Vous êtes sur la page 1sur 28

Week 2: Metrics and normed spaces

Document prepared by Anna Rozanova-Pierrat


1
1 Lecture 2.2: Distance function
1.1 Denitions and examples
Denition 1 Let E be a set and d : E E R be a function. d is a distance on E if:
1. (x, y) E E, d(x, y) 0;
2. (x, y) E E, d(x, y) = 0 if and only if x = y;
3. (x, y) E E, d(x, y) = d(y, x);
4. (x, y, z) E E E, d(x, y) d(x, z) +d(z, y) (triangular inequality).
If point 2 does not hold, d is called a pseudodistance. If point 3 does not hold, d is called a
quasidistance.
Denition 2 Set E with a given distance dened on it, i.e. the pair (E, d), is a metric space. If
d is a pseudodistance on E, then (E, d) is a pseudometric space. If d is a quasidistance on E,
then (E, d) is a quasimetric space.
Example 1 Setting
d(x, y) =
_
0 if x = y,
1 if x ,= y,
where x and y are elements of an arbitrary set E, we obtain a metric space (E, d) (a discrete space
or space of isolated points). Indeed, by denition of d, the rst three points of Denition are satised.
For the last point we have
1. If x = y the triangle inequality becomes:
0 2 if z ,= x or 0 0 if z = x.
2. If x ,= y the triangle inequality becomes:
1 2 if z ,= x and z ,= y or 1 1 if z = x or z = y.
Therefore the triangle inequality is satised.
Example 2 Consider the space C([0, 1]) of all continuous functions on [0, 1]. Let us verify that
d(f, g) = max
x[0,1]
[f(x) g(x)[
is a distance on C([0, 1]):
1
MAS, ECP
2 Week 2: Metrics and normed spaces
1. As the modulus is a positive function on R:
x R [x[ 0,
we have that
(f, g) C([0, 1]) C([0, 1]), d(f, g) 0.
2. We have for all (f, g) C([0, 1]) C([0, 1])
d(f, g) = 0 max
x[0,1]
[f(x) g(x)[ = 0 x [0, 1] 0 [f(x) g(x)[ 0
x [0, 1] [f(x) g(x)[ = 0 x [0, 1] f(x) = g(x).
3. For all (f, g) C([0, 1]) C([0, 1]) we have
d(f, g) = max
x[0,1]
[f(x) g(x)[ = max
x[0,1]
[ [f(x) g(x)][ = max
x[0,1]
[g(x) f(x)[ = d(g, f).
4. For all (f, g, h) C([0, 1]) C([0, 1]) C([0, 1]) we have
d(f, g) = max
x[0,1]
[f(x) g(x)[ = max
x[0,1]
[f(x) h(x) +h(x) g(x)[
max
x[0,1]
([f(x) h(x)[ +[h(x) g(x)[) max
x[0,1]
[f(x) h(x)[ + max
x[0,1]
[h(x) g(x)[
= d(f, h) +d(h, g).
We conclude that (C([0, 1]), d) is a metric space.
Example 3 Consider E = R
n
with n N

and p [1, [. Lets dene for x = (x


1
, . . . , x
n
) E
and y = (y
1
, . . . , y
n
) E
d
p
(x, y) =
_
n

i=1
[x
i
y
i
[
p
_1
p
and d

(x, y) = max
i[1,...,n]
[x
i
y
i
[.
1. We can easily see that d

satises the assertions of Denition 1 and hence, d

is a metric in
R
n
.
2. Let us prove that d
p
is a metric. It is obvious that points 1-3 are true for d
p
for all p [1, [.
We need to prove the triangle inequality 4.
Let x, y, z be three points in R
n
and let A = x z, B = z y. Then x y = A + B and the
triangle inequality 4 takes the form of Minkowskis inequality
_
n

i=1
[A
i
+B
i
[
p
_1
p

_
n

i=1
[A
i
[
p
_1
p
+
_
n

i=1
[B
i
[
p
_1
p
. (1)
The inequality is obvious for p = 1. Suppose that p > 1. To prove Minkowskis inequality (1),
we use Hlders inequality (see Lemma 2 for the proof):
n

i=1
[A
i
B
i
[
_
n

i=1
[A
i
[
p
_1
p
_
n

i=1
[B
i
[
p

_ 1
p

, (2)
where
1
p
+
1
p

= 1, and the following indentity for any a and b in R or C:


([a[ +[b[)
p
= [a[([a[ +[b[)
p1
+[b[([a[ +[b[)
p1
.
Week 2: Metrics and normed spaces 3
Thus, we can write
n

i=1
([A
i
[ +[B
i
[)
p
=
n

i=1
[A
i
[([A
i
[ +[B
i
[)
p1
+
n

i=1
[B
i
[([A
i
[ +[B
i
[)
p1
.
We apply Hlders inequality to each sum in the right-hand part of the equality, using the fact
that (p 1)p

= p:
n

i=1
[A
i
[([A
i
[+[B
i
[)
p1

_
n

i=1
[A
i
[
p
_1
p
_
n

i=1
([A
i
[ +[B
i
[)
(p1)p

_ 1
p

=
_
n

i=1
[A
i
[
p
_1
p
_
n

i=1
([A
i
[ +[B
i
[)
p
_ 1
p

,
from where with
n

i=1
[B
i
[([A
i
[ +[B
i
[)
p1

_
n

i=1
[B
i
[
p
_1
p
_
n

i=1
([A
i
[ +[B
i
[)
p
_ 1
p

,
we nd that
n

i=1
([A
i
[ +[B
i
[)
p

_
n

i=1
([A
i
[ +[B
i
[)
p
_ 1
p

_
_
_
n

i=1
[A
i
[
p
_1
p
+
_
n

i=1
[B
i
[
p
_1
p
_
_
.
Dividing both sides of this inequality by (

n
i=1
([A
i
[ +[B
i
[)
p
)
1
p

, and noticing that 1


1
p

=
1
p
we
nally obtain that
_
n

i=1
([A
i
[ +[B
i
[)
p
_1
p

_
n

i=1
[A
i
[
p
_1
p
+
_
n

i=1
[B
i
[
p
_1
p
.
To nish the proof, we use the fact that [A
i
+ B
i
[ [A
i
[ + [B
i
[ for all i and consequently we
have (1).
3. Consider E = R
n
with the metric
d
2
(x, y) =
_
n

i=1
(x
i
y
i
)
2
_1
2
.
This metric is the Euclidean distance function.
4. Consider E = functions from R to R dened in 0. For f and g in E, we dene
d(f, g) = [g(0) f(0)[.
Let us prove that d is a pseudodistance. For f(x) = x
2
and g(x) = x
3
(f ,= g), by denition of
d, d(f, g) = 0, so point 2 of Denition 1 does not hold. Points 1, 3 and 4 are true thanks to
the properties of the modulus. Hence we conclude that d is a pseudodistance.
Let us prove the inequality of Hlder. Firstly we prove the following technical Lemma:
Lemma 1 Let : [0, [[0, [ be a continuous strictly increasing function. Then there exists the
inverse function
1
and then for all positive a and b it holds
ab
_
a
0
(x)dx +
_
b
0

