Analysis Part 2

www.mathtuition88.com

Book: Measure and Integral by Wheeden and Zygmund

3 Chapter 3

3.1 Q1
Choose c1 to be the largest integer such that c1 b−1 ≤ x. Inductively, choose
cn to be the largest integer such that nk=1 ck b−k ≤ x.
P

Taking limits as n → ∞, we get

∞
X
ck b−k ≤ x.
k=1

By the maximality of cn , we have

n
X n−1
X
ck b−k ≤ x < ck b−k + (cn + 1)b−n
k=1 k=1

for all n. So n
X
x− ck b−k < (cn + 1)b−n − cn b−n = b−n
k=1
for all n.
Taking limits as n → ∞, we get
∞
X
x≤ ck b−k .
k=1

1
 Uniqueness:
Suppose x = ∞
P −k
P∞ −k
k=1 c k b = k=1 dk b . Let n be the least index where
ck 6= dk . That is, cn 6= dn and ck = dk for all k < n.
We have
∞
X ∞
X ∞
X
−n −k −k
(cn − dn )b = dk b − ck b = (dk − ck )b−k .
k=n+1 k=n+1 k=n+1

Note that
∞
X ∞
X
−n −n −k
b ≤ |(cn − dn )b |= |dk − ck |b ≤ (b − 1)b−k = b−n
k=n+1 k=n+1

with strict inequality if there exists |dm − cm | =

6 b − 1 for some m ≥ n + 1.
6 b−1. Then b−n < b−n is a contradiction
Case 1) There exists |dm −cm | =
so in fact no such index n exists. Hence the expansion is unique.
Case 2) |dk − ck | = b − 1 for all k ≥ n + 1. The only possibility is
dk = b − 1 and ck = 0 (or vice versa) for all k ≥ n + 1. So
n
X
x= ck b−k + 0 = (c1 bn−1 + c2 bn−2 + · · · + cn )b−n ,
k=1

where c := c1 bn−1 + c2 bn−2 + · · · + cn ∈ {1, . . . , bn − 1} since in this case

0 < x < 1.
The other expansion is
n
X ∞
X
x= dk b−k + (b − 1)b−k .
k=1 k=n+1

3.2 Q2
3.2.1 (a)
P∞ −k
For this question, we write (0.c1 c2 . . . )3 to represent k=1 ck 3 . For exam-
ple, 0.02̇3 = (0.0222 . . . )3 = 0.13 = 1/3.

2
 ( =⇒ ) Suppose x = (0.c1 c2 . . . )3 ∈ C.
Case 1) x is not an endpoint of any of the intervals removed.
Note that c1 6= 1 since ( 31 , 23 ) = (0.13 , 0.23 ) is removed in the 1st stage.
Note that c2 6= 1 since ( 91 , 29 ) = (0.013 , 0.023 ) and ( 79 , 89 ) = (0.213 , 0.223 )
are removed in the 2nd stage. Inductively, we can note that ck 6= 1 since
(0.b1 b2 . . . bk−1 13 , 0.b1 b2 . . . bk−1 2)3 , where bi 6= 1, are removed in the kth
stage.
Case 2) x is an end point of some interval removed.
If x = 0.b1 b2 . . . bk−1 13 , where bi 6= 1, then we write x = 0.b1 b2 · · ·k−1 02̇3 .
If x = 0.b1 b2 . . . bk−1 23 , where bi 6= 1, we are done already.
We have proved that x ∈ C implies x has some triadic expansion for
which every ck is either 0 or 2.
( ⇐= ) Assume x = (0.c1 c2 . . . )3 has some triadic expansion for which
every ck is either 0 or 2. Let Ck denote the union of the intervals left at the
kth stage, so that C = ∞
T
k=1 Ck . Each ck is either 0 or 2 implies that x is not
removed in the kth stage, for all k. So x ∈ ∞
T
k=1 Ck = C.

3.2.2 (b)

For this question, we are using notation in pg 43 of the textbook.

