Sol20 1

MA40042 Measure Theory and Integration (2020/21): Solutions 1
1. Suppose n ∈ N and X is a set with n elements. Show that the power set P(X) has 2n elements.
Set X = {x1 , . . . , xn }. To pick a subset of X, one has to decide whether to include x1 or not,
whether to include x2 or not, and so on. This is a total of n decisions with two choices at each
stage, so the total number of composite choices is 2 × 2 · · · × 2 = 2n . That is, X has 2n subsets so
P(X) (the collection of all subsets of X) has 2n elements.
Alternatively: for k = 1, . . . , n the set X will have nk subsets

Pnwithnk elements.n So nusing the
Binomial Theorem we have that the number of subsets of Xis k=0 k = (1 + 1) = 2 .
2. Suppose X is a non-empty set and A is an algebra in X. Show that for any k ∈ N, if Ai ∈ A for
i = 1, 2, . . . , k then ∪ki=1 Ai ∈ A.
Proof by induction. The result is true for k = 1. Suppose it holds for some k. Suppose Ai ∈ A
for i = 1, 2, . . . , k + 1. Set B = ∪ki=1 Ai . Then B ∈ A by the inductive hypothesis. Since A is an
algebra so closed under pairwise unions,
∪k+1
i=1 Ai = B ∪ Ai+1 ∈ A,
completing the induction.
3. Which of the following collections M of sets (in X) are σ-algebras? Which ones are algebras?
Explain each answer.
(a) X = {1, 2, 3, 4},
M = {∅, {1}, {2}, {3, 4}, {1, 2}, {1, 3, 4}, {2, 3, 4}, {1, 2, 3, 4}}.
(b) X = {1, 2, 3, . . . } and
M = {A ⊂ X : either A or X \ A is finite}.
(c) X is an uncountable set and
M = {A ⊂ X : either A or X \ A is countable}.
(d) X is any set, M1 ⊂ M2 ⊂ . . . are σ-algebras in X and M = ∪∞

n=1 Mn .
(a) This is a σ-algebra, and hence also an algebra. To check this, observe that:
(i) ∅ ∈ M.
(ii) We can pair the elements of M with their complements, and see that each A ∈ M appears
together with its complement, so A ∈ M implies that Ac ∈ M.
(iii) If A1 , A2 , · · · ∈ M, then since M has finitely many members, there are only finitely many
distinct sets in the sequence A1 , A2 , . . . . Therefore, for some n ≥ 1, the sets A1 , A2 , . . . , An already
include all sets appearing in the sequence A1 , A2 , . . . , and therefore ∪∞ n
i=1 Ai = ∪i=1 Ai . Hence it
is enough to check that M is closed under finite unions. For this, it is enough to check that if
A, B ∈ M, then A ∪ B ∈ M, since the general case follows by induction on n. It can be checked
directly that for any pair of sets in M, their union is in M, and this completes checking that M
is closed under countable unions.
Remark. This is a special case of Question 5 below: the σ-algebra M is generated by the partition
{{1}, {2}, {3, 4}}.
(b) This is an algebra, but not a σ-algebra.
(i) ∅ is finite, so ∅ ∈ M.
(ii) M is closed under complements: Suppose A ∈ M. Then either Ac is finite (in which case
Ac ∈ M) or A is finite (in which case Ac has a finite complement so is in M). Thus if A ∈ M
then Ac ∈ M.
(iii) M is closed under pairwise unions. Suppose A, B ∈ M. If both A and B are finite, then
A ∪ B is finite, and hence A ∪ B ∈ M. If not, then at least one of Ac and B c is finite, and hence
(A ∪ B)c = Ac ∩ B c is finite, showing that A ∪ B ∈ M. Thus if A, B ∈ M then A ∪ B ∈ M.
M is not a σ-algebra, since if we take Ai = {2i}, i = 1, 2, 3, . . . , then each Ai is in M (since
finite), but their union A = ∪∞ i=1 Ai = {2, 4, 6, . . . } is not in M, because neither it nor nor its
complement is finite.
(c) This is a σ-algebra, and hence also an algebra. Here and elsewhere in the course, we count
FINITE sets as being countable.
(i) ∅ ∈ M, since ∅ is finite and therefore countable.
(ii) M is closed under complements, since the condition for a set to be included in M is symmetric
in A and Ac . To go into more detail: if A ∈ M then either Ac is countable (so Ac ∈ M) or A is
countable (so Ac has a countable complement so is in M).
(iii) Let A1 , A2 , · · · ∈ M. If all the Ai ’s are countable, then ∪∞ i=1 Ai is also countable (since a
countable union of countable sets is countable - see the end of this solution) so that ∪∞ i=1 Ai ∈ M.
∞ ∞
If at least one of the Ai ’s, let’s say An , is such that An is countable, then (∪i=1 Ai ) = ∩i=1 Aci ⊂ Acn ,
c c
and hence (∪∞ c ∞

i=1 Ai ) is countable, showing that ∪i=1 Ai ∈ M.
(d) This is an algebra, but in general not a σ-algebra.
(i) Since M1 is an algebra, ∅ ∈ M1 and hence ∅ ∈ M.
(ii) If A ∈ M, then there exists n ≥ 1 such that A ∈ Mn . Since Mn is a σ-algebra, it follows
that Ac ∈ Mn , and hence Ac ∈ M.
(iii) If A, B ∈ M, there exist n1 , n2 ≥ 1 such that A ∈ Mn1 and B ∈ Mn2 . Take n = max{n1 , n2 };
then A, B ∈ Mn . Since Mn is a σ-algebra, A ∪ B ∈ Mn , and hence also A ∪ B ∈ M.
The above shows that M is an algebra. Note that we cannot adapt the argument of (iii) above to an
infinite sequence of sets. (The problem being that if Ai ∈ Mni , we can have sup{ni : i ≥ 1} = ∞.)
An explicit example where M is not a σ-algebra is the following:
X = {1, 2, 3, . . . }
Mn = {A ⊂ X : either A ⊂ {1, . . . , n} or Ac ⊂ {1, . . . , n}}.
Then ∪∞ n=1 Mn is the same as M from part (b) (which we already showed was not a σ-algebra).
We now check for each fixed n that Mn is an algebra. Indeed, clearly ∅ ∈ Mn and Mn is closed
under complementation. Suppose Ai ∈ Mn for i = 1, 2, 3, . . .. If Ai ⊂ {1, . . . , n} for all i then
also ∪∞ ∞ c
i=1 Ai ⊂ {1, . . . , n} so ∪i=1 Ai ∈ Mn . Otherwise, there exists j with Aj ⊂ {1, . . . , n} so that
(∪∞ c ∞ c c
i=1 Ai ) = ∩i=1 Ai ⊂ Aj ⊂ {1, . . . , n}
and therefore again ∪∞

i=1 Ai ∈ Mn . Thus Mn is closed under countable intersections.
In the solution to part (c) we used the following fact: a countable union of countable sets is
countable. Hopefully you’ve seen this before, and in exams etc. for this course you can use it
without proof, but for the sake of interest, and because it is quite important for this course, here
is a reminder of the proof.
Recall we say a set A is said to be countable if its elements can be enumerated as a sequence
A = {a1 , a2 , . . .} or as A = {a1 , a2 , . . . , an } (that is, we are counting finite sets as being countable.)
Suppose A1 , A2 , . . . are countable. For each i ∈ N let ni denote the number of elements of Ai
(possibly infinite). Write Ai = {ai,1 , ai,2 , ai,3 , . . .} (if Ai is infinite) or Ai = {ai,1 , ai,2 , . . . , ai,ni } (if
Ai is finite). Then we can enumerate the elements of the set ∪∞ i=1 Ai in the following sequence:
a1,1 , a1,2 , a2,1 , a1,3 , a2,2 , a3,1 , a1,4 , a2,3 , a3,2 , a4,1 , a1,5 , a2,4 , . . .
where in the sequence we should omit any element ai,j that already appeared in the sequence, and
also any element ai,j with index j > ni should ni be finite. Thus we have enumerated the elements
of ∪∞i=1 Ai in a (possibly terminating) list, so this set is countable as claimed.
4. Given numbers xij ≥ 0 defined for each i ∈ N, j ∈ N, show that ∞

P P∞ P∞ P∞
i=1 ( j=1 xij ) = j=1 ( i=1 xij ).
[Hint: first show a (weak) inequality between the two double sums.]
P P∞
Let a ≥ 0 with a < ∞ i=1 j=1 ij . Then
x
∞
n
! ∞ ∞
!
X X X X
lim xij = xij > a
n→∞
i=1 j=1 i=1 j=1
P P∞
and therefore we can and do pick a number N ∈ N such that Ni=1 j=1 ij > a, so that
x
N m
!! N m
!!
X X X X
a< lim xij = lim xij ,
m→∞ m→∞
i=1 j=1 i=1 j=1
where we have used the algebra of limits theorem. Therefore we can and do pick a number M ∈ N
PN PM
such that i=1 j=1 xij > a. This is a sum with finitely many terms, so we can take the terms
in any order without changing the value of the sum, so that
∞ ∞
! M N
! N M
!
X X X X X X
xij ≥ xij = xij > a.
j=1 i=1 j=1 i=1 i=1 j=1
P
That is, for any a < ∞ ∞
we have a < ∞
P P P∞
i=1 j=1 xij j=1 ( i=1 xij ), and therefore we must have
P∞ P∞ P∞ P∞
i=1 j=1 xij ≤ j=1 ( i=1 xij ); this is the weak inequality mentioned in the hint. Then
setting yij = xji for all i, j, by the same argument with the roles of i and j reversed we also have
∞ ∞
! ∞ ∞
! ∞ ∞
! ∞ ∞
!
X X X X X X X X
xij = yji ≤ yji = xij .
j=1 i=1 j=1 i=1 i=1 j=1 i=1 j=1
P∞ P
∞ P∞
( ∞
P
Hence we have weak inequalities both ways so i=1 j=1 xij = j=1 i=1 xij ) as required.
5. Suppose X is a non-empty set and X = {A1 , A2 , . . . , Ak }, where the sets A1 , . . . , Ak are non-
empty and form a partition of X, i.e. they are pairwise disjoint and ∪ki=1 Ai = X. Show that
σ(X ) = {∪j∈J Aj : J ⊂ {1, 2, . . . , k}}.
Set H = {∪j∈J Aj : J ⊂ {1, 2, . . . , k}}.
Since σ(X ) is a σ-algebra, and X ⊂ σ(X ), for any J ⊂ {1, . . . , k} we have ∪i∈J Ai ∈ σ(X ) (because
each Ai is in X so also in σ(X ), and σ(X ) is closed under finite unions). Hence H ⊂ σ(X ).
For the inclusion the other way, we check H is a σ-algebra. First, note that ∅ ∈ H (by taking
J = ∅.) Also, H is closed under complementation because for any J ⊂ {1, 2, . . . , k}, we have that
(∪i∈J Ai )c = ∪i∈{1,2,...,k}\J Ai ∈ H.
Finally H is closed under countable unions because if J1 , J2 , . . . are subsets of {1, . . . , k} then
∪∞
i=1 (∪j∈Ji Aj ) = ∪j∈∪∞
i=1 Ji
Aj ∈ H.
Therefore H is a σ-algebra, so σ(X ) ⊂ H. Thus σ(X ) = H.
6. Suppose X is a non-empty set and X = {A1 , A2 , A3 , . . .}, where the sets Ai , i ≥ 1 form a countably
infnite partition of X, i.e. they are pairwise disjoint and ∪∞i=1 Ai = X.
(a) Describe the sets in the σ-algebra generated by X .

(b) Describe the sets in the algebra generated by X .
For part (a) the σ-algebra generated by X consists of all sets of the form ∪j∈J Aj with J ⊂ N (the
proof of this is similar to the previous question).
For part (b) the algebra generated by X consists all sets of the form ∪j∈J Aj with J ⊂ N and
either J or N \ J a finite set.
7. Suppose X = (0, 7] and C = {(0, 2], (1, 5]}. Write down the sets in σ(C).
Consider the partition X = {A1 , A2 , A3 , A4 } of X given by
A1 = (0, 1], A2 = (1, 2], A3 = (2, 5], A4 = (5, 7].
We claim σ(C) = σ(X ). Indeed, all the sets A1 , A2 , A3 , A4 are in σ(C) (for example, A1 =
(0, 2] ∩ (1, 5]c ), so X ⊂ σ(C) and hence σ(X ) ⊂ σ(C). Also, all sets in C are in σ(X ) (for example,
(0, 2] = A1 ∪ A2 ), so C ⊂ σ(X ) and hence σ(C) ⊂ σ(X ), justifying the claim.
Hence by Question 5,
σ(C) = σ(X ) = {∪j∈J Aj : J ⊂ {1, 2, 3, 4}}

= {∅, A1 , A2 , A3 , A4 , A1 ∪ A2 , A1 ∪ A3 , A1 ∪ A4 , A2 ∪ A3 , A2 ∪ A4 , A3 ∪ A4 ,
A1 ∪ A2 ∪ A3 , A1 ∪ A2 ∪ A4 , A1 ∪ A3 ∪ A4 , A2 , ∪A3 ∪ A4 , X}
(a total of 24 = 16 sets).
8. Let O := {G ⊂ R : G is open}; F := {F ⊂ R : F is closed}; K := {K ⊂ R : K is compact}.

(Recall that K is compact iff K is closed and bounded.) Recall that the collection B of Borel sets
in R is defined by B = σ(O).
(a) Construct a Borel set that is neither open nor closed, that is, it is in B \ (O ∪ F).
(b) Prove that σ(K) = B.
For part (a), consider the half-open interval I = (0, 1] for example. Then I is not open (because
1 ∈ I but for any ε > 0 the interval (1 − ε, 1 + ε) is not contained in I) and is also not closed
(because 0 ∈ I c but for any ε > 0 the interval (−ε, ε) is not contained in I c ). However, I = ∩∞
i=1 In
with In = (0, 1 + 1/n). Each In is open so for any σ-algebra H containing the class O of open
sets, we have In ∈ H and hence I = ∩∞ n=1 In ∈ H. Therefore I is in all such H so I ∈ σ(O) = B.
For (b), first we show σ(K) ⊂ B. For any K ∈ K we have that K is closed so its complement
K c is open, i.e. K c ∈ O. Hence for any σ-algebra H containing O we have K c ∈ H and
thus K = (K c )c ∈ H. Hence K ⊂ H and hence σ(K) ⊂ H for all such H, and therefore
σ(K) ⊂ σ(O) = B.
We need to show B ⊂ σ(K). Let U ∈ O, i.e. let U ⊂ R be open. Let V = U c , which is closed
but possibly not bounded. Let Vn = U c ∩ [−n, n]. Then Vn is compact, i.e. Vn ∈ K. Thus for any
σ-algebra H containing K we have Vn ∈ H for all n, so
V = ∪∞
n=1 Vn ∈ H
and therefore also U = V c ∈ H. Thus O ⊂ H, and hence also σ(O) ⊂ H, for all such H. But this
shows B = σ(O) ⊂ σ(K), completing the proof of (b).
9. Show that the examples described in Section 3.2 are indeed measures.
(a) In this case µ(A) is the number of elements in A (possibly +∞).
If A = ∅ then the number of elements is zero, so µ(∅) = 0.
If A, B are disjoint finite subsets of X then we can write A = {a1 , . . . , an } and B = {b1 , . . . , bm }
with n = µ(A) and m = µ(B), and with a1 , . . . , an , b1 , . . . , bm distinct. Then
A ∪ B = {a1 , . . . , an , b1 , . . . bm } which has n + m elements so µ(A ∪ B) = n + m.
Now we prove by induction on k that if A1 , . . . , Ak are pairwise disjoint finite sets then µ(∪ki=1 Ai ) =
P k
i=1 µ(Ai ). The result is clearly true for k = 1, and is true for k = 2 by the preceding paragraph.
Suppose it is true for some k ∈ N. Let A1 , . . . , Ak+1 be pairwise disjoint finite sets and set
B = ∪ki=1 Ai . Then ∪k+1 k
i=1 Ai = B ∪ Ak+1 , and B ∩ Ak+1 = ∪i=1 (Ai ∩ Ak+1 ) = ∅, so by the result in
the preceding paragraph
µ(∪k+1
i=1 Ai ) = µ(B ∪ Ak+1 ) = µ(B) + µ(Ak+1 )
Pk+1
which equals i=1 µ(Ai ) by the inductive hypothesis, completing the induction.
Now suppose A1 , A2 , . . . are pairwise disjoint subsets of X. Set I := {i ∈ N : Ai 6= ∅}, the set of
indices i for which the sets Ai are not empty.
If µ(∪∞i=1 Ai ) < ∞ then we must have both that I is finite, and that Ai is finite for each i ∈ I. In
this case, using the finite additivity of µ on finite sets (already proved) we have
X ∞
X
µ(∪∞
i=1 Ai ) = µ(∪i∈I Ai ) = µ(Ai ) = µ(Ai ).
i∈I i=1
On the other hand, if Pµ(∪∞

i=1 Ai ) = ∞ then either I is infinite,P
or there exists some i with Ai
infinite. In both cases i=1 µ(Ai ) = ∞, so we have µ(∪i=1 Ai ) = ∞
∞ ∞
i=1 µ(Ai ) in all cases, so µ is a
measure.
(b) We need to show δx is a measure. Since ∅ has no elements, we cannot have x ∈ ∅ and hence
δx (∅) = 0.
Suppose A1 , A2 , . . . are disjoint subsets of X. If x ∈/ ∪∞
i=1 Ai then for each i we have x ∈
/ Ai so
δx (Ai ) = 0, and hence
X∞ ∞
X
δx (Ai ) = 0 = 0 = µ(∪∞ i=1 Ai ).
i=1 i=1
On the other hand, if x ∈ ∪∞ i=0 Ai then there is just one j with x ∈ Aj (because the Ai are pairwise
disjoint), and hence δx (Ai ) is 1 for i = j and is 0 for all other i. Hence in this case
∞
X
δx (Ai ) = 1 = µ(∪∞
i=1 Ai ).
i=1
Thus we have the countable additivity of δx so it is a measure.

(c) Here (X, M, µ) is a measure space (the word ‘space’ is missing in the notes) and a ≥ 0 is a
constant. For A ∈ M define ν(A) = aµ(A). We need to show ν is a measure.
Since µ is a measure, µ(∅) = 0 so ν(∅) = aµ(∅) = a · 0 = 0.
If A1 , A2 , . . . are pairwise disjoint and all in M, then since µ is a measure we have
∞
X n
X n
X
ν(∪∞
i=1 Ai ) = aµ(∪∞
i=1 Ai ) =a µ(Ai ) = a lim µ(Ai ) = lim (a µ(Ai ))
n→∞ n→∞
i=1 i=1 i=1
Xn n
X X∞
= lim ( aµ(Ai )) = lim ( ν(Ai )) = ν(Ai ).
n→∞ n→∞
i=1 i=1 n=1
Thus ν is countably additive so is a measure as claimed.

(d) SupposeP(X, M) is a measurable space and µ1 , µ2 , . . . are measures on (X, M). For A ∈ M
set ν(A) = ∞ i=1 µi (A). We claim ν is a measure. Indeed, since µi are all measures we have
∞
X ∞
X
ν(∅) = µi (∅) = 0 = 0.
i=1 i=1
Suppose Ai ∈ M for all i ∈ N and A1 , A2 , . . . are pairwise disjoint. Since µi is a measure, we have
µj (Ai ) ≥ 0 for each i, j. Thus using the result from Question 4 we obtain that
∞
X ∞ X
X ∞ ∞ X
X ∞ ∞
X
ν(∪∞
i=1 Ai ) = µj (∪∞
i=1 Ai ) = ( µj (Ai )) = ( µj (Ai )) = ν(Ai ).
j=1 j=1 i=1 i=1 j=1 i=1
Hence ν is countably additive so is a measure.

