Measures, Integrals & Martingales (3rd Edition) : Solution Manual Chapter 1-12

Measures, Integrals & Martingales (3rd edition)
Cambridge University Press, Cambridge 2011

ISBN: 9780521615259
Solution Manual Chapter 112
Rene L. Schilling
Dresden, October 2014

1 Prologue.
Solutions to Problems 1.11.2
Problem 1.1 Name the figures on the left and right Figure 1 and Figure 2, respectively. Figure
1 is a triangle but Figure 2 is a (convex) quadrangle: the hypotenuse has a kink. This is
easily seen by comparing in Figure 2 the slopes of the small triangle in the lower left (it
is 2/5) and the larger triangle on top (it is 3/8 2/5).
Problem 1.2 We have to calculate the area of an isosceles triangle of side-length r, base b,
height h and opening angle = 2/2j . From elementary geometry we know that
cos 2 = h
r and sin 2 = b
2r
so that
1 r2
area (triangle) = hb = r2 cos 2 sin 2 = sin .
2 2
sin
Since we have lim0 = 1 we find
r2
area (circle) = lim 2j sin 2
2j
j 2
sin 2j
= r2 lim 22
j
2j
= r2
just as we had expected.
3
2 The pleasures of counting.
Problem 2.1 (i) We have
x A B x A and x / B
x A and x B c
x A B c .
(ii) Using (i) and de Morgans laws (*) yields

(i)
(A B) C = (A B c ) C c = A B c C c
()
= A (B c C c ) = A (B C)c = A (B C).
(iii) Using (i), de Morgans laws (*) and the fact that (C c )c = C gives
(i)
A (B C) = A (B C c )c
()
= A (B c C)
= (A B c ) (A C)
(i)
= (A B) (A C).
(iv) Using (i) and de Morgans laws (*) gives

(i)
A (B C) = A (B C)c
(*)
= A (B c C c )
= (A B c ) (A C c )
(i)
= (A B) (A C)
(v) Using (i) and de Morgans laws (*) gives

(i)
A (B C) = A (B C)c
(*)
= A (B c C c )
= A Bc C c
= A Bc A C c
(i)
= (A B) (A C)
5
R.L. Schilling: Measures, Integrals & Martingales
Problem 2.2 Observe, first of all, that
A C (A B) (B C). ()
This follows easily from
A C = (A C) X
= (A C c ) (B B c )
= (A C c B) (A C c B c )
(B C c ) (A B c )
= (B C) (A B).
Using this and the analogous formula for C A then gives
(A B C) (A B C)
= (A B C) (A B C)c
= [A (A B C)c ] [B (A B C)c ] [C (A B C)c ]
= [A (A B C)] [B (A B C)] [C (A B C)]
= [A (B C)] [B (A C)] [C (A B)]

(2.1(iv))
= (A B) (A C) (B A) (B C) (C A) (C B)
()
= (A B) (B A) (B C) (C B)
= (AB) (BC)
Problem 2.3 It is clearly enough to prove (2.3) as (2.2) follows if I contains 2 points. De
Morgans identities state that for any index set I (finite, countable or not countable) and
any collection of subsets Ai X, i I, we have
c c
(a) ( Ai ) = Ai
c
and (b) ( Ai ) = Aci .
iI iI iI iI
In order to see (a) we note that

c
a ( Ai ) a / Ai
iI iI
i I a / Ai
i I a Aci
a Aci ,
iI
and (b) follows from

c
a ( Ai ) a / Ai
iI iI
6
Solution Manual. Chapter 112. Last update January 4, 2017
i0 I a / Ai0
i0 I a Aci0
a Aci .
iI
Problem 2.4 (i) The inclusion f (A B) f (A) f (B) is always true since A B A and
A B B imply that f (A B) f (A) and f (A B) f (B), respectively. Thus,
f (A B) f (A) f (B).
Furthermore, y f (A) f (B) means that there is some x A but x / B such that
y = f (x), that is: y f (A B). Thus, f (A) f (B) f (A B).
To see that the converse inclusions cannot hold we consider some non injective f .
Take X = [0, 2], A = (0, 1), B = (1, 2), and f [0, 2] R with x f (x) = c (c is some
constant). Then f is not injective and
= f () = f ((0, 1) (1, 2)) f ((0, 1)) f ((1, 2)) = {c}.
Moreover, f (X) = f (B) = {c} = f (X B) but f (X) f (B) = .
(ii) Recall, first of all, the definition of f 1 for a map f X Y and B Y
f 1 (B) = {x X f (x) B}.
Observe that
x f 1 (iI Ci ) f (x) iI Ci
i0 I f (x) Ci0
i0 I x f 1 (Ci0 )
x iI f 1 (Ci ),
and
x f 1 (iI Ci ) f (x) iI Ci
i I f (x) Ci
i I x f 1 (Ci )
x iI f 1 (Ci ),
and, finally,
x f 1 (C D) f (x) C D
f (x) C and f (x) / D
x f 1 (C) and x / f 1 (D)
x f 1 (C) f 1 (D).
7
Problem 2.5
(i), (vi) For every x we have
1AB (x) = 1 x A B
x A, x B
1A (x) = 1 = 1B (x)

1A (x) 1B (x) = 1

min{1A (x), 1B (x)} = 1

(ii), (v) For every x we have
1AB (x) = 1 x A B
x A or x B
1A (x) + 1B (x) 1

min{1A (x) + 1B (x), 1} = 1

max{1A (x), 1B (x)} = 1

(iii) Since A = (A B) (A B) we see that 1AB (x) + 1AB (x) can never have the value
2, thus part (ii) implies
1A (x) = 1(AB)(AB) (x) = min{1AB (x) + 1AB (x), 1}

= 1AB (x) + 1AB (x)
and all we have to do is to subtract 1AB (x) on both sides of the equation.
(iv) With the same argument that we used in (iii) and with the result of (iii) we get
1AB (x) = 1(AB)(AB)(BA) (x)

= 1AB (x) + 1AB (x) + 1BA (x)
= 1A (x) 1AB (x) + 1AB (x) + 1B (x) 1AB (x)
= 1A (x) + 1B (x) 1AB (x).
Problem 2.6 (i) Using 2.5(iii), (iv) we see that
1A B (x) = 1(AB)(BA) (x)

= 1AB (x) + 1BA (x)
= 1A (x) 1AB (x) + 1B (x) 1AB (x)
= 1A (x) + 1B (x) 21AB (x)
and this expression is 1 if, and only if, x is either in A or B but not in both sets.
Thus
1A B (x) 1A (x) + 1B (x) = 1 1A (x) + 1B (x)mod2 = 1.
8
It is also possible to show that
1A B = 1A 1B .
This follows from

0, if x A B;

0, if x Ac B c ;
1A (x) 1B (x) =

+1, if x A B;

1,
if x B A.
Thus,
1A (x) 1B (x) = 1 x (A B) (B A) = A B.
(ii) From part (i) we see that
1A (B C) = 1A + 1B C 21A 1B C
= 1A + 1B + 1C 21B 1C 21A (1B + 1C 21B 1C )
= 1A + 1B + 1C 21B 1C 21A 1B 21A 1C + 41A 1B 1C
and this expression treats A, B, C in a completely symmetric way, i.e.
1A (B C) = 1(A B) C .
(iii) Step 1: (P(X), , ) is an abelian group.
Neutral element: A = A = A;
Inverse element: A A = (A A) (A A) = , i.e. each element is its own inverse.
Associativity: see part (ii);
Commutativity: A B = B A.
Step 2: For the multiplication we have
Associativity: A (B C) = (A B) C;
Commutativity: A B = B A;
One-element: A X = X A = A.
Step 3: Distributive law:
A (B C) = (A B) (A C).
For this we use again indicator functions and the rules from (i) and Problem 2.5:
1A(B C) = 1A 1B C = 1A (1B + 1C mod 2)

= [1A (1B + 1C )] mod 2
9
= [1A 1B + 1A 1C ] mod 2
= [1AB + 1AC ] mod 2
= 1(AB) (AC) .
Problem 2.7 Let f X Y . One has
f surjective B Y f f 1 (B) = B
B Y f f 1 (B) B.
This can be seen as follows: by definition f 1 (B) = {x f (x) B} so that
f f 1 (B) = f ({x f (x) B}) = {f (x) f (x) B} {y y B}
and we have equality in the last step if, and only if, we can guarantee that every y B
is of the form y = f (x) for some x. Since this must hold for all sets B, this amounts to
saying that f (X) = Y , i.e. that f is surjective. The second equivalence is clear since our
argument shows that the inclusion always holds.
Thus, we can construct a counterexample by setting f R R, f (x) = x2 and B = [1, 1].
Then
f 1 ([1, 1]) = [0, 1] and f f 1 ([1, 1]) = f ([0, 1]) = [0, 1] [1, 1].
On the other hand
f injective A X f 1 f (A) = A
A X f 1 f (A) A.
To see this we observe that because of the definition of f 1
f 1 f (A) = {x f (x) f (A)} {x x A} = A ()
since x A always entails f (x) f (A). The reverse is, for non-injective f , wrong since
then there might be some x0 / A but with f (x0 ) = f (x) f (A) i.e. x0 f 1 f (A)A. This
means that we have equality in () if, and only if, f is injective. The second equivalence
is clear since our argument shows that the inclusion always holds.
Thus, we can construct a counterexample by setting f R R, f 1. Then
f ([0, 1]) = {1} and f 1 f ([0, 1]) = f 1 ({1}) = R [0, 1].
Problem 2.8 Assume that for x, y we have f g(x) = f g(y). Since f is injective, we conclude
that
f (g(x)) = f (g(y)) g(x) = g(y),
and, since g is also injective,

g(x) = g(y) x = y
showing that f g is injective.
10
Problem 2.9 Call the set of odd numbers O. Every odd number is of the form 2k 1
where k N. We are done, if we can show that the map f N O, k 2k 1 is
bijective. Surjectivity is clear as f (N) = O. For injectivity we take i, j N such that
f (i) = f (j). The latter means that 2i 1 = 2j 1, so i = j, i.e. injectivity.
The quickest solution is to observe that N Z = N N N {0} N (N) where

N = {n n N} are the strictly negative integers. We know from Example 2.5(iv)
that N N is countable. Moreover, the map N N N (N), ((i, k)) = (i, k)
is bijective, thus #N (N) = #N N is also countable and so is N {0} since
N N {0}, (n) = (n, 0) is also bijective.
Therefore, N Z is a union of three countable sets, hence countable.
An alternative approach would be to write out Z N (the swap of Z and N is for

notational reasonssince the map ((j, k)) = (k, j) from Z N to N Z is bijective,
the cardinality does not change) in the following form
. . . (3, 1) (2, 1) (1, 1) (0, 1) (1, 1) (2, 1) (3, 1) . . .

. . . (3, 2) (2, 2) (1, 2) (0, 2) (1, 2) (2, 2) (3, 2) . . .
. . . (3, 3) (2, 3) (1, 3) (0, 3) (1, 3) (2, 3) (3, 3) . . .
. . . (3, 4) (2, 4) (1, 4) (0, 4) (1, 4) (2, 4) (3, 4) . . .
. . . (3, 5) (2, 5) (1, 5) (0, 5) (1, 5) (2, 5) (3, 5) . . .
. . . (3, 6) (2, 6) (1, 6) (0, 6) (1, 6) (2, 6) (3, 6) . . .

and going through the array, starting with (0, 1), then (1, 1) (1, 2) (0, 2)
(1, 2) (1, 1), then (2, 1) (2, 2) (2, 3) (1, 3) ... in clockwise oriented
-shapes down, left, up.
In Example 2.5(iv) we have shown that #Q #N. Since N Q, we have a canonical
injection N Q, i i so that #N #Q. Using Theorem 2.7 we conclude that
#Q = #N.
The proof of #(N N) = #N can be easily adaptedusing some pretty obvious

notational changesto show that the Cartesian product of any two countable sets
of cardinality #N has again cardinality #N. Applying this m 1 times we see that
#Qn = #N.
mN Qm is a countable union of countable sets, hence countable, cf. Theorem 2.6.
Problem 2.10 Following the hint it is clear that N N {1}, i (i, 1) is a bijection and
that N {1} N N, (i, 1) (i, 1) is an injection. Thus, #N #(N N).
On the other hand, N N = jN N {j} which is a countable union of countable sets,

thus #(N N) #N.
Applying Theorem 2.7 finally gives #(N N) = #N.
11
Problem 2.11 Since E F the map E F, e e is an injection, thus #E #F .
Problem 2.12 Assume that the set {0, 1}N were indeed countable and that {sj }jN was an
enumeration: each sj would be a sequence of the form (dj1 , dj2 , dj3 , ..., djk , ...) with djk {0, 1}.
We could write these sequences in an infinite list of the form:
s1 = d11 d12 d13 d14 . . . d1k . . .

s2 = d21 d22 d23 d24 . . . d2k . . .
s3 = d31 d32 d33 d34 . . . d3k . . .
s4 = d41 d42 d43 d44 . . . d4k . . .

sk = dk1 dk2 dk3 dk4 . . . dkk . . .

and produce a new 0-1-sequence S = (e1 , e2 , e3 , . . .) by setting

0, m =1
if dm
em = .

1, m =0
if dm
Since S differs from s` exactly at position `, S cannot be in the above list, thus, the above
list did not contain all 0-1-sequences, hence a contradiction.
Problem 2.13 Consider the function f (0, 1) R given by
1 1
f (x) = .
1x x
This function is obviously continuous and we have limx0 f (x) = and limx1 f (x) =
+. By the intermediate value theorem we have therefore f ((0, 1)) = R, i.e. surjectivity.
1 1
Since f is also differentiable and f (x) = + 2 > 0, we see that f is strictly
(1 x) 2 x
increasing, hence injective, hence bijective.
Problem 2.14 Since A1 jN Aj it is clear that c = #A1 # jN Aj . On the other hand,

#Aj = c means that we can map Aj bijectively onto R and, using Problem 2.13, we map R
bijectively onto (0, 1) or (j 1, j). This shows that # jN Aj # jN (j 1, j) #R = c.
Using Theorem 2.7 finishes the proof.
Problem 2.15 Since we can write each x (0, 1) as an infinite dyadic fraction (o.k. if it is
finite, fill it up with an infinite tail of zeroes !), the proof of Theorem 2.8 shows that
#(0, 1) #{0, 1}N .
On the other hand, thinking in base-4 expansions, each element of {1, 2}N can be inter-
preted as a unique base-4 fraction (having no 0 or 3 in its expansion) of some number in
(0, 1). Thus, #{1, 2}N #N.
But #{1, 2}N = #{0, 1}N and we conclude with Theorem 2.7 that #(0, 1) = #{0, 1}N .
12
Problem 2.16 Just as before, expand x (0, 1) as an n-adic fraction, then interpret each element
of {1, 2, . . . , n + 1}N as a unique (n + 1)-adic expansion of a number in (0, 1) and observe
that #{1, 2, . . . , n + 1}N = {0, 1, . . . , n}N .
Problem 2.17 Take a vector (x, y) (0, 1) (0, 1) and expand its coordinate entries x, y as
dyadic numbers:
x = 0.x1 x2 x3 . . . , y = 0.y1 y2 y3 . . . .
Then z = 0.x1 y1 x2 y2 x3 y3 . . . is a number in (0, 1). Conversely, we can zip each z =

0.z1 z2 z3 z4 . . . (0, 1) into two numbers x, y (0, 1) by setting
x = 0.z2 z4 z6 z8 . . . , y = 0.z1 z3 z5 z7 . . .
This is obviously a bijective operation.
Since we have a bijection between (0, 1) R it is clear that we have also a bijection
between (0, 1) (0, 1) R R.
Problem 2.18 We have seen in Problem 2.18 that #{0, 1}N = #{1, 2}N = c. Obviously, {1, 2}N
NN RN and since we have a bijection between (0, 1) R one extends this (using
coordinates) to a bijection between (0, 1)N RN . Using Theorem 2.9 we get
c = #{1, 2}N #NN #RN = c,
and, because of Theorem 2.7 we have equality in the above formula.
Problem 2.19 Let F F with #F = n Then we can write F as a tuple of length n (having n
pairwise different entries...) and therefore we can interpret F as an element of mN Nm .
In this sense, F mN Nm and #F mN Nm = #N since countably many countable
sets are again countable. Since N F we get #F = #N by Theorem 2.7.
Alternative: Define a map F N by
F A (A) = 2a
aA
. It is clear that increases if A gets bigger: A B (A) (B). Let A, B F

be two finite sets, say A = {a1 , a2 , . . . , aM } and {b1 , b2 , . . . , bN } (ordered according to size
with a1 , b1 being the smallest and aM , bN the biggest) such that (A) = (B). Assume,
to the contrary, that A B. If aM bN , say aM > bN , then
2aM 1 aM 1 j
(A) ({aM }) 2aM > = 2
21 j=1
= ({1, 2, 3, . . . aM 1})
(B),
13
which cannot be the case since we assumed (A) = (B). Thus, aM = bN . Now consider
recursively the next elements, aM 1 and bN 1 and the same conclusion yields their equality
etc. The process stops after min{M, N } steps. But if M N , say M > N , then A
would contain at least one more element than B, hence (A) > (B), which is also a
contradiction. This, finally shows that A = B, hence that is injective.
On the other hand, each natural number can be expressed in terms of finite sums of powers
of base-2, so that is also surjective.
Thus, #F = #N.
Problem 2.20 (Let F be as in the previous exercise.) Observe that the infinite sets from P(N),
I = P(N) F can be surjectively mapped onto {0, 1}N : if {a1 , a2 , a3 , . . .} = A N, then
define an infinite 0-1-sequence (b1 , b2 , b3 , . . .) by setting bj = 0 or bj = 1 according to
whether aj is even or odd. This is a surjection of P(N) onto {0, 1}N and so #P(N)
#{0, 1}N . Call this map and consider the family 1 (s), s {0, 1}N in I, consisting
of obviously disjoint infinite subsets of N which lead to the same 0-1-sequence s. Now
choose from each family 1 (s) a representative, call it r(s) I. Then the map s r(s)
is a bijection between {0, 1}N and a subset of I, the set of all representatives. Hence, I
has at least the same cardinality as {0, 1}N and as such a bigger cardinality than N.
Problem 2.21 Denote by the map P(N) A 1A {0, 1}N . Let = (d1 , d2 , d3 , . . .) {0, 1}N
and define A() = {j N dj = 1}. Then = (1A() (j))jN showing that is surjective.
On the other hand,
1A = 1B 1A (j) = 1B (j) j N A = B.
This shows the injectivity of , and #P(N) = #{0, 1}N follows.
14
3 -Algebras.
Problem 3.1 (i) It is clearly enough to show that A, B A A B A, because the case
of N sets follows from this by induction, the induction step being
A1 . . . AN AN +1 = B AN +1 A.

