H2p2sol Alt

NOTES ON
MEASURE THEORY
M. Papadimitrakis
Department of Mathematics
University of Crete
Autumn of 2004
2
Contents
1 -algebras 7
1.1 -algebras. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Generated -algebras. . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Borel -algebras. . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Algebras and monotone classes. . . . . . . . . . . . . . . . . . . . 13
1.5 Restriction of a -algebra. . . . . . . . . . . . . . . . . . . . . . . 16
1.6 Exercises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2 Measures 21
2.1 General measures. . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2 Point-mass distributions. . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Complete measures. . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4 Restriction of a measure. . . . . . . . . . . . . . . . . . . . . . . . 27
2.5 Uniqueness of measures. . . . . . . . . . . . . . . . . . . . . . . . 28
2.6 Exercises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3 Outer measures 33
3.1 Outer measures. . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Construction of outer measures. . . . . . . . . . . . . . . . . . . . 35
3.3 Exercises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4 Lebesgue-measure in R
n
41
4.1 Volume of intervals. . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2 Lebesgue-measure in R
n
. . . . . . . . . . . . . . . . . . . . . . . 43
4.3 Lebesgue-measure and simple transformations. . . . . . . . . . . 46
4.4 Cantors set. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.5 A non-Lebesgue-measurable set in R. . . . . . . . . . . . . . . . 52
4.6 Exercises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5 Borel measures 59
5.1 Lebesgue-Stieltjes-measures in R. . . . . . . . . . . . . . . . . . . 59
5.2 Borel measures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.3 Exercises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3
4 CONTENTS
6 Measurable functions 71
6.1 Measurability. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
6.2 Restriction and gluing. . . . . . . . . . . . . . . . . . . . . . . . . 71
6.3 Functions with arithmetical values. . . . . . . . . . . . . . . . . . 72
6.4 Composition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.5 Sums and products. . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.6 Absolute value and signum. . . . . . . . . . . . . . . . . . . . . . 76
6.7 Maximum and minimum. . . . . . . . . . . . . . . . . . . . . . . 77
6.8 Truncation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
6.9 Limits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.10 Simple functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6.11 The role of null sets. . . . . . . . . . . . . . . . . . . . . . . . . . 84
6.12 Exercises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
7 Integrals 91
7.1 Integrals of non-negative simple functions. . . . . . . . . . . . . . 91
7.2 Integrals of non-negative functions. . . . . . . . . . . . . . . . . . 94
7.3 Integrals of complex valued functions. . . . . . . . . . . . . . . . 97
7.4 Integrals over subsets. . . . . . . . . . . . . . . . . . . . . . . . . 104
7.5 Point-mass distributions. . . . . . . . . . . . . . . . . . . . . . . . 107
7.6 Lebesgue-integral. . . . . . . . . . . . . . . . . . . . . . . . . . . 110
7.7 Lebesgue-Stieltjes-integral. . . . . . . . . . . . . . . . . . . . . . . 117
7.8 Reduction to integrals over R. . . . . . . . . . . . . . . . . . . . 121
7.9 Exercises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
8 Product-measures 135
8.1 Product--algebra. . . . . . . . . . . . . . . . . . . . . . . . . . . 135
8.2 Product-measure. . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
8.3 Multiple integrals. . . . . . . . . . . . . . . . . . . . . . . . . . . 146
8.4 Surface-measure on S
n1
. . . . . . . . . . . . . . . . . . . . . . . 153
8.5 Exercises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
9 Convergence of functions 169
9.1 a.e. convergence and uniformly a.e. convergence. . . . . . . . . . 169
9.2 Convergence in the mean. . . . . . . . . . . . . . . . . . . . . . . 170
9.3 Convergence in measure. . . . . . . . . . . . . . . . . . . . . . . . 173
9.4 Almost uniform convergence. . . . . . . . . . . . . . . . . . . . . 176
9.5 Relations between types of convergence. . . . . . . . . . . . . . . 178
9.6 Exercises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
10 Signed measures and complex measures 187
10.1 Signed measures. . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
10.2 The Hahn and Jordan decompositions, I. . . . . . . . . . . . . . . 189
10.3 The Hahn and Jordan decompositions, II. . . . . . . . . . . . . . 195
10.4 Complex measures. . . . . . . . . . . . . . . . . . . . . . . . . . . 198
10.5 Integration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
CONTENTS 5
10.6 Lebesgue decomposition, Radon-Nikodym derivative. . . . . . . . 204
10.7 Dierentiation of indenite integrals in R
n
. . . . . . . . . . . . . 212
10.8 Dierentiation of Borel measures in R
n
. . . . . . . . . . . . . . . 218
10.9 Exercises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
11 The classical Banach spaces 225
11.1 Normed spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
11.2 The spaces L
p
(X, , ). . . . . . . . . . . . . . . . . . . . . . . . 232
11.3 The dual of L
p
(X, , ). . . . . . . . . . . . . . . . . . . . . . . . 243
11.4 The space M(X, ). . . . . . . . . . . . . . . . . . . . . . . . . . 250
11.5 Exercises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
6 CONTENTS
Chapter 1
-algebras
1.1 -algebras.
Denition 1.1 Let X be a non-empty set and a collection of subsets of X.
We call a -algebra of subsets of X if it is non-empty, closed under
complements and closed under countable unions. This means:
(i) there exists at least one A X so that A ,
(ii) if A , then A
c
, where A
c
= X A, and
(iii) if A
n
for all n N, then
+
n=1
A
n
.
The pair (X, ) of a non-empty set X and a -algebra of subsets of X is
called a measurable space.
Proposition 1.1 Every -algebra of subsets of X contains at least the sets
and X, it is closed under nite unions, under countable intersections, under
nite intersections and under set-theoretic dierences.
Proof: Let be any -algebra of subsets of X.
(a) Take any A and consider the sets A
1
= A and A
n
= A
c
for all n 2.
Then X = A A
c
=
+
n=1
A
n
and also = X
c
.
(b) Let A
1
, . . . , A
N
. Consider A
n
= A
N
for all n > N and get that
N
n=1
A
n
=
+
n=1
A
n
.
(c) Let A
n
for all n. Then
+
n=1
A
n
= (
+
n=1
A
c
n
)
c
.
(d) Let A
1
, . . . , A
N
. Using the result of (b), we get that
N
n=1
A
n
=
(
N
n=1
A
c
n
)
c
.
(e) Finally, let A, B . Using the result of (d), we get that AB = AB
c
.
Here are some simple examples.
Examples.
1. The collection , X is a -algebra of subsets of X.
2. If E X and E is non-empty and dierent from X, then the collection
, E, E
c
, X is a -algebra of subsets of X.
7
8 CHAPTER 1. -ALGEBRAS
3. T(X), the collection of all subsets of X, is a -algebra of subsets of X.
4. Let X be an uncountable set. The collection A X[ A is countable or A
c
is
countable is a -algebra of subsets of X. Firstly, is countable and, hence,
the collection is non-empty. If A is in the collection, then, considering cases, we
see that A
c
is also in the collection. Finally, let A
n
be in the collection for all
n N. If all A
n
s are countable, then
+
n=1
A
n
is also countable. If at least one
of the A
c
n
s, say A
c
n0
, is countable, then (
+
n=1
A
n
)
c
A
c
n0
is also countable. In
any case,
+
n=1
A
n
belongs to the collection.
The following result is useful.
Proposition 1.2 Let be a -algebra of subsets of X and consider a nite
sequence A
n
N
n=1
or an innite sequence A
n
in . Then there exists a nite
sequence B
n
N
n=1
or, respectively, an innite sequence B
n
in with the
properties:
(i) B
n
A
n
for all n = 1, . . . , N or, respectively, all n N,
(ii)
N
n=1
B
n
=
N
n=1
A
n
or, respectively,
+
n=1
B
n
=
+
n=1
A
n
and
(iii) the B
n
s are pairwise disjoint.
Proof: Trivial, by taking B
1
= A
1
and B
k
= A
k
(A
1
A
k1
) for all
k = 2, . . . , N or, respectively, all k = 2, 3, . . . .
1.2 Generated -algebras.
Proposition 1.3 The intersection of any -algebras of subsets of the same X
is a -algebra of subsets of X.
Proof: Let
i
iI
be any collection of -algebras of subsets of X, indexed by an
arbitrary non-empty set I of indices, and consider the intersection =
iI
i
.
(i) Since
i
for all i I, we get and, hence, is non-empty.
(ii) Let A . Then A
i
for all i I and, since all
i
s are -algebras,
A
c

i
for all i I. Therefore A
c
.
(iii) Let A
n
for all n N. Then A
n

i
for all i I and all n N
and, since all
i
s are -algebras, we get
+
n=1
A
n

i
for all i I. Thus,
+
n=1
A
n
.
Denition 1.2 Let X be a non-empty set and c be an arbitrary collection of
subsets of X. The intersection of all -algebras of subsets of X which include
c is called the -algebra generated by c and it is denoted by (c). Namely
(c) =
[ is a -algebra of subsets of X and c .

Note that there is at least one -algebra of subsets of X which includes c and
this is T(X). Note also that the term -algebra used in the name of (c) is
justied by its denition and by Proposition 1.3.
1.3. BOREL -ALGEBRAS. 9
Proposition 1.4 Let c be any collection of subsets of the non-empty X. Then
(c) is the smallest -algebra of subsets of X which includes c. Namely, if
is any -algebra of subsets of X such that c , then (c) .
Proof: If is any -algebra of subsets of X such that c , then is one of
the -algebras whose intersection is denoted (c). Therefore (c) .
Looking back at two of the examples of -algebras, we easily get the following
examples.
Examples.
1. Let E X and E be non-empty and dierent from X and consider c = E.
Then (c) = , E, E
c
, X. To see this just observe that , E, E
c
, X is a
-algebra of subsets of X which contains E and that there can be no smaller
-algebra of subsets of X containing E, since such a -algebra must necessarily
contain , X and E
c
besides E.
2. Let X be an uncountable set and consider c = A X[ A is countable.
Then (c) = A X[A is countable or A
c
is countable. The argument is the
same as before. A X[A is countable or A
c
is countable is a -algebra of
subsets of X which contains all countable subsets of X and there is no smaller
-algebra of subsets of X containing all countable subsets of X, since any such
-algebra must contain all the complements of countable subsets of X.
The next section describes a much more important example.
1.3 Borel -algebras.
Denition 1.3 Let X be a topological space and T the topology of X, i.e. the
collection of all open subsets of X. The -algebra of subsets of X which is
generated by T , namely the smallest -algebra of subsets of X containing all
open subsets of X, is called the Borel -algebra of X and we denote it B
X
:
B
X
= (T ) , T the topology of X.
The elements of B
X
are called Borel sets in X and B
X
is also called the
-algebra of Borel sets in X.
By denition, all open subsets of X are Borel sets in X and, since B
X
is a
-algebra, all closed subsets of X (which are the complements of open subsets)
are also Borel sets in X. A subset of X is called a G
-set if it is a countable
intersection of open subsets of X. Also, a subset of X is called an F
-set if it
is a countable union of closed subsets of X. It is obvious that all G
-sets and
all F
-sets are Borel sets in X.

Proposition 1.5 If X is a topological space and T is the collection of all closed
subsets of X, then B
X
= (T).
Proof: Every closed set is contained in (T ). This is true because (T ) contains
all open sets and hence, being a -algebra, contains all closed sets. Therefore,
T (T ). Since (T ) is a -algebra, Proposition 1.4 implies (T) (T ).
Symmetrically, every open set is contained in (T). This is because (T)
contains all closed sets and hence, being a -algebra, contains all open sets (the
complements of closed sets). Therefore, T (T). Since (T) is a -algebra,
Proposition 1.4 implies (T ) (T).
Therefore, (T) = (T ) = B
X
.
Examples of topological spaces are all metric spaces of which the most fa-
miliar is the euclidean space X = R
n
with the usual euclidean metric or even
any subset X of R
n
with the restriction on X of the euclidean metric. Because
of the importance of R
n
we shall pay particular attention on B
R
n.
The typical closed orthogonal parallelepiped with axis-parallel edges is a set of
the form Q = [a
1
, b
1
] [a
n
, b
n
], the typical open orthogonal parallelepiped
with axis-parallel edges is a set of the formR = (a
1
, b
1
) (a
n
, b
n
), the typical
open-closed orthogonal parallelepiped with axis-parallel edges is a set of the form
P = (a
1
, b
1
] (a
n
, b
n
] and the typical closed-open orthogonal parallelepiped
with axis-parallel edges is a set of the form T = [a
1
, b
1
) [a
n
, b
n
). More
generally, the typical orthogonal parallelepiped with axis-parallel edges is a set
S, a cartesian product of n bounded intervals of any possible type. In all cases
we consider < a
j
b
j
< + for all j = 1, . . . , n and, hence, all orthogonal
parallelepipeds with axis-parallel edges are bounded sets in R
n
.
If n = 1, then the orthogonal parallelepipeds with axis-parallel edges are
just the bounded intervals of all possible types in the real line R. If n = 2, then
the orthogonal parallelepipeds with axis-parallel edges are the usual orthogonal
parallelograms of all possible types with axis-parallel sides.
Since orthogonal parallelepipeds with axis-parallel edges will play a role in
much of the following, we agree to call them, for short, n-dimensional inter-
vals or intervals in R
n
.
Lemma 1.1 All n-dimensional intervals are Borel sets in R
n
.
Proof: For any j = 1, . . . , n, a half-space of the formx = (x
1
, . . . , x
n
) [ x
j
< b
j
or of the formx = (x
1
, . . . , x
n
) [ x
j
b
j
is a Borel set in R
n
, since it is an open
set in the rst case and a closed set in the second case. Similarly, a half-space of
the form x = (x
1
, . . . , x
n
) [ a
j
< x
j
or of the form x = (x
1
, . . . , x
n
) [ a
j
x
j
is a Borel set in R
n
. Now, every interval S is an intersection of 2n of these
half-spaces and, therefore, it is also a Borel set in R
n
.
Proposition 1.6 If c is the collection of all closed or of all open or of all
open-closed or of all closed-open or of all intervals in R
n
, then B
R
n = (c).
Proof: By Lemma 1.1 we have that, in all cases, c B
R
n. Proposition 1.4
implies that (c) B
R
n.
To show the opposite inclusion consider any open subset U of R
n
. For every
x U nd a small open ball B
x
centered at x which is included in U. Now,
1.3. BOREL -ALGEBRAS. 11
considering the case of c being the collection of all closed intervals, take an
arbitrary Q
x
= [a
1
, b
1
] [a
n
, b
n
] containing x, small enough so that it is
included in B
x
, and hence in U, and with all a
1
, . . . , a
n
, b
1
, . . . , b
n
being rational
numbers. Since x Q
x
U for all x U, we have that U =
xU
Q
x
. But the
collection of all possible Q
x
s is countable (!) and, thus, the general open subset
U of R
n
can be written as a countable union of sets in the collection c. Hence
every open U belongs to (c) and, since (c) is a -algebra of subsets of R
n
and B
R
n is generated by the collection of all open subsets of R
n
, Proposition
1.4 implies that B
R
n (c).
Of course the proof of the last inclusion works in the same way with all other
types of intervals.
It is convenient for certain purposes, and especially because functions are
often innitely valued, to consider R = R +, and C = C as
topological spaces and dene their Borel -algebras.
The -neighborhood of a point x Ris, as usual, the interval (x, x+) and
we dene the -neighborhood of +to be (
1
, +] and of to be [,
1
).
We next say that U R is open in R if for every point of U there is an
-neighborhood of the point included in U. It is trivial to see (justifying the
term open) that the collection of all sets open in R is a topology of R, namely
that it contains the sets and R and that it is closed under arbitrary unions
and under nite intersections. It is obvious that a set U R is open in R if
and only if it is open in R. It is also obvious that, if a set U R is open in R,
then U R is open in R.
The next result says, in particular, that we may construct the general Borel
set in R by taking the general Borel set in R and adjoining none or any one or
both of the points +, to it.
Proposition 1.7 We have
B
R
=
_
A, A +, A , A +, [ A B
R
_
.
Also, if c is the collection containing + or and all closed or all open
or all open-closed or all closed-open or all intervals in R, then B
R
= (c).
Proof: (a) Consider the collection = A R[ A B
R
. This collection
obviously contains . If A , then R A B
R
and, since R is open in R, we
get R A = (R A) R B
R
. Hence, R A . If A
n
for all n N,
then all A
n
s are included in R and are contained in B
R
. Therefore
+
n=1
A
n
is
included in R and it is contained in B
R
and, hence,
+
n=1
A
n
. This proves
that is a -algebra of subsets of R. We now observe that all open subsets of
R are also open subsets of R and, hence, belong to . Proposition 1.4 implies
that all Borel sets in R belong to and, by denition of , we get that all
A B
R
are contained in B
R
.
The set [, +) is open in R and, hence, the set + is contained in
B
R
. Similarly, and, hence, +, are contained in B
R
.
We conclude that
_
A, A+, A, A+, [ A B
R
_
B
R
.
If U is open in R, then A = U R is open in R and, thus, U can be written
U = A or U = A + or U = A or U = A +, for some
A which is open in R. This means that all sets open in R are contained in the
collection
_
A, A +, A , A +, [ A B
R
_
. It is a trivial
matter to prove that this collection is a -algebra of subsets of R and, hence,
by Proposition 1.4, B
R

_
A, A+, A, A+, [ A B
R
_
.
Therefore, the rst statement of this proposition is proved.
(b) Let c =
_
+, (a, b] [ < a b < +
_
.
We have already seen that + B
R
and since (a, b] = (a, b+1) (b, b+1)
is the dierence of two open sets in R we get that (a, b] B
R
. Hence c B
R
and, by Proposition 1.4, (c) B
R
.
As we have seen in the proof of Proposition 1.6, every open set A in R is a
countable union of intervals of the form (a, b]. Therefore, every open set A in
R is contained in (c).
In particular, the set R is contained in (c) and, hence, (, +] =
R+ is contained in (c). Thus, also = R (, +] belongs to
(c).
In the proof of (a) we have seen that every U open in R can be written as
U = A or U = A+ or U = A or U = A+, for some A
which is open in R. By the last two paragraphs, every U open in R is contained
in (c) and Proposition 1.4 implies that B
R
(c).
This concludes the proof of the second statement for this particular choice
of c and the proof is similar for all other choices.
We now turn to the case of C = C . The -neighborhood of a point
x = (x
1
, x
2
) = x
1
+ix
2
C is, as usual, the open disc B(x; ) = y = (y
1
, y
2
)
C[ [y x[ < , where [y x[
2
= (y
1
x
1
)
2
+ (y
2
x
2
)
2
. We dene the -
neighborhood of to be the set y C[ [y[ >
1
, the exterior of a
closed disc centered at 0 together with the point . We say that a set U C
is open in C if for every point of U there is an -neighborhood of the point
included in U. The collection of all sets which are open in C contains and C
and is closed under arbitrary unions and under nite intersections, thus forming
a topology in C. It is clear that a set U C is open in C if and only if it is
open in C and that, if a set U C is open in C, then U C is open in C.
As in the case of R, we may construct the general Borel set in C by taking
the general Borel set in C and at most adjoining the point to it.
Proposition 1.8 We have
B
C
=
_
A, A [ A B
C
_
.
Also, if c is the collection of all closed or all open or all open-closed or all
closed-open or all intervals in C = R
2
, then B
C
= (c).
Proof: The proof is very similar to (and slightly simpler than) the proof of
Proposition 1.7. The steps are the same and only minor modications are
needed.
1.4. ALGEBRAS AND MONOTONE CLASSES. 13
1.4 Algebras and monotone classes.
Denition 1.4 Let X be non-empty and / a collection of subsets of X. We call
/ an algebra of subsets of X if it is non-empty, closed under complements
and closed under unions. This means:
(i) there exists at least one A X so that A /,
(ii) if A /, then A
c
/ and
(iii) if A, B /, then A B /.
Proposition 1.9 Every algebra of subsets of X contains at least the sets
and X, it is closed under nite unions, under nite intersections and under
set-theoretic dierences.
Proof: Let / be any algebra of subsets of X.
(a) Take any A / and consider the sets A and A
c
. Then X = A A
c
/
and then = X
c
/.
(b) It is trivial to prove by induction that for any n N and any A
1
, . . . , A
n
/
it follows A
1
A
n
/.
(c) By the result of (b), if A
1
, . . . , A
n
/, then
n
k=1
A
k
= (
n
k=1
A
c
k
)
c
/.
(d) If A, B /, using the result of (c), we get that A B = A B
c
/.
Examples.
1. Every -algebra is also an algebra.
2. If X is an innite set then the collection A X[ A is nite or A
c
is nite
is an algebra of subsets of X.
We have already dealt with the (n-dimensional) intervals in R
n
, which are
cartesian products of n bounded intervals in R. If we allow these intervals to
become unbounded, we get the so-called generalized intervals in R
n
, namely
all sets of the form I
1
I
n
, where each I
j
is any, even unbounded, interval
in R. Again, we have the subcollections of all open or all closed or all open-
closed or all closed-open generalized intervals. For example, the typical open-
closed generalized interval in R
n
is of the form P = (a
1
, b
1
] (a
n
, b
n
],
where a
j
b
j
+ for all j. The whole space R
n
is an open-closed
generalized interval, as well as any of the half spaces x = (x
1
, . . . , x
n
) [ x
j
b
j
and x = (x
1
, . . . , x
n
) [ a
j
< x
j
. In fact, every open-closed generalized interval
is, obviously, the intersection of 2n such half-spaces.
Proposition 1.10 The collection / = P
1
P
k
[ k N, P
1
, . . . , P
k
are
pairwise disjoint open-closed generalized intervals in R
n
is an algebra of sub-
sets of R
n
.
In particular, the following are true:
(i) The intersection of two open-closed generalized intervals is an open-closed
generalized interval.
(ii) For all open-closed generalized intervals P, P
1
, . . . , P
m
there are pairwise dis-
joint open-closed generalized intervals P
1
, . . . , P
k
so that P (P
1
P
m
) =
P
1
P
k
.
(iii) For all open-closed generalized intervals P
1
, . . . , P
m
there are pairwise dis-
joint open-closed generalized intervals P
1
, . . . , P
k
so that P
1
P
m
= P
1

P
k
.
Proof: (a) The intervals (a, b] and (a
, b
] are not disjoint if and only if a
< b
,
where a
= max(a, a
) and b
= min(b, b
). In case a
< b
, then (a, b](a
, b
] =
(a
, b
]. Now if P = (a
1
, b
1
] (a
n
, b
n
] and P
= (a
1
, b
1
] (a
n
, b
n
],
then P and P
are not disjoint if and only if for all j = 1, . . . , n we have that

(a
j
, b
j
] and (a
j
, b
j
] are not disjoint. Hence if P, P
are not disjoint, then a
j
< b
j
for all j, where a
j
= max(a
j
, a
j
) and b
j
= min(b
j
, b
j
), and then P P
= P
,
where P
= (a
1
, b
1
] (a
n
, b
n
]. This proves (i).
If A =
k
i=1
P
i
, where the P
1
, . . . , P
k
are pairwise disjoint, and A
=
l
j=1
P
j
,
where the P
1
, . . . , P
l
are also pairwise disjoint, are two elements of /, then
A A
1ik,1jl
P
i
P
j
. The sets P
i
P
j
are pairwise disjoint and they
all are open-closed generalized intervals, as we have just seen.
Hence, / is closed under nite intersections.
(b) Consider the open-closed generalized interval P = (a
1
, b
1
] (a
n
, b
n
]. It
is easy to see that P
c
can be written as the union of 2n (some may be empty)
pairwise disjoint open-closed generalized intervals. To express this in a concise
way, for every I = (a, b] denote I
(l)
= (, a] and I
(r)
= (b, +] the left
and right compementary intervals of I in R (they may be empty). If we write
P = I
1
I
n
, then P
c
is equal to
I
(l)
1
R R I
(r)
1
R R
I
1
I
(l)
2
R R I
1
I
(r)
2
R R

I
1
I
n2
I
(l)
n1
R I
1
I
n2
I
(r)
n1
R
I
1
I
n1
I
(l)
n
I
1
I
n1
I
(r)
n
.
Hence, for every open-closed generalized interval P the complement P
c
is an
element of /.
Now, if A =
k
i=1
P
i
, where the P
1
, . . . , P
k
are pairwise disjoint, is any ele-
ment of /, then A
c
=
k
i=1
P
c
i
is a nite intersection of elements (P
c
i
s) of /.
Because of the result of (a), A
c
/ and / is closed under complements.
(c) If A, A
/, then, because of the results of (a) and (b), AA
= (A
c
A
c
)
c
/ and / is closed under nite unions.

Therefore / is an algebra and (ii) and (iii) are immediate.
If A
n
is a sequence of subsets of a set X and A
n
A
n+1
for all n, we say
that the sequence is increasing. In this case, if A =
+
n=1
A
n
, we write
A
n
A.
If A
n+1
A
n
for all n, we say that the sequence A
n
is decreasing and, if
also A =
+
n=1
A
n
, we write
A
n
A.
1.4. ALGEBRAS AND MONOTONE CLASSES. 15
Denition 1.5 Let X be a non-empty set and / a collection of subsets of X.
We call / a monotone class of subsets of X if it is closed under countable
increasing unions and closed under countable decreasing intersections. That is,
if A
1
, A
2
, . . . / and A
n
A, then A / and, if A
1
, A
2
, . . . / and
A
n
A, then A /.
It is obvious that every -algebra is a non-empty monotone class.
Proposition 1.11 The intersection of any monotone classes of subsets of the
same set X is a monotone class of subsets of X.
Proof: Let /
i
iI
be any collection of monotone classes of subsets of X,
indexed by an arbitrary non-empty set I of indices, and consider the intersection
/=
iI
/
i
.
Let A
1
, A
2
, . . . / with A
n
A. Then A
n
/
i
for all i I and all n N
and, since all /
i
s are monotone classes, we get that A /
i
for all i I.
Therefore A /.
The proof in the case of a countable decreasing intersection is identical.
Denition 1.6 Let X be a non-empty set and c be an arbitrary collection of
subsets of X. Then the intersection of all monotone classes of subsets of X
which include c is called the monotone class generated by c and it is
denoted by /(c). Namely
/(c) =
/[ / is a monotone class of subsets of X and c /.

There is at least one monotone class including c and this is T(X). Also note
that the term monotone class, used for /(c), is justied by Proposition 1.11.
Proposition 1.12 Let c be any collection of subsets of the non-empty X. Then
/(c) is the smallest monotone class of subsets of X which includes c. Namely,
if / is any monotone class of subsets of X such that c /, then /(c) /.
Proof: If / is any monotone class of subsets of X such that c /, then /
is one of the monotone classes whose intersection is /(c). Thus, /(c) /.
Theorem 1.1 Let X be a non-empty set and / an algebra of subsets of X.
Then /(/) = (/).
Proof: (/) is a -algebra and, hence, a monotone class. Since / (/),
Proposition 1.12 implies /(/) (/).
Now it is enough to prove that /(/) is a -algebra. Since / /(/),
Proposition 1.4 will immediately imply that (/) /(/) and this will con-
clude the proof.
(a) /(/) is non-empty because / /(/).
(b) Fix any A / and consider the collection /
A
= B X[ AB /(/).
It is very easy to show that /
A
includes / and that it is a monotone class
of subsets of X. In fact, if B / then A B / and thus B /
A
. Also, if
B
1
, B
2
, . . . /
A
and B
n
B, then A B
1
, A B
2
, . . . /(/) and A B
n

A B. Since /(/) is a monotone class, we nd that A B /(/). Thus,
B /
A
and /
A
is closed under countable increasing unions. In a similar way
we can prove that /
A
is closed under countable decreasing intersections and
we conclude that it is a monotone class.
Proposition 1.12 implies that /(/) /
A
. This means that:
i. A B /(/) for all A / and all B /(/).
Now x any B /(/) and consider /
B
= A X[ A B /(/). As
before, /
B
is a monotone class of subsets of X and, by i., it includes /. Again,
Proposition 1.12 implies /(/) /
B
, which means:
ii. A B /(/) for all A /(/) and all B /(/).
(c) We consider the collection / = A X[ A
c
/(/). As before, we
can show that / is a monotone class of subsets of X and that it includes /.
Therefore, /(/) /, which means:
iii. A
c
/(/) for all A /(/).
It is implied by ii. and iii. that /(/) is closed under nite unions and
under complements.
(d) Now take A
1
, A
2
, . . . /(/) and dene B
n
= A
1
A
n
for all n. From
ii. we have that B
n
/(/) for all n and it is clear that B
n
B
n+1
for all n.
Since /(/) is a monotone class,
+
n=1
A
n
=
+
n=1
B
n
/(/).
Hence, /(/) is a -algebra.
1.5 Restriction of a -algebra.
Proposition 1.13 Let be a -algebra of subsets of X and Y X be non-
empty. If we denote
Y = A Y [ A ,
then Y is a -algebra of subsets of Y .
Proof: Since , we have that = Y Y .
If B Y , then B = A Y for some A . Since X A , we get that
Y B = (X A) Y Y .
If B
1
, B
2
, . . . Y , then, for each k, B
k
= A
k
Y for some A
k
. Since
+
k=1
A
k
, we nd that
+
k=1
B
k
= (
+
k=1
A
k
) Y Y .
Denition 1.7 Let be a -algebra of subsets of X and let Y X be non-
empty. The -algebra Y , dened in Proposition 1.13, is called the restric-
tion of on Y .
In general, if c is any collection of subsets of X and Y X, we denote
cY = A Y [ A c
1.5. RESTRICTION OF A -ALGEBRA. 17
and call cY the restriction of c on Y .
Theorem 1.2 Let c be a collection of subsets of X and Y X be non-empty.
Then
(cY ) = (c)Y,
where (cY ) is the -algebra of subsets of Y generated by cY .
Proof: (a) If B cY , then B = A Y for some A c (c) and, thus,
B (c)Y . Hence, cY (c)Y and, since, by Proposition 1.13, (c)Y is
a -algebra of subsets of Y , Proposition 1.4 implies (cY ) (c)Y .
(b) Now, dene the collection
= A X[ A Y (cY ).
We have that , because Y = (cY ).
If A , then AY (cY ). Therefore, XA , because (XA)Y =
Y (A Y ) (cY ).
If A
1
, A
2
, . . . , then A
1
Y, A
2
Y, . . . (cY ). This implies that
(
+
k=1
A
k
) Y =
+
k=1
(A
k
Y ) (cY ) and, thus,
+
k=1
A
k
.
We conclude that is a -algebra of subsets of X.
If A c, then AY cY (cY ) and, hence, A . Therefore, c
and, by Proposition 1.4, (c) . Now, for an arbitrary B (c)Y , we have
that B = A Y for some A (c) and, thus, B (cY ). This implies
that (c)Y (cY ).
If X is a topological space with the topology T and if Y X, then, as is well-
known (and easy to prove), the collection T Y = U Y [ U T is a topology
of Y which is called the relative topology or the subspace topology of Y .
Theorem 1.3 Let X be a topological space and let the non-empty Y X have
the subspace topology. Then
B
Y
= B
X
Y.
Proof: If T is the topology of X, then T Y is the subspace topology of Y .
Theorem 1.2 implies that B
Y
= (T Y ) = (T )Y = B
X
Y .
Thus, the Borel sets in the subset Y of X (with the subspace topology) are
just the intersections with Y of the Borel sets in X.
Example.
It is clear from Propositions 1.7 and 1.8 that
B
R
= B
R
R and B
C
= B
C
C.
These two equalities are also justied by Theorem 1.3, since the topology of
R coincides with its subspace topology as a subset of R and the topology of C
coincides with its subspace topology as a subset of C.
1.6 Exercises.
1. Let X be a non-empty set and A
1
, A
2
, . . . X. We dene
limsup
n+
A
n
=
+
k=1
_
+
j=k
A
j
_
, liminf
n+
A
n
=
+
k=1
_
+
j=k
A
j
_
.
Only in case liminf
n+
A
n
= limsup
n+
A
n
, we dene
lim
n+
A
n
= liminf
n+
A
n
= limsup
n+
A
n
.
Prove the following.
(i) limsup
n+
A
n
= x X[ x A
n
for innitely many values of n.
(ii) liminf
n+
A
n
= x X[ x A
n
for all large enough n.
(iii) (liminf
n+
A
n
)
c
= limsup
n+
A
c
n
and (limsup
n+
A
n
)
c
=
liminf
n+
A
c
n
.
(iv) liminf
n+
A
n
limsup
n+
A
n
.
(v) If A
n
A
n+1
for all n, then lim
n+
A
n
=
+
n=1
A
n
.
(vi) If A
n+1
A
n
for all n, then lim
n+
A
n
=
+
n=1
A
n
.
(vii) Find an example where liminf
n+
A
n
,= limsup
n+
A
n
.
(viii) If A
n
B
n
for all n, then limsup
n+
A
n
limsup
n+
B
n
and
liminf
n+
A
n
liminf
n+
B
n
.
(ix) If A
n
= B
n
C
n
for all n, then limsup
n+
A
n
limsup
n+
B
n
limsup
n+
C
n
, liminf
n+
B
n
liminf
n+
C
n
liminf
n+
A
n
.
2. Let / be an algebra of subsets of X. Prove that / is a -algebra if and
only if it is closed under increasing unions.
3. The inclusion-exclusion formula.
Let (X, , ) be a measure space. Prove that for all n and A
1
, . . . , A
n

(
n
j=1
A
j
) +
k even
1i1<<i
k
n
(A
i1
A
i
k
)
=
k odd
1i1<<i
k
n
(A
i1
A
i
k
).
4. Let X be non-empty. In the next three cases nd (c) and /(c).
(i) c = .
(ii) Fix E X and let c = F [ E F X.
(iii) Let c = F [ F is a two-point-subset of X.
5. Let c
1
, c
2
be two collections of subsets of the non-empty X. If c
1
c
2

(c
1
), prove that (c
1
) = (c
2
).
6. Let Y be a non-empty subset of X.
(i) If / is an algebra of subsets of X, prove that /Y is an algebra of
subsets of Y .
1.6. EXERCISES. 19
(ii) If / is a monotone class of subsets of X, prove that /Y is a mono-
tone class of subsets of Y .
(iii) If T is a topology of X, prove that T Y is a topology of Y .
7. (i) Let be a -algebra of subsets of X and let Y X be non-empty. If
Y , prove that Y = A Y [ A .
(ii) Let X be a topological space and Y be a non-empty Borel set in X.
Prove that B
Y
= A Y [ A B
X
.
8. Push-forward of a -algebra.
Let be a -algebra of subsets of X and let f : X Y . Then the
collection
B Y [ f
1
(B)
is called the push-forward of by f on Y .
(i) Prove that the collection B Y [ f
1
(B) is a -algebra of
subsets of Y .
Consider also a -algebra
of subsets of Y and a collection c of subsets

of Y so that (c) =
.
(ii) Prove that, if f
1
(B) for all B c, then f
1
(B) for all
B
.
(iii) If X, Y are two topological spaces and f : X Y is continuous, prove
that f
1
(B) is a Borel set in X for every Borel set B in Y .
9. The pull-back of a -algebra.
Let
be a -algebra of subsets of Y and let f : X Y . Then the

collection
f
1
(B) [ B
is called the pull-back of
by f on X.
Prove that f
1
(B) [ B
is a -algebra of subsets of X.
10. (i) Prove that B
R
n is generated by the collection of all half-spaces in R
n
of the form x = (x
1
, . . . , x
n
) [ a
j
< x
j
, where j = 1, . . . , n and a
j
R.
(ii) Prove that B
R
n is generated by the collection of all open balls B(x; r)
or of all closed balls B(x; r), where x R
n
and r R
+
.
11. (i) Prove that B
R
is generated by the collection of all (a, +], where
a R.
(ii) Prove that B
C
is generated by the collection of all open discs B(x; r)
or of all closed discs B(x; r), where x C and r R
+
.
12. Let X be a metric space with metric d. Prove that every closed F X is a
G
-set by considering the sets U

n
= x X[ d(x, y) <
1
n
for some y F.
Prove, also, that every open U X is an F
-set.
13. (i) Suppose that f : R
n
R. Prove that x R
n
[ f is continuous at x
is a G
-set in R
n
.
(ii) Suppose that f
k
: R
n
R is continuous in R
n
for every k. Prove that
x R
n
[ f
k
(x) converges is an F
-set, i.e. a countable intersection

of F
-sets.
14. Let c be an arbitrary collection of subsets of the non-empty X. Prove
that for every A (c) there is some countable subcollection T c so
that A (T).
Chapter 2
Measures
2.1 General measures.
Denition 2.1 Let (X, ) be a measurable space. A function : [0, +]
is called a measure on (X, ) or, simply, a measure on if
(i) () = 0,
(ii) (
+
n=1
A
n
) =
+
n=1
(A
n
) for all sequences A
n
of pairwise disjoint sets
which are contained in .
The triple (X, , ) of a non-empty set X, a -algebra of subsets of X and
a measure on is called a measure space.
Note that the values of a measure are non-negative real numbers or +.
Property (ii) of a measure is called -additivity and sometimes a mea-
sure is also called -additive measure to distinguish from a so-called nitely
additive measure which is dened to satisfy () = 0 and (
N
n=1
A
n
) =
N
n=1
(A
n
) for all N N and all pairwise disjoint A
1
, . . . , A
N
.
Proposition 2.1 Every measure is nitely additive.
Proof: Let be a measure on the -algebra . If A
1
, . . . , A
N
are pair-
wise disjoint, we consider A
n
= for all n > N and we get (
N
n=1
A
n
) =
(
+
n=1
A
n
) =
+
n=1
(A
n
) =
N
n=1
(A
n
).
Examples.
1. The simplest measure is the zero measure which is denoted o and is dened
by o(A) = 0 for every A .
2. Let X be an uncountable set and consider = A X[ A is countable or A
c
is countable. We dene (A) = 0 if A is countable and (A) = 1 if A
c
is count-
able.
Then it is clear that () = 0 and let A
1
, A
2
, . . . be pairwise dis-
joint. If all of them are countable, then
+
n=1
A
n
is also countable and we get
(
+
n=1
A
n
) = 0 =
+
n=1
(A
n
). Observe that if one of the A
n
s, say A
n0
, is
21
22 CHAPTER 2. MEASURES
uncountable, then for all n ,= n
0
we have A
n
A
c
n0
which is countable. There-
fore (A
n0
) = 1 and (A
n
) = 0 for all n ,= n
0
. Since (
+
n=1
A
n
)
c
( A
c
n0
) is
countable, we get (
+
n=1
A
n
) = 1 =
+
n=1
(A
n
).
Theorem 2.1 Let (X, , ) be a measure space.
(i) (Monotonicity) If A, B and A B, then (A) (B).
(ii) If A, B , A B and (A) < +, then (B A) = (B) (A).
(iii) (-subadditivity) If A
1
, A
2
, . . . , then (
+
n=1
A
n
)
+
n=1
(A
n
).
(iv) (Continuity from below) If A
1
, A
2
, . . . and A
n
A, then (A
n
) (A).
(v) (Continuity from above) If A
1
, A
2
, . . . , (A
1
) < + and A
n
A, then
(A
n
) (A).
Proof: (i) We write B = A (B A). By nite additivity of , (B) =
(A) +(B A) (A).
(ii) From both sides of (B) = (A) +(B A) we subtract (A).
(iii) Using Proposition 1.2 we nd B
1
, B
2
, . . . which are pairwise disjoint and
satisfy B
n
A
n
for all n and
+
n=1
B
n
=
+
n=1
A
n
. By -additivity and mono-
tonicity of we get (
+
n=1
A
n
) = (
+
n=1
B
n
) =
+
n=1
(B
n
)
+
n=1
(A
n
).
(iv) We write A = A
1
+
k=1
(A
k+1
A
k
), where all sets whose union is taken in
the right side are pairwise disjoint. Applying -additivity (and nite additivity),
(A) = (A
1
) +
+
k=1
(A
k+1
A
k
) = lim
n+
[(A
1
) +
n1
k=1
(A
k+1
A
k
)] =
lim
n+
_
A
1
n1
k=1
(A
k+1
A
k
)
_
= lim
n+
(A
n
).
(v) We observe that A
1
A
n
A
1
A and continuity from below implies
(A
1
A
n
) (A
1
A). Now, (A
1
) < + implies (A
n
) < + for all n
and (A) < +. Applying (ii), we get (A
1
) (A
n
) (A
1
) (A) and,
since (A
1
) < +, we nd (A
n
) (A).
Denition 2.2 Let (X, , ) be a measure space.
(i) is called nite if (X) < +.
(ii) is called -nite if there exist X
1
, X
2
, . . . so that X =
+
n=1
X
n
and
(X
n
) < + for all n N.
(iii) is called seminite if for every E with (E) = + there is an
F so that F E and 0 < (F) < +.
(iv) A set E is called of nite (-)measure if (E) < +.
(v) A set E is called of -nite (-)measure if there exist E
1
, E
2
, . . .
so that E
+
n=1
E
n
and (E
n
) < + for all n.
Some observations related to the last denition are immediate.
1. If is nite then all sets in are of nite -measure. More generally, if
E is of nite -measure, then all subsets of it in are of nite -measure.
2. If is -nite then all sets in are of -nite -measure. More generally,
if E is of -nite -measure, then all subsets of it in are of -nite
-measure.
3. The collection of sets of nite -measure is closed under nite unions.
4. The collection of sets of -nite -measure is closed under countable unions.
5. If is -nite, applying Proposition 1.2, we see that there exist pairwise
disjoint X
1
, X
2
, . . . so that X =
+
n=1
X
n
and (X
n
) < + for all n.
2.2. POINT-MASS DISTRIBUTIONS. 23
Similarly, by taking successive unions, we see that there exist X
1
, X
2
, . . . so
that X
n
X and (X
n
) < + for all n. We shall use these two observations
freely whenever -niteness appears in the sequel.
6. If is nite, then it is also -nite. The next result is not so obvious.
Proposition 2.2 Let (X, , ) be a measure space. If is -nite, then it is
seminite.
Proof: Take X
1
, X
2
, . . . so that X
n
X and (X
n
) < + for all n. Let
E have (E) = +. From E X
n
E and continuity of from below, we
get (E X
n
) +. Therefore, (E X
n0
) > 0 for some n
0
and we observe
that (E X
n0
) (X
n0
) < +.
Denition 2.3 Let (X, , ) be a measure space. E is called (-)null if
(E) = 0.
The following is trivial but basic.
Theorem 2.2 Let (X, , ) be a measure space.
(i) If E is -null, then every subset of it in is also -null.
(ii) If E
1
, E
2
, . . . are all -null, then their union
+
n=1
E
n
is also -null.
Proof: The proof is based on the monotonicity and the -subadditivity of .
2.2 Point-mass distributions.
Before introducing a particular class of measures we shall dene sums of non-
negative terms over general sets of indices. We shall follow the standard practice
of using both notations a(i) and a
i
for the values of a function a on a set I of
indices.
Denition 2.4 Let I be a non-empty set of indices and a : I [0, +]. We
dene the sum of the values of a by
iI
a
i
= sup
_
iF
a
i
[ F non-empty nite subset of I
_
.
If I = , we dene
iI
a
i
= 0.
Of course, if F is a non-empty nite set, then
iF
a
i
is just equal to the sum
N
k=1
a
i
k
, where F = a
i1
, . . . , a
iN
is an arbitrary enumeration of F.
We rst make sure that this denition extends a simpler situation.
Proposition 2.3 If I is countable and I = i
1
, i
2
, . . . is an arbitrary enume-
ration of it, then
iI
a
i
=
+
k=1
a
i
k
for all a : I [0, +].
Proof: For arbitrary N we consider the nite subset F = i
1
, . . . , i
N
of I.
Then, by the denition of
iI
a
i
, we have
N
k=1
a
i
k
=
iF
a
i

iI
a
i
.
Since N is arbitrary, we nd
+
k=1
a
i
k

iI
a
i
.
Now for an arbitrary non-empty nite F I we consider the indices of
the elements of F provided by the enumeration I = i
1
, i
2
, . . . and take the
maximal, say N, of them. This means that F i
1
, i
2
, . . . , i
N
. Therefore
iF
a
i

N
k=1
a
i
k

+
k=1
a
i
k
and, since F is arbitrary, by the denition of
iI
a
i
, we nd that
iI
a
i

+
k=1
a
i
k
.
Proposition 2.4 Let a : I [0, +]. If
iI
a
i
< +, then a
i
< + for
all i and the set i I [ a
i
> 0 is countable.
Proof: Let
iI
a
i
< +. It is clear that a
i
< +for all i (take F = i) and,
for arbitrary n, consider the set I
n
= i I [ a
i

1
n
. If F is an arbitrary nite
subset of I
n
, then
1
n
card(F)

iF
a
i

iI
a
i
. Therefore, the cardinality
of the arbitrary nite subset of I
n
is not larger than the number n
iI
a
i
and,
hence, the set I
n
is nite. But then, i I [ a
i
> 0 =
+
n=1
I
n
is countable.
Proposition 2.5 (i) If a, b : I [0, +] and a
i
b
i
for all i I, then
iI
a
i

iI
b
i
.
(ii) If a : I [0, +] and J I, then
iJ
a
i

iI
a
i
.
Proof: (i) For arbitrary nite F I we have
iF
a
i

iF
b
i

iI
b
i
.
Taking supremum over the nite subsets of I, we nd
iI
a
i

iI
b
i
.
(ii) For arbitrary nite F J we have that F I and hence
iF
a
i

iI
a
i
.
Taking supremum over the nite subsets of J, we get
iJ
a
i

iI
a
i
.
Proposition 2.6 Let I =
kK
J
k
, where K is a non-empty set of indices and
the J
k
s are non-empty and pairwise disjoint. Then for every a : I [0, +]
we have
iI
a
i
=
kK
_
iJ
k
a
i
_
.
Proof: Take an arbitrary nite F I and consider the nite sets F
k
= F J
k
.
Observe that the set L = k K[ F
k
,= is a nite subset of K. Then,
using trivial properties of sums over nite sets of indices, we nd
iF
a
i
=
kL
_
iF
k
a
i
_
. The denitions imply that
iF
a
i

kL
_
iJ
k
a
i
_

kK
_
iJ
k
a
i
_
. Taking supremum over the nite subsets F of I we nd
iI
a
i

kK
_
iJ
k
a
i
_
.
Now take an arbitrary nite L K and arbitrary nite F
k
J
k
for each
k L. Then
kL
_
iF
k
a
i
_
is, clearly, a sum (without repetitions) over
the nite subset
kL
F
k
of I. Hence
kL
_
iF
k
a
i
_

iI
a
i
. Taking
supremum over the nite subsets F
k
of J
k
for each k L, one at a time, we
get that
kL
_
iJ
k
a
i
_
iI
a
i
. Finally, taking supremum over the nite
subsets L of K, we nd
kK
_
iJ
k
a
i
_
iI
a
i
and conclude the proof.
After this short investigation of the general summation notion we dene a
class of measures.
Proposition 2.7 Let X be non-empty and consider a : X [0, +]. We
dene : T(X) [0, +] by
(E) =
xE
a
x
, E X.
2.3. COMPLETE MEASURES. 25
Then is a measure on (X, T(X)).
Proof: It is obvious that () =
x
a
x
= 0.
If E
1
, E
2
, . . . are pairwise disjoint and E =
+
n=1
E
n
, we apply Propositions
2.3 and 2.6 to nd (E) =
xE
a
x
=
nN
_
xEn
a
x
_
=
nN
(E
n
) =
+
n=1
(E
n
).
Denition 2.5 The measure dened in the statement of the previous proposi-
tion is called the point-mass distribution on X induced by the function
a. The value a
x
is called the point-mass at x.
Examples.
1. Consider the function which puts point-mass a
x
= 1 at every x X. It is
then obvious that the induced point-mass distribution is
(E) =
_
card(E), if E is a nite X,
+, if E is an innite X.
This is called the counting measure on X.
2. Take a particular x
0
X and the function which puts point-mass a
x0
= 1 at
x
0
and point-mass a
x
= 0 at all other points of X. Then the induced point-mass
distribution is
x0
(E) =
_
1, if x
0
E X,
0, if x
0
/ E X.
This is called the Dirac measure at x
0
or the Dirac mass at x
0
.
Of course, it is very easy to show directly, without using Proposition 2.7,
that these two examples, and
x0
, constitute measures.
2.3 Complete measures.
Theorem 2.2(i) says that a subset of a -null set is also -null, provided that
the subset is contained in the -algebra on which the measure is dened.
Denition 2.6 Let (X, , ) be a measure space. Suppose that for every E
with (E) = 0 and every F E it is implied that F (and hence (F) = 0,
also). Then is called complete and (X, , ) is a complete measure space.
Thus, a measure is complete if the -algebra on which it is dened contains
all subsets of -null sets.
If (X,
1
,
1
) and (X,
2
,
2
) are two measure spaces on the same set X,
we say that (X,
2
,
2
) is an extension of (X,
1
,
1
) if
1

2
and
1
(E) =
2
(E) for all E
1
.
Theorem 2.3 Let (X, , ) be a measure space. Then there is a unique small-
est complete extension (X, , ) of (X, , ). Namely, there is a unique measure
space (X, , ) so that
(i) (X, , ) is an extension of (X, , ),
(ii) (X, , ) is complete,
(iii) if (X, , ) is another complete extension of (X, , ), then it is an exten-
sion also of (X, , ).
Proof: We shall rst construct (X, , ). We dene
= A F [ A and F E for some E with (E) = 0.
(a) We prove that is a -algebra. We write = , where the rst belongs
to and the second is a subset of with () = 0. Therefore .
Let B . Then B = A F for A and F of some E with
(E) = 0. Write B
c
= A
1
F
1
, where A
1
= (A E)
c
and F
1
= E (A F).
Then A
1
and F
1
E. Hence B
c
.
Let B
1
, B
2
, . . . . Then for every n, B
n
= A
n
F
n
for A
n
and F
n

of some E
n
with (E
n
) = 0. Now
+
n=1
B
n
= (
+
n=1
A
n
) (
+
n=1
F
n
), where
+
n=1
A
n
and
+
n=1
F
n

+
n=1
E
n
with (
+
n=1
E
n
) = 0. Therefore
+
n=1
B
n
.
(b) We now construct . For every B we write B = A F for A and
F of some E with (E) = 0 and dene
(B) = (A).
To prove that (B) is well dened we consider that we may also have B =
A
for A
and F
of some E
with (E
) = 0 and we must prove

that (A) = (A
). Since A B A
, we have (A) (A
) + (E
) =
(A
) and, symmetrically, (A
) (A).
(c) To prove that is a measure on (X, ) let = as in (a) and get () =
() = 0. Let also B
1
, B
2
, . . . be pairwise disjoint. Then B
n
= A
n
F
n
for
A
n
and F
n
E
n
with (E
n
) = 0. Observe that the A
n
s are pairwise
disjoint. Then
+
n=1
B
n
= (
+
n=1
A
n
) (
+
n=1
F
n
) and
+
n=1
F
n

+
n=1
E
n

with (
+
n=1
E
n
) = 0. Therefore (
+
n=1
B
n
) = (
+
n=1
A
n
) =
+
n=1
(A
n
) =
+
n=1
(B
n
).
(d) We now prove that is complete. Let B with (B) = 0 and let
B
B. Write B = A F for A and F E with (E) = 0 and

observe that (A) = (B) = 0. Then write B
= A
, where A
= and
F
= B
where E
= A E with (E
) (A) + (E) = 0. Hence

B
.
(e) To prove that (X, , ) is an extension of (X, , ) we take any A and
write A = A , where with () = 0. This implies that A and
(A) = (A).
(f) Now suppose that (X, , ) is another complete extension of (X, , ). Take
any B and thus B = A F for A and F E with (E) = 0.
But then A, E and (E) = (E) = 0. Since is complete, we get that also
F and hence B = A F .
Moreover, (A) (B) (A) + (F) = (A), which implies (B) =
(A) = (A) = (B).
2.4. RESTRICTION OF A MEASURE. 27
(g) It only remains to prove the uniqueness of a smallest complete extension of
(X, , ). This is obvious, since two smallest complete extensions of (X, , )
must both be extensions of each other and, hence, identical.
Denition 2.7 If (X, , ) is a measure space, then its smallest complete ex-
tension is called the completion of (X, , ).
2.4 Restriction of a measure.
Proposition 2.8 Let (X, , ) be a measure space and let Y . If we dene
Y
: [0, +] by
Y
(A) = (A Y ), A ,
then
Y
is a measure on (X, ) with the properties that
Y
(A) = (A) for every
A , A Y , and that
Y
(A) = 0 for every A , A Y = .
Proof: We have
Y
() = ( Y ) = () = 0.
If A
1
, A
2
, . . . are pairwise disjoint,
Y
(
+
j=1
A
j
) =
_
(
+
j=1
A
j
) Y
_
=
+
j=1
(A
j
Y )
_
=
+
j=1
(A
j
Y ) =
+
j=1

Y
(A
j
).
Therefore,
Y
is a measure on (X, ) and its two properties are trivial to
prove.
Denition 2.8 Let (X, , ) be a measure space and let Y . The measure
Y
on (X, ) of Proposition 2.8 is called the restriction of on Y .
There is a second kind of restriction of a measure. To dene it we recall
that the restriction Y of the -algebra of subsets of X on the non-empty
Y X is dened as Y = A Y [ A .
Lemma 2.1 Let be a -algebra of subsets of X and let Y be non-empty.
Then
Y = A [ A Y .
Proof: We set
= A [ A Y . If B Y , then B = A Y for
some A . Since Y , we nd that B and B = A Y Y and,
hence, B
. Conversely, if B
, then B and B Y and, if we set

A = B , we have B = A Y Y .
Proposition 2.9 Let (X, , ) be a measure space and let Y be non-empty.
We consider Y = A [ A Y and dene Y : Y [0, +] by
(Y )(A) = (A), A Y.
Then Y is a measure on (Y, Y ).
Proof: Obvious.
Denition 2.9 Let (X, , ) be a measure space and let Y be non-empty.
The measure Y on (Y, Y ) of Proposition 2.9 is called the restriction of
on Y .
Informally speaking, we may describe the relation between the two restric-
tions of as follows. The restriction
Y
assigns value 0 to all sets in which
are included in the complement of Y while the restriction Y simply ignores
all those sets. Both restrictions
Y
and Y assign the same values (the same
to the values that assigns) to all sets in which are included in Y .
2.5 Uniqueness of measures.
The next result is very useful when we want to prove that two measures are
equal on a -algebra . It says that it is enough to prove that they are equal on
an algebra which generates , provided that an extra assumption of -niteness
of the two measures on the algebra is satised.
Theorem 2.4 Let / be an algebra of subsets of the non-empty set X and let
, be two measures on (X, (/)). Suppose there exist A
1
, A
2
, . . . / with
A
n
X and (A
k
), (A
k
) < + for all k.
If , are equal on /, then they are equal also on (/).
Proof: (a) Suppose that (X), (X) < +.
We dene the collection / = E (/) [ (E) = (E). It is easy to
see that / is a monotone class. Indeed, let E
1
, E
2
, . . . / with E
n
E.
By continuity of measures from below, we get (E) = lim
n+
(E
n
) =
lim
n+
(E
n
) = (E) and thus E /. We do exactly the same when
E
n
E, using the continuity of measures from above and the extra assumption
(X), (X) < +.
Since / is a monotone class including /, Proposition 1.12 implies that
/(/) /. Now Theorem 1.1 implies that (/) / and thus (E) = (E)
for all E (/).
(b) The general case.
For each k, consider the restrictions of , on A
k
. Namely,
A
k
(E) = (E A
k
),
A
k
(E) = (E A
k
)
for all E (/). All
A
k
and
A
k
are nite measures on (X, ), because
A
k
(X) = (A
k
) < + and
A
k
(X) = (A
k
) < +. We clearly have that
A
k
,
A
k
are equal on / and, by the result of (a), they are equal also on (/).
For every E (/) we can write, using the E A
k
E and the continuity
of measures from below, (E) = lim
n+
(E A
k
) = lim
n+
A
k
(E) =
lim
n+
A
k
(E) = lim
n+
(E A
k
) = (E).
Thus, , are equal on (/).
2.6. EXERCISES. 29
2.6 Exercises.
1. Let (X, , ) be a measure space and A
1
, A
2
, . . . . Prove (
+
n=1
A
n
) =
lim
n+
(
n
k=1
A
k
).
2. Let (X, , ) be a measure space and Y . Prove that
Y
is the only
measure on (X, ) with the properties:
(i)
Y
(E) = (E) for all E with E Y ,
(ii)
Y
(E) = 0 for all E with E Y
c
.
3. Positive linear combinations of measures.
Let ,
1
2
be measures on the measurable space (X, ) and [0, +).
(i) Prove that : [0, +], which is dened by
()(E) = (E)
for all E , is a measure on (X, ). The measure is called the
product of by .
(ii) Prove that
1
+
2
: [0, +], which is dened by
(
1
+
2
)(E) =
1
(E) +
2
(E)
for all E , is a measure on (X, ). The measure
1
+
2
is called the
sum of
1
and
2
.
Thus we dene (positive) linear combinations
1
1
+ +
n
n
.
4. Let X be non-empty and consider a nite A X. If a : X [0, +]
satises a
x
= 0 for all x / A, prove that the point-mass distribution
on X induced by a can be written as a positive linear combination (see
Exercise 2.6.3) of Dirac measures:
=
1
x1
+ +
k
x
k
.
5. Let X be innite and dene for all E X
(E) =
_
0, if E is nite,
+, if E is innite.
Prove that is a nitely additive measure on (X, T(X)) which is not a
measure.
6. Let (X, , ) be a measure space and E be of -nite measure. If
D
i
iI
is a collection of pairwise disjoint sets in , prove that the set
i I [ (E D
i
) > 0 is countable.
7. Let X be uncountable and dene for all E X
(E) =
_
0, if E is countable,
+, if E is uncountable.
Prove that is a measure on (X, T(X)) which is not seminite.
8. Let (X, , ) be a complete measure space. If A , B X and
(AB) = 0, prove that B and (B) = (A).
9. Let be a nitely additive measure on the measurable space (X, ).
(i) Prove that is a measure if and only if it is continuous from below.
(ii) If (X) < +, prove that is a measure if and only if it is continuous
from above.
10. Let (X, , ) be a measure space and A
1
, A
2
, . . . . Prove that
(i) (liminf
n+
A
n
) liminf
n+
(A
n
),
(ii) limsup
n+
(A
n
) (limsup
n+
A
n
), if (
+
n=1
A
n
) < +,
(iii) (limsup
n+
A
n
) = 0, if
+
n=1
(A
n
) < +.
11. Increasing limits of measures are measures.
Let
n
be a sequence of measures on (X, ) which is increasing. Namely,
n
(E)
n+1
(E) for all E and all n. We dene
(E) = lim
n+
n
(E)
for all E . Prove that is a measure on (X, ).
12. Let I be a set of indices and a, b : I [0, +].
(i) Prove that
iI
a
i
= 0 if and only if a
i
= 0 for all i I.
(ii) If J = i I [ a
i
> 0, prove that
iI
a
i
=
iJ
a
i
.
(iii) Prove that, for all [0, +],
iI
a
i
=
iI
a
i
.
(iv) Prove that
iI
(a
i
+b
i
) =
iI
a
i
+
iI
b
i
.
13. Tonellis Theorem for sums.
Let I, J be two sets of indices and a : I J [0, +]. Using Proposition
2.6, prove that
iI
_
jJ
a
i,j
_
=
(i,j)IJ
a
i,j
=
jJ
_
iI
a
i,j
_
.
Recognize as a special case the
iI
(a
i
+b
i
) =
iI
a
i
+
iI
b
i
for every a, b : I [0, +].
2.6. EXERCISES. 31
14. Let X be non-empty and consider the point-mass distribution dened
by the function a : X [0, +]. Prove that
(i) is seminite if and only if a
x
< + for every x X,
(ii) is -nite if and only if a
x
< + for every x X and the set
x X[ a
x
> 0 is countable.
15. Let (X, , ) be a measure space.
(i) If A, B and (AB) = 0, prove that (A) = (B).
(ii) We dene A B if A, B and (AB) = 0. Prove that is an
equivalence relation on .
We assume that (X) < + and dene
d(A, B) = (AB)
for all A, B .
(iii) Prove that d is a pseudometric on . This means: 0 d(A, B) < +,
d(A, B) = d(B, A) and d(A, C) d(A, B) +d(B, C) for all A, B, C .
(iv) On the set / of all equivalence classes we dene
d([A], [B]) = d(A, B) = (AB)
for all [A], [B] / . Prove that d([A], [B]) is well-dened and that d is
a metric on / .
16. Let be a seminite measure on the measurable space (X, ). Prove that
for every E with (E) = + and every M > 0 there is an F so
that F E and M < (F) < +.
17. The saturation of a measure space.
Let (X, , ) be a measure space. We call the set E X locally mea-
surable if E A for all A with (A) < +. We dene
= E X[ E is locally measurable.
(i) Prove that

and that

is a -algebra. If =

, then (X, , )
is called saturated.
(ii) If is -nite, prove that (X, , ) is saturated.
We dene
(E) =
_
(E), if E ,
+, if E

.
(iii) Prove that is a measure on (X,
), and hence (X,
, ) is an exten-
sion of (X, , ).
(iv) If (X, , ) is complete, prove that (X,
, ) is also complete.
(v) Prove that (X,
, ) is a saturated measure space.

(X,
, ) is called the saturation of (X, , ).

18. The direct sum of measure spaces.
Let (X
i
,
i
,
i
)
iI
be a collection of measure spaces, where the X
i
s are
pairwise disjoint. We dene
X =
iI
X
i
, = E X[ E X
i

i
for all i I
and
(E) =
iI
i
(E X
i
)
for all E .
(i) Prove that (X, , ) is a measure space. It is called the direct sum
of (X
i
,
i
,
i
)
iI
and it is denoted
iI
(X
i
,
i
,
i
).
(ii) Prove that is -nite if and only if the set J = i I [
i
,= o is
countable and
i
is -nite for all i J.
19. Characterisation of point-mass distributions.
Let X ,= . Prove that every measure on (X, T(X)) is a point-mass
distribution.
20. The push-forward of a measure.
Let (X, , ) be a measure space and f : X Y . We consider the -
algebra
= B Y [ f
1
(B) , the push-forward of by f on Y
(see Exercise 1.6.7). We dene
(B) = (f
1
(B)) , B
.
Prove that
is a measure on (Y,
). It is called the push-forward of

by f on Y .
21. The pull-back of a measure.
Let (Y,
) be a measure space and f : X Y be one-to-one and onto

Y . We consider the -algebra = f
1
(B) [ B
, the pull-back of
by f on X (see Exercise 1.6.8). We dene

(A) =
(f(A)) , A .
Prove that is a measure on (X, ). It is called the pull-back of
by
f on X.
Chapter 3
Outer measures
3.1 Outer measures.
Denition 3.1 Let X be a non-empty set. A function
: T(X) [0, +] is
called outer measure on X if
(i)
() = 0,
(ii)
(A)
(B) if A B X,
(iii)
(
+
n=1
A
n
)
+
n=1
(A
n
) for all sequences A
n
of subsets of X.
Thus, an outer measure on X is dened for all subsets of X, it is monotone and
-subadditive.
We shall see now how a measure is constructed from an outer measure.
Denition 3.2 Let
be an outer measure on the non-empty set X. We say

that the set A X is
-measurable if
(E A) +
(E A
c
) =
(E)
for all E X.
We denote
the collection of all
-measurable subsets of X.
Thus, a set is
-measurable if and only if it decomposes every subset of X into

two disjoint pieces whose outer measures add to give the outer measure of the
subset.
Observe that E = (EA) (E A
c
) and by the -subadditivity
of
we have
(E)
(E A) +
(E A
c
) + 0 + 0 + . Therefore, in
order to check the validity of the equality in the denition, it is enough to check
the inequality
(E A) +
(E A
c
)
(E). Furthermore, it is enough to

check this last inequality in the case
(E) < +.
Theorem 3.1 (Caratheodory) If
is an outer measure on X, then the collec-

tion
of all
-measurable subsets of X is a -algebra. If we denote the

restriction of
on
, then (X,
, ) is a complete measure space.

33
34 CHAPTER 3. OUTER MEASURES
Proof: (a)
(E ) +
(E
c
) =
() +
(E) =
(E) and
.
If A
, then
(EA
c
)+
(E(A
c
)
c
) =
(EA)+
(EA
c
) =
(E)
for all E X. This says that A
c

and
is closed under complements.

(b) Let now A, B
and take an arbitrary E X. To check A B
write
(E(AB)) +
(E(AB)
c
) =
(E(AB)) +
(E(A
c
B
c
))
and use the subadditivity of
for the rst term to get
(E (A B
c
)) +
(E (BA
c
)) +
(E (AB)) +
(E (A
c
B
c
)). Now combine the rst
and third term and also the second and fourth term with the
-measurability
of B to get =
(EA)+
(EA
c
), which is =
(E) by the
-measurability
of A.
This proves that A B
and by induction we get that
is closed
under nite unions. Since it is closed under complements, it is an algebra of
subsets of X and, hence, it is also closed under nite intersections and under
set-theoretic dierences.
(c) Let A, B
with AB = and get for all E X that
(E(AB)) =
([E (A B)] A) +
([E (A B)] A
c
) =
(E A) +
(E B). By
an obvious induction we nd that if A
1
, . . . , A
N

are pairwise disjoint and

E X is arbitrary then
(E(A
1
A
N
)) =
(EA
1
)+ +
(EA
N
).
If now A
1
, A
2
, . . .
are pairwise disjoint and E X is arbitrary, then, for

all N,
(EA
1
)+ +
(EA
N
) =
(E(A
1
A
N
))
(E(
+
n=1
A
n
))
by the monotonicity of
. Hence
+
n=1
(EA
n
)
(E(
+
n=1
A
n
)). Since
the opposite inequality is immediate after the -subadditivity of
, we conclude
the basic equality
+
n=1
(E A
n
) =
(E (
+
n=1
A
n
))
for all pairwise disjoint A
1
, A
2
, . . .
and all E X.
(d) If A
1
, A
2
, . . .
are pairwise disjoint and E X is arbitrary, then, by

the result of (b),
N
n=1
A
n

for all N. Hence
(E) =
(E (
N
n=1
A
n
)) +
(E (
N
n=1
A
n
)
c
)
N
n=1
(E A
n
) +
(E (
+
n=1
A
n
)
c
), where we used
the basic equality for the rst term and the monotonicity of
for the second

term. Since N is arbitrary,
(E)
+
n=1
(E A
n
) +
(E (
+
n=1
A
n
)
c
) =
(E (
+
n=1
A
n
)) +
(E (
+
n=1
A
n
)
c
) by the basic equality.
This means that
+
n=1
A
n

.
If A
1
, A
2
, . . .
are not necessarily pairwise disjoint, we write B

1
= A
1
and B
n
= A
n
(A
1
A
n1
) for all n 2. Then, by the result of (b),
all B
n
s belong to
and they are pairwise disjoint. Therefore, by the last

paragraph,
+
n=1
A
n
=
+
n=1
B
n

. We conclude that
is a -algebra.
(e) We now dene :
[0, +] as the restriction of
, namely (A) =
(A) for all A
.
Using E = X in the basic equality, we get that for all pairwise disjoint
A
1
, A
2
, . . .
,
+
n=1
(A
n
) =
+
n=1
(A
n
) =
(
+
n=1
A
n
) = (
+
n=1
A
n
).
3.2. CONSTRUCTION OF OUTER MEASURES. 35
Since () =
() = 0 we see that (X,
, ) is a measure space.
(f) Let A
with (A) = 0 and B A. Then
(B)
(A) = (A) = 0
and for all E X we get
(E B) +
(E B
c
)
(B) +
(E) =
(E).
Hence B
and is complete.
As a by-product of the proof of Caratheodorys theorem we get the useful
Proposition 3.1 Let
be an outer measure on the non-empty X.

