Lecture Notes on Measure Theory and Integration

2 — Measures

Joel Feldman

University of British Columbia

October 10, 2019

1
2 Measures
2.1 A Problem
In an ideal world, we would like to be able to extend the definition “for all a ≤ b, the
length of the interval [a, b] is b − a” by assigning a size µ(E) to every subset E ⊂ R
with the properties that

(i) µ : P(R) = E E ⊂ R → [0, ∞]

(ii) Countable additivity: If Ei i∈N is a sequence of disjoint subsets of R, then
[  X
µ Ei = µ(Ei )
i∈N i∈N

(To get a union of finitely many sets, just make Ei = ∅ for most i’s.)
(iii) Translation invariance: For all E ⊂ R and y ∈ R

µ(E + y) = µ(E) where E + y = x+y x∈E

(iv) For any a ≤ b,

µ [a, b] = µ (a, b] = µ [a, b) = µ (a, b) = b − a

There is a problem with this. It is impossible to simultaneously satisfy all of these

conditions. They are mutually inconsistent. Here is an example which shows this.
Example 2.1. In this example we will express
[
[0, 1) = Nr
r∈Q∩[0,1)

with the Nr ’s
◦ all disjoint and
◦ all having the same size. Say µ(Nr ) = µN , for all r.
Once we have done so, we will have

X
X
1 = µ [0, 1) = µ Nr = µN
r∈Q∩[0,1) r∈Q∩[0,1)
(
0 if µN = 0
=
∞ if µN > 0

Measures 2 October 10, 2019

which is a contradiction.
Now we construct the Nr ’s. Define an equivalence relation1 on [0, 1) by
x∼y ⇐⇒ x−y ∈Q
and define, for each x ∈ [0, 1), the equivalence class of x to be

[x] = y ∈ [0, 1) y ∼ x, i.e. y − x ∈ Q
As is the case for all equivalence relations, for each pair x, y ∈ [0, 1), either [x] = [y]
or [x] ∩ [y] = ∅. Consequently [0, 1) is the disjoint union of all equivalence classes.
Now, let2 N contain exactly one member of each equivalence class and define, for
each r ∈ Q ∩ [0, 1),

Ñr = N + r = x + r x ∈ N

[

Nr = Ñr ∩ [0, 1) Ñr ∩ [1, 2) − 1
≥r ≥0 <r 0≤x+r−1<1
 z }| { z }| { [  z }| { z }| {
= x+r x ∈ N, 0 ≤ x + r < 1 x+r−1 x ∈ N, 1 ≤ x+r < 2

0 1
0 N 1 N +r Nr Nr

Then
(a) Nr ⊂ [0, 1) for each r ∈ Q ∩ [0, 1)
(b) if r, s ∈ Q ∩ [0, 1) with r 6= s, then Nr ∩ Ns = ∅ because
if y ∈ Nr , (i.e. y = x + r or x + r − 1 with x ∈ N)
and y ∈ Ns (i.e. y = x′ + s or x′ + s − 1 with x′ ∈ N)
then x − x′ = s − r + (−1 or 0 or 1) ∈ Q
which implies that x ∼ x′
and x 6= x′ (since r 6= s and 0 ≤ r, s < 1 so that −1 < s − r < 1)
so that N contains two different elements from the equivalence class [x]
which is a contradiction
1
An equivalence relation x ∼ y has the properties
x∼x
x ∼ y =⇒ y ∼ x
x ∼ y, y ∼ z =⇒ x ∼ z

2
We’re using the axiom of choice here.

Measures 3 October 10, 2019

 (c) µ(Nr ) = µ(N) because



µ(Nr ) = µ Ñr ∩ [0, 1) + µ Ñr ∩ [1, 2) − 1



= µ Ñr ∩ [0, 1) + µ Ñr ∩ [1, 2)


= µ Ñr = N + r
= µ(N)
[
(d) [0, 1) = Nr because
r∈Q∩[0,1)

y ∈ [0, 1) =⇒ ∃ x ∈ N with x ∈ [y] i.e. x ∼ y

=⇒ y = x + r for some r ∈ Q ∩ (−1, 1)
=⇒ y ∈ Nr if r ∈ [0, 1)
∈[1,2)
z }| {
or y = (x + r} +1) −1 ∈ Nr+1 if r ∈ (−1, 0) i.e. r + 1 ∈ (0, 1)
| {z
∈[0,1)

So these Nr ’s satisfy all of the required properties.

We have just seen that it is impossible to define a size function µ(E) which
simultaneously has all four of the properties (i), (ii), (iii), (iv) above. We have to
weaken one or more of those properties. In this course, we will deal with this problem
in two different ways.
• the σ-algebra approach: under this approach, we weaken property (i) and only
require that µ(E) be defined for certain sets E (that include all sets that we are
likely to encounter in practice).
• the outer measure approach:
S
under this approach, we weaken property (ii) and
P
only require that µ i Ei = i µ(Ei ) hold for certain collections of sets (that
include all sets that we are likely to encounter in practice).

2.2 σ-algebras
Let X be a nonempty set. We denote by P(X) the set of all subsets of X.
Definition 2.2 (Algebras). A nonempty collection A of subsets of X is an algebra
if
(i) A, B ∈ A =⇒ A ∪ B ∈ A
(ii) A ∈ A =⇒ Ac = X \ A ∈ A

Measures 4 October 10, 2019

Remark 2.3 (More about algebras). An algebra A of subsets of X also (trivially)
obeys
(i’) A1 , · · · , An ∈ A =⇒ A1 ∪ · · · ∪ An ∈ A

c
(iii) A1 , · · · , An ∈ A =⇒ A1 ∩ · · · ∩ An = Ac1 ∪ · · · ∪ Acn ∈ A
(iv) ∅ = A ∩ Ac ∈ A
(v) X = ∅c ∈ A
Definition 2.4 (σ-algebras). A collection A of subsets of X is a σ -algebra if it is
an algebra that is closed under countable unions. That is,
∞
[

An n∈N
⊂ A =⇒ An ∈ A
n=1

We shall eventually define a measure on a σ-algebra A to be a function µ : A → [0, ∞]

that obeys certain properties.
Example 2.5. For any nonempty set X,
• P(X) is a σ-algebra

• ∅ , X is a σ-algebra
Example 2.6. For any uncountable3 set X,

A = E ⊂ X E is countable or E c is countable

is a σ-algebra. If E c is countable4 , E is said to be co-countable. That A is a σ-algebra

follows easily from the observation that, if En ∈ A for all n ∈ N, then
S
◦ either at least one Em is co-countable, in which case n∈N En ⊃ Em is co-
countable
S
◦ or every En is countable, in which case n∈N En is countable.

2.3 Techniques for Building σ-algebras

• T
If I is any (index) set and, for each i ∈ I, Ai is a σ-algebra of subsets of X then
i∈I Ai is a σ-algebra.
T
Proof. ◦ For each i ∈ I, we have ∅ ∈ Ai , so i∈I Ai is not empty.
3
This example is not interesting if X is not uncountable — then A reduces to P(X).
4
For a review of countablility, see the notes “Cardinality”. By definition, a set is countable if it
has the same number of elements as a subset of the natural numbers. In particular, finite sets are
countable.

Measures 5 October 10, 2019

  T
◦ If En n∈N
⊂ i∈I Ai then,
 [
for each i ∈ I, En n∈N
⊂ Ai =⇒ En ⊂ Ai
n∈N
[ \
=⇒ En ⊂ Ai
n∈N i∈I
T
So i∈I Ai is closed under countable unions.
◦ Similarly for complements.

