Week 3: Banach Spaces and Linear

Continuous Operators
Document prepared by Anna Rozanova-Pierrat
1
1 Lecture 3.2: Banach spaces
Denition 1 A normed vector space that is complete is called a Banach space.
Example 1 1. R
n
is a Banach space for any norm dened on it.
2. (C([a, b], | |
L
2) is not a Banach space.
3. (C([a, b], | |

) is a Banach space (|f|

= max
axb
[f(x)[).
Are L
p
Banach spaces? The answer is given in the Fischer-Riesz Theorem:
Theorem 1 (Fischer-Riesz) Let p [1, ], R
n
, () > 0. Then L
p
() is a Banach space.
Note that: () is the measure of .
Remark 1 The theorem can be reformulated in terms of Cauchy sequences as:
For p [1, ], let (f
k
)
kN
be a Cauchy sequence in L
p
(). Then there exists f L
p
() such that
lim
k
|f
k
f|
L
p
()
= 0.
Proof. Proof for p = . Let (f
n
) be a Cauchy sequence in L

(), thus we can write

|f
m
f
n
|
L

()
0 m, n ,
what means that
there exists
nm
of full measure ((
nm
) = ()) such that
sup
xnm
[f
m
(x) f
n
(x)[ 0 m, n .
We note

n,m

nm
, then (

) = 0 and

nm
for all n and m in N. Therefore,
sup
x

[f
m
(x) f
n
(x)[ 0 m, n .
Thus, (f
n
(x)) is a Cauchy sequence (in R) for all x

. Thanks to Cauchy criteria for numeric

sequences, we conclude that there exists
lim
n
f
n
(x) = f(x) x

,
i.e. almost everywhere on .
1
MAS, ECP
2 Week 3: Banach Spaces and Linear Continuous Operators
We need to verify that f L

() and that f
n
f for n in L

().
For any k in N

, there exists an integer N such that n > m > N implies

[f
m
(x) f
n
(x)[ < 1/k x

.
Taking m to the limit yields
[f(x) f
n
(x)[ < 1/k n > N(k) x

.
Therefore,
sup
x

[f(x) f
n
(x)[ < 1/k n > N(k),
i.e.
|f f
n
|
L
= inf

:(\

)=0
sup
x

[f(x) f
n
(x)[ 1/k n > N(k).
We summarize:
f f
n
L

n > N(k),
|f f
n
|
L
1/k n > N(k) f
n
f for n in L

().
Since L

is a linear vector space and in addition f

n
L

(), it follows that

f = f
n
+ (f f
n
) L

().
Proof for p [1, [
Let (f
n
) be a Cauchy sequence in L
p
(), i.e.
|f
m
f
n
|
L
p
()
0 m, n .
Step 1 There exists a subsequence (f
n
k
) such that
|f
n
k+1
f
n
k
|
L
p
()
<
1
2
k
.
Since (f
n
) is a Cauchy sequence, there exists N
1
N such that m > n > N
1
implies |f
n
f
m
|p <
1
2
.
Let n
1
= N
1
. There exists N
2
N such that m > n > N
2
implies |f
n
f
m
|p <
1
2
2
. Let n
2
=
maxn
1
, N
2
.
(f
n
k
) is constructed by induction.
Step 2 (f
n
k
) converges in L
p
: Let
g
m
=
m

