Functional Analysis Notes

MTH327 Functional Analysis
Abdullah Naeem Malik

March 12, 2014
ii
Contents
Introduction vii
I Preliminaries 1
1 Set Theory 5
2 Group Theory 13
3 Linear Algebra and Set Topology 19
3.1 Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Normed Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3 Set Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3.1 Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3.2 Subspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3.3 Metric Spaces and Norm Spaces . . . . . . . . . . . . . . 55
3.4 Complete Norm Spaces . . . . . . . . . . . . . . . . . . . . . . . 56
II Hilbert Spaces 63
3.5 Pre-Hilbert Space . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.6 Hilbert Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
A Appendix 85
A.1 Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
iii
iv CONTENTS
Preface
v
vi PREFACE
Introduction
Functional analysis is an abstract branch of mathematics that originated from
classical analysis. The name is derived from the word "functional" which is a
particular generalisation of the real-valued function. Historically, the impetus
came from problems related to ordinary and partial dierential equations, nu-
merical analysis, calculus of variations, approximation theory, integral equations
and so on. Ordinarily, one deals with limiting processes in nite dimensional
vector spaces (R or R
n
) but problems arising in the above applications required
a calculus in spaces of functions (which are innite dimensional vector spaces).
Functional analysis is a branch of mathematical analysis, the core of which
is formed by the study of vector spaces endowed with some kind of limit-related
structure (e.g. inner product, norm, topology, etc.) and the linear operators
acting upon these spaces and respecting these structures in a suitable sense.
The historical roots of functional analysis lie in the study of spaces of functions
and the formulation of properties of transformations of functions such as the
Fourier transform as transformations dening continuous, unitary etc. operators
between function spaces. This point of view turned out to be particularly useful
for the study of dierential and integral equations.
The usage of the word functional goes back to the calculus of variations, im-
plying a function whose argument is a function and the name was rst used in
Hadamards 1910 book on that subject. However, the general concept of func-
tional had previously been introduced in 1887 by the Italian mathematician and
physicist Vito Volterra. The theory of nonlinear Functionals was continued by
students of Hadamard, in particular Frchet and Lvy. Hadamard also founded
the modern school of linear functional analysis further developed by Riesz and
the group of Polish mathematicians around Stefan Banach.
In modern introductory texts to functional analysis, the subject is seen as
the study of vector spaces endowed with a topology, in particular innite dimen-
sional spaces. In contrast, linear algebra deals mostly with nite dimensional
spaces, and does not use topology. An important part of functional analysis is
the extension of the theory of measure, integration, and probability to innite
dimensional spaces, also known as innite dimensional analysis.
vii
viii INTRODUCTION
Part I
Preliminaries
1
3
This part of the book will focus on the ground-work, building our way to
the areas which Functional Analysis is famous for. We will start with set theory
in order to set some notions straight, work our way with some group theory.
Some Set Topology as needed has been introduced, as well. The gentle introduc-
tion continues its way slowly by turning from an introduction of norm spaces
and vector spaces, regressing to familiar concepts of metric spaces and then
amplifying the theory of norm and vector spaces with the addition of Cauchy
Sequences.
4
Chapter 1
Set Theory
An attempt to dene all of the symbols and mathematics used in this book has
been undertaken and every eort has been made to ensure that the notes are
self-contained. However, this portion is not intended to be a substitute for any
rigorous treatment. The purpose of this section is to serve as an introductory
reminder for what follows as the main purpose of this book since most of the
students come from a dierent backgrounds. Hence, this section may be skipped,
if necessary. Sections where a slight introduction is necessary have been placed
under appropriate headings. Proofs and extensive examples are intentionally
omitted and have been left as an exercise to the reader. The reader is encouraged
to try to attempt such proofs, apart from the exercise since they will serve as
"warm ups".
Denition 1 A set is any well-dened collection of objects.
The words "well-dened", "collection" and "objects" may be vague and we
will not indulge in commenting any further. An intuitive understanding is as-
sumed. The word "class" will be reserved for a collection of sets.
The stated collection is denoted by capital letters while the members are
listed using curly braces, separated by commas. If an element j in a set ,
then the statement j belongs to or j is in is mathematically written
as j . This very primary concept is the corner stone to place nearly all of
the foundation of mathematics in deriving for ourselves a structure, (a Hilbert
space), which is the main purpose of the book.
Example 2 The empty set . This special case is denoted by ?. All other
sets are non-empty.
Example 3 The collection of those collections of objects that are not members
of themselves is mathematically written as r [ r , r
Elements which do not belong to a set A will always belong to its complement
A
{
. The complement of a set is always dened relative to a universal set.
5
6 CHAPTER 1. SET THEORY
The indexing set 1
n
:= 1, 2, 3, ..., : will be used for convenience. In cases
where an innite number of items have to be indexed, the set 1 = N will be
used. This is particularly important for using countable sets.
Remark 4 We will assume that the set of natural numbers N := 1, 2, 3, ...
exists from which the set of reals can be constructed using the Dedekind Cuts.
For a rigorous treatment on the subject, see Tom M. Apostols Calculus. Also,
keeping in line with the tradition in Pakistan, the usual (classical) logical con-
ventions of exclusive true/false are assumed. "I", "if and only if" and "= "
will mean the usual two-way implication, implying an equivalency. In deni-
tions, the word "if" will refer to "i". The symbol "\" will mean "for all" and
its equivalents whereas the symbol "" will mean "there exists" and its equiva-
lents. Familiarity with the principle of induction and the concept of innity is
assumed. The Axiom of Choice will be used carelessly. The symbols Z, Q, R
and C will denote, respectively, the set of integers, rationals, real and complex
numbers. Bars over elements will denote conjugates.
Denition 5 The intersection, union, dierence and symmetric dier-
ence of any two sets and 1 is dened, respectively, as ' 1 = r [ r
and r 1, 1 = r [ r or r 1, 1 = r [ r and r , 1
and 1 = r [ r or r 1 but r , 1.
Denition 6 Let and 1 be two sets. 1 is called a subset of if \/ 1,
/
This is written as 1 for proper subsets. The notation "_" is reserved
for improper subsets i.e. 1 or 1 = .
Remark 7 Two sets are considered equal if they are subsets of each other.
We will denote the power set of a set A (the collection of all subsets of A)
with T(A), instead of 2
.
Denition 8 Let and 1 be two sets. Then, product set or Cartesian
product of and 1, written 1, consists of all ordered pairs (a, /) where
a and / 1 i.e. 1 = (a, /) : a , / 1
It is clear that if has : elements and 1 has : elements, then 1 has ::
elements. Intuitively, this makes sense and we will make do with the intuitive
concept of cardinality only and not indulge in more rigour since the length
of the notes might get unnecessarily compromised. The use of the concept of
cardinality is particularly limited in the notes, anyway, except, for instance, in
places where a countable set is used in a denition, among others. For rigorous
treatment of the same, see any standard text book of set theory.
Example 9 If = 1, 2, 3 and 1 = 4, 5, 6, then,
1 = (1, 4), (1, 5), (1, 6), (2, 4), (2, 5), (2, 6), (3, 4), (3, 5), (3, 6)
7
It makes sense to dene =
2
. Furthermore, (1) C and
(1 C) are equal (see exercise).
Remark 10 An alternative way to dene an ordered pair (a, /) := a, a, /
was proposed in 1921. This denition is called the Kuratowski denition and it
is now the standard denition of an ordered pair in set theory.
Denition 11 Let and 1 be two sets. Then, a binary relation 1 is a
subset of 1.
For a , / 1, if a is related to /, then (a, /) 1. This is compactly
written as a1/. If a is not related to /, then a / 1/. A binary relation is, shortly,
said to be dened on if it is a subset of
2
.
Example 12 For 1 = (1, 1), (2, 2), (3, 3), ... N N, a1/ =a = /
Denition 13 Let < be a binary relation. A strict weak ordering is a binary
relation < on a set o with the following propoerties:-
< is transitive (a < / and / < c =a < c)
< is irreexive (a < / =/ a)
< is asymmetric (a a)
Example 14 < is said to be an order on N, written as (N, <), if the elements
of N N satisfy all the above properties in which the former member of the
ordered pair is less in magnitude than its latter member, as is the order "",
which is usually read as "contained in" for sets.
Denition 15 Symmetric, transitive and reexive relations are called equiva-
lence relations.
The denition makes sense because elements of a set with an equivalence
relation generate a class within a set of "equal" elements.
Example 16 For a, / Z, 1 is an equivalence relation if 1 ZZ such that
a / is an integral multiple of 5.
Denition 17 The elements of 1 _ T(A) are called pairwise disjoint if
\, 1 1, either = 1 or 1 = ?.
Denition 18 A collection of proper subsets of the set A is called a parti-
tion of A if
1. The elements of are pairwise disjoint and
2.
_
.
= A
where the second bullet indicates that the union is taken over the subsets
of A. In the exercise, you will be asked to show that this is a second way of
thinking of equal objects. The example below should make this clearer.
Example 19 For the set of rationals Q := ZZ such that r = (a, /) Q =a =
r/, we can get a partition, the elements of which are denoted by [(:, t)] := (a, /)
[ a, / Z and /: = at, when the relation (a, /)~(c, d) holds for ad = /c. Such a
partition is also an equivalence class (proof ?).
A set A which can be partitioned under an equivalence class ~ is denoted
by A,~. For the set of rationals, we have Q,~ = [
_
1
2
,
_
2
3
,
_
3
4
, ..., [1] , [2] , ....

This is called the quotient set. This is an important concept since a set endowed
with a structure, for example a space or a group, if partitioned under ~, will
inherit the structure. The partitioning can be thought of as a division of the
set, which suggests the use of the symbol , in a quotient set, and the name. As
the example illustrates, the quotient set may be thought of as a set with all the
"equivalent" points identied and clumped together.
Denition 20 Let _ be a binary relation. A partial order is a binary relation _
over a set o if \a, /, c o
a _ a (_ is reexive)
a _ / and / _ a = a = / (_ is antisymmetric)
a _ / and / _ c = a _ c (_ is transitive)
Example 21 For any set A, T(A) forms a partial order (is a partially ordered
set) under the partial order "_"
A nal property called totality is added to this list to dene _ as a total
order. This mathematically says that a _ / or / _ a \a, / o. This signies a
comparison of all elements in o.
Denition 22 Let (o, _) be a totally ordered set and o . is said to be
bounded above if / o such that a _ / \a
Example 23 1, 2, 3, 4 N is bounded above by 5 whereas 2, 3, 4, ... is not
bounded.
Denition 24 Let (o, _) be a totally ordered set and o is bounded above.
, o is said to be the least upper bound, or supremum of if
, is an upper bound of
c < , =c is not an upper bound of
In such a case, , = sup. Such a , is unique (proof?)
9
Denition 25 Let A and 1 be two sets. Then, a function ) from A to 1 is
an object such that every r A is uniquely associated with an object )(r) 1 .
This is represented by ) : A 1 . The identity map will be represented
by
^
1 throughout the text. This is a map such that 1(r) = r. Confusion with
numerical one should not arise since the usage will be clear from context.
Example 26 A specic type of function called Boolean function used in Boolean
algebra may be dened as ) : 0, 1 0, 1. For : N, this denition may
be extended by using an :-tuple input formed by taking the Cartesian product
of 0, 1 :-times i.e. 0, 1
n
A function is a relation that uniquely associates members of one set with
members of another set. A function is therefore a many-to-one (or sometimes
one-to-one) relation. More rigorously, ) is one-to-one if )(r) = )(j) implies
r = j for r, j A. A function is onto if )() = 1. The set of values at
which a function is dened is called its domain. The domain of ), , will be
reserved by the symbol T()) . The set of values that the function can produce
is called its range or image. The set )() _ 1 is called the image of under
) and will be denoted by ()) . The same symbol for "imaginary part of" is
reserved for complex numbers only and hence confusion with functions is out
of question. This image is a subset of the codomain 1. For 1 1 , the set
)
1
(1) := r A [ )(r) 1 is called the inverse image of ) and does not
require for the function to have an inverse (see exercise).
The term map is synonymous with function. However, every function is a
mapping but not every mapping (or map) is a function since a mapping might
also take one element and map it to many others. The notation is reserved
for function composition. A function will be called bijective or as having a
one-to-one correspondence if it is one-to-one and onto.
A set is countable if it can be put in a one-to-one correspondence with the
set of natural numbers. For instance, the set of natural numbers (trivially),
integers and rationals are countable whereas the set of reals are not. For a
beautiful proof of the usage of the concept, see Cantors diagonal argument
in any standard text book of Set Theory. In notation, this correspondence is
denoted by N ~ Z, N ~ Q. We shall make little use of this.
Technically, a sequence is any function from the natural numbers to any
given set. So, if we have )(:) = :, we can have a sequence of natural numbers,
or an arithmetic sequence of unit dierence. Any such range is called the range
of the sequence. As seen, this range can be an integer, a real number, a natural
number, a complex number, a function or even a set. We will talk about the
convergence of a sequence and their doing in metric and norm spaces in detail,
later. For now, the ideas of Calculus will suce i.e. a sequence r
n
converges to
a point r if lim
no
r
n
= r
Exercise
1. For any subsets , 1 and C of a set A, prove the following:
(a) ' 1 = 1 '
(b) 1 = 1
(c) (' 1) ' C = ' (1 ' C)
(d) ( 1) C = (1 C)
(e) ' (1 C) = (' 1) (' C)
(f) (1 ' C) = ( 1) ' ( C)
(g) T() ' T(1) _ T(' 1)
(h) T() T(1) = T( 1)
(i) 1 = 1
{
(j) _ 1 =1
{
_
{
(k) (' 1)
{
=
{
1
{
(l) ( 1)
{
=
{
' 1
{
(m) ' C = C = C
(n) C = = C
2. For any set A, prove that T(A) is unique. Do this by showing that there
are two such classes and then show that they are the same by employing
set-theoretic arguments
3. For any sets , 1 prove that 1 = 1 = 1 = hence the use of
=
2
is justied.
4. If , 1 and C are non-empty sets, prove that there exists a one-to-one
correspondence between
(a) 1 and 1
(b) (1) C and (1 C)
(c) (1) C and the ordered triples (a, /, c) where a , / 1 and
c C
5. Prove that an equivalence relation yields a partition of a set A and con-
versely.
6. Show that the mapping ) : A 1 is bijective if and only if there exists
a mapping ) : 1 A such that q ) = ) q = 1
7. Let be any indexing set. For ) : A 1 , A and 1 1 , prove
the following: (remember, the following are sets, so youll have to use
set-theoretic arguments)
(a) )()
1
(1)) _ 1.
(b) _ )
1
()())
11
(c) )(
_
I
I
) =
_
I
)(
I
)
(d) )(
I
) _

I
)(
I
)
(e) Equality holds in d if ) is one-to-one
(f) )
1
(1
c
) =
_
)
1
(1)
_
c
(g) )
1
(
_
I
1
I
) =
_
I
)
1
(1
I
)
(h) )
1
(
I
1
I
) =

I
)
1
(1
I
)
8. For ) : A 1 and q : 1 7, show that
(a) ) and q are one-to-one implies q ) is one-to-one
(b) ) and q are onto implies q ) is onto
(c) q ) is one-to-one implies ) is one-to-one
(d) q ) is onto implies q is onto
9. Show that ) : A 1 is one-to-one if and only if q : 1 A such that
q ) =
^
1
10. Show that ) : A 1 is onto if and only if q : 1 A such that

