You are on page 1of 50

Measure theory (MA 550) lecture notes

Jan-April 2012 semester, IIT Guwahati


Shyamashree Upadhyay

Contents
1 Algebras and σ-algebras 2
1.1 Construction of σ-algebras . . . . . . . . . . . . . . . . . . . . 3
1.2 The Borel σ-algebra . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Measures 6
2.1 Properties of measures . . . . . . . . . . . . . . . . . . . . . . 6

3 Outer Measures 8
3.1 Measurable sets . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.2 Properties of the lebesgue measure . . . . . . . . . . . . . . . 16
3.3 Completeness of a measure . . . . . . . . . . . . . . . . . . . . 19

4 Measurable functions 20
4.1 Some basic facts about measurable functions . . . . . . . . . . 21
4.2 Example of a lebesgue measurable, non-borel set . . . . . . . . 27

5 The integral 28
5.1 Properties of integral of non-negative simple functions . . . . . 29
5.2 Properties of integral of [0, ∞]-valued measurable functions . . 31
5.3 Properties of integral of real-valued measurable functions . . . 32
5.4 Properties of integral of [−∞, ∞]-valued measurable functions 33

6 Limit theorems 34

7 Modes of Convergence 36

1
8 Signed measures 37
8.1 Some Properties of signed measures . . . . . . . . . . . . . . . 38

9 Complex measures 40
9.1 Integration with respect to a finite signed or complex measure 41

10 Absolute Continuity 42

11 Functions of Bounded variation 44


11.1 Some general properties of functions of bounded variation . . . 45

12 Absolute continuity of functions 46

13 Product measures 47
13.1 Product of finite number of measure spaces . . . . . . . . . . . 50

1 Algebras and σ-algebras


Definition 1.0.1. Let X be an arbitrary set. A collection A of subsets of
X is called an algebra (sometimes also called a field ) on X if the following 3
conditions are satisfied simultaneously:
(a) X ∈ A.
(b) For each A ∈ A, Ac ∈ A.
(c) For each finite sequence A1 , A2 , . . . , An of sets that belong to A, the set
∪ni=1 Ai belongs to A. 

Note that in definition 1.0.1 above, condition (c) is equivalent to the


following condition:
(d) For each finite sequence A1 , A2 , . . . , An of sets that belong to A, the set
∩ni=1 Ai belongs to A.

Definition 1.0.2. Let X be an arbitrary set. A collection A of subsets of X


is called a σ-algebra (sometimes also called a σ-field ) on X if the following 3
conditions are satisfied simultaneously:
(a) X ∈ A.
(b) For each A ∈ A, Ac ∈ A.
(c) For each infinite sequence {Ai }∞ ∞
i=1 of sets that belong to A, the set ∪i=1 Ai
belongs to A. 

2
Note that in definition 1.0.2 above, condition (c) is equivalent to the
following condition:
(d) For each infinite sequence {Ai }∞ ∞
i=1 of sets that belong to A, the set ∩i=1 Ai
belongs to A.
Remark 1.0.3. (i) The empty set belongs to any algebra or σ-algebra.
(ii) Each σ-algebra is an algebra on X. But the converse is NOT true in general,
see item no. 4 in example(s) 1.0.4 below.
(iii) Condition (a) in both the definitions above can be replaced by saying that
‘A is non-empty’.
Example(s) 1.0.4. 1. Let X be any set and A be the power set P(X) of
X. Then A is a σ-algebra on X.
2. Let X be any set and A = {∅, X}. Then A is a σ-algebra on X.
3. Let X be an infinite set and A be the collection of all finite subsets of X.
Then A is NOT an algebra (because X ∈ / A) and hence NOT a σ-algebra.
4. Let X be an infinite set and A be the collection of all subsets A of X
such that either A or Ac is finite. Check that A is an algebra but is NOT a
σ-algebra. This is an exercise.
5. Let X be any set and A be the collection of all subsets A of X such that
either A or Ac is countable. Check that A is a σ-algebra.
6. X an uncountable set. Let A be the collection of all countable subsets of
X. Then A is NOT an algebra (because X ∈ / A) and hence NOT a σ-algebra.
7. Let X = R, A = {A ⊂ R|A is a union of f initely many intervals of the type
(a, b], (a, ∞) or (−∞, b]}. It is easy to check that each set that belongs to A
is the union of a finite disjoint collection intervals of the 3 types mentioned
above. Check that A is an albegra over X, but it is NOT a σ-algebra on X
(because intervals of the type (c, d) are unions of sequences of sets belonging
to A, but do not themselves belong to A). 

1.1 Construction of σ-algebras


In this subsection, we will see how to construct a σ-algebra out of a given
(arbitrary) collection F of subsets of a given set X.
Theorem 1.1.1. Let X be a set and F be an arbitrary collection of subsets
of X. Then there exists a smallest σ-algebra on X that contains F .
Proof: Let C be the collection of all σ-algebras on X that contain F .
Clearly then C is non-empty since it contain the σ-algebra P(X) which is the

3
power set of X. Take the intersection of all σ-algebras in C, this intersection
will be a σ-algebra (see lemma 1.1.2 below for a proof), call it D. It is now
easy to check that D has the required properties. 

Lemma 1.1.2. Let X be a set. The intersection of an arbitrary collection


of σ-albegras on X is a σ-algebra on X.
Proof: Exercise. 

Definition 1.1.3. Given a set X and an arbitrary collection F of subsets


of X. The smallest σ-algebra on X that contains F is unique (the proof of
uniqueness follows from the proof of theorem 1.1.1 above) and is called the
σ-algebra generated by F ., often denoted by σ(F ).
By the phrase ‘smallest σ-algebra on X that contains F ’, we mean a σ-
algebra on X that includes F and every σ-algebra on X that includes F also
includes it. 
Remark 1.1.4. A natural question that arises after looking at the statement
of lemma 1.1.2 is that if we replace the term ‘intersection’ by the term ‘union’
in the statement of lemma 1.1.2, does the new statement hold true?
The answer to this is NO in general. As an example, take X = {1, 2, 3},
C1 = {X, ∅, {1}, {2, 3}} and C2 = {X, ∅, {2}, {1, 3}}. Then check that both C1
and C2 are σ-algebras but C1 ∪ C2 is NOT.

1.2 The Borel σ-algebra


In this subsection, we will see discuss an important example of a σ-algebra
on Rd where d is any fixed natural number.
Definition 1.2.1. Let F (Rd ) be the collection of all open subsets of Rd . Let
B(Rd ) := σ(F (Rd )). We call B(Rd ) the Borel σ-algebra on Rd . Elements of
B(Rd ) are called the Borel subsets of Rd . We denote the σ-algebra B(R1 ) by
B(R). 
Proposition 1.2.2. The σ-algebra B(R) is generated by each of the following
collection of sets:—
(a) the collection F1 of all closed subsets of R.
(b) the collection F2 of all subintervals of R of the form (−∞, b].
(c) the collection F3 of all subintervals of R of the form (a, b].

4
Proof: Let Bi := σ(Fi ) for all i ∈ {1, 2, 3}. We will prove this proposition
by showing that B(R) ⊇ B1 ⊇ B2 ⊇ B3 ⊇ B(R).
Since B(R) contains the family of all open subsets of R and is closed
under complementation, it follows that B(R) ⊇ F1 . It then follows from
lemma 1.2.3 below that B(R) ⊇ B1 .
The sets of the form (−∞, b] are closed and hence belong to F1 . So
F1 ⊇ F2 and hence it follows from lemma 1.2.3 below that B1 ⊇ B2 .
Since (a, b] = (−∞, b] ∩ ((−∞, a])c , it follows by similar arguments as
before that B2 ⊇ B3 . It now remains to show that B3 ⊇ B(R).
Note that each open interval of R can be wriiten as a countable union
1
of sets from the collection F3 , because (c, d) = ∪∞n=n0 (c, d − n ] where n0 is
a fixed positive integer. Also each open subset of R is a countable (it is an
exercise to show why we can take a countable union here) union of open in-
tervals of R. Thus each open subset of R belongs to B3 . Hence B3 ⊇ B(R). 

Lemma 1.2.3. Let X be a set. Then


(a) If G, G ′ are two collections of subsets of X such that G ⊇ G ′ , then σ(G) ⊇
σ(G ′ ).
(b) If A is a σ-algebra on X which contains a collection G of subsets X, then
A ⊇ σ(G).

Proof: Exercise. 
Observe that proposition 1.2.2 above can be extended naturally to the Borel
σ-algebra on Rd for any natural number d. Hence we have the following
proposition:—

Proposition 1.2.4. The σ-algebra B(Rd ) is generated by each of the follow-


ing collection of sets:—
(a) the collection F1 of all closed subsets of Rd .
(b) the collection F2 of all subsets of Rd of the form {(x1 , . . . , xd )|xi ≤ b for
some b ∈ R and some i ∈ {1, . . . , d}}.
(c) the collection F3 of all subsets of Rd of the form {(x1 , . . . , xd )|a < xi ≤ b
for some a, b ∈ R such that a < b and some i ∈ {1, . . . , d}}.

Proof: Exercise. 

5
2 Measures
Definition 2.0.5. Let X be a set and A be a σ-algebra on X. A function
µ : A → [0, ∞] is said to be countably additive if
µ(∪∞ ∞
i=1 Ai ) = Σi=1 µ(Ai )

for each countable collection {Ai }∞


i=1 of disjoint sets that belong to A. 
Definition 2.0.6. Let X be a set and A be a σ-algebra on X. A measure
on (X, A) is a function µ : A → [0, ∞] that satisfies both of the following
properties:— (i) µ(∅) = 0. (ii) µ is countably additive. 
Definition 2.0.7. Let X be a set and A be a σ-algebra on X. The pair
(X, A) is called a measurable space. Let µ be a measure on (X, A). The
triple (X, A, µ) is called a measure space. 
Example(s) 2.0.8. 1. X anyset, A a σ-algebra on X. Let µ : A → [0, ∞]
n if A is finite and |A| = n
be defined as follows: µ(A) =
∞ otherwise
Then µ is a measure, called the counting measure on (X, A).
2. Let X be a nonempty set and A be a σ-algebra  on X. Let x ∈ X. Let
1 if x ∈ A
δx : A → [0, ∞] be defined as follows: δx (A) =
0 otherwise
Then δx is a measure, called the point mass measure on (X, A) concen-
trated at x.
3. X any set, A a σ-algebra on X. Let µ : A → [0, ∞] be defined as follows:
0 if A = ∅
µ(A) =
∞ otherwise
Check that µ is a measure.
4. Let X be a set with at least 2 elements in it. Consider the σ-algebra
P(X) ofthe powers et of X. Let µ : P(X) → [0, ∞] be defined as follows:
0 if A = ∅
µ(A) =
1 otherwise
Check that µ is NOT a measure.


2.1 Properties of measures


Proposition 2.1.1. Let (X, A, µ) is be a measure space. Let A, B ∈ A be
such that A ⊆ B. Then µ(A) ⊆ µ(B). If in addition, the subset A satisfies
the property that µ(A) < ∞, then µ(B − A) = µ(B) − µ(A).

6
Proof: The sets A and B − A(= B ∩ Ac ) are disjoint and both belong to
A. Now use the countable additivity of µ. 

Remark 2.1.2. The property of µ as metioned in Proposition 2.1.1 above (i.e.,


if A ⊆ B, then µ(A) ⊆ µ(B)) is called monotonicity of µ.

Remark 2.1.3. In Proposition 2.1.1 above, the extra condition that µ(A) < ∞
is given because it might sometimes happen that µ(B −A) < ∞ but µ(B), µ(A)
are both equal to ∞. For example, let X = R, A = B(R) and µ be the length
measure on X. Take B = R, A = the set of all irrationals in R. Then B −A = Q
which a countable subset of R, hence µ(B − A) = 0. But µ(B) = µ(A) = ∞.

Proposition 2.1.4. Let (X, A, µ) is be a measure space. If {Ak } is an


arbitrary (not necessarily disjoint) sequence of sets in A, then

µ(∪∞ ∞
k=1 Ak ) ≤ Σk=1 µ(Ak )

Proof: Let B1 := A1 and Bk := Ak − (∪i = 1k−1 Ai ) for all k ≥ 2. Then


each Bk ∈ A, ∪k Bk = ∪k Ak , Bk ’s are disjoint and Bk ⊆ Ak ∀k. These facts
together with the properties of monotonicity and countable additivity of µ
yeild the required result. 

Proposition 2.1.5. Let (X, A, µ) is be a measure space. A sequence {Ak }


of sets in A is called increasing if Ak ⊆ Ak+1 ∀k ≥ 1 and decreasing if
Ak ⊇ Ak+1 ∀k ≥ 1.
(a) If {Ak } is an increasing sequence of sets in A, then µ(∪k Ak ) = limk→∞ µ(Ak ).
(b) If {Ak } is an decreasing sequence of sets in A and if µ(An ) < ∞ for some
n, then µ(∩k Ak ) = limk→∞ µ(Ak ).

