space C([a, b]) = {f : [a, b] −→ R|f is

continuous function} over R is a normed space under the norm
Z b 1/2
2
∥f ∥ = |f |
a

8. Show that the vector space R2 = {(x1 , x2 ) | xi ∈ R} over R is a

normed space under the norm ∥x∥ = |x1 | + |x2 |.

9. If ∥ ∥1 and ∥ ∥2 are 2 norms on a vector space V then check

whether the function ∥ ∥ : V −→ R defined as ∥x∥ = ∥x∥1 + ∥x∥2
is a norm on V .

10. Let (X, ∥ ∥X ) and (Y, ∥ ∥Y ) be normed spaces. Then prove that
X × Y is a normed space under the norm ∥(x, y)∥ = ∥x∥X + ∥y∥Y .

11. Show that a metric d induced by a norm on a normed space

(V, ∥ ∥) is translation invariant. (Hint: to show (a) d(x + u, y +
u) = d(x, y) and (b) d(ax, ay) = |a| d(x, y) ∀x, y ∈ V , for every
scalar a where u is a fixed vector in V ).

12. Let {xn } and {yn } be sequences in a normed space (V, ∥ ∥).
If lim xn = x ∈ V and lim yn = y ∈ V then prove that
n−→∞ n−→∞
lim xn + yn = x + y
n−→∞

13. Let {xn } be a sequence in a normed space (V, ∥ ∥) and {λn } be

a sequence of real numbers. If lim xn = x ∈ V and lim λn =
n−→∞ n−→∞
λ ∈ R then prove that lim λn xn = λ x
n−→∞

14. Prove that in a normed space, if {xn } is a Cauchy sequence then

{∥xn ∥} is a Cauchy sequence of real numbers.

15. Prove that in a normed space, every Cauchy sequence is bounded.

60
CHAPTER 3. NORMED SPACES

16. Prove that in a normed space, a Cauchy sequence is convergent

if and only if it has a convergent subsequence.

61
FUNCTIONAL ANALYSIS

62
Chapter 4

Banach Space

Unit Structure :
4.1 Introduction
4.2 Objective
4.3 Few definitions and examples
4.4 Equivalent Norms and Finite-Dimensional Spaces
4.6 Arzela-Ascoli theorem
4.6 LET US SUM UP
4.7 Chapter End Exercise

4.1 Introduction
In this chapter, you will be introduced to the notion of a Banach
Space. The concept of Banach space was introduced by the Polish
mathematician Stefan Banach in 1922. Banach spaces are fundamental
parts of functional analysis. Banach thought of, when a norm is de-
fined on a vector space, how to deal with Cauchy sequences and hence
about completeness. This chapter has 3 sections, of which in the first
section, you will find several examples on Banach spaces, along with a
characterization of Banach Spaces. In an attempt to obtain a criterion
for determining when a Cauchy sequence with respect to one norm will
also be a Cauchy sequence with respect to other norm, you will be in-
troduced to the notion of equivalent norms in the second section and
interesting results on finite dimensional normed spaces are obtained.
Further, through Riesz lemma, the concept of compactness is linked
to subspaces of finite dimensional normed spaces. In the last section
of this chapter, you will be introduced to Ascoli-Arzela theorem and
the purpose of this theorem is to show a sequence of continuous func-
tions on campact space has a uniformly convergent subsequence, under
certain conditions.

63
FUNCTIONAL ANALYSIS

4.2 Objectives
The main objective of this chapter is to learn the Banach spaces and
interesting results of finite dimensional normed spaces. After
going through this chapter you will be able to:
• Define a Banach space.
• Learn how to check a normed space is a Banach space under the given
norm.
• Prove that Lp spaces are Banach spaces (Riesz-Fisher theorem).
• Define equivalent norms on a normed space.
• Prove that on a finite dimensional normaed space, any two norms are
equivalent.
• Learn Riesz Lemma and results related to it.
• Prove Ascoli-Arzela theorem.

4.3 Few definitions and examples

In previous chapter, you have learnt the concept of a normed space
and at the end, it was mentioned that “A Cauchy sequence in a normed
space need not be a convergent sequence”. Here is a counter example
for it.

Example 18. In the normed space C[0, 1] under the norm ∥f ∥ =

Z 1
|f (t)|dt, consider the sequence {fn (t)} where each function fn :
0
[0, 1] −→ R is defined as follows:

1


 0 if 0 ⩽ t ⩽

 2
1 1 1


fn (t) = 2nt − n if
2
⩽t⩽
2
+
2n

1 1



1
 if + ⩽t⩽1
2 2n
Show that the sequence {fn (t)} is a Cauchy sequence in C[0, 1]. Is this
sequence convergent? Justify.

fn (t) ∈
Solution: Clearly, C[0, 1] ∀n ∈ N.
1 1 1
Also, ∀t ∈ , + , it is easy to see that |fn (t)| ⩽ 1. So,
2 2 2n
|fm (t) − fn (t)| ⩽ |fm (t)| + |fn (t)| ⩽ 1 + 1 = 2. −→ (∗)
Now, with m > n, we have,

64
CHAPTER 4. BANACH SPACE
Z 1
∥fm − fn ∥ = |fm (t) − fn (t)|dt
0
Z 1 Z 1+ 1 Z 1
2 2 2n
= |fm (t)−fn (t)|dt+ |fm (t)−fn (t)|dt+ |fm (t)−fn (t)|dt
1 1 1
0 2 2
+ 2n
1 1
Z
2
+ 2n
= 0+ |fm (t) − fn (t)|dt + 0 by definition of fn (t)
1
2
1 1
Z + 2n
2 1
⩽2 dt = 2 using inequality (∗)
1
2
2n
1
Thus, ∥fm − fn ∥ ⩽ −→ 0 as m, n −→ ∞
n
Hence, the sequence {fn (t)} is a Cauchy sequence in C[0, 1].
Assume that this Cauchy sequence {fn } is convergent in C[0, 1].
Then ∃ a function g ∈ C[0, 1] such that lim fn = g. It is easy to see
n−→∞
Z 1+ 1 Z 1
2 2n
that lim |fn (t) − g(t)|dt = 0 and lim |1 − g(t)|dt = 0.
n−→∞ 0 n−→∞ 1 1
+ 2n
2
Thus, 
0 1
if 0 ⩽ t <

g(t) = 2
1
1
 if <t⩽1
2
1
It is clear that the function g is discontinuous at t = and hence
2
g∈/ C[0, 1], which contadicts g ∈ C[0, 1]. So, our assumption that the
Cauchy sequence {fn } is convergent in C[0, 1] must be wrong. Hence,
the Cauchy sequence Z {fn } is not convergent in normed space C[0, 1]
1
under norm ∥f ∥ = |f (t)|dt.
0
Now, you will come to know when a normed space is called, a Banach
space.
Definition 4.1. A Banach space is a normed space in which every
Cauchy sequence is convergent.
You know that, a metric space is called a complete space if every
Cauchy sequence of points in it converges to a point in the space. In
view of this definition, the normed space (V, ∥.∥) is said to be complete
if V is complete as a metric space with the metric d(u, v) = ∥u −
v∥ ∀ u, v ∈ V . Hence, you can redefine the definition 4.1 as a Banach
space is a complete normed space or a Banach space is a normed space
which is a complete metric space.
Now, you will notice that examples of Banach spaces are in abun-
dance. Initially, you will find an example of a normed space which is
not a Banach space.
Example 19. Show that the vector space C[0, 1] = {f : [0, 1] −→ R|
f is a continuous function} over R is not a Banach space under norm

65
FUNCTIONAL ANALYSIS
Z 1
∥f ∥ = |f (t)|dt.
0
Solution: Refer Example 14 to show that C[0, 1] is a normed space
under defined norm. Refer Example 18 to show that a Cauchy sequence
in C[0, 1] is not convergent on it.
Thus, C[0, 1] is not a Banach space under defined norm.

You can generalize above Example 19 to get the following result.

Theorem 4.3.1. Prove that the vector space C[a, b] = {f : [a, b] −→

R(or C)| f is a continuous function} over R(or C) is not a Banach
Z b 1/p
p
space under norm ∥f ∥ = |f (t)| dt where 1 ⩽ p < ∞
a

Proof. Left to the reader.

But you will see that C[0, 1] is a Banach space with respect to sup
norm.

Example 20. Show that the vector space C[0, 1] = {f : [0, 1] −→

R(or C)| f is a continuous function} over R(or C) is a Banach space
under norm ∥f ∥ = sup {|f (x)|}.
x∈[0,1]
Solution: Refer Example 15 to show that C[0, 1] is a normed space
under defined norm. Now, consider a Cauchy sequence {fn } in C[0, 1].
Let ϵ > 0. Then {fn } being a Cauchy sequence, ∃ m0 ∈ N such
ϵ
that ∀ l, n ≥ m0 and ∀ x ∈ [0, 1], we have, ∥fl − fn ∥ < i.e.
3
ϵ
sup {|fl (x) − fn (x)|} < and hence ∀ l, n ≥ m0 and ∀ x ∈ [0, 1]; we
x∈[0,1] 3
ϵ
have, |fl (x) − fn (x)| < −→ (∗)
3
This shows that for fixed but arbitrary x ∈ [0, 1], {fn (x)} is a Cauchy
sequence in R(or C). As every Cauchy sequence in R(or C) is conver-
gent, ∃ a function f : [0, 1] −→ R(or C) such that lim fn (x) = f (x),
n−→∞
for each fixed x ∈ [0, 1]. So, making l −→ ∞ in inequality (∗), we get,
ϵ
∀ n ≥ m0 and for each fixed x ∈ [0, 1]; |f (x) − fn (x)| < . Taking
3
supremum over x ∈ [0, 1], (as m0 is independent of x and x is arbi-
ϵ
trary), we get, ∥f − fn ∥ < ∀ n ≥ m0 −→ (∗∗)
3
Thus, fn −→ f uniformly, i.e. ∥fn − f ∥ −→ 0 as n −→ ∞.
Claim: f ∈ C[0, 1].
Consider a sequence {xn } in [0, 1] such that xn −→ x where x ∈ [0, 1].
ϵ
From inequality (∗∗), in particular n = m0 gives, ∥f − fm0 ∥ < . And
3
therefore, for n sufficiently large, we have,
|f (xn )−f (x)| ≤ |f (xn )−fm0 (xn )|+|fm0 (xn )−fm0 (x)|+|fm0 (x)−f (x)|

66
CHAPTER 4. BANACH SPACE

ϵ ϵ
which implies |f (xn ) − f (x)| < + |fm0 (xn ) − fm0 (x)| +
3 3
ϵ ϵ ϵ
and hence |f (xn )−f (x)| < + + = ϵ as fm0 ∈ C[0, 1] and xn −→ x
3 3 3
So, f (xn ) −→ f (x) if xn −→ x which implies f is continuous and we
are done. Thus, the Cauchy sequence {fn } in C[0, 1] is convergent to
f ∈ C[0, 1]. Hence, C[0, 1] is a Banach space under defined norm.
Remark 4.3.1. From above Example 20, it is clear that in the space
C[0, 1] under norm ∥f ∥ = sup {|f (x)|},
x∈[0,1]
fn −→ f
⇐⇒ fn (x) −→ f (x) uniformly on [0,1]
⇐⇒ ∀ϵ > 0, ∃ m0 ∈ N (independent of x) such that ∀x ∈ [0, 1] and
∀n ⩾ m0 , we have, |fn (x) − f (x)| < ϵ

You can generalize above Example 20 to get the following results.

Theorem 4.3.2. Prove that the vector space C[a, b] = {f : [a, b] −→
R(or C)| f is a continuous function} over R(or C) is a Banach space
under norm ∥f ∥ = sup {|f (x)|}.
x∈[a,b]

Proof. Left to the reader.

Theorem 4.3.3. Prove that the vector space C(X) over R(or C) is a
Banach space under norm ∥f ∥ = sup{|f (x)|} where X is a compact
x∈X
space.

Proof. Left to the reader.

Example 21. Show that the vector space R = {x|x ∈ R} over R is a

Banach space under the norm ∥x∥ = |x|=absolute value of x ∈ R.

Solution: Refer Example 6 of previous chapter to show that R is a

normed space under defined norm.
Claim: Every Cauchy sequence in R is convergent in R.
Let {xn } be a Cauchy sequence. Then for any ϵ > 0 ∃ n0 ∈ N such
ϵ
that ∀n, m ≥ n0 we have |xn − xm | < ←−(1)
2
{xn } being Cauchy, is bounded and hence {xn } has a convergent sub-
sequence {xnk } ←− (using Bolzano Weirstrass theorem)
So suppose {xnk } converges to l ∈ R. Then ∃ k0 ∈ N such that ∀k ≥ k0 ,
we have, |xnk − l| < 2ϵ ←−(2)
Note that nk ≥ k ≥ k0
Choose p=max{k0 , n0 }. Then p ≥ k0 and p ≥ n0 ∀n, m ≥ p using (1),
we have, |xn − xm | < 2ϵ ←−(3)
ϵ
∀k ≥ p using (2), we have, |xnk − l| < 2
So in particular, for k=p, we have, |xnp − l| < 2ϵ

67
FUNCTIONAL ANALYSIS

Now, in particular, for m=np ≥ p, ∀n ≥ p by (3), we get, |xn − xnp | < 2ϵ

Consider |xn − l| =|(xn − xnp ) + (xnp − l)| ≤ |xn − xnp | + |xnp − l|
< 2ϵ + 2ϵ ∀n ≥ p
∴ ∀ϵ > 0 there exists p ∈ N such that ∀n ≥ p, we have, |xn − l| < ϵ
Thus, the Cauchy sequence {xn } in R is convergent in R and hence R
is a Banach space under defined norm.
Example 22. Show that the vector space Cn = {(x1 , · · · , xn ) | xi ∈ C}
n
X 1/p
p
over C is a Banach space under the norm ∥x∥ = |xi | where
i=1
1 ⩽ p < ∞.
Solution: Refer Example 8 to show that Cn is a normed space under
defined norm. Now, consider a Cauchy sequence {xm } in Cn . As xm ∈
(m) (m)
Cn is an n-tuple, denote xm = (x1 , · · · , xn ). Let ϵ > 0. Then
{xm } being a Cauchy sequence, ∃ m0 ∈ N such that ∀ l, m ≥ m0 , we
(m) (l)
have, ∥xm − xl ∥ < ϵ which implies ∥(x1 − x1 , · · · , x(m)
n − x(l)
n )∥ =
X n 1/p
(m) (l)
|xi − xi |p <ϵ −→ (∗)
i=1
(m) (l)
Eq.(∗) implies that |xi −xi | < ϵ ∀ l, m ≥ m0 and ∀ i = 1, · · · , n
(r)
This shows that for fixed but arbitrary i, {xi } is a Cauchy sequence in
(r)
C. As every Cauchy sequence in C is convergent, {xi } must converge,
(r)
say to, zi ∈ C. Thus lim xi = zi ∈ C ∀ i = 1, · · · , n −→ (∗∗)
r−→∞
Making l −→ ∞ in Eq.(∗) and then using Eq.(∗∗), we get, ∀ m ≥ m0 ,
Xn 1/p
(m) p
|xi − zi | < ϵ which implies ∥xm − z∥ < ϵ where z =
i=1
(z1 , · · · , zn ) ∈ Cn . It follows that the Cauchy sequence {xm } in Cn
is convergent to z ∈ Cn . Hence, Cn is a Banach space under defined
norm.
Example 23. Show that the vector space Cn = {(x1 , · · · , xn ) | xi ∈ C}
over C is a Banach space under the norm ∥x∥ = max{| x1 |, · · · , | xn |}.
(This norm is referred as ∥x∥∞ on Cn ).
Solution: Refer Example 9 to show that Cn is a normed space under
defined norm. Now, consider a Cauchy sequence {xm } in Cn . As xm ∈
(m) (m)
Cn is an n-tuple, denote xm = (x1 , · · · , xn ). Let ϵ > 0. Then
{xm } being a Cauchy sequence, ∃ m0 ∈ N such that ∀ l, m ≥ m0 ,
(m) (l)
we have, ∥xm − xl ∥ < ϵ which implies ∥(x1 − x1 , · · · , x(m) (l)
n − xn )∥ =
(m) (l)
max{|x1 −x1 |, · · · , |x(m) (l)
n −xn |} < ϵ −→ (∗)
(m) (l)
Eq.(∗) implies that |xi −xi | < ϵ ∀ l, m ≥ m0 and ∀ i = 1, · · · , n
(r)
This shows that for fixed but arbitrary i, {xi } is a Cauchy sequence in
(r)
C. As every Cauchy sequence in C is convergent, {xi } must converge,
(r)
say to, zi ∈ C. Thus lim xi = zi ∈ C ∀ i = 1, · · · , n −→ (∗∗)
r−→∞
Making l −→ ∞ in Eq.(∗) and then using Eq.(∗∗), we get, ∀ m ≥ m0 ,

68
CHAPTER 4. BANACH SPACE

(m)
max{|x1 − z1 |, · · · , |x(m) n − zn |} < ϵ which implies ∥xm − z∥ < ϵ where
z = (z1 , · · · , zn ) ∈ Cn . It follows that the Cauchy sequence {xm } in Cn
is convergent to z ∈ Cn . Hence, Cn is a Banach space under defined
norm.

Example 24. Show that the vector space Cn = {(x1 , · · · , xn ) | xi ∈ C}

X n 1/2
2
over C is a Banach space under the norm ∥x∥ = |xi | .
i=1

Solution: Left to the reader.

