A Primer on Measure Theory

Pramook Khungurn
January 23, 2022

This is a primer on measure theory and Lebesgue integration. The materials are taken from Bartle’s “The
Elements Of Integration And Lebesgue Measure” [Bartle, 1995], Billingsley’s “Probability and Measure”
[Billingsley, 1995], and Hunter’s note on measure theory [Hunter, 2011].

1 Introduction
ˆ Why do we care about measure theory and Lebesgue integration?

– They expand the class of functions for which integrations are defined compared to what can be
achieved by Riemann integration.
– Theorems relating to the intechange of limits and integrals are valid under less stringent conditions
(again, compared to Reimann integration).
– In particular, the dominated convergence theorem 1 is a very powerful tool. For examples, it can
be use to easily show that
Z ∞ −nx
e
lim √ dx = 0
n→∞ 0 x

and
Z ∞ Z ∞
d
x2 e−tx dx = − x3 e−tx dx.
dx 0 0

– Measure theory is the foundation of modern probability theory, and the dominated convergence
theorem shows up everywhere in it.

ˆ How is Lebesgue integration different from Reimann integration?

– Riemann integrals are defined in terms of approximating a function with constant functions over
intevals.
– An interval is a subset of the real line which is of one of the following forms:

[a, b] = {x ∈ R : a ≤ x ≤ b},
[a, b) = {x ∈ R : a ≤ x < b},
(a, b] = {x ∈ R : a < x ≤ b},
(a, b) = {x ∈ R : a < x < b}.

The real number a and b are said to be the endpoints of the interval, and b − a is the length of
the interval.
1 https://en.wikipedia.org/wiki/Dominated_convergence_theorem

1
 – A step function ϕ is a linear combination of a finite number of characteristic functions of intevals.
n
X
ϕ(x) = cj χEj (x)
i=1

where cj ∈ R and each Ej is an interval with endpoints aj and bj .

– The integral of a step function ϕ is defined to be
Z n
X
ϕ= cj (bj − aj ).
i=1

– If f is a bounded function on [a, b], then the Reimann integral is defined to the limit of the
integrals of step functions that approximate f .
– The lower Rieman integral is defined to be the supremum of integrals of all step functions φ
such that φ(x) ≤ f (x) for all x ∈ [a, b] and φ(x) = 0 for all x 6∈ [a, b].
ˆ The Lebesgue integral is defined similarly, with some differences.

– Intervals are replaced by a larger collection of sets (called measurable sets).

– The notion of “length” is generalize to the notation of measure.
* Here, the measure is a function µ that maps a set of a non-negative real number.
– The step function is replaced by the simple function, which is a finite linear combination of
characteristic functions of measurable sets.
n
X
ϕ(x) = cj χEj (x)
j=1

where each Ej is a measurable set. The integral of φ is defined to be

Z n
X
ϕ= cj µ(Ej ).
j=1

– If f is a non-negative function defined on R, then the Lebesgue integral of f is the supremum

of all simple functions φ such that φ(x) ≤ f (x) for all x ∈ R.
* This notation can later be generalized to functions taking both signs.

ˆ When studying integration, it is convenient to work with the extended real number system R =
R ∪ {−∞, ∞}.

– For any x ∈ R, we have that −∞ < x < ∞.

– We say that the length of the real line is ∞.
– We define the supremum of non-empty set of real numbers which does not have an upper bound
to be ∞, and the infemum of the a non-empty set of real numbers which does not have a lower
bound to be −∞.
* In this way, all non-empty sets of real numbers (or subsets of R) have unique supremums and
infemums.

2
 – The arithematic operations between the infiniites and real numbers are as follows:

(±∞) + (±∞) = x + (±∞) = (±∞) + x = ±∞

(±∞)(±∞) = +∞
(±∞)(∓∞) = −∞

±∞, if x > 0,

(±∞)x = x(±∞) = 0, if x = 0,

∓∞, if x < 0


for any (finite) real number x.

– Note that we do not define (±∞)−(±∞). We also do not define quotients when the denominators
are ±∞.
ˆ If (xn ) is a sequence of extended real numbers, define the limit superior and limit inferior by
   
lim sup xn = lim sup xm = inf sup xm
n→∞ n→∞ m≥n n m≥n
   
lim inf xn = lim inf xm = sup inf xm .
n→∞ n→∞ m≥n n m≥n

If the limit superior and limit inferere of a sequence both exist and are equal, then the limit of the
sequence exists and is equal to that value.

2 Sigma-Algebras and Measures

2.1 Basic Definitions
ˆ Let us denote the power set of set X with P(X).
ˆ A σ-algebra is the domain upon which we define measures. It is a collection of sets with some nice
properties.
Definition 1. A σ-algebra (or a σ-field) on a set X is a collection X ∈ P(X) of subsets of X, called
measurable sets, such that the following properties hold.
1. ∅, X ∈ X .
2. It is closed under complementation: if A ∈ X , then Ac = X − A ∈ X .
S∞
3. It is closed under countable unions: if (An ) is a sequence of sets in X , then n=1 An ∈ X .

ˆ We can show that a σ-algebraSis closed under countableSintersections as well. To see this, we note that
∞ ∞ c c


Acn ∈ X for all n ∈ N, and so i=1 Acn ∈ X As result, A
i=1 n ∈ X . Applying de Morgan’s law, we
T∞ S∞ c
c
have that i=1 An = A
i=1 n ∈ X .
ˆ Definition 2. A measurable space (X, X ) is a non-empty set X equipped with a σ-algebra X on X.
ˆ Definition 3. Let A be a non-empty collection of subsets of X. The σ-algebra generated by A,
denoted by σ(A) is the smallest σ-algebra that contains A. In other words,
\n o
σ(A) = à ⊆ P(X) : A ⊆ à and à is a σ-algebra .

ˆ Definition 4. The Borel algebra is the σ-algebra B generated by all the open intervals (a, b) in R.
Any set in B is called a Borel set.

3
ˆ Observe that we can write any open interval (a, b) as a countable unions of closed intervals:
∞  
[ 1 1
(a, b) = a + ,b −
n n
n≥N

where N is an integer such that b − a − N2 > 0. As a result, B is also generated by the collection of
close intervals [a, b] in R. The same is also true for half-open intervals of the form (a, b] and [a, b).

ˆ Let X be the set R of extended real numbers. If E is a Borel set, then define

E1 = E ∪ {∞}
E2 = E ∪ {−∞}
E3 = E ∪ {−∞, ∞}

Let B the collection of all sets E, E1 , E2 , and E3 as E varies over B. We have that B is a σ-algebra,
and it is called the extended Borel algebra.

ˆ A “measure” encapsulates the notion of length, area, volume, mass, etc. of a set.

Definition 5. Let (X, X ) be a measurable space. A measure is a function µ : X → [0, ∞] with the
following properties.
1. µ(∅) = 0.
2. µ is countably additive. That is, for a sequence (En ) of disjoint sets, it holds that
∞
[  X∞
µ En = µ(En ).
n=1 n=1

ˆ If µ(E) < ∞ for all E ∈ X , we say that µ is finite.

ˆ A probability measure is a finite measure with µ(X) = 1.

S∞
ˆ If there exists a sequence (En ) of sets in X with i=1 En = X and such that µ(En ) < ∞ for all n,
then we say that µ is σ-finite.

ˆ Here is an example of a measure that is σ-finite but not finite. Let X = N, and X = P(N). Define µ(E)
to be the number of elements in E with the convention that µ(E) = ∞ when E is infinite. Obviously,
µ(N) = ∞. However, N = {1} ∪ {2} ∪ · · · , and µ({n}) = 1 for all n ∈ N. The measure µ is called the
counting measure on N.
ˆ Lemma 6. Let µ be a measure defined on a σ-algebra X . Let E, F ∈ X be such that E ⊆ F , then
µ(E) ≤ µ(F ). If µ(E) < ∞, then µ(F − E) = µ(F ) − µ(E).
Proof. Since F = E ∪ (F − E) and E ∩ (F − E) = ∅, it follows that

µ(F ) = µ(E) + µ(F − E).

Because µ(F − E) ≥ 0, it follows that µ(F ) ≥ µ(E). If µ(E) < ∞, we can subtract from both sides of
the equation.

ˆ A sequence of sets (En ) is increasing if En ⊆ En+1 for all n.

ˆ A sequence of sets (En ) is decreasing if En ⊇ En+1 for all n.

4
ˆ Lemma 7. If (En ) is an increasing sequence of measurable sets, then
[ ∞ 
µ En = lim µ(En ).
n→∞
n=1

If (En ) is a decreasing sequence of measurable sets and µ(E1 ) < ∞, then

\∞ 
µ En = lim µ(En ).
n→∞
n=1

Proof. Let (En ) be increasing. Set F0 = E1 , and Fn = En+1 − En for all n ≥ 1. We have that (Fn ) is
a sequence of disjoint sets. So,
[∞  [ ∞  X ∞
µ En = µ Fn = µ(Fn )
n=1 n=0 n=0
Sn
Also because En = i=0 Fi , we have that
n
X
µ(En ) = µ(Fi ).
i=0

So,
∞
[  X∞ n
X
µ En = µ(Fn ) = lim µ(Fi ) = lim µ(En ).
n→∞ n→∞
n=1 n=0 i=0

Next, let (En ) be decreasing and µ(E1 ) < ∞. Let Fn = E1 − En . We have that (Fn ) is increasing and
µ(Fn ) = µ(E1 ) − µ(En ). If follows that
[ ∞ 
µ Fn = lim µ(Fn ) = lim µ(E1 ) − µ(En ) = µ(E1 ) − lim µ(En )
n→∞ n→∞ n→∞
n=1

Now,
∞
\ ∞
[
En = E1 − Fn
n=1 n=1

So,
∞
\  ∞
[   
µ En = µ(E1 ) − µ Fn = µ(E1 ) − µ(E1 ) − lim µ(En ) = lim µ(En )
n→∞ n→∞
n=1 n=1

are required.

ˆ Definition 8. A measure space is a triple (X, X , µ) where X is a non-empty set, X is a σ-algebra

on X, and µ is a measure on X .
ˆ Definition 9. In a measure space (X, X , µ), a set N ∈ X is set to be of measure zero or a null set
if µ(N ) = 0. A property that holds on N c is said to hold µ-almost everywhere. In the context where
µ is clear, we says that a property holds just almost everywhere.
ˆ For examples, we say that two functions f and g are equal almost everywhere if f (x) = g(x) for all
x 6∈ N where N is a set of measure zero. We also say that a sequence of functions (fn ) converges
almost everywhere in X if limn→∞ fn (x) exists for all x 6∈ N .

