Block 1
Block 1
FUNCTIONAL ANALYSIS
Indira Gandhi National Open
University School of Sciences
Block
1
NORMED LINEAR SPACES
Block Introduction 5
Notations and Symbols 6
UNIT 1
Inner Products and Norms 9
UNIT 2
Continuous Linear Maps and Functionals 55
UNIT 3
Hahn-Banach theorems 81
Course Design Committee
Dr. A.K Vijayrajan, Faculty Members
Kerala School of Mathematics, Calicut School of Sciences, IGNOU
Prof. Sujatha Varma
Prof. K. Parthasarathy, (Retd.) Prof. Deepika
Ramanujam Institute, University of Madras, Chennai Dr. S. Venkataraman
Prof. Parvin Sinclair
Prof. M.S. Balasubhramanian , (Retd.) Prof. Poornima Mital
University of Calicut, Kerala
Prof. V. Muruganandhan
NISER, Orissa
Acknowledgements: Dr. Suvankar Biswas for his comments on the units. Mr. Santosh Kumar Pal
for word processing the Manuscript and Mr Surender Singh Chauhan for the CRC.
Disclaimer – Any materials adapted from web-based resources in this course are being used for educational purposes
only and not for commercial purpose.
, 2023
ISBN-
All right reserved. No part of this work may be reproduced in any form, by mimeograph or any other means, without
permission in writing from the Indira Gandhi National Open University.
Further information on the Indira Gandhi National Open University courses may be obtained from the University’s
Office at Maidan Garhi, New Delhi-110 068.
Printed and published on behalf of the Indira Gandhi National Open University, New Delhi by the Director, School of
2 Sciences.
COURSE INTRODUCTION
Welcome to the course “Functional Analysis”. This course is a 4-credit course. In this
course, we introduce you to a branch of mathematics known as Functional Analysis. Under
this topic we study function spaces which are sets of functions with additional structures.
This provides a major link between mathematics and its applications. The important
notions that are dealt in this course are Banach spaces, Hilbert spaces and linear
functionals on these spaces.
This material is divided into 4 Blocks. In Block 1, we familiarise you to some basic
concepts in Functional analysis. We introduce to the concept of a “norm” which is another
distance measuring concept like a metric. Any linear space having a norm defined on it is
called a normed linear space. Any norm function defines a metric on a normed space and
thereby any normed linear space is a metric space. A detailed study of normed spaces
requires the study of maps from one normed spaces to another normed space. In this block,
we consider maps which are linear and bounded (also called continuous). We define
continuous linear maps from one normed linear space to another. We also familiarise you
to one of the important theorems in Functional Analysis, known as Hahn-Banach theorem.
Block 2 deals with Banach spaces. Normed spaces which are complete with respect to the
metric induced by the corresponding norm are called Banach spaces. The Banach spaces
plays a crucial role in the study of function spaces. Here we consider four important
theorems – Open mapping theorem, closed graph theorem, Uniform boundedness principle
and Bounded inverse theorem. In this block we also deal with space of bounded linear
maps defined from a normed space to the space K, the scalar field of real or complex
numbers. These spaces are called dual spaces. We consider dual of dual spaces which are
called reflexive spaces.
Block 3 introduces to another type of normed spaces known as inner product spaces. An
inner product space which is complete, under the norm induced by the inner product, is
called a Hilbert space. Here we discuss the fundamental properties of inner product spaces
more specifically for Hilbert spaces. The inner product enables us to introduce the concept
of orthogonality. We discuss orthonormal sets. Another important theorem for Hilbert
spaces namely, Riesz representation theorem is studied in this block. Finally we consider
operators on Hilbert spaces. Here we define adjoint of an operator and study four important
classes of operators, namely, self-adjoint, unitary, normal operators and compact. We also
discuss an important subclasse of self-adjoint operators, viz, compact self-adjoint-
operators.
Each Block is divided into 3 or 4 units based on different themes and each unit starts with
listing of contents and objectives. In the list of objectives in each unit we have given the
main concepts and results which we thought you should have achieved after reading this
material. Each unit is divided into sections, which may be divided into sub-sections. These
sections/sub-sections are numbered sequentially, as are the exercises and important
equations, within a unit. Regarding the exercises in each unit, these have been placed at
many places in a unit. They are meant for you to assess if you have understood the
concepts and processes that have been discussed till that point. Therefore, you must
attempt to solve all the exercises in the unit as you go along, before you look up the
solution.
Since the material in the different units is heavily interlinked, there will be cross
referencing. For this we will be using the notation Sec. x.y to mean Section y of Unit x.
3
Further, the end of an example is shown by the sign *** and the end of a proof of,
theorem/proposition/corollary is shown by the mark .
After reading the unit please go back to the objectives to see that you have not missed any
concept. Each unit also contains worked out examples. At the end, we have recapitulated
the learning points of the unit in the form of summary. This is followed by Hints/Solutions
of the exercises. We advise you to look into the answers only after you have tried them
sincerely.
e
It is important to learn the concepts in each block to understand the concepts covered in the
further blocks. Another compulsory component of this course is its assignment, which
covers the whole course. Your academic counselor at the study centre will evaluate it, and
return it to you with detailed comments. Thus, the assignment is meant to be a teaching as
well as an assessment aid. Further, you will not be allowed to take the examination of this
course till you submit your assignment response. So please submit it well in time.
Now, a word about notation. The notations and symbols used in this course material are
given in a separate list. You are advised to follow this. You may note that different text
books use different notations.
This course assumes the knowledge of linear algebra and basic metric space theory
presented in IGNOU courses on “Linear Algebra” MTE-02 of “Real Analysis” MMT-004,
Block 1. If you have any difficulty with any concept or techniques given in Chapter-1. The
course material that we have sent you is self-sufficient. If you have a problem in
understanding any portion, please ask your academic counselor for help. In addition to
these exercises, you may like to also look up some other books in the library of your study
centre, and try to solve some exercises from these books too.
1. Kreyszig, Ervin: Introductory functional analysis and applications by John Wiley &
Sons. Inc.
2. G. Backman and L. Narici’s: Functional Analysis, Dover Publications
3. S. Kesavan: Functional Analysis, Hindustan Book Agency (Trim Series)
4. S. Kumaresan and D. Sukumar: Functional Analysis, A first course, Narosa
publications.
5. B.V. Limaye: Functional Analysis, New Age International Publishers.
6. Karen Saxe: Beginning Functional Analysis, Springer Verlag.
4
NOTATIONS AND SYMBOLS (used in Block 1)
6
BLOCK 1 NORMED LINEAR SPACES
This is the first block of the course “Functional Analysis” MMT-006. In this block we
introduce you to the concept of normed linear spaces which are linear spaces (vector
spaces) with a new structure formed by a norm function. This structure induces a metric
structure which is compatible with the linear structure. Thus every normed linear space is a
metric space. We shall then discuss linear maps defined on these spaces which are
continuous with respect to the metric defined on it. In fact these continuous maps separates
points on the spaces. This is established by an important theorem known as Hahn Banach
theorem which is covered in this block.
This block is divided into 3 units. In Unit 1 we familiarize you with the metric induced by
the norm. This unit deals with basic properties of normed linear spaces. We shall also
introduce you to inner product spaces. These spaces are normed linear spaces, but not vice-
versa. Inner product space allows us to study the concept of angle, perpendicularly and so
on. You will learn that finite dimensional normed linear spaces have nice properties in
common with Euclidean spaces R n . Many examples of infinite dimensional normed linear
spaces are given in this unit. We call them standard spaces. These examples will be
mentioned throughout all the units in this material.
Unit 2 covers continuous linear maps from one normed linear space to another normed
linear space. We define norm of a continuous linear map and prove that the continuity and
boundedness are the same for normed linear spaces. The continuous/bounded linear map
from a normed linear space to its basic field is known as bounded/continuous linear
functional. We discuss examples of continuous linear maps defined on standard spaces and
also show how to compute their norms.
In Unit 3, we discuss two important theorems in Functional Analysis both known as,
Hahn-Banach Theorem – Hahn-Banach Extension Theorem and Hahn-Banach Separation
Theorem. These theorems are very important as they are used in proving many other
important results.
You may like to view the following Video lectures on certain topics related to the content
covered in this block.
1) https://youtu.be/KwYqKWc-Rqo
2) https://youtu.be/y97GvfAbl3U
3) https://youtu.be/bIYBliUwtTU
4) https://youtu.be/2xWpLmDE52E
5) https://www.youtube.com/watch?v=FkJMfsNg2cM&list=PLyqSpQzTE6M8mjwWBz0vXpmJQwR46JI
dS&index=10&t=471s
7
UNIT 1
Structure Page No
1.1 Introduction
Objectives
1.2 Definition, Properties and Examples
1.3 Standard Examples
1.4 Subspaces
1.5 Quotient spaces
1.6 Summary
1.7 Hints/Solutions
1.1 INTRODUCTION
You are already familiar with linear spaces such as the real line R , complex
n
plane C and Euclidean spaces R in general. The importance of these spaces
is that we can define algebraic structures as well as the concept of distance or
length. On the real line, for example the modulus function, is used to measure
the distance and for other Euclidean spaces, there is an analogous distance
formula. Can we have spaces other than Euclidean spaces on which we can
have both an algebraic structure and the concept of a distance? In an attempt
to find an answer to these questions, the concept of “norm” was introduced, a
concept which is, in fact, analogous to the concept of length. Any linear space
having a norm defined on it is called a normed linear space or simply a
normed space. On Euclidean spaces, besides the concept of length, angles
and perpendicularity are also studied. An inner product, helps us in defining
these. In this unit, you will study normed linear spaces and inner product
spaces.
Objectives
The objectives of this unit are:
• To define a normed linear space and an inner product space and give
examples;
• To calculate the norm of an element in standard normed linear spaces;
R n , C n , l p , 1 ≤ p < ∞, c0 , C [ a, b], L p ( R), 1 ≤ p < ∞;
• To state when a subset of a normed linear space is a subspace;
• To give examples of subspaces of standard normed linear spaces;
• To state and prove Riesz lemma;
Block 1 Normed Linear Spaces
• to define quotient spaces of a normed linear space.
• to explain the convergence in quotient spaces of a normed linear space X
and relate this to convergence in X ;
• to state and prove a necessary and sufficient condition for a quotient
space to be complete.
In the real case, conjugate symmetry is just symmetry. An inner product space
is a vector space with an inner product on it.
We have to always remember that inner product of two elements in the vector
space is a scalar.
Now let us look at some properties which we can deduce from the definition.
x, αy + βz = α x, y + β x, z
2
x, y ≤ x, x y , y
10
Unit 1 Inner Products and Norms
Proof: We prove this one by one.
(i) We have to show that <, > is conjugate linear in the second variable.
Let, x, y, z ∈ X and α, β ∈ K
Then by definition, x, αy + β z = αy + β z , x
(by linearity in the
= αy , x + β z , x first variable)
= α y , x + β z , x = α x , y + β x, z
Conversely suppose x = 0.
Then 0, y = 0 + 0, y = 0, y + 0 , y .
0, y = 0.
Let x, y ∈ X .
Consider z = y , y x − x , y y .
Then: 0 ≤ z, z = y, y x − x, y y, y, y x − x, y y
2
= y, y x, x + x, y . x, y . y, y − y , y x, y x, y
− x, y y, y y, x
= y, y { x, x y , y − x, y
2
}
2
Since y , y ≥ 0, we get x, x y , y − x, y ≥ 0.
2
x, y ≤ x, x y, y . for y ≠ 0.
11
Block 1 Normed Linear Spaces
Proof: Let us start with the right hand side (R.H.S) and simplify using linearity
in the first variable and conjugate linearity in the second variable.
