You are on page 1of 12

3.

The L
p
spaces (1 p < )
In this section we discuss an important construction, which is extremely useful
in virtually all branches of Analysis. In Section 1, we have already introduced the
space L
1
. The rst construction deals with a generalization of this space.
Definitions. Let (X, A, ) be a measure space, and let K be one of the elds
R or C.
A. For a number p (1, ), we dene the space
L
p
K
(X, A, ) =
_
f : X K : f measurable, and
_
X
[f[
p
d <
_
.
Here we use the convention introduced in Section 1, which denes
_
X
hd = ,
for those measurable functions h : X [0, ], that are not integrable.
Of course, in this denition we can allow also the value p = 1, and in this case
we get the familiar denition of L
1
K
(X, A, ).
B. For p [1, ), we dene the map Q
p
: L
1
K
(X, A, ) [0, ) by
Q
p
(f) =
_
X
[f[
p
d, f L
1
K
(X, A, ).
Remark 3.1. The space L
1
K
(X, A, ) was studied earlier (see Section 1). It
has the following features:
(i) L
1
K
(X, A, ) is a K-vector space.
(ii) The map Q
1
: L
1
K
(X, A, ) [0, ) is a seminorm, i.e.
(a) Q
1
(f +g) Q
1
(f) +Q
1
(g), f, g L
1
K
(X, A, );
(b) Q
1
(f) = [[ Q
1
(f), f L
1
K
(X, A, ), K.
(iii)

_
X
f d

Q
1
(f), f L
1
K
(X, A, ).
Property (b) is clear. Property (a) immediately follows from the inequality [f +g[
[f[ +[g[, which after integration gives
_
X
[f +g[ d
_
X
_
[f[ +[g[

d =
_
X
[f[ d +
_
X
[g[ d.
In what follows, we aim at proving similar features for the spaces L
p
K
(X, A, )
and Q
p
, 1 < p < .
The following will help us prove that L
p
is a vector space.
Exercise 1

. Let p (1, ). Then one has the inequality

(s +t)
p
2
p1
(s
p
+t
p
), s, t [0, ).
Hint: The inequality is trivial, when s = t = 0. If s + t > 0, reduce the problem to the case
t + s = 1, and prove, using elementary calculus techniques that
min
t[0,1]

t
p
+ (1 t)
p

= 2
1p
.
Proposition 3.1. Let (X, A, ) be a measure space, let K be one of the elds
R or C, and let p (1, ). When equipped with pointwise addition and scalar
multiplication, L
p
K
(X, A, ) is a K-vector space.
300
3. L
p
spaces (1 p < ) 301
Proof. It f, g L
p
K
(X, A, ), then by Exercise 1 we have
_
X
[f +g[
p
d
_
X
_
[f[ +[g[
_
p
d 2
p1
_ _
X
[f[
p
d +
_
X
[g[
p
d
_
< ,
so f +g indeed belongs to L
p
K
(X, A, ).
It f L
p
K
(X, A, ), and K, then the equalities
_
X
[f[
p
d =
_
X
[[
p
[f[
p
d = [[
p

_
X
[f[
p
d
clearly prove that f also belongs to L
p
K
(X, A, ).
Our next task will be to prove that Q
p
is a seminorm, for all p > 1. In this
direction, the following is a key result. (The above mentioned convention will be
used throughout this entire section.)
Theorem 3.1 (Holders Inequality for integrals). Let (X, A, ) be a measure
space, let f, g : X [0, ] be measurable functions, and let p, q (1, ) be such
that
1
p
+
1
q
= 1. Then one has the inequality
1
(1)
_
X
fg d
_ _
X
f
p
d
_
1/p

_ _
X
g
q
d
_
1/q
.
Proof. If either
_
X
f
p
d = , or
_
X
g
p
d = , then the inequality (1) is
trivial, because in this case, the right hand side is . For the remainder of the
proof we will assume that
_
X
f
p
d < and
_
X
g
q
d < .
Use Corollary 2.1 to nd two sequences (
n
)

n=1
, (
n
)

n=1
L
1
R,elem
(X, A, ),
such that
0
1

2
. . . and 0
1

2
. . . ;
lim
n

n
(x) = f(x)
p
and lim
n

n
(x) = g(x)
q
, x X.
By the Lebesgue Dominated Convergence Theorem, we will also get the equalities
(2)
_
X
f
p
d = lim
n
_
X