P
(e) Now we take X countable, and for A ⊂ X put µ(A) = i∈A mi , where mi ≥ 0 are given for
i ∈ X.
Then µ = ∞
P
i=1 mi δi (pointwise).
By part P
(b) δi is a measure for each i, so by (c) mi δi is a measure for each i, and therefore by part
(d) µ = ∞ i=1 mi δi is also a measure.
10. Let µ be a measure on (R, B) with µ(R) < ∞. For x ∈ R, set F (x) = µ((−∞, x]). Show that F
is nondecreasing and right continuous.
Suppose x < y. Then (−∞, x] ⊂ (−∞, y] (−∞, x] ∪ (x, y] = (−∞, y] and the sets (−∞, x] and
(x, y] are disjoint, so since µ is a measure µ((−∞, x])+µ((x, y]) = µ((−∞, y]) and also µ((x, y]) ≥ 0
so µ((−∞, x]) ≤ µ((−∞, y]), i.e. F (x) ≤ F (y), so F is nondecreasing.
We need to show that F is right continuous, i.e. that for all x ∈ R and ε > 0 there exists δ > 0
with |µ(y) − µ(x)| < ε for x < y < x + δ.
Let x ∈ R. Then (−∞, x] = ∩∞
n=1 (−∞, x + 1/n] so by downwards continuity of measure,
F (x) = µ(−∞, x] = lim µ(−∞, x + 1/n] = lim F (x + 1/n).

n→∞ n→∞
Hence, given ε > 0, we can find n such that F (x + 1/n) < F (x) + ε. But then using the fact that
F is nondecreasing, we also have |F (y) − F (x)| < ε for x < y < x + 1/n, as required.
11. (a) Give an example of a measure space (X, M, µ) and a sequence of sets A1 ⊃ A2 ⊃ A3 . . . with
each Ai ∈ M, such that µ(∩∞ i=1 Ai ) 6= limn→∞ µ(An ).
In view of Theorem 2.2 (iv), any such example must have µ(An ) = ∞ for all n. One example
would be to take X = N with M = P(X) and µ the counting measure as in example 2.3(a).
If we take An = {n, n + 1, n + 2, . . .} we have µ(An ) = +∞ for all n, but ∩∞
i=1 Ai = ∅ so
∞
µ(∩i=1 Ai ) = 0.
(b) Give an example of a measurable space (X, M) and a set function µ : M → [0, ∞] such that
µ is finitely additive but not countably additive.
Set X = N and M := P(X). for all A ⊂ X, set
(
0 if A is finite
µ(A) =
∞ otherwise
This is finitely additive since any finite union of finite sets is finite, but it is not countably
∞
P∞ since if we set Ai = {i} for i ∈ N then the Ai are pairwise disjoint and µ(∪i=1 Ai ) = ∞
additive
but i=1 µ(Ai ) = 0.
12. Suppose X is a non-empty set and X = {A1 , A2 , . . .} is a partition of X with Ai 6= ∅ for each
i. Suppose (a1 , a2 , a3 , . . .) is a sequence of nonnegative numbers, with ai = 0 for all i such that
Ai = ∅. Show that there is a unique measure µ on the measurable space (X, σ(X )) with µ(Ai ) = ai
for all i ∈ N.
By Question 6, the σ-algebra σ(X ) consists of all sets of the form ∪j∈J Aj with J ⊂ N. If µ is a
measure satisfying µ(Ai ) = ai for all i ∈ N, then for all J ⊂ N we have by P countable additivity
P
(or finite additivity if J is finite; see Theorem 2.2 (i)) that µ(∪i∈J Ai ) = i∈J µ(Ai ) = i∈J ai ,
and since this gives a formula for µ(B) for all B ∈ σ(X ), the measure µ is unique.
We still need to check that the set function µ on σ(X ) given by
X
µ(∪i∈J Ai ) = ai , ∀j ⊂ N
i∈J
is a measure. We are assuming that A1 , A2 , . . . are all non-empty. Then

X
µ(∅) = µ(∪i∈∅ Ai ) = ai = 0
i∈∅
since it is an empty sum. To prove countable additivity, suppose B1 , B2 , . . . are a sequence of

pairwise disjoint sets in σ(X ). Write Bn = ∪i∈Jn Ai for each n. Then the sets J1 , J2 , . . . are
pairwise disjoint, since if Jm = Jn for some m 6= n, picking some i ∈ Jm ∩ Jn we would have
Bm ∩ Bn ⊃ Ai 6= ∅, contradicting the disjointness of Bn and Bm . Hence
X ∞ X
X ∞
X
µ(∪∞ ∞
n=1 Bn ) = µ(∪n=1 ∪i∈Jn Ai ) = µ(∪i∈∪∞
n=1 Jn
Ai ) = ai = ai = µ(Bn )
i∈∪∞
n=1 Jn n=1 i∈Jn n=1
which gives the countable additivity.

13. Let F : (−∞, ∞) → R be a non-decreasing, right continuous function. Let I denote the set of
bounded half-open intervals in R (as in lectures). For I ∈ I, put
λF (I) = F (b) − F (a), where I has endpoints a and b.
(a) Check that λF (I) ≥ 0 for all I ∈ I.
(b) Show that λF is finitely sub-additive on I.
(c) Show that λF is finitely additive on I.
(d) Show that λF is countably additive on I.
(a) If I = ∅ then λF (I) = 0 (this should have been stated more clearly in the question).
If I = (a, b] for finite a < b, then λF (I) = F (b) − F (a) ≥ 0 since F is non-decreasing.
(b) We adapt the proof of LemmaP 4.3(i). We need to show that if A ⊂ ∪ki=1 Ai with all of
k
A, A1 , . . . , Ak in I, then λF (A) ≤ i=1 λF (Ai ). It is enough to prove this in the case where
A, A1 , . . . , Ak are all non-empty, so that we can and do write A = (a, b] and Ai = (ai , bi ] for
1 ≤ i ≤ k. We may also assume without loss of generality that bk ≥ max1≤i≤k−1 bi .
We prove the result by induction on k. If A ⊂ A1 then a ≥ a1 and b ≤ b1 , so F (a) ≥ F (a1 ) and
F (b) ≤ F (b1 ) (since F is assumed non-decreasing), and hence λF (A1 ) − λF (A) = F (b1 ) − F (a1 ) −
F (b) + F (a) ≥ 0, so the statement holds if k = 1.
Suppose now that k ≥ 2 and that the statement holds for k − 1.P Suppose A ⊂P∪ki=1 Ai . If
k−1 k
A∩Ak = ∅ then A ⊂ ∪k−1i=1 Ai so by the inductive hypothesis λF (A) ≤ i=1 λF (Ai ) ≤ i=1 λF (Ai ),
completing the induction in this case.
If A ∩ Ak 6= ∅ then ak < b ≤ bk and (a, ak ] ⊂ ∪k−1
i=1 Ai (where we define (a, ak ] = ∅ if ak ≤ a).
Hence by the inductive hypothesis and the fact that F is non-decreasing,
k−1
X k
X
λF (A) = (F (b) − F (ak )) + (F (ak ) − F (a)) ≤ λF (Ak ) + λF (Ai ) = λF (Ai ),
i=1 i=1

k
(c) We need to show that for allPkk ∈ N, if I = ∪i=1 Ii with I, I1 , . . . , Ik all in I and I1 , . . . , Ik
pairwise disjoint, then λF (I) ≤ i=1 λF (Ii ).
As in part (b), it suffices to prove this for the case where I, I1 , . . . , Ik are all non-empty. We
can then write I = (a, b] and Ii = (ai , bi ] for each i, and assume without loss of generality that
b1 ≤ b2 ≤ · · · ≤ b k .
Since I ⊂ ∪ki=1 Ii we have a1 ≤ a and bk ≥ b. Since I1 ⊂ I and Ik ⊂ I we have a1 ≥ a and bk ≤ b.
Thus a1 = a and bk = b.
For 1 ≤ i ≤ k − 1, we have bi ≤ ai+1 (because Ii ∩ Ii+1 = ∅) and bi ≥ ai+1 (because otherwise, the
interval (bi , ai+1 ) would be non-empty and contained in I \ ∪kj=1 Ij , contradicting the assumption
I = ∪kj=1 Ij ). Therefore bi = ai+1 . Hence
k k k−1
!
X X X
λF (Ii ) = (F (bi ) − F (ai )) = F (bk ) + (F (bi ) − F (ai+1 ) − F (a1 )
i=1 i=1 i=1
= F (b) + 0 − F (a) = λF (I),
as required. Alternatively the result can be proved by induction, similarly to part (b).
(d) We modify the proof of Theorem 4.4. Suppose A, A1 , A2 , . . . ∈ I with A ⊂ ∪∞ i=1 Ai . We may
assume A 6= ∅ (otherwise the subadditivity result is trivial). Write A = (a, b] and Ai = (ai , bi ] for
each i (we may omit all i with Ai = ∅ from the list of Ai ).
Let ε > 0 be fixed. Since F is right-continuous, we can find a0 ∈ (a, b) such that F (a0 ) < F (a) + ε,
and hence with A0 := (a0 , b] we have
λF (A) = F (b) − F (a) ≤ F (b) − F (a0 ) + ε = λF (A0 ) + ε. (1)
For each n = 1, 2, . . . , by the right continuity of F , we can find b0n > bn with F (b0n ) < F (bn ) + ε
2n
.
Then setting A0n := (an , b0n ] we have
ε ε
λF (A0n ) = F (b0n ) − F (an ) ≤ F (bn ) + n
− F (a) = λF (An ) + n . (2)
2 2
Now
[a0 , b] ⊂ A ⊂ ∪∞ ∞ 0
n=1 An ⊂ ∪n=1 (an , bn ).
Therefore, by compactness of [a0 , b] (Heine-Borel), there exists N ≥ 1 such that
[a0 , b] ⊂ ∪N 0
n=1 (an , bn ).
It follows that A0 = (a0 , b] ⊂ ∪N 0

n=1 (an , bn ], so by the finite subadditivity established in part (b),
N
X
0
λF (A ) ≤ λF (A0n ).
n=1
Hence we have
N
! N
! N ∞
!
(1) X (2) X X ε X
λF (A) ≤ λF ((A0 ) + ε ≤ λF (A0n ) + ε ≤ λF (An ) + ε + n
≤ λF (An ) + 2ε.
n=1 n=1 n=1
2 n=1
Since ε > 0 was arbitrary, we can let ε ↓ 0, and we obtain the result.
14. Recall from lectures that I is the class of half-open intervals in R and U is the class of finite
disjoint unions of sets in I. Show that U is an algebra.
Deduce that U is in fact the algebra generated by I (where generated algebras are defined analo-
gously to generated σ-algebras).
[Hint: It is easier to show first that U is closed under pairwise intersections rather than unions.]
Following the hint, first suppose A, B ∈ U and show A∩B ∈ U. Write A = ∪ki=1 Ii and B = ∪`j=1 Jj
with I1 , . . . , Ik pairwise disjoint intervals in I, and J1 , . . . , J` pairwise disjoint intervals in I. Then
A ∩ B = ∪ki=1 ∪`j=1 (Ii ∩ Jj ) = ∪ki=1 ∪`j=1 Ki,j

where we set Ki,j := Ii ∩ Jj for each (i, j). Since I is a π-system we have Ki,j ∈ I for each (i, j).
Also for (i0 , j 0 ) 6= (i, j) we have
Ki,j ∩ Ki0 ,j 0 = Ii ∩ Jj ∩ Ii0 ∩ Jj 0 = ∅,
because either i 6= i0 (so Ii ∩ Ii0 = ∅) or j 6= j 0 (so Jj ∩ Jj 0 = ∅).

Therefore the sets Ki,j , 1 ≤ i ≤ k, 1 ≤ j ≤ ` are pairwise disjoint so A ∩ B ∈ U.
Next we show U is closed under complementation. Given A ∈ U, write A = ∪ki=1 Ii with I1 , . . . , Ik
pairwise disjoint non-empty intervals in I, taken in order so that, writing Ii = (ai , bi i, we have
a1 ≤ a2 ≤ · · · ≤ ak . Since the intervals are disjoint we must have bi ≤ ai+1 for i = 1, . . . , k − 1.
Then setting b0 = −∞ and ak+1 = +∞, we have
Ac = ∪ki=0 (bi , ai+1 i,
which is a finite union of disjoint sets in I (some of which may be empty) and therefore is in U.
Finally we show U is closed under pairwise unions. Given A, B ∈ U we have (by the above) that
Ac ∈ U and B c ∈ U, so that by the De Morgan law,
(A ∪ B)c = Ac ∩ B c ∈ U
and hence also A ∪ B = ((A ∪ B)c )c ∈ U. So U is an algebra.

Let A be the algebra generated by I. Then A is the intersection of all algebras containing I so
A ⊂ U (since we’ve just shown that U is one such algebra).
Conversely, since A is an algebra we have for any I1 , . . . , Ik ∈ I that ∪ki=1 Ii ∈ A. Therefore U ⊂ A
so U = A.
15. Let µ be a measure on (R, B) with the property that µ(A) ∈ Z for all A ∈ B. Suppose also that
0 < µ(R) < ∞. Let k = µ(R), and for i ∈ N with 1 ≤ i ≤ k set
xi := inf{x ∈ R : µ((−∞, x]) ≥ i},
with the infimum of the empty set taken to be +∞.
(a) Show that µ((−∞, n]) = k for large enough n and µ((−∞, −n]) = 0 for large enough n.
Deduce that −∞ < xi < ∞ for each i.
(b) For 1 ≤ i ≤ k show that if x < xi then µ((−∞, x]) < i and if x > xi then µ((−∞, x]) ≥ i.
(c) Let F = {x1 , .P
. . , xk } (there may be some repetitions in this enumeration of F ). Show that
µ((−∞, x]) = kj=1 δxj ((−∞, x]) for all x ∈ R \ F .
(d) Let C be the collection of all intervals (−∞, x], x ∈ R \ F . Show that B ⊂ σ(C).
(e) Use the Uniqueness Lemma to show that µ(A) = ki=1 δxi (A) for all A ∈ B.
P
SOLUTION:
(a) By upwards continuity of measure we have limn→∞ µ((−∞, n]) = µ(R) = k, so we can find
some n0 such that µ((−∞, n]) ∈ (k −1/2, k +1/2) for n ≥ n0 . But then since µ((−∞, n]) ∈ Z
we have µ(−∞, n] = k for n ≥ n0 .
Note that µ((−∞, −1]) ≤ µ(R) < ∞. By downwards continuity of measure we have
limn→∞ µ((−∞, −n]) = µ(∅) = 0 so there exists n1 such that for n ≥ n1 we have µ((−∞, −n]) <
1/2 and hence µ((−∞, −n]) = 0 because µ((−∞, −n]) ∈ Z.
For 1 ≤ i ≤ k we have µ((−∞, n0 ]) = k ≥ i and hence xi ≤ n0 . We also have µ((−∞, −n1 ]) =
0 so µ(−∞, x] = 0 for x ≤ −n1 , and hence xi ≥ −n1 .
(b) By definition of inf, if µ((−∞, x]) ≥ i then x ≥ xi , and hence if x < xi then µ(−∞, x]) < i.
Again by definition of inf, if x > xi there exists y < x with µ((−∞, y]) ≥ i. But then
µ((−∞, x]) ≥ µ((−∞, y]) ≥ i.
(c) If µ((−∞, x]) = k then x ≥ xk so x ≥ xi for all i ≤ k, so ki=1 δxi ((−∞, x]) = k = µ(−∞, x].
P
If µ((−∞, x])P = 0 then x ≤ x1 by (b), so in fact x < x1 since x ∈ / F . Then x < xi for
k
1 ≤ i ≤ k, so i=1 δxi (x) = 0 = µ(−∞, x].
Suppose now that µ((−∞, x]) = j with 1 ≤ j < k. By (b), x ≥ xj (otherwise we’d have
µ((−∞, x]) < j) and x ≤ xj+1 (otherwise we’d have µ((−∞, x]) ≥ j + 1.) Since x ∈ / F,
therefore xj < x < xj+1 , so xi < x for i ≤ j and xi > x for i ≥ j, and hence
k
X
δxi ((−∞, x]) = j = µ((−∞, x]).
i=1
(d) We first show I ⊂ σ(C), where I is defined in lectures (just before Theorem 2.9). Since F is
finite, for all x ∈ R we can take xn ↓ x with xn ∈/ F . Then (−∞, x] = ∩∞n=1 (−∞, xn ] ∈ σ(C).
Hence for −∞ < a < b < ∞ we have (a, b] = (−∞, b] \ (−∞, a] ∈ σ(C). Thus I ⊂ σ(C).
Since σ(C) is a σ-algebra, using Theorem 2.9 we then have B = σ(I) ⊂ σ(C).
(e) First we note that kj=1 δxj is a measure by Question 9. For x, y ∈ R \ F we have (−∞, x] ∩
P
(−∞, y] P= (−∞, min(x, y)] which is in C, so that C is a π-system. By part (c) the measures
k
µ and j=1 δxj agree on all sets in C, so by the uniqueness lemma they agree on all sets in
σ(C), and hence (by part (d)) on all sets in B, as required. We can take the set Fn (in the
statement of the Uniqueness lemma) to be the sets (−∞, n], for each n ∈ N.
16. Suppose X is a non-empty set, A is an algebra in X and µ : A → [0, ∞] is a pairwise additive
Show that µ is finitely additive, i.e. if A1 , . . . , Ak ∈ A are pairwise disjoint then
set function. P
µ(∪i=1 Ai ) = ki=1 µ(Ai ) (pairwise additivity is the case k = 2 of this).
k
Proof by induction. Clearly the result is true for k = 1. Suppose it is true for some k ≥ 1, and
suppose A1 , . . . , Ak+1 ∈ A are pairwise disjoint. Set A = ∪ki=1 Ai and set B = Ak+1 . Then by the
pairwise disjointness,
B ∩ A = Ak+1 ∩ (∪ki=1 Ai ) = ∪ki=1 (Ak+1 ∩ Ai ) = ∅.
Therefore by the the assumed pairwise additivity, followed by the inductive hypothesis,
k
! k+1
X X
k+1
µ(∪i=1 Ai ) = µ(A ∪ B) = µ(A) + µ(B) = µ(Ai ) + µ(Ak+1 ) = µ(Ai ),
i=1 i=1
17. Show that if A ⊂ R is countable then A ∈ B and λ1 (A) = 0.
Let x ∈ R. For n ∈ N set Un = (x − 1/n, x + 1/n). Each Un is open so is in B (since B is the
σ-algebra generated by the collection O of open sets in R). Therefore since B is a σ-algebra, also
{x} = ∩∞
n=1 Un ∈ B.
Now suppose A ⊂ R is countable. Then we can write A = ∪∞ i=1 {xi } for some sequence of real
numbers (x1 , x2 , . . .) (in the case where A is finite we could take xi = x1 for all but finitely many
i). Since {xi } ∈ B for each i, and since B is a σ-algebra, we have
A = ∪∞
i=1 {xi } ∈ B.
Next we want to show λ1 (A) = 0. For x ∈ R, setting Jn = (x − 1/n, x + 1/n] we have {x} ⊂ Jn
so λ1 ({x}) ≤ λ1 (Jn ) = 2/n. Since n is arbitrarily large this shows that λ1 ({x}) ≤ 0. Then with
A and xi as above, since A = ∪∞ i=1 {xi }, using countable subadditivity of Lebesgue measure (see
Theorem 3.3) we have
∞
X
λ1 (A) ≤ λ1 ({xi }) = 0.
i=1
Also λ1 (A) ≥ 0 since λ1 is a measure. So λ1 (A) = 0.
18. Show that for any interval I with left endpoint a and right endpoint b we have λ1 (I) = b − a
(regardless of whether a, b ∈ I or not).
λ1 is defined as the unique measure on (R, B) such that λ1 ((a, b]) = b − a for all a < b. By
downwards continuity we have for any a ∈ R that
λ1 ({a}) = λ1 (∩∞
n=1 (a − 1/n, a]) = lim λ1 ((a − 1/n, a]) = lim (1/n) = 0.
n→∞ n→∞
Hence for finite a < b, since [a, b] = {a} ∪ (a, b] (disjoint union) we have
λ1 ([a, b]) = λ1 ((a, b]) + λ1 ({a}) = (b − a) + 0 = b − a.
Also, since (a, b) ∪ {b} = (a, b] (disjoint union) we have λ1 ((a, b)) + λ1 ({b}) = λ1 ((a, b]), so
λ1 (a, b) = λ1 ((a, b]) − λ1 ({b}) = (b − a) − 0.
Finally [a, b) = (a, b) ∪ {a} (disjoint union) so
λ1 ([a, b)) = λ1 ((a, b)) + λ1 ({a}) = (b − a) + 0.
Also λ1 ([a, ∞)) ≥ λ1 ((a, ∞)) = ∞, and λ1 (−∞, a) ≥ λ1 ((−∞, a − 1]) = ∞.
19. * Give an example of a Borel set A ⊂ R with λ1 (A) > 0 but with no non-empty open interval
contained in A.
We could take A = [0, 1) \ Q. Note that Q ∈ B since Q is countable. Also, the set Q ∩ (0, 1]
is countable, so λ1 (Q ∩ (0, 1]) = 0. Since A ∪ (Q ∩ (0, 1]) = (0, 1] (disjoint union), we have
λ1 (A) + λ1 (Q ∩ (0, 1]) = λ1 (0, 1] = 1, so
λ1 (A) = 1 − λ1 (Q ∩ (0, 1]) = 1 > 0.
However, A does not contain any non-empty open interval since Q is dense in the real line, so any
non-empty interval of the form (a, b) contains at least one element of A.
20. Given ε > 0, give an example of an open set U ⊂ R with λ1 (U ) < ε that is dense in R, i.e. has
non-empty intersection with every non-empty open interval in R.
Enumerate the rationals Q as Q = {q1 , q2 , q3 , . . .} (this can be done because Q is countably
infinite). Let δ ∈ (0, ε/2). For i ∈ N Set
δ δ
Ii := (qi − i
, qi + i ),
2 2
and set U := ∪∞ i=1 Ii . Then U is open since if x ∈ U , we can find i such that x ∈ Ii , but then since
Ii is an open interval we can find η > 0 with (x − η, x + η) ⊂ Ii ⊂ U .
Also, U is dense in R because Q is dense in R and Q ⊂ U .
Finally, by subadditivity of Lebesgue measure λ1 ,
∞ ∞
X X 2δ
λ1 (U ) ≤ λ1 (Ii ) = = 2δ,
i=1 i=1
2i
which is less than ε by our choice of δ.
21. Show that for any (not necessarily Borel) A ⊂ R there exists a Borel set B ⊂ R with A ⊂ B and
λ1 (B) = λ∗ (A). [The outer measure λ∗ (A) is defined in the same manner for non-Borel A as it is
for Borel A.]
By definition of Lebesgue outer measure λ∗ , if λ∗ (A) < ∞Pthen for each n ∈ N we can find a
covering In,1 , In,2 , . . . of A by half-open intervals, such that ∞ ∗
i=1 λ(In,i ) ≤ λ (A) + 1/n.
Let An = ∪∞ ∞
i=1 In,i . Since each interval In,i is in B, and B is a σ-algebra we have An = ∪i=1 In,i ∈ B.
That is, An is a Borel set. By subadditivity of measure,
∞
X
λ1 (An ) ≤ λ1 (Ii ) ≤ λ∗ (A) + n−1 .
i=1
Now set B = ∩∞ n=1 An . Then B ∈ B (i.e. B is a Borel set), because it is a countable intersection
of sets in B. Then A ⊂ B, but for all n we have B ⊂ An so λ1 (B) ≤ λ1 (An ) ≤ λ∗ (A) + 1/n.
Therefore λ1 (B) ≤ λ∗ (A), but also λ∗ (A) ≤ λ∗ (B) = λ1 (B) since A ⊂ B, so λ1 (B) = λ∗ (A).
If λ∗ (A) = ∞ then we can just take B = R.
22. Suppose A ⊂ R is a Borel set with λ1 (A) > 0. Using the fact that λ1 (A) = λ∗ (A), show that for
any ε > 0 there exists a non-empty interval I with λ1 (A ∩ I) ≥ (1 − ε)λ1 (I).
Without loss of generality we may assume A is bounded. For if not, then let An = A ∩ (−n, n].
By upwards continuity λ1 (An ) → λ1 (A) as n → ∞, and since λ1 (A) > 0, we can choose n with
λ1 (An ) > 0. If the result holds for this An then it also holds for A since An ⊂ A. Hence, from
now on we assume A is bounded.
Proof by contradiction; supose there exists ε > 0 such that for every interval I we have λ1 (I ∩A) <
(1 − ε)λ1 (I). Now fix this ε. Then for every bounded interval I we have
λ1 (I ∩ Ac ) = λ1 (I) − λ1 (I ∩ A) > ελ1 (I) ≥ ελ1 (I ∩ A).

