Sei sulla pagina 1di 177

April 24, 2018

Measure and Integration


A First Course

M.Thamban Nair
Department of Mathematics
Indian Institute of Technology Madras

March 2018
.
Preface

The concepts from the theory of measure and integration are vital to any
advanced courses in analysis and its applications, specially in the the ap-
plications of Functional Analysis to other areas such as Harmonic Analysis,
Partial Differential Equations and Integral Equations, and in the theoreti-
cal investigations in applied mathematics. Therefore, an early introduction
to such concepts become essential in the masters program in mathematics.
This book is an attempt towards that goal using minimum background in
mathematical analysis.
It is essentially an updated version of the notes which the author has been
using for teaching courses on measure and integration theory several times
for the last 30 years, during 1987–1995 at Goa University and afterwards at
IIT Madras. The topics covered in this book are standard ones. However,
the reader will definitely find that the presentation of the concepts and
results is different from the standard texts.
It starts by a short introduction on Riemann integration to motivate the
necessity of the concept of integration of functions that are more general than
those allowed in Riemann integration and then introduces the concept of
Lebesgue measurable sets that is more general than the concept of intervals.
Once we have this family of measurable sets, and the concept of a Lebesgue
measure, it becomes almost obvious that one need not restrict the theory
of integration to the subsets of the real line, but can be developed on any
set together with a sigma algebra on it. Thus, the concept of a measure
on a measurable space allows us to have a theory of integration in a very
general setting which has immense potential for application to diverse areas
of mathematics and its applications.
Although the theory of integration is very vast, the attempt in this book
is to introduce the students to this modern subject in a simple and natural
manner so that they can pursue the subject further with confidence, and
also apply the concepts in other branches of mathematics such as those

iii
iv Preface

mentioned in the first paragraph.


This book can be used for a one semester course of about 45 lectures for
the first or second semester of a postgraduate programme in mathematics. If
it is to be taught for the final year of undergraduate program in mathemat-
ics, then the last two chapters, which are on product measure and Fourier
transform, can be omitted.
As Lebesgue measure on the real line is introduced in the beginning, no
pre-requisite is assumed, except the mathematical maturity to appreciate
and grasp concepts in analysis.

Acknowledgements: While teaching this course as well as during the


preparation of the notes, I have benefited a lot by the contributions of my
students in terms of their questions in classes and also during the clarifi-
cation of their doubts. Also, I thank my PhD student Ajoy Jana and Dr.
Sivanandan of IIT Delhi for reading the manuscript carefully and bringing
to my notice many typos and corrections in the earlier draft of the text.

November 2015 M. Thamban Nair


.
Contents

Preface iii

1 Review of Riemann Integral 1


1.1 Definition and Some Characterizations . . . . . . . . . . . . . 1
1.2 Advantages and Some Disadvantages . . . . . . . . . . . . . . 8
1.3 Notations and Conventions . . . . . . . . . . . . . . . . . . . 11

2 Lebesgue Measure 13
2.1 Lebesgue Outer Measure . . . . . . . . . . . . . . . . . . . . . 13
2.2 Lebesgue Measurable Sets . . . . . . . . . . . . . . . . . . . . 21
2.3 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3 Measure and Measurable Functions 35


3.1 Measure on an Arbitrary σ-Algebra . . . . . . . . . . . . . . . 35
3.2 Lebesgue measure on Rk . . . . . . . . . . . . . . . . . . . . . 40
3.3 Generated σ-Algebra and Borel σ-Algebra . . . . . . . . . . . 41
3.4 Restrictions of σ-Algebras and Measures . . . . . . . . . . . . 43
3.5 Complete Measure Space and Completion . . . . . . . . . . . 46
3.6 Some Properties of Measures . . . . . . . . . . . . . . . . . . 49
3.7 Measurable Functions . . . . . . . . . . . . . . . . . . . . . . 53
3.7.1 Almost everywhere properties . . . . . . . . . . . . . . 60
3.7.2 Further properties of measurable functions . . . . . . 64
3.8 Simple measurable functions . . . . . . . . . . . . . . . . . . . 67
3.9 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

vi
Contents vii

4 Integral of Positive Measurable Functions 74


4.1 Integral of Simple Measurable Functions . . . . . . . . . . . . 74
4.2 Integral of Positive Measurable Functions . . . . . . . . . . . 81
4.3 Riemann Integral as Lebesgue Integral . . . . . . . . . . . . . 87
4.4 Monotone Convergence Theorem (MCT) . . . . . . . . . . . . 88
4.5 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.6 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

5 Integral of Complex Measurable Functions 103


5.1 Integrability and some properties . . . . . . . . . . . . . . . . 103
5.2 Riemann Integral as Lebesgue Integral . . . . . . . . . . . . . 110
5.3 Dominated Convergence Theorem (DCT) . . . . . . . . . . . 112
5.4 Lp Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.4.1 Hölder’s and Minkowski’s inequalities . . . . . . . . . 121
5.4.2 Completeness of Lp (µ) . . . . . . . . . . . . . . . . . . 126
5.4.3 Denseness of Cc (Ω) in Lp (Ω) for 1 ≤ p < ∞ . . . . . . 131
5.5 Indefinite Integral and its Derivative . . . . . . . . . . . . . . 134
5.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
5.7 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

6 Integration on Product Spaces 152


6.1 Product Measure . . . . . . . . . . . . . . . . . . . . . . . . . 152
6.2 Fubini’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 161
6.3 Counter Examples . . . . . . . . . . . . . . . . . . . . . . . . 164
6.3.1 σ-finiteness condition cannot be dropped . . . . . . . 164
6.3.2 Product of complete measures need not be complete . 165
6.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165

References 167

Index 168
1

Review of Riemann Integral

In this chapter, the definition of Riemann integral of a bounded


function f : [a, b] → R is reviewed, and its limitations are pointed
out so as to convince the reader about the necessity of a more
general integral.

1.1 Definition and Some Characterizations


Let f : [a, b] → R be a bounded function. The idea of Riemann integral of
f is to associate a unique number γ to f such that, in case f (x) ≥ 0 for all
x ∈ [a, b], then γ can be thought of as the the area of the region bounded by
the graph of f , x-axis, and the ordinates at a and b. For this purpose, first we
consider a partition P of [a, b], that is, a finite set P := {xi : i = 0, 1, . . . , k}
such that a = x0 < x1 < x2 < . . . < xk = b, usully written as

P : a = x0 < x1 < x2 < . . . < xk = b,

and consider the sums


k
X k
X
L(P, f ) := mi ∆xi , U (P, f ) := Mi ∆xi ,
i=1 i=1

where, for i = 1, . . . , k, ∆xi = xi − xi−1 ,

mi = inf{f (x) : xi−1 ≤ x ≤ xi } and Mi = sup{f (x) : xi−1 ≤ x ≤ xi }.

The quantities L(P, f ) and U (P, f ) are called the lower sum and upper sum
associated with (P, f ). We may observe that in the definition of mi and Mi ,
we used the fact that f is a bounded function.
Note that if f (x) ≥ 0 for all x ∈ [a, b], then L(P, f ) is the total area of
the rectangles with lengths mi and widths xi − xi−1 , and U (P, f ) is the total

1
2 Review of Riemann Integral

area of the rectangles with lengths Mi and widths xi − xi−1 , for i = 1, . . . , k.


Thus, it is intuitively clear that the required area, say γ, under the graph
of f must satisfy the relation:

L(P, f ) ≤ γ ≤ U (P, f )

for all partitions P of [a, b]. With this requirement in mind, we introduce
the following definition.

Definition 1.1.1 A bounded function f : [a, b] → R is said to be Riemann


integrable on [a, b] if there exists a unique γ ∈ R such that

L(P, f ) ≤ γ ≤ U (P, f )

for all partitions P of [a, b]. If such a γ exists, then it is called the Riemann
integral of f and it is denoted by
Z b
f (x)dx. ♦
a

We shall see that every bounded function f : [a, b → R having atmost


a finite number of discontinuities is Riemann integrable. However, every
bounded function f on [a, b] need not be Riemann integrable, as the following
example shows.
Example 1.1.2 Let f : [0, 1] → R be the Dirichlet’s function, that is,

0 if x rational,
f (x) =
1 if x irrational.

Note that, for any partition P of [a, b], we have L(P, f ) = 0 and U (P, f ) = 1.
Thus, every number α ∈ [0, 1] satisfies L(P, f ) ≤ α ≤ U (P, f ) for every
partition P of [0, 1]. Hence, f is not Riemann integrable. ♦
Note that, in the definition of Riemann integral, we used the fact that f
is a bounded function.
Let P be the set of all partitions of [a, b]. Clearly

L(P, f ) ≤ U (P, f ) ∀ P ∈ P.

Denoting

m = inf{f (x) : a ≤ x ≤ b}, M = sup{f (x) : a ≤ x ≤ b},


Definition and Some Characterizations 3

for any partition P = {xi : i = 0, 1, . . . , k}, we have


k
X k
X
L(P, f ) = mi ∆xi ≥ m∆xi = m(b − a)
i=1 i=1

and
k
X k
X
U (P, f ) = Mi ∆xi ≤ M ∆xi = M (b − a)
i=1 i=1
so that
m(b − a) ≤ L(P, f ) ≤ U (P, f ) ≤ M (b − a).
Thus, the set {L(P, f ) : P ∈ P} is bounded above by M (b − a), and the set
{U (P, f ) : P ∈ P} is bounded below by m(b − a). Hence,
L(f ) := sup{L(P, f ) : P ∈ P},
U (f ) := inf{U (P, f ) : P ∈ P}
exist as real numbers. Using the quantities L(f ) and U (f ), we have the
following characterization of Riemann integrability.
Theorem 1.1.3 A bounded function f : [a, b] → R is Riemann integrable
on [a, b] if and only if L(f ) = U (f ).

Proof. Suppose f : [a, b] → R is Riemann integrable on [a, b], and let γ


be the Riemann integral of f . Then, from the relations
L(P, f ) ≤ γ ≤ U (P, f ) ∀P ∈ P (∗)
we have
L(f ) ≤ γ ≤ U (f ).
Consequently,
L(P, f ) ≤ L(f ) ≤ γ ≤ U (f ) ≤ U (P, f ) ∀P ∈ P.
Now, since γ is the only number satisfying (∗), we have L(f ) = γ = U (f ).
Conversely, suppose L(f ) = U (f ). Then we have
L(P, f ) ≤ L(f ) = U (f ) ≤ U (P, f ) ∀P ∈ P.
Thus, γ := L(f ) = U (f ) satisfies (∗). Further, if γ̃ ∈ R satisfies
L(P, f ) ≤ γ̃ ≤ U (P, f ) ∀P ∈ P,
then, from the definition of L(f ) and U (f ), it follows that L(f ) ≤ γ̃ ≤ U (f ).
Hence, γ̃ = γ. Thus, we have proved that there exists a unique γ ∈ R
satisfying (∗).
4 Review of Riemann Integral

Remark 1.1.4 As you would have observed, the proof of Theorem 1.1.3
was very easy. We gave its proof in detail, mainly because of the fact that
in standard text books, the Riemann integrability of a bounded function
f : [a, b] → R is defined by requiring L(f ) = U (f ). ♦
Given two partitions P1 and P2 of [a, b], we can consider a new partition
P by using all the partition points of P1 and P2 , taking repeated points
only once. Such a partition will be called the partition obtained by
combining P1 and P2 , and is usually denoted by P1 ∪ P2 .
Given any two partitions P and Q of [a, b], it can be shown that
L(P, f ) ≤ L(P ∪ Q, f ) ≤ U (P ∪ Q, f ) ≤ U (Q, f );
consequently, L(f ) ≤ U (f ).

The characterization given in Theorem 1.1.5 below is useful in deducing


many properties of Riemann integral.
Theorem 1.1.5 Let f : [a, b] → R be a bounded function. Then f is
Riemann integrable if and only if for every ε > 0, there exists a partition P
of [a, b] such that U (P, f ) − L(P, f ) < ε.

Proof. Suppose f is Riemann integrable and let ε > 0 be given. By the


definition of L(f ) and U (f ), there exist partitions P1 and P2 of [a, b] such
that
L(f ) − ε/2 < L(P1 , f ) and U (P2 , f ) < U (f ) + ε/2.
Let P = P1 ∪ P2 , the partition obtained by combining P1 and P2 . Then, we
have
L(f ) − ε/2 < L(P1 , f ) ≤ L(P, f ) ≤ U (P, f ) ≤ U (P2 , f ) < U (f ) + ε/2.
Since L(f ) = U (f ) (cf. Theorem 1.1.3), it follows that
U (P, f ) − L(P, f ) < [U (f ) + ε/2] − [L(f ) − ε/2] = ε.

Conversely, suppose that for every ε > 0, there exists a partition P of


[a, b] such that U (P, f ) − L(P, f ) < ε. Since
L(P, f ) ≤ L(f ) ≤ U (f ) ≤ U (P, f ),
we have
U (f ) − L(f ) ≤ U (P, f ) − L(P, f ) < ε.
This is true for every ε > 0. Hence, L(f ) = U (f ), and hence f is Riemann
integrable.
Definition and Some Characterizations 5

Here is an an immediate consequence of the above theorem (Exercise).


Corollary 1.1.6 A bounded function f : [a, b] → R is Riemann integrable
if and only if there exists a sequence (Pn ) of partitions of [a, b] such that

U (Pn , f ) − L(Pn , f ) → 0 as n → ∞,

and in that case the sequences (U (Pn , f )) and (L(Pn , f )) converge to the
Rb
same limit a f (x)dx.

Next we give another characterization of Riemann integrability. For that


purpose we introduce the following definition.
Definition 1.1.7 Let P : a = x0 < x1 < · · · < xn = b be a partition of
[a, b] and let T := {ti : i = 1, . . . , n} with ti ∈ [xi−1 , xi ], i = 1, . . . , n. The
set T is called a tag set for P . Given a function f : [a, b] → R, the sum
n
X
S(P, T, f ) := f (ti )∆xi
i=1

is called the Riemann sum of f associated with (P, T ). The quantity

|P | := max{xi − xi−1 : i = 1, . . . , n}

is called the mesh of the partition P . ♦

We observe that for a given partition of [a, b], the Riemann sums may
vary as the tag sets vary.
It is obvious that, if f : [a, b] → R is a bounded function, then

L(P, f ) ≤ S(P, T, f ) ≤ U (P, f )

for any partition P and tag set T for P . Therefore, by Theorem 1.1.5, we
have the following result.
Theorem 1.1.8 Let f : [a, b] → R be a bounded function. If f is Riemann
integrable, then for every ε > 0, there exists a partition P such that
Z b

S(P, T, f ) − f (x)dx <ε

a

for every tag set T for P .

In fact, the converse to Theorem 1.1.8 is also true.


6 Review of Riemann Integral

Theorem 1.1.9 For a function f : [a, b] → R, if there exists γ ∈ R such


that for every ε > 0, there exists a partition P of [a, b] satisfying

|S(P, T, f ) − γ| < ε
Rb
for every tag set T of P , then f is Riemann integrable and γ = a f (x)dx.

Proof. Let ε > 0 be given and let P : a = x0 < x1 < · · · xk = b be as in


the hypothesis of the theorem. Then we have

γ − ε < S(P, T, f ) < γ + ε (1)

for any tag set T corresponding to P . Let ui , vi ∈ [xi−1 , xi ] for i = 1, . . . , k


be such that

Mi − ε < f (ui ) and f (vi ) < mi + ε for i = 1, . . . , k. (2)

Now, consider the tages

T1 = {ui : i = 1, . . . , k} and T2 = {vi : i = 1, . . . , k}

for the partition P . Then from (2), we have

U (P, f ) − ε(b − a) < S(P, T1 , f ) and S(P, T2 , f ) < L(P, f ) + ε(b − a).

This, together with (1), implies

U (P, f ) − ε(b − a) < γ + ε and γ − ε < L(P, f ) + ε(b − a).

In particular,

U (P, f ) − γ < ε[(b − a) + 1], γ − L(P, f ) < ε[(b − a) + 1]

and
U (P, f ) − L(P, f ) < 2ε[(b − a) + 1].
Hence, by Theorem
R b 1.1.5, it follows that f is Riemann
Rb integrable, and by
the definiton of a f (x)dx (Definition 1.1.1), γ = a f (x)dx.

But, if we know that f is Riemann integrable, then


R b the following theorem
is better suited for obtaining approximations for a f (x) dx. For its proof
we refer [4].
Definition and Some Characterizations 7

Theorem 1.1.10 Suppose f : [a, b] → R is a bounded is a Riemann inte-


grable function. Then for every ε > 0, there exists a δ > 0 such that
Z b
S(P, T, f ) − f (x)dx < ε

a

for any partition P of [a, b] with |P | < δ and for every tag set T for P .

The conclusion in the above theorem is usually written as


Z b
lim S(P, T, f ) = f (x)dx.
|P |→0 a

Here is an immediate consequence of Thoerem 1.1.10.

Corollary 1.1.11 Suppose f : [a, b] → R is a Riemann integrable function


and (Pn ) is a sequence of partitions of [a, b] such that |Pn | → 0 as n → ∞.
If Tn is a tag set for Pn for each n ∈ N, then
Z b
S(Pn , Tn , f ) → f (x)dx as n → ∞.
a

Looking at Theorem 1.1.9, one may ask the following question.

If (Pn ) is a sequence of partitions of [a, b] such that lim S(Pn , Tn , f ) =


n→∞
γ for some γ ∈ R, where Tn is a tag set for Pn for each R b n ∈ N,
then, is it true that f is Riemann integrable and γ = a f (x)dx?

The answer is in negative as the following example shows.


Example 1.1.12 Consider the Dirichlet function f : [0, 1] → R, of Example
1.1.2. That is,

0, x ∈ Q,
f (x) :=
1, x 6∈ Q.

Let xi := i/n for i = 0, 1, . . . , n, and let ti be any rational point in the


interval [xi=1 , xi ]. In this case we have lim |Pn | = 0 and S(Pn , Tn , f ) = 0
n→∞
for all n ∈ N so that lim S(Pn , Tn , f ) = 0. However, f is not Riemann
n→∞
integrable. ♦
8 Review of Riemann Integral

1.2 Advantages and Some Disadvantages


We may observe, in view of Corollary 1.1.6, that if (Pn ) is a sequence of
partitions of [a, b] such that (U (Pn , f )) and (L(Pn , f )) converge to the same
limit say γ, then f is Riemann integrable, and
Z b
γ= f (x)dx.
a
If f : [a, b] → R is a continuous function, then it can be shown that for any
sequence (Pn ) of partitions of [a, b] satisfying
|Pn | → 0 as n → ∞,
we have U (Pn , f ) − L(Pn , f ) → 0 as n → ∞. Thus, by Corollary 1.1.6, we
can conclude the following:

Every continuous function f : [a, b] → R is Riemann integrable.

The following results are also true; for their proofs, the reader may refer to
Ghorpade and Limaye [4] or Rudin [10].

1. Every bounded function f : [a, b] → R having atmost a finite number


of discontinuities is Riemann integrable.
2. Every monotonic function f : [a, b] → R is Riemann integrable.

Thus, the set of all Riemann integrable functions is very large. In fact
we have the following theorem, known as Lebesgue’s criterion for Riemann
integrability, whose proof depends on some techniques involving the concepts
of oscillation of a function; refer Delninger [2] for its proof.

Lebesgue’s criterion for Riemann integrability: A bounded


function f : [a, b] → R is Riemann integrable if and only if the
set of points at which f is discontinuous is of “measure zero”.

Here, the concept of a set of measure zero is used in the sense of the following
definition.
Definition 1.2.1 A set E ⊆ R is said to be of measure zero if for every
ε > 0, there exists a countable family {In } of open intervals such that
[ X
E⊆ In and `(In ) < ε,
n n

where `(In ) is the length of the interval In . ♦


Advantages and Some Disadvantages 9

Example 1.2.2 We show that every countable subset of R is of measure


zero. To see this, consider a countable set E = {an : n ∈ Λ}, where Λ is
{1, . . . , k} for some k ∈ N or Λ = N. For ε > 0, let

In := (an − ε/2n+1 , an + ε/2n+1 ), n ∈ Λ.

Then X X
E ⊆ ∪n∈Λ In and `(In ) = (ε/2n ) ≤ ε. ♦
n∈Λ n∈Λ

Can an uncountable set be of measure zero? We shall answer this ques-


tion affirmatively in the next chapter. Functions with only a finite number
of discontinuity in [a, b] are plenty. Here is an example of a function with
countably infinite number of discontinuities.
Example 1.2.3 Let I = [a, b], S := {an : n ∈ N} ⊆ I and let f : I → R be
defined by f (an ) = 1/n for all n ∈ N and f (x) = 0 for x 6∈ S. Clearly, this
function is not continuous at any x ∈ S. We show that f is continuous at
every x ∈ I \ S.
Let x0 ∈ I \ S. Then f (x0 ) = 0. For ε > 0, we have to find a δ > 0 such
that |x − x0 | < δ implies |f (x)| < ε.
For δ > 0, let Jδ := (x0 − δ, x0 + δ) ∩ I. For ε > 0, let k ∈ N be such
that 1/k < ε. Choose δ > 0 such that

a1 , a2 , . . . , ak 6∈ Jδ .

For instance, we may choose 0 < δ < min{|x0 − ai | : i = 1, . . . , k}. Then we


have
Jδ ∩ {a1 , a2 , . . .} = {ak+1 , ak+2 , . . .}.
Hence, for x ∈ Jδ , we have either f (x) = 0 or f (x) = 1/n for some n > k.
Thus,
1
|f (x)| ≤ < ε.
k
Thus we have proved that f is continuous at x0 .
If we take S = Q, the set of all rational numbers in I, then f is continuous
at every irrational number in I and discontinuous at every rational number
in I. ♦
Although the set set of Riemann integrable functions on [a, b] is quite
large, this class lacks some desirable properties. For example observe the
following draw backs of Riemann integrability and Riemann integration:
10 Review of Riemann Integral

1. If (fn ) is a sequence of Riemann integrable functions on [a, b] and if


fn (x) → f (x) as n → ∞ for every x ∈ [a, b], then it is not necessary
that f is Riemann integrable.

2. EvenRif the functionR f above is Riemann integrable, it is not necessary


b b
that a fn (x)dx → a f (x)dx as n → ∞.

To illustrate the last two statements consider the following examples.


Example 1.2.4 Let {r1 , r2 , . . .} be an enumeration of the set rational num-
bers in [0, 1]. For each n ∈ N, let

0 if x ∈ {r1 , . . . , rn },
fn (x) =
1 if x 6∈ {r1 , . . . , rn }.
Then each fn is Riemann integrable, as it is continuous except at a finite
number of points. Note that fn (x) → f (x) as n → ∞ for every x ∈ [0, 1],
where f : [0, 1] → R is the Dirichlet’s function, that is,

0 if x rational,
f (x) =
1 if x irrational.
We hae seen in Example 1.1.2 that f is not Riemann integrable, which also
follows from the Lebesgue’s criterion of Riemann
R1 integrability, since f is
discontinuous everywhere. Thus, though 0 fn (x)dx = 1 for every n ∈ N
and fn (x) → f (x) for every x ∈ [0, 1], we cannot even talk about the integral
of f . ♦
Example 1.2.5 For n ∈ N, let fn : [0, 1] → R be defined by
fn (x) = nχ(0,1/n] (x), x ∈ [0, 1],
where χE denotes the characteristic function of E, that is,

1 if x ∈ E,
χE (x) =
0 if x 6∈ E.
R1
Then we see that fn (x) → f (x) ≡ 0 as n → ∞, but 0 fn (x) dx = 1 for
Rb Rb
every n ∈ N. Hence a fn (x)dx 6→ a f (x)dx. ♦
Example 1.2.6 Consider fn : [0, 1] → R defined by
ne−nx if x ∈ (0, 1],

fn (x) =
0 if x = 0.
Then
R1 we have fn (x) → f (x) = 0 as n → ∞ for
R every x ∈ [0,
R 1]. Note that
f (x)dx = 1 − e −n → 1 as n → ∞. Thus, 1 f (x)dx 6→ 1 f (x)dx. ♦
0 n 0 n 0
Notations and Conventions 11

In Examples 1.2.5 and 1.2.6, we see that, although the sequence (fn (x))
converges for each x ∈ [0, 1], it is not uniformly bounded, that is, there does
not exist an M > 0 such that |fn (x)| ≤ M for all x ∈ [a, b] and for all n ∈ N.
In fact, in both the examples, the sequence is (fn (1/n)) is unbounded. In
this context, it is worth mentioning the following theorem, known as Arzela’s
dominated convergence theorem.

Theorem 1.2.7 (Arzela’s dominated convergence theorem) Suppose


(fn ) is sequence of Riemann integrable functions defined on [a, b] such that
fn (x) → f (x) as n → ∞ for each x ∈ [a, b] for some Riemann integrable
function f . If (fn ) is uniformly bounded, then
Z b Z b
fn (x)dx → f (x)dx as n → ∞.
a a

immediate

Most of the known proofs for the above theorem, in the context of Rie-
mann integration, are quite involved. For a recent elementary proof, one
may refer [5].
Note that, in Arzela’s theorem, we assumed that the f is Riemann inte-
grable. In this course we shall have a new type of integral, which includes
Riemann integral, called the Lebesgue integral, and derive Arzela’s theorem
(see Theorem 5.3.1) as a consequence of a more general result. In fact,
the assumption of Riemann integrability of f is not required, but then, the
integral of f is to be understood in the Lebesgue sense.

In the next section we introduce certain notations and conventions which


we shall use thoroughout the book.

1.3 Notations and Conventions


By saying that {An } is a countable family of sets, we mean a family {An :
n ∈ Λ} of sets An , n ∈ Λ, where either Λ = {1, 2, . . . , k} for some k ∈ N or
Λ = N, the set of natural numbers. Then, ∪{An : n ∈ Λ} and ∩{An : n ∈ Λ}
will be written as ∪n An and ∩n An , respectively. Also, forP a countable set
{an ∈ R : nP∈ Λ} of non-negative real numbers, the series n∈Λ an will be
written as n an .
By an interval we mean a subset I of R which is any of the following
forms (a, b), [a, b), (a, b], [a, b] for any a, b ∈ R with a < b or of the forms
12 Review of Riemann Integral

(a, ∞), [a, ∞), (−∞, a), (−∞, a] for any a ∈ R. Here, for a ∈ R,

(a, ∞) := {x ∈ R : a < x}, [a, ∞) := {x ∈ R : a ≤ x},

(−∞, a) := {x ∈ R : x < a}, (−∞, a] := {x ∈ R : x ≤ a}.


The intervals (a, b), [a, b), (a, b], [a, b] are bounded intervals with end points
a and b, and the intervals (a, ∞), [a, ∞), (−∞, a), (−∞, a], (−∞, ∞) are
unbounded intervals.
Length of an interval I is denoted by `(I). Thus, for a bounded interval
I of end points a, b with a < b, `(I) is b − a. If I is an unbounded interval,
P `(I) = ∞. Also, for a
then we say thatPits length is infinity, and we write
divergent series n∈Λ an with an ≥ 0, we write n∈Λ an = ∞.
Although ∞ is not a real number, we shall use the symbol ∞ as an
extended real number.
We have already used the symbol ∞ to denote the sum of a divergent
series of nonnegative real numbers. Also, ∞ can be considered to the supre-
mum of a set of real numbers which is not bounded above.
Similarly, we may use the symbol −∞ for the sum of a divergent series
of nonpositive real numbers and for infimum of a set of real numbers which
is not bounded below.
Thus, we have the set R̃ := R ∪ {∞, −∞}, called the set of all extended
real numbers. The order relations, addition and multiplication are defined
on R̃ are defined as follows:

For every a ∈ R,

±∞ if a > 0,
∞ > a, −∞ < a, ±∞±a = ±∞, ±∞×a =
∓∞ if a < 0.
Further,

±∞ × 0 = 0, ±∞ ± ∞ = ±∞, (±∞) × (±∞) = ∞.

However, ±∞ ∓ ∞ is not deefined.

In the above, addition and multiplication are commutaitve. Also, for any
a ∈ R, we denote

(a, ∞] := (a, ∞) ∪ {∞}, [−∞, a) := (−∞, a) ∪ {−∞},

[a, ∞] := [a, ∞) ∪ {∞}, [−∞, a] := (−∞, a] ∪ {−∞}.


2

Lebesgue Measure

This chapter is a stepping stone for the subject of measure and


integration. Here, we define the concept of an outer measure of a
subset of the real line as a generalization of the concept of length
of an interval, and then define the concept of measure of certain
subsets of the real line which must satisfy certain intuitively
desirable properties. The properties of the class of subsets which
can be measured is a motivation for the concept of a general
measure in the context of an arbitrary set which we shall consider
in the subsequent chapters.

2.1 Lebesgue Outer Measure


In Chapter 1 we defined the concept of a set of measure zero for subsets of
R while stating a characterization of Riemann integrability of bounded real
valued functions defined on closed and bounded intervals (Definition 1.2.1).
We have also observed that every countable subset of R is of measure zero.
In particular set of rationals in any interval is of measure zero. So, we may
ask the following questions:

1. If a subset E of R contains an open interval (of nonzero length), then


can it be of measure zero?

2. If a subset E of R contains no open interval, is it of measure zero?

3. Can an uncountable subset E of R be of zero measure?

We shall find answers to these questions after defining the concept of


Lebesgue outer measure. In fact we shall see that a set is of measure zero
if and only if its Lebesgue outer measure is zero, and we shall answer the
above raised questions in the following manner:

13
14 Lebesgue Measure

1. If a subset E of R contains an open interval (of nonzero length), then


it cannot be of measure zero.

2. If a subset E of R contains no open interval, then it can be of measure


zero.

3. An uncountable subset of R can be of measure zero.

Definition 2.1.1 The Lebesgue outer measure of a set E ⊆ R is defined


as X
m∗ (E) := inf `(In ),
IE
n

where the infimum is taken over the collection IE of all countable family
{In } of open intervals which covers E, that is, E ⊆ ∪n In . ♦

Note that m∗ (E) ≥ 0, and m∗ (E) can take the value ∞ as well. Thus,
m∗ can be thought of as a function from the family of all subsets of R into
the set [0, ∞].

From the definition of the Lebesgue outer measure m∗ , it follows that,


if (In ) is a sequence of open intervals, then the inequality
X
m∗ (∪In ) ≤ `(In )
n

holds (Exercise). Now, we shall give detaied proofs for some of the properties
of the Lebesgue outer measure.

Theorem 2.1.2 The following results hold.

(i) If I is an open interval and A ⊆ I, then m∗ (A) ≤ `(I).

(ii) m∗ (∅) = 0.

(iii) If E is a countable subset of R, then m∗ (E) = 0.

Proof. (i) Let A ⊆ I, where I is an open interval.


Taking the singleton family {In } with In = I and n = 1, we obtain
m∗ (A) ≤ `(I).
(ii) For every ε > 0, we have ∅ ⊆ (−ε, ε). Hence by (i), m∗ (∅) ≤ 2ε.
This is true for every ε > 0. Hence, m∗ (∅) = 0.
Lebesgue Outer Measure 15

(iii) First let E be a finite set, say E = {a1 , . . . , ak } ⊆ R. Then for every
ε > 0, E ⊆ ∪ki=1 Ii , where Ii = (ai − ε, ai + ε). Hence, m∗ (E) ≤ 2kε. Since
this is true for every ε > 0, m∗ (E) = 0. Next suppose that E is a countably
infinite set, say E = {ai : i ∈ N}. Then taking

In := (an − ε/2n+1 , an + ε/2n+1 ),

we have E ⊆ ∪n∈N In so that


X X
m∗ (E) ≤ `(In ) = (ε/2n ) ≤ ε.
n∈N n∈N

Since this is true for every ε > 0, we obtain m∗ (E) = 0.

The following theorem is useful in deriving many properties of the outer


measure.
Theorem 2.1.3 Let E ⊆ R. For every ε S> 0, there exists a countable
family {In } of open intervals such that E ⊆ n In and
X
`(In ) ≤ m∗ (E) + ε.
n

Strict inequality occurs in the above if m∗ (E) < ∞.

Proof. If m∗ (E) = ∞, then the conclusion is true trivially. So, assume


that m∗ (E) < ∞. Then the results follow from the definition of infimum of
a subset of R.

By Theorem 2.1.3, for E ⊆ R, m∗ (E) = 0 if and only if for every ε > 0,


there exists a countable family {In } of open intervals such that
[ X
E⊆ In and `(In ) ≤ ε.
n n

Thus, we can conclude that

E is of measure zero if and only if its outer measure is zero.

Corollary 2.1.4 Let E ⊆ R. For every ε > 0, there exists an open set G
in R such that
E ⊆ G and m∗ (G) ≤ m∗ (E) + ε.
16 Lebesgue Measure

Proof. Let ε > 0 be given. By Theorem S 2.1.3, there exists a countable


family {In } of open intervals such that E ⊆ n In and
X
`(In ) ≤ m∗ (E) + ε.
n

Take G = n In . Then, by the definition of m∗ , m∗ (G) ≤


S P
n `(In ). Thus,
we have m∗ (G) ≤ m∗ (E) + ε.

Remark 2.1.5 In view of Corollary 2.1.4, one may ask the following ques-
tion:

If E ⊆ R, then for every ε > 0, does there exists an open set


G ⊃ E such that m∗ (G \ E) < ε?

We shall see, in Theorem 2.2.19, that the answer to the above question
is in affirmative only if E belongs to a class of subsets, called Lebesgue
measurable sets. ♦

Notation: Sets considered in this chapter are subsets of R. Also, for E ⊆ R,


we shall use the notation IE to denote the class of all countable family of
open intervals {In } such that E ⊆ ∪n In .
For subsets A and B of R, we define

A + B = {x + y : x ∈ A, y ∈ B}.

Also for E ⊆ R and a ∈ R, we denote, E + a = E + {a} so that

E + a := {x + a : x ∈ E}.

Theorem 2.1.6 The following results hold.


(i) A ⊆ B ⇒ m∗ (A) ≤ m∗ (B).
(ii) m∗ (A ∪ B) ≤ m∗ (A) + m∗ (B).
(iii) If E ⊆ A and m∗ (E) = 0, then m∗ (A \ E) = m∗ (A).
(iv) If E ⊆ R and x ∈ R, then m∗ (E + x) = m∗ (E).

Proof. (i) Let A ⊆ B and let {In } ∈ IB . Then {In } ∈ IA . Hence


X
m∗ (A) ≤ `(In ).
n
Lebesgue Outer Measure 17

Now, taking infimum over all {In } ∈ IB , we have m∗ (A) ≤ m∗ (B).


(ii) If at least one of m∗ (A) and m∗ (B) is infinity then the result holds.
Next, assume that both m∗ (A) and m∗ (B) are finite. Hence, by Theorem
2.1.3, given ε > 0 there exist {In } an {Jn } in IA and IB , respectively, such
that X ε X ε
`(In ) < m∗ (A) + , `(Jn ) < m∗ (B) + .
n
2 n
2
Then, the collection {In }∞ ∞
n=1 ∪{Jk }k=1 of open intervals cover A∪B. There-
fore,
X X
m∗ (A ∪ B) ≤ `(In ) + `(Jn )
n n

∗ ε  ∗ ε
≤ m (A) + + m (B) +
2 2
= m∗ (A) + m∗ (B) + ε.
This is true for all ε > 0, so that m∗ (A ∪ B) ≤ m∗ (A) + m∗ (B).
(iii) If E ⊆ A and m∗ (E) = 0, then by (i) and (ii),
m∗ (A) ≤ m∗ (E) + m∗ (A \ E) = m∗ (A \ E) ≤ m∗ (A).
Hence, m∗ (A \ E) = m∗ (A).

P Suppose E ⊆ R and x ∈ R. Given ε > 0, let {In } in IE be such


(iv)
that n `(In ) ≤ m∗ (E) + ε. Note that each In + x is an open interval and
E + x ⊆ ∪n (In + x). Hence,
X X
m∗ (E + x) ≤ `(In + x) = `(In ) = m∗ (E) + ε.
n n

This is true for every ε > 0. Hence, m∗ (E


+ x) ≤ m∗ (E). Since E =
(E + x) + (−x), it follows from the above that m∗ (E) ≤ m∗ (E + x). Thus
the proof is complete.

The property (i) in Theorem 2.1.6 is called the monotonicity


property of m∗ , and the property (iv) is called the translation
invariance of m∗ .

Making use of the monotonicity of m∗ , we deduce the following.


Corollary 2.1.7 Let E ⊆ R. Then there exists a set G which is a countable
intersection of open sets in R such that
E ⊆ G and m∗ (G) = m∗ (E).
18 Lebesgue Measure

Proof. By Corollary 2.1.4, for each n ∈ N, there exists an open set Gn


in R such that
1
E ⊆ Gn and m∗ (Gn ) ≤ m∗ (E) + .
n
Take G = n Gn . Then, E ⊆ G and m∗ (E) ≤ m∗ (G) ≤ m∗ (Gn ) for every
T
n ∈ N, so that
1
m∗ (G) ≤ m∗ (E) + ∀ n ∈ N.
n
Letting n tend to infinity, we obtain, m∗ (G) ≤ m∗ (E). Thus, we have proved
m∗ (G) = m∗ (E).

The following corollary is immediate from Theorem 2.1.6(ii).

Corollary 2.1.8 For subsets A1 , . . . , An of R,


n
[  Xn
m∗ Ak ≤ m∗ (Ak ).
k=1 k=1

More generally, we have the following theorem.

Theorem 2.1.9 Suppose Ak ⊆ R for k ∈ N. Then



[  ∞
X

m Ak ≤ m∗ (Ak ).
k=1 k=1

Proof. If m∗ (Ak ) = ∞ for some k ∈ N, then the result holds trivially.


Hence, assume that m∗ (Ak ) < ∞ for every k ∈ N. Then, by Theorem 2.1.3,
for each k ∈ N and ε > 0, there exists {Ik,n } ∈ IAk such that

X
`(Ik,n ) < m∗ (Ak ) + ε/2k .
n=1

Note that ∪∞ ∞
k=1 Ak ⊆ ∪k=1 ∪n Ik,n . Therefore,


[  X∞ X ∞
X ∞
X
m∗ Ak ≤ `(Ik,n ) ≤ (m∗ (Ak ) + ε/2k ) = m∗ (Ak ) + ε.
k=1 k=1 n k=1 k=1

This is true for all ε > 0. Hence the result follows.

The property of m∗ in Theorem 2.1.9 is called the countable


subadditivity of m∗ .
Lebesgue Outer Measure 19

Now comes a result that we are waiting for.


Theorem 2.1.10 The Lebesgue outer measure of any interval is its length.

Proof. Let I be an interval.


Case 1: Suppose I = [a, b] with −∞ < a < b < ∞, and let ε > 0. Let
Iε := (a − ε, b + ε). Then,

m∗ (I) ≤ m∗ (Iε ) ≤ b − a + 2ε.

