U.U.D.M.

Project Report 2016:26

Lebesgue Theory
A Brief Overview

Ralf Pihlström

Examensarbete i matematik, 15 hp
Handledare: Anders Öberg
Examinator: Jörgen Östensson
Juni 2016

Department of Mathematics
Uppsala University
 Abstract
We present different concepts in measure theory. We construct the
Lebesgue measure on R. We define the Lebesgue integral and prove
some famous convergence theorems. We compare the Riemann integral
with the Lebesgue integral and prove some famous results. We give
some examples of how to use the Lebesgue integral.
By doing the above, we hope to give a brief overview of the Lebesgue
theory.

Contents
1 Introduction 2

2 Outline 4

3 Measure on a σ-algebra 5
3.1 σ-algebra of sets . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.2 Limits of sequences of sets . . . . . . . . . . . . . . . . . . . . 6
3.3 Generated σ-algebras . . . . . . . . . . . . . . . . . . . . . . . 7
3.4 Borel σ-algebras . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.5 Measure on a σ-algebra . . . . . . . . . . . . . . . . . . . . . 8
3.6 Measures of a sequence of sets . . . . . . . . . . . . . . . . . . 10
3.7 Measurable space and measure space . . . . . . . . . . . . . . 12

4 Outer measures 13
4.1 Construction of outer measures . . . . . . . . . . . . . . . . . 14

5 Lebesgue measure on R 15
5.1 The Borel σ-algebra of R . . . . . . . . . . . . . . . . . . . . 17
5.2 The invariance of the Lebesgue measure space . . . . . . . . . 17
5.3 Existence of non Lebesgue measurable sets . . . . . . . . . . . 19

6 Measurable functions 20
6.1 Measurability of functions . . . . . . . . . . . . . . . . . . . . 20
6.2 Measurability of sequences of functions . . . . . . . . . . . . . 21
6.3 Measurability of the positive and negative part of f . . . . . 22
6.4 Measurability of the restriction and extension of f . . . . . . 23
6.5 Almost everywhere . . . . . . . . . . . . . . . . . . . . . . . . 23

7 The Lebesgue Integral 24

7.1 Simple functions and approximations . . . . . . . . . . . . . . 24
7.2 Integration of simple functions . . . . . . . . . . . . . . . . . 25
7.3 Integration of nonnegative functions . . . . . . . . . . . . . . 25
7.4 Integration of measurable functions . . . . . . . . . . . . . . . 26
7.5 Properties of the Lebesgue integral . . . . . . . . . . . . . . . 26

2
 7.6 Convergence theorems . . . . . . . . . . . . . . . . . . . . . . 30

8 Comparison with the Riemann integral 38

8.1 Riemann integrability . . . . . . . . . . . . . . . . . . . . . . 38
8.2 The upper and lower envelope integrals . . . . . . . . . . . . 40
8.3 Riemann integrability and Lebesgue integrability . . . . . . . 43
8.4 Lebesgue’s integrability condition . . . . . . . . . . . . . . . . 44
8.5 Where Riemann falls short . . . . . . . . . . . . . . . . . . . . 44
8.6 Some examples of how to use the Lebesgue integral . . . . . . 46

1 Introduction
In 1904 Henri Lebesgue invented a new way of integrating functions. His
theory of integration was a generalization of that of Riemann’s—a larger set
of functions could be integrated and the problem of limits interacting badly
with integrals was solved.
In the center of the Lebesgue integral stands the following idea: the limit
of an integral should equal the integral of the limit. In other words,
Z Z
lim fn = lim fn . (1)
n→∞ n→∞

(This equality holds true in the Riemann sense too, but under less mild
conditions.) The aim of this paper is to establish (1). The following is some
background that will provide a starting point.
Suppose we have (1), and consider any nonnegative function f . Let
0 ≤ f1 ≤ f2 ≤ . . . be any sequence of simple functions1 approximating f
from below. (Such a sequence is guaranteed to exist.) Since integrals should
be sensitive to size,
Z Z
lim fn = sup{ fn }, n = 1, 2, 3, . . . .
n→∞

But Z Z
lim fn = f,
n→∞

so for any nonnegative function f we are prone to define

Z Z
f = sup{ fn },

where (fn ) is any sequence of simple functions approximating f from below.

But this is to say that
Z Z
f = sup{ ϕ}, (2)

1
A simple function is a function with finite range

3
where the supremum is taken over all simple function ϕ such that 0 ≤ ϕ ≤ f .
(For a negative functionR we simply consider −f .)
It remains to define ϕ where ϕ is simple. As we know, Riemann defined
his integral by partitioning the domain. Let us try something different.
Instead of partitioning the domain, let us partition the range. This was
Lebesgue’s breakthrough idea. It is the Lebesgue integral in a nutshell.
See Figure 1.

Figure 1: Riemann integration (blue) and Lebesgue integration (red). The Rie-
mann integral partitions the domain of f , while the Lebesgue integral partitions
the range of f

In a letter to Paul Montel, Lebesgue summarized his approach:

I have to pay a certain sum, which I have collected in my pocket.

I take the bills and coins out of my pocket and give them to the
creditor in the order I find them until I have reached the total
sum. This is the Riemann integral. But I can proceed differently.
After I have taken all the money out of my pocket I order the
bills and coins according to identical values and then I pay the
several heaps one after the other to the creditor. This is my
integral. —Henri Lebesgue

Inspired by Lebesgue, let us partition the range of ϕ. Then

Z X
ϕ= ai m(Di )

where
• ai is the different values ϕ assumes,

• Di are the points where ϕ assumes ai , and

4
 • m(Di ) is the size or measure of Di .

Note that Di can be written as

Di = {x : ϕ(x) = ai } = ϕ−1 (ai ).

We call Di the preimage of ai under ϕ.

If we want to integrate a large set of functions, measuring the preimage
can be quite intricate. This is why Lebesgue invented measure theory.

2 Outline
Section 3, Measure on a σ-algebra
We define a σ-algebra. We define sequences of sets and their limit.
We define generated σ-algebras. We define the Borel σ-algebra. We
define the concept of a measure on a σ-algebra. We define measure
sequences and their limits, and some important results regarding them.
We define the concept of measurable space and measure space.

Section 4, Outer measure

We define the concept of an outer measure µ∗ . We classify sets which
are outer measurable, and show that the collection of all such sets is
a σ-algebra. We show that µ∗ is a measure when restricted to the
collection of all outer measurable sets. We demonstrate a concrete
way of constructing an outer measure.

Section 5, Lebesgue measure on R

We construct the Lebesgue measure on R—using the Lebesgue outer
measure. We show that all open sets in R are Lebesgue measurable.
We prove some properties of the Lebesgue measurable space. We prove
the existence of non Lebesgue measurable sets in R.

Section 6, Measurable functions

We define measurable functions. We define measurability of sequences
of functions. We define the positive and negative part of f , the re-
striction and extension of f , and their measurability. We also define
the concepts of real-valued and convergence almost everywhere.

Section 7, Almost everywhere

In this section we introduce the concept of almost everywhere.

Section 8, The Lebesgue Integral

We define the Lebesgue integral. We investigate some of its properties.
We prove some famous convergence theorems, including Lebesgue’s
monotone convergence theorem, Fatou’s theorem and Lebesgue’s dom-
inated convergence theorem.

5
Section 9, Comparison with the Riemann integral
We compare the Riemann integral with the Lebesgue integral. We
show that if a bounded function is Riemann integrable over an interval
[a, b], it is Lebesgue integrable. We show that f is Riemann integrable
if and only if the set of discontinuity points has measure 0. We prove
the counterpart to the Lebesgue dominated convergence theorem in
the Riemann sense. We demonstrate in a few examples how to use the
Lebesgue integral.

3 Measure on a σ-algebra
Notations. N = {1, 2, 3, . . .}, Z = {0, ±1, ±2, . . .}, R is the real line. P(X)
is the collection of all subsets of a set X.
Let X be any set. Suppose we want a function that measures subsets of
X. Let us call such a function a measure, and denote it by

m : P(X) → [0, ∞]. (3)

m should meet certain criteria that we normally associate with measur-

ing. (Note that one such criterion is already made implicit in (3); since no
measure can be negative, the range of m is positive or zero.)
m should be countably additive, meaning that for any disjoint sets A and
B in P(X),
m(A ∪ B) = m(A) + m(B).
Furthermore, m should be invariant under translation. This means that
when we translate a set in space, its measure should stay the same.
Ideally the domain of m would be P(X), so that every subset of X could
be measured. Unfortunately this is not possible. In 1914 Felix Hausdorff
showed that is impossible to define a measure on P(R) that is countably
additive and at the same time invariant under translation. This is why we
introduce σ-algebras. As we will see, a σ-algebra is a collection of subsets of
X that is measurable with respect to some measure m.

3.1 σ-algebra of sets

Definition 1 (Algebra of subsets). Let X be a set. A collection A of subsets
of X is called an algebra of subsets of X if it satisfies the following conditions:
1 ◦ X ∈ A,

2 ◦ A ∈ A ⇒ Ac ∈ A,

3 ◦ A, B ∈ A ⇒ A ∪ B ∈ A.
Lemma 1. If A is an algebra of subsets of X, then

6
 (1) A, B ∈ A ⇒ A ∩ B ∈ A,

(2) A, B ∈ A ⇒ A \ B ∈ A.

Proof. A ∩ B = (Ac ∪ B c )c ∈ A, A \ B = A ∩ B c ∈ A.

Definition 2 (σ-algebra of subsets). An algebra A of subsets of a set X is

called a σ-algebra if is satisfies the following additional condition:

4◦ (An : n ∈ N) ⊂ A ⇒
S
n∈N An ∈ A.

Example 1. P(X) is a σ-algebra of subsets of X. It is the greatest σ-

algebra of subsets of X in the sense that if A is a σ-algebra of subsets of X
and if P(X) ⊂ A, then A = P(X).

Lemma
T 2. If A is a σ-algebra of subsets of set X, then (An : n ∈ N) ⊂ A
⇒ n∈N An ∈ A.
T S
c
Proof. n∈N An = n∈N An ∈ A.

3.2 Limits of sequences of sets

Definition 3 (Increasing and decreasing sequences of sets). Let (An : n ∈ N)
be a sequence of subsets of X. (An : n ∈ N) is said to be an increasing
sequence and we write An ↑ if An ⊂ An+1 for n ∈ N. We say that (An :
n ∈ N) is a decreasing sequence and write An ↓ if An+1 ⊂ An for n ∈ N.
A sequence (An : n ∈ N) is called monotone if it is either increasing or
decreasing. For an increasing sequence, we define
[
lim An = An = {x ∈ X : x ∈ An for some n ∈ N}.
n→∞
n∈N

For a decreasing sequence (An : n ∈ N), we define

\
lim An = An = {x ∈ X : x ∈ An for every n ∈ N}.
n→∞
n∈N

In order to define the limit of sequences of sets, we must define the limit
inferior and the limit superior of sequences of sets.

Definition 4 (Limit inferior, limit superior). The limit inferior and the
limit superior of a sequence (An : n ∈ N) of subsets of a set X is defined by
[ \
lim inf An = Ak ,
n→∞
n∈N k≥n
\ [
lim sup An = Ak .
n→∞
n∈N k≥n

7
Definition 5 (Convergent sequence). Let (An : n ∈ N) be an arbitrary
sequence of subsets of a set X. We say that the sequence converges if
lim inf n→∞ An = lim supn→∞ An , and we set

lim An = lim inf An = lim sup An .

n→∞ n→∞ n→∞

If this does not hold, we say that the sequence diverges.

Remark 1. It can be shown that for any sequence (An : n ∈ N), both
lim inf n→∞ An and lim supn→∞ An lies in A. In particular if (An : n ∈ N)
converges, then limn→∞ An ∈ A.

3.3 Generated σ-algebras

It is useful to define the idea of a smallest σ-algebra.

Definition 6 (Collection of indexed sets). Let A be any set. We call {Eα :

α ∈ A} a collection of sets indexed by A.