We shall derive a contradiction. By the definition of outer measure λ∗ and the factP that λ∗ (A) =
∞
λ1 (A) > 0, we can cover A by a collection of intervals I1 , I2 , I3 , . . . such that i=1 λ1 (Ii ) ≤
(1 + ε/2)λ1 (A). Since we are now assuming A is bounded, all the intervals Ii are also bounded
(otherwise we have λ1 (Ii ) = ∞). Then
∞
X ∞
X
(1 + ε/2)λ1 (A) ≥ λ1 (Ii ) = (λ1 (Ii ∩ A) + λ1 (Ii \ A))
i=1 i=1
∞
X ∞
X
≥ (λ1 (Ii ∩ A) + ελ1 (Ii ∩ A)) = (1 + ε) λ1 (Ii ∩ A)
i=1 i=1
≥ (1 + ε)λ1 (A),
which is a contradiction since λ1 (A) > 0. We justify the final inequality above as follows. Since
Ai ⊂ ∪∞ i=1 Ii (because the Ii are a covering of A) we have A = P∪
∞
i=1 (Ii ∩ A) and therefore by
countable subadditivity of measure (Theorem 3.3 (iii)) λ1 (A) ≤ ∞ i=1 λ1 (Ii ∩ A).
23. Suppose X is a non-empty set and D is a π-system in X. Show that for any k ∈ N, if Ai ∈ D for
i = 1, 2, . . . , k then ∩ki=1 Ai ∈ D.
Proof by induction. The result is true for k = 1. Suppose it is true for some k. Suppose Ai ∈ D
for 1 ≤ i ≤ k + 1. Set B = ∪ki=1 Ai . By the inductive hypothesis B ∈ D. Since D is a π-system
k+1
∩i=1 Ai = B ∩ Ai+1 ∈ D
which completes the induction.

24. Suppose A ⊂ R is a bounded Borel set. Show that for all ε > 0 there exists a set U which is a
finite union of intervals, such that λ1 (A4U ) < ε, where A4U := (A \ U ) ∪ (U \ A).
Since λ1 (A) = λ∗ (A) (the Lebesgue outer measure of A), which is finite because A is bounded, by
the definiton of Lebesgue outer measure
P∞ there exists a sequence of half-open intervals I1 , I2 , I3 , . . .
∞
such that A ⊂ ∪i=1 Ii := S, and i=1 λ1 (Ii ) < λ1 (A) + ε/2.
By subadditivity of the measure λ1 (Theorem 3.3(iii)), λ1 (S) ≤ ∞
P
i=1 λ1 (Ii ) < λ1 (A) + ε/2. In
particular λ1 (S) < ∞.
For each n set Sn := ∪ni=1 Ii . Then Sn ⊂ Sn+1 for all n, and ∪∞
n=1 Sn = S. Therefore by the upward
continuity of the measure λ1 (Theorem 3.3(i)) we have λ1 (Sn ) → λ1 (S) as n → ∞, and since
λ1 (S) < ∞, we can (and do) choose N such that λ1 (SN ) > λ1 (S) − ε/2.
Set U = SN . Then A ⊂ S and U ⊂ S, so that A \ U ⊂ S \ U , and thus
λ1 (A \ U ) ≤ λ1 (S \ U ) = λ1 (S) − λ1 (U ) < ε/2,
and similarly
λ1 (U \ A) ≤ λ1 (S \ A) = λ1 (S) − λ1 (A) < ε/2.
Combining the last two displays shows that λ1 (U 4A) = λ1 (U \ A) + λ1 (A \ U ) < ε.
25. * Using either the Uniqueness lemma or the definition of Lebesgue outer measure, show that
λ1 has the scaling property: for any real number c 6= 0 and any Borel set B ∈ B, we have
λ1 (cB) = |c|λ1 (B). [Here cB := {cx : x ∈ B}.]
Define ν(B) := λ1 (cB), B ∈ B. It is easy to check that ν is a measure; indeed, ν(∅) = λ1 (c∅) =
λ1 (∅) = 0, and if A1 , A2 , . . . ∈ B are pairwise disjoint then so are cA1 , cA2 , . . . and hence
∞
X ∞
X
ν(∪∞
n=1 An ) = λ1 (c ∪∞
n=1 An ) = λ1 (∪∞
n=1 (cAn )) = λ1 (cAn ) = ν(An ).
n=1 n=1
Likewise, |c|λ1 (B) is a measure by Exercise 9 (c). We check that these two measures agree on the
class I of bounded half-open intervals: if I ∈ I, writing I = (a, b], we have
cI = (ca, cb] if c > 0; cI = [cb, ca) if c < 0.
In either case the length is |c|(b − a), so ν(I) = λ1 (cI) = |c|(b − a) = |c|λ1 (I). Also I is a π-system,
and R = ∪∞ n=1 (−n, n] (a countable increasing union of sets in I with finite λ1 -measure). Hence
the Uniqueness Lemma (Theorem 5.5) implies that ν(A) = |c|λ1 (A) for all Borel sets A.
ALTERNATVE METHOD using outer measure. Since B is Borel we have λ1 (B) = λ∗ (B), the
Lebesgue outer measure of B. Assume for now that λ1 (B) < ∞. P Then by definition of outer
∞
measure we can find intervals I1 , I2 . . . ∈ I such that B ⊂ ∪∞ I
i=1 i and i=1 λ(Ii ) < λ1 (B) + ε. But
∞
then also cB ⊂ ∪i=1 cIi and it is easy to see that λ1 (cI) = |c|λ(I) for all I ∈ I (using Question 18
when c < 0). Therefore by countable subadditivity of measure (Theorem 3.3(iii)),
∞
X ∞
X
λ1 (cB) ≤ λ1 (cIi ) = |c|λ(Ii ) ≤ |c|(λ1 (B) + ε).
i=1 i=1
Therefore since ε > 0 is arbitrary we have λ1 (cB) ≤ |c|λ1 (B), and this is also true when λ1 (B) =
∞. But applying the same argument using the set cB and the constant c−1 , we have that λ1 (B) =
λ1 (c−1 (cB)) ≤ |c−1 |λ1 (cB), and hence λ1 (cB) ≥ |c|λ1 (B). Combining the inequalities shows that
λ1 (cB) = |c|λ1 (B).
26. Suppose µ is a translation invariant measure on (R, B). Set γ := µ((0, 1]) and assume 0 < γ < ∞.
(a) Show that µ((0, 1/n]) = γ/n for all n ∈ N.
Since (0, 1] = ∪ni=1 ( i−1
n n
, i ] (disjoint union) and µ is a measure we have
n
X i−1 i 1
γ = µ((0, 1]) = µ(( , ]) = nµ((0, ])
i=1
n n n
where we have used tranlation invariance in the last step. This gives the result.
(b) Show that µ((0, q]) = γq for all rational q > 0.
i−1 i
For any such q we can write q = m/n with m ∈ N, n ∈ N. Then (0, q] = ∪m
i=1 ( n , n ] (disjoint
union) so that using translation invariance, and (a), we have
m
X i−1 i 1
µ((0, q]) = µ(( , ]) = mµ((0, ]) = m(γ/n) = γq.
i=1
n n n
(c) Let I 0 be the class of half-open intervals in R with rational endpoints, i.e. the class of intervals
of the form (q, r] with q ∈ Q, r ∈ Q and q ≤ r. Show that µ(I) = γλ1 (I) for all I ∈ I 0 .
For I = (q, r] with q < r and q, r ∈ Q, by translation invariance µ(I) = µ((0, r − q]). Since
r − q is also rational, we therefore have by part (b) that µ(I) = γ(r − q) = γλ1 (I). Obviously
if I = ∅ then µ(I) = γλ1 (I) = 0.
(d) Show that σ(I 0 ) = B. You may use without proof the fact that Q is dense in R, that is,
every non-empty open interval in R contains at least one rational number.
Let I be the class of bounded half-open intervals as defined in lectures. By a result from
lectures B = σ(I). In particular I ⊂ B. Therefore I 0 ⊂ I ⊂ B and hence (since B is a
σ-algebra) σ(I 0 ) ⊂ B.
Conversely, given I = (a, b] ∈ I, for all x ∈ (a, ∞) we can find rational q, r with a < q < x <
r, and hence
(a, ∞) = ∪{q,r∈Q:a<q<r<∞} (q, r]
which is a countable union of sets in I 0 , and therefore is in σ(I 0 ). Similarly (b, ∞) ∈ σ(I 0 ),
so I = (a, ∞) \ (b, ∞) ∈ σ(I 0 ). Hence I ⊂ σ(I 0 ). Therefore since σ(I 0 ) is a σ-algebra,
B = σ(I) ⊂ σ(I 0 ). Combined with the previous paragraph this shows σ(I 0 ) = B
(e) Use the Uniqueness lemma to show that µ(B) = γλ1 (B) for all B ∈ B.
Since λ1 is a measure on (R, B), also γλ1 is a measure on (R, B) (see Exercise 9(c)). By
part (c), the measures µ and γλ1 agree on I 0 which is a π-system (because if q, r, s, t are all
rational with q ≤ r and s ≤ t then (q, ri ∩ (s, ti = (max(q, s), min(r, t)i ∈ I 0 )
Moreover, setting Fn = (−n, n] we have Fn ∈ I 0 and Fn ⊂ Fn+1 for all n, and ∪∞ n=1 Fn = R
and γλ1 (Fn ) = 2γn < ∞ for all n.
Therefore we can apply the Uniqueness lemma to deduce that the measures γλ1 and µ agree
on σ(I 0 ) = B (using part (d)), that is γλ1 (B) = µ(B) for all B ∈ B.
27. Suppose F is a function with the properties assumed in Exercise 13.
(a) Prove that there is a unique measure µF on (R, B) with the property that µF ((a, b]) =
F (b) − F (a) for all a, b ∈ R with a < b. (You may assume without proof Carathéodory’s
extension theorem, along with the results of Exercise 13).
To get a semi-algebra in R we need to allow unbounded intervals. These were not covered in
Exercise 13 so instead we shall break up R into unit intervals.
Given n ∈ Z, set Wn := (n, n + 1]. Let I (n) denote the class of intervals I ∈ I such that
I ⊂ Wn . Then I (n) is a semi-algebra in Wn by a similar argument to Lemma 4.2. By Exercise
(n) (n)
13, λF is a pre-measure on (Wn , I (n) ), where λF ((a, b]) := F (b) − F (a), defined for all a, b
(n)
with n ≤ a ≤ b ≤ n + 1. Also λF (Wn ) < ∞.
Since the σ-algebra in Wn generated by I (n) is BWn (the collection of Borel subsets of Wn ),
(n) (n)
by Carathéodory’s Extension Theorem, λF has a unique extension to a measure µF on
(Wn , BWn ).
P (n) (n)
For A ∈ B set µF (A) = n∈Z µF (A ∩ Wn ). Since µF is a measure on (Wn , BWn ), also
(n)
A 7→ µF (A ∩ Wn ) is a measure on (R, B). Hence µF is a measure on (R, B) by Question
9(d). For any interval (a, b] ∈ I, we have (a, b] = ∪n∈Z ((a, b] ∩ Wn ), and only finitely many
of the sets in this union are non-empty. Hence from the finite additivity in Question 13,
X X (n)
λF ((a, b]) = λF ((a, b] ∩ Wn ) = µF ((a, b] ∩ Wn ) = µF ((a, b]),
n∈Z n∈Z
where the last equality holds because µ is a measure. Therefore µF does indeed agree with
λF on I. (Here λF is as defined in Question 13.)
(b) Given y ∈ R, show that the µF -measure of the one-point set {y} is µF ({y}) = F (y) − F (y−),
where F (y−) = limz↑y F (z).
Set An := (y − 1/n, y]. Then An ⊃ An+1 for all n, and An ∈ B for all n, and µF (A1 ) =
F (y) − F (y − 1) < ∞. Also {y} = ∩∞ n=1 An . Therefore by the downward continuity of the
measure µF , we have
µ({y}) = µ(∩∞
n=1 An ) = lim µ(An ) = lim (F (y) − F (y − 1/n)) = F (y) − F (y−).
n→∞ n→∞
(c) Show that µF ([a, b]) = F (b) − F (a−), and also find the formulas for µF ((a, b)) and µF ([a, b)),
when −∞ < a < b < ∞.
Since [a, b] = {a} ∪ (a, b], a disjoint union, using the previous part we have
µ([a, b]) = µ({a}) + µ((a, b]) = (F (a) − F (a−)) + (F (b) − F (a)) = F (b) − F (a−).
Also
µ((a, b)) = µ((a, b]) − µ({b}) = F (b) − F (a) − (F (b) − F (b−)) = F (b−) − F (a).
Finally µ([a, b)) = µ((a, b)) + µ({a}) = F (b−) − F (a−).
28. Suppose X is a non-empty set, S is a semi-algebra in X and π is a pre-measure on (X, S).
(a) Show that if k ∈ N and Ai ∈ S for i = 1, 2, . . . , k then ∪ki=1 Ai is a finite union of disjoint sets
in S.
Proof by induction on k. The statement is obviously true for k = 1. Suppose it true for given
k (and for all A1 , . . . Ak ∈ S). Suppose A1 , . . . , Ak+1 ∈ S. Using the inductive hypothesis,
write ∪ki=1 Ai = ∪`j=1 Bj with B1 , . . . , B` pairwise disjoint sets in S, and using the definition
of a semi-algebra write Ack+1 = ∪nm=1 Cm with C1 , . . . , Cn pairwise disjoint sets in S. Then
Ack+1 ∩ (∪ki=1 Ai ) = (∪nm=1 Cm ) ∩ ∪`j=1 Bj

= ∪1≤m≤n,1≤j≤` (Cm ∩ Bj )
which is a union of finitely many pairwise disjoint sets in S (using the fact that S is a π-
system so each Cm ∩ Bj is in S). Moreover each of the sets Cm ∩ Bj is disjoint from Ak+1
because Cm ⊂ Ack+1 . Hence
∪k+1 c k
i=1 Ai = Ak+1 ∪ (Ak+1 ∩ (∪i=1 Ai )) = Ak+1 ∪ ∪1≤m≤n,1≤j≤` (Cm ∩ Bj )
expresses ∪k+1
i=1 Ai as a finite union of pairwise disjoint sets in S.
Show that if A, A1 , . . . , Ak ∈ S with A1 , . . . , Ak pairwise disjoint and ∪ki=1 Ai ⊂ A, then
(b) P
k
i=1 π(Ai ) ≤ π(A).
First let us show that (∪ki=1 Ai )c is a finite union of disjoint sets in S. We do this by induction
on k. It is true for k = 1 by the definition of a semi-algebra. Suppose it true for given k (and
for all A1 , . . . Ak ∈ S). Suppose A1 , . . . , Ak+1 ∈ S. Using the inductive hypothesis, write
(∪ki=1 Ai )c = ∪`j=1 Bj with B1 , . . . , B` pairwise disjoint sets in S, and using the definition of a
semi-algebra write Ack+1 = ∪nm=1 Cm with C1 , . . . , Cn pairwise disjoint sets in S. Then
(∪ki=1 Ai )c = Ack+1 ∩ (∪ki=1 Ai )c = (∪nm=1 Cm ) ∩ (∪`j=1 Bj ) = ∪1≤m≤n,1≤j≤` (Cm ∩ Bj ),
which is a finite union of pairwise disjoint sets in S (using the fact that S is a π-system so
each Cm ∩ Bj is in S). This completes the induction.
Now suppose A, A1 , . . . , Ak ∈ S with A1 , . . . , Ak pairwise disjoint and ∪ki=1 Ai ⊂ A. By the
statement just proved, we may write (∪ki=1 Ai )c = ∪`j=1 Bj with B1 , . . . , B` pairwise disjoint
sets in S. Then
A = ∪ki=1 Ai ∪ A \ ∪ki=1 Ai = ∪ki=1 Ai ∪ (A ∩ ∪`j=1 Bj ) = ∪ki=1 Ai ∪ ∪`j=1 (A ∩ Bj )

which is a finite disjoint union of sets in S, so by finite additivity

k
X `
X k
X
π(A) = π(Ai ) + π(A ∩ Bj ) ≥ π(Ai ).
i=1 j=1 i=1
P∞
(c) Show that π is countably additive, i.e. π(∪∞ n=1 An ) = n=1 π(An ) whenever A1 , A2 , . . . ∈ S
are pairwise disjoint with ∪∞ A
n=1 n ∈ S.
Set A = ∪∞ previous part, for all k we have ki=1 π(Ai ) ≤ π(A)
P
A
n=1 n . Then A ∈ S, so by the P
and therefore taking the limit we have ∞ Pi )∞ ≤ π(A). Conversely since P
i=1 π(A π is a pre-
measure it is countably sub-additive so π(A) ≤ i=1 π(Ai ). Therefore π(A) = ∞ i=1 π(Ai )
as required.
29. Let I denote the class of half-open intervals in R, together with the empty set (as in the lecture
notes). Define the set-function: π : I → [0, ∞] by
(
0 if A = ∅;
π(A) :=
∞ A 6= ∅.
Show that π has more than one extension to a measure on B = σ(I). What condition (of the
uniqueness theorem) failed here?
One such extension would be the counting measure µ defined in Example 3.2(a). Since any non-
empty interval has infinitely many elements, this µ is also an extension of π to B.
However, for any constant c ∈ (0, ∞] the set function cµ is also a measure, which also extends π.
Since π(F ) = ∞ for all nonempty F ∈ I, there is no sequence of sets Fn ∈ I with π(Fn ) < ∞
and Fn ↑ R, so Theorem 5.5 (Uniqueness lemma) is not applicable.
30. Suppose that (X, M, µ) and (Y, N , ν) are σ-finite measure spaces and let R denote the class
of measurable rectangles in X × Y . Define the ‘area’ function ρ(A × B) := µ(A)ν(B) for all
measurable rectangles A × B in X × Y . Verify that ρ is σ-finite on R, i.e. that there exist sets
Fn ∈ R such that Fn ⊂ Fn+1 and ρ(Fn ) < ∞ for all n ∈ N, and ∪∞ n=1 Fn = X × Y .
Since (X, M, µ) and (Y, M, µ) are σ-finite measure spaces, there are sets Gn ∈ M and Hn ∈ N
defined for n ∈ N, with µ(Gn ) < ∞ and ν(Hn ) < ∞ for all n, and with ∪∞ n=1 Gn = X and
∪∞ H
n=1 n = Y .
n n
For each n ∈ N, set G̃n := ∪Pi=1 Gi and set H̃n := ∪i=1 Hi . Then for all n we have G̃n ⊂ G̃n+1 ,
n
and G̃n ∈ M, and µ(G̃n ) ≤ i=1 µ(Gn ) < ∞; likewise H̃n ⊂ H̃n+1 , and H̃n ∈ N , and ν(H̃n ) ≤
P n ∞ ∞
i=1 ν(Hn ) < ∞. Also ∪n=1 G̃n = X and ∪n=1 H̃n = Y .
Now set Fn := G̃n × H̃n , for each n ∈ N. Then for all n we have Fn ⊂ Fn+1 , and Fn ∈ R, and
ρ(Fn ) = µ(G̃n )ν(H̃n ) < ∞, and ∪∞
n=1 Fn = X × Y , as required.
31. Let (X, M) be a measurable space. Suppose f : X → [0, ∞) and g : X → [0, ∞) are measurable
functions. Define the set A ⊂ X × R × R by A := {(x, s, t) : f (x) > s, g(x) > t}. Let B denote
the Borel σ-algebra in R. Show that A ∈ M ⊗ B ⊗ B.
We have
A = ∪q,r∈Q {(x, s, t) : f (x) > q > s, g(x) > r > t}

