Lecture notes in Functional analysis

BY

Prof. Jamal Shenan

 Chapter one
Metric Space

1.1 Definitions and Examples:

Introduction:
By abstract space we mean a set of unspecified elements that satisfy certain axioms, and
by choosing different set of axioms we obtain different type of abstract space, the most
famous spaces in functional analysis are metric spaces, normed spaces, inner product
spaces.
In abstract space the notation of distance just as we have in R or C has no meaning
therefore we cannot speak of limit, convergence, or continuity as these concepts involves
distance notation. To introduce the distance notation in abstract space X , we define a
function d on X  X called distance function or metric which satisfies the four
conditions of distance as we have in the following definition.
Metric (distance) function:
let X be anon-empty set , a function d : X  X  is said to be a metric on X if it
satisfies the following conditions, for all x, y, z in X ,

[M1 ] d ( x, y)  0 [ Non negativity ]

[M 2 ] d ( x, y)  0  x  y

[M 3 ] d ( x, y)  d ( y, x)
y
[ symmetry ] d (x , y )

[M 4 ] d ( x, y)  d ( x, z)  d ( z, y) [ Triangle inequality ] d (z , y )
x z
Metric space: d (x , z )

The set X along with the metric d is called metric space denoted by ( X , d ) .
Remark:
Several different metrics can be defined on a single set.

2
Before giving examples on metric spaces we need the following inequalities :

1. triangle inequality: Let z ,w  K ( or ) then z  w  z  w

zw z w
2. Let z, w  K , then   .
1 z  w 1 z 1 w

3. Holder inequality (finite form) : Let z i ,w i  K ( or ) (i  1, 2,..., n ) then

1/ p 1/ q
n
 n p  n q
 zi wi    zi  
1 1
 wi  where, p  1 and   1 . p and q are called
i 1  i 1   i 1  p q

Conjugate exponents . For p  q  2 we obtain

4. Cauchy-Schwartz inequality:
1/2 1/2
n
 n 2  n 2
i 1
zi wi    zi    wi  .
 i 1   i 1 
5. Minkowski inequality (finite form): Let z i ,w i  K ( or ) (i  1, 2,..., n ) then
1/ p 1/ p 1/ p
 n p  n p  n p
 
 i 1
zi  wi 

   zi 
 i 1 
   wi 
 i 1 
, for 1  p   , and

n n n

 zi  wi   zi   wi , for 0  p  1 .
p p p

i 1 i 1 i 1

Not, if ai , bi (i  1, 2,..., n) are non-negative real numbers, then

1/ p 1/ p 1/ p
 n p  n p  n p
   ai  bi       ai       b i  .
 i 1   i 1   i 1 

Examples on Metric space:

1) Usual metric space: (real numbers) is a metric space with the metric
d:   defined by d (x , y )  x  y  x , y  , d is called usual metric on

[M 1 ] d (x , y )  x  y  0
[M 2 ] d (x , y )  0  x  y  0  x  y

3
 [M 3 ] d ( x, y)  x  y    y  x   y  x  d ( y, x)

[M 4 ] d (x , y )  x  y  x  z  z  y  x  z  z  y  d (x , z )  d (z , y )
 ( , d ) is a metric space.

another metric defined on is d ( x, y)  x  y but the function d :   defined by

d ( x, y)   x  y  , is not a metric on
2
as M 4 is not satisfied.

2) (the complex plane) is a metric space with the metric defined by,
d :   , d ( z1 , z2 )  z1  z2 , z1 , z2 
2
3) (Euclidean plane) is a metric space with the metric d : 2
 2
 defined as

d ( X ,Y )  x1  y1  x2  y2 X  ( x1 , x2 ), Y  ( y1 , y2 )
2 2

,
or

d ( X ,Y )   x1  y1    x2  y2 
2 2 2
. (This cannot be used in case )

, d ) is a metric space, where d ( X , Y )  x1  y1  x2  y2  x3  y3 ,

3 2 2 2
4) (

X  (x 1 , x 2 , x 3 ), Y  ( y 1, y 2 , y 3 )  3
(set of ordered triples of real numbers).
n
5) (Euclidean space) ( set of all ordered n-tuples of real numbers) is a metric space
1/2
 n 2
with metric d :
n
 n
 defined by d ( X , Y )   
 i 1
xi  yi  , where


X ,Y  n
. X  ( x1 , x2 ,..., xn ), Y  ( y1 , y2 ,..., yn )

Proof:
1/2
 n 2
[ M1 ] d ( X , Y )    xi  yi  0
 i 1 
1/2
 n 2
[ M 2 ] d ( X , Y )  0    xi  yi   0  xi  yi  0  xi  yi , i  1, 2,..., n
2

 i 1 
 X Y
1/2 1/2
 n 2  n 2
[ M 3 ] d ( X , Y )    xi  yi     yi  xi   d (Y , X ) .
 i 1   i 1 
4
 [M 4 ] let X  ( x1 , x2 ,..., xn ), Y  ( y1, y2 ,..., yn ), Z  ( z1, z2,..., z n) , then
1/2
 n 2
1/2
 n 2
d ( X , Y )    xi  yi      xi  zi    zi  yi  
 i 1   i 1 
1/2 1/2
 n 2  n 2
   xi  zi     zi  yi  , by Minkowski inequality
 i 1   i 1 
 d ( X , Z )  d (Z , Y ) .
( n
, d ) is a metric space.
n n
Similarly ( , d ) is a metric space with the same metric d . (unitary space).
n
Others metrices on are :
1/ 2
 n 2
i) d 2 (X ,Y )    x i  y i 
 i 1 
n
ii) d 3 (X ,Y )   x i  y i
2

i 1

iii) d 4 (X ,Y )  max  x 1  y 1 , x 2  y 2 ,..., x n  y n  (Why minimum can't be used)

In fact [M 4 ] is d4 ( X , Y )  max  x1  z1  z1  y1 ,..., xn  zn  zn  yn 

 max  x1  z1 ,..., xn  zn   max  z1  y1 ,..., zn  yn 

 d ( X , Z )  d (Z , Y ) .

6) C a, b  : Space of all continuous real valued functions defined on  a, b is a metric

space with the metric d : C a, b  C a, b   defined by

d (f , g )  max f (x )  g (x ) f , g C a, b  .
a  x b

 M 2  d (f , g )  0  amax
 x b
f (x )  g (x )  0 f , g C a, b 

 f (x )  g (x )  0
 f ( x)  g ( x) f , g C a,b  and x  a, b 
 f  g.
M4  d ( f , g )  max f ( x)  g ( x)  max f ( x)  h( x)  h( x)  g ( x)
a  x b a  x b

 max f ( x)  h( x)  max h( x)  g ( x)  d ( f , h)  d (h, g ) .

a  x b a  x b

5
   c  a, b , d  is a metric space.

Note that amin f ( x)  g ( x) is not a metric on c  a, b as [M 2 ] is not satisfied .

 x b

7) B (A ) : Space of all bounded functions on a given set A is a metric space with the

metric d : B (A )  B (A )  defined by d ( f , g )  sup f ( x)  g ( x) , f , g  B( A) .

xA

8) Discrete (trivial) metric space:

Let X be a non-empty set, defined a mapping d : X  X  by
0 if x  y
d (x , y )   ,
1 if x  y
then  X , d  is a metric space called discrete metric space or trivial metric space.

Proof:
M 1  , M 2  and M 2  are clearly holds , we prove  M 4  ,
i) If x  y  z , then d (x , y )  d (x , z )  d (z , y )  0 , hence d (x , y )  d (x , z )  d (z , y )
ii) If x  y  z , then d (x , y )  d (x , z )  1, d (z , y )  0 , hence d (x , y )  d (x , z )  d (z , y )
iii) If x  y  z , then d (x , y )  d (x , z )  d (z , y )  1, hence d (x , y )  d (x , z )  d (z , y )

Remarks:
1) Example 8 shows that every non empty set can be regarded as a metric space.
2) The mapping d may be defined as
0, if x  y
d (x , y )  
r , if x  y , r  0 (+ve real number)

9) Sequence space L : It is the space of all bounded sequences of real or complex

numbers.
L  x  (x i ) : x i  c , i  1, 2,..., c is a positive real no .

For any X  (x i ), Y  ( y i )  L , define a mapping d : L  L  by

d (x , y )  sup x i  y i , then  L , d  is a metric space .
i

6
Proof:
The mapping d is well defined, as
d (x , y )  sup x i  y i  sup x i  sup y i   , as X  (x i ),Y  ( y i )  L are
i i i

bounded  sup x i   and sup y i  .

i i

 d (x , y ) is a finite real numbers for members X andY of L .

i)  M 1  d (x , y )  0
ii)  M 2  d (x , y )  0  sup x i  y i  0  x i  y i  0  x i  y i i 
i

x  y.
iii)  M 3  d (x , y )  d ( y , x )

iv)  M 4 d (x , y )  sup x i  y i  sup x i  z i  z i  y i

i i

 sup x i  z i  sup z i  y i  d (x , z )  d (z , y ).
i i


 L , d  is a metric space .
10) The space C: It is the space of all convergent sequence

 
c  x  (x i ) : lim x i  x for X  (x i ),Y  ( y i ) c , defined
i 
d : c c  by

d (x , y )  sup x i  y i , then (c , d ) is a metric space.

i

Note that the supremum exists because every convergent sequence is bounded.

Remark:
A mapping d : c c  defined by
d (x , y )  lim x n  y n x  ( x n ) c , y  ( y n ) c ,
n 

doesn't define a metric on c .

1
In fact if x n  and y n  0 , then lim x n  y n  0 but x n  y n .
n n 

Thus  M 2  is not satisfied.

11) The space S : It is the space of all possible sequences
S  X  (x i ) : X is a sequence . .
For X  (x i ),Y  ( y i ) in S , define a mapping d : S  S  by

7
 
1 xi  yi
d (x , y )   ,
i 1 2 1  x i  y i
i

then (S , d ) is a metric space.

1 1
Note that the factor ensures convergence of the series as lim  0 and thus the
2i i  2i
mapping d is well defined.
Proof: To prove  M 4  we use the inequality
   
  .,
1    1  1 

1 xi  yi
 M 4  d (x , y )  
i 1 2 1  x i  y i
i


1 x i  z i   z i  y i 

i 1 2 1   x i  z i    z i  y i 
i


1 xi zi 
1 zi  yi
 i  i
i 1 2 1  x i  z i i 1 2 1  z i  y i

 d (x , z )  d (z , y ) .
Then (S , d ) is a metric space.


x
p p
12 ) The space L : The space of all sequences (x i ) such that is the series i
i 1

 p  1 , is convergent . That is
 

L  X  (x i ) :  x i  , p  1 .
p p

 i 1 
For X  (x i ),Y  ( y i ) L p define a mapping d : L p  L p  by
1
  p
d (x , y )    x i  y i
p
 ,
 i 1 
Then (L p , d ) is a metric space .
Proof: The mapping d is well defined.
In fact for 1  p   , let X ,Y  L , then
p

1
 
p 
d (x , y )    x i   y  
p

 i 1 

8
 1 1
 
  

  xi   yi
p p p p
   , by Minkowski inequality .
 i 1   i 1 
We prove  M 4  .
For X  (x i ),Y  ( y i ), Z  (z i ) in L
p

1
 

d (x , y )    x i  y i
p p

 i 1 
1
 

   x i  z i   z i  y i  
p p

 i 1 
1 1
  p   p
  xi z i   z i  yi
p p
 by Minkowski inequality .
 i 1   i 1 
 d (x , z )  d (z , y ) .
p
Then (L , d ) is a metric space.
1
  2 2
Particular Case: when p  2 , we have d (x , y )    x i  y i  , and this space
 i 1 
(L2 , d ) is called a Hilbert space.
Remark:

If 0  p  1 , then d may defined by d (x , y )   x i  y i
p
, here we use the inequality
i 1
  

x  yi   xi  yi 0  p  1.
p p p
i ,
i 1 i 1 i 1

Pseudometric space (Semi metric)

let X be anon-empty set , a function d : X  X  is said to be a pesudometric on X if

it satisfies the following conditions, for all x, y, z in X ,
[M1 ] d ( x, y)  0 [ Non negativity ]
[M 2 ] d (x , y )  0  x  y
[M 3 ] d ( x, y)  d ( y, x) [ symmetry ]
[M 4 ] d ( x, y)  d ( x, z )  d ( z, y) [ Triangle inequality ]

9
That is d is called a pesudometric on a set X , if  M 1  ,  M 3  ,  M 4  are satisfied , and  M 2  ,
is not necessary satisfied, ie d (x , y )  0 need not imply that x  y . Obviously every
metric space is a pseudometric but converse need not to be true.

Examples:.
1) The space c , of all convergent sequence is a psedometric space with a psedometric
d defined by
d (X ,Y )  lim x n  y n ( x n ),( y n ) c .
n 

2) is a pseudometric space with a pseudometric d defined by

d (x , y )  x 2  y 2 x , y  ,

Diameter of a metric space:

Let (X , d ) be a metric space and A be a subset of X , then diameter of A is defined by
  A   sup d (x , y ) .
x , y A

Bounded metric space:

A metric space (X , d ) is said to be bounded if there exists a positive number M s.t
d (x , y )  M for every pair of points or , a set A is said to be bounded if   A    .

Examples:.
1) Consider X  with usual metric then (X , d ) is unbounded since the diameter of
  is infinite.
2) Consider X  with discrete metric then (X , d ) is bounded since    1  .
Remark:
The above examples show that the boundedness of a metric space depends on the
choice of the metric d .

Distance between sets:

Let (X , d ) be a metric space and A , B be two subsets of X , then the distance between
A and B is defined by
D (A , B )  inf d (a, b ) ,
aA
b B

01
and the distance D (x , B ) from a point x to a subset B is defined by
D (x , B )  inf d (x , b )
b B

Note : The distance D between A and B doesn’t define a metric on the power set of X
(see Q8/p16)

1.2 Open set, closed set:

Open ball:
Let (X , d ) be a metric space, and x 0 X , the subset B (x 0 , r ) of X , defined by
B (x 0 , r )  x  X : d (x , x 0 )  r  , r  0 is called open ball (sphere) with center
x 0 , and radius r .

Closed ball:
The subset B (x 0 , r )  x  X : d (x , x 0 )  r  is called closed ball (sphere) with center x0
and radius r .

Ball:
The subset S (x 0 , r )  x  X : d (x , x 0 )  r  is called ball or sphere with radius r and
center x 0 .

Remark:
B (x 0 , r ) , B (x 0 , r ) and S (x 0 , r ) are non-empty subsets of (X , d ) , clearly x 0  B (x 0 , r )
as d (x 0 , x 0 )  0  r for every positive real no. r .

Examples:
1- The open ball B (x 0 , r ) on with the usual metric d is the open interval
(x 0  r , x 0  r ) ie the r  nbd of x 0 .
Sol: B (x 0 , r )  x  : d (x , x 0 )  r 
 x  : x x0  r 
 x  : x 0  r  x  x 0  r 
  x 0  r , x 0  r .

00
Similarly, B (x 0 , r )   x 0  r , x 0  r  and S (x 0 , r )  x 0  r , x 0  r .
2- Consider the discrete metric space (X , d ) then,
i) B (x 0 , r )  x  X : d (x , x 0 )  r 
1
 x 0  if r  1 e .g B (1, )  1 
2
 X if r  1 e .g B (1, 2)  X
ii) B (x 0 , r )  x  X : d (x , x 0 )  r 
 x 0  if r  1
 X if r  1
iii) S (x 0 , r )  x  X : d (x , x 0 )  r 
 x 0  if r  0

 X  x 0  if r  1 .
Remark:
It is clear from the above examples that the shape of the open balls change as we change
the metric d on the set X , that is if is a metric space with the usual metric then the
open balls as in example 1 , and if is a metric space with the discrete metric d then the
open balls as in example 2 .
3- Consider the metric space  2
,d  with the metric

x 1  y 1   x 2  y 2  , X   x 1, x 2   ,Y   y 1 , y 2  
2 2
d (X ,Y )  2 2
,
then,


B (Y , r )  X  2
: x 1  y 1   x 2  y 2 
2 2

r ,

which is the set of all points X   x 1 , x 2  

2
that lies inside a circle of radius r and
center Y   y 1 , y 2  .
Open set:
Let (X , d ) be a metric space, then the subset M of X is said to be open set , if for each
x  M , r  0 , s.t B (x , r )  M , ie x  B (x , r )  M .
Thus M is open if each point of M is the center of open ball which contained in M .
Closed set:
A subset K of X is said to be closed if its complement in X is open, ie K is closed iff
K c  X  K is open set in X .
02
Examples:
1- Every set of a discrete metric space is open and close.

Proof: Let M be a subset of discrete metric space X , and x  M , then

x  B (x , r )  x   M , for r  1 ie each point in M is the center of open ball contained
in M .
Also, if F is a subset of X , then X  F is a subset of X , which is open set.
 F is closed subset of X .
2- The subset [0,1) is not open set in w.r.to usual metric but it is open set in w.r.to
discrete metric.
In fact, if is a metric space with the usual metric d , then the open ball centered at 0 ,
is not contained in the set [0,1) ,ie r s.t B (0, r )  (0  r ,0  r )  [0,1) . But if [0,1)
is subset of with respect to discrete metric, then by Example 1 [0,1) is open.

Q1/ P 23:
Theorem 1.1: every open ball is an open set and every closed ball is closed set.

Proof:
Let y  B  x , r   d  x , y   r  r  d  x , y   0,
Let r   r  d  x , y   0 , we show B  y , r   B  x , r 
let z  B  y , r   d  y , z   r 
Now, d  x , z   d  x , y   d  y , z  by triangle inequality
 d x , y   r 
 d x , y   r  d x , y   r
 z  B  x , r   B  y , r   B  x , r   B  x , r  is open set .
Similarly we can prove every closed ball is closed set. #

03
Q4/ P 24:
Theorem 1.2 : A subset M of a metric space (X , d ) is open iff it is a union of open balls.

Proof:
Let M  (X , d ) be open, then each point of M is the center of an open ball contained in
M , and hence M is the union of all these open balls.
Conversely: If M is the union of open balls, we show M is open .
Let x  M , as M is the union of open balls
 x contained in some open ball, say B  y , r1  ie x  B  y , r1  , but every open ball is
open set.
 B  y , r1  is open set, as x  B  y , r1 

 x is the center of open ball, say B  x , r  s.t x  B  x , r   B  y , r1   M .

 M is open set. #

Theorem 1.3: In any metric space (X , d )

a) The empty set  and the whole space X are open set.
b) The union of any arbitrary family of open sets is open.
c) The intersection of a finite family of open set is open.

Proof:
a) To show that  is an open we are to prove that each point in  is the center of an open
ball Contained in  , since  has no elements,   is open. The full space X is
obviously open, since each open ball centered at each of its points is contained in X .
b) let G i , i  A  1, 2,3,... be arbitrary collection of open sets in a metric space (X , d ) ,
and let G i  G i , we show G is open .
i A

let x G  x  G i  x G i , for at least one i  A , as G i is open,

i A

 ri  0, s.t B (x , ri )  G i for at least one i  A ,

 B (x , ri )  G i  G .  G is open .
i A

c) let G i , i  A  1, 2,3,...n  be a finite collection of open sets in a metric space (X , d )

n
and let G  G i , we show G is open .
i 1

let x G  x G i , i  1, 2,..., n
04
 ri  0, s.t B (x , ri )  G i , i  1, 2,..., n as G i is open.
Let r  min r1 , r2 ,..., rn  , so r  0, and B (x , r )  B (x , ri )  G i i  1, 2,..., n .
n
 B (x , r )  Gi  G
i 1

 G is open set. #

Remarks:
1) As a consequence of the above Theorem, it follows that the collection of open sets in
a metric space is closed under the formation of arbitrary unions and finite
intersections.
2) The intersection of any arbitrary collection of open sets in a metric space is not
 1 1  
necessary an open set. For, let G   , , n   be infinite collection of open
 n n  
intervals in w.r.to usual metric. G is the collection of open sets as each open
 1 1 
interval is an open set, but  ,   0 which is not open.
n  n n
3) If  is the collection of open subset of a metric space  X , d  , then  satisfies the
following axioms
i)  , X 
ii) The union of any arbitrary collection of members of  , is a member of  .
iii) The intersection of finitely collection of members of  , is a member of  .
is a topology for X , and thus A metric space is a topological space .  

Interior points:
Let  X , d  be a metric space and M be a subset of X , then x  M is said to be an
interior points of M if x is the center of open ball contained in M , ie if it is possible to
find r  0 , s.t x  B (x , r )  M .

Interior of a set:
The set of all interior points of M is called interior of a set M , denoted by M 0 .

Exterior of a set:
The interior points of X  M (complement of M ) are called the exterior of M denoted
by Ext . M , thus Ext . M   X  M  .
0

05
Example:
Let  ,d  be a metric space with the usual metric d and M  0,1 , then M 0   0,1 , and
Ext . M    M    ,0  1,   , since  M   ,0  1,   .
0

Results on interior of a set:

Let  X , d  be a metric space, and M  X then
 M 0 M ,
 M 0 is the largest open subset of M ,
 M is open iff M 0  M ( M is open iff each points of M is an interior points)
 Let M and N be subsets of a metric space  X , d  , then
i. M  N  M 0  N 0
ii.  M 0 N 0    M 
0
N

iii.  M 0 N 0    M 
0
N

Not that  M 0 N 0    M  is not always true.

0
N
For example in , if M  1, 2 , and N   2,3 , then M0 N 0  1, 2  2,3  1,3  2
but  M N   1,3  1,3 .
0 0

Cluster point ( accumulation points ) ( limit point):

Let  X , d  be a metric space, and M  X , then x  X is called a cluster point of M , if
each open ball centered at X contains at least one point of M other than x .ie, x  X
is a cluster point of M , if
 B (x , r )  x  M   or r  0, y  M s .t y  B (x , r ) .
The set of all limit points of M (cluster points ) is called the derived set of M denoted by
M .
Examples:
1) Let M   0,1 be an open interval in . then the derived set of M is M   0,1 .
 
2) Let M  1, , ,... , then M   0
1 1
 2 3 
3) Let  X , d  be the discrete metric space, and x  X , then X    as
x  B (x , r )  x  for r  1, thus the open ball centered at x contains only x .
4) has no cluster points, ie    .
5) Every point of is a cluster point , ie  .
06
 6) Every point of is a cluster point , ie  .

Closure of a set:
Let M be a subset of a metric space  X , d  , then the closure of M denoted by M is given
by M  M M .