1
(y)dy. (3)
The equality in (3) takes place if and only if b = (a).
4 Week 2: Metrics and normed spaces
0 0
a a
b
b
y = (x) y = (x)
S
1
S
1
S
2
S
2
Figure 1 For the left-hand gure b < (a) and for the right-hand gure b > (a). Clearly S < S
1
+ S
2
.
0
a
b
y = (x)
S
1
S
2
Figure 2 b = (a). Clearly S = S
1
+ S
2
.
0
A
b
y = (x)
x =
1
(y)
Figure 3 An example of
_
b
0

1
(y)dy = for b > A.
Proof. The proof is based on the geometric sense of the integral as a area of the subgraph and on
the three following gures, where S
1
=
_
a
0
(x)dx, S
2
=
_
b
0

1
(y)dy, and S = ab is the area of the
rectangle.
Remark 1 If A = sup
x[0,[
(x) < (see Fig. 3), then for b > A we have
_
b
0

1
(y)dy = what
does not contradict (3).
Lemma 2 (inequality of Hlder) Let p ]1, [ and
1
p
+
1
p

= 1. For all x = (x
1
, . . . , x
n
) R
n
and y = (y
1
, . . . , y
n
) R
n
it holds
n

i=1
[x
i
y
i
[
_
n

i=1
[x
i
[
p
_1
p
_
n

i=1
[y
i
[
p

_ 1
p

. (4)
Proof. The proof of Hlders inequality is based on the following fact:
Week 2: Metrics and normed spaces 5
for all p [1, [ and any positive constants a and b, it holds
ab
a
p
p
+
b
p

, (5)
where
1
p
+
1
p

= 1, i.e., p

=
p
p1
.
For p = p

= 2 inequality (5) is obvious: since (a b)


2
0 and thus a
2
2ab + b
2
0, from where
ab
a
2
2
+
b
2
2
.
Let us prove (5) in the general case. We use Lemma 1. Lets take (x) = x
p1
. Since p > 1, (0) = 0
and is a continuous and strictly increasing function. Therefore,
1
(y) = y
1
p1
and from (7) we
obtain
ab
_
a
0
x
p1
dx +
_
b
0
y
1
p1
dy =
a
p
p
+
b
p

. (6)
The equality in (6) takes place if and only if b = a
p1
which is equivalent to b
p

= a
p

(p1)
= a
p
.
Take a =
x
i
dp(x,0)
and b =
y
i
d
p
(y,0)
. Thanks to (5) we nd
[x
i
y
i
[
d
p
(x, 0)d
p
(y, 0)

[x
i
[
p
p[d
p
(x, 0)]
p
+
[y
i
[
p

[d
p
(x, 0)]
p

,
from where, by taking the sum over all i from 1 to n,

n
i=1
[x
i
y
i
[
d
p
(x, 0)d
p
(y, 0)

n
i=1
[x
i
[
p
p[d
p
(x, 0)]
p
+

n
i=1
[y
i
[
p

[d
p
(x, 0)]
p

=
1
p
+
1
p

= 1.
Therefore, we obtain (4) by the multiplication of the last inequality by d
p
(x, 0)d
p
(y, 0).
Problem 1 1. Prove (geometrically) that if : [0, [[0, [ is a continuous strictly decreasing
function with
_
1
0
(x)dx < , then there exists the inverse function
1
and then for all a 0
and b > 0 it holds
ab
_
a
0
(x)dx
_

b

1
(y)dy. (7)
The equality in (7) takes place if and only if b = (a).
2. Prove the inequality
ab
a
p
p
+
b
p

,
where a 0, b > 0, p ]0, 1[ and p

=
p
p1
< 0.
3. Consider d
p
for p ]0, 1[. Is it a metric on R
n
?
Denition 3 Let (E, d) be a metric space. Let A E. The distance between a set A and a
point x X is dened by d(A, x) = inf
aA
d(a, x).
Problem 2 Prove that
1. x A implies d(A, x) = 0, but not conversely;
2. d(A, x) is a continuous function of x (for xed A) (see Week 1 and Subsection 2.2 for the
denition of the continuous function);
6 Week 2: Metrics and normed spaces
3. d(A, x) = 0 if and only if x is a contact point of A;
4. A = A F, where F is the set of all points x such that d(A, x) = 0.
Problem 3 Let there be two subsets of a metric space (E, d).Then the number z(A, B) = inf
(a,b)AB
d(a, b)
is called the distance between A and B. Show that z(A, B) = 0 if A B ,= , but not conversely.
Hence, z(A, B) is not a distance on T(E), the set of all subsets of E. (A / E but A E, thus
A T(E)).
Show that for A and B two non-empty closed subsets of a metric space (E, d), the following function
d
H
(A, B) = max sup
aA
inf
bB
d(a, b), sup
bB
inf
aA
d(a, b) ,
is a distance on the set of all closed subsets in E (d
H
is a pseudodistance in E). Note that d
H
(A, B)
is called Hausdor distance.
2 Lecture 2.3: Underlying topology to a metric space. Com-
pleteness
2.1 Topology in a metric space
Denition 4 Let (E, d) be a metric space. Given x in E, dene the open ball around x with radius
r > 0 by
B
r
(x) = y E[ d(x, y) < r.
Then, we dene a topology on E by (see Week 1)
T = O E[ x O r > 0, B
r
(x) O. (8)
Theorem 1 Every metric space (E, d) is a normal space, and thus, a Hausdor space, and thus, a
T
1
-space.
Proof. Let X and Y be any two disjoint closed subsets of (E, d). Every point x X has an open
neighborhood O
x
disjoint from Y , and hence is at a positive distance r
x
from Y (recall Problem 2).
Similarly, every point y Y is at a positive distance r
y
from X. Consider the open sets
U =
xX
B
rx
2
(x), V =
yY
B
ry
2
(y).
We have X U, Y V . Moreover, U and V are disjoint. In fact, suppose the contrary that there
is a point z U V . Then there are points x
0
X, y
0
Y such that
d(x
0
, z) <
r
x
0
2
and d(z, y
0
) <
r
y
0
2
.
Assume that r
x
0
r
y
0
. Then we have
d(x
0
, y
0
) d(x
0
, z) +d(z, y
0
) <
r
x
0
2
+
r
y
0
2
r
y
0
,
i.e., x
0
B
ry
0
(y
0
). This contradicts the denition of r
y
0
, and consequently, this shows that there is
no point z U V .
Week 2: Metrics and normed spaces 7
Denition 5 A topological space (X, T) is said to be metrizable if the topology T can be specied
by means of some metric (if (X, T) is homeomorphic to some metric space).
Example 4 The topological space (X, T
t
) is not metrizable.
Finally, specifying a metric in E, we dene a topology in E. Thus, any metric denes a topology,
but not conversely. Therefore all properties of topological spaces hold for metric spaces. In
what follows we only give the properties which are specic for metric spaces.
Denition 6 Two distances are equivalent if they dene the same topology.
Denition 7 Let (X, d
x
) and (Y, d
Y
) be two metric spaces. A map f : X Y is called an isometry
or distance preserving if for any a, b X it holds
d
Y
(f(a), f(b)) = d
X
(a, b).
An isometry is automatically injective. Clearly, every isometry between metric spaces is a topological
embedding.
A global isometry, isometric isomorphism, is a bijective isometry.
Denition 8 Two metric spaces (X, d
x
) and (Y, d
Y
) are called isometric if there is a bijective
isometry from X to Y .
Problem 4 Give an example of two isometric metric spaces.
Denition 9 Given two metric spaces (X, d
X
) and (Y, d
Y
), the function f : X Y is said to be
Lipschitz continuous if there exists a constant K 0, called a Lipschitz constant such that,
x
1
, x
2
X, d
Y
(f(x
1
), f(x
2
)) Kd
X
(x
1
, x
2
).
If 0 K < 1 the function is called a contraction.
Example 5 If f : X Y is a bijection of two metric spaces (X, d
X
) and (Y, d
Y
) and, in addition,
f and f
1
are Lipschitz continuous (f is bi-Lipschitz):
K
1
> 0 and K
2
> 0 : K
1
d
X
(x
1
, x
2
) d
Y
(f(x
1
), f(x
2
)) K
2
d
X
(x
1
, x
2
) x
1
, x
2
X,
then f is isomorphism of (X, d
X
) on (Y, d
Y
). If, in addition, K
2
= K
1
= 1, then f is an isometry.
Example 6 Let us consider two dierent metrics on X: d
1
and d
2
. Thus we have two metric spaces
(X, d
1
) and (X, d
2
). The mapping f : (X, d
1
) (X, d
2
), f(x) = 3x is bi-Lipschitz, then two metrics
d
1
and d
2
are equivalent.
Problem 5 Let (X, d
X
) = (C([a, b]), d

