You are on page 1of 16

Real Analysis Chapter 3 Solutions Jonathan Conder

n=1 is an increasing sequence in M. For each n ∈ N define Fn := En \ En−1 (with E0 := ∅). Clearly
1. Suppose (En )∞

X N
X
ν(∪∞ ∞
n=1 En ) = ν(∪n=1 Fn ) = ν(Fn ) = lim ν(Fn ) = lim ν(EN ).
N →∞ N →∞
n=1 n=1

n=1 is a decreasing sequence in M and ν(E1 ) < ∞, then


If (En )∞

ν(∩∞ ∞ ∞
n=1 En ) = ν(E1 \ (E1 \ ∩n=1 En )) = ν(E1 ) − ν(∪n=1 (E1 \ En )) = ν(E1 ) − lim ν(E1 \ En ) = lim ν(En ).
n→∞ n→∞

2. Let E be measurable, and suppose that E is ν-null but |ν|(E) 6= 0. Then ν + (E) − ν − (E) = ν(E) = 0 and ν + (E) +
ν − (E) = |ν|(E) > 0, so ν + (E) = ν − (E) > 0. Since ν + ⊥ ν − , there exist disjoint measurable sets A and B covering
X such that A is ν + -null and B is ν − -null. In particular ν + (E ∩ B) = ν + (E ∩ B) + ν + (E ∩ A) = ν + (E) > 0 but
ν − (E ∩ B) ≤ ν − (B) = 0. This implies that ν(E ∩ B) > 0, which is a contradiction because E ∩ B ⊆ E and E is ν-null.
Conversely, suppose that |ν|(E) = 0. If F ⊆ E is measurable, then |ν|(F ) ≤ |ν|(E) = 0 so ν + (F ) + ν − (F ) = 0. In
particular ν + (F ) = 0 = ν − (F ), so ν(F ) = 0 and hence E is ν-null.
Suppose that ν ⊥ µ, so that X = A ∪ B for some disjoint measurable sets A and B such that A is ν-null and B is
µ-null. From above, |ν|(A) = 0, so A is |ν|-null (because |ν| is positive) and hence |ν| ⊥ µ.
Now suppose that |ν| ⊥ µ, so that X = A ∪ B for some disjoint measurable sets A and B such that A is |ν|-null and
B is µ-null. Since ν + ≤ |ν| and ν − ≤ |ν| pointwise, A is ν + -null and ν − -null. Therefore ν + ⊥ µ and ν − ⊥ µ.
Finally, suppose that ν + ⊥ µ and ν − ⊥ µ. Then X = A ∪ B = E ∪ F for two pairs A, B and E, F of disjoint
measurable sets such that A is ν + -null, E is ν − -null and B and F are µ-null. Note that B ∪ F is µ-null, because
every subset of B ∪ F = B ∪ (F \ B) is the disjoint union of two µ-null sets. Moreover X \ (B ∪ F ) = A ∩ E, which
is both ν + -null and ν − -null, and hence ν-null. This shows that ν ⊥ µ.

3. (a) Let φ ∈ L+ be a simple function, and write φ = ni=1 ai χEi . Then


P

Z Xn n
X n
X n
X Z Z

φ d|ν| = ai |ν|(Ei ) = +
ai (ν (Ei ) + ν (Ei )) = +
ai ν (Ei ) + ai ν (Ei ) = φ dν + φ dν − . (1)
− +

i=1 i=1 i=1 i=1

Hence, if f ∈ L1 (ν) then


Z Z Z  Z Z
φ dν + + φ dν − φ ∈ L+ simple with φ ≤ |f | ≤ |f | dν + + |f | dν − < ∞

|f | d|ν| = sup

because L1 (ν) = L1 (ν + ) ∩ L1 (ν − ). Therefore L1 (ν) ⊆ L1 (|ν|). Conversely, if f ∈ L1 (|ν|) it is clear from (1) that
φ dν + ≤ φ d|ν| ≤ |f | d|ν| for all simple φ ∈ L+ with φ ≤ |f |, so |f | dν + ≤ |f | d|ν| < ∞. This shows that
R R R R R

L1 (|ν|) ⊆ L1 (ν + ), and a similar argument shows that L1 (|ν|) ⊆ L1 (ν − ). Therefore L1 (|ν|) ⊆ L1 (ν).
(b) Let A and B be disjoint measurable sets covering X such that A is ν + -null and B is ν − -null. Then gχA = 0
ν + -a.e. and gχB = 0 ν − -a.e. for every measurable function g. In particular, from (1) φ d|ν| = φ dν + for all
R R

simple functions φ ∈ L+ with φ ≤ f + χB or φ ≤ f − χB . This implies that


Z Z Z Z Z Z
− −
f χB d|ν| = f χB d|ν| − f χB d|ν| = f χB dν − f χB dν = f χB dν + ,
+ + + +

f χA dν − . Moreover |χA − χB | = χX because A ∪ B = X and A ∩ B = ∅. Therefore


R R
and similarly f χA d|ν| =
Z Z Z
f dν = f (χA + χB ) dν + − f (χA + χB ) dν −

1
Real Analysis Chapter 3 Solutions Jonathan Conder

Z Z
+ −

= f χB dν − f χA dν
Z Z

= f χB d|ν| − f χA d|ν|

Z

= f (χB − χA ) d|ν|
Z
≤ |f (χB − χA )| d|ν|
Z
= |f | d|ν|.
R R
(c) Define g := χB − χA . Then |g| ≤ 1 and hence |ν|(E) = | E g dν| ≤ sup{| E f dν| | |f | ≤ 1} because
Z Z
g dν = (χB − χA )χE dν
E Z Z
= χB χE dν − −χA χE dν −
+

Z Z
= χ(B∩E) dν + χ(A∩E) dν −
+

= ν + (B ∩ E) + ν − (A ∩ E)
= ν + (A ∩ E) + ν + (B ∩ E) + ν − (A ∩ E) + ν − (B ∩ E)
= ν + (E) + ν − (E)
= |ν|(E) ≥ 0.

This completes the proof if |ν|(E) = ∞. Assume that |ν|(E) < ∞ and let f be a measurable function with
|f | ≤ 1. Then f χE ∈ L1 (|ν|) = L1 (ν) because E |f | d|ν| ≤ χE d|ν| = |ν|(E) < ∞. Hence, by the previous
R R
R R R R
exercise | E f dν| ≤ |f χE | d|ν| ≤ χE d|ν| = |ν|(E). Therefore sup{| E f dν| | |f | ≤ 1} ≤ |ν|(E).