) q =
^
1
Y
11. A permutation on any set A is a mapping : A A. Let o(A) be a
set of permutations on A. Show that, for c, ,, o(A),
(a) c , o(A)
(b) c (, ) = (c ,)
(c)
^
1 o(A) such that c
^
1 =
^
1 c = c
(d) \c o(A), c
1
o(A) such that c c
1
= c
1
c =
^
1
Chapter 2
Group Theory
Again, only a few denitions are presented just for recollections sake and this
section, too, may be entirely skipped without any loss of continuity; this topic
is not a part of the basic curriculum of the subject. A general exposition of
groups is considered from the set-theoretic point of view. Only one lemma and
one proof has been incorporated with proof.
Denition 27 Let o be a set. Then, a binary operation + on o is a function
such that + : o o o.
This may be extended to include more than two elements but is beyond the
scope of these notes.
Denition 28 Let G be a set. Then, G, together with a binary operation +,
written as (G, +), is a group if it satises the following axioms:-
\ a, /, c G, (a + /) + c = a + (/ + c) i.e. + is associative.
c G such that a + c = a, \ a G
\ a G, / G such that a + / = c
Remark 29 The third axiom is possible only when the uniqueness of c is jus-
tied (see exercise).
(G, +) will be shortened to G, where possible. Also, the "+" symbol will be
skipped and juxtaposition will be used instead. An Abelian group is a group
in which + is commutative i.e. \a, / G, a + / = / + a. Strictly speaking,
it is not generally true that the existence of a right identity and right inverse
implies the existence of a left identity and left inverse in a binary relation unless
the relation is associative (see exercise). This explains why most denitions of
a group assume that both left and right inverse exist and the left and right
identities hold.
13
14 CHAPTER 2. GROUP THEORY
Example 30 Z is a group under the binary operation + : Z Z Z where
for a, / G, +(a, /) = a +/
Lemma 31 For a group G and for q
1
, q
2
, ..., q
n
G,
q
1
q
2
...q
n
=
n
I=1
q
I
This lemma might sound trivial but the reader is reminded that such an im-
pression comes from taking for granted that (q
1
q
2
) q
3
= q
1
(q
2
q
3
). For example,
(5 3) 2 ,= 5 (3 2). Of course, in a group, this has to be true. This the-
orem basically states that the brackets can be removed without any ambiguity.
It thus makes sense to dene q
n
= qq...q :-times. Note that the above is valid
for a nite :.
Proof. We use induction to show that for every : any meaningful product
q
1
q
2
...q
n
=
n
I=1
q
I
. Since the axioms of a group dictate that this is true for
: = 1, 2 and 3 we move on to consider a general :. Let : < :. Then, q
1
q
2
...q
n
=
(q
1
q
2
...q
n
) (q
n+1
q
n+2
...q
n
)
=
_
n
I=1
q
I
__
nn
I=1
q
n+I
_
=
_
n
I=1
q
I
___
nn1
I=1
q
n+I
_
q
n
_
=
__
n
I=1
q
I
__
nn1
I=1
q
n+I
__
q
n
=
_
n1
I=1
q
I
_
q
n
=
n
I=1
q
I
The denition of a isomorphism will be skipped entirely since little use is
made of them in the notes. This point is just presented so as to remark the
following: an isomorphism is a concept in which two groups are equal. The ele-
ments behave in a particularly similar fashion. This corresponds to saying that
the groups have a similar multiplication table. The reader will recall that the
real numbers under multiplication and addition are isomorphic under the iso-
morphism )(r) = c
r
. Lattices are isomorphic when orders between elements are
preserved. Graphs are isomorphic when they both have a similar structure, with
vertices and nodes being the same. Metric spaces are isomorphic when distance
is preserved. Topological spaces are isomorphic when arbitrarily small distances
are preserved between the two topological spaces. In short, two mathematical
structures are isomorphic or the same if their underlying structures are similar,
with a disregard to nature of the elements. This concept is very important in
mathematics.
15
Denition 32 A subgroup H of a group (G, +) is a set together with all the
axioms of the group.
In short, if we restrict the action of the binary operators with the axioms
mentioned to a set H, we get a subgroup. Needless to say, the binary operation
induced on H is that of G. Mathematically, a subgroup is represented by H _ G.
In analogy to set theory, the improper subgroups of a group are c and G. The
identities and inverses of the group and subgroup should necessarily coincide
(proof?). One way to check whether we have a bonade subgroup is by seeing
that it satises the axioms. Another way is to use the following:
Theorem 33 H is a subgroup of G == for any a, / H, a/
1
H.
Proof. We should rst prove that H is non-empty. If it is a subgroup, then
it will at least contain c. Since H is a subgroup, if / H, then /
1
H.
Also, if a, /
1
H, then a/
1
H. conversely, suppose that for any a, / H,
a/
1
H. First we prove associativity. For a, /, c H, we know that a, /, c G,
so associativity is trivially proved for any three elements in H. From Since
a, / H implies a/
1
H, we can reverse the roles of the elements and can
conclude that /a
1
H. Since we know that these two elements belong to the
group, from this we have
_
a/
1
_ _
/a
1
_
= aca
1
H. This step is valid since /
is an element of a group and weve already established associativity. Hence, we
have aca
1
= aa
1
= c H. Finally, 1, a H implies 1a
1
= a
1
H. Lastly,
a H and /
1
H implies a
_
/
1
_
1
H or that a/ H, establishing that H
is closed.
Denition 34 Let F be a set. F is eld with two binary operations + and .,
denoted by (F, +, .), satisfying the following axioms:-
(F, +) is an abelian group
(F
+
, .) is also an abelian group.
\ a, /, c F, the left distributive law a.(/ + c) = a./ + a.c and the right
distributive law (/ +c).a = /.a + c.a hold
Here, F
+
= F 0. x F
n
is a vector dened by an :-ordered pair for
: N. The same rules for the eld are applied to each ,th-tuple for 1 _ , _ :.
Exercise
1. Prove that the set 22 of matrices
_
a /
c d
_
forms a group under multi-
plication for ad ,= /c. (see appendix for a review of matrices).
2. Prove that o(A), the set of permutations on any set A, is a group.
3. Decide whether or not the following is a eld. If it is, state a proof. If it
isnt, state a counter-example:
(a) Q under usual multiplication and addition.
(b) Z under usual multiplication and addition.
(c) R 0 under usual multiplication and addition.
(d) C under usual multiplication and addition.
(e) Set of continuous functions with point-wise addition and multiplica-
tion.
4. Let (G, +) be an Abelian group, a nonempty set and '(, G) the set
of all functions ) : G. Dene + : '(o, G)'(o, G) '(o, G)
as () + q) : G given by )(a) + q(a) G for a . Prove that
'(, G) is an Abelian group.
5. In a group (G, +) and for any a, /, c G, prove that following
(a) the identity c and the inverse a
1
for any element a are both unique.
(b) a
2
= a =a = c
(c) a
2
= c =G is Abelian
(d) a/ = ac = / = c and /a = ca = / = c (this is called the left and
right cancellation law)
(e)
_
a
1
_
1
= a
(f) (a/)
1
= /
1
a
1
(g) for unknown r, j G, the solutions to the equations ar = / and
ja = / exists and is unique
6. Show that the intersection of countably nite groups is a group but that
the union of countably nite groups is not necessarily a group. (Hint: start
with two groups).
7. Prove that the existence of left inverse and left identity, as has been dened
for the denition of a group, is implied by the existence of a right inverse
and right identity.
8. Let ~ be an equivalence relation on a group G such that q
1
~ q
2
and
/
1
~ /
2
imply q
1
/
1
~ q
2
/
2
for all q
I
, /
I
G where i = 1, 2. Then, prove
that the set G, ~ of all equivalence classes of G under ~ is a group under
the binary operation dened by [q] [/] = [q/] .
9. Prove that if G is an Abelian group, then so is G, ~ .
10. Let (G, +) be a group. Prove that if G is Abelian, then for q, / G and
\:, :
17
(a) (q/)
n
= q
n
/
n
.
(b) q
n+n
= q
n
q
n
(c) (q
n
)
n
= q
nn
Chapter 3
Linear Algebra and Set
Topology
This is where the course begins and attention will be paid to rigour and detail.
Basically, linear algebra has to do with the algebra of matrices, vectors and
the spaces formed by the collection of either. The idea of a three-dimensional
vector can be viewed as a member of R
3
. This is called the Euclidean space. The
addition, subtraction, scalar and cross multiplication of two vectors is generally
well-known and so is the algebra of matrices (a recollection is added in the
appendix). However, a mathematical treatment such as the one at our disposal
is far more general and has in store some great surprises. Vectors are not vectors
anymore in the usual sense yet sound very familiar. Their algebra, too, looks
downright distinct but a closer analysis reveals surprising commonalities. In
mathematics, the general idea is look for a common structure, formulate some
rules such structures obey and then see where else the structure lies. Of course,
intuition may play a role in the converse.
Anyway, vector spaces are exactly such spaces. It is to be remarked that
mathematically, a space is any set endowed with a particular structure. Hence,
vector spaces are just that a set of vectors with additional structure.
3.1 Vector Spaces
Addition and scalar multiplication of vectors can be intuitively understood.
For a set of vectors \ , by addition we mean a rule for associating with each
pair of objects u, v \ an object u + v, called the sum of u and v; by scalar
multiplication, we mean a rule for associating with each scalar c and each object
u \ an object cu, called the scalar multiple of u by c. A scalar is an object
that is a part of a eld F. We say that the vectors are scaled over the eld F.
If a restriction F = R or C is added, then we have a linear vector space.
More rigorously,
19
20 CHAPTER 3. LINEAR ALGEBRA AND SET TOPOLOGY
Denition 35 Let \ be a non-empty set and F be a eld. Then, \ is a vector
space over F if for + : \ \ \ and . : F \ \
1. u +v \ , as implied by the denition of the function +.
2. u +v = v +u
3. u + (v +w) = (u +v) +w
4. There is an object 0 in \ , called a zero vector for \ , such that 0 + u =
u +0 = u for all u in \ .
5. For each u in \ , there is an object u in \ , called a negative of u, such
that u + (u) = (u) +u = 0.
6. If c is any scalar and u is any object in \ , then cu is in \ , as implied by
the scalar multiplication function.
7. c(u +v) = cu +cv
8. (c + ,)u = cu +,u
9. c(,u) = (c,)(u)
10. 1u = u
for u, v, w \ and c, ,, 1 F. The rst ve axioms for the vector space
may be compressed to say that \ is an Abelian group for vector multiplication.
The fth axiom may be derived from the fact that c = 1 is also a scalar
in F, following from the tenth axiom. One can imagine that such concepts can
readily be interpreted physically for the usual convention of "arrows" in physics.
However, the above axiomatisation is a vast generalisation of other spaces, as
well. Functions and even sequences can be interpreted as vectors in the above
sense, as can be seen from the examples.
Exercise 36 Prove the following:
1. 0x = 0
2. c0 = 0
3. (1)x = x
Example 37 The set R
n
is a vector space over the set of reals R under the
"usual" vector addition and scalar multiplication. For x = (r
1
, r
2
, ..., r
n
) and
y = (j
1
, j
2
, ..., j
n
) R
n
and for c R, we can dene vector addition and scalar
multiplication as follows: + : R
n
R
n
R
n
such that
+(x, y) = x +y
= (r
1
, r
2
, ..., r
n
) + (j
1
, j
2
, ..., j
n
)
= (r
1
+ j
1
, r
2
+ j
2
, ..., r
n
+ j
n
)
3.1. VECTOR SPACES 21
and for . : R R
n
R
n
such that
. (c, x) = cx = c(r
1
, r
2
, ..., r
n
) = (cr
1
, cr
2
, ..., cr
n
)
To begin proving that these operations dened in the manner above do indeed
form a vector space, the denitions will have to be applied directly. To see this,
clearly,
x +y =(r
1
+ j
1
, r
2
+j
2
, ..., r
n
+ j
n
) R
n
since r
I
+j
I
R \i 1
n
. Next, x +y = y +x because addition is commutative
in the set of reals. Again, inverses will exist for any x = (r
1
, r
2
, ..., r
n
) since we
can construct x = (r
1
, r
2
, ..., r
n
) by resorting to the fact that every tuple,
being real, has an inverse. Furthermore, since 0 R, then 0 =(0, 0, ..., 0) R
n
and is an additive identity. Therefore, R
n
is an abelian group. Axiom 6 is
satised by denition. For Axiom 7,
c(x +y) = c(r
1
+ j
1
, r
2
+ j
2
, ..., r
n
+ j
n
)
using the rule of addition. Since the element (r
I
+j
I
) is a member of the set of
reals, therefore (c(r
1
+ j
1
) , c(r
2
+ j
2
) , ..., c(r
n
+ j
n
)) is justied. Further-
more, R is a eld, hence c(r
I
+j
I
) = cr
I
+ cj
I
. Thus,
c(r
1
+ j
1
, r
2
+ j
2
, ..., r
n
+j
n
)
= (c(r
1
+ j
1
) , c(r
2
+ j
2
) , ..., c(r
n
+ j
n
))
= (cr
1
+cj
1
, cr
2
+ cj
2
, ..., cr
n
+ cj
n
)
Now, since cr
I
and cj
I
are real numbers, we can split (cr
1
+ cj
1
, cr
2
+cj
2
, ..., cr
n
+ cj
n
)
to get (cr
1
, cr
2
, ..., cr
n
) + (cj
1
, cj
2
, ..., cj
n
), which is, according to our den-
ition of the scalar multiplication, c(r
1
, r
2
, ..., r
n
) + c(j
1
, j
2
, ..., j
n
). Thus, we
have justied c(x+y) = cx+cy. For the 8th Axiom, we follow a similar series
of steps:
(c + ,)x
= (c + ,) (r
1
, r
2
, ..., r
n
)
= ((c +,) r
1
, (c + ,) r
2
, ..., (c + ,) r
n
)
= (cr
1
+,r
1
, cr
2
+ ,r
2
, ..., cr
n
+,r
n
)
= (cr
1
, cr
2
, ..., cr
n
) + (,r
1
, ,r
2
, ..., ,r
n
)
= c(r
1
, r
2
, ..., r
n
) + , (r
1
, r
2
, ..., r
n
)
= cx +,x
The last two axioms can be similarly proved.
Notice how everything reduces to the pre-established axioms and the fact
that the base is a eld and that the tuples form from the same eld. This
painstaking mode of argument was only meant to serve as a motivation for how
a vector space ought to be proved: the axioms must hold!
To prove that R
n
is a vector space over the eld Q, we can similarly resort
to the fact that each tuple, that is, every real number when added or multiplied
by a rational number will yield another real number.
Equivalently, we can take :-tuple of the Complex plane C and generate for
ourselves a vector space. Try and prove it rst for : = 1 and then proceed via
induction.
Example 38 The space C[a, /], the set of all continuous real valued functions
dened on the interval 1 = [a, /], forms a real vector space with the algebraic
operations dened in the usual way:
(r +j)(t) = r(t) + j(t)
(cr)(t) = c(r(t)) for / R
The use of r and j as function is rather intentional. We will prove that the
sum of continuous numbers is continuous. Since were using the real line, well
use the c c denition of continuity, dened in real analysis. To recall,
Denition 39 For , 1 R, a function r : 1 is continuous at a point
t
0
if for every c 0, there exists a c 0 such that [r(t) r(t
0
)[ < c whenever
[t t
0
[ < c. A function is continuous if it is continuous at every point of the
domain.
Now, if the function r is continuous, then for any point t
0
, we have [r(t) r(t
0
)[ <
c,2 whenever [t t
0
[ < c
1
and if j is continuous at the same point, then we have
[j(t) j(t
0
)[ < c,2 whenever [t t
0
[ < c
2
. Take c = min(c
1
, c
2
). Then,
[(r + j) (t) (r + j) (t
0
)[ = [r(t) + j(t) r(t
0
) j(t
0
)[
= [[r(t) r(t
0
)] + [j(t) j(t
0
)][
_ [r(t) r(t
0
)[ +[j(t) j(t
0
)[
< c,2 + c,2 = c.
The same can be said for the continuity of (cr)(t) if we let [r(t) r(t
0
)[ <
c,c. Hence the denitions for the addition and scalar multiplication are justied.
It remains rather traditional now to prove that the space of continuous func-
tions is a vector space by showing that under the dened "vector" addition and
scalar multiplication, the remaining 8 axioms hold. In particular, one should
prove that the zero function is continuous; any additive inverse of a continuous
function is also continuous.
Hopefully, you have just proved that this collection of functions forms a
vector space. In other words, functions can be viewed as vectors in a sense a
single point in space. Such a space is also called the function space.
Wherever possible, vectors will be made bold. However, this will not be
strictly followed, especially when it comes to functions and sequences, since this
may be a cause of confusion with some pre-established notations. This should
in no way mean that functions are not vectors.
3.1. VECTOR SPACES 23
Example 40 The set of polynomials 1[a, /] of order at most : over the set
[a, /] is also a vector space. Try to prove this yourself. Take the element
1(r) =
n
I=0
c
I
r
I
and prove that this forms a vector space under the "ordinary" addition and
scalar multiplication.
Example 41 The space |
2
(the space of two summable convergent and hence
bounded sequences) with the algebraic operations dened similar to those for
:-tuples in connection with sequences, that is, for , |
2
such that
o
I=1
[
I
[
2
<
we have
+ & = (
1
,
2
, ..) + (
1
,
2
, ...) = (
1
+
1
,
2
+
2
, ..)
and
c = (c
1
, c
2
, ..)
forming a vector space.
Proof. To show that these operations are well-dened i.e. the sum of bounded
sequences is a bounded sequence and the scalar of convergent sequence is a
convergent sequence, the proof will be based on exact analogy as above but this
will have to wait until we know what a bounded sequence really is. Clearly, we
can have +(& + ) = ( + &) +, by resorting to the associativity of each ith
element. Next, the sequence 0 = (0, 0, 0, ...) converges and hence belongs to |
2
so that
+0 = (
1
,
2
, ..) + (0, 0, ...)
= (
1
+ 0,
2
+ 0, ..)
= (
1
,
2
, ..)
=
Furthermore,
o
I=1
[
I
[
2
=
o
I=1
[
I
[
2
so that for every , we have . For commutativity, again, we resort to the fact
the base eld is commutative so that
+ & = (
1
+
1
,
2
+
2
, ..)
= (
1
+
1
,
2
+
2
, ..)
= & +
Now that additive operation forms an Abelian group, we can similarly prove
that c( + &) = c + c&, (c + ,) = c + ,, c(,) = (c,) and 1 =
These are known as Hilbert sequence spaces. This should not to be confused
with the Hilbert space, which will be studied in the coming chapters. A remark,
however, has to be in order: this was the rst example of a Hilbert space
presented in history. From this, one can get a sense that Hilbert spaces are
not just about vectors in the usual sense and that their development had rather
broad applications in mind.
In order to avoid confusion, we will refer to the space in the example as the
sequence space. In this example, we have |
for j = 2. This j can range between

1 _ j < and the 2 in [
I
[
2
< gets replaced accordingly. For j = ,
the only requirement is that we have bounded sequences i.e. we have [
I
[ _ c
r
where c
r
is a real number which may depend on r but does not depend on i.
This is natural since the the bound will depend on the sequence r but has to
be valid for all i. In fact, we might even have a divergent series if this constant
depended on each i, giving us an unbounded sequence.
The quick eye should have noted the idea of boundedness is entirely depen-
dent on how distance is dened. This topic can be put aside until norm spaces
are dened.
If the sequences for j = 2 are real-valued, then the space is a collection
of all convergent sequences on the real line, which forms a vector space, as
mentioned. Instead of real numbers
I
, we can also have complex numbers as
the base elements of the sequences. Note that this should not be confused with
the Lebesgue Space 1
, which deals with functions with a j-norm. The |
space
is a special case of the 1
space, which will sadly not be covered in these notes

since it is assumed that students will have very little knowledge about measures
in general and Lebesgue measurable functions in particular. This should in no
way imply that 1
spaces are of little importance.

We havent dened the modulus and norm operations ([.[ and |.|, respec-
tively) and have made due with our "usual" understanding of them. They will
now be dened more rigorously.
3.2 Normed Spaces
Denition 42 The modulus of a real number r is a function dened such that
[r[ =
_
r if r _ 0
r if r < 0
This modulus simply converts negative numbers to positive numbers and lets
positive numbers be. In reality, it actually measures the distance of a number
from the zero point. In other words, the magnitude of the one-dimensional ray
formed on the real line is measured by this denition.
The above denition was meant to serve as a gentle introduction what the
modulus (actually, norm) is about. But enough with gentle introductions! Using
this denition, the reader is invited to ex his muscles by proving the following:
3.2. NORMED SPACES 25
Exercise 43 For any real a _ 0, then [r[ _ a == a _ r _ a
Exercise 44 [r[ = 0 == r = 0
Exercise 45 [rj[ = [r[ [j[
Exercise 46 [r[ = [r[
Exercise 47

