Chapter 3

Construction of the Uniform

Probability on ((0, 1], B0)

Having carefully constructed the Borel -algebra B0 of subsets of ⌦ = (0, 1], we can now
prove that there exists a uniform probability on ((0, 1], B0 ). A number of di↵erent approaches
to the construction exist.
We will take the bottom up approach (as opposed to the top down approach). That is, we will
define the uniform probability for a nice collection of sets, namely the half-open intervals,
and then show that it is possible to extend the uniform probability to the -algebra generated
by the half-open intervals, namely B0 .
This approach also has the advantage that it works in a wide variety of situations, as opposed
to the top down approach which is unique to R.

3.1 The Monotone Class Theorem

Recall from Section 1.3 that a probability is defined on a -algebra which is a collection of
sets satisfying certain properties, namely that it contains the sample space, it is closed under
complements, and it is closed under countable unions.
We then proved that as a consequence of the definition, a -algebra is necessarily closed
under finite unions, countable intersections, and finite unions.
We will now establish a result from set theory known as the monotone class theorem. This
theorem holds in quite general circumstances and, in particular, is not exclusive to R. The
existence of the uniform probability on ((0, 1], B0 ) will then follow as a corollary to the
monotone class theorem.
In order to state and prove the monotone class theorem, we need to consider di↵erent classes
(or collections) of subsets of the sample space ⌦ and not just -algebras.

49
Definition 3.1.1. A class C of subsets of ⌦ is closed under finite intersections if
n
\
Ai 2 C
i=1

for every n < 1 and for every A1 , . . . , An 2 C.

Example 3.1.2. If ⌦ = {1, 2, 3, 4} and

C = {;, {1}, {2}, {3}, {1, 2}, {1, 3}, {2, 3}, {1, 2, 3}} ,

/ C.
then C is closed under finite intersections. Note that C is not a -algebra since ⌦ 2

Definition 3.1.3. A class C of subsets of ⌦ is closed under increasing limits if

1
[
Ai 2 C
i=1

for every collection A1 , A2 , . . . 2 C with A1 ✓ A2 ✓ · · · .

A1
A3 A2
1
[
Ai
i=1

Example 3.1.4. As in the previous example, if ⌦ = {1, 2, 3, 4} and

C = {;, {1}, {2}, {3}, {1, 2}, {1, 3}, {2, 3}, {1, 2, 3}} ,

then C is closed under increasing limits.

Definition 3.1.5. A class C of subsets of ⌦ is closed under finite di↵erences if for every A,
B 2 C with A ✓ B, then B \ A 2 C.

Example 3.1.6. As in the previous example, if ⌦ = {1, 2, 3, 4} and

C = {;, {1}, {2}, {3}, {1, 2}, {1, 3}, {2, 3}, {1, 2, 3}} ,

then C is closed under finite di↵erences.

50
Example 3.1.7. We return to Problem 1.2. Let A0 denote the family of all sets of the form
m
[
A= (aj , bj ]
j=1

where 0  a1  b1 < a2 < b2 < · · · < am < bm  1 and m 2 N. If a1 = b1 , then we follow

the usual convention that (a1 , a1 ] = ;. You showed in Problem 1.2 that (0, 1] 2 A0 and that
A0 is closed under complements and finite unions. It immediately follows that A0 is closed
under finite intersections. To show that A0 is closed under finite di↵erences observe that if
A = (a, b] and B = (c, d] with A ✓ B, then B \ A = (c, a] [ (b, d]. However, A0 is not closed
under increasing limits since, for example,
1
[
(0, 1) = (0, 1 1/j ] .

j=1

Exercise 3.1.8. Carefully verify that if F is a -algebra of subsets of ⌦, then F is (i) closed
under finite intersections, (ii) closed under increasing limits, and (iii) closed under finite
di↵erences.

We are now in a position to state the monotone class theorem.

Theorem 3.1.9 (Monotone Class Theorem). Let ⌦ be a sample space, and let C be a class
of subsets of ⌦. Suppose that C is closed under finite intersections and that C contains ⌦
(that is, ⌦ 2 C). If D is the smallest class containing C which is closed under increasing
limits and finite di↵erences, then
D = (C).

Before proving this theorem, we present an example to show that the hypothesis ⌦ 2 C is
vital in the monotone class theorem.