4. Let A and B be disjoint measurable sets covering X such that A is ν + -null and B is ν − -null. If E is measurable then

λ(E) ≥ λ(E ∩ B) = ν(E ∩ B) + µ(E ∩ B) ≥ ν(E ∩ B) = ν + (E ∩ B) = ν + (E ∩ B) + ν + (E ∩ A) = ν + (E).

and similarly

µ(E) ≥ µ(E ∩ A) = µ(E ∩ A) + ν + (E ∩ A) = λ(E ∩ A) + ν − (E ∩ A) ≥ ν − (E ∩ A) = ν − (E ∩ A) + ν − (E ∩ B) = ν − (E).

5. Clearly v1 + v2 is a signed measure, and v1 + v2 = v1+ − v1− + v2+ − v2− = (v1+ + v2+ ) − (v1− + v2− ). Since v1+ + v2+ and
v1− + v2− are positive measures, the previous exercise implies that v1+ + v2+ ≥ (v1 + v2 )+ and v1− + v2− ≥ (v1 + v2 )− .
Therefore |v1 + v2 | = (v1 + v2 )+ + (v1 + v2 )− ≤ v1+ + v2+ + v1− + v2− = |v1 | + |v2 |.

7. (a) Let A and B be disjoint measurable sets covering X such that A is ν + -null and B is ν − -null. Then

ν + (E) = ν + (E ∩ B) + ν + (E ∩ A) = ν + (E ∩ B) = ν(E ∩ B) ≤ sup{ν(F ) | F ∈ M, F ⊆ E}.

Moreover, if F ∈ M and F ⊆ E then

ν(F ) = ν + (F ) − ν − (F ) ≤ ν + (F ) ≤ ν + (E).

Therefore ν + (E) ≥ sup{ν(F ) | F ∈ M, F ⊆ E}, so ν + (E) = sup{ν(F ) | F ∈ M, F ⊆ E}. A similar argument


shows that −ν − (E) = ν(E ∩ A) and hence ν − (E) = − inf{ν(F ) | F ∈ M, F ⊆ E}.

2
Real Analysis Chapter 3 Solutions Jonathan Conder

(b) Clearly |ν|(E) ≤ sup{ ni=1 |ν(Ei )| | n ∈ N and (Ei )ni=1 is a pairwise disjoint sequence in M covering E}, as
P

|ν(E ∩ A)| + |ν(E ∩ B)| = | − ν − (E ∩ A)| + |ν + (E ∩ B)|


= ν − (E ∩ A) + ν + (E ∩ B)
= ν + (E ∩ A) + ν − (E ∩ A) + ν + (E ∩ B) + ν − (E ∩ B)
= |ν|(E ∩ A) + |ν|(E ∩ B)
= |ν|(E).

Now let n ∈ N and (Ei )ni=1 is a pairwise disjoint sequence in M covering E. Then
n
X n
X n
X n
X
|ν(Ei )| ≤ (|ν + (Ei )| + |ν − (Ei )|) = ν + (Ei ) + ν − (Ei ) = ν + (E) + ν − (E) = |ν|(E).
i=1 i=1 i=1 i=1
Pn
This shows that |ν|(E) ≥ sup{ i=1 |ν(Ei )| | n ∈ N and (Ei )ni=1 is a pairwise disjoint sequence in M covering E},
and hence |ν|(E) = sup{ ni=1 |ν(Ei )| | n ∈ N and (Ei )ni=1 is a pairwise disjoint sequence in M covering E}.
P

8. Suppose ν  µ. If E ∈ M and µ(E) = 0 then ν(F ) = µ(F ) = 0 for all F ∈ M with F ⊆ E, so E is ν-null. By exercise
2, this implies that |ν|(E) = 0. Therefore |ν|  µ.
Now suppose that |ν|  µ. If E ∈ M and µ(E) = 0, then ν + (E) + ν − (E) = 0 and hence ν + (E) = ν − (E) = 0. This
shows that ν +  µ and ν −  µ.
Finally, suppose that ν +  µ and ν −  µ. If E ∈ M and µ(E) = 0, then ν(E) = ν + (E) − ν − (E) = 0 − 0 = 0, and
hence ν  µ.

9. Suppose that νn ⊥ µ for all n ∈ N. For each n ∈ N there exists a measurable set En such that En is νn -null and Enc
is µ-null. Define E := ∩∞ En and note that E c = ∪∞ c c
n=1 En ; in particular E is µ-null. It is also clear that E is a null
P∞ n=1 P∞
set with respect to n=1 νn . Therefore n=1 νn ⊥ µ. The second part is trivial.

11. (a) If f ∈ L1 (µ) then {f } is uniformly integrable, because the finite signed measure E 7→ E f dµ is absolutely
R

continuous with respect to µ. Hence, if {fi }i∈I is a finite subcollection of L1 (µ) and ε ∈ (0, ∞), for each i ∈ I
there exists δi ∈ (0, ∞) such that | E fi dµ| < ε for all E ∈ M with µ(E) < δi . Set δ := min{δi }i∈I , so that
R

| E fi dµ| < ε for all i ∈ I and E ∈ M with µ(E) < δ. This shows that {fi }i∈I is uniformly integrable.
R

(b) Let ε ∈ (0, ∞), and choose N ∈ N such that |fn − f | dµ < 2ε for all n ∈ N with n ≥ N. Define f0 := f and
R
ε
I := {n}N
R
n=0 . Since {fn }n∈I is uniformly integrable, there exists δ ∈ (0, ∞) such that | E fn dµ| < 2 for all n ∈ I
and E ∈ M with µ(E) < δ. If n ∈ N \ I and E ∈ M then
Z Z Z Z Z Z Z

fn dµ = (fn − f ) dµ + f dµ ≤ (fn − f ) dµ + f dµ ≤ |fn − f | dµ + f0 dµ

E E E E E E E

fn dµ| < ε for all n ∈ N and E ∈ M with µ(E) < δ.


R R R
Since E |fn − f | dµ ≤ |fn − f | dµ, this implies that | E
Therefore {fn }∞
n=1 is uniformly integrable.

13. (a) Let E ∈ M and suppose that µ(E) = 0. Then E = ∅ so m(E) = 0. Therefore m  µ. Suppose there exists an
R
extended µ-integrable function f such that dm = f dµ. If x ∈ X then f (x) = {x} f dµ = m({x}) = 0, so f = 0.
R
Therefore m(X) = X f dµ = 0, which is a contradiction because m(X) = 1.

