Master of Science in Mathematics

FUNCTIONAL ANALYSIS
[MMSS-32]
SEMESTER - III

Department of Mathematics
School of Sciences
Tamil Nadu Open University
577, Anna Salai, Saidapet,
Chennai – 600 015, Tamil Nadu.
Name of the Programme: M.Sc., Mathematics
Course Code: MMSS - 32
Course Title: Functional Analysis

Curriculum Design
Dr. E. Kumar
Assistant Professor
School of Sciences
Tamil Nadu Open University, Chennai – 15

Course Writer
Dr. G. Palani,
Assistant Professor,
PG & Research Department of Mathematics,
Dr Ambedkar Govt, Arts College,
Vysarpadi, Chennai – 39.

Course Coordinator
Dr. E. Kumar
Assistant Professor
School of Sciences
Tamil Nadu Open University, Chennai – 15

Total No. of Pages: 210

First Edition: December 2022
ISBN No: 978-93-5706-422-4

©Tamil Nadu Open University

All rights reserved. No part of this work can be reproduced in any form, by mimeograph or any other
means, without permission in writing from the Tamil Nadu Open University. Course Writer is the
sole responsible person for the contents presented/available in the Course Materials. Further
information on the Tamil Nadu Open University Academic Programmes may be obtained from the
University Office at 577, Anna Salai, Saidapet, Chennai-600 015 / www.tnou.ac.in

@ TNOU, 2022, “Functional Analysis” is made available under a Creative

Commons Attribution -Share Alike 4.0 License (International)

http://creativecommons.org/licenses/by-sa/4.0

Printed by:
 Date:02.12.2022
Message

My Dear Beloved Learner,

At the outset, I am pleased to invite you to the Tamil Nadu Open University, a leading Open and Distance
Education Institution in India. I am doubly happy to know that you have joined a Programme in the University
for upgrading your educational qualification. The University has been striving hard to democratize the tertiary
education to all sections of the society with a singular motto "education to anyone at any time" since its inception
in 2002. You can feel proud that you are a learner of the University that has got the UGCs approval vide F.No.l-
6/2018 (DEB) dated 03.10.2018 for of fering 81 UG and PG Programmes through distance mode. As you are
aware, the Government ofTamilnadu vide G.O. (MS) No.107 assured that those, who have secured a Degree in
the Tamil Nadu Open University through 10+2+3 Pattern are duly eligible for government jobs.

With dedicated and uncompromising learner supporting services, the University facilitates the pertinent
learning needs such as (a) socially relevant updated curriculum, (b) resourceful Self-Learning Materials
prepared by the eminent academics, (c) conduction of induction programmes for the first year learners, (d)
arranging counseling classes and examination centres within your vicinity, (e) reliable grievance redressal
mechanism and so on. In order to provide equal education on par with on campus education, the University
holds its Learner Support Centres (LSCs) at 91 Government Arts and Science Colleges across the State, as per
the approval of the Government ofTamilnadu.

At present, the University operates eight Regional Centres each one in Chennai, Coimbatore, Dharmapuri,
Madurai, Tirunelveli, The Nilgiris, Tiruchirappalli and Yilupuram. These Regional Centres facilitate the robust
learner supporting services. Learners can get their essential services at the respective Regional Centre rather
than approaching the Head Quarters. Moreover, our University Website holds a separate page for " Student
Portal'; in which you can get almost all the details relating to your Programme with just clicking the mouse. To
reduce the time and cost, the Online Admission facility is also extended to those who are inclined to join in any
Programme being offered by the University from their places.

At this juncture, I assure that the Tamil Nadu Open University will always extend its constant academic
supports from admission to placement ef fectively.

You can get more details atwww.tnou.ac.in or 044-24306600.

I wish you the best of luck in all of your future endeavours!

Principal Secretary

I
Higher Education Department

I
Ema;1, ,eg;[email protected] Ph: 044-24306645 URL:www.tnou.ac.in
Fax:91-44-24306640
 Tamil Nadu Open University
Department of Mathematics
School of Science,
Chennai – 15
M.Sc., Mathe matics - Syllabus – II year – III Semester (Distance Mode)
COURSE TITLE : FUNCTIONAL ANALYSIS

COURSE CODE : MMSS- 32

COURSE CREDIT : 4

COURSE OBJECTIVES
While studying the FUNCTIONAL ANALYSIS, the Learner shall be able to:

CO 1: To highlight the relationship between algebraic structures and distance

structures.
CO 2: Understand the fundamental results of normed linear spaces and able to prove
the theorems.
CO 3: Recognize Hilbert space and its properties.
CO 4: To impart knowledge about adjoint operator on a Hilbert space.
CO 5: To introduce Operator theory and its application to finite dimensional Spectral
Theory.
COURSE LEARNING OUTCOMES
After completion of the FUNCTIONAL ANALYSIS, the Learner will be able to:
CLO 1: To learn to recognize the fundamental properties of normed spaces
and of the transformations between them.
CLO 2: Ability to prove open mapping theorem, closed graph theorem and uniform
boundedness theorem.
CLO 3: Develop the ideas of orthogonal complement and prove Gram-Schmidt
orthogonalization process.
CLO 4: Discuss the concept of projection, self-adjoint operators, normal and
unitary operators.
CLO 5: Analysis and ability to prove the fixed point theorems.
BLOCK I: BANACH SPACES
Banach Spaces - Definition and Examples - Continuous linear transformations
BLOCK II: FUNDAMENTAL THEOREMS IN NORMED LINEAR SPACES
The Hahn-Banach theorem - The natural imbedding of N in N** - The open mapping
theorem -Closed graph theorem - The conjugate of an operator - Uniform
boundedness theorem.
BLOCK III: HILBERT SPACES
Hilbert Spaces - Definition and Properties - Schwarz inequality - Orthogonal
complements -Orthonormal sets - Bessel’s inequality - Gram–Schmidt
orthogonalization process - The conjugate space H*- Riesz - Representation theorem.
BLOCK IV: OPERATOR ON A HILBERT SPACE
The adjoint of an operator - Self-adjoint operators - Normal and unitary operators -
Projections.
BLOCK V: SPECTRAL AND FIXED POINT THEOREMS
Matrices - Determinants and the spectrum of an operator - Spectral theorem - Fixed
point theorems and some applications to analysis.
REFERENCE BOOKS:
1. Simmons G.F., ”Introduction to Topology and Modern Analysis”, Tata Mc-Graw
Hill Pvt.Ltd., New Delhi, 2011.
2. Kreyszig E.,” Introductory Functional Analysis with Applications, John Wiley &
Sons, New York, 2007.
3. Limaye B. V.,” Functional Analysis”, New Age International Ltd., Publishers,
Second Edition, New Delhi,1996.
4. Coffman C. and Pedrick G., “First Course in Functional Analysis”, Prentice-Hall of
India, New Delhi, 1995.
5. Conway J.B., “A Course in Functional Analysis”, Springer-Verlag, New York,
2008.
6. Bollobas B., ”Linear Analysis”, Cambridge University Press, Indian Edition, New
York, 1999.
7.. Nair M.T., “Functional Analysis, A First course”, Prentice Hall of India, New
Delhi, 2010
Web Resource:
https://www.youtube.com/watch?v=ku1Hwu_HzxM
https://www.youtube.com/watch?v=L5YdadqWqRA
https://www.youtube.com/watch?v=p2YBXe3saKc
https://www.youtube.com/watch?v=hADX7WM-6aE
https://www.youtube.com/watch?v=aXeQsS AI8
https://www.youtube.com/watch?v=jWkzBaJDSmY
https://www.youtube.com/watch?v=zdO4Fe_r4io
https://www.youtube.com/watch?v=umVq9lqVrC8
https://www.youtube.com/watch?v=g-jgGXR3w_8
https://www.youtube.com/watch?v=rc0pSoZpvL8
https://www.youtube.com/watch?v=wOnD1L8gxh0
https://www.youtube.com/watch?v=MIou5JEMW5Q
Contents

1 BANACH SPACES-I 3

1.1 Banach Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.2 Some Important Inequalities ..................................................... 10

1.3 Examples of Banach Spaces ..................................................... 14

2 BANACH SPACES-II 27

2.1 Continuous Linear Transformations ......................................... 28

3 FUNDAMENTAL THEOREMS ON NORMED LINEAR SPACES-I 41

3.1 The Hahn Banach Theorem ...................................................... 42

3.2 The Natural Imbedding of N in N ................................................. 52

4 FUNDAMENTAL THEOREMS ON NORMED LINEAR

SPACES-II 59

4.1 The Open Mapping Theorem .................................................... 60

4.2 The Conjugate of an Operator................................................... 70

5 HILBERT SPACES-I 79

5.1 The De nion and Simple Properties ......................................... 80

5.2 Orthogonal Complements............................................................ 90

6 HILBERT SPACES-I 101

6.1 Orthonormal Sets ....................................................................... 102

6.1.1 Complete Orthonormal Sets ....................................... 108

6.2 The Conjugate Space H .................................................................. 114

vii
viii CONTENTS

7 OPERATOR ON HILBERT SPACE-I 127

7.1 The Adjoint of an Operator .................................................... 128

7.2 Self-Adjoint Operators ................................................................ 136

8 OPERATOR ON HILBERT SPACE-II 145

8.1 Normal and Unitary Operators ............................................... 146

8.2 Projections................................................................................... 153

9 FINITE DIMENSIONAL SPECTRAL THEORY 169

9.1 Eigen Values and Eigen Vectors ............................................ 170

9.2 Matrix of linear transformation .............................................. 172

9.3 Determinants and the Spectrum of an Operator ..................... 179

10 SPECTRAL THEOREM AND FIXED POINT THEOREM 185

10.1 The Spectral Theorem ............................................................ 186

10.2 Fixed Point Theorems ............................................................ 195

Functional Analysis
M.Sc.(Mathematics)-TNOU-II Year-III Sem
CONTENTS 1

Block-I
Unit-1: Banach Spaces-I

Unit-2: Banach Spaces-II

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
2
Block-I

UNIT-1

BANACH SPACES-I

Structure
Objective
Overview
1. 1 Banach Spaces
1. 2 Some Important Inequalities
1. 3 Examples of Banach Spaces
Let us Sum Up
Check Your Progress
Suggested Readings

Overview
In this unit, we will illustrate the basic concepts of normed linear

spaces and some examples of Banach Spaces.

M.Sc.(Mathematics)-II Year-III Sem 3 Functional Analysis

 4 1.1. Banach Spaces

Objectives
After successful completion of this lesson, students will be able to

ˆ Dene Banach space and illustrate with an example.

ˆ Derive Holder's inequality.

1.1. Banach Spaces

In this section, we shall study denitions of normed linear space and

Banach Space. A few examples discussed to illustrating the denitions.

Denition 1.1. A normed linear space is a linear space N in which to each vector
x there corresponds a real number, denoted k xk and called the norm of x , in such
a manner that

(1) k xk 0 and k xk= 0 , x = 0:

(2) k x + yk k xk+kyk ;

(3) k xk= j jk xk .

The non-negative real number k xk is to be thought of as the length of

the vector x. If we regard k xk as a real function dened on N, then this

function is called the norm on N .

Example 1.1. If N is a normed linear space with respect to the distance dened
as d( x; y) = k x yk; N becomes a metric space.

Solution. Now, we shall verify the axioms of metric space.

Axiom 1: x; y 2 N ) x y2N
) k x yk 0
) d( x; y)  0

Axiom 2: d( x; y) = 0 , k x yk= 0
, x y=0
, x=y

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
1.1. Banach Spaces 5

Axiom 3: d( x; y) = k x yk
= k (y x)k
= ky xk
= d(y; x)

Axiom 4: If x; y; z 2 N , then d( x; y) = kx yk
= kx z+z yk
 k x zk+kz yk
Hence; d( x; y)  d( x; y) + d(y; z)

Thus, if follows that d is a metric on N. Hence normed linear space N becomes

a metric space.

Denition 1.2. A complete normed linear space is called a Banach Space.

Property 1.1. If N is a normed linear space and x; y 2 N , then

jk xk kykj  k x yk .

Proof.

k xk= k x y + yk  k x yk+kyk
) k xk kyk  k x yk (1.1)
and kyk k xk  ky xk
 k ( x y)k
 k x yk (1.2)
) From (1.1) and (1.2); we have jk xk kykj  k x yk 

Theorem 1.1. Suppose that N is a normed linear space the norm is a continuous
funciton.

Proof. First, we recall the denition of a continuous function.

If xn !x and f ( xn ) ! f ( x) , then we say that f is a continuous function.

Suppose xn ! x in N , then we shall prove that k xn k! k xk .

Since xn ! x which implies that xn x ! 0 and hence, we have
k xn xk! 0 .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
6 1.1. Banach Spaces

By Property 1.1, jk xn k k xkj  k xn xk .

Hence fk xn kg ! k xk .
Thus, norm is a continuous funciton. 
Theorem 1.2. If N is a normed linear space, then addition and scalar
multiplication are jointly continuous.

Proof. Suppose f xn g ! x and fyn g ! y in N , then

k( xn + yn ) ( x + y)k = k( xn x) + (yn y)k
 k xn xk+kyn yk! 0

Hence f xn + yn g ! x + y .
Thus, addition is a continuous function.
Assume that n ! and xn ! x , then
k n xn xk = k n xn nx + nx xk
= k n ( xn x) + ( n ) xk
 j n jk xn xk+j n jk xk! 0

) f n xn g ! x .

Hence scalar multiplication is a continuous function. 

Theorem 1.3. Let M be a closed linear subspace of a normed linear space N .
If the norm of a coset x + M in the quotient space N= M is dened by
k x + Mk = inf fk x + mk: m 2 Mg (1.3)

then N= M is a normed linear space. Further, if N is a Banach space, then so is

N= M .

Proof. First, we shall prove that N= M is a normed linear space, for this we shall
verify all the axioms of normed linear space.
Axiom 1: Since k x + mk is a non-negative real number and every set of
non-negative real numbers is bounded below, it follows that inffk x + mk: m 2 M g
exists and non-negative, that is
k x + Mk  0 8 x + M 2 N= M

Axiom 2: Let x + M = M (the zero element of N= M ). Then x 2 M . Hence,

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
1.1. Banach Spaces 7

k x + Mk = inf fk x + mk: m 2 M; x 2 Mg
= inf fkyk; y 2 Mg = 0
[* M being a subspace contains zero vector whose norm is
a real number 0]

Thus, x + M = M ) k x + M k= 0:
Converse: Assume that k x + M k= 0 . Then, we will show that x + M = M
(zero element of N= M ).
Now k x + M k= 0 ) inf fk x + mk: x 2 M g = 0
9 a sequence fmk g1k= in M such that
1

k x + mk k ! 0 as k ! 1:
) lim mk = x
k!1
) x2M [* M is closed and fmk g is a sequence in M converging to x]
) x 2 M [* M is a subspace]
) x + M = M [the zero element of N= M]

Thus, we have k x + M k= 0 ) x + M = M (the zero of N/M)

Axiom 3: Let x + M; y + M 2 N= M . Then
k( x + M) + (y + M)k = k( x + y) + Mk
= inf fk x + y + mk: m 2 Mg
= inf fk x + y + m + m0 k; m; m0 2 Mg
[* M is a subspace]
= inf fk x + m + y + m0 k; m; m0 2 Mg
 inf fk x + mk+ky + m0 k; m; m0 2 Mg
= inf fk x + mk; m 2 Mg + inf fky + m0 k; m0 2 Mg
= k x + Mk+ky + Mk

Axiom 4: k ( x + M)k = inf fk ( x + m)k; m 2 Mg

= inf fj jk x + mk: m 2 Mg
= j jinf fk x + mk; m 2 Mg
= j jj x + Mk

Hence N= M is a normed linear space.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
8 1.1. Banach Spaces

It remains to prove that, if N is complete then N= M is also a complete.

We know that a Cauchy sequence is a convergent sequence if and only if it has
a convergent subsequence.
Let xn + M be a Cauchy sequence in N= M so that xn 2 M .
We construct a convergent subsequence of this sequence as follows:
1
Since f sn + M g is a Cauchy sequence, for  = there exists a positive integer
2
n1 such that
1
n; m  n1 ) k( sn + M ) ( sm + M )k<
2

Let S n1 =x 2N 1

!2
1
Similarly for  = 2
there exists a positive integer n2 > n1 such that
!2
1
n; m  n2 ) k( sn + M) ( sm + M)k<
2

!
1 2
Let S n2 = x2 2 N , then k( x2 + M ) ( x1 + M )k< :
2
In general having chosen x1 ; x2 ; : : : xk and n1 ; n2 ; : : : ; nk . let nk+1 > nk be
such that
1 k +1
!

n; m  nk ) k( sn + M ) ( sm + M )k<
2

Let S n +1 k
= xk+ 2 N .
1

Thus, we have obtained a subsequence f xk + M g of f sn + M g such that

1 k
!

k( xk+1 + M) ( xk + M)k< ; k = 1; 2; 3; : : :
2

Now choose any vector y1 in x1 + M i:e:; y1 = x1 + m1 where m1 2 M .

1
since k( x1 + M ) ( x2 + M )k <
2
1
) inf fk x x2 + mk; m 2 M g <
1
2
1
) 9 m 2 M such that k x x2 + m0 k <
0 1
2
1
) k( x + m ) ( x m0 + m1 )k <
1 1 2
2
1
) ky y2 k <
1
2

1
Thus for y1 2 x1 + M , we have obtained y2 2 x2 + M such that ky1 y2 k< .
2

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
1.1. Banach Spaces 9
1
Similarly, we can nd y3 2 x3 + M such that ky2 y3 k< .
22
1
Continuing as above, we get kyn yn+1 k< n .
2
1
Thus, we have a sequence fyn g in N such that kyn yn+1 k< n .
2
We claim that fyn g is a Cauchy sequence in N .
1
For a given  > 0 we can choose a positive integer n0 so large that n0 <.
2 1

Now for m < n and both m; n  n0 , we have

kym yn k = k(ym ym+1 ) + (ym+1 ym+2 ) + : : : + (yn 1 yn )k

 kym ym+1 k+kym+1 ym+2 k+ : : : + kyn 1 yn k
1 1 1
< m
2
+ 2m+ +


+ 2n
1 1

n 1
X
1
= 2n 1
i= m
1
X
1
< n 1
i= m 2
(1=2)m 1 1
= 1
=
1=2 2m < n <
2 1 0 1

Hence, fyn g is a Cauchy sequnece in N , but N is complete and therefore there

exists a vector y in N such that yn ! y .
It remains to prove that xn + M ! y + M 2 N= M . We have

k( xn + M) (y + M)k = k xn y + Mk
= inf fk xn y + mk: m 2 Mg
 k xn + m yk 8m 2 M:

Since yn = xn + mn for some mn 2 M , we conclude that

k( xn + M) (y + M)k kyn yk! 0 (* yn ! y)

Hence xn + M ! y + M 2 N= M .
Thus N= M is complete. 

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
 10 1.2. Some Important Inequalities

1.2. Some Important Inequalities

In this section, we shall discuss some important inequalities like

Cauchy's inequality, Holder's inequality and Minkowsi inequality.

Theorem 1.4 (Cauchy's inequality).

Let x = ( x ; x ; : : : ; xn )
1 2 and y = (y ; y ; : : : ; yn )
1 2 be n -tuples of scalars (real
or complex) and dene the norm by
2 32
6Xn 7
k xk =
6 7
6
6
6
4 j xi j27
7
7
5 : Then
i= 1
n
X
j xi jjyi j  k xkkyk:
i= 1

and deduce that k x + yk k xk+kyk .

Proof. If x = 0; y = 0 then the above result is obvious as both sides becomes

zero. Hence, we assume that x 6= 0; y 6= 0:
Now, dene ai ; bi are as follows:
!2

ai = kj xxikj
!2
jyi j
bi = kyk

Clearly, both ai and bi are non-negative real numbers. Also, we know that the
geometric mean of two non-negative real numbers is less than or equal to their
arithmetic mean so that
p ai + bi
ai bi  for i = 1; 2; : : : ; n
2
" #
j x j jy j 1 j xi j jyi j
Thus, i
i 
2 2

k xk kyk 2 k xk
+ kyk 2 2

Since j xi yi j= j xi jjyi j , summing the above inequalities as i varies from 1 to n , we

get
2Xn Xn 3
6 7
n
6
6 2
6
6 j xi j
277 jyi j 777 " #
X
j xi yi j 1 66666 i= i = 7
7 1 k xk kyk 2 2

k xkkyk
 6666
2 66 k xk
1
+ k2
7
7
7 =
yk 7777 2 k xk
7
1
+ kyk
2
=1 2 2
i= 1 6 6 7 7
4 5

n
X
Hence, j xi yi j  k xkkyk
i= 1

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
1.2. Some Important Inequalities 11

Deduction: 0 1
BXn C
k x + yk = j xi + yi j
2
B
B C
2C
B
B C
C
@ A
i= 1
n
X
= (j xi + yi j) (j xi + yi j)
i= 1
Xn
 (j xi + yi j) (j xi j+jyi j)
i= 1
Xn n
X
 j xi + yi jj xi j+ j xi + yi jjyi j
i= 1 i= 1
 k x + ykk xk+k x + ykkyk (by Cauchy's inequality)
) k x + yk 2
 (k x + yk)(k xk+kyk)

If k x + yk= 0 , then the above is obviously true.

In case k x + yk6= 0 then we can divide both sides by it.
) k x + yk k xk+kyk 
1 1
Lemma 1.1. If 1 < p < 1; 1 < q < 1 such that +
p q
=1 and a; b be two
non-negative real numbers then
a
a1= p b1=q 
p
+ qb :

Proof. If a = 0 and b =0 then above is obviously true and hence we assume

that a 6= 0; b 6= 0 ,
Consider the function f (z) where z  0 dened as under:
f (z) = (1 ) + z z where 0 <  < 1
f 0 (z) =  z = (1 1
z 1 )
since  1 < 0 ) z 1
< 1 if z > 1:
or z 1
>1 if z < 1
) 1 z 1
> 0 if z > 1;
1 z 1
< 0 if z < 1
) f 0 (z) = ve for z < 1
= +ve for z > 1

Thus, f (z) is minimum when z = 1 and also f (1) = 0

f (z) = (1 ) + z z  0
) z  (1 ) + z

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
12 1.2. Some Important Inequalities

a
Put z = in the above relation, where a and b are non-negative real numbers.
b
a a
  (1 ) + 
b b
or a b1   (1 )b + a

1 1 1 1
Put  = ; 1  = so that + = 1.
p q p q
a 1 1
) a1= p b1=q 
p
+ bq where +
p q
= 1:
1 1
If we put p = 2 and q = 2 , then + = 1.
p q
p a+b
Then, the above inequality reduces to ab  .
2
i:e:; G: M:  A: M: 
Theorem 1.5 (Holder's inequality and Minkowski inequality).
Let x = ( x ; x ; : : : ; xn )
1 2 and y = (y ; y ; : : : ; yn )
1 2 denote n -tuples of scalars
(real or complex). Dene
2 31= p
6Xn
k xk p =
6
6
6
6 j xi j p 77777 for p  1: Then
4 5
i= 1

2 31= p 2 31=q
n n n
1
+ q1 = 1
X 6X 7 6X 7
(i) j xi yi j 
6
6
6
6
4 j xi j p 7775
7 6
6
6
6
4 yi j jq 777
5
7
= k xk p kykq if p > 1 and
p
i= 1 i= 1 i= 1
: [Holder0 s inequality]
2 31= p 2 31= p 2 31=q
6Xn 7 6Xn 7 6Xn 7
j xi + yi j p 7757 j xi j p 7757 + jq 777
6 7 6 7 6 7
(ii) 6
6
6
4  6
6
6
4
6
6
6
4 yi j 5
i= 1 i= 1 i= 1
i:e:; k x + yk p  k xk p +kykq (1  p < 1) [ Minkowski inequality]

Proof. (i) If x = 0 or y = 0 , then the inequality is quite obvious, so we assume

that x 6= 0 and y 6= 0 .
By the above lemma, for ai  0 and bi  0 , we have
aip bqi
ai bi 
p
+ q

j xi j
Put ai =
k xk p
and bi = kjyyki j , Thus, we get
q
j xi j jyi j 1 j xi j p 1 jyi jq


k xk p kykq
 +
p k xk pp q kykqq

Above result is true for each i and hence summing from i = 1 to i = n , we get

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
1.2. Some Important Inequalities 13
n
X n
X n
X
j xi jjyi j j xi j p jyi jq
i= 1 i= 1
k xk p kykq
 p + i= 1 q
pk xk p qkykq
k xk p kykqq 1
p
= + =p + 1q = 1
pk xk pp qkykqq
n
X
i:e:; j xi yi j  k xk p kykq
i= 1

This proves (i) .

(ii) If p = 1 , then
n
X n
X n
X
k x + yk 1 = j xi + yi j j xi j+ jyi j= k xk +kyk
1 1

i= 1 i= 1 i= 1

Thus, the inequality holds good for p = 1. Now, assume that p > 1 and put
1
q
= 1 1 so that q > 1 . Then
p
n
X n
X
k x + yk pp = j xi + yi j p = j xi + yi jj xi + yi j p 1

i= 1 i= 1
n
X
 (j xi j+jyi j) j xi + yi j p 1

i= 1
n
X n
X
= j xi jj xi + yi j p + 1
jyi jj xi + yi j p 1

i= 1 i= 1
2 31= p 2 31=q
6Xn Xn

6
6
6
6 j xi j p 77777 6
6
6
6
6 j xi + yi jq( p
7
7
1) 7
77
4 5 4 5
i= 1 i= 1
2 31= p 2 31=q
6Xn Xn
+ 6
6
6
6 yi j j p 77777 6
6
6
6
6 j xi + yi jq( p 1) 7
7
7
7
7
4 5 4 5
i= 1 i= 1
2 31=q 2 31=q
6Xn 7 6Xn 7
= k xk 6
6
6
p4
6 j xi + yi j p) 7775
7
+ kyk 6
6
6
p4
6 j xi + yi j p) 7775
7

i= 1 i= 1
 
= k x + yk pp=q k xk p +kyk p
i:e:; k x + yk p  k xk p +kyk p

This completes the proof of (ii) . 

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
 14 1.3. Examples of Banach Spaces

1.3. Examples of Banach Spaces

In this section we shall describe some of the main examples of Banach

Spaces. In each of these, the linear operations are understood to be dene

either coordinatewise or pointwise, whichever is appropriate is a Banach

space.

Example 1.2. The space R and C -the real numbers and the complex numbers
are the simplest of all normed linear spaces. The norm of a number x is of course
dened k xk= j xj , and each space is a Banach space.
Solution. Let us verify all the axioms of normed linear space one by one.
Axiom 1: kx yk = jx yj 0

Axiom 2: kx yk = 0 , j x yj= 0 , x = y:

Axiom 3: Let z; w 2 C and let z and w denote their complex conjucates. Then
kz + wk 2
= (z + w)(z + w)
= zz + zw + zw + ww
 jzj +2jzjjwj+jwj 2 2

= (jzj+jwj) 2

Hence jz + wj  jzj+jwj

Axiom 4: k xk = j jj xj= j jk xk:

Thus both C and R are both normed linear spaces. Also, we know that these
spaces are complete. Hence C and R are complete normed linear spaces.
i:e:; R and C are Banach spaces. È
Example 1.3. The linear space Rn and C n of all n -tuples x = ( x1 ; x2 ; : : : : xn ) of
real and complex numbers are Banach known as Euclidean and Unitary spaces
respectively. The norm of x is dened by
0 11=2
BXn C
k xk =
B C
B
B
B
@ j xi j
2C
C C
A
i= 1

Solution. First. we shall prove that it is a normed linear space. For this, we shall
verify all the postulates of normed linear spaces.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
1.3. Examples of Banach Spaces 15

Axiom 1: j xi j  0 8i
) j xi j 2
 0 8i
n
X
) j xi j2
 0
i= 1
0 11=2
BXn C
B C
) B
B
B
@ j xi j 2C
C
C
A  0
i= 1
) k xk  0
0 11=2
BXn C
k xk= 0 =0
B C
Axiom 2: , B
B
B
@ j xi j 2C
C
C
A
i= 1
n
X
, j xi j = 0
2

i= 1
, j xi j = 0 8i
2

, j xi j= 0 8i
, xi = 0 8i
, x = ( x ; x ; : : : ; xn ) = (0; 0; : : : ; 0)
1 2

, x=0
0 11=2 0 11=2
BXn C B n
X C
Axiom 3: k xk = B
B
B
B
@ j xi j
C
2C
C
C
A = j jB
B
B
B
@
2
j xi j
2C
C
C
C
A
i= 1 i= 1
0 11=2
BXn C
= j j B
B
B
B
@ j xi j 2C
C
C
C
A
i= 1
= j jk xk

Axiom 4: Let x = ( x1 ; x2 ; : : : ; xn ) and y = (y ; y ; : : : ; yn )

1 2 be any two numbers
in Rn or C n . then
k x + yk 2
= k( x ; x ; : : : ; xn ) + (y ; y ; : : : ; yn )k
1 2 1 2
2

= k x + y ; x + y ; : : : ; xn + yn k
1 1 2 2
2

n
X n
X
= j xi + yi j = 2
j xi + yi jj xi + yi j
i= 1 i= 1
Xn
 j xi + yi j(j xi j+jyi j)
i= 1
Xn n
X
= j xi + yi jj xi j+ j xi + yi jjyi j
i= 1 i= 1
= k x + ykk xk+k x + ykkyk
= k x + yk(k xk+kyk)

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
16 1.3. Examples of Banach Spaces

If k x + yk= 0 , then the above inequality is obviously true.

If k x + yk6= 0 , then we can divide both sides by k x + yk , we get
k x + yk  k xk+kyk

Thus, Rn or C n is a normed linear spaces.

It remains to prove that the completeness of C n or Rn .
Let f xn g be a Cauchy sequence in C n (or Rn ) . Since each xm is an n -tuple of
complex (or) real numbers, we shall write it as
 
xm = x1m ; x2m ; : : : ; xnm

so that xkm is the kth coordinate of xm . Let  > 0 be given. Since f xn g is a

Cauchy sequence, there exists a positive integer m0 such that
l; m  m0 ) k xm xl k< 
) k xm xl k2 <  2
n
X
) j xi m
( )
xi(l) j2 <  2 (1.4)
i= 1
) j xi(m) xi(l) j2 <  2 (i = 1; 2; : : : ; n)
) j xi(m) xi(l) j<  (i = 1; 2; : : : ; n)

n o
Thus, xi(m) be a Cauchy sequence of complex (or real) numbers for each xed
but arbitrary i . Since C or R is complete, each of these sequence converges to
a point say zi in C (or R) so that
lim x(m) = zi (i = 1; 2; : : : ; n)
m!1 i

where each zi belongs to C n (or Rn ) , so that

z = (z1 ; z2 ); : : : ; zn ) 2 C n (or Rn ) .

Now, we shall prove that the Cauchy sequence f xm g in C n (or Rn ) converges

to z 2 C n (or Rn ) . From (1.4) we have
n
X
j xi m
( )
xi(l) j2 <  2 (1.5)
i= 1

Let l ! 1 in (1.5), n
X
j xi m
( )
zi j2 <  2
i= 1
) k xm zk <  2 2

) k xm zk < 

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
1.3. Examples of Banach Spaces 17

Thus, f xm g is a Cauchy sequence in C n (or Rn ) converges to a point z in

C n (or Rn ) . Hence C n (or Rn ) are complete normed linear spaces and consequently
they are Banach spaces. È
Example 1.4. let p be a real number such that 1  p < 1 . We denote by lnp the
space of all n -tuples x = ( x1 ; x2 ; : : : ; xn ) of scalars, with the norm dened by
0 11= p
BXn C
k xk p = j xi j p CCCA
B
B C
B
B
@
i= 1

Then lnp is a Banach space.

Solution. First. we shall prove that it is a normed linear space. For this, we shall
verify all the postulates of normed linear spaces.
Axiom 1: j xi j  0 8i
) j xi j p  0 8i
n
X
) j xi j p  0
i= 1
0 11= p
BXn C
j xi j p CCCA
B C
) B
B
B
@  0
i= 1
) k xk p  0

0 11= p
BXn C
k xk p = 0 j xi j p CCAC =0
B C
Axiom 2: , B
B
B
@
i= 1
n
X
, j xi j p = 0
i= 1
, j xi j p = 0
8i
, j xi j= 0 8i
, xi = 0 8i
, x = ( x ; x ; : : : ; xn ) = (0; 0; : : : ; 0)
1 2

, x=0
0 11= p
BXn C
Axiom 3: k xk p = B
B
B
B
@ j xi j p CCCA
C

i= 1
0 11= p
B n
X
= j j p j xi j
B
B
B
B
p CCCCC
@ A
i= 1
0 11= p
BXn C
= j j B
B
B
B
@ j xi j p CCCA
C

i= 1
= j jk xk p

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
18 1.3. Examples of Banach Spaces

Axiom 4: Let x = ( x1 ; x2 ; : : : ; xn ) and y = (y1 ; y2 ; : : : ; yn ) be any two scalars in

lnp . then p
X n Xn
k x + yk p = j xi + yi j =
p p 1 j xi + yi jj xi + yi j
i= 1 i= 1
n
X
 (j xi j+jyi j) j xi + yi j p 1

i= 1
n
X n
X
= j xi jj xi + yi j p + 1
jyi jj xi + yi j p 1

i= 1 i= 1
2 31= p 2 31=q
6Xn Xn

6
6
6
6 j xi j p 77777 6
6
6
6
6 j xi + yi jq p(
7
7
1) 7
7
7
4 5 4 5
i= 1 i= 1
2 31= p 2 31=q
6Xn Xn
+ 6
6
6
6 yi j j p 77777 6
6
6
6
6 j xi + yi jq( p
7
7
1) 7
77
4 5 4 5
i= 1 i= 1
2 31=q 2 31=q
6Xn 7 6Xn 7
= k xk 6
6
6
p 64 j xi + yi j p) 7757
7
+ kyk 6
6
6
p 64 j xi + yi j p) 7757
7

i= 1 i= 1
 
= k x + yk pp=q k xk p +kyk p
i:e:; k x + yk p  k xk p +kyk p

Thus, lnp is a normed linear spaces.

Let f xn g be a Cauchy sequence in lnp .Since each xm is an n -tuple of complex
(or) real numbers, we shall write it as
 
xm = x1m ; x2m ; : : : ; xnm

so that xkm is the kth coordinate of xm . Let  > 0 be given. Since f xn g is a

Cauchy sequence, there exists a positive integer m0 such that
l; m  m0 ) k xm xl k p < 
) k xm xl k p <  p
n
X
) j xi m
( )
xi(l) j p <  p (1.6)
i= 1
) j xi(m) xi(l) j p <  p (i = 1; 2; : : : ; n)
) j xi(m) xi(l) j<  (i = 1; 2; : : : ; n)

n o
Thus, xi(m) be a Cauchy sequence of complex (or real) numbers for each xed
but arbitrary i . Since C or R is complete, each of these sequence converges to
a point say zi in C (or R) so that
lim x(m) = zi (i = 1; 2; : : : ; n)
m!1 i

where each zi belongs to lnp , so that

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
1.3. Examples of Banach Spaces 19

z = (z1 ; z2 ); : : : ; zn ) 2 lnp .

Now, we shall prove that the Cauchy sequence f xm g in lnp converges to z 2 lnp .
From (1.6) we have
n
X
j xi m
( )
xi(l) j p <  p (1.7)
i= 1

Let l ! 1 in (1.7),
n
X
j xi m
( )
zi j p <  p
i= 1
) k xm zk pp <  p
) k xm zk p < 

Thus, f xm g is a Cauchy sequence in lnp converges to a point z in lnp .

Hence lnp are complete normed linear spaces and consequently they are Banach
spaces. È
Example 1.5. Let p be a real number such that 1  p < 1 . We denote by l p the
1
X
space of all sequences x = f x1 ; x2 ; : : : ; xn ; : : :g of scalars such that j xn j p < 1 ,
n= 1
with the norm dened by
0 11= p
BX1 C
k xk p = j xn j p CCAC
B
B C
B
B
@
n= 1

Then l p is a Banach space.

