2 Set Functions

Notation Let R∗ = R ∪ {−∞, +∞} and R+ = {x ∈ R : x ≥ 0} ∪ {+∞}.

Here +∞ and −∞ are symbols satisfying obvious conditions:

for any real number x ∈ R : −∞ < x < +∞,

(±∞) + (±∞) = x + (±∞) = (±∞) + x = ±∞,
(±∞)(±∞) = +∞ and (±∞)(∓∞) = −∞,

 ±∞ if x > 0
(±∞)x = x(±∞) = 0 if x = 0
∓∞ if x < 0,

x x
= = 0.
+∞ −∞
We see that (+∞) + (−∞) is not defined. Also 10 is still not defined.
Example 6 Let I ∗ be the collection of all intervals of types (a, b), (a, b], [a, b)
and [a, b] ⊆ R∗ . Possible set functions are
(i) ` : I ∗ → R+ , `(a, b) = b − a,
(ii) If f : R → R is a non-negative Riemann integrable function then
Rb
define F : I ∗ → R+ by F ((a, b)) = a f (t)dt.
*Definition Let A be a collection of subsets of a non-empty set X. We say
that the extended real-valued set function φ : A → R∗ is
non-negative if φ(A) ≥ 0 for all A ∈ A,
non-positive if φ(A) ≤ 0 for all A ∈ A,
finite at A if |φ(A)| < ∞,
finite if |φ(A)| < ∞ for all A ∈ A,
S all A ∈ A there exist a countable collection {An }n≥1 ⊆ A
σ-finite if for
such that A ⊆ n≥1 An and |φ(An )| < ∞ for all n ≥ 1.

2.1 (Finitely) additive functions

Definition φ : A → R∗ is finitely additive if
(i) φ(φ) = 0,
(ii) if {An }1≤n≤N ⊆ A is a finite collection of disjoint sets and N
S
n=1 An ∈
A then
N
! N
[ X
φ An = φ(An ).
n=1 n=1

Example 7 Let X = (0, 1], and A = {(a, b] : 0 ≤ a < b ≤ 1}. Define

1
 
b − a if a 6= 0
µ((a, b]) =
+∞ if a = 0.
This is an additive function. Verification is left to the student.
2.2 σ-additive set functions.
Definition µ : A → R∗ is σ-additive if
(i) µ(φ) = 0,
(ii) If {An }n≥1 ⊆ A is a countable collection of disjoint sets and ∞
S
n=1 An ∈
A then
∞
! ∞
[ X
µ An = µ(An ).
n=1 n=1

Notes (A) Since all but a finite number of the sets An in (ii) may be empty
the definition for σ-additive contains that for additive, that is, σ-additive
implies additive.
1
, n1 for each n ≥ 1. Then

(B) If we look at Example 7, take An = n+1
∞ ∞  
X X 1 1
µ(An ) = − =1
n=1 n=1
n n+1
while
∞
!
[
µ An = µ((0, 1]) = +∞.
n=1

Hence being additive does not imply being σ-additive.

A is a σ-ring then the definition is simplified since necessarily we
(C) If S
will have ∞ n=1 An ∈ A.
(D) If A is a σ-field then µ cannot take both values +∞ and −∞.
*Proof of (D) Assume otherwise so there exist F, G ∈ A such that µ(F ) =
+∞ and µ(G) = −∞.
Since A is a σ-field we have X ∈ A and so µ(X) should be defined.
But we can decompose X in two ways. In the first X = F ∪ F c and so
µ(X) = µ(F ) + µ(F c ). We cannot have µ(F c ) = −∞ since (+∞) + (−∞) is
not defined. So whatever the value for µ(F c ) we have µ(X) = +∞. Yet we
also have X = G ∪ Gc and so µ(X) = µ(G) + µ(Gc ). This time we cannot
have µ(Gc ) = +∞ and so µ(X) = −∞. This is the required contradiction.


2
Definition A map F : R → R is a distribution function if F is monoton-
ically increasing, so if x < y then F (x) ≤ F (y), and right continuous, so
limx→x0 + F (x) = F (x0 ). (*It might well be that limx→x0 − F (x) 6= F (x0 ).)

Rx
A limited collection of such F are given by F (x) = 0 f (t)dt, where f ≥ 0.
(c.f. example 6(ii)).
Notation If F : R → R is a distribution function then define the set function
µF : P → R+ by

µF ((a, b]) = F (b) − F (a).

