Tutorial 12: Radon-Nikodym Theorem 1

12. Radon-Nikodym Theorem

In the following, (, T) is an arbitrary measurable space.
Denition 96 Let and be two (possibly complex) measures on
(, T). We say that is absolutely continuous with respect to ,
and we write << , if and only if, for all E T:
(E) = 0 (E) = 0
Exercise 1. Let be a measure on (, T) and M
1
(, T). Show
that << is equivalent to [[ << .
Exercise 2. Let be a measure on (, T) and M
1
(, T). Let
> 0. Suppose there exists a sequence (E
n
)
n1
in T such that:
n 1 , (E
n
)
1
2
n
, [(E
n
)[
Dene:
E

= limsup
n1
E
n

n1
_
kn
E
k
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 2
1. Show that:
(E) = lim
n+

_
kn
E
k

= 0
2. Show that:
[[(E) = lim
n+
[[

_
kn
E
k

3. Let be a measure on (, T). Can we conclude in general that:
(E) = lim
n+

_
kn
E
k

4. Prove the following:

www.probability.net
Tutorial 12: Radon-Nikodym Theorem 3
Theorem 58 Let be a measure on (, T) and be a complex
measure on (, T). The following are equivalent:
(i) <<
(ii) [[ <<
(iii) > 0, > 0, E T, (E) [(E)[ <
Exercise 3. Let be a measure on (, T) and M
1
(, T) such
that << . Let
1
= Re() and
2
= Im().
1. Show that
1
<< and
2
<< .
2. Show that
+
1
,

1
,
+
2
,

2
are absolutely continuous w.r. to .
Exercise 4. Let be a nite measure on (, T) and f L
1
C
(, T, ).
Let S be a closed proper subset of C. We assume that for all E T
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 4
such that (E) > 0, we have:
1
(E)
_
E
fd S
1. Show there is a sequence (D
n
)
n1
of closed discs in C, with:
S
c
=
+
_
n=1
D
n
Let
n
C, r
n
> 0 be such that D
n
= z C : [z
n
[ r
n
.
2. Suppose (E
n
) > 0 for some n 1, where E
n
= f D
n
.
Show that:

1
(E
n
)
_
E
n
fd
n

1
(E
n
)
_
E
n
[f
n
[d r
n
3. Show that for all n 1, (f D
n
) = 0.
4. Prove the following:
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 5
Theorem 59 Let be a nite measure on (, T), f L
1
C
(, T, ).
Let S be a closed subset of C such that for all E T with (E) > 0,
we have:
1
(E)
_
E
fd S
Then, f S -a.s.
Exercise 5. Let be a -nite measure on (, T). Let (E
n
)
n1
be
a sequence in T such that E
n
and (E
n
) < + for all n 1.
Dene w : (, T) (R, B(R)) as:
w

=
+

n=1
1
2
n
1
1 +(E
n
)
1
E
n
1. Show that for all , 0 < w() 1.
2. Show that w L
1
R
(, T, ).
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 6
Exercise 6. Let be a -nite measure on (, T) and be a nite
measure on (, T), such that << . Let w L
1
R
(, T, ) be such
that 0 < w 1. We dene =
_
wd, i.e.
E T , (E)

=
_
E
wd
1. Show that is a nite measure on (, T).
2. Show that = + is also a nite measure on (, T).
3. Show that for all f L
1
C
(, T, ), we have f L
1
C
(, T, ),
fw L
1
C
(, T, ), and:
_
fd =
_
fd +
_
fwd
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 7
4. Show that for all f L
2
C
(, T, ), we have:
_
[f[d
_
[f[d
__
[f[
2
d
_
1
2
(())
1
2
5. Show that L
2
C
(, T, ) L
1
C
(, T, ), and for f L
2
C
(, T, ):

_
fd

_
().|f|
2
6. Show the existence of g L
2
C
(, T, ) such that:
f L
2
C
(, T, ) ,
_
fd =
_
fgd (1)
7. Show that for all E T such that (E) > 0, we have:
1
(E)
_
E
gd [0, 1]
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 8
8. Show the existence of g L
2
C
(, T, ) such that g() [0, 1]
for all , and (1) still holds.
9. Show that for all f L
2
C
(, T, ), we have:
_
f(1 g)d =
_
fgwd
10. Show that for all n 1 and E T,
f

= (1 +g +. . . +g
n
)1
E
L
2
C
(, T, )
11. Show that for all n 1 and E T,
_
E
(1 g
n+1
)d =
_
E
g(1 +g +. . . +g
n
)wd
12. Dene:
h

= gw
_
+

n=0
g
n
_
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 9
Show that if A = 0 g < 1, then for all E T:
(E A) =
_
E
hd
13. Show that h = + = A
c
and conclude that (A
c
) = 0.
14. Show that for all E T, we have (E) =
_
E
hd.
15. Show that if is -nite on (, T), and is a nite measure
on (, T) such that << , there exists h L
1
R
(, T, ), such
that h 0 and:
E T , (E) =
_
E
hd
16. Prove the following:
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 10
Theorem 60 (Radon-Nikodym:1) Let be a -nite measure on
(, T). Let be a complex measure on (, T) such that << .
Then, there exists some h L
1
C
(, T, ) such that:
E T , (E) =
_
E
hd
If is a signed measure on (, T), we can assume h L
1
R
(, T, ).
If is a nite measure on (, T), we can assume h 0.
Exercise 7. Let f = u +iv L
1
C
(, T, ), such that:
E T ,
_
E
fd = 0
where is a measure on (, T).
1. Show that:
_
u
+
d =
_
{u0}
ud
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 11
2. Show that f = 0 -a.s.
3. State and prove some uniqueness property in theorem (60).
Exercise 8. Let and be two -nite measures on (, T) such
that << . Let (E
n
)
n1
be a sequence in T such that E
n
and
(E
n
) < + for all n 1. We dene:
n 1 ,
n

=
E
n

= (E
n
)
1. Show that there exists h
n
L
1
R
(, T, ) with h
n
0 and:
E T ,
n
(E) =
_
E
h
n
d (2)
for all n 1.
2. Show that for all E T,
_
E
h
n
d
_
E
h
n+1
d
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 12
3. Show that for all n, p 1,
(h
n
h
n+1
>
1
p
) = 0
4. Show that h
n
h
n+1
-a.s.
5. Show the existence of a sequence (h
n
)
n1
in L
1
R
(, T, ) such
that 0 h
n
h
n+1
for all n 1 and with (2) still holding.
6. Let h = sup
n1
h
n
. Show that:
E T , (E) =
_
E
hd (3)
7. Show that for all n 1,
_
E
n
hd < +.
8. Show that h < + -a.s.
9. Show there exists h : (, T) R
+
measurable, while (3) holds.
10. Show that for all n 1, h L
1
R
(, T,
E
n
).
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 13
Theorem 61 (Radon-Nikodym:2) Let and be two -nite
measures on (, T) such that << . There exists a measurable
map h : (, T) (R
+
, B(R
+
)) such that:
E T , (E) =
_
E
hd
Exercise 9. Let h, h

: (, T) [0, +] be two non-negative and

measurable maps. Let be a -nite measure on (, T). We assume:
E T ,
_
E
hd =
_
E
h

d
Let (E
n
)
n1
be a sequence in T with E
n
and (E
n
) < + for
all n 1. We dene F
n
= E
n
h n for all n 1.
1. Show that for all n and E T,
_
E
hd
F
n
=
_
E
h

d
F
n
< +.
2. Show that for all n, p 1, (F
n
h > h

+ 1/p) = 0.
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 14
3. Show that for all n 1, (F
n
h ,= h

) = 0.
4. Show that (h ,= h

h < +) = 0.
5. Show that h = h

-a.s.
6. State and prove some uniqueness property in theorem (61).
Exercise 10. Take = and T = T() = , . Let be
the measure on (, T) dened by () = 0 and () = +. Let
h, h

: (, T) [0, +] be dened by h() = 1 ,= 2 = h

(). Show
that we have:
E T ,
_
E
hd =
_
E
h

d
Explain why this does not contradict the previous exercise.
Exercise 11. Let be a complex measure on (, T).
1. Show that << [[.
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 15
2. Show the existence of some h L
1
C
(, T, [[) such that:
E T , (E) =
_
E
hd[[
3. If is a signed measure, can we assume h L
1
R
(, T, [[)?
Exercise 12. Further to ex. (11), dene A
r
= [h[ < r for all r > 0.
1. Show that for all measurable partition (E
n
)
n1
of A
r
:
+

n=1
[(E
n
)[ r[[(A
r
)
2. Show that [[(A
r
) = 0 for all 0 < r < 1.
3. Show that [h[ 1 [[-a.s.
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 16
4. Suppose that E T is such that [[(E) > 0. Show that:

1
[[(E)
_
E
hd[[

1
5. Show that [h[ 1 [[-a.s.
6. Prove the following:
Theorem 62 For all complex measure on (, T), there exists h
belonging to L
1
C
(, T, [[) such that [h[ = 1 and:
E T , (E) =
_
E
hd[[
If is a signed measure on (, T), we can assume h L
1
R
(, T, [[).
Exercise 13. Let A T, and (A
n
)
n1
be a sequence in T.
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 17
1. Show that if A
n
A then 1
A
n
1
A
.
2. Show that if A
n
A then 1
A
n
1
A
.
3. Show that if 1
A
n
1
A
, then for all M
1
(, T):
(A) = lim
n+
(A
n
)
Exercise 14. Let be a measure on (, T) and f L
1
C
(, T, ).
1. Show that =
_
fd M
1
(, T).
2. Let h L
1
C
(, T, [[) be such that [h[ = 1 and =
_
hd[[.
Show that for all E, F T:
_
E
f1
F
d =
_
E
h1
F
d[[
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 18
3. Show that if g : (, T) (C, B(C)) is bounded and measurable:
E T ,
_
E
fgd =
_
E
hgd[[
4. Show that:
E T , [[(E) =
_
E
f

hd
5. Show that for all E T,
_
E
Re(f

h)d 0 ,
_
E
Im(f

h)d = 0
6. Show that f

h R
+
-a.s.
7. Show that f

h = [f[ -a.s.
8. Prove the following:
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 19
Theorem 63 Let be a measure on (, T) and f L
1
C
(, T, ).
Then, =
_
fd dened by:
E T , (E)

=
_
E
fd
is a complex measure on (, T) with total variation:
E T , [[(E) =
_
E
[f[d
Exercise 15. Let M
1
(, T) be a signed measure. Suppose that
h L
1
R
(, T, [[) is such that [h[ = 1 and =
_
hd[[. Dene
A = h = 1 and B = h = 1.
1. Show that for all E T,
+
(E) =
_
E
1
2
(1 +h)d[[.
2. Show that for all E T,

(E) =
_
E
1
2
(1 h)d[[.
3. Show that
+
=
A
= (A ).
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 20
4. Show that

=
B
= (B ).
Theorem 64 (Hahn Decomposition) Let be a signed measure
on (, T). There exist A, B T, such that A B = , = A B
and for all E T,
+
(E) = (A E) and

(E) = (B E).
Denition 97 Let be a complex measure on (, T). We dene:
L
1
C
(, T, )

= L
1
C
(, T, [[)
and for all f L
1
C
(, T, ), the Lebesgue integral of f with respect
to , is dened as:
_
fd