• If X is any nonempty set and E ⊂ P(X), then the σ -algebra generated by E is

\
M(E) = A A is a σ-algebra containing E

(Note that A A is a σ-algebra containing E is not empty — P(X) is always
an element.) That is, M(E) is the smallest σ-algebra that contains E. That
smallest σ-algebra always exists.
Example 2.7 (Borel σ-algebra). If X is a metric space (or, more generally, a topolog-
ical space) then the Borel σ -algebra on X, denoted BX , is the σ-algebra generated
by the family of open subsets of X (or the family of closed subsets of X). BX always
contains
• all open subsets of X
• all closed subsets of X
• all countable intersections of open subsets of X (these are called5 Gδ sets)
• all countable unions of closed subsets of X (these are called6 Fσ sets)
Lemma 2.8. Let X be a nonempty set.
(a) If E, F ⊂ P(X) and E ⊂ M(F ), then M(E) ⊂ M(F ).
(b) If E, F ⊂ P(X) and E ⊂ M(F ) and F ⊂ M(E), then M(E) = M(F ).
Proof. (a) is obvious as M(F ) is a σ-algebra containing E.
(b) is trivial from (a).

5
The “G” stands for “Gebeit”, which is German for “neighbourhood” and the δ stands for
“Durchschnitt”, which is German for “intersection”.
6
The “F ” stands for “Fermé”, which is French for “closed” and the σ stands for “somme”, which
is French for “sum” or “union”.

Measures 6 October 10, 2019

Proposition 2.9. BR , the Borel σ-algebra on R, is generated by

(a) (a, b) a < b

(b) [a, b] a < b
 
(c) (a, b] a < b or [a, b) a < b
 
(d) (a, ∞) a ∈ R or (−∞, a) a ∈ R
 
(e) [a, ∞) a ∈ R or (−∞, a] a ∈ R
 
Proof. (a) Let O = A ⊂ R A open and E = (a, b) a < b .
• M(E) ⊂ BR :
E ⊂ O ⊂ M(O) = BR =⇒ M(E) ⊂ BR
• BR ⊂ M(E): As M(O) = BR it suffices to prove that O ⊂ M(E). Let A ∈ O.
It now suffices to prove that A is a countable union of open intervals, since that
implies that A ∈ M(E).
Proof that A is a countable union of open intervals. Let x ∈ A.
A open =⇒ ∃ a < x and b > x such that x ∈ (a, b) ⊂ A
=⇒ ∃ ρ, σ ∈ Q with a < ρ < x, x < σ < b so that x ∈ (ρ, σ) ⊂ A
[
=⇒ A = (ρ, σ)
ρ,σ∈Q
ρ<σ
(ρ,σ)⊂A

=⇒ A ∈ M(E)
S
since ρ,σ∈Q
ρ<σ
is a countable union.
(ρ,σ)⊂A

(b), (c) The proofs of all three parts of (b) and (c) are very similar. The proof of
one of the three parts is problem #8 of Problem Set 1.
(d), (e) The proofs of all four parts
 (d) and (e) are very similar.
 Here is the proof of
′
the first part of (d). Let O = A ⊂ R A open and E = (a, ∞) a ∈ R .
• M(E ′) ⊂ BR :
E ′ ⊂ O ⊂ M(O) = BR =⇒ M(E ′ ) ⊂ BR

• BR ⊂ M(E ′ ): Let E = (a, b) a < b . We already know, by part (a), that
M(E) = BR , so it suffices to prove that E ⊂ M(E ′). Let α < β. Then
\ ∞ 
c 1

c
(α, β) = (−∞, β) ∩ (α, ∞) = [β, ∞) ∩ (α, ∞) = β − n, ∞ ∩ (α, ∞)
n=1

So (α, β) ∈ M(E ). Hence E ⊂ M(E ), so that, part (a), BR = M(E) ⊂ M(E ′).
′ ′

Measures 7 October 10, 2019

2.4 Measures
Let X be a set.

Definition 2.10. A measure on the σ-algebra M ⊂ P(X) is a function

µ : M → [0, ∞]

such that
(i) µ(∅) = 0
(ii) If {Ej }j∈N ⊂ M is a countable collection of disjoint subsets of X, then

[  ∞
X
µ Ej = µ(Ej )
j∈N j=1

(ii) is called countable additivity.

Definition 2.11. A finitely additive measure on the algebra A ⊂ P(X) is a

function
µ : A → [0, ∞]
such that
(i) µ(∅) = 0
(ii’) If {E1 , · · · , En } is a finite collection of disjoint sets in A, then
n
[  n
X
µ Ej = µ(Ej )
j=1 j=1

(ii’) is called finite additivity.

Definition 2.12. A measure µ on the σ-algebra M ⊂ P(X) is called

(i) finite if µ(X) < ∞
(ii) σ
S-finite if there is a countable collection {En }n∈N ⊂ M of subsets with X =
∞
n=1 En and with µ(En ) < ∞ for all n ∈ N.
(iii) semifinite if for each E ∈ M with µ(E) = ∞, there is an F ∈ M with
0 < µ(F ) < ∞ and F ⊂ E.
(iv) Borel if X is a metric space (or, more generally a topological space) and M
is BX , the σ-algebra of Borel subsets of X.

Measures 8 October 10, 2019

Remark 2.13. (a) If µ is a finite measure on the σ-algebra M ⊂ P(X) and if
E ∈ M, then
≥0
z }| {
µ(X) = µ(E ∪ E c ) = µ(E) + µ(E c ) =⇒ µ(E) ≤ µ(X) < ∞

(b) Here is some terminology and notation.

(X, M) measurable space

(X, M, µ) measure space

Example 2.14 (counting measure). If X is any set and M = P(X), then

µ(E) = # points in E

is called the counting measure on X.

Example 2.15 (point mass). If X is any set, x0 ∈ X, and M is any σ-algebra for
X, then (
1 if x0 ∈ E
µ(E) =
0 if x0 ∈/E
is called a point mass or Dirac measure at x0 .

Example 2.16. Let X be any set, M be any σ-algebra for X, and f : X → [0, ∞].
Then X
µ(E) = f (x) for all E ∈ M
x∈E
P
is a measure, if x∈E is interpreted appropriately. See the definition below.
P
Definition 2.17 (the meaning of f (x)). Let E be any set and f : E → [0, ∞].
x∈E
X
• If f (x) = ∞ for at least one x ∈ E, then f (x) = ∞
x∈E
For the rest of this definition, assume that f : E → [0, ∞). Set

P = x ∈ E f (x) > 0
X
• If P is a finite set, then f (x) is, of course, the sum of the finite number of
x∈E 
(strictly positive) numbers in f (x) x ∈ P .

Measures 9 October 10, 2019

 • If P is a countable set, and P = {x1 , x2 , x3 , · · · } is any ordering for P , then

X N
X
f (x) = lim f (xn )
N →∞
x∈E n=1

The right hand side is independent of the choice of ordering. Note that the fact
that f (xn ) ≥ 0, for all n, is critical for this independence of ordering. See problem
#1 on Problem Set 2.
P
• If P is an uncountable set, then x∈E f (x) = ∞. This is reasonable because
there must existSan N ∈ N with PN = x ∈ E f (x) ≥ N1 infinite (because
P
otherwise P = N ∈N PN is countable) and any reasonable definition of f (x)
x∈E
should obey
P X X 1
f (x) ≥ f (x) ≥ =∞
x∈E x∈PN x∈P
N
N

Example 2.18. Let X be any uncountable set and set


M = E ∈ P(X) E is countable or co-countable

See Example 2.5. Then

( )
0 if E is countable
µ(E) = for all E ∈ M
1 if E is co-countable

is a measure. This is because

S
• If E1 , E2 , E3 , · · · are all countable, then j∈N Ej is also countable so that
=0
X z }| { S 
µ(Ej ) = 0 and µ j∈N j = 0
E
j∈N

• If
S E1 , E2 , E3 , · · · are all countable, except for Ej0 , which is co-countable, then
j∈N Ej is co-countable so that

=1 =0
X z }| { X z }| { S 
µ(Ej ) = µ(Ej0 ) + µ(Ej ) = 1 and µ j∈N j = 1
E
j∈N j∈N
j6=j0

• It is impossible to have two disjoint co-countable sets E1 , E2 because E2 ⊂ E1c .

