You are on page 1of 168

The Theory of Measures and

Integration
A Solution Manual for Vestrup (2003)
Jianfei Shen
School of Economics, The University of New South Wales
Sydney, Australia
I hear, I forget;
I see, I remember;
I do, I understand.
Old Proverb
Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
1 Set Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 -Systems, z-Systems, and Semirings . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 -Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 z-System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.3 Semiring. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3 o-Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.4 The Borel o-Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2 Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.1 Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2 Continuity of Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.3 A Class of Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3 Extensions of Measures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.1 Extensions and Restrictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.2 Outer Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.3 Carathodorys Criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.4 Existence of Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.5 Uniqueness of Measures and Extensions . . . . . . . . . . . . . . . . . . . . . . . . 63
3.6 The Completion Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.7 The Relationship between o(A) and M(j
+
) . . . . . . . . . . . . . . . . . . . . . 67
3.8 Approximations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.9 A Further Description of M(j
+
) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4 Lebesgue Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.1 Lebesgue Measure: Existence and Uniqueness. . . . . . . . . . . . . . . . . . . 71
4.2 Lebesgue Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.3 Translation Invariance of Lebesgue Measure . . . . . . . . . . . . . . . . . . . . 73
v
vi CONTENTS
5 Measurable Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.1 Measurability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2 Combining Measurable Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.3 Sequences of Measurable Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.4 Almost Everywhere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.5 Simple Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.6 Some Convergence Concepts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.7 Continuity and Measurability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.8 A Generalized Denition of Measurability . . . . . . . . . . . . . . . . . . . . . . 90
6 The Lebesgue Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.1 Stage One: Simple Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.2 Stage Two: Nonnegative Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.3 Stage Three: General Measurable Functions . . . . . . . . . . . . . . . . . . . . . 104
6.4 Stage Four: Almost Everywhere Dened Functions . . . . . . . . . . . . . . 109
7 Integrals Relative to Lebesgue Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.1 Semicontinuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
8 The L
p
Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
8.1 1
]
Space: The Case 1 6 < o . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
8.2 The Riesz-Fischer Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
8.3 1
]
Space: The Case 0 < < 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
8.4 1
]
Space: The Case = o. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
8.5 Containment Relations for 1
]
Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 125
8.6 Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
8.7 More Convergence Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
8.8 Prelude to the Riesz Representation Theorem . . . . . . . . . . . . . . . . . . 129
9 The Radon-Nikodym Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
9.1 The Radon-Nikodym Theorem, Part I . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
10 Products of Two Measure Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
10.1 Product Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
10.2 The Fubini Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
11 Arbitrary Products of Measure Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
11.1 Notation and Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
11.2 Construction of the Product Measure. . . . . . . . . . . . . . . . . . . . . . . . . . . 147
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
List of Figures
1.1 Inclusion between classes of sets A _ 2

. . . . . . . . . . . . . . . . . . . . . . . . 1
3.1 L T M(v) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
6.1
1
and
2
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.2
n
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
8.1
]
and
q
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
8.2
n

0, but
n
,
1
p
0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
8.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
vii
Preface
Sydney, Jianfei Shen
ix
Acknowledgements
xi
1
SET SYSTEMS
Remarks
Remark 1.1. Klenke (2008, Fig. 1.1, p.7) provides a chart to indicate the rela-
tionships among the set systems. Here I replicate his chart; see Figure 1.1.
o-eld
o-ring
Ring

A
-closed
L-closed
Semiring

A
T =

n
i=1
T
i
-closed
Field

A
-closed
L-closed
z-system

A
_ T ==T A

o
n=1

n
A
o-L-stable A -stable
A o-L-stable
L-stable
Figure 1.1. Inclusion between classes of sets A_ 2

Semiring
S
-stable
!Ring See part (g) of Exercise 1.22;
1
2 CHAPTER 1 SET SYSTEMS
-ring
2A
!-eld See part (b) of Exercise 1.43;
Ring
2A
!Field A and A is closed under dierence implies that
A ==
c
= A;
-systme
T
-stable
!-eld See Exercise 1.10.
Remark 1.2. This notes is for Exercise 1.34 (p.17). See Klenke (2008, Example
1.40, p.18-19). We construct a measure for an innitely often repeated random
experiment with nitely many possible outcomes (Product measure, Bernoulli
measure). Let S be the set of possible outcomes. For s S, let
x
> 0 be the
probability that s occurs. Hence

xS

x
= 1. For a xed realization of the
repeated experiment, let :
1
(o). :
2
(o). . . . S be the observed outcomes. Hence
the space of all possible outcomes of the repeated experiment is = S
N
. We
dene the set of all sequences whose rst n values are :
1
(o). . . . . :
n
(o):
:
1
(o). . . . . :
n
(o)| =

o
t
: :
i
(o
t
) = :
i
(o) for any i = 1. . . . . n
_
. (1.1)
Let C
0
= {]. For n N, dene the class of cylinder sets that depend only on
the rst n coordinates
C
n
= {:
1
(o). . . . . :
n
(o)| : :
1
(o). . . . . :
n
(o) S] . (1.2)
and let C :
_
o
n=0
C
n
.
We interpret :
1
(o). . . . . :
n
(o)| as the event where the outcome of the rst
experiment is :
1
(o), the outcome of the second experiment is :
2
(o) and -
nally the outcome of the n-th experiment is :
n
(o). The outcomes of the other
experiments do not play a role for the occurrence of this event. As the indi-
vidual experiments ought to be independent, we should have for any choice
:
1
(o). . . . . :
n
(o) 1 that the probability of the event :
1
(o). . . . . :
n
(o)| is the
product of the probabilities of the individual events.
1.1 -Systems, -Systems, and Semirings
1.1.1 -Systems
I Exercise 1.3 (1.1.1). Let = (. |. Let P consists of along with the rsc
subintervals of . P is a -system of subsets of (. |.
Proof. Let = (a. b| and T = (c. J| be P-sets. Then either T = P, or
T = (a c. b . J| P. |L
I Exercise 1.4 (1.1.2). Must be in every -system?
Solution. Not necessary. For example, let
SECTION 1.1 t-SYSTEMS, 2-SYSTEMS, AND SEMIRINGS 3
= (0. 1|. = (0. 1,2|. T = (1,4. 1|. C = (1,4. 1,2|.
and let P = {. T. C]. Then P is a -system on , and P. Generally, if
T = for any . T in a -system, then does not in this -system. |L
I Exercise 1.5 (1.1.3). List all -systems consisting of at least two subsets of
{o
1
. o
2
. o
3
].
Solution. These -systems are:
v

{o
i
]. {o
i
. o
}
]
_
, (i. ) {1. 2. 3]
2
and = i ;
v

{o
i
]. {o
1
. o
2
. o
3
]
_
;
v

{o
i
. o
}
]. {o
1
. o
2
. o
3
]
_
;
v

{o
i
]. {o
i
. o
}
]. {o
1
. o
2
. o
3
]
_
;
v

. {o
i
]. {o
i
. o
}
]
_
, i = 1. 2. 3, and = i ;
v

. {o
i
. o
}
]. {o
1
. o
2
. o
3
]
_
;
v

. {o
i
]. {o
i
. o
}
]. {o
1
. o
2
. o
3
]
_
. |L
I Exercise 1.6 (1.1.4). If P
k
consists of the empty set and the k-dimensional
rectangles of any one form, then P
k
is a -system of subsets of R
k
.
Proof. Let A. B P
k
be two k-dimensional rectangles of any form. We also
write A =
1

2

k
and B = T
1
T
k
, where
i
and T
i
are rsc
intervals for every i {1. . . . . n]. We also assume that A = and B = ; for
otherwise A B = P
k
is trivial. Then
A B = (
1

k
) (T
1
T
k
) =
k

i=1
(
i
T
i
) P
k
since
i
T
i
is a rsc interval in R. |L
I Exercise 1.7 (1.1.5). Let P consist of and all subsets of R
k
that are neither
open nor closed. Then P is not a -system of subsets of R
k
.
Proof. To get some intuition, let k = 1. Consider two P-sets: = (0. 1,2|
and T = 1,4. 1). Note that neither nor T are open or closed on R, but their
intersection T = 1,4. 1,2| is closed on R, and is not in P.
Now consider the k-dimensional case. Let A. B P; let A =

k
i=1

i
and
B =

k
i=1
T
i
; particularly, we let
i
= (a
i
. b
i
| and T
i
= c
i
. J
i
), where a
i
< c
i
<
b
i
< J
i
. Then (a
i
. b
i
| c
i
. J
i
) = c
i
. b
i
| = , and A B =

k
i=1
(
i
T
i
) =

k
i=1
c
i
. b
i
| is closed on R
k
. |L
I Exercise 1.8 (1.1.6). For each in a nonempty index set , let P

be a -
system over .
4 CHAPTER 1 SET SYSTEMS
a. The collection
_

is a -system on .
b. Let A _ 2

. Suppose that {P

: ] is the exhaustive list of all the -


system that contain A. In other words, each P

_ A, and any -system that


contains Acoincides with some P

. Then
_

is a -system that contains


A. If Q is a -system containing A, then
_

_ Q. The minimal -system


generated by A always exists.
c. Suppose that P is a -system with P _ A, and suppose that P is contained in
any other -system that contains A. Then P =
_

, with notation as in
(b). The minimal -system containing A [which always exists] is also unique.
Proof. (a) Suppose T. C
_

. Then T. C P

for every . Since P

is a -system, we have TC P

for all . Consequently, TC


_

,
i.e.,
_

is a -system on .
The analogous statement holds for rings, o-rings, algebras and o-algebras.
However, it fails for semirings. A counterexample: let = {1. 2. 3. 4], A
1
=
{. . {1]. {2. 3]. {4]], and A
2
= {. . {1]. {2]. {3. 4]]. Then A
1
and A
2
are
semirings but A
1
A
2
= {. . {1]] is not.
(b) Since 2

{P

: ] = (A), the family (A) is nonempty. It follows


from (a) that
_

is a -system containing A. Finally, if Q is a -system


containing A, then Q (A), hence
_

_ Q.
(c) Since
_

is the -system generated by A, we have


_

_ P;
since P is contained in any other -system that contains A, we have P _
_

. |L
1.1.2 -System
I Exercise 1.9 (1.1.7). This exercise explores some equivalent denitions of a
z-system.
1
a. L is a z-system i L satises (z
1
), (z
t
2
), and (z
3
).
b. Every z-system additionally satises (z
4
), (z
5
), and (z
6
).
c. L is a z-system i L satises (z
1
), (z
t
2
), and (z
5
).
1
The conditions are:
(
1
) L;
(
2
) L ==
c
L;
(
0
2
) , B L & _ B ==B ~ L;
(
3
) For any disjoint
n

1
nD1
_ L,
_
1
nD1

n
L;
(
4
) , B L & B = ==LB L;
(
5
) V
n

1
nD1
_ L,
n
]==
_
1
nD1

n
L;
(
6
) V
n

1
nD1
_ L,
n
]==
_
1
nD1

n
L.
SECTION 1.1 t-SYSTEMS, 2-SYSTEMS, AND SEMIRINGS 5
d. If a collection L is nonempty and satises (z
2
) and (z
3
), then L is a z-system.
Proof. (a) Let L be a z-system. Then L by (z
1
) and (z
2
). Suppose that
. T L and _ T. Then T
c
L by (z2) and T
c
= . By (z
3
), T
c
L =
T
c
L L L L L. By (z
2
) again, T = (T
c
L )
c
L.
To show the inverse direction, we need only to show that (z
1
) and (z
t
2
) imply
(z
2
): if L, then
c
= L.
(b) Let L be a z-system, so it satises (z
1
)(z
3
) and (z
t
2
). To verify that (z
4
)
holds, rst notice that =
c
L. If . T L and T = , then L T =
L T L L L L.
To see that (z
5
), let {
n
] _ L be increasing. Let T
1
=
1
and T
n
=
n

n-1
for n > 2. Then {T
n
] _ L by (z
t
2
) and is disjoint. Hence,
_

n
=

T
n
L.
Finally, if {
n
] _ L is decreasing, then {
c
n
] _ L is increasing. Hence
_

c
n

L by (z
5
). Then
_

n
= (
_

c
n
)
c
L.
(c) If L is a z-system, it follows from (a) and (b) that (z
t
2
) and (z
5
) hold. Now
suppose that (z
1
), (z
t
2
), and (z
5
) hold. It follows from the only if part of (a) that
(z
1
) and (z
t
2
) imply (z
2
). To see (z
3
) also hold, let {
n
]
o
n=1
_ L be a disjoint
sequence. We can construct a nondecreasing sequence {T
n
]
o
n=1
by letting T
n
=
_
n
i=1

i
. Notice that T
n
L for all n. Hence,
_
o
n=1
T
n
=
_
o
n=1

n
, and by (z
5
),
we have (z
3
).
(d) If L = and satises (z
2
) and (z
3
), then there exists some L and so
= L
c
L by (z
4
). |L
I Exercise 1.10 (1.1.8). If L is a z-system and a -system, then
_
o
n=1

n
L
whenever
n
L for all n N. That is, L is closed under countable unions.
Proof. This exercise proves that a z-system which is
_
-stable is a o-eld (see
Figure 1.1). Let {
n
]
o
n=1
_ L. Let T
1
=
1
and T
n
=
c
1

c
2

c
n-1

n
for all n > 2. Since L is a z-system, {
c
1
. . . . .
c
k-1
] _ L; since L is a -system,
T
n
L. It follows from (z
3
) that
_
o
n=1

n
=
_
o
n=1
T
n
L. |L
I Exercise 1.11 (1.1.9). A z-system is not necessarily a -system.
Proof. For example, let = (0. 1|. The following collection is a z-system:
L = {. . (0. 1,2|. (1,4. 1|. (1,2. 1|. (0. 1,4|] .
However, L is not a -system because (0. 1,2| (1,4. 1| = (1,4. 1,2| L. |L
I Exercise 1.12 (1.1.10). Find all z-systems over = {o
1
. o
2
. o
3
. o
4
] with at
least three elements.
Solution.

. . {o
i
]. {o
}
. o
k
. o
I
]
_
i = = k =

. . {o
i
. o
}
]. {o
k
. o
I
]
_
i = = k = .
|L
6 CHAPTER 1 SET SYSTEMS
I Exercise 1.13 (1.1.11). The collection consisting of and the rsc intervals is
not a z-system on R.
Proof. This is not a z-system, but is a semiring. Consider a nontrival rsc in-
terval (a. b|. Note that (a. b|
c
= (o. a| L(b. o) is not a rsc interval, and so is
not in this collection. |L
I Exercise 1.14 (1.1.12). Suppose that for each in a nonempty index set ,
L

is a z-system over .
a. The collection
_

is a z-system on .
b. Suppose that A _ 2

is such that Ais contained in each L

, and suppose that


{L

: ] is the exhaustive list of all the z-system that contain A. Then


_

is a z-system that contains A. If Q is a z-system on that contains


A, then
_

_ Q. The minimal z-system generated by A always exists.


c. Let L denote a z-system over with L _ A and where L is contained in any
other z-system also containing A. Then L =
_

, with notation as in (b).


Therefore, the z-system generated by A always exists and is unique.
Proof. (a) It is clear that
_

. Suppose
_

, then L

for any . Hence,


c
L

for any . So
c

, i.e.,
_

is closed under complementation. To see


_

is closed under disjoint


unions, let {
n
]
o
n=1
_
_

be a disjoint sequence. Then {


n
]
o
n=1
_ L

for
any implies
_
o
n=1

n
L

for any , which implies that


_
o
n=1

n

_

.
(b) From (a) we know
_

is a z-system, and since A _ L

, V ,
we know that A _
_

; hence,
_

is a z-system that contains A.


_

_ Q because Q {L

: ].
(c) Since L is contained in any other z-system containing A, and
_

is
such a z-system, so L _
_

. Since L {L

: ], so
_

_ L. |L
1.1.3 Semiring
I Exercise 1.15 (1.1.13). Is A = {] L{(0. .| : 0 < . 6 1] a semiring over (0. 1|?
Solution. A is not a semiring on (0. 1|. Take (0. .| and (0. ,| with . < ,. Then
(0. ,| (0. .| = (.. ,| A since . > 0 by denition. |L
I Exercise 1.16 (1.1.14). This exercise explores some alternative denitions of
a semiring.
a. Some dene A to be a semiring i A is a nonempty -system such that both
1. J A and 1 _ J imply the existence of a nite collection C
0
. C
1
. . . . . C
n

A with 1 = C
0
_ C
1
_ _ C
n
_ J and C
i
C
i-1
A for i = 1. . . . . n. This
denition of a semiring is equivalent to our denition of a semiring.
SECTION 1.1 t-SYSTEMS, 2-SYSTEMS, AND SEMIRINGS 7
b. Some dene A to be a semiring by stipulating (SR1), (SR2), and the following
property: . T A implies the existence of disjoint A-sets C
0
. C
1
. . . . . C
n
with
T =
_
n
i=0
C
i
. Note that here T is not necessarily a proper dierence.
If A is a semiring by this denition, then A is a semiring by our denition,
but the converse is not necessarily true.
Proof. (a) We rst show that (SR1), (SR2), and (SR3) imply the above denition.
(SR1) and (SR2) imply that A is a nonempty -system (since A). Let 1. J
A and 1 _ J. By (SR3) there exists disjoint D
1
. . . . . D
n
A such that J
1 =
_
n
i=1
D
i
. Let C
0
= 1 and C
i
= 1 L D
1
L L D
i
for i = 1. . . . . n. Then
1 = C
0
_ C
1
_ _ C
n
= J, and C
i
C
i-1
= D
i
A.
Now suppose (a) holds. (SR1): Since A is nonempty, there exists 1 A;
since 1 _ 1, there exists a nite collection 1 = C
0
_ C
1
_ _ C
n
_ 1,
which implies that C
0
= C
1
= = C
n
, and so C
i
C
i-1
= A. (SR2) holds
trivially. (SR3): Let . T A and _ T. Then by the assumption, there exists
a nite collection C
0
. C
1
. . . . . C
n
A with = C
0
_ C
1
_ _ C
n
_ T, and
T
n
= C
n
C
n-1
A. Then {T
i
]
n
i=1
_ A is disjoint, and
L
_
_
n
_
i=1
T
i
_
_
= L
_
_
n
_
i=1
(C
i
C
i-1
)
_
_
= L (T ) = T.
(b) Some authors do apply this denition, for example, see Aliprantis and Bor-
der (2006); Dudley (2002). The proof is obvious. |L
I Exercise 1.17 (1.1.15). Let A consist of as well as all rsc rectangles (a. b|.
The collection of all nite disjoint unions of A-sets is a semiring over R
k
.
Proof. We prove a more general theorem. See Bogachev (2007, Lemma 1.2.14,
p.8).
For any semiring , the collection of all nite unions of sets in forms a ring
R.
Proof. It is clear that the class R admits nite unions. Suppose that =
_
n
i=1

n
and T =
_
k
}=1
T
k
, where
i
. T
i
. Then we have T =
_
i6n,}6k

i
T
}
R. Note that
i
T
}
A, V i {1. . . . . n] and {1. . . . . k],
since a semiring is
_
-stable. Hence R admits nite intersections. In addition,
T =
n
_
i=1
_
_

i

k
_
}=1
T
}
_
_
=
n
_
i=1
k
_
}=1
_

i
T
}
_
.
Since the set
i
T
}
=
i

_

i
T
}
_
is a nite union of sets in , one has

i
T
}
R. Furthermore,
_
k
}=1
_

i
T
}
_
because is
_
-stable. Finally, the
nite list

i
T
}
_
i1,...,n,}1,...,k
is disjoint; hence, T is a nite disjoint
union of sets in .
8 CHAPTER 1 SET SYSTEMS
Now, since A is a semiring [which is a well known fact], we conclude that
the collection of all nite disjoint unions of A-sets is a ring over R
k
[a ring is a
semiring, see Exercise 1.22 (p.10)]. |L
I Exercise 1.18 (1.1.16). An arbitrary intersection of semirings on is not
necessarily a semiring on .
Solution. Unlike the other kinds of classes of families of sets (e.g., Exer-
cise 1.8 and Exercise 11.2), the intersection of a collection of semirings need
not be a semiring. For example, let = {0. 1. 2], A
1
= {. . {0]. {1]. {2]], and
A
2
= {. . {0]. {1. 2]]. Then A
1
and A
2
are semirings (in fact, A
2
is a eld), but
their intersection A = A
1
A
2
= {. . {0]] is not a semiring as {0] = {1. 2]
is not a disjoint union of sets in A.
Generally, let A
1
and A
2
be two semirings, and A
1
and A
2
. Then
A
1
A
2
, and which means that the complement of every element in A
1
A
2
should be expressed as nite union of disjoint sets in A
1
A
2
. As we have seen
in the example, this is a demanding requirement.
Of course, there is no pre-requirement that should be in a semiring. See
the next Exercise 1.19. |L
I Exercise 1.19 (1.1.17). If A is a semiring over , must A?
Solution. Not necessarily. In face, the simplest example of a semiring (a ring,
a o-ring) is just {]. |L
I Exercise 1.20 (1.1.18). Let A denote a semiring. Pick n N, and let
.
1
. . . . .
n
A. Then there exists a nite collection {C
1
. . . . . C
n
] of disjoint
A-sets with
_
n
i=1

i
=
_
n
}=1
C
}
.
Proof. When n = 1, write
1
= (
1
) and invoke (SR3). Now assume
that the result is true for n N. Consider n 1.

n1
_
i=1

i
=
_
_

n
_
i=1

i
_
_

n1
=
_
_
n
_
}=1
C
}
_
_

n1
=
n
_
}=1
_
C
}

n1
_
.
Now for each , there exists disjoint sets {D
}
1
. . . . . D
}
k
j
] _ A such that
C
}

n1
=
k
j
_
i=1
D
}
i
.
Then {D
}
i
: = 1. . . . . m. r = 1. . . . . m
}
] is a nite pairwise disjoint subset of A,
and

n1
_
i=1

i
=
n
_
}=1
n
j
_
i=1
D
}
i
. |L
SECTION 1.1 t-SYSTEMS, 2-SYSTEMS, AND SEMIRINGS 9
I Exercise 1.21 (1.1.19). Other books deal with a system called a ring. We will
not deal with rings of sets in this text, but since the reader might refer to other
books that deal with rings, it is worthy to discuss the concept. A collection R of
subsets of a nonempty set is called a ring of subsets of i
(R1) R = ,
(R2) . T R implies L T R, and
(R3) . T R implies T R.
That is, a ring is a nonempty collection of subsets closed under unions and
dierences.
a. is in every ring.
b. R is a ring i R satises (R1), (R2), and
(R4) . T R with _ T implies T R.
c. Every ring satises
(R5) . T R implies ^T R.
d. Every ring is a -system.
e. Every ring is closed under nite unions and nite intersections.
f. R is a ring i R a nonempty -system that satises (R4) along with
(R6) . T R and T = imply L T R.
g. R is a ring i R is a nonempty -system that satises (R5).
h. Suppose that {R

: ] is the exhaustive list of all rings that contain A.


Then
_

is a ring that contains A, and


_

is contained in any
ring that contains A. The minimal ring containing A is always exists and is
unique.
i. The collection of nite unions of rsc intervals is a ring on R.
j. Let be uncountable. The collection of all amc subsets of is a ring on .
Proof. (a) By (R1), there exists some set R, it follows from (R3) that
= R.
(b) We need only to prove that (R3) ==(R4) under (R1) and (R2).
v (R3) == (R4) is obvious.
v (R4) == (R3): Let . T R, and note that T = (T L ) R since
_ L T, and T L R by (R2).
10 CHAPTER 1 SET SYSTEMS
(c) Let . T R. By (R3), T R, and T R; by (R2), ( T) L(T )
R. Observe that ^T = ( T) L (T ), and we complete the proof.
(d) Let . T R. It is clear that T = ( L T) (^T). Note that LT R
[by (R2)], ^T R [by (c)], and ( L T) (^T) R [by (R3)]. Therefore,
T R and R is a -system.
(e) Just follows (R2) and (d).
(f) To see the only if part, suppose R is a ring. Then (d) means that R is a
nonempty -system, (R3) ==(R4) [by part (b)], and (R2) ==(R6) [by denition].
Now we prove the if part. Note that
(R1) By assumption;
(R2) Let . T R. We can write L T as
L T = ( T) L (T ) L ( T)
=
_
( T)
_
L
_
T ( T)
_
L ( T) .
Now (R4) implies that
_
( T)
_
R, and
_
T ( T)
_
R; (R6) implies
that L T R.
2
(R3) Let . T R. Then T = L ( T) = (
c
) L ( T
c
) =
(
c
L T
c
) = ( T)
c
= ( T). Clearly, T _ , so (R4) implies
that T R.
(g) To see the only if part, suppose that R is a ring. Then (R1) and (d) implies
R is a nonempty -system, and we have (R5) by (c).
For the inverse direction, suppose that R satises the given assumptions.
(R1) R = by assumption;
(R2) Let . T R. Then L T = (^T) L ( T) = (^T) ^( T). Since
R is a -system, T R. Thus, (R5) implies (R2).
(R3) Let . T R. Note that T = (^T) . Then (R5) implies that
^T R, and (^T) R since R is a -system.
3
(h) Similar to Exercise 1.8 and Exercise 11.2.
(i) See Exercise 11.5 (p.147).
(j) (R1) is trivial. (R2) holds because every nite (in fact, countable) union
of amc sets is amc (see, e.g., Rudin 1976). To see (R3), let . T be amc. Since
T = ( T) _ , and T _ , we know that T is amc. |L
IExercise 1.22 (1.1.20). This problemexplores the relationship between semir-
ings and rings.
2
For ~B = ~(B), see part (g) of this exercise
3
Vestrup (2003, p.6) hints that ~B = Z(B).
SECTION 1.1 t-SYSTEMS, 2-SYSTEMS, AND SEMIRINGS 11
a. Every ring is a semiring. However, not every semiring is a ring.
b. Let A denote a semiring on , and let R consist of the nite disjoint unions
of A-sets. Then R is closed under nite intersections and disjoint unions.
c. If . T A and _ T, then T R.
d. A, T R, and _ T imply T R.
e. . T R and _ T imply T R.
f. R is the minimal ring generated by A.
g. A semiring that satises (R2) is a ring.
Proof. (a) Let R be a ring. Then (R1) [R = ] and (R3) [R is closed under
dierences] imply that there exists R such that = R. Thus, (SR1)
is satised. To see that R satises (SR2) [R is a -system], refer Exercise 1.21
(d). Finally, (R4) [Exercise 1.21 (b)] implies (SR3).
To see a semiring is not necessary a ring, note that the collection :
{. (a. b| [ a. b R] is a semiring, but is not a ring: let o < a < b < c < J <
o, then (a. b| L (c. J| .
Note that a semiring is a ring if for any . T we have L T
[Figure 1.1 (p.1), and part (g) of this exercise]. Any semiring generates a ring as
in the Claim in Exercise 11.5 (p.147).
(b) Let A be a semiring on , and let
R :
_
_
_
n
_
i=1

i
:
i
A and n N
_
_
_
.
To prove R is closed under nite intersections, let =
_
n
}=1

}
, and T =
_
n
k=1
T
k
, where the
}
s are disjoint and in A, as are the T
k
s. Then
T =
_
_
n
_
}=1

}
_
_

_
_
n
_
k=1
T
k
_
_
=
n
_
}=1
n
_
k=1
_

}
T
k
_
=
_
16j6m
16k6n
_

}
T
k
_ (1)
R.
where (1) holds because the
_

}
T
k
_
s are disjoint and in A [by (SR2)]. Since
the intersection of any two sets in R is in R, it follows by induction that so is
the intersection of nitely many sets in R.
A disjoint union of nitely many sets in R is clearly in R.
(c) Let . T A and _ T. Then by (SR3), there exists disjoint C
1
. . . . . C
k
A
with T =
_
k
i=1
C
i
. Thus, T R by denition.
12 CHAPTER 1 SET SYSTEMS
(d) Let A, T R, and _ T. Then,
T
(2)
=
_
_
n
_
i=1

i
_
_
=
n
_
i=1
(
i
) =
n
_
i=1
_

i
(
i
)
_ (3)
R.
where (2) follows the fact that T R [the
i
s are in A and disjoint], and
(3) follows part (c) in this problem [note that
i
A. A, and by (SR2),

i
A].
(e) Let . T R and _ T. Then
T =
_
_
n
_
k=1
T
k
_
_

_
_
n
_
}=1

}
_
_
=
n
_
k=1
_
_
_T
k

_
_
n
_
}=1

}
_
_
_

_ =
n
_
k=1
_
_
n
_
}=1
_
T
k

}
_
_
_
.
Note that T
k
.
}
A, then T
k

}
A [(SR2)], and by part (c),
T
k

}
= T
k

_
T
k

}
_
R.
Furthermore, by part (b),
_
n
}=1
_
T
k

}
_
R, and so T R.
(f) Let m(A) be the class of rings containing A, and let C m(A). By denition,
if R, then =
_
n
i=1

i
, where {
i
]
n
i=1
_ A are disjoint. Then C since
C is a ring containing A. Hence, R is the minimal ring containing A.
(g) Let A be a semiring satisfying (R2) [. T A == L T A]. Then A
is nonempty since A by denition of a semiring. By (R2), A is
_
-stable;
hence, to prove A is a ring, we need only to prove that A is closed under
dierence. Let . T A. Then
T = ( T)
1
=
k
_
i=1
C
i
A.
where {C
i
]
k
i=1
_ A are disjoint, and equality (1) follows (SR3). |L
I Exercise 1.23 (1.1.21). Let be innite, and let A _ 2

have cardinality
0
.
We will show that the ring generated by A has cardinality
0
.
a. Given C _ 2

, let C
+
denote the collection of all nite unions of dierences of
C-sets. If card(C) =
0
, then card(C
+
) =
0
. Also, C implies C _ C
+
.
b. Let A
0
= A, and dene A
n
= A
+
n-1
for n > 1. Then A _
_
o
n=0
A
n
_ m(A),
where m(A) is the minimal ring generated by A and where [without loss of
generality] A. Also, card(
_
o
n=0
A
n
) =
0
.
c.
_
o
n=0
A
n
is a ring on , and from the fact that m(A) is the minimal ring
containing A, we have
_
o
n=0
A
n
= m(A), and thus card(m(A)) =
0
.
d. We may generalize: if A is innite, then card(A) = card(m(A)).
SECTION 1.1 t-SYSTEMS, 2-SYSTEMS, AND SEMIRINGS 13
Proof. (a) Let C
t
:{C
i
C
}
: C
i
. C
}
C]. Since card(C) =
0
[C is countable],
we can write C as
C = {C
n
]
o
n=1
.
We now show that card(C
t
) =
0
. Notice that for any C
n
C, we can construct
a bijection on N onto C
n
C :{C
n
C
i
: C
i
C] as follows

C
n
(i ) = C
n
C
i
.
but which means that C
n
C is countable. Then,
C
t
=
_
C
n
C
C
n
C|
is a countable union of countable sets, so it is countable [under the Axiom of
Choice, see (Hrbacek and Jech, 1999, Corollary 3.6, p. 75)].
Now we show that for any n N, the set C
+
n
dened by
C
+
n
=
_
_
_
n
_
i=1
C
t
i
: C
t
i
C
t
and C
t
i
= C
t
}
whenever i =
_
_
_
is countable. We prove this claimwith the Induction Principle on n N. Clearly,
this claim holds with n = 1 since in this case, C
+
1
= C
t
. Assume that it is true
for some n N. We need to prove the case of n 1. However,
C
+
n1
= C
+
n
L

C
t
.
where

C
t
:

C
t
C
t
: C
t
= C
t
i
V i 6 n
_
.
Because C
t
is countable, we conclude that

C
t
_ C
t
is amc. Therefore, C
+
n1
is
countable. Hence, by the Induction Principle, card(C
+
n
) =
0
for any n N, and
C
+
=
_
nN
C
+
n
(1.3)
is countable.
We now show that if C, then C _ C
+
. Let C C, then C C
t
because
C = C ; therefore,
C _ C
t
_ C
+
.
[Remember that C
t
= C
+
1
and (11.1).]
(b) By the denition of A
n
, we know A
1
= A
+
, the collection of all nite unions
of dierences of Asets. Since A, we know from part (a) that A _ A
+
= A
1
;
therefore,
A _
o
_
n=0
A
n
. (1.4)
14 CHAPTER 1 SET SYSTEMS
We are nowready to prove that
_
o
n=0
_ m(A). We use the Induction Principle
to prove that
A
i
_ m(A). V i N. (Pi )
Clearly, P0 holds as A
0
= A _ m(A). Now assume Pn holds. We need to prove
Pn 1. Notice that A
n1
= A
+
, the collection of all nite unions of dierences
of A
n
-sets, we can write a generic element of A
n1
as

n1
=
n
_
}=1

t
}
.
where
t
}
=
t
n

tt
n
, and
t
n
.
tt
n
A
n
. Since A
n
_ m(A) by Pn, we know that

}
=
t
n

tt
n
m(A) by (R3); therefore,
n1
=
_
n
}=1

t
}
m(A) by (R2).
This proves Pn 1. Then, by the Induction Principle, we know that A
n
_ m(A),
V n N; therefore,
o
_
n=0
A
n
_ m(A). (1.5)
Combine (11.2) and (1.5) we have
A _
o
_
n=0
A
n
_ m(A). (1.6)
To prove card(
_
o
n=0
A
n
) =
0
, we rst use the Induction Principle again to
prove that A
n
is countable, V n N. Clearly, A
1
= A
+
is countable by part (a).
Assume A
n
is countable, then A
n1
= A
+
is countable by part (a) once again.
Therefore,
_
o
n=0
A
n
is countable [under the Axiom of Choice].
(c) Clearly,
_
o
n=0
A
n
:

A = , so (R1) is satised. To see (R2) and (R3), let
. T

A. Then there exist m. n N such that A
n
and T A
n
. We have
shown in part (a) that
A
n1
= A
+
n
_ A
n
[along with the Induction Principle]. Therefore, either A
n
_ A
n
[if m 6 n] or
A
n
_ A
n
[if n 6 m]. Without loss of generality, we assume that m 6 n, i.e.,
A
n
_ A
n
; therefore, A
n
== A
n
. Therefore, . T A
n
implies that
L T = ( ) L (T ) A
+
n
= A
n1
_

A.
[this proves (R2)], and
T =
_
_
n
A
_
i=1

i
_
_

_
_
n
B
_
}=1
T
}
_
_
=
n
A
n
B
_
i=1
_

i
T
}
_
A
+
n
_
o
_
n=1
A
n
.
[this proves (R3)]. Hence,

A is a ring, and

A = m(A); furthermore, we have
card(m(A)) =
0
.
(d) Straightforward. |L
SECTION 1.2 FIELDS 15
1.2 Fields
I Exercise 1.24 (1.2.1). The collection F = { _ : is nite or
c
is nite]
is a eld on .
Proof. F because
c
= is nite; let F . If is nite,
c
F as
(
c
)
c
= is nite; if
c
is nite
c
F . Thus, F is closed under complements.
Finally, let . T F . There are two cases: (i) both and T are nite, then LT
is nite, whence L T F ; (ii) at least one of
c
or T
c
is nite. Assume that
T
c
is. We have ( L T)
c
=
c
T
c
_ T
c
, and thus ( L T)
c
is nite, so that
gain L T F . |L
I Exercise 1.25 (1.2.2). Let F _ 2

be such that F and T F


whenever . T F . Then F is a eld on .
Proof. We need to check F satises (F1)(F3). F by assumption. Let
= and T F . Then T
c
= T F . Let . T F . Then
c
. T
c
F . Since
( L T)
c
=
c
T
c
=
c
T F , we must have L T =
_
( L T)
c
_
c
F . |L
I Exercise 1.26 (1.2.3). Every z-system that is closed under arbitrary dier-
ences is a eld.
Proof. We only need to show that it is closed under nite unions, and it comes
from the previous exercise. |L
I Exercise 1.27 (1.2.4). Let F _ 2

satisfy (F1) and (F2), and suppose that F


is closed under nite disjoint unions. Then F is not necessarily a eld.
Solution. For example, let = {1. 2. 3. 4], and
F =

. . {1. 2] . {3. 4] . {2. 3] . {1. 4]


_
.
F satises all the requirements, but which is not a eld since, for example,
{1. 2] L {2. 3] = {1. 2. 3] F . |L
I Exercise 1.28 (1.2.5). Suppose that F
1
_ F
2
_ F
3
_ , where F
n
is a eld
on for each n N. Then
_
o
n=1
F
n
is a eld on .
Proof. (F1) F
n
, for each n N, so
_
o
n=1
F
n
; [Of course, it is enough
to check that F
n
for some F
n
.] (F2) Suppose
_
o
n=1
F
n
. Then there exist
n N such that F
n
. So
c
F
n
==
c

_
o
n=1
F
n
; (F3) Let . T
_
o
n=1
F
n
.
Then J m N such that F
n
, and J n N such that T F
n
. Hence,
L T F
n
L F
n
_
_
o
n=1
F
n
. |L
IExercise 1.29 (1.2.6). The collection consisting of R
k
, , and all k-dimensional
rectangles of all forms fails to be a eld on R
k
.
16 CHAPTER 1 SET SYSTEMS
Solution. Consider k = 1 and a. b|, where a. b R. Then a. b|
c
= (o. a) L
(b. o) is not a interval.
The k > 2 case can be generalized easily. For example, let
A =
k

i=1
a
i
. b
i
| .
Then A
c
is not a rectangle. |L
I Exercise 1.30 (1.2.7). The collection consisting of and the nite disjoint
unions of k-dimensional rsc subrectangles of the given k-dimensional rsc rect-
angle (a. b| is a eld on .
Proof. A more general proposition can be found in Folland (1999, Proposi-
tion 1.7). Denote the set system given in the problem as , a semiring, and
the collection of and the nite disjoint unions of k-dimensional rsc sub-
rectangles as A First =
_
i
1
i
by denition, where 1
i
. If . T
and T
c
=
_
n
i=1
C
i
, where C
i
. Then T =
_
n
i=1
( C
i
) and L T =
( T) LT. Hence T A and LT A. It now follows by induction that
if
1
. . . . .
n
, then
_
n
i=1

i
A. It is easy to see that A is closed under
complements. |L
I Exercise 1.31 (1.2.8). An arbitrary intersection of elds on is a eld on .
Proof. Let {F

: ] be a set of elds on , where is some arbitrary set


of indexes. Then
(F1)
_

since F

for any .
(F2) Let T
_

, then
c
F

, for any ; hence


c

.
(F3) Let T. C
_

. Then T. C F

, V . Hence, TLC F

, V ,
and T C
_

.
|L
I Exercise 1.32 (1.2.9). Let be arbitrary, and let A _ 2

. There exists a
unique eld F on with the properties that (i) A _ F , and (ii) if G is a eld with
A _ G, then F _ G. This eld F is called the [minimal] eld [on ] generated
by A.
Proof. Let {F

: ] be the exhaustive set of elds on containing A. Then


_

is the desired eld. |L


I Exercise 1.33 (1.2.10). Let
1
. . . . .
n
be disjoint. What does a typical
element in the minimal eld generated by {
1
. . . . .
n
] look like?
Solution. Refer to Ash and Dolans-Dade (2000, Exercise 1.2.8). To save nota-
tion, let F denote the minimal eld generated by A :{
1
. . . . .
n
]. We consider
an element of F {. ]. We can write a typical element T F as follows,
SECTION 1.2 FIELDS 17
T = T
1
+ T
2
+ + T
n
.
where + is an set operation either L or , and T
i

1
. . . . .
n
.
c
1
. . . . .
c
n
_
for
each i {1. . . . . m]. |L
I Exercise 1.34 (1.2.11). Let S be nite, and denote the set of sequences of
elements of S. For each o , write
o =
_
:
1
(o) . :
2
(o) . . . .
_
.
so that :
k
(o) denotes the k-th term of o for all k N. For n N and H _ S
n
,
let
C
n
(H) :{o [ :
1
(o) . . . . . :
n
(o) H] .
Let
F :

C
n
(H)

n N. H _ S
n
_
.
Then F is a eld of subsets of S
o
. [The sets C
n
(H) are called cylinders of rank
n, and F is collection of all cylinders of all ranks.]
Proof. See Remark 1.2 (p.2) for more details about Cylinders. To prove F is a
eld, note that
(F1) F . Consider C
o
(S
o
); then o C
o
(S
o
), V o , which means
_ C
o
(S
o
). Hence,
= C
o
_
S
o
_
F .
(F2) To prove that F is closed under complements, consider any C
n
(H) F .
By denition,
C
n
(H) :{o [ :
1
(o) . . . . . :
n
(o) H] .
Then,
_
C
n
(H)
_
c
=

o : :
1
(o). . . . . :
n
(o)| H
_
=

o : :
1
(o). . . . . :
n
(o)| H
c
_
= C
n
_
H
c
_
F .
(-system) Finally, we need to prove F is closed under nite intersections.
4
4
It is hard to prove that F is closed under nite unions. See below for my rst but failed
try.
(Wrong!) Let C
m
(G) , C
n
(1) F, where n, n N and G _ S
m
, 1 _ S
n
. By denition,
C
m
(G) LC
n
(1) =
_
o