This is true for all ε > 0. Hence, m∗ (I) ≤ b − a. Thus, it remains to show
that m∗ (I) ≥ b − a. For this, it is enough to show that
X
b−a≤ `(In ) ∀ {In } ∈ II , (∗)
n

because, in that case we can take infimum over all such {In } ∈ II and obtain
b − a ≤ m∗ (I). So, let {In } ∈ II . If `(In ) = ∞ for some n ∈ N, then (∗)
holds trivially. So, we may assume that each In is of finite length. By the
compactness of I, there exists a finite sub-collection {In1 , , . . . , Ink } of {In }
such that I ⊆ ∪ki=1 Ini . Let Ini := (ai , bi ) for i ∈ {1, . . . , k}. We may assume,
without loss of generality, that

(ai , bi ) ∩ [a, b] 6= ∅ and ai+1 < bi , i ∈ {1, . . . , k − 1}.

Then
k
X k
X k−1
X
`(Ini ) = (bi − ai ) = bk − a1 + (bi − ai+1 ) ≥ bk − a1 ≥ b − a.
i=1 i=1 i=1

Thus,
k
X X
b−a≤ `(Ini ) ≤ `(In ).
i=1 n

Thus, we have proved (∗).


Case 2: Suppose I is an interval of finite length, with end points a and b
with a < b, which is not necessarily a closed interval. Then for sufficiently
small ε > 0, we have [a + ε, b − ε] ⊆ I ⊆ [a − ε, b + ε]. Hence,

m∗ ([a + ε, b − ε]) ≤ m∗ (I) ≤ m∗ ([a − ε, b + ε])

so that by Case 1,

b − a − 2ε ≤ m∗ (I) ≤ b − a + 2ε.
20 Lebesgue Measure

Since this is true for every ε > 0, it follows that m∗ (I) = b − a.


Case 3: Suppose I is of infinite length. Then for every M > 0 there exists a
closed interval IM of length M such that IM ⊆ I. By Case 1, m∗ (IM ) = M .
Thus,
M = m∗ (IM ) ≤ m∗ (I)
for all M > 0 so that m∗ (I) = ∞.

Corollary 2.1.11 Every interval is an uncountable set.

Proof. Follows from Theorem 2.1.2(iii) and Theorem 2.1.10.

Can an uncountable set be of measure 0?

The answer is in affirmative as the following example of Cantor’s ternary


set or simply Cantor set shows.
Example 2.1.12 (Cantor ternary set) Let us first recall how the Cantor
ternary set is constructed. Consider the unit interval [0, 1]. Let C1 be the
set obtained from I after removing its “middle third” J1 := ( 31 , 23 ) from
C0 := [0, 1]. That is
h 1i h2 i
C1 = 0, ∪ ,1 .
3 3
Next, let C2 be the set obtained from C1 after removing the “middle thirds”
from each of the two subintervals in C1 . Let the removed set be J2 . Thus,
h 1i h2 1i h2 7i h8 i
C2 = 0, ∪ , ∪ , ∪ ,1 .
9 9 3 3 9 9
Continue this procedure to obtain C3 , C4 and so on. At the nth stage, we
obtain Cn = Cn−1 \ Jn , where Jn is the union of the middle thirds of each
of the subintervals in Cn−1 . Now the Cantor set C is defined by

\
C= Cn .
n=1

Note that
C1 ⊃ C2 ⊃ C3 ⊃ · · · .
and m∗ (C1 ) = 2/3, m∗ (C2 ) = (2/3)2 , m∗ (C3 ) = (2/3)3 , etc., and more
generally,  2 n
m∗ (Cn ) = , n ∈ N.
3
Lebesgue Measurable Sets 21

Hence,
k
\   2 k
m∗ (C) ≤ m∗ Cn = m∗ (Ck ) = ∀ k ∈ N.
3
n=1

Thus, m∗ (C) = 0. To see that C is an uncountable set,Pfirst we recall


∞ an
that every number a ∈ (0, 1] can be written as a series n=1 3n , where
an ∈ {0, 1, 2} and an 6= 0 for infinitely n ∈ N. For example, as per this, the
number 31 is represented as

1 2 2 2
= 2 + 3 + 4 + ···
3 3 3 3
2
and the number 3 is represented as

2 1 2 2 2
= + 2 + 3 + 4 + ··· .
3 3 3 3 3
Pk aj
More generally, if x = j=1 3j for some k ∈ N with ak 6= 0, then ak ∈ {1, 2},
and we represent a3kk as

ak ak − 1 X 2
k
= +
3 3k 3k+j
j=1

so that x has an infinite expansion. Taking such representations of a ∈ [0, 1],


it can be seen that, for each n ∈ N,

an ∈ {0, 2} ⇐⇒ a 6∈ Jn .

Hence,

n X bn o
C= b= : bn ∈ {0, 2}, n ∈ N .
3n
n=1

Thus, C is in one-one correspondence with the family of all sequence (bn )


with bn ∈ {0, 2}, and hence, C is an uncountable set. ♦

2.2 Lebesgue Measurable Sets


Suppose A1 and A2 are disjoint subsets of R. We may expect that

m∗ (A1 ∪ A2 ) = m∗ (A1 ) + m∗ (A2 ). (1)

Is it true for any two disjoint sets A1 and A2 ?


22 Lebesgue Measure

Suppose for a moment that (1) is true for any two disjoint sets A1 and
A2 . Then we also have
n
[  n
X

m Ai = m∗ (Ai ) (2)
i=1 i=1

for any pairwise disjoint sets A1 , . . . , An . Now, consider a denumerable


disjoint family {An }∞
n=1 , that is, An ∩ Am = ∅ whenever n 6= m. Then using
equality (2) above, monotonicity (Theorem 2.1.6 (i)), and the subadditivity
(Theorem 2.1.9) of m∗ , we obtain
n
X n
[  ∞
[  X∞
m∗ (Ai ) = m∗ Ai ≤ m∗ Ai ≤ m∗ (Ai )
i=1 i=1 i=1 i=1

for every n ∈ N, so that


n
X ∞
[  X∞
m∗ (Ai ) ≤ m∗ Ai ≤ m∗ (Ai )
i=1 i=1 i=1

for every n ∈ N. Taking limit as n → ∞, we obtain



[  X∞
m∗ Ai = m∗ (Ai ). (3)
i=1 i=1

We now show that (3) does not hold for certain denumerable disjoint family
{An }∞
n=1 of subsets of R.

Theorem 2.2.1 There exists a subset E of R such that if En := E + rn


for n ∈ N with {rn : n ∈ N} = [−1, 1] ∩ Q, then {En : n ∈ N} is a disjoint
family and
[∞ 
1 ≤ m∗ En ≤ 3.
n=1

In particular,

[  ∞
X

m En 6= m∗ (En ).
n=1 n=1

Proof. Consider a relation ∼ on R by defining

x ∼ y ⇐⇒ x − y ∈ Q.
Lebesgue Measurable Sets 23

It can be easily seen that ∼ is an equivalence relation on R. Hence, R is the


disjoint union of equivalence classes. Let E be the subset of [0, 1] such that
its intersection with each equivalence class is a singleton set. Such a set E
exists by using the axiom of choice on the collection

E := {[x] ∩ [0, 1] : x ∈ [0, 1]},

where [x] is the equivalence class of x. We note that if x ∼ y, then the


rational number r := x − y satisfies −1 ≤ r ≤ 1. Let {r1 , r2 , . . .} be the set
of all rational numbers in [−1, 1]. Let En := E + rn for n ∈ N. Then {En }
is a disjoint family. Indeed, if En ∩ Em 6= ∅ for some n 6= m, then there
exists x, y ∈ E such that x + rn = y + rm so that x ∼ y; consequently, x = y
so that rn = rm , which is not possible. Further, we have

[
[0, 1] ⊆ En ⊆ [−1, 2]
n=1

so that

[ 
1 ≤ m∗ En ≤ 3.
n=1

Since m∗ (En ) = m∗ (E + rn ) = m∗ (E) for all n ∈ N, the particular case also


follows.

By the above theorem, the relation (3), and hence the relation (1) does
not hold for some disjoint family of sets involved. Thus, we have have also
proved the following result.

Theorem 2.2.2 There exist disjoint subsets A1 and A2 of R such that

m∗ (A1 ∪ A2 ) < m∗ (A1 ) + m∗ (A2 ).

The above discussion motivates us to look for a family of sets in which


the relation (1), and hence (2) and (3) hold for all possible disjoint family
of sets involved.
Definition 2.2.3 A set E ⊆ R is said to be Lebesgue measurable if for
every A ⊆ R,
m∗ (A) = m∗ (A ∩ E) + m∗ (A ∩ E c ).
We denote the set of all Lebesgue measurable sets by M. ♦
The proof of the following theorem is easy and hence left as exercise.
24 Lebesgue Measure

Theorem 2.2.4 The following results hold.

(i) E ∈ M ⇐⇒ m∗ (A) ≥ m∗ (A ∩ E) + m∗ (A ∩ E c ) ∀ A ⊆ R.

(ii) ∅ ∈ M.

(iii) E ∈ M ⇒ E c ∈ M.

(iv) m∗ (E) = 0 ⇒ E ∈ M.

Since countable sets are of zero outer measure, Theorem 2.2.4 implies:

M contains all countable sets. In particular, Q ∈ M and Qc ∈ M.

Theorem 2.2.5 Let A1 , A2 be subsets of R such that A1 ∩ A2 = ∅. If one


of A1 , A2 belongs to M, then

m∗ (A1 ∪ A2 ) = m∗ (A1 ) + m∗ (A2 ).

Proof. Suppose A1 ∩ A2 = ∅. Assume A1 ∈ M. Then

m∗ (A1 ∪ A2 ) = m∗ ((A1 ∪ A2 ) ∩ A1 ) + m∗ ((A1 ∪ A2 ) ∩ Ac1 ).

Since (A1 ∪A2 )∩A1 ) = A1 and (A1 ∪A2 )∩Ac1 ) = A2 , we obtain the result.

The following corollary is immediate from the above theorem.

Corollary 2.2.6 If {A1 , . . . , An } is a disjoint family in M, then


n
[  n
X

m Ai = m∗ (Ai ).
i=1 i=1

The proof of the following theorem is along the same lines as we have
deduced (3) from (2). However, we give its details here as well.

Theorem 2.2.7 Let {An : n ∈ N} be a disjoint family in M. Then



[  ∞
X

m An = m∗ (An ).
n=1 n=1
Lebesgue Measurable Sets 25

Proof. If {An } is a finite family, then by Corollary 2.2.6,


n
[  n
X

m Ai = m∗ (Ai )
i=1 i=1

for every n ∈ N. Hence, by using the monotonicity of m∗ , we have



X ∞
[  n
[  n
X
∗ ∗ ∗
m (An ) ≥ m An ≥ m Ai = m∗ (Ai )
n=1 n=1 i=1 i=1

for all n ∈ N. Letting n tend to infinity, the result follows.

The property of m∗ given in Theorem 2.2.7 is called countable


additivity of m∗ on M.

Corollary 2.2.8 There exists E ⊆ R which does not belong to M. In


particular, M is not the whole of 2R , the power set of R.

Proof. Follows from Theorem 2.2.7 and Theorem 2.2.9.

The following theorem shows that non-measurable sets are plenty.


Theorem 2.2.9 Let A ⊆ R be such that m∗ (A) > 0. Then there exists a
subset E ⊆ A such that E 6∈ M.

Proof. First let us assume that A is a bounded set. We consider the


equivalence relation ∼ on R as in the proof of Theorem 2.2.9, that is,
x ∼ y ⇐⇒ x − y ∈ Q.
Let E be the subset of A such that its intersection with each equivalence
class [x] with x ∈ A is a singleton set. Such a set E exists by using the
axiom of choice on the collection
E := {[x] ∩ A : x ∈ A},
where [x] is the equivalence class of x. Let a = inf A and b = sup A. Then
A ⊆ [a, b]. We note that if x ∼ y, then the rational number r := x − y
satisfies a − b ≤ r ≤ b − a. Let {r1 , r2 , . . .} be the set of all rational numbers
in [a − b, b − a]. Let En := E + rn for n ∈ N. Then {En } is a disjoint family
and, for every x ∈ E and n ∈ N, 2a − b ≤ x + rn ≤ 2b − a so that

[
A⊆ En ⊆ [2a − b, 2b − a].
n=1
26 Lebesgue Measure

Hence,

[ 
0 < m(A) ≤ m∗ En ≤ 3(b − a).
n=1
Since ∗ m∗ (E m∗ (E)
for all n ∈ N, it follows that
 Sm (En ) = + rn ) =
∞ P∞
m∗ n=1 En

6= n=1 m (En ). In particular E 6∈ M.
Next, suppose that A is not bounded. Then for every α > 0, Aα :=
A∩[−α, α] is a bounded set. Since A = ∪α>0 Aα and m∗ (A) > 0, there exists
β > 0 such that m∗ (Aβ ) > 0. By the arguments in the above paragraph,
there exists Eβ ⊆ Aβ ⊆ A such that Eβ 6∈ M.
This completes the proof.

We have seen that M contains all countable subsets of R. Since Cantor


ternary set is of zero outer measure, it also belongs to M. Now we show w
that M contains a lot more subsets of R.
Theorem 2.2.10 Let A1 , A2 ∈ M. Then A1 ∪ A2 ∈ M. More generally, if
{A1 , . . . , An } ⊆ M, then ∪ni=1 Ai ∈ M.

Proof. Let A ⊆ R. We have to show that


m∗ (A) ≥ m∗ (A ∩ (A1 ∪ A2 )) + m∗ (A ∩ Ac1 ∩ Ac2 ) . (∗)
Note that A ∩ (A1 ∪ A2 ) = (A ∩ A1 ) ∪ (A ∩ A2 ∩ Ac1 ), so that
m∗ (A ∩ (A1 ∪ A2 )) ≤ m∗ (A ∩ A1 ) + m∗ (A ∩ A2 ∩ Ac1 ).
Therefore, the right hand side of (∗) is less than or equal to
m∗ (A ∩ A1 ) + m∗ (A ∩ A2 ∩ Ac1 ) + m∗ (A ∩ Ac1 ∩ Ac2 ) . (∗∗)
Now, since A2 ∈ M, we get
m∗ (A ∩ A2 ∩ Ac1 ) + m∗ (A ∩ Ac1 ∩ Ac2 ) = m∗ (A ∩ Ac1 ).
Thus, the expression in (∗∗) is les than or equal to
m∗ (A ∩ A1 ) + m∗ (A ∩ Ac1 )
which is equal to m∗ (A), since A1 ∈ M. Thus,
m∗ (A ∩ (A1 ∪ A2 )) + m∗ (A ∩ Ac1 ∩ Ac2 ) ≤ m∗ (A)
which completes the proof of (∗). The last part follows by repeated appli-
cation of the first part.
Lebesgue Measurable Sets 27

Corollary 2.2.11 If A1 , A2 ∈ M, then A1 \ A2 ∈ M.

Proof. Let A1 , A2 ∈ M. Since

A1 \ A2 = A1 ∩ Ac2 = (Ac1 ∪ A2 )c ,

the result follows from Theorem 2.2.10, as compliment of any element in M


is an element in M.

By Theorem 2.2.10 we can say that M is closed under finite unions. Next
we show that M is closed under countable unions. For this purpose we shall
make use of the following lemma which is more general than Theorem 2.2.5.

Lemma 2.2.12 Let {A1 , . . . , An } be a disjoint family in M. Then for any


A ⊆ R,
 [ n  X n

m A∩ Ai = m∗ (A ∩ Ai ).
i=1 i=1

Proof. Let n = 2. Since A1 ∈ M,

m∗ (A ∩ (A1 ∪ A2 )) = m∗ (A ∩ (A1 ∪ A2 ) ∩ A1 ) + m∗ (A ∩ (A1 ∪ A2 ) ∩ Ac1 ).

But,

A ∩ (A1 ∪ A2 ) ∩ A1 = A ∩ A1 , A ∩ (A1 ∪ A2 ) ∩ Ac1 = A ∩ A2 .

Hence, we have

m∗ (A ∩ (A1 ∪ A2 )) = m∗ (A ∩ A1 ) + m∗ (A ∩ A2 ).

Thus, the result is proved for n = 2. The result for general n follows by
induction.

Now we prove the result that we had promised.

Theorem 2.2.13 If En ∈ M for n ∈ N, then ∞


S
n=1 En ∈ M.

S
Proof. Let E = En , where En ∈ M for n ∈ N, and let A ⊆ R. We
have to show that

m∗ (A) ≥ m∗ (A ∩ E) + m∗ (A ∩ E c ) . (∗)
28 Lebesgue Measure

S
We write E as a disjoint union E = An where A1 = E1 and for n ≥ 2,

An = En \ ∪n−1
i=1 Ei .

Let Fn = ∪ni=1 Ai . Note that Fn ∈ M, by Theorem 2.2.10 and Corollary


2.2.11. Hence,
m∗ (A) ≥ m∗ (A ∩ Fn ) + m∗ (A ∩ Fnc ) .
Now, Lemma 2.2.12 implies
n
X
m∗ (A ∩ Fn ) = m∗ (A ∩ Ai )
i=1

and the relation Fnc ⊇ Ec implies m∗ (A ∩ Fnc ) ≥ m∗ (A ∩ E c ). Thus,


n
X
m∗ (A) ≥ m∗ (A ∩ Ai ) + m∗ (A ∩ E c ) .
i=1

This is true for all n ∈ N. Letting n tend to infinity,



X
m∗ (A) ≥ m∗ (A ∩ Ai ) + m∗ (A ∩ E c )
i=1
S∞ S∞
Therefore, using the fact that i=1 (A ∩ Ai ) = A ∩ i=1 Ai and the subad-
ditivity of m∗ ,

X
m∗ (A) ≥ m∗ (A ∩ Ai ) + m∗ (A ∩ E c )
i=1

[
≥ m∗ (A ∩ Ai ) + m∗ (A ∩ E c )
i=1
= m∗ (A ∩ E) + m∗ (A ∩ E c ) .

Thus the proof is complete.


T∞
Corollary 2.2.14 If En ∈ M for n ∈ N, then n=1 En ∈ M.

Proof. Let En ∈ M for n ∈ N. By De Morgan’s law,



\ ∞
[ c
En = Enc .
n=1 n=1

Hence, the result follows from Theorem 2.2.13 together with the fact that
E ∈ M implies E c ∈ M.
Lebesgue Measurable Sets 29

Definition 2.2.15 The function m∗ restricted to M is called the Lebesgue


measure on R, and it is denoted by m. For E ∈ M, m(E) := m∗ (E) is
called the Lebesgue measure of E. ♦
Let us list some of the important properties of the family M that we
have proved:

∅ ∈ M;
E ∈ M ⇒ E c ∈ M; [ [ X
{Ei } disjoint family in M ⇒ Ei ∈ M & m∗ ( Ei ) = m∗ (Ei )
i i i

Remark 2.2.16 We may observe that, for a set A ∈ M, the family

MA := {E ⊆ A : E ∈ M}

also has the properties listed in the above box by replacing M by MA . ♦

The following theorem shows that the class M is very large.

Theorem 2.2.17 The following results hold.

(i) For any a ∈ R, the intervals (a, ∞), [a, ∞), (−∞, a) and (−∞, a]
belong to M.

(ii) For any a, b ∈ R with a < b, the intervals (a, b) and [a, b] belong to M.

(iii) Open subsets of R belong to M.

(iv) Closed subsets of R belong to M.

(v) Countable intersections of open sets belong to M.

(vi) Countable unions of closed sets belong to M.

Proof. We first prove that (a, ∞) ∈ M for any a ∈ R, and then deduce
other results by using some of the properties of m∗ . So, let a ∈ R and
E = (a, ∞). Let A ⊆ R and ε > 0. By Theorem 2.1.3, there exists a
countable family {In } of open intervals such that
[ X
A⊆ In , `(In ) ≤ m∗ (A) + ε.
n n
30 Lebesgue Measure

Note that
[ [
A∩E ⊆ (In ∩ (a, ∞)), A ∩ Ec ⊆ (In ∩ (−∞, a]).
n n

Hence, taking In0 := In ∩ (a, ∞) and In00 := In ∩ (−∞, a], we have


X X
m∗ (A ∩ E) + m∗ (A ∩ E c ) ≤ m∗ (In0 ) + m∗ (In00 )
n n
X
= [m∗ (In0 ) + m∗ (In00 )].
n

Note that In0 and In00 are intervals such that In0 ∩ In00 = ∅ and In0 ∪ In00 = In so
that
m∗ (In0 ) + m∗ (In00 ) = `(In0 ) + `(In00 ) = `(In ).
Thus, we have proved that
X
m∗ (A ∩ E) + m∗ (A ∩ E c ) ≤ `(In ) ≤ m∗ (A) + ε.
n

This is true for any ε > 0, so that we get


X
m∗ (A ∩ E) + m∗ (A ∩ E c ) ≤ `(In ) ≤ m∗ (A).
n

Hence, (a, ∞) = E ∈ M.
Next we observe that for any a ∈ R,
∞ 
\ 1 
[a, ∞) = a− ,∞ ,
n
n=1

(−∞, a) = R \ [a, ∞), (−∞, a] = R \ (a, ∞),


and for any a, b ∈ R with a < b,

(a, b) = (a, ∞) ∩ (−∞, b), [a, b] = [a, ∞) ∩ (−∞, b].

Thus, we have proved (i) and (ii). Now, using the facts that M is closed
under countable unions, countable intersections and complementation, and
the fact that every open subset of R is a countable union of open intervals,
the results listed in (iii)-(vi) follow.

In view of the results (v) and (vi) in Theorem 2.2.17, we recall the
following definition from Real Analysis.
Lebesgue Measurable Sets 31

Definition 2.2.18 A subset of R is said to be a Gδ set if it is a countable


intersection of open sets, and a subset of R is said to be an Fσ set if it is a
countable union of closed sets. ♦
Since a subset of R is closed if and only if its complement is open, using
De Morgan’s law, it follows that a subset of R is a Gδ set if and only if its
complement is an Fσ set.
Note that a Gδ -set need not be open and an Fσ set need not be closed.
For example, the set [0, 1) is neither open nor closed, but it is both a Gδ -set
and an Fσ set, as
∞  ∞ h
\ 1  [ n i
[0, 1) = − ,1 , [0, 1) = 0, .
n n+1
n=1 n=1

In fact, every interval is both Gδ -set and Fσ -set.


Now, we prove a companion result to Corollary 2.1.4 and Corollary 2.1.8,
which also asnwers the question raised in Remark 2.1.5.
Theorem 2.2.19 Let E ⊆ R. Then the following are equivalent.

(i) E ∈ M.

(ii) For every ε > 0, there exists an open set G in R such that

E ⊆ G and m∗ (G \ E) ≤ ε.

(iii) There exists a Gδ -set set G in R such that

E ⊆ G and m∗ (G \ E) = 0.

Proof. (i)⇒ (ii): Suppose E ∈ M and ε > 0 be given. First, let us


assume that m∗ (E) < ∞. By Corollary 2.1.4, there exists an open set G in
R such that
E ⊆ G and m∗ (G) ≤ m∗ (E) + ε.
Since E and G are in M and G \ E = G ∩ E c is also in M. Therefore,

m∗ (E) + m∗ (G \ E) = m∗ (G) ≤ m∗ (E) + ε.

Since m∗ (E) < ∞, we obtain m∗ (G \ E) ≤ ε. For the case when m∗ (E)


is not necessarily finite, we write E = ∪∞ ∗
n=1 En , where m (En ) < ∞ for
every n ∈ N. For example, we can take En := E ∩ [−n, n] for n ∈ N.
Since m∗ (En ) < ∞ for each n ∈ N, by the above part, there exists open set
32 Lebesgue Measure

Gn ⊇ En such that m∗ (Gn \ En ) < ε/2n . Taking G = ∪∞


n=1 Gn , we have G
is open, G ⊇ E and

G \ E = [∪∞ ∞ ∞
n=1 Gn ] \ [∪n=1 En ] ⊆ ∪n=1 (Gn \ En ).

Therefore,
∞ ∞
X X ε
m∗ (G \ E) ≤ m∗ (Gn \ En ) ≤ = ε.
2n
n=1 n=1

Thus, (ii) holds.


(ii)⇒ (iii): Assume (ii). Then, for every n ∈ N, there exists an open set
Vn in R such that
1
E ⊆ Vn and m∗ (Vn \ E) ≤ .
n
T
Let G = n Vn . Then E ⊆ G ⊆ Vn and G \ E ⊆ Vn \ E for all n ∈ N so that

1
m∗ (G \ E) ≤ ∀ n ∈ N.
n
Letting n tend to infinity, we obtain m∗ (G \ E) = 0. Thus, (iii) holds.
(iii)⇒ (i): Assume (iii). Then there exists a Gδ -set set G in R such that
E ⊆ G and m∗ (G \ E) = 0. Therefore, G \ E ∈ M. We know that G ∈ M
so that by Corollary 2.2.11,

E = G \ (G \ E) ∈ M.

This completes the proof.

Corollary 2.2.20 Let E ⊆ R. Then E ∈ M if and only if there exist a


Gδ -set G and an Fσ -set F such that

F ⊆E⊆G and m∗ (G \ F ) = 0.

Proof. Suppose E ∈ M. Then by Theorem 2.2.19, there exists a Gδ set


G such that
E ⊆ G and m∗ (G \ E) = 0.
Also, by taking E c in place of E, there exists a Gδ -set H such that E c ⊆ H
and m∗ (H \ E c ) = 0. Then F := H c is an Fσ set satisfying

F = Hc ⊆ E and E \ F = E \ H c = E ∩ H = H \ E c
Problems 33

so that m∗ (E \ F ) = m∗ (H \ E c ) = 0. Since G \ F = (G \ E) ∪ (E \ F ) we
obtain m∗ (G \ F ) = 0.
Conversely, suppose that there exists a Gδ -set G and an Fσ -set F such
that F ⊆ E ⊆ G and m∗ (G \ F ) = 0. In particular, G \ E ⊆ G \ F so
that m∗ (G \ E) = 0 and hence G \ E ∈ M. Therefore, by Corollary 2.2.11,
E = G \ (G \ E) ∈ M.

Using Theorem 2.2.19, we obtain the following theorem. The details of


its proof are left as an exercise.

Theorem 2.2.21 Let E ⊆ R. Then the following are equivalent.

(i) E ∈ M.

(ii) For every ε > 0, there exists a closed set F in R such that

F ⊆ E and m∗ (E \ F ) ≤ ε.

(iv) There exists an Fσ -set set F in R such that

F ⊆ E and m∗ (E \ F ) = 0.

2.3 Problems
1. Prove that, in Definition 2.1.1, m∗ (E) remains the same if we take IE to be
the collection of all countable family {In } of intervals where In ’s are intervals
of finite length, not necessary open intervals.
2. Show that, if E ⊆ A and m∗ (E) = 0, then m∗ (A ∪ E) = m∗ (A).
3. Show that, for every a, b ∈ R with a < b, the intervals [a, b), (a, b], [a, b] are
Gδ -sets, and [a, b), (a, b], (a, b) are Fσ -sets .
4. From Theorem 2.1.9, deduce that outer measure of every countable set is 0.
5. Deduce Corollary 2.1.8 from Theorem 2.1.9
6. Justify the statement: There exists a countable disjoint family {En } of sub-
sets of R such that
[∞  X ∞
m∗ En 6= m∗ (En ).
n=1 n=1

7. If E is a subset of an interval I such that m∗ (E) = 0, then prove that E c is


dense in I.
8. Prove Theorem 2.2.4.
34 Lebesgue Measure

9. Suppose {A1 , . . . , An } is a disjoint family of subsets of R such that A1 , . . . , An−1


belong to M. Then show that
n
[  Xn
m∗ Ai = m∗ (Ai ).
i=1 i=1

f = M∪{E0 }, where E0 6∈ M. Prove that if {An } is a countable disjoint


10. Let M
family in M,
f then
[  X
m∗ An = m∗ (An ).
n n

11. Justify the statement: There exists a countably infinite disjoint family N of
subsets of R such that N ∩ M = ∅.
12. Deduce Corollary 2.2.6 from Theorem 2.2.7.
13. Show that if E ∈ M, then

m(E) = inf{m(G) : G open E ⊆ G}.

14. Supply details of the proof of Theorem 2.2.21.


15. If E ⊆ R such that with m∗ (E) < ∞. Prove that the following are equivalent:
(a) E ∈ M.
(b) There exists a Gδ set G ⊇ E such that E = G \ E0 , where m∗ (E0 ) = 0.
(c) There exists an Fσ set F ⊆ E such that E = F ∪E0 , where m∗ (E0 ) = 0.
16. If A, B ∈ M with A ⊆ B and if m(B) < ∞, then show that

m(B \ A) = m(B) − m(A).

17. If A, B ∈ M, prove that

m(A ∪ B) + m(A ∩ B) = m(A) + m(B).

18. Find an open dense subset A of R with m(A) < ∞.


[Hint: Use density of the set of rational numbers.]
19. Find a closed subset B of R with empty interior and with m(B) = ∞.
20. Find a subset E of [0, 1] which is a countable union of closed nowhere dense
sets such that m(E) = 1.
[Hint: Use last exercise. Recall that a set A is said to be nowhere dense if its
closure has empty interior.]
21. Verify the statement in Remark 2.2.16.
22. Prove Theorem 2.2.21.
3

Measure and Measurable Functions

In the last chapter we defined the concept of measure of certain


subsets of the real line. These subsets include most of the usual
sets that we deal with in anlaysis, such as intervals, open sents,
closed sets and so on. We have also seen, that not every subset of
R can be measured in a satisfactory manner. However, the sets
that can be measured follow certain nice set theoretic rules. In
this chapter we build the measure theory on an arbitrary set, by
making use of the essential properties that measureable subsets
of R satisfy.

3.1 Measure on an Arbitrary σ-Algebra


Let M be the class of all Lebesgue measurable subsets of R and let m be
the Lebesgue measure on R, that is, the restriction of the Lebesgue outer
measure m∗ to M. Recall the following properties:

1. R ∈ M,

2. E ∈ M implies E c ∈ M,
S
3. En ∈ M for n ∈ N implies n En ∈ M,, and
[  X
4. {En } is a countable disjoint family in M implies m En = m(En ).
n n

Motivated by the above properties of the family M of all measurable


sets of R and the Lebesgue measure m on R, we introduce the following two
definitions.

35
36 Measure and Measurable Functions

Definition 3.1.1 Let X be a set and A be a family of subsets of X. Then


A is called a σ-algebra on X if

(a) X ∈ A,

(b) A ∈ A implies Ac := X \ A ∈ A,
S
(c) An ∈ A for n ∈ N implies n An ∈ A.

The pair (X, A) is called a measurable space, and members of A are called
measurable sets. ♦

If we replace the countable union in the condition (c) in the above defi-
nition with finite unions, then the resulting family is called an algebra. We
shall use the concept of an algebra in Chapter 6.
Let us first make an easy observation.

Theorem 3.1.2 Let X be a set and A be a σ-algebra on X.

(i) A, B ∈ A implies A \ B ∈ A,
T
(ii) An ∈ A for n ∈ N implies n An ∈ A.

Proof. For subsets A, B, A1 , A2 , . . . of X, we have


\  [ c
A \ B = A ∩ B c = (Ac ∪ B)c and An = Acn ∈ A.
n n

Hence the results in (i) and (ii) follow from the definition of the σ-algebra.

Definition 3.1.3 Let (X, A) be measurable space. Then a function


µ : A → [0, ∞] is called a measure on (X, A) if

(a) µ(∅) = 0, and

(b) for any countable disjoint family {An } in A,


[  X
µ An = µ(An ).
n n

The triple (X, A, µ) is called a measure space. For A ∈ A, µ(A) is called


the measure of A. ♦
Measure on an Arbitrary σ-Algebra 37

Convention: In the due course, we adopt the following convention:

• If the σ-algebra A is understood from the context, then instead of


saying that (X, A) is a measurable space, we may say that X is a
measurable space.

• If the σ-algebra A and the measure µ are understood from the context,
then instead of saying that (X, A, µ) is a measure space, we may say
that X is a measure space.

• If the σ-algebra A is understood from the context, then instead of


saying that µ is a measure on the measurable space (X, A), we may
say that µ is a measure on X.

Remark 3.1.4 A measure as in Definition 3.1.3 is called a positive mea-


sure, since there are notions such as signed measures and complex measures.
In this first course we do not consider such measures. Hence, we use the
terminology measure for the positive measure. ♦

Theorem 3.1.5 Let (X, A, µ) be a measure space and A, B ∈ A. Then

(i) A ⊆ B implies µ(A) ≤ µ(B),

(ii) A ⊆ B and µ(A) < ∞ imply µ(B \ A) = µ(B) − µ(A).

Proof. Suppose A, B ∈ A such that A ⊆ B. Then B is the disjoint union


of A and B \ A. By Theorem 3.1.2, B \ A ∈ A, and by the definition of the
measure,
µ(B) = µ(A) + µ(B \ A).
From this (i) and (ii) follow.

The property (i) in Theorem 3.1.5 is called the monotonicity


property of µ.

Definition 3.1.6 Let (X, A, µ) be a measure space.

(a) If µ(X) = 1, then µ is called a probability measure.

(b) If µ(X) < ∞, then µ is called a finite measure, and (X, A, µ) is


called a finite measure space.
38 Measure and Measurable Functions

(c) If there exists Xn ∈ A, n ∈ N such that X = ∞


S
n=1 Xn and µ(Xn ) < ∞
for each n ∈ N, then µ is called a σ-finite measure, and (X, A, µ) is
called a σ-finite measure space. 

The assertive statements in the following examples may be verified by


the reader.
Example 3.1.7 The triple (R, M, m) is measure space, and it is a σ-finite
measure space. ♦
Example 3.1.8 For A ∈ M, MA defined as in Remark 2.2.16 is a σ-algebra,
and the map mA defined by
mA (E) := m(E), E ∈ MA ,
is a σ-finite measure on (A, MA ). ♦
Example 3.1.9 For any set X, the family {∅, X} and the power set 2X
are σ-algebras, and they are the smallest and largest (in the sense of set
inclusion), respectively, of all σ-algebras on X. ♦
Example 3.1.10 Given any any measurable space (X, A), the function µ
defined by µ(A) = 0 for all A ∈ A is a measure on (X, A). This measure is
called the zero measure on X. ♦
Example 3.1.11 If (X, A, µ) is a measure space, then for every α ∈ R, α >
0,
E 7→ αµ(E), E ∈ A,
defines a measure on (X, A), and if 0 < µ(X) < ∞, then
µ(E)
E 7→ , E ∈ A,
µ(X)
is a probability measure on (X, A). ♦
Example 3.1.12 Let X be a non-empty set and A = {∅, X}. Let µ be
defined by µ(∅) = 0 and µ(X) = 1. Then (X, A, µ) is a measure space. ♦
Example 3.1.13 Let X be any set, A = 2X , the power set of X, and µ be
defined by  #
E if E is a finite set
µ(E) =
∞ if E is an infinite set.
Here, E # denotes the number of elements in E. Then µ is a measure space
on (X, A). This measure is called the counting measure on X. Note that
the counting measure on X is
Measure on an Arbitrary σ-Algebra 39

1. finite if and only if X is a finite set and

2. σ-finite if and only if X is a countable set.


Example 3.1.14 Let X be any set, A be the power set of X, x0 ∈ X, and
µ be defined by 
1 if x0 ∈ E
µ(E) =
0 if x0 6∈ E.
Then (X, A, µ) is a measure space, and the measure µ is called the Dirac
measure on X centered at x0 . ♦

Let us have an illustration of the Dirac measure: Think of a thin


wire of negligible weight. Suppose a bead of weight 1 unit kept
at some point x0 on the wire. Then weight of a part, say E, of
the wire is 1 if x0 belongs to E, and the weight is zero if x0 does
not belong to E.

Example 3.1.15 Let X be a set, A be the power set of X, xi ∈ X and


wi ≥ 0 for i ∈ {1, . . . , k}. For E ⊆ X, let ∆E = {i : xi ∈ E}, and let µ be
defined by X
µ(E) = wi .
i∈∆E

Then (X, A,Pµ) is a measure space. Note that µ is a probability measure if


and only if ki=1 wi = 1. ♦
Analogous to the illustration of the Dirac measure we have the following
corresponding to Example 3.1.15:

Think of a thin wire of negligible weight. Suppose a finite num-


ber of beads of weights w1 , . . . , wk are kept at points x1 , . . . , xk
respectivelyPon the wire. Then the weight of a part, say E, of
the wire is i∈∆E wi where ∆E = {i : xi ∈ E}.

Another illustration of Example 3.1.15:

Let us assume that the wire costs nothing, but the beads at
x1 , . . . , xk cost rupees w1 , . . . , wk respectively. Then the cost
of a part, say E, of the wire together with the beads on it is
P
i∈∆E wi where ∆E = {i : xi ∈ E}.
40 Measure and Measurable Functions

An important example of a measure which is closely related to the


Lebesgue measure on R is the Lebesgue measure on Rk for k ∈ N with
k ≥ 2.

3.2 Lebesgue measure on Rk


Before giving the definition, let us introduce a few notations: Let k ∈ N
with k ≥ 2 and ai , bi ∈ R with ai < bi for i = 1, . . . , k. Then the volume of
the open rectangle
R := (a1 , b1 ) × · · · × (ak , bk )
is the number
vol(R) := (b1 − a1 ) × · · · × (bk − ak ).

Definition 3.2.1 For E ⊆ Rn , the Lebesgue outer measure of E is


defined as X
m∗k (E) := inf vol(Rn ),
n

where the infimum is taken over the collection of all countable family {Rn }
of open rectangles which covers E, that is, E ⊆ ∪n Rn . ♦

It can be shown, as in the case of Lebesgue outer measure m∗ , that

1. m∗k (∅) = 0;

2. A ⊆ Rk , B ⊆ Rk , A ⊆ B ⇒ m∗k (A) ≤ m∗k (B);

3. An ⊆ Rk , n ∈ N ⇒ m∗k ( ∞
S P∞ ∗
n=1 An ) ≤ n=1 mk (An ).

Further, the family Mk of all sets E ⊆ Rk which satisfy

m∗k (A) = m∗k (A ∩ E) + m∗k (A ∩ E c )

for all A ⊆ Rk is a σ-algebra, and the map

E 7→ mk (E) := m∗k (E)

is a measure on Mk , called the Lebesgue measure on Rk .


Remark 3.2.2 Note that the Lebesgue measure mk on Rk is obtained by
restricting the map E 7→ m∗k (E) to the σ-algebra Mk . Likewise, we now
describe a procudere of obtaining a measure on a set X by first defining a
Generated σ-Algebra and Borel σ-Algebra 41

general outer measure µ∗ on all subsets of X and then restricting it to a


general class of measurable sets constructed out of µ∗ .
Let X be a set. A function µ∗ : 2X → [0, ∞] is called an outer measure
on X if the following three conditions are satisfied:

(i) µ∗ (∅) = 0,

(ii) A ⊆ B ⇒ µ∗ (A) ≤ µ∗ (B), and


S  P
(iii) For every countable family {An } in 2X , µ∗ n n ≤
A ∗
n µ (An ).

