~ TEXTS AND READINGS 52

~ IN MATHEMATICS

Functional Analysis

s. Kesavan

I1,Jr,i1@HINDUSTAN
UillJ UBOOK AGENCY
TEXTS AND READINGS
IN MATHEMATICS 52
Functional Analysis
Texts and Readings in Mathematics

Advisory Editor
C. S. Seshadri, Chennai Mathematical Institute, Chennai.

Managing Editor
Rajendra Bhatia, Indian Statistical Institute, New Delhi.

Editors
R. B. Bapat, Indian Statistical Institute, New Delhi.
V. S. Borkar, Tata Inst. of Fundamental Research, Mumbai.
Probal Chaudhuri, Indian Statistical Institute, Kolkata.
V. S. Sunder, Inst. of Mathematical Sciences, Chennai.
M. Vanninathan, TIFR Centre, Bangalore.
 Functional Analysis

s. Kesavan
Institute of Mathematical Sciences
Chennai

fl:[gUgl HINDUSTAN
U U2J UBOOKAGENCY
Published in India by

Hindustan Book Agency (India)

P 19 Green Park Extension
New Delhi 110 016
India

email: info@hindbook.com
http://www.hindbook.com

Copyright © 2009, Hindustan Book Agency (India)

No part of the material protected by this copyright notice may be

reproduced or utilized in any form or by any means, electronic or
mechanical, including photocopying, recording or by any information
storage and retrieval system, without written permission from the
copyright owner, who has also the sole right to grant licences for
translation into other languages and publication thereof.

All export rights for this edition vest exclusively with Hindustan Book
Agency (India). Unauthorized export is a violation of Copyright Law
and is subject to legal action.

ISBN 978-81-85931-87-6 ISBN 978-93-86279-42-2 (eBook)

DOI 10.1007/978-93-86279-42-2
Preface

Functional analysis studies linear spaces provided with suitable topo-

logical structures and (continuous) linear transformations between such
spaces. It has far reaching applications in several disciplines. For in-
stance, the modern theory of (partial) differential equations and the
numerical approximation of their solutions rely a lot on functional ana-
lytic techniques.

Functional analysis is now an integral part of the curriculum in any

post graduate course in Mathematics. Ideally it should be taught after
having covered courses in linear algebra, real and complex analysis, and
topology. An introduction to the theory of measure and integration is
also helpful since the richest examples in functional analysis come from
function spaces whose study demands such a knowledge.

The present book grew out of notes prepared by myself while lectur-
ing to graduate students at the Tata Institute of Fundamental Research
(Bangalore Centre) and the Institute of Mathematical Sciences, Chen-
nai. The material presented in this book is standard and is ideally suited
for a course which can be followed by masters students who have cov-
ered the necessary prerequisites mentioned earlier. While covering all
the standard material, I have also tried to illustrate the use of various
theorems via examples taken from differential equations and the calcu-
lus of variations, either through brief sections or through the exercises.
In fact, this book is well suited for students who would like to pursue a
research career in the applications of mathematics. In particular, famil-
iarity with the material presented in this book will facilitate studying
my earlier book, published nearly two decades ago, 'Topics in Func-
tional Analysis and Applications' (Wiley Eastern, now called New Age
International), which serves as a functional analytic introduction to the
theory of partial differential equations.

The first chapter of the present book gives a rapid revision of linear
algebra, topology and measure theory. Important definitions, examples
and results are recalled and no proofs are given. At the end of each
section, the reader is referred to a standard text on that topic. This
chapter has been included only for reference purposes and it is not in-
tended that it be covered in a course based on this book.
vi

Chapter 2 introduces the notion of a normed linear space and that

of continuous linear transformations between such spaces.

Chapter 3 studies the analytic and geometric versions of the Hahn-

Banach theorem and presents some applications of these.

Chapter 4 is devoted to the famous 'trinity' in functional analysis

- the Banach-Steinhaus, the open mapping and the closed graph theo-
rems, which are all consequences of Baire's theorem for complete metric
spaces. Several applications are discussed. The notion of an 'unbounded
linear mapping' is introduced.

In my opinion, most texts do not emphasize the importance of weak

topologies in a course on functional analysis. That a bounded sequence
in a reflexive space admits a weakly convergent sequence is the corner
stone of many an existence proof in the theory of (partial) differential
equations. These topologies also provide nice counter-examples to show
the inadequacy of sequences in a general topological space. For instance,
we will see that two topologies on a set could be different while having
the same convergent sequences. We will also see an example of a com-
pact topological space in which a sequence does not have any convergent
subsequence. Chapter 5 deals with weak and weak* topologies and their
applications to the notions of reflexivity, separability and uniform con-
vexity.

Chapter 6 introduces the Lebesgue spaces and also presents the the-
ory of one of the simplest classes of Sobolev spaces.

Chapter 7 is devoted to the study of Hilbert spaces.

Chapter 8 studies compact operators and their spectra.