_
z
1
(o), . . . , z
m
(o)
_
G
_
_

z
1
(o), . . . , z
n
(o)j 1
_
1
=
_
o

_
z
1
(o), . . . , z
m^n
(o)
_
(1 LG)
_
2
= C
m^n
(G
m^n
L1
m^n
)
F,
18 CHAPTER 1 SET SYSTEMS
Consider two cylinders, C
n
(G) and C
n
(H), where m. n N, G _ S
n
, and
H _ S
n
. We need to prove that C
n
(G) C
n
(H) F . In fact,
C
n
(G) C
n
(H) = C
nn
_
(G
n.n
H
n.n
)
_
G
n-(n.n)
L H
n-(n.n)
_
_
F .
where, for example, G
m^n
in equality (2), G
m^n
_ S
m^n
, G
m.m^n/
_ S
m.m^n/
, and
G
m^n
G
m.m^n/
= G.
To see why equality (1) holds, we need the following facts:
Claim 1. Suppose that n 6 n, 1 = G 1
nm
, and G _ S
m
. Then C
m
(G) _ C
n
(1).
Proof. Pick any o
0
C
n
(1). By denition,
_
z
1
_
o
0
_
, . . . , z
n
_
o
0
_
_
1 = G 1
nm
,
which means that
_
z
1
_
o
0
_
, . . . , z
m
_
o
0
_
_
G ==o
0
C
m
(G).
Claim 2. If G _ 1 _ S
n
, then C
n
(G) _ C
n
(1).
Proof. Straightforward.
Claim 3. For any n, n N, and G _ S
m
, 1 _ S
n
, we have
C
m
(G) LC
n
(1) _ C
m^n
(G
m^n
L1
m^n
).
Proof. Without loss of any generality, we assume that n.n = n. Pick any o
0
C
m
(G) L
C
n
(1). Then,
_
z
1
(o
0
), . . . , z
m
_
o
0
_
_
G, or
_
z
1
(o
0
), . . . , z
n
_
o
0
_
_
1.
From Claim 2, we have
_
z
1
(o
0
), . . . , z
m
_
o
0
_
_
G L1
m
, or
_
z
1
(o
0
), . . . , z
n
_
o
0
_
_
(G L1
m
) 1
nm
,
where 1
m
_ S
m
, and 1 = 1
m
1
nm
. Then, by Claim 1, if n.n = n, we have
o
0
C
m
(G L1
m
) .
Claim 4. For any n, n N, and G _ S
m
, 1 _ S
n
, we have
C
m
(G) LC
n
(1) _ C
m^n
(G
m^n
L1
m^n
).
Proof. We still assume that n . n = n. Pick any o
0
C
m^n
(G
m^n
L1
m^n
) =
C
m
(G L1
m
). By denition,
_
z
1
_
o
0
_
, . . . , z
m
_
o
0
_
_
G L1
m
;
that is,
_
z
1
_
o
0
_
, . . . , z
m
_
o
0
_
_
G or
_
z
1
_
o
0
_
, . . . , z
m
_
o
0
_
_
1
m
. lL
SECTION 1.2 FIELDS 19
where G
n.n
. H
n.n
_ S
n.n
, G
n-(n.n)
_ S
n-(n.n)
, H
n-(n.n)
_ S
n-(n.n)
, G =
G
n.n
G
n-(n.n)
, H = H
n.n
H
n-(n.n)
, and we dene G
0
= H
0
= . |L
I Exercise 1.35 (1.2.12). Suppose that A is a semiring on with A.
The collection of nite disjoint unions of A-sets is a eld on . [Compare with
Example 3 and Exercise 1.30.]
Proof. Let A be a semiring, and A. Let F be the collection of nite
disjoint unions of A-sets, tha is, F i for some n Nwe have =
_
n
i=1

i
,
where
i
s are disjoint A-sets. F is a eld: (i) F since = L F .
(ii) Let F . Then =
_
n
i=1

i
, where n N and {
i
]
n
i=1
_ A. To prove

c
F , we need only to prove
c
i
F since
c
=
_
n
i=1

c
i
, and A is a semiring
[
_
-stable]. But
c
i
F is directly from (SR3) and the fact that A since

c
i
=
i
=
_
n
i
}=1
C
i
}
, where
_
C
i
}
_
n
i
}=1
_ A is disjoint, and n
i
N, V i
{1. . . . . n], that is, each
c
i
is a nite disjoint union of A-sets. Thus, F is closed
under complements.
Instead of proving that F satises (F3) directly, we prove that F is a -
system. Let T
1
. T
2
F . Then
T
1
T
2
=
_
_
n
_
i=1

i
_
_

_
_
k
_
}=1

}
_
_
=
n
_
i=1
_
_
k
_
}=1
_

i

}
_
_
_
=
_
i,}
_

i

}
_
.
Note that
i

}
A by (SR2). Hence T
1
T
2
F .
|L
I Exercise 1.36 (1.2.13). Let :
t
. Given A
t
_ 2

0
, let
-1
(A
t
) =
{
-1
(
t
) :
t
A
t
], where
-1
(
t
) is the usual inverse image of
t
under .
a. If A
t
is a eld on
t
, then
-1
(A
t
) is a eld on .
b. (A) may not be a eld over
t
even if A is a eld on .
Proof. (a) Let A
t
be a eld on
t
. (i) Since =
-1
(
t
) and
t
A
t
, we have
that
-1
(A
t
). (ii) If
-1
(A
t
), then =
-1
(
t
) for some
t
A
t
.
Therefore,
c
=
-1
(
t
)|
c
=
-1
((
t
)
c
), and (
t
)
c
A
t
since A
t
is a eld.
It follows that
c

-1
(A
t
), so that
-1
(A
t
) is closed under complements.
(iii) To see that
-1
(A
t
) is closed under nite unions, let {
i
]
n
i=1
_ A, where
n N. Therefore, for each i {1. . . . . n], there is
t
i
A
t
with
i
=
-1
(
i
).
Therefore,
n
_
i=1

i
=
n
_
i=1

-1
_

t
i
_
=
-1
_
_
n
_
i=1

t
i
_
_

-1
(A
t
).
since
_
n
i=1

t
i
A
t
.
20 CHAPTER 1 SET SYSTEMS
(b) The simplest case is that is not onto [surjective]. In this case, ()
t
;
that is,
t
A
t
, and so A
t
is not a eld on
t
. |L
I Exercise 1.37 (1.2.14). Let be innite, and let A _ 2

have cardinality
0
.
Let (A) denote the minimal eld generated by A [Exercise 1.32]. We will show
that card((A)) =
0
.
a. Given a collection C, let C
+
denote the collection of
i. nite unions of C-sets,
ii. nite unions of dierences of C-sets, and
iii. nite unions of complements of C-sets.
If C, then C _ C
+
. If card(C) =
0
, then card(C
+
) =
0
.
Proof. |L
I Exercise 1.38 (1.2.15). Some books work with a system of sets called an
algebra. An algebra on is a nonempty collection of subsets of that satises
(F2) and (F3).
a. F is an algebra on i F is a ring on with F .
b. F is an algebra i F is a eld. Thus algebra and eld are synonymous.
Proof.
(a: H)) Suppose F is an algebra. Then,
(R1) F = by assumption.
(R2) F is
_
-stable follows (F3).
(R3) The assumption of F and (F2) imply that if . T F , then
c
=
F and T
c
= T F . Then
_
_

c
L T
_ (F3)
F
_
(F2)
==
_
_

c
L T
_
c
F
_
== T F | .
This proves that F is closed under dierence.
(a: (H) Suppose F is a ring and F . To prove F is an algebra on , note
that
(A1) F = since F is a ring.
(F2) Let F . Because F and (R3), we have
c
= F . This
proves that F is closed under dierence.
(F3)
_
-stability follows (R2).
SECTION 1.3 c-FIELDS 21
(b) We need only to prove that F is an eld if F is an algebra since the reverse
direction is trivial.
Suppose F is an algebra. We want to show F . Since F = by denition
of an algebra, there must exist F . Then
c
F by (F2), and so = L
c

F by (F3). |L
1.3 -Fields
I Exercise 1.39 (1.3.1). A collection F of sets is called a monotone class i
(MC1) for every nondecreasing sequence {
n
]
o
n=1
of F -sets we have
_
o
n=1

n

F , and (MC2) for every nonincreasing sequence {
n
]
o
n=1
of F -sets we have
_
o
n=1

n
F .
a. If F is both a eld and a monotone class, then F is a o-eld.
b. A eld is a monotone class if and only if it is a o-eld.
Proof. See Chung (2001, Theorem 2.1.1).
a. Let F is both a eld and a MC. Let {
n
]
o
n=1
_ F , then T
n
=
_
n
i=1

n
F
since F is a eld, T
n
_ T
n1
, and
_
o
n=1

n
=
_
o
n=1
T
n
F .
b. We only need to show the IF part. But it is trivial: A o-led is a eld and a
MC.
|L
I Exercise 1.40 (1.3.2). This problem discusses some equivalent formulations
of a o-eld.
a. F satises (S1), (S2), and closure under amc intersections i F is a o-eld.
b. Every eld that is closed under countable disjoint unions is a o-eld.
c. If F satises (S1), closure under dierences, and closure under countable
unions or closure under countable intersections, then F is a o-eld.
Proof. (a) For the ONLY IF part, let {
n
]
o
n=1
_ F and F satisfy (S1) and (S2).
Then
c
n
F for any n N; hence,
_
o
n=1

n
=
__
o
n=1

c
n
_
c
F . The IF part
is proved by the same logic.
(b) We need only to prove F is closed under countable unitions. Let F be a
eld, and {
n
]
o
n=1
_ F . Let
T
k
=
k

_
_
n-1
_
i=1

i
_
_
c
.
It is clear that {T
n
]
o
n=1
_ F is disjoint, and
_
o
n=1
T
n
=
_
o
n=1

n
. This completes
the proof.
22 CHAPTER 1 SET SYSTEMS
(c) We only need to prove (S2), that is, F is closed under complementation. Let
F . By (S1), F , then
c
= F since by assumption, F is closed
under dierence. |L
I Exercise 1.41 (1.3.3). Prove the following claims.
a. A nite union of o-elds on is not necessarily a eld on .
b. If a nite union of o-elds on is a eld, then it is a o-eld as well.
c. Given o-elds F
1
F
2
on , it is not necessarily the case that
_
o
n=1
F
n
is a o-eld.
Proof. (a) Let {F
i
]
n
i=1
be a class of o-elds, and consider
_
n
i=1

i
, where
i

F
i
. Note that it is possible that
_
n
i=1

i
F
}
for any , so
_
n
i=1

i

_
n
i=1
F
i
.
For example (Athreya and Lahiri, 2006, Exercise 1.5, p.32), let
= {1. 2. 3] . F
1
=

{1] . {2. 3] . .
_
. F
2

{1. 2] . {3] . .
_
.
It is easy to verify that F
1
and F
2
are both o-elds, but F
1
L F
2
is not a eld
since {1] L {3] = {1. 3] F
1
L F
2
.
(b) Without loss of any generality, we here just consider two o-elds, F
1
and
F
2
, on . Consider a sequence {
n
]
o
n=1
_ F
1
L F
2
. Then we can construct two
sequences, one in F
1
and one in F
2
. Particularly, the sequence of sets

1
n
_
_ F
1
is constructed as follows:

1
n
=
_
_
_

n
. if
n
F
1
. otherwise.
The sequence of sets

2
n
_
_ F
2
is constructed similarly. Then
_
o
k=1

1
k
F
1
and
_
o
n=1

2
n
F
2
since both F
1
and F
2
are o-elds, and
o
_
n=1

n
=
_
_
o
_
n=1

1
n
_
_
L
_
_
o
_
n=1

2
n
_
_
.
If F
1
L F
2
is a eld, we have
o
_
n=1

n
=
_
_
o
_
n=1

1
n
_
_
L
_
_
o
_
n=1

2
n
_
_
F
1
L F
2
.
(c) See Broughton and Hu (1977) for a more general result. Let = N and
for all n N, let
F
n
= o
_

{1] . . . . . {n]
_
_
.
Since

{1] . . . . . {m]
_


{1] . . . . . {n]
_
when m < n, we have F
1
F
2
. It it
clear that {1] . {2] . . . .
_
o
n=1
F
n
, but
SECTION 1.3 c-FIELDS 23
o
_
n=1
{n] = {1. 2. . . .]
o
_
n=1
F
n
.
since there does not exist a F
n
such that {1. 2. . . .] F
n
, for any n N. |L
I Exercise 1.42 (1.3.5). A subset _ R is called nowhere dense i every open
interval 1 contains an open interval J such that J = . Clearly and all
subsets of a nowhere dense set are nowhere dense. A subset _ R is called a
set of the rst category i is a countable union of nowhere dense sets.
a. An amc union of sets of the rst category is of the rst category.
b. Let F =

_ R : or
c
is a set of the rst category
_
. Then F is a o-eld of
subsets of R.
Proof. Refer Gamelin and Greene (1999, Section 1.2) for the more detailed
denitions and discussion of nowhere dense and the rst category set.
(a) Consider a countable sequence of sets of the rst category, {
n
]
o
n=1
. Then

n
=
_
o
i=1

n
i
for any n N, where

n
i
_
o
i=1
are nowhere dense. Clearly, the
amc unions of amc unions is still amc, which proves the claim.
(b) Let F =

_ R : or
c
is a set of the rst category
_
. Then F since
is of the rst category and =
c
. To see F is closed under complemen-
tation, let F . (i) If is of the rst category, then
c
F since (
c
)
c
=
is of the rst category; (ii) If
c
is of the rst category, then
c
F by the
denition of F . In any case, F implies that
c
F .
Finally, to see F is o-
_
-stable, let {
n
]
o
n=1
be a sequence of F -sets. There
are two cases: (i) Each
n
is of the rst category. Then part (a) of this exercise
implies that
_
o
n=1

n
F . (ii) Some
c
n
is of the rst category. In this case,
we assume without loss of generality that
c
1
is of the rst category, and we
have that
__
o
n=1

n
_
c
=
_
o
n=1

c
n
_
c
1
. It is trivial that
__
o
n=1

n
_
c
is of the
rst category since
c
1
is, and every subset of the rst category is of the rst
category. Particularly, let
c
1
=
_
o
n=1
T
n
, where the T
n
s are nowhere dense
sets. Since
__
o
n=1

n
_
c
_
c
1
, we must can rewrite
__
o
n=1

n
_
c
as
_
_
o
_
n=1

n
_
_
c
=
o
_
n=1
C
n
.
where every C
n
is a subset of T
n
and some C
n
s maybe be empty. Note that then
every C
n
is nowhere dense no matter C
n
= or not. Consequently,
__
o
n=1

n
_
c
is of the rst category by denition. |L
I Exercise 1.43 (1.3.6). A o-ring of subsets of is a nonempty collection of
subsets of that is closed under dierences as well as countable unions.
a. Every o-ring is closed under nite unions and amc intersections.
24 CHAPTER 1 SET SYSTEMS
b. F is a o-eld i F is a o-ring with F .
c. State and prove an existence and uniqueness result regarding the [minimal]
o-ring generated by a collection A of subsets of .
Proof. (a) Let R be a o-ring. We rst prove that R. Since R = , there
exists R; moreover, since R is closed under dierence, we have =
R. Now consider an arbitrary sequence of R-sets
1
. . . . .
n
. . . . . ..
Because R is o-
_
-stable, we know that
n
_
i=1

i
= (
1
L
2
L L
n
) L (L L ) R.
which proves that R is
_
-stable.
To see R is closed under amc intersections, let {
n
]
o
n=1
_ R. Then =
_
o
n=1

n
R. Let

t
n
=
n
. V n N.
Then

t
n
_
o
n=1
_ R,
_
o
n=1

t
n
R, and

_
_
o
_
n=1

t
n
_
_
=
o
_
n=1

n
R
since
__
o
n=1

t
n
_
R. [Basically, I let be the universal space, and
t
n
be
the complements of
n
in .]
(b) Suppose that F is a o-eld. Then F be (S1). To see F is closed under
dierence, let . T F . Then (S2) implies that T
c
F . Since F is
_
-stable, we
have T = T
c
F . The fact that F is o-
_
-stable follows (S3).
Now suppose that F is a o-ring with F . We need only to prove that F
satises (S2). Let F . Since F and F is closed under dierence, we
have

c
= F .
(c) Standard. Omitted. |L
I Exercise 1.44 (1.3.9).
a. If A _ A
t
_ o(A), then o(A
t
) = o(A).
b. For any collection = A _ 2

, (A) _ z(A) _ o(A).


c. If the nonempty collection A is nite, then o(A) = (A).
d. For arbitrary collection A, we have o(A) = o
_
(A)
_
.
e. For arbitrary collection A, we have
_
o(A)
_
= o
_
(A)
_
.
Proof. (a) On the rst hand, A _ A
t
implies that o(A) _ o(A
t
); on the sec-
ond hand, A
t
_ o(A) implies that o(A
t
) _ o
_
o(A)
_
= o(A). We thus get the
equality.
SECTION 1.3 c-FIELDS 25
(b) Let (A), (A), and (A) denote the collection of all -systems, z-
systems, and o-elds of subsets of that contain A, respectively. With this,
we may dene
(A) =
_
P(A)
P. z(A) =
_
L/(A)
L. and o(A) =
_
F(A)
F .
It is easy to see that any o-eld containing A is a z-system containing A; hence
(A) _ (A), and so z(A) _ o(A).
(c) It is clear that (A) _ o(A); since 0 <[A[ < o, the eld (A) is nite and
so it is a o-eld. Then o(A) _ (A).
(d) On the rst hand, (A) _ o(A) implies that o
_
(A)
_
_ o
_
o(A)
_
= o(A).
On the second hand, A _ (A) and so o(A) _ o
_
(A)
_
.
(e) It follows from (d) that o
_
(A)
_
= o(A). By denition,
_
o(A)
_
is the
minimal eld containing o(A). But o(A) itself is a eld; hence
_
o(A)
_
=
o(A). |L
I Exercise 1.45 (1.3.16). Let F = o(A), where = A _ 2

. For each T F
there exists a countable subcollection A
B
_ A with T o (A
B
).
Proof. Let
B =

T F : JA
B
_ A such that A
B
is countable and T o (A
B
)
_
. (1.7)
It is clear that B _ F . For any T A, take A
B
= {T]; then A
B
= {T] is
countable and T o
_
{T]
_
= {. . T. T
c
]; hence A _ B. We now show that B
is a o-eld. Obviously, B since F and o
_
{]
_
= {. ]. If T B,
then T F and there exists a countable A
B
_ A such that T o (A
B
); but
which mean that T
c
F and T
c
o (A
B
), i.e., T
c
B. Similarly, it is easy to
see that B is closed under countable unions. Thus, B is a o-eld containing A,
and so F _ B. We thus proved that B = F and the get the result. |L
I Exercise 1.46 (1.3.18). Given = A _ 2

and = T _ , let A T =
{ T : A] and let o(A) T = { T : o(A)].
a. o(A) T is a o-eld on T.
b. Next, dene o
B
(A T) to be the minimal o-eld over T generated by the
class A T. Then o
B
(A T) = o(A) T.
Proof. This claim can be found in Ash and Dolans-Dade (2000, p. 5).
(a) T o(A) T as o(A). If C o(A) T, then C = T with o(A);
hence TC =
c
T o(A)T. To see that o(A)T is closed under countable
unions, let {C
n
]
o
n=1
_ o(A) T. Then each C
n
=
n
T with
n
o(A). Hence,
26 CHAPTER 1 SET SYSTEMS
o
_
n=1
C
n
=
o
_
n=1
(
n
T) =
_
_
o
_
n=1

n
_
_
T o(A) T.
(b) First, A _ o(A), hence AT _ o(A)T. Since o(A)T is a o-eld on T by
(a), we have o
B
(A T) _ o(A) T. To establish the reverse inclusion we must
show that T o
B
(A T) for all o(A). We use the good sets principle.
Let
G = { o(A) : T o
B
(A T)] .
We now show that G is a o-eld containing A. It is evident that G. If G,
then T o
B
(A T) and o(A); hence,
c
T = T( T) o
B
(A T)
implies that
c
G. To see G is closed under countable unions, let {
n
]
o
n=1
_ G
with
n
T o
B
(A T) for all n N. Then
_
_
o
_
n=1

n
_
_
T =
o
_
n=1
(
n
T) o
B
(A T) .
Since A _ G, we have o(A) _ G; hence o(A) = G: every set in o(A) is good. |L
I Exercise 1.47 (1.3.19). Suppose that A = {
1
.
2
. . . .] is a disjoint sequence
of subsets of with
_
o
n=1

n
= . Then each o(A)-set is the union of an at
most countable subcollection of
1
.
2
. . . ..
Proof. Let
C =

o(A) : is an at most countable union of A-sets


_
.
It is easy to see that C since =
_
o
n=1

n
. If C, then =
_
iJ

i
,
where J is at most countable. Hence
c
=
__
o
n=1

n
_

__
iJ

i
_
is an at most
countable union of A-sets, that is, C is closed under complements. It is also
easy to see that C is closed under countable unions and A _ C. Hence, C is a
o-eld and o(A) = C. |L
I Exercise 1.48 (1.3.20). Let P denote a -system on , and let L denote
a z-system on with P _ L. We will show that o (P) _ L. Let z(P) de-
note the z-system generated by P, and for each subset _ we dene
G

= {C _ : C z(P)].
Proof. See Vestrup (2003, Claim 1, p. 82). |L
I Exercise 1.49 (1.3.21). Let F denote a eld on , and let M denote a mono-
tone class on [See Exercise 1.39]. We will show that F _ M implies that
o (F ) _ M. Let m(F ) denote the minimal monotone class on generated by
F . That is, m(F ) is the intersection of all monotone classes on containing the
collection F .
a. To prove the claim, it is sucient to show that o (F ) _ m(F ).
SECTION 1.3 c-FIELDS 27
b. If m(F ) is a eld, then o (F ) _ m(F ).
c. m(F ).
d. Let G = { _ :
c
m(F )]. G is a monotone class on and m(F ) _ G.
e. m(F ) is indeed closed under complements.
f. Let G
1
=

_ : L T m(F ) for all T F


_
. Then G
1
is a monotone class
such that F _ G
1
and m(F ) _ G
1
.
g. Let G
2
=

T _ : L T m(F ) for all m(F )
_
. Then G
2
is a monotone
class such that F _ G
2
, and m(F ) _ G
2
.
h. m(F ) is closed under nie unions, and hence is a eld.
Proof. Halmos Monotone Class Theorem is proved in every textbook. See
Billingsley (1995, Theorem 3.4), Ash and Dolans-Dade (2000, Theorem 1.3.9),
or Chung (2001, Theorem 2.1.2), among others. The above outline is similar to
Chung (2001).
(a) By denition. In fact, o (F ) = m(F ).
(b) By Exercise 1.39: A eld is a o-eld i it is also an M.C. If m(F ) is a eld,
then it is a o-eld containing F ; hence, o (F ) _ m(F ).
(c) F _ m(F ).
(d) Let {
n
]
o
n=1
_ G be monotone; then

c
n
_
o
n=1
is also monotone. The DeMor-
gan identities
_
_
o
_
n=1

n
_
_
c
=
o
_
n=1

c
n
. and
_
_
o
_
n=1

n
_
_
c
=
o
_
n=1

c
n
show that G is a M.C. Since F is closed under complements and F _ m(F ), it
is clear that F _ G. Hence m(F ) _ G by the minimality of m(F ).
(e) By (d), m(F ) _ G, which means that for any m(F ), we have
c
m(F ).
Hence, m(F ) is closed under implementation.
(f) Let G
1
=

_ : L T m(F ) for all T F
_
. If {
n
]
o
n=1
_ G
1
is mono-
tone, then {
n
L T]
o
n=1
is also monotone. The identities
_
_
o
_
n=1

n
_
_
L T =
n
_
n=1
(
n
L T) . and
_
_
o
_
n=1

n
_
_
L T =
n
_
n=1
(
n
L T)
show that G
1
is a M.C. Since F is closed under nite unions and F _ m(F ), it
follows that F _ G
1
, and so m(F ) _ G
1
by the minimality of m(F ).
(g) As in (f) we can show G
2
is a M.C. By (f), m(F ) _ G
1
, which means that for
any m(F ) and T F we have L T m(F ). This in turn means that
F _ G
2
. Hence, m(F ) _ G
2
.
28 CHAPTER 1 SET SYSTEMS
(h) Since m(F ) _ G
2
, for any T m(F ) and m(F ), we have LT m(F );
that is, m(F ) is closed under nite unions. |L
1.4 The Borel -Field
I Exercise 1.50 (1.4.1). Show directly that
5
o(A
3
) = o(A
+
3
), o(A
4
) = o(A
T
),
and o(A
+
4
) = o(A
10
).
Proof. (i) It is clear that o
_
A
+
3
_
_ o (A
3
). We only need to show that o (A
3
) _
o
_
A
+
3
_
. Since .. o) =
_
r
n
. o), where {r
n
]
o
n=1
_ Q, we complete the proof. |L
IExercise 1.51 (1.4.2). All amc subsets of R are Borel sets. All subsets of R that
dier from a Borel set by at most countably many points are Borel sets. That is,
if the symmetric dierence C^T is amc and T B, then C B.
Proof. Let = {.
n
]
o
n=1
_ R. Then =
_
o
n=1
{.
n
]. {.
n
] is a Borel set. |L
I Exercise 1.52 (1.4.3). The Borel o-eld on (0. 1| is denoted by B
(0,1j
and is
dened as the o-eld on (0. 1| generated by the rsc subintervals of (0. 1|. B
(0,1j
may be equivalently dened by {T (0. 1| [ T B].
Proof. It follows from Exercise 1.46 that o
B
(A T) = o(A) T for any =
A _ 2

and = T _ . In particular, we have B


(0,1j
= B T. |L
I Exercise 1.53 (1.4.4). B is generated by the compact subsets of R.
Proof. Denote
A
11
= { _ R
n
: is compact].
Let A
11
. Every compact set is closed (Heine-Borel Theorem); hence A
10
.
It follows that o(A
11
) _ o(A
10
). Now let A
10
. The sets
1
= 1. 1|
n
,
1 N, are compact; hence the countable union =
_
o
1=1

1
is in o(A
11
). It
follows that A
10
_ o(A
11
) and thus o(A
10
) _ o(A
11
). |L
5
Notation: A
3
= intervals of the form x, o), A
4
= intervals of the form (x, o), A
7
=
intervals of the form o, b), and A
10
= closed subsets of R.
2
MEASURES
Remark 2.1 (The de Finetti Notation). I nd the de Finetti Notation is very
excellent. Here I cite Pollard (2001, Sec.4, Ch.1).
Ordinary algebra is easier than Boolean algebra. The correspondence ==
1

between _ and their indicator functions,


1

(.) =

1 if .
0 if . .
transforms Boolean algebra into ordinary pointwise algebra with functions.
The operations of union and intersection correspond to pointwise maxima
() and pointwise minima (.), or pointwise products:
1
_
i

i
(.) =
_
i
1

i
(.). and (2.1)
1
_
i

i
(.) =

i
1

i
(.) =

i
1

i
(.). (2.2)
Complements corresponds to subtraction from one:
1

c (.) = 1 1

(.). (2.3)
Derived operations, such as the set theoretic dierence T :T
c
and
the symmetric dierence, ^T :( T)L(T ), also have simple algebraic
counterparts:
1
~B
(.) =
_
1

(.) 1
B
(.)
_

:max {0. 1

(.) 1
B
(.)] . (2.4)
1
ZB
(.) =[1

(.) 1
B
(.)[ . (2.5)
The algebra looks a little cleaner if we omit the argument .. For example,
the horrendous set theoretic relationship
_
_
n
_
i=1

i
_
_
^
_
_
n
_
i=1
T
i
_
_
_
n
_
i=1
(
i
^T
i
)
29
30 CHAPTER 2 MEASURES
corresponds to the pointwise inequality

i=1
1

i

n

i=1
1
B
i

6
n
_
i=1

i
1
B
i

.
whose verication is easy: when the right-hand side takes the value 1 the in-
equality is trivial, because the left-hand side can take only the values 0 or 1;
and when right-hand side takes the value 0, we have 1

i
= 1
B
i
for i , which
makes the left-hand side zero.
2.1 Measures
I Exercise 2.2 (2.1.1). This problem deals with some other variants of proper-
ties (M1)(M3).
a. Some dene a probability measure 1 on a o-eld A of subsets of by
stipulating that (i) 0 6 1() 6 1 for all A, (ii) 1 () = 1, and (iii) 1 is
countably additive. This is a special case of our denition of a measure.
b. If (M1) and (M3) hold for a set function j dened on a eld Awith j() < o
for some A, then j is a measure on A.
Proof. (a) If j: A 0. o| is a measure, dene a new set-valued function
1 : A

R as
1() =
j()
j()
. V A.
(b) We only need to check (M2): j() = 0. Since A is a eld, A. Consider
the following sequence {. . . . . .]. (M3) implies that
j() = j( L L L ) = j()

j().
Since j() < o, we have j() = 0. |L
IExercise 2.3 (2.1.2). Let = {o
1
. . . . . o
n
], and let
1
. . . . .
n
0. o|. Dene
j on 2

as in Example 2. Then (. 2

. j) is a measure space, and j is o-nite


i
n
< o for each n N.
Proof. To prove (. 2

. j) is a measure space, we only need to prove that j is


a measure on 2

since 2

is a o-eld. Clearly (M1) and (M2) hold. To see (M3)


hold, let
1
. . . . .
n
2

be disjoint (Since is nite, we need only to check


the nite additivity). Then
SECTION 2.1 MEASURES 31
j
_
_
n
_
i=1

n
_
_
=

k
: k is such that o
k

i
for some i {1. . . . . n]
_
=
n

i=1

k
: k is such that o
k

i
_
=
n

i=1
j(
i
) .
If
i
= o for at least one i , then j is not o-nite. If each
i
is nite then
j is o-nite: take
i
= {o
i
], where i {1. . . . . n]. |L
I Exercise 2.4 (2.1.3). Let A = {. ], j() = 0, and j() = o. Then
(. A. j) is a measure space, but j fails to be o-nite.
Proof. {. ] is a (trivial) o-eld. (M1) and (M2) hold. Now check (M3):
j(L ) = j() = 0 j() = j() j() .
Notice that = L or = , but by the hypothesis, j() = o. |L
IExercise 2.5 (2.1.4). Let be uncountable. Let A = { _ : is amc or
c
is amc].
Write j() = 0 if is amc and j() = o if
c
is amc. Then (. A. j) is a
measure space, and j is not o-nite.
Proof. We show A is a o-eld rst. F since
c
= is amc. If A, then
either or
c
is amc. If is amc,
c
A because (
c
)
c
= is amc; if
c
is
amc,
c
A by denition of A. To see that A is closed under countably union,
let {
n
]
o
n=1
A. There are two cases: (i) Each
n
is amc. Then
_
o
n=1

n
is amc,
whence is a A-set, and (ii) At least one
n
is such that
c
n
is amc. Without
loss of generality, we assume
c
1
is amc. Since (
_
o
n=1

n
)
c
=
_
o
n=1

c
n
_
c
1
, it
follows that
_
o
n=1

n
A.
We then show that j is a measure on A. It is clear that j is nonnegative and
j() = 0. Now let {
n
]
o
n=1
_ A be disjoint. If each
n
is amc, then
_
o
n=1

n
is amc, and so j(
_
o
n=1

n
) = 0 =

o
n=1
j(
n
); if there is at least one
n
, say

1
, so that
c
1
is amc, then (
_
o
n=1

n
)
c
is amc. Hence, o = j(
_
o
n=1

n
) =

o
n=1
j(
n
) = j(
1
)

o
n=2
j(
n
) = o.
Since is uncountable, which cannot be covered by a sequence of countable
A-sets. Therefore, in any cover of , there exists a set so that
c
is amc. But
which means that j is not o-nite since j() = o. |L
IExercise 2.6 (2.1.5). Let be arbitrary, and let A = { _ : is amc or
c
is amc].
Dene j over A by stating that j() = 0 if is amc, and j() = 1 if
c
is amc.
a. j is not well-dened if is amc, but j is well-dened if is uncountable.
b. j is o-nite measure on the o-eld A when is uncountable.
32 CHAPTER 2 MEASURES
Proof. (a) If is amc, we can nd a set such that both and
c
are amc.
But then (i) j() = 0 since is amc, and (ii) j() = 1 since
c
is amc. A
contradiction.
However, if is uncountable, then the previous issue will not occur because
if both and
c
are amc, then L
c
= is amc. A contradiction.
(b) We have proved in Exercise 2.5 that A is a o-eld. To prove that j is o-
nite, consider {. . . . . .]. |L
I Exercise 2.7 (2.1.6). Suppose that A is a nite o-eld on . Suppose that j
is dened on A such that (M1), (M2), and (M4) hold. Then (. A. j) is a measure
space.
Proof. Since A is a nite o-eld, any countable union of A-sets must take the
following form

1
L
2
L L
n
L L L
Then the proof is straightforward. |L
I Exercise 2.8 (2.1.7). Let A = { _ : is nite or
c
is nite]. Dene j on
A by
j() =

0 if is nite
1 if
c
is nite.
a. j fails to be well-dened when is nite.
b. If is innite, then j satises (M1), (M2), and (M4).
c. Let card() =
0
. Then j is nitely additive but not countably subadditive.
d. When is uncountable, j is a measure. Is j o-nite?
Proof. (a) Let be nite, and both and
c
are nite. Then j() = 0 and
j() = 1 occurs.
(b) The nonnegativity and j() = 0 are obvious. To see nite additivity, let
{
i
]
n
i=1
_ A be disjoint, and
_
n
i=1

i
A. If each
i
is nite, then
_
n
i=1

i
is nite, whence j(
_
n
i=1

i
) = 0 =

n
i=1
j(
i
); if, say,
c
1
is nite, then

_
n
i=1

i
|
c
=
_
n
i=1

c
i
_
c
1
is nite, and j(
_
n
i=1

i
) = 1. Notice that
}
_
c
1
for all = 2. 3. . . . . n since {
i
]
n
i=1
is disjoint. Hence
2
.
3
. . . . .
n
are all nite
if
c
1
is nite. Therefore,

n
i=1
j(
i
) = 1 = j(
_
n
i=1

i
).
(c) Since card() =
0
, is innitely countable. Then j is well-dened and
nitely additive by part (b). To show j fails to be countably subadditive, let
= {o
1
. o
2
. . . .], and
n
= {o
n
]. Hence j(
n
) = 0 and so

o
n=1
j(
n
) = 0. But
_
o
n=1

n
= and j(
_
o
n=1

n
) = 1 since (
_
o
n=1

n
)
c
= is nite.
SECTION 2.1 MEASURES 33
(d) j is o-nite when is uncountable. Just consider the following sequence
of sets {. . . . . .]. Note that j() = 1 < o as
c
= is nite, and
j() = 0 < o as is nite. Finally, = L L L . |L
I Exercise 2.9 (2.1.8). Let card() =
0
and A = 2

. Let
j() =

0 if is nite
o if is innite.
Then j is well-dened, j satises (M1), (M2), and (M4), and that (M3) fails. Also,
j is o-nite.
Proof. It is straightforward to see that j is well-dened, nonnegative, and
j() = 0. Use the ways as in the previous exercise, we prove that j is nite
additive, but not countable additive. To prove j is o-nite, note that card() =

0
( is innitely countable), can be expressed as = {o
1
. o
2
. . . .]; hence,
we can just consider the following sequence {
1
= {o
1
].
2
= {o
2
]. . . .]. |L
IExercise 2.10 (2.1.9). (M5) is not true if the hypothesis j() < ois omitted.
Proof. Suppose _ T with . T. T A. Then j(T) = j() j(T ).
If j() = o, then j(T) = o since j(T ) > 0. Then j(T) j() =
(o) (o) is undened. |L
I Exercise 2.11 (2.1.10). Let j denote a measure on a o-eld A, and let
.
1
.
2
. . . . A.
a. j() =

o
k=1
j(
k
) when the
k
s are disjoint with
_
o
k=1

k
= .
b. j(
1
^
2
) = 0 i j(
1
) = j(
2
) = j(
1

2
).
c. j(
2
) = 0 forces both j(
1
L
2
) = j(
1
) and j(
1
^
2
) = 0.
d. j(
2
) = 0 forces j(
1

2
) = j(
1
).
Proof. (a) We have =
__
k

k
_
=
_
k
(
k
), and {
k
] A is
disjoint. Hence,
j() = j
_
_
o
_
k=1
(
k
)
_
_
=
o

k=1
j(
k
) .
(b) If j(
1
) = j(
2
) = j(
1

2
), then
j(
1
) = j(
1

2
) j(
1

2
) ==j(
1

2
) = 0.
j(
2
) = j(
1

2
) j(
1

2
) ==j(
1

2
) = 0.
Therefore, j(
1
^
2
) = j(
1

2
) j(
1

2
) = 0.
If j(
1
^
2
) = 0, then j(
1

2
) = j(
1

2
) = 0. But then j(
1
) =
j(
1

2
) j(
1

2
) = j(
1

2
) and j(
2
) = j(
1

2
).
34 CHAPTER 2 MEASURES
(c) 0 6 j(
1

2
) 6 j(
2
) = 0 implies that j(
1

2
) = 0. By the inclusion-
exclusion principle,
j(
1
L
2
) = j(
1
) j(
2
) j(
1

2
) = j(
1
).
(d) Since
1
L
2
= (
1
^
2
) L(
1

2
), and (
1
^
2
) (
1

2
) = , we have
j(
1
L
2
) = j(
1
^
2
) j(
1

2
) = j(
1
^
2
)
= j(
1

2
) j(
1

2
)
= j(
1

2
) . |L
I Exercise 2.12 (2.1.11). Let (. A. j) be a measure space such that there is
T A with 0 < j(T) < o. Fix such a T, and dene j
B
: A R by the
formula j
B
() = j( T) ,j(T).
a.
_
. A. j
B
_
is a measure space.
b. Suppose in addition that is the disjoint union of an amc collection of sets
T
n
A such that each T
n
has nite measure, and suppose that j is nite.
Then for all A we have j() =

n
j
B
n
() j(T
n
). Also, for each i N
we have
j
B
i
() =
j

(T
i
) j()

n
j

(T
n
) j()
.
This formula is known as Bayes Rule.
Proof. (a) If suces to show that j
B
is a measure on A since A is a o-eld.
(M1) To see j
B
() > 0, note that j(T) > 0 and j( T) > 0. (M2) To see
j
B
() = 0, note that j
B
() =
j( T)
j(T)
= 0. (M3) For countable additivity,
let {
n
]
o
n=1
_ A be disjoint. Then
j
B
(
o
_
n=1

n
) =
j
_
_
_T
_
_
o
_
n=1

n
_
_
_

_
j(T)
=
j
_
_
o
_
n=1
(
n
T)
_
_
j(T)
=
o

n=1
j(
n
T)
j(T)
=
o

n=1
j(
n
T)
j(T)
=
o

n=1
j
B
(
n
) .
(b) By the assumption, we can write as =
_
o
n=1
T
n
, where {T
n
]
o
n=1
_ A is
disjoint, and j(T
n
) < o. Since j is nite, j() < o. For the rst claim,
SECTION 2.2 CONTINUITY OF MEASURES 35

n
j
B
n
() j(T
n
) =

n
j( T
n
)
j(T
n
)
j(T
n
) =

n
j( T
n
)
= j
_
_
_
n
( T
n
)
_
_
= j( )
= j().
For the Bayes Rule,
j
B
i
() =
j(T
i
)
j()
=
j(T
i
)
j()
j()

n
j
B
n
() j(T
n
)
=
j

(T
i
) j()

n
j
B
n
() j(T
n
)
. |L
I Exercise 2.13 (2.1.12). Let S = {s
1
. . . . . s
n
], and let {
u
: u S] denote a
collection of nonnegative numbers with

uS

u
= 1. Let denote the set of
sequences of S. For each o , write o = (:
1
(o). :
2
(o). . . .). Given n N and
H _ S
n
, let
C
n
(H) =
_
o :
_
:
1
(o). . . . . :
n
(o)
_
H
_
.
Such a set is called a cylinder of rank n. Let F = {C
n
(H): n N. H _ S
n
], so
that F consists of all cylinders of all ranks. Dene j : F