Example 25. Let 1 ⩽ p < ∞. Show that the sequence space lp =

∞
X
p
{x1 , · · · , xn , · · · } | xi | < ∞ and xi ∈ C over C is a Banach
i=1
∞
X 1/p
p
space under the norm ∥x∥ = |xi | .
i=1
Solution: Refer Example 10 to show that lp is a normed space under
defined norm. Now, consider a Cauchy sequence {xm } in lp . As xm ∈ lp ,
∞
X
(m) (m) (m)
denote xm = {x1 , · · · , xn , · · · } where | xi |p < ∞ . Let ϵ > 0.
i=1
Then as {xm } is a Cauchy sequence, ∃ m0 ∈ N such that ∀ l, m ≥ m0 , we
(m) (l)
have, ∥xm − xl ∥ < ϵ which implies ∥{x1 − x1 , · · · , x(m) (l)
n − xn , · · · }∥ =
X ∞ 1/p
(m) (l) p
|xi − xi | < ϵ. −→ (∗)
i=1
(m) (l)
Eq.(∗) implies that |xi − xi | < ϵ ∀ l, m ≥ m0 and ∀ i
(r)
This shows that for fixed but arbitrary i, {xi } is a Cauchy sequence in
(r)
C. As every Cauchy sequence in C is convergent, {xi } must converge,
(r)
say to, zi ∈ C. Thus lim xi = zi ∈ C ∀i −→ (∗∗)
r−→∞
k
X (m) (l)
From Eq.(∗), it is clear that, ∀ k ∈ N, |xi −xi |p < ϵp ∀ l, m ≥ m0 .
i=1
k
X
(m)
Making l −→ ∞ and then using Eq.(∗∗), we get, |xi −zi |p < ϵp
i=1
∞
X
(m)
∀ m ≥ m0 . Further, making k −→ ∞, we get, |xi − zi | < ϵp
p

i=1
∀ m ≥ m0 . This implies that (z − xm ) ∈ lp where z = {z1 , · · · , zn , · · · }.
So, z = ((z − xm ) + xm ) ∈ lp .
X∞ 1/p
(m) p
Now, ∥xm − z∥ = |xi − zi | < ϵ ∀ m ≥ m0 . It follows that
i=1
the Cauchy sequence {xm } in lp is convergent to z ∈ lp . Hence, lp is a
Banach space under defined norm.

69
FUNCTIONAL ANALYSIS

p
Example 26. Show that the sequence space l = {x1 , · · · , xn , · · · } sup

{| x1 |, · · · , | xn |, · · · } < ∞ and xi ∈ C over C is a Banach space
under the norm ∥x∥ = sup{| x1 |, · · · , | xn |, · · · }.
(This norm is denoted as ∥x∥∞ on lp ).
Solution: Left to the reader.

To show that the normed space Lp (E) is a Banach space, we first

prove characterization of a Banach space in the forem of lemma, for
which you are introduced with following terms.
Definition 4.2. A sequence {xk } in a normed space (V, ∥ ∥) is said to
be summable to the sum s if the sequence {sn } of the partial sums of
∞
X
the series xk converges to s ∈ V . i.e. ∥sn − s∥ −→ 0 as n −→ ∞.
k=1
∞
X
In this case, we write, s = xk .
k=1

Definition 4.3. The sequence {xk } in a normed space (V, ∥ ∥) is said

∞
X
to be absolutely summable if ∥xk ∥ < ∞.
k=1

Lemma 4.3.1. A normed space (V, ∥ ∥) is a Banach space if and only

if every absolutely summable sequence in V is summable.

Proof. Let the normed space (V, ∥ ∥) be a Banach space. Consider an

X∞
absolutely summable sequence {xk } in V . Then ∥xk ∥ = M < ∞
k=1
X∞
where M > 0. Thus, ∀ ϵ > 0, ∃ r ∈ N such that ∥xk ∥ < ϵ. So, if
k=r
n
X ∞
X
sn = xk is nth partial sum of the series xk then ∀ n > m ⩾ r,
k=1 k=1
n
X n
X ∞
X
we have, ∥sn − sm ∥ = xk ⩽ ∥xk ∥ ⩽ ∥xk ∥ < ϵ. It
k=m+1 k=m+1 k=r
∞
X
follows that, the sequence {sn } of partial sums of the series xk is a
k=1
Cauchy sequence in Banach space V and hence must converge to some
element; say s ∈ V . Thus, {xk } is summable in V and we are done.
Conversely, in a normed space (V, ∥ ∥), assume that every absolutely
summable sequence in V is summable. Consider a Cauchy sequence
{xk } in V . Then, for each k ∈ N, ∃ nk ∈ N such that ∀ n, m ⩾ nk , we
1
have, ∥xn − xm ∥ < k −→ (∗)
2
70
CHAPTER 4. BANACH SPACE

Choose nk such that nk+1 > nk . Then {xnk } is a subsequence of Cauchy

sequence {xn }.
Define y1 = xn1
y2 = xn2 − xn1
..
.
yk = xnk − xnk−1
..
.
1
Then as nk > nk−1 , by (∗), we have, ∥yk ∥ < where k > 1.
2k−1
∞
X k
X
th
Consider the series yk . Its k partial sum is sk = yi = xnk . As
k=1 i=1
∞
X 1
k−1
is a geometric series with first term 0.5 and common ratio 0.5,
k=2
2
∞
X 1 0.5
we have, k−1
= = 1.
k=2
2 1 − 0.5
∞ ∞ ∞
X X X 1
Now ∥yk ∥ = ∥y1 ∥+ ∥yk ∥ ⩽ ∥y1 ∥+ k−1
= ∥y1 ∥+1 = M < ∞
k=1 k=2 k=2
2
which implies that the sequence {yk } is absolutely summable and hence
summable (by hypothesis). So, the sequence {sk } converges to some
s ∈ V and hence the subsequence {xnk } is convergent. By Theorem
3.4.6, it follows that, the Cauchy sequence {xn } in V is convergent.
Therefore, the normed space (V, ∥ ∥) is a Banach space.

Recall the following results which you studied in measure theory.

Fatou’s Lemma : Let {fn } be a sequence ofZ non-negativeZmeasurable
functions and lim fn = f a.e. on E. Then f ⩽ lim fn
n−→∞ E n−→∞ E

Lesbesgue Dominated Convergence Theorem : Let g be an (Les-

besgue) integrable function on E. Let {fn } be a sequence of measurable
functions such that |fn | ⩽ g on E and lim fn = f a.e. on E. Then
Z Z n−→∞

f = lim fn
E n−→∞ E

Example 27. Let 1 ⩽ p < ∞ & E be a (bounded closed interval in R)

measurable set. Show that the vector space Lp (E) = {f : E −→ R|f
is Lebesgue measurable function on E and | f |p is Lebesgue integrable
Z 1/p
p
over E} over R is a Banach space under norm ∥f ∥p = |f | .
E
(This norm is referred as p-norms on Lp (E)).
Solution: Refer Example 12 to show that Lp (E) is a normed space
under defined norm.
Consider an absolutely summable sequence {fk } in Lp (E). Then

71
FUNCTIONAL ANALYSIS

∞
X
∥fk ∥p = M < ∞ where M > 0.
k=1

n
X
1. Define a sequence {gn } of functions where gn (x) = |fk |.
k=1
Clearly, for each x, {gn (x)} is an increasing sequence of (extended)
real numbers and ∃ some (extended) real number g(x) such that
gn (x) −→ g(x) ∀x ∈ E

2. The function g is measurable, since the functions gn (x) are mea-

surable.

3. By Minkowski’s inequality (for finite sums) in Lemma 3.3.3, we

have,
n
X Xn Z
∥gn ∥p = |fk | ⩽ ∥fk ∥p < M and hence |gn |p ⩽ M p .
k=1 p k=1 E

4. As gn ⩾ 0 and lim gnp = g p , by Fatou’s lemma, we have,

Z Zn−→∞
p
g ⩽ lim gnp ⩽ M p
E n−→∞ E
It follows that g p is integrable and hence g(x) is finite a.e. on E.

5. Thus, we find that, for each x, for which g(x) is finite, the se-
quence {fn (x)} is an absolutely summable sequence of real num-
bers and therefore, must be summable to a real number, say s(x).

6. Define s(x) = 0 for those x where g(x) = ∞. Then, the function

n
X
s so defined is the limit a.e. of partial sums sn (x) = fk (x).
k=1
i.e. sn (x) −→ s(x) a.e. Hence, s is a measurable function.
(note that |sn (x) − s(x)|p −→ 0 a.e. ∀ x)
n
X
7. Clearly, |sn (x)| ⩽ |fk (x)| = gn (x) ⩽ g(x). Then as g ∈ Lp (E),
k=1
we have, s ∈ Lp (E). ←− since, if h ∈ Lp & |f | ⩽ |h| then f ∈ Lp .

8. It is easy to see that |sn (x) − s(x)|p ⩽ 2p (g(x))p .

9. As 2p g p is an integrable function and |sn (x) − s(x)|p −→ 0 a.e.

∀
R x, by Lesbesgue Dominated Convergence Theorem, we have,
p
|s
E n
− s| −→ 0 and hence ∥sn − s∥p −→ 0

72
CHAPTER 4. BANACH SPACE

∞
X
10. So, the sequence {sn } of partial sums of series fk converges to
k=1
s ∈ Lp (E). i.e. the absolutely summable sequence {fk } in Lp (E)
is summable in Lp (E).

Thus, by Lemma 4.3.1, Lp (E) is a Banach space under defined norm.

Example 28. Let E be a (bounded closed interval in R) measurable
set. Show that the vector space L∞ (E) = {f : E −→ R|f is a mea-
surbale function on E and ess sup|f | < ∞} over R is a Banach space
under norm ∥f ∥∞ = ess supE |f (x)| = inf {m > 0 | |f (x)| ⩽ m a.e. on
E}.
Solution: Refer Example 13 to show that L∞ (E) is a normed space
under defined norm.
Consider a Cauchy sequence {fn } in L∞ (E). Then |fn (x)−fm (x)| ⩽
∥fn − fm ∥∞ except on a set An,m ⊂ [a, b] = E with m(An,m ) = 0.
If A = ∪n,m An,m then m(A) = 0 and |fn (x) − fm (x)| ⩽ ∥fn − fm ∥∞
∀ n, m and ∀x ∈ (E − A).
Therefore, it follows that, {fn } converges uniformly to a bounded
limit f outside A and the result is proved by observing the fact that
the convergence in L∞ (E) is equivalent to uniform convergence outside
a set of measure zero.
Thus, as {fn } −→ f outside A, we have, L∞ (E) is a Banach space
under defined norm.
The Examples 27 and 28 are well known by following theorem.
Theorem 4.3.4. Riesz-Fischer theorem:
For 1 ⩽ p ⩽ ∞, Lp spaces are Banach spaces.

Proof. Combine the answers to the Examples 27 and 28.

Now, you will see next two results related to quotient space.
Theorem 4.3.5. Let M be a closed subspace of a Banach space (V, ∥ ∥).
For each coset x+M in quotient space V /M , define ∥x+M ∥ = inf {∥x+
m∥ | m ∈ M }. Then V /M is a Banach space under the norm ∥x + M ∥.

Proof. Refer Theorem 3.4.3 to show that V /M is a normed space under

defined norm. Now, let {sn + M } be a Cauchy sequence in V /M . Then
sn ∈ V . We know that, a Cauchy sequence is convergent if and only if
it has a convergent subsequence. So in order to show that {sn + M } is
convergent, it is sufficient to show that it has a convergent subsequence.
We construct a subsequence in the following manner:
1
As {sn + M } is Cauchy, given ϵ = > 0, ∃ n1 ∈ N such that
2
1
∀ n, m ⩾ n1 , we have, ∥(sn + M ) − (sm + M )∥ < ϵ =
2
Set sn1 = x1 ∈ V .

73
FUNCTIONAL ANALYSIS

1
Similarly, ϵ = > 0, ∃ n2 ∈ N with n2 > n1 such that ∀ n, m ⩾ n1 ,
22
1
we have, ∥(sn + M ) − (sm + M )∥ < ϵ = 2
2
Set sn2 = x2 ∈ V .
In general, having chosen x1 , · · · , xk and n1 , · · · , nk , let nk+1 > nk be
1
such that ∀ n, m ⩾ nk , we have, ∥(sn + M ) − (sm + M )∥ < ϵ = k+1
2
Set snk+1 = xk+1 ∈ V .
Thus, we have obtained a subsequence {xk + M } of {sn + M } such that
1
∥(xk + M ) − (xk+1 + M )∥ < k for k = 1, 2, · · ·
2
Claim: This subsequence converges to an element of V /M .
1
Let y1 ∈ x1 + M . Choose y2 ∈ x2 + M such that ∥y1 − y2 ∥ < .
2
1
Next, choose y3 ∈ x3 + M such that ∥y2 − y3 ∥ < 2 . Continuing this
2
1
process, we obtain a sequence {yn } in V such that ∥yn − yn+1 ∥ < n .
2
1
Note that given ϵ > 0, we can choose m0 ∈ N so large that m0 −1 < ϵ.
2
Then for n > m ⩾ m0 , we have,
∥ym − yn ∥ = ∥(ym − ym+1 ) + (ym+1 − ym+2 ) + · · · + (yn−1 − yn )∥
⩽ ∥(ym − ym+1 )∥ + ∥(ym+1 − ym+2 )∥ + · · · + ∥(yn−1 − yn )∥
n−1
X 1
<
i=m
2i
∞
X 1 1
< i
= m−1
,being a geometric series
i=m
2 2
1
< m0 −1
2
<ϵ
∴ {yn } is a Cauchy sequence in Banach space V and hence ∃ y ∈ V
such that ∥yn − y∥ −→ 0 as n −→ ∞.
Consider ∥(xn + M ) − (y + M )∥ = ∥(xn − y) + M ∥
= inf {∥(xn − y) + m∥ | m ∈ M }
⩽ ∥(xn − y) + m∥ ∀m∈M
As yn = xn + mn for some mn ∈ M , we conclude that,
∥(xn + M ) − (y + M )∥ ⩽ ∥yn − y∥ −→ 0 as n −→ ∞.
=⇒ xn + M −→ y + M ∈ V /M
=⇒ The Cauchy sequence {sn + M } has a subsequence {xn + M }
which is convergent in V /M .
=⇒ The Cauchy sequence {sn + M } is convergent in V /M .
=⇒ The normed space V /M is complete and we are done.

Theorem 4.3.6. Let M be a closed subspace of a normed space (V, ∥.∥).

If M and V /M are Banach spaces then V is a Banach space

Proof. Consider a Cauchy sequence in V . Let ϵ > 0 be given. Then ∃

74
CHAPTER 4. BANACH SPACE

n0 ∈ N such that ∀m, n ⩾ n0 , we have, ∥xn − xm ∥ < ϵ. Also, {xn + M }

is a sequence in V /M .
Consider ∥(xn + M ) − (xm + M )∥ = ∥(xn − xm ) + M ∥
= inf {∥(xn − xm ) + y∥ | y ∈ M }
⩽ ∥(xn − xm ) + y∥ ∀y∈M
But y = 0 ∈ M . So, ∥(xn + M ) − (xm + M )∥ ⩽ ∥(xn − xm )∥ < ϵ
∀m, n ⩾ n0 . This implies that {xn + M } is a Cauchy sequence in
Banach space V /M and so it must converge to some z + M ∈ V /M for
some z ∈ V . Hence ∥(xn + M ) − (z + M )∥ = ∥(xn − z) + M ∥ −→ 0 as
n −→ ∞
Now, for each n ∈ N, ∃ yn ∈ M such that
∥xn − z + M ∥ = inf {∥(xn − z) + yn ∥ | yn ∈ M } ⩽ ∥(xn − z) + yn ∥ and
1
thus ∥xn − z + M ∥ < ∥(xn − z) + yn ∥ + .
n
Consider ∥yn − ym ∥ = ∥(xn − z + yn ) − (xm − z + ym ) − (xn − xm )∥
∴ ∥yn − ym ∥ ⩽ ∥(xn − z + yn )∥ + ∥(xm − z + ym )∥ + ∥(xn − xm )∥ and
thus ∥yn − ym ∥ ⩽ ∥xn − xm ∥ < ϵ ∀m, n ⩾ n0 . This implies that {yn }
is a Cauchy sequence in Banach space M and so it must converge to
some y ∈ M . Hence yn −→ y in M as n −→ ∞.
Now, since as n −→ ∞, we have ∥xn − z + yn ∥ −→ 0 and hence
xn −→ (z − yn ) in V . So, lim xn = lim (z − yn ) = z − y = x ∈ V .
n−→∞ n−→∞
∴ the Cauchy sequence {xn } in V is convergent to x ∈ V . Hence V is
complete and we are done.
This section is concluded with following result.
Theorem 4.3.7. Every complete subspace of a normed space is closed.
Proof. Consider a normed space (V, ∥.∥). Let M be a complete subspace
ofV . Let
z be any limit point of M . Then ∀ n ∈ N, the open ball
1 1
B z, = x ∥x − z∥ < must contain atleast one point yn other
n n
than z. So, ∃ y1 , y2 , · · · , yn , · · · in M such that

∥y1 − z∥ < 1
1
∥y2 − z∥ <
2
..
.
1
∥yn − z∥ <
n
..
.

Claim: the sequence {yn } of points in M converges to z.

