Homework 1

Real Analysis
Joshua Ruiter
March 23, 2018

Note on notation: When I use the symbol ⊂, it does not imply that the subset is proper.
In writing A ⊂ X, I mean only that a ∈ A =⇒ a ∈ X, leaving open the possibility that
A = X. I do not use the symbol ⊆.
Proposition 0.1 (Exercise 1). The middle-thirds Cantor set C is totally disconnected and
perfect.
Proof. First we show that C is totally disconnected. Let x, y ∈ C such that x 6= y. Then
|x − y| > 0, so there exists k ∈ N such that (1/3)k < |x − y| (because limk (1/3)k = 0). Since
(1/3)k < |x − y|, x and y must be in different subintervals of Ck , since the length of each
interval in Ck is (1/3)k . Thus there is a whole interval of points in [0, 1] \ C between x and
y, so C is totally disconnected.
Now we show that C is perfect. Let x ∈ C. We will show that x is not an isolated point
by finding a nearby point in C in a ball of any radius centered at x. Let r > 0, then choose k
large enough that (1/3)k ≤ r. Since C = ∩j Cj , we have x ∈ Ck . Since each subinterval of Ck
has length (1/3)k and x is in one such subinterval. Let y be an endpoint of that subinterval
so that x 6= y (there are two endpoints, so this is always possible). Then we have x 6= y
and |x − y| < (1/3)k < r, so y ∈ Br (x) ∩ C. Thus x is not an isolated point, and hence C is
perfect.
Proposition 0.2 (Exercise 2a). Let C be the middle-thirds Cantor set. Then for x ∈ [0, 1],
x ∈ C if and only if x has a ternary expansion
∞
X
x= ak 3−k
k=1

where ak ∈ {0, 2} for all k.

Proof. First, suppose that x ∈ [0, 1] has such a ternary expansion, where each ak ∈ {0, 2}.
Since a1 ∈ {0, 2}, we know that 0 ≤ x ≤ 1/3 or 2/3 ≤ x ≤ 1, so x ∈ C1 . More generally, we
can see by induction that ak ∈ {0, 2} =⇒ x ∈ Ck . Thus x ∈ Ck for all Pk, so x −k∈ C.
Now suppose that x ∈ C. We know x has some ternary expansion ∞ a
k=1 k 3 with ak ∈
{0, 1, 2} for all k. Consider the remaining intervals after just the first stage of construction,
[0, 1/3] ∪ [2/3, 1]. If x ∈ [0, 1/3), then a1 = 0. If x = 1/3, then x also has the ternary

1
expansion P ∞
P
k=2 2/3. If x ∈ [2/3, 1), then a1 = 2, and if x = 1, then x has the ternary
expansion ∞ k=1 2/3.
By induction on the recursive construction, at the kth stage, if x is in the left half of a
“split” interval, then ak = 0, except possibly at the right endpoint. But at that endpoint, x
has a finite ternary expansion terminating with a 1/3k term. But this term can be rewritten
as a ternary expansion involving only 0’s and 2’s by the following substitution:
∞
1 X 2
k
=
3 j=k+1
3j

And for x in the right half of a “split” interval, we get ak = 2, except possibly at the right
endpoint. But again, if we get a ternary expansion involving a 1 at such an endpoint, it
must be a finite expansion terminating in a 1/3k term, which we can expand as above. Thus
x has the required ternary expansion.
Proposition 0.3 (Exercise 2b). Let F be the Cantor-Lebesgue function on C. F is well
defined and continuous, and F (0) = 0, and F (1) = 1.
Proof. First we show that F is well defined. To do this, we need to show that for x =
P ∞ −k
k=1 ak 3 ∈ C, the value of F converges. By the previous proposition, we know that x ∈ C
has a ternary expansion where each ak is 0 or 2, so if we define bk =Pak /2, each bk is 0 or 1.
Then bk /2k ≤ 1/2k for each k. We know that the geometric series ∞ k
k=1 1/2 converges, so
by the comparison test,
∞
X bk
F (x) =
k=1
2k
converges as well. Thus F is well defined.
Now we show that F is continuous on C. Let x0 ∈ C. To show that F is continuous at
x0 , we will show that for  > 0, there exists δ > 0 such that
|x − x0 | < δ =⇒ |F (x) − F (x0 )| < 
Fix  > 0. Then there exists n ∈ N such that 2−n−1 <  < 2−n . Let δ = 3−n . Let
∞
X X∞
−k
x= ak 3 x0 = ck 3−k
k=1 k=1