Proof: (a) Let B1 := A1 and Bk := Ak − Ak−1 for all k ≥ 2. Then each


Bk ∈ A, Bk ’s are disjoint, Ak = ∪kj=1 Bj and ∪k Bk = ∪k Ak . Hence µ(∪k Ak ) =
µ(∪k Bk ) = Σ∞ k k
j=1 µ(Bj ) = limk→∞ (Σj=1 µ(Bj )) = limk→∞ (µ(∪j=1 Bj )) =
limk→∞ µ(Ak ). (The disjointness of the Bk ’s and the countable additivity
of µ have been used here for the 2nd and the 4-th equality.)
(b) Without loss of generality, we may assume that µ(A1 ) < ∞. Let Ck :=
A − 1 − Ak ∀ k. Then {Ck } is an increasing sequence of sets in A and

7
∪∞ ∞
k=1 Ck = A1 − (∩k=1 Ak ). Hence by part(a) of this proposition, we have
µ(∪k Ck ) = limk→∞ µ(Ck ).
Hence µ(A1 − (∩∞ k=1 Ak )) = limk→∞ µ(A1 − Ak ). Now since µ(A1 ) < ∞,
Ak ⊆ A1 ∀ k and ∩k Ak ⊆ A1 , we can apply the result present in the last line
of proposition 2.1.1, which will give us the required result. 
Exercise: Give an example where part (b) of proposition 2.1.5 above does
not hold true, if the condition that µ(An ) < ∞ for some n is violated.

Definition 2.1.6. A measure on (Rd , B(Rd )) is called a Borel measure on Rd .


More generally, if X is a borel subset of Rd and A := {Y ∈ B(Rd )|Y ⊆ X},
then any measure on (X, A) is called a Borel measure on X. 

3 Outer Measures
Definition 3.0.7. Let X be a set and P(X) be the power set of X. An
outer measure on X is a function µ∗ : P(X) → [0, ∞] which satisfies all the
3 properties mentioned below:—
(a) µ∗ (∅) = 0.
(b) If A ⊆ B ⊆ X, then µ∗ (A) ≤ µ∗ (B). (monotonicity of µ∗ )
(c) If {An } is an infinite sequence of subsets of X, then µ(∪∞n=1 An ) =
Σ∞n=1 µ(A n ). (countable subadditivity of µ ∗
) 

Remark 3.0.8. 1. The domain of an outer measure is not any arbitrary σ-


algebra, it is always the power set σ-algebra.
2. For any set X, a measure on P(X) is always an outer measure, but the
converse is not true in general. We will soon see some example (In fact, an outer
measure can fail to be a measure because the countable additivity might fail to
hold!).

Example(s) 3.0.9. 1. Let X be any set and µ∗ : P(X) → [0, ∞] be given


by: 
∗ 0 if A = ∅
µ (A) =
1 otherwise
Check that µ∗ is an outer measure.
2. Let X be any set and µ∗ : P(X) → [0, ∞] be given by:
0 if A is countable
µ∗ (A) =
1 otherwise

8
Check that µ∗ is an outer measure.
3. Let X be an infinite set and µ∗ : P(X) → [0, ∞] be given by:
0 if A is finite
µ∗ (A) =
1 otherwise
Check that µ∗ is NOT an outer measure.
4. The lebesgue outer measure λ∗ on R:
Given any subset A of R, let CA := the set of all sequences (possibly finite)
{(ai , bi )}i of bounded open intervals such that A ⊆ ∪i (ai , bi ).
Define λ∗ : P(R) → [0, ∞] as:

λ∗ (A) = Inf {Σi (bi − ai )|{(ai , bi )}i ∈ CA }


. Note that given any A ∈ R, the set CA on which this infimum is taken is
always non-empty. There exists some subsets A of R for which this infimum
can be equal to ∞, for example when A = R or any unbounded interval.
The function λ∗ as defined above is an outer measure on R and it assigns
to each subinterval of R its length (this statement needs a proof which is
given in Proposition 3.0.10 below).
5. The lebesgue outer measure λ∗d on Rd :
A bounded open d-dimensional interval of Rd is of a subset of Rd of the
form I1 × · · · × Id where Ij is an interval of the type (aj , bj ) for each j ∈
{1, . . . , d}. If R is any bounded open d-dimensional interval of Rd given by
R = (a1 , b1 ) × · · · × (ad , bd ), then the volume V ol(R) of R is defined as the
product Πdj=1(bj − aj ).
Given any subset A of Rd , let CA := the set of all sequences (possibly finite)
{Ri } of bounded open d-dimensional intervals of Rd such that A ⊆ ∪i Ri .
Define λ∗d : P(Rd ) → [0, ∞] as:

λ∗d (A) = Inf {Σi V ol(Ri )|{Ri }i ∈ CA }


.
A d-dimensional interval of Rd is of a subset of Rd of the form I1 × · · ·× Id
where Ij is an interval for each j ∈ {1, . . . , d}. The function λ∗d as defined
above is an outer measure on Rd and it assigns to each d-dimensional interval
of Rd its volume (this statement needs a proof which is omitted because it is
similar to the proof of Proposition 3.0.10 below).


9
Proposition 3.0.10. The function λ∗ as defined in item no. 4 of example(s)
3.0.9 above is an outer measure on R and it assigns to each subinterval of R
its length.

Proof: Since λ∗ is a non-negative function, we have λ∗ (∅) ≥ 0. On the other


hand, given any ǫ > 0, consider the finite (singleton) sequence {(−ǫ/2, ǫ/2)}
of bounded open interval(s) of R which contains the set ∅. Clearly then
{(−ǫ/2, ǫ/2)} ∈ C∅ . It now follows from the definition of λ∗ (the fact that it
is the infimum of a set) that λ∗ (∅) ≤ ǫ. Since the ǫ > 0 that we began with
was arbitray and 0 ≤ λ∗ (∅) ≤ ǫ, it follows that λ∗ (∅) = 0.
Now let A ⊆ B ⊆ R. It is easy to see that CB ⊆ CA and hence it follows
that λ∗ (A) ≤ λ∗ (B).
Now let {An }n be a sequence of subsets of R. We will show that λ∗ (∪n An ) ≤
Σn λ∗ (An ) and that will finish the proof of the fact that λ∗ is an outer mea-
sure. If Σn λ∗ (An ) = ∞, there is nothing to prove. So let us assume that
Σn λ∗ (An ) < ∞ and prove that λ∗ (∪n An ) ≤ Σn λ∗ (An ).
Let ǫ > 0 be arbitrary. For each fixed n ∈ N, it follows from the defini-
tion of λ∗ (An ) that there exists a sequence {(an,i , bn,i )}i (possibly finite) of
bounded open intervals of R such that An ⊆ ∪i (an,i , bn,i ) and Σi (bn,i − an,i ) <
λ∗ (An ) + 2ǫn .
We can combine these sequences {{(an,i , bn,i )}i }∞ n=1 to form a single se-
quence {(aj , bj )}j which will satisfy the following property: ∪n (An ) ⊆ ∪j (aj , bj )
and Σj (bj − aj ) < Σn [λ∗ (An ) + 2ǫn ] = Σn λ∗ (An ) + ǫ.
It follows from the definition of λ∗ (the fact that it is the infimum of a set)
that λ∗ (∪n An ) ≤ Σj (bj −aj ). But from the last paragraph, we also know that
Σj (bj − aj ) < Σn λ∗ (An ) + ǫ. Hence we have λ∗ (∪n An ) < Σn λ∗ (An ) + ǫ and
this holds for every ǫ > 0. Taking limits as ǫ → 0, we get that λ∗ (∪n An ) ≤
Σn λ∗ (An ). Hence λ∗ is an outer measure.
We will now prove that λ∗ assigns to each subinterval if R its length.
It is enough to show this for closed intervals of the type [a, b] because any
bounded interval I (which is not necessarily of the type [a, b]) contains and
is contained in closed intervals of the type [a, b] whose lengths are arbitrarily
close to the length of I. And finally, any unbounded interval in R has infinite
outer measure because it includes arbitraily long intervals of the form [a, b].
Let us now show that λ∗ ([a, b]) = b − a for any two real numbers a, b such
that a < b.
It is easy to see that λ∗ ([a, b]) ≤ b − a because [a, b] ⊆ (a − n1 , b + n1 ) ∀ n ≥
1. It now remains to prove that λ∗ ([a, b]) ≥ b − a. Let {(ai , bi )}i be an

10
arbitrary sequence of bounded open intervals of R such that [a, b] ⊆ ∪i (ai , bi ).
Since [a, b] is compact, there exists a positive integer N such that [a, b] ⊆
∪N N
i=1 (ai , bi ). Hence it follows that b − a ≤ Σi=1 (bi − ai ) (How?→ Exercise.
Hint: Show that if for any positive integer n, [α, β] ⊆ ∪ni=1 (αi , βi ), then
β − α ≤ Σni=1 (βi − αi ). Use induction on n and the fact that if any interval
is contained in a union of certain intervals, then at least 2 intervals in that
union must overlap.).

Therefore we have, b − a ≤ ΣN i=1 (bi − ai ) ≤ Σi=1 (bi − ai ). Since the se-
quence {(ai , bi )}i of bounded open intervals of R covering [a, b] was arbitrary
and b − a ≤ Σ∞ i=1 (bi − ai ), by taking infimum on the right hand side over all
elements in C[a,b] , we get that b − a ≤ λ∗ ([a, b]). Hence proved. 

3.1 Measurable sets


Definition 3.1.1. Let X be a set and µ∗ be an outer measure on X. A
subset B of X is said to be µ∗ -measurable if

µ∗ (A) = µ∗ (A ∩ B) + µ∗ (A ∩ B c )

holds for each subset A of X. 

Note that it follows from the defining properties of an outer measure that

µ∗ (A) ≤ µ∗ (A ∩ B) + µ∗ (A ∩ B c )

holds for any two subsets A, B of X. Thus to check for the µ∗ -measurability
of a subset B of X, we only need to check that

µ∗ (A) ≥ µ∗ (A ∩ B) + µ∗ (A ∩ B c ) (3.1.1)

holds for each subset A of X. But if µ∗ (A) = ∞, the inequality 3.1.1


anyways holds true. So the µ∗ -measurability of a subset B of X can be
verified by checking that inequality 3.1.1 holds true for each A ⊆ X which
satisfies µ∗ (A) < ∞.

Definition 3.1.2. A lebesgue measurable subset of R (or of Rd ) is one that


is measurable with respect to the lebesgue outer measure λ∗ (or λ∗d , as the
case may be). 

11
Remark 3.1.3. 1. The sets ∅ and X are measurable with respect to every
outer measure µ∗ on X.
2. Let X be a set and µ∗ be an outer measure on X. Then each subset B of X
which satisfies µ∗ (B) = 0 or µ∗ (B c ) = 0 is µ∗ -measurable. Proof of this is an
exercise.

Now comes the main theorem of this section, which is the following:

Theorem 3.1.4. Let X be a set and µ∗ be an outer measure on X. Let Mµ∗


be the collection of all µ∗ -measurable subsets of X. Then
(a) Mµ∗ is a σ-algebra.
(b) The restriction of the outer measure µ∗ on Mµ∗ is a measure on Mµ∗ .

Proof: (a) It follows from remark 3.1.3 above that X ∈ Mµ∗ . Note also that
the equation µ∗ (A) = µ∗ (A ∩ B) + µ∗ (A ∩ B c ) remains unchanged if the roles
of B and B c are interchanged. Hence Mµ∗ is closed under complementation.
We will now prove that Mµ∗ is closed under taking countable unions. We
claim that it is enough to show that Mµ∗ is closed under taking countable
unions of disjoint sets. The proof of this claim is an exercise. [Hint: First
show that Mµ∗ is an algebra and then replace a countable union of arbitrary
sets with a countable union of disjoint sets in a standard way.]
Let {Bi }i be a sequence of sets in Mµ∗ such that the Bi ’s are disjoint.
We need to show that ∪i Bi ∈ Mµ∗ , that is, we need to show that for every
subset A of X,

µ∗ (A) = µ∗ (A ∩ (∪i Bi )) + µ∗ (A ∩ (∪i Bi )c ).

It follows from the countable sub-additivity of µ∗ that


µ∗ (A) ≤ µ∗ (A ∩ (∪i Bi )) + µ∗ (A ∩ (∪i Bi )c )
= µ∗ (∪i (A ∩ Bi )) + µ∗ (A ∩ (∪i Bi )c ) ≤ Σi µ∗ (A ∩ Bi ) + µ∗ (A ∩ (∪i Bi )c ).
So if we can show that Σi µ∗ (A ∩ Bi ) + µ∗ (A ∩ (∪i Bi )c ) ≤ µ∗ (A), then
we will be done (because in that case, all the inequalities right above this
paragraph will turn into equalities). Hence we are reduced to showing that

Σi µ∗ (A ∩ Bi ) + µ∗ (A ∩ (∪i Bi )c ) ≤ µ∗ (A) (3.1.2)

for every A ⊆ X.
Suppose we can prove that for each A ⊆ X and for each n ∈ N,

µ∗ (A) = Σni=1 µ∗ (A ∩ Bi ) + µ∗ (A ∩ (∩ni=1 Bic )) (3.1.3)

12
then we will be done because if we replace the term µ∗ (A ∩ (∩ni=1 Bic ) in the
RHS of equation 3.1.3 by µ∗ (A ∩ (∩∞ c
i=1 Bi ), then we are actually reducing the
RHS, and getting the following:

µ∗ (A) ≥ Σni=1 µ∗ (A ∩ Bi ) + µ∗ (A ∩ (∩∞ c


i=1 Bi )).