1
Let ϵ > 0. Choose m ∈ N such that < ϵ. Then
m
1 1
∀ n ⩾ m, we have, ∥yn − z∥ < ⩽ <ϵ
n m
75
FUNCTIONAL ANALYSIS

Thus, ∀ϵ > 0, ∃ m ∈ N such that ∥yn − z∥ < ϵ ∀ n ⩾ m

=⇒ the sequence {yn } converges to z. Hence the claim.
As every convergent sequence in a metric space is a Cauchy sequence,
the sequence {yn } is a Cauchy sequence in M . But, since M is complete,
we have, every Cauchy sequence in M converges to a point in M . So,
yn −→ z implies z ∈ M .
Thus, we have shown that, every limit point of M belongs to M
and consequently, M is closed.

4.4 Equivalent Norms and

Finite-Dimensional Spaces

You have seen that there are many norms on the same finite di-
mensional vector space X. It is interesting to see that all these norms
on X lead to same topology for X, that is, the open subsets of X are
the same, regardless of the particular choice of a norm on X. In this
section, you will see the notion of equivalent norms and basic results
related to it.

Definition 4.4. Two norms ∥ ∥1 and ∥ ∥2 on a normed space X are

said to be equivalent and written as ∥ ∥1 ∼ ∥ ∥2 , if ∃ positive real
numbers a and b (independent of x ∈ X) such that
a ∥x∥1 ⩽ ∥x∥2 ⩽ b ∥x∥1 ∀ x ∈ X

With little effort you can show that equivalent of norms is an equiv-
alence relation on the set of all norms over a given space.

Two norms ∥ ∥1 and ∥ ∥2 on a normed space X are equivalent if

and only if any (Cauchy) sequence in X converges with respect to ∥ ∥1
converges with respect to ∥ ∥2 and conversely.

From following result, you will come to know that although one can
define many different norms on finite dimensional linear spaces, there
is only one topology derived from these norms.

Theorem 4.4.1. On a finite dimensional normed space X, any two

norms are equivalent.

Proof. Let X be a finite dimensional vector space over F with dim(X)=n.

Then X ∼ = Fn , since if B = {v1 , · · · , vn } is a basis for X then each
Xn
x ∈ X can be uniquely represented as x = aj vj for some unique
j=1
scalars aj ∈ F which gives an element x̄ = (a1 , · · · , an ) ∈ Fn .

76
CHAPTER 4. BANACH SPACE

n
X 1/2
n 2
Now, by Euclidean norm in F , we have, ∥x̄∥2 = |aj |
j=1
n
X n
X 1/2
2
and for each x = aj vj ∈ X, define ∥x∥2 = |aj | . Then
j=1 j=1
∥x̄∥2 = ∥x∥2 .
Suppose ∥.∥ is a norm on X. Then, for each x ∈ X,
Xn Xn n
X n
1/2 X 1/2
2 2
∥x∥ = aj vj ⩽ |aj | ∥vj ∥ ⩽ ∥vj ∥ |aj |
j=1 j=1 j=1 j=1
n
X 1/2
If M = ∥vj ∥2 then M > 0 such that ∥x∥ ⩽ M ∥x∥2 ∀ x ∈ X
j=1
which gives the one half of the desired inequality.
Now, to establish the other inequlaity, define S = {x̄ = (a1 , · · · , an ) ∈
Fn | ∥x̄∥2 = 1}. Then S is closed and bounded, and hence is compact
(by Heine-Borel theorem) with respect to the Euclidean norm.
Define f : S −→ R as f (x̄) = ∥x∥. As B is a linearly independent
set and since x̄ ∈ S, i.e. x̄ ̸= 0, all aj cannot vanish simultaneously on
S so that f (x̄) > 0 on S.
Clearly, |f (x̄) − f (ȳ)| = ∥x∥ − ∥y∥ ⩽ ∥x − y∥ ⩽ M ∥x − y∥2 .
It follows that f is continuous on S. Thus, f is a positive valued
continuous function on the compact set S and therefore, f attains its
minimum m > 0 at some point on the compact set S. Consequently,
whenever x̄ ∈ S, we have, f (x̄) = ∥x∥ ⩾ m.
Thus, for each 0 ̸= ū = (c1 , · · · , cn ) ∈ Fn ,
n
X ū
∥u∥ = cj vj = ∥ū∥2 f ⩾ m ∥ū∥2 = m ∥u∥2
j=1
∥ū∥2
Therefore, ∃ positive real numbers m and M such that
m ∥u∥2 ⩽ ∥u∥ ⩽ M ∥u∥2 ∀ u ∈ X
This implies that, any given norm ∥.∥ is equivalent to the 2-norm
∥.∥2 . Since, equivalence of norms is an equivalence relation, it follows
that any two norms on X are equivalent.

Now, you will see some immediate consequences of this theorem in

the form of following corollaries.
Corollary 4.4.1. If V is a finite dimensional normed space, then V is
complete.

Proof. Let V be a finite dimensional vector space over F where F =

R or C with dim(V )=n > 0 and {e1 , · · · , en } be basis for V . Then
Xn
each x ∈ V can be uniquely represented as x = aj ej for some
j=1
unique scalars aj ∈ F. It can be easily proved that ∥x∥0 = max |ai | is a
i

77
FUNCTIONAL ANALYSIS

norm on V . This norm is called the zeroth norm.

Since, by Theorem 4.4.1, on a finite dimensional normed space V ,
any two norms are equivalent, to prove V is complete, it suffices to
prove the completeness of V with respect to this zeroth norm.
Let {yn } be any Cauchy sequence in V . Consider the ith term of
n
X (i)
this sequence {yn } which is yi = ak ek for some uniquely determined
k=1
(i) (i)
scalarsa1 , · · · , an
in F.
Since, {yn } is Cauchy, we have, ∥yn − ym ∥0 −→ 0 as m, n −→ ∞.
n
X
(n) (m)
=⇒ ak − ak ek −→ 0 as m, n −→ ∞.
k=1 0
(n) (m)
=⇒ max |ak − ak | −→ 0 as m, n −→ ∞.
k
(n) (m)
=⇒ |ak − ak | −→ 0 as m, n −→ ∞.
(m)
=⇒ {ak } is Cauchy sequence in F for k = 1, · · · , n,
As F is complete, ∃ scalars a1 , · · · , an in F such that
(m)
ak −→ ak as m −→ ∞ −→ (∗)
(m) ∞
∴ the Cauchy sequence {ak }m=1 converges for some ak (k = 1, · · · , n).
X n
Let y = ak ek then y ∈ V . To show that ym −→ y as m −→ ∞,
k=1
let ϵ > 0 be given. Then using (∗), we get,
Xn
(m) (m)
∥ym −y∥0 = ak −ak ek = max |ak −ak | −→ 0 as m −→ ∞
k
k=1 0
which implies ym −→ y as m −→ ∞
∴ the Cauchy sequence {ym } in V converges to y ∈ V and hence V is
complete.

Corollary 4.4.2. If M is any finite dimensional subspace of a normed

space V , then M is closed.

Proof. Let M be a finite dimensional subspace of a normed space V . As

by Corollary 4.4.1, every finite dimensional normed space is complete,
we have, M is complete subspace of V . Further, as by Theorem 4.3.7,
every complete subspace of a normed space is closed, we have, M is
closed.

Now, you will see another interesting result (about closed subspaces)
in the form of lemma, which is due to the famous Hungarian mathe-
matician Riesz. This result/lemma is to prove very important theorem
that relates finite dimensional normed spaces with compactness of its
bounded subset.

Lemma 4.4.1. (Riesz Lemma): (F.Riesz, 1918)

Let M be a closed proper subspace of a normed space V and let a ∈ R

78
CHAPTER 4. BANACH SPACE

be such that 0 < a < 1. Then ∃ a vector xa ∈ V such that ∥xa ∥ = 1

and ∥x − xa ∥ ⩾ a ∀x ∈ M

Proof. Let M be a closed proper subspace of a normed space V . Then

we have M = M and ∃ x1 ∈ (V − M ). i.e. x1 ∈ / M = M.
Define h = inf ∥x − x1 ∥ = d(x1 , M ). Then h ⩽ ∥x − x1 ∥ ∀x ∈ M .
x∈M
Clearly, h > 0 because h = 0 implies x1 ∈ M = M , a contradiction.
h
Let a ∈ R be such that 0 < a < 1. Then > h.
a
h
By definition of infimum, ∃ x0 ∈ M such that ∥x0 − x1 ∥ ⩽ . Clearly,
a
1
x0 ̸= x1 . Further, as ∥x0 − x1 ∥ > h > 0, we have, > 0.
∥x0 − x1 ∥
x1 − x0
Define xa = . Then xa ∈ V such that ∥xa ∥ = 1.
∥x0 − x1 ∥
Let x ∈ M be arbitrary. Then (∥x1 − x0 ∥ x + x0 ) ∈ M as x, x0 ∈ M
and hence ∥(∥x1 − x0 ∥ x + x0 ) − x1 ∥ ⩾ h.
Now,

x1 x0
∥x − xa ∥ = x − +
∥x1 − x0 ∥ ∥x1 − x0 ∥
1
= ∥(∥x1 − x0 ∥ x + x0 ) − x1 ∥
∥x1 − x0 ∥
h
⩾
∥x1 − x0 ∥
⩾a

Thus, ∃ xa ∈ V such that ∥xa ∥ = 1 and ∥x − xa ∥ ⩾ a ∀x ∈ M .

The Riesz Lemma 4.4.1 states that for any closed proper subspace
M of a normed space V , ∃ points in the unit sphere S(0, 1) = {x ∈
V | ∥x∥ = 1} whose distance from M is as near as we please to 1(but
not 1). There may not be a point, though, whose distance is exactly 1.

We conclude this section with the following required result, proved

using Reisz lemma.
Theorem 4.4.2. In a normed space (V, ∥.∥), if the set S = {x ∈
V | ∥x∥ = 1} is compact then V is finite dimensional.

Proof. We know that, in a metric space, a subset is compact if and only

if every sequence has a convergent subsequence,
Let the set S = {x ∈ V | ∥x∥ = 1} in normed space (V, ∥.∥) be com-
pact. Then every sequence in S must have a convergent subsequence.
Suppose, if possible, V is not finite dimensional. Choose x1 ∈ S.
Then ∥x1 ∥ = 1.
Let V1 be the subspace spanned by x1 . Then V1 is a proper subspace

79
FUNCTIONAL ANALYSIS

of V and V1 is finite dimensional. It follows that V1 is closed, since by

Corollary 4.4.2 if M is any finite dimensional subspace of a normed
space V then M is closed.
Applying Riesz Lemma 4.4.1 to this closed proper subspace V1 of
1
V , we get, ∃ x2 ∈ V such that ∥x2 ∥ = 1 and ∥x2 − x∥ ⩾ ∀ x ∈ V1 .
2
1
=⇒ ∃ x2 ∈ S and ∥x2 − x1 ∥ ⩾ as x1 ∈ V1 .
2
Let V2 be the subspace spanned by x1 , x2 . Then V2 is a proper sub-
space of V and V2 is finite dimensional. So, by Corollary 4.4.2, V2 is
closed. Applying Riesz Lemma 4.4.1 to this closed proper subspace V2
1
of V , we get, ∃ x3 ∈ V such that ∥x3 ∥ = 1 and ∥x3 − x∥ ⩾ ∀ x ∈ V2 .
2
1
=⇒ ∃ x3 ∈ S and ∥x3 − x2 ∥ ⩾ as x2 ∈ V2 .
2
Continuing this argument, we obtain an infinite sequence {xn } of
1
vectors in S such that ∥xn − xm ∥ ⩾ .
2
=⇒ the sequence {xn } can have no convergent subsequence, which con-
tradicts the hypothesis that, S is compact.
Hence, the assumption that V is not finite dimensional must be
wrong and we msut have V is finite dimensional.

4.5 Arzela-Ascoli theorem

You are already familiar with Bolzano-Weierstrass theorem which
states that every bounded sequence of real/complex numbers contains a
convergent subsequence. In this section, you will see something similar
is true for sequence of functions, but in connection with the additional
concept of equicontinuity.

Recall that the set C(X) is the set of all K-valued continuous func-
tions on a compact metric space X where K = R or C. For f, g ∈ C(X),
let d∞ (f, g) = supx∈X {|f (x) − g(x)|}. It is easy to see that d∞ is a met-
ric on C(X) and is called as sup metric on C(X).

In the metric space (X, d), for x ∈ X and r > 0, recall, the set
B(x, r) = {y ∈ X|d(x, y) < r}, called as the open ball about x of ra-
dius r.

Also, recall that a subset E of a metric space X is said to be totally

bounded if ∀ϵ > 0, ∃ x1 , · · · , xn ∈ E such that E ⊂ ∪nj=1 B(xj , ϵ).

Further, recall that for functions, fn , f ∈ C(X), we say that the se-
quence {fn } converges uniformly to f on X if ∀ϵ > 0, ∃ n0 ∈ N(depends

80
CHAPTER 4. BANACH SPACE

only on ϵ) such that ∀ x ∈ X, ∀ n ⩾ n0 , we have, |fn (x) − f (x)| < ϵ.

Now, you will see two kinds of boundedness.

Definition 4.5. Let S be subset of C(X). We say that S is pointwise
bounded on X if S is bounded at each x ∈ X, that is, if ∃ a finite-
valued function ϕ defined on X such that |f (x)| < ϕ(x) ∀ f ∈ S.
Definition 4.6. Let S be subset of C(X). We say that S is uniformly
bounded on X if ∃ a number M such that |f (x)| < M ∀ f ∈ S and
∀ x ∈ X.
It is easy to see that a uniformly bounded subset S of C(X) is
always pointwise bounded on X. The converse holds (as seen in Ascoli’s
theorem, the following result) under a certain condition which we now
introduce.
Definition 4.7. A subset S of C(X) is said to be equicontinuous at
x ∈ X if for every ϵ > 0, ∃ δ > 0 such that for every y ∈ X with
d(x, y) < δ and ∀f ∈ S, we have, |f (x) − f (y)| < ϵ, where δ may
depend on x, but not on f ∈ S.
Here d denotes the metric of X.
A subset S of C(X) is said to be equicontinuous on X if S is equicon-
tinuous at every x ∈ X.

You can verify that the functions belonging to the equicontinuous

collection are uniformly continuous.
Example 29. Define a sequence {fn } on R by fn (x) = sin(x + n).
Show that the family {fn |n ∈ N} is equicontinuous on R.
Solution: Note that |cosθ| ⩽ 1 and for small θ, |sinθ|
⩽ |θ|.
′
x − x′

′ 2n + x + x
Also, sin(n + x) − sin(n + x ) = 2 cos sin
2 2
Let ϵ > 0 be given and x ∈ R. With x′ ∈ R,
Conisder
2n + x + x′ x − x′

′
|fn (x) − fn (x )| = 2 cos sin
2 2
′

x−x
⩽2
2
′
= |x − x |
If |x − x′ | < δ = ϵ then |fn (x) − fn (x′ )| < ϵ ∀ n ∈ N.
Thus, the family {fn |n ∈ N} is equicontinuous at x ∈ R and hence is
equicontinuous on R.
A condition sufficient to ensure that a sequence of continuous func-
tions on compact space X has a uniformly convergent subsequence will
come out of the following result (which is known as Arzela’s theorem).

81
FUNCTIONAL ANALYSIS

Theorem 4.5.1. Let S be subset of C(X) where X is a compact metric

space. Suppose that S is pointwise bounded on X and is equicontinuous
at every x ∈ X. Then
(a) (Ascoli, 1883) S is uniformly bounded on X. In fact, S is
totally bounded in the sup metric on C(X).

(b) (Arzela, 1889) Every sequence in S contains a uniformly

convergent subsequence.
Proof. (a) Let ϵ > 0. Since (X, d) is a compact metric space, we
have, ∃ x1 , · · · , xn ∈ X and positive numbers δ1 , · · · , δn such that
X = ∪{B(xi , δi ) | i = 1, · · · , n}. So, to every x ∈ X there corresponds
at least one xi with x ∈ B(xi , δi ). Also, since S is equicontinuous at
every x ∈ X, ∀f ∈S and ∀ x ∈ X, we have, d(x, xi ) < δi (1 ⩽ i ⩽ n)
which implies |f (x) − f (xi )| < ϵ. Further, since S is pointwise bounded
on X, S is bounded at each x ∈ X. So, ∃ Mi < ∞ (1 ⩽ i ⩽ n) such
that |f (xi )| ⩽ Mi ∀f ∈S. Define M = max{M1 , · · · , Mn } + ϵ. Then
it follows that |f (x)| ⩽ M ∀f ∈S and ∀ x ∈ X. Thus, S is uniformly
bounded on X.
Now, let EM = {k ∈ K| |k| ⩽ M } where K ∈ R or C and for f ∈S,
n n
define e(f ) = (f (x1 ), · · · , f (xn )) ∈ EM . It is easy to see that EM is
totally bounded. Hence we can cover it by a finite union of open balls
of radius ϵ, say V1 , · · · , Vm . If j = 1, · · · , m and Vj ∩ {e(f )|f ∈ S} =
̸ ∅,
choose fj ∈ S such that e(fj ) ∈ Vj .
Claim: S is union of open balls of radius 5ϵ about these f1 , · · · , fm .
Let f ∈S. Then e(f ) ∈ Vj for some j = 1, · · · , m. Since, e(fj ) ∈ Vj and
the radius of Vj′ s is ϵ, we see that |f (xi ) − fj (xi )| < 2ϵ ∀i = 1, · · · , n.
Now, each x ∈ X belongs to some B(xi , δi ), i = 1, · · · , n, which implies
|fj (x) − fj (xi )| < ϵ and |f (x) − f (xi )| < ϵ.
Consider

|f (x) − fj (x)| = |f (x) − f (xi ) + f (xi ) − fj (xi ) + fj (xi ) − fj (x)|

⩽ |f (x) − f (xi )| + |f (xi ) − fj (xi )| + |fj (xi ) − fj (x)|
< ϵ + 2ϵ + ϵ

so that d∞ (f, fj ) = supx∈X {|f (x) − fj (x)|} ⩽ 4ϵ < 5ϵ.