For convenience, let bk = ak − ck . Then

     
X∞ ∞
X  X ∞  X ∞ 
|x − x0 | =  ak 3−k − ck 3−k  =  (ak − ck )3−k  =  bk 3−k 
     
     
k=1 k=1 k=1 k=1
−n
If |x − x0 | < δ = 3 , then b1 , b2 , . . . bn are all zero. Then the first n terms of |F (x) − F (x0 )|
are also zero:
   
X ∞ X∞  X ∞ 
−k −k  −k−1 
|F (x) − F (x0 )| =  (ak /2)2 − (ck /2)2  =  (ak − ck )2
 

   
k=1 k=1 k=1
   
X ∞   X ∞ 
−k−1  −k−1 
= bk 2 = bk 2  ≤ 2−n−1 < 
 
   
k=1 k=n+1

2
Thus F is continuous at x0 .
Now we show that F (0) = 0 and F (1) = 1. In the ternary expansion for zero, each
ak = 0. In the ternary expansion of 1, each ak = 2. Then
∞
X ∞
X ∞
X
0= (0)3−k ; F (0) = (0)2−k
= 0=0
k=1 k=1 k=1
∞ ∞ ∞  k
X X X 1
1= −k
(2)3 ; F (1) = (1)2−k
= =1
k=1 k=1 k=1
2

since both are geometric series.

Proposition 0.4 (Exercise 2c). The Cantor-Lebesgue function F : C → [0, 1] is surjective.
P∞ −k
Proof. Let y ∈ [0, 1]. Then y has a binary expansion
P∞ y = k=1 bk 2 , where each bk is zero
or one. Let ak = 2bk for each k. Then let x = k=1 ak 3−k . We can see from the definition of
F that F (x) = y. Furthermore, each ak is zero or two, so x is in the Cantor set by Exercise
2a. Hence F is surjective.
Proposition 0.5 (Exercise 3a). The complement of Cξ in [0, 1] is the union of open intervals
of total length 1.
Proof. At each stage of the construction, we remove an open interval of length ξ/l from the
middle of each remaining interval, where l is the length of each interval. This means that at
each stage, the total length of each interval after the removal is l(1 − ξ). Since each closed
interval from which we remove this interval is the same, the total sum of the remaining
intervals after each stage is multiplied by (1 − ξ). We begin with [0, 1], so after the first
stage, the remaining length is (1 − ξ), and after the second stage the remaining length is
(1 − ξ)2 . Generally, the remaining length after the kth stage is (1 − ξ)k . Thus, the sum of the
lengths of all removed intervals up to the kth stage is 1 − (1 − ξ)k . In the limit as k → ∞,
this length of remaining intervals goes to 1 for ξ ∈ (0, 1).
Proposition 0.6 (Exercise 3b). m∗ (Cξ ) = 0
Proof. After k iterations of the removal process, there are 2k intervals each with length
1 − (1 − ξ)k
2k
and the total length of these intervals is 1−(1−ξ)k . That is, for each k ∈ N, therenis a coveringo
P∞
of Cξ of disjoint cubes (intervals) Qj with total length 1 − (1 − ξ)k . Thus inf j=1 |Qj |
is less than or equal to 1 − (1 − ξ)k for all k ∈ N. Since limk 1 − (1 − ξ)k = 0, this means
that this infimum is at most zero. But the infimum cannot be less than zero, since it is an
infimum over sums in which all terms are positive. Thus m∗ (Cξ ) = 0.
Proposition
P∞ 0.7 (Exercise 4a). Let Cˆ be the Cantor-type set with constants (lk )∞
k=1 , such
k−1
that k=1 lk 2 < 1. Then
∞
X
ˆ =1−
m(C) lk 2k−1
k=1

3
Proof. The complement of Cˆ in [0, 1] is a countable union of disjoint intervals:
∞ 2[k−1
[
[0, 1] \ Cˆ = B(c1,1 , l1 /2) ∪ B(c2,1 , l2 /2) ∪ B(c2,2 , l2 /3) ∪ . . . = B(ck,j , lk /2)
k=1 j=1