= ∪q,r∈Q (f −1 ((q, ∞)) ∩ g −1 ((r, ∞)) × (−∞, q) × (−∞, r))
which is a countable union of sets in M ⊗ B ⊗ B and hence is itself in M ⊗ B ⊗ B.
32. Prove that if W ⊂ R is a Borel set, and f : W → R is an increasing function, then f is Borel-
measurable.
We need to show for all α ∈ R that f −1 ((α, ∞]) ∈ B.
Fix α ∈ R. Let T := f −1 ((α, ∞]) = {x ∈ R : f (x) > α}. If T = ∅ then T ∈ B. So assume T is
non-empty and let t = inf(T ).
If x ∈ W with x > t, then there exists y < x with y ∈ T , so f (y) > α, so f (x) > α since f is
increasing, so x ∈ T .
If x < t, then x ∈
/ T since t is a lower bound for T .
Therefore T is either the set (t, ∞) ∩ W or the set [t, ∞) ∩ W . But both (t, ∞) and [t, ∞) are in
B, and also we assume W ∈ B, so (t, ∞) ∩ W ∈ B and [t, ∞) ∩ W ∈ B. Thus T ∈ B.
33. Suppose (X, M) is a measurable space and Y is a set and f : X → Y is a function. Let
F := {E ⊂ Y : f −1 (E) ∈ M}, and prove F is a σ-algebra in Y .
We check the conditions for F to be a σ-algebra:
• f −1 (Y ) = {x ∈ X : f (x) ∈ Y } = X ∈ M, so Y ∈ F.
• If A ∈ F, then f −1 (Y \ A) = {x ∈ X : f (x) ∈ Y \ A} = {x ∈ X : f (x) ∈
/ A} = X \ f −1 (A) ∈
M, so Y \ A ∈ F.
• If A1 , A2 , A3 , . . . ∈ F, then f −1 (∪∞ ∞ ∞
n=1 An ) = {x ∈ X : f (x) ∈ ∪n=1 An } = ∪n=1 {x ∈ X :
∞ −1 ∞
f (x) ∈ An } = ∪n=1 f (An ) ∈ M, so ∪n=1 An ∈ F.
Thus, F is indeed a σ-algebra in Y .
34. (a) Let (X, M) be a measurable space, and let fn : X → R be measurable functions. Show that
the set of points
{x ∈ X : lim fn (x) exists in R}
n→∞
is in M.
(b) Taking (Ω, F, P) to be a probability space, and random variables (i.e., measurable functions)
Y1 , Y2 , . . . : Ω → R show that for any constant µ ∈ R the set:
( n
)
1X
ω ∈ Ω : lim Yi (ω) = µ
n→∞ n
i=1
1
Pn
is in F. Deduce that expressions like P[limn→∞ n i=1 Yi = µ] are meaningful.
(a) The complement of the set in question is:
{x ∈ X : lim inf n→∞ fn (x) = +∞} ∪ {x ∈ X : lim supn→∞ fn (x) = −∞}

∪ {x ∈ X : lim inf n→∞ fn (x) < lim supn→∞ fn (x)}
=: A+ ∪ A− ∪ B.
Hence it is enough to show that each of the sets A+ , A− , B is in M.

We have ∞ n
\ o
A+ = x ∈ X : lim inf fn (x) > k .
n→∞
k=1
We have shown in Thm 9.12 that lim inf n→∞ fn is measurable, hence each set of the intersection
above is in M. Therefore, A+ ∈ M. By a similar argument, A− ∈ M.
We write B as
S
B = r∈Q {x ∈ X : lim inf n→∞ fn (x) < r < lim supn→∞ fn (x)}
S
= r∈Q ({x ∈ X : lim inf n→∞ fn (x) < r} ∩ {x ∈ X : r < lim supn→∞ fn (x)})
(where Q is the set of rational numbers). This represents B as a countable union of sets that are
the intersection of two sets. For each fixed r ∈ Q, the two sets are in M, since lim inf n→∞ fn and
lim supn→∞ fn are measurable. Hence B ∈ M, and the statement is proved.
(b) RVs are measurable functions by definition, so fn := n1 ni=1 Yi , n = 1, 2, . . . are measurable
P
functions by Theorems 7.2.1 and 7.2.2. We have
{ω ∈ Ω : lim fn (ω) = µ} = {ω ∈ Ω : lim sup fn (ω) = µ} ∩ {ω ∈ Ω : lim inf fn (ω) = µ}.

n→∞ n→∞ n→∞
Since lim supn→∞ fn is measurable, and the first set is the inverse image of the Borel set {µ}
under this function, this set is in F. Similarly, the second set is also in F. Hence the expression
P[limn→∞ n1 ni=1 Yi = µ] is meaningful, because P is defined on the set in question.
P
35. Let (X, M) be a measurable space.

(a) Show that if E ∈ M, then its indicator function 1E defined by 1E (x) = 1 for x ∈ E and
1E (x) = 0 for x ∈ / E, is a measurable function.
(b) Let f : XP→ R be function with finite range f (X) = {α1 , . . . , αn } (with α1 , . . . , αn distinct),
so that f = ni=1 αi 1Ai , where Ai = {x ∈ X : f (x) = αi }. Show that f is measurable if and only
if A1 , . . . , An ∈ M.
(a) Let f = 1E . Then 

X
 if α < 0
−1
f ((α, ∞]) = E if 0 ≤ α < 1

if α ≥ 1

∅
and since ∅ ∈ M and X ∈ M, if E ∈ M we have for all α ∈ R that f −1 ((α, ∞]) ∈ M, so f is
measurable.
(b) First suppose Ai ∈ M for 1 ≤ i ≤ n. Then each of the functions 1Ai is measurable by part
(a). ThereforePαi 1Ai is also measurable (for each i) by Corollary 9.9. Hence by Theorem 9.10 the
function f = ni=1 αi 1Ai is measurable.
Conversely, suppose Ai ∈ / M for some i. Then for this choice of i we have f −1 ({αi }) = Ai ∈
/ M,
and therefore f is not measurable, since if it were measurable, by Theorem 9.6 we would have
f −1 (E) measurable for all Borel E ⊂ R, in particular for E = {αi } (which is a Borel set).
36. * Suppose (X, M, µ) is a measure space and f : X → [0, ∞] is measurable.
(a) Prove that if a ∈ (0, ∞) then µ(f −1 [a, ∞]) ≤ a−1 f dµ.
R
(b) Prove that if f dµ = 0, then µ(f −1 ((0, ∞])) = 0.

R
(a) Since f is measurable the set A := f −1 ([a, ∞]) = ∩∞ n=1 f

−1
((a − 1/n, ∞]) is in M (being a
countable intersection of sets in M). Alternatively, A ∈ M by Theorem 9.6.
We claim A × (0, a) ⊂ hyp(f ). Indeed, for (x, y) ∈ A × (0, a) we have f (x) ≥ a > y so
(x, y) ∈ hyp(f ). Therefore
Z
f dµ = (µ ⊗ λ1 )(hyp(f )) ≥ (µ ⊗ λ1 )(A × (0, a)) = µ(A)λ(0, a) = aµ(A),
and hence µ(f −1 ([a, ∞])) = µ(A) ≤ a−1 f dµ as claimed.

R
(b) Now suppose f dµ = 0. Then since f −1 ((0, ∞]) = ∪∞ −1

R
n=1 f ([(1/n), ∞]), by part (a) and
countable subadditivity of measure (Theorem 3.3 (iii)) we have
∞
X X∞ Z
−1 −1
µ(f ((0, ∞])) ≤ µ(f ([(1/n), ∞])) ≤ n f dµ = 0.
n=1 n=1
R∞ R∞
37. (a) Suppose g : R → R is integrable and t ∈ R. Show that −∞ g(x − t)dx = −∞ g(x)dx.
R b+t Rb
(b) Deduce that for any a, b ∈ R with a < b, a+t g(x − t)dx = a g(x)dx.
R∞ R∞
(c) Let c > 0. Show that −∞ g(cx)dx = c−1 −∞ g(x)dx.
R R
(a) Set h(x) = g(x − t). We need to show R hdλ1 = R gdλ1 , where λ1 is Lebesgue measure.
First suppose g is nonnegative and simple (this is probably better
Pn than starting with indicator
functions as in the hint). By Lemma 10.7(a) we can write g = i=1 αi 1Ai with all of the αi ≥ 0
and Ai ∈ B.PFor all x ∈ R, note that x − t ∈ A ⇔ x ∈ A + t, and hence 1A (x − t) = 1A+t (x).
Hence h = ni=1 αi 1Ai +t . Then using Lemma 10.7(b) and also the translation invariance of λ1
(Theorem 7.1) we have
Z n
X n
X Z
hdλ1 = αi λ1 (Ai + t) = αi λ1 (Ai ) = gdλ1 .
R i=1 i=1 R
Now suppose g is nonnegative. Let (gn )n≥1 be a sequence of simple functions with 0 ≤ gn ↑ g
pointwise (see Theorem 9.15). Set hn (x) = gn (x + t) for x ∈ R. Then hn is simple and 0 ≤ hn ↑ h
pointwise, so by MON and the previous case,
Z Z Z Z
hdλ1 = lim hn dλ1 = lim gn dλ1 = gdλ1 .
R n→∞ R n→∞ R
For general integrable g we have that +

R +h (x) =
+
R g+ (x − t) for
R all x, andR h− (x − t) = g − (x − t) for
all x. Hence by the previous case h dλ1 = g dλ1 and h− dλ1 = g − dλ1 . Hence
Z Z Z Z Z Z
+ − + −
hdλ1 = h dλ1 − h dλ1 = g dλ1 − g dλ1 = gdλ1 .
(b) Set f (x) := g(x)1(a,b) (x). Then f (x − t) = g(x − t)1(a+t,b+t) (x). Hence by part (a),
Z b+t Z ∞ Z ∞ Z ∞ Z b
g(x − t)dx = g(x − t)1(a+t,b+t) (x)dx = f (x − t)dx = f (x)dx = g(x)dx.
a+t −∞ −∞ −∞ a
(c) First suppose g ≥ 0 and g is simple. Then using Lemma 10.7(a) we can write g = ni=1 ai 1Ai
P
for some a1 , . . . , an ≥ 0 and A1 , . . . , An ∈ B pairwise disjoint. We then also have for all x ∈ R
that n n
X X
g(cx) = ai 1Ai (cx) = ai 1c−1 Ai (x),
i=1 i=1
where cA := {cx : x ∈ Ai }. Using Lemma 10.7(b) and Exercise 25 we obtain that

Z ∞ Xn n
X Z ∞
−1 −1 −1
g(cx)dx = ai λ1 (c Ai ) = c ai λ1 (Ai ) = c g(x)dx
−∞ i=1 i=1 −∞
Now suppose g ≥ 0. Then we can take gn simple with gn ↑ g pointwise. Hence gn (cx) ↑ g(cx) for
all x ∈ R. Then by MON and the case already considered
Z Z Z Z
−1 −1
g(cx)dx = lim gn (cx)dx = lim c gn (x)dx = c g(x)dx.
n→∞ n→∞
Finally for general integrable g, applying the previous case to g + and g − we have
Z Z Z Z Z Z
+ − −1 + −1 − −1
g(cx)dx = g (cx)dx − g (cx)dx = c g (x)dx − c g (x)dx = c g(x)dx.
Rh
38. Suppose f : R → R is continuous at 0. Show that h−1 0
f (x)dx → f (0) as h ↓ 0.
Given ε > 0, choose δ > 0 such that |f (x) − f (0)| ≤ ε for all x ∈ [0, δ]. Then for x ∈ [0, δ] we have
g1 (x) := f (0) − ε ≤ f (x) ≤ f (0) + ε := g2 (x)
so that for h ∈ [0, δ] we have

Z h Z h Z h
g1 (x)dx ≤ f (x)dx ≤ g2 (x)dx,
0 0 0
Rh Rh
that is, h(f (0) − ε) ≤ 0 f (x)dx ≤ h(f (0) + ε), and therefore f (0) − ε ≤ h−1 0 f (x)dx ≤ f (0) + ε,
which shows the desired convergence.
39. Suppose
R∞ f : R → R is integrable. Show that there exists integrable g : R → R such that
−∞
|f (x) − g(x)|dx < ε, and g has bounded support (i.e., there exists n ∈ N with g(x) = 0
whenever |x| > n).
Setting fn := |f |1(−n,n) we have fn ↑ |f | pointwise so by MON, we have as n → ∞ that
Z n Z Z
f (x)dx = fn dλ1 → |f |dλ1 < ∞
−n
RN R∞
so we can choose N such that −N |f (x)|dx > −∞ |f (x)|dx − ε.
Take g = f 1(−N,N ) . Then g has bounded support and
Z ∞ Z N Z ∞ Z ∞ Z N
|f (x) − g(x)|dx = |f (x)|dx + |f (x)|dx = |f (x)|dx − |f (x)|dx < ε.
−∞ −∞ N −∞ −N
40. Suppose f : R → R is integrable (with respect to Lebesgue measure). The purpose of this question
is to show in stages that
R∞
(*) Given ε > 0 there exists a step function g : R → R such that −∞ |g(x) − f (x)|dx < ε.
(a) Show that (*) holds in the special case where f = 1A for some bounded Borel set A.
(b) Show that (*) holds when f is simple and nonnegative with bounded support
(c) Show that (*) holds when f (x) ≥ 0 for all x and f has bounded support.
(d) Show that (*) holds whenever f (x) ≥ 0 for all x (and f is integrable).
(e) Finally, show that (*) holds whenever f is integrable.
(a) Assume f = 1A for some bounded Borel set A. By Question 24, given ε > 0 we can find a
set U which is a finite union of bounded intervals such that λ(A4U ) < ε/2. Also we can take
these intervals to be half-open, and then since U is an algebra by Qusestion 14, the set U is in
U and therefore is in fact a finite union of pairwise disjoint half-open bounded intervals, denoted
I1 , . . . , Ik say (alternatively we can see that the intervals may be taken disjoint using Question 28
(a)).
Pk
Clearly R∞ 1 U = i=1 1Ii is a stepR function, and since |1A (x) − 1U (x)| = 1A4U (x) for all x ∈ R we
have −∞ |1A (x) − 1U (x)|dx = 1A4U (x)dx = λ1 (A4U ) < ε, so that (*) holds taking g = 1U .
(b) Now suppose f is simple, f ≥ 0. By Lemma 10.7 we can write f = ì=1 ai 1Ai with Ai all
P
bounded and measurable, and ai ≥ 0 for all i. Assume the ai are not all zero (otherwise f ≡ 0
whichR is itself a step function). Then using part (a) we can find step P functions h1 , . . . , h` such
that |hi − 1Ai | ≤ ε/(` max(a1 , . . . , a` )) for each i. Then setting h = ì=1 ai hi gives us a step
function with
Z Z X `
X Z
|f − h|dλ1 = ai (1Ai − hi ) dλ1 ≤ ai |1Ai − hi |dλ1 ≤ ε.
i=1
(c) Suppose f ≥ 0 with bounded support, and with f ∈R L1 . TakeR fn simple with 0 ≤ fn ≤ f
pointwise and fn ↑ fR pointwise. Then by MON, we have fn dλ1 ↑ f dλ1 , so given ε > 0 we can
R n such that |f − fn |dλ1 ≤ ε/2. Fix this n. By part (b), we can choose a step function g
choose
with |g − fn |dλ1 < ε/2, and then
Z Z Z
|f − g|dλ1 ≤ |f − fn |dλ1 + |fn − g|dλ1 < ε.
1
R f ≥ 0 and
(d) Suppose R f ∈ L . Then taking fn = f 1[−n,n] we have
R that 0 ≤
R fn ↑ f pointwise, so
by MON fn dλ1 → f dλ1 . Given ε > 0 we can choose fn with fn dλ1 > f dλ1 − ε/2. Fix this
n.
R
Since fn has bounded support, by part (c) we can find a step function g with |fn − g|dλ < ε/2.
Then Z Z Z
|f − g|dλ1 ≤ |f − fn |dλ1 + |fn − g|dλ1 < ε.
This gives us (*) in this case.
1 + 1 − 1
R f ∈ +L , we have f ∈
(e) For general R L and−f ∈ L . By part (d) we can find step functions
h1 , h2 with |h1 − f |dλ1 < ε/2 and |h2 − f |dλ1 < ε/2, so that g := h1 − h2 is a step function
with
Z Z Z
|g − f |dλ1 = |h1 − h2 − (f − f )|dλ1 ≤ (|h1 − f + | + |f − − h2 |)dλ1 < ε.
+ −
41. (a) Let (X, M) and (Y, N ) be measurable spaces. Show that that for all A ⊂ X × Y with
A ∈ M ⊗ N , and all y ∈ Y , the horizontal cross-section A[y] of A defined by
A[y] := {x ∈ X : (x, y) ∈ A}
satisfies A[y] ∈ M.
(b) Suppose f : X → [0, ∞] is such that hyp(f ) ∈ M ⊗ B. Show that f is a measurable function.
(a) Fix y ∈ Y . Let F be the collection of A ⊂ X × Y such that A[y] ∈ M.

We claim that F is a σ-algebra. Indeed, ∅[y] = ∅ ∈ M, so ∅ ∈ F. Also, if A ∈ F, then
(Ac )[y] = {x ∈ X : (x, y) ∈ Ac } = {x ∈ X : (x, y) ∈ A}c = (A[y] )c ∈ M,
so Ac ∈ F. Also if An ∈ F for n = 1, 2, 3, . . ., then setting A = ∪∞

n=1 An we have
A[y] = {x ∈ X : (x, y) ∈ ∪∞ ∞ ∞
n=1 An } = ∪n=1 {x ∈ X : (x, y) ∈ An } = ∪n=1 ((An )[y] ) ∈ M,
so A ∈ F. Thus we have verified the claim.

We claim also that R ⊂ F, where R is the collection of measurable rectangles in X × Y .
Indeed if A ∈ M and B ∈ N then
(
A if y ∈ B
(A × B)[y] =
∅ otherwise.
By the two preceding claims, F is a σ-algebra with R ⊂ F, and therefore M ⊗ N = σ(R) ⊂

F. In other words, every A ∈ M ⊗ N is in F, which is what we needed.
(b) Suppose f : X → [0, ∞] with hyp(f ) ∈ M ⊗ B. Then for all y > 0,
f −1 ((y, ∞]) = {x ∈ X : f (x) > y} = {x ∈ x : (x, y) ∈ hyp(f )} = (hyp(f ))[y]
which is in M by part (a). If y < 0 then f −1 ((y, ∞]) = X ∈ M, and for y = 0 we have
f −1 ((0, ∞]) = ∪∞
n=1 f
−1
((1/n, ∞]) ∈ M.
Therefore f −1 ((y, ∞]) ∈ M for all y ∈ R, so f is measurable.
42. Let (Ω, F, µ) be a probability space. Let Rf : Ω → [0, ∞] be measurable, i.e. f is a nonnegative
random variable. For t ≥ 0 define L(t) := Ω e−tf (ω) µ(dω) (the Laplace transform of f ).
(a) Show that limt→∞ L(t) = µ({ω ∈ Ω : f (ω) = 0}). Here we make the convention that e−∞ = 0.
(b) Show that limt↓0 L(t) = µ({ω ∈ Ω : f (ω) < ∞}).
(c) Show that limt↓0 (t−1 (L(0) − L(t))) = f dµ if the integral on the right is finite . [Hint: use
R
the fact that 1 − e−x ≤ x for x ≥ 0].