) and (Y, d
Y
) = (R, d), where d is the usual distance on R.
Show that continuously derivable functions on [a, b] are Lipschitz continuous.
2.2 Convergence
We know the denition of the convergence introduced for topological spaces and thus it is the same
in the metric spaces:
A sequence of points (x
n
) in a metric space (E, d) is said to converge to a point l E if every open
neighborhood B

(l) of l contains all points x


n
starting from a certain index:
> 0 N N, n N d(x
n
, l) < .
8 Week 2: Metrics and normed spaces
Clearly, in a metric space
Proposition 1 1. x
n
converges to l in a metric space (E, d) if and only if lim
n
d(x
n
, l) = 0.
2. a function f : (X, d
X
) (Y, d
Y
) is continuous, if from x
n
x for n in X follows that
f(x
n
) f(x) for n in Y .
It is an immediate consequence of the denition of a limit that
1. No sequence can have two distinct limits;
2. If a sequence (x
n
) converges to a point l, then so does every subsequence of (x
n
).
In addition to Week 1, in metric spaces we have
Theorem 2 A necessary and sucient condition for a point l to be a limit point of a set F is that
there exist a sequence (x
n
) of distinct points of F converging to l.
Proof. The condition is necessary, since if l is a limit point of F, then every open neighborhood
B1
n
(l) contains at least one point x
n
F B1
n
(l) dierent of l. We note that
B
1
(l) B1
2
(l) . . . B1
n
(l) . . .
Therefore, these points x
n
can be chosen dierent and these points form a sequence (x
n
) converging
to l. The suciency is obvious.
2.3 Complete metric spaces
Denition 10 In a metric space (E, d), we call Cauchy sequence, a sequence (u
n
) such that
> 0, N > 0, m, n > N d(u
m
, u
n
) < .
Denition 11 A metric space (E, d) is called complete if all Cauchy sequences of elements of E
converge in E.
Example 7 1. R is complete. Q isnt.
2. (C([a, b]), d

) is complete. (C([a, b]), d


2
) isnt. See Example 3 to denitions of d

and d
2
.
Proposition 2 1. Every convergent sequence (x
n
) in (E, d) is a Cauchy sequence in (E, d).
2. If (x
n
) is the Cauchy sequence in (E, d) and if there exists a subsequence x
n
k
such that
x
n
k
x for k + in (E, d), then x
n
x for n + in (E, d).
Proof. Ideas for the proof are given in Figure 4.
2.4 Completion of a metric space
Denition 12 Let (E, d) be a metric space. A complete metric space (G, d) is called a completion
of E if E G and its closure E
G
= G, i.e., if E is a dense subset of G.
Example 8 The space of all real numbers R is the completion of the space of all rational numbers
Q.
Week 2: Metrics and normed spaces 9
x
x
x
n
x
m
x
n
k
x
k
>
>
<
< 2
< 2
>
Figure 4 As distances d(x
n
, x) < and d(x, x
m
) < , then by the triangle inequality d(x
n
, x
m
) < 2 (the
left-hand schema). As d(x
n
k
, x
k
) < and d(x
n
k
, x) < , then by the triangle inequality d(x
k
, x) < 2
(the right-hand schema).
Theorem 3 Every metric space (E, d) has a completion. This completion is unique in the following
sense: if there are two completions E
1
and E
2
, then they are isometric.
For the proof see A.N. Kolmogorov, S.V. Fomin Introductory Real Analysis.
2.5 Separable spaces
Denition 13 A metric space is said to be separable if it has a countable dense subset.
Example 9 R
n
for n N

contains the countable dense set of all points x = (x


1
, . . . , x
n
) with
rational coordinates. C[a, b] are separable spaces.
The space of sequences
2
= x = (x
1
, . . .)[ d
2
(x, y)
2
=

i1
[x
i
y
i
[
2
< contains the
countable dense set of all points x = (x
1
, . . .) with only nit number of nonzero coordiantes,
which are rational.
The space C[a, b] of all continuous functions on [a, b] with a metric
d(g, f) = max
x[a,b]
[g(x) f(x)[
contains the countable dense set of polynomials with rational coecients.
Let

= bounded sequences x = (x
1
, . . .)[ d

(x, y) = sup
k
[x
k
y
k
[.

is an example of a nonseparable space. Let us show that

contains an uncountable dense


set.
In fact, consider the set F of all sequences consisting exclusively of zeros and ones. In this
case, F has the power of the continuum, since there is a bijection between F and the set of all
subsets of the set of natural numbers N: for all A N we associate x
A
= x
n
such that
x
n
=
_
1, if n A,
0, if n / A.
We note that the distance between any two points of F equals 1:
d

(x
A
, x
B
) = 1 if A ,= B.
Suppose we surround each point of F by an open sphere of radius
1
2
, thereby obtaining an
uncountably innite family of pairwise disjoint spheres. Then if some set M is dense in