Where ck,j is the center of the jth interval with length lk . Each of these intervals is measur-
able, with measure lk for some k. There are 2k−1 such intervals of length lk for each k. Thus
by countable additivity of m, we have
∞
X
[0, 1] \ Cˆ = lk 2k−1
k=1

which converges to less than 1 by hypothesis. By additivity,

ˆ + m(C)
m([0, 1] \ C) ˆ = m([0, 1]) = 1

Thus
∞
X
ˆ = 1 − m([0, 1] \ C)
m(C) ˆ =1− lk 2k−1
k=1

ˆ then there exists a sequence (xn )∞ such that

Proposition 0.8 (Exercise 4b). If x ∈ C, n=1
xn 6∈ Cˆ and xn → x and xn ∈ In where In is a subinterval in the complement of Cˆ with
|In | → 0.
Proof. Let x ∈ C.ˆ At each stage of the construction of C,ˆ there remain 2−1 disjoint closed
intervals, and x is in exactly one of these intervals. At the kth stage, that closed interval
has length

1 − kj=1 lk 2k−1
P

2k−1
since the numerator is the total length of remaining intervals, and the denominator is the
number of remaining intervals. When we remove the open middle interval of length lk from
that closed interval, let xk be the midpoint of that interval. Then

1 − kj=1 lk 2k−1
P
1
|xk − x| ≤ k−1
≤ k−1
2 2
Thus xk → x, and by construction xk ∈6 C.ˆ We let Ik be the interval of which xk is the
ˆ c ˆ we
midpoint, so Ik ⊂ (C) . We know that |Ik | → 0 since |Ik | = lk and by definition of C,
have
k
X
lj 2j−1 < 1
j=1

for all k.

4
Proposition 0.9 (Exercise 4c). Cˆ is perfect, and contains no open interval.
Proof. Let x ∈ C. ˆ By part (b), there exists a sequence (xn )∞ ˆ
n=1 with xn 6∈ C and xn → x.
Thus, for every r > 0, there exists N ∈ N such that n ≥ N =⇒ xn ∈ B(x, r). Hence x
is not an isolated point, so Cˆ has no isolated points. It is clear that Cˆ is closed, since its
complement is a union of open intervals by construction. Thus Cˆ is perfect.
Now we show that Cˆ contains no open interval. Suppose that Cˆ contains an open interval
(a, b), and let x ∈ (a, b). Then by part (b) we have a sequence (xn ) with xn 6∈ Cˆ and xn → x.
Let r = min(|x − a|, |x − b|). Then there exists N ∈ N such that n ≥ N implies xn ∈ B(x, r).
But by construction of r, B(x, r) ⊂ (a, b), so for n ≥ N , xn ∈ (a, b). This is a contradiction,
since xn 6∈ Cˆ and (a, b) ⊂ C.
ˆ Thus Cˆ contains no open interval.

Proposition 0.10 (Exercise 4d). Cˆ is uncountable.

ˆ At the kth stage of construction of C,
Proof. Let x ∈ C. ˆ we separate each closed subinterval
ˆ
into two subintervals. We define a funtion f : C → [0, 1] by
∞
X ak
f (x) =
k=1
2k

where ak = 0 if x is in a left subinterval at stage k, and ak = 1 if x is in a right subinterval

at stage k. We claim that f is surjective. To see this, let y ∈ [0, 1]. Then y has a (not
necessarily unique) binary expansion
∞
X bk
y=
k=1
2k

Then we use the bk to choose a subinterval remaining at each stage of the construction of
ˆ Let Ik be this chosen subinterval. For example, if b1 = 0, we choose the left of the two
C.
subintervals left after the first removal of a middle interval of length l1 . Now we take the
intersection over all such intervals, ∩∞
k=1 Ik , and we know that this intersection is non-empty,
since the endpoints of each Ik are never removed. If we choose some x ∈ ∩k Ik , then f (x) = y
by construction. Hence f is surjective.
Since there is a surjective function Cˆ 7→ [0, 1], we know that Cˆ cannot have smaller
cardinality than [0, 1]. Hence Cˆ is uncountable.
Proposition 0.11 (Exercise 5a). Let E ⊂ Rd and let

On = {x : d(x, E) < 1/n}

If E is compact, then

m(E) = lim m(On )

n→∞

Proof. Let E ⊂ Rd be compact. We have On ⊃ On+1 and clearly E ⊂ ∩n On . We also claim

that ∩n On ⊂ E. Suppose that x ∈ ∩n On . Then
1
d(x, E) = inf{d(x, y) : y ∈ E} <
n

5
for all n ∈ N. Thus, for every n, the B(x, 1/n) ∩ E 6= ∅. Thus x is a limit point of E. Since
E is closed, it contains all of its limit points, so x ∈ E. Thus ∩n On ⊂ E, so we have equality.
Thus we have a sequence On of measurable sets with On & E. Since E is closed and
bounded, each On is also bounded, so m(On ) < ∞ for all n. Then by Corollary 3.3, it follows
that

m(E) = lim m(On )

n→∞

Proposition 0.12 (Exercise 5b). The above proposition need not hold when E is closed and
unbounded, or when E is open and bounded.