− i x − i y, x − i y .
= x, x + x , y + y , x + y , y
− x, x + x, y + y , x − y , y
+ i x , x + x, y − y , x − i y , y
− i x, x + x, y − y, x + i y, y
= 4 x, y (after simplification)
***
Consider x , z = y , z x, z − y, z = 0 ∀ z ∈ X .
x − y, z = 0 ∀ z ∈ X .
x − y = 0 x = y.
***
In the case of R 2 , we have x. y = x1 y1 + x2 y 2 and x.x = x12 + x22 . But x12 + x22 is
the square of the distance of x from the origin. This number is called the norm
of the vector and is denoted by x .
1
2 2 2
If x, y are any two elements in R , [( x1 − y1 ) + ( x2 − y2 ) ] 2 gives the distance
between the x and y and it is nothing but x − y . In short the dot product or
the inner product gives a norm which in turn gives a distance.
1
x = x, x 2
.
(i) x = 0 x , x = 0 x = 0.
Suppose x = 0. Then x , x = 0 x = 0.
2
Consider α x = αx, αx .
2 2
=α x .
αx = α x .
(iii) Let x, y ∈ X .
2
Then x+ y = x + y, x + y .
= x, x + x, y + y , x + y, y
2 2
= x + y + x, y + x, y
13
Block 1 Normed Linear Spaces
2 2
= x + y + 2 Re x, y ,
2 2
≤ x + y + 2 x, y
2 2
= x + y +2 x y
=( x + y ) 2
x+ y ≤ x + y .
....
All the above three properties prove that is a norm on X .
***
x, y = x1 y1 + x2 y2 + ....xn y n
i) ⋅ 2
≥ 0, since all the terms on the R.H.S. are non-negative.
⋅ 2
= 0 x k = 0 for all k. This means that x = 0 and thus the condition in
(i) is satisfied.
1/ 2
α 2
∞
αx 2 = αx k = α x k )1 / 2
2
ii)
k =1 k =1
=α x 2
x+y 2 = ( x k + yk )
2 1/ 2
∴ x+y 2 = ( x ) + ( y )
k
2 1/ 2
k
2 1/ 2
= x2+ y 2
Example 2: Show that the Euclidean space R n is a normed space with the
1/ 2
n 2
norm defined by x 2 = x k where x = ( x 1 , x 2 ,..., x n ).
k =1
Solution: We can check that , defined as above satisfies all the properties
of an inner product. (See E1)
(
Thus K , ,
n
)is an inner product space.
The corresponding induced norm . is given by
x = x, x
1
2
1
( n
= [ x12 + ... + xn2 ] 2 . Thus K , . is a normed linear space. )
***
[(
x = x12 + x 22 ) 3/ 2
+ x3 ]
3 1/ 3
Let us wait for some time and introduce more tools before we come to this
question.
Another important property that arises naturally is the ‘distance’ or the ‘metric’.
Recall that a metric on a set X is a function d : X × X to R satisfying the
15
Block 1 Normed Linear Spaces
following: For x, y , z ∈ X ,
Proof: The proof is left as an exercise for you to verify (See E2).
Remark 2: Thus if we start with an inner product on a linear space, this inner
product induces a norm which in turn induces a metric.
We shall be talking about open sets, closed sets, compact sets, completeness
etc in normed linear spaces/inner product spaces.
All these properties are discussed with respect to the metric induced by the
norm.
2 2 2
Schwarz Inequality: x, y ≤ x y or x, y ≤ x y .
2 2 2 2
Polarization Identity: 4 x, y = x + y − x − y + i x + iy − i x − iy .
(complex case).
Now let us look at some basic properties of inner products and norms. Since
d ( x, y) := x − y is the metric induced by the norm, we shall use only x − y to
denote the metric.
First let us see what we have to prove. Then they becomes easy.
Remember both inner products and norms are functions. So what we have to
prove is that these functions are continuous. Here we shall use a
characterization of continuity of functions on metric spaces given by the
following: ( X , d X ), (Y , d Y ) are metric spaces. A function f : X → Y is
16
Unit 1 Inner Products and Norms
continuous if and only if xn → x in X f ( xn ) → f ( x) in Y i.e.
d X ( x n , x ) → 0 d Y ( f ( x n ), f ( x )) → 0 as n → ∞ .
In the case of norm, we can think of x as f (x) and in the case of inner
product we can think of x, y as g ( x, y ), a function of two variables. Thus what
we have to prove is
xn → x in X xn → x in R and
xn → x & yn → y in X xn , yn → x, y in K
Consider: x, y∈ X .
Then: x = x − y + y x ≤ x − y + y
x − y ≤ x−y (1)
Similarly − y = x − y − x.
y x−y + −x = x−y + x
y − x ≤ x− y
−( x − y )≤ x−y (2)
Since x n → x in X , we have x n − x → 0 as n → ∞.
. is a continuous function on X .
Let x n → x and y n → y in X .
17
Block 1 Normed Linear Spaces
x n , y n − x, y = xn , y n − x, y n + x, y n − x, y
= x n − x, y n + x, y n − y ≤ xn − x, y n + x, y n − y
1 1 1 1
≤ xn − x 2
yn 2
+ x 2
yn − y 2
(By Schwarz inequality)
Thus we get
1 1 1 1
xn yn − x, y ≤ k 2
xn − x 2
+ x 2
yn − y 2
x n y n → x, y as n → ∞.
, is a continuous function on X × X .
***
2 2 2 2
x + y + x − y = 2( x + y )
Proof: For x , y ∈ X ,
2 2
consider x + y + x−y
= x + y, x + y + x − y, x − y
= x , x + x , y + y , x + y , y + x, x − x , y − y , x + y , y
= 2 x, x + 2 y , y
(
=2 x + y
2 2
)
***
18
Unit 1 Inner Products and Norms
law. However if the norm does satisfy the parallelogram law, then we can
define an inner product as:
x, y =
1
4
[ 2 2 2
x + y − x − y + i x + iy − i x − iy
2
] in the complex case.
So far we have discussed important properties of inner products and norms.
We have considered the corresponding induced metric also.
E2) Let X be a normed linear space. Show that for any x, y , the function
d ( x, y ) = x − y defines a metric on X .
Next we shall see important examples of normed linear spaces and inner
products.
Example 3: Let X = c 00 , be the set of all scalar sequences which have only a
finite number of non zero entries. Define a norm on X which makes it a
normed linear space.
x + y = ( x1 + y1 , x 2 + y 2 ,....)
α x = (α x1 , α x2 ,....)
Then vector addition and scalar multiplication as defined above gives a vector
space structure in c00.
Also the collection of vectors, (e1 , e2 , e3 ...) where e n is the sequence having 1 at
the nth place and 0 everywhere else, for n = 1, 2.. is a countable collection of
linearly independent vectors in c00. So c00 is an infinite dimensional vector
space.
∞
x , y = xn y n .
n=1
∞
Again, if K = R , x, y is just x
n =1
n yn .
19
Block 1 Normed Linear Spaces
Since x, y have only finitely many non zero terms, the sum is a finite sum and
so x, y is well defined.
(
Thus c00 , , ) is an inner product space.
Under the corresponding induced norm . , given by
1 1
1 ∞ 2
∞ 2
2
x = x, x 2
= xn x n = xn ,
n=1 n=1
Before we see more examples, we prove certain inequalities which are very
useful in the proving many results.
Before we state the inequalities, we prove another inequality which is used for
proving the inequalities.
a p bq
ab ≤ + (4)
p q
1 1
where p, q are such that + = 1.
p q
1 t
x (t ) = + − t1/ p , t ≥ 0
q p
1 1 1/ p −1
Then x ′( t ) = − .t .
p p
1
1 1 −
= − . t q , t ≥ 0.
p p
−1
q
If t < 1, then x ′(t ) < 0 , as, t > 1.
Also, if t ≥ 1, x′(t ) ≥ 0.
1
p 1 t
Hence t ≤ + , ∀ t ≥ 0.
q p
ap
Taking t = , we get
bq
ap
a 1 p
q/p
≤ +b
b q p
Multiplying by b q , we get
q
q−
p bq a p
a .b ≤ + .
q p
1 1 q q
But + = 1 +1 = q q − = 1
p q p p
a p bq
Hence ab ≤ + . This proves the inequality (4).
p q
Next we prove two important inequalities theorem using the result shown
above.
1 1
Let q be such that + = 1.
p q
1 1
∞
∞ p
p
∞ q
q
Then a n bn ≤ a n bn
n =1 n=1 n=1
1 1
n p
p
n q
q
α = ai , β = bi
i =1 i =1
21
Block 1 Normed Linear Spaces
ai bi
Taking a = and b = in the inequality (4), we get
α β
p q
ai bi ai bi
. ≤ p
+
α β pα q βq
n
1 n
1 n
ai b
p q
a i bi ≤ αβ p
+ i
i =1 pα i =1 qα q i =1
1 1
= αβ + = αβ
p q
1 1
n p
p
n q
q
= ai bi
i =1 i =1
1 1
∞ p
∞ p
∞ q
q
Letting n → ∞, we get a n bn ≤ a n bn
n =1 n =1 n =1
We have
1 1 1
∞ p
p
∞ p
p
∞ p
p
an + bn ≤ a n + bn .
n=1 n=1 n=1
1 1
Let q be the conjugate index of p, i.e. + = 1. Since 1 < p < ∞, q also
p q
satisfies 1 < q < ∞.
First we consider the finite case. Let a1 ,..., a n and b1 ,..., bn be scalars.
(a ) p
Consider i + bi
i =1
22
Unit 1 Inner Products and Norms
n
= ( ai + bi ) (a ) p −1
i + bi
i =1
n n
= ai ( ai + bi ) + bi ( a i + bi )
p −1 p −1
i =1 i =1
1 1
But 1 − = . Hence
q p
1 1 1
n p
p
n p
p
n p
p
1 1 1
n p
p
n p
p
n p
p
ai + bi ≤ ai + bi
i =1 i =1 i =1
Letting n → ∞, we get
1 1 1
∞ p
p
∞ p
p
∞ p
p
an + bn ≤ an + bn
i =1 n=1 n=1
x
2
satisfying n < ∞. Define a norm on X which makes it a normed linear
n =1
space.
Here, the first thing we have to do is to check that the vector addition and
scalar multiplication as defined above are meaningful i.e. we have to check
that x + y & α x are in fact elements of l 2 whenever x, y are in l 2 . To prove
this, all we have to do is prove that the sequences x + y & α x satisfy the
convergence condition of elements in l 2 .
∞ ∞
x, y ∈ l 2 xn < ∞ & y
2 2
n < ∞.
n=1 n=1
x
2
n + yn < ∞.
n =1
x n + y n ≤ xn + y n
n=1 n=1 n=1
∞ ∞
2 2
Since both xn and y n are finite, we have
n =1 n =1
1
∞ 2
2
xn + y n <∞
n=1
∞
xn + y n < ∞
2
n =1
α xn = α x
2 2 2
l 2 is well defined. Moreover n < ∞.
n =1 n =1
∞
For x = ( x1 , x2 ....xn ...), y = ( y1 , y2 ... yn ) define x, y = xn y n .
n=1
Here again we have to first prove x, y is well defined i.e. the infinite sum
24
Unit 1 Inner Products and Norms
∞
x
n=1
n y n is a well determined scalar for any x, y ∈ l 2 .
But this follows immediately from Holder’s Inequality (Theorem 2), for by
Holder’s inequality, the right hand side is absolutely convergent and hence
convergent, meaning, x, y is well defined.
1
∞ 2
2
x = xn whenever
n=1
x = ( x1 , x2 ,... xn ...) ∈ l 2 .