Solution. First. we shall prove that it is a normed linear space. For this, we shall
verify all the postulates of normed linear spaces.
Axiom 1: j xi j  0 8i
) j xi j p  0 8i
1
X
) j xi j p  0
i= 1
0 11= p
BX1 C
j xi j p CCCA
B C
) B
B
B
@  0
i= 1
) k xk p  0
0 11= p
BX1 C
k xk p = 0 j xi j p CCCA =0
B C
Axiom 2: , B
B
B
@
i= 1
1
X
, j xi j p = 0
i= 1

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
 20 1.3. Examples of Banach Spaces

, j xi j p = 0 8i
, j xi j= 0 8i
, xi = 0 8i
, x = ( x ; x ; : : : ; xn ; : : :) = (0; 0; : : : ; 0; : : :)
1 2

, x=0
0 11= p
BX1 C
Axiom 3: k xk p = B
B
B
B
@ j xi j p CCCA
C

i= 1
0 11= p
B 1
X C
= j B
B
B
B
@ jp j xi j p CCCA
C

i= 1
0 11= p
BX1 C
= j j B
B
B
B
@ j xi j p CCCA
C

i= 1
= j jk xk p

Axiom 4: Let x = ( x1 ; x2 ; : : : ; xn ; : : :) and y = (y1 ; y2 ; : : : ; yn ; : : :) be any two

sequence in l p . then
2 31= p 2 31=q 2 31=q 2 31=q
n
X 6Xn 7 6Xn 7 6Xn 7 6Xn 7
j xi + yi j p j xi j p 7775 j xi + yi j p 7775 + jq 777 j xi + yi j p 7775
6 7 6 7 6 7 6 7
 6
6
6
4
6
6
6
4
6
6
6
4 yi j 5
6
6
6
4
i= 1 i= 1 i= 1 i= 1 i= 1
2 31= p 2 31=q 2 31=q 2 31=q
6X1 X1 X1 X1

6
6
6
6 j xi j p 77777 6
j xi + yi j
6
6
6
6 + j p 77777 6
6
6
6
6 yi jq 77777 6
6
6
6
6
7
j xi + yi j p 7775
7
4 5 4 5 4 5 4
i= 1 i= 1 i= 1 i= 1
= p =q
k xk p k x + yk p +kyk p k x + yk p p =q

Hence letting n ! 1 , we obtain

1
X h i
k x + yk pp = j xi + yi j p  k xk p +kyk p k x + yk pp=q
i= 1
i:e:; k x + yk pp p=q  k xk p +kyk p
i:e:; k x + yk p  k xk p +kyk p

Thus, l p is a normed linear spaces.

Let f xn g be a Cauchy sequence in l p .Since each xm is a sequence of scalars,
we shall write it as
n o
xm = x1m ; x2m ; : : : ; xnm ; : : :

1
X
p
where xn(m) < 1:
n= 1
Since f xn g be a Cauchy sequence in l p . Let  > 0 be given. Since f xn g is a

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
1.3. Examples of Banach Spaces 21

Cauchy sequence, there exists a positive integer m0 such that

n; m  m0 ) k xn xn k p < 
in particular; n  m0 ) k xn xm0 k p < 

Thus, if n  m0 , then
k xn k p = k xn xm + xm k p o o

 k xn xm k p +k xm k p o o

  + k xm k p = A where A =  + k xm k p
o o

i:e:; k xn k p < A for n  mo (1.8)

n o
Thus for xed i , the sequence xi(n) is a Cauchy sequence in C or R and
consequently it must converge to a number say zi . Let z = fz1 ; z2 ; : : : ; zn ; : : :g .
We now assert that z 2 l p and that the Cauchy sequence f xn g converges to z .
First we shall prove that z 2 l p . From (1.8), we have
n  m0 k xn k pp < Ap
1
X
) j xi n j p <
( )
p
i= 1

Hence for any positive integer L , we have

L
X
p
xi(n) < Ap (n  m0 ) (1.9)
i= 1

But for i = 1; 2; : : : ; L we have xi(n) ! zi as n ! 1 .

Hence letting n ! 1 (1.9), we obtain
L
X
jzi j p  Ap (L = 1; 2; 3; : : :)
i= 1

L
X
It follows that jzi j p  A p < 1:
i= 1
This proves that z = fzn g is in l p .
For n; m  m0
k xn xm k pp <  p
1
X
p
) xi(n) xi(m) < p
i= 1

Hence for every positive integer L , we have

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
 22 1.3. Examples of Banach Spaces

L
X
p
xi(n) xi(m) < p (n; m  m0 )
i= 1

Letting m ! 1 and using lim xi(m) = zi , we obtain

m!1
L
X
p
xi(n) xi(m) < p 8n  m 0

i= 1

and all positive integer L and so

1
X
p
xi(n) xi(m) < p
i= 1
) k xn zk pp <  p
) k xn zk <  (n  m ) 0

It follows that Cauchy sequence f xn g converges to z 2 l p . Hence l p is complete

and so it is a Banach space. È
Example 1.6. The space of all n -tuples x = ( x1 ; x2 ; : : : ; xn ) of scalars with norm
of x dened as
k xk1 = maxfj x j; j x j; : : : ; j xn jg
1 2

n and it is a Banach space.

This space is denoted by l1
n is a normed linear space.
Solution. First, we shall prove that l1
Axiom 1: j xi j  0 8i
) maxfj x j; j x j; : : : ; j xn jg  0
1 2

) k xk1  0

Axiom 2: k xk1 = 0 , maxfj x j; j x j; : : : ; j xn jg = 0

1 2

, j x j= 0; j x j= 0; : : : ; j xn j= 0
1 2

, x = 0; x = 0; : : : ; xn = 0
1 2

, x = ( x ; x ; : : : ; xn ) = (0; 0; : : : ; 0)
1 2

, x=0

Axiom 3: If is any scalar, then

k xk1 = maxfj x j; j x j; : : : ; j xn jg
1 2

= j jmaxfj x j; j x j; : : : ; j xn jg
1 2

= j jk xk1

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
1.3. Examples of Banach Spaces 23

Axiom 4: Let x = ( x1 ; x2 ; : : : ; xn ) and y = (y1 ; y2 ; : : : ; yn ) be any two two

n . then
elements in l1
x+y = maxfj x1 + y1 j; j x2 + y2 j; : : : ; j xn + yn jg
k x + yk1  maxfj x j+jy j; j x j+jy j; : : : ; j xn j+jyn jg
1 1 2 2

 maxfj x j; j x j; : : : ; j xn jg + maxfjy j; jy j; : : : ; +jyn jg

1 2 1 2

 k xk1 +kyk1

n is a normed linear spaces.

Thus, l1
Now. we shall prove that it is a complete space. Let f xn g be a Cauchy sequence
n . since each x is an n -tuple of scalars, we shall write
in l1 m
 
xm = x1(m) ; x2(m) ; : : : ; xn(m) :

Let  > 0 be given. Then there exists a positive integer m0 such that
l; m  m0 ) k xm xi k< 
n o
) max x m x l ; x m x l ; : : : ;
(
1
) ( )
1
(
2
) ( )
2
xn(m) xn(l) < 
) xi m xi l <  i = 1; 2; : : : ; n
( ) ( )

n o
This shows that for xed i; xi(m) is a Cauchy sequence of complex
or real numbers. Since C or R is complete, it must converge to some
point zi 2 C (or R) . We assert that the Cauchy sequence f xn g converges to
z = (z1 ; z2 ; : : : ; zn ) . where each zi belongs to C or R .

We knwo that j xi(m) xi(l) j<  8i .

Let l ! 1 in the above so that lim xi(l) = zi .
l!1
) we have j xi m ( )
zi j<  8i .
n o
) max j x m (
1
)
z1 j< ; j x2(m) z2 j< ; : : : ; j xn(m) zn j< 

) k xm zk1 <  .
n converges to a point z 2 ln .
This shows that the Cauchy sequence f xm g in l1 1
n
Therefore l1 is a complete normed linear space. i:e:; It is a Banach space. È
Note 1.1. The linear space of all bounded sequences x = f x1 ; x2 ; : : : ; g of scalars
in l1 with norm is dened as
k xk = supj xn j

Then l1 is a Banach space.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
24 1.3. Examples of Banach Spaces

Example 1.7. The space C (X ) is dened as the linear space of all bounded
continuous scalar valued functions dened on X and it is a Banach Space with
the norm of a functon f 2 C (X ) dened as
k f k = sup fj f ( x)j: x 2 X g

The addition and scalar multiplication dened as

(f + g)( x) = f ( x) + g( x);
( f )x = f ( x)

Solution.

(1) k f k is clearly non-negative as

j f ( x)j> 0 8 x 2 X:
(2) k f k= 0 , sup fj f ( x)j= 0 8 x 2 X g
, j f ( x)j= 0 8 x 2 X
, f ( x) = 0 8 x 2 X
, f = 0 (i:e:; zero function)
(3) k f k = sup fj( f )( x)j: x 2 X g
= sup fj jj f ( x)j: x 2 Xg
= j jsup fj f ( x)j: x 2 Xg
= j jk f k
(4) k f + gk = sup fj( f + g) xj: x 2 X g
= sup fj f ( x) + g( x)j: x 2 Xg
 sup fj f ( x)j: x 2 X g + sup fjg( x)j: x 2 X g
= k f k+kgk

Hence C (X ) is a normed linear space.

Now,we shall prove that C (X ) is a complete space.
Let f fm g be a Cauchy sequence in C (X ) therefore for given  > 0 there exists
n0 such that for m; p  n0 .
) k fm fn k< 
) sup fj( fm fn )( x)j: x 2 X g < 
) sup fj( fm ( x) fn ( x))j: x 2 X g < 
) j fm ( x) fn ( x)j<  8 x 2 X

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
1.3. Examples of Banach Spaces 25

Thus, the Cauchy's condition for uniform convergence of the sequence of

bounded continuous scalar valued functions. Therefore sequences f fm g must
converge to a bounded continuous f on X . Hence C (X ) is complete and it
is a Banach space. È
Note 1.2. The space B(X ) is dened as the linear space of all bounded scalar
valued functions dened on X and it is a Banach space with the norm of the
function f 2 B(X ) dened as
k f k = sup fj f ( x)j x 2 X g

Let us Sum Up:

In this unit, the students acquired knowledge to

ˆ the concepts of normed linear space and Banach space.

ˆ illustrate the examples of Banach space.

Check Your Progress:

1. State and Prove Holder's inequality.

2. State and Prove Cauchy's inequality.

3. Let N be a non-zero normed linear space, and prove that N is a

Banach space if and only if f x : k xk= 1g is complete.

4. Let a Banach space B be the direct sum of the linear subspaces M

and N, so that B = M  N . If z = x + y is the unique expression of

a vector z in B as the sum of vectors x and y in M and N, then

a new norm can be dened on the linear space B by kzk

0 k xk kyk . = +
0
Prove that this actually is a norm. If B symbolizes the linear space

B equipped with this new norm, prove that B0 is a Banach space if

M and N are closed in B.

5. Prove that norm is a continuous function.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
26 1.3. Examples of Banach Spaces

Suggested Readings:

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

Tata Mc-Graw Hill Pvt. Ltd., New Delhi, 2011.

2. Kreyszig E.,” Introductory Functional Analysis with Applications,

John Wiley & Sons, New York, 2007.

3. Limaye B. V.,” Functional Analysis”, New Age International Ltd.,

Publishers, Second Edition, New Delhi,1996.

4. Co man C. and Pedrick G., “First Course in Functional Analysis”,

Prentice-Hall of India, New Delhi, 1995..

5. Conway J.B., “A Course in Functional Analysis”, Springer-Verlag,

New York, 2008.

6. Bollobas B.,”Linear Analysis”, Cambridge University Press, Indian

Edition, New York, 1999.

7. Nair M.T., “Functional Analysis, A First course”, Prentice Hall of

India, New Delhi, 2010.

Web Resources:

// / / / /
1. https: archive.nptel.ac.in courses 111 106 111106147 /
// / =
2. https: www.youtube.com watch?v bgQ7Wn-etK0

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
Block-I

UNIT-2

BANACH SPACES-II

Structure
Objective
Overview
2. 1 Continuous Linear Transformations
Let us Sum Up
Check Your Progress
Suggested Readings

Overview
In this unit, we will illustrate the basic concepts of normed linear

spaces and some examples of Banach Spaces.

Objectives
After successful completion of this lesson, students will be able to

ˆ discuss continuous linear transformations.

ˆ understand the concept of bounded linear transformations.

M.Sc.(Mathematics)-II Year-III Sem 27 Functional Analysis

 28 2.1. Continuous Linear Transformations

2.1. Continuous Linear Transformations

Let N and N0 be normed linear spaces with the same scalars, and

let T be a linear transformation of N into N0 . When we say that T is

continuous, we mean that it is continuous as a mapping of the metric space

N into the metric space N 0 , i:e:; xn ! x ) T ( xn ) ! T ( x) . Our main aim

in this section is to convert the requirement of continuity into several more

useful equivalent forms and to show that the set of all continuous linear

transformations N into N0 can itself made into a normed linear space in a

natural way.

Theorem 2.1. Let N and N 0 be normed linear spaces and T a linear

transformation from N into N0 . Then the following conditions on T are all
equivalent to one another:

1. T is continuous.

2. T is continuous at the origin, in the sense that xn ! 0 ) T ( xn ) ! 0 ;

3. there exists a real number K  0 with the property that kT ( x)k K k xk for
every x 2 N ;

4. if S = f x : k xk 1g is the closed unit sphere in N , then its image T (S ) is

a bounded set in N 0 .

Proof.
(1) , (2) : Let T be continuous and suppose f xn g be a sequence in N such
that xn ! 0 as n ! 1: Then by the condition of continuity of T , we have
T ( xn ) ! T (0) = 0: Hence T is continuous at the origin.

Conversely, let T be continuous at the origin and f xn g be any sequence in N

such that xn ! x 2 N . Then
xn x!0 ) T ( xn x) = 0 [* T is continuous at origin]
) T ( xn ) T ( x) = 0
) T ( xn ) = T ( x)

This shows that T is a continous mapping.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
2.1. Continuous Linear Transformations 29

(2) , (3) : Given that T is continuous at the origin.

i:e:; xn ! 0 ) T ( xn ) ! 0:
Now, we shall prove that T is bounded, i:e:; there exists a real number K  0
such that
kT ( x)k  K k xk 8x 2 N

Let us assume that no such K exists satisfying the condition kT ( x)k K k xk .

Then for each positive integer n , we can nd a vector xn such that
kT ( x)k > nk xk
1
) kT ( xn )k > 1
nk xn k
1
) T ( xn ) > 1
nk xn k
!
xn
) T > 1
nk xn k

xn kx k 1
Put yn = . Then kyn k= n = ! 0 as n ! 1 .
nk xn k nk xn k n
Thus yn ! 0 , but T (yn ) does not tends to 0 , since kT (yn )k> 1 .
Hence T is not continuous at the origin which is a contradiction. Hence T
must be bounded.
Conversely, let T be bounded so that there exists a real number K  0 such
that
kT ( x)k  K k xk 8x 2 N (2.1)

Now, our aim is to prove that T is continuous at the origin.

Let f xn g be any sequence in N such that xn ! 0 . Then
k xn k ! k0k= 0: (2.2)

From (2.1), we have

kT ( xn )k  K k xn k 8x 2 N (2.3)

From (2.2) and (2.3), we have kT ( xn )k! 0 which implies that T ( xn ) ! 0 and
consequently T is continuous at the origin.
(3) , (4) : Assume that kT ( x)k K k xk for every x 2 N and let x be any point

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
30 2.1. Continuous Linear Transformations

in the closed unit sphere S so that k xk 1 . Then kT ( x)k K for all x 2 S ,
which implies that T (S ) is bounded set in N 0 .
Conversely, let T (S ) be bounded so that there exist a real number K  0 such
that
kT ( x)k  K 8x 2 S (2.4)

If k xk= 0 , then kT ( x)k= 0 and so clearly kT ( x)k K k xk .

x
If k xk6= 0 , then 2S.
k xk
Thus from (2.4), we have !
x
T  K
k xk
1
) kT ( x)k  K
k xk
kT ( x)k  K k xk

Thus, kT ( x)k K k xk 8 x 2 N . Hence T is bounded.

This completes the proof of the theorem. 
Denition 2.1. Suppose T : N ! N 0 is a linear transformation such that there
exists a real number K  0 such that
kT ( x)k  K k xk

for every x 2 N , then K is called a bounded for T and such a T is often referred
to as a bounded linear transformation

Note 2.1. According to theorem (2.1), we have T is bounded if and only if T is

continuous, so these two adjectives can be used interchangeably.

Denition 2.2. Suppose T is a linear transformation, then the norm of T is

dened by
kT k = supfkT ( x)k: k xk 1g (2.5)

When N 6= 0 , this formula is equivalent to

kT k = supfkT ( x)k: k xk= 1g (2.6)

Remark 2.1. If T : N ! N 0 is a continuous linear transformation then

kT ( x)k K k xk for all x 2 N .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
2.1. Continuous Linear Transformations 31

Proof. From (2.1) there exists a real number K  0 such that kT ( x)k K k xk
for all x 2 N . Hence the set of all bounds for T equals to the set of all radii of
closed sphere centred on the origin which contains T (S ) . Hence
kT k = inf fK : K  0 and kT ( x)k K k xk 8 xg

Hence kT ( x)k K k xk for all x 2 N . 

Notation:
We denote the set of all continuous (or bounded) linear transformations

of N into N0 by B(N; N 0) , where the letter “B00 is intended to suggest

the adjective “bounded”.

Theorem 2.2. If N and N 0 are normed linear spaces, then the set B ( N; N 0 )
of all continuous linear transformations of N into N0 is itself a normed linear
space with respect to the pointwise linear operations
(T 1 + T 2 )( x) = T 1 ( x) + T 2 ( x);
( T )( x) = T ( x)

and the norm dened by

kT k = supfkT ( x)k: k xk 1g

Further, if N 0 is a Banach space, then B(N; N 0) is also a Banach space.

Proof. We know that the set S of all linear transformations from one linear
space to another linear space is itself a linear space with respect to the pointwise
B
operations. Therefore in order to prove that (N; N 0 ) is a linear space, it is
sucies to prove that (N; N 0 ) is a subspace of S .
B
Let T 1 ; T 2 2 (N; N 0 ) . Then T 1 ; T 2 are bounded and so there exist real
number K1  0 and K2  0 such that
kT ( x)k 
1 K1 k xk
and kT ( x)k 
2 K2 k xk

for all x 2 N .
If ; are any two scalars, then

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
 32 2.1. Continuous Linear Transformations

k( T 1 + T 2 ) ( x)k = k( T ) ( x) + ( T ) ( x)k
1 2

= k T ( x) + T ( x)k
1 2

 j jkT ( x)k+j jkT ( x)k

1 2

 j jK k xk+j jK k xk
1 2

= (j jK + j jK ) k xk
1 2

Thus, T 1 + T 2 is bounded and so T 1 + T 2 2 B ( N; N 0 ) .

This proves thatB(N; N 0) is a linear subspace of S .
Next, our wish is to prove that B (N; N 0 ) is a normed linear space. For this,
we shall verifty the norm postulates one by one.
Let T 2 B(N; N 0) then by denition
kT k = supfkT ( x)k: k xk 1g

Axiom 1: Since kT k= supfkT ( x)k: k xk 1g and kT ( x)k 0 .

Hence, we conclude that kT k 0 .
Axiom 2: kT k = supfk(T ( x)k: k xk 1; x 2 N g )
= sup kTk(xxk)k : x =6 0; x 2 N
( )
kT ( x)k
kT k= 0 , sup :x= 6 0; x 2 N = 0
k xk
1
, kT ( x)k= 0 8 x 2 N; x 6= 0
k xk
, kT ( x)k= 0 8 x 2 N; x 6= 0
, T ( x) = 0; 8 x 2 N; x 6= 0
, T = 0 i:e; zero transformation

Axiom 3: k T k = supfk( T ) xk: k xk 1g

= supfj jkT ( x)k: k xk 1g
= j jsupfkT ( x)k: k xk 1g
= j jkT k

Axiom 4: kT 1 + T 2 k = supfk(T 1 + T 2 ) xk: k xk 1g

= supfkT 1 ( x) + T 2 ( x)k: k xk 1g
 supfkT ( x)k: k xk 1g + supfkT ( x)k: k xk 1g
1 2

= kT k+kT k
1 2

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
2.1. Continuous Linear Transformations 33

Hence kT 1 + T 2 k kT 1 k+kT 2 .
Hence B(N; N 0) is a normed linear space.
Next, we wish to prove that B (N; N 0 ) is a Banach space whenever N 0 is a
Banach space.
In otherwords, we have to prove that B(N; N 0) is complete whenever N 0 is
complete.
Let fT n g be a Cauchy sequence in B ( N; N 0 ) .
) kTm Tnk <  8m; n  n 0 (2.7)

Now for any xed x in N . Then

kT m ( x) T n ( x)k = k(Tm T n )( x)k
 kT m T n kk( x)k (2.8)

From (2.7) and (2.8), it follows that kT m T n k! 0 .

) fTn g is a Cauchy sequence in N 0 for each x 2 N . But N 0 is complete

and hence this sequence is convergent.
Therefore there exists a vector y = T ( x) in N 0 such that T n ( x) converges to
T ( x) for xed x . i:e:; lim T n ( x) = T ( x) .

Dene T : N ! N 0 by
T ( x) = lim T n ( x)

Now, we shall prove that T is linear and bounded.

T is linear: Let x; y 2 N and ; be any scalars. Then
T ( x + y) = lim T n ( x + y)
= lim T n ( x) + T n (y)
 

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

= T ( x) + T (y)

Thus, T is linear.
T is bounded:
kT ( x)k = klim T n ( x)k
= limkTn ( x)k
 lim (kT n kk xk)
 sup (kT n k) k xk (2.9)

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
34 2.1. Continuous Linear Transformations

But fT n g is a Cauchy sequence.

) kTm T n k<  .

) jkTm k kTn kj  kTm T n k< :

Thus, the above relation shows that kT n k is a Cauchy sequence on real line
and hence convergent and bounded so that there exists a K  0 such that
) supkTn k  K for some K 0

Hence, from (2.9), we get

kT ( x)k  sup (kT n k) k xk
 K k xk 8 x

Hence T is bounded.
) T is both linear and bounded and hence T 2 B ( N; N 0 ) .
Finally, we show that T n ! T .
Let  > 0 be given. Since fT n g is a Cauchy sequence there exists a positive
integer mo such that
m; n  mo ) kT m T n k< =2 (2.10)

Now, let x 2 N be such that k xk 1: Then since T n ( x) ! T ( x) . We can choose

a positive integer mx > mo such that

kT ( x) T m ( x)k< =2
x
(2.11)

Hence for all n  mo and k xk 1 , we have

kT n ( x) T ( x)k = kTn ( x) T m ( x) + T m ( x)
x x
T ( x)k
 kT n ( x) T m ( x)k+kT m ( x) T ( x)k
x x

= kTn ( x) Tm ( x)k+kTm ( x) T ( x)k

x x

= k(Tn Tm )( x)k+k(Tm T )( x)k

x x

 k(T n T m )kk xk+k(T m T )kk xk

x x

 k(T n T m )k+k(T m T )k (* k xk 1)

x x

< =2 + =2 = 

Thus, kT n ( x) T ( x)k<  8n  mo and x 2 N such that k xk 1 .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
2.1. Continuous Linear Transformations 35

Hence supfkT n ( x) T ( x)k: x 2 N; k xk 1gg <  (n  mo ):

i:e:; supfk(T n T )( x)k: x 2 N; k xk 1gg <  (n  mo ):

i:e:; kT n T k<  n  mo :

It follows that T n ! T and hence B(N; N 0) is complete.

This completes the proof of the theorem. 
Denition 2.3. Let N be a normed linear space. We call a continuous linear
transformation of N into itself an operator on N , and we denote the normed
B
linear space of all operators on N by (N ) instead of (N; N ) . B
Remark 2.2. If N is a Banach space then B(N ) is also a Banach space. Further
if we dene a multiplication in (N ) as B
T
T 0 ( x) = T T 0 ( x)

then B(N ) becomes an algebra, in which multiplication to the norm by

kT T 0 k  kT kkT 0 k

For if,
kT T 0 k = supfk T T 0 ( x)k: k xk 1g

= supfkT T 0 ( x)
k: k xk 1g

 supfkT kkT 0 ( x)k: k xk 1g
= kT ksupfkT 0 ( x)k: k xk 1g
= kT kkT 0 k

Remark 2.3. By Theorem 1.2, addition and scalar multiplication in (N ) are B

B
jointly continuous as (N ) is a normed linear space. From the previous remark,
we conclude that multiplication is also jointly continuous.
i:e:; fT n g ! T and fT n0 g ! T 0


then T n T n0 ! T T 0 .
For if,
kT n T n0 T T 0k = kTn Tn0 TnT 0 + TnT 0 T T 0k
= kTn Tn0 T0


+ (Tn T ) T 0k
= kTn Tn0 T 0 + (T n


T ) T 0k
 kT n kkT n0 T 0 k+kT n T kkT 0 k! 0
) T n T n0 ! TT0

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
36 2.1. Continuous Linear Transformations

Remark 2.4. If N 6= f0g , then the identity transformation I is an identity for the
B
algebra (N ) . In this case, we have kI k= 1 .
For if,
kI k = supfI ( x) : k xk 1g
= supf x : k xk 1g = 1

Denition 2.4. Suppose N and N 0 are normed linear space and T : N ! N 0 is

a one-to-one linear transformation such that kT ( x)k= k xk for every x 2 N . Then
T is said to be isometric isomorphism.

A normed linear space N is said to be isometrically isomorphic to N 0 if there

exists an isometric isomorphism of N onto N 0 .

Example 2.1. If M is a closed linear subspace of a normed linear space N and

let T is the natural mapping of N onto N= M dened by T ( x) = x + M . Show
that T is a continuous linear transformation for which kT k 1 .

Solution. Since M is a closed linear subspace of N and by theorem 1.3, N= M

is also a normed linear space with the norm of coset x + M dened as
k x + Mk = inf fk x + mk: m 2 Mg

First, we shall prove that T is linear:

T (cx + y) = cx + y + M
= c( x + M ) + y + M
= cT ( x) + T (y)

Thus, T is linear.
Next, we shall prove that T is continuous ( i:e:; bounded).
kT ( x)k = k x + Mk
= inf f x + m : m 2 Mg
 k x + mk 8m 2 M

Now M being a subspace, therefore 0 2 M and hence choose m = 0 , we get

kT ( x)k  k x + 0k= k xk= 1
k xk 8 x 2 N

Hence by denition, T is continuous (or bounded).

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
2.1. Continuous Linear Transformations 37

kT k = supfkT ( x)k: k xk 1g

 supfk xk: k xk 1g = 1
) kT k  1: È
Let us Sum Up:
In this unit, the students acquired knowledge to

ˆ the concepts of characterization of normed linear space into a

normed linear space.

ˆ the concept of isometrically isomorphism.

Check Your Progress:

1. If T is a continuous linear transformation of a normed linear space

N into a normed linear space N0 , and if M is its null space, show

that T induces a natural trnasformation T0 of N= M into N0 and that

kT 0 = kT k .
2. Let N and N0 be a normed linear spaces over the same eld of

scalars and T be a linear transformation of N into N0 . Prove that

T is bounded if and only if T is continuous.

Suggested Readings:

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

Tata Mc-Graw Hill Pvt. Ltd., New Delhi, 2011.

2. Kreyszig E.,” Introductory Functional Analysis with Applications,

John Wiley & Sons, New York, 2007.

3. Limaye B. V.,” Functional Analysis”, New Age International Ltd.,

Publishers, Second Edition, New Delhi,1996.

4. Co man C. and Pedrick G., “First Course in Functional Analysis”,

Prentice-Hall of India, New Delhi, 1995..

5. Conway J.B., “A Course in Functional Analysis”, Springer-Verlag,

New York, 2008.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
38 2.1. Continuous Linear Transformations

6. Bollobas B.,”Linear Analysis”, Cambridge University Press, Indian

Edition, New York, 1999.

7. Nair M.T., “Functional Analysis, A First course”, Prentice Hall of

India, New Delhi, 2010.

Web Resources:

// / / / /
1. https: archive.nptel.ac.in courses 111 106 111106147 /
// / =
2. https: www.youtube.com watch?v imYQJOgUx7Y

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
2.1. Continuous Linear Transformations 39

Block-II
Unit-3: Fundamental Theorems on Normed Linear Spaces-I

Unit-4: Fundamental Theorems on Normed Linear Spaces-II

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
40
Block-II

UNIT-3

FUNDAMENTAL THEOREMS

ON NORMED LINEAR

SPACES-I

Structure
Objective
Overview
3. 1 The Hahn-Banach Theorem
3. 2 The Natural Imbedding of N in N 
Let us Sum Up
Check Your Progress
Suggested Readings

Overview
In this unit, we dene linear functionals on a normed linear space,

then we prove the main theorem namely Hahn Banach theorem which

asserts that any functional dened on a linear subspace of a normed linear

M.Sc.(Mathematics)-II Year-III Sem 41 Functional Analysis

 42 3.1. The Hahn Banach Theorem

space can be extended linearly and continuously to the whole space without

increasing its norm.

Objectives
After successful completion of this lesson, students will be able to

ˆ discuss continuous linear functionals.

ˆ understand the concept of the Hahn Banach theorem.

3.1. The Hahn Banach Theorem

Denition 3.1. Suppose N is a normed linear space over C or R . Then the set
of all continuous linear transformations of N into C or R according as N is real
B B
or complex is denoted as N  instead of (N; R) or (N; C ) . This space N 
is called the conjugate space of N . The elements of N  are called continuous
linear functionals or simply functionals.

Remark 3.1. If these functionals are added and multiplied by scalars pointwise,
and if the norm of a functional f 2 N  is dened by
k f k = supfj f ( x)j: k xk 1g
= inf fK : K  0 and j f ( x)j K k xk for all xg;

then N  is a Banach space.

Lemma 3.1. Let M be a linear subspace of a normed linear space N , and let
f be a functional dened on M . If x0 is a vector not in M , and if
M0 = M + [ x0 ]

is the linear subspace spanned by M and X0 , then f can be extended to a

functional f0 dened on M0 such that k f0 k= k f k .

Proof. We shall prove the lemma for real and complex scalars separately.
Case 1: Let N be a real normed linear space.
Since x0 is not in M , each vector w in M0 is uniquely expressed in the form
w = x + x0 with x 2 M . We dene our f0 by setting

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
3.1. The Hahn Banach Theorem 43

f0 (w) = f0 ( x + x0 ) = f ( x) + r0

where r0 is any real number. It is easy to see that for every choice of real number
r0 , f0 is linear on M0 such that
f0 ( x) = f ( x) for all x 2 M

For if, ; 2 R and x; y 2 M , then

f0 ( ( x + r0 ) + (y + x0 )) = f0 ( x + y + ( + ) x) 0

= f ( x + y) + ( + ) x 0

= f ( x) + f (y) + r + r 0 0

= ( f ( x) + r ) + ( f (y) + r )
0 0

= f ( x + r ) + f (y + r )
0 0 0 0

Thus, f0 extends f linearly to M0 . Now, we shall prove that k f0 k= k f k .

k f k = supfj f ( x)j: x 2 M ; k xk 1g
0 0 0

 supfj f ( x)j: x 2 M; k xk 1g

0

= supfj f ( x)j: x 2 M; k xk 1g

= kfk
Thus, k f k  k f k
0

Next our problem is now to choose r0 in such a way that k f0 k k f k .

For this purpose, we rst observe that if x1 ; x2 are any two vectors in M , then

f ( x2 ) f ( x1 ) = f ( x2 x1 )
j f ( x x) j  2

k f kk x x k  2 1

k f kk( x + x ) ( x + x ) k
 2 0 1 0

k f k(k x + x k+k ( x + x )k)

 2 0 1 0

= k f kk x + x k+k f kk x + x k 2 0 1 0

Thus; f ( x ) k f kk x + x k 
1 1 f ( x ) + k f kk x + x k
0 2 2 0

Since this inequality holds good for arbitrary x1 ; x2 2 M , we see that

supf f (y) k f kky + x0 kg  inf f f (y) + k f kky + x0 kg
y2 M y2 M

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
44 3.1. The Hahn Banach Theorem

Choose r0 to be any real number such that

supf f (y) k f kky + x0 kg  r0  inf f f (y) + k f kky + x0 kg
y2 M y2 M

It follows that
f (y) k f kky + x k  0 r0  f (y) + k f kky + x0 k 8y 2 M (3.1)

with this choice of r0 we shall prove that k f0 k k f k .

Let w = x + x0 be an arbitrary vector in M0 .
x
Putting y = in (3.1), we get

x x 
x x
f kfk +x 0  r0  f + kfk +x 0 (3.2)

If > 0 , then right hand inequality in (3.2) gives,

1 1
 f ( x) + k f kk x +
r0 x0 k
) f ( x) + r  k f kk x + x k 0 0

) f ( x + r )  k f kk x + x k
0 0 0

) f (w)  k f kkwk
0

If < 0 , then left hand

 
inequality in (3.2) gives,
x x
r0  f kfk +x 0

1 1
= f ( x) kfk kx + x0 k
" #
1 1 1 1
= f ( x) + k f kk x + x0 k * <0) =

Now, we multiply both sides of the above inequality by .

Since < 0 , the inequality will be reversed.
Hence we obtain
 f ( x) + k f kk x + x k
r0 0

or f ( x) + r  +j f kk x + x k 0 0

or f (w)  k f kkwk0

Thus we have shown that 6= 0 , then

f (w)  k f kkwk
0 w 2 M0 (3.3)

[of course when = 0; k f k= k f k ]

0

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
3.1. The Hahn Banach Theorem 45

Replacing w by w , we get
f0 ( w)  k f kk wk
i:e:; f0 (w)  k f kkwk (3.4)

From (3.3) and (3.4), we conclude that

j f (w)j  k f kkwk
0 (3.5)

Thus, (3.5) shows that f0 is bounded and so f0 is a linear functional on M0 .

Since k f k= supfj f0 (w)j: w 2 M0 ; kwk 1g .
It follows from (3.5) that k f0 k k f k .
Finally, we have k f0 k= k f k .
This proves the lemma for real scalars.
Case 2: Let N be a complex normed linear spaces.
Let N be a normed linear space over C and let f be a complex valued linear
functional dened on a subspace M of N .
Let g = Re ( f ) and h = im ( f ) so that f ( x) = g( x) + ih( x) for every x 2 M .
Then an easy computations shows that both f and g are real valued functionals
on the real space M .
For if,
f ( x + y) = f ( x) + f (y)
) g( x + y) + ih( x + y) = g( x) + ih( x) + g(y) + ih(y)
) g( x + y) = g( x) + g(y) and h( x + y) = h( x) + h(y)

and if 2 R , then
f ( x) = f ( x)
) g( x) + ih( x) = g( x) + ih( x)
 

) g( x) = g( x) and h( x) = h( x)

Thus g and h are linear on M . Further

jg( x)j  j f ( x)j k f kk xk

and so boundedness of f implies boundedness of g and similarly of h . Hence,

g; h are real linear functionals on the real space M .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
46 3.1. The Hahn Banach Theorem

Also for all x 2 M , we have

g(ix) + ih(ix) = f (ix) = i f ( x) = h( x) + ig( x)

whence equating real and imaginary parts, we get

g(ix) = h( x) and h(ix) = g( x)

Consequently, f ( x) = g( x) ig(ix) = h(ix) + ih( x) .

Let f ( x) = g( x) ig(ix) .
Since g is a real-valued functional on the real space M , hence by Case 1 , g
can be extended to a real valued functional g0 on the real space M0 in such a
way that kg0 k= kgk .
Dene f0 for x 2 M0 by
f0 ( x) = g0 ( x) g0 (ix)

Now, we shall prove that f0 is a linear on the complex space M0 such that
f0 = g on M

f0 ( x + y) = g0 ( x + y) ig0 (ix + iy) = g0 ( x) + g0 (y) ig0 (ix) ig0 (iy)

= g0 ( x) ig0 (ix) + g0 (y) ig0 (iy)
= f0 ( x) + f0 (y)

Also, if a; b 2 R , then
f0 (a + ib) = g0 (ax + ibx) ig0 ( bx + iax)
= ag0 ( x) + bg0 (ix) i( b)g0 ( x) iago (ix)
= (a + ib) g0 ( x) ig0 (ix)
 

= (a + ib) f0 ( x)

Thus, f0 is linear on M0 . Also g0 = g on M implies f0 = f on M .

It remains to prove that k f0 k= k f k .
Let x 2 M0 be arbitrary and write
f0 ( x) = rei

where r  0 and  real. Then

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
3.1. The Hahn Banach Theorem 47
 
j f ( x)j =
0 e i
rei =e i f ( x ) = f e i x
0 0

= g (e i x)
0
 
 g e i x
0

 kg k e i x
0

= kg k e i k xk= kg kk xk
0 0

= kgkk xk k f kk xk

This shows that f0 is bounded (hence a functional on M0 ) and that k f0 k k f k .

Also, as in Case 1 , it is obvious that k f0 k k f k .
Therefore k f0 k= k f k .
This completes the proof of the lemma. 
Lemma 3.2 (Zorn's lemma). Every non-empty partial ordered set in which each
chain has an upper bound has a maximal element.

Theorem 3.1 (The Hahn-Banach Theorem). Let M be a linear subspace of a

normed linear space N , and let f be a functional dened on M . Then f can be
extended to a functional f0 dened on the wholse space N such that k f0 k= k f k .

Proof. Given M is the linear subspace of N . Thus, by above lemma for any
x 2 N and x 2= M we can have an extension of f on M + f xg such that k f k is
preserved for extension.
Consider the set of all possible extensions of f on all the subspace of ( M +any
other elements of N not in M ) of N which contains M .
This set of extensions of f say G can be partially ordered as follows:
Let g1 ; g2 2 G we dene the relation “ 00 by saying that g1  g2 means
domain of g1 contained in the domain of g2 and g1 = g2 ( x) 8 x 2 domain of g1 .
( G;  ) is partially ordered:
( i ) Reexivity: g1  g2 8 g1 2 G .
( ii ) Antisymmetry: Let g1  g2 and g2  g1 .
) domain of g is contained in domain of g and domain of g is contained
1 2 2

in domain of g1 .
) domain of g1 = domain of g 2 and

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
48 3.1. The Hahn Banach Theorem

g1 ( x) = g2 ( x) 8 x in domain of g 1

g2 ( x) = g1 ( x) 8 x in domain of g 2

) g1 = g2

( ii ) Transitivity: Let g1  g2 and g2  g3 .