Remark 2. Let {Aα : α ∈ A} be a collection of σ-algebras of subsets of X

where A is an arbitrary indexing set. It is routine to verify that ∩α∈A Aα is
a σ-algebra of subsets of X.

Theorem 1. Let C be any collection of subsets of X. There exists a smallest

σ-algebra of A0 of subsets of X containing C, smallest in the sense that if
A is a σ-algebra of subsets of X containing C then A0 ⊂ A. Similarly there
exists a smallest algebra containing C.

Proof. P(X) is a σ-algebra of X containing C. Let {Aα : α ∈ A} be the

collection
T of all σ-algebras of subsets of X containing C. By Remark 2,
α∈A Aα is a σ-algebra that contains C. It is the smallest
T such σ-algebra,
since for any σ-algebra A containing C, we have A ⊃ α∈A Aα .

Definition 7 (σ-generated set). For any collection C of subsets of a set X,

we write σ(C) for the smallest σ-algebra of subsets of X containing C and
call it the σ-algebra generated by C.

3.4 Borel σ-algebras

Let us review some topology.

Definition 8 (Topology, topological space). Let X be a set. A collection of

subsets of X is called a topology on X if it satisfies:

1 ◦ ∅ ∈ D,

2 ◦ X ∈ D,

8
 3 ◦ {Eα : α ∈ A} ⊂ D ⇒
S
α∈A Eα ∈ D,

4 ◦ E1 , E2 ∈ D ⇒ E1 ∩ E2 ∈ D.

The pair (X, D) is called a topological space, and the sets in D are called the
open sets of the topological space.

Definition 9 (Borel set). We call the σ-algebra of D the Borel σ-algebra of

subsets of the topological space (X, D) and we write BX or B(X) for it. We
call its members the Borel sets of the topological space.

Definition 10. Let (X, D) be a topological space. A set E ⊂ X is called a

Gδ -set if it is the intersection of countably many open sets. A subset E of
X is called an Fσ -set if it the union of countably many closed sets.

If introducing the concept of topology seems mysterious, let us justify

it here. We will use the Borel sets on R when we talk about the Lebesgue
measure on R. Since B(R) is a σ-algebra, it necessarily contains all open
sets, all closed sets, all unions of open sets, all unions of closed sets, all
intersections of closed sets, and all intersections of open sets. So by starting
with open sets, we can—by the virtue of B(R) being a σ-algebra—generate
a very large subset of P(R). The Borel sets on R can be thought of as all
conceivable subsets of R.

3.5 Measure on a σ-algebra

Notation 1. Let R = {−∞}∪R∪{∞} be the extended real number system.

Definition 11. Let C be a collection of subsets of a set X. Let γ be a non-

negative extended real-valued set function on C. We say that

(a) γ is monotone on C if γ(E1 ) ≤ γ(E2 ) for E1 , E2 ∈ C such that E1 ⊂

E2 .

(b) γ is additive on C if γ(E1 ∪ E2 ) = γ(E1 ) + γ(E2 ) for E1 , E2 ∈ C such

that E1 ∩ E2 = ∅.

(c) γ is finitely additive on C if γ(∪nk=1 Ek ) = nk=1 γ(Ek ) for every dis-

P
joint finite sequence (Ek : k = 1, . . . , n) in C such that ∪nk=1 Ek ∈ C.
Pn
(d) γ is countably additive on C if γ(∪nk=1 Ek ) = k=1 γ(Ek ) for every
disjoint sequence (En : n ∈ N) in C such that ∪nn∈N En ∈ C.

(e) γ is subadditive on C if γ(E1 ∪ E2 ) ≤ γ(E1 ) + γ(E2 ) for E1 , E2 ∈ C

such that E1 ∪ E2 ∈ C.

(f ) γ is finitely subadditive on C if γ(∪nk=1 Ek ) ≤ nk=1 γ(Ek ) for every

P
finite sequence (Ek : k = 1, . . . , n) in C such that ∪nk=1 Ek ∈ C.

9
 (g) γ is countable subadditive on C if γ(∪nn∈N En ) ≤ n∈N γ(En ) for every
P
sequence (En : n ∈ N) in C such that ∪n∈N En ∈ C.
Lemma 3. For any sequence (En : n ∈ N) in an algebra A of subsets of X,
there exists a disjoint sequence (Fn : n ∈ N) in A such that
N
[ N
[ [ [
(1) En = Fn and (2) En = Fn .
n=1 n=1 n∈N n∈N
S
In particular, if A is a σ-algebra, then n∈N Fn ∈ A.
Proof. Let F1 = E1 and Fn = En \ (∪n−1
k=1 Ek ) ∈ A, and use induction.

Lemma 4. Let γ be a nonnegative extened real-valued set function on an

algebra A of subsets of a set X.
(a) If γ is additive on A, it is (1) finitely additive, (2) monotone and (3)
finitely subadditive on A.
(b) If γ is countably additive on A, then it is countably subadditive on A.
Proof. (a). The proof of (1) and (2) is routine and independent of (3).
Let us prove (3). Let (Ek : k = 1, . . . , n) be a finite sequence in A, then
using Lemma 3, (1) and (2),
n
[ n
[ n
X n
X
γ( Ek ) = γ( Fk ) = γ(Fk ) ≤ γ(Ek ).
k=1 k=1 k=1 k=1

This proves (a). The proof of (b) is done similarly.

Proposition 1. Let γ be a nonnegative extended real-valued set function on

an algebra Aof subsets of a set X. If γ is additive and countably subadditive
on A then γ is countably additive on A.
Definition 12 (Measure). Let A be a σ-algebra of subsets of a set X. A
set function µ defined on A is called a measure if it satisfies the following
conditions:
1 ◦ µ(E) ∈ [0, ∞] for every E ∈ A,
2 ◦ µ(∅) = 0,
3 ◦ µ is countably additive.
Remark 3. Note that
(1) µ is finitely additive,
(2) µ is monotone,
(3) E1 , E2 ∈ A, E1 ⊂ E2 , µ(E1 ) < ∞ ⇒ µ(E2 \ E1 ) = µ(E2 ) − µ(E1 ),
(4) µ is countably subadditive.

10
3.6 Measures of a sequence of sets
Theorem 2 (Monotone convergence theorem for sequences of measurable
sets). Let µ be a measure on a σ-algebra A of subsets of a set X. Let (En :
n ∈ N) be a monotone sequence in A.

(a) If En ↑, then lim µ(En ) = µ( lim En ).

n→∞ n→∞

(b) If En ↓, then lim µ(En ) = µ( lim En ), provided that there exists a

n→∞ n→∞
set A ∈ A with µ(A) < ∞ such that E1 ⊂ A.

Sn0 ) = ∞ for some n0 ∈ N,

Proof. (a) Suppose En ↑. Then µ(En ) ↑. If µ(E
lim µ(En ) = ∞. Since En is increasing, En0 ⊂ n∈N En = lim En , and so
n→∞ n→∞
µ( lim En ) ≥ µ(En0 ) = ∞.
n→∞
If µ(En ) < ∞ for all n ∈ N, consider the disjoint sequence (Fn : n ∈ N)
S A definedS by Fn = En \ En−1 for n ∈ N where E0 = ∅. Then since
in
n∈N En = n∈N Fn ,
[ X X
µ( lim En ) = µ( Fn ) = µ(Fn ) = [µ(En ) − µ(En−1 )]
n→∞
n∈N n∈N n∈N
n
X
= lim [µ(Ek ) − µ(Ek−1 )] = lim [µ(En ) − µ(E0 )]
n→∞ n→∞
k=1
= lim µ(En ).
n→∞

(b) Let (Fn : n ∈ N) be disjoint sequence in A defined by Fn = En \ En+1

for n ∈ N. Then
\ [
E1 \ En = Fn . (4)
n∈N n∈N
T
To see this, let x ∈ E1 \ n∈N En . Since En ↓, there exists S a smallest
n0 ∈ N such that x ∈/ EnS
0 +1 . Then x ∈ En0 \ En0 +1 = F n0 ⊂ n∈N Fn .
Conversely, if x ∈ n∈N Fn , then x ∈ Fn0 = En0 \ En0 +1 T for some
n0 ∈ N. Thus Tx ∈ En0 ⊂ E1 , and since x ∈ / En0 +1 , we have x ∈
/ n∈N En .
Thus x ∈ E1 \ n∈N En . This proves (1).
By (1),
! !
\ [
µ E1 \ En = µ Fn , (5)
n∈N n∈N

where, since En ↓,
! !
\ \
µ E1 \ En = µ(E1 ) − µ En = µ(E1 ) − µ( lim En ). (6)
n→∞
n∈N n∈N

11
Now
!
[ X X
µ Fn = µ(Fn ) = µ(En \ En+1 ) (7)
n∈N n∈N n∈N
X n
X
= [µ(En ) − µ(En+1 )] = lim [µ(En ) − µ(En+1 )] (8)
n→∞
n∈N k=1
= lim [µ(E1 ) − µ(En+1 )] = µ(E1 ) − lim µ(En+1 ). (9)
n→∞ n→∞

Substituting (3) and (4) into (2), we arrive at the desired result.

Theorem 3. Let µ be a measure on a σ-algebra of subsets of a set X.

(a) For an arbitrary sequence (En : n ∈ N) in A, we have

µ(lim inf En ) ≤ lim inf µ(En ).

n→∞ n→∞

(b) If there exists A ∈ A with µ(A) < ∞ such that En ⊂ A for n ∈ N,

then
µ(lim sup En ) ≥ lim sup µ(En ).
n→∞ n→∞

(c) If both lim En and lim µ(En ) exist, then

n→∞ n→∞

µ( lim En ) ≤ lim µ(En ).

n→∞ n→∞

(d) If lim En exists and if there exists A ∈ A with µ(A) < ∞ such that
n→∞
En ⊂ A for n ∈ N then lim µ(En ) and
n→∞

µ( lim En ) = lim µ(En ).

n→∞ n→∞

Proof. 1. lim inf En = ∪n∈N ∩k≥n Ek = limn→∞ ∩k≥n Ek . Using (a) in

n→∞
Theorem 2, we have
\ \ \
µ(lim inf Ek ) = lim inf µ( Ek ) = lim sup µ( Ek ) ≤ lim inf µ(En ),
n→∞ n→∞ n→∞ n→∞
k≥n k≥n k≥n

2. Assume what is given. Similarly as above,

[ [ [
µ(lim sup Ek ) = lim sup µ( Ek ) = lim sup µ( Ek ) ≤ lim sup µ(En ),
n→∞ n→∞ n→∞ n→∞
k≥n k≥n k≥n

since ∪k≥n Ek ⊂ En ⊂ A.
3. By (a),

µ( lim En ) = µ(lim inf En ) ≤ lim inf µ(En ) = lim µ(En ).

n→∞ n→∞ n→∞ n→∞

12
 4. By (b) and (a),

lim sup µ(En ) ≤ µ(lim sup En ) = µ(lim inf En ) ≤ lim inf µ(En ).
n→∞ n→∞ n→∞ n→∞

But lim inf n→∞ µ(En ) ≤ lim supn→∞ µ(En ) for any sequence (En : n ∈
N), and the result follows.

3.7 Measurable space and measure space

Definition 13 (Measurable space, A-measurable set). Let A be a σ-algebra
of subsets of a set X. The pair (X, A) is called a measurable space. A subset
E of X is said to be A-measurable if E ∈ A.

Definition 14 (Measure space). If µ is a measure on a σ-algebra of subsets

of a set X, we call the triple (X, A, µ) a measure space.

Remark 4. Note that while (X, P(X)) is a measurable space, it is not a

measure space.

Definition 15 (Null set). Given a measure µ on a σ-algebra of A of subsets

of a set X, a subset E of X is called a null set with respect to the measure µ
if E ∈ A and µ(E) = 0. In this case we also say that E is a null set in the
measure space (X, A, µ).

Definition 16 (Complete σ-algebra). Given a measure µ on a σ-algebra

of A of subsets of a set X, we say that the σ-algebra of A is complete with
respect to the measure µ if an arbitrary subset E0 of a null set with respect
to µ is a member of A.