=BA
Let A, B A. Then, by (2 ) also Ac , B c A and, by (3 ) and (2 )
A B = (Ac B c )c = (Ac B c . . .)c A.
Alternative: Of course, the last argument also goes through for N sets:
A1 A2 . . . AN = (Ac1 Ac2 . . . AcN )c

= (Ac1 . . . AcN . . .)c A.
(ii) By (2 ) we have A A Ac A. Use Ac instead of A and observe that (Ac )c = A

to see the claim.
(iii) Clearly Ac , B c A and so, by part (i), A B = A B c A as well as A B =
(A B) (B A) A.
Problem 3.2 (iv) Let us assume that B and B X. Then B c / {, B, X}. Since with B
also B c must be contained in a -algebra, the family {, B, X} cannot be one.
(vi) Set AE = {E A A A}. The key observation is that all set operations in AE are
now relative to E and not to X. This concerns mainly the complementation of sets!
Let us check (1 )(3 ).
Clearly = E AE . If B A, then B = E A for some A A and the complement
of B relative to E is E B = E B c = E (E A)c = E (E c Ac ) = E Ac AE as
Ac A. Finally, let (Bj )jN AE . Then there are (Aj )jN A such that Bj = E Aj .
Since A = jN Aj A we get jN Bj = jN (E Aj ) = E jN Aj = E A AE .
(vii) Note that f 1 interchanges with all set operations. Let A, Aj , j N be sets in A.
We know that then A = f 1 (A ), Aj = f 1 (Aj ) for suitable A, Aj A . Since A is,
by assumption a -algebra, we have
= f 1 () A as A
15
Ac = (f 1 (A )) = f 1 (A ) A A A
c c c
as
1 1
Aj = f (Aj ) = f Aj A as Aj A
jN jN jN jN
which proves (1 )(3 ) for A.
Problem 3.3 (i) Since G is a -algebra, G competes in the intersection of all -algebras C G
appearing in the definition of A in the proof of Theorem 3.4(ii). Thus, G (G) while
G (G) is always true.
(ii) Without loss of generality we can assume that A X since this would simplify
the problem. Clearly {, A, Ac , X} is a -algebra containing A and no element can be
removed without losing this property. Thus {, A, Ac , X} is minimal and, therefore,
= ({A}).
(iii) Assume that F G. Then we have F G (G). Now C = (G) is a potential

competitor in the intersection appearing in the proof of Theorem 3.4(ii), and as
such C (F), i.e. (G) (F).
Problem 3.4 (i) {, (0, 12 ), {0}[ 21 , 1], [0, 1]}. We have 2 atoms (see the explanations below):
(0, 12 ), (0, 21 )c .
(ii) {, [0, 41 ), [ 14 , 43 ], ( 34 , 1], [0, 43 ], [ 14 , 1], [0, 41 ) ( 34 , 1], [0, 1]}. We have 3 atoms (see be-
low): [0, 14 ), [ 14 , 43 ], ( 34 , 1].
(iii) same solution as (ii)
Parts (ii) and (iii) are quite tedious to do and they illustrate how difficult it can be to
find a -algebra containing two distinct sets.... imagine how to deal with something that
is generated by 10, 20, or infinitely many sets. Instead of giving a particular answer, let
us describe the method to find ({A, B}) practically, and then we are going to prove it.
1. Start with trivial sets and given sets: , X, A, B.
2. now add their complements: Ac , B c
3. now add their unions and intersections and differences: A B, A B, A B, B A
4. now add the complements of the sets in 3.: Ac B c , Ac B c , (A B)c , (B A)c
5. finally, add unions of differences and their complements: (AB)(B A), (AB)c
(B A)c .
All in all one should have 16 sets (some of them could be empty or X or appear several
times, depending on how much A differs from B). Thats it, but the trouble is: is this
construction correct? Here is a somewhat more systematic procedure:
Definition: An atom of a -algebra A is a non-void set A A that contains no other

set of A.
16
Since A is stable under intersections, it is also clear that all atoms are disjoint sets! Now
we can make up every set from A as union (finite or countable) of such atoms. The task
at hand is to find atoms if A, B are given. This is easy: the atoms of our future -algebra
must be: AB, B A, AB, (AB)c . (Test it: if you make a picture, this is a tesselation
of our space X using disjoint sets and we can get back A, B as union! It is also minimal,
since these sets must appear in ({A, B}) anyway.) The crucial point is now:
Theorem. If A is a -algebra with N atoms (finitely many!), then A consists of exactly

2N elements.
Proof. The question is how many different unions we can make out of N sets. Simple
answer: we find (Nj ), 0 j N different unions involving exactly j sets (j = 0 will, of
course, produce the empty set) and they are all different as the atoms were disjoint. Thus,
j=0 ( j ) = (1 + 1) = 2 different sets.
N
we get N N N
It is clear that they constitute a -algebra.
This answers the above question. The number of atoms depends obviously on the relative
position of A, B: do they intersect, are they disjoint etc. Have fun with the exercises and
do not try to find -algebras generated by three or more sets..... (By the way: can you
think of a situation in [0, 1] with two subsets given and exactly four atoms? Can there
be more?)
Problem 3.5 (i) See the solution to Problem 3.4.
(ii) If A1 , . . . , AN X are given, there are at most 2N atoms. This can be seen by
induction. If N = 1, then there are #{A, Ac } = 2 atoms. If we add a further
set AN +1 , then the worst case would be that AN +1 intersects with each of the 2N
atoms, thus splitting each atom into two sets which amounts to saying that there are
2 2N = 2N +1 atoms.
Problem 3.6 O1 Since contains no element, every element x admits certainly some
neighbourhood B (x) and so O. Since for all x Rn also B (x) Rn , Rn is
clearly open.
O2 Let U, V O. If U V = , we are done. Else, we find some x U V . Since U, V

are open, we find some 1 , 2 > 0 such that B1 (x) U and B2 (x) V . But then we
can take h = min{1 , 2 } > 0 and find
Bh (x) B1 (x) B2 (x) U V,
i.e. U V O. For finitely many, say N , sets, the same argument works. Notice that
already for countably many sets we will get a problem as the radius h = min{j j
N} is not necessarily any longer > 0.
17
O2 Let I be any (finite, countable, not countable) index set and (Ui )iI O be a family
of open sets. Set U = iI Ui . For x U we find some j I with x Uj , and
since Uj was open, we find some j > 0 such that Bj (x) Uj . But then, trivially,
Bj (x) Uj iI Ui = U proving that U is open.
The family On cannot be a -algebra since the complement of an open set U , Rn is

closed.
Problem 3.7 Let X = R and set Uk = ( k1 , k1 ) which is an open set. Then kN Uk = {0} but a
singleton like {0} is closed and not open.
Problem 3.8 We know already that the Borel sets B = B(R) are generated by any of the
following systems:
{[a, b) a, b Q}, {[a, b) a, b R},

{(a, b) a, b Q}, {(a, b) a, b R}, O1 , or C 1
Here is just an example how to solve the problem. Let b > a. Since (, b)(, a) = [a, b)
we get that
{[a, b) a, b Q} ({(, c) c Q})

B = ({[a, b) a, b Q}) ({(, c) c Q}).
On the other hand we find that (, a) = kN [k, a) proving that
{(, a) a Q} ({[c, d) c, d Q}) = B

({(, a) a Q}) B
and we get equality.
The other cases are similar.
Problem 3.9 Let B = {Br (x) x Rn , r > 0} and let B = {Br (x) x Qn , r Q+ }. Clearly,
B B On
(B ) (B) (On ) = B(Rn ).
On the other hand, any open set U On can be represented by
U= B. ()
BB , BU
Indeed, U BB , BU B follows by the very definition of the union. Conversely, if x U

we use the fact that U is open, i.e. there is some B (x) U . Without loss of generality
we can assume that is rational, otherwise we replace it by some smaller rational .
Since Qn is dense in Rn we can find some q Qn with x q < /3 and it is clear that
B/3 (q) B (x) U . This shows that U BB , BU B.
18
Since #B = #(Qn Q) = #N, formula () entails that
On (B ) (On ) = (B)
and we are done.
Problem 3.10 (i) O1 : We have = A OA , A = X A OA .
O1 : Let U = U A OA , V = V A OA with U, V O. Then U V = (U V )A

OA since U V O.
O2 : Let Ui = Ui A OA with Ui O. Then i Ui = (i Ui ) A OA since i Ui O.
(ii) We use for a set A and a family F P(X) the shorthand A F = {A F F F}.
Clearly, A O A (O) = A B(X). Since the latter is a -algebra, we have
(A O) A B(X) i.e. B(A) A B(X).
For the converse inclusion we define the family
= {B X A B (A O)}.
It is not hard to see that is a -algebra and that O . Thus B(X) = (O)
which means that
A B(X) (A O).
Notice that this argument does not really need that A B(X). If, however, A B(X)
we have in addition to A B(X) = B(A) that
B(A) = {B A B B(X)}
Problem 3.11 (i) As in the proof of Theorem 3.4 we set
m(E) = M. ()
M monotone class
M E
Since the intersection M = iI Mi of arbitrarily many monotone classes Mi , i I, is

again a monotone class [indeed: if (Aj )jN M, then (Aj )jN is in every Mi and so
are j Aj , j Aj ; thus j Aj , j Aj M] and () is evidently the smallest monotone
class containing some given family E.
(ii) Since E is stable under complementation and contains the empty set we know that
X E. Thus, and, by the very definition, is stable under taking complements
of its elements. If (Sj )jN , then (Sjc )jN and
c
Sj m(E), ( Sj ) = Sjc m(E)
j j j
which means that j Sj .
19
(iii) E : if E E, then E m(E). Moreover, as E is stable under complementation,

E c m(E) for all E E, i.e. E .
m(E): obvious from the definition of .
m(E) (E): every -algebra is also a monotone class and the inclusion follows from
the minimality of m(E).
Finally apply the -hull to the chain E m(E) (E) and conclude that m(E)
(E).
Problem 3.12 (i) Since M is a monotone class, this follows from Problem 3.11.
(ii) Let F Rn be any closed set. Then Un = F + B1/n (0) = {x + y x F, y B1/n (0)}
is an open set and nN Un = F . Indeed,
Un = B1/n (x) = {z Rn x z < 1

n for some x F }
xF
which shows that Un is open, F Un and F n Un . On the other hand, if z Un

for all n N, then there is a sequence of points xn F with the property z xn <
1 n
n 0. Since F is closed, z = limn xn F and we get F = n Un .
Since M is closed under countable intersections, F M for any closed set F .
(iii) Identical to Problem 3.11(ii).
(iv) Use Problem 3.11(iv).
20
4 Measures.
Problem 4.1 (i) We have to show that for a measure and finitely many, pairwise disjoint
sets A1 , A2 , . . . , AN A we have
(A1 A2 . . . AN ) = (A1 ) + (A2 ) + . . . + (AN ).
We use induction in N N. The hypothesis is clear, for the start (N = 2) see

Proposition 4.3(i). Induction step: take N + 1 disjoint sets A1 , . . . , AN +1 A, set
B = A1 . . . AN A and use the induction start and the hypothesis to conclude
(A1 . . . AN AN +1 ) = (B AN +1 )
= (B) + (AN +1 )
= (A1 ) + . . . + (AN ) + (AN +1 ).
(iv) To get an idea what is going on we consider first the case of three sets A, B, C.
Applying the formula for strong additivity thrice we get
(A B C) = (A (B C))
= (A) + (B C) ( A (B C) )

= (AB)(AC)
= (A) + (B) + (C) (B C) (A B)

(A C) + (A B C).
As an educated guess it seems reasonable to suggest that

n
(A1 . . . An ) = (1)k+1 ( Aj ).
k=1 j
{1,...,n}
#=k
We prove this formula by induction. The induction start is just the formula from
Proposition 4.3(iv), the hypothesis is given above. For the induction step we observe
that
= +
{1,...,n+1} {1,...,n,n+1} {1,...,n,n+1}
#=k #=k, n+1/ #=k, n+1
()
= +
{1,...,n} {1,...,n}
#=k # =k1, = {n+1}
21
Having this in mind we get for B = A1 . . . An and An+1 using strong additivity
and the induction hypothesis (for A1 , . . . , An resp. A1 An+1 , . . . , An An+1 )
(B An+1 ) = (B) + (An+1 ) (B An+1 )

n
= (B) + (An+1 ) ( (Aj An+1 ))
j=1
n
= (1)k+1 ( Aj ) + (An+1 )
k=1 j
{1,...,n}
#=k
n
+ (1)k+1 (An+1 Aj ).
k=1 j
{1,...,n}
#=k
Because of () the last line coincides with

n+1
(1)
k+1
( Aj )
k=1 j
{1,...,n,n+1}
#=k
and the induction is complete.
(v) We have to show that for a measure and finitely many sets B1 , B2 , . . . , BN A we
have
(B1 B2 . . . BN ) (B1 ) + (B2 ) + . . . + (BN ).
We use induction in N N. The hypothesis is clear, for the start (N = 2) see

Proposition 4.3(v). Induction step: take N + 1 sets B1 , . . . , BN +1 A, set C =
B1 . . . BN A and use the induction start and the hypothesis to conclude
(B1 . . . BN BN +1 ) = (C BN +1 )
(C) + (BN +1 )
(B1 ) + . . . + (BN ) + (BN +1 ).
Problem 4.2 (i) The Dirac measure is defined on an arbitrary measurable space (X, A) by

0, if x / A
x (A) = , where A A and x X is a fixed point.

1, if x A

(M1 ) Since contains no points, x / and so x () = 0.
(M2 ) Let (Aj )jN A a sequence of pairwise disjoint measurable sets. If x jN Aj ,

there is exactly one j0 with x Aj0 , hence

x Aj = 1 = 1 + 0 + 0 + . . .
jN
= x (Aj0 ) + x (Aj )
jj0
= x (Aj ).
jN
22
If x / jN Aj , then x / Aj for every j N, hence

x Aj = 0 = 0 + 0 + 0 + . . . = x (Aj ).
jN jN

0, if #A #N
(ii) The measure is defined on (R, A) by (A) = where A = {A R

1, if #Ac #N

#A #N or #Ac #N}. (Note that #A #N if, and only if, #Ac = #RA > #N.)
(M1 ) Since contains no elements, it is certainly countable and so () = 0.
(M2 ) Let (Aj )jN be pairwise disjoint A-sets. If all of them are countable, then
A = jN is countable and we get

Aj = (A) = 0 = (Aj ).
jN jN
If at least one Aj is not countable, say for j = j0 , then A Aj0 is not countable
and therefore (A) = (Aj0 ) = 1. Assume we could find some other j1 j0 such
that Aj0 , Aj1 are not countable. Since Aj0 , Aj1 A we know that their complements
Acj0 , Acj1 are countable, hence Acj0 Acj1 is countable and, at the same time, A.
Because of this, (Acj0 Acj1 )c = Aj0 Aj1 = cannot be countable, which is absurd!
Therefore there is at most one index j0 N such that Aj0 is uncountable and we get
then

Aj = (A) = 1 = 1 + 0 + 0 + . . . = (Aj0 ) + (Aj ).
jN jj0

#A, if A is finite
(iii) We have an arbitrary measurable space (X, A) and the measure A = .

, else

(M1 ) Since contains no elements, # = 0 and = 0.
(M2 ) Let (Aj )jN be a sequence of pairwise disjoint sets in A. Case 1: All Aj are
finite and only finitely many, say the first k, are non-empty, then A = jN Aj is
effectively a finite union of k finite sets and it is clear that
A = A1 + . . . + Ak + + + . . . = Aj .
jN
Case 2: All Aj are finite and infinitely many are non-void. Then their union A =
jN Aj is an infinite set and we get
A = = Aj .
jN
Case 3: At least one Aj is infinite, and so is then the union A = jN Aj . Thus,
A = = Aj .
jN
23
(iv) On a countable set = {1 , 2 , . . .} we define for a sequence (pj )jN [0, 1] with
jN pj = 1 the set-function
P (A) = pj = pj j (A), A .
j j A jN
(M1 ) P () = 0 is obvious.
(M2 ) Let (Ak )kN be pairwise disjoint subsets of . Then
P (Ak ) = pj j (Ak )
kN kN jN
= pj j (Ak )
jN kN
= pj ( j (Ak ))
jN kN
= pj j ( Ak )
jN k
= P ( Ak ).
k
The change in the order of summation needs justification; one possibility is the
argument used in the solution of Problem 4.6(ii). (Note that the reordering theorem
for absolutely convergent series is not immediately applicable since we deal with a
double series!)
(v) This is obvious.
Problem 4.3 On (R, B(R)) the function is not be a measure, since we can take the
sets A = (1, ), B = (, 1) which are disjoint, not countable and both have non-
countable complements. Hence, (A) = (B) = 1. On the other hand, A B is
non-countable and has non-countable complement, [1, 1]. So, (A B) = 1. This
contradicts the additivity: (A B) = 1 2 = (A) + (B). Notice that the choice of
the -algebra A avoids exactly this situation. B is the wrong -algebra for .
On Q (and, actually, any possible -algebra thereon) the problem is totally different:
if A is countable, then Ac = Q A is also countable and vice versa. This means that
(A) is, according to the definition, both 1 and 0 which is, of course, impossible.
This is to say: is not well-defined. makes only sense on a non-countable set X.
Problem 4.4 (i) If A = {, R}, then is a measure.
But as soon as A contains one set A which is trivial (i.e. either or X), we have
actually Ac A which is also non-trivial. Thus,
1 = (X) = (A Ac ) (A) + (Ac ) = 1 + 1 = 2
and cannot be a measure.
24
(ii) If we equip R with a -algebra which contains sets such that both A and Ac can
be infinite (the Borel -algebra would be such an example: A = (, 0) Ac =
[0, )), then is not well-defined. The only type of sets where is well-defined is,
thus,
A = {A R #A < or #Ac < }.
But this is no -algebra as the following example shows: Aj = {j} A, j N, are

pairwise disjoint sets but jN Aj = N is not finite and its complement is R N not
finite either! Thus, N / A, showing that A cannot be a -algebra. We conclude
that can never be a measure if the -algebra contains infinitely many sets. If we
are happy with finitely many sets only, then here is an example that makes into a
measure A = {, {69}, R {69}, R} and similar families are possible, but the point is
that they all contain only finitely many members.
Problem 4.5 Denote by one-dimensional Lebesgue measure and consider the Borel sets Bk =
(k, ). Clearly k Bk = , k N, so that Bk . On the other hand,
(Bk ) = inf (Bk ) = 0 = ()

k
which shows that the finiteness condition in Theorem 4.4 (iii ) and (iii ) is essential.
Problem 4.6 (i) Clearly, = a + b A [0, ] (since a, b 0!). We check (M1 ), (M2 ).
(M1 ) Clearly, () = a() + b() = a 0 + b 0 = 0.
(M2 ) Let (Aj )jN A be mutually disjoint sets. Then we can use the -additivity
of , to get
( Aj ) = a( Aj ) + b( Aj )
jN jN jN
= a (Aj ) + b (Aj )
jN jN
= (a(Aj ) + b(Aj ))
jN
= (Aj ).
jN
Since all quantities involved are positive and since we allow the value + to be
attained, there are no convergence problems.
(ii) Since all j are positive, the sum jN j j (A) is a sum of positive quantities and,
allowing the value + to be attained, there is no convergence problem. Thus,
A [0, ] is well-defined. Before we check (M1 ), (M2 ) we prove the following
Lemma. Let ij , i, j N, be real numbers. Then
sup sup ij = sup sup ij .

iN jN jN iN
25
Proof. Observe that we have mn supjN supiN ij for all m, n N. The right-hand
side is independent of m and n and we may take the sup over all n
sup mn sup sup ij m N

nN jN iN
and then, with the same argument, take the sup over all m
sup sup mn sup sup ij m N.

mN nN jN iN
The opposite inequality, , follows from the same argument with i and j inter-
changed.
(M1 ) We have () = jN j j () = jN j 0 = 0.
(M2 ) Take pairwise disjoint sets (Ai )iN A. Then we can use the -additivity of
each of the j s to get
( Ai ) = j j ( Ai )
iN jN iN
N
= lim j j (Ai )
N j=1 iN
N M
= lim j lim j (Ai )
N j=1 M i=1
N M
= lim lim j j (Ai )
N M j=1 i=1
N M
= sup sup j j (Ai )
N N M N j=1 i=1
where we used that the limits are increasing limits, hence suprema. By our lemma:
M N
( Ai ) = sup sup j j (Ai )
iN M N N N i=1 j=1
M N
= lim lim j j (Ai )
M N i=1 j=1
M
= lim j j (Ai )
M i=1 jN
M
= lim (Ai )
M i=1
= (Ai ) .
iN
26
Problem 4.7 Set (A) = (A F ). We know, by assumption, that is a measure on (X, A).
We have to show that is a measure on (X, A). Since F A, we have F A A for all
A A, so is well-defined. Moreover, it is clear that (A) [0, ]. Thus, we only have
to check
(M1 ) () = ( F ) = () = 0.
(M2 ) Let (Aj )jN A be a sequence of pairwise disjoint sets. Then also (Aj F )jN A
are pairwise disjoint and we can use the -additivity of to get
( Aj ) = (F Aj ) = ( (F Aj ))
jN jN jN
= (F Aj )
jN
= (Aj ).
jN
Problem 4.8 Since P is a probability measure, P (Acj ) = 1 P (Aj ) = 0. By -subadditivity,
P ( Acj ) P (Acj ), = 0
jN jN
and we conclude that

c
P ( Aj ) = 1 P ([ Aj ] ) = 1 P ( Acj ) = 1 0 = 0.
jN jN jN
Problem 4.9 Note that
Aj Bk = (Aj Bk ) (Aj Bj )
j k j k j

Bj j
Since j Bj j Aj we get from -subadditivity
( Aj ) ( Bj ) = ( Aj Bk )
j j j k
( (Aj Bj ))
j
(Aj Bj ).
j
Problem 4.10 (i) We have A and () = 0, thus N .

(ii) Since M A (this is essential in order to apply to M !) we can use the monotonicity
of measures to get 0 (M ) (N ) = 0, i.e. (M ) = 0 and M N follows.
(iii) Since all Nj A, we get N = jN Nj A. By the -subadditivity of a measure we
find
0 (N ) = ( Nj ) (Nj ) = 0,
jN jN
hence (N ) = 0 and so N N .
27
Problem 4.11 (i) The one-dimensional Borel sets B = B 1 are defined as the smallest -algebra
containing the open sets. Pick x R and observe that the open intervals (x k1 , x+ k1 ),
k N, are all open sets and therefore (x k1 , x + k1 ) B. Since a -algebra is stable
under countable intersections we get {x} = kN (x k1 , x + k1 ) B.
Using the monotonicity of measures and the definition of Lebesgue measure we find
2 k
0 ({x}) ((x k1 , x + k1 )) = (x + k1 ) (x k1 ) = k 0.
[Following the hint leads to a similar proof with [x k1 , x+ k1 ) instead of (x k1 , x+ k1 ).]
(ii) a) Since Q is countable, we find an enumeration {q1 , q2 , q3 , . . .} and we get trivially

Q = jN {qj } which is a disjoint union. (This shows, by the way, that Q B
as {qj } B.) Therefore, using part (i) of the problem and the -additivity of
measures,
(Q) = ( {qj }) = ({qj }) = 0 = 0.
jN jN jN
b) Take again an enumeration Q = {q1 , q2 , q3 , . . .}, fix > 0 and define C() as stated
in the problem. Then we have C() B and Q C(). Using the monotonicity
and -subadditivity of we get
0 (Q) (C())
= ( [qk 2k , qk + 2k ))

kN
([qk 2k , qk + 2k ))

kN
= 2 2k
kN
1
= 2 2
= 2.
1 1
2
As > 0 was arbitrary, we can make 0 and the claim follows.
(iii) Since 0x1 {x} is a disjoint union, only the countability assumption is violated.
Lets see what happens if we could use -additivity for such non-countable unions:
0 = 0 = ({x}) = ( {x}) = ([0, 1]) = 1

0x1 0x1 0x1
which is impossible.
Problem 4.12 Without loss of generality we may assume that a b; set = a + b . Then
(B) = 0 if, and only if, a / B and b / B. Since {a}, {b} and {a, b} are Borel sets, all null
sets of are given by
N = {B {a, b} B B(R)}.
(This shows that, in some sense, null sets can be fairly large!).
28
Problem 4.13 Let us write N for the family of all (proper and improper) subsets of null sets.
We note that sets in N can be measurable (that is: N A) but need not be measurable.
(i) Since N, we find that A = A A for every A A; thus, A A . Let us check
that A is a -algebra.
(1 ) Since A A , we have A .
(2 ) Let A A . Then A = A N for A A and N N. By definition, N M A
where (M ) = 0. Now
A = (A N )c = Ac N c
c
= Ac N c (M c M )
= (Ac N c M c ) (Ac N c M )
= (Ac M c ) (Ac N c M )
where we used that N M , hence M c N c , hence M c N c = M c . But now we

see that Ac M c A and Ac N c M N since Ac N c M M and M A is
a null set: (M ) = 0.
(3 ) Let (Aj )jN be a sequence of A -sets. From its very definition we know that
each Aj = Aj Nj for some (not necessarily unique!) Aj A and Nj N. So,

Aj = (Aj Nj ) = ( Aj ) ( Nj ) = A N.
jN jN jN jN
Since A is a -algebra, A A. All we have to show is that Nj is in N. Since

each Nj is a subset of a (measurable!) null set, say, Mj A, we find that
N = jN Nj jN Mj = M A and all we have to show is that (M ) = 0. But
this follows from -subadditivity,
0 (M ) = ( Mj ) (Mj ) = 0.
jN jN
Thus, A N A .
(ii) As already mentioned in part (i), A A could have more than one representation,
e.g. A N = A = B M with A, B A and N, M N. If we can show that
is independent of the representation of A .
(A) = (B) then the definition of
Since M, N are not necessarily measurable but, by definition, subsets of (measurable)
null sets M , N A we find
A A N = B M B M ,
B B M = A N A N
and since A, B, B M , A N A, we get from monotonicity and subadditivity of

measures
(A) (B M ) (B) + (M ) = (B),
29
(B) (A N ) (A) + (N ) = (A)
which shows (A) = (B).

