Real Analysis MAA 6616

Lecture 5
Lebesgue Outer Measure
Let I ⊂ R be an interval. The length ℓ(I) is defined to be ℓ(I) = ∞ if I is unbounded and
ℓ(I) = b − a if I has finite left endpoint a and finite right endpoint b.
Let E ⊂ R be any set in R and let {Ij }∞ j=1 be a collection of bounded open intervals that cover
∞
[ X∞
the set E. That is E ⊂ Ij . We associate to such a cover ℓ(Ij ) (which could be ∞). Define
j=1 j=1
the outer measure of E as
 
X∞ ∞ 
[
m∗ (E) = inf ℓ(Ij ) : E ⊂ Ij .
 
j=1 j=1

The following properties follow directly from the definition.

▶ m∗ (∅) = 0
▶ The outer measure is monotone. That is, if E ⊂ F, then m∗ (E) ≤ m∗ (F)

Proposition
If C ⊂ R is countable, then m∗ (C) = 0.

Proof. ∞
ϵ ϵ ϵ
  [
Suppose C = {xn }. Let ϵ > 0. For each n, let In = xn − , xn + . Then C ⊂ In and ℓ(In ) = .
2n+1 2n+1 n=1
2n
∞ ∞
ϵ
= ϵ. Therefore m∗ (C) ≤ ϵ. Since ϵ > 0 is arbitrary, we have m∗ (C) = 0.
X X
Hence ℓ(In ) =
n=1 n=1
2n
Proposition
Let I ⊂ R be an interval. Then m∗ (I) = ℓ(I). In particular m∗ (R) = ∞.

Proof.
• Case I = [a, b], closed and bounded. Let ϵ > 0. Then I ⊂ (a − ϵ, b + ϵ) and it follows from the definition of outer
measure that m∗ (I) ≤ (b + ϵ) − (a − ϵ) = ℓ(I) + 2ϵ. Since ϵ > 0 is arbitrary, then m∗ (I) ≤ ℓ(I).
To prove the proposition in this case, we need to prove that m∗ (I) ≥ ℓ(I). First suppose that I covered by finitely many open
n
X
and bounded interval I1 , · · · , In . We are going to show that ℓ(Ij ) ≥ ℓ(I). Since a ∈ I, then there exists
j=1
j1 ∈ {1, · · · , n} such that a ∈ Ij1 = (a1 , b1 ). If b ≤ b1 , then ℓ(I) < ℓ(Ij1 ) and we are done. If not, then a < b1 < b,
so that b1 ∈ I and there is j2 ∈ {1, · · · , n} such that b1 ∈ Ij2 = (a2 , b2 ). If b ≤ b2 , then ℓ(I) < ℓ(Ij1 ) + ℓ(Ij2 ). If
not, we can repeat this process which must end after at most k steps with k ≤ n to get

k
X n
X
ℓ(I) < ℓ(Ijp ) < ℓ(Ij )
p=1 j=1

Now if {Ip }∞ n
p=1 is any countable collection of open and bounded sets that cover I, then it has a finite subcover {Ip }p=1 for I
(Heine-Borel). It follows then from the previous paragraph that

n
X ∞
X
ℓ(I) < ℓ(Ip ) ≤ ℓ(Ij )
p=1 j=1

This shows that m∗ (I) ≥ ℓ(I) and therefore m∗ (I) = ℓ(I) in this case.
Proof.
• Case I bounded interval. Let a and b be the left and right endpoints of I. For ϵ > 0 (small) let

ϵ ϵ ϵ ϵ
J1 = [a + , b− ] and J2 = [a − , b+ ]
2 2 2 2

Hence J1 and J2 are closed bounded intervals with ℓ(J1 ) = ℓ(I) − ϵ, ℓ(J2 ) = ℓ(I) + ϵ and J1 ⊂ I ⊂ J2 . It follows from
the first case (applied to J1 and J2 ) and the monotinicity of the outer measure that

∗ ∗ ∗
ℓ(I) − ϵ = ℓ(J1 ) = m (J1 ) ≤ m (I) ≤ m (J2 ) = ℓ(J2 ) = ℓ(I) + ϵ

Therefore m∗ (I) = ℓ(I) since ϵ > 0 is arbitrary.

• Case I unbounded interval. If I = R, then m∗ (I) = ∞ = ℓ(I). Suppose I ̸= R and it is unbounded above so that
ℓ(I) = ∞. Let c ∈ I. For each n ∈ N the closed and bounded interval In = [c, c + n] is contained in I. It follows from the
monotonocity of the outer measure that

∗ ∗
m (I) ≥ m (In ) = n for all n ∈ N

Therefore m∗ (I) = ∞ = ℓ(I).

For set E ⊂ R and s ∈ R, the translate of E is the set E + s = {x = e + s : e ∈ E}.
Note: Let I ⊂ R be an interval with endpoints a, b, then I + s is an interval with endpoints
a + s, b + s and ℓ(I + s) = ℓ(I).

Proposition
For every E ⊂ R and s ∈ R, we have m∗ (E + s) = m∗ (E). The outer measure m∗ is
invariant under translations.

