Homework 3

Real Analysis
Joshua Ruiter
March 23, 2018

Proposition 0.1 (Exercise 13a). Let A ⊂ Rd be closed and B ⊂ Rd be open. Then A is a

Gδ set and B is an Fσ set.
Proof. Let An = {x : d(x, A) < 1/n}. We know that An is open because we can write is as
a union of open balls,
[  1
An = B a,
a∈A
n

S equality. If x ∈ An , then d(x, a) < 1/n for some a ∈ A, so

We pause to justify this
x ∈ B(a, 1/n). If x ∈ a B(a, 1/n), then d(x, a) < 1/n for all a ∈ A, so d(x, A) < 1/n.
Now we claim that A = ∩n∈N An . Let a ∈ A. Then B(a, 1/n) ⊂ An for all n, so
a ∈ ∩n B( a, 1/n) ⊂ ∩n An . Thus A ⊂ ∩n An .
Now suppose that x ∈ ∩n An . Then d(x, A) < 1/n for all n, so d(x, A) = 0. Then since
A is closed, and {x} is compact, by the contrapositive of Lemma 3.1, {x} and A are not
disjoint. But the only point at which they might intersect is x, hence x ∈ A. Putting this
together, we have established that A can be written as a countable intersection of open sets;
hence A is a Gδ set.
Now let B be open. Then let A = Rd \ B be the complement. As A is closed, A is a
countable intersection of open sets (A = ∩n An ), as shown above. Then let Bn be the closed
set Rd \ An , and we have
\ [ [
B = Rd \ A = Rd \ An = (Rd \ An ) = Bn
n n n

Thus we have written B as a countable union of closed sets, so B is an Fσ set.

Proposition 0.2 (Exercise 13b). There exists a set, namely Q, which is Fσ but not Gδ .
Proof. Consider the set Q of rationals. We can enumerate the rationals as {qi }∞
i=1 . Then we
can write Q as the countable union of singleton sets:
∞
[
Q= {qi }
i=1

1
Singleton sets are closed, so Q is an Fσ set.
We claim that Q is not a Gδ set. Suppose to the contrary that Q can be written as
a countable intersection of open sets, Q = ∩n On . Then On is an open set containing all
rationals, and we know that because On is an open set in R, it can be written as a disjoint
union of open intervals,

On = (a1 , b1 ) ∪ (a2 , b2 ) . . .

where a1 ≤ b1 ≤ a2 ≤ b2 ≤ . . .. If for any i we have bi 6= ai+1 , then there is a rational between

bi and ai+1 , but then On would not contain Q. Hence each bi = ai+1 . Clearly, we cannot
have bi ∈ Q, since Q ⊂ On , so O is a union of disjoint open intervals covering all of R except
for countably many irrationals. Then ∩n On countains all but countably many irrationals.
Hence ∩n On is not equal to Q, which contradicts our starting assumption. Hence Q is not
a Gδ set.

Corollary 0.3 (to Exercise 13b). The set R \ Q is not Gδ .

Proof. If R \ Q were Gδ , then its complement, Q would be Fσ , but we just showed that Q is
not Fσ .

Proposition 0.4 (Exercise 13c). There exists a set which is a Borel set but not Gδ or Fσ .

Proof. Let B = (−∞, 0) ∩ Q and let C = (0, ∞) \ Q, and set A = B ∪ C. We claim that A
is neither Gδ nor Fσ . Suppose that A is Fσ , that is, A can be written as the countable union
of closed sets, A = ∪n An . For each An , set

A+
n = An ∩ [0, ∞)
A−
n = An ∩ (−∞, 0]

Then we can rewrite A as

[ [
A= A+
n ∪ A−
n
n n

In particular, since B, C are disjoint, B = n A−

S S +
n and C = n An . But then C is an Fσ set,
but C cannot be a Fσ set for the same reasons that R \ Q is not Fσ . Thus A is not Fσ .
Now suppose that A is Gδ . Then R \ A is Fσ , so ((−∞, 0) \ Q) ∪ ((0, ∞) ∩ Q) ∪ {0} is Fσ .
We can then do the same construction as above, supposing R \ A to be a countable union of
−
An , and defining A+ n , An . Then again by the disjoint-ness, we would have to conclude that
(0, ∞) ∩ Q is Fσ , but once again, this set is not Fσ for the same reasons that R \ Q is not
Fσ . Hence A is not Gδ .
We have shown that A is neither Fσ nor Gδ . But we claim that A is a Borel set. As
shown in part (b), Q is Fσ , so likewise B is Fσ . For analogous reasons, R \ C is Fσ , so C is
Gδ . Hence A is a union of an Fσ set and a Gδ set. But both the Fσ and Gδ sets are Borel
sets, so their union is contained in the Borel sets (as the Borel sets form a σ-algebra). Hence
A is a Borel set.