,
there must be at least one point of M in each of the spheres. It follows that M cannot be
countable and hence that

cannot be separable.
10 Week 2: Metrics and normed spaces
2.6 Compactness in metric spaces
Since metric spaces are topological spaces, all results and denitions of the compactness in the
topological spaces hold for metric spaces as well. Let us just detail the specic properties of the
compactness in metric spaces.
Denition 14 Let (E, d) be a metric space containing a subset M and > 0. A set A E is said
to be an -net for the set M if,
x M there is at least one point a A such that d(x, a) .
It is possible that A M = , but if A is an -net for M, it is possible to construct 2-set B M.
Example 10 The set of all points with integer coordinates is a
1

2
-net of R
2
.
Denition 15 In a metric space (E, d) a subset M is called total ly bounded if for all > 0 there
exists a nit -net of M.
We notice that:
1. If a set M is totally bounded, then its closure M is also totally bounded.
2. Every subset of a totally bounded set is itself totally bounded.
Every totally bounded set is bounded, being the union of a nite number of bounded sets. The
converse is not true, as shown in the following example:
Example 11 The unit sphere S in the space
2
S = x = (x
1
, . . . , x
n
, . . .)
2
[d
2
(x, 0) =

n=1
x
2
n
= 1
is bounded but not totally bounded. In fact, let us consider in S the points
e
1
= (1, 0, 0, . . . , 0, 0, . . .),
e
2
= (0, 1, 0, . . . , 0, 0, . . .),
. . . . . . . . . . . . . . . . . . . . . . . .
e
n
= (0, 0, 0, . . . , 1, 0, . . .),
. . . . . . . . . . . . . . . . . . . . . . . . ,
where the nth coordinate of e
n
is one and the others are all zero. The distance between any two points
e
n
and e
m
(n ,= m) is

2. Hence S cannot have a nite -net with <

2
2
.
Example 12 In Euclidean space R
n
, total boundedness is equivalent to boundedness. In fact, if M
is bounded in R
n
, then M is contained in some suciently large cube Q. Partitioning Q into smaller
cubes of side , we nd that the vertices of the little cubes form a nite (

n
2
)-net for Q and hence
for any set contained in Q.
Example 13 Let P be the set of points x = (x
1
, x
2
, . . . , x
n
, . . .) in
2
satisfying the inequalities
[x
1
[ 1, [x
2
[
1
2
, . . . , [x
n
[
1
2
n1
, . . .
Week 2: Metrics and normed spaces 11
The set P, called the Hilbert cube, gives an example of an innite-dimensional totally bounded set.
Let us prove it. Given any > 0, we choose n such that
1
2
n1
<

2
.
We associate each point x = (x
1
, . . . , x
n
, . . .) in P with the point
x

= (x
1
, x
2
, . . . , x
n
, 0, 0, . . .) P. (9)
Then
d(x, x

) =

i=n+1
x
2
i

i=n
1
4
i
<
1
2
n1
<

2
.
The set P

of all points in P of the form (9) is a bounded set in the n-dimentional space, and,
consequently, P

is totally bounded. Let A be a nite (/2)-net in P

. Then A is a nite -net for


the whole set P.
We give now the main theorems on the compactness in the metric spaces. The proofs can be found,
for example, in A.N. Kolmogorov, S.V. Fomin Introductory Real Analysis.
In a metric space, compact and sequentially compact are equivalent:
Theorem 4 Let (E, d) be a metric space. Then a subset M is compact if and only if it is sequentially
compact.
In particular for the nite dimensional spaces there is the Theorem of Heine-Borel
Theorem 5 (Heine-Borel) In R
n
(or C
n
) a set is compact if and only if it is bounded and closed.
For the general case there is the following theorem:
Theorem 6 A metric space (E, d) is compact if and only if it is totally bounded and complete.
The direct corollary is
Corollary 1 Let (E, d) be a metric space of innite dimension. Then the unit ball B
1
(0) = x
E[d(x, 0) 1 is not compact.
For the relatively compactness we have the following results:
Theorem 7 A subset M of a complete metric space (E, d) is relatively compact if and only if it is
totally bounded.
Example 14 Any bounded subset of R
n
is totally bounded and hence relatively compact (this is a
version of the Bolzano-Weierstrass theorem).
Example 15 (Relatively compact sets in C([a, b])). If (E, d) = (C([a, b]), d

) there is a crite-
rion for relative compactness, called Arzelas theorem.
Theorem 8 (Arzelas theorem) A set of continuous functions dened on a closed interval [a, b]
is relatively compact in (C([a, b]), d

) if and only if is uniformly bounded :


C 0 such that x [a, b] and [(x)[ < C,
and uniformly equicontinuous:
> 0 > 0 such that
1)
2) x
1
, x
2
[a, b] such that d(x
1
, x
2
) <
[(x
1
) (x
2
)[ < .
12 Week 2: Metrics and normed spaces
Denition 16 A topological space is called local ly compact if every point has a compact neigh-
borhood.
Example 16 R
n
is locally compact.
3 Lecture 2.4: Normed vector spaces
3.1 Denition and examples
Denition 17 Let E be a vector space and N : E R a function. N is called a norm on E if
1. x E, N(x) = 0 x = 0
2. (x, ) E R, N(x) = [[N(x)
3. (x, y) E E, N(x +y) N(x) +N(y) (triangular inequality)
If point 1 does not hold, (nevertheless, assertion 2 implies N(0) = 0) N is called a seminorm.
Remark 2 1. Assertions 1 and 3 imply N(x) 0 for all x in E (take y = x in assertion 3).
2. By the mathematical induction from assertion 3, it follows that
N(x
1
+. . . +x
n
) N(x
1
) +. . . +N(x
n
).
From a geometrical point of view, the last inequality can be interpeted as the length of a segment
between two points x
1
and x
1
+x
2
+. . . +x
n
is smaller than the length of the broken line based
on the points y
i
= x
1
+. . . +x
i
for i = 1, . . . , n (see Fig. 5).
x
1 x
1
+. . . +x
n
x
2
Figure 5 Generalized triangular inequality
Denition 18 Let N be a norm on a vector space E. The couple (E, N) is called a normed vector
space. For elements x in E N(x) is usually noted |x|
E
.
Example 17 1. The modulus of a linear function f : E R is a seminorm in E:
for x E N(x) = [f(x)[ satises 2) and 3), but not 1) in Denition 17.
In particular, if E = R (or C), the seminorm N(x) = [x[ is a norm in E (here it is important
that the dimension of E equals to 1). Thus, the notion of a norm can be considered as a
generalization of the notion of modulus.
2. In R
n
| |
lp
and | |

for 1 p < also dene the norms. For p = 2 we nd the Euclidean


norm
|x| =

_
n

k=1
[x
k
[
2
.
Week 2: Metrics and normed spaces 13
3. Consider E =

introduced in Example 9. For u E, dene N(u) = sup


iN
[u
i
[. Then N(u)
is a norm on

.
4. Let us consider the space of sequences l
p
= x = (x
1
, . . .)[

i1
[x
i
[
p
< for 1 p < .
Then |x|
lp
=
_

i1
[x
i
[
p
_1
p
is a norm on l
p
.
5. For the space C([a, b]) of continuous functions on [a, b] we can consider
N