4 Outer measures
In this section we will introduce the outer measure. By its help, we will
connect the two concepts measure and σ-algebra.

Definition 17 (Outer measure). Let X be a set. A set function µ∗ defined

on the σ-algebra P(X) of all subsets of X is called an outer measure on X
if it satisfies the following conditions:
(1) it is nonnegative extended real-valued,
(2) µ∗ (0) = 0,
(3) it is monotone,
(4) it is countably subadditive.

Let us define measurability with respect to a measure.

13
Definition 18 (µ∗ -measurable set). Let µ∗ be an outer measure on a set
X. We say that E ∈ P(X) is measurable with respect to µ∗ (or simply µ∗ -
measurable) if it satisfies the so-called Caratheodory condition:

µ∗ (A) = µ∗ (A ∩ E) + µ∗ (A ∩ E c ) for every A ∈ P(X).

The set A is called a testing set in the Caratheodory condition. We write

M(µ∗ ) for the collection of all µ∗ -measurable sets E ∈ P(X).

In some sense a measurable set works as a ”ruler” of other sets.

Remark 5. By subadditivity, to verify the Caratheodory condition it suf-

fices to show that

µ∗ (A ∩ E) + µ∗ (A ∩ E c ) ≤ µ∗ (A) for every A ∈ P(X).

Lemma 5. Let µ∗ be an outer measure on a set X.

(a) If E1 , E2 ∈ M(µ∗ ), then E1 ∪ E2 ∈ M(µ∗ )

(b) µ∗ (E1 ∪ E2 ) = µ∗ (E1 ) + µ∗ (E2 ) for every disjoint E1 , E2 ∈ M(µ∗ ).

Proof. Let us prove (b). Let E1 ∪ E2 be the testing set in the Caratheodory
condition. Since E1 and E2 are disjoint, µ∗ (E1 ∪ E2 ) = µ∗ ((E1 ∪ E2 ) ∩ E1 ) +
µ∗ ((E1 ∪ E2 ) ∩ E1c ) = µ∗ (E1 ) + µ∗ (E2 ).

Theorem 4. Let µ∗ be an outer measure on a set X. Then µ∗ is additive

on P(X) if and only if every member of P(X) is µ∗ -measurable.

Proof. (⇒) Suppose µ∗ is additive and let E ∈ P(X). Then A ∩ E and

A ∩ E c are disjoint with union A, so the Caratheodroy condition is satisfied.
Thus P(X) ⊂ M(µ∗ ). (⇐) Follows from (b) of Lemma 5.

Remark 6. Let µ∗ be an outer measure on a set X. It can be seen that

M(µ∗ ) is a σ-algebra of subsets of X.

Theorem 5. Let µ∗ be an outer measure on a set X. If we let µ be the

restriction of µ∗ to the σ-algebra M(µ∗ ), then µ is a measure on M(µ∗ )
and furthermore (X, M(µ∗ ), µ) is a complete measure space.

Proof. µ∗ is countably subadditive on P(X) and thus µ is countably subad-

ditive on M(µ∗ ). By Lemma 5, µ∗ is additive on M(µ∗ ), and thus countably
additive on M(µ∗ ) by Proposition 1. Let E ∈ M(µ∗ ) and µ(E) = 0. Let
E0 ⊂ E. By monotonicity, µ(E0 ) = 0, and

µ∗ (A ∩ E0 ) + µ∗ (A ∩ E0c ) ≤ µ∗ (A),

so E0 ∈ M(µ∗ ). Thus the space is complete.

14
 This is take-away: an outer measure induces a σ-algebra. And since the
measure is just the restriction of the outer measure, one may say that a
measure induces a σ-algebra. This way, the concept of a σ-algebra becomes
more clear.

Definition 19 (Borel outer measure). An outer measure µ∗ on a topological

space X is called a Borel outer measure if B X ⊂ M(µ∗ ).

4.1 Construction of outer measures

Definition 20 (Covering class). A collection B of subsets of a set X is called
a covering class if it satisfies the following conditions:

1 ◦ there exists (Vn : n ∈ N) ⊂ B such that n∈N Vn = X,

S

2 ◦ ∅ ∈ B.
S
For every E ∈ P(X), if (Vn : n ∈ N) ∈ B such that E ⊂ n∈N Vn then
(Vn : n ∈ N) is a covering sequence for E.

Theorem 6. Let B be a covering class of subsets of a set X. Let γ be an

arbitrary set function on B such that

1 ◦ γ is nonnegative extended real-valued,

2 ◦ γ(∅) = 0.

Let us define a set function µ∗ on P(X) by setting for every E ∈ P(X),

X [
µ∗ (E) = inf{ γ(Vn : n ∈ N) : (Vn : n ∈ N) ⊂ B, E ⊂ Vn }.
n∈N

Then µ∗ is an outer measure on X, called the outer measure based on γ.

Proof. Let us verify that µ∗ satisfies the conditions in Definition 17.

(1) Clearly µ∗ (E) ∈ [0, ∞].
(2) ∅ ⊂ (∅) ⇒ µ∗ (∅) = 0.
(3) For any E1 , E2 ∈ P(X) such that E1 ⊂ E2 , we indeed have µ∗ (E1 ) ≤
∗
µ (E2 ) since any covering sequence of E2 is a covering sequence of E1 .
(4) Let En be a sequence in P(X). Let  > 0 be given. Then for
P n ∈ N, there
each
∗
exists a sequence Vn,k such that En ⊂ ∪k∈N Vn,k and

k∈N γ(Vn,k ) ≤ µ (En ) + 2n . Then
!
∗
[ XX X   X ∗
µ En ≤ γ(Vn,k ) ≤ µ∗ (En ) + n = µ (En ) + .
2
n∈N n∈N k∈N n∈N n∈N

The result now follows from the arbitrariness of .

15
5 Lebesgue measure on R
In this section we construct the Lebesgue measure on R (using the Lebesgue
outer measure) and demonstrate some of its properties.

Definition 21. Let J 0 be the collection of ∅ and all open intervals in R.

Let J be the collection of all intervals in R. For an interval I in R with
endpoints a, b ∈ R, a < b, we define l(I) = b − a. For an infinite interval I
in R we define l(I) = ∞. We set l(∅)P = 0. For a countable disjoint collection
{In : n ∈ N} we define l(∪n∈N In ) = n∈N l(In ). As in the previous section,
we set
X [
µ∗ (E) = inf{ γ(Vn ) : (Vn : n ∈ N) ⊂ J0 , E ⊂ Vn },
n∈N

and call it the Lebesgue outer measure on R. We write ML for the σ-algebra
M(µ∗L ) of µ∗L -measurable sets E ∈ B(R) and call it the Lebesgue σ-algebra
of subsets of R. Members of the σ-algebra ML are called ML -measurable or
Lebesgue measurable sets. We call (R, ML ) the Lebesgue measurable space.
We write µL for the restriction of µ∗L to ML and call it the Lebesgue measure
on R. We call (R, ML , µL ) the Lebesgue measure space on R.

Theorem 7. µ∗L (I) = l(I) for every interval I in R.

Proof. 1. If I is finite and closed, I = [a, b] for some a, b ∈ R. Consider

the covering sequence ((a − , b + ), ∅, ∅, . . .) in J 0 for I. It follows that
µ∗L (I) ≤ l(I).
Next we show that for any covering sequence (In : n ∈ N) in J 0 for I,
we have X
l(In ) ≥ l(I). (10)
n∈N

If any of the intervals is infinite, (10) holds. Thus consider the case
where every member is finite. Let us drop those members in the covering
sequence that is disjoint from I and contained in any other member of the
sequence. The resulting sequence Jn is a covering sequence of I, and since I
is compact, Jn has a finite subcover. Renumber the members of Jn so that
Jk = (ak , bk ) for k = 1, . . . , N and a1 ≤ a2 ≤ . . . ≤ aN . In fact since none of
the members are contained in another, we have a1 < a2 < . . . < aN . Let us
show that a2 < b1 . Assume not, then since J1 and J2 lies in I, there exist
x1 ∈ (a1 , b1 ) ∩ I and x2 ∈ (a2 , b2 ) ∩ I such that a1 < x1 < b1 ≤ a2 < x2 < b2 .
Note that [x1 , x2 ] ⊂ I. Since b1 ≤ a2 , there exists at least on point in [x1 , x2 ]

16
that is not covered by (J1 , . . . , Jn ). Thus we have a2 < b1 . Similarly,

a1 < a2 < b1
a2 < a3 < b2
...
aN −1 < aN < bN −1
aN < bN .

We get
N
X
l(Jk ) = (b1 − a1 ) + (b2 − a2 ) + . . . + (bN − aN )
k=1
> (a2 − a1 ) + (a3 − a2 ) + . . . + (bN − an )
= bN − a1 ≥ b − a = l(I).

Thus n∈N l(In ) ≥ N

P P
k=1 l(Jk ) ≥ l(I), establishing (10). By the definition
of infimum, (10) implies that µ∗L (I) ≥ l(I). Thus for any closed and finite
interval I, we have µ∗L (I) = l(I).
2. For any open interval I = (a, b), we have

µ∗L ((a, b)) ≤ µ∗L ([a, b]) ≤ µ∗L ({a}) + µ∗L ((a, b)) + µ∗L ({b}) = µ∗L ((a, b)),

since the Lebesgue measure of a singleton is 0. Thus µ∗L ((a, b)) = µ∗L ([a, b]) =
l([a, b]) = l((a, b)).
3. If I is finite and I = (a, b], µ∗L ((a, b]) ≤ µ∗L ((a, b)) and µ∗L ((a, b]) ≥
µ∗L ((a, b)) by monotonicity. Similarly if I = [a, b).
4. If I is infinite of the type I = (a, ∞), a ∈ R, then (a, ∞) ⊂ (a, n) for
every n ∈ N and thus µ∗L ((a, ∞)) ≥ µ∗L ((a, n)) = n − a. By the arbitrariness
of n ∈ N, µ∗L ((a, ∞)) = ∞ = l(a, ∞). Similarly for other types.

Remark 7. From Theorem 7, one can show that every interval in R is

Lebesgue measurable. This begs the question: are all sets in R Lebesgue
measurable? As we will see, the answer is no. In practice, however, any
subset of R that we can think of will be Lebesgue measurable. This is
why we introduce the Borel σ-algebra of R. Its role is so important that it
deserves a separate section.

5.1 The Borel σ-algebra of R

Definition 22 (Borel σ-algebra of R). The Borel σ-algebra of R, written
B(R), is the σ-algebra generated by the open sets in R. That is, if D is the
collection of all open sets in R, then B(R) = σ(D).

Theorem 8. Every Borel set in R is a Lebesgue measurable set.

17
Proof. Let D be the collection of all open sets in R. Since any open set
in R can be written as a union of countable many open intervals, we have
D ⊂ ML and thus σ(D) = B R ⊂ ML .

It will be useful to know that B(R) can be generated by intervals on R:

Proposition 2. The Borel σ-algebra on R can be generated by any of the
following collections of intervals:
{(−∞, b) : b ∈ R}, {(−∞, b] : b ∈ R}, {(a, ∞) : a ∈ R}, {[a, ∞) : b ∈ R}.
By generated, we mean that for example B(R) = σ({(−∞, b) : b ∈ R}).
Remark 8. We will not prove it here, but there are Lebesgue measurable
sets on R that are not members of the Borel σ-algebra on R, so the inclusion
in Theorem 8 is a true inclusion.