Much of the fun in learning mathematics comes from actually doing

it! Every chapter from Chapters 2 through 8 has a fairly large collection
of exercises at the end. These illustrate the results of the text, show
the optimality of the hypotheses of the various theorems via various ex-
amples or counterexamples, or develop simple versions of theories not
elaborated upon in the text. They are of varying degrees of difficulty.
 vii

Occasionally, some hints for the solution are provided. It is hoped that
the students will benefit by solving them.

Since this is meant to be a first course on functional analysis, I have

kept the bibliographic references to a minimum, merely citing important
texts where the reader may find further details of the topics covered in
this book.

No claim of originality is made with respect to the material presented

in this book. My treatment of this subject has been influenced by the
writings of authors like Simmons (whose book was my first introduction
to Functional Analysis) and Rudin. These works figure in the bibliogra-
phy of this book. I would also like to mention a charming book, hardly
known outside the francophonic mathematical community, viz. 'Analyse
Fonctionnelle' by H. Brezis.

The preparation of this manuscript would not have been possible

but for the excellent facilities provided by the Institute of Mathematical
Sciences, Chennai, and I wish to place on record my gratitude. I also
thank the Hindustan Book Agency and the editor of the TRIM Series,
Prof. R. Bhatia, for their kind cooperation in bringing out this volume.
I must thank the anonymous referees who painstakingly went through
the first draft of the manuscript. I have tried to incorporate many of
their constructive suggestions in the current version.

Finally, I thank my family for its constant support and encourage-

ment.

Chennai, s. Kesavan
May 2008.
Notations
Certain general conventions followed throughout the text regarding
notations are described below. All other specific notations are explained
as and when they appear in the text.

• The set of natural numbers is denoted by the symbol N, the in-

tegers by Z, the rationals by Ql, the reals by lR and the complex
numbers by C.

• Sets (including vector spaces and their subspaces) and also linear
transformations between vector spaces are denoted by upper case
latin letters.

• Elements of sets (and, therefore, vectors as well) are denoted by

lower case latin letters.

• Scalars are denoted by lower case greek letters.

• To distinguish between the scalar zero and the null (or zero) vector,
the latter is denoted by the zero in boldface, i. e. o. This is also
used to denote the zero linear transformation and the the zero
linear functional.

• Column vectors in Euclidean space are denoted by lower case latin

letters in boldface and matrices are denoted by upper case latin
letters in boldface.

• Elements of £P spaces (cf. Chapter 6) are equivalence classes of

functions under the equivalence relation of 'equality almost every-
where'. To emphasize this fact, all elements of £P spaces (and
hence of Sobolev spaces as well) are denoted by lower case latin
letters in the sanserif font. A generic representative of that equiva-
lence class is denoted by the same lower case latin letter (in italics).
Thus, if f E Ll(O, 1), a generic representative of this class will be
denoted by f and will feature in all computations involving this
element. For instance, we have

Ilflll = fal If( t) I dt.

• The norm in a normed linear space V will be denoted by 11.11, or

by 11.llv, if we wish to distinguish it from other norms that may
x

be entering the argument. Similarly the innerproduct in a Hilbert

space H will, in general, be denoted by (.,.) (or by (., .)H in case
we wish to stress the role played by H).
Contents

1 Preliminaries 1
1.1 Linear Spaces . . . . . . 1
1.2 Topological Spaces . . . 11
1.3 Measure and Integration 18

2 Normed Linear Spaces 26

2.1 The Norm Topology 26
2.2 Examples . . . . . . 28
2.3 Continuous Linear Transformations. 36
2.4 Applications to Differential Equations 52
2.5 Exercises 57

3 Hahn-Banach Theorems 69
3.1 Analytic Versions . . . 69
3.2 Geometric Versions .. 77
3.3 Vector Valued Integration 81
3.4 An Application to Optimization Theory 85
3.5 Exercises 92

4 Baire's Theorem and Applications 97

4.1 Baire's Theorem . . . . . . . . . . 97
4.2 Principle of Uniform Boundedness 98
4.3 Application to Fourier Series . . . 101
4.4 The Open Mapping and Closed Graph Theorems 104
4.5 Annihilators . . . . . . . . . . . . 108
4.6 Complemented Subspaces . . . . 113
4.7 Unbounded Operators, Adjoints . 116
4.8 Exercises 125
xii

5 Weak and Weak* Topologies 132

5.1 The Weak Topology · 132
5.2 The Weak* Topology . · 139
5.3 Reflexive Spaces · .. · 145
5.4 Separable Spaces · .. · 147
5.5 Uniformly Convex Spaces · 152
5.6 Application: Calculus of Variations · 156
5.7 Exercises ............. · 158

6 LP Spaces 162
6.1 Basic properties . · 162
6.2 Duals of LP Spaces · 168
6.3 The Spaces LP(n) .170
6.4 The Spaces W1,P(a, b) .178
6.5 Exercises ...... · 188

7 Hilbert Spaces 195

7.1 Basic Properties · .......... · 195
7.2 The Dual of a Hilbert Space . . . . . .202
7.3 Application: Variational Inequalities .207
7.4 Orthonormal Sets . .210
7.5 Exercises .223

