Brief Notes on Measure Theory

John K. Hunter, January 27, 2007

Measure Spaces

Suppose that X is a non-empty set. A collection A of subsets of X is

called a σ-algebra on X if it contains X and is closed under the operations
of taking complements, countable unions, and countable intersections:

• ∅, X ∈ A;

• if E ∈ A, then E c ∈ A;

• if En ∈ A for n ∈ N, then
∞
[ ∞
\
En ∈ A, En ∈ A.
n=1 n=1

Of course, the fact that A is closed under countable intersections is im-

plied by the fact that it is closed under complements and countable unions.
If X is a set and A is a σ-algebra on X, then we call (X, A) a measurable
space, and elements of A are called measurable sets.
1 Example If X is any nonempty set, then the largest σ-algebra on X is
the power set P(X) consisting of all subsets of X. The smallest σ-algebra is
{∅, X}.
If F is any collection of subsets of X, then the smallest σ-algebra contain-
ing F is called the σ-algebra generated by F. This σ-algebra is the intersection
of all σ-algebras that contain F.
2 Example If X = R is the set of real numbers and G is the collection of
open sets, then the σ-algebra generated by G is called the Borel σ-algebra
on R, which we denote by B(R). Elements of B(R) are called Borel sets.
The σ-algebra B(R) is also generated by the collection F of closed sets, or
by various collection of intervals, such as {(a, ∞) | a ∈ R}. We define the
Borel σ-algebra B(Rd ) on Rd in a similar way. More generally, if X is any
topological space, then the Borel σ-algebra B(X) is the σ-algebra generated
by the open sets of X.

1
3 Remark It is surprisingly complicated to obtain B(R) by starting from
the open sets and taking successively complements, countable unions, and
countable intersections. This process requires uncountably many iterations
before it gives B(R).
If (X, A) is a measurable space, then a map
µ : A → [0, ∞]
is called a measure if:
• µ(∅) = 0;
• if {En | n ∈ N} is a collection of disjoint sets in A, meaning that
Em ∩ En = ∅ for m 6= n, then
∞
! ∞
[ X
µ En = µ (En ) .
n=1 n=1

Here, we use the obvious conventions for measures and sums that are
equal to ∞.
The second property in the definition of a measure is called ‘countable addi-
tivity’, and it expresses the idea that the ‘volume’ of a disjoint union should
be the sum of the ‘volumes’ of its individual parts. We call (X, A, µ) a
measure space.
It follows from the definition that: (a) if E ⊂ F are measurable sets,
then µ(E) ≤ µ(F ); (b) if E1 ⊂ E2 ⊂ E3 ⊂ . . . is an increasing sequence of
measurable sets, then
∞
!
[
µ En = lim µ(En );
n→∞
n=1

(c) if E1 ⊃ E2 ⊃ E3 ⊃ . . . is a decreasing sequence of measurable sets with

finite measure for some n ∈ N (see the following example), then
∞
!
\
µ En = lim µ(En ).
n→∞
n=1

4 Example Let X be any set. Define ν : P(X) → [0, ∞] by

ν(E) = # points in E.

2
Then ν is a measure on P(X) called counting measure. If X = N and

En = {1, 2, 3, . . . , n} ,

then (En ) is an increasing sequence of sets whose union is N. We have

ν(En ) = n, ν(N) = ∞, and ν(En ) → ν(N) as n → ∞. On the other hand, if

En = {n, n + 1, n + 2, . . .} ,

then (En ) is an decreasing sequence of sets whose intersection is ∅. We have

ν(En ) = ∞, ν(∅) = 0, and ν(En ) 6→ ν(∅) as n → ∞.
5 Example There is a unique measure m : B(R) → [0, ∞] on R, called
Lebesgue measure, such that

m (I) = Length(I)

for any interval I ⊂ R. Here, the length of an interval is defined in the usual
way; for example, Length ((a, b]) = b − a. More generally, there is a unique
Lebesgue measure m : B(Rd ) → [0, ∞] on Rd such that

m (I) = Volume(I)

for any d-dimensional rectangle I ⊂ Rd . By a rectangle I in Rd , we mean a

set of the form I = I1 × I2 ×, . . . × Id , where the Ik ⊂ R are intervals, and
the volume of I is the product of the lengths of its sides Ik . The Lebesgue
measure is of a set is invariant under translations and rotations.
We say that a measure space (X, A, µ) is finite if the set X has finite
measure, and σ-finite if X is a countable union of sets of finite measure. For
example, any probablity space, which is a measure space such that µ(X) = 1
is finite, and Rd equipped with Lebesgue measure is σ-finite but not finite.
A set E has measure zero if E ∈ A and µ(E) = 0. A property that holds
except on a set of measure zero is said to hold almost everywhere (or a.e. for
short). For example, we say that two functions f, g : X → Y are equal a.e.
if µ{x ∈ X | f (x) 6= g(x)} = 0.
6 Example The Lebesgue measure of any singleton {a} ⊂ R is zero. It
follows from the countable additivity of a measure that the Lebesgue measure
of any countable set is zero; for example, the set of rationals has Lebesgue
measure zero. There are, however, many uncountable sets whose Lebesgue