We shall show that the function in (5) defines a norm on C [a, b].
f = 0.
f (t ) = 0 ∀ t ∈ [ a , b ]
f = 0.
Thus f = 0 ⇔ f = 0.
Consider α f = sup α f (t )
t∈[ a , b ]
= sup α f (t ) = α sup f (t )
t∈[ a , b ] t ∈[ a , b ]
=α f .
f + g = sup ( f + g ) (t ) = sup f (t ) + g (t )
t∈[ a ,b ] t∈[ a ,b ]
≤ sup f (t ) + sup g (t )
t∈[ a ,b ] t∈[ a ,b ]
= f + g .
Remark 7: This norm is called the sup norm on C([ a, b ]). This norm is also
called the uniform norm on C ([ a , b]), since, convergence under this norm
gives uniform convergence i.e.,
Suppose f n → f under the above norm. Then given ∈> 0. ∃ N such that
fn − f < ε ∀ n ≥ N.
= ( f n (t ) − f (t ) ) < ε, ∀ n ≥ N , ∀t ∈ [ a, b]
Example 6: Let X = c[a, b], the linear space of all continuous scalar valued
functions on [a, b] as a linear space already mentioned in the example, above.
b
f , g = f (t ) g (t ) dt
a
( )
Thus C ([a, b]), , is an inner product space.
f = f (t ) dt ,
b 2 2
a
We have seen the sequence spaces c00 and l 2 . Now we look at some more of
them.
Example 6 (The Space B): B is the set of bounded scalar sequences. Let us
show that B is a normed linear space.
Solution: Since the sum of two bounded sequences and scalar multiples of
bounded sequences are again bounded, B has a vector space structure.
27
Block 1 Normed Linear Spaces
Claim: . ∞
is a norm. The verification is left as an exercise, see (E 7)
2 2
x+ y + x− y =4
=2 x( 2
+ y
2
)
Which means . ∞
does not satisfy the parallelogram law and hence this norm
is not induced by an inner product.
Thus B, . ( ∞
) is a normed linear space but is not an inner product space.
***
Example 6 (The Space c): Let us show that the set 'c' , the set of all
convergent scalar sequences, is a normed linear space.
Solution: Since the sum of two convergent sequences and scalar multiples of
convergent sequences are convergent, c has a vector space structure.
Example 7 (The space c0 ): Let us check wether the set c0 , is the set of all
scalar sequences converging to 0, is a normed linear space.
Solution: Since the limit of the sum is the same as the sum of the limits (i.e.
lim( xn + yn ) = lim xn + lim yn and for any scalar α, lim(αxn ) = α lim xn , c0 has
n→∞ n n n n
a vector space structure.
p
xn ≤ ∞.
n =1
(We have already seen the space l 2 . ) Then l p is a normed linear space The
space B , that we have discussed in example 5 is usually denoted by l ∞ .
28
Unit 1 Inner Products and Norms
Solution: Remember, l , . ( 2
2
)is an inner product space and the norm is
given by an inner product.
x
2
convergence condition n < ∞. whereas in l p , the sequences satisfies
n =1
∞
=1 x
p
the condition n < ∞ where x = ( x1 , x2 ,...).
n
You may note that since l p -spaces are sets of scalar sequences the sum and
scalar multiplication are defined in the usual way-namely termwise addition
and termwise scalar multiplication.
But the important property that we have to check is that these operations are
well defined in the sense, just as we have checked that the sum of two square
summable is square summable, we have to prove that for all p, 1 ≤ p < ∞,
( p = 2 has already been discussed), sum of two p-summable sequences is
again p-summable.
But this follows directly from the Minkowski’s inequality (why don’t you check it
for yourself?).
1
∞ p
p
x p
= xn
n=1
∞
x
p
By definition it follows that the p-norm is well defined since n is a well
n =1
determined non-negative real number. The triangle inequality is just the
Minkowiski inequality.
***
Remark 7: Another important fact is that all these sequences spaces are
infinite dimensional spaces since the collection (e1 , e2 ,....en ,...) with
en = (0,0,...,1,0,0...), where 1 appears at the nth place is a countably infinite
linearly independent set.
29
Block 1 Normed Linear Spaces
Example 9: Let X be the linear space of all polynomials in one variable t with
scalar coefficients. For p ∈ X with p (t ) = a0 + a1t + ... + a n t n , define
p = sup{ p (t ) : 0 ≤ t ≤ 1}, p 1 = a0 + a1 + ... + an .
Solution: It can be seen that ⋅ is the sup norm on C [0,1] and that X is a
subspace of C[0,1] . An element p of X is uniquely determined by the finite
sequence {a j } of its coefficients, and p 1 is just the l1 norm of this sequence.
Thus, we see that ⋅ and ⋅ 1 are norms on X .
p (t ) ≤ a0 + a1 t + ... + a n t n ≤ a0 + a1 + ... + an = p 1.
p = sup{ p (t ) : 0 ≤ t ≤ 1} = 1.
***
Let m denote Lebesgue measure. If 1 ≤ p < ∞, then the set of all measurable
p
functions f with f dm < ∞ is denoted by Lp . These are known as
Lebesgue spaces for 1 ≤ p < ∞. We have seen that Lp is a normed linear
space with the norm ⋅ p
defined by ⋅ p
= ( g p
dm )
1/ p
.
30
Unit 1 Inner Products and Norms
For 1 < p < ∞, the triangle inequality is proved using Holder’s inequality,
namely
fg dm = ( f p
dm ) ( f
1/ p q
dm )
1/ q
1 1
where + = 1. The case p = q = 2 of the Holder’s inequality is called
p q
Schwarz inequality.
f ∞
= ess sup ( f )
Remark 8: Consider x = (1,0,0,...) y = (0,1,0,...), then for 1 ≤ p < ∞, p ≠ 2, Recall that ' a . e ' is
1/ p 1/ p the abbreviation for
x p
= 1, y p
= 1, x + y p
=2 and x − y p
=2 , showing that the almost everywhere
parallelogram law is not satisfied. [Refer Block-3,
MMT-004]
Thus, for p ≠ 2 , the l p spaces are just normed linear spaces and not inner
product spaces, meaning, we have an infinite collection of normed linear
spaces which are not inner product spaces.
As mentioned earlier, the norm induces a metric and hence we can talk about
completeness of normed linear spaces. We will discuss this in detail in the
next block.
(Remember, an inner product induces a norm and this norm in turn induces a
metric).
1/ p
n
= x j
p
E10) Let 0 < p < 1 and x , x ∈ K n . Show that . is not a
p
j
norm on K n if n ≥ 2.
We have seen some examples of normed linear spaces and inner product
spaces. In the next section you will see many more examples.
1.4 SUBSPACES
Suppose we are asked the following questions:
(i) Given a normed linear space, can we in some way find simpler ways of
studying the properties of the space compared with studying the
properties on the whole space as such.
(ii) Given a normed linear space, can we get new spaces from the space?
x x
Let y = . Then yi = i for all i = 1, 2,...
x p x p
1/ p
∞ p
Now, x p = x j ≥ xi for all i = 1, 2, ...
j =1
x
So, yi = i ≤ 1 for all i = 1, 2, ...
x p
r p
yi ≤ yi [As,1 ≤ p < r < ∞]
32
Unit 1 Inner Products and Norms
∞ ∞
yi ≤ yi
r p
i =1 i =1
( y r
) ≤( y )
r
p
p
x
y( r
)
r
≤ (1) p As y p
=
p
= 1
x p
x
y( r
)
r
≤1 y r
≤ 1
x
r
≤1
p
x r
≤ x p
for 1 ≤ p < r < ∞.
So, xn − x r → 0 as n → ∞. Hence, xn → x in l r .
Let us now give some important definitions and prove some more important
theorems related to subspaces.
c00 ⊂ p
+l ⊂ c ⊂ c ⊂ l∞ ,
+ 0 + +
where all the inclusions are proper. Also, the above inclusion says, under the
sup norm each member can be considered as induced normed linear sub
space of the next member.
Recall that you are familiar with the function called the distance function which
is denoted by ‘dist ( A, B)' -distance between two subsets A and B of a metric
space.
dist ( A, B ) = inf { a − b : a ∈ A, b ∈ B}
xr = 1 and r ≤ dist ( xr , Y ) ≤ 1
inf { x − y : y ∈ Y } = 0.
1
x − yn < . (6)
n
x − y0 ≤ dist ( x, Y ) r .
This is because if
x − y 0 > dist ( x, Y ) / r ∀y ∈ Y ,
1 ≥ 1 r ≥ 1, contradiction!
r
34
Unit 1 Inner Products and Norms
Thus for any x ∈ X , x ∉ Y , we can get a corresponding y1 in Y such
that x − y1 ≤ dist ( x, Y ) r . (7)
x − y1
Set xr = . Then x r = 1.
x − y1
x − y1
Also dist ( x r , Y ) = dist , Y
x− y
1
dist ( x − y1 , Y )
= ( by definition of infimum)
x − y1
1
= inf { (x − y1 ) − y : y ∈ y}
x − y1
1
= inf { x − ( y1 + y ) : y ∈ Y } (since y1 , y ∈ Y and Y is a subspace,
x − y1
y1 + y ∈ Y and also as y varies over all
1
= inf { x − y : y ∈ Y } of Y, y1 + Y , also varies in overall of Y )
x − y1
1
= . dist ( x,Y )
x − y1
≥r by (7) above.
r ≤ dist ( x r , Y ) ≤ 1.
***
1
Remark 8: Let n ∈ N n ≠ 1, Then 0 < 1 − <1
n
35
Block 1 Normed Linear Spaces
vector on the unit sphere of X) such that the distance between Y and that
vector is close to 1.
Before coming to one of the most important results in normed space, we will
prove the following two results.
Suppose k ≠ 0 .
y
Then kx + y = k x +
k
≥ k dist ( x, Y ) since y / k ∈ Y .
Proof: Since Y is finite dimensional, it has a finite basis say {y1 ,..., y n }.
Then x n − x m = k n y1 − k m y1 = k n − k m y1
Thus k n − k m → 0 as n, m → ∞ in K.
Hence {k n } is convergent in K.
Suppose k n → k in K.
Set x = ky 1 . Then x ∈ Y .
36
Unit 1 Inner Products and Norms
Also xn − x = k n y1 − ky1
= k n − k y1 → 0 as n → ∞ .
{xn } is convergent in Y
xm = k m y1 + k m2 y 2 + ... + k mn y n , m = 1,2,....
xm − x p = k m y1 + z m − (k p y1 + z p )
= (k m − k p )y1 + (z m − z p )
≥ km − k p dist ( y1 , Z )
Hence k m − k p ≤ xm − x p dist ( y1 , Z )
Suppose k n → k . (10)
Also z m − z p = (xm − k m y1 ) − (x p − k p y1 )
≤ x m − x p + k m y1 − k p y1
37
Block 1 Normed Linear Spaces
= xm − x p + k m − k p y1
Thus {z m } is Cauchy in Z.
Suppose z m → z in Z . (11)
Then x ∈ Y .
Also x m − x = k m y1 + z m − ky1 − z
≤ k m y1 − ky1 + z m − z
= km − k y1 + z m − z → as m → ∞.
(ii) { }
The subset x ∈ X : x ≤ 1 of X is compact (equivalently the closed
unit ball in X is compact).
(The equivalence of (ii) and (iii) can be stated as: A normed space X
is finite dimensional if and only if the closed unit ball in X is compact).
{ }
The subset x ∈ X : x ≤ 1 is both closed and bounded and hence by (i), it is
compact.