3
Real Analysis Chapter 3 Solutions Jonathan Conder

(b) Suppose that µ has a Lebesgue decomposition λ + ρ with respect to m, with λ ⊥ m and ρ  m. Then ρ({x}) = 0
and hence λ({x}) = µ({x}) = 1 for all x ∈ X. There exist disjoint measurable sets A and B covering X such
that A is λ-null and m(B) = 0. If x ∈ A then 1 = λ({x}) = 0 so A = ∅. Therefore m(B) = m(X) = 1, which
contradicts m(B) = 0.

16. For each n ∈ N let En := {x ∈ X | f (x) < −n−1 }, so that


Z Z
−1 −1
−n λ(En ) = −n dλ ≥ f dλ = ν(En ) ≥ 0
En En

and hence µ(En ) ≤ λ(En ) = 0. It follows that µ(∪∞ ∞


n=1 En ) = 0, so f ≥ 0 µ-a.e. because ∪n=1 En = {x ∈ X | f (x) < 0}.
Now set F := {x ∈ X | f (x) ≥ 1}. Since ν is σ-finite, there is a sequence (Fn )∞n=1 of subsets of F which cover F such
that ν(Fn ) < ∞ for all n ∈ N. Since
Z Z
ν(Fn ) = f dλ ≥ 1 dλ = λ(Fn ) = µ(Fn ) + ν(Fn )
Fn Fn

it is clear that µ(Fn ) = 0 for all n ∈ N, in which case µ(F ) = 0 and hence f < 1 µ-a.e. Without loss of generality
f, 1 − f ∈ L+ , so for each E ∈ M
Z Z
(1 − f ) dλ + ν(E) = 1 dλ = λ(E) = µ(E) + ν(E).
E E

In particular E (1 − f ) dλ = µ(E) for all E ∈ M with ν(E) < ∞. This result extends to all E ∈ M because ν is
R

σ-finite, using the monotone convergence theorem and the additivity of µ. Therefore dµ
dλ = (1 − f ). Since µ  λ (in
dλ 1
fact µ ≤ λ) and λ  µ (as ν  µ), it follows that dµ = 1−f . This implies that

dν dν dλ f
= = .
dµ dλ dµ 1−f

17. Define a measure λ on M by λ(E) := E f dµ. Then λ is finite because f ∈ L1 (µ). Define ρ := λ|N . Clearly ρ  ν, so
R

by the Radon-Nikodym theorem there is an extended ν-integrable function g such that ρ(E) = E g dν for all E ∈ N.
R

In fact g ∈ L1 (ν) since ρ(X) < ∞. Moreover g is unique up to equality ν-a.e. by the Radon-Nikodym theorem.

19. There exist positive measures ρ and σ and complex-valued functions f ∈ L1 (ρ) and g ∈ L1 (σ) such that dν = f dρ,
d|ν| = |f | dρ, dµ = g dσ and d|µ| = |g| dσ. Suppose that ν ⊥ µ. For each pair a, b ∈ {r, i} there exist disjoint sets
Aab , Bab ∈ M such that Aab ∪ Bab = X, Aab is νa -null and Bab is µb -null. It follows that A := (Arr ∪ Ari ) ∩ (Air ∪ Aii )
is both νr -null and νi -null. In particular, if E ∈ M and E ⊆ A then 0 = νr (E) = Re( E f dρ) = E Re(f ) dρ. By
R R

applying this to the sets {x ∈ A | Re(f (x)) > n−1 } and {x ∈ A | Re(f (x)) < −n−1 } for all n ∈ N, we can show that
Re(f )χA = 0 ρ-a.e., and similarly Im(f )χA = 0 ρ-a.e., giving |f |χA = 0 ρ-a.e. and hence A is |ν|-null. Moreover

B := Ac = (Arr ∪ Ari )c ∪ (Air ∪ Aii )c = (Brr ∩ Bri ) ∪ (Bir ∩ Bii )

is both µr -null and µi -null, and a similar argument shows that B is |µ|-null. Therefore |ν| ⊥ |µ|.
Conversely, suppose that |ν| ⊥ |µ|. There exist disjoint sets A, B ∈ M such that A ∪ B = X, A is |ν|-null and B is
|µ|-null. It follows that |f |χA = 0 ρ-a.e., so that Re(f )χA = 0 = Im(f )χA ρ-a.e. and hence A is both νr -null and
νi -null. Similarly B is both µr -null and µi -null, which implies that νa ⊥ µb for all a, b ∈ {r, i}. This shows that ν ⊥ µ.

4
Real Analysis Chapter 3 Solutions Jonathan Conder

Now suppose that ν  λ, and let A ∈ M satisfy λ(A) = 0. If E ∈ M and E ⊆ A then λ(E) = 0 and hence νa (E) = 0
for each a ∈ {r, i}. It follows that E Re(f ) dρ = E Im(f ) dρ = 0 for all E ∈ M with E ⊆ A, so |f |χA = 0 ρ-a.e.
R R

which implies that |ν|(A) = 0. This shows that |ν|  λ.


Conversely, if |ν|  λ then |ν(A)| ≤ |ν|(A) = 0, and hence νa (A) = 0, for each a ∈ {r, i} and all A ∈ M with
λ(A) = 0. This implies that νr  λ and νi  λ, so ν  λ.

21. If n ∈ N and E1 , E2 , . . . , En ∈ M are disjoint with E = ∪nk=1 Ek , then E1 , E2 , . . . are disjoint, E = ∪∞


k=1 Ek and
P∞ Pn ∞
k=1 |ν(Ek )| = k=1 |ν(Ek )| where En+k := ∅ for all k ∈ N. This implies that µ1 (E) ≤ µ2 (E). Moreover, if (Ek )k=1
is a sequence of disjoint measurable sets such that E = ∪∞ k=1 Ek , then