5.2 The invariance of the Lebesgue measure space

The Lebesgue measure on R is invariant under different transformations.
Definition 23. Let X be a linear space over the field of scalars R.
(a) For E ⊂ X and x0 ∈ X, we write
E + x0 = {x + x0 : x ∈ E}.
(b) For a ∈ R, we write
αE = {αx : x ∈ E}.
(c) For a collection C of subsets of X, x ∈ X, and α ∈ R, we write
C + x = {E + x : E ∈ C} and αC = {αE : E ∈ C}.
Theorem 9 (Translation invariance of the Lebesgue measure space). The
Lebesgue measure space (R, ML , µL ) is translation invariant, that is, for
every E ∈ ML and x ∈ R we have E + x ∈ ML and µL (E + x) = µL (E).
Let ML + x = {E + x : E ∈ ML }. Then ML + x = ML for every x ∈ R.
Proof. Let E ∈ ML and x ∈ R. Let us show that E + x ∈ ML by verifying
the Caratheodory condition for E + x. (In the proof we will use that µ∗L (E +
x) = µ∗L (E), for every E ∈ B(R), which is not difficult to prove.) We get
µ∗L (A ∩ (E + x)) + µ∗L (A ∩ (E + x)c )
= µ∗L (A ∩ (E + x) − x) + µ∗L (A ∩ (E + x)c − x)
= µ∗L ((A − x) ∩ E) + µ∗L ((A − x) ∩ E c )
= µ∗L (A − x) = µ∗L (A).
This shows that E + x ∈ ML and therefore µL (E + x) = µ∗L (E + x) =
µ∗L (E) = µL (E).
Now since E + x ∈ ML for every E ∈ ML , we have ML + x ⊂ ML .
But ML = ML + (−x) + x ⊂ ML + x, and we are done.

18
Theorem 10. For E ∈ B(R) and a ∈ R, let αE = {y ∈ R : y =
ax for some x ∈ E}. Then µ∗L (αE) = |α|µ∗L (E).

Proof. Let S = (In : n ∈ N) be a sequence in J 0 , and let S be the collection

of all such sequences. For a ∈ R, α 6= 0, define a function Mα : S → S by
setting Mα (S) = αS. Clearly Mα is one-to-one.S For an arbitrary E ∈ B(R),
let S E be all sequences in S such that E ⊂ n∈N In . Then Mα (S E )= S αE
is a one-to-one mapping
S of S E onto S αE , where S αE is all sequences in S
such that αE ⊂ n∈N αIn .
Let λ bePa nonnegative extended real-valued set function on S defined
by λ(S) = n∈N l(In ) for S = (In : n ∈ N). Then
X X
λ(Mα (S)) = l(αIn ) = |α| l(In ) = |α|λ(S).
n∈N n∈N

Now µ∗L (E) = inf S∈SE λ(S) and µ∗L (αE) = inf T ∈SαE λ(T ). But Mα (S E )
is one-to-one, so inf T ∈SαE λ(T ) = inf S∈SE λ(Mα (S)) = inf S∈SE |α|λ(S) =
|α|µ∗L (E). Thus µ∗L (αE) = |α|µ∗L (E) when α 6= 0. For α = 0 the equality is
trivial.

Theorem 11 (Positive homogeneity of the Lebesgue measure space). For

every set E ∈ ML and α ∈ R, we have αE ∈ ML and µL (αE) = |α|µL (E).
For every α ∈ R, let αML = {αE : E ∈ ML }. Then αML = ML for every
α ∈ R such that α 6= 0.

Proof. For α = 0 the theorem is trivial. Assume therefore α 6= 0. Let

E ∈ ML and A ∈ B(R), then α1 A ∈ B(R) and

1 1 1
µ∗L ( A) = µ∗L ( A ∩ E) + µ∗L ( A ∩ E c ). (11)
α α α
By Theorem 10,
1 1 ∗
µ∗L ( A) = µ (A),
α |α| L
1 1 ∗ α
µ∗L ( A ∩ E) = µL (A ∩ ),
α |α| E
1 1 1
µ∗L ( A ∩ E c ) = µ∗ ( A ∩ (αE c )).
α |α| L α

Substituting these into (11), we get the Caratheodory condition for αE, so
that αE ∈ ML . Then by Theorem 10, µL (αE) = |α|µL (E). Since αE ∈ ML
for every E ∈ ML , we have αML ⊂ ML . But ML = α α1 ML ⊂ α ML , and
we conclude.

19
5.3 Existence of non Lebesgue measurable sets
Not all sets in R are Lebesgue measurable.

Let us define addition modulo 1 of x, y ∈ [0, 1) by

(
◦ x+y if x + y < 1,
x+y=
x + y − 1 if x + y ≥ 1.
◦
It is easily seen that + is commutative and associative. For E ∈ [0, 1)
◦ ◦
and y ∈ [0, 1), let E + y = {z ∈ [0, 1) : z = x + y for some x ∈ E}.
◦
Remark 9. Let E ⊂ [0, 1) and E ∈ ML . It can be proven that E + y ∈ ML
◦
and µL (E + y) = µL (E) for every y ∈ [0, 1).

Theorem 12. [0, 1) ⊂ R contains a non Lebesgue measurable set.

Proof. For x, y ∈ [0, 1), let us define a equivalence relation on [0, 1) by

x ∼ y if and only if x − y is a rational number.

Let {Eα : α ∈ A} be the collection of equivalence classes of ∼.

Let P be the subset of [0, 1) constructed by for each α ∈ A, picking an
element from Eα . Let {rn : n ∈ Z+ } be an enumeration of the rational
numbers in [0, 1) with r0 = 0. Let
◦
Pn = P + rn for n ∈ Z+ .

Let us show that {Pn : n ∈ Z+ } is a disjoint collection. Assume not.

Then for m 6= n, there exists x ∈ Pm ∩ Pn . Then x ∈ Pm and x ∈ Pn so that
◦ ◦
x = pm + rm = pn + rn for some pm , pn ∈ P . This implies that pm − pn is
rational, so pm ∼ pn . By the construction of P , we must then have pm = pn ,
implying that rm = rn , and then m = n; contradiction, proving that indeed
{Pn : n ∈ Z+ } is disjoint. Next, let us show that
[
Pn = [0, 1). (12)
n∈Z+

S Let us note that since Pn ⊂ [0, 1) for each n ∈ N, we trivially have

n∈Z+ Pn ⊂ [0, 1). To prove the reverse inclusion, let x ∈ [0, 1). Then
x ∈ Eα for some α. Since P contains an element from each equivalent class,
there exists p ∈ P such that p ∈ Eα . Thus x ∼ p so that x and p differ by
some rational number rn , n ∈ Z+ . If x ≥ p then x = p + rn ∈ Pn . If x < p,
◦
x = p − rn . Let rm = 1 − rn ∈ [0, 1), then x = p + rm − 1 = p + rm so that
x ∈ Pm . This shows that (12) holds.

20
 Finally, let us show that P ∈
/ ML . Assume the contrary, then by Remark
◦
9, Pn = P + rn ∈ ML and µL (Pn ) = µL (P ). Thus
 
[ X X
1 = µL ([0, 1)) = µL  Pn  = µL (Pn ) = µL (P ). (13)
n∈Z+ n∈Z+ n∈Z+

If µL (P ) = 0, then (13) says 1 = 0. If µL (P ) > 0, then (13) says 1 = ∞.

Contradiction and we conclude.

6 Measurable functions
It is natural to develop a concept of measurable functions, that is, functions
whose preimage is measurable.
A measurable function pulls back measurable sets to measurable sets,
much like a continuous function pulls back open sets to open sets.

6.1 Measurability of functions

Definition 24 (Measurable function). Let (X, A) and (Y, B) be measurable
spaces. A function f : X → Y is measurable if f −1 (B) ∈ A for every B ∈ B.
Definition 25 (Measurable function on R, Borel set). If (X, A) is a mea-
surable space, then f : X → R is measurable if f −1 (B) ∈ A for every Borel
set B ∈ B(R).
Since B(R) = σ({[−∞, b) : b ∈ R}) by Proposition 2, we can make the
following definition.
Definition 26 (Measurable function on R). Let (X, A) be an arbitrary mea-
surable space and let D ∈ A. An extended real-valued function f defined on
D is said to be A-measurable on D if {x ∈ D : f (x) ≤ α} ∈ A for every
α ∈ R.
Every real-valued continuous function is measurable:
Theorem 13. Let f : R → R be continuous. Then f is measurable.
Proof. Let O be an arbitrary open set in R. Since f is continuous, f −1 (O)
is open in R and then f −1 (O) ∈ BR , since any open set in R is a countable
union of disjoint open intervals. By Theorem 8, f −1 (O) is measurable.

Theorem 14. Each of the following conditions are equivalent.

(a) D1 = {x ∈ D : f (x) ≤ α} ∈ A for every α ∈ R,
(b) D2 = {x ∈ D : f (x) > α} ∈ A for every α ∈ R,
(c) D3 = {x ∈ D : f (x) ≥ α} ∈ A for every α ∈ R,
(d) D4 = {x ∈ D : f (x) < α} ∈ A for every α ∈ R.

21
Proof. 1. (a) ⇔ (b). D1 and D2 partition D. Thus if D1 ∈ A then D2 =
D \ D1 ∈ A. Similarly, if D2 ∈ A then D1 ∈ A.
2. (c) ⇔ (d) as above.
3. (d) ⇒ (a). Note that
\ 1
D1 = {D : f < α + }. (14)
n
n∈N

Now if f satisfies (d), then every intersection in (14) lies in A.

4. (b) ⇒ (c). Note that
\ 1
D3 = {D : f > α − }. (15)
n
n∈N

Now if f satisfies (b), every intersection in (15) lies in A.

Theorem 15. If f is measurable, then |f | is measurable.

Proof. {x : |f (x)| < α} = {x : f (x) < α} ∩ {x : f (x) > −α} ∈ A, by

Theorem 14.

6.2 Measurability of sequences of functions

Theorem 16. Let (fn : n ∈ N) be a sequence of measurable functions. For
x ∈ X, put

g1 (x) = sup fn (x),

n∈N
g2 (x) = inf fn (x),
n∈N
g3 (x) = lim sup fn (x),
n→∞
g4 (x) = lim inf fn (x).
n→∞

Then gi (i = 1, 2, 3, 4) are measurable.

Proof. 1. Note that

[
{x : g1 (x) > α} = {x : sup fn (x) > α} = {x : fn (x) > α}. (16)
n∈N n∈N

Since each set in (16) is measurable, it follows that g1 is measurable.

2. Similarly as above,
[
{x : g2 (x) < α} = {x : inf fn (x) < α} = {x : fn (x) < α}. (17)
n∈N
n∈N

22
 3. By the definition of lim sup,
 
g3 (x) = lim sup fm (x) .
n→∞ m≥n

4. Similarly as above,
 
g4 (x) = lim inf fm (x) .
n→∞ m≥n

Theorem 17. Let (fn : n ∈ N) be a sequence of measurable functions. The

functions
min fn and max fn
n=1,...,N n=1,...,N

are measurable.
Tn
Proof.
Tn {minn=1,...,N fn > α} = n=1 {fi > α}, {maxn=1,...,N fn < α} =
{f
n=1 i < α}.

6.3 Measurability of the positive and negative part of f

Definition 27 (Positive and negative part of f ). Let f be measurable. The
positive part f + and the negative part f − of f are nonnegative functions
defined by

f + (x) = (f ∨ 0)(x) = max{f (x), 0}, (18)

−
f (x) = −(f ∧ 0)(x) = −min{f (x), 0}. (19)

Remark 10. Note that f (x) = f + (x) − f − (x).

Proposition 3. Let f be measurable.

(a) f + and f − are measurable.
(b) The limit of a convergent sequence of measurable functions is mea-
surable.

Proof. 1. Consider the sequence fn (x) = (f1 (x) = f (x), 0, 0, . . .). By Theo-
rem 17,
max fn (x) = max{f (x), 0} = f +
n=1,...,N

is measurable. Likewise for f − .

2. If (fn : n ∈ N) converges, lim and lim sup are equal.

23
6.4 Measurability of the restriction and extension of f
Lemma 6. Let (X, A) be a measurable space.
(a) If f is an extended real-valued measurable function on a set D ∈ A,
then for every D0 ⊂ D such that D0 ∈ A, the restriction of f to D0 is a
measurable function on D0 .
(b) Let (Dn : n ∈ N) be a sequence in A and let D = ∪n∈N Dn . Let
f be an extended real-valued function on D. If the restriction of f to Dn is
A-measurable on Dn for every n ∈ N, then f is A-measurable on D.