Proof. First, we let E be the closed and unbounded set Q lying in R. We know that
m(Q) = 0. (One way to see this is by using Theorem 3.2 on the collection of singleton
sets of rationals.) However, On = {x : d(x, Q) < 1/n} is R for all n, since no matter how
small 1/n, every real number is always within 1/n of some rational. But m(R) = ∞, so
limn m(On ) = ∞.
Now we construct an open, bounded set that doesn’t satisfy the above property. Let
 > 0. Consider the rationals in [0, 1], and order them q1 , q2 , . . .. Around qk , form an open
ball of radius 2−k . Then let
∞
[
E= B(qk , 2−k )
k=1

E is open, since it is a union of open sets, and EP is bounded within [−1/2, 3/2]. We have
an upper bound on the measure of E by m(E) ≤ ∞ k=1 2
−k+1
 = 2. Now consider On . For
each k, we can see that B(qk , 2  + 1/n) ⊂ On , since everything in B(qk , 2−k  + 1/n) is
−k

within 1/n of B(qk , 2−k ). But for any x ∈ [0, 1], there is a rational within 1/n of x, so
∞
[
[0, 1] ⊂ B(qk , 2−k  + 1/n)
k=1

Thus m(On ) ≥ 1 for all n. Hence limn m(On ) ≥ 1. So for any  < 1/2, we have m(E) ≤ 1/2
but limn m(On ) ≥ 1.

Proposition 0.13 (Exercise 6). Let B1 ⊂ Rd be the unit ball, B1 = {x ∈ Rd : |x| < 1}. Let
B = B(x, r) be any ball in Rd with center x and radius r > 0. Then m(B) = rd m(B1 ).

Proof. We observe that B = x + rB1 , that is, B is equal to a translation of a dilation of the
unit ball. By translation and dilation properties of m, we have

m(B) = m(x + rB1 ) = m(rB1 ) = rd m(B1 )

6
Proposition 0.14 (Exercise 7). Let E ⊂ Rd be measurable, and let δ = (δ1 , δ2 , . . . δd ) with
each δi > 0. We define

δE = {(δ1 x1 , δ2 x2 , . . . δd xd ) : (x1 , x2 , . . . xd ) ∈ E}

Then δE is measurable, and more specifically,

d
!
Y
m(δE) = δi m(E)
i=1

Qd Qd
Proof. First we observe that if Q = i=1 [ai , bi ] is a closed rectangle, then δQ = i=1 [δi ai , δi bi ].
Then
d d d
! d
! d
!
Y Y Y Y Y
|δQ| = (δi bi − δi ai ) = δi (bi − ai ) = δi (bi − ai ) = δi |Q|
i=1 i=1 i=1 i=1 i=1

Let  > 0. Since E is measurable, there exists an open set O such that m∗ (E \O) ≤ . Let
{Qj }∞
j=1 be a covering of E \O by closed cubes. Then {δQj } is a covering of δE \δO by closed
cubes. (To see this, let y ∈ δE \ δO. Then y = (δ1 x1 , . . . δd xd ) where x = (x1 , . . . xd ) ∈ E \ O.
Then x ∈ Qj for some j, so then y ∈ δQj .) We note that δO must be open since O is open.
Thus
∞ d
! ∞ d
!
X Y X Y
m∗ (δE \ δO) ≤ |δQj | = δi |Qj | ≤ δi 
j=1 i=1 j=1 i=1
Q
Since i δi is some constant, we can make m∗ (δE \ δO) arbitrarily small, hence δE is
measurable.
Now let {Qj }∞
j=1 be a covering of E by closed cubes. Then {δQj } is a cover of δE by
closed cubes. (For analogous reasons to the above parenthetical note.) We also define

δ −1 E = {(δ1−1 x1 , . . . δd−1 xd ) : (x1 , . . . xd ) ∈ E}

Using this notation, we can say that if {Rj }∞ j=1 is a covering of δE by closed cubes, then
{δ −1 Rj } is a cover of E by closed cubes. With this identification, we can say that for any
covering {Qj } of E by closed cubes,
(∞ ) ( d ∞
)
X Y X
m(δE) = inf |δQj | = inf δi |Qj |
j=1 i=1 j=1
d
! (∞ ) d
!
Y X Y
= δi inf |Qj | = δi m(E)
i=1 j=1 i=1