(i) If B X has
(B) = 0, then B is
-measurable.
(ii) For all pairwise disjoint
-measurable A
1
, A
2
, . . . and all E X
+
n=1
(E A
n
) =
(E (
+
n=1
A
n
)).
Proof: The proof of (i) is in part (f) of the proof of the theorem of Caratheodory
and the proof of (ii) is the basic equality in part (c) of the same proof.
3.2 Construction of outer measures.
Denition 3.3 Let X be a non-empty set. A collection ( of subsets of X is
called a -covering collection for X if ( and there exist X
1
, X
2
, . . . (
so that X =
+
n=1
X
n
.
Theorem 3.2 Suppose we have a -covering collection ( for X and an arbi-
trary function : ( [0, +] with () = 0. If we dene
(E) = inf
_
+
j=1
(C
j
) [ C
1
, C
2
, . . . ( so that E
+
j=1
C
n
_
for all E X, then
is an outer measure on X.
Before the proof, observe that in the denition of
(E) the set over which the

inmum is taken is not empty, since there is at least one countable covering of
E, in fact even of X, with elements of (. This is clear from the denition of a
-covering collection.
Proof: For the covering implies
() () +() + = 0.
Let A B X and take an arbitrary covering B
+
j=1
C
n
with C
1
, . . . (.
Then we also have A
+
j=1
C
n
and by the denition of
(A) we get
(A)
+
j=1
(C
j
). Therefore, by the denition of
(B) we nd
(A)
(B).
Finally, lets prove
(
+
n=1
A
n
)

+
n=1
(A
n
) for all A
1
, A
2
, . . . X.
If the right side is = +, the inequality is clear. Therefore we assume that
the right side is < + and, hence, that
(A
n
) < + for all n. By the
denition of each
(A
n
), for every > 0 there exist C
n,1
, C
n,2
, . . . ( so that
A
n

+
j=1
C
n,j
and
+
j=1
(C
n,j
) <
(A
n
) +

2
n
.
Then
+
n=1
A
n

(n,j)NN
C
n,j
and, using an arbitrary enumeration of
N N and Proposition 2.3, we get by the denition of
(
+
n=1
A
n
) that
(
+
n=1
A
n
)

(n,j)NN
(C
n,j
). Proposition 2.6 implies
(
+
n=1
A
n
)
+
n=1
(
+
j=1
(C
n,j
)) <
+
n=1
(
(A
n
) +

2
n
) =
+
n=1
(A
n
) + . Since is
arbitrary, we conclude that
(
+
n=1
A
n
)
+
n=1
(A
n
).
3.3. EXERCISES. 37
3.3 Exercises.
1. Let
be an outer measure on X and Y X. Dene
Y
(E) =
(EY )
for all E X and prove that
Y
is an outer measure on X and that Y is
Y
-measurable.
2. Let
1
,
2
be outer measures on X and [0, +). Prove that
1
+
2
and max(
1
,
2
) are outer measures on X, where these are
dened by the formulas
(
)(E) =
(E), (
1
+
2
)(E) =
1
(E) +
2
(E)
and
max(
1
,
2
)(E) = max(
1
(E),
2
(E))
for all E X.
3. Let X be a non-empty set and consider
() = 0 and
(E) = 1 if
, = E X. Prove that
is an outer measure on X and nd all the
-measurable subsets of X.
4. For every E N dene (E) = limsup
n+
1
n
card(E 1, 2, . . . , n).
Is an outer measure on N?
5. Let
n
be a sequence of outer measures on X. Let
(E) = sup
n
n
(E)
for all E X and prove that
is an outer measure on X.
6. Let
be an outer measure on X. If A
1
, A
2
, . . .
with A
n
A
n+1
for
all n and E X, prove that lim
n+
(A
n
E) =
+
n=1
(A
n
E)
_
.
7. Extension of a measure, I.
Let (X,
0
,
0
) be a measure space. For every E X we dene
(E) = inf
_
+
j=1
0
(A
j
) [ A
1
, A
2
, . . .
0
, E
+
j=1
A
j
_
,
(i) Prove that
is an outer measure on X. We say that
is induced
by the measure
0
.
(ii) Prove that
(E) = min
_
0
(A) [ A
0
, E A.
(iii) If (X,
, ) is the complete measure space which results from
by
the theorem of Caratheodory (i.e. is the restriction of
on
), prove
that (X,
, ) is an extension of (X,
0
,
0
).
(iv) Assume that E X and A
1
, A
2
, . . .
0
with E
+
j=1
A
j
and
(A
j
) < + for all j. Prove that E
if and only if there is some

A
0
so that E A and
(A E) = 0.
(v) If is -nite, prove that (X,
, ) is the completion of (X,

0
,
0
).
(vi) Let X be an uncountable set,
0
= A X[ A is countable or A
c
is
countable and
0
(A) = (A) for every A
0
. Prove that (X,
0
,
0
) is
a complete measure space and that
= T(X). Thus, the result of (v)

does not hold in general.
(vii) Prove that (X,
, ) is always the saturation (see exercise 2.5.15)

of the completion of (X,
0
,
0
).
8. Measures on algebras.
Let / be an algebra of subsets of the non-empty X. We say that : /
[0, +] is a measure on (X, /) if
(i) () = 0 and
(ii) (
+
j=1
A
j
) =
+
j=1
(A
j
) for all pairwise disjoint A
1
, A
2
, . . . / with
+
j=1
A
j
/.
Prove that if is a measure on (X, /), where / is an algebra of subsets of
X, then is nitely additive, monotone, -subadditive, continuous from
below and continuous from above (provided that, every time a countable
union or countable intersection of elements of / appears, we assume that
this is also an element of /).
9. Extension of a measure, II.
Let /
0
be an algebra of subsets of the non-empty X and
0
be a measure
on (X, /
0
) (see exercise 3.3.8). For every E X we dene
(E) = inf
_
+
j=1
0
(A
j
) [ A
1
, A
2
, . . . /
0
, E
+
j=1
A
j
_
,
(i) Prove that
is an outer measure on X. We say that
is induced
by the measure
0
.
(ii) Prove that
(A) =
0
(A) for every A /
0
.
(iii) Prove that every A /
0
is
-measurable and hence (/

0
)
.
Thus, if (after Caratheodorys theorem) is the restriction of
on
,
the measure space (X,
, ) is a complete measure space which extends

(X, /
0
,
0
).
If we consider the restriction (X, (/
0
), ), then this is also a measure
space (perhaps not complete) which extends (X, /
0
,
0
).
(iv) If (X, (/
0
), ) is another measure space which is an extension of
(X, /
0
,
0
), prove that (E) (E) for all E (/
0
) with equality in
case (E) < +.
(v) If the original (X, /
0
,
0
) is -nite, prove that is the unique measure
on (X, (/
0
)) which is an extension of
0
on (X, /
0
).
10. Regular outer measures.
Let
be an outer measure on X. We say that
is a regular outer
measure if for every E X there is A
so that E A and
(E) = (A) (where is the usual restriction of
on
).
Prove that
is a regular outer measure if and only if
is induced by
some measure on some algebra of subsets of X (see exercise 3.3.9).
3.3. EXERCISES. 39
11. Measurable covers.
Let
be an outer measure on X and be the induced measure (the

restriction of
) on
. If E, G X we say that G is a
-measurable
cover of E if E G, G
and for all A
for which A G E
we have (A) = 0.
(i) If G
1
, G
2
are
-measurable covers of E, prove that (G

1
G
2
) = 0
and hence (G
1
) = (G
2
).
(ii) Suppose E G, G
and
(E) = (G). If
(E) < +, prove

that G is a
-measurable cover of E.
12. We say E R has an innite condensation point if E has uncount-
ably many points outside every bounded interval. Dene
(E) = 0 if E
is countable,
(E) = 1 if E is uncountable and does not have an innite

condensation point and
(E) = + if E has an innite condensation

point. Prove that
is an outer measure on R and that A R is
-
measurable if and only if either A or A
c
is countable. Does every E R
have a
-measurable cover? Is
a regular outer measure? (See exercises

3.3.10 and 3.3.11).
13. Consider the collection ( of subsets of N which contains and all the
two-point subsets of N. Dene () = 0 and (C) = 2 for all other C (.
Calculate
(E) for all E N, where
is the outer measure dened as

in Theorem 3.2, and nd all the
-measurable subsets of N.
Chapter 4
Lebesgue-measure in R
n
4.1 Volume of intervals.
We consider the function vol
n
(S) dened for general intervals S, which is just
the product of the lengths of the edges of S: the so-called (n-dimensional)
volume of S. In this section we shall investigate some properties of the volume
of intervals.
Lemma 4.1 Let P = (a
1
, b
1
] (a
n
, b
n
] and, for each k = 1, . . . , n, let
a
k
= c
0
k
< c
1
k
< < c
m
k
k
= b
k
. If we set P
i1,...,in
= (c
i11
1
, c
i1
1
] (c
in1
n
, c
in
n
]
for 1 i
1
m
1
, . . . , 1 i
n
m
n
, then
vol
n
(P) =
1i1m1,...,1inmn
vol
n
(P
i1,...,in
).
Proof: For the second equality in the following calculation we use the distribu-
tive property of multiplication of sums:
1i1m1,...,1inmn
vol
n
(P
i1,...,in
)
=
1i1m1,...,1inmn
(c
i1
1
c
i11
1
) (c
in
n
c
in1
n
)
=
m1
i1=1
(c
i1
1
c
i11
1
)
mn
in=1
(c
in
n
c
in1
n
)
= (b
1
a
1
) (b
n
a
n
) = vol
n
(P).
Referring to the situation described by Lemma 4.1 we shall use the expres-
sion: the intervals P
i1,...,in
result from P by subdivision of its edges.
Lemma 4.2 Let P, P
1
, . . . , P
l
be open-closed intervals and P
1
, . . . , P
l
be pair-
wise disjoint. If P = P
1
P
l
, then vol
n
(P) = vol
n
(P
1
) + +vol
n
(P
l
).
41
42 CHAPTER 4. LEBESGUE-MEASURE IN R
N
Proof: Let P = (a
1
, b
1
] (a
n
, b
n
] and P
j
= (a
j
1
, b
j
1
] (a
j
n
, b
j
n
] for every
j = 1, . . . , l.
For every k = 1, . . . , n we set
c
0
k
, . . . , c
m
k
k
= a
1
k
, . . . , a
l
k
, b
1
k
, . . . , b
l
k
,
so that a
k
= c
0
k
< c
1
k
< < c
m
k
k
= b
k
. This simply means that we rename
the numbers a
1
k
, . . . , a
l
k
, b
1
k
, . . . , b
l
k
in increasing order and so that there are no
repetitions. Of course, the smallest of these numbers is a
k
and the largest is b
k
,
otherwise the P
1
, . . . , P
l
would not cover P.
It is almost obvious that
(a) every (a
j
k
, b
j
k
] is the union of some successive among (c
0
k
, c
1
k
], . . . , (c
m
k
1
k
, c
m
k
k
],
(b) none of (c
i
k
1
k
, c
i
k
k
] intersects two disjoint among (a
j
k
, b
j
k
]s.
We now set
P
i1,...,in
= (c
i11
1
, c
i1
1
] (c
in1
n
, c
in
n
]
for 1 i
1
m
1
, . . . , 1 i
n
m
n
.
It is clear that the P
i1,...,in
s result from P by subdivision of its edges. It is
also almost clear that
(c) the intervals among the P
i1,...,in
which belong to a P
j
result from it by
subdivision of its edges (this is due to (a)).
(d) every P
i1,...,in
is included in exactly one from P
1
, . . . , P
l
(this is due to (b)).
We now calculate, using Lemma 4.1 for the rst and third equality and
grouping together the intervals P
i1,...,in
which are included in the same P
j
for
the second equality:
vol
n
(P) =
1i1m1,...,1inmn
vol
n
(P
i1,...,in
)
=
l
j=1
_

Pi
1
,...,in
Pj
vol
n
(P
i1,...,in
)
_
=
l
j=1
vol
n
(P
j
).
Lemma 4.3 Let P, P
1
, . . . , P
l
be open-closed intervals and P
1
, . . . , P
l
be pair-
wise disjoint. If P
1
P
l
P, then vol
n
(P
1
) + +vol
n
(P
l
) vol
n
(P).
Proof: We know from Proposition 1.10 that P (P
1
P
l
) = P
1
P
k
for some pairwise disjoint open-closed intervals P
1
, . . . , P
k
. Then P = P
1

P
l
P
1
P
k
and Lemma 4.2 now implies that vol
n
(P) = vol
n
(P
1
) + +
vol
n
(P
l
) +vol
n
(P
1
) + +vol
n
(P
k
) vol
n
(P
1
) + +vol
n
(P
l
).
Lemma 4.4 Let P, P
1
, . . . , P
l
be open-closed intervals. If P P
1
P
l
,
then vol
n
(P) vol
n
(P
1
) + +vol
n
(P
l
).
Proof: We rst write P = P
1
P
l
where P
j
= P
j
P are open-closed intervals
included in P. We then write P = P
1
(P
2
P
1
)
_
P
l
(P
1
P
l1
)
_
.
4.2. LEBESGUE-MEASURE IN R
N
. 43
Each of these l pairwise disjoint sets can, by Proposition 1.10, be written as a
nite union of pairwise disjoint open-closed intervals: P
1
= P
1
and
P
j
(P
1
P
j1
) = P
j
1
P
j
mj
for 2 j l.
Lemma 4.2 for the equality and Lemma 4.3 for the two inequalities imply
vol
n
(P) = vol
n
(P
1
) +
l
j=2
_
mj
m=1
vol
n
(P
j
m
)
_
vol
n
(P
1
) +
l
j=2
vol
n
(P
j
)
l
j=1
vol
n
(P
j
).
Lemma 4.5 Let Q be a closed interval and R
1
, . . . , R
l
be open intervals so that
Q R
1
R
l
. Then vol
n
(Q) vol
n
(R
1
) + +vol
n
(R
l
).
Proof: Let P and P
j
be the open-closed intervals with the same edges as Q and,
respectively, R
j
. Then P Q R
1
R
l
P
1
P
l
and by Lemma
4.4, vol
n
(Q) = vol
n
(P) vol
n
(P
1
) + +vol
n
(P
l
) = vol
n
(R
1
) + +vol
n
(R
l
).
4.2 Lebesgue-measure in R
n
.
Consider the collection ( of all open intervals in R
n
. Since we can write R
n
=
+
k=1
(k, k) (k, k), the collection is -covering for R
n
.
Next we consider
(R) = vol
n
(R) = (b
1
a
1
) (b
n
a
n
)
for every R = (a
1
, b
1
) (a
n
, b
n
) (.
If we dene
m
n
(E) = inf
_
+
j=1
vol
n
(R
j
) [ R
1
, R
2
, . . . ( so that E
+
j=1
R
n
_
for all E R
n
, then Theorem 3.2 implies that m
n
is an outer measure on R
n
.
Now Theorem 3.1 implies that the collection L
n
=
m
n
of m
n
-measurable sets
is a -algebra of subsets of R
n
and, if m
n
is dened as the restriction of m
n
on
L
n
, then m
n
is a complete measure on (X, L
n
).
Denition 4.1 (i) L
n
is called the -algebra of Lebesgue-measurable sets
in R
n
,
(ii) m
n
is called the (n-dimensional) Lebesgue-outer measure in R
n
and
(iii) m
n
is called the (n-dimensional) Lebesgue-measure in R
n
.
Our aim now is to study properties of Lebesgue-measurable sets in R
n
and
especially their relation with the Borel sets or even more special sets in R
n
, like
open sets or closed sets or unions of intervals.
N
Theorem 4.1 Every interval S in R
n
is Lebesgue-measurable and m
n
(S) =
vol
n
(S).
Proof: (a) Let Q = [a
1
, b
1
] [a
n
, b
n
].
Since Q (a
1
, b
1
+ ) (a
n
, b
n
+ ), we get by the denition
of m
n
that m
n
(Q) vol
n
((a
1
, b
1
+ ) (a
n
, b
n
+)) = (b
1
a
1
+
2) (b
n
a
n
+ 2). Since > 0 is arbitrary, we nd m
n
(Q) vol
n
(Q).
Now take any covering, Q R
1
R
2
of Q by open intervals. Since Q
is compact, there is l so that Q R
1
R
l
, and Lemma 4.5 implies that
vol
n
(Q) vol
n
(R
1
) + + vol
n
(R
l
)

+
k=1
vol
n
(R
k
). Therefore vol
n
(Q)
m
n
(Q), and hence m
n
(Q) = vol
n
(Q).
Now take any general interval S and let a
1
, b
1
, . . . , a
n
, b
n
be the end-points of
its edges. Then Q
S Q
, where Q
= [a
1
+, b
1
] [a
n
+, b
n
] and
Q
= [a
1
, b
1
+] [a
n
, b
n
+]. Hence m
n
(Q
) m
n
(S) m
n
(Q
),
namely (b
1
a
1
2) (b
n
a
n
2) m
n
(S) (b
1
a
1
+2) (b
n
a
n
+2).
Since > 0 is arbitrary we nd m
n
(S) = vol
n
(S).
(b) Consider an open-closed interval P and an open interval R. Take the open-
closed interval P
R
with the same edges as R. Then m
n
(RP) m
n
(P
R
P) =
vol
n
(P
R
P) and m
n
(R P
c
) m
n
(P
R
P
c
). Now Proposition 1.10 implies
P
R
P
c
= P
R
P = P
1
P
k
for some pairwise disjoint open-closed intervals
P
1
, . . . , P
k
. Hence m
n
(R P
c
) m
n
(P
1
) + + m
n
(P
k
) = vol
n
(P
1
) + +
vol
n
(P
k
). Altogether, m
n
(R P) + m
n
(R P
c
) vol
n
(P
R
P) + vol
n
(P
1
) +
+vol
n
(P
k
) and, by Lemma 4.2, this is = vol
n
(P
R
) = vol
n
(R). We have just
proved that m
n
(R P) +m
n
(R P
c
) vol
n
(R).
(c) Consider any open-closed interval P and any E R
n
with m
n
(E) < +.
Take, for arbitrary > 0, a covering E
+
j=1
R
j
of E by open intervals so that
+
j=1
vol
n
(R
j
) < m
n
(E) +. Then m
n
(E P) +m
n
(E P
c
)
+
j=1
m
n
(R
j
P) +
+
j=1
m
n
(R
j
P
c
) =
+
j=1
[m
n
(R
j
P) + m
n
(R
j
P
c
)] which, by the
result of (b) is
+
j=1
vol
n
(R
j
) < m
n
(E) +. This implies that m
n
(E P) +
m
n
(E P
c
) m
n
(E) and P is Lebesgue-measurable.
If T is any interval at least one of whose edges is a single point, then m
n
(T) =
vol
n
(T) = 0 and, by Proposition 3.1, T is Lebesgue-measurable.
Now any interval S diers from the open-closed interval P, which has the
same sides as S, by nitely many Ts, and hence S is also Lebesgue-measurable.
Theorem 4.2 Lebesgue-measure is -nite but not nite.
Proof: We write R
n
=
+
k=1
Q
k
with Q
k
= [k, k] [k, k], where
m
n
(Q
k
) = vol
n
(Q
k
) < + for all k.
On the other hand, for all k, m
n
(R
n
) m
n
(Q
k
) = (2k)
n
and hence
m
n
(R
n
) = +.
Theorem 4.3 All Borel sets in R
n
are Lebesgue-measurable.
Proof: Theorem 4.1 says that, if c is the collection of all intervals in R
n
, then
c L
n
. But then B
R
n = (c) L
n
.
Therefore all open and all closed subsets of R
n
are Lebesgue-measurable.
4.2. LEBESGUE-MEASURE IN R
N
. 45
Theorem 4.4 Let E R
n
. Then
(i) E L
n
if and only if there is A, a countable intersection of open sets, so
that E A and m
n
(A E) = 0.
(ii) E L
n
if and only if there is B, a countable union of compact sets, so that
B E and m
n
(E B) = 0.
Proof: (i) One direction is easy. If there is A, a countable intersection of open
sets, so that E A and m
n
(A E) = 0, then, by Proposition 3.1, A E L
n
and thus E = A (A E) L
n
.
To prove the other direction consider, after Theorem 4.2, Y
1
, Y
2
, . . . L
n
so
that R
n
=
+
k=1
Y
k
and m
n
(Y
k
) < + for all k. Dene E
k
= E Y
k
so that
E =
+
k=1
E
k
and m
n
(E
k
) < + for all k.
For all k and arbitrary l N nd a covering E
k

+
j=1
R
k,l
j
by open
intervals so that
+
j=1
vol
n
(R
k,l
j
) < m
n
(E
k
) +
1
l2
k
and set U
k,l
=
+
j=1
R
k,l
j
.
Then E
k
U
k,l
and m
n
(U
k,l
) < m
n
(E
k
) +
1
l2
k
from which
m
n
(U
k,l
E
k
) <
1
l2
k
.
Now set U
l
=
+
k=1
U
k,l
. Then U
l
is open and E U
l
and it is trivial to see
that U
l
E
+
k=1
(U
k,l
E
k
) from which we get
m
n
(U
l
E)
+
k=1
m
n
(U
k,l
E
k
) <
+
k=1
1
l2
k
=
1
l
.
Finally, dene A =
+
l=1
U
l
to get that E A and m
n
(AE) m
n
(U
l
E) <
1
l
for all l and thus
m
n
(A E) = 0.
(ii) If B is a countable union of compact sets, so that B E and m
n
(EB) = 0,
then, by Proposition 3.1, E B L
n
and thus E = B (E B) L
n
.
Now take E L
n
. Then E
c
L
n
and by (i) there is an A, a countable
intersection of open sets, so that E
c
A and m
n
(A E
c
) = 0.
We set B = A
c
, a countable union of closed sets, and we get m
n
(E B) =
m
n
(A E
c
) = 0. Now, let B =
+
j=1
F
j
, where each F
j
is closed. We then write
F
j
=
+
k=1
F
j,k
, where F
j,k
= F
j
([k, k] [k, k]) is a compact set. This
proves that B is a countable union of compact sets: B =
(j,k)NN
F
j,k
.
Theorem 4.4 says that every Lebesgue-measurable set in R
n
is, except from
a m
n
-null set, equal to a Borel set.
Theorem 4.5 (i) m
n
is the only measure on (R
n
, B
R
n) with m
n
(P) = vol
n
(P)
for every open-closed interval P.
(ii) (R
n
, L
n
, m
n
) is the completion of (R
n
, B
R
n, m
n
).
Proof: (i) If is any measure on (R
n
, B
R
n) with (P) = vol
n
(P) for all
open-closed intervals P, then it is trivial to see that (P) = + for any un-
bounded generalised open-closed interval P: just take any increasing sequence
N
of open-closed intervals having union P. Therefore (
m
j=1
P
j
) =
m
j=1
(P
j
) =
m
j=1
m
n
(P
j
) = m
n
(
m
j=1
P
j
) for all pairwise disjoint open-closed generalised
intervals P
1
, . . . , P
m
. Therefore the measures and m
n
are equal on the alge-
bra / =
m
j=1
P
j
[ m N, P
1
, . . . , P
m
pairwise disjoint open-closed generalised
intervals. By Theorem 2.4, the two measures are equal also on (/) = B
R
n.
(ii) Let (R
n
, B
R
n, m
n
) be the completion of (R
n
, B
R
n, m
n
).
By Theorem 4.3, (R
n
, L
n
, m
n
) is a complete extension of (R
n
, B
R
n, m
n
).
Hence, B
R
n L
n
and m
n
(E) = m
n
(E) for every E B
R
n.
Take any E L
n
and, using Theorem 4.4, nd a Borel set B so that B E
and m
n
(E B) = 0. Using Theorem 4.4 once more, nd a Borel set A so that
(E B) A and m
n
(A (E B)) = 0. Therefore, m
n
(A) = m
n
(A (E B)) +
m
n
(E B) = 0.
Hence we can write E = B L, where B B
R
n and L = E B A B
R
n
with m
n
(A) = 0. After Theorem 2.3, we see that E has the form of the typical
element of B
R
n and, thus, L
n
B
R
n. This concludes the proof.
Theorem 4.6 Suppose E L
n
with m
n
(E) < +. For arbitrary > 0, there
are pairwise disjoint open intervals R
1
, . . . , R
l
so that m
n
(E(R
1
R
l
)) < .
Proof: We consider a covering E
+
j=1
R
j
by open intervals such that
+
j=1
vol
n
(R
j
) < m
n
(E) +

2
.
Now we consider the open-closed interval P
j
which has the same edges as
R
j
, and then E
+
j=1
P
j
and
+
j=1
vol
n
(P
j
) < m
n
(E) +

2
.
Take m so that
+
j=m+1
vol
n
(P
j
) <

2
and observe that E(P
1
P
m
)
+
j=m+1
P
j
and (P
1
P
m
)E
_
+
j=1
P
j
_
E. Hence m
n
(E(P
1
P
m
))
+
j=m+1
vol
n
(P
j
) <

2
and m
n
((P
1
P
m
)E) m
n
(
+
j=1
P
j
)m
n
(E) <

2
.
Altogether,
m
n
(E(P
1
P
m
)) < .
Proposition 1.10 implies that P
1
P
m
= P
1
P
l
for some pairwise
disjoint open-closed intervals P
1
, , P
l
, so that
m
n
(E(P
1
P
l
)) < .
We consider R
k
to be the open interval with the same edges as P
k
so that
l
k=1
R
k

l
k=1
P
k
and m
n
((
l
k=1
P
k
) (
l
k=1
R
k
))

l
k=1
m
n
(P
k
R
k
) = 0.
This easily implies that
m
n
(E(R
1
R
l
)) < .
4.3 Lebesgue-measure and simple transforma-
tions.
Some of the simplest and most important transformations of R
n
are the trans-
lations and the linear transformations.
4.3. LEBESGUE-MEASURE AND SIMPLE TRANSFORMATIONS. 47
Every y R
n
denes the translation
y
: R
n
R
n
by the formula
y
(x) = x +y, x R
n
.
Then
y
is a one-to-one transformation of R
n
onto R
n
and its inverse transfor-
mation is
y
.
y
is linear only if y = 0. For every E R
n
we dene
y +E = y +x[ x E (=
y
(E)).
Every > 0 denes the dilation l
: R
n
R
n
by the formula
l
(x) = x, x R
n
.
Then l
is a linear one-to-one transformation of R

n
onto R
n
and its inverse
transformation is l 1
. For every E R
n
we dene
E = x[ x E (= l
(E)).
If S is any interval in R, then any translation transforms it onto another
interval (of the same type) with the same volume. In fact, if a
1
, b
1
, . . . , a
n
, b
n
are the end-points of the edges of S, then the translated y +S has y
1
+a
1
, y
1
+
b
1
, . . . , y
n
+ a
n
, y
n
+ b
n
as end-points of its edges. Therefore vol
n
(y + S) =
_
(y
1
+b
1
)(y
1
+a
1
)
_

_
(y
n
+b
n
)(y
n
+a
n
)
_
= (b
1
a
1
) (b
n
a
n
) = vol
n
(S).
If we dilate the interval S with a
1
, b
1
, . . . , a
n
, b
n
as end-points of its edges
by the number > 0, then we get the interval S with a
1
, b
1
, . . . , a
n
, b
n
as end-points of its edges. Therefore, vol
n
(S) = (b
1
a
1
) (b
n
a
n
) =
n
(b
1
a
1
) (b
n
a
n
) =
n
vol
n
(S).
Another transformation is r, reection through 0, with the formula
r(x) = x, x R
n
.
This is one-to-one onto R
n
, linear and it is the inverse of itself. We dene
E = x[ x E (= r(E))
for all E R
n
. If S is any interval with a
1
, b
1
, . . . , a
n
, b
n
as end-points of its
edges, then S is an interval with b
1
, a
1
, . . . , b
n
, a
n
as end-points of its
edges and vol
n
(S) = (a
1
+b
1
) (a
n
+b
n
) = vol
n
(S).
After all these, we may say that n-dimensional volume of intervals is invari-
ant under translations and reection and it is positive-homogeneous of degree n
under dilations.
We shall see that the same are true for n-dimensional Lebesgue-measure of
Lebesgue-measurable sets in R
n
.
Theorem 4.7 (i) L
n
is invariant under translations, reection and dilations.
That is, for all A L
n
we have that y+A, A, A L
n
for every y R
n
, > 0.
(ii) m
n
is invariant under translations and reection and positive-homogeneous
of degree n under dilations. That is, for all A L
n
we have that
m
n
(y +A) = m
n
(A), m
n
(A) = m
n
(A), m
n
(A) =
n
m
n
(A)
for every y R
n
, > 0.
N
Proof: (a) Let E R
n
and y R
n
. Then for all coverings E
+
j=1
R
j
by open
intervals we get y+E
+
j=1
(y+R
j
). Hence m
n
(y+E)
+
j=1
vol
n
(y+R
j
) =
+
j=1
vol
n
(R
j
). This implies that m
n
(y + E) m
n
(E). Now, applying this to
y+E translated by y, we get m
n
(E) = m
n
(y+(y+E)) m
n
(y+E). Hence
m
n
(y +E) = m
n
(E)
for all E R
n
and y R
n
.
Similarly, E
+
j=1
(R
j
), which implies m
n
(E)
+
j=1
vol
n
(R
j
) =
+
j=1
vol
n
(R
j
). Hence m
n
(E) m
n
(E). Applying this to E, we also get
m
n
(E) = m
n
((E)) m
n
(E) and thus
m
n
(E) = m
n
(E)
for all E R
n
.
Also, E
+
j=1
(R
j
), from which we get m
n
(E)
+
j=1
vol
n
(R
j
) =
+
j=1
vol
n
(R
j
) and hence m
n
(E)
n
m
n
(E). Applying to
1
and to E,
we nd m
n
(E) = m
n
(
1
(E)) (
1
)
n
m
n
(E), which gives
m
n
(E) =
n
m
n
(E).
(b) Suppose now that A L
n
and E R
n
.
Then m
n
(E(y+A))+m
n
(E(y+A)
c
) = m
n
_
y+[(y+E)A]
_
+m
n
_
y+
[(y +E) A
c
]
_
= m
n
_
(y +E) A
_
+m
n
_
(y +E) A
c
_
= m
n
(y +E) =
m
n
(E). Therefore y +A L
n
.
Also m
n
(E(A)) +m
n
(E(A)
c
) = m
n
_
[(E) A]
_
+m
n
_
[(E)
A
c
]
_
= m
n
_
(E) A
_
+ m
n
_
(E) A
c
_
= m
n
(E) = m
n
(E). Therefore
A L
n
.
Finally m
n
(E (A)) +m
n
(E (A)
c
) = m
n
_
[(
1
E) A]
_
+m
n
_
[(
1
E)
A
c
]
_
=
n
m
n
_
(
1
E)A
_
+
n
m
n
_
(
1
E)A
c
_
=
n
m
n
(
1
E) = m
n
(E). Therefore
A L
n
.
(c) If A L
n
, then m
n
(y + A) = m
n
(y + A) = m
n
(A) = m
n
(A), m
n
(A) =
m
n
(A) = m
n
(A) = m
n
(A) and m
n
(A) = m
n
(A) =
n
m
n
(A) =
n
m
n
(A).
Reection and dilations are special cases of linear transformations of R
n
. As
is well known, a linear transformation of R
n
is a function T : R
n
R
n
such
that
T(x +y) = T(x) +T(y) , x, y R
n
,
and every such T has a determinant, det(T) R. In particular, det(r) = (1)
n
and det(l
) =
n
.
We recall that T is one-to-one and onto R
n
if and only if det(T) ,= 0 and
that, if T = T
1
T
2
, then det(T) = det(T
1
) det(T
2
).
Theorem 4.8 Let T : R
n
R
n
be a linear transformation. If A L
n
, then
T(A) L
n
and m
n
(T(A)) = [ det(T)[ m
n
(A).
4.3. LEBESGUE-MEASURE AND SIMPLE TRANSFORMATIONS. 49
Proof: At rst we assume that det(T) ,= 0.
(a) If T has the form T(x
1
, x
2
, . . . , x
n
) = (x
1
, x
2
, . . . , x
n
) for a certain
R 0, then det(T) = and, if P = (a
1
, b
1
] (a
2
, b
2
] (a
n
, b
n
], then
T(P) = (a
1
, b
1
] (a
2
, b
2
] (a
n
, b
n
] or T(P) = (b
1
, a
1
] (a
2
, b
2
]
(a
n
, b
n
], depending on whether > 0 or < 0. Thus T(P) is an interval and
m
n
(T(P)) = [[m
n
(P) = [ det(T)[m
n
(P).
If T(x
1
, x
2
, . . . , x
i1
, x
i
, x
i+1
, . . . , x
n
) = (x
i
, x
2
, . . . , x
i1
, x
1
, x
i+1
, . . . , x
n
)
for a certain i ,= 1, then det(T) = 1 and, if P = (a
1
, b
1
] (a
2
, b
2
]
(a
i1
, b
i1
] (a
i
, b
i
] (a
i+1
, b
i+1
] (a
n
, b
n
], then T(P) = (a
i
, b
i
] (a
2
, b
2
]
(a
i1
, b
i1
] (a
1
, b
1
] (a
i+1
, b
i+1
] (a
n
, b
n
]. Thus T(P) is an interval
and m
n
(T(P)) = m
n
(P) = [ det(T)[m
n
(P).
If T(x
1
, . . . , x
i1
, x
i
, x
i+1
, . . . , x
n
) = (x
1
, . . . , x
i1
, x
i
+ x
1
, x
i+1
, . . . , x
n
) for
a certain i ,= 1, then det(T) = 1 and, if P = (a
1
, b
1
] (a
i1
, b
i1
]
(a
i
, b
i
] (a
i+1
, b
i+1
] (a
n
, b
n
], then T(P) is not an interval any more but
T(P) = (y
1
, . . . , y
n
) [ y
j
(a
j
, b
j
] for j ,= i, y
i
y
1
(a
i
, b
i
] is a Borel set
and hence it is in L
n
. We dene the following three auxilliary sets: L =
(a
1
, b
1
] (a
i1
, b
i1
] (a
i
+ a
1
, b
i
+ b
1
] (a
i+1
, b
i+1
] (a
n
, b
n
],
M = (y
1
, . . . , y
n
) [ y
j
(a
j
, b
j
] for j ,= i, a
i
+ a
1
< y
i
a
i
+ y
1
and N =
(y
1
, . . . , y
n
) [ y
j
(a
j
, b
j
] for j ,= i, b
i
+ a
1
< y
i
b
i
+ y
1
. It is easy to see
that all four sets, T(P), L, M, N, are Borel sets and T(P) M = , L N = ,
T(P) M = LN and that N = M+x
0
, where x
0
= (0, . . . , 0, b
i
a
i
, 0, . . . , 0).
Then m
n
(T(P)) +m
n
(M) = m
n
(L) +m
n
(N) and m
n
(M) = m
n
(N), implying
that m
n
(T(P)) = m
n
(L) = m
n
(P) = [ det(T)[m
n
(P), because L is an interval.
(b) Now, let T be any linear transformation of the above three types. We
have shown that m
n
(T(P)) = [ det(T)[m
n
(P) for every open-closed interval
P. If R = (a
1
, b
1
) (a
n
, b
n
) it is easy to see, just as in the case of
open-closed intervals, that T(R) is a Borel set. We consider P
1
= (a
1
, b
1
]
(a
n
, b
n
] and P
2
= (a
1
, b
1
] (a
n
, b
n
] and, from P
2
R
P
1
we get T(P
2
) T(R) T(P
1
). Hence [ det(T)[m
n
(P
2
) m
n
(T(R))
[ det(T)[m
n
(P
1
) = [ det(T)[m
n
(R) and, taking the limit as 0+, we nd
m
n
(T(R)) = [ det(T)[m
n
(R) for every open interval R.
(c) Let, again, T be any linear transformation of one of the three types in (a).
Take any E R
n
and consider an arbitrary covering E
+
j=1
R
j
by open in-
tervals. Then T(E)
+
j=1
T(R
j
) and hence m
n
(T(E))
+
j=1
m
n
(T(R
j
)) =
[ det(T)[
+
j=1
m
n
(R
j
). Taking the inmum over all coverings, we conclude
m
n
(T(E)) [ det(T)[m
n
(E).
(d) If T is any linear transformation with det(T) ,= 0, there are linear trans-
formations T
1
, . . . , T
N
, where each is of one of the above three types so that
T = T
1
T
N
. Applying the result of (c) repeatedly, we nd m
n
(T(E))
[ det(T
1
)[ [ det(T
N
)[m
n
(E)[ = [ det(T)[m
n
(E) for every E R
n
. In this in-
equality, use now the set T(E) in the place of E and T
1
in the place of T,
and get m
n
(E) [ det(T
1
)[m
n
(T(E)) = [ det(T)[
1
m
n
(T(E)). Combining
the two inequalities, we conclude that
m
n
(T(E)) = [ det(T)[m
n
(E)
N
for every linear transformation T with det(T) ,= 0 and every E R
n
.
(e) Let A L
n
. For all E R
n
we get m
n
(E T(A)) + m
n
(E (T(A))
c
) =
m
n
_
T(T
1
(E) A)
_
+ m
n
_
T(T
1
(E) A
c
)
_
= [ det(T)[[m
n
(T
1
(E) A) +
m
n
(T
1
(E)A
c
)] = [ det(T)[m
n
(T
1
(E)) = m
n
(E). This says that T(A) L
n
.
Moreover,
m
n
(T(A)) = m
n
(T(A)) = [ det(T)[m
n
(A) = [ det(T)[m
n
(A).
If det(T) = 0, then V = T(R
n
) is a linear subspace of R
n
with dim(V )
n 1. We shall prove that m
n
(V ) = 0 and, from the completeness of m
n
, we
shall conclude that T(A) V is in L
n
with m
n
(T(A)) = 0 = [ det(T)[m
n
(A)
for every A L
n
.
Let f
1
, . . . , f
m
be a base of V (with m n 1) and complete it to a base
f
1
, . . . , f
m
, f
m+1
, . . . , f
n
of R
n
. Take the linear transformation S : R
n
R
n
given by
S(x
1
f
1
+ +x
n
f
n
) = (x
1
, . . . , x
n
).
Then S is one-to-one and hence det(S) ,= 0. Moreover
S(V ) = (x
1
, . . . , x
m
, 0, . . . , 0) [ x
1
, . . . , x
m
R.
We have S(V ) =
+
k=1
Q
k
, where Q
k
= [k, k] [k, k] 0 0.
Each Q
k
is a closed interval in R
n
with m
n
(Q
k
) = 0. Hence, m
n
(S(V )) = 0
and, then, m
n
(V ) = [ det(S)[
1
m
n
(S(V )) = 0.
If b, b
1
, . . . , b
n
R
n
, then the set
M = b +
1
b
1
+ +
n
b
n
[ 0
1
, . . . ,
n
1
is the typical closed parallelepiped in R
n
. One of the vertices of M is b and
b
1
, . . . , b
n
are the edges of M which start from b. For such an M we dene the
linear transformation T : R
n
R
n
by T(x) = T(x
1
, . . . , x
n
) = x
1
b
1
+ +x
n
b
n
for every x = (x
1
, . . . , x
n
) R
n
. We also consider the translation
b
and observe
that
M =
b
_
T(Q
0
)
_
,
where Q
0
= [0, 1]
n
is the unit qube in R
n
. Theorems 4.7 and 4.8 imply that M
is Lebesgue-measurable and
m
n
(M) = m
n
_
T(Q
0
)
_
= [ det(T)[m
n
(Q
0
) = [ det(T)[.
The matrix of T with respect to the standard basis e
1
, . . . , e
n
of R
n
has as
columns the vectors T(e
1
) = b
1
, . . . , T(e
n
) = b
n
. We conclude with the rule that
the Lebesgue-measure of a closed parallelepiped is given by the absolute value of
the determinant of the matrix having as columns the sides of the parallelepiped
starting from one of its vertices.
4.4. CANTORS SET. 51
4.4 Cantors set.
Since a set x consisting of only one point of R
n
is a degenerate interval,
we see that m
n
(x) = vol
n
(x) = 0. Now, every countable subset of R
n
has
Lebesgue-measure zero: if A = x
1
, x
2
, . . . then m
n
(A) =
+
k=1
m
n
(x
k
) = 0.
The aim of this section is to provide an uncountable set in Rwhose Lebesgue-
measure is zero.
We start with the interval
I
0
= [0, 1],
then take
I
1
=
_
0,
1
3
_
2
3
, 1
,
next
I
2
=
_
0,
1
9
_
2
9
,
1
3
_
2
3
,
7
9
_
8
9
, 1
,
and so on, each time dividing each of the intervals we get at the previous stage
into three subintervals of equal length and keeping only the two closed subin-
tervals on the sides.
Therefore we construct a decreasing sequence I
n
of closed sets so that
every I
n
consists of 2
n
closed intervals all of which have the same length
1
3
n
.
We dene
C =
+
n=1
I
n
and call it the Cantors set.
C is a compact subset of [0, 1] with m
1
(C) = 0. To see this observe that for
every n, m
1
(C) m
1
(I
n
) = 2
n
1
3
n
which tends to 0 as n +.
We shall prove by contradiction that C is uncountable. Namely, assume that
C = x
1
, x
2
, . . .. We shall describe an inductive process of picking one from
the subintervals constituting each I
n
. It is obvious that every x
n
belongs to I
n
,
since it belongs to C.
At the rst step choose the interval I
(1)
to be the subinterval of I
1
which
does not contain x
1
. Now, I
(1)
includes two subintervals of I
2
and at the second
step choose the interval I
(2)
to be whichever of these two subintervals of I
(1)
does not contain x
2
. (If both do not contain x
2
, just take the left one.) And
continue inductively: if you have already chosen I
(n1)
from the subintervals
of I
n1
, then this includes two subintervals of I
n
. Choose as I
(n)
whichever of
these two subintervals of I
(n1)
does not contain x
n
. (If both do not contain
x
n
, just take the left one.)
This produces a sequence I
(n)
of intervals with the following properties:
(i) I
(n)
I
n
for all n,
(ii) I
(n)
I
(n1)
for all n,
(iii) vol
1
(I
(n)
) =
1
3
n
0 and
(iv) x
n
/ I
(n)
for all n.
From (ii) and (iii) we conclude that the intersection of all I
(n)
s contains a single
point:
+
n=1
I
(n)
= x
0
N
for some x
0
. From (i) we see that x
0
I
n
for all n and thus x
0
C. Therefore,
x
0
= x
n
for some n N. But then x
0
I
(n)
and, by (iv), the same point x
n
does not belong to I
(n)
.
We get a contradiction and hence C is uncountable.
4.5 A non-Lebesgue-measurable set in R.
We consider the following equivalence relation in the set [0, 1). For any x, y
[0, 1) we write x y if and only if xy Q. That is an equivalence relation
is easy to see:
(a) x x, because x x = 0 Q.
(b) If x y, then x y Q, then y x = (x y) Q, then y x.
(c) If x y and y z, then x y Q and y z Q, then x z =
(x y) + (y z) Q, then x z.
Using the Axiom of Choice, we form a set N containing exactly one element
from each equivalence class of . This means that:
(i) for every x [0, 1) there is exactly one x N so that x x Q.
Indeed, if we consider the equivalence class of x and the element x of N from
this equivalence class, then x x and hence x x Q. Moreover, if there are
two x, x N so that xx Q and xx Q, then x x and x x, implying
that N contains two dierent elements from the equivalence class of x.
Our aim is to prove that N is not Lebesgue-measurable.
We form the set
A =
_
rQ[0,1)
(N +r).
Diferent (N +r)
s are disjoint:
(ii) if r
1
, r
2
Q [0, 1) and r
1
,= r
2
, then (N +r
1
) (N +r
2
) = .
Indeed, if x (N + r
1
) (N + r
2
), then x r
1
, x r
2
N. But x x r
1
and x xr
2
, implying that N contains two dierent (since r
1
,= r
2
) elements
from the equivalence class of x.
(iii) A [0, 2).
This is clear, since N [0, 1) implies N +r [0, 2) for all r Q [0, 1).
Take an arbitrary x [0, 1) and, by (i), the unique x N with x x Q.
Since 1 < x x < 1 we consider cases: if r = x x [0, 1), then x = x +r
N+r A, while if r = xx (1, 0), then x+1 = x+(r+1) N+(r+1) A.
Therefore, for every x [0, 1) either x A or x + 1 A. It is easy to see that
exactly one of these two cases is true. Because if x A and x + 1 A,
then x N + r
1
and x + 1 N + r
2
for some r
1
, r
2
Q [0, 1). Hence,
xr
1
, x+1 r
2
N and N contains two dierent (since r
2
r
1
,= 1) elements
4.5. A NON-LEBESGUE-MEASURABLE SET IN R. 53
of the equivalence class of x. Thus, if we dene the sets
E
1
= x [0, 1) [ x A, E
2
= x [0, 1) [ x + 1 A
then we have proved that
(iv) E
1
E
2
= [0, 1), E
1
E
2
= .
From (iv) we shall need only that [0, 1) E
1
E
2
.
We can also prove that
(v) E
1
(E
2
+ 1) = A, E
1
(E
2
+ 1) = .
In fact, the second is easy because E
1
, E
2
[0, 1) and hence E
2
+ 1 [1, 2).
The rst is also easy. If x E
1
then x A. If x E
2
+ 1 then x 1 E
2
and then x = (x 1) + 1 A. Thus E
1
(E
2
+ 1) A. On the other hand,
if x A [0, 2), then, either x A [0, 1) implying x E
1
, or x A [1, 2)
implying x 1 E
2
i.e. x E
2
+ 1. Thus A E
1
(E
2
+ 1).
From (v) we shall need only that E
1
, E
2
+ 1 A.
Now suppose that N is Lebesgue-measurable. By (ii) and by the invari-
ance of m
1
under translations, we get that m
1
(A) =
rQ[0,1)
m
1
(N + r) =
rQ[0,1)
m
1
(N). If m
1
(N) > 0, then m
1
(A) = +, contradicting (iii). If
m
1
(N) = 0, then m
1
(A) = 0, implying by (v) that m
1
(E
1
) = m
1
(E
2
+ 1) = 0,
hence m
1
(E
1
) = m
1
(E
2
) = 0, and nally from (iv), 1 = m
1
([0, 1)) m
1
(E
1
) +
m
1
(E
2
) = 0.
We arrive at a contradiction and N is not Lebesgue-measurable.
N
4.6 Exercises.
1. If A L
n
and A is bounded, prove that m
n
(A) < +. Give an example
of an A L
n
which is not bounded but has m
n
(A) < +.
2. The invariance of Lebesgue-measure under isometries.
Let T : R
n
R
n
be an isometric linear transformation. This means that
T is a linear transformation satisfying [T(x) T(y)[ = [x y[ for every
x, y R
n
or, equivalently, TT
= T
T = I, where T
is the adjoint of T
and I is the identity transformation.
Prove that, for every E L
n
, we have m
n
(T(E)) = m
n
(E).
3. A parallelepiped in R
n
is called degenerate if it is included in a hyper-
plane of R
n
, i.e. in a set of the form b + V , where b R
n
and V is a
linear subspace of R
n
with dim(V ) n 1.
Prove that a parallelepiped M is degenerate if and only if m
n
(M) = 0.
4. State in a formal way and prove the rule
volume = base area height
for parallelepipeds in R
n
.
5. Regularity of Lebesgue-measure.
Suppose that A L
n
.
(i) Prove that there is a decreasing sequence U
j
of open sets in R
n
so
that A U
j
for all j and m
n
(U
j
) m
n
(A). Conclude that m
n
(A) =
infm
n
(U) [ U open A.
(ii) Prove that there is an increasing sequence K
j
of compact sets in
R
n
so that K
j
A for all j and m
n
(K
j
) m
n
(A). Conclude that
m
n
(A) = supm
n
(K) [ K compact A.
The validity of (i) and (ii) for (R
n
, L
n
, m
n
) is called regularity. We shall
study this notion in chapter 5.
6. An example of an m
1
-null uncountable set, dense in an interval.
Let Q [0, 1] = x
1
, x
2
, . . .. For every > 0 we dene
U() =
+
_
j=1
_
x
j

2
j
, x
j
+

2
j
_
, A =
+
n=1
U
_
1
n
_
.
(i) Prove that m
1
(U()) 2.
(ii) If <
1
2
, prove that [0, 1] is not a subset of U().
(iii) Prove that A [0, 1] and m
1
(A) = 0.
(iv) Prove that Q [0, 1] A and that A is uncountable.
7. Let A = Q [0, 1]. If R
1
, . . . , R
m
are open intervals so that A
m
j=1
R
j
,
prove that 1
m
j=1
vol
1
(R
j
). Discuss the contrast to m
1
(A) = 0.
4.6. EXERCISES. 55
8. Prove that the Cantors set is perfect: it is closed and has no isolated
point.
9. The Cantors set and ternary expansions of numbers.
(i) Prove that for every sequence a
n
in 0, 1, 2 the series
+
n=1
an
3
n
converges to a number in [0, 1].
(ii) Conversely, prove that for every number x in [0, 1] there is a sequence
a
n
in 0, 1, 2 so that x =
+
n=1
an
3
n
. Then we say that 0.a
1
a
2
. . . is a
ternary expansion of x and that a
1
, a
2
, . . . are the ternary digits of
this expansion.
(iii) If x [0, 1] is a rational
m
3
N
, where m 1(mod3) and N N, then x
has exactly two ternary expansions: one is of the form 0.a
1
. . . a
N1
1000 . . .
and the other is of the form 0.a
1
. . . a
N1
0222 . . . .
If x [0, 1] is either irrational or rational
m
3
N
, where m 0 or 2(mod3)
and N N, then it has exactly one ternary expansion which is not of
either one of the above forms.
(iv) Let C be the Cantors set. If x [0, 1], prove that x C if and only
if x has at least one ternary expansion containing no ternary digit equal
to 1.
10. The Cantors function.
Let I
0
= [0, 1], I
1
, I
2
, . . . be the sets used in the construction of the Cantors
set C. For each n N dene f
n
: [0, 1] [0, 1] as follows. If, going from
left to right, J
n
1
, . . . , J
n
2
n
1
are the 2
n
1 subintervals of [0, 1] I
n
, then
dene f
n
(0) = 0, f
n
(1) = 1, dene f
n
to be constant
k
2
n
in J
n
k
for all
k = 1, . . . , 2
n
1 and to be linear in each of the subintervals of I
n
in such
a way that f
n
is continuous in [0, 1].
(i) Prove that [f
n
(x)f
n1
(x)[
1
32
n
for all n 2 and all x [0, 1]. This
implies that for every x [0, 1] the series f
1
(x) +
+
k=2
(f
k
(x) f
k1
(x))
converges to a real number.
(ii) Dene f(x) to be the sum of the series appearing in (i), and prove
that [f(x) f
n
(x)[
1
32
n
for all x [0, 1]. Therefore, f
n
converges to f
uniformly in [0, 1].
(iii) Prove that f(0) = 0, f(1) = 1 and that f is continuous and increasing
in [0, 1].
(iv) Prove that for every n: f is constant
k
2
n
in J
n
k
for all k = 1, . . . , 2
n
1.
(v) Prove that, if x, y C and x < y and x, y are not end-points of the
same complementary interval of C, then f(x) < f(y).
This function f is called the Cantors function.
11. The dierence set of a set.
(i) Let E R with m
1
(E) > 0 and 0 < 1. Considering a covering
of E by open intervals of total length <
1
1
(E), prove that there is a
non-empty open interval (a, b) so that m
1
(E (a, b)) (b a).
(ii) Let E R be a Lebesgue-measurable set with m
1
(E) > 0. Applying
N
(i) with =
3
4
, prove that E(E+z)(a, b) ,= for all z with [z[ <
1
4
(ba).
(iii) Let E R be a Lebesgue-measurable set with m
1
(E) > 0. Prove
that the set D(E) = x y [ x, y E, called the dierence set of E,
includes some open interval of the form (, ).
12. Another construction of a non-Lebesgue-measurable set in R.
(i) For any x, y R dene x y if x y Q. Prove that is an
equivalence relation in R.
(ii) Let L be a set containing exactly one element from each of the equiv-
alence classes of . Prove that R =
rQ
(L + r) and that the sets
L +r, r Q, are pairwise disjoint.
(iii) Prove that the dierence set of L (see exercise 4.6.8) contains no ra-
tional number ,= 0.
(iv) Using the result of exercise 4.6.8, prove that L is non-Lebesgue-
measurable.
13. Non-Lebesgue-measurable sets are everywhere, I.
We shall prove that every E R with m
1
(E) > 0 includes at least one
non-Lebesgue-measurable set.
(i) Consider the non-Lebesgue-measurable set N [0, 1] which was con-
structed in section 4.5 and prove that, if B N is Lebesgue-measurable,
then m
1
(B) = 0. In other words, if M N has m
1
(M) > 0, then M is
non-Lebesgue-measurable.
(ii) Consider an arbitrary E R with m
1
(E) > 0. If = 1m
1
(N), then
0 < 1, and consider an interval (a, b) so that m
1
(E(a, b)) (ba)
(see exercise 4.6.8). Then the set N
= (b a)N + a is included in [a, b],

has m
1
(N
) = (1 ) (b a) and, if M
has m
1
(M
) > 0, then M
is non-Lebesgue-measurable.
(iii) Prove that E N
14. Non-Lebesgue-measurable sets are everywhere, II.
(i) Consider the set L from exercise 4.6.9. Then E =
rQ
(E (L + r))
and prove that the dierence set (exercise 4.6.8) of each E(L+r) contains
no rational number ,= 0.
(ii) Prove that, for at least one r Q, the set E(L+r) is non-Lebesgue-
measurable (using exercise 4.6.8).
15. Not all Lebesgue-measurable sets in R are Borel sets and not all continuous
functions map Lebesgue-measurable sets onto Lebesgue-measurable sets.
Let f : [0, 1] [0, 1] be the Cantors function constructed in exercise 4.6.7.
Dene g : [0, 1] [0, 2] by the formula
g(x) = f(x) +x, x [0, 1].
(i) Prove that g is continuous, strictly increasing, one-to-one and onto
[0, 2]. Its inverse function g
1
: [0, 2] [0, 1] is also continuous, strictly
increasing, one-to-one and onto [0, 1].
4.6. EXERCISES. 57
(ii) Prove that the set g([0, 1] C), where C is the Cantors set, is an open
set with Lebesgue-measure equal to 1. Therefore the set E = g(C) has
Lebesgue-measure equal to 1.
(iii) Exercises 4.6.10 and 4.6.11 give non-Lebesgue-measurable sets M
E. Consider the set K = g
1
(M) C. Prove that K is Lebesgue-
measurable.
(iv) Using exercise 1.5.6, prove that K is not a Borel set in R.
(v) g maps K onto M.
16. More Cantors sets.
Take an arbitrary sequence
n
so that 0 <
n
<
1
2
for all n. We split
I
0
= [0, 1] into the three intervals [0,
1
2
1
], (
1
2
1
,
1
2
+
1
), [
1
2
+
1
, 1] and
form I
1
as the union of the two closed intervals. Inductively, if we have
already constructed I
n1
as a union of certain closed intervals, we split
each of these intervals into three subintervals of which the two side ones
are closed and their proportion to the original is
1
2

n
. The union of the
new intervals is the I
n
.
We set K =
+
n=1
I
n
.
(i) Prove that K is compact, has no isolated points and includes no open
interval.
(ii) Prove that K is uncountable.
(iii) Prove that m
1
(I
n
) = (1 2
1
) (1 2
n
) for all n.
(iv) Prove that m
1
(K) = lim
n+
(1 2
1
) (1 2
n
).
(v) Taking
n
=

3
n
for all n, prove that m
1
(K) > 1 .
(Use that (1 a
1
) (1 a
n
) > 1 (a
1
+ + a
n
) for all n and all
a
1
, . . . , a
n
[0, 1]).
(vi) Prove that m
1
(K) > 0 if and only if
+
n=1
n
< +.
(Use the inequality you used for (v) and also that 1 a e
a
for all a.)
17. Uniqueness of Lebesgue-measure.
Prove that m
n
is the only measure on (R
n
, B
R
n) which is invariant
under translations (i.e. (E + x) = (E) for all Borel sets E and all x)
and which satises (Q
0
) = 1, where Q
0
= [1, 1] [1, 1].
18. Let E R be Lebesgue-measurable and A be a dense subset of R. If
m
1
(E(E +x)) = 0 for all x A, prove that m
1
(E) = 0 or m
1
(E
c
) = 0.
19. Let E R be Lebesgue-measurable and > 0. If m
1
(E(a, b)) (ba)
for all intervals (a, b), prove that m
1
(E
c
) = 0.
N
Chapter 5
Borel measures
5.1 Lebesgue-Stieltjes-measures in R.
Lemma 5.1 If a < b + and F : (a, b) R is increasing, then
(i) for all x [a, b) we have F(x+) = infF(y) [ x < y,
(ii) for all x (a, b] we have F(x) = supF(y) [ y < x,
(iii) if a < x < y < z < b, then F(x) F(x) F(x+) F(y) F(z)
F(z) F(z+),
(iv) for all x [a, b) we have F(x+) = lim
yx+
F(y),
(v) for all x (a, b] we have F(x) = lim
yx
F(y).
Proof: (i) Let M = infF(y) [ x < y. Then for every > M there is some
t > x so that F(t) < . Hence for all y (x, t) we have M F(y) < . This
says that F(x+) = M.
(ii) Similarly, let m = supF(y) [ y < x. Then for every < m there is some
t < x so that < F(t). Hence for all y (t, x) we have < F(y) m. This
says that F(x) = m.
(iii) F(x) is an upper bound of the set F(y) [ y < x and a lower bound of
F(y) [ x < y. This, by (i) and (ii), implies that F(x) F(x) F(x+) and,
of course, F(z) F(z) F(z+). Also, if x < y < z, then F(y) is an element
of both sets F(y) [ x < y and F(y) [ y < z. Therefore F(y) is between the
inmum of the rst, F(x+), and the supremum of the second set, F(z).
(iv) By the result of (i), for every > F(x+) there is some t > x so that
F(x+) F(t) < . This, combined with (iii), implies that F(x+) F(y) <
for all y (x, t). Thus, F(x+) = lim
yx+
F(y).
(v) By (ii), for every < F(x) there is some t < x so that < F(t) F(x).
This, combined with (iii), implies < F(y) F(x) for all y (t, x). Thus,
F(x) = lim
yx
F(y).
Consider now a
0
, b
0
with a
0
< b
0
+ and an increasing function
F : (a
0
, b
0
) R and dene a non-negative function acting on subintervals of
59
60 CHAPTER 5. BOREL MEASURES
(a
0
, b
0
) as follows:
((a, b)) = F(b) F(a+), ([a, b]) = F(b+) F(a),
((a, b]) = F(b+) F(a+), ([a, b)) = F(b) F(a).
The mnemonic rule is: if the end-point is included then approach it from the
outside while if the end-point is not included then approach it from the inside.
We use the collection of all open subintervals of (a
0
, b
0
) and the function
to dene, as an application of Theorem 3.2, the following outer measure on
(a
0
, b
0
):
F
(E) = inf
_
+
j=1
((a
j
, b
j
)) [ E
+
j=1
(a
j
, b
j
), (a
j
, b
j
) (a
0
, b
0
) for all j
_
,
for every E (a
0
, b
0
).
Theorem 3.1 implies that the collection of
F
-measurable sets is a -algebra
of subsets of (a
0
, b
0
), which we denote by
F
, and the restriction, denoted
F
,
of
F
on
F
is a complete measure.
Denition 5.1 The measure
F
is called the Lebesgue-Stieltjes-measure
induced by the (increasing) F : (a
0
, b
0
) R.
If F(x) = x for all x R, then (S) = vol
1
(S) for all intervals S and, in
this special case,
F
coincides with the 1-dimensional Lebesgue-measure m
1
on
R. Thus, the new measure is a generalization of Lebesgue-measure.
Following exactly the same procedure as with Lebesgue-measure, we shall
study the relation between the -algebra
F
and the Borel sets in (a
0
, b
0
). In
the following Lemmas 5.2 - 5.6 all intervals that appear are included in (a
0
, b
0
).
Lemma 5.2 Let P = (a, b] (a
0
, b
0
) and a = c
0
< c
1
< < c
m
= b. If
P
i
= (c
i1
, c
i
], then (P) = (P
1
) + + (P
m
).
Proof: A telescoping sum: (P
1
) + +(P
m
) =
m
i=1
(F(c
i
+) F(c
i1
+)) =
F(b+) F(a+) = ((a, b]).
Lemma 5.3 If P, P
1
, . . . , P
l
are open-closed subintervals of (a
0
, b
0
), P
1
, . . . , P
l
are pairwise disjoint and P = P
1
P
l
, then (P) = (P
1
) + +(P
l
).
Proof: Exactly one of P
1
, . . . , P
l
has the same right end-point as P. We rename
and call it P
l
. Then exactly one of P
1
, . . . , P
l1
has right end-point coinciding
with the left end-point of P
l
. We rename and call it P
l1
. We continue until
the left end-point of the last remaining subinterval, which we shall rename P
1
,
coincides with the left end-point of P. Then the result is the same as the result
of Lemma 5.2.
Lemma 5.4 If P, P
1
, . . . , P
l
0
, b
0
), P
1
, . . . , P
l
are pairwise disjoint and P
1
P
l
P, then (P
1
) + +(P
l
) (P).
5.1. LEBESGUE-STIELTJES-MEASURES IN R. 61
Proof: We know that P (P
1
P
l
) = P
1
P
k
for some pairwise disjoint
open-closed intervals P
1
, . . . , P
k
. By Lemma 5.3 we get (P) = (P
1
) + +
(P
l
) +(P
1
) + +(P
k
) (P
1
) + +(P
l
).
Lemma 5.5 Suppose that P, P
1
, . . . , P
l
0
, b
0
)
and P P
1
P
l
. Then (P) (P
1
) + +(P
l
).
Proof: We write P = P
1
P
l
where P
j
= P
j
P are open-closed intervals
included in P. Then write P = P
1
(P
2
P
1
)
_
P
l
(P
1
P
l1
)
_
.
Each of these l pairwise disjoint sets can be written as a nite union of pairwise
disjoint open-closed intervals: P
1
= P
1
and
P
j
(P
1
P
j1
) = P
j
1
P
j
mj
for 2 j l.
Lemma 5.3 (for the equality) and Lemma 5.4 (for the two inequalities) imply
(P) = (P
1
) +
l
j=2
_
mj
m=1
(P
j
m
)
_
(P
1
) +
l
j=2
(P
j
)
l
j=1
(P
j
).
Lemma 5.6 Let Q be a closed interval and R
1
, . . . , R
l
be open subintervals of
(a
0
, b
0
). If Q R
1
R
l
, then (Q) (R
1
) + + (R
l
).
Proof: Let Q = [a, b] and R
j
= (a
j
, b
j
) for j = 1, . . . , l. We dene for > 0
P
= (a , b], P
j,
= (a
j
, b
j
].
We shall rst prove that there is some
0
> 0 so that for all <
0
P
P
1,
P
l,
.
Suppose that, for all n, the above inclusion is not true for =
1
n
. Hence, for all n
there is x
n
(a
1
n
, b] so that x
n
/
l
j=1
(a
j
, b
j
1
n
]. By the Bolzano-Weierstrass
theorem there is a subsequence x
n
k
converging to some x. Looking carefully
at the various inequalities, we get x [a, b] and x /
l
j=1
(a
j
, b
j
). This is a
contradiction and the inclusion we want to prove is true for some
0
=
1
n0
. If
<
0
, then the inclusion is still true because the left side becomes smaller while
the right side becomes larger.
Now Lemma 5.5 gives for <
0
that
F(b+) F((a )+)
l
j=1
_
F((b
j
)+) F(a
j
+)
_
and, using Lemma 5.1,
(Q) = F(b+) F(a)
l
j=1
_
F(b
j
) F(a
j
+)
_
=
l
j=1
(R
j
).
Theorem 5.1 Let F : (a
0
, b
0
) R be increasing. Then every subinterval S of
(a
0
, b
0
) is
F
-measurable and
F
(S) = (S).
Proof: (a) Let Q = [a, b] (a
0
, b
0
).
Then
F
(Q) ((a , b + )) = F((b + )) F((a )+) for all small
enough > 0 and, thus,
F
(Q) F(b+) F(a) = ([a, b]).
For every covering Q
+
j=1
R
j
by open subintervals of (a
0
, b
0
), there is (by
compactness) l so that Q
l
j=1
R
j
. Lemma 5.6 implies (Q)
l
j=1
(R
j
)
+
j=1
(R
j
). Hence (Q)
F
(Q) and we conclude that
(Q) =
F
(Q)
for all closed intervals Q (a
0
, b
0
).
If P = (a, b] (a
0
, b
0
), then
F
(P) ((a, b + )) = F((b + )) F(a+)
for all small enough > 0. Hence
F
(P) F(b+) F(a+) = (P).
If R = (a, b) (a
0
, b
0
), then
F
(R) ((a, b)) = (R).
(b) Now let P = (a, b], R = (c, d) be included in (a
0
, b
0
) and take P
R
= (c, d].
We write
F
(R P) =
F
_
(P
R
P) ((d , d) P)
_

F
(P
R
P) +
F
((d , d)) (P
R
P) + F(d) F((d )+) by the results of (a). The
same inequalities, with P
c
instead of P, give
F
(R P
c
)
F
(P
R
P
c
) +
F(d) F((d )+). Taking the sum, we nd
F
(R P) +
F
(R P
c
)
(P
R
P) +
F
(P
R
P
c
) + 2[F(d) F((d )+)].
Now write P
R
P
c
= P
1
P
l
for pairwise disjoint open-closed intervals
and get (P
R
P) +
F
(P
R
P
c
) (P
R
P) +
l
j=1
F
(P
j
) (P
R
P) +
l
j=1
(P
j
) = (P
R
) by the results of (a) and Lemma 5.3.
Therefore
F
(R P) +
F
(R P
c
) (P
R
) + 2[F(d) F((d )+)] =
F((d )+) F(c+) +2[F(d) F((d )+)] and, taking limit,
F
(RP) +
F
(R P
c
) F(d) F(c+) = (R).
We proved that
F
(R P) +
F
(R P
c
) (R)
for all open intervals R and open-closed intervals P which are (a
0
, b
0
).
(c) Now consider arbitrary E (a
0
, b
0
) with
F
(E) < +. Take a covering
E
+
j=1
R
j
by open subintervals of (a
0
, b
0
) so that
+
j=1
(R
j
) <
F
(E) +.
By -subadditivity and the result of (b) we nd
F
(E P) +
F
(E P
c
)
+
j=1
_
F
(R
j
P) +
F
(R
j
P
c
)
_
+
j=1
(R
j
) <
F
(E) +.
Taking limit:
F
(E P) +
F
(E P
c
)
F
(E) concluding that P
F
.
If Q = [a, b] (a
0
, b
0
), we take any a
k
in (a
0
, b
0
) so that a
k
a and, then,
Q =
+
k=1
(a
k
, b]
F
. Moreover, by the results of (a),
F
(Q) =
F
(Q) = (Q).
If P = (a, b] (a
0
, b
0
), we take any a
k
in (a, b] so that a
k
a and
we get that
F
(P) = lim
k+
F
([a
k
, b]) = lim
k+
(F(b+) F(a
k
)) =
F(b+) F(a+) = (P).
If T = [a, b) (a
0
, b
0
), we take any b
k
in [a, b) so that b
k
b and we
get that T =
+
k=1
[a, b
k
]
F
. Moreover,
F
(T) = lim
k+
F
([a, b
k
]) =
lim
k+
(F(b
k
+) F(a)) = F(b) F(a) = (T).
5.1. LEBESGUE-STIELTJES-MEASURES IN R. 63
Finally, if R = (a, b) (a
0
, b
0
), we take any a
k
and b
k
in (a, b) so that
a
k
a, b
k
b and a
1
b
1
. Then R =
+
k=1
[a
k
, b
k
]
F
. Moreover
F
(R) =
lim
k+
F
([a
k
, b
k
]) = lim
k+
(F(b
k
+)F(a
k
)) = F(b)F(a+) = (R).
0
, b
0
) R be increasing. Then
F
is -nite and it
is nite if and only if F is bounded. Also,
F
_
(a
0
, b
0
)
_
= F(b
0
) F(a
0
+).
Proof: We consider any two sequences a
k
and b
k
in (a
0
, b
0
) so that a
k
a
0
,
b
k
b
0
and a
1
b
1
. Then (a
0
, b
0
) =
+
k=1
[a
k
, b
k
] and
F
([a
k
, b
k
]) = F(b
k
+)
F(a
k
) < + for all k. Hence
F
is -nite.
By continuity of
F
from below, F(b
0
) F(a
0
+) = lim
k+
(F(b
k
+)
F(a
k
)) = lim
k+
F
([a
k
, b
k
]) =
F
((a
0
, b
0
)).
Hence, if
F
is nite, then F(b
0
) < + and F(a
0
+) < +. This im-
plies that all values of F lie in the bounded interval [F(a
0
+), F(b
0
)] and F is
bounded. Conversely, if F is bounded, then the limits F(a
0
+), F(b
0
) are nite
and, by the previous calculation, we get
F
((a
0
, b
0
)) = F(b
0
)F(a
0
+) < +.
It is easy to prove that the collection of all subintervals of (a
0
, b
0
) generates
the -algebra of all Borel sets in (a
0
, b
0
). Indeed, let c be the collection of all
intervals in R and T be the collection of all subintervals of (a
0
, b
0
). It is clear
that T = c(a
0
, b
0
) and Propositions 1.14 and 1.15 imply that
B
(a0,b0)
= B
R
(a
0
, b
0
) = (c)(a
0
, b
0
) = (T).
0
, b
0
) R be increasing. Then all Borel sets in
(a
0
, b
0
) belong to
F
.
Proof: Theorem 5.1 implies that the collection T of all subintervals of (a
0
, b
0
) is
included in
F
. By the discussion of the previous paragraph, we conclude that
B
(a0,b0)
= (T)
F
.
0
, b
0
) R be increasing. Then for every E (a
0
, b
0
)
we have
(i) E
F
if and only if there is A (a
0
, b
0
), a countable intersection of open
sets, so that E A and
F
(A E) = 0.
(ii) E
F
if and only if there B, a countable union of compact sets, so that
B E and
F
(E B) = 0.
Proof: The proof is exactly the same as the proof of the similar Theorem 4.4.
Only the obvious changes have to be made: m
n
changes to
F
and m
n
to
F
,
R
n
changes to (a
0
, b
0
), vol
n
changes to and L
n
changes to
F
.
Therefore every set in
F
is, except from a
F
-null set, equal to a Borel set.
0
, b
0
) R be increasing. Then
(i)
F
is the only measure on
_
(a
0
, b
0
), B
(a0,b0)
_
with
F
((a, b]) = F(b+)F(a+)
for all intervals (a, b] (a
0
, b
0
).
(ii)
_
(a
0
, b
0
),
F
,
F
_
is the completion of
_
(a
0
, b
0
), B
(a0,b0)
,
F
_
.
Proof: The proof is similar to the proof of Theorem 4.5. Only the obvious
notational modications are needed.
It should be observed that the
F
measure of a set x consisting of a single
point x (a
0
, b
0
) is equal to
F
(x) = F(x+) F(x), the jump of F at x.
In other words, the
F
-measure of a one-point-set is positive if and only if F is
discontinuous there. Also, observe that the
F
-measure of an open interval is 0
if and only if F is constant in this interval.
It is very common in practice to consider the increasing function F with the
extra property of being continuous from the right. In this case the measure of
an open-closed interval takes the simpler form
F
((a, b]) = F(b) F(a).
This is not a serious restriction. Given any increasing F : (a
0
, b
0
) R we
may dene the function F
0
: (a
0
, b
0
) R by the formula F
0
(x) = F(x+) for all
x (a
0
, b
0
) and it is immediate from Lemma 5.1 that F
0
is increasing, continuous
from the right, i.e. F
0
(x+) = F
0
(x) for all x, and F
0
(x+) = F(x+), F
0
(x) =
F(x) for all x. This implies that F
0
, F have the same jump at every x and,
in particular, they have the same continuity points. Now it is obvious that
F
0
, F induce the same Lebesgue-Stieltjes-measure on (a
0
, b
0
), simply because
the corresponding functions (S) (from which the construction of the measures
F0
,
F
starts) have the same values at every interval S.
Summarising, given any increasing function there is another increasing func-
tion which is continuous from the right so that the Lebesgue-Stieltjes-measures
induced by the two functions are equal.
5.2 Borel measures.
Denition 5.2 Let X be a topological space and (X, , ) be a measure space.
The measure is called a Borel measure on X if B
X
, i.e. if all Borel
sets in X are in .
The Borel measure is called locally nite if for every x X there is
some open neighborhood U
x
of x (i.e. an open set containing x) such that
(U
x
) < +.
Observe that, for to be a Borel measure, it is enough to have that all open
sets or all closed sets are in . This is because B
X
is generated by the collections
of all open or all closed sets and because is a -algebra.
Examples
The Lebesgue-measure on R
n
and, more generally, the Lebesgue-Stieltjes mea-
sure on any generalized interval (a
0
, b
0
) (induced by any increasing function)
are locally nite Borel measures. In fact, the content of the following theo-
rem is that the only locally nite Borel measures on (a
0
, b
0
) are exactly the
Lebesgue-Stieltjes measures.
5.2. BOREL MEASURES. 65
Lemma 5.7 Let X be a topological space and a Borel measure on X. If is
locally nite, then (K) < + for every compact K X.
If is a locally nite Borel measure on R
n
, then (M) < + for every
bounded M R
n
.
Proof: We take for every x K an open neighborhood U
x
of x so that (U
x
) <
+. Since K
xK
U
x
and K is compact, there are x
1
, . . . , x
n
so that
K
n
k=1
U
x
k
. Hence, (K)
n
k=1
(U
x
k
) < +.
If M R
n
is bounded, then M is compact and then (M) (M) < +.
Theorem 5.6 Let a
0
< b
0
+ and c
0
(a
0
, b
0
). For every locally
nite measure on (a
0
, b
0
) there is a unique increasing and continuous from
the right F : (a
0
, b
0
) R so that =
F
on B
(a0,b0)
and F(c
0
) = 0. For any
other increasing and continuous from the right G : (a
0
, b
0
) R, it is true that
=
G
if and only if G diers from F by a constant.
Proof: Dene the function
F(x) =
_
((c
0
, x]), if c
0
x < b
0
,
((x, c
0
]), if a
0
< x < c
0
.
By Lemma 5.7, F is real valued and it is clear that F is increasing, by the
monotonicity of . Now take any decreasing sequence x
n
so that x
n
x. If
c
0
x, by continuity of from above, lim
n+
F(x
n
) = lim
n+
((c
0
, x
n
]) =
((c
0
, x]) = F(x). Also, if x < c
0
, then x
n
< c
0
for large n, and, by continuity of
from below, lim
n+
F(x
n
) = lim
n+
((x
n
, c
0
]) = ((x, c
0
]) = F(x).
Therefore F is continuous from the right at every x.
If we compare and the induced
F
at the intervals (a, b], we get
F
((a, b]) =
F(b)F(a) = ((a, b]), where the second equality becomes trivial by considering
cases. Theorem 5.5 implies that
F
= on B
(a0,b0)
.
If G is increasing, continuous from the right with
G
= (=
F
) on B
(a0,b0)
,
then G(x) G(c
0
) =
G
((c
0
, x]) =
F
((c
0
, x]) = F(x) F(c
0
) for all x c
0
and, similarly, G(c
0
) G(x) =
G
((x, c
0
]) =
F
((x, c
0
]) = F(c
0
) F(x) for all
x < c
0
. Therefore F, G dier by a constant. Hence, if F(c
0
) = 0 = G(c
0
), then
F, G are equal on (a
0
, b
0
).
If the locally nite Borel measure on (a
0
, b
0
) satises the ((a
0
, c
0
]) < +,
then we may make a dierent choice for F than the one in Theorem 5.6. Add
the constant ((a
0
, c
0
]) to the function of the theorem and get the function
F(x) = ((a
0
, x]), x (a
0
, b
0
).
This last function is called the cumulative distribution function of .
A central notion related to Borel measures is the notion of regularity, and
this is because of the need to replace the general Borel set (a somewhat obscure
object) by open or closed sets.
Let E be a Borel subset in a topological space X and a Borel measure on
X. It is clear that (K) (E) (U) for all K compact and U open with
K E U. Hence
sup(K) [ K compact E (E) inf(U) [ U open E.
Denition 5.3 Let X be a topological space and a Borel measure on X. Then
is called regular if the following are true for every Borel set E in X:
(i) (E) = inf(U) [ U open E,
(ii) (E) = sup(K) [ K compact E.
Therefore, is regular if the measure of every Borel set can be approximated
from above by the measures of larger open sets and from below by the measures
of smaller compact sets.
Proposition 5.1 Let O be any open set in R
n
. Then there is an increasing
sequence K
m
of compact subsets of O so that int(K
m
) O and, hence, K
m

O also.
Proof: Dene the sets
K
m
= x O[ [x[ m and [y x[
1
m
for all y / O,
where [x[
2
= x
2
1
+ +x
2
n
for all x = (x
1
, . . . , x
n
).
The set K
m
is bounded, since [x[ m for all x K
m
.
If x
j
is a sequence in K
m
converging to some x, then from [x
j
[ m for
all j we get [x[ m, and from [y x
j
[
1
m
for all j and for all y / O we get
[y x[
1
m
for all y / O. Thus x K
m
and K
m
is closed.
Thus K
m
is a compact subset of O and, clearly, K
m
K
m+1
O for all m.
Hence, int(K
m
) int(K
m+1
for every m.
Now take any x O and a small enough ball y [ [yx[ < 2 O. Consider
M so large that M max([x[ +,
1
). It is trivial to see that B(x; ) K

M
and
thus x int(K
M
). Therefore int(K
m
) O.
Theorem 5.7 Let X be a topological space with the properties that every open
set in X is the union of an increasing sequence of compact sets and that there
is an increasing sequence of compact sets whose interiors cover X.
Suppose that is a locally nite Borel measure on X. Then:
(i) For every Borel set E and every > 0 there is an open U and a closed F so
that F E U and (U E), (E F) < . If also (E) < +, then F can
be taken compact.
(ii) For every Borel set E in X there is A, a countable intersection of open sets,
and B, a countable union of compact sets, so that B E A and (A E) =
(E B) = 0.
(iii) is regular.
Proof: (a) Suppose that (X) < +.
Consider the collection o of all Borel sets E in X with the property expressed
in (i), namely that for every > 0 there is an open U and a closed F so that
F E U and (U E), (E F) < .
5.2. BOREL MEASURES. 67
Take any open set O X and arbitrary > 0. If we consider U = O, then
(U O) = 0 < . By assumption there is a sequence K
m
of compact sets
so that K
m
O. Therefore, O K
m
and, since (O K
1
) (X) < +,
continuity from above implies that lim
m+
(O K
m
) = 0. Therefore there
is some m so that, if F = K
m
, (O F) < .
Thus all open sets belong to o.
If E o and > 0 is arbitrary, we nd an open U and a closed F so
that F E U and (U E), (E F) < . Then F
c
is open, U
c
is closed,
U
c
E
c
F
c
and (F
c
E
c
) = (E F) < and (E
c
U
c
) = (U E) < .
This implies that E
c
o.
Now, take E
1
, E
2
, . . . o and E =
+
j=1
E
j
. For > 0 and each E
j
take
open U
j
and closed F
j
so that F
j
E
j
U
j
and (U
j
E
j
), (E
j
F
j
) <

2
j
.
Dene B =
+
j=1
F
j
and the open U =
+
j=1
U
j
so that B E U. Then
U E
+
j=1
(U
j
E
j
) and E B
+
j=1
(E
j
F
j
). This implies (U E)
+
j=1
(U
j
E
j
) <
+
j=1
2
j
= and, similarly, (E B) < . The problem now
is that B is not necessarily closed. Consider the closed sets F
j
= F
1
F
j
,
so that F
j
F
j+1
for all j. Then E F
j+1
E F
j
for all j and, since
(E F
1
) (X) < +, continuity from below implies lim
j+
(E F
j
) =
(
+
j=1
(EF
j
)) = (EB) < . Therefore there is some j so that (EF
j
) < .
The inclusion F
j
E is clearly true.
We conclude that E =
+
j=1
o and o is a -algebra.
Since o contains all open sets, we have that B
X
o and nish the proof of
the rst statement of (i) in the special case (X) < +.
(b) Now, consider the general case, and take any Borel set E in X which is
included in some compact set K X. For each x K we take an open
neighborhood U
x
of x with (U
x
) < +. By the compactness of K, there exist
x
1
, . . . , x
n
K so that K
n
k=1
U
x
k
. We form the open set G =
n
k=1
U
x
k
and
have that
E G, (G) < +.
We next consider the restriction
G
of on the G, which is dened by the
formula
G
(A) = (A G)
for all Borel sets A in X. It is clear that
G
is a Borel measure on X which is
nite, since
G
(X) = (G) < +.
By (a), for every > 0 there is an open U and a closed F so that F E U
and
G
(U E),
G
(E F) < . Since E G, we get ((G U) E) =
(G(U E)) =
G
(U E) < and (EF) = (G(EF)) =
G
(EF) < .
Therefore, if we consider the open set U
= G U, we get F E U
and
(U
E), (E F) < and the rst statement of (i) is now proved with no
restriction on (X) but only for Borel sets in X which are included in compact
subsets of X.
(c) We take an increasing sequence K
m
of compact sets so that int(K
m
) X.
For any Borel set E in X we consider the sets E
1
= E K
1
and E
m
= E
(K
m
K
m1
) for all m 2 and we have that E =
+
m=1
E
m
. Since E
m
K
m
,
(b) implies that for each m and every > 0 there is an open U
m
and a closed
F
m
so that F
m
E
m
U
m
and (U
m
E
m
), (E
m
F
m
) <

2
m
. Now dene the
open U =
+
m=1
U
m
and the closed (!, why?) F =
+
m=1
F
m
so that F E U.
As in the proof of (a), we easily get (U E), (E F) < .
This concludes the proof of the rst statement of (i).
(d) Let (E) < +. Take a closed F so that F E and (E F) < ,
and consider the compact sets K
m
of part (c). Then the sets F
m
= F K
m
are compact and F
m
F. Therefore, E F
m
E F and by continuity of
from above, (E F
m
) (E F). Thus there is a large enough m so that
(E F
m
) < . This proves the second statement of (i).
(e) Take open U
j
and closed F
j
so that F
j
E U
j
and (U
j
E), (EF
j
) <
1
j
.
Dene A =
+
j=1
U
j
and B =
+
j=1
F
j
so that B E A. Now for all j we
have (A E) (U
j
E) <
1
j
and (E B) (E F
j
) <
1
j
. Therefore
(A E) = (E B) = 0. We dene the compact sets K
j,m
= F
j
K
m
and
observe that B =
(j,m)NN
K
j,m
. This is the proof of (ii).
(f) If (E) = +, it is clear that (E) = inf(U) [ U open and E U.
Also, from (ii), there is some B =
+
m=1
K
m
, where all K
m
are compact, so
that B E and (B) = (E) = +. Consider the compact sets K
m
=
K
1
K
m
which satisfy K
m
B. Then (K
m
) = (B) = (E) and thus
sup(K) [ K compact and K E = (E).
If (E) < +, then, from (a), for every > 0 there is a compact K and
an open U so that K E U and (U E), (E K) < . This implies
(E) < (K) and (U) < (E) + and thus the proof of (iii) is complete.
Examples
1. Proposition 5.1 implies that the euclidean space R
n
satises the assumptions
of Theorem 5.7. Therefore, every locally nite Borel measure on R
n
is regular.
2. Let X be an open subset of R
n
with the subspace topology and we consider
any O X which is open in X. Then O is open in R
n
and, by Proposition 5.1,
there is an increasing sequence K
m
of compact sets so that int
R
n(K
m
) O.
The set int
R
n(K
m
) is the interior of K
m
with respect to R
n
but, since K
m
X,
it coincides with the interior int
X
(K
m
) of K
m
with respect to X. Theorem 5.7
implies again that every locally nite Borel measure on X is regular.
3. Let X be an closed subset of R
n
with the subspace topology and take any
O X which is open in X. Then O = O
X for some O
R
n
which is open
in R
n
and, by Proposition 5.1, there is an increasing sequence K
m
of compact
subsets of O
so that int
R
n(K
m
) O
, where the set int

R
n(K
m
) is the interior
of K
m
with respect to R
n
. We set K
m
= K
m
X and have that each K
m
is a
compact subset of O. Moreover, int
X
(K
m
) = int
R
n(K
m
) X for every m and,
thus int
X
(K
m
) O. Theorem 5.7 implies, now, that every locally nite Borel
measure on X is regular.
5.3. EXERCISES. 69
5.3 Exercises.
1. If < x
1
< x
2
< < x
N
< + and 0 <
1
, . . . ,
N
< +, then nd
(and draw) the cumulative distribution function of =
N
k=1
x
k
.
2. The Cantors measure.
Consider the Cantors function f extended to R by f(x) = 0 for all x < 0
and f(x) = 1 for all x > 1. Then f : R [0, 1] is increasing, continuous
and bounded.
(i) f is the cumulative distribution function of
f
.
(ii) Prove that
f
(C) =
f
(R) = 1.
(iii) Each one of the 2
n
subintervals of I
n
(look at the construction of C)
has
f
-measure equal to
1
2
n
.
3. Let be a locally nite Borel measure on R such that ((, 0]) < +.
Prove that there is a unique f : R R increasing and continuous from
the right so that =
f
and f() = 0. Which is this function?
4. Linear combinations of regular Borel measures.
If ,
1
,
2
are regular Borel measures on the topological space X and
[0, +), prove that and
1
+
2
are regular Borel measures on X.
5. Prove that every locally nite Borel measure on R
n
is -nite.
6. The support of a regular Borel measure.
Let be a regular Borel measure on the topological space X. A point
x X is called a support point for if (U
x
) > 0 for every open
neighborhood U
x
of x. The set
supp() = x X[ x is a support point for
is called the support of .
(i) Prove that supp() is a closed set in X.
(ii) Prove that (K) = 0 for all compact sets K (supp())
c
.
(iii) Using the regularity of , prove that
_
(supp()
c
)
_
= 0.
(iv) Prove that (supp())
c
is the largest open set in X which is -null.
7. If f is the Cantors function of exercise 5.3.2, prove that the support
(exercise 5.3.6) of
f
is the Cantors set C.
8. Supports of Lebesgue-Stieltjes-measures.
Let F : R R be any increasing function. Prove that the complement
of the support (exercise 5.3.6) of the measure
F
is the union of all open
intervals on each of which F is constant.
9. Let a : R [0, +] induce the point-mass-distribution on (R, T(R)).
Then is a Borel measure on R.
(i) Prove that is locally nite if and only if
RxR
a
x
< + for all
R > 0.
(ii) In particular, prove that, if is locally nite, then x R[ a
x
> 0 is
countable.
10. Restrictions of regular Borel measures.
Let be a -nite regular Borel measure on the topological space X and
Y be a Borel subset of X. Prove that the restriction
Y
is a regular Borel
measure on X.
11. Continuous regular Borel measures.
Let be a regular Borel measure on the topological space X so that
(x) = 0 for all x X. A measure satisfying this last property is called
continuous. Prove that for every Borel set A in X with 0 < (A) < +
and every t (0, (A)) there is some Borel set B in X so that B A and
(B) = t.
12. Let X be a separable, complete metric space and be a Borel measure
on X so that (X) = 1. Prove that there is some B, a countable union of
compact subsets of X, so that (B) = 1.
Chapter 6
Measurable functions
6.1 Measurability.
Denition 6.1 Let (X, ) and (Y,
) be measurable spaces and f : X Y .