Proof. 1. {D0 : f ≤ α} = {D : f ≤ α} ∩ D0 ∈ A.
2. {D : f ≤ α} = {∪n∈N Dn : f ≤ α} = ∪n∈N {Dn : f ≤ α} ∈ A.

Proposition 4. Let (X, A, µ) be a complete measure space.

(a) Every extended real-valued function f defined on a null set N is A-
measurable on N.
(b) Let f and g be two extended real-valued functions defined on a set
D ∈ A such that f = g a.e. on D. If f is A-measurable on D then so is g.

Proof. Let us prove (b). Suppose f = g a.e. on D. Then there exists a null
set N such that N ⊂ D and f = g on D \ N . Since f is measurable on D,
it is measurable on the subset D \ N by (a) of Lemma 6. Since f = g on
D \ N , g is measurable on D \ N . But since N is a null set and the measure
space is complete, g is measurable on N by (a). Thus g is measurable on
D \ N and on N and therefore measurable on (D \ N ) ∪ N = D according
to (b) of Lemma 6.

6.5 Almost everywhere

If a property P holds for every x ∈ D \ A, where A is a null set, it is
customary to say that P holds almost everywhere on D.

Definition 28 (Real-valued a.e.). Let (X, A, µ) be a measure space. Let f

be an extended real-valued A-measurable function on a set D ∈ A. We say
that f is real-valued a.e. on D if there exists a null set (X, A, µ) such that
N ⊂ D and f (x) ∈ R for every x ∈ D \ N .

Definition 29 (Existence and convergence of limit a.e.). Let (X, A, µ) be

a measure space. Let (fn : n ∈ N) be a sequence of extended real-valued
A-measurable functions on a set D ∈ A. We say that limn→∞ fn exists a.e.
(brief for almost everywhere) on D if there exists a null set (X, A, µ) such
that N ⊂ D and limn→∞ fn (x) exists for every x ∈ D \ N . We say that
(fn : n ∈ N) converges a.e. limn→∞ fn exists and limn→∞ fn ∈ R for every
x ∈ D \ N.

Definition 30 (Uniform convergence a.e.). Let (fn : n ∈ N) be a sequence

of extended real-valued measurable functions on a set D ∈ A and let f be a

24
real-valued measurable function on D. We say that (fn : n ∈ N) converges
almost uniformly on D to f if for every δ > 0 there exists a measurable subset
E of D such that µ(E) < δ and (fn : n ∈ N) converges uniformly on D \ E
to f .

Theorem 18 (D.E. Egoroff). Let D ∈ A and µ(D) < ∞. Let (fn : n ∈ N)

be a sequence of extended real-valued measurable functions on D and let f be
a real-valued measurable function on D. If (fn : n ∈ N) converges to f a.e.
on D, then (fn : n ∈ N) converges to f almost uniformly on D.

7 The Lebesgue Integral

7.1 Simple functions and approximations
Definition 31 (Simple function). Let ϕ be a real-valued function defined on
X. If the range of ϕ is finite, we say that ϕ is a simple function.

Definition 32 (Canonical representation of a simple function). Let ϕ be a

simple function on a set D ⊂ X. Let {ai : i = 1, . . . , n} be the set of distinct
values assumed by ϕ on D and let Di = {x ∈ D : ϕ(x) = ai } for i = 1, . . . , n.
{Di : i = 1, . . . , n} is a disjoint collection and ∪ni=1 Di = D. The expression
n
X
ϕ(x) = ai 1Di (x) for x ∈ D,
i=1

is called the canonical representation of ϕ.

The following important theorem shows that any measurable function

can be approximated from below by a sequence of simple functions. 2

Theorem 19. Let f : X → [0, ∞] be measurable. There exists a sequence

of real-valued simple functions s1 , s2 , . . . on X such that 0 ≤ s1 ≤ s2 ≤ . . . f ,
sn (x) → f (x) pointwise.

Proof. Let us construct a sequence of simple functions ϕ1 , ϕ2 , . . . that con-

verges to the identity from below. Then sn = ϕ ◦ f is a sequence of simple
functions that converges to f from below.
Step 1. Let b c denote the floor function. Define ϕ : [0, ∞] → [0, ∞) by
(
b2n tc/2n , if t ∈ [0, n),
ϕn (t) =
n, if t ∈ [n, ∞].

Note that ϕn is simple and therefore measurable. Let us show ϕn ≤ ϕn+1 .

Let n ∈ N and make a case study.
2
Michael J. Fairchild, www.mikef.org/files/SimpleApproximation.pdf, retrieved 2016-
03-17.

25
 (i ) If t ∈ [0, n), then
ϕn (t) = b2n tc/2n = 2b2n tc/2n+1 ≤ b2n+1 tc/2n+1 = ϕn+1 (t),
where we have used that nbtc ≤ bntc. (This can be proven by induction and
using the fact that if a < b then bac < bbc. Idea proof: bac < a < b, and
bbc < b. Thus bac and bbc are two integers not exceeding b, and so by the
definition of floor, bac < bbc.)
(ii ) If t ∈ [n, n + 1), then
ϕn (t) = n ≤ btc = 2n+1 btc/2n+1 ≤ b2n+1 tc/2n+1 ≤ ϕn+1 (t).
(iii ) If t ∈ [n + 1, ∞], then
ϕn (t) = n < n + 1 = ϕn+1 (t).
This shows that for all t ∈ [0, ∞] we have 0 ≤ ϕ1 (t) ≤ ϕ2 (t) ≤ . . . ≤ t.
We show next that φn (t) → t for t ∈ [0, ∞]. If t ∈ [0, ∞), then 2n t − 1 <
n n
< b22ntc , implying t − 21n < b22ntc . If n > t, then
n
b2n tc and therefore 2 2t−1
n

t − 21n < ϕn (t) ≤ t. This implies ϕn (t) → t for t ∈ [0, ∞). Lastly, if t = ∞
then ϕn (t) = ∞. Thus ϕn (t) → t for t ∈ [0, ∞].
Step 2. sn : X → [0, ∞) defined by sn = ϕ◦f is simple and measurable,
and 0 ≤ sn (x) = ϕn (f (x)) ≤ ϕn+1 (f (x)) = sn+1 (x). Moreover sn (x) =
ϕn (f (x)) ≤ f (x), since f (x) ∈ [0, ∞], proving the first part of the theorem.
Lastly, sn (x) = ϕn (f (x)) → f (x).

7.2 Integration of simple functions

Definition 33 (Lebesgue integral of a simple function). Let ϕ = ni=1 ai 1Di
P
be the canonical representation of a simple function on a set D ∈ A. The
Lebesgue integral of ϕ on D with respect to µ is defined by
Z Xn
ϕ(x) µ(dx) = ai µ(Di ),
D i=1

or, briefly,
Z n
X
ϕ dµ = ai µ(Di ).
D i=1

7.3 Integration of nonnegative functions

Definition 34 (Lebesgue integral of a nonnegative function). Let f be mea-
surable and nonnegative on a set D. We define the Lebesgue integral on D
with respect to µ by
Z Z
f dµ = sup{ ϕ dµ : ϕ ∈ Φ},
D D
where Φ is the collection of all simple functions ϕ such that 0 ≤ ϕ ≤ f .

26
7.4 Integration of measurable functions
Definition 35 (LebesgueR integral). Let R f be measurable on a set D. If at
least one of the integrals D f + dµ and D f − dµ are finite, we define
Z Z Z
f dµ = +
f dµ − f − dµ.
D D D

If both integrals are finite, we say that f is integrable on D in the Lebesgue

sense, with respect to µ, and write f ∈ L(µ). (Note that µ need not be the
Lebesgue measure.)

7.5 Properties of the Lebesgue integral

Remark 11. Let f be a function.

(a) If f is measurable and bounded on D, and if µ(D) < ∞, then f ∈ L(µ)

on D.

(b) If a ≤ f (x) ≤ b for x ∈ D and µ(D) < ∞, then

Z
aµ(D) ≤ f dµ ≤ bµ(D).
D

(c) If f and g ∈ L(µ) on D, and if f (x) ≤ g(x) for x ∈ D, then

Z Z
f dµ ≤ g dµ.
D D

(d) If f ∈ L(µ) on D, then cf ∈ L(µ) on D, for every finite constant c,

and Z Z
c f dµ = cf dµ.
D D

(e) If µ(D) = 0 and f is measurable, then

Z
f dµ = 0.
D

(f ) If f ∈ L(µ) on D, A ∈ A, and A ⊂ D, then f ∈ L(µ) on A.

Proof. Let us prove (c). Firstly, let f1 and f2 be nonnegative and measurable
functions on a set D with f1 ≤ f2 . We claim that
Z Z
f1 dµ ≤ f2 dµ. (20)
D D

27
 Indeed, for i = 1 and 2, let Φi be the collection of all nonnegative simple
functions ϕ on D such that 0 ≤ ϕ ≤ fi . Since 0 ≤ ϕ ≤ f1 ≤ f2 on D, we
have Φ1 ⊂ Φ2 . Then
Z Z
{ ϕ dµ : ϕ ∈ Φ1 } ⊂ { ϕ dµ : ϕ ∈ Φ2 }.
D D

Let
Z
S1 = { ϕ dµ : ϕ ∈ Φ1 },
ZD
S2 = { ϕ dµ : ϕ ∈ Φ2 }.
D

Let U2 be any upper bound to S2 . Since S1 ⊂ S2 , U2 is an upper bound

to S1 too. In particular, then, sup S2 is an upper bound to S1 .
But sup S1 is the smallest upper bound to S1 , so

sup S1 ≤ sup S2 ,

or, Z Z
f1 dµ ≤ f2 dµ.
D D
This proves (20). Note that the result now follows since
Z Z Z Z Z Z
+ − + −
f dµ = f dµ − f dµ ≤ g dµ − g dµ = g dµ.
D D D D D D

Theorem 20. (a) Suppose f is measurable and nonnegative on X. For A ∈

A, define Z
v(A) = f dµ.
A
Then v is countably additive on A.
(b) The same conclusion holds if f is Lebesgue measurable on X.
S
Proof. 1. Let (An : n ∈ N) be a disjoint sequence in A with A = n∈N An .
To prove (a), we have to show that
X
v(A) = v(An ). (21)
n∈N

Suppose f is the characteristic function on a set D ⊂ X. Then

Z
v(A) = f dµ = 1 · µ(A ∩ D) + 0 · µ(A ∩ Dc ) = µ(A ∩ D),
A

and since µ is countably additive, v is too.

28
 Suppose f is simple, and let f = m
P
i=1 ai 1Di be its canonical represen-
tation.
Since {Di ∩A, : i = 1, . . . , m} is a disjoint collection with m
S
i=1 (Di ∩A) =
A, the restriction of ϕ to A is given by
m
X
ϕ= ai 1Di ∩A .
i=1

Therefore
Z m
X m
X X
v(A) = f dµ = ai µ (Di ∩ A) = ai µ(Di ∩ An )
A i=1 i=1 n∈N
XX m XZ X
= [ ai µ(Di ∩ An )] = f dµ = v(An ),
n∈N i=1 n∈N An n∈N

where the fourth equality can be proven by induction.