On this, if we take limits as n 7→ ∞, we get that Σi µ∗ (A ∩ Bi ) + µ∗ (A ∩


(∪i Bi )c ) ≤ µ∗ (A) for every A ⊆ X. Thus it is enough to prove the equality
3.1.3.
We will prove 3.1.3 using induction on n. For n = 1, the equality follows
from the fact that B1 is µ∗ -measurable. Then use the induction hypothesis,
the µ∗ -measurability of the sets Bi and the disjointness of Bi ’s to get the
required result. I leave this part of the proof to the reader as an exercise.
[Hint: For any fixed N ∈ N, it follows from the µ∗ -measurability of BN +1
that µ∗ (A ∩ (∩N c ∗ N c ∗ N c c
i=1 Bi )) = µ (A ∩ (∩i=1 Bi ) ∩ BN +1 ) + µ (A ∩ (∩i=1 Bi ) ∩ BN +1 ).
Then using disjointness of the Bi ’s, it follows that A∩(∩N c
i=1 Bi )∩BN +1 equals
N c c
A ∩ BN +1 . It follows from basic set theory that A ∩ (∩i=1 Bi ) ∩ BN +1 equals
N +1 c
A ∩ (∩i=1 Bi ).]
Hence Mµ∗ is a σ-algebra. It now remains to prove part (b). For this, it
is enough to prove that for any sequence {Bi }i of sets in Mµ∗ which are dis-
joint, we have µ∗ (∪i Bi ) ≥ Σi µ∗ (Bi ). But this is obvious if we put A = ∪i Bi
in equation 3.1.2. 

Notation 3.1.5. Recall the lebesgue outer measure λ∗ on R. We denote by


Mλ∗ the collection of all lebesgue measurable subsets of R, i.e., all subsets of
R which are measurable with respect to the outer measure λ∗ .
Similarly, we denote by Mλ∗d the collection of all lebesgue measurable
subsets of Rd , i.e., all subsets of Rd which are measurable with respect to the
lebesgue outer measure λ∗d . 

Proposition 3.1.6. Every Borel subset of R is lebesgue measurable, i.e.,


B(R) ⊆ Mλ∗ .

Proof: It is enough to show that any interval of the type (−∞, b] belongs
to Mλ∗ because intervals of the type (−∞, b] generate the σ-algebra B(R).
Let B be such an interval. It is enough to prove that

λ∗ (A) ≥ λ∗ (A ∩ B) + λ∗ (A ∩ B c )

13
holds for each A ⊆ R having the property that λ∗ (A) < ∞.
Let A ⊆ R be such that λ∗ (A) < ∞ and let ǫ > 0 be arbitrary. It
then follows from the definition of λ∗ (A) (the infimum property) that there
exists a sequence {(an , bn )}n of open intervals such that A ⊆ ∪n (an , bn ) and
Σn (bn − an ) < λ∗ (A) + ǫ.
For each fixed n, consider the sets (an , bn ) ∩ B and (an , bn ) ∩ B c . These
two are disjoint (one of them can possibly be empty) intervals which form a
partition of the interval (an , bn ). We can choose (How?→ Exercise) open
intervals (cn , dn ) and (en , fn ) such that

(an , bn ) ∩ B ⊆ (cn , dn )

(an , bn ) ∩ B c ⊆ (en , fn )

and
ǫ
dn − cn + fn − en ≤ bn − an + .
2n
The sequence {(cn , dn )}n then covers A ∩ B and the sequence {(en , fn )}n
covers A ∩ B c . Hence by the infimum property of λ∗ , it follows that λ∗ (A ∩
B) ≤ Σn (dn − cn ) and λ∗ (A ∩ B c ) ≤ Σn (fn − en ).
Furthermore, Σn (dn −cn )+Σn (fn −en ) ≤ Σn (bn −an )+ǫ. All these together
imply that λ∗ (A ∩ B) + λ∗ (A ∩ B c ) ≤ Σn (bn − an ) + ǫ < (λ∗ (A) + ǫ) + ǫ =
λ∗ (A) + 2ǫ. Since ǫ was arbitrary, the proof follows. 
The following is an analog of Proposition 3.1.6 above, whose proof is similar,
hence omitted.

Proposition 3.1.7. Every borel subset of Rd is lebesgue measurable.

Proof: Omitted. 

Definition 3.1.8. The restriction of λ∗ (resp. of λ∗d ) to Mλ∗ (resp. to Mλ∗d )


is a measure and it is called the lebesgue measure on R (resp. on Rd ). It is
denoted by λ (resp. λd ). 

Some natural Questions: 1) Is every subset of R lebesgue measurable?


2) We have proved that B(R) ⊆ Mλ∗ , but is the converse true?

14
The answer to both these questions is NO in general. We will give an
example for the answer to the 1-st question soon in example 3.1.9 below.
The example corresponding to the 2-nd question will be given when we study
measurable functions.

Example(s) 3.1.9. There is a subset of the interval (0, 1) that is not


lebesgue measurable. Define a binary relation ∼ on R by letting x ∼ y iff
x − y ∈ Q. Check that ∼ so defined is an equivalence relation.
Each equivalence class under ∼ is of the form Q+x for some x ∈ R. Hence
each equivalence class is dense in R. So each equivalence class intersects the
interval (0, 1) non-trivially. It is also easy check that distinct equivalence
classes under ∼ are disjoint.
Let D := the set of all distinct equivalence classes of R under ∼. Each
element of D intersects (0, 1) non-trivially, pick (any) one element from
(0, 1) ∩ (Q + x) for each Q + x ∈ D, call it ex . Define

E := {ex |Q + x ∈ D}.

Note that E is a well defined set since distinct equivalence classes under ∼
are disjoint. We will now prove that E is NOT lebesgue measurable.
Let {rn } be an enumeration of the rational numbers in the interval (−1, 1).
For each n, let En := E + rn . Check that
(a) En ’s are disjoint.
(b) ∪n En ⊆ (−1, 2) and
(c) (0, 1) ⊆ ∪n En .
This checking is left to the reader as an exercise.
Now suppose that E is lebesgue measurable. Then for each n, the set En
will be lebesgue measurable (see Proposition 3.2.2 for a reasoning.). It follows
from (a) that λ(∪n En ) = Σn λ(En ). Again since λ is translation invariant (see
Proposition 3.2.2 for a reasoning.), we have λ(En ) = λ(E) ∀ n.
So if λ(E) = 0, then λ(∪n En ) = 0 which contradicts assertion (c). And
if λ(E) 6= 0, then λ(∪n En ) = ∞ which contradicts assertion (b). Thus E is
not lebesgue measurable. It follows from Proposition 3.1.6 above that E is
NOT a BOREL set. 

Exercise: Prove that the cantor set is a borel set and has lebesgue measure
0.

15
3.2 Properties of the lebesgue measure
Definition 3.2.1. For each A ⊆ Rd and each x ∈ Rd , let
A + x := {a + x|a ∈ A}.
The set A + x is called the translate of A by x. 
Proposition 3.2.2. The lebesgue outer measure λ∗d on Rd is translation-
invariant, i.e., λ∗d (A) = λ∗d (A+x) ∀ A ⊆ Rd and for all x ∈ Rd . Furthermore,
a subset B of Rd is lebesgue measurable iff B + x is lebesgue measurable for
all x ∈ Rd .
Proof: The first part of the proposition follows from the definition of λ∗d
and the fact that the volume of a d-dimensional interval of Rd remains invari-
ant under translation. The second assertion follows from the first, together
with the definition of a lebesgue measurable set (note that the operations of
intersection and translation commute with each other!). All the necessary
checkings for proving the second assertion are left to the reader as exercises.

The property of the lebesgue measure given in Proposition 3.2.3 below is
called the regularity property of the lebesgue measure.
Proposition 3.2.3. Let A be a lebesgue measurable subset of Rd . Then
(a) λd (A) = inf {λd (U)|U is open and A ⊆ U} and
(b) λd (A) = sup{λd(K)|K is compact and K ⊆ A}.
Proof: It is enough to prove that
(a) λd (A) ≥ inf {λd (U)|U is open and A ⊆ U} and
(b) λd (A) ≤ sup{λd(K)|K is compact and K ⊆ A}
because the other two inequalities are obvious (use the monotonicity of λ!).
Part (a): The required inequality clearly holds if λd (A) = ∞. So without loss
of generality, we may assume that λd (A) < ∞. Let ǫ > 0 be arbitrary. It
follows from the definition of λ∗d (A) (the infimum property) that there exists
a sequence {Ri }i of bounded open d-dimensional intervals of Rd such that
A ⊆ ∪i Ri and Σi vol(Ri ) < λ∗d (A) + ǫ. Since A is a lebesgue measurable, it
follows that λ∗d (A) = λd (A) and hence we have Σi vol(Ri ) < λd (A) + ǫ.
Let Uǫ := ∪i Ri . Then Uǫ is open, A ⊆ Uǫ and λd (Uǫ ) = λd (∪i Ri ) ≤
Σi λd (Ri ) = Σi vol(Ri ) < λd (A) + ǫ. Since inf {λd (U)|U is open and A ⊆
U} ≤ λd (Uǫ ), we get that inf {λd (U)|U is open and A ⊆ U} < λd (A) + ǫ.
Taking limits as ǫ → 0, we get the required result.

16
Part (b): Let us first prove it for the case when A is a bounded subset of
Rd . Let C be a closed and bounded subset of Rd such that A ⊆ C (such a
C exists because A is bounded). Let ǫ > 0 be arbitrary. Since C is a closed
subset of Rd , it is borel and hence lebesgue measurable. Since A and C are
both lebesgue measurable, so is C − A (:= C ∩ Ac ). It follows from part(a)
that λd (C − A) = inf {λd (U)|U is open and C − A ⊆ U}. Hence by the
infimum property, there exists an open set U0 such that C − A ⊆ U0 and
λd (U0 ) < λd (C − A) + ǫ.
Let K0 := C − U0 (= C ∩ U0c ). Check that K0 is compact and K0 ⊆ A.
Note that C ⊆ K0 ∪ U0 . Hence λd (C) ≤ λd (K0 ) + λd (U0 ). Now consider the
two inequalities:
λd (C) ≤ λd (K0 ) + λd (U0 )
and
λd (U0 ) < λd (C − A) + ǫ.
These two together imply that λd (C) < λd (K0 ) + λd (C − A) + ǫ. Now since
A is bounded, we have λd (C − A) = λd (C) − λd (A) and hence it follows from
the above that λd (A) − ǫ < λd (K0 ).
Therefore λd (A) − ǫ < λd (K0 ) ≤ sup{λd (K)|K is compact and K ⊆ A}.
Since ǫ > 0 was arbitrary, taking limits as ǫ → 0 in the above, we get that
λd (A) ≤ sup{λd (K)|K is compact and K ⊆ A}.
Let us now work out the case when A is an unbounded subset of Rd . Let b
be an arbitray real number such that b < λd (A). We will produce a compact
subset K of A such that b < λd (K). Let {Aj } be an increasing sequence of
bounded measurable subsets of A such that A = ∪j Aj (such a sequence {Aj }
exists, for example, we can take Aj to be A ∩ {x ∈ Rd : ||x|| ≤ j}). Then
λd (A) = limj→∞ λd (Aj ). Now since b < λd (A) and λd (A) = limj→∞ λd (Aj ),
we can choose j0 such that b < λd (Aj0 ). The set Aj0 is bounded measurable,
hence by the bounded-version of part(b), we can produce a compact subset
K of Aj0 such that λd (Aj0 ) ≤ λd (K). Hence b < λd (K). Since K ⊆ A and
b was an arbitray number < λd (A), it follows that λd (A) ≤ λd (K) which in
turn is ≤ sup{λd(K)|K is compact and K ⊆ A}. 
The following 2 propositions were NOT proved in the class, they were only
stated. We omit their proofs:

Proposition 3.2.4. The lebesgue measure is the only measure on (Rd , B(Rd ))
which assigns to each d-dimensional interval its volume.

17
Proposition 3.2.5. Let µ be a non-zero measure on (Rd , B(Rd )) that is
translation-invariant (i.e.,µ(A + x) = µ(A) for all A ∈ B(Rd ) and for all
x ∈ Rd ) and that is finite on the bounded borel subsets of Rd . Then there
exists a positive number c such that µ(A) = cλd (A) for each A ∈ B(Rd ).
The following lemma gives us a way of computing the lebesgue measure
of open subsets of Rd . Since these open subsets generate B(Rd ), we are in a
better world!
Lemma 3.2.6. Each open subset of Rd is the union of a countable disjoint
collection of half-open cubes, each of which is of the following form:
ji ji + 1
{(x1 , . . . , xd ) :
≤ xi < ∀ i ∈ {1, . . . , d}}
2k 2k
for some positive integer k and some integers j1 , . . . , jd (some of the ji ’s may
be negative).
Proof: For each fixed positive integer k, let Ck denote the collection of all
cubes of the form
ji ji + 1
{(x1 , . . . , xd ) : k ≤ xi < ∀ i ∈ {1, . . . , d}}
2 2k
where j1 , . . . , jd are arbitrary integers.
It is easy to see that for each k, Ck is a countable partition of Rd . Let U
be an arbitrary open subset of Rd . Let us now construct a collection D of
cubes inductively as follows:
Set D0 = ∅ and for each k ∈ N, set Dk to be equal to the collection of all
those cubes in Ck which are contained in U and which are disjoint from all
the cubes in D1 , . . . , Dk−1 .
Let D := ∪∞ k=1 Dk . Clearly D is a countable collection of disjoint cubes and
the union of all the cubes in D is contained in U. So if we can show that U
is contained in the union of all cubes in D, then we will be done.
Let x ∈ U. Since for each k, Ck is a partition of Rd , it follows that for
each k, there exists a cube (say, c0,k ) in Ck such that x ∈ c0,k . In fact, there
exists k large enough such that x ∈ c0,k ⊆ U where c0,k is a cube in Ck .
Let k0 be the smallest k for which there exists a cube c0,k0 in Ck0 with
the property that x ∈ c0,k0 ⊆ U. It is then easy to check that the cube c0,k0
belongs to D. Hence x belongs to the union of all cubes in D. 