5
 ˆ Definition 10. A measure space (X, X , µ) is complete if every subset of a set of measure zero is
measureable.
ˆ Theorem 11. Let (X, X , µ) be a measure space. Define (X, X , µ) by

X = {A ∪ M : A ∈ X , M ⊆ N where N ∈ X and µ(N ) = 0}

and

µ(A ∪ M ) = µ(A).

Then, (X, X , µ) is a complete measure space such that X ⊆ X and µ is the unique extension of µ to
X.
Proof (sketch). The hardest bit of the proof is to show that X is close under complementation. Let
A ∈ X , N ∈ X be a set of measure zero, and M ⊆ N . We have that (A ∪ M )c = Ac ∩ M c . Because
M c = N c ∪ (N − M ), we have that

(A ∪ M )c = Ac ∩ M c = Ac ∩ (N c ∪ (N − M )) = (Ac ∩ N c ) ∪ (Ac ∩ (N − M )).

We note that Ac ∩ N c ∈ X and Ac ∩ (N − M ) ⊆ N , so (A ∪ M )C ∈ X . The other parts of the proof

seems straightforward, and we refer to [Hunter, 2011] for a longer proof sketch.

2.2 Why Do We Need the Sigma-Algebra?

ˆ Why is the σ-algebra necessary? In other words, why do we care to limit ourselves to only some subsets
of P(X)? It turns out that we can arrive at contradictions if we want to define measures on all subsets.
ˆ Suppose that we want to define a probability measure µ on the set [0, 1) such that we define µ for all
subsets of X. Then, there exists a set A where the measure cannot consistenly be defined.
To construct this set, we first partition the elements of [0, 1) into equivalent classes. Let x ∈ [0, 1), we
let [x] = {y ∈ [0, 1) : (y − x) mod 1 ∈ Q ∩ [0, 1)} be the equivalence class containing x. Consider the
collection {[x] : x ∈ [0, 1)} of equivalent classes. By the axiom of choice, we can construct the set A
that contains exactly one element from each equivalent class.
Let x ∈ [0, 1). Then, x ∈ [y] for some y ∈ [0, 1), and y − x ∈ Q. Let Tq (A) = {(x + q) mod 1 : x ∈ A} be
S copy of A by q. We have that there exists one q ∈ Q ∩ [0, 1) such that x ∈ Tq (A).
a circlically translated
As a result, [0, 1) = q∈Q∩[0,1) Tq (A).
Moreover, we have that Tq (A) ∩ Tr (A) = ∅ for any q 6= r ∈ Q ∩ [0, 1). To see this, let there be x
such that x ∈ Tq (A) and x ∈ Tr (A). Then, (x + q) mod 1 ∈ A and (x + r) mod 1 ∈ A. Note that
(x + q) mod 1 and (x + r) mod 1 must belong in the same equivalent class, say [y] where y ∈ A. Since
both numbers are in A, it must be that (x + q) mod 1 = (x + r) mod 1 = y. So, q = r.
A sensible probability measure should be translation invariant. In other words, it is natural to require
that µ(Tq (A)) = A for all q.
Then, we run into
P a problem. What value should we assign to µ(A)? If µ(A) = 0, we have that
1 = µ([0, 1)) = q∈Q∩[0,1) µ(Tq (A)) = 0. However, if µ(A) > 0, then µ([0, 1)) = ∞. In both cases, we
cannot make µ([0, 1)) = 1.
ˆ In general, this problem crops up a lot as a consequence of the Banach–Tarski paradox.2
ˆ To avoid this problem, we have to forego one of the nice things: the axiom of choice, the requirement
that the measure is invariant to translations, or the fact that we can define measures on all subsets.
The approach taken by the math world is to choose the last by limitting the sets we can define measures
on to those in a σ-algebra.
2 https://en.wikipedia.org/wiki/Banach%E2%80%93Tarski_paradox

6
2.3 Pi-Systems and Lambda-Systems
ˆ We take a detour to explore structures that are related to the σ-algebra that will be useful in further
studies.
ˆ Definition 12. A collection of sets A ⊆ P(X) is a π-system if it is closed under finite intersections.
In other words, if A, B ∈ A, then A ∩ B ∈ A.
ˆ Definition 13. A collection of sets A ⊆ P(X) is a λ-system it satisfies the following properties.

1. X ∈ A.
2. It is closed under complementation. That is, A ∈ A implies Ac ∈ A.

S∞ under countable disjoint unions. That is, if (An ) is a sequence of disjoint sets in A,
3. It is closed
then i=n An ∈ A.
ˆ Note that the π-system and the σ-system both have weaker conditions than that of the σ-algebra.

ˆ Lemma 14. A λ-system is closed under proper set differences. In other words, If A is a λ-system,
A, B ∈ A and A ⊆ B, then B − A ∈ A.
Proof. We have that B c ∈ A, and A ∩ B c = ∅. As a result, A ∪ B c ∈ A and so is (A ∪ B c )c = B − A.

ˆ Lemma 15. A collection A ⊆ P(X) that is both a π-system and a λ-system is a σ-algebra.

Proof. We only need to show that A is closed under countable (general) unions. Let (An ) be a sequence
of sets in A. Let Bn = An ∩ Acn−1 ∩ Acn−2 ∩ · · · Ac1 . We have that B
Sn∞∈ A because
S∞ it is formed by finite
intersections. Moreover, the Bn ’s are disjoint from one other, so n=1 Bn = n=1 An is a member of
A as well.

ˆ Definition 16. Let A ⊆ P(X) be a collection of sets. The λ-system generated by A, denoted by
λ(A) is the smallest λ-system that contains A. In other words,
\ 
λ(A) = L ⊆ P(X) : L is a λ-system and A ⊆ L .

.
ˆ Note that since a σ-algebra is a λ-system, we have that λ(A) ⊆ σ(A) for all A ⊆ P(X).

ˆ The following theorem goes in the opposite direction and is useful in proving many uniqueness theorems.

Theorem 17 (Dynkin’s π–λ theorem). If A is a π-system, B is a λ-system, and A ⊆ B, then

σ(A) ⊆ B.

Proof. We will show that λ(A) is a σ-algebra. If this is the case, then σ(A) ⊆ λ(A) ⊆ B, and the
theorem holds.
By the last lemma, it is sufficient to show that λ(A) is closed under intersection. Doing this is rather
convoluted. First, for any A ⊆ X, define

GA = {B ⊆ X : A ∩ B ∈ λ(A)}.

We will show that, if A ∈ λ(A), then GA is a λ-system.

To see this, we first note that X ∈ GA because A ∩ X = A ∈ λ(A).
Next, we show that GA is closed under complementation. Let B ∈ GA . We have that A ∩ B ∈ λ(A)
Because A ∩ B ⊆ A, it follows from Lemma 14 that A − (A ∩ B) = A ∩ B c ∈ λ(A). Hence, B c ∈ GA .

7
 S∞
Next, we show that GA is closed under countable disjoin unions.
S∞ Let B = n=1 Bn where each
Bn ∈ GA . We
S∞have that A ∩ Bn ∈ λ(A) for all n, and so is n=1 (A ∩ Bn ) = A ∩ ∪∞
n=1 Bn = A ∩ B.
Hence, B = n=1 Bn .
We have now established that A ∈ λ(A) =⇒ GA is a λ-system.
We shall now show that A ∈ A =⇒ λ(A) ⊆ GA . To see this, let B ∈ A. It follows that A ∩ B ∈ A
because A is a π-system. Because A ⊆ λ(A), it follows that A ∩ B ∈ λ(A). In other words, B ∈ GA ,
and so A ⊆ GA . Because A ∈ A ⊆ λ(A), it follows that GA is a λ-system that contains A. Thus,
λ(A) ⊆ GA .
Now,
A ∈ A =⇒ λ(A) ⊆ GA
A ∈ A =⇒ (B ∈ λ(A) =⇒ B ∈ GA )
(A ∈ A) ∧ (B ∈ λ(A)) =⇒ B ∈ GA .
Because B ∈ GA iff A ∈ GB , it follows that:
(A ∈ A) ∧ (B ∈ λ(A)) =⇒ A ∈ GB
B ∈ λ(A) =⇒ (A ∈ A =⇒ A ∈ GB )
B ∈ λ(A) =⇒ A ⊆ GB .
Because GB is a λ-system, it follows that λ(A) ⊆ GB . So,
B ∈ λ(A) =⇒ λ(A) ⊆ GB .
Hence,
A ∈ λ(A) ∧ B ∈ λ(A) =⇒ A ∈ λ(A) ∧ λ(A) ⊆ GB
A ∈ λ(A) ∧ B ∈ λ(A) =⇒ A ∈ GB
A ∈ λ(A) ∧ B ∈ λ(A) =⇒ A ∩ B ∈ λ(A)
A, B ∈ λ(A) =⇒ A ∩ B ∈ λ(A).
It follows that λ(A) is a π-system, and thus a σ-algebra.

ˆ It follows that, if A is a π-system, then λ(A) = σ(A).

ˆ Here’s an application of the Dynkin’s theorem.
Theorem 18. Let µ1 and µ2 be two finite measures on σ(A) where A ⊆ P(X) is a π-system that
contains X. If they agree on A, then they also agree on σ(A).
Proof. Let E = {E ∈ σ(A) : µ1 (E) = µ2 (E)}. We will show that E is a λ-system. If so, it follows that
σ(A) ⊆ E, and we would be done.
By the assumption of the theorem, X ∈ A, so µ1 (X) = µ2 (X). Thus, X ∈ E.
Next, let E ∈ E. We have that
µ1 (E c ) = µ1 (X − E) = µ1 (X) − µ1 (E) = µ2 (X) − µ2 (E) = µ2 (X − E) = µ2 (E c ).
As a result, E c ∈ E. (Note that we can do this because the measure is finite, so we can apply Lemma 6.)
S∞
Lastly, let (En ) be a sequence of disjoint sets in E, and let E = n=1 En . It follows that
[n  X ∞ ∞
X n
[ 
µ1 (E) = µ1 En = µ1 (En ) = µ2 (En ) = µ2 En = µ2
n=1 n=1 n=1 n=1

So, E ∈ E as well.