=
_
fhd[[
where h L
1
C
(, T, [[) is such that [h[ = 1 and =
_
hd[[.
Exercise 16. Let be a complex measure on (, T).
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 21
1. Show that for all f : (, T) (C, B(C)) measurable:
f L
1
C
(, T, )
_
[f[d[[ < +
2. Show that for f L
1
C
(, T, ),
_
fd is unambiguously dened.
3. Show that for all E T, 1
E
L
1
C
(, T, ) and
_
1
E
d = (E).
4. Show that if is a nite measure, then [[ = .
5. Show that if is a nite measure, denition (97) of integral
and space L
1
C
(, T, ) is consistent with that already known
for measures.
6. Show that L
1
C
(, T, ) is a C-vector space and that:
_
(f +g)d =
_
fd +
_
gd
for all f, g L
1
C
(, T, ) and C.
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 22
7. Show that for all f L
1
C
(, T, ), we have:

_
fd

_
[f[d[[
Exercise 17. Let , M
1
(, T), let C.
1. Show that [[ = [[.[[
2. Show that [ +[ [[ +[[
3. Show that L
1
C
(, T, ) L
1
C
(, T, ) L
1
C
(, T, +)
4. Show that for all E T:
_
1
E
d( +) =
_
1
E
d +
_
1
E
d
5. Show that for all f L
1
C
(, T, ) L
1
C
(, T, ):
_
fd( +) =
_
fd +
_
fd
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 23
Exercise 18. Let f : (, T) [0, +] be non-negative and measur-
able. Let and be measures on (, T), and [0, +]:
1. Show that + is a measure on (, T) and:
_
fd( +) =
_
fd +
_
fd
2. Show that if , then:
_
fd
_
fd
Exercise 19. Let M
1
(, T),
1
= Re() and
2
= Im().
1. Show that [
1
[ [[ and [
2
[ [[.
2. Show that [[ [
1
[ +[
2
[.
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 24
3. Show that L
1
C
(, T, ) = L
1
C
(, T,
1
) L
1
C
(, T,
2
).
4. Show that:
L
1
C
(, T,
1
) = L
1
C
(, T,
+
1
) L
1
C
(, T,

1
)
L
1
C
(, T,
2
) = L
1
C
(, T,
+
2
) L
1
C
(, T,

2
)
5. Show that for all f L
1
C
(, T, ):
_
fd =
_
fd
+
1

_
fd

1
+i
__
fd
+
2

_
fd

2
_
Exercise 20. Let M
1
(, T). Let A T. Let h L
1
C
(, T, [[)
be such that [h[ = 1 and =
_
hd[[. Recall that
A
= (A ) and

|A
=
|(F
|A
)
where T
|A
= A E , E T T.
1. Show that we also have T
|A
= E : E T , E A.
2. Show that
A
M
1
(, T) and
|A
M
1
(A, T
|A
).
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 25
3. Let E T and (E
n
)
n1
be a measurable partition of E. Show:
+

n=1
[
A
(E
n
)[ [[
A
(E)
4. Show that we have [
A
[ [[
A
.
5. Let E T and (E
n
)
n1
be a measurable partition of A E.
Show that:
+

n=1
[(E
n
)[ [
A
[(A E)
6. Show that [
A
[(A
c
) = 0.
7. Show that [
A
[ = [[
A
.
8. Let E T
|A
and (E
n
)
n1
be an T
|A
-measurable partition of E.
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 26
Show that:
+

n=1
[
|A
(E
n
)[ [[
|A
(E)
9. Show that [
|A
[ [[
|A
.
10. Let E T
|A
T and (E
n
)
n1
be a measurable partition of E.
Show that (E
n
)
n1
is also an T
|A
-measurable partition of E,
and conclude:
+

n=1
[(E
n
)[ [
|A
[(E)
11. Show that [
|A
[ = [[
|A
.
12. Show that
A
=
_
hd[
A
[.
13. Show that h
|A
L
1
C
(A, T
|A
, [
|A
[) and
|A
=
_
h
|A
d[
|A
[.
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 27
14. Show that for all f L
1
C
(, T, ), we have:
f1
A
L
1
C
(, T, ) , f L
1
C
(, T,
A
) , f
|A
L
1
C
(A, T
|A
,
|A
)
and:
_
f1
A
d =
_
fd
A
=
_
f
|A
d
|A
Denition 98 Let f L
1
C
(, T, ) , where is a complex measure
on (, T). let A T. We call partial Lebesgue integral of f with
respect to over A, the integral denoted
_
A
fd, dened as:
_
A
fd

=
_
(f1
A
)d =
_
fd
A
=
_
(f
|A
)d
|A
where
A
is the complex measure on (, T),
A
= (A ), f
|A
is
the restriction of f to A and
|A
is the restriction of to T
|A
, the
trace of T on A.
Exercise 21. Prove the following:
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 28
Theorem 65 Let f L
1
C
(, T, ), where is a complex measure
on (, T). Then, =
_
fd dened as:
E T , (E)

=
_
E
fd
is a complex measure on (, T), with total variation:
E T , [[(E) =
_
E
[f[d[[
Moreover, for all measurable map g : (, T) (C, B(C)), we have:
g L
1
C
(, T, ) gf L
1
C
(, T, )
and when such condition is satised:
_
gd =
_
gfd
Exercise 22. Let (
1
, T
1
), . . . , (
n
, T
n
) be measurable spaces, where
n 2. Let
1
M
1
(
1
, T
1
), . . .,
n
M
1
(
n
, T
n
). For all i N
n
,
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 29
let h
i
belonging to L
1
C
(
i
, T
i
, [
i
[) be such that [h
i
[ = 1 and
i
=
_
h
i
d[
i
[. For all E T
1
. . . T
n
, we dene:
(E)

=
_
E
h
1
. . . h
n
d[
1
[ . . . [
n
[
1. Show that M
1
(
1
. . .
n
, T
1
. . . T
n
)
2. Show that for all measurable rectangle A
1
. . . A
n
:
(A
1
. . . A
n
) =
1
(A
1
) . . .
n
(A
n
)
3. Prove the following:
Theorem 66 Let
1
, . . . ,
n
be n complex measures on measurable
spaces (
1
, T
1
), . . . , (
n
, T
n
) respectively, where n 2. There exists
a unique complex measure
1
. . .
n
on (
1
. . .
n
, T
1
. . .T
n
)
such that for all measurable rectangle A
1
. . . A
n
, we have:

1
. . .
n
(A
1
. . . A
n
) =
1
(A
1
) . . .
n
(A
n
)
www.probability.net
Tutorial 12: Radon-Nikodym Theorem 30
Exercise 23. Further to theorem (66) and exercise (22):
1. Show that [
1
. . .
n
[ = [
1
[ . . . [
n
[.
2. Show that |
1
. . .
n
| = |
1
| . . . |
n
|.
3. Show that for all E T
1
. . . T
n
:

1
. . .
n
(E) =
_
E
h
1
. . . h
n
d[
1
. . .
n
[
4. Let f L
1
C
(
1
. . .
n
, T
1
. . . T
n
,
1
. . .
n
). Show:
_
fd
1
. . .
n
=
_
fh
1
. . . h
n
d[
1
[ . . . [
n
[
5. let be a permutation of 1, . . . , n. Show that:
_
fd
1
. . .
n
=
_

(n)
. . .
_

(1)
fd
(1)
. . . d
(n)
www.probability.net
Solutions to Exercises 31
Solutions to Exercises
Exercise 1. Let be a measure on (, T) and M
1
(, T). Sup-
pose that << . Let E T be such that (E) = 0. Let (E
n
)
n1
be a measurable partition of E. For each n 1, we have E
n
E and
consequently (E
n
) (E). It follows that (E
n
) = 0 for all n 1,
and from << we obtain that (E
n
) = 0 for all n 1. Hence:
+

n=1
[(E
n
)[ = 0
This being true for all measurable partition (E
n
)
n1
of E, it follows
from denition (94) that [[(E) = 0. We have proved the implication
that (E) = 0 [[(E) = 0 and consequently [[ << . Conversely,
if [[ << and (E) = 0, then [[(E) = 0. From [(E)[ [[(E) we
conclude that (E) = 0. So << . We have proved the equivalence
between << and [[ << . Note that is assumed to be a
measure, and not a complex measure.
Exercise 1
www.probability.net
Solutions to Exercises 32
Exercise 2.
1. Dene B
n
=
kn
E
k
for n 1. By assumption, (E
k
) 1/2
k
for all k 1 and consequently:
(B
n
)
+

k=n
(E
k
)
+

k=n
1
2
k
=
1
2
n1
< +
It follows that (B
n
) 0 as n +. Furthermore, since E
is dened as E =
n1
B
n
and B
n+1
B
n
for all n 1, we
have B
n
E. From (B
1
) < + and theorem (8), we obtain
(B
n
) (E) as n +. We have proved that:
(E) = lim
n+

_
kn
E
k

= 0
2. If B
n
=
kn
E
k
, then E =
n1
B
n
and B
n+1
B
n
for all
n 1. From theorem (57), the total variation [[ of the complex
measure is a nite measure. In particular [[(B
1
) < +,
www.probability.net
Solutions to Exercises 33
and applying theorem (8), it follows that [[(B
n
) [[(E) as
n +. Furthermore, since E
n
B
n
for all n 1, we have:
[(E
n
)[ [[(E
n
) [[(B
n
)
and in particular lim[[(B
n
) . We have proved that:
[[(E) = lim
n+
[[

_
kn
E
k

3. Let be a measure on (, T) and B
n
=
kn
E
k
for n 1. Since
E =
n1
B
n
and B
n+1
B
n
for all n 1, it is very tempting
to conclude that (B
n
) (E) as n +. However, a careful
reading of theorem (8) shows that we cannot safely apply this
theorem, unless (B
1
) < +(or at least (B
p
) < + for some
p 1), which in general is not true. So in general, we cannot
www.probability.net
Solutions to Exercises 34
conclude that:
(E) = lim
n+

_
kn
E
k

When = or = [[, we crucially used the assumption that

(E
k
) 1/2
k
for all k 1, and the fact that [[ is a nite
measure, to obtain (B
1
) < +.
4. Let be a measure on (, T) and be a complex measure on
(, T). The fact that << is equivalent to [[ << , has
already been proved in exercise (1). Suppose the condition:
> 0, > 0, E T, (E) [(E)[ < (4)
holds. Let E T be such that (E) = 0. Applying (4), for all
> 0, there exists > 0 such that if E

T satises (E

) ,
then [(E

)[ < . Since (E) = 0, we have (E) for all

> 0 and consequently [(E)[ < for all > 0. So (E) = 0.
This shows that is absolutely continuous with respect to ,
www.probability.net
Solutions to Exercises 35
and we have proved that (4) << . Conversely, suppose
that << , and that condition (4) does not hold. There exists
> 0 such that for all > 0 we can nd some E

T with
the property that (E

) and [(E

)[ . Taking of the
form = 1/2
n
for n 1, there exists a sequence (E
n
)
n1
in T,
such that (E
n
) 1/2
n
and [(E
n
)[ for all n 1. Dening
E = limsup E
n
=
n1

kn
E
k
, we have (E) = 0 from 1.
and [[(E) from 2. This contradicts the fact that [[ << ,
or equivalently the fact that << . We have proved that
<< (4), which completes the proof of theorem (58).
Exercise 2
www.probability.net
Solutions to Exercises 36
Exercise 3.
1. Let be a measure on (, T) and M
1
(, T). Suppose that
<< . Let E T be such that (E) = 0. Then (E) = 0. In
particular,
1
(E) = Re((E)) = 0 and
2
(E) = Im((E)) = 0.
This shows that
1
<< and
2
<< .
2. From 1. we have
1
<< . From exercise (1), this is equivalent
to [
1
[ << . Hence, if E T is such that (E) = 0, then

1
(E) = 0 and [
1
[(E) = 0. It follows that:

+
1
(E) =
1
2
([
1
[(E) +
1
(E)) = 0
and

1
(E) =
1
2
([
1
[(E)
1
(E)) = 0
We conclude that
+
1
<< and

1
<< . We prove similarly
that
+
2
and

2
are absolutely continuous with respect to .
Exercise 3
www.probability.net
Solutions to Exercises 37
Exercise 4.
1. Since S is a closed proper subset of C, its complement S
c
is an
open subset of C, which is not empty. Let z = x + iy S
c
.
There exists > 0 such that B(z, ) S
c
. Let x

, y

Q be
such that [x x

[ < /2

2 and [y y

[ < /2

2, and dene
z

= x

+iy

. Then:
[z z

[ =
_
[x x

[
2
+[y y

[
2
< /2
Let

Q be such that [z z

[ <

< /2. Then it is clear

that z B(z

) and furthermore, for all z

C such that
[z

, we have:
[z z

[ [z z

[ +[z

[ < 2

<
It follows that z

B(z

) B(z, ) S
c
, where

B(z

)
denotes the closed disc with center z

and radius

. Hence, for
all z S
c
, we are able to nd a closed disc D
z
in C, such that
z D
z
S
c
, and furthermore, such closed disc can be chosen
www.probability.net
Solutions to Exercises 38
to have a rational radius (

Q), and a center with rational

coordinates (x

, y

Q). In particular, to each D

z
where z S
c
,
can be associated a triple (x
z
, y
z
,
z
) in Q
3
, dening a mapping
which is injective. Q
3
being a countable set, it follows that
T = D
z
: z S
c
is at most countable (and non-empty),
and consequently there exists a surjective map : N

T.
Dening D
n
= (n), from S
c
=
zS
c D
z
we obtain:
S
c
=
_
DD
D =
+
_
n=1
(n) =
+
_
n=1
D
n
2. Since is a nite measure and (E
n
) > 0, it is always possible
to write the complex number
n
as
n
= (E
n
)
1
_
E
n

n
d.
Consequently, using theorem (24), we have:

1
(E
n
)
_
E
n
fd
n

1
(E
n
)
_
E
n
[f
n
[d
Since E
n
= f D
n
= [f
n
[ r
n
, we have the inequality
www.probability.net
Solutions to Exercises 39
[f
n
[1
E
n
r
n
1
E
n
, and consequently:
1
(E
n
)
_
E
n
[f
n
[d
1
(E
n
)
_
r
n
1
E
n
d = r
n
We have proved that:

1
(E
n
)
_
E
n
fd
n

1
(E
n
)
_
E
n
[f
n
[d r
n
3. Let n 1 and E
n
= f D
n
. Suppose (E
n
) > 0. Then:
1
(E
n
)
_
E
n
fd S (5)
by assumption. However, from 2.:

1
(E
n
)
_
E
n
fd
n

r
n
or equivalently:
1
(E
n
)
_
E
n
fd D
n
(6)
www.probability.net
Solutions to Exercises 40
Since D
n
S
c
, (5) and (6) form a contradiction. It follows that
the assumption (E
n
) > 0 is absurd and therefore (E
n
) = 0.
We have proved that (f D
n
) = 0 for all n 1.
4. Let be a nite measure on (, T) and f L
1
C
(, T, ). Let S
be a closed subset of C such that for all E T with (E) > 0:
1
(E)
_
E
fd S
We claim that f S -a.s. If S = C, there is nothing further
to prove. We assume that S is a proper (closed) subset of C.
Let (D
n
)
n1
be a sequence of closed discs in C as in 1. Then
S
c
=
n1
D
n
and from 3. (f D
n
) = 0 for all n 1. From
f S
c
=
n1
f D
n
we obtain:
(f S
c
)
+

n=1
(f D
n
) = 0
It follows that if N = f S
c
, then N T, (N) = 0 and
www.probability.net
Solutions to Exercises 41
f() S for all N
c
. This shows that f S -a.s. We have
proved theorem (59).
Exercise 4
www.probability.net
Solutions to Exercises 42
Exercise 5.
1. Let . Since E
n
, in particular =
n1
E
n
. There
exists p 1 such that E
p
. Hence:
w() =
+

n=1
1
2
n
1
1 +(E
n
)
1
E
n
()
1
2
p
1
1 +(E
p
)
> 0
Furthermore:
w() =
+

n=1
1
2
n
1
1 +(E
n
)
1
E
n
()
+

n=1
1
2
n
= 1
2. w is R-valued, measurable, and from theorem (19):
_
[w[d =
_
wd =
+

n=1
1
2
n
1
1 +(E
n
)
_
1
E
n
d < +
So w L
1
R
(, T, ).
Exercise 5
www.probability.net
Solutions to Exercises 43
Exercise 6.
1. The fact that =
_
wd is a measure on (, T) stems from
a direct application of theorem (21). However, the result is
pretty straightforward, with or without theorem (21): it is clear
that () = 0 and furthermore from the monotone convergence
theorem (19):
(E) =
_
1
E
wd =
+

n=1
_
1
E
n
wd =
+

n=1
(E
n
)
for any E T and (E
n
)
n1
measurable partition of E. Since w
is non-negative and is an element of L
1
R
(, T, ), we have:
() =
_
wd =
_
[w[d < +
So is a nite measure.
2. Since both and are nite measures on (, T), they are com-
plex measures with values in R
+
. So = + is a complex
www.probability.net
Solutions to Exercises 44
measure on (, T) (M
1
(, T) is a vector space), and it has val-
ues in R
+
. It follows that is a nite measure. Alternatively,
you may wish to argue that is a measure (as the sum of two
measures), and that () = () + () < + since both
and are nite.
3. Let f : (, T) [0, +] be a non-negative and measurable
map, and consider the equality:
_
fd =
_
fd +
_
fwd (7)
Since = + and =
_
wd, this equality is true whenever
f is of the form f = 1
E
with E T. By linearity, equation (7)
is also true whenever f is a simple function on (, T). If f is an
arbitrary non-negative and measurable map, from theorem (18)
there exists a sequence (s
n
)
n1
of simple functions on (, T),
such that s
n
f. Applying equation (7) for each n 1, we
www.probability.net
Solutions to Exercises 45
obtain:
_
s
n
d =
_
s
n
d +
_
s
n
wd (8)
Since w is non-negative, (s
n
w)
n1
is a non-decreasing sequence
of non-negative and measurable maps, converging simply (i.e.
pointwise ) to fw. In short, we have s
n
w fw, and from the
monotone convergence theorem (19), taking the limit in (8) as
n +, we conclude that equation (7) is also true for f.
Suppose now that f L
1
C
(, T, ). Applying (7) to [f[, we
obtain:
_
[f[d +
_
[f[wd =
_
[f[d < +
and consequently f L
1
C
(, T, ) and fw L
1
C
(, T, ). If f
is real-valued, Applying equation (7) once more to f
+
and f

,
we obtain:
_
f
+
d =
_
f
+
d +
_
f
+
wd (9)
www.probability.net
Solutions to Exercises 46
and:
_
f

d =
_
f

d +
_
f

wd (10)
Subtracting (10) to (9) (all terms being nite, w being non-
negative and f
+
w, f

w being nite), we see that equation (7)

is true for f, whenever f L
1
R
(, T, ). If f = u + iv where u
and v are elements of L
1
R
(, T, ), we conclude that equation (7)
is true for f by the linearity of the integral, and the fact that it
is true for u and v. This proves that equation (7) is in fact true
for all f L
1
C
(, T, ).
4. Let f L
2
C
(, T, ). From the Cauchy-Schwarz inequality (42):
_
[f[d
__
[f[
2
d
_
1
2
__
1
2
d
_
1
2
=
__
[f[
2
d
_
1
2
(())
1
2
In particular, being a nite measure,
_
[f[d < + and f is
www.probability.net
Solutions to Exercises 47
also an element of L
1
C
(, T, ).
1
Applying 3. to [f[, we have:
_
[f[d
_
[f[d +
_
[f[wd =
_
[f[d
It follows that:
_
[f[d
_
[f[d
__
[f[
2
d
_
1
2
(())
1
2
5. being a nite measure, from 4. the inequality
_
[f[
2
d < +
implies
_
[f[d < +. So L
2
C
(, T, ) L
1
C
(, T, ). Further-
more, given f L
2
C
(, T, ), from 4. and theorem (24):

_
fd

_
[f[d
_
()|f|
2
1
This shows that L
2
C
(, F, ) L
1
C
(, F, ) whenever is a nite measure.
We dont need |f| L
1
C
(, F, ) for equation (7) to be true (see proof of 3.)
www.probability.net
Solutions to Exercises 48
6. Consider the map : L
2
C
(, T, ) C dened by:
f L
2
C
(, T, ) , (f) =
_
fd
Since L
2
C
(, T, ) L
1
C
(, T, ) is well-dened, and it is
clearly linear. Furthermore from 5., [(f)[
_
()|f|
2
for
all f L
2
C
(, T, ). So is also continuous. Applying theo-
rem (55), there exists g

L
2
C
(, T, ) such that (f) =
_
f g

d
for all fs. Taking g = g

L
2
C
(, T, ), we obtain:
f L
2
C
(, T, ) ,
_
fd =
_
fgd
7. Let E T be such that (E) > 0. being a nite measure, the
map 1
E
is an element of L
2
C
(, T, ). From 6. we have:
_
E
gd =
_
1
E
gd =
_
1
E
d = (E)
www.probability.net
Solutions to Exercises 49
Furthermore, since 0 (E) (E) + (E) = (E), we obtain:
0
_
E
gd (E)
Finally since (E) > 0, we see that (E)
1
_
E
gd [0, 1].
8. Since is a nite measure, we have L
2
C
(, T, ) L
1
C
(, T, ),
as can be seen from the Cauchy-Schwarz inequality (42). In
particular, g is an element of L
1
C
(, T, ). Furthermore, the
interval [0, 1] is a closed subset of C, and for all E T with
(E) > 0, we have:
1
(E)
_
E
gd [0, 1]
Applying theorem (59), it follows that g [0, 1] -almost surely.
There exists N T with (N) = 0 such that g() [0, 1] for
all N
c
. Dene h = g1
N
c. Then h L
2
C
(, T, ) and
h() [0, 1] for all . Furthermore, for all f L
2
C
(, T, )
www.probability.net
Solutions to Exercises 50
we have from 6.:
_
fd =
_
fgd =
_
fg1
N
d +
_
fg1
N
c
d =
_
fhd
Renaming h by g, we have found g L
2
C
(, T, ) such that
g() [0, 1] for all and (1) still holds.
9. Let f L
2
C
(, T, ). Since g L
2
C
(, T, ), from the Cauchy-
Schwarz inequality (42):
_
[fg[d
__
[f[
2
d
_
1
2
__
[g[
2
d
_
1
2
< +
It follows that fg L
1
C
(, T, ). From 3. we have:
_
fgd =
_
fgd +
_
fgwd (11)
all three integrals being well-dened. From 6. we have:
_
fd =
_
fgd (12)
www.probability.net
Solutions to Exercises 51
From (11) and (12), using the linearity of the integral, we obtain:
_
f(1 g)d =
_
fgwd
10. Let n 1 and E T. Let f = (1 +g +. . . +g
n
)1
E
. Then f is
a measurable map and furthermore, since 0 g 1, f is also
bounded. being a nite measure on (, T), we conclude that
f L
2
C
(, T, ).
11. Let n 1 and E T. Let f = (1 + g +. . . + g
n
)1
E
. From 10.
f is an element of L
2
C
(, T, ). Applying 9. we obtain:
_
f(1 g)d =
_
fgwd
or equivalently:
_
E
(1 g
n+1
)d =
_
E
g(1 +g +. . . +g
n
)wd (13)
www.probability.net
Solutions to Exercises 52
12. Let A = 0 g < 1 and dene:
h