Example:
Let M   0,1 be a subset of  ,d  , with usual metric d then
M M M    0,1 0,1  0,1 .

Results on closure of a set:

Let  X , d  be a metric space, and M  X then

 M M .
 M is smallest closed set containing M

 is closed iff M  M . (It contains each of its cluster points)

N is closed as N  N N N
is closed as  
is not closed as  

 M M
 If M , and N are subsets of a metric space  X , d  , then
i. M N M N
ii. M N  M N 
iii. M N  M N
Note that M N   M N  is not always true. for example if M  1, 2  , and
N   2,3 , then M N  1, 2 2,3  2 , while, M N        .

boundary point:
A point x of a metric space (X , d ) is said to be boundary point or a frontier point of
A , iff it is neither an interior point nor an exterior point of A .ie every open ball centered
at x contains points of A as well as points not in A .
07
Set of all boundary points of A = A  A 0 .

Example: Q11/ P 24:

a) Set of boundary points of  1,1 ,  1,1 ,  1,1 is the set 1,1
b) Set of boundary points of   0
   .
c) Set of boundary points of the disks z : z  1 and z : z  1 is the circle z : z  1 .

Dense set:
Let  X , d  be a metric space, a subset M of X is said to be dense in X (everywhere
dense) if M  X . ie every point of X is either a point of M or a limit point of M .

Equivalent dentitions of dense:

1) M is dense in X if x  X ,   0, y  M s.t d (x , y )   .
2) M is dense in X if x  X , x n   M s.t nlim

xn  x .

Examples:
1) M   0,1 is dense in X  0,1 as M  X
2) is dense in , as 
3) Is dense in itself, as  in general , every closed set is dens in itself.
4) The set A  1, , ,... is non-dense in
1 1
as A  A 0  .
 2 3 

Separable space:
The metric space  X , d  is said to be separable if it has a countable subset which is
dense in X .
thus X is separable , if  a subset M  X
i. M is countable , ie M is finite, or  a one-one correspondence with Z  , ie  1  1
function f :  M
ii. M  X

Examples:
1) is separable space since  , is countable, and  .

08
2) discrete metric space  X , d  is separable iff X is countable as
X X X X  X , thus X is a dense in itself and hence it is Separable iff X is
countable.
3) The subset M   0,1 is not dense in X  0,1 as M   0,1 is uncountable , however
M X .
4) Space L is not separable as it cannot have dense subsets which are countable.

5) The space L p with 1  p   , is separable.

Proof :
consider the set S  Y   y 1 , y 2 ,..., y n , 0, 0,.. / y i  , 1  i  n , n  
p
Clearly, S is a countable subset of L . We show S is a dense in L p .
ie x  L and   0, y  S s.t d (x , y )   .
p

Let x   x 1 , x 2 ,..., x n , x n 1 ,...  L p and

y   y 1 , y 2 ,..., y n ,0,0,...  S  L p ,
then
1 1
 
  n 

d (x , y )   xi  yi    xi  yi 
p p

p p p
xi  . (1)
 i 1   i 1 i  n 1 

Now as x   x i   L p , then the series x   (convergent) , so there exist an integer
p
i
i 1

n s.t

p
 
p
xi
i  n 1 2 , (2)
also, is dense in , then for each x i  , y i  close to x i as we wish ie
d ( x i , y i )   , for each  . Hence, we can find y  S satisfying
n
p
 xi  yi 
p
. (3)
i 1 2

Use (2) and (3) in (1), we have

1 1
 n 
    p p p
d (x , y )   xi  yi 
p
   2  2   .
p p
xi
 i 1 i  n 1   
Thus L p is separable.

09
1.3 Continuous mapping in metric space:

If  X , d 1  and Y , d 2  be two metric spaces, then the Mapping f : X Y is said to be

continuous at a point x 0  X , if for each   0,   0 such that whenever
d1 (x , x 0 )   , x  X , then d 2  f (x ), f (x 0 )    , ie x 0  B  x ,    f (x 0 )  B  f (x ),   .
A function f : X Y is said to be continuous on X , if it is continuous at each point of
X , or for each   0,   0 such that B  x ,    f 1
 B f x ,   .

Example:
Let T : X  Y  be a mapping form the metric space  X , d  into a metric space
Y , d  , defined by Tx  x , then T is continuous on X , where d is the usual metric.
In fact, if y  X , then d Tx ,Ty   x  y  d  x , y  choose    , then whenever
d  x , y    , then d Tx ,Ty    ,  T is continuous at y  X . since y is arbitrary ,
 T is continuous.

Theorem 1.4: Let X and Y be two metric space, then a mapping f : X Y is

continuous iff the inverse image of each open subset of Y is an open subset of X .

proof:
let f : X Y be continuous and G be an open subset in Y , then we show f 1
G  is
an open subset of X .
Let x  f 1 G  , and f 1 G  be non-empty subset of X we show   0 . S.t
B x ,   f G  .
1

Now, x  f 1 G  ,  f  x  G , as G open in Y .
   0 . s.t B  f  x  ,    G .
But as f is continuous  for each   0,   0 such that
B  x ,   f 1
 B f  x  ,     f 1 G  .
 B x ,   f 1
G 
f 1
G  is open in X
Conversely: Let the inverse image of each open set in Y be open in X , we show
f : X Y is continuous.
Let x  X , so that f  x  Y .
21
Let B  f  x  ,   be an open ball in Y , hence it is open set in Y (every open ball is open
set).
f 1
 B f  x  ,    is open in X ,

that is ,
x f 1
 B f  x  ,      0 s.t B  x ,    f  B f
1
x  ,   .
f is continuous. #

Similarly we can prove that f : X Y is continuous iff the inverse image of each closed
subset of Y is a closed subset of X .
Note that: the image of an open set under a continuous mapping need not to be open ,
for example if f :  is defined by f (x )  sin x , then clearly f is continuous but f
maps the open interval (0, 2 ) onto the closed interval [1,1] .

1.4 Completeness:

Convergent sequence:
A sequence  x n  in a metric space  X , d  is said to be convergent to a limit x  X , iff for
each   0, N  0 , s.t d  x n , x    , n  N , ie d  x n , x   0 as n   .
Equivalently x n  x iff for each open ball B  x ,   centered at x , N  0, s.t
x n  B  x ,   n  N .

Examples:
1) Let  x n  be a sequence in the metric space with the usual metric d (x , y )  x  y ,
n
defined by x n  , then clearly ,  x n  is convergent to 1 .
n 1

2) Let  x n    1  be a sequence in X  (0,1]  with the usual metric d (x , y )  x  y ,

n 
then  x n  is not convergent (divergent) in X as 0  (0,1] .
3) Let  X , d  be the discrete metric space , then the only convergent sequence in X is
that which consists of single repeated points (constant) for all large n .
In act if X   x n  is convergent to a limit x  X , then d  x n , x   0 as n   .

20
 0 if x  y
But d ( x , y )   .
1 if x  y
 0  d  x n , x   0,  d  x n , x   0,  x n  x
 X   x  , as n is very larg.

Remark: we see the sequence  x n   

n 
 is non-constant for all large n , this means
 n 1 
that the sequence is not convergent under the discrete metric while it is convergent in
with respect to the usual metric. This example illustrate the important point that a
convergent sequence under one metric may be divergent under a different metric.

Bounded sequence:
A sequence  x n  in a metric space  X , d  is said to be bounded sequence, if  a ve
number M s.t d  x n , x m   M , n , m  .

Theorem 1.5: Let  X , d  be a metric space, then

a) A convergent sequence in X is bounded and its limit is unique.
a) If x n  x and y n  y in X , then d  x n , y n   d  x , y  .

Proof:
a) Let  x n  be a convergent sequence in X , and x n  x then for each   0, N  0

s.t . d  x n , x    , n  N .
Taking   1 , we can find N  0 , s.t d  x n , x   1, n  N .
Let d  x 1 , x   a1 , d  x 2 , x   a2 ,..., d  x N 1 , x   aN 1 , and
Let a  max a1 , a2 ,..., aN 1 ,1 , then
d  x n , x   a, n  N .
Now, d  x n , x m   d  x n , x   d  x , x m  by triangle inequality
 a  a  2a
 d  x n , x m   2a , n , m  .
  x n  is bounded.
Assuming x n  x and x n  z in X , then we show for the uniqueness that d  x , z   0.
From the triangle inequality we have.
0  d x , z   d x , x n   d x n , z   0  0  0 .

22
 d x , z   0 x z
 the limit is unique.
b) Let x n  x  d  x n , x   0 as n   ,
and let y n  y  d  y n , y   0 as n   .
Now, d  x n , y n   d  x n , x   d  x , y   d  y n , y  by triangle inequality
 d  x n , y n   d  x , y   d  x n , x   d  y n , y   0 as n   .
Similarly,
d x , y   d x , x n   d x n , y n   d  y n , y 
 d  x , y   d  x n , y n   d  x , x n   d  y n , y   0 as n   .
 d x , y   d x n , y n   d x , x n   d  y n , y   0

 d  x n , y n   d  x , y  .#

Theorem 1.6: Let M be anon empty subset of a metric space  X , d  , then

b) x  M iff there is a sequence  x n  in M s.t x n  x .
b) M is closed iff for every sequence  x n  in M , with x n  x implies that x  M
( M is closed iff every convergent sequence in M converges in M )

Proof :
a) Let x  M we show  x n  be a sequence in M . s.t x n  x .
Now, M  M M .
If x  M , then the sequence  x , x , x ,... in M converges to x .

If X is a limit point of M , then each of the balls B  x , 1  contains an x n  M , thus

n  
 1 1
x n  B  x , .  0 d x n , x   ,
 n n

 0  d  x n , x   0 as n  
1
as n  ,
n
The sequence  x n  in M is convergent to x .
Conversely: If  x n  in M and x n  x then either x  M or every open ball centered at
x contains points x n  x , so that x is a limit point of M , hence x  M  ie either
x  M or x  M   x  M .

23
b) We know that M is closed iff M  M , then form (a) M is closed iff there is a
sequence  x n  and x n  x .  x  M  M .#

Theorem 1.7: A mapping T : X Y of a metric space  X , d  into a metric space Y , d 

is continuous at a point x 0  X , iff x n  x 0 Tx n Tx 0 .

Proof:
Assuming T to be continuous at x 0 , then for a given   0,   0 s.t
d (x , x 0 )    d (Tx ,Tx 0 )   , x  X .
Let x n  x 0  N  0 , s.t d (x n , x 0 )   , n  N .
 d (Tx ,Tx 0 )   , n  N .
 Tx n Tx 0 .
Conversely: Let x n  x 0 implies Tx n Tx 0 , we prove T is continuous at x 0 .
Suppose T is not continuous at x 0 , then there is an   0 , s.t for every   0 , there is
an x  x 0 , satisfying , d (x , x 0 )    d (Tx ,Tx 0 )   .
1 1
In particular , for   , there is an x n satisfying, d (x n , x 0 )  , but d (Tx ,Tx 0 )  
n n
clearly, d (x n , x 0 )  0 as n   ie x n  x 0 , but Tx n  doesn’t converges to Tx 0 ,

this contradicts Tx n Tx 0 ,  T is continuous . #

Cauchy sequence:

A sequence  x n  in a metric space  X , d  is said to be Cauchy sequence in X , if for

each   0 , N  0 , s.t d  x n , x m    n , m  N . ie d  x n , x m   0 as n , m   .

Theorem 1.8: Every convergent sequence in a metric space  X , d  is a Cauchy

sequence, but converse need not to be true.

Proof:
let  x n  be convergent sequence in a metric space  X , d  and x n  x , then for each

  0 , N  0 , s. t d  x n , x   n  N
2 ,

24
 
and d x m , x   m  N .
2
Now,
d  x n , x m   d x n , x   d x , x m 
 d x n , x   d x m , x 
 
  
2 2
 d x n , x m    n , m  N
  x n  is a Cauchy sequence.
Conversely: let X   0,1 be a subset in  ,d  with the usual metric d  x , y   x  y .
Let  x n    1  be a sequence in X , then clearly  x n  is not convergent in X ,
n 
but it is a Cauchy sequence.
In fact , d  x n , x m   x n  x m
 xn  xm
1 1
   0 , as n , m   .
m n
1
  x n     is a Cauchy sequence in  0,1 but not convergent in  0,1 .#
n 
Question: When the converse of the above Theorem is true.

Q2/p32. If  x n  is a cauchy sequence and has a convergent subsequence say x nk  x ,

then  x n  convergent to x .

Complete metric space:

A metric space  X , d  is said to be complete metric space if every Cauchy sequence in X
converges in X .

Theorem 1.9: A subspace M of a complete metric space X is itself complete iff M

is closed in X .

Proof:
let M be a complete subspace of a complete metric space  X , d  , we show M is closed ,
ie M  M .
Let x  M

25
  a sequence  x n  in M s.t x n  x ,
since every convergent sequence is Cauchy sequence
  x n  is a Cauchy sequence in M , by completeness of M ,
  x n  is convergent in M , ie x n  x , x  M ,  x  M ,
 M  M but M  M  M  M  M is closed.
Conversely: Let M be closed in X , and  x n  be a Cauchy sequence in M , we show  x n 
is convergent in M .
As  x n  is a Cauchy sequence in M hence it is a Cauchy sequence in X , but X is
complete.
  x n  is convergent in X , ie x n  x  X , hence by Theorem 7 x  M , but M is

closed  x  M   x n  is convergent in M .  M is complete. #

Examples on complete metric space:

1) The real line and the complex place are complete metric space.
2) is complete subspace of as is closed in .
3) is not complete subspace in , for if is complete , by Theorem 10, is closed
subset of , but is not closed  is not complete . That is why in calculus we
use rather than .
4) A discrete metric space  X , d  is complete.
Let X   x n  be a cauchy sequence in the discrete metric space  X , d  , then

d  x n , x m   0 as n , m   .

0 if x  y
But d ( x , y )   .
1 if x  y

 The only choice is d  x n , x m   0,  x n  x m .

 X   x n  consists o single repeated points as n is very large.
 X   x n  is convergent.

 X , d  is complete.

26
 n
5) The Euclidean space is complete .

Proof:
 ( m )   (1) 
(2) (m ) (m )
 (m ) (m ) (m )

let  X    X , X ,..., X ,...  be a Cauchy sequence in n
, where X   x 1 , x 2 ,..., x n  ,
     
then for each   0 , N  0 , s.t
(1)
 (1) (1) (1) 
 (m ) (k )  X   x 1 , x 2 ,... x n 
d  X , X    , m , k  N  
  (2)
 (2) (2) (2)

 n (m ) (k )  X   x 1 , x 2 ,... x n 
   x i  x i    m , k  N  
 i 1 
(m ) (k )
 x i  x i   , m , k  N for each fixed i  1, 2,..., n .
(m )
 (m ) (m ) (m ) 
X   x 1 , x 2 ,... x n 
 
 (m )  (k )
 (k ) (k ) (k )
X   x 1 , x 2 ,... x n 

  x i  is a Cauchy sequence in , for each fixed i  1, 2,..., n ,
   
but is complete.
 (m ) 
  x i  is convergent to x i in for each i  1, 2,..., n .
 
ie x 1( m )  x 1 , x 2( m )  x 2 ,..., x n ( m )  x n .
  x 1( m ) , x 2( m ) ,..., x n ( m )    x 1 , x 2 ,..., x n   X  n

X (m )
 X in. n

 n
, d  is complete metric space.

n
Similarly the unitary space can be shown a complete metric space.

6) The space L is complete:

(1)
 (1) (1) 
Proof: X   x 1 , x 2 ,... 
 
L  X   x n  / x n  c , for each n  , and (2)
 (2) (2) 
X   x 1 , x 2 ,... 
d (X ,Y )  sup x n  y n .  
n

 (m )
  (1) (2) (m )

let  X    X , X ,..., X ,...  be a Cauchy sequence in L , (m )
 (m ) (m ) 
    X   x 1 , x 2 ,... 
 
(m )
 (m ) (m ) 
where X   x 1 , x 2 ,...,  , (k )
 (k ) (k ) 
X   x 1 , x 2 ,... 
 
 
 (m ) (n ) 
then for each   0 , N  0 , s.t d  X , X    , m , n  N
 
 sup x i( m )  x i( n )   m , n  N
i

27
 (m ) (n )
 x i  x i   , m , n  N and for each fixed i  1, 2,..., .

 (m ) 
  x i  is a Cauchy sequence in , for each fixed i  1, 2,..., ,
 
but is complete.
 (m ) 
  x i  is convergent to x i in for each i  1, 2,..., n .
 
(m )
ie xi xi .
 (m ) (m ) 
  x 1 , x 2 ,...,    x 1 , x 2 ,...  X
 
(m )
X X .
we show X   x 1 , x 2 ,...  L
 (m ) 
Now, d  X , X    , m  N
 
(m ) (m )
 sup x i  x i    x i  x i   m  N , i  1, 2,....
i

since  X   L ,
(m )

 
(m )
  a real no K m say s.t x i  K m , i  1, 2,.... .

Now, by Cauchy inequality

(m ) (m ) (m ) (m )
xi  xi xi xi  xi xi  xi  Km
(m )
 X   x i   L ,  X  X  L  .
 L is complete.

L

7) The space L p is complete:

Proof:
1
 
   p
L  X  (x i ) :  x i  ,1  p    , and d  X ,Y    xi  yi
p p p
 .
 i 1   i 1 
 ( m )   (1) (2) (m )
   (m ) (m ) (m )
let  X    X , X ,..., X ,...  be a Cauchy sequence in L p , where X   x 1 , x 2 ,...,  ,
     
28
then for each   0 , N  0 , s.t
(1)
 (1) (1) 
 (m ) (n )  X   x 1 , x 2 ,... 
d  X , X    , m , n  N  
  (2)
 (2) (2) 
1 X   x 1 , x 2 ,... 
  p   
   x i( m )  x i( n )   
p
m , n  N
 i 1 
(m ) (n ) (m )
 (m ) (m ) 
 x i  x i   , m , n  N and for each fixed i  1, 2,..., . X   x 1 , x 2 ,... 
 
 (m ) 
(k )
 (k ) (k ) 
  x i  is a Cauchy sequence in X   x 1 , x 2 ,... 
, for each fixed i  1, 2,..., ,  
 
but is complete.
 (m ) 
  x i  is convergent to x i in for each i  1, 2,..., n .
 
(m )
ie xi xi .
 (m ) (m ) 
  x 1 , x 2 ,...,    x 1 , x 2 ,...  X
 
(m )
X X .
we show X   x 1 , x 2 ,...  L p
 (m ) 
Now, d  X , X    , m  N
 
 (m ) p

  xi xi p m  N .
i 1
(m ) (m )
 X  X  L p , also X  L p
(m )
 (m ) 
X  X X  X L
p

 
(m )
 X is convergent to X  L p
 L p is complete metric space.

29
8) The space c a, b  is complete with metric d given by d  f , g   max f  x   g  x  .
a  x b

Proof:
c a, b   f / f is continuous on a, b   , and d  f , g   max f  x   g  x  .
a  x b

let  f m  be any Cauchy sequence in c a, b  ,  f m    f 1 , f 2 ,..., f m ,... . then for each   0 ,
N  0 , s.t d  f m , f n    , m , n  N .
 max f m  x   f n  x    , m , n  N .
x a ,b 

 f m  x 0   f n  x 0    for each fixed x 0  a, b  and m , n  N .

 the sequence  f  x    f  x  ,..., f  x  ,... , is a Cauchy sequence
m 0 1 0 m 0 in since is
complete .
  f m  x 0   is convergent , for each x 0  a, b  let f m  x 0   f  x 0  as m   .

  f m  x   converges uniformly to f  x  x  a, b  ie f m  x   f  x  as m   .

To show to f  x c a, b  , we use the well-known Theorem which state that if the
sequence  f m  of continuous functions on a, b  converges on a, b  and converges is
uniform ( see the definition below ) on a, b  , then the limit function f is continuous on
a, b . [ see Q.9 page 40 an its answer page 625 in the text book].
Therefore f m  f , and f c a, b  .
 c a, b  is compete metric space.

Uniformly convergence:
A sequence of functions  f n  in c a, b  is said to be convergence uniformly on an
interval a, b  to a function f , if for any   0 , there exists on integer N (independent on
 ) s.t f n  x   f  x    , n  N , x  a, b  .

In the above example we have show c a, b  is a complete metric space with respect to
the metric d  f , g   max f  x   g  x  , in the next example we show that this space is
a  x b
b
not complete with respect to the metric d  f , g    f  x   g  x  dx .
a

31
 b
9) The space c a, b  is incomplete with metric d given by d  f , g    f  x   g  x  dx .
a

Proof:
For a special case consider the interval  1,1 and the sequence  f n  defined on  1,1 by

 1  1 
 
m 
,1  ,1
n 
0 if  1  x  0

 1 f m  mx
f n (x )  nx if 0  x  f n  nx
 n .
 1
1 if  x 1
n
1 1
1 0 1
m n

We show the sequence  f n  is a Cauchy sequence but not convergent in C  1,1 .

Clearly  f n  is a sequence of continuous functions (see the figure)
1 1
For m  n , ie  , the graph of f m and f n are shown in the Fig.
m n
1

Now, d  f m , f n    f m  x   f n  x  dx .
1

it is clear from the Figure that d  f m , f m  is the area of shadowed triangle shown in the
Figure and as m , n   , the triangle tends to zero ie d  f m , f m   0 .
  f n  is a Cauchy sequence .
Another proof for  f n  is a Cauchy sequence is as follows:
1
d  f m , f m    f m  x   f n  x  dx
1
1 1
m n
  f m  x   f n  x  dx   f m  x   f n  x  dx
o 1
m
1 1
m n
 
1
mx  nx dx   1  nx dx
1
m
1 1
x2  m  nx 2  n 1 1
 m  n     x    
 2 0  2 1
m
2n 2m

30
 d  f m , f m   0 as m , n   .
  f n  is a Cauchy sequence .
Now we show C  1,1 is not convergent.
1
d f n , f    f n  x   f  x  dx
1
1
0 n 1
  f  x  dx  
1 0
nx  f  x  dx   1  f  x  dx
1
n

Now if f n  f as n   then d  f n , f   0 would imply that each of the integral

approaches zero, and we have
0 if  1  x  0
f (x )   .
1 if 0  x  1
Clearly the function f (x ) is not continuous of x  0 .
ie lim f (x )  0 , lim f (x )  1 ,  f (x ) is not continuous
x 0 x 0

 f c  1,1  c  1,1 is not complete.

