Chapter 2 Normed Spaces.

Banach Spaces

2.2 Normed Space. Banach Space

2.2-1 Definition. A normed space X is a vector space with a norm defined on

it. A Banach space is a complete normed space ( complete in the metric
defined by the norm; Note (1) below ). A norm on a real or complex
vector space X is a real-valued function on X whose value at an xX is
defined by || x || which has the properties
(N1) || x ||  0.
(N2) || x || = 0 if and only if x = 0.
(N3) || x || = |  | || x ||.
(N4) || x + y ||  || x || + || y ||. For all x, yX and scalar .

Notes. (1) A norm on X defines a metric d on X which is given by d(x, y) = || x – y ||

where x, yX and called the metric induced by the norm. Hence normed
spaces and Banach spaces are metric spaces.
(2) For all x, yX , | || x || – || y || |  || x – y ||.
(3) The norm is continuous, that is, x  || x || is a continuous
mapping of ( X, || . || ) into R.

Examples.
2.2-2 Euclidean space Rn, unitary space Cn. These are Banach spaces with
n
norm defined by || x || = ( | i |2 )
1
2 . This norm induces the metric
i 1
n
d( x, y ) = || x – y || = ( | i   i |2 ) 2 , where x = (1 , 2 ,......, n ) and
1

i 1

y = ( 1,  2 ,......,  n ) .

p
2.2-3 Space . This space is a Banach spaces with norm defined by

|| x || = ( | i | p )
1
p
. This norm induces the metric d( x, y ) = || x – y ||
i 1

= ( | i   i | )
1
p p
, where x = (i ) and y = ( i ) .
i 1


2.2-4 Space . This space is a Banach spaces with norm defined by
|| x || = sup | i | . This norm induces the metric d( x, y ) = || x – y || =
iN

sup | i   i | , where x = (i ) and y = ( i ) .

iN

1
2.2-5 Space C[a, b]. This space is a Banach spaces with norm defined by
|| x || = max | x(t ) | . This norm induces the metric d( x, y ) = || x – y ||
tJ

= max | x(t )  y(t ) | . where J = [a, b].

tJ

2.2-6 Incomplete normed space. The space X of all continuous real-valued

1

functions on J = [0, 1] with norm defined by || x || =  | x(t ) | dt is an

0
incomplete normed space. This norm induces the metric d( x, y ) = || x – y ||
1

=  | x(t )  y(t ) | dt .
0

2.2-7 Lemma ( Translation invariance). 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 ) for all x, yX and scalar .
Proof. Let d be a metric induced by a norm, then
(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 )

2.2-8 Space s. Let s be the vector space of all sequences with the metric


d( x, y ) = 
i 1
|i  i |
2 (1|i  i |)
i , for any x = (i ) and y = ( i ) in s. Show
that this metric can not be obtained from a norm.
Proof. Let  be a scalar such that |  |  1. Then
 

d(x, y ) =  i 1
|i  i |
2 (1|i  i |)
i = 
i 1
| | |i  i |
2 (1| | |i  i |)
i

 || 
i 1
|i  i |
2 (1|i  i |)
i = |  | d( x, y ). Hence by Lemma 2.2-7 this

metric is not induced by a norm

Note. The a bove example shows that metric spaces need not be normed spaces.

H. W. 5-15. H.W.* 8, 12, 14.

2
2.3 Further Properties of Normed Spaces

2.3-1 Definition. A subspace Y of a normed space X is a subset of X considered as

a vector space with a norm on Y obtained by restrecting the norm on X to Y.
2.3-2 Theorem. A subspace Y of a Banach space X is complete if and only if Y is
closed.
Notes. (i) A sequence (xn) in a normed space X is convergent if there is xX
such that lim || xn – x || = 0.
n

(ii) A sequence (xn) in a normed space X is Cauchy if for every  > 0

there is kN such that || xn – xm || <  for all n, m > k.
2.3-3 Definition. Let (xn) be a sequence in a normed space X, and let (Sn) be a
sequence of partial sums, where Sn = x1 + x2 + ……+ xn. If Sn converges to

S, the infinite series x
i 1
i
is said to be convergent and S is called the sum of
 
the series and write S =  xi . If || x1 || + || x2 || + …… converges, then
i 1
x
i 1
i

is said to be absolutely convergent.

2.3-4 Definition. If a normed space X contains a sequence (en) with the property
that for every xX there is a unique sequence of scalers (n) such that
lim || x – (1e1 + 2e2 + ……+ nen ) || = 0, then (en) is called a Schauder
n

basis for X. The series  e
i 1
i i
is called the expansion of x with respect to (en).
p
2.3-5 Example. (en) is a Schauder basis for , where e1 = ( 1, 0, 0, 0, ….),
e2 = ( 0, 1, 0, 0, …….), ………, en = ( 0, 0, 0,……, 1, 0, 0,.….),.…………..


| 
p
Proof. Let x = (i ) be any fixed element in . Then i | p is finite.
i 1

Then there is a sequence (i) of scalers such that

|| x – (1e1 + 2e2 + ……+ nen) || = || ( 1, 2,…, n, n+1, n+2,….) – ( 1e1 +

2e2 + …+ nen ) || = || ( 0, 0,…, 0, n+1, n+2,….) || = (  | i | p )
1
p
 0 as
i  n 1

n. Now assume that there is a sequence (i) of scalers such that || x – (1e1
n
+ 2e2 + ……+ nen) || 0 as n. Then, ||  (
i 1
i   i )ei || = || ( 1e1 +

2e2 + ……+ nen ) – ( 1e1 + 2e2 + ……+ nen ) ||  || x – ( 1e1 + 2e2

+…..+ nen ) || + || x – ( 1e1 + 2e2 +……+ nen ) ||  0 as n. Hence
n 
||  (
i 1
i   i )ei ||  0 as n which implies that 0 = ||  (
i 1
i   i )ei ||
p
= || (i - i ) ||. Thus (i ) = (i ). Therefore, (en) is a Schauder basis for 

3
2.3-6 Theorem( Completion ). Let X = ( X, || . || ) be a normed space. Then there
 
is a Banach space X and an isometry A from X onto a subspace W of X
 
which is dense in X . The space X is unique, except for isometries.