) domain of g1 is contained in domain of g2 with
g1 ( x) = g2 ( x) 8 x in the domain of g 1

and domainn of g2 is contained in the domain of g3 with

g2 ( x) = g3 ( x) 8 x in the domain of g2

and hence for all x in the domain of g1 as domain of g1 is contained in the

domain of g2 .
Thus, domain of g1 is contained in domain of g2 and g1 ( x) = g ( x) 8 x
2 in
the domainn of g1 .
Hence g1  g2 .
Therefore G , is a partially ordered.
The set G is not only partially ordered but every chain in G has an upper
bound as follows:
Let G0 be a chain in G . Then union of all domains of g0 s 2 G0 is the domain
of the upper bound of elements of G0 . Thus every chain in G has an upper
bound.
By Zorn's lemma, there exists a maximal element in G say f0 . Now our wish
is to prove that f0 is the required extension to whole of N .
If it is not so then there exists an x 2 N and not in M such that f0 can be
extended to domain f0 + f xg . i:e:; M + f xg by our lemma. But this violates the
maximality of f0 .
Hence f0 is the required extension of f0 to whole of N such that k f k= k f0 k .
This completes the proof of the theorem. 
Theorem 3.2. If N is a normed linear space and x0 is a non-zero vector in N ,
then there exists a functional f0 in N  such that f0 ( x0 ) = k x0 k and k f0 k= 1:

Proof. Let M be the space spanned by x0 so that every vector y in M can be

uniquely written as y = x0 .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
3.1. The Hahn Banach Theorem 49

) kyk = j jk x k 0

Dene f on M by f (y) = f ( x0 ) = kx k
0

Clearly f : M ! F so that f is a functional.

Now, we shall prove that f is linear and continuous.
f is linear: Let y1 ; y2 2 M so that y1 = x0 ; y2 = x 0

cy1 + y2 = c( x ) + x
0 0

= (c + ) x 0

) f (cy1 + y2 ) = f (c + ) x 0

= (c + )k x k 0

= c k x k+ k x k
0 0

= c f (y ) + f (y )
1 2

Thus, f is linear.
f is continuous:
j f (y)j = j f ( x )j= j jk x k
0 0

= j j
k x k= kyk
0

) j f (y)j = kyk
or j f (y)j  2kyk

Hence f is bounded (continuous).

k f k = supfj f (y)j: kyk= 1g
= supfkyk: kyk= 1g
kfk = 1

also, f ( x0 ) = k x0 k by denition choosing = 1.

Hence by Hahn-Banach theorem, the functional f can be extended to a
functional f0 preserving norm. i:e:;
f0 ( x0 ) = f ( x0 ) = k x0 k and k f0 k= k f k= 1 . 
Note 3.1. The above theorem shows that N  separates the vectors in N .
For, if x and y are any two distinct vectors, so that x y 6= 0 , then there exists
a functional f0 in N  such that f ( x y) 6= 0 , or equivalently, f ( x) 6= f (y) .
Theorem 3.3. If M is a closed linear subspace of a normed linear space N

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
50 3.1. The Hahn Banach Theorem

and x0 is a vector not in M , then there exists a functional f0 in N  such that

f0 ( M ) = 0 and f0 ( x0 ) 6= 0 .

Proof. As we already proved in Example 2.1, that T : N ! M dened as

T ( x) = x + M is a continuous linear transformation for which kT k 1 .

Now, if m 2 M then
T (m) = m+M =M=0 (zero of N= M )

Thus, T ( x0 ) i:e:; x0 + M 6= 0 is a non-zero vector in N= M .

Hence by Theorem 3.2, there exists a functional f0 in (N= M ) such that
f ( x0 + M ) = k x + Mk6= 0
0

Now dene f0 on N as
f0 ( x) = f (T ( x)) = f ( x + M )

f0 is linear:
f0 (cx + y) = f (T (cx + y)) = f ((cx + y) + M )
= f [c( x + M ) + (y + M )]
= c f ( x + M ) + f (y + M )
= c f0 ( x) + f0 (y)

Thus, f0 is linear.
f0 is bounded:
j f ( x)j = j f (T )( x)j
0 0

 k f kkT kk xk
 k f kk xk [* kT k 1]

Now as f is bounded being a functional, it follows that f0 is bounded.

Hence f0 is a functional on N .
Also f0 (m) = f (T (m)) = f (0) = 0 8m 2 M .
Hence f0 ( M ) = 0:
f0 ( x0 ) = f (T ( x0 )) = f ( x0 + M ) 6= 0 . 
Example 3.1. Let M be a closed linear subspace of a normed linear space N
and x0 be a vector not in M . If d is the distance from x0 to M , show that

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
3.1. The Hahn Banach Theorem 51

there exist a functional f0 in N  such that f0 ( x) = 0 8 x 2 M , f0 ( x0 ) = 1 and

k f0 k= 1=d .
Solution. Consider the space M0 = f x + x : x 2 M g .
Clearly M0 is a subspace of N and every element y 2 M0 is uniquely
expressible as y = x + x0 . If not then choose y = x1 + 1 x0 .
x + x0 =x + 1 1 x0
x x1
) x0 =
1

) x0 2 M which is a contradiction as x0 2= M .

Hence the above representation is unique.

Dene a functional f0 on M0 such that f (y) = f0 ( x + x0 ) = (say).
f (y) =

Now, we can easily seen that f dened as above is a linear functional and because
of the uniqueness of y this mapping is well dened.
Also f ( x) = f ( x + 0 x0 ) = 0 8 x 2 M .
f ( x0 ) f (1 + 1
x0 ) = 1

Now d = distance of x0 from M .

) d = inf fk x 0 xk: x 2 M g .

Now, we shall establish that k f k= 1=d .

k f k = supfj(f (y)j: kyk 1; y 2 M g ) 0

= sup j fk(yyk)j : y =6 0; y 2 M 0

( )
j j
= sup k x + x k : x 2 M; 2 R; x 6= 0 or =6 0
0
( )
j j
= sup j jk x= + x k : x 2 M; 2 R; or 6= 0
0
( )
1 x
= sup
kx zk
:z= 2 M;
0

= inf fk x0 zk: z 2 M g 1

= d 1
= 1=d

Hence f is a linear functional on M0 such that

f ( x) = 0 8 x 2 M i:e:; f ( M ) = 0:
f ( x0 ) = 1 and k f k= 1=d:

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
 52 3.2. The Natural Imbedding of N in N 
Therefore by Hahn-Banach theorem this functional f dened on M0 can be
extended to a functional f0 on N and hence belonging to N  such that
f0 (y) = f (y) 8y 2 M and k f k= k f k= 1=d
0 0

or f0 ( x) = f ( x) = 0 8 x 2 M i:e:; f ( M ) = 0 0

and f0 ( x) = f ( x) = 1 and k f k= 1=d: 0

Thus, we proved the required result. È

3.2. The Natural Imbedding of N in N 

We have seen earlier that the conjugate space N of a normed linear

space N is itself a normed linear space. So we can form the conjugate

  
space ( N ) of N . We denote this space by N
 is call it as the second

conjugate space of N . The importance of the space N  lies in the fact

that each vector x in N gives rise to a functional Fx in N  and that

there exists an isometric isomorphism of N into N  , called the natural

imbedding of N into N , as the following theorem shows.
Theorem 3.4. Let N be an arbitrary normed linear space. Then each vector x
in N induces a functional Fx on N  dened by
Fx( f ) = f ( x) 8 f 2 N

such that kF x k= k xk .

Further the mapping J : N ! N  by

J ( x) = Fx 8x 2 N

denes an isometric isomorphism of N into N .

Proof. First we show that F x is actually a functional on N  , so we must show

that F x is linear and bounded.
F x is linear:

Let f ; g 2 N  and ; be any scalars. Then

Fx( f + g) = ( f + g) x
= f ( x) + g( x)
= F x ( f ) + F x (g)

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
3.2. The Natural Imbedding of N in N  53

This shows that F x is linear.

F x is bounded:

For any f 2 N  , we have

jF x ( f )j = j f ( x)j= k f kk xk

Thus, the constant k xk is bounded for F x . Hence F x is a functional on N  .

Now, we shall prove that kF x k= k xk . We have
kF x k = supfjF x ( f )j: k f k 1g
= supfk f kk xk: k f k 1g
 k xk

To prove the reverse inequality, we rst consider this case when x = 0 .

In this case, we have kF0 k= k0k= 0:
But kF x k= 0 always.
Hence kF0 k= k0k i:e:; kF x k= k xk when x = 0 .
Now let x be any non-zero vector. Then by theorem there exists a functional
F in N  such that
F ( x) = k xk and kF k= 1:

But
kF x k = supfjF x ( f )j: k f k= 1g
= supfj f ( x)j: k f k= 1g

Also, since k xk= jF x ( x)j supfj f ( x)j: k f k= 1g = kF x k .

Thus, we have kF x k k xk .
Hence kF xk= k xk .
Finally, we shall show that J is an isometric isomorphism. i:e:; we must show
that J is a linear transformation as well as an isometry.
J is linear:

For any x; y 2 N and any scalar , we have

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
54 3.2. The Natural Imbedding of N in N 
F x +y ( f ) = f ( x + y) = f ( x) + f (y)
= F x ( f ) + Fy ( f )
 
= F x + Fy ( f )
and F x ( f ) = f ( x) = F x ( f ) = ( F x )( f ) 8 f 2 N

It follows that F x+y = F x + Fy and F f = Fx

and so J ( x + y) = F x+y = F x + Fy = J ( x) + J (y) ; J ( x) = F x = F x = J ( x) .

This shows that J is linear.
It remains to prove that J is an isometry.
The equatios k J ( x)k= kF x k= k xk shows that J is norm preserving and hence
an isometry.
In fact, if x; y 2 N , then
k J ( x) J (y)k = kF x Fy k
= kF x y k= k x yk

Thus, J preserves distances also and so it is an isometry.

Also, J ( x) J (y) = 0 ) J(x y) = 0 ) x y=0 ) x = y.

Thus, J ( x) = J (y) ) x = y and hence J is one-to-one.

Hence J denes an isometric isomorphism of N into N  .
This completes the proof of the theorem. 
Note 3.2. The mapping J as dened in the above theorem is called the natural
embedding (or) canonical mapping of N into N  , since it enables us to regard
N as a part of N  without changing any of its structures as a normed linear
space. We write N  N  where the set inclusion is to be understood in the sense
just explained.
Denition 3.2. A normed linear space N is said to be reexive if N = N  .
Note 3.3. The space l p (and L p ) for 1 < p < 1 are reexive, for lp
 
= lq and
p = l p = lq = l p .
l  

We can also prove that the space lnp for 1  p  1 are also reexive.
Since N  is complete and hence N is complete, if it is reexive.
If N is complete, it is not necessarily reexive.
For if, C0 = l1 and C0 = l1 = l1 .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
3.2. The Natural Imbedding of N in N  55

If X is compact Hausdor space it can be seen that C (X ) is reexive if and

only if X is a nite set.

Denition 3.3. The weak topology on a normed linear space N is the weakest
topology on N with respect to which all the functions in N  remains continuous.

Note 3.4. If B is a Banach space, and if S = f x : k xk 1g is its closed unit

sphere, then B is reexive if and only if S is compact in the weak topology. The
proof is not reqired for us.

Denition 3.4. The weak  topology on N  is dened to be the weak topology

on N  generated by all the induced functionals F x in N  .

Denition 3.5. The conjugate space N  of a normed linear space is a metric

space. The topology derived from this metric on N  is called the strong topology.

Note 3.5. N  is the set of all scalar valued linear functions dened as N  which
are continuous with respect to its strong topology. The weak topology on N  is
the weakest topology on N  with respect to which all the functions in N  are
continuous, and clearly this is weakter than it's strong topology.
Since N  is the conjugate space of N , the natural imbedding enables us to
consider N as part of N  . We now form the weakest topology on N  with
respect to which all the functions in N -regarded as a subset of N  remains
continuous. This is the weak topology and it is evidently weaker than the weak
topology. The weak topology can be given a more explicit description, in which
its dening subbasic open sets are displayed. Consider a vector x in N and
its induced function F x in N  . The weak topology on N  is the weakest
topology under which all such F 0x s are continuous. If f0 is an arbitrary element
in N  and if  > 0 is given then the set
S ( x; f0 ;  )= f f : f 2 N and jF x ( f ) F x ( f0 )j <  g
S ( x; f ;  ) = f f : f 2 N 
0 and j f ( x) f0 ( x)j <  g

is an open set (in fact neighborhood of f0 ) in the weak topology. Furthermore,

the class of all sets of this kind, for all x0 s; f00 s; and  0 s is the dening open
subbase for the weak topology. All nite intersection of these sets constitute an
open base for this topology, and the open sets themselves are all unions of these
nite intersections.

Theorem 3.5. If N is a normed linear space, then the closed unit sphere S  in
N  is a compact Hausdor space in the weak topology.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
56 3.2. The Natural Imbedding of N in N 
Proof. First we shall prove that N  is a Hausdor space with respect to its weak
topology. If f and g are distinct functionals in N  , then there must exists a
vector x in N such that f ( x) 6= g( x) ; for if we put  = j f ( x) g( x)j=3 then
S ( x; f ;  ) and S ( x; g;  ) are distinct neightborhoods of f and g in the weak
topology.
Dene S  = f f : f 2 N  and k f k 1g . Then S  is the closed unit sphere in
N  . We can easily prove that S  is compact in the strongly topology if and only if
N is nite dimensional. If N is complete, then we can prove that S  is compact
in the weak topology if and only if N is reexive. With each vector x in N , we
associate a compact space C x , where C X is the closed interval [ k xk; k xk] or
the closed disc fz : jzj k xkg , according as N is real or complex. By Tychno 's
theorem, the product C of all the C 0x s is also a compact space. For each x , the
values f ( x) for all f 0 s in S  lie in C x . This enables us to imbed S  in C by
regarding each f in S  as identical with the array of all it's values at the vectors
x in N . It is clear from that the denitions of the topologies concerned with the
weak  topology on S  equals its topology as a subspace of C ; and since C is
compact, it sucies to show that S  is closed as a subspace of C . We show that
if g is in S  , then g is in S  . If we consider g to be a function dened on the
index set N , then since g is in C we have jg( x)j k xk for every x in N .
It is therefore sucies to show that g is linear as a function dened on N .
Let  > 0 be given, and let x and y be any two vectors in N . Every basic
neighborhood of g intersects of S  , so there exists an f in S  such that
jg( x) f ( x)j < =3; jg(y) f (y)j< =3; and jg( x + y) f ( x + y)j< =3

Since f is linear, f ( x + y) f ( x) f (y) = 0 , and we therefore have

jg( x + y) g( x) g(y)j = j[g( x + y) f ( x + y)] [g( x) f ( x)] [g(y) f (y)]j
 jg( x + y) f ( x + y)j + jg( x) f ( x)j + jg(y) f (y)j
< =3 + =3 + =3 = 

The fact that this inequality is true for every  > 0 now implies that
g( x + y) = g( x) + g(y) . We can show in same fashion that g( x) = g( x) for
every scalar , so g is linear and the theorem is proved. 

The following theorem is an obvious consequence of the previous

results.

Theorem 3.6. Let N be a normed linear space and let S  be the compact

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
3.2. The Natural Imbedding of N in N  57

Hausdor space obtained by imposing the weak  topology on the closed unit
sphere in N  . Then the mapping x ! F , where F x ( f ) = f ( x) for each f in
S  , is an isometric isomorphism of N into C(S  ) . If N is a Banach spacce,
this mapping is an isometric isomorhism of N onto a closed linear subspace of
C (S  ) .

Let us Sum Up:

In this unit, the students acquired knowledge to

ˆ prove the Hahn-Banach theorem.

ˆ imbedding of N in N .

Check Your Progress:

1. State and Prove Hahn-Banach theorem.

2. Let M be a closed linear subspace of a normed linear space N and

x0 be a vector not in M. If d is the distance from x0 to M, show

that there exist a functional f0 

in N such that f0 ( x) = 0 8x 2 M ,
f0 ( x0 ) = 1 and k f k= 1=d .
0

3. Dene reexive normed linear space.

Suggested Readings:

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

Tata Mc-Graw Hill Pvt. Ltd., New Delhi, 2011.

2. Kreyszig E.,” Introductory Functional Analysis with Applications,

John Wiley & Sons, New York, 2007.

3. Limaye B. V.,” Functional Analysis”, New Age International Ltd.,

Publishers, Second Edition, New Delhi,1996.

4. Co man C. and Pedrick G., “First Course in Functional Analysis”,

Prentice-Hall of India, New Delhi, 1995..

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
58 3.2. The Natural Imbedding of N in N 
5. Conway J.B., “A Course in Functional Analysis”, Springer-Verlag,

New York, 2008.

6. Bollobas B.,”Linear Analysis”, Cambridge University Press, Indian

Edition, New York, 1999.

7. Nair M.T., “Functional Analysis, A First course”, Prentice Hall of

India, New Delhi, 2010.

Web Resources:

// / / / /
1. https: archive.nptel.ac.in courses 111 106 111106147 /
// / / /
2. https: www.math.uzh.ch gorodnik functional
analysis lecture6.pdf

// / /
3. https: www.imsc.res.in kesh hahn.pdf

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
Block-II

UNIT-4

FUNDAMENTAL THEOREMS

ON NORMED LINEAR

SPACES-II

Structure
Objective
Overview
4. 1 The Open Mapping Theorem
4. 2 The conjugate of an operator
Let us Sum Up
Check Your Progress
Suggested Readings

Overview
In this unit, we derive open-mapping theorem, closed graph

theorem and uniform boundedness theorem.

M.Sc.(Mathematics)-II Year-III Sem 59 Functional Analysis

 60 4.1. The Open Mapping Theorem

Objectives
After successful completion of this lesson, students will be able to

ˆ prove open mapping theorem.

ˆ understand the concept of conjugate of an operator.

4.1. The Open Mapping Theorem

In this section, we have our rst encounter with basic theorems which

require that the spaces concerned be complete. The following rather

technical lemma is the key to these theorems.

Lemma 4.1. If B and B0 are Banach spaces, and if T is a continuous linear

transformation of B onto B0 , then the image of each open sphere centred on the
origin in B contains an open sphere centred on the origin in B0 .

Proof. Let S ( x; r) and S 0 ( x; r) denote the open spheres with centre x and radius
r in Banach space B and B0 respectively. Also S r and S r0 will denote these
spheres when the centre is the origin. First we shall prove that
S ( x; r ) = x + Sr and S r = rS 1

For this, we have

y 2 S ( x; r ) , ky xk< r
, kzk< r where z = y x
, y = x + z and kzk< r
, y 2 x + Sr ( )
k xk
and S r = f x : k xk< rg = x : < 1 r
= fry : kyk< 1g = rS 1

and so
T (S r ) = T (rS 1 ) = rT (S 1 )

Hence it is enough to prove that T (S 1 ) contains some S r0 .

We begin by proving that T (S 1 ) contains some S r0 .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
4.1. The Open Mapping Theorem 61

For each positive integer n consider open spheres S n in B . Then it is clear

that
1
[
B = Sn
n= 1

Since T is onto B0 , this gives

B0 = T ( B)
0 1
B[1 C
= T
B
B
B
B
@
C
S n CCCA
n= 1
1
[
= T (S n )
n= 1

Since B0 is complete, it is of second category. Hence by Baire Category theorem,

T (S n0 ) 0 6= ; for some n0 , that is T (S n0 ) has an interior point y0 which may

be assumed to lie in T (S n0 ) .
The existene of such a point y0 is proved as follows:
y is an interior point of T (S n0 ) .

) there exists an open set G such that y 2 G  T (S n ) . 0

0
But y 2 T (S n0 ) which implies that y is an adherent point of T (S n0 ) .
i:e:; the neighborhood G of y must contain a point y0 of T (S n0 ) .


Thus, y 2 T (S n0 ) is such that y0 2 G  T (S n0 ) which implies that y0 is an


interior point of T (S n0 ) .
The mapping f : B0 ! B0 by f (y) = y y0 is a homeomorphism. For f is
evidently one-one onto and if yn 2 B0 is such that yn ! y , then
f (yn ) = yn y0 ! y y = f (y)
0

and f (yn ) 1
= yn + y0 ! y + y = f (y)
0
1

so that f and f 1 are both continuous. We use the mapping f to show that 0
is an interior point of T (S n0 ) y0 . We have y0 is an interior point of T (S n0 ) .
) there exists an open set G such that y0 2 G  T (S n ) . 0
 
) f (y0 ) 2 f (G)  f T (S n0 ) .

) y0 y0 = 0 2 f (G)  T (S n ) 0
y0 .

) 0 is an interior point of T (S n ) 0
y0 .


We assert that T (S n0 ) y0  T S 2n0 .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
62 4.1. The Open Mapping Theorem

Let y 2 T (S n0 ) y0 . Then there exists x 2 S n0 such that y = T ( x) y0 .

But y0 2 T (S n0 ) implies that y0 = T ( x0 ) for some x0 2 S n0 .
Thus y = T ( x) T ( x0 ) = T ( x x0 ) , where x; x0 2 Sn . 0

Also
x; xn 2 Sn 0
) k xk< n ; k x k< n 0 0 0

) k x x k k xk+k x k< 2n
0 0 0

) x x 2S n 0 2 0

) y 2 T (S n ) 2 0

Thus, we have shown that

y 2 T (S n0 ) y0 ) y 2 T (S n ) 2 0

and therefore T (S n0 ) y0  T (S n ) = 2n T (S ) 2 0 0 1

) T (S n0 ) y0  2n T (S ) 0 1

Since f is a homeomorphism,
 
f T (S n0 )  f (T (S n0 )
) T (S n0 ) y0 = T (S n0 ) y0  2n T (S ): 0 1 (4.1)

The mapping g : B0 ! B0 by g( x) = 2n0 x

is easily seen to be a homeomorphism and so
 
g(T (S 1 )) = g T (S 1 )
) 2n T (S )0 1 = 2n0 T (S 1 )

By denition of g and (4.1) implies that

T (S n0 ) y0  2n T (S )
0 1 (4.2)

Thus 0 is an interior point of T (S 1 ) .

Hence there exists  > 0 such that S 0  T (S 1 ) .
We conclude the proof by showing that S 0  T (S 3 ) which is clearly equivalent
0  T (S ) .
to S = 3 1

Let y1 be an arbitarary point of S 0 so that kyk<  . Then y 2 T (S 1 ) which

implies that y is an adherent point of T (S 1 ) and hence there exists a vector
y1 2 T (S 1 ) such that ky y1 k< =2 .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
4.1. The Open Mapping Theorem 63

But y1 2 T (S 1 ) ) y1 = T (X1 ) for some x1 2 S 1 so that k x1 k< 1 .

0  T (S ) and since ky y k< =2 , we have
Again, we observe that S = 2 1=2 1

y y1 2 0
S =  T (S = )
2 1 2

Therefore as before there exists a vector y2 in T (S 1=2 ) such that

k(y y1 ) y2 k < =2

1
where y2 = T ( x2 ) and k x2 k< . Continuing in this way, we obtain a sequence
2
f xn g in B such that k xn k< 1=2n 1 , and
ky (y + y + : : : + yn )k < =2n
1 2

where yn = T ( xn ) . If we put sn = x1 + x2 + : : : + xn , then

k sn k = k x + x + : : : + xn k
1 2

 k x k+k x k+ : : : + k xn k
1 2

1 1
< 1 + + ::: + n 1 < 2
2 2

Also for n > m , we have

k sn sm k = k xm+ + xm+ + : : : + xn k
1 2

 k xm+ k+k xm+ k+ : : : + k xn k

1 2

1 1 1
< m + m+ + : : : + n
2 2 2 1
!
1

1 1
m 1 n m
= 2 1 1=22
= 2m1 1 2n 1m 1
! 0 as m; n ! 1

Hence f sn g is a Cauchy sequence in B and since B is complete there exists a

vector x in B such that
lim s
n!1 n
= x

and so k xk= klim sn k= limk sn k 2 < 3 which implies that x 2 S 3 . Now

y1 + y2 + : : : + yn = T ( x1 ) + T ( x2 ) + : : : + T ( xn )
= T ( x1 + x2 + : : : + xn )

Since T is continuous

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
64 4.1. The Open Mapping Theorem

x = lim sn
) T ( x) = T (lim sn )
= lim T ( sn )
= lim T ( x1 + x2 + : : : + xn )
= lim(y1 + y2 + : : : + yn )
= y

Thus y = T ( x) where k xk< 3 , so that y 2 T (S 3 ) .

Now we have proved that y 2 S 0 ) y 2 T (S 3 )
and so S 0  T (S 3 ) which is what we wished to prove.
This completes the proof of the lemma. 
Theorem 4.1 (The open mapping theorem). Let B and B0 be Banach spaces.
If T is a continuous linear transformation of B onto B0 , then T is an open
mapping.

Proof. Given that the linear transformation T : B ! B0 is continuous and onto.

Now, we want to show that T is an open map, that is T (G) is an open set in
B0 for every open set G in B .
Let y 2 T (G) be arbitrary, then y = T ( x) for some x 2 G .
Since G is an open set in B , there exists an open sphere S ( x; r) in B centered
at x such that S ( x; r)  G .
But, from the previous lemma, we have S ( x; r) = x + S r , where S r is an open
sphere in B centered at origin. Then
x + Sr G

By lemma there exists an open sphere S 0 in B0 centered at origin such that

S 0  T (S r ) .

) y + S 0  y + T (S r ) = T ( x) + T (S r ) = T ( x + S r ) .
i:e:; S 0 (y;  )  T ( x + S r )  T (G) .

Thus we have show that to each point y 2 T (G) , there exists an open sphere in
B0 centered at y and contained in T (G) and consequently T (G) is an open set.

Thus, T is an open mapping and which completes the proof of the

theorem. 

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
4.1. The Open Mapping Theorem 65

Theorem 4.2. A one-to-one continuous linear transformation of one Banach

space onto another is a homeomorphism. In particular, if a one-to-one linear
transformation T of a Banach space onto itself is continuous, then its inverse
T 1
is automatically continuous.

Proof. Since T is one-to-one and continuous, we need to show that T is an

open mapping. The hypothesis of this theorem satisifes all the conditions of open
mapping theorem. Thus, T is an open mapping.
Further that the hypothesis of open mapping theorem is not sucient to infer
that T is an homeomorphism, since T is not given as one-to-one. But, in this
theorem additional property given as T is one-to-one.
Thus, T is one-to-one, onto and open mapping. Hence T is a
homeomorphism then its inverse T is automatically continuous.
1

Denition 4.1. Suppose that linear space L is the direct sum of the subspaces
M and N , so that L = M  N . Then the map E : L ! L dened as E ( x) = y
where x = y + z; y 2 M and z 2 N is called projection of L on M .
Remark 4.1.
Projection is a linear transformation on L and E 2 = E . i:e:; E (E ( x)) = E ( x) .
In fact,this property idempotent characterize projection.
Projection on L can be described geometrically as follows:

1. A Projection E determines a pair of linear subspaces M and N such that

L = M  N where
M = f E ( x ) : x 2 Lg
and N = f x : E ( x) = 0g

are the range and null space of E .

2. A pair of linear subspaces M and N such that L = M  N determines a

projection E whose range and null space are M and N (if z = x + y is
the unique represention of a vector in L as a sum of vectors in M and N ,
then E is dened by E ( x) = x ).

The above remarks show that the study of projection on L is equivalent

to the study of pairs of linear subspaces which are disjoint and span L. For

projections on Banach spaces, something more is required than linearity

and ideompotence. We make it precise in the following denition.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
66 4.1. The Open Mapping Theorem

Denition 4.2. A projection on a Banach space B is an idempotent operator E

on B .
In otherwords, E is a projection on B if

(i) E 2 = E i:e:; E is a projection on B in the algebraic sense, and

(ii) E is continuous

Theorem 4.3. If P is a projection on a Banach space B , and if M and N are

its range and null space, then M and N are closed linear subspaces of B such
that B = M  N .

Proof. By denition of Projection on a Banach space

(i) P is a projection in the algebraic sense and

(ii) P is continuous.

The condition (i) gives that B = M  N except the fact that M and N are
closed, where M and N are the range and null space of P respectively.
In order to prove that M and N are closed subspaces of B , we make use of
the condition (ii) .
By denition of a null space,
N = f x : P( x) = 0g = P (f0g)
1

Since P is continuous and f0g is closed in B , it follows that P 1 (f0g) is a closed

subspace of B . Again
M = f x : P( x) = xg = f x : (I P)( x) = 0g

i:e:; M is the null space of I P , But since the identity map I is always
continuous and hence I P is also continuous, which is also closed.

Thus, M and N are closed subspaces of B . 

Theorem 4.4. Let B be a Banach space, and let M and N be closed linear
subspaces of B such that B = M  N . If z = x + y is the unique representation
of a vector in B as a sum of vectors in M and N , then the mapping of P dened
by P( x) = x is a projection on B whose range and null space are M and N .

Proof. Since B = M  N .
Thus P dened by P( x) = x .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
4.1. The Open Mapping Theorem 67

) z 2 B has a unique representation as z = x + y with x 2 M and y 2 N .

Also the mapping dened by P( x) = x is an idempotent map whose range and

null spaces are M and N respectively. All these above we have proved in the
above theorem.
In order to prove P is a projection on Banach space, it is enough to prove that
P is continuous.

To prove P is continuous from B into itself, we renorm B to make B0 as

follows:
Dene kzk0 = k xk+kyk

then B0 is a Banach space. Further

kP(z)k = k xk
 k xk+kyk= kzk0
) kP(z)k  kzk0 8z 2 B0

Hence P is bounded and therefore continuous mapping of B0 into B . It sucies

to prove that B0 and B have the same topology.
If T denotes the identity mapping of B0 onto B , then
kT (z)k = kzk= k x + yk
 k xk+kyk= kzk0

This shows that T is continuous and one-to-one linear transformations of B0 into

B . Thus T is a homeomorphism.

i:e:; B and B0 have the same topology.

) P is continuous from B0 into B .

Thus, P is continuous from B into itself and hence P is a projection. 

Remark 4.2. This theorem raises some interesting and signicent questions. Let
M be a closed linear subspace of a Banach space B . We know that, there is
always at least one algebraic projection dened on B whose range is M and
there may be a great many. However, it might well happen that none of these are
continuous, and that consequently none are projections is our present sense. In
the light of our theorems, this is equivalent to saying that there might not exist any
closed linear subspace N such that B = M  N . What sorts of Banach spaces
have the property that is awkaed situation cannot occur? We shall see in the

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
68 4.1. The Open Mapping Theorem

next chapter that a Hilbert space-which is a special type of Banach space has this
property. We shall also see that the property is closedly linkedly to the satisfying
geometric structure which sets Hilbert spaces apart from general Banach spaces.

Denition 4.3. Suppose T is a linear transformation from a Banach space B into

a Banach space B0 . Then the graph of T is dened to be the set of all ordered
pairs of the form (x; T ( x)) which is a subset of B  B0 .

Remark 4.3. Let B and B0 be Banach spaces. If we dene a metric on the

product B  B0 by
d (( x1 ; y1 ); ( x2 ; y2 )) = max fk x1 x2 k; ky1 y2 kg

then the resulting topology is easily seen to be the same in the product topology,
and convergence with respect to this metric is equivalent to coordinatewise
convergence. If T is continuous, then its graph is closed as a subset of B  B0 .
In the present context, the converse is also true.

Theorem 4.5 (The closed graph theorem). If B and B0 are Banach spaces, and
if T is a linear transformation of B into B0 , then T is continuous if and only if
its graph is closed.

Proof. The condition is necessary:

Given that T is continuoius. i:e:; xn ! x0 ) T ( xn ) ! T ( x0 ) .
To prove that the graph of T i:e:; TG = f( x; T ( x)) : x 2 Bg is closed. It is
enough to prove that TG = TG .
For this, let (x0 ; y0 ) 2 TG ) 9 a sequence f( xn ; T ( xn ))g in T such that
f( xn ; T ( xn ))g ! ( x ; y ) 0 0

) xn ! x ; T ( xn ) ! y
0 0

or T ( xn ) ! T ( x ) 0

) T (x ) = y 0 0

Now x0 2 B ) ( x0 ; T ( x0 )) 2 TG .
or ( x0 ; y0 ) 2 TG .
Hence ( x0 ; y0 ) 2 TG ) ( x0 ; y0 ) 2 TG .
) TG  TG
But always, TG  TG .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
4.1. The Open Mapping Theorem 69

Thus TG = TG and hence TG is closed.

The Condition is sucient:
Given TG is closed and we will prove that T is continuous.
Let us dene another norm on the space B to be denoted as k
k1 as follows:
For x 2 B; dene k xk1 = k xk+kT ( x)k

It is easy to verify that the new norm satises all the axioms of norm. Hence the
linear space B equipped with this new norm will be denoted by B1 .
Thus B1 is a normed linear space.
Now kT ( x)k k xk+kT ( x)k= k xk1
) kT ( x)gj k xk or kT ( x)k 1
k xk
1 1

Hence T considered as a mapping from B1 to B0 is bounded and

consequently continuous. It is therefore sucient if we prove that B and B1
have the same topology. i:e:; they are homeomorphic which will imply that T
is continuous from B into B0 .
First we shall show that B1 is a complete normed linear space.
Let f xn g be a Cauchy sequence in B1 .
) k xn xm k1 ! 0 as m; n ! 1 .

) k xn xm k+kT ( xn T ( xm )k! 0 as m; n ! 1 .

) k xn xm k+kT ( xn ) T ( xm )k! 0 as m; n ! 1 .

) k xn xm k! 0 as m; n ! 1 .

and T ( xn ) T ( xm )k! 0 as m; n ! 1 .
Thus f xn g is a Cauchy sequence in B and fT ( xn )g is a Cauchy sequence in
B0 . But both B and B0 are Banach space and hence complete.
) xn ! x 2 B and ( xn ) ! y 2 B0 .
Now f( x; T ( xn ))g is a Cauchy sequence in the graph of T which is given to be
closed.
) ( xn ; T ( xn )) ! ( x; y) 2 TG
But ( x; y) 2 TG ) y = T ( x) .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
 70 4.2. The Conjugate of an Operator

Now k xn xk1 = k xn xk+kT ( xn x)k

= k xn xk+kT ( xn ) T ( x)k
= k xn xk+kT ( xn y)k! 0
) xn ! x in B1

Thus B1 is complete and hence it is a Banach space.

Now we shall establish that there is a homeomorpohism between B and B1 .
Consider I : B1 ! B such that I ( x) = x 8 x 2 B .
i:e:; I as an identity map, which is one-to-one and onto.
kI ( x)k = k xk k xk+kT ( x)k k xk 8 x 2 B
1 1

Thus I is bounded i:e: continuous and also it is one-to-one and onto.

Hence I is a homeomorphism from one Banach space to another.
Now T is continuous from B1 to B0 and hence it is continuous on its
hoemomorphic image B .
Hence T : B ! B1 is continuous.
This completes the proof of the theorem. 

4.2. The Conjugate of an Operator

We shall see in this section that each operator T on a normed linear

space N induces a corresponding operator denoted by T and called the

conjuage of T , on the conjugate space N . Our rst task is to dene T  and

our second is to investigate the properties of the mapping T ! T .


Theorem 4.6 (the Uniform Boundedness theorem). Let B be a Banach space

and N a normed linear space. If fT i g is a non-empty set of continuous linear
transformations of B into N with the property that fT i( x) g is a bounded subset
of N for each vector x in B , then fkT i kg is a bounded set of numbers that is Ti
is bounded as a subset of B(B; N ) .
Proof. For each positive integer n , dene
Fn = fx : x 2 B and kT i ( x)k n 8ig (4.3)

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
4.2. The Conjugate of an Operator 71

Then Fn is a closed subset of B as shown below:

x 2 Fn , kT i ( x)k n 8i
, T i ( x) 2 S n 8i; 0

where S n0 denotes the closed sphere in N with centre 0 and radius n:

 
, x 2 Ti 1
S n0 8i
\  
, x2 Ti 1
S n0 ;
\
i
 
so that Fn = Ti 1
S n0
i

which is closed, being an intersection of closed sets is closed.

1
[
Further, B = Fn .
n=1
1
[
For if, B 6= Fn , then there is some x 2 B such that x 2= Fn for any n .
n= 1
) kT i ( x)k> n 8i (* (4.3))
) the set fT i ( x)g is not bounded;

which contradicts the hypothesis. Hence we must have

1
[
B = Fn
n= 1

so that the complete space B is the union of a sequence of its subsets.