Let x ∈ C and x = ∞ −k
P
k=1 ck 3 , where each ck is either 0 or 2. If x = 0,
f (x) = ( 21 ck )2−k is clearly true.
P

Define xn = nk=1 ck 3−k . Note that either xn = 0, or xn is the endpoint

P

of some open interval Ijn (jth interval removed, ordered from left to right, at
stage n), where j = nk=1 ( 12 ck )2n−k . See example1 .
P

Since fn is continuous,
n
−n
X 1
fn (xn ) = j2 = ( ck )2−k
k=1
2
1
For example, if xn = 0.223 , it is the endpoint of (0, 213 , 0.223 ) = I32 .

3
for all n.
Using the fact that {fn } converges uniformly to continuous f , we have
∞
X 1
f (x) = lim fn (xn ) = ( ck )2−k .
n→∞
k=1
2

3.3 Q4
Let Ck denote the union of the intervals left at the kth stage, so that C =
T∞
k=1 Ck .
Note that
|Ck | = (1 − θ)|Ck−1 |
holds for all k ≥ 1, where |C0 | = 1. Since C is covered by the intervals in
any Ck , we have
|C|e ≤ |Ck | = (1 − θ)k
for all k. Since 0 < 1 − θ < 1, we see that |C|e = 0.
To show C is perfect, consider x ∈ C. Each Ck is closed, so C is closed.
x lies in an interval Ik in Ck for every k. Let xk ∈ C \ {x} be an endpoint of
Ik . Then
|x − xk | ≤ |Ck | = (1 − θ)k → 0
as k → ∞.
Thus {xk } is a sequence in C \ {x} that converges to x, and thus x is a
limit point.

3.4 Q5
Let Dk denote the union of the intervals left at the kth stage, so that the
resultant set is D = ∞
T
k=1 Dk .
At each stage k, the length of the intervals removed is 2k−1 δ3−k . Thus,
k
X
|Dk | = 1 − 2j−1 δ3−j .
j=1

4
Since Dk & D and |D1 | < ∞, by Monotone Convergence Theorem for
measure,
|D| = lim |Dk | = 1 − δ.
k→∞

Observe that D1 cannot contain an interval of length greater than 1/2,

since the interval removed is in the middle. Inductively, Dk cannot contain
an interval of length greater than 1/2k . Thus D cannot contain an interval
of length greater than 1/2k for all k, so D contains no intervals.
Since each Dk is closed, D is closed. To show D is perfect, consider x ∈ D.
x lies in an interval Ik in Dk for every k. Let xk ∈ D \ {x} be an endpoint
of Ik . Then
|x − xk | ≤ |Ik | ≤ 1/2k → 0 as k → ∞.

Thus {xk } is a sequence in D \ {x} that converges to x. Hence D is

perfect.

3.5 Q9
Pn P∞
Let  > 0. Since limn→∞ k=1 |Ek |e = k=1 |Ek |e < ∞, there exists N such
that for n ≥ N ,
∞
X ∞
X n
X
|Ek |e = |Ek |e − |Ek |e < .
k=n+1 k=1 k=1
S∞ T∞
Write Uj = k=j Ek , so that lim sup Ek = j=1 Uj . Since lim sup Ek ⊆
UN +1 ,
∞
X
| lim sup Ek |e ≤ |UN +1 |e ≤ |Ek |e < .
k=N +1

Since  > 0 is arbitrary, | lim sup Ek |e = 0 so lim sup Ek has measure zero.
Since lim inf Ek ⊆ lim sup Ek , | lim inf Ek |e = 0 as well, so lim inf Ek has
measure zero as well.