k=1
[f
n
k+1
f
n
k
[.
We note that g
m
0. We have
|g
m
|
L
p = |
m

k=1
[f
n
k+1
f
n
k
[|
L
p
m

k=1
|f
n
k+1
f
n
k
|
L
p 1
1
2
m
< 1.
Then the theorem of Beppo-Levi (see Week 2) provides g L
p
such that g
m
(x) g(x) a.e. and
|g
m
g|
L
p 0.
Week 3: Banach Spaces and Linear Continuous Operators 3
For l and k with l > k, we have
[f
n
l
(x) f
n
k
(x)[ [f
n
l
(x) f
n
l1
(x)[ + . . . +[f
n
k
+1
(x) f
n
k
(x)[ g(x) g
k1
(x) (1)
Consequently (f
n
k
(x)) is a Cauchy sequence in R. Note f(x) its limit.
Taking l to the limit gives
[f(x) f
n
k
(x)[ g(x) g
k1
(x) g(x)
Hence:
[f(x) f
n
k
(x)[
p
0 for a.e.x,
[f(x) f
n
k
(x)[
p
g(x)
p
.
From the Dominated Convergence Theorem (see Week 2) we derive
f = f
n
k
+ (f f
n
k
) L
p
and |f f
n
k
|
p
0.
Let > 0 be arbitrary and n N. Let m > n be such that m = n
k
for a given k. Since (f
n
) is a
Cauchy sequence, |f
n
f
m
|
L
p < /2. Since (f
n
k
) converges toward f, |f
m
f|
L
p < /2. Hence
|f
n
f|
L
p |f
n
f
m
|
L
p +|f
m
f|
L
p < .
Thus (f
n
) converges toward f in L
p
.
Problem 1 Prove that the space C([0, 2]) of continuous function on [0, 2] equiped with the norm
|f| =
__
2
0
[f(t)[
2
dt
_
1
2
is not a Banach space.
Indication: Consider the limit of the sequence (f
n
) of elements of C([0, 2]) dened by
f
n
(t) =
_

_
0, 0 t 1
1
n
1 + n(t 1), 1
1
n
t 1
1, 1 t 2
.
Problem 2 Let

k=1
x
k
be a series of elements x
k
of a normed space (E, | |) and S
n
=

n
k=1
x
k
be its partial sum.
We say that
1.

k=1
x
k
is convergent in E if there exists x E such that S
n
x for n in E:
|S
n
x|
E
0 n .
2.

k=1
x
k
is absolutely convergent in E if

k=1
|x
k
|
E
< .
Prove the following theorem:
Theorem 2 The normed space (E, ||
E
) is complete if and only if every absolutely convergent series
in E,
i.e. x
n
E :

n
|x
n
|
E
< ,
converges in E.
4 Week 3: Banach Spaces and Linear Continuous Operators
Indication:
1. If E is complete and x
n
is absolutely convergent, show that the sequence of partial sums S
n
=

n
k=1
x
k
is a Cauchy sequence.
2. To show that if every absolutely convergent series x
n
E converges in E then E is complete,
rst, we see that (x
n
) is a Cauchy sequence, and then extract for a xed a subsequence (x
n
k
)
such that
|x
n
k
x
n
k+1
|

2
k
.
We show, thanks to the absolute convergence of (x
np
), that the partial sums of the sequence
y
1
= x
n
1
,
y
2
= x
n
2
x
n
1
,
. . . . . . . . .
y
p
= x
np
x
n
p1
,
. . . . . . . . .
converge in E and consequently (x
np
) converges in E.
To conclude, we use the fact that if (x
n
) is a Cauchy sequence containing a convergent (in E)
subsequence, then the sequence converges (in E).
Remark 2 We note that we can use Theorem 2 to prove the case p = in Fischer-Riesz Theorem:
Let f
n
be an absolutely convergent series.We dene

n
= x /f
n
(x) < |f
n
|
of full measure and take

n
.
By Week 2,
(/

) =
_
n
(/
n
) = 0,
and

n
f
n
(x) converges on

since it is absolutely convergent in R.

2 Lecture 3.2: Linear operators
Denition 2 Let (X, | |) and (Y, | |) be two normed vector spaces. Let D be a vector subspace of
X.
A mapping f : D X Y is called a linear operator (or a linear function, or a linear mapping)
if for all u and v in D and all in R
f(u + v) = f(u) + f(v)
f(u) = f(u)
D is called the domain of f. We say that f is densely dened if D is dense in X.
Week 3: Banach Spaces and Linear Continuous Operators 5
1 0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Figure 1 Graphe of the function g(t) = e
1/(1t
2
)
if [t[ < 1 and 0 if [t[ 1.
Example 2 Let X = L
1
(R) and Y = L

(R).
Dene g(t) = e
1/(1t
2
)
if [t[ < 1 and 0 if [t[ 1 (see Fig 1). The function g is in C

We dene the convolution operator

F : X Y
f
R R
x
_
R
f(y)g(x y)dy
,
which can briey be written as
F(f) = (f g)(x) =
_
R
g(x y)f(y)dy. (2)
F is a linear form or a linear functional.
Let us justify that f g L