A set E ⊆ X is said to be µ∗ -measurable if for every A ⊆ X,

µ∗ (A) = µ∗ (A ∩ E) + µ∗ (A ∩ E c )}.

It can be shown that

• the family A of all µ∗ -measurable subsets of X is a σ-algebra, and

• the map µ := µ∗ |A , the restriction of µ∗ to A, is a measure on A.

The above construction of measurable sets out of a general outer measure


is due to Caratheodory (see e.g. [?]). ♦

3.3 Generated σ-Algebra and Borel σ-Algebra


We know that an arbitrary family of subsets of a non-empty set X is not a
σ-algebra. However, an arbitrary family of subsets of X is associated with
a unique σ-algebra in a certain way.

Theorem 3.3.1 Let X be a set and S be a family of subsets of X. Then


the intersection of all σ-algebras containing S is a σ-algebra on X, and it is
the smallest σ-algebra on X containing S.

Proof. Let F be the family of all σ-algebras on X containing S, that is,

F = {A : A is a σ-algebra on X and S ⊆ A}.


T
Let AS be the intersection of all members of F, that is, AS = A∈F A.
First we show that AS is a σ-algebra on X.
42 Measure and Measurable Functions

Since X ∈ A for every A ∈ F, we have X ∈ AS . Now, let A ∈ AS .


Then A ∈ A for every A ∈ F. Since each A is a σ-algebra, Ac ∈ A for every
A ∈ F. Hence, A ∈ AS . Next, let {An : n ∈ Λ} be a countable family in
AS . Then {An : n ∈ Λ} ⊆ A for every A ∈ F. Since each A is a σ-algebra,
∪n∈Λ An ∈ A for every A ∈ F. Therefore, ∪n∈Λ An ∈ AS . Thus, AS is a
σ-algebra.
Since S ⊆ A for every A ∈ F, and AS is the intersection of members
of F, we have S ⊆ AS , and AS is the smallest (in terms of set inclusion)
σ-algebra containing S.

Corollary 3.3.2 Intersection of any family of σ-algebras on a set X is


again a σ-algebra on X.

Proof. If {Aα : α ∈ Λ} is a family of σ-algebras on a set X, where Λ


is some index set, then taking taking
T S as either {∅} or {X} or {∅, X} in
Theorem 3.3.1, we obtain that α∈Λ Aα is a σ-algebra on X.

Definition 3.3.3 Let X be a set and S be a family of subsets of X. The


smallest σ-algebra containing S is called the σ-algebra generated by S,
and is denoted by AS . ♦

If the set under consideration is a topological space, there is a natural


σ-algebra associated with it.
Definition 3.3.4 Let (Y, T ) be a topological space. Then the σ-algebra
generated by the topology T is called the Borel σ-algebra on Y , or the
Borel field on Y , and the members of the Borel σ-algebra are called Borel
sets. ♦
We may recall that, a topological space is a pair (Y, T ) consisting of a set
Y together with a topology T , that is a family T of subsets of Y , satisfying
the following properties:

(1) {∅, X} ⊆ T ,

(2) A, B ∈ T ⇒ A ∩ B ∈ T ,
[
(3) S ⊆ T ⇒ A∈T
A∈S

Recall also that, given a topological space (Y, T ), the members of the topol-
ogy T are called open sets in Y , and a set A ⊆ Y is called a closed set if
Restrictions of σ-Algebras and Measures 43

Ac is an open set. If the topology T on Y is understood from the context,


then instead of saying “(Y, T ) is a topological space”, we say that “Y is a
topological space”.
We shall denote the Borel σ-algebra on a topological space Y by BY . If
Y = Rk , then we shall denote BRk by Bk . Note that m|B1 is a measure on
(R, B1 ). Also, by the observations after Definition 3.2.1, m|Bk is a measure
on (R, Bk ).
Since M contains all open sets and B1 is the smallest σ-algebra containing
all open sets, we have B1 ⊆ M. We shall see that B1 is a proper subfamily
of M (Theorem 3.8.5).

3.4 Restrictions of σ-Algebras and Measures


Recall from Example 3.1.8 that if A ∈ M, then MA is a σ-algebra on A and
and the map mA : E 7→ m(E) measure on (A, MA ). The s-algebra MA and
the measure mA can be thought of as restrictions of M and m, respectively.
Analogously, we can define restrictions of a general σ-algebra and a measure
on it to a measurable subset.
First we may observe the following easily verifiable fact.
Theorem 3.4.1 If (X, A, µ) is a measure space, X0 ∈ A and A0 is σ-
algebra on X0 such that A0 ⊆ A, then the restriction of µ to A0 , that is,
µ0 : A0 → [0, ∞] defined by
µ0 (E) = µ(E), E ∈ A0 ,
is a measure on (X0 , A0 ).
Theorem 3.4.2 Suppose (X, A) is a measurable space and X0 ∈ A. Then
A0 = {A ∩ X0 : A ∈ A}
is a σ-algebra on X0 . Further, if µ is a measure on (X, A), then the restric-
tion of µ to A0 , that is, µ0 : A0 → [0, ∞] defined by
µ0 (E) = µ(E), E ∈ A0 ,
is a measure on (X0 , A0 ).

Proof. Clearly X0 ∈ A0 . Let E ∈ A0 . Then there exists A ∈ A such


that E = A ∩ X0 , and
X0 \ E = X0 \ A ∩ X0 = X0 ∩ (A ∩ X0 )c .
44 Measure and Measurable Functions

Since (A ∩ X0 )c ∈ A, we have X0 \ E ∈ A0 . Next, let {En } be a countably


infinite family in A0 . Let An ∈ A such that En = An ∩ X0 for n ∈ N. Then

[ ∞
[ ∞
[ 
En = (An ∩ X0 ) = An ∩ X0 .
n=1 n=1 n=1
S∞ S∞
Since n=1 An ∈ A, we have n=1 En ∈ A0 .
The second part follows from the fact that µ is a measure on A.

Definition 3.4.3 Let (X, A) be a measurable space and X0 ∈ A. Then


the σ-algebra A0 given in Theorem 3.4.2 is called the restriction of the
σ-algebra A to X0 . If µ is a measure, then the measure µ0 := µ|A0 on
(X0 , A0 ) is called the restriction of the measure µ to (X0 , A0 ). ♦

The following statement can be verified easily.

Given a measurable space (X, A) and X0 ∈ A, the restriction of A


to X0 is the largest σ-algebra on X0 contatained in A. n fact, the
restriction of A to X0 is {A ⊆ X0 : A ∈ A}.

Notation: In the due course, we shall also denote the restriction of a σ-


algebra to a measurable set X0 ∈ A by AX0 , whereas the corresponding
restriction of the measure µ will be denoted by the same notation µ, but
we say that µ is a measure on AX0 or, by abusing the terminology, µ is a
measure on X0 .

In the context of Lebesgue measureable space (R, M, m), a question of


interest would be the following:

Does there exist a σ-algebra A on R properly containing M such that


the restriction of the Lebesgue outer measuere m∗ to A is a measure?

The answer is in negative. We deduce this from the following character-


ization of Lebesgue measurability.
Restrictions of σ-Algebras and Measures 45

Theorem 3.4.4 A subset E of R is Lebesgue measurable if and only if


m∗ (I) ≥ m∗ (I ∩ E) + m∗ (I ∩ E c )
for every open interval I.

Proof. Let E ⊆ R. Clearly, if E is Lebesgue measurable, then


m∗ (I) ≥ m∗ (I ∩ E) + m∗ (I ∩ E c ) (1)
holds for every open interval I. Conversely, suppose (1) holds for every open
interval I. We have to show that
m∗ (A) ≥ m∗ (A ∩ E) + m∗ (A ∩ E c ) (2)
for every A ⊆ R.
Let ε > 0 be given. By the definition of m∗ , there exists a countable
family {In } of open intervals such that
[ X
A⊆ In and `(In ) ≤ m∗ (A) + ε. (3)
n n

Hence, using (1) and (3),


[ [
m∗ (A ∩ E) + m∗ (A ∩ E c ) ≤ m∗ ( In ∩ E) + m∗ ( In ∩ E c )
n n
X
≤ [m∗ (In ∩ E) + m∗ (In ∩ E c )]
n
X
≤ m∗ (In )
n
≤ m∗ (A) + ε.
Thus,
m∗ (A ∩ E) + m∗ (A ∩ E c ) ≤ m∗ (A) + ε
for every ε > 0. Consequently, (2) holds for every A ⊆ R.
Theorem 3.4.5 Suppose there is a σ-algebra A on R such that M ⊆ A and
m∗ is a measure on A. Then A = M.

Proof. We have to prove that A ⊆ M. For this, let E ∈ A. Since A


contains intervals, and m∗ is a measure on A, we have
m∗ (I) = m∗ (I ∩ E) + m∗ (I ∩ E c )
for every open interval I. Therefore, by Theorem 3.4.4, E ∈ M. Thus,
A ⊆ M.
46 Measure and Measurable Functions

Thus, we have also proved the following theorem.

Theorem 3.4.6 If A is a σ-algebra on R properly containing M and µ is


a measure on (R, A) such that µ := m∗ |A and m = µ|M , then A = M and
µ = m.

3.5 Complete Measure Space and Completion


Definition 3.5.1 A measure µ on a measurable space (X, A) is called a
complete measure if for every A ∈ A,

E⊆A and µ(A) = 0 implies E ∈ A.

If µ is a complete measure on (X, A), then (X, A, µ) is called a complete


measure space. ♦

Thus, we may state the following:

A measure space (X, A, µ) is not complete if and only if there


exists A ∈ A and E ⊆ A such that µ(A) = 0 and E 6∈ A.

Example 3.5.2 Recall that (see Theorem 2.2.4(iv)), if A ⊆ R such that


m∗ (A) = 0, then A ∈ M. Hence, it follows that (R, M, m) is complete. We
shall see that (R, B, m) is not complete (Theorem 3.8.5). ♦
Example 3.5.3 Let X = {a, b, c} and A = {∅, X, {a}, {b, c}}. Let µ on A
be defined

µ(∅) = µ({b, c}) = 0 and µ(X) = µ({a}) = 1.

Then µ is a measure on (X, A); it is not complete, as

µ({b, c}) = 0, {b} ⊆ {b, c} but {b} 6∈ A. ♦

Example 3.5.4 For any set nonempty set X, we know that A = {∅, X} is
a σ-algebra and µ defined by µ(∅) = 0 and µ(X) = 0 is a measure on A.
Then we see that µ is complete if and only if X is a singleton set. ♦

The following theorem shows that every measure space can be completed.

Theorem 3.5.5 Let (X, A, µ) be a measure space. Let

N := {E ⊆ X : ∃ B ∈ A with E ⊆ B and µ(B) = 0}


Complete Measure Space and Completion 47

and
à := {A ∪ E : A ∈ A, E ∈ N }.
Then the following are true.

(i) Ã is a σ-algebra on X containing A;

(ii) µ̃ defined on à by

µ̃(A ∪ E) = µ(A) for A ∈ A, E ∈ N

is a complete measure on (X, Ã) with µ̃(A) = µ(A) for every A ∈ A;

(iii) (X, Ã, µ̃) is the smallest complete measure space containing (X, A, µ),
in the sense that if (X, Â, µ̂) is a complete measure space with A ⊆ Â
and µ̂(A) = µ(A) for every A ∈ A, then à ⊆ Â.

Proof. First we observe that A ⊆ Ã and N ⊆ Ã.


(i) Since A ⊆ Ã , X ∈ Ã. Suppose A ∈ A and E ∈ N . Then (A ∪ E)c =
Ac ∩ E c . Since E ⊆ B for some B ∈ A with µ(B) = 0. Hence, E c ⊇ B c so
that E c = B c ∪ (E c \ B c ) = B c ∪ (E c ∩ B). Thus,

(A ∪ E)c = Ac ∩ E c = Ac ∩ [B c ∪ (E c ∩ B)] = [Ac ∩ B c ] ∪ [Ac ∩ (E c ∩ B)].

Note that Ac ∩ B c ∈ A and Ac ∩ (E c ∩ B) ⊆ B with µ(B) = 0. Hence,


E)c ∈ Ã. Next, let An ∈ A and En ∈ N for n ∈ N. We have to show
(A ∪ S
that ∞ n=1 (An ∪ En ) ∈ Ã. Note that


[ ∞ ∞
[  [ 
(An ∪ En ) = An ∪ En .
n=1 n=1 n=1

Since En ⊆ Bn for some Bn ∈ A with µ(Bn ) = 0, we have


∞ ∞ ∞ ∞ ∞
!
[ [ [ [ X
En ⊆ Bn with Bn ∈ A and µ Bn ≤ µ(Bn ) = 0.
n=1 n=1 n=1 n=1 n=1

Therefore, ∞
S S∞
n=1 En ∈ N so that n=1 (An ∪ En ) ∈ Ã. Thus, we have proved
that à is a σ-algebra on X containing A.

(ii) Clearly, µ̃(∅) = 0. Next, let An ∈ A and En ∈ N for n ∈ N be


such that {An ∪ En } is a disjoint family. Then, {An } and {En } are disjoint
48 Measure and Measurable Functions

families and hence


∞ ∞ ∞
! !
[ [  [ 
µ̃ (An ∪ En ) = µ̃ An ∪ En
n=1 n=1 n=1
∞ ∞
!
[ X
= µ An = µ(An )
n=1 n=1

X
= µ̃(An ∪ En ).
n=1

Thus, µ̃ is a measure on Ã.

(iii) For showing the completeness of µ̃, let F ⊆ A ∪ E for some A ∈ A


and E ∈ N with µ̃(A ∪ E) = 0. We have to show that F ∈ Ã. Note that
F = F ∩ (A ∪ E) = (F ∩ A) ∪ (F ∩ E),
where
F ∩ A ⊆ A with µ(A) = µ̃(A ∪ E) = 0
so that F ∩ A ∈ N . Also, F ∩ E ⊆ E with E ∈ N so that F ∩ E ∈ N . Since
N ⊆ à and à is a σ-algebra, F = (F ∩ A) ∪ (F ∩ E) ∈ Ã.

(iv) Suppose that (X, Â, µ̂) is a complete measure space with A ⊆ Â and
µ̂(A) = µ(A) for every A ∈ A. Since A ⊆ Â and µ̂ is a complete measure,
we have N ⊆ Â. Thus, using the fact that  is a σ-algebra, we also obtain
à ⊆ Â.

Definition 3.5.6 The measure space (X, Ã, µ̃) defined in Theorem 3.5.5 is
called the completion of (X, A, µ). ♦
Theorem 3.5.7 The σ-algebra M of Lebesgue measurable sets is the com-
pletion of the Borel σ-algebra B1 .

Proof. Recall from Corollary 2.2.20 that for every E ∈ M, there exist a
Gδ -set G ⊇ E and an Fσ -set F ⊆ E such that m(G \ F ) = 0. We know that
Gδ sets and Fσ sets are Borel sets. Hence,
E = F ∪ (E \ F ), where E \ F ⊆ G \ F with m(G \ F ) = 0.
Thus M is the completion of the Borel σ-algebra B1 .

We shall see in Corollary 3.8.5, using the concept of a measurable func-


tion, that (R, B1 , m) is not complete, and B1 ( M.
Some Properties of Measures 49

3.6 Some Properties of Measures


Here are two theorems which are simple consequences the countable addi-
tivity of a measure.
Theorem 3.6.1 Let (X, A, µ) be a measure space and An ∈ A such that
An ⊆ An+1 for all n ∈ N. Then

[ 
µ(An ) → µ Ai as n → ∞.
i=1

Proof. We write ∞
S S∞
i=1 Ai as a disjoint union i=1 Ei by taking E1 = A1
and Ei = Ai \ Ai−1 for i = 2, 3, . . .. Then

[  ∞
[  ∞
X n
X n
[ 
µ Ai = µ Ei = µ(Ei ) = lim µ(Ei ) = lim µ Ei .
n→∞ n→∞
i=1 i=1 i=1 i=1 i=1
Sn
But, i=1 Ei = An . Hence,

[ 
µ Ai = lim µ(An ).
n→∞
i=1

This completes the proof.

Theorem 3.6.2 Let (X, A, µ) be a measure space and An ∈ A such that


An ⊇ An+1 for all n ∈ N and µ(Ak ) < ∞ for some k ∈ N. Then

\ 
µ(An ) → µ Ai as n → ∞.
i=1

Proof. Since ∞
T T∞
i=1 Ai = i=k Ai for any k ∈ N, we may assume without
loss of generality that µ(A1 ) < ∞, instead of the assumption µ(Ak ) < ∞
for some k ∈ N. Now, let Bn = A1 \ An , n ∈ N. Then Bn ⊆ Bn+1 for all
n ∈ N, so that by the Theorem 3.6.1,

[ 
µ Bi = lim µ(Bn ).
n→∞
i=1

But, ∞
S T∞
i=1 Bi = A1 \ i=1 Ai . Therefore, since µ(A1 ) < ∞, by Theorem
3.1.5 (ii), we have

[   ∞
\  ∞
\ 
µ B i = µ A1 \ Ai = µ(A1 ) − µ Ai ,
i=1 i=1 i=1
50 Measure and Measurable Functions

µ(Bn ) = µ(A1 \ An ) = µ(A1 ) − µ(An ).


Thus, µ(An ) → µ ∞
T 
i=1 Ai as n → ∞.

Remark 3.6.3 It is also to be mentioned that the conclusion in Theorem


3.6.2 need not hold if µ(An ) = ∞ for every n ∈ N. To see this consider the
example of (R, M, m) and An = [n, ∞) for n ∈ N. In this T case, we have
An ⊇ An+1 for everyTn ∈ N, m(An ) = ∞ for every n ∈ N and ∞ n=1 An = ∅.
Thus, m(An ) 6→ m( ∞ A
n=1 n ).
Also, the assumption µ(Ak ) < ∞ for some k ∈ N is not a necessary
condition. To see this, consider An = [1
T∞ − n1 , ∞). Note that µ(An ) = ∞
and An ⊇ AT n+1 for every n ∈ N, and n=1 An = [1, ∞). Thus, we have
∞ 
µ(An ) → µ n=1 An . ♦
Now, let us introduce some set theoretic notions.
Definition 3.6.4 A sequence (An ) of sets is called
(a) monotonically increasing if An ⊆ An+1 for every n ∈ N,
(b) monotonically decreasing if An ⊇ An+1 for every n ∈ N. ♦

We maySobserve that, for any family {Aα : α ∈ Λ} of susets of a set


X, the set α∈Λ Aα is an upper bound of {Aα : α ∈ Λ} with respect to
the partial order of “inclusion” of sets on the T
power set of X; and it is, in
fact, the least upper bound. Similarly, the set α∈Λ Aα is a lower bound of
{Aα : α ∈ Λ}, and it is the greatest lower S
bound. Thus, for a monotonically
increasing sequence (An ) of sets, the set ∞ i=1 Ai can be considered as the
limitTof (An ), and for a monotonically decreasing sequence (An ) of sets, the
set ∞ i=1 Ai can be considered as the limit of (An ).
Motivated by these considerations, we have the following definition.
Definition 3.6.5 Let (An ) be a sequence of subsets of a set X. Then the
limit superior of (An ) and limit inferior of (An ), denoted by lim sup An
n→∞
and lim inf An , respectively, are defined by
n→∞
∞ [
\ ∞ ∞ \
[ ∞
lim sup An = Ai , lim inf An = Ai .
n→∞ n→∞
k=1 i=k k=1 i=k
If lim sup An = lim inf An , then we say that the limit of (An ) exists, and
n→∞ n→∞
the common set is called the limit of (An ), denoted by lim An , i.e.,
n→∞
lim An = lim sup An = lim inf An . ♦
n→∞ n→∞ n→∞
Some Properties of Measures 51

For a sequence (An ) of subsets of a set X and x ∈ X, the following


results can be verified easily (Exercise):

1. x ∈ lim sup An ⇐⇒ x ∈ An for infinitely many n ∈ N,


n→∞

2. x ∈ lim inf An ⇐⇒ x ∈ An for all but finitely many n ∈ N.


n→∞

3. If (An ) is monotonically increasing, then



[
lim sup An = lim inf An = Ai
n→∞ n→∞
i=1
S∞
and hence lim An = i=1 Ai .
n→∞

4. If (An ) is monotonically decreasing, then



\
lim inf An = lim sup An = Ai
n→∞ n→∞
i=1
T∞
and hence lim An = i=1 Ai .
n→∞

Thus, Theorem 3.6.1 and Theorem 3.6.2 can be restated as follows:

If (X, A, µ) is a measure space and (An ) is a sequence in A which


is either monotonically increasing or monotonically decreasing
with µ(A1 ) < ∞, then

µ( lim An ) = lim µ(An ).


n→∞ n→∞

What can we say if (An ) is a general sequence of sets from A?

Theorem 3.6.6 Let (X, A, µ) be a measure space and (An ) be a sequence


of sets in A. Then we have the following.

(i) µ(lim inf An ) ≤ lim inf µ(An ).


n→∞ n→∞
S∞
(ii) If µ( n=1 An ) < ∞, then

µ(lim sup An ) ≥ lim sup µ(An ).


n→∞ n→∞
52 Measure and Measurable Functions

(iii) If lim An and lim µ(An ) exist, then


n→∞ n→∞

µ( lim An ) ≤ lim µ(An ).


n→∞ n→∞
S∞
(iv) If lim An exists and if µ( n=1 An ) < ∞, then lim µ(An ) exists and
n→∞ n→∞

µ( lim An ) = lim µ(An ).


n→∞ n→∞
S T
Proof. For k ∈ N, let Bk = n≥k An and Ck = n≥k An . Then, we have
Bk ⊇ Bk+1 and Ck ⊆ Ck+1 for all k ∈ N.
(i) By Theorem 3.6.1,

[ 
µ(lim inf An ) = µ Ck = lim µ(Ck ).
n→∞ k→∞
k=1

But, µ(Ck ) ≤ µ(Ak ). Therefore,

lim µ(Ck ) = lim inf µ(Ck ) ≤ lim inf µ(Ak )


k→∞ k→∞ k→∞

so that we obtain the required inequality.


(ii) Suppose µ( ∞
S
n=1 An ) < ∞, i.e., µ(B1 ) < ∞. Then, by Theorem
3.6.2,

\ 
µ(lim sup An ) = µ Bk = lim µ(Bk ).
n→∞ k→∞
k=1
But, µ(Bk ) ≥ µ(Ak ). Therefore,

lim µ(Bk ) = lim inf µ(Bk ) ≥ lim sup µ(Ak )


k→∞ k→∞ k→∞

so that we obtain the required inequality.


(iii) This follows from (i).
(iv) By (ii),

lim sup µ(An ) ≤ µ(lim sup An ) = µ( lim An ) = µ(lim inf An )


n→∞ n→∞ n→∞ n→∞

and by (i), µ(lim inf An ) ≤ lim inf µ(An ). Hence,


n→∞ n→∞

lim sup µ(An ) ≤ µ( lim An ) ≤ lim inf µ(An ).


n→∞ n→∞ n→∞

Thus, lim µ(An ) exists and µ( lim An ) = lim µ(An ).


n→∞ n→∞ n→∞
Measurable Functions 53

3.7 Measurable Functions


Suppose that f is a real valued function defined on a subset E of R. Then,
it can be shown that, f is continuous on E if and only if for every open set
G of R, the set f −1 (G) is open in E, that is, f −1 (G) = E ∩ V for some
open subset V of R. Here, the topology on R is considered to be the usual
topology.
Now, let A be either B1 or M, and AE be the restriction of A to E.
Thus, in case E ∈ A,

f continuous ⇒ f −1 (G) ∈ AE for every open set G of R.

Since AE contains sets which are not open, the converse of the above state-
ment is not true. In other words, the set

F := {f : E → R : f −1 (G) ∈ AE for every open set G ⊆ R}

contains more functions than continuous functions. For example, for the
function f : R → R defined by

1, x > 0,
f (x) =
0, x ≤ 0,

we have 

 (0, ∞), 1 ∈ G, 0 6∈ G,
(−∞, 0], 0 ∈ G, 1 6∈ G,

f −1 (G) =

 R, {0, 1} ⊆ G,
{0, 1} ∩ G = ∅.

∅,
for every open set G ⊆ R. Thus, taking E = R, f −1 (G) ∈ B1 for every open
set G ⊆ R, but the function is not continuous.
In view of the above observations, we have the following definition.
Definition 3.7.1 Let (X, A) be a measurable space and let K be either R
or C. Then f : X → K is said to be a measurable function if f −1 (G) ∈ A
for every open set G ⊆ K.
A measurable function f : X → K is said to be a real measurable
function if K = R, and it said to be a complex measurable function if
K = C. ♦

More generally, we defne measurability of functions taking values in a


topological space.
54 Measure and Measurable Functions

Definition 3.7.2 Let (X, A) be a measurable space and (Y, T ) be a topo-


logical space. A function f : X → Y is said to be measurable with respect
to the pair (A, T ) if f −1 (G) ∈ A for every G ∈ T . ♦
Definition 3.7.3 For a given topological space (Y, T ), a function f : R → Y
is called
(i) Lebesgue measurable if f is measurable with respect to the σ-
algebra M on R, and
(ii) Borel measurable if f is measurable with respect to the σ-algebra
B1 on R. ♦

Some of the topological spaces that we frequently use in the study of


measure and integration are the following:

• R with usual topology,


• C with usual topology,
• Rk with usual topology.
• [−∞, ∞] := R ∪ {∞, −∞}, the extended real line, with topology gen-
erated by open subsets of R (under usual topology) and intervals of
the form (a, ∞] and [−∞, b) for a, b ∈ R.
Thus, a subset G of R̃ := [−∞, ∞] is open in R̃ if and only if for every
x ∈ G there exists an interval I of the form (a, b), (a, ∞] or [−∞, b)
such that x ∈ I ⊆ G.

In Definition 3.7.2, if Y = R̃, then we say that f is extended real valued


measurable function.
Using the definition of a σ-algebra, the following statement can be easily
verified.

Let (X, A) be a measurable space and let (Y, T ) be a topological


space such that there is a countable subfamily S of T with the
property that every open set is a countable union of sets from S.
Then, a function f : X → Y is measurable with respect to (A, T )
if and only if f −1 (B) ∈ A for every B ∈ S.

The topological spaces R, C and R̃ have the property stated in the above
box. In the case of Y = R we may take S as the set of all open intervals,
Measurable Functions 55

and if Y = R̃, then we may take S as the set of all intervals of the form
(a, b), (a, ∞] or [−∞, b). In the case of Y = C, we may take S as the set of
all open balls, that is, sets of the form {z ∈ C : |z − α| < ε} for α ∈ C and
ε > 0.

Convention: Let (Y, T ) be a topological space. If the topology T is under-


stood from the context, then instead of saying that (Y, T ) is a topological
space, we may simply say that Y is a topological space.
An important class of functions in measure and integration is the class
of all characteristic functions.
Definition 3.7.4 Given a set X, the characteristic function of a subset
E of X is the function χE : X → R defined by

1, x ∈ E,
χE (x) = 
0, x 6∈ E.

A characteristic function of a set E is also called an indicator function,


and in that case, it is usually denoted by 1E instead of χE .

With the above definition, we have the following theorem.


Theorem 3.7.5 Let X be a measurable space and E ⊆ X. Then χE is
measurable if and only if E is measurable.

Proof. We observe that for an open set G,



 E, 1 ∈ G, 0 6∈ G,
 c

E , 1 6∈ G, 0 ∈ G,
χ−1 (G) =
E

 X, 1 ∈ G, 0 ∈ G,
1 6∈ G, 0 6∈ G.

∅,

Thus, χE is measurable if and only if E is measurable.

The proof of the following theorem is easy, and hence it is left as an


exercise (see Problem ??).
Theorem 3.7.6 Suppose f : X → Y is a measurable function.

(i) If E ∈ A, then f|E : E → Y is a measurable function with respect to


the restricted σ-algebra AE .

(ii) If Y is either R or C or R̃, then for any c ∈ R, cf is measurable.


56 Measure and Measurable Functions

Theorem 3.7.7 Let X be a measurable space, Y be a set and f : X → Y .


Then the following results hold.

(i) S := {E ⊆ Y : f −1 (E) ∈ A} is a σ-algebra on Y .

(ii) If Y is a topological space, then f is measurable if and only if S con-


tains the Borel σ-algebra on Y .

Proof. (i) The fact that S is a σ-algebra on Y follows from the relations

f −1 (∅) = ∅, f −1 (Y ) = X, f −1 (Ac ) = [f −1 (A)]c ,


[  ∞
[
f −1 An = f −1 (An )
n=1 n=1

for subsets A, An of Y for n ∈ N.


(ii) Suppose that Y is a topological space. If f is measurable. Hthen
S contains all open sets. Since Borel σ-algebra BY on Y is the smallest σ-
algebra containing all open sets, S contains BY as well. Converse is obvious
as the Borel σ-algebra on Y contains all open sets in Y .

As an immediate consequence of by Theorem 3.7.7 (ii), we have the


following theorem.

Theorem 3.7.8 Let X be a measurable space and Y be a topological space.


Then,

{f : f −1 (G) ∈ A, G open in Y } = {f : f −1 (B) ∈ A, B ∈ BY }.

In particular, given any function f : X → Y ,

f is measureable ⇐⇒ f −1 (B) ∈ A ∀ B ∈ BY . ♦

In view of the above remark, we have the following general definition of


measurability in a general context.
Definition 3.7.9 Suppose (X1 , A1 ) and (X2 , A2 ) are two measurable spaces.
A function f : X1 → X2 is said to be measurable with respect to the pair
(A1 , A2 ) of σ-algebras if f −1 (B) ∈ A1 for every B ∈ A2 . ♦
Measurable Functions 57

The following theorem shows how a measure on one measurable space can
induce a measure on another measurable space using a measurable function
in the sense discussed in Remark 3.7.9. This theorem is important in the
context of probability theory.
Theorem 3.7.10 Let (X1 , A1 ) and (X2 , A2 ) be a measurable spaces and
f : X1 → X2 be such that f −1 (A) ∈ A1 for every A ∈ A2 . Let µ be a
measure on (X1 , A1 ). Then ν : A2 → [0, ∞] defined by

ν(A) = µ(f −1 (A)), A ∈ A2 ,

is a measure on (X2 , A2 ).

Proof. Clearly, ν(∅) = µ(f −1 (∅)) = 0. Next, let {An } be a countable


disjoint family in A2 . Then
[ [
f −1 ( An ) = f −1 (An )
n n

and {f −1 (An )} is also a disjoint family. Hence,


[ [ [ X X
ν( An ) = µ(f −1 ( An )) = µ( f −1 (An )) = µ(f −1 (An )) = ν(An ).
n n n n n

This completes the proof.

Definition 3.7.11 The measure ν defined as in Theorem 3.7.10 is called


the measure induced by µ and f . ♦
Remark 3.7.12 In the context of probability theory, we may have a sample
space Ω, a σ-algebra A of events, a probability measure µ and a random
variable f , which by definition, is a function f : Ω → R measurable with
respect to A. Then the measure ν on (R, BR ) defined by

ν(B) = µ(f −1 (B)), B ∈ BR ,

is called the probability distribution of the random variable f , and the func-
tion F : R → R defined by

F (x) = µ(f −1 (−∞, x]), x ∈ R,

is called the distribution function or cumulative distribution function induced


by the random variable f . Note that

F (x) = µ({ω ∈ Ω : f (ω) ≤ x}), x ∈ R. ♦


58 Measure and Measurable Functions

Recall that in the definition of measurability is given using open sets


in the topological space. However, for a real or an extended real valued
function, the measurability can be characterized using a very special type
of open sets.
Theorem 3.7.13 For a function f : X → R̃ := [−∞, ∞], the following are
equivalent:

(i) f is measurable.

(ii) {x ∈ X : f (x) > a} ∈ A ∀a ∈ R.

(iii) {x ∈ X : f (x) ≥ a} ∈ A ∀a ∈ R.

(iv) {x ∈ X : f (x) < a} ∈ A ∀a ∈ R.

(v) {x ∈ X : f (x) ≤ a} ∈ A ∀a ∈ R.

Proof. Recall that for any a ∈ R, (a, ∞] is an open set in R̃. Now, for
a ∈ R, we observe that

{x ∈ X : f (x) > a} = f −1 ((a, ∞]),

{x ∈ X : f (x) ≥ a} = f −1 ([a, ∞]),


{x ∈ X : f (x) < a} = f −1 ([−∞, a)),
{x ∈ X : f (x) ≤ a} = f −1 ([−∞, a]).
Note also that

\ 1 
[a, ∞] = a− ,∞ ,
n
n=1

\  1
[−∞, a) = R̃ \ [a, ∞], [−∞, a] = − ∞, a + ,
n
n=1
and for any a, b ∈ R with a < b,

(a, b) = (a, ∞] ∩ [−∞, b).

From these observations together with the properties of a σ-algebra, the


equivalence of (i)-(v) follows.

For the next theorem, and also for use in the due course, we define a few
notions.
Measurable Functions 59

Definition 3.7.14 Let (fn ) be a sequence of functions from a set X to


[−∞, ∞]. Then supremum, infimum, limit superior, limit inferior
of (fn ), denoted by
sup fn , inf fn , lim sup fn , lim inf fn ,
n n n n

respectively, are defined by


(sup fn )(x) = sup fn (x),
n n
(inf fn )(x) = inf fn (x),
n n
(lim sup fn )(x) = lim sup fn (x),
n n
(lim inf fn )(x) = lim inf fn (x)
n n

for x ∈ X. ♦
We may observe the followng.

• If lim supn fn = lim inf n fn , then lim fn (x) exists for every x ∈ X, and
n

lim fn (x) := lim sup fn (x) = lim inf fn (x)


n n n

for every x ∈ X.
Theorem 3.7.15 Let fn : X → [−∞, ∞] be measurable for each n ∈ N.
Then
sup fn , inf fn , lim sup fn , lim inf fn
n n n n

are measurable functions.

Proof. For a ∈ R, we note that


[
{x ∈ X : sup fn (x) > a} = {x ∈ X : fn (x) > a},
n n
[
{x ∈ X : inf fn (x) < a} = {x ∈ X : fn (x) < a}.
n
n
By Theorem 3.7.13, {x ∈ X : fn (x) > a} and {x ∈ X : fn (x) < a}
are measurable sets. Hence, by the properties of the σ-algebra, sup fn and
n
inf fn are also measurable. Since
n

lim sup fn = inf sup fn , lim inf fn = sup inf fn


n k n≥k n k n≥k

it also follows that lim supn fn and lim inf n fn are measurable functions.
60 Measure and Measurable Functions

As a corollary to Theorem 3.7.15, we obtain the following result.


Theorem 3.7.16 Let (fn ) be a sequence of extended real valued measurable
functions on X. If (fn (x)) converges for every x ∈ X, then the function
f : X → [−∞, ∞] defined by

f (x) = lim fn (x), x ∈ X,


n→∞

is measurable.

Definition 3.7.17 Suppose (fn ) is a sequence of functions defined on a


measurable space X with values in Y which is either of R, C or [−∞, ∞].
Then we say that (fn ) converges pointwise on a subset E of X if for each
x ∈ E, the sequence (fn (x)) converges. If (fn ) converges pointwise on E
and f : E → Y is defined by f (x) = lim fn (x), x ∈ E, then we write
n→∞

fn → f pointwise on E. 

Remark 3.7.18 In the above definition, we used a special convention that


if (an ) is a sequence in [−∞, ∞] which diverges to ∞ (respectively, −∞),
then we say that (an ) converges to ∞ (respectively, −∞). ♦

3.7.1 Almost everywhere properties


In measure theory, there are weaker forms of convergence than pointwise
convergence. One such convergence is the almost everywhere convergence.
Definition 3.7.19 Suppose (fn ) is a sequence of measurable functions de-
fined on a measure space (X, A, µ) with values in Y which is either R, C or
[−∞, ∞]. Then we say that (fn ) converges almost everywhere on X, ,
in short (fn ) converges a.e. on X, if the set

E := {x ∈ X : (fn (x)) converges}

belongs to A and µ(X \ E) = 0.


If f : X → Y is such that f (x) = lim fn (x) for every x ∈ E, then we
n→∞
say that (fn ) converges to f alomost everywhere, and write this fact
as fn → f a.e. ♦
Example 3.7.20 For each n ∈ N, let fn : [0, 1] → R be defined by

(−1)n if x ∈ Q,

fn (x) =
1 if x 6∈ Q.
Measurable Functions 61

Note that for every x ∈ [0, 1] \ Q, (fn (x)) converges to the constant function
1 and for x ∈ [0, 1]∩Q, (fn (x)) diverges. Since and m(Q) = 0, (fn ) converges
almost everywhere, but it does not converge pointwise.
Next, let fn : [0, 1] → R be defined by

1 if x ∈ Q,
gn (x) =
(−1)n if x 6∈ Q

for each n ∈ N. In this case, (gn ) neither converges pointwise nor converges
a.e. on [0, 1]. ♦

The concept of almost everywhere can be defined in a more genral con-


text.
Definition 3.7.21 Let (X, A, µ) be a measure space, E ∈ A and P be a
property on the elements of X. Then we say that P holds almost every-
where on E if the set

E := {x ∈ E : P holds at x}

belongs to A and µ(X \ E) = 0, and, in that case, we write P holds a.e.


on E. The fact “P holds a.e. on X” is simply written as P holds a.e. ♦

For example, if f is a function defined on a measurable space (X, A, µ)


with values in R or R̃ or C, then we say that f = 0 a.e. on E if the set
A := {x ∈ E : f (x) 6= 0} belongs to A and µ(A) = 0. For instance, if
f : R → R is defined by

1 if x ∈ Q,
f (x) =
0 if x 6∈ Q,
then f = 0 a.e. on R.
Now, in view of the concept of almost everywhere convergence, we may
ask whether we can modify Theorem 3.7.16 by relaxing the pointwise con-
vergence to almost everywhere convergence.
In this regard we have the following theorem.
Theorem 3.7.22 Let (fn ) be a sequence of extended real valued measurable
functions on a complete measure space (X, A, µ). If fn → f a.e., then f is
measurable.

Proof. Let E = {x ∈ X : fn (x) → f (x)}. By assumption, E ∈ A and


µ(E c ) = 0. We have to prove that f −1 (G) ∈ A for any open set G in
62 Measure and Measurable Functions

[−∞, ∞]. So, let G be an open subset of [−∞, ∞]. Note that
f −1 (G) = {x ∈ E : f (x) ∈ G} ∪ {x ∈ E c : f (x) ∈ G}.
Since fn → f pointwise on E, the set {x ∈ E : f (x) ∈ G} is in AE , and
since {x ∈ E c : f (x) ∈ G} ⊆ E c and µ(E c ) = 0, by the completeness of µ,
{x ∈ E c : f (x) ∈ G} ∈ A. Thus, f −1 (G) ∈ A. is union of two sets in A.

If the measure µ in the above theorem is not complete, then it is necessary


that f is measurable (see Problem 23). However, (fn ) still converges almost
every where to some measurable function, as the following theorem shows.
Theorem 3.7.23 Let (fn ) be a sequence of extended real valued measur-
able functions which converges a.e. on X. Then there exists a measurable
function f such that fn → f a.e.