3
measure is zero; for example, the standard Cantor set has Lebesgue measure
zero. In general, the Lebesgue measure of a subset N ⊂ Rd is zero if and
only if for every  > 0 there is a countable collection {In | n ∈ N} of, not
necessarily disjoint, d-dimensional rectangles such that
∞
[ ∞
X
N⊂ In , Volume(In ) < .
n=1 n=1

A measure space (X, A, µ) is complete if every subset of a set of measure

zero is measurable (when its measure is necessarily zero). Every measure
space (X, A, µ) has a unique completion (X, A, µ), which is the smallest
complete measure space such that A ⊃ A and µ|A = µ.
7 Example Lebesgue measure on the Borel σ-algebra (R, B(R), m) is not
complete, meaning that there are Borel sets of Lebesgue measure zero which
contain subsets that are not Borel sets. The completion of the Borel σ-
algebra with respect to Lebesgue measure is the σ-algebra L(R) of Lebesgue
measurable sets. Similarly, the σ-algebra L(Rd ) of Lebesgue measurable sets
in Rd is the completion of the σ-algebra B(Rd ) of Borel measurable sets with
respect to Lebesgue measure.
8 Remark It is not possible to extend Lebesgue measure to a countably
additive measure defined on all subsets of R. To ‘construct’ a nonmeasurable
set, let x ∼ y be the equivalence relation on R defined by x − y ∈ Q. Then
R is a disjoint union of uncountably many equivalence classes of ∼, each
of which contains countably many elements. Using the axiom of choice, we
pick an uncountable set E ⊂ R containing exactly one element from each
equivalence class of ∼. One can show that E is not Lebesgue measurable.
(This follows from the fact that Lebesgue measure is translation invariant
and countably additive, and R is a countable disjoint union of translates of
E by rational numbers.) Solovay (1970) proved that the axiom of choice is
required to show the existence of a set which is not Lebesgue measurable.
9 Remark In Rd with d ≥ 3 it is possible to decompose a ball (using the
axiom of choice) into a finite number of (non-Lebesgue measurable) subsets
and reassemble these pieces into two balls of the same volume. This Banach-
Tarski paradox shows that in d ≥ 3 space-dimensions it is not even possible
to define a finitely-additive translation and rotation invariant set function on
all subsets of Rd whose value on any ball is equal to its volume. There exist

4
finitely-additive translation and rotation invariant measures on R2 , but they
still have some counter-intuitive properties. For example, Laczkovich (1990)
proved that it is possible to decompose a circular disc into a finite number of
pieces (in Laczkovich’s proof, approximately 1050 non-Lebesgue measurable
sets) and reassemble them into a square of the same area.

Measurable Functions

If (X, A), (Y, B) are measurable spaces, we say that a function f : X → Y

is measurable if f −1 (E) ∈ A for every E ∈ B.
If B is generated by F then f is measurable if and only if f −1 (E) ∈ A
for every E ∈ F. For example, if X, Y are topological spaces equipped
with their Borel σ-algebras, then every continuous function f : X → Y is
measurable.
In the case when Y = R and

f : X → R,

we equip R with its Borel σ-algebra B(R). Then f is measurable if and only
if f −1 ((a, ∞)) ∈ A for every a ∈ R. A function

f :X→C

is measurable if and only if its real and imaginary parts <f and =f are
measurable.
It is often convenient to consider extended real-valued functions

f : X → R,

where R = {−∞} ∪ R ∪ {∞}, with the appropriate Borel σ-algebra. Then f

is measurable if and only if f −1 ((a, ∞]) is measurable for every a ∈ R.
10 Example The characteristic function (or indicator function) of a subset
E ⊂ X is the function χE : X → R defined by

1 if x ∈ E,
χE (x) =
0 if x ∈
/ E.

Then χE is a measurable function if and only if E is a measurable set.