8 Compact Operators 231

8.1 Basic Properties · ......... .231
8.2 Riesz-Fredholm Theory . . . . . . .238
8.3 Spectrum of a Compact Operator . .244
8.4 Compact Self-Adjoint Operators · 251
8.5 Exercises . . . . . . . . . . . . . .258

Bibliography 265

Index 266
Chapter 1

Preliminaries

1.1 Linear Spaces

Functional Analysis is the study of vector spaces endowed with topo-

logical structures (that are compatible with the linear structure of the
space) and of (linear) mappings between such spaces. Throughout this
book we will be working with vector spaces whose underlying field is the
field of real numbers IR or the field of complex numbers C.
For completeness, we will recall some basic definitions and some im-
portant results.
Definition 1.1.1 A vector space or a linear space over a field IF
(whose elements are called scalars) is a set V, whose elements are called
vectors, on which two operations - addition and scalar multiplica-
tion - are defined such that the following properties hold:
Addition: (x, y) E V x V I---t X + Y E V such that
(i) (commutativity) for all x and y E V, we have

x+y = y+x;
(ii) (associativity) for all x, y and Z E V, we have

x + (y + z) = (x + y) + z;
(iii) there exists a unique vector 0 E V, called the zero or the null
vector, such that, for every x E V,

x+o = x;

(iv) for every x E V , there exists a unique vector -x E V such that

x + (-x) = o.
2 1 Preliminaries

Scalar Multiplication: (a, x) E IF x V I-t ax E V such that

(v) for every x E V, Ix = x where 1 is the multiplicative identity in IF;
(vi) for all a and f3 E IF and for every x E V,
a(f3x) = (af3)x;
(vii) for all a and f3 E IF and for every x E V,
(a + (3)x = ax + f3x;
(viii) for every a E IF and for all x and Y E V,
a(x+y) = ax+ay . •
Remark 1.1.1 The conditions (i) - (iv) above imply that V is an abelian
group with respect to vector addition. The conditions (vii) and (viii)
above are known as the distributive laws. •

Example 1.1.1 Let N 2: 1 be a positive integer. Define

~N = {X=(Xl, ... ,XN)lxiE~ foralll:Si:SN}.
We define addition and scalar multiplication componentwise, i. e. if x =
(Xl, ... , XN) and Y = (Yl, ... , YN) are elements of ~N and if a E ~, we
define
X+y= (Xl+Yl, ... ,XN+YN)
and
ax = (axl, ... ,axN).
It is now easy to see that ~N is a vector space over ~ with the zero
vector being that element in ~N with all its components zero. In the
same way, we can define cP as a vector space over C.
Setting N = 1, we see that ~ (respectively q is a vector space over
itself, the scalar multiplication being the usual multiplication operation .
•
Definition 1.1.2 Let V be a vector space and let W c V. Then W is
said to be a subspace of V if W is a vector space in its own right for
the same operations of addition and scalar multiplication. •

Definition 1.1.3 Let V be a vector space and let Xl, ... , Xn be vectors
in V. A linear combination of these vectors is any vector of the form
alxl + ... +anxn , where the ai, 1 :S i :S n are scalars. A linear relation
between these vectors is an equation of the form
alXl + ... + anXn = 0.•
1.1 Linear Spaces 3

Definition 1.1.4 A finite set of vectors in a vector space is said to be

linearly independent if there does not exist any linear relation between
them other than the trivial one, i. e, when all the scalar coefficients are
zero. If there exists a non-trivial linear relation, then the set of vectors
is said to be linearly dependent. An infinite set of vectors is said to
be linearly independent if there does not exist any finite linear relation
amongst vectors in that set. •

Given a vector space V and a set S of vectors, the collection of all

finite linear combinations of vectors from S will form a subspace of V.
In fact, it is clear that this subspace is the smallest subspace containing
S and is called the linear span of the set S or the subspace generated by
S. This subspace is denoted by span{S}.

Definition 1.1.5 A maximal linearly independent subset of a vector

space is called a basis . •

In other words, a basis is a linearly independent subset such that, if

any other vector is adjoined to the set, the enlarged set becomes linearly
dependent. This implies that every vector in the space can be expressed
as a (finite) linear combination of the members of the basis. Thus, the
vector space is generated by its basis.

Proposition 1.1.1 (i) Every vector space has a basis.

(ii) Any two bases of a vector space have the same cardinality . •

The above proposition leads us to the following definition.