n
d and
_
X
g
q
d = lim
n
_
X

n
d.
Remark that the functions f
n
=
1/p
n
, g
n

1/q
n
, n 1 are also elementary (because
they obviously have nite range). It is obvious that we have
0 f
1
f
2
. . . , and 0 g
1
g
2
. . . ;
lim
n
f
n
(x) = f(x), and lim
n
g
n
(x)] = g(x), x X.
With these notations, the equalities (2) read
(3)
_
X
f
p
d = lim
n
_
X
(f
n
)
p
d and
_
X
g
q
d = lim
n
_
X
(g
n
)
q
d.
Of course, the products f
n
g
n
, n 1 are again elementary, and satisfy
0 f
1
g
1
f
2
g
2
. . . ;
lim
n
[f
n
(x)g
n
(x)] = f(x)g(x), x X.
Using the General Lebesgue Monotone Convergence Theorem, we then get
_
X
fg d = lim
n
_
X
f
n
g
n
d.
1
Here we use the convention
1/p
=
1/q
= .
302 CHAPTER IV: INTEGRATION THEORY
Using (3) we now see that, in order to prove (1), it suces to prove the inequalities
_
X
f
n
g
n
d
_ _
X
(f
n
)
p
d
_
1/p

_ _
X
(g
n
)
q
d
_
1/q
, n 1.
In other words, it suces to prove (1), under the extra assumption that both f and
g are elementary integrable.
Suppose f and g are elementary integrable. Then (see III.1) there exist pair-
wise disjoint sets (D
j
)
m
j=1
A, with (D
j
) < , j = 1, . . . , m, and numbers

1
,
1
, . . . ,
m
,
m
[0, ), such that
f =
1

D1
+ +
m

Dm
g =
1

D1
+ +
m

Dm
Notice that we have
fg =
1

D1
+ +
m

Dm
,
so the left hand side of (1) is the given by
_
X
fg d =
m

j=1

j
(D
j
).
Dene the numbers x
j
=
j
(D
j
)
1/p
, y
j
=
j
(D
j
)
1/q
, j = 1, . . . , m. Using these
numbers, combined with
1
p
+
1
q
= 1, we clearly have
(4)
_
X
fg d =
m

j=1
(x
j
y
j
).
At this point we are going to use the classical Holder inequality (see Appendix D),
which gives
m

j=1
(x
j
y
j
)
_
m

j=1
(x
j
)
p
_
1/p

_
m

j=1
(y
j
)
q
_
1/q
,
so the equality (4) continues with
_
X
fg d
_
m

j=1
(x
j
)
p
_
1/p

_
m

j=1
(y
j
)
q
_
1/q
=
=
_
m

j=1
(
j
)
p
(D
j
)
_
1/p

_
m

j=1
(
j
)
q
(D
j
)
_
1/q
=
=
_ _
X
f
p
d
_
1/p

_ _
X
g
q
d
_
1/q
.
Corollary 3.1. Let (X, A, ) be a measure space, let K be one of the elds
R or C, and let p, q (1, ) be such that
1
p
+
1
q
= 1. For any two functions
f L
p
K
(X, A, ) and g L
q
K
(X, A, ), the product fg belongs to L
1
K
(X, A, ) and
one has the inequality

_
X
fg d

Q
p
(f) Q
q
(g).
3. L
p
spaces (1 p < ) 303
Proof. By Holders inequality, applied to [f[ and [g[, we get
_
X
[fg[ d Q
p
(f) Q
q
(g) < ,
so [fg[ belongs to L
1
+
(X, A, ), i.e. fg belongs to L
1
K
(X, A, ). The desired inequal-
ity then follows from the inequality

_
X
fg d

_
X
[fg[ d.
Notation. Suppose (X, A, ) is a measure space, K is one of the elds R or C,
and p, q (1, ) are such that
1
p
+
1
q
= 1. For any pair of functions f L
p
K
(X, A, ),
g L
q
K
(X, A, ), we shall denote the number
_
X
fg d K simply by f, g). With
this notation, Corollary 3.1 reads:

f, g)

Q
p
(f) Q
q
(g), f L
p
K
(X, A, ), g L
q
K
(X, A, ).
The following result gives an alternative description of the maps Q
p
, p (1, ).
Proposition 3.2. Let (X, A, ) be a measure space, let K be one of the elds
R or C, let p, q (1, ) be such that
1
p
+
1
q
= 1. and let f L
p
K
(X, A, ). Then
one has the equality
(5) Q
p
(f) = sup
_

f, g)