=
]r]
]]
for [j[ , = 0
Exercise 48 [r j[ = [j r[
Exercise 49

r
2
= r
2
Exercise 50 [r[ =
_
r
2
Exercise 51 [r + j[ _ [r[ +[j[
Exercise 52 [r j[ _ [r[ +[j[
Exercise 53 [r[ [j[ _ [r j[
Exercise 54 [[r[ [j[[ _ [r j[
We will only prove the triangle inequality, since it will be made fair use of.
We start with [:[ _ : _ [:[ and [t[ _ t _ [t[ for :, t R. Adding these
two together, we have ([:[ +[t[) _ : +t _ [:[ +[t[ and this is exactly the rst
statement for : + t = r and [:[ +[t[ = a.
Denition 55 Let be a linear space over a eld F, where F = R or C. A
norm on is a real-valued function |.| : [0, ) such that
N1: |x| _ 0 for all x and |x| = 0 if and only if x = 0
N2: |cx| = [c[ |x| for all c F, x
N3: |x +y| _ |x| +|y| for arbitrary x, y
Imagine that these denitions apply to two dimensional vectors. Then, jus-
tify to yourself that they apply to three dimensional vectors as well. Theres
no reason to stop there and we can continue with :-dimensional spaces, as well
(dont worry, the example done below covers just that). Moving on, as can be
sensed, this is a generalisation of the concept of "measure" of a vector or its
length. A linear vector space together with a norm dened on it is called a
normed space. For our purposes, we will restrict the eld to be R or C. This
restriction leads us to the name "linear normed space" and this will be denoted
by (, |.|). Since the use of the extra brackets is tedious, we will often shorten
it to just the name of the set. Bear in mind that the norm on the same set
makes a good deal of a dierence.
There is absolutely no reason why we should restrict ourselves to two elds.
Using the eld of rational numbers might prove to be physically easy but imagine
using the quaternions or even the eld of functions as scalars!
The rst part of N1 is the positivity property or positive deniteness, N2
is called the Absolute Homogenity Axiom whereas condition N3 is called the
Triangle Inequality or the Subadditive property. The second part of N1 ensures
that any two given vectors are two separate points. The rst intuitive example
of the magnitude of two vectors will serve as a motivation for the denition.
Example 56 Let R
2
be the Euclidean space. The length of a vector, understood
to be derived from the Pythagorean theorem, is |x| =
_
(r
1
)
2
+ (r
2
)
2
. Prove
that this is indeed a normed space. We will prove the more general norm
|x| =
_
(r
1
)
2
+ (r
2
)
2
+ ... + (r
n
)
2
for the space R
n
. Clearly, this norm is positive since were dealing with square
roots. The only way
_
(r
1
)
2
+ (r
2
)
2
+ ... + (r
n
)
2
or |x| could be zero is when
r
1
= r
2
= ... = r
n
= 0. This in turn implies that x = (0, 0, ..., 0) or x = 0. To
prove N2, we have
|cx| =
_
(cr
1
)
2
+ (cr
2
)
2
+ ... + (cr
n
)
2
since cx = (cr
1
, cr
2
, ..., cr
n
). We then have
_
(cr
1
)
2
+ (cr
2
)
2
+ ... (cr
n
)
2
=
_
c
2
_
(r
1
)
2
+ (r
2
)
2
+... (r
n
)
2
_
=
_
c
2
_
_
(r
1
)
2
+ (r
2
)
2
+... (r
n
)
2
_
= [c[
_
_
(r
1
)
2
+ (r
2
)
2
+ ... (r
n
)
2
_
= [c[ |x|
N3 follows from Minkowskis inequality, which has been introduced in the
next lecture.
The discussion of : = 1 in R
n
as a denition of the modulus [r[ =
_
r
2
can
now tell us that the real numbers are vectors in one dimension on the real line.
For : = 2, the proof demanded in the start of the proof can be demonstrated.
Note that the norm dened as above, the Euclidean norm, is not the only
norm on R
n
. For instance,
Example 57 we can dene |.| : R
n
[0, ) such that for x R
n
,
|x| =
n
I=1
[r
I
[
3.2. NORMED SPACES 27
Then, clearly r
I
_ 0 and
|x| = 0
==
n
I=1
[r
I
[ = 0
== r
I
= 0\i
== x = 0
Furthermore,
|cx| =
n
I=1
[cr
I
[
=
n
I=1
[c[ [r
I
[
= [c[
n
I=1
[r
I
[
= [c[ |x|
N3 follows by inductively applying the triangle inequality for real numbers. That
is,
[r
1
+ r
2
+ ... + r
n
[
_ [r
1
[ +[r
2
+ ... +r
n
[
.
.
.
_ [r
1
[ +[r
2
[ +... +[r
n
[
Hence, we have
|x +y| =
n
I=1
[r
I
+ j
I
[
_
n
I=1
[r
I
[ +
n
I=1
[j
I
[
= |x| +|y|
.
Other norms on the same set yield dierent norm spaces, as discussed in the
lectures. Rephrasing the lectures, two dierent norms on the same set do not
necessarily generate the same norm space. The norms are dened according to
dierent uses and priorities.
Example 58 The norm of |
was hinted at in the example for vector spaces;

for any sequence r |
, we have
|r| =
_
o
I=1
[
I
[
_
1/
for 1 _ j < . The proof of this sequence space being a bona de norm
space follows a similar pattern as that for R
n
. Here, one can use the fact that
[c
I
[
= [c[
[
I
[
courtesy of the denition of the modulus, given that c is a

real number (prove it). The proof of N3 is the proof of the Minkowski inequality,
which is introduced in the 2nd lecture.
Example 59 The denition for j = can be traced from lecture 1 and can be
proved by the reader.
Example 60 C[a, /], the space of continuous functions in the interval [a, /] for
a < / is a normed space under the norm such that for any r(t) C[a, /], we
have
|r(t)| = max
|[o,b]
[r(t)[
N1 holds clearly since (a) the modulus is always positive and (b) if the maximum
value of a non-negative quantity is 0, then that quantity is itself 0. That is,
max
|[o,b]
[r(t)[ = 0 == r(t) = 0(t)
where 0(t) = 0 \t. Next, the scalar c does not range over t, hence
max
|[o,b]
[(cr)(t)[ = max
|[o,b]
[cr(t)[ = [c[ max
|[o,b]
[r(t)[
proving N2. N3 follows by the use of the triangle inequality of real numbers.
We can also dene a norm on C[a, /] such that
|x| = |r(t)| =
b
_
o
[r(t)[ dt
The reader is invited to prove that this is norm space.
The concept of a norm is important because length of a vector, distance be-
tween two vectors and hence the idea of convergence can be dened accordingly.
Needless to say, the idea of convergence is essential in almost all areas of mathe-
matics. Also, based on the denition of the norm and the vectors, convergence,
magnitude and distance can then be decided for, as we shall see.
3.3. SET TOPOLOGY 29
3.3 Set Topology
Denition 61 Let A be any non-empty subset. A metric dened on A is a
function d : A A [0, ) such that
D1 d(r, j) _ 0
D2 d(r, j) = 0 == r = j
D3 d(r, j) = d(j, r)
D4 d(r, j) _ d(r, .) + d(., j)
for r, j, . A. A metric space is a pair (A, d) where A is a set and d is
a metric on A. Metric generalises the concept of distance between two points.
A natural question that should arise is the following: how is the metric and a
norm related? In a sense, the distance between two vectors can also be decided
using the metric function. This is can be observed as follows: the dierence
between two vectors yields a third vector, which connects the tips of the two.
The length of this vector is the distance between the tips of the vectors, which
is what vectors are. Mathematically, d (x, y) = |x y|. But we are getting
ahead of ourselves! For now, here are a few examples of metric spaces.
Example 62 A trivial metric which can be dened on any set is the discrete
metric d(r, j) = 1 for r ,= j and 0 otherwise. d(r, j) _ 0, d(r, j) = 0 == r =
j and d(r, j) = d(j, r) are satised by denition. d(r, j) _ d(r, .) +d(., j) can
be veried exhaustively by considering cases r ,= j, r = j, r ,= . and r = ..
Example 63 On the real line R, we can dene the usual metric d(r, j) =
[r j[ (aha!). This can be generalised for the Euclidean plane R
n
with metric
d(x, y) =
_
(r
1
j
1
)
2
+ (r
2
j
2
)
2
+... + (r
n
j
n
)
2
For : = 1, we can get
_
(r
1
j
1
)
2
= [r j[ , which shouldve been proved by
the student in the exercise in the previous topic. For : = 2, we have the familiar
Pythagorean Theorem in plane. Note that the metric d is also the Pythagorean
Theorem in :-dimension. Dimensionality will be rigorously dened later. For
now, well make do with the normal understanding of the word.
Example 64 Another metric denable on the same set is
d(x, y) =
n
I=1
[r
I
j
I
[
This is called the taxicab or Manhattan metric.
This example illustrates the important fact that from a given set with more
than one element, we can obtain various metric spaces by choosing dierent
metrics, just like the norm space. The reader is invited to prove that both are
certied metrics by satisfying the axioms rst and then by induction.
Example 65 A metric similar to the usual metric can also be dened for the
complex plane C. Try to recall the denition of the modulus [.[ =
_
r
2
+ j
2
for
. = r + ij and then derive a metric which can be generalised for C
n
Example 66 The sequence space |
for 1 _ j < forms a metric space under

the metric
d(r, j) =
_
o
I=1
[
I
I
[
_
1/
Example 67 As a set |
o
we take the set of all bounded sequences of complex
(or real) numbers; that is, every element r of |
o
is a complex (resp. real)
sequence r = (
1
,
2
, ...), briey r = (
I
). If we have r = (
I
) and j = (
I
), we
can have the metric dened by d(r, j) = sup where = c
I
[ c
I
= [
I
I
[.
This can be compactly written as
sup
IN
[
I
I
[
The supremum exists since the set is bounded (why?) and is unique, as asked
in the initial chapter. To prove that this is a metric is easy: D1, D2 and D3
can be easily satised. For D4, we can construct the sets c
I
[ c
I
= [
I
I
[,
,
I
[ ,
I
= [
I
j
I
[ and
I
[
I
= [j
I
I
[ for r = (
I
), j = (
I
) and . = (j
I
)
but before that, we need [
I
I
[ = [
I
j
I
+ j
I
I
[ _ [
I
j
I
[ + [j
I
I
[ \i
which we can denote with c
I
, ,
I
and
I
respectively. Since c
I
_ ,
I
+
I
is valid
\i, we can have supc
I
_ sup,
I
+ sup
I
or d(r, j) _ d(j, .) + d(., r).
Example 68 What about sequences that are unbounded? We can also have a
metric on the space : of all bounded and unbounded sequences dened as
d(x, y) =
o
I=1
1
2
I
[
I
I
[
1 +[
I
I
[
for r = (
I
) and j = (
I
). Proving D1 to D3 is easy. For D4, let . = (j
I
) such
that
[
I
I
[ _ [
I
j
I
[ +[j
I
I
[
==
1
1 +[
I
j
I
[
+
1
1 +[j
I
I
[ .
_
1
2 +[
I
I
[
_
1
1 +[
I
I
[
Note that
d(r, j) =
o
I=1
1
2
I
[
I
I
[
1 +[
I
I
[
=
o
I=1
1
2
I
1 + 1 +[
I
I
[
1 +[
I
I
[
=
o
I=1
1
2
I
_
1
1
1 +[
I
I
[
_
=
o
I=1
1
2
I

o
I=1
1
2
I
1
1 +[
I
I
[
= 1
o
I=1
1
2
I
1
1 +[
I
I
[
Since we have
1
2
i
1
1+]
i
q
i
]
+
1
2
i
1
1+]q
i
i
].
_
1
2
i
1
1+]
i
i
]
, we can equivalently have
1
o
I=1
1
2
I
1
1 +[
I
j
I
[
+ 1
o
I=1
1
2
I
1
1 +[j
I
I
[ .
_ 1
o
I=1
1
2
I
1
1 +[
I
I
[
which is what is required.
Example 69 For C[a, /], we have a bona de metric space under the metric
d(r, j) = max
|[o,b]
[r(t) j (t)[, which the reader is required to prove.
In real analysis, you might have proved that every bounded function has a
supremum using the fact that every function which is bounded above will have
a least upper bound. You should have also noticed, while trying to prove the
above example, that we have a maximum over here, instead of the supremum.
This is because not every function has a supremum but it will have a maximum.
This leads us to
Example 70 the space 1[a, /] of real, bounded functions. By denition, each
element r(t) 1[a, /] is a function dened and bounded on a given set [a, /] and
the metric is dened by
d(r, j) = sup
|[o,b]
[r(t) j (t)[
Now,
d(r, j) = 0
== sup
|[o,b]
[r(t) j (t)[ = 0
== [r(t) j (t)[ = 0\t [a, /]
since if the supremum of non-negative numbers zero, then all the numbers are
themselves zero. Then, we have r(t) = j (t) \t [a, /]. Hence, r = j. Second,
the supremum of non-negative numbers is non-negative, which the reader is
required to rigorously prove. For D3,
d(r, j) = sup
|[o,b]
[r(t) j (t)[
= sup
|[o,b]
[(r(t) j (t))[
= sup
|[o,b]
[r(t) + j (t)[
= sup
|[o,b]
[j (t) r(t)[
= d (j, r)
Finally,
sup
|[o,b]
[r(t) j (t)[ _ sup
|[o,b]
[r(t) . (t)[ + sup
|[o,b]
[. (t) j (t)[
This can be made rigorous by resorting to the denition of order and applying
the triangle inequality.
Exercise 71 Show that another metric can be obtained on : by replacing 1/2
I
with j
I
0 such that j
I
<
Hlder Inequality
We will prove Hlders inequality on the real numbers. The proof method can
then be extended to norm spaces. The inequality is valid for j 1 and a such
that
1
+
1
j
= 1. In a sense, j and are inverses or conjugates of each other.
These are called conjugate exponents. For instance, j = = 2. 1 and are
also regarded as conjugate exponents.
Theorem 72 For convergent (why?) real sequences (r
I
) and (j
I
)
o
I=1
[r
I
j
I
[ _
_
o
I=1
[r
I
[
_
1/
_
o
I=1
[j
I
[
j
_
1/j
Proof. From
1
+
1
j
= 1, we have
+j
j
= 1
j j 1 = 1
j( 1) ( 1) = 0 or (j 1) ( 1) = 1
Thus,
1
1
= 1 so that from a function n = t
1
, we can have n
1/(1)
= t
or t = n
j1
. Now let a, / R for a, / 0. Then, we can think of a/ as an area
of a rectangle with sides a and /.
Let )(t) = t
1
. Then,
a/
_
o
_
0
) (t) dt +
b
_
0
)
1
(t) dt
=
o
_
0
t
1
dt +
b
_
0
t
j1
dt
=
a
j
+
/
j
Assume that we have the sequence (

I
) and (j
I
) such that
o
I=1
[
I
[
=
o
I=1
[j
I
[
j
= 1
Set a = [
I
[ and / = [j
I
[. Then, we have the inequality [
I
[ [j
I
[ _
]
i
]
p
+
]q
i
]
q
j
.
Summing such i objects, we get the inequality
o
I=1
[
I
[ [j
I
[ _
o
I=1
[
I
[
j
+
o
I=1
[j
I
[
j
from which we have

o
I=1
[
I
[ [j
I
[ =
o
I=1
[
I
j
I
[ _
1
j
+
1
= 1 (3.1)
Now, let
I
=
ri
(]ri]
p
)
1=p
and j
I
=
i
(]i]
q
)
1=q
. Then, we have
I
=
r
p
i
(]ri]
p
)
and j
j
I
=

q
i
(]i]
q
)
which implies [
I
[
=
]ri]
p
(]ri]
p
)
and [j
I
[
j
=
]i]
q
(]i]
q
)
which, on summing the i index, will yield
o
I=1
[
I
[
=
[r
I
[
([r
I
[
)
and
o
I=1
[j
I
[
j
=
[j
I
[
j
([j
I
[
j
)
both of which equal 1. Hence,
I
=
ri
(]ri]
p
)
1=p
and j
I
=
i
(]i]
q
)
1=q
is a valid
substitution. Multiplying these two and placing in the equality (3.1), we get
o
I=1
r
I
([r
I
[
)
1/
j
I
([j
I
[
j
)
1/j
_ 1
o
I=1
[r
I
j
I
[ _
_
o
I=1
[r
I
[
_
1/
_
o
I=1
[j
I
[
j
_
1/j
This proof is valid only for |
spaces and not norm spaces in general.

Exercise 73 Show that the geometric mean of two positive numbers does not
exceed their arithmetic mean.
Continuing with the real sequences, we have the special name for j = 2 in
the Hlder inequality. That is, the Cauchy-Schwarz inequality is
o
I=1
[r
I
j
I
[ _
_
o
I=1
[r
I
[
2
_
1/2
_
o
I=1
[j
I
[
2
_
1/2
We will now prove the Minkowski Inequality for the same conditions as the
Hlder inequality. Specically, for j _ 1
Theorem 74
_
o
I=1
[r
I
+ j
I
[
_
1/
_
_
o
I=1
[r
I
[
_
1/
+
_
o
I=1
[j
I
[
j
_
1/j
Proof. For j = 1, we can directly apply the triangle inequality. For j 1
[r
I
+ j
I
[
= [r
I
+ j
I
[ [r
I
+ j
I
[
1
_ ([r
I
[ +[j
I
[) [r
I
+ j
I
[
1
Summing over i,
o
I=1
[r
I
+j
I
[
_
o
I=1
[r
I
[ [r
I
+ j
I
[
1
+
o
I=1
[j
I
[ [r
I
+j
I
[
1
From the Hlder inequality, we have
o
I=1
[r
I
[ [r
I
+j
I
[
1
_
_
o
I=1
[r
I
[
_
1/
_
o
I=1
[r
I
+ j
I
[
(1)j
_
1/j
=
_
o
I=1
[r
I
[
_
1
_
o
I=1
[r
I
+ j
I
[
_
1/j
and
o
I=1
[j
I
[ [r
I
+j
I
[
1
_
_
o
I=1
[j
I
[
_
1/
_
o
I=1
[r
I
+j
I
[
_
1/j
because j = j +
Hence we have
o
I=1
[r
I
+ j
I
[
_
_
_
_
o
I=1
[r
I
[
_
1/
+
_
o
I=1
[j
I
[
_
1/
_
_
_
o
I=1
[r
I
+ j
I
[
_
1/j
==
_
o
I=1
[r
I
+ j
I
[
_
11/j
_
_
o
I=1
[r
I
[
_
1/
+
_
o
I=1
[j
I
[
_
1/
==
_
o
I=1
[r
I
+ j
I
[
_
1/
_
_
o
I=1
[r
I
[
_
1/
+
_
o
I=1
[j
I
[
_
1/
Now that you have these inequalities at your disposal, prove that |
is a
metric space
Balls and Spheres
To make use of sequences in a space, we need machinery which can tell us
whether or not the sequence converges or not. Other than that, the space itself
can give us good hints about the behaviour of sequences and series. This is
possible when we know what open and closed sets are. In a metric space (A, d),
this is
Denition 75 Given a point r
0
A and a real number r 0, we dene three
types of sets:
1. 1(r
0
; r) = r A [ d(r, r
0
) < r (Open ball)
2.

1(r
0
; r) = r A [ d(r, r
0
) _ r (Closed ball)
3. o(r
0
; r) = r A [ d(r, r
0
) = r (Sphere)
Intuitively, it is clear that in all three cases, r
0
is the centre and r the radius.
Mathematically, the open ball of radius r is the set of all points in A whose
distance from the centre of the ball is than r. Try to prove that
o(r
0
; r) =