This proves that S is totally bounded in the sup metric d∞ .

(b) We prove (b) using the following results of metric space.

(i) The subset A of metric space X is totally bounded if and only if
every sequence of points of A contains a Cauchy subsequence.
(ii) The sequence {fn } in C(X) converges to f ∈ C(X) (with respect
to sup metric on C(X)) if and only if {fn } converges uniformly to
f on X.
Consider a sequence {fn |fn ∈ C(X), n = 1, 2, · · · } in S. Then by

82
CHAPTER 4. BANACH SPACE

Ascoli theorem, {fn } ⊆ C(X) is totally bounded. Hence, by result

(i), {fn } has a Cauchy subsequence {fnk } (with respect to sup metric
on C(X)). Since, by Theorem 4.3.3, C(X) is complete , the sequence
{fnk } is convergent to some f ∈ C(X). By result (ii), this implies that
{fnk } converges uniformly to f on X and we are done.
(b) (Alternative proof of (b) without using Ascoli theorem)
Consider a sequence {fn |fn ∈ C(X), n = 1.2, · · · } in S. Let A be a
countable dense subset of X. Then S is pointwise bounded on A.
Claim: {fn } has a subsequence {fnk } such that {fnk (x)} converges for
every x ∈ A.
Let {xi }, i = 1, 2, 3, · · · be points of A, arranged in a sequence. Since,
{fn (x1 )} is bounded, ∃ a subsequence, which we shall denote by {f1,k },
such that {f1,k (x1 )} converges as k −→ ∞.
Now, consider sequences S1 , S2 , S3 , · · · , which we represent by the
array

S1 : f1,1 f1,2 f1,3 f1,4 ···

S2 : f2,1 f2,2 f2,3 f2,4 ···
S3 : f3,1 f3,2 f3,3 f3,4 ···
··· ··· ··· ··· ··· ···

and which have the following properties:

(1) Sn is a subsequence of Sn−1 , for n = 2, 3, 4, · · ·
(2) {fn,k (xn )} converges, as k −→ ∞
(3) Order in which the functions appear is the same in each sequence;
i.e., if one function precedes another in S1 , they are in same relation
in every Sn , until one or the other is deleted. Hence, when going
from one row in the above array to the next below, functions may
move to the left but never to the right.

We now go down the diagonal of the array; i.e., we consider the

sequence P: f1,1 f2,2 f3,3 f4,4 · · ·
By (3), the sequence P(except possibly its first n-1 terms) is a subse-
quence of Sn , for n = 1, 2, 3, · · · . Hence, (2) implies that {fn,n (xi )}
converges, as n −→ ∞, for every xi ∈ A.

For convenience, put fni = gi . We shall prove that {gi } converges

uniformly on X.
Let ϵ > 0 be given. As S is equicontinuous, choose δ > 0 such that
d(x, y) < δ implies |fn (x) − fn (y)| < ϵ ∀n. Since A is dense in X
and X is compact, ∃ finitely many points x1 , · · · , xm in A such that
X = ∪{B(xi , δ) | i = 1, · · · , m}.
Since {gi (x)} converges for every x ∈ A, ∃ n0 ∈ N such that |gi (xs )−
gj (xs )| < ϵ whenever i ⩾ n0 , j ⩾ n0 , 1 ⩽ s ⩽ m.

83
FUNCTIONAL ANALYSIS

Also, if x ∈ X then clearly, x ∈ B(xs , δ) for some s, so that |gi (x) −

gi (xs )| < ϵ for every i. If i ⩾ n0 and j ⩾ n0 then it follows that

|gi (x) − gj (x)| = |gi (x) − gi (xs ) + gi (xs ) − gj (xs ) + gj (xs ) − gj (x)|
⩽ |gi (x) − gi (xs )| + |gi (xs ) − gj (xs )| + |gj (xs ) − gj (x)|
<ϵ+ϵ+ϵ

Thus, the subsequence {gi } of {fn } converges uniformly on X.

4.6 LET US SUM UP

1. A Banach space is a normed space in which every Cauchy se-

quence is convergent. In other words, a Banach space is a com-
plete normed space.

2. Every Banach space is a normed space but a normed space need

not be a Banach space.

3. The normed space C[0, 1] = {f : [0, 1] −→ R| f is a continu-

Z function} over R is not a Banach space under norm ∥f ∥ =
ous
1
|f (t)|dt.
0

4. Prove that the vector space C[a, b] = {f : [a, b] −→ R(or C)| f is

a continuous function} over R(or C) is not a Banach space under
Z b 1/p
p
norm ∥f ∥ = |f (t)| dt where 1 ⩽ p < ∞
a

5. The vector space C(X) = {f : X −→ R(or C)| f is bounded

continuous function on X} over R(or C) is a Banach space under
norm ∥f ∥ = sup{|f (x)|}.
x∈X

6. The vector space R = {x|x ∈ R} over R is a Banach space under

the norm ∥x∥ = |x|=absolute value of x ∈ R.

7. The vector space Cn = {(x1 , · · · , xn ) | xi ∈ C} over C is a Banach

space under the norms
 n 1/p
X
p
|x | if 1 ⩽ p < ∞


i
∥x∥p = i=1

max{| x1 |, · · · , | xn |} if p = ∞


84
CHAPTER 4. BANACH SPACE

X∞
p
8. The sequence space l = {x1 , · · · , xn , · · · } | xi |p < ∞ and
i=1

xi ∈ C over C is a Banach space under the norms

 ∞ 1/p
X
p
|xi | if 1 ⩽ p < ∞


∥x∥p = i=1

sup{| x1 |, · · · , | xn |, · · · } if p = ∞


9. A sequence {xk } in a normed space (V, ∥ ∥) is said to be summable

to the sum s if the sequence {sn } of the partial sums of the series
X∞
xk converges to s ∈ V . i.e. ∥sn − s∥ −→ 0 as n −→ ∞. In
k=1
∞
X
this case, we write, s = xk .
k=1

10. The sequence {xk } in a normed space (V, ∥ ∥) is said to be abso-

∞
X
lutely summable if ∥xk ∥ < ∞.
k=1

11. A normed space (V, ∥ ∥) is a Banach space if and only if every

absolutely summable sequence in V is summable.
12. The vector space Lp (E) = {f : E −→ R|f is Lebesgue measur-
able function on E and | f |p is Lebesgue integrable over E} over
R is a Banach space under norms
 Z 1/p
 p
|f | if 1 ⩽ p < ∞

∥f ∥p = E
ess sup|f |

if p = ∞
where ess supE |f (x)| = inf {m > 0 | |f (x)| ⩽ m a.e. on E} and
L∞ (E) = the class of all those measurable functions f defined on
E which are essentially bounded on E with ess sup|f | < ∞.
13. (Riesz-Fischer theorem): Lp spaces are Banach spaces where
1 ⩽ p ⩽ ∞.
14. Let M be a closed subspace of a Banach space (V, ∥ ∥). For
each coset x + M in quotient space V /M , define ∥x + M ∥ =
inf {∥x + m∥ | m ∈ M }. Then V /M is a Banach space under the
norm ∥x + M ∥.
15. Let M be a closed subspace of a normed space (V, ∥.∥). If M and
V /M are Banach spaces then V is a Banach space.
16. Every complete subspace of a normed space is closed.

85
FUNCTIONAL ANALYSIS

17. Two norms ∥ ∥1 and ∥ ∥2 on a normed space X are said to be

equivalent and written as ∥ ∥1 ∼ ∥ ∥2 , if ∃ positive real numbers
a and b (independent of x ∈ X) such that
a ∥x∥1 ⩽ ∥x∥2 ⩽ b ∥x∥1 ∀ x ∈ X.
18. On a finite dimensional normed space X, any two norms are
equivalent.
19. If V is a finite dimensional normed space, then V is complete.
20. (Riesz Lemma): Let M be a closed proper subspace of a normed
space V and let a ∈ R be such that 0 < a < 1. Then ∃ a vector
xa ∈ V such that ∥xa ∥ = 1 and ∥x − xa ∥ ⩾ a ∀x ∈ M .
21. In a normed space (V, ∥.∥), if the set S = {x ∈ V | ∥x∥ = 1} is
compact then V is finite dimensional.
22. Let S be subset of C(X). We say that S is pointwise bounded
on X if S is bounded at each x ∈ X, that is, if ∃ a finite- valued
function ϕ defined on X such that |f (x)| < ϕ(x) ∀ f ∈ S.
23. Let S be subset of C(X). We say that S is uniformly bounded on
X if ∃ a number M such that |f (x)| < M ∀ f ∈ S and ∀ x ∈ X.
24. A subset S of C(X) is said to be equicontinuous at x ∈ X if for
every ϵ > 0, ∃ δ > 0 such that for every y ∈ X with d(x, y) < δ
and ∀f ∈ S, we have, |f (x) − f (y)| < ϵ, where δ may depend on
x, but not on f ∈ S. Here d denotes the metric of X.
25. Let S be subset of C(X) where X is a compact metric space.
Suppose that S is pointwise bounded on X and is equicontinuous
at every x ∈ X. Then
(a) (Ascoli, 1883) S is uniformly bounded on X. In fact, S
is totally bounded in the sup metric on C(X).
(b) (Arzela, 1889) Every sequence in S contains a uniformly
convergent subsequence.

4.7 Chapter End Exercise

1. Show that the vector space C = {z|z ∈ C} over C is a Banach

space under the norm ∥z∥ = |z|=absolute value of z ∈ C.
2. Show that the vector space Rn = {(x1 , · · · , xn ) | xi ∈ R} over R is
X n 1/2
2
a Banach space under the (Euclidean) norm ∥x∥ = |xi | .
i=1

86
CHAPTER 4. BANACH SPACE

3. Show that the vector space Rn = {(x1 , · · · , xn ) | xi ∈ R} over R

X n 1/p
p
is a Banach space under the norm ∥x∥ = |xi | where
i=1
1 ⩽ p < ∞.

4. Show that the vector space Rn = {(x1 , · · · , xn ) | xi ∈ R} over R

is a Banach space under the norm ∥x∥ = max{| x1 |, · · · , | xn |}.

5. Show that the normed space C([a, b]) = {f : [a, b] −→ R|f is

Z b 1/2
2
continuous function} over R under the norm ∥f ∥ = |f |
a
is not a Banach space.

6. Let (X, ∥ ∥X ) and (Y, ∥ ∥Y ) be normed spaces. Then prove that

X × Y is a Banach space under the norm ∥(x, y)∥ = ∥x∥X + ∥y∥Y .

7. Let V be a non-zero normed space and let S = {x ∈ V | ∥x∥ ⩽ 1}.

Prove that V is complete if and only if S is complete.

8. Prove that every finite dimensional normed space is Banach and

hence deduce each finite dimensional subspace of a normed space
is closed.

9. Give a counter example to show that any two norms on an infinite

dimensional normed space are not equivalent.

10. Show that the p-norms on Rn are equivalent where 1 ⩽ p ⩽ ∞.

11. Let M be a closed proper subspace of the normed space V . Then

prove that for every real number a > 0, ∃ an element y ∈ V with
∥y∥ = 1 such that ∥x − y∥ > 1 − a ∀ x ∈ M .

12. Let M be a closed proper subspace of a normed space V . Then

prove that for each a ∈ (0, 1), ∃ a point xa in V but not in M
(not necessarily unique) such that ∥xa ∥ = 1 and dist(xa , M ) =
inf ∥xa − y∥ ⩾ a.
y∈M

13. Let V be a finite dimensional normed space and r > 0. Then

prove that the closed ball B[0; r] = {x ∈ V | ∥x∥ ⩽ r} is compact.

14. Let V be a normed space such that the closed ball B[x0 ; r] = {x ∈
V | ∥x − x0 ∥ ⩽ r} is compact for some x0 ∈ V and r > 0. Then
prove that V is finite dimensional.

15. Prove that in a finite dimensional normed space V , any proper

subset M of V is compact if and only if M is closed and bounded.

16. Let V be a normed space. Prove that the closed unit ball in V is
compact if and only if V is finite dimensional.

87
FUNCTIONAL ANALYSIS

17. Prove that a normed space V is finite dimensional if and only if

every bounded closed subset in V is compact.

18. If M is a compact subset of a Banach space V then show that M

is also compact.

19. If M is any finite dimensional subspace of a normed space V , then

M is closed.

20. Let fn ∈ C(X) for n = 1, 2, 3, · · · where X is a compact metric

space. If {fn } converges uniformly on X then prove that {fn } is
equicontinuous on X.

21. If {fn } is a pointwise bounded sequence of complex functions on

a countable set E, then prove that {fn } has a subsequence {fnk }
such that {fnk (x)} converges for every x ∈ E.

22. Let X be a compact metric space. Prove that a closed subspace

of C(X) is compact if and only if it is uniformly bounded and
equicontinuous.

23. Show that the family {sin nx}∞

n=1 is not an equicontinuous subset
of C[0, π].

88
Chapter 5

BOUNDED LINEAR
TRANSFORMATIONS
AND DUAL SPACES

Unit Structure :
5.1 Introduction
5.2 Objective
5.3 Definitions, notations, theorems
5.4 Separable spaces
5.5 LET US SUM UP
5.6 Chapter End Exercise

5.1 Introduction
The bounded linear transformations on normed linear spaces are
important operators , that satisfy many properties as a function be-
tween two metric spaces like continuity and their collections B(X, Y )
can be made into a normed linear space under pointwise addition and
scalar multiplication.Completeness of the normed space B(X, Y ) is in-
herited via the completeness of the space Y .The Dual space of X is a
complete metric space even if X is complete or not and hence it satisfies
the properties of being a complete space.The significance of dual spaces
of lp , Lp , Rn is that it is useful to know the general form of bounded
linear functionals on spaces of practical importance.For Hilbert spaces,
Riesz’s theorem elucidates the form of such bounded linear functionals
in simple manner.The separable spaces are somewhat simpler than the
′
non separable spaces.The separability of the dual space X implies that
the space X is separable.

89
FUNCTIONAL ANALYSIS

5.2 Objectives

After going through this Chapter, you will be able to

• Define a bounded linear transformations between two normed linear
spaces.
• Characterise the bounded linear transformations as continuous func-
tions
• Identify the algebraic structure of the bounded linear transformations
as normed linear space
• Define the dual space of X and describe its properties like complete-
ness and seperability

5.3 Definitions, notations, theorems

Definition 1: A metric space is a pair (X, d), where X is a set and

d is a metric on X, that is d is a distance function defined on X × X
such that for all x, y, zϵX we have
(1) d is real valued , finite and nonnegative.
(2) d(x, y) = 0 if and only if x = y
(3) d(x, y) = d(y, x)
(4) d(x, y) ≤ d(x, z) + d(z, y)
Examplep1: Euclidean space Rn with metric d defined as
d(x, y) = (x1 − y1 )2 + ... + (xn − yn )2
Example 2: Sequence space l∞ with metric d defined as
d(x, y) = sup{|xi − yi | : i ≥ 1}
1
Example 3: lp space with metric d defined as d(x, y)=( ∞ p p
P
i=1 |xi −yi | )

Definition 2: Let X, Y be normed linear spaces and and T be a

linear transformation of X into Y then T is said to be bounded linear
transformation, if there exists a real number K ≥ 0 such that
||T (x)|| ≤ K||x|| for every xϵ X. The set of all continuous or bounded
linear transformations from X into Y is denoted by B(X,Y)

Example 1: The identity operator I : X −→ X defined by I(x) = x

on a nonzero normed space X is bounded linear operator with ||I(x)|| =
||x|| ≤ K||x|| with K = 1
Example 2: The 0 operator 0 : X −→ Y defined by 0(x) = 0Y is
bounded linear operator with ||0(x)|| = ||0Y || = 0 ≤ 0||x|| with K = 0

Now we shall see equivalent characterisations of a bounded linear trans-

formation T existing between two normed spaces.