Measures 10 October 10, 2019

Example 2.19. Let X be any infinite set and set M = P(X) and
( )
0 if E is finite
µ(E) = for all E ∈ M
∞ if E is infinite
This is a finitely additive measure, but not a measure.
Theorem 2.20. Let (X, M, µ) be a measure space and E, F , E1 , E2 , · · · ∈ M.
(a) (Monotonicity) If E ⊂ F , then µ(E) ≤ µ(F ).
S  P
∞
(b) (Subadditivity) µ E
n=1 n ≤ ∞ n=1 µ(En ) (Note that the En ’s need
not be disjoint.)
(c) (Continuity from below) If E1 ⊂ E2 ⊂ E3 · · · , then
S 
∞
µ n=1 En = lim µ(En )
n→∞

(d) (Continuity from above) If µ(E1 ) < ∞ and E1 ⊃ E2 ⊃ E3 · · · , then

T 
∞
µ n=1 En = lim µ(En )
n→∞
≥0

z }| {
Proof. (a) µ(F ) = µ E ∪ (F \ E) = µ(E) + µ(F \ E) ≥ µ(E)
k−1


(b) Set F1 = E1 , F2 = E2 \ E1 , F3 = E3 \ (E1 ∪ E2 ), · · · , Fk = Ek \ ∪n=1 En , · · · .
Then the Fk ’s are all disjoint and, for all N ∈ N, we have ∪N N
n=1 Fn = ∪n=1 En , so that
∞
[ [∞ S  S  X ∞ X∞
∞ ∞
Fn = En =⇒ µ n=1 En = µ n=1 Fn = µ(Fn ) ≤ µ(En )
n=1 n=1 n=1 n=1

(c) Writing
∞
[ S 
∞
En = E1 ∪ j=2 Ej \ Ej−1
n=1
we have
S  ∞
X
∞

µ n=1 En = µ(E1 ) + µ Ej \ Ej−1
j=2
Sn

µ E
j=1 j

z n
X
}| {


= lim µ(E1 ) + µ Ej \ Ej−1
n→∞
j=2

= lim µ(En )
n→∞

Measures 11 October 10, 2019

(d) Set, for n ∈ N, Fn = E1 \ En . Then F1 ⊂ F2 ⊂ F3 ⊂ · · · so that, by part (c),
S 
µ ∞ n=1 n = lim µ(Fn ) = lim µ(E1 \ En )
F
n→∞ n→∞

Subtracting µ(En ) from both sides of µ(E1 ) = µ(En ) + µ(E1 \ En ) (which we are
allowed to do only because we know that µ(En ) is finite), we have µ(E1 \ En ) =
µ(E1 ) − µ(En ) and hence
S   
µ ∞ n=1 n = lim µ(E1 ) − µ(En ) = µ(E1 ) − lim µ(En )
F (∗)
n→∞ n→∞

As Fn = E1 ∩ Enc ,
we have
∞
[ S  T c T 
∞ c ∞ ∞
Fn = E1 ∩ E
n=1 n = E1 ∩ E
n=1 n = E1 \ E
n=1 n
n=1

So the left hand side of (∗) is

S   T  T 
∞ ∞ ∞
µ n=1 Fn = µ E1 \ n=1 En = µ(E1 ) − µ n=1 En
and the claim follows.
Example 2.21. The hypothesis that µ(E1 ) < ∞ in part (d) of Theorem 2.20 (con-
tinuity from above) really is needed. For example if
X=N
M = P(X)
µ = counting measure

En = j ∈ N j ≥ n for each n ∈ N
then
µ(En ) = ∞ for all n ∈ N so that lim µ(En ) = ∞
n→∞

but ∞
\ T 
∞
En = ∅ so that µ n=1 En =0
n=1
Definition 2.22. Let (X, M, µ) be a measure space.
(a) A null set is a set E ∈ M with µ(E) = 0.
(b) If f : X → {true, false} is a statement about points of X and


µ x ∈ X f (x) = false =0
then we say that f is true almost everywhere (with respect to µ). It is
standard to abbreviate “almost everywhere” by “a.e.”.

Measures 12 October 10, 2019

Remark 2.23. The following “silliness” is completely consistent with all of our
definitions so far.
Let (X, M, µ) be a measure space. Let N ∈ M obey µ(N) = 0 (i.e. N is a null
set). As N has “size” zero, we would certainly expect that any smaller subset Z ⊂ N
would also have “size” zero. But we have not required that

N ∈ M, µ(N) = 0, Z ⊂ N =⇒ Z ∈ M

So it is possible that µ(Z) is not defined, even though it ought to be zero. That’s
silly.
Here is an extreme example of this.

X = any nonempty set M = {∅, X} µ(∅) = µ(X) = 0

The entire world has size zero, and yet not a single nontrivial subset has a defined
size.

Definition 2.24. A measure space (X, M, µ) is said to be complete if

Z ⊂ N ∈ M, µ(N) = 0 =⇒ Z ∈ M

The following theorem says that the silliness of Remark 2.23 can always be fixed —
any measure can be completed.

Theorem 2.25 (Completion). Let (X, M, µ) be a measure space. Set


N = N ∈ M µ(N) = 0

M = E ∪ Z E ∈ M, Z ⊂ N for some N ∈ N
µ̄ : M → [0, ∞] with
µ̄(E ∪ Z) = µ(E) for all E ∈ M and Z ⊂ N for some N ∈ N

Then
(a) M is a σ-algebra.
(b) µ̄ is a well-defined, complete measure on M, called the completion of µ.
(c) µ̄ is the unique extension of µ to M. That is, µ̄ is the unique measure with
domain of definition M that obeys µ̄(E) = µ(E) for all E ∈ M.

Measures 13 October 10, 2019

Proof.
• M is a σ-algebra:
◦ Both M and N are closed under countable unions. So M is closed under
countable unions.
◦ Let E ∈ M and Z ⊂ N ∈ N . We must show that (E ∪ Z)c ∈ M. That is,
we must express (E ∪ Z)c as the union of a set in M and a subset of a null
set. Define
N′ = N \ E ∈ M Z′ = Z \ E ⊂ N ′
In the figure below, Z is outlined in red and Z ′ is shaded.

X
N′
E
N
Z
Z′

Then X is the disjoint union of E, Z ′ , N ′ \ Z ′ and (E ∪ N ′ )c and µ(N ′ ) = 0

so that
∈M ⊂N ′
z }| { z }| {
(E ∪ Z)c = (E ∪ Z ′ )c = (E ∪ N ′ )c ∪ (N ′ \ Z) ∈ M

• µ̄ is well-defined :
Let

E, E ′ ∈ M, Z ⊂ N ∈ N, Z′ ⊂ N ′ ∈ N with E ∪ Z = E ′ ∪ Z ′

We need to prove that µ̄(E ∪ Z) = µ̄(E ′ ∪ Z ′ ), i.e. that µ(E) = µ(E ′ ). As

E ∪ Z = E ′ ∪ Z ′,
⊂Z ′ ⊂N ′
z }| {
µ(E) = µ(E ∩ E ′ ) + µ(E \ E ′ ) = µ(E ∩ E ′ )

µ(E ′ ) = µ(E ′ ∩ E) + µ(E ′ \ E) = µ(E ∩ E ′ )

| {z }
⊂Z⊂N

Measures 14 October 10, 2019

 • µ̄ is a complete measure:
This is problem #8 on Problem Set 2.
• Unique extension:
Let µ′ be any extension of µ to M. We have to show that µ′ = µ̄.
◦ If E ∈ M, then µ′ (E) = µ(E) by hypothesis.
◦ If Z ⊂ N ∈ N , then µ′ (Z) ≤ µ′ (N) = µ(N) = 0.
◦ In general, if Ē = E ∪ Z ∈ M, with E ∈ M, Z ⊂ N ∈ N , then
=µ(E) =0
z }| { z }| {
µ(E) = µ′ (E) ≤ µ′ (Ē) ≤ µ′ (E ∪ N) ≤ µ′ (E) + µ′(N) = µ(E)
=⇒ µ′ (Ē) = µ(E) = µ̄(Ē)

Measures 15 October 10, 2019

2.5 Outer Measures
We now have a reasonable abstraction of “volume” — namely measure. But, so
far, we have only pretty trivial and artificial examples. We now start building the
procedure that we’ll use to define the “Lebesgue measure”, which is our usual notion
of volume in Rd . The strategy that we’ll use to define the volume of a set A ⊂ Rd is
• approximate the volume of A from above by covering A by a union of (at most
countably many) rectangles. Certainly the volume of A is at most the sum of the
volumes of the rectangles.
• Make the rectangles finer and finer. Define the

“outer volume of A” = inf vol(S) S simple , A ⊂ S
• approximate the volumes of A from below by the volumes of simple sets contained
in A. Define the

“inner volume of A” = sup vol(S) S simple , S ⊂ A
• If the “inner volume” equals the “outer volume”, we have the volume of A.
We now abstract these ideas.
Definition 2.26. An outer measure on X is a function
µ∗ : P(X) → [0, ∞]
such that
(i) µ∗ (∅) = 0
(ii) If A ⊂ B, then µ∗ (A) ≤ µ∗ (B).
(iii) If {Aj }j∈N is a countable collection of (not necessarily disjoint) subsets of X,
then ∞
S  X
µ∗ ∞ A
j=1 j ≤ µ∗ (Aj )
j=1

This is called countable subadditivity.