Remark:
The above examples show that the completeness of a metric space depends on the
choice of the metric d .

Another example to show that the space C  1,1 is not complete with respect to the
b
metric d  f , g    f  x   g  x  dx is given by,
a

10) The space C is complete:

The space C is the space of all convergent sequence and as every convergent sequence is
bounded  C is a subspace of L , as L is complete, so we need to show only C is
closed [A subspace of a complete metric space is itself complete iff it is closed] [see
Theorem 10]
To show C is closed , let X   x j  C , then there exist a sequence  X  m   in C , where
X (m )
  x 1( m ) , x 2( m ) ,... , s.t X (m )
 X .[see Theorem 7]
 for each   0 , N  0 , s.t

32
 

d X (m )

, X  sup x j m   x j 
j N 2
, m  N ,  (1, 2,3,...)


 x j m   x j  For each j   (1, 2,3,...) and m  N .
2
In particular for m  N , we have
N  
xj x j  for each j .
2
But when m  N the sequence X
N  (N ) (N )

C , ie the sequence x 1 , x 2 ,..., x j ,... is
(N )

convergent  for each   0 , N 1  0 , s.t

x j N   x  j  N 1
2
Now, we show X   x j  C , ie convergent .
Use triangle inequality , we have
x j  x  x j  x j N   x j N   x   j  N 1

 X   x j  is convergent  X C

C C but C  C C C  c is closed  C is complete.

1.5 Completion:

Isometric mapping:
Let  X , d  , and  X , d  be two metric spaces a mapping T : X  X of X into X is said to
   

 

be isometric (isometry) if d  x , y   d Tx ,Ty  , x , y  X .

It is clear that a isometric mapping T : X  X .
i. Preserves the distances.
ii. It is one to one mapping, (for Tx  Ty ,  d  x , y   0  x  y )

iii. It is continuous mapping. (for each   0 ,    s.t d  x , y     d Tx ,Ty    ).

Isomorphism:
An isometric mapping which is also onto is called an isomorphism, ie isomorphism is a
bijective isometry.

33
Isometric space:
 
The space X and X are said to be isometric or X is isometric with X , if there exist an

isomorphism between them, ie  a bijective isometry of X into X .
Remark:

If X and X are isometric , then if one has the property then so does the other even if the

algebraic properties of X and X are different.

Examples:
1) The metric space  2

,d and  ,d  where d is the usual metric, are isometric and the
isometry is given by, T : 2
 defined by T  x , y   x  iy  z .
In fact if X 1   x 1 , y 1   2
, X 2  x 2 , y 2   2
,
then d  X 1 , X 2    y 2  y 1    x 2  x 1  ,
2 2

and d TX 1 ,TX 2   d  z 1 , z 2    y 2  y 1    x 2  x 1 

2 2
,
ie d  X 1 , X 2   d TX 1 ,TX 2  , also clearly T is 1-1 and onto.

Q4/p46.
 
Theorem 1.10:If X and X are isometric, and X is complete then, X is complete.ie
completeness is preserved under isomorphism.

proof:

let  X , d  , and  X , d  be isomorphism, and T : X  X be an isomorphism.

  

 
 
Let X be complete we show X is complete. Let  y n  be a Cauchy sequence in X ,

then for given   0 , N  0 , s.t d  y n , y m    n , m  N .

but T is onto , y n , y m  X ,  T 1  y n  ,T 1  y m   X

 as T is an isometry, d T  1
 y n  ,T 1  y m    d  y n , y m   
 T 1
 y n   is a Cauchy sequence in X but X is complete .

 T 1
 y n   is convergent in X . To a limit say x  X , ie T
1
 y n   x as n  

 y n T  x  as n   , x  X T  x   X .
34
  
  y n  converges to T  x  in X ,  X is complete. #

Homeomorphism:
Let  X , d  , and  X , d  be two metric spaces a mapping T : X  X is said be a
  

 
homeomorphism if .
i. T is both one to one and onto.
ii. T and T 1 are both continuous.
Two metric spaces are said to be homeomorphic if there exists a homeomorphism
between them.

It is clear that if X and X are isometric, then they are homeomorphic.

In fact, let X and X are isometric , to show that they are homeomorphic, it is sufficient to

show that T 1 is continuous. Let T : X  X be an isometry, let
 
 
y 1 , y 2  X ,  T 1 y 1 ,T 1 y 2  X as T is onto and d T 1 y 1 ,T 1 y 2  d  y 1 , y 2  as T is an

isometry, let    , then whenever d  y 1 , y 2    , then

 
d T 1 y 1 ,T 1 y 2   , y 1 , y 2  X , ie T 1
is continuous.
ie isomorphism  homeomorphism but converse, need not to be true, for example,
2
Let T :   1,1 defined by T (x )  tan 1 x , then clearly T is a homeomorphism, but it

2 2
is not an isomorphism, as d  x , y   x  y , and d Tx ,Ty   tan 1 x  tan 1 y  x  y .
 
Q5/p46.
Show that Completeness is not preserved under homeomorphism?

Solution:
2
T :   1,1 defined by T (x )  tan 1 x , is a homeomorphism .

The space  ,d  is a metric space, and complete, but  1,1 is not complete, since it is
not closed ( a subs pace of a complete space is itself complete iff it is closed).

Remark:
Homeomorphism preserve convergence because of bicontinuity, but they don’t
necessarily preserve Cauchy sequence. For example, consider the usual metric space
X   0,1 and 1,   Y then T : X Y defined by Tx 
1
is a homeomorphism.
x

35
We observe that  x n     is a Cauchy sequence in X , where T  x n     n  is not
1
n 
Cauchy in Y .

Completion of metric space:

The rational number is not complete , but it can be enlarged to the real line , which
is complete, thus we say that is the completion of and is the smallest completion
of containing as contains as a dense subspace, ie  .
A another example for completion , Let X   0,1 and d  x , y   x  y x , y  X , then
 
 X , d  is not complete space, but X X X   0,1  X ie X contains X as a dense
 
subspace. X is a dense in X , and X is complete metric space , because it is closed,

hence X is the completion of X .

Here we claim is how to enlarge the metric space  X , d  , to obtain a new metric space
  
 X , d  which is complete and contains  X , d  as a dense subs pace.
 

Remark:
 
If X contains X as a dense subspace, and X contains W as a dense subspace then X
 
and W are isometric , ie if X  X and X W then W is isometric to X .

Completion:
A metric space  X , d  , is said to be a completion of a metric space  X , d  , if the
 

 
following conditions are satisfied

i. X is complete.

ii. X contains a dense subspace W hat is isometric with X .

Examples:

1) The completion of is itself.

2) The completion of is .
3) The completion of a,   is a,  

36
4) The completion of a, b  is a, b  .

For any arbitrary metric space X , does there exists metric space, which contain a dense
subspace W isometric with X . Following Theorem shows the existence of such
completion.

Theorem 1.11: every metric space has a completion, and the completion is unique up to
an isometry.

Ie If X is a metric space not necessary complete , then  a complete metric space  X , d 

 

 

which has a dense subspace W isometric with X , and this space X is unique up to
isometry, ie if there is another complete metric space X which contains a dense

subspace W isometric with X , then X and X are isometric.

Proof :
We sub divide the proof into the following steps.
a) Construct a metric space  X , d 
 

 

b) Construction of an isometry T : X W  X :
 
c) Show that W is a dense in X ie W  X .

d) Show that X is complete.

e) Show the uniqueness of X up to an isometry.

Let  X , d  be a given metric space, and S denote the set of all Cauchy sequence in
 X , d  , ie S   x n  /  x n  is a cauchysequnce in X  .
define a relation R in S by  x n  R  y n   nlim

d x n , y n   0 .

Then the relation R is an equivalence relation on S , indeed

i. R is reflexive as d  x n , y n   0 .
ii. R is symmetric since d  x n , y n   d  y n , x n  .
iii. R is transitive.
If  x n  R  y n  and  y n  R  z n  then  x n  R  z n  since d  x n , z n   d  x n , y n   d  y n , z n 
37
Thus the relation R decompose the set S into disjoint equivalence classes.

Let x  the equivalence class of all Cauchy sequence in S which are equivalence to
x n  .

Let X denote all the collection of all these equivalence classes of Cauchy sequences in
X .

a) Construct a metric space  X , d  :

 

 
  
 
  
Consider the mapping d : X  X  defined by d  x , y   nlim d x n , y n 
  
 
x n   x Where  y n   y . It is easy to show that
(i) This limit exists nlim d  x n , y n  exists. [See the proof in page 42 of the text book.]

 
(ii) This limit is independent of the sequences chosen form x and y ie
if  x n   z n  and  y n  t n  , then nlim d  x n , y n   lim d  z n , t n  .
 n 

[See the proof in page 43 of the text book.]

(iii)  X , d  is metric space .
 

 

 M 2  d  x , y   0
  
 lim d  x n , y n   0
  n 

 x n  R  y n 
 
x y .

 M 3  d  x , y   nlim  
     
d  x n , y n   lim d  y n , x n   d  y , x  .
 
 n   
M 4  d  x n , y n   d x n , z n   d z n , y n 
 lim d  x n , y n   lim d  x n , z n   lim d  z n , y n 
n  n  n 


     
 d x , y   d x ,z  d z , y  .
     

b) Construction of an isometry T : X W  X :
  
define a mapping T : X W  X by Tx  x , where x is the equivalence class which
contains the constant sequence  x , x , x ,... then X and W are isomorphic .

38
  
In fact, T is 1-1 T is onto , as for any x W x  X s.t Tx  x , and
 
 
d Tx ,Ty   d  x , y   lim d  x n , y n   d  x , y  , as x n  x n and y n  y n .
  n 
Hence W and X are isometric.
 
c) Show that W is a dense in X ie W  X .
  
Let x  X , and let  x n   X , then  x n  is Cauchy sequence in X .

 for each   0, N  0 , s.t d  x n , x m   , n , m  N .
2

In particular, for m  N , we have d  x n , x N   , n  N .
2

Now as x N  X ,  Tx N  x N   x N , x N ,... w .

   
d  x , x N   lim d  x n , x N     , as when n  , n  N , N   .
  n  2
 

    
 d  x , x N    and  x N W  x W  ,  x W  X W .
 

d) X is complete:
let  x n  be a Cauchy sequence in X .  d  x n , x m   0 , as n , m   .
    

   

Since W is dense in X ,  for each x n  X ,  z n W s.t d  x n , z n   .

       1
n  
Now by triangle inequality.

               
d z m ,z n   d z m ,x m  d x m ,x n  d x n ,z n 
       
1     1
 d x m ,x n  
m   n

   
 d  z m , z n   0 , as n , m  
 
  
  z m  is a Cauchy sequence in W .
 
  
Now T : X w is isometric,  as T is bijective z m  X , s.t z m  T 1  z m 
 

but  z m  is a Cauchy sequence in W .



 
  z m  is a Cauchy sequence in X . ie d  z m , z n   0 as n, m  

39
   
Let x be the equivalence class in X s.t  z m   x
   
We show x n  x in X . [ X is the class contains the sequence  z m  ].

         
d x n ,x   d x n ,z n  d z n ,x 
     
1   
 d z n ,x  .
n  
   
Since z n W  the sequence z n   z n , z n ,...  z n as z n  X , and  z m   x .

   1
 d  x n , x    lim d  z n , z m   0 , as n   .
  n m 
   
 lim x n  x in X  X is complete .
n 

e) Uniqueness of X up to an isometric;
Let,  X , d  be another completion of  X , d  we show X and X are isometric .


As X is the completion of X .
 X is complete and X contains a dense Subspace W isometric with X .
 
Let x , y  X we have sequences  x n  and  y n  in W as W is dense in X .
   
   
d x n,x   0 and d  y n , y   0 as n   .
   
We know if x n  x and y n  y then d  x n , y n   d  x , y  .

 x n  x and y n  y ,  d  x n , y n   d  x , y 
   
 The distances on X and W are preserved, but W is isometric with W
 
and W is dense subspace of X , ie X W .

 The distances on X and X are preserved (same).
 All the completions are isomorphic.
This complete the proof . #

41
1.6 Compactness of metric space:

The concept of compactness is a natural generalization of finiteness. Compact sets are

important since they have several basic properties similar to those of a finite set.

Open cover:
Let Y be a subset of a metric space  X , d  , and C  G :   I  , where I is any index set,
be a family of open subset of X such that Y  G , then we say that the family C is an
 I

open cover of Y .

Open subcover:
If C   C also is an open cover of Y , then C  is called an open subcover of Y .

Finite subcover:
If the number of members in C  is finite, then it is called a finite subcover of Y .

Compact metric space:

A subset Y of a metric space  X , d  is said to be compact, if every open cover of Y has
a finite subcover. ie for each family of open subsets G  of X , for which Y  G ,
 I
n

there exist a finite subfamily say G1 ,G 2 ,...,G n  s.t Y  G i .

i 1
n

A metric space  X , d  is itself compact, if whenever X  G , then X  G i .

 I i 1

Examples:
1)  ,d  with the usual metric d is not compact metric space .
The family C   n , n  : n  N  of open subsets of is an open cover of , since
  n , n  and the family C    2n , 2n  : n  N  is a subcover of , as,
n N

  2n , 2n  , and C   C , but C  is not finite .  is not compact.

n N

40
2) Each finite subsets in a metric space  X , d  is compact.
proof:
let G i : i  I  be an open cover of a finite set E  x 1 , x 2 ,..., x n  .
E  G i as x j  E for j  1, 2,..., n ,
i I

xj  G i ie x j  E for all j  1, 2,..., n and some i  I .

i I

ie x 1 G i 1 ,..., x n G in
n
 E  Gi 1 G i 2 ... G in , E  G ij .
j 1

 E is compact.

3) The discrete metric space  X , d  , when X is a finite set is compact.

Q2/p81.
4) The discrete metric space  X , d  , when X is infinite set is not compact.

Proof:
let  X , d  be infinite metric space, with discrete metric d . then each singleton set is
open in X . Thus for each x  X , x  is open set.
Also G x  x  , x  X  is an open cover of X .
ie X  x  and as X is infinite, the open cover G x has no finite subcover.
x X

5) A subset of is compact iff it is closed and bounded (Hein-Borel Theorem). e.g

(i) 0,1 is compact subset of
(ii) 0,1 is not compact (not closed).
(iii) ; ; are not compact (unbounded).
Note that  0,1 is compact subset of with usual metric, but with discrete metric is not
compact.

Equivalent definition of compactness:

A metric space  X , d  is said to be compact, if every sequence in X has a convergent
subsequence in X . A subset Y of X is compact if every sequence  y n  in Y has a
convergent subsequence  y nk  whose limit is an element in Y
42
Remark:
If Y  n then the following definitions of compactness are equivalent.
(1) Every open cover of Y has a finite subcover.
(2) Every sequence in Y has a convergent subsequence in Y .
(3) Y is closed and bounded.

Theorem 1.12: A compact subset of a metric space  X , d  is closed and bounded .

Proof :
let  X , d  be a metric space, and Y  X be compact we show Y is closed and bounded .
first we show Y is closed.
Let x Y ,   a sequence  x n  in Y , with x n  x , but Y is compact,   a
subsequence  x nk  in Y and x nk  x , x Y Y is closed.
[a sequence  s n  converges to s iff every subsequence. Converges to s ]
Now we show Y is bounded:
If Y is unbounded, it would contain an unbounded sequence  y n  and if  y nk  is a
subsequence in  y n  then  y nk  is also unbounded. But  y nk  is convergent subsequence
of  y n  as Y is compact, and every convergent sequence is bounded , ie  y nk  is
bounded which is a contradiction Y is bounded. #
Note that the converse of above Theorem is true only, for subsets of . (Hein - Borel
theorem). However the converse is not necessary true for metric spaces other than .

Example:
 1 1 
Let X  0,1 , and let  X , d  be a discrete metric space, let A  1, , ,... be a subset of
 2 3 
 X , d  then A is closed , and bounded but not compact .
(i) A is closed. ( every subset of discrete metric space is closed and open ).
for x , y  A , x  y , d  x , y   1 ,
1   
1 1 
Let   , then B  x ,    y  X :d  y , x     
2  2    2
ie the open all centered at x  A , contains only x
 x is not a limit point of A .
 A    ,  A  A ,  A is closed.

43
(ii) A is bounded
d  x , y   1 , x , y  A , ie S  A   1 .  A is bounded
(iii) A is not compact.
A is infinite subset of the discrete metric space  X , d  , and every discrete metric space
 X , d  , where X is infinite is not compact.

Q.9/ P.82
Theorem1.13: A closed subset of a compact metric space is compact.

Proof:
let Y be a closed subset of a compact metric space X . we show Y is compact
Let x n  be a sequence in Y , as X is compact and Y is a subset of X ,  x n  is a
sequence in X .
 x n  has a convergent subsequence in X .
ie x nk  x , x  X ,  x n  x , x  X .
We know that Y is closed iff for every sequence x n Y , x n  x ,  x Y .
Thus x n  x , and Y is closed ,  x Y .#
Y is compact.

Theorem 1.14: The continuous image of a compact set is compact

Proof: let  X , d  , Y , d  be two metric spaces, and T : X Y be continuous mapping of

X in to Y .
Let M be a compact subset of X , we show that image of M , T  M  is a compact subset
of Y .
Let  y n  be a sequence in T  M   X , as T  M  is the image of M and  y n  is a
sequence in T  M    a sequence  x n  in M s.t Tx n  y n , n  N ,
But M is compact,   x n  contains a convergent subsequence  x nk  converges in M
ie x nk  x , x  M .
As T is continuous at x  M ,
 x nk  x implies that Tx nk Tx .  y nk Tx .
  y n  has a convergent subsequence  y nk  convergent to Tx T  M  .  T  M  is
compact . #

44
 Chapter Two
Normed Space

2.1 Definitions and Examples:

Introduction:
In Chapter one we introduced the notation of metric on an abstract space, so that we can
speak of distance between elements, convergence of sequences, but we can't add two
elements because any algebraic operation is not defined, for such things we need a
different structure (different abstract object) , we need to impose further axioms here,
includes the scalar addition and multiplication, thus we define a linear space (vector
space).
Vector space alone doesn't included any metric or any topology , and metric space alone
doesn't involve algebraic operation, both of them give a normed space.

A normed space: is a vector space with a metric defined by a norm ( the generalization
of the concept of magnitude or length of a vector in the plane or in three dimensional
space).
We start with the definition of vector space

Vector space (liner space):

1) X is an abelian group under addition of vectors ie
(i) x  y  X ( X is closed under addition of vectors)
(ii) x  y  y  x
(iii)  0  X s.t x  0  x x  X ( 0 is the zero vector)
(iv) For every x  X ,  x  X s.t x   x   0
2) X is closed under scalar multiplication ie x  X and   K ,  x  X
3) Addition and scalar multiplication satisfy the following conditions
 ,   K , x , y  X .
(i)    x     x    x 
(ii)     x   x   x
(iii)   x  y    x   y
(iv) 1x  1 ( 1 being the unit element of K )

45
Subspace:

A non-empty S of a vector space X is a subspace, iff

(i) If x 1 , x 2  S , then x 1  x 2  S (closed under addition).
(ii) If x  S and   K , then  x  S . (closed under scalar multiplication)

Linear combination:
A linear combination of vectors x 1 ,..., x n or a vector space X is a vector of the form
1x 1  ...   n x n where 1 ,...,  n are scalars.

Span:
If M  X , M  x 1 ,..., x n  and Y  set of alllinear combinations of x 1,..., x n  , then Y
is a subspace of X , and we say that Y is spanned by M or M spans Y ie any element of
Y is a linear combination of the elements of M .

Linearly independent:
A set of vectors x 1 ,..., x n  is said to be L.T if 1x 1  ...   n x n  0 for
i  0 i  1, 2,..., n , ie 1  ...   n  0 .
Otherwise, they are linearly dependent, ie if 1x 1  ...   n x n  0 for at least one i ,s  0.

Example: the set of vectors 1,0,0  ,  0,1,0  ,  0,0,1 of 3

are linearly independent.

Basis:
A subset B of a vector space X is said to be a basis for X if
(1) B is linearly independent set.
(2) B spans (generates) X or ( X is spanned by B ) ie any element of X can be written
as a linear combination of B ).

Example: let X  3
, and let B  1,0,0  ,  0,1,0  ,  0,0,1 , then B is a basis for 3

(i) B is L.I set of vectors.

(ii) Any vector  x 1 , x 2 , x 3   3
can be written as
 x 1, x 2 , x 3   x 1 1,0,0   x 2  0,1,0  x 3  0,0,1 ie B spans 3
.

Remarks:
(1) If X has a basis of n elements , then every other basis of X , has n -elements.
(2) Every vector space X  0 has a basis.

46
(3) If B is a basis for X , then every non-zero x  X has a unique representation as a
linear combination of elements of B with non-zero scalars as coefficients.

Dimension of a vector pace:

The number of elements in any basis of a vector space is called dimension of X .
If X  0 then X is said to be of dimension zero.
If X  0 and X has a basis consisting of a finite no of elements, then the space is said
to be finite dimensional and we write dim X  n .
A vector space X is said to be infinite dimensional if the number of elements in its basis
is infinite.

Examples:
n n
(1) and are n -dimensional.
(2) c a, b  and L 2 are infinite dimensional.

Remark:
If X is of finite dimension n then X contains a linearly independent set of n vectors
any set of n  1or more vectors of X is linearly dependent.
ie the largest set of linearly independent has n elements and any set of n  1 elements
is linearly dependent.

Norm:
A norm of a vector space X (read or complex) is a real valued function on X
. : X  s.t x , y  X ,   K .
The following properties are satisfied.
N 1  x 0
N 2  x 0 x 0
N 3  x   x
N 4  x y  x  y (triangle inequality).
Here x and y are arbitrary vectors in X .
The vector space X together with the norm . is called a normed space . denoted by
X , .  .
If K  , then X is called real normed space and if K  then it is called a complex
normed space.

Remark:

47
The distance between any vectors x , y  X is defined by means of a norm, ie
d  x , y   x  y . Thus the norm in any metric spaced X defines a metric d on X which
is given by d  x , y   x  y .