2
Lemma 0.5 (for Exercise 14a). Let X be a topological space with subsets A1 , . . . An . Then
n
[ n
[
Ai = Ai
i=1 i=1

that is, the closure of a finite union is equal to the union of the closures.

Proof. Suppose x ∈ ni=1 Ai . Then x ∈ i Ai or x is a limit point of this union. If x is in this

S S
union, then x ∈ Ai for some i, so then x ∈ Ai and we’re done. So suppose x is a limit point
of this union. Then there exists a sequence xn converging to x where each xn ∈ Ai for some
i. Since there are only finitely many Ai , there must be some Ai containing infinitely many
xn . Choose the subsequence
Sn of S
xn lying within Ai . Then this subsequence also converges to
n
x, so x ∈ Ai . Hence S A
i=1 i ⊂ i=1 Ai .
Now suppose x ∈ i Ai . Then x ∈ Ai for some i or x is a limit point of some Ai . If
x ∈ Ai , then x is in the union over all Ai and hence in the closure of that union. If xS
is a limit
point of Ai , then x is also a limit point of the union over all Ai . Hence ni=1 Ai ⊃ ni=1 Ai .
S
We have shown two-way containment of these sets, thus they are equal.

Proposition 0.6 (Exercise 14a). The outer Jordan content J∗ (E) of a set E ⊂ R is defined
by
( N )
X
J∗ (E) = inf |Ij |
j=1

SN
where Ij are intervals such that E ⊂ j=1 Ij . We claim that J∗ (E) = J∗ (E) for E ⊂ R.

Proof. Every covering of E by intervals {Ij } is a covering of E, so J∗ (E) is an infimum over

a subset of the set over which J∗ (E) is an infimum, hence J∗ (E) ≤ J∗ (E).
Let {Ij } be a covering of E by intervals. Then E is covered by the intervals {I j },
since the closure of a finite union is equal to the union of the closures (see above lemma).
But |Ij | = |Ij |, so every covering of E extends to a covering of E with the same sum, so
J∗ (E) ≤ J∗ (E).
Thus we have inequalities going both ways, so the quantities J∗ (E), J∗ (E) are equal.

Proposition 0.7 (Exercise 14b). There exists a countable subset E ⊂ [0, 1] such that
J∗ (E) = 1 and m∗ (E) = 0. In particular, E = Q ∩ [0, 1].

Proof. Let E = Q ∩ [0, 1]. As we know, m∗ (Q) = 0 so m∗ (E) = 0. Suppose that {Ij } is a
covering of E by finitely many intervals,
n
[
E= (ai , bi )
i=1

(The intervals might not be open, they might be of the form [ai , bi ) or something, but this
doesn’t affect the argument.) This complement of this union must not contain any intervals
contained in [0, 1], since if there is such an interval, then that interval contains a rational,

3
contradicting the fact that E contains all rationals in [0, 1]. Thus the intervals covering E
must contain something of the form
[0, a1 ) ∪ (a1 , a2 ) ∪ (a2 , a3 ) ∪ . . . ∪ (an , 1]
where a1 , . . . an are irrational. But the sum over the lengths of these intervals is one, so for
any covering of E by intervals, the sum over the lengths of those intervals is at least one.
Hence J∗ (E) ≥ 1. Of course, the single interval [0, 1] is a covering for E, so J∗ (E) ≤ 1, which
with the previous inequality yields J∗ (E) = 1.
Proposition 0.8 (Exercise 15). Define, for E ⊂ Rd ,
(∞ ∞
)
X [
mR ∗ (E) = inf |Rj | : E ⊂ Rj
j=1 j=1

where Rj are closed rectangles. Then for m∗ (E) = mR d

∗ (E) for E ⊂ R .