Example 3.1.10. As in the previous example, suppose that ⌦ = {1, 2, 3, 4} and let

C = {;, {1}, {2}, {3}, {1, 2}, {1, 3}, {2, 3}, {1, 2, 3}}

so that C is closed under finite intersections. Furthermore, C is also closed under increasing
limits and finite di↵erences as a simple calculation shows. Therefore, if D denotes the smallest
class containing C which is closed under increasing limits and finite di↵erences, then clearly
D = C itself. However, C is not a -algebra; the conclusion of the monotone class theorem
suggests that D = (C) which would force C to be a -algebra. This is not a contradiction
since the hypothesis that ⌦ 2 C is not met. Suppose that

C 0 = C [ ⌦.

Then C 0 is still closed under finite intersections. However, it is no longer closed under finite
di↵erences. As a calculation shows, the smallest -algebra containing C 0 is now (C 0 ) = 2⌦ .

51
Proof. We begin by noting that the intersection of classes of sets closed under increasing
limits and finite di↵erences is again a class of that type. Hence, if we take the intersection
of all such classes, then there will be a smallest class containing C which is closed under
increasing limits and by finite di↵erences. Denote this class by D. Also note that a -algebra
is necessarily closed under increasing limits and by finite di↵erences. Thus, we conclude that
D ✓ (C). To complete the proof we will show the reverse containment, namely (C) ✓ D.
For every set B ✓ ⌦, let

DB = {A ✓ ⌦ : A 2 D and A \ B 2 D}.

Since D is closed under increasing limits and finite di↵erences, a calculation shows that DB
must also closed under increasing limits and finite di↵erences.
Since C is closed under finite intersections, C ✓ DB for every B 2 C. That is, suppose that
B 2 C is fixed and let C 2 C be arbitrary. Since C is closed under finite intersections, we
must have B \ C 2 C. Since C ✓ D, we conclude that B \ C 2 D verifying that C 2 DB
for every B 2 C. Note that by definition we have DB ✓ D for every B 2 C and so we have
shown

C ✓ DB ✓ D

for every B 2 C. Since DB is closed under increasing limits and finite di↵erences, we conclude
that D, the smallest class containing C closed under increasing limits and finite di↵erences,
must be contained in DB for every B 2 C. That is, D ✓ DB for every B 2 C. Taken together,
we are forced to conclude that D = DB for every B 2 C.
Now suppose that A 2 D is arbitrary. We will show that C ✓ DA . If B 2 C is arbitrary,
then the previous paragraph implies that D = DB . Thus, we conclude that A 2 DB which
implies that A \ B 2 D. It now follows that B 2 DA . This shows that C ✓ DA for every
A 2 D as required. Since DA ✓ D for every A 2 D by definition, we have shown

C ✓ DA ✓ D

for every A 2 D.
The fact that D is the smallest class containing C which is closed under increasing limits and
finite di↵erences forces us, using the same argument as above, to conclude that D = DA for
every A 2 D.
Since D = DA for all A 2 D, we conclude that D is closed under finite intersections.
Furthermore, ⌦ 2 D and D is closed by finite di↵erences which implies that D is closed
under complementation. Since D is also closed by increasing limits, we conclude that D is
a -algebra, and it is clearly the smallest -algebra containing C. Thus, (C) ✓ D and the
proof that D = (C) is complete.

52
3.2 Continuity of Probability
Recall from Definition 3.1.3 that a class C of subsets of ⌦ is closed under increasing limits if
1
[
Ai 2 C
i=1

for every collection A1 , A2 , . . . 2 C with A1 ✓ A2 ✓ · · · .

In addition to being important for the monotone class theorem, understanding increasing
limits of sets is crucial for establishing uniqueness and existence results for distribution
functions, the topic of Section ??
Definition 3.2.1. Let (⌦, F, P) be a probability space. We say that the sequence of events
{Aj , j 2 N} increases to A if A1 ✓ A2 ✓ · · · and
1
[
A= Ai .
i=1

Theorem 3.2.2 (Continuity of Probability). Suppose that (⌦, F, P) is a probability space.

If {Aj , j 2 N} is a sequence of events increasing to A, then

lim P (An ) = P (A) .

n!1

Proof. Since Aj ✓ Aj+1 we see that Aj \ Aj+1 = Aj . Therefore, we consider the event
Cj+1 = Aj+1 \ Acj , namely that part of Aj+1 not in Aj . For notational convenience, take
A0 = ; so that C1 = A1 . Notice that C1 , C2 , . . . are disjoint with
n
[ 1
[ 1
[
Cj = An and Cj = Aj = A.
j=1 j=1 j=1

Therefore, using the fact that P is countably additive, we conclude

1
! 1 n n
!
[ X X [
P (A) = P Cj = P (Cj ) = lim P (Cj ) = lim P Cj = lim P (An )
n!1 n!1 n!1
j=1 j=1 j=1 j=1

as required.
Definition 3.2.3. Let (⌦, F, P) be a probability space. We say that the sequence of events
{Bj , j 2 N} decreases to B if B1 ◆ B2 ◆ · · · and
1
\
B= Bi .
i=1