Proof. By Theorem 1.6-2 there is a complete metric space X and an isometry
 
A : X  W = A(X), where W is dense in X and X is unique except for
    
isometries. Let x , y  X , ( xn) x , ( yn)  y and zn = xn + yn. Since ( xn) and
 
( yn) are Cauchy sequences then ( zn) is a Cauchy sequences. Define z = x +
 
y to be the equivalence class for which ( zn) is a representative; thus ( zn)  z .
 
This definition is independent of the Cauchy sequences in x and y . In fact, if
(xn) ~ (wn) and (yn) ~ (vn) then || ( xn + yn ) - ( wn + vn ) ||  || xn - wn || +
  
|| yn - vn ||  0 as n. Hence ( xn + yn ) ~ ( wn + vn ) so that z = x + y is
  
well defined. Similarly, we define the product  x  X of a scalar  and x to
be the equivalence class for which (xn) is a representative. Similarly as in the
definition of addition one can see that the definition of product is independent
 
of the choice of a representative of x . The zero element of X is the
equivalence class contains all Cauchy sequences which converges to zero.

Then X under these operations is a vector space(how?). Let  be a scalar and
x, y X. Then there are sequences ( xn) and ( yn) in X such that xnx, yn y.
 
Let ( xn)  x and ( yn)  y . By using the definitions of scalar product and
   
vector addition in X we have Ax + Ay =  x + y =  x  y = A(x + y).
Hence A is an isometry that preserves the algebraic operations.
 
For any y = AyW, we have || . ||1 a norm on W as || y ||1 = || y ||. The

corresponding metric on W is the restriction of d to W since A is isometric.
    
Then we can extend the norm || . ||1 to X by setting || x ||2 = d ( 0 , x ) for all
   
x  X . || x ||2 is a norm (why?). Therefore, X is the desired space

H. W. 2-10, 12-15. H.W.*2, 8, 10, 12.

4
2.4 Finite Dimensional Normed Spaces and Subspaces

2.4-1 Lemma. Let {x1, x2,…, xn } be a linearly independent set of vectors in a

normed space X. Then there is a number c > 0 such that for any choice of
n
scalars 1, 2,…, n, we have || 1x1 + 2x2 + ..… + nxn ||  c | 
i 1
i |.

Proof. We first show that there is c > 0 such that for any choice of scalars 1,
n
2,…, n with | 
i 1
i | = 1, we have || 1x1 + 2x2 + … + nxn ||  c ……….(1)

Assume that (1) does not hold, then for all m > 0 there are scalars 1( m ) ,  2( m ) ,
n 1
……,  n( m ) such that | 
i 1
i
(m)
| = 1 and || 1x1 + 2x2 + ..… + nxn || < m .

Then there is a sequence ( ym ) of vectors, ym = 1( m ) x1 +  2( m ) x2 + ..… +

n
 n( m ) xn with | 
i 1
i
(m)
| = 1 and || ym ||  0 as m   …………………..(2)
n
Now since | 
i 1
i
(m)
| = 1, then | i( m ) |  1 for all i and all m. Hence for any

fixed i, ( i( m ) ) = (  i(1) , i(2) , ……. ) is a bounded sequence. Then ( 1( m ) ) has
a convergent subsequence, say ( 1( m,1) ) that converges to 1. Let ( y1, m ) be
the corresponding sequence which is a subsequence of ( ym ) where
y1, m = 1( m,1) x1 +  2( m ) x2 + ..… +  n( m ) xn. Also since (  2( m ) ) is a bounded
sequence, then it has a convergent subsequence (  2( m,2) ) that converges to 2.
Let ( y2, m ) be the corresponding sequence which is a subsequence of ( y1, m )
where y2, m = 1( m,1) x1 +  2( m,2) x2 + 3( m ) x3 +..… +  n( m ) xn. By the same
way after n steps we get ( yn, m ) as a subsequence of ( ym ) with yn, m =
1( m,1) x1 +  2( m,2) x2 + ..… +  n( m,n ) xn, where | 1( m,1) | + |  2( m,2) | + ..… +
|  n( m,n ) | = 1 and 1( m,1) 1,  2( m,2) 2, .……,  n( m,n ) n as m  .
Hence as m  , yn, m  y = 1x1 + 2x2 + ..… + nxn. Also, 1 = lim [
m
n
| 1( m,1) | + |  2( m,2) | + .… + |  n( m,n ) | ] =  |  i | . Then there is at least one i 0.
i 1

However, {x1, x2,…, xn } is linearly independent, then y  0 …………….(3)

Since the norm is continuous, then || yn, m ||  || y ||. But ( yn, m ) is a
subsequence of ( y m ), then by (2) || yn, m ||  0. Hence || y || = 0 and so y = 0.
From this and (3) we have a contradiction. Therefore, (1) is proved.
Now, if | 1| + | 2 | + ..…+ | n | = 0 then i = 0 for all i. Then || 1x1 + 2x2
n n
i
+ ..… + nxn ||  c | 
i 1
i | . Finally, if s = | 
i 1
i |  0, take i =
s
, i=
n n
i
 | i | = | 
1
1, 2, …, n. Then i | = 1. By substituting i = in (1) we
i 1 s i 1 s

5
 
get that || 1 x1 +  2 x2 + ..… + n xn ||  c; that is, || 1x1 + 2x2 + ..…
s s s
n
+ nxn ||  c s = c | 
i 1
i | . This completes the proof