The following corollary is immediate:

18
Corollary 3.2.7. The lebesgue measure of ant open subset of Rd is its vol-
ume.

The following exercises were discussed in the class:


1) Show that the lebesgue measure of any circle in R2 is 0.
2) Show that the lebesgue measure of any curve in R2 is 0.
3) Show that the lebesgue measure of any straight line (finite as well as
infinite) in R2 is 0.

3.3 Completeness of a measure


Definition 3.3.1. Let (X, A, µ) be a measure space. The measure µ is said
to be complete if the relations A ∈ A, µ(A) = 0 and B ⊆ A together imply
that B ∈ A. 

Definition 3.3.2. Let (X, A, µ) be a measure space. A subset B of X is


called µ-negligible if there exists a subset A of X such that A ∈ A, µ(A) = 0
and B ⊆ A. 

Remark 3.3.3. 1) A measure µ is complete if and only if every µ-negligible


subset of X belongs to A.
2) Let µ∗ be an outer measure on a set X. Consider the σ-algebra Mµ∗ as
defined in theorem 3.1.4 above. We know from the same theorem that the
restriction of µ∗ to Mµ∗ is a measure. Using the exercise given in item no.2
of remark 3.1.3 above, prove that this measure (the restriction of µ∗ to Mµ∗ ) is
complete.

Definition 3.3.4. Let (X, A) be a measurable space and let µ be a measure


on A. The completion of A under µ is the collection Aµ of subsets A of X
for which there exists sets E and F in A such that

• E ⊆ A ⊆ F.

• µ(F − E) = 0.

Remark 3.3.5. Clearly A ⊆ Aµ .

19
Definition 3.3.6. Let µ̄ : Aµ → [0, ∞] be defined as:

µ̄(A) := µ(E) (which in turn = µ(F ))

for any A ∈ Aµ , where E and F are elements of A which satisfy the 2


conditions given in definition 3.3.4 above. We first need to check µ̄ is a well
defined function, in the sense that for any A ∈ Aµ , the common value of µ(E)
and µ(F ) depends only on the set A and the measure µ, NOT on the choice
of the sets E and F . This checking is left to the reader as an exercise [Hint:
Prove that if E and F are elements of A which satisfy the 2 conditions given in
definition 3.3.4 above, then µ(E) = µ(F ) = sup{µ(B)|B ∈ A and B ⊆ A}.].
We call µ̄ the completion of µ. 
The following 2 propositions were stated in the class, their proofs were
omitted:
Proposition 3.3.7. Let (X, A, µ) be a measure space. Then
1) Aµ is a σ-algebra which contains A.
2) µ̄ is a measure on Aµ that is complete and
3) The restriction of µ̄ to A is µ.
Proposition 3.3.8. The lebesgue measure on (Rd , Mλ∗d ) is the completion
of the lebesgue measure on (Rd , B(Rd )).

4 Measurable functions
The following proposition is needed to define measurable functions:
Proposition 4.0.9. Let (X, A) be a measurable space. Let A ∈ A. For a
function f : A → [−∞, ∞], the following conditions are equivalent:
(a) For each real t, the set {x ∈ A|f (x) ≤ t} ∈ A.
(b) For each real t, the set {x ∈ A|f (x) < t} ∈ A.
(c) For each real t, the set {x ∈ A|f (x) ≥ t} ∈ A.
(d) For each real t, the set {x ∈ A|f (x) > t} ∈ A.
Proof: Since for any real t, {x ∈ A|f (x) < t} = ∪n {x ∈ A|f (x) ≤ t − n1 },
(a) ⇒ (b) follows. Again since {x ∈ A|f (x) ≥ t} = A − {x ∈ A|f (x) < t},
it follows that (b) ⇒ (c). The proofs of the other 2 implications ((c) ⇒ (d)
and (d) ⇒ (a)) are similar and hence left to the reader as as exercise. 

20
Definition 4.0.10. Let (X, A) be a measurable space. Let A ∈ A. A
function f : A → [−∞, ∞] is said to be measurable with respect to A if
it satisfies one (and hence all 4) of the equivalent conditions of proposition
4.0.9 above.
A function that is measurable with respect to A is called A-measurable,
or if the σ-algebra A is clear from the context, it is simply called measurable.


Definition 4.0.11. When X = Rd , a function that is measurable with re-


spect to B(Rd ) is called Borel-measurable or a Borel function. Similarly, a
function that is measurable with respect to Mλ∗d is called lebesgue-measurable.


Clearly, every borel function on Rd is lebesgue measurable.

Example(s) 4.0.12. 1) Let f : Rd → R be continuous. For each real t, the


set {x ∈ Rd |f (x) < t} is open, hence borel. Thus f is a borel function.
2) Let I be a subinterval of R. Let f : I → R be non-decreasing. CHECK
THAT for each real t, the set {x ∈ I|f (x) < t} is a borel set (it is either an
interval or a singleton or the empty set!). Hence f is a borel function.
3) Let (X, A) be a measurable space. Let B ⊆ X. Let χB : X → {0, 1} be
defined as:

0 if x ∈ /B
χB (x) =
1 otherwise
χB is called the chracteristic function of B. CHECK THAT χB is A-
measurable if and only if B ∈ A.
4) A function is called simple if it takes only finitely many values. Let
(X, A) be a measurable space. Let f : X → [−∞, ∞] be simple. Let
range(f ) = {α1 , . . . , αn }. PROVE THAT f is A-measurable if and only if
{x ∈ X|f (x) = αi } ∈ A for all i ∈ {1, . . . , n}. 

4.1 Some basic facts about measurable functions


Proposition 4.1.1. Let (X, A) be a measurable space. Let A ∈ A and f, g
be two [−∞, ∞]-valued measurable functions on A. Then the sets
{x ∈ A|f (x) < g(x)},
{x ∈ A|f (x) ≤ g(x)} and
{x ∈ A|f (x) = g(x)}
belong to A.

21
Proof: Note that f (x) < g(x) if and only if there exists r ∈ Q such that
f (x) < r < g(x). Hence

{x ∈ A|f (x) < g(x)} = ∪r∈Q [{x ∈ A|f (x) < r} ∩ {x ∈ A|r < g(x)}].

So the set {x ∈ A|f (x) < g(x)} is a countable union of sets that belong to
A, hence {x ∈ A|f (x) < g(x)} ∈ A.
Similarly, we can prove that {x ∈ A|g(x) < f (x)} ∈ A. Note that
{x ∈ A|f (x) ≤ g(x)} = A − {x ∈ A|g(x) < f (x)}, hence belongs to A. Fi-
nally, {x ∈ A|f (x) = g(x)} = {x ∈ A|f (x) ≤ g(x)} − {x ∈ A|f (x) < g(x)}.


Definition 4.1.2. Let (X, A) be a measurable space. Let A ∈ A and f, g


be two [−∞, ∞]-valued functions having common domain A. The maximum
and minimum of f and g, denoted by f V g and f ∧ g respectively, are
functions from A to [−∞, ∞] defined by

(f V g)(x) = max(f (x), g(x))∀ x ∈ A

and

(f ∧ g)(x) = min(f (x), g(x))∀ x ∈ A.


If {fn }n is a sequence of [−∞, ∞]-valued functions on A, then Supn fn , Infn fn , limsupn fn
and liminfn fn are defined as follows:
(Supn fn )(x) := sup{fn (x)|n ∈ N},
(Infn fn )(x) := inf {fn (x)|n ∈ N},
limsupn fn := Infk gk where gk := Supn≥k fn and
liminfn fn := Supk hk where hk := Infn≥k fn .
All these functions have domain A. Let A0 := {x ∈ A|limsupn fn (x) =
liminfn fn (x)}. For each x ∈ A0 , we define the function limn fn as

limn fn (x) := limsupn fn (x) [= liminfn fn (x)].

Note that the domain of limn fn is A0 , not A. 


Proposition 4.1.3. Let (X, A) be a measurable space. Let A ∈ A and f, g
be two [−∞, ∞]-valued measurable functions on A. Then the functions f V g
and f ∧ g are measurable.

22
Proof: For the measurability of f V g, use the identity

{x ∈ A|(f V g)(x) ≤ t} = {x ∈ A|f (x) ≤ t} ∩ {x ∈ A|g(x) ≤ t}


for any real number t. The measurability of f ∧ g is left to the reader as an
exercise. 

Example(s) 4.1.4. There exists sequences {fn }n of functions such that each
fn is continuous on the given domain, but the function limn fn does not exist
because the functions limsupn fn and liminfn fn do not agree at any point of
the domain!
Take X = R, A = B(R), A = [0, 1] and
1
fn (x) = 1 + ∀ x ∈ A if n is even and
n

1
fn (x) = −1 − ∀ x ∈ A if n is odd.
n

Exercise: Construct an example of a sequence {fn }n of functions such that
each fn is discontinuous at every point in its domain, but the function limn fn
exists and is continuous.
Proposition 4.1.5. Let (X, A) be a measurable space. Let A ∈ A. Let {fn }n
be a sequence of [−∞, ∞]-valued measurable functions on A. Then the func-
tions Supn fn , Infn fn , limsupn fn and liminfn fn are measurable. Moreover,
the function limn fn (whose domain is {x ∈ A|limsupn fn (x) = liminfn fn (x)})
is measurable.
Proof: Since {x ∈ A|Supn fn (x) ≤ t} = ∩n {x ∈ A|fn (x) ≤ t} for any real t,
it follows that Supn fn is measurable. The measurability of Infn fn , limsupn fn
and liminfn fn are left to the reader as exercises.
Let A0 := {x ∈ A|limsupn fn (x) = liminfn fn (x)}, the domain of limn fn .
It is an exercise to the reader to show that A0 ∈ A. Then note that for
any real t,
{x ∈ A0 |limn fn (x) ≤ t} = A0 ∩ {x ∈ A|limsupn fn (x) ≤ t}.
Hence limn fn is measurable. 

23
Proposition 4.1.6. Let (X, A) be a measurable space. Let A ∈ A. Let f
and g be [0, ∞]-valued measurable functions on A. Let α be a real number
≥ 0. Then αf and f + g are measurable.

Proof: If α = 0, the αf is the identically zero function, hence (It is an


exercise for the reader to show that any real valued constant function is
measurable) measurable. If α > 0, then for any real t, {x ∈ A|αf (x) < t} =
{x ∈ A|f (x) < αt }. Hence αf is measurable.
For the measurability of f + g, note that for each real t, f (x) + g(x) < t
if and only if f (x) < t − g(x) if and only if there exists r ∈ Q such that
f (x) < r < t − g(x). So for any real t,

{x ∈ A|(f + g)(x) < t} = ∪r∈Q [{x ∈ A|f (x) < r} ∩ {x ∈ A|g(x) < t − r}].

Hence the measurability of f + g follows. 

Proposition 4.1.7. Let (X, A) be a measurable space. Let A ∈ A. Let f


and g be R-valued measurable functions on A. Let α be a any real number.
Then αf, f + g, f − g, f g, fg are measurable where the domain of fg is {x ∈
A|g(x) 6= 0} and the domains of all the other functions is A.

Proof: The proof of the measurability of αf and f + g are similar to


the proof of proposition 4.1.6 above, the only extra case that needs to be
considered is the case when α < 0. In this situation, note that for any real
t, {x ∈ A|αf (x) < t} = {x ∈ A|f (x) > αt }. Hence αf is measurable. The
measurability of f − g follows from the identity f − g = f + (−1)g.
For the measrability of f g, note that f g = 21 [(f + g)2 − f 2 − g 2]. We
will show that if h : A → R is measurable, then so is h2 , the measurability
of f g will follow then from the identity f g = 12 [(f + g)2 − f 2 − g 2 ]. If
t ≤ 0, then {x ∈√A|h2 (x) < t} √= ∅ and if t > 0, then {x ∈ A|h2 (x) <
t} = {x ∈√A| − t < √h(x) < t}. Since h is measrable, we have that
{x ∈ A| − t < h(x) < t} ∈ A. Hence h2 is measurable.
Let Ã:= {x ∈ A|g(x) 6= 0}, the domain of fg . It is an exercise to
the reader to show that Ã∈ A. Then note that for each real t, the set
{x ∈ A|( fg )(x) < t} = D1 ∪ D2 where D1 and D2 are elements of A given by

D1 := {x ∈ A|g(x) > 0} ∩ {x ∈ A|f (x) < tg(x)}

24
and

D2 := {x ∈ A|g(x) < 0} ∩ {x ∈ A|f (x) > tg(x)}.


f
Hence g
is measurable. 