90
CHAPTER 5. BOUNDED LINEAR TRANSFORMATIONS AND DUAL
SPACES

Theorem 5.3.1. Let X, Y be normed linear spaces and T is a linear

transformation of X into Y . Then the following statements are equiv-
alent
(1)T is continuous
(2)T is continuous at the origin, in other words T(xn )−→ 0 as xn −→
0
(3)There exists a real number K ≥ 0 such that ||T (x)|| ≤ K||x|| for
every
xϵ X
(4) T carries the closed unit sphere in X to a bounded set in Y
Proof: (1) ⇒ (2)
Suppose that T is continuous . Since we have T(0)=0 , therefore T is
continuous at the point x=0. If T is continuous at the origin then xn
−→ x if and only if xn − x −→0. This implies that T (xn − x) −→0.This
is true if and only if T (xn ) −→T(x). So T is continuous.
(2) ⇒ (3)
Assume that for each positive integer n , we can find a point xn in X
such that ||T (xn )|| ≥ n||xn ||. This implies that ||T (xn /nxn )|| > 1. Put
yn = xn /n||xn ||. Then observe that yn →0 as n → ∞ but T (yn ) ↛0.
Therefore, T is not continuous at the origin.This is a contradiction to
our hypothesis. Therefore There exists a real number K ≥ 0 such that
||T (x)|| ≤ K||x|| for every
xϵX
(3) ⇒ (4)
Let x be any point belonging to the closed unit sphere in X. This implies
that ||x|| ≤ 1.Hence,by hypothesis ||T (x)|| ≤ K||x|| ≤ K. Therefore, by
definition, T carries the closed unit sphere in X to a bounded set in Y.
(4) ⇒ (1)
We first show that (4) ⇒ (3). If x = 0 then T (x)=0, beacuse T is Linear
transformation. Hence ||T (x)|| ≤K||x||. If x ̸=0 then x/||x|| = 1. Put
y = x/||x|| , then y belongs to the closed unit sphere and therefore by
hypothesis, ||T (y)|| ≤K for some real number K≥0. This implies that
||T (x/||x||) ≤K. Thus, ||T (x)|| ≤K||x||. Therefore, ||T (x)|| ≤K||x||for
every x ϵX. Now, if xn is any convergent sequence in X such that xn
−→ x, then ||T (xn − x)|| ≤K||xn − x|| implies that T (xn ) − T (x) −→0
as xn − x −→0. Hence T is continuous.
Note:(1) T is continuous iff T is bounded
(2) If T is continuous then T carries the closed unit sphere to a bounded
set in Y , in this case, we denote the norm of T by ||T || and it is defined
as
||T ||= sup { ||T (x)|| : ||x|| ≤1 } .
We also have ||T ||= inf {K : K≥0 and ||T (x)|| ≤K||x||. From this, we
conclude that ||T (x)|| ≤ ||T || ||x|| for all x

91
FUNCTIONAL ANALYSIS

Now we shall see that B(X, Y ) forms a normed linear space and it
is complete , when Y is complete space.

Theorem 5.3.2. If X, Y are normed linear spaces, then B(X,Y) is

a normed linear space with respect to pointwise addition and scalar
multiplication and the norm defined as ||T ||= sup { ||T (x)|| : ||x|| ≤1
}.Further, if Y is a Banach space then B(X,Y) is also a Banach space.
Proof: To show that B(X,Y) is a normed linear space, let T, U be any
two linear transformations belonging to B(X,Y), then T+U is defined
as (T + U )(x):=T (x) + U (x) and for any scalar α in F , we have (α
T )(x):=α T (x).Therefore we have (U+T)(xn )=U(xn )+T(xn ) and (α
T )(xn ):=α T (xn ).
Therefore (T + U )(xn ) −→0 and (α T )(xn ) −→0 as xn −→0
Hence, T+U and αT both are continuous at the origin. This implies
that T+Uϵ B(X,Y) and αTϵ B(X,Y)
We have (T+U)(x)=T(x)+U(x)=U(x)+T(x)=(U+T)(x), therefore vec-
tor addition is commutative.
For any S,T,Uϵ B(X,Y), we have [(S+T)+U](x)=(S+T)(x)+U(x)=
S(x)+T(x)+U(x)=[S+(T+U)](x) for every x ϵ X.Therefore, (S+T)+U=S+(T+U)
There exist 0 linear transformation in B(X,Y) defined as 0(x)=0Y for
every x ϵ X.
For every TϵB(X,Y), there exist an additive inverse -T ϵB(X,Y) such
that T+(-T)=0.
we have (-T)(x)=-T(x) for every x ϵX.
Scalar multiplication is associative and distributive.For all α, β ϵ F and
T, U ϵ B(X,Y), [α(βU)](x)=α(βU(x))=α βU(x)= (α β)U(x).This im-
plies that α(βU)=(α β)U. [(α+β)U](x)=(α+β)U(x)=αU(x)+βU(x)=(αU+βU)(x).
Hence, (α+β)U=αU+βU. Now [α(T+U)](x)=α(T+U)(x)=αT(x)+αU(x)=
(αT+αU)(x). Therefore, we have α(T+U)=αT+αU.
Further, we have (1.U)(x)=1.U(x)=U(x).Hence 1.U=U.
Now we shall show that for Tϵ B(X,Y), ||T ||= sup { ||T (x)|| : ||x|| ≤1
} is norm on the linear space B(X,Y).
(a) For every Tϵ B(X,Y), we have sup { ||T (x)|| : ||x|| ≤1 }≥0, there-
fore ||T || ≥0
(b)||T ||=0 if and only if sup { ||T (x)|| : ||x|| ≤1}=0. This is possible
iff ||T (x)||=0. This is true iff T(x)=0 and therefore T=0.
(c)|| αT||=sup { ||(αT)(x)|| : ||x|| ≤1}=| α |sup { ||T (x)|| : ||x|| ≤1}=|
α | ||T || for every α ϵF and For every Tϵ B(X,Y)
(d)||T +U ||= sup { ||(T+U)(x)|| : ||x|| ≤1}=sup { ||T(x)+U(x)|| : ||x||
≤1}
≤sup { ||T(x)|| +||U(x)|| : ||x|| ≤1}≤ sup { ||T(x)|| :||x|| ≤1}+sup {
||U(x)|| :||x|| ≤1}=||T ||+||U ||

92
CHAPTER 5. BOUNDED LINEAR TRANSFORMATIONS AND DUAL
SPACES
Therefore, ||T + U || ≤ ||T ||+||U || for every T, U ϵ B(X,Y)
Next, we show that B(X,Y) is complete, when Y is complete.
Let Tn be any Cauchy sequence of linear transformations in B(X,Y).
For any vector x ϵ X , we have ||Tn (x)-Tm (x)||=|| (Tn − Tm )(x)|| ≤ ||
Tn -Tm || —— x ||.This implies that {Tn ( x)} is a Cauchy sequence
in Y.Since Y is Complete therefore, there exist a vector T(x) in Y
such that Tn ( x)−→T(x) .This defines a function T: X−→Y by x
−→T(x).This function is a linear transformation from X into Y , for
if x1 ,x2 ϵ X, we have Tn (x1 + x2 )=Tn (x1 )+Tn (x2 ) and Tn (α x)=α
Tn (x).Hence T(x1 +x2 )=T(x1 )+T(x2 ) and T(α x)=αT(x).Now we show
that T is continuous and Tn −→T as n −→ ∞
It is enough to show that T is bounded linear transformation.
Consider ||T (x)||=|| limn→∞ Tn (x)||=limn→∞ ||Tn (x)|| ≤ sup( ||Tn ||
||x||)= (sup||Tn ||) ||x||. Since the norms of the terms of the Cauchy
Sequence in a normed linear space is a bounded set, therefore there
exists K=sup||Tn || ≥0 such that
||T (x)|| ≤K ||x||.Hence, T is a bounded linear transformation.Now we
show that ||Tn − T || −→0.Let ϵ >0 be any number , let N be a positive
integer such that n, m ≥N ⇒ ||Tn − Tm || < ϵ. Now, if ||x|| ≤1 and m, n
≥N, then
|| Tn (x)-Tm (x)||=||(Tn -Tm )(x)|| ≤ ||Tn − Tm || ||x|| ≤ ||Tn − Tm || < ϵ
Now hold n fixed and letting m −→ ∞, we obtain
|| Tn (x)-Tm (x)|| −→ || Tn (x)-T(x)||.This implies that || Tn (x)-T(x)|| <
ϵ for all n ≥N and every x such that ||x|| ≤1.Hence, ||Tn − T || < ϵ for
every n ≥N.Therefore, we have ||Tn − T || −→0 as n −→ ∞.
Definition 3: If X is any normed linear space then then the set of
all continuous linear transformations from X into R or C is denoted
by B(X,R) or B(X,C), according to X is real or complex vector space.
′ ′
Denote it by X , it is called as the dual space of X.The elements of X
are called as continuous linear functionals.
Note: (1) functional defined on normed linear space X is a scalar- val-
ued continuous linear functional defined on X.
′
(2) X is a normed linear space with norm defined by
||f ||:=sup{ |f||x||
(x)|
: x ϵX,x ̸=0 }=sup{ |f (x)|:||x|| = 1 }
′
(3) Since R and C are complete normed linear spaces therefore X is a
Banach space.
Definition 4: A bijective linear operator from a normed space X onto
the normed space Y is called as an isomorphism if it preserves the
norm, that is, for every x ϵX, ||T (x)||=||x||. In this case, X is said to
be isomorphic to Y and X,Y are called isomorphic normed spaces.
Theorem 5.3.3. The dual space of Rn is Rn
Proof: Rn is a normed linear space with p the norm defined as follows:
n
For every x=(x1 ,x2 ,...,xn )ϵ R , ||x||= x21 + x22 + ... + x2n . We recall
the theorem , which states that if the dimension of a normed linear space

93
FUNCTIONAL ANALYSIS

is finite, then every linear operator on X is bounded. Therefore, we have

′
Rn =Rn ∗ .Given any f ϵ Rn , if {e1 , e2 , ..., en }Pis a standardPbasis for
Rn , then f (x) = f (x1 e1 + x2 e2 + ... + xn en )= ni=1 xi f (ei )= ni=1 xi γi ,
where γi =f (ei ) for Pevery i.Therefore,
Pn by the Cauchy-Schwarz inequality
we have |f (x)| ≤ i=1 |xi γi | ≤ ( i=1 xi ) ( i=1 γi ) =||x|| ( ni=1 γi2 )1/2
n 2 1/2
Pn 2 1/2 P
|f |
≤ ( ni=1 γi2 )1/2 .Now taking supremum over all x of norm
P
Hence ||x||
1, we get, ||f || ≤ ( ni=1 γi2 )1/2 , because of the equality obtained
P
Pn 2for
x=(γ1 , γ2 , ..., γn ) in the above inequality, we must have ||f ||=( i=1 γi )1/2 .This
implies that the norm of f is the Euclidean norm on Rn
′
Hence the mapping ϕ: Rn −→ Rn , defined as ϕ(f )=(γ1 ,γ2 ,...,γn ) is
norm preserving bijective linear map, hence it is an isomorphism.

Here, we shall try to identify dual spaces of some of the normed linear
spaces.
Theorem 5.3.4. The dual space of l1 is l∞ .
Proof: Consider a Schauder basis for l1 , namely (ek ), where ek =(δkj ),
where δkj =1 if j = k and δkj =0, if j ̸= k.Then every x ϵ l1 can be
uniquely
represented as x= ∞
′
l1 be any linearPfunctional.
P
k=1 xk ek . Let f ϵ
Since f is linear and bounded , therefore, we have f (x)= ∞ k=1 xk γk ,
where γk = f (ek ).Here the γk =f (ek ) are uniquely determined by f .
Also, ||ek ||=1 and |γk |=|f (ek ) ≤ ||f || ||ek ||=||f ||.Taking supremum
on both th sides, we get, sup{|γk | : k ≥ 1} ≤ ||f ||...(1)
Hence,P(γk ) ϵ l∞ . Further, if d=(δk ) ϵ l∞ then define g on l1 as follows:
g(x)= ∞ k=1 xk δk , where x=(xk ) ϵ l1 . Observe that g is linear as well
as bounded P map, because we have
|g(x)| ≤ ∞
P∞
k=1 |xk ||δk | ≤ sup{|δk | : k ≥ 1} k=1 |xk |=||x|| sup{|δk | :
1
k ≥ 1}.Hence g ϵ l .Finally, we prove P that the norm of f is the norm
on the space l∞ .Since we have, f (x)= ∞ k=1 xk γk , where γk = f (ek ),
therefore,P
|f (x)|=| ∞
P∞
k=1 xk γk | ≤ sup{|γk | : k ≥ 1} k=1 |xk |=||x|| sup{|γk | : k ≥
1}.Thus, we get
|f (x)|
||x||
≤ sup{|γk | : k ≥ 1}.Taking supremum over all x of norm 1, we
get, ||f || ≤ sup{|γk | : k ≥ 1}...(2). From (1) and (2), we conclude that
||f ||=sup{|γk | : k ≥ 1}, which is the norm on l∞ .This shows that the
′
bijective linear mapping of l1 onto l∞ defined by f −→ (γk ) is an
isomorphism.
Theorem 5.3.5. The dual space of lp is lq for 1 < p < ∞ and p1 + 1q = 1

Proof: A Schauder basis for lp is (ek ),Pwhere ek = (δkj ). Then every x

ϵ lp has a unique representation x=P ∞ p′
k=1 xk ek . Let f ϵ l , since f is
linear and bounded, therefore f (x)= ∞ k=1 xk γk , where γk =f (ek )...(*)
q
Let q be the conjugate of p and define yn =(βk ) with βk = |γγkk| , if k
(n) (n)

94
CHAPTER 5. BOUNDED LINEAR TRANSFORMATIONS AND DUAL
SPACES
(n)
≤ n and γk ̸=0 and βk =0, if k > n or γk = 0. By substituting this in
(*) we get,
f (yn )= ∞
P (n) Pn q
k=1 βk γk = k=1 |γk |
Using the definition of βkn and (q − 1)p = q, we obtain,
≤ ||f || ||yn ||=||f || ( ∞ =||f || ( ∞
P (n) p 1/p P (q−1)p 1/p
f (yn ) P k=1 |βk | ) k=1 |γk | ) =
n q 1/p
||f || ( k=1 |γk |P)
n
|γk |q ≤ ||f || ( nk=1 |γk |q )1/p .Therefore,
P
Hence,
Pn f (yqn )= k=1P
( k=1 |γk | )1−1/p =( nk=1 |γk |q )1/q P ≤ ||f ||.Since n was arbitrary, there-
fore letting n −→ ∞, we obtain ( ∞ q 1/q
k=1 |γk | ) ≤ ||f ||...(**)
q
Hence (γk ) ϵ l .
Conversely, for any (ηk ) ϵ lq , we can define corresponding bounded
p
linear functionalP∞g on l as follows:
g(x)=g(ψk )= k=1 ψk ηk , where x=(ψk ) ϵ lp . Then by the Hölder in-
equality,P we have
|g(x)|=| ∞
P∞ p 1/p
( ∞ q 1/q
P
k=1 ψk ηk | ≤ ( k=1′ |ψk | ) k=1 |ηk | ) .Thus g is linear
p
and bounded. Hence, g ϵ l .Now, we prove P∞ that the norm P∞of f isp the
q 1/p
norm
P∞ on the space l . therefore, |f (x)|=| x
k=1 k kγ | ≤ ( k=1 |xk | )
∞
( k=1 |γk |q )1/q =||x|| ( k=1 |γk |q )1/q .
P

Therefore, we have |f||x|| (x)|

≤( ∞ q 1/q
P
k=1 |γk | ) .Now taking supremum over
P∞
all x ofPnorm 1, we obtain , ||f || ≤ ( k=1 |γk |q )1/q . From (**), we have
||f ||=( ∞ q 1/q
k=1 |γk | ) ′
Therefore, the mapping of lp onto lq defined by f −→ (γk ) is linear,
bijective and norm preserving. Hence it is an isomorphism.

Its practically important to know the general form of bounded linear

fucntionals on various spaces.For general Banach spaces, such formulas
and their derivation can sometimes be complicated. But, for Hilbert
space the situation is simple as described by the following result:
Representation of functionals on Hilbert spaces:

Riesz’s representation of bounded linear functionals on Hilbert spaces:

Theorem 5.3.6. Every bounded linear functional f on a Hilbert space

H can be represented in terms of an inner product as follows:
f (x)=< x, z >, where z is uniquely determined by f and ||z|| = ||f ||.

Proof: We prove the following claims

(a) f has representation as f (x)=< x, z >
(b) z is uniquely determined by f
(c) ||z||=||f ||

(a) If f =0 then f (x)=0=< x, 0 > and ||z||=||f ||=0 for z=0. There-
fore, assume that f ̸=0. In this case, z ̸=0 since, otherwise f =0. Now
< x, z >= 0 for every x in the nullspace of f , denoted by N (f ).So, con-

95
FUNCTIONAL ANALYSIS

sider N (f ) and its orthogonal complement N (f )⊥ . Since f is bounded

functional, therefore N (f ) is a closed vector subspace of H.Further f
̸=0 implies that N (f )̸=H. So that N (f )⊥ ̸=0 by the following theorem:
Let Y be any closed subspace of a Hilbert space H .Then H=Y⊕ Y ⊥
Hence N (f )⊥ contains some z0 ̸=0.Put v = f (x)z0 − f (z0 )x; where, x
ϵ H is arbitrary.Applying f , we obtain f (v)=f (x)f (z0 )−f (z0 )f (x)=0.
This shows that v ϵ N (f )
Since z0 ⊥ N (f ), we have 0=< v, z0 >=< f (x)z0 − f (z0 )x, z0 >
=f (x) < z0 , z0 > −f (z0 ) < x, z0 >=f (x)||z0 ||2 − f (z0 ) < x, z0 >
Since ||z0 ||2 ̸=0, we can solve for f (x).The solution is given by
f (x)= <zf 0(z,z00)> < x, z0 >. Therefore

z= <zf 0(z,z00)> z0 . Since x was arbitrary , therefore f (x)=< x, z >.

(b) We prove that z is uniquely determined in (a). Suppose that for
all x ϵ H, f (x)=< x, z1 >=< x, z2 >. Then < x, z1 − z2 >= 0 for all
x.Choose x = z1 −z2 , we have < x, z1 −z2 >=< z1 −z2 , z1 −z2 >=||z1 −
z2 ||2 =0.
Hence z1 − z2 = 0, this implies that z1 =z2
(c) Now we show that ||z||=||f ||. Assume that f ̸=0 then z ̸=0
Put x = z in f (x)=< x, z >. This implies that ||z||2 =< z, z >=f (z)
≤ ||f || ||z||
Dividing by ||z|| ̸=0, we get ||z|| ≤ ||f ||. It remains to show that
||f || ≤ ||z||, from Schwarz inequality, we have |f (x)|=| < x, z > | ≤
||x||||z||
This implies that ||f ||=sup{| < x, z > | : ||x|| = 1} ≤ ||z||.Hence
||f ||=||z||.