X ν(Ek )
f := χE
|ν(Ek )| k
k=1

is well-defined and satisfies |f | ≤ 1 (by disjointness). If a ∈ {r, i} and b ∈ {+, −} then 1 ∈ L1 (νab ) so
Z n
Z X n ∞
ν(Ek ) X ν(Ek ) b X ν(Ek ) b
f dνab = lim b
χE dν = lim ν (Ek ) = ν (Ek )
n→∞ |ν(Ek )| k a n→∞ |ν(Ek )| a |ν(Ek )| a
k=1 k=1 k=1

by the dominated convergence theorem. It follows that


Z Z Z
f dν = f dνr + i f dνi
Z Z Z Z
= f dνr+ − f dνr− + i f dνi+ − i f dνi−

X ν(Ek ) +
= (ν (Ek ) − νr− (Ek ) + iνi+ (Ek ) − iνi− (Ek ))
|ν(Ek )| r
k=1

X ν(Ek )
= (νr (Ek ) + iνi (Ek ))
|ν(Ek )|
k=1

X ν(Ek )
= ν(Ek )
|ν(Ek )|
k=1

X |ν(Ek )|2
=
|ν(Ek )|
k=1
X∞
= |ν(Ek )|.
k=1

This implies that | E f dν| = ∞


R P
k=1 |ν(Ek )|, so µ2 (E) ≤ µ3 (E). If f is a complex-valued measurable function on X

R R R
such that |f | ≤ 1 then | E f dν| ≤ E |f | d|ν| ≤ E 1 d|ν| = |ν|(E), so µ3 (E) ≤ |ν|(E). Conversely, if f := d|ν| (which
exists because ν  |ν|) then |f | = 1 |ν|-almost everywhere and hence |f | ≤ 1 everywhere (without loss of generality).
By generalising the chain rule to complex measures, it follows that
Z Z Z Z Z
dν 2

µ3 (E) ≥ f dν = f d|ν| = f f d|ν| = |f | d|ν| = 1 d|ν| = ||ν|(E)| = |ν|(E).
E E d|ν| E E E

This shows that µ3 (E) = |ν|(E), so it remains to show that µ3 (E) ≤ µ1 (E). If f is a complex-valued measurable
function on X such that |f | ≤ 1, there exists a sequence (φk )∞
k=1 of simple functions which converges pointwise to f

5
Real Analysis Chapter 3 Solutions Jonathan Conder

such that (|φk |)∞


k=1 is increasing to |f |. For each k ∈ N let
nk
X
φk = ckj χEkj
j=1

be the standard representation of φk . By the dominated convergence theorem, if a ∈ {r, i} and b ∈ {+, −} then
Z Z Xnk
b b
f dνa = lim φk dνa = lim ckj νab (Ekj ∩ E)
E k→∞ E k→∞
j=1

and hence
Z Z Z
f dν = f dνr + i f dνi
E ZE Z E Z Z
= f dνr+ − f dνr− + i f dνi+ − i f dνi−
E E E E
 
Xnk Xnk nk
X nk
X
= lim  ckj νr+ (Ekj ∩ E) − ckj νr− (Ekj ∩ E) + i ckj νi+ (Ekj ∩ E) − i ckj νi− (Ekj ∩ E)
k→∞
j=1 j=1 j=1 j=1
nk
X
= lim ckj (νr+ (Ekj ∩ E) − νr− (Ekj ∩ E) + iνi+ (Ekj ∩ E) − iνi− (Ekj ∩ E))
k→∞
j=1
Xnk
= lim ckj (νr (Ekj ∩ E) + iνi (Ekj ∩ E))
k→∞
j=1
Xnk
= lim ckj ν(Ekj ∩ E).
k→∞
j=1

Since |φk | ≤ 1 and Ek1 , Ek2 , . . . , Eknk are disjoint for each k ∈ N, it follows that

nk nk nk
Z X X X

f dν = lim
k→∞ ckj ν(Ekj ∩ E) ≤ lim
|ckj ||ν(Ekj ∩ E)| ≤ lim |ν(Ekj ∩ E)| ≤ µ1 (E).
k→∞ j=1 k→∞

E j=1 j=1

Therefore µ3 (E) ≤ µ1 (E), as required.


R
22. Assuming kf k1 > 0, there exists R ∈ (0, ∞) such that BR (0) |f | dm > 0 (otherwise the monotone convergence theorem
implies that |f | dm = limN →∞ BN (0) |f | dm = 0). If x ∈ Rn \ BR [0], then BR (0) ⊆ B2|x| (x) so
R R

Z Z Z
1 1 1
Hf (x) ≥ A2|x| |f |(x) = |f | dm ≥ |f | dm = |f | dm.
m(B2|x| (x)) B2|x| (x) m(B2|x| (0)) BR (0) |x|n m(B2 (0)) BR (0)

Note that C := m(B12 (0)) BR (0) |f | dm is positive and independent of x. Now, if α ∈ (0, 2R C n such that
R
n ) and x ∈ R

R < |x| < ( C 1/n then Hf (x) ≥ C|x|−n > C α = α and hence B n
α) C (C
α
)1/n (0) \ BR [0] ⊆ {x ∈ R | Hf (x) > α}. Thus
 
n
  C n Cm(B1 (0))
m({x ∈ R | Hf (x) > α}) ≥ m B( C )1/n (0) − m(BR [0]) = − R m(B1 (0)) > .
α α 2α

23. Let x ∈ Rn and r ∈ (0, ∞), so that Br (x) is a ball with x ∈ Br (x). Then m(B1r (x)) Br (x) |f | dm ≤ H ∗ f (x) and hence
R

Hf (x) ≤ H ∗ f (x). Conversely, let y ∈ Rn and r ∈ (0, ∞) such that x ∈ Br (y). Clearly Br (y) ⊆ B2r (x), so
2n
Z Z Z
1 1
|f | dm ≤ |f | dm = |f | dm ≤ 2n Hf (x)
m(Br (y)) Br (y) m(Br (y)) B2r (x) m(B2r (x)) B2r (x)
and hence H ∗ f (x) ≤ 2n Hf (x).

6
Real Analysis Chapter 3 Solutions Jonathan Conder

24. Let ε ∈ (0, ∞) and take r ∈ (0, ∞) such that |f (y) − f (x)| < ε for all y ∈ Br (x). Then
Z Z Z
1 1 1 εm(Br (x))
|f (y) − f (x)| dy = |f (y) − f (x)| dy < ε dy = = ε.

m(Br (x)) Br (x) m(Br (x)) Br (x) m(Br (x)) Br (x) m(Br (x))

1
R
This shows that limr↓0 m(Br (x)) Br (x) |f (y) − f (x)| dy = 0. Therefore x ∈ Lf .

25. (a) Define µ : BRn → [0, ∞] by µ(A) = m(E ∩ A). Clearly µ is a measure with Lebesgue-Radon-Nikodym represen-
tation dµ = χE dm. Let A ∈ BRn and suppose m(A) < ∞. Given ε ∈ (0, ∞), there exists an open set U ⊆ Rn
containing A such that m(U ) < m(A) + ε and hence m(U \ A) < ε. It follows that

µ(U ) = m(E ∩ U ) ≤ m(E ∩ A) + m(U \ A) < µ(A) + ε.