What about if the integral is infinite?
(a) Fix a sequence tn ↑ ∞. Let ω ∈ Ω. Then
(
1 if f (ω) = 0;
lim e−tn f (ω) =
n→∞ 0 if f (ω) > 0.
Since tn > 0 for all n, we have the domination |e−tn f (ω) | ≤ 1, and here 1 ∈ L1 (µ), since
R
Ω
1 dµ =
µ(Ω) = 1 < ∞. Therefore, by the Dominated Convergence Theorem
Z Z
−tn f n→∞
L(tn ) = e dµ −→ 1C dµ = µ(C),
Ω Ω
where C = {ω ∈ Ω : f (ω) = 0}. Since this convergence holds for any choice of tn with tn ↑ ∞, it
follows that L(t) → µ(C) as t → ∞, as required.
[Here we are using: if F : [0, ∞) → R is a function and a ∈ R are such that F (tn ) → a for any
sequence (tn )n∈N with tn ↑ ∞ as n → ∞, then F (t) → a as t → ∞.]
(b) Fix a sequence tn ↓ 0. We have
(
1 if f (ω) < ∞;
lim e−tn f (ω) =
n→∞ 0 if f (ω) = ∞.
As in RPart (a), since tn > 0 for all n, we have the domination |e−tn f (ω) | ≤ 1, and here 1 ∈ L1 (µ),
since Ω 1 dµ = µ(Ω) = 1 < ∞. Therefore, by the Dominated Convergence Theorem
Z Z
−tn f n→∞
L(tn ) = e dµ −→ 1D dµ = µ(D),
Ω Ω
where D = {ω ∈ Ω : f (ω) < +∞}. Since our choice of tn satisfying tn ↓ 0 is arbitrary, it follows
that L(t) → µ(D) as t ↓ 0, as required.
[Here we are using: if F : (0, ∞) → R is a function and a ∈ R with F (tn ) → a for any sequence
(tn )n∈N with tn ↓ 0 as n ↓ 0, then F (t) → a as t ↓ 0. This is similar to the fact that if a function
is sequentially continuous at 0, then it is continuous at 0, which you should have seen in first year
Analysis.]
(c) Let us take a sequence tn ↓ 0. Then
Z Z
t−1
n (L(0) − L(tn )) = t−1
n (1
−tn f (ω)
−e )µ(dω) = gn dµ
Ω
where we set gn (ω) := t−1 −tn f (ω)
n (1 − eR ). Then gn (ω) → f (ω) as n → ∞, and also (using the
hint) gn (ω) ≤ f (ω). Therefore if f dµ < ∞ we can use Dominated convergence with dominating
function f to deduce that
Z Z
−1
lim tn (L(0) − L(tn )) = f (ω)µ(dω) = f dµ.
n→∞ Ω
R
If f dµ = ∞ we can no longer use Dominated Convergence (DOM). However, in fact we have
0 ≤ gn (ω) ≤ gn+1 (ω) for all ω, so we Rcan use Monotone Convergence (MON) to deduce that R in
−1 −1
this case limn→∞ tn (L(0) − L(tn )) = f dµ = ∞. Thus in all cases tn (L(0) − L(tn )) → f dµ
as n → ∞. Since this holds for any sequence tn ↓ 0, it follows that t−1 (L(0) − L(tn )) → f dµ as
R
t ↓ 0.
43. Let (X, M, µ) be a measure space.
R
(a) Show that if f : X → [−∞, ∞] is measurable, E ∈ M, E |f | dµ = 0, then f = 0 a.e. on E.
R
(b) Show that if f ∈ L1 (µ) with E f dµ = 0 for all E ∈ M, then f = 0 a.e. on X.
R R
(c) Show that if f ∈ L1 (µ) with X f dµ = X |f | dµ, then either f ≥ 0 a.e. on X, or f ≤ 0 a.e.
on X.
(d) Show that if f : X → R and g : X → R are measurable functions, then {x ∈ X : f (x) 6=
g(x)} ∈ M.
R
(a) By the assumption given, setting g = |f |1E we have gdµ = 0. Also g ≥ 0 pointwise. There-
fore by Question 36 (b) we have µ(g −1 ((0, ∞])) = 0, so that µ({x ∈ E : f (x) 6= 0} = 0, or in other
words f = 0 µ-a.e. on E.
R
(b) Suppose f ∈ L1 (µ) with E
f dµ = 0 for all E ∈ M.
R
Take E = {x ∈ X : f (x) ≥ 0}. Then f 1 E ≥ 0 pointwise and by the stated condition f 1E dµ =
−1
R
E
f = 0. Hence by Question 36 (b), µ({x ∈ X : f (x)1 E (x) > 0}) = 0, that is, µ(f ((0, ∞])) = 0.
R R
Also setting g = −f , we have E gdµ = − E f dµ = 0 for all E, so by the same argument as above
we have µ(g −1 ((0, ∞])) = 0, that is, µ(f −1 ([−∞, 0))) = 0. Therefore
µ({x ∈ X : x 6= 0}) = µ(f −1 ((0, ∞]) ∪ f −1 ([−∞, 0))) ≤ µ(f −1 ((0, ∞])) + µ(f −1 ([−∞, 0))) = 0,
so µ({x ∈ X : x 6= 0}) = 0, or in other words f = 0 a.e.
(c) Suppose f ∈ L1 (µ) with X f dµ = X |f | dµ. Set I + = f + dµ and I − = f − dµ. Then
R R R R
I + ≥ 0, I − ≥ 0 and since |f | = f + − f − , our assumption tells us that

|I + − I − | = I + + I −
which fails unless either I + = 0 or I − = 0 (or both). But if I + = 0 then by Question 36 (b) we
have f + = 0 µ-almost everywhere, or in other words f ≤ 0 µ-a.e. Similarly, if I − = 0 then by
Question 36 (b) we have f − = 0 µ-almost everywhere, or in other words f ≥ 0 µ-a.e.
(d) By Corollary 9.9 and Theorem 9.10 the function f − g is measurable, so by Lemma 9.5 the
set {x ∈ X : f (x) 6= g(x)} = (f − g)−1 (R \ {0}) is in M.
summable, i.e. ∞
P
44. Suppose f : N → R is absolutely n=1 |f (n)| < ∞. Let µ be counting measure on
R P ∞
(N, P(N)). Show that N f dµ = i=1 f (i).
First suppose f ≥ 0 (pointwise). For n ∈ N, define fn (i) = f (i) for i ≤ n, with fn (i) = 0 for
i ≥ n. Then
n
X
fn = f (i)1{i}
i=1
so fn is simple, so by Lemma 10.7,

Z n
X n
X
fn dµ = f (i)µ({i}) = f (i),
i=1 i=1
since µ is counting measure so µ({i}) = 1 for each i.

Since we assume f ≥ 0 we have fn ↑ f pointwise and so by MON,
Z Z n
X ∞
X
f dµ = lim fn dµ = lim f (i) = f (i).
n→∞ n→∞
i=1 i=1
Pn
Now drop the assumption f ≥ 0. Define fn as before. Then we still have f = i=1 f (i)1{i} , so
by linearity we still have
Z n
X Z n
X
fn dµ = f (i) 1{i} dµ = f (i).
i=1 i=1
Set g(i) = |f (i)| for each i ∈ N. Then |fn | ≤ g pointwise and also g ≥ 0 so by the case already
considered,
Z ∞
X
gdµ = g(i)
i=1
which is finite by our assumption that f is absolutely summable. Also, fn → f pointwise. There-
fore we can apply DOM to obtain that
Z Z n
X ∞
X
f dµ = lim fn dµ = lim f (i) = f (i).
n→∞ n→∞
i=1 i=1
45. Suppose f : R → R is in L1 . Let ε > 0. Using R ∞ Question 40, show there exists a continuous
function g : R → R such that kf − gk1 < ε, i.e. −∞ |f (x) − g(x)|dx < ε.
Pk
R ∞ Question 40 we can find a step function w : R → R of the form w = i=1 ai 1Ii such that
By
−∞
|w(x) − f (x)|dx < ε/2, where a1 , . . . , ak ∈ R and I1 , . . . , Ik are intervals.
Suppose 1 ≤ i ≤ k. Set hi := ai 1Ii . Denote the left endpoint of II by bi and the right endpoint
by ci . Take hi,n (x) = hi (x) for x ∈ Ii and for x ≤ bi − 1/n, and for x ≥ ci + 1/n, with the value
of hi,n interpolated linearly between x = bi − 1/n and x = bi , and the value of hi,n interpolated
linearly between x = ci and x = ci + 1/n.
R∞
Then hi,n is continuous and |hi,n −hi | ≤ |ai |1[bi −1/n,bi ]∪[ci ,ci +1/n] pointwise so that −∞ |hi,n −hi |dx ≤
R∞
2|ai |/n. Thus khi,n − hi k1 → 0 as n → ∞, where we set kF k1 = −∞ |F (x)|dx for all F ∈ L1 .
Set hn (x) = ki=1 hi,n (x). Then hn is continuous and since w = ni=1 hi , using Theorem 15.5
P P
repeatedly we have
Xk Xn
khn − wk1 = k (hi,n − hi )k1 ≤ khi,n − hi k1
i=1 i=1
which tends to zero as n → ∞. Hence we can choose N such that khN − wk1 < ε/2. Then hN is
continuous and khN − f k1 ≤ khN − wk1 + kw − f k1 < ε. So taking g = hN gives the function g
we want.
46. (a) Let d ∈ N. Define Lebesgue measure on Rd .
(b) RSuppose f : [0, 1] → [0, ∞) is a non-negative Borel function. How is the Lebesgue integral
1
0
f (x)dx defined?
R1
(c) Suppose f : [0, 1] → R is a Borel function. How is the Lebesgue integral 0 f (x)dx defined?
(d) Let f : [0, 1] → R be continuously differentiable with derivative f 0 (x), x ∈ [0, 1] (interpreted
as a 1-sided derivative for x = 0 or x = 1). Define the curve γ ⊂ R2 by γ := {(x, f (x)) : x ∈
[0, 1]}.
For n ∈ N and i ∈ {0, 1, . . . , n} set xi,n = i/n and yi,n = f (xi,n ). Let γn (a polygonal
approximation to γ) be the union of the straight line segments (in R2 ) from (xi−1,n , yi−1,n )
to (xi,n , yi,n ), 1 ≤ i ≤ n. Let |γn | denote the total length of these line segments. The aim of
the next two parts is to show that
R1p
limn→∞ |γn | = 0 1 + f 0 (x)2 dx. (∗)
(i) Use the mean value theorem to show that there exist numbers zi,n defined for i =
1, 2, . . . , n with zi,n ∈ (xi−1,n , xi,n ) such that
n q
X 1
|γn | = 1 + f 0 (zi,n )
i=1
n
(ii) Using the previous part, complete the proof of (∗).

Q
d
(a) Lebesgue measure is the unique measure λd on the Borel σ-algebra in Rd such that λd (a ,
i=1 i i b ] =
Qd
i=1 (bi − ai ) for all a1 < b1 , . . . , ad < bd .
R1
(b) We define 0 f (x)dx = λ2 ({(x, y) : 0 < y < f (x)}).
R1 R1
(c) Set f + (x) = max(f (x), 0) and f − (x) = max(−f (x), 0). Then 0 f (x)dx = 0 f + (x) −
R1 −
0
f (x)dx with the convention ∞ − x = ∞, x − ∞ = −∞ and ∞ − ∞ = 0 (or undefined).
p
(d) (i) The length of the ith segment of γn , denoted ì,n say, is equal to (yi,n − yi−1,n )2 + (xi,n − xi−1,n )2 .
By the mean value theorem
i+1 i
yi,n − yi−1,n = f ( ) − f ( ) = (1/n)f 0 (zi,n )
n n
for some zi,n ∈ (xi−1,n , xi,n ). Hence
q q
ì,n = ((1/n)f (zi,n )) + (1/n) = (1/n) 1 + f 0 (zi,n )2
0 2 2
Pn Pn p
and γn = i=1 ì,n = i=1 (1/n) 1 + f 0 (zi,n )2 .
p R xi,n
(ii) Set gn (x) := 1+ f 0 (zi,n )2 for x ∈ [xi−1,n , xi,n ). Then ì,n = xi−1,n
gn (x)dx so
n
X Z 1
|γn | = ì,n = gn (x)dx.
i=1 0
p x ∈ [0, 1) choose i so that xi−1,n ≤ x < xi,n and set zn (x) = zi,n . Then gn (x) =0
For
1 + f 0 (zn (x))p
2 and |z (x) − x| ≤ 1/n, so z (x) → x as n → ∞, and by continuity of f
n n
also gn (x) → 1 + f (x) as n → ∞. Also setting K = supx∈[0,1] |f 0 (x)| we have K < ∞
0 2
(since the√continuous image of aR compact interval is compact: this was Lemma 12.4) and
1√
|gn (x)| ≤ 1 + K 2 , and clearly 0 1 + K 2 dx < ∞ . Hence by the dominated convergence
theorem, Z 1 Z 1p
lim |γn | = lim gn (x)dx = 1 + f 0 (x)2 dx
n→∞ n→∞ 0 0
as required.
47. Let f : R → R be integrable. Suppose {hn }n≥1 is a sequence in R such that hn → 0.
RK
(a) Show that for any K ∈ (0, ∞) we have −K |f (x + hn ) − f (x)|dx → 0 as n → ∞. [Hint: first
suppose f is continuous, and recall that any continuous real-valued function on a compact
interval is bounded.]
R∞
(b) Show that −∞ |f (x + hn ) − f (x)|dx → 0 as n → ∞.
(a) First assume f is continuous. In that case, setting gn (x) = |f (x + hn ) − f (x)| we have
gn → 0 pointwise as n → ∞, and (assuming n is large enough so that |hn | ≤ 1) |gn (x)| ≤
|fn (x) + fn (x + hn )| ≤ 2M , where we set M = sup−K−1≤x≤K+1 |f (x)| which is finite by the
RK
hint. Since −K (2M )dx = 4KM < ∞, by DOM we have
Z K Z K
|f (x + hn ) − f (x)|dx = gn (x)dx → 0 as n → ∞.
−K −K
In the general case (f maybe not continuous) Rwe have to use Question 45. Given ε > 0,
∞
take w : R → R such that w is continuous and −∞ |w(x) − f (x)|dx < ε/3, and set wn (x) =
RK
w(x + hn ). Then −K |wn (x) − w(x)|dx → 0 as n → ∞ by the argument just given. Also, for
all n, using Question 37 we have
Z ∞ Z ∞ Z ∞
|wn (x) − fn (x)|dx = |w(x + hn ) − f (x + hn )|dx = |w(x) − f (x)|dx < ε/3,
−∞ −∞ −∞
so for large enough n, setting fn (x) = f (x + hn ) we have

Z K Z K Z K Z K
|fn − f |dx ≤ |fn − wn |dx + |wn − w|dx + |w − f |dx < ε.
−K −K −K −K
R
(b) Let ε > 0 and choose K such that |f (x)|dx < ε. This can be done by the solution
R\[−K,K]
R K+1
to Question 39. Set fn (x) = f (x + hn ). Then using (b), choose N so that −(K+1) |fn (x) −
f (x)|dx < ε for all n ≥ N , and also |hn | ≤ 1 for all n ≥ N . Then for n ≥ N , we have
Z ∞ Z Z K+1
|fn (x) − f (x)|dx ≤ |fn (x) − f (x)|dx + |fn (x) − f (x)|dx
−∞ [−K−1,K+1]c −K−1
Z Z
≤ |f (x + hn )|dx + |f (x)|dx + ε
[−K−1,K+1]c [−K−1,K+1]c
Z
≤2 |f (x)|dx + ε ≤ 3ε
[−K,K]c
which gives the result.

48. Let f : [0, 1] → [0, ∞) be a bounded Borel function. Suppose for each n ∈ N that we are given
kn ∈ N and numbers
0 = x0,n < x1,n < x2,n < · · · < xkn ,n = 1,
with ∆n := max{xi,n − xi−1,n : 1 ≤ i ≤ n} → 0 as n → ∞. Set
mi,n := inf f (x); and Mi,n := sup f (x);

x∈(xi−1,n ,xi,n ] x∈(xi−1,n ,xi,n ]
then the lower Riemann sum of the subdivision is `n := ki=1

Pn
mi,n (xi,n − xi−1,n ) and the upper
Pkn
Riemann sum of the subdivision is un := i=1 Mi,n (xi,n − xi−1,n ).
Define simple functions gn , hn : (0, 1] → [0, ∞) by
kn
X kn
X
gn = mi,n 1(xi−1,n ,xi,n ] ; hn = Mi,n 1(xi−1,n ,xi,n ] .
i=1 i=1
R1 R1
(a) Show that 0
gn (x)dx = `n and 0
hn (x)dx = un .
(b) Assume now that {x ∈ [0, 1] : f is not continuous at
R 1 x} is a Borel subset of R with zero
Lebesgue measure. Show that limn→∞ `n = limn→∞ un = 0 f (x)dx.
(a) The functions gn and hn are simple so by the general formula for the integral of a simple
function (see Lemna 10.7),
Z 1 kn
X kn
X
gn = mi,n λ1 ((xi,n , xi+1,n ]) = mi,n (xi−1,n − xi,n ) = `n ,
0 i=1 i=1
and similarly
Z 1 kn
X
hn = Mi,n (xi−1,n − xi,n ) = un .
0 i=1
(b) Let x ∈ (0, 1). If f is continuous at x then gn (x) → f (x) as n → ∞ and hn (x) → f (x) as
n → ∞. Therefore by the assumption given we have gn → f λ1 -almost everywhere on (0, 1).
Therefore
R by theR Dominated convergence
R theorem
R (stronger version in Corollary 12.9) we have
`n = gn dλ1 → f dλ1 and un = hn dλ1 → f dλ1 . We can take our dominating function to be
a constant since f is assumed bounded.
49. (a) Show that {(x, y) ∈ R2 : x < y} ∈ B ⊗ B.

(b) Let c ∈ (0, ∞). Show that {(x, y) ∈ R2 : x < y ≤ x + c} ∈ B ⊗ B.
(c) Suppose µ is a probability measure on (R, B). For x ∈ R, let F (x) = µ((−∞, x]).
R∞
Let c ∈ R. Use Fubini’s Theorem to show that −∞ (F (x + c) − F (x))dx = c.
(a) Given b ∈ R let Bb := {(x, y) ∈ R2 : x + b < y}. Then
Bb = ∪q∈Q {(x, y) ∈ R2 : x + b < q < y} = ∪q∈Q ((−∞, q − b) × (q, ∞)),

so that Bb is a countable union of sets in B2 and therefore is itself in B2 .
Taking b = 0 gives us the result.
(b) With Bb as defined above, given c ∈ (0, ∞) we have
{(x, y) ∈ R2 : x < y ≤ x + c} = B0 \ Bc ∈ B2 .
(c) First suppose c ≥ 0. We shall apply Fubini’s theorem (in fact Tonelli’s theorem) to the product
of the measure spaces (R, B, λ1 ) (where λ1 is Lebesgue measure) and (R, B, µ). For (x, y) ∈ R × R
set f (x, y) = 1 if x < y ≤ x + c (or equivalently, if y − c ≤ x < y), and otherwise f (x, y) = 0.
Then f ≥ 0 and f is (B ⊗ B)-measurable by part (b), so we can apply Tonelli’s theorem. We have
Z Z Z Z
f (x, y)µ(dy) λ1 (dx) = 1(x,x+c] (y)µ(dy) λ1 (dx)
R R R R
Z ∞ Z ∞
= [µ((−∞, x + c] − µ((−∞, x])]dx = [F (x + c) − F (x)]dx (3)
−∞ −∞
and by Tonelli’s theorem, this is equal to the same double integral taken in the opposite order,
which comes to
Z Z Z Z Z
f (x, y)λ(dx) µ(dy) = 1[y−c,y) (x)λ1 (dx) µ(dy) = cµ(dy) = c (4)
R R R R R
where the last equality is because µ is a probability measure. The equality between the expressions
(3) and (4) gives us the result, for the case c > 0.
Now we consider the case with c < 0. Put C = −c, and set G(x) = F (x + c) − F (x). Then by
Question 37,
Z ∞ Z ∞ Z ∞ Z ∞
G(x)dx = G(x + C)dx = (F (x) − F (x + C))dx = − (F (x + C) − F (x))dx
−∞ −∞ −∞ −∞
and by the case of this result that we already proved, the last expression equals −C = c as
required.
50. Let (X, M, µ) be a probability space, and let f : X → [0, ∞) be a measurable function.
(a) Let h(t) = µ(f −1 ((t, ∞))) for t ∈ [0, ∞). Show that h is a decreasing, right continuous function
of t.
(b) Let g : X × [0, ∞) → R be defined by
g(x, t) = 1{f (x)>t} , (x, t) ∈ X × [0, ∞).
Show that g is M ⊗ B-measurbale.