For its proof we shall make use of the following proposition.


Proposition 3.7.24 Let Y be a topological space, E ∈ A, and f1 : E → Y
and f2 : E c → Y be measurable with respect to the σ-algebras AE and AE c ,
respectively. Then f : X → Y defined by

f1 (x), x ∈ E,
f (x) =
f2 (x), x ∈ E c
is a meaurable function.

Proof. Let G be an open set in Y . Then


{x ∈ X : f (x) ∈ G} = {x ∈ E : f (x) ∈ G} ∪ {x ∈ E c : f (x) ∈ G}
= {x ∈ E : f1 (x) ∈ G} ∪ {x ∈ E c : f2 (x) ∈ G}.
Since f1 and f2 are measurable with respect to the σ-algebras AE and AE c ,
respectively, we have
{x ∈ E : f1 (x) ∈ G} ∈ AE ⊆ A, {x ∈ E : f2 (x) ∈ G} ∈ AE c ⊆ A.
Hence, {x ∈ X : f (x) ∈ G} ∈ A for every open set G in Y ; consequently, f
is measurable.

Proof of Theorem 3.7.23. Let E := {x ∈ X : (fn (x)) converges}.


By assumption, E ∈ A with µ(E c ) = 0. Therefore, by Theorem 3.7.16, the
function f0 : E → Y defined by
f0 (x) = lim fn (x), x ∈ E,
n→∞
Measurable Functions 63

is measurable with respect to the restricted σ-algebra AE . Let



f0 (x), x ∈ E,
f (x) =
0, x ∈ Ec.

Then, by Proposition 3.7.24, f is measurable. Since fn → f pointwise on E


and µ(E c ) = 0, fn → f a.e. on X.

Remark 3.7.25 When we say that a sequence (fn ) of measurable functions


converges almost every where to f , we tacitly assume that f is meaurable,
with the understanding that f is defined by

f0 (x), x ∈ E,
f (x) =
0, x ∈ Ec,

where E := {x ∈ X : (fn (x)) converges} and f0 (x) := lim fn (x), x ∈ E. ♦


n→∞
We know that almost everywhere convergence is weaker than pointwise
convergence (see Example 3.7.19) and pointwise convergence is weaker than
uniform convergence. However, every sequence of functions which converges
almost everywhere also converges uniformly on every set of finite measure
except on a set of arbitrarily small measure. This is the so called Egorov’s
theorem stated below.

Theorem 3.7.26 (Egorov’s theorem) Let (fn ) be a sequence of extended


real valued measurable functions on X which converges to a measurable func-
tion f almost everywhere on X. If µ(X) < ∞, then for every ε > 0, there
exists a measurable set E ⊆ X such that µ(X \E) < ε and fn → f uniformly
on E.

Proof. We assume, without loss of generality, that fn → f pointwise.


Thus, for each x ∈ X and for each n ∈ N, there exists kx,n ∈ N such that

|fm (x) − f (x)| < 2−n ∀ m ≥ kx,n .


S∞
Therefore, X = k=1 Fn,k , where

Fn,k := {x ∈ X : |fm (x) − f (x)| < 2−n ∀ m ≥ k}.

Note that
Fn,k ⊆ Fn,k+1 ∀k ∈ N
so that
µ(Fn,k ) → µ(∪∞
j=1 Fn,j ) = µ(X) as k → ∞.
64 Measure and Measurable Functions

In particular, since µ(X) < ∞,

µ(X \ Fn,k ) = µ(X) − µ(Fn,k ) → 0 as k → ∞.

Hence, for a given ε > 0, there exists kn such that

µ(X \ Fn,kn ) < ε2−n ∀ n ∈ N.

Let E = ∩∞
n=1 Fn,kn . Then


X
µ(X \ E) = µ(∪∞
n=1 (X \ Fn,kn ) ≤ µ(X \ Fn,kn ) < ε.
n=1

Note that if x ∈ E, then x ∈ Fn,kn for all n ∈ N so that

|fm (x) − f (x)| < 2−n ∀ m ≥ kn .

Thus fn → f uniformly on E.

3.7.2 Further properties of measurable functions


We have already observed that, given a measurable space X, a function
f : X → R is measurable if and only if f −1 (I) is a measurable set for every
open interval I. Theorem 3.7.27 below is analogous to it for R2 -valued
functions due to the following facts:

• A set S ⊆ R2 is open if and only if for every (x1 , x2 ) ∈ S, there exist


open intervals I1 and I2 containing x1 and x2 , respectively, such that
I1 × I2 ⊆ S, and

• Every open subset of R2 is a countable union of open rectangles.

Theorem 3.7.27 Let X be a measurable space. Then a function f : X → R2


is measurable if and only if f −1 (I1 × I2 ) is a measurable set for every open
rectangle I1 × I2 in R2 .

Theorem 3.7.28 Let X be a measurable space and f, g are real measurable


functions on X. Then the function h : X → R2 defined by

h(x) = (f (x), g(x)), x ∈ X,

is measurable.
Measurable Functions 65

Proof. Since every open subset of R2 is a countable union of open rect-


angles, it is sufficient to show that for open intervals I1 and I2 , h−1 (I1 × I2 )
is measurable. Note that

h−1 (I1 × I2 ) = {x ∈ X : h(x) ∈ I1 × I2 }


= {x ∈ X : f (x) ∈ I1 and g(x) ∈ I2 }
= f −1 (I1 ) ∩ g −1 (I2 ).

Since f and and g are measurable functions, h−1 (I1 ×I2 ) = f −1 (I1 )∩g −1 (I2 )
is a measurable set.

Theorem 3.7.29 Let X be a measurable space, and Y and Z be topological


spaces. If f : X → Y is measurable and g : Y → Z is continuous, then
g ◦ f : X → Z is measurable.

Proof. Suppose f : X → Y is measurable and g : Y → Z is continuous.


Let V be an open set in Z. By continuity of g, the set g −1 (V ) is open in Y ,
and by measurability of f , f −1 (g −1 (V )) is measurable in X. Hence, from
the identity
(g ◦ f )−1 (V ) = f −1 (g −1 (V ))
the set (g ◦ f )−1 (V ) is measurable in X.

As corollaries to the above theorem we have the following two theorems.

Theorem 3.7.30 Let X be a measurable space.

(i) If f and g are real measurable functions on X and λ ∈ R, then f + g,


f g and λf are real measurable functions, and f + ig is a complex
measurable function.

(ii) If f and g are real measurable functions on X, then f +ig is a complex


measurable function.

(iii) If f is a complex measurable function on X, then the functions Ref ,


Imf, |f |, defined by

[Ref ](x) := Ref (x), [Imf ](x) = Imf (x), |f |(x) := |f (x)|,

respectively, for x ∈ X, are real measurable functions on X.

(iv) If f and g are complex measurable functions on X and λ ∈ C, then


f + g, f g and αf are complex measurable functions on X.
66 Measure and Measurable Functions

Proof. (i) By Theorem 3.7.28, the function x 7→ (f (x), g(x)) from X to


R2 is measurable. Now, the results follow from Theorem 3.7.29 by making
use of the following facts:
(a) The functions (α, β) 7→ α + β, (α, β) 7→ αβ from R2 to R and the
function α 7→ λα from R to R are continuous.
(b) The function (α, β) 7→ α + iβ from R2 to C is continuous.
(c) The functions z 7→ Rez, z 7→ Imz and z 7→ |z| from C to R are
continuous.

Remark 3.7.31 From (i) and (iv) in Theorem 3.7.30, it follows that the set
VR of all real measurable functions and the set VC of all complex measurable
functions are vector spaces over R and C, respectively. Sine R ⊆ C, VC is a
vector space over R as well. ♦
Theorem 3.7.32 Suppose (fn ) is a sequence of complex measurable func-
tions which converges pointwise to a function f on X. Then f is measurable.

Proof. For each n ∈ N, let gn = Re (fn ) and hn = Im (fn ). By Theorem


3.7.30 (iii), gn and hn are real measurable functions. Since fn → f poitwise,
it can be easily seen that gn → g := Re (f ) and hn → h := Im (f ) point-
wise. By Theorem 3.7.16, g and h are real measurable functions. Hence, by
Theorem 3.7.30 (ii), f = g + ih is complex measurable.

Remark 3.7.33 In Theorem 3.7.30, we avoided the case of extended real


valued functions, because f + g need not be defined at some points. Also, so
far we have not considered any topology on [−∞, ∞] × [−∞, ∞]. However,
for non-negative extended real valued measurable functions f and g, we
shall show the measurability of f + g, after considering the notion of simple
measurable functions. ♦
We shall have occasions to use the following definition and the subsequent
theorem.
Definition 3.7.34 Given a function f : X → [−∞, ∞] defined on a set X,
its positive part, negative part and absolute value or modulus, denoted
by f + , f − and |f |, respectively, are defined by

f + (x) = max{f (x), 0},


f − (x) = max{−f (x), 0},
|f | = max{f (x), −f (x)}

for x ∈ X. ♦
Simple measurable functions 67

We may observe that


f = f + − f −, |f | = f + + f − .
Theorem 3.7.35 Let f : X → [−∞, ∞]. If f is measurable, then f + , f −
and |f | are measurable. In case f is real valued, then f is measurable if and
only if f + and f − are measurable.

Proof. Suppose f is measurable. By Theorem 3.7.6, the function −f is


measurable. Now, measurability of f + , f − and |f | follows from Theorem
3.7.15, since
f + = sup{f, 0}, f − = sup{−f, 0}, |f | = sup{f, −f }.
Next, suppose that f is real valued and f + , f − are measurable. Then by
Theorem 3.7.30(i), f = f + − f − is also measurable.

3.8 Simple measurable functions


In this section, we consider a class of functions which are more general than
characteristic functions.
Definition 3.8.1 Let X be a nonempty set. A function ϕ : X → R is called
a simple function if it takes only a finite number of values. ♦

If α1 , . . . , αn are the distinct values of a simple function ϕ : X → R, then


ϕ can be represented as
Xn
ϕ= αi χEi ,
i=1
where Ei = {x ∈ X : ϕ(x) = αi }, i = 1, . . . , n. The above representation of
ϕ is called its canonical representation,.

Note that if ϕ = ni=1 αi χEi is the canonical representation of a simple


P
function ϕ, then {E1 , . . . , En } is a decomposition of X, that is, X = ∪ni=1 Ei
and {E1 , . . . , En } is a disjoint family.
Pn In case α1 , . . . , αn are the nonzero
distinct values of ϕ, then also ϕ = i=1 αi χEi is the canonical representation
of ϕ, but X = ∪ni=0 Ei with E0 = {x ∈ X : f (x) = 0}.
Suppose a1 , . . . , an in R and Ai ⊆ X, i = 1, . . . , n. Then we see that the
function ϕ : X → R defined by
n
X
ϕ= ai χAi
i=1
68 Measure and Measurable Functions

takes only a finite number of values, and hence, it is a simple function. Thus,

A function ϕ : X → R is simple if and only if P there are a1 , . . . , an


in R and Ai ⊆ X, i = 1, . . . , n, such that ϕ = ni=1 ai χAi .

Recall that a function f : R → R is called a step function if there are


disjoint intervals I1 , . . . , In and a1 , . . . , an in R such that
n
X
f= ai χIi .
i=1

Thus, every real valued step function defined on R is a simple function.

Theorem 3.8.2 Let (X, A) be a measurable P space and ϕ be a simple function


on X with canonical representation ϕ = ni=1 αi χEi . Then ϕ is measurable
if and only if Ei ∈ A for i = 1, . . . , n.

Proof. Suppose ϕ is a measurable function. Then, each Ei is a mea-


surable set as it is the inverse image of the closed (singleton) set {αi }.
Conversely, if each Ei is a measurable set, then by Theorem ??, ϕ is a
measurable function.

The next theorem is one of the most important results in the theory of
measure and integration, and it would help us inferring many properties of
a measurable functions.

Theorem 3.8.3 Let (X, A) be a measurable space and f : X → [0, ∞] be a


measurable function.Then there exists a sequence (ϕn ) of simple measurable
functions on X such that

(i) 0 ≤ ϕn ≤ ϕn+1 for every n ∈ N and

(ii) ϕn → f pointwise on X.

In fact, ϕn defined by

n2n
X i−1
ϕn = χEi,n + nχFn , n ∈ N,
2n
i=1
Simple measurable functions 69

where
n i−1 i o
Ei,n := x ∈ X : n ≤ f (x) < n , i = 1, . . . , n2n ,
2 2
Fn := {x ∈ X : f (x) ≥ n},

satisfy the requirements (i) and (ii) above.

Proof. Note that, for each n ∈ N, {Ei,n } ∪ {Fn } is a disjoint family of


measurable sets. Let x ∈ X. If f (x) = ∞, then ϕn (x) = n for all n ∈ N,
so that ϕn (x) → f (x) as n → ∞. In case 0 ≤ f (x) < ∞, then there exists
k ∈ N such that f (x) ≤ k. Hence, for every n ≥ k, x ∈ Ei,n for some
i ∈ {1, 2, . . . , n2n }, and in that case

1
|f (x) − ϕn (x)| ≤ .
2n
Therefore, in this case also, ϕn (x) → f (x) as n → ∞.
Next suppose that x ∈ Ei,n for some n ∈ N and for some i ∈ {1, 2, . . . , n2n }.
Then ϕn (x) = (i − 1)/2n , and
 
i−1 i−1 1
ϕn+1 (x) ∈ , + n+1 .
2n 2n 2

Thus, ϕn (x) ≤ ϕn+1 (x). If x ∈ Fn , then ϕn (x) = n and

1
ϕn+1 (x) ∈ {n, n + }.
2n+1
Thus, we get ϕn (x) ≤ ϕn+1 (x) for every x ∈ X and for every n ∈ N.

As a corollary to the above theorem, we obtain the following theorem.

Theorem 3.8.4 Let f and g be extended real valued non-negative measur-


able functions on a measurable space (X, A). Then f + g is measurable.
More generally, if (fn ) is a sequence of extended real valued non-negative
measurable functions on (X, A), then f := ∞
P
n=1 n is measurable.
f

Proof. By Theorem 3.8.3, there exist increasing sequences (ϕn ), (ψn ) of


non-negative simple measurable functions on X which converge pointwise
to f and g, respectively. Hence, by Theorem 3.7.16, both f and g are
measurable. Also, since ϕn and ψn are measurable functions, by Theorem
70 Measure and Measurable Functions

3.7.30 (i), ϕn + ψn are measurable, and since ϕn + ψn → f + g pointwise,


again by Theorem 3.7.16, f + g is measurable.
Second part of the theorem also follows from Theorem 3.7.16, by consid-
ering sequence of partial sums of the series f := ∞
P
f
n=1 n .

We end this chapter by proving one of the results which we promised


earlier.

Theorem 3.8.5 The Lebesgue measure on the Borel σ-algebra B1 is not


complete. In particular, B1 is a proper subfamily of M.

Proof. In Example 2.1.12 we have considered the Cantor ternary set C


and showed that m∗ (C) = 0. As C is a countable intersection of finite
unions of closed intervals, C ∈ B1 . Hence, it is enough to identify a subset
of C which is not in B1 . For this, first we represent each x ∈ [0, 1] in binary
expansion as

X ϕn (x)
x= ,
2n
n=1

with ϕn (x) ∈ {0, 1} such that the sequence (ϕn (x)) is not eventually con-
stant. Now, define f : [0, 1] → R by

X 2ϕn (x)
f (x) = , x ∈ [0, 1].
3n
n=1

Note that f is an injective function and its range is contained in the Cantor
set C. We observe that, for each n ∈ N, ϕn = χEn , where En is a finite
union of subintervals of [0, 1]. Hence, each ϕn is a Lebesgue measurable
function. Therefore, by Theorem 3.8.4, f is also a Lebesgue measurable
function. Now, let E0 ⊆ [0, 1] be a non-Lebesgue measurable set, and let
F0 = f (E0 ). Thus, F0 ⊆ C and m(C) = 0 so that F0 ∈ M. Now, if F0 ∈ B1 ,
then by the Lebesgue measurability of f (see Theorem 3.7.8),

E0 = f −1 (F0 ) ∈ M;

arriving at a contradiction. Thus, we have proved that the subset F0 of C


does not belong to B1 . The prticular case follows, since F0 ∈ M \ B1 .
Problems 71

3.9 Problems
1. Show that the condition (a) in Definition 3.1.3 can be replaced by “∃ A0 ∈ A
such that µ(A0 ) < ∞.”
2. Let (X, A, µ) be a measure space. Prove that for every A, B ∈ A,

µ(A ∪ B) + µ(A ∩ B) = µ(A) + µ(B).

3. Let m∗k and Mk be as in Section 3.2. Prove that Mk is a σ-algebra and m∗k
restrired to Mk is a measure.
rem-gen-meas
4. Let X be a set. Let A, µ∗ and µ be as in Remark 3.2.2. Prove the following:

(a) A is a σ-algebra on X, and


(b) µ := µ∗ |A is a measure on A.
(c) If A ⊆ X and µ∗ (A) = 0, then A ∈ A.

5. Let (X, A, µ) be a measure space. Prove that, if α is a non-negative real


number, then the function E 7→ αµ(E) is a also measure on (X, A).
More, generally, if µ1 , . . . , µk are measures on a measurable space (X, A) and
if α1 , . . . , αk are non-negative real numbers, then prove that
k
X
E 7→ µi (E), E ∈ A,
i=1

is a measure on (X, A).


6. Let A be a σ-algebra on X and X0 ∈ A. Show that

{E ⊆ X0 : E ∈ A} = {E ∩ X0 : E ∈ A}.

7. Let X be an uncountable set and A ⊆ 2X such that A ∈ A if and only if


either A or Ac is countable. Define µ : A → [0, ∞] such that

0 if A is countable,
µ(A) =
1 if A is uncountable .

Show that A is a σ-algebra and µ is a measure on (X, A).


8. Let (X, A, µ) be a measure space. Prove that µ is σ-finite if and
S only if
there exists a countable disjoint family {An } ⊆ A such that X = n An and
µ(An ) < ∞ for every n ∈ N.
9. Let S is a family of subsets of a set X and let AS be the σ-algebra generated
by S. Show that AAS = AS .
72 Measure and Measurable Functions

10. Prove that the σ-algebra generated by the family of all closed subsets of a
topological space Y is the Borel σ-algebra on Y .
11. Let (An ) and (Bn ) are sequences of subsets of a set X such that there exists
k ∈ N with An = Bn for all n ≥ k. Show that

lim inf An = lim inf Bn and lim sup An = lim sup Bn .


n→∞ n→∞ n→∞ n→∞

12. Let (xn ) be a sequence in R which converges to some a ∈ R. Prove that,


lim {xn } exists, and if a 6= xn for every n ∈ N, then lim {xn } =
6 {a}. Here,
n→∞ n→∞
{x} denotes the singleton set containing x.
13. Let (En ) be a sequence of subsets of a set X and E ⊆ X. Prove that

lim En = E ⇐⇒ lim χEn (x) = χE (x) ∀ x ∈ X.


n→∞ n→∞

14. For subsets A, B of a set X, the set A∆B := (A \ B) ∪ (B \ A) is called the


symmetric difference of A and B. Now, let (X, A, µ) be a measure space.
Prove the following:
(a) For A, B, C ∈ A, µ(A∆B) ≤ µ(A∆C) + µ(C∆B).
(b) For A, B ∈ A, A ∼ B ⇐⇒ µ(A∆B) = 0 defines an equivalent relation
on A.
15. Find examples to show that the inequalities in Theorem 3.6.6 cannot be
replaced by equalities.
16. Let f be a real valued function on a measurable space X. Prove that f is
Borel (resp. Lebesgue) measurable if and only if f −1 (B) is a Borel (resp.
Lebesgue) measurable for every closed set B ⊆ R.
17. Let (X, A) be a measurable space. Prove that a function f : X → R is
measurable if and only if f −1 (I) ∈ A for every open interval I.
18. Prove that if f is a real function on (X, A) such that {x ∈ X : f (x) ≥ r} ∈ A
for every rational number r, then f is measurable.
19. Prove that if f and g are real measurable functions on (X, A), then the
following sets are measurable:
(a) {x ∈ X : f (x) < g(x)}
(b) {x ∈ X : f (x) = g(x)}
(c) {x ∈ X : f (x) ≤ g(x)}
20. Prove that if (fn ) is a sequence of real measurable functions on (X, A), then
the set {x ∈ X : lim fn (x) exists} is a measurable set.
n→∞

21. Let X and Y be topological spaces and f : X → Y . Prove the following.


Problems 73

(a) If f is continuous, then it is measurable with respect to the Borel σ-


algebra BX on X.
(b) f is measurable with respect to the Borel σ-algebra BX on X if and
only if f −1 (B) ∈ BX for every B ∈ BY .
22. Show that, if f is measurable on a complete measure space and g = f a.e.,
then g is measurable. Show also that the above conclusion need not hold if
the measure space is not complete.
23. Let (fn ) be a sequence of extended real valued measurable functions which
converges a.e. on X to a function f . Justify the following statements:
(a) f need not be measurable.
(b) There exists a measurable function g such that f = g a.e.
24. Suppose (X, A) is a measurable space and Y and Z be topological spaces with
associated σ-algebras BY and BZ , respectively. Prove that, if f : X → Y and
g : Y → Z are measurable functions, then g ◦ f : X → Z is measurable.
Pn
25. Prove that if ϕ = i=1 αi χAi is the canonical representation of a simple
function ϕ andP if Ai ∩ E is nonempty for every i = 1, . . . , n, then for any
n
E ∈ A, χE ϕ = i=1 αi χAi ∩E is the canonical representation of χE ϕ.
26. Let X be a set and ϕ : X → R. Show that ϕ is a simple function if and
only if there are distinct nonzero real numbers α1 ,P. . . , αn and disjoint family
n
{Ei : i = 1, . . . , n} of subsets of X such that ϕ = i=1 αi χEi .
27. Let (X, A) be a measurable space and f : X → [0, ∞] be a bounded measur-
able function. Prove that there exists a sequence (ϕn ) of simple measurable
functions on X which converges to f uniformly. Also, prove that the above
conclusion does not hold if f is unbounded.
4

Integral of Positive Measurable


Functions

In this chapter we introduce one of the most important concepts


in measure theory, neamely, the integral of nonnegative mea-
surable functions, and prove a basic theorem called Monotone
convergence theorem. Unlike in many of the standard texts on
measure and integration, we introduce the integration in a more
intuitive manner by using the fact that every nonnegative mea-
surable function is a pointwise limit of an increasing sequence
of simple measurable functions. The method of showing that
the definition is well-defined is used in proving the Monotone
convergence theorem as well.

4.1 Integral of Simple Measurable Functions


Throughout this chapter, (X, A, µ) is a measure space.
Definition 4.1.1 Let ϕ be a non-negative simple measurable function on
X with the canonical representation
n
X
ϕ := α i χ Ai ,
i=1

i.e., α1 , . . . , αn are the distinct values of ϕ and Ai := {x ∈ X : ϕ(x) = αi }


for i ∈ {1, .R. . , n}. Then the integral of ϕ over X with respect to µ,
denoted by X ϕdµ, is defined by
Z n
X
ϕ dµ := αi µ(Ai ). ♦
X i=1

74
Integral of Simple Measurable Functions 75

Remark 4.1.2 (a) In Definition 4.1.1, we assumed that ϕ is non-negative,


n
X
for otherwise the sum αi µ(Ai ) may not be well-defined, since it can
i=1
happen that for some i 6= j, µ(Ai ) = µ(Aj ) = ∞ and αi = −αj 6= 0 so that
we end up in having expression of the form ∞ − ∞.
(b) If ϕ takes the value 0, then the term corresponding to this value need
not be written in the canonical representation and in the expression for the
integral of ϕ. Thus, there is nothing wrong in assuming that α1 , . . . , αn are
the nonzero values of ϕ. But, then
n
[
Ai = X \ A0 , where A0 = {x ∈ X : ϕ(x) = 0}.
i=1

R
(c) If αi > 0 and µ(Ai ) = ∞ for some i ∈ {1, . . . , n}, then X ϕ dµ = ∞.
For example, if ϕ = χ[0,∞) , then
Z
ϕ dm = µ([0, ∞) = ∞. ♦
R

In view of the above remark, we have following definition.


Definition 4.1.3 Let ϕ be a non-negative simple
R measurable function on
X. Then ϕ is said to be integrable on X if X ϕ dµ < ∞. ♦
Pn
If ϕ = i=1 αi χAi is the canonical representation
Pof a non-negative
simple measurable function ϕ, then for E ∈ A, χE ϕ = ni=1 αi χAi ∩E is the
canonical representation of χE ϕ (see Problem 25) so that
Z n
X
χE ϕ dµ = αi µ(Ai ∩ E).
X i=1

This observation motivates to have the following definition.


Definition 4.1.4 If ϕ is a non-negative simple measure function on and if
E ∈ A, then we define
Z Z
ϕ dµ := χE ϕ dµ,
E X

and it is called the integral of ϕ over E with respect to µ. ♦


76 Integral of Positive Measurable Functions

Example 4.1.5 Let X = [a, b], A = M[a,b] , the Lebesgue σ-algebra re-
stricted to [a, b], and let µ is the Lebesgue measure. Let A = [0, 1] ∩ Q and
B = [a, b] ∩ Qc . Then we have
Z Z
χA dµ = µ(A) = 0 and χB dµ = µ(B) = b − a.
X X

Recall that the functions χA and χB are not Riemann integrable. ♦

Exercise 4.1.6 Let E ∈ A. Let µE be the measure induced by µ on


(E, AE ), where AE is the σ-algebra restricted
R to E. Prove
R that, for any
non-negative simple measurable function ϕ, E ϕdµ = E ϕ|E dµE , where
ϕ|E is the restriction of ϕ to E. ♦

CONVENTION: In the following, when we speak about measurable func-


tions, it is meant that they are defined on a measurable space (X, A), and
when we speak about integrals, they are with respect to a measure µ on
(X, A).

In the due course, we shall use the following properties of the character-
istic functions:

(1) For disjoint sets A and B, χA∪B = χA + χB .

(2) For any sets A and B, χA∩B = χA χB .

(3) If A ⊆ B, then χB\A = χB − χA .

(4) For any sets A and B, χA∪B = χA + χB − χA χB .

In the above (1) and (2) can be seen easily; (3) follows from (1) by observing
that B = A ∪ (B \ A), and (4) is a consequence of (1) and (2), since

A ∪ B = (A \ A ∩ B) ∪ (A ∩ B) ∪ (B \ A ∩ B).

Lemma 4.1.7 Let {B1 , . . . , Bk } be a disjoint family ofPmeasurable sets and


let β1 , . . . , βk be non-negative real numbers. Let ψ := ki=1 βi χBi . Then

Z k
X
ψ dµ = βi µ(Bi ).
X i=1
Integral of Simple Measurable Functions 77

Proof. In case β1 , . . . , βk are distinct, the given representation of ψ is its


canonical representation and hence the proof follows from the definition.
Next assume that some of β1 , . . . , βk are repeated. Suppose α1 , . . . , αn
are the distinct numbers among β1 , . . . , βk . For each i ∈ {1, . . . , n}, let
∆i := {j : βj = αi }.
Then it is clear that
k
X n  X
X  n
X X n
X
ψ= βj χBj = βj χBj = αi χBj = αi χAi ,
j=1 i=1 j∈∆i i=1 j∈∆i i=1
S P
where Ai = j∈∆i Bj . Note that µ(Ai ) = j∈∆i µ(Bj ) for i = 1, . . . , n and
{A1 , . . . , An } is a disjoint family. Thus,
Z n
X n
X X
ψ dµ = αi µ(Ai ) = αi µ(Bj )
X i=1 i=1 j∈∆i
n
X X k
X
= βj µ(Bj ) = βi µ(Bi ).
i=1 j∈∆i i=1

This completes the proof.

Example 4.1.8 Let X = [a, b], A = M[a,b] , the Lebesgue σ-algebra re-
stricted to [a, b],
P and let µ is the Lebesgue measure. Let ϕ be a step func-
tion, say ϕ = ni=1 αi χIi with αi ≥ 0 for i = 1, . . . , n, where I1 , . . . , In are
disjoint intervals whose union is [a, b] and α1 , . . . , αn be non-negative real
numbers. Then we have
Z X n
ϕ dµ = αi `(Ii ).
X i=1

Note that the above integral is same as the Riemann integral of ϕ. ♦


Example 4.1.9 Let X be a finite set, A = 2X and µ be the counting
measure. Then every function f : X → [0, ∞) can be represented as
X
f= f (x)χ{x} .
x∈X

Thus, every function f on X is a simple measurable function, and its integral


is given by Z X
f dµ = f (x). ♦
X x∈X
78 Integral of Positive Measurable Functions

Example 4.1.10 Using the result in Example 4.1.9, every finite sum of
non-negative real numbers can be represented as an integral. This can be
seen as follows: Let a1 , . . . , an be in [0, ∞), X = {1, . . . , n}, A = 2X and µ
be the counting measure on X. Define f : X → R by

f (j) = aj , j ∈ {1, . . . , n}.

Then f is a simple function and it has the representation


n
X
f= aj χ{j} .
j=1

Hence,
Z n
X
f dµ = aj .
X j=1

We shall see, in the due course, that every countable sum of non-negative
real numbers can also be represented as an integral. ♦

In Lemma 4.1.7, we assumed that {B1 , . . . , Bk } is a disjoint family. We


shall drop this assumption by making use of the following result.
Theorem 4.1.11 Let ϕ and ψ be non-negative simple measurable functions
and c be a non-negative real number. Then
Z Z Z
(ϕ + ψ) dµ = ϕ dµ + ψ dµ,
X X X
Z Z
c ϕ dµ = c ϕ dµ.
X X
Pn Pm
Proof. Let ϕ := i=1 αi χAi and ψ := j=1 βj χBj be the canonical
representations of ϕ and ψ, respectively. Let

Eij = Ai ∩ Bj , i ∈ {1, . . . , n}, j ∈ {1, . . . , m}.

Since ni=1 Ai = X = m
S S
j=1 Bj , we have

m
[  [m
Ai = Ai ∩ Bj = (Ai ∩ Bj ), i ∈ {1, . . . , n},
j=1 j=1

n
[  [n
Bj = Bj ∩ Ai = (Ai ∩ Bj ), j ∈ {1, . . . , m}.
i=1 i=1
Integral of Simple Measurable Functions 79

Thus,
n
X n
X m
X  n X
X m
ϕ= αi χAi = αi χAi ∩Bj = αi χAi ∩Bj ,
i=1 i=1 j=1 i=1 j=1

m
X m
X n
X  m X
X n
ψ= αi χBj = βj χAi ∩Bj = βj χAi ∩Bj ,
j=1 j=1 i=1 j=1 i=1
n X
X m n X
X m
ϕ+ψ = (αi + βj )χAi ∩Bj , cϕ = cαi χAi ∩Bj .
i=1 j=1 i=1 j=1

Now, since {Ai ∩ Bj : i = 1, . . . , n; j = 1, . . . , m} is a disjoint family of


measurable sets, by Lemma 4.1.7, we have
Z n X
X m
(ϕ + ψ)dµ = (αi + βj )µ(Ai ∩ Bj )
X i=1 j=1
n X
X m n X
X m
= αi µ(Ai ∩ Bj ) + βj µ(Ai ∩ Bj )
i=1 j=1 i=1 j=1
Xn X m Xm Xn
= αi µ(Ai ∩ Bj ) + βj µ(Ai ∩ Bj )
i=1 j=1 j=1 i=1
n
X m
X
= αi µ(Ai ) + βj µ(Bj )
i=1 j=1
Z Z
= ϕdµ + ψdµ,
X X

and
Z n X
X m n X
X m Z
cϕ dµ = cαi µ(Ai ∩ Bj ) = c αi µ(Ai ∩ Bj ) = c ϕ dµ.
X i=1 j=1 i=1 j=1 X

Thus, the proof is complete.

Corollary 4.1.12 If ϕ1 , . . . , ϕn are simple non-negative measurable func-


tions on X, then
Z X n  n Z
X
ϕi dµi = ϕi dµi .
X i=1 i=1 X

Proof. This is immediate from Theorem 4.1.11.


80 Integral of Positive Measurable Functions

Following corollary is a generalization of Lemma 4.1.7.

Corollary 4.1.13 Let B1 , . . . , Bk be measurable sets and β1 , . . . , βk be non-


negative real numbers. Then

k
Z X  k
X
βi χBi dµ = βi µ(Bi ).
X i=1 i=1

Proof. Since ϕi := βi χBi is a simple non-negative measurable function


for each i = 1, . . . , k, the result follows from Coroally 4.1.12.

Corollary 4.1.14 Let ϕ and ψ be non-negative simple measurable functions


on X such that ϕ ≤ ψ. Then
Z Z
ϕdµ ≤ ψdµ.
X X

Proof. We write ψ = ϕ+ (ψ−ϕ) and observe that ψ−ϕ is a non-negative


simple measurable function. Hence, by Theorem 4.1.11,
Z Z Z
ψ dµ = ϕ dµ + (ψ − ϕ) dµ.
X X X
R
Since X (ψ − ϕ) dµ ≥ 0, the result follows.

Corollary 4.1.15 Let (ϕn ) be a sequence of non-negative simple measurable


functions
R  X. Suppose 0 ≤ ϕn ≤ ϕn+1 for all n ∈ N. Then the sequence
on
X ϕn dµ converges in [0, ∞].

R 
Proof. In view of Corollary 4.1.14, the sequence X ϕn dµ is mono-
tonically increasing, and hence it converges in [0, ∞].

Corollary 4.1.16 Let ϕ be a non-negative simple measurable function on


X. Then Z
ν(E) := ϕ dµ, E ∈ A,
E

defines a measure on A.
Integral of Positive Measurable Functions 81

ν(∅) = 0. Let {En } be a disjoint familyPof measurable


Proof. Clearly, S
sets, and let E = n En . We have to show that ν(E) = n ν(En ). Let
ϕ = ki=1 αi χAi be the canonical representation of ϕ. Then
P

k
X k
X
χE ϕ = αi χE χAi = αi χE∩Ai .
i=1 i=1

Then,
Z Z k
X
ν(E) = ϕ dµ = χE ϕ dµ = αi µ(E ∩ Ai ). (1)
E X i=1
But,
h [  i [  X
µ(E ∩ Ai ) = µ En ∩ Ai = µ (En ∩ Ai ) = µ(En ∩ Ai ), (2)
n n n

as {En ∩ Ai : n ∈ N} is a disjoint family in A. Now, (1) and (2) imply


Z XX k XZ X
ν(E) = χE ϕdµ = αi µ(En ∩ Ai ) = ϕdµ = ν(En ).
X n i=1 n En n

This completes the proof that ν is a measure.

Definition 4.1.17 The measure ν defined Corollary 4.1.16 is called the


measure induced by ϕ. ♦

4.2 Integral of Positive Measurable Functions


Suppose f : X → [0, ∞] is a measurable function. By Theorem 3.8.3,
there exists an increasing sequence (ϕn ) of non-negative simple measurable
functions
R which converges to f pointwise. By Corollary 4.1.15, the sequence
X ϕ n dµ converges in [0, ∞]. So, we may define
Z Z
f dµ := lim ϕn dµ.
X n→∞ X

For this definition to be meaningful, it is necessary to prove that the above


limit is independent of the sequence (ϕn ), as long as it is increasing and
converging pointwise to f . That is, we have to show that, if (ϕn ) and
(ψn ) are increasing sequences of non-negative simple measurable functions
converging to f pointwise, then .
Z Z
lim ϕn dµ = lim ψn dµ.
n→∞ X n→∞ X
82 Integral of Positive Measurable Functions

Theorem 4.2.1 Let f : X → [0, ∞] be a measurable function and (ϕn ) be


an increasing sequence of non-negative simple measurable functions on X
which converges to f pointwise. Then
Z Z
lim ϕn dµ = sup ϕdµ,
n→∞ X ϕ∈Sf X

where Sf is the set of all non-negative simple measurable functions ϕ satis-


fying ϕ ≤ f .
In particular, if (ϕn ) and ()ψn ) are pointwise convergent sequences of
non-negative simple measurable functions on X having the same limit, then
Z Z
lim ϕn dµ = lim ψn dµ.
n→∞ X n→∞ X

R Proof. By the definition


R of Sf , we have ϕn ∈ Sf for all n ∈ N. Therefore,
ϕ
X n dµ ≤ supϕ∈Sf X ϕdµ for all n ∈ N and hence
Z Z
lim ϕn dµ ≤ sup ϕdµ.
n→∞ X ϕ∈Sf X

To show the other way inequality, let ϕ ∈ Sf . We are going to show that
Z Z
r ϕdµ ≤ lim ϕn dµ ∀ r ∈ (0, 1), (∗)
X n→∞ X

so that by letting r → 1, the result will follow. For this, let r ∈ (0, 1) and
for each n ∈ N, let

En = {x ∈ X : rϕ(x) ≤ ϕn (x)}.

Clearly, En ∈ A and En ⊆ En+1 for every n ∈ N. Further, X = ∞


S
n=1 En .
To see this, let x ∈ X. If ϕ(x) = 0, then x ∈ En for all n ∈ N. Next,
suppose ϕ(x) > 0. Then r ϕ(x) < ϕ(x) ≤ f (x). Since ϕn (x) → f (x), there
exists k ∈ N such that r ϕ(x) ≤ ϕk (x) ≤ f (x). Hence, x ∈ Ek . Now, let ν
be the measure induced by ϕ, that is,
Z
ν(E) = ϕdµ, E ∈ A.
E

Then we have
Z Z Z
r ν(En ) = rϕdµ ≤ ϕn dµ ≤ ϕn dµ.
En En X
Integral of Positive Measurable Functions 83

By taking limit and using Theorem 3.6.1,


Z Z
r ϕdµ = r ν(X) = r lim ν(En ) ≤ lim ϕn dµ.
X n→∞ n→∞ X

Thus, we have proved (∗), and the proof is complete.

Definition 4.2.2 Let f : X → [0, ∞] be a measurable function. Then the


integral of f over X is defined as
Z Z
f dµ := sup ϕ dµ,
X ϕ∈Sf X

where Sf is the set of all non-negative simple measurable functions ϕ satis-


fying ϕ ≤ f . If E ∈ A, then the integral of f over E is defined as
Z Z
f dµ := χE f dµ. ♦
E X

Notation: Given an extended real valued non-negative measurable func-


tion f , we shall use the notation Sf to denote the set of all non-negative
simple measurable functions ϕ satisfying ϕ ≤ f .

In view of Theorem 4.2.1, we have the following theorem.

Theorem 4.2.3 Let f be an extended real valued non-negative measurable


function on X. If (ϕn ) is an increasing sequence of non-negative simple
measurable functions which converges pointwise to f , then
Z Z
f dµ = lim ϕn dµ.
X n→∞ X

Example 4.2.4 Let (an ) be a sequence of non-negative real numbers. Let


X = N, A = 2N and µ be the counting measure on N. Define f : X → R by

f (j) = aj , j ∈ N.