In the general case, thus, we have that for every measurable simple
function ϕ such that 0 ≤ ϕ ≤ f ,
Z XZ XZ X
ϕ dµ = ϕ(s) dµ ≤ f dµ = v(An ). (22)
A n∈N An n∈N An n∈N

Then by definition of the Lebesgue integral as a supremum,

Z X
v(A) = f dµ ≤ v(An ).
A n∈N

Therefore, in order to establish (21), it remains to be shown that

Z X
v(A) = f dµ ≥ v(An ).
A n∈N

If v(An ) = ∞ for some n ∈ N, then v(A) ≥ ∞

P
n=1 v(An ) because
Z Z
v(A) = f dµ ≥ f dµ = v(An ),
A An
P∞
P that n=1 v(An ) = ∞, also because f is non-
and f is nonnegative. Note
negative. Thus v(A) ≥ ∞ n=1 v(An ) reads ∞ ≥ ∞.
In the case were v(An ) = ∞ for some n ∈ N, (21) is thus established.
Suppose v(An ) < ∞ for every n ∈ N. Given  > 0, we can choose a
measurable function ϕ with 0 ≤ ϕ ≤ f such that
Z Z Z Z
ϕ dµ ≥ f dµ − , ϕ dµ ≤ f dµ − . (23)
A1 A1 A2 A2

29
 Hence
Z Z Z Z
v(A1 ∪ A2 ) = f dµ ≥ ϕ dµ = ϕ dµ + ϕ dµ
A1 ∪A2 A1 ∪A2 A1 A2
≥ v(A1 ) + v(A2 ) − 2,

where we in the second equality used (22). By the arbitrariness of , this

shows that
v(A1 ∪ A2 ) ≥ v(A1 ) + v(A2 ).
From this it follows that for every n ∈ N,

v(A1 ∪ . . . ∪ An ) ≥ v(A1 ) + . . . + v(An ). (24)

Since A ⊃ A1 ∪ . . . ∪ An , (24) implies

X
v(A) ≥ v(An ).
n∈N

2. Let us write
Z Z Z
v(A) = f dµ = f + dµ − f − dµ,
A A A

and see that (b) follows from (a).

Let A = A1 ∪ A2 . We will prove that v(A1 ∪ A2 ) = v(A1 ) + v(A2 ).
(From this the countable additivity follows by induction.) Indeed, if f is
measurable then f + and f − are measurable and nonnegative, so that
Z Z
v(A1 ∪ A2 ) = +
f dµ − f − dµ
A1 ∪A2 A1 ∪A2
Z Z Z Z 
+ + − −
= f dµ + f dµ − f dµ + f dµ
A1 A2 A1 A2
Z Z  Z Z 
+ − + −
= f dµ − f dµ + f dµ − f dµ
A1 A1 A2 A2
= v(A1 ) + v(A2 ).

Corollary 1. v(A) is a measure.

Corollary 2. If A ∈ A, B ⊂ A, and µ(A \ B) = 0, then

Z Z
f dµ = f dµ.
A B

Proof. By Theorem 20 and (e) of Remark 11,

Z Z Z Z Z
f dµ = f dµ = f dµ + f dµ = f dµ.
A B∪(A\B) B A\B B

30
Theorem 21. If f ∈ L(u) on E, then |f | ∈ L(u) on E and
Z Z
f dµ ≤ |f | dµ.
E E
Proof. Write E = A ∪ B, where f (x) ≥ 0 on A and f (x) < 0 on B. Then A
and B partitions E so by (b) of Theorem 20,
Z Z Z Z Z
|f | dµ = |f | dµ + |f | dµ = f + dµ + f − dµ.
E A B A B
But A ⊂ E and A ∈ A, so f ∈ L(u) on A by (f) of Remark 11, implying
that Z
f + dµ < ∞.
A
Similarly, Z
f − dµ < ∞.
B
This shows that |f | ∈ L(µ).
For the last part, note that since f ≤ |f | and −f ≤ |f |, for all x ∈ E,
Z Z Z Z Z
f dµ ≤ |f | dµ, − f dµ = (−f ) dµ ≤ |f | dµ,
E E E E E
where we have used (c) and (d) of Remark 11.
Theorem 22. Suppose f is measurable on E, |f | ≤ g, and g ∈ L(µ) on E.
Then f ∈ L(µ) on E.
Proof. Note that |f | ≤ g on E implies f + ≤ g and f − ≤ g on E. Then
Z Z
f + dµ < g dµ < ∞,
E E
and Z Z
f − dµ < g dµ < ∞,
E E
implying that f ∈ L(µ) on E.

7.6 Convergence theorems

Theorem 23 (Lebesgue’s monotone convergence theorem). Suppose E ∈ A.
Let (fn : n ∈ N) be a sequence of measurable functions such that
0 ≤ f1 (x) ≤ f2 (x) ≤ . . . , x ∈ E. (25)
Let f be defined by
f (x) = lim fn (x), x ∈ E.
n→∞
Then Z Z
lim fn dµ = f dµ.
n→∞ E E

31
 In particular, if fn ∈ L(µ) is nondecreasing and fn → f pointwise, then
f ∈ L(µ). As we will se, this is not true in the Riemann sense.

Proof. Since (fn : n ∈ N) is an increasing sequence of nonnegative functions,

lim fn (x)
n→∞

exists in [0, ∞) for every x ∈ E and thus f is measurable by Theorem 16.

Since fn ≤ f on E, we have
Z Z
fn dµ ≤ f dµ
E E

by (c) of Remark 11. Also fn ≤ fn+1 implies

Z Z
fn dµ ≤ fn+1 dµ,
E E
R
and
R thus ( E fn dµ : n ∈ N) is an increasing sequence bounded above by
E f dµ. Consequently,
Z Z
lim fn dµ ≤ f dµ. (26)
n→∞ E E

Choose c such that 0 < c < 1. Let ϕ be a simple measurable function

such that 0 ≤ ϕ ≤ f . Put

En = {x ∈ E : fn (x) ≥ cϕ(x)}, n ∈ N.

Since (fn : n ∈ N) is an increasing sequence, fn (x) ≥ cs(x) implies

fn+1 (x) ≥ cϕ(x), and therefore (En : n ∈ N) is an increasing sequence.
Since En ⊂ E for every n ∈ N, we have
[
En ⊂ E. (27)
n∈N

In fact, the reverse inclusion of (27) also holds, that is,

[
En ⊃ E. (28)
n∈N

To prove this, let us show that if x ∈ E then x ∈ En for some n ∈ N. Let

therefore x ∈ E. If f (x) = 0, then since ϕ ≤ f we have ϕ(x) = 0 and then
trivially x ∈ En for some n ∈ N since fn (x) ≥ cϕ(x) = 0 and fn (x) ≥ 0.
Suppose f (x) 6= 0. Since 0 ≤ fn ≤ f and 0 < c < 1, we have f (x) >
cϕ(x). Since fn (x) ↑ f (x), there exists N ∈ N, depending on x, such that
fN (x) ≥ cϕ(x). Thus x ∈ EN , so that (28) is established. This implies that
[
En = E. (29)
n∈N

32
 Let us define a set function v on A by setting
Z
v(A) = ϕ dµ,
A

for every A ∈ A. By Theorem 20, v is countably additive and thus a measure

on X.
For every n ∈ N, since E ⊃ En and fn ≥ cϕ on En , we have
Z Z Z
fn dµ ≥ fn dµ ≥ c ϕ dµ = cv(En ). (30)
E En En

Since (En : n ∈ N) is an increasing sequence, we have

[
lim En = En = E,
n→∞
n∈N

where the last equality comes from (29). Then by Theorem 2, we have

lim v(En ) = v( lim En ) = v(E).

n→∞ n→∞

Letting n → ∞ in (30), we then see that

Z
lim fn dµ ≥ c lim v(En ) = cv(E).
n→∞ E n→∞

Since this holds for every 0 < c < 1, letting c → 1, we get

Z Z
lim fn dµ ≥ v(E) = ϕ dµ. (31)
n→∞ E E

Since (31) holds for an arbitrary nonnegative simple function ϕ on E

such that 0 ≤ ϕ ≤ f , and since f is nonnegative, we have
Z Z Z
lim fn dµ ≥ sup ϕ dµ = f dµ.
n→∞ E 0≤ϕ≤f E E

Combining this inequality with that of (26), we are done.

Theorem 24. Suppose f = f1 + f2 where fi ∈ L(µ) on E (i = 1, 2). Then

f ∈ L(µ) on E, and
Z Z Z
f dµ = f1 dµ + f2 dµ.
E E E

Proof. Suppose first that f1 ≥ 0, f2 ≥ 0. If f1 and f2 are simple, the result

follows immediately. Otherwise, choose monotonically increasing sequences
0 00
(sn ), (sn ) of nonnegative measurable simple functions which converge to f1 ,

33
 0 00
f2 . By Theorem 19, this is possible. Put sn = sn + sn . Then by Theorem
20 Z Z Z
0 00
sn dµ = sn dµ + sn dµ,
E E E
and letting n → ∞ and applying Theorem 23 the result follows.
Next, suppose f1 ≥ 0, f2 ≤ 0. Put

A = {x : f (x) ≥ 0}, B = {x : f (x) < 0}.

Then f1 and −f2 are nonnegative on A. Hence

Z Z Z Z Z Z
f1 dµ = (f + (−f2 )) dµ = f dµ + (−f2 ) dµ = f dµ − f2 dµ,
A A A A A A
(32)

where we have used Theorem 20.

Similarly, −f , f1 and −f2 are nonnegative on B, so that
Z Z Z
(−f2 ) dµ = f1 dµ + (−f ) dµ,
B B B

or,
Z Z Z
f1 dµ = f dµ − f2 dµ. (33)
B B B

Adding (32) and (33), we get

Z Z Z Z Z Z
f1 dµ + f1 dµ = f dµ − f2 dµ + f dµ − f2 dµ,
A B A A B B

or, again using Theorem 20,

Z Z Z
f1 dµ = f dµ − f2 dµ.
E E E

In the general case, E can be decomposed into four pairwise disjoint sets
Ei where f1 (x) and f2 (x) have constant sign, and we proceed as above.

Corollary 3. Suppose E ∈ M. If (fn : n ∈ N) is a sequence of nonnegative

measurable functions and
∞
X
f (x) = fn (x), (x ∈ E),
n=1

then Z ∞ Z
X
f dµ = fn dµ.
E n=1 E

34
Proof. By Theorem 24, for any finite n ∈ N, we have
Z X n Xn Z
fj (x) dµ = fj (x) dµ.
E j=1 j=1 E

Since ( nj=1 fj : j ∈ N) is a monotonically increasing sequence, Theorem

P
23 yields
Xn Z Z Xn
lim fj (x) dµ = lim fj (x) dµ.
n→∞ E E n→∞ j=1
j=1

Theorem 25 (Fatou’s theorem). Suppose E ∈ M. If (fn : n ∈ N) is a

sequence of nonnegative measurable functions and

f (x) = inf fn (x), (x ∈ E),

n→∞

then Z Z
f dµ ≤ inf fn dµ.
E n→∞ E

Proof. By definition,
 
inf fn = lim inf fk .
n→∞ n→∞ k≥n

Since (inf k≥n fk : n ∈ N) is a increasing sequence of nonnegative func-

tions, Theorem 23 may be applied and
Z Z  
inf fn dµ = lim inf fk dµ
E n→∞ E n→∞ k≥n
Z  
= lim inf fk dµ
n→∞ E k≥n
Z
≤ inf fn dµ.
E n→∞

We have now arrived at the most important theorem in this paper.

Theorem 26 (Lebesgue’s dominated convergence theorem). Suppose E ∈
A. Let (fn : n ∈ N) be a sequence of measurable functions such that

fn (x) → f (x), (x ∈ E)

as n → ∞. If there exists a function g ∈ L(µ) on E such that

|fn (x)| ≤ g(x) (n = 1, 2, 3, . . . , x ∈ E),

then Z Z
lim fn dµ = f dµ.
n→∞ E E

35
Proof. Since fn + g ≥ 0, Fatou’s theorem shows that
Z Z
(f + g) dµ ≤ lim inf (fn + g) dµ,
E n→∞ E
or, Z Z
f dµ ≤ lim inf fn dµ,
E n→∞ E
Similarly g − fn ≥ 0, so by Fatou’s,
Z Z
(g − f ) dµ ≤ lim inf (g − fn ) dµ,
E n→∞ E
or, Z  Z 
− f dµ ≤ lim inf − fn dµ ,
E n→∞ E
or Z Z
f dµ ≥ lim sup fn dµ.
E n→∞ E
Thus Z Z Z Z
f dµ ≥ lim sup fn dµ ≥ lim inf fn dµ ≥ f dµ.
E n→∞ E n→∞ E E

Theorem 27 (Bounded convergence theorem). Let (fn : n ∈ N) be a

uniformly bounded sequence of real-valued measurable functions a on a set
D ∈ A with µ(D) < ∞. Let f a bounded real-valued measurable function on
D. If (fn : n ∈ N) converges to f a.e. on D, then
Z
lim |fn − f | dµ = 0, (34)
n→∞ D

and in particular
Z Z
lim fn dµ = f dµ. (35)
n→∞ D D

Proof. Since (fn : n ∈ N) is uniformly bounded on D, there exists M > 0

such that |fn (x)| ≤ M for all x ∈ D. Since f also is bounded on D, we may
choose M so that |f (x)| ≤ M for all x ∈ D. Since (fn : n ∈ N) converges to
f a.e. on D and µ(D) < ∞, we may apply Theorem 18. According to this
theorem, for every δ > 0 there exists a measurable subset E of D such that
µ(E) < δ and such that (fn : n ∈ N) converges to f uniformly on D \ E,
that is, for every  > 0 there exists N ∈ N such that |fn (x) − f (x)| <  for
all x ∈ D \ E whenever n ≥ N . Then
Z Z Z Z Z
|fn − f | dµ = |fn − f | dµ + |fn − f | dµ =  dµ + 2M dµ
D D\E E D\E E
≤  µ(D \ E) + 2M µ(E) <  µ(D) + 2M δ.