Definition 5: Let (X, A, µ) be measure space and 1 ≤ p < ∞.The

space Lp (X) consists ofR equivalence classes of measurable functions
f : X −→ R such that |f |p dµ < ∞ , where two measurable func-
tions are equivalent, ifR they are equal µ a.e.The Lp norm of f ϵLp (X) is
defined by ||f ||Lp = ( |f |p dµ)1/p

We say that fn −→ f in Lp if ||f − fn ||Lp −→ 0.For example, the

characteristic function χQ of the rationals on R is equivalent to 0 in
Lp (R)
p
Example 1: If N is equipped with counting measure,
P∞ thenp L (N) con-
sists of all sequences {xn ϵR : nϵN} such that n=1 |xn | < ∞ with
norm
||(xn )||Lp = ( ∞ p 1/p
P
n=1 |xn | )

Definition 6: Let (X, A, µ) be a measure space.The space L∞ (X)

consists of pointwise a.e.- equivalence classes of essentially bounded
measurable functions f : X −→ R with norm ||f ||L∞ =ess sup |f |,
where for any measurable function f : X −→ R, the essential supre-

96
CHAPTER 5. BOUNDED LINEAR TRANSFORMATIONS AND DUAL
SPACES
mum of f on X is
ess supf =inf{aϵR : µ{xϵX : f (x) > a} = 0} , equivalently
ess sup f =inf{sup g: g = f pointwise a.e}

Now we shall establish the isomorphism between Lq (X) and the dual
space of Lp (X) for 1 < p < ∞.
Proposition 5.3.1. Suppose that (X,R A, µ) be a measure space and
1 < p ≤ ∞. If f ϵ Lq (X), then F (g)= f gdµ defines a bounded linear
functional F : Lp (X) −→ R with ||F ||Lp′ =||f ||Lq .If X is σ-finite then
the same results hold for p = 1.
Proof: Using Hölder’s inequality, we have for 1 ≤ p ≤ ∞ that
|F (g)| ≤ ||f ||Lq ||g||Lp .This implies that F is a bounded linear func-
tional on Lp with ||F ||Lp′ ≤ ||f ||Lq .In proving the reverse inequality, we
may assume that f ̸=0, otherwise the result is trivial.
First assume that 1 < p < ∞.Let g = (sgnf )( ||f|f||L| q )q/p , then, g ϵ Lp ,
since f ϵ Lq , and ||g||Lp =1.Also, since pq =q−1, F (g)= (sgnf )f ( ||f|f||L| q )q−1
R

dµ=||f ||Lq
Since ||g||Lp = 1, we have ||F ||Lp′ ≥ |F (g)| so that ||F ||Lp′ ≥ ||f ||Lq .
If p = ∞, we get the same conclusion by taking g=sgn(f )ϵL∞ .Therefore,
in these cases the supremum defining ||F ||Lp′ is actually attained for
suitable function g.
Now, suppose that p = 1 and X is σ-finite. For ϵ > 0,
let A = {xϵX : |f (x)| > ||f ||L∞ − ϵ}. Then 0 < µ(A) ≤ ∞. Moreover,
since X is σ-finite, there is an increasing sequence of sets An of finite
measure whose union is A such that µ(An ) −→ µ(A), so we can find a
subset B⊂A such that 0 < µ(B) < ∞.
Let g=sgn(f ) χ(B)µ(B)
. Then gϵ L1 (X) and ||g||L1 =1 and
1
R
F (g) = µ(B) B
|f |dµ ≥ ||f ||L∞ − ϵ. This implies that , ||F ||L1′ ≥
||f ||L∞ − ϵ and therefore we have ||F ||L1′ ≥ ||f ||L∞ , this is because ϵ
was arbitrary.
′
This result shows that the map F :Lq (X) −→ Lp (X) defined by F (g)= f gdµ
R
′
is an isometry from Lq into Lp
F is onto, when 1 < p < ∞, so that every bounded linear functional
on Lp arises in this way from an Lq function.
Theorem 5.3.7. Let (X, A, µ) be a measure space.If 1 < p < ∞ then
q p ′ R
F :L (X) −→ L (X) defined by F (g)= f gdµ is an isometric isomor-
phism of Lq (X) onto the dual space of Lp (X).
Proof: Suppose that X has a finite measure and let F :Lp −→ R
be a bounded linear functional on Lp (X). If A ϵA, then χA ϵLp (X).
S∞X has finite measure, define ν : A −→ R by ν(A) = FP
Since, (χA ). If
A= i=1 Ai is a disjoint union of measurable sets , then χA = ∞
i=1 χAi
and the dominated convergence theorem implies that

97
FUNCTIONAL ANALYSIS

||χA − ∞
P
i=1 χAi ||Lp −→ 0 as n −→ ∞.Hence, since F is continuous
linear functional on LpP ,
ν(A) = F (χA ) = F ( ∞
P∞ P∞
i=1 χA i
)= i=1 F (χA i
)= i=1 ν(Ai ). This im-
plies that ν is a signed measure on (X, A).
If µ(A) = 0, then χA is equivalent to 0 in Lp and therefore ν(A) = 0,
by the linearity of F.Thus ν is absolutely continuous with respect to µ.
By the Radon-Nikodym theorem,that is stated as follows:
Let ν be a σ−finite signed measure and µ be a σ− finite measure on
a measurable space (X, A).Then there exists unique σ− finite signed
measures νa , νs such that ν = νa + νs , where νa << µ and νs ⊥ µ.
Further, there exists a measurable function f : X R −→ R uniquely de-
fined upto µ a.e. equivalence, such that νa (A) = A f dµ for every AϵA,
where the integral is well defined as an extended real number.
The decomposition ν = νa + νs is called the Lebesgue decomposition
of ν and the representation of an absolutely continuous signed measure
ν << µ as dν = f dµ is the Radon-Nikodym theorem.We call the func-
tion f here as the Radon-Nikodym derivative of ν with respect to µ
dν
and denote it by f = dµ .

Thus there R is a function f : X −→ R such that dν = f dµ and

F (χA ) = f χA dµ forRevery A ϵ A.Hence, by the linearity andRbound-
edness of F , F (ϕ) = f ϕdµ for all simple functions ϕ, and | f ϕdµ|
≤ M ||ϕ||Lp , where M = ||F ||Lp′ .
Taking ϕ = sgnf , which is a simple function, we observe that f ϵ
L1 (X).We may then extend the integral of f against bounded func-
tions by continuity. If g ϵ L∞ (X), then from the following theorem:[
Suppose that (X, A, µ) is a measure space and 1 ≤ p ≤ ∞. Then the
simple functions that belong to Lp (X) are dense in Lp (X).], there exist
a sequence of simple fucntions ϕn with |ϕn | ≤ |g| such that ϕn −→ g
in L∞ , and therefore, also in Lp . Since |f ϕn | ≤ ||g||L∞ |f | ϵ L1 (X), the
dominated convergence theoremR and the continuity R of F implies
R that
F (g) = limn→∞ F (ϕn )=limn→∞ f ϕn dµ= f gdµ and that | f gdµ| ≤
M ||g||Lp for every g ϵ L∞ (X)
Now we prove that f ϵ Lq (X). Let {ϕn } be a sequence of simple func-
tions such that ϕn −→ f pointwise a.e. as n −→ ∞ and |ϕn | ≤ |f |.
Define gn =(sgnf )( ||ϕ|ϕnn||L| q )q/p .Then gn ϵL∞ (X) and ||gn ||Lp = 1.Further,
R gn |
f gn =|f
and |ϕn gn |dµ = ||ϕn ||Lq .Therefore, by Fatou’s lemma and inequality R
|ϕn | ≤ |f |, Rwe have, ||f ||Lq ≤ limn→∞ inf ||ϕn ||Lq ≤ limn→∞ inf |ϕn gn |dµ
≤ limn→∞ |f gn |dµ ≤ M .
Thus f ϵ Lq .Since the simple functions are dense in Lp and R g is a contin-
uous functional on Lp , if f ϵ Lq , it follows that F (g) = f gdµ for every
g ϵ Lp (X).By the previous proposition, this implies that ||F ||Lp′ =||f ||Lq .
This proves the result, when X has a finite measure.
If X is σ − f inite then there is an increasing sequence {An } of sets

98
CHAPTER 5. BOUNDED LINEAR TRANSFORMATIONS AND DUAL
SPACES
∞
S
with finite measure such that An = X.By the previous result, there
n=1
is a unique function R
fn ϵ Lq (An ) such that F (g) = An fn gdµ for all g ϵ Lp (An ). If m
≥ n then the functions fm , fn are equal pointwise a.e. on An and the
dominated convergence theorem implies that f = limn→∞ fn ϵ Lq (X)
is the required function.
Finally , if X is not σ−f inite then for each σ−f
R inite subset A ⊂ X, let
fA ϵLq (A) be the function such that F (g) = A fA gdµ for every gϵLp (A).
Define M = sup{||fA ||Lq (A) : A ⊂ X is σ − f inite} ≤ ||F ||Lp (X)′ , and
choose an increasing sequence of sets An such that ||fAn ||Lq (An ) −→ M
as n −→ ∞ ∞
S
.Define B = An , verify that fB is the required function.
n=1

5.4 Separable spaces

Definition 7: Let (X, d) be a metric space then X is said to be

separable if there exist a countable subset A of X such that A = X
Examples:
1] the real line R is separable because Q is a countable subset of R such
that Q = R.
Now , we shall see some of the examples of separable and nonseparable
spaces through the following results.
Theorem 5.4.1. The space l∞ is not separable.
Proof: Let y = (η1 , η2 , ...) be a sequence of 0,1.Then clearly, y ϵ
l∞ .Corresponding to y, we associate the real number ỹ, whose binary
representation is η21 + η222 + η233 + ...
Observe that every ỹϵ[0, 1] has a unique binary representation.Hence,
there are uncountably many sequences made up of 0, 1.The metric on
l∞ shows that if x ̸= y then d(x, y) = 1.If we choose a small ball of
radius 31 centred at these sequences then these balls do not intersect
and there are uncountably many such balls. If A is any dense set in l∞
then every such non-intersecting balls contain an element of A. This
shows that A cant be countable set.Since A is an arbitrary dense set,
we conclude that l∞ cant have countable dense set.Therefore, l∞ is not
separable.
Theorem 5.4.2. The space lp is separable for 1 ≤ p < ∞.
Proof: Let M be the set of all sequences y of the form y = (η1 , η2 , ..., ηn , 0, 0, ...)
where n is any positive integer and ηi ’s are rational.Observe that M
is countable.We claim that M is dense in lp .Let x = (ζi ) ϵlp be any
element.Then for every ϵ > 0 there is an n depending on ϵ such that

99
FUNCTIONAL ANALYSIS

P∞
i=n+1 |ζi |p < ϵp /2.This is because LHS is the remainder of a con-
vergent series. Since, the rationals are dense in R, for each ζi there
is
Pnrational number ηi close to it.Hence, we can find yϵM satisfying
p p
|ζ
i=1 i − η i | < ϵ /2.This implies
P∞ that p
n
[d(x, y)] = i=1 |ζi − ηi | + i=n+1 |ζi | < ϵp .Thus, we have d(x, y) <
p p
P
ϵ.Therefore, M is dense in lp .
′
Theorem 5.4.3. If the dual space X of a normed space is separable ,
then X itself is separable.
′
Proof: We assume that X is separable.Then the unit sphere
′ ′
U = {f ϵX : ||f || = 1} ⊂ X also contains a countable dense subset
say (fn ) . Since fn ϵU , we have ||fn || = sup{|fn (x)| : ||x|| = 1} = 1.
Therefore, by definition of supremum, we can find points xn ϵX of norm
1 such that |fn (xn )| ≥ 21 . Let Y = span(xn ).Then Y is separable,
because Y has a countable dense subset span(xn ), which consists of all
linear combinations of x′n s with coefficients, whose real and imaginary
parts are rational numbers.We claim that Y = X.Suppose that Y ̸=
X.Since, Y is closed in X, therefore, by the following lemma:
Lemma 5.4.1. (Existence of functional). Let Y be proper closed sub-
space of a normed space X.Let x0 ϵX −Y be arbitrary and δ = inf {||ỹ −
′
x0 || : ỹϵY }, the distance from x0 to Y. Then there exists an f˜ϵX such
that ||f˜|| = 1, f˜(y) = 0 for all yϵY , f˜(x0 ) = δ
′
We have f˜ϵX with ||f˜|| = 1 and f˜(y) = 0 for all yϵY
Since xn ϵY , we have f˜(xn ) = 0 and for all n,
1
2
≤ |fn (xn )| = |fn (xn )−f˜(xn )| = |(fn −f˜)(xn )| ≤ ||fn −f˜|| ||xn ||, where
||xn || = 1.Hence ||fn − f˜|| ≥ 12 ,but this contradicts the assumption that
(fn ) is dense in U , because we have f˜ϵU

5.5 Let us Sum Up

(1) If T is a linear transformation existing between two normed
spaces X and Y then T is bounded iff T is continuous iff T is continu-
ous at the origin iff T maps closed unit sphere to a bounded set.
(2) B(X, Y ) is normed linear space and it is complete, if Y is complete.
(3) The dual space of Rn is Rn .
(4) The dual space of l1 is l∞ .
(5) For 1 < p < ∞ the dual space of lp is lq .
(6) Every bounded linear functional on Hilbert space can be represented
in terms of an inner product.
(7) For 1 < p < ∞, the dual space of Lp (X) is Lq (X).If X is σ-finite
then L1 (X)′ = L∞ (X) .
(8) The space l∞ is not separable.

100
CHAPTER 5. BOUNDED LINEAR TRANSFORMATIONS AND DUAL
SPACES
(9) For 1 ≤ p < ∞, the space lp is separable.
′
(10)If the dual space X of a normed space is separable then X is sep-
arable.

5.5 List of References:

(1) Introductory Functional Analysis with applications by Erwin Kreyszig,
Wiley India

(2) Educational Resources , the University of California , Davis

5.6 CHAPTER END EXERCISES

(1) Show that C([a, b]) is separable.

(Hint: This follows from the Weierstrass approximation theorem that
states that P([a, b])={f ϵC([a, b]) : f is a polynomial with real coef-
ficients } is dense in C([a, b]).Further, Q([a, b]) = {f ϵC([a, b]) : f is
a polynomial with rational coefficients }, we can show that Q([a, b] is
countable and it is dense in P([a, b])

(2) C([a, b], R) with supremum norm is Banach space.

Proof: Let {fn } be any Cauchy sequence in C([a, b], R). This means
that given ϵ > 0 there exists an integer N > 0 such that ||fn − fm ||∞ <
ϵ
2
, whenever, n, m ≥ N .That is , given any ϵ > 0 there exists an integer
N > 0 such that |fn − fm | < 2ϵ for all n, m ≥ N and every xϵ[a, b].Thus
{fk (x)} is a Cauchy sequence of real numbers for every xϵ[a, b].Since
R is complete.Therefore {fk (x)} converges to some real number for
each x; we will denote this value by f (x).This defines a new function
f such that fn −→ f pointwise. We prove that fn −→ f uniformly
on [a, b]. Since fn is a sequence of continuous function therefore its
uniform limit f is also continuous.Let ϵ > 0 be any number.Then there
exists an N such that |fn (x) − fm (x)| < 2ϵ for every choice of xϵ[a, b]
and n, m ≥ N . If m ≥ N and xϵ[a, b] then fn (x)ϵ(fm (x)− 2ϵ , fm (x)+ 2ϵ ),
for all n ≥ N .Therefore f (x)ϵ[fm (x) − 2ϵ , fm (x) + 2ϵ ], and hence |f (x) −
fm (x)| ≤ 2ϵ < ϵ.Since x was arbitrary, therefore we are done.

(3) Let X be a normed space of all polynomials on I = [0, 1] with

norm given by ||x|| = max|x(t)|, for tϵI. A differentiation operator T
is defined on X by T (x(t)) = x′ (t).Show that T is linear operator and
it is not bounded.
(Hint: Consider xn (t) = tn , where nϵN. Then ||xn || = 1 and ||T (xn )|| =
n)

(4) Let T be a bounded linear operator from a normed space X onto a

101
FUNCTIONAL ANALYSIS

normed space Y .If there is positive b such that ||T (x)|| ≥ b||x|| for all
xϵX then show that T −1 exists and it is bounded.

(5) Show that the dual space of c0 is l1

2
(6) Show that every bounded
P∞ linear functional f 2on l can be repre-
sented in the form f (x) = j=1 ξj ζj for z = (ζj )ϵl
(Hint: Use Riesz representation theorem to express any bounded linear
functional f on Hibert space l2 as f (x) =< x, z >)

(7) Show that any Hilbert Space is isomorphic with its second dual
′′
space H = (H ′ )′ (This property is called reflexivity of H)
Proof: We shall prove that the canonical mapping C : H −→ H ′′ de-
fined by C(x) = gx is onto, where gx is a functional on X ′ defined for
fixed xϵX as gx (f ) = f (x) for f ϵX ′ , by showing that for every gϵH ′′
there exist an xϵH such that g = C(x)
Define A : H ′ −→ H by A(f ) = z, where z is determined by the Riesz
representation theorem f (x) =< x, z >, we know that A is bijective
and isometric.A(αf1 + βf2 ) = αAf1 + βAf2 implies that A is conju-
gate linear.Observe that H ′ is complete and its an Hilbert space with
inner product defined by < f1 , f2 >=< Af2 , Af1 >.For all functionals
f1 , f2 , f3 and scalars α we have
< f1 + f2 , f3 >=< Af3 , A(f1 + f2 ) >= < A(f1 + f2 ), Af3 >=
< Af1 , Af3 > + < Af2 , Af3 >=< Af3 , Af1 > + < Af3 , Af2 >
=< f1 , f3 > + < f2 , f3 >
< αf1 , f2 >=< Af2 , A(αf1 ) >=< Af2 , αAf1 >= α < f1 , f2 >
< f1 , f2 >=< Af2 , Af1 >= < Af1 , Af2 > = < f2 , f1 >
< f1 , f1 >=< Af1 , Af1 >=< z, z >≥ 0 and < f1 , f1 >= 0 implies that
z = 0, hence f1 = 0
Let gϵH” be arbitrary.Let its Riesz representation be g(f ) =< f, f0 >=<
Af0 , Af >.We know that f (x) =< x, z > , where z = Af . Writing
Af0 = x, we therefore have < Af0 , Af >=< x, z >= f (x).Together
with g(f ) = f (x) implies that g = C(x), by the definition of C.Since
gϵH” was arbitrary, C is onto, so that H is reflexive.