2.4-2 Theorem. Every finite dimensional subspace Y of a normed space X is

complete. In particular, every finite dimensional normed space is complete.
Proof. Let ( ym) be any Cauchy sequence in Y. Suppose that dimY = n and
{e1, e2,…, en } be any basis for Y. Then each ym has a unique representation of
the form ym = 1( m ) e1 +  2( m ) e2 + ..… +  n( m ) en. Since ( ym) is a Cauchy
sequence, then for every  > 0, there is kN such that || ym - yr || <  for all
m, r > k. From this and by Lemma 2.4-1, there is c > 0 such that
n n
 > || ym - yr || = ||  (
i 1
i
( m)
  i( r ) ) ei ||  c  |  i( m )   i( r ) | . This implies
i 1

that for any fixed i and all m, r > k, | i( m)  i( r ) | < . Then for any fixed
c
i = 1, 2, …, n, the squence (  i( m ) ) = (  i(1) ,  i(2) ,….. ) is a Cauchy squence
of numbers. Hence it converges, say to i. Use these n limits to define
y = 1e1 + 2e2 + ..… + nen. Then yY. Furthermore, || ym - y || =
n n
||  (i( m)  i ) ei || 
i 1
| 
i 1
i
( m)
  i | || ei ||. However,  i( m ) i as m  ,

then || ym - y ||  0, that is ym  y. Hence ( ym) converges in Y. Therefore, Y

is complete
2.4-3 Theorem. Every finite dimensional subspace Y of a normed space X is closed.
Note. Infinite dimensional subspaces need not be closed. To see this, consider
the space C[0, 1] and the subspace Y = span(x0, x1, x2,….. ), where xj(t) = t j.
Then Y is not closed ( why? ).
2.4-4 Definition. A norm || . || on a vector space X is said to be equivalent to a norm
|| . ||0 on X if there are positive numbers a and b such that for all xX we have,
a || x ||0  || x ||  b || x ||0.
Note. Equivalent norms on X defines the same topology for X. The proof is
left to the reader.
2.4-5 Theorem. On a finite dimensional vector space X, any norm || . || is equivalent
to any other norm || . ||0.
Proof. Let dim X = n and {e1, e2,…, en } be any basis for X. Then every xX
has a unique representation x = 1e1 + 2e2 + ..… + nen, where 1, 2,…,
n are scalars. By lemma 2.4-1, there is a positive constant c such that
n n n
|| x ||  c  |  i | . On the other hand || x ||0   |  i | || ei ||0  k  |  i | , where
i 1 i 1 i 1
n

| 
k
k = max {|| e1 ||0, || e2 ||0,…., || en ||0 }. Together, || x ||0  k i |  || x ||.
i 1 c
c
Hence a || x ||0  || x ||, where a = . Similarly, || x ||  b || x ||0. Together,
k
a || x ||0  || x ||  b || x ||0. Hence || . || and || . ||0 are equivalent
H. W. 4-6, 8,9. H.W.*4, 8.

6
2.5 Compactness and Finite Dimension.

2.5-1 Definition. A metric space X is said to be compact if every sequence in X has

a convergent subsequence. A subset M of X is said to be compact if M is
compact considered as a subspace of X; that is every sequence in M has a
convergent subsequence whose limit belongs to M.
2.5-2 Lemma. A compact subset M of a metric space is closed and bounded.
____
Proof. Let x be any element in M . Then there is a sequence ( xn ) of elements
in M such that xnx. However, M is compact, then xM. Hence M is closed.
Suppose that M is not bounded, then it contains unbounded sequence ( yn )
such that d( yn, b ) > n, where b is any fixed element. Hence ( yn ) has not a
bounded subsequence which implies that it has not a convergent subsequence.
This contradicts the compactness of M. Therefore, M is bounded
Remark. The converse of Lemma 2.5-2 need not be true in general.
2
Proof. Consider the sequence ( en ) in , where en = ( nj ). This sequence is
bounded since || en || = 1. To show that M is closed we show that M has no
accumulation point. First we show that any point in M can’t be an
accumulation point. Let en be any fixed point in M, then for any m  n, we
have d(en, em ) = || ( 0, 0,…, 0, 1, 0, 0, ….0, 1, 0, 0,…., 0 ) || = 2. Then
B(en, 1) \ { en }  M = . Hence any point in M can’t be an accumulation
point. Second we show that any point in M c can’t be an accumulation point.
Let x be any fixed point in M c. Then B(x, 1 ) \ { x }  M has at most one
2
element (why?). Then x is not an accumulation point [ See 1.3 Problem 6 ].
Hence any point in M c can’t be an accumulation point. Therefore, M has no
____
accumulation point. Hence M = M and so M is closed. Moreover, M has no
convergent subsequence because M has no accumulation poin
2.5-3 Theorem. In a finite dimensional normed space X, any subset M of X is
compact if and only if M is closed and bounded.
Proof. By lemma 2.5-2, compactness implies closedness and boundedness.
Conversely, suppose that M is closed and bounded. Let dim X = n and
{e1, e2,…, en } be a basis for X. Consider any sequence ( xm ) in M, where
xm = 1( m ) e1 +  2( m ) e2 + ..… +  n( m ) en. Since M is bounded so is ( xm ), say
n
|| xm ||  k for all m. By lemma 2.4-1 there is c > 0 such that k  ||  i( m ) ei ||
i 1
n
 c  | i( m ) | . Hence for any fixed i = 1, 2, …, n, the sequence ( i( m ) ) is a
i 1

bounded sequence of numbers then it has a convergent subsequence, say it

converges to i. Then as in the proof of lemma 2.4-1 we show that ( xm ) has a
n
subsequence ( zm ) which converges to z = 
i 1
i ei. However, M is closed then

zM. Therefore, M is compact

7
2.5-4 F. Riesz;s Lemma. Let Y and Z be subspaces of a normed space X, and
suppose that Y is closed and a proper subset of Z. Then for any real number
  (0, 1) there is zZ such that || z || = 1 and || z – y ||   for all yY.
Proof. Let v be any element in Z – Y, and let a = inf { || v – y || : yY}.
Since Y is closed, then a > 0 ( why?). Now, take any   (0, 1), then by
a
definition of infimum there is y0 Y such that a  || v – y0 ||  …..(1)