(iii) We check (M1 ) and (M2 )
(M1 ) Since = A , A, N, we have
() = () = 0.
(M2 ) Let (Aj )jN A be a sequence of pairwise disjoint sets. Then Aj = Aj Nj
for some Aj A and Nj N. These sets are also mutually disjoint, and with
the arguments in (i) we see that A = A N where A A , A A, N N stand
for the unions of Aj , Aj and Nj , respectively. Since
does not depend on the
special representation of A -sets, we get
( Aj ) =
(A ) = (A) = ( Aj )
jN jN
= (Aj )
jN
(Aj )
=
jN
showing that
is -additive.
(iv) Let M be a
null set, i.e. M A and
(M ) = 0. Take any B M . We have to
show that B A and
(B) = 0. The latter is clear from the monotonicity of
once
we have shown that B A which means, once we know that we may plug B into
.
Now, B M and M = M N for some M A and N N. As
(M ) = 0 we also
know that (M ) = 0. Moreover, we know from the definition of N that N N for
some N A with (N ) = 0. This entails
B M = M N M N A
and (M N ) (M ) + (N ) = 0.
Hence B N as well as B = B A . In particular,

(B) = () = 0.
(v) Set C = {A X A, B A, A A A B, (B A) = 0}. We have to show

that A = C.
Take A A . Then A = A N with A A, N N and choose N A, N N and
(N ) = 0. This shows that
A A = A N A N = B A
and that (B A) = ((A N ) A) (N ) = 0. (Note that (A N ) A =

(A N ) Ac = N Ac N and that equality need not hold!).
Conversely, take A C. Then, by definition, A A B with A, B A and
(B A) = 0. Therefore, N = B A is a null set and we see that A A B A,
i.e. A A N. So, A = A (A A) where A A and A A N showing that
A A .
30
Problem 4.14 (i) Since B is a -algebra, it is closed under countable (disjoint) unions of its
elements, thus inherits the properties (M1 ), (M2 ) directly from .
(ii) Yes [yes], since the full space X B so that (X) = (X) is finite [resp. = 1].
(iii) No, -finiteness is also a property of the -algebra. Take, for example, Lebesgue mea-
sure on the Borel sets (this is -finite) and consider the -algebra C = {, (, 0), [0, ), R}.
Then C is not -finite since there is no increasing sequence of C-sets having finite
measure.
Problem 4.15 By definition, is -finite if there is an increasing sequence (Bj )jN A such
that Bj X and (Bj ) < . Clearly, Ej = Bj satisfies the condition in the statement of
the problem.
Conversely, let (Ej )jN be as stated in the problem. Then Bn = E1 . . .En is measurable,
Bn X and, by subadditivity,
n
(Bn ) = (E1 . . . En ) (Ej ) < .
j=1
Remark: A small change in the above argument allows to take pairwise disjoint sets Ej .
31
5 Uniqueness of measures.
Problem 5.1 Since X D and since complements are again in D, we have = X c D.
If A, B D are disjoint, we set A1 = A, A2 = B, Aj = j 3. Then (Aj )jN D is a

sequence of pairwise disjoint sets, and by (3 ) we find that
A B = Aj D.
jN
Since (1 ) = (3 ), (2 ) = (2 ) and since (3 ) (3 ), it is clear that every -algebra

is also a Dynkin system; that the converse is, in general, wrong is seen in Problem 5.2.
Problem 5.2 Consider (3 ) only, as the other two conditions coincide: (j ) = (j ), j = 1, 2.

We show that (3 ) breaks down even for finite unions. If A, B D are disjoint, it is
clear that A, B and also A B contain an even number of elements. But if A, B have
non-void intersection, and if this intersection contains an odd number of elements, then
A B contains an odd number of elements. Here is a trivial example:
A = {1, 2} D, B = {2, 3, 4, 5} D,
whereas
A B = {1, 2, 3, 4, 5} / D.
This means that (3 ) holds, but (3 ) fails.
Problem 5.3 Mind the misprint: A B must be assumed and is missing in the statement of
the problem! We verify the hint first. Using de Morgans laws we get
R Q = R (R Q) = R (R Q)c = (Rc (R Q))c = (Rc (R Q))c
where the last equality follows since Rc (R Q) = .
Now we take A, B D such that A B. In particular A B = A. Taking this into account

and setting Q = A, R = B we get from the above relation
c
B A = ( B c A ) D

D

D
where we repeatedly use (2 ) and (2 ).
33
Problem 5.4 (i) Since the -algebra A is also a Dynkin system, it is enough to prove (D) =
D for any Dynkin system D. By definition, (D) is the smallest Dynkin system
containing D, thus D (D). On the other hand, D is itself a Dynkin system, thus,
because of minimality, D (D).
(ii) Clearly, G H (H). Since (H) is a Dynkin system containing G, the minimality
of (G) implies that (G) (H).
(iii) Since (G) is a -algebra, it is also a Dynkin system. Since G (G) we conclude
(again, by minimality) that (G) (G).
Problem 5.5 Clearly, ({A, B}) ({A, B}) is always true.
By Theorem 5.5, ({A, B}) = ({A, B}) if {A, B} is -stable, i.e. if A = B or A = B c or if

at least one of A, B is X or .
Let us exclude these cases. If A B = , then
({A, B}) = ({A, B}) = {, A, Ac , B, B c , A B, Ac B c , X}.
If A B , then
({A, B}) = {, A, Ac , B, B c , X}
while ({A, B}) is much larger containing, for example, A B.
Problem 5.6 We prove the hint first. Let (Gj )jN G as stated in the problem, i.e. satisfying
(1) and (2), and define the sets FN = G1 . . . GN . As G A, it is clear that FN A
(but not necessarily in G...). Moreover, it is clear that FN X.
We begin with a more general assertion: For any finite union of G-sets A1 . . . AN we
have (A1 . . . AN ) = (A1 . . . AN ).
Proof. Induction Hypothesis: (A1 . . . AN ) = (A1 . . . AN ) for some N N and

any choice of A1 , . . . , AN G.
Induction Start (N = 1): is obvious.
Induction Step N N + 1: We have by the strong additivity of measures and the -

stability of G that
(A1 . . . AN AN +1 )
= ((A1 . . . AN ) AN +1 )
= (A1 . . . AN ) + (AN +1 ) ((A1 . . . AN ) AN +1 )
= (A1 . . . AN ) + (AN +1 ) ((A1 AN +1 ) . . . (AN AN +1 ))

G G
= (A1 . . . AN ) + (AN +1 ) ((A1 AN +1 ) . . . (AN AN +1 ))

34
= (A1 . . . AN AN +1 )
where we used the induction hypothesis twice, namely for the union of the N G-sets
A1 , . . . , AN as well as for the N G-sets A1 AN +1 , . . . , AN AN +1 . The induction is
complete.
In particular we see that (FN ) = (FN ), (FN ) (G1 )+. . .+(GN ) < by subadditiv-
ity, and that (think!) (GFN ) = (GFN ) for any G G (just work out the intersection,
similar to the step in the induction....). This shows that on the -stable system
G = {all finite unions of sets in G}
and coincide. Moreover, G G A so that, by assumption A = (G) (G) (A)

A, so that equality prevails in this chain of inclusions. This means that G is a generator
of A satisfying all the assumptions of Theorem 5.7, and we have reduced everything to
this situation.
Problem 5.7 Intuition: in two dimensions we have rectangles. Take I, I J . Call the lower
left corner of I a = (a1 , a2 ), the upper right corner b = (b1 , b2 ), and do the same for I using
a , b . This defines a rectangle uniquely. We are done, if I I = . If not (draw a picture!)
then we get an overlap which can be described by taking the right-and-upper-most of the
two lower left corners a, a and the left-and-lower-most of the two upper right corners b, b .
That does the trick.
Now rigorously: since I, I J , we have for suitable aj , bj , aj , bj s:
n n
I = [aj , bj ) and I = [aj , bj ).
j=1 j=1
We want to find I I , or, equivalently the condition under which x I I . Now
x = (x1 , . . . , xn ) I xj [aj , bj ) j = 1, 2, . . . , n
aj xj < bj j = 1, 2, . . . , n
and the same holds for x I (same x, but I no typo). Clearly aj xj < bj , and, at the
same time aj xj < bj holds exactly if
max(aj , aj ) xj < min(bj , bj ) j = 1, 2, . . . , n

n
x [ max(aj , aj ), min(bj , bj )).
j=1
This shows that I I is indeed a rectangle, i.e. in J . This could be an empty set (which
happens if I and I do not meet).
Problem 5.8 First we must make sure that t B is a Borel set if B B. We consider first
rectangles I = [[a, b)) J where a, b Rn . Clearly, t I = [[ta, tb)) where ta, tb are just the
35
scaled vectors. So, scaled rectangles are again rectangles, and therefore Borel sets. Now
fix t > 0 and set
Bt = {B B n t B B n }.
It is not hard to see that Bt is itself a -algebra and that J Bt B n . But then we get
B n = (J ) (Bt ) = Bt B n ,
showing that Bt = B n , i.e. scaled Borel sets are again Borel sets.
Now define a new measure (B) = n (t B) for Borel sets B B n (which is, because of
the above, well-defined). For rectangles [[a, b)) we get, in particular,
[[a, b)) = n ((t [[a, b))) = n [[ta, tb))

n
= ((tbj ) (taj ))
j=1
n
= t (bj aj )
j=1
n
= tn (bj aj )
j=1
= tn n [[a, b))
which shows that and tn n coincide on the -stable generator J of B n , hence theyre
the same everywhere. (Mind the small gap: we should make the mental step that for any
measure a positive multiple, say, c , is again a measurethis ensures that tn n is a
measure, and we need this in order to apply Theorem 5.7. Mind also that we need that
is finite on all rectangles (obvious!) and that we find rectangles increasing to Rn , e.g.
[k, k) . . . [k, k) as in the proof of Theorem 5.8(ii).)
Problem 5.9 Define (A) = 1 (A). Obviously, is again a finite measure. Moreover, since
1 (X) = X, we have
(X) = (X) < and, by assumption, (G) = (G) G G.
Thus, = on G = G {X}. Since G is a -stable generator of A containing the (trivial)

exhausting sequence X, X, X, . . ., the assertion follows from the uniqueness theorem for
measures, Theorem 5.7.
Problem 5.10 The necessity of the condition is trivial since G (G) = B, resp., H (H) = C.
Fix H H and define
(B) = P (B H) and (B) = P (B)P (H).
Obviously, and are finite measures on B having mass P (H) such that and coincide
on the -stable generator G {X} of B. Note that this generator contains the exhausting
sequence X, X, X, . . .. By the uniqueness theorem for measures, Theorem 5.7, we conclude
= on the whole of B.
36
Now fix B B and define
(C) = P (B C) and (C) = P (B)P (C).
Then the same argument as before shows that = on C and, since B B was arbitrary,
the claim follows.
37
6 Existence of measures.
Problem 6.1 We know already that B[0, ) is a -algebra (it is a trace -algebra) and, by
definition,
= {B (B) B B[0, )}
if we write B = {b b B[0, )}.
Since the structure B (B) is stable under complementation and countable unions it is
clear that is indeed a -algebra.
One possibility to extend defined on would be to take B B(R) and define B + =

B [0, ) and B = B (, 0) and to set
(B) = (B + (B + )) + ((B ) B )
which is obviously a measure. We cannot expect uniqueness of this extension since does
not generate B(R)not all Borel sets are symmetric.
Problem 6.2 By definition we have
(Q) = inf { (Bj ) (Bj )jN A, Bj Q}.

j jN
(i) Assume first that (Q) < . By the definition of the infimum we find for every > 0
a sequence (Bj )jN A such that B = j Bj Q and, because of -subadditivity,
(B ) (Q) (Bj ) (Q) .

j
Set B = k B 1/k A. Then B Q and (B) = (B) = (Q).
Now let N A and N B Q. Then
B N B (B Q) = B [(B Qc )c ] = B [B c Q]
=BQ
= Q.
So,
(Q) (N ) = (B) (N ) = (B N ) = (B N ) (Q)
which means that (N ) = 0.
39
If (Q) = , we take the exhausting sequence (Aj )jN A with Aj X and

(Aj ) < and set Qj = Aj Q for every j N. By the first part we can find sets
Cj A with Cj Qj , (Cj ) = (Qj ) and (N ) = 0 for all N A with N Cj Qj .
Without loss of generality we can assume that Cj Aj , otherwise we replace Cj by
Aj Cj . Indeed, Cj Aj Qj , Cj Aj A,
(Qj ) = (Cj ) (Aj Cj ) (Qj )
and Cj Qj (Cj Aj )Qj , i.e. we have again that all measurable N (Cj Aj )Qj
satisfy (N ) = 0.
Assume now that N C Q, C = j Cj and N A. Then Nj = N Cj A and we

have N = j Nj as well as
Nj = N Cj (C Q) Cj = Cj Q = Cj Qj .
Thus (Nj ) = 0 and, by -subadditivity, (N )

j=1 (Nj ) = 0.
= A . We know from Theorem 6.1 that

(ii) Define is a measure on A and,
because of the monotonicity of , we know that for all N A with
(N ) we have
M N (M ) (N ) =
(N ) = 0.
It remains to show that M A . Because of (6.4) we have to show that
Q X (Q) = (Q M ) + (Q M ).
Since is subadditive we find for all Q X
(Q) = ((Q M ) (Q M ))
(Q M ) + (Q M )
= (Q M )
(Q),
which means that M A .
(iii) Obviously, (X, A ,

) extends (X, A, ) since A A and
A = . In view of Problem
4.13 we have to show that
A = {A N A A, N N} ()
with N = {N X N is subset of an A-measurable null set or, alternatively,
A = {A X A, B A, A A B, (B A) = 0}. ()
We are going to use both equalities and show in () and in () (which

is enough since, cf. Problem 4.13 asserts the equality of the right-hand sides of
(), ()!).
40
: By part (ii), subsets of A-null sets are in A so that every set of the form A N
with A A and N being a subset of an A null set is in A .
: By part (i) we find for every A A some A A such that A A and

A A is an A null set. By the same argument we get B A, B (A )c and
B (A )c = B A = A B c is an A null set. Thus,
B c A A
and
A B c (A A ) (A B c ) = (A A ) (B (A )c )
which is the union of two A null sets, i.e. A B c is an A null set.
Problem 6.3 (i) A little geometry first: a solid, open disk of radius r, centre 0 is the set
Br (0) = {(x, y) R2 x2 + y 2 < r2 }. Now the n-dimensional analogue is clearly
{x Rn x21 + x22 + . . . + x2n < r2 } (including n = 1 where it reduces to an interval).
We want to inscribe a box into a ball.
n
Claim: Q (0) = [ n , n ) B2 (0). Indeed,
j=1
2 2 2
x Q (0) x21 + x22 + . . . + x2n + + ... + < (2)2
n n n
x B2 (0),
and the claim follows.
Observe that n (Q (0)) = nj=1 2

n
> 0. Now take some open set U . By translating
it we can achieve that 0 U and, as we know, this movement does not affect n (U ).
As 0 U we find some > 0 such that B (0) U , hence
n (U ) n (B (0)) (Q (0)) > 0.
(ii) For closed sets this is, in general, wrong. Trivial counterexample: the singleton {0}
is closed, it is Borel (take a countable sequence of nested rectangles, centered at 0
and going down to {0}) and the Lebesgue measure is zero.
To get strictly positive Lebesgue measure, one possibility is to have interior points,
i.e. closed sets which have non-empty interior do have positive Lebesgue measure.
Problem 6.4 (i) Without loss of generality we can assume that a < b. We have [a+ k1 , b) (a, b)
as k . Thus, by the continuity of measures, Theorem 4.4, we find (write = 1 ,
for short)
1 1
(a, b) = lim [a + , b) = lim (b a ) = b a.
k k k k
Since [a, b) = b a, too, this proves again that
({a}) = ([a, b) (a, b)) = [a, b) (a, b) = 0.
41
(ii) The hint says it all: H is contained in the union y + kN Ak for some y and we have
2 (Ak ) = (2 2k ) (2k) = 4 k2k . Using the -subadditivity and monotonicity of
measures (the Ak s are clearly not disjoint) as well as the translational invariance of
the Lebesgue measure we get

0 2 (H) 2 ( Ak ) (Ak ) = 4 k2k = C
k=1 k=1 k=1
where C is the finite (!) constant 4

k=1 k2
k
(check convergence!). As was arbitrary,
we can let it 0 and the claim follows.
n n
(iii) n-dimensional version of (i): We have I = (aj , bj ). Set Ik = [aj + k1 , bj ). Then
j=1 j=1
Ik I as k and we have (write = n , for short)
n n
1
(I) = lim (Ik ) = lim (bj aj ) = (bj aj ) .
k k j=1 k j=1
n-dimensional version of (ii): The changes are obvious: Ak = [2k , 2k )[k, k)n1
and n (Ak ) = 2n 2k k n1 . The rest stays as before, since the sum
k=1 k
n1 k
2
still converges to a finite value.
Problem 6.5 (i) All we have to show is that 1 ({x}) = 0 for any x R. But this has been
shown already in problem 6.3(i).
(ii) Take the Dirac measure: 0 . Then {0} is an atom as 0 ({0}) = 1.
(iii) Let C be countable and let {c1 , c2 , c3 , . . .} be an enumeration (could be finite, if C

is finite). Since singletons are in A, so is C as a countable union of the sets {cj }.
Using the -additivity of a measure we get
(C) = (jN {cj }) = ({cj }) = 0 = 0.