Examples on normed space:

(1) is a normed space with respect to the norm x  x , x  .

in fact, d  x , y   x  y  x  y x , y  X .
n
(2) is a normed space the following norms defined by,
1
 n 2 2
    x i  , x  (x 1 , x 2 ,...x n )
n
(i) : , defined by x
 i 1 
1 1

1
 n 2 2
Note that d  x , y   x  y    x i  y i 
 i 1 
N 1  x 1
0
N 2  x 1
 0  xi  0  xi  0  x  0 for each i
1 1
 n 2 2  n 2 2
N 3  x 1    x i      x i    x
 i 1   i 1 
1
 n 2 2
N 4  x  y 1   xi  yi 
 i 1 
1 1
 n 2 2  n 2 2
   x i     y i  by Minkowski inequality.
 i 1   i 1 
 x 1 y 1.
n
(ii) x 2   x i .
i 1

(iii) x 3  max  x 1 , x 2 ,..., x n  .

1
 n p
  xi
p
(iv) x  .
 i 1 
4

(3) The unitary space n is a normed space.

(4) Space L p is a normed space with norm defined by . : L p 
1
 1

  p d x , y   x  y   x  y

p 
x   xi  i
p p
 ,  i   .
 i 1   i 1 
 
48
Proof:
 
L p is a vector space , if X   x 1 , x 2 ,... , Y   y 1, y 2 ,...  L p ,   x i   and  y i
p
.
p

i 1 i 1

Then , x  y   x 1  y 1 , x 2  y 2 ,... , and  x   x 1 , x 2 ,... .

Thus we have,
  

 xi  yi   xi  yi
p p p

i 1 i 1 i 1
 
 x  y  L p and  x   x  Scalars ,   x  L p
p p p
i i  ,
i 1 i 1

Thus L p is closed under addition and scalar multiplication .

N 1  x 0
N 2  x  0  xi  0, i  x i  0, i  x  0 .
p

1 1
 p    

x
p
N 3  x    x i    
p
  x
p p
i
 i 1   i 1 
1 1 1
     p   p
N 4  x  y    x i  y i   xi   yi
p p p p
  by Minkowski inequality.
 i 1   i 1   i 1 
 x  y .

(5) Space L is a normed space with norm . : L  , defined by x  sup x i ,

i

for any x , y  L  x i  c , y i  c ,  x i  y i  x i  y i  2c ,  x  y  L .
also  x  L ,  L is V.s and it is easy to show it is a n.s.

(6) c a, b  is a normed space with the following norms.

i. f 1
 sup f (x ) .
x a ,b 
b

ii. f 2
  f (x ) dx .
a
1
b 2
   f (x ) dx  .
2
iii. f 3
a 

Theorem15: Every normed space is a metric space with respect to the metric
d  x , y   x  y but converse need not to be true.

Proof:
let  X , .  be a normed space, and d  x , y   x  y .

49
(as every norm on X defines a metric d on X given by d  x , y   x  y
M 1  d  x , y   0 as x  y  0
M 2  d x , y   0  x  y  0  x  y  0  x  y

M 3  d  x , y   x  y    y  x   1 y  x  d  y , x 

M 3  d x , y   x  z  z  y  x  z  z  y  d x , z   d z , y 

  X , d  is a metric space . #
To show that converse need not to be true we give the following example.
We know that the space S of all sequences is a metric space with the metric d given by

1 xi  yi
d x , y    X   x 1 , x 2 ,... , Y   y 1 , y 2 ,...
i 1 2 1  x i  y i
i

Now this metric space S , d  can't be a normed space

Suppose it is a norm space, and this norm defines a metric d by d  x , y   x  y ,
then we see that  N 2  is not satisfied.

1 x i  y i 
1  xi  yi
  x  y   d  x ,  y      ,
i 1 2 1   x i   y i i 1 2 1   x i  y i
i i


1 xi  yi
and  x  y   d x , y    2
i 1
i
1 x i  y i
  x  y    x  y
 S , d  is a metric space but not a norm space , ie the given metric is not obtained from
any norm  S , .  is not a normed space.

Theorem16: A metric d induced by a norm on a normed space X satisfies

(a) d  x  a, y  a   d  x , y 
(b) d  x , y    d  x , y  .

Proof:
(a) d  x  a, y  a   x  a   y  a   x  y  d  x , y 
(b) d  x , y    x   y   x  y   d  x , y  #
This shows that the metric on a vector space X can be obtained from a norm if it
satisfies the above conditions (a) and (b) .

Q13/ P 66:

51
Show that the discrete metric on a vector space X  0 can't be obtain from a norm.

Proof:
Let d be obtain from a norm, then it must satisfy the condition , d  x , y    d  x , y  .
For   2, x  y ,
d  x ,  y   d  2x , 2 y   1 , and  d  x , y   2 .
Thus above condition doesn't hold.
 d can't be obtain from a norm

Theorem 17: Norm is a continuous real valued function.

Proof: first we show that, x  y  x y

x  x y y  x y  y
 x  y  x y ,
also, y   y  x  y x  x  y  x  x  y  x
 y  x  x y
 x  y  x y
 x  y  x y .

We know that f :  is continuous if f  x   f  y    where x  y   , but

. :X  , hence
Chose    , we see for each   0,   0 s.t whenever x  y   then x  y  .

 . is a continuous real function. #

Seminorm (pseudonorm):
A seminorm on a vector space X is a mapping P : X  satisfying  N 1   N 3   N 4  , but
 N 2  is not necessary satisfied.

Remark:
If P is a pseudonorm then clearly P (0)  0 , as P (0x )  0P (x ) from  N 3   P (0)  0 .

Convergent sequence:
A sequence  x n  in a normed space X is said to be convergent to a limit x  X if for
each   0, N  0 s.t x n  x   , n  N . ie x n  x  0 as n  .

50
Cauchy sequence:
A sequence  x n  in a normed space X said to be Cauchy sequence if for each
  0, N  0 s.t x n  x m   , n , m  N .

Note that according to Theorem 9, every convergent sequence is a normed space is

Cauchy sequence by converse is not true.

Banach space :
A normed space X , in which every Cauchy sequence is convergent is aid to be Banach
space.
Note that according to Theorem 10 , a subspace of a Banach space is itself a Banach
space iff it is closed.

Examples of Banach spaces:

(1) n is a Banach space with respect to each norm. for the proof, take the norm
1
 n
x   xi
 i 1
2 2
 , x 

n

, and let x 
m
 be any Cauchy sequence in n
then for each

  0, N  0 s.t x  m   x  k    m , k  N .
1
 n 2
   x i  m   x i  k  2     x i    x i     i  1, 2,3,..., n
m k

 i 1 
and complete of the proof is same as example 5 in section (1.3).
(2) L p is a Banach space.
(3) L is a Banach space.
(4) C a, b  is a Banach space with respect to the norm given by f  sup f  x  ,
x a ,b 

b
but not a Banach space with respect to the norm f   f  x dx .
a

Proof :
Let  f n  be a sequence in X  C 0,1 defined by
 1
n if 0  t 
n2
f n t   
 1 if
1
t 1
 t n2
1 1
 f n  is Cauchy sequence as, for m  n, 2
 2.
m n

52
 1 1
1 n2 m2 1
1 1 1 1 1
f n  f m   f n t   f n t dt   n  m dt    m dt    dt    0 as
0 0 1 t 1 t t m n
n2 m2

n, m   .
But  f n  is not convergent in X
1
1 n2 1
1
f n f   f n t   f t dt   n  f t dt    f t dt .
0 0 1 t
n2

f n f  0 would imply that each integral conveys zero , and have

 1
 if 0  t  1
f t    t
0 if t  0

ie f is not continuous on 0,1  f C 0,1 .
  f n  is not convergent in C 0,1
C 0,1 is not Banach space.

Convex set :
A subset A of a normed X is said to be convex if the line segment joining any two
points x , y  A lies entirely in A .
The line segment joining the points x and y is the set
Lx , y  z  2 / z   x  1    y , 0    1.
Thus A is convex if Lx , y  A , x , y  A .
or ,  x  1    y  A , x , y  A
The points x and y are called boundary points of L x , y and any other points z  Lx , y is
called interior point of L x , y .

Further properties of Normed space:

Convergence of a series for the normed space:

Given a sequence x n  , n  1 in a normed space X , we say that the series



x
k 1
k  x 1  x 2  ...  x k  ... is convergent in X if the sequence of partial sum  s n  is

convergent in X .ie
53
lim  s n   s , where, s n  s1  s 2  ...  s n , n  1, 2,... ,
n 

or s n  s in X as n   , or s n  s  0 , as n   .

Here s is the sum of the series, ie s   x k  x 1  x 2  ...  x k  ... , for some s  X .
k 1

is convergent to x  X , iff its sequence of partial sum  s n 


Thus the series xk 1
k

convergent to s  X .

If s n  s , s  X , then x
k 1
k is not convergent in X .
 
If the series 
k 1
xk is convergent, then we say that the series x
k 1
k is absolutely
convergent .

Note that in calculus every absolutely convergent is convergent, but in normed space X ,
absolute convergent may not imply convergent.

Example: Q.7/P71

Let Y   z   L
n

has only finite number of non-zero terms .
Let  y n  be a sequence in Y defined by .

  n  1 n 
y n  y j n  , where y j  2 if j  n , and y j  0 if j  n .
n
 1   1 
ie y 1  1, 0, 0,... , y 2   0, , 0,...  , y 3   0, 0, ,...  ,...
4  9   
then clearly each y n Y .

Now ,  y n is A.C but not convergent in Y .
n 1
In fact,

1

n 1
y n  y 1  y 2  ...  y n  ... ,and y n  sup y j n  
j n2
 
1
 yn  2 ,
n 1 n 1 n

which is P  series with P  2 and hence it is convergent   y n is A.C.
n 1

Now, we show y
n 1
n is not convergent in Y . .

 1 1 1 
s1  y 1 , s 2  y 1  y 2 ,..., s n  y 1  y 2  ...  y n  1, , ,..., 2 , 0, 0,.. 
 4 8 n 
54
It is clear that s n  s as n   but s Y as it consist of infinitely many non-zero
terms.

Q.8+9/P71

Theorem18: A normed space X is a Banach space iff every absolutely convergent

series in X is convergent.

Proof:

Assume X is a Banach space and x
n 1
n is A.C in X , we show it is convergent in X .

Consider  s n  the sequence of partial sum,

n
where s n  x 1  x 2  ...  x n   x k ,
k 1

Then as n  m ,
s n  s m  x 1  x 2  ...  x m  x m 1  ...  x n  x 1  x 2  ...  x m
n n
 
k  m 1
xk  
k  m 1
x k  0 as m   (the series is A.C)

 s n  s m  0 as n  m   .
  s n  is a Cauchy sequence in X .
As X is a Banach space,   s n  is convergent in X ie it has a limit s in X

 The series x
n 1
n is convergent in X .#

Conversely: Suppose that X is a normed space, and every absolutely convergent series
in X is convergent in X we show X is compete .
Let  x n  be a Cauchy sequence in X ,
1
 for each ve the integer k , ve the integer n k s.t x n  x m  , n , m  nk . k
2
Now, we show the series


     
x n1  x n2  x n1  x n3  x n2  ...  x nk  x nk 1  ... =  x n1   x nk 1  x nk ,
k 1
is convergent to some x .
In fact ,
 
x n1  x n2  x n1  ...  x n1  x n2  x n1  ...

55
 
 x n1   x nk 1  x nk
k 1

1
 x n1   k 
k 1 2

 The series x n1   x nk 1  x nk
k 1
  is A.C in X and therefore converges (by hypothesis )
to some x  X
 its sequence of partial sum  s k  also converges to x  X but

   
s k  x n1  x n2  x n1  x n3  x n2  ...  x nk  x nk 1  x nk 
 s k  x nk  x , x  X .
 The subsequence x n k  of the sequence x n  is convergent in X .
But we know if a Cauchy sequence in a metric space or a normed space has a
convergent subsequence then the whole sequence is convergent
 x n  is convergent in X .
 X is a Banach space . #
Schauder basis :

A Hamel basis for a general Vector space, in which every vector is represented by a finite
linear combination of the basis vector and this representation is unique in a Hamel basis
we can't have the concept of infinite sum (which involves the notation of convergence ).
For a normed space the concept of convergence is meaning full, we can have the concept
of infinite sum, and we can define another type of basis known as a schauder basis.

Let X be an infinite dimensional normed space, a sequence e n  of elements (vectors) in

X is said to be a schauder basis for X , if for every x  X , there is a unique sequence of
scalars  n  such that
n n

 e
i 1
i i x as n   , ie  e
i 1
i i  x  0 as n   ,

 
here x is the sum f the series  i e i , ie x  i ei .
i 1 i 1

Remarks:
(1) If X is a finite dimensional normed space, dim X  n then the schauder basis is just
a Hamel basis .

(2) Clearly a schauder basis for X is L.I, since if  e
i 1
i i  0 then  i , s  0 .

56
We know that every vector space has a Hamel basis, as for the scheduler basis, a normed
space may or may not have it .
The following Theorem given a necessary but not sufficient condition for a normed
space to have a schauder basis

Q.10/P71

Theorem19: If a normed space has a schauder basis, then it is separable.

For example L can't have a schauder basis since L is not separable.

Proof:
Let X be a normed space and e1 , e 2 ,... is a schauder basis of X .
we show that X is separable, ie  a countable dense subset M of X .
 

Define M   y : y   y i e i , y i   .
 i 1 
Then clearly M is subset of X , which is countable since is countable , we show M is
dense in X .

Let x  X , then x has a unique representation, x   x i e i , x i  .
i 1

Since is dense in and x i  ,

  a sequence   in
ij s.t as ij  x i as j   .
ie x i  ij  0 as j   .
Recall that M is dense in X if

As, ij   y j    ij e i  M . x  X ,   0, y  M s.t d (x , y )   , or
i 1
x  X , x n   M s.t lim x n  x .
Thus the sequence  y j  is a sequence n 

in M .
  
Now , y j  x  ij ei   x i ei   ij  x i ei  0
i 1 i 1 i 1
as j  

Thus, y j  x  0 as j   .
 M is dense in X and therefore M is separable. #

Finite dimensional normed space:

A normed space is said to be finite dimensional if the underlying vector space is finite
dimensional.

Theorem20: Every finite dimensional normed space is Banach space.

57
In order to prove the theorem we shall be using the following lemma.

Lemma:
Let x 1 , x 2 ,..., x n  be a linearly independent set of vectors, in a normed space X , then for
any scalars 1 ,  2 ,...,  n  there exist a number c  0 s.t
1x 1   2 x 2  ...   n x n  c  1   2  ...   n 
Proof of Theorem:
Let Y be a finite dimensional normed space , and  y m  be an arbitrary Cauchy sequence
in Y . We show  y m  is convergent to a limit y in Y .
Let dimY  n and e1 , e 2 ,..., e n  be a basis for Y , then each y m has a unique
representation of the form. y m  1 e1   2 e 2  ...   n e n .
m m m

Since  y m  is a Cauchy sequence , for each   0, N  0 s.t y m  y r   , m , r  N .

From the above Lemma, we have for some c  0 ,
  
  y m  y r  1 m   1 r  e1  ...   n m    n r  e n 

 c 1 m   1 r   ...   n m    n r  
n
 c   i m    i r    , m , r  N .
i 1
n

   i m    i r    
i 1 c
 i m   i r     For each fixed 1  i  n .

 the sequence     is a Cauchy sequence for each i in

i
m
or .
but is complete ,

 i m 
     ,   ,... i  1, 2,..., n is convergent in
i
1
i
2
.
Let  i denote the limit,
 i m   i for each fixed i.
 i m   i  0 as n   for each fixed i .
Now, using these n limits 1 ,  2 ,...,  n ,
we define y  1e1   2e 2  ...   ne n , then y Y .
As e1 , e 2 ,..., e n  forms a basis for Y .
We show y m  y .

58
  
n n
ym  y  i m   i e i   i m    i e i  0 as m 
i 1 i 1

 y m  y and thus  y m  is convergent in Y since  y m  was an arbitrary.

 Y is complete. #

Theorem21: Every finite dimensional subspace Y of a normed space X is closed in

X .(every finite dimensional normed space is closed).

Proof:
Let Y be finite dimensional subspace of X , then by Theorem 20 Y is complete . we
show it is closed .
Let y Y   a sequence  y n  in Y s.t y n  y since every convergent sequence is
Cauchy ,
  y n  is Cauchy sequence in Y , as Y is complete ,
  y n  is convergent in Y .
ie y Y Y is closed . #
Theorem22: Any subset Y of a finite dimensional normed space X is compact iff
Y is closed and bounded.

Proof:
Completeness implies closedness and boundedness so, we prove the converse only,
Let Y be closed and bounded. and let dim X  n .
let e1 , e 2 ,..., e n  be a basis for X we show Y is compact.
Let  y m  be a sequence in Y , we show it has a convergent subsequence in Y
Now, each y m has the representation
y m  1 m e1   2 m e 2  ...   n m e n ,
Since Y is bounded so is  y m 
 y m  K , m . as e1 , e 2 ,..., e n  is L.I.
 By Lemma (lemma in Th21 )
K  y m  1 m e1   2 m e 2  ...   n m e n

 
 c 1 m    2 m   ...   n m  , c  0 .
n
K
   i m   .
i 1 c

59
 the sequence of real scalars  i  m
 is bounded for each fixed i , 1 i  n .
We know from balzano-weierstrass Theorem that every bounded sequence of real
numbers has a convergent subsequence.
 i
m 

has a convergent subsequence for each fixed i .
Thus 1 m    11 , 1 2 ,... has a convergent subsequence say  1 m     11 ,  1 2 ,... converges
to a limit say 1 .
ie  1 m   1 as m   .
 
 1 m e1 has a convergent subsequence  1 e1 converges to 1e1 .  m 

   e   
1
m
1 2
m
2e  ...   n m e n 
has a convergent subsequence  1 m e1   2 m e 2  ...   n m e n  ,
that converges to 1e1  2e 2  ...  ne n as m  

  y m   1 m e1   2 m e 2  ...   n m e n 
n

y   y    that converges to y   i e i .
n
has a convergent subsequence where m 
nm nm i ei
i 1 i 1

But Y is closed  y Y .
 every sequence  y m  in Y has a convergent subsequence  y n m  that converges to a

limit y in Y .#

Equivalent norms:
A norm . on a vector space X is said to be equivalent to a norm . 0 , if  a positive
numbers a and b such that , a x 0
 x b x 0
x  X .

Example:
 max  x 1 , x 2  and  x 1  x 2 are equivalent.
2 2
Let X  2
, then , x 1
x 2

Proof:
Let x 1
 x 1 , then clearly, x 1  x 2 .
 x 2  x1 ,
2 2

thus x 1  x 1  x 2  x 1  x 1 ,
2 2 2 2 2

 x1  x1  x 2  2 x1 ,
2 2

or x 1
 x 2
 2 x 1
.
 x 1
 x 2
.

61
 Theorem23: All norms defined on a finite dimensional normed space are equivalent.

Proof:
Let . and . 0 be two norms on a finite dimensional normed space X ,
as X is a finite dimensional normed space, Let dimX  n and e1 , e 2 ,..., e n  be a basis
for X , then every x  X has a unique representation of the form,
x  1e1  ...   ne n ,
By Lemma (lemma in Theorem 21 ) as e1 , e 2 ,..., e n  are L.I
 1e1   2e 2  ...   ne n  c  1   2  ...   n 
n
 x  c  i
i 1
n
1
 x   i , (1)
c i 1

On the other hand the triangle inequality gives

x 0  1e1   2e 2  ...   ne n 0
 1 e1 0
 ...   n e n 0
n
  i ei 0
, let K  max ei 0
i
i 1
n
 K  i
i 1
n
1
 x 0
  i (2)
K i 1

From (1) and (2), We have

1 1 c
x 0 x ,or x 0
 x ,
K c K
or a x 0
 x , (3)
c
where a  .
K
By interchanging the roles of . and . 0
in (1) and (2) we obtain
x b x 0
, (4)
Form (3) and (4) we have , a x 0
 x b x 0
.
 . and . 0 are equivalent . #

Remark:

60
The above Theorem implies that convergence and divergence of a sequence in a finite
dimensional normed space doesn't depend on the particular choice of a norm on that
space .

Linear operator:

Operator:
Let X and Y be two normed spaces then a mapping T : X Y is called an operator .
D T   X is The domain of T which is a vector space, and R T  Y is the range of T
which is also a vector space over the same field.

Remarks:
(1) R T   T  x  Y / x  D T  , If R T  Y , then T is called onto.
(2) N T   x  D T  / Tx  0 , is called the null space of T or Kernal of T ,
N T   D T 

Linear operator:
An operator T is called linear operator if x , y  D T  , and scalars   K  or  ,
T satisfies
(i) T  x  y   Tx Ty
(ii) T  x   Tx
ie T preserves the vector addition and scalar multiplication .
Clearly, the conditions (i) and (ii) are equivalent to ,T  x   y   Tx  Ty

Examples:
1- Identity operator:
Let X be a normed space, the operator I x : X  X defined by
I x x  x , x  X is called identity operator and it is linear a operator.
R T   X , ie T is onto operator, and N T   0 .
2- Zero operator:
let X and Y be two normed space, the operator T : X Y defined by Tx  0, x  X is
called zero operator , and it is a linear operator, R T   0 . ie T is not onto and
N T   x  D T  / Tx  0  X .
3- Let T :  defined by Tx  x 2 , thenT is not linear operator since

62
T  x  y    x  y   x 2  2xy  y 2  Tx Ty
2

.
4- let T : c a, b   c a,b  defined by ,
t
T f t    f t  dt , f c a, b  , t  a, b  , then T is a linear operator.
a

5- let T : Pn a,b   Pn a,b  , defined by

f t  , f  Pn a, b  , where Pn a, b  is the space of all polynomials of

d
T f t  
dt
degree less or equal to n , then T is a linear operator.

Theorem25: Let T be a linear operator. Then:

(a) the range R T  is a vector space.
(b) The null space N T  is a vector space.
(c) If dim D T   n   then dim R T   n .

Proof:
(a) R T  is a vector space.
Let y 1 , y 2  R T   x 1 , x 2  D T  s.t y 1  Tx 1 and y 2  Tx 2 .
Now , x 1  x 2  D T  as D T  is a vector space
 T  x 1  x 2   Tx 1 Tx 2  y 1  y 2 , asT is linear,
So, x 1  x 2  D T  s.t T  x 1  x 2   y 1  y 2 ,  y 1  y 2  R T  .
Also, let y  R T   x  D T  s.t Tx  y
Let   K ,   x  D T  by linearity of T , T  x    y   y  R T  ,
so R T  is a vector space.

(b) N T  is a vector space

Let x 1 , x 2  N T  ,  Tx 1  0 , and Tx 2  0 .
Since T is linear, T  x 1  x 2   Tx 1 Tx 2  0 ,  x 1  x 2  N T  .
Also, let x  N T  Tx  0 , by linearity of T , T  x   Tx  0,   x  N T  .