(c) Use Fubini’s theorem to show that
Z Z ∞
f dµ = h(t)dt.
X 0
(a) If t < u then f −1 ((u, ∞)) ⊂ f −1 ((t, ∞)) so h(u) = µ(f −1 ((u, ∞))) ≤ µ(f −1 ((t, ∞))) = h(t), so
h is decreasing.
If tn ↓ t then setting An = f −1 ((tn , ∞)) we have An ⊂ An+1 for all n and ∪∞
n=1 An = f
−1
((t, ∞)) :=
A. Hence by upward continuity of measure,
lim h(tn ) = lim µ(An ) = µ(∪∞
n=1 An ) = µ(A) = h(t)
n→∞ n→∞
which shows that h is right continuous (this argument is similar to the solution to Question 10).
(b) By Question 35 (a), it suffices to show that the set {(x, t) ∈ X × [0, ∞) : f (x) > t} is in the
product σ-algebra M ⊗ B. This is the case becuase
{(x, t) ∈ X×[0, ∞) : f (x) > t} = ∪q∈Q {(x, t) : f (x) > q > t} = ∪q∈Q [f −1 ((q, ∞))×[0, q)] ∈ M⊗B.
(c) We apply Fubini’s (in fact Tonelli’s) theorem to the function g which is nonnegative and
product-measurable by the previous part. Let λ be Lebesgue measure on the Borel sets in [0, ∞).
We have
Z Z Z Z ∞ Z Z
g(x, t)λ(dt) µ(dx) = 1[0,f (x)) (t)dt µ(dx) = f (x)µ(dx) = f dµ, (5)
X [0,∞) X 0 X X
while
Z Z Z Z
g(x, t)µ(dx) λ(dt) = 1(t,∞) (f (x))µ(dx) λ(dt)
[0,∞) X [0,∞) X
Z Z ∞
−1

= µ(f (t, ∞)) λ(dt) = h(t)dt (6)
[0,∞) 0
and by Tonelli’s theorem, the expressions in (5) and (6) are equal.
51. Let (X, M) be a measurable space and suppose f : X → [0, ∞] and g : X → [0, ∞] are Borel
functions. Show that
Z ∞Z ∞ Z
µ({x ∈ X : f (x) > s, g(x) > t})dsdt = f (x)g(x)µ(dx).
0 0 X
By Question 36 (c), the set A ⊂ X × R × R given by A = {(x, s, t) : f (x) > s, g(x) > t} is in
M ⊗ B ⊗ B. Therefore the function 1A is measurable with respect to M ⊗ B ⊗ B, by Question
35. Therefore by Tonelli’s theorem,
Z ∞Z ∞ Z ∞Z ∞Z
µ({x ∈ X : f (x) > s, g(x) > t})dsdt = 1A ((x, s, t))µ(dx)dsdt
0 0 0 0 X
Z Z ∞Z ∞
= 1A ((x, s, t))dsdtµ(dx)
X 0 0
Z Z ∞Z ∞
= 1[0,f (x)) (s)1[0,g(x)) (t)dsdtµ(dx)
X 0 0
Z Z ∞
f (x)1[0,g(x)) (t)dtµ(dx)
X 0 Z
f (x)g(x)µ(dx).
X
P∞ 4/3
52. Suppose (aP
n )n≥1 and (bn )n≥1 are sequences of nonnegative numbers, such that A := n=1 an < ∞
and B := ∞ 4
n=1 bn < ∞. Show that
∞
X
an bn ≤ A3/4 B 1/4
n=1
(you may use results from lectures without proof).

We apply Hölder’s inequality on the measure space (N, P(N), µ), where µ is the counting measure
on (N, P(N)), with p = 4/3 and q R= 4 so 1/pP+ 1/q = 1. Setting
R p f (n) = anR and g(n) = bn for
q
n ∈ N, using Question 44 we have f gdµ = n an bn , and f dµ = A and g dµ = B. Hence
by Hölder’s inequality
∞
X Z Z 1/p Z 1/q
p q
an b n = f gdµ ≤ f dµ g dµ = A1/p B 1/q .
n=1
53. (a) Let α ∈ R be a fixed constant. Let f (x) = xα for x ∈ (0, 1]. Determine the values of p ∈ [1, ∞)
(depending on α), such that f ∈ Lp ([0, 1]).
(b) Let α ∈ R, and let g(x) = xα for x ∈ [1, ∞). Determine the values of p ∈ [1, ∞) (depending
on α) such that g ∈ Lp ([1, ∞)).
p
P∞Let p ∈ [1, ∞) and let f ∈ L (R). Let (an )n≥1 and (bn )n≥1 be real-valued sequences such that
(c)
n=1 |an | < ∞. Show that the sequence of functions
n
X
fn (x) := ak f (x − bk )
k=1
converges in Lp (R).
R1
(a) To have f ∈: Lp ([0, 1]) we need to have 0 |f (x)|α dx < ∞. Since f (x) = xα > 0, this condition
R1 R1
amounts to 0 xαp dx < ∞. If αp > −1 then 0 xαp dx = [xαp+1 ]10 /(αp + 1) < ∞. If αp < −1,
we get the same indefinite Rintegral but now the integral comes to +∞ because xαp+1 diverges at
1
x = 0. If αp = −1 then 0 xαp dx = [log x]10 = +∞. To sum up, f ∈ Lp ([0, 1]) if and only if
αp > −1.
R∞ R∞
(b) Similarly to (a), to have g ∈ LpR([1, ∞)) we need 1 xαp dx < ∞. If αp = −1 then 1 xαp dx =
∞ αp
[log x]∞
1 = +∞. If αp 6= −1 then 1 x dx = [x
αp+1 ∞
]1 /(αp + 1) which is finite if αp < −1 but
p
infinite if αp > −1. So this time we have g ∈ L ([1, ∞)) if and only if αp < −1.
(c) We show that the sequence is a Cauchy sequence in Lp . For each k set gk (x) = f (x − bk )
for x ∈ R. Then gk ∈ Lp with kgk kp = kf kp by Question 37. By Minkowski’s inequality and a
straightforward induction argument, we have for any h1 , . . . , hj ∈ Lp that kh1 + h2 + · · · hj kp ≤
P j
i=1 khi kp . Hence for any n < m we have
m
X m
X
kfm − fn kp = ak gk ≤ kak gk kp
k=n+1 p k=n+1
and since kαhkp = |α|khkp for any real α and any h ∈ Lp , we therefore have for n < m that
m
X m
X ∞
X
kfm − fn kp ≤ |ak |kgk kp = kf kp |ak | ≤ kf kp |ak |
k=n+1 k=n+1 k=n+1
which tends to zero as n → ∞ since it is the tail of a convergent series. This shows that fn is a
Cauchy sequence in Lp , so by the Riesz-Fischer theorem fn converges in Lp to a limit function in
Lp .
54. Suppose that (X, M, µ) is a measure space, and 1 ≤ p < q < ∞.
(a) Show that if µ is a probability measure and f ∈ Lq (µ), then kf kp ≤ kf kq .

[Hint: note that f = f · 1, and apply Hölder’s inequality]
(b) Show that if µ(X) < ∞ then Lq (µ) ⊂ Lp (µ).
(c) Give an example to show that if µ(X) = ∞, then we might not have Lq (µ) ⊂ Lp (µ).
(a) Following the hint, set g(x) = 1 for all x ∈ X. Assume f ∈ Lq (µ). Then f p = f p · g, so by
Hölder’s inequality Z
p
kf kp = |f |p dµ = kf p · gk1 ≤ k(f p )kq/p kgkr
where
R r is the conjugate exponent to q/p. Since g ≡ 1 and µ is a probability measure, kgkr =
( 1dµ)1/r = 1, so
Z p/q Z p/q
p p q/p q
kf kp ≤ (|f | ) = |f | = kf kpq
so the result follows.

(b) If µ(X) is finite (but now not equal to 1), we can still take g ≡ 1 as above; with r as above
we now have that kgkr = µ(X)1/r < ∞. Therefore repeating the argument above shows that if
f ∈ Lq (µ) then
kf kpp ≤ kf kpq × kgkr < ∞
so that f ∈ Lp (µ) as required.
(c) We can use part (b) of the previous question. Take X = [1, ∞) with Lebesgue measure and
f (x) = x−β with q −1 < β ≤ p−1 , so that pβ ≤ 1 < qβ. Then f ∈ Lq \ Lp , so here we do not have
Lq ⊂ Lp .
55. * Let (X, M, µ) be a σ-finite measure space. Let p ∈ (1, ∞).

P∞Suppose f : X → R and (for all
n ∈ N) fn : X → R are measurable functions, and assume n=1 kfn kp < ∞. For all n ∈ N and
x ∈ X, set
n
X X ∞
gn (x) = |fk (x)| and g∞ (x) = |fk (x)|.
k=1 k=1
(i) Show that kgn kp → kg∞ kp as n → ∞, and deduce that kg∞ kp < ∞.
(ii) Show that the function h(x) := ∞
P
n=1 fn (x) is well-defined and finite µ-a.e., that is, the sum
converges for µ-a.e. x ∈ X.
(i) By using Minkowski’s inequality repeatedly we have

n
X n
X ∞
X
kgn kp ≤ k|fk |kp = kfk kp ≤ kfk kp ,
k=1 k=1 k=1
p
which is finite by assumption. Also for each x ∈ X we have that ( nk=1 |fk (x)|) is nonnega-
P
tive and nondecreasing in k. Therefore by definition and by MON,
Z n
!p Z n
!p
X X
kgn kpp = |fk | dµ → |fk | dµ, as n → ∞.
X k=1 X k=1
Therefore this limit is finite (being the limit of a bounded sequence). The limit in the display
above is equal to kg∞ kpp . Taking pth roots gives us the required result that kgn kp → kg∞ kp
and kg∞ kp < ∞.
R
(ii) Clearly g∞ (x) ≥ 0 for all x ∈ X, and the previous part we have (g∞ (x))p µ(dx) < ∞, so by
Lemma 10.5(b), g(x) < ∞ for µ-almost all x ∈ X.
P∞
Therefore
P∞ for µ-almost all x ∈ X, we have n=1 |fn (x)| is convergent so that the sum
f
n=1 n (x) converges (absolutely).
P∞
[In case you need reminding about absolute convergence: Fix x such that Pn|fn (x)| < ∞.
n=1
that ∞
P
Given ε > 0 we can choose N so P n=N |f (x)|
n P < ε. Taking s n := k=1 fk (x), for
m m
m > n ≥ N we have |sm − sn | = | k=n+1 fk (x)| ≤ k=n+1 |f Pk (x)| < ε. Hence (sn )n∈N is a
Cauchy sequence so converges to a finite limit, i.e. the series ∞ n=1 fn (x) is convergent.]
56. (a) (i) Let n ∈ N and set [n] = {1, . . . , n}. Show that if f ∈ `∞ ([n]) then kf k∞ = limp→∞ kf kp .
(ii) Is the corresponding statement true for `∞ (N), that is, is it always true for f ∈ `∞ (N) that
kf k∞ = limp→∞ kf kp ?
(b) Suppose X is finite or countable, equipped with counting measure. Show that for any f :
X → R and p ∈ [1, ∞) we have kf kp ≥ kf k∞ .
(a) (i) Suppose f ∈ `∞ ([n]). Write fi for f (i), i ∈ [n]. Then using Question 44 we have kf kp =
1/p
( ni=1 |fi |p ) , for 1 ≤ p < ∞, and kf k∞ = max(|f1 |, . . . , |fn |). Choose j ∈ [n] so that this
P
maximum is achieved at |fj |. Then
n
X
kf kpp = |fi |p ≥ |fj |p = kf kp∞ .
i=1
On the other hand |fi | ≤ |fj | for all i ∈ [n], so

n
X
kf kpp ≤ |fj |p = nkf kp∞ .
i=1
Hence kf k∞ ≤ kf kp ≤ n1/p kf k∞ , and since n1/p → 1 as p → ∞ this shows that kf kp → kf k∞ as
p → ∞.
(ii) No, we could take f (i) = 1 for all i ∈ N. Then on the measure space (N, P(N), µ) with µ
being counting measure, we have kf k∞ = 1 but kf kp = ∞ for all p.
(b) Again writing fi for f (i), if kf k∞ = 0 there is nothing to prove (since kf kp ≥ 0 by definition)
so assume kf k∞ > 0. Given ε > 0, we can choose j with |fj | > (1 − ε)kf k∞ . Then
∞
X
kf kpp ≥ |fi |p ≥ |fj |p > (1 − ε)kf kp∞ .
i=1
Hence kf kp ≥ (1 − ε)1/p kf k∞ , and since ε > 0 is arbitrary this shows that kf kp ≥ kf k∞ .
57. Let n ∈ N and m ∈ N. Recalling Definitions 8.8 and 8.11 from the notes,
(a) show that Bn ⊗ Bm = Bn+m ;
(b) show that λn ⊗ λm = λn+m .
(a) Define Rd as in Definition 8.8, the collection of ‘bounded boxes’ in Rd . Each element of Rn+m
is the Cartesian product of an element of Rn and an element of Rm , and hence is in the collection
of sets A := {A × B : A ∈ Bn , B ∈ Bm }. That is, Rn+m ⊂ A ⊂ σ(A) = Bn ⊗ Bm . Hence
Bn+m = σ(Rn+m ) ⊂ Bn ⊗ Bm .
We have to show the reverse inclusion Bn ⊗ Bm ⊂ Bn+m . Since Bn ⊗ Bm is generated by the
measurable rectangles in Rn ×Rm , it is enough to show that every measurable rectangle in Rn ×Rm
is an (n + m)-dimensional Borel set.
We show this in steps. First define
Bem = {B ∈ B m : Rn × B ∈ Bn+m }.
It is clear that Rm ⊂ Bem , since the product of Rn and an element of Rm is an (n + m)-dimensional

box, which is in Bn+m . We check that Bem is a σ-algebra in Rm .
• It is clear that Rm ∈ Bem , since Rn × Rm = Rn+m ∈ Bn+m .
• If B ∈ Bem , then we can write
Rn × (Rm \ B) = (Rn × Rm ) \ (Rn × B) = Rn+m \ (Rn × B) ∈ Bn+m ,

| {z }
∈Bn+m
showing that Rm \ B ∈ Bem as well.

• If B1 , B2 , · · · ∈ Bem , we have
Rn × (∪∞ ∞ n
N =1 BN ) = ∪N =1 (R × BN ) ∈ Bn+m ,
| {z }
∈Bn+m
showing that ∪∞ N =1 BN ∈ Bm .
e
Therefore, Bem is a σ-algebra, and it contains Rm , so it contains Bm . Hence we have shown that
if B ∈ Bm , then Rn × B ∈ Bn+m .
Next, by virtually the same argument we may show that if A ∈ Bn , then A × Rm ∈ Bn+m .
Combining these two statements, and using the fact that Bn+m is closed under pairwise intersec-
tions, we have that for A ∈ Bn and B ∈ Bm that
A × B = (A × Rm ) ∩ (Rn × B) ∈ Bn+m .
This shows that all measurable rectangles in Rn × Rm are in Bn+m , and hence Bn ⊗ Bm ⊂ Bn+m .
This completes the proof of part (a).
(b) We use the Uniqueness lemma (Theorem 5.5). The collection Rn+m is a π-system, and every
set in Rn+m is of the form A × B with A ∈ Rn , B ∈ Rm . For such a set, we have
(λn ⊗ λm )(A × B) = λn (A)λm (B) = λn+m (A × B).
Thus Rn+m is a π-system that generates Bn+m , with the measures λn ⊗ λm and λn+m equal on all
sets in Rn+m Also the sets AN := (−N, N ]n+m are in Rn+m , satisfy λn+m (AN ) < ∞ and AN ⊂
AN +1 for all N , and also ∪∞
N =1 AN = R
n+m
. Hence by the Uniqueness lemma, λn ⊗ λm = λn+m on
the σ-algebra generated by the (n + m)-dimensional boxes, which is Bn+m .
58. Suppose V (·) were a volume function on subsets of R3 satifying conditions 1-7 described in Section
0.3 (pretending that such a V existed). Show that V (·) would satisfy 0 < V (B) < ∞, where B is
the unit ball as in Section 0.3.
We did not say whether B is the open ball {x ∈ R3 : 0 < |x| < 1} or the closed ball {x ∈ R3 : 0 <
|x| ≤ 1}, so let us argue in a manner that covers both cases.
By Assumption 5 (pairwise additivity) and induction we have V (∪ni=1 Ai ) = ni=1 V (Ai ) whenver
P
A1 , . . . , An are pairwise disjoint. To spell out the inductuion: The statement is true for n = 1, and
suppose for some general n it is true. Then given A1 , . . . , An+1 pairwise disjoint, set D = ∪ni=1 Ai .
Then D and An+1 are disjoint, so by Assumption 5 and the inductive hypothesis,
n+1
X
V (∪n+1
i=1 Ai ) = V (D ∪ An+1 ) = V (D) + V (An+1 ) = V (Ai ),
i=1

Also, for any A ⊂ A0 ⊂ Rd , we have V (A0 ) = V (A) + V (A0 \ A) ≥ V (A).
We have B ⊂ (−2, 2]3 = ∪64 d
i=1 Ai where A1 , . . . , A64 are a collection of disjoint translates of (0, 1] .
By translation invariance (Assumption 4) and Assumption 3, we have for each i that
V (Ai ) = V ((0, 1]d ) ≤ V ([0, 1]d ) = 1
so
64
X
3
V (B) ≤ V ((−2, 2] ) ≤ V (Ai ) ≤ 64 < ∞.
i=1
To show V (B) > 0,plet C0 = [−1/2, 1/2]3 . Then C0 ⊂ B since the vector (1/2, 1/2, 1/2) has length
|(1/2, 1/2, 1/2)| = 1/4 + 1/4 + 1/4 < 1. Hence, by translation invariance
V (B) ≥ V (C0 ) = V ([0, 1]d ) = 1 > 0.
59. Let S be the collection of boxes in Rd .
(a) Show that S is closed under intersections.

(b) Write out a careful proof of the fact that for any box S ∈ S in Rd , the complement S c is a
disjoint union of at most 2d boxes.
(a) First suppose d = 1. Let S be a box
S = {x ∈ R : a < x ≤ b}, with some −∞ ≤ a ≤ b ≤ ∞;
and let T be a box
T = {x ∈ R : c < x ≤ d}, with some −∞ ≤ c ≤ d ≤ ∞.
Then we have
S ∩ T = {x ∈ R : max{a, c} < x ≤ min{b, d}}.
This is again a box, as required.
uppose now d ≥ 2, and S = I1 × · · · × Id and T = J1 × · · · × Jd . Then x ∈ S ∩ T if and only if
for each i = 1, . . . , d, the i-th coordinate of x is in Ii ∩ Ji . By the previous paragraph, Ii ∩ Ji is a
one-dimensional box, and hence S ∩ T is a d-dimensional box.
(b) If d = 1, we have
(a, b]c = (−∞, a] ∪ (b, ∞),

(−∞, b]c = (b, ∞)
c
(a, ∞) = (−∞, a]
c
(−∞, ∞) = ∅.
Hence the result is true in d = 1 dimensions.