8
2.4 Algebras and Monotone Classes
ˆ In this section, we take another detour on a structure that is similar to the σ-algebra and a theorem
about it that is similar to the Dynkin’s π-λ theorem.
ˆ Definition 19. A family A of subsets of a set X is said to be an algebra or a field on X if the
following properties are satisfied.
1. ∅, X ∈ A.
2. If E ∈ A, then E c = X − E ∈ A.
Sn
3. If E1 , E2 , . . . , En ∈ A, then i=1 Ei ∈ A.
ˆ Note that an algebra is a π-system because A ∩ B = (Ac ∪ B c )c . However, it is not a λ-system.
ˆ Definition 20. A class M of subsets of X is called a monotone class if it is closed under the
formation of monotone unions and intersections. In other words,
S∞
– If A1 , A2 , . . . ∈ M and Ai ⊆ Ai+1 for all i, then i=1 Ai ∈ M, amd
T∞
– If A1 , A2 , . . . ∈ M and Ai ⊇ Ai+1 for all i, then i=1 Ai ∈ M.
ˆ Note that a σ-algebra is a monotone class because it is already closed under arbitrary countable unions
and intersections.
ˆ Lemma 21. Let A be a subset of P(X). If A is both an algebra and a monotone class, then it is a
σ-algebra.
Proof. We only have to show Snthat A is closed under arbitrary countable unions. Let (An ) be a sequence
of sets in A. Define Bn = i=1 Ai . We have that S∞Bn ∈ A for all n because A is an algebra. Moreover,
because
S∞ B n ⊆
S∞Bn+1 for all n, it follows
S∞ that n=1 Bn ∈ A because A is a monotone class. Because
n=1 A n = n=1 B n , we have that n=1 A n ∈ A too, so A is a σ-algebra.

ˆ Definition 22. Let A be a subset of P(X). The monotone class generated by A, denoted by
m(A), is intersection of all monotone classes that contains A. That is,
\
m(A) = {M : M is a monotone class and A ⊆ M}.

ˆ Theorem 23 (Halmos’s monotone class lemma). Let A be an algebra and M be a monotone

class. Then, A ⊆ M implies σ(A) ⊆ M. In particular, σ(A) = m(A).
Proof. Since A ⊆ σ(A) and σ(A) is a monotone class that contains A, it follows that m(A) ⊆ σ(A).
As a result, it suffices to show that σ(A) ⊆ m(A). For this, it suffices to show that m(A) is an algebra
(which will imply that it is a σ-algebra according to Lemma 21).
(Two basic sets) Because A ⊆ m(A), it follows that ∅ and X are both in m(A).
(Closure under complementation) Consider G = {A : Ac ∈ m(A)}. Because m(A) is a monotone
class, it follows that G is a monotone class as well. It follows that m(A) ⊆ G. This means that, for
any A ∈ m(A), it means that A ∈ G, which means that Ac ∈ m(A). As a result, m(A) is closed under
complementation.
(Closure under finite union) Define G1 = {A : A ∪ B ∈ m(A) for all B ∈ A}. We have that G1 is a
monotone class. To see this, suppose that (An ) is an increasing sequence of sets in G1 . It follows from
definition that (An ∪ B) is also an increasing sequence of sets in m(A). As a result,
∞
[ ∞
[ 
(An ∪ B) = An ∪ B ∈ m(A),
n=1 n=1

9
 S∞
which implies that n=1 An ∈ G1 . The condition involving countable monotone intersections can be
shown similarly.
Because A is an algebra, it follows that
A ∈ A ∧ B ∈ A =⇒ A ∪ B ∈ A.
Because A ⊆ m(A), we can also say that
A ∈ A ∧ B ∈ A =⇒ A ∪ B ∈ m(A)
 
A ∈ A =⇒ B ∈ A =⇒ A ∪ B ∈ m(A)
A ∈ A =⇒ ∀B∈A [A ∪ B ∈ m(A)],
A ∈ A =⇒ A ∈ G1 .
In other words, A ⊆ G1 . Because G1 is a monotone class and G1 contains A, it follows that m(A) ⊆ G1 .
Now,
m(A) ⊆ G1 ≡ A ∈ m(A) =⇒ A ∈ G1
≡ A ∈ m(A) =⇒ ∀B∈A [A ∪ B ∈ m(A)]
 
≡ A ∈ m(A) =⇒ B ∈ A =⇒ A ∪ B ∈ m(A)
≡ A ∈ m(A) ∧ B ∈ A =⇒ A ∪ B ∈ m(A).

Define G2 = {B : A ∪ B ∈ m(A) for all A ∈ m(A)}. It follows from the above statement that A ⊆ G2 .
Moreover, we can show again that G2 is a monotone class by repeating the same argument we used for
G1 . As a result, m(A) ⊆ G2 . In other words,
B ∈ m(A) =⇒ B ∈ G2
B ∈ m(A) =⇒ ∀A∈m(A) [A ∪ B ∈ m(A)]
 
B ∈ m(A) =⇒ A ∈ m(A) =⇒ A ∪ B ∈ m(A)
A ∈ m(A) ∧ B ∈ m(A) =⇒ A ∪ B ∈ m(A),
which means that m(A) is closed under finite unions. So, m(A) is an algebra.

3 Lebesgue Measures
The Lebesgue measure on R is a measure that corresponds to the notion of “length” on the real line. We
will construct it in this section.

3.1 Length as a Premeasure

ˆ The natural notion of length can be defined as follows.
– Let ` denote the length function.
– The length of the half-open interval (a, b] is defined to be b − a.
– The lengths of (−∞, b], (a, +∞), and (−∞, ∞) are defined to be ∞.
– The length of the union of a finite number of disjoint sets of intervals of these forms is defined to
be the sum of the corresponding lengths.
[n  Xn
` (ai , bn ] = (bi − ai ).
i=1 i=1

10
ˆ Let F be the collection of subsets of R that contains all intervals of the forms

(a, b], (−∞, b], (a, ∞), (∞, ∞), (1)

and all the finite unions of such intervals.

ˆ By the notion above, we have that ` is a function of signature F → [0, ∞]. However, we cannot claim
that it is a measure (1) we have not defined how to deal with countable unions of such intervals yet,
and (2) we have not identified the σ-algebra upon which it acts.
ˆ Furthermore, we cannot claim that F is a σ-algebra. However, F is an algebra.

ˆ Lemma 24. F is an algebra on R.

Proof. We have that:

– ∅ ∈ F because ∅ = {x : 1 < x ≤ 1} = (1, 1].
– R = (−∞, ∞) ∈ F by definition.
– F is closed under complementation.
* (−∞, b]c = (b, ∞) ∈ F.
* (a, ∞)c = (−∞, a] ∈ F.
* (a, b]c = (∞, a] ∪ (b, ∞) ∈ F.
* ∅ and (−∞, ∞) are in F.
– F is closed under finite unions by definition.

ˆ We now turn back to the length function `. It turns out that we can show that it has more nice
properties that are worthy of being encapsulated with an abstraction.
ˆ Definition 25. Let A be an algebra on X. A premeasure on A is an extended real valued function
µ defined on A that satisfies the following properties.
1. µ(∅) = 0.
2. µ(E) ≥ 0 for all E ∈ A.
S∞
3. If (En ) is any disjoint sequence of sets in A such that i=n En belongs to A, then
∞
[  X∞
µ En = µ(En ).
n=1 n=1

ˆ Lemma 26. The length function ` is a premeasure on F.

Proof. First, `(∅) = `((1, 1]) = 1 − 1 = 0.

Second, if E ∈ F, then E is a finite union of intervals of the forms in (1). We can subdivide these
intervals into disjoint pieces, and the length of the union is the sum of the length of the individual
pieces, each of which is non-negative. So, we have that µ(E) ≥ 0.
S∞
S∞suppose that (En ) is a sequence of disjoint sets in F such that i=n En is also in F. Note
Lastly,
that i=n En is a finite union of intervals, so we can again partition them into disjoint pieces, each of
which is of a form in (1). Each piece is now a union of a countable collection of disjoint elements of F.
WLOG, we may treat each of the En ’s as an interval which is disjoint from any other. Our goal now
would be to show that the lengths of the constituent intervals add up to the length of the piece.

11
 A piece can be any of the 4 types. We will only deal with the (a, b] type in this proof as the proof of
other types are similar. Suppose, then, that
∞
[
(a, b] = (aj , bj ]
j=1

where the intervals are disjoint. Consider the first n intervals. We may assume that

a ≤ a1 < b1 ≤ a2 < b2 ≤ · · · ≤ an < bn ≤ b.

We have that
n
X n
X
`((ai , bi ]) = (bi − ai ) ≤ bn − a1 ≤ b − a = `((a, b]).
i=1 i=1

Because n is arbitrary, we have that

∞
X
`((ai , bi ]) ≤ `((a, b]).
i=1

For
P the other direction, let ε > 0 be arbitrary. Let (εj ) be a sequence of positive numbers with
εj < ε/2. Consider the interval Ij = (aj − εj , bk + εj ). The collection {Ij } of open sets is a cover of
the interval [a, b]. Since [a, b] is compact, it has a finite subcover, say, I1 , I2 , . . . , Im . By reordering
and discarding some intervals, we may assume that

a1 − ε1 < a
b < bm + εm
aj − εj < bj−1 + εj−1 .

If follows that
m
X m
X ∞
X
b − a ≤ (bm + εm ) − (a1 − ε1 ) ≤ [(bj + εj ) − (aj − εj )] ≤ ε + (bj − aj ) ≤ ε + (bj − aj ).
j=1 j=1 j=1

Since ε > 0 is arbitrary, it follows that

∞
X
`((a, b]) ≤ `((ai , bi ]).
i=1
P∞
As a result, `((a, b]) = i=1 `((ai , bi ]).
Combining the results of all cases, we can conclude that ` is countably additive in F.

3.2 Length as an Outer Measure

ˆ The good news is that, given a premeasure µ on an algebra A, we can show that µ can be extended to
a measure on a σ-algebra.
– In other words, there exist a σ-algebra A∗ containing A and a measure µ∗ defined on A∗ such
that µ∗ (E) = µ(E) for all E ∈ A.
As a result, we can extend our ` so that it becomes a measure on a σ-algebra.
ˆ The way to extend a premeasure µ is as follows.