We say that f is (,
)measurable if f
1
(E) for all E
.
Example
A constant function is measurable. In fact, let (X, ) and (Y,
) be mea-
surable spaces and f(x) = y
0
Y for all x X. Take arbitrary E
. If
y
0
E, then f
1
(E) = X . If y
0
/ E, then f
1
(E) = .
Proposition 6.1 Let (X, ) and (Y,
) measurable spaces and f : X Y .

Suppose that c is a collection of subsets of Y so that (c) =
. If f
1
(E)
for all E c, then f is (,
)measurable.
Proof: We consider the collection o = E Y [ f
1
(E) .
Since f
1
() = , it is clear that o.
If E o, then f
1
(E
c
) = (f
1
(E))
c
and thus E
c
o.
If E
1
, E
2
, . . . o, then f
1
(
+
j=1
E
j
) =
+
j=1
f
1
(E
j
) , implying that
+
j=1
E
j
o.
Therefore o is a -algebra of subsets of Y . c is, by hypothesis, included in
o and, thus,
= (c) o. This concludes the proof.

Proposition 6.2 Let X, Y be topological spaces and f : X Y be continuous
on X. Then f is (B
X
, B
Y
)measurable.
Proof: Let c be the collection of all open subsets of Y . Then, by continuity,
f
1
(E) is an open and, hence, Borel subset of X for all E c. Since (c) = B
Y
,
Proposition 6.1 implies that f is (B
X
, B
Y
)measurable.
6.2 Restriction and gluing.
If f : X Y and A X, then the function f
A
: A Y , dened by f
A
(x) =
f(x) for all x A, is the usual restriction of f on A.
71
72 CHAPTER 6. MEASURABLE FUNCTIONS
Lemma 6.1 Let be a -algebra of subsets of X and A and consider
A
= E A[ E . Then
A
is a -algebra of subsets of A.
Proof: It is clear that
A
.
If E
A
, then E A and E and hence A E A and A E .
Thus A E
A
.
If E
j

A
for all j, then E
j
A and E
j
for all j. Therefore
+
j=1
E
j
A
and
+
j=1
E
j
and thus
+
j=1
E
j

A
.
Denition 6.2 Let be a -algebra of subsets of X and A . The -algebra
A
of subsets of A, which was dened in the statement of Lemma 6.1, is called
the restriction of on A.
Proposition 6.3 Let (X, ), (Y,
) be measurable spaces and f : X Y .

Suppose A
1
, . . . , A
n
are pairwise disjoint with A
1
A
n
= X.
f is (,
)measurable if and only if f

Aj
is (
Aj
,
)measurable for all

j = 1, . . . , n.
Proof: Let f be (,
)measurable. For all E
we have f
1
Aj
(E) = f
1
(E)
A
j

Aj
because the set f
1
(E)A
j
belongs to and is included in A
j
. Hence
f
Aj
is (
Aj
,
)measurable for all j.

Now, let f
Aj
be (
Aj
,
)measurable for all j. For every E
we have
that f
1
(E) A
j
= f
1
Aj
(E)
Aj
and, hence, f
1
(E) A
j
for all j.
Therefore f
1
(E) = (f
1
(E) A
1
) (f
1
(E) A
n
) , implying that f
is (,
)measurable.
In a free language: measurability of a function separately on complementary
(measurable) pieces of the space is equivalent to measurability on the whole space.
There are two operations on measurable functions that are taken care of
by Proposition 6.3. One is the restriction of a function f : X Y on some
A X and the other is the gluing of functions f
Aj
: A
j
Y to form a single
f : X Y , whenever the nitely many A
j
s are pairwise disjoint and cover
X. The rules are: restriction of measurable functions on measurable sets are
measurable and gluing of measurable functions dened on measurable subsets
results to a measurable function.
6.3 Functions with arithmetical values.
Denition 6.3 Let (X, ) be measurable space and f : X R or R or C or
C. We say f is measurable if it is (, B
R
or B
R
or B
C
or B
C
)measurable,
respectively.
In the particular case when (X, ) is (R
n
, B
R
n) or (R
n
, L
n
), then we use
the term Borel-measurable or, respectively, Lebesgue-measurable for f.
If f : X R, then it is also true that f : X R. Thus, according to the
denition we have given, there might be a conict between the two meanings
6.3. FUNCTIONS WITH ARITHMETICAL VALUES. 73
of measurability of f. But, actually, there is no such conict. Suppose, for
example, that f is assumed (, B
R
)measurable. If E B
R
, then E R B
R
and, thus, f
1
(E) = f
1
(E R) . Hence f is (, B
R
)measurable. Let,
conversely, f be (, B
R
)measurable. If E B
R
, then E B
R
and, thus,
f
1
(E) . Hence f is (, B
R
)measurable.
The same question arises when f : X C, because it is then also true that
f : X C. Exactly as before, we may prove that f is (, B
C
)measurable if
and only if it is (, B
C
)measurable and there is no conict in the denition.
Proposition 6.4 Let (X, ) be measurable space and f : X R
n
. Let, for
each j = 1, . . . , n, f
j
: X R denote the j-th component function of f. Namely,
f(x) = (f
1
(x), . . . , f
n
(x)) for all x X.
Then f is (, B
R
n)measurable if and only if every f
j
is measurable.
Proof: Let f be (, B
R
n)measurable. For all intervals (a, b] we have
f
1
j
((a, b]) = f
1
(R R(a, b] R R)
which belongs to . Since the collection of all (a, b] generates B
R
, Proposition
6.1 implies that f
j
is measurable.
Now let every f
j
be measurable. Then
f
1
((a
1
, b
1
] (a
n
, b
n
]) = f
1
1
((a
1
, b
1
]) f
1
n
((a
n
, b
n
])
which is an element of . The collection of all open-closed intervals generates
B
R
n and Proposition 6.1, again, implies that f is (, B
R
n) measurable.
In a free language: measurability of a vector function is equivalent to mea-
surability of all component functions.
The next two results give simple criteria for measurability of real or complex
valued functions.
Proposition 6.5 Let (X, ) be measurable space and f : X R. Then f is
measurable if and only if f
1
((a, +)) for all a R.
Proof: Since (a, +) B
R
, one direction is trivial.
If f
1
((a, +)) for all a R, then f
1
((a, b]) = f
1
((a, +))
f
1
((b, +)) for all (a, b]. Now the collection of all intervals (a, b] generates
B
R
and Proposition 6.1 implies that f is measurable.
Of course, in the statement of Proposition 6.5 one may replace the intervals
(a, +) by the intervals [a, +) or (, b) or (, b].
If f : X C, then the functions (f), (f) : X R are dened by
(f)(x) = (f(x)) and (f)(x) = (f(x)) for all x X and they are called
the real part and the imaginary part of f, respectively.
Proposition 6.6 Let (X, ) be measurable space and f : X C. Then f is
measurable if and only if both (f) and (f) are measurable.
Proof: An immediate application of Proposition 6.4.
The next two results investigate extended-real or extended-complex valued
functions.
Proposition 6.7 Let (X, ) be measurable space and f : X R. The follow-
ing are equivalent.
(i) f is measurable.
(ii) f
1
(+), f
1
(R) and, if A = f
1
(R), the function f
A
: A R is
A
measurable.
(iii) f
1
((a, +]) for all a R.
Proof: It is trivial that (i) implies (iii), since (a, +] B
R
for all a R.
Assume (ii) and consider B = f
1
(+) and C = f
1
() =
(A B)
c
. The restrictions f
B
= + and f
C
= are constants and
hence are, respectively,
B
measurable and
C
measurable. Proposition 6.3
implies that f is measurable and thus (ii) implies (i).
Now assume (iii). Then f
1
(+) =
+
n=1
f
1
((n, +]) and then
f
1
((a, +)) = f
1
((a, +]) f
1
(+) for all a R. Moreover,
f
1
(R) =
+
n=1
f
1
((n, +)) . For all (a, +) we have f
1
A
((a, +)) =
f
1
((a, +))
A
, because the last set belongs to and is included in A.
Proposition 6.5 implies that f
A
is
A
measurable and (ii) is now proved.
Proposition 6.8 Let (X, ) be measurable space and f : X C. The follow-
ing are equivalent.
(i) f is measurable.
(ii) f
1
(C) and, if A = f
1
(C), the f
A
: A C is
A
measurable.
Proof: Assume (ii) and consider B = f
1
() = (f
1
(C))
c
. The restric-
tion f
B
is constant and hence
B
measurable. Proposition 6.3 implies that
f is measurable. Thus (ii) implies (i).
Now assume (i). Then A = f
1
(C) since C B
C
. Proposition 6.3
implies that f
A
is
A
measurable and (i) implies (ii).
6.4 Composition.
Proposition 6.9 Let (X, ), (Y,
), (Z,
) be measurable spaces and let f :

X Y , g : Y Z. If f is (,
)measurable and g is (
)measurable,
then g f : X Z is (,
)measurable.
Proof: For all E
we have (g f)
1
(E) = f
1
_
g
1
(E)
_
, because
g
1
(E)
.
Hence: composition of measurable functions is measurable.
6.5. SUMS AND PRODUCTS. 75
6.5 Sums and products.
The next result is: sums and products of real or complex valued measurable
functions are measurable functions.
Proposition 6.10 Let (X, ) be a measurable space and f, g : X R or C be
measurable. Then f +g, fg are measurable.
Proof: (a) We consider H : X R
2
by the formula H(x) = (f(x), g(x)) for all
x X. Proposition 6.4 implies that H is (, B
R
2)measurable. Now consider
, : R
2
R by the formulas (y, z) = y+z and (y, z) = yz. These functions
are continuous and Proposition 6.2 implies that they are (B
R
2, B
R
)measurable.
Therefore the compositions H, H : X R are measurable. But
( H)(x) = f(x) +g(x) = (f +g)(x) and ( H)(x) = f(x)g(x) = (fg)(x) for
all x X and we conclude that f +g = H and fg = H are measurable.
(b) In the case f, g : X Cwe consider (f), (f), (g), (g) : X R, which,
by Proposition 6.6, are all measurable. Then, part (a) implies that (f+g) =
(f) + (g), (f + g) = (f) + (g), (fg) = (f)(g) (f)(g), (fg) =
(f)(g) + (f)(g) are all measurable. Proposition 6.6 again, gives that
f +g, fg are measurable.
If we want to extend the previous results to functions with innite values,
we must be more careful.
The sums (+) +(), () +(+) are not dened in R and neither is
+ dened in C. Hence, when we add f, g : X R or C, we must agree
on how to treat the summation on, respectively, the set B = x X[ f(x) =
+, g(x) = or f(x) = , g(x) = + or the set B = x X[ f(x) =
, g(x) = . There are two standard ways to do this. One is to ignore the
bad set and consider f +g dened on B
c
X, on which it is naturally dened.
The other way is to choose some appropriate h dened on B and dene f +g = h
on B. The usual choice for h is some constant, e.g h = 0.
Proposition 6.11 Let (X, ) be a measurable space and f, g : X R be
measurable. Then the set
B = x X[ f(x) = +, g(x) = or f(x) = , g(x) = +
belongs to .
(i) The function f +g : B
c
R is
B
c measurable.
(ii) If h : B R is
B
measurable and we dene
(f +g)(x) =
_
f(x) +g(x), if x B
c
,
h(x), if x B,
,
then f +g : X R is measurable.
Similar results hold if f, g : X C and B = x X[ f(x) = , g(x) = .
Proof: We have
B =
_
f
1
(+) g
1
()
_
_
_
f
1
() g
1
(+)
_
.
(i) Consider the sets A = x X[ f(x), g(x) R, C
1
= x X[ f(x) =
+, g(x) ,= or f(x) ,= , g(x) = + and C
2
= x X[ f(x) =
, g(x) ,= + or f(x) ,= +, g(x) = . It is clear that A, C
1
, C
2
,
that B
c
= A C
1
C
2
and that the three sets are pairwise disjoint.
The restriction of f +g on A is the sum of the real valued f
A
, g
A
. By Propo-
sition 6.3, both f
A
, g
A
are
A
measurable and, by Proposition 6.10, (f +g)
A
=
f
A
+ g
A
is
A
measurable. The restriction (f + g)
C1
is constant +, and is
thus
C1
measurable. Also the restriction (f +g)
C2
= is
C2
measurable.
Proposition 6.3 implies that f +g : B
c
R is
B
cmeasurable.
(ii) This is immediate after the result of (i) and Proposition 6.3.
The case f, g : X C is similar, if not simpler.
For multiplication we make the following
Convention. () 0 = 0 () = 0 in R and 0 = 0 = 0 in C.
Thus, multiplication is always dened and we may state that: the product
of measurable functions is measurable.
Proposition 6.12 Let (X, ) be a measurable space and f, g : X R or C be
measurable. Then the function fg is measurable.
Proof: Let f, g : X R.
Consider the sets A = x X[ f(x), g(x) R, C
1
= x X[ f(x) =
+, g(x) > 0 or f(x) = , g(x) < 0 or f(x) > 0, g(x) = + or f(x) <
0, g(x) = , C
2
= x X[ f(x) = , g(x) > 0 or f(x) = +, g(x) <
0 or f(x) > 0, g(x) = or f(x) < 0, g(x) = + and D = x X[ f(x) =
, g(x) = 0 or f(x) = 0, g(x) = . These four sets are pairwise disjoint,
their union is X and they all belong to .
The restriction of fg on A is equal to the product of the real valued f
A
, g
A
,
which, by Propositions 6.3 and 6.10, is
A
measurable. The restriction (fg)
C1
is constant + and, hence,
C1
measurable. Similarly, (fg)
C2
= is
C2
measurable. Finally, (fg)
D
= 0 is
D
measurable.
Proposition 6.3 implies now that fg is measurable.
If f, g : X C, the proof is similar and slightly simpler.
6.6 Absolute value and signum.
The action of the absolute value on innities is: [ + [ = [ [ = + and
[[ = +.
Proposition 6.13 Let (X, ) be a measurable space and f : X R or C be
measurable. Then the function [f[ : X [0, +] is measurable.
Proof: Let f : X R. The function [ [ : R [0, +] is continuous and,
hence, (B
R
, B
R
)measurable. Therefore, [f[, the composition of [ [ and f, is
measurable.
The same proof applies in the case f : X C.
6.7. MAXIMUM AND MINIMUM. 77
Denition 6.4 For every z C we dene
sign(z) =
_
_
_
z
|z|
, if z ,= 0,
0, if z = 0,
, if z = .
If we denote C
= C 0, , then the restriction sign

C
: C
C is
continuous. This implies that, for every Borel set E in C, the set sign
1
C
(E)
is a Borel set contained in C
. The restriction sign

{0}
is constant 0 and the
restriction sign
{}
is constant . Therefore, for every Borel set E in C, the sets
sign
1
{0}
(E), sign
1
{}
(E) are Borel sets. Altogether, sign
1
(E) = sign
1
C
(E)
sign
1
{0}
(E) sign
1
{}
(E) is a Borel set in C. This means that sign : C C is
(B
C
, B
C
)measurable.
All this applies in the same way to the function sign : R R with the
simple formula
sign(x) =
_
1, if 0 < x +,
1, if x < 0,
0, if x = 0.
Hence sign : R R is (B
R
, B
R
)measurable.
For all z C we may write
z = sign(z) [z[
and this is called the polar decomposition of z.
Proposition 6.14 Let (X, ) be a measurable space and f : X R or C be
measurable. Then the function sign(f) is measurable.
Proof: If f : X R, then sign(f) is the composition of sign : R R and f
and the result is clear by Proposition 6.9. The same applies if f : X C.
6.7 Maximum and minimum.
Proposition 6.15 Let (X, ) be measurable space and f
1
, . . . , f
n
: X R be
measurable. Then the functions max(f
1
, . . . , f
n
), min(f
1
, . . . , f
n
) : X R
are measurable.
Proof: If h = max(f
1
, . . . , f
n
), then for all a R we have h
1
((a, +]) =
n
j=1
f
1
j
((a, +]) . Proposition 6.7 implies that h is measurable and
from min(f
1
, . . . , f
n
) = max(f
1
, . . . , f
n
) we see that min(f
1
, . . . , f
n
) is also
measurable.
The next result is about comparison of measurable functions.
Proposition 6.16 Let (X, ) be a measurable space and f, g : X R be
measurable. Then x X[ f(x) = g(x), x X[ f(x) < g(x) .
If f, g : X C is measurable, then x X [ f(x) = g(x) .
Proof: Consider the set A = x X[ f(x), g(x) R . Then the functions
f
A
, g
A
are
A
measurable and thus f
A
g
A
is
A
measurable. Hence the
sets x A[ f(x) = g(x) = (f
A
g
A
)
1
(0) and x A[ f(x) < g(x) =
(f
A
g
A
)
1
((, 0)) belong to
A
. This, of course, means that these sets
belong to (and that they are subsets of A).
We can obviously write x X[ f(x) = g(x) = x A[ f(x) = g(x)
_
f
1
() g
1
()
_ _
f
1
(+) g
1
(+)
_
. In a similar
manner, x X [ f(x) < g(x) = x A[ f(x) < g(x)
_
f
1
()
g
1
((, +])
_ _
f
1
([, +)) g
1
(+)
_
.
The case of f, g : X C and of x X [ f(x) = g(x) is even simpler.
6.8 Truncation.
There are many possible truncations of a function.
Denition 6.5 Let f : X R and consider , R with . We dene
f
()
()
(x) =
_
_
_
f(x), if f(x) ,
, if f(x) < ,
, if < f(x).
We write f
()
instead of f
()
()
and f
()
instead of f
(+)
()
.
The functions f
()
()
, f
()
, f
()
are called truncations of f.
Proposition 6.17 Let (X, ) be a measurable space and f : X R be a
measurable function. Then all truncations f
()
()
are measurable.
Proof: The proof is obvious after the formula f
()
()
= min
_
max(f, ),
_
.
An important role is played by the following special truncations.
Denition 6.6 Let f : X R. The f
+
: X [0, +] and f
: X [0, +]
dened by the formulas
f
+
(x) =
_
f(x), if 0 f(x),
0, if f(x) < 0,
f
(x) =
_
0, if 0 f(x),
f(x), if f(x) < 0,
are called, respectively, the positive part and the negative part of f.
It is clear that f
+
= f
(0)
and f
= f
(0)
. Hence if is a -algebra of subsets
of X and f is measurable, then both f
+
and f
are measurable. It is
also trivial to see that at every x X either f
+
(x) = 0 or f
(x) = 0 and that

f
+
+f
= [f[, f
+
f
= f.
There is another type of truncations used mainly for extended-complex val-
ued functions.
6.9. LIMITS. 79
Denition 6.7 Let f : X R or C and consider r [0, +]. We dene
(r)
f(x) =
_
f(x), if [f(x)[ r,
r sign(f(x)), if r < [f(x)[.
The functions
(r)
f are called truncations of f.
Observe that, if f : X R, then
(r)
f = f
(r)
(r)
.
Proposition 6.18 Let (X, ) be a measurable space and f : X R or C a
measurable function. Then all truncations
(r)
f are measurable.
Proof: Observe that the function
r
: R R with formula
r
(x) =
_
x, if [x[ r,
r sign(x), if r < [x[,
is continuous on R and hence (B
R
, B
R
)measurable. Now
(r)
f =
r
f is
measurable.
The proof in the case f : X C is similar.
6.9 Limits.
The next group of results is about various limiting operations on measurable
functions. The rule is, roughly: the supremum, the inmum and the limit of a
sequence of measurable functions are measurable functions.
Proposition 6.19 Let (X, ) be a measurable space and f
j
a sequence of
measurable functions f
j
: X R. Then all the functions sup
jN
f
j
,
inf
jN
f
j
, limsup
j+
f
j
and liminf
j+
f
j
are measurable.
Proof: Let h = sup
jN
f
j
: X R. For every a R we have h
1
((a, +]) =
+
j=1
f
1
j
((a, +]) . Proposition 6.7 implies that h is measurable.
Now inf
jN
f
j
= sup
jN
(f
j
) is also measurable.
And, nally, limsup
j+
f
j
= inf
jN
_
sup
kj
f
k
_
and liminf
j+
f
j
=
sup
jN
_
inf
kj
f
k
_
are measurable.
Proposition 6.20 Let (X, ) be a measurable space and f
j
a sequence of
measurable functions f
j
: X R. Then the set
A = x X[ lim
j+
f
j
(x) exists in R
belongs to .
(i) The function lim
j+
f
j
: A R is
A
measurable.
(ii) If h : A
c
R is
A
cmeasurable and we dene
( lim
j+
f
j
)(x) =
_
lim
j+
f
j
(x), if x A,
h(x), if x A
c
,
then lim
j+
f
j
: X R is measurable.
Similar results hold if f
j
: X C for all j and we consider the set A =
x X[ lim
j+
f
j
(x) exists in C.
Proof: (a) Suppose that f
j
: X R for all j.
Proposition 6.19 implies that limsup
j+
f
j
and liminf
j+
f
j
are both
measurable. Since lim
j+
f
j
(x) exists if and only if limsup
j+
f
j
(x) =
liminf
j+
f
j
(x), we have that
A = x X[ limsup
j+
f
j
(x) = liminf
j+
f
j
(x)
and Proposition 6.16 implies that A .
(i) It is clear that the function lim
j+
f
j
: A R is just the restriction of
limsup
j+
f
j
(or of liminf
j+
f
j
) to A and hence it is
A
measurable.
(ii) The proof of (ii) is a direct consequence of (i) and Proposition 6.3.
(b) Let now f
j
: X C for all j.
Consider the set B = x X[ lim
j+
f
j
(x) exists in C and the set C =
x X[ lim
j+
f
j
(x) = . Clearly, B C = A.
Now, C = x X[ lim
j+
[f
j
[(x) = +. Since [f
j
[ : X R for all
j, part (a) implies that the function lim
j+
[f
j
[ is measurable on the set on
which it exists. Therefore, C .
B is the intersection of B
1
= x X [ lim
j+
(f
j
)(x) exists in R and
B
2
= x X[ lim
j+
(f
j
)(x) exists in R. By part (a) applied to the se-
quences (f
j
), (f
j
) of real valued functions, we see that the two functions
lim
j+
(f
j
), lim
j+
(f
j
) are both measurable on the set on which each of
them exists. Hence, both B
1
, B
2
(the inverse images of R under these functions)
belong to and thus B = B
1
B
2
.
Therefore A = B C .
We have just seen that the functions lim
j+
(f
j
), lim
j+
(f
j
) are
measurable on the set where each of them exists and hence their restrictions to
B are both
B
measurable. These functions are, respectively, the real and the
imaginary part of the restriction to B of lim
j+
f
j
and Proposition 6.6 says
that lim
j+
f
j
is
B
measurable. Finally, the restriction to C of this limit
is constant and thus it is
C
measurable. By Proposition 6.3, lim
j+
f
j
is
A
measurable.
This is the proof of (i) in the case of complex valued functions and the proof
of (ii) is immediate after Proposition 6.3.
(c) Finally, let f
j
: X C for all j.
For each j we consider the function
g
j
(x) =
_
f
j
(x), if f
j
(x) ,= ,
j, if f
j
(x) = .
If we set A
j
= f
1
j
(C) , then (g
j
)
Aj
= (f
j
)
Aj
is
Aj
measurable. Also
(g
j
)
A
c
j
is constant j and hence
A
c
j
measurable. Therefore g
j
: X C is
measurable.
6.10. SIMPLE FUNCTIONS. 81
It is easy to show that the two limits lim
j+
g
j
(x) and lim
j+
f
j
(x)
either both exist or both do not exist and, if they do exist, they are equal. In
fact, let lim
j+
f
j
(x) = p C. If p C, then for large enough j we shall have
that f
j
(x) ,= , implying g
j
(x) = f
j
(x) and thus lim
j+
g
j
(x) = p. If p = ,
then [f
j
(x)[ +. Therefore [g
j
(x)[ min([f
j
(x)[, j) + and hence
lim
j+
g
j
(x) = = p in this case also. The converse is similarly proved. If
lim
j+
g
j
(x) = p C, then, for large enough j, g
j
(x) ,= j and thus f
j
(x) =
g
j
(x) implying lim
j+
f
j
(x) = lim
j+
g
j
(x) = p. If lim
j+
g
j
(x) = ,
then lim
j+
[g
j
(x)[ = +. Since [f
j
(x)[ [g
j
(x)[ we get lim
j+
[f
j
(x)[ =
+ and thus lim
j+
f
j
(x) = .
Therefore A = x X [ lim
j+
g
j
(x) exists in C and, applying the result
of (b) to the functions g
j
: X C, we get that A . For the same reason, the
function lim
j+
f
j
, which on A is equal to lim
j+
g
j
, is
A
measurable.
6.10 Simple functions.
Denition 6.8 Let E X. The function
E
: X R dened by
E
(x) =
_
1, if x E,
0, if x / E,
is called the characteristic function of E.
Observe that, not only E determines its
E
, but also
E
determines the set E
by E = x X[
E
(x) = 1 =
1
E
(1).
The following are trivial:
E
+
F
=
E\F
+(+)
EF
+
F\E

E
F
=
EF

E
c = 1
E
for all E, F X and all , C.
Proposition 6.21 Let (X, ) be a measurable space and E X. Then
E
is
measurable if and only if E .
Proof: If
E
is measurable, then E =
1
E
(1) .
Conversely, let E . Then for an arbitrary F B
R
or B
C
we have
1
E
(F) = if 0, 1 / F,
1
E
(F) = E if 1 F and 0 / F,
1
E
(F) = E
c
if 1 / F
and 0 F and
1
E
(F) = X if 0, 1 F. In any case,
1
E
(F) and
E
is
measurable.
Denition 6.9 A function dened on a non-empty set X is called a simple
function on X if its range is a nite subset of C.
The following proposition completely describes the structure of simple func-
tions.
Proposition 6.22 (i) A function : X C is a simple function on X if
and only if it is a linear combination with complex coecients of characteristic
functions of subsets of X.
(ii) For every simple function on X there are m N, dierent
1
, . . . ,
m
C
and non-empty pairwise disjoint E
1
, . . . , E
m
X with
m
j=1
E
j
= X so that
=
1
E1
+ +
m
Em
.
This representation of is unique (apart from rearrangement).
(iii) If is a -algebra of subsets of X, then is measurable if and only if
all E
k
s in the representation of described in (ii) belong to .
Proof: Let =
n
j=1
Fj
, where
j
C and F
j
X for all j = 1, . . . , n.
Taking an arbitrary x X, either x belongs to no F
j
, in which case (x) = 0,
or, by considering all the sets F
j1
, . . . , F
j
k
which contain x, we have that (x) =
j1
+ +
j
k
. Therefore the range of contains at most all the possible sums
j1
+ +
j
k
together with 0 and hence it is nite. Thus is simple on X.
Conversely, suppose is simple on X and let its range consist of the dierent
1
, . . . ,
m
C. We consider E
j
= x X[ (x) =
j
=
1
(
j
). Then
every x X belongs to exactly one of these sets, so that they are pairwise
disjoint and X = E
1
E
m
. Now it is clear that =
m
j=1
Ej
, because
both sides take the same value at every x.
If =
i=1
i
is another representation of with dierent
i
s and non-
empty pairwise disjoint E
i
s covering X, then the range of is exactly the set
1
, . . . ,
m
. Hence m
= m and, after rearrangement,
1
=
1
, . . . ,
m
=
m
.
Therefore E
j
=
1
(
j
) =
1
(
j
) = E
j
for all j = 1, . . . , m. We conclude
that the representation is unique.
Now if all E
j
s belong to the -algebra , then, by Proposition 6.21, all
Ej
s are measurable and hence is also measurable. Conversely, if is
measurable, then all E
j
=
1
(
j
) belong to .
Denition 6.10 The unique representation of the simple function , which is
described in part (ii) of Proposition 6.22, is called the standard representa-
tion of .
If one of the coecients in the standard representation of a simple function is
equal to 0, then we usually omit the corresponding term from the sum (but then
the union of the pairwise disjoint sets which appear in the representation is not,
necessarily, equal to the whole space).
Proposition 6.23 Any linear combination with complex coecients of simple
functions is a simple function and any product of simple functions is a simple
function. Also, the maximum and the minimum of real valued simple functions
are simple functions.
Proof: Let , be simple functions on X and p, q C. Assume that
1
, . . . ,
n
are the values of and
1
, . . . ,
m
are the values of . It is obvious that the
possible values of p + q are among the nm numbers p
i
+ q
j
and that the
possible values of are among the nm numbers
i
j
. Therefore both functions
p + q, have a nite number of values. If , are real valued, then the
possible values of max(, ) and min(, ) are among the n+m numbers
i
,
j
.
6.10. SIMPLE FUNCTIONS. 83
Theorem 6.1 (i) Given f : X [0, +], there exists an increasing sequence
n
of non-negative simple functions on X which converges to f pointwise on
X. Moreover, it converges to f uniformly on every subset on which f is bounded.
(ii) Given f : X C, there is a sequence
n
of simple functions on X which
converges to f pointwise on X and so that [
n
[ is increasing. Moreover,
n
converges to f uniformly on every subset on which f is bounded.

If is a -algebra of subsets of X and f is measurable, then the
n
in
(i) and (ii) can be taken to be measurable.
Proof: (i) For every n, k N with 1 k 2
2n
, we dene the sets
E
k
n
= f
1
__
k 1
2
n
,
k
2
n
_
, F
n
= f
1
((2
n
, +])
and the simple function
n
=
2
2n
k=1
k 1
2
n

E
k
n
+ 2
n
Fn
.
For each n the sets E
1
n
, . . . , E
2
2n
n
, F
n
are pairwise disjoint and their union is
the set f
1
((0, +]), while their complementary set is G = f
1
(0). Observe
that if f is measurable then all E
k
n
and F
n
belong to and hence
n
is
measurable.
In G we have 0 =
n
= f, in each E
k
n
we have
n
=
k1
2
n
< f
k
2
n
=
n
+
1
2
n
and in F
n
we have
n
= 2
n
< f.
Now, if f(x) = +, then x F
n
for every n and hence
n
(x) = 2
n
+ = f(x). If 0 f(x) < +, then for all large n we have 0 f(x) 2

n
and hence 0 f(x)
n
(x)
1
2
n
, which implies that
n
(x) f(x). Therefore,
n
f pointwise on X.
If K X and f is bounded on K, then there is an n
0
so that f(x) 2
n0
for
all x K. Hence for all n n
0
we have 0 f(x)
n
(x)
1
2
n
for all x K.
This says that
n
f uniformly on K.
It remains to prove that
n
is increasing. If x G, then
n
(x) =
n+1
(x) = f(x) = 0. Now observe the relations
E
2k1
n+1
E
2k
n+1
= E
k
n
, 1 k 2
2n
,
and
(
2
2(n+1)
l=2
2n+1
+1
E
l
n+1
) F
n+1
= F
n
.
The rst relation implies that, if x E
k
n
then
n
(x) =
k1
2
n
and
n+1
(x) =
(2k1)1
2
n+1
or
2k1
2
n+1
. Therefore, if x E
k
n
, then
n
(x)
n+1
(x).
The second relation implies that, if x F
n
, then
n
(x) = 2
n
and
n+1
(x) =
(2
2n+1
+1)1
2
n+1
or . . . or
2
2(n+1)
1
2
n+1
or 2
n+1
. Hence, if x F
n
, then
n
(x)
n+1
(x).
(ii) Let A = f
1
(C), whence f = on A
c
. Consider the restriction f
A
: A C
and the functions
((f
A
))
+
, ((f
A
))
, ((f
A
))
+
, ((f
A
))
: A [0, +).
If f is measurable, then A and these four functions are
A
measurable.
By the result of part (i) there are increasing sequences p
n
, q
n
, r
n
and
s
n
of non-negative (real valued) simple functions on A so that each converges
to, respectively, ((f
A
))
+
, ((f
A
))
, ((f
A
))
+
and ((f
A
))
pointwise on A
and uniformly on every subset of A on which f
A
is bounded (because on such
a subset all four functions are also bounded). Now it is obvious that, if we
set
n
= (p
n
q
n
) + i(r
n
s
n
), then
n
is a simple function on A which is
A
measurable if f is measurable. It is clear that
n
f
A
pointwise on A
and uniformly on every subset of A on which f
A
is bounded.
Also [
n
[ =
_
(p
n
q
n
)
2
+ (r
n
s
n
)
2
=
_
p
2
n
+q
2
n
+r
2
n
+s
2
n
and thus the
sequence [
n
[ is increasing on A.
If we dene
n
as the constant n on A
c
, then the proof is complete.
6.11 The role of null sets.
Denition 6.11 Let (X, , ) be a measure space. We say that a property P(x)
holds -almost everywhere on X or for -almost every x X, if the set
x X[ P(x) is not true is included in a -null set.
We also use the short expressions: P(x) holds -a.e. on X and P(x) holds
for -a.e. x X.
It is obvious that if P(x) holds for -a.e. x X and is complete then
the set x X[ P(x) is not true is contained in and hence its complement
x X[ P(x) is true is also in .
Proposition 6.24 Let (X, , ) be a measure space and (X, , ) be its comple-
tion. Let (Y,
) be a measurable space and f : X Y be (,
)measurable.
If g : X Y is equal to f -a.e on X, then g is (,
)measurable.
Proof: There exists N so that x X[ f(x) ,= g(x) N and (N) = 0.
Take an arbitrary E
and write g
1
(E) = x X[ g(x) E = x
N
c
[ g(x) Ex N [ g(x) E = x N
c
[ f(x) Ex N [ g(x) E.
The rst set is = N
c
f
1
(E) and belongs to and the second set is N.
By the deniton of the completion we get that g
1
(E) and hence g is
(,
)measurable.
In the particular case of a complete measure space (X, , ) we have the
rule: if f is (,
)measurable on X and g is equal to f -a.e. on X, then g

is also (,
)measurable on X.
Proposition 6.25 Let (X, , ) be a measure space and (X, , ) be its com-
pletion. Let f
j
be a sequence of measurable functions f
j
: X R or C.
If g : X R or C is such that g(x) = lim
j+
f
j
(x) for -a.e. x X, then g
is measurable.
Proof: x X[ lim
j+
f
j
(x) does not exist or is ,= g(x) N for some
N with (N) = 0.
6.11. THE ROLE OF NULL SETS. 85
N
c
belongs to and the restrictions (f
j
)
N
c are all
N
cmeasurable. By
Proposition 6.20, the restriction g
N
c = lim
j+
(f
j
)
N
c is
N
cmeasurable.
This, of course, means that for every E
we have x N
c
[ g(x) E .
Now we write g
1
(E) = x N
c
[ g(x) E x N [ g(x) E. The
rst set belongs to and the second is N. Therefore g
1
(E) and g is
measurable.
Again, in the particular case of a complete measure space (X, , ) the rule
is: if f
j
is a sequence of measurable functions on X and its limit is equal
to g -a.e. on X, then g is also measurable on X.
Proposition 6.26 Let (X, , ) be a measure space and (X, , ) be its comple-
tion. Let (Y,
) be a measurable space and f : A Y be (

A
,
)measurable,
where A with (A
c
) = 0. If we extend f to X in an arbitrary manner, then
the extended function is (,
)measurable.
Proof: Let h : A
c
Y be an arbitrary function and let
F(x) =
_
f(x), if x A,
h(x), if x A
c
.
Take an arbitrary E
and write F
1
(E) = x A[ f(x) E x
A
c
[ h(x) E = f
1
(E) x A
c
[ h(x) E. The rst set belongs to
A
and hence to , while the second set is A
c
. Therefore F
1
(E) and F is
(,
)measurable.
If (X, , ) is a complete measure space, the rule is: if f is dened -a.e.
on X and it is measurable on its domain of denition, then any extension of f
on X is measurable.
6.12 Exercises.
1. Let (X, ) be a measurable space and f : X R. Prove that f is
measurable if f
1
((a, +]) for all rational a R.
2. Let f : X R. If g, h : X R are such that g, h 0 and f = g h on
X, prove that f
+
g and f
h on X.
3. Let (X, ) be a measurable space and f : X R or C be measurable.
We agree that 0
p
= +, (+)
p
= 0 if p < 0 and 0
0
= (+)
0
= 1. Prove
that, for all p R, the function [f[
p
is measurable.
4. Prove that every monotone f : R R is Borel-measurable.
5. Translates and dilates of functions.
Let f : R
n
Y and take arbitrary y R
n
and (0, +). We dene
g, h : R
n
Y by
g(x) = f(x y), h(x) = f
_
x
_
for all x R
n
. g is called the translate of f by y and h is called the
dilate of f by .
Let (Y,
) be a measurable space. Prove that, if f is (L

n
,
)measurable,
then the same is true for g and h.
6. Functions with prescribed level sets.
Let (X, ) be a measurable space and assume that the collection E
R
of subsets of X has the properties:
(i) E
for all , with ,

(ii)
R
E
= X,
R
E
= ,
(iii)
,>
E
= E
for all R.
Consider the function f : X R dened by f(x) = inf R[ x E
.
Prove that f is measurable and that E
= x X[ f(x) for
every R.
How will the result change if we drop any of the assumptions in (ii) and
(iii)?
7. Not all functions are Lebesgue-measurable and not all Lebesgue-measurable
functions are Borel-measurable.
(i) Prove that a Borel-measurable g : R R is also Lebesgue-measurable.
(ii) Find a non-Lebesgue-measurable function f : R R.
(iii) Using exercise 4.6.12, nd and a function g : R R which is
Lebesgue-measurable but not Borel-measurable.
8. Give an example of a non-Lebesgue-measurable f : R R so that [f[ is
Lebesgue-measurable.
6.12. EXERCISES. 87
9. Starting with an appropriate non-Lebesgue-measurable function, give an
example of an uncountable collection f
i
iI
of Lebesgue-measurable func-
tions f
i
: R R so that sup
iI
f
i
10. (i) Prove that, if G : R R is continuous and H : R R is Borel-
measurable, then H G : R R is Borel-measurable.
(ii) Using exercise 4.6.12, construct a continuous G : R R and a
Lebesgue-measurable H : R R so that HG : R R is not Lebesgue-
measurable.
11. Let (X, , ) be a measure space and f : X R or C be measurable.
Assume that (x X[ [f(x)[ = +) = 0 and that there is M < +
so that (x X[ [f(x)[ > M) < +.
Prove that for every > 0 there is a bounded measurable g : X R
or C so that (x X[ g(x) ,= f(x)) < . You may try a suitable
truncation of f.
12. We say that : X C is an elementary function on X if it has count-
ably many values. Is there a standard representation for an elementary
function?
Prove that for any f : X [0, +), there is an increasing sequence
n
of elementary functions on X so that

n
f uniformly on X. If is a
-algebra of subsets of X and f is measurable, prove that the
n
s can
be taken measurable.
13. We can add, multiply and take limits of equalities holding almost every-
where.
Let (X, , ) be a measure space.
(i) Let f, g, h : X Y . If f = g -a.e. on X and g = h -a.e. on X, then
f = h -a.e. on X.
(ii) Let f
1
, f
2
, g
1
, g
2
: X R. If f
1
= f
2
-a.e. on X and g
1
= g
2
-a.e.
on X, then f
1
+g
1
= f
2
+g
2
and f
1
g
1
= f
2
g
2
-a.e. on X.
(iii) Let f
j
, g
j
: X R so that f
j
= g
j
-a.e. on X for all j N. Then
sup
jN
f
j
= sup
jN
g
j
-a.e. on X. Similar results hold for inf, limsup
and liminf.
(iv) Let f
j
, g
j
: X R so that f
j
= g
j
-a.e. on X for all j N. If A =
x X[ lim
j+
f
j
(x) exists and B = x X[ lim
j+
g
j
(x) exists,
then AB N for some N with (N) = 0 and lim
j+
f
j
=
lim
j+
g
j
-a.e. on A B. If, moreover, we extend both lim
j+
f
j
and lim
j+
g
j
by a common function h on (AB)
c
, then lim
j+
f
j
=
lim
j+
g
j
-a.e. on X.
14. Let (X, , ) be a measure space and (X, , ) be its completion.
(i) If E , then there is A so that
E
=
A
-a.e. on X.
(ii) If : X C is a measurable simple function, then there is a
measurable simple function : X C so that = -a.e. on X.
(iii) Use Theorem 6.1 to prove that, if g : X R or C is measurable,
then there is a measurable f : X R or C so that g = f -a.e. on X.
15. Let X, Y be topological spaces of which Y is Hausdor. This means that,
if y
1
, y
2
Y and y
1
,= y
2
, then there are disjoint open neighborhoods
V
y1
, V
y2
of y
1
, y
2
, respectively. Assume that is a Borel measure on X so
that (U) > 0 for every open U X. Prove that, if f, g : X Y are
continuous and f = g -a.e. on X, then f = g on X.
16. The support of a function.
(a) Let X be a topological space and a continuous f : X C. The set
supp(f) = f
1
(C 0) is called the support of f. Prove that supp(f)
is the smallest closed subset of X outside of which f = 0.
(b) Let be a regular Borel measure on the topological space X and
f : X C be a Borel-measurable function. A point x X is called
a support point for f if (y U
x
[ f(y) ,= 0) > 0 for every open
neighborhood U
x
of x. The set
supp(f) = x X[ x is a support point for f
is called the support of f.
(i) Prove that supp(f) is a closed set in X.
(ii) Prove that (x K[ f(x) ,= 0) = 0 for all compact sets K
(supp(f))
c
.
(iii) Using the regularity of , prove that f = 0 -a.e on (supp(f))
c
.
(iv) Prove that (supp(f))
c
is the largest open set in X on which f = 0
-a.e.
(c) Assume that the appearing in (b) has the additional property that
(U) > 0 for every open U X. Use exercise 6.12.15 to prove that for
any continuous f : X C the two denitions of supp(f) (the one in (a)
and the one in (b)) coincide.
17. The restriction of a -algebra.
(a) Let be a -algebra of subsets of X and A X.
(i) We dene
A
= E A[ E to be the restriction of on A.
Prove that
A
is a -algebra of subsets of A and that, in case A , this
denition coincides with the Denition 6.2.
(ii) Now let c be a collection of subsets of X and let c
A
= EA[ E c.
Prove that, if = (c), then
A
= (c
A
).
(b) Let X be a topological space and A X. Consider A equipped with
the relative topology - namely, a set is open in A if and only if it is the
intersection of some open set in X with A. Prove that B
A
= (B
X
)
A
.
18. The Theorem of Lusin.
We shall prove that every Lebesgue-measurable function which is nite
m
n
-a.e. on R
n
is equal to a continuous function except on a set of arbi-
trarily small m
n
-measure.
6.12. EXERCISES. 89
(i) For each a < a + < b < b we consider the function
a,b,
: R R
which: is 0 outside (a, b), is 1 on [a+, b] and is linear on [a, a+] and on
[b, b] so that it is continuous on R. Now, let R = (a
1
, b
1
) (a
n
, b
n
)
and, for small enough > 0, we consider the function
R,
: R
n
R by
the formula
R,
(x
1
, . . . , x
n
) =
a1,b1,
(x
1
)
an,bn,
(x
n
).
If R
= (a
1
+ , b
1
) (a
n
+ , b
n
), prove that
R,
= 1 on
R
,
R,
= 0 outside R, 0
R,
1 on R
n
and
R,
is continuous
on R
n
. Therefore, prove that for every > 0 there is > 0 so that
m
n
(x R
n
[
R,
(x) ,=
R
(x)) < .
(ii) Let E L
n
with m
n
(E) < +. Use Theorem 4.6 to prove that for
every > 0 there is a continuous : R
n
R so that 0 1 on R
n
and m
n
(x R
n
[ (x) ,=
E
(x)) < .
(iii) Let be a non-negative Lebesgue-measurable simple function on R
n
which is 0 outside some set of nite m
n
-measure. Prove that for all > 0
there is a continuous : R
n
R so that 0 max
R
n on R
n
and
m
n
(x R
n
[ (x) ,= (x)) < .
(iv) Let f : R
n
[0, +] be a bounded Lebesgue-measurable function
which is 0 outside some set of nite m
n
-measure. Use Theorem 6.1 to
prove that f =
+
k=1
k
uniformly on R
n
, where all
k
are Lebesgue-
measurable simple functions with 0
k

1
2
k
on R
n
for all k. Now
apply the result of (iii) to each
k
and prove that for all > 0 there
is a continuous g : R
n
R so that 0 g max
R
n f on R
n
and
m
n
(x R
n
[ g(x) ,= f(x)) < .
(v) Let f : R
n
[0, +] be a Lebesgue-measurable function which is 0
outside some set of nite m
n
-measure and nite m
n
-a.e. on R
n
. By taking
an appropriate truncation of f prove that for all > 0 there is a bounded
Lebesgue-measurable function h : R
n
[0, +] which is 0 outside some
set of nite m
n
-measure so that m
n
(x R
n
[ h(x) ,= f(x)) < . Now
apply the result of (iv) to nd a continuous g : R
n
R so that m
n
(x
R
n
[ g(x) ,= f(x)) < .
(vi) Find pairwise disjoint open-closed qubes P
k
so that R
n
=
+
k=1
P
k
and
let R
k
be the open qube with the same edges as P
k
. Consider for each k
a small enough
k
> 0 so that m
n
(x R
n
[
R
k
,
k
(x) ,=
R
k (x)) <

2
k+1
.
(vii) Let f : R
n
[0, +] be Lebesgue-measurable and nite m
n
-a.e.
on R
n
. If R
k
are the qubes from (vi), then each f
R
k : R
n
[0, +]
is Lebesgue-measurable, nite m
n
-a.e. on R
n
and 0 outside R
k
. Ap-
ply (v) to nd continuous g
k
: R
n
R so that m
n
(x R
n
[ g
k
(x) ,=
f(x)
R
k (x)) <

2
k+1
.
Prove that m
n
(x R
n
[
R
k
,
k
(x)g
k
(x) ,= f(x)
R
k (x)) <

2
k
.
Dene g =
+
k=1
R
k
,
k
g
k
and prove that g is continuous on R
n
and that
m
n
(x R
n
[ g(x) ,= f(x)) < .
(viii) Extend the result of (vii) to all f : R
n
Ror Cwhich are Lebesgue-
measurable and nite m
n
-a.e. on R
n
.
19. Let f : R
n
R be continuous at m
n
-a.e. x R
n
. Prove that f is
Lebesgue-measurable on R
n
.
Chapter 7
Integrals
7.1 Integrals of non-negative simple functions.
In this whole section (X, , ) will be a xed measure space.
Denition 7.1 Let : X [0, +) be a non-negative -measurable simple
function. If =
m
k=1
E
k
is the standard representation of , we dene
_
X
d =
m
k=1
k
(E
k
)
and call it the integral of over X with respect to or, shortly, the
integral of .
We can make the following observations.
(i) If one of the values
k
of is equal to 0, then, even if the corresponding set
E
k
has innite -measure, the product
k
(E
k
) is equal to 0. In other words,
the set where = 0 does not matter for the calculation of the integral of .
(ii) We also see that
_
X
d < + if and only if (E
k
) < + for all k for
which
k
> 0. Taking the union of all these E
k
s we see that
_
X
d < + if
and only if (x X[ (x) > 0) < +. In other words, has a nite integral
if and only if = 0 outside a set of nite -measure.
(iii) Moreover,
_
X
d = 0 if and only if (E
k
) = 0 for all k for which
k
> 0.
Taking, as before, the union of these E
k
s we see that
_
X
d < + if and
only if (x X [ (x) > 0) = 0. In other words, has vanishing integral if
and only if = 0 outside a -null set.
Lemma 7.1 Let =
n
j=1
Fj
, where 0
j
< + for all j and the sets
F
j
are pairwise disjoint. Then
_
X
d =
n
j=1
j
(F
j
).
The representation =
n
j=1
Fj
in the statement may not be the standard
representation of . In fact, the
j
s are not assumed dierent and it is not
assumed either that the F
j
s are non-empty or that they cover X.
91
92 CHAPTER 7. INTEGRALS
Proof: (a) In case all F
j
s are empty, then their characteristic functions are 0
on X and we get = 0 = 0
X
as the standard representation of . Therefore
_
X
d = 0 (X) = 0 =
n
j=1
j
(F
j
), since all measures are 0. In this
particular case the result of the lemma is proved.
(b) In case some, but not all, of the F
j
s are empty, we rearrange so that
F
1
, . . . , F
l
,= and F
l+1
, . . . , F
n
= . (We include the case l = n.) Then we
have =
l
j=1
Fj
, where all F
j
s are non-empty, and the equality to be
proved becomes
_
X
d =
l
j=1
j
(F
j
).
In case the F
j
s do not cover X we introduce the non-empty set F
l+1
=
(F
1
F
l
)
c
and the value
l+1
= 0. We can then write =
l+1
j=1
Fj
for
the assumed equality and
_
X
d =
l+1
j=1
j
(F
j
) for the one to be proved.
In any case, using the symbol k for l or l + 1 we have to prove that, if
=
k
j=1
Fj
, where all F
j
are non-empty, pairwise disjoint and cover
X, then
_
X
d =
k
j=1
j
(F
j
).
It is clear that
1
, . . . ,
k
are all the values of on X, perhaps with repeti-
tions. We rearrange in groups, so that
1
= =
k1
=
1
,
k1+1
= =
k1+k2
=
2
,
. . .
k1++km1+1
= =
k1++km
=
m
are the dierent values of on X (and, of course, k
1
+ +k
m
= k). For every
i = 1, . . . , m we dene E
i
=
k1++ki
j=k1++ki1+1
F
j
= x X[ (x) =
i
, and
then
=
m
i=1
Ei
is the standard representation of . By denition
_
X
d =
m
i=1
i
(E
i
) =
m
i=1
i
_
k1++ki
j=k1++ki1+1
(F
j
)
_
=
m
i=1
_
k1++ki
j=k1++ki1+1
j
(F
j
)
_
=
k
j=1
j
(F
j
).
Lemma 7.2 If , are non-negative measurable simple functions and 0
< +, then
_
X
( +) d =
_
X
d +
_
X
d and
_
X
d =
_
X
d.
Proof: (a) If = 0, then = 0 = 0
X
is the standard representation of
and hence
_
X
d = 0 (X) = 0 =
_
X
d.
Now let 0 < < +. If =
m
j=1
Ej
is the standard representation
of , then =
m
j=1
Ej
is the standard representation of . Hence
_
X
d =
m
j=1
j
(E
j
) =
m
j=1
j
(E
j
) =
_
X
d.
7.1. INTEGRALS OF NON-NEGATIVE SIMPLE FUNCTIONS. 93
(b) Let =
m
j=1
Ej
and =
n
i=1
Fi
be the standard representations
of and . It is trivial to see that X =
1jm,1in
(E
j
F
i
) and that the
sets E
j
F
i
are pairwise disjoint. It is also clear that + is constant
j
+
i
on each E
j
F
i
and thus
+ =
1jm,1in
(
j
+
i
)
EjFi
.
Lemma 7.1 implies that
_
X
( +) d =
1jm,1in
(
j
+
i
)(E
j
F
i
)
=
1jm,1in
j
(E
j
F
i
) +
1jm,1in
i
(E
j
F
i
)
=
m
j=1
j
_
n
i=1
(E
j
F
i
)
_
+
n
i=1
i
_
m
j=1
(E
j
F
i
)
_
=
m
j=1
j
(E
j
) +
n
i=1
i
(F
i
) =
_
X
d +
_
X
d.
Lemma 7.3 If , are non-negative measurable simple functions so that
on X, then
_
X
d
_
X
d.
Proof: Let =
m
j=1
Ej
and =
n
i=1
Fi
be the standard representa-
tions of and . Whenever E
j
F
i
,= , we take any x E
j
F
i
and nd
j
= (x) (x) =
i
. Therefore, since in the calculation below only the
non-empty intersections really matter,
_
X
d =
m
j=1
j
(E
j
) =
1jm,1in
j
(E
j
F
i
)
1jm,1in
i
(E
j
F
i
) =
n
i=1
i
(F
i
) =
_
X
d.
Lemma 7.4 Let be a non-negative measurable simple function and A
n
an increasing sequence in with

+
n=1
A
n
= X. Then
_
X

An
d
_
X
d.
Proof: Let =
m
j=1
Ej
be the standard representation of . Then
An
=
m
j=1
Ej
An
=
m
j=1
EjAn
. Lemma 7.1 implies that
_
X

An
d =
m
j=1
j
(E
j
A
n
).
For each j we see that (E
j
A
n
) (E
j
) by the continuity of from
below. Therefore
_
X

An
d
m
j=1
j
(E
j
) =
_
X
d.
Lemma 7.5 Let ,
1
,
2
, . . . be non-negative measurable simple functions
so that
n

n+1
on X for all n.
(i) If lim
n+
n
on X, then lim
n+
_
X

n
d
_
X
d.
(ii) If lim
n+
n
on X, then
_
X
d lim
n+
_
X

n
d.
Proof: Lemma 7.3 implies that
_
X

n
d
_
X

n+1
d for all n and hence the
limit lim
n+
_
X

n
d exists in [0, +].
(i) Since, by Lemma 7.3,
_
X

n
d
_
X
d, we get lim
n+
_
X

n
d
_
X
d.
(ii) Consider arbitrary [0, 1) and dene A
n
= x X[ (x)
n
(x) .
It is easy to see that A
n
is increasing and that
+
n=1
A
n
= X. Indeed, if there is
any x /
+
n=1
A
n
, then
n
(x) < (x) for all n, implying that 0 < (x) (x)
which cannot be true.
Now we have that
An

n
on X. Lemmas 7.2, 7.3 and 7.4 imply that
_
X
d =
_
X
d
= lim
n+
_
X
An
d lim
n+
_
X
n
d.
We now take the limit as 1 and get
_
X
d lim
n+
_
X

n
d.
Lemma 7.6 If
n
and
n
are two increasing sequences of non-negative
measurable simple functions and if lim
n+
n
= lim
n+
n
on X, then
lim
n+
_
X

n
d = lim
n+
_
X

n
d.
Proof: For every k we have that
k
lim
n+
n
on X. Lemma 7.5 im-
plies that
_
X

k
d lim
n+
_
X

n
d. Taking the limit in k, we nd that
lim
n+
_
X

n
d lim
n+
_
X

n
d.
The opposite inequality is proved symmetrically.
7.2 Integrals of non-negative functions.
Again in this section, (X, , ) will be a xed measure space.
Denition 7.2 Let f : X [0, +] be a measurable function. We dene
the integral of f over X with respect to or, shortly, the integral of f
by
_
X
f d = lim
n+
_
X
n
d,
where
n
is any increasing sequence of non-negative measurable simple
functions on X such that lim
n+
n
= f on X.
Lemma 7.6 guarantees that
_
X
f d is well dened and Theorem 6.1 implies the
existence of at least one
n
as in the denition.
Proposition 7.1 Let f, g : X [0, +] be measurable functions and let
[0, +). Then
_
X
(f +g) d =
_
X
f d+
_
X
g d and
_
X
f d =
_
X
f d.
Proof: Consider arbitrary increasing sequences
n
and
n
of non-negative
measurable simple functions on X with lim
n+
n
= f, lim
n+
n
= g
on X. Then
n
+
n
is an increasing sequence of non-negative measurable
7.2. INTEGRALS OF NON-NEGATIVE FUNCTIONS. 95
simple functions with lim
n+
(
n
+
n
) = f + g on X. By denition and
Lemma 7.2,
_
X
(f + g) d = lim
n+
_
X
(
n
+
n
) d = lim
n+
_
X

n
d +
lim
n+
_
X

n
d =
_
X
f d +
_
X
g d.
Also,
n
is an increasing sequence of non-negative measurable simple
functions on X such that lim
n+
n
= f on X. Lemma 7.2 implies again
that
_
X
f d = lim
n+
_
X

n
d = lim
n+
_
X

n
d =
_
X
f d.
Proposition 7.2 Let f, g : X [0, +] be measurable functions such that
f g on X. Then
_
X
f d
_
X
g d.
Proof: Consider arbitrary increasing sequences
n
and
n
of non-negative
measurable simple functions with lim
n+
n
= f, lim
n+
n
= g on X.
Then for every k we have that
k
f g = lim
n+
n
on X. Lemma 7.5
implies that
_
X

k
d lim
n+
_
X

n
d =
_
X
g d. Taking the limit in k we
conclude that
_
X
f d
_
X
g d.
Proposition 7.3 Let f, g : X [0, +] be measurable functions on X.
(i)
_
X
f d = 0 if and only if f = 0 -a.e. on X.
(ii) If f = g -a.e. on X, then
_
X
f d =
_
X
g d.
Proof: (i) Suppose that
_
X
f d = 0. Dene A
n
= x X[
1
n
f(x) =
f
1
([
1
n
, +]) for every n N. Then
1
n
An
f on X and Proposition 7.2 says
that
1
n
(A
n
) =
_
X
1
n
An
d
_
X
f d = 0. Thus (A
n
) = 0 for all n and, since
x X[ f(x) ,= 0 =
+
n=1
A
n
, we nd that (x X[ f(x) ,= 0) = 0.
Conversely, let f = 0 -a.e. on X. Consider an arbitrary increasing sequence
n
of non-negative measurable simple functions with lim
n+
n
= f on
X. Clearly,
n
= 0 -a.e. on X for all n. Observation (iii) after Denition 7.1
says that
_
X

n
d = 0 for all n. Hence
_
X
f d = lim
n+
_
X

n
d = 0.
(ii) Consider A = x X[ f(x) = g(x) . Then there is some B so that
A
c
B and (B) = 0. Dene D = B
c
A. Then f
D
, g
D
are measurable
and f
D
= g
D
on X. Also, f
B
= 0 -a.e. on X and g
B
= 0 -a.e. on X.
By part (i), we have that
_
X
f
B
d =
_
X
g
B
d = 0 and then Proposi-
tion 7.1 implies
_
X
f d =
_
X
(f
D
+ f
B
) d =
_
X
f
D
d =
_
X
g
D
d =
_
X
(g
D
+g
B
) d =
_
X
g d.
The next three theorems, together with Theorems 7.9 and 7.10 in the next
section, are the most important results of integration theory.
Theorem 7.1 (The Monotone Convergence Theorem) (Lebesgue, Levi)
Let f, f
n
: X [0, +] (n N) be measurable functions on X so that
f
n
f
n+1
-a.e. on X and lim
n+
f
n
= f -a.e. on X. Then
lim
n+
_
X
f
n
d =
_
X
f d.
Proof: (a) Assume that f
n
f
n+1
on X and lim
n+
f
n
= f on X.
Proposition 7.2 implies that
_
X
f
n
d
_
X
f
n+1
d
_
X
f d for all n and
hence the lim
n+
_
X
f
n
d exists and it is
_
X
f d.
(i) Take an arbitrary increasing sequence
n
of non-negative measurable
simple functions so that lim
n+
n
= f on X. Then for every k we have
k
f = lim
n+
f
n
. We now take an arbitrary [0, 1) and dene the
set A
n
= x X[
k
(x) f
n
(x) . It is clear that A
n
is increasing
and X =
+
n=1
A
n
. It is also true that
k
An
f
n
on X and, using Lemma
7.5,
_
X

k
d =
_
X

k
d = lim
n+
_
X

k
An
d lim
n+
_
X
f
n
d.
Taking limit as 1, we nd
_
X

k
d lim
n+
_
X
f
n
d. Finally, taking
limit in k, we conclude that
_
X
f d lim
n+
_
X
f
n
d and the proof has
nished.
(ii) If we want to avoid the use of Lemma 7.5, here is an alternative proof of the
inequality
_
X
f d lim
n+
_
X
f
n
d.
For each k take an increasing sequence
k
n
of non-negative measurable
simple functions so that lim
n+
k
n
= f
k
on X. Next, dene the non-negative
measurable simple functions
n
= max(
1
n
, . . . ,
n
n
).
It easy to see that
n
is increasing, that
n
f
n
f on X and that
n
f on X. For the last one, take any x X and any t < f(x). Find k so that
t < f
k
(x) and, then, a large n k so that t <
k
n
(x). Then t <
n
(x) f(x)
and this means that
n
(x) f(x).
Thus
_
X
f d = lim
n+
_
X

n
d lim
n+
_
X
f
n
d.
(b) In the general case, Theorem 2.2 implies that there is some A with
(A
c
) = 0 so that f
n
f
n+1
on A for all n and lim
n+
f
n
= f on A. These
imply that f
n
A
f
n+1
A
on X for all n and lim
n+
f
n
A
= f
A
on X.
From part (a) we have that lim
n+
_
X
f
n
A
d =
_
X
f
A
d.
Since f = f
A
-a.e. on X and f
n
= f
n
A
-a.e. on X, Proposition 7.3
implies that
_
X
f d =
_
X
f
A
d and
_
X
f
n
d =
_
X
f
n
A
d for all n. Hence,
lim
n+
_
X
f
n
d = lim
n+
_
X
f
n
A
d =
_
X
f
A
d =
_
X
f d.
Theorem 7.2 Let f, f
n
: X [0, +] (n N) be measurable on X so that
+
n=1
f
n
= f -a.e. on X. Then
+
n=1
_
X
f
n
d =
_
X
f d.
Proof: We write g
n
= f
1
+ + f
n
for each n. g
n
is an increasing sequence
of non-negative measurable functions with g
n
f -a.e. on X. Proposition
7.1 and Theorem 7.1 imply that
n
k=1
_
X
f
k
d =
_
X
g
n
d
_
X
f d.
Theorem 7.3 (The Lemma of Fatou) Let f, f
n
: X [0, +] (n N) be
measurable. If f = liminf
n+
f
n
-a.e. on X, then
_
X
f d liminf
n+
_
X
f
n
d.
Proof: We dene g
n
= inf
kn
f
k
. Then each g
n
: X [0, +] is measura-
ble, the sequence g
n
is increasing and g
n
f
n
on X for all n. By hypothesis,
f = lim
n+
g
n
-a.e. on X. Proposition 7.2 and Theorem 7.1 imply that
_
X
f d = lim
n+
_
X
g
n
d liminf
n+
_
X
f
n
d.
7.3. INTEGRALS OF COMPLEX VALUED FUNCTIONS. 97
7.3 Integrals of complex valued functions.
Let (X, , ) be a xed measure space.
Denition 7.3 Let f : X R be a measurable function and consider its
positive and negative parts f
+
, f
: X [0, +]. If at least one of

_
X
f
+
d
and
_
X
f
d is < +, we dene
_
X
f d =
_
X
f
+
d
_
X
f
d
and call it the integral of f over X with respect to or, simply, the
integral of f.
We say that f is integrable on X with respect to or, simply, integrable
if
_
X
f d is nite.
Lemma 7.7 Let f : X R be a measurable function. Then the following
are equivalent:
(i) f is integrable
(ii)
_
X
f
+
d < + and
_
X
f
d < +
(iii)
_
X
[f[ d < +.
Proof: The equivalence of (i) and (ii) is clear from the denition.
We know that [f[ = f
+
+ f
and, hence, f
+
, f
[f[ on X. Therefore,
_
X
[f[ d =
_
X
f
+
d +
_
X
f
d and
_
X
f
+
d,
_
X
f
d
_
X
[f[ d. The
equivalence of (ii) and (iii) is now obvious.
Proposition 7.4 Let f : X R be a measurable function. If f is inte-
grable, then
(i) f(x) R for -a.e. x X and
(ii) the set x X[ f(x) ,= 0 is of -nite -measure.
Proof: (i) Let f be integrable. Lemma 7.7 implies
_
X
[f[ d < +. Consider
the set B = x X[ [f(x)[ = + . For every r (0, +) we have that
r
B
[f[ on X and hence r(B) =
_
X
r
B
d
_
X
[f[ d < +. This implies
that (B)
1
r
_
X
[f[ d and, taking the limit as r +, we nd (B) = 0.
(ii) Consider the sets A = x X[ f(x) ,= 0 and A
n
= x X[ [f(x)[
1
n
.
From
1
n
An
[f[ on X, we get
1
n
(A
n
) =
_
X
1
n
An
d
_
X
[f[ d < +.
Thus (A
n
) < + for all n and, since A =
+
n=1
A
n
, we conclude that A is of
-nite -measure.
Denition 7.4 Let f : X C be measurable. Then [f[ : X [0, +] is
measurable and we say that f is integrable on X with respect to or,
simply, integrable, if
_
X
[f[ d < +.
Proposition 7.5 Let f : X C be measurable. If f is integrable, then
(i) f(x) C for -a.e. x X and
(ii) the set x X[ f(x) ,= 0 is of -nite -measure.
Proof: Immediate application of Proposition 7.4 to [f[.
Assume now that f : X C is a measurable integrable function. By
Proposition 7.5, the set D
f
= x X[ f(x) C = f
1
(C) has a -null
complement. The function
f
D
f
=
_
f, on D
f
0, on D
c
f
: X C
is measurable and f
D
f
= f -a.e. on X. The advantage of f
D
f
over f
is that f
D
f
is complex valued and, hence, the (f
D
f
), (f
D
f
) : X R
are dened on X. We also have that [(f
D
f
)[ [f
D
f
[ [f[ on X and
similarly [(f
D
f
)[ [f[ on X. Therefore
_
X
[(f
D
f
)[ d
_
X
[f[ d < +,
implying that (f
D
f
) is an integrable real valued function. The same is true for
(f
D
f
) and thus the integrals
_
X
(f
D
f
) d and
_
X
(f
D
f
) d are dened
and they are (nite) real numbers.
Denition 7.5 Let f : X C be a measurable integrable function and let
D
f
= x X[ f(x) C. We dene
_
X
f d =
_
X
(f
D
f
) d +i
_
X
(f
D
f
) d
and call it the integral of f over X with respect to or just the integral
of f.
We shall make a few observations regarding this denition.
(i) The integral of an extended-complex valued function is dened only if the
function is integrable and then the value of its integral is a (nite) complex
number. Observe that the integral of an extended-real valued function is dened
if the function is integrable (and the value of its integral is a nite real number)
and also in certain other cases when the value of its integral can be either +
or .
(ii) We used the function f
D
f
, which changes the value of f to the value
0, simply because we need complex values in order to be able to consider their
real and imaginary parts. We may allow more freedom and see what happens
if we use a function
F =
_
f, on D
f
h, on D
c
f
: X C,
where h is an arbitrary
D
c
f
measurable complex valued function on D
c
f
. It is
clear that F = f
D
f
-a.e. on X and hence (F) = (f
D
f
) -a.e. on X.
Of course, this implies that (F)
+
= (f
D
f
)
+
and (F)
= (f
D
f
)
-
a.e. on X. From Proposition 7.3,
_
X
(F) d =
_
X
(F)
+
d
_
X
(F)
d =
_
X
(f
D
f
)
+
d
_
X
(f
D
f
)
d =
_
X
(f
D
f
) d. Similarly,
_
X
(F) d =
_
X
(f
D
f
) d. Therefore there is no dierence between the possible denition
_
X
f d =
_
X
(F) d+i
_
X
(F) d and the one we have given. Of course, the
function 0 on D
c
f
is the simplest of all choices for h.
(iii) If f : X C is complex valued on X, then D
f
= X and the denition
takes the simpler form
_
X
f d =
_
X
(f) d +i
_
X
(f) d.
We also have
_
_
X
f d
_
=
_
X
(f) d,
_
_
X
f d
_
=
_
X
(f) d.
The next is helpful and we shall make use of it very often.
Lemma 7.8 If f : X C is integrable, there is F : X C so that F = f
-a.e. on X and
_
X
F d =
_
X
f d.
Proof: We take F = f
D
f
, where D
f
= f
1
(C).
Theorem 7.4 Let f, g : X R or C be measurable so that f = g -a.e. on
X and
_
X
f d is dened. Then
_
X
g d is also dened and
_
X
g d =
_
X
f d.
Proof: (a) Let f, g : X R. If f = g -a.e. on X, then f
+
= g
+
-
a.e. on X and f
= g
-a.e. on X. Proposition 7.3 implies that

_
X
f
+
d =
_
X
g
+
d and
_
X
f
d =
_
X
g
d. Now if
_
X
f
+
d or
_
X
f
d is nite, then,
respectively,
_
X
g
+
d or
_
X
g
d is also nite. Therefore

_
X
g d is dened and
_
X
f d =
_
X
g d.
(b) Let f, g : X C and f = g -a.e. on X.
If f is integrable, from [f[ = [g[ -a.e. on X and from Proposition 7.3, we
nd
_
X
[g[ d =
_
X
[f[ d < + and, hence, g is also integrable.
Now, Lemma 7.8 says that there are F, G : X C so that F = f and G = g
-a.e. on X and also
_
X
F d =
_
X
f d and
_
X
Gd =
_
X
g d. From f = g
-a.e. on X we see that F = G -a.e. on X. This implies that (F) = (G)
-a.e. on X and, from (a),
_
X
(F) d =
_
X
(G) d. Similarly,
_
X
(F) d =
_
X
(G) d.
Therefore,
_
X
f d =
_
X
F d =
_
X
(F) d+i
_
X
(F) d =
_
X
(G) d+
i
_
X
(G) d =
_
X
Gd =
_
X
g d.
Theorem 7.5 Let f : X R or C be measurable. Then the following are
equivalent:
(i) f = 0 -a.e. on X
(ii)
_
X
[f[ d = 0
(iii)
_
X
f
A
d = 0 for every A .
Proof: If
_
X
[f[ d = 0, Proposition 7.3 implies that [f[ = 0 and, hence, f = 0
-a.e. on X.
If f = 0 -a.e. on X, then f
A
= 0 -a.e. on X for all A . Theorem 7.4
implies that
_
X
f
A
d = 0.
Finally, let
_
X
f
A
d = 0 for every A .
(a) If f : X Rwe take A = f
1
([0, +]) and nd
_
X
f
+
d =
_
X
f
A
d = 0.
Similarly,
_
X
f
d = 0 and thus
_
X
[f[ d =
_
X
f
+
d +
_
X
f
d = 0.
(b) If f : X C, we rst take A = X and nd
_
X
f d = 0. This says, in
particular, that f is integrable. We take some F : X C so that F = f -a.e.
on X.
For every A we have F
A
= f
A
-a.e. on X and, from Theorem 7.4,
_
X
F
A
d =
_
X
f
A
d = 0. This implies
_
X
(F)
A
d =
_
X
(F
A
) d =
(
_
X
F
A
d) = 0 and, from part (a), (F) = 0 -a.e. on X. Similarly,
(F) = 0 -a.e. on X and thus F = 0 -a.e. on X. We conclude that f = 0
-a.e. on X.
Theorem 7.6 Let f : X R or C be measurable and R or C.
(i) If f : X R, R and
_
X
f d is dened, then
_
X
f d is also dened
and
_
X
f d =
_
X
f d.
(ii) If f is integrable, then f is also integrable and the previous equality is
again true.
Proof: (i) Let f : X R and
_
X
f d be dened and, hence, either
_
X
f
+
d <
+ or
_
X
f
d < +.
If 0 < < +, then (f)
+
= f
+
and (f)
= f
. Therefore, at least
one of
_
X
(f)
+
d =
_
X
f
+
d and
_
X
(f)
d =
_
X
f
d is nite. This
means that
_
X
f d is dened and
_
X
f d =
_
X
(f)
+
d
_
X
(f)
d =
_
_
X
f
+
d
_
X
f
d
_
=
_
X
f d.
If < < 0, then (f)
+
= f
and (f)
= f
+
and the previous
argument can be repeated with no essential change.
If = 0, the result is trivial.
(ii) If f : X Ris integrable and R, then
_
X
[f[ d = [[
_
X
[f[ d < +,
which means that f is also integrable. The equality
_
X
f d =
_
X
f d has
been proved in (i).
If f : X C is integrable and C, the same argument gives that f is
also integrable.
We, now, take F : X C so that F = f -a.e. on X. Then, also F = f
-a.e. on X and Theorem 7.4 implies that
_
X
F d =
_
X
f d and
_
X
F d =
_
X
f d. Hence, it is enough to prove that
_
X
F d =
_
X
F d.
From (F) = ()(F) ()(F) and from the real valued case we get
that
_
X
(F) d = ()
_
X
(F) d ()
_
X
(F) d.
Similarly,
_
X
(F) d = ()
_
X
(F) d +()
_
X
(F) d.
From these two equalities
_
X
F d =
_
X
(F) d +i
_
X
(F) d =
_
X
F d.
Theorem 7.7 Let f, g : X R or C be measurable and consider any
measurable denition of f +g.
(i) If f, g : X R and
_
X
f d,
_
X
g d are both dened and they are not
opposite innities, then
_
X
(f +g) d is also dened and
_
X
(f +g) d =
_
X
f d +
_
X
g d.
(ii) If f, g : X R or C are integrable, then f + g is also integrable and the
previous equality is again true.
Proof: (i) Considering the integrals
_
X
f
+
d,
_
X
f
d,
_
X
g
+
d,
_
X
g
d, the
assumptions imply that at most the
_
X
f
+
d,
_
X
g
+
d are + or at most the
_
X
f
d,
_
X
g
d are +.
Let
_
X
f
d < + and
_
X
g
d < +.
Proposition 7.4 implies that, if B = x X[ f(x) ,= , g(x) ,= ,
then (B
c
) = 0. We dene the functions F = f
B
and G = g
B
. Then
F, G : X (, +] are measurable and F = f and G = g -a.e. on X.
The advantage of F, G over f, g is that F(x) + G(x) is dened for every
x X.
Observe that for all measurable denitions of f +g, we have F +G = f +g
-a.e. on X. Because of Theorem7.4, it is enough to prove that the
_
X
(F+G) d
is dened and that
_
X
(F +G) d =
_
X
F d +
_
X
Gd.
From F = F
+
F
F
+
and G = G
+
G
G
+
on X we get F +G
F
+
+G
+
on X. Hence (F +G)
+
F
+
+G
+
on X and similarly (F +G)
+G
on X.
From (F + G)
+ G
on X we nd
_
X
(F + G)
d
_
X
F
d +
_
X
G
d < +. Therefore,
_
X
(F +G) d is dened.
We now have (F +G)
+
(F +G)
= F +G = (F
+
+G
+
) (F
+G
) or,
equivalently, (F +G)
+
+F
+G
= (F +G)
+F
+
+G
+
.
Proposition 7.1 implies that
_
X
(F+G)
+
d+
_
X
F
d+
_
X
G
d =
_
X
(F+G)
d+
_
X
F
+
d+
_
X
G
+
d.
Because of the niteness of
_
X
(F +G)
d,
_
X
F
d,
_
X
G
d, we get
_
X
(F +G) d =
_
X
(F +G)
+
d
_
X
(F +G)
d
=
_
X
F
+
d +
_
X
G
+
d
_
X
F
d
_
X
G
d
=
_
X
F d +
_
X
Gd.
The proof in the case when
_
X
f
+
d < + and
_
X
g
+
d < + is similar.
(ii) By Lemma 7.8, there are F, G : X C so that F = f and G = g -a.e.
on X. This implies that for all measurable denitions of f + g we have
F + G = f + g -a.e. on X. Now, by Theorem 7.4, it is enough to prove that
F +G is integrable and
_
X
(F + G) d =
_
X
F d +
_
X
Gd.
Now
_
X
[F + G[ d
_
X
[F[ d +
_
X
[G[ d < + and, hence, F + G is
integrable.
By part (i) we have
_
X
(F + G) d =
_
X
(F) d +
_
X
(G) d and the
same equality with the imaginary parts. Combining, we get
_
X
(F + G) d =
_
X
F d +
_
X
Gd.
Theorem 7.8 Let f, g : X R be measurable. If
_
X
f d and
_
X
g d are
both dened and f g on X, then
_
X
f d
_
X
g d.
Proof: From f g = g
+
g
g
+
we get f
+
g
+
. Similarly, g
.
Therefore, if
_
X
g
+
d < +, then
_
X
f
+
d < + and, if
_
X
f
d < +,
then
_
X
g
d < +.
Hence we can subtract the two inequalities
_
X
f
+
d
_
X
g
+
d,
_
X
g
d
_
X
f
d
and nd that
_
X
f d
_
X
g d.
Theorem 7.9 Let f : X R or C be measurable.
(i) If f : X R and
_
X
f d is dened, then
_
X
f d
_
X
[f[ d.
(ii) If f : X C is integrable, then the inequality in (i) is again true.
Proof: (i) We write [
_
X
f d[ = [
_
X
f
+
d
_
X
f
d[
_
X
f
+
d+
_
X
f
d =
_
X
[f[ d.
(ii) Consider F : X C so that F = f -a.e. on X. By Theorem 7.4, it is
enough to prove [
_
X
F d[
_
X
[F[ d.
If
_
X
F d = 0, then the inequality is trivially true. Let 0 ,=
_
X
F d C
and take = sign(
_
X
F d) ,= 0. Then
_
X
F d
=
_
X
F d =
_
X
F d =
_
_
X
F d
_
=
_
X
(F) d
_
X
[(F)[ d
_
X
[F[ d =
_
X
[F[ d.
Theorem 7.10 (The Dominated Convergence Theorem) (Lebesgue) Con-
sider the measurable f, f
n
: X R or C (n N) and g : X [0, +].
Assume that f = lim
n+
f
n
-a.e. on X, that, for all n, [f
n
[ g -a.e. on
X and that
_
X
g d < +. Then all f
n
and f are integrable and
_
X
f
n
d
_
X
f d.
Proof: From the [f
n
[ g -a.e. on X we nd
_
X
[f
n
[ d
_
X
g d < + and
hence f
n
is integrable. Also, from [f
n
[ g -a.e. on X and f = lim
n+
f
n
-a.e. on X, we get that [f[ g -a.e. on X and, for the same reason, f is also
integrable.
We may now take F, F
n
: X R or C so that F = f and F
n
= f
n
-a.e. on
X for all n. We, then, have [F
n
[ g -a.e. on X and F = lim
n+
F
n
-a.e.
on X and it is enough to prove
_
X
F
n
d
_
X
F d.
(i) Let F, F
n
: X R. Since 0 g + F
n
, g F
n
on X, the Lemma of Fatou
implies that
_
X
g d
_
X
F d liminf
n+
_
X
(g F
n
) d
and hence
_
X
g d
_
X
F d
_
X
g d + liminf
n+
_
X
F
n
d.
Since
_
X
g d is nite, we get that
_
X
F d liminf
n+
_
X
F
n
d and
hence
limsup
n+
_
X
F
n
d
_
X
F d liminf
n+
_
X
F
n
d.
This implies
_
X
F
n
d
_
X
F d.
(ii) Let F, F
n
: X C. From [(F
n
)[ [F
n
[ g -a.e. on X and from
(F
n
) (F) -a.e. on X, part (i) implies
_
X
(F
n
) d
_
X
(F) d. Simi-
larly,
_
X
(F
n
) d
_
X
(F) d and, from these two,
_
X
F
n
d
_
X
F d.
Theorem 7.11 (The Series Theorem) Consider the measurable f, f
n
:
X R or C (n N). If
+
n=1
_
X
[f
n
[ d < +, then
(i)
+
n=1
f
n
(x) exists for -a.e. x X,
(ii) if f =
+
n=1
f
n
(x) -a.e. on X, then
_
X
f d =
+
n=1
_
X
f
n
d.
Proof: (i) Dene g =
+
n=1
[f
n
[ : X [0, +] on X. From Theorem 7.2 we
have
_
X
g d =
+
n=1
_
X
[f
n
[ d < +. This implies that g < + -a.e. on
X, which means that the series
+
n=1
f
n
(x) converges absolutely, and hence
converges, for -a.e. x X.
(ii) Consider s
n
=
n
k=1
f
k
for all n. Then lim
n+
s
n
= f -a.e. on X
and [s
n
[ g on X. Theorem 7.10 implies that
n
k=1
_
X
f
k
d =
_
X
s
n
d
_
X
f d.
Theorem 7.12 (Approximation) Let f : X R or C be integrable. Then
for every > 0 there is some measurable simple function : X R or C
so that
_
X
[f [ d < .
Proof: If f : X [0, +] is measurable with
_
X
f d < +, there is
an increasing sequence
n
of non-negative measurable simple functions
so that
n
f on X and
_
X

n
d
_
X
f d. Therefore, for some n we have
_
X
f d <
_
X

n
d
_
X
f d. Thus
_
X
[f
n
[ d =
_
X
(f
n
) d < .
Now if f : X R is integrable, then
_
X
f
+
d < + and
_
X
f
d < +.
From the rst case considered, there are non-negative measurable simple
functions
1
,
2
so that
_
X
[f
+

1
[ d <

2
and
_
X
[f

2
[ d <

2
. We
dene the simple function =
1

2
: X R and get
_
X
[f [ d
_
X
[f
+
1
[ d +
_
X
[f
2
[ d < .
Finally, let f : X C be integrable. Then there is F : X C so that
F = f -a.e. on X. The functions (F), (F) : X R are both integrable,
and hence we can nd real valued measurable simple functions
1
,
2
so that
_
X
[(F)
1
[ d <

2
and
_
X
[(F)
2
[ d <

2
. We dene =
1
+i
2
and
get
_
X
[f [ d =
_
X
[F [ d < .
7.4 Integrals over subsets.
Let A and f : X R or C be measurable. In order to dene an
integral of f over A we have two natural choices. One way is to take f
A
, which
is f in A and 0 outside A, and consider
_
X
f
A
d. Another way is to take the
restriction f
A
of f on A and consider
_
A
f
A
d with respect to the restricted
on (A,
A
). The following lemma says that the two procedures are equivalent
and give the same results.
Lemma 7.9 Let A and f : X R or C be measurable.
(i) If f : X R and either
_
X
f
A
d or
_
A
f
A
d exists, then the other also
exists and they are equal.
(ii) If f : X C and either
_
X
[f
A
[ d or
_
A
[f
A
[ d is nite, then the other
is also nite and the integrals
_
X
f
A
d and
_
A
f
A
d are equal.
Proof: (a) Take a non-negative measurable simple function =
m
j=1
Ej
with its standard representation. Now
A
=
m
j=1
EjA
: X [0, +)
has
_
X

A
d =
m
j=1
j
(E
j
A). On the other hand,
A
=
m
j=1
EjA
:
A [0, +) (where we omit the terms for which E
j
A = ) has exactly the
same integral
_
A
A
d =
m
j=1
j
(E
j
A).
(b) Now let f : X [0, +] be measurable. Take an increasing sequence
n
of non-negative measurable simple
n
: X [0, +) with
n
f.
Then
n
A
is increasing and
n
A
f
A
. Also, (
n
)
A
is increasing
and (
n
)
A
f
A
. Hence, by (a) we get,
_
X
f
A
= lim
n+
_
X

n
A
d =
lim
n+
_
A
(
n
)
A
d =
_
A
f
A
d.
(c) If f : X R is measurable, then f
+
A
= (f
A
)
+
and f
A
=
7.4. INTEGRALS OVER SUBSETS. 105
(f
A
)
and also (f
A
)
+
= (f
+
)
A
and (f
A
)
= (f
)
A
. Hence, by (b) we
get
_
X
(f
A
)
+
d =
_
X
f
+
A
d =
_
A
(f
+
)
A
d =
_
X
(f
A
)
+
d and, similarly,
_
X
(f
A
)
d =
_
X
(f
A
)
d. These show (i).