(ii) (iii) Suppose X is not finite dimensional. Then we can find an infinite
linearly independent set in X . Suppose {z1 , z 2 ,..., z n ,...} is an infinite linearly
independent set in X .
38
Unit 1 Inner Products and Norms
Then each Z n is a finite dimensional subspace of X and hence by Theorem 6
is closed in X .
1 1
dist ( xn , Z n ) ≥ taking r =
2 2
1
x n − xi ≥ , i = 1,...n − 1 (since x1 ,..., x n−1 ∈ Z n )
2
1
xm − xn ≥
2
1
Since x m − x n ≥ , ∀ m, n, m ≠ n, this sequence cannot have a convergent
2
{
subsequence. Since xn = 1 ∀ n, xn ∈ x ∈ X : x ≤ 1 ∀ n and thus in }
{x ∈ X : x ≤ 1}we have produced a sequence which has no convergent
{ }
subsequence. But this means the set x ∈ X : x ≤ 1 is not compact. Thus: X
{ }
is not finite dimensional x ∈ X : x ≤ 1 is not compact.
(
Thus km, j y j + k m,1 y1 + ... + k m, j −1 y j −1 + km, j +1 y j +1 + ... + k m, n yn )
≥ k m, j dist ( y j , Z j ), j = 1....
M
≤ , m = 1,2,...
dist ( y j , Z j )
The sequence k m , j { } ∞
m=1
is a bounded sequence in K for j = 1,2,..., n.
Set xm p = k m p ,1 y1 + ... + k m p , n y n .
Then by construction, x m p { } ∞
p =1
is a subsequence of {x m }, and hence lies in E.
Set x = k1 y1 + ... + k n y n .
k m p ,1 − k1 y1 + ... + k m p , n − k n yn
→ 0 as p → ∞, since k m p , j → k j , i = 1,..., n
Proof: Let x ∈ X and x1 ∈ E1 . Since E1 is open, there is some r > 0 such that
U ( x1 , r ) ⊂ E1 . But U ( x1 + x, r ) = U ( x1 , r ) + x ⊂ E1 + x.
Let x ∈ U ( o , r ) x ≤ r .
Consider x + a
x + a − a = x ≤ r x + a ∈ U (a, r )
x = x + a − a ∈ U ( a , r ) − a.
Now let x ∈ U ( a , r ) − a. 41
Block 1 Normed Linear Spaces
x = y − a, y ∈ U ( a , r ).
y ∈ U (a , r ) y − a ≤ r x ≤ r x ∈ U (o, r )
Given a vector space V If X is a non-trivial vector space and a is any non-zero vector in X , then
over a field K , the span 2
[a] – the span of a, is always a subspace of X . In R itself we have an
of a set S of vectors (not
necessarily infinite) is infinite collection of subspaces.
defined to be the
intersection W of all Now suppose X is a normed space and Y is a non-trivial subspace of X. As
subspaces of V that a group X is abelian and so is Y and hence the quotient
contains S . W is X / Y = {x + Y : x ∈ Y } is also an abelian group.
referred to as the
subspace spaned by S Now, for α ∈ K and x + Y ∈ X / Y , if we define α ( x + Y ) = αx + Y , then
or by the vectors in S . α ( x + Y ) ∈ X / Y . This is well defined and X / Y become a vector space over
K.
42
Unit 1 Inner Products and Norms
For x + Y ∈ X / Y ,
x + Y = inf { x + y : y ∈ Y }. (15)
Proof: Since x + y ≥ 0, ∀ y ∈ Y ,
x +Y ≥ 0 ∀ x +Y ∈ X /Y
{
Hence 0 = inf x + y : y ∈ Y . }
since 0 ∈ Y , Y is a
= 0+0 =0
subspace.)
Thus x + Y = 0 x + Y = 0
Conversely suppose
x +Y = 0.
x + Y = 0 inf { x + y : y ∈ Y } = 0
But x + Y is zero in X / Y ⇔ x ∈ Y .
{
By definition of infimum, inf x + y : y ∈ Y = 0 .}
1
∀ n ∈ N , ∃ y n ∈ Y such that x + y n < .
n
Hence x + Y = 0 ⇔ x + Y = 0.
Here what we observe is as y varies over the whole of Y , αy also varies over
the whole of Y if α ≠ 0. Hence
[As, Y is a subspace on K and
α(x + Y ) = αx + Y = inf { αx + y : y ∈ Y }
α ∈ K , we can write the
general element as αy where
= inf { αx + αy : y ∈ Y } α ≠ 0 and y ∈ Y ]
= inf {α x + y : y ∈ Y }
= α inf { x + y : y ∈ Y }
= α x+ y
Let x + Y , y + Y ∈ X / Y
( x + Y ) + ( y + Y ) = ( x + y) + Y
= inf { ( x + y ) + z : z ∈ Y } (15)
Let ε > 0.
x + y1 ≤ x + Y + ε / 2 and
∃ y 2 ∈ Y such that
y + y 2 ≤ y + Y + ε / 2.
Hence ( x + Y ) + ( y + Y )
= ( x + y ) + Y = inf { ( x + y ) + z : ε z Y }
≤ ( x + y ) + y1 + y 2
≤ ( x + y1 ) + ( y + y 2 )
44
Unit 1 Inner Products and Norms
≤ x + y1 + y + y 2
≤ x + Y + ε + y + Y + ε / 2.
2
= x + Y + y + Y + ε.
since ԑ is arbitrary.
(x + Y ) + ( y + Y ) ≤ x + Y + y + Y .
Thus the triangle inequality is also satisfied. All these prove the quotient norm
is in fact a norm.
For an example of a quotient space, let us consider the simple question: for
m > n, can we consider the quotient space R m / R n ?
n m n
First of all, how do we consider R as a subspace of R , since R consists
m
of n-tuples and R consists of m-tuples.
n n
By adjoining (m−n) zeroes to the elements of R , we can consider R as a
m
subspace of R .
{
R m / R n = ( x1 ,..., xm ) + R n : ( x1 ,..., x m) ∈ R m }
Suppose ( x1 ,..., xm ) ∈ R m and ( y1 ,..., yn ) ∈ R n ,
( x1 + y1,..., xn + yn , xn +1,..., xm )
n
( x1 ,..., xn ,0,...,0) + (0,...0, xn +1 ,..., x m ) + R .
(m−n) zeros n zeros
= R n + ( 0,...0, xn +1 ,..., xm ) + R n
n
But R is the zero element in R m / R n .
45
Block 1 Normed Linear Spaces
n
Ultimately the coset ( x1 ,..., xm ) + R can be identified with the coset
n m -n
(0,...0, x n +1 ,..., x m ) + R . Here ( xn+1 ,..., xm ) varies in R .
n
Hence, identifying the coset (0,...0, xn +1 ,..., xm ) + R with element
( xn +1 ,..., xm ) in R m − n , we can identify the quotient space R m R n with the
space R m − n . Hence for m > n, R m / R n is a quotient space.
f = sup{ f ( x ) : x ∈ T }
This norm is called the sup norm on B (T ) , it satisfies all the properties of a
norm.
Since complete subspaces are closed we can talk about the quotient space
B(T ) C(T ) . Giving the quotient norm on B(T ) C(T ) , B(T ) C(T ) becomes a
normed linear space.
Remember: In general there are bounded functions which are not continuous.
Now let us ask: under what conditions the quotient space will also become
complete.
46
Unit 1 Inner Products and Norms
We will answer this question a little later. However we now look at an
important property about convergence of sequences in the quotient space
X Y.
Let ε > 0 .
x n + Y → x + y for the ε, ∃ N ∈ N
(x n − x ) + Y ≤ ε , ∀ n ≥ N.
2
(xn − x ) + y n ≤ ( xn − x) + Y + ε 2 .
(xn − x ) + y n ≤ ( xn − x) + Y + ε 2 ∀ n
< ε 2 + ε 2 ∀n ≥ N .
∃ (x n − x ) + y n < ε ∀ n ≥ N .
Let ε > 0 .
∃ n ∈ N such that
(x n + y n ) − x < ε , ∀ n ≥ N.
( x n − x ) + y n < ε , ∀ n ≥ N.
(x n − x ) + Y ≤ (x n − x ) + y n ∀ n
47
Block 1 Normed Linear Spaces
< ε ∀n ≥ N .
(x n + Y ) − ( x + Y ) = ( x n − x ) + Y < ε ∀ n > N .
x n + Y → x + Y in X Y .
With this we come to the end of this unit. Let us summarise the points covered
in this unit.
1.6 SUMMARY
In this unit we have covered the following points:
6. Stated and proved the result that the quotient space X / Y is complete if
and only if Y is complete.
1.7 SOLUTIONS/ANSWERS
E1) For X = ( x1 , x2 , x3 ) ∈ R3 , define the function . , given by
[
x = ( x12 + x22 ) 3 / 2 + x3 ]
3 1/ 3
, x∈ X.
Where x = ( x1 , x 2 , x 3 ), y = ( y1 , y 2 , y 3 )
We have x + y = ( x1 + y1 , x 2 + y 2 , x 3 + y 3 ) . Also
[
x + y == (( x1 + y1 ) 2 + ( x2 + y 2 ) 2 )3 / 2 + ( x3 + y3 ) 2 ]1/ 2
1/ 2
2
2
((x 1 + y1 ) + ( x 2 + y 2 ) ) 2 1/ 2
= (x i + yi ) 2
i =1
1/ 2 1/ 2
2 2
≤ (x i ) 2 + ( yi ) 2
i=1 i =1
= ( x 12 + x 22 )1 / 2 + ( y12 + y 22 )1/ 2
1/ 3 1/ 3 1/ 3
2 3 2 3 2 3
ai + bi ≤ ai + bi
i =1 i =1 i=1
Then we have
[
x + y ≤ (( x12 + x22 )1 / 2 + ( y12 + y 22 )1 / 2 ) 3 + x3 + y3 ]
3 1/ 3
3 3
≤ ((x 12 + x 22 ) 3 / 2 + x 3 )1 / 3 + (( y12 + y 22 ) 3 / 2 + y 3 )1 / 2
= x + y .
E2) For x , y ∈ X , d ( x , y ) := x − y
(i) x = y x − y = 0 d x , y = 0.
Suppose d x , y = 0 .
x− y =0 x= y
49
Block 1 Normed Linear Spaces
(ii) d ( y, x) = y − x = − ( x − y )
= −1 x − y = x − y
= d ( x, y ), ∀x, y ∈ X .
(iii) For x , y , z ∈ X ,
d ( x, y ) = x − y = x − z + z − y
≤ x−z + z− y
= d ( x, z ) + d ( z , y ).
Therefore d is a metric.
x
2
To check for addition, we have to show that n + y n is convergent
n
2
for x = {xn } and y = { y n } in l . We have
x + y n ≤ x n + y n + 2 Re xn y n
2 2 2
n
n n n
yn 1/ 2
∞ 2
∞
2 (by
≤ xn + y n + 2 xn y n
2 2
Cauchy-
n n n=1 n=1 Schwarz
2
Hence x + y ∈ l inequality)
αx = α xn
2 2
n
n
Hence αx ∈ l 2 .
E4) We first have to show that the map < , >: l 2 × l 2 → K, defined by
∞
< x, y > = xn y n is well defined where x = { x n }n∞=1 and y = { y n }∞n =1 .
n =1
∞
To show that we have to check that the infinite series x
n =1
n yn
50
Unit 1 Inner Products and Norms
n n
x
k =1
k yk = xk yk
k =1
≤ ( x ) ( y ) k
2 y2
k
2 y2
dfd
y2 y2
∞ 2
∞
2
≤ xk y k
k =1 k =1
<∞ since x, y ∈ l 2 .