5
3.6 Q11
Suppose that |E|e < +∞.
( =⇒ ) Assume E is measurable. Let  > 0. There exists an open set G
such that E ⊆ G and |G \ E|e < .
Since G is open, it can be written as a countable union of nonoverlapping
(closed) intervals, say G = ∞
S
k=1 Ik .
We have
∞
X ∞
[
|Ik | = | Ik | = |G| ≤ |E|e + |G \ E|e < ∞.
k=1 k=1

Thus, there exists N such that

∞
[ ∞
X
Ik = |Ik | < .
k=N +1 k=N +1
SN S∞
Let S = k=1 Ik , N1 = k=N +1 Ik , and N2 = G \ E.
Then E = (S ∪ N1 ) \ N2 with |N1 |e , |N2 |e <  as desired.
( ⇐= ) Assume E = (S ∪ N1 ) \ N2 . Since S is measurable, we can pick
an open set G with S ⊆ G and |G \ S| < . Pick another open set G1 with
N1 ⊆ G1 and
|G1 | < |N1 |e +  < 2.

Note that (G ∪ G1 ) \ E ⊆ (G \ S) ∪ G1 ∪ N2 .
Thus

|(G ∪ G1 ) \ E|e ≤ |G \ S|e + |G1 | + |N2 |e <  + 2 +  = 4.

Since G ∪ G1 is open, and E ⊆ G ∪ G1 , this means that E is measurable.

3.7 Q12
Lemma 3.1. If A ⊆ R and Z ⊆ R with |Z| = 0, then |A × Z| = 0. Similarly,
|Z × A| = 0.

6
Proof. Let  > 0. Since |Z| = 0, there exists intervals {Ik } such that Z ⊆
S∞ P∞
k=1 Ik and k=1 |Ik | < .
Write An = A ∩ [−n, n]. Then A = ∞
S
n=1 An . Note that

∞
[ ∞
[
An × Z ⊆ [−n, n] × Ik = ([−n, n] × Ik )
k=1 k=1

so ∞
X
|An × Z|e ≤ 2n|Ik | = 2n.
k=1

Since  > 0 is arbitrary, |An × Z| = 0 for each n. So

∞
[ ∞
X
|A × Z|e = | (An × Z)|e ≤ |An × Z|e = 0.
n=1 n=1

Thus |A × Z| = 0 as desired.

Now, since E1 is measurable, E1 = H1 ∪ Z1 , where H1 is of type Fσ and

|Z1 | = 0. Similarly, write E2 = H2 ∪ Z2 where H2 is of type Fσ and |Z2 | = 0.
Then

E1 × E2 = (H1 × H2 ) ∪ (H1 × Z2 ) ∪ (Z1 × H2 ) ∪ (Z1 × Z2 ).

Note that H1 × H2 is of type Fσ , while the other terms have measure zero
by the previous lemma. Thus E1 × E2 is measurable.
Case 1) Suppose |E1 | and |E2 | are both finite.
Since E1 , E2 are measurable, for each k ∈ N there are open sets Sk ⊇ E1 ,
Tk ⊇ E2 such that |Sk \ E1 | < 1/k, |Tk \ E2 | < 1/k. We may assume
0
Sk+1 ⊆ Sk , Tk+1 ⊆ Tk (if Sk+1 6⊆ Sk , then define Sk+1 = Sk+1 ∩ Sk instead).
S
Since Sk is open, Sk = k∈N Ii for some nonoverlapping closed intervals.
S
Similarly, Tk = j∈N Jj for some nonoverlapping closed intervals.

7
 So
[
|Sk × Tk | = (Ii × Jj )
(i,j)∈N×N
X
= |Ii × Jj |
i,j∈N
X
= |Ii ||Jj |
i,j∈N
X X
=( |Ii |)( |Jj |)
i∈N j∈N

= |Sk ||Tk |.
T∞ T∞
Write S = k=1 Sk , T = k=1 Tk . Then |S \ E1 | = |T \ E2 | = 0.
Hence

|E1 × E2 | = |S × T |
= lim |Sk × Tk |
k=∞

= lim |Sk ||Tk |

k→∞

= |E1 ||E2 |,

where the second equality follows by MCT for measure, since Sk ×Tk & S ×T
and |Sk ×Tk | < ∞ for some k since |E1 |, |E2 | are both finite. The last equality
also follows by MCT for measure.
Case 2) Suppose one of |E1 |, |E2 | are infinite.
If |E1 | = ∞ and |E2 | > 0, then write E1n = E1 ∩ [−n, n].

|E1 × E2 | = lim |E1n × E2 |

n→∞

= lim |E1n ||E2 |

n→∞

= |E1 ||E2 |
= ∞,

8
where the first equality follows by MCT for measure, since E1n ×E2 % E1 ×E2 .
If |E1 | = ∞ and |E2 | = 0, |E1 × E2 | = 0 by our first lemma.