1
Let z = c( v – y0), where c = . Then || z || = 1. Let y be any fixed
|| v - y0 ||
element in Y, then || z – y || = || c( v – y0) – y || = c|| v – y0 – c -1y || = c|| v – y1||,
where y1 = y0 + c -1y which belongs to Y. Hence || v – y1||  a. Use (1) to get
|| z – y || = c|| v – y1||  ca   a = 
a
Note. In a finite dimensional space the closed unit ball is compact.

2.5-5 Theorem. If a normed space X has the property that the closed unit ball
M = { x X: || x ||  1 } is compact, then X is finite dimensional.
Proof. Suppose that M is compact but dimX = . Choose any x1X with
|| x1 || = 1. This x1 generates a one dimensional subspace X1 of X which is
closed and a proper subspace of X since dimX = . By Riesz’s lemma there is
1
x2X with || x2 || = 1 and || x2 – x1 ||   = . x1 and x2 generate a two
2
dimensional subspace X2 of X which is closed and a proper subspace of X. By
1
Riesz’s lemma there is x3X with || x3 || = 1 and for all xX2, || x3 – x ||  .
2
1 1
In particular || x3 – x1 ||  and || x3 – x2 ||  . Proceeding by induction we
2 2
1
obtain a sequence ( xn ) of elements xnM such that || xm – xn ||  for m  n.
2
Hence ( xn ) can’t have a convergent subsequence. This contradicts the
compactness of M. Therefore, the dimension of X is finite

2.5-6 Theorem. Let X and Y be metric spaces and T : X  Y be a continuous

mapping. Then the image of a compact subset of M under T is compact.
Proof. Let ( yn ) be any sequence in T(M). Then for any n there exists xnM
such that yn = Txn. Since M is compact, then ( xn ) contains a subsequence
( xnk) which converges in M. However, T is continuous then the image of ( xnk )
is a subsequence of ( yn ) which converges in T(M). Hence T(M) is compact
2.5-7 Corollary. A continuous mapping T of a compact subset M of a metric space
X into R assumes a maximum and a minmum at some points of M.
Proof. By Theorem 2.5-6, T(M)  R is compact and so is closed and
bounded, so that inf(T(M)) T(M) and sup(T(M)) T(M). Then the inverse
image of these two points consists of points of M at which Tx is minimum or
maximum, respectively
H. W. 1, 2, 5-7, 9, 10. H.W.*6, 10.

8
2.6 Linear Operators.

2.6-1 Definition. A linear operator T is a mapping such that (i) The domain D(T) of
T is a vector space and the range R(T) lies in a vector space over the same
field. (ii) For all x, yD(T) and scalars , T(x + y) = Tx +Ty and T(x) = Tx.
The null space of T is N(T) = { xD(T): Tx = 0 }.
Note. The linear operator is a homomorphism of vector space into another
vector space.

2.6-2 Theorem. Let T be alinear operator, then

(a) The range R(T) is a vector space.
(b) If dimD(T) = n, then dimR(T)  n.
(c) The null space N(T) is a vector space.
Proof. Left to the reader.

2.6-3 Theorem. Let X and Y be vector spaces both real or both complex. 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 if and only if Tx = 0 implies x = 0.
(b) If T -1 exists, then it is a linear operator.
(c) If dimD(T) = n and T -1 exists, then dimD(T) = dimR(T).
Proof. Left to the reader.

2.6-4 Lemma. Let X, Y and Z be vector spaces and let T : X  Y and S : Y  Z

be bijective linear operators. Then (ST) -1 exists and (ST) -1 = T -1S -1.
Proof. Left to the reader.

H. W. 5, 6, 11-15. H.W.* 14.

9
2.7 Bounded and Continuous Linear Operators.

2.7-1 Definition. Let X and Y be normed spaces both real or both complex. Let
T : D(T)  Y be a linear operator with D(T)  X. The operator T is said to be
bounded if there is a number c such that for all xD(T). || Tx ||  c || x ||.
Notes. 1) The smallest c such that the above inequality holds for all nonzero
||Tx||
xD(T) is defined as || T ||, thus || T || = sup{ || x|| : xD(T), x  0 }.
If D(T) = {0} then we define || T || = 0.
2) From the definition above, || Tx ||  || T || || x || with c = || T ||.

2.7-2 Lemma Let T : D(T)  Y be a bounded linear operator. Then

|| T || = sup{|| Tx || : xD(T), || x || =1}, and || T || satisfies (N1) to (N4).
Proof. Left to the reader.

Examples.
2.7-3 Identity operator. The identity operator I : X  X on a normed space X  {0}
is a bounded linear operator with || I || = 1.
Proof. Left to the reader.

2.7-4 Zero operator. The zero operator 0: XY on a normed space X is a bounded
linear operator with || 0 || = 0.
Proof. Left to the reader.