Definition 1.1.6 A vector space V is said to be finite dimensional if

it admits a basis with a finite number of elements. Otherwise, it is said
to be infinite dimensional. The dimension of a vector space is the
number of elements in a basis, if it is finite dimensional, and infinity if
it is infinite dimensional and is denoted dim(V) . •

Example 1.1.2 The space ]RN (respectively, eN) has a basis which
is defined as follows. Let 1 ::; i ::; N. Let ei be the vector whose i-th
component is unity and all other components are zero. Then {el' ... , eN}
is a basis for]RN (respectively, eN) and is called the standard basis . •

Example 1.1.3 Let P denote the collection of all polynomials in one

variable with real coefficients. Let p(x) = "L~=o aixi and q(x) = "L~o bix i
4 1 Preliminaries

where the ai and bi are real numbers and x is the variable. Let a E R
Assume, without loss of generality, that m ::; n. Define
n
(p + q)(x) = I)ai + bdxi
i=O

where we set bi = 0 for m < i ::; n if m < n. Define

n
(ap)(x) = L aaixi.
i=O

With these operations, P becomes a vector space over R It is easy

to check that the collection of monomials {p ~o where Po(x) == 1 and
Pi(X) = xi for i > 1, forms a basis for P. Thus, P is an infinite dimen-
sional vector space. •

We will come across numerous examples of infinite dimensional vec-

tor spaces in the sequel (cf. Section 2.2, for instance).
Let V be a vector space and let Wi, 1 ::; i ::; n be subspaces. The
span of the Wi is the subspace of all vectors of the form

v= WI + ... + Wn
where Wi E Wi for each 1 :::; i :::; n. The spaces are said to be independent
if an element in the span is zero if, and only if each, Wi = O. In particular,
it follows that, if the Wi are independent, then for all 1 ::; i, j ::; n such
that i f. j, we have Wi n Wj = {O}. Further, every element in the span
will have a unique decomposition into vectors from the spaces Wi'
Definition 1.1.7 Let V be a vector space and let Wi, 1 ::; i ::; n be
subspaces. Then, V is said to be the direct sum of the Wi if the spaces
Wi are independent and their span is the space V. In this case we write

We now study mappings between vector spaces which preserve the

linear structure.

Definition 1.1.8 Let V and W be vector spaces (over the same base
field). A linear transformation, or linear operator, is a mapping
T : V -> W such that for all x and y E V and for all scalars a and (3,
we have
T(ax + (3y) = aT(x) + (3T(y).
1.1 Linear Spaces 5

If W is the base field (which is a vector space over itself), then a linear
transformation from V into W is called a linear functional on V. •

Definition 1.1.9 Let V and W be vector spaces and let T : V W

-t

be a linear transformation. The image of T is a subspace of Wand is

called the range of T. The dimension of the range is called the rank of
T. The set
{x E V IT(x) = O}
is a subspace of V and is called the null space or kernel of T . •

Definition 1.1.10 Let V and W be vector spaces and T a linear trans-

formation between them. The transformation is said to be invertible if
T is a bijection. •

It is easy to see that a linear transformation which is an injection

maps a linearly independent set onto a linearly independent set. In
particular, if T : V - t W is an injection, then, necesarily, dim(V) ~
dim(W). On the other hand, if T : V - t W is a surjection, clearly,
dim(V) ::::: dim(W). Thus, if T is invertible, then the two spaces must
have the same dimension.
We now focus our attention on finite dimensional spaces. Let V
be a space of dimension n with basis {VI, ... , vn } and W a space of
dimension m with basis {WI, ... , w m }. A linear map T : V - t W is
completely defined, once it is defined on a basis. So let us write
m

T(vj) = L tijWi, 1 ~ j ~ n. (1.1.1)

i=1

The coefficients (tij) in the above relation form a matrix with m rows
and n columns. Such a matrix is referred to as an m x n matrix. The
j-th column of the matrix represents the coefficients in the expansion
of T(vj) in terms of the basis { ~1 of W. Of course, if we change
the bases for V and W, the same linear transformation will be given by
another matrix. In particular, let dim(V) = n and let T : V - t V be a
linear operator. Let T be represented by the n x n matrix (also known
as a square matrix of order n) T = (tij) with respect to a given basis. If
we change the basis, then T will be represented by another n x n matrix
T = (tij) and the two will be connected by a relation of the form:
6 1 Preliminaries