R as follows: if for
some n N and H _ S
n
we have = {o : (:
1
(o). . . . . :
n
(o)) H], write
j() =

i
1
. . . . .
i
n
: (i
1
. . . . . i
n
) H
_
.
a. j is well dened.
2.2 Continuity of Measures
I complete the Claim 1 of Vestrup (2003, p. 43) before working out the exer-
cises for this section. Note that we sometimes take the following notation (see
Rosenthal, 2006, p. 34):
liminf
n
=
n
ev.|:
n
eventually,
limsup
n
=
n
i. o.|:
n
innitely often.
Ash (2009) provides an excellent treatment of limsup and liminf for real num-
ber sequences.
Claim 5. Let {
n
]
o
n=1
denote a sequence of subsets of . Then we have the
following properties:
a. liminf
n
=

o : o is in all but nitely many of


1
.
2
. . . .
_
.
b. limsup
n
=

o : o is in innitely many of
1
.
2
. . . .
_
.
36 CHAPTER 2 MEASURES
c. liminf
n
_ limsup
n
.
d. If {
n
]
o
n=1
is nondecreasing, then lim
n
exists and equals
_
o
n=1

n
.
e. If {
n
]
o
n=1
is nonincreasing, then lim
n
exists and equals
_
o
n=1

n
.
f. If
1
.
2
. . . . are disjoint, then lim
n
exists and equals .
Proof. (b) If o limsup
n
, then for all k N, there exist some n > k such
that o
n
. Hence, o is in innitely many of
1
.
2
. . . .. Conversely, if o is in
innitely many of
1
.
2
. . . ., then for all k N, there exists n > k such that
o
n
. Therefore, o
_
o
k=1
_
o
n=k

n
= limsup
n
.
(e) Since
i1
_
i
, we get
_
k>n

k
=
n
. Therefore,
limsup
n
=
o
_
n=1
_
_
_
k>n

k
_
_
=
o
_
n=1

n
.
Likewise
liminf
n
=
o
_
n=1
_
_
_
k>n

k
_
_
_
o
_
k=1

k
= limsup
n
_ liminf
n
.
Thus equality prevails and so lim
n
=
_
o
n=1

n
. |L
I Exercise 2.14 (2.2.1). liminf 1

n
= 1
liminf
n
and limsup1

n
= 1
limsup
n
.
Proof. We rst show
1
_
1
nDk

n
= inf
n>k
1

n
. (2.6)
1
_
1
nDk

n
= sup
n>k
1

n
. (2.7)
To prove (2.6), we must show that the two functions are equal. But
1
_
1
nDk

n
(o) = 1 == o
o
_
n=k

n
== o
n
for all n > k
== 1

n
(o) = 1 for all n > k
== inf
n>k
1

n
(o) = 1.
Similarly, (2.7) holds since
SECTION 2.2 CONTINUITY OF MEASURES 37
1
_
1
nDk

n
(o) = 1 == o
o
_
n=k

n
== o
n
for some n > k
== 1

n
(o) = 1 for some n > k
== sup
n>k
1

n
(o) = 1.
Hence,
1
liminf
n
= 1
_
1
nD1
(
_
1
kDn

k)
= sup
n>1
1
_
1
kDn

k
= sup
n>1
inf
k>n
1

k
= liminf 1

n
.
1
limsup
n
= 1
_
1
nD1
(
_
1
kDn

k)
= inf
n>1
1
_
1
kDn

k
= inf
n>1
sup
k>n
1

k
= limsup1

n
.
Alternatively, we have
liminf 1

n
(o) = 1 == 1

n
(o) = 1 ev. == o
n
ev.
== o liminf
n
== 1
liminf
n
(o) = 1.
limsup1

n
(o) = 1 == 1

n
(o) = 1 i. o. == o
n
i. o.
== o limsup
n
== 1
limsup
n
(o) = 1. |L
I Exercise 2.15 (2.2.2). Show that liminf
n
_ limsup
n
without using the
representations of liminf
n
and limsup
n
given in parts (a) and (b) of Claim
1.
Proof. Notice that
o
o
_
k=1
o
_
n=k

n
==J k
+
N such that o
n
. V n > k
+
==V k N. J n > k such that o
n
==o
o
_
k=1
o
_
n=k

n
.
where the rst == holds because if k < k
+
, then o
n
for all n > k
+
; if
k > k
+
, then o
n
for all n > k. |L
IExercise 2.16 (2.2.3). (liminf
n
)
c
= limsup
c
n
and (limsup
n
)
c
= liminf
c
n
.
Proof. These results are analog to liminf .
n
= limsup(.
n
) and limsup.
n
=
liminf (.
n
). We have two methods to prove these claims. Here is the Method
1:
(liminf
n
)
c
=
_
_
o
_
k=1
o
_
n=k

n
_
_
c
=
o
_
k=1
_
_
o
_
n=k

n
_
_
c
=
o
_
k=1
o
_
n=k

c
n
= limsup
c
n
.
38 CHAPTER 2 MEASURES
(limsup
n
)
c
=
_
_
o
_
k=1
o
_
n=k

n
_
_
c
=
o
_
k=1
_
_
o
_
n=k

n
_
_
c
=
o
_
k=1
o
_
n=k

c
n
= liminf
c
n
.
Here is Methods 2:
. (liminf
n
)
c
==
_
(J N N) (V n > N) (.
n
)
_
== (V N N) (J n > N)
_
.
c
n
_
== . limsup
c
n
:
. (limsup
n
)
c
==
_
(V N N) (Jn > N) .
n
_
== (J N N) (V n > N)
_
.
c
n
_
== . liminf
c
n
. |L
I Exercise 2.17 (2.2.4). If T
n
=
n
for at most nitely many n N, then
liminf
n
= liminf T
n
and limsup
n
= limsupT
n
. In other words, liminf
n
,
limsup
n
[and lim
n
] are not changed if a nite number of
k
s are altered.
Proof. Just follow Claim 1. |L
I Exercise 2.18 (2.2.5). We have the following relations:
a. limsup(
n
LT
n
) = limsup
n
LlimsupT
n
and liminf(
n
T
n
) = liminf
n

liminf T
n
.
b. limsup(
n
T
n
) _ limsup
n
limsupT
n
and liminf(
n
LT
n
) _ liminf
n
L
liminf T
n
. Both containment relations can be strict.
c. 1
limsup(
n
LB
n
)
= max{1
limsup
n
. 1
limsupB
n
] and 1
liminf(
n
B
n
)
= min{1
liminf
n
. 1
liminf B
n
].
Proof. (a) By denition, limsup(
n
L T
n
) =
n
L T
n
i. o.| =
n
i. o.| L
T
n
i. o.|, and liminf(
n
T
n
) =
n
T
n
ev.| =
n
ev.| T
n
ev.|.
(b) We have
o limsup(
n
T
n
) ==o is in innitely many of
1
T
1
.
2
T
2
. . . .
+
==o is in innitely many of
1
.
2
. . . . and T
1
. T
2
. . . .
==o limsup
n
and o limsupT
n
==o limsup
n
limsupT
n
.
where (+) holds with == rather than == because, e.g., let
o

i
if i is odd
T
}
if is even:
(2.8)
then o is in innitely many of
1
.
2
. . . ., and o is in innitely many of
T
1
. T
2
. . . .. However, if
n
T
n
= for all n N, then o is not in any of

n
T
n
.
SECTION 2.2 CONTINUITY OF MEASURES 39
Likewise,
o liminf
n
L liminf T
n
==o is in all but nitely many of
1
.
2
. . . . or T
1
. T
2
. . . .
++
==o is in all but nitely many of (
1
L T
1
). (
2
L T
2
). . . .
==o liminf(
n
L T
n
).
where (++) holds with == rather than == because, e.g., consider (2.8);
then o is in all of
1
L T
1
,
2
L T
2
, . . ., but there does not exist N such that o
is in all
n
for all n > N or N
t
such that o is in all T
n
for all n > N
t
.
(c) We have
1
limsup(
n
LB
n
)
(o) = 1 == o limsup(
n
L T
n
)
== o
_
limsup
n
L limsupT
n
_
== max{1
limsup
n
(o). 1
limsupB
n
(o)] = 1.
1
liminf(
n
B
n
)
= 1 == o liminf(
n
T
n
)
== o
_
liminf
n
liminf T
n
_
== min

1
liminf
n
(o). 1
liminf B
n
(o)
_
= 1. |L
I Exercise 2.19 (2.2.6). If
n
and T
n
T, then
n
L T
n
L T,

c
n

c
,
n
T
n
T,
n
T
n
T, and
n
^T
n
^T.
Proof. (i) We have limsup(
n
L T
n
) = limsup
n
L limsupT
n
= L T,
liminf(
n
L T
n
) _ liminf
n
L liminf T
n
= L T, and liminf(
n
L T
n
) _
limsup(
n
L T
n
) = L T. Therefor,
n
L T
n
L T.
(ii) Notice that limsup
c
n
= (liminf
n
)
c
=
c
, and on the other hand
liminf
c
n
= (limsup
n
)
c
=
c
.
(iii) We have liminf(
n
T
n
) = liminf
n
liminf T
n
= T, limsup(
n

T
n
) _ limsup
n
limsupT
n
= T, and limsup(
n
T
n
) _ liminf(
n
T
n
).
(iv) Note that
n
T
n
=
n
T
c
n
. We have known that T
c
n
T
c
, so
n
T
n

T
c
= T by (ii) and (iii).
(v)
n
^T
n
= (
n
T
n
)L(T
n

n
). Since
n
T
n
T and T
n

n
T
by (iv), we have
n
^T
n
( T) L (T ) = ^T by (i). |L
I Exercise 2.20 (2.2.7). If
n
is T or C as n is even or odd, then liminf
n
=
T C, and limsup
n
= T L C.
Proof. We have
liminf
n
=
n
ev.| = T C.
limsup
n
=
n
i. o.| = T L C. |L
I Exercise 2.21 (2.2.8). limsup
n
liminf
n
= limsup(
n

c
n1
) =
limsup(
c
n

n1
).
40 CHAPTER 2 MEASURES
Proof. We have
. limsup(
n

c
n1
)
== (V N N)(J n > N)(.
n
and .
c
n1
)
== (.
n
i. o.) and (.
c
n
i. o.)
== . (limsup
n
limsup
c
n
) = limsup
n
liminf
n
.
The other equality can be proved similarly. |L
I Exercise 2.22 (2.2.9). a. limsup
n
liminf
k
(
n

c
k
) = .
b. limsup
k

k
= liminf
k
(
k
).
c. limsup
n
_
liminf
k

k

n
_
= .
d. limsup
n
(
n
) = liminf
n

n
and limsup
n
(
n
) = limsup
n

n
.
e. limsup
n
(^
n
) =
_
liminf
n

n
_
L limsup
n
(
n
).
f.
n
implies that limsup
n
(^
n
) = limsup
n
(
n
).
g. For arbitrary set 1. J. G and H we have (1^J) ^(G^H) = (1^G) ^(J^H).
We also have for any set that
limsup
k

k
liminf
k

k
= liminf
k

k
^limsup
k

k
=
_
liminf
k

k
^limsup
k

k
_
^(^)
=
_
liminf
k

k
^
_
^
_
limsup
k

k
^
_
.
Proof. (a)
SECTION 2.2 CONTINUITY OF MEASURES 41
limsup
n
liminf
k
(
n

c
k
)
= limsup
n
_
liminf
k
(
n

c
k
)
_
= limsup
n
_
liminf
k

n
liminf
k

c
k
_
_
by Exercise 2.18(a)
_
= limsup
n
_

n
liminf
k

c
k
_
= limsup
n
_
_

_
limsup
k

k
_
c
_
_
_
_
limsup
n

n
_

_
_
limsup
n
_
limsup
k

k
_
c
_
_
_
by Exercise 2.18(b)
_
=
_
limsup
n

n
_

_
limsup
k

k
_
c
= .
(b) liminf
k
(
k
) = liminf
k
(
c
k
) = liminf
k
liminf
k

c
k
=
(limsup
k

k
)
c
= limsup
k

k
.
(c) limsup
n
(liminf
k

k

n
) _ limsup
n
(liminf
k

k
)|(limsup
n

c
n
) =
_
liminf
k

k
_

_
liminf
n

n
_
c
= .
(d) limsup
n
(
n
) =
_
o
k=1
_
o
n=k
_

c
n
_
=
_
o
k=1
_

__
o
n=k

c
n
_
_
=
__
o
k=1
_
o
n=k

c
n
_
=
_
limsup
n

c
n
_
=
_
liminf
n

n
_
c
= liminf
n

n
,
and limsup
n
(
n
) =
_
o
k=1
_
o
n=k
(
n

c
) =
_
o
k=1
_
__
o
n=k

n
_

c
_
=
limsup
n

n
.
(e) limsup
n
(^
n
) = limsup
n
(
n
) L (
n
)| = limsup
n
(
n
) L
limsup
n
(
n
) = ( liminf
n

n
) L limsup
n
(
n
).
(f) If
n
, then limsup
n
= liminf
n
= . Hence, limsup
n
(^
n
) =
_
liminf
n

n
_
Llimsup
n
(
n
) = ( )Llimsup
n
(
n
) = limsup
n
(
n

).
(g) We rst show for all . T. C 2

,
(^T) ^C = ^(T^C) . (2.9)
This equation hold because
1
1
This proof is not elegant. See Resnick (1999, Exercise 1.43).
42 CHAPTER 2 MEASURES
(^T)^C =
_
( T) L (T )
_
^C
=
_
_
( T) L (T )
_
C
_
L
_
C
_
( T) L (T )
_
_
=
_
_
_
T
c
_
L (
c
T)
_
C
c
_
L
_
_
_
T
c
_
L (
c
T)
_
c
C
_
=
_
( T
c
C
c
) L (
c
T C
c
)
_
L
_
_
_
T
c
_
c
| (
c
T)
c
_
C
_
+
=
_
( T
c
C
c
) L (
c
T C
c
)
_
L
_
_
( T) L (
c
T
c
)
_
C
_
=
_
( T
c
C
c
) L (
c
T C
c
)
_
L
_
( T C) L (
c
T
c
C)
_
=
_

_
_
T
c
C
c
_
L (T C)
_
_
_
_

_
_
T C
c
_
L
_
T
c
C
_
_
_
=
_

_
(T L C) |
_
T
c
L C
c
_
_
c
_
_
_

c
(T^C)
_
++
=
_
(T^C)
c
_
_
_

c
(T^C)
_
= ^(T^C) .
where equality (+) holds because
_
T
c
_
c
(
c
T)
c
=
_

c
L T
_

_
L T
c
_
=
_
_

c
L T
_

_
L
_
_

c
L T
_
T
c
_
= ( T) L (
c
T
c
).
and equality (++) holds because
(T L C) |
_
T
c
L C
c
_
=
_
(T L C) T
c
_
L
_
(T L C) C
c
_
=
_
T
c
C
_
L
_
T C
c
_
.
By (2.9), we have
(1^J) ^(G^H) = 1^
_
J^(G^H)
_
= 1^
_
J^(H^G)
_
= 1^
_
(J^H) ^G
_
= 1^
_
G^(J^H)
_
= (1^G) ^(J^H) .
Now it suces to show that (liminf
k

k
)^(limsup
k

k
) =
_
limsup
k

k
_

_
liminf
k

k
_
. Notice that
(liminf
k
)^(limsup
k
)
=
_
_
limsup
k
_

_
liminf
k
_
_
L
_
_
liminf
k
_

_
limsup
k
_
_
=
_
_
limsup
k
_

_
liminf
k
_
_
L
_
_
liminf
k
_

_
liminf
c
k
_
_
+++
=
_
limsup
k
_

_
liminf
k
_
.
SECTION 2.2 CONTINUITY OF MEASURES 43
where (+ + +) holds because (liminf
k

k
) (liminf
k

c
k
) =
k
ev.|
c
k
ev.| =
. |L
I Exercise 2.23 (2.2.10). Let = N, and let A = 2

. Dene j on A by j() =
number of points in if is nite; dene j() = o if is innite.
a. j is a measure on A. This measure is called the counting measure.
b. There exists a nonincreasing sequence {
n
]
o
n=1
of A-sets with j(
n
) = o
for all n N but j
_
lim
n

n
_
= 0, thus (M9) accordingly fails to hold, hence
the assumption that some
k
must have nite measure cannot be dropped.
Proof. (a) can be found in Vestrup (2003, Example 2, p. 37). For (b), let
n
=
{n. n 1. . . .] for each n N, then
n
, j(
n
) = o, but j
_
lim
n

n
_
=
0. |L
IExercise 2.24 (2.2.11). Let (. A. j) denote a measure space. Let {
x
: . R. . > 0]
denote a collection of A-sets.
a. Suppose that 0 < . < , implies
x
_
,
. Then (i)
_
x>0

x
A, (ii) .
1
< .
2
implies j(
x
1
) 6 j(
x
2
), and (iii) j(
x
) j(
_
,>0

,
) as . o.
b. Suppose that 0 < . < , implies
x
_
,
. Also, further assume that j(
z
) <
o for some : > 0. Then (i)
_
x>0

x
A, (ii) .
1
< .
2
implies j(
x
1
) > j(
x
2
),
and (iii) j(
x
) j(
_
,>0

,
) as . o. The assumption j(
z
) < o for
some : > 0 cannot be dropped. This and (a) generalize (M8) and (M9) from
monotone sequences of sets to monotone [uncountable] collections of sets.
Proof. (a) Denote {
x
: . R. . > 0] as A
R
. Dene a subset A
N
of A
R
as
follows:
A
N
= {
n
A: n N. n > 0].
Then, for every . R and . > 0, there exists n N such that . 6 n (by the
Archimedan property; see Rudin 1976, Theorem 1.20); that is,
x
_
n
. Thus,
_
A
R
=
_
A
N
A.
(b) Dene B = {
1{n
: n N. n > 0]. Then for every . R and . > 0, there
exists n N with n > 0 such that . < 1,n; that is,
1{n
_
x
. Thus,
_
A
R
=
_
B. |L
I Exercise 2.25 (2.2.12). Let (. A. j) denote a measure space.
a. j is o-nite i there is a nondecreasing sequence
1
_
2
_ of A-sets
with j(
n
) < o for all n N and
_
o
n=1

n
= .
b. j is o-nite i there is a disjoint sequence
1
.
2
. . . . of A-sets with j(
n
) <
o for all n N and
_
o
n=1

n
= .
44 CHAPTER 2 MEASURES
c. Let j
1
. . . . . j
n
denote o-nite measures on A. Then there exists a sequence
{
n
]
o
n=1
of A-sets such that (i) j
i
(
}
) < o for all i = 1. . . . . n and N
and (ii) =
_
o
n=1

n
. These sets may be chosen to be nondecreasing or
disjoint.
d. Does (c) hold if we have countably many o-nite measures on A as compared
to nitely many o-nite measures on A.
Proof. (a) The if part is trivial. So assume that j is o-nite. Then there exists
{T
n
] _ A with j(T
n
) < o for all n N and
_
T
n
= . Let
n
=
_
n
k=1
T
k
.
Then {
n
] is nondecreasing and
_
T
n
=
_

n
= .
(b) Again, the if part is trivial. So assume that j is o-nite. Let {T
n
] as in (a).
Let
1
= T
1
, and
n
= T
n
(
_
n-1
i=1
T
i
) for n > 2. Then {
n
] is disjoint and
_
T
n
=
_

n
= .
(c) Let j
1
. . . . . j
n
be o-nite. Then for each i = 1. . . . . n, there exists {
ik
] A
such that j
i
(
ik
) < ofor all k Nand
_
k

ik
= . Now let
n
=
_
n
I=1

In
.
For each i and ,
j
i
(
}
) = j
i
_
_
n
_
I=1

I}
_
_
6
n

I=1
j
i
(
}
) < o. (2.10)
and
_

n
= . It follows from (a) and (b) that {
n
] may be chosen as nonde-
creasing or disjoint.
(d) (c) may not hold if we have countably many o-nite measures on A since
(2.10) may fail. |L
I Exercise 2.26 (2.2.13). Let j denote a measure on a o-eld A, and let

1
.
2
. . . . A be such that j(
_
o
}=1
liminf
k
(
}

c
k
)) < o for some N N.
Use (M10) and parts (a)(c) of Exercise 2.22 to show the following claims:
a. lim
n
j(liminf
k
(
n

c
k
)) exists and equals zero.
b. lim
n
j(
n
limsup
k

k
) exists and equals zero.
c. lim
n
j((liminf
k

k
)
n
) exists and equals zero.
Proof. (a) Since there exists N Nsuch that j(
_
o
}=1
liminf
k
(
}

k
)) < o,
it follows from (M10) and Exercise 2.22(a) that
SECTION 2.2 CONTINUITY OF MEASURES 45
j
_
liminf
n
_
liminf
k
(
n

k
)
_
_
6 liminf
n
j
_
liminf
k
(
n

k
)
_
6 limsup
n
j
_
liminf
k
(
n

k
)
_
6 j
_
limsup
n
_
liminf
k
(
n

k
)
_
_
= j()
= 0.
Thus, lim
n
j(liminf
k
(
n

c
k
)) = 0.
(b) Notice that
n
limsup
k

k
= liminf
k
(
n

k
) by Exercise 2.22(b). Then
(b) follows from (a) immediately.
(c) Using (M10) and Exercise 2.22(c), we get (c). |L
I Exercise 2.27 (2.2.16). Let A be a eld on , and suppose that j: A

R
satises (M1) with j() < o, (M2), (M4) (and hence (M5)), and in addition is
continuous from above at . Then j is a measure.
Proof. Let {T
n
] A be disjoint, and
_
n
T
n
A. For n > 2, let C
n
=
_
o
k=n
T
k
.
Then {C
n
] is nonincreasing and converges to
_
n
C
n
=
_
n
o
_
k=n
T
k
= limsup
n
T
n
= .
Then lim
n
j(C
n
) = 0; that is,
0 = lim
n
j
_
_
o
_
k=n
T
k
_
_
= lim
n
j
_
_
_
_
_
o
_
k=1
T
k
_
_

_
_
n-1
_
i=1
T
i
_
_
_
_
_
= j
_
_
_
n
T
n
_
_
lim
n
j
_
_
n-1
_
i=1
T
i
_
_
= j
_
_
_
n
T
n
_
_
lim
n
n-1

i=1
j(T
i
)
= j
_
_
_
n
T
n
_
_

n=1
j(T
n
).
i.e., j(
_
n
T
n
) =

n
j(T
n
). |L
I Exercise 2.28 (2.2.17). Let = (0. 1|, and let F consist of and the nite
disjoint unions of rsc subintervals of (0. 1|. Then F is a eld. Dene j on F as
follows: j() = 1 if there exists c

> 0 with (1,2. 1,2 c

| _ and j() = 0
46 CHAPTER 2 MEASURES
otherwise. Then j is well-dened and satises (M1), (M2), and (M4), but j is not
countably additive.
Proof. We rst show that F is a eld. Suppose that = (a
1
. a
t
1
| L (a
n
. a
t
n
|,
where the notation is so chosen that a
1
6 a
n
. If the (a
i
. a
t
i
| are disjoint, then

c
= (0. a
1
| L (a
t
1
. a
2
| L L (a
t
n-1
. a
n
| L (a
t
n
. 1| and so lies in F (some of these
intervals may be empty, as a
t
i
and a
i1
may coincide). If T = (b
1
. b
t
1
| L L
(b
n
. b
t
n
|, then (b
}
. b
t
}
| again disjoint, then
T =
n
_
i=1
n
_
}=1
_
(a
i
. a
t
i
| (b
}
. b
t
}
|
_
:
each intersection here is again an interval or else the empty set, and the union
is disjoint, and hence T F .
Nevertheless, F is not a o-eld: It does not contain the singleton {.], even
though each is a countable intersection
_
n
(. 1,n. .| of F -sets.
The set function j dened above is not countably additive. Counter the
rational number on (0. 1| starting 1: {1. .
1
. .
2
. . . .]. This set is countable. Con-
sider the collection {(.
1
. 1|. (.
2
. .
1
|. . . .]. Then j(
_
(.
i
. .
i-1
|) = 1, however,

o
i=1
j((.
i
. .
i-1
|) = 0. |L
I Exercise 2.29 (2.2.18). Let (. A. j) be a measure space. Suppose that j is
nonatomic: A and j() > 0 imply that there exists T _ with T A with
0 < j(T) < j().
a. The measure j of Example 2 in Section 2.1 is atomic.
b. Suppose A is such that j() > 0, and let c > 0 be given. Then there exists
T A with T _ and 0 < j(T) < c.
c. Let A be such that j() > 0. Given any 0 6 6 j() there exists a set
T A with T _ and j(T) = .
Proof. (a) Let = {o
0
]. Then 2

and j() = 1. The only subsets of (in


A) is and . But j() = 0 and j() = j(). So the unit mass at o
0
is atomic.
(b) Take an arbitrary
1
A with
1
_ and 0 < j(
1
) < j(). Since j() =
j(
1
) j(
1
) and j(
1
) > 0 (otherwise j(
1
) = j()), we know that
either
0 < j(
1
) 6 j(),2. (2.11)
or
0 < j(
1
) 6 j(),2. (2.12)
If (2.11) holds, take an arbitrary
2
Awith
2
_
1
and 0 < j(
2
) < j(
1
).
Then either
0 < j(
2
) 6 j(
1
),2 6 j(),2
2
.
or
SECTION 2.3 A CLASS OF MEASURES 47
0 < j(
1

2
) 6 j(
1
),2 6 j(),2
2
.
If (2.12) holds, take
2
A with
2
_
1
and 0 < j(
2
) < j(
1
). Then
either
0 < j(
2
) 6 j(
1
),2 6 j(),2
2
.
or
0 < j(
2
) 6 j(
1

2
),2 6 j(
1
),2 6 j(),2
2
.
Thus, there exists
2
A with
2
_ such that 0 < j(
2
) 6 j(),2
2
.
Then by mathematical induction principle, we can show that there exists

n
A with
n
_ and 0 < j(
n
) 6 j(),2
n
. By letting n suciently large
and setting T =
n
, we get the result.
(c) If = 0, set T = ; if = j(), set T = . So we assume that 0 < < j().
Let C denote the family of collection D of countable disjoint A-sets con-
tained in such that

TD
j(D) 6 . Notice that C is well-dened by (b). For
D. E C, write D 4 E i (i)

TD
j(D) 6

TE
j(1), and (ii) D _ E.
It is clear that (C. 4) is a partially ordered set since 6 and _ are partial
orderings. For any chain D _ C, there exists an upper bound
_
D. It follows
from Zorns Lemma that there exists a maximal element F C.
Let T =
_
F . We nally show that j(T) =

TF
j(J) = . Assume that
j(T) < . Then j(T) > 0 for otherwise j() = j(T)j(T) = < j().
Take an arbitrary c > 0 such that c < j(T). It follows from (b) that there
exists C A with C _ T and 0 < j(C) < c. Let G = F L {C]. Then all sets
in G are disjoint and

GG
j(G) =

TF
j(J) j(C) < j(T) c < :
that is, G C. Further,

GG
j(G) =

TF
j(J) j(C) >

TF
j(J) since
j(C) > 0, and F G. Thus, F - G. In Contradict the fact that F is maximal
in C. |L
I Exercise 2.30 (2.2.19). Let (. A. j) be a measure space. Let T A and
A
B
= { A: _ T]. Then A
B
is a o-led on T, and the restriction of j to A
B
is a measure on A
B
.
Proof. Automatically, T A
B
. If A
B
, then A, so T A and
T _ T, i.e., T A
B
. Finally, if {
n
] _ A
B
, then
n
A and
n
_ T
for each n. Thus,
_

n
A and
_

n
_ T, i.e.,
_

n
A
B
. It is trivial to verify
that j is a measure on A
B
. |L
2.3 A Class of Measures
I Exercise 2.31 (2.3.1). Let
1
. . . . .
n
> 0. Fix n real numbers .
1
< .
2
< <
.
n
, and dene
48 CHAPTER 2 MEASURES
J(.) =

0 if . < .
1

1

}
if there exists 1 6 6 n such that .
}
6 . < .
}1
1 if . > .
n
.
^
T
is a measure on A
1
, and
^
T
((a. b|) =

}
: is such that a < .
}
6 b
_
if (a. b| {.
1
. . . . . .
n
] =
0 otherwise.
This is an important framework in discrete probability theory.
Proof. We rst show that ^
T
takes the given form. If (a. b| {.
1
. . . . . .
n
] = ,
then either a < b < .
1
< .
2
< < .
n
or .
1
< .
2
< < .
n
< a < b. Hence,
J(a) = J(b) = 0 or J(a) = J(b) = 1, so ^
T
((a. b|) = J(b) J(a) = 0. If
(a. b| {.
1
. . . . . .
n
] = , then there exists i. = 1. . . . . n such that a < .
i
< .
}
6
b. Hence,
^
T
((a. b|) = J(b) J(a) = (
1

}
) (
1

i
)
=

} : o~x
j
6b

}
.
To see that ^
T
is a measure, notice that (M1) and (M2) are satised auto-
matically. Let {(a
n
. b
n
|]
o
n=1
_ A
1
is disjoint, and assume that a
1
< b
1
6 a
2
<
b
2
6 6 a
n
< b
n
6 , whence,
^
T
_
_
o
_
n=1
(a
n
. b
n
|
_
_
=

k
: k is such that .
k

n
for some n N
_
=
o

n=1

k
: k is such that .
k

n
_
=
o

n=1
^
T
_
(a
n
. b
n
|
_
. |L
IExercise 2.32 (2.3.2). This problem generalizes Example 4. Let : R
k
R be
such that is continuous and nonnegative. Further suppose that
_
o
-o

_
o
-o
(t) dt <
o. Dene a function J : R
k
R by
J(x) =
_
x
1
-o

_
x
k
-o
(t) dt. x R
k
.
Then ^
T
is a measure on the semiring A
k
, and for all (a. b| A
k
we have
^
T
_
(a. b|
_
=
_
b
1
o
1

_
b
k
o
k
(t
1
. . . . . t
k
) dt
k
dt
1
.
SECTION 2.3 A CLASS OF MEASURES 49
Proof. J is continuous. To derive ^
T
k
, we use the mathematical induction.
If n = 1, then ^
T
1
((a. b|) = J(b) J(a) =
_
b
o
(t ) dt . Let us assume that the
hypothesis hold for n = k, and consider n = k 1:
^
T
kC1
()
=

xV()
s

(x)J
k1
(x)
=

xV
1
()
s

(x)J
k1
(x)

xV
2
()
s

(x)J
k1
(x)
=

V(

)
s

(x
+
. a
k1
)J
k1
(x
+
. a
k1
)

V(

)
s

(x
+
. b
k1
)J
k1
(x
+
. b
k1
)
=

V(

)
(1) s

(x
+
)J
k1
(x
+
. a
k1
)

V(

)
s

(x
+
)J
k1
(x
+
. b
k1
)
=

V(

)
s

(x
+
)
_
J
k1
(x
+
. b
k1
) J
k1
(x
+
. a
k1
)
_
=

V(

)
s

(x
+
)
_
b
kC1
o
kC1
J
k
(x
+
) dt
k1
=
_
b
kC1
o
kC1
_
_

xV(

)
s

(x
+
)J
k
(x
+
)
_
_
dt
k1
=
_
b
1
o
1

_
b
kC1
o
kC1

_
t
1
. . . . . t
k1
_
dt
k1
dt
1
> 0.
Hence, J
k
S
k
. |L
I Exercise 2.33 (2.3.3). Let J
1
. . . . . J
k
S. For each x R
k
, write J(x) =

k
i=1
J
i
(.
i
).
a. J S
k
, hence ^
T
is a measure on A
k
.
b. ^
T
_
(a. b|
_
=

k
i=1
_
J
i
(b
i
) J
i
(a
i
)
_
for all (a. b| A
k
.
Proof. The continuity of J is clear. For J
i
, we have ^
T
i
_
(a
i
. b
i
|
_
= J
i
(b
i
)
J
i
(a
i
). We can derive the form of ^
T
as in Example 3. |L
I Exercise 2.34 (2.3.4). Suppose that J
i
S
k
i
for i = 1. . . . . n. Suppose that
J : R

n
iD1
k
i
R is such that
J(x
(1)
. . . . . x
(n)
) =
n

i=1
J
i
(x
(i)
)
for each x
(1)
R
k
1
. . . ., and x
(n)
R
k
n
. Then ^
T
is a measure on A

n
iD1
k
i
. Also,
^
T
(
1

n
) =

n
i=1
^
T
i
(
i
) for each
1
A
1
. . . ., and
n
A
k
n
.
50 CHAPTER 2 MEASURES
Proof. We use mathematical induction. If i = 1 then J(x
(1)
) = J
1
(x
(1)
), and
^
T
(
1
) = ^
T
1
(
1
). If i = 2 then J(x
(1)
. x
(2)
) = J
1
(x
(1)
) J
2
(x
(2)
). Consider
any (x
(1)
. x
(2)
) V (
1

2
). By denition,
s

2
(x
(1)
. x
(2)
) =

1 if [{i. : .
(1)
i
= a
(1)
i
. .
(2)
}
= a
(2)
}
][ is even
1 if [{i. : .
(1)
i
= a
(1)
i
. .
(2)
}
= a
(2)
}
][ is odd.
s

2
(x
(1)
. x
(2)
)J(x
(1)
. x
(2)
) = s

2
(x
(1)
. x
(2)
)
_
J
1
(x
(1)
)J
2
(x
(2)
)
_
. (2.13)
v If [{i : .
(1)
i
= a
(1)
i
][ = #
(1)
is even, and [{i : .
(2)
i
= a
(2)
i
][ = #
(2)
is even, too,
then {i. : .
(1)
i
= a
(1)
i
. .
(2)
}
= a
(2)
}
][ = #
(1,2)
is even, and s

1
(x
(1)
) = s

2
(x
(2)
) =
s

2
(x
(1)
. x
(2)
) = 1. Therefore, by (2.13)
s

2
(x
(1)
. x
(2)
)J(x
(1)
. x
(2)
) = J(x
(1)
. x
(2)
)
= J
1
(x
(1)
)J
2
(x
(2)
)
=
_
s

1
(x
(1)
)J
1
(x
(1)
)
_

_
s

2
J
2
(x
(2)
)
_
(2.14)
v If #
(1)
and #
(2)
are both odd, then #
(1,2)
is even, and so
s

2
(x
(1)
. x
(2)
)J(x
(1)
. x
(2)
) = J(x
(1)
. x
(2)
)
= J
1
(x
(1)
)J
2
(x
(2)
)
=
_
J
1
(x
(1)
)
_

_
J
2
(x
(2)
)
_
=
_
s

1
(x
(1)
)J
1
(x
(1)
)
_

_
s

2
J
2
(x
(2)
)
_
(2.15)
v If one of the #
(1)
. #
(2)
is even, and the other is odd, then #
(1,2)
is odd. (2.14)
holds in this case.
Hence, for any (x
(1)
. x
(2)
)
1

2
, (2.14) hold. Therefore, we have

(x
.1/
,x
.2/
)V (
1

2
)
s

2
(x
(1)
. x
(2)
)J
_
x
(1)
. x
2
_
=

(x
.1/
,x
.2/
)V (
1

2
)
_
s

1
(x
(1)
)J
1
(x
(1)
)
_

_
s

2
(x
(2)
)J
2
(x
(2)
)
_
=

x
.1/
V(
1
)
_
s

1
(x
(1)
)J
1
(x
(1)
)
_

_
_
_

x
.2/
V(
2
)
s

2
(x
(2)
)J
2
(x
(2)
)
_

_
=

x
.1/
V(
1
)
_
s

1
(x
(1)
)J
1
(x
(1)
)
_
^
T
2
(
2
)
= ^
T
1
(
1
) ^
T
2
(
2
).
(2.16)
Now suppose the claim holds for n = k, and consider n = k 1. In this case,
SECTION 2.3 A CLASS OF MEASURES 51
J(x
(1)
. . . . . x
(k)
. x
(k1)
) =
_
_
k

i=1
J
i
(x
(i)
)
_
_
J
k1
(x
(k1)
).
Just like Step 2, we have
s

kC1
iD1

i
(x
(1)
. . . . . x
(k1)
)J(x
(1)
. . . . . x
(k1)
)
=
_
_
s

k
iD1

i
(x
(1)
. . . . . x
(k)
)
k

i=1
J
i
(x
(i)
)
_
_

_
s

kC1
(x
(k1)
)J
k1
(x
(k1)
)
_ (2.17)
for every (x
(1)
. . . . . x
(k1)
)

k1
i=1

i
. Therefore,
^
T
(
1

k1
)
=

(x
.1/
,...,x
.kC1/
)V
_

kC1
iD1

i
_
s

kC1
iD1

i
(x
(1)
. . . . . x
(k1)
)J(x
(1)
. . . . . x
(k1)
)
=

(x
.1/
,...,x
.kC1/
)V
_

kC1
iD1

i
_
_
_
s

k
iD1

i
_
x
(1)
. . . . . x
(k)
_

i=1
J
i
(x
(i)
)
_
_

_
s

kC1
(x
(k1)
)J
k1
(x
(k1)
)
_
=

(x
.1/
,...,x
.k/
)V
_

k
iD1

i
_
_
_
s

k
iD1

i
_
x
(1)
. . . . . x
(k)
_

i=1
J
i
(x
(i)
)
_
_

_
_
_

x
.kC1/
V (
kC1)
s

kC1
(x
(k1)
)J
k1
(x
(k1)
)
_

_
=
_
_
k

i=1
^
T
i
(x
(i)
)
_
_
^
T
kC1
(x
(k1)
)
=
k1

i=1
^
T
i
(x
(i)
).
Since J
i
S
k
i
, we have ^
T
i
(
i
) > 0; thus, ^
T
(
1

n
) =

n
i=1
^
T
i
(
i
) >
0. The continuity of J(x
(1)
. . . . . x
(k1)
) is obvious. Hence,
J(x
(1)
. . . . . x
(n)
) S

n
iD1
k
i
. |L
3
EXTENSIONS OF MEASURES
Remark 3.1 (p. 82). If z(P), then G

is a z-system.
Proof. Let
G

= {C _ : C z(P)] . (3.1)
(z
1
) = z(P) == G

. (z
2
0 ) Suppose C
1
_ C
2
with C
1
. C
2
G.
Then we have C
1
_ C
2
and C
1
. C
2
z(P) by assumption. Since
every z-system is closed under proper dierences, we have
( C
2
) ( C
1
) = (C
2
C
1
) z(P) .
so C
2
C
1
G

. (z
3
) Let {C
n
]
o
n=1
denote a disjoint collection of G

-sets, so that
{ C
n
]
o
n=1
is a disjoint sequence of z(P)-sets. Since z(P) is a z-system and
hence satises (z3), we have

_
_
o
_
n=1
C
n
_
_
=
o
_
n=1
( C
n
) z(P) .
so that
_
o
n=1
C
n
G

. |L
Remark 3.2 (p. 90). 1^Q = 1^S ==1^1 = Q^S.
Proof. First observe that ^T = i = T. To see this, note that
= ^T = ( T) L (T ) == T = T = | .
but
1
T = == _ T.
T = ==T _ :
Thus = T.
For the reverse inclusion, let = T. Then
1
The proof is as follows: Let ~B = B
c
= . Let x . Then x B
c
==x B.
53
54 CHAPTER 3 EXTENSIONS OF MEASURES
^T = ( T) L (T ) = L = .
Now we prove the claim. Since 1^Q = 1^S, we have
_
1^Q
_
^(1^S) = . (3.2)
It follows from Exercise 2.22(g) that we can rewrite (3.2) as (1^1) ^
_
Q^S
_
=
, and which gives the result: 1^1 = Q^S. |L
3.1 Extensions and Restrictions
I Exercise 3.3 (3.1.1). Let (. A. j) denote a measure space. Pick 1 A and
dene A
T
= {J A: J _ 1]. Then A
T
is a o-eld on 1, A
T
= { 1: A],
and the restriction j
T
of j [from A] to A
T
is a measure. That is,
_
1. A
T
. j
T
_
is
a measure space and j
T
= j on A
T
.
Proof. Automatically, 1 A
T
. If A
T
, then A and _ 1; hence
1 A
T
. If {
n
]
o
n=1
_ A
T
, then
_
o
n=1

n
_ 1 and
_
o
n=1

n
A, i.e.,
_
o
n=1

n
A
T
. Therefore, A
T
is a o-eld.
We rst show
{J A: J _ 1] = A
T
_ A
t
T
= { 1: A] .
If J A
T
then J A and J _ 1. Since J = J 1, we get J A
t
T
. For the
converse inclusion direction, let T A
t
T
. Then there exists A such that
1 = T. It is obvious that 1 A and 1 _ 1, so 1 = T A
T
.
j
T
is a measure [on A
T
] because j is a measure [on A]. [See Exercise 3.4(b).]
|L
I Exercise 3.4 (3.1.2). Prove Claim 1 and 2.
Claim 1 Assume the notation of the denition. If j is o-nite on G, then v is
o-nite on G as well.
Claim 2 Suppose that G, and let G _ H _ 2

. Let v : H

R denote a
measure. Then the restriction of v to G is a measure.
Proof. (Claim1) By denition, j is the restriction of v [fromH] to G, so v() =
j() for all G. Since j is o-nite on G, there exists a sequence of G-sets,
{
n
]
o
n=1
, such that =
_
o
n=1

n
and v(
n
) = j(
n
) < o for each n N.
Hence, v is o-nite on G.
(Claim 2) Since j is the restriction of v fromH to G, we have j() = v() for
all G. (M1) To see the nonnegativity, let G _ H. Since v is a measure,
v(T) > 0 for all T H; particularly, j() = v() > 0 for all G. (M2)
j() = v() = 0. (M3) Let
_
o
n=1

n
G. Then
SECTION 3.2 OUTER MEASURES 55
j
_
_
o
_
n=1

n
_
_
= v
_
_
o
_
n=1

n
_
_
=
o

n=1
v(
n
) =
o

n=1
j(
n
). |L
3.2 Outer Measures
IExercise 3.5 (3.2.1). Let (. A. j) denote an arbitrary measure space. Dene
v on 2

by writing v(T) = inf {j(): T _ . A] for each T _ . Then v is


an outer measure.
Proof. The nonnegativity of v is evident since j() is a measure. To see (O2),
observe that _ , so v() 6 j() = 0. By (O1), v() = 0. To see (O3), let
T _ C _ . Hence {j(): C _ . A] _ {j(): T _ . A], which means
that inf {j(): C _ . A] > inf {j(): T _ . A], and so v(T) 6 v(C).
To see v is countable subadditivity, let {T
n
]
o
n=1
_ 2

. We just consider the


case that v(T
n
) < o for all n N. For each n, there exists c > 0 and
n
A
such that T
n
_
n
and
v(T
n
) c,2
n
> j(
n
).
Also
_
o
n=1
T
n
_
_
o
n=1

n
. Thus,
v
_
_
o
_
n=1
T
n
_
_
6 j
_
_
o
_
n=1

n
_
_
6
o

n=1
j(
n
) 6
o

n=1
_
_
v(T
n
)
c
2
n
_
_
=
o

n=1
v(T
n
) c. |L
I Exercise 3.6 (3.2.2). Let v : 2


R be an outer measure, and suppose in
addition that v is nitely additive: v( L T) = v() v(T), where . T _ are
disjoint. Then v is a measure. That is, (. 2

. v) is a measure space.
Proof. (M1) and (M2) are satised automatically. To see (M3) (countable addi-
tivity), let {
n
]
o
n=1
_ 2

be disjoint. Then
v
_
_
o
_
n=1

n
_
_
> v
_
_
1
_
n=1

n
_
_
=
1

n=1
v(
n
).
for every N N. Now let N ] o and yield
v
_
_
o
_
n=1

n
_
_
>
o

n=1
v(
n
). (3.3)
Combining (3.3) with (O4) (countable subadditivity) yields the result. |L
I Exercise 3.7 (3.2.3). Suppose that and are outer measures [relative to
some common background set ], and suppose that we dene a new function
v : 2


R for all _ by writing v() = max {(). ()]. Then v is an outer
measure [relative to ].
56 CHAPTER 3 EXTENSIONS OF MEASURES
Proof. (O1) and (O2) are straightforward. To see (O3), let _ T _ . Then
v() = max {(). ()] 6 max {(T). (T)] = v(T).
To see (O4), let {
n
]
o
n=1
_ 2

. Then
v
_
_
o
_
n=1

n
_
_
= max
_

_
_
o
_
n=1

n
_
_
.
_
_
o
_
n=1

n
_
_
_

_
6 max
_
_
_
o

n=1
(
n
).
o

n=1
(
n
)
_
_
_
6
o

n=1
max {(
n
). (
n
)]
=
o

n=1
v(
n
). |L
I Exercise 3.8 (3.2.4). Let v denote an outer measure, and let _ . Dene a
new set function v

on 2

by writing v

(T) = v(T ) for each T _ . Then


v

is an outer measure relative to .