Given A ∈ BRn and ε ∈ (0, ∞), choose a sequence (Ak )∞


k=1 of Borel sets such that m(Ak ) < ∞ for all k ∈ N and
∪k=1 Ak = A. For each k ∈ N there exists an open set Uk ⊆ Rn containing Ak such that µ(Uk ) < µ(Ak ) + 2−k ε.

Clearly A ⊆ ∪∞ ∞ n
k=1 Uk and µ(∪k=1 Uk \ A) < ε. This implies that µ(A) = inf{µ(U ) | U ⊆ R is open and A ⊆ U }.
Therefore µ is regular, so for almost every x ∈ Rn

m(E ∩ Br (x)) µ(Br (x))


DE (x) = lim = lim = χE (x).
r↓0 m(Br (x)) r↓0 m(Br (x))

In particular DE (x) = 1 for almost all x ∈ E and DE (x) = 0 for almost all x ∈ E c .
(b) Given α ∈ [0, 14 ) define Eα := {(x, y) ∈ (0, ∞)2 | y ≤ x tan(2πα)}. Fix r ∈ (0, ∞) and note that Eα ∩ Br (0) is
a sector of the disk Br (0) between the angles 0 and 2πα. Hence m(Eα ∩ Br (0)) = αm(Br (0)) and DEα (0) = α.
For α ∈ [ 41 , 1) just include some of the other quadrants (0, ∞) × (−∞, 0), (−∞, 0) × (0, ∞) or (−∞, 0)2 in Eα .
Now define E := ∪∞
n=1 [2
−n , 2−n + 2−n−1 ] and fix N ∈ N. Then E ∩ B ∞
2−N (0) = ∪n=N +1 [2
−n , 2−n + 2−n−1 ] so


X ∞
X
−n −n −n−1
m(E ∩ B2−N (0)) = m([2 ,2 +2 ]) = 2−n−1 = 2−N −1 ,
n=N +1 n=N +1

while m(B2−N (0)) = 2 · 2−N = 21−N . This implies that

m(E ∩ B2−N (0)) 2−N −1 1


= 1−N = 2−2 = .
m(B2−N (0)) 2 4

On the other hand E ∩ B2−N +2−N −1 (0) = (∪∞


n=N +1 [2
−n , 2−n + 2−n−1 ]) ∪ [2−N , 2−N + 2−N −1 ), which implies that


X ∞
X
m(E ∩ B2−N +2−N −1 (0)) = m([2−N , 2−N + 2−N −1 )) + m([2−n , 2−n + 2−n−1 ]) = 2−n−1 = 2−N ,
n=N +1 n=N

but m(B2−N +2−N −1 (0)) = 2 · (2−N + 2−N −1 ) = 21−N + 2−N = 3 · 2−N and hence

m(E ∩ B2−N +2−N −1 (0)) 2−N 1


= = .
m(B2−N +2−N −1 (0)) 3 · 2−N 3

Since 2−N and 2−N + 2−N −1 converge to 0 as N → ∞, it follows that DE (0) does not exist.

7
Real Analysis Chapter 3 Solutions Jonathan Conder

28. (a) If x ∈ R then


( n )
X
TF (x) = sup |F (xk ) − F (xk−1 )| n ∈ N and x0 , x1 , . . . , xn ∈ R with x0 < x1 < · · · < xn = x


k=1
( n )
X
= sup |µF ((−∞, xk ]) − µF ((−∞, xk−1 ])| n ∈ N and x0 , x1 , . . . , xn ∈ R with x0 < x1 < · · · < xn = x


k=1
( n )
X
= sup |µF ((xk−1 , xk ])| n ∈ N and x0 , x1 , . . . , xn ∈ R with x0 < x1 < · · · < xn = x


k=1
( n )
X
≤ sup |µF (Ek )| n ∈ N and E1 , E2 , . . . , En ∈ BR are disjoint with ∪k=1 Ek ⊆ (−∞, x]
n


k=1
( n )
X
≤ sup |µF (Ek )| n ∈ N and E1 , E2 , . . . , En ∈ BR are disjoint with ∪nk=1 Ek = (−∞, x]


k=1

= |µF |((−∞, x])


= G(x),

where the penultimate equality follows from exercise 21.


(b) If a, b ∈ R and a < b then
( n )
X
TF (b) − TF (a) = sup |F (xk ) − F (xk−1 )| n ∈ N and x0 , x1 , . . . , xn ∈ R with a = x0 < x1 < · · · < xn = b


k=1

and hence

|µF ((a, b])| = |µF ((−∞, b]) − µF ((−∞, a])| = |F (b) − F (a)| ≤ TF (b) − TF (a) = µTF ((a, b]).

The set A of all finite disjoint unions of left-open, right-closed intervals is an algebra of subsets of R. If x ∈ R

X n X n Xn
|µF ((−∞, x])| = µF ((x − k, x + 1 − k]) ≤ |µF ((x−k, x+1−k])| ≤ µTF ((x−k, x+1−k]) = µTF ((−∞, x])


k=1 k=1 k=1

and similarly |µF ((x, ∞))| ≤ µTF ((x, ∞)). By the same argument, it follows that |µF (E)| ≤ µTF (E) for all
E ∈ A (because |µF (∅)| = µTF (∅)). Define C := {E ∈ BR | |µF (E)| ≤ µTF (E)}. If (Ek )∞ k=1 is an increasing
sequence in C, then |µF (∪k=1 Ek )| = | limk→∞ µF (Ek )| = limk→∞ |µF (Ek )| ≤ limk→∞ µTF (Ek ) = µTF (∪∞

k=1 Ek )
and hence ∪k=1 Ek ∈ C. Similarly C is closed under countable decreasing intersections (because µTF and the real

and imaginary parts of µF are finite). Therefore C contains the monotone class generated by A, which is BR by
the monotone class lemma and the fact that BR is generated by A. Thus |µF (E)| ≤ µTF (E) for all E ∈ BR .
(c) If E ∈ BR then, by exercise 21,
(∞ )
X
∞ ∞
|µF |(E) = sup |µF (Ek )| (Ek )k=1 is a sequence of disjoint Borel sets such that E = ∪k=1 Ek


k=1
(∞ )
X
∞ ∞
≤ sup µTF (Ek ) (Ek )k=1 is a sequence of disjoint Borel sets such that E = ∪k=1 Ek


k=1

= sup{µTF (E)}

8
Real Analysis Chapter 3 Solutions Jonathan Conder

= µTF (E).