(R).
It is easy to see that g C

(R) and bounded. Thus, g L

(R). Therefore,
[(f g)(x)[
_
R
[g(x y)[[f(y)[dy max
xR
[g(x)[
_
R
[f(y)[dy |g|
L
|f|
L
1.
Denition 3 A set X is bounded if there exists r > 0, such that X B
r
(0).
Let X and Y be two normed vector spaces. Let f : D X Y be a linear operator. We say that f
is bounded if the image of any bounded set in X is a bounded set in Y .
Remark 3 Linear operator A : X Y is bounded if and only if A is bounded on a non-trivial ball
centered in 0 in X : B
r
(0) = x X[|x| r.
Proof. If A is bounded, it is obvious that AB
r
(0) is bounded for all r > 0. (Note that AB
r
(0)
means A applied to B
r
(0)).
Suppose that A is bounded on B
r
(0) for a xed r ,= 0. We proceed in two steps:
Let R > 0 and R ,= r. For B
R
(0) = B
r
(0) with =
R
r
, we see that AB
R
(0) is bounded.
Let M be a bounded set in X. Thus
R > 0 : M B
R
(0)
and therefore, AM AB
R
(0). Since AB
R
(0) is bounded, it implies that AM is bounded too.
.
6 Week 3: Banach Spaces and Linear Continuous Operators
Proposition 1 Let X and Y be two normed vector spaces. For a linear operator A : X Y the
following assertions are equivalent:
1. A is continuous,
2. A is bounded,
3. C 0 : |Ax|
Y
C|x|
X
x X.
Proof. (1) (2) Let A be continuous. Thus, in particular, A is continuous in 0:
> 0 = () > 0 : |x|
X
|Ax|
Y
< ,
where we have used A0 = 0, since A is linear. Thus, A is bounded on B

(0) and hence A is bounded.

(2) (3) Let A be bounded. If x = 0 (3) is obvious. Let x ,= 0. We normalize it by introducing
x
0
=
x
|x|
X
|x
0
|
X
= 1 x
0
B
1
(0).
Since B
1
(0) is bounded, AB
1
(0) is bounded too. Consequently,
C 0 : |Ax
0
|
Y
C,
or equivalently, by the linearity of A and the norm,
|Ax
0
|
Y
=
_
_
_
_
_
A
_
x
|x|
X
__
_
_
_
_
Y
=
_
_
_
_
_
1
|x|
X
Ax
_
_
_
_
_
Y
=
|Ax|
Y
|x|
X
C,
which gives (3).
(3) (1) Let x
n
x for n in X . We have, due to the linearity of A, that
|Ax
n
Ax|
Y
= |A(x
n
x)|
Y
C|x
n
x|
X
0 n ,
what implies that |Ax
n
Ax|
Y
0 for n , i.e. Ax
n
Ax for n in Y .
Example 3 (linear bounded operators, Fredholm operators)
1. Let X = Y = C([a, b]) with its usual norm: |f|
C([a,b]
= max
axb
[f(x)[. We dene the
operator A by the formula:
f C([a, b])
_
t y(t) =
_
b
a
K(t, s)f(s)ds C([a, b])
_
where Af = y and K, called kernel of A, is a xed continuous function on [a, b] [a, b]. The
operator A is called the Fredholm operator.
By its denition A is linear. Let us show that A is bounded, or equivalently, continuous:
[y(t)[
_
b
a
[K(t, s)f(s)[ds M(b a)|f|
C([a,b]
t [a, b],
where we have used
As K is a continuous function on the compact [a, b] [a, b], K is bounded on [a, b] [a, b]:
M > 0 [K(t, s)[ M (t, s) [a, b] [a, b].
Week 3: Banach Spaces and Linear Continuous Operators 7
[f(s)[ |f|
C([a,b]
s [a, b].
Thus, we obtain for C = M(b a) that
|y|
C([a,b])
C|f|
C([a,b])
.
2. Let X = Y = L
2
(]a, b[). We dene the Fredholm operator A such that
y(t) = A(f(s)) =
_
b
a
K(t, s)f(s)ds L
2
(]a, b[),
where the kernel K L
2
(]a, b[]a, b[):
_
]a,b[]a,b[
[K(t, s)[
2
dt ds = M
2
< .
Let us show that A is bounded, or equivalently, continuous. We use the Cauchy-Schwartz
inequality (or the Hlder inequality for p = 2):
[y(t)[
_
b
a
[K(t, s)[[f(s)[ds