1(r
0
; r) 1(r
0
; r). Remember, these are two sets and you will
have to employ set-theoretic arguments.
In Complex Analysis or in Real Analysis, an open set is one in which every
element has an open neighbourhood contained in that set, using the c c de-
nition. In complete analogy (generalisation, rather), we have
Denition 76 ' A is said to be open if it contains an open ball about each
of its points. 1 A is said to be closed 1
{
= A 1 is open
According to Real or Complex Analysis, the open set should contain c-
neighbourhood. We can have this when we replace r with c in an open ball.
Thus, 1(r
0
; c) is an c-neighbourhood of r
0
where c 0. We will agree to
call a simple neighbourhood of r
0
as a subset of A which contains an c-
neighbourhood of r
0
. This c can be arbitrary (but positive!). The letter will
be reserved for such neighbourhoods. This should not be a cause of confusion
for normed spaces since the denotation should be clear from context.
Trivially, every neighbourhood of r
0
contains r
0
. This will be called an
interior point. We can collect all interior points of an open set ' and give it
a special name the interior of '. This will be denoted by '
o
. You might
nd that Int(') is reserved for such a set in some literature. This set has some
interesting properties, as we shall see.
Clearly, if any neighbourhood of r
0
is contained in an open set ', then
' is also a neighborhood of r
0
(proof?). An alternative denition of open sets
is as follows, with a trivial proof:
Exercise 77 A set ' is open in a metric space (A, d) if and only if every point
is an interior point.
Theorem 78 '
o
is the largest open set contained in '
Proof. Clearly, '
o
_ ' by denition. To prove that this is the largest such
set, assume that there exists another open set O such that '
o
_ O _ '. Then,
let r = O '
o
== r is not an interior point of A. In particular, it is
not an interior point of O. Since r is arbitrary, therefore O is not an open set,
establishing the required theorem.
It is not dicult to show that the collection t of all open subsets of A has
the following properties:
(t
l
) ? t, A t.
(t
2
) The union of any members of t is a member of t.
(t
3
) The intersection of nitely many members of t is a member of t.
Proof. Vacuously, every point of ? is an interior point. Hence, it is open. Sec-
ond, a ball of any radius, when and if constructed around any point of A will nat-
urally be contained in A. Hence, A is open, too. This settles t
l
. Let 1
1
(r
0
; r
1
)
and 1
2
(j
0
; r
2
) be two open sets. Assign the value max (r
1
, r
2
, d (r
0
, j
0
)) to r.
Then, let be a collection of r such that d (r, j) < r and for a j 1
1
(r
0
; r
1
) '
1
2
(j
0
; r
2
). This set is clearly open for the interior point j. The argument can be
applied to arbitrary sets. Moving on to t
3
, if 1
1
(r
0
; r
1
) 1
2
(j
0
; r
2
) = ?, then
we are done. Assume 1
1
(r
0
; r
1
)1
2
(j
0
; r
2
) ,= ?. Fix j 1
1
(r
0
; r
1
)1
2
(j
0
; r
2
),
let r = min (r
1
, r
2
) and take \r 1
1
(r
0
; r
1
) 1
2
(j
0
; r
2
) such that d (r, j) < r
and we have our open set with interior point j. This cannot be extended inde-
nitely since min(r
1
, r
2
, ...) may be zero, giving us a singleton as the intersection,
or even the empty set. As an example, consider the interval
_
1
n
,
1
n
_
=
n
.
Then,

n
n
= 0
Open sets also play a role in connection with continuous mappings, where
continuity is a natural generalisation of the continuity known from complex and
real analysis and is dened as follows:
Denition 79 Let (A, d
1
) and (1, d
2
) be metric spaces. A mapping T : A
1 is said to be continuous at a point r
0
A if for every c 0 there exists a
c 0 such that d
2
(T (r) , T (r
0
)) < c whenever d
1
(r, r
0
) < c.
T is said to be continuous if it is continuous at every point of A. Alterna-
tively, this denition could be phrased as follows:
Theorem 80 A mapping T of a metric space A into a metric space 1 is con-
tinuous if and only if the inverse image of any open subset of 1 is an open
subset of A.
Proof. Let 1 1 be an open set and let T
1
(1) = r [ T(r) 1. We need
to prove that T
1
(1) is open. Let T(r
0
) 1. Since T (r
0
) is an interior point,
we have d
2
(T (r) , T (r
0
)) < c \c 0. Since T is continuous, this ensures the
existence of a c such that d
1
(r, r
0
) < c. Hence for any c or for any open set, we
can nd a c or an open set (r
0
; c) = r [ d
1
(r, r
0
) < c. Hence the inverse
image of every open set is open.
The converse of the proof is trivial. We start with 1(T (r
0
) ; c), an open set,
such that T
1
(1) = is open by suggesting that this set satises d (r, r
0
) < c
for every r , guaranteeing the existence of the required c.
In analogy to interior points, we dene the boundary point of a closed set:
a boundary point or a limit point (synonymous with point of accumulation)
r
0
A of a set A is such that \c 0, an open ball 1(r
0
; c) contains
points of other than r
0
. Notice that r
0
need not be a member of . The set
of all limit points of a set is denoted by
J
. The closure of a set , written as
C|() or even

is '
J
. If a point is not a limit point, then it is an isolated
point.
Proposition 81 If a set contains isolated points only, then that set is nite
Proof. Let (A, d) be a metric space. Assume that we have a set ' A
which is innite. Since we have innite elements, we can collect elements r such
that d (r, r
0
) < c for all c. Such a collection implies that r
0
is a limit point,
contradicting our hypothesis. Therefore, ' is nite.
The crucial point of the proof relies on the fact that this collection is valid
for all c
Corollary 82 Every neighbourhood of a limit point contains innitely many
points.
Exercise 83 A set is closed if and only if it contains all its boundary points.
Denition 84 Let (A, d) be a metric space and A. The diameter of a
set is c () = dia:() = supd (r, j) [ r, j
Example 85 For the usual (Euclidean) metric space, dia:(Z) = dia:(P) =
dia:(Q) = dia:(R) = whereas for =
_
j [ j = r
2
, 0 _ r _ 5
_
= [0, 25]
the diameter for this set is 25.
That is, the greatest distance between any two points. If the set were circular
in shape, then this denition would make sense. A set is called bounded if its
diameter is nite.
Proposition 86 In the usual metric space (R, d), a subset is bounded if and
only if it is bounded above and bounded below
Proof. dia:() = supd (r, j) [ r, j = c < == d (r, j) _ c for all
r, j == [r j[ _ c for all r, j == c _ rj _ c == / _ r _ /
for all r.
Denition 87 A subset ' of a metric space A is said to be dense in A if
' = A or A is said to be separable if it has a countable subset which is dense

in A.
That is, we can have limit points of within it and the result can equal to
the parent set. One good example is the set of rationals and the real set. We
all know that the set of rationals is not complete. As studied in Real Analysis,
Dedekind cuts or the simple addition of limits to every Cauchy sequence can
construct the set of reals. That is,

Q = R or that the set of rationals are dense
in the set of reals. The complex plane, too, can be separated from the irrational
real and imaginary parts against the rational ones. This has importance in the
theory of operators, which will be glimpsed upon.
More technically, if ' is dense in A, then \r
0
A and \c 0, 1(r
0
; c) will
contain points of '; or, in other words, in this case there is no point r
0
A
which has a neighbourhood that does not contain points of '. This is the direct
consequence of the denition of a limit point.
To hit the point home, we will prove that the space |
is separable for
1 _ j < . We will proceed as follows: we will rst construct a countable
subset then basing our argument on the fact that Q is dense in R, construct
limits for every sequence of elements of |
which will be limit points of sequences

in the constructed subset.
Proof. Let ' be the set of all sequences r of the form r = (
1
,
2
, . . . ,
n
, 0, 0, . . .)
where : is any integer. Now, we can assume that
I
Q \i since were only
discussing real or complex sequences. Since Q is countable, Q
n
is therefore
countable, leading us to a countable '. That justies one part of the deni-
tion. To prove that

' = |
. Take j = (j
I
) |
. Since this is convergent, we

have
o
|=n+1
[j
|
[
<
c
2
Since

Q = R, for each j
|
there is a rational
|
close to it. Hence, we can nd
an r ' such that
n
|=1
[
|
j
|
[
<
c
2
Since
d (r, j) =
_
o
|=1
[
|
j
|
[
_
1/
We therefore have
[d (r, j)]
=
o
|=n+1
[j
|
0[
+
n
|=1
[
|
j
|
[
<
c
2
+
c
2
= c
or that d (r, j) < c. That is, every sequence r will have a limit point j. Hence,
' = |
This should in no way mean that every collection of convergent sequences

forms a separable set.
Proposition 88 The space |
o
is not separable.
As a reminder, it is mentioned that the proof that |
o
forms a metric space
was left to the reader.
Proof. Let r = (
1
,
2
, . . .) be a sequence of zeros and ones. Then r converges
and hence r |
o
. With r we associate j =
_
i
2
i
_
. Clearly, j [0, 1]. Since
d (r, j) = sup
I

I
2
I
= sup
I
I
2
I
=
1
2
I
If these sequences are the centre of an open ball of diameter
1
2
i+1
then these
balls do not intersect and we have uncountably many of them, since [0, 1] is
uncountable (this is because the set of reals is uncountable and any interval
is isomorphic to the real line). If ' |
o
such that

' = |
o
, then each of
these nonintersecting balls must contain an element of ' implying that ' is
uncountable, contradicting the hypothesis that ' is countable.
Proposition 89 A discrete metric space (A, d) is separable if and only if A is
countable.
Proof. Let A = r
1
, r
2
, ..., r
n
be a countable set and let ' = r
I
, ..., r
be
a subset for 1 _ i, , _ :. Then, for d (r
I
, r
) < c, we have r
I
= r
if c is close to
zero. Hence, any open ball 1(r
|
; c) for 1 _ / _ : will contain only the element
r
|
and no point is a limit point. Hence, no proper subset of A can have a limit
point. Therefore, any ' will not be dense in A. Since there are no limit points
in A, A
J
= ?. We therefore have

A = A ' A
J
= A. Hence, A is dense in A
and, therefore, separable.
Conversely, assume that A is separable, that is, ' A such that

' = A
but '
J
is empty for the same reason as above. Therefore,

' = ' ' '
J
= A
implies ' = A is the only possible subset. Since ' (or A) does not have any
limit points, every point is an isolated point. Hence, the set is countable.
Simply put, no proper subset of separable A can have limit points if the
metric is discrete. It is surprising that we can have limit points in a specic
metric space whenever we can count the elements of that space, and conversely.
3.3.1 Sequences
Denition 90 A sequence (r
n
) in a metric space (A, d) is said to converge
or to be convergent if there is an r A such that lim
no
d(r
n
, r) = 0. Alter-
natively, a sequence is called convergent if \c 0, such that d(r
n
, r) < c
whenever : .
In such a case, r is called the limit of (r
n
) and r
n
is said to converge to r.
This is denoted by lim
no
r
n
= r or r
n
r as : . If we cannot nd an
for any given , or that if the sequence fails to be convergent, we say that
this sequence diverges. A sequence r
n
is bounded if its range r
n
is bounded.
If a sequence converges, its limit is unique
Proof. Let lim
no
r
n
= |
1
and lim
no
r
n
= |
2
be two limits. Then, \c 0, we can
nd
1
and
2
such that d(r
n
, |
1
) < c,2 and d(r
n
, |
2
) < c,2 for :
1
,
2
.
Let = max (
1
,
2
). Then, d (|
1
, |
2
) < d (r
n
, |
1
) + d (r
n
, |
2
) < c,2 + c,2 = c
whenever : , \c 0. The condition d (|
1
, |
2
) < c implies |
1
= |
2
Of particular interest is the convergence of sequence of functions )
n
. How-
ever, in this case, we also have to consider the domain of the functions, as well.
In the ordinary notion of continuity, this convergence will depend on each point
of the domain, giving the name "point-wise convergence". Apart from this no-
tion of continuity, we also have the notion of uniform continuity, in which the
elements of the domain do not matter. Thus, in uniform continuity, we have
Denition 91 A sequence of functions )
n
(r) converges uniformly if \c 0
\r, such that d()
n
(r) , ) (r)) < c whenever : .
This type of convergence is important when dealing with spaces involving
continuous functions. For now, well use it in this chapter to establish the
completeness of the vector space of continuous functions, which occurs rather
at the closing of the chapter but for that, a theorem.
Theorem 92 If a series of functions converges uniformly, then the limit is
continuous
Proof. Let )
n
(r) ) (r) uniformly. Then, we have an such that \c 0 \r,
d()
n
(r) , ) (r)) <
t
3
whenever : . We also have continuous )
n
(r) so that
\c 0, c such that d()
n
(r) , )
n
(j)) <
t
3
whenever d (r, j) < c. The uniform
continuity is valid for all r T()) and, in particular, whenever d (r, j) < c.
Hence, whenever d (r, j) < c, we have
d() (r) , ) (j)) _ d ()
n
(r) , ) (r)) +d ()
n
(r) , )
n
(j)) + d ()
n
(j) , ) (j))
<
c
3
+
c
3
+
c
3
= c
so that ) (r) is continuous.
Other than a sequence, we can also take out some members of the sequence
to form a subsequence. That "order" of the sequence has to be maintained so
that in a subsequence formed by r
n
and, say, r
n+5
, we can have r
n
= a
1
and
r
n+5
= a
2
. For r
I
, a certied subsequence is r
n1
, r
n2
, ... where :
1
is some i, :
2
is another natural number greater than i and so on.
Exercise 93 If a sequence converges to a point, then a subsequence will con-
verge to that point.
Exercise 94 r
n
r == r
n
k
r for all subsequences r
n
k
Proposition 95 If a sequence converges, then it is bounded.
Proof. Let r
n
r. Then, we can be assured that we will denitely have a
(very large) natural number such that d (r
n
, r) < c \: and \c 0. Let
: = max (d (r
1
, r) , d (r
2
, r) , ..., d (r
, r) , c). Then, d (r
n
, r) < : \: which
really means that every element of the sequence is bounded.
The converse, however, is usually false. Consider the series
(1)
n
2
. This series
is bounded by 1 yet does not converge as it keeps on alternating between 1,2
and 1,2.
Now, try to prove that the sum of two bounded series and a scalar multiple
of a bounded sequence is bounded, to complete the proof that a collection of
such sequences forms a vector space.
Corollary 96 If a sequence is unbounded, then it is divergent.
Proposition 97 If r is a limit point of a subset of a metric space (A, d),
then there exists a sequence such that r
n
r.
Proof. What we need to do is construct a sequence that converges to this limit
point. Since r is a limit point, then we can rest assured that we have an open
ball centred at r of c radius contained in A and containing points other than r,
by denition. Hence, we can collect such points and call them r
n
. Therefore,
d (r
n
, r) < c. What we need now is to prove that we have an such that this is
valid for : . Since epsilon was arbitrary, we can let it depend on the index
:. So, c = 1,:, say. From a collection of the natural numbers, we will always
have an such that c < 1. This can be seeing by applying the Archimedean
property of real numbers. Now, we have 1 c or 1 ,: or : , which
establishes the proof.
Proposition 98 Every real sequence has a monotone subsequence
Proof. A r
n
sequence is monotonic if r
|
_ r
|+1
or if r
|
_ r
|+1
for all /.
Hence we can collect such points and form the required subseuqence. Details
are left to the reader.
Theorem 99 Let ' be a nonempty subset of a metric space (A, d). Then
1. r

' == r
n
' such that r
n
r
2. ' is closed == r
n
' such that r
n
r implies that r '.
Proof. For bullet 1, weve proven that any limit point will have a sequence
convergent to it. The converse is a trivial result of the denition of convergence
and limit points. Bullet two follows by observing that if ' is closed, then
' =

'
Theorem 100 Every real, bounded sequence has at least one convergent subse-
quence
Proof. If we call the bound r, then we can let a monotonic sequence converge
to that point, establishing the theorem. Details are left to the reader.
This is the famous Bolzano-Weistrass theorem and will be made use of ex-
tensively.
Proposition 101 If r
n
r and j
n
j in A, then d (r
n
, j
n
) d (r, j).
Proof. \c 0, we can nd
1
and
2
such that d(r
n
, r) < c,2 and d(j
n
, j) <
c,2 for :
1
,
2
. Let = max (
1
,
2
). Then,
d (r
n
, j
n
) _ d(r
n
, r) + d(r, j) + d(j
n
, j)
== d (r
n
, j
n
) d(r, j) _ d(r
n
, r) + d(j
n
, j)
Also,
d(r, j) _ d (r, r
n
) + d (r
n
, j
n
) + d (j
n
, j)
== [d (r, r
n
) + d (j
n
, j)] _ d (r
n
, j
n
) d(r, j)
Using the two inequalities, we have
[d (r
n
, j
n
) d(r, j)[ _ d(r
n
, r) + d(j
n
, j) < c,2 + c,2 = c
i.e. [d (r
n
, j
n
) d(r, j)[ < c \: .
Notice that in this proof, weve treated d (r
n
, j
n
) as a sequence with the
index :.
Denition 102 A sequence (r
n
) in a metric space (A, d) is said to Cauchy if
\c 0, such that d(r
n
, r
n
) < c whenever :, : .
This can understood to mean that after a certain number of elements, the
elements of the sequence become arbitrarily close together. However, this in
no way means that every Cauchy sequence converges. Find a counter example.
First, try to prove that every convergent sequence is Cauchy i.e. if a sequence
approaches to a point, then after a certain number of elements of the sequence,
the elements of the sequence themselves will come arbitrarily close together.
Proposition 103 Every Cauchy sequence is bounded
Proof. Let r
n
be Cauchy. Then, we can be assured that we will denitely
have a (very large) natural number such that d (r
n
, r
n
) < c \:, : and
\c 0. Let
max
1I,
(d (r
I
, r
) , c) = c
Then, d (r
n
, r
n
) < c \:, :.
Some spaces in which every Cauchy sequence does converge have a special
place in functional analysis, playing a crucial role in most theorems. Any such
space is said to be complete if every Cauchy sequence in it converges (that
is, has a limit which is an element in that set). If any space has a Cauchy
sequence which does not converge, then that space is incomplete. For instance,
the set of rationals, as mentioned that the completion of the set of rationals
will be assumed via Dedekind cuts, forming the reals. We can prove that every
Cauchy, real sequence converges:
Proof. If we have a Cauchy sequence of real numbers r
n
, we know that it is
bounded and if the Cauchy sequence is bounded, then it will have a convergent
subsequence r
n
k
, as was proved earlier. To make things rigorous, let c 0. We
know that we have a 1 so that d (r
n
k
, |) <
t
2
whenever / 1 and also so
that d (r
n
, r
n
) <
t
2
whenever :, : . Notice that :
|
is a sequence which
increases without bound, so that we dont have :
|
1 but instead have / 1.
Now, we can pick a / 1 so that :
|
. Then, for every :
d (r
n
, |) _ d (r
n
, r
n
k
) + d (r
n
k
, |) =
c
2
+
c
2
= c
This leads us to the complex numbers and the generalised Euclidean space.
Proof. Let x
|
R
n
be a Cauchy sequence in the generalised Euclidean space.
Here, the subscript k indicates the index, instead of :, which is reserved for
dimensionality. Then, we have |x
|
x
| <
t
n
\,, / for N This norm
is the usual norm
|x| =
_
(r
1
)
2
+ (r
2
)
2
+ ... + (r
n
)
2
which, as already mentioned, is a generalised version of the Pythagorean theo-
rem. Hence, from |x
|
x
| <
t
n
, we have
_
_
r
(|)
1
r
()
1
_
2
+
_
r
(|)
2
r
()
2
_
2
+... +
_
r
(|)
n
r
()
n
_
2
<
c
:
==
_
r
(|)
1
r
()
1
_
2
+
_
r
(|)
2
r
()
2
_
2
+... +
_
r
(|)
n
r
()
n
_
2
< c
2
==
_
r
(|)
I
r
()
I
_
2
<
c
2
:
2
==
_
_
r
(|)
I
r
()
I
_
2
<
c
:
==

r
(|)
I
r
()
I
<
c
:
for all the i tuples. Since every i tuple is a real number, were actually talking
about the set of reals, which is complete. That is, every Cauchy sequence con-
verges. Hence,

r
(|)
I
r
()
I
< c will converge to, say, r

I
. We can thus construct
a number x = (r
1
, r
2
, ..., r
n
) R
n
. Now,
|x
|
x|
=
_
_
r
(|)
1
r
1
_
2
+
_
r
(|)
2
r
2
_
2
+ ... +
_
r
(|)
n
r
n
_
2
_
_
_
r
(|)
1
r
1
_
2
+
_
_
r
(|)
2
r
2
_
2
+ ... +
_
_
r
(|)
n
r
n
_
2
=

r
(|)
1
r
1
r
(|)
2
r
2
+ ... +
r
(|)
n
r
n
<
c
:
+
c
:
+ ... +
c
:
= c
That is, |x
|
x| < c. For complex numbers, the proof is established by the
fact that R
2 ~
= C.
What weve done over here is that weve taken an arbitrary Cauchy sequence,
constructed a limit and then proved that this sequence converges to this limit.
The
~
= means that the two sets are isomorphic or similar in structure to each
other. Loosely speaking, their operations, their behaviour, their multiplication
and addition rules behave the same in each individual space.
Theorem 104 A subspace ' of a complete metric space (A, d) is itself complete ==
' =

'.
Proof. If ' is complete, we can have a limit point onto which a sequence
converges belonging to ', establishing the existence of limit points in '. Thus,
' =