jN jN
(iv) If y1 , y2 , . . . , yN are atoms of mass P ({yj }) 1

k we find by the additivity and mono-
tonicity of measures
N N
P ({xj })
k j=1
N
= P ( {yj })
j=1
= P ({y1 , . . . , yN }) P (R) = 1
so N
k 1, i.e. N k, and the claim in the hint (about the maximal number of atoms
of given size) is shown.
(k) (k)
Now denote, as in the hint, the atoms with measure of size [ k1 , k1
1
) by y1 , . . . yN (k)
where N (k) k is their number. Since
[ k1 , k1
1
) = (0, )
kN
42
we exhaust all possible sizes for atoms.
There are at most countably many (actually: finitely many) atoms in each size range.
Since the number of size ranges is countable and since countably many countable sets
make up a countable set, we can relabel the atoms as x1 , x2 , x3 , . . . (could be finite)
and, as we have seen in exercise 4.6(ii), the set-function
= P ({xj }) xj
j
(no matter whether the sum is over a finite or countably infinite set of js) is indeed
a measure on R. But more is true: for any Borel set A
(A) = P ({xj }) xj (A)

j
= P ({xj })
j xj A
= P (A {x1 , x2 , . . .}) P (A)
showing that (A) = P (A) (A) is a positive number for each Borel set A B.
This means that B [0, ]. Let us check M1 and M2 . Using M1 , M2 for P and
(for them they are clear, as P, are measures!) we get
() = P () () = 0 0 = 0
and for a disjoint sequence (Aj )jN B we have
( Aj ) = P ( Aj ) ( Aj )
j j j
= P (Aj ) (Aj )
j j
= (P (Aj ) (Aj ))
j
= (Aj )
j
which is M2 for .
Problem 6.6 (i) Fix a sequence of numbers k > 0, k N0 such that kN0 k < . For
example we could take a geometric series with general term k = 2k . Now define
open intervals Ik = (k k , k + k ), k N0 (these are open sets!) and call their
union I = kN0 Ik . As countable union of open sets I is again open. Using the
-(sub-)additivity of = 1 we find
()
(I) = ( Ik ) (Ik ) = 2k = 2 k < .
kN0 kN0 kN0 kN0
By 6.4(i), (I) > 0.
43
Note that in step () equality holds (i.e. we would use -additivity rather than -
subadditivity) if the Ik are pairwise disjoint. This happens, if all k < 1
2 (think!), but
to be on the safe side and in order not to have to worry about such details we use
sub-additivity.
(ii) Take the open interior of the sets Ak , k N, from the hint to 6.4(ii). That is, take
the open rectangles Bk = (2k , 2k ) (k, k), k N, (we choose = 1 since we are
after finiteness and not necessarily smallness). That these are open sets will be seen
below. Now set B = kN Bk and observe that the union of open sets is always open.
B is also unbounded and it is geometrically clear that B is connected as it is some
kind of lozenge-shaped staircase (draw a picture!) around the y-axis. Finally, by
-subadditivity and using 6.4(ii) we get
2 (B) = 2 ( Bk ) 2 (Bk )
kN kN
= 2 2k 2 k
kN
= 4 k 2k < .
kN
It remains to check that an open rectangle is an open set. For this take any open
rectangle R = (a, b) (c, d) and pick (x, y) R. Then we know that a < x < b and
c < y < d and since we have strict inequalities, we have that the smallest distance of
this point to any of the four boundaries (draw a picture!) h = min{a x, b x, c
y, d y} > 0. This means that a square around (x, y) with side-length 2h is inside R
and what were going to do is to inscribe into this virtual square an open disk with
radius h and centre (x, y). Since the circle is again in R, we are done. The equation
for this disk is
(x , y ) Bh (x, y) (x x )2 + (y y )2 < h2
Thus,

x x x x 2 + y y 2 < h

and y y x x 2 + y y 2 < h
i.e. x h < x < x + h and y h < y < y + h or (x , y ) (x h, x + h) (y h, y + h),

which means that (x , y ) is in the rectangle of sidelength 2h centered at (x, y). since
(x , y ) was an arbitrary point of Bh (x, y), we are done.
(iii) No, this is impossible. Since we are in one dimension, connectedness forces us to go
between points in a straight, uninterrupted line. Since the set is unbounded, this
means that we must have a line of the sort (a, ) or (, b) in our set and in both
cases Lebesgue measure is infinite. In all dimensions n > 1, see part (ii) for two
dimensions, we can, however, construct connected, unbounded open sets with finite
Lebesgue measure.
44
Problem 6.7 Fix > 0 and let {qj }jN be an enumeration of Q [0, 1]. Then
U = U = (qj 2j1 , qj + 2j1 ) [0, 1]

jN
is a dense open set in [0, 1] and, because of -subadditivity,

(U ) (qj 2j1 , qj + 2j1 ) = j
= .
jN jN 2
Problem 6.8 Assume first that for every > 0 there is some open set U N such that (U ) .
Then
(N ) (U ) > 0,
which means that (N ) = 0.
Conversely, let (N ) = inf { j (Uj ) Uj O, jN Uj N }. Since for the Borel set N

we have (N ) = (N ) = 0, the definition of the infimum guarantees that for every > 0
there is a sequence of open sets (Uj )jN covering N , i.e. such that U = j Uj N . Since
U is again open we find because of -subadditivity
(N ) (U ) = ( Uj ) (Uj ) .
j j
Attention: A construction along the lines of Problem 3.12, hint to part (ii), using open
sets U = N + B (0) is, in general not successful:
it is not clear that U has finite Lebesgue measure (o.k. one can overcome this by
considering N [k, k] and then letting k ...)
U N
and not N (unless N is closed, of course). If, say, N is a dense set of [0, 1],
this approach leads nowhere.
Problem 6.9 Observe that the sets Ck =

j=k Aj , k N, decrease as k we admit less and
less sets in the union, i.e. the union becomes smaller. Since P is a probability measure,
P (Ck ) 1 and therefore Theorem 4.4(iii) applies and shows that

P ( Aj ) = P ( Ck ) = lim P (Ck ).
k=1 j=k k=1 k
On the other hand, we can use -subadditivity of the measure P to get

P (Ck ) = P ( Aj )
j=k P (Aj )
j=k
but this is the tail of the convergent (!) sum

j=1 P (Aj ) and, as such, it goes to zero as
k . Putting these bits together, we see

P ( Aj ) = lim P (Ck ) lim P (Aj ) = 0,
k=1 j=k k k j=k
45
Problem 6.10 (i) We can work out the optimal A-cover of (a, b):
Case 1: a, b [0, 1). Then [0, 1) is the best possible cover of (a, b), thus (a, b) =
[0, 1) = 12 .
Case 2: a, b [1, 2). Then [1, 2) is the best possible cover of (a, b), thus (a, b) =
[1, 2) = 12 .
Case 3: a [0, 1), b [1, 2). Then [0, 1) [1, 2) is the best possible cover of (a, b),
thus (a, b) = [0, 1) + [1, 2) = 1.
And in the case of a singleton {a} the best possible cover is always either [0, 1) or
[1, 2) so that ({a}) = 1
2 for all a.
(ii) Assume that (0, 1) A . Since A A , we have [0, 1) A , hence {0} = [0, 1)(0, 1)
A . Since (0, 1) = ({0}) = 21 , and since is a measure on A (cf. step 4 in the
proof of Theorem 6.1), we get
1 1 1
= [0, 1) = [0, 1) = (0, 1) + {0} = + = 1
2 2 2
leading to a contradiction. Thus neither (0, 1) nor {0} are elements of A .
Problem 6.11 Since A A , the only interesting sets (to which one could extend ) are those
B R where both B and B c are uncountable. By definition,
(B) = inf { (Aj ) Aj A, Aj B}.

j j
The infimum is obviously attained for Aj = R, so that (B) = (B c ) = 1. On the

other hand, since is necessarily additive on A , the assumption that B A leads to a
contradiction:
1 = (R) = (R) = (B) + (B c ) = 2.
Thus, A = A .
46
7 Measurable mappings.
Problem 7.1 We have x1 (z) = z + x. According to Lemma 7.2 we have to check that
x1 ([a, b)) B n [a, b) J
since the rectangles J generate B n . Clearly,
x1 ([a, b)) = [a, b) + x = [a + x, b + x) J B n ,
Problem 7.2 We had = {A X T 1 (A ) A} where A was a -algebra of subsets of X.

Let us check the properties (1 )(3 ).
(1 ) Take X . Then T 1 () = A, hence .

c
(2 ) Take any B . Then T 1 (B) A and therefore T 1 (B c ) = (T 1 (B)) A since all
set-operations interchange with inverse maps and since A is a -algebra. This shows
that B c .
(3 ) Take any sequence (Bj )jN . Then, using again the fact that A is a -algebra,
T 1 (j Bj ) = j T 1 (Bj ) A which proves that j Bj .
Problem 7.3 (i) First of all we remark that Ti1 (Ai ) is itself a -algebra, cf. Example 3.3(vii).
If C is a -algebra of subsets of X such that Ti (X, C) (Xi , Ai ) becomes measur-

able, we know from the very definition that T 1 (Ai ) C. From this, however, it is
clear that T 1 (Ai ) is the minimal -algebra that renders T measurable.
(ii) From part (i) we know that (Ti , i I) necessarily contains Ti1 (Ai ) for every i I.
Since i Ti1 (Ai ) is, in general, not a -algebra, we have ( i Ti1 (Ai )) (Ti , i
I). On the other hand, each Ti is, because of Ti1 (Ai ) i Ti1 (Ai ) (Ti , i I)
measurable w.r.t. ( i Ti1 (Ai )) and this proves the claim.
Problem 7.4 We have to show that
f (F, F) (X, (Ti , i I)) measurable

i I Ti f (F, F) (Xi , Ai ) measurable.
47
Now
i I (Ti f )1 (Ai ) F i I f 1 (Ti1 (Ai )) F
f 1 ( iI Ti1 (Ai )) F
()
[f 1 ( iI Ti1 (Ai ))] F
()
f 1 ([ iI Ti1 (Ai )]) F.
Only (*) and (**) are not immediately clear. The direction in (*) is trivial, while
follows if we observe that the right-hand side, F, is a -algebra. The equivalence
(**) is another case of Problem 7.8 (see there for the solution!).
Problem 7.5 Using the notation of the foregoing Problem 7.4 we put I = {1, 2, . . . , m}, Tj =
j Rm R, j (x1 , . . . , xm ) = xj is the coordinate projection, Aj = B(R). Since each j
is continuous, we have (1 , . . . , m ) B(Rm ) so that Problem 7.4 applies and proves
f is B(Rm )-measurable
fj = j f is B(R)-measurable for all j = 1, 2, . . . , m.
Remark. We will see, in fact, in Chapter 13 (in particular in Theorem 13.10) that we
have the equality (1 , . . . , m ) = B(Rm ).
Problem 7.6 In general the direct image T (A) of a -algebra is not any longer a -algebra. (1 )
and (3 ) hold, but (2 ) will, in general, fail. Here is an example: Take X = X = N, take
any -algebra A other than {, N} in N, and let T N N, T (j) = 1 be the constant map.
Then T () = but T (A) = {1} whenever A . Thus, {1} = T (Ac ) [T (A)]c = N {1}
but equality would be needed if T (A) were a -algebra. This means that 2 fails.
Necessary and sufficient for T (A) to be a -algebra is, clearly, that T 1 is a measurable
map T 1 X X.
Problem 7.7 Consider for t > 0 the dilation mt Rn Rn , x t x. Since mt is continuous, it

is Borel measurable. Moreover, m1
t = m1/t and so
t B = m1
1/t (B)
which shows that n (t B) = n m1

1/t (B) = m1/t ( )(B) is actually an image measure
n
n
of n . Now show the formula first for rectangles B = [aj , bj ) (as in Problem 5.8) and
j=1
deduce the statement from the uniqueness theorem for measures.
Problem 7.8 We have
T 1 (G) T 1 ((G)) (T 1 (G)) T 1 ((G)).

is itself a -algebra
48
For the converse consider T (X, (T 1 (G))) (Y, (G)). By the very choice of the -
algebras and since T 1 (G) (T 1 (G)) we find that T is (T 1 (G))/(G) measurable
mind that we only have to check measurability at a generator (here: G) in the image
region. Thus,
T 1 ((G)) (T 1 (G)).
Problem 7.9 (i) Note the misprint: we need to assume that [n, n) < for all n N.
Monotonicity: If x 0 y, then F (x) 0 F (y).
If 0 < x y, we have [0, x) [0, y) and so 0 F (x) = [0, x) [0, y) = F (y).
If x y < 0, we have [y, 0) [x, 0) and so 0 F (y) = [y, 0) [x, 0) = F (x),

i.e. F (x) F (y) 0.
Left-continuity: Let us deal with the case x 0 only, the case x < 0 is analogous (and
even easier). Assume first that x > 0. Take any sequence xk < x and xk x as k .
Without loss of generality we can assume that 0 < xk < x. Then [0, xk ) [0, x) and
using Theorem 4.4(iii) implies
lim F (xk ) = lim [0, xk ) = [0, x) = F (x).

k k
If x = 0 we must take a sequence xk < 0 and we have then [xk , 0) [0, 0) = . Again
by Theorem 4.4, now (iii), we get
lim F (xk ) = lim [xk , 0) = () = 0 = F (0).

k k
which shows left-continuity at this point, too.
We remark that, since for a sequence yk y, yk > y we have [0, yk ) [0, y], and not
[0, y), we cannot expect right-continuity in general.
(ii) Since J = {[a, b), a b} is a semi-ring (cf. Proposition 6.4) it is enough to check that
F is a premeasure on J . This again amounts to showing (M1 ) and (M2 ) relative
to J (mind you: F is not a measure as J is not a -algebra....). We do this in the
equivalent form of Theorem 4.4, i.e. we prove (i), (ii) and (iii) of Theorem 4.4:
(i) F () = F [a, a) = F (a) F (a) = 0 for any a.
(ii) Let a b c so that [a, b), [b, c) J are disjoint sets and [a, c) = [a, b) [b, c) J
(the latter is crucial). Then we have
F [a, b) + F [b, c) = F (b) F (a) + F (c) F (b)

= F (c) F (a)
= F [a, c)
= F ([a, b) [b, c)).
49
(iii) (Sufficient since F is finite for every set [a, b)). Now take a sequence of intervals
[ak , bk ) which decreases towards some [a, b) J . This means that ak a, ak a
and bk b, bk b because the intervals are nested (gives increasing-decreasing
sequences). If bk > b for infinitely many k, this would mean that [ak , bk )
[a, b] / J since b [ak , bk ) for all k. Since we are only interested in sequences
whose limits stay in J , the sequence bk must reach b after finitely many steps
and stay there to give [a, b). Thus, we may assume directly that we have only
[ak , b) [a, b) with ak a, ak a. But then we can use left-continuity and get
lim F [ak , b) = lim (F (b) F (ak )) = F (b) F (a)

k k
= F [a, b).
Note that F takes on only positive values because F increases.
This means that we find at least one extension. Uniqueness follows since F [k, k) =
F (k) F (k) < and [k, k) R.
(iii) Now let be a measure with [n, n) < . The latter means that the function
F (x), as defined in part (i), is finite for every x R. Now take this F and define,
as in (ii) a (uniquely defined) measure F . Let us see that = F . For this, it is
enough to show equality on the sets of type [a, b) (since such sets generate the Borel
sets and the uniqueness theorem applies....)
If 0 a b,
F [a, b) = F (b) F (a) = [0, b) [0, a)

= ([0, b) [0, a))
= [a, b)
If a b 0,
F [a, b) = F (b) F (a) = [b, 0) ([a, 0))

= [a, 0) [b, 0)
= ([a, 0) [b, 0))
= [a, b)
If a 0 b,
F [a, b) = F (b) F (a) = [0, b)) ([a, 0))

= [a, 0)) + [0, b)
= ([a, 0) [0, b))
= [a, b)
50
(iv) F R R with F (x) = x, since [a, b) = b a = F (b) F (a).

0, x 0
(v) F R R, with, say, F (x) = = 1(0,) (x) since 0 [a, b) = 0 whenever

1, x > 0

a, b < 0 or a, b > 0. This means that F must be constant on (, 0) and (0, ) If
a 0 < b we have, however, 0 [a, b) = 1 which indicates that F (x) must jump by 1 at
the point 0. Given the fact that F must be left-continuous, it is clear that it has, in
principle, the above form. The only ambiguity is, that if F (x) does the job, so does
c + F (x) for any constant c R.
(vi) Assume that F is continuous at the point x. Then
({x}) = ( [x, x + k1 ) )
kN
4.4
= lim ([x, x + k1 ))
k
def
= lim (F (x + k1 ) F (x))
k
= lim F (x + k1 ) F (x)
k
()
= F (x) F (x) = 0
where we used (right-)continuity of F at x in the step marked ().
Now, let conversely ({x}) = 0. A similar calculation as above shows, that for every
sequence k > 0 with k
F (x+) F (x) = lim F (x + k ) F (x)

k
def
= lim [x, x + k )
k
4.4
= ( [x, x + k ))
kN
= ({x}) = 0
which means that F (x) = F (x+) (x+ indicates the right limit), i.e. F is right-
continuous at x, hence continuous, as F is left-continuous anyway.
(vii) Then hint is indeed already the proof. Almost, that is... Let be some measure as
specified in the problem. From part (iii) we know that the Stieltjes function F = F
then satisfies
[a, b) = F (b) F (a) = 1 [F (a), F (b))

(#)
= 1 (F ([a, b)))
(##)
= 1 F ([a, b)).
The crunching points in this argument are the steps (#) and (##).
51
(#) This is o.k. since F was continuous, and the intermediate value theorem for
continuous functions tells us that intervals are mapped to intervals. So, no
problem here, just a little thinking needed.
(##) This is more subtle. We have defined image measures only for inverse maps,
i.e. for expressions of the type 1 G1 where G was measurable. So our job is
to see that F can be obtained in the form F = G1 where G is measurable. In
other words, we have to invert F . The problem is that we need to understand
that, if F (x) is flat on some interval (a, b) inversion becomes a problem (since
then F 1 has a jumphorizontals become verticals in inversions, as inverting is
somehow the mirror-image w.r.t. the 45-degree line in the coordinate system.).
So, if there are no flat bits, then this means that F is strictly increasing, and it
is clear that G exists and is even continuous there.
If we have a flat bit, lets say exactly if x [a, b] and call F (x) = F (a) = F (b) = C
for those x; clearly, F 1 jumps at C and we must see to it that we take a version
of F 1 , say one which makes F 1 left-continuous at Cnote that we could assign
any value from [a, b] to F 1 (C)which is accomplished by setting F 1 (C) = a.
(Draw a graph to illustrate this!)
There is A canonical expression for such a generalized left-continuous inverse of
an increasing function (which may have jumps and flat bitsjumps of F become
just flat bits in the graph of F 1 , think!) and this is:
G(y) = inf{x F (x) y}
Let us check measurability:
y0 {G } G(y0 )
def
inf{F y0 }
()
F () y0
y0 [F (), ).
Since F is monotonically increasing, we find also in step (), hence
{G } = [F (), ) B(R)
which shows that G is measurable. Even more: it shows that G1 (x) = inf{G
} = F (x). Thus, 1 F = 1 G1 = is indeed an image measure of 1 .
(viii) We have F (x) = F0 (x) = 1(0,) (x) and its left-continuous inverse G(y) in the sense
of part (vii) is given by

+, y>1

G(y) = 0, 0<y1.

,
y0
52
This function is clearly measurable (use B to accommodate ) and so the claim

holds in this case. Observe that in this case F is not any longer continuous but only
left-continuous.
Problem 7.10 (i) We find the following picture:

E0 = [0, 1]
E1 = [0, 13 ] [ 32 , 1]
E2 = [0, 19 ] [ 92 , 13 ] [ 23 , 79 ] [ 98 , 1]
1 2 1
E3 = [0, 27 ] [ 27 , 9 ] [ 29 , 27
7 8 1
] [ 27 , 3]
[ 23 , 27
19
] [ 20 , 7 ] [ 89 , 27
27 9
25
] [ 26
27
, 1]
(ii) Each En is a finite union of 2n closed and bounded intervals. As such, En is itself
a closed and bounded set, hence compact. The intersection of closed and bounded
sets is again closed and bounded, so compact. This shows that C is compact. That
C is non-empty follows from the intersection principle: if one has a nested sequence
of non-empty compact sets, their intersection is not empty. (This is sometimes
formulated in a somewhat stronger form and called: finite intersection property.
The general version is then: Let (Kn )nN be a sequence of compact sets such that
each finite sub-family has non-void intersection, then n Kn ). This is an obvious
generalization of the interval principle: nested non-void closed and bounded intervals
have a non-void intersection.
(iii) At step n we remove open middle-third intervals of length 3n . To be precise, we
partition En1 in pieces of length 3n and remove every other interval. The same
effect is obtained if we partition [0, ) in pieces of length 3n and remove every other
piece. Call the taken out pieces Fn and set En = En1 Fn , i.e. we remove from En1
even pieces which were already removed in previous steps. It is clear that Fn exactly
3n , 3n ), k N0 which comprises exactly every other
consists of sets of the form ( 3k+1 3k+2
set of length 3n . Since we do this for every n, the set C is disjoint to the union of
these intervals over k N0 and n N.
(iv) Since En consists of 2n intervals I1 . . . I2n , each of which has length 3n (prove
this by a trivial induction argument!), we get
2 n
(En ) = (I1 ) + . . . + (I2n ) = 2n 3n = ( )
3
where we also used (somewhat pedantically) that
[a, b] = ([a, b) {b}) = [a, b) + {b} = b a + 0 = b a.
Now using Theorem 4.4 we conclude that (C) = inf n (En ) = 0.