: g L
q
K
(X, A, ), Q
q
(g) 1
_
.
Proof. Let us denote the right hand side of (5) simply by P(f). By Corollary
3.1, we clearly have the inequality
P(f) Q
p
(f).
To prove the other inequality, let us rst observe that in the case when Q
p
(f) = 0,
there is nothing to prove, because the above inequality already forces P(f) = 0.
Assume then Q
p
(f) > 0, and dene the function h : x K by
h(x) =
_

_
[f(x)[
p
f(x)
if f(x) ,= 0
0 if f(x) = 0
It is obvious that h is measurable. Moreover, one has the equality [h[ = [f[
p1
,
which using the equality qp = p + q gives [h[
q
= [f[
qpq
= [f[
p
. This proves that
h L
q
K
(X, A, ), as well as the equality
Q
q
(h) =
_ _
X
[h[
q
d
_
1/q
=
_ _
X
[f[
p
d
_
1/q
= Q
p
(f)
p/q
.
If we dene the number = Q
p
(f)
p/q
, then the function g = h has Q
q
(g) = 1,
so we get
P(f)

_
X
fg d

=
1
Q
p
(f)
p/q

_
X
fhd

.
Notice that fh = [f[
p
, so the above inequality can be continued with
P(f)
1
Q
p
(f)
p/q
_
X
[f[
p
d =
Q
p
(f)
p
Q
p
(f)
p/q
= Q
p
(f).
Corollary 3.2. Let (X, A, ) be a measure space, let K be one of the elds
R or C, and let p (1, ). Then the Q
p
is a seminorm on L
p
K
(X, A, ), i.e.
(a) Q
p
(f
1
+f
2
) Q
p
(f
1
) +Q
p
(f
2
), f
1
, f
2
L
p
K
(X, A, );
(b) Q
p
(f) = [[ Q
p
(f), f L
p
K
(X, A, ), K.
304 CHAPTER IV: INTEGRATION THEORY
Proof. (a). Take q =
p
p1
, so that
1
p
+
1
q
g L
q
K
(X, A, ), with Q
q
(g) 1. Then the functions f
1
g and f
2
g belong to
L
1
K
(X, A, ), and so f
1
g +f
2
g also belongs to L
1
K
(X, A, ). We then get

f
1
+f
2
, g)

_
X
(f
1
g +f
2
g) d

_
X
f
1
g d +
_
X
f
2
g d

_
X
f
1
g d

_
X
f
2
g d

f
1
, g)

f
2
, g)

.
Using Proposition 3.2, the above inequality gives

f
1
+f
2
, g)

Q
p
(f
1
) +Q
p
(f
2
).
Since the above inequality holds for all g L
q
K
(X, A, ), with Q
q
(g) 1, again by
Proposition 3.2, we get
Q
p
(f
1
+f
2
) Q
p
(f
1
) +Q
p
(f
2
).
Property (b) is obvious.
Remarks 3.2. Let (X, A, ) be a measure space, and K be one of the elds R
or C, and let p [1, ).
A. If f L
p
K
(X, A, ) and if g : X K is a measurable function, with g = f,
-a.e., then g L
p
K
(x, A, ), and Q
p
(g) = Q
p
(f).
B. If we dene the space
N
K
(X, A, ) =
_
f : X K : f measurable, f = 0, -a.e.
_
,
then N
K
(X, A, ) is a linear subspace of L
p
K
(X, A, ). In fact one has the equality
N
K
(X, A, ) =
_
f L
p
K
(X, A, ) : Q
p
(f) = 0
_
.
The inclusion is trivial. Conversely, f L
p
K
(X, A, ) has Q
p
(f) = 0, then the
measurable function g : X [0, ) dened by g = [f[
p
will have
_
X
g d = 0. By
Exercise 2.3 this forces g = 0, -a.e., which clearly gives f = 0, -a.e.
Definition. Let (X, A, ) be a measure space, let K be one of the elds R or
C, and let p [1, ). We dene
L
p
K
(X, A, ) = L
p
K
(X, A, )/N
K
(X, A, ).
In other words, L
p
K
(X, A, ) is the collection of equivalence classes associated with
the relation =, -a.e. For a function f L
p
K
(X, A, ) we denote by [f] its
equivalence class in L
p
K
(X, A, ). So the equality [f] = [g] is equivalent to f = g,
-a.e. By the above Remark, there exists a (unique) map | . |
p
: L
p
K
(X, A, )
[0, ), such that
|[f]|
p
= Q
p
(f), f L
p
K
(X, A, ).
By the above Remark, it follows that | . |
p
is a norm on L
p
K
(X, A, ). When K = C
the subscript C will be ommitted.
Conventions. Let (X, A, ), K, and p be as above We are going to abuse a
bit the notation, by writing
f L
p
K
(X, A, ),
if f belongs to L
p
K
(X, A, ). (We will always have in mind the fact that this notation
signies that f is almost uniquely determined.) Likewise, we are going to replace
Q
p
(f) with |f|
p
.
3. L
p
spaces (1 p < ) 305
Given p, q (1, ), with
1
p
+
1
q
= 1, we use the same notation for the (correctly
dened) map
. , . ) : L
p
K
(X, A, ) L
q
K
(X, A, ) K.
Remark 3.3. Let (X, A, ) be a measure space, let K be either R or C, and
let p, q (1, ) be such that
1
p
+
1
q
= 1. Given f L
p
K
(X, A, ), we dene the map