Proof.
The collection of open and bounded intervals {Ij }∞
j=1 covers E if and only the collection of open and bounded translate
intervals {Ij + s}∞ ∗ ∗
j=1 covers the translate E + s. This implies m (E + s) = m (E).

Proposition
Let {Ej }∞
j=1 be a countable collection of subsets of R. Then

 
∞
[ ∞
X
∗
m  Ej  ≤ m∗ (Ej ) . (∗)
j=1 j=1

The outer measure m∗ is subadditive.

Proof. X ∗
X ∗

First note that if m Ej = ∞, then (∗) holds trivially. Suppose then m Ej < ∞. Let ϵ > 0. It follows from
j j
the definition of m∗ that for each j, there is a countable collection of open and bounded intervals {Ij,k }∞
k=1 such that
[∞
Ej ⊂ Ij,k and
k=1
∞
X ∗
ϵ
ℓ(Ij,k ) < m Ej + .
k=1
2j

The collection of open and bounded intervals {Ij,k }(j,k)∈N×N is countable and

 
∞
[ ∞
[ ∞
[
Ej ⊂  Ij,k  .
j=1 j=1 k=1

Furthermore,    
∞ ∞ ∞
∗
[ X X
m  Ej  ≤  ℓ(Ij,k )
j=1 j=1 k=1

∞  ∞
ϵ

∗ ∗
X
X

≤ m Ej + = m Ej + ϵ
2 j
j=1 j=1

And (∗) follows since ϵ > 0 is arbitrary.

Measurable Sets

A set E ⊂ R is said to be measurable if for every set S ⊂ R, the following equality holds:

m∗ (S) = m∗ (S ∩ E) + m∗ (S ∩ Ec ) , (∗∗)

where Ec = R\E is the R-complement of E.

▶ ∅ and R are measurable.
▶ E is measurable if and only if Ec is measurable.
▶ If E is measurable and T ⊂ R is disjoint from E , then m∗ (E ∪ T) = m∗ (E) + m∗ (T).
Indeed E ∪ T = [(E ∪ T) ∩ E] ∪ [(E ∪ T) ∩ Ec ] and it follows from (∗∗) and from
E ∩ T = ∅ that
m∗ (E ∪ T) = m∗ ((E ∪ T) ∩ E) + m∗ ((E ∪ T) ∩ Ec ) = m∗ (E) + m∗ (T).
▶ E is measurable if and only if for every set S ⊂ R, m∗ (S) ≥ m∗ (S ∩ E) + m∗ (S ∩ Ec ).
This is a consequence of the definition of measurable set and the subadditivity of the
outer measure.
▶ Any set E with outer measure 0 (m∗ (E) = 0) is measurable. Indeed, let S ⊂ R. Then
S ∩ E ⊂ E implies that m∗ (S ∩ E) = 0, and S ∩ Ec ⊂ S implies m∗ (S ∩ Ec ) ≤ m∗ (S).
Hence, by the subadditivity we have

m∗ (S) ≤ m∗ (S ∩ E) + m∗ (S ∩ Ec ) ≤ 0 + m∗ (S)

and (∗∗) holds.

Proposition
If E1 and E2 are measurable sets, then E1 ∪ E2 is measurable. In general if E1 , · · · , En are
measurable sets, then their union is also measurable.

Proof.
Let S ⊂ R be any set. The following identities (verification left as exercise) will be used

c
c c
c
S ∩ (E1 ∪ E2 ) = (S ∩ E1 ) ∪ S ∩ E1 ∩ E2 ; S ∩ (E1 ∪ E2 ) = S ∩ E1 ∩ E2

Note that the subadditivity of m∗ and the first identity above imply that

∗ ∗ c
∗
m (S ∩ E1 ) + m S ∩ E1 ∩ E2 ≥ m (S ∩ (E1 ∪ E2 )) .

Using the measurability of E1 and then of E2 and the above identities, we have

m∗ (S) = m∗ (S ∩ E1 ) + m∗ S ∩ E1c = m∗ (S ∩
E1 ) + m∗ S ∩ E1c ∩ E2 + m∗ S ∩ E1c ∩ E2c

≥ m∗ (S ∩ (E1 ∪ E2 )) + m∗ S ∩ E1c ∩ E2c

≥ m∗ (S ∩ (E1 ∪ E2 )) + m∗ (S ∩ (E1 ∪ E2 )c ) ≥ m∗ (S)

Therefore E1 ∪ E2 satisfies (∗∗) and it is measurable. The case of finite union of measurable sets follows by induction.
Proposition
Let E1 , · · · , En be measurable sets that are mutually disjoint (Ei ∩ Ej = ∅ if i ̸= j) and let
S ⊂ R be any set. Then
    
n
[ n
X n
[ n
X
∗ ∗ ∗
m S ∩  Ej  = m (S ∩ Ej ) and m  Ej  = m∗ (Ej )
j=1 j=1 j=1 j=1