Then f can be represented as f = ∞


P
j=1 aj χ{j} . Note that

n
X
f = lim aj χ{j} = lim ϕn ,
n→∞ n→∞
j=1
84 Integral of Positive Measurable Functions

where ϕn = nj=1 aj χ{j} . Clearly, (ϕn ) is an increasing sequence of non-


P
negative simple measurable functions which converges to f pointwise. Hence,
by Theorem 4.2.3, we have
Z Z n
X ∞
X
f dµ = lim ϕn dµ = lim aj = aj .
X n→∞ X n→∞
j=1 j=1

Thus, every countable sum of non-negative real numbers can be represented


as an integral. ♦

Theorem 4.2.5 Let f and g be extended real valued non-negative measur-


able functions on X and c ≥ 0. Then f +g and cf are measurable functions,
and Z Z Z
(f + g)dµ = f dµ + g dµ.
X X X

Proof. By Theorem 3.8.3, there exist increasing sequences (ϕn ), (ψn ) of


non-negative simple measurable functions on X which converge pointwise to
f and g, respectively. Then (ϕn + ψn ) and (cϕn ) are increasing sequences of
non-negative simple measurable functions which converge pointwise to f + g
and c f , respectively. Hence, f + g and c f are measurable functions, and by
Theorem 4.2.3,
Z Z
(f + g)dµ = lim (ϕn + ψn )dµ
X n→∞ X
Z Z 
= lim ϕn dµ + ψn dµ
n→∞
Z X X
Z
= lim ϕn dµ + lim ψn dµ
n→∞ X n→∞ X
Z Z
= f dµ + g dµ,
X X

and Z Z Z Z
c f dµ = lim cϕn dµ = lim c ϕn dµ = c f dµ.
X n→∞ X n→∞ X X

This completes the proof.

Notation:
R OnceR the measure
R µ under
R consideration
R is understood, we may
write E f dµ as E f and X f dµ as X f or f .
In the following theorem, we list some of the properties of the integral.
Integral of Positive Measurable Functions 85

Theorem 4.2.6 Suppose that f, g are extended real valued non-negative


measurable functions defined on a measure space X, c ∈ [0, ∞) and E ∈ A.
Then we have the following.
R R
(i) f ≤ g ⇒ Ef ≤ E g.
R R
(ii) A, B ∈ A, A ⊆ B ⇒ A f ≤ B f.
R R R
(iii) A, B ∈ A, A ∩ B = ∅ ⇒ A∪B f = A f + B f .
R R
(iv) E c f = c E f.
R
(v) f (x) = 0 ∀ x ∈ E ⇒ E f = 0.
R
(vi) µ(E) = 0 ⇒ E f = 0.
R R
(vii) E f = supϕ∈S f E
ϕ.

Proof. (i) Suppose f ≤ g. Then we have Sf ⊆ Sg . Hence,


Z Z Z Z
f = sup ϕ ≤ sup ϕ= g.
X ϕ∈Sf X ϕ∈Sg X X

Also, f ≤ g implies χE f ≤ χE g so that from the above,


Z Z Z Z
f= χE f ≤ χE g = g.
E X X E

(ii) Let A, B ∈ A be such that A ⊆ B. Then, χA f ≤ χB f . Hence, (i)


implies Z Z Z Z
f= χA f ≤ χB f = f.
A X X B

(iii) Let A, B ∈ A be such that A ∩ B = ∅. Then χA∪B = χA + χB so


that, by Theorem 4.2.5,
Z Z Z   Z Z Z Z
f= χA∪B f = χA + χB f = χA f + χB f = f+ f.
A∪B X X X X A B
R R
(iv) Suppose ϕ ∈ SRf . We know
R that E c ϕ = c E ϕ. Taking supremum
over all ϕ ∈ Sf , we get c f = c f so that, by Theorem 4.2.5, we also have
Z Z Z Z
c f = c χE f = c χE f = c f.
E X E
86 Integral of Positive Measurable Functions

R R
(v) If f (x) = 0 for all x ∈ E, then χE f = 0. Hence, E f= X χE f = 0.
(vi) Suppose µ(E) = 0 and ϕ ∈ SχE f . Then we have
Z Z
ϕ = χE ϕ and ϕ = χE ϕ = 0.
X

Hence, Z Z Z
f= χE f = sup ϕ = 0.
E X ϕ∈Sχ f X
E
R R
(vii) Let ϕ R∈ Sf . Then,
R χ E ϕ ≤ χ E f so that by (i), χ E ϕ ≤ X χE f .
Thus, supϕ∈Sf E ϕ ≤ E f . To show the reverse inequality, let ϕ ∈ SχE f .
Then ϕ ∈ Sf and ϕ = χE ϕ. Hence,
Z Z Z Z Z Z
f= χE f = sup ϕ = sup χE ϕ ≤ sup χE ϕ = sup ϕ.
E X ϕ∈Sχ f X ϕ∈Sχ f X ϕ∈Sf X ϕ∈Sf E
E E

This completes the proof.

Theorem 4.2.7 Let f be an extended real valued non-negative measurable


function on X. Then
Z
f = 0 ⇐⇒ f = 0 a.e. on X
X

Proof. Let A = {x ∈ X : f (x) 6= 0} and B = {x ∈ X : f (x) = 0}.


Then A ∩ B = ∅ and A ∪ B = X. Suppose µ(A) = 0. Hence, by Theorem
4.2.6(iii), (v), (vi), we have
Z Z Z
f= f+ f = 0.
X A B
R
Conversely, suppose that X f = 0. Observe that

[ n 1o
A= An where An := x ∈ X : f (x) ≥ .
n
n=1

We prove that µ(An ) = 0 for every n ∈ N so that µ(A) = 0. Note that


χAn f ≥ n1 χAn for every n ∈ N. Hence, by Theorem 4.2.6(i), we have
Z Z Z
µ(An )
0= f≥ f= χAn f ≥ ∀ n ∈ N.
X An X n
Thus, µ(An ) = 0 for every n ∈ N, and thus the proof is over.
Riemann Integral as Lebesgue Integral 87

4.3 Riemann Integral as Lebesgue Integral


Suppose f : [a, b] → [0, ∞) is a Riemann integrable function. We show that
f is measurable with respect to the σ-algebra M[a,b] and
Z b Z
f (x)dx = f dm.
a [a,b]

Let (Pn ) be a partition of [a, b] such that Pn+1 is a refinement of Pn and


|Pn | → 0 as n → ∞. Then we know that
Z b Z b
L(Pn , f ) → f (x)dx and U (Pn , f ) → f (x)dx
a a

as n → ∞, where L(Pn , f ) and U (Pn , f ) denote the lower sum and upper
sum corresponding to the partition Pn . Note that if
(n) (n) (n)
Pn : a = x0 < x1 . . . < xkn = b,

(n) (n) (n) (n) (n)


Ii = [xi−1 , xi ), mi = inf f (x), Mi = sup f (x),
(n) (n)
x∈Ii x∈Ii

for i = 1, . . . , kn , then
kn Z
(n) (n) (n)
X
L(Pn , f ) := mi (xi − xi−1 ) = ϕn dm
i=1 [a,b]

and
kn Z
(n) (n) (n)
X
U (Pn , f ) := Mi (xi − xi−1 ) = ψn dm,
i=1 [a,b]

where
kn kn
(n) (n)
X X
ϕn = mi χ (n) , and ψn = Mi χ (n)
I I
i i
i=1 i=1

for i = 1, . . . , kn . Thus,
Z Z b Z Z b
ϕn dm → f (x)dx and ψn dm → f (x)dx
[a,b] a [a,b] a

as n → ∞. Let
ϕ = sup ϕn , ψ = inf ψn .
n∈N n∈N
88 Integral of Positive Measurable Functions

Since Pn+1 is a refinement of Pn , we have

ϕn ≤ ϕn+1 ≤ ϕ ≤ f ≤ ψ ≤ ψn+1 ≤ ψn ∀ n ∈ N.

Since |Pn | → 0 as n → ∞, we also have the convergence

ϕn → ϕ and ψn → ψ pointwise.

Hence, ϕ and ψ are measurable functions and


Z Z
(ψ − ϕ)dm ≤ (ψn − ϕn )dm ∀ n ∈ N.
[a,b] [a,b]

We have already seen that


Z Z Z
(ψn − ϕn )dm = ψn dm − ϕn dm → 0
[a,b] [a,b] [a,b]
R
as n → ∞. Therefore, [a,b] (ψ − ϕ)dm = 0, so that by Corollary 4.2.7, ϕ = ψ
a.e. Thus, we also obtain f = ϕ a.e. Since Lebesgue measure is complete, f
is measurable, and we have
Z b Z Z Z
f (x)dx = lim ϕn dm = ϕdm = f dm.
a n→∞ [a,b] [a,b] [a,b]

Thus, we have prove the following theorem.


Theorem 4.3.1 If f : [a, b] → [0, ∞) is a Riemann integrable function, then
f is measurable with respect to the σ-algebra M[a,b] and
Z b Z
f (x)dx = f dm.
a [a,b]

4.4 Monotone Convergence Theorem (MCT)


Theorem 4.4.1 (MCT) Let (fn ) be a sequence of extended real valued
non-negative measurable functions defined on X such that

fn ≤ fn+1 for all n ∈ N.


Z
Let f (x) = lim fn (x), x ∈ X. Then f is measurable, lim fn exists and
n→∞ n→∞ X
Z Z
lim fn = f.
n→∞ X X
Monotone Convergence Theorem (MCT) 89

Proof. Since fn (x) → f (x) for every x ∈ X, by Theorem 3.7.16, the


function f : X → [0, ∞] is measurable. Further, we have
fn ≤ fn+1 ≤ f ∀ n ∈ N,
and by Theorem 4.2.6(i),
Z Z Z
fn ≤ fn+1 ≤ f ∀ n ∈ N.
X X X
Z
Hence, lim fn exists and
n→∞ X
Z Z
lim fn ≤ f.
n→∞ X X

Following the same lines of arguments as in the proof of Theorem 4.2.1 with
the sequence (fn ) in place of (ϕn ) we obtain
Z Z Z
f := sup ϕdµ ≤ lim fn .
X ϕ∈Sf X n→∞ X

For the sake of completion, let us imitate the arguments here briefly:
Let ϕ ∈ Sf , and for 0 < r < 1 and n ∈ N, let
En = {x ∈ X : rϕ(x) ≤ fn (x)}.
Then (En ) is an increasing sequence in A and X = ∞
S
n=1 En . Consider the
measure induced by ϕ, that is,
Z
ν(E) = ϕdµ, E ∈ A.
E

Then we have
Z Z Z
r ν(En ) = rϕdµ ≤ fn dµ ≤ fn dµ.
En En X

Taking limit,
Z Z
r ϕdµ = r ν(X) = lim r ν(En ) ≤ lim fn dµ.
X n→∞ n→∞ X

Since this is true for every r ∈ (0, 1) and for every ϕ ∈ Sf ,


Z Z Z
f := sup ϕdµ = lim fn dµ..
X ϕ∈Sf X n→∞ X

This completes the proof of the theorem.


90 Integral of Positive Measurable Functions

Remark 4.4.2 Traditionally, MCT (Theorem 4.4.1) is proved first and then
the result in Theorem 4.2.1 is observed. We followed the reverse path as it
is felt that Theorem 4.2.1 is a good motivation for defining the integral. ♦
Corollary 4.4.3 Let (fn ) be a sequence
P of extended real valued non-negative
measurable functions on X and f := ∞ n=1 fn . Then for every E ∈ A,
Z X∞ Z
f= fn .
E n=1 E

Proof. Let gn = ni=1 fi for n ∈ N. Then 0 ≤ gn ≤ gn+1 for every n ∈ N


P
and gn → f pointwise on X. Hence, by monotone convergence theorem and
Theorem 4.2.5,
Z Z n Z
X ∞ Z
X
f = lim gn = lim fi = fi .
X n→∞ X n→∞ X X
i=1 i=1

This completes the proof.

Corollary 4.4.4 Let f : X → [0, ∞) be a measurable function and let


Z
ν(E) := f dµ, E ∈ A.
E

Then ν is a measure on X. Further, if g : X → [0, ∞] is any measurable


function, then Z Z
gdν = gf dµ. (∗)
X X

Proof. Clearly, ν(∅) = 0. Suppose {En } is a countable disjoint family


in A. Then, Z Z
[ 
ν Ei := S f dµ = χSi Ei f dµ.
i i Ei X

Now, we observe that


n
X
χSi Ei f = lim χSn Ei
f = lim χEi f,
n→∞ i=1 n→∞
i=1

where ( ni=1 χEi f ) is an increasing sequence of non-negative measurable


P
functions. Hence by MCT (Theorem 4.4.1),
Z Z X n Xn Z ∞ Z
X
χSi Ei f dµ = lim χEi f = lim χEi f = χEi f.
X n→∞ X n→∞ X X
i=1 i=1 i=1
Monotone Convergence Theorem (MCT) 91

Thus,
[  ∞
X
ν Ei = ν(Ei ).
i i=1

This completes the first part of the theorem.


Next, we observe that the relation (∗) holds if g is a characteristic func-
tion of a measurable set. Indeed, if g = χE for some measureable set, then
Z Z
gdν = χE dν = ν(E)
X X

and Z Z Z
gf dµ = χE f dµ = f dµ
X X E

so that, by the definition of ν, we obtain (∗). Now, using the property of


the integral, (∗) holds for all simple non-negative measurable functions as
well. Since any measurable function g : X → [0, ∞] is a pointwise limit of an
increasing sequence of simple non-negative measurable functions, the proof
of (∗) can be completed by invoking Theorem 4.2.1 or MCT.

Corollary 4.4.5 (Fatou’s lemma) Let (fn ) be a sequence of extended real


valued non-negative measurable functions on a measure space (X, A, µ).
Then Z Z
(lim inf fn ) ≤ lim inf fn .
X n n X

Proof. For each k ∈ N, let gk = inf n≥k fn . Then gk ≤ gk+1 for all k ∈ N,
and lim gk = f := lim inf fn . Hence, by MCT (Theorem 4.4.1),
k→∞ n
Z Z
lim gk = f.
k→∞ X X
R R
But, since gk ≤ fk for all k ∈ N, we have X gk ≤ X fk so that
Z Z Z Z
f = lim gk = lim inf gk ≤ lim inf fk .
X k→∞ X k X k X

This completes the proof.

Strict inequality can hold in Fatou’s lemma. To see this, consider the
following two examples.
92 Integral of Positive Measurable Functions

Example 4.4.6 Consider the measure space (X, A, µ) = (R, M, m) and

fn := χ[n,∞) , n ∈ N.
R
Then, we have lim fn (x) = 0 for every x ∈ R and X fn = ∞ for all n ∈ N.
n→∞
Hence, Z Z
(lim inf fn ) dµ = ( lim fn ) dµ = 0,
X n X n→∞
Z
lim inf fn dµ = ∞.
n X

Thus, in this example, strict inequality holds in Fatou’s lemma. ♦


Example 4.4.7 Consider the measure space (X, A, µ) = ([0, 1], M[0,1] , m)
and
fn := n χ(0, 1 ] , n ∈ N.
n
R
Then, we have lim fn (x) = 0 for every x ∈ X, whereas X fn = 1 for all
n→∞
n ∈ N. Hence,
Z Z
(lim inf fn ) dµ = ( lim fn ) dµ = 0,
X n X n→∞
Z
lim inf fn dµ = 1.
n X

Thus, strict inequality holds in Fatou’s lemma. ♦

In MCT we assumed that fn ≤ fn+1 for every n ∈ N, that is, for each
x ∈ X, fn (x) ≤ fn+1 (x) for every n ∈ N. A natural question is whether the
conclusion of MCT holds if the above condition is replaced by

fn ≤ fn+1 a.e. for each n ∈ N.

We give an affirmative answer by modifying f appropriately.

Theorem 4.4.8 (MCT) Let (fn ) be a sequence of extended real valued


non-negative measurable functions defined on X such that fn ≤ fn+1 a.e.
for every n ∈ N. Then (fn ) converges a.e. to a measurable function f and
Z Z
lim fn = f.
n→∞ X X
Monotone Convergence Theorem (MCT) 93

Proof. We note that



\
E = {x ∈ X : fn (x) ≤ fn+1 (x) for every n ∈ N} = En ,
n=1

where En := {x ∈ X : fn (x) ≤ fn+1 (x)} for each n ∈ N. By the assumption


that fn ≤ fn+1 a.e., we have En ∈ A and µ(Enc ) = 0 for each n ∈ N. Hence,
  X∞
µ(E c ) = µ ∪∞ E
n=1 n
c
≤ µ(Enc ) = 0.
n=1

Let g(x) = lim fn (x) for x ∈ E. Then g is measurable with respect to AE


n→∞
and, by Proposition 3.7.24, the function f defined by

g(x), x ∈ E,
f (x) =
0, x ∈ Ec
is measureable. Since µ(E c ) = 0 and µ(Enc ) = 0 for every n ∈ N, we have
Z Z Z Z
f dµ = f dµ + f dµ = gdµ,
X E Ec E
Z Z Z Z
fn dµ = fn dµ + fn dµ = fn dµ,
X E Ec E
and since fn ≤ fn+1 on E and lim fn (x) = g(x) for every x ∈ E, by
n→∞
Theorem 4.4.1, Z Z
gdµ = lim fn dµ.
E n→∞ E

Hence, Z Z Z Z
f dµ = gdµ = lim fn dµ = lim fn dµ.
X E n→∞ E n→∞ X

This completes the proof.

The relation (∗) in Corollary 4.4.4 is usually written as



dν = f dµ or = f,

and f is called the Radon-Nikodym derivative of ν with respect to µ.

Note that if µ and ν are as in Corollary 4.4.4, then for every E ∈ A,


µ(E) = 0 ⇒ ν(E) = 0.
94 Integral of Positive Measurable Functions

Definition 4.4.9 Suppose µ and ν are measures on a measurable space


(X, A). Then ν is said to be absolutely continuous with respect to µ
if for every A ∈ A,
µ(E) = 0 ⇒ ν(E) = 0,

and this fact is written as ν << µ. ♦

A question naturally arises is the following:

If µ and ν are measures such that ν << µ, then


Z does there exist a
measurable function f ≥ 0 satisfying ν(E) = f dµ ∀ E ∈ A?
E

The answer to the above question is in affirmative, and this result is called
the Radon-Nikodym theorem.

Theorem 4.4.10 (Radon-Nikodym theorem) If µ and ν are σ-finite


measures on a measurable space X such that ν << µ, then there exists a
non-negaive measurable function f , unique upto a.e., such that
Z
ν(E) = f dµ ∀ E ∈ A.
E

As the proof of this thisrem is a bit long, we relegate it to an appendix


to this chapter (Section ??).
We may observe the following:

If µ and ν are measures as in Radon-Nikodym


R theorem,
then ν is a finite mesaure if and only if X f dµ < ∞.

Also, it is to be mentioned that the existence of f in the Radon-Nikodym


theorem is not guaranteed if one of the measues µ and ν is not σ-finite. For
example, let ν be the Lebesgue measure and µ be the counting measure on
(R, M). Clearly, for E ∈ M,

µ(E) = 0 ⇒ E=∅ ⇒ ν(E) = 0.


Monotone Convergence Theorem (MCT) 95

Hence, ν << µ. However, thereR does not exist a non-negative measurable


function f such that ν(E) = E f dµ for every E ∈ M. To see this, let
E = {x} for some x ∈ R. Then
Z
ν(E) = 0 and f dµ = f (x)
E

for every measurable function f . Thus,


R if there exists a non-negative mea-
surable function f such that ν(E) = E f dµ for every E ∈ M, then f must
be the zero funciton. This forces ν(E) = 0 for every E ∈ M, which is not
true.
Remark 4.4.11 Radon-Nikodym theorem is important in probability theory
due to the following considerations:
Recall from Remark 3.7.12 that if (Ω, A, µ) is a probability space , and
h : Ω → R is a random variable,, that is a measurable functions, then the
distribution of h is the probability measure ν defined on (R, BR ) by

ν(B) = µ(h−1 (B)), B ∈ B,

and the distribution function of h is the function F : R → R defined by

F (t) := µ{x ∈ X : h(x) ≤ t}, t ∈ R,

that is, F (t) = ν((−∞, t]) for t ∈ R.


Thus, if we can establish that ν is absolutely continuous with respect
to the Borel measure m on R, then as a consequence of Radon-Nikodym
theorem, there exists a non-negative Borel measurable function f : R → R
such that Z
F (t) = f dm, t ∈ R.
(−∞, t]

Such a function f , if exists, is called the probability density function of the


random variable h.
Another consequence of Radon-Nikodym theorem is the following: Sup-
pose f is a non-negative measurable function on (X, A, µ) and let ν be the
measure as in Corollary 4.4.4. That is,
Z
ν(E) = f dµ, E ∈ A.
E

Let A0 be another σ-algebra on X, which is a subfamily of A. Note that f


need not be measurable with respect to A0 . Let ν0 : A0 → [0, ∞] be defined
96 Integral of Positive Measurable Functions

by Z
ν0 (E) = f dµ, E ∈ A0 .
E
Then ν0 is a measure on A0 and ν0 << µ0 , where µ0 is the restriction of the
measure µ to A0 . Hence, by Radon-Nikodym theorem (Theorem 4.4.10),
there exists a function g : X → [0, ∞], measurable respect to A0 , such that
Z Z
ν0 (E) = g dµ0 = g dµ, E ∈ A0 .
E E

In the context of probability theory, where f is a random variable, the


function g is called the conditional expectation of f . ♦

4.5 Problems
In the following it is assumed that a measure space (X, A, µ) is given.

1. Suppose ϕ is a non-negative simple measurable function and E is a measur-


able set. Let µE be the restriction
R of µR to the restricted σ-algebra AE :=
{A ⊆ E : A ∈ A}. Show that E ϕdµ = E ϕdµE .
2. Let ϕ and ψR be non-negative
R simple measurable
R functions such that ϕ ≤ ψ.
Prove that X (ψ − ϕ)dµ = X ψ dµ − X ϕ dµ.
3. RLet f and g be non-negative
R measurable R such that f ≤ g and
R functions
X
g dµ < ∞. Prove that X
(g − f )dµ = X
g dµ − X
f dµ.
4. ProveRthat if Rf ≥ 0 is a measurable function and E ∈ A is such that µ(E) = 0,
then X f = E c f .
P∞ P∞ P∞ P∞
5. Let aij ≥ 0 for all i, j ∈ N. Show that i=1 j=1 aij = j=1 i=1 aij .
6. Suppose X = {x1 , x2 , . . .}, and
P w1 , w2 , . . . are non-negative reals. Let w(xi ) =
wi for i ∈ N and µ(E) = x∈E w(x) for E ⊆ X. Show that µ is a mea-
sure on (X, 2X ), and
R for every P∞ extended real valued non-negative measurable
function f on X, X f dµ = i=1 f (xi )wi .
7. Give details of the proof of the last inequality in the proof of Theorem 4.4.1.
8. Let E1 and E2 be in A such that E1 ∪ E2 = X, f1 and f2 be extended
 to AE1 and AE2 , respectively.
real valued measurable functions with respect
f1 (x), x ∈ E1 ,
Then f : X → [−∞, ∞] defined by f (x) = is meaurable.
f2 (x), x ∈ E2
9. P
Let (fRn ) is a sequence of complex valued measurable functions such that

n=1 X |fn − f | dµ converges. Then fn → f a.e. - Why?

10. Let f : X → [0, ∞] be measurable and An := {x ∈ X : f (x) ≥ n}. Prove the


following:
Appendix 97

Z
(a) If f dµ < ∞, then µ(An ) → 0 as n → ∞.
X

X
(b) If µ is a finite measure, then µ(An ) < ∞.
n=1
P∞ P∞
(Hint: For (b), observe nµ(An \ An+1 ) = n=1 µ(An ).)
n=1

11. Let f : X → C be measurable such that X |f |p dµ < ∞ for some p > 0.


R

Let En = {x ∈ X : |f (x)| ≥ n}. Show that µ(En ) = o(1/np ), i.e.,


lim np µ(En ) = 0.
n→∞
(Hint: WritePAk := {x ∈ X : k ≤ |f (x)| < k + 1} and observe that
p ∞ R
|f |p dµ.)
R
X
|f | dµ = k=1 Ak

12. Let (fn ) be a sequence of non-negative measurable


R functions which con-
verges pointwise and let f be its limit. If lim X fn dµ exists and it is equal
R R n→∞ R
X
f dµ < ∞, then prove that lim E fn dµ = E f dµ for every E ∈ A.
n→∞
(Hint: Use Fatou’s lemma for fn χE and fn − fn χE .)
13. Justify the statement: If f : R → [0, ∞) is a continuous function, then
Z Z n
f dm = lim f (x)dx.
R n→∞ −n

14. Let (X, A, µ) be a measure space with µ(X) < ∞ and f be a non-negative
measurable function on X. Let g(t) := µ({x ∈ X : f (x) ≤ t}), t ∈ R.
Prove that there exists a measure ν on the Borel σ-algebra on R such that
ν((−∞, t]) = g(t) for all t ∈ R.

4.6 Appendix
Proof of Theorem 4.4.10: Radon-Nikodym theorem. We first prove
the existence part of the theorem when µ and ν are finite measures, and then
use this part for proving the existence part when µ and ν are σ-finite mea-
sures. Finally uniqueness part is proved.
Step (1): Suppose µ and ν be finite measures on (X, A) such that ν << µ.
In this case, we prove following:

There
R exists a non-negaive measurable function f such that ν(E) =
E f dµ for all E ∈ A.

Let F be the class of measurable functions f : X → [0, ∞] such that


Z
f dµ ≤ ν(A) ∀ A ∈ A.
A
98 Integral of Positive Measurable Functions

Clearly, F is nonempty as zero function belongs to F, and


Z
α := sup f dµ ≤ ν(X) < ∞.
f ∈F X

We prove the following:


R
(i) There exists f ∈ F such that X f dµ = α.

(ii) The function λ : A → [0, ∞) be defined by


Z
λ(A) := ν(A) − f dµ, A ∈ A,
A

is a measure.

(iii) If λ 6= 0, then there exists  ∈ F and c > 0 such that

µ(Â) > 0 and f + cχ ∈ F.

(iv) If λ is not the zero measure, then observation in (c) contradicts (a).

Hence, λ(A) = 0 for every A ∈ A; equivaletnly,


Z
ν(A) = f dµ ∀ A ∈ A.
A

Proof of (i): We observe that

f, g ∈ F ⇒
max{f, g} ∈ F.
R
Let (fn ) be a sequence in F such that X fn dµ → α. For each n ∈ N, let

gn := max{f1 , . . . , fn } ∈ F.

Note that g1 ≤ g2 ≤ · · · . Then f := lim gn is measurable, and by MCT,


n→∞
Z Z
lim gn = f
n→∞ A A

for every A ∈ A. Since gn ∈ F, it also follows that f ∈ F. Further,


Z Z Z
α := lim fn ≤ lim gn = f dµ ≤ α.
n→∞ X n→∞ X X
Appendix 99

Thus, Z
f dµ = α.
X

Proof of (ii): Let λ : A → [0, ∞) be defined by


Z
λ(A) := ν(A) − f dµ, A ∈ A.
A

Since µ and ν are finite measures, it can be easily shown that λ is a finite
measure on (X, A).
Proof of (iii): Suppose λ is not the zero measure. Then there exsits A ∈ A
such that λ(A) > 0. In particular, there exists A0 ∈ A and k ∈ N such that
1
λ(A0 ) > µ(A0 ).
k
Let
1
S := {A ∈ A : λ(A) > µ(A)}
k
and
G := {A ∈ S : B ⊆ A ⇒ B ∈ S}.
We show that G 6= ∅. Clearly A0 ∈ S. Let

1
β0 := inf{λ(B) − µ(B) : B ∈ A, B ⊆ A0 }.
k
If β0 ≥ 0, then A0 ∈ G and we are done.
Suppose β0 < 0. Since µ and ν are finite measures, β0 6= −∞. Hence,
−∞ < β0 < 0. Let B0 ∈ A be such that B0 ⊆ A0 and
1 β0
λ(B0 ) − µ(B0 ) < .
k 2
Let A1 := A0 \ B0 . Then we see that λ(A1 ) − k1 µ(A1 ) > 0. We may observe
that, if B ∈ A is such that B ⊆ A1 , then λ(B)− k1 µ(B) ≥ β20 . (For, otherwise
B ∪ B0 would violate the definition of β0 .)
If A1 ∈ G, then we are done. Otherwise, we may repeat the above
procedure with A1 in place of A0 to obtain a B1 and then to get A2 :=
A1 \ B1 . Thus, we get either a finite number of sets A1 , . . . , An , B1 , . . . , Bn
in A and numbers β1 , . . . , βn , βn+1 such that
1
Ai ∈ S, Bi ⊆ Ai , βi := inf{λ(B) − µ(B) : B ∈ A, B ⊆ Ai }
k
100 Integral of Positive Measurable Functions

with βi+1 ≥ βi /2 for i = 1, . . . , n, and An+1 = An \ Bn ∈ G or obtain


sequences (An ) and (Bn ) of sets in A and sequence (βn ) of real numbers
such that
1
An ∈ S, Bn ⊆ An , βn := inf{λ(B) − µ(B) : B ∈ A, B ⊆ An }
k
with βn+1 ≥ βn /2 for all n ∈ N and An+1 = An \ Bn 6∈ G for any n ∈ N. In
the latter case, for each n ∈ N,

(a) λ(An+1 ) − k1 µ(An+1 ) ≥ λ(An ) − k1 µ(An ) ≥ λ(A0 ) − k1 µ(A0 ) > 0,

(b) βn+1 ≥ β0 /2n .


T
Define A∞ := n∈N An , we have

λ(A∞ ) = lim λ(An ) and µ(A∞ ) = lim µ(An ).


n→∞ n→∞

Hence, by (a) above,


1 1
λ(A∞ ) − µ(A∞ ) ≥ λ(A0 ) − µ(A0 ) > 0.
k k
Let B ∈ A be such that B ⊆ A∞ . Then for each n ∈ N, B ⊆ An+1 and
1
λ(B) − µ(B) ≥ βn+1 ≥ β0 /2n ≥ 0.
k
Hence, A∞ ∈ G. Thus, we have shown that there exists  ∈ G.
Next, let h := f + k1 Â. Observe that
Z Z Z
1
B ⊆ Â ⇒ hdµ ≤ f dµ + µ(B) ≤ f dµ + λ(B) ≤ ν(B),
B B k B

and Z Z
B ⊆ X \ Â ⇒ hdµ = f dµ ≤ ν(B).
B B
Hence,
Z Z Z
hdµ = hdµ + hdµ ≤ ν(B ∩ A) + ν(B \ A) = ν(B).
B B∩A0 B\A0

Thus, we hav shown that h ∈ F. Since  ∈ G, we also have µ(Â) > 0.


Proof of (iv): By (iii), µ(Â) > 0 and f + k1 Â ∈ F. Hence, the relation h ≤ f
a.e.cannot be true. Therefore, λ must be the zero measure.
Appendix 101

Step (2): Suppose µ and ν are σ-finite measures, not necessarily finite,
on (X, A) such that ν << µ.
By this assumption, there exists a countable
S mutually disjoint collection
{Xn } of measurable sets such that X = n∈N Xn , and µ(Xn ) < ∞ and
ν(Xn ) < ∞ for each n ∈ N. For each n ∈ N, consider the finite measures µn
and νn on (X, A) defined by

µn (E) := µ(E ∩ Xn ) and νn (E) := ν(E ∩ Xn ) for E ∈ A.

By Step (1), for each n ∈ N, there exists a non-negaive measurable function


fn such that Z
νn (E) = fn dµn ∀ E ∈ A.
E

Now, we define f : X → [0, ∞] by

f (x) = fn (x) whenever x ∈ E ∩ Xn .

Since {Xn } is aScountable mutually disjoint collection of measurable sets


such that X = n∈N Xn , the above function is well-defined and is measur-
able. Hence, for E ∈ A,
Z Z Z
f dµ = f dµ = S f dµ
S ∞ ∞
E E∩( n=1 Xn ) n=1 (E∩Xn )

XZ ∞ Z
X ∞
X
= f dµ = fn dµn = ν(E ∩ Xn )
n=1 E∩Xn n=1 E∩Xn n=1
= ν(E).

Step 3: The function f obtianed in Step (2) is is unique a.e.


To see this, suppose f and g are non-negative functions such that
Z Z
ν(E) = f dµ and ν(E) = g dµ ∀ E ∈ A.
E E

As in Step (2), let {Xn } be S


a countable mutually disjoint collection of mea-
surable sets such that X = n∈N Xn , and let E ∈ A. Then we have
Z Z
f dµ = g dµ ∀ n ∈ N.
E∩Xn E∩Xn
102 Integral of Positive Measurable Functions

Since the integrals on both sides of the above equations are finite, we have
Z Z
χXn (f − g) dµ = (f − g) dµ = 0 ∀ n ∈ N. (∗)
E E∩Xn

For n ∈ N, let hn := χXn (f − g) and

En+ := {x ∈ X : hn (x) ≥ 0}, En− := {x ∈ X : hn ≤ 0}.

Since (∗) is true for every E ∈ A, it also follows that


Z Z Z Z
+ −
hn dµ = hn dµ = 0 and hn dµ = (−hn ) dµ = 0
+ −
X En X En

for all n ∈ N. Therefore, for each n ∈ N, hn = 0 a.e., that is, f = g a.e. on


Xn . Consequently, f = g a.e. on X.
5

Integral of Complex Measurable


Functions

In the last chapter, we considered integral of measurable func-


tions which are extended real valued and nonnegative. In this
chapter, we extend the concept of integral to real or complex
valued measurable functions, provided the expressions involved
are well-defined. Also, we shall prove one of the most important
theorems in the theory of measure and inegration, namely, the
dominated convergence theorem, and derive many important and
useful consequences of this theorem.

5.1 Integrability and some properties


Throughout this chapter we consider a measure space (X, A, µ). Recall from
Definition 3.7.34 that if f is a real valued function defined on X, then
f = f + − f −,
where f + and f − are the positive part and negative part, respectively, of f ,
that is, for x ∈ X,

f + (x) := max{f (x), 0}, f − (x) := max{−f (x), 0}.


So, it is natural to extend the definition of integral to real measurable
functions as in the following defintion.
Definition 5.1.1 Suppose f is a real measurable function on X. If at least
one of X f + dµ and X f − dµ is finite,Rthen the integral of f over X with
R R

respect to the measure µ, denoted by X f dµ, is defined by


Z Z Z
f dµ = +
f dµ − f − dµ. ♦
X X X

103
104 Integral of Complex Measurable Functions

Since |f | = f + + f − , we have f + dµ + f − dµ and hence,


R R R
X |f | dµ = X X

Z Z Z
f + dµ and f − dµ are finite if and only if |f | dµ is finite,
X XZ Z Z X
and in that case f dµ = f + dµ − f − dµ.
X X X

R
Note that X |f | dµ is defined in the case of complex measurable functions
f as well. Hence, we introduce the following definition.
Definition 5.1.2 A real orZ complex measurable function f on X is said to
be integrable (over X) if |f |dµ < ∞. ♦
X

Notation: We shall denote the set of all complex measurable functions


on X which are integrable by L(X, A, µ), or by L(µ) or L(X). If Ω is a
Lebesgue measurable subset of R, then an f ∈ L(Ω) is called a Lebesgue
integrable function on Ω.

If f is a real measurable function, then by extending the codomain of f


to C, f can be thought of as a complex valued measurable function. Hence,
we have the following:

Z if and only if f ∈ L(µ),


A real measurableZ function fZ is integrable
and in that case, f dµ = f + dµ − f − dµ.
X X X

Suppose f is a complex measurable function on X. Since f = Ref +iImf ,


we have

|Ref | ≤ |f |, |Imf | ≤ |f | and |f | ≤ |Ref | + |Imf |.

Hence,
f ∈ L(µ) ⇐⇒ Ref ∈ L(µ) and Imf ∈ L(µ).

Thus, we have the following definition.


Integrability and some properties 105

Definition 5.1.3 If f ∈ L(µ), then the integral of f over X with respect


to µ is defined by
Z Z Z
f dµ := Ref dµ + i Imf dµ.
X X X

If E ∈ A, then the integral of f over E with respect to µ is defined as


Z Z
f dµ := χE f dµ. ♦
E X

Notation: OnceRthe measureR is understood from


R the context, we shall also
use the notation X f or by f for the integral X f dµ.
Let us prove some simple properties of the integral of complex valued
measurable functions.

Theorem 5.1.4 If f and g are in L(µ) and c ∈ C, then f + g and cf are


in L(µ), and
Z Z Z Z Z
(f + g) = f+ g, cf = c f.
X X X X X

Proof. Suppose f and g are in L(µ) and c ∈ C. Since

|f + g| ≤ |f | + |g| and |cf | = |c| |f |,

both f + g and αf belong to L(µ).


Now to prove the equalities of the integrals, we consider two cases.
Case 1: f and g are real valued and c ∈ R.
Note that
f + g = (f + − f − ) + (g + − g − ).
Thus,
(f + g)+ − (f + g)− = (f + − f − ) + (g + − g − ),
so that
(f + g)+ + f − + g − = (f + g)− + f + + g + .
Hence, by Theorem 4.2.5,
Z Z Z Z Z Z
+ − − − +
(f + g) + f + g = (f + g) + f + g+.
X X X X X X
106 Integral of Complex Measurable Functions

Since each integral on both the sides of the above equation is finite, we
obtain,
Z Z Z Z Z Z
− −
+
(f + g) − (f + g) = +
f − f + +
g − g−.
X X X X X X

Thus, Z Z Z
(f + g) = f+ g.
X X X
Next, note that

cf = (c+ − c− )(f + − f − ) = (c+ f + + c− f − ) − (c+ f − + c− f + )

so that
(cf )+ − (cf )− = (c+ f + + c− f − ) − (c+ f − + c− f + )
and hence

(cf )+ + (c+ f − + c− f + ) = (cf )− + (c+ f + + c− f − ).

Again, by Theorem 4.2.5, we have


Z Z Z Z Z Z
+ − − + −
+
(cf ) + c f + c f = (cf ) + + +
c f + c− f − .
X X X X X X

Since each integral on both the sides of the above equation is finite, we
obtain,
Z Z Z
cf = (cf )+ − (cf )−
X ZX ZX Z Z
= + +
c f − c+ f − + c− f − − c− f +
X Z X
Z X Z XZ
h i h i
− −
= c + +
f − f −c +
f − f−
X X X X
hZ Z i
= (c+ − c− ) f+ − f−
Z X X

= c f.
X

Case 2: f and g are complex valued.