Proof. (O1) and (O2) are satised automatically. If T _ C _ , then
v

(T) = v(T ) 6 v (C ) = v

(C)
by the monotonicity of v. To see (O4), let {T
n
] _ 2

. Then
v

_
_
o
_
n=1
T
n
_
_
= v
_
_
_
_
_
o
_
n=1
T
n
_
_

_
_
_ = v
_
_
o
_
n=1
(T
n
)
_
_
6
o

n=1
v(T
n
)
=
o

n=1
v

(T
n
). |L
I Exercise 3.9 (3.2.5). Let {v
n
]
o
n=1
denote a sequence of outer measures [rel-
ative to some common ], and let {a
n
]
o
n=1
denote a sequence of nonnegative
numbers. For each _ , let v() =

o
n=1
a
n
v
n
(). Then v is an outer mea-
sure relative to .
Proof. (O1) and (O2) are satised obviously. If _ T _ , then
v() =
o

n=1
a
n
v
n
() 6
o

n=1
a
n
v
n
(T) = v(T).
To see (O4), let {
k
] _ 2

. Then
SECTION 3.3 CARATHODORYS CRITERION 57
v
_
_
o
_
k=1

k
_
_
=
o

n=1
a
n
v
n
_
_
o
_
k=1

k
_
_
6
o

n=1
a
n
_
_
o

k=1
v
n
(
k
)
_
_
=
o

k=1
o

n=1
a
n
v
n
(
k
)
=
o

k=1
v(
k
). |L
3.3 Carathodorys Criterion
I Exercise 3.10 (3.3.1). Show directly that if . T M(v), then LT. T
M(v).
Proof. (i) The following method is from Bear (2002). Let . T M(v) and let
T _ be any test set. Let T = T
1
L T
2
L T
3
L T
4
as indicated in Figure 3.1. We
need to show
v (T ) = v
_
T ( L T)
_
v
_
T ( L T)
c
_
. (3.4)
or, in terms of Figure 3.1,
v(T ) = v(T
1
L T
2
L T
3
) v(T
4
). (3.4
t
)
T
1
T
2
T
3
T
4
T
Figure 3.1. LB M()
Cutting the test set T
1
L T
2
with T gives
v(T
1
L T
2
) = v(T
2
) v(T
1
). (3.5)
Similarly, cutting T
3
L T
4
with T gives
v (T
3
L T
4
) = v(T
3
) v(T
4
). (3.6)
58 CHAPTER 3 EXTENSIONS OF MEASURES
Cutting T with gives
v(T ) = v(T
1
L T
2
) v (T
3
L T
4
) . (3.7)
Combining (3.5), (3.6), and (3.7) we can write
v(T ) = v(T
1
) v(T
2
) v(T
3
) v(T
4
). (3.8)
Now cut T
1
L T
2
L T
3
with and then use (3.5):
v(T
1
L T
2
L T
3
) = v(T
1
L T
2
) v(T
3
) = v(T
1
) v(T
2
) v(T
3
). (3.9)
From (3.9) and (3.8) we have the desired equality (3.4
t
).
(ii) It is clear that M(v) i
c
M(v). Thus, . T M(v) implies that
T L
c
M(v) by the previous result. Since T = (T L
c
)
c
M(v), we get
T M(v). |L
I Exercise 3.11 (3.3.2). Suppose that may be written as
_
o
n=1

n
, where
{
n
]
o
n=1
is a nondecreasing sequence of subsets of . If _ is such that

k
M(v) for all k exceeding some constant k

, then M(v).
Proof. By the Outer Measure Theorem, M(v) is a o-eld on . Therefore,
o
_
k=k
A
1
(
k
) =
_
_
o
_
k=k
A
1

k
_
_
M(v).
But since {
k
]
o
k=1
is a nondecreasing sequence, we have
o
_
k=k
A
1

k
=
o
_
k=1

k
= .
which means that

_
_
o
_
k=k
A
1

k
_
_
= = M(v). |L
I Exercise 3.12 (3.3.3). Let v denote an outer measure such that v() < o,
and further suppose that if _ with v() < o, then there exists T M(v)
such that _ T and v() = v(T). Then 1 M(v) i v() = v(1) v(1
c
).
Proof. If 1 M(v), then v(T ) = v(T 1)v (T 1
c
); in particular, this holds
for T = , so v() = v(1) v(1
c
).
For the other direction, suppose v() = v(1) v(1
c
). Since v() < o, we
get v(1). j(1
c
) < oby the monotonicity of v. Then there exist T
t
. T
tt
M(v)
such that 1
c
_ T
t
, v(1
c
) = v(T
t
), and 1 _ T
tt
, v(1) = v(T
tt
). Let T = (T
t
)
c

M(v). Then T _ 1 and


v(T) = v((T
t
)
c
) = v() v(T
t
) =
_
v(1) v(1
c
)
_
v(1
c
) = v(1).
SECTION 3.3 CARATHODORYS CRITERION 59
Hence, there exist T. T
tt
M(v) such that T _ 1 _ T
tt
, and v(T) = v(1) =
v(T
tt
).
Notice that 1 is the union of T and a subset of T
tt
T. If we can show that
every subset of T
tt
T is in M(v), then 1 M(v) (since M(v) is a o-eld). For
every C _ T
tt
T, we have
v(C) 6 v(T
tt
T) = v(T
tt
) v(T) = 0.
Therefore, v(C) = 0, i.e., C M(v), and so 1 M(v). |L
I Exercise 3.13 (3.3.4). Suppose that v is an arbitrary outer measure, and let
. T _ with M(v). Show that v( L T) v ( T) = v() v(T).
Proof. M(v) implies that
v( L T) = v
_
( L T)
_
v
_
( L T)
c
_
= v() v
_
T
c
_
. (3.10)
and
v(T) = v(T ) v
_
T
c
_
. (3.11)
Combining (3.10) with (3.11) we get v( L T) v ( T) = v() v(T). |L
I Exercise 3.14 (3.3.5). Let v denote an outer measure, and let {
n
]
o
n=1
denote
a nondecreasing sequence of M(v)-sets. Show that v(lim(
n
)) = limv(

n
) for any _ . State and prove an analogous result for nonincerasing
sequences of M(v)-sets.
Proof. Let {
n
]
o
n=1
be a nondecreasing sequence of M(v)-sets. Then {
n
]
forms a nondecreasing sequence, so lim(
n
) =
_
(
n
) = (
_

n
).
Let T
1
=
1
and T
n
=
n

n-1
for n > 2. Then {T
n
] _ M(v) is disjoint and
_

n
=
_
T
n
. Thus
v
_
lim(
n
)
_
= v
_

_
_
T
n
_
_
=
o

n=1
v( T
n
)
= lim
n
n

i=1
v( T
n
)
= lim
n
v
_
_
_
_
_
n
_
i=1
T
i
_
_
_
_
_
= lim
n
v(
n
).
If {
n
] is a nonincreasing sequence of M(v) sets, then {
c
n
] is a nondecreas-
ing sequence of M(v) sets. Thus v(lim(
n
)) = limv(
n
). |L
I Exercise 3.15 (3.3.6). Let v denote an outer measure such that the following
holds: if _ with v() < o, then there is T M(v) with _ T and
60 CHAPTER 3 EXTENSIONS OF MEASURES
v() = v(T). Then, for any nondecreasing sequence {
n
]
o
n=1
of subsets of , we
have v(lim
n
) = limv(
n
).
Proof. If there exists
k
such that v(
k
) = o, then v(lim
n
) = v(
_

n
) >
v(
k
) = o, and so v(lim
n
) = o; on the other hand, v (
1
) 6 v (
2
) 6
and v(
k
) = o imply that limv(
n
) = o.
Now let v(
n
) < o for all n N. Then there exists T
n
M(v) such that

n
_ T
n
and v(
n
) = v(T
n
) for each n N. We rst show that we can choose
{T
n
] so that it is nonincreasing.
Consider T
n
and T
n1
. If T
n
T
n1
, then v(T
n
) = v(T
n1
) since v(T
n
) =
v(
n
) 6 v(
n1
) = v(T
n1
) always holds. But then v(
n
) = v(
n1
) and so we
can just let T
n1
= T
n
after we having chosen T
n
.
Thus, v(limT
n
) = limv(T
n
) exists. Since {
n
]
o
n=1
is nondecreasing, we
have v(
_
n
i=1

i
) = v(
n
) = v(T
n
) for all n N. Take the limit and we get
v
__
o
i=1

i
_
= v(limT
n
) = limv(T
n
) = limv(
n
). |L
I Exercise 3.16 (3.3.7). In each of the following parts, (i) describe the outer
measure j
+
on 2

induced by the given j, (ii) describe the collection M(j


+
) and
determine if M(j
+
) is a o-eld, and (iii) check to see whether j
+
= j on the
given collection A.
a. = {1. 2. 3], A =

. {1] . {2. 3] .
_
, and j is a measure on A such that
j() = 1 and j
_
{1]
_
= 0.
Solution. (a) Since j is a measure on A, by (nite) additivity, j() = 0 and
j
_
{2. 3]
_
= j() j
_
{1]
_
= 1. Then, j
+
() = 0, j
+
_
{1]
_
= 0, j
+
_
{2. 3]
_
= 1,
j
+
() = 1, j
+
_
{2]
_
= j
_
{2. 3]
_
= 1, j
+
(3) = j
_
{2. 3]
_
= 1, j
+
_
{1. 2]
_
=
j
+
_
{1. 3]
_
= j() = 1. |L
3.4 Existence of Extensions
I Exercise 3.17 (3.4.1). Let k N and refer to the measure z
k
: B
k


R that
assigns the value

k
i=1
(b
i
a
i
) to every (a. b| A
k
, as given in this sections
example. For this measure z
k
, we have the following;
a. z
k
_
(a. b|
_
= z
k
_
(a. b)
_
= z
k
_
a. b)
_
= z
k
_
a. b|
_
=

k
i=1
(b
i
a
i
).
b. z
k
applies to any k-dimensional rectangle that contains a k-dimensional open
set and is unbounded in at least one dimension gives o.
c. z
k
applied to any bounded B
k
-set yields a nite number.
d. It might be thought that if _ T with T B
k
with z
k
(T) = 0, then z
k
()
must exists and equal 0. Show that if there exists an uncountable set C B
k
with z
k
(C) = 0, then z
k
() need not even exist, let along equal zero, and thus
z
k
violates our intuition in this regard.
SECTION 3.4 EXISTENCE OF EXTENSIONS 61
Proof. (a) Observe rst that {x] = lim
n
_
x 1,n. x
_
for any x R
k
. Therefore,
z
k
_
{x]
_
= z
k
_
lim
n
_
x 1,n. x
_
_
= lim
n
z
k
_
x 1,n. x
_
= lim
n
1,n
k
= 0.
Now by (M5),
z
k
(a. b) = z
k
_
(a. b| {b]
_
= z
k
(a. b| z
k
_
{b]
_
= z
k
(a. b| .
(b) Write (a. b| = (a
1
. b
1
| (a
k
. b
k
|, and assume that b
1
a
1
= o. Since (a. b|
contains an open set, b
i
a
i
> 0 for each i = 1. . . . . k. Therefore, z(a. b| = o.
(c) Let B
k
be bounded. Then there exists a bounded (a. b| containing .
Hence z
k
() 6 z
k
(a. b| < o.
(d) By the Continuum Hypothesis, if C is countable then[C[ > c. Hence

2
C

>
2
c
> c. However, [B
k
[ = c. |L
I Exercise 3.18 (3.4.2). This problem reviews the Extension Theorem.
a. Where or how is the fact that j() = 0 used?
b. What happens if A = 2

?
Solution. (a) M
_
j
_
and (b) The Extension Theorem holds if and only if j
is a measure on A. |L
IExercise 3.19 (3.4.3). Consider the Extension Theoremframework. If we have
j() < ofor each A, it might not be the case that the measure extension
j
+
c(A)
assigns nite measure to every set in o(A). However, if j is o-nite on A,
then the measure extension j
+
c(A)
is o-nite on o(A), and the measure extension
j
+
M(

)
is o-nite on M(j
+
).
Proof. Let = R and A consist of and all bounded rsc intervals (a. b|. Let
j(a. b| = b a for all (a. b| A. Then j() < o for each A. However,
R o (A), and j
+
c(A)
(R) = o. The other two claims are obvious. |L
I Exercise 3.20 (3.4.4). There is a measure v : B

R with v(R) = 1 and
v(a. b| =
_
b
o
(2)
-1{2
e
-z
2
{2
d:.
where the integral is the familiar Riemann integral from calculus.
Proof. According to the approach of Georgakis (1994), let , = :s, d, = : ds,
then
62 CHAPTER 3 EXTENSIONS OF MEASURES
_
_
b
o
e
-z
2
{2
d:
_
2
=
_
b
o
_
_
b
o
e
-(z
2
,
2
){2
d,
_
d:
=
_
b
o
_
_
b
o
e
-z
2
(1x
2
){2
: ds
_
d:
=
_
b
o
_
_
b
o
e
-z
2
(1x
2
){2
: d:
_
ds
=
_
b
o
_
_
1

_
1 s
2
_e
-z
2
(1x
2
){2

b
o
_
_
ds
=
_
b
o
1
1 s
2
e
-o
2
(1x
2
){2
ds
_
b
o
1
1 s
2
e
-b
2
(1x
2
){2
ds. |L
I Exercise 3.21 (3.4.5). Consider the Extension Theorem framework again.
a. If _ T _ with T M(j
+
) and j
+
M(

)
(T) = 0, then M(v
+
) and
j
+
M(

)
() = 0.
b. If in (a) we replace every occurrence of M(j
+
) [including instances where it
appears as a subscript] with o(A), then the claim is not necessarily true.
Proof. (a) Assume the hypotheses. To see M(j
+
), note that for any T _
with j
+
(T ) < o, we have j
+
(T ) 6 j
+
(T T) 6 j
+
(T) = 0 by monotonicity
of j
+
. Therefore, j
+
(T ) > j
+
(T ) j
+
(T
c
) = j
+
(T
c
) always holds.
(b) The same reason as in Exercise 3.17(d). |L
I Exercise 3.22 (3.4.6). Let F denote a eld on , and let j: F

R denote a
measure. Let j
+
: 2


R be given by
j
+
() = inf
_
_
_
o

n=1
j(
n
): {
n
] is a F -covering of
_
_
_
. _ .
Then j
+
is an outer measure, and the restriction of j
+
to the o-eld M(j
+
) is a
measure. With these facts, we have that j
+
= j on F and F _ M(j
+
). Finally,
there exists a measure extension of j to o(F ).
Proof. Notice that j() = 0 and j() > 0 for all F since j is a measure
on F . So j
+
is an outer measure by Example 1 in Section 3.2. The other parts
are standard. |L
I Exercise 3.23 (3.4.7). Let F denote a eld on . Suppose that v : F

R is
such that (i) v() > 0 for all F , (ii) v is nitely additive, and (iii) if {
n
]
o
n=1
is
a nonincreasing sequence of F -sets with lim
n
= , the limv(
n
) = 0. Dene
v
+
: 2


R for all _ by writing
v
+
() = inf
_
_
_
o

n=1
v(
n
): {
n
]
o
n=1
is an F -covering of
_
_
_
.
SECTION 3.5 UNIQUENESS OF MEASURES AND EXTENSIONS 63
a. v
+
is an outer measure.
b. M(v
+
) is a o-eld on .
c. The restriction of v
+
to M(v
+
) is a measure on M(v
+
).
d. F _ M(v
+
).
e. There exists a measure extension of v to o (F ).
Proof. (a) It suces to show that v () = 0 by Example 1 of Section 3.2. Take
a sequence {. . . . .]. Then 0 = limv () = v().
(b) (e) are from the Outer Measure Theorem. |L
3.5 Uniqueness of Measures and Extensions
I Exercise 3.24 (3.5.1). If j
1
and j
2
are nite measures with domain o(P)
(where P denotes a -system on ), if can be expressed as an amc union of
P-sets, and if j
1
= j
2
on P, then j
1
= j
2
on o(P).
Proof. Assume the hypotheses. Then o
1
is o-nite with respect to P and j
1
=
j
2
on P. By the Uniqueness Theorem, j
1
= j
2
on o(P). |L
I Exercise 3.25 (3.5.2). Let j
1
and j
2
denote nite measures with domain
o(P), where P is a -system on , and further suppose that j
1
= j
2
on P.
Then j
1
= j
2
.
Proof. I am not sure about this exercise. If P, then by letting
L = { o(P): j
1
() = j
2
()] .
we can easily to show that L is a z-system with P _ L. Then the result is
trivial. |L
I Exercise 3.26 (3.5.3). Let = {o
1
. o
2
. o
3
. o
4
], let A consist of , {o
1
. o
2
],
{o
1
. o
3
], {o
2
. o
4
], {o
3
. o
4
], and , and let j: A R be dened as follows:
j() = 6, j() = 0, and j({o
1
. o
2
]) = j({o
1
. o
3
]) = j({o
2
. o
4
]) = j({o
3
. o
4
]) =
3.
a. A is neither a -system nor a semiring, and o(A) = 2

.
b. j is a measure.
c. Dene two new distinct measures v and on the o-eld 2

by the following:
v({o
1
]) = v({o
4
]) = 1, v({o
2
]) = v({o
3
]) = 2, ({o
2
]) = ({o
3
]) = 1, and
({o
1
]) = ({o
4
]) = 2. Then v and are distinct measure extensions of j
from A to o(A) = 2

.
64 CHAPTER 3 EXTENSIONS OF MEASURES
d. Let j
+
: 2

R denote the outer measure induced by j. Then j


+
= j on A
and j
+
is a measure.
e. j
+
, v, and are distinct measures.
Proof. (a) A is not a -system because {o
1
. o
2
] {o
1
. o
3
] = {o
1
] A, and so
A is not a semiring. o(A) = 2

since every singleton can be expressed as a


intersection of A-sets.
(b) Easy to check.
(c) For example, v({o
1
. o]) = v({o
1
]) v({o
2
]) = 1 2 = j({o
1
. o
2
]).
(d) For example, j
+
({o]) = inf{j({o
1
. o
2
]). j({o
1
. o
3
])] = 3.
(e) Trivial to see that they are distinct. |L
I Exercise 3.27 (3.5.5). We assume the setup of Exercise 3.22. The aim of this
exercise is to show that if v is a measure with domain o(F ) such that v = j on
the eld F , then v coincides with the measure extension of j to o(F ) guaranteed
by Exercise 3.22.
a. If T o(F ), then v(T) 6 j
+
(T).
b. If J o(F ) and j
+
(J) < o, then v(J) = j
+
(J).
c. If j is o-nite on F , then v(1) = j
+
(1) for all 1 o(F ). This gives the
uniqueness of the measure extension whose existence is guaranteed by Exer-
cise 3.22.
Proof. (a) Let F
c
denote the family of all countable unions of F -sets. If
F
c
, i.e., =
_

n
with
n
F for all n, then, by letting T
1
=
1
and T
n
=

n
(
_
i6n

i
) for n > 2, we can rewrite as a disjoint union of F -sets {T
n
].
Thus,
v() = v
_
_
T
n
_
=

v(T
n
) =

j(T
n
) =

j
+
(T
n
) = j
+
_
_
T
n
_
= j
+
().
Now take an arbitrary T o(F ), and we show that
j
+
(T) = inf
_

j(
n
): {
n
] is an F -covering of T
_
= inf

j
+
(): T _ F
c
_
dened as .
Firstly, T _ implies that j
+
(T) 6 j
+
() and so j
+
(T) 6 . Secondly, for all
F
c
, there exists {
n
] _ F
c
such that =
_

n
, and so we get j
+
() =
j
+
(
_

n
) 6

j
+
(
n
) =

j(
n
); thus, 6 j
+
(T). Therefore,
j
+
(T) = inf

j
+
(): T _ F
c
_
= inf {v(): T _ F
c
] > v(T).
SECTION 3.5 UNIQUENESS OF MEASURES AND EXTENSIONS 65
(b) It suces to showthat v(J) > j
+
(J) by part (a). Since j
+
(J) = inf{j
+
(): J _
F
c
], for a given c > 0, there exists C J
c
with J _ C, such that
j
+
(J) c > j
+
(C).
Hence,
j
+
(J) 6 j
+
(C) = v(C) = v(J) v(C J) 6 v(J) j
+
(C J)
= v(J) j
+
(C) j
+
(J)
< j(J) c.
Since c > 0 is arbitrary, we get j
+
(J) 6 v(J) whenever J o(F ) and j
+
(J) <
o.
(c) If j is o-nite on F , then there exists {
n
] _ F such that =
_

n
and
j(
n
) < o for all n. Without loss of generality, we can assume that {
n
] is
disjoint F -sets. Then by (b), for every 1 o(F ) we get
v(1) = v
_
_
(1
n
)
_
=

v(1
n
) =

j
+
(1
n
) = j
+
(1). |L
I Exercise 3.28 (3.5.6). A lattice on is a collection L _ 2

such that (i)


L, (ii) L, (iii) L is closed under (nite) unions, and (iv) L is a -system.
We also dene the following two collections: D = {T : . T L. _ T], and
U will denote the collection of all nite disjoint unions of D-sets.
a. D is a -system.
b. U is a -system.
c. U is closed under complementation.
d. U coincides with the minimal eld containing the lattice L.
e. Let A denote a o-eld on that contains L. Suppose that j and v are mea-
sures with domain A such that j = v on L. Furthermore, suppose that
=
_

n
, where
n
L and j(
n
) < o for each n N. Then j = v
on o(L).
Proof. (a) Write D
i
= T
i

i
with
i
. T
i
L and
i
_ T
i
, for i = 1. 2. Then
D
1
D
2
= (T
1

1
) (T
2

2
)
= (T
1

c
1
) (T
2

c
2
)
= (T
1
T
2
) (
1
L
2
)
= (T
1
T
2
)
_
(T
1
T
2
) (
1
L
2
)
_
D.
66 CHAPTER 3 EXTENSIONS OF MEASURES
(b) Let U
1
= D
1
L L D
n
and U
2
= 1
1
L L 1
n
, where D
1
. . . . . D
n
D are
disjoint, and 1
1
. . . . . 1
n
D disjoint. Then
U
1
U
2
=
n
_
i=1
n
_
}=1
(D
i
1
}
).
Since D
i
1
}
D (D is a -system), U
1
U
2
is a nite disjoint unions of
D-sets, and so is in U.
(c) Pick an arbitrary U U. Then there exists disjoint D
1
. . . . . D
n
D such
that U =
_
n
i=1
D
i
. If n = 1, then U = D
1
= T
1

1
, where
1
. T
1
L and

1
_ T
1
. Thus, U
c
= ( ) L ( T) U. Let us assume that U
c
U when
U = D
1
L L D
n
and consider n 1. Then
_
_
n1
_
i=1
D
i
_
_
c
=
_
_
n
_
i=1
D
i
_
_
c
D
c
n1
U.
since (
_
n
i=1
D
i
)
c
U by the induction hypothesis, D
c
n1
U as in the case of
n 1, and U is a -system.
(d) Notice that = U, so U is a eld by (b) and (c). If L, then
= U, so L _ U. Thus, (L) _ U. It is easy to see that U _ (L).
(e) Let =
_
o
n=1

n
with
n
L and j(
n
) < o for each
n
. Let U U;
then U = D
1
L L D
n
for some disjoint D
1
. . . . . D
n
D. Hence,
j(U) = j
_
_
o
_
n=1
(
n
U)
_
_
=
o

n=1
j(
n
U)
=
o

n=1
n

i=1
j(
n
D
i
)
|L
3.6 The Completion Theorem
I Exercise 3.29 (3.6.1). Let k > 2, and let z
k
: B
k


R denote the unique mea-
sure with domain B
k
that assigns the value

k
i=1
(b
i
a
i
) to each k-dimensional
rsc rectangle (a. b|. Let = {x R
k
: .
2
= = .
k
= 0]. Then B
k
, is
uncountable, and z
k
() = 0, hence

T
k

= 2
c
.
Proof. B
k
since = lim
n
(n. n| (1,n. 1,n|
k-1
|; is uncountable since
R is uncountable; z
k
() = 0 since z
k
() = (o) 0 0 = 0.
Since B
k
is complete (the completion of B
k
), and B
k
_ B
k
, we know
that every subset of is in B
k
. There are 2
c
subsets of , so

B
k

> 2
c
; on
SECTION 3.7 THE RELATIONSHIP BETWEEN c(A) AND M(

) 67
the other hand, there are 2
c
subsets on R, i.e.,

B
k

6 2
c
. It follows from the
Cantor-Bernstein theorem that

B
k

= 2
c
. |L
I Exercise 3.30 (3.6.2). Let (. F . v) denote a measure space. If . T F with
_ 1 _ T and v(T ) = 0, then 1

F and v(1) = v() = v(T).
Proof. We rst show 1

F . Since . T F , and F is a o-eld, we get T
F . Now we can write 1 as
1 = L
_
(T ) (T 1)
_
.
Since F , (T ) (T 1) _ T F , and v(T ) = 0, we have
(T ) (T 1) N
0
(v); thus 1

F .
To show v(1) = v() = v(T), we only need to show that v() = v(T) since
v(1) = v() by denition. If v() < o or v(T) < o, then 0 = v(T ) =
v(T) v() implies that v(T) = v(). If v() = o, then by the monotonicity
of a measure, v(T) > v() = o, and so v(T) = o= v() |L
I Exercise 3.31 (3.6.3). Let (. F . v) denote a measure space. Furthermore,
let F
1
denote a sub-o-eld of F . Then there exists a minimal o-eld F
2
such
that F
1
_ F
2
_ F and N(v) _ F
2
. Also, F
2
i there exists T F
1
with
^T N(v).
Proof. Let F
2
= o(F
1
. N(v)). It is clear that F
1
. N(v) _ F
2
. Since F
1
. N(v) _ F ,
we have that
F
2
= o(F
1
. N(v)) _ o(F ) = F .
We next show that (. F
2
. v) is complete (where v is restricted on F
2
). |L
3.7 The Relationship between .A/ and M.

/
I Exercise 3.32 (3.7.1). Let be uncountable, let A denote the o-led { _
: is amc or
c
is ams], and dene j: A

R by stipulating that j() de-
notes the number of points in if is nite and j() = o if is innite.
a. (. A. j) is a non-o-nite measure space.
b. (. A. j) is complete.
c. Letting j
+
denote the outer measure induced by j, the (o-eld) M(j
+
) coin-
cides with 2

.
Proof. (a) For every sequence {
n
] _ A with j(
n
) < o, i.e.,
n
is nite for
all n, their union
_
n

n
is amc. Hence, (. A. j) is non-o-nite.
(b) Let _ T _ with T A and j(T) = 0. Then T must be empty and so
= = T A.
68 CHAPTER 3 EXTENSIONS OF MEASURES
(c) Take an arbitrary T _ with j
+
(T ) < o; that is, T is nite. Then for
every subset _ , we have that T and T
c
are both nite and[T [ =
[T [ [T
c
[. Hence, j
+
(T ) = j
+
(T )j
+
(T
c
), i.e., M(j
+
). Thus,
M(j
+
) = 2

.
Note that here

A = A = M(j
+
) = 2

, so the o-niteness is essential. |L


3.8 Approximations
IExercise 3.33 (3.8.1). The assumption v(T) < ois not superuous in Claim
4, and the assumption v() < o is not superuous in Claim 6.
Proof. Let A be the semiring consisting of and all bounded rsc (a. b|. Let
us consider (R. B. z). For Claim 4, take T = R. It is evident that for any nite
disjoint A-sets (a
1
. b
1
|. . . . . (a
n
. b
n
|, we have z(R^
_
n
i=1
(a
i
. b
i
|) = o. For Claim
6, let = R. then for any bounded set 1 B, there exists (a. b| A containing
1, so z(1) 6 z(a. b| = b a < o= z(R). |L
I Exercise 3.34 (3.8.2). Let v : B
k


R be nonnegative and nitely additive
with v(R
k
) < o. Suppose that v() = sup{v(1): 1 _ . 1 compact] for each
B
k
. Then v is a nite measure.
Proof. It suces to show that v is countably additive. |L
3.9 A Further Description of M.

/
I Exercise 3.35 (3.9.1). Countable superadditivity: If
1
.
2
. . . . _ are dis-
joint, then j
+
(
_
o
n=1

n
) >

o
n=1
j
+
(
n
).
Proof. Fix an arbitrary c > 0. For every
n
, nd C
n
o(A) with C
n
_
n
such
that
j
+
c(A)
(C
n
) c,2
n
> j
+
(
n
).
Since
_
o
n=1
C
n
_
_
o
n=1

n
, we have
o

n=1
j
+
(
n
) <
o

n=1
j
+
c(A)
(C
n
) c = j
+
c(A)
_
_
o
_
n=1
C
n
_
_
c 6 j
+
_
_
o
_
n=1

n
_
_
c.
where the rst equality holds since {C
n
] is disjoint o(A)-sets. Since c > 0 is
arbitrary, we get the countable superadditivity. |L
I Exercise 3.36 (3.9.2). For any _ , j
+
() = inf{j
+
M(

)
(T): _ T
M(j
+
)] = j
++
().
Proof. Dene
SECTION 3.9 A FURTHER DESCRIPTION OF M(

) 69
C =
_
j
+
c(A)
(T): _ T o(A)
_
and D =
_
j
+
M(

)
(T): _ T M(j
+
)
_
.
It is clear that C _ D, so inf C > inf D. Next, pick J D. Then J must be of
the form j
+
M(

)
(T), where _ T M(j
+
). Write T = C LD, where C o(A),
D _ N and N is a j
+
c(A)
-null set. Thus, there exists C L N o(A) such that
_ T _ C L N and
j
+
c(A)
(C L N) 6 j
+
c(A)
(C) j
+
c(A)
(N) = j
+
M(

)
(C) 6 j
+
M(

)
(T).
Denote j
+
c(A)
(C LN) = c. Hence, for every J D, there exists c C with c 6 J.
It follows that inf C 6 inf D. |L
I Exercise 3.37 (3.9.3). For any _ , j
+
() = sup{j
+
M(

)
(T): T _ . T
M(j
+
)].
Proof. Dene
C =
_
j
+
c(A)
(T): T _ . T o(A)
_
. D =
_
j
+
M(

)
(T): T _ . T M(j
+
)
_
.
First, C _ D implies that supC 6 supD. Next, pick J D. Then J must be
of the form j
+
M(

)
(T), where T _ and T M(j
+
). Write T = C L D with
C o(A), D _ N, and N is a j
+
c(A)
-null set. Thus, there exists C o(A) such
that C _ T _ , and
j
+
c(A)
(C) = j
+
M(

)
(C L D) = j
+
M(

)
(T).
Denote c = j
+
c(A)
(C). So for every J D there exists c C such that c = J.
Therefore, supC > supD. |L
I Exercise 3.38 (3.9.4). For any _ , there is 1 o(A) such that 1 _ and
j
+
c(A)
(1) = j
+
().
Proof. For every n N, there exists 1
n
o(A) with 1
n
_ such that
j
+
c(A)
(1
n
) > j
+
() 1,n.
Let 1 =
_
o
n=1
1
n
. Then 1 o(A), 1 _ , and for all n we have
j
+
c(A)
(1) > j
+
c(A)
(
n
) > j
+
() 1,n:
hence, j
+
c(A)
(1) > j
+
(). Since j
+
c(A)
(1) 6 j
+
() holds, we get the result. |L
I Exercise 3.39 (3.9.5). The inmum that denes j
+
() and the supremum
that denes j
+
() are achieved for each _ . That is, there exist C. T o(A)
with C _ _ T and j
+
c(A)
(C) = j
+
() and j
+
c(A)
(T) = j
+
().
Proof. We have proved the inner measure in the previous exercise, so we
focus on the outer measure. For every n N, there exists T
n
o(A) with
70 CHAPTER 3 EXTENSIONS OF MEASURES
_ T
n
and j
+
() 1,n > j
+
c(A)
(T
n
). Let T =
_
o
n=1
T
n
. Then T o(A), _ T,
and for all n,
j
+
c(A)
(T) 6 j
+
c(A)
(T
n
) 6 j
+
() 1,n:
that is, j
+
c(A)
(T) 6 j
+
(). Since the other direction is clear, we get the result.
|L
I Exercise 3.40 (3.9.6). Let _ and let {
n
] denote a disjoint sequence of
M(j
+
)-sets. Then we have j
+
(
_
o
n=1
(
n
)) =

o
n=1
j
+
(
n
).
Proof. There exists 1 M(j
+
) with 1 _
_
o
n=1
(
n
) = (
_
o
n=1

n
) and
j
+
M(

)
(1) = j
+
(
_
o
n=1
(
n
)) by Exercise 3.38 (since o(A) = M(j
+
)). Thus,
j
+
_
_
o
_
n=1
(
n
)
_
_
= j
+
M(

)
(1) =
o

n=1
j
+
M(

)
(1
n
) 6
o

n=1
j
+
(
n
).
Then the desired result follows from Exercise 3.35. |L
I Exercise 3.41 (3.9.7). If . T _ are disjoint, then j
+
( L T) 6 j
+
()
j
+
(T) 6 j
+
( L T).
Proof. Let J o(A) with T _ J with j
+
(T) = j
+
c(A)
(J). Let 1 o(A) with
1 _ LT such that j
+
(LT) = j
+
c(A)
(1). Since 1J _ 1T _ (LT)T = ,
it follows that
j
+
( L T) = j
+
c(A)
(1) 6 j
+
c(A)
(1 J) j
+
c(A)
(J) 6 j
+
() j
+
(T).
Dually, let H o(A) with H _ and j
+
c(A)
(H) = j
+
(). Let G o(A) with
L T _ G and j
+
c(A)
(G) = j
+
( L T). Since T _ G H, it follows that
j
+
( L T) = j
+
c(A)
(G) = j
+
c(A)
(H) j
+
c(A)
(G H) > j
+
() j
+
(T). |L
I Exercise 3.42 (3.9.8). If M(j
+
) and T _ , then j
+
M(

)
() = j
+
(T
) j
+
(T
c
).
Proof. Applying Exercise 3.41 to T and T
c
, we obtain
j
+
() 6 j
+
(T ) j
+
(T
c
) 6 j
+
().
Since M(j
+
), we have j
+
() = j
+
() = j
+
M(

)
(), and thus we get the
result. |L
4
LEBESGUE MEASURE
4.1 Lebesgue Measure: Existence and Uniqueness
I Exercise 4.1 (4.1.1). Let . R and k > 2. Then z({.]) = z
k
({.] R
k-1
) = 0.
Next, for any {1. . . . . k 1] and x R
}
we have z
k
({x] R
k-}
) = 0.
Proof. Since the sequence {{.] (n. n|
k-1
] is increasing and converges to
{.] R
n-1
, we have
z
k
({.] R
k-1
) = z
k
(lim
n
{.] (n. n|
k-1
) = 0 = z({.]).
The other claim can be proved in the same way. |L
I Exercise 4.2 (1.4.2). Enumerate the rationals in (0. 1| by {q
1
. q
2
. . . .]. Given
arbitrarily small c > 0, remove the interval
n
= (q
n
c,2
n1
. q
n
c,2
n1
)(0. 1|.
Let =
_
o
n=1

n
. Then

z() 6 c, despite the fact that is an open dense subset
of (0. 1|. Also, we have

z((0. 1| ) > 1 c, even though (0. 1| is a nowhere
dense subset of (0. 1|.
Proof. For every
n
we have 0 <

z(
n
) 6

z(q
n
c,2
n1
. q
n
c,2
n1
) = c,2
n
;
hence,

z() =

z
_
_
o
_
n=1

n
_
_
6
o

n=1
z (
n
) 6
o

n=1
c
2
n
= c.
and so

z((0. 1| ) = 1

z() > 1 c. |L
IExercise 4.3 (4.1.3). There cannot exist a closed subset of (0. 1| whose interior
is empty, yet has

z-measure of one.
Proof. |L
IExercise 4.4 (4.14). z
k
is nonatomic: any B
k
with z
k
() > 0 has a proper
subset T B
k
with 0 < z
k
(T) < z
k
(). This forces

z
k
to be nonatomic as well.
71
72 CHAPTER 4 LEBESGUE MEASURE
Proof. Take any L
k
with

z
k
() > 0. Since L
k
= B
k
, the completion of B
k
,
there exists C. M B
k
, where M is a subset of z
k
-null set, such that = CLT.
Therefore, z
k
(T) =

z
k
() > 0. Since z
k
is nonatomic, there exists T B
k
with
0 < z
k
(T) < z
k
(C) =

z
k
(), that is,

z
k
is nonatomic. |L
I Exercise 4.5 (4.1.5). Let k > 2. There exists an uncountable set U B
k
with
z
k
(U) = 0.
Proof. Let U = {.]R, where . R. Then U is uncountable, and z(U) = 0. |L
I Exercise 4.6 (4.1.6). [L
k
[ = 2
c
and [B
k
[ = c for each k N.
Proof. Since [R
k
[ = c and L
k
_ 2
R
k
, we rst have [L
k
[ 6 2
c
; on the other hand,
there exists an uncountable set U _ R
k
such that z
k
(U) = 0 (if k = 1, consider
the Cantor set; if k > 2, consider the set U in the preceding exercise), so [U[ = c
by the Continuum Hypothesis. Since L
k
is complete, we have [L
k
[ > [2
U
[ = 2
c
.
It follows from the Cantor-Bernstein Theorem that [L
k
[ = 2
c
. |L
I Exercise 4.7 (4.1.7). Assume that L = 2
R
; in particular every one of the 2
c
subsets of 0. 1| is a Lebesgue set. Let T = {

z(): _ 0. 1|.

z() ]. Consider-
ation of the set T (which is in L be assumption) leads to contradiction.
Proof. T _ 0. 1| so

z(T) exists. We now have a contradiction:

z(T) T i

z(T) T. |L
I Exercise 4.8 (4.1.8). Let k > 2. Every line is in B
k
, has z
k
-measure zero, and
hence has

z
k
-measure.
Proof. Let be the line. Take two points a and b in , and denote a. b| as the
segment on . Then a. b| is closed in R
k
and so is in B
k
. Enumerate the points
with rational coordinates on a. b|; then it is easy to see that

z
k
(a. b|) = 0.
Write as an increasing limit of line segments containing a. b|. Then we get
the result. |L
I Exercise 4.9 (4.1.9). Let B
(0,1j
= o({(a. b| : (a. b| _ (0. 1|]).
a. B
(0,1j
= {T _ (0. 1| : T B] and B
(0,1j
= {T _ (0. 1| : T L].
b. Construct Lebesgue measure on both B
(0,1j
and B
(0,1j
. Call these measures
z
(0,1j
and

z
(0,1j
, and denote B
(0,1j
by L
(0,1j
.
c. z
(0,1j
as constructed is the measure restriction of z from B to B
(0,1j
and

z
(0,1j
as constructed is the measure restriction of

z from L to L
(0,1j
.
Proof. See, for example, Resnick (1999, Theorem 1.8.1). |L
SECTION 4.3 TRANSLATION INVARIANCE OF LEBESGUE MEASURE 73
4.2 Lebesgue Sets
No exercise.
4.3 Translation Invariance of Lebesgue Measure
I Exercise 4.10 (4.3.2). Let L be such that

z() > 0, and let c 0. 1).
There exists an open interval U such that

z( U) > cz(U).
1
Proof. It follows fromthe Approximation Theoremfor Lebesgue measure that

z() = inf{z(G): G open, _ G].