Proposition 2.27. Let X be a set, S ⊂ P(X) and ρ : S → [0, ∞] be such that
∅∈S X∈S ρ(∅) = 0
Define, for each A ⊂ X,
nP  S∞ o
∗ ∞
µ (A) = inf j=1 ρ(Sj ) Sj j∈N
⊂ S, A ⊂ j=1 Sj

Then µ∗ is an outer measure.

Measures 16 October 10, 2019

Example 2.28 (Lebesgue outer measure on R).

X=R
 
S = {∅, R} ∪ (a, b) a, b ∈ R, a < b ∪ (a, ∞), (−∞, a) a ∈ R





ρ(∅) = 0 ρ (a, b) = b − a ρ (a, ∞) = ρ (−∞, a) = ρ R = ∞

Proof of Proposition 2.27.

• That µ∗ (∅) = 0 is obvious since ∅ ∈ S and ρ(∅) = 0.
• Let A ⊂ B. We need to show that µ∗ (A) ≤ µ∗ (B).
 S S∞
If Sj j∈N ⊂ S with B ⊂ ∞ j=1 Sj . Then we also have A ⊂ j=1 Sj . Consequently,
nP  S∞ o
∞
j=1 ρ(Sj ) Sj j∈N
⊂ S, B ⊂ S
j=1 j
nP  S∞ o
∞
⊂ j=1 ρ(Sj ) Sj j∈N ⊂ S, A ⊂ j=1 Sj

so that
nP  S∞ o
∗ ∞
µ (B) = inf j=1 ρ(Sj ) Sj j∈N ⊂ S, B ⊂ j=1Sj
nP  S∞ o
∞
≥ inf j=1 ρ(S j ) S j j∈N ⊂ S, A ⊂ S
j=1 j = µ∗ (A)

• Countable subadditivity
◦ Let ε > 0. Let, for each n ∈ N, A(n) ⊂ X with µ∗ (A(n) ) < ∞. (The other case
is trivial.)
 (n) S (n)
◦ For each n ∈ N, there exists Sj j∈N ⊂ S such that A(n) ⊂ ∞ j=1 Sj and
P∞ (n) ∗
j=1 ρ(Sj ) ≤ µ (A
(n)
) + 2εn .
 (n) S S (n)
◦ Then Sj j,n∈N ⊂ S and n∈N A(n) ⊂ j,n∈N Sj so that

S  ∞ 
∞
X (n)

X ε
µ∗ n=1 A
(n)
≤ ρ Sj ≤ µ∗ (A(n) ) + n
j,n∈N n=1
2
∞
X
=ε+ µ∗ (A(n) )
n=1

As this is true for all ε > 0, we have

S  ∞
X
∗ ∞ (n)
µ n=1 A ≤ µ∗ (A(n) )
n=1

Measures 17 October 10, 2019

 Next we’ll have some motivation for a trick which will allow us to avoid repeating
the previous arguments for “inner measure”. Let A ⊂ Rn . Choose a definition of a
“simple set” (for example a countable union of rectangles) covering A. Fix a simple
set E with A ⊂ E ⊂ Rn . For this motivation, we’ll restrict to bounded A’s and E’s.
Then define

S to be a simple set with S ⊂ A ⇐⇒ T = E \ S is a simple set covering E \ A

E
T

A S

Then we’ll define


inner measure(A) = sup vol(S) S simple set with S ⊂ A

= sup vol(E \ T ) T simple set with E \ A ⊂ T

= sup vol(E) − vol(T ) T simple set with E \ A ⊂ T

= vol(E) − inf vol(T ) T simple set with E \ A ⊂ T
= µ∗ (E) − µ∗ (E \ A)

and A will have a well–defined volume if

inner measure(A) = outer measure(A)

i.e. µ∗ (E) − µ∗ (E \ A) = µ∗ (A)
i.e. µ∗ (A) + µ∗ (E \ A) = µ∗ (E)

We’ll use something like this as the definition of “A is measurable”. Technically, it

is convenient to not restrict to a single7 E or even to E’s that are nice sets or that
contain all of A. Here is the standard definition.
7
For example, if we were restrict to a single E, there would be the danger that whether or not
“inner measure(A) = outer measure(A)” is true could depend on the choice of E.

Measures 18 October 10, 2019

Definition 2.29. Let µ∗ be an outer measure on a set X. A subset A ⊂ X is said
to be µ∗ -measurable if

µ∗ (E) = µ∗ (E ∩ A) + µ∗ (E ∩ Ac ) for all E ⊂ X

Remark 2.30.
(a) That µ∗ (E) ≤ µ∗ (E ∩ A) + µ∗(E ∩ Ac ) is always true, by axiom (iii) of Definition
2.26.
(b) That µ∗ (E) ≥ µ∗ (E ∩ A) + µ∗ (E ∩ Ac ) is trivial for µ∗ (E) = ∞. So, to verify
that A is measurable, it suffices to verify µ∗ (E) ≥ µ∗ (E ∩ A) + µ∗ (E ∩ Ac ) for
all E with µ∗ (E) < ∞.
(c) We will show, in Problem Set 4 #2, that if µ∗ is induced from a premeasure
(to be defined - most of our measures will be) and if µ∗ (X) is finite, then A is
µ∗ -measurable if and only if µ∗ (A) + µ∗ (X \ A) = µ∗ (X).

THEOREM 2.31 (Carathéodory8 ). Let µ∗ be an outer measure on the set X and


M∗ = A ⊂ X µ∗ (E) = µ∗ (E ∩ A) + µ∗ (E ∩ Ac ) ∀ E ⊂ X

be the set of µ∗ -measurable subsets of X. Then

(a) M∗ is a σ-algebra.
(b) The restriction, µ∗↾ M∗ , of µ∗ to M∗ is a complete measure.
(c) If N ⊂ X obeys µ∗ (N) = 0, then N ∈ M∗ .

Proof.
• Step 1 : M∗ is nonempty
since ∅ ∈ M∗ since µ∗ (E) = µ∗ (E ∩ ∅) + µ∗(E ∩ ∅c ) for all E ⊂ X.
| {z } | {z }
=0 =µ∗ (E)
• Step 2 : M∗ is closed under complements.
This is obvious since

µ∗ (E) = µ∗ (E ∩ A) + µ∗ (E ∩ Ac ) ⇐⇒ µ∗ (E) = µ∗ (E ∩ Ac ) + µ∗ (E ∩ {Ac }c )

| {z }
A

• Step 3 : M∗ is closed under finite unions.

It suffices to prove that A, B ∈ M∗ =⇒ A ∪ B ∈ M∗ .
8
Constantin Carathéodory (1873–1950) was the son of a Greek diplomat who spent most of his
career in Germany.