12
 Definition 27. Given a premeasure µ defined on an algebra A on set X, define µ∗ : P(X) → [0, ∞]
to be
X∞ 
∗
µ (B) = inf µ(Ej ) : (Ej ) ∈ C(B)
j=1
S∞
where C(B) is the set of all sequences (Ej ) of sets in A such that B ⊆ j=1 Ej .

ˆ The following lemma is useful for proving other results

P∞ 28. For any set B ⊆ X and any real number ε > 0, there exists a sequence (Ej ) ∈ C(B) such
Lemma
that j=1 µ(Ej ) ≤ µ∗ (B) + ε.

P∞ is false. Then, there exists ε > 0 such that, for all (Ej ) ∈ C(B),
Proof. Suppose that the lemma
itPmust be the case that j=1 µ(Ej ) > µ∗ (B) + ε. It follows that µ∗ (B) + ε is a lower bound of
{ µ(Ej ) : Ej ∈ C(B)}. So,
∞
X 
∗
µ (B) = inf µ(Ej ) : (Ej ) ∈ C(B) ≥ µ∗ (B) + ε,
j=1

and we have arrived at a contradiction.

ˆ Lemma 29. The function µ∗ has the following properties.

(a) µ∗ (∅) = 0.
(b) µ∗ (B) ≥ 0 for any B ⊆ X.
(c) If A ⊆ B, then µ∗ (A) ≤ µ∗ (B).
(d) If A ∈ A, then µ∗ (A) = µ(A).
(e) If (Bn ) is a sequence of subsets of X, then
[∞  X∞
µ∗ Bn ≤ µ∗ (Bn ).
n=1 i=1

Proof. Let us call a sequence (Ej ) ∈ C(B) a cover of B.

P
measure, so µ(Ej ) ≥ 0 for all j. Hence,
For (b), recall that µ is a P µ(Ej ) ≥ 0 for any sequence (Ej ).
As a result, µ∗ (B) = inf{ µ(Ej ) : (Ej ) ∈ C(∅)} ≥ 0.
For (a), note that the sequence (∅, ∅, . . . ) is a cover of ∅, so µ∗ (∅) = inf{ µ(Ej ) : (Ej ) ∈ C(∅)} ≤ 0.
P
However, from (b), we have that µ∗ (∅) ≥ 0, so µ∗ (∅) = 0.
S
For (c), let A ⊆ B. Let (Ej ) ∈ C(B). We have that A ⊆ B ⊆ Ej , so (Ej ) ∈ C(A) too. It follows
that C(A) ⊇ C(B), which implies that
X  X 
µ(Ej ) : (Ej ) ∈ C(A) ⊇ µ(Ej ) : (Ej ) ∈ C(B) ,

and so
X  X 
µ∗ (A) = inf µ(Ej ) : (Ej ) ∈ C(A) ≤ inf µ(Ej ) : (Ej ) ∈ C(B) = µ∗ (B).

For (d), let A ∈ A. We have that {A, ∅, ∅, . . . ) is a cover of A. As a result,

µ∗ (A) ≤ µ(A) + µ(∅) + µ(∅) + · · · = µ(A).

13
 S
Let (Ej ) be a cover of A. We have that A = (A ∩ Ej ). Because µ is a measure,
∞
X ∞
X
µ(A) ≤ µ(A ∩ Ej ) ≤ µ(Ej ).
j=1 j=1

µ(Ej ) : (Ej ) ∈ C(A)} = µ∗ (A). Hence, µ(A) = µ∗ (A).

P
It follows that µ(A) ≤ inf{
For (e), let ε > 0 be arbitrary. For each n, apply Lemma 28 to choose a sequence (Enk ) of sets in A
such that (Enk ) covers Bn and
∞
X ε
µ(Enk ) ≤ µ∗ (Bn ) + .
2n
k=1
S
Since {Enk : n, k ∈ N} is a countable collection from A whose union contains Bn , it follows that
∞
[  ∞ X
X ∞ ∞
X
µ∗ Bn ≤ µ(Enk ) ≤ ε + µ∗ (Bn ).
n=1 n=1 k=1 k=1

Since ε can be arbitrarily small, we have that Property (e) holds.

ˆ Definition 30. An outer measure µ∗ on a set X is a function µ∗ : P(X) → [0, ∞] such that the
following properties hold.
1. µ∗ (∅) = 0.
2. If E ⊆ F ⊆ X, then µ∗ (E) ≤ µ∗ (F ).
3. µ∗ is countably subadditive. In other words, if {Ei ⊆ X : i ∈ N} is a countable collection of
subsets of X, then
∞
[  ∞
X
µ∗ Ei ≤ µ∗ (Ei ).
i=1 i=1

ˆ Lemma 29 shows that µ∗ is an outer measure on X if µ is a premeasure on an algebra A on X. The

outer measure µ∗ is called the outer measured generated by µ.

3.3 Measure from Outer Measure

ˆ µ∗ is defined for arbitrary subsets of X, so it is also defined on countable unions of subsets of X as
well. However, it is not yet a fully fledged measure because we cannot yet find a σ-algebra on which it
is countably additive on.
ˆ The following criterion is used to classify members of such a σ-algebra.
Definition 31. A subset E of X is said to be µ∗ -measurable if

µ∗ (A) = µ∗ (A ∩ E) + µ∗ (A ∩ E c )

for every subset A of X.

ˆ A µ∗ -measurable set E splits any set A into pieces whose output measures add up to the outer measure
of A. In other words, a set is µ∗ -measurable if it splits other sets in a “nice” way.
ˆ Theorem 32 (Carathéodory’s extension theorem). Let µ∗ be an outer measure on X. The col-
lection of µ∗ -measurable sets is a σ-algebra on X. Moreover, µ∗ is a measure on this collection.

14
Proof. Let X ∗ denote the set of µ∗ -measurable sets on X. We shall show that X ∗ is a σ-algebra, and
µ∗ is a measure on it.
(∞, X ∈ X ∗ ) Because µ∗ is a premeasure, we have that µ∗ (∅) = 0. For any set A ⊆ X, we have that

µ∗ (A) = µ∗ (∅) + µ∗ (A) = µ∗ (A ∩ ∅) + µ∗ (A ∩ X).

Because ∅ and X are complements of each other, it follows that they are µ∗ -measurable and so belong
to X ∗ .
(Closure under complementation) Let E be a µ∗ -measurable set. It follows that

µ∗ (A) = µ∗ (∅) + µ∗ (A ∩ E) = µ∗ (A ∩ E c )

This implies that E c is also µ∗ -measurable.

(Closure under finite unions) This step is required to show closure under countable union. Suppose
that E and F are µ∗ -measurable. Let A ⊆ X. We need to show that

µ∗ (A) = µ∗ (A ∩ (E ∪ F )) + µ∗ (A ∩ (E ∪ F )c ).

Because A = (A ∩ (E ∪ F )) ∪ (A ∩ (E ∪ F )c ) and µ∗ is subadditive, we have that

µ∗ (A) ≤ µ∗ (A ∩ (E ∪ F )) + µ∗ (A ∩ (E ∪ F )c ).

Thus, it remains to show that µ∗ (A) ≥ µ∗ (A∩(E ∪F ))+µ∗ (A∩(E ∪F )c ). Because E is µ∗ -measurable,
we have that

µ∗ (A) = µ∗ (A ∩ E) + µ∗ (A ∩ E c ).

Since both A ∩ E and A ∩ E c are subsets of X and B is µ∗ -measurable, we have

µ∗ (A ∩ E) = µ∗ (A ∩ E ∩ F ) + µ∗ (A ∩ E ∩ F c )
µ∗ (A ∩ E c ) = µ∗ (A ∩ E c ∩ F ) + µ∗ (A ∩ E c ∩ F c ).

As a result,

µ∗ (A) = µ∗ (A ∩ E ∩ F ) + µ∗ (A ∩ E ∩ F c ) + µ∗ (A ∩ E c ∩ F ) + µ∗ (A ∩ E c ∩ F c )
= µ∗ (A ∩ E ∩ F ) + µ∗ (A ∩ E ∩ F c ) + µ∗ (A ∩ E c ∩ F ) + µ∗ (A ∩ (E ∪ F )c ).

Because E ∪ F = (E ∩ F ) ∪ (E ∩ F c ) ∪ (E c ∩ F ), we have that

A ∩ (E ∪ F ) = (A ∩ E ∩ F ) ∪ (A ∩ E ∩ F c ) ∪ (A ∩ E c ∩ F ).

So,

µ∗ (A ∩ (E ∪ F )) ≤ µ∗ (A ∩ E ∩ F ) ∪ µ∗ (A ∩ E ∩ F c ) ∪ µ∗ (A ∩ E c ∩ F ).

Thus,

µ∗ (A) = µ∗ (A ∩ E ∩ F ) + µ∗ (A ∩ E ∩ F c ) + µ∗ (A ∩ E c ∩ F ) + µ∗ (A ∩ (E ∪ F )c )
≥ µ∗ (A ∩ (E ∪ F )) + µ∗ (A ∩ (E ∪ F )c ).

It follows that µ∗ (A) = µ∗ (A ∩ (E ∪ F )) + µ∗ (A ∩ (E ∪ F )c ), and so E ∪ F is also measurable.

(Closure under finite intersections) We briefly mention that closure under complementation and
closure under finite union implies closure under complementation. This is a consequence of de Morgan’s
law: E ∩ F = (E c ∪ F c )c .

15
(Closure under disjoint countable unions implies closure under countable S unions) We need
to show closure under countable union. However, it suffices to only show that Ej is measurable for
all sequences (Ej ) where the sets are measurable and disjoint. To see this, let Fj denote the union of
the first j sets:
j
[
Fj = Ej ,
i=1

and let G0 = E1 and Gj = Fj+1 − Fj = Fj+1 ∩ Fjc

for j ≥ 1. It follows that, Gj is measurable for all
j because X ∗ is closed under finite unions and intersections. Moreover, the Gj ’s are disjoint, and
∞
[ ∞
[
Ej = Gj .
j=1 j=0
S∞ S∞
Hence, the measurability of j=0 Gj imples the measurability of j=1 Ej .
(Finite additivity) In order to establish closure under countable unions, it is useful to show that
µ∗ is additive for finite disjoint unions. Let E and F be any disjoint measurable sets. Because E is
measurable, we have that

µ∗ (E ∪ F ) = µ∗ ((E ∪ F ) ∩ E) + µ∗ ((E ∪ F ) ∩ E c ) = µ∗ (E) + µ∗ ((E ∪ F ) − E) = µ∗ (E) + µ∗ (F ).