'.
Conversely, if ' is closed, then it contains all its limits points. For a Cauchy
sequence r
n
in ', r
n
r A since A is complete. Since r is a limit point
of r
n
, it is a limit point of ', establishing that any Cauchy sequence in '
converges in '.
Theorem 105 If a Cauchy sequence r
n
has a convergent subsequence r
n
k

r, then r
n
r
Proof. Since we have a convergent subsequence, we therefore have d (r, r
n
k
) <
c ,2 for all c, where we are promised the existence of a natural number
1
such
that :
|
. We can also enumerate these r
n
k
s such that d (r
n
k
, r
n
) < c,2
for : because we have a Cauchy sequence. It remains trivial then to see
that d (r, r
n
) _ d (r, r
n
k
) + d (r
n
k
, r
n
) < c,2 + c,2 = c
Theorem 106 Let (A, d
1
) and (1, d
2
) be metric spaces. A mapping T : A
1 is continuous at a point r A == r
n
r implies T (r
n
) Tr
Proof. If T is continuous at r, then for every c 0 there exists a c 0 such
that d
2
(T (j) , T (r)) < c whenever d
1
(j, r) < c. If r
n
r, then we can label
the points j when we have an : so that d
1
(r
n
, r) < c, which is possible
when d
2
(T (r
n
) , T (r)) < c, for every c 0 and : .
The converse of the proof is trivial.
Now that we have some machinery regarding Cauchy sequence, lets see some
other examples of complete metric spaces, which include |
, |
o
, c and C [a, /].
Proof. The completeness of |
will follow a similar pattern; we will take an

arbitrary Cauchy sequence, construct a limit in |
and then show that this

sequence converges to that limit. So we have
d (r
n
r
n
) =
_
(n)
I

(n)
I
_
1/
< c
from which we have
(n)
I

(n)
I
< c
Whether real or complex, each
(n)
I

I
because each ith sequence is a member
of a space of convergent Cauchy sequences. Constructing r = (
I
), we need to
show that this sequence of sequences is bounded under the j-norm for it to
belong to this space. Since were talking about a sequence of sequences, we can
have
_
(n)
I

(n)
I
_
1/
<
t
p
_
2
. Using this, we have
_
(n)
I
_
=
_
|
I=1
(n)
I
+
(n)
I

(n)
I
_
_
_
|
I=1
(n)
I
+
|
I=1
(n)
I

(n)
I
_
<
_
c
p
_
2
_
+
_
c
p
_
2
_
= c
We will then have d (r

n
, r) =
_
(n)
I

I
_
1/
< c, implying that the
Cauchy sequence converges. In particular, we have
d (r
n
, r) =
_
(n)
I

I
_
< c
= c
implying that the sequence
_
(n)
I

I
_
belongs to the space. Since r
n
was a
member and r
n
r is a member, we can use the fact the vector addition is
closed to say that r = r
n
(r
n
r) belongs to the space.
In a similar way, we can prove that |
o
is complete.
Proof. Taking an arbitrary Cauchy sequence r
n
=
_
(n)
I
_
, we have
d (r
n
, r
n
) = sup
I
(n)
I

(n)
I
< c
for all :, : . The denition of supremum implies that

(n)
I

(n)
I
< c for
all i. Again, the ith Cauchy sequence
(n)
I
converges to, say, ith limit
I
. Now,
we construct (
I
). To prove that (
I
) |
o
, observe that
(n)
I

(n)
I
< c ==

(n)
I

I
_ c
. This is valid whenever : . For any sequence j = (j
I
) |
o
, we have
[j
I
[ _ c
. Similarly,
(n)
I
|
o
implies that

(n)
I
_ c
n
where this constant
depends on the :th term of r
n
. Then,
[
I
[
_

(n)
I

I
(n)
I
_ c +c
n
= /
== [
I
[ is bounded
== sup[
I
[ exists
== sup
(n)
I

I
exists
Since

(n)
I

I
_ c, we have sup
(n)
I

I
_ c whenever : and
c 0
An analogous proof holds for :-tuples x = (
1
,
2
, ...,
n
) and y = (j
1
, j
2
, ..., j
n
)
under the metric
d(r, j) = max
I
[
I
j
I
[
The space c is a subspace of |
o
and involves not only bounded sequences
but also sequences which are convergent. Hence, the collection of convergent
sequences also forms a vector space. Try to prove rst that the sum of two
convergent sequences is a convergent sequence and that a scalar multiple of a
convergent sequence is convergent. This space is complete, as well.
Proof. We can pursue this proof by the normal way or take a detour and apply
the fact that subspace of a complete metric space is itself complete if and only if
the space is closed. Hence, if we can prove that c is closed in |
o
, were done. We
take a limit point r of c. This r = (
I
). Then, we clearly have a sequence r
n
in
c which converges to this point. But c is a collection of convergent sequences.
Hence, this r belongs to c. Therefore, c is closed.
The space C[a, /] is also complete under its usual norm
Proof. For any Cauchy sequence (r
n
) we have
d (r
n
, r
n
) = max
|[o,b]
[r
n
(t) r
n
(t)[ < c
which says that we can have a t
0
[a, /] such that [r
n
(t
0
) r
n
(t
0
)[ < c. These
are, then, not continuous functions but points in a complete (complex or real)
space. Hence, we can have a r
n
(t
0
) r(t
0
) but this is a real or complex
number, not a function. What we need is an arbitrary t, which we can still get
by remembering that we had a maximum function. So, we then associate for
each t, r
n
(t) r(t). Notice that the convergence will depend on t. So, in this
case, we have point-wise convergence. To show that the limit r(t) belongs to the
space, we let : in d (r
n
, r) to get d (r
n
, r) _ d (r
n
, r
n
) + d (r
n
, r) < c.
Hence, we have
d (r
n
, r) = max
|[o,b]
[r(t) r
n
(t)[ < c
This convergence is now uniform since the limit is valid for any t [a, /]. Since if
continuous functions converge uniformly, then the limit will also be continuous,
as has been proved earlier. Therefore, r(t) is continuous or that r(t) C[a, /].
That is, any Cauchy sequence of continuous functions has a limit in the same
space, establishing completeness.
If we change the norm to the integral norm, the space C[a, /] becomes in-
complete. We will see the case for a = 0 and / = 1.
The above proof indicates why the metric is also referred to as the uniform
metric.
Theorem 107 In space C[a, /], convergence is always uniform
Proof. From r
n
(t) r(t), we have
d (r
n
, r) = max
|[o,b]
[r
n
(t) r(t)[ < c
== [r
n
(t) r(t)[ < c \t
i.e. the epsilon is independent of the points of the domain.
For incomplete spaces, we start o with the usual set of rationals.
Proof. To show that the set of rationals Q is incomplete, we take the Cauchy
sequence r
n
=
_
1 +
1
n
_
n
. The limit of this sequence is the exponential c. We
show that this sequence is indeed Cauchy. Let : = : + for an integer . For
suciently large :, : under the metric d (r, j) = [r j[, we have
d (r
n
, r
n
) =
_
1 +
1
:
_
n
_
1 +
1
: +
_
n+j
_
1 +
1
:
_
n
_
1 +
1
: +
_
n+j
|=0
_
:
/
__
1
:
_
|
n+j
|=0
_
: +
/
__
1
: +
_
|
_
1 + 1 +
n1
2!n
+
(n1)(n2)
3!n
2
+ ... +
1
n
n
_
_
1 + 1 +
n+j1
2!(n+j)
+
(n+j1)(n+j2)
3!(n+j)
2
+... +
1
(n+j)
n+q
_
: 1
2!:
+
(: 1) (: 2)
3!:
2
+... +
1
:
n

: + 1
2! (: + )

(: + 1) (: + 2)
3! (: +)
2
...
1
(: + )
n+j
To this last experssion, we can apply the triangle inequality to each term indi-
vidually. Hence, for a large :, we have smaller and smaller distances between :
and :. In other words, for any c, we can always nd a value for : so that the
Cauchy condition is certied.
From real analysis, it is well known that c = 1 + 1 +
1
2!
+
1
3!
+ .... Hence,
well show that lim
no
r
n
= c =
1
0!
+
1
1!
+
1
2!
+
1
3!
+... is irrational. Well proceed
by contradiction. Assume c =

j
. Then,
!c = ! +
!
1!
+
!
2!
+ ... +
!
!
+1
where 1 =
1
j+1
+
1
(j+1)(j+2)
+ ... is the remainder of the terms. Since !c =
( 1)!j is an integer and ! +
j!
1!
+
j!
2!
+... +
j!
j!
is also the nite sum of integers
(implying that it, itself, is an integer), hence 1 is an integer but
1
+ 1
+
1
( + 1) ( + 2)
+... <
1
+ 1
+
1
( + 1)
2
+ ... =
1
so that 1 <
1
j
means that 1 is not an integer. This is our contradiction so we
can conclude that c ,=

j
In summary, we have lim
no
r
n
, Q where r
n
is Cauchy, implying that Q is
incomplete.
Other than the rationals, the set Z is also incomplete under the metric
d (:, :) = [1,:1,:[
For, from a Cauchy sequence r
n
, we have
d (r
n
, r
n
) =
1
r
n

1
r
n
< c
This sequence does not converge for c =
1
for : .
The space of polynomials 1[a, /] is also not complete.
Proof. The Cauchy sequence
r
n
=
n
|=0
r
/!
converges uniformly to the function c
r
which is not a polynomial. For, if :, :
, we have
[r
n
r
n
[ =
|=0
r
/!

n
|=0
r
/!
max(n,n)
|=min(n,n)
r
(/ + 1)!
which can be made as small as we like. Notice, also that the the epsilon will not
depend on r.
The space C[0, 1] under the norm
|r(t)| =
1
_
0
[r(t)[ dt
is incomplete.
Proof. If we have a sequence
r
n
=
_
0 if 0 _ t _
1
2
1 if
1
n
+
1
2
_ t _ 1
of functions, we get a Cauchy sequence if :, : 1,c . This is because
|r
n
(t) r
n
(t)| =
1
_
0
[r
n
(t) r
n
[ dt
=
1
_
1/2
[r
n
(t) r
n
[ dt
=
1
_
1/2
[1,:1,:[ dt
_ ([1,:[ +[1,:[)
1
_
1/2
dt
< (2c) (1,2) = c
Now, for any r C, from |r
n
(t) r(t)| < c and : , we should have
1
_
0
[r
n
(t) r[ dt =
1
_
0
[r
n
(t) r[ dt
=
1/2
_
0
[r
n
(t) r[ dt +
1/2+1/n
_
1/2
[r
n
(t) r[ dt +
1
_
1/2+1/n
[r
n
(t) r[ dt
=
1/2
_
0
[r(t)[ dt +
1/2+1/n
_
1/2
[r
n
(t) r[ dt +
1
_
1/2+1/n
[r(t) 1[ dt < c
In other words, each integral is less than c. Now, recalling that the choice of c
is arbitrary, we have r(t) = 0 for 0 _ t < 1,2 from
1/2
_
0
[r(t)[ dt < c. t = 1,2 fails
because of the middle integral in the range (1,2, 1,2+1:) where : is very large.
Finally, if the last integral has to be arbitrarily small, we can safely say that
r(t) = 1 in the range (1,2 + 1:, 1]. From this, we can observe that lim
|1/2
r(t)
fails to exist because lim
|1/2
r(t) ,= lim
|1/2
+
r(t), implying that the function is not
continuous, so that we cannot have any limit for the particular Cauchy sequence
r
n
(t)
Other than this machinery, we can also complete a metric space. Intuitively,
this is done by "adding" limit points so that every Cauchy sequence converges.
Of course we cant include elements in the set on our own accord but what we
can do is make the set "equal" to another complete subset of another space.
This "equality" is not the true equality which we are wired to think of and is
based on the denition of a type of isomorphism which follows this paragraph.
This strange "equality" of sets makes the two sets discernable with respect to
their structure and properties but the substance itself diers. For instance, as is
known, the set Q is constructed from Z(Z 0). The removal of 0 ensures
that zero is excluded in the denominator. The resulting set Q is not TRULLY
equal to an extension of the set of integers so that it is unfair to state that
Z Q, strictly speaking. Z happens to a single set whereas Q is the Cartesian
product of both. What can actually be said is that Z1 Q. However, there
exists an isomorphism between Z1 and Z, so that the sets are equal in
their properties and structure but not substance, as you can clearly see. For all
practical purposes, people usually dont beat about the bush and simply state
Z Q, which is safe to say because of the concept of isomorphism. To make
the point relatable to completion, it is not true, strictly speaking, that

Q = R
because Q has holes in it and we have absolutely no authority to add points
to complete the set of rational numbers but what we can do is make the set
Q isomorphic to subset of complete set. Thus, while were not really adding
points, were still making the set complete by accounting for the missing holes,
which is why we have the colloquial "addition of points".
Here is the promised denition:
Denition 108 A mapping T from a metric space (A, d) into
_
^
A,
^
d
_
is said
to be an isometry if T \r, j A,
^
d(T (r) , T (j)) = d(r, j).
If such a bijective mapping T : A
^
A can be found, then A is said to be
isometric with
^
A. This is the isomorphism for two metric spaces. Intuitively,
an isometry preserves distances so nearby points in one space and equivalently
near in another metric space. Remember, in metric spaces, one is concerned
with distance between two points so that if two metric spaces have the same
structure and properties i.e. must be isometric, then the distance between two
points must be conserved and nothing else matters not the names of the points,
at least.
In order to complete any metric space, we can show that it is isomorphic
to a dense subspace of a complete metric space and that this complete metric
space must necessarily exist. Furthermore, this metric space is unique (up to
isomorphism).
Theorem 109 For a metric space (A, d), there exists a complete metric space
_
^
A,
^
d
_
which has a subspace \ that is isometric with
^
A such that

\ =
^
A.
Furthermore, this space is unique except for isometries.
Proof. First, we focus on the construction of
_
^
A,
^
d
_
. Let r
n
and r
n
be Cauchy
sequences in A. We will call two Cauchy sequences equivalent if they have the
same limit i.e.
lim
no
d (r
n
, r
n
) = 0
This will be written as (r
n
) ~ ( r
n
) . We can then gather all such equivalent
sequences and form an equivalent class. Indeed, (r
n
) ~ (r
n
) is trivial, so this
relation is reexive. Also, since the arguments of a metric function are symme-
try, the relation ~ is symmetric. Finally, if (r
n
) ~ (j
n
) and (j
n
) ~ (.
n
), we
have
d (r
n
, .
n
) _ d (r
n
, j
n
) + d (r
n
, j
n
)
Taking limits on both sides and using the fact that the metric function is always
positive, we have
lim
no
d (r
n
, .
n
) = 0
so that (r
n
) ~ (.
n
), implying transitivity. Thus, we can have for ourselves
an equivalence class ^ r = r
n
of Cauchy sequences. We can collect all such
equivalence classes ^ r, ^ j, ... and form the set
^
A. For this set, we can have the
metric function
^
d (^ r, ^ j) = lim
no
d (r
n
, j
n
)
where r
n
^ r and j
n
^ j. Note that this is not equal to zero since r
n
and j
n
are
members of a dierent equivalence class. To show that this limit is well-dened
or that this denition is sensible and not ambiguous with dierent results for
the same choice of inputs, we will rst show that this limit exists and then show
that it is independent of the choice of representatives. First, we have
d (r
n
, j
n
) _ d (r
n
, r
n
) + d (r
n
, j
n
) + d (j
n
, j
n
)
==
d (r
n
, j
n
) d (r
n
, j
n
) _ d (r
n
, r
n
) + d (j
n
, j
n
)
Similarly,
d (r
n
, j
n
) _ d (r
n
, r
n
) + d (r
n
, j
n
) + d (j
n
, j
n
)
==
d (r
n
, j
n
) d (r
n
, j
n
) _ d (r
n
, r
n
) + d (j
n
, j
n
)
==
(d (r
n
, r
n
) + d (j
n
, j
n
)) _ d (r
n
, j
n
) d (r
n
, j
n
)
this is basically / _ a and / _ a so that we have [a[ _ /. Hence,
[d (r
n
, j
n
) d (r
n
, j
n
)[ _ d (r
n
, r
n
) + d (j
n
, j
n
)
Now, since r
n
is Cauchy, we have d (r
n
, r
n
) < c,2 and similarly d (j
n
, j
n
) <
c,2. This in turn implies that for :, :
[d (r
n
, j
n
) d (r
n
, j
n
)[ < c
so that
lim
no
d (r
n
, j
n
) = lim
no
d (r
n
, j
n
)
Hence,
^
d (^ r, ^ j) is just as valid for any Cauchy sequence, making the deni-
tion valid. Now, we prove that
_
^
A,
^
d
_
is a metric space.
^
d (^ r, ^ j) = 0 ==
lim
no
d (r
n
, j
n
) = 0 == lim
no
r
n
= lim
no
j
n
so that (r
n
) ~ (j
n
), making them
members of the same equivalence class. Since members of an equivalence class
are either disjoint or the same, therefore ^ r = ^ j. Next, since d (r
n
, j
n
) _ 0, we
have
^
d (^ r, ^ j) _ 0. Furthermore, d (r
n
, j
n
) = d (j
n
, r
n
) so that
^
d (^ r, ^ j) =
^
d (^ j, ^ r).
Finally,
d (r
n
, .
n
) _ d (r
n
, j
n
) + d (j
n
, .
n
)
so that
^
d obeys the triangle inequality.
We have just proved that for any metric space (A, d), we will have another
metric space (
^
A,
^
d) by accounting for the limits of the Cauchy sequences, made
possible by clumping all Cauchy sequences with common limits. Let \
^
A
and let T : A \ be a mapping such that T(a) = ^ a where ^ a is an equivalence
class of constant Cauchy sequences. Here, \ is a subclass of constant Cauchy
sequences. Since if two sequences are both constant and converge to the same
limit, then the two sequences are equal. Thus, every equivalence class of constant
Cauchy sequences will be a singleton so that
^
/ will only contain the Cauchy
sequence (/, /, ...). We will now prove that this is an isometry.
First, notice that the mapping is clearly onto. This can be understood by
recalling how we arrived at
^
A and hence \. Next, for T(/
1
) = T(/
2
), we have
^
/
1
=
^
/
2
so that the mapping is one-to-one. Hence T is bijective. Finally, T is
an isometry since
^
d (T (a) , T (/)) =
^
d
_
^ a,
^
/
_
= lim
no
d (a
n
, /
n
) = d (a, /)
To show that this \ is dense in
^
A. For that, we need to show that the limit
points of \ are in
^
A. That is, if ^ r
^
A, we should have
^
d(^ r, r) < c \c 0
contained in \ for r \. For ^ r A and (r
n
) ^ r. Now, for any Cauchy
sequence r
n
the inequality
d (r
n
, r
) < c,2
will be valid \c 0 whenever : . For the constant sequence (r
, r
, ...) =
^ r
\, we have
^
d(^ r, r) = lim
no
d (r
n
, r
) _ c,2 < c
so that every neighbourhood of ^ r will contain a point of \.
To show that
^
A is complete, let (^ r
n
) be any Cauchy sequence in
^
A. Now,
since \ is dense in
^
A, every point ^ r
n