Therefore, by Baire Category theore, there exists an integer n0 such that Fn0
has non-empty interior. Since Fn0 is closed,
F n0 = F n0

and so Fn0 must have non-empty interior, that is there exists some x0 2 Fn00 so
that Fn0 is a neighborhood of x0 .
Since Fn0 is closed, there exists a closed sphere
S = f x 2 B : k x x0 k r0 g  Fn0

Now, if kyk< 1 , then for arbitrary but xed

!
i
z
kT i (y)k = Ti where z = r0 y
r0
1 1
= r0
kT i (z)k= kT i (z + x
r0
0 x0 )k

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
72 4.2. The Conjugate of an Operator

1
= r0
kT i (z + x ) 0 T i ( x0 )k
1
 [kT i (z + x0 )k+kT i ( x0 )k]
r0
1 2n0
 (n0 + n0 ) = ; * z + x and xo 2 Fn
0
r0 r
2n0
Thus, kT i (y)k  if kyk 1:
r

2n0
) kT k= supfkTi (y)k: kyk 1g  .
r
It follows that fkT i kg is a bounded set of numbers.
Thus, the proof of the theorem is complete. 
Note 4.1. The above theorem is usually known as the Banach Steinhaus theorem.

Theorem 4.7. A non-empty subset X of a normed linear space N is bounded if

and only if f (X ) is a bounded set of numbers for each f in N .

Proof. Suppose S is a bounded subset of N so that there exists a positive

constant K1 such that
k xk  K1 8x 2 X (4.4)

To show that f (X ) is bounded for each f 2 N  .

Now f 2 N  ) f is bounded
) there exists K > 0 such that j f ( x)j K k xk 8 x 2 N (4.5)
2 2

From (4.4) and (4.5), we conclude that

j f ( x)j  K1 K2 8 x 2 X:

Hence f (X ) is a bounded set of numbers for each f 2 N  .

Conversely, let f (X ) be a bounded set of numbers for each f in N  . To show
that X is bounded. For convenience, we exhibit the vectors in X by writing
X = f xi g . We now use the natural imbedding xi ! F x to pass from X to the i

corresponding subset fF x g of N  .
i

F x is dened by F x ( f ) = f ( x) .

Hence our assumption that f (X ) = f f ( xi )g is bounded for each f is equivalent

to the assumption that fF x ( f )g is bounded set for each f 2 N  and since N 
i

is complete, it follows from the uniform boundedness theorem that fF x g is a i

bounded subset of N  , that is fkF x kg is a bounded set of numbers. Since the

i

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
4.2. The Conjugate of an Operator 73

natural imbedding preserves norms, we have

kF x k = k xi k
i

for each xi 2 X . It follows that fk xi kg is a bounded set of numbers, that is

X = f xi g is a bounded subset of N , thus completes the proof of the theorem. 

Remark 4.4. Let L be the linear space of all scalar-valued linear functions
dened on N . The conjugate space N  is clearly a linear subsapce of L . Let T
be a linear transformation of N into itself, which is not necessarily continuous.
We use T to dene a linear transformation T 0 of L into itself as follows:
If f is in L , then T 0 ( f ) is dened by
 
T 0 ( f ) ( x) = f (T ( x))

We leave it to the reader to verify that T 0 ( f ) actually is linear as a function

dened on N , and also that T 0 is linear as mapping of L into itself.
The following natural question now present itself. Under what circumstances
does T 0 map N  into N  ?. This quest has a simple and elegant answer:
T 0 (N  )  N  if and only if T is continuous. If we keep Theorem 4.7 in mind,
the proof of this statement is very easy; for if S is the closed unit sphere in N ,
then T is continuous ) T (S ) is bounded ) f (T (S )) is bounded for each f
 
in N  ) T 0 ( f ) (S ) is bounded for each f in N  ) T 0 ( f ) is in N  for each
f in N  .

Denition 4.4. Suppose T is a continuous linear operator on N and T 0 the

restriction of N  into itself. If we denote this restriction by T  , then T  is
called the conjuage of T . Hence
(T  ( f )) ( x) = f (T ( x)) (4.6)

where f is a functional on N .

Theorem 4.8. If T is an operator on a normed linear space N , then its

conjugate T  dened by (4.6) is an operator on N and the mapping T ! T 
B B
is an isometric isomorphism of (N ) into (N  ) which reverses products and
preserves the identity transformation.

Proof. Let f ; g be a linear functionals on N . Then

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
74 4.2. The Conjugate of an Operator

(T  ( f+ g)) ( x) = (f + g)(T ( x))

= f (T ( x)) + g(T ( x))
= (T  ( f ))( x) + (T  (g))( x)
(T  ( f )) ( x) = ( f )(T ( x))
= ( f (T ( x)))
= (T  ( f )) ( x)

Hence T  is linear.
kT  k = sup fkT  ( f )k: k f k 1g
= sup fkT  ( f )k( x) : k f k 1 and k xk 1g
= sup fk f (T ( x))k: k f k 1 and k xk 1 g
 sup fk f kkT kk xk: k f k 1 and k xk 1 g
 kT k

Since kT k= sup fkT ( x)k: k xk 1g , we see at once from Theorem 3.2 that equality
holds here, that is, that
kT  k = kT k

The mapping T ! T  is thus a norm-preserving mapping of B(N ) into B(N ) .

We continue in this vein by observing that the mapping T ! T  also has the
following pleasant algebraic properties:
( T 1 + T 2 ) = T 1 + T 2
(T 1 T 2 ) = T 2 T 1 (4.7)
and I  = I

The proof of these facts are easy consequences of the denitions. We illustrate
the principle involved by proving (4.7). It must be shown that
(T 1 T 2 ) ( f ) = T 2 T 1 ( f ) for each f in N 

and this means that

(T 1 T 2 ) ( f )( x) =

T 2 T 1 ( f )( x) for each f in N 

A simple computation now show that

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
4.2. The Conjugate of an Operator 75

(T 1 T 2 ) ( f )( x) = f ((T 1 T 2 )( x))
= f (T 1 (T 2 ( x)))
=

T 1 ( f ) (T 2 ( x))
=

T 2 (T 1 ( f )) ( x)
=


T 2 T 1 ( f ) ( x) 

Let us Sum Up:

In this unit, the students acquired knowledge to

ˆ prove the open mapping theorem.

ˆ understand the concept of the conjugate of an operator.

Check Your Progress:

1. State and Prove open mapping theorem.

2. State and Prove closed graph theorem.

3. State and prove Banach Steinhaus theorem.

4. Dene graph of f .

Suggested Readings:

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

Tata Mc-Graw Hill Pvt. Ltd., New Delhi, 2011.

2. Kreyszig E.,” Introductory Functional Analysis with Applications,

John Wiley & Sons, New York, 2007.

3. Limaye B. V.,” Functional Analysis”, New Age International Ltd.,

Publishers, Second Edition, New Delhi,1996.

4. Co man C. and Pedrick G., “First Course in Functional Analysis”,

Prentice-Hall of India, New Delhi, 1995..

5. Conway J.B., “A Course in Functional Analysis”, Springer-Verlag,

New York, 2008.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
76 4.2. The Conjugate of an Operator

6. Bollobas B.,”Linear Analysis”, Cambridge University Press, Indian

Edition, New York, 1999.

7. Nair M.T., “Functional Analysis, A First course”, Prentice Hall of

India, New Delhi, 2010.

Web Resources:

// / / /
1. https: archive.nptel.ac.in courses 111 106 111106147/ /
// / /
2. https: people.math.sc.edu schep Openmapping.pdf

// / / /
3. http: www.math.lsa.umich.edu rauch 555 openmappingthm.pdf

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
4.2. The Conjugate of an Operator 77

Block-III
Unit-5: Hilbert Spaces-I

Unit-6: Hilbert Spaces-II

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
78
Block-III

UNIT-5

HILBERT SPACES-I

Structure
Objective
Overview
5. 1 The Denion and Simple Properties
5. 2 Orthogonal Complements
Let us Sum Up
Check Your Progress
Suggested Readings

Overview
In this unit, we derive open-mapping theorem, closed graph

theorem and uniform boundedness theorem.

Objectives
After successful completion of this lesson, students will be able to

ˆ prove open mapping theorem.

ˆ understand the concept of conjugate of an operator.

M.Sc.(Mathematics)-II Year-III Sem 79 Functional Analysis

 80 5.1. The Definion and Simple Properties

5.1. The Denion and Simple Properties

The Banach spaces studied in the previous Block are little more than

linear spaces provided with a reasonable notion of the length of a vector.

The main geometric concpet missing in an abstract space of this type is

that of the angle between two vectors. The theory of Hilbert spaces does

not hinge on angles in general, but rather on some means on telling when

two vectors are orthogonal.

Denition 5.1. A Hilbert space is a complex Banach space whose norm arise
from the inner product, which is a function ( ; ) : H  H ! C are called inner
product satises the following conditions:

1. ( x + y; z) = ( x; z) + (y; z) for all x; y; z 2 H and ; 2 C

2. ( x; y) = (y; x) where the bar denotes the complex conjugate.

3. ( x; x) = k xk2

Example 5.1. The space l2n with the inner product of two vectors
x = ( x1 ; x2 ; : : : ; xn ) and y = (y1 ; y2 ; : : : ; yn )

n
X
dened by ( x; y) = xi yi
i= 1

is a Hilbert space.

Solution. Let us verify the conditions given in the denition as follows:

Axiom 1: n
X
( x + y; z) = ( xi + yi )zi
i= 1
Xn n
X
= xi zi + yi zi
i= 1 i= 1
= ( x; z) + (y; z)

Axiom 2: 0
n 1
B
BX C
C
( x; y) = B
B
B
B
@
C
( xi yi )
C
C
C
A
i= 1
= ( x1 y1 + x2 y2 + : : : + xn yn )

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
5.1. The Definion and Simple Properties 81

= x1 y1 + x2 y2 + : : : + xn yn
= x1 y1 + x2 y2 + : : : + xn yn
= y1 x1 + y2 x2 + : : : + yn xn
= (y; x)

n
X
Axiom 3: ( x; x) = xi xi
i= 1
Xn
= j xi j = k xk
2 2

i= 1

Hence l2 is a Hilbert space. È

Example 5.2. Consider the space l2n consisiting of all sequence x = f xn g such
1
X
that j xn j . Dene the inner product on
2
l2 as follows:
n= 1
If x = f xn g and y = fyn g 2 l2 then
1
X
( x; y) = xn yn (5.1)
n= 1

and l2 is a Hilbert space.

Solution. First let us show that the inner product (5.1) is well-dened. For this
we have to show that (5.1) is a convergent series and having its sum as a complex
number. By Cauchy's inequality, we get
0 11=2 0 11=2
n
X BXn C BXn C
B C B C
xi yi  B
B
B
@ j xi j 2C
C C
A
B
B
B
@ yi j j 2C
CC
A
i= 1 i= 1 i= 1
0 11=2 0 11=2
BX1 C BX1 C
B C B C
 B
B
B
@ j xn j 2C
C
C
A
B
B
B
@ yn j j 2C
C
C
A
n= 1 n= 1

1
X 1
X
Since j xn j and
2
jyn j are convergent, the monotonic increasing sequence
2

n= 1 n= 1
of partial sums of the series (5.1) is bounded above. Hence (5.1) is convergent so
that the inner product (5.1) is well dened. È
Theorem 5.1. In a Hilbert space H , prove that
(i) ( x y; z) = ( x; z) (y; z)

(ii) ( x; y + z) = ( x; y) + ( x; z)

(iii) ( x; y z) = ( x; y) ( x; z)

(iv) ( x; 0) = 0 and (0; x) = 0 for every x 2 H

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
82 5.1. The Definion and Simple Properties

Proof.
(i) ( x y; z) = ( x+( )y; z)
= ( x; z) + ( )(y; z)
= ( x; z) (y; z)

(ii) ( x; y + z) = ( y + z; x)
= ( y; x) + ( z; x)
= (y; x) + (z; x)
= ( x; y) + ( x; z)

(iii) ( x; y z) = ( y+( )z; x)

= (y; x) + ( ) (z; x)
= ( x; y) ( x; z)

(iv) (0; x) = (0 0; x)
= 0(0; x) = 0

where 0 is the zero element of H .

and ( x; 0) = (0; x) = 0 = 0 . 
Note 5.1. In a Hilbert space, we have
( x; y + z) = ( x; 1y + 1z)
= 1( x; y) + 1( x; z)
= ( x; y) + ( x; z)

Theorem 5.2 (Schwarsz inequality). If x and y are any two vectors in a Hilbert
space H , then

j( x; y)j  k xkkyk

Proof. If y = 0 , then kyk= 0 and j( x; y)j= 0 . Therefore, in this case both sides
vanish and the result is quite obvious.
Now, let y 6= 0 . For any scalar  , we have

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
5.1. The Definion and Simple Properties 83

( x + y; x + y)  0
) ( x; x + y) + (y; x + y)  0
) ( x; x) + ( x; y) + (y; x) + (y; y)  0
k xk +( x; y) + (y; x) + jj kyk  0
2 2 2
(5.2)

Since y 6= 0 ) kyk6= 0 , therefore putting

( x; y)
 = kyk 2
in (5.2)

we get
( x; y) j( x; y)j
( x; y) 2

k xk 2
( x; y) (y; x) +

kyk
2
 0
kyk 2
kyk kyk 2 2 2

j( x; y)j ( x; y)(y; x) j( x; y)j

2 2

) k xk
kyk
2

kyk 2
+ kyk 2 2
 0
j( x; y)j j( x; y)j j( x; y)j
2 2 2

) k xk
kyk
2

kyk
+ kyk
2 2 2
 0
j( x; y)j 2

) k xk 2
 0
kyk 2

) k xk kyk j( x; y)j 2 2 2
 0
) j( x; y)j  k xk
kyk 

Theorem 5.3. In a Hilbert space the inner product is jointly

continuous. i:e:; xn ! x; yn ! y ) ( xn ; yn ) ! ( x; y)

Proof.

j( xn ; yn ) ( x; y)j = j( xn ; yn ) ( xn ; y) + ( xn ; y) ( x; y)j
= j( xn ; yn y) + ( xn x; y)j
 j( xn ; yn yj + j( xn x; y)j
 k xn kkyn yk+k xn xkkyk! 0 as n ! 1
) j( xn ; yn ) ( x; y)j ! 0 as n ! 1

Hence ( xn ; yn ) ! ( x; y) .
Thus, the inner product in a Hilbert space is jointly continuous. 

From elementary geometry, we know the parallelogram law which states

that the sum of the squares of the sides of a parallelogram equals the sum

of the squares of the diagonals. We shall show that the norm on a Hilbert

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
84 5.1. The Definion and Simple Properties

space satises the following important parallelogram law.

Theorem 5.4. If x and y are any two vectors in a Hilbert space H , then
 
k x + yk +k x2
yk2 = 2 k xk2 +kyk2

Proof. For any x; y 2 X , we get

k x + yk 2
= ( x + y; x + y)
= ( x; x + y) + (y; x + y)
= ( x; x) + ( x; y) + (y; x) + (y; y)
= k xk +( x; y) + (y; x) + kyk
2 2
(5.3)
kx yk = ( x y; x y)
2

= ( x; x y) + (y; x y)
= ( x; x) ( x; y) (y; x) + (y; y)
= k xk ( x; y) (y; x) + kyk
2 2
(5.4)

Adding (5.3) and (5.4), we get

 
k x + yk +k x
2
yk2 = 2 k xk2 +kyk2 

Lemma 5.1 (Polarization of identity). If x; y are any two vectors in a Hilbert

space H , then
4( x; y) = k x + yk k x 2
yk2 +ik x + iyk2 ik x iyk2

Proof.

k x + yk 2
= k xk +( x; y) + (y; x) + kyk
2 2
(5.5)
and k x yk 2
= k xk ( x; y) (y; x) + kyk
2 2
(5.6)

Subtracting (5.6) from (5.5), we get

k x + yk k x
2
yk2 = 2( x; y) + 2(y; x) (5.7)

Replacing y by iy in (5.7), we get

k x + iyk k x 2
iyk2= 2( x; iy) + 2(iy; x)
= 2i( x; y) + 2i(y; x)
Hence k x + iyk2 k x iyk =2
2i( x; y) + 2i(y; x) (5.8)

Multiplying both sides of (5.8), we get

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
5.1. The Definion and Simple Properties 85

ik x + iyk2 ik x iyk2 = 2( x; y) 2(y; x) (5.9)

Adding (5.8) and (5.9), we get

k x + yk k x2
yk2 +ik x + iyk2 ik x iyk2 = 4( x; y) 

Theorem 5.5. If B is a complex Banach space whose norm obeys the

parallelogram, and if an inner product on B by
4( x; y) = k x + yk k x2
yk2 +ik x + iyk2 ik x iyk2

then B is a Hilbert space.

Proof. It is given that for all x; y 2 B ,

k x + yk k x 2
yk2 = 2k xk +2kyk
2 2
(5.10)
and 4( x; y) = k x + yk k x 2
yk2 +ik x + iyk2 ik x iyk2 (5.11)

Now B will be a Hilbert space if we show that the inner product on B dened
by (5.11) satises the following properties:

(i) ( x; x) = k xk2

(ii) ( x; y) = (y; x)

(iii) ( x + y; z) = ( x; z) + (y; z)

(iv) ( x; y) = ( x; y)

Note that the properties (iii) and (iv) are equivalent to the single property
( x + y) = ( x; z) + (y; z)

Now, we shall prove them one by one.

(i) Replacing y by x in (5.11), we get
4( x; x) = k2 xk k0k +ik x + ixk ik x
2 2 2
ixk2
= 4k xk 0 + ik(1 + i) xk ik(1
2 2
i) xk2
= 4k xk +2ik xk 2ik xk
2 2 2

= 4k xk 2

) ( x; x) = k xk
2

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
86 5.1. The Definion and Simple Properties

(ii) To prove (ii) we shall take complex conjugates of both sides of (5.11). Since
k x + iyk2 etc are all real numbers, therefore they will not change on while taking
conjugates. So, we have
4( x; y) = k x + yk k x yk ik x + iyk +ik x iyk
2 2 2 2

= ky + xk k (y x)k iki(y ix)k +ik i(y + ix)k

2 2 2 2

= ky + xk ky xk iky ixk +ik(y + ixk

2 2 2 2

= 4(y; x)
) ( x; y) = (y; x)

(iii) Replacing x by ( x + y) and y by z in (5.11), we get

4( x + y; y) = k( x + y) + zk k( x + y)2
zk2 +ik( x + y) + izk2 ik( x + y) izk2 (5.12)

Now, k( x + y)zk2 = k( x + z) + yk2 .

Replacing x by x + z in (5.10), we get
k( x + z) + yk +k( x + z) yk
2 2
= 2k x + zk +2kyk 2 2

k( x + z) + yk 2
= 2k x + zk +2kyk k( x + z) 2 2
yk2 (5.13)

Again
k( x + z) yk2 = k(z y) + xk 2

= 2kz yk +2k xk k(z y) xk by (5.13)

2 2 2

= 2k (y z)k +2k xk k f( x + y) zgk

2 2 2

= 2ky zk +2k xk k( x + y) zk
2 2 2
(5.14)

Substitute the value of k( x + z) yk2 from (5.4) in (5.3), we get

n o
k( x + y) + zk 2
= 2k x + zk +2kyk 2 2
2ky zk2 +2k xk2 k( x + y) zk2
k( x + y) + zk k( x + y) zk
2 2
= 2k x + zk +2kyk 2 2
2ky zk2 2k xk2 (5.15)

Interchanging x and y in (5.15), we get

k( x + y) + zk k( x + y)
2
zk2 = 2ky + zk +2k xk 2 2
2k x zk2 2kyk2 (5.16)

Adding (5.15) and (5.16), we get

k( x + y) + zk k( x + y)
2
zk2 = k x + zk k x 2
zk2 +ky + zk2 ky zk2 (5.17)

Replacing z by iz in (5.17) and multiplying both the sides by i , we get

ik( x + y) + izk2 ik( x + y) izk2 = ik x + izk ik x 2
izk2 +iky + izk2 iky izk2 (5.18)

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
5.1. The Definion and Simple Properties 87

Adding (5.17) and (5.18), we get

k( x + y) + zk k( x + y) zk +ik( x + y) + izk ik( x + y) izk
2 2 2 2

n o
= k x + zk k x zk +ik x + izk ik x izk
2 2 2 2

n o
+ ky + zk ky zk +iky + izk iky izk
2 2 2 2
(5.19)

By (5.12), the left hand side of (5.19) is equal to 4(x + y; z) and the right hand
side of (5.19) is equal to 4(x; z) + 4(y; z) . Hence from (5.19), we have
4( x + y; z) = 4( x; z) + 4(y; z)
) ( x + y; z) = ( x; z) + (y; z)

(iv) We shall establish this results in several steps according as the scalars
takes di erent types of values.
(a) is a positive integer:
As proved in (iii) , we have
( x + z; y) = ( x; y) + (z; y) (5.20)

Replacing z by x in (5.20), we get

( x + x; y) = ( x; y) + ( x; y)
) (2 x; y) = 2( x; y)

Thus the result (iv) is true for = 2 . Now, assumes that (iv) is true for any
positive integer n i:e: assume that
(nx; y) = n( x; y) (5.21)

Then
((n + 1) x; y) = (nx + x; y)
= (nx; y) + ( x; y)
= n( x; y) + ( x; y)
= (n + 1)( x; y)

Thus (iv) is true for n + 1 if it was true for n . Hence by induction (iv) is true
for all positive integers .
(b) is a negative integer.
First we shall prove that ( x; y) = ( x; y) .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
88 5.1. The Definion and Simple Properties

Replacing x by x in (5.11), wee get

4( x; y) = k x + yk k x yk +ik x + iyk ik x iyk
2 2 2 2

= k ( x y)k k ( x + y)k +ik ( x iy)k ik ( x + iy)k

2 2 2 2

= k( x y)k k x + yk +ik x iyk ik x + iyk

2 2 2 2

n o
= k x + yk k x yk +ik x + iyk ik x iyk
2 2 2 2

= 4( x; y)
) ( x; y) = ( x; y)

(c) is a rational number.

Let = qp where p and q are integers and q 6= 0 . We have
!
p p
( x; y) = q
x; y = ( pz; y) where
q
= z:
= p(z; y) (5.22)
Also (qz; y) = q(z; y)
1
) (z; y) = q
(qz; y) (5.23)

Substituting the value of (z; y) from (5.23) in (5.22), we get

p
( x; y) = q
(qz; y)
= ( x; y)

(d) is a complex number.

First we shall show that (ix; y) = i( x; y) .
Replacing x by ix in in (5.11), wee get
4(ix; y) = kix + yk kix yk +ikix + iyk ikix iyk
2 2 2 2

= ki( x iy)k ki( x + iy)k +iki( x + y)k iki( x y)k

2 2 2 2

= k x iyk k x + iyk +ik x + yk ik x yk

2 2 2 2

= i k x iyk +i k x + iyk +ik x + yk ik x yk

2 2 2 2 2 2

n o
= i k x + yk k x yk +ik x + iyk ik x iyk
2 2 2 2

= i4( x; y)
) (ix; y) = i( x; y)

Now, let = 1 +i 2 , where 1 and 2 are real numbers. Then, we have

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
5.1. The Definion and Simple Properties 89

( x; y) = (( 1 +i 2 ) x; y)
= ( 1 x+i 2 x; y)
= ( 1 x; y) + (i 2 ; y)
= ( 1 x; y) + i( 2 x; y)
= 1 ( x; y) + i 2 ( x; y)
= ( 1 +i 2 )( x; y)
= ( x; y)

Thus, we have proved that ( x; y) = ( x; y) for each scalar .

Hence B is a Hilbert space and completes the proof of the theorem. 
Denition 5.2 (Convex sets). Let L be a linear space real or complex. A
non-empty subset S of L is said to be convex if x; y 2 S ) (1 ) x + y 2 S
where is any real number such that 0   1 .

Note 5.2. Taking = 1=2 , we see that if S is a convex subset of a linear space
L , then x; y 2 S ) ( x + y)=2 2 S .

Theorem 5.6. A closed convex subset C of a Hilbert space H contains a unique

vector of smallest norm.

Proof. Let d = inf fk xk: x 2 C g . Then there must exist a sequence of vector f xn g
in C such that k xn k! d .
Consider the two vectors xm and xn belonging to the sequence f xn g . Since C
is a convex subset of H , therefore
xm ; xn 2 C ) ( xm + xn )=2 2 C

Hence by denition of d , we have

k( xm + xn )=2k  d
) k xm + xn k  2d (5.24)

By the parallelogram law, we have

k xm + xn k +k xm2
xn k2= 2k xm k +2k xn k 2 2

k xm xn k = 2k xm k +2k xn k k xm + xn k
2 2 2 2

 2k xm k +2k xn k 4d 2 2 2
(5.25)

Now k xm k! d and k xn k! d . Therefore

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
 90 5.2. Orthogonal Complements

2k xm k2 +2k xn k2 4d2 ! 2d2 + 2d2 4d2 = 0

Hence as m; n ! 1 , we have k xm xn k2 ! 0 .

Therefore f xn g is a Cauchy sequence in C . Since H is complete and C is a

closed subset of H , therefore C is also complete.
Hence the Cauchy sequence f xn g in C is also a convergent sequence in C ,
Therefore there exists a vector x in C such that xn ! x .Now,
k xk = klim xn k
= limk xn k
= d

Hence x is a vector in C with smallest norm d . It remains to prove that the

uniqueness of x .
Uniqueness of x : Suppose x and x0 are two vectors in C such that k xk= d and
k x0 k= d . Then to show that x = x0 .
Since C is convex, therefore x; x0 2 C ) ( x + x0 )=2 2 C .
Hence by the denition of d , we have k( x + x0 )=2k d so that k x + x0 k 2d .
Now by the parallelogram law, we have
kx x0 k2 = 2k xk +2k x0 k k x + x0 k
2 2 2

 2k xk +2k x0 k 4d
2 2 2

= 2d + 2d 4d ! 0
2 2 2

Thus; k x x0 k2  0
But k x x0 k2  0

Hence we have k x x0 k2 = 0 ) x x0 =0 ) x = x0 . 

5.2. Orthogonal Complements

Denition 5.3. Let x and y be vectors in a Hilbert space H . Then x is said to

be orthogonal to y (written as x?y ) if ( x; y) = 0 .

Note 5.3. The relation of orthogonality in a Hilbert space is symmetric

i:e:; x?y ) y? x .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
5.2. Orthogonal Complements 91

For if, x?y ) ( x; y) = 0 .

) ( x; y) = 0 ) (y; x) = 0 ) y? x
Note 5.4. If x is orthogonal to y , then every scalar multiple of x is orthogonal
to y . Let be any scalar, then ( x; y) = ( x; y) =
0 = 0 . Therefore
x?y ) x?y .

Note 5.5. The zero vector is orthogonal to every vector. For every vector x in
H , we have (0; x) = 0 . Therefore 0? x for all x in H .

Note 5.6. The zero vector is the only vector which is orthogonal to itself. We
have x? x ) ( x; x) = 0 ) k xk2 = 0 ) x = 0 .
Hence if x? x , then x must be a zero vector.

Theorem 5.7 (Pythagoream theorem). If x and y are any two orthogonal vectors
in a Hilbert space H , then
k x + yk = k x
2
yk2 = k xk2 +kyk2

Proof. Given that x?y ) ( x; y) = 0 . Then, we must have y? x i:e:; (y; x) = 0

Now
k x + yk = ( x + y; x + y) = ( x; x) + ( x; y) + (y; x) + (y; y)
2

= k xk +0 + 0 + kyk = k xk +kyk
2 2 2 2

Also k x yk = ( x y; x y) = ( x; x) ( x; y) (y; x) + (y; y)

2

= k xk +0 + 0 + kyk = k xk +kyk
2 2 2 2

This proves the theorem. 

Denition 5.4. A vector x is said to be orthogonal to a non-empty subset S of
a Hilbert space H (written x?S ) if x?y for every y in S .
Two non-empty subsets S 1 and S 2 of a Hilbert space H are said to be
orthogonal (written as S 1 ?S 2 ) if x?y 8 x 2 S 1 and 8y 2 S 2 .

Notation: S? is the set of all vectors in H which is orthogonal to S .

Denition 5.5. Let S be any non empty subset of a Hilbert space H . Then S ?


is a subspace of H . We dene S ? ? , written as S ?? , by
n o
S ?? = x 2 H : ( x; y) = 0 8y 2 S ?

Theorem 5.8. If S ; S 1 ; S 2 are non-empty subsets of a Hilbert space H , then

prove the following:

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
92 5.2. Orthogonal Complements

(i) f0g? = H .
(ii) H ? = 0.
(iii) S \ S ?  f0g .
(iv) S 1  S ) S?  S? .
2 2 1

(v) S  S ?? .

Proof.
(i) First we shall prove that H  f0g? .
Let x 2 H . Since ( x; 0) = 0 , therefore x 2 f0g? .
Thus x 2 H ) x 2 f0g? .
Hence H  f0g? .
But f0g?  H .
Hence H = f0g? .
(ii) Let x 2 H . Then by denition of H ? , we have
( x; y) = 0 8y 2 H

Taking y = x , we get
( x; x) = 0 ) k xk2 = 0 ) x = 0

Thus, x 2 H ? ) x = 0 . Therefore H ? = f0g .

(iii) Let x 2 S \ S ? . Then x 2 S and x 2 S ? .
Since x 2 S ? , therfore x is orthogonal to every vector in S . In particular x
is orthogonal x , because x 2 S . Now
( x; x) = 0 ) k xk2 = 0 ) x = 0

Thus 0 is the only vector which can belong to both S and S ? . Therefore
S \ S ?  f0g

If S is a subspace of H , then 0 2 S . Also S ? is a subspace of H . Therefore

0 2 S ? . Thus, if S is a subspace of H , then 0 2 S \ S ? .
Therefore, in this case, we have S \ S ? = f0g .
(iv) Let S 1  S 2 . We have

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
5.2. Orthogonal Complements 93

x 2 s?2 ) x is orthogonal to every vector in S 2

) x is orthogonal to every vector in S 1 (* S 1  S 2 )
) x 2 S 1?

) S 2?  S? .
1

(v) Let x 2 S . Then ( x; y) = 0 8 y 2 S ? .

Therefore by denition of S ? ? ; x 2 S ?? .
Thus S  S ?? 
Note 5.7. If M and N are subspaces of a Hilbert space H and M ?N , then
M \ N = f0g .

For if, suppose x 2 M \ N , then x 2 M and x 2 N . Therefore

M ?N ) ( x; x) = 0 ) k xk2 = 0 ) x = 0 .

Theorem 5.9. Let S be a non-empty subset of a Hilbert space H . Then S ? is

a closed linear subspace of H .

Proof. By denition
S? = f x 2 H : ( x; y) = 0 8 y 2 S g

Since (0; y) = 0 8y 2 S , thus 0 2 S and hence S is non-empty.

Let x1 ; x2 2 S ? and let ; be any two scalars. Then (x1 ; y) = 0 8y 2 S and
( x2 ; y) = 0 8y 2 S .
For every y 2 S , we have
( x1 + x2 ; y) = ( x1 ; y) + ( x2 ; y)
= (0) + (0) = 0
) x1 + x2 2 S?

Hence S ? is a subspace of H .
Now, we shall show that S ? is a closed subset of H . We shall prove it by
showing that if x is any limit point of S ? , therefore there certainly exists a
sequence f xn g of points of S ? such that xn ! x . Now let y 2 S . Since
xn 2 S ? for every n , therefore ( xn ; y) = 0 for every n .

Consequently we have

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
94 5.2. Orthogonal Complements

lim( xn ; y) = 0
) (lim xn ; y) = 0
) ( x; y) = 0

Thus if x is a limit point of S ? , then ( x; y) = 0 8y 2 S . Therefore x 2 S ? .

Hence S ? is a closed subset of H . 
Theorem 5.10. Let M be a closed linear subspace of Hilbert space H , let x be
a vector not in M , and let d be the distance from x to M . Then there exists a
unique vector y0 in M such that k x y0 k= d .

Proof. By the denition of distance from x to M , we have

d ( x; M ) = d = inf fk x zk: z 2 M g

There must exist a sequence of vectors fyn g in M such that k x yn k! d .

Consider the two vectors ym and yn belonging to the sequence fyn g . Since
M is a linear subspace of H , therefore ym ; yn 2 M ) (ym + yn )=2 2 M .
ym + yn
Hence by the denition of d , we have x d
2
so that k2 x (ym + yn )k 2d .
Applying the parallelogram law for the vectors x ym and x yn
k( x ym ) ( x yn )k2 = 2k x ym k2 +2k x yn k2 k( x ym ) + ( x yn )k2
kyn ym k2  2k x ym k2 +2k x yn k2 k2 x (ym + yn k ) 2

 2k x ym k2 +2k x yn k2 4d2
= 2d2 + 2d2 4d2 ! 0
Thus; kyn ym k2 ! 0

Therefore fyn g is a Cauchy sequence in M . Since H is complete and M is

closed subspace of H , therefore M is also complete.
Hence the Cauchy sequence fyn g in M is also a convergent sequence in M .
Therefore there exists a vector y0 in M such that yn ! y0 . Now
kx y0 k = k x lim yn k
= klim(x yn )k
= limk x yn k= d

Hence y0 is a vector in M such that k x y0 k= d . It remains to prove the

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
5.2. Orthogonal Complements 95

uniqueness.
Uniqueness of y0 :
Suppose y1 and y2 are two vectors in C such that k x y1 k= d and
k x y2 k= d . Then to show that y1 = y2 .
Since C is convex, therefore y1 ; y2 2 C ) (y1 + y2 )=2 2 C .
y1 + y2
Hence by the denition of d , we have x  d so that
2
k2 x (y + y )k 2d .
1 2

Now by the parallelogram law, we have

k( x y1 ) ( x y2 )k2 = 2k( x y1 )k2 +2k( x y2 )k2 k( x y ) + ( x y )k
1 2
2

ky 2 y1 k2  2k( x y1 )k2 +2k( x y2 )k k2 x (y + y )k

2
1 2
2

= 2d2 + 2d2 4d2 ! 0

Thus; ky2 y1 k2  0
But ky2 y1 k2  0

Hence we have ky2 y1 k2 = 0 ) y2 y1 =0 ) y2 =y 1 . 

Theorem 5.11. If M is a proper closed linear subspace of a Hilbert space H ,
then there exists a non-zero vector z0 in H such that z0 ? M .

Proof. Since M is a proper subspace of H , therefore there exists a vector x in

H which is not in M . Let d be the distance from x to M . Then by denition
of distance we have
d = inf fk x yk: y 2 M g

Clearly y 6= x , because x 2= M . Therefore d > 0:

Since M is a proper subspace of H , therfore by Theorem 5.9 there exists a
vector y0 in M such that k x y0 k= d .
Now put z0 = x y0 , then kz0 k= k x y0 k= d > 0 .
Therefore z0 is a non-zero vector. It remains to prove that z0 ? M .
Let y be an arbitrary vector in M . Now, we shall prove that z0 ?y .
For any scalar , we have
z0 y = x y0 y=x (y0 + y)

Since M is a subspace of H and y0 ; y 2 M , therefore y0 + y 2 M . Then by

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
96 5.2. Orthogonal Complements

denition of d , we have k x (y0 + yk d . Now,

kz0 yk= k x (y0 + y)k d = kz k 0

 ) kz 0 yk 2
kz k0
2

) (z y; z y) (z ; z ) 
0 0 0 0 0
) (z ; z )0 0 (z ; y) (y; z ) + (y; y) (z ; z ) 
0 0 0 0 0
) (z ; y) (z ; y) + (y; y) 
0 0 0 (5.26)

The relation (5.26) is true for all scalars . Let us take = (z0 ; y) where is
an arbitrary real number. Then (z0 ; y) . Putting the values of ; in (5.26),
we get
(z0 ; y)(z0 ; y) (z0 ; y)(z0 ; y) + 2
(z0 ; y)(z0 ; y)kyk2  0
) 2 k(z0 ; y)k2 + 2 k(z0 ; y)k2 kyk2  0
h i
) k(z ; y)k kyk 2  0
0
2 2
(5.27)

The relation (5.27) is true for all real . Suppose that (z0 ; y0 ) 6= 0 . Then taking
positive and so small that kyk2 < 2 .
h i
We see from (5.27) that k(z0 ; y)k2 kyk2 2 < 0 . This contradicts (5.27).
Hence we must have (z0 ; y) = 0 which means that z0 ?y . Thus z0 ?y 8 y 2 M .
Therefore z0 ? M . 
Theorem 5.12. If M and N are closed linear subspaces of a Hilbert space H
such that M?N , then the linear subspace M + N is also closed.

Proof. Let z be any limit point of M + N . In order to show that M + N is closed,

it is enough to show that z 2 M + N .
Since z is a limit point of M + N , therefore there certainly exists a sequence
fzn g of points of M + N such that zn ! z . Since M?N , therefore the subspaces
M and N are disjoint i:e: M \ N = f0g and so the subspace M + N is the
direct sum of the subspaces M and N . Consequently each zn can be uniquely
written as zn = xn + yn where xn 2 M and yn 2 N . Consider the two vectors
zm = xm + ym ; zn = xn + yn belonging to the sequence fzn g . Let us check the
norm of
zm zn = ( xm xn ) + (ym yn )

Since xm xn 2 M and ym yn 2 N and M ?N , therefore (xm xn )?(ym yn ) .