(Closure under disjoint countable unions) Let (Ej ) be a sequence of disjoint measurable sets.
Let
j
[ ∞
[
Fj = Ej , F = Ej .
i=1 i=1

We need to show show that, for any set A ⊆ X, it is true that µ∗ (A) = µ∗ (A ∩ F ) + µ∗ (A ∩ F c ).
Because µ∗ is subadditive, we already know that µ∗ (A) ≤ µ∗ (A ∩ F ) + µ∗ (A ∩ F c ). So, we only need
to show that µ∗ (A) ≥ µ∗ (A ∩ F ) + µ∗ (A ∩ F c ).
For any j, we have that Fj is measurable because of closure under finite unions. So

µ∗ (A) = µ∗ (A ∩ Fj ) + µ∗ (A ∩ Fjc )
 [j 
∗
=µ A∩ Ei + µ∗ (A ∩ Fjc )
i=1
n
[ 
= µ∗ (A ∩ Ei ) + µ∗ (A ∩ Fjc ).
i=1

Because A ∩ E1 , A ∩ E2 , . . . , and A ∩ Ej are mutually disjoint, we have that

j
X
µ∗ (A) = µ∗ (A ∩ Ej ) + µ∗ (A ∩ Fjc ).
i=1

Moreover, because Fj ⊆ F , it follows that A ∩ Fjc ⊇ A ∩ F c . Hence, µ∗ (A ∩ Fjc ) ≥ µ∗ (A ∩ F c ). Thus,

j
X
µ∗ (A) ≥ µ∗ (A ∩ Ei ) + µ∗ (A ∩ F c ).
i=1

16
 Taking the limit as j → ∞, we have that
∞
X
µ∗ (A) ≥ µ∗ (A ∩ Ei ) + µ∗ (A ∩ F c )
i=1
∞
[ 
∗
≥µ (A ∩ Ei ) + µ∗ (A ∩ F c )
i=1
 ∞
[ 
= µ∗ A ∩ Ei + µ∗ (A ∩ F c )
i=1
= µ∗ (A ∩ F ) + µ∗ (A ∩ F c ),
which implies that F is µ∗ -measurable.
(Countable additivity) In the proof of closure under countable union, we established that
∞
X
µ∗ (A) ≥ µ∗ (A ∩ Ei ) + µ∗ (A ∩ F c )
i=1
∞
[ 
∗
≥µ (A ∩ Ei ) + µ∗ (A ∩ F c )
i=1
= µ∗ (A ∩ F ) + µ∗ (A ∩ F c )
≥ µ∗ (A).
Therefore, it must be the case that
∞
X ∞
[ 
∗ ∗ ∗
µ (A ∩ Ei ) + µ (A ∩ F ) = µ c
(A ∩ Ei ) + µ∗ (A ∩ F c ),
i=1 i=1

and this is true for any set A ⊆ X. Taking A = F , we have that

X∞ ∞
[ 
∗ ∗ ∗
c
µ (F ∩ Ei ) + µ (F ∩ F ) = µ (F ∩ Ei ) + µ∗ (F ∩ F c )
i=1 i=1
∞
X ∞
[ 
µ∗ (Ei ) + µ∗ (∅) = µ∗ Ei + µ∗ (∅)
i=1 i=1
∞
X ∞
[ 
∗ ∗
µ (Ei ) = µ Ei ,
i=1 i=1

which shows that µ∗ is countably additive.

ˆ In the proof of Theorem 32, there is a useful fact that we will use later, so let us extract it into a
lemma here.
Lemma 33. For any sequence of sets (Ej ) where each set belongs to an algebra A, there exists a
sequence of disjoint sets (Hj ) such that, for each j, we have that Hj ∈ A, and Hj ⊆ Ej . Moreover,
∞
[ ∞
[
Hj = Ej
j=1 j=1

Proof. Choose Hj to be Gj−1 as defined in the proof of Theorem 32. We have that all the properties
we want have already been proven except for Hj ⊆ Ej . However, this should be clear because
c c c
Hj = Gj−1 = Fj ∩ Fj−1 = (Ej ∪ Fj−1 ) ∩ Fj−1 = Ej ∩ Fj−1 ⊆ Ej .
We are done.

17
ˆ Proposition 34. The measure space (X, X ∗ , µ∗ ) in Theorem 32 is complete.
Proof. Let E be a µ∗ -measurable set of measure zero. Let B ⊆ E. We need to show that B is also
µ∗ -measurable, and that µ∗ (B) = 0. The latter statement is immediate because µ∗ is subadditive.
Let A ⊆ X. First, 0 ≤ µ∗ (A ∩ B) ≤ µ∗ (A ∩ E) ≤ µ∗ (E) = 0, so µ∗ (A ∩ B) = 0. Now,

A ∩ Bc ⊆ A
µ∗ (A ∩ B c ) ≤ µ∗ (A)
µ∗ (A ∩ B) + µ∗ (A ∩ B c ) ≤ µ∗ (A).

However, because µ∗ is subadditive, we have that µ∗ (A ∩ B) + µ∗ (A ∩ B c ) ≥ µ∗ (A). It follows that

µ∗ (A ∩ B) + µ∗ (A ∩ B c ) = µ∗ (A), and B is also µ∗ -measurable.

ˆ Theorem 35. Let µ be a premeasure on an algebra A on X. Let µ∗ be the outer measure generated
by µ, and let A∗ be the collection of µ∗ -measurable sets. Then, A ⊆ A∗ , and (X, A∗ , µ∗ ) is complete
measure space with the property that µ∗ (E) = µ(E) for all E ∈ A.
Proof. That µ∗ is a σ-algebra on A∗ follows from Theorem 32. That (X, A∗ , µ∗ ) is a complete measure
space follows from Proposition 34. That µ∗ (E) = µ(E) for all E ∈ A follows from Lemma 29. It
remains to show that A ⊆ A∗ .
Let E ∈ A. Let A ⊆ X. Let (Ej ) be a collection of sets in A that covers A. According to Lemma 28,
we can choose (Ej ) so that
∞
X ∞
X
∗
µ (Ej ) = µ(Ej ) ≤ µ∗ (A) + ε
j=1 j=1

for any arbitrary ε > 0. Because (Ej ) covers A, it follows that (Ej ∩ E) covers A ∩ E, and (Ej ∩ E c )
covers A ∩ E c . As a result,
∞
X ∞
X
µ∗ (A ∩ E) ≤ µ(Ej ∩ E) = µ∗ (Ej ∩ E),
j=1 j=1
∞
X X∞
µ∗ (A ∩ E c ) ≤ µ(Ej ∩ E c ) = µ∗ (Ej ∩ E c ).
j=1 j=1

Adding the above two inequalities, we have that

∞
X ∞
X
µ∗ (A ∩ E) + µ∗ (A ∩ E c ) ≤ µ∗ (Ej ∩ E) + µ∗ (Ej ∩ E c )
j=1 j=1
∞
X  ∗
µ (Ej ∩ E) + µ∗ (Ej ∩ E c ) .

=
j=1

c
Because Ej ∩ E and Ej ∩ E are disjoint, we have that
∞
X ∞
 X
µ∗ (A ∩ E) + µ∗ (A ∩ E c ) ≤
 ∗
µ (Ej ∩ E) + µ∗ (Ej ∩ E c ) = µ(Ej ) ≤ µ∗ (A) + ε.
j=1 j=1

As ε is arbitrary, it follows that µ∗ (A ∩ E) + µ∗ (A ∩ E c ) ≤ µ(A). Moreover, because µ∗ is subadditive,

we have that µ∗ (A ∩ E) + µ∗ (A ∩ E c ) ≥ µ∗ (A). It follows that E is measurable, and A ⊆ A∗ .

ˆ Theorem 36 (Hahn’s extension theorem). If µ is a σ-finite premeasure on an algebra A, then µ∗

is the unique extension of µ that is a measure on A∗ .

18
 Proof. Let ν be an extension of µ that is a measure on A∗ . We have that ν(E) = µ(E) = µ∗ (E) for
all E ∈ A. We need to show that ν(E) = µ∗ (E) for all E ∈ A∗ .
Let E ⊆ A∗ . Let (Ej ) be a sequence of sets that covers E such that Ej ∈ A for all j. We have that
∞
[  ∞
X ∞
X
ν(E) ≤ ν Ej ≤ ν(Ej ) = µ(Ej ).
j=1 j=1 j=1

It follows that ν(E) is a lower bound of the set { µ(Ej ) : Ej ∈ C(E)}. As a result, ν(E) ≤ µ∗ (E) for
P
any E ∈ A∗ .
S
Since µ is σ-finite, there exists a sequence (Ej ) with each Ej ∈ A such that j Ej = X and µ(Ej ) =
µ∗ (Ej ) = ν(Ej ) is finite for all j. Applying Lemma 33 to (Ej ), we have a disjoint sequence (Hj ) that
covers X. Morever, µ(Hj ) ≤ µ(Ej ) < ∞.
For each Hj , we have that

µ∗ (Hj ) = ν(Hj ) (Hj ∈ A)

∗ c c
µ ((Hj ∩ E) ∪ (Hj ∩ E )) = ν((Hj ∩ E) ∪ (Hj ∩ E ))
µ∗ (Hj ∩ E) + µ∗ (Hj ∩ E c ) = ν(Hj ∩ E) + ν(Hj ∩ E c ) (both µ∗ and ν are additive)

Because µ∗ (Hj ∩ E) ≥ ν(Hj ∩ E) and µ∗ (Hj ∩ E c ) ≥ ν(Hj ∩ E c ), the equality can hold only if
µ∗ (Hj ∩ E) = ν(Hj ∩ E) and µ∗ (Hj ∩ E c ) = ν(Hj ∩ E c ).
Lastly,
 ∞
[  ∞
[  X∞
∗ ∗ ∗ ∗
µ (E) = µ (E ∩ X) = µ E∩ Hj =µ (E ∩ Hj ) = µ∗ (E ∩ Hj ).
j=1 j=1 j=1

Similarly,
∞
X
ν(E) = ν(E ∩ Hj ).
j=1

Because we just show that µ∗ (Hj ∩ E) = ν(Hj ∩ E) for all j, it follows that µ∗ (E) = ν(E).

3.4 Lebesgue Measure on the Real Line

ˆ Recall that the length function ` is defined on F, the collection of all finite unions of intervals of the
form

(a, b], (−∞, b], (a, ∞), (−∞, ∞),

We have that F is an algebra (Lemma 24) and ` is a premeasure on it (Lemma 26).