2.7-5 Differentiation operator. Let X be the normed space of all polynomials on

J = [0, 1] with || x || = max | x(t ) | . The differential operator T: XX defined
tJ
by Tx(t) = x /(t) is an unbounded linear operator.
Proof. For all x, yX and all scalars , T(x + y) = (x + y)/(t) = x/(t) +
y/(t) =  Tx + Ty. Hence T is linear. Let xn(t) = t n, where nN. Then || xn ||
n 1
= max | xn (t ) | = max | t | = 1 and || Txn || = max | nt | = n. Then || Txn || = n.
n
tJ tJ tJ || xn ||
But n was arbitrary in N, then there is no c such that || Txn ||  c for all xn.
|| xn ||
Therefor, T is not bounded
1

2.7-6 Integral operator. Let T:C[0, 1]C[0, 1] be defined as Tx(t) =  k (t , ) x( )d .
0

Where k is a given continuous function on G = JJ, which is called the kerrnal

of T and J = [0, 1]. Then T is a bounded linear operator.
1

Proof. For all x, yC[0, 1] and all scalars , T(x + y) =  k (t , )[ x  y]( )d
0
1 1

=   k (t , ) x( )d +  k (t , ) y ( )d =  Tx + Ty. Hence T is linear. Since k is

0 0

11
 continuous on the closed bounded square G, then k is bounded, that is there is
c > 0 such that | k(t, ) |  c for all (t, )G. Furthermore, | x(t) |  max | x(t ) |
tJ
1 1

= || x ||. Hence || Tx || = max |  k (t , )x ( )d |  max  | k (t , ) | |x( ) | d 

tJ tJ
0 0
1

c  || x || d = c || x ||. Hence T is bounded

0

2.7-7 Matrix. Let A = (jk ) be an mn real matrix. Define T: Rn  Rm by Tx = Ax.

Then T is a bounded linear operator.
Proof. For all x, y Rn and all scalars , T(x + y) = A(x + y) = Ax + Ay
=  Tx + Ty. Hence T is linear. By using Cauchy Shwartz inequality we
 1   11 12 ...... 1n   1 
    
 2    21  22 ......  2 n    2 
have that for all x =  .   Rn, || Tx || 2 = || Ax || 2 = ||  . . .   .  ||
2

    
 .  . . .  . 
    m 2 ......  mn    n 
 n  m1
 n

   1k  k 
 k 1 
 n 
   2kk  2
m n m n n m n

      2 12

l2 ] 2 ) 2  
1

2 2
= ||  k 1  || = [ jk k ] ([ jk ] [ = || x || 2 jk .
 .  j 1 k 1 j 1 k 1 l 1 j 1 k 1
 
 . 
 n 
   mk  k 
 k 1 
m n
Hence || Tx ||  c || x || , where c =
2 2 2 2
 
j 1 k 1
2
jk . Hence T is bounded

2.7-8 Theorem. If a normed space X is finite dimensional, then every linear operator
on X is bounded.
Proof. Let dim X = n and {e1, e2,…, en } be a basis for X. Then every xX
has a unique representation x = 1e1 + 2e2 + ..… + nen, where 1, 2,…,
n n
n are scalars. By linearity of T, || Tx || = ||   iTei ||  |  i | ||Tei || 
i 1 i 1
n
max || Tek ||
k
| 
i 1
i | …………………………………………………………(1)
n n
By Lemma2.4-1, there is c > 0 such that || x || = ||   i ei ||  c  |  i | . By (1),
i 1 i 1
1
|| Tx ||  c max || Tek || || x ||. Therefore, || Tx ||  m|| x ||, where M = 1c max || Tek || .
k k

Hence T is bounded

11
2.7-9 Theorem. Let T : D(T)  Y be a linear operator, where D(T)  X and X
and Y are normed spaces then, (a) T is continuous if and only if T is bounded.
(b) If T is continuous at a single point, it is continuous.
Proof. (a) For T = 0 the statement is trivial. Let T  0. Then || T ||  0.
Suppose that T is bounded and let x0 be any fixed point in D(T). Let  > 0 be

given. Then, since T is linear, there is  = ||T || > 0 such that if x  D(T) and
|| x - x0 || < , then || Tx - Tx0 || = || T(x - x0) ||  || T || || x - x0 || < || T || = .
Therefore, T is continuous at x0, but x0 was arbitrary, then T is continuous.
Conversely, assume that T is continuous at an arbitrary point x0D(T). Then
for every  > 0 there is  > 0 such that if xD(T) and || x - x0 || < , then
|| Tx - Tx0 || < ………………………………………………………………(2)
y
Let y  0 be any element in D(T) and set x = x0 + 2|| y || . Then || x - x0 || =
y
|| 2|| y|| || < . By (2) and by using the linearity of T we have,  > || Tx - Tx0 || =
y  ||Ty||
||T(x - x0) || = ||T( 2|| y|| )|| = 2|| y|| . Then ||Ty ||  2 || y || = c || y ||, where c = 2 .
Hence T is bounded
(b) Assume that T is continuous at a single point, then by the second part of the
proof of (a) T is bounded and so by (a) T is continuous

2.7-10 Corollary. Let T : D(T)  Y be a bounded linear operator, where D(T)  X

and X and Y are normed spaces. Then (a) xn  x [where xn, x D(T)] implies
Txn  Tx. (b) The null space N(T) is closed.
Proof. Left to the reader.
Note. For any bounded linear operators T : X  Y and S : Y  Z, where X,
Y and Z are normed spaces we have, || TS ||  || T || || S || and || T n ||  || T || n.

2.7-11 Theorem. Let T : D(T)  Y be a bounded linear operator, where D(T) lies
in a normed space X and Y is a Banach space. Then T has an extension
~ _______ ~ ~
T : D(T )  Y where T is a bounded linear operator of norm || T || = || T ||.
_______
Proof. Let x  D(T ) be arbitrary. Then there is a sequence ( xn ) of elements
in D(T) such that xn  x. By using the linearity and boundedness of T we have,
||Txn – Txm || = ||Txn – Txm ||  || T || || xn - xm ||. Hence ( Txn ) is a Cauchy
sequence, because ( xn ) is. But Y is complete, then ( Txn ) converges, say
~ ~ ~
Txn  yY. Define T by T x = y. We show that T is well defined.
Suppose xn  x and zn  x. Then vm  x, where (vm ) = ( x1, z1, x2, z2,….. ).
Hence (Tvm) converges and the two subsequences (Txn ) and (Tzn) of (Tvm)
~ _______
must have the same limit. Hence T is well defined on D(T ) . Let x and x / be
_______
any elements in D(T ) and  any scalar. Then there are sequences ( xn ) and
( xn/ ) of elements in D(T) such that xn x and xn/  x /. As above Txn yY