5
 A function φ : X → R is a simple function if
N
X
φ(x) = cn χEn (x)
n=1

for some c1 , . . . , cN ∈ R and E1 , . . . , EN ∈ A. Note that, according to this

definition, a simple function is measurable.
We say that a sequence of functions fn : X → R converges pointwise to
a function f : X → R if fn (x) → f (x) as n → ∞ for every x ∈ X. (Here,
we use the obvious conventions about sequences that converge to ±∞.) We
say that fn → f pointwise a.e. if fn (x) → f (x) as n → ∞ except for x ∈ N
where N ⊂ X is a set of measure zero.
11 Theorem If fn : X → R are measurable functions and fn → f pointwise
as n → ∞, then f : X → R is measurable. If X is a complete measure space
and fn → f pointwise a.e. as n → ∞, then f : X → R is measurable.

Integration

If φ : X → [0, ∞) is P a nonnegative simple function on a measure space

(X, A, µ), given by φ = N n=1 cn χEn where 0 ≤ cn < ∞ and En ∈ A, then
we define the integral of φ with respect to the µ by
Z XN
φ dµ = cn µ (En ) .
n=1

We use the convention that 0 · ∞ = 0, meaning that the integral of a function

that is 0 on a set of measure ∞, or ∞ on a set of measure 0, is equal to 0.
12 Example The characteristic function χQ : R → R of the rationals is
not Riemann integrable on any compact interval of non-zero length, but it is
Lebesgue integrable with
Z
χQ dm = 1 · m(Q) = 0.

If f : X → [0, ∞] is a nonnegative, measurable, extended real-valued

function, we define
Z Z 
f dµ = sup φ dµ | 0 ≤ φ ≤ f , φ simple .

6
13 Remark In this definition, we approximate the function f from below by
simple functions. In contrast with the definition of the Riemann integral, it
is not necessary to approximate a measurable function from both above and
below in order to define its integral.
If f : X → R, we write f = f+ − f− in terms of its positive and negative
parts
f+ = max{f, 0}, f− = max{−f, 0},
and define Z Z Z
f dµ = f+ dµ − f− dµ,
R R
provided that at least one of the integrals f+ dµ, f− dµ is finite.
We also have Z Z Z
|f | dµ = f+ dµ + f− dµ.

We say that f : X → R is integrable if

Z
|f | dµ < ∞,
R R
which occurs if and only if both f+ dµ, f− dµ are finite.
If f : X → C is a complex valued function f = g+ih, then f is measurable
if and only if its real and imaginary parts g, h : X → R are measurable, and
integrable if and only if g, h are integrable. In that case, we define
Z Z Z
f dµ = g dµ + i h dµ.

If A ⊂ X is measurable, then we define

Z Z
f dµ = f χA dµ.
A

In contrast with the Riemann integral, where integrals over non-rectangular

subsets of R2 already present problems, it is trivial to define the Lebesgue
integral over arbitrary measurable subsets.
This Lebesgue integral has all the usual properties of an integral. For
example, if f, g : X → R and f ≤ g then
Z Z
f dµ ≤ g dµ,

7
and if f, g : X → C, λ ∈ C then
Z  Z
 
 f dµ ≤ |f | dµ,
 
Z Z
λf dµ = λ f dµ,
Z Z Z
(f + g) dµ = f dµ + g dµ.

14 Example Suppose that X = N and ν : P(N) → [0, ∞] is counting

measure on N. If f : N → R and f (n) = xn , then
Z ∞
X
f dν = xn ,
N n=1

where the integral converges if and only if the series is absolutely conver-
gent. Thus, the theory of absolutely convergent series is a special case of
the Lebesgue integral. Note that conditionally convergent series, such as the
alternating harmonic series, do not correspond to a Lebesgue integral, since
both their positive and negative parts diverge.
15 Example Any Riemann integrable function f : [a, b] → R is integrable
with respect to Lebesgue measure, and the Riemann integral is equal to the
Lebesgue integral, Z b Z
f (x) dx = f dm.
a [a,b]

Thus, all of the usual results from elementary calculus remain valid for the
Lebesgue integral on R. A Lebesgue integrable function, however, need not
be Riemann integrable. In fact, a bounded function f : [a, b] → R is Riemann
integrable if and only if it is Lebesgue measurable and the set of disconti-
nuities {x ∈ [a, b] | f is discontinuous at x} has Lebesgue measure zero. For
example, the characteristic function of a Cantor set with non-zero measure
is Lebesgue integrable, but it is not Riemann integrable, nor is any modifi-
cation of the function on a set of measure zero Riemann integrable. We will
often write an integral with respect to Lebesgue measure on R or Rd as
Z
f dx.