(v) Fix > 0 and choose n so big that 3n < . Then En consists of 2n disjoint intervals
of length 3n < and cannot possibly contain a ball of radius . Since C En , the
53
same applies to C. Since was arbitrary, we are done. (Remark: an open ball in R
with centre x is obviously an open interval with midpoint x, i.e. (x , x + ).)
(vi) Fix n and let k = 0, 1, 2, . . . , 3n1 1. We saw in (c) that at step n we remove the
intervals Fn , i.e. the intervals of the form
3k + 1 3k + 2
( , ) = (0. . . . 1 000 . . . , 0. . . . 2 000 . . . )
3n 3n
n n
where we used the ternary representation of x. These are exactly the numbers in [0, 1]
whose ternary expansion has a 1 at the nth digit. As 0.. . .1 = 0.. . .022222 . . .
has two representations, the left endpoint stays in. Since we do this for every step
n N, the claim follows.
(vii) Take t C with ternary representation t = 0.t1 t2 t3 . . . tj . . ., tj {0, 2} and map it to

tj tj
the binary number b = 0. t21 t22 t23 . . . 2 with digits bj = 2 {0, 1}. This gives a bijection
between C and [0, 1], i.e. both have as infinitely many points, i.e. #C = #[0, 1].
Despite of that
(C) = 0 1 = ([0, 1])
which is, by the way, another proof for the fact that -additivity for the Lebesgue
measure does not extend to general uncountable unions.
Problem 7.11 One direction is easy: if f = g T with g Y R being measurable, we have
f 1 (B(R)) = (g T )1 (B(R)) = T 1 (g 1 (B(R))) T 1 (A) = (T ).
Conversely, if f is (T )-measurable, then whenever T (x) = T (x ), we have f (x) = f (x );

for if not, let B be a Borel set in R with f (x) B and f (x ) / B. Then f 1 (B) = T 1 (C)
for some C A, with T (x) C but T (x ) / Cwhich is impossible. Thus, f = g T for
some function g from the range T (X) of T . But, by assumption, T (X) = Y .
For any Borel set S R, T 1 g 1 (S) = f 1 (S) = T 1 (A) for some suitable A A, so
A = g 1 (S) proving the measurability of g.
Remark. Originally, I had in mind the above solution (taken from Dudleys book [14],
Theorem 4.2.8), but recently I found a much simpler solution (below) which makes the
whole Remark following the statement of Problem 7.11 obsolete; moreover, it is not any
longer needed to have T surjective. However, this requires that you read through Theorem
8.8 from the next chapter.
Alternative solution: One direction is easy: if f = gT with g Y R being measurable,

we have
f 1 (B(R)) = (g T )1 (B(R)) = T 1 (g 1 (B(R))) T 1 (A) = (T ).
For the converse assume in ...
54
Step 1: ...first that f = 1B is a step function consisting of a single step. Then
1B (T )-measurable B (T )
A A B = {T A} = T 1 (A)
1B = 1A T
so that g = 1A Y R does the job.
j=1 j 1Bj then, by Step 1, g = j=1 j 1Aj with the obvious notation
Step 2: If f = N N
Bj = {T Aj }, Aj A suitable, and f = g T and g Y R.
Step 3: If f 0 is measurable, then we use Theorem 8.8 which says that we can approx-
imate f as an increasing limit of (T )-measurable (!) elementary functions (have a look
at the proof of 8.8!), say f = supj fj and each fj is of the form of the functions from Step
2. Thus, there are again functions gj such that fj = gj T and we get
f = lim sup fj = lim sup gj T = (lim sup gj ) T

j j j
which means that g = lim supj gj Y R does the job.
Step 4: If f is just measurable, consider positive and negative parts f = f + f and

construct, according to Step 3, g Y R such that g T = f . Then g = g + g does
the trick.
55
8 Measurable functions.
Problem 8.1 We remark, first of all, that {u } = u1 ([x, )) and, similarly, for the other
sets. Now assume that {u } A for all . Then
{u > } = u1 ((, )) = u1 ( [ + k1 , ) )
kN
1
= u ([ + k1 , ))
kN
= {u + k1 } A
kN
by assumption A
since A is a -algebra.
Conversely, assume that {u > } A for all . Then
{u } = u1 ([, )) = u1 ( ( k1 , ) )
kN
1
= u (( k1 , ))
kN
= {u > k1 } A.
kN
by assumption A
since A is a -algebra. Finally, as
{u > }c = {u } and {u }c = {u < }
we have that {u > } A if, and only if, {u } A and the same holds for the sets
{u }, {u < }.
Problem 8.2 Recall that B B if, and only if B = B C where B B and C is any of the
following sets: , {}, {}, {, }. Using the fact that B is a -algebra and using
this notation (that is: B-sets
carry an asterisk ) we see
(1 ) Take B = B, C = to see that = B;

(2 ) Let B B. Then (complements are to be taken in B
(B )c = (B C)c
= Bc C c
57
= (R B) (R C)
= (R B {, +}) (R C)
= ((R B) (R C)) ({, +} (R C))
= (R B) ({, +} (R C))
which is again of the type B-set union a set of the list , {}, {}, {, }, hence
it is in B.

(3 ) Let Bn B and Bn = Bn Cn . Then
B = Bn = (Bn Cn ) = Bn Cn = B C
nN nN nN nN
with B B and C from the list , {}, {}, {, }, hence B B.

A problem is the notation B = B(R).

While the left-hand side can easily be defined by
(8.5), B(R)
has a well-defined meaning as the (topological) Borel -algebra over the set R,

which is defined via the open sets in R.
i.e. the -algebra in R To describe the open sets
(R)
of R we use require, that each point x U O(R)
admits an open neighbourhood
B(x) inside U . If x , we take B(x) as the usual open -interval around x with > 0
sufficiently small. If x = we take half-lines [, a) or (b, +] respectively with a, b
sufficiently large. Thus, O(R)
adds to O(R) a few extra sets and open sets are therefore
of the form U = U C with U O(R) and C being of the form [, a) or (b, +] or
or R or unions thereof.
Thus, O(R) = R O(R)

and therefore
B(R) = R B(R)

(this time in the proper topological sense).
Problem 8.3 (i) Notice that the indicator functions 1A and 1Ac are measurable. By Corollary
8.10 sums and products of measurable functions are again measurable. Since h(x)
can be written in the form h(x) = 1A (x)f (x) + 1Ac (x)g(x), the claim follows.
(ii) The condition fj Aj Ak = fk Aj Ak just guarantees that f (x) is well-defined if we set

f (x) = fj (x) for x Aj . Using j Aj = X we find for B B(R)
f 1 (B) = Aj f 1 (B) = Aj fj1 (B) A.

jN jN
A
An alternative solution would be to make the Aj s disjoint, e.g. by setting C1 = A1 ,

Ck = Ak (A1 Ak1 ). Then
f = 1Cj f = 1Cj fj
j j
and the claim follows from Corollaries 8.10 and 8.9.
58
Problem 8.4 Since 1B is B-measurable if, and only if, B B the claim follows by taking B B
such that B / A (this is possible as B A.
Problem 8.5 By definition, f E if it is a step-function of the form f = N

j=0 aj 1Aj with some
aj R and Aj A. Since
f + = aj 1Aj and f = aj 1Aj ,

0jN 0jN
aj 0 aj 0
f are again of this form and therefore simple functions.
The converse is also true since ff+ f see (8.9) or Problem 8.6and since sums and
differences of simple functions are again simple.
Problem 8.6 By definition
u+ (x) = max{u(x), 0} and u (x) = min{u(x), 0}.
Now the claim follows from the elementary identities that for any two numbers a, b R
a + 0 = max{a, 0} + min{a, 0} and a = max{a, 0} min{a, 0}
which are easily verified by considering all possible cases a 0 resp. a 0.
Problem 8.7 Assume that 0 u(x) c for all x and some constant c. Choose j N such that
j > c. Then the procedure used to approximate u in the proof of Theorem 8.8see page
62, line 9 from aboveguarantees that fj (x) u(x) 2j for all values of x; note that
the case u j does not occur! This means that sup fj u+ 2j , i.e. we have uniform
convergence.
The general case is now obtained by considering positive and negative parts u = u+ u
which are bounded since u u c.
Problem 8.8 If we show that {u > } is an open set, it is also a Borel set, hence u is measurable.
Let us first understand what openness means: {u > } is open means that for x {u > }
we find some (symmetric) neighbourhood (a ball) of the type (x h, x + h) {u > }.
What does this mean? Obviously, that u(y) > for any y (x h, x + h) and, in other
words, u(y) > whenever y is such that x y < h. And this is the hint of how to use
continuity: we use it in order to find the value of h.
u being continuous at x means that
> 0 > 0 y x y < u(x) u(y) < .
Since u(x) > we know that for a sufficiently small we still have u(x) + . Take this
and find the corresponding . Then
u(x) u(y) u(x) u(y) < x y <
59
and since + u(x) we get
+ u(y) < x y <
i.e. u(y) > for y such that x y < . This means, however, that h = does the job.
Problem 8.9 The minimum/maximum of two numbers a, b R can be written in the form
1
min{a, b} = (a + b a b)
2
1
max{a, b} = (a + b + a b)
2
which shows that we can write min{x, 0} and max{x, 0} as a combination of continuous
functions. As such they are again continuous, hence measurable. Thus,
u+ = max{u, 0}, u = min{u, 0}
are compositions of measurable functions, hence measurable.
Problem 8.10 The fj are step-functions where the bases of the steps are the sets Ajk and Aj .
Since they are of the form, e.g. {k2j u < (k + 1)2j } = {k2j u} {u < (k + 1)2j }, it
is clear that they are not only in A but in (u).
Problem 8.11 Corollary 8.11 If u are measurable, it is clear that u = u+ u is measurable

since differences of measurable functions are measurable.
(For the converse we could use the previous Problem 8.10, but we give an alter-
native proof...) Conversely, let u be measurable. Then sn u (this is short for:
limn sn (x) = u(x) and this is an increasing limit) for some sequence of simple
functions sn . Now it is clear that s+n u+ , and s+n is simple, i.e. u+ is measurable.
As u = u+ u we conclude that u = u+ u is again measurable as difference of two
measurable functions. (Notice that in no case can occur!)
Corollary 8.12 This is trivial if the difference u v is defined. In this case it is measurable
as difference of measurable functions, so
{u < v} = {0 < u v}
etc. is measurable.
Let us be a bit more careful and consider the case where we could encounter ex-
pressions of the type . Since sn u for simple functions (they are always
R-valued...) we get
()
{u v} = {sup sn u} = {sn u} = {0 u sn }
n n n
and the latter is a union of measurable sets, hence measurable. Now {u < v} =
{u v}c and we get measurability after switching the roles of u and v. Finally
{u = v} = {u v} {u v} and {u v} = {u = v}c .
60
Let me stress the importance of in () above: we use here
x {sup sn u} sup sn (x) u(x)

n n
()
sn (x) u(x) n
x {sn u}
and this would be incorrect if we had had <, since the argument would break down
at () (only one implication would be valid: ).
Problem 8.12 If u is differentiable, it is continuous, hence measurable. Moreover, since u

exists, we can write it in the form
u(x + k1 ) u(x)
u (x) = lim 1
k
k
i.e. as limit of measurable functions. Thus, u is also measurable.
Problem 8.13 It is sometimes necessary to distinguish between domain and range. We use the
subscript x to signal the domain, the subscript y for the range.
(i) Since f Rx Ry is f (x) = x, the inverse function is clearly f 1 (y) = y. So if we

take any Borel set B B(Ry ) we get B = f 1 (B) Rx . Since, as we have seen,
(f ) = f 1 (B(Ry )), the above argument shows that f 1 (B(Ry )) = B(Rx ), hence
(f ) = B(Rx ).

(ii) The inverse map of g(x) = x2 is multi-valued, i.e. if y = x2 , then y = x. So

g 1 [0, ) R, g 1 (y) = y. Let us take some B B(Ry ). Since g 1 is only
defined for positive numbers (squares yield positive numbers only!) we have that

g 1 (B) = g 1 (B[0, )) = B [0, )( B [0, )) (where we used the obvious

notation A = { a a A} and A = {a a A} whenever A is a set). This shows
that

(g) = { B ( B) B B, B [0, )}

= { B ( B) B [0, ) B}
where we used the notation of trace -algebras in the latter identity.
(It is an instructive exercise to check that (g) is indeed a -algebra. This is, of
course, clear from the general theory since (g) = g 1 ([0, ) B), i.e. it is the pre-
image of the trace -algebra and pre-images of -algebras are always -algebras.
(iii) A very similar calculation as in part (ii) shows that
(h) = {B (B) B B, B [0, )}

= {B (B) B [0, ) B}.
61
(iv) As warm-up we follow the hint. The set {(x, y) x + y = } is the line y = x in
the x-y-plane, i.e. a line with slope 1 and shift . So {(x, y) x + y } would be
the points above this line and {(x, y) x + y } = {(x, y) x + y [, ]} would
be the points in the strip which has the lines y = x and y = x as boundaries.
More general, take a Borel set B B(R) and observe that
F 1 (B) = {(x, y) x + y B}.
This set is, in an abuse of notation, y = B x, i.e. these are all lines with slope 1 (135
degrees) and every possible shift from the set Bit gives a kind of stripe-pattern. To
sum up:
(F ) = {all 135-degree diagonal stripes in R2 with base B B(R)}.
(v) Again follow the hint to see that {(x, y) x2 + y 2 = r} is a circle, radius r, centre
(0, 0). So {(x, y) x2 + y 2 r} is the solid disk, radius r, centre (0, 0) and {(x, y)
R x2 + y 2 r} = {(x, y) x2 + y 2 [r, R]} is the annulus with exterior radius R and
interior radius r about (0, 0).
More general, take a Borel set B [0, ), B B(R), i.e. B [0, ) B(R) (negative
radii dont make sense!) and observe that the set {(x, y) x2 + y 2 B} gives a ring-
pattern which is supported by the set B (i.e. we take all circles passing through
B...). To sum up:
(G) ={a set consists of all circles in R2 about (0, 0)

passing through B [0, ) B(R)}.
Problem 8.14 Assume first that u is injective. This means that every point in the range u(R)
comes exactly from one uniquely defined x R. This can be expressed by saying that
{x} = u1 ({u(x)}) but the singleton {u(x)} is a Borel set in the range, so {x} (u)
as (u) = u1 (u(R) B).
Conversely, assume that for each x we have {x} (u). Fix an x0 and call u(x0 ) = .
Since u is measurable, the set {u = } = {x u(x) = } is measurable and, clearly,
{x0 } {u = }. But if we had another x0 x1 {u = } this would mean that we could
never produce {x0 } on its own as a pre-image of some set, but we must be able to do so
as {x0 } (u), by assumption. Thus, x1 = x0 . To sum up, we have shown that {u = }
consists of one point only, i.e. we have shown that u(x0 ) = u(x1 ) implies x0 = x1 which is
just injectivity.
Problem 8.15 Clearly u R [0, ). So lets take I = (a, b) [0, ). Then u1 ((a, b)) =
(b, a) (a, b). This shows that for = u1
(a, b) = u1 ((a, b)) = ((b, a) (a, b)) = (b, a) + (a, b)
62
= (a (b)) + (b a) = 2(b a) = 2((a, b)).
This shows that = 2 if we allow only intervals from [0, ), i.e.
(I) = 2(I [0, )) for any interval I R.
Since a measure on the Borel sets is completely described by (either: open or closed or
half-open or half-closed) intervals (the intervals generate the Borel sets!), we can invoke
the uniqueness theorem to guarantee that the above equality holds for all Borel sets.
Problem 8.16 clear, since u(x 2) is a combination of the measurable shift 2 and the
measurable function u.
this is trivial since u eu is a continuous function, as such it is measurable and
combinations of measurable functions are again measurable.
this is trivial since u sin(u + 8) is a continuous function, as such it is measurable
and combinations of measurable functions are again measurable.
iterate Problem 8.12
obviously, sgn x = (1) 1(,0) (x) + 0 1{0} (x) + 1 1(0,) (x), i.e. a measurable
function. Using the first example, we see now that sgn u(x 7) is a combination of
three measurable functions.
Problem 8.17 Let A R be such that A / B. Then it is clear that u(x) = 1A (x) 1Ac (x)
is NOT measurable (take, e.g. A = {f = 1} which should be measurable for measurable
functions), but clearly, f (x) = 1 and as constant function this IS measurable.
Problem 8.18 We want to show that the sets {u } are Borel sets. We will even show that
they are intervals, hence Borel sets. Imagine the graph of an increasing function and the
line y = cutting through. Essentially we have three scenarios: the cut happens at a
point where (a) u is continuous and strictly increasing or (b) u is flat or (c) u jumpsi.e.
has a gap; these three cases are shown in the following pictures:
6 6

- -
b a c b a
6

-
b a
63
From the three pictures it is clear that we get in any case an interval for the sub-level sets
{u } where is some level (in the pics = or = ), you can read off the intervals on
the abscissa where the dotted lines cross the abscissa.
Now lets look at the additional conditions: First the intuition: From the first picture,
the continuous and strictly increasing case, it is clear that we can produce any interval
(, b] to (, a] by looking at {u } to {u } my moving up the -line to level .
The point is here that we get all intervals, so we get a generator of the Borel sets, so we
should get all Borel sets.
The second picture is bad: the level set {u } is (, b] and all level sets below will
only come up to the point (, c], so there is no chance to get any set contained in (c, b),
i.e. we cannot get all Borel sets.
The third picture is good again, because the vertical jump does not hurt. The only
problem is whether {u } is (, b] or (, b) which essentially depends on the
property of the graph whether u(b) = or not, but this is not so relevant here, we just
must make sure that we can get more or less all intervals. The reason, really, is that jumps
as we described them here can only happen countably often, so this problem occurs only
countably often, and we can overcome it therefore.
So the point is: we must disallow flat bits, i.e. (u) is the Borel -algebra if, and only, if
u is strictly increasing, i.e. if, and only if, u is injective. (Note that this would have been
clear already from Problem 8.14, but our approach here is much more intuitive.)
64
9 Integration of positive functions.
Problem 9.1 We know that for any two simple functions f, g E+ we have I (f + g) = I (f ) +
I (g) (=additivity), and this is easily extended to finitely many, say, m different positive
simple functions. Observe now that each j 1Aj is a positive simple function, hence
m m m m
I j 1Aj = I (j 1Aj ) = j I (1Aj ) = j (Aj ) .
j=1 j=1 j=1 j=1
Put in other words: we have used the linearity of I .
Problem 9.2 We check Properties 9.8(i)(iv).

(i) This follows from Properties 9.3 and Lemme 9.5 since 1A d = I (1A ) = (A).
(ii) This follows again from Properties 9.3 and Corollary 9.7 since for un E+ with
u = supn un (note: the sups are increasing limits!) we have
u d = sup un d = sup I (un )

n n
= sup I (un )
n
= sup I (un )
n
= u d.
(iii) This follows again from Properties 9.3 and Corollary 9.7 since for un , vn E+ with
u = supn un , v = supn vn (note: the sups are increasing limits!) we have
(u + v) d = n
lim (un + vn ) d = lim I (un + vn )
n
= lim (I (un ) + I (vn ))

n
= lim I (un ) + lim I (vn )

n n
= u d + v d.
(iv) This was shown in step 1 of the proof of the Beppo Levi theorem 9.6
Problem 9.3 Consider on the space ([1, 0], ), (dx) = dx is Lebesgue measure on [0, 1], the
sequence of tent-type functions

0, 1 x k1 ,
fk (x) = (k N),

k (x + k ), k x 0,
3 1 1
65
(draw a picture!). These are clearly monotonically increasing functions but, as a sequence,
we do not have fk (x) fk+1 (x) for every x! Note also that each function is integrable
(with integral 12 k) but the pointwise limit is not integrable.
Problem 9.4 Following the hint we set sm = u1 + u2 + . . . + um . As a finite sum of positive

measurable functions this is again positive and measurable. Moreover, sm increases to
s =
j=1 uj as m . Using the additivity of the integral (9.8 (iii)) and the Beppo Levi
theorem 9.6 we get

uj d = sup sm d = sup sm d
j=1 m m
= sup (u1 + . . . + um ) d
m
m
= sup uj d
m j=1

= uj d.
j=1
Conversely, assume that 9.9 is true. We want to deduce from it the validity of Beppo
Levis theorem 9.6. So let (wj )jN be an increasing sequence of measurable functions with
limit w = supj w. For ease of notation we set w0 0. Then we can write each wj as a
partial sum
wj = (wj wj1 ) + + (w1 w0 )
of positive measurable summands of the form uk = wk wk1 . Thus,

m
wm = uk and w = uk
k=1 k=1
and, using the additivity of the integral,

m
9.9
w d = uk d = sup uk d = sup wm d.
k=1 m k=1 m
Problem 9.5 Set (A) = 1A u d. Then is a [0, ]-valued set-function defined for A A.
(M1 ) Since 1 0 we have clearly () = 0 u d = 0.
(M1 ) Let A = jN Aj a disjoint union of sets Aj A. Then

1Aj = 1A
j=1
and we get from Corollary 9.9

(A) = ( 1Aj ) u d = (1Aj u) d
j=1 j=1

= 1Aj u d
j=1

= (Aj ).
j=1
66
Problem 9.6 This is actually trivial: since our -algebra is P(N), all subsets of N are measur-
able. Now the sub-level sets {u } = {k N u(k) } are always N and as such
they are P(N), hence u is always measurable.
Problem 9.7 We have seen in Problem 4.6 that is indeed a measure. We follow the instruc-
tions. First, for A A we get
1A d = (A) = j (A) = 1A dj .
jN jN
By the linearity of the integral, this easily extends to functions of the form 1A + 1B
where A, B A and , 0:
(1A + 1B ) d = 1A d + 1B d
= 1A dj + 1B dj
jN jN
= (1A + 1B ) dj
jN
and this extends obviously to simple functions which are finite sums of the above type.
f d = f dj f E+ .
jN
Finally, take u M+ and take an approximating sequence un E+ with supn un = u. Then

we get by Beppo Levi (indicated by an asterisk )

u d = sup un d = sup un dj
n n j=1
m
= sup sup un dj
n m j=1
m
= sup sup un dj
m n j=1
m
= sup lim un dj
m n
j=1
m
= sup lim un dj
m j=1 n
m

= sup lim un dj
m j=1 n

= u dj
j=1
where we repeatedly used that all sups are increasing limits and that we may swap any
two sups (this was the hint to the hint to Problem 4.6.)
67
Problem 9.8 Set wj = u uj . Then the wj are a sequence of positive measurable functions. By
Fatous lemma we get
limjinf wj d limjinf wj d
= lim inf ( u d uj d)
j
= u d lim sup uj d
j
(see, e.g. the rules for lim inf and lim sup in Appendix A). Thus,
u d lim sup uj d limjinf wj d

j
= lim inf (u uj ) d
j
= (u lim sup uj ) d
j
and the claim follows by subtracting the finite value u d on both sides.
Remark. The uniform domination of uj by an integrable function u is really im-

portant. Have a look at the following situation: (R, B(R), ), (dx) = dx denotes
Lebesgue measure, and consider the positive measurable functions uj (x) = 1[j,2j] (x).
Then lim supj uj (x) = 0 but lim supj uj d = lim supj j = 0 d.
Problem 9.9 (i) Have a look at Appendix A, Lemma A.2.
(ii) You have two possibilities: the set-theoretic version:
( lim inf Aj ) = ( Aj )
j k jk