36
Since this holds for all n ≥ N , we have
Z
sup |fn − f | dµ ≤  µ(D) + 2M δ
n≥N D

which implies
Z
lim sup |fn − f | dµ ≤  µ(D) + 2M δ,
n→∞ m≥n D

or,
Z
lim sup |fn − f | dµ ≤  µ(D) + 2M δ.
n→∞ D

By the arbitrariness of of  > 0 and δ > 0, this implies

Z
lim sup |fn − f | dµ = 0.
n→∞ D
R
Now since D |fn − f | dµ ≥ 0 for every n ∈ N, we have
Z
lim inf |fn − f | dµ ≥ 0.
n→∞ D

Then Z Z
0 ≤ lim inf |fn − f | dµ ≤ lim sup |fn − f | dµ = 0,
n→∞ D n→∞ D
showing (60).
Let us note that (61) can be derived from (60). Indeed,
Z Z Z Z
| fn dµ − f dµ| = | (fn − f ) dµ| ≤ |fn − f | dµ. (36)
D D D D

The equalities in (36) may need justification. For the first one, note
that since fn and f are measurable and bounded, and µ(D) < ∞, we have
fn , f ∈ L(µ) by Remark 11 (a). Then −fn ∈ L(µ) by (d) of the same
Remark. We then use Theorem 24. For the last equality, we use Theorem
21.
Applying (60) to the right most member of (36), we get
Z Z
lim | fn dµ − f dµ| = 0.
n→∞ D D

This shows (61).

Lemma 7. Let (X, A, µ) be a measure space. Let f and g be bounded real-

valued measurable functions on Ra set D ∈ A with µ(D) < ∞.
(a) If f ≥ 0 a.e. on D and R D f dµ = R0, then f = 0 a.e. on D.
(b) If f ≤ g a.e. on D and D f dµ = D g dµ, then f = g a.e. on D.

37
Proof. 1. Suppose f ≥ 0 on D. Let

D0 = {x ∈ D : f (x) = 0} and D1 = {x ∈ D : f (x) > 0}.

Let us show that

f = 0 a.e. on D ⇔ µ(D1 ) = 0. (37)

Suppose f = 0 a.e. on D. Then there exists a null set E such that E ⊂ D

and f = 0 on D \ E. Then D \ E ⊂ D0 = D \ D1 , so that D1 ⊂ E and
µ(D1 ) = 0.
Conversely suppose µ(D1 ) = 0, then D1 ⊂ D is a null set and f = 0 on
D0 = D \ D1 , so that f = 0 a.e. on D. Thus (37) is established.
R
Before proceeding, let us note that if µ(D) = 0, then D f dµ = 0 and
µ(D1 ) = 0 so that f = 0 a.e. on D by (37). So in order to show (a) may
assume µ(D) > 0.
To show that f = 0 a.e. on D, assume the contrary. By (37) this implies
µ(D1 ) > 0. Now
[ 1
D1 = {D : f > 0} = {D : f ≥ },
k
k∈N

so that X 1
0 < µ(D1 ) ≤ µ{D : f ≥ }.
k
k∈N

This implies that there exists k0 ∈ N such that

1
µ{D : f ≥ } > 0.
k0
Let us define a simple function ϕ on D by
(
1
for x ∈ {D : f ≥ k10 },
ϕ(x) = k0
0 for x ∈ D \ {D : f ≥ k10 }.

Then clearly ϕ ≤ f on D and

Z Z
1 1
f dµ ≥ ϕ dµ = µ{D : f ≥ } > 0;
D D k0 k0
R
a contradiction to the assumption that D f dµ = 0. Therefore f = 0 a.e.
on D. R
Consider finally the general case where f ≥ 0 a.e. on D and D f dµ = 0.
Then there exists a null set E such that E ⊂ D and f ≥ 0 on D \ E, and
Z Z Z Z
0= f dµ = f dµ + f dµ = f dµ.
D E D\E D\E

38
 R
This shows that f ≥ 0 on D \ E and D\E f dµ = 0. But then f = 0 a.e.
on D \ E by our result above. Thus there exists a null set F such that
F ⊂ D \ E and such that f = 0 on (D \ E) \ F = D \ (E ∪ F ). Thus f = 0
a.e. on D. R R
2.R If f ≤ g a.e. on D, then g−f ≥ 0 a.e. on D. If also D f dµ = D g dµ,
then D (g − f ) dµ = 0. Then by (a), g − f = 0 a.e. on D.

8 Comparison with the Riemann integral

8.1 Riemann integrability
Definition 36 (Darboux sums, Darboux integrals). Let f : [a, b] → R be a
bounded function. Suppose |f (x)| ≤ M for x ∈ I = [a, b] for some M ≥ 0.
Let B be the collection of all partitions of I. With P ∈ B given by P =
{x0 , . . . , xn }, let Ik = [xk−1 , xk ], mk = inf x∈Ik f (x) and Mk = supx∈Ik f (x)
for k = 1, . . . , n. The lower and upper Darboux sums of f corresponding to
the partition P are defined by
n
X n
X
S(f, P) = mk l(Ik ) and S(f, P) = Mk l(Ik ).
k=1 k=1

Let
S(f ) = sup S(f, P) and S(f ) = inf S(f, P).
P∈B P∈B

We call S(f ) and S(f ) the lower and upper Darboux integrals of f on I.
Definition 37. A bounded real-valued function f on I = [a, b] ⊂ R is Rie-
mann integrable on I if and only if
S(f ) = S(f ).
In this case we set Z b
f (x) dx := S(f ) = S(f ).
a
Theorem 28. If f is a continuous real-valued function on I = [a, b], then
S(f ) = S(f ) and consequently f is Riemann integrable on I.
Proof. If f is continuous on the compact set I, it is uniformly continuous
on I. Thus for every  > 0 there exists δ > 0 such that for all x, x00 ∈ [a, b],

|x − x00 | ≤ δ ⇒ |f (x0 ) − f (x00 )| ≤ . (38)
b−a
Let P0 ∈ B be such that |P0 | < δ. Let P0 = {x0 , . . . , xn } where a =
x0 < . . . < xn = b. In preceding, let us note that (38) implies

sup f − inf f ≤ . (39)
[xk−1 ,xk ] [xk−1 ,xk ] b−a

39
 Indeed, let α, β ∈ [xk−1 , xk ]. By (38),

f (α) − f (β) ≤ ,
b−a
or,

f (α) ≤ + f (β).
b−a
Then
f 
[xk , xk−1 ] ≤ + f (β). (40)
b−a
Thus the right hand side of (40) is an upper bound to [xk , xk−1 ] f . But
sup[xk ,xk−1 ] f is the smallest upper bound to [xk , xk−1 ] f , so

sup f≤ + f (β). (41)
[xk ,xk−1 ] b−a

Rewriting (41), we have


sup f− ≤ f (β). (42)
[xk ,xk−1 ] b−a

Thus the left hand side of (42) is a lower bound to [xk , xk−1 ] f , and since
inf[xk , xk−1 ] f is the largest such lower bound,

sup f− ≤ inf f,
[xk ,xk−1 ] b − a [xk ,xk−1 ]

establishing (39). Then

n
!
X
S(f, P0 ) − S(f, P0 ) = sup f− inf f (xk − xk−1 ) (43)
[xk−1 ,xk ] [xk−1 ,xk ]
k=1

≤ (b − a) = . (44)
b−a
By a similar argument as above, we see that (43) implies

inf S(f, P) − sup S(f, P) ≤ . (45)

P∈B P∈B

By the arbitrariness of  > 0, (45) implies

inf S(f, P) = S(f ) = S(f ) = sup S(f, P).

P∈B P∈B

This shows that f is Riemann integrable on I by Definition 37.

40
8.2 The upper and lower envelope integrals
Definition 38 (Limit inferior, limit superior of f ). For x0 ∈ R and δ > 0,
let
U (x0 , δ) = (x0 − δ, x0 + δ),
and
U0 (x0 , δ) = (x0 − δ, x0 + δ) \ {x0 }.
Let f be an extended real-valued function on a set D ∈ R and let x0 ∈ D.
The limit inferior and limit superior of f as x approaches x0 are defined by

lim inf f (x) = lim inf f and lim sup f (x) = lim sup f.
x→x0 δ→0 U0 (x0 ,δ)∩D x→x0 δ→0 U0 (x0 ,δ)∩D

We have

f (x0 ) ∈ R,
f is continuous at x0 ⇔
lim inf f (x) = lim sup f (x) = f (x0 ).
x→x0 x→x0

Definition 39 (Lower and upper envelope). Let f be an extended real-valued

function on a set D ∈ R. Let x0 ∈ D. We define the lower envelope and
upper envelope of f at x0 by

f∗ (x0 ) = lim inf f and f ∗ (x0 ) = lim sup f.

δ→0 U (x0 ,δ)∩D δ→0 U0 (x0 ,δ)∩D

Remark 12. For an extended real-valued function f on a set D ⊂ R and

for x0 ∈ D,

f∗ (x0 ) ≤ lim inf f (x) ≤ lim sup f (x) ≤ f ∗ (x0 ). (46)

x→x0 x→x0

f∗ (x0 ) ≤ f (x0 ) ≤ f ∗ (x0 ). (47)

(
f (x0 ) ∈ R,
f is continuous at x0 ⇔ (48)
f∗ (x0 ) = f ∗ (x0 ).

Proof. Let us prove (48), since (46) and (47) follows directly from the defi-
nition.
Suppose f is continuous at x0 ∈ D. Then for every  > 0 there exists
δ > 0 such that for all x ∈ U (x0 , δ) ∩ D,

f (x0 ) −  < f (x) < f (x0 ) + .

Note that the first of these inequalities implies

f (x0 ) −  ≤ inf f (x).

U (x0 ,δ)∩D

41
 Now since inf U (x0 ,δ)∩D f (x) ↑ as δ ↓ 0 (as the infimum is taken over a
smaller set), we have

f (x0 ) −  ≤ inf f (x) ≤ lim inf f (x) ≤ f∗ (x0 ). (49)

U (x0 ,δ)∩D δ→0 U (x0 ,δ)∩D

By a similar argument we show

f ∗ (x0 ) ≤ f (x0 ) + . (50)

If we then combine (49) and (50), we get f ∗ (x0 ) − f∗ (x0 ) ≤ 2. Since
 > 0 was chosen arbitrarily, we have f∗ (x0 ) = f ∗ (x0 ).
Conversely, if f (x0 ) ∈ R and f∗ (x0 ) = f ∗ (x0 ), then

f∗ (x0 ) = f ∗ (x0 ) = f (x0 )

by (47). Then by (46),

lim inf f (x) = lim sup f (x) = f (x0 ),

x→x0 x→x0

proving the continuity of f at x0 .

RLemma 8. Let f beR a bounded real-valued function on I = [a, b]. Then

∗ dµ = S(f ).
I f∗ dµ = S(f ) and I f
R ∗ R
Proof. Let us prove I f dµL = S(f ). (The equality I f∗ dµL = S(f ) is
proved similarly.) Consider the upper Darboux integral of f on I,

S(f ) = inf S(f, P).