Next we shall verify the three properties for inner product space. We
note that
∞
x, x = xn xn ≥ 0
n =1
∞
x, y = x n y n = y , x
n =1
∞
αx + β y , z = α ( xn + y n )z n where z = {z n }
n =1
∞
= [α ( xn ) z n + β( y n ) z n ]
n =1
∞ ∞
= α xn z n + β y n z n
n =1 n =1
= α x, z + β y , z
E6) We will show that if x ∈ l r for some r ≥ 1, then x ∈ l p for all p > r.
x
r
Let x ∈ l r , then n converges where x = {xn }. That means xn < 1
n
p r
xn ≤ xn for n ≥ n0
51
Block 1 Normed Linear Spaces
x
p
This implies that n < ∞. Hence x ∈ l p .
E7) The properties (i) and (ii) for norm is automatically satisfied. To check
the triangle inequality, note that
x + y = sup x n + y n , x = {xn }, y = { y n }
n
≤ x + y.
E8) By definition
x p
(
= x1p + x2p )1/ p
where x = ( x1 , x 2 )
Here x1 = 2, x2 = 3
For p = 1, x 1 = 5
For p = 2, x 2
(
= 2 2 + 322 )
1/ 2
= 13
For p = 4, x 4
= (2 4 + 34 )1 / 2 = 97
For p = ∞, x ∞
= max {2, 3} = 3 .
1/ p 1/ p
1 2 p 1 2p
E9) x p
= (t ) dt = t dt
0 0
1/ p
t 2 p +1 1
=
2 p + 1
0
1/ p
1
=
2 p +1
1
=
(2 p + 1)1 / p
→ 1 = x ∞ as p → ∞ .
Hence,
xp+ y p
= 2 < 21/ p = x + y p ,
E11) We shall show that it is not necessary. For example Take X = K and
consider the set
1
E1 = n + : n ∈ N , E2 = {− n : n ∈ N}.
n
These sets have no limit points. Hence they are closed. Since
1 1
p+ = n + p + + (− n) ∈ E1 + E2 , for all p, n ≥ 1
n+ p n + p
1
But lim p + = p i.e.
n→∞ n+ p
1
p = n+ + ( − m)
n
i.e. np = n 2 + 1 − mn
i.e. np + mn − n 2 = 1
1
p = n + + (−m),
n
yni = z ni − xni → z − x.
54
UNIT 2
Structure Page No
2.1 Introduction
Objectives
2.2 Characterisation of Continuity of Linear Maps
2.3 Spaces of Bounded Linear Operators
2.4 Summary
2.5 Hints/Solutions
2.1 INTRODUCTION
In applications, it is very important to know whether a given linear operator is
continuous or not. For instance, suppose a physical problem is modelled in the
form of an operator equation
Ax = y,
where A : X → Y is linear operator between normed linear spaces X and Y .
The problem may be to find x ∈ X from the knowledge of y ∈ Y , or to find
y ∈ Y from the knowledge of x ∈ X . The given information is often known as
‘data’ and the information which is to be obtained is known as a ‘solution’. The
questions of existence and uniqueness of a solution are purely set-theoretic or
algebraic ones. But, in applications, exact data are not usually available. Also,
in numerical approximation procedures, one normally considers approximate
data. In such situations, ‘stability’ of the solution is of great importance. It
amounts to knowing whether the operator A or its inverse A−1 is continuous.
In this unit our main study is about continuous linear operators between
normed linear spaces and their various properties. We shall also discuss
briefly a class of discontinuous operators which are closely related to
continuous operators, namely, the class of closed operators.
Objectives
The objectives of this unit are:
• To characterise continuity of linear maps on normed spaces;
Block 1 Normed Linear Spaces
• To identify continuous linear maps on standard spaces;
Before proceeding to discuss about continuity, let us briefly recall some ideas
on linear maps.
For
α, β ∈ F , x1 , x2 ∈ X ,
T (α x1 + β x2 ) = αT ( x1 ) + βT ( x2 ).
The understanding is that the linear maps can be used to classify vector
spaces. As a simple example, any two n-dimensional spaces are isomorphic.
What we are now interested in is the study of linear maps between normed
spaces.
In the case of normed spaces, in addition to being vector spaces, they also
have metric structures. When we have additional structures, it is natural to
investigate how maps behave with respect to all these structures.
Important classes of maps that are usually discussed reflecting the metric or
topological structures are the continuous maps.
Consider T ( x m ) − T ( x ) .
As m → ∞, k m, j → k j , j = 1,..., n.
i.e. T ( x m ) → T ( x) in Y as m → ∞.
Thus, whenever x m → x in X , T ( x m ) → T ( x ) in Y .
It is natural to ask whether this is true for infinite dimensional spaces. The
following proposition shows that this is not true, in general.
57
Block 1 Normed Linear Spaces
Proof: Since X is infinite dimensional, we can find an infinite linearly
independent subset {e n } of X .
en
Let xn = for n = 1,2,...
n . en
b if x∈L
T ( x) =
0 if x ∈ B , x ∉ L.
Then T is defined on all the basis vectors of X and thus T defines a linear
map on X .
Here the sequence {x n } is such that
xn → 0 as n → ∞.
But T ( x n ) = b ≠ 0 ∀n.
What is so special about linear maps which are continuous from one normed
space to another?
Proof: First we prove that conditions (i) and (v) are equivalent.
Suppose (i) holds.
58
Unit 2 Continuous Linear Maps and Functionals
T is bounded on U ( 0 , r ) for same r > 0.
Thus ∃ K > 0 such that
T ( x) ≤ K ∀ x ∈U (0, r )
rx
If x ≠ 0. Set y = . Then y = r .
x
Hence by assumption T ( y ) ≤ K .
rx
T ≤ K .
x
r
T ( x) ≤ K .
x
K
T ( x) ≤ x.
r
Taking α = K / r , we see α > 0 &
T ( x ) ≤ α x , for all x ∈ X
T ( x ) ≤ α ∀ x ∈U (0,1).
T is bounded on U ( 0,1).
(v) (iv).
Let x1 , x2 ,∈ X be arbitrary.
Let ε > 0.
By (v) ∃α > 0 = T ( x ) ≤ α x ∀ x ∈ X .
By this,
T ( x1 ) − T ( x2 ) = T ( x1 − x2 ) ≤ α x1 − x 2 .
T ( x1 ) − T ( x2 ) ≤ α x1 − x2
< α. ε = ε.
α
Hence given ε < 0, ∃ δ > 0 such that
59
Block 1 Normed Linear Spaces
T ( x1 ) − T ( x2 ) ≤ ε whenever x1 − x2 < δ.
Thus T is uniformly continuous.
(iv) (iii).
Since any uniformly continuous function is continuous, T is continuous on X .
(iii) (ii).
Since continuity on X continuity at 0, T is continuous at 0.
(ii) (i).
Suppose T is continuous at 0.
Then corresponding to 1, ∃ an r0 > 0 such that T ( x ) − T ( 0) < 1 whenever
r0
x − 0 < r0 . Taking r = .
2
We get T ( x ) < 1 whenever x ≤ r .
T is bounded on U ( 0 , r ).
Suppose F is continuous on X .
= T −1 ({0}).
But {0} is closed in Y and T is continuous T −1 ({0}) is closed in X i.e.
Z (T ) is closed in X .
Thus the quotient space X / Z ( F ) is a normed space under the quotient norm.
= T ( x + z)
since T is continuous.
This is true for every z ∈ Z (T ).
~
Hence T ( x + Z (T ) ) ≤ α x + z ∀z ∈ Z (T )
60
Unit 2 Continuous Linear Maps and Functionals
~
∴ T ( x + Z ( f ) ≤ α.inf { x + z : z ∈ Z (T )}
= α. x + Z (T )
~
Hence, by the above equivalent conditions T is continuous.
~
Conversely suppose Z (T ) is closed and T defined as above is continuous.
≤α x.
This is true ∀ x ∈ X .
Hence T is continuous.
β x ≤ F ( x) ≤ α x ∀ x ∈ X .
T ( x) ≤ α x ∀ x ∈ X and
T −1 ( y) ≤ β1 y ∀ y ∈ Y . (2)
T −1 (T ( x)) ≤ β1 T ( x) ∀ x ∈ X .
i.e. x ≤ β1 T ( x) , ∀ x ∈ X . (3)
Conversely, suppose the condition holds, i.e. ∃ scalars α, β > 0 such that
β x ≤ T ( x) ≤ α x , x ∈ X .
x = 0. Thus T is 1−1.
By Theorem 2, F is continuous.
Now consider F −1 : R ( F ) → X .
F −1 ( y ) = F −1 (F ( x) )
1
= x ≤ F ( x) (by assumption)
β
1
= y.
β
1
i.e. F −1 ( y) ≤ y , ∀ y ∈ R( F ).
β
−1
F is continuous.
Thus F : X → Y is injective and continuous and F −1 : R ( F ) → X is also
continuous and hence F is a homeomorphism.
β x ≤ T ( x) ≤ α x , x ∈ X . (4)
1
By (4), x n − x m ≤ T ( xn − xm )
β
1 1
= T ( xn ) − T ( x m ) = y n − y m
β β
62
Unit 2 Continuous Linear Maps and Functionals
→ 0, as n, m → ∞.
{ x n } is cauchy in X .
x n → x in X . (5)
Let y = T ( x).
Then, y n − y = T ( x n ) − T ( x )
= T ( xn − x)
Hence Y is complete.
Now reversing the roles of X and Y and using the function T −1 and inequality
(4) again, we can prove Y is complete X is complete. (Try yourself !)
In Theorem 1, we have shown that every linear map from a finite dimensional
normed space to any other normed space is continuous.
We give a definition.
1
Definition 1: Suppose . , . are norms on a linear space X . Then these
norms are said to be equivalent if ∃ scalars α, β > 0 such that
1
β x ≤ x ≤ α x , ∀ x ∈ X.
Note that equivalent norms induce the same topology, the identity map is a
homeomorphism.
I : ( X , . ) → X , . .
′
This is a bijective linear map and X is a finite dimensional I and I −1 are
continuous, i.e., I : (X , . ) → X , . is a homeomorphism.
′
Hence by Proposition 2, ∃ scalars α, β > 0 such that
1
β x ≤ I ( x) ≤ α x , ∀ x ∈ X . (6)
Remark: Recall if X is a linear space over a field F , then a nonzero linear map
f : X → F is called a linear functional, i.e. a linear map from a linear space
to its base field is called a linear functional.
By assumption R(T ) is finite dimensional and hence has a finite basis, say
64 { y1 ,..., y n }. Since yi ∈ R(T ), ∃ x ∈ X such that yi = T ( xi ), i = 1,..., n.
Unit 2 Continuous Linear Maps and Functionals
Claim: Span {x1 + Z (T ),..., xn + Z (T )} = X .
Z (T )
T ( x) = k1 y1 + ... + k n yn , ki ∈ K, i = 1,..., n.
= k1T ( x1 ) + ... + k nT ( xn ).
= T ( k1 x1 + ... + k n xn ).
x + Z (T ) = ( k1 x1 + ... + k n xn ) + Z (T ).
= k1 ( x1 + Z (T ) ) + ... + k n (xn + Z (T ) )
~ ~ ~
So, if we define T : X / Z (T ) → Y as T (x + Z (T ) ) = T ( x), x ∈ X , then T is will
defined and continuous. Hence by Theorem 2, T is continuous.
n m
Then M is a linear map from K → K .
Since both are finite dimensional, we have two facts: Any linear map is
continuous and all norms are equivalent.
n m
Thus M : K → K is continuous under any norm on K n and K m .
***
p p p
Example 3: Let X = l ,1 ≤ p ≤ ∞ , define the mapping T : l → l as follows.