For general d we prove the result by induction on d. Assume it is true for some d, and let S be a
box in Rd+1 . Then we can write
d+1
Y
S= Ii = S ∗ × Id+1 ,
i=1
Qd
where I1 , . . . , Id+1 are 1-dimensional boxes and where we set S ∗ = i=1 Ii , a box in Rd . Then
S c = ((S ∗ )c × R) ∪ (S ∗ × Id+1
c
)
and this is a disjoint union. By the inductive hypothesis we have S ∗ = ∪2d i=1 Ti for some disjoint
d
collection of (possibly empty) boxes T1 , . . . , T2d in R . Also by the 1-dimensional case considered
c
above, Id+1 = J1 ∪ J2 where J1 , J2 are disjoint (possibly empty) boxes in R. Hence
∗ ∗
S c = ∪2d
i=1 (Ti × R) ∪ (S × J1 ) ∪ (S × J2 )
which expresses S c as a union of 2(d+1) (possibly empty) disjoint boxes, completing the induction.
60. Complete the proof of Lemma 3.1.2 from lectures. That is: show that S ∈ B for all boxes S (even
for unbounded ones), and show that σ(S) = B (by first showing that every open set is a countable
union of boxes).
We first show that S ∈ B for any box S. This was done in the notes (Section 3.1) for the case where
S is bounded. Given a general (possibly unbounded) box S, set Rn = (−n, n]d and Sn = Rn ∩ S.
Then Sn is a box by Lemma 3.1.1 (since it is the intersection of two boxes), and moreover
S = S ∩ Rd = S ∩ (∪∞ ∞ ∞
n=1 Sn ) = ∪n=1 (S ∩ Rn ) = ∪n=1 Sn .
Each Sn is a bounded box, so each Sn ∈ B by the case already considered. Hence S is a countable
union of sets in B, and since B is a σ-algebra therefore S ∈ B.
Thus S ⊂ B and hence σ(S) ⊂ B.
For an inclusion the other way, we show that any open set U ⊂ Rd is in σ(S) by showing that it
is a countable union of boxes. Let S 0 be the class of rational (bounded) boxes, i.e. boxes of the
form ni=1 (ai , bi ] with ai < bi and ai , bi ∈ Q for i = 1, 2, . . . , d. Since Q (and therefore Q2d ) is
Q
countable, there are countably many such boxes. We claim that
U = ∪S∈S 0 :S⊂U S. (7)
To see this, it suffices to show that U ⊂ ∪S∈S 0 :S⊂U S, since the inclusion the other way is obvious.
Let x ∈ U . Then we can pick r (depending on x) such that the ball B(x, r) is contained in U , and
then we can pick a rational box S ∈ S 0 (also depending on x) with x ∈ S ⊂ B(x, r), and hence
S ⊂ U so x ∈ ∪S∈S 0 :S⊂U S. Since this holds for all x ∈ U we have U ⊂ ∪S∈S 0 :S⊂U S as required.
Since (7) expresses U as a countable union of sets that are in S (and hence in σ(S)), and σ(S) is
a σ-algebra, it follows that U ∈ σ(S). Thus G ⊂ σ(S) (where G is the class of open sets in Rd )
and hence B = σ(G) ⊂ σ(S), as required.
61. The following problem arises in the construction of a probability measure representing an infinite
sequence of independent coin tosses, with probability of Heads 0 < p < 1. Let Ω = {0, 1}N , where
N = {1, 2, 3, . . . } denotes the natural numbers. (We think of a sequence ω = (ω1 , ω2 , . . . ) ∈ Ω as
the possible outcome of the coin tossing, where 1 represents Heads and 0 represents Tails.) Given
k ≥ 1 and εi ∈ {0, 1}, i = 1, . . . , k, we define the set
Eε1 ,...,εk := {ω ∈ Ω : ωi = εi , i = 1, . . . , k}. (8)
(a) For k ≥ 1 fixed, let Ak denote the collection of sets that are finite unions of sets of the form
(8). Check that Ak is an algebra in Ω.
(b) Let A := ∪∞ k=1 Ak . Is A also an algebra in Ω? Is it a σ-algebra?
(c) Define Pk (∅) = 0 and
k
Y
Pk [Eε1 ,...,εk ] := pεi (1 − p)1−εi .
i=1
Explain how Pk can be defined on all of Ak so that it becomes finitely additive on Ak .
(a) Let Xk denote the collection of all sets of the form Eε1 ,...,εk (indexed by 0–1 sequences of length
k). There are 2k sets in Xk , and these are pairwise disjoint. In fact, Xk is a partition of Ω. because
the union of all 2k of the sets is X. Therefore by Question 5, Ak is the σ-algebra σ(Xk ).
(b) We claim that for each k that Ak ⊂ Ak+1 . Indeed, each set in Xk is of the form Eε1 ,...,εk =
Eε1 ,...,εk ,0 ∪ Eε1 ,...,εk ,1 , and hence is in Ak+1 . Hence Xk ⊂ Ak+1 , so Ak = σ(Xk ) ⊂ Ak+1 as claimed.
We saw in course of the solution of Exercise 3(d) on Sheet 1, that if A1 ⊂ A2 ⊂ . . . are algebras,
then ∪∞
k=1 Ak is also an algebra. So A is also an algebra in Ω.
A is not a σ-algebra. For example, setting ω0 = (0, 0, 0, . . .) ∈ Ω, we have for all k that {ω0 } ∈/ Ak
since Ak is an infinite set and {ω0 } is a finite set; however denoting the k-vector (0, 0, . . . , 0) by
0k , we have {ω0 } = ∩∞k=1 E0k ∈ σ(A), so σ(A) 6= A so A is not a σ-algebra.
(c) Any A ∈ Ak can be written as a finite disjoint union of some of the sets Eε1 ,...,εk , and such a
representation is unique. Therefore, we can define P[A] for A ∈ Ak by adding up the “masses”
of each constituent set Eε1 ,...,εk contained in A. This is additive, because disjoint sets A, B ∈ Ak
have disjoint constituents. This is a special case of Question 12.
Thus Pk is a measure on (Ω, Ak ). It is actually a probability measure because by the binomial
theorem,
k k
k
X k j k−j
X Y
1 = (p + (1 − p)) = p (1 − p) = pεi (1 − p)1−εi
j=0
j k i=1
ε=(ε1 ,...,εk )∈{0,1}
= P[Ω]
62. Suppose X is a non-empty set. A semi-algebra in X is a non-empty collection R of subsets of X

such that (i) R is a π-system, and (ii) if R ∈ R then Rc can be expressed as a finite disjoint union
of sets in R.
(a) Suppose R is a semi-algebra in X. Let V be the collection of all subsets of X that can be
expressed as a finite disjoint union of sets in R. Show that V is an algebra.
(b) Suppose µ : R → [0, ∞] is a finitely additive set function. Show that there is a unique finitely
additive set function µ̄ : V → [0, ∞] such that µ̄(A) = µ(A) for all A ∈ R. (see Lemma 3.2
for the concept of finite additivity)
(a) We follow the same steps as the proof of Theorem 3.3.

Step 1. We first show that V is closed under pairwise intersections. Suppose A = ∪ki=1 Ri ,
B = ∪`j=1 Sj (disjoint unions) with all Ri , Sj ∈ A. Then
A ∩ B = ∪ki=1 ∪`j=1 ( Ri ∩ Sj )∈V by definition of V.

| {z }
∈ A by Assumption (i)
Step 2. We note that if R ∈ R, then by assumption (ii), Rc ∈ V.
Step 3. We now show that V is closed under complements. Suppose A = ∪ki=1 Ri (disjoint union)
with all Ri ∈ R. Then
Ac = ∩ki=1 Ric ∈V by Step 1 and induction.

|{z}
∈ V by Step 2
This verifies that V is closed under complements.

Step 4. We check that X ∈ V. Pick R ∈ R (we can do this since we assumed R is nonempty).
Then X = R ∪ Rc ∈ V by assumption (ii).
Step 5. We check that V is closed under pairwise unions: if A, B ∈ V, we have
c
A∪B =( ∩ B}c
|A {z )c ∈ V by Step 2.
∈ V by Steps 1 and 3
Thus by Steps 3,4 and 5, V is an algebra.

(b) We follow the proof of Theorem 3.4.
writing any A ∈ V in the form A = ∪ki=1 Ri (disjoint union of members of R),
We define µ̄ on V, by P
and putting µ̄(A) = ki=1 µ(Ri ). This is the only possible way to define µ̄(A) with the required
properties, so uniqueness of the extension (if it exsits at all) is immediate.
We check that µ̄ is well-defined, that is, the value of µ̄(A) does not depend on the way we
represented A as a union. Suppose A = ∪ki=1 Ri = ∪`j=1 Sj with all Ri and Sj in R. Then
Ri = ∪`j=1 (Ri ∩ Sj ) and Sj = ∪ki=1 (Ri ∩ Sj ).
Using the assumption that µ is finitely additive on R, this implies:

k
X k X
X ` `
X
µ(Ri ) = µ(Ri ∩ Sj ) = µ(Sj ).
i=1 i=1 j=1 j=1
So any two possible definitions of µ̄(A) agree.

Now we check that µ̄ is finitely additive on V. Suppose A, B1 , . . . , Bn ∈ V with B1 , . . . , Bn
pairwise disjoint and A = ∪ni=1 Bi . Write Bi = ∪m
j=1 Ri,j as a disjoint union of members of R. Then
i
A = ∪ki=1 ∪mj=1 Ri,j , and hence

i
mi
k X k
defn. of µ̄(A) X defn. of µ̄(Bi ) X
µ̄(A) = µ(Ri,j ) = µ̄(Bi ).
i=1 j=1 i=1
63. Given a function F with the properties assumed in Exercise 13, show that the pre-measure λ̄F
on the algebra U, established in Exercise ??, is σ-finite. Deduce from Carathéodory’s extension
¯ to the Borel sets B in R.
theorem that λ̄F has a unique extension λ̄ F
64. (Basic properties of λ̄ ¯ ) Consider the measure λ̄ ¯ constructed in Exercise 63. Use the relation
F F
¯
λ̄F ((a, b]) = F (b) − F (a) and general properties of measures to deduce the following relations:
¯ ({y}) = F (y)−F (y−),
(a) For any y ∈ R show that the measure of the one-point set {y} equals λ̄ F
where F (y−) = limz↑y F (z).
¯ ([a, b]) = F (b) − F (a−) and also find the formulas for λ̄
(b) Show that λ̄ ¯ ((a, b)) and λ̄
¯ ([a, b)),
F F F
when −∞ < a < b < ∞.
(a) We have that {y} = ∩∞ n=1 (y − 1/n, y] (intersection of a decreasing sequence of sets), and
¯ ((y − 1, y]) = F (y) − F (y − 1) < ∞, so by downwards continuity of measure (Theorem 2.2 (iii))
λ̄F
we have
¯ ({y}) = λ̄
λ̄ ¯ (∩∞ (y − 1/n, y] = lim λ̄
¯ (y − 1/n, y] = lim (F (y) − F (y − 1/n)) = F (y) − F (y−).
F F n=1 F
n→∞ n→∞
¯ is a measure we have
(b) Since [a, b] = {a} ∪ (a, b] (a disjoint union) and λ̄ F
¯ ([a, b]) = λ̄
λ̄ ¯ ({a}) + λ̄
¯ ((a, b]) = (F (a) − F (a−)) + (F (b) − F (a)) = F (b) − F (a−).
F F F
¯ ({b}) < ∞,
Also, since λ̄ F
¯ ((a, b)) = λ̄
λ̄ ¯ ((a, b]) − λ̄
¯ ({b}) = F (b−) − F (a)
F F F
and
¯ ([a, b)) = λ̄
λ̄ ¯ (a, b) + λ̄
¯ ({a}) = F (b−) − F (a−).
F F F
65. Set I0 = (0, 1], let λ̄ be the 1-dimensional Lebesgue pre-measure, and let λ∗ be the associated
outer measure. Let A ⊂ I0 . The Lebesgue inner measure of A is given by λ∗ (A) := 1 − λ∗ (I0 \ A).
The aim of this question is to show that if the Lebesgue outer and inner measures of A coincide,
then A is Lebesgue measurable. Therefore, assume throughout that λ∗ (A) = λ∗ (A).
(a) Show that for any 1-dimensional box I ∈ S with I ⊂ I0 , we have λ∗ (I) = λ∗ (A∩I)+λ∗ (Ac ∩I),
where here we are taking complements in I0 .
(b) Show that for any T ⊂ I0 , we have λ∗ (A ∩ T ) + λ∗ (Ac ∩ T ) ≤ λ∗ (T ), where again we are
taking complements in I0 .
(c) Using Lemma 3.10.1 or otherwise, deduce that A is λ∗ -measurable.
(a) By Lemma 3.10.2 (subadditivity of outer measure), taking E1 = A∩I, E2 = Ac ∩I and En = ∅
for n ≥ 3, we have λ∗ (I) ≤ λ∗ (A ∩ I) + λ∗ (Ac ∩ I).
We need to prove an inequality the other way. Since I ∈ S, by Lemma 3.10.3 I is λ∗ -measurable.
Hence, writing I c := I0 \ I, and Ac := I0 \ A, we have λ∗ (A) = λ∗ (A ∩ I) + λ∗ (A ∩ I c ) and
λ∗ (Ac ) = λ∗ (Ac ∩ I) + λ∗ (Ac ∩ I c ). Therefore
λ∗ (A ∩ I) + λ∗ (Ac ∩ I) = λ∗ (A) − λ∗ (A ∩ I c ) + λ∗ (Ac ) − λ∗ (Ac ∩ I c )
= 1 − λ∗ (A ∩ I c ) − λ∗ (Ac ∩ I c ) (9)
where for the last line we used the assumption that λ∗ (A) = λ∗ (A). But now using the subaddi-
tivity of outer measure again, we have λ∗ (A ∩ I c ) + λ∗ (Ac ∩ I c ) ≥ λ∗ (I c ), so that by (9),
λ∗ (A ∩ I) + λ∗ (Ac ∩ I) ≤ 1 − λ∗ (I c ) = λ∗ (I)
as required (for the last equality we have again used the λ∗ -measurability of I.
(b) Using
P∞ the definition of outer measure, given ε > 0 we can cover T by intervals J1 , J2 , . . . in S
∗
with i=1 λ(Ji ) ≤ λ (T ) + ε. Therefore using part (a) yields
∞
X
λ∗ (T ) + ε ≥ λ(Ji )
i=1
∞ ∞
! ∞
X X X
= (λ∗ (Ji ∩ A) + λ∗ (Ji ∩ Ac )) = λ∗ (Ji ∩ A) + λ∗ (Ji ∩ Ac )
i=1 i=1 i=1
≥ λ∗ (T ∩ A) + λ∗ (T ∩ Ac )
where for the last line we used subadditivity of outer measure (Lemma 3.10.2), and the fact that
T ∩ A ⊂ ∪∞ c ∞ c
i=1 (Ji ∩ A) and T ∩ A ⊂ ∪i=1 (Ji ∩ A ). Since ε > 0 is arbitrary, we may deduce that
∗ ∗ ∗ c
λ (T ) ≥ λ (T ∩ A) + λ (T ∩ A ).
(c) We show that A is λ∗ -measurable, considered as a subset of the whole of R. Let V ⊂ R. Since
A ⊂ I0 , we have V \ A = (V ∩ I0 \ A) ∪ (V \ I0 ), so using subadditivity of outer measure once
more, then using part (b) (taking T = V ∩ I0 ), gives
λ∗ (V ∩ A) + λ∗ (V \ A) ≤ λ∗ (V ∩ A) + λ∗ (V ∩ I0 \ A) + λ∗ (V \ I0 )
≤ λ∗ (V ∩ I0 ) + λ∗ (V \ I0 ) = λ∗ (V ),
where the last equality follows from the fact that I0 ∈ S so I0 is λ∗ -measurable by Lemma 3.10.3.
Together with the subadditivity of outer measure, this shows that λ∗ (V ) = λ∗ (V ∩ A) + λ∗ (V \ A),
so A is λ∗ -measurable.
66. In the proof of Theorem 3.11.1, let S 0 be the class of rational boxes in Rd , i.e. the class of boxes
with all corners having rational coordinates (not quite what it says in the printed notes: see
the online correction). Provide a careful proof that if µ is a translation invariant measure with
µ((0, 1]d ) = 1, then µ(R) = λ(R) for all R ∈ S 0 (the argument was sketched in a lecture).
First consider boxes of the form Bn = (0, 1/n] with n ∈ N. For each n, clearly we can write
d
B1 = ∪ni=1 (Bn + xi ) (disjoint union) where x1 , . . . , xnd ∈ R; in other words, B1 is a disjoint union
of nd translates of B1 . Therefore by additivity and translation invariance,
n d nd
X X
1 = µ(B1 ) = µ(Bn + xi ) = µ(Bn ) = nd µ(Bn )
i=1 i=1
so µ(Bn ) = n−d = λ(Bn ).

Now let B = di=1 (ai , bi ] be an arbitrary rational box in Rd (we are considering only bounded
Q
boxes here). Then bi − ai is rational for each i so we can choose a ‘common denominator’ m ∈ N
with m(bi − ai ) ∈ N for each i. Then we can choose y1 , . . . , yk ∈ Rd such that B = ∪ki=1 (Bm + yi )
(disjoint union); that is, B is a disjoint union of translates of Bm . Therefore
k
X
µ(B) = µ(Bm + yi ) = kµ(Bm )
i=1
and similarly λ(B) = kλ(Bm ); therefore since µ(Bm ) = λ(Bm ) we have µ(B) = λ(B).
67. The aim of this question is to prove Theorem 3.11.2. Suppose ρ is a rotation on Rd , i.e. pre-
multiplication by a d × d matrix M with M T = M −1 (viewing elements of Rd as column vectors).
Show that |ρ(x)| = |x| for all x ∈ Rd , where for x = (x1 , . . . , xd ) ∈ Rd we put |x| =
(a) p
x21 + · · · + x2d .
Consider x as a column vector i.e. x = (x1 , . . . , xd )0 ; then |x|2 = x0 x so |ρ(x)|2 = (M x)0 (M x) =
x0 M T M x = x0 x = |x|2 .
(b) Show that ρ(A) ∈ B for all A ∈ B.
Suppose U ∈ G (the class of open sets). Then ρU is also open since if y ∈ ρU , then
ρ−1 y ∈ U so for some r > 0 we have B(ρ−1 y, r) ⊂ U (since it is assumed open), so B(y, r) =
ρ(B(ρ−1 y, r)) ⊂ ρU . Hence ρ(U ) is open, so ρ(U ) is Borel.
Let F be the class of sets A such that ρ(A) is Borel. This is a σ-algebra because:
• ρ(Rd ) = Rd is a Borel set, so Rd ∈ F.
• If A ∈ F then ρ(Rd \ A) = ρ(Rd ) \ ρ(A) which is a Borel set, so Rd \ A ∈ F.
• If A1 , A2 , . . . ∈ F then ρ(∪∞ ∞
n=1 An ) = ∪n=1 ρ(An ) ∈ F.
Therefore F contains the σ-algebra generated by the open sets, i.e. the Borel σ-algebra.
(c) Define a measure µ on B by µ(A) = λ(ρA) for all A ∈ B. Show that µ is translation invariant.
For A ∈ B and x ∈ Rd we have ρ(A + x) = ρ(A) + ρx (distributive law for matrix multipli-
cation) so
µ(A + x) = λ(ρ(A + x)) = λ(ρA + ρx) = λ(ρA) = µ(A)
where the penultimate inequality is because λ is translation invariant. Thus µ is translation
invariant.
(d) Show that λ is rotation invariant, i.e. λ(ρA) = λ(A) for all Borel A ⊂ Rd (and for any
rotation ρ).
By part (c) along with Theorem 3.11.1(iii) (uniqueness of λ as a translation invariant Borel
measure), our µ is a constant multiple of λ, say µ = cλ. But for the unit ball centred at the
origin (denoted B) we have ρB = B, by part (a). Therefore µ(B) = λ(ρB) = λ(B), and
hence c = 1.
In the next few questions let λ, respectively λ∗ , denote Lebesgue measure, respectively Lebesgue
outer measure, in R (or in Rd ).
68. * Use the Uniqueness Lemma to show that d-dimensional Lebesgue measure λ has the scaling
property: for any real number c 6= 0 and any Borel set B ∈ B, we have λ(cB) = |c|d λ(B).
Define ν(B) := λ(cB), B ∈ B. It is easy to check that ν is a measure; indeed, ν(∅) = λ(c∅) =
λ(∅) = 0, and if A1 , A2 , . . . ∈ B are pairwise disjoint then so are cA1 , cA2 , . . . and hence
∞
X ∞
X
ν(∪∞
n=1 An ) = λ(c ∪∞
n=1 An ) = λ(∪∞
n=1 (cAn )) = λ(cAn ) = ν(An ).
n=1 n=1
Likewise, |c|d λ(B) is a measure. We check that these two measures are equal on boxes: if S is a
box, and ai , bi are the endpoints of its sides, then the endpoints of the sides of cB are:
cai , cbi if c > 0;

cbi , cai if c < 0.
The length of the i-th side, in either case is |c|(bi − ai ). Therefore we get
d
Y d
Y
d
ν(S) = λ(cB) = (|c|(bi − ai )) = |c| (bi − ai ) = |c|d λ(S).
i=1 i=1
The collection of boxes is closed under intersection. We also have Rd = ∪∞ d

n=1 (−n, n] (a countable
union of boxes). Therefore, the Uniqueness Lemma (Theorem 3.9) implies that ν ≡ |c|d λ all Borel
sets.
69. Suppose A ⊂ R with λ∗ (A) = 0. Show that A is Lebesgue measurable.