f
: L
q
K
(X, A, ) g f, g) K.
According to Proposition 3.2, the map
f
is linear, continuous, and has norm
|
f
| = |f|
p
. If we denote by L
q
K
(X, A, )

the Banach space of all linear continu-

ous maps L
q
K
(X, A, ) K, then we have a correspondence
(6) L
p
K
(X, A, ) f
f
L
q
K
(X, A, )

which is linear and isometric. This correspondence will be analyzed later in Section
5.
Notation. Given a sequence (f
n
)

n=1
, and a function f, in L
p
K
(X, A, ), we
are going to write
f = L
p
- lim
n
f
n
,
if (f
n
)

n=1
converges to f in the norm topology, i.e. lim
n
|f
n
f|
p
= 0.
The following technical result is very useful in the study of L
p
spaces.
Theorem 3.2 (L
p
Dominated Convergence Theorem). Let (X, A, ) be a mea-
sure space, let K be one of the elds R or C, let p [1, ) and let (f
n
)

n=1
be a
sequence in L
p
K
(X, A, ). Assume f : X K is a measurable function, such that
(i) f = -a.e.- lim
n
f
n
;
(ii) there exists some function g L
p
K
(X, A, ), such that
[f
n
[ [g[, -a.e., n 1.
Then f L
p
K
(X, A, ), and one has the equality
f = L
p
- lim
n
f
n
.
Proof. Consider the functions
n
= [f
n
[
p
, n 1, and = [f[
p
, and = [g[
p
.
Notice that
= -a.e.- lim
n

n
;
[
n
[ , -a.e., n 1;
L
1
+
(X, A, ).
We can apply the Lebsgue Dominated Convergence Theorem, so we get the fact that
L
1
+
(X, A, ), which gives the fact that f L
p
K
(X, A, ). Now if we consider
the functions
n
= [f
n
f[
p
, and = 2
p1
_
[g[
p
+ [f[
p
_
, then using Exercise 1, we
have:
0 = -a.e.- lim
n

n
;
[
n
[ , -a.e., n 1;
L
1
+
(X, A, ).
Again using the Lebesgue Dominated Convergence Theorem, we get
lim
n
_
X

n
d = 0,
306 CHAPTER IV: INTEGRATION THEORY
which means that
lim
n
[f
n
f[
p
d,
n
_
|f
n
f|
p
_
p
= 0, so we clearly have f = L
p
- lim
n
f
n
.
Our main goal is to prove that the L
p
spaces are Banach spaces. The key result
which gives this, but also has some other interesting consequences, is the following.
Theorem 3.3. Let (X, A, ) be a measure space, let K be one of the elds R
or C, let p [1, ) and let (f
k
)

k=1
be a sequence in L
p
K
(X, A, ), such that

k=1
|f
k
|
p
< .
Consider the sequence (g
n
)

n=1
L
p
K
(X, A, ) of partial sums:
g
n
=
n

k=1
f
k
, n 1.
Then there exists a function g L
p
K
(X, A, ), such that
(a) g = -a.e.- lim
n
g
n
;
(b) g = L
p
- lim
n
g
n
.
Proof. Denote the sum