ˆ We can apply Theorem 32 to generate a measure space (R, F ∗ , `∗ ), which is complete by construction
(Proposition 34). Moreover, ` is σ-finite because the real line can be covered by intervals of length 1.
Thus, by Theorem 36, `∗ is the only extension of ` on F ∗ .
ˆ The elements of the σ-algebra F ∗ are called the Lebesgue measurable sets. The measure `∗ is
called the Lebesgue measure on R.
ˆ The smallest σ-algebra containing F is the Borel algebra B.The restriction of Lebesgue measure to the
Borel sets is called the Borel measure.
ˆ Because B ⊆ F ∗ , we have that F ∗ is more extensive than B because it contains more sets. However,
they can be practically ignored.

19
 Proposition 37. Let A be a Lebesgue measurable subset of R. There exists a Borel measurable subset
B of R such that A ⊆ B, and `∗ (B − A) = 0.

Proof. Let A be a Lebesgue measurable subset of R. Let us assume for now that `∗ (A) is finite. For
arbitrary ε > 0 and for each n ∈ N, let S (En,j ) be a sequence of sets in F such that (En,j ) covers A and
`(En,j ) ≥ `∗ (A) + ε/n. Let Bn = EP
P
n,j . We have that Bn is a Borel set because
T∞ it is a countable
union of intervals. Moreover, `∗ (Bn ) ≤ `(Bn,j ) ≤ `∗ (A) + ε/n. Take B = n=1 Bn . We have that
B is a Borel set because it is a countable Snintersection of Borel sets. We also have that A ⊆ B because
each Bn covers A. Moreover, let Bn0 = i=1 Bi . We have that `∗ (B10 ) is finite, and the sequence Bn0 is
decreasing, and `∗ (Bn0 ) ≤ `∗ (Bn ) ≤ µ∗ (A) + ε/n By Lemma 7, we have that
 
ε
`∗ (B) = lim `∗ (Bn0 ) ≤ lim `∗ (Bn ) ≤ lim `∗ (A) + = `∗ (A).
n→∞ n→∞ n→∞ n

However, because A ⊆ B, it follows that `∗ (B) ≥ `∗ (A), and thus `∗ (B) = `∗ (A). Because `∗ (B) is
finite, it follows that `∗ (B − A) = `∗ (B) − `∗ (A) = 0.
Let us now remove the assumption that `∗ (A) is finite. Let Ak = A ∩ [k, k + 1) for any n ∈ Z. We have
that the Ak ’s are disjoint. Moreover `∗ (Ak ) ≤ `∗ ([k, k + 1)) = 1. Because Ak is finite,Sthere exists a
∞
Borel set Bk such that Ak ⊆ Bk and `∗ (Bk ) = `∗ (Ak ) and `∗ (Bk − Ak ) = 0. Take B = k=−∞ Bk . It
Sj union of Borel sets. Moreover, A ⊆ B obviously.
follows that B is aTBorel set because it is a countable
∞
Lastly, B − A = k=−∞ (Bk − Ak ). Let Cj = k=−j (Bj − Aj ). We have that Cj is an increasing
sequence of sets. By Lemma 7,
∞
[ 
∗ ∗
` (B − A) = ` Cj = lim `∗ (Cj ) = 0.
j→∞
j=0

We are done.

3.5 Lebesgue Measures in Higher Dimensions

ˆ Note that the construction in this section can be extended to Rk .

ˆ Here, the length of intervals becomes the volume of “rectangles” of the form

I1 × I2 × · · · × Ik

where Ij is an interval of the forms in (1). The volume is then defined to be

v(I1 × I2 × · · · × Ik ) = `(I1 ) `(I2 ) · · · `(In ).

ˆ The measure generated in this way is called the Lebesgue measure on Rk . There are also Borel
measure on Rk and Borel algebra on Rk . All the results proved thus far do hold on them.

4 Measurable Functions
An important component of measure theory is the definition of integrals of functions and the study of their
properties. Measurable functions are nice functions upon which integrals can be defined.

4.1 Measurable Real-Valued Functions

ˆ Consider a fixed measurable space (X, X ).

20
ˆ Definition 38. A function f : X → R is said to be X -measurable (or simply measurable) if, for
every real number α, the set {x ∈ X : f (x) > α} is measurable (in other words, belongs to X ).
ˆ Proposition 39. For a function f : X → R, the following statements are equivalent.

(a) For every α ∈ R, the set Aα = {x ∈ X : f (x) > α} belongs to X .

(b) For every α ∈ R, the set Bα = {x ∈ X : f (x) ≤ α} belongs to X .
(c) For every α ∈ R, the set Cα = {x ∈ X : f (x) ≥ α} belongs to X .
(d) For every α ∈ R, the set Dα = {x ∈ X : f (x) < α} belongs to X .
Proof. Because Aα = Bαc , we have that (a) and (b) are equivalent. The same can be said for (c) and
(d).
We will now show that (a) implies (c). Let α ∈ R. Consider the sequenceT∞ (Aa−1/n ) for n ∈ N, all of
which belongs to X because of (a). So, there countable intersection n=1 Aα−1/n = C also belongs to
X.
S∞
Next, we will show that (c) implies (a). This is a result of the fact that Aα = n=1 Cα+1/n .

ˆ A constant function f (x) = c is measurable. This is because, if c ≤ α, then x ∈ X : f (x) > α = ∅. On

the other hand, if c > α, then x ∈ X : f (x) > α = X.

ˆ If E ∈ X , then the characteristic function χE , defined by

(
1, x ∈ E
χE (X) = ,
0, x 6∈ E

is measureable. This is because {x ∈ X : χE (x) > α} is either X, E, or ∅ depending on whether α is

[1, ∞), [0, 1) or (−∞, 0), respectively.
ˆ If X = R, and X is the Borel algebra B, then any continuous function f : R → R is Borel measurable.
This is because (α, ∞) is an open set, so f ( (α, ∞)) = {x ∈ R : f (x) > α} is open as well. (This is a
fact from real analysis.)
ˆ If X = R and X = B, then any monotone function is Borel measurable. This is because the set
{x ∈ R : f (x) > α} are of the form {x ∈ R : x > β} or {x ∈ R : x ≥ β} or R or ∅, all of which are
Borel measurable.
ˆ Lemma 40. Let f and g be measurable real-valued functions and let c be a real number. Then the
functions

cf, f 2 , f + g, f g, |f |, 1/f, min(f, g), max(f, g)

are also measurable. For the case of 1/f , we assume that f (x) 6= 0 for all x.

Proof. (a) If c = 0, then cf is a constant function and so is measurable. If c > 0, then

{x ∈ X : cf (x) > α} = {x ∈ X : f (x) > α/c}.

The RHS is measurable, so cf is measurable. The case where c < 0 can be handled similarly.
(b) If α < 0, then {x ∈ X : (f (x))2 > α} = X ∈ X . If α ≥ 0, then
√ √
{x ∈ X : (f (x))2 > α} = {x ∈ X : f (x) > a} ∪ {x ∈ X : f (x) < − a}.

Both sets on the RHS are measurable, so f 2 is measurable.

21
 (c) We have that
[
{x ∈ X : f (x) + g(x) < α} = {x ∈ X : f (x) < q} ∩ {x ∈ X : g(x) < r}.
q+r<b;q,r∈Q

The RHS is a countable union of measurable sets. Hence, f + g is measurable.

(d) f g is measurable because f g = 21 [(f + g)2 − f 2 − g 2 ].
(e) If α < 0, then {x ∈ X : f (x) > 0} = X, which is measurable. If α ≥ 0, then

{x ∈ X : |f (x)| > α} = {x ∈ X : f (x) < −α} ∪ {x ∈ X : f (x) < α}.

Both sets on the RHS are measurable. As a result, |f | is measurable.

(f ) If f (x) 6= 0 for all x, we have that

{x ∈ X : 1/b < f (x) < 0}
 if b < 0,
{x ∈ X : 1/f (x) < α} = {x ∈ X : f (x) < 0} if b = 0,

{x ∈ X : f (x) < 0} ∪ {x ∈ X : f (x) > 1/α} if b > 0.


In all cases, the RHS is a measurable set. It follows that 1/f is also measurable if f (x) 6= 0 for all x.
(g) and (h) We have that

{x ∈ X : min(f, g)(x) < α} = {x ∈ X : f (x) < α} ∪ {x ∈ X : g(x) < α},

{x ∈ X : max(f, g)(x) < α} = {x ∈ X : f (x) < α} ∩ {x ∈ X : g(x) < α}.

So min(f, g) and max(f, g) are measurable.

ˆ Definition 41. For any function f : X → R, let f + and f − be non-negative functions defined by:

f + (x) = max(f (x), 0) f − (x) = max(−f (x), 0).

We call f + the positive part of f , and f − the negative part.

ˆ It should be clear that, if f is measurable, then f + and f − are also measurable.

4.2 Measurable Extended Real-Valued Functions

ˆ Working with functions with extended real values makes it more convenience to work with limits of
sequences of functions.

ˆ Definition 42. An extended real-valued function on X is X -measurable if x ∈ X : f (x) > α ∈ X for

all α ∈ R. The collection of extended real-valued X -measurable functions is denoted by M (X, X ).
ˆ If f ∈ M (X, X ), then
∞
\
{x ∈ X : f (x) = ∞} = {x ∈ X : f (x) > n}
n=1
\∞
{x ∈ X : f (x) = −∞} = {x ∈ X : f (x) < −n},
n=1

so these sets are also in X automatically.

22
ˆ Lemma 43. An extended real-valued function f is measurable if and only if (1) the sets
A = {x ∈ X : f (x) = ∞},
B = {x ∈ X : f (x) = −∞}
are measurable, and (2) the real-valued function f1 defined by
(
f (x), x ∈ A ∪ B,
f1 (x) =
0, x 6∈ A ∪ B

is measurable.
Proof. TODO

ˆ The consequence to the last lemma and the lemmas in the last section is that, if f is M (X, X ), then
the functions
cf, f 2 , |f |, f + , f −
are also in M (X, X ).
ˆ For cf , we use the convention that 0(±∞) = 0.
ˆ For f + g, note that the sum is not well-defined when f (x) and g(x) are infinities with different signs.
ˆ Lemma 44. Let (fn ) be a sequence of functions in M (X, X ). Define
f (x) = inf {fn (x)},
n≥1

F (x) = sup {fn (x)},

n≥1
∗
f (x) = lim inf {fn (x)},
n→∞
∗
F (x) = lim sup {fn (x)}.
n→∞

Then f , F , f ∗ , F ∗ all belong to M (X, X ).