Proposition 4.1.8. Let (X, A) be a measurable space. Let A ∈ A. For a


function f : A → R, the following are equivalent:
(a) f is A-measurable.
(b) f −1 (U) ∈ A for all U open in R.
(c) f −1 (C) ∈ A for all C closed in R.
(d) f −1 (B) ∈ A for all B borel in R.

Proof: Let F := {B ⊆ R|f −1 (B) ∈ A}. Prove that F is a σ-algebra over


R (exercise).
Note that (a) ⇔ f −1 ((−∞, t]) ∈ A ∀t real ⇔ B(R) ⊆ F . Hence
(a) ⇔ (d). The proofs of (d) ⇒ (b) and (d) ⇒ (c) are obvious. Again
since B(R) is generated by the collection of all open subsets of R, it follows
that (b) ⇒ (d). The proof of (c) ⇒ (d) is similar. 

Definition 4.1.9. Let X be a set. Let f be an extended real valued (i.e.,[−∞, ∞]-
valued) function on X.Let

f + := max{f (x), 0} ∀ x ∈ X
and

f + := −min{f (x), 0} ∀ x ∈ X.
Then f + and f − are extended real valued functions called the positive and
the negative parts of f . Thus f + = f V 0 and f − = (−f ) V 0. 

Remark 4.1.10. Let (X, A) be a measurable space. Let A ∈ A. Let f


be a [−∞, ∞]-valued function defined on A. Note that f = f + − f − and
|f | = f + + f − . Then f and |f | are measurable if and only if f + and f − are
measurable (Proof:exercise).

Proposition 4.1.11. Let (X, A) be a measurable space. Let A ∈ A. Let f


be a [−∞, ∞]-valued measurable function on A. Then there exists a sequence

25
{fn }n of simple, A-measurable, R-valued functions on A such that

f (x) = limn fn (x) ∀ x ∈ A.


Proof: Apply proposition 4.1.12 below to the positive and negative parts
of f separately. 

Proposition 4.1.12. Let (X, A) be a measurable space. Let A ∈ A. Let f


be a [0, ∞]-valued measurable function on A. Then there exists a sequence
{fn }n of simple, A-measurable, [0, ∞)-valued functions on A that satisfy

(1) f1 (x) ≤ f2 (x) ≤ · · ·

and

(2) f (x) = limn fn (x)


for each x ∈ A.
Proof: For each n ∈ N and for each k ∈ {1, 2, . . . , n2n }, let
k−1 k
An,k := {x ∈ A| ≤ f (x) < }.
2n 2n
The measurability of f implies that An,k ∈ A for all n, k. Define a sequence
{fn }n of functions from A to R as follows:
For each  n ∈ N, let
k−1
2n
if x ∈ An,k for some k ∈ {1, 2, . . . , n2n }
fn (x) := n
n if x ∈ A − (∪n2
k=1 )An,k
Check that each fn is simple and A-measurable (exercise). We claim that
the fn ’s so defined satisfy the conditions (1) and (2) for each x ∈ A.
Let us first prove condition (1): Fix any n ∈ N. If x ∈ An,k for some
k, then k−12n
≤ f (x) < 2kn . So x ∈ An,k if and only if either x ∈ An+1,2k−1
or x ∈ An+1,2k . Hence for any x ∈ An,k , either fn+1 (x) equals 2k−1
2n+1
2k
or 2n+1 .
k−1 2k−1 2k
But since x ∈ An,k , fn (x) = 2n < 2n+1 < 2n+1 . Hence for any x ∈ An,k ,
fn (x) < fn+1 (x). So condition (1) is satisfied. For x ∈ A − (∪k An,k ), I leave
it to the reader as an exercise to prove that fn (x) ≤ fn+1 (x).
Let us now prove condition (2): For those x ∈ A for which f (x) = ∞,
it is easy to see that x will belong to the set ∩n [A − (∪k An,k )]. Hence
limn fn (x) = ∞ (= f (x)). Now consider those x ∈ A for which f (x) < ∞.

26
Then there exists N ∈ N such that f (x) ≤ N. So for any n ≥ N, x
will belong to An,k for some k. For such an x, we have fn (x) = k−1 2n
and
1
|f (x) − ( k−1
2n
)| < 2n
. Hence for such x, |f (x) − fn (x)| → 0 as n → ∞. 

4.2 Example of a lebesgue measurable, non-borel set


Let K denote the cantor set. Recall the construction of K: Let K0 := [0, 1],
and for each positive integer n, let Kn be the compact set constructed out of
Kn−1 by removing the open middle thirds of each of the intervals present in
it. The cantor set K was given by K = ∩n Kn .
The Cantor singular function is the function f : [0, 1] → [0, 1] defined
as follows: For each x ∈ ( 31 , 32 ), let f (x) = 21 . Thus f is now defined at
each point removed from [0, 1] in the construction of K1 . Next define f at
each point removed from K1 in the construction of K2 by letting f (x) = 14
if x ∈ ( 19 , 92 ) and leting f (x) = 43 if x ∈ ( 97 , 89 ). Continue in this way, leting
n
f (x) be 21n , 23n , 25n , . . . , 2 2−1
n on the various intervals removed from Kn−1 in
the construction of Kn . Now f is defined on the open set [0, 1] − K, it is
non-decreasing, and has values in [0, 1]. Extend f to all of [0, 1] by letting
f (0) = 0 and letting

f (x) = sup{f (t) : t ∈ [0, 1] − K and t < x}

if x ∈ K − {0}. This completes the definition of the Cantor function.


It is easy to check that f is non-decreasing and continuous, and it is clear
that f (1) = 1. The intermediate value theorem thus implies that for each
y ∈ [0, 1], there is at least one x ∈ [0, 1] such that f (x) = y, and so we can
define a function g : [0, 1] → [0, 1] by

g(y) := inf {x ∈ [0, 1]|f (x) = y}.

The continuity of f implies that f (g(y)) = y holds for each y ∈ [0, 1]; hence
g is injective. It is an exercise for the reader to check that all the values
of g lie in the Cantor set. The facts that f is non-decreasing and f (g(y)) =
y∀y ∈ [0, 1] imply that g is non-decreasing. It now follows from example 2
of 4.0.12 that g is Borel measurable.

Proposition 4.2.1. There is a Lebesgue measurable subset of R that is not


a Borel set.

27
Proof: Let g be the function constructed above, let E be the subset of [0, 1]
as in example 3.1.9 and let B := g(E). Then B is a subset of the Cantor set
and so is Lebesgue measurable (recall that the Cantor set if lebesgue mea-
surable, has lebesgue measure 0 and that Lebesgue measure on the σ-algebra
of Lebesgue measurable sets is complete). If B were a Borel set, then g −1 (B)
will also be Borel (see Proposition 4.1.8). However the injectivity of g im-
plies that g −1(B) = E, which is not lebesgue measurable and hence is not
a Borel set. Consequently the lebesgue measurable set B is not a Borel set. 

5 The integral
Let (X, A, µ) be a measure space. Let S denote the collection of all real-
valued simple A-measurable functions on X, and by S+ the collection of
non-negative functions in S.
If f ∈ S+ and is given by f = Σm i=1 ai χAi , where a1 , . . . , am are non-
negativeRreal numbers and A1 , . . . , Am are disjoint subsets of X that belong to
A, then f dµ, the integral of f with respect to µ, is defined to be Σm i=1 ai µ(Ai )
(note that this sum is eitherRa non-negative real number or +∞).
We need to check that f dµ depends only on f , and not on the ai ’s
and Ai ’s. Suppose that f is also given by Σni=1 bj χBj where b1 , . . . , bn are non-
negative real numbers and B1 , . . . , Bn are disjoint subsets of X that belong to
A. We then need to show that Σm n
i=1 ai µ(Ai ) = Σj=1 bj µ(Bj ). We can assume
without loss of generality that ∪m n
i=1 Ai = ∪j=1 Bj (if necessary eliminate those
sets Ai for which ai = 0 and those sets Bj for which bj = 0). Then the
additivity of µ and the fact that ai = bj if Ai ∩ Bj 6= ∅ imply that

Σm m m m n m n
i=1 ai µ(Ai ) = Σi=1 ai µ[Ai ∩(∪i=1 Ai )] = Σi=1 ai µ[Ai ∩(∪j=1 Bj )] = Σi=1 ai µ[∪j=1 (Ai ∩Bj )]

= Σm n m n n m n m
i=1 Σj=1 ai µ(Ai ∩Bj ) = Σi=1 Σj=1 bj µ(Ai ∩Bj ) = Σj=1 Σi=1 bj µ(Ai ∩Bj ) = Σj=1 bj µ[∪i=1 (Bj ∩Ai )]

= Σnj=1 bj µ[Bj ∩ (∪m n n n


i=1 Ai )] = Σj=1 bj µ[Bj ∩ (∪j=1 Bj )] = Σj=1 bj µ(Bj ).
R
Hence f dµ does not depend on the representation of f used in its definition.

28
Next we define the integral of an arbitrary [0, ∞]-valued A-measurable
function on X. For such a function f let
Z Z
f dµ := sup{ gdµ : g ∈ S+ and g ≤ f }.

It is easy to see that for functions in S+ , this agrees with the previous
definition.
Finally, let f be an arbitrary [−∞, ∞]-valued R A-measurable function on
X. Recall the decomposition f = f + − f − . If f + dµ and f − dµ are
R
R both
finite, then f Ris called integrable
R − (or µ-integrable), and its integral f dµ is
definedRto be f + dµ R −− f dµ. The integral of f is said toRexist if at least
+
one ofR f dµ and R − f dµ is finite, and again in that case f dµ is defined
+
to be f dµ − f dµ.
Suppose that f : X → [−∞, ∞] is A-measurable and that A ∈ A. Then
f is integrable overR A if the function f χA is integrable in the abobe R sense,
and in that case, A f dµ, the integral of f over A, is defined to be f χA dµ.
Likewise, if A ∈ A and if f is a measurable function whose domain is A
(rather than the entire space X), then the integral of f over A is defined to
be the integral (if it exists) of the function on X that agrees with f on A
and vanishes on Ac .
In case X = Rd and µ = λd , one often refers to Lebesgue integrability and
the LebesgueR integral. The Lebesgue integral of a function f on R is often
denoted by f (x)dx.
We define L1 (X, A, µ, R) (or simply L1 ) to be the set of all real-valued
(NOT [−∞, ∞]-valued!) integrable functions on X. We will show later that
L1 (X, A, µ, R) is a vector space and the integral is a linear functional on
L1 (X, A, µ, R).

5.1 Properties of integral of non-negative simple func-


tions
Proposition 5.1.1. Let (X, A, µ) be a measure space, let f and g belong to
S+ ,Rand let α beR a non-negative real number. Then
(a) R αf dµ = α Rf dµ, R
(b) (f + g)dµ = f dµ + gdµ, and R R
(c) if f (x) ≤ g(x) holds at each x ∈ X, then f dµ ≤ gdµ.
Proof: Suppose that f = Σm i=1 ai χAi , where a1 , . . . , am are non-negative
real numbers and A1 , . . . , Am are disjoint subsets of X that belong to A,

29
and that g = Σnj=1 bj χBj , where b1 , . . . , bn are non-negative real numbers
and B1 , . . . , Bn are disjoint subsets of X that belong to A. We can assume
without loss of generality that ∪m n m
i=1 Ai = ∪j=1 Bj . Note that αf = Σi=1 αai χAi
and f + g = Σm n
i=1 [Σj=1 (ai + bj )χAi ∩Bj ]. Then parts (a) and (b) follow from
the calculations
Z Z
m m
αf dµ = Σi=1 αai µ(Ai ) = αΣi=1 ai µ(Ai ) = α f dµ

and Z
(f + g)dµ = Σm n
i=1 Σj=1 (ai + bj )µ(Ai ∩ Bj )

= Σm n m n
i=1 Σj=1 ai µ(Ai ∩ Bj ) + Σi=1 Σj=1 bj µ(Ai ∩ Bj )

Z Z
Σm
i=1 ai µ(Ai ) + Σnj=1 bj µ(Bj ) = f dµ + gdµ.

I leave part (c) as an exercise to the reader: the proof used parts (a)
and (b). 

Proposition 5.1.2. Let (X, A, µ) be a measure space, let f belong to S+ ,


and let {fn }n be a non-decreasing sequence of
R functions inR S+ for which
f (x) = limn fn (x) holds at each x ∈ X. Then f dµ = limn fn dµ.

Proof: It follows from Proposition 5.1.1 that


Z Z Z
f1 dµ ≤ f2 dµ ≤ . . . ≤ f dµ;
R R R
hence limn fn dµ exists and satisfies limn fn dµ ≤ f dµ. We now need to
prove the reverse inequality. Let ǫ be a number such that 0 < ǫ < 1. We
shall construct a non-decreasing
R sequence
R {gn }n of Rfunctions Rin S+ such that
gn ≤ fn ∀n and limRn gn dµ = (1 R− ǫ) f dµ. Since gn dµ ≤ fn dµ, this will
implyR that (1 − ǫ) Rf dµ ≤ limn fn dµ, and, since ǫ was arbitrary, it follows
that f dµ ≤ limn fn dµ.