_
b
a
[K(t, s)[
2
ds

_
b
a
[f(s)[
2
ds.
Therefore
[y(t)[
2

_
b
a
[K(t, s)[
2
ds|f|
2
L
2,
and integrating the last inequality over ]a, b[ on t, we obtain
|y|
2
L
2
_
b
a
_
b
a
[K(t, s)[
2
dsdt|f|
2
L
2.
Since |K|
L
2
(]a,b[]a,b[)
= M, we conclude that
|y|
L
2 M|f|
L
2.
Example 4 (unbounded operator) Let us consider A =
d
dt
: D L
2
(]0, 2[) L
2
(]0, 2[) with
domain D = C
1
(]0, 2[). We note that A is not dened for all f L
2
(]0, 2[), but its domain D is
dense in L
2
(]0, 2[). Let us take x
n
D such that
x
n
(t) =
1

2
e
int
, n Z.
Then
|x
n
|
L
2
(]0,2[)
=

_
_
]0,2[

2
e
int

2
d = 1, x
n
B
1
(0),
but
|Ax
n
|
L
2
(]0,2[)
= [n[|x
n
|
L
2
(]0,2[)
= [n[ n .
Therefore, A is not bounded.
Actually, what is a norm of an operator? We will dene it in the next lecture.
8 Week 3: Banach Spaces and Linear Continuous Operators
3 Lecture 3.4: /(X, Y ): The space of linear continuous op-
erators
Proposition 2 Let X and Y be two normed vector spaces. For a linear operator A : X Y we
dene:
= sup
x=0
|Ax|
Y
|x|
X
, = sup
x
X
1
|Ax|
Y
, = sup
x
X
=1
|Ax|
Y
,
= infC R
+
[|Ax|
Y
C|x|
X
x X.
Then = = = .
Proof. Let M
r
= sup
x
X
=r
|Ax|
Y
. Then M
1
= and M
r
= rM
1
= r.
1. = :
= sup
r>0
sup
x
X
=r
|Ax|
Y
|x|
X
= sup
r>0
M
r
r
= M
1
= .
2. = :
= sup
0<r1
M
r
= M
1
sup
0<r1
r = M
1
= .
3. = : Let (x) =
Ax
Y
x
X
. As 0, by denition of the supremum, we have
= sup
x=0
(x) = infC R
+
[(x) C = .
Corollary 1 An linear operator A is bounded if and only if one of , , or is nite.
Denition 4 Let X and Y be two normed vector spaces.
The set of linear and continuous operators from X to Y is denoted /(X, Y ).
It is easy to verify that /(X, Y ) is a linear vector space.
Denition 5 Let A /(X, Y ). The norm of A in /(X, Y ) is dened by
|A|
L(X,Y )
= sup
x=0
|Ax|
Y
|x|
X
= sup
x
X
1
|Ax|
Y
= sup
x
X
=1
|Ax|
Y
= infC R
+
[|Ax|
Y
C|x|
X
x X.
Therefore, (/(X, Y ), | |
L(X,Y )
) is a normed vector space.
Problem 3 Prove that | |
L(X,Y )
is a norm in /(X, Y ).
Remark 4 /(X, X) is a normed algebra: for a product AB of linear continuous operators, dened
as (AB)x = A(Bx), we have |AB| |A||B| (if A and B are linear continuous, then AB is linear
continuous as a composition of two linear continuous mappings).
Remark 5 In a nite dimensional space all linear operators are continuous!
Theorem 3 If Y is a Banach space, then /(X, Y ) is a Banach space.
Proof. Let (A
n
) be a Cauchy sequence in /(X, Y ), i.e.,
> 0 N() > 0 m, n N() |A
n
A
m
|
L(X,Y )
< . (3)
Week 3: Banach Spaces and Linear Continuous Operators 9
We need to show that there exists A /(X, Y ) such that A
n
A for n in /(X, Y ).
Let us use for the denition of the norm:
|A
n
A
m
|
L(X,Y )
< sup
x
X
=1
|A
n
x A
m
x|
Y
.
Moreover, we have for all m, n N()
sup
x
X
=1
|A
n
x A
m
x|
Y
|A
n
x A
m
x|
Y
x : |x|
X
= 1, (4)
in other words, (A
n
x) is a Cauchy sequence in Y . Y is a Banach space, thus (A
n
x) is convergent in
Y : there exists an element y Y such that y = lim
n
A
n
x. Let us note it as Ax: y = Ax. Then
we obtain:
1. A is linear: the limit lim
n
is linear and A
n
are linear for all n.
2. A
n
A for n in /(X, Y ): in (4) we x x and n and pass to m , thus we nd with
the limit that
n N() |(A
n
A)x|
Y
= |A
n
x Ax| for |x| = 1.
Taking a supremum on x (note that N does not depend on x!) we obtain
|A
n
A|
L(X,Y )
< .
3. A /(X, Y ): as |A
n
A|
L(X,Y )
< for n N() then A
n
A is bounded for n N(),
therefore, A = A
n
+ (AA
n
) /(X, Y ) as a sum of linear bounded operators.
Corollary 2 If X is a Banach space, then /(X, X) is complete.
We give now the BLT-theorem (see H. Brezis Functional Analysis, Sobolev Spaces and Partial
Dierential Equations):
Theorem 4 (BLT: Bounded Linear Transformation) Let X be a normed vector space, D be
a dense subspace of X and Y be a Banach space. Let f : D X Y be a densely-dened linear
continuous operator. There exists a unique continuous linear operator F : X Y that extends f
and that has the same norm.
3.1 Bounded operators and theorem of Banach-Steinhaus
We give the following notions of bounded operators, using in Section 4.3.
Denition 6 Let X and Y be two normed spaces. A set of bounded linear operators A
i