where P is called the change of basis matrix and represents the linear
transformation which maps one basis to another. The matrix p-l rep-
resents the inverse of this change of basis m p~n and is the inverse
matrix of P. In this case, the matrices T and T are said to be simi-
lar. The identity matrix I represents the identity mapping x 1-+ x for
all x E V for any fixed basis of V. For a given basis, if T : V ~ V is
invertible, then the matrix representing T- 1 will be the inverse of the
matrix representing T.
A square matrix is said to be diagonal if all its off-diagonal entries
are zero. A n x n square matrix A = (aij) is said to be upper trian-
gular (repectively, lower triangular) if aij = 0 for all 1 :S j < i :S n
(repectively, aij = 0 for all 1 :S i < j :S n). It can be shown that every
matrix is similar to an upper triangular matrix. A matrix is said to be
diagonalizable if it is similar to a diagonal matrix.
Given an m x n matrix and two vector spaces of dimensions nand m
respectively along with a basis for each of them, the matrix can be used,
as in relation (1.1.1), to define a linear transformation beteween these
two spaces. Thus, there is a one-to-one correspondence between ma-
trices and linear transformations between vector spaces of appropriate
dimension, once the bases are fixed.
Definition 1.1.11 If T = (tij) is an m x n matrix, then the n x m
matrix T' = (tji), formed by interchanging the rows and the columns of
the matrix T, is called the transpose of the matrix T. If T = (tij) is
an m x n matrix with complex entries, then the n x m matrix T* = (tTj )
where tTj = tji (the bar denoting complex conjugation), is called the
adjoint of the matrix T . •
If x and y E ]Rn (respectively, en), then y'x (respectively, y*x) repre-
sents the 'usual' scalar product of vectors in ]Rn (respectively, en) given
by L~~l XiYi (respectively, L~=l x/fh). If the scalar product is zero, we
say that the vectors are orthogonal to each other and write x ~ y. If
W is a subspace and x is a vector orthogonal to all vectors in W, we
write x ~ W.
Definition 1.1.12 Let T be an m x n matrix. Then its row rank is
defined as the number of linearly independent row vectors of the matrix
and the column rank is the number of independent column vectors of
the matrix. •
The column rank is none other than the rank of the linear transfor-
mation defined by T and the row rank is the rank of the transformation
1.1 Linear Spaces 7

defined by the transpose. We have the following important result.

Proposition 1.1.2 For any matrix, the row and column ranks are equal
and the common value is called the rank of the matrix. •

Definition 1.1.13 The nullity of a matrix is the dimension of the null

space of the linear transformation defined by the matrix. •

Proposition 1.1.3 Let T be an m x n matrix. The the sum of the rank

of T and the nullity of T is equal to n.

Corollary 1.1.1 Ann x n matrix is invertible if, and only if, its nullity
is zero or, equivalently, its rank is n. Equivalently, a linear operator on
a finite dimensional space is one-to-one if, and only if, it is onto . •

n
Definition 1.1.14 Let T be an n x matrix with complex entries and
T* its adjoint. The matrix is said to be normal if

TT* = T*T.

The matrix is said to be unitary if

TT* = T*T = I.

The matrix is said to be self-adjoint or hermitian if T = T* . •

Definition 1.1.15 Let T be an n x n matrix with complex entries. It is

said to be positive semi-definite if for every n x 1 matrix with complex
entries, i.e, a column vector, x, we have

x*Tx ~ O.

The matrix T is said to be positive definite if, in addition, the above

inequality is strict if xi- O. •

Remark 1.1.2 A hermitian matrix is equal to its adjoint and the inverse
of a unitary matrix is its adjoint. A matrix T, with real entries, which
is equal to its transpose is called symmetric and one whose inverse is
its own transpose is called orthogonal. In case the matrix T has real
entries, we can still define positive semi-definiteness (or positive definite-
ness) by considering real column vectors x in the above definition. •
8 1 Preliminaries

We now introduce an important notion, viz. that of the determinant.

Before we do this, we need some notation. Let Sn denote the set of all
permutations of n symbols. A transposition is a permutation wherein
two symbols exchange places with each other and all other symbols are
left invariant. It is known that every permutation is the product (in the
sense of composition of mappings) of transpositions. A permutation is
even if it is the product of an even number of transpositions and odd if
it is the product of an odd number of transpositions. The signature of a
permutation CT, denoted sgn CT, is +1 if the permutation is even and -1
if it is odd.
Definition 1.1.16 Let T = (tij) be an n x n matrix. The determinant
of T, denoted det(T), is given by the formula

det(T) = L (sgn CT)t1,a(1) ... tn,a(n)' •

aESn

We list below the important properties of the determinant.

Proposition 1.1.4 (i) If I is the identity matrix of order n, then det(l) =
l.
(ii) If T and S are two n x n matrices, then

det(ST) = det(S).det(T).
In particular, T is invertible if, and only if, det(T) =I- 0 and
det(T- 1 ) = (det(T))-l.
(iii) If T is an n x n matrix, then

det(T') = det(T) . •

Definition 1.1.17 An invertible matrix is also said to be a non-singular

matrix. Otherwise, it is said to be singular. •

If T is a non-singular matrix of order n, then, given any n x 1 column

vector b, there exists a unique n x 1 column vector x such that

Tx = b.
This is because the corresponding linear transformation is invertible and
hence onto (which gives the existence of the solution) and one-to-one
(which gives the uniqueness of the solution). If T is singular, this is no
longer the case and we have the following result.
1.1 Linear Spaces 9

Proposition 1.1.5 (FredhOlm Alternative) Let T be a singular ma-

trix of order n and let b be an n x 1 column vector. Then, either the
system of n linear equations in n unknowns (written in matrix notation)

Tx = b
has no solution or has an infinite number of solutions. The latter pos-
sibility occurs if, and only if

b'u = 0

for all (column) vectors u such that T/u = o.•

We now come to a very important notion in linear algebra and func-
tional analysis.