Then by the Pythagorean theorem , we have

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
5.2. Orthogonal Complements 97

k( xm xn ) + (ym yn )k2= k xm xn k2 +kym yn k2

kzm zn k = k xm
2
xn k2 +kym yn k2 (5.28)

Now fzn g is a convergent sequence in the Hilbert space H and every convergent
sequence is a Cauchy sequence. Therefore as m; n ! 1 , we have
kzm zn k2 ! 0
) k xm xn k +kym
2
yn k2 ! 0 (from (5.28))
) k xm xn k2 ! 0 and kym yn k ! 0 2

Thus f xn g and fyn g are Cauchy sequences in M and N respectively.

Now every closed subspace of a complete metric space is itself. Since H is
complete and M; N are closed subspaces of H , therefore M; N are complete.
Consequently the Cauchy sequences f xn g and fyn g in M and N are convergent
sequences in M and N . Therefore there exists vectors x and y in M and N
such that xn ! x and yn ! y .
Now z = lim zn = lim( xn + yn ) = lim xn + lim yn = x + y 2 M + N .
Thus, z is a limit point of M + N and z 2 M + N . Therefore M + N is
closed. 
Theorem 5.13. If M is a linear subspace of a Hilbert space H , show that M is
closed , M = M?? .

Proof.
Let M be a subspace of a Hilbert space H and let M = M ?? = M ? ? .

By Theorem 5.9, we know that M ? ? is a closed subspace of H . Therefore
M = M ?? is a closed subspace of H .

Conversely, suppose that M is a closed subspace of H . Then to prove that

M = M ?? .

But, we know that M  M ?? .

Suppose that this inclusion is proper i:e:; M 6= M ?? .
Now M ?? is a Hilbert space and M is a proper closed subspace of M ?? .
Therefore there exists a non-zero vector z0 in M ?? such that z0 ? M or z0 2 M ? .
Now z0 2 M ? and z0 2 M ?? implies that
z0 2 M ? \ M ?? (5.29)

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
98 5.2. Orthogonal Complements

Since M ? is a subspace of H , then by theorem, we have

M ? \ M ?? = f0g (5.30)

From (5.29) and (5.30), we conclude that z = 0 . This contradicts the fact that z
is a non-zero vector. Therefore the inclusion M  M ?? cannot be proper and we
must have M = M ?? . 
Theorem 5.14 (Projection theorem). If M is a closed linear subspace of a
Hilbert space H , then H = M  M ? .

Proof. If M is a subspace of a Hilbert space H , then we know that

M \ M ? = f0g . Therefore in order to show that H = M  M ? , it is enough
to prove that H = M + M ? .
Now by Theorem 5.8, M ? is a closed subspace of H . Also M is given to be
a closed subspace of H . Therefore by the Theorem 5.11, M + M ? is a closed
subspace of H .
Let us put
N = M + M?: (5.31)

From (5.31), we see that M  N and M ?  N . Then by Theorem 5.8, we have

N?  M? and N ?  M ??
N ?  M ? \ M ?? = f0g
) N? = f0g
 ?
) N? = f0g?
) N ?? = H
h
) N = H * N = M + M?
i
is a closed subspace of H ) N ?? = N
Thus; N = M + M ? = H
Finally H = M + M ? and M \ M ? = f0g
) H = M  M?

This completes the proof of the theorem. 

Let us Sum Up:
In this unit, the students acquired knowledge to

ˆ prove the parallelogram law.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
5.2. Orthogonal Complements 99

ˆ understand the concept of orthogonal vectors in a Hilbert space.

Check Your Progress:

p
1. If L is an inner product space, show that ( x; x) has the properties

of a norm.

2. If x and y are any two vectors in a Hilbert space, then show that

(i) k x + yk k x2
yk2 = 4 Re ( x; y) .

(ii) ( x; y) = Re ( x; y) + i Re ( x; iy)

3. If M is a linear subspace of a Hilbert space H, show that M is

closed , M = M ?? .

4. State and Prove Schwarz Inequality.

5. State and Prove Polarization identity.

Suggested Readings:

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

Tata Mc-Graw Hill Pvt. Ltd., New Delhi, 2011.

2. Kreyszig E.,” Introductory Functional Analysis with Applications,

John Wiley & Sons, New York, 2007.

3. Limaye B. V.,” Functional Analysis”, New Age International Ltd.,

Publishers, Second Edition, New Delhi,1996.

4. Co man C. and Pedrick G., “First Course in Functional Analysis”,

Prentice-Hall of India, New Delhi, 1995..

5. Conway J.B., “A Course in Functional Analysis”, Springer-Verlag,

New York, 2008.

6. Bollobas B.,”Linear Analysis”, Cambridge University Press, Indian

Edition, New York, 1999.

7. Nair M.T., “Functional Analysis, A First course”, Prentice Hall of

India, New Delhi, 2010.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
100 5.2. Orthogonal Complements

Web Resources:

// / / / /
1. https: archive.nptel.ac.in courses 111 106 111106147 /
// /
2. https: mathworld.wolfram.com HilbertSpace.html

// / =
3. https: www.youtube.com watch?v lsFTQOSgCpk

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
Block-III

UNIT-6

HILBERT SPACES-II

Structure
Objective
Overview
6. 1 Orthonormal Sets
6. 2 The Conjugate space H
Let us Sum Up
Check Your Progress
Suggested Readings

Overview
In this unit, we shall extend the notion of orthonormal basis of

nite dimensional linear spaces to the Hilbert spaces.

Objectives
After successful completion of this lesson, students will be able to

ˆ understand the concept of orthonormal sets.

ˆ explain Gram-Schmidt Orthogonalization process.

M.Sc.(Mathematics)-II Year-III Sem 101 Functional Analysis

 102 6.1. Orthonormal Sets

6.1. Orthonormal Sets

Denition 6.1. Let H be a Hilbert space. If x 2 H is such that k xk= 1 .

i:e: ( x; x) = 1 , then x is said to be a unit vector or a normal vector

Denition 6.2. A non-empty subset fei g of a Hilbert space H is said to be an

orthonormal set if
(i) kei k= 1 for every i .

(ii) i 6= j ) ei ?e j i:e:; (ei ; e j ) = 0

Thus a non-empty subset of a Hilbert space H is said to be an orthonormal set

if it consists of mutually orthogonal unit vectors. Obviously an orthonormal set
cannot contain zero vector because k0k= 0 .

Existence of an Orthonormal Set: Every Hilbert space H which is not equal

to zero space possesses an orthonormal set.

Let 0 6= x 2 H . Then k xk6= 0 . Let us normalize x by considering the

x
vector e= . Obviously e is a unit vector. Thus, we have
k xk
x
kek =
k xk
= k1xk
k xk= 1

Thus e is a unit vector and the set feg containing only one vector is

necessarily an orthonormal set.

It should be noted that if f xi g is a non-empty set of mutually orthogonal

non-zero vectors in H and if in this set each xi is replaced by the

xi
corresponding unit vector ei = k xi k
. Then the resulting set fei g is an

orthonormal set.

Examples of Orthonormal sets

Example 6.1. Let the Hilbert space be l2n . Consider the subset fe1 ; e2 ; : : : ; en g
of l2n , where ei is the n -tuple with 1 in the ith place and 00 s elsewhere. This
collection of any non-empty subset of this collection forms an orthonormal set in
l2n .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
6.1. Orthonormal Sets 103

Example 6.2. Let the Hilbert space be l2 . Consider the subset fe1 ; e2 ; : : : ; en ; : : :g
of l2 , where ei is the n -tuple with 1 in the ith place and 00 s elsewhere. This
collection of any non-empty subset of this collection forms an orthonormal set in
l2 .

Theorem 6.1. Let fe1 ; e2 ; : : : ; en g be a nite orthonormal set in a Hilbert space

H . If x is any vector in H , then
Xn
j( x; ei )j2  k xk2 (6.1)
i= 1
X n
further x ( x; ei )ei ?e j (6.2)
i= 1

for each j .

Proof.
(i) Since fe1 ; e2 ; : : : ; en g is a nite orthonormal set and hence we have
(ei ; e j ) = 0 when i 6= j; (ei ; e j ) = 1 when i = j .
n
X
Choose the vector y = x ( x; ei )ei . We have
0 i= 1 1
B n
X n
X C
kyk 2
= (y; y) =
B
B
@x
B
B ( x; ei )ei ; x ( x; ei )
C
ei CCCA
i= 1 i= 1
n
X n
X n X
X n
= ( x; x) ( x; ei )(ei ; x) ( x; e j )( x; e j ) + ( x; ei )( x; e j )(ei ; e j )
i= 1 j= 1 i= 1 j= 1
n
X n
X n
X
= k xk 2
j( x; ei )j 2
j( x; ei )j + 2
j( x; ei )j 2

i= 1 i= 1 i= 1
n
X
= k xk 2
j( x; ei )j 2

i= 1

n
X
Now kyk2  0 . Therefore k xk2 j( x; ei )j  0 .
2

i= 1
n
X
Thus, we have j( x; ei )j  k xk .
2 2

i= 1
This proves the result (i) .
(ii) For each j where 1  j  n , we have

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
104 6.1. Orthonormal Sets

0 1 0 1
B n
X C BXn C
B
B
@x
B
B ( x; ei )ei ;
C
e j CCCA = ( x; e j )
B
B
B
B
@ ( x; ei )ei ;
C
e j CCCA
i= 1 i= 1
n
X
= ( x; e j ) ( x; ei )(ei ; e j )
i= 1
= ( x; e j ) ( x; e j )
= 0
n
X
Hence x ( x; ei )ei ?e j for each j . 
i= 1
Note 6.1. The result (i) is known as Bessel's inequality for nite orthonormal
sets.

Now our aim is to prove that the above theorem holds good even in the

case of an arbitrary orthonormal set. The following theorem will enable us

to dene suitably the sums involved in (i) and (ii) of Theorem 6.1 in the

case of an arbitrary orthonormal set.

Theorem 6.2. If fei g is an orthonormal set in a Hilbert space H , and if x is any

vector in H , then the set S = fei : ( x; ei ) 6= 0g is either empty or countable.

Proof. For each positive integer n(, consider the set )

k xk 2

Sn = ei : j( x; ei )j > 2

If the set S n contains n or more than n vectors, then we have

X
k xk 2

j( x; ei )j 2
> n
n
= k xk 2
(6.3)
e 2S
i n

But by Theoorem 6.1, we have

X
j( x; ei )j 2
 k xk 2

e 2S
i n

This contradicts (6.3). Therefore the set S n contains at most n 1 vectors.

Thus for each positive integer n , the set S n is nite. Now suppose ei 2 S .
Then ( x; ei ) 6= 0 . However small may be the vvalue of j( x; ei )j2 , we can take n
k xk 2

so large that j( x; ei )j2 > . Therefore if each ei 2 S , then ei must belong to

n
some S n . So we can write
1
[
S = Sn
n= 1

Thus S can be expressed as a countable union of nite sets. Therefore S itself

a countable set.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
6.1. Orthonormal Sets 105

If we have (x; ei ) = 0 for every i , then S is empty. Otherwise either S is a

nite set or it is a countably innite set. 
Theorem 6.3 (Bessel's inequality). If fei g is an orthonormal set in a Hilbert
space H , then
X
j( x; ei )j2
 k xk 2

for every vector x in H .

Proof. Let S = fei : ( x; ei ) 6= 0g .

Then by Theorem 6.2, we have either S is empty or it is countable.
X
If S is empty, then we have (x; ei ) = 0 8i . In this case we dene j( x; ei )k2

to be the number 0 and we have 0  k xk2 .

Thus if S is empty, when we have
X
j( x; ei )j2
 k xk 2

Now we suppose that S is not empty. Then either S is nite or it is countably

innite. If S if nite, then we can write S = fe1 ; e2 ; : : : ; en g for some positive
integer n .
In this case,we dene
X n
X
j( x; ei )j2
= j( x; ei )j  k xk
2 2
(by Bessel's inequality for nte cases)
i= 1

Finally suppose that S is counably innite. Let the vectors in S be arranged in

a denite order:
S = fe ; e ; : : : ; en ; : : :g
1 2

1
X
j( x; ei )j =
P
In this case let us dene 2
j( x; ei )j . But this sum will be dened
2

i= 1
1
X
only if we can show that the series j( x; ei )j is convergent and its sum does not
2

i= 1
change by any rearrangement of its terms i:e: by any rearrangement of the vectors
n
X
in the set S . By Bessel's inequality for nite cases we have j( x; ei )j  k xk .
2 2

i= 1
Since this relation is true for every positive integer n , therefore it must also be
true in the limit. So we have
1
X
j( x; ei )j2
 k xk 2
(6.4)
i= 1

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
106 6.1. Orthonormal Sets

1
X
From the above, we see that the series j( x; ei )j is convergent. Since all terms
2

i= 1
of this series are positive, therefore it is absolutely convergent and so its sum
will not change by any rearrangement of its. So we are justied in dening
X 1
X
j( x; ei )j =
2
j( x; ei )j .
2

i= 1
Thus, from (6.4), we have
X 1
X
j( x; ei )j
2
= j( x; ei )j  k xk
2 2

i= 1

This completes the proof. 

Theorem 6.4. If fei g is an orthonormal set in a Hilbert space H , and if x is an
arbitrary vector in H , then
X
x ( x; ei )ei ? e j

for each j .

Proof. Let S = fei : ( x; ei ) 6= 0g . Then by Theorem 6.2, either S is empty or it

is countable.
If S is empty, then we have (x; ei ) = 0 8 i . In this case, we dene
P
( x; ei )ei
to be the vector 0 and we have
X
x ( x; ei )ei = x 0 = x:

So, in this case we are to prove that x?e j for each j . Since S is empty, therefore
( x; e j ) = 0 8 j i:e:; x?e j 8 j . Hence the result.
Now, we assume that S is non-empty. Then either S is nite or countably
innite. If S is nite, then we can write S = fe1 ; e2 ; : : : ; en g for some positive
n
X
integer n . In this case, we dene
P
( x; ei )ei = ( x; ei )ei and we are to prove
i= 1
that
n
X
x ( x; ei )ei ? e j
i= 1

We already proved in Theorem 6.1.

Finally Assume that S is countably innite. Let the vectors in S be arranged
in a denite order
S = fe ; e ; : : : ; en ; : : :g
1 2

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
6.1. Orthonormal Sets 107
n
X
We put S n = ( x; ei )ei . For m > n ,we have
i= 1
m
X
2 m
X
k sm sn k 2
= ( x; ei )ei = j( x; ei )j
2

i= n + 1 i= n + 1

1
X
But by Bessel's inequality the series j( x; ei )j is convergent.
2

i= n + 1
m
X
Therefore as m; n ! 1 , we have j( x; ei )j ! 0 . This implies that
2

i= n + 1
k sm sn k2 ! 0 . Therefore the sequence f sn g is a Cauchy sequence in H . But
H is complete, so there exists a vector s in H such that sn ! s which can be
1
X
written in the form s = ( x; en )en .
n==1
Let us now dene
X 1
X
( x; ei )ei = ( x; en )en
n= 1

It remains to prove that the sum is well dened i:e: it does not depend on the
rearrangement of the vectors in S . Now, we observe that
 X 
x ( x; ei )ei ; e j = (x s; e j ) = ( x; e j ) ( x; s)
 
= ( x; e j ) lim sn ; e j
= ( x; e j ) lim( sn ; e j )
0 1
BXn C
If e j 2= S , then ( sn ; e j ) = B
B
B
B
@ ( x; ei )ei ;
C
e j CCCA = 0. Therefore lim(sn ; ei ) = 0. So,
  i= 1
x
P
( x; ei )ei ; e j = ( x; e j ) = 0 .
0 1
BXn C
If e j 2 S then ( sn ; e j ) = B
B
B
B
@ ( x; ei )ei ;
C
e j CCCA = ( x; e j ) if n > j . Therefore
i= 1
lim( sn ; e j ) = ( x; e j ) . So, in this case we have
 X 
x ( x; ei )ei ; e j = ( x; e j ) ( x; e j ) = 0

 
Thus, we have x
P
( x; ei )ei ; e j = 0 for each j.
P
i:e:; x ( x; ei )ei ; e j ?e j for each j .
P
Now we shall conclude the proof by showing that our denition of ( x; ei )ei
is valid.
Let the vectors in S be rearranged in any manner:
S = f f ; f ; : : : ; fn ; : : :g
1 2

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
 108 6.1. Orthonormal Sets

n
X
Put s0n = ( x; fi ) fi .
i= 1
As above, the sequence f s0n g converges to a vector s0 in H . We write
1
X
s0 = ( x; fn ) fn
n= 1

Now our wish is to prove that s = s0 .

For any arbitrarily chosen positive number  , let n0 be a poistive integer so
large that if n  n0 , then
k sm sk < ; k s0m s0 k< ;
1
X
and j( x; ei )j < 
2 2

i= n 0 + 1

For some positive integer m0 > n0 all terms of sn0 occur among those of s0m0 .
Therefore s0m0 sn0 is a nite sum consisting of terms of the type (x; ei )ei for
i = n0 + 1; n0 + 2; : : : This gives
1
X
k s0m 0
sn0 k2  j( x; ei )j < 
2 2

i= n 0 + 1
) k s0 m0 sn0 k  

Now,
k s0 sk = k( s0 s0m0 ) + ( s0m0 sn0 ) + ( sn0 s)k
 k s0 s0m k+k s0m 0 0
sn0 k+k sn0 sk
  +  +  = 3

Since  is arbitrary, we can choose it as small as we please.

Therefore k s0 sk< 3 ) s0 s=0 ) s0 = s.
This completes the proof of the theorem. 

6.1.1. Complete Orthonormal Sets

In three dimensional space the unit vectors e1 = (1; 0; 0); e = (0; 1; 0)
2 and

e3 (0; 0; 1) are orthonormal and as such fe ; e ; e g

1 2 3 consitute an orthonormal

set. Now the set fe ; e g

1 2 is also orthonormal, but this can be extended to

fe ; e ; e g
1 2 3 so that the extended set is orthonormal. On the otherhand, it is

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
6.1. Orthonormal Sets 109

not possible us to extend the set fe ; e ; e g

1 2 3 by adjoining to it any other unit

vector which is orthogonal to each of the vectors of the given set. Such a

set will be said to be complete orthonormal set in R3 .

Denition 6.3. An orthonormal set S in a Hilbert space H is said to be complete

if there exists no other orthonormal set containing S . That is S must be a
maximal orthonormal set.

Theorem 6.5. Every orthonormal set in a Hilbert space is contained in some

complete orthonormal set. Further, every non-zero Hilbert space contains a
complete orthonormal set.

Proof. Let S be an orthonormal set in a Hilbert space H . Let P be the class

of all orthonormal set containing S . Then P is non-empty because at least S
belongs to P . Now P is a partially ordered set with respect to set inclusion.
Let T be any totally ordered subset of P and let T fA ;  2 g . Now we shall
prove that fA :  2 g is an upper bound for T in P .
S

Clearly for every  , we have

[
A  fA :  2 g

fA :  2 g is in P .
S
Our Claim is that
Since each A contains S , therefore fA :  2 g also contains S . Now,
S

it remains to prove that fA :  2 g is an orthonormal set.

S

Let x; y be any two distinct vectors belonging to fA :  2 g . Then there

S

exists A and A belonging to T such that x 2 A and y 2 A . But T is

totally ordered. Therefore either A  A or A  A .
Without loss of generality let us take A  A . Then x; y 2 A . But A is an
orthonormal set. Therefore x?y and k xk= 1; kyk= 1 . Therefore fA :  2 g
S

is an orthonormal set. Thus the set fA :  2 g is an orthonormal set

S

containing S and each A  fA :  2 g . Therefore fA :  2 g is

S S

an upper bound for T in P . Thus P satises all the conditions of Zorn's lemma.
Therefore there must exist a maximal element in P . Let it be M . Then M is
a complete orthonormal set containing S . For if it is not so then there will exist
an orthonormal set containing S and also containing M propertly. This will
contradict the maximal character of M .
For the second part of the theorem, let H be a non-zero Hilbert space. Let x
be a non-zero vector in H . Then S = f x : k xk= 1g is an orthonormal set in H .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
110 6.1. Orthonormal Sets

Therefore by the rst part of this theorem there exists a complete orthonormal set
in H containing S .
This completes the proof of the theorem. 
Theorem 6.6. Let H be a Hilbert spoace, and let fei g be an orthonormal set in
H . Then the following conditions are all equivalent to one another:

1. fei g is complete;

2. x?fei g ) x = 0 ;

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 k xk2 =

P
j( x; ei )j 2
.

Proof.
(1) ) (2) : Given that the orthonormal set fei g is complete. Now, we shall prove
that x?fei g ) x = 0 .
x
Suppose x?fei g and x 6= 0 . Then e = is a unit vector such that
k xk
e?fei g i:e:; (e; ei ) = 0 . Therefore fe; ei g is an orthonormal set containing fei g .
This contradicts the hypothesis that fei g is a complete orthonormal set. Therefore
x?fei g ) x = 0 .

(2) ) (3) : Given that x?fei g ) x = 0 . Now, we are to show that if x is an

arbitrary vector in H , then x = ( x; ei )ei . By Theorem 6.4, we have the vector
P

P
x ( x; ei )ei is orthogonal to every vector in the set fei g .
P
i:e:; x ( x; ei )ei ?fei g .
( x; ei )ei = 0 ) x = ( x; ei )ei .
P P
Therefore by hypothesis we have x
(3) ) (4) : Given that for any vector x in H , we must have x = P( x; ei )ei .
Now, we shall prove that k xk2 = j( x; ei )j2 . We have
P

k xk2
= ( x; x)
0 1
B
BX X C
C
= B
B
B
B
@
( x; )ei ; ( x; e j ) C
e j CCAC
i j
XX
= ( x; ei )( x; e j )(ei ; e j )
i j
X
= ( x; ei )( x; ei )
i
X
= j( x; ei )j 2

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
6.1. Orthonormal Sets 111

(4) ) (1) : Given that if x is an arbitrary vector in H , then k xk2 =

P
j( x; ei )j .
2

Now our aim is to prove that fei g is complete.

Suppose that the orthonormal set fei g is not complete. Then fei g is a proper
subset of orthonormal set fei ; eg . By hypothesis we have
X
kek 2
= j(e; ei )j = 0 (* e?ei for each i)
2

Now kek2 = 0 contradicts the fact that e is a unit vector. Therefore the
orthonormal set fei g must be complete. 

Some standard terminology: Let fei g be a complete orthonormal set in

a Hilbert space H and let x be any vector in H. Then with respect

the complete orthonormal set the scalars ( x ei ) are called the Fourier ;
coecients of x , the expression x
P
= ;
( x ei )ei is called the Fourier expansions
of x and the equation
X
=
2
k xk2
( x; ei )

is called Parseval's equation or Parseval's identity.

Theorem 6.7. Let f x1 ; x2 ; : : : ; xn ; : : :g be a linearly independent set of vectors

in a Hilbert space H . Then there exists an orthonormal set of vectors
e1 ; e2 ; : : : ; en ; : : : such that, for every n the linear subspcae of H spanned by
fe ; e ; : : : ; en g
1 2 is the same as that spanned by x1 ; x2 ; : : : ; xn .

Proof. The set f x1 ; x2 ; : : : ; xn g is linearly independent set and hence none of the
xi0 s is a zero vector and none of the xi is expressible as a linear combination of
preceeding ones.
Now x1 6= 0 ) k x k6= 0 .
1

x1
Choose e1 = then ke1 k6= 0 as x1 6= 0 .
k x1 k
x 1
Also ke1 k= 1 =
k x k= 1 .
k x1 k k x1 k 1
Thus the set fe1 g is an orthonormal set.
Now, consider y2 = x2 ( x2 ; e1 )e1
Clearly y2 6= 0 because if it were zero, then x2 = ( x2 ; e1 )e1 . But e1 is
expressible in terms of x1 and hence x2 is expressible in terms of preceeding
vector x1 . This is not possible as the set of xi0 s is a linearly independent set.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
112 6.1. Orthonormal Sets

Also (y2 ; e1 ) = ( x2 ( x2 ; e1 )e1 ; e1 )

= ( x2 ; e1 ) (( x2 ; e1 )e1 ; e1 )
= ( x2 ; e1 ) ( x2 ; e1 )(e1 ; e2 )
= ( x2 ; e1 ) ( x2 ; e1 )
= 0

Thus y2 is orthogonal to e1 .
Now choose
e2= kyy k = k xx (( xx ;; ee ))ee k
2 2 2 1 1

2 2 2 1 1

y 1
ke k = =
ky k= 1
2
2
ky k ky k 2 2
2

Since y2 is orthogonal to e1 , and hence e2 is orthogonal to e1 .

) fe ; e g is an orthonormal set. Also e is a linear combination of x ; e or
2 1 2 2 1

a linear combination of x2 ; x1 . Thus, we get x2 is a linear combination of e1 ; e2 .

This amounts to saying that the linear space spanned by f x1 ; x2 g is same as
that spanned by orthonormal set fe1 ; e2 g .
Now consider y3 = x3 ( x3 ; e1 )e1 ( x2 ; e2 )e2 .
In choosing y3 we have subtracted from x3 its components along e1 and e2 .
Clearly y3 6= 0 because if it were zero then x3 would be a linear combination
of e1 ; e2 which in turn means a linear combination of x1 ; x2 . But this is not
possible because of linear independence of the set of xi0 s .
Also (y3 ; e1 ) = ( x3 ( x3 ; e1 )e1 ( x3 ; e2 )e2 ; e1 )
= ( x3 ; e1 ) ( x3 ; e1 )(e1 ; e1 ) ( x3 ; e2 )(e2 ; e1 )
= ( x3 ; e1 ) ( x3 ; e1 )
= 0

In a similar way, we can prove that (y3 ; e2 ) = 0 .

Now choose
e3 = kyy k = k xx (( xx ;; ee ))ee
3 3 3 1 1 ( x3 ; e2 )e2
( x3 ; e2 )e2 k
3 3 3 1 1

y 1
ke k = = 3

ky k= 1
3
ky k ky k 3 3
3

Thus y2 is orthogonal to both e1 and e2 and fe1 ; e2 g is an orthonormal set. As

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
6.1. Orthonormal Sets 113

y3 is orthogonal to both e1 and e2 and hence e3 is orthogonal to both e1 and

e2 .

Therefore the set fe1 ; e2 ; e3 g is an orthonormal set.

Now e3 is a linear combination of x3 ; e1 ; e2 or linear combination of
x3 ; x1 ; x2 . Hence x3 is a linear combination of e1 ; e2 and e3 .

This amount to saying that the linear space spanned by x1 ; x2 ; x3 is the same
as that spanned by orthonorml set e1 ; e2 ; e3 .
Continuing like thhis we can construct an orthonormal set of non-zero vectors
fe1 ; e2 ; : : : ; e j such that e j in the above set given by
x j ( x j ; e1 )e1 ( x j ; e2 )e2 : : : ( x j ; e j 1 )e j 1
ej =
k x j ( x j ; e1 )e1 ( x j ; e2 )e2 : : : ( x j ; e j 1 )e j 1 k
r
X
Consider yr+1 = xr+1 ( xr+1 ; ei )ei .
i= 1
Clearly yr+1 6= 0 for otherwise xr+1 will be a linear combination of
e1 ; e2 ; : : : ; er or a linear combination of x1 ; x2 ; : : : ; xr which is not possible
because of linear independence of the set of xi0 s .
Also yr+1 is orthogonal to each 0e j for j = 1; 2; : : : ; r . 1
B r
X C
) (yr+ ; e j ) =
B C
1
B
@ xr+1
B
B ( xr+1 ; ei )ei ; e j CCCA
i= 1
r
X
= ( xr+1 ; e j ) ( xr+1 ; ei )(ei ; e j )
i= 1

Summing with respect to the indices i , then all the terms will vanish except one
term, because (ei ; ei )=1 if i6= j and (ei ; ei ) = 0 if i = j;
) (yr+ ; e j ) = ( xr+ ; e j ) ( xr+ ; e j ) = 0
1 1 1

Thus yr+1 is orthogonal to each e1 ; e2 ; : : : ; er .

r
X
xr+1 ( xr+1 ; ei )ei
yr+1 i= 1
Choose er+1 = = .
kyr+1 k Xr
xr+1 ( xr+1 ; ei )ei
i= 1
Also er+1 is a linear combination of xr+1 ; e1 ; e2 ; : : : ; er or a linear combination
of xr+1 ; x1 ; x2 ; : : : ; xr . Thus, we get xr+1 is a linear combination of
e1 ; e2 ; : : : :er+1 .

This amounts to saying that the linear space spanned by f x1 ; x2 ; : : : ; xr+1 g is

the same as spanned by orthonormal set fe1 ; e2 ; : : : ; er+1 g .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
 114 6.2. The Conjugate Space H .
Therefore by induction we prove the existence of an orthonormal set.
This completes the proof of the theorem. 
Note 6.2. Some of the useful orthonormal sets in analysis are Legendre
polynomials, Hermite functions and Laguerre functions. These are obtained by
Gram-Schmidt process to sequence of simple functions.

6.2. The Conjugate Space H .

Let H be a Hilbert space. A continuous linear transformations from H
into C is called a continuous linear functional or more briey a functional
on H . Thus, if we say that f is a functional on H, then f will be a

continous linear functioal on H. The set B(H; C) of all continous linear

functionals on H is denoted by H and is called the conjugate space of

H . The elements of H  are called continuous linear functionals or more

briey functionals. Thus, if f 2 H then f is called a functional in H and
 

as mentioned above as f is a continous linear functional on H. In this

section we shall show that there exists a natural correspondence between

H and its conjugate space H .

Theorem 6.8. Let y be a xed vector in a Hilbert space H and let fy be a scalar
valued function on H dened by the formula fy ( x) = ( x; y) 8 x 2 H . Show that
fy is a functional in H . Prove that
kyk = k fy k

Proof. We know that (x; y) is a scalar i:e: a complex number for all
x 2 H therefore f x is a mapping from H to C . In order to prove that fy is
a functional on H we must establish that it is linear as well as continuous.
f is linear: Let x1 ; x2 2 H and c 2 C , then
fy (cx1 + x2 ) = (cx1 + x2 ; y)
= (cx1 ; y) + ( x2 ; y)
= c( x1 ; y) + ( x2 ; y)
= c fy ( x1 ) + fy ( x2 )

Hence fy is linear.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
6.2. The Conjugate Space H. 115

f is continuous: j fy ( x)j= j( x; y)j k xkkyk

But y is a xed vector and let kyk= k .

) j fy ( x)j  kk xk: (6.5)

Hence fy is bounded which in turn implies that fy is continuous.

f is norm preserving: i:e:; k fy k= kyk .
From the previous chapter, we have the following denitions of norm of a
continuous linear trasnformations.
kT k = sup fkT ( x)k: k xk 1g
= sup fk
(
T ( x)k: k xk= 1g
)
kT ( x)k
= sup k xk : x =6 0

Now, from (6.5), we have

j fy ( x)j
 kyk; k xk6= 0
(
k xk )
j fy ( x)j
) sup : x 6= 0  kyk
k xk
i:e:; k fy k  kyk

We will show that the above relation takes the form of equality. If y = 0 , then
fy ( x) = ( x; y) = ( x; 0) = 0 8 x .

) fy is zero functional. Also kyk= 0 when y = 0 .

Hence k fy k= kyk= 0 .
But if y 6= 0 then kyk6= 0 .
Now by deniton
k fy k = supfj fy ( x)j: k xk= 1g
 j fy ( x)j: k xk= 1

Choose

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
116 6.2. The Conjugate Space H.
x = kyyk
!
y
) k fy k  fy
kyk
!
y
= kyk
;y

= k1yk (y; y)
= k1yk
kyk 2

= kyk

Therefore k fy k kyk .
Hence, we have k fy k= kyk 8 y and thus fy is norm preserving. 
Theorem 6.9 (Riesz Representation Theorem). Let H be a Hilbert space and
let f be an arbitrary functional in H , then there exists a unique vector y in H
such that
f ( x) = ( x; y)

for every x in H .

Proof. Uniquness of y : Let us assume that there exists y in H such that

f ( x) = ( x; y) 8 x 2 H . Now, we will establish that this y is unique.

If y be not unique, then let y1 ; y2 be the vectors in H corresponding to an

arbitrary functional f in H  such that
f ( x) = ( x; y1 ) 8 x 2 H
f ( x) = ( x; y2 ) 8 x 2 H
) ( x; y ) 1 = ( x; y2 ) 8 x
) ( x; y y )
1 2 = 0 8 x
) y y 1 2 = 0
) y =y 1 2

Hence y is unique.
Now we shall prove the existence of y .
Existence of y :
Case 1: If f = O i:e: a zero functional, then

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
6.2. The Conjugate Space H. 117

f ( x) = O( x) = 0 8 x 2 H
Also f ( x) = ( x; y) 8 x 2 H
) ( x; y) = 0 8 x and hence y = 0

Thus when f = O then there eixsts y = 0 .

Case 2: Now consider the case when f 6= O so that f ( x) 6= 0 for some x 2 H .
Consider the set M = f x : f ( x) = 0; x 2 H g .
Clearly M is non-empty as at least O 2 M and we know that null space is
a subspace of H and it is also a proper subspace because f ( x) 6= 0 for some
x 2 H and all such x do not belong to M . Again we know that the null space of
continuous linear transformation is closed.
Hence M is proper closed linear subspace of a Hilbert space H , therefore
there exists a non-zero vector y0 2 H such that y0 ? M or y0 2 M ? and
(y0 ; x) = 0 8 x2M

Now we have to prove that there exists a vector y in H such that

f ( x) = ( x; y) 8 x 2 H . But x 2 H may be such that
(i) x 2 M

(ii) x = y0

(iii) x may be any other vector in H .

We shall prove that any suitable scalar multiple of y0 will be our required y ,
such that f ( x) = ( x; y) 8 x 2 H .
(i) x 2 M ) f ( x) = 0 . (by denition).
But f ( x) = ( x; y)
= ( x; y0 )
= ( x; y0 ) (* x 2 M and y0 ? M )

Thus f ( x) = 0 and ( x; y) = 0 where y = y0 .

) f ( x) = ( x; y) .

(ii) When x = y0 . Then by denition

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
118 6.2. The Conjugate Space H .
f ( x) = ( x; y)
f (y0 ) = (y ; y )
0 0 choosing y = y0
) f (y0 ) = (y ; y ) 0 0

= ky k 0
2

) = kfy(yk ) 0

2
0

i:e:; = kfy(yk ) 0

2
0

Hence for x = y0 there exist a vector y = y0 where = kfy(yk ) 0

2
and
f ( x) = ( x; y) .
0

(iii) When x 2 H [ x 2= M and x 6= y0 ]

We know that H = M  M ? .
Therefore any vector x 2 H can be uniquely expressed as the sum of a vector
m in M and a vector y0 in M ? where is any scalar.
) x = m + y0
But f (m) = 0 (by denition of M )
) f (x y0 ) = 0
f ( x) f (y0 ) = 0
f ( x)
) = f (y0 )
Also f ( x) = f (m + y0 )
= f (m) + f (y0 )
= (m; y) + (y0 ; y)
= (m; y) + ( y0 ; y)
= (m + y0 ; y)
= ( x; y)

Thus, in all the cases we proved that there exists a unique vector y such that
f ( x) = ( x; y) 8 x 2 H . Where y is a sclar multiple of y0 such that y0 ? M and
M is the null space of functional f . 
Theorem 6.10. The mapping  : H ! H  dened by (y) = fy where
fy ( x) = ( x; y) every x 2 H is an additive, one-to-one onto isometry but not
linear.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
6.2. The Conjugate Space H. 119

Proof.
(i) Let us prove that  is additive.
For this we have to show that
 (y 1 + y ) = (y ) + (y ) 8 y ; y 2 H
2 1 2 1 2

But from the denition, we have  (y1 + y2 ) = fy1 +y2 .

Hence for any x 2 H , we get
fy1 +y2 = ( x; y1 + y2 )
= ( x; y1 ) + ( x; y2 )
= fy1 ( x) + fy2 ( x)
= ( fy1 + fy2 )( x)

Hence fy1 +y2 = fy + fy

1 2
which implies that  (y1 + y2 ) = (y1 ) + (y2 ) .
(ii)  is one-to-one: Let y1 y2 2 H .
Then (y1 ) = fy1 and (y2 ) = fy2 . Then
= (y ) (y ) 1 2

) fy = fy 1 2

) fy ( x) = fy (X ) for all x 2 H
1 2

) ( x; y ) = ( x; y ) 1 2

) ( x; y y ) = 0 8 x 2 H 1 2

) (y y ; y y ) = 0 Choose x = y y
1 2 1 2 1 2

) ky y k = 0 1 2
2

) y = y 1 2

Thus  is one-to-one.
(iii)  is onto: Let f 2 H  . Then by the Theorem 6.9, there exists a vector
y 2 H such that f (y) = ( x; y) .