102
Chapter 6

Four Pillars of Functional

Analysis

Unit Structure :
6.1 Introduction
6.2 Objective
6.3 Few Definitions and Notations
6.4 Hahn-Banach Theorem
6.5 Uniform Boundedness Principle
6.6 Open Mapping Theorem
6.7 Closed Graph Theorem
6.8 Applications of Hahn-Banach theorem
6.9 Chapter End Exercise

6.1 Introduction
In this chapter we shall see four important theorems, which are
also called sometimes called as four pillars of Functional Analysis. The
Hahn-Banach theorem, the Open Mapping Theorem, Closed Graph
Theorem and Uniform Boundedness Principle.
Hahn-Banach Theorems: It is so much important because it pro-
vides us with the linear functionals to work on various spaces as Func-
tional Analysis is all about the study of functionals.

Open Mapping Theorem: It provides us with the open sets in

the topology of the range of the mapping.

Uniform Boundedness Principle: An application of Baire Cat-

egory theorem. It is further used many times as the uniformity is an
important property.

103
FUNCTIONAL ANALYSIS

Closed Graph Theorem: Closeness of the graph of a map is

enough to prove its boundedness or continuity. This fact is further
used many times.

6.2 Objectives
After going through this chapter you will be able to:
• State and prove Hahn-Banach theorems
• State and prove Open Mapping theorem
• State and prove Closed Graph theorem
• State and prove Uniform Boundedness theorem

6.3 Few Definitions and Notations

Let X be a normed space over the field K(R or C).
Definition 6.1. Let X be a normed space over K. A mapping f :
X → K is said to be linear functional on X if:
f (αx + βy) = αf (x) + βf (y), ∀x, y ∈ X, α, β ∈ K
Definition 6.2. A linear functional f , as defined in Definition 6.1,
is said to be bounded if there exists M > 0 in R such that

|f (x)| ≤ M ∥ x ∥, ∀x ∈ X
.
Note: The branch of analysis of functionals as defined above was ba-
sically called as functional analysis initially! The bounded linear func-
tional is a special case of bounded liner operator and hence all proper-
ties related to bounded linear operators holds true for bounded linear
functionals also. Here is a small activity for you to recollect these
properties.
Activity 1: Let f be a linear functional on a normed space X.
1. f is continuous iff ker (f ) is closed in X.

.....................................................................
.....................................................................
.....................................................................
.....................................................................

104
CHAPTER 6. FOUR PILLARS OF FUNCTIONAL ANALYSIS

.....................................................................
.....................................................................

2. Consider Rn with usual norm and let a = (a1 , a2 , . . . , an ) be a

vector in Rn . Define f : Rn → R such that f (x) = x.a, where x =
(x1 , x2 , . . . , xn ) and x.a denote the scalar product of x with a. Show
that f is a bounded linear functional and ∥ f ∥=∥ a ∥.

.....................................................................
.....................................................................
.....................................................................
.....................................................................
.....................................................................
.....................................................................

3. f is continuous iff f is bounded.

.....................................................................
.....................................................................
.....................................................................
.....................................................................

4. If f is bounded linear functional then |f (x)| ≤∥ f ∥∥ x ∥, ∀x ∈ X.

.....................................................................
.....................................................................
.....................................................................
.....................................................................

Remark: If f is a bounded linear complex functional then f need not

be linear.

Definition 6.3. Let X be a linear space over R. A functional p is said

to be sublinear functional on X if it satisfies the following properties:
(i) p(x + y) ≤ p(x) + p(y), ∀x, y ∈ X
(ii) p(αx) = αp(x), ∀α ≥ 0 in R, x ∈ X
Definition 6.4. Let X be a linear space over K and Z be it’s subspace.
Let f : Z → K be a linear functional. Then f˜ : X → K is said to be

105
FUNCTIONAL ANALYSIS

an extension of f if f (x) = f˜(x), ∀x ∈ Z. Infact f is also called as

restriction of f˜ on Z and is denoted as f˜|Z = f .

6.4 Hahn-Banach Theorem

Theorem 6.4.1. (Hahn-Banach Lemma) Let X be a real vector

space and p be a sublinear functional on X. Let Z be a subspace of X
and f be linear functional defined on Z such that

f (x) ≤ p(x), ∀x ∈ Z

Then, there exists a linear functional f˜ on X such that f˜|Z = f and

f˜(x) ≤ p(x), ∀x ∈ X

Proof: Consider a set L of all linear extensions (Zα , gα ) of (Z, f ) such

that,
gα (x) ≤ p(x), ∀x ∈ Zα
Since (M, f ) ∈ L, clearly L = ̸ ϕ.
Now, we define a relation ”≤” on L such that:
(Zα , gα ) ≤ (Zβ , gβ ) ⇔ Zα ⊂ Zβ and
gβ is a extension of gα i.e. gβ |Zα = gα

Activity 2: Check that the relation ” ≤ ” defined above is reflexive,

antisymmetric and transitive.

.....................................................................
.....................................................................
.....................................................................
.....................................................................

From the above activity we can conclude that (L, ≤) is a partial ordered
set. Let Q be any totally ordered subset of L and let

Z ′ = β ∪ {Zβ : (Zβ , gβ ) ∈ Q}

We see that Z ′ is a subspace. Define g ′ : Z ′ → R by

g ′ (x) = gβ (x), ∀x ∈ Zβ

Clearly, g ′ is a linear functional on Z ′ and g ′ |Z = f .

Claim: (Z ′ , g ′ ) is an upper bound of Q (Try it yourself)
By Zorn’s lemma, Q has a maximal element, say (Z0 , g0 ).

106
CHAPTER 6. FOUR PILLARS OF FUNCTIONAL ANALYSIS

To prove that, (Z0 , g0 ) = (X, f˜). It is enough to prove that Z0 = X.

Suppose Z0 ̸= X. Then there exists x0 ∈ X − Z0 . Consider the linear
space spanned by Z0 and x0 , Z ′ = Z0 + [x0 ].
Each element z ∈ Z ′ can be uniquely expressed as z = x + αx0 , where
x ∈ Z0 and α ∈ R.
Define g1 : Z1 → R as g1 (x + αx0 ) = g0 (x) + αK, where K is a real
constant.
We can see that g1 is a linear functional on Z1 (Verify this!) And
g1 |Z = g0
Thus, Z1 is a linear subspace of X containing Z and g1 |Z = f
The constant K can be chosen appropriately so that g1 (y) ≤ p(y), ∀y ∈
Z1 .
This means, 9Z1 , g1 ) ∈ Q and (Z0 , g0 ) ≤ (Z1 , g1 ), Z0 ̸= Z1 . This
contradicts the maximality of (Z0 , g0 ). Hence our assumption that Z0 ̸=
X is not true. Hence the proof.

Theorem 6.4.2. (Hahn-Banach) Let X be a normed space over a

field K and Z be a subspace of X.Then, for every bounded linear func-
tional f on Z, there exists a bounded linear functional f˜ on X such
that, f˜|Z = f and ∥f˜∥ = ∥f ∥

6.5 Uniform Boundedness Principle

The uniform boundedness principle (or uniform boundedness theo-

rem) by S. Banach and H. Steinhaus (1927) is one of the fundamental
results in functional analysis. Together with the Hahn-Banach theo-
rem, the open mapping theorem and the closed graph theorem, it is
considered as one of the cornerstones of the field.
The uniform boundedness principle answers the question of whether
a ”point-wise bounded” sequence of bounded linear operators must also
be ”uniformly bounded”. As the proof of the Uniform Boundedness
Principle is an application of Baire’s Category Theorem. So, we shall
prove the Baire’s category theorem first. Following are the basic con-
cepts needed for Baire’s theorem:

Definition 6.5. (Nowhere Dense or Rare) A subset M of a metric

space X is said to be Nowhere dense in X if its closure M has no
interior points. That is, int(M ) ̸= ∅, M contains no open ball.

Example 30. The set of all integers Z is nowhere dense set in R

Example 31. Let (R, d) be the usual metric space, then every singleton
is nowhere dense in R since {a} = {a} for every a ∈ R. And int(a) = ∅
since it contains no open interval.

107
FUNCTIONAL ANALYSIS

Definition 6.6. (Meager or of First Category) A subset M in metric

space X is said to be of First Category if M is the union of countably
many sets which are all nowhere dense in X.
Example 32. Since Q is countable and for every a ∈ Q, {a} being
nowhere dense,
[
Q= {a}
a∈Q

is of first category.
Definition 6.7. (Nonmeager or of Second Category) A subset M in
metric space X is said to be of Second Category if M is not of first
category in X.
Theorem 6.5.1. (Baire’s Category Theorem) If a metric space X ̸= ∅
̸ ∅ is complete and
is complete, it is of second category. Hence, if X =
∞
[
X= Ak (Ak closed)
k=1

then atleast one Ak contains a nonempty open subset.

Proof. Suppose the metric space X ̸= ∅ is of first category in itself.

Then
∞
[
X= Mk ,
k=1

where each Mk is rare in X. Since M1 is rare in X, so by definition, M1

does not contain a nonempty open set. But (X, d) is complete, so it
C
will contain a nonempty set. So, M1 ̸= X. Therefore, M1 = X − M1
C
which is nonempty and open. At the point p1 ∈ M1 , we can get an
open ball
C
B1 = B(p1 , ϵ1 ) ⊂ M1 ,
where ϵ < 21 .
Further, M2 is rare in X, so that M2 does not contain a nonempty open
set. Hence, it does not contain open ball B(p1 , ϵ21 ). This implies that
CT
M2 B(p1 , ϵ21 ) is not empty and open. Now, we may choose an open
ball in this set, say,
C \ ϵ1 ϵ1
B2 = B(p2 , ϵ2 ) ⊂ M2 B(p1 , ), ϵ2 < .
2 2
Continuing this process, we obtain a sequence of balls by induction,

Bk = B(pk , ϵk ), ϵk < 2−k

108
CHAPTER 6. FOUR PILLARS OF FUNCTIONAL ANALYSIS

such that Bk Mk = ∅ and Bk+1 ⊂ B(pk , ϵ2k ) ⊂ Bk , k = 1, 2, ....

T
As ϵk < 2−k and the space X is complete, the sequence (pk of the
centers is Cauchy and converges, say, pk → p ∈ X. Also, for every m
and n > m we have Bn ⊂ B(pm , ϵ2m ), so that

d(pm , p) ≤d(pm , pn ) + d(pn , p)

ϵm ϵm
< + d(pn , p) →
2 2
as n approaches to ∞. Hence, p belongs to Bm for every m. Since
C S
Bm ⊂ Mm , we get p ̸∈ Mm for every m, so that p ̸∈ Mm = X. This
contradicts p ∈ X. Hence proved.

Theorem 6.5.2. Let {Tn } be a sequence of bounded linear operators

Tn : X → Y from a Banach space X into a normed space Y . If for
every x ∈ X, {Tn (x)} is bounded, say,

∥Tn (x)∥ ≤ cx , n = 1, 2, 3, ... (pointwise boundedness),

where cx is a real number, then the sequence ∥Tn ∥ is also bounded, that
is, there is a positive real c such that

∥Tn ∥ ≤ c, n = 1, 2, 3, ... (unif orm boundedness),

Proof. For every x ∈ X, {Tn (x)} is bounded sequence in Y , that is,

∥Tn (x)∥ ≤ cx where cx ≥ 0.

Suppose that Ak = {x ∈ X : ∥Tn (x)∥ ≤ k}

Step 1: We claim that Ak is nonempty, closed and X = ∞
S
k=1 Ak . Since
0 ∈ X and Tn is linear so Tn (0) = 0 ≤ k for each k. This implies that
0 ∈ Ak for each k which implies that Ak is nonempty.
Let {xn } be a convergent sequence in Ak with xm → x as n → ∞.
This means that for every fixed n, we have ∥Tn (xm )∥ ≤ k and obtain
∥Tn (x)∥ ≤ k by applying limits for m → ∞ because Tn is continuous
and so is the norm. Hence, x ∈ A Sk∞and therefore Ak is closed.
Also, Ak ⊆ X (for all k) ⇒ k=1 Ak ⊆ X. On the other hand, let
x ∈ X, ∥Tn (x)∥ is a real number. Using Archimedean property, there
is a positive integer n0 such that Tn (x)∥ ≤ cx ≤ n0 . So,
x ∈ An0 ⊆ ∞
S S∞
k=1 A k ⇒ X ⊆ k=1 Ak

Hence,
∞
[ ∞
[
X= Ak = Ak , (Since Ak = Ak )
k=1 k=1

In view of Baire’s Category Theorem, every complete space is of second

category. That is, at least one of Ak is not nowhere dense. Hence, some

109
FUNCTIONAL ANALYSIS

Ak contains an open ball, say, B0 = B(x0 , r) ⊆ Ak0 .

Step 2: Let x ∈ X be arbitrary, not zero. We set
z = x0 + γx ,
r
where γ = 2∥x∥
. Then
r
z = x0 + x
2∥x∥
r
z − x0 = x
2∥x∥
r r
∥z − x0 ∥ = ∥x∥ = < r
2∥x∥ 2
That is, z ∈ B(x0 , r) ⊆ Ak0 .
As, z, x0 ∈ Ak0 ⇒ ∥Tn (z)∥ ≤ k0 , ∥Tn (x0 )∥ ≤ k0 . We have,
2∥x∥
x= (z − x0 ).
r
Therefore,
2∥x∥
∥Tn (x)∥ = ∥Tn (z − x0 )∥
r
2∥x∥
= ∥Tn (z) − Tn (x0 )∥
r
2∥x∥
≤ [∥Tn (z)∥ + ∥Tn (x0 )∥]
r
2∥x∥
≤ (2k0 )
r
4k0
= ∥x∥
r
Hence,
4k0 ∥Tn (x)∥ 4k0
∥Tn (x)∥ ≤ ∥x∥ ⇒ ≤ = c.
r ∥x∥ r
which implies
∥Tn (x)∥
sup ≤ c ⇒ ∥Tn ∥ ≤ c.
x∈X,x̸=0 ∥x∥

6.6 Open Mapping Theorem

Open mapping theorem is one of the basic theorems for the develop-
ment of the general theory of normed linear spaces. The theorem gives
conditions under which a linear mapping is open. In this theorem, we
begin to appreciate the importance of the completeness condition for
normed linear spaces.

110
CHAPTER 6. FOUR PILLARS OF FUNCTIONAL ANALYSIS

Definition 6.8. (Open Mapping) Let X and Y be metric spaces, a

mapping T : D(T ) → Y with domain D(T ) ⊂ X is called an open
mapping if for every open set in D(T ) the image is an open set in Y .
Definition 6.9. (Continuous Mapping) A continuous mapping T :
X → Y has the property that for every open set in Y the inverse
image is an open set in X.
Remark 6.6.1. Continuous mappings are not necessarily open map-
pings.
Example 33. Consider a mapping T : R → R defined by T (x) =
Sinx. As Sine function is continuous, T is a continuous mapping but
T [(0, 2π)] = [−1, 1] that is, T maps an open set (0, 2π) onto [−1, 1]
which is not open.
Example 34. Define T : R → R by T (x) = x2 . Then T is continuous
mapping but not an open mapping because it maps an open set (−1, 1)
onto [0, 1) which is not open.
Lemma 6.6.1. (Open Unit Ball) Let T be a bounded linear operator
from a Banach space X onto a Banach space Y . Then the image of the
open ball B0 = B1 (0) ⊂ X, that is, T (B0 ), contains an open ball with
center 0 in Y .