In particular, if x ∈ R then G(x) = |µF |((−∞, x]) ≤ µTF ((−∞, x]) = TF (x). Therefore G = TF by part (a).
Since |µF | and µTF are finite, it is straightforward to show that M := {E ∈ BR | |µF |(E) = µTF (E)} is a σ-algebra
which contains (−∞, x] for all x ∈ R. Therefore BR ⊆ M and hence |µF |(E) = µTF (E) for all E ∈ BR .

29. The total variation of µF as a complex measure is the same as the total variation of µF as a signed measure, so

TF (x) + F (x) = µTF ((−∞, x]) + µF ((−∞, x]) = |µF |((−∞, x]) + µF ((−∞, x]) = 2µ+
F ((−∞, x])

and hence µP ((−∞, x]) = P (x) = 21 (TF (x) + F (x)) = µ+ +


F ((−∞, x]) for all x ∈ R. Therefore µP = µF , because

{E ∈ BR | µP (E) = µ+
F (E)} is a σ-algebra containing a set which generates BR . Similarly µN = µF .
P∞ −k
30. Let q : N → Q be a surjection and define f := k=1 2 χ(q(k),∞) . Given x ∈ R it is clear that
X
f (x) = 2−k
k∈N
q(k)<x

and hence f (x) ≤ f (y) for all y ∈ (x, ∞). Let ε ∈ (0, ∞) and choose N ∈ N such that 2−N < ε. There exists δ ∈ (0, ∞)
such that (x − δ, x) ∪ (x, x + δ) does not contain q(1), q(2), . . . , q(N ). If y ∈ (x − δ, x) then

X X ∞
X
f (x) ≥ f (y) = 2−k = f (x) − 2−k ≥ f (x) − 2−k > f (x) − ε,
k∈N k∈N k=N +1
q(k)<y y≤q(k)<x

which shows that limy↑x f (y) = f (x). On the other hand, if y ∈ (x, x + δ) then
X X
f (x) ≤ f (y) = 2−k = f (x) + 2−k ,
k∈N k∈N
q(k)<y x≤q(k)<y

which implies that limy↓x f (y) = f (x) provided that x ∈


/ Q. However, if x = q(n) for some n ∈ N then
X
f (x) + 2−n ≤ f (y) = f (x) + 2−n + 2−k < f (x) + 2−n + ε
k∈N
x<q(k)<y

and hence limy↓x f (y) = f (x) + 2−n . This shows that Q is the set of discontinuities of f.

31. (a) By the usual differentiation rules F and G are differentiable on R \ {0}. Moreover,

F (h) − F (0) h2 sin(h−1 )


F 0 (0) = lim = lim = lim h sin(h−1 ) = 0
h→0 h h→0 h h→0

by the squeeze theorem. Similarly G0 (0) = 0.


(b) If x ∈ R \ {0} then F 0 (x) = 2x sin(x−1 ) − cos(x−1 ) and hence |F 0 (x)| ≤ 2|x|| sin(x−1 )| + | cos(x−1 )| ≤ 2|x| + 1.
Therefore F 0 ≤ 3 on [−1, 1]. Hence, by the mean value theorem, if a, b ∈ [−1, 1] and a < b there exists c ∈ (a, b)
such that |F (b) − F (a)| = |F 0 (c)||b − a| ≤ 3(b − a). It follows that F ∈ BV ([−1, 1]) because
( n )
X
TF (1) − TF (−1) = sup |F (xk ) − F (xk−1 )| n ∈ N and x0 , x1 , . . . , xn ∈ R with − 1 = x0 < x1 < · · · < xn = 1


k=1

9
Real Analysis Chapter 3 Solutions Jonathan Conder

( n )
X
≤ sup 3(xk − xk−1 ) n ∈ N and x0 , x1 , . . . , xn ∈ R with − 1 = x0 < x1 < · · · < xn = 1


k=1

= sup{3xn − 3x0 | n ∈ N and x0 , x1 , . . . , xn ∈ R with − 1 = x0 < x1 < · · · < xn = 1}


= sup{3(1 − (−1))}
= 6.

For each k ∈ N ∪ {0} define xk := (π(k + 21 ))−1/2 ∈ [0, 1]. If n ∈ N, then


n n n n n n
X X X X X 1 X 1
|G(xk ) − G(xk−1 )| = |x2k (−1)k − x2k−1 (−1)k−1 | = 2 2
(xk + xk−1 ) ≥ 2
xk = ≥ .
k=1 k=1 k=1 k=1 k=1
π(k + 21 ) k=1 2πk
P∞ 1
Since k=1 k diverges, this implies that G ∈
/ BV ([−1, 1]).

33. Define an increasing, right continuous function G : R → R by



lim F (y), x < b
y↓x
G(x) :=
F (b), x ≥ b.

Then G(b) − G(a) ≤ F (b) − F (a) and G0 = F 0 almost everywhere on (a, b). Moreover, there exists a regular measure
µG on R such that µG ((a, b]) = G(b) − G(a) for all a, b ∈ R with a < b. Let dµG = dλ + f dm be the Lebesgue-Radon-
Nikodym representation of µG (note that λ ≥ 0 by the proof of Lebesgue-Radon-Nikodym), so that
G(x + h) − G(x) µG ((x, x + h])
lim = lim = f (x)
h↓0 h h↓0 m((x, x + h])

and
G(x + h) − G(x) G(x − h) − G(x) G(x) − G(x − h) µG ((x − h, x])
lim = lim = lim = lim = f (x)
h↑0 h h↓0 −h h↓0 h h↓0 m((x − h, x])

for almost all x ∈ R. In particular


Z Z b Z b
0
F (b) − F (a) ≥ G(b) − G(a) = µG ((a, b]) ≥ f dm = G dm = F 0 dm.
(a,b] a a