= gw
_
+

k=0
g
k
_
and h
n
= gw(

n
k=0
g
k
) for n 1. Then, for all E T,
(h
n
1
E
)
n1
is a non-decreasing sequence of non-negative and
measurable maps, converging simply to h1
E
. By the monotone
convergence theorem (19), we have
_
h
n
1
E
d
_
h1
E
d, i.e.
lim
n+
_
E
g(1 +g +. . . +g
n
)wd =
_
E
hd (14)
Furthermore for all A, (1 g
n+1
()) 1 as n +,
and if , A, since 0 g 1 we have 1 g
n+1
() = 0 for
all n 1. It follows that (1 g
n+1
)1
E
1
EA
, and being
a nite measure, the condition [(1 g
n+1
)1
E
[ 1 allows us to
www.probability.net
Solutions to Exercises 53
apply to dominated convergence theorem (23) to obtain:
lim
n+
_
E
(1 g
n+1
)d =
_
1
EA
d = (E A) (15)
Using (14) and (15), taking the limit in (13) as n +:
(E A) =
_
E
hd
13. Let with h() = + = g()w()

k=0
g
k
(). Since
0 g 1 and 0 < w 1, the series

+
k=0
g
k
() cannot be
convergent, and consequently g() = 1. So A
c
and we have
proved that h = + A
c
. Conversely, suppose that A
c
.
Since 0 g 1, we have g() = 1. Hence

+
k=0
g
k
() = +,
and since w() > 0 it follows that h() = +. This shows that
A
c
h = + and nally that h = + = A
c
. Applying
12. to E = A
c
, we obtain:
0 = (A
c
A) =
_
A
c
hd = (+)(A
c
)
www.probability.net
Solutions to Exercises 54
from which we conclude that (A
c
) = 0.
14. Let E T. From 12. we have:
(E A) =
_
E
hd
From 13. we have (A
c
) = 0. Since by assumption, is abso-
lutely continuous with respect to (i.e. << ), we also have
(A
c
) = 0. It follows that:
(E) = (E A) +(E A
c
) =
_
E
hd
15. Let be a -nite measure on (, T) and be a nite measure
on (, T) such that << . From 14. we have found a map
h : (, T) [0, +] non-negative and measurable, such that:
E T , (E) =
_
E
hd (16)
www.probability.net
Solutions to Exercises 55
Furthermore, from 13. we have (h = +) = 0. It follows
that property (16) will also hold, if we replace h by h1
{h<+}
.
Hence, without loss of generality, we can assume that h satisfy-
ing (16) has values in R
+
. Since is a nite measure, taking
E = in (16) we obtain:
_
[h[d =
_
hd = () < +
So h L
1
R
(, T, ). We have proved the existence of a map
h L
1
R
(, T, ) such that h 0 and property (16) holds.
16. Let be a -nite measure on (, T), and be a complex
measure on (, T) such that << . If is in fact a nite
measure, then 15. guarantees the existence of h L
1
C
(, T, )
such that:
E T , (E) =
_
E
hd (17)
In fact, the result in 15. is slightly stronger, and allows us to
choose h with values in R
+
. If is a signed measure (i.e. it
www.probability.net
Solutions to Exercises 56
has values in R), then it can be written as =
+

, where

+
and

are respectively the positive part and negative part

of . Since
+
and

are nite measures (see exercise (12) of

Tutorial 11), which are absolutely continuous with respect to
(see exercise (3)), there exist h
+
, h

elements of L
1
R
(, T, )
with values in R
+
, such that
+
=
_
h
+
d and

=
_
h

d.
Dening h = h
+
h

, we obtain an element of L
1
R
(, T, )
for which (by linearity of the integral) property (17) holds. In
the general case when is an arbitrary complex measure,
can be written as =
1
+ i
2
where
1
,
2
are two signed
measures which are absolutely continuous with respect to (see
exercise (3)). Hence, there exist h
1
, h
2
in L
1
R
(, T, ) such that

1
=
_
h
1
d and
2
=
_
h
2
d. Dening h = h
1
+ ih
2
, we
obtain an element of L
1
C
(, T, ) for which (by linearity of the
integral) property (17) holds. This proves the complex version
of the Radon-Nikodym theorem (60).
Exercise 6
www.probability.net
Solutions to Exercises 57
Exercise 7.
1. The positive part u
+
of u is dened as u
+
= max(0, u). It
follows that u
+
= u1
{u0}
and consequently:
_
u
+
d =
_
u1
{u0}
d =
_
{u0}
ud
2. By assumption, using E = u 0 T, we have:
_
{u0}
fd = 0 =
_
{u0}
ud +i
_
{u0}
vd
It follows in particular that
_
{u0}
ud = 0 and consequently
using 1.,
_
u
+
d = 0. Since u
+
is non-negative, this implies
that u
+
= 0 -a.s. (See Exercise (7) of Tutorial 5.). Similarly,
from u

= max(u, 0) = u1
{u0}
we obtain:
_
u

d =
_
u1
{u0}
d =
_
{u0}
ud
www.probability.net
Solutions to Exercises 58
and from:
_
{u0}
fd = 0 =
_
{u0}
ud +i
_
{u0}
vd
we see that
_
u

d = 0 and nally u

= 0 -a.s. An identical
proof will show that v
+
= 0 -a.s. and v

= 0 -a.s. Having
proved that u
+
, u

, v
+
and v

are all -almost surely equal

to zero, there exist sets N
1
, N
2
, N
3
and N
4
, elements of T,
with (N
1
) = (N
2
) = (N
3
) = (N
4
) = 0 and such that
u
+
() = u

() = v
+
() = v

() = 0 for all N
c
1
. . . N
c
4
.
Taking N = N
1
. . . N
4
, we have found N T with (N) = 0
such that f() = 0 for all N
c
. This shows that f = 0 -a.s.
3. Suppose there exist two maps h
1
, h
2
L
1
C
(, T, ) which satisfy
the conclusion of theorem (60), i.e. such that:
E T , (E) =
_
E
h
1
d =
_
E
h
2
d
www.probability.net
Solutions to Exercises 59
Dening f = h
1
h
2
L
1
C
(, T, ), we obtain:
E T ,
_
E
fd = 0
and from 2. we conclude that f = 0 -a.s., or equivalently
h
1
= h
2
-a.s. This shows that the Radon-Nikodym deriva-
tive of with respect to (i.e. the element h of L
1
C
(, T, )
which satises the conclusion of theorem (60)), is unique up to
-almost sure equality.
Exercise 7
www.probability.net
Solutions to Exercises 60
Exercise 8.
1. Let n 1. We have
n
() = (E
n
) = (E
n
) < +. So
n
is
a nite measure, and in particular a complex measure on (, T).
Furthermore, if E T is such that (E) = 0, then (E
n
E) = 0
and it follows from << that (E
n
E) = 0 i.e.
n
(E) = 0.
This shows that
n
<< , and the assumptions of theorem (60)
are therefore all satised. There exists h
n
L
1
C
(, T, ) such
that:
E T ,
n
(E) =
_
E
h
n
d
Furthermore,
n
being a nite measure, the map h
n
can be
chosen to lie in L
1
R
(, T, ), with h
n
0.
2. Let E T and n 1. By assumption, E
n
E
n+1
. Hence:
_
E
h
n
d = (E
n
E) (E
n+1
E) =
_
E
h
n+1
d
www.probability.net
Solutions to Exercises 61
3. Let n, p 1. Since h
n
, h
n+1
have values in R (in fact R
+
), the
dierence h
n
h
n+1
is meaningful, and from 2. we have:
_
E
(h
n
h
n+1
)d 0
Applying this inequality to E = h
n
h
n+1
> 1/p we obtain:
1
p
(h
n
h
n+1
>
1
p
)
_
E
(h
n
h
n+1
)d 0
from which we conclude that (h
n
h
n+1
> 1/p) = 0.
4. Let n 1. From:
h
n
> h
n+1
=
+
_
p=1
h
n
h
n+1
>
1
p

and the fact that (h

n
h
n+1
> 1/p) = 0 for all p 1, we
conclude that (h
n
> h
n+1
) = 0. So h
n
h
n+1
-.a.s.
www.probability.net
Solutions to Exercises 62
5. Given n 1, let N
n
= h
n
> h
n+1
. Dene N =
n1
N
n
.
Then, (N) = 0 and replacing all h
n
s by h
n
1
N
c, we obtain a
sequence (h
n
)
n1
in L
1
R
(, T, ) such that 0 h
n
h
n+1
(this
time everywhere), where the new h
n
s are -almost equal to our
original h

n
s, and therefore such that equation (2) still holds.
6. Let h = sup
n1
h
n
and E T. From (2), for all n 1 we have:
(E
n
E) =
_
1
E
h
n
d (18)
From (E
n
E) E and theorem (7), (E
n
E) (E) as
n +. From 1
E
h
n
1
E
h and the monotone convergence
theorem (19),
_
1
E
h
n
d
_
1
E
hd as n +. Taking the
limit in (18) as n +, we conclude that:
E T , (E) =
_
E
hd
www.probability.net
Solutions to Exercises 63
7. Let n 1. From 6. we have:
_
E
n
hd = (E
n
) < +
8. From (+)1
{h=+}
h and 7. we obtain:
(+)(E
n
h = +)
_
E
n
hd < +
It follows that (E
n
h = +) = 0 for all n 1. From
E
n
h = + h = + and theorem (7), we obtain:
(h = +) = lim
n+
(E
n
h = +) = 0
We conclude that h < + -a.s.
9. Replacing h by h1
{h<+}
, we obtain a measurable map with
values in R
+
, which is -almost surely equal to our original h,
and therefore such that equation (3) still holds.
www.probability.net
Solutions to Exercises 64
10. h has values in R
+
and is measurable, while from 7.:
_
hd
E
n
=
_
E
n
hd < +
So h L
1
R
(, T,
E
n
).
Exercise 8
www.probability.net
Solutions to Exercises 65
Exercise 9.
1. Let n 1 and E T. We have:
_
E
hd
F
n
=
_
F
n
E
hd =
_
F
n
E
h

d =
_
E
h

d
F
n
Furthermore:
_
E
hd
F
n
=
_
1
E
h1
E
n
1
{hn}
d n(E
n
) < +
2. Let n 1 and p 1. Applying 1. to E = h > h