Then for any c > 0, there exists an open set G containing such that z(G) <

z() cz(G), i.e., (1 c)z(G) <



z(). Thus, for an arbitrary c 0. 1), there
exists an open set G containing such that
cz(G) 6

z().
Write G as an countable disjoint unions of open intervals: G =
_
G
n
. Then

z() =

z( G) since _ G. We thus obtain
cz(G) = cz
_
_
o
_
n=1
G
n
_
_
=
o

n=1
cz(G
n
) 6

z() =

z
_
_
o
_
n=1
( G
n
)
_
_
=
o

n=1

z( G
n
).
Hence, for some N N, we must have cz(G
1
) 6

z( G
1
). Let U = G
1
and
we are done. |L
I Exercise 4.11 (4.3.3). Let L contain an open interval. Then there exists
a > 0 such that (a. a) is contained in D() = {. , : .. , ].
Proof. Let (b. c) _ ; then (b c. c b) _ D(). Let a = c b and so (a. a) _
D(). |L
I Exercise 4.12 (4.3.4). Let L be such that

z() > 0. Then there exists
a > 0 such that (a. a) is contained in D() = {. , : .. , ].
Proof. It follows from Exercise 4.10 that there exists an open interval U _ R
such that

z( U) > 3z(U),4.
We next show that a can be taken as z(U),2.
1
Exercise 4.104.16 are from Halmos (1974).
74 CHAPTER 4 LEBESGUE MEASURE
(i) For an arbitrary . (z(U),2. z(U),2), the set U L (U .) is an open
interval containing ( U) L (( U) .), and
z(U L (U .)) < z(U) z(U),2 =
3
2
z(U).
(ii) (U)L((U).) is an interval. Suppose that (U)((U).) = ;
then

z(( U) L (( U) .)) =

z( U)

z(( U) .)
= 2

z( U)
>
3
2
z(U).
which contradicts the fact that

z(( U) L (( U) .)) 6 z(U L (U .)) <
3z(U),2.
(iii) Thus, for every . (z(U),2. z(U),2), there exists , ( U) ((
U) .); that is, there exists ,. : such that , = : .. But then . = , :
and so . D(). Therefore, if we let a = z(U),2, then (a. a) _ D(). |L
I Exercise 4.13 (4.3.5). Let be a dense subset of R. Then c = {ca: a ] is
dense for any c = 0.
Proof. Take an arbitrary point . R and an arbitrary open interval (.c. .c).
Now consider (. c,c. . c,c). Since is dense, there exists a such that
a (. c,c. . c,c). Thus, ca (. c. . c) and ca c, i.e., c is dense in
R. |L
I Exercise 4.14 (4.3.6). Let be an irrational number.
a. Let = {n m : n. m Z]. Then is a dense subset of R.
b. Let T = {n m : n. m Z. n even]. Then T is a sense subset of R.
c. Let C = {n m : n. m Z. n odd]. Then C is a sense subset of R.
Proof. (a) For every positive integer i there exists a unique integer n
i
(which
may be positive, negative, or zero) such that 0 6 n
i
i < 1; we write .
i
=
n
i
i . If U is any open interval, then there is a positive integer k such that
j(U) > 1,k. Among the k 1 numbers, .
1
. . . . . .
k1
, in the unit interval, there
must be at least two, say .
i
and .
}
, such that [.
i
.
}
[ < 1,k. It follows that
some integral multiple of .
i
.
}
, i.e. some element of , belongs to the interval
U, and this concludes the proof of the assertion concerning .
(b) If is irrational, then ,2 is also irrational. Then D = {n(m,2) : n. m Z]
is dense by (a); then 2D = {n m : n. m Z] is dense by Exercise 4.13.
SECTION 4.3 TRANSLATION INVARIANCE OF LEBESGUE MEASURE 75
(c) Notice that C = T1, and translates of dense sets are obviously dense. |L
I Exercise 4.15 (4.3.7). For .. , R write . - , i . , = {nm : n. m
Z], where is a xed irrational number as in the previous exercise. Then - is an
equivalence relation, and hence R may be partitioned into disjoint equivalence
classes.
Proof. We rst show that - is reexive. For every . R, we have . . =
0 = 0 0 , i.e. . - .. We next show that - is symmetric. If . - ,, then
. , = n m and so , . = n m , i.e. , - .. Finally, we verify that
- is transitive. Let . - , - :. Then . , = n m and , : = q, where
n. m. . q Z. Thus, . : = (n ) (mq) , i.e. . - :. |L
I Exercise 4.16 (4.3.8). We now invoke (AC) to form a set 1
0
consisting of ex-
actly one element from each of the equivalence classes in the previous exercise.
We will now show that 1
0
L.
a. There exist Borel subsets of 1
0
.
b. Let J _ 1
0
be a Borel set. Then D(J) cannot contain any nonzero elements
of , where is the set in Exercise 4.14.
c. By (b), there cannot exist an open interval containing the origin that is con-
tained in D(J), hence z(J) = 0.
d. From (c), we have z
+
(1
0
) = 0.
e. If a
1
and a
2
are distinct elements of = {nm : m. n Z], then 1
0
a
1
and
1
0
a
2
are disjoint.
f. R =
_
{1
0
a: a ], the countable union being disjoint.
g. If 1
0
L, then

z(1
0
a) = 0 for each a , hence

z(R) = 0. Therefore, since


the assumption that 1
0
L leads to an absurdity, it must be the case that
1
0
L, and hence there exists a subset of R that fails to be a Lebesgue set.
Proof. (a) Every singleton is a Borel set.
(b) If there exists . = 0 and . D(J) , there there exists ,. : J such
that , : and , = :. But then , - : and , = :, which contradicts the
construction of 1
0
.
(c) If z(J) > 0, then there exists a > 0 such that (a. a) _ D(J) by Exer-
cise 4.12. Then there exists . such that . (a,2. a) since is dense, which
contradicts (b). Thus, z(J) = 0.
(d) z
+
(J) = 0 follows from the denition immediately.
(e) If : = .
1
a
1
= .
2
a
2
, where .
1
. .
2
1
0
, then .
1
.
2
= a
2
a
1
. Then
.
1
- .
2
and .
1
= .
2
(since a
1
= a
2
). A contradiction.
76 CHAPTER 4 LEBESGUE MEASURE
(f) It suces to show that R _
_
{1
0
a: a ]. Take an arbitrary r R. Since
- is an equivalence relation on R, there is an equivalence class .|
~
containing
r. In particular, r - ., i.e. r . = n m for some n. m Z. Hence, r =
. n m 1
0
(n m). The union is countable since is countable.
(g) If 1
0
L, then 1
0
a L for all a , and

z(1
0
a) =

z(1
0
) = z
+
(1
0
) = 0.
But then

z(R) =

z
_
_
_
o
(1
0
a)
_
_
=

z(1
0
a) = 0.
A contradiction. |L
5
MEASURABLE FUNCTIONS
5.1 Measurability
I Exercise 5.1 (5.1.1). Let :

R be F ,B
+
-measurable. Let ,

R, and let
h:

R be such that
h(o) =

_
(o) if (o) > 0
, if (o) < 0.
Then h is F ,B
+
-measurable.
Proof. Dene :

R

R by letting
(.) =

_
. if . > 0
, if . < 0.
We rst show that is B
+
,B
+
-measurable by demonstrating that
-1
(t. o|
B
+
for each t R. If , < 0, then

-1
(t. o| =

R if t < ,
0. o| if t ,. 0|
(t
2
. o| if t > 0.
If , > 0, then

-1
(t. o| =

R if t < 0
o. 0) L (t
2
. o| if t 0. ,)
(t
2
. o| if t > ,.
Therefore, is B
+
,B
+
-measurable, and so h = is F ,B
+
-measurable. |L
I Exercise 5.2 (5.1.2). There exists a continuous function : R R and a
subset _ R such that L but
-1
() L.
77
78 CHAPTER 5 MEASURABLE FUNCTIONS
Proof. Do according with the hints. |L
I Exercise 5.3 (5.1.5). Suppose that :

R is F ,B
+
-measurable.
a. If F = 2

, can be any function from into



R.
b. If F = {. ], then must be constant.
c. If F = o({
1
. . . . .
n
]), where
1
. . . . .
n
are disjoint subsets of such that
=
_
n
i=1

i
, then must have the form =

n
i=1
c
i
1

i
, where c
1
. . . . . c
n

R.
Proof. (a) is trivial. For (b), if takes two dierent values, say, ,
1
and ,
2
and
,
1
< ,
2
, then
-1
,
2
. o| {. ]; that is, is not F ,B
+
-measurable. For (c),
note that
-1
(c
i
) o({
1
. . . . .
n
]). |L
I Exercise 5.4 (5.1.6). If :

R is such that
-1
({.]) F for every .

R,
then is not necessarily F ,B
+
-measurable.
Proof. Let = R and F = L. Let L, and let
(.) =

. if . .
. if . .
Then
-1
({.]) L for any . R, but fails to be L,B
+
-measurable. |L
I Exercise 5.5 (5.1.7). If _ R is any type of interval and :

R is
monotone, then is both Borel and Lebesgue measurable.
Proof. Without loss of generality, we suppose that is increasing in the
sense that .
1
< .
2
implies that (.
1
) 6 (.
2
). Then for any r

R,

-1
_
r. o|
_
= .. o), where . = inf {. R : (.) > r]. Hence,
-1
is Bo-
real, and so is Lebesgue measurable. |L
I Exercise 5.6 (5.1.11). Let :

R, and suppose that =
_
o
n=1

n
, where

1
.
2
. . . . are disjoint F -sets [F is a o-eld on ]. Let F
n
= { F :
n
] for
each n N. Then F
n
is a o-eld for each n N. Let
n
denote the restriction of
from to
n
, n N. Then is F ,B
+
-measurable i
n
is F
n
,B
+
-measurable
for each n N.
Proof. Assume that each
n
is F
n
,B
+
-measurable. Let T B
+
. Then

-1
(T) =
o
_
n=1

-1
n
(T) F
since each
-1
n
(T) F
n
F . Now assume that is F ,B
+
-measurable. Take
any
n
and T B
+
. Then

-1
n
(T) =
n

-1
(T) F . |L
SECTION 5.2 COMBINING MEASURABLE FUNCTIONS 79
I Exercise 5.7 (5.1.12). Show that the function :

R

R given in Example 4
is B
+
,B
+
-measurable by suitably appealing to (MF6).
Proof. Let
1
= o. 0),
2
= {0], and
3
= (0. o|. Let
i
=
i
for i =
1. 2. 3. Since both
1
and
3
are continuous, they are B
+
,B
+
-measurable; since

2
is constant, it is B
+
,B
+
-measurable. By (MF6), is B
+
,B
+
-measurable. |L
I Exercise 5.8 (5.1.13). The minimal o-eld F on such that :

R is
F ,B
+
-measurable is
-1
_
B
+
_
.
Proof. It suces to show that
-1
_
B
+
_
is a o-eld on since by F ,B
+
-
measurability of , any o-led F includes
-1
_
B
+
_
. First,
-1
_
B
+
_
since

-1
_

R
_
= . If
-1
_
B
+
_
, there exists T B
+
such that
-1
(T) = , then

-1
_

R T
_
= implies that
-1
_
B
+
_
is closed under complements. To
see that
-1
_
B
+
_
is closed under countable union, let {
n
]
o
n=1
_
-1
_
B
+
_
. So
there exists T
n
B
+
for each n N with
-1
(T
n
) =
n
. Therefore,

-1
_
_
o
_
n=1
T
n
_
_
=
o
_
n=1

-1
(T
n
) =
o
_
n=1

n
implies
-1
_
B
+
_
is closed under countable unions. |L
I Exercise 5.9 (5.1.14). The word continuous in (MF4) may be replaced by
either of lower semicontinuous and upper semicontinuous.
Proof. For a detailed discussion of semicontinuous functions, see Ash (2009,
Section 8.4). Let :

R be low semicontinuous (LSC), then
-1
(t. o| is open
for any t

R. Therefore,
-1
(t. o| F and so is F ,B
+
-measurable. Now let
be upper semicontinuous (USC), then is LSC and so is F ,B
+
-measurable;
then =
_

_
is F ,B
+
-measurable. |L
5.2 Combining Measurable Functions
I Exercise 5.10 (5.2.1). If :

R is F ,B
+
-measurable, then [ [ is F ,B
+
-
measurable. However, if [ [ is F ,B
+
-measurable, then is not necessarily
F ,B
+
-measurable.
Proof. Since [ [ =


-
, and is F ,B
+
-measurable if and only if

and
-
are measurable, we know that [ [ is F ,B
+
-measurable. To see that the
converse is not true take F and let
(o) = 1

(o) 1

c (o) =

1 if o
1 if o
c
.
80 CHAPTER 5 MEASURABLE FUNCTIONS
It is not F ,B
+
-measurable since
-1
(0. o| = F . But [ [ = 1 is F ,B
+
-
measurable. |L
I Exercise 5.11 (5.2.2). Let n N, and let
1
. . . . .
n
denote F ,B
+
-measurable
functions with common domain F .
a. Both max{
1
.
2
] and min{
1
.
2
] are F ,B
+
-measurable functions.
b. Both max{
1
. . . . .
n
] and min{
1
. . . . .
n
] are F ,B
+
-measurable functions.
Proof. (a) Let g = max{
1
.
2
]. For an arbitrary . R, we have

o : max{
1
.
2
](o) < .
_
= {o :
1
(o) < .] {o :
2
(o) < .] F .
and

o : min{
1
.
2
] > .
_
= {o :
1
(o) > .] {o :
2
(o) > .] F .
(b) We do the max case. Let g
n
= max{
1
. . . . .
n
]. The claim holds for n = 1
and 2 by (a). Assume that it is true for n N. Then for n 1, we have
{o : g
n1
(o) < .] =
_
_
n
_
i=1
{o :
i
(o) < .]
_
_
{o :
n1
(o) < .] F
by the induction hypothesis. |L
I Exercise 5.12 (5.2.3). Let (. F . j) denote a measure space, and let :

R denote a F ,B
+
-measurable mapping. Let v : B
+


R be such that v(T) =
j(
-1
(T)) for every T B
+
. That is, v = j
-1
. Then (

R. B
+
. v) is a measure
space. Furthermore, even if j is o-nite, v is not necessarily o-nite.
Proof. It is clear that v is well dened since is F ,B
+
-measurable. To see v
is a measure on B
+
, note that (i) v(T) = j(
-1
(T)) > 0 for every T B
+
, (ii)
v() = j() = 0, and (iii) For a disjoint sequence {T
n
] _ B
+
, we know that the
sequence {
-1
(T
n
)] _ F is disjoint; then the countable additivity follows. |L
I Exercise 5.13 (5.2.4). This exercise concerns itself with (MF9).
a. Prove part (b) of (MF9) by suitably adapting the proof of (a).
b. Prove part (b) of (MF9) by using (a) of (MF9) and (MF7).
c. Show that {o : (o) g(o) < .] =
_
i
1
,i
2
Q;i
1
i
2
~x
{o : (o) <
r
1
] {o : g(o) < r
2
].
d. Repeat part (c) for g by proving an analogous identity.
e. Let ,

R, n > 2, and for i = 1. . . . . n, let
i
:

R denote a F ,B
+
-
measurable function. Let h:

R be dened for all o by the rule
SECTION 5.2 COMBINING MEASURABLE FUNCTIONS 81
h(o) =

1
(o)
n
(o) if
1
(o)
n
(o) is dened
, if
1
(o)
n
(o) is undened.
Show that h is F ,B
+
-measurable.
Proof. (a) Let ,

R, and let . R. Dene

,
=

-1
(o) g
-1
(o)
_
L
_

-1
(o) g
-1
(o)
_
if , > .
if , 6 ..
Observe that (i)
,
_ , and (ii) the assumption of F ,B
+
-measurability for
and g forces
,
F (and hence
,
F ). Next,
h
-1
(.. o| = {o : h(o) > .]
=

o
,
: h(o) > .
_
L

o
,
: h(o) > .
_
=

o
,
: (o) g(o) > .
_
L
,
=

o
,
: (o) > . g(o)
_
L
,
F .
(b) Note that g is F ,B
+
-measurable since g is. Then g is F ,B
+
-measurable.
It follows form (MF7)(c) that h is F ,B
+
-measurable.
(c) Let 1 = {o : (o) g(o) < .], and 1 =
_
i
1
,i
2
Q;i
1
i
2
~x
{o : (o) <
r
1
] {o : g(o) < r
2
]. If o 1, then (o) g(o) < .. Take c R

such
that
(o) c| g(o) c| = ..
Such an c exists since (o). g(o). . R. Then there exists r
1
Q such that
(o) < r
1
< (o) c since Q is dense in R; similarly, there exists r
2
Q such
that g(o) < r
2
< g(o) c. Thus,
r
1
r
2
< (o) c g(o) c = .:
that is, o 1. The other direction is evident.
(d) For g, we have
{o : (o) g(o) < .] =
_
i
1
,i
2
Q
i
1
-i
2
~x
{o : (o) < r
1
] {o : g(o) < r
2
].
Let 1 denote the left hand side of the above display, and let 1 denote the right
hand side. If o 1, then (o) g(o) < .. Pice c R

such that (o) c|


g(o) c| = .. Pick r
1
. r
2
Qsuch that (o) < r
1
< (o) c and g(o) < r
2
<
g(o) c. Then r
1
r
2
< ., i.e., o 1. The other direction is evident.
(e) The claim holds for n = 2. Let us assume that it holds for n N. We now
consider the n 1 case. Dene
82 CHAPTER 5 MEASURABLE FUNCTIONS

,
=

{o :
1
(o)
n1
(o) is undened] if , > .
if , 6 ..
It is clear that
,
F . Next,
h
-1
(.. o| = {o : h(o) > .]
= {o
,
:
1
(o)
n1
(o) > .] L
,
.
It suces to show that {o
,
:
1
(o)
n1
(o) > .] F . Notice
that if
1
(o)
n1
(o) is dened, then
1
(o)
n
(o) is dened,
too. Thus,
1

n
is F ,B
+
-measurable on
,
. It follows from (MF8) that
h
-1
(.. o| F . |L
I Exercise 5.14 (5.2.5). This exercise concerns itself with (MF10).
a. Directly prove (a) of (MF10) assuming that and g are real valued instead
of extended real valued.
b. Prove (b) of (MF10) assuming that and g are real valued and that g is
nonzero on .
c. Using the previous part, now prove (b) of (MF10) in full generality.
Proof. (a) For every . R, we have
(g)
-1
(.. o| =
_
o :
_
(o) g(o)
_
2

_
(o) g(o)
_
2
> 4.
_
F :
that is, g is F ,B
+
-measurable.
(b) Write ,g = (1,g). Then 1,g is real valued and F ,B
+
-measurable. It
follows from (a) that ,g is F ,B
+
-measurable.
(c) Take an arbitrary . R. Dene

,
=

{o : (o),g(o) is undened] if , > .


if , 6 ..
Notice that
{o : (o),g(o) is undened] =
-1
(o) g
-1
(o)| L g
-1
(0):
thus,
,
F . Next,
h
-1
(.. o| = {o : (o),g(o) > .]
= {o
,
: (o),g(o) > .] L
,
.
So it suces to show that {o
,
: (o),g(o) > .] F . It can be proved
case by case. |L
SECTION 5.3 SEQUENCES OF MEASURABLE FUNCTIONS 83
5.3 Sequences of Measurable Functions
I Exercise 5.15 (5.3.1). Prove (a)-(e) of (MF13).
Proof. (a) Pick an arbitrary o . If (o) > 0, we have

(o) = (o) and

-
(o) = 0. Then


(o)
-
(o) = (o).
Next, if (o) < 0, we get

(o) = 0 and
-
(o) = (o). Then


(o)
-
(o) = (o).
Hence, [ [ =

-
.
Now if =

and
-
= 0, then [ [ =

-
= , i.e., > 0. The other
claim is similar.
(b) If c > 0, then c(o) > 0 i (o) > 0. Hence,
(c )

(o) =

c(o) if c(o) > 0


0 if c(o) < 0
=

c(o) if (o) > 0


0 if (o) < 0
= c

(o).
If c < 0, then c(o) > 0 i (o) 6 0. Thus,
(c )

(o) =

c(o) if (o) 6 0
0 if (o) > 0
= c
-
(o).
(c) Suppose that (o) = g(o) > 0; then ( g)

(o) = ( g)
-
(o) = 0, but


(o) g

(o) =

(o) > 0, and
-
(o) g
-
(o) = g(o) > 0.
To see 0 6 ( g)

6

g

(the rst inequality always holds), observe


that for every o ,
( g)

(o) =

(o) g(o) if (o) g(o) > 0


0 if (o) g(o) < 0.
and


(o) g

(o) =

(o) g(o) if (o) > 0. g(o) > 0


(o) if (o) > 0. g(o) < 0
g(o) if (o) < 0. g(o) > 0
0 if (o) < 0. g(o) < 0.
For instance, if (o) > 0, g(o) < 0, and (o) g(o) > 0, then ( g)

(o) =
(o)g(o) < (o) =

(o)g

(o). All other cases can be analyzed similarly.


(d) If [g[ 6 , i.e., g

g
-
6 , then 0 6 g

. g
-
6 .
(e) For every o , we have (o) = g(o) h(o) 6 g(o); thus

6 g

= g.
Similarly, for every o , we have (o) > h(o), and so
-
6 (h)
-
= h. |L
84 CHAPTER 5 MEASURABLE FUNCTIONS
I Exercise 5.16 (5.3.2). The class of F ,B
+
-measurable functions is not nec-
essarily closed under uncountable suprema and inma. The following outline
gives a simple instantiation of this claim. Let = R and F = B.
a. Let 1 denote a non-Borel set as constructed in Section 4.5. Argue that 1
cannot be at most countable.
b. For each . 1, dene
x
: R

R by writing
x
(o) = 1
x
(o) for each o R.
Then
x
is F ,B
+
-measurable for each . 1, but sup
xT

x
= 1
T
, hence
sup
xT

x
is not F ,B
+
-measurable.
Proof. (a) Every singleton set {.] R is a Borel set; thus, if 1 is at most
countable, it would be a Borel set.
(b) For every . 1, the function
x
is B,B
+
-measurable by (MF3) since {.]
B. However, sup
xT

x
= 1
T
is not B,B
+
-measurable since 1 B. |L
5.4 Almost Everywhere
I Exercise 5.17 (5.4.1). In (MF15), the completeness of
_
. F . j
_
is not a re-
dundant assumption.
Proof. Note that the proof of (MF15) dependents on (MF14), which depends
on the completeness of the measure space. |L
I Exercise 5.18 (5.4.2). Suppose that : R R is a dierentiable function.
Then
t
is a Borel measurable function.
Proof. Let
n
(.) = (. 1,n) for all . R and n N. Then
n
is a Borel
measurable function for each n N since is continuous. Therefore,
(. 1,n) (.)
1,n
= n
_

n
(.) (.)
_
is Borel measurable for each n N. Thus
t
= lim
n
n(
n
) is Borel measur-
able. |L
5.5 Simple Functions
I Exercise 5.19 (5.5.1). Refer to (MF18).
a. If :

R is a general F ,B
+
-measurable function, then there exists a se-
quence {s
n
]
o
n=1
of F ,B
+
-measurable and nite-valued simple functions such
that s
n
, having the additional property that 0 6 [s
1
[ 6 [s
2
[ 6 .
b. If in (a) the function is also bounded, then s
n
uniformly on .
SECTION 5.5 SIMPLE FUNCTIONS 85
c. The following converse to (MF18) holds: if :

R is such that there exists
a sequence {s
n
] of simple F ,B
+
-measurable functions with s
n
then is
F ,B
+
-measurable.
Proof. (a) Write =


-
. Then there exist nondecreasing, nonnegative
F ,B
+
-measurable and nite-valued simple functions {s

n
] and {s
-
n
] such that
s

n


and s
-
n

-
. Let s
n
= s

n
s
-
n
for all n, and consider {s
n
].
(b) We rst consider a nonnegative F ,B
+
-measurable bounded function :
0. o). Fix an c > 0. For every n N and k = 1. . . . . n2
n
, dene

n,k
=
_
o :
k 1
2
n
6 (o) <
k
2
n
_
and T
n
= {o : (o) > n] .
Take N
1
N so that (o) s
1
1
(o) < 1,2
1
1
6 c for all o
_
n2
n
k=1

1
1
,k
. Now
pick N N such that N > N
1
and (o) N < c for all o T
1
. This proves
that s
n
uniformly. This result can be easily extended.
(c) Follows from (MF11). |L
I Exercise 5.20 (5.5.2). Consider the measure space (R
k
. L
k
.

z
k
). Let : R
k

R denote a Lebesgue measurable function. We will show that there exists a Borel
measurable function g: R
k


R with [g[ 6 [ [ and = g

z
k
a. e.
a. Let > 0. There exists a sequence {s
n
] of nonnegative nite-valued Lebesgue
measurable simple functions such that 0 6 s
1
6 s
2
6 6 and s
n
.
Pick m N, and write s
n
=

n
m
}=1
c
n}
1

mj
, where 0 6 c
n1
. . . . . c
nn
m
< o and

n1
. . . . .
nn
m
are disjoint L
k
-sets with
_
n
m
}=1

n}
= R
k
. Write the set
n}
as
T
n}
L C
n}
, where T
n}
B
k
and C
n}
is contained in some z
k
-null set N
n}
.
Dene s
+
n
=

n
m
}=1
c
n}
1
B
mj
.
b. For each m N, s
+
n
is a Borel measurable simple function such that 0 6 s
+
n
6
s
n
and s
+
n
= s
n

z
k
a. e.
Dene N =
_
o
n=1
N
n
, where N
n
= {x R
k
: s
n
(x) = s
+
n
(x)] for each m N,
and let g = sup
nN
s
+
n
.
c. N is

z
k
-null, 0 6 g 6 , and g =

z
k
a. e.
d. g is Borel measurable, hence the proof is complete in the nonnegative case.
e. The claim holds when : R
k


R is an arbitrary Lebesgue measurable func-
tion.
Proof. (a) Follows from (MF18) immediately.
(b) s
+
n
is B
k
,B
+
-measurable since T
n}
B
k
for every = 1. . . . . n
n
(by (MF16)).
Dene
N
n
=
_
x R
k
: s
+
n
(x) = s
n
(x)
_
.
Then D _
_
n
m
}=1
C
n}
and so

z
k
(D) 6

z
k
(
_
n
m
}=1
C
n}
) = 0; that is, s
+
n
= s
n

z
k
a. e.
The other claims are trivial.
86 CHAPTER 5 MEASURABLE FUNCTIONS
(c) It follows from (b) that

z
k
(N
n
) = 0 for every m N. Thus,

z
k
(N) 6

o
n=1

z
k
(N
n
) = 0.
Fix an arbitrary x R
k
. Then s
+
n
(x) 6 s
n
(x) 6 (x). Hence, g(x) =
sup
n
s
+
n
(x) 6 (x), i.e., g 6 . Notice that g(x) = (x) only probably on
N, and since

z
k
(N) = 0, we conclude that g =

z
k
a. e.
(d) It follows from(b) that s
+
n
is B
k
,B
+
-measurable. Then g is B
k
,B
+
-measurable
since g = sup
nN
s
+
n
.
(e) Write =

-
. Then there exist nonnegative B
k
,B
+
-measurable func-
tions g
1
and g
2
such that 0 6 g
1
6

, g
1
=


z
k
a. e., and 0 6 g
2
6
-
,
g
2
=
-

z
k
a. e. Let g = g
1
g
2
. Then g =

z
k
a. e., and
[g[ = [g
1
g
2
[ 6 [g
1
[ [g
2
[ = g
1
g
2
6

-
= [ [. |L
I Exercise 5.21 (5.5.3). In Exercise 5.20, show that the Borel measurable func-
tion g: R
k


R may be chosen such that [g[ > [ [ and = g

z
k
a. e.
Proof. Just let
n}
= T
n}
C
n}
, and
_
n
m
}=1
T
n}
= R
k
. Then s
+
n
> s
n
for every
m, and so g > when > 0. As in Exercise 5.20(e), we also get the other
results. |L
5.6 Some Convergence Concepts
I Exercise 5.22 (5.6.1). This result concerns uniqueness.
a. If
n
ja. e. and
n
g ja. e., then = g ja. e..
b. If
n
in j-measure and
n
g in j-measure, then = g ja. e..
Proof. (a) Write (o) = g(o)| as the union of four F -sets:

1
= (o) = g(o).
n
(o) (o). g
n
(o) g(o)|.

2
= (o) = g(o).
n
(o) (o). g
n
(o) ,g(o)|.

3
= (o) = g(o).
n
(o) ,(o). g
n
(o) g(o)|.

4
= (o) = g(o).
n
(o) ,(o). g
n
(o) ,g(o)|.
Since limits of sequences of numbers are unique,
1
= . Observe that each of

2
,
3
, and
4
are contained in v-null sets; for example,
2
_ g
n
(o) , g(o)|.
Thus, j(o) = g(o)| = 0.
(b) Fix m N. Suppose that there exists o such that [(o) g(o)[ > 1,m.
For an arbitrary n N, if [
n
(o) (o)[ < 1,(2m), then
[
n
(o) g(o)[ > [(o) g(o)[ [
n
(o) (o)[ > 1,(2m).
Thus,
SECTION 5.6 SOME CONVERGENCE CONCEPTS 87
[(o) g(o)[ > 1,m| _ [
n
(o) (o)[ > 1,(2m)| L [
n
(o) g(o)[ > 1,(2m)|.
For any xed c > 0, N N such that j[
n
(o) (o)[ > 1,(2m)| < c,2
n1
and
j[
n
(o) g(o)[ > 1,(2m)| < c,2
n1
. Then j[(o) g(o)[ > 1,m| < c,2
n
. Since
j(o) = g(o)| = j
_
_
o
_
n=1
[(o) g(o)[ > 1,m|
_
_
6
o

n=1
j[(o) g(o)[ > 1,m|
6
o

n=1
c,2
n
= c.
we have = g ja. e. |L
I Exercise 5.23 (5.6.2). Suppose that
n

and g
n

g.
a.
n

0 and [
n
[

[ [.
b. If a. b R, then a
n
b

a b.
c. If a. b R, then a
n
bg
n

a bg.
d.

n


and
-
n

-
.
e. If j() < o and j > 0, there is M > 0 with j[g(o)[ > M| < j.
f. If j() < o, then
n
g

g.
g. If j() < o, then
n
g
n

g.
h. If j() = o, then
n
g
n
does not necessarily converge to g in j-measure.
i. It is not necessarily the case that
n
,g
n

,g, even if g(o) = 0 and g


n
(o) =
0 for every o and n N. However, if j() < o, the result follows.
Proof. (a)
n

i for any c > 0, we have lim


n-o
j[
n
(o) (o)[ > c| = 0;
that is
lim
n-o
j
_
[(
n
(o) (o)) 0[ > c
_
= 0.
i.e.,
n

0.
To see the second implication, note that [[
n
(o)[ [(o)[[ 6 [
n
(o) (o)[,
and so
_
[[
n
(o)[ [(o)[[ > c
_
_
_
[
n
(o) (o)[ > c
_
.
Therefore, j[
n
(o) (o)[ > c| = 0 implies that j[[
n
(o)[ [(o)[| > c| = 0;
that is, [
n
[

[ [.
88 CHAPTER 5 MEASURABLE FUNCTIONS
(b) It is clear that
_
[
n
(o) (o)[ > c
_
=
_
[a
n
(o) a(o)[ > [a[c
_
=
_
[(a
n
(o) b) (a(o) b)[ >[a[ c
_
.
(c) It suces to show that
n
g
n

g. By the triangle inequality, we have


[(
n
g
n
)(o) ( g)(o)[ 6[
n
(o) (o)[ [g
n
(o) g(o)[ .
Therefore, [(
n
g
n
)(o) ( g)(o)[ > c| [
n
(o) (o)[ > c,2| L [g
n
(o)
g(o)[ > c,2|, and so
n
g
n

g. Then, by (b), it is evident that a


n
bg
n

a bg.
(d) Observe that
[

n
(o)

(o)[ = [ max{
n
(o). 0] max{(o). 0][
6 [ max{
n
(o). 0] (o)[
= [(o) max{
n
(o). 0][
6 [(o)
n
(o)[.
Thus, [

n
(o)

(o)[ > c| _ [
n
(o) (o)[ > c|, and so

n


. Similarly,

-
n

-
.
(e) Let
n
= [g(o)[ > m| for every m N. Observe that {
n
] _ F and

n
(since g(o) R). By monotonicity, we get j(
1
) 6 j() < o; thus,
lim
n
j(
n
) = j(lim
n

n
) = j() = 0. Given j > 0, there exists M N such
that j(

) < j; that is, j[g(o)[ > M| < j.


(f) Observe that for an arbitrary M N,
_
[
n
(o)g(o) (o)g(o)[ > c
_
=
_
[
n
(o) (o)[ [g(o)[ > c
_
=
_
[
n
(o) (o)[ [g(o)[ > c. [g(o) > M
_
L
_
[
n
(o) (o)[ [g(o)[ > c. [g(o)[ 6 M
_
_
_
[g(o)[ > M
_
L
_
[
n
(o) (o)[ > c,M
_
.
It follows from (e) that for an arbitrary > 0, there exists M such that
j[g(o)[ > M| < ,2. Since
n

, there exists N
0
N such that j[
n
(o)
(o)[ > c,M| < ,2. Let N = max{M. N
0
]. Then j[
n
(o)g(o) (o)g(o)[ >
c| < ; that is,
n
g

g.
(h) Observe that how the assumption j() < o is used.
(i) Notice that
[g(o) > M| = [1,g(o)[ < 1,M|. |L
I Exercise 5.24 (5.5.3). Suppose that
n
a.e.
and g
n
a.e.
g.
a.
n

a.e.
0 and [
n
[
a.e.
[ [.
SECTION 5.6 SOME CONVERGENCE CONCEPTS 89
b. If a. b R, then a
n
b
a.e.
a b.
c. If a. b R, then a
n
bg
n
a.e.
a bg.
d.

n
a.e.


and
-
n
a.e.

-
.
e.
n
g
n
a.e.
g.
f. If g. g
n
= 0 ja. e. for each n N, then
n
,g
n
a.e.
,g.
Proof. (a) We have
j
n
(o) (o) ,0| = j
n
(o) ,(o)| = 0.
Similarly for [
n
[.
(b) If a, the claim holds trivially. So assume that a = 0. Then
ja
n
(o) b ,a(o) b| = j
n
(o) ,(o)| = 0.
(c) For every o with a
n
(o) bg(o) ,a(o) bg(o), if
n
(o) (o), then
g
n
(o) ,g(o). Therefore,
a
n
(o) bg(o) ,a(o) bg(o)| _
n
(o) ,(o)| L g
n
(o) ,g(o)|.
and so a
n
bg
n
a.e.
a bg.
(d) Take an arbitrary o such that
n
(o) (o). If (o) > 0, then there
exists N N such that
n
(o) > 0 for all n > N, and then

n
(o)

(o). If
(o) < 0, then there exists N N such that
n
(o) < 0 for all n > N, and then


n
(o) = 0 for all n > N. Thus,

n
(o) 0 =

(o). Finally, if (o) = 0, then
there exists N Nsuch that [
n
(o)(o)[ < c. In this case, for all n > N, either


n
(o) =
n
(o) > 0 and
n
(o) < c, or

n
(o) = 0; thus,

n
(o)

(o) = 0.
We thus proved that

n
(o)

(o) whenever
n
(o) (o). In other
words, we have


n
(o) ,

(o)| _
n
(o) ,(o)|:
that is,

n
a.e.


. Similarly, we get
-
n
a.e.

-
.
(e)
n
(o)g
n
(o) ,(o)g(o)| _
n
(o) ,(o)| L g
n
(o) ,g(o)|.
(f)
n
(o),g
n
(o) ,(o),g(o)| _
n
(o) ,(o)| L g
n
(o) ,g(o)|. |L
I Exercise 5.25 (5.5.4). Suppose for each n N that
n
= g
n
ja. e..
a. If
n
a.e.
, then g
n
a.e.
.
b. If
n

, then g
n

.
90 CHAPTER 5 MEASURABLE FUNCTIONS
Proof. (a) We have
g
n
(o) ,(o)| _
_
_
o
_
n=1
g
n
(o) =
n
(o)|
_
_
L
n
(o) ,(o)|.
Thus, g
n
a.e.
if g
n
a.e.
.
(b) We have
[g
n
(o) (o)[ > c| _
_
_
o
_
n=1
g
n
(o) =
n
(o)|
_
_
L [
n
(o) (o)[ > c|:
thus, g
n

. |L
IExercise 5.26 (5.6.5). Prove the following statements connecting convergence
in j-measure with convergence ja. e.
a.
n

i
n
j

for every subsequence {n


}
].
b.
n

i each subsequence of {
n
] has a sub-subsequence that converges
to ja. e..
Proof. See Resnick (1999, Theorem 6.3.1). Here is the basic procedure of the
proof:
n

i {
n
] is Cauchy in measure, i.e., j[
i

x
[| 0 as r. s o.
Then there exists a subsequence {
n
k
] which converges a.s. |L
I Exercise 5.27 (5.6.6). Suppose that : R R is continuous.
a. If
n
a.e.
, then
n
a.e.
.
b. If
n

, then
n

.
Proof. (a) There exists a null set N F with j(N) = 0 such that if o N
c
,
then
n
(o) (o). By continuity of , if o N
c
, then
_

n
(o)
_

_
(o)
_
.
(b) Let {
n
k
] be some subsequence of {
n
]. It suces to nd an a.s.
convergence subsequence {
n
k.i/
]. But we know that {
n
k
] has some a.s.
convergent subsequence {
n
k.i/
] such that
n
k.i/
a.e.
. Thus,
n
k.i/
a.e.

by (a). |L
5.7 Continuity and Measurability
5.8 A Generalized Definition of Measurability
I Exercise 5.28 (5.8.1). Regarding the measure v:
SECTION 5.8 A GENERALIZED DEFINITION OF MEASURABILITY 91
a. v is really a measure.
b. If j is nite, then so is v.
c. If j is o-nite, v need not be o-nite.
Proof. (a) (i) v(
t
) > 0 for all
t
F
t
since v(
t
) = j(
-1
(
t
)) > 0 for all

t
F
t
. (Ii) v() = (j
-1
)() = j() = 0. (iii) Let {
t
n
] _ F
t
be disjoint.
Then
v
_
_

t
n
_
= j
_

-1
_
_

t
n
_
_
= j
_
_

-1
(
t
n
)
_
=

j
_

-1
(
t
n
)
_
=

v(
t
n
).
(b) If j() < o, then v(
t
) = j(
-1
(
t
)) = j() < o.
(c) Suppose that there exists a unique sequence {
n
] of F -sets such that =
_

n
and j(
n
) < ofor all n N; suppose that there does not exist a sequence
{
t
n
] of F
t
-sets such that
-1
(
t
n
) =
n
for all n N. Then v is not o-nite. |L
I Exercise 5.29 (5.8.2). Modify (MF14) and prove it in the more general setting
of this section.
Proof. (MF14) can be modied in the following way:
(MF14
0
) Let (. F . j) denote a complete measure space. Pick nonempty in
F , and let be dened ja. e. on and F ,F
t
-measurable. If g is dened
ja. e. on and = g ja. e. on , then g is F ,F
t
-measurable.
Proof of (MF
t
). Let T = {o dom( ) dom(g) : (o) = g(o)], so that j(
T) = 0. Take an arbitrary
t
F
t
, and observe that
g
-1
(
t
) =
_
g
-1
(
t
) T
_
L
_
g
-1
(
t
) ( T)
_
=
_

-1
(
t
) T
_
L
_
g
-1
(
t
) ( T)
_
.
Since
-1
(
t
) F and T F , we have
-1
(
t
) T F . Next, g
-1
(
t
) (T)
is a subset of T, and j( T) = 0. Since (. F . j) is complete, we have
g
-1
(
t
) ( T) F . Thus, g
-1
(
t
) F and so g is F ,F
t
-measurable. |L
6
THE LEBESGUE INTEGRAL
6.1 Stage One: Simple Functions
I Exercise 6.1 (6.1.1). Let 1 F .
a. j(1) = 0 i
s
T
(s) = 0 for every s S.
b. For any c > 0,
s
T
(c) = cj(1).
Proof. (a) Write s =

n
i=1
c
i
1

i
. If j(1) = 0, then
s
T
(s) =

n
i=1
c
i
j(
i
1) =

n
i=1
c
i
0 = 0 since j(
i
1) 6 j(1) = 0 implies that j(
i
1) = 0. For the
converse direction, observe that if j(1) > 0, then
s
T
(1
T
) = j(1) > 0, where,
of course, 1
T
S.
(b) Write c S as c = c1

. Thus,
s
T
(c) = cj( 1) = cj(1). |L
I Exercise 6.2 (6.1.2). Let t. s. s
1
. s
2
. . . . S. Why cant we say that
s
T
(as
bt ) = a
s
T
(s) b
s
T
(t ) for every a. b R, as compared to saying that the
result holds for every 0 6 a. b < o? Also, why cant we necessarily write

s
T
(

o
i=1
c
i
s
i
) =

o
i=1
c
i

s
T
(s
i
)? [What is the domain of
s
T
?]
Proof. This is because the domain of
s
T
is S: the collection of nite-valued
nonnegative F ,B
+
-measurable simple functions with domain . Hence, if a <
0 and b < 0, then as bt S. We cant necessarily write
s
T
(

o
i=1
c
i
s
i
) =

o
i=1
c
i

s
T
(s
i
) because it is possible that

o
i=1
c
i
s
i
(o) = ofor some o . |L
I Exercise 6.3 (6.1.3). Let 1 F be such that j(1) > 0. Then
s
T
(s) = 0 i
s = 0 ja. e. on 1. In particular,
s
(s) = 0 i s = 0 ja. e. [on ].
Proof. Assume rst that s = 0 ja. e. on 1. Let 1
1
= {o 1 : s(o) = 0] and
1
2
= 1 1
1
. Then j(1
2
) = 0, and so

s
T
(s) =
s
T
1
LT
2
(s) =
s
T
1
(s) = 0.
Conversely, assume that
s
T
(s) = 0. Let
n
= {o 1 : s(o) > 1,n] for
each n N. Then, for every n N, we have
n
F and
1
n
1

n
6 s on 1;
93
94 CHAPTER 6 THE LEBESGUE INTEGRAL
hence
s
T
(
1
n
1

n
) 6
s
T
(s) by (S4), whence j(
n
),n 6 0 by (S6). It follows that
j(
n
) = 0 for all n N. Since
{o 1 : s(o) > 0] =
o
_
n=1

n
.
we have j({o 1 : s(o) > 0]) = 0; hence s = 0 ja. e.. Replacing 1 by we
get the second claim. |L
I Exercise 6.4 (6.1.4). Let 1 F and s S.
a. If j(1) < o, then
s
T
(s) < o, but the converse is not necessarily true.
b. If j(1) = oand
s
T
(s) = o, then j({o 1 : s(o) > 0]) > 0, but the converse
is not necessarily true.
c. Let j(1) = o. Then
s
T
(s) < o i j({o 1 : s(o) > 0]) < o.
Proof. (a) Let s =

n
i=1
c
i
1

i
, where 0 6 c
i
< o and
_
n
i=1

i
= . If j(1) <
o, then j(
i
1) < j(1) < o, for all i . Thus,
s
T
(s) =

n
i=1
c
i
j(
i
1) < o
since the nite summation of nite terms is nite.
To see the converse is not necessarily true, consider t = 1
T
c . Then
s
T
(t ) =
0, and which holds no matter whether j(1) = o or not as long as we assume
that 0 o= 0.
(b) If s = 0 ja. e. on 1, then
s
T
(s) = 0 by the previous exercise; hence,
j({o 1 : s(o) > 0]) > 0.
To see the converse is not necessarily true, let t = 1
T
. Then j({o 1 :
t (o) > 0]) = j(1). By letting 0 < j(1) < o, we see that j({o 1 : t (o) > 0]) >
0, but j(1) < o and
s
T
(t ) = j(1) < o.
(c) If j(1) = o, then
n

i=1
c
i
j(
i
1) < o == j(
i
1) < o for all i with c
i
> 0
== j({o 1 : s(o) > 0]) < o. |L
I Exercise 6.5 (6.1.5). Suppose that {1
n
]
o
n=1
is a nonincreasing sequence of
F -sets. Also, let s S.
a. It is not necessarily the case that lim
n

s
T
n
(s) =
s
lim
n
T
n
(s).
b. If
s
T
n
(s) < o for some n N, then lim
n

s
T
n
(s) =
s
lim
n
T
n
(s).
c. If {1
n
]
o
n=1
is no longer nonincreasing but is still such that lim
n
1
n
exists, state
conditions under which lim
n

s
T
n
(s) =
s
lim
n
T
n
(s).
Proof. (a) Let (. F . j) = (N. 2
N
. j), where j is the counting measure; let s =
1