(c) If dim D T   n , then dim R T   n .

To show that dim R T   n , we show that any set of  n  1 vector s in R T  is linearly
dependent.
63
Let  y 1, y 2 ,..., y n 1 a set of vectors in R T  .  x 1, x 2 ,..., x n 1  D T  s.t
Tx 1  y 1 , Tx 2  y 2 ,..., Tx n 1  y n 1 .
Since dim D T   n
 The set of vector x 1 , x 2 ,..., x n 1 is linearly dependent set.
 1 ,  2 ,...,  n 1  K not all zero s.t 1x 1   2 x 2  ...   n 1x n 1  0 .
Now T 1x 1   2 x 2  ...   n 1x n 1   0 as T  0   0 .
 1Tx 1   2Tx 2  ...   n 1Tx n 1  0 ,  1 y 1  2 y 2  ...   n 1 y n 1  0 , and not all  ,s are zero.
  y 1 , y 2 ,..., y n 1 is linearly dependent  R T   n .#

Inverse of linear operators:

We know that the operator T : X Y is injective (one-one) operator, if

Tx 1  Tx 2  x 1  x 2 , or x 1  x 2  Tx 1  Tx 2 .
If T : D T   R T  is 1  1 , then there exists a mapping T 1 : R T   D T  ,
which maps every y 0  R T  onto x 0  D T  for which T 1 y 0  x 0  Tx 0  y 0 .
T 1
is called the inverse of T , and T 1Tx  x , x  D T  , and TT 1x  x , x  R T  .

Theorem26: Let X and Y be two normed spaces, and let T : D T  Y be a

linear operator with D T   X and R T  Y then :
(a) T 1 : R T   D T  exists iff Tx  0  x  0 ie N T   0 .
(b) If T 1 exists , then it is linear .
(c) If dim D T   n and T 1 exists, then dim R T   n .

Proof:
(a) Let T : D T  Y be a linear operator. Let T 1 exists  T is 1  1 .
so, Tx 1  Tx 2  x 1  x 2 , let Tx  0, Tx  T 0  x  0 .
conversely: let Tx  0  x  0 , we show T is 1  1 . let Tx 1  Tx 2 ,
since T is linear, then T  x 1  x 2   Tx 1 Tx 2  0 ,
 x 1  x 2  0,  x 1  x 2 ,  T is 1  1 ,  T 1 exists.

(b) Let T 1 exists , then we need to show T 1 is linear , T 1 : R T   D T  .

Let x 1 , x 2  D T  ,  y 1 , y 2  R T  s.t y 1  Tx 1 and y 2  Tx 2 .
SinceT 1 exists ,  x 1  T 1 y 1 and x 2  T 1 y 2

64
Now, T is linear,
 T  x 1  x 2   Tx 1 Tx 2  y 1  y 2
 T 1  y 1  y 2   x 1  x 2  T 1 y 1 T 1 y 2 .
Also, T  x 1   Tx 1   y 1
 T 1  y 1    x 1  T 1 y 1 .
So T 1 is liner operator.

(c) Let dim D T   n , and T 1 exists

 by Theorem25(c, ) dim R T   dim D T 
Since T 1 : R T   D T  is linear.  dim D T   dim R T  .
So, dim D T   dim R T   n .#

Theorem27: Let T : X Y and S :Y  Z be 1  1 linear operators.

Then ST  : Z  X exists , and ST   T 1S 1 .
1 1

Proof:
Since T and S are 1  1 ,  ST : X  Z is 1  1 .  ST  exists .
1

Now , ST ST  : Z  Z .

1

 ST ST   I z
1

 S 1 ST ST   S 1I z

1
 S 1
 T ST   S 1
1

 T 1T ST   T 1S 1

1

 ST   T 1S 1 .#
1

Bounded operator:

Let X and Y be normed space, and T : D T  Y be a linear operator, where D T   X ,

then the operator T is said to e bounded if x  D  T there is a fixed positive real
number c such that ,
Tx  c x (1)
This formula shows that a bounded linear operator maps bounded sets in D T  on to
bounded sets in Y .

What is the smallest possible c such that (1) holds x  D T  ?

Form (1) we have
65
 Tx
 c , x  D T  , x  0
x
Tx
 sup  c, x 0
x D T  x
Tx
 The smallest possible c  sup , x 0 .
x
norm of the linear operator:
 Tx 
the quantity sup  , x  0  is called the norm of the linear operator T and written
x D T   x 

Tx
T  sup , here supremum exists because T is bounded linear operator .
x D T  x
x 0

Tx
Note that, T  sup  c , and with T  c , then we have form (1)
x D T  x
x 0

Tx  T x .

Lemma:
Tx
letT be a bounded linear operator then T  sup , x  0 is equivalent
x D T  x

to the norm of T given by T  sup Tx . and the norm of T satisfies the

x D T 
x 1

axioms of the norm.

Proof:
  1
let x  a and set y    x , x  0 , then clearly y  D T  and y  x / a  1 .
a 
Now ,
Tx Tx
T  sup  sup
x D T  x x D T  a
x 0 x 0

1 
 sup T  x  asT is linear
x D T  a 
x 0

 sup Ty
y D T  ,
y 1

66
 T  sup Tx
writing x for y on the right, we have x D T  .
x 1

T satisfies the axioms of the norm.

N 1  T  sup Tx  0 ,
x D T 
x 0

N 2  T  0  Tx  0 x  D T 
 Tx  0, x  D T 
 T  0 (the zero operator).
N 3  T  sup Tx  sup  Tx ,
x D T  x D T 
x 1 x 1

as Tx Y and Y is a normed space

 T   sup Tx   T
x D T 
x 1

N 4  T1 T 2  sup T1 T 2  x  sup T1x T 2 x

x 1 x 1

 sup T1x  sup T 2 x , x  D T1  , x  D T 2 

x 1 x 1

 T1  T 2
 T1  T 2  T1  T 2 .

Examples:
1- Identity operator: The identity operator I : X  X defined by Ix  x , x  X is
bounded linear operator with I  1,
Ix x
Ix  c x ,  c    1.
x x
Ix x
Thus the smallest c is 1, and, I  sup  sup 1
x D T  x x D T  x
x 0 x 0

2- Zero operator. The zero operator 0 : X Y , on a normed space X is bounded and has
norm 0  0 here Tx  0, x  D T  , and Tx  c x  c  0 .
Tx 0
also, T  sup  sup  0.
x D T  x x D T  x
x 0 x 0

3- The differential operator. The differential operator T : X 0,1  X 0,1

67
 d
defined by Tx t   x t  , with x  max x (t ) , where
t 0,1
dx
X 0,1 : Spaceof all polynomials defined on[0,1].
This operator is linear but not bounded .

Proof:
Let x t   t 2 , Tx t   2t .
x t   t 2  max t 2  1 , and
t 0,1

Tx t   2t  max 2t  2
t 0,1

Tx 2
 T  sup  2 c  2.
x 1
Now if we choose x  t , then c  3
3

In general , if x  t , Tx  nt n 1 ,  c  n .
n


So, we cannot find a fixed c  s.t Tx  c x .
Now we consider general properties of bounded linear operators.

Theorem28: If X is a finite dimensional normed space, then every linear

operator is bounded.

Proof:
Let dimX  n , and e1 , e 2 ,..., e n  be a basis for X ,
n
 for any x  X , x    i e i ,
i 1

 n
 n

so, Tx  T    i e i    iTe i as T is linear .

 i 1  i 1
n n n
Tx   Te
i 1
i i    iTe i
i 1
   i Te i
i 1
n
 K   i , where K  max Te i (1)
i
i 1

By Lemma (lemma in Theorem 21 ) , since e1 , e 2 ,..., e n  are linearly independent then
c  0 s.t,
1e1  1e 2  ...  1e n  c  1   2  ...   n 
n n
1 1
  i   e i i  x (2)
i 1 c i 1 c

68
 K
(1) and (2) imply that Tx  x  c  x for any x  X ,
c
 T is bounded . #
Remark:
The above Theorem doesn't hold for infinite dimensional normed space, for example the
space X 0,1 is infinite dimensional normed space , and the differentiation operator
defined on X 0,1 is linear but not bounded.

continuous operator:

Let T : D T  Y be any operator (not necessarily linear), where D T   X and X and

Y are normed spaces, the operator T is said to be continuous at x 0  D T  , if for every
  0,   0 , s.t
Tx Tx 0   , x  D T  whenever x  x 0   .
T is continuous on X if it is continuous at every point x  D T  .
Alternative definition of continuous is,
T is continuous if x n  x in D T   T  x n  T  x  in Y .

Theorem29: Let X and Y be normed spaces, and T be a linear operator on X

into Y . then the following conditions on T are equivalent.
(a) T is bounded
(b) T is continuous
(c) T is continuous at the origin.

Proof:
let T be a bounded linear operator , on X we show a  b , b  c , and c  a .
(a b )
Let T be bounded , we show it is continuous .
Let x 0  D T  we show T is continuous at x 0 .
Let x  x 0   , then ,
Tx Tx 0  T  x  x 0  as T is linear
T x x0 . asT is bounded
Choose    T ,
we see whenever x  x 0   , then Tx Tx 0   ,  T is continuous.

69
b  c  is trivial . If T is continuous then T is continuous at every point, hence at the
origin .
c  a 
Let T be continuous at the origin, then we show T is bounded .
Since T is continuous at the origin then , x n  0 Tx n T 0  0 ,
( T is linear T  0x   0Tx  0 ).
Assume T is not bounded .
Hence for each positive integer n , we can find a vector x n  D T  s.t Tx n  n x n
1 Tx n
 Tx n  1,   1  x   x 
n xn n xn
 x 
 T n   1 ( T is a linear operator).
n x
 n 
xn
Let y n  ,  T y n  1 .
n xn
xn 1
Now, yn    0 as n   .
n xn n
 y n  0 as n   ,but Ty n  0 as Ty n  1 .
Hence T is not continuous at the origin , which is a contradiction to our assumption, so
that T must be bounded. #

Remarks:
1- according to above Theorem T is bounded  T is continuous, so these two
adjectives can be used interchangeably.
2- A above Theorem shows that if T is linear and T is continuous at the origin then T
is continuous which is not true if T is not linear.

Theorem30: Let T be a bounded linear operator, then The Null space N T 

is closed.

Proof:
Let x  N T  .
  a sequence  x n  in N T  s.t x n  x .
Since T is bounded linear operator and thus continuous,  Tx n Tx
Now ,  x n   N T   Tx n  0 , Tx  0 , as Tx n  0 ,  x  N T  .
 N T   N T  also N T   N T  ,  N T   N T 

71
 N T  is closed. #
Q.6/P101

Show that the rang R T  of a bounded linear operator T : X Y need not to be closed
in Y .

Proof:
Let T : L  L defined by
x x x 
T  x 1 , x 2 ,..., x i ,...   1 , 2 ,..., i ,...  , then
 1 2 i 
 x x x  x x x 
R T    1 , 2 ,..., i ,...   L T  x 1 , x 2 ,..., x i ,...   1 , 2 ,..., i ,...   .
 1 2 i   1 2 i 
We show R T  is not closed in Y .
T is bounded and linear (see Q5 P 101 )
Let  z k  , k  1 be a sequence in L defined by
z k  L , and Tz k  1,1,1...,1,
 z k   1, 2,..., k , k , k ,... then  k k
, ,...  .
 k 1 k  2 
We observe as k   then , Tz k  1,1,1,1,.... 
Now 1,1,1,1,....  R T  , because, 1,1,1,1,....  Tz ,  z  1, 2,3,...  L .
Thus there is a sequence Tz k  with Tz k  z , but z  R T   R T  is not closed .

Q.8/P101

Show that the inverse T 1 : R T   X of a bounded linear operator T : X Y need not

to be bounded.

Proof:
We take the same operator in above question.
We see that Tx  0,   1 , 2 ,..., i ,...   0 ,   x 1 , x 2 ,..., x i ,...  0
x x x
 1 2 i 
1 1
 T exists.  T y  x  y  Tx .

Now,  z k   1, 2,..., k , k , k ,... , Tz k  1,1,1...,1, 

k k
, ,...   y k .
 k 1 k  2 
1
T y k  z k .
 T 1 y k  z k  1, 2,..., k , k , k ,...  k ,
and,

70
  k k 
y k  1,1,1...,1, , ,...   1 .
 k 1 k  2 
Thus if T 1 y k  c y k , then k  c , that is there exists no fixed number c such that
T 1 y k  c y k , y k  L .
1
T is not bounded.

Linear functional:

Functional:
A functional is an operator whose range lies on the real line or the complex plane we
denote functional by lower case letters f , g , h ,... .
Thus a functional f is an operator f : D  f   X  K ,where X is a normed space over a
field K and K  if X is real and K  c if X is complex.
As functional are operators so the previous definitions apply, we shall need in particular
the following definitions.

Linear and bounded functional:

A functional f : D  f   K is said to be linear on D  f  if,
f  x   y    f  x    f  y  x , y  D  f  ,  ,   K .
And said to be bounded if there exists a positive Real number c such that f  x   c x .
Furthermore, the norm of f is given by
f x 
f  sup , (1)
x D  f  x
x 0
or
f  sup f  x  (2)
x 1

f x 
And f  c  sup is the smallest value of c such that
x D  f  x
x 0

f x   f x . (3)
All the results and Theorems on linear bounded operators remain also valid for
functional,
Thus if f : D  f   K then
(i) f is continuous iff it is bounded.
(ii) f is continuous at the origin iff it is continuous at each point of D  f  .

72
Examples:

(1) The norm . : X  … of a normed space  X , .  is a functional on X which is not

linear, as  x   y   x   y x , y  X ,  ,   .
(2) Dot product:
Let f : 3  defined by
f  x   x .a  x 1a1  x 2a2  x 3a3 , x  (x 1 , x 2 , x 3 )  3
, and a  a1 , a2 , a3  3
is fixed .
3
Then f is linear and bounded functional on
In fact, f is linear since,
f  x   y    x   y  .a   x .a   y .a
   x .a     y .a 
  f  x    f  y  , x , y  3
,  ,   ..
Now f  x   x .a  x a .
 f is bounded
We show norm of f , f  a
From (2) we have,
f  sup f  x   sup x a  a
x D  f  x D  f 
x 1 x 1

 f  a , on the other hand by taking x  a using (3) we have

2
f (x ) f (a ) a .a a
f      a
x a a a
ie f  a hence f  a
ie c  a is the smallest values of c such that f (x )  c x .

(3) Definite integrals:

Let f : c a, b   Defined by
b
f  x    x t  dt , x  c a , b  .
a

Then f is linear and bounded functional on c a, b  , the space of all continuous
functional on c a, b  .

73
In fact, f is linear, let x , y c a,b  ,  ,   , then
b
f  x   y     x   y t  dt
a
b b
   x t  dt    y t  dt
a a

  f  x    f  y .
Now we show f is bounded,
We know from the first mean value Theorem that if x is continuous functional on a, b  ,
then there exists c  a, b  s.t,
b
f (x )   x t  dt  x t b  a   b  a  max x t  ,
t a ,b 
a
b
 f x    x t  dt  b  a  max x t   b  a  x ,
t a ,b 
a

 f is bounded.
We show f  b  a
From (2) , f  sup f  x   b  a  sup x  b  a 
x 1 x 1

 f  b  a  ,
on the other hand for x  x 0  1 , we have form (3)
f x 0  b
f   f  x 0    1 dt  b  a 
x0 a

ie f  b  a   f  b  a  .

(4) Space c a, b  .

Another functional defined on c a, b  is obtained if we choose a fixed t 0  a, b  and
f : c a, b   defined by ,
f  x   x t 0  , x c a,b  .
Then f is linear and bounded
In fact, f is linear
f  x   y    x   y  t 0    x  t 0    y  t 0 
  f  x    f  y  , x , y c a,b  and  ,   .

Also, f  x   x t 0   tmax x t   x
 a ,b  
 f x   x .
74
 f is bounded , we show f  1.
Form (2) we have
f  sup f  x   sup x 1
x D  f  x 1
x 1

f x 
 f  1 , also from (3) f  , x  D  f 
x
In particular for x  x 0  1
f x 0 
 f   f  x 0   x 0 t 0   1 .
x0
 f  1.

(5) Space l 2
Let f : L2  defined by

f  x    x i ai , where x   x i   l 2 , and a  ai   l 2 is fixed.
i 1

f is linear and bounded .

in fact, f is linear , as

f  x   y     x i   y i  ai
i 1
 
   x i ai    y i ai
i 1 i 1

  f  x    f  y  x , y  l 2 and  ,   .
Also f is bounded since
 
f x    x i ai   x i ai
i 1 i 1
1 1
 2

 2

   x i     ai 
2 2
By Cauchy-Schwarz inequality
 i 1   i 1 
 x a .
 f is bounded, if is easy to show that f a .

6) define a linear functional f : L   by

 
f  x    x i , x   x i   L , ie x i   , then the functional f is bounded with the
i 1 i 1

norm x 1   x i , but it is not bounded with the norm x 
 sup x i , . x  L 
i 1 1i 

75
Solution:

Let x  L  and x 1   x i , then
i 1
 
f x   x
i 1
i   xi  x
i 1
1 , x  L 

 f x   x 1 , x  L  is bounded on L  and f  1 .
On the other hand f is unbounded on L  with .  .
Let x  e1  1,0,... then ,
clearly x  L  and x 
 sup x i  1 .
1i 

and f  x    x i  1  f  x   x   1.
i 1

Let x  e1  e 2  1,0,0,...   0,1,0,...  1,1,0,...

Then, x 
 sup x i  1
1i 

And f  x    x i  2  f  x   2 x  .
i 1

If we choose x  e1  e 2  e 3 then we obtain

f x   3 x  ,
if x  e1  e 2  e3  ...  e n , then f  x   n x  .
  no fixed n such that f  x   n x  , x  L  .
 L  is unbounded on L  .

Algebraic dual space X  .

Let X be a normed space, then the set of all linear functional defined on X is called the
algebraic dual space of X , denoted by X  , ie
X   f : X  K and f is linear .

Remark:

The set X  is a vector space, with the algebraic operations defined as follows:
If f 1 , f 2  X  , ie f 1 , f 2 are linear functional on X , then the sum f 1  f 2 is defined by
 f 1  f 2  x   f 1  x   f 2  x  , x  X ,
also , if   K and f  X  , then the Scalar product is defined by
  f  x   f  x  , x  X
With these two operations , we can show that X  is a vector space, ie
76
f 1  f 2  X  and  f  X  .
In fact.
 f 1  f 2  x 1   x 2   f 1  x 1   x 2   f 2  x 1   x 2 
  f 1  x 1    f 1  x 2     f 2  x 1    f 2  x 2   , as f 1 and f 2 are linear
  f 1  x 1   f 2  x 1    f 1  x 2   f 2 x 2 
   f 1  f 2  x 1     f 1  f 2  x 2 
 f 1 f 2X ,
Similarly
  f 1  x 1   x 2     f 1  x 1      f 1  x 2 
ie f 1 X  .

Second algebraic dual space X ** :

The set of all linear functional on X  is called the second duals space of X denoted by
X  ,
ie X    g : X   and g is linear functional .

Thus g  X means g is a linear functional defined on X .

Theorem30: (dimension of X  )
Let X be an n -dimensional vector space, and B  e1 ,..., e n  be a basis for X
then , B   f 1 ,..., f n  is a basis for the algebraic dual X  of X , where
0 if i  j
f i e j    i j   (  i j is the Kronecker delta).
1 if i  j
 
Consequently, dim X  dim X  dim X  n .
The basis B  is called the dual basis of B .

Proof:
First we show B  is linearly independent.
Let B1f 1  B 2f 2  ...  B n f n  0 (zero vector of X  )
  B1f 1  B 2f 2  ...  B n f n  x   0  x   0, x  X ,  0  x   0  .
With x  e j  X we have ,
B1f 1 e j   B 2f 2 e j   ...  B n f n e j   0 j  1, 2,..., n

  B i f i e j   0,   B i S i j  0 j  1, 2,..., n
n n

i 1 i 1

 B j  0, j  1, 2,..., n

77
 f 1 , f 2 ,..., f n are linearly independent.
Now we show B  spans X  , ie every f  X 
Can be expressed uniquely as a linear combination of f 1 , f 2 ,..., f n
Let x  X , as e1 ,..., e n  is a basis of X
  Scalars 1 ,  2 ,...,  n s.t
n
x   j e j .
j 1

 n  n
 f  x   f   j e j     j f e j  as f is linear .
 j 1  j 1
Let f e j    j
n
 f x     j j .
j 1

On the other hand

x  1e1   2e 2  ...   ne n
 f j  x   1f j e1    2f j e 2   ...   n f j e n    j , j  1, 2,..., n

where  j  f e j 
n
 f x     j f j x  ,
j 1

 f  1 f 1  2 f 2  ...  n f n .
 B  spans X  , and hence B  is a basis of X  .
dim X   number of distinct elements of B   n .
Now as X   is the dual of X  , and X  is n-finite dimensional , then by this theorem
dim X   dim X    dim X  n .#

Corollary: Let X be a finite dimensional space , if f  x 0   0 for x 0  X , and all f  X  ,

then x 0  0 .

Proof:
Let e1 ,...,e n  be a basis of X , then by Theorem 30, f 1 ,..., f n  is a basis for dual X  .
Let, x 0  X then , x 0  1e1  e 22  ...  e n n .
n
 f  x 0   1f e1    2f e 2   ...   n f e n   i f e i , f  X 

i 1
n
   i f e i   0 f  X 

i 1
n

 f e   0 ,

 i j i j  1, 2,..., n , as f j  X .
i 1

 i  0, i  1, 2,..., n , as f j ei    i j

78
 x 0  0. #

In previous chapter we have define the isomorphism T of a metric space X onto a

metric space X as bijective mapping which preserve distances,
d Tx ,Ty   d  x , y  x , y  X .
An isomorphism T of a vector space X onto a vector space X over the same field, is a
bijective mapping which preserves the two algebraic operations of vector space.
ie x , y  X , and Scalars  ,
T  x  y   Tx Ty , and T  x   Tx .
ie T : X  X is a bijective linear operator, X is called isomorphic with X , and X and X
are called isomorphic vector spaces.
An isomorphism of a normed space X onto a normed space X is a bijective linear
operator T : X  X which preserves the norm that is , for all x  X , Tx  x . X is
then called isomorphic with X .