= sup ( Aj )
k jk

(Aj ) jk
hence, inf jk (Aj )
sup inf (Aj )

k jk
= lim inf (Aj )

j
which uses at the point the continuity of measures, Theorem 4.4.
The alternative would be (i) combined with Fatous lemma:
( lim inf Aj ) = 1lim inf j Aj d

j
= lim inf 1Aj d

j
lim inf 1Aj d

j
68
(iii) Again, you have two possibilities: the set-theoretic version:
( lim sup Aj ) = ( Aj )
j k jk
#
= inf ( Aj )
k jk

(Aj ) jk
hence, supjk (Aj )
inf sup (Aj )

k jk
= lim sup (Aj )

j
which uses at the point # the continuity of measures, Theorem 4.4. This step uses
the finiteness of .
The alternative would be (i) combined with the reversed Fatou lemma of Problem
9.8:
( lim sup Aj ) = 1lim supj Aj d

j
= lim sup 1Aj d

j
lim sup 1Aj d

j
(iv) Take the example in the remark to the solution for Problem 9.8. We will discuss it
here in its set-theoretic form: take (R, B(R), ) with denoting Lebesgue measure
(dx) = dx. Put Aj = [j, 2j] B(R). Then
lim sup Aj = [j, 2j] = [k, ) =

j k jk k
But 0 = () lim supj (Aj ) = lim supj j = is a contradiction. (The problem is

that [k, ) = !)
Problem 9.10 We use the fact that, because of disjointness,

1 = 1X = 1Aj
j=1
so that, because of Corollary 9.9,

u d = ( 1Aj ) u d = (1Aj u) d
j=1 j=1

= 1Aj u d.
j=1
Assume now that (X, A, ) is -finite with an exhausting sequence of sets (Bj )j A such
that Bj X and (Bj ) < . Then we make the Bj s pairwise disjoint by setting
A1 = B1 , Ak = Bk (B1 Bk1 ) = Bk Bk1 .
69
Now take any sequence (ak )k (0, ) with k ak (Ak ) < e.g. ak = 2k /((Ak ) + 1)
and put

w(x) = ak 1Ak .
j=1
Then w is integrable and, obviously, w(x) > 0 everywhere.
Problem 9.11 (i) We check (M1 ), (M2 ). Using the fact that N (x, ) is a measure, we find
N () = N (x, ) (dx) = 0 (dx) = 0.
Further, let (Aj )jN A be a sequence of disjoint sets and set A = j Aj . Then
N (A) = N (x, j Aj ) (dx) = N (x, Aj ) (dx)

j
9.9
= N (x, Aj ) (dx)
j
= N (Aj ).
j
(ii) We have for A, B A and , 0,
N (1A + 1B )(x) = (1A (y) + 1B (y)) N (x, dy)
= 1A (y) N (x, dy) + 1B (y) N (x, dy)
= N 1A (x) + N 1B (x).
Thus N (f + g)(x) = N f (x) + N g(x) for positive simple f, g E + (A). Moreover, since
by Beppo Levi (marked by an asterisk ) for an increasing sequence fk u

sup N fk (x) = sup fk (y) N (x, dy) = sup fk (y) N (x, dy)
k k k
= u(y) N (x, dy)
= N u(x)
and since the sup is actually an increasing limit, we see for positive measurable
u, v M+ (A) and the corresponding increasing approximations via positive simple
functions fk , gk :
N (u + v)(x) = sup N (fk + gk )(x)

k
= sup N fk (x) + sup N gk (x)

k k
= N u(x) + N v(x).
Moreover, x N 1A (x) = N (x, A) is a measurable function, thus N f (x) is a mea-

surable function for all simple f E + (A) and, by Beppo Levi (see above) N u(x),
u M+ (A), is for every x an increasing limit of measurable functions N fk (x).
Therefore, N u M+ (A).
70
(iii) If u = 1A , A A, we have
1A (y) N (dy) = N (A) = N (x, A) (dx)
= N 1A (x) (dx).
By linearity this carries over to f E + (A) and, by a Beppo-Levi argument, to

u M+ (A).
Problem 9.12 Put

(A) = u 1A+ d + (1 u) 1A d.
If A is symmetric w.r.t. the origin, A+ = A and A = A. Therefore,
(A) = u 1A d + (1 u) 1A d = 1A d = (A).
This means that extends . It also shows that () = 0. Since is defined for all sets
from B(R) and since has values in [0, ], it is enough to check -additivity.
For this, let (Aj )j B(R) be a sequence of pairwise disjoint sets. From the definitions

it is clear that the sets (Aj ) are again pairwise disjoint and that j (Aj ) = ( j Aj ) .
Since each of the set-functions
B u 1B d, C (1 u) 1C d
is -additive, it is clear that their sum will be -additive, too.
The obvious non-uniqueness of the extension does not contradict the uniqueness theorem
for extensions, since does not generate B(R)!
71
10 Integrals of measurable functions and null
sets.
Problem 10.1 Let u, v be integrable functions and a, b R. Assume that either u, v are real-
valued or that au + bv makes sense (i.e. avoiding the case ). Then we have
au + bv au + bv = a u + b v K(u + v)
with K = max{a, b}. Since the RHS is integrable (because of Theorem 10.3 and Prop-
erties 9.8) we have that au + bv is integrable by Theorem 10.3. So we get from Theorem
10.4 that
(au + bv) d = au d + bv d = a u d + b v d
and this is what was claimed.
Problem 10.2 We follow the hint and show first that u(x) = x1/2 , 0 < x < 1, is Lebesgue
integrable. The idea here is to construct a sequence of simple functions approximating u
from below. Define

0, if x (0, n1 )
un (x) =

u( j+1 ), if x [ j , j+1 ),
j = 1, . . . n 1
n n n
n1
un = u( j+1
n )1 j j+1
[n, n )
j=1
which is clearly a simple function. Also un u and limn un (x) = supn un (x) = u(x) for
all x.
Since P (A) is just (A (0, 1)), the integral of un is given by
n1
j+1 j j+1
un dP = IP (un ) = u( n )[ n , n )
j=1
n1
j+1
= n 1
n
j=1
n1
1
n 1
j=1
and is thus finite, even uniformly in n! So, Beppo Levis theorem tells us that
u dP = sup un dP sup 1 = 1 <

n n
73
showing integrability.
Now u is clearly not bounded but integrable.
Problem 10.3 True, we can change an integrable function on a null set, even by setting it to
the value + or on the null set. This is just the assertion of Theorem 10.9 and its
Corollaries 10.10, 10.11.
Problem 10.4 We have seen that a single point is a Lebesgue null set: {x} B(R) for all
x R and ({x}) = 0, see e.g. Problems 4.11 and 6.4. If N is countable, we know that
N = {xj j N} = jN {xj } and by the -additivity of measures

(N ) = {xj } = ({xj }) = 0 = 0.
jN jN jN
The Cantor set C from Problem 7.10 is, as we have seen, uncountable but has measure
(C) = 0. This means that there are uncountable sets with measure zero.
In R2 and for two-dimensional Lebesgue measure 2 the situation is even easier: every
line L in the plane has zero Lebesgue measure and L contains certainly uncountably
many points. That 2 (L) = 0 is seen from the fact that L differs from the ordinate
{(x, y) R2 x = 0} only by a rigid motion T which leaves Lebesgue measure invariant
(see Chapter 5) and 2 ({x = 0}) = 0 as seen in Problem 6.4.
Problem 10.5 (i) Since {u > c} {u c} and, therefore, ({u > c}) ({u c}), this
follows immediately from Proposition 10.12. Alternatively, one could also mimic the
proof of this Proposition or use part (iii) of the present problem with (t) = t, t 0.
(ii) This will follow from (iii) with (t) = tp , t 0, since ({u > c}) ({u c}) as
{u > c} {u c}.
(iii) We have, since is increasing,
({u c}) = ({(u) (c)})
= 1{x (u(x))(c)} (x) (dx)

(u(x))
= 1{x (u(x))(c)} (x) (dx)
(u(x))
(u(x))
1{x (u(x))(c)} (x) (dx)
(c)
(u(x))
(dx)
(c)
1
= (u(x)) (dx)
(c)
(iv) Let us set b = u d. Then we follow the argument of (iii):
({u b}) = 1{x u(x)b} (x) (dx)
74
u(x)
= 1{x u(x)b} (x) (dx)
u(x)
u(x)
1{x u(x)b} (x) (dx)
b
u
d
b
1
= u d
b
and substituting u d for b shows the inequality.
(v) Using the fact that is decreasing we get {u < c} = {(u) > (c)}mind the
change of the inequality signand going through the proof of part (iii) again we
used there that increases only in the first step in a similar role as we used the
decrease of here! This means that the argument of (iii) is valid after this step and
we get, altogether,
({u < c}) = ({(u) > (c)})
= 1{x (u(x))>(c)} (x) (dx)

(u(x))
= 1{x (u(x))>(c)} (x) (dx)
(u(x))
(u(x))
1{x (u(x))>(c)} (x) (dx)
(c)
(u(x))
(dx)
(c)
1
= (u(x)) (dx)
(c)

(vi) This follows immediately from (ii) by taking = P , c = V X, u = X EX and
p = 2. Then
1
P (X EX EX) X EX dP
2
( V X)2
1 1
= 2 V X = 2.
VX
Problem 10.6 We mimic the proof of Corollary 10.13. Set N = {u = } = {up = }. Then
N = kN {up k} and using Markovs inequality (MI) and the continuity of measures,
Theorem 4.4, we find
4.4
(N ) = ( {up k}) = lim ({up k})
kN k
MI 1
lim u d = 0.
p
k k

<
For arctan this is not any longer true for several reasons:
... arctan is odd and changes sign, so there could be cancelations under the integral.
75
... even if we had no cancelations we have the problem that the points where u(x) =
are now transformed to points where arctan(u(x)) =
2 and we do not know how the
measure acts under this transformation. A simple example: Take to be a measure
of total finite mass (that is: (X) < ), e.g. a probability measure, and take the
function u(x) which is constantly u +. Then arctan(u(x)) =
2 throughout, and
we get

arctan u(x) (dx) = d = d = (X) < ,
2 2 2
but u is nowhere finite!
Problem 10.7 : since the Aj are disjoint we get the identities

1j Aj = 1Aj and so u 1j Aj = u 1Aj ,
k=1 k=1
hence u1An = u1An u1j Aj = u1j Aj showing the integrability of each u1An by
Theorem 10.3. By a Beppo Levi argument (Theorem 9.6) or, directly, by Corollary 9.9
we get

u d = u1Aj d = u1Aj d
j=1 Aj j=1 j=1
= u1j Aj d < .
The converse direction follows again from Corollary 9.9, now just the other way
round:

u1j Aj d = u1Aj d = u1Aj d
j=1 j=1

= u d <
j=1 Aj
showing that u1j Aj is integrable.
Problem 10.8 One possibility to solve the problem is to follow the hint. We go here a different
(shorter) direction.
(i) Observe that uj v 0 is a sequence of positive and integrable functions. Applying

Fatous lemma (in the usual form) yields (observing the rules for lim inf, lim sup from
Appendix A, compare also Problem 9.8):
limjinf uj d v d = limjinf (uj v) d
lim inf (uj v) d

j
= lim inf uj d v d
j
and the claim follows upon subtraction of the finite (!) number v d.
76
(ii) Very similar to (i) by applying Fatous lemma to the positive, integrable functions
w uj 0:
w d lim sup uj d = limjinf (w uj ) d

j
lim inf (w uj ) d
j
= w d lim sup uj d
j
Now subtract the finite number w d on both sides.
(iii) We had the counterexample, in principle, already in Problem 9.8. Nevertheless...
Consider Lebesgue measure on R. Put fj (x) = 1[2j,j] (x) and gj (x) = 1[j,2j] (x).
Then lim inf fj (x) = 0 and lim sup gj (x) = 0 for every x and neither admits an inte-
grable minorant resp. majorant.
Problem 10.9 Note the misprint in the statement: the RHS should read
j=0 P (u
j)
We can safely assume that u 0 (since integrability of u is equivalent to the integrability

of u). Then

u(x) = u(x)1{ju<j+1} (x) j1{ju<j+1} (x)
j=0 j=0

= j(1{ju} (x) 1{j+1u} (x)).
j=0
N
Since for fixed x, u(x) < , we have N 1{N +1u} (x) 0. Therefore, we can use Abels
summation trick and get
N
j(1{ju} (x) 1{j+1u} (x))
j=0
= 0 (1{0u} (x) 1{1u} (x)) + 1 (1{1u} (x) 1{2u} (x))

+ + N (1{N u} (x) 1{N +1u} (x))
= 1{1u} (x) + 1{2u} (x) + + 1{N u} (x) N 1{N +1u} (x)
and this proves

j1{ju<j+1} (x) = 1{ju} (x).
j=0 j=1
Therefore,

u = u 1{ju<j+1} (j + 1)1{ju<j+1}
j=0 j=0

2j1{ju<j+1}
j=0
77

= 2 1{ju} (x) 2u.
j=1
The claim follows from this, the fact that const. dP = const. and Corollary 9.9:

P ({u j}) = 1{uj} dP = 1{uj} dP.
j=0 j=0 j=0
Problem 10.10 For u = 1B and v = 1C we have, because of independence,
uv dP = P (A B) = P (A)P (B) = u dP v dP.
For positive, simple functions u = j j 1Bj and v = k k 1Ck we find
uv dP = j k 1Aj 1Bk dP
j,k
= j k P (Aj Bk )
j,k
= j k P (Aj )P (Bk )
j,k
= ( j P (Aj ))( k P (Bk ))

j k
= u dP v dP.
For measurable u M+ (B) and v M+ (C) we use approximating simple functions uk

E + (B), uk u, and vk E + (C), vk v. Then, by Beppo Levi,
uv dP = lim
k
uk vk dP = lim
k
uk dP lim
j
vj dP
= u dP v dP.
Integrable independent functions: If u L1 (B) and v L1 (C), the above calculation

when applied to u, v shows that u v is integrable since
uv dP u dP v dP < .
Considering positive and negative parts finally also gives
uv dP = u dP v dP.
Counterexample: Just take u = v which are integrable but not square integrable, e.g.
u(x) = v(x) = x1/2 . Then (0,1) x1/2 dx < but (0,1) x1 dx = , compare also Problem
10.2.
Problem 10.11 (i) Assume that f is A -measurable. The problem at hand is to construct
A-measurable upper and lower functions g and f . For positive simple functions this
78
is clear: if f (x) = N
j=0 j 1Bj (x) with j 0 and Bj A , then we can use Problem
4.13(v) to find Bj , Cj A with (Cj Bj ) = 0
Bj Bj Cj j 1Bj j 1Bj j 1Cj
and summing over j = 0, 1, . . . , N shows that f f g where f, g are the appropriate

lower and upper sums which are clearly A measurable and satisfy
({f g}) (C0 B0 CN BN )

(C0 B0 ) + + (CN BN )
= 0 + + 0 = 0.
(Bj ), we have
Moreover, since by Problem 4.13 (Bj ) = (Cj ) =

j (Bj ) = j
(Bj ) = j (Cj )
j j j
which is the same as

f d = f d
= g d.
(ii), (iii) Assume that u is A -measurable; without loss of generality (otherwise consider
positive and negative parts) we can assume that u 0. Because of Theorem 8.8 we
know that fk u for fk E + (A ). Now choose the corresponding A-measurable
lower and upper functions fk , gk constructed in part (i). By considering, if necessary,
max{f1 , . . . , fk } we can assume that the fk are increasing.
Set u = supk fk and v = lim inf k gk . Then u, v M(A), u u v, and by Fatous

lemma
v d = limkinf gk d limkinf gk d
= lim inf fk d

k
= u d

v d.
Since fk u we get by Beppo Levi and Fatou
u d = sup fk d = limkinf fk d
k
= lim inf gk d
k
lim inf gk d
k
= v d
u d
79
This proves that u d = v d = u d. This answers part (iii) by considering

positive and negative parts.
It remains to show that {u v} is a -null set. (This does not follow from the above
integral equality, cf. Problem 10.16!) Clearly, {u v} = {u < v}, i.e. if x {u < v} is
fixed, we deduce that, for sufficiently large values of k,
fk (x) < gk (x), k large
since u = sup fk and v = lim inf k gk . Thus,
{u v} {fk gk }
k
but the RHS is a countable union of -null sets, hence a null set itself.
Conversely, assume first that u u v for two A-measurable functions u, v with

u = v a.e. We have to show that {u > } A . Using that u u v we find that
{u > } {u > } {v > }
but {v > }, {u > } A and {u > } {v > } {u v} is a -null set. Because of

Problem 4.13 we conclude that {u > } A .
Problem 10.12 Note the misprint in the statement: for the estimate (E) + (F )
(E F ) the sets E, F should be disjoint!
Throughout the solution the letters A, B are reserved for sets from A.
(i) a) Let A E B. Then (A) (B) and going to the supAE and inf EB proves
(E) (E).
b) By the definition of and we find some A E such that
(E) (A) .
Since Ac E c we can enlarge A, if needed, and achieve
(E c ) (Ac ) .
Thus,
(X) (E) (E c )
(E) (A) + (E c ) (Ac )
2,
and the claim follows as 0.
c) Let A E and B F be arbitrary majorizing A-sets. Then A B E F and
(E F ) (A B) (A) + (B).
80
Now we pass on the right-hand side, separately, to the inf AE and inf BF , and
obtain
(E F ) (E) + (F ).
d) Let A E and B F be arbitrary minorizing A-sets. Then A B E F and
(E F ) (A B) = (A) + (B).
Now we pass on the right-hand side, separately, to the supAE and supBF ,
where we stipulate that A B = , and obtain
(E F ) (E) + (F ).
(ii) By the definition of the infimum/supremum we find sets An E An such that

1
(A) (An ) + (A) (An ) .
n
Without loss of generality we can assume that the An increase and that the An
decrease. Now A = n An , A = n An are A-sets with A A A . Now,
(An ) (A ) as well as (An ) (E) which proves (A ) = (E). Analogously,
(An ) (A ) as well as (An ) (E) which proves (A ) = (E).
(iii) In view of Problem 4.13 and (i), (ii), it is clear that
{E X (E) = (E)} =
{E X A, B A, A E B, (B A) = 0}
but the latter is the completed -algebra A . That A = A =

is now trivial
since and coincide on A .
Problem 10.13 Let A A and assume that there are non-measurable sets, i.e. P(X) A. Take
some N / A which is a -null set. Assume also that N A = . Then u = 1A and
w = 1A + 2 1N are a.e. identical, but w is not measurable.
This means that w is only measurable if, e.g. all (subsets of) null sets are measurable,
that is if (X, A, ) is complete.
Problem 10.14 The function 1Q is nowhere continuous but u = 0 Lebesgue almost everywhere.
That is
{x 1Q (x) is discontinuous} = R
while
{x 1Q 0} = Q is a Lebesgue null set,
that is 1Q coincides a.e. with a continuous function but is itself at no point continuous!
The same analysis for 1[0,) yields that
{x 1[0,) (x) is discontinuous} = {0}
81
which is a Lebesgue null set, but 1[0,) cannot coincide a.e. with a continuous function!
This, namely, would be of the form w = 0 on (, ) and w = 1 on (, ) while it
interpolates somehow between 0 and 1 if < x < . But this entails that
{x w(x) 1[0,) (x)}
cannot be a Lebesgue null set!
Problem 10.15 Let (Aj )jN A be an exhausting sequence Aj X such that (Aj ) < . Set

1
f (x) = 1Aj (x).
j=1 2j ((A j ) + 1)
Then f is measurable, f (x) > 0 everywhere, and using Beppo Levis theorem

1
f d = ( 1Aj ) d
j=1 2j ((A j ) + 1)

1
= 1Aj d
j=1 2 ((Aj ) + 1)
j
(Aj )
= j
j=1 2 ((Aj ) + 1)