Here λ∗ denotes Lebesgue outer measure. It follows from the definition of outer measure that
whenever E ⊂ F ⊂ R we have λ∗ (E) ≤ λ∗ (F ). Hence for any T ⊂ R we have T ∩ A ⊂ A, and
therefore using the assumption λ∗ (A) = 0 we have λ∗ (T ∩ A) ≤ λ∗ (A) = 0.
Also, since T \ A ⊂ T we have λ∗ (T \ A) ≤ λ∗ (T ).
Combining the last two inequalities gives us that λ∗ (T ∩ A) + λ∗ (T \ A) ≤ λ∗ (T ), for all T ⊂ R,
and as discussed in lectures this is enough to show that A is Lebesgue measurable.
70. Let 0 ≤ a1 ≤ a2 ≤ . . . and 0 ≤ b1 ≤ b2 ≤ . . . be sequences in [0, ∞] with limits an ↑ a ∈ [0, ∞] and

bn ↑ b ∈ [0, ∞].
(a) Show that (an + bn ) ↑ (a + b).
(b) Show that (an bn ) ↑ (ab).
(a) Clearly (an + bn ) is increasing in n; we need to show an + bn → a + b. Since an ≤ a and bn ≤ b,
clearly an + bn ≤ a + b.
If a + b < ∞, then given ε > 0 we can find n0 such that for n ≥ n0 we have an > a − ε/2, and
we can find n1 such that for n ≥ n1 we have bn > b − ε/2. Then for n ≥ max(n0 , n1 ), we have
a + b ≥ an + bn ≥ a + b − ε. Therefore an + bn → a + b.
If a + b = ∞ then either a = ∞ or b = ∞. In the first case with a = ∞ we have for all n that
an + bn ≥ an , and hence an + bn → ∞. The other case is treated similarly.
(b) Suppose 0 < a < ∞ and 0 < b < ∞. Then for all n we have
0 ≤ (ab) − an bn = (a − an )b + an (b − bn ) ≤ (a − an )b + a(b − bn )
and given ε > 0 we can choose n2 such that for n ≥ n2 we have a−an < ε/(2b) and b−bn < ε/(2a),
so that 0 ≤ (ab − an bn ) < ε. This gives the convergence in this case.
If a = 0 then an = 0 for all n so an bn = 0 for all n so an bn → 0 = ab (note that if b = ∞ we use
the rule 0 × ∞ = 0 to say that we still have ab = 0).
Similarly, if b = 0 then an bn → 0 = ab.
If a = ∞ and b > 0 (possibly b = ∞ too), pick b̃ with 0 < b̃ < b. Then given K > 0 there
exists n3 such that for n ≥ n3 we have both an ≥ K/b̃, and bn > b̃, so that an bn > K. Thus
an bn → ∞ = ab. Similarly if b = ∞ and a > 0, then an bn → ∞ = ab.
71. * The next problem takes an initial step towards R → R is

R 1 proving that if a function f : [0, 1]
Riemann intergable, then its Riemann integral 0 f (x) dx equals its Lebesgue integral [0,1] f dλ.
Let f : [0, 1] → [0, ∞) be a bounded function that is Riemann-integrable. For each n ≥ 1, consider
the subdivision of [0, 1] into 2n equal intervals: xi := i/2n , i = 0, . . . , 2n . Recall that if
mi := inf f (x); and Mi := sup f (x);

x∈(xi−1 ,xi ] x∈(xi−1 ,xi ]
Pn 1
then the lower Riemann sum Pn of the subdivision is `n := i=1 mi 2n and the upper Riemann sum
1
of the subdivision is un := i=1 Mi 2n .
(a) Define simple functions sn , tn : [0, 1] → [0, ∞) such that
0 ≤ s1 ≤ s2 ≤ · · · ≤ f ≤ · · · ≤ t2 ≤ t1 (pointwise),
R R
and `n = [0,1] sn dλ and un = [0,1] tn dλ.
R R
(b) Show that limn→∞ [0,1] sn dλ = limn→∞ [0,1] tn dλ.
(c) Show that f := supn≥1 sn and f := inf n≥1 tn are measurable, and f ≤ f ≤ f .
(a) Define
2n 2n
X X
sn = mi 1(xi−1 ,xi ] and tn = Mi 1(xi−1 ,xi ] .
i=1 i=1
Since the subdivisions are nested, we have sn ≤ sn+1 and tn+1 ≤ tn . Also, since mi is the infimum
of f (y) over y ∈ (xi−1 , xi ], we have mi ≤ f (y) for y ∈ (xi−1 , xi ]. This shows that sn ≤ f . By a
similar argument f ≤ tn . This proves that
s1 ≤ s2 ≤ · · · ≤ f ≤ · · · ≤ t2 ≤ t1 (pointwise).
Also the possible values taken by sn (x), x ∈ [0, 1], are the numbers 0, m1 , m2 , . . . , m2n (a finite
set of values) so sn has finite range. Also the function sn is Borel-measurable since each of the
intervals (xi−1 , xi ] is in B. Thus sn is simple, and a similar argument shows tn is simple.
We have n n
2
X 2
X Z
`n = mi (xi − xi−1 ) = mi λ((xi−1 , xi ]) = sn dλ.
i=1 i=1 [0,1]
R
Similarly, un = t
[0,1] n
dλ.
R1
(b) Since f is Riemann integrable, we have limn→∞ `n = 0
f (x) dx = limn→∞ un , so the statement
follows from part (a).
(c) f = supn≥1 sn is measurable (as the supremum of a sequence of measurable functions: see
Theorem 4.3), and similarly, f = inf n≥1 tn is measurable. Since sn ≤ f ≤ tn for all n ≥ 1, we get
f ≤ f ≤ f.
1
72. * For x ∈ R let f (x) = 1+x2
. Recall that f ∈ L1 (λ). Show that if {hn }n≥1 is a sequence in R such
that hn → 0, then Z ∞
lim |f (x + hn ) − f (x)| dλ(x) = 0.
n→∞ −∞
Since f is continuous at every x ∈ R, we have that |f (x + hn ) − f (x)| → 0 as n → ∞ at every

x ∈ R. We need to dominate the integrand, uniformly in n, in order to use the Dominated
Convergence Theorem. Since hn → 0, we may assume that |hn | ≤ 1 for all n (otherwise discard
finitely many terms where this is not true, and this does not affect the limit). Then, using that
f (x) is decreasing on [0, ∞) and increasing on (−∞, 0], we have:

1
 if −1 ≤ x ≤ 1;
1
|f (x + hn ) − f (x)| ≤ max{f (x + hn ), f (x)} ≤ g(x) := f (x − 1) = 1+(x−1)2 if 1 < x < ∞;
 1
f (x + 1) = 1+(x+1) if −∞ < x < −1.

2
It is clear that g ∈ L1 (R), since G is bounded on [−1, 1] and the rate of decay at ±∞ is like x−2 .
Therefore, we can apply the Dominated Convergence Theorem to get
Z Z
lim |f (x + hn ) − f (x)| dλ(x) = 0 dλ(x) = 0.
n→∞ R R
73. Show that the conclusion of the preceding question remains true if f is of the form:
g(x)
f (x) = ,
1 + x2
where g : R → R is a bounded continuous function, that is, g is continuous and there exists a
constant M > 0 such that |g(x)| ≤ M for every x ∈ R.
Since g is continuous at every x ∈ R, we have that |f (x+hn )−f (x)| → 0 as n → ∞ at every x ∈ R.
We need to dominate the integrand, uniformly in n, in order to use the Dominated Convergence
Theorem. Since hn → 0, we may assume that |hn | ≤ 1 for all n (otherwise discard finitely many
terms where this is not true, and this does not affect the limit). We bound
g(x+hn ) g(x) g(x+hn )−g(x) g(x) g(x)

|f (x + hn ) − f (x)| = 1+(x+hn )2
− 1+x2
≤ 1+(x+hn )2
+ 1+(x+hn )2
− 1+x2
2M 1 1
≤ 1+(x+hn )2
+M 1+(x+hn )2
− 1+x2
.
As in the solution to Question 72, the second term is dominated by the function M h(x), where

1
 if −1 ≤ x ≤ 1;
1
h(x) := 1+(x−1)2 if 1 < x < ∞;

 1
1+(x+1)2
if −∞ < x < −1.
Similarly, the first term is dominated by 2M h(x). We saw in the solution to Question 72 that
h ∈ L1 (R). Therefore, we can apply the Dominated Convergence Theorem to get
Z Z
lim |f (x + hn ) − f (x)| dλ(x) = 0 dλ(x) = 0.
n→∞ R R
74. * Suppose that f : [0, 1] → R is a Lipschitz continuous function, i.e.

|f (y) − f (x)|
sup < ∞.
0≤x<y≤1 y−x
Suppose also that the right derivative f+0 (x) := limh↓0 (h−1 (f (x + h) − f (x))) exists for Lebesgue-
almost all x ∈ [0, 1).
Rh
(a) Show that f (0) = limh↓0 h−1 0 f (x)dx.
R1
(b) Show that f (1) − f (0) = 0 f+0 (x)dx.
(a) Since f is continuous, given ε > 0 we can choose δ > 0 such that for 0 < x < δ we have
|f (x) − f (0)| < δ. Then for 0 < h < δ we have
Z h Z h Z h
−1 −1 −1
f (0) − h f (x)dx = h (f (0) − f (x))dx ≤ h |f (0) − f (x)|dx ≤ δ.
0 0 0
This shows the required convergence.

(b) Given h ∈ (0, 1/2), for 0 < x ≤ 1 − h set gh (x) = h−1 (f (x + h) − f (x)), and for 1 − h ≤ x ≤ 1
set gh (x) = 0. Then |gh (x)| ≤ K for all x ∈ (0, 1), where K is the supremum appearing in the
R1
definition of Lipschitz continuous function. Since 0 Kdx < ∞, we can apply DOM, to obtain
Z 1 Z 1 Z 1
f+0 (x)dx
= lim gh (x)dx = lim gh (x)dx
0 0 h↓0 h↓0 0
Z 1−h Z 1−h
−1
= lim h f (x + h)dx − f (x)dx
h↓0 0 0
Z 1 Z 1−h
−1
= lim h f (x)dx − f (x)dx
h↓0 h 0
Z 1−h Z 1 Z h Z 1−h
−1
= lim h f (x)dx + f (x)dx − f (x)dx + f (x)dx
h↓0 h 1−h 0 h
Z 1 Z h
−1 −1
= lim h f (x)dx − h f (x)dx = f (1) − f (0),
h↓0 1−h 0
R1
where the last equality follows from part (a) and a similar argument to show h−1 1−h
f (x)dx →
f (1) as h ↓ 0.
Note that in the above we have used a slightly stronger version of DOM than the one given in the
notes, namely that the statement of Theorem 5.5 still holds with the phrase ‘fn (x) → f (x) for
every x ∈ X’ in the hypothesis of the theorem replaced by ‘fn (x) → f (x) for µ-almost all x ∈ X’.
We leave the proof of this, starting from Theorem 5.5, as a further exercise.
75. Suppose f : R → R is given by f (x) = x2 . What are (i) f −1 ((4, 9]) and (ii) f (R)?
f −1 ((4, 9]) = {x ∈ R : 4 < x2 ≤ 9} = [−3, −2) ∪ (2, 3].
f (R) = [0, ∞).
76. Suppose f : R → R is in L1 . The purpose of this question is to show in stages that

L1
R ∞ there exists a sequence of continuous functions gn : R → R such that gn −→ f as n → ∞, i.e.
(*)
−∞ n
|g − f |dx → 0 as n → ∞.
(a) Show that the statement (*) holds in the special case where f = 1I for some bounded interval
I.
(b) Recalling Question 24, now show that the statement (*) holds when f is simple and nonneg-
ative and f has bounded support (i.e., there exists n ∈ N with f (x) = 0 whenever |x| > n),
(c) Now show that the statement (*) holds when f (x) ≥ 0 for all x and f has bounded support
as well as being in L1 .
(d) Show that (*) holds whenever f (x) ≥ 0 for all x (and f ∈ L1 ).
(e) Finally, show that (*) holds whenever f ∈ L1 .
(a) Suppose f = 1I for I an interval with left endpoint a and right endpoint b. Take gn (x) = f (x)
for x ∈ I and for x ≤ a − 1/n, and for x ≥ b + 1/n, with the value of gn interpolated linearly
between x = a − 1/n and x = a, and the value of gn interpolated linearly between x = b and
x = b + 1/n.
R∞ L1
Then gn is continuous and |gn − f | ≤ 1[a−1/n,a]∪[b,b+1/n] so that −∞ |gn − f |dx ≤ 2/n, so gn −→ f
as n → ∞,
(b) First assume f = 1A for some bounded Borel set A. Then by Question 24, given ε > 0
we can find a set U which is a finite union of disjoint (bounded) intervals denoted I1 , . . . , Ik
say, such that λ(A4U ) < ε/2 (the fact that the intervals may be taken disjoint follows from
Question 28 (a)). By part (a) of this question we can find continuous functions g1 , . . . , gk such
that |1Ii − gi |dλ < ε/(2k). Setting g = ki=1 gi we have that g is continuous (since it is a finite
R P
sum of continuous functions) and
Z Z Z Xk
|f − g|dλ ≤ |f − 1U |dλ + |1U − gi |dλ
i=1
k
Z X
= λ(A4U ) + (1Ii − gi ) dλ
i=1
k Z
X
≤ λ(A4U ) + |1Ii − gi |dλ < ε.
i=1
P`
Now suppose f is simple, say f = i=1 ai 1Ai with Ai all bounded and measurable. Assume
the ai are not all zero (otherwise f = R which is continuous). By the preceding case we can find
0
continuous functions h1 , . . . , h` with |hi −1Ai | ≤ ε/(` max(|a1 |, . . . , |a` |)) for each i. Then setting
h = ì=1 ai hi gives us a continous function with
P
Z Z X `
X Z
|f − h|dλ = ai (1Ai − hi )dλ ≤ ai |1Ai − hi |dλ ≤ ε.
i=1
(c) Suppose f ≥ 0 with bounded support, and with f ∈ L1 . Now suppose only that f ≥ 0 with
bounded suppaort, and with f ∈ L1R. Take fRn simple with 0 ≤ fn ≤ f pointwise and fn ↑ f
pointwise.
R Then by MON, we have fn dµ ↑ f dµ, so given ε > 0 we can choose n such that
R |f − fn | ≤ ε/2. Fix this n. By what we previously proved, we can choose continous F with
|F − fn |dµ < ε/2, and then
Z Z Z
|f − F |dµ ≤ |f − fn |dµ + |fn − h|dµ < ε.
Then (*) follows.
1
R f ≥ 0 and
(d) Suppose R f ∈ L . Then taking fn = f 1[−n,n] we have
R that R0 ≤ fn ↑ f pointwise, so
by MON fn dµ → f dµ. Given ε > 0 we can choose fn with fn dλ > f dλ − ε/2. Fix this n.
R
Since fn has bounded support, by part (b) we can find continuous F with |fn − F |dλ < ε/2.
Then Z Z Z
|f − F |dµ ≤ |f − fn |dµ + |fn − F |dµ < ε.
This gives us (*) in this case.
(e) For general f ∈ L1 , we have f + ∈ L1 and f − ∈ L1 . By part (c) we can find gn , hn ∈ L1

+ −
R ∞ kgn − f k1 → 0 and khn − f k1 → 0 (where for any Borel h : R → R we write khk1 for
with
−∞
|h(x)|dx, so that fn := gn − hn is a continuous function with
kfn − f k1 ≤ kgn − f + k1 + khn − f − k1 → 0.

(a) Let A be the class of sets A satisfying (i) and (ii). It can be shown that A is a σ-algebra,
and that R ⊂ A, where R is the class of measurable rectangles in X × Y . It follows that
M ⊗ N = σ(R) ⊂ A.
Unfortunately the question asks us to show the reverse inclusion A ⊂ M ⊗ N . This is not
actually true in general – the question is mistaken.
Here is a counterexample to show this. Take both (X, M) and (Y, N ) to be the measurable space
(R, B). Let E ⊂ R with E ∈/ B. (By Corollary 7.2 such a set exists) Define the set S ⊂ R2 by
S := {(u, u) : u ∈ E}.
Then for each x ∈ R we have Sx = S[x] = {x} if x ∈ E, and Sx = S[x] = ∅ if x ∈

/ E, so in all cases
Sx and S[x] are in B. However, we assert (and prove below) that S ∈ / B ⊗ B, which proves that
the question was mistaken.
Let B2 be as defined in Definition 8.8. Then B ⊗ B = B2 ; we verify this in Question 57. Let
L = {(x, y) ∈ R2 : x = y}, the diagonal line in R2 through the origin tilted up at 45 degrees. For
A ∈ B2 set π(A) := {x ∈ R : (x, x) ∈ A} (the set π(A) is the projection of A ∩ L onto the x-axis).
We claim that for all A ∈ B2 we have π(A) ∈ B. This holds because if A is a product of intervals
then π(A) is an interval so is in B. Also, the collection of sets A ⊂ R2 with π(A) ∈ B is a σ-
algebra, because π(Ac ) = π(A)c for all A ⊂ R2 and π(∪i Ai ) = ∪i π(Ai ) for all sequences A1 , A2 , . . .
of subsets of R2 . Therefore the collection of sets A ⊂ R2 with π(A) ∈ B contains the σ-algebra
generated by all products of intervals, which is B2 . This justifies the claim.
For S given above we have π(S) = E ∈
/ B, so by the preceding claim S ∈
/ B2 , as asserted earlier.

Sol20 1

Uploaded by

Document Information

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

Sol20 1

Uploaded by

Copyright:

Available Formats

MA40042 Measure Theory and Integration (2020/21): Solutions 1

completing the induction.

(a) X = {1, 2, 3, 4},

(b) X = {1, 2, 3, . . . } and

(c) X is an uncountable set and

(d) X is any set, M1 ⊂ M2 ⊂ . . . are σ-algebras in X and M = ∪∞

and hence (∪∞ c ∞

and therefore again ∪∞

4. Given numbers xij ≥ 0 defined for each i ∈ N, j ∈ N, show that ∞

Therefore H is a σ-algebra, so σ(X ) ⊂ H. Thus σ(X ) = H.

(a) Describe the sets in the σ-algebra generated by X .

A1 = (0, 1], A2 = (1, 2], A3 = (2, 5], A4 = (5, 7].

σ(C) = σ(X ) = {∪j∈J Aj : J ⊂ {1, 2, 3, 4}}

8. Let O := {G ⊂ R : G is open}; F := {F ⊂ R : F is closed}; K := {K ⊂ R : K is compact}.

On the other hand, if Pµ(∪∞

Thus we have the countable additivity of δx so it is a measure.

Thus ν is countably additive so is a measure as claimed.

Hence ν is countably additive so is a measure.

F (x) = µ(−∞, x] = lim µ(−∞, x + 1/n] = lim F (x + 1/n).

is a measure. We are assuming that A1 , A2 , . . . are all non-empty. Then

since it is an empty sum. To prove countable additivity, suppose B1 , B2 , . . . are a sequence of

which gives the countable additivity.

completing the induction.

λF (A) = F (b) − F (a) ≤ F (b) − F (a0 ) + ε = λF (A0 ) + ε. (1)

Therefore, by compactness of [a0 , b] (Heine-Borel), there exists N ≥ 1 such that

It follows that A0 = (a0 , b] ⊂ ∪N 0

A ∩ B = ∪ki=1 ∪`j=1 (Ii ∩ Jj ) = ∪ki=1 ∪`j=1 Ki,j

Ki,j ∩ Ki0 ,j 0 = Ii ∩ Jj ∩ Ii0 ∩ Jj 0 = ∅,

because either i 6= i0 (so Ii ∩ Ii0 = ∅) or j 6= j 0 (so Jj ∩ Jj 0 = ∅).

Ac = ∪ki=0 (bi , ai+1 i,

and hence also A ∪ B = ((A ∪ B)c )c ∈ U. So U is an algebra.

xi := inf{x ∈ R : µ((−∞, x]) ≥ i},

with the infimum of the empty set taken to be +∞.

which is less than ε by our choice of δ.

λ1 (I ∩ Ac ) = λ1 (I) − λ1 (I ∩ A) > ελ1 (I) ≥ ελ1 (I ∩ A).

which completes the induction.

λ1 (A \ U ) ≤ λ1 (S \ U ) = λ1 (S) − λ1 (U ) < ε/2,

Combining the last two displays shows that λ1 (U 4A) = λ1 (U \ A) + λ1 (A \ U ) < ε.

cI = (ca, cb] if c > 0; cI = [cb, ca) if c < 0.

which is a finite disjoint union of sets in S, so by finite additivity

A = ∪q,r∈Q {(x, s, t) : f (x) > q > s, g(x) > r > t}

which is a countable union of sets in M ⊗ B ⊗ B and hence is itself in M ⊗ B ⊗ B.

Thus, F is indeed a σ-algebra in Y .

{x ∈ X : lim inf n→∞ fn (x) = +∞} ∪ {x ∈ X : lim supn→∞ fn (x) = −∞}

Hence it is enough to show that each of the sets A+ , A− , B is in M.

{ω ∈ Ω : lim fn (ω) = µ} = {ω ∈ Ω : lim sup fn (ω) = µ} ∩ {ω ∈ Ω : lim inf fn (ω) = µ}.

35. Let (X, M) be a measurable space.

(a) Let f = 1E . Then 

(b) Prove that if f dµ = 0, then µ(f −1 ((0, ∞])) = 0.

(a) Since f is measurable the set A := f −1 ([a, ∞]) = ∩∞ n=1 f

and hence µ(f −1 ([a, ∞])) = µ(A) ≤ a−1 f dµ as claimed.

(b) Now suppose f dµ = 0. Then since f −1 ((0, ∞]) = ∪∞ −1

For general integrable g we have that +

where cA := {cx : x ∈ Ai }. Using Lemma 10.7(b) and Exercise 25 we obtain that

g1 (x) := f (0) − ε ≤ f (x) ≤ f (0) + ε := g2 (x)

so that for h ∈ [0, δ] we have

This gives us (*) in this case.

(a) Fix y ∈ Y . Let F be the collection of A ⊂ X × Y such that A[y] ∈ M.

(Ac )[y] = {x ∈ X : (x, y) ∈ Ac } = {x ∈ X : (x, y) ∈ A}c = (A[y] )c ∈ M,

so Ac ∈ F. Also if An ∈ F for n = 1, 2, 3, . . ., then setting A = ∪∞

so A ∈ F. Thus we have verified the claim.

By the two preceding claims, F is a σ-algebra with R ⊂ F, and therefore M ⊗ N = σ(R) ⊂

f −1 ((y, ∞]) = {x ∈ X : f (x) > y} = {x ∈ x : (x, y) ∈ hyp(f )} = (hyp(f ))[y]