(f) = |f|

= max
axb
[f(x)[, N
p
(f) = |f|
p
=
_
_
b
a
[f(x)[
p
dx
_1
p
,
where 1 p < . Then the spaces (C([a, b]), N

) and (C([a, b]), N


p
) are normed spaces.
6. Let us consider the space Lip of all Lipschitz continuous functions f from a normed space X
to R. We can dene the norm by
|f|
Lip
= [f(x
0
)[ + sup
x=y
[f(x) f(y)[
|x y|
.
Proposition 3 Let (X, | |) be a linear vector normed space. Then the ball
B
r
(0) = x X[|x| r
is convex and symmetric set in X.
The converse is also true: let B
r
(0) be a convex and symmetric ball dened by a function N : X R:
B
r
(0) = x X[N(x) r.
Then N is a norm on X.
Problem 6 Prove it!
Remark 3 Using Proposition 3, it is easy to verify that | |
p
is a norm on C([a, b]): [x[
p
for p > 1
is a convex and symmetric over 0 function, thereby the ball B
r
(0) is convex and symmetric, what
implies that | |
p
is a norm.
We can see that in the same linear vector space E we can dene dierent norms. In the next Section
we will answere the question: if there are two dierent norms | |
1
and | |
2
on E, what about the
properties of (E, | |
1
) and (E, | |
2
)?
3.2 Converging sequences and continuous applications
Let us reformulate in terms of the norm the denitions of converging sequences and continuous
applications from Week 1, given in terms of topology (see Lecture 2.5 for relations between a metric,
a norm and a topology):
1. (E, | |) be a normed linear vector space and (x
n
) be a sequence of elements of E. We say that
(x
n
) converges to l E if
d(x
n
, l) = |x
n
l| 0 for n +.
14 Week 2: Metrics and normed spaces
2. Let f be a mapping of a normed space (X, | |) to a normed space (Y, | |). The mapping f
is continuous if from the convergence of x
n
to l in X for n + follows that the sequence
f(x
n
)
nN
converge to f(l) in Y :
|x
n
l|
X
0 for n + |f(x
n
) f(l)|
Y
0 for n + (10)
Example 18 Let (X, | |
X
) be a normed linear vector space. The norm | |
X
is a continuous
function on X:
if |x
n
l|
X
0 for n + |x
n
|
X
|l|
X
for n +.
To prove it is sucient to notice that for all x and y in X it holds the following inequality:
[ |x|
X
|y|
X
[ |x y|
X
.
Therefore, we have that
[ |l|
X
|x
n
|
X
[ |l x
n
|
X
and, since |l x
n
|
X
0 for n +, it implies that |l|
X
|x
n
|
X
0 for n +.
Remark 4 Any linear continuous mapping f : (X, | |
X
) (Y, | |
Y
) is Lipschitz continuous:
K 0 |f(x) f(y)|
Y
K|x y|
X
x, y X.
If in addition, K < 1, f is a contraction.
4 Lecture 2.5: Underlying metric and topology to a normed
space
We can recognize in Example 17 the formulas of the metrics given in Example 3 written for a distance
between x and 0.
4.1 Metric and a norm
Proposition 4 Let (E, | |) be a normed space. Then the function
(x, y) = |x y| for (x, y) E E (11)
is a metric in E. Moreover, the metric is absolutely homogenous, i.e.
(x, y) = [[(x, y) for (x, y, ) E E C, (12)
and invariant by translation (see Fig. 6)
(x +z, y +z) = (x, y). (13)
Proposition 5 A metric d in a space E is dened by a norm in E by formula (11) if and only if d
satises (12) and (13).
Problem 7 If d satises (12) and (13), prove that N(x) = d(x, 0) is a norm.
Week 2: Metrics and normed spaces 15
x
x
y
y
z
0
Figure 6 Translation of two vectors x and y by a vector z.
4.2 Equivalent norms
Denition 19 Let | |
1
and | |
2
be two norms in a vector space E. The norm | |
2
is called
stronger than the norm | |
1
if there exists a positive constant C > 0 such that
x E |x|
1
C|x|
2
.
Example 19 Let us consider the space C([a, b]) of continuous functions on [a, b] with two following
norms:
|f|

= max
axb
[f(x)[, |f|
2
=
_
_
b
a
[f(x)[
2
dx
_1
2
We nd that | |

is stronger than | |
2
:
|f|
2

_
_
b
a
1dx
_1
2
max
axb
[f(x)[ =

b a|f|

.
A stronger norm will provide a stronger topology. If there are two norms | |
1
and | |
2
on E and
| |
2
is stronger than | |
1
, it means that
1. the convergence in (E, | |
2
) implies the convergence in (E, | |
1
) (but not converse!)
2. if a set F is dense in (E, | |
2
) then F is also dense in (E, | |
1
) (but not converse!)
Denition 20 Let ||
1
and ||
2
be two norms in a vector space E. The norms are called equivalent
if there exist two positive constants c > 0 and C > 0 such that
x E c|x|
2
|x|
1
C|x|
2
.
We notice that norms are equivalent i associated balls can be included in one another (after a
possible homothetic transformation).
Theorem 9 If E is a nite-dimensional vector space dim(E) < , then all norms in E are equiv-
alent.
Proof. Since E is a nite-dimensional vector space, there exists a basis u
i
, 1 i n such that
for all x E there exist unique
i
(i = 1, . . . , n) such that
x =
n

i=1

i
u
i
.
16 Week 2: Metrics and normed spaces
Let us denote by | |
2
the usual Euclidean norm:
|x|
2
= |
n

i=1

i
u
i
|
2
=
_
n

i=1
[
i
[
2
_1
2
.
We want to prove that any norm | | in E is equivalent to the Euclidean norm | |
2
. We start with
the proof of the existence of c > 0 such that c|x| |x|
2
. Using the triangular inequality, we have
that
|x| = |
n

i=1

i
u
i
|
n

i=1
[
i
[|u
i
|.
As for all i |u
i
| are positive numbers, we can apply Lemma 2 for p = 2, which gives Cauchy-Schwartz
inequality in R
n
,
|x|
n

i=1
[
i
[|u
i
|
_
n

i=1
[
i
[
2
_1
2
_
n

i=1
|u
i
|
2
_1
2
= |x|
2
_
n

i=1
|u
i
|
2
_1
2
.
We set c = (

n
i=1
|u
i
|
2
)

1
2
and nally obtain that
c|x| |x|
2
.
Let us justify the existance of C > 0 such that |x|
2
C|x|. The inequality c|x| |x|
2
implies
that if a sequence (x
n
) converges with respect to | |
2
, then (x
n
) converges with respect to | | too.
Consequently, the norm | | is continuous in (E, | |
2
) (as application from E to R
+
) and attains
its minimum m > 0 on the unit sphere
S
1
(0) = x E[ |x|
2
= 1,
(which is compact by the Heine-Borel theorem):
min
x
2
=1
|x| = m > 0.
We can thus write that for |x|
2
= 1
|x|
2
m = 1 m |x|. (14)
Suppose now that x =
y
y
2
, i. e. |x|
2
= 1, but y is not necessary in S
1
(0). Consequently, from (14)
we nd using the linear property of the norm that
|y|
2