If ( x1 , x2 ,..., x n ,...) ∈ l p , define
= ( x1 , x2 ,..., xn ,...) p .
i.e. we have
T ( x) p
= x p , ∀ x ∈l p .
p
By Theorem 2, T is a continuous linear map on l . This T is called the right
shift operator.
***
Remark 2: Note that by giving specific value for p , we get a specific example
5
i.e. if we take p = 5, we get a particular example on the space l .
Then xn ∞
= 1 ∀n (. ∞
is the sup norm )
1 n −1
xn (t ) = nt , ∀n,
x1n = n, ∀n
∞
Example 6: Let X = c00 , the space of all scalar sequences having only finitely
many non-zero terms.
Consider the space with the one-norm, i.e. if x = ( x1 , x2 ,...) ∈ c00 , then
x 1 = x(1) + x ( 2) + ...
(note, since there are only finitely many non-zero terms, the sum on the right
side is a finite sum.)
Define T0 : c00 → K as
Also T0 ( x) = x1 + x2 + ...
∞
≤ xn = x 1 .
n =1
T0 is continuous.
67
Block 1 Normed Linear Spaces
n ∞ 2
1 1 π
= ≤ 2 = .
2
xn 2 2
j =1 j n =1 n 6
n
1
Now T0 ( xn ) = .
j =1 j
n
1
T0 ( x n ) = → ∞ as n → ∞.
j =1 j
E1) Consider the space c00 . For x = ( x1 , x 2 ... xn ,...) ∈ c00 define
∞
T ( x ) = xn . Then show that T is a linear functional which is not
n =1
In the next section we consider the set of all bounded linear maps from one
normed linear space to another.
68
Unit 2 Continuous Linear Maps and Functionals
the set BL( X , K) the set of all bounded linear functionals on X as X ′,
called the dual space of X .
By Theorem 2, the first condition is equivalent to the fifth condition.
T ( x) ≤ α x , ∀ x ∈ X .
Many a time, we check this inequality to check the boundedness (= continuity)
of a linear map between normed spaces.
Note the difference between the boundedness of (ordinary) maps and those of
linear maps.
such that f (t ) ≤ M ∀ t.
T ( x) ≤ M ∀ x ∈ X .
You have seen that the set of all linear maps from a vector space X to a vector
spaces, i.e. L( X , Y ), is again a vector space under addition and scalar
Since normed spaces are basically vector spaces, the set of all linear maps is
also a vector space.
69
Block 1 Normed Linear Spaces
But here what we are interested in is: if X , Y are normed spaces, is BL( X , Y )
also a normed linear space?
Suppose we define addition as above. We have to check fist, whether it is
meaningful i.e., if, T , S ∈ BL( X , Y ), is T + S defined as above also in
BL( X , Y ) .
Similarly for δ ∈ K, T ∈ BL( X , Y ), is δT in BL( X , Y ) .
S ( x) ≤ β x ∀ x ∈ X .
Consider (T + S ) x = T ( x) + S ( x)
≤ T ( x ) + S ( x)
≤ α x + β x = (α + β ) x ∀ x ∈ X .
T + S ∈ BL( X ,Y ).
Similarly if δ ∈ K, T ∈ BL( X , Y ).
(δT ) ( x) = δ(T ( x ) ) = δ T ( x )
≤ δα x ∀x∈ X.
δT is bounded δT ∈ BL( X , Y ).
Also ‘+’ gives an abelian group structure on BL( X , Y ) and scalar multiplication
satisfies all the properties of a vector space. Thus BL( X , Y ) is a vector space
over K.
Recall if X is n-dimensional and Y is m-dimensional, then L( X , Y ) is mn-
dimensional. Hence if X is n-dimensional normed linear space and Y is m-
dimensional, then BL( X , Y ) also is an mn-dimensional space. Since
BL ( X , Y ) = L ( X , Y ) in this case.
Now we will see how we can make BL( X , Y ) into a normed linear space.
Again remember bounded linear maps take bounded sets onto bounded sets.
T = sup{ T ( x) : x ∈ X , x ≤ 1}.
70
Unit 2 Continuous Linear Maps and Functionals
Hence {T ( x) : x ∈ X : x ≤ 1}is a bounded set in Y which means
{ T ( x) : x ∈ X , x ≤ 1}
is a bounded set of non-negative reals and hence the
sup{ T ( x) : x ∈ X , x ≤ 1}is well defined.
Now we shall check the properties (i), (ii) and (iii) of a norm.
i) We note that if X = {0}, there is nothing to prove.
Conversely suppose T = 0.
T ( x) = 0, ∀ x ∈ X with x ≤ 1.
x x x
Let x ≠ 0. Consider . Then = = 1.
x x x
x
Hence T
x = 0.
1
T ( x ) = 0.
x
T ( x ) = 0. T ( x ) = 0
Thus T ( x) = 0 ∀x ∈ X .
T = 0.
(ii) Let α ∈ K and T ∈ BL( X , Y ).
Consider α T = sup{ α T ( x) : x ∈ X , x ≤ 1}
= sup{ α T ( x) : x ∈ X , x ≤ 1}
= α sup{ T ( x ) : x ∈ X , x ≤ 1}
= α T.
(T + S )( x ) , T ( x) + S ( x) ≤ T ( x) + S ( x )
Hence
71
Block 1 Normed Linear Spaces
T + S = sup { (T + S )( x) : x ∈ X , x ≤ 1}
= sup { T ( x) + S ( x ) : x ∈ X , x ≤ 1}
≤ sup { T ( x) + S ( x ) : x ∈ X , x ≤ 1}
≤ sup { T ( x) : x ∈ X , x ≤ 1}
+ sup { S ( x ) : x ∈ X , x ≤ 1}
= T + S.
x
Hence T
x ≤ sup { T ( y) : y ∈ X , y ≤ 1}.
= T.
1
T ( x) ≤ T .
x
T ( x) ≤ T x.
Hence the claim.
Remark 4: For bounded linear maps, this inequality above is of great practical
use.
T = sup { T ( x ) : x ∈ X , x = 1}
= sup { T ( x) : x ∈ X , x < 1}
β = sup { T ( x) : x ∈ X , x = 1}
r = sup { T ( x) : x ∈ X , x < 1}
Since {x : x = 1} ⊂ {x : x ≤ 1},
β≤ T (7)
72
Unit 2 Continuous Linear Maps and Functionals
since {x : x < 1} ⊂ {x ∈ x ≤ 1},
r≤ T. (8)
tx x x
T ( x) = T ≤ sup{ T ( z ) : z ∈ X , z = t}
x t t
If t = 1, then T ( x ) ≤ sup{ T ( z ) : z ∈ X , z = 1} x
=β x.
Thus T ( x) ≤ β x , ∀ x ∈ X .
α0 ≤ β (9)
T ( x) ≤ r x , ∀ x ∈ X .
α 0 ≤ r. (10)
Claim: T ≤ α0.
Hence T = α 0 = β.
Hence T = α 0 = r .
Thus we get
T = α 0 = β = r.
≤ sup{α x : x ∈ X , x ≤ 1} = α
73
Block 1 Normed Linear Spaces
T ≤ α. (11)
T ≤ α0.
This proves the claim.
Many times it is not possible to get the exact value of a bounded linear map,
however we can get a bound.
1
Consider T( f ) =
0
f (t ) dt
1
≤ sup f (t ) dt
0 t∈[ 0 ,1]
1
= f ∞
. dt = f ∞
∀t ∈ C ([0,1]).
0
f 0 (t ) = 1 ∀t ∈ [0,1].
Then f 0 = 1.
1
Also T ( f 0 ) =
0
f (t ) dt = 1.
T ( f 0 ) = 1.
{
T = sup T ( f ) : f ∈ C ([ 0,1]) : T ∞
}
=1
≥ T ( f 0 ) = 1.
i.e. T ≥ 1. (13)
***
Now recall that a bounded linear map T from a normed linear space X to K is
called a bounded linear functional. We shall see some more examples of
bounded linear functionals and also discuss some properties.
1
Example 8: Consider the linear functional T : L [0,1] → R , f → tf (t )dt. Find
0
T .
1 1
Solution: We have Tf = tf (t ) dt ≤ tf (t ) dt ≤ f
0
0
1
0 for 0 ≤ t ≤ 1 − 1 / n
f n (t ) =
n for 1 − 1 / n < t ≤ 1
1 1
1
Tf n = tf n (t )dt = n tdt = 1 − → 1 as n → ∞,
0 1−1 / n
2n
and
1 1
f n 1 = f n (t ) dt = n dt = 1.
0 1−1 / n
***
T ( y n ) = T ( y n − y + x0 + y − x0 )
= f ( y n − y + x0 ) + f ( y ) − f ( x0 ).
Since y n → y + x0 → x0 ,
T ( y n − y + x0 ) → f ( x0 )
and 75
Block 1 Normed Linear Spaces
T ( y n ) → T ( y ).
Before proceeding, a new definition is necessary.
Definition 2: The linear functional T on the normed linear space X is called
bounded if there exists a real constant k , such that, for all x ∈ X ,
T ( x) ≤ k x (14)
Proof: First we shall show that continuity implies boundedness, and the proof
will be by contradiction; thus, suppose T is continuous but not bounded. The
negation of being bounded is that for any natural number n, however large,
there is some point, xn say, such that
T ( xn ) > n xn .
Consider now the vectors
xn
yn =
n xn
with norms
1
yn = .
n
Clearly, then the sequence
yn → 0
Since any linear functional maps the zero vector into the zero scalar and T is
continuous, this implies
T ( y n ) → f ( 0) = 0
But
1
T ( yn ) = T ( xn )
n xn
and
T ( xn ) > n xn ,
which implies
T ( yn ) > 1
76
Unit 2 Continuous Linear Maps and Functionals
To prove the converse, that boundedness implies continuously, we need only
note that boundedness certainly implies continuity at the origin and apply the
preceeding theorem.
Now you can try the following exercise
With this we come to the end of this unit. Let us summarise the points covered
in this unit.
2.6 HINTS/SOLUTIONS
∞
E1) Since x ∈ c00 , x
n =1
n contains only a finite number of non-zero terms.
1 for 1 ≤ j ≤ n
( xn ) j =
0 for j > n
Then xn ∞
= 1 and T ( xn ) = xn ( j ) = n .
77
Block 1 Normed Linear Spaces
Therefore supT ( x) ≥ n for all n. This shows that T cannot bounded on
u (0, r ) for any r > 0. Hence T is not continuous.
xn ≤ x ∞
T ( x) = lim xn
n
≤ x∞.
E3) To show the 'if ' part we suppose that T is continuous. Let {xn } be a
Cauchy sequence in X . To show that {T ( xn )} is Cauchy. Let ε > 0 be
given. Then
T ( x n ) − T ( xm ) = T ( x n − x m )
≤ α xn − x m
ε
Given > 0 there exist N such that
α
ε
xn − xm < for all n ≥ N .
α
To show the “the” other part, let us suppose that T is not continuous.
Then we show that there exists a {z n } which is Cauchy but {T ( z n )} is
not Cauchy.
78
Unit 2 Continuous Linear Maps and Functionals
xn 1
yn = , zn = yn
xn n
1
Then y n = 1 and z n = → 0. Hence z n → 0, but
n
1 1
T ( zn ) = T ( yn ) = T ( xn ) > n
n n xn
To show that the result is not true for non-linear map, consider the map
1
f : R → R given by f ( x) = , x ≠ 0 and f (0) = 0. Then check whether
x
1
f is linear. For the second part consider the sequence in R. Then
n
1 1
is a Cauchy sequence in R . But f = n and { n } is not Cauchy
n n
in R .