3.8 Q17
Let f be the Cantor-Lebesgue function, which is continuous.

Lemma 3.2. f (C) = [0, 1], where C is the Cantor set.

Proof. f (C) ⊆ [0, 1] is clear.

Let y ∈ [0, 1]. Write y in its binary expansion, i.e.
∞
X
y= ck 2−k
k=1

where ck = 0 or 1.
Consider x = ∞ −k
P
k=1 (2ck )3 . Since 2ck = 0 or 2, by Exercise 2, x ∈ C.
P∞
Furthermore f (x) = k=1 ck 2−k = y. So y ∈ f (C). Hence [0, 1] ⊆ f (C).

Then, we have |f (C)| = 1. Since any set in R with positive outer measure
contains a non-measurable set, f (C) contains a non-measurable set A.
Note that f −1 (A) ⊆ C so |f −1 (A)| = 0. In particular f −1 (A) is measur-
able. So
f (f −1 (A)) = A

gives the desired counterexample.

3.9 Q20
Let E be a nonmeasurable subset of [0, 1] whose rational translates are dis-
joint. Consider the translates of E by all rational numbers r, 0 < r < 1,
denoted Er = {x + r : x ∈ E}.
S S
Note that | r Er |e ≤ 2 since r Er ⊆ [0, 2].

9
 Note that |Er |e = |E|e by Exercise 18, furthermore |E|e > 0 since E is
P
nonmeasurable. So r |Er |e = ∞.
Thus the inequality is strict.

3.10 Q23
S∞
Let Z ⊆ R with |Z| = 0. Write Zn = Z ∩ [−n, n], then Z = n=1 Zn . Clearly
|Zn | = 0 for each n.
S∞
Thus there exists intervals {Ik } (depending on n) such that Zn ⊆ k=1 Ik
and ∞
P
k=1 |Ik | < . We may assume each Ik ⊆ [−n, n].
For each x ∈ Ik = [ak , bk ], we have

 a2 ≤ x2 ≤ b2k if 0 ≤ ak ≤ bk
 k


0 ≤ x2 ≤ max{a2k , b2k } if ak ≤ 0 ≤ bk


b 2 ≤ x 2 ≤ a2

if ak ≤ bk ≤ 0.
k k

If ak ≤ 0 ≤ bk , we discard Ik = [ak , bk ] and replace it with two intervals

[ak , 0] ∪ [0, bk ] instead2 . Thus we may assume 0 ≤ ak ≤ bk or ak ≤ bk ≤ 0 for
all Ik = [ak , bk ].
Thus, x2 ∈ Jk where Jk is an interval with

|Jk | = |a2k − b2k | = |ak − bk ||ak + bk | ≤ |Ik |(|ak | + |bk |) ≤ |Ik |(2n).
2
S∞ P∞
Note that this will not affect Zn ⊆ k=1 Ik and k=1 |Ik | < .

10
 So
∞
[
2
|{x : x ∈ Zn }|e ≤ | Jk |e
k=1
∞
X
≤ |Jk |
k=1
∞
X
≤ 2n |Ik |
k=1

< 2n.

Since  > 0 is arbitrary, |{x2 : x ∈ Zn }|e = 0, for all n. Note that

{x2 : x ∈ Z} = ∞ 2
S
n=1 {x : x ∈ Zn }, so

∞
X
2
|{x : x ∈ Z}|e ≤ |{x2 : x ∈ Zn }|e = 0.
n=1

11