k=1
|f
n
|
p
simply by S. For each integer n 1,
dene the function h
n
: X [0, ], by
h
n
(x) =
n

k=1
[f
n
(x)[, x X.
It is clear that h
n
L
p
R
(X, A, ), and we also have
(7) |h
n
|
p

n

k=1
|f
k
|
p
S, n 1.
Notice also that 0 h
1
h
2
. . . . Dene then the function h : X [0, ] by
h(x) = lim
n
h
n
(x), x X.
Claim: h L
p
R
(X, A, ).
To prove this fact, we dene the functions = h
p
and
n
= (h
n
)
p
, n 1. Notice
that, we have
0
1

2
. . . ;

n
L
1
R
(X, A, ), n 1;
lim
n

n
(x) = (x), x X;
sup
__
X

n
d : n 1
_
M
p
.
Using the Lebesgue Monotone Convergence Theorem, it then follows that h
p
=
L
1
R
(X, A, ), so h indeed belongs to L
p
R
(X, A, ).(7) gives
Let us consider now the set N = x X : h(x) = . On the one hand, since
we also have
N = x X : (x) < ,
and is integrable, it follows that N A, and (N) = 0. On the other hand, since

k=1
[f
n
(x)[ = h(x) < , x X N,
3. L
p
spaces (1 p < ) 307
it follows that, for each x X N, the series

k=1
f
k
(x) is convergent. Let us
dene then g : X K by
g(x) =
_

k=1
f
k
(x) if x X N
0 if x N
It is obvious that g is measurable, and we have
g = -a.e.- lim
n
g
n
.
Since we have
[g
n
[ =

k=1
f
k

k=1
[f
k
[ = h
n
h, n 1,
using the Claim, and Theorem 3.2, it follows that g indeed belongs to L
p
K
(X, A, )
and we also have the equality g = L
p
- lim
n
g
n
.
Corollary 3.3. Let (X, A, ) be a measure space, and let K be one of the
elds R or C. Then L
p
K
(X, A, ) is a Banach space, for each p [1, ).
Proof. This is immediate from the above result, combined with the complete-
ness criterion given by Remark II.3.1.
Another interesting consequence of Theorem 3.3 is the following.
Corollary 3.4. Let (X, A, ) be a measure space, let K be one of the elds
R or C, let p [1, ), and let f L
p
K
(X, A, ). Any sequence (f
n
)

n=1

L
p
K
(X, A, ), with f = L
p
- lim
n
f
n
, has a subsequence (f
n
k
)

k=1
such that f =
-a.e.- lim
k
f
n
k
.
Proof. Without any loss of generality, we can assume that f = 0, so that we
have
lim
n
|f
n
|
p
= 0.
Choose then integers 1 n
1
< n
2
< . . . , such that
|f
n
k
|
p

1
2
k
, k 1.
If we dene the functions
g
m
=
m

k=1
f
n
k
,
then by Theorem 3.3, it follows that there exists some g L
p
K
(X, A, ), such that
g = -a.e.- lim
m
g
m
.
This measn that there exists some N A, with (N) = 0, such that
lim
m
g
m
(x) = g(x), x X N.
In other words, for each x XN, the series

k=1
f
n
k
(x) is convergent (to some
number g(x) K). In particular, it follows that
lim
k
f
n
k
(x) = 0, x X N,
so we indeed have 0 = -a.e.- lim
k
f
n
k
.
The following result collects some properties of L
p
spaces in the case when the
undelrying measure space is nite.
308 CHAPTER IV: INTEGRATION THEORY
Proposition 3.3. Suppose (X, A, ) is a nite measure space, and K is one
of the elds R or C.
(i) If f : X K is a bounded measurable function, then f L
p
K
(X, A, ),
p [1, ).
(ii) For any p, q [1, ), with p < q, one has the inclusion L
q
K
(X, A, )
L
p
K
(X, A, ). So taking quotients by N
K
(X, A, ), one gets an inclusion of
vector spaces
(8) L
q
K
(X, A, ) L
p
K
(X, A, ).
Moreover the above inclusion is a continuous linear map.
Proof. The key property that we are going to use here is the fact that the
constant function 1 =
X
is -integrable (being elementary -integrable).
(i). This part is pretty clear, because if we start with a bounded measurable
function f : X K and we take M = sup
xX
[f(x)[, then the inequality [f[
p