Proof. Let E ⊂ Rd . Since every covering of E by closed cubes is also a covering by closed
rectangles, it is immediate that mR∗ (E) ≤ m∗ (E) for all E, since the former is an infimum
over a subset of the set over which the latter is an infimum. We just need to show that the
opposite inequality also holds.
If mR R R
∗ (E) = ∞, then m∗ (E) ≤ m∗ (E) holds trivially, so suppose that m∗ (E) < ∞. Let
 > 0. Then there exists a covering {Rj }∞j=1 of E by closed rectangles such that
∞
X
|Rj | < mR
∗ (E) + 
j=1

For each rectangle Rj , m∗ (Rj ) = |Rj |, so by definition of m∗ , there exists a covering {Qij }∞
i=1
of Rj by closed cubes such that
∞
X
|Qij | < |Rj | + /2j
i=1

Then {Qij } is a countable covering of E by closed cubes, so

∞ X
X ∞ ∞
X ∞
X
j
mR /2j = mR


m∗ (E) ≤ |Qij | ≤ |Rj | + /2 = ∗ (E) ++ ∗ (E) + 2
j=1 i=1 j=1 j=1

Since  > 0 is arbitrary, we get m∗ (E) ≤ mR

∗ (E), and hence we have equality.

Proposition 0.9 (Exercise 16, Borel-Cantelli Lemma). Let {Ek }∞

k=1 be a countable family
d
of measurable subsets of R such that
∞
X
m(Ek ) < ∞
k=1

Define
E = {x ∈ Rd : x ∈ Ek for infinitely many k} = lim sup(Ek )
k→∞

Then E is measurable, and m(E) = 0.

4
Proof. First, we claim that
∞
!
\ [
E= Ek
n=1 k≥n
S
Suppose that x ∈ E, that is, x ∈ Ek for infinitely many k. Then for any n ∈ N, x ∈ k≥n Ek ,
henceTx isSin the intersection over all such sets. For the other inclusion, suppose that
x ∈ ∞ n=1 k≥n Ek . If x were in only finitely many Ek , then we could take the max over
those k and we would have x 6∈ ∪k≥N where N is that maximum. But by assumption, x is
in every such union, so it must be that x is in infinitely many Ek . Hence x ∈ E. Thus we
have two way containment, and hence equality S of sets.
Each Ek is measurable by hypothesis, so k≥n Ek is measurable, and hence the countable
intersection of such sets is also measurable. Thus E is measurable. Now we claim that
m(E) = 0.
Let  > 0. Since we have that the series ∞
P
k=1 m(Ek ) converges and is finite, we can
always go out far enough that the tail of the series is less than . Formally, there exists
N ∈ N such that
∞
X
m(Ek ) < 
k=N

Using our established expression for E, we have the inclusion

∞
!
\ [ [
E= Ek ⊂ Ek
n=1 k≥n k≥N

And then by monotonicity of measure,

! ∞
[ X
m(E) ≤ m Ek ≤ m(Ek ) < 
k≥N k=N

Thus m(E) <  for every  > 0, hence m(E) = 0.

Proposition 0.10 (Exercise 17). Let {fn } be a sequence of measurable functions on [0, 1]
with |fn (x)| < ∞ for a.e. x. Then there exists a sequence cn of positive real numbers such
that
fn (x)
→ 0 a.e. x
cn
Proof. For any sequence cn of positive reals, let En be the sequence of sets given by
   
 fn (x)  1
En = x :    >
cn  n
We claim that for every n there exists cn ∈ R such that

m∗ (En ) < 2−n

5
Suppose no such cn exists. Then for every cn ∈ R+ , we have
n cn o
m∗ x : |fn (x)| > ≥ 2−n
n
In particular, for cn = n, 2n, 3n, . . . we have

{x : |fn (x)| > 1} ⊃ {x : |fn (x)| > 2} ⊃ {x : |fn (x)| > 3} ⊃ . . .

−n
2 ≤ m({x : |fn (x)| > 1}) ≤ m({x : |fn (x)| > 2}) ≤ m({x : |fn (x)| > 3}) ≤ . . .