P ∈B

By how infimum is defined, for every m ∈ N, there exists Pm ∈ B such that

1
S(f ) ≤ S(f, Pm ) < S(f ) + . (51)
m
For the sequence of partitions (Pm : m ∈ N), we then have

lim S(f ) = S(f, Pm ). (52)

n→∞

1
If necessary, let us add partition points to Pm so that |Pm | ≤ m . Note that
this not affect the validity of (51) and (52), since refining a partition does
not increase the upper Darboux sum.
Let Pm = {xm1 , xm2 , . . . , xmn } and let Imk = [xmk−1 , xmk ] for k =
1, . . . , n. For each m ∈ N, define a function ψm on I by
n
X
ψm = sup f · 1Imk . (53)
k=1 Imk

42
 Let E be the countable set consisting of all partition points in (Pm : m ∈
N), that is,
(Pm : m ∈ N) = (x11 , x22 , . . .).
Let us show that

lim ψm (x) = f ∗ (x), for x ∈ I \ E. (54)

m→∞

Let x0 ∈ I \ E. Then for each m ∈ N there exists a subinterval in Pm

which contain x0 in the open interval between two partitions points, that is,
for each m ∈ N there exists k ∈ N, 1 ≤ k ≤ mk such that x0 ∈ Im ◦ .
k
Pick a δ 0 > 0 so small that for all m ≥ M 0 , where M 0 ∈ N, we have
U (x0 , δ 0 ) ⊂ Im
◦ . Then
k

f ∗ (x0 ) = lim sup f ≤ sup f ≤ sup f ≤ sup f = ψm (x0 ), (55)

δ→0 U (x0 ,δ) U (x0 ,δ 0 ) ◦
Im Im k
k

where each inequality comes from the fact that we take the sup over succes-
sively greater sets, and the last equality from how we defined ψm .
Since supU (x0 ,δ)∩I f ”shrinks” from above to f ∗ (x0 ) as δ ↓ 0, we have
that for an arbitrary  > 0 there exists δ > 0 such that

| sup f − f ∗ (x0 )| = sup f − f ∗ (x0 ) < .

U (x0 ,δ)∩I U (x0 ,δ)∩I

By the same reason,

sup f ≥ f ∗ (x0 ).
U (x0 ,δ)∩I

Combining these two inequalities, we get

f ∗ (x0 ) ≤ sup f < f ∗ (x0 ) + . (56)

U (x0 ,δ)∩I

Since |Pm | → 0 as m → ∞ and since x0 ∈ Im◦ for some k ∈ N for every

k
m ∈ N, there exists M > 0 such that for m ≥ M ,

Imk ⊂ U (x0 , δ) ∩ I. (57)

Let N = max(M, M 0 ), then by (55), (57), and (56), we have that for all
m≥N
f ∗ (x0 ) ≤ ψm (x0 ) = sup f ≤ sup f < f ∗ (x0 ) + ,
Im k U (x0 ,δ)∩I

proving (54).
Consider the sequence of simple functions (ψm : m ∈ N) defined by (53).
Then Z X n
ψm dµ= sup f · l(Imk ) = S(f, Pm );
I k=1 Imk

43
the upper Darboux sum of f corresponding to Pm . Then by (52),
Z
lim ψm dµ = S(f ). (58)
n→∞ I

Since f is bounded on I, we have |f (x)| ≤ M on I for some M ≥ 0.

Then by (53), we have |ψm (x)| ≤ M . By (54), we have that ψm converges
to f ∗ a.e. on I. Then by Theorem 32,
Z Z
lim ψm dµ = f ∗ dµ. (59)
n→∞ I I

By combining (58) and (59), we have

Z
f ∗ dµ = S(f ).
I

Lemma 9. Let (X, A, µ) be a measure space. Let f1 and f2 be bounded

real-valued measurable functions on a set D ∈ A with µ(D) < ∞. If f1 = f2
a.e. on D, then Z Z
f1 , dµ = f2 dµ.
D D

8.3 Riemann integrability and Lebesgue integrability

Theorem 29. Let f be a bounded real-valued function on I = [a, b]. If f is
Riemann integrable on I, then f is ML -measurable and Lebesgue integrable
on I and moreover Z Z b
f (x) dx = f dµ.
a I

Proof. If f is Riemann integrable on I, then S(f ) = S(f ) by Definition 37.

By Lemma 8, Z Z
f∗ dµ = f ∗ dµ.
I I
Since f∗ and f∗
are BR -measurable on I, they are ML -measurable on I
by Theorem 8. Since f∗ ≤ f ∗ and their Lebesgue integrals are equal, f∗ = f ∗
a.e. on I by (b) of Lemma 7. Since f∗ ≤ f ≤ f ∗ by Remark 12, we have
f = f∗ = f ∗ a.e. on I. Since the Lebesgue measure space is complete, f is
Lebesgue measurable on I by (b) of Remark 4.
f is Lebesgue measurable and f = f ∗ a.e. on I, we have I f dµ =
R
R Since
∗
I f dµ by Lemma 9. But by Lemma 8 and Definition 37,
Z Z b
∗
f dµ = S(f ) = f (x) dx.
I a
Rb R
Thus a f (x) dx = I f dµ.

44
8.4 Lebesgue’s integrability condition
Theorem 30. Let f be bounded real-valued function on I = [a, b] and let E
be the set of all points of discontinuity of f in I. Then

f is Riemann integrable on I
⇔
∗
f∗ = f , a.e. on I
⇔
µL (E) = 0.

Proof. 1. If f is Riemann integrable on I, f∗ = f ∗ a.e. as seen above.

If f∗ = f ∗ a.e. on I, then by Lemma 9,
Z Z
f∗ dµ = f ∗ dµ,
I I

implying Riemann integrability by Definition 37 and Theorem 8.

2. By Remark 12 we have that f is continuous at x0 ∈ I if and only if
f∗ (x0 ) = f ∗ (x0 ). Then f ∗ = f∗ on I \ E 0 for some null set E 0 if and only if f
is continuous on I \ E 0 for some set null E 0 if and only if f is discontinuous
at E 0 for some set null E 0 if and only if E 0 = E.

8.5 Where Riemann falls short

In a Riemann setting, uniform convergence tells us that we are allowed to
switch places between the integral and the limit.

Theorem 31 (Uniform Convergence and Integration in the Riemann sense).

Let (fn (x) : n ∈ N) be a sequence of continuous functions defined on the
interval [a, b] and assume that fn converges uniformly to a function f . Then
f is Riemann-integrable and
Z b Z b
lim fn (x) dx = f (x) dx.
n→∞ a a

Proof. Since the fn ’s are continuous and converge uniformly to f , the limit
function must be continuous. By Theorem 28, f is Riemann-integrable and
Z b Z b Z b Z b
fn (x) dx − f (x) dx ≤ fn (x) dx − f (x) dx
a a a a
Z b
≤ |fn (x) − f (x)| dx
a
≤ (b − a) sup |fn (x) − f (x)| → 0.

45
 Note that Theorem 31 does not say ¬(uniform) ⇒ ¬(interchange). How-
ever in practice, to be sure that interchanging is justified it is uniform con-
vergence that we check.
Even when the sequence is uniformly bounded, pointwise convergence is
not enough. Recall the Bounded convergence theorem:
Theorem 32 (Bounded convergence theorem). Let (fn : n ∈ N) be a
uniformly bounded sequence of real-valued measurable functions a on a set
D ∈ A with µ(D) < ∞. Let f a bounded real-valued measurable function on
D. If (fn : n ∈ N) converges to f a.e. on D, then
Z
lim |fn − f | dµ = 0, (60)
n→∞ D

and in particular
Z Z
lim fn dµ = f dµ. (61)
n→∞ D D

The bounded convergence theorem does not hold in a Riemann setting.

We will prove this in Proposition 5, but first we need the following Lemma.
Lemma 10. Let f be the Dirichlet function on [0, 1], that is,
(
1, if x is rational,
f (x) =
0, otherwise,

for all x ∈ [0, 1].

f is not Riemann integrable on [0, 1].
Proof. Take an arbitrary partition P = {x0 , x1 , . . . , xn } of the interval [0, 1].
Between any two points xj and xj+1 there exists an irrational number.
Therefore the inf over [xj , xj+1 ] is 0, so that

S(f ) = sup S(f, P) = 0.

P∈B

Between any two points xj and xj+1 there exists a rational number.
Therefore the sup over [xj , xj+1 ] must be 1. This means that
n
X
S(f, P) = 1 · (xk − xk−1 )
k=1
= (x1 − x0 ) + (x2 − x1 ) + . . . + (xn − xn−1 )
= xn − x0 = 1 − 0 = 1,

so that
S(f ) = inf S(f, P) = 1.
P∈B
Then by Definition 37, f is not Riemann integrable.

46
Proposition 5. There exists a uniformly bounded sequence (fn : n ∈ N)
of real-valued functions defined on the set [0, 1] that converges to a bounded
function f , but where
Z 1 Z 1
lim fn dx 6= lim fn dx. (62)
n→∞ 0 0 n→∞

Proof. Let f be the Dirichlet function on [0, 1]. Then f is clearly bounded.
Let {rn }∞
n=1 be any enumeration of the rationals in [0, 1], and define
(
1, if x = rk , for some 1 ≤ k ≤ n,
fn (x) =
0, otherwise.

Then (fn ) is uniformly bounded by 1, and it converges pointwise to f .

Moreover, Z 1
lim fn (x) dx = 1.
n→∞ 0

However f is not Riemann integrable by Lemma 10, so (62) holds.

8.6 Some examples of how to use the Lebesgue integral

Example 2. Compute Z 1
1
lim dµ.
n→∞ 0 nx + 1
1
Solution. Let fn (x) = nx+1 , then fn (x) → 0 and |fn (x)| ≤ g(x) = 1 for all
x ∈ [0, 1] and for all n ∈ N. Since g is continuous and bounded on [0, 1], it
is Lebesgue integrable on [0, 1] by Theorem 28 and 29.
Therefore the integral is equal to 0. 

Remark 13. Note that in the Riemann sense, we could not so easily be
sure of the equality
Z 1 Z 1
1 1
lim dµ = lim dµ.
n→∞ 0 nx + 1 0 n→∞ nx + 1

One way of being sure, is to know that (fn (x)) converges uniformly to 0
and then use [0, 1] Theorem 31. The problem is, it doesn’t:
1
sup | | = 1 6= 0,
nx + 1
(To see this, note that |fn (x)| ≤ 1, so 1 is an upper bound to |fn (x)|. Now
since fn (0) = 1, it is indeed the smallest upper bound.)

For the next example we will need the following result.

47
 x n


Proposition 6. Let fn (x) = 1 + n , x ∈ [0, ∞).
Then (fn ) is non-decreasing.

Proof. Fix x ∈ [0, ∞) and take x1 = 1, x2 = x3 = . . . = xn+1 = 1 + nx in the

AM-GM inequality
√ x1 + . . . + xn+1
n+1 x1 · . . . · xn+1 ≤ .
n+1
Then n
 x  n+1 x
1+ ≤1+ ,
n n+1
proving fn (x) ≤ fn+1 (x).

Example 3. Compute
Z ∞
x n −2x
lim 1+ e dµ.
n→∞ 0 n

n
Solution. Let fn (x) = 1 + nx e−2x , then each fn is measurable by Theo-
rem 13, and
0 ≤ f1 (x) ≤ f2 (x) ≤ . . . , x ∈ [0, ∞),
n
since 1 + nx → ex , monotonically, from below, by Proposition 6.

By Lebesgue’s monotone convergence, the integral is equal to 1. 

References
[1] Yeh, J. Real Analysis, Theory of Measure and Integration (2nd. ed.),
World Scientific Publishing Co. Pte. Ltd. 2006

[2] Burkill, J.C. The Lebesgue Integral, Cambridge University Press 1971

[3] Rudin, W. Principles of Mathematical Analysis (3rd. ed.), McGraw-

Hill. 1976

48