Definition 1.1.18 Let T be a square matrix of order n with complex

entries. A complex number A is said to be an eigenvalue of T if there
exists an n x 1 vector u =1= 0 such that

Tu = AU.

Such a vector u is called an eigenvector of T associated to the eigen-

value A. The set of all eigenvectors associated to an eigenvalue A is a
subspace of en and is called the eigenspace associated to A.•

From the above definition, we see that A E e is an eigenvalue of a

square matrix T if, and only if the matrix T - ),1 is not invertible. Thus,
A will be an eigenvalue of T if, and only if,

det(T - ),1) = O. (1.1.2)

The expression on the left-hand side of (1.1.2) is a polynomial of degree

equal to the order of T and is called the characteristic polynomial
of T. Thus, every eigenvalue is a root of the characteristic polynomial
and so every matrix of order n has a non-empty set of at most n distinct
eigenvalues. Counting multiplicity, there are exactly n eigenvalues. The
equation (1.1.2) is called the characteristic equation of the matrix T.

Definition 1.1.19 Let A be an eigenvalue of a matrix T. Its algebraic

multiplicity is its multiplicity as a root of the characteristic polyno-
mial. Its geometric multiplicity is the dimension of the eigenspace
associated to it. •
10 1 Preliminaries

Proposition 1.1.6 The geometric multiplicity of an eigenvalue does

not exceed its algebraic multiplicity . •

Definition 1.1.20 The set of all eigenvalues of a matrix is called its

spectrum. The maximum of the absolute values of the eigenvalues is
called its spectral radius . •

The sum of the diagonal entries of a square matrix T is called its

trace and is denoted by tr(T). It is easy to see from the characteristic
equation that the trace is the sum of all the eigenvalues taking into
account their multiplicities. Similarly the determinant of a matrix is the
product of all its eigenvalues (again counting mUltiplicity).

Proposition 1.1.7 (i) The eigenvalues of the adjoint of a matrix are

the complex conjugates of the eigenvalues of the original matrix.
(ii) The eigenvalues of a hermitian matrix are all real.
(iii) The eigenvalues of a hermitian and positive definite matrix are pos-
itive.
(iv) The eigenvalues of a unitary matrix lie on the unit circle of the
complex plane. •
The eigenvalues of a hermitian matrix admit a 'variational charac-
terization'. Let T be a hermitian matrix of order n. Its eigenvalues are
all real and let them be numbered in increasing order as follows:

Let Vi, 1 :s: i :s: n be a collection of eigenvectors, where Vi is associated

to Ai. Set Vo = {O} and

for 1 :s: i :s: n.

We define the Rayleigh quotient associated to the matrix T as follows:
x*Tx
RT(X) = --,
x*x
x 1= 0,
which is real valued, since T is hermitian.

Proposition 1.1.8 (i) The eigenvectors Vi can be chosen such that

if i = j,
if i 1= j.
1.2 Topological Spaces 11

(ii) For each 1 :::; i :::; n, we have

Ai RT( Vi)
RT(X)
maxxE\!;
mind Vi-I RT(X)
minwcc n , dim(W)=i maxxEW RT(X).

In particular,

By analogy, if T is a real symmetric matrix, a similar result holds

with the adjoint being replaced by the transpose in the definition of the
Rayleigh quotient and en
being replaced by]Rn in the formulae above.

Remark 1.1.3 A matrix is diagonalizable if, and only if, it admits a

basis of eigenvectors. In particular, if all the eigenvalues of a matrix are
distinct, it is diagonalizable. All normal matrices are diagonalizable. In
particular, hermitian matrices are diagonalizable. •

For more details on linear spaces, the reader is referred to, for in-
stance, Artin [1].

1.2 Topological Spaces

In this section, we recall the important definitions and results of topology

which will be used in the sequel.

Definition 1.2.1 A topology on a set X is a collection T of subsets

of X such that
(i) X and (/) are in T;
(ii) the union of any collection of members in T is a member of T;
(iii) the intersection of any finite collection of members ofT is a member
ofT.
The pair {X, T} is said to be a topological space and the members of
the topology T are called open sets. The complements of open sets are
called closed sets. If x E X, then a neighbourhood of x is any open
set containing x. •

It is clear from the above definition that X and 0 are both open and
closed. Further, a finite union and an arbitrary intersection of closed
12 1 Preliminaries

sets is closed. It then follows that given any set A eX, there is a
smallest closed set containing it. This is called the closure of the set
A and is usually denoted by A. If A = X, we say that A is dense in
X. Similarly, given any set A eX, there is a largest open set contained
in A. This set is called the interior of A and is denoted by A o . A set
A c X is said to be nowhere dense if (A)O =.0.
If {X, T} is a topological space and if Y eX, then Y inherits a
natural topology from that of X. The open sets are those of the form
Un Y, where U is open in X.