12
 ~ ~
and Txn/  y /Y. Then T x = y and T x / = y /. Since T is linear then
T(xn + xn/ ) = Txn + Txn/  y + y /. But (xn + xn/ )  x + x /, then by
~ ~ ~ ~ ~
the deffinition of T , T ( x + x ) = y + y =  T x + T x . Therefore, T is
/ / /

~ ~
linear. But T x = Tx for all xD(T) because x x and Tx  Tx. So that T
is a linear extension of T.
_______
For any x D(T ) there is a sequence ( xn ) of elements in D(T) such that xn x
~
and as above Txn yY and T x = y. Since || Txn ||  || T || || xn || for all n.
~
Letting n   and using the continuity of the norm to get || T x || = || y || =
~ ~
lim || Txn ||  lim|| T || || xn || = || T || || x ||. Hence T is bounded and || T ||  || T ||.
~ ~ _______
However, || T || = sup{|| T x ||: x D(T ) , || x || =1}  sup{||Tx ||: xD(T), || x ||
~
=1} = || T ||. Hence || T || = || T ||

H. W. 1, 2, 5-10. H.W.* 6, 8, 10.

13
2.8 Linear Functionals.

2.8-1 Definition. A linear functional is a linear operator with domain in a vector

space and range in the scalar field of that vector space.
2.8-2 Definition. A bounded linear functional is a bounded linear operator with
domain in a normed space and range in the scalar field of the normed space.
2.8-3 Theorem. A linear functional f with D(f) in a normed space is continuous if
and only if f is bounded.
Examples.
2.8-4 Norm. The norm || . || : X  R on a normed space ( X, || . || ) is a functional on
X which is not linear.
Proof. Left to the reader.
2.8-5 Dot Product. Let a = (1, 2, 3) be a fixed element in R3. Define f :R3 R by
f(x) = xa for all xR3. Then f is a bounded linear functional with || f || = || a ||.
Proof. For all x, yX and all scalars , f(x + y) = (x + y)a = (xa) + ya =
3
f(x) + f(y). Hence f is linear. For any x = (1, 2, 3)R3, | f(x) | = ( (i i )2 )
1
2

i 1
3 3
 ( i 2 ) 2 (  i 2 ) = || x || || a ||. Hence f is bounded and || f ||  || a ||.
1 1
2

i 1 i 1

| f ( x )| | f ( a )| || a ||2
However, || f || = sup{ || x|| : xD(f), x  0 }  ||a|| = || a || = || a ||.
Therefore, || f || = || a || .
b

2.8-6 Definite Integral. Let J = [a, b] and f :C[a, b]R be defined by f(x) =  x(t )dt .
a

Then f is a bounded linear functional with || f || = b – a.

Proof. f is a linear functional (why?) For any xC[0, 1] we have, | f(x) | =
b

|  x(t )dt |  (b-a) max | x(t ) | = (b-a) || x ||. Hence f is bounded and || f ||  b-a.
tJ
a
b
| f ( x0 )|
Now choose x0C[a, b] with x0 = 1, then || f ||  || x0 || = | f(x0)| =  dt = b-a.
a

Hence || f || = b – a
2.8-7 Space C[a, b]. Let f :C[a, b]R be defined by f(x) = x(t0) where t0 is a fixed
point in J = [a, b]. Then f is a bounded linear functional with || f || = 1.
Proof. f is a linear functional (why?) For any xC[0, 1], | f(x) | = | x(t0) | 
max | x(t ) | = || x ||. Hence f is bounded and || f ||  1. Now, choose x0C[a, b]
tJ
| f ( x0 )|
with x0 = 1, then || f ||  || x0 || = | f(x0)| =| x(t0) | = 1. Hence || f || = 1
2 2 2
2.8-8 Space . Let (i) be a fixed point in and f : R be defined by


 
2
f(x) = i i for all x  . Then f is a bounded linear functional.
i 1

Proof. Left to the reader.

H. W. 2, 3, 7-13. H.W.* 3, 10, 12.

14
2.9 Linear Operators and Functionals on Finite dimensional Spaces.

Remark. Let X* be the set of all linear functionals defined on a vector space
X. For any f, gX* and any scalar , define (f + g) and (f) by (f + g)(x) =
f(x) + g(x) and (f)(x) = f(x) for all xX. Then X* is a vector space called
the algebraic dual space of X.
X** is the space of all linear functionals defined on X*, which is called the
second algebraic dual space.

2.9-1 Theorem. Le X be an n-dimensional vector space and E = {e1, e2,…, en } be a

basis for X. Then the set of linear functionals F = { f1, f2,…, fn } given by
0 if j  k
fk(ej) = 
 is a basis for X* and so dimX* = dimX.
1 if j  k
n
Proof. To show that F is a linearly independent set, let 
k 1
k f k = 0. Then
n n


k 1
k f k ( x) = 0 for all xX. In particular for any ej, 0 = 
k 1
k f k (e j ) = j. This

implies that j = 0 for all j =1, 2,...,n. Therefore, F is a linearly independent set.
n
Let f be any fixed element in X*. For any x =  e
k 1
k k
X, by using the linearity
n
of f we have f(x) = 
k 1
k f (ek ) ………………………………………………(1)

On the other hand for any fixed 1  j  n,

n n
fj(x) = fj(   k ek ) =  k f j (ek ) = j …..………………………………(2)
k 1 k 1
n n
By (1) and (2), f(x) = f
k 1
k ( x) f (ek ) for all xX. Hence f = f
k 1
k f (ek ) .