30
We now construct the sequence {gn }n . Suppose a1 , . . . , ak are the non-
zero values of f , and that A1 , . . . , Ak are the sets on which these values occur.
Thus f = Σki=1 ai χAi . For each n and i let
A(n, i) := {x ∈ Ai : fn (x) ≥ (1 − ǫ)ai }.
Then each A(n, i) ∈ A, and for each i the sequence {A(n, i)}∞ n=1 is non-
decreasing and satisfies Ai = ∪n A(n, i). If we let gn = Σki=1 (1 − ǫ)ai χA(n,i) ,
then gn belongs to S+ and satisfies the required properties (verify!). 

5.2 Properties of integral of [0, ∞]-valued measurable


functions
Proposition 5.2.1. Let (X, A, µ) be a measure space, let f be a [0, ∞]-valued
A-measurable function on X, and let {fn }n be a non-decreasing sequence of
functions
R in SR+ for which f (x) = limn fn (x) holds at each x ∈ X. Then
f dµ = limn fn dµ.
Proof: It is clear that
Z Z Z
f1 dµ ≤ f2 dµ ≤ . . . ≤ f dµ;
R R R
hence limn fn dµ exists and satisfies R lim n fn dµ ≤ f dµ. We now need to
prove the reverse inequality.
R Since f dµ is the supremum of those elements
of [0, ∞] of the form gdµ, where g ranges over the setRof functions that R be-
long to S+ and satisfy g ≤ f , it follows that for proving f dµ ≤ limn fn dµ,
it is enough to R arbitrary function g in S+ which satisfies g ≤ f
R check that an
also satisfies gdµ ≤ limn fn dµ. Let g be such a function. Then {g ∧ fn }n
is a non-decreasing sequence of functions inR S+ for whichR g = limn (g ∧ fn ).
R then follows Rfrom proposition 5.1.2 Rthat gdµ = Rlimn (g ∧ fn )dµ. Since
It
(g ∧ fn )dµ ≤ fn dµ, it follows that gdµ ≤ limn fn dµ. 

Proposition 5.2.2. Let (X, A, µ) be a measure space, let f and g be [0, ∞]-
valued A-measurable functions on X, and let α be a non-negative real number.
ThenR R
(a) R αf dµ = α Rf dµ, R
(b) (f + g)dµ = f dµ + gdµ, and R R
(c) if f (x) ≤ g(x) holds at each x ∈ X, then f dµ ≤ gdµ.

31
Proof: Choose non-decreasing sequences {fn }n and {gn }n of functions in
S+ such that f = limn fn and g = limn gn (there exists such sequences,
see Proposition 4.1.12). Then {αfn }n and {fn + gn }n are non-decreasing
sequences of functions in S+ that satisfy αf = limn αfn and f +g = limn (fn +
gn ). The proof of parts (a) and (b) are now left to the reader as exercises.
For part (c), note that if f ≤ g, then the class of functions h in S+ that
satisfy h ≤ f isRcontainedR in the class of functions h in S+ that satisfy h ≤ g;
it follows that f dµ ≤ gdµ. 

5.3 Properties of integral of real-valued measurable


functions
Lemma 5.3.1. Let (X, A, µ) be a measure space, and let f1 , f2 , g1 , and g2 be
non-negative
R real-valued
R integrable
R Rfunctions on X such that f1 −f2 = g1 −g2 .
Then f1 dµ − f2 dµ = g1 dµ − g2 dµ.
R R R R
Proof: Clearly f1 +g2 = g1 +f2 and hence f1 dµ+ g2 dµ R= g1 dµ+ R f2 dµ.
Since
R all the
R integrals involved are finite, this implies that f1 dµ − f2 dµ =
g1 dµ − g2 dµ. 

Proposition 5.3.2. Let (X, A, µ) be a measure space, let f and g be real-


valued integrable functions on X, and let α be a real number. Then
(a) Rαf and f + Rg are integrable,
(b) R αf dµ = α fR dµ, R
(c) (f + g)dµ = f dµ + gdµ, and R R
(d) if f (x) ≤ g(x) holds for all x ∈ X, then f dµ ≤ gdµ.
R R
Proof: The integrability of αf and the relation αf dµ = α f dµ are clear
if α = 0. If α > 0, then (αf )+ = αf + and (αf )− = αf − . And if α < 0,
then (αf )+ = −αf − and (αf )− = −αf + . The rest of the proof about the
function αf is left to the reader as an exercise.
Now consider the sum of f and g. Note that (f + g)+ ≤ f + + g + and
(f + g)− ≤ f − + g − . Thus
Z Z Z
(f + g) dµ ≤ f dµ + g + dµ < +∞
+ +

32
and
Z Z Z
− −
(f + g) dµ ≤ f dµ + g − dµ < +∞,

and so f + g is integrable. Since f + g is equal to (f + g)+ − (f + g)− and


also to f + + g + − (f − + g − ), it follows from lemma 5.3.1 above that
Z Z Z
(f + g)dµ = (f + g )dµ − (f − + g − )dµ,
+ +

R R R
and hence that (f + g)dµ = f dµ + gdµ.
If f (x) ≤ g(x)
R holds at each x ∈ X, then
R g−fRis a non-negative
R integrable
function; hence (g − f )dµ ≥ 0, and so gdµ − f dµ = (g − f )dµ ≥ 0. 

5.4 Properties of integral of [−∞, ∞]-valued measur-


able functions
Proposition 5.4.1. Let (X, A, µ) be a measure space, and f be a [−∞, ∞]-
valued A-measurable function on X. Then f is integrable if and only if |f |
is integrable. If these functions are integrable, then
Z Z
| f dµ| ≤ |f |dµ.

Proof: Exercise. 

Remark 5.4.2. In Proposition 5.4.1 above, the assumption about the measur-
ability of f is important because there are functions that are not measurable,
hence not integrable, but have an integrable absolute value. Exercise:Find
such an example (example of such a function was discussed in the class!).

The following results were proved in the class, which I am stating with-
out proof over here:

Proposition 5.4.3. Let (X, A, µ) be a measure space, and let f , g be [−∞, ∞]-
valued
R A-measurable
R functions on X that agree
R almost R everywhere. If either
f dµ or gdµ exists, then both exist, and f dµ = gdµ.

33
Proposition 5.4.4. Let (X, A, µ) be a measure space,
R and f be a [−∞, ∞]-
valued A-measurable function on X that satisfies |f |dµ = 0. Then f van-
ishes almost everywhere.
Proposition 5.4.5. Let (X, A, µ) be a measure space, and f be a [−∞, ∞]-
valued integrable function on X. Then |f (x)| < +∞ holds at almost every x
in X.
Corollary 5.4.6. Let (X, A, µ) be a measure space, and f be a [−∞, ∞]-
valued A-measurable function on X. Then f is integrable if and only if there
is a function in L1 (X, A, µ, R) that is equal to f almost everywhere.

DECLARATION : From this point onwards, no proofs will be


provided in these notes. Only statements of results will be given.

6 Limit theorems
Theorem 6.0.7. The Monotone Convergence theorem: Let (X, A, µ)
be a measure space. Let f and f1 , f2 , . . . be [0, ∞]-valued A-measurable func-
tion on X. Suppose that the relations
(1) f1 (x) ≤ f2 (x) ≤ · · · and
(2) f (x) = limn fn (x) R R
hold at almost every x ∈ X. Then f dµ = limn fn dµ.
Corollary 6.0.8. Levi’s theorem: Let (X, A, µ) be a measure space. Let
Σ∞k=1 fk be an infinite
R series whose terms Rare [0, ∞]-valued A-measurable func-
tions on X, then (Σ∞ f
k=1 k )dµ = Σ∞
k=1 ( fk dµ).
Proof: Exercise. Hint: Apply theorem 6.0.7 to the sequence of partial
sums of the given series. 

Remark 6.0.9. Let (X, A, µ) be a measure space and f : X → [0, ∞] be


A-measurable. Let ν : A → [0, ∞] be a function defined by
Z
ν(A) := f dµ.
A

CHECK THAT ν is a measure on (X, A). Hint: Use Levi’s theorem.


It is also clear that ν is a finite measure if and only if f is µ-integrable.

34
Lemma 6.0.10. Fatou’s lemma: Let (X, A, µ) be a measure space. Let
{fn }n be a sequence of [0, ∞]-valued A-measurable function on X. Then
Z Z
(liminfn fn )dµ ≤ liminfn fn dµ.

Remark 6.0.11. In Fatou’s lemma, the functions fn ’s should be non-negative


valued. There exist example(s) where if we take the fn ’s to be NOT non-negative
valued, Fatou’s lemma fails to hold. Here is one example:
Let X = [0, ∞), A = B([0, ∞)) and µ = the lebesgue measure. For each
n ∈ N, let
fn (x) := − n1 if x ∈ [n, 2n] and 0 otherwise.
CHECK THAT the inequality in Fatou’s lemma fails to hold→ exercise.
Example(s) 6.0.12. There are cases when the inequality in Fatou’s lemma
is strict! For example, take X = R, A = B(R) and µ = the lebesgue measure.
For each n ∈ N, let
1
fn (x) := n
if x ∈ [0, n] and 0 otherwise.
CHECK THAT the inequality in Fatou’s lemma is strict for this example →
exercise. 
Example(s) 6.0.13. Application of Fatou’s lemma: 1) Let X = [0, 1],
A = B([0, 1]) and µ = the lebesgue measure. For each n ∈ N, let
1
fn (x) := n
if x ∈ Q ∩ [0, 1] and x otherwise.
CHECK THAT each fn is A-measurable and integrable (exercise).
Let f : X → [0, ∞] be defined as
f (x) := 0 if x ∈ Q ∩ [0, 1] and x otherwise.
Exercise: UsingRFatou’s lemma, prove that f in integrable and provide an
upper bound for f dµ.
2) In example 1 above, replace [0, 1] by (0, 1) and for each n ∈ N, let
fn (x) := xn if x ∈ Q ∩ (0, 1) and x otherwise.
CHECK THAT each fn is A-measurable and integrable (exercise).
Let f : X → [0, ∞] be defined as

35
f (x) := 0 if x ∈ Q ∩ [0, 1] and x otherwise.

Exercise: UsingRFatou’s lemma, prove that f in integrable and provide an


upper bound for f dµ.


Theorem 6.0.14. Lebesgue’s Dominated Convergence theorem: Let


(X, A, µ) be a measure space. Let g : X → [0, ∞] be integrable. Let f and
f1 , f2 , . . . be [−∞, ∞]-valued A-measurable function on X. Suppose that the
relations
(1) |fn (x)| ≤ g(x) and
(2) f (x) = limn fn (x) R
hold atR almost every x ∈ X. Then f and f1 , f2 , . . . are integrable and f dµ =
limn fn dµ.

Example(s) 6.0.15. The following is an exercise:


Application of Lebesgue’s Dominated Convergence theorem: Let
X, A, µ, f and the fn ’s be as in example
R 1 of 6.0.13. Use theorem 6.0.14 to
show that f is integrable. RCompute fn dµ for each n and using theorem
6.0.14, provide a value for f dµ. 

Remark 6.0.16. Think about more and more apllications of theorem 6.0.14.

7 Modes of Convergence
Definition 7.0.17. Let (X, A, µ) be a measure space, let f and f1 , f2 , . . . be
real-valued A-measurable function on X. We say that {fn }n converges to f
in measure if
limn µ({x ∈ X : |fn (x) − f (x)| > ǫ}) = 0
holds for each ǫ > 0. 

Definition 7.0.18. The sequence {fn }n converges to f almost everywhere if


f (x) = limn fn (x) holds at µ-almost every point x ∈ X. 

Remark 7.0.19. In general, convergence in measure neither implies nor is


implied by convergence almost everywhere (see the examples below).

Example(s) 7.0.20. Convergence almost everywhere;Convergence


in measure: Consider the measure space (R, B(R), λ) where λ denotes the

36
lebesgue measure. Let fn := χ[n,∞) ∀ n ∈ N. Clearly fn → 0 almost every-
where (in fact, everywhere). But observe that fn 9 0 in measure (Exercise).


Example(s) 7.0.21. Convergence in measure;Convergence almost


evrywhere: Let X = [0, 1), A = B([0, 1)) and µ = λ(the lebesgue measure).
Let f1 := χ[0,1) and the next two functions (that is, f2 and f3 ) be given
by f2 := χ[0, 1 ) , f3 := χ[ 1 ,1) . Similarly, let the next 4 functions (that is,
2 2
f4 , f5 , f6 , f7 ) be given by f4 := χ[0, 1 ) , f5 := χ[ 1 , 1 ) , f6 := χ[ 1 , 3 ) , f7 := χ[ 3 ,1) ,
4 4 2 2 4 4
. . . and so on. CHECK THAT fn → 0 in measure, but fn 9 0 almost
everywhere [λ] (Exercise). 

Proposition 7.0.22. Let (X, A, µ) be a measure space, let f and f1 , f2 , . . .


be real-valued A-measurable function on X. If µ is finite and if fn → f
almost everywhere [µ], then {fn }n converges to f in measure.

Proposition 7.0.23. Let (X, A, µ) be a measure space, let f and f1 , f2 , . . .


be real-valued A-measurable function on X. If {fn }n → f in measure, then
there exists a subsequence of {fn }n that converges to f almost everywhere.