iI

/(X, Y ) is said to be bounded on a vector x X, if
|A
i
x| C i I, where C = C(x) 0.
Denition 7 A set of linear bounded operators A
i

iI
/(X, Y ) is uniformly bounded if
|A
i
|
L(X,Y )
C i I.
10 Week 3: Banach Spaces and Linear Continuous Operators
We notice that if A
i

iI
/(X, Y ) is uniformly bounded then
|A
i
x|
Y
C|x|
X
x X i I.
Theorem 5 (Banach-Steinhaus) Let X be a Banach space and Y be a normed space. Let
A
i

iI
/(X, Y ) be a set of linear bounded operators. A
i

iI
is uniformly bounded if and only if
A
i

iI
is bounded on all vectors x from X.
Remark 6 The proof of the Banach-Steinhaus theorem is based on the Baire theorem (see Week 1
for the denition of a nowhere dense set):
Theorem 6 (Baire) A complete metric space E cannot be represented as the union of a countable
number of nowhere dense sets.
4 Lecture 3.5: Dual space and weak topology
4.1 Denition
Denition 8 A linear mapping f of a normed space X to R is called functional or a linear form
on X.
Denition 9 We note X

= /(X, R) the space of linear continuous functionals on X. It is called

the dual space of X.
The norm on X

is dened by
f X

|f| = sup
x=0
[f(x)[
|x|
.
Remark 7 X

is always a Banach space (whether X is a Banach space or not).

Denition 10 We denote the value of f X

on x X by < f, x >= f(x).

4.2 Hahn-Banach theorem and its corollaries
We give without proof the following theorem (see H. Brezis Functional Analysis, Sobolev Spaces
and Partial Dierential Equations):
Theorem 7 (Hahn-Banach) Let X be a real vector space and : X R
+
be a semi-norm. In
addition, let L be a subspace of X and l : L R be a functional on L such that
l(x) (x) x L.
Then there exists a linear functional : X R such that
1. (x) = l(x) x L ([
L
= l)
2. (x) (x) x X.
There is a direct corollary of the Hahn-Banach theorem:
Week 3: Banach Spaces and Linear Continuous Operators 11
Corollary 3 Given a real normed linear space X, let L be a subspace of X and l a bounded linear
functional on L. Then l can be extended to a bounded linear functional on the whole space X
without increasing its norm, |l|
L(L,R)
= ||
L(X,R)
.
In what follows we use three corollaries from the Hahn-Banach theorem:
Corollary 4 Let X be a normed vector space, x X and x ,= 0. Then there exists a linear
continuous functional f X