Algebraically reflexive:
A vector space X is said to be a algebraically reflexive of X   , if there exist an
isomorphism from X onto X   .

Theorem31: Every finite dimensional vector space is algebraically reflexive.

Proof:
Define a mapping C : X  X   , by C  x   g x , x  X , where
g x  X ** , ie g x : X *  is a linear functional defined by g x  f   f  x  , f  X  .
Such mapping is called the canonical mapping.
To Show that X is algebraically reflexive, We show C is an isomorphism.
1) C is Linear operator from X into X   :
We show C  x   y   C (x )  C ( y ) . For  ,   K and x , y  X .
Now

g  x   y  f   f  x   y 
 f x   f  y  , as f is linear.
  g x f    g y f 
  g x  f     g y   f 
  g x   g y   f  .
 g x  y   g x   g y
 C  x   y    g x   g y  C  x   C  y  .

79
2) C is one-one.
Let x , y  X , and
C x   C  y ,
 gx  g y
 g x f   g y  f  f  X 
f  x   f  y  f  X 
f  x   f  y   0 f  X 
f  x  y   0 f  X  as f is linear
 x  y  0, by the corollary of Theorem 30.
x  y
 C is one to one.

3) C is onto.
we show R C   X  
For x 0  X , let C  x 0   0 , then by definition of C we have,
C  x 0   g x 0  0  g x 0  f   f  x 0  f  X 

 x 0  0 , by the corollary of Theorem 30

ie C  x 0   0  x 0  0 , it follows that C has an inverse, C 1 : R C   X , where R C  is
the range of .
We know that if C 1 exists then , dim R C   dim X .
From Theorem 30 we have, dim X   dim X  dim X   .
 dim R C   dim X **.
But R C  is a vector space, and a proper subset of X   , which has a dimension less

than dim X .
 R C   X 
.
 X is algebraically reflexive. #

Dual space X  :

Let X be a normed space, then the set of all linear and bounded functional defied on X ,
is called the dual of X denoted by X  .
ie X   f : X  K where f is linear and bounded .

81
Remark:
In a finite dimensional normed space X , X   X .(why)


clearly X   X  , and we know that in a finite dimensional normed space X , every linear
operator is bounded ie X   X  consequently, X   X  .

Theorem32: The dual space X  of X is a normed space with the norm defined
by
f x 
f  sup  sup f  x  .
x X x x X
x 0 x 1

Proof:
First we show X  is a vector space,.
The sum and product of vectors in X  is defined as usual
 f 1  f 2  x   f 1  x   f 2  x  x  X , and  f  x    f  x  x  X ,   K .
We show f 1  f 2  X  and  f  X  , ie f 1  f 2 and  f are linear and bounded functional on
X .
Linearity was shown before, we show boundedness.
Let f 1 and f 2 are in X  ,
Then f 1 and f 2 are bounded on X , ie K 1  0, K 2  0 s.t ,
f 1  x   K 1 x and f 2  x   K 2 x , x  X .
Now,
f 1  f 2  x  f 1 x   f 2 x 
 f 1 x   f 2 x 
 K1 x  K 2 x  K1  K 2  x
 f 1  f 2  x  K1  K 2  x
,  1
K K2  0
 f 1  f 2 is bounded on X  f 1  f 2  X 
Similarly, we can show that  f is bounded.
If is easy to show that f satisfy the for axioms of norm,
and that X  is a normed space. #
The space B  X ,Y  :
Let X and Y be two normed spaces, we denote by B  X ,Y  , the set of all bounded ( or
continuous ) linear operators from X into Y , ie

80
B  X ,Y   T : X  R T  Y and T is bounded and linear .

Theorem33: The space B  X ,Y  of all bounded linear operators from a normed

space X into a normed space Y is itself a normed space with the norm defined
by ,
Tx
T  sup  sup Tx . (1)
x D  f  x x D  f 
x 0 x 1

Further if Y is a Banach space, then B  X ,Y  is a Banach space.

Proof :
First, B  X ,Y  is a vector space , the sum T1 T 2 and the scalar product T are defined
as, if T1 ,T 2  B  X ,Y  and   K  K  or  , then T1 T 2 x  T1x T 2x , and
T  x  Tx .
In fact,
Let T1 ,T 2  B  X ,Y  , then T 1 and T 2 are bounded linear operators,
 k 1  0 and k 2  0 such that
T1x  k 1 x and T 2 x  k 2 x , x  X .
Let  ,  are any two Scalars, then
T1  T 2  x  T1  x   T 2  x
 T1x  T 2 x
 T1x  T 2 x
  T1x   T 2 x
  k1 x   k 2 x
  k1   k 2  x
  T1   T 2 is bounded, also we have shown before  T1   T 2 is linear
  T1   T 2  B  X ,Y  .
 B  X ,Y  is closed under addition of vectors and scalars multiplication of vector,
consequently , B  X ,Y  is a vector space.
B  X ,Y  is a normed space as the norm defined by (1) satisfies the axioms of the norm
(already shown).
Now we assume Y is complete and we show B  X ,Y  is complete also.
Let T n  be a cauchy sequence in B  X ,Y  . Then
We show that T n  converges to an operator T  B  X ,Y  .

82
As T n  is a Cauchy sequence in B  X ,Y  , then for each   0, N  0 , s.t
T n T m   n , m  N .
For each x  X , we have
T n x T m x  T n T m  x
 T n T m x as T n T m  are bounded
 x (2)
  for any fixed x and    x .
 T n x T m x    n , m  N
 T n x  is a Cauchy sequence in Y
SinceY is complete ,
 T n x  Converges say T n x Tx  y , Tx Y .
Now we show T n  T .
First we show T is linear and bounded.
If x , y  X , and  ,  are any Scalars , then
T  x   y   limT n  x   y 
n 

 lim T n x  T n y 
n 

  limT n x   limT n y
n  n 

 T n x  T n y
T is linear,
Now T n x Tx , n  N as m  N
 T m x Tx ie lim T m x  Tx
m 

Thus ,we have form (2)

T n x Tx  T n x  lim T m x
m 

 lim T n x  T m x , as . is continuous.
m 

  x n  N (3)
 T n T  x   x n  N ,   0.
 T n T  is bounded with n  N
 T  T n  T n T  is bounded .
 T  B  X ,Y  .
A gain form (3) we have
sup T n x Tx  sup  x , n N
x D  f  x D  f 
x 1 x 1

83
 T n T   n  N .
 T n  T in B  X ,Y 
 T n  is convergent sequence in B  X ,Y  .
 B  X ,Y  is a Banach space. #

Remark:
B  X ,Y  is the set of all bounded linear operators form X intoY , if Y  K  K  or  ,
then B  X ,Y   X  , the set of all bounded linear functional on X . Form the above
Theorem B  X ,Y  is a Banach space ,if Y is a Banach space, and if Y  K then clearly
K  or are Banach spaces and consequently X  is a Banach space, hence we have
the following Theorem.

Theorem34: The dual space X  of a normed space X is a Banach space

(whether or not X is).

Now we consider some spaces and find out what their duals look like.

Examples:
n n n n
(1) Dual space of is isomorphic with or briefly we say the dual space of is

Proof:
We know that n is a finite dimensional normed space, hence   the algebraic
n n

dual of n is identical with the dual space of n .

n
Now we show n  is isomorphic with n ie the mapping T :  n is an
isomorphism.
Let e1 , e 2 ,..., e n  be a basis for n , then every x   x 1 , x 2 ,..., x n   n can be expressed
n
uniquely in the form x   x i e i .
i 1
n
n
If f  ie f : n
 is linear, then f  x    x i f e i  .
i 1
n
Now, Define a mapping T :  n
by,
Tf   f e1  ,..., f e n   , f  n
.
We show T is an isomorphism.
Clearly, T is linear and bijective. Also by Cauchy Schwarz inequality we have
n
f  x    x i f e i 
i 1

84
 1 1
 n   n 2
   f e i   
2 2
  x i 
2

 i 1   i 1 
 x Tf .
Taking the supremum over all x of norm 1 ,
we obtain f  T f we show now f  T f as f is bounded linear functional.
 f x   f x x  n
.
In particular choose x  x 0   f e1  ,..., f e n   .
1
 n 2
    f e i    , and
2
Then x 0
 i 1 
n n
f  x 0    f e i  f e i     f e i  
2

i 1 i 1
n

  f e  
2

f x 0 
1
i
 n 2
    f e i   
2
 f   i 1
1
 Tf ,
x0  n 2  i 1 
   f e i   
2

 i 1 
 f  Tf ,
n
 f  Tf f 
n
 is an isomorphic with n
.

(2) The dual space of n is n .

Proof is similar as example (1).

(3) The dual space of L1 is L .

Proof:
we know that e i  , where e i   i j  is ashauder basis for L p and thus for L1 , where
0 if i  j
Si j   .
1 if i  j

Let x   x 1 , x 2 ,..., x i ,...  L1 then x   x i e i ,
i 1

and let f  L1 (dual space of L1 ) , where f : L 

1
is bounded and linear functional.

 f  x    x i f e i  .
i 1

Define a mapping T : L1  L by

Tf   f e1  ,..., f e i    y .

85
First we show that the sequence y  L
As f is bounded
 f e i   f e i 1

Where e i  L  for each i  , e i 1   S i j  1
j 1

(Note , if x   x i   L  , then x 1   x i )
i 1

 f e i   f 
 sup f e i   f  
i

 the sequence y   f e1  ,..., f e i    L ,

and Tf  y 
 sup f e i   f . (1)
i

{ Note , if x   x i   L , then x 
 sup x i }.
i N

Now we show T is an isomorphism from L1 onto L .

Clearly T is 1  1 and onto, we show Tf   f ie y   f .

We have, f  x    x i f e i 
i 1

  
  
 

 f x    x i f e i    sup f e j    xi 
  sup f e j  x
  j 
1
i 1  j   i 1
Taking the supremum over all x of norm 1 , we obtain.
f  sup f e j   y 
j

 f  y , 
(2)
and hence from (1) and (2) we have f  y  .
 L1  L .

1 1
(4) The dual space of L p , 1  p   ,is Lq , 1  q   and  1
p q

Proof:
Proceed as example (3) , e j  is a scheduler basis for L p , and f  L p1 , f : L p 

Let x   x 1 , x 2 ,..., x i ,...  L then x   x i e i ,
p

i 1

 f  x    x i f e i  .
i 1

Define a mapping T : L p   Lq by Tf   f e1  ,..., f e i  ,...  y .

86
 1
  q 
q
We show y  L , ie   f e i   .
q

 i 1 
Consider x   x i   L p , with
 f e  q
 , f e i   0
i

x i   f e i  .

0 , f e i   0
Clearly, x i  so x   x i   L p
 
f  x    x i f e i  =  f e 
q
i .
i 1 i 1

  f e i   f  x   f  x 
q

i 1

f x p
1
 
 p
f  xi
p
 
 i 1 
1
   f e  q 
p
 p

 
f   
i
 i 1  f e i   
   
 
1
 
q 1 p  p
 f 

 f ei 
i 1


1
 
 p

 f e i  1 1
q
f   , q  1 p  q as   1.
 i 1  p q


 f e 
q
1
i
 
 q

 f e 
q
 f  i 1
1
 i 
 
 p  i 1 
 f e 
q
 i 
 i 1 
1
 
 q

 f e 
q
  i   y  f  (1)
 
q
i 1

 y  Lp .
Now we show T is an isomorphism,
Clearly T is 1  1 and onto , we show Tf  f
ie y q
 f

f  x    x i f e i 
i 1

87
 1 1
   p
  q 
q
  xi   f e i  
p
 by holder inequality
 i 1   i 1 
 x p
y q
.
Taking the supremum over all x of norm 1 ,
we get
f  y q
(2)
Form (1) and (2) we obtain
f  y q
 Tf q

 T is an isomorphism of L p  onto Lq and L p   Lq .

Fundamental Theorems for normed space :

(1) The Hahn-Banach Theorem:

The Hahn-Banach Theorem is an extension Theorem for linear functional defined on a
subspace of a linear functionals on the whole space.
Note here that a functional f on a space X which is linear on a subspace Y may not e
linear on the whole space.
For example, let f : 2  defined by f  x , y   x  y 2 is not linear on 2 , while it
is linear on the subspace M   x , y   2 / y  0  2 , thus we speak of extension of a
linear functional f on a subspace M to a linear functional on the whole space X .

Extension of function:
Let M be a subspace of a normed space X and f be a linear functional on M . then
by extension of f to the whole space X , we mean a linear functional F on X , such
that F  x   f  x  , x  M ie F / M  f .
If F is an extension of f , then f is said to be restriction of F .
For example, F : 2  defined by F  x , y   x  y 2 , is an extension of f : M 
defined by f  x , y   x , where , M   x , y   2
/ y  0 .
Clearly, F  x , y   f  x , y  ,   x , y   M ie y  0 .

Sublinear functional :
Let X be a vector space over a field K ,  K  or  a mapping P : X  is called
a sublinear functional on X if
(i) P  x  y   P  x   P  y  x , y  X (subadditive )
(ii) P  x    P  x  , x  X and   K .
Example:

88
If X is a normed space, then the norm is a sublinear functional, ie . : X  satisfies
(i) x  y  x  y , x  X and
(ii)  x   x ,  

Lemma (Hahn-Banach Lemma:)

Let X be a vector space over a field K , and P be a sublinear functional on X .let f
be a linear functional on a subspace M of X , such that f  x   P  x  , x  M .
Then there exists a linear functional F on X s.t F  x   f  x  , x  M ie F is an
extension of f and F  x   P  x  , x  X .
(see the proof of this Lemma page 219 in Kreyszing book)

Theorem35: (Hahn-Banach Theorem for normed spaces):

Let X be a normed space over a field K and M be a subspace of X . Then for
every bounded linear functional f on M , there exists a bounded linear
functional F on X s.t F  x   f  x  , x  M ( ie F is an extension of f )
and F  f , where
f  sup f x  and F X
 sup F  x 
M
x M x M
x 1 x 1

Proof:
Let f be bounded linear functional f on M , hence
f x   f x x  M .
Define P  x   f x , x  X ,
then P is a sublinear on X ,
Indeed , P  x  y   f x  y
 f x  y 
 f x  f y
 P x   P  y  . x , y  X
and P  x   f  x   f x   P  x  , x  X ,  scalars  .
 P is a sublinear functional on X and
f  x   f x  P  x  , x  M .
Hence, by Hahn-Banach Lemma, if follows that there exists a linear functional F on
X which is an extension of f , ie
F  x   f  x  , x  M and satisfies ..

89
 f x   P x   f x x  X (1)
Taking the supremum over all x  X of norm 1, we obtain form (1)
F  f . (2)
As f is bounded  F is also bonded.
 F x   F x
F x 
 F  x  X
x
In particular, let x  x 1  M , x 1  0
F x 1  f x 1 
 F    f (3)
x1 x1
From (2) and (3) we have F  f .#
The Hahn-Banach Theorem provides guarantee that the dual space X  of any normed
space X consists of sufficiently many bounded linear functional , this follows from the
following Theorem.

Theorem36: Let X be a normed space, and x 0  0 , be any element in X , then

there exists an F  X  ie F is a bounded linear functional on F s.t F  1 and
F x 0   x 0 .

Proof:
Consider the set M  x : x   x 0 ,   K  .
Then , clearly M is a subspace of X
Define f : M  K by
f  x   f  x 0    x 0 ,   K
Clearly, f is linear functional on M further for x  M , we have ,
f  x   f  x 0    x 0   x 0  x

 f is bounded and f 1 .
 by the Hahn-Banach Theorem,  a bounded linear functional F on X s.t
f  x   F  x  x  M and f  F .
Hence F  x 0   f  x 0   f 1x 0  1 x 0  x 0 .
And F  f  1 . #

91
(2) Open mapping Theorem:

Open mapping:
Let X and Y be two normed spaces a mapping T : X Y is said to be open mapping if
T maps open sets to open set, ie for every open set G in X , the image T G  is an
open in Y .

Remark:
We should note the difference between open mapping and the continuous mapping T
which has the property that for every open set in Y the inverse image is an open set in
X .
Also, we remark that open mapping is not necessary imply continuous mapping also
continuous mapping is not necessary imply open mapping .
For example, the mapping T :  defined by Tx  sin x is continuous but not open
mapping as T maps  0, 2  onto  1,1 ,  0, 2  is open interval hence open set in Y ,
while  1,1 is closed in Y .

Lemma:
Let X and Y be Banach spaces, and T be a bounded (continuous ) linear operator
from X onto Y . Then the image of each open ball centered at the origin in X contains
an open ball centered at the origin in Y .
ie if S  0, r  is an open ball in X , then T S  0, r    S   0, r  , where S   0, r  is an open
ball centered at the origin in Y . (see the proof page 286 from Kreysing book).

Theorem37: (The open mapping Theorem)

Let X and Y be Banach spaces, and T be a bounded (continuous) linear
operator from X onto Y . Then T is an open mapping.

Proof:
Let T : X Y be a bounded linear operator and let G be an open set in X , we show
T G  is an open set in Y .
Let y T G  , we show y is the centered of an open ball contained in T G  .
As y T G  ,  y  Tx for some x G  X since G is an open set in X , so there
exists an open ball B  x , r  in X centered at x such that B  x , r   G .
Now it is clear that B  x , r   S  0, r   x , in fact
z  S  0, r   x  z  x  S  0, r   z  x  r  z  B  x , r  .
 B  x , r   S  0, r   x , where S  0, r  is an open ball in X centered at the origin , by
above lemma, there exists an open ball S   0, r   centered at the origin such that
90
 S   0, r   T S  0, r  
 y  S   0, r   y T S  0, r  
 T  x  T S  0, r  
 T  x  S  0, r   , as T is linear
But y  S   0, r   B   y , r   .
 B   y , r   T  x  S  0, r    T  B  x , r    T G  .
 y  B   y , r   T G  .
 T G  is an open set.
 T is an open mapping. #
Most of the application of the open mapping Theorem depends on the following
corollary.

Corollary:
Let X and Y be two Banach spaces, and T : X Y be a bounded linear operator of
X onto Y , then if T is 1  1 then T 1 is a bounded (continuous) linear operator form
Y onto X .

Proof:
Since T : X Y is bijective,  T 1 exists and T 1 :Y  X is linear (shown before) , we
show T 1 is continuous.
Now by the open mapping Theorem .
If G is an open set in X then T G  is an open in Y .
Thus for each open set G in X , T 1  G   T G  is open in Y
1

1
T is continuous ie bounded
1
T is linear and bounded.

Remark:
We know that if T is bijective mapping , and T and T 1 are continuous then T is a
homeomorphism thus if T : X Y be a bounded linear operator of a Banach space X
onto a Banach space Y , then T is a homeomorphism , this is clear from the above
Corollary.

Example:
Let X  c 0,1 with the norm f 
 sup f  x  .
0 x 1

92
and Y  c 0,1 , with the norm   f  x dx .
1
f 1 0

We have shown before that Y , c 0,1 is incomplete.

Let I : X Y be the identity operator show that I is not open mapping

Proof:
I is linear and bounded, but Y is not a Banach space, otherwise by the above Theorem
I is open mapping.
To show that I is not open mapping ,we show I 1 is not continuous.
The operator I is a bijective so we can consider its inverse mapping in I 1 :Y  X .
Let  x n  be a sequence in c 0,1 , defined by
 1
1  nt , 0 t 
x n t   
n
0 1
,  t 1
 n
Note that as n   , x n  0 in Y .
And I 1x n  xn 
 sup x n t   1
 0t 1

ie (maximum value of x n t  occurs at t  0 , which is 1 )

 I 1x n  1 in X
 I 1 is not continuous .
 I is not open mapping.

(3) Closed graph Theorem:

Let X and Y be two normed spaces, over a field K , then the Cartesian product.
X Y   x , y  / x  X and y Y  is also a normed space.
First of all X Y is a vector space, under the algebraic operations given by
 x 1, y 1    x 2 , y 2    x 1  x 2 , y 1  y 2  ,  x 1, y 1  ,  x 2 , y 2   X Y ,
and   x , y     x ,  y  ,  x , y   X Y ,   K .
And the norm . for X Y is given by
x , y   max x x
, y y 
 x , y   X Y .
It is easy to prove that the norm  x , y  satisfies the four axioms of norm more over if
X and Y are Banach spaces, then so is X Y .
Indeed, let  z n  with z n   x n , y n  be a Cauchy sequence in X Y ,
 z n  z m  0 as n , m  
but z n  z m   x n , y n    x m , y m    x n  x m , y n  y m  .

93
 z n  z m  max x n xm x
, yn ym y 0 as n, m   ,

 xn xm x
 0 and yn  ym y
 0 as n , m   ,
  x n  and  y n  are Cauchy sequence in Y ,
and hence x n  x in X and y n  y in Y ,
and hence  x n , y n    x , y  ie z n  z in X Y ,where z   x , y   X Y .
 X Y is a Banach space.

Graph of T :
Let X and Y be two normed spaces, and T : X Y be a linear operator , the set given
by , G T    x ,Tx  : x  X  is called the graph of T . Clearly G T   X Y .
If  x , y  G T  , then x  X and Tx  y .

Closed linear operator:

Let X and Y be normed spaces, and let T : X Y be a linear operator, then T is
called a closed linear operator iff its graph G T  is closed subspace of the normed space
X Y .

Lemma:
Let X and Y be normed spaces, over the field K , then a linear operator T : X Y is
closed iff for every sequence x n   X with x n  x and Tx n  y , we have x  X and
Tx  y .

Theorem38: (Closed graph Theorem)

Let X and Y be two Banach spaces, and T be a linear operator of X into .
Then T is closed iff T is bounded.

Proof:
Let T : X Y be a bounded linear operator, we show T is closed.
Let  x n   X s.t x n  x and Tx n  y
as X is complete  x  X and as T is bounded and hence continuous
x n  x  Tx n Tx hence Tx  y
Thus x  X and Tx  y .  T is closed .
Conversely: Let T be closed, then the graph G T  of T is closed linear subspace of
the Banach space X Y , we know that a subspace of a Banach space is a Banach space ,
iff it is closed.
As T is closed .  G T  is closed and hence complete ie a Banach subspace of X Y .
94
Define a mapping f : G T   X by f   x ,Tx    x .
Clearly, f is linear operator, which is also one to one and onto we show it is bounded .
f  x ,Tx   x x
 max x x
, y y 
  x ,Tx 
 f is bounded with f  1 .
as f is bijective bounded linear operator from G T  onto X .
f 1
: X  G T  is defined by f 1
 x    x ,Tx  .
1
By a consequence of open mapping Theorem, the mapping f is continuous hence
bounded linear operator.
 f 1 is continuous.
We show finally T is bounded
Tx x
 max x x
, Tx y 
  x ,Tx 
 f 1
x 
1
 f 1
x x  X as f is bounded.
 T is bounded . This completes the proof of the Theorem.