In this case, we observe that

f + g = Re (f + g) + iIm (f + g) = [Re f + Re g] + i[Im f + Im g]


Integrability and some properties 107

Hence, using Case 1,


Z Z Z
(f + g) = [Re f + Re g] + i [Im f + Im g]
X X X
hZ Z i hZ Z i
= Re f + Re g + i Im f + Im g
X X X X
hZ Z i hZ Z i
= Re f + i Im f + Re g + i Im g
Z X Z X X X

= f+ g.
X X

If c ∈ C with c = α + iβ, where α, β ∈ R, then we have


cf = (α + iβ)(Re f + i Im f ) = (α Re f − β Im f ) + i(α Im f + β Re f ).
Now, using the Case 1, we get
Z Z Z
cf = (α Re f − β Im f ) + i (α Im f + β Re f )
X XZ Z X Z Z
= α Re f − β Im f + iα Im f + iβ Re f
X Z X ZX X

= (α + iβ) Re f + i(α + iβ) Im f


Z X Z X
h i
= c Re f + i Im f
Z X X

= c f.
X

This completes the proof.

Exercise 5.1.5 For real integrable functions f and g, we have


f + g = (f + − f − ) + (g + − g − ) = (f + + g + ) − (f − + g − )
and
Z Z Z Z Z Z
− −
+ +
(f + g ) − (f + g ) = +
f +g − f − g− + −

Z Z  Z Z 
+ − + −
= f − f + g − g
Z Z
= f + g.
Z Z Z
From the above, can you conclude (f + g) = f+ g? ♦
108 Integral of Complex Measurable Functions

Theorem 5.1.6 If f ∈ L(µ), then


Z Z

f ≤ |f |.

X X
R
Proof. Let f ∈ L(µ), and let θ ∈ R be such that X f = X f eiθ . Then,
R

using Theorem 5.1.4,


Z Z Z Z Z
f = e−iθ −iθ −iθ
Im (e−iθ f ).

f= e f= Re(e f ) + i
X X X X X

Since the first term above is real, we must have X Im (e−iθ f ) = 0, and
R

hence,
Z Z
Re(e−iθ f )

f =

X
ZX Z
−iθ
= [Re(e f )] − +
[Re(e−iθ f )]−
ZX X

≤ [Re(e−iθ f )]+
ZX

≤ |f |.
X

This completes the proof.

The following corollary is immediate from the above theorem.


R
Corollary 5.1.7 If f is measurable and f = 0 on E ∈ A, then E f dµ = 0.

Theorem 5.1.8 The following results hold.

(i) Suppose f and g are complex measurable functions such that f = 0


a.e. and g = 0 a.e. Then f + g = 0 a.e.
R
(ii) Suppose f ∈ L(µ) is such that E f = 0 for all E ∈ A. Then f = 0
a.e.

Proof. (i) We know that, if h is a complex measurable function, then


h = 0 a.e. if and only if the set Eh := {x ∈ X : h(x) 6= 0} is of measure
zero. Note that
Ef +g ⊆ Ef ∪ Eg .
Hence, µ(Ef +g ) ≤ µ(Ef ) + µ(Eg ) = 0 so that µ(Ef +g ) = 0.
Integrability and some properties 109

(ii) First observe that it is enough to prove for the case of a real valued f .
So, let f be a real valued measurable function. Let E = {x ∈ X : f (x) ≥ 0}.
By hypothesis, Z Z
f+ = f =0
X E

so that by Theorem 4.2.7, f + = 0 a.e. Similarly, we have f − = 0 a.e. Hence,


by (i), f = f + − f − = 0 a.e.

Remark 5.1.9 Theorem 5.1.4 shows that L(µ) is a vector space over C and
the map Z
f 7→ f dµ
X
is a linear functional on L(µ). Further, the map
Z
f 7→ |f |dµ
X

is a seminorm on L(µ).

By a seminorm on a real or complex vector space space V , we


mean a function p : V → R such that

p(u + v) ≤ p(u) + p(v), p(αv) = αp(v)

for all u, v ∈ V and for all scalar α. It can be easily shown that
if p is a seminorm on a vector space V , then

p(0) = 0 and p(v) ≥ 0 ∀ v ∈ V.

To have an example of a seminorm other than the one given above, let
us consider the vector space B(Ω), the vector space of all bounded real or
complex valued functions defined on a set Ω. Then, for each ω ∈ Ω, the
map f 7→ f (ω) is a linear functional on B(Ω) and the map f 7→ |f (ω)| is a
seminorm on B(Ω).
Now, in view of Theorem 4.2.7 and Theorem 5.1.8, we have
Z
|f |dµ = 0 ⇐⇒ f = 0 a.e.
X

So, let
Z := {f ∈ L(µ) : f = 0 a.e}.
110 Integral of Complex Measurable Functions

By Theorem 5.1.8, Z is a subspace of L(µ). Consider the quotient space

L(µ) := L(µ)/Z.

By abusing the notation, we shall denote the equivalent class [f ] in L(µ) by


f . Thus, the map Z
f 7→ |f | dµ
X
defines a norm on L(µ).

By a norm on a real or complex vector space space V , we mean


a seminorm p : V → R which also satisfies the condition

v ∈ V, p(v) = 0 ⇒ v = 0.

The usual notation for a norm is k · k.

As another example of a norm other than the one given above, let us con-
sider the vector space B(Ω) of all bounded real or complex valued functions
defined on a set Ω. Then, the map

f 7→ kf k := sup |f (x)|
x∈Ω

is a norm on B(Ω). In later sections, we shall deal with some other spaces
with norms. ♦

5.2 Riemann Integral as Lebesgue Integral


Recall from Theorem 4.3.1 that if f : [a, b] → [0, ∞) is a Riemann integrable
function, that is, integrable over [a, b] with respect to the Lebesgue measure
on [a, b], then it is Lebesgue integrable and
Z b Z
f (x)dx = f dm.
a [a,b]

Now, we show that this result still holds if f is a real valued Riemann
integrable function.
So, suppose that f : [a, b] → R is a Riemann integrable function. Then,
we may recall from the theory of Rieman integration that |f | is also Riemann
integrable and Z b Z b


f (x)dx ≤ |f (x)|dx.
a a
Riemann Integral as Lebesgue Integral 111

Now, let us observe that


1 1
f + = (|f | + f ), f − = (|f | − f ).
2 2
Hence, both f + and f − are Riemann integrable, and
Z b
1 b
Z Z b 
+
f (x)dx = |f (x)|dx + f (x)dx , (1)
a 2 a a
Z b Z b Z b
1  
f − (x)dx = |f (x)|dx − f (x)dx . (2)
a 2 a a
By the result in Section 4.3, f + and f − are Lebesgue measurable and
Z b Z Z b Z

+
f (x)dx = +
f dm, f (x)dx = f − dm.
a a,b] a a,b]

Therefore, f is Lebesgue integrable, and


Z Z Z
f dm = +
f dm − f − dm
[a,b] [a,b] [a,b]
Z b Z b
= f + (x)dx − f − (x)dx.
a a

Now, using relations (1) and (2), we obtain


Z Z b
f dm = f (x)dx.
[a,b] a

Thus, we have proved the following theorem.


Theorem 5.2.1 If f : [a, b] → R is a Riemann integrable function, then it
is Lebesgue integrable, and
Z b Z
f (x)dx = f dm.
a [a,b]

It is to be remarked that the concept of Riemann integral can be extended


naturally to complex valued functions as well. Suppose f is a complex valued
bounded function defined on an interval [a, b]. Then f is said to be Riemann
integrable if Ref and Imf are Riemann integrable, and in that case we define
the Riemann integral of f by
Z b Z b Z b
f (x)dx := Ref (x)dx + i Imf (x)dx.
a a a

Thus, we can also assert the following.


112 Integral of Complex Measurable Functions

Theorem 5.2.2 If f : [a, b] → C is a Riemann integrable function, then it


is Lebesgue integrable, and
Z b Z
f (x)dx = f dm.
a [a,b]

Notation: If the measure space is R with Lebesgue measure m, and if J


is an interval with end points a and b, where a < b, then for every complex
measurable function f on J, we use the notations
Z b Z Z a Z
f dm := f dm and f dm := − f dm.
a J b J

5.3 Dominated Convergence Theorem (DCT)


Recall that, in Monotone Convergence Theorem, we required the sequence
(fn ) of functions to be non-negative and monotonically increasing. Now,
we shall prove convergence results without these assumptions; but using
some other conditions. First, let us prove the following theorem, which is
analogous to a theorem in the theory of Riemann integration (see, e.g., [?]).

Theorem 5.3.1 Let (fn ) be a sequence of complex measurable functions


which converges uniformly to a function f on X and let µ(X) < ∞. Then
Z
lim |fn − f |dµ = 0.
n→∞ X

Further, if f ∈ L(µ), then fn is integrable for all large enough n, and


Z Z
lim fn dµ = f dµ.
n→∞ X X

Proof. Let ε > 0 be given and let N ∈ N be such that

|fn (x) − f (x)| ≤ ε ∀n ≥ N ∀ x ∈ X.

Then, for all n ≥ N ,


Z Z
|fn − f |dµ ≤ εdµ = εµ(X).
X X
Z
Thus, lim |fn − f | = 0.
n→∞ X
Dominated Convergence Theorem (DCT) 113

Note that fn − f ∈ L(µ) for all n ≥ N . Hence, if f ∈ L(µ), then


fn = f + (fn − f ) ∈ L(µ) and
Z Z Z Z

fn dµ − f dµ = (fn − f ) dµ ≤ |fn − f | dµ.

X X X X
Z Z
for all n ≥ N . Hence, lim fn dµ = f dµ.
n→∞ X X

Now, we would like to relax the conditions in Theorem 5.3.1 on (fn ) and
X. The resulting theorem is the Dominated Convergence Theorem, one of
the most important one in the theory of integration.

Theorem 5.3.2 (DCT) Suppose (fn ) is a sequence of complex measurable


functions such that (fn ) converges pointwise on X, and there exists g ∈ L(µ)
satisfying |fn | ≤ |g| for all n ∈ N. Let f (x) := lim fn (x), x ∈ X. Then fn
n→∞
and f are integrable,
Z Z Z
lim |fn − f | = 0 and lim fn = f.
n→∞ X n→∞ X X

Proof. Since |fn | ≤ |g| for some g ∈ L(µ) and (fn (x)) converges for every
x ∈ X, it follows that |f | ≤ |g| so that fn ∈ L(µ) and f ∈ L(µ). Also, we
have
|fn − f | ≤ |fn | + |f | ≤ 2|g| ∀ n ∈ N.
Thus, 2|g| − |fn − f | ≥ 0 for all n ∈ N, and 2|g| − |fn − f | → 2|g| as n → ∞.
Hence, by Fatou’s lemma,
Z Z
2|g| = lim inf (2|g| − |fn − f |)
X X n
Z
≤ lim inf (2|g| − |fn − f |)
n X
Z Z
= 2|g| − lim sup |fn − f |.
X n X

Thus, Z Z
0 ≤ lim inf |fn − f | ≤ lim sup |fn − f | ≤ 0.
n X n X
Z
Consequently, lim |fn − f | exists and it is equal to 0.
n→∞ X
114 Integral of Complex Measurable Functions

The following example shows that the condition (b) in the Dominated
Convergence Theorem cannot be dropped.
Example 5.3.3 For each n ∈ N, let fn : R → R be defined by

1/n if 0 < x < n,
fn (x) =
0 otherwise.
R
Then, (fn ) converges to the zero function uniformly. But R fn dm = 1 for
every n ∈ N. Note that, (fn ) is not dominated by an integrable function. ♦

The conclusions in DCT hold good if we replace pointwise convergence


by convergence almost everywhere provided we define f appropriately.
Theorem 5.3.4 (DCT) Suppose (fn ) is a sequence of complex measurable
functions such that (fn ) converges a.e. on X, and there exists g ∈ L(µ)
satisfying |fn | ≤ |g| for all n ∈ N. Let E := {x ∈ X : fn (x) → f (x)} and
(
lim fn (x), x ∈ E,
f (x) := n→∞
0, x ∈ X \ E.

Then fn and f are integrable,


Z Z Z
lim |fn − f | = 0 and lim fn = f.
n→∞ X n→∞ X X

Proof. The conclusion follow by applying Theorem 5.3.2 on (E, AE , µE )


and then using Proposition 3.7.24.

The following theorem is an immediate consequence of DCT, and hence


we omit its proof.
Theorem 5.3.5 (Bounded Convergence Theorem) Suppose (fn ) is a
sequence of complex measurable functions which converges pointwise to a
function f on X. If µ(X) < ∞ and (fn ) is uniformly bounded, that is,
there exists M > 0 such that |fn (x)| ≤ M for all x ∈ X, then fn and f are
integrable,
Z Z Z
lim |fn − f | = 0 and lim fn = f.
n→∞ X n→∞ X X

Example 5.3.6 Let


nx
fn (x) = , x ∈ [0, 1].
1 + n2 x2
Dominated Convergence Theorem (DCT) 115

We observe that fn (x) → 0 for each x ∈ [0, 1] and


1
0 ≤ fn (x) ≤ ∀ x ∈ [0, 1], n ∈ N.
2
Hence, by BCT, Z 1 Z
fn (x)dx = f dm → 0.
0 [0,1]

Note that the convergence of (fn ) to the zero function is not uniform, and
hence, Theorem 5.3.1 cannot be applied. ♦

For the next consequence of the DCT, we shall make use of the following
lemma.
Z
Lemma 5.3.7 Suppose f : X → [0, ∞] is such that f is finite. Then
X
µ({x : f (x) > n}) → 0 as n → ∞ and f is real valued almost everywhere,
i.e., µ({x : f (x) = ∞}) = 0.

Proof. Let A = {x ∈ X : f (x) = ∞}. Then A = ∩∞ n=1 An with An :=


{x : f (x) > n}, n ∈ N. Note that
Z Z Z Z
f≥ f= f χAn ≥ nχAn = nµ(An )
X An X X

for all n ∈ N. Hence,


Z
1
µ(An ) ≤ f →0 as n → ∞.
n X

Since µ(A) ≤ µ(An ) for all n ∈ N, it follows that µ(A) = 0.

Theorem 5.3.8 Suppose (fn ) is a sequence of complex measurable functions


∞ Z
X ∞
X
such that |fn | converges. Then fn converges a.e. to an integrable
n=1 X n=1
function and
∞ ∞ Z
Z !
X X
fn = fn .
X n=1 n=1 X

Proof. By assumption

Z X ∞ Z
X
|fn | = |fn | < ∞.
X n=1 n=1 X
116 Integral of Complex Measurable Functions

P∞ P∞
Hence, by Lemma 5.3.7, n=1 |fn | is finite a.e. In particular, n=1 fn
converges a.e. Let
n
X ∞
X
gn := fj and h = |fn |.
j=1 n=1
P∞
Thus, (gn ) converges to g := n=1 fn a.e. and |gn | ≤ h for all n ∈ N with
h ∈ L(µ). Hence, by DCT (Theorem 5.3.4),
Z Z ∞
Z X 
lim gn dµ = g dµ = fn .
n→∞ X X X n=1

But,
Z n
Z X  n Z
X ∞ Z
X
lim gn = lim fj = lim fj = fj .
n→∞ X n→∞ X n→∞ X X
j=1 j=1 j=1

R  P∞  P
∞ R
Thus, X n=1 n =
f j=1 X fj and the proof is completed.

Now we consider some more consequences of DCT.


Theorem 5.3.9 Let f ∈ L(µ). For n ∈ N, let

f (x), |f (x)| ≤ n,
fn (x) =
n, |f (x)| > n.

Then for each n ∈ N, fn is a bounded measurable function and


Z
|f − fn |dµ → 0 as n → ∞.
X

Proof. It is clear that |fn | ≤ n and |fn | ≤ |f | for all n ∈ N. Also, fn → f


pointwise. Indeed, since f is complex valued, for each x ∈ X, there exists
k ∈ N Rsuch that |f (x)| ≤ k, so that fn (x) = f (x) for all n ≥ k. Hence, by
DCT, X |f − fn |dµ → 0.

The following theorem is a consequence of Theorem 5.3.9.


Theorem 5.3.10 If f ∈ L(µ), then for every ε > 0, there exists a bounded
g ∈ L(µ) such that Z
|f − g|dµ < ε.
X
Dominated Convergence Theorem (DCT) 117

Exercise 5.3.11 Prove that the set {f ∈ L(µ) : f bounded} is a dense


subspace of L(µ). ♦

Recall
R that if f is a measurable function and E ∈ A with µ(E) = 0,
Rthen E |f |dµ = 0. Suppose µ(E) 6= 0, but µ(E) is small. Can we say that
E |f | dµ small? Not necessarily, as the following example shows.

Example 5.3.12 Let f : R → R be defined by


1/x2 , x 6= 0,

f (x) =
0, otherwise.
Let An = {x ∈ R : 0 < x < 1/n}, n ∈ N. Then we have m(An ) = 1/n and
Z Z Z
|f |dm = χAn |f |dm ≥ n2 χAn dm = n2 m(An ) = n.
An R R
R
Thus, m(An ) → 0, but An |f |dm → ∞. ♦
Such situations does not happen if f ∈ L(µ).
Theorem 5.3.13 Let f ∈ L(µ). Then for every ε > 0, there exists δ > 0
such that Z
A ∈ A, µ(A) < δ ⇒ |f |dµ < ε.
A

Proof. Let ε > 0 be given. If f is bounded, say, |f | ≤ M0 for some


M0 > 0, then for every A ∈ A,
Z
|f |dµ ≤ M0 µ(A)
A

so that Z
ε
µ(A) < ⇒ |f |dµ < ε.
M0 A
Now, let us consider the general case of f ∈ L(µ).R In this case, by Theorem
5.3.9, there exists a bounded g ∈ L(µ) such that X |f − g|dµ < ε/2. Hence,
for A ∈ A,
Z Z Z Z
ε
|f |dµ ≤ |g|dµ + |f − g|dµ ≤ |g|dµ + .
A A A A 2
Let M > 0 be such that |g| ≤ M . Then we have
Z Z
ε ε
|f |dµ ≤ |g|dµ + ≤ M µ(A) + .
A A 2 2
R
Thus, µ(A) < ε/2M implies A |f |dµ < ε.
118 Integral of Complex Measurable Functions

Corollary 5.3.14 Let f ∈ L(µ).Z If (An ) is a sequence of sets in A such


that µ(An ) → 0 as n → ∞, then |f |dµ → 0 as n → ∞.
An

Proof. Let (An ) be a sequence of sets in A such that µ(An ) → 0 as


n → ∞. Let ε > 0 be given. By Theorem 5.3.13, there exists δ > 0 such
that Z
A ∈ A, µ(A) < δ ⇒ |f |dµ < ε. (∗)
A

Since µ(An ) → 0, there exists N ∈ N such that

µ(An ) < δ ∀ n ≥ N.
R
Hence, by (∗) above, An |f |dµ < ε for all n ≥ N .

Corollary
R 5.3.15 Let f ∈ L(µ) and An = {x ∈ X : |f (x)| > n} for n ∈ N.
Then An |f |dµ → 0 as n → ∞.

Proof. By Lemma 5.3.7, Rwe know that µ(An ) → 0 as n → ∞. Hence, by


Corollary 5.3.14, we obtain An |f |dµ → 0 as n → ∞.
R
Recall from Theorem 5.1.8(b) that, if f ∈ L(µ) is such that E f = 0 for
all E ∈ A, then f = 0 a.e. Now, in the case when X is an interval and µ is
the Lebesgue measure, then we obtain the same conclusion with measurable
sets replaced by subintervals.

Theorem 5.3.16 Let J be an interval and let f Zbe a complex Lebesgue


x
integrable function on J. For any given a ∈ J, if f dm = 0 for every
a
x ∈ J, then f = 0 a.e. on J
Z
Proof. First we observe that f dm = 0 for every interval I ⊆ J. In-
I
deed, for every open interval with end points c, d, where c < d,
Z d Z d Z c
f dm = f dm − f dm = 0.
c a a
R
Next, we show that G∩J f dm = 0 for every open set G ⊆ R:

S∞Let G be an open set in R. We know that, G can be written as G =


n=1 In , where {In } is a countable family of disjoint open intervals, which
Lp Spaces 119

may also contain empty sets. Let Jn := In ∩ J. Then G ∩ J = ∞


S
n=1 Jn , and
{Jn } is a countable disjoint family of intervals or emplty sets. Let
fn = χJn f, n ∈ N.
Then we have Z Z
fn dm = f dm = 0 ∀ n ∈ N. (1)
G∩J Jn
Pk
Let hk := n=1 fn , k ∈ N. Then hk → f pointwise on G ∩ J and
k
X k
X
|hk | ≤ |fn | = χIn |f | = |f | ∀ k ∈ N.
n=1 n=1

Hence, by DCT and (1),


Z Z k Z
X
f dm = lim hk dm = lim fn dm = 0. (2)
G∩J k→∞ G∩J k→∞
n=1 G∩J

Now, let E be any measurable subset of J. Then, for every n ∈ N there


exists an open set Gn ⊇ E such that m(Gn \ E) < 1/n. Hence, by (2),
Z Z Z

f dm = f dm − f dm


E Gn ∩J (Gn \E)∩J
Z

= f dm

(Gn \E)∩J
Z
≤ |f |dm.
(Gn \E)∩J
R
Since m(Gn \ E) → 0, by Corollary 5.3.14, (Gn \E)∩J |f |dm → 0. Therefore,
R
we can conclude that E f dm = 0. This is true for every measurable set
E ⊆ J. Therefore, by Theorem 5.1.8(b), f = 0 a.e.

5.4 Lp Spaces
Let (X, A, µ) be a measure space. For 1 ≤ p < ∞, we denote by Lp (X, A, µ)
the set of all complex measurable functions f on X such that |f |p is inte-
grable. In short, we may denote this set by Lp (µ) or by Lp (X). Thus,
Z
p
f ∈ L (µ) ⇐⇒ |f |p dµ < ∞.
X
120 Integral of Complex Measurable Functions

Definition 5.4.1 A measurable function f on the measure space (X, A, µ)


is said to be an essentially bounded function if there exists Mf > 0 such
that |f | ≤ Mf a.e. on X. ♦

The set of all essentially bounded functions is denoted by L∞ (X, A, µ)


or simply by L∞ (µ). Thus,
f ∈ L∞ (µ) ⇐⇒ ∃ Mf > 0 such that |f | ≤ Mf a.e. onX.

For a complex measurable function f on X, we denote


Z 1/p
kf kp := |f |p dµ < ∞ if 1 ≤ p < ∞,
X
and
kf k∞ := inf{Mf > 0 : |f | ≤ Mf a.e. on X}.
Note that, for a complex measurable function f on X, kf kp can be ∞. Thus,
for 1 ≤ p ≤ ∞, and for any complex measurable function f on X,
f ∈ Lp (µ) ⇐⇒ kf kp < ∞.
Let
Zp := {f ∈ Lp (µ) : f = 0 a.e}.
Theorem 5.4.2 For 1 ≤ p ≤ ∞, Lp (µ) is a vector space over C and Zp is
a subspace of Lp (µ).

Proof. Let 1 ≤ p ≤ ∞. Note that for f, g ∈ Lp (µ),


|f + g| ≤ |f | + |g| ≤ 2 max{|f |, |g|}.
Hence,
kf + gk ≤ 2 max{kf k∞ , kgk∞ }
and for 1 ≤ p < ∞,
|f + g|p ≤ 2p max{|f |p , |g|p }
so that Z nZ Z o
p p p
|f + g| dµ ≤ 2 max |f | dµ, |g|p dµ .
X X X
Hence, f + g ∈ Lp (µ).
It is easy to see that α f ∈ p
for every f ∈ Lp (µ)
L (µ)
and α ∈ C. Using these, all axioms of a vector space can be verified. It
can also be verified that, for f, g ∈ Lp (µ), f = 0 a.e. and g = 0 a.e. imply
f + g = 0 a.e., and αf = 0 a.e. for every α ∈ C. Hence, Zp is a subspace of
Lp (µ).
Lp Spaces 121

Consider the quotient space

Lp (µ) := Lp (µ)/Zp .

Notation: As in the case of the space L(µ) considered in Remark 5.1.9, we


use the same symbol f for f ∈ Lp (µ) and for the corresponding equivalent
class [f ]. If Ω is a Lebesgue measurable subset of R, then we shall denote the
spaces Lp (Ω, MΩ , m) and Lp (Ω, MΩ , m) by Lp (Ω) and Lp (Ω), respectively.

Note that the spaces L(µ) and L(µ) introduced earlier are the spaces
L1 (µ) and L1 (µ), respectively.

We shall show that

• f 7→ kf kp is a norm on Lp (µ) and the corresponding metric is complete.

5.4.1 Hölder’s and Minkowski’s inequalities


One of the crucial inequality required is the Young’s inequality:

Lemma 5.4.3 (Young’s inequality) Let a, b be non-negative real numbers


and p, q ∈ (1, ∞) be such that p1 + 1q = 1. Then

ap bq
ab ≤ + .
p q

Proof. Recall from real analysis that the function ϕ : R → R defined by

ϕ(x) = ex , x ∈ R,

is convex, that is, for every x, y ∈ R and 0 < λ < 1,

ϕ(λx + (1 − λ)y) ≤ λϕ(x) + (1 − λ)ϕ(y).

Taking λ = 1/p we have 1 − λ = 1/q and


x
+ yq ex ey
ep ≤ + .
p q
x y
Now, taking x > 0 and y > 0 such that a = e p and b = e q , that is,
p q
x = ln(ap ) and y = ln(bq ), we obtain ab ≤ ap + bq .
122 Integral of Complex Measurable Functions

Remark 5.4.4 Observe that the Young’s inequality for p = 2 and q = 2


follows from the identity (a − b)2 = a2 + b2 − 2ab. ♦
In fact, Young’s inequality considered in Lemma 5.4.3 is a particular case
of Jensens’s inequality.
Theorem 5.4.5 (Jensen’s inequality) Let f be a real valued integrable
function on a probability space (X, A, µ), that is, with µ(X) = 1, and I be
an open intervalRsuch that f (x) ∈ I for all x ∈ X. Let g : I → R be a convex
function. Then X f dµ ∈ I and
Z  Z
g f dµ ≤ g ◦ f dµ.
X X

Proof. For a, b, c ∈ I with a < b < c, we have


b−a
b = (1 − λ)a + λc with λ = .
c−a
By the convexity of g,
g(b) ≤ (1 − λ)g(a) + λg(c),
so that
g(b) − g(a) ≤ λ[g(c) − g(a)]. (1)
Thus,
g(b) − g(a) g(c) − g(a)
≤ . (2)
b−a c−a
c−b
Writing µ = 1 − λ = c−a , (1) implies
g(b) − g(a) ≤ (1 − µ)[g(c) − g(a)]
so that µ[g(c) − g(a)] ≤ g(c) − g(b), that is,
g(c) − g(a) g(c) − g(b)
≤ . (3)
c−a c−b
The relations (2) and (3) give
g(b) − g(a) g(c) − g(b)
≤ . (4)
b−a c−b
Note that the right hand side of (4) is independent of a. For a fixed b, let
β be the supremum of the left hand side of (4) as a varies such that a < b.
Then we have
g(b) − g(a) g(c) − g(b)
≤β≤ (5)
b−a c−b
Lp Spaces 123

for all a ∈ I with a < b. The above relations can also be written as
g(b) − g(a) g(b) − g(c)
≤β≤ , (6)
b−a b−c
which is also true for all c ∈ I with b < c. The relations in (5) and (6) show
that
g(b) − g(t) ≤ β(b − t) ∀ t ∈ I.
R
Now, taking b = X f dµ and t = f (x) for x ∈ X, we obtain
Z  Z 
g f dµ − g(f (x)) ≤ β f dµ − f (x) .
X X

On integration, taking into account that µ(X) = 1, we get


Z  Z Z Z 
g f dµ − (g ◦ f )dµ ≤ β f dµ − f dµ = 0.
X X X X

Thus, we obtain the required inequality.

To see that the Young’s inequality is a particular case of Jensens’s in-


equality, let x1 , . . . xn be real numbers and λ1 , . . . λn be non-negative real
numbers such that λ1 + · · · + λn = 1. Then, considering the convex func-
tion g(t) = et , X = {x1 , . . . , xn } with measure µ on the power set of X by
defining µ({xi }) = λi , i = 1, . . . , n, and taking ai = exi , i = 1, . . . , n, we
obtain Z  X n  Pn
g xdµ = g xi λi = e i=1 xi λi = aλ1 1 · · · aλnn ,
X i=1
Z n
X n
X
g ◦ f dµ = g(xi )λi = λi ai .
X i=1 i=1
Thus, by Jensens’s inequality, for non-negative real numbers a1 , . . . , an and
λ1 , . . . , λn with λ1 + · · · + λn = 1, we have

aλ1 1 · · · aλnn ≤ λ1 a1 + · · · λn an .
2
Taking n = 2, p = 1/λ1 , q = 1/λ2 , a = aλ1 1 , b = aλ2 we obtain the Young’s
inequality in Lemma 5.4.3.
Recall the notation
Z 1/p
p
kf kp := |f | dµ for 1 ≤ p < ∞,
X
124 Integral of Complex Measurable Functions

for complex measurable functions f on X. In the following, for 1 ≤ p ≤ ∞,


we take q ∈ [1, ∞] such that

1 1
+ =1
p q

with the convention that 1/∞ = 0.

Theorem 5.4.6 (Hölder’s inequality) Let f and g be complex measurable


functions on X and 1 ≤ p ≤ ∞. Then
Z
|f g|dµ ≤ kf kp kgkq ,
X

1 1
where q is such that p + q = 1.

Proof. For the case p = 1 and p = ∞, it is easy to see that the inequality
holds. Hence, assume that 1 < p < ∞. Then 1 < q < ∞. First we observe
that if one of kf kp and kgkq is zero or infinity, then the inequality holds.
Hence, we assume that 0 < kf kp < ∞ and 0 < kgkq < ∞. For x ∈ X,
taking a = |f (x)|/kf kp and b = |g(x)|/kgkq in the Young’s inequality, we
have
|f (x)g(x)| |f (x)|p |g(x)|q
≤ + .
kf kp kgkq pkf kpp qkgkqq
Now, taking integrals over X, we get
p |g|q
R R
|f |
Z
1
|f g| ≤ X p + X q = 1.
kf kp kgkq X pkf kp qkgkq
Z
Hence, |f g| ≤ kf kp kgkq , which completes the proof.
X

Theorem 5.4.7 (Minkowski’s inequality) Let f and g be complex measur-


able functions on X and 1 ≤ p ≤ ∞. Then

kf + gkp ≤ kf kp + kgkp .

Proof. We observe that the inequality for p = 1 and p = ∞ follow from


the inequality |f + g| ≤ |f | + |g|. So, let 1 < p < ∞ and q > 1 be such that
1 1
p + q = 1.
Lp Spaces 125

We note that the inequality holds if kf + gkp = 0. Hence, assume that


kf + gkp 6= 0. Note that
Z Z
p
|f + g| dµ = |f + g|p−1 |f + g|dµ
X ZX Z
p−1
≤ |f + g| |f |dµ + |f + g|p−1 |g|dµ.
X X
By Hölder’s inequality, we have
Z Z 1/q
p−1
|f + g| |f | ≤ kf kp |f + g|(p−1)q = kf kp kf + gkp/q
p ,
X X
Z Z 1/q
|f + g|p−1 |f | ≤ kgkp |f + g|(p−1)q = kf kp kf + gkp/q
p .
X X
Thus, Z  
kf + gkpp = |f + g|p dµ ≤ kf kp + kgkp kf + gkp/q
p .
X
p/q
Now canceling out kf + gkp , as kf + gkp 6= 0, we obtain the required
inequality.

Now, the proof of the following theorem is immediate.


Theorem 5.4.8 For 1 ≤ p ≤ ∞, Lp (µ) is a vector space over C and
f 7→ kf kp
is a norm on on Lp (µ). In particular,
(f, g) 7→ kf − gkp
is a metric on Lp (µ).
Remark 5.4.9 If X = N, A = 2N and µ is the counting measure, then
every complex valued function on X is measurable. Hence, in this case, for
any p ∈ [1, ∞], we have Zp = {0} and
Lp (µ) = Lp (µ) = `p (N).
Here, `∞ (N) is the space of all bounded sequences in CP and for 1 ≤ p < ∞,
`p (N) is the space of all sequences (an ) in C such that ∞ p
n=1 |an | < ∞.
Also, if X = {1, . . . , k} for some k ∈ N, A = 2N and µ is the counting
measure, then every complex valued function on X is measurable, and hence,
for any p ∈ [1, ∞], Zp = {0} and
Lp (µ) = Lp (µ) = Ck .

126 Integral of Complex Measurable Functions

5.4.2 Completeness of Lp (µ)


We prove that metric induced by the norm k·kp on Lp (µ) is complete. Before
that, let us observe the following result which will be used in the sequel.
Lemma 5.4.10 Let 1 ≤ p < ∞ and let (fn ) be a Cauchy sequence in Lp (µ)
having a subsequence which converges almost everywhere to a measurable
function f . Then f ∈ Lp (µ) and (fn ) converges to f in Lp (µ).

Proof. Suppose (fn ) is a Cauchy sequence in Lp (µ) with 1 ≤ p < ∞


having a subsequence (fnk ) which converges a.e. to a measurable function
f . Then for each n ∈ N, |fnk − fn | → |f − fn | a.e. on X. Therefore, by
Fatou’s lemma,
Z Z
|f − fn |p ≤ lim inf |fnk − fn |p .
X k X

Now, let ε > 0 be given. Since (fn ) is a Cauchy sequence in Lp (µ), there
exists N ∈ N be such that kfnk − fn kp < ε for all n, k ≥ N . Thus,
Z Z
p
|f − fn | ≤ lim inf |fnk − fn |p < εp ∀ n ≥ N,
X n X

showing that f ∈ Lp and kf − fn kp → 0 as n → ∞.

We shall also make use of the following result.


Theorem 5.4.11 Let Ω be any nonempty set. Then the map

(f, g) 7→ kf − gk∞ := sup |f (t) − g(t)|, f ∈ B(Ω),


t∈Ω

is a complete metric on B(Ω), the space of all bounded complex valued func-
tions on Ω.

Proof. Let (fn ) be a Cauchy sequence in B(Ω). Then for every t ∈ Ω, the
sequence (fn (t)) is a Cauchy sequence of complex numbers. Hence, (fn (t))
converges for every t ∈ Ω. Let

f (t) := lim fn (t), t ∈ Ω.


n→∞

We show that f ∈ B(Ω) and kfn − f k∞ → 0.


Now, let k ∈ N be such that kfn − fk k∞ ≤ 1 for all n ≥ k. Then we have

|fn (t)| ≤ kfn − fk k∞ + kfk k∞ < 1 + kfk k∞


Lp Spaces 127

for all n ≥ k and for all t ∈ Ω. Hence,


|f (t)| ≤ max{1 + kfk k∞ , kf1 k∞ , kf2 k∞ , . . . , kfk k∞ } ∀t ∈ Ω
showing that f ∈ B(Ω). Next, let ε > 0 be given. Let N ∈ N be such that
kfn − fm k∞ ≤ ε for all n, m ≥ N . Let t ∈ Ω and let m ≥ N be such that
|f (t) − fm (t)| < ε. Then we have
|f (t) − fn (t)| ≤ |f (t) − fm (t)| + |fm (t) − fn (t)| < 2ε
for all n ≥ N . Note that this N is independent of t. Hence we have proved
that kf − fn k∞ < 2ε for all n ≥ N . Thus, the Cauchy sequence (fn )
converges in B(Ω).

Theorem 5.4.12 For 1 ≤ p ≤ ∞, the metric (f, g) 7→ kf − gkp on Lp (µ)


is complete.

Proof. First we consider the case 1 ≤ p < ∞: Let (fn ) be a Cauchy


sequence in Lp (µ). By Lemma 5.4.10, it is enough to show that (fn ) has a
subsequence which converges almost every where.
Since (fn ) is a Cauchy sequence, for each i ∈ N, there exists ni ∈ N such
that kfn − fni kp < 1/2i for all n ≥ ni . Without loss of generality we may
assume that ni ≤ ni+1 for all i ∈ N. Then we have kfni+1 − fni kp < 1/2i for
all i ∈ N. Note that
k−1
X
fnk = fn1 + (fni+1 − fni ).
i=1
P∞
Thus, it is enough to show that i=1 (fni+1 −fni ) converges a.e. on X. Now,
let
k
X X∞
gk := |fni+1 − fni |, g := |fni+1 − fni |.
i=1 i=1
Then gk (x) → g(x) asZ k → ∞Zfor every x ∈ X, and by Monotone Conver-
gence Theorem, lim gkp = g p . But,
k→∞ X X

Z 1/p k
gkp
X
= kgk kp ≤ kfni − fni−1 kp ≤ 1.
X i=1

gp
R
Hence, X < ∞. Therefore, by Lemma 5.3.7, g(x) < ∞ a.e. on X. Thus
P ∞
i=1 (fni+1 − fni ) converges a.e. on X, completing the proof for the case
1 ≤ p < ∞.
128 Integral of Complex Measurable Functions

Next we consider the case p = ∞. Let (fn ) be a Cauchy sequence in


L∞ (µ).Then the sets

Ak := {x ∈ X : |fk (x)| > kfk k∞ }

and
Bm,n := {x ∈ X : |fn (x) − fm (x)| > kfn − fm k∞ }
are of measure zero for every k, m, n ∈ N. Hence, the set
[   [ 
E := Ak ∪ Bm,n
k m,n

is of measure zero, and for each x ∈ Ω := E c , |fk (x)| ≤ kfk k∞ for all k ∈ N,
and
|fn (x) − fm (x)| ≤ kfn − fm k∞ ∀ m, n ∈ N.
Since (fn ) is cauchy in L∞ (µ), it follows that (f˜n ) with f˜n = fn |Ω is a
Cauchy sequence in B(Ω) with respect to k · k∞ . By Theorem 5.4.11, B(Ω)
is complete with respect to k · k∞ . Hence, there exists f˜ ∈ B(Ω) such that
kf˜n − f˜k∞ → 0 as n → ∞. Defining f (x) = f˜(x) for x ∈ Ω and f (x) = 0
for x 6∈ Ω, it follows that f ∈ L∞ (µ), and kfn − f k∞ → 0 as n → ∞.

Note that, as part of the proof of the above theorem, we have proved
the following.
Proposition 5.4.13 If 1 ≤ p < ∞, then every Cauchy sequence in Lp (µ)
has a subsequence which converges almost everywhere.
In particular, if (fn ) is a sequence in Lp (µ) which converges to f in
Lp (µ), then (fn ) has a subsequence which converges to f a.e.

Before closing this subsection, let us observe certain relations between


Lp -spaces.
Theorem 5.4.14 If µ(X) < ∞, then for 1 ≤ p ≤ r ≤ ∞,

Lr (µ) ⊆ Lp (µ).