This shows that the result is not true for non-linear maps.
tn 1
E4) Hint: To prove that z n → x1 , note that z n − x1 = sup = → 0. as
n n
n → ∞.
n
f ( zn ) = z′n (t ) t =1 = 1 − t n −1 = 0.
n t =1
But f ( x1 ) = 1 ≠ 0 i.e. x1 ∉ Z ( f ) .
Tf ( x ) ≤ f ( x ) ≤ f ∞
.
Therefore T = 1.
h( y) = kf ( y ) + Tg ( y)
Tf ( x) = k Tf ( x) + Tg ( x).
Tf ( x ) = f ( y ) ≤ f
∴ T is bounded and T ≤ 1.
To find T , we take g = T ( f ).
g ( x + x0 ) ≤ f ( x) ≤ f i.e. g ≤ f
Also f ( x) = g ( x + x0 )
≤ g
Therefore f ≤ g
Thus f = g . Hence T = 1.
80
UNIT 3
HAHN
HAHNBANACH THEOREMS
Structure Page No
3.1 Introduction
Objectives
3.2 Hahn-Banach Separation Theorem
3.3 Hahn-Banach Extension Theorem
3.4 Summary
3.5 Solutions/Answers
3.1 INTRODUCTION
There are two main versions of the Hahn Banach theorems, one is called
Hahn-Banach separation theorem which is geometric in nature and the other
is called Hahn-Banach extension theorem which is analytic in nature. Both the
theorems are concerned with the existence of bounded linear functionals with
special properties. In Sec. 3.2 we discuss Hahn Banach separation theorem
which deals with the separation of two disjoint convex subsets of a normed
space by a closed hyper plane. In Sec. 3.3 we consider Hahn-Banach
extension theorem which ensures the existence of a norm preserving linear
extension of a functional on a subspace of a normed linear space.
The Hahn Banach theorem is a central tool in the study of functional analysis.
Hans Hahn, an Austrian mathematician and Stefan Banach of poland, proved
the result independently in the 1920’s. Giving credit to both, the result is called
the Hahn Banach theorem.
Objectives
The objectives of the unit are:
• to state and prove Hahn-Banach separation theorem
• to state and prove Hahn-Banach extension theorem
• to give examples to show that a Hahn-Banach extension is not unique
• to state conditions under which Hahn-Banach extension are unique.
To understand this we need to recall certain results which we have learnt from
the earlier units. You will be introduced to hyperplanes/hyperspaces and the
connection between these and non-zero bounded linear functionals.
Remark 1: In this unit as well as in other units also, we use the notation ' f '
instead of T for a bounded linear functional.
27th September, 1879 Proposition 1: Let X be a linear space over C . Regarding X as a real linear
to 24th July, 1934 space (i.e., the scalar multiplication is restricted to reals), consider a real linear
functional u : X → R.
Define f : X → C as
f ( x) = u ( x) − iu (ix), x ∈ X
Then f is a complex linear functional on X .
You may find the Remark 2: This result gives a sort of formula for converting a real linear
following link useful:
https://www.youtube.co
functional to a complex linear functional or complexifying a real linear
m/watch?v=FkJMfsNg2 functional.
cM&list=PLyqSpQzTE6
M8mjwWBz0vXpmJQw Proof: First we note that, since X is a linear space over C , if x ∈ X then ix
R46JIdS&index=10&t= also belongs to X . Also, because u is real linear, we cannot write u (ix) as
471s
equal to iu (x).
= α1 [u ( x1 ) − iu (ix1 )] + α 2 [u ( x2 ) − iu (ix2 )]
= α1 f ( x1 ) + α 2 f ( x2 )
Thus f is also real linear.
Now to prove f is complex linear, it is enough to prove f (ix) = if ( x), x ∈ X .
So let x ∈ X , we consider
E1) Prove that f as defined in the above proposition is complex linear i.e. if
α, β ∈ C, then f (αx + βy) = αf ( x) + βf ( y), ∀x, y ∈ X .
Remark 3: If X is one dimensional then the trivial space {0} is the only hyper
space in X .
We also observe that a proper subspace is maximal if and only if the span of
Z ∪ {a} equals X for each a ∉ Z . Further if Z is a hyperspace in X , for each
x ∈ X , ∃ unique z ∈ Z and k ∈ K such that
x = z + ka.
In other words
X = Z ⊕ [a] for any a ∉ Z .
Here [ a ] is 1-dimensional. Thus, Z is a hyperspace in X if Z has co-
dimension 1 in the sense that X is the direct sum of Z and an1-dimensional
space or, equivalently, the quotient X / Z has dimension 1.
f ( x)
Consider z = x − a.
f (a )
Then f ( z ) = 0 z ∈ Z ( f ).
f ( x)
Thus x = z + a ∈ Z ⊕ [a]. s
f ( a )
Thus X ⊆ Z ⊕ [a].
83
Block 1 Normed Linear Spaces
But Z ⊕ [a ] ⊆ X . Hence X = Z ⊕ [a].
This means Z is a maximal proper subspace of X i.e. Z is a hyper space in X
.
Now let g be any linear functional on X such that Z ( g ) = Z ( f ) & f (a) = g (a).
Then
f ( x)
g ( x) = g ( z ) + g (a )
f (a )
= f ( z ) + f ( x) = 0 + f ( x ) = f ( x ).
This means f is completely determined by Z ( f ) and f (a), where a ∉ Z ( f ).
1 (1 − s)
So x1 = y0 − x2
s s
and
1 s
x2 = y0 − x1 .
(1 − s ) (1 − s )
Suppose x ∈ span {Y , x1 }
x = y + kx1 , for some y ∈ Y and k ∈ K.
1 1 − s
= y + k y 0 − x 2
s s
k 1− s
= y + y0 − x2
s s
∈ span {Y , x 2 }.
Thus span {Y , x1 } ⊆ span {Y , x 2 }.
Set tx1 + (1 − t ) x = y, y ∈ Y .
( y − tx1 )
Then x = ∈ span {Y , x1}, a contradiction.
1− t
Hence tx1 + (1 − t ) x ∉ Y ∀ t ∈ (0,1). Since x1 & x ∉ Y , we get
tx1 + (1 − t ) x ∉ Y ∀ t ∈ [0,1].
Similarly since x ∉ span {Y , x2 }, proceeding as above we get tx 2 + (1 − t ) x ∉ Y
for t ∈ [0,1].
This completes the proof.
Remark 6: You may observe that Proposition 4 already gives that Y c is path
connected and hence connected.
Now we prove a ‘big’ theorem which leads to the Hahn Banach Separation
Theorem.
85
Block 1 Normed Linear Spaces
Theorem 1: Let X be a normed space over K and Y be a subspace of X . Let
E be a non-empty open convex subset of X such that E I Y = φ. Then there
exists a closed hyperspace Z in X such that Y ⊂ Z and E I Z = φ.
We prove the theorem in different steps. Each step we call as a claim and
each claim has subclaims the proof of which are given in different steps.
We also prove the theorem first by taking K = R which we call the real part
and then for K = C which we call the complex part.
P I E = ( U F ) I E = U ( F I E ) = Φ
F∈F F∈F
Thus P is a subspace of X such that
Y ⊂ P & P I E = φ P ∈ W.
Also by definition of P, F ≤ P, ∀F ∈ F .
Thus every totally ordered subset in W has an upper bound.
Hence by Zorn’s Lemma, W has a maximal element, say Z .
Since Z ∈ W, Y ⊂ Z and Z I E = φ.
Claim 1: Z is a hyper space.
Here we have to be careful. By appealing to Zorn’s lemma, we have obtained
a maximal element in W in the sense that there exists no proper subspace in
X between Z and X , having the required property.
86
Unit 3 Hahn-Banach Theorems
Now we have to prove it is also a maximal subspace of X having the required
properties.
On the contrary suppose that Z is not a hyperspace, then we arrive at a
contradiction.
For that we proceed as follows:
Set S = Z + ∪{αE : α > 0}.
Since E is open, αE is open for any α > 0 and since an open set + a subset
of X is always open, S is open. [by Proposition 8 in Unit 1]
Similarly − S is also open.
Further both S and − S are non-empty as E is non-empty.
Now we claim the following:
Subclaim 1: S ∩ − S = φ.
We prove this in step 1 by assuming the contrary and arriving at a
contradiction.
Step 1: Suppose that S ∩ − S ≠ φ
Suppose p ∈ S ∩ − S .
( z1 + z 2 ) α1 α2
− = x1 + x2
α 1 + α 2 α1 + α 2 α1 + α 2
E is convex right hand side lies in E.
Z is a subspace left hand side lies in Z .
( z1 + z 2 ) α1 α2
− = x1 + x2 ∈ E ∩ Z .
α 1 + α 2 α1 + α 2 α1 + α 2
Thus S ∩ − S ≠ φ E ∩ Z ≠ φ, a contradiction.
Thus S ∩ − S = φ.
Next we claim the following:
Subclaim 2: S ∪ − S ⊂ Z c .
+
We prove this in Step 2 by assuming the contrary and arriving at a
contradiction.
Step 2: Suppose there exists t ∈ S ∪ − S which also lies in Z .
Then t ∈ S U −S t = z1 + α1 x1 or
t = − z 2 − α 2 x2 , z1 z 2 ∈ Z , x1 x2 ∈ E.
α1 > 0, α 2 > 0.
t also lies in Z .
t − z1 t − z1
If t = z1 + α1 x1 then = x1 ∈ E , ∈ Z.
α1 α1 1
87
Block 1 Normed Linear Spaces
Thus E I Z ≠ φ, a contradiction. Similarly if t = − z 2 − α 2 x2 , even then we get a
contradiction.
Thus no element of S U − S lies in Z , i.e., S U −S ⊂ Z c .
Hence S U −S ⊂ Z c .
+
Thus ∃ x0 ∈ Z c ⊂ X such that
x0 ∉ S U −S .
x ∈ E and x = p + kx0 .
where p ∈ Z & k ∈ R.
If k = 0, then x = p E I Z ≠ φ, a contradiction.
1 1
If k > 0, then x0 = x− p
k k
1 1
=− p + x ∈ S,
k k
a contradiction since x0 ∉ Z .
1 1
Similarly if k < 0, x0 = − p + x ∈ −S.
k k
again a contradiction.
This means E I span {Z , x 0 } = φ.
Thus Z is hyperspace.
88
Unit 3 Hahn-Banach Theorems
Further, in the theorem we need a closed subspace.
Let us now check whether the hyperspace Z is a closed subspace. By
construction, E I Z = φ.
Z ∈ W.
f ( x) = u ( x) − iu (ix ), x ∈ X ,
as in Proposition 1. Then as proved in Proposition 1, f is a complex linear
functional on X .
But by Theorem 4 of Unit 2, since Z = Z (u ) is closed, u is a real continuous
linear functional on X and f is a continuous complex linear functional on X ,
i.e. f ∈ X 1 .
Re f ( x1 ) < t < Re f ( x 2 )
for all x1 ∈ E1 and x2 ∈ E2 .
90
Unit 3 Hahn-Banach Theorems
2
Recall: A hyperplane is just a translate of a hyperspace. In R , a hyperplane
is any straight. In R 3 , a hyperplane is any plane.
Let p1 , p 2 ∈ Re f ( E1 ).
tp1 + (1 − t ) p 2 = tf ( x1 ) + (1 − t ) f ( x 2 )
= f (tx1 + (1 − t ) x 2 ) (Q f is linear)
∈ f ( E1 ) since E1 is convex.
Thus, p1 , p 2 ∈ Re f ( E1 )
tp1 + (1 − t ) p 2 ∈ Re f ( E1 ) ∀ t ∈[0,1].