^
A and \c 0, we can nd a point
^ .
n
\ so that
^
d(^ r
n
, ^ .
n
) < c. We can choose this c = 1,: so that the sequence
(^ .
n
) becomes Cauchy. This can be observed as follows:
^
d(^ .
n
, ^ .
n
) _
^
d(^ .
n
, ^ r
n
) +
^
d(^ r
n
, ^ r
n
) +
^
d(^ r
n
, ^ .
n
)
< 1,: +
^
d(^ r
n
, ^ r
n
) + 1,:
Since the element of \, ^ .
n
, is Cauchy, (.
n
) = T
1
(^ .
n
) is also Cauchy in A.
If (.
n
) is contained in the class ^ r, then
^
d(^ r
n
, ^ r) _
^
d(^ r
n
, ^ .
n
) +
^
d(^ .
n
, ^ r)
< 1,: +
^
d(^ .
n
, ^ r)
= 1,: + lim
no
d (.
n
, .
n
)
Since the sequence (.
n
) is an element of the equivalence class of Cauchy se-
quences ^ r and ^ .
n
is an equivalence class of Cauchy sequences and is contained
in \, we have (.
n
, .
n
, .
n
, ...) ^ .
n
and thus the inequality can be made as small
as we like, implying that the limit of ^ r
n
is ^ r
If
_
~
A,
~
d
_
is another complete space with a subspace
~
\ which is isometric
with A such that
~
\ is dense in
~
A. Then, for any ~ r, ~ j
~
A , we apply the same
method as above to get
~
d (~ r, ~ j)
~
d (~ r
n
, ~ j
n
)
_
~
d (~ r, ~ r
n
) +
~
d (~ j, ~ j
n
)
so that
~
d (~ r
n
, ~ j
n
)
~
d (~ r, ~ j), implying that
^
A and
~
A are isometric.
Just for a sneak peak:
Denition 110 A series associated with a sequence r
n
is the sum of all the
terms of the sequence.
More compactly, a series can be written using the summation symbol r
n
.
A series converges if this sum is nite.
Example 111 If a rabbit hops innite hopes such that every hop is half the
distance of its previous hop, then the total distance will be represented by a
1
2
n
where a is the distance covered by the rst hop.
Exercise 112 Show that the open set of reals (a, /) is incomplete whereas [a, /]
is complete.
The trick over here is to construct a particular Cauchy sequence which con-
verges to a or /. This does mean that the sequence does not converge in the
particular open subset. For this, r
n
= / 1,: or even r
n
= a +1,: does fairly
well. The fact that the latter interval is closed automatically qualies it for
completeness. For a particular construction, see the related theorem.
3.3.2 Subspaces
A subspace of a vector space is so that the subset inherits the structure. So,
for a vector space \ , if we have a \ and if this satises all the axioms of a
vector space using the addition and scalar multiplication dened for \ , we have
a subspace. Needless to say, the improper subspaces are the trivial subspace 0
or the space \ itself. Any other subset, if it satises the ten axioms using the
induced vector addition and scalar multiplication, is a subspace. However, this
method of verication is too lengthy. We can make use of the fact that vector
addition is a binary operation. As already established, we can have a subgroup
H of a G if and only if \a, / H, a/
1
H. This will take care of the fact
that we have an Abelian group for vector multiplication. But what about scalar
multiplication? For that we have a theorem:
Theorem 113 A subspace of a vector space \ is a nonempty subset of \
== for all u, v and all scalars c, , we have cu ,v .
The proof for groups has been worked out. Try to prove this on your own.
Example 114 If we have a vector space of the usual plane R
2
, a subspace would
be R. Note that dimension does not necessarily have to decrease. For instance,
the reals over the eld of quotients is a vector space but for the subspace of the
set of rationals over the eld of rationals, we have the same dimensions.
But what do we mean by dimension? And is the dimension of c, the space
of all convergent sequences the same, less or greater than the dimensionality of
|
o
, of which it is a subspace?
Denition 115 A linear combination of vectors r
1
, r
2
, ..., r
n
of a vector
space A is an expression of the form c
1
r
1
+ c
2
r
2
+... + c
n
r
n
where c
I
F.
Example 116 For scalars 2, 3 and vectors (1, 2) and (2, 4), we have the linear
combination 2 (1, 2) + 3 (2, 4) = (2, 4) + (6, 12) = (8, 16).
The word "linear" indicates that the vectors are not non-linear i.e. we do not
have any square, square roots, or under-roots. For vectors, this consideration is
meaningless, all the same. Of course this does not mean were discussing trivial
matters and just beating about the bush. Nevertheless, from this we will more
forward to dene dimensions but for now, another denition.
Denition 117 For any nonempty subset of a vector space \ , the set of all
linear combinations of vectors of called the span of , written span().
Example 118 The span of the vector (1, 2, 3) is c(1, 2, 3) i.e. all vectors that
are parallel to it. Note that in this case, anti-parallel vectors are also included
(anti-parallel in the sense that the vectors are parallel with an opposite direction)
Example 119 span(2, 4) , (1, 3) = c(2, 4) + , (1, 3)
For any such , span is a subspace of the vector space \ .
Proof. For any x, y , cx ,y .
Were just a step away from dening dimensions.
Denition 120 A linear combination of vectors r
1
, r
2
, ..., r
n
is linearly in-
dependent in a vector space \ if c
1
r
1
+c
2
r
2
+...+c
n
r
n
= 0 == c
I
= 0 \i.
Such a combination is not linearly independent or is linearly dependent if
c
I
,= 0.
The name derives from the fact that a linear combination of such vectors
can never be produced by any single other vector. For instance, the vectors
(1, 0, 0), (0, 1, 0) and (0, 0, 1) are linearly independent i.e. the usual i, j, k axis
cannot create any one from each other whereas the vectors (1, 2) and (2, 4) are
not. Also, if any third vector were written out, it would depend on these three
linearly independent vectors not more, not less. More formally, an arbitrary
subset of \ is said to be linearly independent if every nonempty nite subset
of is linearly independent. This leads to our long-awaited notion of dimension.
Denition 121 A vector space \ is said to have dimension :, written dim\ =
:, if every vector r \ can be written out as a linear combination of a linearly
independent set of : vectors
By default, this means that any set of : +1 or more vectors of \ is linearly
dependent. By denition, \ = 0 is nite dimensional and dim\ = 0. If : is
not nite, then \ is said to be innite dimensional. Again, the i, j, k axis set
the intuition for the dimension and clearly satisfy this denition, as highlighted
above. Such linearly independent vectors are called basis.
Theorem 122 Let \ be an :-dimensional vector space. Then, for any proper
subspace , dim < dim\ .
3.3.3 Metric Spaces and Norm Spaces
As already explained why, a norm space can be converted into a metric space
by dening d (x, y) = |x y|. This is called the metric induced by the norm.
The norm of any single vector, then, can be viewed as its distance from the zero
vector. In R
n
, this is the origin. However, there is still a distinction between the
two in terms of structure. For instance, objects of a norm space are necessarily
vectors whereas in a metric space, one can have ordinary points. In fact, one does
not need any other relation between the points in a metric space. This should
be easy to understand because the discrete metric can convert any ordinary set
into a metric space, even the set |i:a, ,:a|, Jn:aid, Haroo:. Metric
spaces are "coarse" whereas normed spaces have a richer structure. In short,
every norm space is a metric space but not conversely.
Since we now have a way of "equating" any norm with a metric, convergence
of sequences and related concepts in normed spaces follow readily from the cor-
responding denitions. For instance, a sequence x
n
in a normed space (, |.|)
is convergent if == x such that lim
no
|x
n
x| = 0 == \c 0 , such
that |x
n
x| < c whenever : .
Example 123 To show that the sequence a
n
= 1,:
2
converges to 0 under any
norm, we can let c 1, so that : implies
_
_
_
_
1
:
2
0
_
_
_
_
=
_
_
_
_
1
:
2
_
_
_
_
<
_
_
1,
2
_
_
= c
Thus, if we have any given epsilon, we can nd an so that the criterion of
convergence is satised.
Try to prove that the norm on C[a, /] is uniform.
Lemma 124 A metric d induced by the norm on a normed space (, |.|) is
invariant to translation and scales appropriately. That is,
1. d (x +a, y +a) = d (x, y)
2. d (cx, cy) = [c[ d (x, y)
Proof. d (x +a, y +a) = |(x +a) (y +a)| = |x +a y a| = |x y| =
d (x, y)
d (cx, cy) = |cx cy| = |c(x y)| = [c[ |x y| = [c[ d (x, y)
3.4 Complete Norm Spaces
We have just proved instances of norm spaces which are complete and which
can be completed. Moving on to a more general notion, we have
Denition 125 A Banach Space is a complete normed space (complete in
the metric dened by the norm).
Needless to say, a sequence x
n
in a normed space (, |.|) is Cauchy if
== lim
no
|x
n
x
n
| = 0 == \c 0 , such that |x
n
x
n
| < c
whenever :, : . A subspace of a Banach space is complete ==
= .
Theorem 126 Let (, |.|) be a normed space. Then there is a Banach space
1 and an isometry T from onto a subspace of 1 such that

= 1.The
space 1 is unique, except for isometries.
Lemma 127 Let r
1
, r
2
, ..., r
n
be a set of linearly independent vectors in a
normed space . Then, there is a number c 0 such that for any choice of
scalars c
1
, c
2
, ..., c
n
|c
1
r
1
+c
2
r
2
+ .... + c
n
r
n
| _ c ([c
1
[ +[c
2
[ + ... +[c
n
[)
3.4. COMPLETE NORM SPACES 57
Proof. The lemma holds trivially if all the scalars are zero. Let [c
1
[ + [c
2
[ +
... +[c
n
[ 0. Then |c
1
r
1
+ c
2
r
2
+ .... + c
n
r
n
| _ c ([c
1
[ +[c
2
[ + ... +[c
n
[)
=
|c
1
r
1
+ c
2
r
2
+.... +c
n
r
n
|
([c
1
[ +[c
2
[ + ... +[c
n
[)
_ c
=
|c
1
r
1
+c
2
r
2
+.... +c
n
r
n
|
|[c
1
[ +[c
2
[ + ... +[c
n
[|
_ c
=
_
_
_
_
c
1
[c
1
[ +[c
2
[ + ... +[c
n
[
r
1
+
c
2
[c
1
[ +[c
2
[ +... +[c
n
[
r
2
+.... +
c
n
[c
1
[ +[c
2
[ +... +[c
n
[
r
n
_
_
_
_
_ c
= |,
1
r
1
+ ,
2
r
2
+ .... + ,
n
r
n
| _ c (say)
It should be remarked that
n
=1
= 1
Suppose |,
1
r
1
+ ,
2
r
2
+ .... + ,
n
r
n
| < c with
n
=1
= 1. Then, a
sequence j
n
= ,
(n)
1
r
1
+ ,
(n)
2
r
2
+ .... + ,
(n)
n
r
n
such that j
n
0 with
n
=1
,
(n)
= 1 from which we have

,
(n)
_ 1
Since ,
(n)
is a bounded sequence for every ,, then there must exist a cor-

responding convergent subsequence for each ,, according to the Bolzano Weier-
strass theorem. Let j
1,n
denote the corresponding subsequence for j
n
. Since
j
n
is bounded, j
1,n
is also bounded and has a convergent subsequence j
2,n
(say). Continuing this way, we can obtain : such subsequnces. ,
is a limit for
each , subsequence and, consequently, j
n,n
j = ,
1
r
1
+ ,
2
r
2
+ .... + ,
n
r
n
as : increases without bound. It is clear that ,
,= 0 for each , otherwise j = 0

but j
n
0 ==j
nn
0 ==j = 0 which is a contradiction.
Theorem 128 Every nite dimensional subspace of a normed space is
complete. In particular, every nite dimensional normed space is complete.
Theorem 129 Every nite dimensional subspace of a normal space is
closed in .
Denition 130 The norm |.|
1
on any normed space is said to be equivalent
to another norm |.|
2
if a, / R
+
such that \x
a |x|
1
_ |x|
2
_ / |x|
1
This concept is motivated by the following fact: equivalent norms on
dene the same topology for .
Theorem 131 Any two norms on a nite dimensional space are equivalent
Proof. Let A be an :-dimensional vector space with norms |.|
1
and |.|
2
\x A, x = c
1
c
1
+c
2
c
2
+.... +c
n
c
n
where c
1
, c
2
, ..., c
n
are the basis for A
and c
1
, c
2
, ..., c
n
are scalars
Then,
|x|
1
= |c
1
c
1
+c
2
c
2
+.... + c
n
c
n
|
1
_ |c
1
c
1
|
1
+|c
2
c
2
|
1
+.... +|c
n
c
n
|
1
= [c
1
[ |c
1
|
1
+[c
2
[ |c
2
|
1
+ .... +[c
n
[ |c
n
|
1
Let / = max
1In
|c
I
|
1
Then,
[c
1
[ |c
1
|
1
+[c
2
[ |c
2
|
1
+.... +[c
n
[ |c
n
|
1
_ [c
1
[ / +[c
2
[ / + ... +[c
n
[ /
= ([c
1
[ +[c
2
[ +... +[c
n
[) /
Also,
|x|
2
c
=
|c
1
c
1
+ c
2
c
2
+ .... + c
n
c
n
|
2
c
_ ([c
1
[ +[c
2
[ + ... +[c
n
[)
Combining the two inequalities, we get
|x|
1
_ ([c
1
[ +[c
2
[ +... +[c
n
[) / _
/ |x|
2
c
or
c
/
|x|
1
_ |x|
2
which is the rst half of the inequality for a = c,/. The second inequality
|x|
2
_ / |x|
1
can be obtained in a similar fashion.
Denition 132 A metric space (A, d) is said to be compact if every sequence
in A has a convergent subsequence.
A general property of compact sets is expressed in:
Lemma 133 A compact subset of a metric space (A, d) is closed and bounded.
The converse of this lemma is in general false. However, for a nite dimen-
sional normed space we have:
Theorem 134 In a nite dimensional normed space (, |.|), any subset _
is compact if and only if is closed and bounded.
Theorem 135 Let =

1 be subspaces of a normed space . Then for
every real number 0 (0, 1) there is a / 1 such that |/| = 1, |/ a| _ 0
\a .
This is Rieszs Lemma.
Theorem 136 If a normed space has the property that the closed unit ball
= r [ |r| _ 1 is compact, then is nite dimensional.
Theorem 137 Let (A, d
1
) and (1, d
2
) be metric spaces and T : A 1 a
continuous mapping. Then, the image of a compact subset of A under T is
compact.
Corollary 138 A continuous mapping T of a compact subset of a metric
space A, d
1
into R assumes a maximum and a minimum at some points of .
Denition 139 Let A and 1 be vector spaces over the same eld. Then, an
operator T : A 1 is linear if for all r, j A and scalars c, T(r + j) =
T (r) + T (j) and T(cr) = cT (r).
Example 140 The identity operator
^
1(r) = r is clearly linear.
Example 141 The zero operator
^
0 (r) = 0 is also linear by default.
Example 142 The dierential operator
d
dr
: 1[a, /] 1[a, /]
is linear. For, let )(r) =
n
I=0
c
I
r
I
by any polynomial. Then,
d
dr
)(r) =
n
I=1
,
I
r
I1
1[a, /]
where ,
I
= ic
I
. Then,
J
Jr
[)(r) + q(r)] =
J
Jr
)(r) +
J
Jr
q(r) and
J
Jr
c)(r) =
c
J
Jr
)(r)
Example 143 Another operator T from C[a, /] into itself can be dened by
T(r(t)) =
_
r(t)dt
is linear.
Example 144 An operator T from C[a, /] into itself dened by
T(r(t)) = tr(t)
This operator is linear. For T (cr(t)) = t (cr) (t) = c(tr(t)) = cT (r(t)).
Furthermore, T (r(t) + j (t)) = t (r(t) + j (t)) = tr(t) + tj (t) = T (r(t)) +
T (j (t))
Lemma 145 An operator T : A 1 is linear if and only if \r, j A and
\c, , 1, T(cr + ,j) = cT(r) + ,T(j)
Proof. T(cr + ,j) = T(cr) + T(,j) = cT(r) + ,T(j)
Conversely, by denition, T(cr) = c(T(r)) = cT(r)
For the rst property of linearity, we have T(cr + ,j) = cT(r) + ,T(j) =
T(cr) + T(,j)
If cr = n and ,j = , then from T(cr + ,j) = T(cr) + T(,j), we have
T( +n) = T(n) + T()
Example 146 The operator T : R
3
R
3
dened by T(w) = v w is linear
where v =(
1
,
2
,
3
) is a xed vector in R
3
and "" is the usual cross product
for vectors. This can asily be seen as follows: T(cx + ,y) = v (cx + ,y) =
(v cx) + (v ,y) = c(v x) +, (v y) = cT(x) +,T(y), where we have
resorted to the well-established identities for cross multiplication, which may
easily be veried by resorting to the determinant notation for cross product.
Since this operator is linear, it will and does, indeed, preserve linear dependence.
Thus the check for linearity may be shortened to the statement of this lemma
instead of refering to the original denition. As an immediate consequence,
linear operators preserve linear dependence. Of special importance is the fact
that for any linear operator T and 0 vector, T(0) = 0
Proof.
T(0) = T(r r)
= T(r + (1r))
= T(r) + T(1r)
= T(r) T(r)
= 0
Theorem 147 Let T be a linear operator. Then dimT(T) = : < ==
dim(T) _ :.
Denition 148 Let Let A and 1 be vector spaces and T : A 1 be an
operator. Then, the null space A(T) or kernal of T, denoted by ker T is the
set r [ T (r) = 0
Example 149 The null space for T : R
3
R
3
dened by T(w) = v w is
n [ n = cv .
Theorem 150 Let T be a linear operator. Then ker T is a vector space.
Theorem 151 Let A, 1 be vector spaces, both real or complex. Let T : A
1 be a linear operator with range (T) 1 . Then, the inverse T
1
: (T)
A exists if and only if T (r) = 0 for r = 0. This T
1
is a linear operator.
Furthermore, dimT(T) = : < and T
1
exists == dim(T) = :.
Corollary 152 Let T : A 1 be linear and dimA = dim1 = : < . Show
that (T) = 1 == T
1
exists.
Proof. If we can prove that T(x) = 0 for only x = 0, then were done. Let
c
1
, c
2
, ..., c
n
be the basis for A. Then, T (c
1
) , T (c
2
) , ..., T (c
n
) is linearly
independent so that if T(x) = 0, then we have
T
_
n
I=1
c
I
c
I
_
= 0
or
_
n
I=1
c
I
T (c
I
)
_
= 0
This is possible only when c
I
= 0 because T (c
1
) , T (c
2
) , ..., T (c
n
) is linearly
independent so that x = 0.
Conversely, if T
1
and dimT(T) = dimA = : implies dim(T) = :.
Also, dim1 = :. So, we have dim1 = dim(T) = :. Now, if T was not
surjective, we must have had y 1 such that T(x) ,= y for any x so that
(T) 1 , implying dim1 dim(T), which is the required contradiction.
Corollary 153 Let T : A 1 and o : 1 7 be bijective linear operators,
where A, 1 and 7 are vector spaces. Then the inverse (oT)
1
: 7 A of
the composite oT exists and (oT)
1
= T
1
o
1
Denition 154 Let (, |.|
1
) and (', |.|
2
) be normed spaces and T : '
a linear operator. The operator T is said to be bounded if there is a real number
c 0 such that \r |T (r)|
2
_ c |r|
1
Recall the denition of supremum. It is an upper bound and the lowest of
the upper bounds of a set. We can collect all r such that
]T(r)]
]r]
_ c and dene
a supremum out of it. Hence, we have
Denition 155 The norm of a bounded linear operator T, denoted by
|T|, is dened as |T| = sup
r
]T(r)]
]r]
.
Needless to say, this is valid when |r| , = 0. Also, the norm of |T| can be
taken over normalised vectors so that |T| = sup
]r]=1
|T (r)|.
Example 156 The identity operator
^
1 : on a normed space ,= ?
is bounded and has norm
_
_^
1
_
_
= 1.
Example 157 The zero operator
^
0 : ' on a normed space to norm
space ' is bounded and has norm
_
_^
0
_
_
= 0.
Example 158 Let be the normed space of all polynomials on J = [0, 1]
with norm given r = max
|
|r(t)|. A dierentiation operator T =
J
J|
is dened
on by T (r(t)) = r/(t) where the prime denotes dierentiation with respect
to t. This operator is linear but not bounded. Indeed, let r
n
(t) = t
n
where
: N. Then |r
n
| = 1 and T (r(t)) = r/(t) = :t
n1
so that |T (r
n
)| = : and
]T(rn)]
]rn]
= :. Since : N is arbitrary, this shows that there is no xed number
c such that
]T(rn)]
]rn]
_ c. From this, we conclude that T is not bounded.
Example 159 For T : R
3
R
3
dened by T(w) = v w, we have T(w) =
|v| |w| sin0 _ c |w| where c = |v| so that T is bounded.
Boundedness is typical; it is an essential simplication which we always have
in the nite dimensional case, as follows.
Theorem 160 If a normed space is nite dimensional, then every linear
operator on is bounded.
Theorem 161 Let T be a linear operator. Then, T is continuous if and only
if it is bounded.
Proof. Let T : A 1 be continuous. Then, |T(x) T(x
0
)|
Y
< - whenever
|x x
0
|
< c
or |T(x x
0
)|
Y
< - whenever |x x
0
|
< c
Let x x
0
=
:y
o]y]
X
for a 0. This is justied since the denominator is
bounded and not equal to zero. Then,
_
_
_
_
T(
-y
c |y|
)
_
_
_
_
Y
< - =|T(y)|
Y
< a |y|
or |T(y)|
Y
_ c |y|
for some 0 < c < a