Measures 19 October 10, 2019

 To do so, it suffices to prove that

A, B ∈ M∗ , E ⊂ X =⇒ µ∗ (E) ≥ µ∗ (E ∩ (A ∪ B)) + µ∗ (E ∩ Ac ∩ B c )

But, if A, B ∈ M∗ and E ⊂ X, then

A∈M∗
µ∗ (E) = µ∗ (E ∩ A) + µ∗ (E ∩ Ac )
using E ∩ A as E using E ∩ Ac as E
z
B∈M∗
}| { z }| {
= µ∗ (E ∩ A ∩ B) + µ∗ (E ∩ A ∩ B c ) + µ∗ (E ∩ Ac ∩ B) + µ∗ (E ∩ Ac ∩ B c )
≥ µ∗ (E ∩ (A ∪ B)) + µ∗ (E ∩ (A ∪ B)c )

since A ∪ B = (A ∩ B) ∪ (A ∩ B c ) ∪ (Ac ∩ B) and (A ∪ B)c = Ac ∩ B c .

• Step 4 : µ∗ is finitely additive on M∗ .
Let A, B ∈ M∗ with A ∩ B = ∅. We have to prove that µ∗ (A ∪ B) =
µ∗ (A) + µ∗ (B). For later use, we prove more generally, that, for any Ẽ ⊂ X,
=A =B

A∈M∗ z }| {
z }| {

µ Ẽ ∩ (A ∪ B) = µ∗ Ẽ ∩ (A ∪ B) ∩ A + µ∗ Ẽ ∩ (A ∪ B) ∩ Ac
∗

(using Ẽ ∩ (A ∪ B) as E)
= µ∗ (Ẽ ∩ A) + µ∗ (Ẽ ∩ B) (∗)

Just choosing Ẽ = X gives µ∗ (A ∪ B) = µ∗ (A) + µ∗ (B).

• Step 5 : M∗ is a σ-algebra.
∗
By Problem Set 1, #2,  it suffices to ∗prove that M is closed under countable
disjoint unions. Let An n∈N ⊂ M with An ∩ Am = ∅ for all m 6= n. Let
∗
Sn⊂ X. As M
E
∗
is closed under finite unions, we have that, for each n ∈ N,
j=1 Aj ∈ M so that
 S   S c 
n n
µ∗ (E) = µ∗ E ∩ A
j=1 j + µ ∗
E ∩ A
j=1 j
n
X  S c 
n
= µ∗ (E ∩ Aj ) + µ∗ E ∩ j=1 Aj
j=1

(by (∗) in step 4)

n
X  S c 
∗ ∗ ∞
≥ µ (E ∩ Aj ) + µ E ∩ j=1 Aj
j=1

Measures 20 October 10, 2019

 Taking the limit n → ∞,
∞
X  S c 
∗ ∞
µ (E) ≥ µ∗ (E ∩ Aj ) + µ∗ E ∩ j=1 Aj
j=1
 S   S c 
∗ ∞ ∗ ∞
≥µ E∩ j=1 Aj +µ E∩ j=1 Aj

(by subadditivity)
∗
≥ µ (E)

So
S∞both of the ≥’s in the last displayed equation must be equalities and
∗
j=1 Aj ∈ M .
• Step 6 : µ∗↾ M∗ is countably additive.
Just take E = X in Problem Set 3, #2.
• Step 7 : µ∗↾ M∗ is complete.
Let N ∈ M∗ with µ∗ (N) = 0 and let Z ⊂ N. We must show that Z ∈ M∗ ,
i.e. that µ∗ (E) = µ∗ (E ∩ Z) + µ∗ (E ∩ Z c ) for all E ⊂ X. So let E ⊂ X. Then
⊂N ⊂E
z }| {
z }| {

µ∗ (E) ≤ µ∗ E ∩ Z + µ∗ E ∩ Z c
(∗∗)
≤ µ∗ (N) + µ∗ (E)
= µ∗ (E)

which forces the first ≤ above to be =.

• Step 8 : µ∗ (N) = 0 =⇒ N ∈ M∗ .
Note that, in Step 7, we did not use that N ∈ M∗ in (∗∗). We only used that
µ∗ (N) = 0. So taking Z = N in (∗∗) shows that if µ∗ (N) = 0, then N ∈ M∗ .

Measures 21 October 10, 2019

2.6 Premeasures and Lebesgue Measure
Typically, when you wish to construct a measure, you have
• a collection of really nice sets and
• you know what measure (i.e. volume) you want to assign to each of these really
nice sets and
you want to extend this to a full measure. Here are two examples. After the examples,
we’ll formalize this framework.
Example 2.32.
(a) (will lead to “Lebesgue measure on R”)

{really nice sets} = A

n
n[ o
= {∅} ∪ (aj , bj ] − ∞ ≤ a1 < b1 < a2 < · · ·< bn ≤ ∞, n ∈ N
j=1

µ0 (∅) = 0
S  X n
n
µ0 j=1 (aj , bj ] = (bj − aj ) for all − ∞ ≤ a1 < b1 < a2 < · · · < bn ≤ ∞
j=1

with (a, ∞] defined to be (a, ∞)

(b) (will lead to “Lebesgue-Stieltjes measures on R”) Let F : R → R be a function
with some properties that we will specify at the end of this example.

{really nice sets} = A

n
n[ o
= {∅} ∪ (aj , bj ] − ∞ ≤ a1 < b1 < a2 < · · ·< bn ≤ ∞, n ∈ N
j=1

µ0 (∅) = 0
S  X n


µ0 nj=1 (aj , bj ] = F (bj ) − F (aj ) for all − ∞ ≤ a1 < b1 < a2 < · · ·< bn ≤ ∞
j=1

Let’s think about what conditions we need to impose on the function F . Suppose
that we succeed in extending µ0 to a measure µ. I.e. suppose that we find a
σ-algebra M and a measure µ with
(i) A ⊂ M
(ii) µ(A) = µ0 (A) for all A ∈ A.
Then

Measures 22 October 10, 2019

 • M must contain all Borel sets (since the σ-algebra generated by the (a, b]’s
is BR )



• F (b) − F (a) = µ0 (a, b] = µ a, b] ≥ 0 for all b > a. That is, F must be
nondecreasing.
• By continuity from above

 

lim F a + n1 − F (a) = lim µ0 a, a + n1
n→∞ n→∞


= lim µ a, a + n1
n→∞
T 
=µ ∞ n=1 a, a + 1
n

= µ(∅) = 0

That is, F must be continuous from the right.

• Recall that, in µ0 (a, b] = F (b)−F (a), a = −∞ and b = +∞ are allowed.
So we need to define F (−∞) and F (+∞). By countable additivity,
∞


X

F (∞) − F (0) = µ0 (0, ∞] = µ (0, ∞] = µ (j − 1, j]
j=1
∞
X ∞
X



= µ0 (j − 1, j] = F (j) − F (j − 1)
j=1 j=1
n
X

= lim F (j) − F (j − 1)
n→∞
j=1


= lim F (n) − F (0)
n→∞

and
∞
X




F (0) − F (−∞) = µ0 (−∞, 0] = µ − ∞, 0] = µ (−j − 1, −j]
j=0
∞
X ∞
X



= µ0 (−j − 1, −j] = F (−j) − F (−j − 1)
j=0 j=0
n−1
X

= lim F (−j) − F (−j − 1)
n→∞
j=0


= lim F (0) − F (−n)
n→∞

Measures 23 October 10, 2019

 So we define
F (∞) = lim F (x) F (−∞) = lim F (x)
x→∞ x→−∞

Note that, as F is nondecreasing, both of these limits exist, though the

first could be +∞ and the second could be −∞.
We can now fill in the needed properties of F . We should have started this
example (b) like this:
(b) (will lead to “Lebesgue-Stieltjes measures on R”) Let F : R → R be a nonde-
creasing and right continuous function, and define
F (∞) = lim F (x) (possibly +∞)
x→∞
F (−∞) = lim F (x) (possibly −∞)
x→−∞

Here is a definition which formalizes the above setting.

Definition 2.33. Let X be a set. A premeasure on the algebra A ⊂ P(X) is a
function µ0 : A → [0, ∞] such that
(i) µ0 (∅) = 0
(ii) If {Aj }S
j∈N is a countable collection of disjoint subsets of X with {Aj }j∈N ⊂ A
and if ∞ j=1 Aj ∈ A, then
S  ∞
X
∞
µ0 j=1 Aj = µ0 (Aj )
j=1

(This includes finite additivity, since we can choose many Aj = ∅.)