(d) Finally, let f : X C be measurable. Then [f
A
[ = [f[
A
and [f
A
[ =
[f[
A
. By (b) we have
_
X
[f
A
[ d =
_
X
[f[
A
d =
_
A
[f[
A
d =
_
A
[f
A
[ d,
implying that f
A
and f
A
are simultaneously integrable or non-integrable.
Assuming integrability, there is an F : X C so that F = f
A
-a.e. on
X. It is clear that F
A
= f
A
-a.e. on X and, also, F
A
= f
A
-a.e. on A.
Therefore, it is enough to prove that
_
X
F
A
d =
_
A
F
A
d.
Now part (c) implies
_
X
(F
A
) d =
_
X
(F)
A
d =
_
A
(F)
A
d =
_
A
(F
A
) d. Similarly,
_
X
(F
A
) d =
_
A
(F
A
) d and we conclude that
_
X
F
A
d =
_
A
F
A
d.
Denition 7.6 Let f : X R or C be measurable and A .
(i) If f : X R and
_
X
f
A
d or, equivalently,
_
A
f
A
d is dened, we say
that the
_
A
f d is dened and dene
_
A
f d =
_
X
f
A
d =
_
A
f
A
d.
(ii) If f : X C and f
A
is integrable on X or, equivalently, f
A
is integrable
on A, we say that f is integrable on A and dene
_
A
f d exactly as in (i).
Lemma 7.10 Let f : X R or C be measurable.
(i) If f : X R and
_
X
f d is dened, then
_
A
f d is dened for every A .
(ii) If f : X C is integrable then f is integrable on every A .
Proof: (i) We have (f
A
)
+
= f
+
A
f
+
and (f
A
)
= f
A
f
on
X. Therefore, either
_
X
(f
A
)
+
d
_
X
f
+
d < + or
_
X
(f
A
)
d
_
X
f
d < +. This says that

_
X
f
A
d is dened and, hence,
_
A
f d is
also dened.
(ii) If f : X C is integrable, then
_
X
[f
A
[ d
_
X
[f[ d < + and f
A
is
also integrable.
Proposition 7.6 Let f : X R be measurable and
_
X
f d be dened.
Then either
_
A
f d (, +] for all A or
_
A
f d [, +) for all
A .
Proof: Let
_
X
f
d < +. Then
_
X
(f
A
)
d
_
X
f
d < + and hence

_
A
f d =
_
X
f
A
d > for all A .
Similarly, if
_
X
f
+
d < +, then
_
A
f d < + for all A .
Theorem 7.13 If f : X R and
_
X
f d is dened or f : X C and f is
integrable, then
(i)
_
A
f d = 0 for all A with (A) = 0,
(ii)
_
A
f d =
+
n=1
_
An
f d for all pairwise disjoint A
1
, A
2
, . . . with A =
+
n=1
A
n
,
(iii)
_
An
f d
_
A
f d for all increasing A
n
in with A =
+
n=1
A
n
,
(iv)
_
An
f d
_
A
f d for all decreasing A
n
in with A =
+
n=1
A
n
and
[
_
A1
f d[ < +.
Proof: (i) This is easy because f
A
= 0 -a.e. on X.
(ii) Let A
1
, A
2
, . . . be pairwise disjoint and A =
+
n=1
A
n
.
If f : X [0, +] is measurable, since f
A
=
+
n=1
f
An
on X, Theo-
rem 7.2 implies
_
A
f d =
_
X
f
A
d =
+
n=1
_
X
f
An
d =
+
n=1
_
An
f d.
If f : X C and f is integrable, we have by the previous case that
+
n=1
_
X
[f
An
[ d =
+
n=1
_
An
[f[ d =
_
A
[f[ d < +. Because of f
A
=
+
n=1
f
An
on X, Theorem 7.11 implies that
_
A
f d =
+
n=1
_
An
f d.
If f : X R and
_
X
f
d < +, we apply the rst case and get
+
n=1
_
An
f
+
d =
_
A
f
+
d and
+
n=1
_
An
f
d =
_
A
f
d < +. Sub-
tracting, we nd
+
n=1
_
An
f d =
_
A
f d.
If
_
X
f
+
d < +, the proof is similar.
(iii) Write A = A
1
+
k=2
(A
k
A
k1
), where the sets in the union are pairwise
disjoint. Apply (ii) to get
_
A
f d =
_
A1
f d+
+
k=2
_
A
k
\A
k1
f d =
_
A1
f d+
lim
n+
n
k=2
_
A
k
\A
k1
f d = lim
n+
_
An
f d.
(iv) Write A
1
A =
+
n=1
(A
1
A
n
), where A
1
A
n
is increasing. Apply (iii)
to get
_
A1\An
f d
_
A1\A
f d.
From
_
A1\A
f d +
_
A
f d =
_
A1
f d and from [
_
A1
f d[ < + we im-
mediately get that also [
_
A
f d[ < +. From the same equality we then get
_
A1\A
f d =
_
A1
f d
_
A
f d. Similarly,
_
A1\An
f d =
_
A1
f d
_
An
f d
and hence
_
A1
f d
_
An
f d
_
A1
f d
_
A
f d. Because of [
_
A1
f d[ < +
again, we nally have
_
An
f d
_
A
f d.
We must say that all results we have proved about integrals
_
X
over X
hold without change for integrals
_
A
over an arbitrary A . To see this we
either repeat all proofs, making the necessary minor changes, or we just apply
those results to the functions multiplied by
A
or to their restrictions on A. As
an example let us look at the following version of the Dominated Convergence
Theorem.
Assume that f, f
n
: X R or C are measurable, that g : X [0, +]
has
_
A
g d < +, that [f
n
[ g -a.e. on A and f
n
f -a.e. on A. The
result is that
_
A
f
n
d
_
A
f d.
Indeed, we have then that
_
X
g
A
d < +, that [f
n
A
[ g
A
-a.e. on
X and f
n
A
f
A
-a.e. on X. The usual form of the dominated convergence
theorem (for X) implies that
_
A
f
n
d =
_
X
f
n
A
d
_
X
f
A
d =
_
A
f d.
Alternatively, we observe that
_
A
g
A
d < +, that [(f
n
)
A
[ g
A
-a.e. on
A and (f
n
)
A
f
A
-a.e. on A. The dominated convergence theorem (for A)
implies that
_
A
f
n
d =
_
X
(f
n
)
A
d
_
X
f
A
d =
_
A
f d.
7.5 Point-mass distributions.
Consider the point-mass distribution induced by a function a : X [0, +]
through the formula
(E) =
xE
a
x
for all E X.
We observe that all functions f : X Y , no matter what the (Y,
) is, are
(,
)measurable.
If =
n
j=1
Ej
is any non-negative simple function on X with its stan-
dard representation, then
_
X
d =
n
j=1
j
(E
j
) =
n
j=1
j
(
xEj
a
x
) =
n
j=1
(
xEj
j
a
x
) =
n
j=1
(
xEj
(x)a
x
). We apply Proposition 2.6 to get
_
X
d =
xX
(x) a
x
.
Proposition 7.7 If f : X [0, +] then
_
X
f d =
xX
f(x) a
x
.
Proof: Consider an increasing sequence
n
of non-negative simple functions
so that
n
f on X and
_
X

n
d
_
X
f d.
Then
_
X

n
d =
xX

n
(x) a
x

xX
f(x) a
x
and, taking limit in n,
we nd
_
X
f d
xX
f(x) a
x
.
If F is a nite subset of X, then
xF

n
(x) a
x

xX

n
(x) a
x
=
_
X

n
d. Using the obvious lim
n+
xF

n
(x) a
x
=
xF
f(x) a
x
, we nd
xF
f(x) a
x

_
X
f d. Taking supremum over F,
xX
f(x) a
x

_
X
f d
and, combining with the opposite inequality, the proof is nished.
We would like to extend the validity of this Proposition 7.7 to real valued or
complex valued functions, but we do not have a denition for sums of real valued
or complex valued terms! We can give such a denition in a straightforward
manner, but we prefer to use the theory of the integral developed so far.
The amusing thing is that any series
iI
b
i
of non-negative terms over the
general index set I can be written as an integral
iI
b
i
=
_
I
b d ,
where is the counting measure on I (and we freely identify b
i
= b(i)). This is
a simple application of Proposition 7.7.
Using properties of integrals we may prove corresponding properties of sums.
For example, it is true that
iI
(b
i
+c
i
) =
iI
b
i
+
iI
c
i
,
iI
b
i
=
iI
b
i
for every non-negative b
i
, c
i
and . The proof consists in rewriting
_
I
(b+c) d =
_
I
b d +
_
I
c d and
_
I
b d =
_
I
b d in terms of sums.
For every b R we write b
+
= max(b, 0) and b
= min(b, 0) and, clearly,

b = b
+
b
and [b[ = b
+
+b
.
Denition 7.7 If I is any index set and b : I R, we dene the sum of
b
i
iI
over I by
iI
b
i
=
iI
b
+
i

iI
b
i
only when either
iI
b
+
i
< + or
iI
b
i
< +. We say that b
i
iI
is
summable (over I) if
iI
b
i
is nite or, equivalently, if both
iI
b
+
i
and
iI
b
i
are nite.
Since we can write
iI
b
i
=
iI
b
+
i

iI
b
i
=
_
I
b
+
d
_
I
b
d =
_
I
b d
and also
iI
[b
i
[ =
iI
b
+
i
+
iI
b
i
=
_
I
b
+
d +
_
I
b
d =
_
I
[b[ d ,
we may say that b
i
iI
is summable over I if and only if b is integrable over
I with respect to counting measure or, equivalently, if and only if
iI
[b
i
[ =
_
I
[b[ d < +. Also, the
iI
b
i
is dened if and only if the
_
I
b d is dened
and they are equal.
Further exploiting the analogy between sums and integrals we have
Denition 7.8 If I is any index set and b : I C, we say that b
i
iI
is
summable over I if
iI
[b
i
[ < +.
This is the same condition as in the case of b : I R.
Proposition 7.8 Let b : I R or C. Then b
i
iI
is summable over I if and
only if the set i I [ b
i
,= 0 is countable and, taking an arbitrary enumeration
i
1
, i
2
, . . . of it,
+
k=1
[b
i
k
[ < +.
Proof: An application of Propositions 2.3 and 2.4.
In particular, if b
i
iI
is summable over I then b
i
is nite for all i I. This
allows us to give the
Denition 7.9 Let b : I C be summable over I. We dene the sum of
b
i
iI
over I as
iI
b
i
=
iI
(b
i
) +i
iI
(b
i
).
Therefore, the sum of complex valued terms is dened only when the sum is
summable and, hence, this sum always has a nite value. Again, we can say
that if b : I C is summable over I (which is equivalent to b being integrable
over I with respect to counting measure) then
iI
b
i
=
_
I
b d .
We shall see now the form that some of the important results of general
integrals take when we specialize them to sums. They are simple and straight-
forward formulations of known results but, since they are very important when
one is working with sums, we shall state them explicitly. Their content is the
interchange of limits and sums. It should be stressed that it is very helpful to
be able to recognize the underlying integral theorem behind a property of sums.
Proofs are not needed.
Theorem 7.14 (i) (The Monotone Convergence Theorem) Let b, b
(k)
:
I [0, +] (k N). If b
(k)
i
b
i
for all i, then
iI
b
(k)
i

iI
b
i
.
(ii) Let b
(k)
: I [0, +] (k N). Then
iI
(
+
k=1
b
(k)
i
) =
+
k=1
(
iI
b
(k)
i
).
(iii) (The Lemma of Fatou) Let b, b
(k)
: I [0, +] (k N). If b
i
=
liminf
k+
b
(k)
i
for all i I, then
iI
b
i
liminf
k+
iI
b
(k)
i
.
(iv) (The Dominated Convergence Theorem) Let b, b
(k)
: I [0, +]
(k N) and c : I [0, +]. If [b
(k)
i
[ c
i
for all i and k, if
iI
c
i
< +
and if b
(k)
i
b
i
for all i, then
iI
b
(k)
i

iI
b
i
.
(v) (The Series Theorem) Let b
(k)
: I [0, +] (k N). Assuming that
+
k=1
(
iI
[b
(k)
i
[) < +, then
+
k=1
b
(k)
i
converges for every i. Moreover,
iI
(
+
k=1
b
(k)
i
) =
+
k=1
(
iI
b
(k)
i
).
Observe that the only -null set is the . Therefore, saying that a property
holds -a.e. on I is equivalent to saying that it holds at every point of I.
Going back to the general case, if is the point-mass distribution induced
by the function a : X [0, +], and f : X R, then
_
X
f d is dened
if and only if either
xX
f
+
(x)a
x
=
_
X
f
+
d < + or
xX
f
(x)a
x
=
_
X
f
d < +, and in this case we have

_
X
f d =
_
X
f
+
d
_
X
f
d =
xX
f
+
(x)a
x

xX
f
(x)a
x
=
xX
f(x)a
x
.
Moreover, f is integrable if and only if
xX
[f(x)[a
x
=
_
X
[f[ d < +. This
is also true when f : X C, and in this case we have
_
X
f d =
xX
(f(x)
D
f
(x))a
x
+i
xX
(f(x)
D
f
(x))a
x
,
where D
f
= x X[ f(x) ,= . Since
xX
[f(x)[a
x
< +, it is clear that
f(x) = can happen only if a
x
= 0 and a
x
= +can happen only if f(x) = 0.
But, then f(x)a
x
C for all x X and, moreover, f(x)
D
f
(x)a
x
= f(x)a
x
for
all x X. Therefore, we get
_
X
f d =
xX
(f(x)a
x
) +i
xX
(f(x)a
x
) =
xX
f(x)a
x
.
Now we have arrived at the complete interpretation of sums as integrals.
Theorem 7.15 Let be a point-mass distribution induced by a : X [0, +].
If f : X R or C, then the
_
X
f d exists if and only if the
xX
f(x)a
x
exists
and, in this case,
_
X
f d =
xX
f(x)a
x
.
A simple particular case of a point-mass distribution is the Dirac mass
x0
at x
0
X. We remember that this is induced by a
x
= 1 if x = x
0
and a
x
= 0
if x ,= x
0
. In this case the integrals become very simple:
_
X
f d
x0
= f(x
0
)
for every f. It is clear that f is integrable if and only if f(x
0
) C. Thus,
integration with respect to the Dirac mass at x
0
coincides with the so-called
point evaluation at x
0
.
7.6 Lebesgue-integral.
A function f : R
n
R or C is called Lebesgue-integrable if it is integrable
with respect to m
n
.
It is easy to see that every continuous f : R
n
R or C which is 0 outside
some bounded set is Lebesgue-integrable. Indeed, f is then Borel-measurable
and if Q is any closed interval in R
n
outside of which f is 0, then [f[ K
Q
,
where K = max[f(x)[ [ x Q < +. Therefore,
_
R
n
[f[ dm
n
Km
n
(Q) <
+ and f is Lebesgue-integrable.
Theorem 7.16 (Approximation) Let f : R
n
R or C be Lebesgue-integra-
ble. Then for every > 0 there is some continuous function g : R
n
R or C
which is 0 outside some bounded set so that
_
R
n
[g f[ dm
n
< .
Proof: (a) Let < a < b < + and for each (0,
ba
2
) consider the
continuous function
a,b,
: R [0, 1] which is 1 on (a + , b ), is 0 outside
(a, b) and is linear in each of [a, a +] and [b , b].
Let R = (a
1
, b
1
) (a
n
, b
n
) be an open interval in R
n
. Consider, for
small > 0, the open interval R
= (a
1
+, b
1
) (a
n
+, b
n
) R.
Then it is clear that, by choosing small enough, we can have m
n
(R R
) < .
Dene the function
R,
: R
n
[0, 1] by the formula
R,
(x
1
, . . . , x
n
) =
a1,b1,
(x
1
)
an,bn,
(x
n
).
7.6. LEBESGUE-INTEGRAL. 111
Then,
R,
is continuous on R
n
, it is 1 on R
and it is 0 outside R. Therefore,

_
R
n
[
R,

R
[ m
n
m
n
(R R
) < .
(b) Let E L
n
have m
n
(E) < +. Theorem 4.6 implies that there are
pairwise disjoint open intervals R
1
, . . . , R
l
so that m
n
(E(R
1
R
l
)) <

2
.
The functions
E
and
R1
+ +
R
l
dier (by at most 1) only in the set
E(R
1
R
l
). Hence,
_
R
n
[
l
i=1
Ri

E
[ dm
n
<

2
.
By (a), we can take small enough > 0 so that, for each R
i
, we have
_
R
n
[
Ri,

Ri
[ m
n
<

2l
. This implies
_
R
n
[
l
i=1
Ri,

l
i=1
Ri
[ dm
n
<
l
i=1
2l
=

2
.
Denoting =
l
i=1
Ri,
: R
n
R, we have
_
R
n
[
E
[ dm
n
< . Observe
that is a continuous function which is 0 outside the bounded set
l
i=1
R
i
.
(c) Let now f : R
n
R or C be Lebesgue-integrable. From Theorem 7.12
we know that there is some L
n
measurable simple : R
n
R or C so
that
_
R
n
[ f[ dm
n
<

2
. Let =
m
j=1
Ej
be the standard representa-
tion of , where we omit the possible value = 0. From
m
j=1
[
j
[m
n
(E
j
) =
_
R
n
[[ dm
n

_
R
n
[f[ dm
n
+
_
R
n
[f [ dm
n
< +, we get that m
n
(E
j
) < +
for all j. By part (b), for each E
j
we can nd a continuous
j
: R
n
R so
that
_
R
n
[
j

Ej
[ dm
n
<

2m|j|
.
If we set g =
m
j=1
j
, then this is continuous on R
n
and
_
R
n
[g f[ dm
n

_
R
n
[g [ dm
n
+
_
R
n
[ f[ dm
n
<
m
j=1
_
R
n
[
j
j

j
Ej
[ dm
n
+

2
<
m
j=1
[
j
[

2m[
j
[
+

2
= .
Since each
j
is 0 outside a bounded set, g is also 0 outside a bounded set.
We shall now investigate the relation between the Lebesgue-integral and the
Riemann-integral. We recall the denition of the latter.
Assume that Q = [a
1
, b
1
] [a
n
, b
n
] is a closed interval in R
n
and consider
a bounded function f : Q R.
If m N is arbitrary and Q
1
, . . . , Q
l
are arbitrary closed intervals which
have pairwise disjoint interiors and so that Q = Q
1
Q
l
, then we say that
= Q
1
, . . . Q
l
is a partition of Q. If P, P
1
, . . . , P
l
are the open-closed intervals with the same
sides as, respectively, Q, Q
1
, . . . , Q
l
, then Q
1
, . . . Q
l
is a partition of Q if and
only if the P
1
, . . . , P
l
are pairwise disjoint and P = P
1
P
l
. Now, since f is
bounded, in each Q
j
we may consider the real numbers m
j
= inff(x) [ x Q
j
and M
j
= supf(x) [ x Q
j
. We then dene the upper Darboux-sum and
the lower Darboux-sum of f with respect to as, respectively,
(f; ) =
l
j=1
M
j
vol
n
(Q
j
),
(f; ) =
l
j=1
m
j
vol
n
(Q
j
).
If m = inff(x) [ x Q and M = supf(x) [ x Q, we have that m m
j

M
j
M for every j and, using Lemma 4.2, we see that
mvol
n
(Q) (f; ) (f; ) M vol
n
(Q).
If
1
= Q
1
1
, . . . , Q
1
l1
and
2
= Q
2
1
, . . . , Q
2
l2
are two partitions of Q, we
say that
2
is ner than
1
if every Q
2
i
is included in some Q
1
j
. Then it is
obvious that, for every Q
1
j
of
1
, the Q
2
i
s of
2
which are included in Q
1
j
cover
it and hence form a partition of it. Therefore, from Lemma 4.2 again,
m
1
j
vol
n
(Q
1
j
)
Q
2
i
Q
1
j
m
2
i
vol
n
(Q
2
i
)
Q
2
i
Q
1
j
M
2
i
vol
n
(Q
2
i
) M
1
j
vol
n
(Q
1
j
).
Summing over all j = 1, . . . , l
1
we nd
(f;
1
) (f;
2
) (f;
2
) (f;
1
).
Now, if
1
= Q
1
1
, . . . , Q
1
l1
and
2
= Q
2
1
, . . . , Q
2
l2
are any two partitions of
Q, we form their common renement = Q
1
j
Q
2
i
[ 1 j l
1
, 1 i l
2
.
Then, (f;
1
) (f; ) (f; ) (f;
2
) and we conclude that
mvol
n
(Q) (f;
1
) (f;
2
) M vol
n
(Q)
for all partitions
1
,
2
of Q. We now dene
(
n
)
_
Q
f = sup(f; ) [ partition of Q
(
n
)
_
Q
f = inf(f; ) [ partition of Q
and call them, respectively, the lower Riemann-integral and the upper
Riemann-integral of f. It is then clear that
mvol
n
(Q) (
n
)
_
Q
f (
n
)
_
Q
f M vol
n
(Q).
We say that f is Riemann-integrable over Q if (
n
)
_
Q
f = (
n
)
_
Q
f and
in this case we dene
(
n
)
_
Q
f = (
n
)
_
Q
f = (
n
)
_
Q
f
to be the Riemann-integral of f.
Lemma 7.11 Let Q be a closed interval in R
n
and f : Q R be bounded.
Then f is Riemann-integrable if and only if for every > 0 there is some
partition of Q so that (f; ) (f; ) < .
Proof: To prove the suciency, take arbitrary > 0 and the corresponding .
Then 0 (
n
)
_
Q
f (
n
)
_
Q
f (f; ) (f; ) < . Taking the limit
as 0+, we prove the equality of the upper Riemann-integral and the lower
Riemann-integral of f.
For the necessity, assume (
n
)
_
Q
f = (
n
)
_
Q
f and for each > 0 take
partitions
1
,
2
of Q so that (
n
)
_
Q
f

2
< (f;
1
) and (f;
2
) <
(
n
)
_
Q
f +

2
. Therefore, if is the common renement of
1
and
2
, then
(f; ) (f; ) (f;
2
) (f;
1
) < .
Proposition 7.9 Let Q be a closed interval in R
n
and f : Q R be continuous
on Q. Then f is Riemann-integrable.
Proof: Since f is uniformly continuous on Q, given any > 0 there is a > 0
so that [f(x) f(y)[ <

voln(Q)
for all x, y Q whose distance is < . We take
any partition = Q
1
, . . . , Q
l
of Q, so that every Q
j
has diameter < . Then
[f(x) f(y)[ <

voln(Q)
for all x, y in the same Q
j
. This implies that for every
Q
j
we have M
j
m
j
= maxf(x) [ x Q
j
minf(y) [ y Q
j
<

voln(Q)
.
Hence
(f; ) (f; ) =
l
j=1
(M
j
m
j
) vol
n
(Q
j
) <

vol
n
(Q)
l
j=1
vol
n
(Q
j
) =
and Lemma 7.11 implies that f is Riemann-integrable.
Theorem 7.17 Let Q be a closed interval in R
n
and f : Q R be Riemann-
integrable. If we extend f as 0 outside Q, then f is Lebesgue-integrable and
_
R
n
f dm
n
=
_
Q
f dm
n
= (
n
)
_
Q
f.
Proof: Lemma 7.11 implies that, for all k N, there is a partition
k
=
Q
k
1
, . . . , Q
k
l
k
of Q so that (f;
k
) (f;
k
) <
1
k
. We consider the simple
functions
k
=
l
k
j=1
m
k
j

P
k
j
,
k
=
l
k
j=1
M
k
j

P
k
j
,
where P
k
j
is the open-closed interval with the same sides as Q
k
j
and m
k
j
=
inff(x) [ x Q
k
j
, M
k
j
= supf(x) [ x Q
k
j
.
From (f;
k
) (
n
)
_
Q
f (f;
k
) we get that
(f;
k
), (f;
k
) (
n
)
_
Q
f.
It is clear that
k
f
P

k
for all k, where P is the open-closed interval
with the same sides as Q. It is also clear that
_
R
n
k
dm
n
=
l
k
j=1
m
k
j
vol
n
(P
k
j
) = (f;
k
)
_
R
n
k
dm
n
=
l
k
j=1
M
k
j
vol
n
(P
k
j
) = (f;
k
).
Hence
_
R
n
(
k

k
) dm
n
<
1
k
for all k.
We dene g = limsup
k+
k
and h = liminf
k+
k
and then, of course,
g f
P
h. The Lemma of Fatou implies that
_
R
n
(h g) dm
n
= liminf
k+
_
R
n
(
k

k
) dm
n
= 0.
By Proposition 7.3, g = h m
n
-a.e. on R
n
and, thus, f = g = h m
n
-a.e. on R
n
.
Since g, h are Borel-measurable, Proposition 6.24 implies that f is Lebesgue-
measurable. f is also bounded and is 0 outside Q. Hence [f[ K
Q
, where
K = sup[f(x)[ [ x Q. Thus,
_
R
n
[f[ dm
n
Km
n
(Q) < + and f is
Lebesgue-integrable.
Another application of the Lemma of Fatou gives
_
R
n
(h f
P
) dm
n
liminf
k+
_
R
n
(
k
f
P
) dm
n
= liminf
k+
(f;
k
)
_
R
n
f
P
dm
n
= (
n
)
_
Q
f
_
R
n
f
P
dm
n
.
Hence
_
R
n
hdm
n
(
n
)
_
Q
f and, similarly, (
n
)
_
Q
f
_
R
n
g dm
n
. Since
f = g = h m
n
-a.e. on R
n
, we conclude that
(
n
)
_
Q
f =
_
R
n
f dm
n
.
Theorem 7.17 incorporates the notion of Riemann-integral in the notion
of Lebesgue-integral. It says that the collection of Riemann-integrable func-
tions is included in the collection of Lebesgue-integrable functions and that the
Riemann-integral is the restriction of the Lebesgue-integral on the collection of
Riemann-integrable functions.
Furthermore, Theorem 7.17 provides a tool to calculate Lebesgue-integrals,
at least in the case of R. If a function f is Riemann-integrable over a closed
interval [a, b] R, we have many techniques (integration by parts, change
of variable, antiderivatives) to calculate its
_
b
a
f(x) dx. In case the given f is
Riemann-integrable over intervals [a
k
, b
k
] with a
k
and b
k
+ and we
can calculate the integrals
_
b
k
a
k
f(x) dx, then it is a matter of being able to pass
to the limit
_
b
k
a
k
f(x) dx
_
+
f(x) dx to calculate the integral over R. To do

this we may try to use the Monotone Convergence Theorem or the Dominated
Convergence Theorem.
On the other hand we have examples of bounded functions f : Q R which
are Lebesgue-integrable but not Riemann-integrable.
Example
Dene f(x) = 1, if x Q has all coordinates rational, and f(x) = 0, if at least
one of the coordinates of x is irrational. If = Q
1
, . . . , Q
l
is any partition
of Q, then all Q
j
s with non-empty interior (the rest do not matter because
they have zero volume) contain at least one x with f(x) = 1 and at least one x
with f(x) = 0. Hence, for any such Q
j
we have M
j
= 1 and m
j
= 0. Hence,
(f; ) = vol
n
(Q) and (f; ) = 0 for every and this says that (
n
)
_
Q
f = 1
and (
n
)
_
Q
f = 0. Thus, f is not Riemann-integrable.
On the other hand, f extended as 0 outside Q is 0 m
n
-a.e. on R
n
and hence
it is Lebesgue-integrable on R
n
with
_
R
n
f dm
n
=
_
Q
f dm
n
= 0.
Notation
If Q = [a, b] is any closed interval in R, then the Riemann-integral (
n
)
_
[a,b]
f
of a function f : [a, b] R is, traditionally, denoted by
_
b
a
f,
_
b
a
f(x) dx,
_
b
a
f(t) dt, etc.
After Theorem 7.17 we are allowed to use the same notations for the correspond-
ing Lebesgue-integral
_
[a,b]
f dm
1
. We also observe that m
1
(a) = m
1
(b) = 0
and, hence, the above notations cover all integrals
_
S
f dm
1
, where S is any of
the intervals with end-points a and b. This is also extended to include the cases
of all unbounded intervals (, b), (, b], (a, +), [a, +) and (, +).
Therefore, in all cases of intervals S with end-points a, b R we use any of
_
b
a
f dm
1
,
_
S
f(x) dm
1
(x),
_
b
a
f(x) dm
1
(x),
_
b
a
f,
_
S
f(x) dx,
_
b
a
f(x) dx
for the Lebesgue-integral
_
S
f dm
1
of f over S.
The notation dm
1
(x), dm
1
(t), dx, dt etc. for Lebesgue-measure is used also
in higher dimensions. We may, thus, write
_
A
f(x) dm
n
(x),
_
A
f(x) dx
for the Lebesgue-integral
_
A
f dm
n
of f over the Lebesgue-measurable A R
n
.
The last topic will be the change of Lebesgue-integral under transformations
of the space.
Proposition 7.10 Let T : R
n
R
n
be a linear transformation with det(T) ,=
0. If (Y,
) is a measurable space and f : R

n
Y is (L
n
,
)measurable,
then f T
1
: R
n
Y is also (L
n
,
)measurable.
Proof: For every E
we have (f T
1
)
1
(E) = T(f
1
(E)) L
n
, because
of Theorem 7.18.
Theorem 7.18 Let T : R
n
R
n
be a linear transformation with det(T) ,= 0
and f : R
n
R or C be Lebesgue-measurable.
(i) If f : R
n
R and the
_
R
n
f dm
n
exists, then the
_
R
n
f T
1
dm
n
also
exists and
_
R
n
f T
1
dm
n
= [ det(T)[
_
R
n
f dm
n
.
(ii) If f : R
n
C is integrable, then f T
1
is also integrable and the equality
of (i) is again true.
Proof: (a) Let : R
n
[0, +) be a non-negative Lebesgue-measurable
simple function and =
m
j=1
Ej
be its standard representation. Then
_
R
n
dm
n
=
m
j=1
j
m
n
(E
j
).
It is clear that T
1
=
m
j=1
Ej
T
1
=
m
j=1
T(Ej)
from which
we get
_
R
n
T
1
dm
n
=
m
j=1
j
m
n
(T(E
j
)) = [ det(T)[
m
j=1
j
m
n
(E
j
) =
[ det(T)[
_
R
n
dm
n
.
(b) Let f : R
n
[0, +] be Lebesgue-measurable. Take any increasing
sequence
k
of non-negative Lebesgue-measurable simple functions so that
k
f on R
n
. Then
k
T
1
is increasing and
k
T
1
f T
1
on R
n
. From part (a),
_
R
n
f T
1
dm
n
= lim
k+
_
R
n

k
T
1
dm
n
=
[ det(T)[ lim
k+
_
R
n

k
dm
n
= [ det(T)[
_
R
n
f dm
n
.
(c) Let f : R
n
R and the
_
R
n
f dm
n
exist. Then (f T
1
)
+
= f
+
T
1
and (f T
1
)
= f
T
1
, and from (b) we get
_
R
n
(f T
1
)
+
dm
n
=
[ det(T)[
_
R
n
f
+
dm
n
and
_
R
n
(f T
1
)
dm
n
= [ det(T)[
_
R
n
f
dm
n
. Now (i)
is obvious.
(d) Let f : R
n
C be integrable. From [f T
1
[ = [f[ T
1
and from (b) we
have that
_
R
n
[f T
1
[ dm
n
= [ det(T)[
_
R
n
[f[ dm
n
< +. Hence f T
1
is
also integrable.
We take an F : R
n
C so that F = f m
n
-a.e. on R
n
.
If A = x R
n
[ F(x) ,= f(x) and B = x R
n
[ F T
1
(x) ,= f
T
1
(x), then B = T(A). Hence, m
n
(B) = [ det(T)[m
n
(A) = 0 and, thus,
F T
1
= f T
1
m
n
-a.e. on R
n
. Therefore, to prove (ii) it is enough to prove
_
R
n
F T
1
dm
n
= [ det(T)[
_
R
n
F dm
n
.
We have (F T
1
) = (F)T
1
and, from part (c),
_
R
n
(F T
1
) dm
n
=
[ det(T)[
_
R
n
(F) dm
n
. We, similarly, prove the same equality with the imagi-
nary parts and, combining, we get the desired equality.
The equality of the two integrals in Theorem 7.19 is nothing but the (linear)
change of variable formula. If we write y = T
1
(x) or, equivalently, x =
7.7. LEBESGUE-STIELTJES-INTEGRAL. 117
T(y), then
_
R
n
f T
1
(x) dx =
_
R
n
f(T
1
(x)) dx = [det(T)[
_
R
n
f(y) dy.
Thus, the informal rule for the change of dierentials is
dx = [det(T)[dy.
7.7 Lebesgue-Stieltjes-integral.
Every monotone f : R R is Borel-measurable. This is seen by observing
that f
1
((a, b]) is an interval, and hence a Borel set, for every (a, b]. If, now,
F : R R is another increasing function and
F
is the induced Borel-measure,
then f satises the necessary measurability condition and the
_
R
f d
F
exists
provided, as usual, that either
_
R
f
+
d
F
< + or
_
R
f
d
F
< +.
The same can, of course, be said when f is continuous.
In particular, if f is continuous or monotone in a (bounded) interval S and
it is bounded on S, then it is integrable over S with respect to
F
.
We shall prove three classical results about Lebesgue-Stieltjes-integrals.
Observe that the four integrals which we get from
_
S
f d
F
, by taking S =
[a, b], [a, b), (a, b] and (a, b), may be dierent. This is because the
_
{a}
f d
F
=
f(a)
F
(a) = f(a)(F(a+) F(a)) and
_
{b}
f d
F
= f(b)(F(b+) F(b))
may not be zero.
Proposition 7.11 (Integration by parts) Let F, G : R R be two increasing
functions and
F
,
G
be the induced Lebesgue-Stieltjes-measures. Then
_
(a,b]
G(x+) d
F
+
_
(a,b]
F(x) d
G
= G(b+)F(b+) G(a+)F(a+)
for all a, b R with a b. In this equality we may interchange F with G.
Similar equalities hold for the other types of intervals, provided we use the
appropriate limits of F, G at a, b in the right side of the above equality.
Proof: We introduce a sequence of partitions
k
= c
k
0
, . . . , c
k
l
k
of [a, b] so that
a = c
k
0
< c
k
1
< < c
k
l
k
= b for each k and so that
lim
k+
maxc
k
j
c
k
j1
[ 1 j l
k
= 0.
We also introduce the simple functions
g
k
=
l
k
j=1
G(c
k
j
+)
(c
k
j1
,c
k
j
]
, f
k
=
l
k
j=1
F(c
k
j1
+)
(c
k
j1
,c
k
j
]
.
It is clear that G(a+) g
k
G(b+) and F(a+) f
k
F(b) for all k.
If, for an arbitrary x (a, b] we take the interval (c
k
j1
, c
k
j
] containing x
(observe that j = j(k, x)), then g
k
(x) = G(c
k
j
+) and f
k
(x) = F(c
k
j1
+). Since
lim
k+
(c
k
j
c
k
j1
) 0, we have that c
k
j
x and c
k
j1
x. Therefore,
g
k
(x) G(x+), f
k
(x) F(x)
as k +.
We apply the Dominated Convergence Theorem to nd
l
k
j=1
G(c
k
j
+)(F(c
k
j
+) F(c
k
j1
+)) =
_
(a,b]
g
k
d
F

_
(a,b]
G(x+) d
F
,
l
k
j=1
F(c
k
j1
+)(G(c
k
j
+) G(c
k
j1
+)) =
_
(a,b]
f
k
d
G

_
(a,b]
F(x) d
G
as k +.
Adding the two last relations we nd
G(b+)F(b+) G(a+)F(a+) =
_
(a,b]
G(x+) d
F
+
_
(a,b]
F(x) d
G
.
If we want the integrals over (a, b), we have to subtract from the right side
of the equality the quantity
_
{b}
G(x+) d
F
+
_
{b}
F(x) d
G
which is equal to
G(b+)(F(b+)F(b))+F(b)(G(b+)G(b)) = G(b+)F(b+)G(b)F(b).
Then, subtracting the same quantity from the left side of the equality, this
becomes F(b)G(b) F(a+)G(a+). We work in the same way for all other
types of intervals.
The next two results concern the reduction of Lebesgue-Stieltjes-integrals
to Lebesgue-integrals. This makes calculation of the former more accessible in
many situations.
Proposition 7.12 Assume that F : R R is increasing and has a continuous
derivative on (a, b) for some a, b with a < b +. Then
F
(E) =
_
E
F
(x) dx
for every Borel set E (a, b) and
_
(a,b)
f d
F
=
_
b
a
f(x)F
(x) dx
for every Borel-measurable f : R R or C for which either of the two integrals
exists.
7.7. LEBESGUE-STIELTJES-INTEGRAL. 119
Proof: (i) The assumptions on F imply that it is continuous on (a, b) and that
F
0 on (a, b). For every [c, d] (a, b) we have, by the fundamental theorem
of calculus, that
_
d
c
F
(x) dx = F(d) F(c) =

F
((c, d]). If we apply this to
two strictly monotone sequences c
n
a and d
n
b, we get, by the monotone
convergence theorem, that
_
b
a
F
(x) dx = F(b) F(a+) =

F
((a, b)) < +.
Hence, F
is integrable over (a, b).

We now introduce the Borel measure on R by the formula
(E) =
F
(E (a, b)) +
_
E(a,b)
F
(x) dx
for every Borel set E R. If (c, d] (, a] or if (c, d] [b, +), then
obviously ((c, d]) =
F
((c, d]). If (c, d] (a, b), then, by what we said in the
rst paragraph, again ((c, d]) =
_
(c,d]
F
(x) dx =
F
((c, d]). It is easy, now, to
prove that for every (c, d] we have ((c, d]) =
F
((c, d]). We just need to break
the interval into at most three subintervals.
Theorem 5.5 implies that =
F
and hence
F
(E) =
F
(E (a, b)) +
_
E(a,b)
F
(x) dx for every Borel set E R. This implies
F
(E (a, b)) =
_
E(a,b)
F
(x) dx
for every Borel E R and this can be written
_
(a,b)
E
d
F
=
F
(E (a, b)) =
_
E(a,b)
F
(x) dx =
_
b
a

E
(x)F
(x) dx. Taking linear combinations of character-

istic functions, we nd
_
(a,b)
d
F
=
_
b
a
(x)F
(x) dx for all Borel-measurable

simple functions : R [0, +). Now, applying the Monotone Convergence
Theorem to an appropriate increasing sequence of simple functions, we get
_
(a,b)
f d
F
=
_
b
a
f(x)F
(x) dx
for every Borel-measurable f : R [0, +]. The proof is easily concluded for
any f : R R, by taking its positive and negative parts, and then for any
f : R C, by taking its real and imaginary parts (and paying attention to the
set where f = ).
Proposition 7.13 Assume that F : R R is increasing and G : (a, b) R
has a bounded, continuous derivative on (a, b), where < a < b < +.
Then,
_
(a,b)
Gd
F
= G(b)F(b) G(a+)F(a+)
_
b
a
F(x)G
(x) dx
= G(b)F(b) G(a+)F(a+)
_
b
a
F(x+)G
(x) dx.
Proof: (A) By the assumptions on G we have that it is continuous on (a, b) and
that the limits G(b) and G(a+) exist. We then extend G as G(b) on [b, +)
and as G(a+) on (, a] and G becomes continuous on R. We use the same
partitions
k
as in the proof of Proposition 7.11 and the same simple functions
g
k
=
l
k
j=1
G(c
k
j
+)
(c
k
j1
,c
k
j
]
=
l
k
j=1
G(c
k
j
)
(c
k
j1
,c
k
j
]
.
We have again that [g
k
[ M where M = sup[G(x)[ [ x [a, b] and that
g
k
(x) G(x+) = G(x) for every x (a, b]. By the Dominated Convergence
Theorem,
l
k
j=1
G(c
k
j
)(F(c
k
j
+) F(c
k
j1
+)) =
_
(a,b]
g
k
d
F

_
(a,b]
G(x) d
F
as k +.
By the mean value theorem, for every j with j = 1, . . . , l
k
, we have
G(c
k
j
) G(c
k
j1
) = G
(
k
j
)(c
k
j
c
k
j1
)
for some
k
j
(c
k
j1
, c
k
j
). Hence
l
k
j=1
F(c
k
j1
+)(G(c
k
j
)G(c
k
j1
)) =
l
k
j=1
F(c
k
j1
+)G
(
k
j
)(c
k
j
c
k
j1
) =
_
b
a
k
(x) dx,
where we set
k
=
l
k
j=1
F(c
k
j1
+)G
(
k
j
)
(c
k
j1
,c
k
j
]
.
We have that
k
(x) F(x)G
(x) for every x (a, b) and that [

k
[ K on
(a, b) for some K which does not depend on k. By the Dominated Convergence
Theorem,
_
b
a

k
(x) dx
_
b
a
F(x)G
(x) dx. We combine to get

G(b)F(b+) G(a)F(a+) =
_
(a,b]
G(x) d
F
+
_
b
a
F(x)G
(x) dx.
From both sides we subtract
_
{b}
G(x) d
F
= G(b)(F(b+) F(b)) to nd
G(b)F(b) G(a)F(a+) =
_
(a,b)
G(x) d
F
+
_
b
a
F(x)G
(x) dx,
which is the rst equality in the statement of the proposition. The second
equality is proved in a similar way.
(B) There is a second proof making no use of partitions.
Assume rst that G is also increasing in (a, b). Then its extension as G(a+)
on (, a] and as G(b) on [b, +) is increasing in R. We apply Proposition
7.11 to get
_
(a,b)
Gd
F
= G(b)F(b) G(a+)F(a+)
_
(a,b)
F(x) d
G
,
7.8. REDUCTION TO INTEGRALS OVER R. 121
which, by Proposition 7.12, becomes the desired
_
(a,b)
Gd
F
= G(b)F(b) G(a+)F(a+)
_
b
a
F(x)G
(x) dx.
If G is not increasing, we take an arbitrary x
0
(a, b) and write G(x) =
G(x
0
)+
_
x
x0
G
(t) dt for every x (a, b). Now, (G
)
+
and (G
are non-negative,
continuous, bounded functions on (a, b) and we can write G = G
1
G
2
on (a, b),
where
G
1
(x) = G
(
x
0
) +
_
x
x0
(G
)
+
(t) dt, G
2
(x) =
_
x
x0
(G
(t) dt
for all t (a, b). By the continuity of (G
)
+
and (G
and the fundamental

theorem of calculus, we have that G
1
= (G
)
+
0 and G
2
= (G
0 on (a, b).
Hence, G
1
and G
2
are both increasing with bounded, continuous derivative on
(a, b) and from the previous paragraph we have
_
(a,b)
G
i
d
F
= G
i
(b)F(b) G
i
(a+)F(a+)
_
b
a
F(x)G
i
(x) dx
for i = 1, 2. We subtract the two equalities and prove the desired equality.
It is worth keeping in mind the fact, which is included in the second proof
of Proposition 7.13, that an arbitrary G with a continuous, bounded derivative
on an interval (a, b) can be decomposed as a dierence, G = G
1
G
2
, of two
increasing functions with a continuous, bounded derivative on (a, b). We shall
generalise it later in the context of functions of bounded variation.
7.8 Reduction to integrals over R.
Denition 7.10 Let f : X [0, +] be measurable. Then the function
f
: [0, +) [0, +], dened by
f
(t) = (x X[ t < f(x)),
is called the distribution function of f.
Some properties of
f
are easy to prove. It is obvious that
f
is non-negative
and decreasing on [0, +). Since x X[ t
n
< f(x) x X[ t < f(x) for
every t
n
t, we see that
f
is continuous from the right on [0, +).
Hence, there exists some t
0
[0, +] with the property that
f
is + on
the interval [0, t
0
) (which may be empty) and
f
is nite in the interval (t
0
, +)
(which may be empty).
Proposition 7.14 Let f : X [0, +] be measurable and G : R R be
increasing with G(0) = 0. Then
_
X
G(f(x)) d =
_
[0,+)
f
d
G
.
Moreover, if G has continuous derivative on (0, +), then
_
X
G f d =
_
+
0
f
(t)G
(t) dt +
f
(0)G(0+).
Thus,
_
X
f d =
_
+
0

f
(t) dt.
Proof: (a) Let =
m
j=1
Ej
be a non-negative measurable simple func-
tion on X with its standard representation, where we omit the value 0. Rear-
range so that 0 <
1
< <
m
and then
(t) =
_
_
(E
1
) +(E
2
) + +(E
m
), if 0 t <
1
(E
2
) + +(E
m
), if
1
t <
2

(E
m
), if
m1
t <
m
0, if
m
t
Then
_
[0,+)
d
G
= ((E
1
) +(E
2
) + +(E
m
))
_
G(
1
) G(0)
_
+((E
2
) + +(E
m
))
_
G(
2
) G(
1
)
_

+(E
m
)
_
G(
m
) G(
m1
)
_
= G(
1
)(E
1
) +G(
2
)(E
2
) + +G(
m
)(E
m
)
=
_
X
G((x)) d.
because G((x)) is a simple function taking value G(
j
) on each E
j
and
value G(0) = 0 on (E
1
E
m
)
c
.
(b) Take arbitrary measurable f : X [0, +] and any increasing sequence
n
of non-negative measurable simple
n
: X [0, +) so that
n
f
on X. Then 0 G(
n
(x)) G(f(x)) for every x X and, by the Monotone
Convergence Theorem,
_
X
G(
n
(x)) d
_
X
G(f(x)) d.
Since x X[ t <
n
(x) x X[ t < f(x), we have that
n
(t)
f
(t)
for every t (0, +). Again by the Monotone Convergence Theorem,
_
[0,+)
n
d
G

_
[0,+)
f
d
G
.
7.8. REDUCTION TO INTEGRALS OVER R. 123
By the result of (a), we combine and get
_
X
G(f(x)) d =
_
[0,+)
f
d
G
.
Proposition 7.12 implies the second equality of the statement and the special
case G(t) = t implies the last equality.
Proposition 7.15 Let (X) < + and f : X [0, +] be measurable.
We dene F : R R by
F
f
(t) = (x X [ f(x) t) =
_
(X)
f
(t), if 0 t < +
0, if < t < 0
Then F
f
is increasing and continuous from the right and, for every increasing
G : R R with G(0) = 0, we have
_
X
G(f(x)) d =
_
[0,+)
G(t) d
F
f
+G(+)(f
1
(+)).
Proof: It is obvious that F
f
is increasing. If t
n
t, then x X[ f(x) t
n

x X [ f(x) t. By the continuity of from above, we get F
f
(t
n
) F
f
(t)
and F
f
is continuous from the right.
We take any n N and apply Proposition 7.11 to nd
_
[0,n]
G(t) d
F
f
= G(n+)F
f
(n)
_
[0,n]
F
f
d
G
=
_
[0,n]
(F
f
(n) F
f
) d
G
.
The left side is =
_
[0,+)
G(t)
[0,n]
(t) d
F
f

_
[0,+)
G(t) d
F
f
, by the
Monotone Convergence Theorem.
Similarly, the right side is =
_
[0,+)
(x X[ t < f(x) n)
[0,n]
(t) d
G

_
[0,+)
(x X[ t < f(x) < +) d
G
, again by the Monotone Convergence
Theorem.
Therefore,
_
[0,+)
G(t) d
F
f
=
_
[0,+)
(x X [ t < f(x) < +) d
G
and, adding to both sides the quantity G(+)(x X[ f(x) = +) we nd
_
[0,+)
G(t) d
F
f
+G(+)(x X[ f(x) = +) =
_
[0,+)
f
d
G
.
Now, the equality of the statement is an implication of Proposition 7.14.
7.9 Exercises.
1. The graph and the area under the graph of a function.
Let f : R
n
[0, +] be Lebesgue-measurable. If
A
f
= (x
1
, . . . , x
n
, x
n+1
) [ 0 x
n+1
< f(x
1
, . . . , x
n
) R
n+1
,
G
f
= (x
1
, . . . , x
n
, x
n+1
) [ x
n+1
= f(x
1
, . . . , x
n
) R
n+1
,
prove that A
f
, G
f
L
n+1
and
m
n+1
(A
f
) =
_
R
n
f dm
n
, m
n+1
(G
f
) = 0.
2. An equivalent denition of the integral.
Let f : X [0, +] be measurable. Take all = E
1
, . . . , E
l
,
where l N and the non-empty sets E
1
, . . . , E
l
are pairwise disjoint
and cover X. Such are called partitions of X. Dene (f, ) =
l
j=1
m
j
(E
j
), where m
j
= inff(x) [ x E
j
. Prove that
_
X
f d = sup(f, ) [ is a partition of X.
3. If f, g, h : X R are measurable, g, h are integrable and g f h
-a.e. on X, prove that f is also integrable.
4. The Uniform Convergence Theorem.
Let all f
n
: X R or C be integrable and let f
n
f uniformly on X. If
(X) < +, prove that f is integrable and that
_
X
f
n
d
_
X
f d.
5. The Bounded Convergence Theorem.
Let f, f
n
: X R or C be measurable. If (X) < + and there is
M < + so that [f
n
[ M -a.e. on X and f
n
f -a.e. on X, prove
that
_
X
f
n
d
_
X
f d.
6. Let f, f
n
: X R or C be measurable and g : X [0, +] be
integrable. If [f
n
[ g -a.e. on X for every n and f
n
f -a.e. on X,
prove that
_
X
[f
n
f[ d 0.
7. Let f, f
n
: X [0, +] be measurable with f
n
f -a.e. on X and
f
n
f -a.e. on X. Prove that
_
X
f
n
d
_
X
f d.
8. Let f, f
n
: X [0, +] be measurable and f
n
f -a.e. on X. If
there is M < + so that
_
X
f
n
d < M for innitely many ns, prove
that f is integrable.
7.9. EXERCISES. 125
9. Generalisation of the Lemma of Fatou.
Assume that f, g, f
n
: X R are measurable and
_
X
g
d < +.
If g f
n
-a.e. on X and f = liminf
n+
f
n
-a.e. on X, prove that
_
X
f d liminf
n+
_
X
f
n
d.
10. Let f, f
n
: X [0, +] be measurable with f
n
f -a.e. on X and
_
X
f
1
d < +. Prove that
_
X
f
n
d
_
X
f d.
11. Use either the Lemma of Fatou or the Series Theorem 7.2 to prove the
Monotone Convergence Theorem.
12. Generalisation of the Dominated Convergence Theorem.
Let f, f
n
: X R or C, g, g
n
: X [0, +] be measurable. If
[f
n
[ g
n
-a.e. on X, if
_
X
g
n
d
_
X
g d < + and if f
n
f -a.e.
on X and g
n
g -a.e. on X, prove that
_
X
f
n
d
_
X
f d.
13. Assume that all f, f
n
: X [0, +] are measurable, f
n
f -a.e.
on X and
_
X
f
n
d
_
X
f d < +. Prove that
_
A
f
n
d
_
A
f d for
every A .
14. Let f, f
n
: X R or C be integrable and f
n
f -a.e. on X. Prove
that
_
X
[f
n
f[ d 0 if and only if
_
X
[f
n
[ d
_
X
[f[ d.
15. Improper Integrals.
Let f : [a, b) R, where < a < b +. If f is Riemann-integrable
over [a, c] for every c (a, b) and the limit lim
cb
_
c
a
f(x) dx exists in
R, we say that the improper integral of f over [a, b) exists and we
dene it as
_
b
a
f(x) dx = lim
cb
_
c
a
f(x) dx.
We have a similar terminology and denition for
_
b
a
f(x) dx, the im-
proper integral of f over (a, b].
(i) Let f : [a, b) [0, +) be Riemann-integrable over [a, c] for every
c (a, b). Prove that the Lebesgue-integral
_
b
a
f(x) dx and the improper
integral
_
b
a
f(x) dx both exist and they are equal.
(ii) Let f : [a, b) R be Riemann-integrable over [a, c] for every c (a, b).
Prove that, if the Lebesgue-integral
_
b
a
f(x) dx exists, then
_
b
a
f(x) dx
also exists and the two integrals are equal.
(iii) Prove that the converse of (ii) is not true in general. Look at the
fourth function in exercise 7.9.17.
(iv) If
_
b
a
[f(x)[ dx < + (we say that the improper integral is ab-
solutely convergent), prove that the
_
b
a
f(x) dx exists and is a real
number (we say that the improper integral is convergent.)
16. Using improper integrals (see exercise 7.9.15), nd the Lebesgue-integral
_
+
f(x) dx in the following cases:

1
1 +x
2
, e
|x|
,
1
x
2
[0,+)
,
1
x
,
1
[x[
,
1
_
[x[
[1,1]
.
17. Using improper integrals (see exercise 7.9.15), nd the Lebesgue-integral
_
+
f(x) dx in the following cases:

+
n=1
1
2
n
(n,n+1]
,
+
n=1
(1)
n+1
2
n

(n,n+1]
,
+
n=1
1
n
(n,n+1]
,
+
n=1
(1)
n+1
n

(n,n+1]
.
18. Apply the Lemma of Fatou for Lebesgue-measure on R and the sequences:
(n,n+1)
,
(n,+)
, n
0,
1
n
)
, 1 +sign
_
sin(2
n
x
2
)
_
.
19. Let f : [1, 1] Cbe integrable. Prove that lim
n+
_
1
1
x
n
f(x) dx = 0.
20. The discontinuous factor.
Prove that
lim
t+
1
_
+
a
t
1 +t
2
x
2
dx =
_
0, if 0 < a < +
1
2
, if a = 0
1, if < a < 0
21. Prove that
lim
n+
_
n
0
_
1 +
x
n
)
n
e
x
dx =
_
1
1
, if 1 <
+, if 1
22. Let f : X [0, +] be measurable with 0 < c =
_
X
f d < +.
Prove that
lim
n+
n
_
X
log
_
1 +
_
f
n
_
_
d =
_
+, if 0 < < 1
c, if = 1
0, if 1 < < +.
23. Consider the set A = Q [0, 1] = r
1
, r
2
, . . . and a sequence a
n
of real
numbers so that
+
n=1
[a
n
[ < +. Prove that the series
+
n=1
a
n
_
[x r
n
[
converges absolutely for m
1
-a.e. x [0, 1].
7.9. EXERCISES. 127
24. The measure induced by a function.
Let f : X [0, +] be measurable. Dene : [0, +] by
(E) =
_
E
f d
for all E . Prove that is a measure on (X, ) which is called the
measure induced by f. Prove that
(i)
_
X
g d =
_
X
gf d for every measurable g : X [0, +],
(ii) if g : X R is measurable, then
_
X
g d exists if and only if
_
X
gf d exists and in such a case the equality of (i) is true,
(iii) if g : X C is measurable, then g is integrable with respect to
if and only if gf is integrable with respect to and in such a case the
equality of (i) is true.
25. Assume that f : X R or C is integrable and prove that for every > 0
there is an E with (E) < + and
_
E
c
[f[ d < .
26. Absolute continuity of the integral of f.
Let f : X R or C be integrable. Prove that for every > 0 there is
> 0 so that: [
_
E
f d[ < for all E with (E) < .
(Hint: One may prove it rst for simple functions and then use the ap-
proximation theorem.)
27. Let f : R R or C be Lebesgue-integrable. Prove F(x) =
_
x
f(t) dt is
a continuous function of x on R.
28. Continuity of translations.
Assume that f : R
n
R or C is Lebesgue-integrable. Prove that
lim
R
n
h0
_
R
n
[f(x h) f(x)[ dx = 0.
(Hint: Prove it rst for continuous functions which are 0 outside a bounded
set and then use the approximation theorem.)
29. The Riemann-Lebesgue Lemma.
Assume that f : R R or C is Lebesgue-integrable. Prove that
lim
x+
_
+
f(t) cos(xt) dt = lim

x+
_
+
f(t) sin(xt) dt = 0
in two ways.
Prove the limits when f is the characteristic function of any interval and
then use an approximation theorem.
Prove that [
_
+
f(t) cos(xt) dt[ =

1
2
[
_
+
(f(t

x
) f(t)) cos(xt) dt[
1
2
_
+
[f(t

x
) f(t)[ dt and then use the result of exercise 7.9.28.
30. Let Q R
n
be a closed interval and x
0
Q. If f : Q R is Riemann-
integrable over Q and g : Q R coincides with f on Q x
0
, prove that
g is also Riemann-integrable over Q and that (
n
)
_
Q
g = (
n
)
_
Q
f.
31. Let Q R
n
be a closed interval, R and f, g : Q R be Riemann-
integrable over Q. Prove that f +g, f and fg are all Riemann-integrable
over Q and
(
n
)
_
Q
(f +g) = (
n
)
_
Q
f + (
n
)
_
Q
g, (
n
)
_
Q
f = (
n
)
_
Q
f.
32. Let Q R
n
be a closed interval.
(i) If the bounded functions f, f
k
: Q R are all Riemann-integrable over
Q and 0 f
k
f on Q, prove that (
n
)
_
Q
f
k
(
n
)
_
Q
f.
(ii) Find bounded functions f, f
k
: Q R so that 0 f
k
f on Q and so
that all f
k
are Riemann-integrable over Q, but f is not Riemann-integrable
over Q.
33. Continuity of an integral as a function of a parameter.
Let f : X (a, b) R and g : X [0, +] be such that
(i) g is integrable and, for every t (a, b), f(, t) is measurable,
(ii) for -a.e. x X, f(x, ) is continuous on (a, b),
(iii) for every t (a, b), [f(, t)[ g -a.e. on X.
Prove that F(t) =
_
X
f(, t) d is continuous as a function of t on (a, b).
34. Dierentiability of an integral as a function of a parameter.
Let f : X (a, b) R and g : X [0, +] be such that
(i) g is integrable and, for every t (a, b), f(, t) is measurable,
(ii) for at least one t
0
(a, b), f(, t
0
) is integrable,
(iii) for -a.e. x X, f(x, ) is dierentiable on (a, b) and [
df
dt
(x, t)[ g(x)
for every t (a, b).
Prove that F(t) =
_
X
f(, t) d is dierentiable as a function of t on (a, b)
and that
dF
dt
(t) =
_
X
df
dt
(, t) d, a < t < b,
where
df
dt
: A (a, b) R for some A with (X A) = 0.
35. The integral of Gauss.
Consider the functions f, h : [0, +) R dened by
f(x) =
1
2
_
_
x
0
e
1
2
t
2
dt
_
2
, h(x) =
_
1
0
e
1
2
x
2
(t
2
+1)
t
2
+ 1
dt.
(i) Prove that f
(x) + h
(x) = 0 for every x (0, +) and, hence, that

f(x) +h(x) =

4
for every x [0, +).
(ii) Prove that
_
+
1
2
t
2
dt =
2.
7.9. EXERCISES. 129
36. The distribution (or measure) of Gauss.
Consider the function g : R R dened by
g(x) =
1
2
_
x
1
2
t
2
dt.
(i) Prove that g is continuous, strictly increasing, with g() = 0 and
g(+) = 1 and with continuous derivative g
(x) =
1
2
e
1
2
x
2
, x R.
(ii) The Lebesgue-Stieltjes measure
g
induced by g is called the distri-
bution or the measure of Gauss. Prove that
g
(R) = 1, that
g
(E) =
1
2
_
E
e
1
2
x
2
dx
for every Borel set in R and that
_
R
f d
g
=
1
2
_
+
f(x)e
1
2
x
2
dx
for every Borel-measurable f : R R or C for which either of the two
integrals exists.
37. (i) Prove that the function F : (0, +) R dened by
F(t) =
_
+
0
e
tx
sinx
x
dx
is dierentiable on (0, +) and that
dF
dt
(t) =
1
1+t
2
for every t > 0. Find
the lim
t+
F(t) and conclude that F(t) = arctan
1
t
for every t > 0.
(ii) Prove that the function
sin x
x
is not Lebesgue-integrable over (0, +).
(iii) Prove that the improper integral
_
+
0
sin x
x
dx exists.
(iv) Justify the equality lim
t0+
F(t) =
_
+
0
sin x
x
dx.
(v) Conclude that
_
+
0
sin x
x
dx =

2
.
(vi) Prove that
lim
t+
1
_
+
a
sin(tx)
x
dx =
_
0, if 0 < a < +
1
2
, if a = 0
1, if < a < 0
38. The gamma-function.
Let H
+
= s = x+iy C[ x > 0 and consider the function : H
+
C
dened by
(s) =
_
+
0
t
s1
e
t
dt.
This is called the gamma-function.
(i) Prove that this Lebesgue-integral exists and is nite for every s H
+
.
(ii) Prove that
x
(s) = i
y
(s)
for every s H
+
. This means that is holomorphic in H
+
.
(iii) Prove that (n) = (n 1)! for every n N.
39. The invariance of Lebesgue-integral and of Lebesgue-measure under isome-
tries.
Let T : R
n
R
n
be an isometric linear transformation. This means that
[T(x) T(y)[ = [x y[ for every x, y R
n
or, equivalently, that TT
=
T
T = I, where T
is the adjoint of T and I is the identity transformation.