2j = 1.
j=1
Thus, set P (A) = A f d. We know from Problem 9.5 that P is indeed a measure.
If N N , then, by Theorem 10.9,
10.9
P (N ) = f d = 0
N
so that N NP .
Conversely, if M MP , we see that
f d = 0
M
but since f > 0 everywhere, it follows from Theorem 10.9 that 1M f = 0 -a.e., i.e.
(M ) = 0. Thus, NP N .
ym theorem) that N = NP if
Remark. We will see later (cf. Chapter 19, Radon-Nikod
and only if P = f (i.e., if P has a density w.r.t. ) such that f > 0.
Problem 10.16 Well, the hint given in the text should be good enough.
82
11 Convergence theorems and their
applications.
Problem 11.1 We start with the simple remark that
a bp (a + b)p
(max{a, b} + max{a, b})p
= 2p max{a, b}p
= 2p max{ap , bp }
2p (ap + bp ).
Because of this we find that uj up 2p g p and the right-hand side is an integrable

dominating function.
Proof alternative 1: Apply Theorem 11.2 on dominated convergence to the sequence

j = uj up of integrable functions. Note that j (x) 0 and that 0 j where
= 2p g p is integrable and independent of j. Thus,
lim uj up d = lim j d = lim j d

j j j
= 0 d = 0.
Proof alternative 2: Mimic the proof of Theorem 11.2 on dominated convergence. To do

so we remark that the sequence of functions
j
0 j = 2p g p uj up 2p g p
Since the limit limj j exists, it coincides with lim inf j j , and so we can use Fatous
Lemma to get
2 g d = lim
p p
inf j d
j
lim inf j d
j
= lim inf (2p g p uj up ) d

j
83
= 2p g p d + lim inf ( uj up d)
j
= 2p g p d lim sup uj up d
j
where we used that lim inf j (j ) = lim supj j . This shows that lim supj uj up d = 0,
hence
0 lim inf uj up d lim sup uj up d 0
j j
showing that lower and upper limit coincide and equal to 0, hence limj uj up d = 0.
Problem 11.2 Assume that, as in the statement of Theorem 11.2, uj u and that uj f
L1 (). In particular,
f uj and uj f
(j N) is an integrable minorant resp. majorant. Thus, using Problem 10.8 at below,
u d = lim inf uj d
j

lim inf uj d
j
lim sup uj d
j

lim sup uj d = u d.
j
This proves u d = limj uj d.
Addition: since 0 u uj limj uj + uj 2f L1 (), the sequence u uj has an

integrable majorant and using Problem 10.8 we get
0 lim sup uj u d lim sup uj u d = 0 d = 0

j j
and also (i) of Theorem 11.2 follows...
Problem 11.3 By assumption we have

k
0 fk gk f g,
k
0 Gk fk G f.
Using Fatous Lemma we find
(f g) d = lim
k
(fk gk ) d
= lim inf (fk gk ) d

k
lim inf (fk gk ) d

k
= lim inf fk d g d,
k
84
and
(G f ) d = lim
k
(Gk fk ) d
= lim inf (Gk fk ) d

k
lim inf (Gk fk ) d

k
= G d lim sup fk d.
k
Adding resp. subtracting g d resp. G d therefore yields
lim sup fk d f d lim inf fk d

k k
Problem 11.4 Using Beppo Levis theorem in the form of Corollary 9.9 we find

uj d = uj d < , (*)
j=1 j=1
which means that the positive function

j=1 uj is finite almost everywhere, i.e. the series

j=1 uj converges (absolutely) almost everywhere.
Moreover,
N N
uj d = uj d (**)
j=1 j=1
and, using the triangle inequality both quantities

N N
uj d and uj d
j=n j=n
can be estimated by
N n,N
uj d 0
j=n
because of (*). This shows that both sides in (**) are Cauchy sequences, i.e. they are
convergent.
Problem 11.5 Since L1 () uj 0 we find by monotone convergence, Theorem 11.1, that

uj d 0. Therefore,

= (1)j uj and S = (1)j uj d converge
j=1 j=1
(conditionally, in general). Moreover, for every N N,
N N
N
(1) uj d = (1) uj d S.
j j
j=1 j=1
85
All that remains is to show that the right-hand side converges to d. Observe that for
j=1 (1) uj we have
SN = N j
S2N S2N +2 . . . S
and we find, as Sj L1 (), by monotone convergence that
lim S2N d = d.
N
Problem 11.6 Consider uj (x) = j 1(0,1/j) (x), j N. It is clear that uj is measurable and
Lebesgue integrable with integral
1
uj d = j =1 j N.
j
Thus, limj uj d = 1. On the other hand, the pointwise limit is
u(x) = lim uj (x) 0

j
so that 0 = u d = limj uj d 1.
The example does not contradict dominated convergence as there is no uniform dominating
integrable function.
Alternative: a similar situation can be found for vk (x) = 1
k 1[0,k] (x) and the pointwise limit
v 0. Note that in this case the limit is even uniform and still limk vk d = 1 0 = v d.
Again there is no contradiction to dominated convergence as there does not exist a uniform
dominating integrable function.
Problem 11.7 Let be an arbitrary Borel measure on the line R and define the integral function
for some u L1 () through
I(x) = Iu (x) = u(t) (dt) = 1(0,x) (t)u(t) (dt).

(0,x)
For any sequence 0 < lj x, lj < x from the left and rk x, rk > x from the right we find
j k
1(0,lj ) (t) 1(0,x) (t) and 1(0,rk ) (t) 1(0,x] (t).
Since 1(0,x) u u L1 is a uniform dominating function, Lebesgues dominated conver-

gence theorem yields
I(x+) I(x) = lim I(rk ) lim I(lj )

k j
= 1(0,x] (t)u(t) (dt) 1(0,x) (t)u(t) (dt)
= (1(0,x] (t) 1(0,x) (t))u(t) (dt)
= 1{x} (t)u(t) (dt)
= u(x) ({x}).
Thus I(x) is continuous at x if, and only if, x is not an atom of .

Remark: the proof shows, by the way, that Iu (x) is always left-continous at every x, no
matter what or u look like.
86
Problem 11.8 (i) We have

1
1[1,) (x) dx
x
1
= lim 1[1,n) (x) dx by Beppo Levis thm.
n x
1
= lim dx usual shorthand
n [1,n) x
n 1 n
= lim (R) dx Riemann- exists
n 1 x 1
= lim [log x]n1

n
= lim [log(n) log(1)] =

n
which means that 1

x is not Lebesgue-integrable over [1, ).
(ii) We have
1
1[1,) (x) dx
x2
1
= lim 2 1[1,n) (x) dx by Beppo Levis thm.
n x
1
n [1,n) x2
n 1 n
n 1 x2 1
n
= lim [ x1 ]1
n
= lim [1 n1 ] = 1 <
n
which means that 1

x2
is Lebesgue-integrable over [1, ).
(iii) We have
1
1(0,1] (x) dx
x
1
= lim 1(1/n,1] (x) dx by Beppo Levis thm.
n x
1
n (1/n,1] x
1 1 1
n 1/n x 1/n
1
= lim [2 x]1/n
n

= lim [2 2 n1 ]
n
=2<
which means that 1

x
is Lebesgue-integrable over (0, 1].
(iv) We have
1
1(0,1] (x) dx
x
87
1
= lim 1(1/n,1] (x) dx by Beppo Levis thm.
n x
1
n (1/n,1] x
1 1 1
n 1/n x 1/n
= lim [log x]11/n

n
= lim [log(1) log n1 ]

n
which means that 1

x is not Lebesgue-integrable over (0, 1].
Problem 11.9 We construct a dominating integrable function.
If x 1, we have clearly exp(x ) 1, and (0,1] 1 dx = 1 < is integrable.
If x 1, we have exp(x ) M x2 for some suitable constant M = M < . This function

is integrable in [1, ), see e.g. Problem 11.8. The estimate is easily seen from the fact
that x x2 exp(x ) is continuous in [1, ) with limx x2 exp(x ) = 0.
This shows that exp(x ) 1(0,1) +M x2 1[1,) with the right-hand side being integrable.
Problem 11.10 Take (a, b) where 0 < a < b < are fixed (but arbitrary). We show that the
function is continuous for these . This shows the general case since continuity is a local
property and we can catch any given 0 by some choice of a and bs.
We use the Continuity lemma (Theorem 11.4) and have to find uniform (for (a, b))
3
dominating bounds on the integrand function f (, x) = ( sinx x ) ex . First of all, we
remark that sinx x M which follows from the fact that sin x
x is a continuous function such
that limx sin x
x = 0 and limx0 sin x
x = 1. (Actually, we could choose M = 1...). Moreover,
exp(x) 1 for x (0, 1) and exp(x) Ca,b x2 for x 1use for this the continuity
of x2 exp(x) and the fact that limx x2 exp(x) = 0. This shows that
f (, x) M (1(0,1) (x) + Ca,b x2 1[1,) (x))
and the right-hand side is an integrable dominating function which does not depend on
as long as (a, b). But since f (, x) is obviously continuous, the Continuity
lemma applies and proves that (0,) f (, x) dx is continuous.
Problem 11.11 Fix some number N > 0 and take x (N, N ). We show that G(x) is continuous
on this set. Since N was arbitrary, we find that G is continuous for every x R.
sin(tx) sin(tx) 1
Set g(t, x) = t(1+t2 )
=x (tx) 1+t2 . Then, using that sinu u M , we have
1 1
g(t, x) x M M N (1(0,1) (t) + 2 1[1,) (t))
1+t 2 t
and the right-hand side is a uniformly dominating function, i.e. G(x) makes sense and we
find G(0) = t0 g(t, 0) dt = 0. To see differentiability, we use the Differentiability lemma
88
(Theorem 11.5) and need to prove that x g(t, x) exists (this is clear) and is uniformly
dominated for x (N, N ). We have
sin(tx) cos(tx)
x g(t, x) = x =
t(1 + t2 ) (1 + t2 )
1

1 + t2
1
(1(0,1) (t) + 1[1,) (t))
t2
and this allows us to apply the Differentiability lemma, so
G (x) = x g(t, x) dt = x g(t, x) dt

t0 t0
cos(tx)
= dt
t0 1 + t2
cos(tx)
= dt
R 1 + t2
(use in the last equality that {0} is a Lebesgue null set). Thus, by a Beppo-Levi argu-
ment (and using that Riemann=Lebesgue whenever the Riemann integral over a compact
interval exists...)
1 n 1
G (0) = dt = lim (R) dt
R 1+t2 n n 1 + t2
= lim [tan1 (t)]nn

n
= .
Now observe that

t t
x sin(tx) = t cos(tx) = x cos(tx) = t sin(tx).
x x
Since the integral defining G (x) exists we can use a Beppo-Levi argument, Riemann=Lebesgue
(whenever the Riemann integral over an interval exists) and integration by parts (for the
Riemann integral) to find
x cos(tx)
xG (x) = dt
R 1 + t2
n x sin(tx)
x
= lim (R) dt
n n t(1 + t2 )
n t sin(tx)
t
= lim (R) dt
n n t(1 + t2 )
n sin(tx)
t
= lim (R) dt
n n 1 + t2
n 1
= lim (R) t sin(tx) dt
n n 1 + t2
n n
sin(tx) 1
= lim [ ] lim (R) sin(tx) t dt
n 1 + t2 t=n n n 1 + t2
n 2t
= lim (R) sin(tx) dt
n n (1 + t2 )2
2t sin(tx)
= dt.
R (1 + t2 )2
89
Problem 11.12 (i) Note that for 0 a, b 1

1 1
1 (1 a)b = btb1 dt b dt = ba
1a 1a
so that we get for 0 x k and a = x/k, b = k/(k + 1)

k
x k+1 x
(1 ) 1 , 0xk
k k+1
or,
x k x k+1
(1 ) 1[0,k] (x) (1 ) 1[0,k+1] (x).
k k+1
Therefore we can appeal to Beppo Levis theorem to get
x k x k
lim (1 ) ln x 1 (dx) = sup 1(1,k) (x)(1 ) ln x 1 (dx)
k (1,k) k kN k
x k
= sup [1(1,k) (x)(1 ) ] ln x 1 (dx)
kN k
= 1(1,) (x)ex ln x 1 (dx).
That ex ln x is integrable in (1, ) follows easily from the estimates
ex CN xN and ln x x
which hold for all x 1 and N N.
(ii) Note that x ln x is continuous and bounded in [, 1], thus Riemann integrable. It
is easy to see that x ln x x is a primitive for ln x. The improper Riemann integral
1
ln x dx = lim[x ln x x]1 = 1
0 0
exists and, since ln x is negative throughout (0, 1), improper Riemann and Lebesgue
integrals coincide. Thus, ln x L1 (dx, (0, 1)).
Therefore,
x k
(1 ) ln x ln x, x (0, 1)
k
is uniformly dominated by an integrable function and we can use dominated conver-
gence to get
x k x k
lim (1 ) ln x dx = lim (1 ) ln x dx
k (0,1) k (0,1) k k
= ex ln x dx
(0,1)
Problem 11.13 Fix throughout (a, b) (0, ) and take x (a, b). Let us remark that, just as
in Problem 11.8, we prove that
t dt < < 1 and t dt < > 1.

(0,1) (1,)
90
(i) That the integrand function x (t, x) is continuous on (a, b) is clear. It is therefore
enough to find an integrable dominating function. We have
et tx1 ta1 t (0, 1), x (a, b)
which is clearly integrable on (0, 1) and
et tx1 Ma,b t2 t 1, x (a, b)
where we used that t et , > 0, is continuous and limt t et = 0 to find Ma,b . This
function is integrable over [1, ). Both estimates together give the wanted integrable
dominating function. The Continuity lemma (Theorem 11.4) applies. The well-
definedness of (x) comes for free as a by-product of the existence of the dominating
function.
(ii) Induction Hypothesis: (m) exists and is of the form as claimed in the statement of
the problem.
Induction Start m = 1: We have to show that (x) is differentiable. We want to use

the Differentiability lemma, Theorem 11.5. For this we remark first of all, that the
integrand function x (t, x) is differentiable on (a, b) and that
x (t, x) = x et tx1 = et tx1 log t.
We have now to find a uniform (for x (a, b)) integrable dominating function for
x (t, x). Since log t t for all t > 0 (the logarithm is a concave function!),
et tx1 log t = et tx1 log t

et tx et tb Cb t2 t 1, x (a, b)
(use for the last step the argument used in part (i) of this problem). Moreover,
et tx1 log t ta1 log t

1
= ta1 log Ca t1/2 t (0, 1), x (a, b)
t
where we used the fact that limt0 t log 1t = 0 which is easily seen by the substitution
t = eu and u and the continuity of the function t log 1t .
Both estimates together furnish an integrable dominating function, so the Differen-

tiability lemma applies and shows that
(x) = x (t, x) dt = et tx1 log t dt = (1) (x).

(0,) (0,)
Induction Step m m + 1: Set (m) (t, x) = et tx1 (log t)m . We want to apply the
Differentiability Lemma to (m) (x). With very much the same arguments as in the
91
induction start we find that (m+1) (t, x) = x (m) (t, x) exists (obvious) and satisfies
the following bounds
et tx1 (log t)m+1 = et tx1 (log t)m+1

et tx+m
et tb+m
Cb,m t2 t 1, x (a, b)
et tx1 (log t)m+1 ta1 log tm
1 m+1
= ta1 (log )
t
Ca,m t1/2 t (0, 1), x (a, b)
and the Differentiability lemma applies completing the induction step.
(iii) Using a combination of Beppo-Levi (indicated by B-L), Riemann=Lebesgue (if the

Riemann integral over an interval exists) and integration by parts (for the Riemann
integral, indicated by I-by-P) techniques we get
x(x) = lim et xtx1 dt B-L

n (1/n,n)
n
= lim (R) et t tx dt
n 1/n
n
= lim [et tx ]t=1/n lim (R) t et tx dt
n
I-by-P
n n 1/n
n
= lim (R) et t(x+1)1 dt
n 1/n
= lim et t(x+1)1 dt
n (1/n,n)
= et t(x+1)1 dt B-L
(0,)
= (x + 1).
Problem 11.14 Fix (a, b) (0, 1) and let always u (a, b). We have for x 0 and L N0
eux
xL f (u, x) = xL
ex + 1
eux
= xL x
e +1
ux
e
xL x
e
L (u1)x
=x e
1[0,1] (x) + Ma,b 1(1,) (x) x2
where we used that u 1 < 0, the continuity and boundedness of x eax for x [1, ) and
0. If x 0 we get
eux
xL f (u, x) = xL
ex + 1
92
= xL eux
1[1,0] (x) + Na,b 1(,1) (x) x2 .
Both inequalities give dominating functions which are integrable; therefore, the integral
R x f (u, x) dx exists.
L
To see m-fold differentiability, we use the Differentiability lemma (Theorem 11.5) m-times.
Formally, we have to use induction. Let us only make the induction step (the start is very
similar!). For this, observe that
xn eux xn+m eux

um (xn f (u, x)) = um = x
ex + 1 e +1
but, as we have seen in the first step with L = n + m, this is uniformly bounded by an
integrable function. Therefore, the Differentiability lemma applies and shows that
um xn f (u, x) dx = xn um f (u, x) dx = xn+m f (u, x) dx.

R R R
Problem 11.15 Note the misprint in this problem: the random variable X should be positive.
(i) Since
dm tX
e = X m etX X m
dtm
(m)
m applications of the differentiability lemma, Theorem 11.5, show that X (0+)
exists and that
(m)
X (0+) = (1)m X m dP.
(ii) Using the exponential series we find that

m
(1)k tk (1)k tk
etX X k = Xk
k=0 k! k=m+1 k!

(1)m+1+j tj
= tm+1 X m+1+j .
j=0 (m + 1 + j)!
Since the left-hand side has a finite P -integral, so has the right, i.e.

(1)m+1+j tj
(X ) dP
m+1+j
converges
j=0 (m + 1 + j)!
and we see that

tX
m
(1)k tk
(e Xk ) dP = o(tm )
k=0 k!
as t 0.
(iii) We show, by induction in m, that

m1
(u)k um
eu u 0. (*)
k=0 k! m!
93
Because of the elementary inequality
eu 1 u u 0
the start of the induction m = 1 is clear. For the induction step m m + 1 we note
that
m
(u)k u m1
(y)k
eu = (ey ) dy
k=0 k! 0 k=0 k!
u m1
(y)k
ey dy
0 k=0 k!
(*) u ym
dy
0 m!
um+1
= ,
(m + 1)!
Setting x = tX in (*), we find by integration that

m1 X k dP tm X m dP
( etX (1)k tk ) .
k=0 k! m!
(iv) If t is in the radius of convergence of the power series, we know that
tm X m dP
lim =0
m m!
which, when combined with (iii), proves that
m1 X k dP
X (t) = lim (1)k tk .
m
k=0 k!
Problem 11.16 (i) Wrong, u is NOT continuous on the irrational numbers. To see this, just
take a sequence of rationals qj Q [0, 1] approximating p [0, 1] Q. Then
lim u(qj ) = 1 0 = u(p) = u(lim qj ).

j j
(ii) True. Mind that v is not continuous at 0, but {n1 , n N} {0} is still countable.
(iii) True. The points where u and v are not 0 (that is: where they are 1) are count-
able sets, hence measurable and also Lebesgue null sets. This shows that u, v are
measurable and almost everywhere 0, hence u d = 0 = v d.
(iv) True. Since Q [0, 1] as well as [0, 1] Q are dense subsets of [0, 1], ALL lower resp.
upper Darboux sums are always
S [u] 0 resp. S [u] 1
(for any finite partition of [0, 1]). Thus upper and lower integrals of u have the
value 0 resp. 1 and it follows that u cannot be Riemann integrable.
94
Problem 11.17 Note that every function which has finitely many discontinuities is Riemann
integrable. Thus, if {qj }jN is an enumeration of Q, the functions uj (x) = 1{q1 ,q2 ,...,qj } (x)
are Riemann integrable (with Riemann integral 0) while their increasing limit u = 1Q is
not Riemann integrable.
Problem 11.18 Of course we have to assume that u is Borel measurable! By assumption we

know that uj = u1[0,j] is (properly) Riemann integrable, hence Lebesgue integrable and
j j
u d = uj d = (R) u(x) dx u(x) dx.
[0,j] [0,j] 0 0
The last limit exists because of improper Riemann integrability. Moreover, this limit is
an increasing limit, i.e. a sup. Since 0 uj u we can invoke Beppo Levis theorem and
get

u d = sup uj d = u(x) dx <
j 0
proving Lebesgue integrability.

Problem 11.19 Observe that x2 = k x = k, x 0, k N0 . Thus, Since sin x2 is
continuous, it is on every bounded interval Riemann integrable. By a change of variables,
y = x2 , we get
b b dy b sin y
sin(x2 ) dx = sin y = dy
a a 2 y a 2 y

a
which means that for a = ak = k and b = bk = (k +1) = ak+1 the values a k+1 sin(x2 ) dx
k
are a decreasing sequence with limit 0. Since on [ ak , ak+1 ] the function sin x2 has only
one sign (and alternates its sign from interval to interval), we can use Leibniz convergence
criterion to see that the series

ak+1
sin(x2 ) dx (*)
k ak
converges, hence the improper integral exists.

The function cos x2 can be treated similarly. Alternatively, we remark that sin x2 = cos(x2
/2).
The functions are not Lebesgue integrable. Either we show that the series (*) does not
converge absolutely, or we argue as follows:
sin x2 = cos(x2 /2) shows that sin x2 dx and cos x2 dx either both converge or
diverge. If they would converge (this is equivalent to Lebesgue integrability...) we would
find because of sin2 + cos2 1 and sin , cos 1,

= 1 dx = [(sin x2 )2 + (cos x2 )2 ] dx
0 0

= (sin x2 )2 dx + (cos x2 )2 dx
0 0

sin x2 dx + cos x2 dx < ,
0 0
which is a contradiction.
95
Problem 11.20 Let r < s and, without loss of generality, a b. A change of variables yields
s f (bx) f (ax) s f (bx) s f (ax)
dx = dx dx
r x r x r x
bs f (y) as f (y)
= dy dy
br y ar y
bs f (y) br f (y)
= dy dy
as y ar y
Using the mean value theorem for integrals, E.12, we get
s f (bx) f (ax) bs 1 br 1
dx = f (s ) dy f (r ) dy
r x as y ar y
= f (s ) ln ab f (r ) ln ab .
s r0
Since s (as, bs) and r (ar, br), we find that s and r 0 which means
that
s f (bx) f (ax) s
dx = [f (s ) f (r )] ln ab (M m) ln ab .
r x r0
Problem 11.21 (i) The function x x ln x is bounded and continuous in [0, 1], hence Rie-
mann integrable. Since in this case Riemann and Lebesgue integrals coincide, we may
use Riemanns integral and the usual rules for integration. Thus, changing variables
according to x = et , dx = et dt and then s = (k + 1)t, ds = (k + 1) ds we find,
1
[et (t)] et dt
k
(x ln x)k dx =
0 0

= (1)k tk et(k+1) dt
0
s k s ds
= (1)k ( ) e
0 k+1 k+1
1 k+1 (k+1)1 s
= (1) ( ) s
k
e ds
k+1 0
1 k+1
= (1)k ( ) (k + 1).
k+1
(ii) Following the hint we write

(x ln x)k
xx = ex ln x = (1)k .
k=0 k!
Since for x (0, 1) the terms under the sum are all positive, we can use Beppo Levis
theorem and the formula (k + 1) = k! to get

1
xx dx = (1)k (x ln x)k dx
(0,1) k=0 k! (0,1)

1 1 k+1
= (1)k (1)k ( ) (k + 1)
k=0 k! k+1

1 k+1
= ( )
k=0 k+1

1 n
= ( ) .
n=1 n
96
12 The function spaces Lp, 1 p .
Problem 12.1 olders inequality for r, s (1, ) and

(i) We use H 1
s + 1
t = 1 to get
uqq = uq d = uq 1 d
1/r 1/s
( uqr d) ( 1s d)
1/r
= ( uqr d) ((X))1/s .
Now let us choose r and s. We take

p 1 q 1 1 q
r= > 1 = and =1 =1 ,
q r p s r p
hence
q/p1/q
uq = ( up d) ((X))(1q/p)(1/q)
q/p1/q
= ( up d) ((X))1/q1/p
= up ((X))1/q1/p .
(ii) If u Lp we know that u is measurable and up < . The inequality in (i) then
shows that
uq const up < ,
hence u Lq . This gives Lp Lq . The inclusion Lq L1 follows by taking p q,

q 1.
Let (un )nN Lp be a Cauchy sequence, i.e. limm,n un um p = 0. Since by the
inequality in (i) also
lim un um q (X)1/q1/p lim un um p = 0

m,n m,n
we get that (un )nN Lq is also a Cauchy sequence in Lq .