Theorem 7.0.24. Egoroff ’s theorem: Let (X, A, µ) be a measure space,


let f and f1 , f2 , . . . be real-valued A-measurable function on X. If µ is finite
and if fn → f almost everywhere, then for each ǫ > 0, there exists a subset
Bǫ of X that belongs to A and satisfies
(i) µ(Bǫc ) < ǫ and
(ii) {fn }n converges to f uniformly on Bǫ .

8 Signed measures
Definition 8.0.25. Let (X, A) be a measurable space. A function µ : A →
[−∞, ∞] is called a signed measure if the following two conditions are satisfied
simultaneously:
(i) µ(∅) = 0.
(ii) µ(∪∞ ∞
i=1 ) = Σi=1 µ(Ai ) holds for each infinite sequence {Ai }i of disjoint
sets in A (→ Countable additivity). 

Definition 8.0.26. A signed measure is finite if neither +∞ nor −∞ occurs


among its values. 

37
Remark 8.0.27. Let µ be a signed measure on the measurable space (X, A).
Then for each A ∈ A, the sum µ(A) + µ(Ac ) must be well defined, that is, it
must not be of the form (+∞) + (−∞) or (−∞) + (+∞) and this sum must
equal µ(X). Hence if there exists a set A ∈ A such that µ(A) = ∞, then
µ(X) = ∞. Similarly, if there exists a set A ∈ A such that µ(A) = −∞, then
µ(X) = −∞. Therefore a signed measure can attain atmost one of the values
+∞ or −∞.
Exercise: If B is a set in A for which µ(B) is finite, then for any A ⊆ B
such that A ∈ A, µ(A) is finite.
Example(s) 8.0.28. Let (X, A, µ) be a measure space where µ is a positive
measure. Let f ∈ L1 (X, A, µ, R) where L1 (X, A, µ, R) is the set of all R-
valued integrable functions on X. Define a function ν : A → [−∞, ∞] as
Z
ν(A) := f dµ.
A

Exercise: ν is a signed measure on (X, A) (Hint: Use Lebesgue’s dominated


convergence theorem to prove countable additivity of ν).
Note that such a signed measure is the difference of the two positive
measures ν1 and ν2 defined by
Z Z
ν1 (A) = +
f dµ and ν2 (A) = f − dµ.
A A


Remark 8.0.29. If ν1 and ν2 are two positive measures on (X, A) and if at
least one of them is finite, then ν1 − ν2 is a signed measure on (X, A). We will
soon see (in Jordan Decomposition theorem) that every signed measure arises in
this way.

8.1 Some Properties of signed measures


Lemma 8.1.1. Let (X, A) be a measurable space, and let µ be a signed mea-
sure on (X, A). Then
(a) If {Ak }k is an increasing sequence of sets in A, then µ(∪∞ k=1 Ak ) =
limk µ(Ak ) and
(b) If {Ak }k is an decreasing sequence of sets in A such that µ(An ) < ∞ for
some n, then µ(∩∞ k=1 Ak ) = limk µ(Ak ).

38
Proof: Exercise (The Proof is similar to that of positive measures.) 

Definition 8.1.2. Let µ be a signed measure on (X, A). A subset A of X


is called a positive set for µ if A ∈ A and if each A-measurable subset E of
A satisfies µ(E) ≥ 0. Similarly, a subset A of X is called a negative set for
µ if A ∈ A and if each A-measurable subset E of A satisfies µ(E) ≤ 0. 
Lemma 8.1.3. Let µ be a signed measure on (X, A) and let A be a subset
of X that belongs to A and satisfies −∞ < µ(A) < 0. Then there exists a
negative set B such that B ⊆ A and µ(B) ≤ µ(A).
Theorem 8.1.4. Hahn decomposition theorem: Let (X, A) be a mea-
surable space, and let µ be a signed measure on (X, A). Then there exist
disjoint subsets P and N of X such that P is a positive set for µ, N is a
negative set for µ and X = P ∪ N.
Definition 8.1.5. A Hahn decomposition of a signed measure µ is a pair
(P, N) of disjoint subsets of X such that P is a positive set for µ, N is a
negative set for µ and X = P ∪ N. 
Remark 8.1.6. A signed measure can have more than one Hahn decomposi-
tions. For example, let X = [−1, 1], A = B([−1, 1]) and λ beR the lebesgue
measure on [−1, 1]. Let µ : A → [−∞, ∞] be defined by µ(A) = A xdλ. Then
([−1, 0), [0, 1]) and ([−1, 0], (0, 1]) both are Hahn decompositions for µ.
But the Hahn decomposition of a signed measure is “essentially unique” in
the sense that suppose µ be a signed measure on (X, A) and if (P1 , N1 ) and
(P2 , N2 ) are two distinct Hahn decompositions for µ, then P1 ∩ N2 and P2 ∩ N1
are both positive and negative sets for µ. Hence any A-measurable subset of
P1 ∩ N2 (as well as of P2 ∩ N1 ) has µ measure zero.
Theorem 8.1.7. Jordan Decomposition theorem: Every signed mea-
sure is the difference of two positive measures, at least one of which is finite.
Definition 8.1.8. Let (P, N) be a Hahn decomposition of the signed measure
µ. Let µ+ and µ− be the measures constructed from (P, N), that is, for any
A ∈ A,
µ+ (A) = µ(A ∩ P ) and µ− (A) = −µ(A ∩ N).
Then each A-measurable subset B of A satisfies µ(B) = µ+ (B) − µ− (B) ≤
µ+ (B) ≤ µ+ (A). Hence sup{µ(B) : B ∈ A and B ⊆ A} ≤ µ+ (A). Since

39
µ+ (A) = µ(A ∩ P ), it follows that µ+ (A) = µ(A ∩ P ) = sup{µ(B) : B ∈
A and B ⊆ A}. Similarly, one can prove that the measure µ− satisfies
µ− (A) = sup{−µ(B) : B ∈ A and B ⊆ A}. Thus µ+ and µ− are indepen-
dent of the particular Hahn decomposition used in their construction. The
measures µ+ and µ− are called “the” positive part and “the” negative part
of µ, and the representation µ = µ+ − µ− is called the Jordan decomposition
of µ. 

Definition 8.1.9. The variation of the signed measure µ is the positive


measure |µ| defined by |µ| = µ+ + µ− . 

Exercise: Check that |µ(A)| ≤ |µ|(A) for all A ∈ A.


Exercise: Let (X, A) be a measurable space, and let µ be a signed measure
on (X, A). Let ν be a positive measure on (X, A) such that |µ(A)| ≤ ν(A)
holds for each A ∈ A. Prove that |µ|(A) ≤ ν(A) for all A ∈ A. [Hint: First
prove that |µ|(A) = sup{Σ∞
k=1 |µ(Ak )| : {Ak }k ⊆ A are pairwise disjoint and ∪k
Ak = A}.]

9 Complex measures
Definition 9.0.10. Let (X, A) be a measurable space. A function µ : A →
C is called a complex measure if the following two conditions are satisfied
simultaneously:
(i) µ(∅) = 0.
(ii) µ(∪∞ ∞
i=1 ) = Σi=1 µ(Ai ) holds for each infinite sequence {Ai }i of disjoint
sets in A (→ Countable additivity). 

By definition, a complex measure can take only complex values, no infi-


nite values. Each complex measure µ on (X, A) can of course be written in
the form µ = µ′ + iµ′′ where µ′ and µ′′ are finite signed measures on (X, A).
Hence by the Jordan decomposition theorem for signed measures, each com-
plex measure µ can be written in the form

µ = µ1 − µ2 + iµ3 − iµ4
where µ1 , µ2 , µ3 and µ4 are finite positive measures on (X, A). Such a
representation is called the Jordan decomposition for µ if µ′ = µ1 − µ2 and
µ′′ = µ3 − µ4 are the jordan decompositions of the real and imaginary parts
of µ respectively.

40
Definition 9.0.11. Given a complex measure µ on (X, A), the variation |µ|
of µ is a function: A → [0, ∞] defined as
|µ|(A) := sup{Σnk=1 |µ(Ak )| : {Ak }nk=1 ranges over all f inite partitions of A
into A − measurable sets}. 
Proposition 9.0.12. Let (X, A) be a measurable space and let µ be a complex
measure on (X, A). Then the variation |µ| of µ is a finite measure on (X, A).
Exercise: Let µ be a finite signed measure. Then µ is also a complex mea-
sure. Prove that |µ| as the variation of a signed measure equals |µ| as the
variation of µ as a complex measure.
Exercise: Let ν be a positive measure on (X, A) such that |µ(A)| ≤
ν(A) ∀ A ∈ A, prove that |µ|(A) ≤ ν(A) ∀ A ∈ A.
Definition 9.0.13. The total variation ||µ|| of the complex measure µ is
defined by ||µ|| := |µ|(X). 

9.1 Integration with respect to a finite signed or com-


plex measure
Suppose (X, A) is a measurable space. Let
B(X, A, R) := the vector space of all bounded R-valued A-measurable func-
tions on X and
B(X, A, C) := the vector space of all bounded C-valued A-measurable func-
tions on X.

Definition 9.1.1. Let µ be a finite signed measure on (X, A). Let µ =


µ+ − µ− be the Jordan decomposition of µ. For any f ∈ B(X, A, R), the
integral of f w.r.t µ is defined by:
Z Z Z
f dµ := f dµ − f dµ−
+

R
Clearly f 7→ f dµ is a linear functional on B(X, A, R). 
Definition 9.1.2. Let µ be a complex measure on (X, A). Let µ = µ1 −
µ2 + iµ3 − iµ4 be the Jordan decomposition of µ. For any f ∈ B(X, A, C),
the integral of f w.r.t µ is defined by:
Z Z Z Z Z
f dµ := f dµ1 − f dµ2 + i f dµ3 − i f dµ4

41
R
Clearly f 7→ f dµ is a linear functional on B(X, A, C). 

Lemma 9.1.3. For any f ∈ B(X, A, R) or B(X, A, C), let ||f ||∞ := sup{|f (x)| :
x ∈ X}. Since f is bounded, ||f ||∞ < ∞. If µ is a finite signed/complex
measure on (X, A, then
Z
| f dµ| ≤ ||f ||∞||µ|| ∀ f ∈ B(X, A, R) or B(X, A, C).

10 Absolute Continuity
Definition 10.0.4. Let (X, A) be a measurable space. Let µ and ν be
positive measures on (X, A). Then ν is said to be absolutely continuous
w.r.t µ if for each set A ∈ A satisfying µ(A) = 0, we have ν(A) = 0. We
denote this by ν << µ. 

Definition 10.0.5. A positive measure on (Rd , B(Rd )) is called absolutely


continuous if it is absolutely continuous w.r.t the lebesgue measure λd . 

Example(s) 10.0.6. Let (X, A, µ) be a measure space where µ is a positive


1
R negative function in L (X, A, µ, R). We know that
measure. Let f be a non
the formula ν(A) = A f dµ defines a positive measure on A. Let ν be a
finite measure since f ∈ L1 . Note that µ(A) =R0 implies that µ({x ∈ A :
f χA (x) = 0}) = 0, which in turn implies that A f χA dµ = 0 which is the
same as saying that ν(A) = 0. Thus ν << µ. 

With reference to the example above, the following lemma characterizes


those finite positive measures that are absolutely continuous with respect to
an arbitrary positive measure:

Lemma 10.0.7. Let (X, A) be a measurable space. Let µ be a positive mea-


sure on (X, A). Let ν be a finite positive measure on (X, A). Then ν << µ if
and only if for each ǫ > 0, there exists a δ > 0 such that each A-measurable
set A that satisfies µ(A) < δ also satisfies ν(A) < ǫ.

Theorem 10.0.8. Radon Nikodym theorem (for positive measures): Let


(X, A) be a measurable space. Let µ and ν be σ-finite positive measures on
(X, A). If ν << µ, thenRthere exists an A-measurable function g : X →
[0, ∞) such that ν(A) = A gdµ holds for each A ∈ A. The function g is
unique upto µ-almost everywhere equality.

42
Theorem 10.0.9. Radon Nikodym theorem (for finite signed or complex
measures): Let (X, A) be a measurable space. Let µ be a σ-finite positive
measure on (X, A). Let ν be a finite signed or complex measure on (X, A).
If ν << µ, then there exists g ∈ L1 (X, A, µ, R) if ν is a finite signed measure
(or g ∈ L1 (X, A, µ, C) if ν is a complex measure) such that ν(A) = A gdµ
R

holds for each A ∈ A. The function g is unique upto µ-almost everywhere


equality.

Definition 10.0.10. Let (X, A) be a measurable space. Let µ be a positive


measure on (X, A). Let ν be a finite signed or complex measure on (X, A).
We say that ν is absolutely continuous w.r.t µ (written as ν << µ) if its
variation |ν| is absolutely continuous w.r.t µ. 