Since fy1 ( x) = ( x; y) , we get f = fy so that (y) = fy = f . Hence for f 2 H 

there exists a preimage y in H . Therefore  is onto.
(iv)  is an isometry: Let y1 ; y2 2 H . Then

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
120 6.2. The Conjugate Space H .
(y ) (y )
1 2 = k fy fy k
1 2

= k fy + f y k
1 ( 2)

= k fy y k
1 2

= ky y k
2 2

Thus,  is an isometry.
(iv)  is not linear: Let y 2 H and be any scalar. Then ( y) = f y .
Hence for any x 2 H , we get f y ( x) = ( x; y) = ( x; y) = fy ( x) which gives
f y = fy , so that ( y) = (y) .

Thus the mapping is not linear. Such a mapping is called conjugate linear.
Thus  is conjugate linear.
Hence the proof. 
Theorem 6.11. If H is a Hilbert space, then H is also a Hilbert space with the
inner product dened by
( f x ; fy ) = (y; x) (6.6)

Proof. First we shall verify that (6.6) satises the conditions of an inner product
space.
Let x; y 2 H and ; be complex sacalrs in Theorem 6.10, we have shown
that f y = fy .
Hence f y = fy = fy
 
Now f x + fy ; fz = ( f x + f y ; fz )
But ( f x + f y ; fz ) = (z; x + y)
 
Now z f x+ y = (z; x) + (z; y) = ( f x ; fz ) + ( fy ; fz )
Thus, we have ( f x + fy ; z) = ( f x ; fz ) + ( fy ; fz )
Now, ( f x ; fy ) = (y; x) = ( x; y) = ( fy ; f x )
Also, ( f x ; f x ) = ( x; x) = k xk2 = k f x k2
So that f x > 0 and k f x k= 0 if and only if f x = 0
Now the Hilbert space H is a complete normed linear space. Hence its
conjugate space H  is a Banach space with respect to the norm dened on H  :
Since the norm on H  is induced by the inner product, H  is a Hilbert space
with the inner product ( fy ; f x ) = (y; x) . 

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
6.2. The Conjugate Space H . 121

Corollary 6.1. The conjugate space H of H  is a Hilbert space with the inner
product dened as folllows:

If f ; g 2 H  , let F f and Fg be the corresponding elements of H obtained

by Riesz representation theorem. Then (F f ; Fg ) = (g; f ) denes an inner product
on H  .

Theorem 6.12. Every Hilbert space is reexive.

Proof. We shall show that the natural imbedding on H to H  is an isometric

isomorphism.
Let x be any xed element of H . Let F x be a scalar valued function dened
on H  by F x ( f ) = f ( x) for every f 2 H  . In Riesz representation theorem, we
have proved that F x 2 H  . Thus each vector x 2 H gives rise to a functional
F x in H  . F x is called the functional on H  induced by the vector x . Let
 : H ! H  dened by ( x) = F X for every x 2 H . As in Theorem 3.4, 
is an isometric isomorphism of H into H  . We shall that  is mapping of H
onto H  .
Let T 1 be a mapping from H into H  dened by T 1 ( x) = f x where
f x (y) = (y; x) for every y 2 H and T 2 be a mapping from H  into H  dened
by T 2 ( f ) = F f x , where F f x = ( f ; f x ) for f 2 H  . Then T 2  T 1 is a composition
of T 2 and T 1 from H to H  . By Theorem 6.10, T 1 and T 2 are one-to-one
and onto. Hence T 2  T 1 is the same as the natural imbedding  . For this we
shall show that ( x) = T 2  T 1 )(X ) for every x 2 H .Now
(T 2 T 1 ( x)) = T 2 (T 1 ( x)) = T 2 ( f x ) = F f x

By the denition of  , ( x) = F x . Hence to show that T 2  T 1 =  , we have to

prove that F x = F f x .
For this, let f 2 H  . Then f = fy where f corresponds to y in the
representation F f x ( f ) = ( f ; f x ) = ( fy ; f x ) = ( x; y) .
But ( x; y) = fy ( x) = f ( x) = F x ( f ) . Thus, we get F f x ( f ) = F x ( f ) for every
f 2 H  . Hence the mapping F f x and F x are equal. That is T 2  T 1 =  and so
 is a mapping of H onto H  , so that H is reexive.
From the above, we get
(F x ; Fy ) = (F f x ; F f y ) = ( fy ; f x ) = ( x; y)

Hence  is an isometric isomorphism of H onto H  are congruent. 

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
122 6.2. The Conjugate Space H .
Let us Sum Up:
In this unit, the students acquired knowledge to

ˆ Dene complete orthonormal set and characterize the same.

ˆ understand the concept of Bessle's inequality.

ˆ prove the Hilbert space is reexive.

Check Your Progress:

1. Show that in a Hilbert space H an orthonormal set S is complete if

and only if x?S ) x = 0.

2. If S is a non-empty subset of a Hilbert space, show that S? = S ??? .

3. If M is a linear subspace of a Hilbert space H, show that M is

closed , M = M?? .
4. State and Prove Bessel's Inequality.

5. State and Prove Riesz representation theorem.

Suggested Readings:

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

Tata Mc-Graw Hill Pvt. Ltd., New Delhi, 2011.

2. Kreyszig E.,” Introductory Functional Analysis with Applications,

John Wiley & Sons, New York, 2007.

3. Limaye B. V.,” Functional Analysis”, New Age International Ltd.,

Publishers, Second Edition, New Delhi,1996.

4. Co man C. and Pedrick G., “First Course in Functional Analysis”,

Prentice-Hall of India, New Delhi, 1995..

5. Conway J.B., “A Course in Functional Analysis”, Springer-Verlag,

New York, 2008.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
6.2. The Conjugate Space H . 123

6. Bollobas B.,”Linear Analysis”, Cambridge University Press, Indian

Edition, New York, 1999.

7. Nair M.T., “Functional Analysis, A First course”, Prentice Hall of

India, New Delhi, 2010.

Web Resources:

// / /
1. https: archive.nptel.ac.in courses 111 106 111106147/ / /
// / / / /
2. http: sta .ustc.edu.cn wangzuoq Courses 15F-FA Notes FA25.pdf /
// /
3. https: www.youtube.com watch?v zdO4Fe
r4io =

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
124
6.2. The Conjugate Space H . 125

Block-IV
Unit-7: Operator on Hilbert Space-I

Unit-8: Operator on Hilbert Space-II

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
126
Block-IV

UNIT-7

OPERATOR ON HILBERT

SPACE-I

Structure
Objective
Overview
7. 1 The Adjoint of an Operator
7. 2 Self-Adjoint Operator
Let us Sum Up
Check Your Progress
Suggested Readings

Overview
In this unit, we shall explain the di erence between the adjoint

and conjugate operator.

M.Sc.(Mathematics)-II Year-III Sem 127 Functional Analysis

 128 7.1. The Adjoint of an Operator

Objectives
After successful completion of this lesson, students will be able to

ˆ understand the concept of adjoint operator.

ˆ explain the concept of self-adjoint operator and its properties.

7.1. The Adjoint of an Operator

In this section, we shall introduce the concept of the adjoint of a bounded

linear operator on a Hilbert space. We point out the basic di erence

between the adjoint and the conjugate of an operator.

Note 7.1. Let H be a Hilbert space. The set of all bounded linear transformations
B
known as operators on H is denoted by (H ) . We already know that (H ) is B
B
a complex Banach space. Let T 2 (H ) we already shown that T give rise to
an operator T  called the conjugate of H dened asf
(T  f ) x = f (T ( x)) for every f 2 H

and every x 2 H . We have also shown that the mapping T ! T  is an isometric

B B
isomorphism of (H ) into (H  ) which reverses the product and preseves
the identity transformation. Since H  is a Hilbert space, T  gives rise to a
conjugate operator T  on H  . Since H is reexive and H = H  in the
natural imbedding T  = T .
In the denition of the conjugate T  of T , we have never made use of
this of the correspondence between H and H  . Now we make use of this
correspondence to dene an operator T  on H called the adjoint of T . Though
we are using the same symbol for the conjugate and adjoint operators on H ,
one should note that the conjugate operator is dened as H  , while the adjoint
is dened on H . However, if we identify H and H  under the natural
correspondence, then the adjoint of T and the conjugate of T will coincide.
Hence T  is the conjugate of T if it operates on H  and T  is the adjoint of T ,
if it operates on H .

Theorem 7.1. Let T be an operator on a Hilbert space H . Then there existts a

unique operator T  on H such that for all x; y 2 H; (T x; y) = ( x; T  y) .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
7.1. The Adjoint of an Operator 129

The operator I is called the adjoint of the operator T .

Proof. Existence of adjoint operator: First we shall show that if T is an

operator on H , then there exists a mapping T  of H into itself such that
(T x; y) = ( x; T  y) 8 x; y 2 H (7.1)

Let y be a vector in H and fy its corresponding functional in H  . Let T  fy fz

where T  is the conjugate of the operator T . Corresponding to the functional
fz in H  the vector in H is z . Thus starting with a vector y in H we have
arrived at a vector z in H . Actually we have formed the product of following
three mappings:
y ! fy ! T  fy = fz ! z

If we denote the product of these mapping by T  then T  a mapping of H into

itself and is such that T  y = z . We call this new mapping T  of H into itself the
adjoint of T . It should be noted that if we identify the Hilbert spaces H and H 
by the natural correspondence y ! fy , then the conjugate of T and the adjoint
of T are the same mappings. Now we shall show that the mapping T  which we
have called the adjoint of T satises the property (7.2).
For every x in H , we have
(T  fy x) = fy (T x)
= (T x; y)
Also (T  fy x) = fz ( x)
= ( x; z)
= ( x; T  y)

Thus, we have (T x; y) = ( x; T  y) 8 x; y 2 H . In this way we see that there exists

a mapping T  of H into itself which satisifes the property (7.2). Now we shall
show that this mapping T  of H into itself is an operator on H . For this we
have to show that the mapping T  is linear and continuous (bounded).
T  is linear: Let y1 ; y2 be any two vectors in H and let ; be any two scalars.
For any vector x in H , we have

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
130 7.1. The Adjoint of an Operator

( x; T  ( y1 + y2 )) = (T x; y1 + y2 )
= (T x; y1 ) + (T x; y2 )
= ( x; T  y1 ) + ( x; T  y2 )
= ( x; T  y1 ) + ( x; T  y2 )
= ( x; T  y1 + T  y2 )

Thus, we have
( x; T  ( y1 + y2 )) = ( x; T  y1 + T  y2 ) 8 x 2 H:

But we know that (x; y) = ( x; z) 8 x 2 H ) y = z . Therefore we have

T  ( y1 + y2 ) = T  y1 + T  y2

Thus, the mapping T  is linear.

T  is continuous: For any vector y in H , we have
kT  yk 2
 (T  y; T  y)
= (T T  y; y)
= j(T T  y; y)j
 kT T  ykkyk
 kT kkT  ykkyk
i:e:; kT  yk 2
 kT kkT  ykkyk (7.2)

If kT  yk= 0 , then we have kT  yk kT kkyk because kT kkyk 0 .

If kT  yk6= 0 ,then from (7.2), we have
kT  yk  kT kkyk

Thus, we have
kT  yk  kT kkyk 8y 2 H

Let kT k= K . Then clearly K  0 and we have

kT  yk  K kyk 8y 2 H

Therefore the mapping T  is bounded and so it is continuous.

Thus T  is a continuous linear transformation from H into itself. Therefore
T  is an operator on H . Thus there exists an operator T  on H satisfying the

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
7.1. The Adjoint of an Operator 131

property (7.2).
Uniqueness of the adjoint operator: Now we shall show that if there is any
mapping T 0 of H into itself such that
(T x; y) = ( x; T 0 y) 8 x; y 2 H

then we must have T 0 = T  .

For if
(T x; y) = ( x; T 0 y) and (T x; y) = ( x; T  y) 8 x; y 2 H
) ( x; T 0 y) = ( x; T  y)
) T 0y = T y 8 y2H
) T0 = T

This proves the uniqueness of the adjoint operator.

Hence the proof. 
Note 7.2. In all our future work, the symbol T  will always signify the adjoint
of T and will never signify its conjugate.

Note 7.3. From the above theorem we can conclude that the mapping T ! T  is
B
a mapping of (H ) into itself. This mapping is called the adjoint operation on
B (H ) .

Theorem 7.2. The adjoint operation T ! T  on B (H ) has the following

properties:
1. (T 1 + T 2 ) = T + T1 2

2. ( T ) = T

3. (T 1 T 2 ) = T T 
2 1

4. T  =T
5. kT  k= kT k
6. kT  T k= kT k 2

Proof. (1) For every x; y 2 H , we have

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
132 7.1. The Adjoint of an Operator

( x; (T 1 + T 2 ) y) = ((T 1 + T 2 ) x; ; y)
= (T 1 x + T 2 x; y)
= (T 1 x; y) + (T 2 x; y)
= ( x; T 1 y) + ( x; T 2 y)
= ( x; (T 1 + T 2 )y)

Hence from the uniqueness of adjoint operator, we have

(T 1 + T 2 ) = T 1 + T 2

(2) For every x; y 2 H and be any scalar, then we have

( x; ( T ) y) = (( T ) x; y)
= ( (T x); y)
= (T x; y)
= ( x; T  y)
= ( x; T  y)

Hence from the uniqueness of adjoint operator, we have

( T ) = T

(3) For every x; y 2 H , then we have

( x; (T 1 T 2 ) y) = ((T 1 T 2 ) x; y)
= (T 1 (T 2 x); y)
= (T 2 x; T 1 y)
= ( x; T 2 (T 1 y))
= ( x; (T 2 T 1 )y)

Hence from the uniqueness of adjoint operator, we have

(T 1 T 2 ) = T 2 T 1

(4) For every x; y 2 H , then we have

( x; T  y) = ( x; (T  ) y)
= (T  x; y)
= (y; T  x)
= (T y; x) = ( x; T y)

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
7.1. The Adjoint of an Operator 133

Hence from the unqiueness of adjoint operator, we have

T  = T

(5) For every x 2 H , then we have

kT  xk 2
= (T  x; T  x)
= (T T  x; x)
= j(T T  x; x)j
 kT T  kk xk
 kT kkT  xkk xk
(
) kT  x)k  kT kk xk (7.3)
kT  xk
i:e:; sup :x=6 0  kT k
k xk
i:e:; kT  k  kT k (7.4)

If k xk 1 then from (7.3), we have kT  xk kT k .

) sup fkT  xk: k xk 1g  kT k .
i:e:; kT  k kT k which is same as (7.3).
Replacing T by T  in the above we get
kT  k  kT  k
i:e:; kT k  kT  k (* T  = T ) (7.5)

Thus, from (7.4) and (7.6), we have

kT  k = kT k (7.6)

(6) For every x 2 H , then we have

kT xk2
= j(T x; T x)j
= j(T  T x; x)j
 kT  T xkk xk
 kT  T kk xkk xk (7.7)

Now if k xk 1 then from (7.7), we conclude that

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
134 7.1. The Adjoint of an Operator

kT xk  kT  T k
2

) supfkT xk : k xk 1g
2
= kT  T k

) supfkT xk: k xk 1g = kT  T k
2

) kT k  kT  T k
2
(7.8)
Again kT  T k  kT  kkT k
= kT kkT k
= kT k 2

) kT  T k  kT k 2
(7.9)

Thus, from (7.8) and (7.9), we have

kT  T k = kT k 2

Hence the theorem. 

Corollary 7.1. Prove that kT T  k= kT k2 .

Proof. From the above theorem, we have

kT  T k = kT k 2

Applying the above result for T  in place of T , we have

k(T  ) T  k = kT k 2

) kT  T  k = kT k 2
(* kT  k= kT k)
) kT T  k = kT k 2

Hence the proof. 

Corollary 7.2. Prove that the adjoint operation is one-to-one onto as a mapping
B
of (H ) onto itself.

Proof. Let : B(H) ! B(H) be dened as

(T ) = T 8 T 2 B (H )

Now, we shall prove is one-to-one and onto.

is one-to-one:

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
7.1. The Adjoint of an Operator 135

Let (T 1 ) = (T 2 )
) T 1 = T 2
)


T 1  =

T 2 
) T1 = T2

Hence is one-to-one.
is onto: Let T be any arbitrary operator in B(H) then we have proved that
T  exists and belongs to (H ) and B
(T  ) = (T  ) = T  = T

Hence is onto. 
Theorem 7.3. If 0 and I be zero and identity operators in a Hilbert space H ,
then 0 = 0 and I  = I . Hence prove that if T is non-singular operator on H
then T  is also non-singular and in this case (T  ) 1
= (T 1
) .

Proof. ( x; 0 y) = (0 x; y) = (0; y) = 0 = ( x; 0y) .

Since adjoint is unique, ) 0 = 0
Again ( x; I  y) = (I x; y) = ( x; y) = ( x; Iy) .
) I = I .
T is non-singular ) T is invertible.

) TT 1
=T 1
T =I.
Taking adjoints, we get
   
TT 1
= T 1T = I
) (T ) (T  )
1
= T  (T 1
) = I

Above relation exhibits that T  is invertible and hence non-singular and

therefore, we have (T  ) 1 = (T 1 ) .
i:e:; Inverse of adjoint of T = adjoint of inverse of T . 

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
 136 7.2. Self-Adjoint Operators

7.2. Self-Adjoint Operators

Denition 7.1. A linear operator T on a Hilbert space is said to be self adjoint

if T  = T . In otherwords (T x; y) = ( x; T y) when T is self-adjont.

Note 7.4. 0 and the identity operator are self-adjoint operators.

Theorem 7.4. If fAn g is a sequence of self-adjoint operators on a Hilbert space

H and if fAn g converges to an operator A , then A is self-adjoint operator.

Proof. Let fAn g be a sequence of self-adjoint operators and let An ! A , since

each An is self-adjoint, we have An = An for n = 1; 2; 3; : : : . Now we shall
prove that A is a self-adjoint operator. i:e:; A = A .
A A= A An + An An + An A
kA A k = kA An + An An + An A k
 kA An k+kAn An k+kAn A k
= k (A An )k+kAn An k+kAn A k
= kA An k+kAn An k+k(An A) k
= kA An k+0 + kAn Ak
= 2kA An k! 0 as An ! A

Hence kA A k! 0 (or) A A = 0 i:e:; A = A .

) A is self-adjoint operator. 
Theorem 7.5. The self adjoint operator in B(H) form a closed linear subspace
and therefore a real Banach space which contains the identity transformation.

Proof. Let T 1 ; T 2 be self-adjoint operators on a Hilbert space H so that T 1 = T 1

and T 2 = T 2 . Now, if and be any two real numbers then
( T 1 + T 2 ) = ( T 1 ) + ( T 2 )
= T 1 + T 2 (* and are real numbers)
= T1 + T2

Hence T 1 + T 2 is also self adjoint operator on H when T 1 and T 2 are

self-adjoint. Thus, we conclude that the set of all self-adjoint operators form a

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
7.2. Self-Adjoint Operators 137

real subspace. Now, we have to show that it is closed subset of the Banach space
B (H ) .
Let fT n g be a sequence of self-adjoint operators converging to an operator T .
By Theorem 7.4, we have T is also a self-adjoint operator.
Again we know that the identity operator is a self-adjoint. Hence the set of all
B
adjoint operators in (H ) forms a closed linear sub-space and therefore a real
Banach space which contains the identity operator. 
Theorem 7.6. If T 1 and T2 be self adjoint operators then their product T1 T 2
as self adjoint , T1T2 =T T
2 1 .

In otherwords, the product of two self-adjoint operators is self adjoint if and

only if they commute.

Proof. Given that T 1 and T 2 are self adjoint operators, i:e:; T 1 = T 1 and
T 2 = T 2 .

Assume that T 1 T 2 is self-adjoint, then

(T 1 T 2 ) = T1T2
) T 2 T 1 = T1T2
) T2T1 = T1T2

i:e:; T 1 and T 2 commute.

Conversely, assume that T 1 T 2 = T 2 T 1 . Now,

(T 1 T 2 ) = T 2 T 1 = T 2 T 1 = T 1 T 2 .
Thus, T 1T 2 is self-adjoint. 
Theorem 7.7. If T is an operator on H for which (T x; x) = 0 for all x if and
only if T = 0:

Proof. Let T = 0 . Then for all x 2 H , we get

(T x; x) = (0 x; x) = (0; x) = 0 .
Converse Part: Let (T x; x) = 0 for every x 2 H . Now, we shall prove that T
is the zero operator.
Let x; y 2 H and ; be any scalars. Then, we have

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
138 7.2. Self-Adjoint Operators

(T ( x + y); x + y) = ( T x + T y; x + y) (7.10)
( T x + T y; x + y) = T ( x; x + y) + T (y; x + y) (7.11)
(T x; x + y) = (T x; x) + (T x; y) (7.12)
(T y; x + y) = (T y; x) + (T y; y) (7.13)

Using (7.11), (7.12) and (7.13) in (7.10), we get

(T ( x + y); x + y) = (T x; x) + (T x; y) + (T y; x) + (T y; y)
= j j (T x; x) +
2
(T x; y) + j j2 (T y; y) (7.14)

Since T ( x; x) = 0 for every x 2 H , the left hand side of (7.14) is zero so that we
get
(T x; y) + (T y; x) = 0 (7.15)

for all scalars ; and x; y 2 H .

Taking = 1; = 1 and = i; = 1 successively in (7.15), we get
(T x; y) + (T y; x) = 0 (7.16)
and i(T x; y) i(T y; x) = 0 (7.17)

Multiplying (7.16) by i and then adding to (7.17), we get

2i(T x; y) = 0 for every x; y 2 H (7.18)

since 2i 6= 0 and Taking y = T x in (7.18), we have (T x; T x) = 0 for every

x 2 H which gives T x = 0 for every x in H . This proves that T = 0. 
Theorem 7.8. An operator T on H is self-adjoint , (T x; x) is real for all x .

Proof. If T be self-adjoint then T  = T .

Now (T x; x) = ( x; T  x) = ( x; T x; ) = (T x; x) .
Thus, (T x; x) is real for all x .
Converse: Assume that (T x; x) is given to be real for all x .
) (T x; x) = (T x; x) = ( x; T  x) = (T  x; x) 8 x .
i:e:; (T x; x) = (T  x; x) = 0

i:e:; ((T T  ) x; x) = 0 8 x.
) T T = 0.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
7.2. Self-Adjoint Operators 139

Thus, T = T  i:e:; T is self-adjoint. 

Denition 7.2. If A1 ; A2 belong to a set of self-adjoint operators on a Hilbert
space H , then we dene a relation “less than or equal to”  as under.
A1  A2 if (T 1 x; x)  (T 2 x; x) 8 x 2 H:

In the next theorem we shall prove that the above relation is partially

ordered in the sense that it is reexive, transitive but anti-symmetric which

means that

a R b; b R a ) a=b
i:e:; A1 A ; 2 A2 A )1 A1 =A :
2

Theorem 7.9. The real Banach space of all self-adjoint operators on a Hilbert
space H is a partially ordered set whose linear structure and order structure are
related by the following properties:

(a) A1 A 2 then A1 + A  A2 + A for every A .

(b) A1 A 2 and 0 then A1  A2 .

Let us denoted by B be the set of all self-adjoint opoerators on H . First we shall

prove that this set is partially ordered set i:e:; the relation  is partially ordered.

Reexive: Clearly A1  A1 , becuase

(A1 x; x) = (A1 x; x) 8 x 2 H and A1 2 B:

Hence (A1 x; x)  (A1 x; x) 8 x2 H.

) By denition A1 A 1 i:e: the relation  is reexive.

Transitive: Let A1  A2 ; A2  A3 .
) ( A x; x)  ( A x; x) 8
1 2 x
and (A x; x)  (A x; x) 8
2 3 x

From the above relations, we can see that (A1 x; x)  (A3 x; x) 8 x.

Therefore, by denition A1 A 3 . Hence the relation  is transitive.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
140 7.2. Self-Adjoint Operators

Anti-Symmetric:
 A and A  A
Let A1 2 2 1

) ( A x; x)  ( A x; x) 8 x 2 H
1 2

and (A x; x)  (A x; x) 8 x 2 H
2 1

Hence (A x; x) = (A x; x) 8 x 2 H
1 2

) ((A A ) x; x) = = 8 x 2 H
1 2

) A A = 1 2 0

) A = A 1 2

Therefore the relation is  is anti-symmetric.

 A A1 2

) ( A x; x)  ( A x; x) 8 x 2 H
1 2

) (A x; x) + (Ax; x)  (A x; x) + (Ax; x)
1 2

) ((A + A) x; x)  ((A + A) x; x) 8 x 2 H
1 2

) A +A  A +A 1 2

Again A1  A 2

) ( A x; x) 1  ( A x; x) 8 x 2 H
2

) ( A x; x) 1  ( A x; x)
2

) ( A x; x) 1  ( A x; x) 8 x 2 H
2

) (( A ) x; x) 1  (( A ) x; x) 8 x 2 H
2

) A 1  A 2

Denition 7.3. A self-adjoint operator T on H is said to be a positive operator

if T  0 i:e: if (T x; x)  0 8 x 2 H .

Remark 7.1. Identity and zero operatros are positive operators.

We know that both the identity and zero operators are self-adjoint. Also
(I x; x) = ( x; x) = k xk2  0
(0 x; x) = (0; x) = 0 8 x 2 H

Hence I and 0 are both positive operators.

Now, we shall prove that for any arbitrary operator T , both T T  and T  T are
positive.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
7.2. Self-Adjoint Operators 141

(T T  ) = (T  ) T  = T T 
(T  T ) = T  (T  ) = T T

Thus, both T T  and T  T are self adjoint. Also,

(T T  x; x) = (T  x; T  x) = kT  xk2  0
(T  T x; x) = (T x; T  x) = (T x; T x) = kT xk2  0

Hence T T  and T  T are both positive operators.

Theorem 7.10. If T is a positive operator on a Hilbert space H , then I + T

is non-singular. In particular, I + T  T and I + TT are non-singular for an
arbitrary operator T on H .

Proof. We know that any transformation is non-singular if and only if it is

one-to-one and onto.
(I + T ) is one-to-one:
Let (I + T ) x = (I + T )y
) (I + T )( x y) = 0
) (I + T )z = 0 where z = x y
) Iz + T z = 0
) Tz = z
) (T z; z) = ( z; z) = (z; z) = kzk2

But L:H:S (T z; z)  0 as T is a positive operator whereas R:H:S kzk is

2

negative.
Above is possible only when each side is zero.
kzk 2
= 0

Consequently z = 0 or x y = 0 or x = y .
Hence I + T is one-to-one.
(I + T ) is onto: Since I + T is one-to-one and for showing it to be onto, we have
to establish that range of I + T i:e: R(I + T ) is H itself.
For this, we shall rst establish that R(I + T ) is a closed linear subspace of H .
Let x; y 2 H so that x + y 2 H where ; are scalars.
Let (I + T ) x; (I + T )y both belong to R(I + T ) .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
142 7.2. Self-Adjoint Operators

(I + T ) x + (I + T )y = (I + T ) x + (I + T ) y
= (I + T )( x + y) = (I + T )z where z 2 H
2 R(I + T )

R(I + T ) is a subspace of H .

Now, we shall show that R(I + T ) is a closed linear subspace. i:e: every limit
of it belongs to it or it is sucient to show that it is complete subset. i:e: every
Cauchy sequence in it is convergent.
Let f(I + T ) xn g be a Cauchy sequence in R(I + T ) so that for every  > 0
there exists n0 such that 8m; n  n0 .
k(I + T ) xn (I + T ) xm k  
) k(I + T )( xn xm )k  
) k( xn xm ) + T ( xn xm )k  
) k( xn xm ) + T ( xn xm )k 2
  2
(7.19)

But we know that

k x + yk2
= ( x + y; x + y)
= ( x; x) + ( x; y) + (y; x) + (y; y)
= k xk +kyk +( x; y) + (y; x)
2 2

= k xk +kyk +( x; y) + ( x; y)
2 2

) By (7.19), we have
k( xn xm )k +kT ( xn xm )k + (T ( xn
2 2
xm ); ( xn xm )) + (T ( xn xm ); ( xn xm ))   2

But when T is self adjoint then (T x; x) is real so that (T x; x) = (T x; x) and as

T is positive.
k( xn xm )k2 +kT ( xn xm )k2 +2 (T ( xn xm ); ( xn xm ))   2

All the terms in the L:H:S are positive,

) k xn xm k2   ) k xn
2
xm k 

) f xn g is a convergent sequence and let xn ! x2 H.

Therefore (I + T ) x 2 R(I + T ) is the limit of (I + T ) xn .

Thus, R(I + T ) is a closed linear sub space of H .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
7.2. Self-Adjoint Operators 143

We have to prove that R(I + T ) is H itself. Let us suppose that R(I + T ) is

not equal to H , then it is a proper closed linear sub space of H and in that case
there exists a non-zero vector y0 in H such that y0 ? R(I + T ) .
Now, y0 2 H ) (I + T )y0 2 R(I + T ) .
) y0 ?(I + T )y0 or y0 ?y0 + T y0 .

i:e:; (y0 + T y0 ; y0 ) = 0

i:e:; (y0 ; y0 ) + (T y0 ; y0 ) = 0 .

i:e:; (T y0 ; y0 ) = ky k0
2

) ky k  0 as (T y ; y )  0 as
0
2
0 0 T is positive.

Above is possible only when ky0 k= 0 i:e:; y0 = 0 .

But this will be contradiction as y0 was a non-zero vector in H such that
y0 ?R(I + T ) . Hence our assumption that R(I + T ) is not equal to H is
contradiction.
) R(I + T ) = H .

i:e:; I + T is onto.

Since I + T is both 0ne-to-one and onto, it follows that I + T is non-singular.

We know that T T  and T  are positive operators and as above, we can say
that I + T T  and I + T  T are non-singular. 
Let us Sum Up:
In this unit, the students acquired knowledge to

ˆ Dene adjoint of an operator and prove various properties.

ˆ understand the concept of self-adjoint operator and its properties.

Check Your Progress:

1. If T be an invertible operator than either both or none of T and T 1

are self-adjoint.

2. If T is self adjoint operator on a Hilbert space H then T is self

adjoint if and only if is real.

3. Dene adjoint operator.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
144 7.2. Self-Adjoint Operators

Suggested Readings:

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

Tata Mc-Graw Hill Pvt. Ltd., New Delhi, 2011.

2. Kreyszig E.,” Introductory Functional Analysis with Applications,

John Wiley & Sons, New York, 2007.

3. Limaye B. V.,” Functional Analysis”, New Age International Ltd.,

Publishers, Second Edition, New Delhi,1996.

4. Co man C. and Pedrick G., “First Course in Functional Analysis”,

Prentice-Hall of India, New Delhi, 1995..

5. Conway J.B., “A Course in Functional Analysis”, Springer-Verlag,

New York, 2008.

6. Bollobas B.,”Linear Analysis”, Cambridge University Press, Indian

Edition, New York, 1999.

7. Nair M.T., “Functional Analysis, A First course”, Prentice Hall of

India, New Delhi, 2010.

Web Resources:

// / / / /
1. https: archive.nptel.ac.in courses 111 106 111106147 /
// / /
2. https: web.iitd.ac.in siva lecture
A-adj.pdf

// / =
3. https: www.youtube.com watch?v bjD6w
LT0

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
Block-IV

UNIT-8

OPERATOR ON HILBERT

SPACE-II

Structure
Objective
Overview
8. 1 Normal and Unitary Operators
8. 2 Projections
Let us Sum Up
Check Your Progress
Suggested Readings

Overview
In this unit, we shall study the operators on a Hilbert space with

its properties. Also,we can discuss about the orthogonal projection on a

Hilbert space.

M.Sc.(Mathematics)-II Year-III Sem 145 Functional Analysis

 146 8.1. Normal and Unitary Operators

Objectives
After successful completion of this lesson, students will be able to

ˆ understand the concept of normal opertors and its properties.

ˆ explain the concept of projection on a Hilbert space.

8.1. Normal and Unitary Operators

In this section, we shall dene normal and unitary operators. These

operators are more general operators on H for which a simple and

revealing structure theory is possible. In this section, we present a few

of their more elementary properties, which are necessary for the spectral

theory.

Denition 8.1. An operator N on H is said to be normal if it commutes with its

adjoint, that is NN  = N  N .

Remark 8.1. (i) Every self-adjoint operator is a normal operator but a normal
operator need not be a self-adjoint.

Proof. Assume that T is a self-adjoint operator, then T = T .

Now, T T  = T 2 = T T = T T .
Hence T is normal.
Dene T : H ! H by T = 2iI where I is the identity operator.
Then T  = 2iI  = 2iI
So that T T  = T  T = 4I . Hence T is a normal operator on H . But, T 6= T  .
Hence T is not self-adjoint. 

(ii) If T 2 B(H) is normal then T  is normal.

Proof. Since T is normal, then we have T T  = T  T .
Now, T  T  = T  T  .
Hence T  is normal. 

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
8.1. Normal and Unitary Operators 147

Theorem 8.1. The limit N of any convergent sequence fNk g of normal operators
is normal.

Proof. Given that the sequence of normal operators fNk g converges to N . Now,
we shall prove that N is a normal operator i:e:; NN  = N  N .
kNk Nk = k(Nk N ) k= k(Nk N )k !0

Hence Nk is a normal operator.

kNN  NNk = kNN  Nk Nk + Nk Nk Nk Nk + Nk Nk NNk
 kNN  Nk Nk k+kNk Nk Nk Nk k+kNk Nk N  N k (8.1)

Since Nk is normal
Using the above equation in (8.1), we get
 kNN  Nk Nk k+0 + kNk Nk NNk

Since Nk ! N and Nk ! N  , the right hand side of (8.2) tends to zero. This
implies that kNN  NN  k= 0 .
Hence, we have NN  = NN  and thus, we have N is normal. 
Theorem 8.2. The set of all normal operators on H is a closed subset of B (H )
which contains the set of all self-adjoint operators and is closed under scalar
multiplication.

Proof. Let M be the set of all operators on a Hilbert space H . First we are to
B
prove that M is a closed subset of (H ) . Let N be any limit point of M . Then
to show that N 2 M i:e:; to show that N is a normal operator on H . Since N
is a normal operator on H . Since N is a limit point of M , therefore there exists
a sequence fNk g of disjoint points of M such that Nk ! N .
By above remarks, N is a normal operator and hence N 2 M . So, M is
closed. Now every self-adjoint operator is normal. Therefore the set M contains
the set of all self-adjoiont operators on H .
Finally, we have to show that the set M is closed with respect to scalar
multiplication i:e:; N 2 M ) N 2 M where is any scalar. In other
words, we have to show that if N is a normal operator on H and is any scalar,
then N is normal operator on H . Since N is normal, therefore NN  = N  N .
We have ( N ) = N  .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
148 8.1. Normal and Unitary Operators

( N )( N ) = ( N) N = (NN  )
= ( )(NN  )
= ( )(N  N )
= ( N ) ( N )

Hence N is normal. 
Theorem 8.3. If N1 and N2 are normal operators on a Hilbert space H with
the property that either commutes with the adjoint of the other then N1 + N2 and
N1 N2 are also normal operators.

Proof. Since N1 ; N2 are normal operators, therefore

N1 N1 = N1 N1
and N2 N2 = N2 N2

Also, it is given that

N1 N2 = N2 N1
and N2 N1 = N1 N2

First we shall prove that N1 + N2 is normal. We have

(N1 + N2 )(N1 + N2 ) = (N1 + N2 )(N1 + N2 )
= N1 N1 + N1 N2 + N2 N1 + N2 N2
= N1 N1 + N2 N1 + N1 N2 + N2 N2
= N1 (N1 + N2 ) + N2 (N1 + N2 )
= N1 + N2 (N1 + N2 )

= (N1 + N2 ) (N1 + N2 )

) N1 + N2 is normal.

Again to prove that N1 N2 is normal, we have

(N1 N2 )(N1 N2 ) = N1 N2 N2 N1
= N1 (N2 N2 )N1
= (N1 N2 )(N2 N1 )
= (N2 N1 )(N1 N2 )
= N2 (N1 N1 )N2 )
= (N2 N1 )(N1 N2 )
= (N1 N2 ) (N1 N2 )

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
8.1. Normal and Unitary Operators 149

Therefore, N1 N2 is normal. 
Theorem 8.4. An operator T on a Hilbert space H is normal , kT  xk= kT xk .
for every x .

Proof. We have
T is normal , TT = T T
, TT T T =0
, ((T T  T  T ) x; x) = 0
8 x
, (T T  x; x) = (T  T x; x) 8 x
, (T  x; T  x) = (T x; T  x) 8 x
, (T  x; T  x) = (T x; T x) 8 x
, kT  xk = kT xk 8 x
2 2

, kT  xk= kT xk 8 x

Hence the proof. 

Theorem 8.5. If N is a normal operator on a Hilbert space H then
kN k= kN k
2 2
.

Proof. We know that if T is a normal operator on H , then we have

kT  xk = kT xk 8 x

Taking N in place of T and N x in place of x in above, we get

kNN xk = kN  N xk 8 x
) kN xk = kN  N xk 8
2
x

Now,
n o
N2 = sup kN xk: k xk 1
2

= sup fkN  N xk: k xk 1g

= kN  N k
= kN k 2

This completes the proof. 

Theorem 8.6. Any arbitrary operator T on a Hilbert space H can be uniquely
expressed as T = T + iT
1 2 .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
150 8.1. Normal and Unitary Operators

where T1 and T2 are self-adjoint operators on H.