Prove that, for every E L
n
, we have m
n
(T(E)) = m
n
(E), and that
_
R
n
f T
1
dm
n
=
_
R
n
f dm
n
for every Lebesgue-measurable f : R
n
R or C, provided that at least
one of the two integrals exists.
40. (i) Consider the Cantors set C and the I
0
= [0, 1], I
1
, I
2
, . . . which were
used for its construction. Prove that the 2
n1
subintervals of I
n1
I
n
,
n N, can be described as
(
a
1
3
+ +
a
n1
3
n1
+
1
3
n
,
a
1
3
+ +
a
n1
3
n1
+
2
3
n
),
where each of a
1
, . . . , a
n1
takes the values 0 and 2.
(ii) Let f be the Cantors function, which was introduced in exercise 4.6.7,
extended as 0 in (, 0) and as 1 in (1, +). Prove that f is constant
f =
a
1
2
2
+ +
a
n1
2
n
+
1
2
n
in the above subinterval (
a1
3
+ +
an1
3
n1
+
1
3
n
,
a1
3
+ +
an1
3
n1
+
2
3
n
).
(iii) If G : (0, 1) R is another function with bounded derivative in (0, 1),
prove that
_
(0,1)
Gd
f
= G(1)
+
n=1
a1,...,an1=0, 2
(
a
1
2
2
+ +
a
n1
2
n
+
1
2
n
)
_
G(
a
1
3
+ +
a
n1
3
n1
+
2
3
n
) G(
a
1
3
+ +
a
n1
3
n1
+
1
3
n
)
_
.
(iv) In particular,
_
(0,1)
xd
f
=
1
2
.
(v) Prove that
_
(0,1)
e
itx
d
f
= e
1
2
it
lim
n+
cos
_
t
3
_
cos
_
t
3
n
_
for every t R.
7.9. EXERCISES. 131
41. Let F, G : R R be increasing and assume that FG is also increasing.
Prove that
GF
(E) =
_
E
G(x+) d
F
+
_
E
F(x) d
G
for every Borel set E R and
_
R
f(x) d
GF
=
_
R
f(x)G(x+) d
F
+
_
R
f(x)F(x) d
G
for every Borel-measurable f : R R or C for which at least two of the
three integrals exist.
42. If F : R R is increasing and continuous and f : R [0, +] is
Borel-measurable, prove that
_
R
f F d
F
=
_
F(+)
F()
f(t) dt.
Show, by example, that this may not be true if F is not continuous.
43. Riemanns criterion for convergence of a series.
Assume F : R [0, +) is increasing and g : (0, +) [0, +) is
decreasing. Let a
n
0 for all n and
n[ a
n
g(x) F(x)
for every x (0, +) and
_
(0,+)
g d
F
< +. Prove that
+
n=1
a
n
<
+.
44. Mean values.
Let f : X R or C be integrable and F be a closed subset of R or C.
If
1
(E)
_
E
f d F for every E with 0 < (E), prove that f(x) F
for -a.e. x X.
45. Let E have -nite -measure. Prove that there is an f : X [0, +]
with
_
X
f d < + and f(x) > 0 for every x E.
46. Let f : X [0, +]. Prove that
1
2
nZ
2
n
f
(2
n
)
_
X
f d
nZ
2
n
f
(2
n
)
and, hence, that f is integrable if and only if the
nZ
2
n
f
(2
n
) is nite.
47. Equidistributed functions.
Let f, g : X [0, +] be measurable. The two functions are called
equidistributed if
f
(t) =
g
(t) for every t [0, +).
Prove that, if f, g are equidistributed, then
_
X
f
p
d =
_
X
g
p
d for every
p (0, +).
48. Let , : X [0, +) be two measurable simple functions and let
=
m
j=1
Ej
and =
n
i=1
Fi
be their standard representations
so that 0 <
1
< <
m
and 0 <
1
< <
n
, where we omit the
possible value 0.
If and are integrable, prove that they are equidistributed (exercise
7.9.47) if and only if m = n,
1
=
1
, . . . ,
m
=
m
and (E
1
) =
(F
1
), . . . , (E
m
) = (F
m
).
49. The inequality of Chebychev.
If f : X [0, +] is measurable, prove that
(x X[ t < f(x)) =
f
(t)
1
t
_
X
f d
for every t (0, +). Prove also that, if f is integrable, then
lim
t+
t
f
(t) = 0.
50. If f : X [0, +] is measurable and 0 < p < +, prove that
_
X
f
p
d = p
_
+
0
t
p1
f
(t) dt.
If, also, f < + -a.e. on X, prove that
_
X
f
p
d =
_
[0,+)
t
p
d
F
f
,
where F
f
is dened in Proposition 7.15.
51. The Jordan-content of sets in R
n
.
If E R
n
is bounded we dene its inner Jordan-content
c
(i)
n
(E) = sup
m
j=1
vol
n
(R
j
) [ m N, E
1
, . . . , E
m
pairwise disjoint
open intervals with
m
j=1
R
j
E
and its outer Jordan-content
c
(o)
n
(E) = inf
m
j=1
vol
n
(R
j
) [ m N, E
1
, . . . , E
m
open intervals
with
m
j=1
R
j
E.
(i) Prove that the values of c
(i)
n
(E) and c
(o)
n
(E) remain the same if in the
above denitions we use closed intervals instead of open intervals.
(ii) Prove that c
(i)
n
(E) c
(o)
n
(E) for every bounded E R
n
.
7.9. EXERCISES. 133
The bounded E is called Jordan-measurable if c
(i)
n
(E) = c
(o)
n
(E), and
the value
c
n
(E) = c
(i)
n
(E) = c
(o)
n
(E)
is called the Jordan-content of E.
(iii) If E is bounded and c
(o)
n
(E) = 0, prove that E is Jordan-measurable.
(iv) Prove that all intervals S are Jordan-measurable and c
n
(S) = vol
n
(S).
(v) If E is bounded, prove that it is Jordan-measurable if and only if for
every > 0 there exist pairwise disjoint open intervals R
1
, . . . , R
m
and
open intervals R
1
, . . . , R
k
so that
m
j=1
R
j
E
k
i=1
R
i
and
k
i=1
vol
n
(R
i
)
m
j=1
vol
n
(R
j
) < .
(vi) If E is bounded, prove that E is Jordan-measurable if and only if
c
(o)
n
(E) = 0.
(vii) Prove that the collection of bounded Jordan-neasurable sets is closed
under nite unions and set-theoretic dierences. Moreover, if E
1
, . . . , E
l
are pairwise disjoint Jordan-measurable sets, prove that
c
n
(E) =
l
j=1
c
n
(E
j
).
(viii) Prove that if the bounded set E is closed, then m
n
(E) = 0 implies
c
n
(E) = 0. If E is not closed, then this result may not be true. For
example, if E = Q [0, 1] R, then m
1
(E) = 0, but c
(i)
1
(E) = 0 < 1 =
c
(o)
1
(E) and, hence, E is not Jordan-measurable. (See exercise 4.6.4.)
(ix) If the bounded set E is Jordan-measurable, prove that it is Lebesgue-
measurable and
m
n
(E) = c
n
(E).
(x) Let E be bounded and take any closed interval Q so that E Q. Prove
that E is Jordan-measurable if and only if
E
is Riemann-integrable over
Q and that, in this case,
c
n
(E) = (
n
)
_
Q
E
.
(xi) Let Q be a closed interval, f, g : Q R be bounded and E Q
be Jordan-measurable with c
n
(E) = 0. If f is Riemann-integrable over Q
and f = g on Q E, prove that g is also Riemann-integrable over Q and
that (
n
)
_
Q
f = (
n
)
_
Q
g.
52. Lebesgues characterisation of Riemann-integrable functions.
Let Q R
n
be a closed interval and f : Q R be bounded. Prove that
f is Riemann-integrable if and only if x Q[ f is discontinuous at x is
a m
n
-null set.
Chapter 8
Product-measures
8.1 Product--algebra.
If I is a general set of indices, the elements of the cartesian product
iI
X
i
are all functions x : I
iI
X
i
with the property: x(i) X
i
for every i I. It
is customary to use the notation x
i
, instead of x(i), for the value of x at i I
and, accordingly, to use the notation (x
i
)
iI
for x
iI
X
i
.
If I is a nite set, I = 1, . . . , n, besides writing x = (x
i
)
iI
, we also
use the traditional notation x = (x
1
, . . . , x
n
) for the elements of
iI
X
i
=
n
i=1
X
i
= X
1
X
n
. And if I is countable, say I = N = 1, 2, . . ., we
write x = (x
1
, x
2
, . . .) for the elements of
iI
X
i
=
+
i=1
X
i
= X
1
X
2
.
Denition 8.1 If I is a set of indices, then, for every j I, the function
j
:
iI
X
i
X
j
dened by
j
(x) = x
j
for all x = (x
i
)
iI

iI
X
i
, is called the j-th projection of
iI
X
i
or the
projection of
iI
X
i
onto its j-th component X
j
.
In case I = 1, . . . , n or I = N, the formula of the j-th projection is
j
(x) = x
j
for all x = (x
1
, . . . , x
n
) X
1
X
n
or, respectively, x = (x
1
, x
2
, . . .)
X
1
X
2
=
+
i=1
X
i
.
It is clear that the inverse image
1
j
(A
j
) = x

iI
X
i
[ x
j
A
j
of an
arbitrary A
j
X
j
is the cartesian product
1
j
(A
j
) =
iI
Y
i
, where Y
i
=
_
X
i
, if i ,= j
A
j
, if i = j
In particular, if I = 1, . . . , n, then
1
j
(A
j
) = X
1
X
j1
A
j
X
j+1
X
n
135
136 CHAPTER 8. PRODUCT-MEASURES
and, if I = N, then
1
j
(A
j
) = X
1
X
j1
A
j
X
j+1
.
Denition 8.2 If (X
i
,
i
) is a measurable space for every i I, we consider
the -algebra of subsets of the cartesian product
iI
X
i
iI
i
=
_
1
j
(A
j
) [ j I, A
j

j
_
,
called the product--algebra of
i
, i I.
In particular,
n
i=1

i
is generated by the collection of all sets of the form
X
1
X
j1
A
j
X
j+1
X
n
, where 1 j n and A
j

j
.
Similarly,
+
i=1

i
is generated by the collection of all sets of the form
X
1
X
j1
A
j
X
j+1
, where j N and A
j

j
.
Proposition 8.1 Let (X
i
,
i
) be a measurable space for each i I. Then
iI

i
is the smallest -algebra of subsets of
iI
X
i
for which all projec-
tions
j
:
iI
X
i
X
j
are (,
j
)measurable.
Proof: For every j and every A
j

j
we have that
1
j
(A
j
)

iI

i
and,
hence, every
j
is (
iI

i
,
j
)measurable.
Now, let be a -algebra of subsets of
iI
X
i
for which all projections
j
:
iI
X
i
X
j
are (,
j
)measurable. Then for every j and every A
j

j
we have that
1
j
(A
j
) . This implies that
1
j
(A
j
) [ j I, A
j

j

and, hence,
iI

i
.
i
,
i
) be a measurable space for each i I. If c
i
is a
collection of subsets of X
i
with
i
= (c
i
) for all i I, then
iI

i
= (c),
where
c =
1
j
(E
j
) [ j I, E
j
c
j
.
Proof: Since c
1
j
(A
j
) [ j I, A
j

j

iI

i
, it is immediate that
(c)
iI

i
.
We, now, x j I and consider the
j
:
iI
X
i
X
j
. We have that
1
j
(E
j
) c (c) for every E
j
c
j
. Proposition 6.1 implies that
j
is
((c),
j
)measurable and, since j is arbitrary, Proposition 8.1 implies that
iI

i
(c).
i
,
i
) be measurable spaces. If c
i
is a collection of
subsets of X
i
so that
i
= (c
i
) for every i I, then
iI

i
= (
c), where
c =
iI
E
i
[ E
i
,= X
i
for at most countably many i I and E
i
c
i
if E
i
,= X
i
.
Proof: We observe that
1
j
(E
j
)

c for every j I and every E
j
c
j
and,
hence, c

c (
c). This implies (c) (
c).
8.1. PRODUCT--ALGEBRA. 137
Now take any
iI
E
i

c. We set i
1
, i
2
, . . . = i I [ E
i
,= X
i
and
observe that
iI
E
i
=
+
n=1
1
in
(E
in
) (c).
Thus,

c (c) and, hence, (
c) (c). Proposition 8.2 nishes the proof.

In particular,
n
i=1

i
is generated by the collection of all cartesian products
of the form E
1
E
n
, where E
j
c
j
for all j = 1, . . . , n.
Also,
+
i=1

i
is generated by the collection of all cartesian products of the
form E
1
E
2
, where E
j
c
j
for all j N.
Example
If we consider R
n
=
n
i=1
R and, for each copy of R, we take the collection of
all open-closed 1-dimensional intervals as a generator of B
R
, then Proposition
8.3 implies that the collection of all open-closed n-dimensional intervals is a
generator of
n
i=1
B
R
. But we already know that the same collection is a
generator of B
R
n. Therefore,
B
R
n =
n
i=1
B
R
.
This can be generalised. If n
1
+ +n
k
= n, we formally identify the typical
element (x
1
, . . . , x
n
) R
n
with
_
(x
1
, . . . , x
n1
), . . . , (x
n1++n
k1
+1
, . . . , x
n1++n
k
)
_
,
i.e. with the typical element of
k
j=1
R
nj
. We thus identify
R
n
=
k
j=1
R
nj
.
Now,
k
j=1
B
R
n
j is generated by the collection of all products
k
j=1
A
j
, where
each A
j
is an n
j
-dimensional open-closed interval. This means that
k
j=1
A
j
is,
by the same identication, the typical n-dimensional open-closed interval and,
hence,
k
j=1
B
R
n
j is generated by the collection of all open-closed intervals in
R
n
. Therefore,
B
R
n =
k
j=1
B
R
n
j .
If x
iI
X
i
and J I, we denote, as usual, x
J

iJ
X
i
the restriction
of x on J. Then x
J
c

iJ
c
X
i
is the restriction of x on J
c
and, obviously,
x uniquely determines both x
J
and x
J
c . Conversely, x is uniquely determined
by its restrictions x
J
and x
J
c . Namely, if y

iJ
X
i
and z

iJ
c
X
i
are
given, then there is a unique x
iI
X
i
so that x
J
= y and x
J
c = z. We just
dene x
i
= y
i
, if i J, and x
i
= z
i
, if i J
c
.
This produces an identication between the product spaces
iI
X
i
and
(
iJ
X
i
) (
iJ
c
X
i
). We identify the element x of the rst space with the
pair (y, z) of the second space, whenever y = x
J
and z = x
J
c . Or, in the same
context, we may identify
iI
X
i
and (
iJ
c X
i
) (
iJ
X
i
), by identifying
x with the pair (z, y).
In case I = 1, . . . , n and J = i
1
, . . . , i
m
with 1 i
1
< < i
m
n,
we prefer the vector notation and write x
J
= (x
i1
, . . . , x
im
). For example, if
x = (x
1
, x
2
, x
3
, x
4
, x
5
), then x
{1,3,5}
= (x
1
, x
3
, x
5
) and x
{2,4}
= (x
2
, x
4
). It
is obvious that x = (x
1
, x
2
, x
3
, x
4
, x
5
) uniquely determines and is uniquely de-
termined by the restrictions y = (x
1
, x
3
, x
5
) and z = (x
2
, x
4
) and we identify
x = (x
1
, x
2
, x
3
, x
4
, x
5
) with (y, z) =
_
(x
1
, x
3
, x
5
), (x
2
, x
4
)
_
. It must be stressed
that these are formal identications (logically supported by the underlying bi-
jections) and not actual equalities.
Denition 8.3 Let E

iI
X
i
and J I. For every z

iJ
c X
i
, we
dene
E
z
= y
iJ
X
i
[ x E, where x
J
= y, x
J
c = z = y
iJ
X
i
[ (y, z) E
and call it the z-section of E.
It is clear that every z-section of E is a subset of
iJ
X
i
.
We have E
(x2,x4)
= (x
1
, x
3
, x
5
) [ (x
1
, x
2
, x
3
, x
4
, x
5
) E X
1
X
3
X
5
for the simple example before Denition 8.3.
Denition 8.4 Let f :
iI
X
i
Y and J I. For every z
iJ
c X
i
, we
dene f
z
:
iJ
X
i
Y by the formula
f
z
(y) = f(x) = f(y, z), where x
J
= y and x
J
c = z,
and call it the z-section of f.
For example, f
(x2,x4)
(x
1
, x
3
, x
5
) = f(x
1
, x
2
, x
3
, x
4
, x
5
).
In the case where J
c
= i
0
is a one-point set, for simplicity we prefer to
write E
xi
0
and f
xi
0
, instead of E
(xi
0
)
and f
(xi
0
)
.
Theorem 8.1 Let (X
i
,
i
) be a measurable space for every i I and consider
J I. If a set E

iI
X
i
is
iI

i
measurable, then E
z

iJ
X
i
is
iJ

i
measurable for every z
iJ
c X
i
.
Proof: Consider the collection of all E
iI
X
i
with the property that E
z
is
iJ

i
measurable for every z
iJ
c
X
i
.
Clearly, belongs to .
If E , then (E
c
)
z
= (E
z
)
c
is
iJ

i
measurable for all z
iJ
c X
i
and, hence, E
c
.
8.1. PRODUCT--ALGEBRA. 139
If E
1
, E
2
, . . . , then (
+
n=1
E
n
)
z
=
+
n=1
(E
n
)
z
is
iJ

i
measurable for
all z
iJ
c
X
i
and, hence,
+
n=1
E
n
.
Therefore, is a -algebra of subsets of
iI
X
i
. Consider, now, an arbi-
trary i
0
I and an arbitrary A
i0

i0
. We observe that, if i
0
J, then
(
1
i0
(A
i0
))
z
=
iJ
Y
i
, where Y
i
=
_
A
i0
, if i = i
0
X
i
, if i J i
0
and, if i
0
J
c
, then
(
1
i0
(A
i0
))
z
=
_
iJ
X
i
, if z
i0
A
i0
, if z
i0
/ A
i0
.
In both cases we have that (
1
i0
(A
i0
))
z
is
iJ

i
measurable for every
z

iJ
c X
i
and, hence,
1
i0
(A
i0
) . Since i
0
and A
i0
are arbitrary, by
Denition 8.2 we have that
iI

i
. This says that, if E
iI
X
i
is
iI

i
measurable, then E and, hence, E
z
is
iJ

i
measurable for
every z
iJ
c X
i
.
Theorem 8.2 Let (Y,
), (X
i
,
i
) be measurable spaces for every i I and
consider J I.
If f :
iI
X
i
Y is (
iI

i
,
)measurable, then f
z
:
iJ
X
i
Y is
(
iJ

i
,
)measurable for every z
iJ
c
X
i
.
Proof: Take an arbitrary E
. Then (f
z
)
1
(E) = (f
1
(E))
z
for every
z
iJ
c
X
i
.
If f is (
iI

i
,
)measurable, then f
1
(E)
iI

i
and, by Theorem
8.1, (f
z
)
1
(E) = (f
1
(E))
z

iJ

i
. Since E is arbitrary, we have that f
z
is (
iJ

i
,
)measurable.
The last two theorems say, in informal language, that sets or functions which
are measurable in a product space have all their sections measurable in the ap-
propriate product subspaces.
The converse in not true in general.
Examples
1. Let us consider R
n
=
n
i=1
R, where I = 1, . . . , n, and take a J =
i
1
, . . . , i
m
with 1 i
1
< < i
m
n. We write J
c
= i
1
, . . . , i
nm
with
1 i
1
< < i
nm
n.
We, naturally, identify
iJ
R with R
m
, by writing y = x
J
= (x
i1
, . . . , x
im
)
as y = (y
1
, . . . , y
m
). We similarly identify
iJ
c R with R
nm
, by writing
z = x
J
c = (x
i
1
, . . . , x
i
nm
) as z = (z
1
, . . . , z
nm
).
Therefore,
iJ
B
R
= B
R
m and
iJ
c B
R
= B
R
nm.
Now, if E is a Borel set in R
n
, then, for arbitrary z
iJ
c R = R
nm
,
the z-section E
z
of E is a Borel set in R
m
.
2. Take any A R which is not a Borel set in R and consider the set
E = (x
1
, x
2
) R
2
[ x
1
= x
2
A.
Clearly, all 1-dimensional sections of E are either empty or one-point sets
and, hence, are Borel sets in R. We shall see that E is not a Borel set in R
2
.
Indeed, assume that E is a Borel set in R
2
and consider the invertible linear
transformation T : R
2
R
2
given by the formula
T(x
1
, x
2
) =
_
x
1
+x
2
2
,
x
1
x
2
2
_
.
Then T(E) = (x
1
, 0) [ x
1
A is a Borel set in R
2
and, hence, all 1-
dimensional sections of T(E) must be Borel sets in R. In particular, the (hori-
zontal) section x
1
[ x
1
A = A must be a Borel set in R and we, thus, arrive
at a contradiction.
8.2 Product-measure.
In this section we shall limit ourselves to cartesian products of nitely many
spaces. We x the measure spaces (X
1
,
1
,
1
), . . . , (X
n
,
n
,
n
) and the mea-
surable space (
n
j=1
X
j
,
n
j=1

j
).
From Proposition 8.3 and the paragraph after it, we know that
n
j=1

j
is
generated by the collection

c of all sets of the form
n
j=1
A
j
, where A
j

j
for all j. Since
n
j=1
X
j
belongs to

c and =
n
j=1

c, we obviously have
that this collection is a -covering collection for
n
j=1
X
j
.
The elements of

c play the same role that open-closed intervals play for the
introduction of Lebesgue-measure on R
n
. We agree to call these sets measur-
able intervals in
n
j=1
X
j
, a term which will be justied by Theorem 8.3, and
denote them by
R =
n
j=1
A
j
.
j
,
j
) be a measurable space for every j = 1, . . . , n.
The collection
/ =
R
1

R
m
[ m N,

R
1
, . . . ,

R
m
pairwise disjoint elements of

c
is an algebra of subsets of
n
j=1
X
j
.
Proof: If

R =
n
j=1
A
j
and

R
n
j=1
B
j
are elements of

c, then

R

R
n
j=1
(A
j
B
j
) is an element of

c.
Moreover, if

R =
n
j=1
A
j
is an element of

c, then
R
c
= A
c
1
A
2
A
n

X
1
X
2
X
j1
A
c
j
A
j+1
A
n

X
1
X
2
X
n1
A
c
n
8.2. PRODUCT-MEASURE. 141
is a disjoint union of elements of

c, i.e. an element of /.
Now, if

R
1

R
m
and

R
1

R
k
are any two elements of /, then
(

R
1

R
m
) (

R
1

R
k
) =
1jm,1ik
(

R
j

R
i
), is, by the result
of the rst paragraph, also an element of /. Hence, / is closed under nite
intersections. Also, if

R
1

R
m
is an element of /, then (

R
1

R
m
)
c
=
R
c
1

R
c
m
is, by the result of the second paragraph, a nite intersection of
elements of / and, hence, an element of /.
Therefore, / is closed under nite intersections and under complements.
This implies that it is an algebra of subsets of
n
j=1
X
j
.
For each

R =
n
j=1
A
j

c, we dene the quantity
(

R) =
n
j=1
j
(A
j
),
which plays the role of volume of the measurable interval

R.
Denition 8.5 Let (X
j
,
j
,
j
) be a measure space for every j = 1, . . . , n. For
every E
n
j=1
X
j
we dene
(E) = inf
_
+
i=1
(

R
i
) [

R
i

c for all i and E
+
i=1
R
i
_
Theorem 3.2 implies that the function
: T
_
n
j=1
X
j
_
[0, +] is an
outer measure on
n
j=1
X
j
.
j
,
j
,
j
) be a measure space for every j = 1, . . . , n
and

R,

R
i
be measurable intervals for every i N.
(i) If

R
+
i=1
R
i
, then (

R)
+
i=1
(

R
i
).
(ii) If

R =
+
i=1
R
i
and all

R
i
are pairwise disjoint, then (

R) =
+
i=1
(

R
i
).
Proof: (i) Let

R =
n
j=1
A
j
and

R
i
=
n
j=1
A
i
j
, where A
j
, A
i
j

j
for every
i N and j with 1 j n.
From
n
j=1
A
j

+
i=1
_
n
j=1
A
i
j
_
, we get that
n
j=1
Aj
(x
j
) =
n
j=1
Aj
(x
1
, . . . , x
n
)
i=1
n
j=1
A
i
j
(x
1
, . . . , x
n
) =
+
i=1
_
n
j=1
A
i
j
(x
j
)
_
for every x
1
X
1
, . . . , x
n
X
n
. Integrating over X
1
with respect to
1
, we nd
1
(A
1
)
n
j=2
Aj
(x
j
)
+
i=1
_
1
(A
i
1
)
n
j=2
A
i
j
(x
j
)
_
for every x
2
X
2
, . . . , x
n
X
n
. Integrating over X
2
with respect to
2
, we get
1
(A
1
)
2
(A
2
)
n
j=3
Aj
(x
j
)
+
i=1
_
1
(A
i
1
)
2
(A
i
2
)
n
j=3
A
i
j
(x
j
)
_
for every x
3
X
3
, . . . , x
n
X
n
. We continue until we have integrated all
variables.
(ii) We use equalities everywhere in the above calculations.
The next result justies the term measurable interval for each

R

c.
Theorem 8.3 Let (X
i
,
i
,
i
) be a measure space for every i = 1, . . . , n and
the outer measure of Denition 8.5. Every measurable interval

R =
n
j=1
A
j
is
-measurable and
(

R) = (

R) =
n
j=1
j
(A
j
).
Also,
n
j=1

j
is included in the -algebra of
-measurable subsets of
n
j=1
X
j
.
Proof: (a) If

R is a measurable interval, then

R

c and, from

R

R, we
obviously get
(

R) (

R).
Proposition 8.5 implies (

R)
+
i=1
(

R
i
) for every covering

R
+
i=1
R
i
with

R
i

c for all i N. Hence, (

R)
(

R) and we conclude that
(

R) = (

R).
(b) We take any two measurable intervals

R,

R
and Proposition 8.4 implies that

there are pairwise disjoint measurable intervals

R
1
, . . . ,

R
m
so that

R

R =
R
1

R
m
. By the subadditivity of
, the result of (a) and Proposition 8.5,
(

R

R) +
(

R

R)
(

R

R) +
(

R
1
) + +
(

R
n
)
= (

R

R) +(

R
1
) + +(

R
n
)
= (

R
).
(c) Let

R

c and consider an arbitrary E
n
j=1
X
j
with
(E) < +. For

any > 0 we consider a covering E
+
i=1
R
i
with

R
i

c for all i N, such
that
+
i=1
(

R
i
) <
(E) +. By the result of (b) and the subadditivity of
(E

R) +
(E

R)
+
i=1
_
(

R
i

R) +
(

R
i

R)
_
i=1
(

R
i
) <
(E) +.
Since is arbitrary,
(E

R) +
(E

R)
(E) and we conclude that

R is
-measurable.
Since
n
j=1

j
is generated by the collection of all measurable intervals, it
is included in the -algebra of all
-measurable sets.
Denition 8.6 Let (X
i
,
i
,
i
) be a measure space for each i = 1, . . . , n and
be the outer measure of Denition 8.5. The measure induced from
by
Theorem 3.1 is called the product-measure of
j
, 1 j n, and it is
denoted
n
j=1
j
.
We denote by
n
j=1
j
the -algebra of
-measurable subsets of
n
j=1
X
j
.
Therefore, (
n
j=1
X
j
,
n
j=1
j
,
n
j=1
j
) is a complete measure-space.
Theorem 8.3 implies that
n
j=1

j

n
j=1
j
and
(
n
j=1
j
)
_
n
j=1
A
j
_
=
n
j=1
j
(A
j
)
for every A
1

1
, . . . , A
n

n
.
It is very common to consider the restriction, also denoted by
n
j=1
j
, of
n
j=1
j
on
n
j=1

j
.
Theorem 8.4 Let (X
i
,
i
,
i
) be a measure space for each i = 1, . . . , n. If
1
, . . . ,
n
are -nite measures, then
(i)
n
j=1
j
is the unique measure on (
n
j=1
X
j
,
n
j=1

j
) with the property:
(
n
j=1
j
)
_
n
j=1
A
j
_
=
n
j=1
j
(A
j
) for every A
1

1
, . . . , A
n

n
and
(ii) the measure space (
n
j=1
X
j
,
n
j=1
j
,
n
j=1
j
) is the completion of the mea-
sure space (
n
j=1
X
j
,
n
j=1

j
,
n
j=1
j
).
Proof: (i) We consider the algebra / of subsets of
n
j=1
X
j
described in Propo-
sition 8.4. If is any measure on (
n
j=1
X
j
,
n
j=1

j
) such that (

R) =
(
n
j=1
j
)(

R) for every

R

c, then, by additivity of the measures, we have
that (

R
1

R
m
) = (
n
j=1
j
)(

R
1

R
m
) for all pairwise disjoint
R
1
, . . . ,

R
m

c. Therefore, the measures and
n
j=1
j
are equal on /.
Since all measures
j
are -nite, there exist A
i
j

j
with
j
(A
i
j
) < +
for every i, j and A
i
j
X
j
for every j. This implies that the measurable in-
tervals

S
i
=
n
j=1
A
i
j
have the property that

S
i

n
j=1
X
j
and that (
S
i
) =
(
n
j=1
j
)(
S
i
) =
n
j=1
j
(A
i
j
) < + for every i.
Since
n
j=1

j
= (
c) = (/), Theorem 2.4 implies that and

n
j=1
j
are
equal on
n
j=1

j
.
(ii) We already know that (
n
j=1
X
j
,
n
j=1
j
,
n
j=1
j
) is a complete exten-
sion of (
n
j=1
X
j
,
n
j=1

j
,
n
j=1
j
). Therefore, it is also an extension of the
completion (
n
j=1
X
j
,
n
j=1

j
,
n
j=1
j
) and it is enough to prove that every
E
n
j=1
j
belongs to
n
j=1

j
.
Take any E
n
j=1
j
and assume, at rst, that (
n
j=1
j
)(E) < +.
We take arbitrary k N and we nd a covering E
+
i=1
R
k
i
by pair-
wise disjoint measurable intervals so that
+
i=1
(

R
k
i
) < (
n
j=1
j
)(E) +
1
k
. We
dene B
k
=
+
i=1
R
k
i

n
j=1

j
and have that E B
k
and (
n
j=1
j
)(E)
(
n
j=1
j
)(B
k
) < (
n
j=1
j
)(E)+
1
k
. Now, dene A =
+
k=1
B
k

n
j=1

j
. Then
E A and (
n
j=1
j
)(E) = (
n
j=1
j
)(A). Therefore (
n
j=1
j
)(A E) = 0.
In case (
n
j=1
j
)(E) = +, we consider the specic sets

S
i
, which were
constructed in the proof of part (i), and take the sets E
i
= E

S
i
. These
sets have (
n
j=1
j
)(E
i
) < + and, by the previous paragraph, we can nd
A
i

n
j=1

j
so that E
i
A
i
and (
n
j=1
j
)(A
i
E
i
) = 0. We dene A =
+
i=1
A
i

n
j=1

j
so that E A and, since A E

+
i=1
(A
i
E
i
), we
conclude that (
n
j=1
j
)(A E) = 0.
We have proved that for every E
n
j=1
j
there exists A
n
j=1

j
so
that E A and (
n
j=1
j
)(A E) = 0.
Considering AE instead of E, we nd a set B
n
j=1

j
so that AE B
and (
n
j=1
j
)
_
B (A E)
_
= 0. Of course, (
n
j=1
j
)(B) = 0.
Now we observe that E = (A B) (E B), where A B
n
j=1

j
and
E B B
n
j=1

j
with (
n
j=1
j
)(B) = 0. This says that E
n
j=1

j
.
We shall examine, now, the inuence to the product-measure of replacing
the measure spaces (X
j
,
j
,
j
) by their completions (X
j
,
j
,
j
).
Theorem 8.5 Let (X
j
,
j
,
j
) and (X
j
,
j
,
j
) be a measure space and its com-
pletion for every j = 1, . . . , n.
(i) The measure spaces (X
j
,
j
,
j
) induce the same product-measure space as
their completions (X
j
,
j
,
j
). Namely,
(
n
j=1
X
j
,
n
j=1
j
,
n
j=1
j
) = (
n
j=1
X
j
,
n
j=1
j
,
n
j=1
j
).
Moreover, the above product-measure space is an extension of both measure
spaces (
n
j=1
X
j
,
n
j=1

j
,
n
j=1
j
) and (
n
j=1
X
j
,
n
j=1

j
,
n
j=1
j
), of which
the second is an extension of the rst.
(ii) If each (X
j
,
j
,
j
) is -nite, then (
n
j=1
X
j
,
n
j=1
j
,
n
j=1
j
) is the com-
pletion of both (
n
j=1
X
j
,
n
j=1

j
,
n
j=1
j
) and (
n
j=1
X
j
,
n
j=1

j
,
n
j=1
j
).
Proof: (i) To construct the product-measure space (
n
j=1
X
j
,
n
j=1
j
,
n
j=1
j
),
we rst consider all
n
j=1

j
measurable intervals of the form

R =
n
j=1
A
j
for arbitrary A
j

j
and then dene the outer measure
1
(E) = inf
_
+
i=1
(

R
i
) [

R
i
are
n
j=1
j
measurable intervals and E
+
_
i=1
R
i
_
,
where (

R) =
n
j=1
j
(A
j
) for all

R =
n
j=1
A
j
.
To construct the product-measure space (
n
j=1
X
j
,
n
j=1
j
,
n
j=1
j
), we
now consider all
n
j=1

j
measurable intervals of the form

R =
n
j=1
A
j
for
arbitrary A
j

j
and dene the outer measure
2
(E) = inf
_
+
i=1
(

R
i
) [

R
i
are
n
j=1
j
measurable intervals and E
+
_
i=1
R
i
_
,
where (

R) =
n
j=1
j
(A
j
) for all

R =
n
j=1
A
j
.
Our rst task will be to prove that the two outer measures
1
and
2
are
identical.
We observe that all
n
j=1

j
measurable intervals are at the same time
n
j=1

j
measurable and, hence,
2
(E)
1
(E) for every E R
n
.
Now take any E R
n
with
2
(E) < + and an arbitrary > 0. Then
there exists a covering E
+
i=1
R
i
with
n
j=1

j
measurable intervals

R
i
so that
+
i=1
(

R
i
) <
2
(E) + . For each i, write

R
i
=
n
j=1
A
i
j
with
A
i
j

j
. It is clear that there exist B
i
j

j
so that A
i
j
B
i
j
and
j
(A
i
j
) =
j
(B
i
j
). We form the
n
j=1

j
measurable intervals

R
i
=
n
j=1
B
i
j
and have
R
i

R
i
and (

R
i
) = (

R
i
) for all i. We now have a covering E
+
i=1
i
with
n
j=1

j
measurable intervals, and this implies
1
(E)
+
i=1
(

R
i
) =
+
i=1
(

R
i
) <
2
(E) + . Since is arbitrary, we nd
1
(E)
2
(E). In the
remaining case
2
(E) = + the inequality
1
(E)
2
(E) is obviously true
and we conclude that
1
(E) =
2
(E)
for every E R
n
.
The next step in forming the product measure is to apply the process of
Caratheodory to the common outer measure
1
=
2
and nd the common
complete product-measure space
(
n
j=1
X
j
,
n
j=1
j
,
n
j=1
j
) = (
n
j=1
X
j
,
n
j=1
j
,
n
j=1
j
).
where
n
j=1
j
=
n
j=1
j
is the symbol we use for
, the -algebra of
-
measurable sets, and
n
j=1
j
=
n
j=1
j
is the restriction of
on
.
Theorem 8.3 says that
n
j=1

j
and
n
j=1

j
are included in
n
j=1
j
and,
since every
n
j=1

j
measurable interval is also a
n
j=1

j
measurable inter-
val, we have that
n
j=1

j
is included in
n
j=1

j
. Thus
n
j=1
j

n
j=1
n
j=1
j
.
(ii) The proof is immediate from Theorem 8.4.
The most basic application of Theorem 8.5 is related to the n-dimensional
Lebesgue-measure. The next result is no surprise, since the n-dimensional
Lebesgue-measure of any interval in R
n
is equal to the product of the 1-
dimensional Lebesgue-measure of its edges:
m
n
_
n
j=1
[a
j
, b
j
]
_
=
n
j=1
m
1
([a
j
, b
j
]).
Theorem 8.6 (i) The Lebesgue-measure space (R
n
, L
n
, m
n
) is the product-
measure space of n copies of (R, B
R
, m
1
) and, at the same time, the product-
measure space of n copies of (R, L
1
, m
1
).
(ii) The Lebesgue-measure space (R
n
, L
n
, m
n
) is the completion of both measure
spaces (R
n
,
n
j=1
B
R
, m
n
) = (R
n
, B
R
n, m
n
) and (R
n
,
n
j=1
L
1
, m
n
), of which
the second is an extension of the rst.
Proof: We know that
n
j=1
B
R
= B
R
n, that (R, L
1
, m
1
) is the completion of
(R, B
R
, m
1
) and that m
1
is a -nite measure.
Hence, Theorem 8.5 implies immediately that the n copies of (R, B
R
, m
1
)
and, at the same time, the n copies of (R, L
1
, m
1
) induce the same product-
measure space (R
n
,
n
j=1
m1
,
n
j=1
m
1
), which is the completion of both measure
spaces (R
n
, B
R
n,
n
j=1
m
1
) and (R
n
,
n
j=1
L
1
,
n
j=1
m
1
), of which the second is
an extension of the rst.
Theorem 8.3 says that, for every Borel-measurable interval

R =
n
j=1
A
j
, we
have (
n
j=1
m
1
)(

R) =
n
j=1
m
1
(A
j
). In particular, (
n
j=1
m
1
)(P) = vol
n
(P) for
every open-closed interval P in R
n
and Theorem 4.5 implies that
n
j=1
m
1
= m
n
on B
R
n. Hence
(R
n
, B
R
n,
n
j=1
m
1
) = (R
n
, B
R
n, m
n
).
The proof nishes because (R
n
, L
n
, m
n
) is the completion of (R
n
, B
R
n, m
n
).
It is, perhaps, surprising that, although the measure space (R, L
1
, m
1
) is
complete, the product (R
n
,
n
j=1
L
1
, m
n
) is not complete (when n 2, of
course). It is easy to see this. Take any non-Lebesgue-measurable set A R
and form the set E = A0 0 R
n
. Consider, also, the Lebesgue-
measurable interval

R = R 0 0 R
n
. We have that E

R and
m
n
(

R) = m
1
(R)m
1
(0) m
1
(0) = 0. If we assume that (R
n
,
n
j=1
L
1
, m
n
)
is complete, then we conclude that E
n
j=1
L
1
. We, now, take z = (0, . . . , 0)
R
n1
and, then, the section E
z
= A must belong to L
1
. This is not true and
we arrive at a contradiction.
8.3 Multiple integrals.
The purpose of this section is to give the mechanism which reduces the calcu-
lation of product-measures of subsets of cartesian products and of integrals of
functions dened on cartesian products to the calculation of the measures or,
respectively, the integrals of their sections. The gain is obvious: the reduced
calculations are over sets of lower dimension.
8.3. MULTIPLE INTEGRALS. 147
For the sake of simplicity, we further restrict to the case of two measure
spaces.
Theorem 8.7 Let (X
1
,
1
,
1
) and (X
2
,
2
,
2
) be two measure spaces and
(X
1
X
2
,
12
,
1
2
) be their product-measure space.
If E
12
has -nite
1

2
-measure, then E
x1

2
and E
x2

1
for
1
-a.e. x
1
X
1
and
2
-a.e. x
2
X
2
and the a.e. dened functions
x
1

2
(E
x1
), x
2

1
(E
x2
)
are
1
measurable and, respectively,
2
measurable. Also,
(
1
2
)(E) =
_
X1
2
(E
x1
) d
1
(x
1
) =
_
X2
1
(E
x2
) d
2
(x
2
).
Proof: As shown by Theorem 8.5, it is true that
12
=
12
and
1
2
=
1

2
. It is also immediate that E
x1

2
for
1
-a.e. x
1
X
1
if and only if
E
x1

2
for
1
-a.e. x
1
X
1
and, similarly, E
x2

1
for
2
-a.e. x
2
X
2
if
and only if E
x2

1
for
2
-a.e. x
2
X
2
. Hence, the whole statement of the
theorem remains the same if we replace at each occurence the measure spaces
(X
1
,
1
,
1
) and (X
2
,
2
,
2
) by their completions (X
1
,
1
,
1
) and (X
2
,
2
,
2
).
Renaming, we restate the theorem as follows:
Let (X
1
,
1
,
1
), (X
2
,
2
,
2
) and (X
1
X
2
,
12
,
1

2
) be two complete
measure spaces and their product-measure space. If E
12
has -nite
2
-measure, then E
x1

2
and E
x2

1
for
1
-a.e. x
1
X
1
and
2
-a.e.
x
2
X
2
x
1

2
(E
x1
), x
2

1
(E
x2
)
are
1
2
measurable. Also,
(
1
2
)(E) =
_
X1
2
(E
x1
) d
1
(x
1
) =
_
X2
1
(E
x2
) d
2
(x
2
).
We are, now, going to prove the theorem in this equivalent form and we
denote ^ the collection of all sets E
12
which have all the properties in
the conclusion of the theorem.
(a) Every measurable interval

R = A
1
A
2
belongs to ^.
Indeed,

R
x1
= , if x
1
/ A
1
, and

R
x1
= A
2
, if x A
1
. Hence,
2
(

R
x1
) =
2
(A
2
)
A1
(x
1
) for every x
1
X
1
, implying that the function x
1

2
(

R
x1
) is
1
measurable. Moreover, we have
_
X1
2
(

R
x1
) d
1
=
2
(A
2
)
_
X1
A1
d
1
=
2
(A
2
)
1
(A
1
) = (
1
2
)(

R). The same arguments hold for x
2
-sections.
(b) Assume that the sets E
1
, . . . E
m
^ are pairwise disjoint. Then E =
E
1
E
m
^.
Indeed, from E
x1
= (E
1
)
x1
(E
m
)
x1
for every x
1
X
1
, we have that
E
x1

2
for
1
-a.e. x
1
X
1
and
2
(E
x1
) =
2
((E
1
)
x1
) + +
2
((E
m
)
x1
)
for
1
-a.e. x
1
X
1
. By the completeness of
1
, the function x
1

2
(E
x1
)
is
1
measurable and
_
X1
2
(E
x1
) d
1
(x
1
) =
m
j=1
_
X1
2
((E
j
)
x1
) d
1
(x
1
) =
m
j=1
(
1
2
)(E
j
) = (
1
2
)(E). The same argument holds for x
2
-sections.
(c) Assume that E
n
^ for every n N. If E
n
E, then E ^.
From (E
n
)
x1
E
x1
for every x
1
X
1
, we have that E
x1

2
for
1
-a.e.
x
1
X
1
. Continuity of
2
from below implies that
2
((E
n
)
x1
)
2
(E
x1
) for
1
-a.e. x
1
X
1
1
, the function x
1

2
(E
x1
) is
1
measurable. By continuity of
1
2
from below and from the Monotone
Convergence Theorem, we get (
1

2
)(E) =
_
X1
2
(E
x1
) d
1
(x
1
). The same
can be proved, symmetrically, for x
2
-sections.
(d) Now, x any measurable interval

R with (
1

2
)(

R) < + and consider
the collection ^
R
of all sets E
12
for which E

R ^.
If E
n
^
R
for all n and E
n
E, then E ^
R
.
Indeed, we have that E
n

R E

R and, hence, (E
n

R)
x1
(E

R)
x1
for every x
1
X
1
. This implies that (E

R)
x1

2
for
1
-a.e. x
1
X
1
.
From the result of (a),
_
X1
2
(

R
x1
) d
1
(x
1
) = (
1

2
)(

R) < + and, hence,
2
(

R
x1
) < + for
1
-a.e. x
1
X
1
. Therefore,
2
_
(E
1

R)
x1
_
< + for
1
-
a.e. x
1
X
1
and, by the continuity of
2
from above, we nd
2
_
(E
n

R)
x1
_
2
_
(E

R)
x1
_
for
1
-a.e. x
1
X
1
1
, the function
x
1

2
_
(E

R)
x1
_
is
1
measurable. Another application of continuity
from above gives (
1

2
)(E

R) =
_
X1
2
_
(E

R)
x1
_
d
1
(x
1
) and, since all
arguments hold for x
2
-sections as well, we conclude that E

R ^ and, hence,
E ^
R
.
If E
n
^
R
for all n and E
n
E, then E
n

R E

R and, from the result
of (c), E ^
R
.
We have proved that the collection ^
R
is a monotone class of subsets of
X
1
X
2
.
If the E
1
, . . . , E
m
^
R
are pairwise disjoint and E = E
1
E
m
, then
E

R = (E
1

R) (E
m

R) and, by the result of (b), E ^
R
. From (a),
we have that ^
R
contains all measurable rectangles and, hence, ^
R
contains
all elements of the algebra / of Proposition 8.4. Therefore, ^
R
includes the
monotone class generated by /, which, by Theorem 1.1, is the same as the
-algebra generated by /, namely
1

2
.
This says that E

R ^ for every E
1

2
and every measurable
interval

R with (
1
2
)(

R) < +.
(e) If / is, again, the algebra of Proposition 8.4, an application of the results
of (b) and (d) implies that E F ^ for every E
1
2
and every F /
with (
1
2
)(F) < +.
(f) Now, let E
1

2
with (
1

2
)(E) < +. We nd a covering E
+
i=1
R
i
by measurable intervals so that
+
i=1
(
1
2
)(

R
i
) < (
1
2
)(E)+1 <
+. We dene F
n
=
n
i=1
R
i
/ and we have that (
1

2
)(F
n
) < + for
every n. The result of (e) implies that E F
n
^ and, since, E F
n
E, we
have, by the result of (c), that E ^.
Hence, E ^ for every E
1
2
with (
1
2
)(E) < +.
(g) Now let E
12
with (
1
2
)(E) = 0. We shall prove that E ^.
We nd, for every k N, a covering E
+
i=1
R
k
i
by measurable intervals
so that
+
i=1
(
1
2
)(

R
k
i
) <
1
k
. We dene A
k
=
+
i=1
R
k
i

1
2
and have
that E A
k
and (
1

2
)(A
k
) <
1
k
. We then write A =
+
k=1
A
k

1

2
and have that E A and (
1
2
)(A) = 0. From the result of (f) we have that
A ^ and, in particular, 0 =
_
X1
2
(A
x1
) d
1
(x
1
) =
_
X2
1
(A
x2
) d
2
(x
2
). The
rst equality implies that
2
(A
x1
) = 0 for
1
-a.e. x
1
X
1
. From E
x1
A
x1
and from the completeness of
2
, we see that E
x1

2
and
2
(E
x1
) = 0 for
1
-
a.e. x
1
X
1
. Now, from the completeness of
1
, we get that the function x
1

2
(E
x1
) is
1
measurable. Moreover, (
1
2
)(E) = 0 =
_
X1
2
(E
x1
) d
1
(x
1
)
and the same arguments hold for x
2
-sections. Therefore, E ^.
(h) If E
12
has (
1
2
)(E) < +, then E ^.
Indeed, for every k N we nd a covering E

+
i=1
R
k
i
by measurable
intervals so that
+
i=1
(
1

2
)(

R
k
i
) < (
1

2
)(E) +
1
k
. We dene A
k
=
+
i=1
R
k
i

1
2
and have that E A
k
and (
1
2
)(A
k
) < (
1
2
)(E)+
1
k
.
We then write A =
+
k=1
A
k

1
2
and have that E A and (
1
2
)(A) =
(
1
2
)(E). Hence A E
12
has (
1
2
)(A E) = 0. As in part (g),
we can nd A

1

2
so that A E A
and (
1

2
)(A
) = 0. We set
B = A A

1

2
and we have B E and (
1

2
)(E B) = 0. By the
result of (g), we have E B ^ and, by the result of (f), B ^. By the result
of (b), E = B (E B) ^.
(i) Finally, if E
12
has -nite (
1

2
)-measure, we nd E
n

12
with (
1
2
)(E
n
) < + for every n and so that E
n
E. Another application
of the result of (c) implies that E ^.
Theorem 8.8 Let (X
1
,
1
,
1
) and (X
2
,
2
,
2
) be -nite measure spaces and
(X
1
X
2
,
1
2
,
1
2
) be their (restricted) product-measure space.
If E
1
2
, then E
x1

2
and E
x2

1
for every x
1
X
1
and x
2
X
2
and the functions
x
1

2
(E
x1
), x
2

1
(E
x2
)
are
1
2
measurable. Also,
(
1
2
)(E) =
_
X1
2
(E
x1
) d
1
(x
1
) =
_
X2
1
(E
x2
) d
2
(x
2
).
Proof: Exactly as in the proof of Theorem 8.7, we denote ^ the collection of all
E
1
2
which satisfy all the properties in the conclusion of this theorem.
(a) If

R is any measurable interval, then

R ^.
The proof is identical to the proof of the result of (a) of Theorem8.7. Observe
that, now, all statements hold for every x
1
X
1
and x
2
X
2
and there is no
need of completeness.
(b) If the sets E
1
, . . . E
m
^ are pairwise disjoint, then E = E
1
E
m
^.
The proof is identical to the proof of the result of (b) of Theorem 8.7.
(c) If E
n
^ for every n N and E
n
E, then E ^.
The proof is identical to the proof of the result of (c) of Theorem 8.7.
(d) We x any measurable interval

R = A
1
A
2
with
1
(A
1
) < + and
2
(A
2
) < + and consider the collection ^
R
of all sets E
1
2
for which
E
R ^. The rest of the proof of part (d) of Theorem 8.7 continues unchanged
and we get that ^
R
is a monotone class of subsets of X
1
X
2
which includes
the algebra / of Proposition 8.4. Hence, ^
R
includes
1
2
and this says that
E

R ^ for every E
1
2
and every measurable interval

R = A
1
A
2
with
1
(A
1
) < + and
2
(A
2
) < +.
(e) Since
1
is -nite, we can nd an increasing sequence A
n
1
so that A
n
1

1
,
A
n
1
X
1
and 0 <
1
(A
n
1
) < +for every n. Similarly, we can nd an increasing
sequence A
n
2
so that A
n
2

2
, A
n
2
X
2
and 0 <
2
(A
n
2
) < + for every n
and we form the measurable intervals

R
n
= A
n
1
A
n
2
.
We take any E
1

2
and, from the result of (d), we have that all sets
E
n
= E

R
n
belong to ^. Since E
n
E, an application of the result of (c)
implies that E ^.
Theorem 8.9 (Tonelli) Let (X
1
,
1
,
1
) and (X
2
,
2
,
2
) be measure spaces
and (X
1
X
2
,
12
,
1
2
) be their product-measure space.
If f : X
1
X
2
[0, +] is
12
measurable and if f
1
((0, +]) has
-nite
1

2
-measure, then f
x1
is
2
measurable for
1
-a.e. x
1
X
1
and
f
x2
is
1
measurable for
2
-a.e. x
2
X
2
x
1

_
X2
f
x1
d
2
, x
2

_
X1
f
x2
d
1
are
1
2
measurable. Also,
_
X1X2
f d(
1
2
) =
_
X1
_
_
X2
f
x1
d
2
_
d
1
(x
1
) =
_
X2
_
_
X1
f
x2
d
1
_
d
2
(x
2
).
Proof: (a) A rst particular case is when f =
E
is the characteristic function
of an E
12
with -nite
1
2
-measure.
Theorem 8.7 implies that (
E
)
x1
=
Ex
1
is
2
measurable for
1
-a.e.
x
1
X
1
and the function x
1

_
X2
(
E
)
x1
d
2
=
2
(E
x1
) is
1
measurable.
Finally, we have
_
X1X2
E
d(
1
2
) = (
1
2
)(E) =
_
X1
2
(E
x1
) d
1
(x
1
) =
_
X1
_
_
X2
(
E
)
x1
d
2
_
d
1
(x
1
). The argument for x
2
-sections is the same.
(b) Next, we take =
m
j=1
Ej
to be the standard representation of a simple
: X
1
X
2
[0, +), where we omit the possible value = 0, and which is
12
measurable and so that
m
j=1
E
j
=
1
((0, +]) has -nite
1

2
-
measure. Then,
x1
=
m
j=1
j
(
Ej
)
x1
and
x2
=
m
j=1
j
(
Ej
)
x2
for every
x
1
X
1
and x
2
X
2
. Therefore, this case reduces, by linearity, to (a).
(c) Finally, we take any
12
measurable f : X
1
X
2
[0, +] with
f
1
((0, +]) having -nite
1
2
-measure. We take an increasing sequence
n
of
12
measurable simple functions
n
: X
1
X
2
[0, +] so that
n
f on X
1
X
2
. From
n
f, it is clear that
1
n
((0, +]) has -nite
2
-measure for every n. Part (b) says that every
n
satises the conclusion
of the theorem and, since (
n
)
x1
f
x1
and (
n
)
x2
f
x2
for every x
1
X
1
and
x
2
X
2
, an application of the Monotone Convergence Theorem implies that f
also satises the conclusion of the theorem.
Theorem 8.10 (Fubini) Let (X
1
,
1
,
1
) and (X
2
,
2
,
2
) two measure spaces
and (X
1
X
2
,
12
,
1
2
) their product-measure space.
If f : X
1
X
2
R or C is integrable with respect to
1

2
, then f
x1
is integrable with respect to
2
for
1
-a.e. x
1
X
1
and f
x2
is integrable with
respect to
1
for
2
-a.e. x
2
X
2
x
1

_
X2
f
x1
d
2
, x
2

_
X1
f
x2
d
1
are integrable with respect to
1
and, respectively, integrable with respect to
2
.
Also,
_
X1X2
f d(
1
2
) =
_
X1
_
_
X2
f
x1
d
2
_
d
1
(x
1
) =
_
X2
_
_
X1
f
x2
d
1
_
d
2
(x
2
).
Proof: (a) If f : X
1
X
2
[0, +] is integrable with respect to
1

2
,
Theorem 8.9 gives
_
X1
_
_
X2
f
x1
d
2
_
d
1
=
_
X2
_
_
X1
f
x2
d
1
_
d
2
=
_
X1X2
f d(
1
2
) < +.
This implies
_
X2
f
x1
d
2
< + for
1
-a.e. x
1
X
1
and
_
X1
f
x2
d
1
< +
for
2
-a.e. x
2
X
2
. Thus, the conclusion of the theorem is true for non-negative
functions.
(b) If f : X
1
X
2
R is integrable with respect to
1

2
, the same is true
for f
+
and f
and, by the result of (a), the conclusion is true for these two
functions. Since f
x1
= (f
+
)
x1
(f
)
x1
and f
x2
= (f
+
)
x2
(f
)
x2
for every
x
1
X
1
and x
2
X
2
, the conclusion is, by linearity, true also for f.
(c) If f : X
1
X
2
C is integrable with respect to
1

2
, the same is true
for (f) and (f). By the result of (b), the conclusion is true for (f) and
(f) and, since f
x1
= (f)
x1
+ i(f)
x1
and f
x2
= (f)
x2
+ i(f)
x2
for every
x
1
X
1
and x
2
X
2
, the conclusion is, by linearity, true for f.
(d) Finally, let f : X
1
X
2
C be integrable with respect to
1

2
. Then
the set E = f
1
()
12
has (
1

2
)(E) = 0. Theorem 8.7 implies
that
2
(E
x1
) = 0 for
1
-a.e. x
1
X
1
and
1
(E
x2
) = 0 for
2
-a.e. x
2
X
2
.
If we dene F = f
E
c, then F : X
1
X
2
C is integrable with respect to
2
and, by (c), the conclusion of the theorem holds for F.
Since F = f holds (
1

2
)-a.e. on X
1
X
2
, we have
_
X1X2
F d(
1

2
) =
_
X1X2
f d(
1

2
). We, also, have that F
x1
= f
x1
on X
2
E
x1
and, hence, F
x1
= f
x1
holds
2
-a.e. on X
2
for
1
-a.e. x
1
X
1
. Therefore,
f
x1
2
and
_
X2
f
x1
d
2
=
_
X2
F
x1
d
2
, for
1
-a.e.
x
1
X
1
. This implies
_
X1
_
_
X2
f
x1
d
2
_
d
1
(x
1
) =
_
X1
_
_
X2
F
x1
d
2
_
d
1
(x
1
)
and, equating the corresponding integrals of F, we nd
_
X1X2
f d(
1

2
) =
_
X1
_
_
X2
f
x1
d
2
_
d
1
(x
1
). The argument is the same for x
2
-sections.
The power of the Theorems of Tonelli and of Fubini lies in the resulting
successive integration formula for the calculation of integrals over product spaces
and in the interchange of successive integrations. The function f to which we
may want to apply Fubinis Theorem must be integrable with respect to the
product measure
1

2
. The Theorem of Tonelli is applied to non-negative
functions f which must be
12
measurable and whose set f
1
((0, +])
must be of -nite
1

2
-measure. Thus, the assumptions of Theorem of
Tonelli are, except for the sign, weaker than the assumptions of the Theorem of
Fubini.
The strategy, when we want to calculate the integral of f over the prod-
uct space by means of successive integrations or to interchange successive in-
tegrations, is rst to prove that f is
12
measurable and that the set
(x
1
, x
2
) [ f(x
1
, x
2
) ,= 0 is of -nite
1

2
-measure. We, then, apply the
Theorem of Tonelli to [f[ and have
_
X1X2
[f[ d(
1
2
) =
_
X1
_
_
X2
[f[
x1
d
2
_
d
1
=
_
X2
_
_
X1
[f[
x2
d
1
_
d
2
.
By calculating either the second or the third term in this string of equalities,
we calculate the
_
X1X2
[f[ d(
1

2
). If it is nite, then f is integrable with
respect to the product measure
1
2
and we may apply the Theorem of Fubini
to nd the desired
_
X1X2
f(x
1
, x
2
) d(
1

2
)(x
1
, x
2
) =
_
X1
_
_
X2
f(x
1
, x
2
) d
2
(x
2
)
_
d
1
(x
1
)
=
_
X2
_
_
X1
f(x
1
, x
2
) d
1
(x
1
)
_
d
2
(x
2
).
Of the two starting assumptions, the -niteness of (x
1
, x
2
) [ f(x
1
, x
2
) ,= 0
is usually easy to check. For example, if the measure spaces (X
1
,
1
,
1
) and
(X
2
,
2
,
2
) are both -nite, then the measure space (X
1
X
2
,
12
,
1
2
)
is also -nite and all subsets of X
1
X
2
are obviously of -nite
1

2
-
measure.
The assumption of
12
measurability of f is more subtle and sometimes
dicult to verify.
Theorem 8.11 (Tonelli) Let (X
1
,
1
,
1
) and (X
2
,
2
,
2
) be -nite measure
spaces and (X
1
X
2
,
1

2
,
1

2
) be their (restricted) product-measure
space.
If f : X
1
X
2
[0, +] is
1
2
measurable, then f
x1
is
2
measurable
for every x
1
X
1
and f
x2
is
1
measurable for every x
2
X
2
and the functions
x
1

_
X2
f
x1
d
2
, x
2

_
X1
f
x2
d
1
are
1
2
measurable. Also,
_
X1X2
f d(
1
2
) =
_
X1
_
_
X2
f
x1
d
2
_
d
1
(x
1
) =
_
X2
_
_
X1
f
x2
d
1
_
d
2
(x
2
).
8.4. SURFACE-MEASURE ON S
N1
. 153
Proof The measurability of the sections is an immediate application of Theorem
8.2 and does not need the assumption about -niteness. Otherwise, the proof
results from Theorem 8.8 in exactly the same way in which the proof of Theorem
8.9 results from Theorem 8.7.
Theorem 8.12 (Fubini) Let (X
1
,
1
,
1
) and (X
2
,
2
,
2
) be two -nite mea-
sure spaces and (X
1
X
2
,
1
2
,
1
2
) be their (restricted) product-measure
space.
Let f : X
1
X
2
R or C be
1

2
measurable and integrable with
respect to
1
2
. Then f
x1
is
2
measurable for every x
1
X
1
and integrable
with respect to
2
for
1
-a.e. x
1
X
1
. Also, f
x2
is
1
measurable for every
x
2
X
2
and integrable with respect to
1
for
2
-a.e. x
2
X
2
. The a.e. dened
functions
x
1

_
X2
f
x1
d
2
, x
2

_
X1
f
x2
d
1
are integrable with respect to
1
and, respectively, integrable with respect to
2
and
_
X1X2
f d(
1
2
) =
_
X1
_
_
X2
f
x1
d
2
_
d
1
(x
1
) =
_
X2
_
_
X1
f
x2
d
1
_
d
2
(x
2
).
Proof: Again, the measurability of the sections is an immediate application of
Theorem 8.2 and does not need the assumption about -niteness. Otherwise,
the proof results from Theorem 8.11 in exactly the same way in which the proof
of Theorem 8.10 results from Theorem 8.9.
8.4 Surface-measure on S
n1
.
For every x = (x
1
, . . . , x
n
) R
n
= R
n
0 we write
r = [x[ =
_
x
2
1
+ +x
2
n
R
+
= (0, +), y =
x
[x[
S
n1
,
where S
n1
= y R
n
[ [y[ = 1 is the unit spere of R
n
.
The mapping : R
n
R
+
S
n1
dened by
(x) = (r, y) =
_
[x[,
x
[x[
_
is one-to-one and onto and its inverse
1
: R
+
S
n1
R
n
is given by
1
(r, y) = x = ry.
The numbers r = [x[ and y =
x
|x|
are called the polar coordinates of x and
the mappings and
1
determine an identication of R
n
with the cartesian

product R
+
S
n1
, where every point x ,= 0 is identied with the pair (r, y) of
its polar coordinates.
As usual, we consider S
n1
as a metric subspace of R
n
. This means that
the distance between points of S
n1
is their euclidean distance considered as
points of the larger space R
n
. Namely
[y y
[ =
_
(y
1
y
1
)
2
+ + (y
n
y
n
)
2
,
for every y = (y
1
, . . . , y
n
), y
= (y
1
, . . . , y
n
) S
n1
. No two points of S
n1
have
distance greater that 2 and, if two points have distance 2, then they are opposite
or, equivalently, anti-diametric. The open ball in S
n1
with center y S
n1
and radius r > 0 is the spherical cap S(y; r) = y
S
n1
[ [y
y[ < r, which
is the intersection of the euclidean ball B(y; r) = x R
n
[ [x y[ < r with
S
n1
. In fact, the intersection of an arbitrary euclidean open ball in R
n
with
S
n1
is, if non-empty, a spherical cap of S
n1
.
It is easy to see that there is a countable collection of spherical caps with
the property that every open set in S
n1
is a union (countable, necessarily) of
spherical caps from this collection. Indeed, such is the collection of the (non-
empty) intersections with S
n1
of all open balls in R
n
with rational centers and
rational radii: if U is an arbitrary open subset of S
n1
and we take arbitrary
y U, we can nd r so that B(y; r) S
n1
U. Then, we can nd an open
ball B(x
; r
) with rational x
and rational r
so that y B(x
; r
) B(y; r).
Now, y belongs to the spherical cap B(x
; r
) S
n1
U.
If we equip R
+
S
n1
with the product-topology through the product-metric
d
_
(r, y), (r
, y
)
_
= max([r r
[, [y y
[),
then the mappings and
1
are both continuous. In fact, it is clear that the
convergence (r
k
, y
k
) (r, y) in the product-metric of R
+
S
n1
is equivalent
to the simultaneous r
k
r and y
k
y. Therefore, if x
k
x in R
n
, then
r
k
= [x
k
[ [x[ = r and y
k
=
x
k
|x
k
|

x
|x|
= y and hence (x
k
) = (r
k
, y
k
)
(r, y) = (x) in R
+
S
n1
. Conversely, if (r
k
, y
k
) (r, y) in R
+
S
n1
, then
r
k
r and y
k
y and hence
1
(r
k
, y
k
) = r
k
y
k
ry =
1
(r, y) in R
n
.
We may observe that the open balls in the product-topology of R
+
S
n1
are exactly all the cartesian products (a, b) S(y; r) of open subintervals of R
+
with spherical caps of S
n1
.
Proposition 8.6 Let X be a topological space and Y X with the restricted
topology. This means that a subset of Y is open in Y if and only if it is the
intersection with Y of a set open in X. Then B
Y
= A Y [ A B
X
.
Proof: Consider = A Y [ A B
X
. It is easy to prove that is a -
algebra of subsets of Y and that it contains all subsets of Y which are open in
Y . Therefore, B
Y
and it remains to prove the opposite inclusion.
We set
1
= A X[ A Y B
Y
. It is again easy to see that
1
is a
-algebra of subsets of X and contains all subsets of X which are open in X.
Hence B
X

1
. This means that AY B
Y
for every A B
X
or, equivalently,
that B
Y
.
N1
. 155
The next proposition contains information about the Borel structures of R
n
and of R
+
, S
n1
and their product R
+
S
n1
.
Proposition 8.7 (i) B
R
n
= E B
R
n [ E R
n
.
(ii) B
R
+ = E B
R
[ E R
+
and B
R
+ is generated by the collection of all
open subintervals of R
+
and, also, by the collection of all open-closed subinter-
vals of R
+
.
(iii) B
S
n1 = E B
R
n [ E S
n1
and B
S
n1 is generated by the collection
of all spherical caps.
(iv) B
R
+
S
n1 = B
R
+ B
S
n1.
(v) (E) is a Borel set in R
+
S
n1
for every Borel set E in R
n
and
1
(E)
is a Borel set in R
n

+
S
n1
.
Proof: The equalities of (i),(ii) and (iii) are simple consequences of Proposition
8.6. That B
R
+ is generated by the collection of all open or of all open-closed
subintervals of R
+
is due to the fact that every open subset of R
+
is a countable
union of such intervals. Also, that B
S
n1 is generated by the collection of all
spherical caps is due to the fact that every open subset of S
n1
is a countable
union of spherical caps.
(iv) Both B
R
+
S
n1 and B
R
+ B
S
n1 are -algebras of subsets of the space
R
+
S
n1
. The second is generated by the collection of all cartesian products
of open subintervals of R
+
n1
and all these sets are
open subsets of R
+
S
n1
and, hence, belong to the rst -algebra. Therefore,
the second -algebra is included in the rst. Conversely, the rst -algebra is
generated by the collection of all open subsets of R
+
S
n1
and every such set
is a countable union of open balls, i.e. of cartesian products of open subintervals
of R
+
n1
. Thus, every open subset of R
+
S
n1
is
contained in the second -algebra and, hence, the rst -algebra is included in
the second.
(v) Since is continuous, it is (B
R
n
, B
R
+
S
n1)measurable and, thus,
1
(E)
is a Borel set in R
n

+
S
n1
. The other statement
is, similarly, a consequence of the continuity of
1
.
A set R
n
is called a positive cone if rx for every r R

+
and
every x or, equivalently, if is closed under multiplication by positive
numbers or, equivalently, if is invariant under dilations. If B R
n
, then
the set R
+
B = rb [ r R
+
, b B is, obviously, a positive cone and it is
called the positive cone determined by B. It is trivial to see that, if is
a positive cone and A = S
n1
, then is the positive cone determined by
A and, conversely, that, if A S
n1
and is the positive cone determined by
A, then S
n1
= A. This means that there is a one-to-one correspondence
between the subsets of S
n1
and the positive cones of R
n
.
The next result expresses a simple characterization of open and of Borel
subsets of S
n1
in terms of the corresponding positive cones.
Proposition 8.8 Let A S
n1
.
(i) A is open in S
n1
if and only if the cone R
+
A is open in R
n
.
(ii) A is a Borel set in S
n1
if and only if R
+
n
.
Proof: (i) By the denition of the product-topology, A is open in S
n1
if and
only if R
+
A is open in R
+
S
n1
. By the continuity of and
1
, this
last one is true if and only if R
+
A =
1
(R
+
A) is open in R
n
if and only
if R
+
A is open in R
n
.
(ii) If A is a Borel set in S
n1
then, as a measurable interval, R
+
A is a Borel
set in R
+
S
n1
. Conversely, if R
+
+
S
n1
, then
all its r-sections, and in particular A, are Borel sets in S
n1
. Therefore, A is a
Borel set in S
n1
if and only if R
+
+
S
n1
. Proposition
8.7 implies that this is true if and only if R
+
A =
1
(R
+
A) is a Borel set
in R
n
if and only if R
+
n
.
The following is useful.
Proposition 8.9 (i) If B is open in R
n
, then R
+
B is open in R
n
.
(ii) If B is a Borel set in R
n
, then R
+
B is a Borel set in R
n
.
Proof: (i) Assume that B R
n
is open and take arbitrary x R

+
B. Then
x = rx
for some r R
+
and some x
B. We take > 0 so that B(x
; ) B
and we have that B(x; r) = r B(x
; ) r B R
+
B. Hence, R
+
B is
open in R
n
.
(ii) We consider the collection of all B R
n
with the property that R

+
B
B
R
n
. We easily prove that is a -algebra of subsets of R

n
. Part (i) implies

that contains all open subsets of R
n
and, hence, B
R
n
.
Proposition 8.7 implies that the set M A = ry [ r M, y A is a Borel
set in R
n
for every Borel set A in S

n1
and every Borel set M in R
+
. This
is true because M A =
1
(M A) and M A is a Borel set (measurable
interval) in R
+
S
n1
.
Proposition 8.10 If we dene
n1
(A) = n m
n
_
(0, 1] A
_
for every A B
S
n1, then
n1
is a measure on (S
n1
, B
S
n1).
Proof: We have
n1
() = n m
n
_
(0, 1]
_
= n m
n
() = 0. Moreover, if
A
1
, A
2
, . . . B
S
n1 are pairwise disjoint, then the sets (0, 1] A
1
, (0, 1] A
2
, . . .
are also pairwise disjoint. Hence,
n1
(
+
j=1
A
j
) = n m
n
_
(0, 1]
+
j=1
A
j
_
=
n m
n
_
+
j=1
((0, 1] A
j
)
_
=
+
j=1
n m
n
_
(0, 1] A
j
_
=
+
j=1

n1
(A
j
).
Denition 8.7 The measure
n1
on (S
n1
, B
S
n1), which is dened in Propo-
sition 8.10, is called the (n 1)-dimensional surface-measure on S
n1
.
Lemma 8.1 If we dene
(N) =
_
N
r
n1
dr
for every N B
R
+, then is a measure on (R
+
, B
R
+).
N1
. 157
Proof: A simple consequence of Theorem 7.13.
Lemma 8.2 If we dene
m
n
(E) = m
n
(
1
(E))
+
S
n1
, then m
n
is a measure on the measurable
space (R
+
S
n1
, B
R
+
S
n1).
Proof: The denition makes sense because, by Proposition 8.7,
1
(E) is a
Borel set in R
n
.
Clearly, m
n
() = m
n
(
1
()) = m
n
() = 0. If E
1
, E
2
, . . . are pairwise
disjoint, then
1
(E
1
),
1
(E
2
), . . . are also pairwise disjoint and we nd that
m
n
(
+
j=1
E
j
) = m
n
(
1
(
+
j=1
E
j
)) = m
n
(
+
j=1
1
(E
j
)) =
+
j=1
m
n
(
1
(E
j
))
=
+
j=1
m
n
(E
j
).
Lemma 8.3 The measures m
n
and
n1
are identical on the measurable
space (R
+
S
n1
, B
R
+
S
n1) = (R
+
S
n1
, B
R
+ B
S
n1).
Proof: The equality B
R
+
S
n1 = B
R
+ B
S
n1 is in Proposition 8.7.
If A is a Borel set in S
n1
, then the sets (0, b] A and (0, 1] A are both
Borel sets in R
n
and the rst is a dilate of the second by the factor b > 0. By
Theorem 4.7, m
n
((0, b] A) = b
n
m
n
((0, 1] A) for every b > 0. By a simple
subtraction we nd that m
n
((a, b] A) = (b
n
a
n
)m
n
((0, 1] A) for every a, b
with 0 a < b < +.
Therefore, if A is a Borel set in S
n1
, then
m
n
((a, b] A) = m
n
(
1
((a, b] A)) = m
n
((a, b] A)
= (b
n
a
n
)m
n
((0, 1] A) =
b
n
a
n
n

n1
(A)
=
_
(a,b]
r
n1
dr
n1
(A) = ((a, b])
n1
(A)
= (
n1
) ((a, b] A).
If we dene
(N) = m
n
(N A), (N) = (
n1
)(N A)
for every Borel set N in R
+
, it is easy to see that both and are Borel
measures on R
+
and, by what we just proved, they satisfy ((a, b]) = ((a, b])
for every interval in R
+
. This, obviously, extends to all nite unions of pairwise
disjoint open-closed intervals. Theorem 2.4 implies, now, that the two measures
are equal on the -algebra generated by the collection of all these sets, which,
by Proposition 8.7, is B
R
+. Therefore,
m
n
(N A) = (
n1
)(N A)
for every Borel set N in R
+
and every Borel set A in S
n1
.
Theorem 8.4 implies now the equality of the two measures, because both
measures and
n1
are -nite.
If E R
n
, we consider the set (E) R

+
S
n1
. We, also, consider the
r-sections (E)
r
= y S
n1
[ (r, y) (E) = y S
n1
[ ry E and the
y-sections (E)
y
= r R
+
[ (r, y) (E) = r R
+
[ ry E of (E). We
extend the notation as follows.
Denition 8.8 If E R
n
, we dene, for every r R
+
and every y S
n1
,
E
r
= y S
n1
[ ry E, E
y
= r R
+
[ ry E
and call them the r-sections and the y-sections of E, respectively.
Observe that E may contain 0, but this plays no role. Thus, the sections
of E are, by denition, exactly the same as the sections of (E 0). This is
justied by the informal identication of E 0 with (E 0).
Theorem 8.13 Let E be any Borel set in R
n
. Then, E
r
is a Borel set in S
n1
for every r R
+
and E
y
is a Borel set in R
+
for every y S
n1
and the
functions
r
n1
(E
r
), y
_
Ey
r
n1
dr
are B
R
+measurable and, respectively, B
S
n1measurable. Also,
m
n
(E) =
_
+
0
n1
(E
r
)r
n1
dr =
_
S
n1
_
_
Ey
r
n1
dr
_
d
n1
(y).
Proof: The set E 0 is a Borel set in R
n
, while E
r
= (E 0)
r
and
E
y
= (E 0)
y
.
Lemmas 8.2 and 8.3 imply that m
n
(E) = m
n
(E 0) = m
n
_
(E 0)
_
=
(
n1
)
_
(E 0)
_
. Proposition 8.7 says that (E 0) is a Borel set in
R
+
S
n1
and the rest is a consequence of Theorem 8.8.
The next result gives a simple description of the completion of the measure
space (S
n1
, B
S
n1,
n1
) in terms of positive cones.
Denition 8.9 We denote (S
n1
, o
n1
,
n1
) the completion of the measure
space (S
n1
, B
S
n1,
n1
).
Proposition 8.11 If A S
n1
, then
(i) A o
n1
if and only if R
+
A L
n
if and only if (0, 1] A L
n
,
(ii)
n1
(A) = n m
n
_
(0, 1] A
_
for every A o
n1
.
Proof: (i) If A o
n1
, there exist A
1
, A
2
B
S
n1 with
n1
(A
2
) = 0 so
that A
1
A and A A
1
A
2
. Proposition 8.8 implies that the positive
cones R
+
A
1
and R
+
A
2
are Borel sets in R
n
with R
+
A
1
R
+
A and
(R
+
A) (R
+
A
1
) R
+
A
2
. Lemmas 8.2 and 8.3 or Theorem 8.13 imply
m
n
(R
+
A
2
) = (R
+
)
n1
(A
2
) = 0. Hence, R
+
A L
n
.
N1
. 159
Conversely, let R
+
A L
n
. Then, there are Borel sets B
1
, B
2
R
n
with m
n
(B
2
) = 0, so that B
1
R
+
A and (R
+
A) B
1
B
2
. For every
r R
+
we have that (B
1
)
r
A and A (B
1
)
r
(B
2
)
r
. From Theorem 8.13,
_
+
0

n1
((B
2
)
r
)r
n1
dr = m
n
(B
2
) = 0, implying that
n1
((B
2
)
r
) = 0 for
m
1
-a.e. r (0, +). If we consider such an r, since (B
1
)
r
and (B
2
)
r
are Borel
sets in S
n1
, we conclude that A o
n1
.
If R
+
A L
n
, then (0, 1] A = (R
+
A) B
n
L
n
. Conversely, if
(0, 1] A L
n
, then R
+
A =
+
k=1
k
_
(0, 1] A
_
L
n
.
(ii) We take A o
n1
and A
1
, A
2
B
S
n1 with
n1
(A
2
) = 0 so that A
1
A
and A A
1
A
2
. Then the sets (0, 1] A
1
and (0, 1] A
2
are Borel sets in
R
n
with (0, 1] A
1
(0, 1] A and (0, 1] A (0, 1] A
1
(0, 1] A
2
. Since
m
n
((0, 1] A
2
) =
1
n

n1
(A
2
) = 0, we conclude that
n1
(A) =
n1
(A
1
) =
n m
n
((0, 1] A
1
) = n m
n
((0, 1] A).
The next result is an extension of Theorem 8.13 to Lebesgue-measurable
sets.
Theorem 8.14 Let E L
n
. Then, E
r
o
n1
for m
1
-a.e. r R
+
and
E
y
L
1
for
n1
-a.e. y S
n1
r
n1
(E
r
), y
_
Ey
r
n1
dr
are L
1
measurable and, respectively, o
n1
measurable. Also,
m
n
(E) =
_
+
0
n1
(E
r
)r
n1
dr =
_
S
n1
_
_
Ey
r
n1
dr
_
d
n1
(y).
Proof: We consider Borel sets B
1
, B
2
in R
n
with m
n
(B
2
) = 0, so that B
1
E
and E B
1
B
2
.
Theorem 8.13 implies that, for every r R
+
, (B
1
)
r
and (B
2
)
r
are Borel sets
in S
n1
with (B
1
)
r
E
r
and E
r
(B
1
)
r
(B
2
)
r
. From Theorem 8.13 again,
_
+
0

n1
((B
2
)
r
)r
n1
dr = m
n
(B
2
) = 0 and we get that
n1
((B
2
)
r
) = 0 for
m
1
-a.e. r R
+
. Therefore, E
r
o
n1
and
n1
(E
r
) =
n1
((B
1
)
r
) for m
1
-a.e.
r R
+
.
Similarly, for every y S
n1
, (B
1
)
y
and (B
2
)
y
are Borel sets in R
+
with
(B
1
)
y
E
y
and E
y
(B
1
)
y
(B
2
)
y
. From
_
S
n1
_ _
(B2)y
r
n1
dr
_
d
n1
(y) =
m
n
(B
2
) = 0, we get that
_
(B2)y
r
n1
dr = 0 for
n1
-a.e. y S
n1
. This
implies m
1
((B
2
)
y
) = 0 for
n1
-a.e. y S
n1
and, hence, E
y
L
1
and
_
Ey
r
n1
dr =
_
(B1)y
r
n1
dr for
n1
-a.e. y S
n1
. Theorem 8.13 implies
m
n
(E) = m
n
(B
1
) =
_
+
0

n1
((B
1
)
r
)r
n1
dr =
_
+
0

n1
(E
r
)r
n1
dr and,
also, =
_
S
n1
_ _
(B1)y
r
n1
dr
_
d
n1
(y) =
_
S
n1
_ _
Ey
r
n1
dr
_
d
n1
(y).
The rest of this section consists of a series of theorems which describe the
so-called method of integration by polar coordinates.
Denition 8.10 Let f : R
n
Y . For every r R
+
and every y S
n1
we
dene the functions f
r
: S
n1
Y and f
y
: R
+
Y by the formulas
f
r
(y) = f
y
(r) = f(ry).
f
r
is called the r-section of f and f
y
is called the y-section of f.
The next two theorems cover integration by polar coordinates for Borel-
measurable functions.
Theorem 8.15 Let f : R
n
[0, +] be B
R
nmeasurable. Then, every f
r
is
B
S
n1measurable and every f
y
is B
R
+measurable. The functions
r
_
S
n1
f(ry) d
n1
, y
_
+
0
f(ry)r
n1
dr
are B
R
+measurable and, respectively, B
S
n1measurable. Moreover
_
R
n
f(x) dm
n
(x) =
_
+
0
_
_
S
n1
f(ry) d
n1
(y)
_
r
n1
dr
=
_
S
n1
_
_
+
0
f(ry)r
n1
dr
_
d
n1
(y).
Proof: The results of this theorem and of Theorem 8.13 are the same in case
f =
E
. Using the linearity of the integrals, we prove the theorem in the
case of a simple function : R
n
[0, +]. Finally, applying the Monotone
Convergence Theorem to an increasing sequence of simple functions, we prove
the theorem in the general case.
Theorem 8.16 If f : R
n
R or C is B
R
nmeasurable and integrable with
respect to m
n
, then every f
r
is B
S
n1measurable and, for m
1
-a.e. r R
+
,
f
r
n1
. Also, every f
y
is B
R
+measurable, and
for
n1
-a.e. y S
n1
, f
y
is integrable with respect to m
1
. The a.e. dened
functions
r
_
S
n1
f(ry) d
n1
(y), y
_
+
0
f(ry)r
n1
dr
are integrable with respect to m
1
and, respectively, with respect to
n1
. Also
_
R
n
f(x) dm
n
(x) =
_
+
0
_
_
S
n1
f(ry) d
n1
(y)
_
r
n1
dr
=
_
S
n1
_
_
+
0
f(ry)r
n1
dr
_
d
n1
(y).
Proof: We use Theorem 8.15 to pass to the case of functions f : R
n
R, by
writing them as f = f
+
f
. We next treat the case of f : R

n
C, by writing
f = (f) +i(f), after we exclude, in the usual manner, the set f
1
().
N1
. 161
The next two theorems treat integration by polar coordinates in the case
of Lebesgue-measurable functions. They are proved, one after the other, using
Theorem 8.14 exactly as Theorems 8.15 and 8.16 were proved with the use of
Theorem 8.13.
n
[0, +] be L
n
measurable. Then, for m
1
-a.e.
r R
+
, the function f
r
is o
n1
measurable and, for
n1
-a.e. y S
n1
, the
function f
y
is L
1
measurable. The a.e. dened functions
r
_
S
n1
f(ry) d
n1
(y), y
_
+
0
f(ry)r
n1
dr
are L
1
measurable and, respectively, o
n1
measurable. Moreover
_
R
n
f(x) dm
n
(x) =
_
+
0
_
_
S
n1
f(ry) d
n1
(y)
_
r
n1
dr
=
_
S
n1
_
_
+
0
f(ry)r
n1
dr
_
d
n1
(y).
Theorem 8.18 If f : R
n
R or C is L
n
measurable and integrable with
respect to m
n
, then, for m
1
-a.e. r R
+
, f
r
n1
and, for
n1
-a.e. y S
n1
, f
y
is integrable with respect to m
1
. The a.e.
dened functions
r
_
S
n1
f(ry) d
n1
(y), y
_
+
0
f(ry)r
n1
dr
are integrable with respect to m
1
and, respectively, with respect to
n1
. Also
_
R
n
f(x) dm
n
(x) =
_
+
0
_
_
S
n1
f(ry) d
n1
(y)
_
r
n1
dr
=
_
S
n1
_
_
+
0
f(ry)r
n1
dr
_
d
n1
(y).
Denition 8.11 A set E R
n
is called radial if x E implies that x
E
for all x
with [x
[ = [x[.
A function f : R
n
Y is called radial if f(x) = f(x
) for every x, x
with
[x[ = [x
[.
It is obvious that E is radial if and only if
E
is radial.
If the set E is radial, we may dene the radial projection of E as
E = r R
+
[ x E when [x[ = r.
Also, if f is radial, we may dene the radial projection of f as the function
f : R
+
Y by
f(r) = f(x)
for every x R
n
with [x[ = r.
It is obvious that a radial set or a radial function is uniquely determined
from its radial projection (except from the fact that the radial set may or may
not contain the point 0 and that the value of the function at 0 is not determined
by its radial projection).
Proposition 8.12 (i) The radial set E R
n
is in B
R
n or in L
n
if and only
if its radial projection is in B
R
+ or, respectively, in L
1
. In any case we have
m
n
(E) =
n1
(S
n1
)
_
E
r
n1
dr.
(ii) If (Y,
) is a measurable space, then the radial function f : R

n
Y is
(B
R
n,
)measurable or (L
n
,
)measurable if and only if its radial projection

is (B
R
+,
)measurable or, respectively, (L

1
,
)measurable.
If f : R
n
[0, +] is Borel- or Lebesgue-measurable or if f : R
n
R or
C is Borel- or Lebesgue-measurable and integrable with respect to m
n
, then
_
R
n
f(x) dm
n
(x) =
n1
(S
n1
)
_
+
0
f(r)r
n1
dr.
Proof: (i) If E B
R
n or E L
n
is radial, then, for every y S
n1
, we have
E
y
=

E and, hence, the result is a consequence of Theorems 8.13 and 8.14.
For the converse we may argue as follows: we consider the collection of all
subsets of R
+
which are radial projections of radial Borel sets in R
n
, we then
prove easily that this collection is a -algebra which contains all open subsets
of R
+
and we conclude that it contains all Borel sets in R
+
.
Now, if E is radial and

E L
1
, we take Borel sets M
1
, M
2
in R
+
with
m
1
(M
2
) = 0 so that M
1

E and

E M
1
M
2
. We consider the radial sets
E
1
, E
2
R
n
so that

E
1
= M
1
and

E
2
= M
2
, which are Borel sets, by the
result of the previous paragraph. Then we have E
1
E and E E
1
E
2
.
Since 0 = m
n
(E
2
) =
_
S
n1
_
_
(E2)y
r
n1
dr
_
d
n1
=
n1
(S
n1
)
_
E2
r
n1
dr,
we have
_
E2
r
n1
dr and, hence, m
1
(
E
2
) = 0. This implies that E L
1
.
(ii) The statement about measurability is a trivial consequence of the denition
of measurability and the result of part (i). The integral formulas are conse-
quences of Theorems 8.15 up to 8.18.
8.5. EXERCISES. 163
8.5 Exercises.
1. Consider the measure spaces (R, B
R
, m
1
) and (R, T(R), ), where is
the counting measure. If E = (x
1
, x
2
) [ 0 x
1
= x
2
1, prove that
all numbers (m
1
)(E),
_
R
(E
x1
) dm
1
(x
1
) and
_
R
m
1
(E
x2
) d(x
2
) are
dierent.
2. Consider a
m,n
= 1 if m = n, a
m,n
= 1 if m = n +1 and a
m,n
= 0 in any
other case. Then
+
n=1
_
+
m=1
a
m,n
_
,=
+
m=1
_
+
n=1
a
m,n
_
. Explain,
through the Theorem of Fubini.
3. The graph and the area under the graph of a function.
Suppose that (X, , ) is a measure space and f : X [0, +] is
measurable. If
A
f
= (x, y) X R[ 0 y < f(x)
and
G
f
= (x, y) X R[ y = f(x),
prove that both A
f
and G
f
are B
R
measurable. If, moreover, is
-nite, prove that
( m
1
)(A
f
) =
_
X
f d, ( m
1
)(G
f
) = 0.
4. The distribution function.
Suppose that (X, , ) is a -nite measure space and f : X [0, +] is
measurable. Calculating the measure
G
of the set A
f
= (x, y)
X R[ 0 y < f(x), prove Proposition 7.14.
5. Consider two measure spaces (X
1
,
1
,
1
) and (X
2
,
2
,
2
), a
1
measu-
rable f
1
: X
1
C and a
2
measurable f
2
: X
2
C. Consider the
function f : X
1
X
2
C dened by f(x
1
, x
2
) = f
1
(x
1
)f
2
(x
2
).
Prove that f is
1
2
measurable.
If f
1
1
and f
2
2
, prove that f is integrable with respect to
1
2
and that
_
X1X2
f d(
1
2
) =
_
X1
f
1
d
1
_
X2
f
2
d
2
.
6. The volume of the unit ball in R
n
and the surface-measure of S
n1
.
(i) If v
n
= m
n
(B
n
) is the Lebesgue-measure of the unit ball of R
n
, prove
that
v
n
= 2v
n1
_
1
0
(1 t
2
)
n1
2
dt.
(ii) Set J
n
=
_
1
0
(1 t
2
)
n1
2
dt for n 0 and prove the inductive formula
J
n
=
n1
n
J
n2
, n 2.
(iii) Prove that the gamma-function (dened in Exercise 7.9.38) satises
the inductive formula
(s + 1) = s(s)
for every s H
+
, and that (1) = 1, (
1
2
) =
.
(iii) Prove that
v
n
=

n
2
(
n
2
+ 1)
,
n1
(S
n1
) =
2
n
2
(
n
2
)
.
7. The integral of Gauss and the measures of B
n
and of S
n1
.
Dene
I
n
=
_
R
n
e
|x|
2
2
dx.
(i) Prove that I
n
= I
n
1
for every n N.
(ii) Use integration by polar coordinates to prove that I
2
= 2 and, hence,
that
_
R
n
e
|x|
2
2
dx = (2)
n
2
.
(iii) Use integration by polar coordinates to prove that
(2)
n
2
=
n1
(S
n1
)
_
+
0
e
r
2
2
r
n1
dr
and, hence,
n1
(S
n1
) =
2
n
2
(
n
2
)
, v
n
= m
n
(B
n
) =

n
2
(
n
2
+ 1)
.
8. From
_
n
0
sin x
x
dx =
_
n
0
_ _
+
0
e
xt
dt
_
sin xdx, prove that
_
+
0
sin x
x
dx =

2
.
9. Convolution.
Let f, g : R
n
R or C be L
n
measurable.
(i) Prove that the function H : R
n
R
n
R or C, which is dened by
the formula
H(x, y) = f(x y)g(y),
is L
2n
measurable.
Now, let f and g be integrable with respect to m
n
.
8.5. EXERCISES. 165
(ii) Prove that H is integrable with respect to m
2n
and
_
R
2n
[H[ dm
2n

_
R
n
[f[ dm
n
_
R
n
[g[ dm
n
.
(iii) Prove that, for m
n
-a.e. x R
n
the function f(x )g() is integrable
with respect to m
n
.
The a.e. dened function f g : R
n
R or C by the formula
(f g)(x) =
_
R
n
f(x y)g(y) dy
is called the convolution of f and g.
(iv) Prove that f g is integrable with respect to m
n
, that
_
R
n
(f g) dm
n
=
_
R
n
f dm
n
_
R
n
g dm
n
and _
R
n
[f g[ dm
n

_
R
n
[f[ dm
n
_
R
n
[g[ dm
n
.
(v) Prove that, for every f, g, h, f
1
, f
2
which are Lebesgue-integrable, we
have m
n
-a.e. on R
n
that fg = gf, (fg)h = f(gh), (f)g = (fg)
and (f
1
+f
2
) g = f
1
g +f
2
g.
10. The Fourier transforms of Lebesgue-integrable functions.
Let f : R
n
R or C be Lebesgue-integrable over R
n
. We dene the
function

f : R
n
R or C by the formula
f() =
_
R
n
e
2ix
f(x) dx,
where x = x
1
1
+ x
n
n
is the euclidean inner-product. The function
f is called the Fourier transform of f.