For the following exercises, let (X, A) be a measurable space, µ be a positive


measure on (X, A) and ν be a finite signed or complex measure on (X, A).
Exercise: Show that a signed measure ν is absolutely continuous with
respect to µ if and only if ν + and ν − are absolutely continuous w.r.t µ.
Exercise: Show that a complex measure ν is absolutely continuous with
respect to µ if and only if the measures ν1 , ν2 , ν3 and ν4 appearing in its
jordan decomposition ν = ν1 − ν2 + iν3 − iν4 are absolutely continuous w.r.t
µ.
Exercise: Show that a signed or complex measure ν is absolutely continuous
with respect to µ if and only if each set A ∈ A that satisfies µ(A) = 0 also
satisfies ν(A) = 0.
Exercise: Let ν be a signed or complex measure on (X, A). Let A ∈ A.
(a) Show that |ν|(A) = 0 holds if and only if each A-measurable subset B of
A satisfies ν(B) = 0.
(b) Show that in general the relation ν(A) = 0 does not imply the relation
|ν|(A) = 0.

Definition 10.0.11. Let (X, A) be a measurable space. Let µ be a positive


measure on (X, A). Let ν be a finite signed or complex or σ-finite positive
measure on (X, A) R such that ν << µ. An A-measurable function g on X that
satisfies ν(A) = A gdµ for all A ∈ A is called a Radon-Nikodym derivative
of ν with respect to µ. A Radon-Nikodym derivative of ν with respect to µ

is denoted by dµ , it is unique upto µ-almost everywhere equality. 

Proposition 10.0.12. Let (X, A) be a measurable space. Let µ be a positive


measure on (X, A). Let f ∈ L1 (X, A, µ, R) or L1 (X, A, µ, C). Let ν be the

43
finite signed or complex measure defined by
Z
ν(A) := f dµ.
A
R
Then |ν|(A) = A
|f |dµ holds for each A ∈ A.
Corollary 10.0.13. Let ν be a finite signed or complex measure on a measur-

able space (X, A). Then d|ν| equals the constant function 1 almost everywhere
|ν| on X.
Remark 10.0.14. Let νRbe a finite
R signed or complex measure on a measur-

able space (X, A). Then f dν = f d|ν| d|ν| for each bounded A measurable
function f on X.

11 Functions of Bounded variation


Definition 11.0.15. Let F be a R-valued function whose domain contains
the interval [a, b]. Let S be the collection of all finite sequences {ti }ni=0 such
that
a ≤ t0 < t1 < . . . < tn ≤ b.
Then VF [a, b], the variation of F over [a, b] is defined by

VF [a, b] = Sup{Σni=1 |F (ti ) − F (ti−1 )| : {ti }ni=0 ∈ S}.

The function F is of bounded variation on [a, b] if VF [a, b] is finite. 


The variation VF (−∞, b] of F over the interval (−∞, b] can be defined
similarly, the finite sequences {ti }ni=0 now have the property that:

−∞ < t0 < t1 < . . . < tn ≤ b.

The variation VF (−∞, ∞) of F over the interval (−∞, ∞) can be defined


similarly, the finite sequences {ti }ni=0 now have the property that:

−∞ < t0 < t1 < . . . < tn < ∞.

Definition 11.0.16. Let F be a R-valued function whose domain contains


the interval (−∞, b]. F is said to be of bounded variation on (−∞, b] if
VF (−∞, b] < ∞. 

44
Definition 11.0.17. Let F be a R-valued function whose domain is R. F
is said to be of bounded variation if VF (−∞, ∞) < ∞. 

Definition 11.0.18. If F : R → R is of bounded variation, then the variation


of F is the function VF : R → R defined by

VF (x) := VF (−∞, x] f or each x ∈ R.

Remark 11.0.19. Let µ be a finite signed measure on (R, B(R)). Define a


function Fµ : R → R by letting Fµ (x) = µ((−∞, x]) for all x ∈ R.
If {ti }ni=0 is an increasing (strictly) sequence of real numbers, then

Σni=1 |Fµ (ti ) − Fµ (ti−1 )| = Σni=1 |µ(ti−1 , ti ]| ≤ |µ|(R).

Since µ is a finite signed measure, it now follows that VFµ (−∞, ∞) ≤ |µ|(R) <
∞. Hence the function Fµ is of bounded variation.
The function Fµ also happens to be right continuous and it vanishes at −∞
(that is, limx→−∞ Fµ (x) = 0).

Exercise: Prove that Fµ is continuous on R if and only if µ({x}) = 0 for


each x ∈ R.

11.1 Some general properties of functions of bounded


variation
Proposition 11.1.1. Let F : R → R be of bounded variation. Then
(a) F is bounded.
(b) If −∞ < a < b < ∞, then VF (−∞, b] = VF (−∞, a] + VF [a, b].
(c) If b ∈ R, then VF (−∞, b] = lima→−∞ VF [a, b].
(d) If a < c and F is right continuous at a, then VF [a, c] = limb→a+ VF [b, c].
(→exercise)

Lemma 11.1.2. Let F : R → R be of bounded variation. Let VF be the


function as in definition 11.0.18 above. Then
(a) VF is bounded and non-decreasing.
(b) VF vanishes at −∞ and
(c) If F is right continuous, so is VF .

45
Lemma 11.1.3. Let F : R → R be of bounded variation. Then there are
two bounded, non-decreasing functions F1 and F2 such that F = F1 − F2 .

Remark 11.1.4. Let F : R → R be of bounded variation. Let F1 and F2 be


as in lemma 11.1.3 above. Then
(a) F1 and F2 are right continuous, if F is so.
(b) F1 and F2 vanish at −∞, if F does so.

Proposition 11.1.5. There exists a bijection µ ↔ Fµ between the set of


all finite signed measures on (R, B(R)) and the set of all right-continuous
functions of bounded variation that vanish at −∞, where Fµ : R → R is
given by Fµ (x) = µ((−∞, x]) for eavery x ∈ R.

12 Absolute continuity of functions


Definition 12.0.6. A function f : R → R is called absolutely continuous if
for each ǫ > 0, there exists a δ > 0 such that

Σi |F (ti ) − F (si )| < ǫ

holds whenever {(si , ti )} is a finite sequence of disjoint open intevals for which
Σi (ti − si ) < δ. 

CHECK THAT every absolutely continuous function is continuous, in-


fact, uniformly continuous. However there exist functions that are uniformly
continuous and of bounded variation, but are not absolutely continuous, see
the example below:

Example(s) 12.0.7. Recall the definition of the Cantor singular function


f : [0, 1] → [0, 1] from subsection
 4.2. f can be extended to a function
 0 if x < 0
F : R → R as follows: F (x) := f (x) if x ∈ [0, 1]
1 if x > 1

Exercise: Check that F is uniformly continuous, of bounded variation but
NOT absolutely continuous. 

Remark 12.0.8. An absoluetly continuous function is of bounded variation on


each closed bounded interval. Let F : R → R be absolutely continuous. It
follows from the definition of absolute continuity that for ǫ = 1, there exists a

46
δ > 0 such that Σi |F (ti ) − F (si )| < ǫ = 1 holds whenever {(si , ti )} is a finite
sequence of disjoint open intevals for which Σi (ti − si ) < δ.
Let [a, b] be a closed bounded interval. Exercise: Show that if {ui }ni=0 is a
finite sequence such that

a = u0 < · · · < un = b,

then Σni=1 |F (ui ) − F (ui−1)| ≤ b−a


δ
+ 1. It will then follow easily that F is
of bounded variation on [a, b]. But an absoluetly continuous function is NOT
necessarily of bounded variation on R. For example, the function F (x) = x ∀ x ∈
R is absolutely continuous but not of bounded variation on R.
Proposition 12.0.9. Let µ be a finite signed measure on (R, B(R)). Let
Fµ : R → R be defined as Fµ (x) := µ(−∞, x] for all x ∈ R. Then Fµ is
absolutely continuous if and only if µ << λ where λ denotes the lebesgue
measure on (R, B(R)).
Lemma 12.0.10. If F : R → R is absolutely continuous and of bounded
variation, then VF is absolutely continuous.
Corollary 12.0.11. The functions F : R → R that arise through the formula
Z x
F (x) = f (t)dt
−∞

for some f ∈ L1 (R, B(R), λ, R) are exactly those that are of bounded varia-
tion, absolutely continuous and vanish at −∞.

13 Product measures
Definition 13.0.12. Let (X, A) and (Y, B) be measurable spaces. Consider
the cartesian product X × Y . A subset of X × Y is said to be a rectangle
with measurable sides if it is of the form A × B for some A ∈ A and B ∈ B.
The σ-algebra on X × Y generated by the collection of all rectangles with
measurable sides is called the product of the σ-algebras A and B and is
denoted by A × B. That is,

A × B := σ({A × B|A ∈ A, B ∈ B}).

47
Exercise: Prove that B(R) × B(R) = B(R2 ).
Exercise: Let M1 be the σ-algebra of all lebesgue measurable subsets of
R and M2 be the σ-algebra of all lebesgue measurable subsets of R2 . Show
that M2 6= M1 × M1 .

Definition 13.0.13. Let X and Y be sets and E ∈ X × Y . Then for each


x ∈ X and each y ∈ Y , the sections Ex and E y are defined as

Ex := {y ∈ Y |(x, y) ∈ E}

and
E y := {x ∈ X|(x, y) ∈ E}.
If f is a function on X × Y , then the sections fx and f y are the functions on
Y and X defined by

fx (y) = f (x, y) and f y (x) = f (x, y).

Lemma 13.0.14. Let (X, A) and (Y, B) be measurable spaces.


(a) If E ∈ A × B, then each section Ex ∈ B and E y ∈ A.
(b) If f : X × Y → [−∞, ∞] or C is A × B measurable, then each section fx
is B-measurable and each section f y is A-measurable.

Proposition 13.0.15. Let (X, A, µ) and (Y, B, ν) be σ-finite measure spaces.


If E ∈ A × B, then the function x 7→ ν(Ex ) is A-measurable and the function
y 7→ µ(E y ) is B-measurable.

Theorem 13.0.16. Let (X, A, µ) and (Y, B, ν) be σ-finite measure spaces.


Then there exists a unique measure µ × ν on the σ-algebra A × B such that

(µ × ν)(A × B) = µ(A)ν(B)

holds for each A ∈ A and B ∈ B. Furthermore the measure µ × ν of an


arbitrary set E in A × B is given by
Z Z
(µ × ν)(E) = ν(Ex )dµ = µ(E y )dν.
X Y

The measure µ × ν is called the product of µ and ν.

48
Exercise: Let λ1 be the lebesgue measure on B(R) and λ2 be the lebesgue
measure on B(R2 ). Show that λ2 = λ1 × λ1 .
Exercise: Show that any (d − 1) dimensional hyperplane in Rd has measure
zero under the lebesgue measure λd . [Definition: A (d − 1) dimensional
hyperplane in Rd is a set of the form {x = (x1 , . . . , xd ) ∈ Rd : |Sigmai ai xi =
b} for some b ∈ R and some vector (a1 , . . . , ad ) ∈ Rd such that (a1 , . . . , ad ) 6=
(0, . . . , 0).]

Theorem 13.0.17. Let (X, A, µ) and (Y, B, ν) be σ-finite measure spaces.


Let f : X × Y → [0, ∞]R be A × B-measurable. Then R
(a) the function x 7→ Y fx dν is A-measurable and the function y 7→ X f y dµ
is B-measurable.
(b) f satisfies
Z Z Z Z Z
y
f d(µ × ν) = ( f dµ)dν = ( fx dν)dµ.
X×Y Y X X Y

Remark 13.0.18. The formula in part (b) of theorem 13.0.17 is applicable


to each non-negative A × B-measurable function f , which may or may not be
integrable. One can often determine whether an A × B-measurable function
Rf is integrable or not by using part (b) of theorem 13.0.17 above to calculate
X×Y
|f |d(µ × ν).

Theorem 13.0.19. Fubini’s theorem: Let (X, A, µ) and (Y, B, ν) be σ-


finite measure spaces. Let f : X × Y → [−∞, ∞] be A × B-measurable and
µ × ν-integrable. Then
(a) For µ-almost every x ∈ X, the section fx is ν-integrable and for ν-almost
every y ∈ Y , the section f y is µ-integrable.  R
f dν if fx is ν-integrable
Y x
(b) The functions If and Jf defined by If (x) =
 R y 0 otherwise
y
f dµ if f is µ-integrable
and Jf (y) = X
0 otherwise
belong to L (X, A, µ, R) and L1 (Y, B, ν, R) respectively and
1

(c) The relation


Z Z Z
f d(µ × ν) = Jf dν = If dµ
X×Y Y X

holds.

49
We can deal with a C-valued function on X × Y by separating it into its
real and imaginary parts.
Exercise: Let λ be the lebesgue measure on (R, B(R)) and mu be the
counting measure on (R, B(R)). Let f : R2 → R beR the
R characteristic func-
2
tion
RR of the line {(x, y) ∈ R |y = x}. Show that f (x, y)µd(y)λd(x) 6=
f (x, y)λd(x)µd(y).

13.1 Product of finite number of measure spaces


Let (X1 , A1 , µ1), . . . , (Xn , An , µn ) be σ-finite measure spaces. Then

A1 × · · · × An = σ({A1 × · · · × An |Ai ∈ Ai f or i = 1, . . . , n}).

Using theorem 13.0.16 (applied n − 1 times), we can construct a unique


measure µ1 × · · · µn on A1 × · · · × An that satisfies

(µ1 × · · · µn )(A1 × · · · × An ) = µ(A1 ) · · · µ(An )

whenever Ai ∈ Ai for i = 1, . . . , n. Integrals of functions with respect to


µ1 × · · · µn can be evaluated by repeated applications of Fubini’s theorem.

50

You might also like