Note that if b ≥ a then F (b) ≥ F (a) since F is increasing, and so
µF ((a, b]) ≥ 0. That is, µF is a non-negative set function.
The next result is important.
Theorem 2.1
The set function µF is σ-additive on P.
Proof is in two steps.
(A) µF is additive on P.
LetS(ai , bi ] ∈ P, 1 ≤ i ≤ n be disjoint sets for which ni=1 (ai , bi ] ∈ P.
S
Then ni=1 (ai , bi ] = (a, b] for some a and b. Since we have a finite collection
of intervals we can assume they are ordered such that

a = a1 < b1 = a2 < b2 = a3 < ... < bn = b.

(This is not necessarily possible for an infinite collection.) Then, using
−F (aj ) + F (bj−1 ) = 0 for 2 ≤ j ≤ n, we find

3
 n
!
[
µF (ai , bi ] = µF (a, b] = F (b) − F (a)
i=1
= F (bn ) + (−F (an ) + F (bn−1 ))
+ (−F (an−1 ) + F (bn−2 )) + ...
... + (−F (a2 ) + F (b1 )) − F (a1 )
= (F (bn ) − F (an )) + (F (bn−1 ) − F (an−1 ))
+(F (bn−2 ) − F (an−2 )) + ...
... − F (a2 )) + (F (b1 ) − F (a1 ))
X n
= µF ((ai , bi ]).
i=1

(B) µF is σ-additive on P.
S∞Let {(ai , bi ]}i≥1 be a S
countable collection of disjoint intervals such that
∞
i=1 (ai , bi ] ∈ P. Then i=1 (ai , bi ] = (a, b] for some a and b. We wish to
prove
∞
X
µF (a, b] = µF (ai , bi ].
i=1

The idea of the proof is to use the fact that µF is additive on P. This means
that we have to replace the infinite union by a finite one. This can be done
easily by throwing away all but a finite number of the intervals in the union.
Unsurprisingly, this introduces an error and we can only prove an inequality.
To get the inequality in the other direction we have to work harder and use
another result that allows us to replace an infinite union by a finite one. Such
a result is the Heine-Borel Theorem, Theorem 1.3. But that result deals with
open intervals covering a closed interval while we have open-closed intervals
covering an open-closed interval. We will have to “alter” our intervals so
they are of a form to which we can apply the Heine-Borel Theorem.
(B1) Let N ≥ 1 be given. Then (a, b] ⊇ N
S
i=1 (ai , bi ]. Since this is a finite
union relabel so that a1 < a2 < ... < aN . We then decompose (a, b] as
N
[ N
[ −1
(a, b] = (a, a1 ] ∪ (ai , bi ] ∪ (bi , ai+1 ] ∪ (bN , b].
i=1 i=1

Since µF is additive on P, by part (A), we have

4
 N
X N
X −1
µF (a, b] = µF (a, a1 ] + µF (ai , bi ] + µF (bi , ai·+1 ] + µF (bN , b].
i=1 i=1

On throwing away a number of these terms, we get

N
X
µF (a, b] ≥ µF (ai , bi ].
i=1

Thus we have a sequence of partial sums bounded above. This sequence

is increasing because, as we increase N we are adding non-negative terms,
µF (ai , bi ], to the partial sums. An increasing sequence bounded above con-
verges. Hence the infinite series converges and its sum satisfies
∞
X
µF (ai , bi ] ≤ µF (a, b].
i=1

(B2) Let ε > 0 be given such that a + ε < b. Recall that the distribution
function F is right continuous. This means, that at a point bi , if we move to
the right the function does not change its value by much. That is, we can
find b0i > bi for which
ε
F (bi ) ≤ F (b0i ) < F (bi ) +
. (1)
2i
(Note the factor of 1/2i . This is a “Trick”. Later we will be adding
together infinitely many of these terms and without this factor such a sum
would diverge, however small ε.) Consider
∞
[ ∞
[
[a + ε, b] ⊆ (a, b] = (ai , bi ] ⊆ (ai , b0i ).
i=1 i=1

So we now have a closed interval covered by a countable collection of open

intervals. This is exactly the situation in which we can use the Heine-Borel
Theorem. Thus we can conclude that we have a finite sub-cover of [a + ε, b].
Let M be the largest i occurring in this subcover. Then
M
[
[a + ε, b] ⊆ (ai , b0i ). (2)
i=1