such that
|f|
L(X,R)
= 1 and < f, x >= |x|.
Proof. We apply Corollary 3 of the Hahn-Banach theorem for L = tx[t R X and f
0
L

dened as
f
0
(tx) = t|x|.
We notice that
f
0
(x) = |x|
and if y = tx then
[f
0
(y)[ = [t[|x| = |tx| = |y| |f
0
|
L(L,R)
= 1.
Therefore, by Corollary 3 of the Hahn-Banach theorem, there exists f X

such that f(x) = |x|

and |f| = 1.
Corollary 5 Let X be a normed vector space, L be a subspace of X and x
0
/ L such that d(L, x
0
) =
d > 0. Then there exists f X

such that
1. f(x) = 0 x L,
2. f(x
0
) = 1,
3. |f| =
1
d
.
Proof. Let us take L
1
= L+ < x
0
>, where < x
0
> is a linear space constructed on x
0
taking all its
linear combinations. Thus,
y L
1
!x L and !t R : y = x + tx
0
y L
1
.
We dene f
0
L
1

by the formula:
f
0
(y) = t.
Therefore, if y L it follows that f
0
(y) = 0 (thus (1) ), and we also have f
0
(x
0
) = 1 (thus (2) ).
Let us show that |f
0
| =
1
d
. On the one hand,
[f
0
(y)[ = [t[ =
[t[|y|
|y|
=
|y|
_
_
_
x
t
+ x
0
_
_
_

|y|
d
,
since _
_
_
_
x
t
+ x
0
_
_
_
_
= |x
0

_

x
t
_
| d (as
x
t
L).
Hence, we nd that |f
0
|
1
d
.
On the other hand, we show that |f
0
|
1
d
.
Since d = inf
xL
|x
0
x|, it follows that
(x
n
) L such that d = lim
n
|x
0
x
n
|.
12 Week 3: Banach Spaces and Linear Continuous Operators
As
1 = f
0
(x
0
x
n
) |f
0
||x
0
x
n
|,
we obtain for n that |f
0
|
1
d
.
From |f
0
|
1
d
and |f
0
|
1
d
we conclude |f
0
| =
1
d
. Thanks to Corollary 3 of the Hahn-Banach
theorem, we expand f
0
to f X

which satises all three conditions.

Corollary 6 Let X be a Banach space, L be a subspace of X. L is not dense in X if and only if
f X

f ,= 0 such that f(x) = 0 x L.

Proof. Let L be not dense in X:
L ,= X.
Thus, there exists x
0
X such that d(x
0
, L) = d > 0. Applying Corollary 5,
f X

: f(x
0
) = 1 (f ,= 0) and f(x) = 0 x L.
Let L = X, i.e.,
x X (x
n
) L : x
n
x n .
Then, by the assumption,
f X

(f ,= 0) such that f(y) = 0 y L,

from where we nd, by the continuity of f, that
x X f(x) = lim
n
f(x
n
) = 0 f = 0.
This is the contradiction with f ,= 0. Therefore, L ,= X.
Remark 8 < f, x >= 0 x X implies f = 0.
< f, x >= 0 f X

implies x = 0.
Proof Lets assume the converse. If x ,= 0, by Corollary 4, there exists f X

such that f ,= 0
and < f, x >= |x| , = 0. It is the contradiction.
Remark 9 Can we consider the notation < f, x > as an inner product between elements of two
spaces X

and X? It is obvious bilinear and continuous with respect to x X. Is it continuous with

respect to f too?
4.3 Weak topology
Denition 11 Let X be a normed vector space. Let X

= /(X, R) be its dual space.

The weak topology on X is the nal topology with respect to X

. It is noted (X, X

).
The weak topology has fewer open sets compared to the topology derived from the norm on X. We
note that if X is a normed space, then (X, (X, X

)) is a Hausdor space (see Week 1).