Re f ( E1 ) is convex.
Similarly we can show that Re f ( E 2 ) is convex. But Re f ( E1 ) and Re f ( E 2 )
are convex sets in R they are intervals.
Since Re f ( x1 ) ≠ Re f ( x 2 ), x1 ∈ E1 , x 2 ∈ E 2
Re f ( x ) ≤ t < Re f (a )∀ x ∈ E.
∂A = A − A0 = A I A c
92
Unit 3 Hahn-Banach Theorems
0
Proof: Since E is convex, the interior E is also convex.
Since, it is given that b ∈ ∂E , b ∉ E 0 .
Remark 10: The above result says if E is a convex body and b is a boundary
point of E then there always exists a support hyperplane for E at b.
The inequality Re f ( x ) ≤ Re f (b) again means the support hyperplane
separates E 0 and b !
x
If x ∈ X and f ( x) ≠ 0, consider, a − then
f ( x)
x f ( x)
f a − = f (a ) −
f ( x) f ( x)
= 1 − 1 = 0.
x
Hence a − ∈ Z( f )
f ( x)
x
By hypothesis a − ∉ U (a, r )
f ( x)
x
Hence a − −a ≥ r
f ( x)
x
≥r
f ( x)
x ≥ r f ( x)
1
f ( x) ≤ ⋅ x r.
r
Thus for x ∈ X , with f ( x) ≠ 0, we have
1
f ( x) ≤ ⋅ x ∀ x ∈ X .
r
If f ( x) = 0, for some x, the above inequality is trivial.
1
Thus f ( x) ≤ ⋅ x , ∀ x∈ X.
r
1
f ≤ .
r
94 Conversely suppose
Unit 3 Hahn-Banach Theorems
1
f ≤ .
r
Let x ∈ U (a, r ). Then x − a < r .
Consider now
f ( x ) − 1 = f ( x ) − f (a ) = f ( x − a )
1
≤ f x − a < . r = 1.
r
f ( x ) − 1 < 1.
1 − f ( x ) ≤ 1 − f ( x) ≤ 1 − f ( x) < 1
− f ( x ) < 0 f ( x) ≠ 0.
x ∉ Z ( f ).
Hence U (a, r ) I Z ( f ) = φ.
1
Let E = U X a, . Then E is a non-empty open convex subset of X and it
g
is disjoint from the subspace Z (g ) of X . Hence by the Hahn-Banach
separation theorem, ∃ f ∈ X ′ such that f ( y ) = 0 ∀ y ∈ Z ( g ) and Re f ( x) ≠ 0
for every x ∈ E .
Hence f ( x ) ≠ 0 for every x ∈ E. (2)
Since a ∈ E , f (a) ≠ 0.
95
Block 1 Normed Linear Spaces
f
f / y (a ) = 1 = g (a ). If necessary we an replace and assume f (a) = 1
f (a)
f
(calling also as f ). Now f / y is a mapping from
f (a)
Y → K, f / y (a) = 1 = g (a).
Let y ∈ Y . Consider y − g ( y )a
g ( y − g ( y )a ) = g ( y ) − g ( y ) g (a ) = g ( y ) − g ( y ).1 = 0
y − g ( y )a ∈ Z ( g ) ⊂ Z ( f / y ).
f ( y − g ( y ) a ) = 0 f ( y ) − g ( y ) f ( a) = 0
f ( y ) = g ( y ).
Thus for every y ∈ Y , f ( y ) = g ( y ).
i.e., f / y = g .
Hence g = f /y ≤ f . (3)
1
By (2) f ( x) ≠ 0 ∀ x ∈ E = U X a,
g
1
U X a, I Z ( f ) = φ.
g
1
Using Proposition 7, we get f ≤ = g. (4)
1
g
Let x ∈ X and 0 ≤ x (t ) ≤ 1 ∀ t ∈ T .
Claim: f ( y ) is real.
If K = R, there is nothing to prove.
(Q − 1 ≤ y(t ) ≤ 1).
2
sup y (t ) + ia ≤ 1 + a 2
t ∈T
y + ia ∞
≤1+ a2 (5)
(Here the real a is treated as the constant function taking the value a ).
(
Now consider β + a f ) = (β + af (1) )
2 2
2
≤ α + i (β + af (1)
2
= α + iβ + iaf (1)
2
= f ( y ) + f (ia ) , since f is linear.
2
= f ( y + ia )
2 2
≤ f y + ia ∞
(since f is continuous).
2
≤ f (1 + a 2 ) by (1)
2 2 2
Thus we have β + 2aβ f + a2 f
2 2
≤ f + a2 f .
2 2
β 2 + 2aβ f ≤ f , for every real a.
1
If β ≠ 0, taking a = , we get
2β
2 2
β2 + f ≤ f
β 2 ≤ 0, impossible.
Hence β = 0.
97
Block 1 Normed Linear Spaces
Hence f ( y ) = α i.e. f ( y ) is real.
This proves the claim.
Since y ∞
≤ 1, we have
f ( y) = α ≤ f .
y +1 α + f
Hence f ( x) = f = ≥ 0.
2 2
i.e. f ( x) ≥ 0.
x
Then set z = . Then 0 ≤ z (t ) ≤ 1, ∀ t.
x∞
Hence by what we have proved above
f ( z) ≥ 0
x
f ≥ 0 f ( x) ≥ 0.
x
∞
Thus ∀ x ∈ X with x(t ) ≥ 0, we have
f ( x) ≥ 0.
f is positive.
Converse part: suppose Re x ∈ X whenever x ∈ X , and f is positive.
( − y) ≥ 0
f y ∞
f ( y ) − f ( y) ≥ 0
∞
f ( y .1) − f ( y ) ≥ 0
∞
y ∞
f (1) − f ( y ) ≥ 0
y ∞
f − f ( y) ≥ 0
Here y ∞
f is real f ( y ) also should be real.
Let y1 = Re y & y 2 = Im y.
Since y ∈ X , by hypothesis Re y ∈ X
98
Unit 3 Hahn-Banach Theorems
i.e. y1 ∈ X .
We have r = f (e − is x) = f ( y1 + iy2 )
= f ( y1 ) + if ( y 2 ).
Since y1 is real, by what we have proved above f ( y1 ) is real.
Hence i f ( y 2 ) = r − f ( y1 ) is real.
f ( y 2 ) = 0.
We have:
y1 (t ) ≤ y1 ∞
≤ e −is x .
∞
= x ∞.
− x ∞
≤ y1 (t ) ≤ x ∞
∀ t ∈T
Thus ( )
f ( x ) = f e −is x = f ( y1 )
≤ f x( ∞
) (since f is positive and x ∞
− y (t ) ≥ 0)
= x ∞
f (1)
= x ∞
f
Thus f ( x ) ≤ f x ∞
f is continuous.
f = g = g (1) = f (1)
i.e. f = f (1).
Let Y = {( x1 , x2 ) ∈ X : x2 = 0}
Define g on Y as g ( x1 , x 2 ) = x1 .
For any x = ( x1 , x 2 ) ∈ Y ,
g ( x ) = x1 ≤ x1 + x 2 = x 1 .
g ≤1 (6)
Consider x0 = (1,0), x 0 1 = 1.
Also g ( x 0 ) = 1 g ( x 0 ) = 1.
Hence g ≥ g ( x 0 ) = 1. (7)
f ( x1 , x 2 ) = k1 x1 + k 2 x 2 , k1k 2 ∈ K
2
and ( x1 , x2 ) ∈ K .
If f has to be an extension of g , then f should be of the form.
= max { k1 , k 2 }( x1 + x2 )
= max { k1 , k 2 } ( x1 , x2 ) 2
f ≤ max { k1 , k 2 } (8)
k1 = 1.
So f should be of the form.
all ( x1 , x 2 ) ∈ K 2
But f = g = 1
k2 ≤ 1
Remark 13: The above example shows the Hahn-Banach extensions are in
general not unique.
Thus Hahn-Banach extensions always exist, but may not be unique. So the
natural question is: Are there conditions under which Hahn-Banach extensions
are unique. We address this question little later.
{
Hence a = sup f ( a ) : f ∈ X ′, f ≤ 1 . }
Proof: Take Y = {ka : k ∈ K). Then Y is a closed subspace of X .
Define g on Y as g ( ka) = k a , k ∈ K.
g ( a) = a
= sup{ ka : ka ∈ Y & ka = 1} = 1
i.e. g = 1.
Hence we have:
f ( a ) = g ( a ) = a & f = g = 1.
Thus we have
a = f ( a) = f ( a) (Q f (a) = a ≥ 0)
≤ sup {h( a) : h ∈ X ′, h ≤ 1}
Also h ∈ X ′ h( a) ≤ h a
≤ sup { h a : h ∈ X ′, h ≤ 1}
= a . sup { h : h ∈ X ′, h ≤ 1}
= a .1 = a .
Remark 14: This theorem gives a formula for the norm of an element of X in
terms of bounded linear functionals.
~
Also f = f = 1 since:
~ ~
f ( x) = f ( x + Y ) ≤ f x + Y
102
Unit 3 Hahn-Banach Theorems
~
≤ f x
~
f ≤ f .
≤ f x+ y
~
f ( x + Y ) ≤ f inf { x + y : y ∈ Y } = f x +Y .
~
f ≤ f .
~
Thus f = f .
gj
Set f j = . Then f i ∈ X ′ and f j (ai ) = δ ij , 1 ≤ i, j ≤ m.
dist (a j , Yi )
X = X 0 + X 1, X 0 ∩ X1 = { 0 }.
n
X 1 = I N ( f j ).
j =1
n
y = f j ( x) x j , z = x− y.
j =1
103
Block 1 Normed Linear Spaces
Proof: Define X 1 as given in the theorem. Since each null space is a closed
subspace of X and a finite intersection of closed of subspaces is again a
closed subspace, X 1 is a closed subspace of X .
Now let w ∈ X 0 ∩ X 1
w ∈ X 0 w = α1 x1 + ... + α n xn , α i ∈ K.
w ∈ X 1 f i ( w) = 0, i = 1, ..., n.
Thus 0 = αi , i = 1,..., n w = 0
Now let x ∈ X .
n
Set y = f ( x) x . Since x ,..., x
i =1
i i 1 n is a basis for X 0 , y ∈ X 0
If we set z = x − y , then
f i ( z) = 0, i = 1,..., n z ∈ X 1 x = y + z, y ∈ X 0 , z ∈ X 1 .
Thus X = X 0 + X 1 , X 0 I X 1 = {0}.
Now we will address the question whether the extension obtained in the Hahn-
Banach theorem is unique.
The following proposition shows that the uniqueness holds in special cases.
We shall only give the statement of the theorem, the proof is omitted.
The general uniqueness Theorem is a deep result which is beyond the scope
of this course.
There are other special cases where uniqueness of extension exist. For
example you will see later that the uniqueness exist in the following cases.
3.4 SUMMARY
In this unit, the following points have been covered.
f (ix ) = if ( x) = i (u ( x) + iv ( x)) = −v ( x ) + iu ( x)
v ( x ) = −iu (ix ).
u( x) ≤ f ( x) ≤ f x.
Thus u ≤ f
f (e it x ) = e it f ( x)
Since f (e it x ) is real, f (e it x) ≤ u (e it x )
Therefore
e it f ( x ) = f (e it x ) = u (e it x) ≤ u x
Hence f = u .
f ( x1 , x2 , x3 ) = x1 , x = ( x1 , x2 , x3 ) ∈ R 3
f ( x) ≤ x .
g ( x) = x = ( x1 , x 2 , x3 , x4 )
= x1 + ax2 + bx3 + cx4
107