Fixed point Theorems and their Applications:

Fixed point:
If T is an operator of a Banach space X onto itself, T : X  X , x  X is called a fixed
point of T if Tx  x .

Examples.
1) Let I x : X  X be the identity operator, then the fixed points of I are all values of X
.
2) If T : R  R , defined by Tx  x 2 , the fixed points of T are 0 and 1.
3) T : R 2  R defined by T  x 1 , x 2   x 1 , has infinitely many fixed points.
4) T : R  R defined by Tx  x  a has no fixed points.
x
5) T : R  R defined by Tx  has a unique fixed point that is zero.
2

Contraction mapping.

95
Let T : X  X be an operator of a normed space X into itself, then T is called a
contraction mapping if there exists a constant k , 0  k  1, such that
Tx Ty  k x  y , x , y  X .

Example.
x 1
Let X  x  X , x  1  R , and let T : X  X defined by Tx   , then that T is a
2 x
contraction mapping.
Proof:
x 1 y 1 x y x y 1 1 1 1 
Tx Ty         x y  x y   x y ,
2 x 2 y 2 2 xy 2 xy  2 xy 
1 1 
as x , y  1  0     1
 2 xy 
T is contraction.
1 1 3
The smallest value of k is , i.e.,  k  .
2 2 2
Remark: Every contraction mapping is uniformly continuous.
In fact if T is a contraction mapping, then Tx Ty  k x  y , whenever x  y   .
Choose   k  , then Tx Ty   .

Theorem39: (Banach contraction Theorem)

Every contraction mapping T defined on a Banach space X into itself has a
unique fixed point x  X . Moreover, if x 0 is any point in X and the sequence
x n  is defined by x 1  T  x 0  , x 2  T  x 1  , x 3  T  x 2  ,..., x n  T  x n 1  , ... then
kn
lim x n  x * and x n  x *  x1  x 0 .
n  1 k

Proof:
a) Existence of a fixed point.
Let x 0 be an arbitrary point in X , and we define ,
x 1  T  x 0  , x 2  T  x 1  , x 3  T  x 2  ,..., x n  T  x n 1  , ...
then ,
x 2  Tx 1  T Tx 0   T 2x 0
x 3  Tx 2  T Tx 1   T 3x 0

x n T n x 0.
If m  n , say, m  n  p , p  1, 2,3,...
Then
96
 np
x np  x n  T x 0 T n x 0

 T T n  p 1
x 0  T T n 1
x0

 k T n  p 1x 0 T n 1x 0 as T is contraction mapping.

Continuing this process n-1 times, we have
x n  p  x n  k n T p x 0  x 0 , for n  0,1, 2,3,... and all p .

However,
p 1 p 1 p 2 p 2
T p x 0  x 0  T p x 0 T x 0 T x 0 T x 0 T x 0  ... Tx 0 Tx 0  x 0
p 1 p 1 p 2
 T p x 0 T x0  T x 0 T x 0  ...  Tx 0  x 0
but T p x 0  T p 1x 1
p 1 p 1 p 2 p 2
 T px 0  x 0  T x 1 T x0  T x 1 T x 0  ...  x 1  x 0
Now,
Tx 1 Tx 0  k x 1  x 0
 T 2 x 1 T 2 x 0  T Tx 1  T Tx 0 
 k Tx 1 Tx 0 x 0 , x 1  X
 k 2 x1  x 0
p 1 p 2
 T px1  x 0  k x1  x 0  k x 1  x 0  ...  x 1  x 0
 x 1  x 0 k p 1
k p 2
 ...  1
 x 1  x 0 1  k  k 2  ...  k p 1
,
then,
x n  p  x n  k n x 1  x 0 1  k  k 2  ...  k p 1 
as n  , m  n  p   and so p  .
kn
 x np  x n  x1  x 0 , 0  k 1
1 k
0 as lim k n  0 , 0  k  1 n 

 x n  is cauchy sequence in a Banach space X . Hence x n  must be convergent, say,

lim x n  x * , since T is continuous.
n 

We have Tx *  T nlim

 n 

x n  limT  x n   lim x n 1 , as Tx n  x n 1
n 

Now, nlim x n  lim x n 1  x *

 n 

 Tx *  x *.  x * is a fixed point of T .
b) Uniqueness of fixed point of T .
97
Let x 1* be another fixed point of T . Then Tx 1*  x 1*. We also have Tx * Tx 1*  k x *  x 1* ,
as T is a contraction mapping .
 Tx * Tx 1*  x *  x 1*  k x *  x 1* , 0  k  1. This is possible only if x *  x 1*  0 i.e.,
x *  x 1*  0 or x *  x 1*
Thus T has unique fixed point.

c) Finally the relation.

kn
xn x *  x 1  x 0 is clearly satisfies as,
1 k
kn
x np  x n  x1  x 0 and nlim x n  x * , and lim x n  p  x * . as p   .
1 k  p 

kn
 xn x*  x1  x 0 .
1 k
Example:
x
Tx  has a unique fixed point as T is a contraction mapping.
2
x y 1 1
Tx Ty    x  y , T is a contraction mapping where, k  . Thus by
2 2 2 2
contraction mapping principle, T has a unique fixed point.

Theorem40: Let Y be a closed subset of a Banach space X , and T :Y Y be

a contraction mapping. Then T has a unique fixed point.

Proof:
The proof is clearly from the above Theorem as a subspace of a Banach space is itself
Banach space if and only if it is closed .

There are many application of Banach Contraction Principle, here we mention one of it.

Example: Let X  R , Banach space and x  x and a, b   R , let f : a, b   a, b  be a
differential function such that f   x   k  1. Find a solution of f  x   x .

Solution: By Lagrange's mean-value Theorem, for any x , y  a, b  ,

f x   f  y   f   z  x  y  , y z x .
Hence ,

98
f  x   f  y   f   z  x  y  k x  y , where 0  k  1.
Thus f is a contraction mapping on a, b  into itself. Since a, b  is closed subset of R ,
hence by above Theorem there exists a unique fixed point x *  a, b  , i.e., f  x *   x *.
Therefore, x * is a solution of the equation f  x   x .

Inner product space, Hilbert space

Inner product space, Definitions and Examples:

Introduction:
In the preceding chapters, we studied a norm space, which is a linear space with a metric
defined by a norm.
In a normed space we can find the distance between vectors as we can add vectors and
multiply vectors by Scalars, but we can't find the angle between two vectors, or the dot
product of tow vectors, or orthogonality.
In fact, this can be done by imposing the axioms of inner product (the generalization of
the usual dot product ) on the abstract space, and the resulting space is called inner
product space .
Thus an inner product space is a special type of normed space which have an additional
structure of an inner product.
The dot product of two vectors x   x 1 , x 2 , x 3  and y   y 1 , y 2 , y 3  in three dimensional
Euclidean space 3 is given by x .y  x 1 y 1  x 2 y 2  x 3 y 3 and if x and y are in 3
then x .y  x 1 y 1  x 2 y 2  x 3 y 3 where y denote the conjugate of y .
The dot product of .. and . in satisfy the following properties:
(i) x . x  0 , x  3 and x . x  0  x  0 .
(ii) x . y  y . x , x , y  3 .
(iii)  x  . y    y . x  , x , y  3 ,   K  or 
(iv)  x  y  . z  x . z  y . z , x , y , z  3 .
The dot product of x and y in 3 satisfies the above properties except (ii) which
becomes (ii) x . y  y . x .
99
The inner product is a generalization of dot product to any real or complex vector space
(not necessarily 3 or 3 ).

Inner product:
Let X be a vector space over the field K  or .
A mapping . , . : X  X  K is said to be inner product on X if it satisfies the
following conditions:
 Ip1  x , x  0 x  X and x , x  0  x  0 (positive definite)
 Ip2  x , y  y ,x , x , y  X
 Ip3   x , y   x , y , x , y  X ,   K .
 Ip4  x  y , z  x , z  y , z , x , y , z  X .
Inner product space, is a vector space X , along with an inner product . , . defined on
X .

Remarks:
(1) If X is a real space, then x , y  y , x .
(2)  Ip3  and  Ip 4  can be written as  x   y , z   x , z   y , z ,which shows that
the inner product is linear in the first argument.
(3) x ,  y   y , x   y , x   x , y , if X Is real then, x ,  y   x , y .
(4) x ,  y   z   x , y   x , z which shows that the inner product is conjugate
linear in the second argument.
(5) Remarks (4) and (2) mean that the inner product is sesquilinear ( 1 1 2 times linear).
This is because the conjugate linear is half linear.
(6) An inner product on X defines a norm on X given by x  x , x .
, then x , x  x . x  x 12  x 22  x
2
If X  2
, a metric on X is given by
d x , y   x  y  x  y ,x  y .
(7) The norm induced by an inner product space X satisfies the parallelogram equality
x y
2
 x y
2
2  x
2
 y
2
 , x , y  X
(The sum of squares of diagonals equals the sum of the squares of its sides)
We show the norm induced by an inner product satisfies this equality
x y  x  y,x  y
2

 x ,x  x , y  y ,x  y , y
 x  y  x , y  y ,x ,
2 2
(1)
which on changing y to  y we have
x y  x  y  x , y  y ,x
2 2 2
(2)

011
Adding (1) and (2) we get the parallelogram equality.

Examples:
(1) 2 is an inner product space with inner product defined by
x , y  x 1 y 1  x 2 y 2 , where x   x 1 , x 2   2 and y   y 1 , y 2   2
.
Here d  x , y   x  y  x 1  y 1   x 2  y 2  ,
2 2
x  y ,x  y 

and x  x 12  x 2 2 .
n
In general is an inner product space with inner product defined by
n
x , y  x 1 y 1  x 2 y 2  ...  x n y n   x i y i where x   x 1 , x 2 ,..., x n  and
i 1

y   y 1 , y 2 ,..., y n  .
To prove that x,y is inner product space , we show  Ip 4  and  Ip1  ,  Ip 2  , and  Ip3 
are obvious.
n
x  z , y   x i  z i  y i
i 1
n n
 x i y i  z i y i
i 1 i 1

 x,y  z,y .
Other inner product on 2 can be defined by
x , y  x 1 y 1  x 2 y 1  x 1 y 2  3x 2 y 2 .
This shows that we may have different inner product on a vector space.
n
(2) is an inner product space with inner product defined by.
n
x , y   x i y i , x   x 1 , x 2 ,..., x n   n
y   y 1 , y 2 ,..., y n   n
.
i 1

The space  n
, ., .  is a complex inner product space .
n
 Ip 4  x  y , z   x i  y i  z i
i 1
n n
 x i z i   y i z i
i 1 i 1

 x ,z  y ,z .
 Ip1  ,  Ip 2  and  Ip3  are easy to show, use the properties of conjugate
z 1  z 2  z 1  z 2 and z 1z 2  z 1 z 2 .
1 1
 n  2
 n  2
x  x i x i    xi
2
Note :  .
 i 1   i 1 
010
(3) The space c a, b  the set of all continuous real valued or complex valued functions
on a, b  for vectors f , g c a, b  , defined f , g   f t  g t  dt .
b

 
1

f t  f t  dt
b 2
The norm is f  ,   
a

  
1

f t 
b 2 2

a
dt .
The space c a, b  is not an inner product space with the inner product
f , g  sup f t  g t  (3)
a t b

To prove this space is not inner product space, we show that the norm given by
f   sup f t  can't obtained from The inner product defined by (3).
a t b

Suppose that this norm is obtained from inner product (3) , then it must satisfy the
parallelogram equality.
t a
Consider f , g c a, b  defined by f t   1 , g t   ,
b a
t a t a
then f t   g t   1  and f t   g t   1 
b a b a
f  g  sup f t   g t   2 , and
a t b

f  g  sup f t   g t   1
a t b

 f g  f g  4 1  5 ,
2 2

and f 1 , g 1 ,  2 f  2
 g
2
  2 1 1  4
  f g 2
 f g
2
2  f 2
 g
2
.
 the norm f
can't be obtained from an inner product, or the inner product (3)


doesn't defined the norm f  .

 c a, b  is not an inner product with the inner product (3) .

(4) The space L , with p  2 is not an inner product space.

p

We show the norm of L p with p  2 can't be obtained from an inner product.

We proof this norm doesn't satisfy the parallelogram equality.
The norm of L p is given by
1
  p 
p
x p    x i  , x  x i   Lp .
 i 1 
Let x  1,1,0,0,...  L and y  1, 1,0,0,...  L p .
p

012
 1 1
Then x p
2 p
, y p
2 p
, x y p
2 and x y p
 2.

 x y  x y 8.
2 2
P P

 
2
 4   2 p  .
1
2 x  y
2 2
p p
 
 x y
2
p
 x y
2
p
2 x  2
p
 y
2
p
 for p  2.

Hence . p
doesn't satisfy the parallelogram equality if p  2 .
But we see if p  2 then the parallelogram is satisfied.
 L2 is an inner product with inner product.


x ,y   xi yi , x   x i   L2 , y   y i   L2 . (4)
i 1
1 1
   2
   2
x  x ,x  x i x i    xi
2
Here  .
 i 1   i 1 
Note that the series on the right of (4) is convergent series , as
1 1

    
2
 2

   xi  yi
2 2
x i y i   , by Cauchy-Schwartz)
i 1  i 1   i 1 
  , as x , y  L2

Schwartz inequality: If x and y are two vectors in an inner product space then
x,y  x y .

Proof :
If y  0 , then x , y  x ,0  x ,0x  0 x , x  0 and the inequality is clear.
Let y  0 , then for any scalar  ,we have
0  x  y  x  y , x  y
2

 x , x   x , y   y , x   y , y
 x  x , y   y , x  y  .
2 2
 
y ,x
In particular, choose   2 .
y
Conversantly,
2
x,y 2
0 x 
2
2 , here we used x,y  x,y x,y
y
This verifies the inequality. #
013
 Theorem41: Every inner product in a normed space , but converse is not
necessary true.

Proof:
We know that every inner product defines a norm given by x  x , x . we show this
norm satisfies the 4 axioms of norm.
N 1  x  0 since x , x  0 .
N 2  x  0  x  0 , since x ,x  0  x  0 .
N 3  x   x ,  x   x , x  
2 2 2
x
 x   x
N 4  x y  x  y ,x  y
2

 x ,x  y , y  x , y  y ,x
 x  y  x,y  x, y
2 2

 x  y  2 Re x , y
2 2

 z z 
 x  y 2 x,y ,
2 2
 Re z  
 2 
 x  y 2 x
2 2
y
 x  y 
2
(Schwartz inequality)
 x y  x  y .
For the converse, L p with p  2 is a normed space which is not inner product space.
 inner product space  normed space  metric space. #
Hilbert space:
An inner product space  X , .,,  over the field K is said to be a Hilbert space , if
 X , .,,  is complete with respect to the norm . . given by x  x ,x
.
The Hilbert space is called a real or complex Hilbert space according as the field
K  or . Clearly a Hilbert space is a Banach space.

Examples:
n
(1) n
is a Hilbert space with inner product defined by x , y   x i y i .
i 1

x   x 1 , x 2 ,..., x n   n
, y   y 1 , y 2 ,..., y n   n
.
1
 n 2
2
  xi  .
1
In fact, the induced norm is given by x  x , x 2

 i 1 
014
 n
We have shown before that is a complete with respect to the above norm .
n
(2) n
a Hilbert space with inner product x , y   x i y i
i 1

x   x 1 , x 2 ,..., x n   n
y   y 1 , y 2 ,..., y n   n
, .

2
(3) L is a Hilbert space with the inner product x , y  x i y i ,
i 1

x  x i   L , y   y i   L .
2 2

(4) The space L p , 1  p   , p  2 is not a Hilbert space.

Q.9+10/P135

Polarization identity:
(i) For real inner product space X we have x , y 
1
4
x y
2
 x y
2
.
(ii) For complex inner product space we have
Re x , y 
1
4
x y  x y
2 2
 
i m x,y 
1
4
x  iy  x  iy
2 2
 
Proof :
x  y  x  y ,x  y
2
(i)
 x  x,y  x,y  y
2 2

(1)
Similarly,
x y  x  y ,x  y
2

 x  x,y  x,y  y
2 2
(2)
Subtracting (2) form (1) , we get,
x  y  x  y  2 x , y  2 y ,x .
2 2
(3)
If the space X is real then x , y  y ,x
 x y  x y 4 x,y
2 2

x  iy  i x  iy , x  iy
2
(ii) i
 i  x , x  i x , y  i y , x  i i y , y  i  i 
 i x ,x  x , y  y ,x i y , y ,
also

015
 i x  iy  i  x , x  i x , y  i y , x  i i y , y 
2

 i x , x  x , y  y , x  i y , y .
Thus
i  x  iy 2
 x  iy
2
2 x,y 2 y ,x . (4)
Adding (3) and (4) we get.
 x y 2
 x y
2
  i  x  iy 2
 x  iy
2
 4 x ,y .

Theorem42: Let X be an inner product space. if x n  x and y n  y in X ,

then x n , y n  x , y .

Proof:
xn,yn  x ,y  xn,yn  xn,y  xn,y  x ,y
 xn,yn  y  xn x ,y
 xn,yn  y  xn x ,y
 xn yn  y  xn x y ( Schwartz inequality )
 0 as n   . #
Orthogonality of vectors:
Let H be a Hilbert space , a vector x  H , is said to be orthogonal to a vector y  H , if
x , y  0 , such vectors x and y are called orthogonal vector , written x  y .

Example:
The vectors y   0,1 and x   2,0  in 2 are orthogonal as, x , y  0.2  1.0  0 .
Here y   0,1 lies on the y-axis and x   2, 0  lies on the x- axis which are
perpendicular axes.

Q.2/P135

Theorem43: If in an inner product space X , x  y then x  y  x  y .

2 2 2

Proof:
x y  x  y ,x  y
2

‫ـ‬ ‫ــــــــ‬
 x  x,y  x,y  y
2 2

016
  x  2Re x , y  y
2 2

 x 0 y
2 2

 x  y
2 2
#
n 2 n

x   xi
2
More generally if x 1 ,..., x n are pairwise orthogonal vectors in X , then i .
i 1 i 1

Remarks:
(i) If x  y then y  x
x y  x,y 0  x,y  0
 y ,x  0 y x
ie The relation of orthogonality in H is symmetric.
(ii) x  y   x  y ,
 x , y   x , y   x .y   .0  0
 x  y .
(iii) 0, x  0 therefore 0  x , x  H .
ie the zero vector is orthogonal to every vector .
(iv) x  x ,  x , x  0 ,  x  0  x  0 .
2

Hence the zero vector is the only vector which is orthogonal to itself .

Orthogonal subsets:
 A vector x is said to be orthogonal to a non-empty subset S of a Hilbert space
(written x  S ) if x  y for every y  S .
 Two non-empty subsets S 1 and S 2 of a Hilbert space H are said to be orthogonal
(written S 1  S 2 ) if x  y , and x  S 1 , y  S 2 .

Examples:
1) x  1, 0  in 2 is orthogonal to y-axis . ie to M   0, y  / y  .
2) x-axis is orthogonal to y- axis.

Orthogonal complements:
Let S be a non-empty subset of a Hilbert space H , Then,
 Orthogonal complement of S is denoted by S  ( S orthogonal) and is defined by ,
S   x  H / x  S   x  H / x , y  0 y  S  .
 Orthogonal complement of S  is denoted by S   (S  ) , and is defined by ,
S   x  H / x , y  0 , y  S   .

017
Examples:
(1) If S   x , y   2
/ y  0  , then S

  x , y   2
/ x  0  , and S   S .
 1 
(2) Let M   x , y   2
/ y  2x  , then M 
  x , y   2 / y  x 
 2 
If  x , y   M ,   x , y  .  x , y   0,   x , y  M ,   x , y  .  x , 2x   0 ,  xx   y 2x   0


1
 y  x.
2

Theorem44: Let S be anon-empty subset of a Hilbert space H , then S  is

a closed subspace of H .

Proof:
S   x  H / x , y  0 , y  S 
Since 0, y  0 y  S , therefore 0  S  and so S  is non-empty.
Let x 1 , x 2  S  ,  ,   K then , x 1 , y  x 2 , y  0 , y  S .
We show  x 1   x 2 , y  0 .
Now,
 x 1   x 2 , y   x 1, y   x 2 , y
  x 1, y   x 2 , y
  .0   .0  0 .
  x 1   x 2   S  .
 S  is a subspace of H .

We show S is closed.
Let x  S 

  a sequence  x n  of points of S  s.t x n  x , we show x  S .
Let y  S , as x n  S 
 x n , y  0 , y  S
 lim x n , y  0 , y  S
n 

 lim x n , y  0 [as inner product is continuous function]

n 

 x , y  0 , y  S
 x S  .
 S  is closed subspace of H . #
Remark:
As S  is a closed subspace of H , H is a Hilbert space and hence complete.
 S  is complete , since each closed subspace of a complete space is complete.
018
 S  is a Hilbert space with the inner product in H .

Theorem45 Let S , S 1 be non-empty subspaces of a Hilbert space H , then

(a) H   0
0

(b) H
(c) S S   0
(d) S  S 1  S 1  S  .
(e) S  S 

Proof:

(a) 0  H  , as H  is a subspace of H  0  H  .

Let x  H   x , y  0 , y  H for x  y  H ,
 x ,x  0  x  0  x  0  x  0 ,
2

 H   0
 H   0

(b) 0  H is obvious .



Let x  H , sine x , 0  0 , x  H  x 0 .



 H  0 ,  H  0 .
 

(c) Let x  S S 
 x  S and x  S  , x  S   x , y  0 , y  S
 x , x  0 , x  S
 x  0 ,  x  0 0 .
2

S S   0
as S is a subspace of H , then 0  S , also S  is a subspace of H , so 0  S 
Therefore 0  S  S ,  0  S  S .
 S  S  0 .

(d) Let S  S 1 , and let x  S 1

019
 x,y 0 y  S 1 .
 x,y 0, y  S as, S  S 1 .
x S   S 1  S  .