Week 3: Banach Spaces and Linear Continuous Operators 13
Denition 12 Let (x
n
) be a sequence of elements of X. We say that (x
n
) converges weakly to x,
noted x
n
x, if (x
n
) converges to x in (X, X

).
Remember Week 1? These statements are equivalent:
For all f in X

, f(x
n
) converges to f(x)
(x
n
) converges to x in (X, X

) .
Denition 13 The convergence in the initial topology (dened by the norm in X) is called strong
convergence:
(x
n
) (X, | |
X
), |x
n
x|
X
0 x
n
x strongly.
There are fewer open sets in the weak topology. Hence, if a sequence strongly converges, then it
weakly converges.
Proposition 3 1. Weak limit is unique.
2. If x
n
x, n strongly in X, then x
n
x, n weakly in X (the converse is false).
3. If (x
k
) converges weakly to x, then (x
k
) is bounded:
C > 0 : |x
n
|
X
< C and |x| liminf |x
n
|
X
,
where liminf |x
n
|
X
= lim
n
(inf
mn
x
m
).
4. x
n
x
0

(a) (|x
n
|) is bounded,
(b) < f, x
n
>< f, x
0
> f E, E = X

5. If x
n
x
0
in X and if f
n
f strongly in X

(i.e. |f
n
f|
L(X,R)
0), then
< f
n
, x
n
>< f, x > .
Proof.
1. Let (x
n
) be a sequence in X such that
x
n
x and x
n
x n .
Then for all f X

, since f is continuous,
< f, x >=< f, x > .
Therefore, since f is linear,
< f, x x >= 0 x = x.
2. We use the estimate:
[ < f, x
n
> < f, x > [ = [ < f, x
n
x > [ |f|
L(X,R)
|x
n
x|
X
.
We will show that the converse is false: see Example 5 and a counter example in Week 4.
14 Week 3: Banach Spaces and Linear Continuous Operators
3. The proof needs to consider the sequence (x
k
) as a sequence in X

and to apply the theorem

of Banach-Steinhaus.
4. It is the corollary of the theorem of Banach-Steinhaus.
5. It is the corollary of the following estimation:
[ < f
n
, x
n
> < f, x > [ [ < f
n
f, x
n
> [ +[ < f, x
n
x > [
|f
n
f||x
n
| +[ < f, x
n
x > [ = |f
n
f||x
n
| +[ < f, x
n
> < f, x > [.
Theorem 8 If dimX = m < , X is a normed space and x
n
x
0
weakly in X, then x
n
x
0
strongly in X.
Proof. Let e
n

m
i=1
be a basis of X. Then
x
n
=
m

i=1

(n)
i
e
i
, x
0
=
m

i=1

(0)
i
e
i
.
We dene f
i
X

such that
< f
i
, e
j
>=
i,j
=
_
1 i = j
0 i ,= j
.
Therefore,

(n)
i
=< f
i
, x
n
>< f
i
, x
0
>=
0
i
n ,
and consequently,
x
n
= (
(n)
1
, . . . ,
(n)
m
) x
0
= (
0
1
, . . . ,
0
m
) n ,
what implies that x
n
x
0
strongly in X. (As all norms are equivalent in X, it is sucient to show
that x
n
x
0
by | |
l
).
Example 5 Consider the space C([a, b]) of all functions continuous on [a, b] equipped with the norm
|f|

= max
axb
[f(x)[.
If |f
n
f|

0 in (C([a, b]), it follows that the sequence (f

n
) converges uniformly to f on [a, b].
Thus the strong convergence in C([a, b]) means the uniform convergence.
Let us now consider the weak convergence in C([a, b]). Let (f
n
) be a sequence of functions in C([a, b])
converging weakly to a function f C([a, b]). Among the continuous linear functionals on C([a, b]),
we have the functionals
x
0
, a < x
0
< b, named the Dirac delta functions, which assign to each
function f(x) C([a, b]) its value at some xed point x
0
[a, b].
Let us show that
x
0
C

([a, b]). Indeed, by denition

x
0
is linear mapping from C([a, b]) to R:

x
0
(f + g) = f(x
0
) + g(x
0
) =
x
0
(f) +
x
0
(g) f, g C([a, b]), , R,
and in addition we have
[
x
0
(f)[ = [f(x
0
)[ max
axb
[f(x)[ = |f|