= T +2 T

Proof. Let T1 and T2 = 21i (T T ) .

Then T = T + iT1 2 Now " # 

1
T
1
= 2
(T + T )

1 
= 2
(T + T  )


1 
= 2
(T + T )
1
= 2
(T + T  ) = T 1

) T 1 is self-adjoint. " #

1
Also T
2
= 2i
(T T  )

1
= 2i
(T T  )


1 
= 2i
(T T  )
1 
= 2i
(T T )
1
= 2i
(T T  ) = T 2

) T 2 is self-adjoint.

Thus, T can be expressed in the form T = T 1 + iT 2 , where T 1 and T 2 are

both self-adjoint operators. Now, we shall prove that the expression for T is
unique.
Let T = U + iU
1 2 where U1 and U2 are both and adjoint. We have
T = (U1 + iU2 )
= U1 + (iU2 )
= U1 iU2
= U1 iU2
) T + T = (U1 + iU2 ) + (U1 iU2 ) = 2U1
1
) U1 = 2
(T + T  )
Also T T = (U1 + iU2 ) (U1 iU2 ) = 2iU2
1
) U2 = 2i
(T T  ) = 2iU2
1
) U2 = 2i
(T T  ) = T 2

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
8.1. Normal and Unitary Operators 151

Hence the expression for T is unqiue. 

Note 8.1. The above result is analogous to the result on complex number that
every complex number z can be uniquely written in the form z = a + ib where a
and b are real numbers. In the above resolution T = T 1 + iT 2 , T 1 is called the
real part of T and T 2 is called the impaginary part of T .

Theorem 8.7. If T is an operator on a Hilbert space H , then T is normal ,

its real and imaginary parts commute.

Proof. Let T 1 and T 2 be the real and imaginary parts of T . Then T 1 and T 2
are self-adjoint operators and T = T 1 + iT 2 . Now, we have
T = (T 1 + iT 2 )
= T 1 + (iT 2 )
= T 1 + iT 2
= T 1 iT 2
= T1 iT 2
Now T T  = (T 1 + iT 2 )(T 1 iT 2 )
= T 12 + T 22 + i (T 2 T 1 T1T2) (8.2)
and T  T = (T 1 iT 2 )(T 1 + iT 2 )
= T 12 + T 22 + i (T 1 T 2 T2T1) (8.3)

Suppose T is normal i:e: T T  = T  T . Then from (8.2) and (8.3), we have

T 12 + T 22 + i (T 2 T 1 T1T2) = T 12 + T 22 + i (T 1 T 2 T2T1)
) T2T1 T1T2 = T1T2 T2T1
) 2T 2 T1 = 2T 1 T 2
) T 2 T1 = T1T2

Hence T 1 and T 2 commute.

Conversely suppose that T 1 and T 2 commute.
i:e:; T 2 T 1 =TT 1 2 . Then from (8.2) and (8.3) we see that T T  = T T and
hence T is normal. 
Denition 8.2. An operator U on a Hilbert space H is said to be unitary if
UU  = U U = I

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
152 8.1. Normal and Unitary Operators

Note 8.2. Clearly every unitary operator is normal. Also, we observe that an
operator U is unitary if and only if it is invertible and its inverse is precisely
equal to its adjoint.

Theorem 8.8. If T is an operator on a Hilbert space H , then the following

conditions are all equivalent to one another:

1. T  T =I;
2. (T x; T y) = ( x; y) for all x and y ;

3. kT xk= k xk for all x .

Proof. (1) ) (2) : It is given that T  T = I . We have

(T x; T x) = ( x; T  T y)
= ( x; Iy)
= ( x; y) for all x and y

(2) ) (3) : It is given that (T x; T y) = ( x; y) for all x and y .

Taking y = x , we get
= ( x; x)
(T x; T x)
) kT xk = k xk
2 2

) kT xk = k xk for all x:

(3) ) (1) : It is given that kT xk= k xk for all x . Therefore, we have

kT xk = k xk
) kT xk = k xk
2 2

) (T x; T x) = ( x; x)
) (T  T x; x) = ( x; x)
) ((T  T I ) x; x) = 0 for all x
) T T I = 0
) T T = I

This completes the proof of the theorem. 

Theorem 8.9. An operator T on a Hilbert space H is unitary , it is an
isometric isomorphism of H onto itself.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
 8.2. Projections 153

Proof. Suppose that T is a unitary operator on H . Then T is invertible and

therefore T is onto. Also T  T = I . Therefore by Theorem 8.8, we have
kT xk = k xk 8 x 2 H

Thus T preserves norms and so T is an isometric isomorphism of H onto itself.

Conversely, Suppose that T is an isometric isomorphism of H onto H itself.
Then T is one-to-one and onto. Therefore T 1 exists. Also T is an isometric
isomorphism.
) kT xk = k xk 8 x
) T  T = I (by theorem 8.8)
) (T  T )T 1
= IT 1

) T  (T T ) 1
= T 1

) T I = T 1

) T = T 1

) TT = I = T T

Hence T is unitary. 

8.2. Projections

In the chapter of Banach spaces we have dened a projection on a

Banach space is an idempotent operator on B , that is, if P is a continuous

linear transformation from B into B with the property that P2 = P . If M

and N are the range and the null space of a projection P on B , then
M = fPx : x 2 Bg
and N = f x 2 B : Px = 0g

We have shown that both M and N are closed linear subspaces of B such

that B = M  N. Also, we have shown that x 2 M , Px = x i:e:

the range M of P consists of exactly those vectors in B which are left

unchanged by P. Further we have seen that if z 2B and z = x+y where

x 2 M; y 2 N , then Pz = x. Conversely, we have shown that if ( M; N ) is

any pair of closed linear subspace of B such that B = M  N , then there

exists a projection P on B whose range is M and whose null space is

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
154 8.2. Projections

N. If z 2 B is written as z = x+y where x is in M and y is in N, then

this projection P is dened by P(z) = P( x + y) = x . In this way we have

established a one-to-one correspondence between the projecton on B and

pairs of closed linear subspaces of B which span the whole space and have

only the zero vector in common.

Now, if our space is a Hilbert space H in place of Banach space B,

then we can single out those projecton on H whose range and null spaces

are orthogonal. Such projections are called perpendicular projections on

H and they are of special importance in the theory of Hilbert spaces.

Theorem 8.10. If P is a projection on H with range M and null space N , then

M ?N , P is a self-adjoint; and in this case, N = M ? .

Proof. Assume that P is a projection on a Hilbert space H with range and null
space M and N respectively. Then H = M  N .
Firse we shall show that if M ?N , then P is self-adjoint. Let z be any
arbitrary vector in H . Then z can be uniquely written as z = x + y where
x 2 M and y 2 N . We have
(Pz; z) = ( x; z)
= ( x; x + y)
= ( x; x) + ( x; y)
= ( x; x)

Also, we have
(z; P z) = ( x + y; x)
= ( x; x) + (y; x)
= ( x; x)

Thus, we have
(Pz; z) = (P z; z) 8z 2 H
) ((P P )z; z) = 0 8z 2 H
) P P = 0
) P = P

Hence P is self-adjoint.
Conversely, Assume that P is self-adjoint. Now, we shall prove that M ?N .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
8.2. Projections 155

Let x be any vector in M and y be any vector in N . Then

( x; y) = (Px; y)
= ( x; P y)
= ( x; Py)
= ( x; 0)
= 0

Hence M ?N .
Now, we shall prove that the last part of the theorem. i:e: we shall prove that
if P is a projection on H with range M and null space N then M ?N .
Suppose x 2 N . Then
N?M ) x? M ) x 2 M?

Therefore N  M ? .
Suppose N is a propert subset of M ? .Then N is a closed linear subspace of
the Hilbert space M ? . Therefore by theorem there exists a non-zero vector z0 in
M ? suc that z0 ?N . But z0 2 M ? ) zo ? M . There z0 is such that z0 ? M and
z0 ?N . But H = M  N . Therefore z0 ?H and so we must have z0 = 0 , which
is a contradication to the fact that it is a non-zero vector.
Therefore, N cannot be a proper subset of M ? and hence we must have
N = M? . 
Note 8.3.

1. As we already proved that 0 and I are self-adjoint operators. Since

02 = 0 and I 2 = I , they are projections on H .

2. If P is a projection on a Hilbert space H , then its range M and the null

space N are closed linear subspaces of H and N = M ? . Further by the
orthogonal decomposition theorem, we have H = M  M ? . Therefore to
each projection P , there corresponds a closed linear subspace which is its
range M = fPx : x 2 H g and conversely to each closed subspace M , there
corresponds the projection P with range dened by P( x + y) = Px where
x 2 M and y 2 M ? . Hence we can nd a one-to-one correspondence
between the projection on H and a closed subspace of H .

3. If P is a projection on a Hilbert space H with range M , the null space

N is uniquely determined and it is M ? . Thus there is only one projection

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
156 8.2. Projections

with range M . So instead of saying that P is aprojection having the range

M , we say that P is a projection on M .

Theorem 8.11.
P is the projection on a closed linear subspace M of H , I P is the
projection on M? .

Proof. Let P be a projection on H , then P2 = P and P = P . Now,

(I P) = I P
= I P
Also (I P)2 = (I P)(I P)
= I2 P P + P2
= I P P+P
= I P

Hence I P is a projection on H .

Let, if M be the range of P , then we shall show that range of I P is M ? .

Let N be the range of I E then

x2N ) (I P) x = x
) x Px = x
) Px = 0
) x 2 Null space of P
) x 2 M?

) N  M? .
Again, let x 2 M ? which is null space of P so that
Px = 0
) x Px x =
) (I P) x x =
) x 2 Range of I P
) x 2 N

) M? N.
Hence N = M ? . i:e:; range of I P is M ? .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
8.2. Projections 157

Conversely, Assume that I P is a projection on M ? , then by just proved

above we have I (I P) is a projection on (M ? )? i:e:; P is a projection on
M ?? .

But, we know M be a closed linear subspace of H , M = M ?? .

Thus, we have P is a projection on M . 
Theorem 8.12. If P is the projection on the closed linear subspace of H , then
x2M ) Px = x , kPxk= k xk

Proof. Given that P is a projecton on a Hilbert space with range M . Now, we

shall prove that x 2 M , Px = x .
Suppose Px = x . Then x is in the range of P because Px is in the range of
P . Therefore Px = x x 2 M .

Conversely suppose that x 2 M . Then to show that Px = P. Let Px = y.

Then we must show that y = x . We have
Px = y
) P(Px) = Py
) Px 2
= Py
) Px = Py
) P( x y) = 0
) x y is in the null space of P
) x y 2 M?
) x y = z where z 2 M ?
) x = y+z

Now y = PX ) y is in the range of P i:e:; y is in M .

Thus we have expressed x = y + z where y 2 M; z 2 M ? . But x is in M . So
we can write x = x + 0 where x 2 M and 0 2 M ? . But H = M  M ? . There
we must have y = x; z = 0 .
Next we shall prove that Px = x , kPxk= k xk .
If Px = x , then obviously kPxk= k xk . Conversely suppose that kPxk= k xk .
Then to show that Px = x .
Now, we have
k xk2
= kPx + (I P) xk2 (8.4)

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
158 8.2. Projections

Now Px is in M . Also P is the projection on M which implies that I P is

the projection on M ? . Therefore we have (I P) x 2 M ? . Then by Pythagoream
theorem, we have
kPx + (I P) xk2 = kPxk +k(I
2
P) xk2 (8.5)

From (8.4) and (8.5), we get

k xk2
= kPxk +k(I
2
P) xk2
) k(I P) xk2
= 0
) (I P) x = 0
) x Px = 0
) Px = x

Hence the theorem. 

Theorem 8.13. If P is a projection on a Hilbert space H , then

(i) P is a positive operator i:e: P  0 .

(ii) 0  P  I .

(iii) kPxk k xk 8 x 2 H .
(iv) kPk 1 .

Proof. Given that P is a projection on a Hilbert space H .

Then P2 = P; P = P . Let M be the range of P .
(i) Let x be any arbitaray vector in H . Then
(Px; x) = (PPx; x)
= (Px; P x)
= (Px; Px)
= kPxk  0:
2

Thus (Px; x)  0 8 x 2 H . Therefore P is a positive operatore i:e: P  0 .

(ii) Since P is a projection on H , therefore I P is also a projection on H .
Then by (i) , we have I P  0; i:e:; P  I . But P  0 .
Thus, we have 0  P  I .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
8.2. Projections 159

(iii) Let x be any vector in H . If M is the range of P , then M ? is the range of

(I P) . Now Px is in M and (I P) x is in M ? . Therefore Px and (I P) x
are orthogonal vectors. So applying Pythagoream theorem, we have
kPx + (I P) xk2
= kPxk +k(I
2
P) xk2
) kI ( x)k
2
= kPxk +k(I
2
P) xk2
) k xk 2
= kPxk +k(I
2
P) xk2
) k xk2
 kPxk2

) kPxk  k xk:

(iv) we have
kPk = supfkPxk: k xk 1g
 1 (* by (iii))

This proves the theorem. 

Note 8.4. If P is a projection on a Hilbert space H and x is any vector in H ,
then (Px; x) = kPxk2 .

Denition 8.3. A closed subspace M of a Hilbert space H is said to be invariant

under an operator T if T ( M )  M .

Note 8.5. If a closed subspace M of a Hilbert space H is invariant under an

operator T , then it induces an operator T M on M such that T M ( x) = T ( x) for
every x 2 M . This operator T M is called the restriction of T on M .

Denition 8.4. If M is a closed linear subspace of Hilbert space H such that

both M and M ? are invariant under an operator T on H , then T is said to be
reduced by M and also we say that M reduces T .

Theorem 8.14. A closed linear subspace M of a Hilbert space H is invariant

under the operator T if and only if M? is invariant under T  .

Proof. Assume that M is invariant under T . Then to prove that M ? is invariant

under T  . Let y be any arbitrary vector in M ? . Then to show that T  y is also
in M ? . i:e: T  y is orthogonal to every vector in M .
Let x be any vector in M . Then T x 2 M because M is invariant under T .
Also y 2 M ? ) y is orthogonal to every vector in M .
Therefore y is orthogonal to T x i:e: (T x; x) = 0 ) ( x; T  y) = 0 .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
160 8.2. Projections

Thus, T  y is orthogonal to every vector x in M . Therefore T  y is in M ?

and so M ? is invariant under T  .
Conversely, Assume that M ? is invariant under T  . Now, we shall prove that
M is invariant under T . Since M ? is a closed linear subspace of H invariant
under T  , therefore by above (M ? )? is invariant under (T  ) . But
( M ? )? = M ?? =M
and (T  ) = T  = T

Therefore M is invariant under T . 

Theorem 8.15. A closed linear subspace M of a Hilbert space H reduces an
operator T , M is invariant under both T and T  .

Proof. Let us assume that M reduces T . M reduces T means that both M

and M ? are invariant under T . By Theorem 8.14, if M ? is invariant under T ,
then M ?? = M is invariant under T  . Hence M is invariant under both T and
T  , if M reduces T .

Converse Part: Let us assume that M is invariant under both T and T  . Since
M is invariant under T  , M ? is invariant under (T  ) = T . Thus M and M ?
are both invariant under T . Hence M reduces T . 
Theorem 8.16. If P is the projection on a closed linear subspace M of a Hilbert
space H , then M is invariant under an operator T if and only if T P = PT P .

Proof. Let M be invariant under T . Now, we shall prove that T P = PT P .

Let x be any arbitrary vector in H . If x 2 H , then Px 2 M . Since M is
invariant under T , T Px 2 M . Since P is projection and M is the range of P ,
T Px 2 M implies that P maps T Px into itself. Hence PT Px = T Px for every
x 2 H . Hence PT P = T P .

Conversely, assume that PT P = T P and prove M is invariant under T . Let

x be any vector in M . Since P is projection with range M and x 2 M , we have
Px = x which gives T Px = T x . Using the given condition that PT Px = T x ,
we have PT Px = T Px . Since P maps elements of M into the same element
P(T Px) = T Px means T Px 2 M . Since PX = x , we get T x 2 M . Hence if
x 2 M implies T x 2 M . Therefore T is invariant under T . 
Theorem 8.17. If P is the projection on a closed linear subspace M of a Hilbert
space H , then M reduces an operator T , T P = PT .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
8.2. Projections 161

Proof. Assume that M reduces T .

, M is invariant under both T and T 
, TP = PT P and T  P = PT  P
, T P = PT P and (T  P) = (PT  P)
, T P = PT P and P T  = P T  P
, T P = PT P and PT = PT P

Thus M reduces T .
, T P = PT P and PT = PT P .
Now, suppose M reduces T . Then we have T P = PT P and PT = PT P .
This gives T P = PT .
Conversely, suppose that T P = PT . Multiplying both sides on the left by P ,
this gives PT P = P2 T = PT .
Similarly, multiplying both sides of T P = PT on the right by P , we get
T P2 = PT P i:e:; T P = PT P .

Thus, we have T P = PT ) T P = PT P and PT = PT P .

Thus from the above part, we conclude that M reduces T . 
Theorem 8.18. If M and N are closed linear subspaces of a Hilbert space
H and P and Q are the projections of M and N respectively, then
M ?N , PQ = 0 , QP = 0 .

Proof.
Since P and Q are projections on a Hilbert space H ,
therefore P = P , Q = Q . First we observe that
PQ = 0 , (PQ) = (0) , Q P = 0 , QP = 0

Therefore to prove the theorem it is enough to prove that

M ?N , PQ = 0

Assume that M ?N . If y is any vector in N , then M ?N , y is orthogonal to

every vector in M . So y 2 M ? .
Thus, we have N  M ? .
Now, let z be any vector in H . Then QZ is in the range of Q i:e:Qz 2 N .
Consequently Qz 2 M ? which is the null space of P .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
162 8.2. Projections

Therefore P(Qz) = 0 . Thus PQz = 0 8 z 2 H .

Hence PQ = 0 .
Conversely Assume that PQ = 0 and x 2 M and y 2 N . Since M is the
range of P , therefore Px = x . Also N is the range of Q . Therefore Qy = y .
we have
( x; y) = (Px; Qy)
= ( x; P Qy)
= ( x; PQy)
= ( x; Oy)
= ( x; 0) = 0

Therefore M and N are orthogonal. i:e: M ?N . 

Denition 8.5. Two projections P and Q on a Hilbert space H are said to
orthogonal if PQ = 0 .

Denition 8.6. By Theorem 8.18, two projections P and Q are orthogonal if

and only if their ranges M and N are orthogonal.
Theorem 8.19. If P1 ; P2 ; : : : ; Pn are the projections on closed linear subspaces
M1 ; M2 ; : : : ; Mn of a Hilbert space H , then P = P1 + P2 + : : : + Pn , the P0i s
are pairwise orthogonal (in the sense that Pi P j = 0 whenever i 6= j ). Also, then
P is the projection on M = M1 + M2 + : : : + Mn .

Proof. Given that P1 ; P2 ; : : : ; Pn are projections on H . Therefore P2i = Pi = Pi

for each i = 1; 2; : : : ; n .
Let P = P + P + : : : + Pn
1 2

Then P = (P + P + : : : + Pn )
1 2

= P + P + : : : + Pn
1 2

= P + P + : : : + Pn
1 2

First we shall prove the sucient condition. Assume that Pi P j =0 whenever

i 6= j . Then to prove that P is a projection on H . We have
P2 = PP
= (P + P + : : : + Pn )(P + P + : : : + Pn )
1 2 1 2

= P + P + : : : + Pn [* Pi P j = 0 if i 6= j]
2
1
2
2
2

= P + P + : : : + Pn
1 2

=P

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
8.2. Projections 163

Thus P2 = P = P . Therefore P is a projection on H .

To prove the necessity of the condition. Assume that P is a projection on H .
Then P2 = P = P . We are to prove that Pi P j = 0 if i 6= j . We rst notice that
if T is any projection on H and s is any vector in H , then
(T z; z) = (T T z; z) = (T z; T  z) = (T z; T z) = kT zk2

Now let x belong to the range of some Pi so that Pi x = x . Then

k xk 2
= kPi xk 2

n
X
 k P j xk 2

j= 1
= kP1 xk2 +kP2 xk2 + : : : + kPn xk2
X n
= (Pi x; x)
j= 1
= (P1 x; x) + (P2 x; x) + : : : + (Pn x; x)
= ([P + P + : : : + Pn ] x; x)
1 2

= (Px; x)
= kPxk 2

 k xk 2

From the above, we can conclude that the sign of equality must hold. Therefore
we have
n
X
kPi xk 2
= k P j xk
2

j= 1
) k P j xk 2
= 0 if j 6= i
) k P j xk = 0 if j 6= i
) Pjx = 0 if j 6= i
) x is the null space of P j if j 6= i
) x 2 M ?j if j 6= i

Thus x is orthogonal to the range M j of every P j with j 6= i .

Thus every vector x in the range Pi (i = 1; 2; : : : ; n) is orthogonal to the range
of every P j with j 6= i . Therefore the range of Pi is orthogonal to the range of
every P j with j 6= i . Hence by Theorem 8.19, we have
Pi P j = 0 whenever i 6= j

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
164 8.2. Projections

In order to prove that P is the projection on

M = M1 + M2 + : : : + Mn

It is enough to prove that R(P) = M where R(P) is the range of P .

Let x 2 M . Then x = x1 + x2 + : : : + xn where xi 2 Mi ; (1  i  n) .
But, we have k xk2 = kPxk2 if x is in the range of some Pi . Therefore
xi 2 Mi i:e: the range of P .
) kPxi k 2
= k xi k2

) kPxi k = k xi k
) xi 2 the range of P
) xi 2 R(P) for each i = 1; 2; : : : ; n
) x1 + x2 + : : : + xn 2 R(P)
) x 2 R(P)

Hence x 2 M ) x 2 R(P) . Therefore, we have M  R(P) .

Let us assume that x 2 R(P) . Then
Px = x
) (P + P + : : : + Pn ) x
1 2 = x
) P x + P x + : : : + Pn x
1 2 = x

But Pi x 2 Mi for each i = 1; 2; : : : ; n .

Therefore x 2 M1 + M2 + : : : + Mn and So R(P)  M .
Hence, we have R(P) = M .
This completes the proof of the theorem. 
Let us Sum Up:
In this unit, the students acquired knowledge to

ˆ Dene normal operator and prove various properties.

ˆ understand the concept of perpendicular projections.

Check Your Progress:

1. If T is an arbitrary operator on a Hilbert space H, and if and

are scalars such that j j= j j , then T + T  is normal.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
8.2. Projections 165

2. If T is a normal operator on a Hilbert space H and  is any scalar,

then prove that T I is also normal.

Suggested Readings:

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

Tata Mc-Graw Hill Pvt. Ltd., New Delhi, 2011.

2. Kreyszig E.,” Introductory Functional Analysis with Applications,

John Wiley & Sons, New York, 2007.

3. Limaye B. V.,” Functional Analysis”, New Age International Ltd.,

Publishers, Second Edition, New Delhi,1996.

4. Co man C. and Pedrick G., “First Course in Functional Analysis”,

Prentice-Hall of India, New Delhi, 1995..

5. Conway J.B., “A Course in Functional Analysis”, Springer-Verlag,

New York, 2008.

6. Bollobas B.,”Linear Analysis”, Cambridge University Press, Indian

Edition, New York, 1999.

7. Nair M.T., “Functional Analysis, A First course”, Prentice Hall of

India, New Delhi, 2010.

Web Resources:

// / / / /
1. https: www.impan.pl pmh teach algebra additional normal.pdf /
// /
2. https: www.youtube.com watch?v s-5bCfENHg8 =

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
166
8.2. Projections 167

Block-V
Unit-9: Finite Dimensional Spectral Theory

Unit-10: Spectral Theorem and Fixed Point Theorem

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
168
Block-V

UNIT-9

FINITE DIMENSIONAL

SPECTRAL THEORY

Structure
Objective
Overview
9. 1 Eigen Values and Eigen Vectors
9. 2 Matrix of Linear Transformation
9. 3 Determinants and Spectrum of an Operator
Let us Sum Up
Check Your Progress
Suggested Readings

Overview
In this unit, we shall study that the every operator on a Hilbert

space has an eigen value. Also, we discuss the matrices of linear

transformation.

M.Sc.(Mathematics)-II Year-III Sem 169 Functional Analysis

 170 9.1. Eigen Values and Eigen Vectors

Objectives
After successful completion of this lesson, students will be able to

ˆ understand the concept of eigen values and eigen vectors.

ˆ explain the concept of spectrum of an operator.

9.1. Eigen Values and Eigen Vectors

Denition 9.1. Let T be an operator on a Hilbert space H . Then a scalar 

is called an eigenvalue of T if there exists a non-zero vector x in H such that
T x =  x . Also, if  is an eigenvalue of T , then any non-zero vector x in H
such that T x =  x is called an eigenvector of T belonging to the eigenvalue 

Note 9.1. Eigenvalues are sometimes are called characteristic values, proper
values, or spectral values. Similarly eigenvectors are called characteristic
vectors, proper vectors, spectral vectors.

The set of all eigen values of T is called the spectrum of T and we shall
denote it by (T ) .
If the Hilbert space H has no non-zero vectors at all, then T certainly has
no eigen vectors. In this case the whole theory collapses into trivially. Therefore
throughout the present chapter and next chapter we shall assume that H 6= f0g .

Theorem 9.1. If x is an eigenvector of T corresponding to the eigenvalue  ,

then x is also an eigenvector of T corresponding to the same eigenvalue  .
Here is any non-zero scalar.

Proof. Since x is an eigenvector of T corresponding to the eigenvalue  ,

therefore x 6= 0 and T x =  x .
If is any non-zero scalar, then x 6= 0 ) x 6= 0 . Further,
T ( x) = T x = ( x) = ( ) x = ( x)

Thus, x is an eigenvector of T corresponding to the eigen value  . 

Note 9.2. Theus the corresponding to an eigenvalue  there may correspond
more than one eigenvectors.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
9.1. Eigen Values and Eigen Vectors 171

Theorem 9.2. If x is an eigenvector of T , then x cannot correspond to more

than one eigenvalues of T .

Proof. Let x be an eigenvector of T corresponding to two distinct eigenvalues

1 and 2 of T . Then
Tx =  1 x
Tx =  2 x

Therefore, we have 1 x = 2 x which implies that (1  )x = 0.

2

i:e:; 1 = 2 . 
Theorem 9.3. Let  be an eigenvalue of an operator T on a Hilbert space
H . If M is the set consisting of all eigenvectors of T which correspond to
the eigenvalue  together with the vector 0 , then M is a non-zero closed
linear subspace of H invariant under T . M is called the eigenspace of T
corresponding to the eigenvalue  .

Proof. By denition of an eigenvector, it is a non-zero vector, therefore the set

M necessarily contains some non-zero vectors. Now by denition of M , a
non-zero vector x is in  if and only if T x =  x . Also, the vector 0 is in M .
Thus, the vector 0 denitely satisfy the equation T x =  x . Therefore we have
M = f x 2 H : T x =  xg
i:e:; M = f x 2 H : (T ) x = 0g

Thus M is the null space of the linear transformation T I on H . Therefore

M is a linear subspace of H . The linear transformation T I is denitely a
continuous mapping because both T and I are continuous. We know that the
null space of a continuous linear transformation is closed. Thus, M is closed.
Now, it remains to prove that M is invariant under T . Let x 2 M , then
T x =  x . Since M is a linear subspace of H and hence x 2 M which implies
that  x 2 M . Thus x 2 M ) T x 2 M . Therefore M is invariant under
T . Thus we have proved that each eigenspace of T is a non-zero closed linear
subspace of H invariant under T . 

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
 172 9.2. Matrix of linear transformation

9.2. Matrix of linear transformation

Denition 9.2.
Let H be an n -dimensional Hilbert space and let B = fe1 ; e2 ; : : : ; en g be
an ordered basis for H . Let T be an operator on H . Since each T (e j )) 2 H
and B is a basis for H , therefore each T (e j ) is uniquely expressible as a linear
combination of the vectors e1 ; e2 ; : : : ; en . Let for j = 1; 2; : : : ; n ,
T (e j ) = 1 j e1 + 2 j e2 + : : : + n j en
n
X
= i j ei
i= 1

The n  n matrix whose jth column ( j = 1; 2; : : : ; n) consists of the scalars

1 j; 2 j; : : : ; n j is called the matrix of the operator T relative to the ordered
basis B.

We shall denote it by [T ]B or simpy [T ] if the basis B is understood. Thus,

we have
[T ] = [T ]B = matrix of T relative to the ordered basis B
= [ i j ]nn
n
X
where T (e j ) = i j e j for each j = 1; 2; : : : ; n
i= 1

From the above, we can clearly understood that to write the jth ( j = 1; 2; : : : ; n)
column of the matrix [T ]B , we express as a linear combination of the vectors
e1 ; e2 ; : : : ; en and the resulting scalar coecients gives us the jth column of the
matrix [T ]B .

Theorem 9.4. Let H be an n -dimensional Hilbert space and let B be any

ordered basis for H . If I be the identity operator and 0 be the zero operator on
H , then

(i) [I ]B = [i j ]nn ( i:e: unit matrix of ordedr n ).

(ii) [0]nn = 0 ( i:e: null matrix of order n  n )

Proof. Let B = fe1 ; e2 ; : : : ; en g

(i) we have

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
9.2. Matrix of linear transformation 173

I (e j ) = e j; j = 1; 2; : : : ; n
= 0e1 + 0e2 + : : : + 0en
n
X
= i j e j ; where i j is a Kronecker delta:
i= 1

Thus, by denition, we have [T ]B = [i j ]nn = I i:e: unit matrix of order n .

(ii) we have
0(e j ) = 0e1 + 0e2 + : : : + 0en
n
X
= 0i j ei where 0i j = 0
i= 1

Therefore, by denition, we have [0]B = [0i j ]nn = i:e: null matrix of order
nn. 

Next, our aim is to prove that there is an isomorphism between the

algebra B(H) and the total matrix algebra An . In order to establish this,

we shall prove the folowing theorem which will be use to prove our desired

theorem.

Theorem 9.5. Let H be a nite dimensional Hilbert space of dimension n and

let B = e ; e ; : : : ; en
1 2be an ordered basis for H , if f1 ; f2 ; : : : ; fn are any n
vectors in H , then there exists a unique operator in H such that
T (ei ) = fi ; i = 1; 2; : : : ; n

Proof. Let x 2 H . Since B = e1 ; e2 ; : : : ; en is a basis for H , therefore must

exist unqiue scalars 1 ; 2 ; 3 ; : : : ; n such that
x = 1 e1 + 2 e2 + : : : + n en

For this vector x , let us dene

T ( x) = 1 f1 + 2 f2 + : : : + n fn

Clearly T ( x) is dened above is a unique element of H . Therefore T is well

dened rule for associating with each vector x in H is a unique vector T ( x) in
H . Thus, T is a mapping from H into H . The unique representation of ei 2 H
as a linear combination of the vectors belonging to the basis B is
ei = 0e1 + 0e2 + : : : + 1ei + : : : + 0en

By denition of T , we have

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
174 9.2. Matrix of linear transformation

T (ei ) = 0 f1 + 0 f2 + : : : + 1 fn + : : : + 0 fn
T (ei ) = fi i = 1; 2; : : : ; n

Now, we shall prove that T is a linear transformation. Let ; be any scalars

and let x; y 2 H . Let
x = 1 e1 + 2 e2 + : : : + n en
and y = 1 e1 + 2 e2 + : : : + n en

Then,
T ( x + y) = T [ ( 1 e1 + 2 e2 + : : : + n en ) + ( 1 e1 + 2 e2 + : : : + n en )]
= T (

1 + 1 )e1 + ( 2+ 2 )e2 + : : : + ( n+ n )en


= (( + )f ) + (
1 1 1 2 + 2 ) f2 + : : : + ( n+ n ) fn
= ( f + f + ::: +
1 1 2 2 n fn ) + ( 1 f1 + 2 f2 + : : : + n fn )
= T ( x) + T (y)

Therefore, T is a linear transformation from H into H . Thus, there exists an

operator T on H such that
T (ei ) = fi ; i = 1; 2; : : : ; n

It remains to prove that the uniquess of T . For this, let T 0 be another linear
operator on H such that
T 0 (ei ) = fi i = 1; 2; : : : ; n

For the vector x = 1 e1 + 2 e2 + : : : + n en , we have

T 0 ( x) = T 0 ( 1 e1 + 2 e2 + : : : + n en )
= T 0 (e1 ) + T 0 (e2 ) + : : : + T 0 (en )
= f1 + 2 f2 + : : : + n fn
= T ( x)

Thus, T 0 ( x) = T ( x) 8 x 2 H . Therefore, we have T 0 = T .

This proves the uniqueness of T and completes the proof of the theorem. 

Theorem 9.6. If B is an ordered basis for a nite dimensional Hilbert space H

of dimension n , then the mapping T ! [T ] , which assigns to each operator T
its matrix relative to B , is an isomorphism of the algebra B(H) onto the total
matrix algebra An .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
9.2. Matrix of linear transformation 175

Proof. Let B = fe1 ; e2 ; : : : ; en g . We know that (H ) is the set of all operators B

on H and An is the set of all n  n matrices over the eld of complex numbers.
In an algebra, there are three compositions namely addition, scalar
multiplication and multiplication. Both (H ) and An are algebras. B
Let : B(H) ! An such that
[T ] = [T ]B

Now, our aim is to prove that is an isomorphism. i:e: to prove that one-to-one,
onto and preserves all the three compositions.
Let T 1 ; T 2 2 B(H) and let [T ] = n[ 1 i j ]nn and [T 2 ] = [ i j ]nn . Then
X
T 1 (e j ) = i j ei ; j = 1; 2; : : : ; n (9.1)
i= 1
Xn
and T 2 (e j ) = i j ei ; j = 1; 2; : : : ; n (9.2)
i= 1

is one-to-one:
(T 1 ) = (T 2 )
) [T ] 1 = [T 2 ]
h i h i
) i j n n = i j n n
) ij = i j (i = 1; 2; : : : ; n); ( j = 1; 2; : : : ; n)
n
X Xn
) i j ei = i j ei j = 1; 2; : : : ; n
i= 1 i= 1
) T 1 (e j ) = T 2 (e j ) j = 1; 2; : : : ; n
) T1 = T2

) is one-to-one.
is onto: Let [ i j ]nn be any matrix in An .
For each j = 1; 2; : : : ; n , we have
n
X
i j ei is a vectors belonging to a basis for H .
i= 1
Hence, by Theorem 9.5, there exists a unique operator T on H such that
n
X
T (ei ) = i j ei
i= 1

By denition, we have
[T ] = [ i j ]nn ) (T ) = [ i j ]nn

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
176 9.2. Matrix of linear transformation

Thus, is onto.
preserves addition:
(T 1 + T 2 )(e j ) = T 2 (e j ) + T 2 (e j ); j = 1; 2; : : : ; n
Xn Xn
= i j ei + i j ei
i= 1 i= 1
Xn
= ( i j + i j )ei
i= 1
) [T + T ] =
1 2 [ i j + i j ]nn
= [ i j ]nn + [ i j ]nn
= [T 1 ] + [T 2 ]:

Now, (T 1 + T 2 ) = [T 1 ] + [T 2 ] = (T 1) + (T 2 ) . Therefore, preserves addition.

preserves scalar multiplications: Let be any scalar. We have
( T 1 )(e j ) = T 1 (e j ); j = 1; 2; : : : ; n
X n
= i j ei
i= 1
Xn
= ( i j )ei
i= 1

By denition, we have
[ T1] = [ i j ]nn = [ i j ]nn = [T 1 ]

Now, ( T 1 ) = [T 1 ] = (T 1 ) . Thus, preserves scalar multiplication.

preserves multiplicaiton: We have
(T 1 T 2 )(e j ) = T 1 (T 2 e j ); j = 1; 2; : : : ; n
0 1
BX n C
= B
B
T 1 BB@
C
C
k j ek CCA
k =1
0 1
X n BX n C
= B
B
k j BB@
C
C
ik ei CCA
k =1 i= 1
X n 0BX n 1
C
= B
B
B
B
@
C
C
ik k j ei CCA
k = 1 i= 1

By denition of matrix of an operator, we

2
have 3
6Xn 7
[T 1 T 2 ] = 6
6
6
6
4 ik
7
7
k j 775
i= 1 n n
= [ i j ]nn [ i j ]nn
= [T 1 ][T 2 ]

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
9.2. Matrix of linear transformation 177

Now (T 1 T 2 ) = [T 1 T 2 ] = [T 1 ][T 2 ] = (T 1 ) (T 2 ) .
Thus, preserves multiplications and hence is an isomorphism of the
B
algebra (H ) onto the algebra An .
Hence the theorem. 
Theorem 9.7. Let B be a basis for H , and T an operator whose matrix relative
to B is [ i j ] . Then T is non-singular , [ i j ] is non-singular, and in this case
[ i j ] 1 = [T 1 ] .

Proof.