. Let 1
n
= { N : > n]. Then 1
n
, j(1
n
) = o and j(
_
o
n=1
1
n
) = 0.
In this case,
s
T
n
(1

) = o for each n; but


s
lim
n
T
n
(1

) = 0. This argument and


the following example is modied from Folland (1999, p. 26).
SECTION 6.1 STAGE ONE: SIMPLE FUNCTIONS 95
(b) The same as (M9). This example shows that some niteness assumption is
necessary.
(c) See Vestrup (2003, p. 47-48). |L
I Exercise 6.6 (6.1.6). Let s S and recall (S7), where v
x
is the measure on F
with v
x
(T) =
s
B
(s) for every T F . Then, for any 1 F and t S, we have

s
T
(t : v
x
) =
s
T
(t s: j).
Proof. Let s =

n
i=1
c
i
1

i
and t =

n
}=1
J
}
1
B
j
. Then

s
T
(t : v
x
) =
n

}=1
J
}
v
x
(T
}
1) =
n

}=1
J
}

s
B
j
T
(s: j) =
n

}=1
J
}

s
T
(s1
B
j
: j)
=
s
T
_
_
n

}=1
J
}
s1
B
j
: j
_
_
=
s
T
(t s: j). |L
I Exercise 6.7 (6.1.7). Let F be nonempty, and let s S. Let F

= {1 _
: 1 F ], and let j

denote the restriction of j from F to F

. Finally, let s

denote the restriction of s from to .


a. (. F

. j

) is a measure space.
b. If J denotes this sections functional relative to (. F

. j

), then we have

s
T
(s: j) = J
T
(s

: j

) for every 1 F

.
Proof. (a) We only need to show that F

is a o-eld. (i) F

because F
and _ . (ii) If 1 F

, then 1 _ and 1 F . Thus, 1 _ and


1 F , i.e., 1 F

. (iii) Let {1
n
] _ F

. Then 1
n
_ and 1
n
F for all
n. Thus,
_
1
n
_ and
_
1
n
F ; that is,
_
1
n
F

.
(b) Let s =

n
i=1
c
i
1

i
. Then s

n
i=1
c
i
1

i
;
, and so
J
T
(s

: j

) =
n

i=1
c
i
j

(
i
1) =
n

i=1
c
i
j(
i
1) =
s
T
(s: j)
since j

= j on F

and 1 _ . |L
I Exercise 6.8 (6.1.8). Let F {], and suppose that s : 0. o) is
simple and F ,B
+
-measurable. Let 1 F

, where F

is dened in the previous


exercise. Consider two programs:
Program 1 Extend s from to as follows: Let s
+
S be such that s
+
= s on
and s
+
= 0 on
c
.
Program 2 Do not extend s from to as in Program 1, but instead view s
as a function dened everywhere relative to the measure space (. F

. j

),
where the notation is as in the previous exercise.
96 CHAPTER 6 THE LEBESGUE INTEGRAL
These two programs are equivalent in the sense that
s
T
(s
+
: j) = J
T
(s: j

),
where
s
T
is this sections functional relative to the measure space (. F . j),
and J
T
is this sections functional relative to the measure space (. F

. j

).
Proof. Write s =

n
i=1
c
i
1

i
, where {
i
] is disjoint and
_
n
i=1

i
= . Extend s
to s
+
as
s
+
=
n

i=1
c
i
1

i
01

c .
Then

s
T
(s
+
: j) =
n

i=1
c
i
j(
i
1) 0 j(
c
1) =
n

i=1
c
i
j(
i
1).
Next, consider Program 2. We have
J
T
(s: j

) =
n

i=1
c
i
j

(
i
1) =
n

i=1
c
i
j(
i
1).
Thus,
s
T
(s
+
: j) = J
T
(s: j

). |L
I Exercise 6.9 (6.1.9). Quickly prove the following almost everywhere modi-
cation of (S3) and (S4). As usual, s. t S and all sets are in F .
a. If s = t ja. e. on 1, then
s
T
(s) =
s
T
(t ) for every J _ 1.
b. If s 6 t ja. e. on 1, then
s
T
(s) 6
s
T
(t ) for every J _ 1.
Proof. (a) Let 1
1
= {o 1 : s(o) = t (o)], and 1
2
= 1 1
1
. Then j(1
2
) = 0.
For an arbitrary J _ 1, dene
J
1
= J 1
1
. and J
2
= J 1
2
.
Then j(J
2
) = 0. Observe that
s
T
1
(s) =
s
T
1
(t ) by (S3), and
s
T
2
(s) =
s
T
2
(t ) = 0
by (S2)(a). It follows from (S7) that

s
T
(s) =
s
T
1
(s)
s
T
2
(s) =
s
T
1
(s) =
s
T
1
(t ) =
s
T
1
(t )
s
T
2
(t ) =
s
T
(t ).
(b) Let 1
1
= {o 1 : s(o) 6 t (o)], and 1
2
= 1 1
1
; then j(1
2
) = 0. For an
arbitrary J _ 1, dene J
1
= J 1
1
and J
2
= J 1
2
. Then
s
T
1
(s) 6
s
T
1
(t )
and
s
T
2
(s) =
s
T
2
(t ) = 0. By (S7) we get the result. |L
6.2 Stage Two: Nonnegative Functions
I Exercise 6.10 (6.2.1). Prove
a. (N3
t
) If = g ja. e. on 1, then
n
T
( ) =
n
T
(g) for any J _ 1.
b. (N4
t
) If 6 g ja. e. on 1, then
n
T
( ) 6
n
T
(g) for any J _ 1.
SECTION 6.2 STAGE TWO: NONNEGATIVE FUNCTIONS 97
c. (N7
t
) If .
1
.
2
. . . . N and = lim
n

n
ja. e. on 1, then
n
T
( ) =
lim
n

n
T
(
n
).
d. If 6 M ja. e. on 1 for some M 0. o|, then
n
T
( ) 6 Mj(1).
Proof. (a) Let 1
2
= {o 1 : (o) = g(o)] and 1
1
= 1 1
2
. Let J
1
= J 1
1
and J
2
= J 1
2
. Then j(J
2
) 6 j(1
2
) = 0, i.e., j(J
2
) = 0. Thus,

n
T
( ) =
n
T
1
LT
2
( ) =
n
T
1
( )
n
T
2
( ) =
n
T
1
( ) =
n
T
1
(g) =
n
T
(g).
(b) Similar to (a) and so is omitted.
(c) Let 1
2
= {o 1 : (o) = lim
n

n
(o)] and 1
1
= 1 1
2
. Then j(1
2
) = 0,
and so

n
T
( ) =
n
T
1
( )
n
T
2
( ) =
n
T
1
( ) = lim
n

n
T
1
(
n
) = lim
n

n
T
(
n
).
(d) Let 1
2
= {o 1 : (o) > M] and 1
1
= 1 1
2
. Then 6 M on 1
1
and
j(1
2
) = 0. Thus,

n
T
( ) =
n
T
1
( ) 6
n
T
1
(M) = Mj(1
1
) = Mj(1). |L
I Exercise 6.11 (6.2.2). It was claimed in (N5) that
n
T
(c ) = c
n
T
( ) for every
c 0. o). This result in fact holds in c = o as well:
n
T
(o ) = o
n
T
( ).
Therefore, (N5) holds for all c 0. o|. Similarly, we may allow the numbers
c
1
. . . . . c
n
to be in 0. o| in the statement (N8).
Proof. If = 0 ja. e. on 1, then o = 0 ja. e. on 1 and so
n
T
(o ) =
o
n
T
( ) = 0. So assume that there exists J F withJ _ 1 and j(J) > 0
such that > 0 on J. Then o = oon J. Thus,
n
T
(o ) >
n
T
(o) = o, and
o
n
T
( ) > o
n
T
( ) = o; that is,
n
T
(o ) = o
n
T
( ) = o. |L
I Exercise 6.12 (6.2.3). This exercise concerns Fatous Lemma.
a. Let {
n
] denote a sequence of F -sets. Show that j(liminf
n
) 6 liminf j(
n
)
by using Fatous Lemma applied to the sequence of indicator functions
{1

n
]
o
n=1
.
b. Consider (R. B. z). If s
n
= n
2
1
0,1{nj
for each n N, then
n
(liminf s
n
) = 0
while liminf
n
(s
n
) = o,hence strict inequality may hold in Fatous Lemma.
c. In (b), with 1 B, the sequence 1
T
. 1 1
T
. 1
T
. 1 1
T
. . . . provides another
example where strict inequality holds in Fatous Lemma.
d. It is not necessarily the case that limsup
n
T
(
n
) 6
n
T
(limsup
n
) if j(1) =
o. However, if j(1) < o, the inequality holds, hence we have

n
T
(liminf
n
) 6 liminf
n
T
(
n
) 6 limsup
n
T
(
n
) 6
n
T
(limsup
n
)
by putting everything together.
98 CHAPTER 6 THE LEBESGUE INTEGRAL
Proof. (a) It is evident that {1

n
] _ S _ Nand liminf
n
1

n
Nsince {
n
] _ F .
For every n N, let
g
n
= inf{1

n
. 1

nC1
. . . .].
Then the sequence {g
n
] is nondecreasing and so lim
n
g
n
exists. Thus,
lim
n
g
n
= sup
n
g
n
= sup
n
inf
n>n
1

m
= liminf
n
1

n
.
and g
n
6 1

n
for all n N. Also note that
n
(1

n
) =
s
(1

n
) = j(
n
), and
which implies that

n
(lim
n
g
n
) =
n
(liminf
n
1

n
) =
n
(1
liminf
n

n
) = j
_
liminf
n

n
_
.
where the second equality is from Exercise 2.14 (p. 36). Invoking Lebesgues
Monotone Convergence Theorem (MCT), we have
j
_
liminf
n

n
_
=
n
(lim
n
g
n
) = lim
n

n
(g
n
) = liminf
n

n
(g
n
)
6 liminf
n

n
(1

n
)
= liminf
n
j(
n
).
(b) We rst show that
liminf
n
s
n
(o) =

0 if o = 0
o if o = 0.
Suppose that there exists o = 0 such that liminf
n
s
n
(o) = > 0. Then for
an arbitrary c (0. ), there exists N N such that s
n
(o) > c for all n > N.
However, when n is large enough, s
n
(o) = 0. A contradiction. Thus, liminf
n
s
n
=
0 za. e. on R.
Therefore,
n
(liminf
n
s
n
) = 0. Nevertheless,
n
(s
n
) = n
2
z0. 1,n| = n, and
so liminf
n
(s
n
) = lim
n
n = o.
(c) Write 1 1
T
as 1
T
c . Then liminf
n
{1
T
. 1
T
c . . . .] = 0 and so

n
(liminf
n
{1
T
. 1
T
c . . . .)] = 0.
However,
n
(1
T
) = z(1) and
n
(1
T
c ) = z(1
c
) imply that
liminf
n

n
(1
T
).
n
(1
T
c ). . . .
_
= min
_
z(1). z
_
1
c
_
_
.
(d) We rst extend Fatous Lemma: If there exists g 1
1
and
n
> g on 1, then

n
T
(liminf
n

n
) 6 liminf
n

n
T
(
n
).
In this case, we have
n
g > 0 on 1 and
SECTION 6.2 STAGE TWO: NONNEGATIVE FUNCTIONS 99

n
T
(liminf
n
(
n
g)) 6 liminf
n

n
T
(
n
g)
by Fatous Lemma. So

n
T
(liminf
n

n
)
n
T
(g) 6 liminf
n

n
T
(
n
)
n
T
(g).
The result follows by cancelling
n
T
(g).
Now if
n
6 g, then
n
> g 1
1
, and the extended Fatous Lemma gives

n
T
(liminf
n
(
n
)) 6 liminf
n

n
T
(
n
).
so that

n
T
(liminf
n
(
n
)) > liminf
n

n
T
(
n
):
that is,
n
T
(limsup
n

n
) > limsup
n

n
T
(
n
). |L
I Exercise 6.13 (6.2.4). Let denote an arbitrary nonempty set, and x atten-
tion upon a particular o
0
. Let F = 2

, and dene j: F

R by writing
j() = 1 if o
0
and j() = 0 if o
0
.
a. (. F . j) is a measure space.
b. Every : R is F ,B
+
-measurable.
c. Let 1 F and N. Then
n
T
( ) = (o
0
)1
T
(o
0
).
Proof. (a) It is evident that F is a o-eld, so it suces to show that j is
a measure on F . It is clear that j() > 0 for all F and j() = 0. To
see j is countably additive, let {
n
]
o
n=1
be a disjoint sequence of F -sets. If
o
0

_
o
n=1

n
, then o
0

n
for all n N; hence
j
_
_
o
_
n=1

n
_
_
= 0 =
o

n=1
j(
n
).
Otherwise, if there exists
n
such that o
0

0
, then o
0

_
o
n=1

n
and so
j
_
_
o
_
n=1

n
_
_
= 1 =
o

n=1
j(
n
).
(b) Every : R is trivially F ,B
+
-measurable since F = 2

: for every
T B
+
, we have
-1
(T) F .
(c) If o
0
1, then j(1) = 1
T
(o
0
) = 0; thus,
n
T
( ) = (o
0
)1
T
(o
0
) = 0. If
o
0
1, then j(1) = 1
T
(o
0
) = 1. Hence,

n
T
( ) =
n
o
0

( )
n
T~o
0

( ) =
n
o
0

( ) = (o
0
). |L
100 CHAPTER 6 THE LEBESGUE INTEGRAL
I Exercise 6.14 (6.2.5). Let denote an uncountable set, and let F = { _
: is amc or
c
is amc]. Dene j: F

R for all F by writing j() = 0
if is amc and j() = 1 if
c
is amc.
a. (. F . j) is a measure space.
b. :

R is F ,B
+
-measurable i there is c

R [depending on ] with = c
ja. e. on .
c. Let 1 F and N, then
n
T
( ) = c j(1), where c is the constant such
that = c ja. e. on [as given in (b)].
Proof. (a) (i) F since
c
= is amc. (ii) If F , then either is amc or

c
is amc. If is amc, then
c
F since (
c
)
c
= is amc; if
c
is amc, then

c
F . (iii) Let {
n
] _ F . Then either each
n
is amc or at least one
c
n
is amc.
In the rst case,
_

n
F since countable unions of amc sets is itself amc; in
the second case, let us assume that
c
1
is amc. We have (
_

n
)
c
=
_

c
n
_
c
1
,
so (
_

n
)
c
is amc. It follows that
_

n
F as well.
(b) First suppose that = c ja. e. for some c

R; that is, there exists N F
with j(N) = 0 such that (o) = c for all o N. By denition, N is
amc; thus, every subset of N is in F . With this observation, we see that is
F ,B
+
-measurable.
Conversely, suppose that is F ,B
+
-measurable. Let
C = {t

R :
-1
o. t | is amc].
Note that
-1
(o) F . If
-1
(o)|
c
=
-1
(o. o| is amc, then = o
ja. e.. So we suppose that o C. Also, if o < b < a with a C, then
b C since
-1
o. b| _
-1
o. a|. Thus, C = {o] or C is some type of
unbounded interval containing o.
v If C = {o], then for any t > o, the set
-1
o. t | is not amc, and so
is not in F ; hence (
-1
o. t |)
c
=
-1
(t. o| is amc and is in F since is
F ,B
+
-measurable. Hence, j({o : (o) = o]) = 0; that is, = o
ja. e..
v If C is an interval containing o, let c = supC, so that o < c 6 o. Let
t
1
6 t
2
6 be such that t
n
c and t
n
< c for each n N. Then
-1
o. t
n
|
is amc for each n N. Thus,
_
n

-1
o. t
n
| =
-1
_
_
o. t
n
|
_
=
-1
o. c)
is amc, i.e., c C. If c = o, then =
-1
o. o| is amc; but is uncount-
able so we get a contradiction. Hence, o < c < o. Now, for every J > c,
the set
-1
o. J| is not amc; hence,
-1
(J. o| is amc. Let t
1
> t
2
> be
such that t
n
c and t
n
> c for each n N. Then
-1
(t
n
. o| is amc for each
SECTION 6.2 STAGE TWO: NONNEGATIVE FUNCTIONS 101
n N, and so
-1
(c. o| is amc. In sum, the sets
-1
o. c) and
-1
(c. o|
are both amc; thus,
j({o : (o) = c]) = j(
-1
o. c) L
-1
(c. o|) = 0:
that is, = c ja. e..
(c) Let N F be the set such that j(N) = 0 and (o) = c on N. Let
1
1
= 1 N and 1
2
= 1 1
1
. Then j(1
1
) = 0 and

n
T
( ) =
n
T
1
(c) = cj(1
1
) = cj(1). |L
I Exercise 6.15 (6.2.6). Let denote an arbitrary nonempty set, let _ ,
and let : 0. o| be given. Write
+

o
(o) = sup
_
_
_

oT
(o) : J _ , J nite
_
_
_
.
a. Suppose = {o
1
. . . . . o
n
]. Then

+
o
(o) =

n
i=1
(o
i
), hence the deni-
tion above is consistent with what were used to in the nite case.
b. Suppose = {o
1
. o
2
. . . .] [a countable set]. Then

+
o
(o) =

o
i=1
(o
i
).
Proof. (a) It is easy to see that

n
i=1
(o) 6

+
o
(o) since _ and
is nite here. For the converse inclusion, observe that every J _ is nite;
thus,

oT
(o) 6

o
(o) since (o) > 0. We thus have

+
o
(o) 6

o
(o).
(b) Consider a sequence {
n
]
o
n=1
with
n
= {o
1
. . . . . o
n
] for all n N. Then

n
] . Also observe that for any nite set J _ , there exists
n
containing
J. Thus,
+

o
(o) = sup
_
_
_

o
n
(o)
_
_
_
= lim
n

o
n
(o) =
o

i=1
(o
i
). |L
I Exercise 6.16 (6.2.7). Let denote a nonempty set, let F = 2

, and let
j: F

R be such that j() = the number of points in when is nite, and
j() = o otherwise.
a. (. F . j) is a measure space. The measure j is called the counting measure
since j counts the number of points in each F -set.
b. Every :

R is F ,B
+
-measurable.
c. Given any 1 F and N, we have
n
T
( ) =

+
oT
(o).
Proof. (a) and (b) are straightforward, so I just do (c). If S and > 0 only
on a nite subset of 1, then
n
T
( ) =
s
T
( ) =

+
oT
(o).
102 CHAPTER 6 THE LEBESGUE INTEGRAL
Now let N and let J _ 1 be nite. Then 1
T
S and > 0 only on a
nite subset of J; hence

n
T
( ) >
n
T
( 1
T
) =
n
T
( ) =
+

oT
(o) =

oT
(o).
Since J is an arbitrary nite subset of 1, we have
n
T
( ) >

+
oT
(o). This
gives one inequality.
If

+
oT
(o) = o, the previous inequality forces the result, so we may
assume that

+
oT
(o) < o. For any s S
(
we have
+

oT
s(o) 6
+

oT
(o) < o:
from this deduce that s > 0 only on a nite subset of 1; hence, we may nd
nite J _ 1 with s = 0 on 1 J. Then

s
T
(s) =
+

oT
s(o) =
+

oT
s(o) 6
+

oT
(o) 6
+

oT
(o).
Since the above holds for any s S
(
, it follows that

n
T
( ) = sup
xS
f

s
T
(s) 6
+

oT
(o). |L
I Exercise 6.17 (6.2.12). Let denote a nonempty set, and let F denote a
o-eld on . For each n N, let j
n
denote a measure with domain F . For each
n N, let

n
i=1
j
i
denote the measure that assigns the value

n
i=1
j
i
() to each
F .
a. Let s S and n N. Then
s
T
(s:

n
i=1
j
i
) =

n
i=1

s
T
(s: j
i
).
b. Let N and n N. Then
n
T
( :

n
i=1
j
i
) =

n
i=1

n
T
( : j
i
).
Proof. (a) Let s =

n
}=1
c
}
1

j
S and n N. Then

s
T
(s:
n

i=1
j
i
) =
n

}=1
c
}

_
_
n

i=1
j
i
_

}
1
_
_
_
=
n

i=1
n

}=1
c
i
j
i
_

}
1
_
=
n

i=1

s
T
(s: j
i
).
(b) Let N. By (MF18), there exists a nondecreasing sequence {s
k
]
o
k=1
S
such that s
n
. Then by Lebesgues Monotone Convergence Theorem,
SECTION 6.2 STAGE TWO: NONNEGATIVE FUNCTIONS 103

n
T
( :
n

i=1
j
i
) =
n
T
(lim
n
s
n
:
n

i=1
) =
s
T
(lim
n
s
n
:
n

i=1
j
i
) = lim
n

s
T
(s
n
:
n

i=1
j
i
)
= lim
n
n

i=1

s
T
(s
n
: j
i
)
=
n

i=1
lim
n

s
T
(s
n
: j
i
)
=
n

i=1

n
T
( : j
i
). |L
I Exercise 6.18 (6.2.14). Keep (. F . j) general. Suppose that F
0
_ F is a
o-eld on , and let j
0
denote the restriction of j to j
0
. If a nonnegative F ,B
+
-
measurable function dened on also happens to be F
0
,B
+
-measurable in
addition, then
n
T
( : j) =
n
T
( : j
0
) for every 1 F
0
.
Proof. First consider the case of = 1
T
where 1 F
0
. Then
n
T
( : j) =
j(1) = j
0
(1) =
n
T
( : j
0
). Next, let =

n
i=1
c
i
1

i
S. Then

n
T
( : j) =
n

i=1
c
i

n
T
(1

i
: j) =
n

i=1
c
i

n
T
(1

i
: j
0
) =
n
T
( : j
0
).
Finally, let N. Then there exists a nondecreasing sequence {s
n
]
o
n=1
S
such that s
n
. Hence,

n
T
( : j) =
n
T
(lim
n
s
n
: j) = lim
n

n
T
(s
n
: j) = lim
n

n
T
(s
n
: j
0
) =
n
T
( : j
0
). |L
I Exercise 6.19 (6.2.16, N17). Let denote a nonnegative F
t
,B
+
-measurable
function. We have the two equalities

n
( T : j) =
n
( : j T
-1
) (6.9)
and

n
T
1
(
0
)
( T : j) =
n

0 ( : j T
-1
) V
t
F
t
. (6.10)
Proof. We have
(. F . j) (
t
. F
t
. j T
-1
)
(

R. B
+
.

z)

T
T

v Let = 1

0 , where
t
F
t
. Then T = 1
T
1
(
0
)
, and

n
(1

0 T : j) =
n
(1
T
1
(
0
)
: j) = j(T
-1
(
t
)) = (j T
-1
)(
t
)
=
n
(1

0 : j T
-1
).
104 CHAPTER 6 THE LEBESGUE INTEGRAL
v Let =

n
i=1
c
i
1

0
i
S(
t
. F
t
. j T
-1
). Then T =

n
i=1
c
i
1
T
1
(
0
i
)
. It
follows from (N8) that

n
( T : j) =
n
(
n

i=1
c
i
1
T
1
(
0
i
)
: j) =
n

i=1
c
i

n
(1
T
1
(
0
i
)
: j)
=
n

i=1
c
i

n
(1

0
i
: j T
-1
)
=
n
(
n

i=1
c
i
1

0
i
: j T
-1
)
=
n
( : j T
-1
).
v Let N(
t
. F
t
. j T
-1
). By (MF18), there exists a nondecreasing sequence
{s
n
] _ S(
t
. F
t
. jT
-1
) such that s
n
. Thus, {s
n
T ] _ S(
t
. F
t
. jT
-1
)
is a nondecreasing sequence and s
n
T T . It follows from MCT that

n
( T : j) =
n
(lim
n
s
n
T : j) = lim
n

n
(s
n
T : j)
= lim
n

n
(s
n
: j T
-1
)
=
n
(lim
n
s
n
: j T
-1
)
=
n
( : j T
-1
).
Replace by 1

0 . It suces to show that


( 1

0 ) T = ( T )1
T
1
(
0
)
.
Note that 1

0 is dened on
t
, while 1
T
1
(
0
)
is dened on . For an arbitrary
o , we have
( 1

0 ) T |(o) = (T(o)) 1

0 (T(o)) = ( T )(o) 1
T
1
(
0
)
(o)
= ( T )1
T
1
(
0
)
|(o). |L
6.3 Stage Three: General Measurable Functions
I Exercise 6.20 (6.3.1). Let 1 F be such that j(1) < o, and let M be
such that = 0 on 1
c
and m 6 6 M on 1, where m. M R. Then
g
( )
exists and is nite. Furthermore, we have mj(1) 6
g
( ) 6 Mj(1).
Proof. Let M
t
= max{[M[. [m[]. Then

.
-
6 M
t
on 1. It follows that

n
T
(
-
).
n
T
(

) 6
n
T
(M
t
) = M
t
j(1) < o:
hence,
g
T
( ) eaif. The second claim follows (G6) since that m. M M(1). |L
SECTION 6.3 STAGE THREE: GENERAL MEASURABLE FUNCTIONS 105
I Exercise 6.21 (6.3.2). Let 1 F , c > 0 and let M be such that
g
T
( )
exists and is nite. There exists a subset J _ 1 such that j(J) < o and
[
g
T
( )
g
T
( )[ < c.
Proof. First let N. We rst show that for all . > 0,
j({o 1 : (o) > .]) < o.
Suppose there exists . > 0 such that j({o 1 : (o) > .]) = o. Then

n
T
( ) >
n
oT:((o)>x
(.1
T
) = .j({o 1 : (o) > .]) = o.
A contradiction. For each n N, let
1
n
= {o 1 : (o) > 1,n].
Then for each n N, we have 1
n
] {o 1 : (o) > 0] = 1 and j(1
n
) < o. It
follows that

g
T
( ) =
n
T
( ) =
n
lim
n
T
n
( ) = lim
n

n
T
n
( ) = lim
n

g
T
n
( ).
Thus, there exists 1
1
such that [
g
T
( )
g
T
N
( )[ < c. Let J = 1
1
and we are
done.
Next let M. Since
g
T
( ) < o, we know that
g
T
(

) < oand
g
T
(
-
) <
o. By the previous argument, there exist J
t
. J
tt
_ 1 such that
[
g
T
(

)
g
T
0
(

)[ < c,2 and [
g
T
(
-
)
g
T
00
(
-
)[ < c,2.
Let J = J
t
L J
t
. We get
[
g
T
( )
g
T
( )[ = [
g
T
(

)
g
T
(
-
)
g
T
(

)
g
T
(
-
)[
6 [
g
T
(

)
g
T
(

)[ [
g
T
(
-
)
g
T
(
-
)[
6 [
g
T
(

)
g
T
0
(

)[ [
g
T
(
-
)
g
T
00
(
-
)[
< c. |L
IExercise 6.22 (6.3.3). is such that
g
T
( ) exists and is nite i for any c > 0
there are functions g and h in Msuch that h 6 6 g on 1 and
g
T
(g h) < c.
Proof. Suppose rst that
g
T
( ) eaif. Then both
n
T
(

) and
n
T
(
-
) are -
nite. Given c > 0, let c = c,4j(1). Let
h = c and g = c.
Then

g
T
(g h) =
g
T
(c,2j(1)) = c,2 < c.
Now suppose that for every c > 0 there exist g. h M such that h 6 6 g
on 1 and
g
T
(g h) < c. Since h 6 g, we have g h N, and so
g
T
(g h) =

n
T
(g h) = 0. Then g h = 0 ja. e. on 1, and so g and h are nite and g = h
106 CHAPTER 6 THE LEBESGUE INTEGRAL
ja. e. on 1, and so is nite ja. e. on 1. This proves that
g
T
( ) eaif on
1. |L
I Exercise 6.23 (6.3.4). Let .
1
.
2
. . . . denote a sequence of nonnegative
functions in M. For each n N and 1 F , dene v
n
(1) =
g
T
(
n
: j) and
v(1) =
g
T
( : j). Furthermore, assume that v(). v
1
(). v
2
(). . . . are nite
and
n
ja. e. on . Then
sup{[v(1) v
n
(1)[ : 1 F ] 6
g
([
n
[: j) 0 as n o.
Proof. Since .
1
.
2
. . . . N and
n
a.e.
on , we get
sup
TF
[v(1) v
n
(1)[ = sup
TF

g
T
( : j)
g
T
(
n
: j)

= sup
TF

g
T
(
n
: j)

6 sup
TF

g
T
([
n
[: j)
6
g
([
n
[: j)
0. |L
IExercise 6.24 (6.3.5). Let (. F . j) = (R
k
. B
k
. z
k
) with 1 F , and let M
be such that
g
( ) exists and is nite.
a. Suppose that {
n
] is a sequence of functions in Msuch that

n
(x) =

(x) if x is such that [(x)[ 6 n and [x[ 6 n


0 otherwise.
Then lim
n

g
T
(
n
) exists and equals
g
T
(lim
n

n
).
b. Reset everything in (a), and this time let
n
(x) = (x) exp([x[
2
,n) for each
x R
k
and n N. Then lim
n

g
T
(
n
) =
g
T
( ).
Proof. (a) Since
g
( ) eaif,
g
([ [) is nite by (G7). It is clear that [
n
[ 6 [ [
and
n
(see Figure 6.1), all the claims follow from DCT.
(b) Observe that
n
(see Figure 6.2). |L
IExercise 6.25 (6.3.6). Suppose that j() < o, {
n
] is a sequence of functions
in Msuch that there exists M R with [
n
[ 6 M for each n N, and let M
be such that
n
uniformly on 1. Then lim
n

g
T
(
n
) =
g
T
( ).
Proof. j() < o and M R

implies that
g

(M) = Mj() < o. Then for


each 1 F , we have
g
T
(M) < o. The claim then follows from the DCT since
uniform convergence implies pointwise convergence: lim
n

n
= . |L
SECTION 6.3 STAGE THREE: GENERAL MEASURABLE FUNCTIONS 107
0
.

1
0
.

2
Figure 6.1. (
1
and (
2
.
0
.
= .
3

1
3
Figure 6.2. (
n
-(
I Exercise 6.26 (6.3.7). Let {
n
]
o
n=1
denote a nondecreasing sequence of
functions in M such that
g
T
(
n
) exists and is nite for each n N and
sup
nN

g
T
(
n
) < o. Then
g
T
(lim
n

n
) exists, is nite, and equals lim
n

g
T
(
n
).
This is one form of Beppo Levis Theorem.
Proof. Let g
n
=
n

1
for all n N. Then {g
n
]
o
n=1
N, is nondecreasing,
g
n
] lim
n

1
, and lim
n

g
T
(g
n
) = lim
n

g
T
(
n

1
) = lim
n

g
T
(
n
)
g
T
(
1
) =
sup
n

g
T
(
n
)
g
T
(
1
) < o. Then by MCT,
g
T
(lim
n
g
n
) = lim
n

g
T
(g
n
). Since

g
T
(lim
n
g
n
) =
g
T
(lim
n

n

1
) =
g
T
(lim
n

n
)
g
T
(
1
).
lim
n

g
T
(g
n
) = lim
n

g
T
(
n

1
) = lim
n

g
T
(
n
)
g
T
(
1
).
and
g
T
(
1
) < o, we get the Beppo Levis Theorem. |L
I Exercise 6.27 (6.3.8). Let {
n
], {g
n
], and {h
n
] denote sequences of functions
in Msuch that
n
a.e.
, g
n
a.e.
g, and h
n
a.e.
h for some functions . g. h M.
108 CHAPTER 6 THE LEBESGUE INTEGRAL
Suppose for any {g. g
1
. g
2
. . . . . h. h
1
. h
2
. . . .] that
g
T
() exists and is nite.
Furthermore, suppose that lim
n

g
T
(g
n
) =
g
T
(g) and lim
n

g
T
(h
n
) =
g
T
(h). Also,
assume that g
n
6
n
6 h
n
for every n N.
a.
g
T
(
n
) exists and is nite for all n N,
g
T
( ) exists and is nite, and
lim
n

g
T
(
n
) =
g
T
( ).
b. DCT may be obtained from (a).
Proof. (a) Since
n
6 h
n
and
g
T
(h
n
) < o, we have
g
T
(
n
) 6
g
T
(h
n
) < o for
all n N; that is,
g
T
(
n
) exists and is nite for all n N.
Since
n
6 h
n
for all n, we get lim
n

n
6 lim
n
h
n
, i.e., 6 h; since
g
T
(h) < o,
we have
g
T
( ) exists and is nite.
Since
n
6 h
n
for all n, we have h
n

n
> 0 ja. e.. Fatous Lemma yields

g
T
(h)
g
T
( ) =
g
T
(h ) =
g
T
(lim
n
(h
n

n
))
=
g
T
(liminf
n
(h
n

n
))
6 liminf
n

g
T
(h
n

n
)
= liminf
n
_

g
T
(h
n
)
g
T
(
n
)
_
=
g
T
(h) liminf
n
_

g
T
(
n
)
_
:
that is,
g
T
( ) > limsup
g
T
(
n
).
Finally, observe that g
n
6
n
yields
n
g
n
> 0 ja. e.. Applying Fatous
Lemma once again, we obtain

g
T
( )
g
T
(g) =
g
T
( g) =
g
T
(lim
n
(
n
g
n
))
=
g
T
(liminf
n
(
n
g
n
))
6 liminf
n

g
T
(
n
g
n
)
= liminf
n

g
T
(
n
)
g
T
(g):
that is,
g
T
(
n
) 6 liminf
n

g
T
(
n
). We thus get lim
n

g
T
(
n
) =
g
T
( ).
(b) Observe that if [ [ 6 g, then g 6 6 g. By (a) we get DCT. |L
I Exercise 6.28 (6.3.12). Suppose that
g
_
1
nD1
T
n
( ) exists and is nite, where
{1
n
]
o
n=1
is a disjoint sequence of F -sets. Then
g
_
1
nD1
T
n
( ) =

o
n=1

g
T
n
( ), and
the convergence of the series is absolute.
Proof. Since
g
_
1
nD1
T
n
( ) eaif, (G1-b) implies that each of
g
T
n
( ) eaif. There-
fore,
g
_
1
nD1
T
n
(

).
g
_
1
nD1
T
n
(
-
).
g
T
n
(

).
g
T
n
(
-
) < o. We have
SECTION 6.4 STAGE FOUR: ALMOST EVERYWHERE DEFINED FUNCTIONS 109

g
_
1
nD1
T
n
( ) =
g
_
1
nD1
T
n
(

)
g
_
1
nD1
T
n
(
-
)
=
o

n=1

n
T
n
(

)
o

n=1

n
T
n
(
-
)
=
o

n=1
_

n
T
n
(

)
n
T
n
(
-
)
_
=
o

n=1

g
T
n
( ).
We now show that

o
n=1
[
g
T
n
( )[ converges. Since
g
T
n
( ) exists (and is
nite), by (G7) we have [
g
T
n
( )[ 6
g
T
n
([ [) for any n N; since
g
T
n
( ) eaif
for any n N, we know that
g
T
n
([ [) eaif for any n N. Therefore,
o

n=1

g
T
n
( )

6
o

n=1

g
T
n
([ [) =
g
_
1
nD1
T
n
([ [) < o
since
g
_
1
nD1
T
n
( ) eaif. |L
I Exercise 6.29 (6.3.14). Suppose that {
n
]
o
n=1
is a sequence of elements of N
converging to some N. Furthermore, assume that there is 0 6 M < o such
that
g
(
n
) 6 M for each n N. Then
g
( ) exists, is nite, and is no more
than M.
Proof. Let
n
. Then lim
n

n
= liminf
n

n
. By Fatous Lemma,

g
( ) =
g
(liminf
n

n
) 6 liminf
n

g
(
n
) 6 limsup
n

g
(
n
) 6 M. |L
6.4 Stage Four: Almost Everywhere Defined Functions
I Exercise 6.30 (6.4.1). (L10) If 1
1
(1) and [g[ 6 ja. e. on 1, then
g 1
1
(1). Also, any that is bounded ja. e. on a set 1 with j(1) < oand
is zero ja. e. on 1
c
is in 1
1
(1).
(L16) We have the following, where
t
F
t
:
a. If > 0, then
_
(
1
(
0
)
dj =
_

0
d
_
j
-1
_
.
b. For general ,
_
(
1
(
0
)
dj exists and is nite i
_

0
d
_
j
-1
_
exists
and is nite, and in this case equality obtains.
Proof. We rst prove (L10). 1
1
(1) ==
g
T
(
+
) < o; since [g[ 6
ja. e., we have g
+
6
+
ja. e. and g
+-
6
+
ja. e.. Then the conclusion
follows (G9). |L
7
INTEGRALS RELATIVE TO LEBESGUE MEASURE
7.1 Semicontinuity
I Exercise 7.1 (7.1.1). (SC4b) If (x) = o, then is USC at x i lim
y-x
(y) =
o.
(SC7b)

is USC, and is the minimal USC function > .
(SC9) Let denote a generic nonempty index set. For each , suppose that

is a function from R
k
into

R. We have the following:
a. If

is LSC for each , then sup

is LSC.
b. If

is USC for each , then inf

is USC.
Proof. (SC4b) Assume is USC at x. Pick t > (x) = o. Then there is
> 0 such that (y) < t for each y T (x. ). Since t is arbitrary, we
have lim
y-x
(y) = o. Conversely, assume that lim
y-x
(y) = o and
pick any t > (x) = o. Then there is > 0 such that (y) < t for each
y T (x. ). Since t is generic, is USC at x by denition.
(SC7b) We show that inf
>0
sup
yB(x,)

(y) 6

(x) for each x; then

is
USC by (SC6). Suppose there is x so that the preceding inequality fails, then
there exists t such that inf
>0
sup
yB(x,)

(y) > t >

(x). It follows that
sup
yB(x,)

(y) > t for any > 0, and therefore, there exists y T (x. )
so that

(y) > t for any T (x. ). Now consider an open ball of y, T (y. r) _
T (x. ). We have

(y) = inf

0
>0
sup
zB(y,
0
)
(z) 6 sup
zB(y,i)
(z) 6 sup
zB(x,)
(z).
that is, for any T (x. ), we have sup
zB(x,)
(z) >

(y) > t . But this implies
that

(x) = inf
>0
sup
zB(x,)
(z) > t.
A contradiction.
111
112 CHAPTER 7 INTEGRALS RELATIVE TO LEBESGUE MEASURE
With this outcome, we can show that is LSC. Take any x D
(
. Then
(x) = inf
>0
sup
yB(x,)
_

_
(y) = sup
>0
inf
yB(x,)
(y) = (x).
Since is USC, ( ) is LSC, that is, = ( ) = ( ) is LSC.
(SC9a) Take any x

R
k
and t < sup

(x). Then there exists


t
such
that

0 (x) > t ; since

0 is LSC, there is T (x. ) such that

0 (y) > t for all


y T (x. ). Since sup

(y) >

0 (y), we know that sup

is LSC.
(SC9b)

is USC implies that

is LSC; then sup

) = inf

is
LSC. Hence, inf

is USC.
|L
I Exercise 7.2 (7.1.2). Let 1

R
k
.
a. 1 is open i 1
T
is LSC.
b. 1 is closed i 1
T
is USC.
c. We have 1
T
= 1
T
and 1
T
= 1
T
.
Proof. |L
8
THE L
P
SPACES
8.1 L
p
Space: The Case 1 6 p < C1
I Exercise 8.1 (8.1.1). Pick A. B L
]
. Then A B L
]
and [A B[
]
6
[A[
]
[B[
]
.
Proof. Let h AB, so that h = g for some A and g B. Now
_
[h[
]
=
__
[ g[
]
_
]{]
6
_
__
[ [
]
_
1{]

__
[g[
]
_
1{]
_
]
< o
by Minkowskis Inequality, and hence A B L
]
. The above display also
implies that
[AB[
]
]
6
_
[A[
]
[B[
]
_
]
.
i.e., [AB[
]
6 [A[
]
[B[
]
. |L
I Exercise 8.2 (8.1.2). Prove the Cauchy-Schwarz inequality

k=1
a
k
b
k

6
p
k

k=1
a
2
k
p
n

k=1
b
2
k
.
Proof. Let =
t
= 2; then Hlders Inequality becomes

_
g

6 [ [
2
[g[
2
. (8.1)
Let = {o
1
. . . . . o
n
], F = 2

, j be the counting measure, (o


i
) = a
i
, and
g (o
i
) = b
i
. Then [
_
g[ = [

o
i

(o
i
) g (o
i
) [ = [

n
k=1
a
k
b
k
[,
[ [
2
=
__
[ [
2
_
1{2
=
_
_

o
i

[ (o
i
)[
2
_
_
1{2
=
p
n

k=1
a
2
k
.
113
114 CHAPTER 8 THE 1
P
SPACES
and similarly for [g[
2
. Put these into (8.1) and we get the Cauchy-Schwarz
inequality. See Shirali and Vasudeva (2006, Theorem 1.1.4) for a direct proof.
|L
IExercise 8.3 (8.1.3). Let 1 <
1
. . . . .
n
< obe such that 1,
1
1,
n
=
1, and pick functions
1
1
]
1
. . . . .
n
1
]
n
. We wish to generalize Hlders
Inequality by showing that

n
i=1

i
1
1
and [
_
n
i=1

i
[ 6

n
i=1
[
i
[
]
i
.
a. Show rst that a
1
a
n
6 a
]
1
1
,
1
a
]
n
n
,
n
by generalizing the calculus
result given in the section. [Here 0 6 a
1
. . . . . a
n
< o.]
b. If [
1
[
]
1
= 0 or . . . or [
n
[
]
n
= 0, the claim is trivial.
c. Use (a) to prove the claim when [
1
[
]
1
= = [
n
[
]
n
= 1.
d. Prove the claim when [
1
[
]
1
. . . . . [
n
[
]
n
are positive.
Proof. (a) This is the arithmetic mean-geometric mean inequality, or AM-GM
inequality, for short. Since ln is concave, we have
n

i=1
1

i
ln.
i
6 ln
_
_
n

i=1
1

i
.
i
_
_
.
i.e.,
ln
_
_
n

i=1
.
1{]
i
i
_
_
6 ln
_
_
n

i=1
.
i

i
_
_
==
n

i=1
.
1{]
i
i
6
n

i=1
.
i

i
.
Let .
1{]
i
i
= a
i
, then .
i
= a
]
i
i
and we have the desired result.
(b) Let [
i
[
]
i
= 0; then
i
= 0 ja. e. on . But then

n
i=1

i
= 0 ja. e. on ,
hence

n
i=1

i
1
1
and the desired inequality in this case is actually the trivial
equation 0 = 0.
(c) If [
1
[
]
1
= = [
n
[
]
n
= 1, observe that

i=1

i
(o)