Proof. TODO

ˆ Corollary 45. If (fn ) is a sequence in M (X, X ) which converges to f on X, then f is also in M (X, X ).
Proof. We have that f (x) = lim fn (x) = lim inf fn (x).

ˆ Lemma 46. If f, g ∈ M (X, X ), then f g ∈ M (X, X ) too.

Proof. TODO

ˆ The following lemma establishes the fact that a measurable non-negative function can be approximated
by an increasing sequence of functions, all of which takes on a finite number of real values.
Lemma 47. If f is non-negative function in M (X, X ), then there exists a sequence (ϕn ) in M (X, X )
such that:
(a) 0 ≤ ϕn (x) ≤ ϕn+1 (x) for all x ∈ X and n ∈ N.
(b) f (x) = lim ϕn (x) for each x ∈ X.
(c) Each ϕn has only a fininte number of real values.
Proof. TODO

23
5 Integration
ˆ In this section, we consider a fixed measure space (X, X , µ).
ˆ Denote the set of all X -measurable functions by M = M (X, X ). Denote the set of all non-negative
X -measurable functions by M + = M + (X, X ).
ˆ Definition 48. A real-valued function is simple if it has only a finite number of values.
ˆ A simple measurable function ϕ can be represented by
n
X
ϕ= aj χEj
j=1

where aj ∈ R, and χEj is the characteristic function fo a set Ej in X .

ˆ The standard representation is the representation where the S aj ’s are distinct, and the Ej ’s are
n
disjoint, non-empty subsets of X. Moreover, we require that X = j=1 Ej .

ˆ Definition 49. If ϕ is a simple function in M + (X, X ) with the standard representation ϕ =

P
aj χEj ,
we define the integral of ϕ with respect to µ to be the extended real number
Z Xn
ϕ dµ = αj µ(Ej ).
j=1

ˆ Note that the value of the integral is always well defined because all the aj ’s are positive, so we never
encounter an expression of the form ∞ + (−∞).
ˆ Lemma 50. If ϕ and ψ are simple functions in M + (X, X ) and c ≥ 0, then
Z Z
cϕ dµ = c ϕ dµ,
Z Z Z
(ϕ + ψ) dµ = ϕ dµ + ψ dµ.

Proof. TODO

ˆ Lemma 51. Let ϕ be a simple function in M + (X, X ). For each E ∈ X , define λ(E) to be
Z
λ(E) = ϕ χE dµ.

Then, λ is a measure on X .
Proof. TODO

ˆ Definition 52. If f belongs to M + (X, X ), define the (Lebesgue) integral of f with respect to µ
to be the extended real number
Z Z 
f dµ = sup ϕ dµ ϕ ∈ M + (X, X ) is simple, and ϕ(x) ≤ f (x) for all x ∈ S

For any E ∈ X , define the (Lebesgue) integral of f over E with respect to µ to be the extended
real number
Z Z
f dµ = f χE dµ.
E

24
ˆ Lemma 53. If f, g ∈ M + (X, X ), then
R R
f dµ ≤ g dµ.
Proof. TODO

ˆ Lemma 54. If f ∈ M + (X, X ), E, F ∈ X , and E ⊆ F , then

R R
E
f dµ ≤ F
f dµ.
Proof. TODO

ˆ Theorem 55 (Monotone Convergence Theorem). If (fn ) is a monotonically increasing sequence

of functions in M + (X, X ) which converges to f , then
Z Z
f dµ = lim fn dµ.
n→∞

Proof. TODO

ˆ Corollary 56. If f belongs to M + (X, X ) and c ≥ 0, then cf ∈ M + (X, X ), and

Z Z
cf dµ = c f dµ.

Proof. TODO

ˆ Corollary 57. If f and g belong to M + (X, X ), then f + g ∈ M + (X, X ), and

Z Z Z
(F + g) dµ = f dµ + g dµ.

Proof. TODO

ˆ Theorem 58 (Fatou’s Lemma). If (fn ) is a sequence of functions in M + (X, X ), then

Z   Z
lim inf fn dµ ≤ lim inf fn dµ.
n→∞ n→∞

Proof. TODO

ˆ Corollary 59. If f belongs to M + (X, X ), and λ is defined on X by

Z
λ(E) = f dµ,
E

then λ is a measure.
Proof. TODO

ˆ Corollary
R 60. Let f belong to M + (X, X ). Then, f (x) = 0 µ-almost everyone on X if and only if
f dµ = 0.
Proof. TODO

ˆ Corollary 61. Let f belong to M + (X, X ), and let λ be defined as in Corollary 59. Then, the measure
λ is absolutely continuous with respect to µ in the sense that if E ∈ X , then µ(E) = 0, then λ(E) = 0.
Proof. TODO

25
 ˆ Corollary 62. If (fn ) is a monotonically increasing sequence of functions in M + (X, X ) which con-
verges µ-almost everywhere on X to a function f in M + (X, X ), then
Z Z
f dµ = lim fn dµ.
n→∞

Proof. TODO

ˆ Corollary 63. If (gn ) be a sequence of functions in M + (X, X ), then

Z X ∞  X∞ Z 
gn dµ = gn dµ .
n=1 n=1

Proof. TODO

6 Integrable Functions
ˆ Definition 64. The collection L = L(X, X , µ) of integrable functions consists of all real-valued
X -measurable functions f defined on X, such that both the positive and negative parts (f + and f − ) of
f have finite integrals with respective to µ.
ˆ Definition 65. If f ∈ L(X, X , µ), the integral of f with respect to µ is defined to be
Z Z Z
f dµ = f + dµ + f − dµ.

If E ∈ X , define
Z Z Z
f dµ = f + dµ + f − dµ.
E E E

ˆ Definition 66. Let X be a σ-algebra on a set X. A function µ : X → R is said to be a charge on X

if the following properties are satisfied.
1. µ(∅) = 0.
2. µ is countably additive. This is, for a sequence (En ) of disjoint sets, it holds that
[ ∞  X ∞
µ En = µ(En ).
n=1 n=1

ˆ The difference between a charge and a measure is that a measure is always non-negative, but a charge
can be negative. Moreover, a charge cannot take infinite values.
ˆ Lemma 67. If f belongs to L, and λ(E) = E f dµ, then λ is a charge.
R

Proof. TODO

ˆ The function λ defined as in the above lemma is often called the indefinite integral of f with
respect to µ.
ˆ Since λ is a charge, if (En ) is a disjoint sequence of sets in X whose union is E, then
Z X∞ Z
f dµ = f dµ.
E n=1 En

In other words, the indefinite integral of a function in L is countably additive.

26
 ˆ Theorem 68. A measurable function f belongs fo L if and only if |f | belons to L. Moreover,
Z Z
f dµ ≤ |f | dµ.

Proof. f belongs to L if and only if f + and f − belong to M + . Now, we have that |f |+ = |f | = f + +f − ,

and |f |− = 0. Hence, |f | also belongs to L. The converse can be proven in a similar fasion. Moreover,
Z Z Z Z Z Z
+ − + −
f dµ = f dµ − f dµ ≤ f dµ + f dµ = |f | dµ

as required.

ˆ Corollary 69. If f is measurable, g is integrable, and |f | ≤ |g|, then f is integrable, and

Z Z
|f | dµ ≤ |g| dµ.

Proof. First, we have that |f | ∈ M + , so |f | dµ is well defined. Moreover,

R
R we know
R that it has a finite
value because we can apply Lemma 46 to |f | and |g| to conclude that |f | dµ ≤ |g| dµ. Now, we can
apply Theorem 68 to conclude that f is also integrable.

ˆ Theorem 70. A constant multiple cf and a sum f + g of integrable functions are integrable, and the
integrals are given by:
Z Z
cf dµ = c f dµ,
Z Z Z
(f + g) dµ = f dµ + g dµ.

Proof. TODO

ˆ Theorem 71 (Lebesgue dominated convergence theorem). Let (fn ) be a sequence of integrable

functions which converges almost everywhere to a real-valued measurable function f . If there exists an
integrable function g such that |fn | ≥ g for all n, then f in integrable and
Z
f dµ = lim fn dµ.
n→∞

Proof. TODO

7 Integrals That Involve Limits

ˆ In this section, let f denote a function with signature X × [a, b] → R. Also, assume that f (x, t) is
X -measurable for each t ∈ [a, b].
ˆ Corollary 72. Suppose that, for some t0 in [a, b], we have that

f (x, t0 ) = lim f (x, t)

t→t0

for each x ∈ X, and that there exists an integrable function g such that |f (x, t)| ≤ g(x). Then,
Z Z
f (x, t0 ) dµ(x) = lim f (x, t) dµ(x).
t→t0

27
 Proof. Just apply the dominated convergence theorem.

ˆ Corollary 73. Let the function t 7→ f (x, t) be continuous on [a, b] for all x ∈ X. If there is an
integrable function g on X such that |f (x, t)| ≤ g(x), then the function F defined by
Z
F (t) = f (x, t) dµ(x)

is continuous for t ∈ [a, b].

Proof. Apply the last corollary.

ˆ Corollary 74. Suppose that for some t0 ∈ [a, b], the function x 7→ f (x, t0 ) is integrable on X. Suppose
that ∂f /∂t exists on X × [a, b]. Moreover, let there be an integrable function g on X such that

∂f
(x, t) ≤ g(x).
∂t

Then, the function Z

F (t) = f (x, t) dµ(x)

is differentiable on [a, b], and

Z Z
dF d ∂f
(t) = f (x, t) dµ(x) = (x, t) dµ(x).
dt dt ∂t
Proof. TODO

ˆ Corollary 75. Let the function t 7→ f (x, t) be continuous on [a, R b] for all x ∈ X. Let there be an
integrable function g on X such that |f (x, t)| ≤ g(x). Let F (t) = f (x, t) dµ(x). We have that
Z b Z b Z  Z Z b 
F (t) dt = f (x, t) dµ(x) dt = f (x, t) dt dµ(x)
a a a

where the integrals with respect to t are Reimann integrals.