,
where equality holds if f(x) = const. Hence
x
0
is bounded, thus it is continuous, with norm
|
x
0
|
L(C([a,b]),R)
= 1.
From

x
0
(f
n
)
x
0
(f)
Week 3: Banach Spaces and Linear Continuous Operators 15
follows by the denition of the functional
x
0
that
f
n
(x
0
) f(x
0
).
Hence, if the sequence (f
n
) is weakly convergent in C([a, b]), then
1. (f
n
) is uniformly bounded on [a, b], i.e., there is a constant C 0 such that
[f
n
(x)[ C n N x [a, b],
2. (f
n
) is pointwise convergent on [a, b], i.e., (f
n
(x)) is a convergent numerical sequence for every
xed x [a, b].
We can see that the strong convergence in C([a, b]) implies the weak convergence, but not the converse.
5 Lecture 3.6: Reexivity
Denition 14 The bidual space of X is noted by X

and dened by
X

= (X

.
It is the normed space of a linear continuous functional from X

to R with the norm:

|F|
L(X

,R)
= sup
fX

,f
L(X,R)
1
[ < F, f > [.
Let us x x X. Then for f X

the mapping F
x
: X

R such that f < f, x > is a linear

continuous functional on X

.
There is a natural injection from X to X

. Given x X, associate (x) = F

x
X

dened by
a linear continuous functional on X

: F
x
: f X

< f, x > R. Thus we have the equality:

< F
x
, f >
X

,X
=< f, x >
X

,X
x X, f X

.
The notation < x, y >
X,Y
means that x X, y Y and < x, y >= x(y).
Proposition 4 Let X be a normed space. Then the natural injection : X X

is a linear
isometric (thus continuous) operator.
Proof. The linearity is obvious. Let us prove that
|F
x
|
X
= |x|
X
.
We nd that
|F
x
|
X
= sup
f1
[ < F
x
, f > [ = sup
f1
[ < f, x > [ = |x|
X
.
We need to justify that sup
f1
[ < f, x > [ = |x|
X
.
Let x ,= 0. We see that
sup
f1
[ < f, x > [ |x|
X
sup
x=0
[ < f, x > [
|x|
X
1 [ < f, x > [ |x|
X
.
16 Week 3: Banach Spaces and Linear Continuous Operators
In addition, thanks to Corollary 4,
x
0
X f
0
X

: |f
0
| = 1 < f
0
, x
0
>= |x
0
|.
Thus, sup
f1
[ < f, x > [ = |x|
X
.
As is isometric, it means that X (X) X

, where (X) is a subspace of X

. If is
surjective (and then bijective), X and X

can be identied.
Denition 15 We say that X is reexive if (X) = X

. In this case we write X = X

under-
standing the isometric equivalence.
Example 6 1. R
n
is a reexive space.
2. l
p
and L
p
for 1 < p < are reexive spaces (see Week 6 for the proof).
3. Let c
0
be the space of all sequences x = (x
1
, . . . , x
k
, . . .) converging to zero, with the norm (see
Week 2)
|x| = sup
k
[x
k
[.
Then the space c

0
is isomorphic to the space l
1
of all absolutely summable sequences. So c

0
= l
1
,
and l

1
= l

. Therefore, c
0
,= c

0
.
4. L

and L
1
are not reexive spaces (see Week 6 for the proof).
Next week we will dene the weak topology and will discuss the compactness in the strong, weak
and weak topology. We need these notions to be able to prove the following theorems (see H. Brezis
Functional Analysis, Sobolev Spaces and Partial Dierential Equations for the proof):
Theorem 9 Let X be a Banach space. X is reexive if and only if X

is reexive.
Theorem 10 Let X be a Banach space. X is reexive if and only if each bounded sequence in X
contains a subsequence which converges weakly in X.
Theorem 11 (Kakutani) Let X be a Banach space. X is reexive i the closed unit ball associated
to the norm x X, |x| 1 is compact in the weak topology (X, X

) .
Denition 16 A Banach space X is uniformly convex if for all > 0, there exists > 0 such that
for all x X and y X, |x| 1, |y| 1 and |x y| > implies |x + y|/2 < 1 .
Theorem 12 (Milman-Pettis) Let X be a uniformly convex Banach space. Then X is reexive.