Remark 2.34. Our goal here is to rig the definitions of algebra and premeasure so
that
• it is easy to construct algebras A and premeasures µ0 on A and yet
• it is possible to enlarge an algebra A to a σ-algebra M and to extend the
premeasure µ0 to a full measure µ on M.
S
Now a countable union ∞ j=1 Aj of disjoint subsets of X with {Aj }j∈N ⊂ A is not
required to be in A. But there is no law that says that it can’t be in A. And if
S ∞
j=1 Aj does happen to be in A and if µ0 can be extended to a measure µ on a
σ-algebra M ⊃ A, then we do have to have
S  S  X ∞ ∞
X
∞ ∞
µ0 j=1Aj = µ j=1 Aj = µ(Aj ) = µ0 (Aj )
j=1 j=1

That’s the reason for part (ii) of Definition 2.33.

Measures 24 October 10, 2019

Proposition 2.35. Let F : R → R be nondecreasing and right continuous. Define
F (±∞) = lim F (x). Set
x→±∞

n
n[ o
A = {∅} ∪ (aj , bj ] n ∈ N, −∞ ≤ a1 < b1 < a2 < · · · < bn ≤ ∞
j=1

µ0 (∅) = 0
S  X n
n  
µ0 j=1(aj , bj ] = F (bj )−F (aj )
j=1
for all n ∈ N, −∞ ≤ a1 < b1 < a2 < b2 < · · · < bn ≤ ∞

In the above, replace (a, b] by (a, b) when b = ∞. Then µ0 is a premeasure on A.

Proof.
• Step 1 : A is an algebra
◦ A is closed under complements
since ∅c = (−∞, ∞] and
S c
n
(a ,
j=1 j j b ] = (−∞, a1 ] ∪ (b1 , a2 ] ∪ · · · (bn−1 , an ] ∪ (bn , ∞]

◦ A is closed under finite unions.

To verify that any finite union of (α, β]’s can be expressed as a disjoint
(with gaps) finite union, repeatedly apply

if a < b, a′ < b′ and (a, b] ∩ (a′ , b′ ] 6= ∅ or b = a′ ar a = b′


then (a, b] ∪ (a′ , b′ ] = min{a, a′ } , max{b, b′ }

• Step 2 : µ0 (∅) = 0
by hypothesis.
• Step 3 : µ0 is countably additive where defined.
′
S∞
Let {Im }m∈N ⊂ A with Im ∩ Im′ = ∅ for P∞m 6
= m , and with I = m=1 Im being
in A. We have to show that µ0 (I) = m=1 µ0 (Am ).
The reason that this is not trivial
 is that {Im }m∈N could contain a countable
1 1
collection α + 2n+1 , α + 2n , n ≥ n0 , of shorter and shorter intervals that
“accumulate at α ∈ I”. Furthermore, there can be a countable number of
different such “accumulation” points in I.

Measures 25 October 10, 2019

 First we make some reductions.
It suffices to consider the case in which none of Im ’s is empty. Otherwise,
drop them.
It suffices
 to considerthe case in which each Im is a single interval. Otherwise
replace Im m∈N by J J is an interval in some Im .
It suffices to consider the case in which I is a single interval.
 Otherwise
consider separately each interval I¯ of I together with Im Im ⊂ I¯ .
(Recall that, by hypothesis, I ∈ A.)
The case in which there are only finitely many Im ’s is easy. So let I = S
(a, b] and
′ ∞
with Im ∩ Im′ = ∅ for m
Im = (am , bm ], m ∈ N, P
6
= m and with I = m=1 Im .
∞
We have to show that m=1 F (bm ) − F (am ) = F (b) − F (a).
P
◦ Step 3a: µ0 (I) ≥ ∞ m=1 µ0 (Im )
Let n ∈ N. By relabelling I1 , · · · , In , we may assume that

a ≤ a1 < b1 ≤ a2 < b2 ≤ a3 · · · ≤ an−1 < bn−1 ≤ an < bn ≤ b

So
≤F (b) ≤0 ≤0
n
X z }| { z }| {z }| {
µ0 (Im ) = F (bn ) −F (an ) + F (bn−1 ) −F (an−1 ) + F (bn−2 )
m=1 ≤0 ≤−F (a)
z }| { z }| {
· · · −F (a2 ) + F (b1 ) −F (a1 )
≤ F (b) − F (a) = µ0 (I)
P
This is true for all n, so that µ0 (I) ≥ ∞ m=1 µ0 (Im ).
P∞
◦ Step 3b: µ0 (I) ≤ m=1 µ0 (Im ) if a, b are finite
Here is our strategy. Let ε > 0.
· We want to use compactness to replace the infinite sum by a finite sum.
By compactness, any cover by open sets of a compact set has a finite
subcover.
· But I = (a, b] is not compact, so replace it by [a + δ, b] with δ > 0 chosen
so that F (a + δ) − F (a) < ε.


· And Im = (am , bm ] is not open. So replace it by am , bm + δm with
δm > 0 chosen so that F (b + δm ) − F (bm ) < 2εm .
P
· Using compactness, P we will extract a finite subsum from ∞ m=1 µ0 (Im )
that obeys µ0 (I) ≤ j µ0 (Ij ) + 2ε. That will do it.
Here we go.

Measures 26 October 10, 2019

 · Since F is right continuous, there exists a δ > 0 with F (a+δ)−F (a) < ε
and, for each m ∈ N, there exists a δm > 0 with F (b+ δm ) −F (bm ) < 2εm .
S S S
· Then m∈N (am , bm +δm ) covers m∈N Im = m∈N (am , bm ], which in turn
covers I = (a, b], which in turn covers [a + δ, b]. So there is a cover of
[a + δ, b] by a finite number of (am , bm + δm )’s. Pick one.
· Relabel the (am , bm + δm )’s (and possibly discard some unnecessary
(am , bm + δm )’s) as follows. Choose as (a1 , b1 + δ1 ) the (am , bm + δm )
that contains a + δ (and in particular has a1 < a + δ) with bm + δm as
large as possible. Then choose as (a2 , b2 + δ2 ) the (am , bm + δm ) that
contains b1 + δ1 (and in particular has a2 < a1 + δ1 ) with bm + δm as
large as possible. And so on.
a1 b1 + δ1 a3 b3 + δ3
b
a+δ a2 b2 + δ2 a4 b4 + δ4
So we may assume that
a1 < a + δ ≤ a2 < b1 + δ1 ≤ a3 < b2 + δ2 ··· < b < bn + δn
Then
µ0 (I) = F (b) − F (a) ≤ F (b) − F (a + δ) + ε ≤ F (bn + δn ) − F (a1 ) + ε
n−1
X  
= F (bn + δn ) − F (an ) + F (aj+1 ) − F (aj ) + ε
j=1
n−1
X  
≤ F (bn + δn ) − F (an ) + F (bj + δj ) − F (aj ) + ε
j=1
n
X  ε

≤ F (bj ) − F (aj ) + 2j
+ε
j=1
Xn
≤ µ0 (Ij ) + 2ε
j=1
P
◦ Step 3c: µ0 (I) ≤ ∞m=1 µ0 (Im ) if a = −∞ and/or b = ∞
Let M > 0. By the same argument as in Step 3b
∞


X
F min{M, b} − F max{−M, a} ≤ µ0 (Im )
m=1

Take the limit M → ∞.

Measures 27 October 10, 2019

THEOREM 2.36. Let X be a nonempty set and
A ⊂ P(X) be an algebra,
M = M(A) be the σ-algebra generated by A,
µ0 be a premeasure on A,
µ∗ be the outer measure induced by µ0 and
M∗ be the set of µ∗ -measurable sets.
Recall that
n P∞  S o
µ∗ (E) = inf µ0 (An ) An n∈N ⊂ A, E ⊂ ∞ A
n=1 n
n=1

Then
(a) µ∗ ↾ A = µ0 . That is, µ∗ extends µ0 . That is, µ∗ (A) = µ0 (A) for all A ∈ A.
(b) M ⊂ M∗ and µ ≡ µ∗ ↾ M is a measure that extends µ0 . That is, all sets in
M(A) are µ∗ -measurable and µ(A) = µ0 (A) for all A ∈ A.
(c) If ν is any other9 measure on M such that ν ↾ A = µ0 , then
ν(B) ≤ µ(B) for all B ∈ M
ν(B) = µ(B) if B ∈ M is µ-σ-finite. That is, if B is a countable union of
sets of finite µ-measure.