35. Since F and G are continuous on [a, b], there exists M ∈ (0, ∞) such that |F (x)| ≤ M and |G(x)| ≤ M for all
x ∈ [a, b]. Let ε ∈ (0, ∞) and choose δ ∈ (0, ∞) such that
n n
X ε X ε
|F (bk ) − F (ak )| < and |G(bk ) − G(ak )| < (2)
2M 2M
k=1 k=1
Pn
for every finite collection of disjoint intervals (a1 , b1 ), (a2 , b2 ), . . . , (an , bn ) ⊆ [a, b] with k=1 (bk − ak ) < δ. It follows
that F G is absolutely continuous on [a, b], because (2) implies that
n
X n
X
|F G(bk ) − F G(ak )| ≤ (|F (bk )G(bk ) − F (bk )G(ak )| + |F (bk )G(ak ) − F (ak )G(ak )|)
k=1 k=1
Xn n
X
= |F (bk )||G(bk ) − G(ak )| + |F (bk ) − F (ak )||G(ak )|
k=1 k=1
Xn n
X
≤ M |G(bk ) − G(ak )| + |F (bk ) − F (ak )|M
k=1 k=1

10
Real Analysis Chapter 3 Solutions Jonathan Conder

ε ε
<M + M
2M 2M
= ε.

Therefore, by the fundamental theorem of calculus for Lebesgue integrals, F G is differentiable almost everywhere
Rb
on [a, b], (F G)0 ∈ L1 ([a, b], m) and F (b)G(b) − F (a)G(a) = a (F G)0 dm. Since F and G are differentiable almost
everywhere on [a, b] (again by the fundamental theorem), it is clear that (F G)0 = F G0 + GF 0 almost everywhere on
Rb
[a, b] and hence F (b)G(b) − F (a)G(a) = a (F G0 + GF 0 ) dm.

37. Suppose F is Lipschitz continuous, and choose M ∈ (0, ∞) such that |F (y) − F (x)| ≤ M |y − x| for all x, y ∈ R. If
ε ∈ (0, ∞) and (a1 , b1 ), (a2 , b2 ), . . . , (an , bn ) is a finite collection of disjoint intervals with nk=1 (bk − ak ) < M
ε
P
then
n n m
X X X ε
|F (bk ) − F (ak )| ≤ M |bk − ak | = M (bk − ak ) < M = ε.
M
k=1 k=1 k=1

Therefore F is absolutely continuous, so it is differentiable almost everywhere. It follows that, for almost every x ∈ R,

0
F (y) − F (x) |F (y) − F (x)| M |y − x|
|F (x)| = lim
= lim ≤ lim = lim M = M.
y→x y−x y→x |y − x| y→x |y − x| y→x

Conversely, suppose F is absolutely continuous and there exists M ∈ [0, ∞) such that |F 0 | ≤ M almost everywhere.
Ry Ry Ry
If x, y ∈ R and x ≤ y then |F (y) − F (x)| = | x F 0 dm| ≤ x |F 0 | dm ≤ x M dm = M (y − x) = M |y − x| Therefore F
is Lipschitz continuous with Lipschitz constant M.

39. Set F0 := F, and for each k ∈ N ∪ {0} define a non-negative, increasing, right continuous function Gk : R → R by

 lim F (y), x < a
 y↓a k


Gk (x) := limy↓x Fk (y), x ∈ [a, b)


F (b), x ≥ b.

k

Fix k ∈ N ∪ {0}, and note that G0k = Fk0 almost everywhere on [a, b]. If x ∈ [a, b), there is a sequence (xn )∞ n=1 in
P∞
(x, b) which decreases to x. For each n ∈ N it is clear that k=1 Fk (xn )χ{k} is integrable with respect to the counting
measure ν on N; indeed
Z X∞ ∞
X
Fk (xn )χ{k} dν = Fk (xn ) = F (xn ) < ∞.
k=1 k=1
P∞
Since each Fk is increasing these functions are dominated by k=1 Fk (x1 )χ{k} , and by definition

X ∞
X
lim Fk (xn )χ{k} = Gk (x)χ{k}
n→∞
k=1 k=1

pointwise. Hence, by the dominated convergence theorem



Z X ∞
Z X ∞
X
G0 (x) = lim F (xn ) = lim Fk (xn )χ{k} dν = Gk (x)χ{k} = Gk (x).
n→∞ n→∞
k=1 k=1 k=1
P∞ P∞
Moreover, if x ∈ (−∞, a) then G0 (x) = G0 (a) = k=1 Gk (a) = k=1 Gk (x) and similarly, if x ∈ [b, ∞) then
P∞ P∞
G0 (x) = F0 (b) = k=1 Fk (b) = k=1 Gk (x). There exists an outer regular Radon measure µk on R such that

11
Real Analysis Chapter 3 Solutions Jonathan Conder

µk ((s, t]) = Gk (t) − Gk (s) for all s, t ∈ R with s < t. Let dµk = dλk + fk dm be the Lebesgue-Radon-Nikodym
representation of µk (note that λk ≥ 0 by the proof of Lebesgue-Radon-Nikodym), so that
Gk (x + h) − Gk (x) µk ((x, x + h])
lim = lim = fk (x)
h↓0 h h↓0 m((x, x + h])

and
Gk (x + h) − Gk (x) Gk (x − h) − Gk (x) Gk (x) − Gk (x − h) µk ((x − h, x])
lim = lim = lim = lim = fk (x)
h↑0 h h↓0 −h h↓0 h h↓0 m((x − h, x])

for almost all x ∈ R, and hence G0k = fk almost everywhere. By definition Fk0 ≥ 0 almost everywhere. Note that
∞ ∞ ∞
!
X X X
µk ((s, t]) = µk ((s, t]) = (Gk (s) − Gk (t)) = G0 (s) − G0 (t) = µ0 ((s, t])
k=1 k=1 k=1
P∞
for all s, t ∈ R with s < t. This implies that k=1 µk = µ0 , because both measures are outer regular and every
open subset of R is a disjoint union of countably many left-open, right-closed intervals. For each k ∈ N choose
P∞
Ak ∈ BR such that λk (Ak ) = 0 and m(Ack ) = 0. Then ∩∞ k=1 Ak is null with respect to the measure λk , while
∞ c
R k=1
P∞
its complement ∪k=1 Ak is m-null. Moreover, it is clear that the Borel measure ρ defined by ρ(E) := E k=1 fk dm
is absolutely continuous with respect to m. Therefore λ0 = ∞
P P∞
k=1 λk and f0 = k=1 fk almost everywhere, by the
0
P∞ P∞ 0
Lebesgue-Radon-Nikodym theorem. In particular, F = f0 = k=1 fk = k=1 Fk almost everywhere on [a, b].