Exercise 3.2.4. Suppose that (⌦, F, P) is a probability space. If {Bj , j 2 N} is a sequence

of events decreasing to B, prove that

lim P (Bn ) = P (B) .

n!1

53
3.3 Uniqueness of Probability
The following uniqueness of probability theorem, which is a corollary to the monotone class
theorem and requires the continuity of probability theorem, is the key ingredient to proving
the existence of the uniform probability on ((0, 1], B0 ). Notice that this result holds for any
probability space. In particular, it is not necessary that ⌦ be a subset of R even though that
is the assumption that we are making for most of this book.

Theorem 3.3.1 (Uniqueness of Probability). Let ⌦ be a sample space and suppose that F
is a -algebra of subsets of ⌦. Suppose further that P, Q are two probabilities on (⌦, F). If
P and Q agree on a class C which is closed under finite intersections, and (C) = F, then
P = Q.

Proof. Since F is a -algebra we know that ⌦ 2 F. Since P (⌦) = Q (⌦) = 1 we can assume
without loss of generality that ⌦ 2 C. Define

D = {A 2 F : P (A) = Q (A)}

to be the class on which P and Q agree (and note that ; 2 D and ⌦ 2 D so that D is
non-empty). Using the definition of probability, a calculation shows that D is closed by
finite di↵erences and increasing limits. By assumption, we also have C ✓ D. Therefore, since
(C) = F, it follows from the monotone class theorem that D = F as required.

3.4 Distribution Functions Characterize Probabilities

on (R, B)
Recall that we defined the Borel -algebra of R in Definition 2.3.4. That is,
!
[
B= Bn
n2Z

where
Bn = ({(a, b] : n  a < b  n + 1})
for n 2 Z.
We then showed that there were a number of di↵erent representations of B, including

B = ({(a, b] : 1 < a < b < 1}).

For the purposes of studying distribution functions, it turns out that the most useful rep-
resentation of B is as the -algebra generated by finite unions of the half-closed intervals
including the half-infinite ones.

54
Theorem 3.4.1. If A is the collection of all finite unions of disjoint half-open intervals,
namely
( n )
[
A= (aj , bj ] : 1 < a1 < b1 < a2 < b2 < · · · < an < bn < 1, n 2 N
j=1

[ {( 1, x] : x 2 R}

then A is closed under finite di↵erences and (A) = B.

Proof. Suppose that B = ({(a, b] : 1 < a < b < 1}), and observe that
n
[
( 1, x] = ( j, x].
j=1

This implies (A) ✓ B. On the other hand,

(a, b] = ( 1, b] \ ( 1, a]c

implying B ✓ (A).
To verify that A is closed under finite di↵erences, suppose that (a, b] ✓ (c, d] with a > 1
and observe that
(c, d] \ (a, b] = (c, a] [ (b, d].
If a = c = 1, then (c, d] \ (a, b] = (b, d]. In all cases we see that if A, B 2 A with A ✓ B,
then we have B \ A 2 A as required.

Corollary 3.4.2. If B denotes the Borel -algebra of subsets of R, then

B = ({( 1, x] : x 2 R}).

Proof. Let A0 = {( 1, x] : x 2 R} and observe that A0 ✓ A implying that (A0 ) ✓ B. On

the other hand, observe that if 1 < a < b < 1, then

(a, b] = ( 1, b] \ ( 1, a]

proving that A ✓ A0 so that B ✓ (A0 ). Hence,

B = ({( 1, x] : x 2 R})

as required.

Theorem 3.4.3. Consider the real numbers R with the Borel -algebra B, and let P be a
probability on (R, B). The function F : R ! [0, 1] defined by F (x) = P (( 1, x]), x 2 R,
characterizes P.

55
Proof. Let
( n
)
[
A= (aj , bj ] : 1 < a1 < b1 < a2 < b2 < · · · < an < bn < 1, n 2 N
j=1

[ {( 1, x] : x 2 R}

denote the collection of all finite unions of disjoint half-open intervals as in Theorem 3.4.1.
We observe that if x, y 2 R with x < y, then

P ((x, y]) = P (( 1, y] \ ( 1, x]) = P (( 1, y]) P (( 1, x]) = F (y) F (x),

and so if A 2 A is of the form

n
[
A = ( 1, x] [ (xi , yi ]
i=1

with 1 < x < x1 < y1 < · · · < xn < yn < 1, then

n
X
P (A) = F (x) + [F (yi ) F (xi )].
i=1

Suppose that Q is another probability on (⌦, F) satisfying F (x) = Q (( 1, x]). Repeating

the argument just given shows that Q (A) = P (A) for every A 2 A. Since A is closed under
finite di↵erences and Q = P on A, we conclude from the corollary to the monotone class
theorem that Q = P on B.
Definition 3.4.4. A function F : R ! [0, 1] is called a distribution function if

(i) lim F (x) = 0 and lim F (x) = 1,

x! 1 x!1

(ii) F is right-continuous, and

(iii) F is increasing.