Proof. (a) Let A ∈ A.

◦ µ∗ (A) ≤ µ0 (A) by the definition of µ∗ . (Choose A1 = A and An = ∅ for all n > 1.)
 S n Sn−1 o
◦ If An n∈N ⊂ A with A ⊂ ∞ A
n=1 n , then set Bn = A ∩ An \ j=1 Aj . The Bn ’s
are disjoint sets in A with union A. So, as A ∈ A, and µ0 is a premeasure,

∞ ⊂A n ∞
X z}|{
X
µ0 (A) = µ0 Bn ≤ µ0 (An )
n=1 n=1
 S∞
This is true for any choice of An n∈N
⊂ A with A ⊂ n=1 An , so µ0 (A) ≤ µ∗ (A).
(b) By Carathéodory (Theorem 2.31), M∗ is a σ-algebra and the restriction of µ∗ to
M∗ is a measure. So it suffices to prove that every A ∈ A is µ∗ -measurable.
Let A ∈ A and E ∈ P(X). Let ε > 0. By the definition of µ∗ , there exist
9
See Problem Set 4, #4 for an example of multiple extensions.

Measures 28 October 10, 2019

 S∞
Bn n∈N
⊂ A with E ⊂ n=1 Bn and
∞
X ∞
X
∗
 
µ (E) ≥ µ0 (Bn ) − ε = µ0 (Bn ∩ A) + µ0 (Bn ∩ Ac ) − ε
n=1 n=1
(since µ0 is finitely additive on A)
∞
X X∞
= µ0 (Bn ∩ A) + µ0 (Bn ∩ Ac ) − ε
n=1 n=1
≥ µ∗ (E ∩ A) + µ∗ (E ∩ Ac ) − ε
(since E ∩ A ⊂ ∪n (Bn ∩ A) and E ∩ Ac ⊂ ∪n (Bn ∩ Ac ))
As this is true for all ε > 0, we have µ∗ (E) ≥ µ∗ (E ∩ A) + µ∗ (E ∩ Ac ). So A is
µ∗ -measurable by part (b) of Remark 2.30.
S 
(c) First part: Let B ∈ M and B ⊂ ∞ n=1 An with An n∈N ⊂ A. Then
∞
X ∞
X
ν(B) ≤ ν(An ) = µ0 (An )
n=1 n=1

Taking the inf over the choice of An n∈N
⊂ A gives
∗
ν(B) ≤ µ (B) = µ(B)
(c) Second part:
• Let B ∈ M. First consider the case that µ(B) < ∞. As a preliminary step,
we approximate B by a union of An ’s (sincewe know µ(An ) = ν(An )). Let
ε > 0. By the definition of µ∗ , we may choose An n∈N ⊂ A such that B ⊂ A =
S ∞
n=1 An ∈ M and
X∞ X∞
µ(A) ≤ µ(An ) = µ0 (An ) ≤ µ∗ (B) + ε = µ(B) + ε
n=1 n=1

As A = B ∪ (A ∩ B ), we have µ(A) = µ(B) + µ(A ∩ B c ) ≤ µ(B) + ε and hence

c

µ(A ∩ B c ) ≤ ε
Hence
∈A
S
z }| {  S  S 
n n n
µ(B) ≤ µ(A) = lim µ j=1 Aj = lim µ 0 j=1 Aj = lim ν j=1 Aj
n→∞ n→∞ n→∞
c c
= ν(A) = ν(B) + ν(A ∩ B ) ≤ ν(B) + µ(A ∩ B )
≤ ν(B) + ε
As this is true for all ε > 0, we have µ(B) = ν(B).

Measures 29 October 10, 2019

 S
• Finally consider the case that B = ∞ n=1 Bn with Bn ∈ M and µ(Bn ) < ∞ for
∈ N. We may
all n  assume that the Bn ’s are disjoint. Otherwise replace Bn by
Sn−1 
Bn \ j=1 Bj . So

∞
X ∞
X
µ(B) = µ(Bn ) = ν(Bn ) = ν(B)
n=1 n=1

Corollary 2.37. Let F, G : R → R be nondecreasing and right continuous.

(a) There is a unique Borel measure µF on R such that µF (a, b] = F (b) − F (a)
for all a, b ∈ R with a < b.
(b) µF = µG if and only if F − G is a constant function.
(c) If µ is a Borel measure on R that is finite on all bounded Borel sets, then µ = µF
for 

µ (0, x]
 if x > 0
F (x) = 0 if x = 0



−µ (x, 0] if x < 0


Proof. (a) We have already shown, in Proposition 2.35, that µF (a, b] = F (b)−F (a)
S σ-algebra BR is generated
gives a premeasure and, in Proposition 2.9, that the Borel
by the (a, b]’s. The premeasure is σ-finite since R = ∞ n=−∞ (n, n + 1], so Theorem
2.36 gives both the existence and the uniqueness of the corresponding Borel measure.
(b) We have
µF = µG ⇐⇒ the corresponding premeasures are the same



⇐⇒ µF (a, b] = µG (a, b] for all a < b
⇐⇒ F (b) − F (a) = G(b) − G(a) for all a < b
⇐⇒ F (a) − G(a) = F (b) − G(b) for all a < b

(c) For the specified F ,




µ (0, b]
− µ (0, a]

 if 0 ≤ a < b
F (b) − F (a) = µ (0, b] + µ (a, 0] if a < 0 ≤ b




−µ (b, 0] + µ (a, 0] if a < b < 0


= µ (a, b] for all a < b

Measures 30 October 10, 2019

We have already checked, in Example 2.32(b), that F is nondecreasing and right
continuous. So µF and µ give the same premeasure, and so are the same.

Remark 2.38. Here is some terminology and notation.

• A Borel measure is a measure on the σ-algebra of Borel sets (i.e. the σ-algebra
generated by the open sets).
• We have just seen that every Borel measure on R that is finite on bounded sets is
µF for some nondecreasing, right continuous F .
• For each nondecreasing, right continuous F : R → R, the complete measure
◦ determined (via Carathéodory) by the outer measure, which is, in turn,


◦ determined by the premeasure with µ0 (a, b] = F (b)−F (a), −∞ ≤ a < b ≤ ∞,
is usually also denoted µF and is called the Lebesgue-Stieltjes measure asso-
ciated to F .
• The (complete) Lebesgue-Stieltjes measure associated to the function F (x) = x is
called the Lebesgue measure.
◦ The Lebesgue measure of the set A is denoted m(A).
◦ The domain of the Lebesgue measure is denoted L. It is a σ-algebra that contains
BR .
◦ The Lebesguemeasure is translation invariant. That is, if E ∈ L and t ∈ R,
then E + t = x + t x ∈ E ∈ L and m(E + t) = m(E). That’s Problem Set
6, #1.

Example 2.39. As a simple example, consider

(
0 if x < 0
F (x) =
x if x ≥ 0

Then µF (S) = 0 for all S ⊂ (−∞, 0). In particular every S ⊂ (−∞, 0) is null and
hence measurable. So there are lots of sets that are measurable, but not Borel. On
the other hand the set N ⊂ [0, 1) of Problem set 3, #4 is not measurable. So not all
subsets are measurable.

Measures 31 October 10, 2019

2.7 The Cantor Set Example
Each x ∈ [0, 1] can be represented by the ternary expansion (as opposed to the
∞
P
decimal expansion) x = xn 3−n with each xn ∈ {0, 1, 2}. Such representations are
n=0
not unique. For example
∞
X 2 1
0.2222 · · · = 2 × 3−n = 1 = 1 = 1.0000 · · ·
n=1
3 1− 3

When faced with such a choice, in this example, we always avoid terminal 1’s (so we
express 31 as 0.0222 · · · rather than 0.1000 · · · ) and keep terminal 2’s (so we express
2
3
as 0.2000 · · · rather than 0.1222 · · · ). Given x ∈ [0, 1] we can generate the xn ’s
inductively by the algorithm
Set n = 0, xn = 0.
n
P
Set εn = x − xp 3−p .
p=0

1
Observe that 0 ≤ εn ≤ 3n
.

1
0 if 0 ≤ εn ≤ 3n+1

1 2
Set xn+1 = 1 if 3n+1 < εn < 3n+1

 2 3 1
2 if 3n+1 ≤ εn ≤ 3n+1 = 3n

n = 0 → n + 1.