8
 Convergence Theorems

One of the most basic questions in integration theory is the following: If

fn → f pointwise, when can one say that
Z Z
fn dµ → f dµ?

Some condition is necessary to ensure the convergence of the integrals, as

can be seen from very simple examples.
16 Example Define fn : R → R by

n if 0 < x < 1/n,
fn (x) =
0 otherwise.

Then fn → 0 as n → ∞ pointwise on R, but

Z
fn dx = 1 for every n ∈ N.

The Riemann integral is not sufficiently general to permit a satisfactory

answer to this question. For example, one can construct a monotone in-
creasing sequence of nonnegative continuous functions fn : [0, 1] → R that
is bounded from above by 1 but whose pointwise limit is not even Riemann
integrable.
The Lebsegue integral allows us to formulate simple and widely applicable
conditions for the convergence of the integrals. The most important of these
are that the pointwise convergence is monotone (the Monotone Convergence
Theorem), or that every function fn in the sequence is bounded by the same
integrable function (the Lebesgue Dominated Convergence Theorem).
17 Theorem [Monotone Convergence] Suppose that (fn ) is an increasing
sequence of nonnegative measurable functions fn : X → [0, ∞] converging
pointwise to f : X → [0, ∞], meaning that fn (x) ≤ fn+1 (x) for every n ∈ N
and x ∈ X. Then Z Z
fn dµ → f dµ as n → ∞.

Here, we use the obvious conventions about ∞.

9
18 Theorem [Dominated Convergence] Suppose that (fn ) is an sequence of
measurable functions fn : X → C such that fn → f pointwise and |fn | ≤ g
where g : X → [0, ∞] is an integrable function, meaning that
Z
g dµ < ∞.

Then Z Z
fn dµ → f dµ as n → ∞.

Here, all the integrals are necessarily finite.

Product Measures

Fubini’s theorem provides a simple and general condition for the equality
of multiple and iterated integrals of a function: namely, the function should
be integrable (see Theorem 23 for multiple integrals on Rd ).
Suppose that (X, A), (Y, B) are measurable spaces. The product σ-algebra
on the Cartesian product X × Y is the sigma algebra A ⊗ B generated by
the collection of all measurable rectangles {A × B | A ∈ A, B ∈ B}. Note
that this collection is not itself a σ-algebra; for example, the union of two
rectangles is not in general another rectangle.
19 Example Consider Rm , Rn equipped with the Borel σ-algebras B(Rm ),
B(Rn ). Then Rm × Rn = Rm+n , and one can show that

B(Rm+n ) = B(Rm ) ⊗ B(Rn ).

Thus, the Borel σ-algebra on Rd can be obtained as a d-fold product of the

Borel σ-algebra on R. Unfortunately, this result does not remain quite true
for Lebesgue measurable sets because the product of complete σ-algebras is
not necessarily complete. For example, if N ⊂ R is not a Borel measurable
subset of R, then {0}×N is not a Borel-measurable subset of R2 even though
it is contained in the subset {0k×R of zero Lebesgue measure on R2 . Instead
on can show that the Lebesgue σ-algebra on Rm+n is the completion with
respect to Lebesgue measure of the product of the Lebesgue σ-algebras on
Rm and Rn :
L(Rm+n ) = L(Rm ) ⊗ L(Rn ).

10
19 Theorem Suppose that (X, A, µ), (Y, B, ν) are σ-finite measure spaces.
There is a unique measure µ ⊗ ν : A ⊗ B → [0, ∞] on X × Y such that

(µ ⊗ ν)(A × B) = µ(A)ν(B) for all A ∈ A, B ∈ B.

If f : X × Y → C is a function of (x, y) ∈ X × Y , then for each x ∈ X we

define the x-section fx : Y → C and for each y ∈ Y we define the y-section
f y : Y → C by
fx (y) = f (x, y), f y (x) = f (x, y).

20 Theorem If (X, A, µ), (Y, B, ν) are σ-finite measure spaces and f : X ×

Y → C is a measurable function, then fx : Y → C, f y : X → C are
measurable for every x ∈ X, y ∈ Y . Moreover, the functions g : X → C,
h : Y → C defined by
Z Z
g(x) = fx dν, h(y) = f y dµ

are measurable.
21 Theorem [Fubini] A measurable function f : X × Y → C is integrable if
and only if either one of the iterated integrals
Z Z  Z Z 
y
|f | dµ dν, |fx | dν dµ

is finite. In that case

Z Z Z  Z Z 
y
f dµ ⊗ dν = f dµ dν = fx dν dµ.