1
m
|x||y|
2
=
1
m
_
_
_
_
_
y
|y|
2
_
_
_
_
_
|y|
2
=
1
m
|y|
|y|
2
|y|
2
=
1
m
|y|.
Now we choose C = 1/m.
Corollary 2 Let E be a nite-dimensional vector space. There is only one topology induced by the
norms.
Week 2: Metrics and normed spaces 17
5 Lecture 2.6: An example of a normed space L
p
5.1 A brief summary of the theory of measure and of Lebesgues integral
We give some basic ideas of the constuction of Lebesgues measure, for sets in R
2
following A.N.
Kolmogorov, S.V. Fomin Introductory Real Analysis.
Theorem of Fubini for the direct product of the measures
measure [
R
n+m = measure [
R
n measure [
R
m
allows to have the Lebesgue integral in R
n
.
5.1.1 Denition of a measure
We denote by

the disjoint union.
Denition 21 (-algebra) A -algebra T on a set is a nonempty family of subsets of satisfying
the following axioms :
A1 T contains .
A2 T is stable by the complementarity:
if A T then A T .
A3 T is stable by countable unions:
if n N A
n
T then
nN
A
n
T .
In this denition, it is of course possible to change the axiom (A1) by (A

1):
T contains
and the axiom (A3) by (A

3):
T is stable by countable intersections.
Example 20 Let us consider a set of three elements = a, b, c. We can construct the following
dierent -algebras:
,
, a, b, c,
, b, c, a,
, c, a, b,
, a, b, c, a, b, b, c, c, a, = T()
We can verify that there are no other tribes on this set.
It is easy to show that:
18 Week 2: Metrics and normed spaces
Proposition 6 If T is a -algebra on a set then :
T contains and .
T is stable by complementarity.
T is stable by countable intersections and unions.
T is stable by dierences :
if A T and B T then B A T .
Remark 5 Note that the denition of a -algebra does not coincide with the denition of a topology.
Denition 22 (Abstract measure dened on a -algebra) Let be a set and T be a -algebra
on it. The application A T (A) R is called a measure on T , if it satises the following
conditions:
1. (A) 0 A T ,
2. (A

B) = (A) +(B) A, B T , A B = ,
3. if (B
n
)
nN
a sequence of elements of T such that
B
1
B
2
. . . , B =
n
B
n
, then B T and (B) = lim
n
(B
n
) = sup
nN
(B
n
).
From the denition, it follows the two properties of the measure:
Proposition 7 (Properties of measure)
1. () = 0,
2. (A B) = (A) +(B) (A B).
We also note that
Proposition 8 Let be a set and T be a -algebra on it. A mapping : T R
+
is a measure on
T i it satised two following conditions:
() = 0,
(
nN
A
n
) =

n=0
(A
n
), if A
n
T for all n N and A
i
A
j
= for i ,= j.
5.1.2 Measure of Lebesgue in R
2
Let us consider the set of all rectangles R in R
2
(closed, open and semiclosed or semiopen):
R = (x, y) R
2
[ (a, b, c, d) R
4
a
<

x
<

b, c
<

y
<

d.
By semiclosed or semiopen rectangle we undestand that if one of the inequalities on x or y is strict
<, there exists at least one inequality which is not strict (see Fig. 7).
Let R be a rectangle. We dene its measure (its area) m(R) by
1. If R is empty (a > b or c > d), then m() = 0.
Week 2: Metrics and normed spaces 19
a) b) c)
Figure 7 Examples of rectangles in R
2
: a) a closed rectangle [a, b] [c, d]; b) an open rectangle ]a, b[]c, d[;
c) examples of semiclosed or semiopen rectangles: form the top to the bottom ]a, b[]c, d], [a, b]]c, d]
and ]a, b[[c, d].
2. If R ,= , then m(R) = (b a)(d c).
More generally, we say that
1. For all R the measure m(R) 0 and m(R) R
+
.
2. The measure m(R) is additive: if R =

n
k=1
R
k
(R
i
R
k
= for i ,= k) then
m(R) =
n

k=1
m(R
k
).
We now construct the measure for all subsets of R
2
by two steps:
1. Construction of the measure, denoted m

, of basic sets B which can be presented as a nite


union of disjoint rectangles:
B =
n
_
k=1
R
k
m

(B) =
n

k=1
m(R
k
).
Moreover, for basic sets it holds the following theorem
Theorem 10 All set operations:
, , , and
of two basic sets give a basic set (here AB = (A B) (B A), see Fig 8).
If B
n
B
n
(the sum is not necessary nite), where B and B
n
are basic sets, then
m

(B)

n
m

(B
n
).
Let B
n
be disjoint basic sets for n N, and let B be a basic set, which can be presented
by the disjoint union of B
n
:
B =

_
n=1
B
n
.
Then
m

(B) =

n=1
m

(B
n
).
20 Week 2: Metrics and normed spaces
A
B
Figure 8 Let A and B be sets with a non empty inersection. The set AB = (A B) (B A) is presented
by the blue region.
2. Approximation of a set A in R
2
by a basic set using

(A) = inf
AB=

n
Bn
m(B),
called the exterior measure and dened as an inmum of the measure of all cover of A by
nite or innite union of disjoint rectangles.
Lets rstly consider all sets in R
0
= 0 x 1; 0 y 1 of nite measure. For all sets A R
0
we dene the exterior measure

(A) = inf
AB=

n
Bn
m(B). We notice that
If A is a basic set, then

(A) = m

(A).
If A
n
A
n
R
0
for n J N, then

(A)

(A
n
).
Denition 23 The set A R
0
is measurable by Lebesgue if
> 0 there exists a basic set B such that

(AB) < .
Denition 24 /
R
0
= measurable (by Lebesgue) subsets of R
0
.
Theorem 11 1. /
R
0
is a -algebra.
2. The exterior measure

considered on /
R
0
is a measure. It is called the Lebesgue measure
and noted .
There are properties of Lebesgues measure:
Theorem 12 1. Lebesgue measure is -additive:
if A
n

nN
/
R
0
and A =

n=1
A
n
, then
A /
R
0
and (A) =

n=1
(A
n
).
2. If A
1
A
2
. . . are measurable and A =
n
A
n
, then
A /
R
0
and (A) = lim
n
(A
n
).
We can now generalize the denition of measure in all R
2
: we present R
2
as an union of rectangles
R
n,m
= n < x n + 1, m < y m + 1 of nite measure
R
2
=
n,mZ
R
n,m
,
Week 2: Metrics and normed spaces 21
and we say that A R
2
is measurable if A
n,m
= AR
n,m
is measurable for all n and m in Z. Hence,
we dene
(A) =