Though we have a finite union the intervals are not in P and they are not
disjoint. Nonetheless, since we have a finite union we can relabel such that

5
a ≤ a1 < a + ε and a1 < a2 < a3 ... < aM < b. Also, since the open intervals
in (2) must overlap we have b0i > ai+1 for all 1 ≤ i ≤ M − 1. Using the
sequence {ai } we can decompose (a1 , b] as
M
[ −1
(a1 , b] = (ai , ai+1 ] ∪ (aM , b].
i=1

So, since µF is additive, and this is a finite disjoint union, we get

M
X −1
µF (a1 , b] = µF (ai , ai+1 ] + µF (aM , b].
i=1

But

µF (ai , ai+1 ] = F (ai+1 ) − F (ai ) by definition

≤ F (b0i ) − F (ai ) since ai+1 < b0i
ε
< F (bi ) − F (ai ) + i by definition of b0i
ε 2
= µF (ai , bi ] + i .
2
Similarly µF (aM , b] ≤ µF (aM , b0M ] < µF (aM , bM ] + ε
2M
. Hence

M −1 
X ε ε
µF (a1 , b] < µF (ai , bi ] + i
+ µF (aM , bM ] + M
i=1
2 2
∞
X
< µF (ai , bi ] + ε. (3)
i=1

We now see the importance of weighting the ε in (1) by 1/2i . From above
we have that a ≤ a1 < a + ε. So as ε → 0 we have that a1 → 0 and, again
since F is right continuous, F (a1 ) → F (a). In particular this means that as
ε → 0 we have µF (a1 , b] → µF (a, b]. So letting ε → 0 in (3) we obtain
∞
X
µF (a, b] ≤ µF (ai , bi ].
i=1

Finally, combining parts (A) and (B) we get our result. 

2.3 Extending a σ-additive function

Theorem 2.1 has given us an important example of a σ-additive set func-
tion defined on a semi-ring of subsets of R. Our aim is to enlarge the collection

6
of sets on which the set function is defined and this is what we achieve in
the next few sections. We do this in general but then apply the general re-
sults, in corollaries and examples, to the specific case of µF . To highlight the
difference between the general results and specific, I will mark the specific
results with a †.
Assume that C is a semi-ring.
Theorem 2.2 If µ : C → R+ is a non-negative σ-additive set function on
the semi-ring C, then there is a unique σ-additive set function v on R(C)
such that v = µ on C. (So we say v extends µ.) Further v is non-negative.
Proof
Let A ∈ R(C) so, by Theorem 1.7, A = nk=1 Ek for some disjoint sets
S
Ak ∈ C. Define
n
X
v(A) = µ(Ek ).
k=1

We need first show that this is S well defined. So assume that we have
another decomposition of A as A = m j=1 Fj , a disjoint union of Fj ∈ C. Put
Hkj = Ek ∩ Fj . Since C is a semi-ring we have that Hjk ∈ C for all k and j.
Trivially, Ek ⊆ A = m
S
j=1 Fj but this means that

m
! m
[ [
Ek = Ek ∩ A = Ek ∩ Fj = Hkj .
j=1 j=1
Sn
Similarly Fj = k=1 Hkj . Then

n
X n X
X m
µ(Ek ) = µ(Hkj ) since µ is additive on C,
k=1 k=1 j=1
Xm X n
= µ(Hkj ) interchanging sums,
j=1 k=1
Xm
= µ(Fj ) since µ is additive on C.
j=1

So the result is independent of the decomposition and v is well-defined.

S it is σ-additive take anyScollection {Ek }k≥1 ⊆ R of distinct sets
To show
for which k≥1 ES k ∈ R. Let E = k≥1 Ek . Then by Theorem 1.7 E ∈ R
n
means that E = r=1 Ar for disjoint setsS A r ∈ R. Similarly, by Theorem 1.7
again, Ek ∈ R means that each Ek = m i=1 Bki for disjoint sets Bki ∈ R.
k

7
 Put Crki = Ar ∩ Bki , a collection of disjoint sets from C (having used the
fact that C is a semi-ring.) Note that
mk
[[ n
[
Ar = Crki and Bki = Crki .
k≥1 i=1 r=1