Proof.
The second relation is a special case of the first with S = R. We prove the first relation by induction. The case n = 1 is trivial.
Suppose that the relation holds up to n − 1. Let E1 , · · · , En be measurable and mutually disjoint. Then for S ⊂ R we have
the following identities

     
n n n−1
c
[ [ [
S∩ Ej  ∩ En = S ∩ En and S ∩  Ej  ∩ En = S ∩  Ej  .
j=1 j=1 j=1

It follows from these identities, the measurability of En and the induction hypothesis that

     
n n−1
∗ ∗ ∗
[ [
m S ∩  Ej  = m (S ∩ En ) + m S ∩  Ej 
j=1 j=1
n−1
∗ ∗
X

= m (S ∩ En ) + m S ∩ Ej
j=1
n
∗
X

= m S ∩ Ej
j=1
Proposition
∞
[
Let {Ej }∞
j=1 be a countable collection of measurable sets. Then E = Ej is measurable.
j=1

Proof.
First we prove that E can be expressed as
the disjoint union of a countable family of measurable sets. For each n ∈ N, let
n−1

n−1 c
′
Ek  . Then En′ ∩ Em ′
= ∅ if n ̸= m; En′ is measurable (as a finite intersection of
[ [
En = En \ Ek = En ∩ 
k=1 k=1
∞ ∞
′
[ [
measurable sets); and En = En = E.
n=1 n=1
j
′
[
For each n ∈ N, let Fn = Ej . Then Fn is measurable (finite union of measurable sets). Note that Ec ⊂ Fnc for all n. Now
j=1
let S ⊂ R be any set. Then we use the measurability of Fn and the monotonicity of m∗ to get

∗ ∗ ∗ c
∗ ∗ c

m (S) = m (S ∩ Fn ) + m S ∩ Fn ≥ m (S ∩ Fn ) + m S ∩ E .

Since Fn is the disjoint union of finitely many measurable set, then it follows from a previous proposition that
n   n  
∗ ′ ∗ ∗ ′ ∗ c

m∗ (S ∩ Fn ) =
X X
m S ∩ Ej . Hence, m (S) ≥ m S ∩ Ej + m S ∩ E for all n ∈ N, and it follows
j=1 j=1
from the subadditivity of m∗ that

∞  
∗ ∗ ′ ∗ c
∗ ∗ c

X
m (S) ≥ m S ∩ Ej + m S ∩ E ≥ m (S ∩ E) + m S ∩ E .
j=1

Therefore E is measurable.
σ-Algebra of Measurable Sets

Recall that a σ-algebra in R is a collection of subsets containing R; is closed under formation of

complement and countable union. The previous propositions imply that the collection of
measurable sets forms a σ-algebra.

Theorem
An interval in R is measurable.

Proof.
Consider the case I = (a, ∞) with a ∈ R. Let S ⊂ R be any set. Since m∗ (S\{a}) = m∗ (S), we can assume a ∈ / S.
Let S1 = S ∩ I c
=S ∩ (−∞, 2
 a) and S = S ∩ I = S ∩ (a, ∞). To prove that I is measurable, it is enough to show that
m∗ (S) ≥ m∗ S1 + m∗ S2 .
Let {Ij }∞ 1
j=1 be a countable collection of open and bounded intervals that cover S. For each j ∈ N let Ij = Ij ∩ (−∞, a)
and Ij2 = Ij ∩ (a, ∞). Then Ij1 , Ij2 are open bounded intervals such that ℓ(Ij ) = ℓ(Ij1 ) + ℓ(Ij2 ) and the collections
 
{Ij1 }∞ 2 ∞ ∗ ∗
1 2 P∞
j=1 and {Ij }j=1 cover S and S , respectively. It follows from m as an infimum that m S1 ≤ 1
j=1 ℓ(Ij ) and
 
m∗ S 2 ≤ ∞ 2
P
j=1 ℓ(Ij ). Hence,

    ∞   ∞
∗ 1 ∗ 2 1 2
X X
m S +m S ≤ ℓ(Ij ) + ℓ(Ij ) ≤ ℓ(Ij ) .
j=1 j=1

   
Since {Ij }∞
j=1 is an arbitrary cover of S, it follows m
∗
S1 + m∗ S1 ≤ m∗ (S) and I is measurable.
Finally by using the fact that measurable sets form a σ-algebra, it can be proved that any interval is measurable.
It follows from the preceding results that:
▶ Any open set is measurable (since it can be written as a countable union of intervals).
▶ Any closed set is measurable (as a complement of an open set)
▶ Any Gδ set is measurable (recall that a Gδ set is a set that can be written as the
intersection of a countable collection of open sets)
▶ Any Fσ set is measurable (recall that an Fσ set is a set that can be written as the union of
a countable collection of closed sets)
▶ Recall that a Borel σ-algebra is the σ-algebra generated by open sets (it is contained in
any σ-algebra that contains open sets). Its members are called Borel sets. Since
measurable sets form a σ-algebra, it follows that any Borel set is measurable.
We can summarize the above as:
Theorem
The collection M of measurable sets forms a σ-algebra that contains the σ-algebra of Borel
sets. Each interval is measurable, each open set is measurable, each closed is measurable, each
Gδ is measurable, and each Fσ is measurable.