Then if we take the intersection over all such sets, the measure is still at least 2−n , that is,
∞
!
\
2−n ≤ m {x : |fn (x)| > k}
k=1

But this intersection is {x : |f (x)| = ∞}, which by hypothesis has measure zero. Hence
there must be a cn ∈ R such that

m(En ) < 2−n

(Specifically, we just showed that we can take cn to be a positive integer multiple of n.) Now
we consider the set

E = {x : x ∈ Ek for infinitely many k}

Suppose x 6∈ E. Then x is in at most finitely many Ek , that is,

 
 fk (x)  1
 ck  ≤ k
 

for infinitely many k. Hence for x ∈ [0, 1] \ E, we have

fn (x)
lim=0
cn
n→∞

By the Borel-Cantelli lemma, since we showed that ∞

P P −n
n=1 m(En ) ≤ n2 < 1 < ∞, we
have m(E) = 0. Hence

fn (x)
lim = 0 a.e. x
n→∞ cn

Proposition 0.11 (Exercise 18). Every measurable function is the limit a.e. of a sequence
of continuous functions.
Proof. First we prove this in the case that f is finite-valued. Let f : A → R be a measurable
function (where A ⊂ Rd is measurable). By Theorem 4.3, there exists a sequence of step
functions ψk such that limk ψk (x) = f (x) for almost every x, that is, there exists B ⊂ A
such that ψk (x) → f (x) on B and m(A \ B) = 0.

6
 Let En = B(0, n) ∩ A for n ∈ N. Then ψk is measurable and finite-valued on En so by
Lusin’s theorem there exists a closed set Fn such that Fn ⊂ En and m(En \ Fn ) ≤ 2−n , and
ψk |Fn is continuous. By the Tietze extension theorem, there exists a continuous function
fn : A → R such that fn (a) = f (a) for a ∈ Fn . So we have fn (x) → f (x) for x ∈ Fn .
Suppose that fn does not converge to f almost everywhere, in particular, suppose that
fn (x) does not converge to f (x) for some x ∈ A. Then x ∈ A \ Fn for infinitely many n. We
know that
A \ Fn ⊂ (En \ Fn ) ∪ (Rd \ B(0, n))

x ∈ Rd \ B(0, n) for only finitely many n, we have x ∈ En \ Fn for infinitely many

and sinceP
n. Since ∞ n=1 m(En \ Fn ) ≤ 1 < ∞, by the Borel-Cantelli lemma (Exercise 16 above), we
have
m ({x : x ∈ Ek \ Fk for infinitely many k}) = 0
But
[
{x : x ∈ Ek \ Fk for infinitely many k} = {x ∈ A \ Fn }
n
= {x ∈ A : fn (x) does not converge to f (x)}
hence the set of non-convergence of fn to f has measure zero, so fn → f a.e.
Now suppose that f : A → [−∞, ∞] is a measurable function. Define a function gn :
A → [−∞, ∞] by

f (x) f (x) < ∞

gn (x) = n f (x) = ∞

−n f (x) = −∞


The sequence of functions gn converges pointwise to f on A. Furthermore, each gn is mea-

surable and finite-valued, so by the above there exist continuous functions hnk such that
limk hnk (x) = gn (x) a.e. We claim that
lim hnn (x) = f (x) a.e.
n→∞

for all x ∈ A. Let  > 0. If f (x) is finite, then there exists N such that for all n ∈ N,
k ≥ N =⇒ |gn (x) − hnk (x)| = |f (x) − hnk (x)| <  (this holds for x a.e.)
Since this holds for all n ∈ N, it holds in particular for n ≥ N . Thus
k ≥ N =⇒ |f (x) − hkk (x)| <  (a.e.)
and hence hnn (x) → f (x) a.e. for f (x) finite. Now suppose that f (x) = ∞ and let M > 0.
Then there exist N1 , N2 such that
n ≥ N1 =⇒ gn (x) > M
k ≥ N2 =⇒ |gn (x) − hnk (x)| < 1 (a.e.)

7
Then let N = max(N1 , N2 ). So then

n ≥ N =⇒ hnn (x) > M − 1 (a.e.)

and hence hnn (x) → ∞ = f (x) a.e.. A symmetric argument shows that hn n(x) → −∞
a.e. if f (x) = −∞. Thus the sequence of continuous functions hnn converges pointwise
to f for almost every x ∈ A, so every measurable function is the limit a.e. of continuous
functions.