Therefore, f  span(F). Hence F is a basis for X* and so dimX = dimX*

2.9-2 Lemma. Let C: X  X** be a canonical mapping on a vector space X defined

by Cx = gx, where for any fixed xX, gx: X* K [ K is the real or the complex
field. ] is defined by gx(f) = f(x) for any fX*. Then C is a linear operator
Proof. First we show that gxX**. For any f1,f2X*and any scalar , we have
gx(f1 + f2) = (f1 + f2)(x) = (f1)(x) + f2(x) =  gx(f1) + gx(f2). Hence gx is
linear on X*, that is gxX**. Then for any xX there is a unique gxX**
such that Cx = gx, because if Cx = gx and Cx = hx for some gx, hxX**, then
gx(f) = f(x) and hx(f) = f(x) which is impossible because f is a function. Hence
gx = hx. Therefore, C is well defined. Finally, we show that C is linear. For any
x, yX, any scalar , and any fX* we have, C(x + y)(f) = gx + y(f) =
f(x + y) = (f)(x) + f(y) = gx(f) + gy(f) = C(x)(f) + C(y)(f). Hence,
C(x + y) = Cx + Cy. Therefore, C is linear

Note. If R(C) = X**, then X is said to be algebraically reflexive.

15
2.9-3 Lemma. Let X be an finite dimensional vector space. If x0X has the property
that f(x0) = 0 for all fX*, then x0 = 0.
n
Proof. Let {e1, e2,…, en } be a basis for X and x0 =   j e j X. Then 0 = f(x0)
j 1
n n
= 
j 1
j f (e j ) for all fX*. Hence 
j 1
j f (e j ) = 0 for every choice of f(ej). Then

j = 0 for all j, that is x0 = 0

2.9-4 Theorem. A finite dimensional vector space is algebraically reflexive.
Proof. By lemma 2.9-2 C is linear. Let Cx0 = 0. Then for all fX*, 0 = Cx0(f)
= g x (f) = f(x0). Hence, by lemma 2.9-3, x0 = 0. Therefore, C is 1-1 and so
0

C : R(C)  X exists. Then by theorem 2.6-10 dimR(C) = dimX. By theorem

-1

2.9-1 dimX** = dimX* = dimX, together, dimR(C) = dimX**. However, R(C)

is a vector subspace of X**, then by theorem 2.1-8 R(C) = X**. Hence C is
onto. Therefore, X is algebraically reflexive

2.9-5 Definition. Two vector spaces are said to be isomorphic if there is a bijective
linear map between the two vector spaces. Two normed spaces are said to be
isomorphic if there is a bijective linear map that preserves the norm between
the two normed spaces. If the vector space X is isomorphic with a subspace of
a vector space Y, then we say that X is embeddable in Y.

H. W. 4, 5, 11, 13. H.W.* 13.

16
2.10 Normed Spaces of Operators. Dual Space.

Remark. Let B( X, Y ) denote the set of all bounded linear operators from a
normed space X into a normed space Y. For any T, S B( X, Y ) and any
scalar , define (T+ S) and (T) by (T+ S) (x) = T(x) + S(x) and (T)(x) =
T(x) for all xX. Then B( X, Y ) is a vector space.
2.10-1 Theorem. The vector space B( X, Y ) of all bounded linear operators from a
normed space X into a normed space Y is it self a normed space with norm
defined by || T || = sup{ || x|| : xD(T), x  0} = sup{||Tx ||: xD(T), || x ||=1}.
||Tx||

Proof. By the above remark and lemma 2.7-2 we get the result
2.10-2 Theorem. If Y is a Banach space, then B( X, Y ) is a Banach space for any
normed space X.
Proof. Let ( Tn ) be any Cauchy sequence in B( X, Y ). Then for every  > 0
there is kN such that || Tn – Tm || <  for all n, m > k. Then for all xX and
all n, m > k, || Tnx – Tmx || = || (Tn – Tm)x ||  || Tn – Tm || || x || < || x || ….(1)
Now, for any fixed xX and any given / > 0 we can choose  = x such
that x || x || < /. Then by (1) || Tnx – Tmx || < x|| x || < /. Hence ( Tnx )
is a Cauchy sequence in Y. However Y is complete. Then (Tnx ) converges,
say TnxyY. Define T : X  Y by Tx = y = limTnx. For any x, zX and any
scalar , T(x + z) = Lim Tn(x + z) =  Lim Tnx + Lim Tnz = Tx + Tz.
Hence T is linear. By letting m in (1) and using the continuity of the norm
we obtain, || (Tn – T)x || = || Tnx – Tx ||  || x || for all xX and all n > k …(2)
Hence (Tn – T) is bounded for any n > k. However, Tn is bounded then T = Tn
– (Tn – T) is bounded, that is TB( X, Y ). By (2) for all n > k, || Tn – T || =
sup{|| (Tn – T)x ||: xX, || x || = 1 }  . Hence Tn T. Therefore, B( X, Y ) is
complete
2.10-3 Definition. Let X be a normed space. Then the set of all bounded linear
functionals on X constitutes a normed space with norm defined by
| f ( x )|
|| f || = sup{ || x|| : xX, x  0 } = sup{ | f(x) |: xX, || x || = 1 } which is
called the dual space of X and is denoted by X/.
2.10-4 Theorem. The dual space X/ of a normed space X is a Banach space.
Examples.
2.10-5 Space Rn. The dual space of Rn is Rn.
Proof. Since Rn is of finite dimensional, then by Theorem 2.7-8, (Rn) / = (Rn)*
n n
so that every f(Rn)* can be represented as f(x) =   j j for any x=   j e j Rn,
j 1 j 1

where k = f(ek) and {e1, e2,…, en } is the standard basis for R . By Cauchy n
n n n n
Schwarz inequality | f(x) |   |  j | | j |  (  j2 ) (  k2 ) = || x || (  k2 ) 2 .
1 1 1
2 2

j 1 j 1 k 1 k 1
n
Then || f || = sup{ | f(x) |: xX, || x || = 1 }  (  k2 ) 2 .
1
However,
k 1