=
n

i=1
[
i
(o)[ 6
n

i=1
1

i
[
i
(o)[
]
i
V o .
by the Am-GM inequality. Therefore,
_

n
i=1

6
n

i=1
1

i
_
[
i
[
]
i
=
n

i=1
1

i
[
i
[
]
i
= 1.
This shows that

n
i=1

i
1
1
.
(d) Dene
+
i
=
i
,[
i
[
]
i
for any i = 1. . . . . n. We have
__

+
i

]
i
_
1{]
i
=
__

i
,[
i
[
]
i

]
i
_
1{]
i
=
_
1
[
i
[
]
i
]
i
_
[
i
[
]
i
_
1{]
i
= 1.
SECTION 8.1 1
P
SPACE: THE CASE 1 6 1 ~ o 115
which shows that
+
i
1
]
i
and [
+
i
[
]
i
= 1. By (c),

n
i=1

+
i
1
1
and
_
[

n
i=1

+
i
[ 6 1. Since

n
i=1

i
=
_
n
i=1
[
i
[
]
i
_ _
n
i=1

+
i
_
, we have
_

i=1

=
_
_
n

i=1
[
i
[
]
i
_
_
_

i=1

+
i

6
n

i=1
[
i
[
]
i
< o.
giving

n
i=1

i
1
1
, and

_ n

i=1

6
_

i=1

=
_
_
_
_
_
_
n

i=1

i
_
_
_
_
_
_
1
6
n

i=1
[
i
[
]
i
. |L
I Exercise 8.4 (8.1.4). We have equality in Hlders Inequality i there are
nonnegative numbers and T, not both zero, with [ [
]
= T[g[
]
0
ja. e. on
.
Proof. We have equality in Hlders Inequality i
[ [
[ [
]

[g[
[g[
]
0
=
1

[ [
]
[ [
]
]

t
[g[
]
0
[g[
]
0
]
0
ja. e. on .
which holds i the AM-GM holds equality, that is,
[ [
]
[ [
]
]
=
[g[
]
0
[g[
]
0
]
0
j-a.e. on . |L
I Exercise 8.5 (8.1.5). Given 1
]
_
1 < < o
_
, there is g 1
]
0
with
[g[
]
0 = 1 and
_
g = [ [
]
.
Proof. Let g = (, [ [
]
)
]-1
. Then
_
[g[
]
0
=
_

_

[ [
]
_
]-1

]
0
=
_

[ [
]

]
=
1
[ [
]
]
_
[ [
]
=
[ [
]
]
[ [
]
]
= 1.
i.e., [g[
]
0 = 1. We also have
_
[g[ =
_
[ [
]
[ [
]-1
]
= [ [
]
. |L
I Exercise 8.6 (8.1.6). We now explore conditions for equality in Minkowskis
Inequality. Let . g 1
]
.
a. When = 1, [ g[
]
= [ [
]
[g[
]
i there exists positive F ,B
+
-
measurable h > 0 dened on with h = g ja. e. on {o : (o)g(o) =
0].
b. For 1 < < o, equality obtains i there are nonnegative real numbers
and T, not both zero, such that = Tg ja. e. on .
116 CHAPTER 8 THE 1
P
SPACES
Proof. (a) When = 1, we have
[ g[
1
= [ [
1
[g[
1
==
_
[ g[ =
_
[ [
_
[g[
==
_
_
[ g[ [ [ [g[
_
= 0
== [ g[ = [ [ [g[ ja. e.
== JF ,B
+
-measurable h > 0 dened on
with h = g ja. e. on (o)g(o) = 0|.
(b) When 1 < < o, we have
[ g[
]
]
=
_
[ g[
]
=
_
[ g[ [ g[
]-1
+
6
_
[ [ [ g[
]-1

_
[g[ [ g[
]-1
++
6 [ [
]
[[ g[
]-1
[
]
0 [g[
]
[[ g[
]-1
[
]
0
=
_
[ [
]
[g[
]
_
[ g[
]-1
]
.
Hence, the Minkowskis Inequality holds with equality i (+) and (++) hold with
equality. The result follows from Exercise 8.4 that (++) immediately. |L
I Exercise 8.7 (8.1.7). Let 1 6 . q. r < o be such that 1,r = 1, 1,q. Let
1
]
and g 1
q
. Then g 1
i
and [g[
i
6 [ [
]
[g[
q
.
Proof. Let
t
= ,r and q
t
= q,r. Then 1,
t
1,q
t
= r, r,q = 1. Let

+
=
i
and g
+
= g
i
. Then
_
[
+
[
]
0
=
_
[
i
[
]{i
=
_
[ [
]
< o.
and
_
[g
+
[
q
0
=
_
[g
i
[
q{i
=
_
[g[
q
< o.
i.e.,
+
1
]
0
and g
+
1
q
0
. By the AM-GM inequality, for any o ,

+
(o)g
+
(o)

= [
+
(o)[ [g
+
(o)[ 6
1

t
[
+
(o)[
]
0

1
q
t
[g
+
(o)[
q
0
.
Integrate the left and right sides of the above display, obtaining
_
[g[
i
=
_
[
+
g
+
[ 6
1

t
_
[
+
[
]
0

1
q
t
_
[g
+
[
q
0
< o.
that is, g 1
i
and
+
g
+
1
1
. Then by Hlders Inequality, we have
[
+
g
+
[
1
6 [
+
[
]
0 [g
+
[
q
0 :
SECTION 8.2 THE RIESZ-FISCHER THEOREM 117
therefore,
_
[g[
i
=
_
[
+
g
+
[ 6
__
[
+
[
]
0
_
1{]
0 __
[g
+
[
q
0
_
1{q
0
=
_
__
[ [
]
_
1{]
__
[g[
q
_
1{q
_
i
=
_
[ [
]
[g[
q
_
i
:
that is, [g[
i
6 [ [
]
[g[
q
. |L
IExercise 8.8 (8.1.8). If 1 6 < o, Minkowskis Inequality gives

[ [
]
[g[
]

6
[ g[
]
for every . g 1
]
.
Proof. Write =
_
g
_
g. We rst show that g 1
]
when . g 1
]
.
_
[ g[
]
6
_
[ [
]

_
[g[
]
< o.
Then by the Minkowskis Inequality, we have
[ [
]
= [( g) g[
]
6 [ g[
]
[g[
]
.
Rearrange the above display and we get the desired result. |L
8.2 The Riesz-Fischer Theorem
I Exercise 8.9 (8.2.1). Return to the formal denition of 1
]
.
a. Write out the formal denition of convergence in 1
]
.
b. State and prove the formal version of the Riesz-Fischer Theorem.
Proof. (a) Let F. F
1
. F
2
. . . . 1
]
. {F
n
]
o
n=1
converges to F in 1
]
, F
n
1
p
F, if
and only if lim
n
[F
n
F[
]
= 0.
(b) Straightforward. |L
I Exercise 8.10 (8.2.2). Let (X. j) denote a generic metric space. Let C
b
(X)
denote the collection of continuous real-valued bounded functions on X. For
C
b
(X), write [ [ = sup
xA
[(.)[, the usual supremum norm. Then C
b
(X)
is a Banach space.
Proof. Let {
n
] be a Cauchy sequence in C
b
(X). Then for every c > 0, there
exists N
t
N such that for all n. m > N
t
, we have
[
n

n
[ = sup
xA
[
n
(.)
n
(.)[ < c.
118 CHAPTER 8 THE 1
P
SPACES
Therefore, for every . X, we get [
n
(.)
n
(.)[ < c for all n. m > N
t
; that is,
{
k
(.)] is Cauchy in R. The completeness of R yields

k
(.) (.).
for some (.) R. Now x n > N
t
. Since [ [ is continuous, we get
[
n
(.) (.)[ = lim
n-o
[
n
(.)
n
(.)[ 6 c.
Hence, for every n > N
t
, we have
[
n
[ = sup
xA
[
n
(.) (.)[ 6 c.
What has been just shown is that [
n
[ 0 as n o. Note that this
implies that
n
uniformly on X. Thus, is continuous since every
n
is
continuous. Also,
[ [ 6 [
n
[ [
n
[ < o.
Hence C
b
(X) and so C
b
(X) is a Banach space. |L
I Exercise 8.11 (8.2.3). A function on R
k
is said to vanish at innity i
(x) 0 as [x[ o. Show that the collection of continuous functions on R
k
that vanish at innity is a Banach space relative to the supremum norm given
in the previous exercise.
Proof. Let C
0
(R
k
) denote the collection of continuous functions on R
k
that
vanish at innity. We use an alternative denition (Rudin, 1986, Denition
3.16): A complex function on a locally compact Hausdor space X is said to
vanish at innity if to every c > 0 there exists a compact set 1 _ X such that
[(.)[ < c for all . 1.
Let {
n
] be a Cauchy sequence in C
0
(R
k
), i.e., assume that {
n
] converges
uniformly. Then its pointwise limit function is continuous. Given c > 0,
there exists an n so that [
n
[ < c,2 and there is a compact set 1 so that
[
n
(.)[ < c,2 outside 1. Hence [(.)[ < c outside 1, and we have proved that
vanishes at innity. Thus C
0
(R
k
) is complete. |L
IExercise 8.12 (8.2.4). Let C
c
(R
k
) denote the collection of continuous functions
on R
k
with compact support, and again consider the supremum norm. This
collection is dense in the collection in the previous exercise, but it fails to be a
Banach space.
Proof. Refer Hewitt and Stromberg (1965, 7) and Rudin (1986, p. 69-71). The
support of a (complex) function on a topological space X is the closure of
the set {. X : (.) = 0].
Given C
0
(R
k
) and c > 0, there is a compact set 1 so that [(.)[ < c
outside 1. Urysohns lemma (Rudin, 1986, 2.12) gives us a function g C
c
(R
k
)
SECTION 8.3 1
P
SPACE: THE CASE 0 ~ 1 ~ 1 119
such that 0 6 g 6 1 and g(.) = 1 on 1. Put h = g. Then h C
c
(R
k
) and
[ h[ < c. This proves that C
c
(R
k
) = C
0
(R
k
). |L
I Exercise 8.13 (8.2.5). A sequence {
n
] in 1
]
may converge in th mean to
some 1
]
but at the same time fail to converge pointwise to at any point in
. Therefore, convergence in 1
]
does not in general imply convergence ja. e.
Proof. Consider (0. 1|. B0. 1|. z). Consider the sequence
1
0,1{2j
. 1
1{2,1j
. 1
0,1{4j
. 1
1{4,1{2j
. 1
1{2,3{4j
. 1
3{4,1j
. 1
0,1{Sj
. . . . .
Then
n
1
p
0, but obviously (.) ,0 for all . 0. 1|. |L
8.3 L
p
Space: The Case 0 < p < 1
I Exercise 8.14 (8.3.1). Let . g 1
]
, where 0 < < 1. We know that g
1
]
by the Minkowski-like Inequality result given earlier.
a. We have (a b)
]
6 a
]
b
]
for every 0 < a. b < o.
b. From (a), we have
_
[ g[
]
6
_
[ [
]

_
[g[
]
.
c. If we write [ g[
]
]
for the distance between and g, then this distance
function is truly a metric, if we identify functions equal ja. e. on .
d. Writing [ g[
]
for the distance between and g does not dene a metric
on 1
]
.
Proof. (a) If 0 < a = b < o, we have
(a b)
]
= 2
]
a
]
6 2a
]
= a
]
b
]
.
Next we assume that 0 < a < b < o. Since 0 < < 1, the function .
]
dened
on (0. o) is concave. Write b as a convex combination of a and ab as follows:
b =
a
b
a
b a
b
(a b).
Then
b
]
=
_
a
b
a
b a
b
(a b)
_
]
6
a
b
a
]

b a
b
(a b)
]
:
that is,
(a b)
]
6
b
]1
a
]1
b a
6
(b a)(a
]
b
]
)
b a
= a
]
b
]
.
where the second inequality holds since
(b a)(a
]
b
]
) = b
]1
a
]1
ab
_
a
]-1
b
]-1
_
> b
]1
a
]1
.
120 CHAPTER 8 THE 1
P
SPACES
(b) It follows from (a) that
_
[ g[
]
6
_
([ [ [g[)
]
6
_
([ [
]
[g[
]
) =
_
[ [
]

_
[g[
]
.
(c) We use the informal denition. To see [ g[
]
]
is a metric on 1
]
, we need
to verify:
v 0 6 [ g[
]
]
< o for every . g 1
]
. It is true because by (b):
0 6 [ g[
]
]
=
_
[ g[
]
6
_
[ [
]

_
[g[
]
< o.
v [ [
]
]
= 0 for each 1
]
, and [ g[
]
]
= 0 forces = g ja. e. on
. The rst claim is obvious, so we focus on the second one. If [ g[
]
]
=
_
[ g[
]
= 0, then [ g[
]
= 0 ja. e., then = g ja. e.
v [ g[
]
]
= [g [
]
]
for every . g 1
]
. This is evident.
v [ h[
]
]
6 [ g[
]
]
[g h[
]
]
for every . g. h 1
]
. It also follows from
(b):
[ h[
]
]
=
_
[ h[
]
=
_
[( g) (g h)[
]
6
_
([ g[ [g h[)
]
6
_
[ g[
]

_
[g h[
]
= [ g[
]
]
[g h[
]
]
.
Thus, [ g[
]
]
is a metric on 1
]
when 0 < < 1.
(d) It follows from the Minkowski-like equality that the triangle inequality fails
for [ g[
]
when 0 < < 1. |L
I Exercise 8.15 (8.3.2). Consider the space = (0. 1) and let 0 < < 1. Write
B for the Borel subsets of (0. 1), and write z for Lebesgue measure restricted to
B. We will show that there exists no norm [ [ on 1
]
such that lim
k
[
k
[
]
= 0
forces lim
k
[
k
[ = 0.
a. Suppose that such a norm [ [ exists. Then there is C > 0 such that [ [ 6
C[ [
]
for each 1
]
.
Pick the minimal such C from (a).
b. There is 0 < c < 1 with
_
c
0
[ [
]
=
_
1
c
[ [
]
=
1
2
_
1
0
[ [
]
.
c. Let g = 1
(0,cj
and h = 1
(c,1j
, so that = g h. Then [g[
]
= [h[
]
=
2
-1{]
[ [
]
and [ [ 6 [g[ [h[ 6 C[g[
]
C[h[
]
= C 2
1-1{]
[ [
]
. Use the
minimality of C to obtain C 6 C2
1-1{]
, and deduce C = 0.
SECTION 8.3 1
P
SPACE: THE CASE 0 ~ 1 ~ 1 121
d. Conclude that [ [ = 0 for every 1
]
, and hence the assumption in (a)
entails a contradiction.
Proof. (a) Suppose that for every C > 0 there is 1
]
with [ [ > C[ [
]
.
Then for every k N there is
k
1
]
with [
k
[ > k[
k
[
]
. Dene g
k
=
k
,[
k
[
for each k N. Then [g
k
[ = 1 and for every k N
1 = [g
k
[ > k[g
k
[
]
.
i.e., [g
k
[ < 1,k. Hence, lim
k
[g
k
[
]
= 0, so lim
k
[g
k
[ = 0 by the assumption
that such a norm exists. But lim
k
[g
k
[ = 1.
(b) The function
_
x
0
[ [
]
is continuous and increasing with respect to .. The
claim follows immediately.
(c) It follows from (b) that [g[
]
]
= [h[
]
]
=
1
2
[ [
]
, i.e.,
[g[
]
= [h[
]
= 2
-1{]
[ [
]
.
Since [ [ is a norm, we have
[ [ = [g h[ 6 [g[ [h[ 6 C[g[
]
C[h[
]
= C2
1-1{]
[ [
]
.
The minimality of C implies that C 6 C2
1-1{]
. Hence, C = 0.
(d) By (a) and (c) we get [ [ 6 0[ [
]
= 0; that is, [ [ = 0 for all 1
]
. But
then [ [ is not a norm. A contradiction. |L
I Exercise 8.16 (8.3.3). Let 0 <
0
< o and let 1
]
0
be nonnegative. Let
1
1
= {o : (o) = 0], 1
2
= {o : 0 < (o) 6 1], and 1
3
= {o :
(o) > 1].
a. lim
]-0
C
_
T
2
[ [
]
= j(1
2
).
b. lim
]-0
C
_
T
3
[ [
]
= j(1
3
).
c. lim
]-0
C
_
[ [
]
= j({o : (o) = 0]).
Proof. (a) The function .
]
decreasing with respect to when 0 < . 6 1. By
MCT we have
lim
]-0
C
_
T
2
[ [
]
=
_
T
2
lim
]-0
C
[ [
]
=
_
T
2
1 = j(1
2
).
(b) It follows from DCT.
(c) Let {
n
] be a decreasing sequence converging to 0. Then
lim
n
_
[ [
]
= lim
n
_
_
T
2
[ [
]

_
T
3
[ [
]
_
= lim
n
_
T
2
[ [
]
lim
n
_
T
3
[ [
]
= j(1
2
) j(1
3
)
= j(o) = 0|. |L
122 CHAPTER 8 THE 1
P
SPACES
I Exercise 8.17 (8.3.4). Say j() = 1, and let 1
1
be nonnegative. Write
log 0 = o.
a.
_
log 6 log
_
if log 1
1
.
b. If log 1
1
, then
_
(log )

< o under the assumption 1


1
, so it must
be the case that
_
(log )
-
= o. Conclude that even if log 1
1
,
_
log
still exists and equals o, giving the inequality in (a).
c. (
i
1),r decreases to log as r 0

, hence lim
i-0
C(
_

i
1),r =
_
log .
d. Verify the inequalities
1
r
__

i
1
_
>
1
r
log
_

i
>
1
r
_
log
i
=
_
log .
e. Conclude that lim
i-0
C [ [
i
exists and equals exp(
_
log ). If log 1
1
, this
is interpreted as lim
i-0
C [ [
i
= 0.
Proof. (a) If [ [
1
=
_
= 0, then = 0 ja. e.. Hence,
_
log = log
_
=
o. Now assume that [ [
1
> 0. Since log . 6 . 1 when 0 6 . < o, we have
_
log

[ [
1
6
_ _

[ [
1
1
_
=
_

[ [
1

_
1 = 0:
hence,
_
log 6 log [ [
1
= log
_
.
(b) Observe that
(log )

(o) =

0 if (o) 0. 1|
log (o) if (o) (1. o|.
Also, log (o) < (o) 1 when o (1. o|. Since 1
1
, we have
_
< o.
Thus,
_
(log )

=
_
((o)>1j
log 6
_
((o)j>1
( 1) < o.
Therefore, it must be the case that
_
(log )
-
= o, and so
_
log =
_
(log )

_
(log )
-
= o.
(c) Fix an arbitrary o . We have
lim
i-0
C
(o)
i
1
r
= log (o):
hence, (
i
1),r log as r 0

, and consequently,
_


i
1
r
_
]
_
log
_
as r 0

.
SECTION 8.4 1
P
SPACE: THE CASE 1 = o 123
It follows from the MCT that
lim
i-0
C
_ _


i
1
r
_
=
_
lim
i-0
C
_

i
1
r
=
_

_
lim
i-0
C

i
1
r
=
_

_
log .
Since
_
< o and j() = 1, we get the desired result.
(d) The rst inequality follows from the fact
i
> 0 and under this case
log
i
6
i
1. The second inequality follows from (a) and (b).
(e) |L
8.4 L
p
Space: The Case p D C1
I Exercise 8.18 (8.4.2). Consider the o-nite measure space (. F . j).
a. 1
o
i there is a bounded F ,B
+
-measurable function g on such that
= g ja. e. on .
b. If 1
o
, then [ [
o
= inf{sup
o
[g(o)[ : g is as in (a)].
Proof. (a) First assume that there exists a bounded F ,B
+
-measurable func-
tion g on such that = g ja. e.. Then there exists M > 0 such that [g[ 6 M.
Hence, [ [ 6 M ja. e.; that is, ess sup 6 M, and so 1
o
.
Now suppose that 1
o
. Dene g on by letting
g(o) =

(o) if [(o)[ 6 ess sup


0 otherwise.
This g is bounded, F ,B
+
-measurable, and = g ja. e..
(b) We rst show that sup
o
[g(o)[ > [ [
o
for all g as in (a). Suppose that
sup
o
[g(o)[ < [ [
o
. Dene
:
_
o : [(o)[ > sup
o
[g(o)[
_
.
Then j() > 0; for otherwise [ [
o
6 sup
o
[g(o)[. But which means that
> g on and j() > 0. A contradiction. This shows that
inf
_
sup
o
[g(o)[ : g as in (a)
_
> [ [
o
.
124 CHAPTER 8 THE 1
P
SPACES
We next show the reverse inclusion. Let T : {o : [(o)[ 6 [ [
o
]; then
j(T
c
) = 0. Let g = 1
B
. Then g is bounded, F ,B
+
-measurable, and = g
ja. e.. Furthermore,
sup
o
[( 1
B
)(o)[ 6 [ [
o
.
This prove that [ [
o
{sup
o
: [g(o)[ : g as in (a)], and the proof is com-
pleted. |L
IExercise 8.19 (8.4.4). Quickly prove the = oversion of Hlders Inequality.
Proof. Let 1
o
and g 1
1
. By Claim 1 we get
[g[ 6 ess supg ja. e.
Observe that
ess supg = [g[ess sup = [g[ [ [
o
.
so we have
_
[g[ 6
_
[g[ [ [
o
= [ [
o
[g[
1
< o.
Hence, g 1
1
and [g[
1
6 [ [
o
[g[
1
. |L
I Exercise 8.20 (8.4.5). Let (. F . j) be such that j() < o, and let denote
a bounded F ,B
+
-measurable function on .
a. For every 1 6 6 o we have 1
]
, hence [ [
]
exists.
b. lim
]-o
[ [
]
= [ [
o
.
Proof. (a) Since is bounded, there exists 0 6 M 6 o such that [ [ 6 M.
Then ess sup 6 M; that is, [ [
o
exists. Now consider 1 6 < o. We have
_
[ [
]
6
_
M
]
= M
]
j() < o.
i.e., [ [
]
exists.
(b) If = 0 ja. e., then [ [
]
= [ [
o
= 0 for all , and the claim is trivial.
So assume that = 0 on a set of positive measure, so that [ [
o
> 0. We
rst show that liminf
]-o
[ [
]
> [ [
o
. Let t (0. [ [
o
). By the Chebyshevs
Inequality (Exercise 8.24) we have
[ [
]
> tj[(o)[ > t |
1{]
.
If j[(o)[ > t | = o, the claim is trivial. If j[(o)[ > t | < o, then
liminf
]-o
[ [
]
> liminf
]-o
tj[(o)[ > t |
1{]
= lim
]-o
tj[(o)[ > t |
1{]
= t.
Since t (0. [ [
o
) is arbitrary, we have lim
]-o
[ [
]
> [ [
o
.
We next show that limsup
]-o
[ [
]
6 [ [
o
. It follows from (a) that [ [
o
exists, and so [ [ 6 [ [
o
ja. e.. Then
SECTION 8.5 CONTAINMENT RELATIONS FOR 1
P
SPACES 125
[ [
]
]
=
_
[ [
]
6
_
[ [
]
o
= [ [
]
o
j():
that is, [ [
]
6 [ [
o
j()
1{]
. Then
limsup
]-o
[ [
]
6 [ [
o
.
Summarizing the ndings, we have lim
]-o
[ [
]
= [ [
o
. |L
8.5 Containment Relations for L
p
Spaces
I Exercise 8.21 (8.5.1). Consider the measure space (R. B. z). Let 1 6 < q <
o, and let r be such that 1,q < r < 1,.
a. Dene on R by writing (.) = .
-i
1
(0,1)
(.) for each . R. Then 1
]
but 1
q
. Therefore, we do not in general have 1
]
_ 1
q
when < q.
b. Let g(.) = .
-i
1
(1,o)
(.) for every . R. Then g 1
q
but g 1
]
. Therefore,
we do not in general have 1
q
_ 1
]
when < q.
Proof. (a) We have rq < 1 < r. Hence,
_
[ [
]
=
_
(0,1)
.
-i]
=
.
1-i]
1 r

1
0
=
1
1 r
< o.
_
[ [
q
=
_
(0,1)
.
-iq
=
.
1-iq
1 rq

1
0
= o:
that is, 1
]
, but 1
q
.
(b) We have
_
[g[
q
=
_
(1,o)
.
-iq
=
.
1-iq
1 rq

o
1
=
1
rq 1
< o.
_
[g[
]
=
_
(1,o)
.
-i]
=
.
1-i]
1 rq

o
1
= o:
hence, g 1
q
but g 1
]
. |L
Remark 8.22 (Folland 1999, p.185). Thus we see two reasons why a function
may fail to be in 1
]
: either [ [
]
blows up too rapidly near some point, or it
fails to decay suciently rapidly at innity. In the rst situation the behavior
of [ [
]
becomes worse as increases, while in the second it becomes better.
In other words, if < q, functions in 1
]
can be locally more singular than
functions in 1
q
, whereas functions in 1
q
can be globally more spread out than
functions in 1
]
. See Figure 8.1.
126 CHAPTER 8 THE 1
P
SPACES
0 .
Figure 8.1. (
p
and (
q
.
I Exercise 8.23 (8.5.2). Let 0 < < r < o. Then 1
]
1
o
_ 1
i
, and for any
1
]
1
o
we have [ [
i
6 [ [
]{i
]
[ [
1-]{i
o
.
Proof. Let :{o : [(o)[ 6 [ [
o
], so that j(
c
) = 0. Then
_
[ [
i
=
_

[ [
i
=
_

[ [
i-]
[ [
]
6 [ [
i-]
o
_

[ [
]
= [ [
i-]
o
[ [
]
]
< o.
Hence, 1
i
and [ [
i
6 [ [
]{i
]
[ [
1-]{i
o
. |L
I Exercise 8.24 (8.5.3). For any 0 < < o and 0 < M < o we have
__
[ [
]
_
1{]
> Mj({o : [(o)[ > M])
1{]
.
Proof. Let 1

:{o : [(o)[ > M]. Then


[ [
]
]
=
_
[ [
]
>
_
T
M
[ [
]
> M
]
_
T
M
1 = M
]
j(1

):
that is, j[(o)[ > M| 6 ([ [
]
,M)
]
. |L
I Exercise 8.25 (8.5.4). Let 0 < r < o and assume that 1
i
1
o
, so
that 1
]
for every r < < o by Exercise 8.23. We wish to show that
lim
]-o
[ [
]
= [ [
o
. Follow this outline:
a. Ignoring the trivial case where = 0 ja. e. on , let = 0 on a set of
positive measure, so that [ [
o
> 0. Show that liminf
]-o
[ [
]
> [ [
o
.
b. Show that limsup
]-o
[ [
]
6 [ [
o
.
c. Put (a) and (b) together to prove the claim.
Proof. (a) Pick an arbitrary t (0. [ [
o
). It follows from the Chebyshevs
Inequality (Exercise 8.24) that
SECTION 8.7 MORE CONVERGENCE CONCEPTS 127
[ [
]
> t j[(o)[ > t |
1{]
.
If j[(o)[ > t | = o, then [ [
]
= o for all , and the claim is trivial. If
j[(o)[ > t | < o, then lim
]-o
j[(o)[ > t |
1{]
= 1 and so liminf
]-o
[ [
]
>
t . Since t (0. [ [
o
) is chosen arbitrarily, we get
liminf
]-o
[ [
]
> [ [
o
(8.2)
(b) By Exercise 8.23 we have [ [
]
6 [ [
i{]
i
[ [
1-i{]
o
, for every (r. o).
Hence,
limsup
]-o
[ [
]
6 [ [
o
. (8.3)
(c) Combining (8.2) and (8.3) we get
[ [
o
6 liminf
]-o
[ [
]
6 limsup
]-o
[ [
]
6 [ [
o
.
Hence, lim
]-o
[ [
]
= [ [
o
. |L
I Exercise 8.26 (8.5.5). Let j() = 1 and 1 6 6 q 6 o. Show for arbitrary
that
_
[ [ 6
__
[ [
]
_
1{]
=
__
[ [
q
_
1{q
6 ess sup.
so that [ [
1
6 [ [
]
6 [ [
q
6 [ [
o
.
Proof. It follows from Claim 1 and the assumption that j() = 1. |L
8.6 Approximation
8.7 More Convergence Concepts
I Exercise 8.27 (8.7.1). Prove the following simple claims.
a. Let {
n
]
o
n=1
denote a Cauchy sequence in 1
]
, where 0 < < o. Show that
{
n
]
o
n=1
is a Cauchy sequence in measure: for every c > 0 and > 0 there is
N N such that for every n. m > N we have j({o : [
n
(o)
n
(o)[ >
]) < c.
b. Let .
1
.
2
. . . . 1
]
and suppose that
n
1
p
, where 0 < < o. If g 1
o
,
then g.
1
g.
2
g. . . . 1
]
and
n
g
1
p
g.
Proof. (a) Choose arbitrary c > 0 and > 0. It follows from Chebyshevs
Inequality that
j[
n
(o)
n
(o)[ > | 6
[
n

n
[
]
]

]
.
128 CHAPTER 8 THE 1
P
SPACES
Let c
t
= c
1{]
. Then there exists N Nsuch that [
n

n
[
]
< c
t
when n. m > N
since {
n
] is Cauchy in 1
]
. Hence, when n. m > N we get
j[
n
(o)
n
(o)[ > | <
c
]

]
= c.
(b) We have
[g[
]
]
=
_
[g[
]
=
_
[ [
]
[g[
]
6
_
[ [
]
[g[
]
o
= [g[
]
o
_
[ [
]
= [ [
]
]
[g[
]
o
< o:
that is, g 1
]
. Similarly we can show that
n
g 1
]
for all n N. Finally,
_
[
n
g g[
]
=
_
[
n
[
]
[g[
]
= [g[
]
_
[
n
[
]
0
since
n
1
p
. |L
I Exercise 8.28 (8.7.2). While convergence in th mean implies convergence
in measure, it is not the case that convergence in measure implies convergence
in the mean.
Proof. Consider the probability space (0. 1|. B
0,1j
. z), where z is Lebesgue
measure and set

n
= 2
n
1
(0,1{n)
.
Then
lim
n
z([
n
0[ > c) = lim
n
z(0. 1,n) = 0.
However,
_
[
n
[
]
= 2
n]
,n o.
0 .

4
Figure 8.2. (
n

--0, but (
n
,
L
p
---0.
SECTION 8.8 PRELUDE TO THE RIESZ REPRESENTATION THEOREM 129
Thus, convergence in measure does not imply 1
]
convergence. What can go
wrong is that the nth function in the sequence can be huge on a very small set
(see Figure 8.2). |L
I Exercise 8.29 (8.7.3). It is possible for a sequence {
n
]
o
n=1
in 1
]
to converge
ja. e. to some 1
]
but not in th mean. That is, convergence ja. e. does
not force convergence in th mean.
Proof. Consider the setting in the previous exercise again. |L
I Exercise 8.30 (8.7.4). It is possible for a sequence {
n
]
o
n=1
in 1
]
to converge
in th mean to zero, but {
n
]
o
n=1
converges at no point of .
Proof. Consider (0. 1|. B
0,1j
. z). Set

1
= 1
0,1{2j
.
2
= 1
1{2,1j
.

3
= 1
0,1{3j
.
4
= 1
1{3,2{3j
.
5
= 1
2{3,1j
.

6
= 1
0,1{4j
.
For every > 0,
_
[
1
[
]
=
_
[
2
[
]
=
1
2
.
_
[
3
[
]
=
_
[
4
[
]
=
_
[
5
[
]
=
1
3
.
_
[
6
[
]
=
1
4
.
So
_
[
n
[
]
0 and
n
1
p
0. However, {
n
] converges at no point. |L
I Exercise 8.31 (8.7.5). It is possible to have functions .
1
.
2
. . . . 1
]
1
1
]
2
such that
n
1
p
1
but
n
,
1
p
2
.
Proof. Consider ((0. o). B
(0,o)
. z). Set

n
= n
-1
1
(n,2n)
:
see Figure 8.3. Then
_
[
n
[
]
=
n
n
]
= n
1-]
.
The sequence {n
1-]
] converges if > 1, and diverges if 6 1. Thus,
n
1
p
0
when 1 < < o, but [
n
[
1
fails to converge to 0. |L
8.8 Prelude to the Riesz Representation Theorem
130 CHAPTER 8 THE 1
P
SPACES
0 .
0.2
0.4
0.6
0.8
1
1

5
Figure 8.3.
9
THE RADON-NIKODYM THEOREM
9.1 The Radon-Nikodym Theorem, Part I
I Exercise 9.1 (9.1.1). In the denition of an additive set function, show that
the series

o
n=1
(
n
) must converge absolutely.
Proof. Observe that
_
o
n=1

n
=
_
o
k=1

n
k
for every rearrangement {n
k
]
o
k=1
of the positive integers, hence both

o
n=1
(
n
) and

o
k=1
(
n
k
) should be
dened and equal, that is, the series is unconditionally convergent. By the
Riemann series theorem, it is absolutely convergent. |L
I Exercise 9.2 (9.1.2). In Claim 4, quickly verify that
-
is a nite measure
with support
-
.
Proof. For all F we have
-
_
-
, and the negativity of
-
with
respect to implies that
-
() = (
-
) > 0. Therefore,
-
is nonnega-
tive. Next,
-
() = () = 0. We now exhibit countable additivity for
-
. Let
{
n
]
o
n=1
denote a disjoint sequence of F -sets. Then

-
_
_
o
_
n=1

n
_
_
=
_
_
_
_
_
o
_
n=1

n
_
_

-
_
_
_ =
_
_
o
_
n=1
(
n

-
)
_
_
=
o

n=1
(
n

-
)
=
o

n=1
_
(
n

-
)
_
=
o

n=1

-
(
n
).
This shows that
-
is a measure. Since
-
() = (
-
) R,
-
is a -
nite measure. To see that
-
is a support of
-
, observe that
-
_
(
-
)
c
_
=

_
(
-
)
c

-
_
= () = 0. |L
131
132 CHAPTER 9 THE RADON-NIKODYM THEOREM
I Exercise 9.3 (9.1.3). Suppose that
_

.
-
_
and
_
T

. T
-
_
are Hahn de-
compositions with respect to an additive set function . Then
_

^T

_
=
(
-
^T
-
) = 0.
Proof. We rst do the set operations:

^T

= (

L T

) (

) = (

L T

)
_

-c
T
-c
_
= (

L T

) (
-
L T
-
)
c
= (

L T

) (
-
L T
-
)
=
_

T
-
_
L
_

-
T

_
.
and
_

T
-
_

-
T

_
= . Since

T
-
_

, we have
_

T
-
_
>
0; since

T
-
_

_ T
-
, we have
_

T
-
_

_
6 0; hence,

T
-
_

_
= 0. Similarly,
_

-
T

_
= 0, and so
_

^T

_
= 0.
Using this way, we can also show that (
-
^T
-
) = 0. |L
I Exercise 9.4 (9.1.4). The Jordan decomposition of an additive set function
is unique.
Proof. Let
_

.
-
_
and
_
T

. T
-
_
denote Hahn decomposition of with re-
spect to . Let

(1) =
_
1

_
and

B
(1) =
_
1 T

_
for every 1 F ;
dene
-

and
-
B
similarly. Then =

is the Jordan decomposition of


relative to the Hahn decomposition
_

.
-
_
and =

B

-
B
is the Jordan de-
composition of relative to the Hahn decomposition
_
T

. T
-
_
. We now show
that

B
and
-

=
-
B
. For any 1 F , we have

(1) =
_
1

_
=
_
1
_
T

L T
-
_

_
=
_
1

_
1

T
-
_
=
_
1

_
.
where
_
1

T
-
_
= 0 since: (i) 1

T
-
_

implies that (1

T
-
) > 0; (ii) 1

T
-
_ T
-
implies that
_
1

T
-
_
6 0. Similarly, we
can show

B
=
_
1

_
=

and
-

=
-
B
. |L
I Exercise 9.5 (9.1.5). This problem relates somewhat the notion of absolute
continuity with the familiar c- concepts.
a. Let j and v denote measures with common domain F and such that v is
nite. Then v j i for every c > 0 there is > 0 such that j() < forces
v() < c.
b. The claim in (a) is not necessarily true if v is innite, since the condition v j
does not imply the c- condition.
SECTION 9.1 THE RADON-NIKODYM THEOREM, PART I 133
Proof. (a) Suppose rst that for every c > 0 there is > 0 such that j() <
forces v() < c. We desire to show that v j. If j() = 0 and c > 0 is given
(and the corresponding is found), then j() < , hence v() < c. Since c > 0
is arbitrary, it follows that v() = 0, whence v j.
To show the other direction, suppose that there is c > 0 such that for every
> 0 there is a set F with j() < and v() > c. In particular, there is
c > 0 such that there is a sequence {
n
]
o
n=1
of F -sets with j() < 1,n
2
and
v() > c for each n N. Let = lim
n
=
_
o
n=1
_
o
n=n

n
. For every n N we
have
j() 6 j
_
_
o
_
n=n

n
_
_
6
o

n=n
j(
n
) <
o

n=n
1
m
2
.
so j() = 0. However, we also have
v() = v
_
lim
n
_
> limv(
n
) > c > 0
by property (M10) in Section 2.2. This shows that there is F such that
j() = 0 and v() > 0, so v ,j.
(b) Let = Z, let F = 2

, let v denote the counting measure, so that v


is innite, and let j be such that j
_
{n]
_
=
1
n
2
for each n Z, so that j is
nite. |L
I Exercise 9.6 (9.1.6). Let j, v, v
1
, and v
2
denote measures, each having com-
mon domain F .
a. If v
1
J j and v
2
J j, then v
1
v
2
J j.
b. If v
1
j and v
2
j, then v
1
v
2
j.
c. If v
1
j and v
2
J j, then v
1
J v
2
.
d. If v j and v J j, then v = 0.
e. If j J j, then j = 0.
f. If j and v are o-nite with v j, then v({o :
d
d
(o) = 0]) = 0.
Proof. (a) Let v
1
J j and v
2
J j. Then there exist D
1
F with v
1
(D
1
) =
j
_
D
c
1
_
= 0, and D
2
F with v
2
(D
2
) = j
_
D
c
2
_
= 0. Let D = D
1
D
2
. We show
that D supports j and D
c
supports v
1
v
2
. As for j, we have
j
_
D
c
_
= j
_
D
c
1
L D
c
2
_
6 j
_
D
c
1
_
j
_
D
c
2
_
= 0.
As for v
1
v
2
, we have
(v
1
v
2
) (D) = v
1
(D
1
D
2
) v
2
(D
1
D
2
) 6 j
1
(D
1
) v
2
(D
2
) = 0.
Therefore, (v
1
v
2
) (D) = j(D
c
) = 0, that is, v
1
v
2
J j.
(b) If j() = 0, then (v
1
v
2
)() = v
1
() v
2
() = 0; hence, v
1
v
2
j.
134 CHAPTER 9 THE RADON-NIKODYM THEOREM
(c) Since v
2
J j, there exists D F with v
2
(D) = j(D
c
) = 0; since v
1

j, j(D
c
) = 0 forces v
1
(D
c
) = 0. Therefore, there is D F with v
1
(D
c
) =
v
2
(D) = 0, that is, v
1
J v
2
.
(d) Since v J j, there is D F with v (D) = j(D
c
) = 0. For any 1 F , we
have
v(1) = v (1 D) v
_
1 D
c
_
= 0 0 = 0.
where v (1 D) = 0 since v (1 D) 6 v (D) = 0, and v (1 D
c
) 6 v (D
c
) = 0
since j(D
c
) = 0 and v j.
(e) Let v = j in (d) and we get the result.
(f) We have
vdv,dj = 0| =
_
d{d=0j
dv
dj
dj = 0. |L
I Exercise 9.7 (9.1.8). Let 1
1
(. F . j). Dene v(1) =
_
T
dj for every
1 F .
a. v is an additive set function such that v

(1) =
_
T


dj and v
-
(1) =
_
T

-
dj for every 1 F .
b. If

= {o : (o) > 0] and


-
=
c
, then (

.
-
) is a Hahn decom-
position with respect to v.
Proof. (b) We rst show (b). For every 1 F with 1 _

, we have
v(1) =
_
T
dj =
_
T
C
dj =
_
T
1

C dj =
_
T


dj > 0.
and for every 1 F with 1 _
-
we have
v(1) =
_
T
dj =
_
T

dj =
_
T
1

dj =
_
T

-
dj 6 0.
Hence, (

.
-
) is a Hahn decomposition with respect to v.
(a) It follows from (L6) (p. 251) that v is an additive set function. Now by part
(b) and the uniqueness of Hahn decomposition (Exercise 9.3), we get the de-
sired result. |L
I Exercise 9.8 (9.1.9). Let M denote the collection of additive set functions
with domain F .
a. Mis a linear space over R: for a. b R and
1
.
2
Mwe have a
1
b
2
M.
b. Given M, dene [[ =

()
-
(), where =

-
is the Jordan
decomposition of . Then [ [ is a norm on M.
c. Is Ma Banach space?
SECTION 9.1 THE RADON-NIKODYM THEOREM, PART I 135
Proof. (a) It is clear that a
1
b
2
: F R, and for a disjoint sequence
{
n
] _ F we have
(a
1
b
2
)
_
_
o
_
n=1

n
_
_
= a
1
_
_
o
_
n=1

n
_
_
b
2
_
_
o
_
n=1

n
_
_
=
o

n=1
a
1
(
n
)
o

n=1
b
2
(
n
)
=
o

n=1
_
a(
n
) b
2
(
n
)
_
=
o

n=1
(a
1
b
2
)(
n
).
(b) Clearly, [[ > 0 for all M, and [o[ = 0, where o(1) = 0 for all 1 F .
Now if [[ = 0, then

()
-
() = 0 implies that

() =
-
() = 0.
Since

and
-
are nite measures on F (by Claim 4, p. 373), for every 1 F
we have

(1) 6

() = 0 and
-
(1) 6
-
() = 0; that is,
(1) =

(1)
-
(1) = 0.
We nally show that the triangle inequality. Let
1
.
2
M. Then
[
1

2
[ = (
1

2
)

() (
1

2
)
-
()
6

1
()

2
()
-
1
()
-
()
= [
1
[ [
2
[.
This proves that (M. [ [) is a normed space. |L
I Exercise 9.9 (9.1.10). Let (

.
-
) denote a Hahn decomposition of the ad-
ditive set function , and let =


-
denote the Jordan decomposition. We
have

() = sup{(1) : 1 F . 1 _ ].