Proof. Let µ(X) < ∞. Suppose f ∈ L∞ (µ) and 1 ≤ p < ∞. Then we


have Z
|f |p dµ ≤ kf kp∞ µ(X).
X
Hence,
f ∈ Lp (µ) and kf kp ≤ kf k∞ [µ(X)]1/p .
Lp Spaces 129

In particular,
L∞ (µ) ⊆ Lp (µ) for every p ∈ [1, ∞).
1 1
Also, for p, q ∈ (1, ∞) with + = 1, by Hölder’s inequality,
p q
Z Z 1/p
|f | dµ ≤ |f |p dµ [µ(X)]1/q .
X X

Thus,
Lp (µ) ⊆ L1 (µ) for every p ∈ [1, ∞).
Next, let 1 < p < r < ∞ and f ∈ Lr (µ). Then, again by Hölder’s inequality,
we have Z Z 1/s
|f |p dµ ≤ |f |ps dµ [µ(X)]1/t ,
X X
where s = r/p and t > 1 is such that 1s + 1t = 1. Thus,
Z Z p/r
p
|f | dµ ≤ |f |r dµ [µ(X)]1−p/r
X X

so that
f ∈ Lp (µ) and kf kp ≤ kf kr [µ(X)]1/p−1/r .
Hence Lr (µ) ⊆ Lp (µ).

Exercise 5.4.15 Suppose 1 ≤ s ≤ p ≤ ∞. If f ∈ Lp (µ) is such that


µ({x ∈ X : f (x) 6= 0}) < ∞, then f ∈ Ls (µ). ♦

If µ(X) = ∞, then it is not necessary to hold the relations Lp (µ) ⊆ Lr (µ)


or Lr (µ) ⊆ Lp (µ) (see Problem 13). However, we have the following result.
Theorem 5.4.16 If for 1 ≤ p ≤ r ≤ ∞, then

`p (N) ⊆ `r (N).

Further, if p < r, then the above inclusion is proper.

Proof. Suppose (an ) ∈ `p (N) for 1 ≤ p < ∞. Then, (an ) converges to 0.


In particular, (an ) is a bounded sequence. Hence, `p (N) ⊆ `∞ (N) for every
p with 1 ≤ p < ∞.
Next, let 1 ≤ p < r < ∞ and (an ) ∈ `p (N). Let M > 0 be such that
|an | ≤ M for all n ∈ N. Then

X ∞
X ∞
X
|an |r = |an |p |an |r−p ≤ M r−p |an |p .
n=1 n=1 n=1
130 Integral of Complex Measurable Functions

This shows that `p (N) ⊆ `r (N). To see the last part, let 1 ≤ p < r < ∞,
1
and consider the sequence (1/n p ). Since ∞
P 1 P∞ 1
n=1 n diverges and n=1 nr/p
converges, we have (1/n1/p ) ∈ `r (N) \ `p (N). Also, if an = 1 for all n ∈ N,
then (an ) ∈ `∞ (N) \ `p (N) for any p ∈ [1, ∞).

More generally, for any denumerable set X with counting measure µ, we


have the proper inclusions Lp (µ) ⊆ Lr (µ) (see Problem 14).

We close this subsection with another application of DCT in connection


with Lp functions. We shall also make use of this result in the context of
Fourier transform in Chapter ??.
We may recall from calculus (see [7]) that if Rf : R → R is Riemann
n
integrable on [−n, n] for every n ∈ N and if lim −n |f (x)|dx exists, then
R∞ n→∞
the integral −∞ f (x)dx, called an improper inteegral of f , is defined by
Z ∞ Z n
f (x)dx = lim f (x)dx.
−∞ n→∞ −n

In view of this, we may ask the following question:

Suppose f : R → R is a Lebesgue measurable function such


that Rf is Lebesgue integrable over [−n, n] for every n ∈ N and
lim [−n,n] |f |dm exists. Then, is it true that f ∈ L1 (R) and
n→∞
Z Z
f dm = lim |f |dm?
R n→∞ [−n,n]

The answer is in affirmative. In fact, something more is true as the following


proposition shows.

Proposition 5.4.17 Let (X, A, µ) be a measure space and E ∈ A. For each


n ∈ N, let En ∈ A be such that µ(En ) < ∞ and χEn → 1 pointwise. Let
1 ≤ p < ∞, f ∈ Lp (µ) and fn := χEn f for n ∈ N. Then, for each n ∈ N,
fn ∈ Ls (µ) ∩ Lp (µ) for all s ∈ [1, p] and
Z
|f − fn |p dµ → 0 as n → ∞.
X
Z Z
In particular, if f ∈ L1 (R), then f dµ = lim f dµ.
X n→∞ E
n
Lp Spaces 131

Proof. Let n ∈ N and 1 ≤ p < ∞. Then


Z Z Z
p p
|fn | dµ = |f | dµ ≤ |f |p dµ = kf kpp
X En X

so that fn ∈ Lp (µ). Also, since µ({x ∈ X : fn (x) 6= 0} ≤ µ(En ) < ∞, it can


be seen that fn ∈ Ls (µ) for every s ∈ [1, p] (see Exercise 5.4.15). Further,
Z Z
|f − fn |p dµ = (1 − χEn )|f |p dµ,
X X

(1 − χEn )|f |p → 0 pointwise and (1 − χEn )|f |p ≤ |f |p with |f |p ∈ L1 (µ).


Therefore, by DCT (Theorem 5.3.2),
Z
|f − fn |p dµ → 0 as n → ∞.
X

If f ∈ L1 (µ), then we have


Z Z Z Z
f dµ − f dµ = (f − fn )dµ ≤ |f − fn |dµ.


X En X X
R R R
Since X |f − fn |dµ → 0 as n → ∞, we obtain X f dµ = lim
n→∞ En
f dµ.

5.4.3 Denseness of Cc (Ω) in Lp (Ω) for 1 ≤ p < ∞


We may recall that the space C[a, b] of all complex valued continuous func-
tions defined on [a, b] is a complete metric space with respect to the metric

d(f, g) := sup |f (x) − g(x)|, f, g ∈ C[a, b].


x∈[a,b]

It can be easily shown that C[a, b] is not complete with respect to the metric
Z b
dp (f, g) := |f (x) − g(x)|p dx, f, g ∈ C[a, b],
a

for 1 ≤ p < ∞. Note that C[a, b] ⊆ Lp [a, b]. So, it is natural to ask whether
C[a.b] is dense in Lp [a, b]. We shall answer this affirmatively in a slightly
more general context.
Let Ω be a (Lebesgue) measurable subset of R. By Cc (Ω) we mean the
vector space of all continuous functions f : Ω → C such that there exists a
compact subset K of Ω with f (x) = 0 for all x 6∈ K. Clearly, if Ω itself is
compact, then Cc (Ω) is C(Ω), the space of all continuous functions on Ω.
132 Integral of Complex Measurable Functions

The question is whether every f ∈ Lp (Ω) can be approximated by a


sequence of functions from Cc (Ω). The answer is known to be affirmative if
Ω is a locally compact subset of R, that is, if for each x ∈ Ω, there exists an
open set in Ω whose closure is compact. The proof is quite involved for a
general locally compact subset (cf. Rudin [11]). However, for certain special
cases, the result can be proved rather easily. In the following we shall prove
the result when Ω is a countable disjoint union of intervals. In particular,
Ω can be an open set in R or an interval of any of the following forms:
[a, b], [a, ∞), (−∞, b].
First we prove a result for characteristic functions of intervals.

Proposition 5.4.18 Let Ω be a measurable subset of R and J ⊆ Ω be an


interval of finite length. Let 1 ≤ p < ∞. Then for every ε > 0, there exists
fε ∈ Cc (Ω) such that Z
|χJ − fε |p dm < ε.

Proof. Let a and b be the end points of J with a < b. Let δ > 0 be such
that a + 4δ < b so that a + 2δ < b − 2δ and [a + δ, b − δ] ⊆ J. Let gδ : R → R
be defined by

 1 if x ∈ [a + 2δ, b − 2δ],
 1

gδ (x) = δ (x − a − δ) if x ∈ [a + δ, a + 2δ],
1
(b − δ − x) if x ∈ [b − 2δ, b − δ].
 δ


0 if x 6∈ [a + δ, b − δ].

Then we see that gδ ∈ Cc (Ω) with gδ (x) = 0 for all x 6∈ [a + δ, b − δ]. Hence,
Z Z a+2δ
p 1
|χJ (x) − gδ (x)| dx = 2 [1 − (x − a − δ)]p dx
J a+δ δ
Z 1
= 2δ y p dy
0
= 2δ/(p + 1).

Take δ > 0 be such that p+1 < ε and define fε on Ω by

gδ (x), x ∈ J,
fε (x) =
0, x∈6 J.

− fε |p dm < ε.
R
Then we have fε ∈ Cc (Ω) and Ω |χJ
Lp Spaces 133

Theorem 5.4.19 Let 1 ≤ p < ∞ and Ω be a countable disjoint union of


intervals. Then Cc (Ω) is dense in Lp (Ω).

Proof. Let f ∈ Lp (Ω) and ε > 0. It is enough to prove for the case when
f is a non-negative real valued function. In that case, we know that there
is an increasing sequence (ϕn ) of non-negative simple measurable functions
on Ω such that ϕn → f pointwise. We observe that, for each n ∈ N,

ϕn ∈ Lp (Ω), |f − ϕn |p ≤ 2p |f |p , f − ϕn → 0 pointwise.

Hence, by DCT,
Z Z
p
|f − ϕn | dm = (f − ϕn )p dm → 0.
Ω J

Thus, given ε > 0, there exists a non-negative simple measurable function


ϕ such that
kf − ϕkp < ε.
Hence, it is enough to prove that, corresponding to the above ϕ, there exists
g ∈ Cc (Ω) such that kϕ − gkp < ε. Note that the simple function ϕ is of
the form ϕ = ki=1 αi χAi for some αi ∈ [0, ∞) and measurable sets Ai ⊆ J
P
with m(Ai ) < ∞ for i = 1, . . . , k. Hence, it is enough to prove that for every
measurable set E ⊆ J with m(E) < ∞, there exists g ∈ Cc (J) such that
kχE − gkp < ε.
We know that, there exists an open set G ⊆ R such that E ⊆ G and

m(G) ≤ m(E) + ε, m(G \ E) < ε.

Let G0 = G ∩ Ω. Then E ⊆ G0 and we have

χG0 = χE + χ(G0 \E)

so that
Z Z
kχG0 − χE kpp = |χG0 − χE |p dm = χ(G0 \E) dm = m(G0 \ E) < ε.
Ω Ω

Since G is a countable disjoint union of intervals


S of finite length, by the as-
sumption on Ω, G0 can be written as G0 = n In , where {In } is a countable
disjoint family of intervals of finite length. Since m(G) < ∞,

X [ 
`(Jn ) = m In = m(G0 ) < ∞.
i=1 n
134 Integral of Complex Measurable Functions

P∞ Sk
Let k ∈ N be such that i=k+1 `(Jn ) < ε and Gk = n=1 In . Then

X
Gk ⊆ G, m(G0 \ Gk ) = `(In ) < ε.
i=k+1

Hence,
Z Z
kχG0 − χGk kpp = p
|χG0 − χGk | dm = χ(G0 \G ) dm = m(G0 \ Gk ) < ε.
k
Ω J
Thus, by Minkowski inequality,
kχE − χGk kp ≤ kχE − χG0 kp + kχG0 − χGk kp < 2ε1/p .

But, χGk = ki=1 χIn . Hence, it is enough to prove that for each interval
P
J ⊆ Ω of finite length, there exists g ∈ Cc (Ω) such that kχJ − gkp < ε. This
is proved in Proposition 5.4.18.

5.5 Indefinite Integral and its Derivative


We may recall from calculus that if f : [a, b] → R is a Riemann integrable
function, then an indefinite integral of f is the function g : [a, b] → R defined
by Z x
g(x) = c + f (t)dt, x ∈ [a, b],
a
for some c ∈ R. We know that g is continuous, and if f ∈ C[a, b], then g is
differentiable and g 0 = f . A natural question is:

What can we say about g if f is known to be only Lebesgue


integrable?

In this section we investigate this issue when f is a real or complex Lebesgue


integrable function.
We denote by K the field R of real numbers or the filed C of complex
numbers, and by L[a, b] the set of all integrable K-valued functions defined
on [a, b], where [a, b] is endowed with the Lebesgue measure. Also, we shall
R Rd
denote the integral [c,d] f dm by c f dm.

Definition 5.5.1 For f ∈ L[a, b] and c ∈ K, the function g : [a, b] → K


defined by Z x
g(x) = c + f dm, x ∈ [a, b],
a
is called an indefinite integral of f (with respect to m). ♦
Indefinite Integral and its Derivative 135

As in the case of the result in calculus, we have the following theorem.

Theorem 5.5.2 Let f ∈ L[a, b] and g : [a, b] → K be defined by


Z x
g(x) = f dm, x ∈ [a, b].
a

Then g is continuous.

Proof. Let x ∈ [a, b) and (xn ) be a sequence in [x, b] such that xn → x


as n → ∞. Then we have
Z xn Z b
g(xn ) = f dm = χ[a,xn ] f dm.
a a

Note that χ[a,xn ] f is measurable, |χ[a,xn ] f | ≤ |f | and χ[a,xn ] f → χ[a,x] f point-


wise. Hence, by DCT,
Z b Z b
g(xn ) = χ[a,xn ] f dm → χ[a,x] f dm = g(x)
a a

as n → ∞. Similarly, if x ∈ (a, b] and (xn ) is a sequence in [a, x] such that


xn → x as n → ∞, then we can see that g(xn ) → g(x) as n → ∞. Thus, g
is a continuous function.

In fact, something more is true, as Theorem 5.5.4 below shows.


Definition 5.5.3 A function ϕ : [a, b] → K is said to be absolutely con-
tinuous if for every ε > 0, there exists δ > 0 such that for every family
{Ii : i = 1, . . . , n} of non-overlapping subintervals of [a, b],
n
X n
X
`(Ii ) < δ ⇒ |ϕ(xi ) − ϕ(yi )| < ε,
i=1 i=1

where xi and yi are the endpoints of Ii , i = 1, . . . , n. ♦


The following statements can be easily verified.

• If ϕ is a Lipschitz function, that is, if there exists c > 0 such that


|ϕ(x)−ϕ(y)| ≤ c|x−y| for all x, y ∈ [a, b], then f absolutely continuous.

• If ϕ is absolutely continuous, then ϕ is uniformly continuous.


136 Integral of Complex Measurable Functions

Theorem 5.5.4 Let f ∈ L[a, b] and g : [a, b] → K be defined by


Z x
g(x) = f dm, x ∈ [a, b].
a

Then g is absolutely continuous.

Proof. Let {Ii : i = 1, . . . , n} be a family of non-overlapping subintervals


of [a, b] with left and right endpoints of Ii as xi and yi , respectively. Then
we have
Xn n Z yi
X X n Z yi Z

|g(xi ) − g(yi )| =
f dm ≤
|f |dm = |f |dm,
i=1 i=1 xi i=1 xi A

where A = ni=1 Ii . Note that m(A) =


S Pn
i=1 `(Ii ). Hence, by Theorem
5.3.13, for every ε > 0, there exists δ > 0 such that
n
X n
X
`(Ii ) < δ ⇒ |g(xi ) − g(yi )| < ε.
i=1 i=1

Thus, g is absolutely continuous.

Next we show that if g is an indefinite integral of f ∈ L[a, b], then g is


differentiable almost everywhere and g 0 = f a.e. That is, we shall prove the
follwoing theorem, which is known as a fundamental theorem of Lebesgue
integration
Theorem 5.5.5 (Fundamental theorem of Lebesgue integration) Let
f ∈ L[a, b] and g : [a, b] → K be defined by
Z x
g(x) = f dm, x ∈ [a, b].
a

Then g is differentiable almost everywhere and g 0 = f a.e.

We shall make use of the following lemma, whose proof is relegated to


the Appendix to this chapter (Theorem 5.7.4).
Lemma 5.5.6 If ϕ : [a, b] → R is a monotonically increasing function,
then ϕ is differentiable almost everywhere, ϕ0 is non-negative and Lebesgue
measurable, and
Z b
ϕ0 dm ≤ ϕ(b) − ϕ(a).
a
Indefinite Integral and its Derivative 137

Proof of Theorem 5.5.5. Without loss of generality, we may assume


that f is real valued. Since f ∈ L[a, b], both f + and f − are non-negative
integrable functions, and hence, the functions g1 and g2 defined by
Z x Z x
g1 (x) = +
f dm and g2 (x) = f − dm, x ∈ [a, b],
a a

are non-negative, monotonically increasing, and g = g1 − g2 . By Lemma


5.5.6, g1 and g2 are differentiable almost everywhere, g10 and g20 are non-
negative and Lebesgue measurable, and
Z b
gi0 dm ≤ gi (b) − gi (a) for i = 1, 2.
a

Hence, g is differentiable almost everywhere, |g 0 | ≤ g10 + g20 and


Z b Z b Z b
0
|g |dm ≤ g10 dm + g20 dm ≤ g1 (b) − g1 (a) + g2 (b) − g2 (a).
a a a

Thus, g 0 ∈ L[a, b]. Now, we show that g 0 = f a.e. We split the proof into
two cases:
Case (1): f is bounded. Let M > 0 be such that |f (x)| ≤ M for all
x ∈ (a, b). Let (tn ) be a sequence real numbers in (0, 1] such that tn → 0 as
n → ∞. By We know that

g(x + tn ) − g(x)
→ g 0 (x) a.e.
tn

Let us extend the function f to [a, b+1] by defining its value as 0 on (b, b+1]
and designate the extended function by the same notation f . Note that for
x ∈ (a, b), and for all n ∈ N such that x + tn ∈ [x, b],

x+tn
g(x + tn ) − g(x)
Z
1
fn (x) := = f dm.
tn tn x

Then we have |fn | ≤ M and fn → g 0 almost everywhere. Hence, by DCT,


for every y ∈ [a, b],
Z y Z y
fn dm → g 0 dm.
a a

Note that, for y ∈ [a, b], using the translation invariant property of the
138 Integral of Complex Measurable Functions

Lebesgue measure,
Z y
1 y
Z
fn dm = [g(x + tn ) − g(x)] dm
a tn a
Z y+hn Z y 
1
= g(x) dm − g(x) dm
tn a+tn a
Z y+tn Z a+tn Z y 
1
= g(x) dm − g(x)dm − g(x) dm
tn y y a
Z y+tn Z a+tn 
1
= g(x) dm − g(x) dm .
tn y a

Since g is continuous, we know from real analysis that

1 y+tn 1 a+tn
Z Z
g(x) dm → g(y), g(x) dm → g(a) = 0.
tn y tn a

Hence,
Z y Z y Z y
0
g dm = lim fn dm = g(y) = f dm ∀ y ∈ [a, b].
a n→∞ a a

Thus, Z y
(g 0 − f )dm = 0 ∀ y ∈ [a, b].
a
Therefore, by Theorem 5.3.16, g 0 = f a.e.
Case (2): f not necessarily bounded. Without loss of generality,
assume that f is non-negative. For n ∈ N, let fn be as in Theorem 5.3.9,
that is, 
f (x), f (x) ≤ n,
fn (x) =
n, f (x) > n.
Then fn → f pointwise and f − fn ≥ 0 for all n ∈ N. Let
Z x Z x
gn (x) := fn dm and hn (x) := (f − fn )dm, x ∈ [a, b],
a a

for each n ∈ N. Since, for each n ∈ N, gn and hn are monotonically increasing


functions on [a, b], by Lemma 5.5.6, they are differentiable almost everywhere
and their derivatives gn0 and h0n are measurable almost everywhere. Also, by
Case (1) above, gn0 = fn a.e. for each n ∈ N. Note that

g = gn + hn
Indefinite Integral and its Derivative 139

so that g is differentiable almoste everywhere, and


g 0 = gn0 + h0n = fn + h0n ≥ fn a.e.
Since fn → f pointwise, we have g 0 ≥ f a.e. Therefore, again by Lemma ??,
Z b Z b
0
g(b) = g(b) − g(a) ≥ g dm ≥ f dm = g(b)
a a
Rb 0
Hence, a (g − f )dm = 0. Since g 0 − f ≥ 0 a.e., we have g 0 = f a.e.

We have proved that if f ∈ L[a, b], then its indefinite integral g is


absolutely continuous, differentiable almost everywhere, and its derivative
g 0 is equal to f almost everywhere. Recall again from calculus that if
f : [a, b] → R is a Riemann integrable function and if f = g 0 for some
diffierentiable function g : [a, b] → R, then
Z x
g(x) = g(a) + f (t)dt ∀ x ∈ [a, b].
a

In this connection, we may ask the following question:

If a function g : [a, b] → K is differentiable almost everywhere


with g 0 ∈ L[a, b], then do we have the relation
Z x
g(x) = g(a) + g 0 dm ∀ x ∈ [a, b]?
a

Well, it is too much to expect, in view of the fact that an indefinite integral
of an integrable function is continuous, in fact, absolutely continuous. For
instance, consider the function g := χ[a,c] : [a, b] → R for c = (a + b)/2. Then
g is differentiable almost everywhere with g 0 = 0. Note that
Z x
g(a) + g 0 dm = 1 ∀ x ∈ [a, b] whereas g(x) = 0 ∀ x ∈ [c, b].
a
However, if we assume that the function g is also absolutely continuous, then
we do have the affirmative answer, which is also known as a fundamental
theorem of Lebesgue integration.
Theorem 5.5.7 (Fundamental theorem of Lebesgue integration)
Suppose g : [a, b] → K is absolutely continuous and g is differentibale almost
everywhere with g 0 ∈ L[a, b]. Then
Z x
g(x) = g(a) + g 0 dm.
a
140 Integral of Complex Measurable Functions

For its proof, we make use of the following property of absolutely con-
tinuous functions.
Lemma 5.5.8 Suppose ϕ : [a, b] → K be absolutely continuous. If ϕ0 = 0
a.e., then ϕ is a constant function.

Proof of Theorem 5.5.7. Let


Z x
h(x) = g 0 dm, x ∈ [a, b].
a

By Theorem 5.5.4, h is absolutely continuous and, by Theorem 5.5.5, h is


differentiable almost everywhere and h0 = g 0 a.e. Thus, the function g − h
is absolutely continuous and (g − h)0 = 0 a.e. Therefore, by Lemma 5.5.8,
g − h is a constant function. In particular, g(x) − h(x) = g(a) − h(a) so that
Z x
g(x) − g(a) = h(x) − h(a) = g 0 dm.
a

This completes the proof.

Now, we show that the assumption in Theorem 5.5.7 that g is differen-


tibale almost everywhere with g 0 ∈ L[a, b] is redundant. Thus, in fact, we
have the following theorem in place of Theorem 5.5.7, which is again known
as a fundamental theorem of Lebesgue integration.
Theorem 5.5.9 (Fundamental theorem of Lebesgue integration)
Suppose g : [a, b] → K is absolutely continuous. Then g is differentiable
almost everywhere, g 0 ∈ L[a, b] and
Z x
g(x) = g(a) + g 0 dm.
a

For its proof we need some preparatory results. First a definition.


Definition 5.5.10 A function ϕ : [a, b] → K is said to be of bounded
variation if there exists M > 0 such that
n
X
|ϕ(xi ) − ϕ(xi−1 )| ≤ M
i=1

for every partition P : a = x0 < x1 < · · · < xn = b of [a, b]. The quantity
n
X
V[a,b] (ϕ) := sup |ϕ(xi ) − ϕ(xi−1 )|
P i=1
Indefinite Integral and its Derivative 141

is called the total variation of ϕ, where supremum is taken over all parti-
tions P of [a, b]. ♦
Remark 5.5.11 Suppose γ : [a, b] → C is a continuous function. If γ is of
bounded variation, then one may define the length of the curve γ as its
total variation. ♦
It can be easily seen that

• Every monotonically increasing function ϕ : [a, b] → R is of bounded


variation and its total variation is ϕ(b) − ϕ(a) (Problem 24).

• Every characteristic function on [a, b] is of bounded variation.

As an example of a function which is not of bounded variation, consider


the function (Problem 25) ϕ : [0, 1] → R defined by

sin(1/x), x > 0,
ϕ(x) =
1, x = 0.

It is also important to observe that a continuous function need not be


of bounded variation. For example, consider the function ϕ : [0, 1] → R
defined by

x sin(1/x), x > 0,
ϕ(x) =
0, x = 0.

Then it can be shown that (Problem 26) f ∈ C[0, 1] but not of bounded
variation. However,

• if ϕ : [a, b] → K is Lipschitz continuous, that is, if there exists K > 0


such that |ϕ(x) = ϕ(y)| ≤ K|x − y| for every x, y ∈ [a, b], then ϕ is of
bounded variation.

Theorem 5.5.12 Let f ∈ L[a, b] and g : [a, b] → K be defined by


Z x
g(x) = f dm, x ∈ [a, b].
a

Then g is of bounded variation.


142 Integral of Complex Measurable Functions

Proof. Let a = x0 < x1 < . . . < xn = b be a partition of [a, b]. Then


n
X n
X Z xi
|g(xi ) − g(xi−1 )| = | f dm|
i=1 i=1 xi−1
n
X xiZ
≤ |f |dm
i=1 xi−1
Z b
= |f |dm.
a

Since f ∈ L[a, b], it follows that g is of bounded variation.

Theorem 5.5.13 Every absolutely continuous function on [a, b] is of bounded


variation.

Proof. Let ϕ : [a, b] → K be an absolutely continuous function. Hence,


for a given ε, there exists δ > 0 such that for every family {Ii : i = 1, . . . , n}
of non-overlapping subintervals of [a, b],
n
X n
X
`(Ii ) < δ ⇒ |ϕ(xi ) − ϕ(yi )| < ε,
i=1 i=1

where xi and yi are the endpoints of Ii , i = 1, . . . , n.


Consider a partition P : a = t0 < t1 < · · · < tm = b of [a, b]. Let
{ai : i = 0, 1, . . . , k} ⊆ [a, b] be such that

a = a0 < a1 < · · · < ak = b with ai − ai−1 < δ for i = 1, . . . , k.

Now, consider the combination of the partitions {ti }m k


i=0 and {ai }i=0 , say
`
{si }i=0 , where any two repeated points are taken only once. Then we have

m
X `
X
|ϕ(ti ) − ϕ(ti−1 | ≤ |ϕ(si ) − ϕ(si−1 )|.
i=1 i=1

For each i ∈ {1, . . . , k}, let si,1 , . . . , si,ni be the points from {si }`i=0 that lie
in [ai−1 , ai ]. Then we have
ni
X
|ϕ(si,j ) − ϕ(si,j−1 )| < ε
j=1
Indefinite Integral and its Derivative 143

for each i ∈ {1, . . . , k}, so that


`
X X ni
k X k
X
|ϕ(si ) − ϕ(si−1 )| = |ϕ(si,j ) − ϕ(si,j−1 )| < ε = kε
i=1 i=1 j=1 i=1

and hence m
P
i=1 |ϕ(ti ) − ϕ(ti−1 | ≤ kε. Note that the number kε is inde-
pendent of the partition P . Thus, we have proved that ϕ is of bounded
variation.

We shall make use of the following characterization of functions of bounded


variation, the proof of which is left as an exeercise (see Problem 30).
Proposition 5.5.14 Let ϕ : [a, b] → R be a function of bounded variation,
and let
ϕ1 (x) := V[a,x] (ϕ) and ϕ2 (x) = ϕ1 (x) − ϕ(x)
for x ∈ [a, b]. Then ϕ1 and ϕ2 are monotonically increasing functions.

We have already mentioned that every monotonically increasing function


[a, b] is of bounded variation. This, combined with Proposition 5.5.14 lead
to the following:

A ϕ : [a, b] → R is of bounded variation if and only if there


exists monotonically increasing functions ϕ1 , ϕ2 on [a, b] such
that ϕ = ϕ1 − ϕ2 .
Theorem 5.5.15 Let ϕ : [a, b] → K be of bounded variation. Then ϕ is
differentiable almost everywhere and ϕ0 ∈ L[a, b].

Proof. Assume without loss of generality that ϕ is real valued. Then


by Proposition 5.5.14, ϕ = ϕ1 − ϕ2 where ϕ1 and ϕ2 are monotonically
increasing functions which are differentiable almost everywhere and ϕ01 ≥ 0
and ϕ02 ≥ 0 almost everywhere. Thus, ϕ0 = ϕ01 − ϕ02 a.e. and

|ϕ0 | ≤ ϕ01 + ϕ02 a.e.

Therefore, by Lemma 5.5.6,


Z b Z b Z b
0 0
|ϕ | dm ≤ ϕ1 dm + ϕ02 dm ≤ ϕ1 (b) − ϕ1 (a) + ϕ2 (b) − ϕ2 (a).
a a a

Consequently, ϕ0 ∈ L[a, b].


144 Integral of Complex Measurable Functions

Proof of Theorem 5.5.9. By Theorem 5.5.13, g is of bounded varia-


tion, and by Theorem 5.5.15, g is differentiable a.e. and g ∈ L[a, b]. Hence,
by Theorem 5.5.7, the relation
Z b
g(x) − g(a) = g 0 dm.
a

also holds.

Combining Theorem 5.5.4 and Theorem 5.5.9, we have the following


two theorems which are also known as Fundamental theorem of Lebesgue
integration.

Theorem 5.5.16 (Fundamental theorem of Lebesgue integration)


A function g : [a, b] → K is an indefinite integral of an integrable function
f : [a, b] → K if and only if g is absolutely continuous, and in that case
g 0 = f a.e., and
Z x
g(x) − g(a) = f dm ∀ x ∈ [a, b].
a

Equivalently,

Theorem 5.5.17 (Fundamental theorem of Lebesgue integration)


A function g : [a, b] → K is absolutely continuous if and only if there exists
an integrable function f : [a, b] → K such that
Z x
g(x) − g(a) = f dm ∀ x ∈ [a, b],
a

and in that case g 0 = f a.e.

5.6 Problems
R R R
1. If f ∈ L(µ) such that X
f ≥ 0, then show that X
f= X
Ref.
2. Suppose f is a real measurable function. Prove the following.
(a) If f ≥ 0 a.e., then X f dµ is well defined and X f dµ = X f + dµ.
R R R

(b) If f ≤ 0 a.e., then X f dµ is well defined and X f dµ = − X f + dµ.


R R R

3. Prove that if f is a bounded complex measurable function and µ(X) < ∞,


then f ∈ L(µ).
4. Prove Corollary 5.1.7.
Problems 145

R
5. Show that L(µ) is a vector space over C, and f 7→ X f is a linear functional
on L(µ).
R
6. Show that the set N := {f ∈ L(µ)R : X |f | = 0} is subspace of the vector
space L(µ), and the map [f ] 7→ X |f | is a norm on the quotient space
L(µ)/N .
7. For f ∈ L(R, m), showR that the integral R f (x)e−itx dx exists for each t ∈ R
R

and the function t 7→ R f (x)e−itx dx is continuous and bounded on R.


R R
8. If f ∈ L(µ) such that X f = X |f |, then show that there exists c ∈ C such
that f (x) = c|f (x)| for almost all x ∈ X.
R
9. Prove that [f ] 7→ X |f |dµ is a norm on the quotient space L(µ) := L(µ)/Z,
where Z = {f ∈ L(µ) : f = 0 a.e}.
10. Justify this satement: Suppose (fn ) is a sequence of Riemann integrable
functions on [a, b] such that fn → f pointwise on [a, b] and if (fn ) is uniformly
Rb R
bounded, then f is Lebesgue integrable, and lim a fn (x) dx = X f.
n→∞

11. Prove that, if µ(X) < ∞ and 1 ≤ p < r ≤ ∞, then Lr (µ) ⊆ Lp (µ).
12. Prove that for 1 ≤ p < r ≤ ∞, the inclusion `p (N) ⊆ `r (N) is proper.
13. Let
 1
1

 , x ∈ [1, ∞),
  √ ,
 x ∈ (0, 1),
x x
f (x) = and g(x) =
 
0, x 6∈ [1, ∞)
 
0, x 6∈ (0, 1)

Show that f ∈ L2 (R) \ L1 (R) and g ∈ L1 (R) \ L2 (R).


14. Let X be a denumerable set with counting measure µ on X. Show that for
1 ≤ p < r ≤ ∞, Lp (µ) ⊆ Lr (µ).
15. Prove that for 1 ≤ p < ∞, the metric d(·, ·) defined by
Z b
dp (f, g) := |f (x) − g(x)|p dx, f, g ∈ C[a, b],
a

on C[a, b] is not complete.


16. Prove that for 1 ≤ p < ∞, the space P[a, b] of all polynomial functions on
[a, b] is dense in C[a, b] with respect to the metric (f, g) 7→ kf − gkp .
17. For p, q ∈ [1, ∞], let (fn ) and (gn ) be sequences in Lp (µ) and g ∈ Lq (µ),
respectively such that kfn − f kp → 0 and kgn − gkq → 0 for some f ∈ Lp (µ)
and g ∈ Lq (µ). Prove that kfn gn − f gk1 → 0.
18. Suppose p, q, r ∈ (1, ∞] are such that 1r = p1 + 1q . If f ∈ Lp (µ) and g ∈ Lq (µ),
then prove that f ∈ Lr (µ) and kf kr ≤ kf kp kf kq .
146 Integral of Complex Measurable Functions

19. Let p, q, r ∈ (1, ∞] be such that p ≥ r, q ≥ r and 1r = pθ + 1−θ q . If f ∈


p q r θ 1−θ
L (µ) ∩ L (µ), then prove that f ∈ L (µ) and kf kr ≤ kf kp kf kq .

20. Prove that for 1 ≤ p < ∞ and k ∈ N, the space C k [a, b] is dense in Lp [a, b].
21. Suppose Ω is a subset of Ω such that the topology induced by the usual
metric on R is the discrete topology. Prove the following:

(a) A subset of Ω is compact if and only if it is a finite set.


(b) Cc (Ω) is a dense subset of Lp (Ω).

22. Let f : [a, b] → R be continuous and differentiable on (a, b) with |f 0 (x)| ≤ M


for all x ∈ (a, b) for some M > 0. Show that f is absolutely continuous.
23. If f : [a, b] → R is an absolutely continuous function such that f (x) 6= 0 for
all x ∈ [a, b], then show that 1/f is also absolutely continuous.
24. Every monotonically increasing function F : [a, b] → K is of bounded varia-
tion and its total variation is F (b) − F (a).

sin(1/x), x > 0,
25. Show that the function f : [0, 1] → R defined by f (x) =
1, x=0
is not of bounded variation.

x sin(1/x), x > 0,
26. Show that the function f : [0, 1] → R defined by f (x) =
0, x=0
is continuous, but not of bounded variation.
27. Suppose f : [a, b] → K is of of bounded variation and |f | ≥ c > 0 on [a, b].
Show that 1/f is of bounded variation on [a, b].
28. Suppose f and g are functions of bounded variation on [a, b]. Show that f g
is also of bounded variation on [a, b].
29. Suppose f : [a, b] → R is continuous and and differentiable on (a, b) with
|f 0 (x)| ≤ M for all x ∈ (a, b). for some M > 0. Show that f is also of
bounded variation on [a, b].
30. Prove Proposition 5.5.14.

5.7 Appendix
Definition 5.7.1 Let E ⊆ R. A family I of intervals is called a Vitali
cover of E, if for every x ∈ E and for every ε > 0, there exists I ∈ I such
that
x ∈ I and `(I) < ε.

Appendix 147

Lemma 5.7.2 (Vitali covering lemma) Let E ⊆ R be such that m∗ (E) <
∞. If I is a Vitali cover of E, then for every ε > 0, there exist pairwise
disjoint intervals I1 , . . . , In in I such that

m∗ (E \ ∪ni=1 Ii ) < ε.

Proof. Let ε > 0 be given and let G be an open set such that E ⊆ G and
m∗ (G) < m∗ (E) + ε. Without loss of generality assume that intervals in I
are closed. Since I is a Vitali cover of E, we can also assume that I ⊆ G
for every I ∈ I.
For some n ∈ N, consider pairwise disjoint intervals I1 , . . . , In in I. If
m∗ (E \∪ni=1 Ii ) < ε, then we are done. Otherwise, choose In+1 ∈ I as follows:
Let
κn := sup{`(I) : I ∈ I, I ∩ Ij 6= ∅, j = 1, . . . , n}.
Choose In+1 ∈ I such that `(In ) ≥ `(In+1 ) > κn /2. Sine {In } is a disjoint
family of intervals and since their union is contained in G, we have

X
`(In ) = µ∗ (∪∞ ∗
n=1 ) ≤ m (G) < ∞.
n=1

Hence, there exists N ∈ N such that



X
`(In ) < ε/5.
n=N +1

We show that m∗ (E \ ∪N N N
i=1 Ii ) < ε. Let x ∈ E \ ∪i=1 Ii . Since E \ ∪i=1 Ii ⊆
N N
G \ ∪i=1 Ii , there exists Ix ∈ I such that Ix ⊆ G \ ∪i=1 Ii . In particular,
Ix ∩Ij = ∅, j = 1, . . . , N . Also, there exists some n > N such that Ix ∩In 6= ∅.
For, otherwise, κn > `(Ix ) for all n > N which is not possible, since κn → 0
as n → ∞. Note that

Ix ∩ In 6= ∅ ⇒ `(Ix ) ≤ κn ≤ 2`(In+1 ) ≤ 2`(In ).

Therefore,
Ix ⊆ In + 2`(In )[−1, 1] = Jn , say.
Thus,

E \ ∪N
i=1 Ii ⊆ ∪n=N +1 Jn ,
P∞
where `(Jn ) ≤ 5`(In ), so that m∗ (E \ ∪N
i=1 Ii ) ≤ 5 n=N +1 `(In ) < ε.
148 Integral of Complex Measurable Functions

Theorem 5.7.3 If f : [a, b] → R is monotonically increasing, then f is


differentiable a.e.

Proof. For each x ∈ [a, b], define the quantities


f (x + h) − f (x)
D+ f (x) = lim sup ,
h→0+ h
f (x) − f (x − h)
D− f (x) = lim sup ,
h→0+ h
f (x + h) − f (x)
D+ f (x) = lim inf ,
h→0+ h
f (x) − f (x − h)
D− f (x) = lim inf ,
h→0 + h
Let
A = {x ∈ [a, b] : D+ f (x) > D− f (x)},
B = {x ∈ [a, b] : D− f (x) > D+ f (x)}.
Note that x 6∈ A ∪ B implies

D+ f (x) ≤ D− f (x) and D− f (x) ≤ D+ f (x)

so that
D− f (x) ≤ D− f (x) ≤ D+ f (x) ≤ D+ f (x) ≤ D− f (x)
which implies that all the four quantities D− f (x), D− f (x), D+ f (x), D+ f (x)
are the same, and hence f is differentiable at x. In other words, f is differ-
entiable on [a, b] \ A ∪ B. Hence, it is enough to show that A and b are sets
of measure zero.
Observe that,

x ∈ A ⇐⇒ ∃ s, t ∈ Q such that D+ f (x) > s > t > D− f (x).

Hence,
[
A= As,t where As,t := {x ∈ [a, b] : D+ f (x) > s > t > D− f (x)}.
s,t∈Q

Thus, it is enough to prove that As,t is of measure 0 for each s, t ∈ Q.


Now, let ε > 0 be given. Then, by the definition of m∗ (·), there exists
an open set V ⊆ R such that

As,t ⊆ V and m∗ (V ) ≤ m∗ (As,t ) + ε.