Then

! n
!
[ [
v Ek =v Ar
k≥1 r=1
n
X
= µ(Ar ) by definition of v,
r=1
n mk
!
X [[
= µ Crki
r=1 k≥1 i=1
∞ X
n X
X mk
= µ (Crki ) since µ is σ-additive on C,
r=1 k=1 i=1
∞ X
X mk X
n
= µ (Crki ) interchange allowed since µ ≥ 0,
k=1 i=1 r=1
X∞ Xmk
= µ (Bki ) since µ is additive on C,
k=1 i=1
∞
X
= v(Ek ) by definition of v.
k=1

Hence v is σ-additive on R.
S other extension of µ from C to R. Then for A ∈ R decom-
Let τ be any
posed as A = ni=1 Ei , with Ei ∈ C, we have
n
X
τ (A) = τ (Ei ) since τ is additive on R,
i=1
Xn
= µ(Ei ) since τ = µ on C,
i=1
= v(A) by definition of v.
Hence v is unique. 
Corollary 2.3† The µF on P of Theorem 2.1 can be extended to a σ-additive
function on E that we still denote by µF .

8
Proof By Corollary 1.8 E is the ring generated by P. 

2.4 Measure and Outer Measure.

Definition Any non-negative set function µ : C → R+ which is σ-additive is
a measure on C.
Note Assume that µ is a measure on a ring R and E, F ∈ R with E ⊆ F .
Then F = E ∪ (F \ E), a disjoint union with F \ E ∈ R. So µ(F ) =
µ(E) + µ(F \ E) ≥ µ(E) since µ ≥ 0. Hence E ⊆ F implies µ(E) ≤ µ(F ).
We say that µ is monotonic.
Definition If C is the collection of all subsets of X then λ : C → R+ is an
outer measure on X if
(i) λ(φ) = 0,
(ii) (monotone) If E ⊆ F then λ(E) ≤ λ(F ) ,
(iii) (Countable subadditivity) For any countable collection of subsets
{An } we have
∞
! ∞
[ X
λ An ≤ λ(An ).
n=1 n=1

*Note that in (iii) we do not require the An to be disjoint. We will see later
that, for some outer measures, even if the An are disjoint we don’t necessarily
have equality.
The main result here is
Theorem 2.4
Let R be a ring of subsets of X such that X = ∞
S
i=1 Ei for some Ei ∈ R.
Let µ be a measure on R. For A ⊆ X define
∞
X
µ∗ (A) = inf µ(Aj ) (4)
j=1

where the inf is over all covers of A ⊆ ∞ ∗

S
j=1 Aj by sets Aj ∈ R. Then µ is
an outer measure on X.
Proof
We need show that the conditions of the definition of an outer measure
are satisfied.
Since µ is a measure we have that µ(Aj ) ≥ 0 for all j and so µ∗ (A) ≥ 0
for all A, that is µ∗ : C → R+ where C is the collection of all subsets of X.
(i) The empty set φ is in R so it is covered by itself and thus µ∗ (φ) ≤
µ(φ) = 0. Since we have seen that µ∗ ≥ 0 we can conclude that µ∗ (φ) = 0.

9
 (ii) If E ⊆ F then every cover of F is a cover of E. So the set of values
of ∞
P
j=1 µ(Aj ) for covers of F is contained in the set of values for covers of
E. In general, if A ⊆ B ⊆ R then inf B ≤ inf A. In our case this means that
µ∗ (E) ≤ µ∗ (F ).
∗
P∞ ∗ collection of sets, {A
(iii) Let any countable
∗
n }, be given. If µ (An ) = +∞
for some n then j=1 µ (Aj ) = +∞ since µ ≥ 0. The result is then trivial.
Assume that µ∗ (An ) < +∞ for all n. Let ε > 0 be given. Then by the
definition of infimum
S∞ or equivalently, the least upper bound, we can find a
cover of An ⊆ i=1 Gni , with Gni ∈ R such that
∞
∗
X ε
µ (An ) ≤ µ(Gni ) < µ∗ (An ) + . (5)
i=1
2n
(Note how we have used the same “trick” as in the proof of Theorem 2.1 of
weighting the errors in (5) by 1/2n .) But then

∞ ∞ ∞
!
X X X ε
µ∗ (An ) ≥ µ(Gni ) − n
n=1 n=1 i=1
2
∞
!
[
≥ µ∗ An −ε (6)
n=1

∗
S∞ S∞
having
S∞ used the definition of µ and the fact that n=1 i=1 Gni is a cover for
A
n=1 n . Since (6) is true for all ε > 0 we deduce countable subadditivity.
Corollary 2.5† From µF on E we can define µ∗F the Lebesgue-Stieltjes outer
measure on R.

10