(i) Prove that

f
1
+f
2
=

f
1
+

f
2
and

f =
f.
(ii) Prove that

f g =

f g, where fg is the convolution dened in Exercise
8.5.9.
(iii) If g(x) = f(x a) for a.e. x R
n
, prove that g() = e
2ia
f() for
all R
n
.
(iv) If g(x) = e
2iax
f(x) for a.e. x R
n
, prove that g() =

f( +a) for
all R
n
.
(v) If g(x) = f(x) for a.e. x R
n
, prove that g() =

f() for all
R
n
.
(vi) If T : R
n
R
n
is a linear transformation with det(T) ,= 0 and
g(x) = f(Tx) for a.e. x R
n
, prove that g() =
1
det(T)
f
_
(T
)
1
()
_
for
all R
n
, where T
is the adjoint of T.
(vii) Prove that

f is continuous on R
n
.
(viii) Prove that [
f()[
_
R
n
[f[ dm
n
for every R
n
.
11. Let C be a Cantor-type set in [0, 1] with m
1
(C) > 0. Prove that the set
(x, y) [0, 1] [0, 1] [ xy C is a compact subset of R
2
with positive
m
2
-measure, which does not contain any measurable interval of positive
m
2
-measure.
12. Uniqueness of Lebesgue-measure.
Let and be two locally nite Borel measures on R
n
, which are trans-
lation invariant. Namely: (A+x) = (A) and (A+x) = (A) for every
x R
n
and every A B
R
n.
Working with
_
R
n
R
n

A
(x)
B
(x + y) d( )(x, y), prove that either
= or = for some [0, +).
Conclude that the only locally nite Borel measure on R
n
which has value
1 at the unit cube [0, 1]
n
is the Lebesgue-measure m
n
.
13. Let E [0, 1] [0, 1] have the property that every horizontal section E
x
is countable and every vertical section E
y
has countable complementary
set [0, 1] E
y
. Prove that E is not Lebesgue-measurable.
14. Let (X, , ) be a measure space and (Y,
) be a measurable space. Sup-

pose that for every x X there exists a measure
x
on (Y,
) so that for
every B
the function x
x
(B) is measurable.
We dene (B) =
_
X

x
(B) d(x) for every B
.
(i) Prove that is a measure on (Y,
).
(ii) If g : Y [0, +] is
measurable and if f(x) =

_
Y
g d
x
for every
x X, prove that f is measurable and
_
X
f d =
_
Y
g d.
15. Interchange of successive summations.
If I
1
, I
2
are two sets of indices with their counting measures, prove that
the product-measure on I
1
I
2
is its counting measure.
Applying the theorems of Tonelli and Fubini, derive results about the
validity of
i1I1,i2I2
c
i1,i2
=
i1I1
_

i2I2
c
i1,i2
_
=
i2I2
_

i1I1
c
i1,i2
_
.
16. Consider, for every p (0, +), the function f : R
n
[0, +], dened
by f(x) =
1
|x|
p
.
(i) Prove that f is not Lebesgue-integrable over R
n
.
(ii) Prove that f is integrable over the set A
= x R
n
[ 0 < [x[ if
and only if p > 1.
(iii) Prove that f is integrable over the set B
R
= x R
n
[ [x[ R < +
if and only if p < 1.
17. Suppose that (Y, ) and (X
i
,
i
) are measurable spaces for all i I and
that g : X
i0
Y is (
i0
, )measurable. If we dene f :
iI
X
i
Y
by f
_
(x
i
)
iI
_
= g(x
i0
), prove that f is (
iI

i
, )measurable.
8.5. EXERCISES. 167
18. Integration by parts.
Consider the interval

R = (a, b] (a, b] and partition it into the two
sets
1
= (t, s)

R[ t s and
2
= (t, s)

R[ s < t. Writing
(
G
F
)(

R) = (
G
F
)(
1
) +(
G
F
)(
2
), prove Proposition 7.11.
Chapter 9
Convergence of functions
9.1 a.e. convergence and uniformly a.e. conver-
gence.
The two types of convergence of sequences of functions which are usually stud-
ied in elementary courses are the pointwise convergence and the uniform con-
vergence. We, briey, recall their denitions and simple properties.
Suppose A is an arbitrary set and f, f
n
: A R or C for every n N. We
say that f
n
converges to f pointwise on A if f
n
(x) f(x) for every x A.
In case f(x) is nite, this means that for every > 0 there is an n
0
= n
0
(, x)
so that: [f
n
(x) f(x)[ for every n n
0
.
Suppose A is an arbitrary set and f, f
n
: A C for every n N. We say
that f
n
converges to f uniformly on A if for every > 0 there is an n
0
= n
0
()
so that: [f
n
(x) f(x)[ for every x A and every n n
0
or, equivalently,
sup
xA
[f
n
(x) f(x)[ for every n n
0
. In other words, f
n
converges to
f uniformly on A if and only if sup
xA
[f
n
(x) f(x)[ 0 as n +.
It is obvious that uniform convergence on A of f
n
to f implies pointwise
convergence on A. The converse is not true in general. As a counter-example,
if f
n
=
(0,
1
n
)
for every n, then f
n
converges to f = 0 pointwise on (0, 1) but
not uniformly on (0, 1).
Let us describe some easy properties.
The pointwise limit (if it exists) of a sequence of functions is unique and,
hence, the same is true for the uniform limit.
Assume that f, g, f
n
, g
n
: A C for all n. If f
n
converges to f and
g
n
converges to g pointwise on A, then f
n
+ g
n
converges to f + g and
f
n
g
n
converges to fg pointwise on A. The same is true for uniform conver-
gence, provided that in the case of the product we also assume that the two
sequences are uniformly bounded: this means that there is an M < + so that
[f
n
(x)[, [g
n
(x)[ M for every x A and every n N.
Another well-known fact is that, if f
n
: A C for all n and f
n
is Cauchy
uniformly on A, then there is an f : A C so that f
n
converges to f
169
170 CHAPTER 9. CONVERGENCE OF FUNCTIONS
uniformly on A. Indeed, suppose that for every > 0 there is an n
0
= n
0
() so
that: [f
n
(x) f
m
(x)[ for every x A and every n, m n
0
. This implies
that, for every x, the sequence f
n
(x) is a Cauchy sequence of complex numbers
and, hence, it converges to some complex number. If we dene f : A C by
f(x) = lim
n+
f
n
(x) and if in the above inequality [f
n
(x) f
m
(x)[ we
let m +, we get that [f
n
(x) f(x)[ for every x A and every n n
0
.
Hence, f
n
converges to f uniformly on A.
It is almost straightforward to extend these two notions of convergence to
measure spaces. Suppose that (X, , ) is an arbitrary measure space.
We have already seen the notion of -a.e. convergence. If f, f
n
: X R or
C for every n, we say that f
n
converges to f (pointwise) -a.e. on A
if there is a set B , B A, so that (A B) = 0 and f
n
converges to f
pointwise on B.
If f, f
n
: X Ror Cfor every n, we say that f
n
converges to f uniformly
-a.e. on A if there is a set B , B A, so that (A B) = 0, f and
f
n
are nite on B for all n and f
n
converges to f uniformly on B.
It is clear that uniform convergence -a.e. on A implies convergence -a.e.
on A. The converse is not true in general and the counter-example is the same
as above.
If f
n
converges to both f and f
-a.e. on A, then f = f
-a.e. on A.
Indeed, there are B, B
with B, B
A so that (A B) = (A B
) = 0
and f
n
converges to f pointwise on B and to f
pointwise on B
. Therefore, f
n
pointwise on B B
and, hence, f = f
on B B
.
Since (A (B B
)) = 0, we get that f = f
a.e. on A. This is a common

feature of almost any notion of convergence in the framework of measure spaces:
the limits may be considered unique only if we agree to identify functions which
are equal a.e. on A. This can be made precise by using the tool of equivalence
classes in an appropriate manner, but we postpone this discussion for later.
We can, similarly, prove that if f
n
uniformly
-a.e. on A, then f = f
-a.e. on A.
Moreover, if f, g, f
n
, g
n
: A C -a.e. on A for every n and f
n
converges
to f and g
n
converges to g -a.e. on A, then f
n
+g
n
converges to f +g and
f
n
g
n
converges to fg -a.e. on A. The same is true for uniform convergence
-a.e., provided that in the case of the product we also assume that the two
sequences are uniformly bounded -a.e.: namely, that there is an M < + so
that [f
n
[, [g
n
[ M -a.e. on A for every n N.
9.2 Convergence in the mean.
Assume that (X, , ) is a measure space.
Denition 9.1 Let f, f
n
: X R or C be measurable for all n. We say
that f
n
converges to f in the mean on A if f and f
n
are nite -a.e.
on A for all n and
_
A
[f
n
f[ d 0
9.2. CONVERGENCE IN THE MEAN. 171
as n +.
We say that f
n
is Cauchy in the mean on A if f
n
is nite -a.e.
on A for all n and
_
A
[f
n
f
m
[ d 0
as m, n +.
It is necessary to make a comment regarding the denition. The functions
[f
n
f[ and [f
n
f
m
[ are dened only -a.e. on A. In fact, if all f, f
n
are nite
on B with B A and (AB) = 0, then [f
n
f[ and [f
n
f
m
[ are all dened
on B and are
B
measurable. Therefore, only the integrals
_
B
[f
n
f[ d and
_
B
[f
n
f
m
[ d are well-dened. If we want to be able to write the integrals
_
A
[f
n
f[ d and
_
A
[f
n
f
m
[ d, we must extend the functions [f
n
f[ and
[f
n
f
m
[ on X so that they are measurable and, after that, the integrals
_
A
[f
n
f[ d and
_
A
[f
n
f
m
[ d will be dened and equal to
_
B
[f
n
f[ d
and
_
B
[f
n
f
m
[ d, respectively. Since the values of the extensions outside B
do not aect the resulting values of the integrals over A, it is simple and enough
to extend all f, f
n
as 0 on X B.
Thus, the replacement of all f, f
n
by 0 on X B makes all functions nite
everywhere on A without aecting the fact that f
n
converges to f in the mean
on A or that f
n
is Cauchy in the mean on A.
Proposition 9.1 If f
n
in the mean on A, then

f = f
-a.e. on A.
Proof: By the comment of the previous paragraph, we may assume that all f, f
and f
n
are nite on A. This does not aect either the hypothesis or the result
of the statement.
We write
_
A
[f f
[ d
_
A
[f
n
f[ d +
_
A
[f
n
f
[ d 0 as n +.
Hence,
_
A
[f f
[ d = 0, implying that f = f
-a.e. on A.
Proposition 9.2 Suppose f
n
converges to f and g
n
converges to g in the
mean on A and C. Then
(i) f
n
+g
n
converges to f +g in the mean on A.
(ii) f
n
converges to f in the mean on A.
Proof: We may assume that all f, g, f
n
, g
n
are nite on A.
Then,
_
A
[(f
n
+ g
n
) (f + g)[ d
_
A
[f
n
f[ d +
_
A
[g
n
g[ d 0 as
n +, and
_
A
[f
n
f[ d = [[
_
A
[f
n
f[ d 0 as n +.
It is trivial to prove that, if f
n
converges to f in the mean on A, then
f
n
is Cauchy in the mean on A. Indeed, assuming all f, f
n
are nite on A,
_
A
[f
n
f
m
[ d
_
A
[f
n
f[ d +
_
A
[f
m
f[ d 0 as n, m +. The
following basic theorem expresses the converse.
Theorem 9.1 If f
n
is Cauchy in the mean on A, then there is f : X C
so that f
n
converges to f in the mean on A. Moreover, there is a subsequence
f
n
k
which converges to f -a.e. on A.
As a corollary: if f
n
converges to f in the mean on A, there is a subse-
quence f
n
k
Proof: As usual, we assume that all f, f
n
are nite on A.
We have that, for every k, there is n
k
so that
_
A
[f
n
f
m
[ d <
1
2
k
for
every n, m n
k
. Since we may assume that each n
k
is as large as we like,
we inductively take n
k
so that n
k
< n
k+1
for every k. Therefore, f
n
k
is a
subsequence of f
n
.
From the construction of n
k
and from n
k
< n
k+1
, we get that
_
A
[f
n
k+1
f
n
k
[ d <
1
2
k
for every k. Then, the measurable function G : X [0, +] dened by
G =
_
+
k=1
[f
n
k+1
f
n
k
[, on A
0, on A
c
satises
_
X
Gd =
+
k=1
_
A
[f
n
k+1
f
n
k
[ d = 1 < +. Thus, G < +
-a.e. on A and, hence, the series
+
k=1
(f
n
k+1
(x) f
n
k
(x)) converges for -
a.e. x A. Therefore, there is a B , B A so that (A B) = 0
and
+
k=1
(f
n
k+1
(x) f
n
k
(x)) converges for every x B. We dene the
measurable f : X C by
f =
_
f
n1
+
+
k=1
(f
n
k+1
f
n
k
), on B
0, on B
c
.
On B we have that f = f
n1
+lim
K+
K1
k=1
(f
n
k+1
f
n
k
) = lim
K+
f
nK
and, hence, f
n
k
converges to f -a.e. on A.
We, also, have on B that [f
nK
f[ = [f
nK
f
n1

+
k=1
(f
n
k+1
f
n
k
)[ =
[
K1
k=1
(f
n
k+1
f
n
k
)
+
k=1
(f
n
k+1
f
n
k
)[

+
k=K
[f
n
k+1
f
n
k
[ for all K.
Hence,
_
A
[f
nK
f[ d
+
k=K
_
A
[f
n
k+1
f
n
k
[ d <
+
k=K
1
2
k
=
1
2
K1
0
as K +.
From n
k
+, we get
_
A
[f
k
f[ d
_
A
[f
k
f
n
k
[ d+
_
A
[f
n
k
f[ d 0
as k + and we conclude that f
n
converges to f in the mean on A.
Example
Consider the sequence f
1
=
(0,1)
, f
2
=
(0,
1
2
)
, f
3
=
(
1
2
,1)
, f
4
=
(0,
1
3
)
, f
5
=
(
1
3
,
2
3
)
, f
6
=
(
2
3
,1)
, f
7
=
(0,
1
4
)
, f
8
=
(
1
4
,
2
4
)
, f
9
=
(
2
4
,
3
4
)
, f
10
=
(
3
4
,1)
and so on.
It is clear that
_
(0,1)
[f
n
(x)[ dx 0 as n + (the sequence of integrals
is 1,
1
2
,
1
2
,
1
3
,
1
3
,
1
3
,
1
4
,
1
4
,
1
4
,
1
4
, . . .) and, hence, f
n
converges to 0 in the mean on
(0, 1). By Theorem 9.1, there exists a subsequence converging to 0 m
1
-a.e. on
9.3. CONVERGENCE IN MEASURE. 173
(0, 1) and it is easy to nd many such subsequences: indeed, f
1
=
(0,1)
, f
2
=
(0,
1
2
)
, f
4
=
(0,
1
3
)
, f
7
=
(0,
1
4
)
and so on, is one such subsequence.
But, it is not true that f
n
itself converges to 0 m
1
-a.e. on (0, 1). In fact, if
x is any irrational number in (0, 1), then x belongs to innitely many intervals
of the form (
k1
m
,
k
m
) (for each value of m there is exactly one such value of k)
and, thus, f
n
(x) does not converge to 0. It easy to see that f
n
(x) 0 only
for every rational x (0, 1).
We may now complete Proposition 9.2 as follows.
n
n
converges to g in the
mean on A.
(i) If there is M < + so that [f
n
[ M -a.e. on A, then [f[ M -a.e. on
A.
(ii) If there is an M < + so that [f
n
[, [g
n
[ M -a.e. on A, then f
n
g
n
converges to fg in the mean on A.

Proof: (i) Theorem 9.1 implies that there is a subsequence f
n
k
which con-
verges to f -a.e. on A. Therefore, [f
n
k
[ [f[ -a.e. on A and, hence, [f[ M
-a.e. on A.
(ii) Assuming that all f, g, f
n
, g
n
are nite on A and using the result of (i),
_
A
[f
n
g
n
fg[ d
_
A
[f
n
g
n
fg
n
[ d +
_
A
[fg
n
fg[ d M
_
A
[f
n
f[ d +
M
_
A
[g
n
g[ d 0 as n +.
9.3 Convergence in measure.
n
: X R or C be measurable for all n. We say
that f
n
converges to f in measure on A if all f, f
n
are nite -a.e.
on A and if for every > 0 we have
(x A[ [f
n
(x) f(x)[ ) 0
as n +.
We say that f
n
is Cauchy in measure on A if all f
n
are nite
-a.e. on A and if for every > 0 we have
(x A[ [f
n
(x) f
m
(x)[ ) 0
as n, m +.
We make a comment similar to the comment following Denition 9.1. If
we want to be able to write the values (x A[ [f
n
(x) f(x)[ ) and
(x A[ [f
n
(x) f(x)[ ), we rst extend the functions [f
n
f[ and
[f
n
f
m
[ outside the set B A, where all f, f
n
are nite, as functions dened
on X and measurable. Then, since (A B) = 0, we get that the above
values are equal to the values (x B[ [f
n
(x) f(x)[ ) and, respectively,
(x B[ [f
n
(x) f(x)[ ). Therefore, the actual extensions play no role
and, hence, we may for simplicity extend all f, f
n
as 0 on X B.
Thus the replacement of all f, f
n
by 0 on X B makes all functions nite
everywhere on A and does not aect the fact that f
n
converges to f in measure
on A or that f
n
is Cauchy in measure on A.
A useful trick is the inequality
(x A[ [f(x) +g(x)[ a +b) (x A[ [f(x)[ a)
+ (x A[ [g(x)[ b),
which is true for every a, b > 0. This is due to the set-inclusion
x A[ [f(x) +g(x)[ a +b x A[ [f(x)[ a x A[ [g(x)[ b.
n
in measure on A, then
f = f
-a.e. on A.
Proof: We may assume that all f, f
, f
n
are nite on A.
Applying the above trick we nd that (x A[ [f(x) f
(x)[ )
(x A[ [f
n
(x) f(x)[

2
) + (x A[ [f
n
(x) f
(x)[

2
) 0 as
n +. This implies (x A[ [f(x) f
(x)[ ) = 0 for every > 0.

We, now, write x A[ f(x) ,= f
(x) =
+
k=1
x A[ [f(x) f
(x)[
1
k
.
Since all terms in the union are -null sets, we get (x A[ f(x) ,= f
(x)) = 0
and conclude that f = f
-a.e. on A.
n
n
converges to g in mea-
sure on A and C. Then
(i) f
n
+g
n
converges to f +g in measure on A.
(ii) f
n
converges to f in measure on A.
(iii) If there is M < + so that [f
n
[ M -a.e. on A, then [f[ M -a.e.
on A.
(iv) If there is M < + so that [f
n
[, [g
n
n
g
n
converges to fg in measure on A.
n
are nite on A, since all hypotheses and all
results to be proved are not aected by any change of the functions on a subset
of A of zero -measure.
(i) We apply the usual trick and (x A[ [(f
n
+g
n
)(x) (f +g)(x)[ )
(x A[ [f
n
(x) f(x)[

2
) + (x A[ [g
n
(x) g(x)[

2
) 0 as
n +.
(ii) Also (x A[ [f
n
(x)f(x)[ ) = (x A[ [f
n
(x)f(x)[

||
)
0 as n +.
(iii) We write (x A[ [f(x)[ M + ) (x A[ [f
n
(x)[ M +

2
) +
(x A[ [f
n
(x) f(x)[

2
) = (x A[ [f
n
(x) f(x)[

2
) 0 as
n +. Hence, (x A[ [f(x)[ M +) = 0 for every > 0.
We have x A[ [f(x)[ > M

+
k=1
x A[ [f(x)[ M +
1
k
and, since
all sets of the union are -null, we nd that (x A[ [f(x)[ > M) = 0.
Hence, [f[ M -a.e. on A.
9.3. CONVERGENCE IN MEASURE. 175
(iv) Applying the result of (iii), (x A[ [f
n
(x)g
n
(x) f(x)g(x)[ )
(x A[ [f
n
(x)g
n
(x) f
n
(x)g(x)[

2
) +(x A[ [f
n
(x)g(x) f(x)g(x)[
2
) (x A[ [g
n
(x) g(x)[

2M
) +(x A[ [f
n
(x) f(x)[

2M
) 0
as n +.
If f
n
converges to f in measure on A, then f
n
is Cauchy in measure
on A. Indeed, taking all f, f
n
nite on A, (x A[ [f
n
(x) f
m
(x)[ )
(x A[ [f
n
(x) f(x)[

2
) + (x A[ [f
m
(x) f(x)[

2
) 0 as
n, m +.
Theorem 9.2 If f
n
is Cauchy in measure on A, then there is f : X C
so that f
n
converges to f in measure on A. Moreover, there is a subsequence
f
n
k
n
converges to f in measure on A, there is a subsequence
f
n
k
Proof: As usual, we assume that all f
n
are nite on A.
We have, for all k, (x A[ [f
n
(x) f
m
(x)[
1
2
k
) 0 as n, m +.
Therefore, there is n
k
so that (x A[ [f
n
(x) f
m
(x)[
1
2
k
) <
1
2
k
for
every n, m n
k
k
is as large as we like, we
may inductively take n
k
so that n
k
< n
k+1
for every k. Hence, f
n
k
is a
subsequence of f
n
and, from the construction of n
k
and from n
k
< n
k+1
, we
get that
(x A[ [f
n
k+1
(x) f
n
k
(x)[
1
2
k
) <
1
2
k
for every k. For simplicity, we write
E
k
= x A[ [f
n
k+1
(x) f
n
k
(x)[
1
2
k
and, hence, (E
k
) <
1
2
k
for all k. We also dene the subsets of A:
F
m
=
+
k=m
E
k
, F =
+
m=1
F
m
= limsup E
k
.
Now, (F
m
)
+
k=m
(E
k
) <
+
k=m
1
2
k
=
1
2
m1
and, hence, (F)
(F
m
) <
1
2
m1
for every m. This implies
(F) = 0.
If x AF, then there is m so that x AF
m
, which implies that x AE
k
for all k m. Therefore, [f
n
k+1
(x) f
n
k
(x)[ <
1
2
k
for all k m, so that
+
k=m
[f
n
k+1
(x) f
n
k
(x)[ <
1
2
m1
. Thus, the series
+
k=m
(f
n
k+1
(x) f
n
k
(x))
converges and we may dene f : X C by
f =
_
f
n1
(x) +
+
k=1
(f
n
k+1
f
n
k
), on A F
0, on A
c
F.
By f(x) = f
n1
(x)+lim
K+
K1
k=1
(f
n
k+1
(x)f
n
k
(x)) = lim
K+
f
nK
(x)
for every x A F and, from (F) = 0, we get that f
n
k
converges to f -a.e.
on A.
Now, on A F
m
we have [f
nm
f[ = [f
nm
f
n1

+
k=1
(f
n
k+1
f
n
k
)[ =
[
m1
k=1
(f
n
k+1
f
n
k
)

+
k=1
(f
n
k+1
f
n
k
)[

+
k=m
[f
n
k+1
f
n
k
[ <
1
2
m1
.
Therefore, x A[ [f
nm
(x) f(x)[
1
2
m1
F
m
and, hence,
(x A[ [f
nm
(x) f(x)[
1
2
m1
) (F
m
) <
1
2
m1
.
Take an arbitrary > 0 and m
0
large enough so that
1
2
m
0
1
. If m m
0
,
x A[ [f
nm
(x) f(x)[ x A[ [f
nm
(x) f(x)[
1
2
m1
and, hence,
(x A[ [f
nm
(x) f(x)[ ) <
1
2
m1
0
as m +. This means that f
n
k
Since n
k
+ as k +, we get (x A[ [f
k
(x) f(x)[ )
(x A[ [f
k
(x) f
n
k
(x)[

2
) + (x A[ [f
n
k
(x) f(x)[

2
) 0 as
k + and we conclude that f
n
Example
We consider a variation of the example just after Theorem 9.1. Consider the
sequence f
1
=
(0,1)
, f
2
= 2
(0,
1
2
)
, f
3
= 2
(
1
2
,1)
, f
4
= 3
(0,
1
3
)
, f
5
= 3
(
1
3
,
2
3
)
, f
6
=
3
(
2
3
,1)
, f
7
= 4
(0,
1
4
)
, f
8
= 4
(
1
4
,
2
4
)
, f
9
= 4
(
2
4
,
3
4
)
, f
10
= 4
(
3
4
,1)
and so on.
If 0 < 1, the sequence of the values (x (0, 1) [ [f
n
(x)[ ) is
1,
1
2
,
1
2
,
1
3
,
1
3
,
1
3
,
1
4
,
1
4
,
1
4
,
1
4
, . . . and, hence, converges to 0. Therefore, f
n
con-
verges to 0 in measure on (0, 1). But, as we have already seen, it is not true
that f
n
converges to 0 m
1
-a.e. on (0, 1).
9.4 Almost uniform convergence.
n
: X R or C be measurable for all n N. We
say that f
n
converges to f almost uniformly on A if for every > 0
there is B , B A, so that (A B) < and f
n
on B.
We say that f
n
is Cauchy almost uniformly on A if for every
> 0 there is B , B A, so that (A B) < and f
n
is Cauchy
uniformly on B.
Suppose that some g : X R or C is measurable and that, for every k,
there is a B
k
, B
k
A, with (AB
k
) <
1
k
so that g is nite on B
k
. Now, it
is clear that g is nite on the set F =
+
k=1
B
k
and that (AF) (AB
k
) <
1
k
for all k. This implies that (A F) = 0 and, hence, g is nite -a.e. on A.
From the statement of Denition 9.3. it is implied by the uniform conver-
gence that all functions f, f
n
are nite on sets B , B A with (AB) < .
Since is arbitrary, by the discussion in the previous paragraph, we conclude
that, if f
n
converges to f almost uniformly on A or if it is Cauchy almost
9.4. ALMOST UNIFORM CONVERGENCE. 177
uniformly on A, then all f, f
n
are nite -a.e. on A. Now, if F , F A
with (A F) = 0 is the set where all f, f
n
are nite, then, if we replace all
f, f
n
by 0 on X F, the resulting functions f, f
n
are all nite on A and the
fact that f
n
converges to f almost uniformly on A or that it is Cauchy almost
uniformly on A is not aected.
n
almost uniformly on A,
then f = f
-a.e. on A.
Proof: Suppose that (x A[ f(x) ,= f
(x)) > 0. For simplicity, we set

E = x A[ f(x) ,= f
(x).
We nd B , B A, with (A B) <
(E)
2
so that f
n
converges to f
uniformly on B. We, also, nd B
, B
A, with (A B
) <
(E)
2
so that
f
n
converges to f
uniformly on B
. We, then, set D = B B
and have that

(A D) < (E) and f
n
uniformly on D. This,
of course, implies that f = f
on D and, hence, that D E = .

But, then, E A D and, hence, (E) (A D) < (E) and we arrive
at a contradiction.
n
n
converges to g almost
uniformly on A. Then
(i) f
n
+g
n
converges to f +g almost uniformly on A.
(ii) f
n
converges to f almost uniformly on A.
(iii) If there is M < + so that [f
n
[ M -a.e. on A, then [f[ M -a.e.
on A.
(iv) If there is M < + so that [f
n
[, [g
n
n
g
n
converges to fg almost uniformly on A.

n
are nite on A.
(i) For arbitrary > 0, there is B
, B
A, with (AB
) <

2
so that f
n
converges to f uniformly on B
and there is B
, B
A, with (AB
) <

2
so that g
n
converges to g uniformly on B
. We take B = B
and have
that (A B) < and that f
n
and g
n
converge to f and, respectively, g
uniformly on B. Then f
n
+g
n
converges to f +g uniformly on B and, since
is arbitrary, we conclude that f
n
+ g
n
converges to f +g almost uniformly
on A.
(ii) This is easier, since, if f
n
converges to f uniformly on B, then f
n
(iii) Suppose (x A[ [f(x)[ > M) > 0 and set E = x A[ [f(x)[ > M.
We nd B , B A, with (A B) < (E) so that f
n
converges to f
uniformly on B. Then we have [f[ M -a.e. on B and, hence, (B E) = 0.
Now, (E) = (E B) (A B) < (E) and we arrive at a contradiction.
(iv) Exactly as in the proof of (i), for every > 0 we nd B
1
, B
1
A,
with (A B
1
) < so that f
n
and g
n
converge to f and, respectively, g
uniformly on B
1
. By the result of (iii), [f[ M -a.e. on A and, hence, there
is a B
2
, B
2
A with (A B
2
) = 0 so that [f
n
[, [g
n
[, [f[ M on B
2
. We
set B = B
1
B
2
, so that (A B) = (A B
1
) < . Now, on B we have that
[f
n
g
n
fg[ [f
n
g
n
fg
n
[ + [fg
n
fg[ M[f
n
f[ + M[g
n
g[ and, thus,
f
n
g
n
converges to fg uniformly on B. We conclude that f
n
g
n
converges to
fg almost uniformly on A.
One should notice the dierence between the next result and the correspond-
ing Theorems 9.1 and 9.2 for the other two types of convergence: if a sequence
converges in the mean or in measure, then a.e. convergence holds for some sub-
sequence, while, if it converges almost uniformly, then a.e. convergence holds
for the whole sequence (and, hence, for every subsequence).
Before the next result, let us consider a simple general fact.
Assume that there is a collection of functions g
i
: B
i
C, indexed by the
set I of indices, where B
i
X for every i I, and that f
n
converges to
g
i
pointwise on B
i
, for every i I. If x B
i
B
j
for any i, j I, then, by
the uniqueness of pointwise limits, we have that g
i
(x) = g
j
(x). Therefore, all
limit-functions have the same value at each point of the union B =
iI
B
i
of
the domains of denition. Hence, we can dene a single function f : B C by
f(x) = g
i
(x),
where i I is any index for which x B
i
, and it is clear that f
n
converges
to f pointwise on B.
Theorem 9.3 If f
n
is Cauchy almost uniformly on A, then there is an f :
X C so that f
n
converges to f almost uniformly on A. Moreover, f
n
n
converges to f almost uniformly on A, then f
n
Proof: For each k, there exists B
k
, B
k
A, with (A B
k
) <
1
k
so that
f
n
is Cauchy uniformly on B
k
. Therefore, there is a function g
k
: B
k
C so
that f
n
converges to g
k
uniformly and, hence, pointwise on B
k
.
By the general result of the paragraph just before this theorem, there is an
f : B C, where B =
+
k=1
B
k
, so that f
n
converges to f pointwise on B.
But, (A B) (A B
k
) <
1
k
for every k and, thus, (A B) = 0. If we
extend f : X C, by dening f = 0 on B
c
, we conclude that f
n
converges
to f -a.e. on A.
By the general construction of f, we have that g
k
= f on B
k
and, hence,
f
n
k
. If > 0 is arbitrary, we just take k large
enough so that
1
k
and we have that (A B
k
) < . Hence, f
n
converges
to f almost uniformly on A.
9.5 Relations between types of convergence.
In this section we shall see three results describing some relations between the
four types of convergence: a.e. convergence, convergence in the mean, con-
vergence in measure and almost uniform convergence. Many other results are
consequences of these.
9.5. RELATIONS BETWEEN TYPES OF CONVERGENCE. 179
Theorem 9.4 If f
n
n
con-
verges to f -a.e. on A.
The converse is true under the additional assumption that either
(i) (Egoro) all f, f
n
are nite -a.e. on A and (A) < +
or
(ii) there is a g : A [0, +] with
_
A
g d < + and [f
n
[ g -a.e. on A for
every n.
Proof: The rst statement is inluded in Theorem 9.3.
(i) Assume f
n
converges to f -a.e. on A, all f, f
n
are nite -a.e. on A and
(A) < +. We may assume that all f, f
n
are nite on A and, for each k, n,
we dene
E
n
(k) =
+
_
m=n
_
x A[ [f
m
(x) f(x)[ >
1
k
_
.
If C = x A[ f
n
(x) f(x), then it is easy to see that
+
n=1
E
n
(k) A C.
Since (A C) = 0, we get (
+
n=1
E
n
(k)) = 0 for every k. From E
n
(k)
+
n=1
E
n
(k), from (A) < + and from the continuity of from above, we nd
that (E
n
(k)) 0 as n +. Hence, for an arbitrary > 0, there is n
k
so
that
(E
n
k
(k)) <

2
k
.
We dene
E =
+
_
k=1
E
n
k
(k), B = A E
and have (E)
+
k=1
(E
n
k
(k)) < . Also, for every x B we have that, for
every k 1, [f
m
(x) f(x)[
1
k
for all m n
k
. Equivalently, for every k 1,
sup
xB
[f
m
(x) f(x)[
1
k
for every m n
k
. This implies, of course, that f
n
on B. Since (A B) = (E) < , we conclude that f
n
converges to f almost
uniformly on A.
(ii) If [f
n
[ g -a.e. on A for all n, then also [f[ g -a.e. on A and, since
_
A
g d < +, all f, f
n
are nite -a.e. on A. Assuming, as we may, that all
f, f
n
are nite on A, we get [f
n
f[ 2g -a.e. on A for all n. Using the same
notation as in the proof of (i), this implies that E
n
(k) x A[ g(x) >
1
2k
except for a -null set. Therefore

(E
n
(k)) (x A[ g(x) >
1
2k
)
for every n, k. It is clear that the assumption
_
A
g d < + implies
(x A[ g(x) >
1
2k
) < +.
Therefore, we may, again, apply the continuity of from above to nd that
(E
n
(k)) 0 as n +. From this point, we repeat the proof of (i) word for
word.
Example
If f
n
=
(n,n+1)
for every n 1, then f
n
converges to 0 everywhere on R, but
f
n
does not converge to 0 almost uniformly on R. In fact, if 0 < 1, then
every Lebesgue-measurable B R with m
1
(R B) < must have non-empty
intersection with every interval (n, n + 1) and, hence, sup
xB
[f
n
(x)[ 1 for
every n.
In this example, of course, m
1
(R) = + and it is easy to see that there is
no g : R [0, +] with
_
R
g(x) dx < + satisfying f
n
g m
1
-a.e. on R for
every n. Otherwise, g 1 a.e. on (1, +).
Theorem 9.5 If f
n
n
con-
verges to f in measure on A.
Conversely, if f
n
converges to f in measure on A, then there is a subse-
quence f
n
k
which converges to f almost uniformly on A.
Proof: Suppose that f
n
converges to f almost uniformly on A and take an
arbitrary > 0. For every > 0 there is a B , B A, with (A B) < so
that f
n
Now, there exists an n
0
so that [f
n
(x) f(x)[ < for all n n
0
and
every x B. Therefore, x A[ [f
n
(x) f(x)[ A B and, thus,
(x A[ [f
n
(x) f(x)[ ) < for all n n
0
.
This implies that (x A[ [f
n
(x) f(x)[ ) 0 as n + and f
n
The idea for the converse is already in the proof of Theorem 9.2.
We assume that f
n
converges to f in measure on A and, without loss of
generality, that all f, f
n
are nite on A. Then (x A[ [f
n
(x)f(x)[
1
2
k
)
0 as n + and there is n
k
so that (x A[ [f
n
(x) f(x)[
1
2
k
) <
1
2
k
for
all n n
k
. We may, inductively, assume that n
k
< n
k+1
for all k and, hence,
that f
n
k
is a subsequence of f
n
for which
(x A[ [f
n
k
(x) f(x)[
1
2
k
) <
1
2
k
for every k 1. We set
E
k
= x A[ [f
n
(x) f(x)[
1
2
k
, F
m
=
+
k=m
E
k
.
Then (F
m
) <
+
k=m
1
2
k
=
1
2
m1
for every m.
If x A F
m
, then x A E
k
for every k m so that [f
n
k
(x) f(x)[ <
1
2
k
for every k m. This implies that
sup
xA\Fm
[f
n
k
(x) f(x)[
1
2
k
9.5. RELATIONS BETWEEN TYPES OF CONVERGENCE. 181
for all k m and hence sup
xA\Fm
[f
n
k
(x) f(x)[ 0 as k +. Therefore,
f
n
k
converges to f uniformly on AF
m
and we conclude that f
n
k
converges
to f almost uniformly on A.
Example
We consider the example just after Theorem 9.2. The sequence f
n
converges
to 0 in measure on (0, 1) but it does not converge to 0 almost uniformly on (0, 1).
In fact, if we take any with 0 < 1, then every B (0, 1) with m
1
((0, 1)
B) < must have non-empty intersection with innitely many intervals of the
form (
k1
m
,
k
m
) (at least one for every value of m) and, hence, sup
xB
[f
n
(x)[ 1
for innitely many n.
The converse in Theorem 9.6 is a variant of the Dominated Convergence
Theorem.
Theorem 9.6 If f
n
converges to f in the mean on A, then f
n
converges
to f in measure on A.
The converse is true under the additional assumption that there exists a
g : X [0, +] so that
_
A
g d < + and [f
n
[ g -a.e. on A.
Proof: It is clear that we may assume all f, f
n
are nite on A.
Suppose that f
n
converges to f in the mean on A. Then, for every > 0
we have
(x A[ [f
n
(x) f(x)[ )
1
_
A
[f
n
f[ d 0
as n +. Therefore, f
n
Assume that the converse is not true. Then there is some
0
> 0 and a
subsequence f
n
k
of f
n
so that
_
A
[f
n
k
f[ d
0
for every k 1. Since f
n
k
converges to f in measure, Theorem 9.2 implies
that there is a subsequence f
n
k
l
which converges to f -a.e. on A. From
[f
n
k
l
[ g -a.e. on A, we nd that [f[ g -a.e. on A. Now, the Dominated
Convergence Theorem implies that
_
A
[f
n
k
l
f[ d 0
as l + and we arrive at a contradiction.
Example
Let f
n
= n
(0,
1
n
)
for every n. If 0 < 1, then (x (0, 1) [ [f
n
(x)[ ) =
1
n
0 as n + and, hence, f
n
converges to 0 in measure on (0, 1). But
_
1
0
[f
n
(x)[ dx = 1 and f
n
does not converge to 0 in the mean on (0, 1).
If g : (0, 1) [0, +] is such that [f
n
[ g m
1
-a.e. on (0, 1) for every n, then
g n m
1
-a.e. in each interval [
1
n+1
,
1
n
). Hence,
_
1
0
g(x) dx
+
n=1
_ 1
n
1
n+1
ndx =
+
n=1
n(
1
n

1
n+1
) =
+
n=1
1
n+1
= +.
9.6. EXERCISES. 183
9.6 Exercises.
Except if specied otherwise, all exercises refer to a measure space (X, , ), all
sets belong to and all functions are measurable.
1. Let : C C.
(i) If is continuous and f
n
converges to f -a.e. on A, prove that
f
n
(ii) If is uniformly continuous and f
n
converges to f in measure or
almost uniformly on A, prove that f
n
converges to f in measure
or, respectively, almost uniformly on A.
2. (i) If f
n
converges to f with respect to any of the four types of con-
vergence (-a.e. or in the mean or in measure or almost uniformly) on A
and f
n
converges, also, to f
with respect to any other of the same four

types of convergence, prove that f = f
-a.e. on A.
(ii) If f
n
converges to f with respect to any of the four types of conver-
gence on A and [f
n
[ g -a.e. on A for all n, prove that [f[ g -a.e.
on A.
3. If E
n
A for every n and
En
converges to f in the mean or in measure
or almost uniformly or -a.e. on A, prove that there exists E A so that
f =
E
-a.e. on A.
4. Suppose that E
n
A for every n. Prove that
En
is Cauchy in measure
or in the mean or almost uniformly on A if and only if (E
n
E
m
) 0
as n, m +.
5. Let be the counting measure on (N, T(N)). Prove that f
n
converges
to f uniformly on N if and only if f
n
converges to f in measure on N.
6. A variant of the Lemma of Fatou.
If f
n
0 -a.e. on A and f
n
converges to f in measure on A, prove
that
_
A
f d liminf
n+
_
A
f
n
d.
7. The Dominated Convergence Theorem.
Prove the Dominated Convergence Theorem in two ways, using either the
rst converse or the second converse of Theorem 9.4.
8. A variant of the Dominated Convergence Theorem.
Suppose that [f
n
[ g -a.e. on A, that
_
A
g d < + and that f
n
converges to f in measure on A. Prove that

_
A
f
n
d
_
A
f d.
One can follow three paths. One is to use the result of Exercise 9.6.2.
Another is to reduce to the case of -a.e. convergence and use the original
version of the theorem. The third path is to use almost uniform conver-
gence.
9. Suppose that A is of -nite -measure and f
n
converges to f -a.e. on
A. Prove that, for each k, there exists E
k
A so that f
n
converges to
f uniformly on E
k
and (A
+
k=1
E
k
) = 0.
10. Suppose that E
k
() = x A[ [f
k
(x) f(x)[ for every k and > 0.
If (A) < +, prove that f
n
converges to f -a.e. on A if and only if,
for every > 0, (
+
k=n
E
k
()) 0 as n +.
11. (i) Let h
n
satisfy sup
nN
[h
n
(x)[ < for -a.e. x A. If (A) < +,
prove that for every > 0 there is a B A with (A B) < so that
sup
xB,nN
[h
n
(x)[ < +.
(ii) Let f
n
converge to f in measure on A and g
n
converge to g in
measure on A. If (A) < +, prove that f
n
g
n
converges to fg in
measure on A.
12. Suppose that (A) < + and every f
n
is nite -a.e. on A.
(i) Prove that there is a sequence
n
of positive numbers so that
n
f
n
converges to 0 -a.e. on A.
(ii) Prove that there exists g : A [0, +] and a sequence r
n
in R
+
so that [f
n
[ r
n
g -a.e. on A for every n.
13. Suppose that (A) < + and f
n
(i) Prove that there exists a sequence
n
in R
+
with
n
+ so that
n
f
n
(ii) Prove that there exists g : A [0, +] and a sequence
n
in R
+
with
n
0 so that [f
n
[
n
g -a.e. on A for every n.
14. A characterisation of convergence in measure.
If (A) < +, prove that f
n
converges to f in measure on A if and
only if
_
A
|fnf|
1+|fnf|
d 0 as n +.
In general, prove that f
n
converges to f in measure on A if and only if
inf
>0
+(x A[ [f
n
(x) f(x)[ )
1 + +(x A[ [f
n
(x) f(x)[ )
0
as n +.
15. A variant of Egoros Theorem for continuous parameter.
Let (X) < + and f : X [0, 1] C has the properties:
(a) f(, y) : X C is measurable for every y [0, 1]
(b) f(x, ) : [0, 1] C is continuous for every x X.
(i) If , > 0, prove that x X[ [f(x, y) f(x, 0)[ for all y <
belongs to .
(ii) Prove that for every > 0 there is B X with (X B) < and
f(, y) f(, 0) uniformly on B as y 0+.
16. Let f
n
converge to f in measure on A. Prove that
fn
(t)
f
(t) for
every t [0, +) which is a point of continuity of
f
.
9.6. EXERCISES. 185
17. Prove the converse part of Theorem 9.6 using the converse part of Theorem
9.5.
18. The complete relation between convergence in the mean and convergence
in measure: the Theorem of Vitali.
We say that the indenite integrals of f
n
are uniformly abso-
lutely continuous over A if for every > 0 there exists > 0 so that
[
_
E
f
n
d[ < for all n 1 and all E A with (E) < .
We say that the indenite integrals of f
n
are equicontinuous
from above at over A if for every sequence E
k
of subsets of A with
E
k
and for every > 0 there exists k
0
so that [
_
E
k
f
n
d[ < for all
k k
0
and all n 1.
Prove that f
n
converges to f in the mean on A if and only if f
n
converges to f in measure on A and the indenite integrals of f

n
are
uniformly absolutely continuous on A and equicontinuous from above at
on A.
How is Theorem 9.6 related to this result?
19. The Theorem of Lusin.
If f is Lebesgue-measurable and nite m
n
-a.e. on R
n
, then for every > 0
there is a Lebesgue-measurable B R
n
and a g, continuous on R
n
, so
that m
n
(B
c
) < and f = g on B.
(i) Use Theorem 7.16 to nd a sequence
n
of functions continuous on
R
n
so that
_
R
n
[f
n
[ dm
n
0 as n +. Theorem 9.1 implies that
there is a subsequence
n
k
which converges to f m
n
-a.e. on R
n
.
(ii) Consider the qubes Q
m1,...,mn
= [m
1
, m
1
+1) [m
n
, m
n
+1) for
every choice of m
1
, . . . , m
n
Z and enumerate them as Q
1
, Q
2
, . . .. Then,
these qubes are pairwise disjoint and they cover R
n
. Apply Egoros
Theorem to prove that for each Q
k
there is a closed set B
k
Q
k
with
m
n
(Q
k
B
k
) <

2
k
so that
n
k
k
. Conclude
that the restriction f
B
k
of f on B
k
is continuous on B
k
.
(iii) Take B =
+
k=1
B
k
and prove that m
n
(B
c
) < , that B is closed and
that the restriction f
B
of f on B is continuous on B.
(iv) Use the Extension Theorem of Tietze to prove that there is a g,
continuous on R
n
, so that g = f
B
on B.
20. If f : R
n
C is continuous in each variable separately, prove that f is
Lebesgue-measurable.
Chapter 10
Signed measures and
complex measures
10.1 Signed measures.
Denition 10.1 Let (X, ) be a measurable space. A function : R is
called a signed measure on (X, ) if
(i) either (A) ,= for all A or (A) ,= + for all A ,
(ii) () = 0,
(iii) (
+
j=1
A
j
) =
+
j=1
(A
j
) for all pairwise disjoint A
1
, A
2
, . . . .
If is a signed measure on (X, ) and (A) R for every A , then
is called a real measure. It is obvious that is a non-negative signed
measure (i.e. with (A) 0 for every A ) if and only if is a measure. If
(A) 0 for every A , then is called a a non-positive signed measure.
It is clear that, if is a non-negative signed measure, then is a non-
positive signed measure and conversely. Also, if and
are signed measures

on (X, ) with either (A),
(A) ,= for all A or (A),
(A) ,= +
for all A , then +
, well-dened by ( +
)(A) = (A) +
(A) for all

A , is a signed measure. Similarly, the , dened by ()(A) = (A) for
all A , is a signed measure for every R.
Examples
1. Let
1
,
2
be two measures on (X, ). If
2
(X) < +, then
2
(A)
2
(X) < + for every A . Then, =
1

2
is well-dened and it is a
signed measure on (X, ), because (A) =
1
(A)
2
(A)
2
(A) > for
all A . Similarly, if
1
(X) < +, then =
1

2
is a signed measure on
(X, ) with (A) < + for all A .
Hence, the dierence of two measures, at least one of which is nite, is a
signed measure.
2. Let be a measure on (X, ) and f : X R be a measurable function
187
188 CHAPTER 10. SIGNED MEASURES AND COMPLEX MEASURES
such that the
_
X
f d is dened. Lemma 7.10 says that the
_
A
f d is dened
for every A . If we consider the function : R dened by
(A) =
_
A
f d
for all A , then Proposition 7.6 and Theorem 7.13 imply that is a signed
measure on (X, ).
Denition 10.2 The signed measure which is dened in the previous para-
graph is called the indenite integral of f with respect to and it is
denoted by f. Thus, the dening relation for f is
(f)(A) =
_
A
f d, A .
In case f 0 -a.e. on X, the signed measure f is a measure, since
(f)(A) =
_
A
f d 0 for every A . Similarly, if f 0 -a.e. on X, the f
is a non-positive signed measure.
Continuing the study of this example, we shall make a few remarks. That
the
_
X
f d is dened means either
_
X
f
+
d < + or
_
X
f
d < +.
Let us consider the case
_
X
f
+
d < +rst. Then the signed measure f
+
is a nite measure (because (f

+
)(X) =
_
X
f
+
d < +) and the signed mea-
sure f
is a measure. Also, for every A we have (f

+
)(A) (f
)(A) =
_
A
f
+
d
_
A
f
d =
_
A
f d = (f)(A). Therefore, in the case
_
X
f
+
d <
+, the signed measure f is the dierence of the measures f
+
and f
d, of
which the rst is nite:
f = f
+
f
.
Similarly, in the case
_
X
f
d < +, the signed measure f is the dif-

ference of the measures f
+
and f
, of which the second is nite, since

(f
)(X) =
_
X
f
d < +.
Property (iii) in the denition of a signed measure is called the -additivity
of . It is trivial to see that a signed measure is also nitely additive.
A signed measure is not, in general, monotone: if A, B and A B, then
B = A(B A) and, hence, (B) = (A) +(B A), but (B A) may not be
0!
Theorem 10.1 Let be a signed measure on (X, ).
(i) Let A, B and A B. If (B) < +, then (A) < + and, if
(B) > , then (A) > . In particular, if (B) R, then (A) R.
(ii) If A, B , A B and (A) R, then (B A) = (B) (A).
(iii) (Continuity from below) If A
1
, A
2
, . . . and A
n
A
n+1
for all n, then
(
+
n=1
A
n
) = lim
n+
(A
n
).
(iv) (Continuity from above) If A
1
, A
2
, . . . , (A
1
) R and A
n
A
n+1
for
all n, then (
+
n=1
A
n
) = lim
n+
(A
n
).
10.2. THE HAHN AND JORDAN DECOMPOSITIONS, I. 189
Proof: (i) We have (B) = (A) +(B A).
If (A) = +, then (B A) > and, thus, (B) = +. Similarly, if
(A) = , then (B A) < + and, thus, (B) = .
The proofs of (ii), (iii) and (iv) are the same as the proofs of the correspond-
ing parts of Theorem 2.1.
10.2 The Hahn and Jordan decompositions, I.
Denition 10.3 Let be a signed measure on (X, ).
(i) P is called a positive set for if (A) 0 for every A , A P.
(ii) N is called a negative set for if (A) 0 for every A , A N.
(iii) Q is called a null set for if (A) = 0 for every A , A Q.
It is obvious that an element of which is both a positive and a negative
set for is a null set for . It is also obvious that, if is a measure, then every
A is a positive set for .
Proposition 10.1 Let be a signed measure on (X, ).
(i) If P is a positive set for , P
, P
P, then P
is a positive set for .

(ii) If P
1
, P
2
, . . . are positive sets for , then
+
k=1
P
k
The same results are, also, true for negative sets and for null sets for .
Proof: (i) For every A , A P
we have A P and, hence, (A) 0.

(ii) Take arbitrary A , A
+
k=1
P
k
. We can write A =
+
k=1
A
k
, where
A
1
, A
2
, . . . are pairwise disjoint and A
k
P
k
for every k. Indeed, we may
set A
1
= A P
1
and A
k
= A
_
P
k
(P
1
P
k1
)
_
for all k 2. By the
result of (i), we then have (A) =
+
k=1
(A
k
) 0.
Theorem 10.2 Let be a signed measure on (X, ).
(i) There exist a positive set P and a negative set N for so that P N = X
and P N = .
(ii) (N) (A) (P) for every A .
(iii) If (A) < + for every A , then is bounded from above, while if
< (A) for every A , then is bounded from below.
(iv) If P
is a positive set for and N
is a negative set for with P
= X
and P
= , then PP
= NN
is a null set for .

Proof: (i) We consider the case when (A) < + for every A .
We dene the quantity
= sup(P) [ P is a positive set for .
This set is non-empty since () = 0 is one of its elements. Thus, 0 . We
consider a sequence P
k
of positive sets for so that (P
k
) and form
the set P =
+
k=1
P
k
which, by Proposition 10.1, is a positive set for . This
implies that (P P
k
) 0 for every k and, hence, (P
k
) (P) for every
k. Taking the limit, we nd that
= (P) < +.
This P is a positive set for of maximal -measure and we shall prove that the
set N = X P is a negative set for .
Suppose that N is not a negative set for . Then there is A
0
, A
0
N,
with 0 < (A
0
) < +. The set A
0
is not a positive set or, otherwise, the
set P A
0
would be a positive set with (P A
0
) = (P) + (A
0
) > (P),
contradicting the maximality of P. Hence, there is at least one subset of A
0
in
having negative -measure. This means that
0
= inf(B) [ B , B A
0
< 0.
If
0
< 1, there is B
1
, B
1
A
0
with (B
1
) < 1. If 1
0
< 0,
there is a B
1
, B
1
A
0
with (B
1
) <
0
2
. We set A
1
= A
0
B
1
and have
(A
0
) = (A
1
) + (B
1
) < (A
1
) < +. Observe that we are using Theorem
10.1 to imply (A
1
), (B
1
) R from (A
0
) R.
Suppose that we have constructed sets A
0
, A
1
, . . . , A
n
and B
1
, . . . , B
n

so that
A
n
A
n1
A
1
A
0
N, B
n
= A
n1
A
n
, . . . , B
1
= A
0
A
1
,
so that,

k1
= inf (B) [ B , B A
k1
< 0,
(B
k
) <
_
1, if
k1
< 1
k1
2
, if 1
k1
< 0
for all k = 1, . . . , n and so that
0 < (A
0
) < (A
1
) < < (A
n1
) < (A
n
) < +.
Now, A
n
is not a positive set for for the same reason that A
0
is not a
positive set for . Hence, there is at least one subset of A
n
in having negative
-measure. This means that
n
= inf(B) [ B , B A
n
< 0.
If
n
< 1, there is B
n+1
, B
n+1
A
n
with (B
n+1
) < 1. If 1
n
< 0,
there is a B
n+1
, B
n+1
A
n
with (B
n+1
) <
n
2
. We set A
n+1
= A
n
B
n+1
and have (A
n
) = (A
n+1
) + (B
n+1
) < (A
n+1
) < +. This means that
we have, inductively, constructed two sequences A
n
, B
n
satisfying all the
properties .
Now, the sets B
1
, B
2
, . . . and
+
n=1
A
n
are pairwise disjoint and we have
A
0
= (
+
n=1
A
n
) (
+
n=1
B
n
). Therefore, (A
0
) = (
+
n=1
A
n
) +
+
n=1
(B
n
),
from which we nd
+
n=1
(B
n
) > .
This implies that (B
n
) 0 as n + and, by the third property ,
n1
0
as n +. Now the set A =
+
n=1
A
n
, by continuity from above of , has
(A) = lim
n+
(A
n
) > 0.
Moreover, A is not a positive set for for the same reason that A
0
is not a
positive set for . Hence, there is some B , B A with (B) < 0. But then
B A
n1
for all n and, hence,
n1
(B) < 0 for all n. We, thus, arrive at
a contradiction with the limit
n1
0.
In the same way, we can prove that, if < (A) for every A , then
there is a negative set N for of minimal -measure so that the set P = X N
Thus, in any case we have a positive set P and a negative set N for so
that P N = X and P N = .
(ii) If A , then (P A) 0, because P A P. This implies (P) =
(P A) + (P A) (P A) and, similarly, (N) (N A). Therefore,
(A) = (PA)+(NA) (PA) (P) and (A) = (PA)+(NA)
(N A) (N).
(iii) This a consequence of the result of (ii).
(iv) Now, let P
be a positive set and N
be a negative set for with P
= X
and P
= . Then, since P P
= N
N P N
, the set P P
= N
N
is both a positive set and a negative set for and, hence, a null set for .
Similarly, P
P = N N
is a null set for and we conclude that their union

PP
= NN
is a null set for .

Denition 10.4 Let be a signed measure on (X, ). Every partition of X
into a positive and a negative set for is called a Hahn decomposition of X
for .
It is clear from Theorem 10.2 that if P, N is a Hahn decomposition of X for
, then
(P) = max(A) [ A , (N) = min(A) [ A .
Denition 10.5 Let
1
,
2
be two signed measures on (X, ). We say that they
are mutually singular (or that
1
is singular to
2
or that
2
is singular to
1
) if there exist A
1
which is null for
2
and A
2
which is null for
1
so that A
1
A
2
= X and A
1
A
2
= .
We use the symbol
1
2
to denote that
1
,
2
are mutually singular.
In other words, two signed measures are mutually singular if there is a set
in which is null for one of them and its complement is null for the other.
If
1
,
2
are mutually singular and A
1
, A
2
are as in the Denition 10.5, then
it is clear that
1
(A) =
1
(A A
1
),
2
(A) =
2
(A A
2
)
for every A . Thus, in a free language, we may say that
1
is concentrated
on A
1
and
2
is concentrated on A
2
.
Proposition 10.2 Let ,
1
,
2
be signed measures on (X, ). If
1
,
2
and
1
+
2
is dened, then
1
+
2
.
Proof: Take A
1
, B
1
, A
2
, B
2
so that A
1
B
1
= X = A
2
B
2
, A
1
B
1
=
= A
2
B
2
, A
1
is null for
1
, A
2
is null for
2
and B
1
, B
2
are both null for .
Then B
1
B
2
is null for and A
1
A
2
is null for both
1
and
2
and, hence,
for
1
+
2
. Since (A
1
A
2
) (B
1
B
2
) = X and (A
1
A
2
) (B
1
B
2
) = ,
we have that
1
+
2
.
Theorem 10.3 Let be a signed measure on (X, ). There exist two non-
negative signed measures (i.e. measures)
+
and
, at least one of which is

nite, so that
=
+
,
+
.
If
1
,
2
are two measures on (X, ), at least one of which is nite, so that
=
1
2
and
1
2
, then
1
=
+
and
2
=
.
Proof: We consider any Hahn decomposition of X for : P is a positive set and
N a negative set for so that P N = X and P N = .
We dene
+
,
: [0, +] by
+
(A) = (A P),
(A) = (A N)
for every A . It is trivial to see that
+
,
are non-negative signed measures

on (X, ). If (A) < + for every A , then
+
(X) = (P) < + and,
hence,
+
is a nite measure. Similarly, if < (A) for every A , then
(X) = (N) < + and, hence,
is a nite measure.
Also, (A) = (AP) +(AN) =
+
(A)
(A) for all A and, thus,

=
+
.
If A and A N, then
+
(A) = (AP) = () = 0. Therefore, N is a
null set for
+
. Similarly, P is a null set for
and, hence,
+
.
Now, let
1
,
2
be two measures on (X, ), at least one of which is nite,
so that =
1

2
and
1
2
. Consider A
1
, A
2
, with A
1
A
2
= X and
A
1
A
2
= , so that A
2
is a null set for
1
and A
1
is a null set for
2
.
If A , A A
2
, then (A) =
1
(A)
2
(A) =
2
(A) 0 and, if
A A
1
, then (A) =
1
(A)
2
(A) =
1
(A) 0. Hence, A
1
, A
2
is a Hahn
decomposition of X for . Theorem 10.2 implies that A
1
P = A
2
N is a null
set for . Therefore, for every A , we have
1
(A) =
1
(AA
1
)+
1
(AA
2
) =
1
(A A
1
) =
1
(A A
1
)
2
(A A
1
) = (A A
1
) = (A A
1
P) +(A
A
1
N) = (A A
1
P), since A A
1
N A
1
P. On the other hand,
+
(A) = (A P) = (A A
1
P) + (A A
2
P) = (A A
1
P), since
A A
2
P A
2
N. From the two equalities we get
1
(A) =
+
(A) for every
A and, thus,
1
=
+
. We, similarly, prove
2
=
.
Denition 10.6 Let be a signed measure on (X, ). We say that the pair of
mutually singular measures
+
,
, whose existence and uniqueness is proved in

Theorem 10.3, constitute the Jordan decomposition of .
+
is called the positive variation of and
is called the negative

variation of .
The measure [[ =
+
+
is called the absolute variation of , while

the quantity [[(X) is called the total variation of .
Observe that the total variation of is equal to
[[(X) =
+
(X) +
(X) = (P) (N),

where the sets P, N constitute a Hahn decomposition of X for . Hence, the
total variation of is equal to the dierence between the largest and the smallest
values of .
Moreover, the total variation is nite if and only if the absolute variation is
a nite measure if and only if both the positive and the negative variations are
nite measures if and only if the signed measure takes only nite values.
Proposition 10.3 Suppose (X, , ) is a measure space and let f : X R be
measurable and
_
X
f d be dened. Then the sets P = x X[ f(x) 0
and N = x X[ f(x) < 0 constitute a Hahn decomposition of X for the
signed measure f. Also,
(f)
+
= f
+
, (f)
= f
constitute the Jordan decomposition of f and

[f[ = [f[.
Proof: If A and A P, then (f)(A) =
_
A
f d 0, while, if A N,
then (f)(A) =
_
A
f d 0. Therefore, P is a positive set for f and N is a
negative set for f. Since P N = X and P N = , we conclude that P, N
constitute a Hahn decomposition of X for f.
Now, (f)
+
(A) = (f)(A P) =
_
AP
f d =
_
A
f
P
d =
_
A
f
+
d =
(f
+
)(A) and, similarly, (f)
(A) = (f)(AN) =
_
AN
f d =
_
A
f
N
d =
_
A
f
d = (f
)(A) for every A .

Therefore, (f)
+
= f
+
and (f)
= f
.
Now, [f[ = (f)
+
+ (f)
= f
+
+f
= [f[.
It is easy to see that another Hahn decomposition of X for f consists of
the sets P
= x X[ f(x) > 0 and N
= x X[ f(x) 0.
Proposition 10.4 Suppose (X, , ) is a measure space and f : X R is
measurable so that
_
X
f d is dened. Let E .
(i) E is a positive set for f if and only if f 0 -a.e. on E.
(ii) E is a negative set for f if and only if f 0 -a.e. on E.
(iii) E is a null set for f if and only if f = 0 -a.e. on E.
Proof: (i) Let f 0 -a.e. on E and take any A , A E. Then f 0
-a.e. on A and, hence, (f)(A) =
_
A
f d 0. Thus, E is a positive set
for f. Suppose, conversely, that E is a positive set for f. If n N and
A
n
= x E [ f(x)
1
n
, then 0 (f)(A
n
) =
_
An
f d
1
n
(A
n
). This
implies that (A
n
) = 0 and, since x E [ f(x) < 0 =
+
n=1
A
n
, we conclude
that (x E[ f(x) < 0) = 0. This means that f 0 -a.e. on E.
The proof of (ii) is identical to the proof of (i), and (iii) is a consequence of
the results of (i) and (ii).
We recall that, for every a R, the positive part of a and the negative part
of a are dened as
a
+
= max(a, 0), a
= min(a, 0)
and, hence,
a = a
+
a
, [a[ = a
+
+a
.
It is trivial to prove that
(a +b)
+
a
+
+b
+
, (a +b)
+b
for every a, b R for which a +b is dened.

Denition 10.7 Let (X, ) be a measurable space and A . If A
1
, . . . , A
n

are pairwise disjoint and A =
n
k=1
A
k
, then A
1
, . . . , A
n
is called a (nite)
measurable partition of A.
Theorem 10.4 Let be a signed measure on (X, ) and let [[,
+
and
be
the absolute, the positive and the negative variation of , respectively. Then, for
every A ,
[[(A) = sup
_
n
k=1
[(A
k
)[ [ n N, A
1
, . . . , A
n
measurable partition of A
_
,
+
(A) = sup
_
n
k=1
(A
k
)
+
[ n N, A
1
, . . . , A
n
_
,
(A) = sup
_
n
k=1
(A
k
)
[ n N, A
1
, . . . , A
n
_
.
Proof: We let P, N be a Hahn decomposition of X for . For every pairwise
disjoint A
1
, . . . , A
n
with
n
k=1
A
k
= A we have that
n
k=1
[(A
k
)[ =
n
k=1
[
+
(A
k
)
(A
k
)[
n
k=1
+
(A
k
) +
n
k=1
(A
k
)
=
+
(A) +
(A) = [[(A).
Therefore, the supremum of the left side is [[(A). On the other hand,
AP, AN is a particular measurable partition of A for which [(AP)[ +
[(A N)[ = (A P) (A N) =
+
(A) +
(A) = [[(A) and, hence, the

supremum is equal to [[(A).
The proofs of the other two equalities are identical.
10.3. THE HAHN AND JORDAN DECOMPOSITIONS, II. 195
Lemma 10.1 Let be a signed measure on (X, ) and A . Then, A is a
null set for if and only if it is a null set for both
+
,
if and only if it is a
null set for [[.
Proof Since [[ =
+
+
, the second equivalence is trivial.