-
() = inf{(1) : 1 F . 1 _ ].
for every F .
Proof. Let . 1 F with 1 _ . Then
(1) = (1

) (1
-
) =

(1)
-
(1) 6

(1) 6

().
Thus,

() is an upper bound of {(1) : 1 F . 1 _ ]. We next show that

() is actually in the former set: let 1 =

. Then 1 F , 1 _ , and
(1) = (

) =

().
We then have
136 CHAPTER 9 THE RADON-NIKODYM THEOREM

-
() =

() () = sup{(1) : 1 F . 1 _ ] ()
= sup{(1) () : 1 F . 1 _ ]
= sup{( 1) : 1 F . 1 _ ]
= sup{(J) : J F . J _ ]
= inf{(J) : J F . J _ ]. |L
I Exercise 9.10 (9.1.12). Let j, v, v
1
, v
2
and j denote o-nite measures having
domain F . We have the following claims.
a. If v
1
j and v
2
j, then d(v
1
v
2
) ,dj = dv
1
,djdj
2
,dj ja. s. on .
b. If v j and j j, then v j and
d
dp
=
d
d
d
dp
ja. e. on .
c. If v j and j v, then
d
d
= 1
d{d>0|

1
d{d
ja. e. on .
d. Let j j and v j. Then v j if and only if j
d
dp
> 0.
d
dp
> 0| = 0, in
which case we have
dv
dj
= 1
d{dp>0|

dv,dj
dj,dj
ja. s. on .
Proof. (a) Since v
1
j and v
2
j, we get v
1
v
2
j, and that for every
1 F ,
v
1
(1) =
_
T
dv
1
dj
dj. v
2
(1) =
_
T
dv
2
dj
dj.
and
(v
1
v
2
)(1) =
_
T
d(v
1
v
2
)
dj
dj.
Clearly, (v
1
v
2
)(1) = v
1
(1) v
2
(1). Hence,
d(v
1
v
2
)
dj
=
dv
1
dj

dv
2
dj
.
(b) Let v j and j j. Take an arbitrary 1 F so that j(1) = 0; then
j(1) = 0; then v(1) = 0 and so v j. Next, for every 1 F , we have
v(1) =
_
T
dv
dj
dj. j(1) =
_
T
dj
dj
dj. and v(1) =
_
T
dv
dj
dj.
It follows from (L14) (p. 259) that
v(1) =
_
T
dv
dj
dj =
_
T
dv
dj
dj
dj
dj:
that is,
d
dp
=
d
d
d
dp
.
(c) It follows from (b) that
dv
dv
=
dv
dj
dj
dv
.
Therefore,
SECTION 9.1 THE RADON-NIKODYM THEOREM, PART I 137
dv
dj
= 1
d{d>0|

1
dj,dv
.
(d) |L
IExercise 9.11 (9.1.15). Suppose that j and v are o-nite measures on F . The
Lebesgue decomposition of v with respect to j is unique. That is, if v = v
ac
v
s
where v
ac
and v
s
are o-nite measures with v
ac
j and v
s
J j, and if in
addition v = v
t
ac
v
t
s
where v
t
ac
and v
t
s
are o-nite measures with v
t
ac
j and
v
t
s
J j, then v
ac
= v
t
ac
and v
s
= v
t
s
.
Proof. Since v
s
J j, there exists F such that supports v
s
and
c
supports j; that is,
v
s
(
c
) = j() = 0.
Since v
t
s
J j, there exists T F such that T supports v
t
s
and T
c
supports j;
that is,
v
t
s
(T
c
) = j(T) = 0.
Since v
s
(
c
T
c
) 6 v
s
(
c
) = 0, and v
t
s
(
c
T
c
) 6 v
t
s
(T
c
) = 0, we have that
LT supports both v
s
and v
t
s
. Since j(LT) 6 j() j(T) = 0, we have that
( L T)
c
supports j. Let S : L T, so that
j(S) = v
s
(S
c
) = v
t
s
(S
c
) = 0
We now show that v
ac
= v
t
ac
. For every 1 F , we have
v
ac
(1) = v
ac
(1 S
c
) v
ac
(1 S) = v
ac
(1 S
c
) v
ac
j|
= v
ac
(1 S
c
) v
s
(1 S
c
)
= v(1 S
c
)
= v
t
ac
(1 S
c
) v
t
s
(1 S
c
)
= v
t
ac
(1 S
c
)
= v
t
ac
(1 S
c
) v
t
ac
(1 S)
= v
t
ac
(1).
Hence, v
ac
= v
t
ac
, and so v
s
= v
t
s
. |L
10
PRODUCTS OF TWO MEASURE SPACES
10.1 Product Measures
Remark 10.1.
( T) (C D) = ( C) (T D) .
( L T) (C L D) = ( C) L (T D) .
(T C) = ( T) ( C) .
(T L C) = ( T) L ( C) .
I Exercise 10.2 (10.1.1). Let
1
denote an uncountable set, and let F
1
denote
the o-eld of subsets of
1
that are at most countable or have at most countable
complements. Let
2
and F
2
be identical to
1
and F
1
, respectively. Let D =
{(o
1
. o
2
)
1

2
: o
1
= o
2
]. We have D
o
1
F
2
and D
o
2
F
1
for every
o
1

1
and o
2

2
, but D F
1
F
2
.
Proof. For every o
1
, we have D
o
1
= {o
2
] with o
2
= o
1
. For every o
2

2
, we
have D
o
2
= {o
1
] with o
1
= o
2
. Hence, D
o
1
F
2
and D
o
2
F
1
. |L
I Exercise 10.3 (10.1.2). Let _
1
and T _
2
.
a. Suppose that T = . Then T F
1
F
2
i F
1
and T F
2
.
b. Suppose that T = . Then obviously T F
1
F
2
, but it is not always
the case that F
1
and T F
2
.
Proof. (a) The if part is evident since F
1
F
2
_ F
1
F
2
. Now take an arbitrary
o
1
(such a point exists because T = implies that = ). Then
( T)
o
1
= T F
2
. Similarly we show that F
1
.
(b) If there exists F
1
, then we have = F
1
F
2
. But obviously
F
1
. |L
I Exercise 10.4 (10.1.3). Prove the following set-theoretical facts.
139
140 CHAPTER 10 PRODUCTS OF TWO MEASURE SPACES
a. Let
1
T
1
and
2
T
2
both be nonempty. Then
1
T
1
_
2
T
2
i
1
_
2
and T
1
_ T
2
.
b. Let
1
T
1
and
2
T
2
both be nonempty. Then
1
T
1
=
2
T
2
i
1
=
2
and T
1
= T
2
.
c. Let T,
1
T
1
, and
2
T
2
be nonempty. Then T is the disjoint union
of
1
T
1
and
2
T
2
i either (i) is the disjoint union of
1
and
2
and
T = T
1
= T
2
or (ii) =
1
=
2
and T is the disjoint union of T
1
and T
2
.
d. The only if parts of (a) and (b) do not necessarily hold for empty Cartesian
products. What about (c)?
Proof. (a) The if part is automatic, so we only do the only if part. Suppose
that
1
T
1
_
2
T
2
. If, say,
1
,_
2
, then there exists o
t
1

1

2
.
Take an arbitrary o
2
T
1
. Then (o
t
1
. o
2
)
1
T
1
but (o
t
1
. o
2
)
2
T
2
. A
contradiction.
(b) Using the fact that
1
T
1
=
2
T
2
i
1
T
1
_
2
T
2
and
2
T
2
_

1
T
1
, and the result in (a), we get the desired outcome.
(c) Straightforward. |L
I Exercise 10.5 (10.1.5). Let F
1
and F
2
denote o-elds on
1
and
2
, respec-
tively. It may not be the case that F
1
F
2
is a o-eld on
1

2
.
Proof. Consider (R
k
. B
k
. z
k
) and (R
n
. B
n
. z
n
). Then B
k
B
n
B
k
B
n
. |L
I Exercise 10.6 (10.1.6). Prove Claims 2(b) and 3(b) by mimicking the proofs
of Claims 2(a) and 3(a).
Proof. (2(b)) We show that if 1 F
1
F
2
, then 1
o
2
F
1
for every o
2

2
.
Dene
D = {1 F
1
F
2
: 1
o
2
F
1
for every o
2

2
].
First observe that
1

2
F
1
F
2
_ F
1
F
2
and (
1

2
)
o
2
=
1
F
1
for every o
2

2
. Therefore,
1

2
D. Next, if 1 D, then we have
1
c
F
1
F
2
and (1
c
)
o
2
= (1
o
2
)
c
F
1
for every o
2

2
, so that 1
c
D.
Next, if {1
n
] is a sequence of D-sets, then
_
1
n
F
1
F
2
and (
_
1
n
)
o
2
=
_
(1
n
)
o
2
F
1
for every o
2

2
, whence
_
1
n
D. Therefore, D is a o-eld
on
1

2
and D _ F
1
F
2
. We desire to strengthen this inclusion to an
equality. To do this, let 1
1
F
1
and 1
2
F
2
. Then for every o
2

2
we have
(1
1
1
2
)
o
2
=

1
1
if o
2
1
2
if o
2
1
2
F
1
.
This shows that F
1
F
2
_ D. Since D is a o-led, we have F
1
F
2
_ D. This
yields D = F
1
F
2
.
SECTION 10.1 PRODUCT MEASURES 141
(3(b)) We show that if :
1

2


R be F
1
F
2
,B
+
-measurable, then
o
2
is F
1
,B
+
-measurable for every o
2

2
. To do this, rst consider the case of
= 1
T
, where 1 F
1
F
2
. Next, pick o
1

1
and o
2

2
. We have
(1
T
)
o
2
(o
1
) = 1 == 1
T
(o
1
. o
2
) = 1 == (o
1
. o
2
) 1 == o
1
1
o
2
== 1
T
!
2 (o
1
) = 1.
and hence
(1
T
)
o
2
= 1
T
!
2 .
By Claim 2(b), 1 F
1
F
2
forces 1
o
2
F
1
. Therefore, if = 1
T
where
1 F
1
F
2
, we have that
o
2
is the indicator function 1
T
!
2 of the F
1
-set
1
o
2
, and hence is F
1
,B
+
-measurable. This proves 3(b) when is an indicator
function on
1

2
of a set in F
1
F
2
.
Next, let =

n
i=1
c
i
1
T
i
, where 1
1
. . . . . 1
n
F
1
F
2
are disjoint with union

1

2
and c
1
. . . . . c
n
R, so that is an F
1
F
2
,B
+
-measurable simple
function on
1

2
. For every o
2

2
, observe that
o
2
=

n
i=1
c
i
1
(T
i
)
!
2 ,
a nite linear combination of the indicator functions (1
T
1
)
o
2
. . . . . (1
T
n
)
o
2
, and
each of these is F
1
,B
+
-measurable by the previous paragraph. It follows that

o
2
is F
1
,B
+
-measurable for every o
2

2
. This proves the result when is
an F
1
F
2
,B
+
-measurable simple function on
1

2
.
Next, suppose that is a nonnegative F
1
F
2
,B
+
-measurable function on

1

2
. There exists a nondecreasing sequence {s
n
] of nonnegative nite-
valued F
1
F
2
,B
+
-measurable simple functions on
1

2
with lim
n
s
n
= .
By the previous paragraph, we have that (s
n
)
o
2
is F
1
,B
+
-measurable for every
o
2

2
and n N. Since

o
2
=
_
lim
n
s
n
_
o
2
= lim
n
(s
n
)
o
2
for every o
2

2
, we have that
o
2
is the limit of a sequence of F
1
,B
+
-
measurable functions and hence is itself F
1
,B
+
-measurable. This proves the
result when is a nonnegative F
1
F
2
,B
+
-measurable function on
1

2
.
Finally, if is a general F
1
F
2
,B
+
-measurable function on
1

2
, then
the functions

and
-
, both being nonnegative F
1
F
2
,B
+
-measurable
functions on
1

2
, are such that (

)
o
2
and (
-
)
o
2
are F
1
,B
+
-measurable
for every o
2

2
. Observing that

o
2
= (


-
)
o
2
= (

)
o
2
(
-
)
o
2
for every o
2

2
, we see for every o
2

2
that
o
2
is the dierence of two
F
1
,B
+
-measurable functions on
1

2
and hence is F
1
,B
+
-measurable. This
completes the proof. |L
I Exercise 10.7 (10.1.7). The product of (R
k
. B
k
. z
k
) and (R
n
. B
n
. z
n
) is
(R
kn
. B
kn
. z
kn
).
142 CHAPTER 10 PRODUCTS OF TWO MEASURE SPACES
In other words, B
k
B
n
= B
kn
and z
k
z
n
= z
kn
.
Proof. We rst show that B
k
B
n
= B
kn
visa showing that B
k
B
n

B
kn
(proper subset). Consider the projection
k
: R
kn
R
k
. Let O
k
and
O
kn
be the set of open sets of R
k
and R
kn
, respectively. Endowed with
Tychono topology,
k
is continuous. Hence,

-1
k
(B
k
) =
-1
k
(o(O
k
)) = o(
-1
k
(O
k
)) _ o(O
kn
) = B
kn
.
Similarly, we have
-1
n
(B
n
) _ B
kn
. Therefore,
B
k
B
n
=
-1
k
(B
k
)
-1
n
(B
n
) _ B
kn
.
To see that the containment is strict, observe that the open unit ball D in B
kn
cannot be written as
1

2
with
1
_ R
k
and
2
_ R
n
, let along with
1
B
k
and
2
B
n
. From the above argument, we have
B
k
B
n
= o(B
k
B
n
) _ B
kn
.
Dene A
1
= intervals of the form (o. .|. We have A
kn
1
= A
k
1
A
n
1
_ B
k

B
n
; hence,
B
kn
= o(A
kn
1
) _ o(B
k
B
n
) = B
k
B
n
.
It follows from Claim 4 of Section 4.2 that z
kn
( T) = z
k
()z
n
(T) for
every B
k
and T B
n
. Since (R
k
. B
k
. z
k
) and (R
n
. B
n
. z
n
) are o-nite, by
Claim 6 we have z
kn
= z
k
z
n
. |L
I Exercise 10.8 (10.1.8). Let
1
=
2
= 0. 1|. Let F
1
= F
2
denote the Borel
subsets of 0. 1|. Let j
1
denote Lebesgue measure restricted to F
1
, and let j
2
denote the counting measure on 0. 1|. Let 1 = {(o
1
. o
2
)
1

2
: o
1
= o
2
].
a. 1 F
1
F
2
.
b.
_

1
j
2
(1
o
1
) dj
1
(o
1
) = 1.
c.
_

2
j
1
(1
o
2
) dj
2
(o
2
) = 0.
Proof. (a) We prove 1 F
1
F
2
by showing that 1 is closed in 0. 1| 0. 1|. It
is true because 0. 1| is Hausdor (see Willard, 2004, Theorem 13.7).
(b) Since j
2
is a counting measure, we have
_

1
j
2
(1
o
1
) dj
1
(o
1
) =
_

1
j
2
(o
2
) dj
1
(o
1
) =
_

1
1 dj
1
(o
1
) = 1.
(c) We have
_

2
j
1
(1
o
2
) dj
2
(o
2
) =
_

2
j
1
(o
1
) dj
2
(o
2
) =
_

2
0 dj
2
(o
2
) = 0. |L
I Exercise 10.9 (10.1.10, Cavalieris Principle). If 1. J F
1
F
2
are such that
j
2
(1
o
1
) = j
2
(J
o
1
) for every o
1

1
, then j
1
j
2
(1) = j
1
j
2
(J).
SECTION 10.2 THE FUBINI THEOREMS 143
Proof. We have
j
1
j
2
(1) =
_

1
j
2
(1
o
1
) dj
1
=
_

1
j
2
(J
o
1
) dj
1
= j
1
j
2
(J). |L
10.2 The Fubini Theorems
11
ARBITRARY PRODUCTS OF MEASURE SPACES
11.1 Notation and Conventions
I Exercise 11.1 (11.1.1). Let
1
denote a nonempty set, and let A denote a
nonempty collection of subsets of
1
. Let
2
denote a nonempty set, and let B
denote a nonempty collection of subsets of
2
.
a. Let T _
2
be nonempty. Then o

1
B
(A {T]) = o(A) {T].
b. o(A B) = o(o(A) o(B)).
Proof. (a) Since A {T] _ o(A) {T], and o(A) {T] is a o-eld on
1
T,
we get
o

1
B
(A {T]) _ o(A) {T].
To see the converse inclusion, dene
C :

o(A) : T o
B
(A {T])
_
.
If A, then T A {T] _ o

1
B
(A {T]), so C; thus A _ C. We
then show that C is a o-eld. (i)
1
C. (ii) If C, then ( T)
c
=
c
T
o

1
B
(A {T]), i.e.,
c
C. (iii) If {
n
]
o
n=1
_ C, then (
_

n
) T =
_
(
n
T),
i.e.,
_

n
C. Therefore, o(A) = C.
(b) Since A B _ o(A) o(B), we have
o(A B) _ o(o(A) o(B)).
Next, for every T o(B) we have o(A) {T] = o

1
B
(A {T]) _ o(A B) by
(a). Therefore,
_
Bc(B)
_
o(A) {T]
_
_ o(A B):
that is, o(A) o(B) _ o(A B). But then o(o(A) o(B)) _ o(A B). |L
I Exercise 11.2 (11.1.2). Prove the claim in the Identication Lemma for the
case where 1 is a two-element set, which case is really the only one that we use.
145
146 CHAPTER 11 ARBITRARY PRODUCTS OF MEASURE SPACES
Proof. By the assumption, 1 = {1. 2]. Then
N
0
= F
T
1
F
T
2
.
Let A
1
denote the collection of sets of the form
iT
1

i
, where
i
F
i
for
each i D
1
and at most nitely many
i
s dier from
i
. Then F
T
1
= o(A
1
).
Let A
2
denote the collection of sets of the form
iT
2

i
, where
i
F
i
for
each i D
2
and at most nitely many
i
s dier from
i
. Then F
T
2
= o(A
2
),
and
N
1
= A
1
A
2
.
Therefore, o(N
0
) = o(N
1
) i
o(o(A
1
) o(A
2
)) = o(A
1
A
2
).
The above equality follows from Exercise 11.1(b) immediately. |L
I Exercise 11.3 (11.1.3). Prove the Identication Lemma in full generality for
the case where 1 is an arbitrary set.
Proof. By denition, N
0
is the collection of sets
]1

T
p
, where
T
p
F
T
p
for each 1 and
T
p
=
T
p
for at most nitely many 1. Further, N
1
is
the collection of sets of the form

]1
_

iT
p

i
_
.
where
i
=
i
for at most nitely many i
_
]1
D
]
. For each 1, let A
T
p
denote the collection of
iT
p

i
. We then have
o(N
0
) = o
_

]1
F
T
p
_
.
o(N
1
) = o
_

]1
A
T
p
_
.
Notice that F
T
p
= o(A
T
p
) for every 1. Thus we need to show that
o
_

]1
o(A
T
p
)
_
= o
_

]1
A
T
p
_
.
Generalizing the result in Exercise 11.1(b) yields the desired outcome. |L
I Exercise 11.4 (11.1.4). Show that E
T
is a semiring on
T
.
Proof. Given a nite subset J _ 1, we dene E
T
by writing
E
T
=
_

iT

i
:
i
F
i
for every i J
_
.
SECTION 11.2 CONSTRUCTION OF THE PRODUCT MEASURE 147
Clearly, E
T
. Take two sets T. C E
T
and write them as T =
iT
T
i
and
C =
iT
C
i
, where T
i
. C
i
F
i
for each i J. Then
T C =
_

iT
T
i
_

iT
C
i
_
=

iT
(T
i
C
i
) E
T
.
Finally, suppose that = T _ C (otherwise the proof is trivial). Then T
i
_ C
i
for every i J. It is easy to see that C T is a nite disjoint union of E
T
-
sets. |L
I Exercise 11.5 (11.1.5). Let A denote a semiring on
1
, and let B denote a
semiring on
2
. Then A B is a semiring on
1

2
.
Proof. It is evident that AB contains and is a -system. Now let
1
T
1
_

2
T
2
, where
1
.
2
A and T
1
. T
2
B. Then
1
_
2
and T
1
_ T
2
, and
(
2
T
2
) (
1
T
1
) =
1
(T
2
T
1
)| L (
2

1
) T
2
|.
Observe that T
2
T
1
may be written as a nite disjoint union

k
i=1
D
i
of B-
sets, and
2

1
may be written as a nite disjoint union

I
}=1
C
}
. It follows
that
(
2
T
2
) (
1
T
1
) =
_
_
_
1

_
_
k
_
i=1
D
i
_
_
_

_ L
_
_
_
_
_
I
_
}=1
C
}
_
_
T
2
_

_
=
_
_
k
_
i=1
(
1
D
i
)
_
_
L
_
_
I
_
}=1
(C
}
T
2
)
_
_
.
Hence (
2
T
2
) (
1
T
1
) is a nite disjoint union of sets in A B. |L
11.2 Construction of the Product Measure
I Exercise 11.6 (11.2.1). Refer to the proof of Claim 4.
a. Prove Subclaim 1.
b. Prove Subclaim 2.
c. Why cant we use the same type of proof as used to demonstrate the nite
additivity of j to show that j as dened on H is countably additive?
Proof. (a) We show that if
T
1
F
T
1
, then there exists C
T
1
LT
2
F
T
1
LT
2
with
the property that

-1
T
1
(
T
1

T
c
1
) =
-1
T
1
LT
2
(C
T
1
LT
2

(T
1
LT
2
)
c ). (11.1)
Dene the collection
148 CHAPTER 11 ARBITRARY PRODUCTS OF MEASURE SPACES
C :

T
1
F
T
1
: there is C
T
1
LT
2
F
T
1
LT
2
such that (11.1) holds
_
.
Then C _ F
T
1
.
We rst show that E
T
1
_ C. Take an arbitrary
iT
1

i
E
T
1
. Then

-1
T
1
_
_
_

iT
1

i
_

T
c
1
_
_
=

iJ
Q
i
.
where Q
i
=
i
for each i J
1
and Q
i
=
i
for each i J
c
1
. Dene the set
C
T
1
LT
2
=

iT
1
LT
2
1
i
.
where 1
i
=
i
for each i J
1
and 1
i
=
i
for each i J
2
. We have C
T
1
LT
2

E
T
1
LT
2
_ F
T
1
LT
2
, and

-1
T
1
LT
2
_
C
T
1
LT
2

(T
1
LT
2
)
c
_
=

iJ
Q
i
.
Comparing the last two displayed equations shows that
iT
1

i
C. That is,
we have E
T
1
_ C.
We now turn to showing that C is a o-led on
T
1
. We rst show that
T
1

C. This is because

-1
T
1
(
T
1

T
c
1
) = =
-1
T
1
LT
2
(
T
1
LT
2

(T
1
LT
2
)
c ).
and
T
1
LT
2
F
T
1
LT
2
. We now discuss closure under complementation. Sup-
pose that
T
1
C, and let C
T
1
LT
2
F
T
1
LT
2
be such that (11.1) holds. Then

-1
T
1
(
c
T
1

T
c
1
) =
-1
T
1
_
(
T
1

T
c
1
) (
T
1

T
c
1
)
_
=
-1
T
1
(
T
1

T
c
1
)
-1
T
1
(
T
1

T
c
1
)
=
-1
T
1
LT
2
_

T
1
LT
2

(T
1
LT
2
)
c
_

-1
T
1
LT
2
_
C
T
1
LT
2

(T
1
LT
2
)
c
_
=
-1
T
1
LT
2
_
_

T
1
LT
2

(T
1
LT
2
)
c
_

_
C
T
1
LT
2

(T
1
LT
2
)
c
_
_
=
-1
T
1
LT
2
_
C
c
T
1
LT
2

(T
1
LT
2
)
c
_
.
Since C
c
T
1
LT
2
F
T
1
LT
2
, it follows that
c
T
1
C.
We now show that C is closed under countable intersections. Let {
(n)
T
1
] _ C,
and let C
(n)
T
1
LT
2
F
T
1
LT
2
denote the corresponding sets for
(n)
T
1
that satises
(11.1) for every n N:

-1
T
1
_

(n)
T
1

T
c
1
_
=
-1
T
1
LT
2
_
C
(n)
T
1
LT
2

(T
1
LT
2
)
c
_
.
We have
SECTION 11.2 CONSTRUCTION OF THE PRODUCT MEASURE 149

-1
T
1
_
_
_
_
_
o
_
n=1

(n)
T
1
_
_

T
c
1
_

_ =
-1
T
1
_
_
o
_
n=1
_

(n)
T
1

T
c
1
_
_
_
=
o
_
n=1

-1
T
1
_

(n)
T
1

T
c
1
_
=
o
_
n=1

-1
T
1
LT
2
_
C
(n)
T
1
LT
2

(T
1
LT
2
)
c
_
=
-1
T
1
LT
2
_
_
o
_
n=1
_
C
(n)
T
1
LT
2

(T
1
LT
2
)
c
_
_
_
=
-1
T
1
LT
2
_
_
_
_
_
o
_
n=1
C
(n)
T
1
LT
2
_
_

(T
1
LT
2
)
c
_

_.
Since
_
o
n=1
C
(n)
T
1
LT
2
F
T
1
LT
2
, it follows that
_
o
n=1

(n)
T
1
C. Therefore, C is a
o-led on
T
1
, and E
T
1
_ C. Hence, F
T
1
= C.
(b) We prove that if
T
1
F
T
1
, then

-1
T
1
(
T
1

T
c
1
) =
-1
T
3
_

-1
T
1
,T
3
(
T
1

T
3
~T
1
)
T
c
3
_
. (11.2)
Let
C =

T
1
F
T
1
: (11.2) holds for
T
1
_
.
We rst show that E
T
1
_ C. Pick
iT
1

i
E
T
1
. Then

-1
T
1
_
_
_

iT
1

i
_

T
c
1
_
_
=

iJ
Q
i
.
where Q
i
=
i
for each i J
1
and Q
i
=
i
for each i J
c
1
. Similarly, we have

-1
T
1
,T
3
_
_
_

iT
1

i
_

T
3
~T
1
_
_
=

iT
3
1
i
.
where 1
i
=
i
for each i J
1
and 1
i
=
i
for each i J
3
J
1
. Hence,

-1
T
3
_
_
_
-1
T
1
,T
3
_
_
_

iT
1

T
1
_

T
3
~T
1
_
_

T
c
3
_

_ =
-1
T
3
_

iT
3
1
i
_
=

iJ
S
i
.
where
S
i
=

1
i
if i J
3

i
if i J
c
3
=

i
if i J
1

i
if i J
3
J
1

i
if i J
c
3
=

i
if i J
1

i
if i J
c
1
.
150 CHAPTER 11 ARBITRARY PRODUCTS OF MEASURE SPACES
Therefore,
iT
1

i
C; that is, E
T
1
_ C.
We next show that C is a o-eld on
T
1
. It is clear that

-1
T
1
(
T
1

T
c
1
) = =
-1
T
3
_

-1
T
1
,T
3
(
T
1

T
3
~T
1
)
T
c
3
_
.
so
T
1
C. Now suppose that
T
1
C. Then

-1
T
1
_

c
T
1

T
c
1
_
=
-1
T
1
_
_

T
1

T
c
1
_

T
1

T
c
1
_
_
=
-1
T
1
_

T
1

T
c
1
_

-1
T
1
_

T
1

T
c
1
_
=
-1
T
3
_

-1
T
1
,T
3
_

T
1

T
3
~T
1
_

T
c
3
_
=
-1
T
3
_
_

T
3

T
c
3
_

-1
T
1
,T
3
_

T
1

T
3
~T
1
_

T
c
3
_
_
=
-1
T
3
_
_

T
3

-1
T
1
,T
3
_

T
1

T
3
~T
1
_
_

T
c
3
_
=
-1
T
3
_

-1
T
1
,T
3
_

c
T
1

T
3
~T
1
_

T
c
3
_
:
that is,
c
T
1
C whenever
T
1
C. We nally show that C is closed under
countable intersections. Take an arbitrary sequence {
(n)
T
1
] _ C. Then

-1
T
1
_
_
_
_
_
o
_
n=1

(n)
T
1
_
_

T
c
1
_

_ =
-1
T
1
_
_
o
_
n=1
_

(n)
T
1

T
c
1
_
_
_
=
o
_
n=1

-1
T
1
_

(n)
T
1

T
c
1
_
=
o
_
n=1

-1
T
3
_

-1
T
1
,T
3
_

(n)
T
1

T
3
~T
1
_

T
c
3
_
=
-1
T
3
_
_
_
_

-1
T
1
,T
3
_
_
_
_
_
o
_
n=1

(n)
T
1
_
_

T
3
~T
1
_
_
_
T
c
3
_

_
.
Thus, C is a o-eld containing E
T
1
, and so C = F
T
1
.
(c) Given a sequence {J
n
] of nite subsets of 1, it is not necessarily the case
that
_
o
n=1
J
n
is a nite subset of 1. |L
I Exercise 11.7 (11.2.2). Prove equalities (+) and (++) given in the proof of
Subclaim 3 of Claim 7.
Proof. ./ Take N. Let :
N
mCj

N
m

n1,...,n}
be the bijec-
tion that associates each (o
1
. . . . . o
n}
) with ((o
1
. . . . . o
n
). (o
n1
. . . . . o
n}
)).
We rst prove
SECTION 11.2 CONSTRUCTION OF THE PRODUCT MEASURE 151

-1
N
m
_

N
m

N
c
m
_
=
-1
N
mCj
_

-1
_

N
m

n1,...,n}
_

N
c
mCj
_
. (+)
Dene
C =

N
m
F
N
m
: (+) holds for
N
m
_
.
As usual, we showthat E
N
m
_ C and C is a o-led on
N
m
. Let
N
m
=

n
i=1

i
,
where
i
F
i
for each i {1. . . . . m]. Then

-1
N
m
_
_
_
n

i=1

i
_

N
c
m
_
_
=
1

n

n1

n2
.
and

-1
N
mCj
_
_
_
-1
_
_
_
n

i=1

i
_

n1,...,n}
_
_

N
c
mCj
_

_
=
-1
N
mCj
_
_

1

n

n1

n}
_

N
c
mCj
_
=
1

n

n1

n2
.
Hence, E
N
m
_ C.
We turn to show that C is a o-eld on
N
m
. It is evident that
N
m
C since

-1
N
m
_

N
m

N
c
m
_
= =
-1
N
mCj
_

-1
_

N
m

n1,...,n}
_

N
c
mCj
_
.
Now suppose that
N
m
C. Then

-1
N
m
_

c
N
m

N
c
m
_
=
-1
N
m
_
_

N
m

N
c
m
_

N
m

N
c
m
_
_
= (
1

n

n1

n2
) .
and

-1
N
mCj
_

-1
_

c
N
m

n1,...,n}
_

N
c
mCj
_
=
-1
N
mCj
_
_

N
mCj

_

1

n

n1

n}
_
_

N
c
mCj
_
=
-1
N
mCj
_ _

N
mCj

N
c
mCj
_

_
_

1

n

n}

n}
_

N
c
mCj
_ _
= (
1

n

n1

n2
) :
that is,
N
m
C forces
c
N
m
C. To verify that C is closed under countable
unions, take an arbitrary sequence {
(n)
N
m
] _ C. We then have
152 CHAPTER 11 ARBITRARY PRODUCTS OF MEASURE SPACES

-1
N
m
_
_
_
_
_
o
_
n=1

(n)
N
m
_
_

N
c
m
_

_ =
o
_
n=1

-1
N
m
_

(n)
N
m

N
c
m
_
.
and

-1
N
mCj
_
_
_
_

-1
_
_
_
_
_
o
_
n=1

(n)
N
m
_
_

n1,...,n}
_
_
_
N
c
mCj
_

_
=
-1
N
mCj
_
_
_
_
_
o
_
n=1

-1
_

(n)
N
m

n1,...,n}
_
_
_

N
c
mCj
_

_
=
o
_
n=1

-1
N
mCj
_

-1
_

(n)
N
m

n1,...,n}
_

N
c
mCj
_
=
o
_
n=1

-1
N
m
_

(n)
N
m

N
c
m
_
=
-1
N
m
_
_
_
_
_
o
_
n=1

(n)
N
m
_
_

N
c
m
_

_.
Hence,
_
o
n=1

(n)
N
m
C. Therefore, C is a o-led containing E
N
m
, and so C =
F
N
m
.
() We now prove
1

1
(
Nm

fmC1;:::;mCjg
)
_

1
. . . . .
n
. o
n1
. . . . . o
n}
_
= 1

Nm
_

1
. . . . .
n
. o
n1
. . . . . o
n
_
.
(++)
Dene
D =

N
m
F
N
m
: (++) holds for each o
n1

n1
. . . . . o
n

n
_
.
Once again, we prove this claim by showing that E
N
m
_ D and D is a o-led
on
N
m
.
Let
N
m
=

n
i=1

i
, where
i
F
i
for each i {1. . . . . m]. Then
SECTION 11.2 CONSTRUCTION OF THE PRODUCT MEASURE 153
1

1
_
_

m
iD1

i
_

fmC1;:::;mCjg
_
_

1
. . . . .
n
. o
n1
. . . . . o
n}
_
= 1
==
_

1
. . . . .
n
. o
n1
. . . . . o
n}
_

-1
_
_
_
n

i=1

i
_

n1,...,n}
_
_
==
_

1
. . . . .
n
. o
n1
. . . . . o
n}
_

1

n

n1

n}
==
_

1
. . . . .
n
. o
n1
. . . . o
n
_

1

n
== 1

m
_

1
. . . . .
n
. o
n1
. . . . o
n
_
= 1:
that is, E
N
m
_ D.
We next show that D is a o-eld on
N
m
. It is easy to see that
1

1
(
Nm

fmC1;:::;mCjg
)
_

1
. . . . .
n
. o
n1
. . . . . o
n}
_
= 1 = 1

Nm
_

1
. . . . .
n
. o
n1
. . . . . o
n
_
.
so
N
m
D. Suppose that
N
m
D. Then
1

1
_

c
Nm

fmC1;:::;mCjg
_
_

1
. . . . .
n
. o
n1
. . . . . o
n}
_
= 1
==
_

1
. . . . .
n
. o
n1
. . . . . o
n}
_

-1
_

c
N
m

n1,...,n}
_
==
_

1
. . . . .
n
. o
n1
. . . . . o
n}
_

N
mCj

-1
_

N
m

n1,...,n}
_
== 1

1
(
Nm

fmC1;:::;mCjg
)
_

1
. . . . .
n
. o
n1
. . . . . o
n}
_
= 0
== 1

Nm
_

1
. . . . .
n
. o
n1
. . . . . o
n
_
= 0
== 1

c
Nm
_

1
. . . . .
n
. o
n1
. . . . . o
n
_
= 1.
Hence,
c
N
m
D whenever
N
m
D. Finally, we verify that D is closed under
countable unions. Take an arbitrary sequence {
(n)
N
m
] _ D. Observe that
1

1
_
_
_
1
nD1

.n/
Nm
_

fmC1;:::;mCjg
_
= 1_
1
nD1

1
_

.n/
Nm

fmC1;:::;mCjg
_
= sup
n
1

1
_

.n/
Nm

fmC1;:::;mCjg
_
.
and
1
_
1
nD1

.n/
Nm
= sup
n
1

.n/
Nm
.
Since
1

1
_

.n/
Nm

fmC1;:::;mCjg
_
_

1
. . . . .
n
. o
n1
. . . . . o
n}
_
= 1

.n/
Nm
_

1
. . . . .
n
. o
n1
. . . . . o
n
_
for every n N, we get
154 CHAPTER 11 ARBITRARY PRODUCTS OF MEASURE SPACES
sup
n
1

1
_

.n/
Nm

fmC1;:::;mCjg
_
_

1
. . . . .
n
. o
n1
. . . . . o
n}
_
= sup
n
1

.n/
Nm
_

1
. . . . .
n
. o
n1
. . . . . o
n
_
:
that is,
_
o
n=1

(n)
N
m
D. This proves that D = F
N
m
. |L
References
[1] Aliprantis, Charalambos D. and Kim C. Border (2006) Innite Dimen-
sional Analysis: A Hitchhikers Guide, Berlin: Springer-Verlag, 3rd edition.
[7]
[2] Ash, Robert B. (2009) Real Variables with Basic Metric Space Topology,
New York: Dover Publications, Inc. [35, 79]
[3] Ash, Robert B. and Catherine A. Dolans-Dade (2000) Probability &
Measure Theory, New York: Academic Press, 2nd edition. [16, 25, 27]
[4] Athreya, Krishna B. and Soumendra N. Lahiri (2006) Measure Theory
and Probability Theory: Springer-Verlag. [22]
[5] Bear, H. S. (2002) A Primer of Lebesgue Integration, San Diego, California:
Academic Press, 2nd edition. [57]
[6] Billingsley, Patrick (1995) Probability and Measure, Wiley Series in
Probability and Mathematical Statistics, New York: Wiley-Interscience, 3rd
edition. [27]
[7] Bogachev, Vladimir I. (2007) Measure Theory, I, New York: Springer-
Verlag. [7]
[8] Broughton, A. and B. W. Huff (1977) A Comment on Unions of Sigma-
Fields, American Mathematical Monthly, 84 (7), pp. 553554. [22]
[9] Chung, Kai Lai (2001) A Course in Probability Theory, San Diego, Cali-
fornia: Academic Press, 3rd edition. [21, 27]
[10] Dudley, Richard Mansfield (2002) Real Analysis and Probability, 74
of Cambridge Studies in Advanced Mathematics, New York: Cambridge
University Press, 2nd edition. [7]
[11] Folland, Gerald B (1999) Real Analysis: Modern Techniques and Their
Applications, Pure and Applied Mathematics: A Wiley-Interscience Series
of Texts, Monographs and Tracts, New York: John Wiley & Sons, Inc. 2nd
edition. [16, 94, 125]
155
156 REFERENCES
[12] Gamelin, Theodore W. and Robert Everist Greene (1999) Introduc-
tion to Topology, New York: Dover Publications, Inc. 2nd edition. [23]
[13] Georgakis, Constantine (1994) A Note on the Gaussian Integral,
Mathematics Magazine, 67 (1), p. 47. [61]
[14] Halmos, Paul R. (1974) Measure Theory, 18 of Graduate Texts in Math-
ematics, New York: Springer-Verlag. [73]
[15] Hewitt, Edwin and Karl Stromberg (1965) Real and Abstract Analysis,
25 of Graduate Texts in Mathematics, New York: Springer-Verlag. [118]
[16] Hrbacek, Karel and Thomas Jech (1999) Introduction to Set Theory,
220 of Pure and Applied Mathematics: A Series of Monographs and Text-
books, New York: Taylor & Francis Group, LLC, 3rd edition. [13]
[17] Klenke, Achim (2008) Probability Theory: A Comprehensive Course, Uni-
versitext, London: Springer-Verlag. [1, 2]
[18] Pollard, David (2001) A Users Guide to Measure Theoretic Probability,
Cambridge Series in Statistical and Probabilistic Mathematics: Cambridge
University Press. [29]
[19] Resnick, Sidney I. (1999) A Probability Path, Boston: Birkhuser. [41, 72,
90]
[20] Rosenthal, Jeffrey S. (2006) A First Look at Rigorous Probability The-
ory, Singapore: World Scientic, 2nd edition. [35]
[21] Rudin, Walter (1976) Principles of Mathematical Analysis, New York:
McGraw-Hill Companies, Inc. 3rd edition. [10, 43]
[22] (1986) Real and Complex Analysis, New York: McGraw-Hill Com-
panies, Inc. 3rd edition. [118]
[23] Shirali, Satish and Harkrishan L. Vasudeva (2006) Metric Spaces,
London: Springer-Verlag. [114]
[24] Vestrup, Eric M. (2003) The Theory of Measures and Integration, Wi-
ley Series in Probability and Statistics, Hoboken, New Jersey: Wiley-
Interscience. [i, 10, 26, 35, 43, 95]
[25] Willard, Stephen (2004) General Topology, New York: Dover Publica-
tions, Inc. [142]

You might also like