+ 1/p:
_
E
h

d
F
n
=
_
E
hd
F
n

_
E
h

d
F
n
+
1
p
(F
n
E)
and since
_
E
h

d
F
n
< +, it follows that (F
n
E) = 0.
3. Let n 1. From the equality:
h > h

=
+
_
p=1
h > h

+
1
p

www.probability.net
Solutions to Exercises 66
and the fact that (F
n
h > h

+ 1/p) = 0 for all p 1,

we have (F
n
h > h

) = 0. A similar argument shows that

(F
n
h

> h) = 0. It follows that (F

n
h ,= h

) = 0.
4. By assumption, F
n
= E
n
h n. Hence, F
n
F
n+1
for all
n 1 and
n1
F
n
= h < +. In short, F
n
h < +,
and consequently we have F
n
h ,= h

h ,= h

h < +.
Applying theorem (7), we conclude that:
(h ,= h

h < +) = lim
n+
(F
n
h ,= h

) = 0
5. The assumption made on h and h

, namely:
E T ,
_
E
hd =
_
E
h

d
is symmetric in terms h and h

. Using 4. where h and h

have
been interchanged, we obtain (h ,= h

< +) = 0.
Since the set h ,= h

can be decomposed as:

h ,= h

= (h ,= h

h < +) (h ,= h

< +)
www.probability.net
Solutions to Exercises 67
we conclude that (h ,= h

) = 0, i.e. h = h

-a.s.
6. Let h and h

be two maps satisfying the conclusion of theo-

rem (61). Then in particular, h and h

are non-negative and

measurable, while satisfying:
E T ,
_
E
hd =
_
E
h

d
This exercise shows that h = h

-a.s. In other words, the

Radon Nikodym derivative of with respect to (i.e. the map
h which satises the conclusion of theorem (61)) is unique, up
to -almost sure equality.
Exercise 9
www.probability.net
Solutions to Exercises 68
Exercise 10. The sigma-algebra T has only two elements, and .
If E = , then:
_
E
hd = 0 =
_
E
h

d
If E = , then:
_
E
hd = 1 () = + = 2 () =
_
E
h

d
In any case, we have
_
E
hd =
_
E
h

d. Although h and h

are not
-almost surely equal, this does not contradict exercise (9), as the
measure is not sigma-nite.
Exercise 10
www.probability.net
Solutions to Exercises 69
Exercise 11.
1. Let E T be such that [[(E) = 0. Since [(E)[ [[(E) we
have (E) = 0, and consequently << [[.
2. From theorem (57), the total variation [[ of is a nite measure
on (, T). In particular, it is sigma-nite. being a complex
measure such that << [[, we can apply theorem (60): there
exists h L
1
C
(, T, [[) such that:
E T , (E) =
_
E
hd[[
3. If is in fact a signed measure, then from theorem (60), h can
indeed be chosen to lie in L
1
R
(, T, [[).
Exercise 11
www.probability.net
Solutions to Exercises 70
Exercise 12.
1. Let A
r
= [h[ < r (for some r > 0) and (E
n
)
n1
be a measur-
able partition of A
r
. From exercise (11), for all n 1:
[(E
n
)[ =

_
E
n
hd[[

_
E
n
[h[d[[ r[[(E
n
)
where the rst inequality stems from theorem (24), and the
second from the fact that E
n
[h[ < r. It follows that:
+

n=1
[(E
n
)[ r
+

n=1
[[(En) = r[[(A
r
)
2. [[(A
r
) being the least upper bound of all sums

+
n=1
[(E
n
)[ as
(E
n
)
n1
ranges across all measurable partitions of A
r
, it follows
from 1. that [[(A
r
) r[[(A
r
). When 0 < r < 1, this can only
occur if [[(A
r
) = 0.
www.probability.net
Solutions to Exercises 71
3. From the equality:
[h[ < 1 =
+
_
p=2
[h[ < 1
1
p

and the fact that [[([h[ < 1 1/p) = [[(A

11/p
) = 0 for all
p 2, it follows that [[([h[ < 1) = 0, i.e. [h[ 1 [[-a.s.
4. Let E T be such that [[(E) > 0. We have:

1
[[(E)
_
E
hd[[

1
[[(E)
(E)

=
[(E)[
[[(E)
1
5. Applying theorem (59) to the closed disc S = [z[ 1 and the
nite measure [[, we conclude from 4. that h S [[-a.s. or
equivalently that [h[ 1 [[-a.s.
6. Having proved that [h[ 1 [[-a.s. and [h[ 1 [[-a.s., the set
N = [h[ > 1 [h[ < 1 is such that [[(N) = 0. Replacing h
by h1
N
c + 1
N
, we obtain an element of L
1
C
(, T, [[) such that
www.probability.net
Solutions to Exercises 72
[h[ = 1 (this time everywhere), which is almost surely equal to
our original h, and therefore such that:
E T , (E) =
_
E
hd[[
From 3. of exercise (11), if is a signed measure, then h can be
chosen to lie in L
1
R
(, T, [[). This proves theorem (62).
Exercise 12
www.probability.net
Solutions to Exercises 73
Exercise 13.
1. Suppose A
n
A. Then A
n
A
n+1
for all n 1, and further-
more A =
n1
A
n
. Let and n 1. If 1
A
n
() = 0,
then 1
A
n
() 1
A
n+1
() is clear. If 1
A
n
() = 1, then A
n
and consequently A
n+1
, so 1
A
n+1
() = 1. In any case, we
have 1
A
n
() 1
A
n+1
(). This shows that 1
A
n
1
A
n+1
for all
n 1. Since A
n
A for all n 1, we obtain similarly that
1
A
n
1
A
for all n 1, and consequently sup
n1
1
A
n
1
A
. Let
. If 1
A
() = 0, then 1
A
() sup
n1
1
A
n
() is clear. If
1
A
() = 1 then A =
n1
A
n
, and there exists n 1 such
that A
n
. So 1
A
n
() = 1 sup
n1
1
A
n
(). In any case, we
have 1
A
() sup
n1
1
A
n
(). This shows that:
1
A
= sup
n1
1
A
n
= lim
n+
1
A
n
and nally, we have proved that 1
A
n
1
A
.
2. Suppose that A
n
A. Then A
n+1
A
n
for all n 1 and
A =
n1
A
n
. It follows that A
c
n
A
c
n+1
for all n 1 and
www.probability.net
Solutions to Exercises 74
A
c
=
n1
A
c
n
, or equivalently that A
c
n
A
c
. Applying 1. we
obtain that 1
A
c
n
1
A
c. Since 1
A
c
n
= 1 1
A
n
for all n 1 and
1
A
c = 1 1
A
, we conclude that 1
A
n+1
1
A
n
for all n 1 and
1
A
= lim
n
1
A
n
. We have proved that 1
A
n
1
A
.
3. Suppose that 1
A
n
1
A
and let M
1
(, T). From theo-
rem (62), there exists h L
1
C
(, T, [[) such that:
E T , (E) =
_
E
hd[[
In particular, (A
n
) =
_
1
A
n
hd[[ for all n 1. The hypoth-
esis 1
A
n
1
A
implies in particular that 1
An
h 1
A
h, and
since [1
A
n
h[ [h[ L
1
R
(, T, [[), the dominated convergence
theorem (23) allows us to conclude that:
lim
n+
(A
n
) = lim
n+
_
1
A
n
hd[[ =
_
1
A
hd[[ = (A)
Exercise 13
www.probability.net
Solutions to Exercises 75
Exercise 14.
1. Let f L
1
C
(, T, ) and : T C be dened by:
E T , (E) =
_
E
fd
The fact that M
1
(, T) has already been proved in ex. (3)
of Tutorial 11. For a slightly leaner proof, here is the following:
let E T and (E
n
)
n1
be a measurable partition of E. For
all n 1, we dene A
n
= E
1
. . . E
n
. Then, from 1
A
n
=
1
E
1
+. . . + 1
E
n
we obtain:
(A
n
) =
_
1
A
n
fd =
n

k=1
_
1
E
k
fd =
n

k=1
(E
k
) (19)
Furthermore, from A
n
E we have 1
A
n
1
E
and consequently
1
A
n
f 1
E
f. Since [1
A
n
f[ [f[ L
1
R
(, T, ) for all n 1, it
www.probability.net
Solutions to Exercises 76
follows from the dominated convergence theorem (23) that:
lim
n+
(A
n
) = lim
n+
_
1
A
n
fd =
_
1
E
fd = (E) (20)
Comparing (19) with (20), it appears that the series

+
k=1
(E
k
)
converges to (E). So is indeed a complex measure on (, T).
2. From theorem (62), there is h L
1
C
(, T, [[) with [h[ = 1 and:
E T , (E) =
_
E
hd[[
Let E, F T. We have:
_
E
f1
F
d =
_
EF
fd = (E F) =
_
E
h1
F
d[[
3. Given g : C bounded and measurable, we claim that:
E T ,
_
E
fgd =
_
E
hgd[[ (21)
www.probability.net
Solutions to Exercises 77
From 2., equation (21) is true whenever g is of the form g = 1
F
with F T. By the linearity of the integral, (21) is also true
whenever g is a simple function on (, T). Suppose g is non-
negative and measurable, while being bounded. From theo-
rem (18), there exists a sequence (s
n
)
n1
of simple functions
on (, T), such that s
n
g. Having proved (21) for simple
functions, for all n 1 we have:
_
1
E
fs
n
d =
_
1
E
hs
n
d[[ (22)
From s
n
g we obtain 1
E
fs
n
1
E
fg and 1
E
hs
n
1
E
hg.
Since [1
E
fs
n
[ [f[g L
1
R
(, T, ) (since g is bounded) and
[1
E
hs
n
[ [h[g L
1
R
(, T, [[), it follows from the dominated
convergence theorem (23) that
_
1
E
fs
n
d
_
1
E
fgd and
_
1
E
hs
n
d[[
_
1
E
hgd[[ as n +. Taking the limit in (22)
as n +, we see that (21) is true whenever g is non-negative
and measurable, while being bounded. If g is now an arbitrary
C-valued map which is measurable and bounded, then it can be
www.probability.net
Solutions to Exercises 78
expressed as g = g
1
g
2
+i(g
3
g
4
) where each g
i
is non-negative,
measurable and bounded. From the linearity of the integral, we
conclude that (21) is also true for g, which completes the proof
of our initial claim.
4. Since [h[ = 1, applying (21) to g =

h, we obtain for all E T:
_
E
f

hd =
_
E
h

hd[[ =
_
E
d[[ = [[(E)
5. The total variation [[ of the complex measure being a nite
measure on (, T) (theorem (57)), it has values in R
+
. Hence:
_
E
Re(f

h)d = Re
__
E
f

hd
_
= Re([[(E)) 0
and:
_
E
Im(f

h)d = Im
__
E
f

hd
_
= Im([[(E)) = 0
www.probability.net
Solutions to Exercises 79
6. Dene g
1
= Re(f

h) and g
2
= Im(f

h). Then g
1
and g
2
are
elements of L
1
R
(, T, ), and from 5. we have
_
E
g
1
d 0 while
_
E
g
2
d = 0 for all E T. Since S = R
+
and S = 0 are
closed subset of C, it is very tempting to apply theorem (59) in
an attempt to conclude that g
1
R
+
-a.s. and g
2
= 0 -a.s.
Unfortunately, is not assumed to be a nite measure (it is
not even assumed to be sigma-nite) and theorem (59) should
therefore be forgotten here. Taking E = g
1
< 1/n for some
n 1, we obtain:
0
_
E
g
1
d
1
n
(g
1
< 1/n) 0
from which we see that (g
1
< 1/n) = 0 for all n 1. Since
g
1
< 0 =
n1
g
1
< 1/n, it follows that (g
1
< 0) = 0
and consequently, g
1
R
+
-a.s. Similarly, from
_
E
g
2
d = 0
for all E T, we obtain g
2
R
+
-a.s. and g
2
R
+
-a.s.
It follows that g
2
= 0 -a.s. We have proved that Re(f

h) R
+
-a.s. while Im(f

h) = 0 -a.s., so f

h R
+
-a.s.
www.probability.net
Solutions to Exercises 80
7. From 6. there exists N T with (N) = 0 and f()

h() R
+
for all N
c
. In particular, since [h[ = 1, for all N
c
:
f()

h() = [f()

h()[ = [f()[
It follows that f

h = [f[ -a.s.
8. Let be a measure on (, T) and f L
1
C
(, T, ). Then, from
part 1. of this exercise, =
_
fd is a complex measure on
(, T). Furthermore, from 4. we have:
E T , [[(E) =
_
E
f

hd
Finally, from 7. we have f

h = [f[ -a.s. We conclude that:

E T , [[(E) =
_
E
[f[d
This completes the proof of theorem (63).
Exercise 14
www.probability.net
Solutions to Exercises 81
Exercise 15.
1. The positive part
+
of the signed measure is dened by the
formula
+
= ([[ + )/2 (see exercise (12) of Tutorial 11). It
follows that for all E T:

+
(E) =
1
2
[[(E) +
1
2
_
E
hd[[ =
_
E
1
2
(1 +h)d[[
2. The negative part

of the signed measure is dened as

= ([[ )/2. Hence, for all E T:

(E) =
1
2
[[(E)
1
2
_
E
hd[[ =
_
E
1
2
(1 h)d[[
3. Since h L
1
R
(, T, [[) is R-valued and [h[ = 1, h can only
assume the values 1 and 1. Having dened A = h = 1,
(1 +h)/2 = 0 on A
c
and for all E T we have:

+
(E) =
_
E
1
2
(1 +h)d[[ =
_
AE
1
2
(1 +h)d[[ (23)
www.probability.net
Solutions to Exercises 82
Furthermore, since h = (1 +h)/2 on A:
(A E) =
_
AE
hd[[ =
_
AE
1
2
(1 +h)d[[ (24)
Comparing (23) with (24), we obtain
+
=
A
.
4. Having dened B = h = 1, we have for all E T:

(E) =
_
E
1
2
(1 h)d[[ =
_
BE
1
2
(1 h)d[[
since (1 h)/2 = 0 on B
c
. Furthermore:
(B E) =
_
BE
hd[[ =
_
BE
1
2
(1 h)d[[
since h = (1 h)/2 on B. This shows that

=
B
, and
completes the proof of theorem (64).
Exercise 15
www.probability.net
Solutions to Exercises 83
Exercise 16.
1. Let f : (, T)(C, B(C)) be measurable. From denition (97),
any element of L
1
C
(, T, ) is an element of L
1
C
(, T, [[), and
therefore satises:
_
[f[d[[ < + (25)
Conversely, if f satises the integrability condition (25), then
it is an element of L
1
C
(, T, [[) and therefore an element of
L
1
C
(, T, ).
2. Let f L
1
C
(, T, ). The integral of f w.r. to is dened as:
_
fd

=
_
fhd[[ (26)
where h is any element of L
1
C
(, T, [[) with [h[ = 1 and =
_
hd[[ (there is at least one such h by virtue of theorem (62)).
This denition is potentially ambiguous, as h may not be unique.
www.probability.net
Solutions to Exercises 84
However, if h

is another element of L
1
C
(, T, [[) with [h

[ = 1
and =
_
h

d[[, then for all E T, we have:

(E) =
_
E
hd[[ =
_
E
h

d[[
which implies that
_
E
(h h

)d[[ = 0. Using exercise (7), it

follows that h = h

[[-a.s. and consequently the r.h.s integral

of equation (26) is unchanged, when replacing h by h

. We
conclude that equation (26) is in fact unambiguous, as its r.h.s
integral does not depend on the particular choice of element
h L
1
C
(, T, [[) with [h[ = 1 and =
_
hd[[.
3. Let E T. Then 1
E
: (, T) (C, B(C)) is measurable, and
furthermore:
_
[1
E
[d[[ =
_
1
E
d[[ = [[(E) < +
since [[ is a nite measure on (, T) (see theorem (57)). Using
1. it follows that 1
E
is an element of L
1
C
(, T, ), as dened in
www.probability.net
Solutions to Exercises 85
denition (97). Moreover, we have:
_
1
E
d =
_
1
E
hd[[ =
_
E
hd[[ = (E)
4. If is a nite measure (complex measure with values in R
+
),
then for all E T and (E
n
)
n1
measurable partition of E:
+

n=1
[(E
n
)[ =
+

n=1
(E
n
) = (E)
In particular, (E) is an upper bound of all sums

+
n=1
[(E
n
)[,
as (E
n
)
n1
ranges through all measurable partitions of E. It
follows that [[(E) (E). Since (E) = [(E)[ [[(E) is
clear, we conclude that [[ = .
5. Suppose that is a nite measure. Then is not only a measure,
but also a complex measure. It follows that denition (97) of the
space L
1
C
(, T, ), and of the integral
_
fd (valid for complex
www.probability.net
Solutions to Exercises 86
measures), is potentially in conict with the denitions already
known for measures (denitions (46) and (48)). However, since
= [[, the space L
1
C
(, T, ) of denition (97) being dened
as L
1
C
(, T, [[), coincide with that of denition (46). Further-
more, h = 1 being an element of L
1
C
(, T, [[) with [h[ = 1
and =
_
hd[[, the integral
_
fd of denition (97) can be
expressed as:
_
fd =
_
fhd[[ =
_
fd[[ =
_
fd
where the r.h.s integral is that of denition (48). We conclude
that denition (97) which extends the notion of integral with
respect to complex measures, is consistent with previous deni-
tions laid out for measures.
6. The space L
1
C
(, T, ) being dened as L
1
C
(, T, [[), it is a
C-vector space. Let h L
1
C
(, T, [[) be such that [h[ = 1
and =
_
hd[[. Then, for all f, g L
1
C
(, T, ) and C,
www.probability.net
Solutions to Exercises 87
following denition (97) we have:
_
(f +g)d =
_
(f +g)hd[[
=
_
fhd[[ +
_
ghd[[
=
_
fd +
_
gd
where the second equality stems from the linearity of the inte-
gral, already established for measures.
7. Let f L
1
C
(, T, ) and h be as in denition (97). Then, from
theorem (24), we have:

_
fd

_
fhd[[

_
[fh[d[[ =
_
[f[d[[
Exercise 16
www.probability.net
Solutions to Exercises 88
Exercise 17.
1. Let E T and (E
n
)
n1
be a measurable partition of E. Then:
+

n=1
[(E
n
)[ = [[
+

n=1
[(E
n
)[ [[[[(E)
It follows that [[[[(E) is an upper bound of all

+
n=1
[(E
n
)[
as (E
n
)
n1
ranges through all measurable partitions of E. Since
[[(E) is the smallest of such upper bounds, we obtain the
inequality [[(E) [[[[(E). This being true for all E T,
we have proved that [[ [[[[ for all C. If = 0, then
[[ = [[[[ is clear. If ,= 0, then applying what we have just
proved to

= and

= 1/, we obtain:
[[ =

()

1
[[
[[
and consequently [[[[ [[. This shows that [[ = [[[[
for all M
1
(, T) and C.
www.probability.net
Solutions to Exercises 89
2. Let E T and (E
n
)
n1
be a measurable partition of E. Then:
+

n=1
[( +)(E
n
)[
+

n=1
[(E
n
)[ +
+

n=1
[(E
n
)[ ([[ +[[)(E)
It follows that ([[ + [[)(E) is an upper bound of all sums

+
n=1
[( + )(E
n
)[ as (E
n
)
n1
ranges through all measurable
partitions of E. [ + [(E) being the smallest of such upper
bounds, we have [+[(E) ([[ +[[)(E). This being true for
all E T, we have proved that [ +[ [[ +[[.
3. Let f L
1
C
(, T, )L
1
C
(, T, ). Then f is C-valued, measur-
able, and satises
_
[f[d[[ < + with
_
[f[d[[ < +. Using
2. and 1., for all C:
[ +[ [[ +[[ = [[ +[[[[
Hence, for all E T, we have:
_
1
E
d[ +[
_
1
E
d[[ +[[
_
1
E
d[[
www.probability.net
Solutions to Exercises 90
By linearity, if s is a simple function on (, T), we obtain:
_
sd[ +[
_
sd[[ +[[
_
sd[[
Approximating [f[ by a sequence simple functions (see theo-
rem (18)) and using the monotone convergence theorem (19):
_
[f[d[ +[
_
[f[d[[ +[[
_
[f[d[[ < +
So f L
1
C
(, T, +), and we have proved the inclusion:
L
1
C
(, T, ) L
1
C
(, T, ) L
1
C
(, T, +)
4. Using 3. of exercise (16), we have:
_
1
E
d( +) = ( +)(E)
= (E) +(E)
=
_
1
E
d +
_
1
E
d
www.probability.net
Solutions to Exercises 91
5. Let f L
1
C
(, T, ) L
1
C
(, T, ). We claim that:
_
fd( +) =
_
fd +
_
fd (27)
Note from 3. that f L
1
C
(, T, + ) and all integrals of
equation (27) are therefore well dened. Furthermore from 4.,
(27) is true whenever f is of the form f = 1
E
with E T.
By linearity (proved in 6. of exercise (16)), equation (27) is in
fact true whenever f is a simple function on (, T). Suppose
now that f : (, T) [0, +] is non-negative and measurable,
while being an element of L
1
C
(, T, ) L
1
C
(, T, ). From the-
orem (18), there exists a sequence (s
n
)
n1
of simple functions
on (, T) such that s
n
f. Let h L
1
C
(, T, [ +[) be such
that [h[ = 1 and + =
_
hd[ + [. Then, s
n
h fh and
furthermore [s
n
h[ = [s
n
[ = s
n
f L
1
C
(, T, [ + [). From
the dominated convergence theorem (23), we have:
lim
n+
_
s
n
d( +) = lim
n+
_
s
n
hd[ +[
www.probability.net
Solutions to Exercises 92
=
_
fhd[ +[
=
_
fd( +)
We show similarly that:
lim
n+
_
s
n
d =
_
fd
and:
lim
n+
_
s
n
d =
_
fd
Having proved (27) for all simple functions on (, T), we have:
_
s
n
d( +) =
_
s
n
d +
_
s
n
d
and taking the limit as n +, we see that (27) is also true for f
non-negative, measurable and belonging to L
1
C
(, T, )L
1
C
(, T, ).
If f is an arbitrary element of L
1
C
(, T, ) L
1
C
(, T, ), then it
www.probability.net
Solutions to Exercises 93
can be expressed as f = f
1
f
2
+ i(f
3
f
4
) where each f
i
is non-
negative, measurable and belonging to L
1
C
(, T, ) L
1
C
(, T, ).
Equation (27) being true for each f
i
, it follows by linearity that (27)
is also true for f. We have proved that (27) is true for all elements f
of L
1
C
(, T, ) L
1
C
(, T, ).
Exercise 17
www.probability.net
Solutions to Exercises 94
Exercise 18.
1. Let , be two measures on (, T) and [0, +]. Then:
( +)() = () +() = 0 (28)
Note that from the convention (+) 0 = 0, equation (28) is
still true in the case when = +. Furthermore, if A T and
(A
n
)
n1
is a sequence of pairwise disjoint elements of T with
A =
n1
A
n
, then:
( +)(A) = (A) +(A)
=
+