Definition 1.2.2 A topological space {X, T} is said to be Hausdorff

if for every pair of distinct elements x and y in X, there exist disjoint
open sets U and V such that x E U and y E V. •

Definition 1.2.3 A metric on a set X is a function d : X x X ----t [0,00)

such that
(i) d(x, y) = 0 if, and only if, x = y;
(ii) for all x and y EX, we have

d(x, y) = d(y, x);

(iii) for all x, y and Z E X, we have

d(x, z) ::; d(x, y) + d(y, z). (1.2.1)

The pair {X, d} is called a metric space . •

Remark 1.2.1 The notion of a metric generalizes that of a distance, as

we know it, in the Euclidean spaces JR2 and JR3. The inequality (1.2.1)
is just an abstract version of a familiar theorem from Euclidean plane
geometry which states that the sum of the lengths of two sides of a tri-
angle is greater than the length of the third side. For this reason, it is
called the triangle inequality. Of all the conditions on a metric, this one
will need the most non-trivial verification . •

If {X, d} is a metric space, then it is easy to see that we have a

topology (called the metric topology) induced on X by the metric d
which is defined as follows. A non-empty set U C X is open if, and only
if, for every x E U, there exists r > 0 such that

B(x;r) ~f {y E X I d(x,y) < r} C U.

1.2 Topological Spaces 13

The set B(x; r) described above is called the (open) ball centred at x and
of radius r. It is a simple exercise to check that open balls themselves
are open sets. It is also immediate to see that this topology is Hausdorff.
On]R or e, we have the 'usual' metric defined by

d(x, y) = Ix - YI·
The topology induced by this metric will be called the 'usual' topology
on ]R or e, as the case may be. Similarly, on ]RN (or eN), we have
the 'usual' Euclidean distance which defines a metric on that space: if
x = (x, ... , XN) and Y = (YI, ... , YN) are vectors in]RN (respectively, eN),
then

The topology induced by this metric will be referred to as the 'usual'

topology on ]RN (respectively, eN).
In the metric topology defined above, we see that every open set is
the union of open balls. In a general topological space a collection B
of open sets is called a base for the topology if every open set can be
expressed as the union of members of B.

Definition 1.2.4 Let {X, T} be a topological space and let S be a col-

lection of open sets in X. We say that S is a subbase for the topology
T if every open set can be expressed as unions of finite intersections of
members of s .•

Clearly any topology containing S will have to contain T. Thus T

is the smallest topology containing S. The set of finite intersections of
members of S form a base for the topology.

Definition 1.2.5 Let {X, T} be a topological space and let A be an

arbitrary subset of X. A point x E X is said to be a limit point of A
if every neighbourhood of x contains a point of A (different from x, in
case x E A) . •

Definition 1.2.6 Let {X, T} be a topological space and let {x n } be a

sequence of elements in X. We say that the sequence converges to a
point x E X if for every neighbourhood U of x, we can find a positive
integer N (depending on U) such that Xk E U for all k ~ N. In this
case, we write Xn --t x in X . •
14 1 Preliminaries

Definition 1.2.7 Let {Xi, 'Ii}, i = 1,2, be two topological spaces and
let f : Xl -+ X 2 be a given function. We say that f is continuous
if f-l(U) is an open set in Xl for every open set U in X 2 • If f is
a bijection such that both f and f- l are continuous, then f is said to
be a homeomorphism and the two topological spaces are said to be
homeomorphic to each other. •

The following propositions are easy to prove.

Proposition 1.2.1 Let {X,d} be a metric space and let {x n } be a se-

quence in X. Then Xn -+ x in X if, and only if, for every E > 0, there
exists a positive integer N such that

d(Xk' x) < E for every k 2: N . •

In particular, every convergent sequence is bounded, i. e., it can be

contained in a (sufficiently large) ball.
Proposition 1.2.2 Let {Xi, dd, i = 1,2, be metric spaces and let f :
Xl -+ X 2 be a given function. The following are equivalent:
(i) f is continuous.
°
(ii) For every x E X and for every E > 0, there exists <5 > such that
whenever d l (x, y) < <5, we have d2 (f(x),j(y)) < E.
(iii) If Xn -+ x in Xl, then f(xn) -+ f(x) in X 2 . •
Remark 1.2.2 When property (ii) (or, equivalently, (iii)) holds for a
particular point x E X, we say that f is continuous at x. Thus, f is
continuous if, and only if, it is continuous at each point of X. If f is
°
continuous and, for a given E > 0, the <5 > described in statement (ii)
above does not depend on the point x, then the function is said to be
uniformly continuous on X . •

Let {X, d} be a metric space and let E be an arbitrary subset of X.

Let x EX. Define
d(x, E) = inf d(x, y).
yEE

This is called the distance of the point x from the set E. The following
proposition is easy to prove.

Proposition 1.2.3 Let {X, d} be a metric space and let E eX. Then
(i) for all x and y E X, we have

Id(x, E) - d(y, E)I < d(x, y).