22 Example An application of Fubini’s theorem to counting measure on

N × N implies that if {amn ∈ C | m, n ∈ N} is a doubly-indexed sequence of
complex numbers such that
∞ ∞
!
X X
|amn | < ∞
m=1 n=1

then
∞ ∞
! ∞ ∞
!
X X X X
amn = amn .
m=1 n=1 n=1 m=1

11
 Finally, we state a version of Fubini’s theorem for functions on Rd .
23 Theorem A Lebesgue measurable function f : Rm+n → C is integrable,
meaning that Z
|f (x, y)| dxdy < ∞,
Rm+n
if and only if either one of the iterated integrals
Z Z  Z Z 
|f (x, y)| dx dy, |f (x, y)| dy dx
Rn Rm Rm Rn

is finite. In that case,

Z Z Z  Z Z 
f (x, y) dxdy = f (x, y) dx dy = f (x, y) dy dx,
Rm+n Rn Rm Rm Rn

where all of the integrals are well-defined and finite a.e..

Lp -spaces

R (X, A, µ) is a measure space. If f : X → C is a measurable

Suppose that
function and |f | dµ = 0 then it does not follow that f = 0, only that the
set on which f is non-zero has measure zero. We will henceforth regard two
measurable functions that differ from each other on a set of measure zero
as the ‘same’ function. Thus, we identify a function with the equivalence
class to which it belongs under the equivalence relation of a.e.-equality. For
example, the characteristic
R function χQ of the rationals on R is equivalent to
0. With this convention, |f | dµ = 0 implies that f = 0.
For 1 ≤ p < ∞, we define the space Lp (X) to consist of all (equivalence
classes of) measurable functions f : X → C such that
Z
|f |p dµ < ∞.

For f ∈ Lp (X), we define

Z 1/p
p
kf kp = |f | dµ .

One can also define the space L∞ (X) of essentially bounded measurable
functions, with norm given by the essential supremum. (Here ‘essential’
means ‘up to sets of measure zero’, but we omit the details.)

12
24 Theorem If 1 ≤ p < ∞, then Lp (X) equipped with the norm k · kp is a
Banach space.
The theorem includes the statement that the Lp -norm satisfies the trian-
gle inequality,
kf + gkp ≤ kf kp + kgkp .
This is called Minkowski’s inequality and it does not hold for p < 1, which
explains the restriction on p. It also includes the statement that Lp is com-
plete, meaning that if (fn )∞ p
n=1 is any Cauchy sequence in L , then there exists
f ∈ Lp such that kfn − f kp → 0 as n → ∞. The sequence need not converge
pointwise a.e., but there is a subsequence (fnk )∞
k=1 such that fnk → f point-
wise a.e. as k → ∞. Any function in Lp is the pointwise a.e. and Lp -limit
of a sequence of simple functions.
Finally, we give a theorem which states that Lp -functions on Rd can be
approximated by continuous functions. We consider functions f : Ω → C
where Ω ⊂ Rd is an arbitrary nonempty open set. We denote by Cc (Ω) the
set of continuous functions f : Ω → C with compact support in Ω, meaning
that the closure in Ω of the set {x ∈ Ω | f (x) 6= 0} is a compact set. A
function f ∈ Cc (Ω) is bounded and nonzero on a bounded set, so f ∈ Lp (Ω).
25 Theorem Cc (Ω) is a dense subset of Lp (Ω).
More explicitly, this theorem states that if f ∈ Lp (Ω) then, given any
 > 0, there exists g ∈ Cc (Ω) such that

kf − gkp < .

It follows that Lp (Ω) is the completion of Cc (Ω) with respect to the Lp -norm.

13
 References

A readable, insightful introduction to Lebesgue measure is given in:

E. M. Stein and R. Shakarchi, Real Analysis, Princeton, 2005.
For a leisurely, clear, and well-motivated introduction to Lebesgue measure
and integration on Rn see:
F. Jones, Lebesgue Integration on Euclidean Space, Jones & Bartlett 1993.
A detailed, individual discussion of measure theory on Rn is given in:
E. H. Lieb and M. Loss, Analysis, AMS 1997.
An excellent, but more condensed, standard text is:
G. B. Folland, Real Analysis, Wiley, 1984.
Another good, recent text is:
M. E. Taylor, Measure Theory and Integration, AMS, 2006.

14