n=1
(A
n
) +
+

n=1
(A
n
)
=
+

n=1
(A
n
) +
+

n=1
(A
n
)
=
+

n=1
(A
n
) +(A
n
)
www.probability.net
Solutions to Exercises 95
=
+

n=1
( +)(A
n
)
Note that the third equality is still true if = + or (A) =

n1
(A
n
) = +. It follows that + is countably additive,
and we have proved that it is indeed a measure on (, T). Now,
given f : (, T) [0, +], we claim that:
_
fd( +) =
_
fd +
_
fd (29)
(29) is obviously true when f is of the form f = 1
E
with E
T. By linearity (which is still valid, even if = +), (29)
is also true when f is a simple function on (, T). If f is an
arbitrary non-negative and measurable map, from theorem (18)
there exists a sequence (s
n
)
n1
of simple functions on (, T),
such that s
n
f. Having proved (29) for any simple function,
www.probability.net
Solutions to Exercises 96
for all n 1 we have:
_
s
n
d( +) =
_
s
n
d +
_
s
n
d (30)
From the monotone convergence theorem (19), taking the limit
in (30) as n +, we conclude that (29) is also true for f.
Note that if = + and (u
n
)
n1
is a sequence in [0, +]
converging to some u [0, +], then it is not true in general
that u
n
u. Indeed, consider the case when u
n
= 1/n.
Then u = (+) 0 = 0 while u
n
= (+) (1/n) = +
for all n 1, and (u
n
)
n1
does not converge to u. However,
if u
n
u
n+1
for all n 1, then the convergence u
n
u is
true. Indeed, if u = sup
n1
u
n
= 0, then u
n
= 0 for all n 1
and consequently u
n
= 0 = u. If u ,= 0, then u
n
,= 0 for n
large enough, and consequently u
n
= + = u for n large
enough. All this to say that even in the case when = +,
the convergence
_
s
n
d
_
fd is true.
www.probability.net
Solutions to Exercises 97
2. We claim that:
_
fd
_
fd (31)
Since , (31) is true when f = 1
E
, and E T. By linearity,
(31) is also true when f is a simple function on (, T). If f is an
arbitrary non-negative and measurable map, from theorem (18)
there exists a sequence (s
n
)
n1
of simple functions on (, T),
such that s
n
f. Having proved (31) for any simple function,
for all n 1 we have:
_
s
n
d
_
s
n
d (32)
From the monotone convergence theorem (19), taking the limit
in (32) as n +, we conclude that (31) is also true for f.
Exercise 18
www.probability.net
Solutions to Exercises 98
Exercise 19.
1. Since
1
= Re(), for all F T we have [
1
(F)[ [(F)[.
Hence, if E T and (E
n
)
n1
is a measurable partition of E:
+

n=1
[
1
(E
n
)[
+

n=1
[(E
n
)[ [[(E)
It follows that [[(E) is an upper bound of all

+
n=1
[
1
(E
n
)[,
as (E
n
)
n1
ranges through all measurable partitions of E. Since
[
1
[(E) is the smallest of such upper bounds, [
1
[(E) [[(E).
This being true for all E T, we conclude that [
1
[ [[. We
show similarly that [
2
[ [[.
2. Let E T and (E
n
)
n1
be a measurable partition of E:
+

n=1
[(E
n
)[
+

n=1
[
1
(E
n
)[ +[
2
(E
n
)[ [
1
[(E) +[
2
[(E)
www.probability.net
Solutions to Exercises 99
[[(E) being the supremum of all sums involved on the l.h.s of
this inequality, we conclude that [[(E) [
1
[(E) +[
2
[(E) for
all E T, i.e. that [[ [
1
[ +[
2
[.
3. Let f : (, T) (C, B(C)) be a measurable map. Proving:
L
1
C
(, T, ) = L
1
C
(, T,
1
) L
1
C
(, T,
2
)
amounts to showing the equivalence:
_
[f[d[[ < +
_
[f[d[
1
[ < +,
_
[f[d[
2
[ < + (33)
From [
1
[ [[ and [
2
[ [[ using exercise (18) we obtain:
_
[f[d[
1
[
_
[f[d[[ (34)
and:
_
[f[d[
2
[
_
[f[d[[ (35)
www.probability.net
Solutions to Exercises 100
Furthermore, from [[ [
1
[ +[
2
[ and exercise (18):
_
[f[d[[
_
[f[d([
1
[ +[
2
[) =
_
[f[d[
1
[ +
_
[f[d[
2
[ (36)
The equivalence (33) follows easily from (34), (35) and (36).
4. Let f : (, T) (C, B(C)) be a measurable map. Proving:
L
1
C
(, T,
1
) = L
1
C
(, T,
+
1
) L
1
C
(, T,

1
)
amounts to showing the equivalence:
_
[f[d[
1
[ < +
_
[f[d
+
1
< +,
_
[f[d

1
< + (37)
The positive and negative parts
+
1
and

1
of
1
being dened
as
+
1
= ([
1
[
1
)/2 and

1
= ([
1
[
1
)/2 (see exercise (12)
of Tutorial 11), we have [
1
[ =
+
1
+

1
. Using exercise (18):
_
[f[d[
1
[ =
_
[f[d
+
1
+
_
[f[d

1
www.probability.net
Solutions to Exercises 101
Hence, the equivalence (37) is clear. We show similarly that:
L
1
C
(, T,
2
) = L
1
C
(, T,
+
2
) L
1
C
(, T,

2
)
5. Let f L
1
C
(, T, ). We claim that
_
fd =
_
fd
+
1

_
fd

1
+i
__
fd
+
2

_
fd

2
_
(38)
Note that from 3. and 4. we have:
f L
1
C
(, T,
+
1
)L
1
C
(, T,

1
)L
1
C
(, T,
+
2
)L
1
C
(, T,

2
)
and consequently all integrals in (38) are well-dened. Applying
exercise (17) to the complex measures (in fact signed measures)

1
,
2
and = i, we obtain:
_
fd =
_
fd
1
+i
_
fd
2
(39)
Applying exercise (17) to the complex measures (in fact nite
www.probability.net
Solutions to Exercises 102
measures)
+
1
,

1
and = 1, we obtain:
_
fd
1
=
_
fd
+
1

_
fd

1
(40)
Similarly, we have:
_
fd
2
=
_
fd
+
2

_
fd

2
(41)
Equation (38) follows from (39), (40) and (41).
Exercise 19
www.probability.net
Solutions to Exercises 103
Exercise 20.
1. By denition, the trace of T on A is given by:
T
|A

= A E : E T
Since A is an element of T, it is clear that T
|A
T

, where:
T

= E : E T, E A
For the reverse inclusion, note that if E T

then E can be
written as E = A E and E T. So E is an element of T
|A
.
2. Let E T and (E
n
)
n1
be a measurable partition of E. Then
(A E
n
)
n1
is a measurable partition of A E. Since is a
complex measure of (, T), we have:
(A E) =
+

n=1
(A E
n
) (42)
www.probability.net
Solutions to Exercises 104
i.e. the right-hand-side series converges to (A E). By the
very denition of
A
, (42) can be re-expressed as:

A
(E) =
+

n=1

A
(E
n
) (43)
i.e. the right-hand-side series converges to
A
(E). This shows
that
A
is a complex measure on (, T).
Let E T
|A
and (E
n
)
n1
be a measurable partition of E, i.e. a
sequence of pairwise disjoint elements of T
|A
with E =
+
n=1
E
n
.
From 1., E and every E
n
is an element of T, (while being a
subset of A). being a complex measure on (, T), we have:
(E) =
+

n=1
(E
n
) (44)
i.e. the right-hand-side series converges to (E). Since
|A
is
dened as the restriction of to T
|A
, and since E and all E
n
s
www.probability.net
Solutions to Exercises 105
are elements of T
|A
, (45) can be equivalently expressed as:

|A
(E) =
+

n=1

|A
(E
n
) (45)
i.e. the right-hand-side series converges to
|A
(E). This shows
that
|A
is a complex measure on (A, T
|A
).
3. Let E T and (E
n
)
n1
be a measurable partition of E. Then
(A E
n
)
n1
is a measurable partition of A E. Hence:
+

n=1
[(A E
n
)[ [[(A E)
or equivalently:
+

n=1
[
A
(E
n
)[ [[
A
(E)
www.probability.net
Solutions to Exercises 106
4. From the previous section 3., [[
A
(E) is an upper bound of all
sums

+
n=1
[
A
(E
n
)[, as (E
n
)
n1
ranges through all measurable
partitions of E. Since [
A
[(E) is the smallest of such upper
bounds, we have [
A
[(E) [[
A
(E). This being true for all
E T, we conclude that [
A
[ [[
A
.
5. Let E T and (E
n
)
n1
be a measurable partition of AE. For
all n 1, E
n
A and consequently (E
n
) =
A
(E
n
). Hence:
+

n=1
[(E
n
)[ =
+

n=1
[
A
(E
n
)[ [
A
[(A E)
6. Let (E
n
)
n1
be a measurable partition of A
c
. Then:
+

n=1
[
A
(E
n
)[ =
+

n=1
[(A E
n
)[ = 0
[
A
[(A
c
) being the supremum of all sums

+
n=1
[
A
(E
n
)[, as
www.probability.net
Solutions to Exercises 107
(E
n
)
n1
ranges through all measurable partitions of A
c
, we con-
clude that [
A
[(A
c
) = 0.
7. From 5. it follows that [
A
[(A E) is an upper bound of all
sums

+
n=1
[(E
n
)[, as (E
n
)
n1
ranges through all measurable
partitions of AE. [[(AE) being the smallest of such upper
bounds, we have [[(A E) [
A
[(A E). However, from 6.
we have [
A
[(A
c
) = 0, and consequently:
[
A
[(E) = [
A
[(A E) +[
A
[(A
c
E) = [
A
[(A E)
It follows that [[(A E) [
A
[(E). This being true for all
E T, we see that [[
A
[
A
[. Having proved in 4. that
[
A
[ [[
A
, we conclude that [
A
[ = [[
A
. In other words,
the total variation of the restriction of to A, is equal to the
restriction of the total variation of to A.
8. Let E T
|A
and (E
n
)
n1
be an T
|A
-measurable partition of E.
Since T
|A
T, E T and (E
n
)
n1
is also an T-measurable
www.probability.net
Solutions to Exercises 108
partition of E. Hence:
+

n=1
[(E
n
)[ [[(E) (46)

|A
and [[
|A
being respectively the restrictions of and [[ to
T
|A
, (46) can be re-expressed as:
+

n=1
[
|A
(E
n
)[ [[
|A
(E)
9. Given E T
|A
, it appears from 8. that [[
|A
(E) is an upper
bound of all sums

+
n=1
[
|A
(E
n
)[, as (E
n
)
n1
ranges through
all T
|A
-measurable partitions of E. Since [
|A
[(E) is the small-
est of such upper bounds, we have [
|A
[(E) [[
|A
(E). This
being true for all E T
|A
, we conclude that [
|A
[ [[
|A
.
10. Let E T
|A
and (E
n
)
n1
be an T-measurable partition of E.
From 1. we have E T and E A. It follows that E
n
A
www.probability.net
Solutions to Exercises 109
for all n 1 and consequently E
n
T
|A
. So (E
n
)
n1
is also an
T
|A
-measurable partition of E. Hence:
+

n=1
[
|A
(E
n
)[ [
|A
[(E)
which can be equivalently written as:
+

n=1
[(E
n
)[ [
|A
[(E)
11. Given E T
|A
, it appears from 10. that [
|A
[(E) is an upper
bound of all sums

+
n=1
[(E
n
)[, as (E
n
)
n1
ranges through all
T-measurable partitions of E. Since [[(E) is the smallest of
such upper bounds, we have [[(E) [
|A
[(E), or equivalently
since E T
|A
, [[
|A
(E) [
|A
[(E). This being true for all
E T
|A
, [[
|A
[
|A
[. Having proved in 9. that [
|A
[ [[
|A
,
we conclude that [
|A
[ = [[
|A
.
www.probability.net
Solutions to Exercises 110
12. By assumption, h L
1
C
(, T, [[) is such that [h[ = 1 and
=
_
hd[[. In particular, for all E T:

A
(E) = (A E)
=
_
AE
hd[[
=
_
(h1
E
)1
A
d[[
=
_
(h1
E
)d[[
A
=
_
E
hd[[
A
=
_
E
hd[
A
[
where the rst equality stems from the denition of
A
, the
second from the fact that =
_
hd[[, the third, fourth and
fth from a use of denition (49) and nally the sixth from the
www.probability.net
Solutions to Exercises 111
fact that [[
A
= [
A
[. This being true for all E T, we have
proved that
A
=
_
hd[
A
[.
13. Since h L
1
C
(, T, [[), from denition (49), h
|A
is an element
of L
1
C
(A, T
|A
, [[
|A
). Having proved that [[
|A
= [
|A
[, it follows
that h
|A
L
1
C
(A, T
|A
, [
|A
[)
2
. Furthermore, for all E T
|A
:

|A
(E) = (E)
= (A E)
=
_
AE
hd[[
=
_
(h1
E
)1
A
d[[
=
_
h
|A
(1
E
)
|A
d[[
|A
=
_
E
h
|A
d[[
|A
2
One may argue that |h
|A
| = 1 and |
|A
| is a nite measure. . .
www.probability.net
Solutions to Exercises 112
=
_
E
h
|A
d[
|A
[
where the rst equality stems from the denition of
|A
, the
second from the fact that E A, the third from the fact that
=
_
hd[[, the fourth, fth and sixth from denition (49) an
nally the seventh from the fact that [[
|A
= [
|A
[. This being
true for all E T
|A
, we conclude that
|A
=
_
h
|A
d[
|A
[.
14. Let f L
1
C
(, T, ). From denition (97), this is equivalent to
f L
1
C
(, T, [[). Applying denition (49), we have:
f1
A
L
1
C
(, T, [[), f L
1
C
(, T, [[
A
), f
|A
L
1
C
(A, T
|A
, [[
|A
)
and since [[
A
= [
A
[ and [[
|A
= [
|A
[, we obtain:
f1
A
L
1
C
(, T, [[), f L
1
C
(, T, [
A
[), f
|A
L
1
C
(A, T
|A
, [
|A
[)
Moreover, since [h[ = 1 and =
_
hd[[, from denition (97):
_
f1
A
d =
_
fh1
A
d[[ (47)
www.probability.net
Solutions to Exercises 113
and similarly, since
A
=
_
hd[
A
[ and [[
A
= [
A
[:
_
fd
A
=
_
fhd[
A
[ =
_
fhd[[
A
(48)
Furthermore since
|A
=
_
h
|A
d[
|A
[ and [[
|A
= [
|A
[:
_
f
|A
d
|A
=
_
f
|A
h
|A
d[
|A
[ =
_
(fh)
|A
d[[
|A
(49)
Finally, from denition (49):
_
fh1
A
d[[ =
_
fhd[[
A
=
_
(fh)
|A
d[[
|A
(50)
Comparing (47), (48) and (49) with (50), we conclude that:
_
f1
A
d =
_
fd
A
=
_
f
|A
d
|A
Exercise 20
www.probability.net
Solutions to Exercises 114
Exercise 21. Let f L
1
C
(, T, ), where is a complex measure on
(, T). Let h L
1
C
(, T, [[) be such that [h[ = 1 and =
_
hd[[.
Let =
_
fd, i.e. be the map dened by:
E T , (E) =
_
E
fd
From denitions (98), (97) and (49), for all E T:
(E) =
_
f1
E
d =
_
fh1
E
d[[ =
_
E
fhd[[
It follows that =
_
fhd[[, and applying theorem (63), is therefore
a complex measure on (, T), with total variation [[ given by:
E T , [[(E) =
_
E
[fh[d[[ =
_
E
[f[d[[
Let g : (, T) (C, B(C)) be measurable. Applying theorem (21) to
www.probability.net
Solutions to Exercises 115
[[ =
_
[f[d[[, we obtain:
_
[g[d[[ =
_
[g[[f[d[[
and therefore we have the equivalence:
_
[g[d[[ < +
_
[gf[d[[ < +
i.e.
g L
1
C
(, T, ) gf L
1
C
(, T, )
When such condition is satised, we claim that:
_
gd =
_
gfd (51)
This equality is clearly true when g is of the form g = 1
E
where E T
(such a g would automatically lie in L
1
C
(, T, ) since [[ is a nite
measure). By the linearity of the integral (with respect to complex
measures, such a linearity is proved in exercise (16)), equation (51) is
www.probability.net
Solutions to Exercises 116
also true when g is a simple function on (, T). If g is non-negative
and measurable, while being an element of L
1
C
(, T, ), from theo-
rem (18) there exists a sequence (s
n
)
n1
of simple functions on (, T),
such that s
n
g. Let k be an arbitrary element of L
1
C
(, T, [[) with
[k[ = 1 and =
_
kd[[. Then:
lim
n+
_
s
n
d = lim
n+
_
s
n
kd[[
=
_
gkd[[
=
_
gd
where the rst and third equalities stem from denition (97), and the
second from the dominated convergence theorem (23) (and the fact
that s
n
k gk with [s
n
k[ = s
n
g L
1
C
(, T, [[)). Similarly:
lim
n+
_
s
n
fd = lim
n+
_
s
n
fhd[[
www.probability.net
Solutions to Exercises 117
=
_
gfhd[[
=
_
gfd
where the rst and third equalities stem from denition (97), and
the second from the dominated convergence theorem (23) (and the
fact that s
n
fh gfh with [s
n
fh[ = s
n
[f[ g[f[ L
1
C
(, T, [[)).
Having proved (51) for simple functions, for all n 1:
_
s
n
d =
_
s
n
fd
and taking the limit as n +, we see that (51) is also true when-
ever g is non-negative and measurable, while being an element of
L
1
C
(, T, ). If g is an arbitrary element L
1
C
(, T, ), then it can be
decomposed as g = g
1
g
2
+i(g
3
g
4
) where each g
i
is non-negative
and measurable, while being an element of L
1
C
(, T, ). By linearity,
equation (51) is also true for g.
Exercise 21
www.probability.net
Solutions to Exercises 118
Exercise 22.
1. Let =
1
. . .
n
and T = T
1
. . . T
n
. Then:
E T , (E) =
_
E
hd
where h = h
1
. . . h
n
and = [
1
[ . . . [
n
[ is the product
measure, as dened in denition (62). Each total variation [
i
[
being a nite measure, is also a nite measure, and further-
more [h[ = [h
1
[ . . . [h
n
[ = 1. Moreover, the map h is clearly
measurable with respect to T, as the equality:
B B(C) , h
1
i
(B) =
1
. . . h
1
i
(B) . . .
n
shows that each h
i
(viewed as a map dened on the product
space (, T)) is measurable. It follows that is of the form =
_
hd, where h L
1
C
(, T, ). From theorem (63), we conclude
that is a complex measure on (, T). In fact, theorem (63)
www.probability.net
Solutions to Exercises 119
goes further, asserting that the total variation of is:
E T , [[(E) =
_
E
[h[d =
_
1
E
d = (E)
i.e. [[ = = [
1
[ . . . [
n
[.
2. Let A = A
1
. . . A
n
be a measurable rectangle. We have:
_
A
1
A
2
h
1
h
2
d[
1
[ [
2
[ =
_
h
1
h
2
1
A
1
A
2
d[
1
[ [
2
[
=
_
(h
1
1
A
1
)(h
2
1
A
2
)d[
1
[ [
2
[
=
_
_
_
(h
1
1
A
1
)(h
2
1
A
2
)d[
2
[
_
d[
1
[
=
_
h
1
1
A
1
__
h
2
1
A
2
d[
2
[
_
d[
1
[
=
_
h
1
1
A
1

2
(A
2
)d[
1
[
www.probability.net
Solutions to Exercises 120
=
2
(A
2
)
_
h
1
1
A
1
d[
1
[
=
1
(A
1
)
2
(A
2
)
Where crucially, the third equality stems from Fubini theo-
rem (33). If n = 2, then we have nothing further to prove.
If n > 2, we consider the induction hypothesis, for 2 k n:
_
B
k
g
k
d
k
=
1
(A
1
) . . .
k
(A
k
) (52)
where B
k
= A
1
. . . A
k
,
k
= [
1
[ . . . [
k
[ and g
k
is dened
as g
k
= h
1
. . . h
k
. If we assume that such induction hypothesis
is true for some k with 2 k n 1, then:
_
B
k+1
g
k+1
d
k+1
=
_
(g
k
1
B
k
)(h
k+1
1
A
k+1
)d
k
[
k+1
[
=
_
g
k
1
B
k
__
h
k+1
1
A
k+1
d[
k+1
[
_
d
k
www.probability.net
Solutions to Exercises 121
=
k+1
(A
k+1
)
_
B
k
g
k
d
k
=
1
(A
1
) . . .
k+1
(A
k+1
)
where the second equality stems from Fubini theorem (33) and
the fourth from our induction hypothesis (52). This shows
that (52) is in fact true for all k = 2, . . . , n, and nally:
(A) =
_
B
n
g
n
d
n
=
1
(A
1
) . . .
n
(A
n
)
3. We have proved that is a complex measure on (, T) such
that for all measurable rectangle A = A
1
. . . A
n
:
(A) =
1
(A
1
) . . .
n
(A
n
)
In order to prove theorem (66), it remains to show that such a
measure is unique. Suppose and are two complex measures
on (, T) which coincide on the set of measurable rectangles
www.probability.net
Solutions to Exercises 122
T
1
H . . . H T
n
. We dene:
T = E T , (E) = (E)
Then T
1
H . . . H T
n
T, and T is easily seen to be a Dynkin
system on (, T). Indeed, being a measurable rectangle, we
have T. Furthermore, If A, B T and A B, Then:
(B A) = (A) +(B A) (A)
= (B) (A)
= (B) (A)
= (B A)
and therefore B A T. Moreover, if (A
n
)
n1
is a sequence of
elements of T such that A
n
A, then using exercise (13):
(A) = lim
n+
(A
n
) = lim
n+
(A
n
) = (A)
and therefore A T. So T is indeed a Dynkin system on
(, T). The set of measurable rectangles being closed under
www.probability.net
Solutions to Exercises 123
nite intersection (and being a subset of T), from the Dynkin
system theorem (1), we have:
(T
1
H . . . H T
n
) T
and consequently T = T
1
. . . T
n
T. It follows that T = T
and nally = . This proves theorem (66).
Exercise 22
www.probability.net
Solutions to Exercises 124
Exercise 23.
1. We saw in exercise (22) that the complex measure dened by:
E T , (E) =
_
E
h
1
. . . h
n
d[
1
[ . . . [
n
[ (53)
satises the requirement of theorem (66), and is therefore equal
to the product measure
1
. . .
n
. Furthermore, we proved
using theorem (63) that [[ = [
1
[ . . . [
n
[.
2.
|
1
. . .
n
| = [
1
. . .
n
[()
= [
1
[ . . . [
n
[()
= [
1
[(
1
) . . . [
n
[(
n
)
= |
1
| . . . |
n
|
www.probability.net
Solutions to Exercises 125
3. From (53) and [[ = [
1
[ . . . [
n
[, we obtain:
E T , (E) =
_
E
h
1
. . . h
n
d[[
4. Having shown that =
_
hd[[ with h = h
1
. . . h
n
([h[ = 1), it
follows from denition (97) that for all f L
1
C
(, T, ):
_
fd =
_
fhd[[
or equivalently:
_
fd
1
. . .
n
=
_
fh
1
. . . h
n
d[
1
[ . . . [
n
[
5. Let be a permutation of N
n
and h = h
1
. . . h
n
. Then:
_
fd
1
. . .
n
=
_
fhd[
1
[ . . . [
n
[
=
_

(n)
. . .
_

(1)
fhd[
(1)
[ . . . d[
(n)
[
www.probability.net
Solutions to Exercises 126
=
_

(n)
. . .
_

(1)
fd
(1)
. . . d
(n)
where the second equality stems from exercise (17) of Tutorial 7,
and the third equality from:
_

(1)
fhd[
(1)
[ = h
(2)
. . . h
(n)
_

(1)
fh
(1)
d[
(1)
[
= h
(2)
. . . h
(n)
_

(1)
fd
(1)
followed by an induction argument.
Exercise 23
www.probability.net