Let A be null for [[. For every B , B A, we have that [(B)[ =
[
+
(B)
(B)[
+
(B) +
(B) = [[(B) = 0. Hence, (B) = 0 and A is

null for .
Let A be null for . If A
1
, . . . , A
n
is any measurable partition of A, then
(A
k
) = 0 for all k and, hence,
n
k=1
[(A
k
)[ = 0. Taking the supremum of the
left side, Theorem 10.4 implies that [[(A) = 0 and, thus, A is null for [[.
Proposition 10.5 Let
1
and
2
be two signed measures on (X, ). Then
1
and
2
are mutually singular if and only if each of
+
1
,
1
and each of
+
2
,
2
are mutually singular if and only if [
1
[ and [
2
[ are mutually singular.
Proof: The proof is a trivial consequence of Lemma 10.1.
1
,
2
be signed measures on (X, ) and R. If
1
+
2
is dened, we have
[
1
+
2
[ [
1
[ +[
2
[, [[ = [[[[.
Proof: We take an arbitrary measurable partition A
1
, . . . , A
n
of A and we
have
n
k=1
[(
1
+
2
)(A
k
)[
n
k=1
[
1
(A
k
)[ +
n
k=1
[
2
(A
k
)[ [
1
[(A)+[
2
[(A).
Taking the supremum of the left side, we nd [
1
+
2
[(A) [
1
[(A) +[
2
[(A).
In the same manner,
n
k=1
[()(A
k
)[ = [[
n
k=1
[(A
k
)[. This equality
implies
n
k=1
[()(A
k
)[ [[[[(A) and, taking supremum of the left side,
[[(A) [[[[(A). The same equality, also, implies [[(A) [[
n
k=1
[(A
k
)[
and, taking supremum of the right side, [[(A) [[[[(A).
10.3 The Hahn and Jordan decompositions, II.
In this section we shall describe another method of constructing the Hahn and
Jordan decompositions of a signed measure. In the previous section we derived
the Hahn decomposition rst and, based on it, we derived the Jordan decom-
position. We shall, now, follow the reverse procedure.
Denition 10.8 Let be a signed measure on (X, ). For every A we
dene
[[(A) = sup
_
n
k=1
[(A
k
)[ [ n N, A
1
, . . . , A
n
_
,
+
(A) = sup
_
n
k=1
(A
k
)
+
[ n N, A
1
, . . . , A
n
_
,
(A) = sup
_
n
k=1
(A
k
)
[ n N, A
1
, . . . , A
n
_
.
Lemma 10.2 Let be a signed measure on (X, ). Then,
+
(A) +
(A) = [[(A)
and
+
(A) = sup(B) [ B , B A,
(A) = inf(B) [ B , B A
for every A .
Proof: (a) Take any A and any measurable partition A
1
, . . . , A
n
of A.
Then,
n
k=1
[(A
k
)[ =
n
k=1
(A
k
)
+
+
n
k=1
(A
k
)

+
(A) +
(A).
Taking the supremum of the left side, we get [[(A)
+
(A) +
(A).
Now take arbitrary partitions A
1
, . . . , A
n
and A
1
, . . . , A
n
of A. Then
n
k=1
(A
k
)
+
k=1
n
=1
(A
k
A
k
)
+
,
n
=1
(A
k
)
=1
n
k=1
(A
k
A
k
)
and, adding,
n
k=1
(A
k
)
+
+
n
=1
(A
k
)
1kn,1k
[(A
k
A
k
)[.
Since A
k
A
k
[ 1 k n, 1 k
is a measurable partition of A, we get
n
k=1
(A
k
)
+
+
=1
(A
k
)
[[(A). Finally, taking the supremum of the

left side, we nd
+
(A) +
(A) [[(A).
(b) If B and B A, then B, A B is a measurable partition of A
and, hence, (B) (B)
+
(B)
+
+ (A B)
+

+
(A). This proves that
sup(B) [ B , B A
+
(A).
Let A
1
, . . . , A
n
be any measurable partition of A. If A
i1
, . . . , A
im
are ex-
actly the sets with non-negative -measure and if B
0
=
m
l=1
A
i
l
A, then
n
k=1
(A
k
)
+
=
m
l=1
(A
i
l
) = (B
0
). This implies that
n
k=1
(A
k
)
+
sup(B) [ B , B A and, hence,

+
(A) sup(B) [ B , B A.
We conclude that
+
(A) = sup(B) [ B , B A and a similar argu-
ment proves the last equality.
Theorem 10.5 Let be a signed measure on (X, ). Then, the functions
[[,
+
,
: [0, +], which were dened in Denition 10.8, are measures

on (X, ).
At least one of
+
,
is nite and
= ,
+
+
= [[,
+
.
10.3. THE HAHN AND JORDAN DECOMPOSITIONS, II. 197
Proof: (a) We shall rst prove that [[ is a measure.
It is obvious that [[() = 0 and take arbitrary pairwise disjoint A
1
, A
2
, . . .
and A =
+
j=1
A
j
.
If A
1
, . . . , A
n
is an arbitrary measurable partition of A, then, for ev-
ery j, A
1
A
j
, . . . , A
n
A
j
is a measurable partition of A
j
. This im-
plies,
n
k=1
[(A
k
)[ =
n
k=1
[
+
j=1
(A
k
A
j
)[
n
k=1
+
j=1
[(A
k
A
j
)[ =
+
j=1
n
k=1
[(A
k
A
j
)[
+
j=1
[[(A
j
) and, taking the supremum of the left
side, [[(A)
+
j=1
[[(A
j
).
Fix arbitrary N N and for every j = 1, . . . , N take any measurable par-
tition A
j
1
, . . . , A
j
nj
of A
j
. Then A
1
1
, . . . , A
1
n1
, . . . , A
N
1
, . . . , A
N
nN
,
+
j=N+1
A
j
is a measurable partition of A and, hence, [[(A)
N
j=1
nj
k=1
[(A
j
k
)[ +
[(
+
j=N+1
A
j
)[
N
j=1
nj
k=1
[(A
j
k
)[. Taking the supremum of the right side,
we get [[(A)
N
j=1
[[(A
j
) and, taking the limit as N +, we nd
[[(A)
+
j=1
[[(A
j
).
Hence, [[(A) =
+
j=1
[[(A
j
).
The proofs that
+
and
are measures are identical to the proof we have

just seen.
(b) In case (A) < + for every A , we shall prove that
+
(X) < +.
We claim that for every A with
+
(A) = + and every M > 0, there
exists B , B A, so that
+
(B) = + and (B) M.
Suppose that the claim is not true. Then, there is A with
+
(A) = +
and an M > 0 so that, if B , B A, has (B) M, then
+
(B) <
+. Now, by Lemma 10.2, there is B
1
, B
1
A with (B
1
) M and,
hence,
+
(B
1
) < +. Suppose that we have constructed pairwise disjoint
B
1
, . . . , B
m
subsets of A with (B
j
) M and
+
(B
j
) < + for every
j = 1, . . . , m. Since
+
is a measure, we have
m
j=1
+
(B
j
) +
+
(A
m
j=1
B
j
) =
+
(A) = + and, thus,
+
(A
m
j=1
B
j
) = +. Lemma 10.2 implies that
there is B
m+1
, B
m+1
A
m
j=1
B
j
with (B
m+1
) M and, hence,
+
(B
m+1
) < +.
We, thus, inductively construct a sequence B
m
in of pairwise disjoint
subsets of A with (B
m
) M. But, then, (
+
m=1
B
m
) =
+
m=1
(B
m
) = +
and we arrive at a contradiction.
Using the claimed result and assuming that
+
(X) = +, we nd B
1

with (B
1
) 1 and
+
(B
1
) = +. We, similarly, nd B
2
, B
2
B
1
, with
(B
2
) 2 and
+
(B
2
) = +. Continuing inductively, a decreasing sequence
B
m
is constructed in with (B
m
) m for every m. Then, (
+
l=1
B
l
) =
lim
m+
(B
m
) = + and we arrive at a contradiction.
Therefore,
+
(X) < +.
If < (A) for every A , we prove in the same way that
(X) < +.
(c) Suppose that (A) < + for every A and, hence,
+
(X) < +, by
the result of (b).
We take any A and any B , B A. Then (A B)
+
(A)
and, hence, (A)
+
(A) + (B). Taking the inmum over B and using the
+
(A) < +, we get (A)
+
(A)
(A).
To prove the opposite inequality, we rst assume
(A) < +. For every

B , B A, we have
(A) (A B) and, hence, (B)
(A) (A).
Taking the supremum over B we nd
+
(A)
(A) (A). If
(A) = +,
then, since
+
(A) < +, the
+
(A)
(A) (A) is clearly true.

We conclude that (A) =
+
(A)
(A) for every A and the same can

be proved if we assume that < (A) for every A .
Therefore, =
+
.
(d) The equality [[ =
+
+
is contained in Lemma 10.2.

(e) We, again, assume (A) < + for every A and, hence,
+
(X) < +.
Using Lemma 10.2, we take a sequence B
n
in so that (B
n
)
+
(X)
as n +. Since (B
n
)
+
(B
n
)
+
(X), we have that
+
(B
n
)
+
(X)
as n +. From (B
n
) =
+
(B
n
)
(B
n
), we get
(B
n
) 0 as n +.
We nd a strictly increasing n
k
so that
(B
n
k
) <
1
2
k
for all k. If we
set F
k
=
+
l=k
B
n
l
, then
(F
k
)
+
l=k

(B
n
l
) <
1
2
k1
for every k and F
k
is decreasing. Therefore, the set F =

+
k=1
F
k
has
(F) = 0. We, also, have

that
+
(B
n
k
)
+
(F
k
)
+
(X) and, hence,
+
(F
k
)
+
(X) as k +.
Therefore,
+
(F) =
+
(X).
We have constructed a set F so that
(F) = 0 and
+
(F) =
+
(X).
Since
+
(X) < +, we nd
+
(X F) = 0 and we conclude that
+
.
The decomposition =
+
of the signed measure on (X, ), which is

given in Theorem 10.5, is the same as the Jordan decomposition of , which was
dened in the previous section 10.2. This is justied both by the uniqueness
of the Jordan decomposition of a signed measure and by the result of Theo-
rem 10.4. Using, now, the Jordan decomposition, we shall produce the Hahn
decomposition of a signed measure.
Theorem 10.6 Let be a signed measure on (X, ) and
+
,
be the measures
of Denition 10.8. Then, there exist P, N so that P N = X, P N = ,
P is a positive set for , N is a negative set for and
+
(N) = 0,
(P) = 0.
Proof: Theorem 10.5 implies that
+
and, hence, there are P, N so

that P N = X, P N = and
+
(N) = 0 =
(P).
If A , A P, then (A) =
+
(A)
(A) =
+
(A) 0. Similarly, if
A , A N, then (A) =
+
(A)
(A) =
(A) 0. Hence, P is a
positive set for and N is a negative set for .
10.4 Complex measures.
Denition 10.9 Let (X, ) be a measurable space and a function : C
such that
(i) () = 0,
(ii) (
+
j=1
A
j
) =
+
j=1
(A
j
) for every pairwise disjoint A
1
, A
2
, . . . .
It is trivial to prove, taking real and imaginary parts, that the functions
(), () : R, which are dened by ()(A) = ((A)) and ()(A) =
10.4. COMPLEX MEASURES. 199
((A)) for every A , are real measures on (X, ) and, hence, they are
bounded. That is, there is an M < + so that [()(A)[, [()(A)[ M for
every A . This implies that [(A)[ 2M for every A and we have
proved the
Proposition 10.7 Let be a complex measure on (X, ). Then is bounded,
i.e. there is an M < + so that [(A)[ M for every A .
If
1
and
2
are complex measures on (X, ) and
1
,
2
C, then
1
1
+
2
2
,
dened by (
1
1
+
2
2
)(A) =
1
1
(A) +
2
2
(A) for all A , is a complex
measure on (X, ).
The following are straightforward extensions of Denitions 10.3 and 10.5.
Denition 10.10 Let be a complex measure on (X, ) and A . We say
that A is a null set for if (B) = 0 for every B , B A.
Denition 10.11 Let
1
and
2
be complex or signed measures on (X, ). We
say that
1
and
2
are mutually singular, and denote this by
1
2
, if there
are A
1
, A
2
so that A
2
is null for
1
, A
1
is null for
2
and A
1
A
2
= X,
A
1
A
2
= .
Proposition 10.8 Let be a complex measure on (X, ). If for every A
we dene
[[(A) = sup
_
n
k=1
[(A
k
)[ [ n N, A
1
, . . . , A
n
_
,
then the function [[ : [0, +] is a nite measure on (X, ).
Proof: The proof that [[ is a measure is exactly the same as in part (a) of the
proof of Theorem 10.5.
We take an arbitrary measurable partition A
1
, . . . , A
n
of X and have
n
k=1
[(A
k
)[
n
k=1
[()(A
k
)[ +
n
k=1
[()(A
k
)[ [()[(X) +[()[(X).
Taking the supremum of the left side, [[(X) [()[(X) + [()[(X) < +,
because the signed measures () and () have nite values.
Denition 10.12 Let be a complex measure on (X, ). The measure [[
dened in Proposition 10.8 is called the absolute variation of and the
number [[(X) is called the total variation of .
1
,
2
be complex measures on (X, ) and C.
Then
(i) [
1
+
2
[ [
1
[ +[
2
[ and [[ = [[[[
(ii) [()[, [()[ [[ [()[ +[()[.
Proof: (i) The proof is identical to the proof of Proposition 10.6.
(ii) In the same manner, if A
1
, . . . , A
n
is any measurable partition of A ,
we have
n
k=1
[()(A
k
)[
n
k=1
[(A
k
)[ [[(A) and also
n
k=1
[()(A
k
)[
n
k=1
[(A
k
)[ [[(A). Taking supremum of the left sides of these two inequal-
ities, we nd [()[(A), [()[(A) [[(A).
The last inequality is a consequence of the result of (i).
Lemma 10.3 Let be a complex measure on (X, ) and A . Then A is
null for if and only if A is null for both () and () if and only if A is null
for [[.
Proof: The rst equivalence is trivial. The proof that A is null for if and only
if A is null for [[ is a repetition of the proof of the same result for a signed
measure . See Lemma 10.1.
1
and
2
be complex or signed measures on (X, ).
Then,
1
2
if and only if each of (
1
), (
1
) and each of (
2
), (
2
) are
mutually singular if and only if [
1
[[
2
[.
Proof: Trivial after Lemma 10.3.
Example
We take a measure on (X, ) and a measurable function f : X C which
is integrable over X with respect to . Then,
_
A
f d is, by Lemma 7.10, a
complex number for every A , and Theorem 7.13 implies that the function
: C, which is dened by
(A) =
_
A
f d
for every A , is a complex measure on (X, ).
Denition 10.13 Let (X, , ) be a measure space and f : X C be integrable
with respect to . The complex measure dened in the previous paragraph is
called the indenite integral of f with respect to and it is denoted by
f. Thus,
(f)(A) =
_
A
f d, A .
The next result is the analogue of Proposition 10.3.
Proposition 10.11 Let (X, , ) be a measure space and f : X C be inte-
grable with respect to . Then
[f[(A) =
_
A
[f[ d
for every A . Hence,
[f[ = [f[.
Proof: If A
1
, . . . , A
n
is an arbitrary measurable partition of A , then
n
k=1
[(f)(A
k
)[ =
n
k=1
[
_
A
k
f d[
n
k=1
_
A
k
[f[ d =
_
A
[f[ d. Therefore,
taking the supremum of the left side, [f[(A)
_
A
[f[ d.
Since f is integrable with respect to , it is nite -a.e. on X. If N =
x X[ f(x) ,= , then (N
c
) = 0 and Theorem 6.1 implies that there is
a sequence
m
of measurable simple functions with
m
sign(f) on N
10.5. INTEGRATION. 201
and [
m
[ [sign(f)[ 1 on N. Dening each
n
as 0 on N
c
, we have that all
these properties hold -a.e. on X.
If
m
=
nm
k=1
m
k

E
m
k
is the standard representation of
m
, then [
m
k
[ 1
for all k = 1, . . . , n
m
and, hence, [
_
A
f
m
d[ = [
nm
k=1
m
k
_
AE
m
k
f d[
nm
k=1
[(f)(A E
m
k
)[ [f[(A), where the last inequality is true because
A E
m
1
, . . . , A E
m
nm
is a measurable partition of A. By the Dominated
Convergence Theorem, we get that
_
A
[f[ d =
_
A
fsign(f) d [f[(A).
We conclude that [f[(A) =
_
A
[f[ d for every A .
10.5 Integration.
The next denition covers only the case when both f and have their values
in R.
Denition 10.14 Let be a signed measure on (X, ). If f : X R is
measurable, we say that the integral
_
X
f d of f over X with respect
to is dened if both
_
X
f d
+
and
_
X
f d
are dened and they are neither

both + nor both . In such a case we write
_
X
f d =
_
X
f d
+
_
X
f d
.
We say that f is integrable over X with respect to if the
_
X
f d is
nite.
Proposition 10.12 Let be a signed measure on (X, ) and f : X R be
measurable. Then f is integrable with respect to if and only if f is integrable
with respect to both
+
and
if and only if f is integrable with respect to [[.

Proof:
_
X
f d is nite if and only if both
_
X
f d
+
and
_
X
f d
are nite
or, equivalently,
_
X
[f[ d
+
< + and
_
X
[f[ d
< + or, equivalently,

_
X
[f[ d[[ < + if and only if f is integrable with respect to [[.
Lemma 10.4 Let
1
,
2
be two measures on (X, ) with
1

2
. Then
_
X
f d
1

_
X
f d
2
for every measurable f : X [0, +].
Proof: If =
m
j=1
Ej
is a measurable non-negative simple function
with its standard representation, then we have
_
X
d
1
=
m
j=1
1
(E
j
)
m
j=1
2
(E
j
) =
_
X
d
2
. For the general f we take a sequence
n
of
measurable non-negative simple functions with
n
f on X. We write
the inequality for each
n
and the Monotone Convergence Theorem implies
_
X
f d
1

_
X
f d
2
.
Now, suppose that is a signed measure or a complex measure on (X, )
and the function f : X R or C is measurable. If
_
X
[f[ d[[ < +,
then f is nite [[-a.e. on X and the [[-a.e. dened functions (f) and
(f) satisfy
_
X
[(f)[ d[[ < + and
_
X
[(f)[ d[[ < +. Since, by Proposi-
tion 10.9, [()[ [[ and [()[ [[, Lemma 10.4 implies that all integrals
_
X
[(f)[ d[()[,
_
X
[(f)[ d[()[,
_
X
[(f)[ d[()[ and
_
X
[(f)[ d[()[ are
nite. Proposition 10.12 implies that all integrals
_
X
(f) d(),
_
X
(f) d(),
_
X
(f) d() and
_
X
(f) d() are dened and they all are complex numbers.
Therefore, the following denition is valid.
Denition 10.15 Let be a signed measure or a complex measure on (X, )
and f : X R or C be measurable. We say that f is integrable over X
with respect to if f is integrable with respect to [[. In such a case we say
that the integral
_
X
f d of f over X with respect to is dened and it
is given by
_
X
f d =
_
X
(f) d()
_
X
(f) d() +i
_
X
(f) d() +i
_
X
(f) d().
Of course, when f : X C and is signed, we have
_
X
f d =
_
X
(f) d +i
_
X
(f) d,
and when f : X R and is complex, we have
_
X
f d =
_
X
f d() +i
_
X
f d(),
all under the assumption that
_
X
[f[ d[[ < +.
We shall not bother to extend all properties of integrals with respect to
measures to properties of integrals with respect to signed measures or complex
measures. The safe thing to do is to reduce everything to positive and negative
variations or to real and imaginary parts.
For completeness, we shall only see a few most necessary properties, like the
linearity properties and the appropriate version of the Dominated Convergence
Theorem.
1
,
2
be signed or complex measures on (X, ) and
f, f
1
, f
2
: X R or C be all integrable with respect to these measures. For
every
1
,
2
C,
_
X
(
1
f
1
+
2
f
2
) d =
1
_
X
f
1
d +
2
_
X
f
2
d ,
_
X
f d(
1
1
+
2
2
) =
1
_
X
f d
1
+
2
_
X
f d
2
.
Proof: The proof is straightforward when we reduce everything to real functions
and signed measures.
10.5. INTEGRATION. 203
Theorem 10.7 (Dominated Convergence Theorem) Let be a signed or
complex measure on (X, ), f, f
n
: X R or C and g : X [0, +] be
measurable. If f
n
f and [f
n
[ g on X except on a set which is null for
and if
_
X
g d[[ < +, then
_
X
f
n
d
_
X
f d.
Proof: A set which is null for is, also, null for
+
and
, if is signed,
and null for () and (), if is complex. Moreover, by Lemma 10.4,
_
X
g d
+
,
_
X
g d
< +, if is signed, and

_
X
g d[()[,
_
X
g d[()[ < +,
if is complex.
Therefore, the proof reduces to the usual Dominated Convergence Theorem
for measures.
Theorem 10.8 Let be a signed or complex measure on (X, ) and f : X R
or C be such that the
_
X
f d is dened. Then
_
X
f d

_
X
[f[ d[[.
Proof: We may assume that
_
X
[f[ d[[ < +, or else the inequality is obvious.
If is a signed measure, [
_
X
f d[ = [
_
X
f d
+
_
X
f d
[ [
_
X
f d
+
[ +
[
_
X
f d
[
_
X
[f[ d
+
+
_
X
[f[ d
=
_
X
[f[ d[[.
If is complex, we shall see a proof which is valid in all cases anyway.
Let : X C be a measurable simple function with its standard rep-
resentation =
n
k=1
E
k
and so that [[(E
k
) < + for all k. Then, we
have [
_
X
d[ = [
n
k=1
k
(E
k
)[

n
k=1
[
k
[[(E
k
)[

n
k=1
[
k
[[[(E
k
) =
_
X
[[ d[[.
Consider a sequence
n
of measurable simple functions so that
n

f on X and [
n
[ [f[ on X. The Monotone Convergence Theorem implies
_
X
[
n
[ d[[
_
X
[f[ d[[ and Theorem 10.7, together with
_
X
[f[ d[[ < +,
implies that
_
X

n
d
_
X
f d. Taking the limit in [
_
X

n
d[
_
X
[
n
[ d[[
we prove the
_
X
f d

_
X
[f[ d[[.
A companion to the previous theorem is
Theorem 10.9 Let be a signed or complex measure on (X, ). Then
[[(A) = sup
_
_
A
f d
f is measurable, [f[ 1 on A
_
,
for every A , where the functions f have real values, if is signed, and
complex values, if is complex.
Proof: If f is measurable and [f[ 1 on A, then [f
A
[
A
on X and
Theorem 10.8 implies [
_
A
f d[ = [
_
X
f
A
d[
_
X
[f
A
[ d[[
_
X

A
d[[ =
[[(A). Therefore the supremum of the left side is [[(A).
If is signed, we take a Hahn decomposition of X for . There are P, N
so that P N = X, P N = , P is a positive set and N a negative set for .
We consider the function f with values f = 1 on P and f = 1 on N. Then
[
_
A
f d[ = [(A P) (A N)[ = (A P) (A N) =
+
(A) +
(A) =
[[(A). Therefore, the supremum is equal to [[(A).
If is complex, we nd a measurable partition A
1
, . . . , A
n
of A so that
[[(A)

n
k=1
[(A
k
)[. We, then, dene the function f =
n
k=1
A
k
,
where
k
= sign((A
k
)) for all k. Then, [f[ 1 on A and [
_
A
f d[ =
[
n
k=1
k
(A
k
)[ =
n
k=1
[(A
k
)[ [[(A) . This proves that the supremum
is equal to [[(A).
Finally, we prove a result about integration with respect to an indenite
integral. This is important because, as we shall see in the next section, indenite
integrals are special measures which play an important role among signed or
complex measures.
Theorem 10.10 Let be a measure on (X, ) and f : X R or C be
measurable so that
_
X
f d is dened. Consider the signed measure or com-
plex measure f, the indenite integral of f with respect to .
A measurable function g : X R or C is integrable over X with respect
to f if and only if gf is integrable over X with respect to . In such a case,
_
X
g d(f) =
_
X
gf d.
This equality is true, without any restriction, if f, g : X [0, +] are
measurable.
Proof: We consider rst the case when g, f : X [0, +].
If g =
A
for some A , then
_
X

A
d(f) = (f)(A) =
_
A
f d =
_
X

A
f d. Hence, the equality
_
X
g d(f) =
_
X
gf d is true for characteristic
functions. This extends, by linearity, to measurable non-negative simple
functions g = and, by the Monotone Convergence Theorem, to the general g.
This implies that, in general,
_
X
[g[ d([f[) =
_
X
[gf[ d. From this we see
that g is integrable over X with respect to f if and only if, by denition, g is
integrable over X with respect to [f[ = [f[ if and only if, by the equality we
just proved, gf is integrable over X with respect to .
The equality
_
X
g d(f) =
_
X
gf d can, now, be established by reducing all
functions to non-negative functions and using the special case we proved.
10.6 Lebesgue decomposition, Radon-Nikodym
derivative.
Denition 10.16 Let be a measure and a signed or complex measure on
(X, ). We say that is absolutely continuous with respect to when
(A) = 0 for every A with (A) = 0 and we denote by
.
10.6. LEBESGUE DECOMPOSITION, RADON-NIKODYMDERIVATIVE.205
Example
Let f : X R or C be measurable so that the
_
X
f d is dened (recall
that, in the case of C, this means that f is integrable). Then the indenite
integral f is absolutely continuous with respect to .
This is obvious: if A has (A) = 0, then (f)(A) =
_
A
f d = 0.
Proposition 10.14 Let be a measure and ,
1
,
2
signed or complex mea-
sures on (X, ).
(i) If is complex, then if and only if () and () if and
only if [[ .
(ii) If is signed, then if and only if
+
and
if and only if
[[ .
(iii) If and , then = 0.
(iv) If
1
,
2
and
1
+
2
is dened, then
1
+
2
.
Proof: (i) Since (A) = 0 is equivalent to ()(A) = ()(A) = 0, the rst
equivalence is obvious.
Let and take any A with (A) = 0. If A
1
, . . . , A
n
is any
measurable partition of A, then (A
k
) = 0 for all k and, thus,
n
k=1
[(A
k
)[ = 0.
Taking the supremum of the left side we get [[(A) = 0. Hence, [[ .
If [[ and we take any A with (A) = 0, then [(A)[ [[(A) = 0.
Therefore, (A) = 0 and .
(ii) The argument of part (i) applies without change to prove that if and
only if [[ . Since [[ =
+
+
, it is obvious that
+
and
if
and only if [[ .
(iii) Take sets M, N so that M N = X, M N = , M is a null set for
and N is a null set for . Then, (N) = 0 and imply that N is a null
set for . But, then, X = M N is a null set for and, hence, = 0.
(iv) If A has (A) = 0, then
1
(A) =
2
(A) = 0 and, hence, (
1
+
2
)(A) = 0.
The next result justies the term absolutely continuous at least in the special
case of a nite .
Proposition 10.15 Let be a measure and a real signed measure or a com-
plex measure on (X, ). Then if and only if for every > 0 there is a
> 0 so that [(A)[ < for every A with (A) < .
Proof: Suppose that for every > 0 there is a > 0 so that [(A)[ < for every
A with (A) < . If (A) = 0, then (A) < for every > 0 and, hence,
[(A)[ < for every > 0. Therefore, (A) = 0 and .
Suppose that but there is some
0
> 0 so that, for every > 0, there
is A with (A) < and [(A)[
0
. Then, for every k, there is A
k
with
(A
k
) <
1
2
k
and [[(A
k
) [(A
k
)[
0
. We dene B
k
=
+
l=k
A
l
and, then,
(B
k
) <
1
2
k1
and [[(B
k
) [[(A
k
)
0
for every k. If we set B =
+
k=1
B
k
,
then B
k
B and, by the continuity of [[ from above, we get (B) = 0 and
[[(B)
0
. This says that [[ is not absolutely continuous with respect to
and, by Proposition 10.14, we arrive at a contradiction.
Theorem 10.11 Let be a measure on (X, ).
(i) If ,
, ,
are signed or complex measures on (X, ) so that ,
and
,
and + =
, then =
and =
.
(ii) If f, f
: X R or C are integrable over X with respect to and f = f
,
then f = f
-a.e. on X.
(iii) If f, f
: X R are measurable and

_
X
f d,
_
X
f
d are dened and

f = f
, then f = f
-a.e. on X, provided that , restricted to the set

x X[ f(x) ,= f
(x), is seminite.
Proof: (i) There exist sets M, M
, N, N
with M N = X = M
,
M N = = M
so that N, N
are null for , M is null for and M
is
null for
. If we set K = N N
, then K is null for and K

c
= MM
is null
for both and
. Since ,
, we have that K is null for both and
.
If A , A K, then (A) = (A) + (A) =
(A) +
(A) =
(A). If
A , A K
c
, then (A) = 0 =
(A). Therefore, for every A we have

(A) = (A K) + (A K
c
) =
(A K) +
(A K
c
) =
(A) and, hence,

=
.
A symmetric argument implies that =
.
(ii) We have
_
A
(f f
) d =
_
A
f d
_
A
f
d = (f)(A) (f
)(A) = 0 for
all A . Theorem 7.5 implies f = f
-a.e. on X.
(iii) Let t, s R with t < s and A
t,s
= x X[ f(x) t, s f
(x). If
0 < (A
t,s
) < +, we dene B = A
t,s
. If (A
t,s
) = +, we take B so
that B A
t,s
and 0 < (B) < +. In any case, (f)(B) =
_
B
f d t(B)
and (f
)(B) =
_
B
f
d s(B) and, thus, s(B) t(B). This implies

(B) = 0, which is false. The only remaining case is (A
t,s
) = 0.
Now, we observe that x X[ f(x) < f
(x) =
t,sQ,t<s
A
t,s
, which implies
(x X[ f(x) < f
(x)) = 0. Similarly, (x X[ f(x) > f
(x)) = 0 and
we conclude that f = f
-a.e. on X.
Lemma 10.5 Let , be nite measures on (X, ). If and are not mutually
singular, then there exists an
0
> 0 and an A
0
with (A
0
) > 0 so that
(A)
(A)

0
for every A , A A
0
with (A) > 0.
Proof: We consider, for every n, a Hahn decomposition of the signed measure

1
n
. There are sets P
n
, N
n
so that P
n
N
n
= X, P
n
N
n
= and P
n
is a positive set and N
n
is a negative set for
1
n
.
We set N =
+
n=1
N
n
and, since N N
n
for all n, we get (
1
n
)(N) 0
for all n. Then (N)
1
n
(N) for all n and, since (N) < +, (N) = 0. We
set P =
+
n=1
P
n
and have P N = X and P N = . If (P) = 0, then and
are mutually singular. Therefore, (P) > 0 and this implies that (P
N
) > 0
for at least one N. We dene A
0
= P
N
for such an N and we set
0
=
1
N
for
the same N.
Now, (A
0
) > 0 and, if A , A A
0
, then, since A
0
is a positive set for

0
, we get (A)
0
(A) 0. If also (A) > 0, then
(A)
(A)

0
.
Theorem 10.12 (Lebesgue-Radon-Nikodym Theorem. Signed case.)
Let be a -nite signed measure and be a -nite measure on (X, ). Then
there exist unique -nite signed measures and on (X, ) so that
= +, , .
Moreover, there exists a measurable f : X R so that the
_
X
f d is dened
and
= f.
If f
is another such function, then f
= f -a.e. on X.
If is non-negative, then and are non-negative and f 0 -a.e. on X.
If is real, then and are real and f is integrable over X with respect to .
Proof: The uniqueness part of the statement is a consequence of Theorem 10.11.
Observe that is -nite and, hence, seminite.
Therefore, we need to prove the existence of , and f.
A. We rst consider the special case when both , are nite measures on
(X, ).
We dene ( to be the collection of all measurable f : X [0, +] with
the property
_
A
f d (A), A .
The function 0, obviously, belongs to ( and, if f
1
, f
2
belong to (, then the
function f = max(f
1
, f
2
) also belongs to (. Indeed, if A we consider
A
1
= x A[ f
2
(x) f
1
(x) and A
2
= x A[ f
1
(x) < f
2
(x) and we have
_
A
f d =
_
A1
f d +
_
A2
f d =
_
A1
f
1
d +
_
A2
f
2
d (A
1
) +(A
2
) = (A).
We dene
= sup
_
_
X
f d[ f (
_
.
Since 0 ( and
_
X
f d (X) for all f (, we have 0 (X) < +.
We take a sequence f
n
in ( so that
_
X
f
n
d and dene g
1
= f
1
and,
inductively, g
n
= max(g
n1
, f
n
) for all n. Then all g
n
belong to (. If we set
f = lim
n+
g
n
, then g
n
f and, by the Monotone Convergence Theorem,
_
A
f d (A), A
and
_
X
f d = < +.
Since ( f)(A) = (A)
_
A
f d 0 for all A , the signed measure
f is a nite measure. If f and are not mutually singular, then, by
Lemma 10.5, there is A
0
and
0
> 0 so that
(A)
(A)

1
(A)
_
A
f d =
( f)(A)
(A)

0
for all A , A A
0
with (A) > 0. From this we get
_
A
(f +
0
A0
) d (A)
for all A , A A
0
. Now for any A we have
_
A
(f +
0
A0
) d =
_
AA0
(f +
0
A0
) d+
_
A\A0
(f +
0
A0
) d (AA
0
)+
_
A\A0
(f +
0
A0
) d =
(A A
0
) +
_
A\A0
f d (A A
0
) + (A A
0
) = (A). This implies that
f +
0
A0
belongs to ( and hence +
0
(A
0
) =
_
X
(f +
0
A0
) d . This is
false and we arrived at a contradiction. Therefore, f.
We set = f and = f and we have the decomposition = +
with , . Both and are nite measures and f : X [0, +] is
integrable with respect to , because (X) =
_
X
f d = < + and (X) =
(X)
_
X
f d = (X) < +.
B. We, now, suppose that both , are -nite measures on (X, ).
Then, there are pairwise disjoint F
1
, F
2
, . . . so that X =
+
k=1
F
k
and
(F
k
) < + for all k and pairwise disjoint G
1
, G
2
, . . . so that X =
+
l=1
G
l
and (G
l
) < + for all l. The sets F
k
G
l
are pairwise disjoint, they cover X
and (F
k
G
l
), (F
k
G
l
) < + for all k, l. We enumerate them as E
1
, E
2
, . . .
and have X =
+
n=1
E
n
and (E
n
), (E
n
) < + for all n.
We dene
n
and
n
by
n
(A) = (A E
n
),
n
(A) = (A E
n
)
for all A and all n and we see that all
n
,
n
are nite measures on (X, ).
We also have
(A) =
+
n=1
n
(A), (A) =
+
n=1
n
(A)
for all A .
Applying the results of part A, we see that there exist nite measures
n
,
n
on (X, ) and f
n
: X [0, +] integrable with respect to
n
so that
n
=
n
+
n
,
n

n
,
n
n
,
n
(A) =
_
A
f
n
d
n
for all n and all A . From
n
(E
c
n
) = 0 we get that
n
(E
c
n
) =
n
(E
c
n
) = 0.
Now, since
n
(A) =
n
(A) = 0 for every A , A E
c
n
, the relation
n
(A) =
_
A
f
n
d
n
remains true for all A if we change f
n
and make it 0 on E
c
n
. We,
therefore, assume that
f
n
= 0 on E
c
n
,
n
(A) =
_
AEn
f
n
d
n
for all n and all A .
We dene , : [0, +] and f : X [0, +] by
(A) =
+
n=1
n
(A), (A) =
+
n=1
n
(A), f(x) =
+
n=1
f
n
(x)
for every A and every x X. It is trivial to see that and are measures
on (X, ) and that f is measurable.
The equality = + is obvious.
If A has (A) = 0, then
n
(A) = (A E
n
) = 0 and, hence,
n
(A) = 0
for all n. Thus, (A) = 0 and, thus, .
Since
n
n
, there is R
n
so that R
n
is null for
n
and R
c
n
is null for
n
. But, then R
n
= R
n
E
n
is also null for
n
and R
c
n
= R
c
n
E
c
n
is null for
n
. Since R
n
is obviously null for all
m
, m ,= n, we have that R
n
is null for .
Then R =
+
n=1
R
n
is null for and R
c
=
+
n=1
R
c
n
is null for all
n
and, hence,
for . We conclude that .
The and are -nite, because (E
n
) =
n
(E
n
) < + and (E
n
) =
n
(E
n
) < + for all n.
Finally, for every A , (A) =
+
n=1
n
(A) =
+
n=1
_
AEn
f
n
d
n
=
+
n=1
_
AEn
f d
n
=
+
n=1
_
AEn
f d =
_
A
f d. The fourth equality is true
because
_
En
f d
n
=
_
En
f d for all measurable f : X [0, +]. This
is justied as follows: if f =
A
, then the equality becomes
n
(A E
n
) =
(AE
n
) which is true. Then the equality holds, by linearity, for non-negative
measurable simple functions and, by the Monotone Convergence Theorem,
it holds for all measurable f : X [0, +]. Now, from (A) =
_
A
f d, we
conclude that = f and that .
C. In the general case we write =
+

and both
+
,
are -nite
measures on (X, ). We apply the result of part B and get -nite measures
1
,
2
,
1
,
2
so that
+
=
1
+
1
,
=
2
+
2
and
1
,
2
,
1
,
2
.
Since either
+
or
is a nite measure, we have that either

1
,
1
are nite
or
2
,
2
are nite. We then write =
1

2
and =
1

2
and have that
= + and , .
We also have measurable f
1
, f
2
: X [0, +] so that
1
= f
1
and
2
= f
2
. Then, either
_
X
f
1
d =
1
(X) < + or
_
X
f
2
d =
2
(X) < +
and, hence, either f
1
< +-a.e. on X or f
2
< +-a.e. on X. The function
f = f
1
f
2
is dened -a.e. on X and the
_
X
f d =
_
X
f
1
d
_
X
f
2
d exists.
Now, (A) =
1
(A)
2
(A) =
_
A
f
1
d
_
A
f
2
d =
_
A
f d for all A and,
thus, = f.
Theorem 10.13 (Lebesgue-Radon-Nikodym Theorem. Complex case.)
Let be a complex measure and be a -nite measure on (X, ). Then there
exist unique complex measures and on (X, ) so that
= +, , .
Moreover, there exists a measurable f : X C so that f is integrable over
X with respect to and
= f.
If f
is another such function, then f
= f -a.e. on X.
If is non-negative, then and are non-negative and f 0 -a.e. on X.
If is real, then and are real and f is extended-real valued.
Proof: The measures () and () are real measures and, by Theorem 10.12,
there exist real measures
1
,
2
,
1
,
2
on (X, ) so that () =
1
+
1
, () =
2
+
2
and
1
,
2
and
1
,
2
. We set =
1
+i
2
and =
1
+i
2
and,
then, = + and, clearly, and . There are, also, f
1
, f
2
: X R,
which are integrable over X with respect to , so that
1
= f
1
and
2
= f
2
.
The function f = f
1
+ if
2
: X C is -a.e. dened, it is integrable over X
with respect to and we have (f)(A) =
_
A
f d =
_
A
f
1
d + i
_
A
f
2
d =
1
(A) +i
2
(A) = (A) for all A . Hence, = f.
The uniqueness is an easy consequence of Theorem 10.11.
Denition 10.17 Let be a signed measure or a complex measure and a
measure on (X, ). If there exist, necessarily unique, signed or complex mea-
sures and , so that
= +, , ,
we say that and constitute the Lebesgue decomposition of with
respect to .
is called the absolutely continuous part and is called the singular
part of with respect to .
Let be a signed or complex measure and a measure on (X, ) so that
. If there exists a measurable f : X R or C so that
_
X
f d is
dened and
= f,
then f is called a Radon-Nikodym derivative of with respect to . Any
Radon-Nikodym derivative of with respect to is denoted by
d
d
.
Theorems 10.12 and 10.13 say that, if and are -nite, then has a
unique Lebesgue decomposition with respect to . Moreover, if and are -
nite and , then there exists a Radon-Nikodym derivative of with respect
to , which is unique if we disregard -null sets. This is true because = +0
is, necessarily, the Lebesgue decomposition of with respect to .
We should make some remarks about Radon-Nikodym derivatives.
1. The symbol
d
d
appears as a fraction of two quantities but it is not. It is like
the well known symbol
dy
dx
of the derivative in elementary calculus.
2. Denition 10.17 allows all Radon-Nikodym derivatives of with respect to
to be denoted by the same symbol
d
d
. This is not absolutely strict and it would
be more correct to say that
d
d
is the collection (or class) of all Radon-Nikodym
derivatives of with respect to . It is simpler to follow the tradition and use
the same symbol for all derivatives. Actually, there is no danger for confusion
in doing this, because the equality f =
d
d
, or its equivalent = f, acquires
its real meaning through the (A) =
_
A
f d, A .
3. As we just observed, the real meaning of the symbol
d
d
is through
(A) =
_
A
d
d
d, A ,
which, after formally simplifying the fraction (!), changes into the true equality
(A) =
_
A
d.
4. Theorem 10.11 implies that the Radon-Nikodym of with respect to
, if it exists, is unique when is a seminite measure, provided we disregard
sets of zero -measure.
The following propositions give some properties of Radon-Nikodym deriva-
tives of calculus type.
1
,
2
be complex or -nite signed measures and a
-nite measure on (X, ). If
1
,
2
and
1
+
2
is dened, then
1
+
2

and
d(
1
+
2
)
d
=
d
1
d
+
d
2
d
, a.e. on X.
Proof: We have (
1
+
2
)(A) =
_
A
d1
d
d+
_
A
d2
d
d =
_
A
_
d1
d
+
d2
d
_
d for all
A and, hence,
d(1+2)
d
=
d1
d
+
d2
d
-a.e. on X.
Proposition 10.17 Let be a complex or a -nite signed measure and a
-nite measure on (X, ). If and C or R, then and
d()
d
=
d
d
, a.e. on X.
Proof: We have ()(A) =
_
A
d
d
d =
_
A
_
d
d
_
d for all A and, hence,
d()
d
=
d
d
-a.e. on X.
Proposition 10.18 (Chain rule.) Let be a complex or -nite signed measure
and
, be -nite measures on (X, ). If
and
, then
and
d
d
=
d
d
d
, a.e. on X.
Proof: If A has (A) = 0, then
(A) = 0 and, hence, (A) = 0. Therefore,

.
Theorem 10.10 implies that (A) =
_
A
d
d
=
_
A
d
d
d
d for every
A and, hence,
d
d
=
d
d
d
-a.e. on X.
Proposition 10.19 Let and
be two -nite measures on (X, ). If

and
, then
d
d
d
= 1 , a.e. on X.
Proof: We have (A) =
_
A
d for every A and, hence,
d
d
= 1 -a.e. on X.
The result of this proposition is a trivial consequence of Proposition 10.18.
Proposition 10.20 Let be a -nite measure on (X, ). Then [[ and
d
d[[
= 1 , a.e. on X.
Proof: Proposition 10.11 implies that
d
d||
[[ =
d
d||
[[
= [[ and, hence,
d
d||
= 1 [[ -a.e. on X.
10.7 Dierentiation of indenite integrals in R
n
.
Let f : [a, b] Rbe a Riemann-integrable function. The Fundamental Theorem
of Calculus says that, for every x [a, b] which is a continuity point of f, we
have
d
dx
_
x
a
f(y) dy = f(x). This, of course, means that
lim
r0+
_
x+r
a
f(y) dy
_
x
a
f(y) dy
r
= lim
r0+
_
xr
a
f(y) dy
_
x
a
f(y) dy
r
= f(x).
Adding the two limits, we nd
lim
r0+
_
x+r
xr
f(y) dy
2r
= f(x).
In this (and the next) section we shall prove a far reaching generalisation of
this result: a fundamental theorem of calculus for indenite Lebesgue-integrals
and, more generally, for locally nite Borel measures in R
n
.
Lemma 10.6 (N. Wiener) Let B
1
, . . . , B
m
be open balls in R
n
. There exist
pairwise disjoint B
11
, . . . , B
i
k
so that
m
n
(B
i1
) + +m
n
(B
i
k
)
1
3
n
m
n
(B
1
B
m
).
Proof: From B
1
, . . . , B
m
we choose a ball B
i1
with largest radius. (There may
be more than one balls with the same largest radius and we choose any one of
them.) Together with B
i1
we collect all other balls, its satellites, which intersect
it and call their union (B
i1
included) C
1
. Since each of these balls has radius
not larger than the radius of B
i1
, we see that C
1
B
i1
, where B
i1
is the ball
with the same center as B
i1
and radius three times the radius of B
i1
. Therefore,
m
n
(C
1
) m
n
(B
i1
) = 3
n
m
n
(B
i1
).
The remaining balls have empty intersection with B
i1
and from them we
choose a ball B
i2
with largest radius. Of course, B
i2
does not intersect B
i1
.
Together with B
i2
we collect all other balls (from the remaining ones), its satel-
lites, which intersect it and call their union (B
i2
included) C
2
. Since each of
these balls has radius not larger than the radius of B
i2
, we have C
2
B
i2
, where
B
i2
is the ball with the same center as B
i2
and radius three times the radius of
B
i2
. Therefore,
m
n
(C
2
) m
n
(B
i2
) = 3
n
m
n
(B
i2
).
We continue this procedure and, since at every step at least one ball is
collected (B
i1
at the rst step, B
i2
at the second step and so on), after at most
10.7. DIFFERENTIATION OF INDEFINITE INTEGRALS IN R
N
. 213
m steps, say at the kth step, the procedure will stop. Namely, after the rst
k1 steps, the remaining balls have empty intersection with B
i1
, . . . , B
i
k1
and
from them we choose a ball B
i
k
with largest radius. This B
i
k
does not intersect
B
i1
, . . . , B
i
k1
. All remaining balls intersect B
i
k
, they are its satellites, (since
this is the step where the procedure stops) and form their union (B
i
k
included)
C
k
. Since each of these balls has radius not larger than the radius of B
i
k
, we
have C
k
B
i
k
, where B
i
k
is the ball with the same center as B
i
k
and radius
three times the radius of B
i
k
. Therefore,
m
n
(C
k
) m
n
(B
i
k
) = 3
n
m
n
(B
i
k
).
It is clear that each of the original balls B
1
, . . . , B
m
is either chosen as one
of B
i1
, . . . , B
i
k
or is a satellite of one of B
i1
, . . . , B
i
k
. Therefore, B
1
B
m
=
C
1
C
k
and, hence,
m
n
(B
1
B
m
) = m
n
(C
1
C
k
) m
n
(C
1
) + +m
n
(C
k
)
3
n
_
m
n
(B
i1
) + +m
n
(B
i
k
)
_
.
n
R or C be Lebesgue-measurable. We say
that f is locally Lebesgue-integrable if for every x R
n
there is an open
neighborhood U
x
of x so that
_
Ux
[f(y)[ dm
n
(y) < +.
Lemma 10.7 Let f : R
n
R or C be Lebesgue-measurable. Then f is locally
Lebesgue-integrable if and only if
_
M
[f(y)[ dm
n
(y) < + for every bounded set
M L
n
.
Proof: Let f be locally Lebesgue-integrable and M R
n
be bounded. We
consider a compact set K M. (Such a K is the closure of M or just
a closed ball or a closed cube including M.) For each x K we take an
open neighborhood U
x
of x so that
_
Ux
[f(y)[ dm
n
(y) < +. We, then, take
nitely many x
1
, . . . , x
m
so that M K U
x1
U
xm
. This implies
_
M
[f(y)[ dm
n
(y)
_
Ux
1
[f(y)[ dm
n
(y) + +
_
Uxm
[f(y)[ dm
n
(y) < +.
If, conversely,
_
M
[f(y)[ dm
n
(y) < + for every bounded set M L
n
,
then
_
B(x;1)
[f(y)[ dm
n
(y) < + for every x and, hence, f is locally Lebesgue-
integrable.
Proposition 10.21 Let f, f
1
, f
2
: R
n
R or C be locally Lebesgue-integrable
and C. Then
(i) f is nite m
n
-a.e. on R
n
,
(ii) f
1
+ f
2
is dened m
n
-a.e. on R
n
and any Lebesgue-measurable denition
of f
1
+f
2
is locally Lebesgue-integrable,
(iii) f is locally Lebesgue-integrable.
Proof: (i) Lemma 10.7 implies
_
B(0;k)
[f(y)[ dm
n
(y) < + and, hence, f is
nite m
n
-a.e. in B(0; k) for every k. Since R
n
=
+
k=1
B(0; k), we nd that f is
nite m
n
-a.e. in R
n
.
(ii) By the result of (i), both f
1
, f
2
are nite and, hence, f
1
+ f
2
is dened
m
n
-a.e. on R
n
. We have
_
M
[f
1
(y) + f
2
(y)[ dm
n
(y)
_
M
[f
1
(y)[ dm
n
(y) +
_
M
[f
2
(y)[ dm
n
(y) < + for every bounded M R
n
and, by Lemma 10.7,
f
1
+f
2
is locally Lebesgue-integrable.
(iii) Similarly,
_
M
[f(y)[ dm
n
(y) = [[
_
M
[f(y)[ dm
n
(y) < + for all bounded
M R
n
and, hence, f is locally Lebesgue-integrable.
The need for local Lebesgue-integrability (or for local niteness of measures)
is for denitions like the following one to make sense. Of course, we may restrict
to Lebesgue-integrability if we like.
n
R or C be locally Lebesgue-integrable. The
function M(f) : R
n
[0, +], dened by
M(f)(x) = sup
B open ball, Bx
1
m
n
(B)
_
B
[f(y)[ dm
n
(y)
for every x R
n
, is called the Hardy-Littlewood maximal function of f.
1
, f
2
: R
n
R or C be locally Lebesgue-integrable
and C. Then
(i) M(f
1
+f
2
) M(f
1
) +M(f
2
),
(ii) M(f) = [[M(f).
Proof: (i) For all x and all open balls B x,
1
mn(B)
_
B
[f
1
(y) +f
2
(y)[ dm
n
(y)
1
mn(B)
_
B
[f
1
(y)[ dm
n
(y)+
1
mn(B)
_
B
[f
2
(y)[ dm
n
(y) M(f
1
)(x)+M(f
2
)(x). Ta-
king supremum of the left side, we get M(f
1
+f
2
)(x) M(f
1
)(x) +M(f
2
)(x).
(ii) Also,
1
mn(B)
_
B
[f(y)[ dm
n
(y) = [[
1
mn(B)
_
B
[f(y)[ dm
n
(y) [[M(f)(x)
and, taking the supremum of the left side, M(f)(x) [[M(f)(x). Conversely,
M(f)(x)
1
mn(B)
_
B
[f(y)[ dm
n
(y) = [[
1
mn(B)
_
B
[f(y)[ dm
n
(y) and, taking
the supremum of the right side, M(f)(x) [[M(f)(x).
Lemma 10.8 Let f : R
n
R or C be locally Lebesgue-integrable. Then, for
every t > 0, the set x R
n
[ t < M(f)(x) is open in R
n
.
Proof: Let U = x R
n
[ t < M(f)(x) and x U. Then t < M(f)(x) and,
hence, there exists an open ball B x so that t <
1
mn(B)
_
B
[f(y)[ dm
n
(y). If we
take an arbitrary x
B, then
1
mn(B)
_
B
[f(y)[ dm
n
(y) M(f)(x
) and, thus,
t < M(f)(x
). Therefore, B U and U is open in R

n
.
Since x R
n
[ t < M(f)(x) is open, it is also Lebesgue-measurable.
Theorem 10.14 (Hardy, Littlewood) Let f : R
n
R or C be Lebesgue-
integrable. Then, for every t > 0, we have
m
n
(x R
n
[ t < M(f)(x))
3
n
t
_
R
n
[f(y)[ dm
n
(y).
N
. 215
Proof: We take arbitrary compact K U = x R
n
[ t < M(f)(x) and for
each x K we nd an open ball B
x
x with t <
1
mn(Bx)
_
Bx
[f(y)[ dm
n
(y).
Since K is compact, there exist x
1
, . . . , x
m
so that K B
x1
B
xm
. By
Lemma 10.6, there exist pairwise disjoint B
xi
1
, . . . , B
xi
k
so that
m
n
(B
x1
B
xm
) 3
n
_
m
n
(B
xi
1
) + +m
n
(B
xi
k
)
_
.
Then
m
n
(K) m
n
(B
x1
B
xm
)
3
n
t
_
_
Bx
i
1
[f(y)[ dm
n
(y) + +
_
Bx
i
k
[f(y)[ dm
n
(y)
_
3
n
t
_
R
n
[f(y)[ dm
n
(y).
By the regularity of m
n
, m
n
(U) = supm
n
(K) [ K is compact U and
we conclude that m
n
(U)
3
n
t
_
R
n
[f(y)[ dm
n
(y).
Observe that the quantity m
n
(x R
n
[ t < M(f)(x)) is nothing but the
value at t of the distribution function
M(f)
of M(f). Therefore, another way
to state the result of Theorem 10.14 is
M(f)
(t)
3
n
t
_
R
n
[f(y)[ dm
n
(y).
Denition 10.20 Let (X, , ) be a measure space and g : X R or C be
measurable. We say that g is weakly integrable over X with respect to
if there is a c < + so that
|g|
(t)
c
t
for every t > 0.
Another way to state Theorem 10.14 is: if f is Lebesgue-integrable, then
M(f) is weakly Lebesgue-integrable.
Proposition 10.23 Let (X, , ) be a measure space, g, g
1
, g
2
: X R or C
be weakly integrable over X with respect to and C. Then
(i) g is nite -a.e. on X,
(ii) g
1
+g
2
is dened -a.e. on X and any measurable denition of g
1
+g
2
is weakly integrable over X with respect to ,
(iii) g is weakly integrable over X with respect to .
Proof: (i)
|g|
(t)
c
t
for all t > 0 implies that (x X[ g(x) is innite)
(x X[ n < [g(x)[)
c
n
for all n and, thus, (x X[ g(x) is innite) = 0.
(ii) By (i) both g
1
and g
2
are nite -a.e. on X and, hence, g
1
+g
2
is dened -
a.e. on X. If (x X[ t < [g
1
(x)[)
c1
t
and (x X[ t < [g
2
(x)[)
c2
t
for
all t > 0, then any measurable denition of g
1
+g
2
satises, for every t > 0,
the estimate: (x X[ t < [g
1
(x) + g
2
(x)[) (x X[
t
2
< [g
1
(x)[) +
(x X[
t
2
< [g
2
(x)[)
2c1+2c2
t
.
(iii) If (x X[ t < [g(x)[)
c
t
for all t > 0, then (x X [ t < [g(x)[) =
(x X[
t
||
< [g(x)[)
c||
t
for all t > 0.
Proposition 10.24 Let (X, , ) be a measure space and g : X R or C be
integrable over X with respect to . Then g is also weakly integrable over X
with respect to .
Proof: We have
|g|
(t) = (x X[ t < [g(x)[)
1
t
_
{xX| t<|g(x)|}
[g[ d
1
t
_
X
[g[ d for all t > 0. Therefore,
|g|
(t)
c
t
for all t > 0, where c =
_
X
[g[ d.
Example
The converse of Proposition 10.24 is not true. Consider, for example, the func-
tion g(x) =
1
|x|
n
, x R
n
.
By Proposition 8.12,
_
R
n
[g(x)[ dm
n
(x) =
n1
(S
n1
)
_
+
0
1
r
n
r
n1
dr =
n1
(S
n1
)
_
+
0
1
r
dr = +. But, x R
n
[ t < [g(x)[ = B(0; t
1
n
) and,
hence,
|g|
(t) = v
n
(t
1
n
)
n
=
vn
t
for every t > 0, where v
n
= m
n
(B(0; 1)).
The next result says that the Hardy-Littlewood maximal function is never
(except if the function is zero) Lebesgue-integrable.
Proposition 10.25 Let f : R
n
R or C be Lebesgue-integrable. Then M(f)
is locally Lebesgue-integrable. If M(f) is Lebesgue-integrable, then f = 0 -a.e.
on R
n
.
Proof: Let A = x R
n
[ f(x) ,= 0 and assume that m
n
(A) > 0. Since
A =
+
k=1
(AB(0; k)), we get that m
n
(AB(0; k)) > 0 for at least one k 1.
We set M = AB(0; k) and we have got a bounded set M so that m
n
(M) > 0
and [x[ k for every x M. Since f(x) ,= 0 for every x M, we have that
_
M
[f(y)[ dm
n
(y) > 0.
We take any x with [x[ k and observe that there is an open ball B of
diameter [x[ + k + 1 containing x and including M. If v
n
= m
n
(B(0; 1)), then
M(f)(x)
1
mn(B)
_
B
[f(y)[ dm
n
(y)
2
n
vn(|x|+k+1)
n
_
M
[f(y)[ dm
n
(y)
c
|x|
n
,
with c =
2
n
vn3
n
_
M
[f(y)[ dm
n
(y) > 0. This implies
_
R
n
[M(f)(x)[ dm
n
(x)
c
_
{xR
n
| |x|k}
1
|x|
n
dm
n
(x) = c
n1
(S
n1
)
_
+
k
1
r
n
r
n1
dr = +.
The next result is a direct generalization of the fundamental theorem of
calculus and the proofs are identical.
Lemma 10.9 Let g : R
n
C be continuous on R
n
. Then
lim
r0+
1
m
n
(B(x; r))
_
B(x;r)
[g(y) g(x)[ dm
n
(y) = 0
for every x R
n
.
Proof: Let > 0 and take > 0 so that [g(y) g(x)[ for every y R
n
with
[y x[ < . Then, for every r < ,
1
mn(B(x;r))
_
B(x;r)
[g(y) g(x)[ dm
n
(y)
1
mn(B(x;r))
_
B(x;r)
dm
n
(y) = .
N
. 217
Theorem 10.15 (Lebesgue) Let f : R
n
R or C be locally Lebesgue-integra-
ble. Then,
lim
r0+
1
m
n
(B(x; r))
_
B(x;r)
[f(y) f(x)[ dm
n
(y) = 0
for m
n
-a.e. x R
n
.
Proof: We rst assume that f is integrable.
We take an arbitrary > 0 and, through Theorem 7.16, we nd g : R
n
C,
continuous on R
n
, so that
_
R
n
[g f[ dm
n
< . For all x R
n
and r > 0 we get
1
mn(B(x;r))
_
B(x;r)
[f(y)f(x)[ dm
n
(y)
1
mn(B(x;r))
_
B(x;r)
[f(y)g(y)[ dm
n
(y)+
1
mn(B(x;r))
_
B(x;r)
[g(y)g(x)[ dm
n
(y)+
1
mn(B(x;r))
_
B(x;r)
[g(x)f(x)[ dm
n
(y)
M(f g)(x) +
1
mn(B(x;r))
_
B(x;r)
[g(y) g(x)[ dm
n
(y) +[g(x) f(x)[.
We set A(f)(x; r) =
1
mn(B(x;r))
_
B(x;r)
[f(y) f(x)[ dm
n
(y) and the last in-
equality, together with Lemma 10.9, implies
limsup
r0+
A(f)(x; r) M(f g)(x) + 0 +[g(x) f(x)[.
Now, for every t > 0, we get m
n
(x R
n
[ t < limsup
r0+
A(f)(x; r))
m
n
(x R
n
[
t
2
< M(f g)(x)) + m
n
(x R
n
[
t
2
< [g(x) f(x)[)
23
n
t
_
R
n
[f g[ dm
n
+
2
t
_
R
n
[f g[ dm
n

23
n
+2
t
, where the second inequality
is a consequence of Theorem 10.14. Since is arbitrary, we nd, for all t > 0,
m
n
(x R
n
[ t < limsup
r0+
A(f)(x; r)) = 0.
By the subadditivity of m
n
, m
n
(x R
n
[ limsup
r0+
A(f)(x; r) ,= 0)
+
k=1
m
n
(x R
n
[
1
k
< limsup
r0+
A(f)(x; r)) = 0 and, hence,
m
n
(x R
n
[ limsup
r0+
A(f)(x; r) ,= 0) = 0.
This implies that limsup
r0+
A(f)(x; r) = 0 for m
n
-a.e. x R
n
and, since
liminf
r0+
A(f)(x; r) 0 for every x R
n
, we conclude that
lim
r0+
A(f)(x; r) = 0
for m
n
-a.e. x R
n
.
Now, let f be locally Lebesgue-integrable. We x an arbitrary k 2 and
consider the function h = f
B(0;k)
. Then h is Lebesgue-integrable and, for every
x B(0; k 1) and every r 1, we have A(f)(x; r) = A(h)(x; r). By what
we have already proved, this implies that lim
r0+
A(f)(x; r) = 0 for m
n
-a.e.
x B(0; k 1). Since k is arbitrary, we conclude that lim
r0+
A(f)(x; r) = 0
for m
n
-a.e. x R
n
.
n
R or C be locally Lebesgue-integrable. The set
L
f
of all x R
n
for which lim
r0+
1
mn(B(x;r))
_
B(x;r)
[f(y) f(x)[ dm
n
(y) = 0
is called the Lebesgue set of f.
Example
If x is a continuity point of f, then x belongs to the Lebesgue set of f. The
proof of this fact is, actually, the proof of Lemma 10.9.
n
R or C be locally Lebesgue-integrable. Then,
for every x in the Lebesgue set of f, we have
lim
r0+
1
m
n
(B(x; r))
_
B(x;r)
f(y) dm
n
(y) = f(x).
Proof: Indeed, for all x L
f
we have
1
mn(B(x;r))
_
B(x;r)
f(y) dm
n
(y) f(x)
1
mn(B(x;r))
_
B(x;r)
[f(y) f(x)[ dm
n
(y) 0.
Denition 10.22 Let x R
n
and c be a collection of sets in L
n
with the
property that there is a c > 0 so that for every E c there is a ball B(x; r)
with E B(x; r) and m
n
(E) cm
n
(B(x; r)). Then the collection c is called a
thick family of sets at x.
Examples
1. Any collection of qubes containing x and any collection of balls containing x
is a thick family of sets at x.
2. Consider any collection c all elements of which are intervals S containing x.
Let A
S
be the length of the largest side and a
S
be the length of the smallest
side of S. If there is a constant c > 0 so that
aS
AS
c for every S c, then c is
a thick family of sets at x.
n
R or C be locally Lebesgue-integrable. Then,
for every x in the Lebesgue set of f and for every thick family c of sets at x,
we have
lim
EE,mn(E)0+
1
m
n
(E)
_
E
[f(y) f(x)[ dm
n
(y) = 0
lim
EE,mn(E)0+
1
m
n
(E)
_
E
f(y) dm
n
(y) = f(x).
Proof: There is a c > 0 so that for every E c there is a ball B(x; r
E
) with
E B(x; r
E
) and m
n
(E) cm
n
(B(x; r
E
)). If x L
f
, then for every > 0
there is a > 0 so that r < implies
1
mn(B(x;r))
_
B(x;r)
[f(y)f(x)[ dm
n
(y) < c.
If m
n
(E) < cv
n
n
, where v
n
= m
n
(B(0; 1)), then r
E
< and, hence,
1
mn(E)
_
E
[f(y) f(x)[ dm
n
(y)
1
cmn(B(x;rE))
_
B(x;rE)
[f(y) f(x)[ dm
n
(y) < .
This means that lim
EE,mn(E)0+
1
mn(E)
_
E
[f(y) f(x)[ dm
n
(y) = 0.
By
1
mn(E)
_
E
f(y) dm
n
(y) f(x)

1
mn(E)
_
E
[f(y) f(x)[ dm
n
(y) and by
the rst limit, we prove the second.
10.8 Dierentiation of Borel measures in R
n
.
Denition 10.23 Any signed or complex measure on (R
n
, B
R
n) is called a
Borel signed or complex measure on R
n
.
10.8. DIFFERENTIATION OF BOREL MEASURES IN R
N
. 219
Denition 10.24 Let be a Borel signed measure in R
n
. We say that is
locally nite if for every x R
n
there is an open neighborhood U
x
of x so that
(U
x
) is nite.
This denition is indierent for complex measures, since complex measures
take only nite values.
Proposition 10.26 Let be a Borel signed measure in R
n
. Then, is locally
nite if and only if
+
and
are both locally nite if and only if [[ is locally

nite.
Proof: Since [[ =
+
+
, the second equivalence is trivial to prove. It is also

trivial to prove that is locally nite if [[ is locally nite.
Let be locally nite. For an arbitrary x R
n
we take an open neigh-
borhood U
x
of x so that (U
x
) is nite. Since (U
x
) =
+
(U
x
)
(U
x
), both
+
(U
x
) and
(U
x
) and, hence, [[(U
x
) are nite. Therefore, [[ is locally nite.
Proposition 10.27 Let be a Borel signed measure in R
n
. Then, is locally
nite if and only if (M) is nite for all bounded Borel sets M R
n
.
Proof: One direction is easy, since every open ball is a bounded set. For the
other direction, we suppose that is locally nite and, by Proposition 10.26,
that [[ is also locally nite. Lemma 5.7 implies that [(M)[ [[(M) < +
for all bounded Borel sets M R
n
.
Theorem 10.18 Let be a locally nite Borel signed measure or a complex
measure on R
n
with m
n
. Then,
lim
r0+
(B(x; r))
m
n
(B(x; r))
= 0
for m
n
-a.e. x R
n
.
Proof: If is complex, then [[ is a nite Borel measure on R
n
. Proposition
10.26 implies that, if is signed, then [[ is a locally nite Borel measure on
R
n
. Moreover, Proposition 10.10 implies that [[m
n
. Hence, there exist sets
R, M B
R
n with M R = R
n
, M R = so that R is null for m
n
and M is
null for [[.
We dene A([[)(x; r) =
||(B(x;r))
mn(B(x;r))
, take an arbitrary t > 0 and consider the
set M
t
= x M [ t < limsup
r0+
A([[)(x; r).
Since [[ is a regular measure and [[(M) = 0, there is an open set U so that
M
t
M U and [[(U) < . For each x M
t
, there is a small enough r
x
> 0
so that t < A([[)(x; r
x
) =
||(B(x;rx))
mn(B(x;rx))
and B(x; r
x
) U.
We form the open set V =
xMt
B(x; r
x
) and take an arbitrary compact
set K V . Now, there exist nitely many x
1
, . . . , x
m
M
t
so that K
B(x
1
; r
x1
) B(x
m
; r
xm
). Lemma 10.6 implies that there exist pairwise
disjoint B(x
i1
; r
xi
1
), . . . , B(x
i
k
; r
xi
k
) so that m
n
(B(x
1
; r
x1
) B(x
m
; r
xm
))
3
n
_
m
n
(B(x
i1
; r
xi
1
)) + +m
n
(B(x
i
k
; r
xi
k
))
_
. All these imply that
m
n
(K)
3
n
t
_
[[(B(x
i1
; r
xi
1
)) + +[[(B(x
i
k
; r
xi
k
))
_
3
n
t
[[(U) <
3
n
t
.
By the regularity of m
n
and since K is an arbitrary compact subset of
V , we nd that m
n
(V )
3
n
t
. Since M
t
V , we have that m
n
(M
t
)
3
n
t
and, since is arbitrary, we conclude that M
t
is Lebesgue-measurable and
m
n
(M
t
) = 0. Finally, since x M[ limsup
r0+
A([[)(x; r) ,= 0 =
+
k=1
M1
k
,
we get that limsup
r0+
A([[)(x; r) = 0 for m
n
-a.e. x R
n
. Now, from
0 liminf
r0+
A([[)(x; r), we conclude that lim
r0+
A([[)(x; r) = 0 for m
n
-
a.e. x R
n
.
Lemma 10.10 Let be a locally nite Borel signed measure on R
n
. Then
is -nite and let = + be the Lebesgue decomposition of with respect
to m
n
, where m
n
and m
n
. Then both and are locally nite Borel
signed measures.
Moreover, if f is any Radon-Nikodym derivative of with respect to m
n
,
then f is locally Lebesgue-integrable.
Proof: Since R
n
=
+
k=1
B(0; k) and (B(0; k)) is nite for every k, we nd
that is -nite and Theorem 10.12 implies the existence of the Lebesgue
decomposition of .
Since m
n
, there exist Borel sets R, N with R N = X, R N = so
that R is null for m
n
and N is null for . From m
n
, we see that R is null
for , as well.
Now, take any bounded Borel set M. Since (M) is nite, Theorem 10.1
implies that (MN) is nite. Now, we have (M) = (MR) +(MN) =
(M N) = (M N) + (M N) = (M N) and, hence, (M) is nite.
From (M) = (M) +(M) we get that (M) is also nite. We conclude that
and are locally nite.
Take, again, any bounded Borel set M. Then
_
M
f(x) dm
n
(x) = (M)
is nite and, hence,
_
X
[f(x)[ dm
n
(x) < +. This implies that f is locally
Lebesgue-integrable.
Theorem 10.19 Let be a locally nite Borel signed measure or a Borel com-
plex measure on R
n
. If f is any Radon-Nikodym derivative of the absolutely
continuous part of with respect to m
n
, then
lim
r0+
(B(x; r))
m
n
(B(x; r))
= f(x)
for m
n
-a.e. x R
n
.
Proof: Let = + be the Lebesgue decomposition of with respect to m
n
,
where m
n
, m
n
and = fm
n
. If is signed, Lemma 10.10 implies that
is a locally nite Borel signed measure and f is locally Lebesgue-integrable. If
is complex, then is complex and f is Lebesgue-integrable. Theorems 10.16
and 10.18 imply
lim
r0+
(B(x; r))
m
n
(B(x; r))
= lim
r0+
1
m
n
(B(x; r))
_
B(x;r)
f(y) dm
n
(y)
10.8. DIFFERENTIATION OF BOREL MEASURES IN R
N
. 221
+ lim
r0+
(B(x; r))
m
n
(B(x; r))
= f(x) + 0
= f(x)
for m
n
-a.e. x R
n
.
Theorem 10.20 Let be a locally nite Borel signed measure or a Borel com-
plex measure on R
n
. If f is any Radon-Nikodym derivative of the absolutely
continuous part of with respect to m
n
, then, for m
n
-a.e. x R
n
,
lim
EE,mn(E)0+
(E)
m
n
(E)
= f(x)
for every thick family c of sets at x.
Proof: If is the singular part of with respect to m
n
, then [[m
n
and, by
Theorem 10.18, lim
r0+
||(B(x;r))
mn(B(x;r))
= 0 for m
n
-a.e. x R
n
.
We, now, take any x for which lim
r0+
||(B(x;r))
mn(B(x;r))
= 0 and any thick family
c of sets at x. This means that there is a c > 0 so that for every E c there
is a ball B(x; r
E
) with E B(x; r
E
) and m
n
(E) cm
n
(B(x; r
E
)). For every
> 0 there is a > 0 so that r < implies
||(B(x;r))
mn(B(x;r))
< c. Therefore, if
m
n
(E) < cv
n
n
, where v
n
= m
n
(B(0; 1)), then r
E
< and, hence,
(E)
mn(E)

||(E)
mn(E)

1
c
||(B(x;rE))
mn(B(x;rE))
< . This means that, for m
n
-a.e. x R
n
,
lim
EE,mn(E)0+
(E)
m
n
(E)
= 0
for every thick family c of sets at x.
We combine this with Theorem 10.17 to complete the proof.
10.9 Exercises.
1. Let be a signed measure on (X, ) and let
1
,
2
be two measures on
(X, ) at least one of which is nite. If =
1

2
, prove that
+

1
and

2
.
2. Let be the counting measure on (N, T(N)) and be the point-mass
distribution on N induced by the function a
n
=
1
2
n
, n N. Prove that
there is an
0
> 0 and a sequence E
k
of subsets of N, so that (E
k
) 0
and (E
k
)
0
for all k. On the other hand, prove that .
3. Let
1
,
1
be -nite measures on (X
1
,
1
) and
2
,
2
be -nite measures
on (X
2
,
2
). If
1

1
and
2

2
, prove that
1

2

1

2
and
that
d(
1

2
)
d(
1

2
)
(x
1
, x
2
) =
d
1
d
1
(x
1
)
d
2
d
2
(x
2
)
for (
1
2
)-a.e. (x
1
, x
2
) X
1
X
2
.
4. Let be the counting measure on (R, B
R
).
(i) Prove that m
1
. Is there any f so that m
1
= f ?
(ii) Is there any Lebesgue decomposition of with respect to m
1
?
5. Generalization of the Radon-Nikodym Theorem.
Let be a signed measure and be a -nite measure on (X, ) so that
. Prove that there is a measurable f : X R, so that
_
X
f d
exists and = f.
6. Generalization of the Lebesgue Decomposition Theorem.
Let be a -nite signed measure and a measure on (X, ). Prove that
there are unique -nite signed measures , on (X, ) so that ,
and = +.
7. Let , be two measures on (X, ) with . If = +, prove that
. If f : X [0, +] is measurable and = f, prove that
0 f < 1 -a.e. on X and =
f
1f
.
8. Conditional Expectation.
Let be a -nite measure on (X, ),
0
be a -algebra with
0
and
be the restriction of the measure on (X,
0
).
(i) If f : X R or C is measurable and
_
X
f d exists, prove that
there is a
0
measurable f
0
: X R or, respectively, C so that
_
X
f
0
d
exists and
_
A
f
0
d =
_
A
f d, A
0
.
If f
0
has the same properties as f
0
, prove that f
0
= f
0
-a.e. on X.
10.9. EXERCISES. 223
Any f
0
with the above properies is called a conditional expectation of
f with respect to
0
and it is denoted by
E(f[
0
).
(ii) Prove that
(a) E(f[) = f -a.e. on X,
(b) E(f +g[
0
) = E(f[
0
) +E(g[
0
) -a.e. on X,
(c) E(f[
0
) = E(f[
0
) -a.e. on X,
(d) if g is
0
measurable, then E(gf[
0
) = gE(f[
0
) -a.e. on X,
(e) if
1

0
, then E(f[
1
) = E(E(f[
0
)[
1
) -a.e. on X.
9. Let be a real or complex measure on (X, ). If (X) = [[(X), prove
that = [[.
10. Let be a signed or complex measure on (X, ). We say that A
1
, A
2
, . . .
is a (countable) measurable partition of A , if A
k
for all k,
the sets A
1
, A
2
, . . . are pairwise disjoint and A = A
1
A
2
. Prove
that
[[(A) = sup
_
+
k=1
[(A
k
)[ [ A
1
, A
2
, . . . is a measurable partition of A
_
for every A .
11. A variant of the Hardy-Littlewood maximal function.
Let f : R
n
R or C be locally Lebesgue-integrable. We dene
H(f)(x) = sup
r>0
1
m
n
(B(x; r))
_
B(x;r)
[f(y)[ dm
n
(y)
for every x R
n
.
(i) Prove that the set x R
n
[ t < H(f)(x) is open for every t > 0.
(ii) Prove that
1
2
n
M(f)(x) H(f)(x) M(f)(x) for every x R
n
.
One may dene other variants of the Hardy-Littlewood maximal function
by taking the supremum of the mean values of [f[ over open cubes con-
taining the point x or open cubes centered at the point x. The results are
similar.
12. The Vitali Covering Theorem.
Let E R
n
and let ( be a collection of open balls with the property that
for every x E and every > 0 there is a B ( so that x B and
m
n
(B) < . Prove that there are pairwise disjoint B
1
, B
2
, . . . ( so that
m
n
(E
+
k=1
B
k
) = 0.
13. Points of density.
Let E L
n
. If x R
n
, we set
D
E
(x) = lim
r0+
m
n
(E B(x; r))
m
n
(B(x; r))
whenever the limit exists. If D
E
(x) = 1, we say that x is a density point
of E.
(i) If x is an interior point of E, prove that it is a density point of E.
(ii) Prove that m
n
-a.e. x E is a density point of E.
(iii) For any (0, 1) nd x R and E L
1
so that D
E
(x) = . Also,
nd x R and E L
1
so that D
E
(x) does not exist.
14. Let be a signed or a complex measure on (X, ) and A . Prove that
[[(A) = 0 if and only if (B) = 0 for all B , B A.
15. Let f be the Cantors function on [0, 1] (see Exercise 4.6.7) extended as
0 on (, 0) and as 1 on (1, +) and let
f
be the Lebesgue-Stieltjes
measure on (R, B
R
) induced by f. Prove that
f
m
1
.
16. Let be a signed measure on (X, ). Prove that
+
,
[[ and nd
formulas for Radon-Nikodym derivatives
d
+
d||
and
d
d||
.
17. Let be a nite measure on (X, ). We dene
d(A, B) = (AB), A, B .
(i) Prove that (, d) is a complete metric space.
(ii) If is a real or a complex measure on (X, ), prove that is continuous
on (with respect to d) if and only if is continuous at (with respect
to d) if and only if .
Chapter 11
The classical Banach spaces
11.1 Normed spaces.
Denition 11.1 Let Z be a linear space over the eld F = R or over the eld
F = C and let | | : Z R have the properties:
(i) |u +v| |u| +|v|, for all u, v Z,
(ii) |u| = [[|u|, for all u Z and F,
(iii) |u| = 0 implies u = o, where o is the zero element of Z.
Then, | | is called a norm on Z and (Z, | |) is called a normed space.
If | | is understood, we may say that Z is a normed space.
Proposition 11.1 If | | is a norm on the linear space Z, then
(i) |o| = 0, where o is the zero element of Z,
(ii) | u| = |u|, for all u Z,
(iii) |u| 0, for all u Z.
Proof: (i) |o| = |0 o| = [0[|o| = 0.
(ii) | u| = |(1)u| = [ 1[|u| = |u|.
(iii) 0 = |o| = |u + (u)| |u| +| u| = 2|u| and, hence, 0 |u|.
Proposition 11.2 Let (Z, | |) be a normed space. If we dene d : Z Z R
by
d(u, v) = |u v|
for all u, v Z, then d is a metric on Z.
Proof: Using Proposition 11.1, we have
a. d(u, v) = |u v| 0 for all u, v Z and, if d(u, v) = 0, then |u v| = 0
and, hence, u v = o or, equivalently, u = v.
b. d(u, v) = |uv| = |(uw)+(wv)| |uw|+|wv| = d(u, w)+d(w, v).
Denition 11.2 Let (Z, | |) be a normed space. If d is the metric dened in
Proposition 11.2, then d is called the metric induced on Z by | |.
225
226 CHAPTER 11. THE CLASSICAL BANACH SPACES
Therefore, if (Z, | |) is a normed space, then (Z, d) is a metric space and we
can study all notions related to the notion of a metric space, like convergence
of sequences, open and closed sets and so on.
Open balls have the form B(u; r) = v Z [ |v u| < r.
A sequence u
n
in Z converges to u Z if |u
n
u| 0 as n +. We
denote this by: u
n
u in Z.
A set U Z is open in Z if for every u U there is an r > 0 so that
B(u; r) U. Any union of open sets in Z is open in Z and any nite intersection
of open sets in Z is open in Z. The sets and Z are open in Z.
A set K Z is closed in Z if its complement Z K is open in Z or,
equivalently, if the limit of every sequence in K (which has a limit) belongs to
K. Any intersection of closed sets in Z is closed in Z and any nite union of
closed sets in Z is closed in Z. The sets and Z are closed in Z.
A set K Z is compact if every open cover of K has a nite subcover of K.
Equivalently, K is compact if every sequence in K has a convergent subsequence
with limit in K.
A sequence u
n
in Z is a Cauchy sequence if |u
n
u
m
| 0 as n, m
+. Every convergent sequence is Cauchy. If every Cauchy sequence in Z is
convergent, then Z is a complete metric space.
Denition 11.3 If the normed space (Z, | |) is complete as a metric space
(with the metric induced by the norm), then it is called a Banach space.
If there is no danger of confusion, we say that Z is a Banach space.
There are some special results based on the combination of the linear and
the metric structure of a normed space. We rst dene, as in any linear space,
u +A = u +v [ v A, A = v [ v A
for all A Z, u Z and F. We also dene, for every u Z and every
F 0, the translation
u
: Z Z and the dilation l
: Z Z, by
u
(v) = v +u, l
(v) = v
for all v Z. It is trivial to prove that translations and dilations are one-to-one
transformations of Z onto Z and that
1
u
=
u
and l
1
= l 1
. It is obvious
that u +A =
u
(A) and A = l
(A).
Proposition 11.3 Let (Z, | |) be a normed space.
(i) u +B(v; r) = B(u +v; r) for all u, v Z and r > 0.
(ii) B(v; r) = B(v; [[r) for all v Z, F 0 and r > 0.
(iii) If u
n
u and v
n
v in Z, then u
n
+v
n
u +v in Z.
(iv) If
n
in F and u
n
u in Z, then
n
u
n
u in Z.
(v) Translations and dilations are homeomorphisms. This means that they,
together with their inverses, are continuous on Z.
(vi) If A is open or closed in Z and u Z, then u +A is open or, respectively,
closed in Z.
(vii) If A is open or closed in Z and F0, then A is open or, respectively,
closed in Z.
11.1. NORMED SPACES. 227
Proof: (i) w u+B(v; r) if and only if wu B(v; r) if and only if |wuv| <
r if and only if w B(u +v; r).
(ii) w B(v; r) if and only if
1
w B(v; r) if and only if |

1
w v| < r if and
only if |w v| < [[r if and only if w B(v; [[r).
(iii) |(u
n
+v
n
) (u +v)| |u
n
u| +|v
n
v| 0 as n +.
(iv) |
n
u
n
u| [
n
[|u
n
u| +[
n
[|u| 0 as n +, because
n
is bounded in F.
(v) If v
n
v in Z, then
u
(v
n
) = u + v
n
u + v =
u
(v), by (iii). Also,
l
(v
n
) = v
n
v = l
(v), by (iv). Therefore,

u
and l
are continuous on Z.
Their inverses are also continuous, because they are also a translation,
u
, and
a dilation, l 1
, respectively.
(vi) u +A =
1
u
(A) is the inverse image of A under the continuous
u
.
(vii) A = l
1
1
(A) is the inverse image of A under the continuous l 1
.
As in any linear space, we dene a linear functional on Z to be a function
l : Z F which satises
l(u +v) = l(u) +l(v), l(u) = l(u)
for every u, v Z and F. If l is a linear functional on Z, then l(o) = l(0o) =
0l(o) = 0 and l(u) = l((1)u) = (1)l(u) = l(u) for all u Z. We dene
the sum l
1
+l
2
: Z F of two linear functionals l
1
, l
2
on Z by
(l
1
+l
2
)(u) = l
1
(u) +l
2
(u), u Z
and the product l : Z F of a linear functional l on Z and a F by
(l)(u) = l(u), u Z.
It is trivial to prove that l
1
+l
2
and l are linear functionals on Z and that
the set Z
whose elements are all the linear functionals on Z,

Z
= l [ l is a linear functional on Z,
becomes a linear space under this sum and product. Z
is called the algebraic

dual of Z. The zero element of Z
is the linear functional o : Z F with

o(u) = 0 for every u Z and the opposite of a linear functional l on Z is the
linear functional l : Z F with (l)(u) = l(u) for every u Z.
Denition 11.4 Let (Z, | |) be a normed space and l Z
a linear functional
on Z. Then l is called a bounded linear functional on Z if there is an
M < + so that
[l(u)[ M|u|
for every u Z.
Theorem 11.1 Let (Z, | |) be a normed space and l Z
. The following are

equivalent.
(i) l is bounded.
(ii) l : Z F is continuous on Z.
(iii) l : Z F is continuous at o Z.
Proof: Suppose that l is bounded and, hence, there is an M < + so that
[l(u)[ M|u| for every u Z. If u
n
u in Z, then [l(u
n
)l(u)[ = [l(u
n
u)[
M|u
n
u| 0 as n + and, thus, l(u
n
) l(u) in F as n +. This
says that l is continuous on Z.
If l is continuous on Z, then it is certainly continuous at o Z.
Suppose that l is continuous at o Z. Then, for = 1 there exists a > 0
so that [l(u)[ = [l(u) l(o)[ < 1 for every u Z with |u| = |u o| < .
We take an arbitrary u Z o and have that
_
_

2u
u
_
_
=

2
< . Therefore,
l
_

2u
u
_
< 1, implying that [l(u)[

2
|u|. This is trivially true also for u = o

and we conclude that [l(u)[ M|u| for every u Z, where M =
2
. This says
that l is bounded.
Denition 11.5 Let (Z, | |) be a normed space. The set of all bounded linear
functionals on Z or, equivalently, of all continuous linear functionals on Z,
Z
= l [ l is a bounded linear functional on Z,

is called the topological dual of Z or the norm-dual of Z.
Proposition 11.4 Let (Z, | |) be a normed space and l a bounded linear func-
tional on Z. Then there is a smallest M with the property: [l(u)[ M|u| for
every u Z. This M
0
is characterized by the two properties:
(i) [l(u)[ M
0
|u| for every u Z,
(ii) for every m < M
0
there is a u Z so that [l(u)[ > m|u|.
Proof: We consider
M
0
= infM[ [l(u)[ M|u| for every u Z.
The set L = M[ [l(u)[ M|u| for every u Z is non-empty by assump-
tion and included in [0, +). Therefore M
0
exists and M
0
0. We take a
sequence M
n
in L so that M
n
M
0
and, from [l(u)[ M
n
|u| for every
u Z, we nd [l(u)[ M
0
|u| for every u Z.
If m < M
0
, then m / L and, hence, there is a u Z so that [l(u)[ > m|u|.
Denition 11.6 Let (Z, | |) be a normed space and l a bounded linear func-
tional on Z. The smallest M with the property that [l(u)[ M|u| for every
u Z is called the norm of l and it is denoted by |l|
.
Proposition 11.4, which proves the existence of |l|
, states also its charac-

terizing properties:
1. [l(u)[ |l|
|u| for every u Z,

2. for every m < |l|
there is a u Z so that [l(u)[ > m|u|.