M
p
1, combined with the integrability of 1, will force the inetgrability of [f[
p
, i.e.
f L
p
K
(X, A, ).
(ii). Fix 1 p < q < , as well as a function f L
q
K
(X, A, ). Consider the
number r =
q
p
> 1, and s =
r
r1
, so that we have
1
r
+
1
s
= 1. Since f L
q
K
(X, A, ),
the function g = [f[
q
belongs to L
1
K
(X, A, ). If we dene then the function h = [f[
p
,
then we obviously have g = h
r
, so we get the fact that h belongs to L
r
K
(X, A, ).
Using part (i), we get the fact that 1 L
s
K
(X, A, ), so by Corollary 3.1, it follows
that h = 1 h belongs to L
1
K
(X, A, ), and moreover, one has the inequality
_
X
[f[
p
d =
_
X
hd |1|
s
|h|
r
=
_ _
X
1 d
_
1/s

_ _
X
h
r
d
_
1/r
=
= (X)
1/s

_ _
X
[f[
q
d
_
1/r
= (X)
1/s

_
|f|
q
_
q/r
.
On the one hand, this inequality proves that f L
p
K
(X, A, ). On the other hand,
this also gives the inequality
_
|f|
p
_
p
(X)
1/s

_
|f|
q
_
q/r
= (X)
1
p
q

_
|f|
q
_
p
,
which yields
|f|
p
(X)
1
p

1
q
|f|
q
.
This proves that the linear map (8) is continuous (and has norm no greater than
(X)
1
p

1
q
).
Exercise 2. Give an example of a sequence of continuous functions f
n
: [0, 1]
[0, ), n 1, such that L
p
- lim
n
f
n
= 0, p [1, ), but for which it
is not true that 0 = -a.e.- lim
n
f
n
. (Here we work on the measure space
_
[0, 1], M

([0, 1]), ).)

Exercise 3. Let R
n
be an open set. Prove that C
K
c
() is dense in
L
p
K
(, M

measure, and M

() denotes the collection of all Lebesgue measurable subsets of

.)
3. L
p
spaces (1 p < ) 309
Notations. Let (X, A, ) be a measure space, let K be one of the elds R or
C. We dene the space
N
K,elem
(X, A, ) = L
1
K,elem
(X, A, ) N
K
(X, A, ),
and we dene the quotient space
L
1
K,elem
(X, A, ) = L
1
K,elem
(X, A, ) / N
K,elem
(X, A, ).
In other words, if one considers the quotient map

1
: L
1
K
(X, A, ) L
1
K
(X, A, ),
then L
1
K,elem
(X, A, ) =
1
_
L
1
K,elem
(X, A, )
_
. Notice that we have the obvious
inclusion
L
1
K,elem
(X, A, ) L
p
K
(X, A, ), p [1, ),
so we we consider the quotient map

p
: L
p
K
(X, A, ) L
p
K
(X, A, ),
we can also dene the subspace
L
p
K,elem
(X, A, ) =
p
_
L
p
K,elem
(X, A, )
_
, p [1, ).
Remark that, as vector spaces, the spaces L
p
K,elem
(X, A, ) are identical, since
Ker
p
= N
K
(x, A, ), p [1, ).
With these notations we have the following fact.
Proposition 3.4. L
p
K,elem
(X, A, ) is dense in L
p
K
(X, A, ), for each p
[1, ).
Proof. Fix p [1, ), and start with some f L
p
K
(X, A, ). What we
need to prove is the existence of a sequence (f
n
)

n=1
L
1
K,elem
(X, A, ), such that
f = L
p
- lim
n
f
n
. Taking real and imaginary parts (in the case K = C), it
sueces to consider the case when f is real valued. Since [f[ also belongs to L
p
,
it follows that f
+
= maxf, 0 =
1
2
_
[f[ + f
_
, and f