n,m
(A
n,m
).
This time (A) can take the value +. In this case, Theorem 12 is still true, but for the property
2) we need to add the condition that (A
1
) < +.
Example 21 All open and closed sets on R
n
are measurable.
5.1.3 Measurable functions
Denition 25 Fuction f : B R
n
R is measurable on B if
1. B is measurable,
2. for all a R f
1
(]a, +[) = x B : f(x) > a is measurable.
We can reformulate the denition of the measurable scalar function by
Proposition 9 Function f : R R is measurable if and only if
c R the set x[ f(x) < c is measurable.
There are several examples of measurable functions:
Example 22 1. Continous functions are measurable, as the inverse image of any open set is open
and any open set is measurable.
2. 1
Q
(x) =
_
1 if x Q,
0 if x R Q
is measurable.
3. Let A be a measurable set. Then 1
A
(x) =
_
1 if x A,
0 if x / A
is measurable (the converse is also
true).
There are main properties of measurable functions:
Proposition 10 1. If f is measurable then k R kf is also measurable.
2. If f is measurable then c R, f +c is also measurable.
3. If f and g are measurable then the sum f + g, the product f g and the composition f g are
also measurable.
4. If f ,= 0 is measurable then
1
f
is also measurable.
5. If n N f
n
is measurable then sup f
n
, inf f
n
, limf
n
(if it exists) and

nN
f
n
(if it exists) are
also measurable.
We note as a collorary of Proposition 10, that
If f is measurable, then f
2
and [f[ are measurable. (15)
22 Week 2: Metrics and normed spaces
Denition 26 (Equivalent functions) Let f and g be two measurable functions. We say that f
is equivalent to g (f g) if they take dierent values on a set which Lebesgue measure is zero:
(x : f(x) ,= g(x)) = 0.
In this case, f = g almost everywhere (a.e.).
We notice that the relation gives classes of equivalentness: if f g then g f, and moreover, if
also g h, it implies that f h.
Example 23 The function 1
Q
(x) and the trivial function f 0 belong to the same class of equiva-
lentness.
We nish the introduction to the measurable functions by Lusins Theorem (see Fig 9 for an example):
Theorem 13 (Lusin) Function f is measurable on [a, b] if and only if
> 0 f

C([a, b]) such that (x : f(x) ,= f

(x)) .

f
f

Figure 9 An example of f (the blue line) and f

(the red line) for Lusins Theorem.


5.1.4 Integral of Lebesgue
Denition 27 A measurable function f : R
m
R is called simple if f can be presented as a linear
combination of indicator functions of measurable by Lebesgue sets:
f(x) =
n

k=1
a
k
1
A
k
(x),
where n N (n ), a
k
R for all k and A
k
are measurable by Lebesgue sets.
The notion of simple functions allows to precise the property to be measurable:
Lemma 3 Function f is measurable if and only if there exists a sequence (f
n
)
nN
of simple functions
such that f
n
converge uniformly to f: f
n
f.
The notion of simple functions also allows to introduce the Lebesgue integrals.
Week 2: Metrics and normed spaces 23
Denition 28 Let f be a simple function and be the Lebesgue measure. The function f is called
integrable (by Lebesgue) on a measurable set A, if the sum
_
A
f(x)d
def
=

n
y
n
(A
n
), where A
n
= x[x A, f(x) = y
n

is absolutely convergent.
Example 24 Let f(x) = 1 on [a, b] which is a simple function. Then, by denition
_
[a,b]
1d = 1 ([a, b]) = b a.
Consequently, we dene the Lebesgue integral as a limit of the integrals for simple functions:
Denition 29 A measurable function f is called Lebesgue-integrable on a measurable by Lebesgue
set A if there exists a sequence of simple functions (f
n
)
nN
which are integrable and f
n
f:
_
A
f(x)d
def
= lim
n
_
A
f
n
(x)d,
where is Lebesgue measure.
We dont give the properties of Lebesgues integral, which can be easily found in the literature,
but we would like to notice the dierence of the constuctions of Riemann and Lebesgue integrals
(see Fig. 10). Disretization on x for the Riemann integral is raplaced by discretization by y in the
a a b b
f(a)
f(b)
x
i
x
i+1

i
a = x
0
< . . . < x
i
< . . . < x
n
= b
f(
i
)

i
y
Figure 10 Left-hand gure: construction of the Riemann integral
_
b
a
f(x)dx = lim
n

n
i=0
(x
i+1

x
i
)f(
i
). Right-hand gure: construction of the Lebesgue integral
_
[a,b]
f(x)dx =
lim
y0,n

n
i=0

i
(x[ [f(x)
i
[ < y) with
i
[f(a), f(b)].
Lebesgue integral.
We also notice that if there exists the Riemann integral of [f[, it implies the existance of the Lebesgue
integral of f. In this case, if there also exists the Riemann integral of f, it is equal to the Lebesgue
integral of f.
This is an example of the function which is Lebesgue-integrable, but not Riemann integrable:
Example 25 Let f be the Dirichlet function:
f(x) = 1
Q[0,1]
(x) =
_
1, x Q [0, 1]
0, x (R Q) [0, 1]
.
24 Week 2: Metrics and normed spaces
The Dirichlet function is Lebesgue-integrable:
_
[0,1]
f(x)d = 1 (Q [0, 1]) + 0 ((R Q) [0, 1]) = 1 0 + 0 1 = 0,
where we have used that Q is a countable set and therefore, its Lebesgue measure (Q) = 0. But the
Dirichlet function is not Riemann integrable: f is discontinous in every point of [0, 1], of a positive
measure equal to 1.
These are especially two theorems which we will use in Week 3:
Theorem 14 (Beppo-Levi)
If (f
n
) is an increasing sequence of measurable positive functions taking values in R
+
, then
lim
_
f
n
=
_
limf
n
. (16)
If (f
n
) is an increasing or decreasing sequence of Lebesgue-integrable functions, then it holds (16).
Theorem 15 (Dominated Convergence) Lets consider a measurable and dened on it Lebesgue
measure . If (f
n
) is a sequence of functions which satises:
n N f
n
is Lebesgue measurable on .
Sequence (f
n
) converge almost everywhere to a measurable function f on .
There exists an Lebesgue-integrable function g,
_

g(x)d < , such that


n N [f
n
(x)[ g(x) a. e. in .
Then,
1. f is Lebesgue-integrable on ,
2. lim
n
_

f
n
(x)d =
_

lim
n
f
n
(x)d =
_

f(x)d.
Finally, it is important to conclude from Denitions 28 and 29 that f is Lebesgue-integrable if f is
measurable and if [f[ is Lebesgue-integrable.
5.2 L
p
spaces
Lets consider an open set of R
n
equipped with the Lebesgue measure.
Denition 30 The set of Lebesgue-integrable functions from to R is noted L
1
() or simply L
1
when no confusion is possible. For f L
1
() we dene
|f|
L
1
()
=
_

[f[d.
Each element of L
1
is a classe of equivalent functions, which are equal almost everywhere.
Denition 31 Let 0 < p < , and f : R be measurable functions. The function f belongs to
L
p
() if [f[
p
L
1
(). For f L
p
() we dene
|f|
L
p
()
=
__