17
 n

 2

|| f ||  | f ( i ) | =
k n n
= (  k2 ) 2 . Hence || f || = (  k2 )
1 1
k 1 2 = || (k) ||.
|| ( i ) || n
(  )
1 k 1 k 1
2 2
k
k 1

Now, define T : (Rn) /  Rn by Tf = ( f(ek)) = ( k ). || Tf || = || ( k ) || = || f ||,

then T preserves the norm and so T is one to one. Now, for any f, g(Rn) /
and any scalar , T(f + g) = ((f + g )(ek)) = ((f )(ek)) + (g(ek)) = Tf + Tg.
Hence T is linear. However, dim Rn = dim (Rn)* = n, and since T is linear and
one to one, then T is onto. Therefore, T is an isomorphism. Hence the dual
space of Rn is Rn

1 1 
2.10-6 Space . The dual space of is .
1 1
Proof. Since (ek) is a Scauder basis for , where ek = (kj), then any x
 1 /
has a unique representation x =   k ek . Consider any f( ) , the dual space
k 1
1 
of . Since f is linear and bounded, then f(x) =  
k 1
k k
…………………(3),

where the numbers k = f(ek) are uniquely determined by f. Also || ek || = 1 and


| k | = |f(ek)|  || f || || ek || = || f ||, so that || f ||  sup |  k | . Hence ( k ) …(4)
k
 
Now, by (3) | f(x) |   | k | |  k | 
k 1
| 
k 1
k | sup |  j | = || x || sup |  j | . Then
j j

|| f ||  sup |  j | = || (k) ||. Use this and (4) to get || f || = sup |  j | = || (k )||.
j j

1 
Now, define T : ( ) / by Tf = (f(ek)) = (k). Then || Tf || = ||( k )|| = || f ||,
then T preserves the norm and so T is one to one. On the other hand for any
 1 /
b = (k) we can find g( ) by the following:
1  1
Define g: k ( complex or real field ) by g(x)=   k  k , for any x = (k) .
k 1
 
Since | g(x) |  | 
k 1
k | | k |  | 
k 1
k | sup |  j | = || x || sup |  j | < , then
j j

1
g(x)  k and g is bounded. For any x = (k), y = (k) and any scalar ,
  
g(x + y) =  (
k 1
k  k )  k =    k  k +
k 1
 
k 1
k k
= g(x) + g(y). Hence g is
1 / 
linear. Therefore, g( ) . However x can be written as x =  e
k 1
k k
, then

g(x) =   g (e ) . Then Tg = (g(ek)) = (k). Thus T is onto. Now, for any f,
k 1
k k

1
g( ) / and any scalar , T(f + g) = ((f + g )(ek)) = ((f )(ek)) + (g(ek)) =
Tf + Tg. Hence T is linear. Therefore, T is an isomorphism. Hence the dual
1 
space of is 

18
 p p q
2.10-7 Space . The dual space of is , where 1 < p <  and q is the
conjugate of p.
p p
Proof. Since (ek) is a Scauder basis for , where ek = (kj), then any x
 p
has a unique representation x =   k ek . Consider any f( ) /. Since f is linear
k 1

and bounded, then f(x) =  
k 1
k k
…………………………………………...(5),

where the numbers k =f(ek). Let q be the conjugate of p and consider xn= (  k( n ) )
 | |
if k  n and  k  0
q

 kk
with  k( n ) =  ………………………………...(6)

 0 if k  n or  k  0
 n n
Then by(5) f(xn) =  kn k =
k 1
 |  k |q = |  |  k |q | = |f(xn)|  || f || || xn || =
k 1 k 1
 n
|k | q
= || f || ( |
1 n n
|| f || ( | kn | p ) = || f || ( |  k |( q 1) p ) = || f || ( |  k |q ) p .
1 1 1
p
|p ) p p

k 1 k 1 k k 1 k 1
1
n n n 1
|  |q  || f || ( |  k |q ) . So that ( |  k |q )
1
Hence f(xn) = k
p p
 || f ||, then
k 1 k 1 k 1
n
( |  k |q )
1
q
 || f ||. Since n is arbitrary, then by letting n  we have,
k 1

( |  k |q )
1
q
 || f || ………………………………………………………….(7)
k 1
q p q
Hence ( k ) . Now, define T : ( ) / by Tf = (f(ek)) = (k). To show
q p
that T is onto, let b = (k) , define g: k ( complex or real field ) by
 p
g(x)=   k  k , for any x = (k) . then as in the previous example g is linear.
k 1
 
Moreover, by Holder inequality, since | g(x) |  ( | k | p ) ( |  k |q )
1 1
p q
=
k 1 k 1
 p 
|| x || ( |  k |q )   g (e ) .
1
q
then g is bounded. Hence g ( ) /, and so g(x) = k k
k 1 k 1

Then Tg = (g(ek)) = (k). Therefore, T is onto. By Holder inequality,

   
|f(x)| = |   k  k |  ( | k | p ) ( |  k |q ) = || x || ( |  k |q ) q , so that
1 1 1
p q

k 1 k 1 k 1 k 1
 
|| f ||  ( |  k |q ) q . Hence by (7), || f || = ( |  k |q ) q . Therefore, || Tf || =
1 1

k 1 k 1

||( k )|| = ( |  k |q )
1
q
= || f ||, then T preserves the norm and so T is one to
k 1

one. Finally, T is linear ( the proof is similar to that in the above example ).
p q
Therefore T is an isomorphism. Hence the dual space of is 

H. W. 2, 6, 8, 12, 13. H.W.* 6, 8, 13.

19