= maxf, 0 =
1
2
_
[f[ f
_
both belong to L
p
, so in fact we can assume that f is non-negative. Consider the
function g = f
p
L
1
+
(X, A, ). Use the denition of the integral, to nd a sequence
(g
n
)

n=1
L
1
R,elem
(X, A, ), such that
0 g
n
g, n 1;
lim
n
_
X
g
n
d =
_
X
g d.
This gives the fact that g = L
1
- lim
n
g
n
. Using Corollary 3.4, after replacing
(g
n
)

n=1
with a subsequence, we can also assume that g = -a.e.- lim
n
g
n
. If we
put f
n
= (g
n
)
1/p
, n 1, we now have
0 f
n
f, n 1;
f = -a.e.- lim
n
f
n
.
Obviously, the f
n
s are still elementary integrable, and by the L
p
Dominated Con-
vergence Theorem, we indeed get f = L
p
- lim
n
f
n
.
Comments. A. The above result gives us the fact that L
p
K
(X, A, ) is the com-
pletion of L
p
K,elem
(X, A, ). This allows for the following alternative construction
of the L
p
spaces.
B. For a measurable function f : X K, by the (proof of the) above result,
it follows that the condition f L
p
K
(X, A, ) is equivalent to the equality f =
310 CHAPTER IV: INTEGRATION THEORY
-a.e.- lim
n
f
n
, for some sequence (f
n
)

n=1
of elementary integrable functions,
which is Cauchy in the L
p
norm, i.e.
(c) for every > 0, there exists N

, such that
|f
m
f
n
|
p
< , m, n N

.
One key feature, which will be heavily exploited in the next section, deals with
the Banach space p = 2, for which we have the following.
Proposition 3.5. Let (X, A, ) be a measure space, and let K be one of the
elds R or C.
(i) The map ( . [ . ) : L
2
K
(X, A, ) L
2
K
(X, A, ) K, given by
( f [ g ) =

f, g) =
_
X

fg d, f, g L
2
K
(X, A, ),
denes an inner product on L
2
K
(X, A, ).
(ii) One has the equality
|f|
2
=
_
( f [ f ), f L
2
K
(X, A, ).
Consequently , L
2
K
(X, A, ) is a Hilbert space.
Proof. The properties of the inner product are immediate, from the properties
of integration. The second property is also clear.
Remark 3.4. The main biproduct of the above feature is the fact that the
correspondence (6) is an isometric isomorphism, in the case p = q = 2. This
follows from Riesz Theorem (only the surjectivity is the issue here; the rest has
been discussed in Remark 3.3). If : L
2
K
(X, A, ) K is a linear continuous map,
then there exists some h L
2
K
(X, A, ), such that
(g) = ( h[ g ), g L
2
K
(X, A, ).
If we put f =

h, then the above equality gives
(g) = f, g), g L
2
K
(X, A, ).
i.e. =
f
.
Comments. Eventually (see Section 5) we shall prove that the correspondence
(6) is surjective also in the general case.
The correspondence (6) also has a version for q = 1. This would require the
denition of an L
p
space for the case p = . We shall postpone this until we reach
Section 5. The next exercise hints towards such a construction.
Exercise 4

. Let (X, A, ) be a measure space, let K be one of the elds R or C,

and let f : X K be a bounded measurable function. Dene M = sup
xX
[f(x)[.
Prove the following.
(i) Whenever g L
1
K
(X, A, ), it follows that the function fg also belongs to
L
1
K
(X, A, ), and one has the inequality
|fg|
1
M |g|
1
.
(ii) The map

f
: L
1
K
(X, A, ) g
_
X
fg d K
is linear and continuous. Moreover, one has the inequality |
f
| M.
3. L
p
spaces (1 p < ) 311
Remark 3.5. If we apply the above Exercise to the constant function f = 1,
we get the (already known) fact that the integration map
(9)
1
: L
1
K
(X, A, ) g
_
X
g d K
is linear and continuous, and has norm |
1
| 1. The follwing exercise gives the
exact value of the norm.
Exercise 5. With the notations above, prove that the following are equivalent:
(i) the measure space (X, A, ) is non-degenerate, i.e. there exists A A
with 0 < (A) < ;
(ii) L
1
K
(X, A, ) ,= 0;
(ii) the integration map (9) has norm |
1
| = 1.
Exercise 6

. Let (X, A, ) be a measure space, and let B A be a -algebra.

Consider the measure space (X, B

B
). Let K be either R or C. Prove that for
every p [1, ) one has the inclusion
L
p
K
(X, B,

B
) L
p
K
(X, A, ).
Prove that this inclusion gives rise to an isometric linear map
T : L
p
K
(X, B,

B
) L
p
K
(X, A, ).