Proof. TODO

8 Product Measures and Double Integrals

ˆ In this section, we show that the Cartesian product of two measurable spaces can be made into a
measurable space in a natural fashion.
ˆ Note that this gives us another way to construct a measure on Rk .

– If we force the measure to agree on rectangles, though, the measure would agree with the Lebesgue
measure on the Borel sets in Rk by the Hahn’s extension theorem.
ˆ Definition 76. If (X, X ) and (Y, Y) are measurable spaces, then the set of the form A × B with A ∈ X
and B ∈ B are called a measurable rectangle or simply rectangle in Z = X × Y . Let Z0 denote
the set of all finite unions of rectangles in Z.
ˆ It can be shown that each element of Z0 can be written as a finite union of disjoin rectangles in Z.
(Just use the inclusion–exclusion principle.)

28
ˆ Lemma 77. The collection Z0 is an algebra on Z.
Proof. (∅ ∈ Z0 ) We have that ∅ ∈ X and ∅ ∈ Y. So, ∅ = ∅ × ∅ ∈ Z0 .
(Z ∈ Z0 ) Since X ∈ X and Y ∈ Y, we have that Z = X × Y ∈ Z0 .
(Closure under complementation) If C ∈ Z0 , then C = A × B for some A ∈ X and B ∈ Y. It
follows that C c = (Ac × B) ∪ (A × B c ) ∪ (Ac × B c ), which is a finite unions of rectangles in Z, so
C c ∈ Z0 .
Closure under finite union follows from definition of Z0 .

ˆ Definition 78. Let (X, X ) and (Y, Y) be measurable spaces. Let Z = X × Y denote the σ-algebra of
subsets of Z = X × Y generated by rectangles A × B with A ∈ X and B ∈ Y. A set in Z is called a
Z-measurable set or as a measurable subset of Z.
ˆ Theorem 79 (Product measure theorem). Let (X, X , µ) and (Y, Y, ν) be measure spaces. There
exists a measure π defined on Z = X × Y with
π(A × B) = µ(A) ν(B)
for all A ∈ X and B ∈ Y. If the measure is σ-finite, then then it is unique. The measure π is called
the product of µ and ν.
Proof. TODO

ˆ Definition 80. If E is a subset of Z = X × Y , and x ∈ X, then the x-section of E is the set

Ex = {y ∈ Y : (x, y) ∈ E}.
Similary, the y-section is the set
E y = {x ∈ X : (x, y) ∈ E}.

ˆ Lemma 81. If E is a measurable subset of E, then every section of E is measurable.

Proof. TODO

ˆ Definition 82. For any f : Z → R, and x ∈ X, the x-section of f is the function fx defined on Y as
fx (y) = f (x, y)
for all y ∈ Y . The y-section is defined as
f y (x) = f (x, y)
for all x ∈ X.
ˆ Lemma 83. If f : Z → R is a measurable function, then every section of f is measurable.
Proof. TODO

ˆ Lemma 84. Let (X, X , µ) and (Y, Y, ν) be σ-finite measure spaces. If E ∈ Z = X × Y, then the;
functions defined by
f (x) = ν(Ex ),
g(y) = µ(E y )
are measurable and
Z Z
f dµ = π(E) = g dν.
X Y

29
 Proof. TODO. This requires the monotone class theorem (Theorem 23).

ˆ Theorem 85 (Tonelli’s theorem). Let (X, X , µ) and (Y, Y, ν) be σ-finite measure spaces. Let π be
the product of µ and ν, which is a measure on Z = X ×Y. Let F : Z → R be a non-negative measurable
function. Then, the functions defined on X and Y by
Z
f (x) = Fx dν,
Y
Z
g(y) = F y dµ
X

are measurable. Moreover,

Z Z Z
f dµ = F dπ = g dν.
X Z Y

Equivalently,
Z Z  Z Z Z 
Fx dν dµ = F dπ = F y dµ dν.
X Y Z Y X

Proof. TODO. This requires the monotone class theorem (Theorem 23).

ˆ Theorem 86 (Fubini’s theorem). Let (X, X , µ) and (Y, Y, ν) be σ-finite measure spaces. Let π be
the product of µ and ν, which is a measure on Z = X × Y. Let F : Z → R be an integrable function
with respect to π. Then, the extended real value functions defined almost everywhere by
Z
f (x) = Fx dν,
Y
Z
g(y) = F y dµ
X

have finite integrals. Moreover,

Z Z Z
f dµ = F dπ = g dν.
X Z Y

Equivalently,
Z Z  Z Z Z 
Fx dν dµ = F dπ = F y dµ dν.
X Y Z Y X

Proof. TODO

9 Signed Measures and Their Decompositions

ˆ Definition 87. Let (X, X ) be a measurable space. A function λ : X → R is said to be a signed
measures if the following conditions are satisfied.
(a) λ(∅) = 0.
(b) λ attains at most one of the values ∞ and −∞.
(c) λ is countably additive. In other words, if (An ) is a sequence of disjoint sets in X , then
[ ∞  X ∞
λ An = µ(An ).
n=1 n=1

30
ˆ The condition (b) is introduced to avoid undefined quantities such as ∞ − ∞.
ˆ A signed measure is very similar to a charge, but it can take an infinite value while a charge cannot.
Note that a charge is a signed measure, but not the other way around.
ˆ It can be shown that, if (En ) is an increasing sequence of sets in X , then
∞
[ 
λ En = lim λ(En ).
n→∞
n=1

Similarly, if (Fn ) is a decreasing sequence of sets in X , then

∞
\ 
λ Fn = lim λ(Fn ).
n→∞
n=1

ˆ Definition 88. Let λ be a signed measure on (X, X ). A set P in X is said to be positive with respect
to λ if λ(E ∩ P ) ≥ 0 for any E ∈ X . A set N ∈ X is said to be negative with respect to λ if
λ(E ∩ N ) ≤ 0 for all E ∈ X . A set M ∈ X is said to be a null set for λ if λ(E ∩ M ) = 0 for all
E ∈ X.
ˆ It can be shown that a measurable subset of a positive set is positive, and the union of two positive
sets is also positive.
ˆ Theorem 89 (Hahn decomposition theorem). If λ is a signed measure on X , then there exist
sets P and N in X with X = P ∪ N , P ∩ N = ∅, and such that P is positive and N is negative with
respect to λ.
Proof. TODO

ˆ A pair (P, N ) of measurable sets in the above theorem is said to form a Hahn decomposition of X
with respect to λ.
ˆ In general, Hahn decompositions are not unique. In fact, if M is a null set for λ, then (P ∪ M, N − M )
and (P − M, N ∪ M ) are also Hahn decompositions of X w.r.t λ.
ˆ Lemma 90. If (P1 , N1 ) and (P2 , N2 ) are Hahn decompositions of X w.r.t λ, and E ∈ X , then

λ(E ∩ P1 ) = λ(N ∩ P2 ),
λ(E ∩ N1 ) = λ(N ∩ N2 ).

Proof. TODO

ˆ Definition 91. Let λ be a signed measure on X , and let (P, N ) be a Hahn decomponent w.r.t λ. The
positive and negative variations of λ are the non-negative measures λ+ , λ− defined by:

λ+ (E) = λ(E ∩ P ),
λ− (E) = −λ(E ∩ N ).

The total variation of λ, denoted by |λ|, is defined as:

|λ|(E) = λ+ (E) + λ− (E).

ˆ By Lemma 90, the positive and negative variations are well defined and do not depend on the particular
choice of Hahn decomposition.

31
 ˆ We also have that
λ(E) = λ+ (E) − λ− (E),
and this result is encapsulated in the following theorem.
Theorem 92 (Jordan decomposition theorem). If λ is a signed measure on X , then there exists
measures λ+ and λ− on X , at least one of which is finite, such that the following conditions are
satisifed.
(a) λ = λ+ − λ− .
(b) There exists sets M and N such that M ∪ N = X, M ∩ N = ∅, λ+ (N ) = 0, and λ− (M ) = 0.
(c) If λ = µ − ν where µ and ν are measures, then µ(E) ≥ λ+ (E) and ν(E) ≥ λ− (E).
Proof. TODO

ˆ Let f be an integrable function with respect to a measure µ on X . We can define

Z
λ(E) = f dµ.
E

Lemma 67 tells us that λ is a charge.

ˆ Lemma 93. If f ∈ L(X, X , µ) and λ(E) = E f dµ, then λ+ , λ− and |λ| are given by:
R
Z Z Z
λ+ (E) = f + dµ, λ− (E) = f − dµ, |λ|(E) = |λ| dµ.
E E E

Proof. TODO

10 Radon–Nikodym Theorem
ˆ From Corollary 59, we can create a measure λ from a non-negative extended real-valued measurable
function f by integrating with respect to an existing measure.
ˆ The Radon–Nikodym theorem is the converse of the above corollary. It indicates when a measure λ
can be expressed as an integration of a function f with respect to µ. Hence, it is useful in defining
probability density functions.
ˆ A necessary and sufficient condition for the theorem to hold is given below.
Definition 94. A measure λ on X is said to be absolutely continuous with respect to a measure
µ on X if λ(E) = 0 for all set E ∈ X such that µ(E) = 0. In this case, we write λ  µ. A signed
measure λ is absolsution continuous with respect to a signed measure µ if the total variation |λ| is
absolutely continuous with respect to |µ|.

ˆ Lemma 95. Let λ and µ be finite measures on X . Then, λ  µ if and only if, for every ε > 0, there
exists a δ(ε) > 0 such that λ(E) < ε for all E such that µ(E) < δ(ε).
Proof. TODO

ˆ Theorem 96 (Random–Nikodym theorem). Let λ and µ be σ-finite measures defined on X , and

suppose that λ is absolutely continous with respect to µ. Then, there exists a function f ∈ M + (X, X )
such that
Z
λ(E) = f dµ.
E

Moreover, the function f is uniquely determined µ-almost everywhere.

32
 Proof. TODO

ˆ The function f above is often called the Random–Nikodym derivative of λ with respective to µ,
and it is denoted by dλ/dµ.

References
[Bartle, 1995] Bartle, R. G. (1995). The Elements of Integration and Lebesgue Measure. John Wiley & Sons,
New York.
[Billingsley, 1995] Billingsley, P. (1995). Probability and Measure. John Wiley and Sons, third edition.
[Hunter, 2011] Hunter, J. K. (2011). Measure theory. https://www.math.ucdavis.edu/~hunter/measure_
theory/measure_notes.pdf. Accessed: 2021-12-28.

33