Conversely, |T(y)|
Y
_ c |y|
Let y = x x
0
for |y|
= |x x
0
|
< c
Then, |T(x) T(x
0
)|
Y
< cc = - whenever |x x
0
|
< c
Theorem 162 Let T : ' be a linear operator and , ' are normed
spaces. Then, if T is continuous at a single point, then it is continuous.
Part II
Hilbert Spaces
63
3.5. PRE-HILBERT SPACE 65
3.5 Pre-Hilbert Space
Denition 163 A vector space 1 over F endowed with the operation ., . :
1 1 F, called the inner product, is called an inner product space or
pre-Hilbert space if \x, y, z 1 and c F it satises the following axioms:-
x +y, z = x, z +y, z
cx, y = cx, y
x, y = y, x
x, x _ 0
x, x = 0 == x = 0
An inner product space 1 is compactly written as (1, ., .).
Lemma 164 Let 1 be an inner product with a corresponding norm. For x, y
1, the space satises the Cauchy-Schwartz inequality [x, y[
2
_ x, xy, y
Proof. For y = 0, x, 0 = x, x x = x, x x, x = 0
Let c F. For y ,= 0,
x cy, xcy _ 0
= x, x cy cy, x cy _ 0
= x, x cx, y c[y, x cy, y] _ 0
For c =
y,x)
y,y)
, we have
x, x cx, y c[y, x y, x]
= x, x
y, x
y, y
x, y _ 0
= x, x
x, y
y, y
x, y
= x, x
[x, y[
2
y, y
_ 0
= x, x _
[x, y[
2
y, y
= x, xy, y _ [x, y[
2
= [x, y[
2
_ x, xy, y
If x = cy, then
x cy, x cy
= cy cy, cy cy
= 0, 0
= 0
66
or
x cj, x cy = 0
== x, x cy cy, x cy = 0
== x, x cx, y c[y, x cy, x] = 0
Again, for c =
y,x)
y,y)
, we have
x, x
y, x
y, y
x, y = 0
== x, x
x, y
y, y
x, y
= x, x
[x, y[
2
y, y
= 0
== x, x =
[x, y[
2
y, y
== [x, y[
2
= x, xy, y
i.e. equality will hold if the vectors are multiples of each other.
Example 165 For x, y F
n
where F = R or C,
x, y = (r
1
, r
2
, ..., r
n
), (j
1
, j
2
, ..., j
n
) =
n
I=1
r
I
j
I
satises the above conditions and hence is an inner product space. The rst
axiom of the inner product space is satised since
x +y, z
= (r
1
+j
1
, r
2
+ j
2
, ..., r
n
+ j
n
), (.
1
, .
2
, ..., .
n
)
= (r
1
+ j
1
).
1
+ (r
2
+j
2
).
2
+... + (r
n
+ j
n
).
n
= r
1
.
1
+ j
1
.
1
+ r
2
.
2
+ j
2
.
2
+ ... + r
n
.
n
+j
n
.
n
= (r
1
.
1
+r
2
.
2
+ ... +r
n
.
n
) + (j
1
.
1
+j
2
.
2
+... + j
n
.
n
)
= x, z +y, z
Next,
cx, y
= c(r
1
, r
2
, ..., r
n
), (j
1
, j
2
, ..., j
n
)
=
n
I=1
c r
I
j
I
= c
n
I=1
r
I
j
I
= cx, y
Also,
x, y =
n
I=1
r
I
j
I
=
n
I=1
r
I
j
I
=
n
I=1
j
I
r
I
=
n
I=1
j
I
r
I
=
n
I=1
j
I
r
I
= y, x
x, x _ 0 and x, x = 0 == x = 0 are easy to see. Notice that in the eld of
real numbers and for : = 3, this is the dot product of vectors, as made extensive
use of in Physics. If we let n, then we have the space |
2
.
Example 166 For 1 _ j _ , 1
() = ) : ) is measurable on and |)|
<
where for 1 _ j < , |)|
=
__
|)|
dr
_
1/
and |)|
o
= sup
r
]}(r)]
]r]
. 1
spaces, sometimes called Lebesgue spaces, are dened using natural generalisa-
tions of j-norms for nite-dimensional vector spaces. They are named after the
French mathematician Henri Lebesgue (June 28, 1875 July 26, 1941). It is
from these spaces that Quantum Mechanics adopts its mechinary. 1
2
(, )
is a particular example of a space of square-integrable function. It is important
from an application point of view. Hilbert spaces are named after the Ger-
man mathematician David Hilbert (January 23, 1862 February 14, 1943). As
promised, measure theory will not be an integral portion so lets take a close look
at the special case j = 2 where our ordinary rules of integration will be of aid.
We can dene a dot product of two real-valued functions as
r, j =
_
r(t)j(t)dt
If we have a domain, then the integral becomes denite. If we have complex-
valued functions, then the dot product becomes
r, j =
b
_
o
r(t)j(t)dt
68
To prove that this is an inner product space, see that
x +y, z =
b
_
o
(r + j) (t).(t)dt
=
b
_
o
(r(t) + j(t)) .(t)dt
=
b
_
o
_
r(t).(t) + j(t).(t)
_
dt
=
b
_
o
r(t).(t)dt +
b
_
o
j(t).(t)dt
= x, z +y, z
Next,
x, y =
b
_
o
r(t)j(t)dt
=
b
_
o
j(t)r(t)dt
=
b
_
o
j(t)r(t)dt
=
b
_
o
j(t)r(t)dt
y, x
For the scaling axiom, we have
cx, y =
b
_
o
cr(t)j(t)dt
= c
b
_
o
r(t)j(t)dt
= cx, y
Postivity axioms x, x _ 0 and x, x = 0 == x = 0 can similarly be proved.
Recall the denition of a linear operator. The inner product looks linear in
the rst argument. However, because of the axiom x, y = y, x, we dont
have linearity in the second argument with regards to scalar multiplication. In
fact, we have 1
1
2
linearity, so to speak. This is called sesquilinear:
Proposition 167 Let (A,., .) be an inner product space. Then, for v
1
, v
2
, v
3

A over F and c, , F, the following properties hold:-
cv
1
+ ,v
2
, v
3
= cv
1
, v
3
+ ,v
2
, v
3
v
1
, cv
2
= cv
1
, v
2
v
1
, cv
2
+ ,v
3
= cv
1
, v
2
+

,v
1
, v
3
Proof. For i)
cv
1
+,v
2
, v
3
= cv
1
, v
3
+,v
2
, v
3
= cv
1
, v
3
+ ,v
2
, v
3
For ii), we have

v
1
, cv
2
= cv
2
, v
1
= cv
2
, v
1
= cv
2
, v
1
= cv
1
, v
2
and nally for iii) we have

v
1
, cv
2
+ ,v
3
= cv
2
+,v
3
, v
1
= cv
2
, v
1
+ ,v
3
, v
1
= cv
2
, v
1
+

,v
3
, v
1
= cv
1
, v
2
+

,v
1
, v
3
establishing the required identities.

Theorem 168 Every inner product space is a normed space
Proof. For |.| : \ \ F, dene |x| =
_
x, x. This is justied since
., . : \ \ F carries the same "structure" as ., . : \ \ F
Let \ be an inner product space over F.
\ c F and x, y \
By default, x, x _ 0 and x, x = 0 == x = 0 ==
_
x, x _ 0 and
_
x, x = 0 == x = 0
70
Next,
|cx| =
_
cx, cx
=
_
cx, cx
=
_
c cx, x
=
_
[c[
2
x, x
= [c[
_
x, x
= [c[ |x|
Finally,
|x +y|
2
= x +y, x+j
= x, x +y +y, x +y
= x, x +x, y +y, x +y, y
= |x|
2
+x, y +y, x+|y|
2
=

|x|
2
+x, y +y, x+|y|
2
|r|
2
+[x, y[ +[y, x[ +
|y|
2
= |x|
2
+[x, y[ +[y, x[ +|y|
2
= |x|
2
+[x, y[ +
_
y, xy, x +|y|
2
= |x|
2
+[x, y[ +
_
x, yx, y +|y|
2
= |x|
2
+[x, y[ +
_
x, yr, j +|y|
2
= |x|
2
+[x, y[ +[x, y[ +|y|
2
= |x|
2
+ 2 [x, y[ +|y|
2
_ |x|
2
+ 2 |x| |y| +|y|
2
= (|x| +|y|)
2
i.e.
|x +y|
2
_ (|x| +|y|)
2
== |x +y| _ |x| +|y|
The converse is not true in general since an additional structure is needed.
However, what we can do is we can make use of the following (called polar-
ization identity) to nd out an inner product corresponding to a norm on the
complex (and hence real) numbers.
Theorem 169 For C, r, j =
1
4
_
|r +j|
2
|r j|
2
_
+
I
4
_
|r + ij|
2
|r ij|
2
_
Proof.
|r + j|
2
|r j|
2
= r + j, r + j r j, r j
= 2 r, j + 2 j, r
from which we have |r + ij|
2
|r ij|
2
= 2 r, ij + 2 ij, r
From
4 r, j = 2 r, j + 2 r, j
= 2 r, j + 2 r, j 2 j, r + 2 j, r
= 2 r, j + 2 j, r + 2 r, j 2 j, r
= 2 r, j + 2 j, r i
2
2 r, j + 2i
2
j, r
= 2 r, j + 2 j, r + 2i r, ij + 2i ij, r
= |r +j|
2
|r j|
2
+i
_
|r + ij|
2
|r ij|
2
_
If we have a real space, then the imaginary part will equal to zero and 2 r, j +
2 j, r = 4 r, j because in a real space, conjugate of a real number is equal to
the real number.
Example 170 For the same space F
n
, we have the norm
|x|
2
= x, x =
n
I=1
r
I
r
I
=
n
I=1
[r
I
[
2
from the dot product of a vector with itself. Now we also see why we need
the conjugate and why the order makes sense, notwithstanding the conjugate.
If these were real numbers, of course [r
I
[
2
= r
I
, which brings us back to the
Euclidean norm and, of course, the usual metric.
Example 171 For inner product space 1
2
[a, /], we also have a norm induced
from the dot product:
|x|
2
= x, x
=
b
_
o
r(t)r(t)dt
=
b
_
o
[r(t)[
2
dt
It has already been proved that this norm makes 1
2
[a, /] a normed space.
Denition 172 Two elements x, y of an inner product space are orthogonal
if x, y = 0.
72
This is written as x l y. If, furthermore, the norm of the two elements is
1, then the two are said to be orthonormal to each other. Two inner product
spaces and 1 are orthogonal if \x and \y 1, x l y. This is written
as l 1.
Theorem 173 An orthonormal set is linearly independent
Proof. Let c
1
, c
2
, ..., c
n
be orthonormal. Consider c
1
c
1
+c
2
c
2
+...+c
n
c
n
= 0
Then, for any , 1, 2, ..., :,
_
n
I=1
c
I
c
I
, c
_
= 0
==
n
I=1
c
I
c
I
, c
= 0
== c
, c
= 0
== c
= 0
Since , is arbitrary, therefore c
1
c
1
+ c
2
c
2
+ ... + c
n
c
n
= 0
== c
1
= c
2
= ... = c
n
= 0
Theorem 174 The norm on an inner product space satises the parallelogram
equality |x +y|
2
+|x y|
2
= 2
_
|x|
2
+|y|
2
_
Proof.
|x +y|
2
+|x y|
2
= x +y, x +y +x y, x y
= x, x +y +y, x +y +x, x y y, x y
= x, x +x, y +y, x +y, y +x, x x, y y, x +y, y
= x, x + 0 + 0 +y, y +x, x +y, y
= 2
_
|x|
2
+|y|
2
_
Theorem 175 The polarization identity is valid for any norm space if the norm
space satises the parallelogram law.
Theorem 176 Let 1 be a inner product space and x, y 1 such that x l y.
Then, |x +y|
2
= |x|
2
+|y|
2
Proof.
|x +y|
2
= x +y, x +y
= x, x +y +y, x +y
= x, x+x, y+y, x+y, y
= |x|
2
+ 0 + 0 +|y|
2
= |x|
2
+|y|
2
This is the famous Pythagorean Theorem in :-dimensions since the vector
x is left as it is, without resorting to tuples. Also, if any two vectors (be they
sequences, functions or ordinary lines) are orthogonal, then this identity will
hold. The identity can be extended to : mutually orthogonal vectors:
x
I
, x
= /
I
c
I
for 1 _ i, , _ :
c
I
=
_
1 i = ,
0 i ,= ,
is the Kronecker delta "function" and the constant /
I
depends on the vectors
x
I
and x
. Now,
|x
1
+x
2
+ ... +x
n
|
2
= x
1
+x
2
+ ... +x
n
, x
1
+x
2
+ ... +x
n
= x
1
, x
1
+x
2
+ ... +x
n
+x
2
, x
1
+x
2
+... +x
n
+... +x
n
, x
1
+x
2
+ ... +x
n
= x
1
, x
1
+x
1
, x
2
+ ... +x
1
, x
n
+x
2
, x
1
+... +x
2
, x
n
+ ... +x
n
, x
n
= x
1
, x
1
+x
2
, x
2
+ ...+x
n
, x
n
= |x
1
|
2
+ ... +|x
n
|
2
Proposition 177 |. r|
2
+|. j|
2
=
1
2
|r j|
2
+ 2
_
_
.
1
2
(r + j)
_
_
2
Proof. We will prove that both sides are equal.
. r, . r +. j, . j
=
1
2
r j, r j + 2
_
.
1
2
(r +j) , .
1
2
(r +j)
_
==
., . r r, . r +., . j j, . j
=
1
2
r, r j
1
2
j, r j + 2
_
., .
1
2
(r +j)
_
2
2
_
r +j, .
1
2
(r +j)
_
74
==
., . ., r r, . +r, r +., . ., j j, . +j, j
=
1
2
r, r
1
2
r, j
1
2
j, r +
1
2
j, j + 2., .
2
2
., r + j
_
r, .
1
2
(r + j)
_
_
j, .
1
2
(r + j)
_
==
r, . r, . +
1
2
r, r ., j ., j +
1
2
j, j
=
1
2
r, j
1
2
r, j ., (r +j)
_
r, .
1
2
(r + j)
_
_
j, .
1
2
(r +j)
_
==
2 Re r, . +
1
2
r, r 2 Re ., j +
1
2
j, j
= Re r, j ., (r +j) r, . +
1
2
r, (r + j) j, . +
1
2
j, r +j
==
2 Re r, . +
1
2
r, r 2 Re ., j +
1
2
j, j
= Re r, j ., r ., j r, . +
1
2
r + j, r j, . +
1
2
r +j, j
==
2 Re r, . +
1
2
r, r 2 Re ., j +
1
2
j, j
= Re r, j ., r ., j r, . +
1
2
r, r +
1
2
j, r j, . +
1
2
r, j +
1
2
j, j
==
2 Re r, . 2 Re ., j
= Re r, j 2 Re r, . 2 Re j, . + Re r, j
==
0 = 0
Proposition 178 Under the inner product space R
n
, |x| = |y| == x +y, x y =
0
Proof.
x +y, x y = x, x y +y, x y
= x, x x, y +y, x y, y
= (x, x y, y) + (x, y +y, x)
= (x, x x, x) + (x, y +x, y)
= 0
Proposition 179 In an inner product space, if x, u = x, v for all x, then
u = v
Proof.
x, u x, v = 0
== x, u v = 0 \x
if x = u v, then
u v, u v = 0
== u v = 0
== u = v
Denition 180 Let 1 be a Inner Product space and 1 1. 1 is an ortho-
normal system of 1 if |x| = 1 \x 1 and x l y \x, y 1
This denition is employed when the orthonormal basis (axis) of any space
need to be claried.
Theorem 181 The inner product is continuous
Proof. Let lim
no
r
n
= r and lim
no
j
n
= j be two convergent sequences in
an inner product space 1. Then,
1
such that |j
n
j| <
t
2]rn]
whenever
:
1
and
2
such that |r
n
r| <
t
2]]
whenver :
2
. Needless to say,
these inequalities are valid \c 0.
In order to establish continuity of the inner product, we need to prove that
lim
no
r
n
, j
n
= r, j just like lim
no
) (r
n
) = ) (r) . In other words, we need to
76
prove that the sequence r
n
, j
n
converges to r, j .
[r
n
, j
n
r, j[
= [r
n
, j
n
r
n
, j +r
n
, j r, j[
_ [r
n
, j
n
r
n
, j[ +[r
n
, j r, j[
= [r
n
, j
n
j[ +[r
n
r, j[
_ |r
n
| |j
n
j| +|r
n
r| |j|
< |r
n
|
c
2 |r
n
|
+
c
2 |j|
|j|
= c
i.e. [r
n
, j
n
r, j[ < c whenever : where = max (
1
,
2
)
Corollary 182 If lim
no
r
n
= r in a inner product space and j l r
n
, then j l r
Proof. lim
no
r
n
, j = r, j
== 0 = r, j
== j l r
Just like subspaces for vectors, we have subspaces for inner product spaces
where the inner product is restricted to the vectors of the subspace.
Theorem 183 Every subset of a separable inner product space is separable.
3.6 Hilbert Spaces
Not every sequence in an inner product space converges. In fact, there is a
criterion under which any sequence r
n
converges in an inner product space.
Theorem 184 lim
no
|r
n
| = |r| and lim
no
r
n
, r = r, r == lim
no
r
n
= r
Proof. From the rst condition, we have lim
no
||r
n
| |r|| = 0 and |r
n
r|
2
=
r
n
r, r
n
r
Denition 185 A complete inner product space is known as a Hilbert Space.
Needless to say, every Hilbert Space is a Banach space. One only needs to
prove that every Cauchy sequence in a Hilbert space converges under the norm
|x| =
_
x, x.
Recalling the fact that every metric space can be completed, we can similarly
complete an inner product space and make it into a Hilbert space. Similar to
isometry, the isomorphism of inner product spaces is dened as follows:
Denition 186 Two inner product spaces (A
1
, .
1
) and (A
2
, .
2
) are isomor-
phic if there exists a bijective linear operator T : A
1
A
2
such that
T (r) , T (j)
2
= r, j
1
Such a mapping is called an isomorphism.
3.6. HILBERT SPACES 77
Being bijective and having a similar dot product guarantees that the Inner
Product spaces are the same, except for the labelling of points. Since from inner
product we can have norms and from norms we can have metrics, therefore, T
is also an isometry of A
1
onto A
2
.
Theorem 187 For every inner product space 1, there exists a Hilbert space H
and an isomorphism T : 1 where