Proof. The proof has three steps. We will prove

(a) The closure of the image of the open ball B1 = B(0; 21 ) contains an
open ball B ∗ .
(b) T (Bn ) contains an open ball Vn about 0 ∈ Y , where Bn = B(0; 2−n ) ⊂
X.
(c) T (B0 ) contains an open ball about 0 ∈ Y .
(a) Clearly, we have
∞
[
kB1 ⊂ X
k=1

For any x ∈ X, there is k(k > 2||x||) such that x ∈ kB1 . So,
∞
[
X⊂ kB1 .
k=1

Thus, we have
∞
[
X= kB1 .
k=1

Since T is surjective,
∞
[ ∞
[
Y = T (X) = T ( kB1 ) = T (kB1 ).
k=1 k=1

111
FUNCTIONAL ANALYSIS

Due to the linearity of T , we have

∞
[ ∞
[ ∞
[
Y = T (kB1 ) = kT (B1 ) = kT (B1 ).
k=1 k=1 k=1

Since Y is complete, by the Baire’s Category theorem, we conclude that

a kT (B1 ) contains an open ball. This implies that T (B1 ) also contains
an open ball, namely, there is B ∗ = B(y0 ; ϵ) such that B ∗ ⊂ T (B1 ). It
follows that B ∗ − y0 = B(0; ϵ) ⊂ T (B1 ) − y0 .
(b) We will first prove that B(0; ϵ) = B ∗ − y0 ⊂ T (B0 ). Since B ∗ ⊂
T (B1 ) by (a), we have B ∗ − y0 ⊂ T (B1 ) − y0 . It suffices to prove
T (B1 ) − y0 ⊂ T (B0 ). Let y ∈ T (B1 ) − y0 . Then y + y0 ∈ T (B1 ).
Notice that y0 ∈ T (B1 ) since B ∗ ⊂ T (B1 ). Then there are sequences
un = T wn ∈ T (B1 ) and vn = T zn ∈ T (B1 ) such that un → y + y0 ,
vn → y0 , where wn , zn ∈ B1 . Observing that
1 1
||wn − zn || ≤ ||wn || + ||zn || < + = 1.
2 2
So, wn − zn ∈ B0 . Also, T (wn − zn ) = T wn − T zn = un − vn → y.
Hence, y ∈ T (B0 ). This proves that B(0; ϵ) = B ∗ − y0 ⊂ T (B0 ). Let
Bn = B(0; 21n ). Since T is linear, we have T (Bn ) = 2−n T (B0 ). Let
Vn = B(0; 2ϵn ). Then Vn = 21n B(0; ϵ) ⊂ 21n T (B0 ) = T (Bn ). This proves
(b).
(c) We finally prove that V1 = B(0; 12 ϵ) ⊂ T (B0 ) by showing that every
y ∈ V1 is in T (B0 ). So, let y ∈ V1 . Since V1 ⊂ T (B1 ), there is x1 ∈ B1
such that ∥y−T x1 ∥ < 4ϵ . Then we have y−T x1 ∈ V2 . Since V2 ⊂ T (B2 ),
there is x2 ∈ B2 such that ∥y − T x1 − T x2 ∥ < 8ϵ . Continuing in this
manner, we have, for each n, there are xn ∈ Bn such that
n
X ϵ
∥y − T xk ∥ < .
k=1
2n+1
Let zn = x1 + x2 + ... + xn . The above inequality becomes
ϵ
∥y − T zn ∥ < n+1 , ∀n.
2
Namely, T zn → y. Since xk ∈ Bk , we have ∥xk ∥ < 21k . So, for n > m,
n ∞
X X 1
∥zn − zm ∥ ≤ ∥xk ∥ < → 0, as m → ∞.
k=m+1 k=m+1
2k

Thus, the sequence {zn } is Cauchy. Since X is complete, there is x ∈ X

such that zn → x and x = x1 + x2 + .... Notice that
∞ ∞
X 1 X 1 1
∥x∥ ≤ ∥xk ∥ < + ∥xk ∥ ≤ + = 1.
k=1
2 k=2 2 2
So, x ∈ B0 . Since T is continuous, we have T zn → T x. Hence, y = T x.
That is, y ∈ T (B0 ).

112
CHAPTER 6. FOUR PILLARS OF FUNCTIONAL ANALYSIS

Theorem 6.6.1. (Open Mapping Theorem, Bounded Inverse Theorem)

Let X and Y be Banach spaces. Then any bounded linear operator T
from X onto Y is an open mapping. Consequently, if T is bijective,
then T −1 is continuous and hence bounded.

Proof. Let A ⊂ X be an arbitrary open subset of X. We will show that

the image T (A) is open in Y . That is, for any y = T x ∈ T (A), the set
T (A) contains an open ball centered at y. Let y ∈ T (A). Then y = T x
with x ∈ A. Since A is open, there is r > 0 such that Br (x) ⊂ A. Thus

1
B1 (0) ⊂ (A − x).
r
By Lemma 6.6.1, the image T ( 1r (A − x)) contains an open ball with
center 0. That is, there is ϵ > 0, such that
1
B(0; ϵ) ⊂ T ( (A − x))
r
Since T is linear, we have
1
B(0; ϵ) ⊂ (T (A) − T x).
r
Since y = T x, the above relation implies B(y; rϵ) ⊂ T (A). Hence, T (A)
contains an open ball with center y.

6.7 Closed Graph Theorem

Definition 6.10. (Cartesian product of two normed spaces)

Let (X, ∥.∥1 ) and (X, ∥.∥2 ) be two normed spaces, then X × Y is also a
normed space where the two algebraic operations of a vector space and
the norm on X × Y are defined as usual, that is

ˆ (x1 , y1 ) + (x2 , y2 ) = (x1 + x2 , y1 + y2 )

ˆ α(x, y) = (αx, αy) (α a scalar)

ˆ ∥(x, y)∥ = ∥x∥1 + ∥y∥2

Theorem 6.7.1. For any two Banach spaces X and Y , X × Y is also

a Banach space.

Proof. We show that for any two Banach spaces X and Y ,X × Y =

{(x, y) : x ∈ X, y ∈ Y } is also a Banach spaces. Let {zn } be a Cauchy

113
FUNCTIONAL ANALYSIS

sequence in X × Y , where zn = (xn , yn ). Then for every ϵ > 0, there

exists a positive integer n0 such that for every n ≥ n0 , we have
∥zn − zm ∥ = ∥(xn , yn ) − (xm , ym )∥
= ∥(xn − xm , yn − ym )∥
= ∥xn − xm ∥ + ∥yn − ym ∥ < ϵ
This implies that ∥xn − xm ∥ < ϵ, ∥yn − ym ∥ < ϵ thereby proving that
{xn } is a Cauchy sequence in X and {yn } is a Cauchy sequence in Y .
Let xn → x ∈ X, yn → y ∈ Y , then
∥zn − z∥ = ∥(xn , yn ) − (x, y)∥ = ∥(xn − x, yn − y)∥ = ∥xn − x∥ + ∥yn − y∥
Applying n → ∞,
∥xn −x∥ → 0 and ∥yn −y∥ → 0. This implies that limn→∞ ∥zn −z∥ = 0.
Hence, zn → z = (x, y) ∈ X × Y . Thus, X × Y is a Banach space.
Definition 6.11. (Closed Linear Operator) Let (X, ∥.∥1 ∥) and (Y, ∥.∥2 ∥)
be normed spaces and let T : D(T ) → Y a linear operator with domain
D(T ) ⊂ X. Then T is called a closed linear operator if its graph
G(T ) = {(x, y) : x ∈ D(T ), y = T x} ⊆ X × Y
Theorem 6.7.2. Let X and Y be Banach spaces and T : D(T ) → Y
be a closed linear operator, where D(T ) ⊂ X. If D(T ) is closed in X,
then the operator T is bounded.
Proof. First recall that for any two Banach spaces X and Y , X × Y is
also a Banach space. By assumption, G(T ) is closed in X × Y . Hence,
G(T ) is complete and also D(T ) is closed in X. Therefore, D(T ) is
also complete space. We now consider the mapping P : G(T ) → D(T ),
defined by P (x, T x) = x. We prove that P is bijective, linear and
bounded. To show that P is linear, we consider
P ((x1 , T x1 ) + (x2 , T x2 )) = P (x1 + x2 , T x1 + T x2 )
= x1 + x2
= P (x1 , T x1 ) + P (x2 , T x2 )
And for α ∈ K, we have
P (α(x, T x)) = P (αx, αT x) = αx = αP (x, T x)
Now, P is bounded, because
∥P (x, T x)∥ = ∥x∥ ≤ ∥x∥ + ∥T x∥ = ∥(x, T x)∥
This implies that ∥P (x, T x)∥ ≤ 1.∥(x, T x)∥. P is onto, since for all
x ∈ D(T ) ∃(x, T x) ∈ G(T ) such that P (x, T x) = x. P is one to one,
since
P (x1 , T x1 ) = P (x2 , T x2 ) ⇒ x1 = x2 ⇒ (x1 , T x1 ) = (x2 , T x2 ).

114
CHAPTER 6. FOUR PILLARS OF FUNCTIONAL ANALYSIS

Hence, P is a bounded linear operator from Banach space G(T ) to

Banach space D(T ) and also it is bijective hence P −1 : D(T ) → G(T )
given by P −1 (x) = (x, T x) is a bounded linear operator by Bounded
inverse theorem. That is, there exists b > 0 such that for every x ∈
D(T ), we have
∥P −1 (x)∥ ≤ b∥x∥ ⇒ ∥(x, T x)∥ ≤ b∥x∥ ⇒ ∥x∥ + ∥T x∥ ≤ b∥x∥
Therefore, ∥T x∥ ≤ ∥x∥ + ∥T x∥ ≤ b∥x∥ ⇒ ∥T x∥ ≤ b∥x∥. This implies
that T is bounded.

6.8 Applications of Hahn-Banach theorem

1. Given a normed linear space X over a field K and a non zero
member x0 ∈ X, there is a bounded linear functional F over X such
that F (x0 ) = ∥x0 ∥ and ∥F ∥ = 1.
Proof. Let, M = [x0 ]= the subspace spanned by {x0 } = {αx0 : α is real}.
Define f : M → R by f (αx0 ) = α∥x0 ∥. Clearly, f is linear. Now, for
x ∈ M , x = αx0 for some α. Now, |f x| = |f (αx0 )| = |α∥x0 ∥| =
|α|∥x0 ∥ = ∥αx0 ∥ = ∥x∥. Clearly, ∥f ∥ = 1, that is, f is a bounded
linear functional on M . So, there exists an extension F (a bounded lin-
ear functional) of f over X such that ∥f ∥ = ∥F ∥. But f (x) = F (x),
∀x ∈ M . Now, x0 ∈ M . So, f (x0 ) = F (x0 ). But f (x0 ) = ∥x0 ∥. So,
F (x0 ) = ∥x0 ∥ with ∥F ∥ = 1.
2. Let, X be a normed linear space over a field K and x0 ̸= θ be
an arbitrary member of X and let M0 be an arbitrary positive real.
Then ∃ a bounded linear functional f on X, such that ∥f ∥ = M0 and
f (x0 ) = ∥f ∥∥x0 ∥.
Proof. Let G = [x0 ] = {tx0 : t is real}. Clearly, G is a subspace of X.
Define ϕ : G → R by ϕ(tx0 ) = tM0 ∥x0 ∥. Clearly, ϕ is linear. Now, for
x ∈ G, x = tx0 for some t. Now,
|ϕ(x)| = |ϕ(tx0 )|
= |tM0 |∥x0 ∥
= M0 |t|∥x0 ∥
= M0 ∥tx0 ∥
= M0 ∥x∥.
So, |ϕ(x)| = M0 ∥x∥. Clearly, ∥ϕ∥ = M0 . So, ϕ is a bounded linear
functional on G. So, ∃ an extension f of ϕ over X such that ∥f ∥ =
∥ϕ∥ = M0 . Hence, ∥f ∥ = M0 . But ϕ(x0 ) = M0 ∥x∥. Now, x0 ∈ G,
f (x0 ) = ϕ(x0 ) = M0 ∥x0 ∥ = ∥f ∥∥x0 ∥. So, f (x0 ) = ∥f ∥∥x0 ∥. This
completes the proof.

115
FUNCTIONAL ANALYSIS

f (x)
3. For every x ∈ X, ∥x∥ = supf ̸=Θ ∥f ∥
. [Θ is zero functional on X].
′
Proof. We find a non zero bounded linear functional f0 ∈ X such that
f0 (x) = ∥x∥ and ∥f0 ∥ = 1. Now,
|f (x)| |f0 (x)
sup ≥
f ̸=Θ ∥f ∥ ∥f0 ∥
′
Again, ∀f ̸= Θ ∈ X , |f (x)| ≤ ∥f ∥∥x∥. So,
|f (x)|
≤ ∥x∥.
∥f ∥
That is,
|f (x)|
sup ≤ ∥x∥.
f ̸=Θ∈X ′ ∥f ∥
Hence,
|f (x)|
∥x∥ = sup .
f ̸=Θ∈X ′ ∥f ∥

4. Let M be a closed subspace of normed linear space X such that

M ̸= X. Let u ∈ X \ M and let

d = dist(u, M ) = inf ∥u − M ∥
m∈M

′
Then there is a bounded linear functional f ∈ X such that
(i)f (x) = 0 ∀x ∈ M
(ii) f (u) = 1 and (iii) ∥f ∥ = d1
S
Proof. Clearly, d > 0. Let N = [M {u}]. Clearly, N is a subspace of
X. So, every member of N is of the form m + tu, where m ∈ M, t ∈ R.
Define g : N → R by g(m+tu) = t. Clearly, g is linear. Now, g(m) = 0,
for some m ∈ M, g(u) = 1. For t ̸= 0,

|g(m + tu)| = |t|

|t|∥m + tu∥
=
∥m + tu∥
∥m + tu∥ 1
= ≤ ∥m + tu∥,
∥u − (−m/t)∥ d

as (−m/t) ∈ M . Let x = m + tu ∈ N . So, |g(x)| ≤ d1 ∥x∥ implying that

′
g is bounded and ∥g∥ ≤ d1 . So, g ∈ N . Again, d = inf m∈M ∥u − m∥.
So, there exists a sequence {mn } ∈ M such that ∥u − mn ∥ → d as
n → ∞. Now,

|g(u − mn )| ≤ ∥g∥∥u − mn ∥ ⇒ 1 ≤ ∥g∥∥u − mn ∥,

116
CHAPTER 6. FOUR PILLARS OF FUNCTIONAL ANALYSIS

since g(u − mn ) = g(u) − g(mn ) = 1 − 0 = 1. Letting n → ∞, d1 ≤ ∥g∥,

′
we get ∥g∥ = d1 . So, ∃ a bounded linear functional f ∈ X which is an
extension of g on N such that f (x) = g(x), ∀x ∈ N and ∥f ∥ = ∥g∥.
Thus, we have
(i) f (x) = 0, ∀x ∈ N
(ii) f (u) = 1
(iii) ∥f ∥ = d1 .
5. Let M be a subspace of a normed linear space X and M ̸= X. If
u ∈ X M such that d = dist(u, M ) > 0. Then there is a bounded
′
linear functional F ∈ X such that
(i) F (x) = 0 ∀x ∈ M
(ii) F (u) = d
(iii) ∥F ∥ = 1.
S
Proof. Let, N = [M {u}] = a subspace of X spanned by M and {u}.
Define f : N → R by f (x) = td, where x = m + tu for some m ∈ M
and for some t. That is, f (m + tu) = td. Clearly, f is linear. Now,
f (m) = 0 for m ∈ M , f (u) = d. Now, for t ̸= 0

−m
∥m + tu∥ = ∥ − t −u ∥
t
−m
= |t|∥ − u∥ ≥ |t|d,
t
as −m
t
∈ M and d = dist(u, M ). So, |f (m + tu)| = |t|d ≤ ∥m + tu∥.
Let x = m + tu ∈ N . So,
|f (x)| ≤ ∥x∥ ⇒ ∥f ∥ ≤ 1. (6.1)
′
Clearly, f is bounded with ∥f ∥ ≤ 1. So, f ∈ N . We have d =
dist(u, M ) = inf m∈M ∥u − m∥. Let ϵ > 0. By infimum property there
exists an element m ∈ N such that ∥u − m∥ < d + ϵ.
Put
m−u m 1
v= = − u ∈ N,
∥m − u∥ ∥m − u∥ ∥m − u∥
−1 m
where t = ∥m−u∥
and ∥m−u∥
∈ M . So, ∥v∥ = 1. But

d d
|f (v)| = > ∥v∥, ∥v∥ = 1.
∥m − u∥ d+ϵ
As, ϵ > 0 is arbitrary,
|f (v)| ≥ ∥v∥ ⇒ ∥f ∥ ≥ 1. (6.2)
So, by (6.1) and (6.2), ∥f ∥ = 1. So, there exists a bounded linear
functional F over X which is an extension of f over N such that f (x) =
F (X), ∀x ∈ N and ∥f ∥ = ∥F ∥. So, (i) F (x) = 0, ∀x ∈ M ,
(ii) F (u) = d and (iii) ∥F ∥ = 1.

117
FUNCTIONAL ANALYSIS

6.9 Chapter End Exercises

1. Let f be an additive functional on a normed space X. Prove that

if f is continuous then f is linear.

2. Prove that f : R2 → R defined by f (x, y) = f (x) + f (y), x, y ∈ R,

is a bounded linear functional on (R, ∥ . ∥2 ).

3. If X be a non trivial real normed linear space, that is, X ̸= {0}.

′
Then its first conjugate space X is also nontrivial.

4. Let X is a Banach space and A ⊂ X a dense set. Can we find

a function f : X → R such that, for every x ∈ A, we have
limt→x |f (t)| = ∞?

5. Let X and Y be Banach spaces and T ∈ B(X, Y ). Suppose T is

bijective. Show that there exist real numbers a, b > 0 such that
a∥x∥ ≤ ∥T x∥ ≤ b∥x∥, ∀x ∈ X.

6. Let X, Y and Z be Banach spaces. Suppose that T : X → Y is

linear, that J : Y → Z is linear, bounded and injective, and that
JT ≡ J ◦ T : X → Z is bounded. Show that T is also bounded.

7. Let (X, ∥.∥1 ) and (X, ∥.∥2 ) be Banach spaces. Suppose that

∃C ≥ 0 : ∥x∥2 ≤ C∥x∥1 , ∀x ∈ X.

Show that the two norms ∥.∥1 and ∥.∥2 are equivalent.

118