Appendix 149

Also, by the definition of D− f , for each x ∈ As,t , there exists hx > 0 such
that
[x − h, x] ⊆ V and f (x) − f (x − h) < th ∀ h ∈ (0, hx ].
Let
IV := {[x − h, x] : h ∈ (0, hx ], x ∈ As,t }.
We observe that I is a Vitali cover of As,t . Hence, there exist disjoint
intervals I1 , . . . , In in IV such that
m∗ (As,t \ ∪nj=1 Ij ) < ε.
Writing Ij := [xj − hj , xj ], j = 1, . . . , n, we have
n
X n
X
[f (xj ) − f (xj − hj )] ≤ t hj ≤ tm∗ (V ) < t[m∗ (As,t ) + ε].
j=1 j=1

Now, let
G = As,t ∩ (∪nj=1 Ij◦ ).
Then for every y ∈ G, there exists ky > 0 and jy ∈ {1, . . . , n} such that
[y, y + k] ⊆ Ijy and f (y + k) − f (y) > sk ∀k ∈ (0, ky ].
We observe that
IG := {[y, y + k] : k ∈ (0, ky ], y ∈ G}
is a Vitali cover of As,t . Hence, there exist disjoint intervals J1 , . . . , Jm in
IG such that
m∗ (G \ ∪mi=1 Ji ) < ε.
Let Ji = [yi , yi + ki ], i = 1, . . . , m. Then we have
m
X m
X
[f (yi + ki ) − f (yi )] > s ki .
i=1 i=1

Since each Ji is contained in some Ij , we have


m
X n
X X
[f (yi + ki ) − f (yi )] = [f (yi + ki ) − f (yi )]
i=1 j=1 {i:Ji ⊆Ij }
n
X
= [f (xj ) − f (xj − hj )]
j=1
< t[m∗ (As,t ) + ε].
150 Integral of Complex Measurable Functions

Thus,
m
X m
X
s ki < [f (yi + ki ) − f (yi )] < t[m∗ (As,t ) + ε]. (∗)
i=1 i=1
Since m∗ (G \ m
∪i=1 Ji ) < ε, we have

m∗ (G) ≤ m∗ (∪m ∗ m ∗ m
i=1 Ji ) + m (G \ ∪i=1 Ji ) < m (∪i=1 Ji ) + ε.

Hence,
m
X
ki = m∗ (∪m ∗
i=1 Ji ) > m (G) − ε.
i=1
Also, we have

m∗ (As,t ) ≤ m∗ (G) + m∗ (As,t \ G) = m∗ (G) + m∗ (As,t \ ∪nj=1 Ij ) < m∗ (G) + ε.

Thus,
m
X
ki = m∗ (∪m ∗ ∗
i=1 Ji ) > m (G) − ε > m (As,t ) − 2ε.
i=1
Applying this on (∗), we obtain
m
X

s[m (As,t ) − 2ε] < s ki < t[m∗ (As,t ) + ε].
i=1

Hence, s[m∗ (As,t )−2ε] < t[m∗ (As,t )+ε] for every ε > 0 so that s m∗ (As,t ) ≤
t m∗ (As,t ). Since t < s, it follows that m∗ (As,t ) = 0.
We have shown that m∗ (A) = 0. Following analogues arguments, we
obtain m∗ (B) = 0. This completes the proof.

For both the theorems above, we did not use any measure theoretic argu-
ments except the fact that if {In } is a countable disjoint family of intervals,
then
X∞
∗ ∞
m (∪n=1 In ) = `(In ).
n=1
In the next theorem we require Fatou’s lemma, though the statement of the
theorem is quite measure-theoretic-free.
Theorem 5.7.4 Let ϕ : [a, b] → R be a monotonically increasing function.
Then ϕ0 exists a.e., f 0 is measurable and
Z b
ϕ0 dm ≤ ϕ(b) − ϕ(a).
a
Appendix 151

Proof. By Theorem 5.7.3, ϕ is differentiable a.e. Let E ⊆ [a, b] be such


that ϕ0 (x) exists for all x ∈ E and m∗ ([a, b] \ E) = 0. Let
 0
ϕ (x), x ∈ E,
f (x) =
0, x ∈ [a, b] \ E.

Define ϕ(x) = ϕ(b) for x ∈ [b, b + 1] and for n ∈ N, let

fn (x) = n[ϕ(x + 1/n) − ϕ(x)], x ∈ [a, b].

Note that each fn is non-negative, and fn → g pointwise. Hence, f is


measurable; in particular, f 0 is measurable. Also we have
Z b Z b Z b
fn (x)dx = n ϕ(x + 1/n)dx − n ϕ(x)dx
a a a
Z b+1/n Z b
= n ϕ(y)dy − n ϕ(y)dy
a+1/n a
Z b+1/n Z a+1/n
= n ϕ(y)dy − n ϕ(y)dy
b a

But Z b+1/n Z b+1/n


n ϕ(y)dy = n ϕ(b)dy = ϕ(b),
b b
Z a+1/n Z a+1/n
n ϕ(y)dy ≥ n ϕ(a)dy = ϕ(a).
a a
Hence,
Z b
fn (x)dx ≤ ϕ(b) − ϕ(a).
a
Therefore, by Fatous’ lemma,
Z b Z b
f (x)dx ≤ lim inf fn (x)dx ≤ ϕ(b) − ϕ(a).
a n a

Thus, the proof is complete.


6

Integration on Product Spaces

So far we have been concerned about measurable functions of


one variable and their integration. In this chapter we consider
measurable functions of more than one variable, that is, functions
on measurable spaces of the form X := X1 × X2 × · · · × Xk with
appropriate σ-algebras and measures on them. We shall restrict
our study for the case of k = 2 and when we are already given
certain σ-algebras and measures on the component spaces X1
and X2 . Thus, the ideas is to construct a new σ-algebra and
a measure on cartesian product X1 × X2 using the measures
spaces X1 and X2 , and see how integration of the product space
is related to the integration on the component spaces. For this
purpose we shall use most of the concepts and basic theorems
introduced in the previous chapters.

6.1 Product Measure


Let (X1 , A1 , µ1 ) and (X2 , A2 , µ2 ) be measure spaces. We would like to have
a σ-algebra A1 ⊗ A2 on X1 × X2 and a measure µ1 × µ2 on A1 ⊗ A2 with
the following properties:

• A1 ⊗ A2 ⊇ {A × B : A ∈ A1 , B ∈ A2 }.

• (µ1 × µ2 )(A × B) = µ1 (A)µ2 (B) ∀ A ∈ A1 , A2 ∈ A2 .

• For every non-negative measurable function f on X1 ×X2 , the integrals


Z Z  Z Z 
f (x, y)dµ2 dµ1 , f (x, y)dµ1 dµ2
X1 X2 X2 X1

152
Product Measure 153

are well defined and


Z Z Z 
f d(µ1 × µ2 ) = f (x, y)dµ2 dµ1
X1 ×X2 X X
Z 1 Z 2 
= f (x, y)dµ1 dµ2 .
X2 X1

We would also like to have the above results for a complex valued measurable
function f , whenever f is integrable with respect to the product measure.
We shall show the existence of a measure µ1 × µ2 with the required
properties will be guaranteed whenever µ1 and µ2 are σ-finite measures. In
fact, it is known that such a product measure is unique for σ-finite measures
µ1 and µ2 .
To proceed further we consider a few definitions:
Definition 6.1.1 Sets of the form A × B with A ∈ A1 and B ∈ A2 are
called measurable rectangles, and the σ-algebra generated by the family
of all measurable rectangles is called the product σ-algebra. We denote
the product σ-algebra by A1 ⊗ A2 . ♦
Definition 6.1.2 For any set E ⊆ X1 × X2 and (x, y) ∈ X1 × X2 , let

Ex := {v ∈ X2 : (x, v) ∈ E}, E y := {u ∈ X1 : (u, y) ∈ E}.

The sets Ex and E y are called x-section and y-section, respectively, of


the set E. ♦
It can be easily seen that if

E⊆F ⇒ Ex ⊆ Fx and E y ⊆ F y .

In the following lemma, we state some easily verifiable facts:

Lemma 6.1.3 Let E be a measurable recatangle, say E = A × B. Then,


for each x ∈ X1 , y ∈ X2 , the following are true.
 
B, x ∈ A, A, y ∈ B,
(i) Ex = Ey =
∅, x ∈
6 A, ∅, y ∈
6 B.

(ii) Ex ∈ A2 and E y ∈ A1 for all (x, y) ∈ X1 × X2 ,

(iii) µ2 (Ex ) = µ2 (B)χA (x) and µ1 (E y ) = µ2 (B)χB (y).


154 Integration on Product Spaces

Further, the functions x 7→ µ2 (Ex ) and y 7→ µ1 (E y ) are measurable with


respect to A1 and A2 , respectively, and
Z Z
µ2 (Ex )dµ1 = µ1 (A)µ2 (B) = µ1 (E y )dµ2 .
X1 X2

The results in the above lemma prompt us to ask whether for every
E ∈ A1 ⊗ A2 the following are true or not:

1. Ex ∈ A2 , E y ∈ A1 for every (x, y) ∈ X1 × X2 ,

2. x 7→ µ2 (Ex ) and y 7→ µ1 (E y ) are measurable, and


Z Z
µ2 (Ex )dµ1 = µ1 (E y )dµ2 .
X1 X2

Z
3. E 7→ µ2 (Ex )dµ1 is a measure on A1 ⊗ A2 .
X1

Our first attempt in this chapter is to prove that the above results are
true provided µ1 and µ2 are σ-finite. For this, we shall make use of another
easily verifiable proposition.

Proposition 6.1.4 For every E ⊆ X1 × X2 ,

(E c )x = Exc , (E c )y = (E y )c

and for En ⊆ X1 × X2 , n ∈ N,
[  [ [ y [
En = (En )x , En = (En )y .
x

Further, if {En :∈ N} is a disjoint family of subsets of X1 × X2 , then


{(En )x :∈ N} and {Eny :∈ N} are disjoint families.

Notation: If En ∈ X1 × X2 for some n ∈ N, then we shall use the notation


En,x for (En )x , the x-section of En , and Eny for (En )y , the y-section of En .

Theorem 6.1.5 Let E ∈ A1 ⊗ A2 . Then for every (x, y) ∈ X1 × X2 ,


Ex ∈ A2 and E y ∈ A1 .
Product Measure 155

Proof. Let S be the family of all E ∈ A1 ⊗ A2 such that Ex ∈ A2 for all


x ∈ X1 . We have to show that S = A1 ⊗ A2 . For this, it is enough to show
that S is a σ-algebra containing all measurable rectangles. We have already
observed that if E is a measurable rectangle, then Ex ∈ A1 and E y ∈ A2 for
every (x, y) ∈ X1 × X2 so that S contains all measurable rectangles. The
fact that S is a σ-algebra follows from Proposition 6.1.4.
Similarly we see that the family of all E ∈ A1 ⊗ A2 such that E y ∈ A2
for all y ∈ X1 is A1 ⊗ A2 .

Theorem 6.1.6 Suppose (X1 , A1 , µ1 ) and (X2 , A2 , µ2 ) are σ-finite measure


spaces. Then for every E ∈ A1 ⊗ A2 , the functions

x 7→ µ2 (Ex ), y 7→ µ1 (E y )

are measurable with respect to A1 and A2 , respectively, and


Z Z
µ2 (Ex )dµ1 = µ1 (E y )dµ2 .
X1 X2

For the proof of the above theorem we shall also make use of a lemma
(Lemma 6.1.9) whose statement requires two more definitions.
Definition 6.1.7 A subset of X1 × X2 is called an elementary set if it is
a disjoint union of a finite number of measurable rectangles. ♦

Definition 6.1.8 Let X be a set. A family S of subsets of a set X is called


a monotone class if it has the following two properties:

(1) If Ai ∈ S and Ai ⊆ Ai+1 for all i ∈ N, then ∪Ai ∈ S,

(2) If Ai ∈ S and Ai ⊇ Ai+1 for all i ∈ N, then ∩Ai ∈ S. ♦

We observe the following:

1. Given any family F of subsets of X, there exists a smallest monotone


class containing F, called the monotone class generated by F.

2. A1 ⊗ A2 is a monotone class containing all elementary sets.

Now, we state the required lemma; its proof is given at the end of this
subsection.
156 Integration on Product Spaces

Lemma 6.1.9 The σ-algebra A1 ⊗ A2 is the smallest monotone class con-


taining all elementary sets.

Proof of Theorem 6.1.6. Let S be the class of all E ∈ A1 ⊗ A2 such


that the conclusions in the theorem hold. We prove that S = A1 ⊗ A2 , that
the proof will be complete.
The following facts can be verified easily:

(a) S contains all measurable rectangles.

(b) S contains all elementary sets.

(c) S contains finite disjoint unions of its members.

We claim that S has also the following properties:


(i) If En ∈ S with En ⊆ En+1 for all n ∈ N, then ∪En ∈ S.
(ii) If {En } is a disjoint family in S, then ∪En ∈ S.
(iii) If En ∈ S such that En ⊇ En+1 for all n ∈ N and there exists
a measurable rectangle A × B satisfying E1 ⊆ A × B, µ1 (A) < ∞ and
µ2 (B) < ∞, then ∩En ∈ S.
Note that, if µ1 and µ2 are finite measurers, then (i) and (iii) imply
that S is a monotone class (containing all measurable rectangles), so that
by Lemma 6.1.9, S = A1 ⊗ A2 , and hence the proof is complete in this case.
Proof of (i): Let En ∈ S with En ⊆ En+1 for all n ∈ N. By the definition of
S, for each n ∈ N, the functions

x 7→ µ1 (En,x ), y 7→ µ2 (Eny )

are measurable functions and


Z Z
µ2 (En,x )dµ1 = µ1 (Eny )dµ2 .
X1 X2

Also we have
y
En,x ⊆ En+1,x , Eny ⊆ En+1 ∀n ∈ N
so that
µ2 (En,x ) → µ2 (∪En,x ) = µ2 ((∪En )x ),

µ1 (Eny ) → µ1 (∪Eny ) = µ1 ((∪En )y ).


Product Measure 157

Hence, the functions

x 7→ µ2 ((∪En )x ), y 7→ µ1 ((∪En )y )

are measurable, and by Monotone Convergence Theorem (Theorem 4.4.1),


we have Z Z
lim µ2 (En,x )dµ1 = µ2 ((∪En )x )dµ1 ,
n→∞ X X1
1
Z Z
lim µ1 (Eny )dµ2 = µ1 ((∪En )y )dµ2 .
n→∞ X X2
2

Therefore, ∪En ∈ S.
Proof of (ii): Let {En } be a disjoint family in S. Then ∪En = ∪Fn where
Fn = ∪ni=1 Ei , n ∈ N. Since each Fn is a finite disjoint union of members
of S, we have Fn ∈ S for every n ∈ N. Also, Fn ⊆ Fn+1 for every n ∈ N.
Hence, by (i), ∪En = ∪Fn ∈ S.
Proof of (iii): As in (i),

x 7→ µ1 (En,x ), y 7→ µ2 (Eny )

are measurable functions and


Z Z
µ2 (En,x )dµ1 = µ1 (Eny )dµ2 .
X1 X2

Since A × B ⊇ E1 , we have

E1,x ⊆ (A × B)x , E1y ⊆ (A × B)y .

Hence, by Lemma 6.1.3,

µ2 ((A × B)x ) = µ2 (B)χA (x) < ∞,

µ1 ((A × B)y ) = µ1 (A)χB (y) < ∞.


Also, the condition En ⊇ En+1 for all n ∈ N, consequently, the facts
y
En,x ⊇ En+1,x , Eny ⊇ En+1 ∀n ∈ N,

imply
µ2 (En,x ) → µ2 (∩En,x ) = µ2 ((∩En )x ),
µ1 (Eny ) → µ1 (∩Eny ) = µ1 ((∩En )y ).
158 Integration on Product Spaces

Hence, the functions

x 7→ µ2 ((∩En )x ), y 7→ µ1 ((∩En )y )

are measurable. Again since

µ2 (En,x ) ≤ µ2 ((A × B)x ) = µ2 (B)χA (x),

µ1 (Eny ) ≤ µ1 ((A × B)y ) = µ1 (A)χB (y)


with Z Z
µ2 (B)χA (x)dµ1 = µ2 (B)µ1 (A) = µ1 (A)χB (y)dµ2
X1 X2

for all n ∈ N, by Dominated convergence theorem (Theorem 5.3.2),


Z Z
lim µ2 (En,x )dµ1 = µ2 ((∩En )x )dµ1 ,
n→∞ X X1
1

Z Z
lim µ1 (Eny )dµ2 = µ1 ((∩En )y )dµ2 .
n→∞ X X2
2

Therefore, ∩En ∈ S.
Now, since µ1 and µ2 are σ-finite measures, there exist disjoint families
(n) (n)
{X1 : n ∈ N} and {X2 : n ∈ N} of measurable subsets of X1 and X2 ,
respectively, such that
∞ ∞
(n) (m)
[ [
X1 = X1 , X2 = X2
n=1 m=1

(n) (m)
with µ1 (X1 ) < ∞ and µ2 (X2 ) < ∞ for all n, m ∈ N. For E ∈ A1 ⊗ A2 ,
let
(n) (m)
En,m := E ∩ (X1 × X2 ), n, m ∈ N.
Clearly, E is a disjoint union of {En,m : n, m ∈ N}. Let

A := {E ∈ A1 ⊗ A2 : En,m ∈ S ∀ n, m ∈ N}.

By (i), (ii) (iii), it can be seen (verify) that A is a monotone class containing
all elementary sets. Hence, by Lemma 6.1.9, A = A1 ⊗ A2 . Thus, E ∈ A1 ⊗
A2 implies En,m ∈ S for all n, m ∈ N. Again by (ii) above, E = ∪En,m ∈ S.
That is, for every E ∈ A1 ⊗ A2 , the conclusions of the theorem hold. Thus,
we have proved that S = A1 ⊗ A2 , which completes the prof.
Product Measure 159

Now, the following theorem will lead to the definition of the product
measure.

Theorem 6.1.10 Suppose that (X1 , A1 , µ1 ) and (X2 , A2 , µ2 ) are σ-finite


measure spaces. For E ∈ A1 ⊗ A2 , let
Z Z
µ(E) := µ2 (Ex )dµ1 = µ1 (E y )dµ2 .
X1 X2

Then µ is a measure on A1 ⊗ A2 .

Proof. Clearly, µ(∅) = 0. Let {En : n ∈ N} be a disjoint family in


A1 ⊗ A2 . Then, by Proposition 6.1.4, we have
Z Z
µ(∪En ) = µ2 ((∪En )x )dµ1 = µ2 (∪En,x )dµ1 .
X1 X1

Now, using the fact that {En,x : n ∈ N} is a disjoint family in A2 and the
monotone convergence theorem, we have
Z ∞
X ∞ Z
X ∞
X
µ(∪En ) = µ2 (En,x )dµ1 = µ2 (En,x )dµ1 = µ(En ).
X1 n=1 n=1 X1 n=1

This completes the proof.

Definition 6.1.11 The measure µ in Theorem 6.1.10 is called the product


measure on A1 ⊗ A2 , and it is denoted by µ1 × µ2 . ♦
Now, we prove Lemma 6.1.9. For this, we require the concept of an
algebra which we have not introduced so far.
Definition 6.1.12 A family F of subsets of a set X is called an algebra if
it contains X and if it is closed under complements and finite unions, that
is,
(a) X ∈ F,
(b) A ∈ F ⇒ Ac ∈ F,
(c) A, B ∈ F ⇒ A ∪ B ∈ F. ♦
We may observe that every σ-algebra is an algebra which is also closed
under countable unions.

Proof of Lemma 6.1.9. The proof involves two main steps:


160 Integration on Product Spaces

Step (i): Let F be an algebra on a set X, MF be the monotone class


generated by F and AF be the σ-algebra generated by F. Then MF is a
σ-algebra and MF = AF :
Step (ii): The family E of all elementary sets in A1 ⊗ A2 is an algebra:
Since AE = A1 ⊗ A2 , results in Step 1 and Step 2 will imply the required
result ME = A1 ⊗ A2 .
Proof of Step 1: For showing that MF is a σ-algebra, it is enough to
show that it is an algebra, because, in that case, for any (An ) in MF ,

[ ∞  [
[ n 
An = Ak ∈ MF ,
n=1 n=1 k=1
as
n
[ n
[ n+1
[
Ak ∈ MF and Ak ⊆ Ak ∀ n ∈ N.
k=1 k=1 k=1

Let M
f := {A : Ac
∈ MF }. Then it can be seen that M
f is a monotone class
containing F. Hence, MF ⊆ M. Thus,
f

A ∈ MF ⇒ Ac ∈ MF .

Next, we show that MF is closed under finite unions, i.e., A, B ∈ MF


implies A ∪ B ∈ MF , equivalently, for every A ∈ MF , A
b = MF , where

b := {B ∈ MF : A ∪ B ∈ MF }.
A

So, let A ∈ MF . Note that A b is a monotone class. Therefore, for showing


that A = MF , it is enough to show that F ⊆ A.
b b So, let C ∈ F. Note that
F ⊆ C,b since if D ∈ F, then D ∪ C ∈ F and hence D ∪ C ∈ MF . Thus, C b
is a monotone class containing F. Since MF is the smallest monotone class
containing F, we obtain Cb = MF . Thus,

A ∈ MF ⇒ A∈C
b ⇒ C∈A
b

for any C ∈ F. Consequently,

A ∈ MF ⇒ F ⊆ A.
b

Thus, the proof is completed.


Proof of Step 2: It can be easily seen that E is closed under finite disjoint
unions and finite intersections. Also, for any measurable rectangle A1 × A2 ,

(A1 × A2 )c = (Ac1 × X2 ) ∪ (X1 × Ac2 )


Fubini’s Theorem 161

so that (A1 × A2 )c is a finite disjoint union of measurable rectangles. Hence,


(A1 × A2 )c ∈ E. Since each member A of E is a finite disjoint union of
measurable rectangles, say A = ∪ni=1 Ri , where {Ri : i = 1, . . . , n} is a
disjoint union of rectangles, we have
n
\
Ac = Ric ∈ E.
i=1

Now, if A, B ∈ E, then

A ∪ B = (A \ B) ∪ B = (A ∩ B c ) ∪ B,

which is a finite disjoint union of members of E. Thus, E is closed under


finite unions as well.

6.2 Fubini’s Theorem


Let (X1 , A1 , µ1 ) and (X2 , A2 , µ2 ) be measure spaces.
Definition 6.2.1 Let f be a function defined on X1 × X2 taking values in
another set Y . For each x ∈ X1 , the function fx : X2 → Y defined by

fx (y) = f (x, y), y ∈ X2 ,

is called the x-section of f , and for each y ∈ X2 , the function f y : X1 → Y


defined by
f y (x) = f (x, y), x ∈ X1 ,
is called the y-section of f . ♦

Proposition 6.2.2 Let f be a measurable function on X1 × X2 with respect


to the σ-algebra A1 ⊗ A2 . Then for each (x, y) ∈ X1 × X2 , fx and f y are
measurable with respect to A2 and A1 respectively.

Proof. Let G be an open set in the topological space in which f takes


values. Then we see that

{u ∈ X1 : f y (u) ∈ G} = {u ∈ X1 : f (u, y) ∈ G} = [f −1 (G)]y ,

{v ∈ X2 : fx (v) ∈ G} = {u ∈ X1 : f (x, v) ∈ G} = [f −1 (G)]x .


Since f is measurable, by Proposition 6.1.4, both [f −1 (G)]y and [f −1 (G)]x
are measurable sets. Hence the result.
162 Integration on Product Spaces

Theorem 6.2.3 (Tonelli’s theorem) Let (X1 , A1 , µ1 ) and (X2 , A2 , µ2 ) be


σ-finite measure spaces and let f be an extended real valued and non-negative
measurable function. Then the functions
Z Z
x 7→ f (x, y)dµ2 (y), y 7→ f (x, y)dµ1 (x)
X2 X1

are measurable with respect to A1 and A2 respectively, and


Z Z Z 
f d(µ1 × µ2 ) = f (x, y)dµ2 (y) dµ1 (x)
X1 ×X2 X1 X2
Z Z 
= f (x, y)dµ1 (x) dµ2 (y).
X2 X1

Proof. For x ∈ X1 and y ∈ X2 , let


Z Z
g(x) = fx dµ2 , h(y) = f y dµ1 .
X2 X1

Let us consider first the case f = χE for some E ∈ A1 ⊗ A2 . Then we have


Z Z
g(x) = χEx dµ2 = µ2 (Ex ), h(y) = χEy dµ1 = µ1 (E y ).
X2 X1

Hence, for f = χE , the result is a consequence of Theorem 6.1.6. Next,


let f be a non-negative simple measurable function. In this case, the result
follows by using the linearity of integrals. Now, let f be any non-negative
measurable function. Then, consider an increasing sequence (ϕn ) of simple
measurable functions which converges to f pointwise. If we take
Z Z
gn (x) := ϕn,x dµ2 , hn (y) = ϕyn dµ1 ,
X2 X1

then, by monotone convergence theorem, gn → g and hn → h pointwise.


Again, applying monotone convergence theorem,
Z Z Z Z
gn dµ1 → gdµ1 , hn dµ2 → hdµ2 .
X1 X1 X2 X2

Since
Z Z Z
gn dµ1 = hn dµ2 = fn d(µ1 × µ2 ) ∀ n ∈ N,
X1 X2 X1 ×X2

we have Z Z Z
gdµ1 = hdµ2 = f d(µ1 × µ2 ).
X1 X2 X1 ×X2
This completes the proof.
Fubini’s Theorem 163

Theorem 6.2.4 (Fubini’s theorem) Let (X1 , A1 , µ1 ) and (X2 , A2 , µ2 ) be


σ-finite measure spaces and let f be a complex measurable function on the
product measure space (X1 × X2 , A1 ⊗ A2 , µ1 × µ2 ). Suppose that atleast one
of the integrals
Z Z  Z Z  Z
|fx |dµ2 dµ1 , |f y |dµ1 dµ2 , |f |d(µ1 × µ2 )
X1 X2 X2 X1 X1 ×X2

is finite. Then they are equal and the following hold.

(i) fx ∈ L1 (µ2 ) for a.a x ∈ X1 , f y ∈ L1 (µ1 ) for a.a. y ∈ X2 , and


f ∈ L1 (µ1 × µ2 ),

(ii) x 7→ X2 f (x, y)dµ2 (y) and y 7→ X1 f (x, y)dµ1 (x) belong to L1 (µ1 )
R R

and L1 (µ2 ), respectively.

(iii) The integrals


Z Z  Z Z  Z
y
fx dµ2 dµ1 , f dµ1 dµ2 , f d(µ1 × µ2 )
X1 X2 X2 X1 X1 ×X2

are well defined and are equal.

Proof. Since |fx | = |f |x and |f y | = |f |y , by Theorem 6.2.3, the integrals


Z Z  Z Z  Z
y
|fx |dµ2 dµ1 , |f |dµ1 dµ2 , |f |d(µ1 × µ2 )
X1 X2 X2 X1 X1 ×X2

are equal. Hence, if one of these integrals is finite, all of these integrals are
finite. In particular, results in (i) hold.
To prove (ii) and (iii), first we assume that f ∈ L1 (µ1 ×µ2 ) is real valued.
Note that
Z Z
f + d(µ1 × µ2 ) ≤ |f |d(µ1 × µ2 ) < ∞.
X1 ×X2 X1 ×X2

Hence, the integrals


Z Z  Z Z  Z
(f + )x dµ2 dµ1 , + y
(f ) dµ1 dµ2 , f + d(µ1 × µ2 )
X1 X2 X2 X1 X1 ×X2

are equal and finte. In particular, the functions


Z Z
+
x 7→ f (x, y)dµ2 (y) and y 7→ f + (x, y)dµ1 (x)
X2 X1
164 Integration on Product Spaces

belong to L1 (µ1 ) and L1 (µ2 ) respectively. Hence, (ii) and (iii) hold with f +
in place of f . Similarly, we have the conclusions in (ii) and (iii) with f − in
place of f . Therefore, we have (ii) and (iii) for f as well.
The case for complex valued f follows by writing f as f = Re(f )+iIm(f )
and applying the results for the real valued functions Re(f ) and Im(f ), and
observing the facts that |Re(f )| ≤ |f |, |Im(f )| ≤ |f |, and the linearity of
taking intgrals.

6.3 Counter Examples

6.3.1 σ-finiteness condition cannot be dropped


Let X1 = [0, 1] with Lebesgue measure µ1 and X2 = [0, 1] with counting
measure µ2 . Let D := {(x, y) ∈ X1 × X2 : x = y}, the diagonal set. Since
D = ∩Dn , where
n    
[ j−1 j j−1 j
Dn := , × , , n ∈ N,
n n n n
j=1

it follows that D is a measurable subset of X1 × X2 . Note that for x, ∈ [0, 1],

Dx = {x}, Dy = {y}, µ2 (Dx ) = 1, µ1 (Dy ) = 0,

so that Z Z
µ2 (Dx )dµ1 = µ1 (X1 ) = 1, µ1 (Dy )dµ2 = 0.
X1 X2

Hence, the integrals involved in the definition of product measure are not
equal for the measurable set D.
Also, taking f = χD , the characteristic function of D, we have
Z Z  Z Z  Z
fx dµ2 dµ1 = χDx dµ2 dµ1 = µ2 (Dx )dµ1 = 1,
X1 X2 X1 X2 X1

Z Z  Z Z  Z
y
f dµ1 dµ2 = χDy dµ1 dµ2 = µ1 (Dy )dµ2 = 0.
X2 X1 X2 X1 X2

Thus, the iterated integrals in Fubini’s theorem are not equal for f = χD .
Note that µ2 is not σ-finite.
Problems 165

6.3.2 Product of complete measures need not be complete


Suppose (X1 , A1 , µ1 ) and (X2 , A2 , µ2 ) are complete σ-finite measure spaces
such that there exists A ∈ A1 with µ1 (A) = 0 and there exists B ⊆ X2 such
that B 6∈ A2 . Then we have A × B ⊆ A × X2 with
(µ1 × µ2 )(A × X2 ) = µ1 (A)µ2 (Y ) = 0.
But, A × B 6∈ A1 ⊗ A2 , since for every x ∈ A, (A × B)x = B 6∈ A2 . Thus,
µ1 × µ2 is not complete.
As an example, consider
(X1 , A1 , µ1 ) = (X2 , A2 , µ2 ) = (R, M, m).
We know that Q ∈ A1 with m(Q) = 0 and there exists B ⊆ R such that
B 6∈ M. Thus, (R, M, m) is complete, whereas (R × R, M ⊗ M, m × m) is not
complete. It can be shown that the completion of (R × R, M ⊗ M, m × m)
is the Lebesgue measure space (R2 , M2 , m2 ).

6.4 Problems
1. Prove Lemma 6.1.3.
2. Prove Proposition 6.1.4.
3. Let (X1 , A1 , µ1 ) and (X2 , A2 , µ2 ) be σ-finite measure spaces and let S be
the family of all E ∈ A1 ⊗ A2 such that the functions x 7→ µ2 (Ex ) and
yZ 7→ µ1 (E y ) are measurable
Z with respect to A1 and A2 , respectively, and
µ2 (Ex )dµ1 = µ1 (E y )dµ2 . Show that
X1 X2

(a) S contains all measurable rectangles.


(b) S contains all elementary sets.
(c) S contains finite disjoint unions of its members.
4. Prove that if A1 = BRm and A2 = BRn , then BRm ⊗ BRn = BRm+n .

(Hint: Observe: Every open set in Rm+n is a countable union of sets of the
form A × B where A and B are rectangles in Rm and Rn , respectively, and
prove: BRm+n contains sets of the form A × Rn and Rm × B where A ∈ BRm
and B ∈ BRn . )
5. Let (X1 , A1 , µ1 ) and (X2 , A2 , µ2 ) be σ-finite measure spaces, and f1 : X1 →
R and f2 : X2 → R are measurable functions. Let
f (x, y) = f1 (x)f2 (y), (x, y) ∈ X1 × X2 .
Prove the following:
166 Integration on Product Spaces

(a) f is measurable on the product space (X1 × X2 , A1 × A2 ).


(b) If f1 ∈ L1 (µ1 ) and f2 ∈ L1 (µ2 ), then f ∈ L1 (µ1 × µ2 ) and
Z Z  Z 
f d(µ1 × µ2 ) = f1 dµ1 f2 dµ2 .
X1 ×X2 X1 X2

xy
6. Let I = [−1, 1] and f (x, y) = for (x, y) ∈ I × I \ {(0, 0)} and
(x2 + y 2 )2
f (0, 0) = 0. Show that the integrals
Z Z  Z Z 
f (x, y)dm(x) dm(y), f (x, y)dm(x) dm(y)
I I I I
Z
exist and are equal, but f (x, y)d(m × m)(x, y) does not exist.
I×I

7. Let I = [0, 1], S = I × I \ {(0, 0)} and f : S → R be defined by f (x, y) =


x2 − y 2
for (x, y) ∈ I ×I \{(0, 0)} and f (0, 0) = 0. Show that the integrals
(x2 + y 2 )2
Z Z  Z Z 
f (x, y)dm(x) dm(y), f (x, y)dm(x) dm(y)
I I I I

exist and are unequal.


8. Let f be a non-negative measurable function on a σ-finite measure space
(X, A, µ) and S = {(x, y) ∈ X × R : 0 ≤ y ≤ f (x)}. Show that S ∈ A × M
and Z Z
f (x)dµ(x) = χS (x, y)d(µ × m).
X X×R
Z τ
sin x π
9. Use Fubini’s theorem to prove that lim dx = .
τ →∞ 0 x 2
Z ∞
1
[You may write = e−xy dy.]
x 0

10. Let (X, A, µ) be a complete measure space and f : X → [0, ∞) be an inte-


grable function.
 Let g(t) := µ({x ∈ X : f (x) ≥ t}), t ≥ 0. Using the function
1, 0 ≤ t ≤ f (x),
h(t, x) := show that
0, t > f (x),
Z Z ∞
f dµ = g(t)dt.
X 0
References

[1] K. Chandrasekharan, Classical Fourier Transforms, Springer–Verlag,


[2] C. G. Denlinger, Elements of Real Analysis, Jones and Bartlett Learn-
ing, 2011
[3] de Barra, Measure Theory and Integration, Wiley Eastern, New delhi,
1981.
[4] S.R. Ghorpade and B.V. Limaye, A Course in Calculus and real Anal-
ysis, Springer,
[5] W. A. J. Luxemburg, Arzela’s dominated convergence theorem for the
Riemann integral, The American Mathematical Monthly, Vol. 78, No.
9 (Nov., 1971), pp. 970-979.
https : //sites.math.washington.edu/ morrow/3351 5/dominated.pdf
[6] M.T. Nair, Functional Analysis: A First Course, PHI Learning, New
Delhi, 2002 (Fourth Print: 2014).
[7] M.T. Nair, Calculus of One Variable, Ane Books Pvt. Ltd., New Delhi,
2015.
[8] M.H. Protter, Basic Elements of Real Analysis, Springer, 1998.
[9] H.L. Royden, Real Analysis, 3rd Edition, Prentice-Hall of India, New
Delhi, 1995.
[10] W. Rudin, Principles of Mathematical Analysis (Third Edition), Inter-
national Student Edition, McGraw-Hill Kogakusha Ltd., Tokyo, 1964
[11] W. Rudin, Real and Complex Analysis (Third Edition), McGraw-Hill
Book Co., Singapore, 1987.

167
Index

Cc (Ω), 131 canonical representation, 67


Fδ set, 31 Canter set, 20
Gδ set, 31 Cantor set, 20
Lp - Spaces, 119 characteristic function, 10, 55
V[a,b] (ϕ), 141 closed set, 42
σ-algebra generated by, 42 complete measure, 46
σ-finite measure space, 38 , 46
σ-algebra, 36 completion, 48
x-section, 153, 161 complex measurable function,
y-section, 153, 161 53
Lp (Ω), 121 conditional expectation, 96
Lp (Ω, MΩ , m), 121 continuous, 53
L(X), 104 converges
L(µ), 104 almost everywhere, 60
L∞ (X, A, µ), 120 pointwise, 60
L∞ (µ), 120 countable family, 11
Lp (X), 119 Counter Examples, 164
Lp (X, A, µ), 119 counting measure, 38
Lp (Ω), 121
DCT, 112–114
Lp (Ω, MΩ , m), 121
Dirac measure, 39
Lp (µ), 119
Dirichlet’s function, 2, 7, 10
absolutely continuous, 94, 135 disjoint family, 22
distribution function, 57, 95
algebra, 36
cumulative, 57
almost everywhere, 61
Dominated convergence theo-
convergence, 60
rem, 112–114
converges, 60
Egorov’s theorem, 63
Borel σ-algebra, 42, 70 elementary set, 155
Borel field, 42 essentially bounded, 120
Borel sets, 42
bounded intervals, 12 Fatou’s lemma, 91, 150
bounded variation, 140 finite measure, 37, 38

168
Index 169

finite measure space, 37 measurable function, 53, 54


Fubini’s theorem, 161, 163 measurable sets, 36
Fundamental theorem, 144 measurable space, 36
measure, 36
Hölder’s inequality, 124 measure induced, 57
measure space, 36
improper inteegral, 130
measure zero, 8
indefinite integral, 134
mesh, 5
indicator function, 55
Minkowski inequality, 124
infimum, 59
modulus, 66
Integrability, 103, 104
monotone class, 155
integrable, 104
Monotone convergence theorem,
integral, 103, 105
88
integral of ϕ
monotonicity, 17, 37
over E, 75
over X, 74 negative part, 66
integral of f nowhere dense, 34
over E, 83
over X, 83 outer measure, 41
Integration on Product Spaces,
152 pointwise
converges, 60
Jensen’s inequality), 122 positive part, 66
Jensens’s inequality, 122 probability density function, 95
probability distribution, 57
Lebesgue integral, 11 probability measure, 37
Lebesgue measurable, 23 probability space, 95
Lebesgue measure, 29, 40 probability theory, 57, 95
Lebesgue outer measure, 14, 40 product σ-algebra, 153
length of the curve, 141 Product Measure, 152, 159
limit inferior, 59
limit superior, 59 Radon-Nikodym
locally compact, 132 derivative, 93
lower sum, 1 theorem, 94
random variable, 95
MCT, 92 real measurable function, 53
measurable, 56 restriction of
Borel, 54 the σ-algebra, 44
extended real valued , 54 the measure, 44
Lebesgue , 54 Riemann
rectangles, 153 Integrable, 87, 88
170 Index

Integral, 87
Riemann integrable, 110, 111
Riemann sum, 5

seminorm, 109
simple function, 67
simple measurable functions, 66
step function, 68
subadditivity, 18
supremum, 59
symmetric difference, 72

tag set, 5
Tonelli’s theorem, 162
topological space, 42
total variation, 141
translation invariance, 17

unbounded intervals, 12
upper sum, 1

Vitali
cover, 146
covering lemma, 147

Young’s inequality, 121

zero measure, 38

Potrebbero piacerti anche