The zero linear functional o : Z F is bounded and, since [o(u)[ = 0 0|u|
for every u Z, we have that
|o|
= 0.
On the other hand, if l Z
has |l|
= 0, then [l(u)[ 0|u| = 0 for every

u Z and, hence, l = o is the zero linear functional on Z.
Proposition 11.5 Let (Z, | |) be a normed space and l Z
. Then
|l|
= sup
uZ,u=o
[l(u)[
|u|
= sup
uZ,u=1
[l(u)[ = sup
uZ,u1
[l(u)[ .
Proof: Every u with |u| = 1 satises |u| 1. Therefore, sup
uZ,u=1
[l(u)[
sup
uZ,u1
[l(u)[.
Writing v =
u
u
for every u Z o, we have that |v| = 1. Therefore,
sup
uZ,u=o
|l(u)|
u
= sup
uZ,u=o
l
_
u
u
_
sup
uZ,u=1
[l(u)[.
For every u with |u| 1, we have [l(u)[ |l|
|u| |l|
and, thus,
sup
uZ,u1
[l(u)[ |l|
.
If we set M = sup
uZ,u=o
|l(u)|
u
, then
|l(u)|
u
M and, hence, [l(u)[ M|u|
for all u ,= o. Since this is obviously true for u = o, we have that |l|
M and
this nishes the proof.
Proposition 11.6 Let (Z, | |) be a normed space, l, l
1
, l
2
be bounded linear
functionals on Z and F. Then l
1
+l
2
and l are bounded linear functionals
on Z and
|l
1
+l
2
|
|l
1
|
+|l
2
|
, |l|
= [[|l|
.
Proof: We have that [(l
1
+ l
2
)(u)[ [l
1
(u)[ + [l
2
(u)[ |l
1
|
|u| + |l
2
|
|u| =
(|l
1
|
+ |l
2
|
)|u| for every u Z. This implies that l

1
+ l
2
is bounded and
that |l
1
+l
2
|
|l
1
|
+|l
2
|
.
Similarly, [(l)(u)[ = [[[l(u)[ [[|l|
|u| for every u Z. This implies

that l is bounded and that |l|
[[|l|
. If = 0, then the equality is

obvious. If ,= 0, to get the opposite inequality, we write [[[l(u)[ = [(l)(u)[
|l|
|u|. This implies that [l(u)[

l
||
|u| for every u Z and, hence, that
|l|

l
||
.
Proposition 11.6 together with the remarks about the norm of the zero func-
tional imply that Z
is a linear subspace of Z
and that | |
: Z
R is a
norm on Z
.
Theorem 11.2 If (Z, | |) is a normed space, then (Z
, | |
) is a Banach
space.
Proof: Let l
n
be a Cauchy sequence in Z
. For all u Z, [l
n
(u) l
m
(u)[ =
[(l
n
l
m
)(u)[ |l
n
l
m
|
|u| 0 as n, m +. Thus, l
n
(u) is a Cauchy
sequence in F and, hence, converges to some element of F. We dene l : Z F
by
l(u) = lim
n+
l
n
(u)
for every u Z.
For every u, v Z and F we have l(u + v) = lim
n+
l
n
(u + v) =
lim
n+
l
n
(u) + lim
n+
l
n
(v) = l(u) + l(v) and l(u) = lim
n+
l
n
(u) =
lim
n+
l
n
(u) = l(u). Therefore, l Z
.
There is N so that |l
n
l
m
|
1 for all n, m N. This implies that

[l
n
(u) l
m
(u)[ |l
n
l
m
|
|u| |u| for all u Z and all n, m N and,

taking the limit as n + and, taking m = N, we nd [l(u) l
N
(u)[ |u|
for all u Z. Therefore, [l(u)[ [l
N
(u)[ +|u| (|l
N
|
+1)|u| for every u Z

and, hence, l Z
.
For an arbitrary > 0 there is N so that |l
n
l
m
|
for all n, m N.
This implies [l
n
(u) l
m
(u)[ |l
n
l
m
|
|u| |u| for all u Z and all

n, m N and, taking the limit as m +, we nd [l
n
(u) l(u)[ |u| for
all u Z and all n N. Therefore, |l
n
l|
for all n N and, hence,

l
n
l in Z
.
Denition 11.7 Let Z and W be two linear spaces over the same F and a
function T : Z W. T is called a linear transformation or a linear
operator from Z to W if
T(u +v) = T(u) +T(v), T(u) = T(u)
for all u, v Z and all F.
Denition 11.8 Let (Z, | |
Z
) and (W, | |
W
) be two normed spaces and a
linear transformation T : Z W. We say that T is a bounded linear trans-
formation from Z to W if there exists an M < + so that
|T(u)|
W
M|u|
Z
for all u Z.
Theorem 11.3 Let (Z, ||
Z
) and (W, ||
W
) be two normed spaces and a linear
transformation T : Z W. The following are equivalent.
(i) T is bounded.
(ii) T : Z W is continuous on Z.
(iii) T : Z W is continuous at o Z.
Proof: Suppose that T is bounded and, hence, there is an M < + so that
|T(u)|
W
M|u|
Z
for every u Z. If u
n
u in Z, then |T(u
n
) T(u)|
W
=
|T(u
n
u)|
W
M|u
n
u|
Z
0 as n + and, thus, T(u
n
) T(u) in W
as n +. This says that T is continuous on Z.
If T is continuous on Z, then it is certainly continuous at o Z.
Suppose that T is continuous at o Z. Then, for = 1 there exists a > 0 so
that |T(u)|
W
= |T(u)T(o)|
W
< 1 for every u Z with |u|
Z
= |uo|
Z
< .
We take an arbitrary u Zo and have that
_
_

2uZ
u
_
_
Z
=

2
< . Therefore,
_
_
T
_

2uZ
u
__
_
W
< 1, implying that |T(u)|
W

2
|u|
Z
. This is trivially true
also for u = o and we conclude that |T(u)|
W
M|u|
Z
for every u Z, where
M =
2
. This says that T is bounded.

Proposition 11.7 Let (Z, | |
Z
) and (W, | |
W
bounded linear transformation T : Z W. Then there is a smallest M with
the property: |T(u)|
W
M|u|
Z
for every u Z. This M
0
is characterized by
the two properties:
(i) |T(u)|
W
M
0
|u|
Z
for every u Z,
(ii) for every m < M
0
there is a u Z so that |T(u)|
W
> m|u|
Z
.
Proof: We consider
M
0
= infM[ |T(u)|
W
M|u|
Z
for every u Z.
The set L = M[ |T(u)|
W
M|u|
Z
for every u Z is non-empty by
assumption and included in [0, +). Therefore M
0
exists and M
0
0. We take
a sequence M
n
in L so that M
n
M
0
and, from |T(u)|
W
M
n
|u|
Z
for
every u Z, we nd |T(u)|
W
M
0
|u|
Z
for every u Z.
If m < M
0
, then m / L and, hence, there is a u Z so that |T(u)|
W
>
m|u|
Z
.
Z
) and (W, | |
W
bounded linear transformation T : Z W. The smallest M with the property
that |T(u)|
W
M|u|
Z
for every u Z is called the norm of T and it is
denoted by |T|.
By Proposition 11.7, which proves the existence of |T|, we have:
1. |T(u)|
W
|T||u|
Z
for every u Z,
2. for every m < |T| there is a u Z so that |T(u)|
W
> m|u|
Z
.
The zero linear transformation o : Z W is bounded and, since |o(u)|
W
=
0 0|u|
Z
for every u Z, we have that
|o| = 0.
On the other hand, if T is a bounded linear transformation with |T| = 0,
then |T(u)|
W
0|u|
Z
= 0 for every u Z and, hence, T = o is the zero linear
transformation.
Z
) and (W, | |
W
bounded linear transformation T : Z W. Then
|T| = sup
uZ,u=o
|T(u)|
W
|u|
Z
= sup
uZ,uZ=1
|T(u)|
W
= sup
uZ,uZ1
|T(u)|
W
.
Proof: Every u with |u|
Z
= 1 satises |u|
Z
1. This, clearly, implies that
sup
uZ,uZ=1
|T(u)|
W
sup
uZ,uZ1
|T(u)|
W
.
Writing v =
u
uZ
for every u Z o, we have that |v|
Z
= 1. Therefore,
sup
uZ,u=o
T(u)W
uZ
= sup
uZ,u=o
_
_
T
_
u
uZ
__
_
W
sup
uZ,uZ=1
|T(u)|
W
.
For every u with |u|
Z
1, we have |T(u)|
W
|T||u|
Z
|T| and, thus,
sup
uZ,uZ1
|T(u)|
W
|T|.
If we set M = sup
uZ,u=o
T(u)W
uZ
, then
T(u)W
uZ
M and this implies
|T(u)|
W
M|u|
Z
for all u ,= o. Since this is obviously true for u = o, we have
that |T| M and this nishes the proof.
Z
) and (W, | |
W
bounded linear transformation T : Z W.
If T is onto W and |T(u)|
W
= |u|
Z
for every u Z, then we say that T
is an isometry from Z onto W or an isometry between Z and W.
If |T(u)|
W
= |u|
Z
for every u Z, we say that T is an isometry from
Z into W.
Z
) and (W, | |
W
) be two normed spaces.
(i) If T is an isometry from Z into W, then T is one-to-one.
(ii) If T is an isometry from Z onto W, then T
1
is also an isometry from W
onto Z.
Proof: (i) If T(u) = T(v), then 0 = |T(u) T(v)|
W
= |T(uv)|
W
= |uv|
Z
and, hence, u = v.
(ii) From (i) we have that T is one-to-one and, thus, the inverse mapping T
1
:
W Z exists. If w, w
1
, w
2
W and F, we take the (unique) u, u
1
, u
2
Z
so that T(u) = w, T(u
1
) = w
1
and T(u
2
) = w
2
. Then T(u
1
+ u
2
) = T(u
1
) +
T(u
2
) = w
1
+ w
2
and, hence, T
1
(w
1
+ w
2
) = u
1
+ u
2
= T
1
(w
1
) + T
1
(w
2
).
Also, T(u) = T(u) = w and, hence, T
1
(w) = u = T
1
(w). These
imply that T
1
: W Z is a linear transformation.
Moreover, |T
1
(w)|
Z
= |u|
Z
= |T(u)|
W
= |w|
W
. Therefore, T
1
is an
isometry from W onto Z.
11.2 The spaces L
p
(X, , ).
In this whole section and the next, (X, , ) will be a xed measure space.
Denition 11.11 If 0 < p < +, we dene the space L
p
r
(X, , ) to be the
set of all measurable functions f : X R with
_
X
[f[
p
d < +.
The space L
p
c
(X, , ) is the set of all measurable f : X C under the
same niteness condition.
If R or C are understood, we write L
p
(X, , ) for either L
p
r
(X, , ) or
L
p
c
(X, , ).
Observe that the space L
1
(X, , ) is the set of all functions which are
integrable over X with respect to .
Proposition 11.10 The spaces L
p
r
(X, , ) are linear spaces over R and the
spaces L
p
c
(X, , ) are linear spaces over C.
Proof: We shall use the trivial inequality
(a +b)
p
2
p
(a
p
+b
p
), 0 a, b.
11.2. THE SPACES L
P
(X, , ). 233
This can be proved by (a +b)
p
[2 max(a, b)]
p
= 2
p
max(a
p
, b
p
) 2
p
(a
p
+b
p
).
Suppose that f
1
, f
2
L
p
(X, , ). Then both f
1
and f
2
are nite -a.e. on
X and, hence, f
1
+ f
2
is dened -a.e. on X. If f
1
+ f
2
is any measurable
denition of f
1
+f
2
, then, using the above elementary inequality, [(f
1
+f
2
)(x)[
p
2
p
([f
1
(x)[
p
+[f
2
(x)[
p
) for -a.e. x X and, hence,
_
X
[f
1
+f
2
[
p
d 2
p
_
X
[f
1
[
p
d + 2
p
_
X
[f
2
[
p
d < +.
Therefore f
1
+f
2
L
p
(X, , ).
If f L
p
r
(X, , ) and R or if f L
p
c
(X, , ) and C, then
_
X
[f[
p
d = [[
p
_
X
[f[
p
d < +.
Therefore, f L
p
(X, , ).
Denition 11.12 Let f : X R or C be measurable. We say that f is
essentially bounded over X with respect to if there is M < + so that
[f[ M -a.e. on X.
Proposition 11.11 Let f : X R or C be measurable. If f is essentially
bounded over X with respect to , then there is a smallest M with the property:
[f[ M -a.e. on X. This smallest M
0
is characterized by:
(i) [f[ M
0
-a.e. on X,
(ii) (x X[ [f(x)[ > m) > 0 for every m < M
0
.
Proof: We consider the set A = M[ [f[ M a.e. on X and the
M
0
= infM[ [f[ M a.e. on X.
The set A is non-empty by assumption and is included in [0, +) and, hence,
M
0
exists.
We take a sequence M
n
in A with M
n
M
0
. From M
n
A, we nd
(x X[ [f(x)[ > M
n
) = 0 for every n and, since x X [ [f(x)[ > M
0
=
+
n=1
x X [ [f(x)[ > M
n
, we conclude that (x X[ [f(x)[ > M
0
) = 0.
Therefore, [f[ M
0
-a.e. on X.
If m < M
0
, then m / A and, hence, (x X[ [f(x)[ > m) > 0.
Denition 11.13 Let f : X R or C be measurable. If f is essentially
bounded, then the smallest M with the property that [f[ M -a.e. on X is
called the essential supremum of f over X with respect to and it is
denoted by ess-sup
X,
(f).
The ess-sup
X,
(f) is characterized by the properties:
1. [f[ ess-sup
X,
(f) -a.e. on X,
2. for every m <ess-sup
X,
(f), we have (x X[ [f(x)[ > m) > 0.
Denition 11.14 We dene L
r
(X, , ) to be the set of all measurable
functions f : X R which are essentially bounded over X with respect to .
The space L
c
(X, , ) is the set of all measurable f : X C which are
essentially bounded over X with respect to .
If R or C are understood, we write L
(X, , ) for either L
r
(X, , ) or
L
c
(X, , ).
Proposition 11.12 The space L
r
(X, , ) is a linear space over R and the
space L
c
(X, , ) is a linear space over C.
Proof: If f
1
, f
2
L
(X, , ), then there are sets A

1
, A
2
so that (A
c
1
) =
(A
c
2
) = 0 and [f
1
[ ess-sup
X,
(f
1
) on A
1
and [f
2
[ ess-sup
X,
(f
2
) on A
2
.
If we set A = A
1
A
2
, then we have (A
c
) = 0 and [f
1
+ f
2
[ [f
1
[ + [f
2
[
ess-sup
X,
(f
1
)+ess-sup
X,
(f
2
) on A. Hence f
1
+f
2
is essentially bounded over
X with respect to and
ess-sup
X,
(f
1
+ f
2
) ess-sup
X,
(f
1
) + ess-sup
X,
(f
2
).
If f L
r
(X, , ) and R or f L
c
(X, , ) and C, then there
is A with (A
c
) = 0 so that [f[ ess-sup
X,
(f) on A. We, now, have
[f[ [[ess-sup
X,
(f) on A. Hence f is essentially bounded over X with
respect to and ess-sup
X,
(f) [[ess-sup
X,
(f). If = 0, this inequality,
obviously, becomes equality. If ,= 0, we apply the same inequality to
1
and
f and get ess-sup
X,
(f) = ess-sup
X,
(
1
(f))
1
||
ess-sup
X,
(f). Therefore
ess-sup
X,
(f) = [[ess-sup
X,
(f).
Denition 11.15 Let 1 p +. We dene
p
=
_
_
_
p
p1
, if 1 < p < +
+, if p = 1
1 , if p = +.
We say that p
is the conjugate of p or the dual of p.

The denition in the cases p = 1 and p = + is justied by lim
p1+
p
p1
=
+ and by lim
p+
p
p1
= 1.
It is easy to see that, if p
is the conjugate of p, then 1 p
+ and p is
the conjugate of p
. Moreover, p, p
are related by the symmetric equality

1
p
+
1
p
= 1.
Lemma 11.1 Let 0 < t < 1. For every a, b 0 we have
a
t
b
1t
ta + (1 t)b.
11.2. THE SPACES L
P
(X, , ). 235
Proof: If b = 0 the inequality is obviously true: 0 ta.
If b > 0, the inequality is equivalent to (
a
b
)
t
t
a
b
+1 t and, setting x =
a
b
,
it is equivalent to x
t
tx + 1 t, 0 x. To prove it we form the function
f(x) = x
t
tx on [0, +) and we easily see that it is increasing in [0, 1] and
decreasing in [1, +). Therefore, f(x) f(1) = 1 t for all x [0, +).
Theorem 11.4 (H olders inequalities) Let 1 p, p
+ and p, p
be
conjugate to each other. If f L
p
(X, , ) and g L
p
(X, , ), then fg
L
1
(X, , ) and
_
X
[fg[ d
_
_
X
[f[
p
d
_1
p
_
_
X
[g[
p
d
_ 1
p
, 1 < p, p
< +,
_
X
[fg[ d
_
X
[f[ d ess-sup
X,
(g) , p = 1, p
= +,
_
X
[fg[ d ess-sup
X,
(f)
_
X
[g[ d, p = +, p
= 1.
Proof: (a) We start with the case 1 < p, p
< +.
If
_
X
[f[
p
d = 0 or if
_
X
[g[
p
d = 0, then either f = 0 -a.e. on X or g = 0

-a.e. on X and the inequality is trivially true. It becomes equality: 0 = 0.
So we assume that A =
_
X
[f[
p
d > 0 and B =
_
X
[g[
p
d > 0. Applying
Lemma 11.1 with t =
1
p
, 1 t = 1
1
p
=
1
p
and a =
|f(x)|
p
A
, b =
|g(x)|
p
B
, we have
that
[fg[
A
1
p
B
1
p
1
p
[f[
p
A
+
1
p
[g[
p
B
-a.e. on X. Integrating, we nd
1
A
1
p
B
1
p
_
X
[fg[ d
1
p
+
1
p
= 1
and this implies the inequality we wanted to prove.
(b) Now, let p = 1, p
= +. Since [g[ ess-sup

X,
(g) -a.e. on X, we have
that [fg[ [f[ ess-sup
X,
(g) -a.e. on X. Integrating, we nd the inequality
we want to prove.
(c) The proof in the case p = +, p
= 1 is the same as in (b).

Theorem 11.5 (Minkowskis inequalities) Let 1 p +. If f
1
, f
2

L
p
(X, , ), then
_
_
X
[f
1
+f
2
[
p
d
_1
p
_
_
X
[f
1
[
p
d
_1
p
+
_
_
X
[f
2
[
p
d
_1
p
, 1 p < +,
ess-sup
X,
(f
1
+f
2
) ess-sup
X,
(f
1
) + ess-sup
X,
(f
2
) , p = +.
Proof: The case p = + is included in the proof of Proposition 11.12. Also,
the case p = 1 is trivial and the result is already known. Hence, we assume
1 < p < +.
We write
[f
1
+f
2
[
p
([f
1
[ +[f
2
[)[f
1
+ f
2
[
p1
= [f
1
[[f
1
+f
2
[
p1
+[f
2
[[f
1
+f
2
[
p1
-a.e. on X and, applying H olders inequality, we nd
_
X
[f
1
+f
2
[
p
d
_
_
X
[f
1
[
p
d
_1
p
_
_
X
[f
1
+f
2
[
(p1)p
d
_ 1
p
+
_
_
X
[f
2
[
p
d
_1
p
_
_
X
[f
1
+f
2
[
(p1)p
d
_ 1
p
=
_
_
X
[f
1
[
p
d
_1
p
_
_
X
[f
1
+f
2
[
p
d
_ 1
p
+
_
_
X
[f
2
[
p
d
_1
p
_
_
X
[f
1
+f
2
[
p
d
_ 1
p
.
Simplifying, we get the inequality we want to prove.
Denition 11.16 Let f
n
be a sequence in L
p
(X, , ) and f L
p
(X, , ).
We say that f
n
converges to f in the p-mean if
_
X
[f
n
f[
p
d 0, 1 p < +,
ess-sup
X,
(f
n
f) 0, p = +
as n +. We say that f
n
is Cauchy in the p-mean if
_
X
[f
n
f
m
[
p
d 0, 1 p < +,
ess-sup
X,
(f
n
f
m
) 0, p = +
as n, m +.
The notion of convergence in the 1-mean coincides with the notion of con-
vergence in the mean on X. The following result is an extension of the result of
Theorem 9.1.
Theorem 11.6 If f
n
is Cauchy in the p-mean, then there is f L
p
(X, , )
so that f
n
converges to f in the p-mean. Moreover, there is a subsequence
f
n
k
which converges to f -a.e. on X.
n
converges to f in the p-mean, there is a subsequence
f
n
k
which converges to f -a.e. on X.
11.2. THE SPACES L
P
(X, , ). 237
Proof: (a) We consider rst the case 1 p < +.
First proof. Since each f
n
is nite -a.e. on X, there is A so that (A
c
) = 0
and all f
n
are nite on A.
We have that, for every k, there is n
k
so that
_
X
[f
n
f
m
[
p
d <
1
2
kp
for
every n, m n
k
k
is as large as we like,
we inductively take n
k
so that n
k
< n
k+1
for every k. Therefore, f
n
k
is a
subsequence of f
n
.
From the construction of n
k
and from n
k
< n
k+1
, we get that
_
X
[f
n
k+1
f
n
k
[
p
d <
1
2
kp
for every k. We dene the measurable function G : X [0, +] by
G =
_
+
k=1
[f
n
k+1
f
n
k
[, on A
0, on A
c
.
If
G
K
=
_
K1
k=1
[f
n
k+1
f
n
k
[, on A
0, on A
c
,
then
_ _
X
G
p
K
d
_ 1
p

K1
k=1
_ _
X
[f
n
k+1
f
n
k
[
p
d
_ 1
p
< 1, by Minkowskis in-
equality. Since G
K
G on X, we nd that
_
X
G
p
d 1 and, thus, G < +
-a.e. on X. This implies that the series
+
k=1
(f
n
k+1
(x) f
n
k
(x)) converges
for -a.e. x A. Therefore, there is a B , B A so that (A B) = 0
and
+
k=1
(f
n
k+1
(x) f
n
k
(x)) converges for every x B. We dene the
measurable f : X C by
f =
_
f
n1
+
+
k=1
(f
n
k+1
f
n
k
), on B
0, on B
c
.
On B we have that f = f
n1
+lim
K+
K1
k=1
(f
n
k+1
f
n
k
) = lim
K+
f
nK
and, hence, f
n
k
converges to f -a.e. on X.
We, also, have on B that [f
nK
f[ = [f
nK
f
n1

+
k=1
(f
n
k+1
f
n
k
)[ =
[
K1
k=1
(f
n
k+1
f
n
k
)
+
k=1
(f
n
k+1
f
n
k
)[
+
k=K
[f
n
k+1
f
n
k
[ G for every
K and, hence, [f
nK
f[
p
G
p
-a.e. on X for every K. Since we have
_
X
G
p
d < + and that [f
nK
f[ 0 -a.e. on X, we apply the Dominated
Convergence Theorem and we nd that
_
X
[f
nK
f[
p
d 0
as K +.
From n
k
+, we get
_ _
X
[f
k
f[
p
d)
1
p
_ _
X
[f
k
f
n
k
[
p
d
_1
p
+
_ _
X
[f
n
k
f[
p
d
_1
p
0 as k + and we conclude that f
n
converges
to f in the p-mean.
Second proof. For every > 0 we have that (x X[ [f
n
(x) f
m
(x)[ )
1
_ _
X
[f
n
f
m
[
p
d
_ 1
p
and, hence, f
n
is Cauchy in measure on X. Theorem
9.2 implies that there is a subsequence f
n
k
which converges to some f -a.e.
on X.
Now, for every > 0 there is an N so that
_
X
[f
n
f
m
[
p
d for all
n, m N. Since n
k
+ as k +, we use m = n
k
for large k and apply
the Lemma of Fatou to get
_
X
[f
n
f[
p
d liminf
k+
_
X
[f
n
f
n
k
[
p
d
for all n N. This, of course, says that f
n
converges to f in the p-mean.
(b) Now, let p = +.
For each n, m we have a set A
n,m
with (A
c
n,m
) = 0 and [f
n
f
m
[
ess-sup
X,
(f
n
f
m
) on A
n,m
. We form the set A =
1n,m
A
n,m
and have that
(A
c
) = 0 and [f
n
f
m
[ ess-sup
X,
(f
n
f
m
) on A for every n, m. This
says that f
n
is Cauchy uniformly on A and, hence, there is an f so that f
n
converges to f uniformly on A. Now,

ess-sup
X,
(f
n
f) sup
xA
[f
n
(x) f(x)[ 0
as n +.
If, for every f L
p
(X, , ), we set
N
p
(f) =
_
_ _
X
[f[
p
d
_ 1
p
, if 1 p < +
ess-sup
X,
(f) , if p = +,
then, Propositions 11.10 and 11.12 and Theorem 11.5 imply that the function
N
p
: L
p
(X, , ) R satises
1. N
p
(f
1
+f
2
) N
p
(f
1
) +N
p
(f
2
),
2. N
p
(f) = [[N
p
(f)
for every f, f
1
, f
2
L
p
r
(X, , ) or L
p
c
(X, , ) and R or C, respectively.
The function N
p
has the two properties of a norm but not the third. Indeed,
N
p
(f) = 0 if and only if f = 0 -a.e. on X. The usual practice is to identify
every two functions which are equal -a.e. on X so that N
p
becomes, informally,
a norm. The precise way to do this is the following.
Denition 11.17 We dene the relation on L
p
(X, , ) as follows: we write
f
1
f
2
if f
1
= f
2
-a.e. on X.
Proposition 11.13 The relation on L
p
(X, , ) is an equivalence relation.
Proof: It is obvious that f f and that, if f
1
f
2
, then f
2
f
1
. Now, if
f
1
f
2
and f
2
f
3
, then there are A, B with (A
c
) = (B
c
) = 0 so that
f
1
= f
2
on A and f
2
= f
3
on B. This implies that ((AB)
c
) = 0 and f
1
= f
3
on A B and, hence, f
1
f
3
.
11.2. THE SPACES L
P
(X, , ). 239
As with any equivalence relation, the relation denes equivalence classes.
The equivalence class [f] of any f L
p
(X, , ) is the set of all f
L
p
(X, , )
which are equivalent to f:
[f] = f
L
p
(X, , ) [ f
f.
Proposition 11.14 Let f
1
, f
2
L
p
(X, , ). Then
(i) [f
1
] = [f
2
] if and only if f
1
f
2
if and only if f
1
= f
2
-a.e. on X.
(ii) If [f
1
] [f
2
] ,= , then [f
1
] = [f
2
].
Moreover, L
p
(X, , ) =
fL
p
(X,,)
[f].
Proof: (i) Assume f
1
f
2
. If f [f
1
], then f f
1
. Therefore, f f
2
and,
hence, f [f
2
]. Symmetrically, if f [f
2
], then f [f
1
] and, thus, [f
1
] = [f
2
].
If [f
1
] = [f
2
], then f
1
[f
1
] and, hence, f
1
[f
2
]. Therefore, f
1
f
2
.
(ii) If f [f
1
] and f [f
2
], then f f
1
and f f
2
and, hence, f
1
f
2
. This,
by the result of (i), implies [f
1
] = [f
2
].
For the last statement, we observe that every f L
p
(X, , ) belongs to [f].
Proposition 11.14 says that any two dierent equivalence classes have empty
intersection and that L
p
(X, , ) is the union of all equivalence classes. In other
words, the collection of all equivalence classes is a partition of L
p
(X, , ).
Denition 11.18 We dene
L
p
(X, , ) = L
p
(X, , )/
= [f] [ f L
p
(X, , ).
The rst task is to carry addition and multiplication from L
p
(X, , ) over
to L
p
(X, , ).
1
, f
2
, f
, f
1
, f
2
L
p
(X, , ).
(i) If f
1
f
1
and f
2
f
2
, then f
1
+f
2
f
1
+f
2
.
(ii) If f f
, then f f
.
Proof: (i) There are A
1
, A
2
with (A
c
1
) = (A
c
2
) = 0 so that f
1
= f
1
on
A
1
and both f
1
, f
1
are nite on A
1
and, also, f
2
= f
2
on A
2
and both f
2
, f
2
are
nite on A
2
. Then ((A
1
A
2
)
c
) = 0 and f
1
+f
2
= f
1
+f
2
on A
1
A
2
. Hence,
f
1
+f
2
f
1
+f
2
.
(ii) There is A with (A
c
) = 0 so that f = f
on A. Then, f = f
on A
and, hence f f
.
Because of Proposition 11.14, another way to state the results of Proposition
11.15 is:
1. [f
1
] = [f
1
] and [f
2
] = [f
2
] imply [f
1
+f
1
] = [f
2
+f
2
],
2. [f] = [f
] implies [f] = [f
].
These allow the following denition.
Denition 11.19 We dene addition and multiplication in L
p
(X, , ) as fol-
lows:
[f
1
] + [f
2
] = [f
1
+f
2
], [f] = [f].
It is a matter of routine to prove, now, that the set L
p
(X, , ) becomes
a linear space under this addition and multiplication. If we want to be more
precise, we denote this space L
p
r
(X, , ), if it comes from L
p
r
(X, , ), and we
denote it L
p
c
(X, , ), if it comes from L
p
c
(X, , ). Then L
p
r
(X, , ) is a linear
space over R and L
p
c
(X, , ) is a linear space over C.
The zero element of L
p
(X, , ) is the equivalence class [o] of the function o
which is identically 0 on X. The opposite of an [f] is the equivalence class [f].
The next task is to dene a norm on L
p
(X, , ).
Proposition 11.16 Let f
1
, f
2
L
p
(X, , ). If f
1
f
2
, then N
p
(f
1
) = N
p
(f
2
)
or equivalently
_
X
[f
1
[
p
d =
_
X
[f
2
[
p
d, 1 p < +,
ess-sup
X,
(f
1
) = ess-sup
X,
(f
2
) , p = +.
Proof: It is well known that f
1
= f
2
-a.e. on X implies the rst equal-
ity. Regarding the second equality, we have sets B, A
1
, A
2
with (B
c
) =
(A
c
1
) = (A
c
2
) = 0 so that f
1
= f
2
on B, [f
1
[ ess-sup
X,
(f
1
) on A
1
and
[f
2
[ ess-sup
X,
(f
2
) on A
2
. Then, the set A = B A
1
A
2
has (A
c
) = 0.
Moreover, [f
1
[ = [f
2
[ ess-sup
X,
(f
2
) on A and, hence, ess-sup
X,
(f
1
) ess-
sup
X,
(f
2
). Also, [f
2
[ = [f
1
[ ess-sup
X,
(f
1
) on A and, hence, ess-sup
X,
(f
2
)
ess-sup
X,
(f
1
).
An equivalent way to state the result of Proposition 11.16 is
1. [f
1
] = [f
2
] implies
_
X
[f
1
[
p
d =
_
X
[f
2
[
p
d, if 1 p < +,
2. [f
1
] = [f
2
] implies ess-sup
X,
(f
1
) = ess-sup
X,
(f
2
), if p = +.
These allow the
Denition 11.20 We dene, for every [f] L
p
(X, , ),
|[f]|
p
= N
p
(f) =
_
_ _
X
[f[
p
d
_1
p
, if 1 p < +
ess-sup
X,
(f) , if p = +.
Proposition 11.17 The function | |
p
is a norm on L
p
(X, , ).
Proof: We have |[f
1
] +[f
2
]|
p
= |[f
1
+f
2
]|
p
= N
p
(f
1
+f
2
) N
p
(f
1
) +N
p
(f
2
) =
|[f
1
]|
p
+|[f
2
]|
p
. Also |[f]|
p
= |[f]|
p
= N
p
(f) = [[N
p
(f) = [[|[f]|
p
.
If |[f]|
p
= 0, then N
p
(f) = 0. This implies f = 0 -a.e. on X and, hence,
f o or, equivalently, [f] is the zero element of L
p
(X, , ).
11.2. THE SPACES L
P
(X, , ). 241
In order to simplify things and not have to carry the bracket-notation [f] for
the elements of L
p
(X, , ), we shall follow the traditional practice and write
f instead of [f]. When we do this we must have in mind that the element f of
L
p
(X, , ) (and not the element f of L
p
(X, , )) is not the single function f,
but the whole collection of functions each of which is equal to f -a.e. on X.
For example:
1. when we write f
1
= f
2
for the elements f
1
, f
2
of L
p
(X, , ), we mean the
more correct [f
1
] = [f
2
] or, equivalently, that f
1
= f
2
-a.e. on X,
2. when we write
_
X
fg d for the element f L
p
(X, , ), we mean the integral
_
X
fg d for the element-function f L
p
(X, , ) and, at the same time, all
integrals
_
X
f
g d (equal to each other) for all functions f
L
p
(X, , ) such
that f
= f -a.e. on X,
3. when we write |f|
p
for the element f L
p
(X, , ) we mean the more correct
|[f]|
p
or, equivalently, the expression
_ _
X
[f[
p
d
_ 1
p
, when 1 p < +, and
ess-sup
X,
(f), when p = +, for the element-function f L
p
(X, , ) and
at the same time all similar expressions (equal to each other) for all functions
f
L
p
(X, , ) such that f
= f -a.e. on X.
The inequality of Minkowski takes the form
|f
1
+f
2
|
p
|f
1
|
p
+|f
2
|
p
for every f
1
, f
2
L
p
(X, , ).
H olders inequality takes the form
|fg|
1
|f|
p
|g|
p
for every f L
p
(X, , ) and g L
p
(X, , ).
Theorem 11.7 All L
p
(X, , ) are Banach spaces.
Proof: Let f
n
be a Cauchy sequence in L
p
(X, , ). Then |f
n
f
m
|
p
0 and,
hence,
_
X
[f
n
f
m
[
p
d 0, if 1 p < +, and ess-sup
X,
(f
n
f
m
) 0, if
p = +. Theorem 11.6 implies that the sequence f
n
in L
p
(X, , ) converges
to some f L
p
(X, , ) in the p-mean. Therefore,
_
X
[f
n
f[
p
d 0, if
1 p < +, and ess-sup
X,
(f
n
f) 0, if p = +. This means that
|f
n
f|
p
0 and f
n
converges to the element f of L
p
(X, , ).
Denition 11.21 Let I be a non-empty index set and be the counting measure
on (I, T(I)). We denote
l
p
(I) = L
p
(I, T(I), ).
In particular, if I = N, we denote l
p
= l
p
(N).
If 1 p < +, then, the function b = b
i
iI
: I R or C belongs to l
p
(I)
if, by denition,
_
I
[b[
p
d < + or, equivalently,
iI
[b
i
[
p
< +.
If [b
i
[ = + for at least one i I, then
iI
[b
i
[
p
= +.
Denition 11.22 Let I be an index set and b : I R or C. If 1 p < +,
we say that b = b
i
iI
is p-summable if
iI
[b
i
[
p
< +.
Hence, b = b
i
iI
is p-summable if and only if it belongs to l
p
(I). We also
have
|b|
p
=
_
iI
[b
i
[
p
_1
p
.
When 1 p < +, Minkowskis inequality becomes
_
iI
[b
1
i
+b
2
i
[
p
_ 1
p
iI
[b
1
i
[
p
_1
p
+
_
iI
[b
2
i
[
p
_1
p
for all b
1
= b
1
i
iI
and b
2
= b
2
i
iI
which are p-summable. Similarly, when
1 < p, p
< +, H olders inequality becomes
iI
[b
i
c
i
[
_
iI
[b
i
[
p
_ 1
p
_
iI
[c
i
[
p
_ 1
p
for all p-summable b = b

i
iI
and all p
-summable c = c
i
iI
.
Since the only subset of I with zero -measure is the , we easily see that
b = b
i
iI
is essentially bounded over I with respect to if and only if there
is an M < + so that [b
i
[ M for all i I. It is obvious that the smallest M
with the property that [b
i
[ M for all i I is the M
0
= sup
iI
[b
i
[.
Denition 11.23 Let I be an index set and b : I R or C. We say that
b = b
i
iI
is bounded if sup
iI
[b
i
[ < +.
Therefore, b is essentially bounded over I with respect to or, equivalently,
b l
(I) if and only if b is bounded. Also,

|b|
= ess-sup
I,
(b) = sup
iI
[b
i
[.
The inequality of Minkowski takes the form
sup
iI
[b
1
i
+b
2
i
[ sup
iI
[b
1
i
[ + sup
iI
[b
2
i
[
for all b
1
= b
1
i
iI
and b
2
= b
2
i
iI
which are bounded. When p = 1 and
p
= +, H olders inequality takes the form
iI
[b
i
c
i
[
iI
[b
i
[ sup
iI
[c
i
[
for all summable b = b
i
iI
and all bounded c = c
i
iI
.
The spaces l
p
(I) are all Banach spaces.
As we have already mentioned, a particular case is when I = N. Then
l
p
=
_
x = (x
1
, x
2
, . . .) [
+
k=1
[x
k
[
p
< +
_
, 1 p < +,
11.3. THE DUAL OF L
P
(X, , ). 243
l
=
_
x = (x
1
, x
2
, . . .) [ sup
k1
[x
k
[ < +
_
, p = +.
The corresponding norms are
|x|
p
=
_
+
k=1
[x
k
[
p
_1
p
, 1 p < +,
|x|
= sup
k1
[x
k
[, p = +,
for every x = (x
1
, x
2
, . . .) l
p
.
Another very special case is when I = 1, . . . , n. In this case we have
l
p
r
(I) = R
n
and l
p
c
(I) = C
n
. The norms are
|x|
p
=
_
n
k=1
[x
k
[
p
_1
p
, 1 p < +,
|x|
= max
1kn
[x
k
[, p = +,
for every x = (x
1
, . . . , x
n
) R
n
or C
n
.
11.3 The dual of L
p
(X, , ).
Theorem 11.8 Let g L
p
(X, , ). If 1 < p +, then

|g|
p
= sup
_
_
X
fg d
[ f L
p
(X, , ), |f|
p
1
_
.
If is seminite, the same is true when p = 1.
Proof: (a) Let 1 < p + and, hence, 1 p
< +.
For any f L
p
(X, , ) with |f|
p
1, we have, by H olders inequality,
that [
_
X
fg d[ |f|
p
|g|
p
|g|
p
. Therefore,
sup
_
_
X
fg d
[ f L
p
(X, , ), |f|
p
1
_
|g|
p
.
If |g|
p
= 0, then the inequality between the sup and the |g|
p
, obviously,
becomes equality. Anyway, we have
_
X
[g[
p
d = 0 and, hence, g = 0 -a.e. on

X. This implies that
_
X
fg d = 0 for every f L
p
(X, , ).
Now, let |g|
p
> 0. We consider the function f
0
dened by
f
0
(x) =
_
_
_
|g(x)|
p
1
sign(g(x))
g
p
1
p
, if g(x) is nite and g(x) ,= 0,

0 , if g(x) is innite or g(x) = 0.
Then,
f
0
(x)g(x) =
_
_
_
|g(x)|
p
g
p
1
p
, if g(x) is nite,
0 , if g(x) is innite
and, hence,
_
X
f
0
g d =
1
g
p
1
p
_
X
[g[
p
d = |g|
p
.
If 1 < p, p
< +, then, since p(p
1) = p
,
[f
0
(x)[
p
=
_
_
_
|g(x)|
p
g
p
, if g(x) is nite,
and, hence, |f
0
|
p
=
_ _
X
[f
0
[
p
d
_ 1
p
= 1.
If p = +, p
= 1, then
[f
0
(x)[ =
_
1 , if g(x) is nite and ,= 0,
0 , if g(x) is innite or = 0
and, hence, |f
0
|
= ess-sup
X,
(f
0
) = 1.
We conclude that
|g|
p
= max
_
_
X
fg d
[ f L
p
(X, , ), |f|
p
1
_
.
(b) Let p = 1, p
= +.
For any f L
1
(X, , ) with |f|
1
1, we have [
_
X
fg d[ |f|
1
|g|

|g|
. Therefore,
sup
_
_
X
fg d
[ f L
1
(X, , ), |f|
1
1
_
|g|
.
If |g|
= 0, then g = 0 -a.e. on X. This implies that

_
X
fg d = 0 for
every f L
p
(X, , ) and, thus, the inequality between the sup and the |g|
becomes equality.
Let |g|
> 0. We consider an arbitrary with 0 < < |g|
and, then
(x X [ |g|
< [g(x)[ |g|
) > 0. If is seminite, there exists a

B so that B x X[ |g|
< [g(x)[ |g|
and 0 < (B) < +.

We dene the function f
0
by
f
0
(x) =
_
sign(g(x))B(x)
(B)
, if g(x) is nite,
0 , if g(x) is innite.
Then,
f
0
(x)g(x) =
_
|g(x)|B(x)
(B)
, if g(x) is nite,
and, hence,
_
X
f
0
g d =
1
(B)
_
B
[g[ d |g|
.
11.3. THE DUAL OF L
P
(X, , ). 245
Also,
[f
0
(x)[ =
_
B(x)
(B)
, if g(x) is nite,
and, hence, |f
0
|
1
=
_
X
[f
0
[ d =
1
(B)
_
B
d = 1.
These imply
sup
_
_
X
fg d
[ f L
1
(X, , ), |f|
1
1
_
|g|
for every with 0 < < |g|
and, taking the limit as 0+, we conclude

that
|g|
= sup
_
_
X
fg d
[ f L
1
(X, , ), |f|
1
1
_
.
Denition 11.24 Let 1 p +. For every g L
p
r
(X, , ) or L
p
c
(X, , )
we dene l
g
: L
p
r
(X, , ) R or, respectively, l
g
: L
p
c
(X, , ) C by
l
g
(f) =
_
X
fg d, f L
p
(X, , ).
Proposition 11.18 Let 1 p +. For every g L
p
(X, , ), the function

l
g
of Denition 11.24 belongs to (L
p
(X, , ))
.
Moreover, if 1 < p +, then |l
g
|
= |g|
p
and, if p = 1, then |l
g
|

|g|
. If p = 1 and is seminite, then |l

g
|
= |g|
.
Proof: We have l
g
(f
1
+ f
2
) =
_
X
(f
1
+ f
2
)g d =
_
X
f
1
g d +
_
X
f
2
g d =
l
g
(f
1
) +l
g
(f
2
). Also, l
g
(f) =
_
X
(f)g d =
_
X
fg d = l
g
(f). These imply
that l
g
is a linear functional.
Theorem 11.8 together with Proposition 11.5 imply that, if 1 < p +,
then |l
g
|
= |g|
p
. If is seminite, the same is true, also, for p = 1.
If p = 1, for every f L
1
(X, , ) we have [l
g
(f)[ =
_
X
fg d
|g|
|f|
1
.
Therefore, |l
g
|
|g|
.
Denition 11.25 Let 1 p +. We dene the mapping J : L
p
(X, , )
(L
p
(X, , ))
by
J(g) = l
g
for all g L
p
(X, , ).
Proposition 11.19 The function J of Denition 11.21 is a bounded linear
transformation. If 1 < p +, J is an isometry from L
p
(X, , ) into
(L
p
(X, , ))
. This is true, also, when p = 1, if is seminite.

Proof: For every f L
p
(X, , ) we have l
g1+g2
(f) =
_
X
f(g
1
+ g
2
) d =
_
X
fg
1
d+
_
X
fg
2
d = l
g1
(f) +l
g2
(f) = (l
g1
+l
g2
)(f) and, hence, J(g
1
+g
2
) =
l
g1+g2
= l
g1
+l
g2
= J(g
1
) +J(g
2
).
Moreover, l
g
(f) =
_
X
f(g) d =
_
X
fg d = l
g
(f) = (l
g
)(f) and,
hence, J(g) = l
g
= l
g
= J(g).
Now, |J(g)|
= |l
g
|
|g|
p
and J is bounded. That J is an isometry is
a consequence of Proposition 11.18.
Lemma 11.2 Let l (L
p
(X, , ))
. If E ,
E
= A [ A E is the
restriction of on E and is the restricted measure on (E,
E
), we dene l
E
by
l
E
(h) = l(
h), h L
p
(E,
E
, ),
where

h is the extension of h as 0 on X E.
Then, l
E
(L
p
(E,
E
, ))
and |l
E
| |l|. Moreover,
l(f
E
) = l
E
(f
E
), f L
p
(X, , ),
where f
E
is the restriction of f on E.
Proof: For all h, h
1
, h
2
L
p
(E,
E
, ) we consider the corresponding extensions
h,
h
1
,
h
2
L
p
(X, , ). Since

h
1
+
h
2
and
h are the extensions of h

1
+ h
2
and h, respectively, we have l
E
(h
1
+ h
2
) = l(
h
1
+

h
2
) = l(
h
1
) + l(
h
2
) =
l
E
(h
1
) + l
E
(h
2
) and l
E
(h) = l(
h) = l(
h) = l
E
(h). This proves that l
E
is
linear and [l
E
(h)[ = [l(
h)[ |l||
h|
p
= |l||h|
p
proves that l
E
is bounded and
that |l
E
| |l|.
If f L
p
(X, , ), then

f
E
= f
E
on X and, hence, l
E
(f
E
) = l(
f
E
) =
l(f
E
).
Denition 11.26 The l
E
dened in Lemma 11.2 is called the restriction of
l (L
p
(X, , ))
on L
p
(E,
E
, ).
Theorem 11.9 If 1 < p < +, the function J of Denition 11.21 is an isom-
etry from L
p
(X, , ) onto (L
p
(X, , ))
. If is -nite, then this is true

also when p = 1.
Proof: A. We consider rst the case when is a nite measure: (X) < +.
Let l (L
p
(X, , ))
and 1 p < +.
Since
_
A
[
A
[
p
d = (A) < +, we have that
A
L
p
(X, , ) for every
A . We dene the function : R, if l (L
p
r
(X, , ))
, or : C,
if l (L
p
c
(X, , ))
, by
(A) = l(
A
), A .
We have () = l(
) = l(o) = 0. If A
1
, A
2
, . . . are pairwise disjoint
and A =
+
j=1
A
j
, then
A
=
+
j=1

Aj
. Therefore, |
n
j=1
Aj

A
|
p
p
=
_
X
[
+
j=n+1
Aj
[
p
d =
_
X
[
+
j=n+1
Aj
[
p
d = (
+
j=n+1
A
j
) () = 0, by the
continuity of from above. The linearity and the continuity of l imply, now, that
n
j=1
(A
j
) =
n
j=1
l(
Aj
) = l(
n
j=1
Aj
) l(
A
) = (A) or, equivalently,
that
+
j=1
(A
j
) = (A).
Hence, is a real measure, if l (L
p
r
(X, , ))
, or a complex measure on
(X, ), if l (L
p
c
(X, , ))
.
We observe that, if A has (A) = 0, then (A) = l(
A
) = l(o) = 0,
because the function
A
is the zero element o of L
p
(X, , ). Therefore,
and, by Theorems 10.12 and 10.13, there exists a function g : X R or C (if
11.3. THE DUAL OF L
P
(X, , ). 247
is real or complex, respectively), which is integrable over X with respect to
, so that
l(
A
) = (A) =
_
A
g d =
_
X
A
g d
for every A . By the linearity of l and of the integral, this, clearly, implies
l() =
_
X
g d
for every measurable simple function on X.
This extends to all measurable functions which are bounded on X. In-
deed, let f L
p
(X, , ) be such that [f[ M on X for some M < +.
We take any sequence
n
of measurable simple functions with
n

f and [
n
[ [f[ on X. Then,
n
g fg and [
n
g[ [fg[ M[g[ on
X. Since
_
X
[g[ d < +, the Dominated Convergence Theorem implies that
_
X

n
g d
_
X
fg d. On the other hand, [
n
f[
p
0 on X and [
n
f[
p
([
n
[ +[f[)
p
2
p
[f[
p
on X. The Dominated Convergence Theorem, again, im-
plies that
_
X
[
n
f[
p
d 0 as n + and, hence,
n
f in L
p
(X, , ).
By the continuity of l, we get
_
X

n
g d = l(
n
) l(f) and, hence,
l(f) =
_
X
fg d
for every f L
p
(X, , ) which is bounded on X.
Our rst task, now, is to prove that g L
p
(X, , ).
If 1 < p, p
< +, we consider a sequence

n
of measurable non-
negative simple functions on X so that
n
[g[
p
1
on X. We dene
n
(x) =
_
n
(x)sign(g(x)), if g(x) is nite
0, if g(x) is innite.
Then, 0
n
g =
n
[g[ [g[
p
-a.e. on X and each

n
is bounded on X.
Hence, |
n
|
p
p
=
_
X

p
n
d
_
X

n
[g[ d =
_
X

n
g d = l(
n
) |l||
n
|
p

|l||
n
|
p
, where the last equality is justied by . This implies
_
X

p
n
d =
|
n
|
p
p
|l|
p
and, by the Monotone Convergence Theorem, we get

_
X
[g[
p
d =
lim
n+
_
X

p
n
d |l|
p
. Therefore, g L
p
(X, , ) and
|g|
p
|l|.
If p = 1 and p
= +, we consider any possible t > 0 such that the set

A = x X[ t < [g(x)[ has (A) > 0. We dene the function
f(x) =
_
A
(x)sign(g(x)), if g(x) is nite
0, if g(x) is innite.
Then t(A)
_
A
[g[ d =
_
X
fg d = l(f) |l||f|
1
|l|(A), where the last
equality is justied by . This implies that t |l| and, hence, [g[ |l| -a.e.
on X. Therefore, g is essentially bounded on X with respect to and
|g|
|l|.
We have proved that, in all cases, g L
p
(X, , ) and |g|

p
|l|.
Now, consider an arbitrary f L
p
(X, , ) and take a sequence
n
of
measurable simple functions on X so that
n
f and [
n
[ [f[ on X. We
have already shown, by the Dominated Convergence Theorem, that
n
f
in L
p
(X, , ) and, hence, l(
n
) l(f). Moreover, [
_
X

n
g d
_
X
fg d[
_
X
[
n
f[[g[ d |
n
f|
p
|g|
p
0, since |g|
p
< +. From l(
n
) =
_
X

n
g d, we conclude that
l(f) =
_
X
fg d, f L
p
(X, , ).
This implies, of course, that l(f) = l
g
(f) for every f L
p
(X, , ) and,
hence,
l = l
g
= J(g).
The uniqueness part of Proposition 11.19 implies that, if g
L
p
(X, , )
also satises l = l
g
, then g
= g -a.e. on X.
B. We suppose, now, that is -nite and consider an increasing sequence E
k
in so that E
k
X and (E
k
) < + for all k.
Let l (L
p
(X, , ))
.
For each k, we consider the restriction l
E
k
of l on L
p
(E
k
,
E
k
, ), which is
dened in Lemma 11.2. Since l
E
k
(L
p
(E
k
,
E
k
, ))
and |l
E
k
| |l| and
since (E
k
) < +, part A implies that there is a unique g
k
L
p
(E
k
,
E
k
, )
so that |g
k
|
p
|l
E
k
| |l| and
l(f
E
k
) = l
E
k
(f
E
k
) =
_
E
k
f
E
k
g
k
d
for every f L
p
(X, , ).
For an arbitrary h L
p
(E
k
,
E
k
, ) we consider its extension h
on E
k+1
as
0 on E
k+1
E
k
and, observing that
h =

h
on X, we have
_
E
k
hg
k
d = l
E
k
(h) =
l(
h) = l(
) = l
E
k+1
(h
) =
_
E
k+1
h
g
k+1
d =
_
E
k
hg
k+1
d. By the uniqueness
result of part A, we have that g
k+1
= g
k
-a.e. on E
k
. We may clearly suppose
that g
k+1
= g
k
on E
k
for every k, by inductively changing g
k+1
on a subset of
E
k
of zero -measure.
We dene the measurable function g on X as equal to g
k
on each E
k
.
Therefore, l(f
E
k
) =
_
E
k
f
E
k
g d and, thus,
l(f
E
k
) =
_
E
k
fg d, f L
p
(X, , ).
If 1 < p
< +, then, since [ g

k
[ [g[ on X, the Monotone Convergence Theo-
rem implies that
_
X
[g[
p
d = lim
k+
_
X
[ g
k
[
p
d = lim
k+
_
E
k
[g
k
[
p
d
limsup
k+
|l
k
|
p
|l|
p
< +. Hence, g L
p
(X, , ) and |g|

p
|l|.
If p
= +, we have that, for every k, [g[ = [g

k
[ |g
k
|
|l
k
| |l| -a.e.
on E
k
. This implies that [g[ |l| -a.e. on X and, thus, g L
(X, , ) and
|g|
|l|.
11.3. THE DUAL OF L
P
(X, , ). 249
Hence, in all cases, g L
p
(X, , ) and |g|

p
|l|.
For an arbitrary f L
p
(X, , ), we have |f
E
k
f|
p
p
=
_
X
[f
E
k
f[
p
d =
_
E
c
k
[f[
p
d =
_
X

E
c
k
[f[
p
d 0, by the Dominated Convergence Theorem. By
the continuity of l, we get l(f) = lim
k+
l(f
E
k
) = lim
k+
_
E
k
fg d =
_
X
fg d. The last equality holds since [
_
E
k
fg d
_
X
fg d[ = [
_
E
c
k
fg d[
_ _
E
c
k
[f[
p
d
_1
p
|g|
p
0. We have proved that
l(f) =
_
X
fg d, f L
p
(X, , )
and, hence, l = l
g
. Again, the uniqueness part of Proposition 11.19 implies that,
if also g
L
p
(X, , ) satises l = l
g
, then g
= g -a.e. on X.
C. Now, let 1 < p, p
< + and be arbitrary.

Let l (L
p
(X, , ))
.
We consider any E of -nite -measure and the restriction l
E
of
l on L
p
(E,
E
, ), dened in Lemma 11.2. Since l
E
(L
p
(E,
E
, ))
and
|l
E
| |l|, part B implies that there is a unique g
E
L
p
(E,
E
, ) so that
|g
E
|
p
|l| and
l(f
E
) = l
E
(f
E
) =
_
E
f
E
g
E
d
for every f L
p
(X, , ).
Now, let F E be two sets of -nite -measure with F E. For an
arbitrary h L
p
(F,
F
, ) we consider its extension h
on E as 0 on E F
and, observing that

h =

h
on X, we have
_
F
hg
F
d = l
F
(h) = l(
h) = l(
) =
l
E
(h
) =
_
E
h
g
E
d =
_
F
hg
E
d. By the uniqueness result of part B, we have
that g
F
= g
E
-a.e. on F.
We dene
M = sup
_
_
E
[g
E
[
p
d[ E of -nite -measure
_
and, obviously, M |l|
p
< +. We take a sequence E

n
in , where each
E
n
has -nite -measure, so that
_
En
[g
En
[
p
d M. We dene E =
+
n=1
E
n
and observe that E has -nite -measure and, hence,
_
E
[g
E
[
p
d M. Since
E
n
E, by the result of the previous paragraph, g
En
= g
E
-a.e. on E
n
and,
thus,
_
En
[g
En
[
p
d
_
E
[g
E
[
p
d M. Taking the limit as n +, this

implies that
_
E
[g
E
[
p
d = M.
We set g = g
E
and have that
_
X
[g[
p
d =
_
E
[g
E
[
p
d = M |l|.
Now consider an arbitrary f L
p
(X, , ). The set
F = E x X[ f(x) ,= 0
has -nite -measure. From g
E
= g
F
-a.e. on E, we get M =
_
E
[g
E
[
p
d =
_
E
[g
F
[
p
d
_
E
[g
F
[
p
d +
_
F\E
[g
F
[
p
d =
_
F
[g
F
[
p
d M. Therefore,
_
F\E
[g
F
[
p
d = 0 and, hence, g
F
= 0 -a.e. on F E. If f
F
is the restriction of
f on F, we have that l(f) = l(f
F
) =
_
F
f
F
g
F
d =
_
E
f
F
g
F
d =
_
E
fg
E
d =
_
X
fg d.
Thus, l(f) = l
g
(f) for every f L
p
(X, , ) and, hence, l = l
g
= J(g).
We conclude that J is onto and, combining with the result of Proposition
11.19, we nish the proof.
11.4 The space M(X, ).
Denition 11.27 Let (X, ) be a measurable space. The set of all real mea-
sures on (X, ) is denoted by M
r
(X, ) and the set of all complex measures on
(X, ) is denoted by M
c
(X, ).
If there is no danger of confusion, we shall use the symbol M(X, ) for both
M
r
(X, ) and M
c
(X, ).
We recall addition and multiplication on these spaces. If
1
,
2
M(X, ),
we dene
1
+
2
M(X, ) by (
1
+
2
)(A) =
1
(A) +
2
(A) for all A . We,
also, dene by ()(A) = (A) for all A . If M
r
(X, ) and R,
then M
r
(X, ) and, if M
c
(X, ) and C, then M
c
(X, ).
It is easy to show that M
r
(X, ) is a linear space over R and M
c
(X, ) is a
linear space over C. The zero element of both spaces is the measure o dened by
o(A) = 0 for all A . The opposite to is dened by ()(A) = (A)
for all A .
Denition 11.28 For every M(X, ) we dene
|| = [[(X).
Thus, || is just the total variation of .
Proposition 11.20 | | is a norm on M(X, ).
Proof: Proposition 10.9 implies that |
1
+
2
| = [
1
+
2
[(X) [
1
[(X) +
[
2
[(X) = |
1
| +|
2
| and || = [[(X) = [[[[(X) = [[||.
If || = 0, then [[(X) = 0. This implies that [(A)[ [[(A) = 0 for all
A and, hence, = o is the zero measure.
Theorem 11.10 M(X, ) is a Banach space.
Proof: Let
n
be a Cauchy sequence in M(X, ). This means [
n
m
[(X) =
|
n
m
| 0 as n, m + and, hence, [
n
(A)
m
(A)[ = [(
n
m
)(A)[
[
n

m
[(A) [
n

m
[(X) 0 as n, m +. This implies that the
sequence
n
(A) of numbers is a Cauchy sequence for every A . Therefore,
it converges to a nite number and we dene
(A) = lim
n+
n
(A)
11.4. THE SPACE M(X, ). 251
for all A .
It is clear that () = lim
n+
n
() = 0.
Now, let A
1
, A
2
, . . . be pairwise disjoint and A =
+
j=1
A
j
. We take an
arbitrary > 0 and nd N so that |
n

m
| for all n, m N. Since
+
j=1
[
N
[(A
j
) = [
N
[(A) < +, there is some J so that
+
j=J+1
[
N
[(A
j
) .
From [
n
[ [
n
N
[ +[
N
[ we get that, for every n N,
+
j=J+1
[
n
[(A
j
)
+
j=J+1
[
n
N
[(A
j
) +
+
j=J+1
[
N
[(A
j
)
[
n
N
[(
+
j=J+1
A
j
) +
[
n
N
[(X) + = |
n
N
| +
2.
This implies that, for arbitrary K J + 1 and every n N, we have
K
j=J+1
[
n
(A
j
)[
K
j=J+1
[
n
[(A
j
) 2 and, taking the limit as n +,
K
j=J+1
[(A
j
)[ 2. Finally, taking the limit as K +, we nd
+
j=J+1
[(A
j
)[ 2.
We have [
n
(A)
J
j=1
n
(A
j
)[ = [
+
j=J+1
n
(A
j
)[
+
j=J+1
[
n
(A
j
)[
+
j=J+1
[
n
[(A
j
) 2 for all n N and, taking the limit as n +,
[(A)
J
j=1
(A
j
)[ 2.
Altogether, we have
[(A)
+
j=1
(A
j
)[ [(A)
J
j=1
(A
j
)[ +
+
j=J+1
[(A
j
)[ 4.
Since is arbitrary, we get (A) =
+
j=1
(A
j
) and we conclude that
M(X, ).
Consider an arbitrary measurable partition A
1
, . . . , A
p
of X. We have
that
p
k=1
[(
n

m
)(A
k
)[ |
n

m
| for every n, m N. Taking the
limit as m +, we nd
p
k=1
[(
n
)(A
k
)[ for every n N and, taking
the supremum of the left side, we get
|
n
| = [
n
[(X) .
This means that |
n
| 0 as n +.
11.5 Exercises.
1. Approximation
(i) Let f L
p
(X, , ) and > 0. Using Theorem 6.1, prove that there
exists a measurable simple function on X so that |f |
p
< . If
p < +, then = 0 outside a set of nite -measure.
(ii) Let f L
p
(R
n
, L
n
, m
n
) and > 0. If p < +, prove that there exists
a function g continuous on R
n
and equal to 0 outside some bounded set
so that |f g|
p
< .
2. Let I be any index set and 0 < p < q +. Prove that l
p
(I) l
q
(I)
and that
|b|
q
|b|
p
for every b l
p
(I).
3. Let (X) < + and 0 < p < q +. Prove that L
q
(X, , )
L
p
(X, , ) and that
|f|
p
(X)
1
p
1
q
|f|
q
for every f L
q
(X, , ).
4. Let 0 < p < q < r + and f L
p
(X, , ) L
r
(X, , ). Prove that
f L
q
(X, , ) and, if
1
q
=
t
p
+
1t
r
, then
|f|
q
|f|
t
p
|f|
1t
r
.
5. Let 1 p < r +. We set Z = L
p
(X, , ) L
r
(X, , ) and we dene
|f| = |f|
p
+|f|
r
for every f Z.
(i) Prove that | | is a norm on Z and that (Z, | |) is a Banach space.
(ii) If p < q < r, consider the linear transformation T : Z L
q
(X, , )
with T(f) = f for every f Z (see Exercise 11.5.4). Prove that T is
bounded.
6. Let 0 < p < q < r + and f L
q
(X, , ). If t > 0 is arbitrary,
consider the functions
g(x) =
_
f(x), if [f(x)[ > t
0, if [f(x)[ t
h(x) =
_
0, if [f(x)[ > t
f(x), if [f(x)[ t
.
Prove that g L
p
(X, , ) and h L
r
(X, , ) and that f = g +h on X.
7. Let 1 p < r +. We dene W = L
p
(X, , ) + L
r
(X, , ) =
g +h[ g L
p
(X, , ), h L
r
(X, , ) and
|f| = inf
_
|g|
p
+|h|
r
[ g L
p
(X, , ), h L
r
(X, , ), f = g +h
_
for every f W.
(i) Prove that | | is a norm on W and that (W, | |) is a Banach space.
(ii) If p < q < r, consider the linear transformation T : L
q
(X, , ) W
with T(f) = f for every f L
q
(X, , ) (see Exercise 11.5.6). Prove that
T is bounded.
11.5. EXERCISES. 253
8. Let 0 < p < q < +. Prove that L
p
(X, , ) , L
q
(X, , ) if and
only if X includes sets of arbitrarily small positive -measure and that
L
q
(X, , ) , L
p
(X, , ) if and only if X includes sets of arbitrarily
large nite -measure.
9. Let 1 p < + and f
n
be a sequence in L
p
(X, , ) so that [f
n
[ g
-a.e. on X for every n for some g L
p
(X, , ). If f
n
converges to f
-a.e. on X or in measure, prove that |f
n
f|
p
0.
10. Let 1 p < + and f, f
n
L
p
(X, , ) for all n. If f
n
f -a.e. on X,
prove that |f
n
f|
p
0 if and only if |f
n
|
p
|f|
p
.
11. Let 1 p + and g L
(X, , ). We dene the linear trans-

formation T : L
p
(X, , ) L
p
(X, , ) with T(f) = gf for every
f L
p
(X, , ). Prove that T is bounded, that |T| |g|
and that
|T| = |g|
if is seminite.
12. The inequality of Chebychev.
If 0 < p < + and f L
p
(X, , ), prove that
|f|
(t)
|f|
p
p
t
p
, 0 < t < +.
13. The general Minkowskis Inequality.
Let (X
1
,
1
,
1
) and (X
2
,
2
,
2
) be two -nite measure spaces and 1
p < +.
(i) If f : X
1
X
2
[0, +] is
1
2
measurable, prove that
_
_
X1
_
_
X2
f(, ) d
2
_
p
d
1
_1
p
_
X2
_
_
X1
f(, )
p
d
1
_1
p
d
2
.
(ii) If f(, x
2
) L
p
(X
1
,
1
,
1
) for
2
-a.e. x
2
X
2
and the function x
2

|f(, x
2
)|
p
is in L
1
(X
2
,
2
,
2
), prove that f(x
1
, ) L
1
(X
2
,
2
,
2
) for
1
-
a.e. x
1
X
1
, that the function x
1

_
X2
f(x
1
, ) d
2
is in L
p
(X
1
,
1
,
1
)
and
_
_
X1
_
X2
f(, ) d
2
p
d
1
_1
p
_
X2
_
_
X1
[f(, )[
p
d
1
_1
p
d
2
.

H2p2sol Alt

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

H2p2sol Alt

Transféré par

Droits d'auteur :

Formats disponibles

NOTES ON

[ is a -algebra of subsets of X and c .

-sets are Borel sets in X.

] are not disjoint if and only if a

, then (a, b](a

are not disjoint if and only if for all j = 1, . . . , n we have that

are not disjoint, then a

/, then, because of the results of (a) and (b), AA

/ and / is closed under nite unions.

/[ / is a monotone class of subsets of X and c /.

of subsets of Y and a collection c of subsets

be a -algebra of subsets of Y and let f : X Y . Then the

is called the pull-back of

-set by considering the sets U

-set, i.e. a countable intersection

) = 0 and we must prove

B. Write B = A F for A and F E with (E) = 0 and

) (A) + (E) = 0. Hence

, then B and B Y and, if we set

), and hence (X,

, ) is a saturated measure space.

, ) is called the saturation of (X, , ).

). It is called the push-forward of

) be a measure space and f : X Y be one-to-one and onto

by f on X (see Exercise 1.6.8). We dene

be an outer measure on the non-empty set X. We say

the collection of all

-measurable if and only if it decomposes every subset of X into

(E). Furthermore, it is enough to

is an outer measure on X, then the collec-

-measurable subsets of X is a -algebra. If we denote the

, ) is a complete measure space.

is closed under complements.

and take an arbitrary E X. To check A B

for the rst term to get

and by induction we get that

with AB = and get for all E X that

are pairwise disjoint and

are pairwise disjoint and E X is arbitrary, then, for

are pairwise disjoint and E X is arbitrary, then, by

for all N. Hence

for the second

are not necessarily pairwise disjoint, we write B

and they are pairwise disjoint. Therefore, by the last

[0, +] as the restriction of

(A) for all A

() = 0 we see that (X,

with (A) = 0 and B A. Then

be an outer measure on the non-empty X.

(E) the set over which the

be an outer measure on X and Y X. Dene

is an outer measure on X and nd all the

is an outer measure on X. We say that

, ) is the complete measure space which results from

if and only if there is some

, ) is the completion of (X,

= T(X). Thus, the result of (v)

, ) is always the saturation (see exercise 2.5.15)

is an outer measure on X. We say that

-measurable and hence (/

, ) is a complete measure space which extends

be an outer measure on X. We say that

(E) = (A) (where is the usual restriction of