(e) Let x  S ,  x , y  0 , y  S  .
 x  S   S  S  #

Direct sum:
A vector space X is said to be the direct sum of two subspaces Y and Z of X ,
written , X Y  Z , if each x  X has a unique representation x  y  z , y Y and
z  Z . ie each vector of X can be uniquely expressed as sum of an element of Y and
an element of Z .
Equivalently, X Y  Z X Y  Z , iff
(1) X Y  Z  x  y : x  X and y Y  .
(2) Y Z  0 , ie Y and Z are disjoint.

Example:
M 1   , 0  :    and M 2   0,   :    are two subspace of 2
obviously any
element  ,    2
can be uniquely expressed as sum of an element of M 1 and an
element of M 2 , the unique expression is  ,     ,0   0,   , thus 2
 M1 M 2.

Theorem46 : Let M and N be closed subspaces of a Hilbert space H , such

that M  N . Then M  N is also closed subspace of H .

Proof:
We know that if M and N are subspaces of H , then so is M  N . we show
M  N is closed.
Let z  M  N ,
  a sequence  z n  of points in M  N s.t z n  z .
as  z n  are points in M  N ,
 each z n  M  N can be uniquely expressed as z n  x n  y n , x n  M and y n  N .
Now each convergent sequence is a Cauchy sequence so  z n  is a Cauchy sequence ,
therefore as m , n  
zm zn 0
2

 x m  y m   x n  y n   0
2

 x m  x n    y m  y n   0
2
as m , n   .
001
We know form Pythagorean Theorem if x  y then x  y  x  y
2 2 2
,
x m  x n  M , y m  y n  N and M  N ,
 x m  x n    y m  y n 
 x m  x n    y m  y n   x m  x n
2
 ym yn  0 as m , n   ,
2 2

  x n  and  y n  are Cauchy sequences in M and N which are complete being closed
subspace of H and so,
x  M , y  N s.t x n  x and y n  y .
 zn  xn  yn x  y
Now, z  nlim z n  lim  x n  y n   lim x n  lim y n  x  y .
 n  n  n 

 z  x  y , x M , y N .
 z  M N
 M  N is closed. #

Lemma: Let M be a proper closed subspace of a Hilbert space H , then there exists a
non zero vector z 0 in H such that z 0  M ie z 0 , z  0 z  M .

Theorem47 : Let M be a subspace of a Hilbert space H , then M is closed iff

M  M  .

Proof:
let M be closed subspace of a Hilbert space H , we show M  M  .
By Theorem (45) (e) M  M  . Let M  M  , then M is a proper closed subspace of

M  , and so by Lemma z 0  0  M s.t z 0  M , ie z 0  M  .
Now z 0  M  and z 0  M   z 0  M  M 
since M  is a subspace of H so, by Theorem (45) (c) M  M   0
 z 0 M  M   0
 z 0  0 which is contradiction, that z 0  0
 M  M 
Conversely: Let M be a subspace of a Hilbert space H and M  M    M 


, we
show M is closed .
We know from Theorem (44) that  M   is a closed subspace of H .


 M  M  is a closed subspace of H . #

Theorem48 : (projection Theorem) Let M be a closed subspace of a Hilbert

space H . then H  M  M  .
000
Proof:
Let M be a closed subspace of a Hilbert space H , we know that M M   0 ,
Therefore to show that H  M  M  , we prove that H  M  M  .
Let N  M  M  , we show H  N , obviously, M  N and M   N ,
 N   M  and N   M  (by Theorem (45) (c))
N   M  M   0
 N   0 as 0  N  is a subspace  N   0 ,


as, 0  H .

Or, N   H
Now, N  M  M  is a closed subspace of H
 N   N by Theorem47.
N H
 H  M M  . #
Example :
Let M   , 0 :   and N   0,   :    , then M and N are subspace of 2
,
and M   N , and 2
 M N ,  2
 M  M .

Orthonormal set:
A non-empty subset e i  of a Hilbert space H is said to be an orthonormal set if
(i) e i  1 for every i .
(ii) i  j  ei  e j or i  j  ei ,e j  0 , ie
0 if i  j
ei ,e j  
1 if i  j
Thus a non-empty subset of a Hilbert space H is said to be orthonormal set if it consists
of mutually orthogonal unit vectors.

Remarks:
(1) The orthonormal set can't contain zero vector because 0  0 .
(2) If an orthonormal set M is countable , we can arrange it in a sequence  x n  and call it
orthonormal sequence .

Examples:
(1) Euclidean space 3 :
The three vectors 1,0,0  ,  0,1,0  ,  0,0,1 form an orthonormal set in 3
.
(2) Space L2 : The sequence e n  where e n   n j  has the n element 1 and other zero
th

forms an orthonormal set in L2 .

002
 Theorem49: (Bessel's' inequality for finite orthonormal set )
Let e1 , e 2 ,..., e n  be a finite orthonormal set in a Hilbert space H . if x is any
n


2
 x .
2
vector in H , then x ,ei
i 1

Proof:
n
Let y  x   x , e i e i , then
i 1

0 y  y,y
2

n n
 x   x ,ei ei , x   x ,e j e j
i 1 j 1

n n n n
 x , x  x ,  x ,e j e j   x ,ei ei , x   x ,ei ei ,  x ,e j e j
j 1 i 1 i 1 j 1
n n n n
 x   x ,e j x , e j   x , e i e i , x   x , e i
2
x ,e j e i ,e j .
j 1 i 1 i 1 j 1

Since e1 , e 2 ,..., e n  is an orthonormal set

 ei ,e j  0 i  j , and e j , e j  1
n n n
 0 x   x ,e j x ,e j   x ,e i x ,e i   x ,e j
2
x ,e j
j 1 i 1 j 1

n 2

 x   x ,ei
2
zz  z
2
as,
i 1
n
  x ,ei
2
 x
2
.#
i 1

Remark:
Bessel's' inequality for infinite sequence is If e i  is an orthonormal sequence in a



2
 x
2
Hilbert space H , then x ,ei , for every vector x in H.
i 1

003
 Theorem50: An orthonormal set is linearly independent.

Proof:
Let e1 , e 2 ,..., e n  be an orthonormal set,
Let 1e1   2e 2  ...   ne n  0 ,
Multiplication by a fixed e j gives
n n

 e
k 1
k k , ej   k e k , e j   j e j , e j   j  0
k 1

 The orthonormal set is linearly independent. #

We have seen that orthonormal sequence is L.I set, the remaining problem is how to
obtain an orthonormal sequence if an arbitrary linearly independent sequence given .
This is done by a constructive procedure, the Gram Schmidt process for
orthonormalizing a linearly independent sequence x j in an inner product space.  

Gram Schmidt process:

Let X be an inner product space, let S  x 1 , x 2 ,..., x n  be a linearly independent
sequence . then there exists an orthonormal sequence e1 , e 2 ,..., e n  such that.
Span x 1 , x 2 ,..., x n   span e1 , e 2 ,..., e n 

Processors:
Step 1:
The first element of e k  is
 x  is
x1
e1  , e1  0 as x 1  0 . Since j L.I , then e1 is a unit vector and e1 is an
x1
orthonormal set.

Step 2:
Take u 2  x 2  x 2 , e1 e1 ,
u 2  0 ; also u 2  e1
Since u 2 , e1  x 2  x 2 , e1 e1 , e1
 x 2 , e1  x 2 , e1 e1 , e1 .
0

004
 u2 1
Take e 2  then clearly e 2  1 , and e 2 , e1  u 2 , e1  0 .
u2 u2

Step 3:
Take u 3  x 3  x 3 , e1 e1  x 3 , e 2 e 2
u 3  0 and u 3  e1 as well as u 3  e 2 .
as u 3 , e1  x 3 , e1  x 3 ,e1 e1 ,e1  x 3 ,e 2 e 2 ,e1  0 ,
and u 3 , e 2  x 3 , e 2  x 3 , e1 e1 , e 2  x 3 , e 2 e 2 , e 2  0
u3
We take e 3  .
u3

n th step:
n 1

Take u n  x n   x n , e k e k , un  0
k 1
and is orthogonal to e1 , e 2 ,..., e n 1 ,
un
We take e n  .
un
Thus we get an orthonormal set e1 , e 2 ,..., e n  .

Q.9/P160

Orthonormalize the I.st three terms of the sequence x 1 , x 2 ,... , where x j (t )  t j 1 on

 1,1 , where x , y   x t  y t  dt .
1

1

Solution:
x 1  x 1 (t )  t 0  1
x 2  x 2 (t )  t
x 3  x 3 (t )  t 2

Step 1:
x1
e1  , x1 1
x1
1
 x , x   12dt  2 ,
2
and, x
1

005
 x  2  e1  1
2

Step 2:
u 2  x 2  x 2 , e1 e1
1 1
x 2 , e1  t , e1   t dt  0 ,
1
2
 u2  x 2  t .
1
 u 2 ,u 2  t , t   t 2dt  2
2
Now, u 2 1 3
 u2  2 ,
3
u2 t
 e2    3 t.
u2 2 2
3

Step 3 :
u 3  x 3  x 3 , e1 e1  x 3 , e 2 e 2 , u3  t 2
1 1 1 2
x 3 , e1  t 2 ,  t2 dt  ,
1 3
2 2
and
1
x 3 , e 2   t 2 3 t dt  0 .
1 2
Thus
u 3  x 3  x 3 , e1 e1  0
2 1 1
t 2  . t 2  ,
3 2 3
2
1  1 8
and u 3  u 3 , u 3  2  t 2   
2
0
 3 45
1 8
 u3 
3 5

 e3 
u3

3t 2  1
3
5

3t 2  1   5
.
u3 3 8 8
1 
Thus the first three orthonormal elements are  , 3 2 t ,
5
   .
3t 2  1
2 8 

006
Operators in Hilbert Space:

Adjoint of a bounded linear operator on a Hilbert space:

Let X be a Hilbert space and T : X  X be a bounded linear operator on X into itself.
Then the adjoint operator T * of T is defined by
Tx , y  x ,T * y , x , y  X
Theorem51: The adjoint linear operator T * is linear, bounded and
unique.

Proof:
For all z  X
z ,T * (x  y)  Tz , x  y
 Tz , x  Tz , y
 z ,T *x  z ,T * y
 z ,T *x T * y

 z ,T * (x  y)  z ,T *x T * y  0, z  X

 z , T *  x  y   T *x T * y   0, z  X

 T *  x  y   T *x T * y [ For given y  X , x , y  0 x  X  y  0 ].

z ,T *  x   Tz ,  x   Tz , x   z ,T *x  z , T *  x  ,
or
z ,T *  x   T *  x   0, z  X
 T *  x   T *  x  ,
Thus T * is linear.
Also
 T *x ,T *x  T T *x  , x  T T *x  x , by Cauchy-Schawartz inequality.
2
T *x

Since T is bounded, there exists k  0, such that

T T *x   k T *x
2
 T *x  k T *x x

or T *x  k x .
007
Thus T * is bounded.
Now, we show that T * is unique.
Suppose that for a given T there exists two operators T 1* and T 2* such that.
Tx , y  x ,T1* y ,

and Tx , y  x ,T 2* y
 x ,T1* y  x ,T 2* y  0
 x ,T1* y T 2* y  0 , x  X
 T1* y T 2* y  0 i .e ., T1* y  T 2* y y
 T1*  T 2*
Hence the adjoint of T is unique.

Theorem52: Let T be a bounded linear operator on a Hilbert space

X into itself, then it's adjoint operator T * has the following
properties.
1) I *  I ( I is the identity operator)
2) T  S   T *  S *
*

3) T   T *
*

4) TS   S * T *
*

5) T **  T
6) T *  T
7) T *T  T
2

8) T * 1
 T
 , and if T is invertible i.e., T T
1 * 1
 I , then T * is

invertible ie T  T  I . 1 * *

Proof:
1) Since Ix  x x  X , we have
Ix , y  x , y  x , I * y by definition of I * .
 x , y  I * y  0, x  X .
 y  I * y  0 or I * y  y i .e ., I  I * .
2) T  S   T *  S *
*

008
 T  S  z , x  z , T  S  x , x  X
*

,
also
T  S  z , x  Tz  Sz , x
 Tz , x  Sz , x
 z ,T *x  z , S *x
 z ,T *x  S *x

 z , T  S  x  z ,T *x  S *x
*

 T  S  x  T *x  S *x , x  X
*

 T  S   T *  S *
*

T  z , x  z , T 
*
3) x ,

and T  z , x   Tz , x   z ,T *x  z , T *x
 z , T  x  z ,  T *x
*

 z , T  x   T *x  0
*

 T   T * .
*

4) TS  x   T S  x  
TS  x  , y  T S  x   , y
 Sx ,T * y

 x , S * T * y 

 x ,  S *T *  y ,
On the other hand,
TS  x , y  x , TS  y
*

 x ,  S *T *  y  x , TS  y
*

 TS   S *T *
*

5) Tx , y  x ,T * y  T **x , y
 Tx T **x , y  0

 T T **  x , y  0, y  X
 T  T **
6) T  T *

009
 2
T *x  T *x ,T *x

 T T *x  , x

 T T *x  x by Cauchy-Schawatz Inequality.
T T *x x as T is bounded
or, T *x  T x

 Sup T *x  Sup T x .
x 1 x 1

 T*  T ,
applying this relation to T * , we have
T **  T * , However T **  T

 T  T*

 T  T*

T *T  T
2
7)
T * Tx   T * T x  T
2
x

 T *T  T
2
,
On the other hand
 Tx ,Tx
2
Tx
 T *Tx , x
 T *Tx x
 T *T x x

 Sup Tx  Sup T *x
2 2
x
x 1 x 1

 T  T *T
2

 T *T  T
2

8) Let T be invertible, that is T T 1  T 1T  I , we show T * is invertible

T T   I  I , also T T   T  T
1 * * 1 * 1 * *
by (4)
 T T   T  T  I
1 * 1 * *

consequently,

021
T  T T   I T  * 1 * 1
 T * 
1 * * 1

 T   T 
1 * * 1

Now,
T * T *   T * T   T   I *  I  T *  T *.
1 1 * 1 * 1
T

Definitions:
Let T be a bounded linear operator on a Hilbert space X into itself
1) T is called self-adjoint or Hermitian if T  T * i.e., if Tx , y  x ,Ty .
2) A self-adjoint operator T is called a positive operator if Tx , x  0 , x  X .
It is called strictly positive if Tx , x  0 only if x  0 .
3) T is called normal if T T *  T * T
4) T is called unitary if T T *  T * T  I , where I is the identity operator.

Examples:
1) The identity operator I is a self-adjoint as I *  I .
2) The zero operator is a self-adjoint.
3) Let T : C 0,1  C 0,1 be defined by Tf t   t f t  , t  0,1 , where
b
f , g   f t  g t  dt , f , g C a,b  , t  a ,b  , then T is a self-adjoint operator.
a

T is bounded and linear operator

1 1
Tf , g  Tf t  g t  dt   t f t  g t  dt ,
0 0
1 1
f ,Tg   f t Tg t  dt   f t  t g t  dt ,
0 0
1
  t f t  g t  dt as t   0,1 , t  t .
0

 Tf , g  f ,Tg (1)
Now, the adjoint operator T * of T is defined by Tf , g  f ,T * g (2)
from (1) and (2) T  T * , that is T is self adjoint operator.
4) Let T : x  x be given by T  2iI , where I is the identity operator, then T is normal.
T T *   2iI  2iI    2iI  I *  2i    2iI  I  2i   4I , and
* *

T *T   2iI   2iI   4I
*

 T T *  T *T

020
Properties of self-adjoint, positive, normal and the unitary operator:

1) Every self-adjoint operator is normal, but converse is not true in general.

Proof:
Let T be self-adjoint operator, then
T  T *  T T *  T 2 , and T * T  T 2 .
Hence T * T  T T * .
For the converse, consider the above example T is normal but not self-adjoint, since
T *   2i  I *  2i I  T
*

2ix , y  2i x , y
y , 2ix  2i x , y , i  i
 2i is the adjoint operator of 2i .

2) Every unitary operator is normal but the converse need not be true.

Proof:
Let T be unitary operator, then T T *  T * T  I , thus T is normal.
For the converse, consider the above example T  2iI ,
T *   2iI   I *  2i   I  2i   2iI
* *

T * T  T T *  4I  I , hence T is normal but not unitary.

3) Let T 1 and T 2 be two self-adjoint operators on a Hilbert space X , then their product
T1 T 2 is self-adjoint if and only if T1T 2  T 2 T1 .

Proof:
Let T1T 2 be self-adjoint, then T1T 2   T1T 2 , and we know that T1T 2   T 2* T1* , thus
* *

T1T 2   T 2 T1 as T1 ,T 2 are self-adjoint.

*

 T1 T 2  T 2 T1 .
Conversely, Let T1T 2  T 2 T1 , we show that T1T 2 is self-adjoint.
T1T 2  x , y  x , T 1 T 2  y
*

 x ,T 2*T1* y
 x ,T 2 T1 y
 x ,T1 T 2 y
thus, T1T 2 is the adjoint of T1T 2 , i.e., T1T 2 is self-adjoint.

022
4) An operator T on a Hilbert space X is self-adjoint if and only if Tx , x is real
x  X .

Proof:
Let T be self-adjoint, then T  T , to show that Tx , x is real, we show that
*

Tx , x  Tx , x .
Now,
Tx , x  x ,Tx as x , y  y , x
 T *x , x
 Tx , x x  X
 Tx , x is real x  X .
Conversely, let Tx , x be real x  X , we show that T is self-adjoint.
Now, Tx , x  Tx , x , Tx , x  x ,T *x and Tx , x  x ,Tx
 x ,T *x  x ,Tx
 x ,T *x  x ,Tx  0

 x ,T *x Tx  x , T * T  x  0 x  X
 T *x  Tx x  X , T *  T
 T is self adjoint.

5) If T is bounded linear operator then T * T and T T * are positive.

Proof:
2
T T *x , x  T *x ,T *x  T *x  0, i .e ., T T * is positive.

similarly, T * Tx , x  Tx ,T **x  Tx ,Tx  Tx  0, i .e ., T *T is positive.

2

6) If T 1 and T 2 are normal operators on a Hilbert space X such that T1T 2*  T 2* T1 and
T 2 T1*  T1* T 2 , then T1 T 2 and T1 T 2 are normal.

Proof:
T1 T 2 T1 T 2   T1 T 2  T1* T 2* 
*

 T1 T1* T1T 2* T 2 T1* T 2 T 2*

and

023
T1 T 2  T1 T 2   T1* T 2*  T1 T 2 
*

 T1* T1 T1*T 2 T 2* T1 T 2* T 2
 T1 T1* T 2 T1* T1 T 2* T 2 T 2*
 T1 T 2 T1 T 2 
*

 T1 T 2  is normal.
also T1T 2 T1T 2   T1T 2  T 2*T1*   T1T 2*T 2 T1*  T 2*T1T 2 T1*
*

and T1T 2  T1T 2   T 2* T1*  T1T 2   T 2*T1T 2*T 2  T 2*T1T 2 T1*

*

 T1T 2  is normal.

7) Abounded linear operator T on a Hilbert space X is normal if and only if

T *x  T x ,  X .
Proof:
2
T * x  T x  T *x  T x
2

 T *x ,T *x  Tx ,Tx
 T T *x , x  T *Tx , x
 T T *x , x  T *Tx , x  0
 T T *x T *Tx , x  0

 T T * T *T  x , x  0 x  X
 T T * T *T  0  T T *  T *T
T is normal.

8) If T is a normal operator, then T 2  T .

2

Proof:
Let T be normal, then from (7), we have T 2 x  T Tx   T *Tx x  X

T 2
 T *T

but T *T  T
2

then T 2  T
2
.

024
9) Let T be a bounded linear operator on a Hilbert space X into itself and T * be its
T T * T T *
adjoint. Further suppose that A 
2
and B 
2i
, T  A  iB and T *  A  iB  ,

then T is normal if and only if AB  BA .

Proof:
Let AB  BA , we have
T T *   A  iB  A  iB 
 A 2  B 2  i  BA  AB 
T * T   A  iB  A  iB 
 A 2  B 2  i  AB  BA  ,
Since AB  BA , then T T *  T *T and so T is normal.
Conversely, let T T *  T *T
 A 2  B 2  i  BA  AB   A 2  B 2  i  AB  BA 
 BA  AB  AB  BA
 2BA  2AB
 AB  BA

10) A bounded linear operator T on a Hilbert space X into itself is unitary if and only
if it is an isometry of X onto itself.

Proof:
Let T be abounded linear operator on a Hilbert space X and T is an isometry of X onto
itself, then Tx  x x  X , then we have T *T  I ,
 x  Tx ,Tx  x , x
2 2
In fact Tx  x x , then, Tx
 T *Tx , x  x , x
 T *Tx , x  x , x  0

 T *T  I  x , x  0 x
 T *T  I  0
 T *T  I

025
 1
Now as T is an isometric of X onto itself, then T exists.
 T *T T 1
 IT 1
T 1

1 1
 T *T T T
1 1
 T *I  T as T T I
1
T * T
1
T T * T T I
T T T T  I
* *

T is unitary.
Conversely, Let T be unitary, then T T *  T *T  I
 Tx  x x by above lemma.
T is an isometric of X onto itself.

026
Solve the following questions from the text book(E. KREYSZIG):

Homework1: Homework5: Homework9:

Q3, 4, 5, 7, 8/P8 Q8, 10/P32 Q1, 2, 3, 4, 5, 6, 7, 8, 9/P76

Q11, 12, 13, 14, 15/P9 Q1, 2, 3, 4, 5/P39 Q8/P82
Q6, 7, 8/P40

Homework2: Homework6: Homework10:

Q3, 4 , 5, 6, 7, 8, 9, 10/P16 Q9, 11, 13, 14/P40 Q2, 3, 4, 5, 6, 7/P90

Q11, 12/P17 Q3, 6, 7, 8, 12, 14/P46 Q12, 13, 14, 15/P91

Homework3: Homework7: Homework11:

Q13, 14, 15/P17 Q1, 3/P81 Q1, 3, 4, 5, 7, 9, /P101

Q1, 2/P23 Q10/P82 Q1, 2/P109
Q3, 6, 7, 8, 9/P24 Q4, 7, 8, 10/P64 Q3, 4/110
Q11/P65
Q14, 15/P66

Homework4: Homework8: Homework12:

Q10/P24 Q1, 2, 3, 4, 5, 6/P70 Q3, 4, 5, 6, 7 /110

Q12, 14, 15/P25 Q11, 12, 13, 15/P71 Q10, 11/136
Q1/P31 Q4, 5/P140
Q2, 3, 4, 5, 6/P32 Q7, 8/P141

027