The Cantor set is


C= x ∈ [0, 1] xn 6= 1 for all n ∈ N
For example
rather than

z 0.1000···
}| {
1/3 = 0.0222 · · · ∈ C

not 000 · · ·

z or 2222
}| · · · {
x ∈ (1/3, 2/3) = 0.1 x2 x3 x4 · · · ∈ /C
2/3 = 0.2000 · · · ∈ C

Measures 32 October 10, 2019

So
x =1
1
z }|
{


C = [0, 1] \ 13 , 23 \ 1 2
,
9 9
∪ 7 8
,
9 9
···
| {z } | {z }
x1 =0 x1 =2
x2 =1 x2 =1

We have that C is
· compact (C is [0, 1] with an open set removed)
· nowhere dense, i.e. the interior of its closure is empty, i.e. if x ∈ C̄ = C, ε > 0,
then (x − ε, x + ε) 6⊂ C
· totally disconnected, i.e. the only connected subsets are single points (if α < β
were both in a connected subset, then we would have (α, β) ⊂ C, which doesn’t
happen)
· uncountable (by the same argument as for R)
· of Lebesgue measure zero because





m(C) = m [0, 1] − m (1/3, 2/3) − m (1/9, 2/9) + m (7/9, 8/9) − · · ·
1 2 4
=1− − − −···
3 9 27
∞
X 2n 1 1
=1− = 1−
n=0
3 n+1 3 1 − 23
=0

Remark 2.40.

(a) We can generalize this. Let αj j∈N
⊂ (0, 1). Then
(1) Start with [0, 1]. (measure 1)
(2) Remove the middle α1th . (measure 1 − α1 )
(3) Remove the middle α2th of the remaining pieces. (measure (1 − α1 )(1 − α2 ))
(4) And so on.
This gives a set of measure
∞ n
( )
Y Y n
P
(1 − αj ) = lim (1 − αj ) = lim exp ln(1 − αj )
n→∞ n→∞ j=1
j=1 j=1

We can get any measure in (0, 1) in this way.

Measures 33 October 10, 2019

(b) The Cantor function is defined by

∈{0,1}

 P z}|{
 ∞
X∞  n=0 x2n 2−n
 if x ∈ C
f :x= xn 3−n 7→ PN
xn −n 1
n=0


 n=0 2
2 + 2N+1 if x ∈
/ C, xN +1 = 1


xn ∈ {0, 2} for all n ≤ N
This function is
· increasing,
· continuous (it takes all values in [0, 1] and so cannot have jump discontinuities),
· constant on each subinterval we removed while building C.
It can be used to construct a subset of R that is Lebesgue measurable but not
Borel. See Exercise 9 of Chapter 2 of Folland.

2.8 Regularity Properties of Lebesgue-Stieltjes Measures

Fix a (complete) Lebesgue-Stieltjes measure µF and call it µ. Call its domain Mµ .
By definition, for any E ∈ Mµ ,
n P∞   S o
µ(E) = µ∗ (E) = inf F (bn ) − F (an ) E ⊂ ∞ n=1 (an , bn ]
n=1
∞
n P
S∞ o
= inf µ (an , bn ] E ⊂ n=1 (an , bn ]
n=1

Theorem 2.41 (Regularity). For all E ∈ Mµ ,

n P∞
S
o
µ(E) = inf µ (an , bn ) E ⊂ ∞n=1 a ,
n nb (a)
 n=1
= inf µ(O) O ⊂ R, O open, E ⊂ O (b)

= sup µ(K) K ⊂ R, K compact, K ⊂ E (c)
Proof. (a) Call the right hand side ν(E). Fix any E ∈ Mµ .
• Proof that µ(E) ≤ ν(E):
Let a < b. Choose c0 = a < c1 < c2 < · · · with lim cn = b. Then (a, b) =
S∞ n→∞
n−1 (cn−1 , cn ] with the intervals (cn−1 , cn ] all disjoint, so that
∞
X



µ (a, b) = µ (cn−1 , cn ]
n=1

Measures 34 October 10, 2019

 P
P∞

So every ∞ n=1 µ (an , bn ) can be expressed in the form
′ ′
p=1 µ (ap , bp ] and,
consequently,
µ(E) ≤ ν(E)
• Proof that µ(E) ≥ ν(E):
S  P∞

Let ε > 0 and let E ⊂ ∞ n=1 an , bn with n=1 µ (an , bn ] ≤ µ(E) + ε.
Since F is right continuous there is, for each n ∈ N, a b′n > bn with F (b′n ) ≤
F (bn ) + 2εn . Then
∞
[ ∞
 [

E⊂ an , bn ⊂ an , b′n
n=1 n=1

so that
∞ ∞ ∞ n
X
X
X
εo
ν(E) ≤ µ (an , b′n ) ≤ µ (an , b′n ] ≤ µ (an , bn ] + n
n=1 n=1 n=1
2
( ∞
)
X

= µ (an , bn ] +ε
n=1
≤ µ(E) + 2ε

This is true for all ε > 0, so ν(E) ≤ µ(E).

(b) Call the right hand side ν̃(E). Fix any E ∈ Mµ .
• Proof that µ(E) ≤ ν̃(E):
If E ⊂ O, with O open, then µ(E) ≤ µ(O). Just taking the inf over open
O’s containing E gives µ(E) ≤ ν̃(E).
• Proof that ν̃(E) ≤ µ(E):

Let ε > 0. By part (a), there exist an , bn n∈N such that
∞
[

E⊂O= an , bn
n=1

and ∞
X

ν̃(E) ≤ µ(O) ≤ µ (an , bn ) ≤ µ(E) + ε
n=1

This is true for all ε > 0, so ν̃(E) ≤ µ(E).

(c) This is Problem Set 5, #4(a).

Measures 35 October 10, 2019

Corollary 2.42. Let E ⊂ R. The following are equivalent.
(a) E ∈ Mµ
(b) E = V \N1 where V is Gδ (a countable intersection of open sets) and µ∗ (N1 ) = 0
(c) E = H ∪ N2 where H is a countable union of compact sets and µ∗ (N2 ) = 0
Here µ∗ is the outer measure for µ.

Proof.
• Proof that (a) ⇐⇒ (c):
This is Problem Set 5, #4(b).
• Proof that (b) =⇒ (a):
As V is Gδ it is in BR ⊂ Mµ . As µ∗ (N1 ) = 0, N1 is µ∗ -measurable (see part (c)
of Carathéodory, Theorem 2.31) and hence is in Mµ . So E = V ∩ N1c ∈ Mµ .
• Proof that (a) =⇒ (b):
By Theorem 2.41(b), there is, for each k ∈ Z and j ∈ N, an open set Oj,k
such that E ∩ (k, k + 1] ⊂ Oj,k and

1
µ(Oj,k ) ≤ µ E ∩ (k, k + 1] +
2j+|k|

 

1
=⇒ µ E ∩ (k, k + 1] + µ Oj,k \ E ∩ (k, k + 1] ≤ µ E ∩ (k, k + 1] +
2j+|k|
 
1
=⇒ µ Oj,k \ E ∩ (k, k + 1] ≤
2j+|k|
Set open
covers E
∞
\  z[ }| { 
V = Oj,k ∈ Gδ
j=1 k∈Z

Then E ⊂ V and, setting N1 = V \ E,

 [   X

X 1
µ(N1 ) ≤ µ Oj,k \ E ≤ µ Oj,k \ E ≤
k∈Z k∈Z k∈Z
2j+|k|
3
≤ j
2
for each j ∈ N. So µ(E \ V ) = 0.

Measures 36 October 10, 2019