1.2 Topological Spaces 15

Thus, the function x f-> d(x, E) is a uniformly continuous function on

X;
(ii) if E is a closed set, then, d(x, E) = 0 if, and only if, x E E; more
generally, if E is any subset of X, we have

E = {x E X I d(x, E) = o}.
Definition 1.2.8 Let {X,d} be a metric space. A sequence {x n } in X
is said to be Cauchy if, for every E > 0, there exists a positive integer
N such that

for every k 2: N, l 2: N . •

It is simple to verify that every Cauchy sequence is bounded. It is also

easy to see that every convergent sequence in a metric space is Cauchy.
The converse is not true and this leads to the following important defi-
nition.

Definition 1.2.9 A metric space is said to be complete if every Cauchy

sequence is convergent . •

With their usual metric, the spaces lR and C are complete.

We now introduce an important notion which we will study in detail
in later chapters.

Definition 1.2.10 Let I be an arbitrary indexing set. Let X be a set

and {Xi, 'Iilia be topological spaces. Let li : X -* Xi be given functions.
The weak topology generated by the functions fi, i E I, is the smallest
topology on X such that all the li are continuous. •

From the above definition it follows that a subbase for the weak
topology generated by the li is the collection of all sets of the form
fi-l(U) where U is an arbitrary open set in Xi and the index i ranges
over the indexing set I. A typical neighbourhood of a point x E X will
therefore be a finite intersection of sets of the form f i- 1 (Ui) where Ui is
a neighbourhood of Ji(x) in Xi.

Definition 1.2.11 Let I be an indexing set and let {Xi, 'Ii}, i E I be

topological spaces. Set X = IIiEIXi . Let x = (Xi)iEI. Let Pi : x E X f->
xi E Xi be the i-th coordinate projection. The product topology on
X is the weak topology generated by the coordinate projections, i. e. it is
the smallest topology such that the projections are all continuous. •
16 1 Preliminaries

Thus, sets of the form IIi EI Ui , where Ui = Xi for all i -I- io (an
arbitrary element of I) and Uio is open in X io ' form a subbase for the
product topology. A base for the topology is the collection of all sets
of the form IIiEIUi where Ui = Xi for all but a finite number of indices
and, for those indices, Ui is an open set in Xi.

Definition 1.2.12 Let {X, T} be a topological space and let 0 -I- K c

X. A collection of open sets F is said to be an open cover of K if
the union of the members of F contains K. A sub cover of F is a
subcollection of members of F which is also an open cover of K . •

Definition 1.2.13 Let {X, T} be a topological space and let 0 -I- K c

X. The set K is said to be a compact set if every open cover of K
admits a finite subcover. •

If X is itself a compact set, we say that it is a compact space. We

can also describe compactness via closed sets.

Definition 1.2.14 A collection A of subsets of a set X is said to have

finite intersection property if every finite subcollection has non-
empty intersection. •

The following proposition is easily proved.

Proposition 1.2.4 A non-empty subset K of a topological space is com-

pact if, and only if, every collection of closed sets in K having finite
intersection property has non-empty intersection . •

Definition 1.2.15 Let {X, T} be a topological space and let 0 -I- K c

X. The set K is said to be sequentially compact if every sequence in
K has a convergent subsequence. •

We list important facts about compact sets in the following propo-

sition.

Proposition 1.2.5 (i) Every continuous image of a compact set is com-

pact.
(ii) The product of compact sets is compact.
(iii) Compact subsets of Hausdorff spaces are closed.
(iv) A compact subset of a metric space is closed and bounded.
(v) For the usual topology on ]RN, a subset is compact if, and only if, it
is closed and bounded.
1.2 Topological Spaces 17

(vi) Every continuous real valued function on a compact space is bounded

and attains its maximum and minimum values in that set.
(vii) Every continuous real valued function on a compact metric space is
uniformly continuous . •
Compact metric spaces are very special. In order to characterize
them, we need the following notion.
Definition 1.2.16 A metric space {X, d} is said to be totally bounded
if, for every E > 0, there exists finite set of points {xi 7~ { such that

X C 7~ { B(Xi; E) .•
Proposition 1.2.6 Let {X, d} be a metric space. The following state-
ments are equivalent:
(i) X is compact.
(ii) X is sequentially compact.
(iii) Every infinite subset of X has a limit point.
(iv) X is complete and totally bounded. •

We conclude this section with one final important topological notion.

Definition 1.2.17 Let {X, T} be a topological space. We say that X
is connected if there do not exist non-empty open sets U and V such
that X = U U V and U n V = 0. A subset A c X is said to be
connected if there do not exist disjoint open sets U and V such that
A = An (U U V), An U of 0, A n V of 0.•

Definition 1.2.18 A non-empty subset A of a topological space is said

to be path connected if given any pair of points x and y in A, there
exists a continuous function, : [0,1] --+ A such that ,(0) = x and
,(1) = y.•

Proposition 1.2.7 (i) Every continuous image of a connected set is

connected.
(ii) The product of connected sets is connected.
(iii) Every path connected set is connected. •
In particular, every ball in a metric space is path connected, and
hence, connected.
The only connected sets in lR. are intervals.
For a detailed study of topological spaces, the reader is referred to,
for instance, Simmons [9].