[f[
p
d
_1
p
.
Week 2: Metrics and normed spaces 25
Remark 6 For p = 1 we obtain from Denition 31 that f L
1
() implies that
1. f is measurable,
2. [f[ L
1
(),
which exactly means that f is Lebesgue-integrable. Hence, Denition 31 for p = 1 is equivalent to
Denition 30. Functions that are equal almost everywhere are identied.
Remark 7 The space (C(), | |
L
1
()
) is not complete. By the way,
L
1
() = C()

L
1
,
i.e. L
1
() is the completion of C() by the norm | |
L
1.
Denition 32 Let f be a function from to R. We dene essential supremum of f as a number
ess sup
x
f(x) = inf
(A)=0
_
sup
x\A
f(x)
_
= inf
B: (\B)=0
_
sup
xB
f(x)
_
. (17)
Example 26 Let f(x) = 1
Q[0,1]
(x). By denition of the essential supremum, ess sup
x[0,1]
f(x) = 0,
but sup
x[0,1]
f(x) = 1.
Remark 8 1. It holds
ess sup
x
f(x) = infM R : f(x) M a. e. in .
2. Let R
n
be an open set and f : R be a continuous function. Then it holds
ess sup
x
f(x) = sup
x
f(x).
Problem 8 Prove using Remark 8 1) that
[f(x)[ |f|
L

()
a. e. in .
Denition 33 We note L

() the set of measurable functions from to R for which there exists a


real number C such that for almost every x in , [f(x)[ C, i.e.
ess sup
x
f(x) < .
Functions equal almost everywhere are identied. We note
|f|
L

()
= ess sup
x
f(x).
Proposition 11 Let f L
p
() for 0 < p . Then
|f|
L
p
()
= 0 f = 0 a.e. on .
Proof. For 0 < p < we have:
|f|
L
p
()
= 0
_

[f[
p
d = 0 [f[
p
= 0 a.e. on f = 0 a.e. on .
26 Week 2: Metrics and normed spaces
Problem 9 Prove it for p = :
|f|
L

()
= 0 f = 0 a.e. on .
There are some indications:
1. Use Remark 8 1).
2. Show for A = |f|
L

()
that
A+
1
n
= x : f(x) A+
1
n
has Lebesgue measure (
A+
1
n
) =
() for n = 1, 2, . . ..
3. Consider
A
=

n=1

A+
1
n
which is a full Lebesgue measure: (
A
) = ().
Problem 10 Show that L
p
() for 0 < p is a linear vector space. Indication: use the numerical
inequality
a, b 0 and p > 0 (a +b)
p
2
p
(a
p
+b
p
).
For example, if a b, then
(a +b)
p
(2a)
p
= 2
p
a
p
2
p
(a
p
+b
p
).
Denition 34 Let p [1, ].
A function f belongs to L
p,loc
() when f1
K
belongs to L
p
() for every compact K , where 1
K
is
the characteristic function of K: 1
K
(x) = 1 if x K and 0 otherwise.
Denition 35 Let p [1, ].
We call Hlder conjugate (or dual index) of p, the number p

= 1 +
1
p1
so that
1
p
+
1
p

= 1 (see
Section ) (if p = 1 then p

= and p = then p

= 1).
Note that the Hlder conjugate of 2 is 2.
Proposition 12 (Hlders Inequality) Let p [1, ] and p

be its Hlder conjugate. Let f


L
p
() and g L
p

(). Then fg L
1
and
|fg|
L
1
()
|f|
L
p
()
|g|
L
p

()
.
Problem 11 Using results of Section 35, prove Hlders inequality.
If we summarize all the results, we have
Corollary 3 Let p [1, ]. | | is a norm on L
p
().
5.3 Applications of Hlders inequality
Proposition 13 Let 0 < p q , R
n
, () < . Then
1. L
q
() L
p
(),
2. |f|
L
p
()
()
1
p

1
q
|f|
L
q
()
.
Proof. We denote by Q =
q
p
1 and calculate the dual index of Q:
Q

=
Q
Q1
=
q
q p
.
Week 2: Metrics and normed spaces 27
We now apply Hlders inequality with Q and Q

:
|f|
p
L
p
()
=
_

[f[
p
d =
_

[f[
p
1d
|[f[
p
|
L
Q |1|
L
Q
=
__

([f[
p
)
Q
d
_ 1
Q
__

1d
_ 1
Q

=
__

[f[
q
d
_
p
q
()
qp
q
,
from where
|f|
L
p
()
=
__

[f[
p
d
_1
p

__

[f[
q
d
_1
q
()
qp
pq
,
which gives 2).
If f L
q
, thanks to 2), |f|
L
p
()
< , what implies that f L
p
, i.e. 1) holds.
Let us notice that the condition () < is very important. For instance,
Problem 12 For = R
+
give an example of a function f such that for 0 < p < q
f L
q
(R
+
), but f / L
p
(R
+
).
Proposition 14 Let 0 < p
1
< p < p
2
. Lets dene by the equality
1
p
=

p
1
+
1
p
2
(18)
Note that for p = p
1
= 1 and for p = p
2
= 0. Such ]0, 1[ exists and it is unique. Then from
f L
p
1
() L
p
2
() follows that f L
p
() and
|f|
L
p
()
|f|

L
p
1()
|f|
1
L
p
2()
. (19)
Proof. Let us consider p
2
< . We denote by q =
p
1
p
and calculate the dual index of q using (18):
1 =
p
p
1
+
(1 )p
p
2

p
p
1
< 1 q > 1
q

=
q
q 1

1
q

= 1
1
q
= 1
p
p
1
=
(1 )p
p
2
,
i.e. q

=
p
2
(1 )p
.
We now apply Hlders inequality with q and q

:
|f|
p
L
p
()
=
_

[f[
p
d =
_

[f[
p
[f[
(1)p
d

__

([f[
p
)
p
1
p
d
_p
p
1
__

([f[
(1)p
)
p
2
(1)p
d
_
(1)p
p
2
=
__

[f[
p
1
d
_p
p
1
__

[f[
p
2
d
_
(1)p
p
2
,
from where
|f|
L
p
()
=
__

[f[
p
d
_1
p

__

[f[
p
1
d
_
p
1
__

[f[
p
2
d
_1
p
2
= |f|

L
p
1()
|f|
1
L
p
2()
,
which gives (19).
Problem 13 Prove (19) for p
2
= .
28 Week 2: Metrics and normed spaces
We nish with the interpolation inequality (see H. Brezis Functional Analysis, Sobolev Spaces and
Partial Dierential Equations):
Proposition 15 Let f
i
, i I be a family of functions with f
i
L
p
i
() and
1
p
=

1
p
i
1. Then

f
i
L
p
() and
|

f
i
|
L
p
()

|f
i
|
L
p
i ()
Corollary 4 If f L
p
() L
q
() then f L
r
() for any r such that p r q.

Vous aimerez peut-être aussi