(iii) No, the assertion breaks down completely if the measure has infinite mass. Here
is an example: = Lebesgue measure on (1, ). Then the function f (x) = 1
x is
not integrable over [1, ), but f (x) =
2 1
x2
is. In other words: f / L (1, ) but1
f L2 (1, ), hence L2 (1, ) / L1 (1, ). (Playing around with different exponents

shows that the assertion also fails for other p, q 1....).
97
Problem 12.2 This is going to be a bit messy and rather than showing the streamlined solution
we indicate how one could find out the numbers oneself. Now let be some number in
(0, 1) and let , be conjugate indices: 1
+ 1 = 1 where , (1, ). Then by the Holder
inequality
u d = u u
r r r(1)
d
1 1

( u r
d) ( u r(1)
d)
r r(1)
r r(1)
= ( u r
d) ( ur(1)
d) .
Taking rth roots on both sides yields

(1)
r r(1)
ur ( ur d) ( ur(1) d)
= ur u1
r(1) .
This leads to the following system of equations:

1 1
p = r, q = r(1 ), 1= +

with unknown quantities , , . Solving it yields
1
r 1
q qp qp
= , = = .
1
p 1
q
qr rp
Problem 12.3 v L () means that v(x) (v + ) for all x N = N with (N ) = 0.

Using in step below Theorem 10.9, we get
uv d uv d

= uv d
Nc
u(v + ) d
Nc
= (v + ) u d
Nc
(v + ) u d
and since the left-hand side does not depend on > 0, we can let 0 and find
0
uv d uv d uv d (v + )u1 v u1 .
Problem 12.4 Proof by induction in N .
Start N = 2: this is just H

olders inequality.
older inequality holds for some N 2.

Hypothesis: the generalized H
98
Step N N + 1:. Let u1 , . . . , uN , w be N + 1 functions and let p1 , . . . , pN , q > 1 be such

that p1 1 1
1 + p2 + . . . + pN + q
1
= 1. Set p1 = p1 1 1
1 + p2 + . . . + pN . Then, by the ordinary
Holder inequality,
1/p
u1 u2 . . . uN w d ( u1 u2 . . . uN d) uq
p
1/p
= ( u1 p u2 p . . . uN p d) uq
Now use the induction hypothesis which allows us to apply the generalized Holder in-
1
j=1 j = p/p = 1, to the first factor to
equality for N (!) factors j = pj /p, and thus N
get
1/p
u1 u2 . . . uN w d = ( u1 u2 . . . uN d) uq
p p p
up1 up2 . . . upN uq .
Problem 12.5 Draw a picture similar to the one used in the proof of Lemma 12.1 (note that
the increasing function need not be convex or concave....). Without loss of generality we
can assume that A, B > 0 are such that (A) B which is equivalent to A (B) since
and are inverses. Thus,
B (B) A
AB = () d + () d + B d.
0 0 (B)
Using the fact that increases, we get that
((B)) = B (C) B C (B)
and we conclude that

B (B) A
AB = () d + () d + B d
0 0 (B)
B (B) A
() d + () d + () d
0 0 (B)
B A
= () d + () d
0 0
= (B) + (A).
Problem 12.6 Let us show first of all that Lp -limk uk = u. This follows immediately from
limk u uk p = 0 since the series
k=1 u uk p converges.
Therefore, we can find a subsequence (uk(j) )jN such that
lim uk(j) (x) = u(x) almost everywhere.

j
Now we want to show that u is the a.e. limit of the original sequence. For this we mimic
the trick from the Riesz-Fischer theorem 12.7 and show that the series
K
(uj+1 uj ) = lim (uj+1 uj ) = lim uK
j=0 K j=0 K
99
(again we agree on u0 = 0 for notational convenience) makes sense. So let us employ

Lemma 12.6 used in the proof of the Riesz-Fischer theorem to get
X
X
X X
X
X X
X
X X
X
X
X
X
X (u ) X
X
X X
X
X u X
X
X
X
X j+1 uj X
X X
X j+1 uj X
X
X
X
Xj=0 X
X
Xp XX
Xj=0 X
X
Xp

uj+1 uj p
j=0

(uj+1 up + u uj p )
j=0
<
where we used Minkowskis inequality, the function u from above and the fact that

j=1 uj up < along with u1 p < . This shows that limK uK (x) = j=0 (uj+1 (x)
uj (x)) exists almost everywhere.
We still have to show that limK uK (x) = u(x). For this we remark that a subsequence
has necessarily the same limit as the original sequencewhenever both have limits, of
course. But then,

u(x) = lim uk(j) (x) = lim uk (x) = (uj+1 (x) uj (x))
j k j=0
Problem 12.7 That for every fixed x the sequence
n
un (x) = n1(0,1/n) (x) 0
is obvious. On the other hand, for any subsequence (un(j) )j we have
1 j
un(j) d = n(j) = n(j)p1 c
p p
n(j)
with c = 1 in case p = 1 and c = if p > 1. This shows that the Lp -limit of this
subsequencelet us call it w if it exists at allcannot be (not even a.e.) u = 0.
On the other hand, we know that a sub-subsequence (

uk(j) )j of (uk(j) )j converges point-
wise almost everywhere to the Lp -limit:
k(j) (x) = w(x).

lim u
j
Since the full sequence limn un (x) = u(x) = 0 has a limit, this shows that the sub-sub-
sequence limit w(x) = 0 almost everywherea contradiction. Thus, w does not exist in
the first place.
Problem 12.8 Using Minkowskis and Holders inequalities we find for all > 0
uk vk uv1 = uk vk uk v + uk v uv
100
uk (vk v) + (uk u)v

uk p vk vq + uk up vq
(M + vq )
for all n N . We used here that the sequence (uk p )kN is bounded. Indeed, by
Minkowskis inequality
uk p = uk up + up + up = M.
Problem 12.9 We use the simple identity
un um 22 = (un um )2 d
= (u2n 2un um + um ) d (*)
= un 22 + um 22 2 un um d.
Case 1: un u in L2 . This means that (un )nN is an L2 Cauchy sequence, i.e. that
limm,n un um 22 = 0. On the other hand, we get from the lower triangle inequality for
norms
lim un 2 u2 lim un u2 = 0
n n
so that also limn un 22 = limm um 22 = u22 . Using (*) we find
2 un um d = un 22 + um 22 un um 22
n,m
u22 + u22 0
= 2u22 .
Case 2: Assume that limn,m un um d = c for some number c R. By the very

definition of this double limit, i.e.
> 0 N N un um d c < n, m N ,
we see that limn un un d = c = limm um um d hold (with the same c!). Therefore,
again by (*), we get
un um 22 = un 22 + um 22 2 un um d
n,m
c + c 2c = 0,
i.e. (un )nN is a Cauchy sequence in L2 and has, by the completeness of this space, a limit.
Problem 12.10 Use the exponential series to conclude from the positivity of h and u(x) that

hj uj hN N
exp(hu) = u .
j=0 j! N!
101
Integrating this gives

hN
u d exp(hu) d <
N
N!
and we find that u LN . Since is a finite measure we know from Problem 12.1 that for
N > p we have LN Lp .
Problem 12.11 (i) We have to show that un (x)p = np (x + n)p has finite integral
measurability is clear since un is continuous. Since np is a constant, we have only
to show that (x + n)p is in L1 . Set = p > 1. Then we get from a Beppo-Levi
and a domination argument
(x + n) (dx) (x + 1) (dx)
(0,) (0,)
1 (dx) + (x + 1) (dx)
(0,1) (1,)

1 + lim x (dx).
k (1,k)
Now using that Riemann=Lebesgue on intervals where the Riemann integral exists,
we get
k
lim x (dx) = lim x dx
k (1,k) k 1
= lim [(1 )1 x1 ]1
k
k
= (1 )1 lim (k 1 1)
k
1
= ( 1) <
which shows that the integral is finite.

(ii) We have to show that vn (x)q = nq eqnx is in L1 again measurability is inferred
from continuity. Since nq is a constant, it is enough to show that eqnx is integrable.
Set = qn. Since
lim (x)2 ex = 0 and ex 1 x 0,

x
and since ex is continuous on [0, ), we conclude that there are constants C, C()
such that
C
ex min {1, }
(x)2
1
C() min {1, 2 }
x
1
= C() (1(0,1) (x) + 1[1,) )
x2
but the latter is an integrable function on (0, ).
Problem 12.12 Without loss of generality we may assume that . We distinguish between
the case x (0, 1) and x [1, ). If x 1, then
1 1 1 1/2
= x 1;
x x + x x + x x + x
102
this shows that (x + x )1 is in L1 ((0, 1), dx) if, and only if, < 1.
Similarly, if x 1, then
1 1 1 1/2
= x 1
x x +x x +x x + x
this shows that (x + x )1 is in L1 ((1, ), dx) if, and only if, > 1.
Thus, (x + x )1 is in L1 (R, dx) if, and only if, both < 1 and > 1.
Problem 12.13 If we use X = {1, 2, . . . , n}, x(j) = xj , = 1 + + n we have
n 1/p
( xj )
p
= xLp ()
j=1
and it is clear that this is a norm for p 1 and, in view of Problem 12.18 it is not a norm
for p < 1 since the triangle (Minkowski) inequality fails. (This could also be shown by a
direct counterexample.
Problem 12.14 Without loss of generality we can restrict ourselves to positive functionselse
we would consider positive and negative parts. Separability can obviously considered
separately!
Assume that L1+ is separable and choose u Lp+ . Then up L1 and, because of separability,
there is a sequence (fn )n D1 L1 such that
in L1 in L1
fn up upn up
n n
1/p
if we set un = fn Lp . In particular, un(k) (x) u(x) almost everywhere for some
k
subsequence and un(k) p up . Thus, Riesz theorem 12.10 applies and proves that
in Lp
Lp un(k) u.
k
Obviously the separating set Dp is essentially the same as D1 , and we are done.
The converse is similar (note that we did not make any assumptions on p 1 or p < 1this
is immaterial in the above argument).
Problem 12.15 We have seen in the lecture that, whenever limn u un p = 0, there is
a subsequence un(k) such that limk un(k) (x) = u(x) almost everywhere. Since, by
assumption, limj uj (x) = w(x) a.e., we have also that limj un(j) (x) = w(x) a.e.,
hence u(x) = w(x) almost everywhere.
Problem 12.16 We remark that y log y is concave. Therefore, we can use Jensens inequality
for concave functions to get for the probability measure /(X) = (X)1 1X
d d
(log u) log ( u )
(X) (X)
= log (
u d
)
(X)
103
1
= log ( ),
(X)
Problem 12.17 As a matter of fact,
u(s) ds log u(t) dt u(x) log u(x) dx.

(0,1) (0,1) (0,1)
We begin by proving the hint. log x 0 x 1. So,
y 1 ( log y y log y 1 y)
and y 1 ( log y y log y 1 y).
Assume now that (0,1) u(x) dx = 1. Substituting in the above inequality y = u(x) and
integrating over (0, 1) yields
log u(x) dx u(x) log u(x) dx.

(0,1) (0,1)
Now assume that = (0,1) u(x) dx. Then (0,1) u(x)/ dx = 1 and the above inequality
gives
u(x) u(x) u(x)
log dx log dx
(0,1) (0,1)
which is equivalent to
log u(x) dx log

(0,1)
= log u(x) dx log dx

(0,1) (0,1)
u(x)
= logdx
(0,1)
u(x) u(x)
log dx
(0,1)
1 u(x)
= u(x) log dx
(0,1)
1 1
= u(x) log u(x) dx u(x) log dx
(0,1) (0,1)
1 1
= u(x) log u(x) dx u(x) dx log
(0,1) (0,1)
1
= u(x) log u(x) dx log .
(0,1)
The claim now follows by adding log on both sides and then multiplying by =
(0,1) u(x) dx.
Problem 12.18 Note the misprint: q = p/(p 1) 1

p + 1
q = 1 independent of p (1, ) or
p (0, 1)!
104
(i) Let p (0, 1) and pick the conjugate index q = p/(p 1) < 0. Moreover, s = 1/p
(1, ) and the conjugate index t, 1
s + 1
t = 1, is given by
1
s p 1
t= = = (1, ).
s1 1
p 1 1p
Thus, using the normal H

older inequality for s, t we get
wp
u d = u
p p
d
wp
1/s 1/t
p s pt
( (u w ) d)
p
( w d)
p 1p
= ( uw d) ( wp/(p1) d) .
Taking pth roots on either side yields

1/p (1p)/p
( up d) ( uw d)( wp/(p1) d)
1/q
= ( uw d)( wq d)
(ii) This reversed Minkowski inequality follows from the reversed Holder inequality in
exactly the same way as Minkowskis inequality follows from Holders inequality, cf.
Corollary 12.4. To wit:
(u + v) d = (u + v) (u + v) d
p p1
= u (u + v)p1 d + v (u + v)p1 d
(i)
up (u + v)p1 q + vp (u + v)p1 q .
Dividing both sides by u + vp1 q proves our claim since

1/q 11/p
(u + v)p1 q = ( (u + v)(p1)q d) = ( (u + v)p d) .
Problem 12.19 By assumption, u u C < and u / 0.
(i) We have
Mn = un d C n d = C n (X) (0, ).
Note that Mn > 0.
(ii) By the Cauchy-Schwarz-Inequality,
Mn = un d
n+1 n1
= u 2 u 2 d
105
1/2 1/2
( un+1 d) ( un1 d)

= Mn+1 Mn1 .
(iii) The upper estimate follows from
Mn+1 = un+1 d un u d = u Mn .
Set P = /(X); the lower estimate is equivalent to

d
u
1/n n+1
d (X)
( u n
) d
(X) un (X)
1+1/n
( un dP ) un+1 dP
(n+1)/n
( un dP ) un+1 dP
and the last inequality follows easily from Jensens inequality since P is a probability
measure:
(n+1)/n
n+1
( u dP )
n
u
n n dP = un+1 dP.
(iv) Following the hint we get

1/n n
un ({u > u }) (u ) u ,
0
i.e.
lim inf un u .
n
Combining this with the estimate from (iii), we get
u lim inf (X)1/n un

n
(iii) Mn+1
lim inf
Mn
n
Mn+1
lim sup
n Mn
u .
Problem 12.20 The hint says it all.... Maybe, you have a look at the specimen solution of
Problem 12.19, too.
Problem 12.21 Without loss of generality we may assume that f 0. We use the following
standard representation of f , see (8.7):
N
f = j 1Aj
j=0
106
with 0 = 0 < 1 < . . . < N < and mutually disjoint sets Aj . Clearly, {f 0} =
A1 AN .
Assume first that f E Lp (). Then
N N
> f p d = pj (Aj ) p1 (Aj ) = p1 ({f 0});
j=1 j=1
thus ({f 0}) < .

Conversely, if ({f 0}) < , we get
N N
p p p
f d = j (Aj ) N (Aj ) = N ({f 0}) < .
p
j=1 j=1
Since this integrability criterion does not depend on p 1, it is clear that E + Lp () =

E + L1 (), and the rest follows since E = E + E + .
Problem 12.22 (i) Note that (x) = x1/q is concavee.g. differentiate twice and show that
it is negativeand using Jensens inequality for positive f, g 0 yields
p/q
f g d = gf 1{f 0} f p d
1/q
g q f p 1{f 0} f p d
f d p
f p d
11/q 1/q
( f p d) ( g q d)
where we used 1{f 0} 1 in the last step. Note that f g L1 follows from the fact
that (g q f p 1{f 0} )f p = g q L1 .
(ii) The function (x) = (x1/p + 1)p has second derivative
1p
(x) = (1 + x1/p ) x11/p 0
p
showing that is concave. Using Jensens inequality gives for f, g 0
g p
(f + g) 1{f 0} d = ( 1{f 0} + 1) f 1{f 0} d
p p
f
p
g p 1{f 0} d 1/p
f d(
p
) + 1
{f 0} {f 0} f d
p

1/p 1/p p
= [( p
g d) + ( p
f d) ] .
{f 0} {f 0}
Adding on both sides {f =0} (f +g)p d = {f =0} g p d yields, because of the elementary
inequality Ap + B p (A + B)p , A, B 0, p 1,
(f + g) d
p
1/p 1/p p p/p

[( g p d) + ( f p d) ] + [ g p d]
{f 0} {f 0} {f =0}
1/p 1/p p
[( g p d) + ( f p d) ] .
107

Measures, Integrals & Martingales (3rd Edition) : Solution Manual Chapter 1-12

Uploaded by

Copyright:

Available Formats

Measures, Integrals & Martingales (3rd Edition) : Solution Manual Chapter 1-12

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Measures, Integrals & Martingales (3rd Edition) : Solution Manual Chapter 1-12

Uploaded by

Copyright:

Available Formats

Measures, Integrals & Martingales (3rd edition)

Cambridge University Press, Cambridge 2011

Solution Manual Chapter 112

Dresden, October 2014

just as we had expected.

Problem 2.1 (i) We have

(ii) Using (i) and de Morgans laws (*) yields

(iv) Using (i) and de Morgans laws (*) gives

(v) Using (i) and de Morgans laws (*) gives

Problem 2.2 Observe, first of all, that

This follows easily from

Using this and the analogous formula for C A then gives

= [A (A B C)c ] [B (A B C)c ] [C (A B C)c ]

= [A (A B C)] [B (A B C)] [C (A B C)]

= [A (B C)] [B (A C)] [C (A B)]

In order to see (a) we note that

and (b) follows from

= f () = f ((0, 1) (1, 2)) f ((0, 1)) f ((1, 2)) = {c}.

Moreover, f (X) = f (B) = {c} = f (X B) but f (X) f (B) = .

(ii) Recall, first of all, the definition of f 1 for a map f X Y and B Y

f 1 (B) = {x X f (x) B}.

(i), (vi) For every x we have

(ii), (v) For every x we have

1A (x) = 1(AB)(AB) (x) = min{1AB (x) + 1AB (x), 1}

1AB (x) = 1(AB)(AB)(BA) (x)

Problem 2.6 (i) Using 2.5(iii), (iv) we see that

1A B (x) = 1(AB)(BA) (x)

1A B (x) 1A (x) + 1B (x) = 1 1A (x) + 1B (x)mod2 = 1.

It is also possible to show that

This follows from

(ii) From part (i) we see that

and this expression treats A, B, C in a completely symmetric way, i.e.

(iii) Step 1: (P(X), , ) is an abelian group.

Inverse element: A A = (A A) (A A) = , i.e. each element is its own inverse.

Associativity: see part (ii);

Step 2: For the multiplication we have

Step 3: Distributive law:

1A(B C) = 1A 1B C = 1A (1B + 1C mod 2)

Problem 2.7 Let f X Y . One has

This can be seen as follows: by definition f 1 (B) = {x f (x) B} so that

f f 1 (B) = f ({x f (x) B}) = {f (x) f (x) B} {y y B}

On the other hand

To see this we observe that because of the definition of f 1

f 1 f (A) = {x f (x) f (A)} {x x A} = A ()

f ([0, 1]) = {1} and f 1 f ([0, 1]) = f 1 ({1}) = R [0, 1].

and, since g is also injective,

showing that f g is injective.

The quickest solution is to observe that N Z = N N N {0} N (N) where

Therefore, N Z is a union of three countable sets, hence countable.

An alternative approach would be to write out Z N (the swap of Z and N is for

. . . (3, 1) (2, 1) (1, 1) (0, 1) (1, 1) (2, 1) (3, 1) . . .

The proof of #(N N) = #N can be easily adaptedusing some pretty obvious

mN Qm is a countable union of countable sets, hence countable, cf. Theorem 2.6.

On the other hand, N N = jN N {j} which is a countable union of countable sets,

Applying Theorem 2.7 finally gives #(N N) = #N.

Problem 2.11 Since E F the map E F, e e is an injection, thus #E #F .

s1 = d11 d12 d13 d14 . . . d1k . . .

and produce a new 0-1-sequence S = (e1 , e2 , e3 , . . .) by setting