= H. The space H is unique except
for isomorphisms.
Theorem 188 Let H be a Hilbert space with an orthonormal system o = c
n
:
: N and let r H. Then,
o
n=0
[c
n
, r[
2
_ |r|
2
Proof. Let r
|
:= r
|
n=0
c
n
, r c
n
\/ N, then
r
|
, c
n
=
_
r
|
n=0
c
n
, r c
n
, c
n
_
= r, c
n

_
|
n=0
c
n
, r c
n
, c
n
_
= r, c
n

_
|
n=0
c
n
, r
_
c
n
, c
n
= 0 \: = 0, ..., /
Hence r
|
l c
n
and r
|
l
|
n=0
c
n
, r c
n
By Pythagoras theorem,
|r|
2
= |r
|
|
2
+
_
_
_
_
_
|
n=0
c
n
, r c
n
_
_
_
_
_
2
= |r
|
|
2
+
|
n=0
[c
n
, r[
2
|c
n
|
2
= |r
|
|
2
+
|
n=0
[c
n
, r[
2
_
|
n=0
[c
n
, r[
2
i.e.,
o
n=0
[c
n
, r[
2
_ |r|
2
. This is the famous Bessels inequality.
78
Theorem 189 Let H be a separable Hilbert space and o = c
n
: : N be an
orthonormal system. Then, the following are equivalent
1. o is an orthonormal basis
2. x l o implies x = 0 \x H
3. x =
o
n=0
c
n
, xc
n
\x H
4. x.y =
o
n=0
x,c
n
c
n
, y \x, y H
5. |x|
2
=
o
n=0
[c
n
, x[
2
\x H
Proof. 1.==2.
Let x l o for any x H. If x ,= 0 or
x
]x]
,= 0, then c
n
,
x
]x]
= 0 ==
x
]x]
o i.e. o is not an orthonormal system rather o '
x
]x]
is which is a
contradiction
2.==3.
To prove that x =
o
n=0
x,c
n
c
n
\x H converges, consider
_
c
I
, x
o
n=0
x,c
n
c
n
_
\i
= c
I
, x
_
c
I
,
o
n=0
x,c
n
c
n
,
_
= c
I
, x
_
o
n=0
x,c
n
_
c
I
, c
n
= c
I
, x x,c
I
= 0
Since c
I
,= 0 \i, therefore x
o
n=0
x,c
n
c
n
= 0
or x =
o
n=0
x,c
n
c
n
3.==4.
x.y =
_
o
n=0
x,c
n
c
n
,
o
n=0
y,c
n
c
n
_
=
o
n=0
x,c
n
y,c
n
c
n
, c
n
=
o
n=0
x,c
n
c
n
, y
4.==5.
Set x = y in 4
5.==1.
Suppose S is not an orthonormal basis. Then, x H such that |x| = 1
and o ' x is an orthonormal system.
But since |x|
2
=
o
1 =

n=0
[c
n
, x[
2
and x is orthonormal to c
n
o, then
1 = 0 which is absurd.
Just like subspaces for inner product spaces and vector spaces, we can have
subspaces for Hilbert spaces. However, a subspace of a Hilbert space need not
be complete. All the theorems of completeness and subspaces are applicable
here. In particular,
Theorem 190 Every subspace of a Hilbert space is complete if and only if the
subspace is closed.
Theorem 191 Every nite dimensional subspace of a Hilbert space is complete.
Denition 192 Let H
1
and H
2
be two Hilbert spaces of dimension : and /
respectively. Given two vectors (r
1
, r
2
, ..., r
n
) H
1
and (j
1
, j
2
, ..., j
3
) H
2
.
The tensor product of x and y, written compactly as x y, or even xy is
dened as
x y := (r
1
j
1
, r
1
j
2
, ..., r
1
j
|
, r
2
j
1
, r
2
j
2
, ..., r
2
j
|
, ..., r
n
j
1
, r
n
j
2
, ..., r
n
j
|
)
One can even take the tensor product of two spaces altogether to form a
"bigger" space by taking the tensor of each element of the former space with
each element of the latter space i.e. H
1
H
2
= r j, r H
1
, j H
2
.
This construction accounts for a system of particles..If, however, one wishes to
break a Hilbert space into its constituent orthogonal spaces, then one considers
the direct sum of two spaces. If nitely many Hilbert spaces H
1
, H
2
, ..., H
n
are
given, one can construct their direct sum. Formally, this is done as follows. Let
H
1
and H
2
be two Hilbert spaces over a Field F. The Cartesian product H
1
H
2
can be given a space structure by using the direct sum H
1
H
2
and then turn
this into a Hilbert space by dening the inner product as
(r
1
, r
2
, ..., r
n
) , (j
1
, j
2
, ..., j
n
) = r
1
, j
1
+r
2
, j
2
+... +r
n
, j
n
80
and
c(x, y) = (cx,cy)
Proposition 193 Let H
1
, H
2
, H
3
be Hilbert spaces of dimension :, / and :
respectively, over a eld F. For c F, y, y
t
H
1
, x, x
t
H
2
and w H
3
(x y) w = x (y w)
c(x y) = (cx) y = r (cy)
(x +x
t
) j = (x y) +(x
t
y)
x (y +y
t
) = (x y) + (x y
t
)
Proof. (x y) w
= (r
1
j
1
, r
1
j
2
, ..., r
1
j
|
, r
2
j
1
, r
2
j
2
, ..., r
2
j
|
, ..., r
n
j
1
, r
n
j
2
, ..., r
n
j
|
)(n
1
, n
2
, ..., n
n
)
=
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
r
1
j
1
n
1
, r
1
j
2
n
1
, ..., r
1
j
|
n
1
,
r
2
j
1
n
1
, r
2
j
2
n
1
, ..., r
2
j
|
n
1
, ...,
r
n
j
1
n
1
, r
n
j
2
n
1
, ..., r
n
j
|
n
1,
r
1
j
1
n
2
, r
1
j
2
n
2
, ..., r
1
j
|
n
2
,
r
2
j
1
n
2
, r
2
j
2
n
2
, ...,
r
2
j
|
n
2
, ...,
r
n
j
1
n
2
, r
n
j
2
n
2
, ...,
r
n
j
|
n
2,
.
.
.
r
1
j
1
n
n
, r
1
j
2
n
n
, ..., r
1
j
|
n
n
,
r
2
j
1
n
n
, r
2
j
2
n
n
, ..., r
2
j
|
n
n
, ...,
r
n
j
1
n
n
, r
n
j
2
n
n
, ..., r
n
j
|
n
n
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
=
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
r
1
(j
1
n
1
) , ...,
r
1
(j
|
n
1
) , r
2
(j
1
n
1
) , ...,
r
2
(j
|
n
1
) , ..., r
n
(j
1
n
1
) , ...
r
n
(j
|
n
1,
) , r
1
(j
1
n
2
) , ...,
r
1
(j
|
n
2
) , r
2
(j
1
n
2
) , ...,
r
2
(j
|
n
2
) , ..., r
n
(j
1
n
2
) , ...,
r
n
(j
|
n
2,
)
.
.
.
r
1
(j
1
n
n
) , ...,
r
1
(j
|
n
n
) , r
2
(j
1
n
n
) , ...,
r
2
(j
|
n
n
) , ...,
r
n
(j
1
n
n
) , ...,
r
n
(j
|
n
n
)
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
= x (y w)
Next, c(x y)
= c(r
1
j
1
, r
1
j
2
, ..., r
1
j
|
, r
2
j
1
, r
2
j
2
, ..., r
2
j
|
, ..., r
n
j
1
, r
n
j
2
, ..., r
n
j
|
)
= (cr
1
j
1
, cr
1
j
2
, ..., cr
1
j
|
, cr
2
j
1
, cr
2
j
2
, ..., cr
2
j
|
, ..., cr
n
j
1
, cr
n
j
2
, ..., cr
n
j
|
)
Then, ((cr
1
) j
1
, ..., (cr
1
) j
|
, (cr
2
) j
1
, ..., (cr
2
) j
|
, ...,
_
cr
(n)
_
j
1
, ...,
_
cr
(n)
_
j
|
) =
(cx) y
and (r
1
(cj
1
) , ..., r
1
(cj
|
) , r
2
(cj
1
) , ..., r
2
(cj
|
) , ..., r
n
(cj
1
) , ..., r
n
(cj
|
)) =
x (cy)
For the third proposition,
(x +x
t
) y
=
((r
1
+r
t
1
) j
1
, (r
1
+ r
t
1
) j
2
, ..., (r
1
+ r
t
1
) j
|
, (r
2
+r
t
2
) j
1
, ...,
(r
2
+ r
t
2
) j
|
, (r
3
+r
t
2
) j
2
, ..., (r
n
+ r
t
n
) j
1
, ..., (r
n
+ r
t
n
) j
|
)
= (r
1
j
1
+r
t
1
j
1
, ..., r
1
j
|
+ r
t
1
j
|
, ..., r
n
j
|
+ r
t
n
j
|
)
= (r
1
j
1
, ..., r
1
j
|
, ..., r
n
j
|
) + (r
t
1
j
1
, ..., r
t
1
j
|
, ..., r
t
n
j
|
)
= (x y) + (x
t
y)
The fourth proposition follows in a similar manner.
H
1
H
2
is a Hilbert space. Firstly, the set of the tensor product of elements
from two dierent elds yields a eld itself. Secondly, one can construct a vector
space from two others since every vector space is free and nally, one can dene
the inner product ., .
11_12
: H
1
H
2
F
1
F
2
by v
1
v
2
, u
1
u
2
11_12
=
v
1
, u
1
2
, u
2
for v
1
, u
1
H
1
and v
2
, u
2
H
2
. Completeness can be shown
by using the same inner product.
If H
1
and H
2
have orthonormal bases c
n
and c/
|
, respectively, then
c
n
c/
|
is an orthonormal basis for H
1
H
2
. Furthermore, the dimension of
H
1
H
2
is the product (as cardinal numbers) of the Hilbert dimensions i.e. the
dimension of such a space is : /.
Proof. Let c
1
, c
2
, ..., c
n
and c
t
1
, c
t
2
, ..., c
t
|
be linearly independent basis for H
1
and H
2
respectively. Then, From the basis (c
1
, c
2
, ..., c
n
) (c
t
1
, c
t
2
, ..., c
t
|
) we can
form the sum c
1
c
t
1
c
1
c
t
1
+ c
1
c
t
2
c
1
c + ... + c
1
c
t
|
c
1
c
t
|
+ c
2
c
t
1
c
2
c
t
1
+ c
2
c
t
2
c
2
c
t
2
+ ... +
c
2
c
t
|
c
t
2
c
t
|
+ ... + c
n
c
t
1
c
n
c
t
1
+c
n
c
t
2
c
n
c
t
2
+ ... + c
n
c
t
|
c
t
n
c
t
|
If (c
1
, c
2
, ..., c
n
) (c
t
1
, c
t
2
, ..., c
t
|
) = 0
then c
1
c
1
(c
t
1
c
t
1
+ c
t
2
c
t
2
+ ... + c
t
2
c
t
|
) + c
2
c
2
(c
t
1
c
t
1
+ c
t
2
c
t
2
+ ... + c
t
2
c
t
|
) + ... +
c
n
c
n
(c
t
1
c
t
1
+ c
t
2
c
t
2
+ ... + c
t
2
c
t
|
) = 0
=(c
1
c
1
+c
2
c
2
+ ... +c
n
c
n
) (c
t
1
c
t
1
+c
t
2
c
t
2
+ ... + c
t
2
c
t
|
) = 0
=(c
1
c
1
+c
2
c
2
+ ... + c
n
c
n
) = 0 or (c
t
1
c
t
1
+c
t
2
c
t
2
+ ... +c
t
2
c
t
|
) = 0
=c
1
, c
2
, ..., c
n
= 0 or c
t
1
, c
t
2
, ..., c
t
n
= 0
=c
1
c
t
1
, c
1
c
t
2
, ..., c
1
c
t
|
, ..., c
n
c
t
|
= 0
=(c
1
, c
2
, ..., c
n
) (c
t
1
, c
t
2
, ..., c
t
|
) is linearly independent
This completes the proof.
Theorem 194 The collection of all operators on a Hilbert space form a space
themselvces
This is known as the dual space.
Lemma 195 The hermitian conjugate of any vector in a Hilbert space H is an
element of the dual space of H
A dual space of any Hilbert space H is a collection of all functionals on H.
The Hermitian of any vector is seen by applying the Hermitian operator on the
vector with the property that T(x), y = x, T*(y) = T*(x), y =< r, T(j)
82
The addition of operators is commutative. However, the products of op-
erators is not always commutative. The product is, however, associative. A
functional, on the other hand, is a mapping from a vector space to a eld.
Many properties of functionals are analogous to those of operators.
Denition 196 Let H
1
, H
2
be Hilbert spaces and T : H
1
H
2
be a bounded,
linear operator. Then, the Hilbert adjoint operator T*: H
2
H
1
of T is such
that, \x H
1
and y H
2
T(x), y = x, T*(y)
Theorem 197 The Hilbert adjoint operator T* of T is unique
Proof. Let T*
1
: H
2
H
1
and T*
2
: H
2
H
1
be Hilbert adjoints of
T : H
1
H
2
Then, \x H
1
and y H
2
T(x), y
= x, T*
1
(y)
= x, T*
2
(y)
== T*
1
(y) = T*
2
(y) \y H
2
== T*
1
= T*
2
Denition 198 A bounded linear operator T : H H on a Hilbert space H
is said to be
self-adjoint or Hermitian if T*= T
unitary if T is bijective and T*= T
1
Thus, if T is self-adjoint, then T(x), y = x, T(y)
Proposition 199 Let H
1
and H
2
be Hilbert spaces and T : H
1
H
2
, o :
H
1
H
2
be bounded, linear operators and c be any scalar. Then,
1. T*(y), x = y, T(x)
2. (o +T)*= T*+o*
3. (cT)*= cT*
4. (T*)*= T
5. T*T =
^
0 =T =
^
0
6. (oT)*= T*o*
Proof. T*(y), x = x,T*(y) = T(x), y = y,T(x)
Next, \x, y, we have
(T +o) *(x), y = x, (T + o) (y)
= x, T(y) + o(y)
= x, T(y) +x, o(y)
= T*(x), y +o*(x), y
= (T* +o*)(x), y
Hence (T + o)*= T*+o*
For 3.,
(cT)*(y), x = y, (cT)(x)
= y, (cT)(x)
= y, cT(x)
= cy, T(x)
= cT*(y), x
= cT*(y), x
Apply 1. and then the denition of the adjoint to get 4.
For 5., T* (T (y)) , x =
^
0(y), x
_
=0, x = 0
=T* (T (y)) , x = T (y) , T(x) = 0
Since x, y ,= 0 are arbitrary, therefore T(x) =0 \x hence T =
^
0
Lastly, (oT)*(x), y = (x, o (T (y)) = (o*(x), T (y)
= (T* (o* (x)) , y \x, y hene (oT)*= T*o*
Example 200 The usual dot product ., . : C
n
C is dened as x, y =
x
T
y where x,y are column vectors.
Denition 201 Let T be an operator and x be an element of a Hilbert space.
If there is a scalar c such that T(x) = cx, then c is called the eigenvalue T
corresponding to that eigenvector
Theorem 202 The eigenvalues of every Hermitian operator are real
Proof. Let H be a Hilbert space over F and T(x) = cx for c F and x H
Then,
T(x), y = T*(x), y
= cx, y = cx, y
= cx, y = cx, y
= c = c if x, y , = 0
84
Denition 203 Let H be a Hilbert space and l H. An operator
^
1 : H l
is called the projection operator if
^
1
=
^
1 and
^
1
2
=
^
1.
The product of two commuting projection operators is also a projection
operator.
Proof. Let
^
1
1
and
^
1
2
be two projection operators. Then,
_
^
1
1
^
1
2
_
=
^
1
2
^
1
1
=
^
1
2
^
1
1
=
^
1
1
^
1
2
And
_
^
1
1
^
1
2
_
2
=
^
1
1
^
1
2
^
1
1
^
1
2
=
^
1
1
^
1
1
^
1
2
^
1
2
=
^
1
2
1
^
1
2
2
=
^
1
1
^
1
2
justifying the requirments for projectivity of the operator
^
1
1
^
1
2
.
The sum of two projection operators is not necessarily a projection operator
itself. Two projection operators are orthogonal if their product is zero. Thus,
^
1
^
1
I
= c
I
^
1
I
. The sum of two projection operators is a projection operator if
and only if the projection operators are mutually orthogonal
Proof.
_
^
1
1
+
^
1
2
_
=
^
1
1
+
^
1
2
=
^
1
1
+
^
1
2
and
_
^
1
1
+
^
1
2
_
2
=
_
^
1
1
+
^
1
2
__
^
1
1
+
^
1
2
_
=
^
1
2
1
+
^
1
1
^
1
2
+
^
1
2
^
1
1
+
^
1
2
2
=
^
1
1
+
^
0 +
^
0 +
^
1
2
Appendix A
Appendix
A.1 Matrices
Let F be a eld (this requirement can be considerably weakened but for our
purposes, it suces). Then, an array of elements of the form
_
_
_
_
_
_
_
_
a
11
a
12
... a
1n
a
21
a
22
... a
2n
. .
. .
. .
a
n1
a
n2
... a
nn
_
_
_
_
_
_
_
_
with a
I
F is called a matrix with : rows and : columns. In a short way, we
will write such matrices as (a
I
).
If : = :, then the matrix is a square matrix. Otherwise, the matrix is an
: : matrix. Two : : matrices (a
I
) and
(/
I
) are equal = a
I
= /
I
\i, ,. A diagonal of the matrix consists of
elements a
II
. For a square : matrix, if a
I
= 0 for i ,= ,, then the matrix is a
diagonal matrix. If a
II
= 1 for a diagonal matrix, then such a matrix is called
the identity matrix. This is the multiplicative identity of matrices. The additive
identity of matrices is (a
I
) = 0\i, , formed using the addition rule (a
I
+ /
I
)
for each i, ,. Note, this is possible only when the matrices and 1 have the
same dimension. For = (a
I
) with dimension an : : and 1 = (/
I
) with
dimension : j, the product 1 = C :=
_
n
|=1
a
I|
/
|
_
is dened to be the
:j matrix. Multiplication is not commutative, in general. Multiplication by a
scalar c is dened as c = (ca
I
). The transpose of is the matrix
|
:= (a
I
).
The operation of a matrix on a vector can be treated as matrix multiplication
if a vector is considered to be a : 1 matrix.
85

Functional Analysis Notes

Uploaded by

Copyright:

Available Formats

You might also like

Functional Analysis Notes

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Functional Analysis Notes

Uploaded by

Copyright:

Available Formats

MTH327 Functional Analysis

Abdullah Naeem Malik

, ..., [1] , [2] , ....

10. Show that ) : A 1 is onto if and only if q : 1 A such that

for j = 2. This j can range between

, which deals with functions with a j-norm. The |

space, which will sadly not be covered in these notes

spaces are of little importance.

was hinted at in the example for vector spaces;

courtesy of the denition of the modulus, given that c is a

for 1 _ j < forms a metric space under

Assume that we have the sequence (

from which we have

spaces and not norm spaces in general.

' = A or A is said to be separable if it has a countable subset which is dense

which will be limit points of sequences

. Since this is convergent, we

This should in no way mean that every collection of convergent sequences

< c will converge to, say, r

will follow a similar pattern; we will take an

and then show that this

We will then have d (r

= 1 from which we have

is a bounded sequence for every ,, then there must exist a cor-

,= 0 for each , otherwise j = 0

for some 0 < c < a

() = ) : ) is measurable on and |)|

For ii), we have

and nally for iii) we have

establishing the required identities.

You might also like