3.5 Null Sets

Definition 3.5.1. Let (⌦, F, P) be a probability space. A null set (or a negligible set) for
P is a subset A ✓ ⌦ such that there exists a B 2 F with A ✓ B and P (B) = 0.
Note. Suppose that B 2 F with P (B) = 0. Let A ✓ B as shown below.

56
If A 2 F, then we can conclude that P (A) = 0. However, if A 62 F, then P (A) does not
make sense.
In either case, A is a null set. Thus, it is natural to define P (A) = 0 for all null sets.
Example 3.5.2. Suppose that ⌦ = {1, 2, 3} and
F = ({1}) = {;, {1}, {2, 3}, ⌦}.
If we define P ({1}) = 1, then P ({2, 3}) = 0 so that {2, 3} 2 F is a null set for P. However,
since {2} ✓ {2, 3} we see that {2} is a null set for P with {2} 2/ F. Similarly, {3} is a null
set for P with {3} 2/ F.
Theorem 3.5.3. Suppose that (⌦, F, P) is a probability space so that P is a probability on
the -algebra F. Let N denote the set of all null sets for P. If
F 0 = F [ N = {A [ N : A 2 F, N 2 N },
then F 0 is a -algebra, called the P-completion of F, and is the smallest -algebra containing
F and N . Furthermore, P extends uniquely to a probability on F 0 (denoted by P0 ) by setting
P0 (A [ N ) = P (A)
for A 2 F, N 2 N .

Proof. To show that F 0 is a -algebra, we need to verify that the three conditions in the
definition of -algebra are met. To that end, we will first show that ⌦ 2 F 0 . Since ; 2 F
has P (;) = 0 and ; ✓ ;, we conclude that ; 2 N is a null set. If we now write ⌦ = ⌦ [ ;
then we have expressed ⌦ as a union of an event in F (namely ⌦) and a null set (namely
;). This shows ⌦ 2 F 0 . We will now show that F 0 is closed under complements. That is,
suppose E 2 F 0 so that E = A [ N for some event A 2 F and some null set N 2 N . Since
N is a null set, we know there exists some event B 2 F with P (B) = 0. We now observe
that N c = B c [ (B \ N ) and so
E c = Ac \ N c = (Ac \ B c ) [ (N c \ (B \ N ))
Since F is a -algebra and A, B 2 F , we know that Ac \ B c 2 F. Moreover, we know that
(N c \ (B \ N )) is a null set since (N c \ (B \ N )) ✓ B \ N ✓ B. This shows that E c 2 F 0
since E c can be expressed as the union of an event in F and a null set. Finally, suppose that
E1 , E2 , . . . 2 F 0 are disjoint so that Ej = Aj [ Nj where A1 , A2 , . . . is a sequence of disjoint
events and N1 , N2 , . . . are null sets. Since Nj is a null set, there exist events Bj 2 F, j 2 N,
with Nj ✓ Bj and P (Bj ) = 0. Since
1
[ 1
[
Nj ✓ Bj
j=1 j=1

and countable subadditivity of P implies

1
! 1
[ X
P Bj  P (Bj ) = 0,
j=1 j=1

57
we see that [1
j=1 Nj is a null set. Therefore,

1 1 1
! 1
!
[ [ [ [
Ej = (Aj [ Nj ) = Aj [ Nj 2 F0 (3.5.1)
j=1 j=1 j=1 j=1

since [1 1 0
j=1 Aj 2 F and [j=1 Nj is a null set. We also note that the fact that F = F [ N is a
0
-algebra implies that F must be the smallest -algebra containing F and N . Finally, to
show that P0 is a probability on (⌦, F 0 ), we begin by noting that

P0 (⌦) = P0 (⌦ [ ;) = P (⌦) = 1.

Moreover, using (3.5.1), we find that if E1 , E2 , . . . 2 F 0 are disjoint, then

1
! 1
! 1
!! 1
! 1
[ [ [ [ X
P0 E j = P0 Aj [ Nj =P Aj = P (Aj )
j=1 j=1 j=1 j=1 j=1
1
X
= P0 {Aj [ Nj }
j=1
X1
= P0 {Ej }
j=1

and the proof is complete.

58