T is non-singular , there exists an operator T 1

on H such that
T 1T = I = TT 1

, [T 1
][T ] = I = [T T ] 1

, [T 1
][T ] = [i j ] = [T ][T ] 1

, [T 1
][ i j ] = [i j ] = [ i j ][T ] 1

Thus, the matrix [ i j ] is non-singular and [ i j ] 1

= [T 1
].
Hence the theorem. 
Denition 9.3. Let A and B be operators on a Hilbert space H . Then B is
said to be similar to A , if there exists a non-singular operator C on H such that
B = C 1 AC .

Note 9.3. We can easily observe that the relation of similarity on the set B (H )
is an equivalance relation.

Theorem 9.8. Two matrices in An are similar , they are the matrices of a
single operator on H relative to (possibly) di erent bases.

Proof. Now, we shall show that if T is an operator on an n -dimensional Hilbert

space H and if B and B0 are two ordered bases for H , then the matrix of T
relative to B is similar to the matrix of T relative to B0 .
Let B = fe1 ; e2 ; : : : ; en g and B0 = f f ; f ; : : : ; fn g .
1 2 Let [T ]B =[ i j ]nn and
[T ]B = [ i j ]nn . Then, we have
0

n
X
T (e j ) = i j ei ; j = 1; 2; : : : ; n (9.3)
i= 1
Xn
and T ( f j ) = i j fi ; j = 1; 2; : : : ; n (9.4)
i= 1

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
178 9.2. Matrix of linear transformation

Let S be the operator on H dened by

n
X
S (e j ) = i j ei ; j = 1; 2; : : : ; n (9.5)
i= 1

Now, we have
T ( f j) = T (S e j )
0 1
BX n C
= T B@
B
B
B
C
C
k j ek CCA
k =1
X n
= k j T (ek )
k =1
X n X n
= kj ik ei
k =1 i= 1
X n 0BX n 1
C
= B
B
B
B
@
C
C
ik k j CCA ei (9.6)
i= 1 k = 1

Also, we have
n
X
T ( f j) = k j fk
k =1
X n
= k j S (ek )
k =1
X n n
X
= kj i j ei
k =1 i= 1
X n 0BXn 1
C
= B
B
B
B
@ ik i j
C
C
A ei
C
C (9.7)
i= 1 k = 1

Thus, from
0
(9.6) and
1
(9.7), we have
0 1
n BX
X n C n BX
X n C
B
B
B
B
@ ik
C
C
k j CCA ei = B
B
B
B
@ ik
C
C
i j CCA ei
i= 1 k = 1 i= 1 k = 1
n
X X n
) ik k j = ik k j
k =1 k =1
) [ ik ]nn [ k j ]nn = [ ik ]nn [ k j ]nn
) [ i j ]nn [ i j ]nn = [ i j ]nn [ i j ]nn (* change of sux is immaterial)
) [ i j ] 1 [ i j ][ i j ] = [ i j ] [ i j ][ i j ]
1

) [ i j ] 1 [ i j ][ i j ] = [i j ][ i j ]
) [ i j ] 1 [ i j ][ i j ] = [ i j]

Thus, [ i j ] and [ i j ] are similar matrices.

Also, we observe that

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
 9.3. Determinants and the Spectrum of an Operator 179

[ i j] = [ i j ] 1 [ i j ][ i j ]
i:e: [T ]B 0 = [ i j ] 1 [T ]B [ i j ] (9.8)

where [ i j ] is the matrix of the operator S relative to the basis B . S is the

opoerator which maps the B onto the basis B0 . The above relation gives us a
formula which enables us to write the matrix of T relative to the basis B0 , when
we already knew the matrix of T relative to the basis B .
Converse part: Suppose that [ i j ] and [ i j ] are two n  n similar matrices.
Then there exists a non-singular matrix [ i j ]nn such that
[ i j ] 1 [ i j ][ i j ] = [ i j]

Let B = fe1 ; e2 ; : : : ; en g be any ordered basis for H and let T be the operator on
H whose matrix relative to B is [ i j ] i:e: [T ]B = [ i j ] .

Let S be the operator on H whose matrix relative to B is [ i j ] . Since the

matrix [ i j ] is non-singular and hence the operator S is alos non-singular.
Let B0 = fS e1 ; S e2 ; : : : ; S en g . Since S is a non-singular operator and hence
it carries a basis onto a basis and so B0 is also a basis for H , Thus, we have
[S ]B = [ i j ] .
Hence by (9.8), we have
[T ]B 0 = [ i j ] 1 [T ]B [ i j ]
= [ i j ] 1 [ i j ][ i j ]
= [ i j]

Thus, [ i j ] and [ i j ] are the matrices of T relative to the bases B and B0

respectively. 

9.3. Determinants and the Spectrum of an Operator

Let [] i j ] be an nn matrix. The determinant of this matrix which we

denote by det ([ i j ]) , a scalar associated with it in such a way that

1. det ([i j ]) = 1 ;

2. det ([ i j ][ i j ]) = det ([ i j ][ i j ]) ;

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
180 9.3. Determinants and the Spectrum of an Operator

3. [ i j ] 6= 0 , [ i j ] is a non-singular;

 
4. det [ i j ] [i j ] is a polynomial, with complex coe cients, of

degree n in the variable .

The determinant function det is thus a scalar-valued functio of matrices

which has certain properties. In elementary work, the determinant of a

matrix is usually written as follows:

n11 12 1

n
det ([ i j ]) =
21 22 2

n n


nn
1 2

Theorem 9.9. Let T be an operator on H . If B and B0 are bases on H , then

the matrices [ i j ] and [ i j ] of T relative to the bases B and B0 may be entirely
di erent but they have the same determinant.

Proof. By Theorem 9.8, there exists a non-singular matrix [ i j] such that

[ i j ] = [ i j ] 1 [ i j ][ i j ]
   
) det [ i j ] = det [ i j ] [ i j ][ i j ] 1

     
= det [ i j ] 1
det [ i j ] det [ i j ]
     
= det [ i j ] 1
det [ i j ] det [ i j ]
   
= det [ i j ] 1 [ i j ] det [ i j ]
   
= det [i j ] det [ i j ]
 
= det [ i j ]

Thus, the two matrices have the same determinant. 

Denition 9.4. Let T be an operator on an n -dimensional Hilbert space H .
Then the determinant of T is the determinant of the matrix of T relative to any
ordered basis for H .

Theorem 9.10. Let S and T be operators on a nite-dimensional Hilbert space

H of dimension n . Then

(i) det(I ) = 1 , where I is the identity operator;

(ii) det(S T ) = (detS )(detT ) ;

(iii) det(S ) 6= 0 , T is non-singular; and

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
9.3. Determinants and the Spectrum of an Operator 181

(iv) det(T I ) is a polynomial, with complex coecients, of degree n in the

variable  .

Proof. Let B be any ordered basis for H . Let T be any opoerator on H , then
by dention we have det(T ) = det([T ]B ) .
(i) , we have det(I ) = det([I ]B ) = det(i j ) = 1

(ii) , we have det(S T ) = det([S T ]B )

= det ([S ]B [T ]B )
= det ([S ]B ) det ([T ]B )
= (detS )(detT )

(iii) , T is non-singular , [T ]B is non-singular

, det ([T ]B ) 6= 0
, det [T ] 6= 0

(iv) det(T I ) = det[(T I )]B

= det([T ])B det([[I ]])B
= det([T ])B det([I ])B
= det([ i j ]) det(i j )
= det([ i j ]) det(i j )
= det([ i j ] i j )
11 



12 1 n



n
= 21 22 2

n1 n



2 nn 

On Expanding, we get a polynomial with complex coecients of degree n in

. 
Theorem 9.11. An operator T on a nite-dimensional Hilbert space H is
singular , there exists a non-zero vector x in H such that T x = 0 .

Proof. Suppose that there exists a non-zero vector x in H such that T x = 0.

Then,we have

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
182 9.3. Determinants and the Spectrum of an Operator

Tx = T0

Since x 6= 0 therefore x and 0 are two distinct elements in H and they have the
same image under T . Thus, the mapping T is not one-to-one. Hence T is not
non-singular i:e:; T is singular.
Conversely, assume that T is singular. Suppose there exists no non-zero vector
x in H such that T x = 0 .
Tx = 0 ) x=0

Then T must be one-to-one. For

Ty = Tz
) Ty Tz = 0
) T (y z) = 0
) y z = 0
) y = z

Since H is nite-dimensional, therefore T is one-to-one implies that T is onto

and so T is non-singular, which contradicts our hypothesis that T is singular.
Hence there must exists a non-zero vector x such that T x = 0 .
This completes the proof of the theorem. 

Next, our aim is to prove that every operator T on a nite-dimensional

Hilbert space has an eigen value.

Theorem 9.12. If T is an arbitrary operator T ona nite dimensional Hilbert

space H , then the eigen values of T consitute a non-empty nite subset of the
complex plane. Furthermore, the number of points in this does not exceed the
dimension n of the space H .

Proof. Let T be an operator on a nite-dimensional Hilbert space H of

dimension n .
A scalar  is an eigen value of T .
) there exists a non-zero vector x such that T x =  x .

) there exists a non-zero vector x such that (T I ) = 0 .

) the operator T I is singular.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
9.3. Determinants and the Spectrum of an Operator 183

) det(T I ) = 0 .
Hence  is an eigenvalue of T if and only if  satises the equation
det(T I ) = 0 .
Let B be any ordered basis for H . Then
det(T I ) = det ([T I ]B )
= det ([T ])B [I ]B )
 
= det [T ])B [i j ]B

Let [T ]B = [ i j ]B . Then in its expanded form the equation det(T I ) = 0 takes

the form
11 



12 1 n



n
21 22 2
= 0 (9.9)













n1 n



2 nn 

The equation (9.9) is called the characteristic equation of the operator T . If

we expand the determinant on the left hand side of (9.9), we have a polynomial
equation with complex coecents of degree n in the variable  . By the
fundamental theorem of algebra the equation (9.9) has a root in the eld of
complex numbers. Hence any operator T on H has an eigenvalue.
Further, the equation (9.9) has exactly n roots in the complex eld. Some
of these roots may of course be repeated, in which case there are fewer than n
distinct roots. Hence T has an eigenvalue and the number of distinct eigenvalues
of T is less than or equal to n i:e: the number of eigen values does not exceed
the dimension of the space. 

Let us Sum Up:

In this unit, the students acquired knowledge to

ˆ Dene eigenvalues and eigenvectors.

ˆ understand the concept of spectrum of an operator.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
184 9.3. Determinants and the Spectrum of an Operator

Check Your Progress:

1. If T is an arbitrary operator T ona nite dimensional Hilbert space

H , then the eigen values of T consitute a non-empty nite subset of

the complex plane. Furthermore, the number of points in this does

not exceed the dimension n of the space H.

2. Let B be a basis for H , and T an operator whose matrix relative to

B is [ i j ] . Then T is non-singular , [ i j ] is non-singular, and in
this case [ i j ]
1
= [T 1 ] .

Suggested Readings:
1. Simmons G.F., “Introduction to Topology and Modern Analysis”,

Tata Mc-Graw Hill Pvt. Ltd., New Delhi, 2011.

2. Kreyszig E.,” Introductory Functional Analysis with Applications,

John Wiley & Sons, New York, 2007.

3. Limaye B. V.,” Functional Analysis”, New Age International Ltd.,

Publishers, Second Edition, New Delhi,1996.

4. Co man C. and Pedrick G., “First Course in Functional Analysis”,

Prentice-Hall of India, New Delhi, 1995..

5. Conway J.B., “A Course in Functional Analysis”, Springer-Verlag,

New York, 2008.

6. Bollobas B.,”Linear Analysis”, Cambridge University Press, Indian

Edition, New York, 1999.

7. Nair M.T., “Functional Analysis, A First course”, Prentice Hall of

India, New Delhi, 2010.

Web Resources:
// / /
1. https: archive.nptel.ac.in courses 111 106 111106147 / / /
2.

// / / /
https: courses.cs.washington.edu courses cse521 16sp 521-lecture-8.pdf /

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
Block-V

UNIT-10

SPECTRAL THEOREM AND

FIXED POINT THEOREM

Structure
Objective
Overview
10. 1 The Spectral Theorem.
10. 2 Fixed Point Theorem.
Let us Sum Up
Check Your Progress
Suggested Readings

Overview
In this unit, we shall study the spectral theorem, xed theorem

and some applications to analysis.

M.Sc.(Mathematics)-II Year-III Sem 185 Functional Analysis

 186 10.1. The Spectral Theorem

Objectives
After successful completion of this lesson, students will be able to

ˆ understand the concept of spectral theorem and spectral

resolution

ˆ explain the concept of xed point theorem

10.1. The Spectral Theorem

The essence of spectral theorem is that every normal operator on a nite

dimensional Hilbert space H is made of projections in a way which clearly

reveals the geometric nature of its action on the vectors in H. Now, we

shall discuss some properties of the spectra of normal operators.

Theorem 10.1. If T is a normal operator on a Hilbert space H , then x is

an eigenvector of T with eigenvalue  , x is an eigenvector of T  with
eigenvalue  .

Proof. Since T is a normal operator on a Hilbert space H , therefore T I is

also a normal operator on H , where  is any scalar. Now,
(T I ) = T I 
= T I

Since T I is normal and hence, we have

k(T I ) xk = k(T I ) xk 8 x 2 H
, k(T I ) xk = k(T  I  ) xk 8 x 2 H
, kT x  xk = kT  x  xk 8 x 2 H (10.1)

From (10.1), we notice that T x  x = 0 if and only if T  x  x = 0 .

Therefore, x is an eigenvector of T with eigenvalue  if and only if it is an
eigenvector of T  with eigenvalue  . 
Theorem 10.2. If T is a normal operator on a Hilbert space H , then the
eigenspaces of T are pairwise orthogonal.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
10.1. The Spectral Theorem 187

Proof. Let Mi and M j for i 6= j be eigenspaces of normal operator T on H

corresponding to the distinct eigen values i and  j .
Let xi be any vector in Mi and x j be any vector in M j . Then we have
T xi = i xi
and T x j = jxj

From the Theorem 10.1, we have

i ( xi ; x j ) = (i xi ; x j )
= (T xi ; x j )
= ( xi ; T  x j )
= ( xi ;  j x j )
=  j ( xi ; x j )
) (i  j )( xi ; x j )& = 0
i:e:; xi xj = (* i 6=  j )
i:e:; xi = xj

Therefore, Mi is orthogonal to M j .
Hence Mi0 s are pairwise orthogonal. 
Theorem 10.3. If T is a normal, then each Mi reduces T .

Proof. Let Mi be an eigenspace of T corresponding to the eigenvalue i . In

order to prove that Mi reduces T , it is enough to show that Mi is invariant both
under T and T  .
Now, Mi is invariant under T because Mi is an eigenspace of T . Now, our
wish is to prove that Mi is invariant under T  . For this, let us take any vector xi
in Mi . Then T xi = i xi .
By Theorem 10.1, we have T  xi =  xi .
Since Mi is a linear subspace of H , therefore xi 2 Mi and  is some scalar,
which implies that  xi 2 Mi . Thus, we have
xi 2 Mi ) T  x =  xi 2 Mi

Thus, Mi is also invariant under T  .

Hence Mi reduces T . 
Theorem 10.4. If T is normal, then Mi0 s span H .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
188 10.1. The Spectral Theorem

Proof. Since T is a normal operator on H and Mi0 s are eigenspaces of T .

Therefore, Mi0 s are pairwise orthogonal and P0i s are projections on Mi0 s . But
P0i s are pairwise orthogonal. Let
M = M1 + M2 + : : : + Mm

Then by theorem, M is a closed linear subspace of H and it's associated

projection is
P = P1 + P2 + : : : + Pm

Since T is a normal operator on H , therefore each eigen Mi of T reduces T .

Also, Pi is the projection on the closed linear subspace Mi of H . Therefore Mi
reduces T implies that Pi T = T Pi .
Thus, Pi T = T Pi for each Pi . Therefore, we have
TP = T (P1 + P2 + : : : + Pm )
= T P1 + T P2 + : : : + T Pm
= P1 T + P T + : : : + Pm T
2

= (P1 + P + : : : + Pm )T
2

= PT

Now, PT = T P and P is the projection on M . Therefore, M reduces T and so

M ? is invariant under T . Let U be an operator on a nite dimensional Hilbert
space M ? and U x = T x for all x 2 M ? .
If x is an eigenvector for U corresponding to the eigenvalue  , then x 2 M ?
and U x =  x . Therefore T x =  x and so it is also an eigenvector for T .
Therefore each eigenvector for U is also an eigenvector for T . But T has no
eigenvectors in M ? since all the eigenvectors for T are in M and M \ M ? = f0g .
So U is an operator on a nite dimensional Hilbert space M ? and U has no
eigenvector and no eigenvalue. Therefore we have M ? = f0g .
If M ? 6= f0g , then every operator on a non-zero nite dimensional Hilbert
space have an eigenvalue. Now
M? = f0g ) M =H

Thus, M1 + M2 + : : : + Mm = H and so Mi0 s span H .

Hence the theorem. 
Theorem 10.5 (The Spectral Theorem). Let T be an operator on a nite

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
10.1. The Spectral Theorem 189

dimensional Hilbert space H . Let 1 ; 2 ; : : : ; m be the distinct eigenvalues

of T ; let M1 ; M2 ; : : : ; Mm be their corresponding eigenspaces; and let
P1 ; P2 ; : : : ; Pm be the projections on these eigenspaces. Then the following three
statements are equivalent to one another:

(i) The Mi0 s are pairwise orthogonal and span H ;

(ii) The P0i s are pairwise orthogonal, P1 + P2 + : : : + Pm =I and

T =  P +  P + : : : + m Pm
1 1 2 2

(iii) T is normal.

Proof. (i) ) (ii) . Since the Mi0 s are pairwise orthogonal and span H , therefore
each vector x in H can be uniquely expressed in the form
x = x1 + x2 + : : : + xm (10.2)

where xi 2 Mi for each i = 1; 2; : : : ; m .

Now, it is given that Mi0 s are pairrwise orthogonal. Also P0i s are the
projections on Mi0 s . Therefore P0i s are pairwise orthogonal.
i:e: i 6= j ) Pi P j = 0 .
If x is any vector in H , then from (10.2), for each i , we have
Pi x = Pi ( x1 + x2 + : : : + xm )
= Pi x1 + Pi x2 + : : : + Pi xm ) (10.3)

Now xi 2 Mi which is in the range of Pi . Therefore Pi x = xi . Further

let j 6= i . Then Mi ? M j . Since x j 2 M j for each j , therefore if i 6= j ,
then x j ? Mi i:e:; x j 2 Mi? . But Mi? is the null space of Pi . Therefore
xj 2 Mi? ) Pi x j = 0 . Thus Pi xj j = 0 if i 6= j and Pi xi = xi . Therefore
(10.3), we have
Pi x = xi (10.4)

Now, for all x in H , we have

Ix = x
= x1 + x2 + : : : + xm
= P1 x + P2 x + : : : + Pmx
= (P1 + P2 + : : : + Pm) x
) I = P1 + P2 + : : : + Pm

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
190 10.1. The Spectral Theorem

Again for all x in H , we have

Tx = T ( x1 + x2 + : : : + xm )
= T x1 + T x2 + : : : + T xm
= x +  x + : : : + m xm
1 1 2 2

=  P x +  P x + : : : + m Pm x
1 1 2 2

= ( P +  P + : : : + m Pm ) x
1 1 2 2

) T =  P +  P + : : : + m Pm
1 1 2 2

(ii) ) `(iii) . Give that T = 1 P1 + 2 P2 + : : : + m Pm where P0i s are pairwise

orthogonal projections. Now, our wish is to prove that T is normal.
Since each Pi is projection, therefore Pi = Pi and P2i = Pi . Also P0i s are
pairwise orthogonal. Therefore i 6= j ) Pi P j = 0 . Now
T = ( P +  P + : : : + m Pm )
1 1 2 2

=  P +  P + : : : + m Pm
1 1 2 2

We have
TT = ( P +  P + : : : + m Pm )( P +  P + : : : + m Pm )
1 1 2 2 1 1 2 2

= j j P + j j P + : : : + jm j Pm (* Pi = Pi )
1
2
1 2
2
2
2 2

Similarly, we can prove that T  T = j j P + j j P + : : : + jm j Pm .

1
2
1 2
2
2
2

Thus, we have T T  = T  T and hence T is normal.

(iii) ) (i) : Suppose T is a normal operator on H . Then we shall prove that
Mi0 s are pairwise orthogonal and Mi0 s span H .

By Theorem 10.2, we have Mi0 s are pairwise orthogonal.

Also, by Theorem 10.4, we have Mi0 s span H .
This completes the proof of the theorem. 
Denition 10.1. Let T be an operator on a Hilbert space H . If there
exists distinct complex numbers 1 ; 2 ; : : : ; m and non-zero pairwise orthogonal
projections P1 ; P2 ; : : : ; Pm such that
T =  P +  P + : : : + m Pm
1 1 2 2

and P + P + : : : + Pm = I
1 2

then the above expression for T is called a spectral resolution for T .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
10.1. The Spectral Theorem 191

Note 10.1. The spectral theorem ensures that every normal operator T on a
non-zero nite dimensional Hilbert space H has a spectral resolution.

The following theorem shows that the spectral resolution of a normal

operator on a nite dimensional non-zero Hilbert space is unique.

Theorem 10.6. The spectral resolution of a normal operator on a nite

dimensional non-zero Hilbert space is unique.

Proof. Let T be a normal operator on a nite dimensional non-zero Hilbert

space H . Let
T =  P +  P + : : : + m Pm
1 1 2 2 (10.5)

be a spectral resolution of T . Then 1 ; 2 ; : : : ; m are distinct complex numbers

and P0i s are non-zero pairwise orthogonal projections such that
P1 + P2 + : : : + Pm = I (10.6)

Our aims is to prove that the scalars 1 ; 2 ; : : : ; m are precisely the distinct
eigenvalues of T .
First, we shall prove that for each i , i is an eigenvalue of T .
Since Pi 6= 0 , therefore there exists a non-zero vector x in the range of Pi .
But Pi is a projection. Therefore Pi x = x . Now
Tx = (1 P1 + 2 P2 + : : : + m Pm ) x
= (1 P1 + 2 P2 + : : : + m Pm )Pi x (* x = Pi x)
=  P Pi x +  P Pi x + : : : + m Pm Pi x
1 1 2 2

= i Pi x2
(* Pi P j = 0; if i 6= j)
= i Pi x (* P2i = Pi ; Pi being a projection)
= i x

Thus x is a non-zero vector such that x = i x . Therefore i is an eigenvalue of

T.

Now, we shall prove that each eigenvalue of T is an element of the set

f1 ; 2 ; : : : ; m g . Since T is an operator on a nite-dimensional Hilbert space,
therefore T must possess an eigenvalue.
Let  be an eigenvalue of T , then there exist a non-zero vector x in H such
that

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
192 10.1. The Spectral Theorem

T x = x
) T x = I x [* I x = x]
) ( P +  P + : : : + m Pm ) x = (P + P + : : : + Pm ) x
1 1 2 2 1 2

) ( )P x + ( )P x + : : : + (m )Pm x = 0

1 1 2 2

Operating on this with Pi and use the fact that P2i = Pi and Pi P j =0 if i 6= j ,
we get
(i )Pi x = 0 for i = 1; 2; : : : ; m

If i 6=  for each i , then we have Pi x = 0 for each i . Therefore

P1 x + P2 x + : : : + Pm x = 0
) (P + P + : : : + Pm ) x
1 2 = 0
) Ix = 0
) x = 0

This contradicts the fact that x is a non-zero vector. Hence  must be equal to
i for some i .
Thus, we have proved that in the spectral resolution (10.5) of T the scalars
0i s are precisely the distinct eigenvalues of T .
Therefore, if
T = 1 Q1 + 2 Q2 + : : : + k Qk (10.7)

is another spectral resolution of T , then the scalars 0i s are precisely the distinct
eigenvalues of T . Therefore renaming the projections Q0i s , if necessary we can
write (10.7) in the form
T =  Q +  Q + : : : + m Qm
1 1 2 2

Now, we shall prove that in the spectral resolution of (10.5) of T the P0i s are
uniquely determined as specic polynomials in T . We have
T0 = I = P + P + : : : + Pm
1 2

T =  P +  P + : : : + m Pm
1 1 2 2

T 2
= ( P +  P + : : : + m Pm )( P +  P + : : : + m Pm )
1 1 2 2 1 1 2 2

=  P +  P + : : : + m Pm
2
1 1
2
2 2
2

Similarly, we can prove that T n = n P + n P + : : : + nm Pm

1 1 2 2 where n is any

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
10.1. The Spectral Theorem 193

non-negative integer.
Therefore if p(z) be any polynomial, with complex coecient in the complex
variable z , then taking a linear combinations of the above relations, we get
m
X
p(T ) = p(1 )P1 + p(2 )P2 + : : : + p(m )Pm = p(i )Pi
i= 1

Now suppose that pi is a polynomial such that pi ( j ) = i j i:e:; pi (i ) = 1 and

pi (i ) = 0 if j 6= i . Then taking p j in place of p , we get
m
X
p j (T ) = p j (i )Pi
i= 1
Xm
= i j Pi = P j
i= 1

Thus, for each j , we have P j = p j (T ) , which is a polynomial in T . But we must

show the existence of such a polynomial p j over the eld of complex numbers.
Clearly
(z 1 )(z 2 ) : : : (z  j 1 )(z  j+1 ) : : : (z m )
p j (z) =
( j 1 )( j 2 ) : : : ( j  j 1 )( j  j+1 ) : : : ( j m )

serves the purpose i:e: p j (i ) = i j .

If we apply the same arguments for Q0i s then we shall get Q j = p j (T ) for
each j .
Therefore P j = Qj for each j . Hence the two spectral resolution of T are
the same. 
Theorem 10.7. If T is a normal operator on a nite dimensional Hilbert space
H , then there exists an orthonormal basis for H relative to which the matrix of
T is diagonal.

Proof. Let T =  P +  P + : : : + m Pm
1 1 2 2 be the spectral resolution of T . Then
1 ; 2 ; : : : ; m are precisely the distinct eigenvalues of T and P1 ; P2 ; : : : ; Pm are
the projections on M1 ; M2 ; : : : ; Mm which are the eigenspaces of the eigenvalues
1 ; 2 ; : : : ; m respectively. Also we have
P1 + P2 + : : : + Pm = I

Since Mi0 s are eigenspaces of a normal operator, therefore they are pairwise
orthogonal i:e:; Mi ? M j for i 6= j .
Now each Mi is a nite-dimensional non-zero Hilbert space. Therefore
each Mi contains a complete orthonormal set which will also be a basis for

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
194 10.1. The Spectral Theorem

it. Let B1 ; B2 ; : : : ; Bm be orthonormal bases for the spaces M1 ; M2 ; : : : ; Mm

respectively.
m
[
Now our claim is that B = Bi i:e: it is the union of the B0i s which are
i= 1
orthonormal basis for H . Clearly B is an orthonormal set, since each Bi is
an orthonormal set and any vector in Bi is orthonormal to any vector in B j , if
i 6= j .

Observe that the vectors in Bi are some elements of Mi and the vectors in B j
are some elements of M j . The eigenspaces Mi and M j are orthogonal if i 6= j .
Since B is an orthonormal set, therefore B is linearly independent. Now B
will be a basis for H , if we prove that B generates H .
Let x be any vector in H . Then
x = I x = (P1 + P2 + : : : + Pm ) x
= P1 x + P2 x + : : : + Pm x
= x1 + x2 + : : : + xm where xi = Pi x

Since Pi x is the range of Pi , therefore xi is in Mi . For each i , the vector xi can

be expressed as a linear combination of the vectors in Bi which is a basis for Mi .
Hence H is generated by B . Therefore B is an orthonormal basis for H . Since
each non-zero vector in Mi is an eigenvector of T , therefore each vector in Bi
is an eigenvector for T . Consequently each vector in B is an eigenvector for T .
Thus, B is an orthonormal basis for H and each vector in B is an eigenvector
for T . Let us nd the matrix of T relative to the basis B . Let
B = fe ; e ; : : : :en g
1 2

Since each vector in B is an eigenvector of T , therefore

T e1 = 1 e1 ; T e2 = 2 e2 ; : : : ; T en = n en :

where 1 ; ; : : : ; n are some scalars. Now

2

T e1 = e1 = e + 0e + : : : + 0en
1 2

T e2 = e2 = 0e + e + : : : + 0en
1 2

..
.
T en = en = 0e + 0e + : : : +
1 2 en

Thus, by denition of matrix of T relative to the basis B , we have

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
 10.2. Fixed Point Theorems 195
0 1
B
B
B
B 1 0 : : : 0 CCCC
B
B C
C
B
B
B0 : : : 0 CCCCC
[T ]B = B
B
B
B
B
2
C
C
: : : : : : : : : : : :CCCCC
B
B
B
B
B
B C
C
@
0 0 : : : nA

which is a diagonal matrix. 

10.2. Fixed Point Theorems

Let f be a continuous mapping of the closed interval [ 1; 1] into itself.

The Figure 10.1. suggests that the graph of f must touch or cross the

indicated diagonal, or more precisely, that there must exist a point x0 in

[ 1; 1] with the property that f ( x0 ) = x0 .

Figure 10.1

The proof is quite simple, consider the continuous function F dened

on [ 1; 1] by

F ( x) = f ( x) x

Note that F ( 1)  0 and F (1)  0 .

Thus, by applying Weierstrass intermediate value theorem, there must

exists a point x0 in [ 1; 1] such that F ( x0 ) = 0 i:e:; f ( x0 ) = x0 .

Note 10.2. A topological space X is called a xed point space if every continuous
mapping f on X into itself has a xed point in the sense that f ( x0 ) = x0 .

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
196 10.2. Fixed Point Theorems

From the above illustration, we shown that [ 1; 1] is a xed point space.

Furthermore, the closed disc f( x; y) : x2 + y2  1g in the Euclidean Plane R2
is also a xed point space.

Theorem 10.8 (Brouwer's Fixed Point Theorem). The closed unit sphere
S = f x : k xk 1g in Rn is a xed point space.

Theorem 10.9 (Schauder's Fixed Point Theorem). Every convex compact

subspace of a Banach space is a xed point space.

Denition 10.2. Let X be an arbitrary metric space with metric d . A mapping

T of X into itself is called a contraction if there exists a positive real number
r < 1 with the property that d(T x; T y)  rd( x; y) for all points x and y in X .

Lemma 10.1. If T is a contraction dened on a complete metric space X , then

T has a unique xed point.

Proof. Let x0 be an arbitrary point in X and write

x1 = T x0
x2 = T 2 x0 = Tx 1

..
.
xn = T n x0 = T xn 1

If m < n , then
d( xm ; xn ) = d (T m x0 ; T n x0 )
= d T m x0 ; T m T n m x0

 rm d( x0 ; T n m x0 ) = rm d( x0 ; xn m )
 rm [d( x0 ; x1 ) + d( x1 ; x2 ) + : : : + d( xn m 1 ; xn m )]
 rm d( x0 ; x1 )[1 + r + : : : + rn m 1 ]
rm d( x ; x )
 0 1
:
1 r

Since r < 1 , it is clearly from the fact that f xn g is a Cauchy sequence, and by the
completeness of X , there exists a point x in X such that xn ! x .
Since T is continuity and to infer that x is a xed point, we have
Tx = T (lim xn )
= lim T xn
= lim xn+1
= x

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
10.2. Fixed Point Theorems 197

Thus x is a xed point. Now, we shall prove that the uniqueness of x .

If y is also a xed point, i:e: T y = y .
d( x; y) = d(T x; T y)  rd( x; y)

Since r < 1 , this implies that d( x; y) = 0 or y = x . 

@f
Theorem 10.10 (Picard's Theorem.). If f ( x; y) and are continuous in a
@y
closed rectange R = f( x; y) : a1  x  a2 and b1  y  b2 g and if ( x0 ; y0 ) is
an interior point of R , then the di erential equation
dy
dx
= f ( x; y) (10.8)

has a unique solution y = g( x) which passes through ( x0 ; y0 ) .

@f
Proof. Since f ( x; y) and are continuous in R , they are bounded, and hence
@y
there exist constants K and M such that
j f ( x; y)j  K (10.9)
@f
 M (10.10)
@y

for all points (x; y) in R .

If ( x; y1 ) and ( x; y2 ) are two points in R , then the mean value theorem ensures
that
@
j f ( x; y )
1 f ( x; y2 )j = jy 1 y2 j
@y
f ( x; y 1 + (y 2 y1 )) (10.11)

for some  such that 0 <  < 1 .

Thus, from (10.10) and (10.11), we have
j f ( x; y )
1 f ( x; y2 )j  M jy1 y2 j (10.12)

for all points (x; y1 ) and ( x; y2 ) in R .

At this stage, we conveninently replace our problem by an equivalent problem
to an integral equation.
If y = g( x) satises (10.8) and has the property that g( x0 ) = y0 , then
integrating (10.8) from x0 to x which gives
Z x
g( x) g( x0 ) = f (t; g(t))dt
x0
Z x
i:e:; g( x) = y0 + f (t; g(t))dt (10.13)
x0

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
198 10.2. Fixed Point Theorems

Conversely, if y = g( x) satises (10.13), then it is evidently that g( x0 ) = y0 .

Di erentiating (10.13),we get
dy
dx
= f ( x; y)

It remains to prove that the integral equation (10.13) has a unique solution.
For this, we choose a positive number a such that Ma < 1 and the closed
rectangle R0 determined by j x x0 j a and jy y0 j Ka is contained in R .
Let X be the set of all continuous real functions y = g( x) dened on the
closed interval j x x0 j a such that jg(y) y0 j Ka . Clearly, X is a closed
C
subspace of the complete metric space [ x0 a; x0 + a] and hence it is also a
complete metric space.
Consider the mapping T of X onto itself dened by T g = h , where
Z x
h( x) = y0 + f (t; g(t))dt
x0

It is very clear from (10.9) the Zfact that T maps X into itself. For if
x
jh( x) y0 j = f (t; g1 (t))

f (t; g2 (t)) dt  Ka
x0

Furthermore, from (10.12), we haveZ

x
jh ( x)
1 h2 ( x)j = [ f (t; g1 (t)) f (t; g2 (t))]dt
x0
 Ma supjg1 ( x) g2 ( x)j

Since Ma < 1 , this exhibits that T is a contraction on a complete metric space

X.

Thus, by Lemma 10.1, we have the equation T g = g has a unique solution.

This completes the proof of the theorem. 

Let us Sum Up:

In this unit, the students acquired knowledge to

ˆ Dene xed point.

ˆ understand the concept of spectrum theoerm.

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
10.2. Fixed Point Theorems 199

Check Your Progress:

1. State and Prove Spectrum theorem.

2. If T is normal, then the Mi0 s span H.

Suggested Readings:

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

Tata Mc-Graw Hill Pvt. Ltd., New Delhi, 2011.

2. Kreyszig E.,” Introductory Functional Analysis with Applications,

John Wiley & Sons, New York, 2007.

3. Limaye B. V.,” Functional Analysis”, New Age International Ltd.,

Publishers, Second Edition, New Delhi,1996.

4. Co man C. and Pedrick G., “First Course in Functional Analysis”,

Prentice-Hall of India, New Delhi, 1995..

5. Conway J.B., “A Course in Functional Analysis”, Springer-Verlag,

New York, 2008.

6. Bollobas B.,”Linear Analysis”, Cambridge University Press, Indian

Edition, New York, 1999.

7. Nair M.T., “Functional Analysis, A First course”, Prentice Hall of

India, New Delhi, 2010.

Web Resources:

// /
1. https: mathworld.wolfram.com FixedPointTheorem.html

2.

// / / / /
http: www.u.arizona.edu mwalker econ519 Econ519LectureNotes FixedPointTheorems.pdf

zzzzz

Functional Analysis
M.Sc.(Mathematics)-II Year-III Sem
Document Information

Analyzed document Functional analysis.pdf (D151795707)

Submitted 2022-12-02 12:41:00

Submitted by

Submitter email [email protected]

Similarity 8%

Analysis address [email protected]

Sources included in the report

Romen_Maths_Thesis.pdf
14
Document Romen_Maths_Thesis.pdf (D47516061)

Jaya Mary P Full Thesis.pdf

12
Document Jaya Mary P Full Thesis.pdf (D34303728)

URL: https://www.ddegjust.ac.in/2019/4/mal%20641_19042019.pdf
346
Fetched: 2021-11-16 22:59:14

URL: https://cpb-us-
w2.wpmucdn.com/sites.udel.edu/dist/b/10716/files/2022/02/hoefer_functional_anal... 12
Fetched: 2022-05-14 14:05:56

07170511.Assignment 1.pdf
4
Document 07170511.Assignment 1.pdf (D103826761)

Dissertation.pdf
2
Document Dissertation.pdf (D108354499)

Kristoffer-Jansen(01715HW7).pdf
1
Document Kristoffer-Jansen(01715HW7).pdf (D22770086)

mathematics.pdf
3
Document mathematics.pdf (D123770633)

P_USHA_ MATHS.pdf
14
Document P_USHA_ MATHS.pdf (D58922462)

Week05_SebastianLarsson.pdf
2
Document Week05_SebastianLarsson.pdf (D80973024)

MA40057_Problem_Sheet_5.pdf
2
Document MA40057_Problem_Sheet_5.pdf (D97826501)