Proposition 0.12 (Exercise 19a). Let A, B ⊂ Rd such that either A or B is open. Then
A + B is open.

Proof. Let x ∈ A + B. Then there exist a ∈ A, b ∈ B such that a + b = x. Since A is

open, there exists  > 0 such that B(a, ) ⊂ A. Then b + B(a, ) ⊂ A + B. Note that
b + B(a, ) = B(b + a, ), since translating the center of a ball is the same as translating
the ball by the same amount. Hence B(a + b, ) = B(x, ) ⊂ A + B, so A + B is open.
By interchanging the labels for A and B, we can see that if B is open the A + B must be
open.

Lemma 0.13 (for Exercise 19b). If A, B ⊂ Rd are compact, then A + B is compact.

Proof. Let xn be a sequence in A + B. Then for each n, there exist an ∈ A, bn ∈ B such

that xn = an + bn . Since A, B are compact, they are sequentially compact, thus an , bn have
convergent subsequences ank , bnk . Then xnk is a convergent subsequence of xn . Hence A + B
is sequentially compact, so it is compact.

Lemma 0.14 (for Exercise 19b). Every closed set in Rd can be written as a countable union
of compact sets.

Proof. Let A ⊂ Rd be closed. Then A ∩ B(0, n) is compact, and A = ∞

S
n=1 (A ∩ B(0, n).

Proposition 0.15 (Exercise 19b). Let A, B ⊂ Rd be closed. Then A + B is measurable.

Proof. We will show that A + B is FσS. Each A,SB is closed, so they can be written as a
countable union of compact sets, A = i Ai , B = j Bj . Then
[
A+B = (Ai + Bj )
i,j

By a previous lemma, since Ai , Bj are compact, Ai + Bj is compact. Hence we have written

A + B as a countable union of closed sets, so A + B is Fσ . Thus A + B is measurable.

Proposition 0.16 (Exercise 19c). There exist closed sets A, B such that A+B is not closed.

Proof. Let

A = {2 + 1/2, 3 + 1/3, 4 + 1/4, . . .} = {n + 1/n : n = 2, 3, . . .}

B = {−1, −2, −3, −4, . . .} = {−n : n ∈ N}

8
Then A and B are both closed. But A + B is not closed, since it does not contain all of its
limit points. In particular, the sequence
 
1 1 1
, , ,...
2 3 4

is in A + B. However, its limit (zero) is not in A + B, since there are no positive integers in
A.

Proposition 0.17 (Exercise 20a). There exist closed sets A, B ⊂ R such that m(A) =
m(B) = 0 and m(A + B) > 0.

Proof. Let A = C be the middle thirds Cantor set, and let B = (1/2)C (the dilation of the
Cantor set by 1/2). Let x ∈ [0, 1]. Then x has a ternary representation,
∞
X
x= ak 3−k
k=1

where each ak ∈ {0, 1, 2}. We form a partition of K0 , K1 , K2 of N where Ki = {k ∈ N : ak =

i}. Since the infinite sum for x converges and the terms are positive, we can rearrange terms
to rewrite it as
∞
X X X X X X
ak 3−k = (0)3−k + (1)3−k + (2)3−k = 3−k + (2)3−k
k=1 k∈K0 k∈K1 k∈K2 k∈K1 k∈K2

Then notice that k∈K1 (2)3−k ∈ C by Exercise 4a, and k∈K2 3−k ∈ (1/2)C (since multi-
P P
plying it by 2 puts it in C for the same reason). Hence x is a sum of something in A and B,
so x ∈ A + B. Thus [0, 1] ⊂ A + B. Then by monotonicity, m(A + B) ≥ 1. However, as we
know, m(A) = m(B) = 0.

Proposition 0.18 (Exercise 20b). There exist closed sets A, B ⊂ R2 such that m(A) =
m(B) = 0 and m(A + B) > 0.

Proof. Let A = [0, 1] × {0} and B = {0} × [0, 1]. Then A + B = [0, 1] × [0, 1] since any
(x, y) ∈ [0, 1] × [0, 1] is equal to (x, 0) + (0, y). Hence m(A + B) = 1, but m(A) = m(B) =
0.