40. By construction Fn is continuous and Fn (R) = [0, 1] for all n ∈ N. Therefore, if n ∈ N then
∞ ∞
n

X X X
−k −k
G − 2 Fk = 2 Fk ≤ 2−k = 2−n


k=1 k=n+1 k=n+1

and hence G is the uniform limit of a sequence of continuous functions. This implies that G is continuous. If x, y ∈ R
and x < y there exists n ∈ N such that [an , bn ] ⊆ (x, y), so that Fn (x) = 0 while Fn (y) = 1 and hence
X X X
G(x) = Fn (x) + 2−k Fk (x) < Fn (y) + 2−k Fk (x) ≤ Fn (y) + 2−k Fk (y) = G(y).
k∈N k∈N k∈N
k6=n k6=n k6=n

The complement of the Cantor set C is open, so if x ∈ (0, 1) \ C then F is constant on a neighbourhood of x,
implying that F 0 (x) = 0. Since m(C) = 0 and F is constant on (−∞, 0) and (1, ∞), this shows that F 0 = 0 almost
everywhere. If n ∈ N, the preimages of (0, 1) and (0, 1) \ C under the map x 7→ bx−a n
n −an
have measures bn − an and
P∞ k−1 −k 0
k=1 2 3 (bn − an ) = bn − an , which implies that Fn = 0 almost everywhere by the chain rule. Therefore, by the
previous exercise G0 = ∞ −k 0
P
k=1 2 Fk = 0 almost everywhere.

41. (a) If x, y ∈ [0, 1] and x < y then F (y) = m([0, y] ∩ A) = m([0, x] ∩ A) + m((x, y] ∩ A) > F (x), so F is strictly
increasing on [0, 1]. Let ε ∈ (0, ∞). If (a1 , b1 ), (a2 , b2 ), . . . , (an , bn ) is a finite collection of disjoint intervals such
that nk=1 (bk − ak ) < ε then
P

n
X n
X n
X n
X
|F (bk ) − F (ak )| = |m([0, bk ] ∩ A) − m([0, ak ] ∩ A)| = m((ak , bk ] ∩ A) ≤ m((ak , bk ]) < ε,
k=1 k=1 k=1 k=1

so F is absolutely continuous on R. Hence there exists an outer regular Radon measure µF on R such that
Rb
µF ((a, b]) = F (b) − F (a) = a F 0 dm for all a, b ∈ R with a < b. Moreover
Z b
F (b) − F (a) = m([0, b] ∩ A) − m([0, a] ∩ A) = m((a, b] ∩ A) = χA dm
a

12
Real Analysis Chapter 3 Solutions Jonathan Conder

for all a, b ∈ R with a < b. This implies that dµF = F 0 dm = χA dm, since the subset of BR on which these (finite)
measures agree is a σ-algebra containing a set which generates BR . By the Lebesgue-Radon-Nikodym theorem
F 0 = χA almost everywhere, so F 0 = 0 on [0, 1] \ A, which has positive measure because m([0, 1] ∩ A) < m([0, 1]).
(b) Let ε ∈ (0, ∞). If (a1 , b1 ), (a2 , b2 ), . . . , (an , bn ) is a finite collection of disjoint intervals such that nk=1 (bk −ak ) < ε
P

n
X n
X
|G(bk ) − G(ak )| = |m([0, bk ] ∩ A) − m([0, bk ] \ A) − m([0, ak ] ∩ A) + m([0, ak ] \ A)|
k=1 k=1
n
X
= |m((ak , bk ] ∩ A) − m((ak , bk ] \ A)|
k=1
n
X
≤ (m((ak , bk ] ∩ A) + m((ak , bk ] \ A))
k=1
Xn
= m((ak , bk ])
k=1

< ε,

so G is absolutely continuous on R. In particular G is differentiable on some E ⊆ R with m(E c ) = 0. Moreover


Z b
G0 dm = G(b) − G(a)
a
= m([0, b] ∩ A) − m([0, b] \ A) − m([0, a] ∩ A) + m([0, a] \ A)
= m((a, b] ∩ A) − m((a, b] \ A)
Z b Z b
= χA dm − χAc dm
a a
Z b
= (χA − χAc ) dm
a

for all a, b ∈ R such that a < b. This implies that G0 = (χA − χAc ) almost everywhere, by the Lebesgue-Radon-
Nikodym theorem applied to each of the measures G0 dm and (χA − χAc ) dm (which are equal because they are
finite and agree on a generating set of BR ). If a, b ∈ [0, 1], a < b and G is monotone on (a, b), then G0 ≥ 0 or
G0 ≤ 0 on (a, b) ∩ E (because the difference quotient at each point is either non-negative or non-positive). This
is a contradiction because χA − χAc takes on the values 1 and −1 on A and Ac respectively, both of which have
positive measure and hence G0 (x) = 1 and G0 (y) = −1 for some x, y ∈ (a, b) ∩ E.

42. (a) Suppose F is convex and let s, t, s0 , t0 ∈ (a, b) such that s ≤ s0 < t0 and s < t ≤ t0 . Then tt−s
0 −s ∈ (0, 1] and hence

       
t−s 0 t−s 0 t−s t−s 0 t−s
F (t) = F (t − s) + s = F t + 1− 0 s ≤ 0 F (t ) + 1 − 0 F (s),
t0 − s t0 − s t −s t −s t −s
−s 0 −s 0
because if t−s
= 1 then t = t0 . Moreover st0 −s
t0 −s ∈ [0, 1), and st0 −s = 0 iff s0 = s, so
 0  0
s0 − s s0 − s s0 − s
     
0 s −s 0 s −s 0 0
F (s ) = F (t − s) + s = F t + 1− 0 s ≤ 0 F (t ) + 1 − 0 F (s).
t0 − s t0 − s t −s t −s t −s
It follows that
t−s t−s
F (t) − F (s) ≤ F (t0 ) − 0 F (s)
t0 − s t −s

13
Real Analysis Chapter 3 Solutions Jonathan Conder

s0 − s −1 s0 − s
   
t−s 0 t−s 0 0
≤ 0 F (t ) − 0 1− 0 F (s ) − 0 F (t )
t −s t −s t −s t −s
s0 − s
 
t−s
= 0 F (t0 ) − (t − s)(t0 − s − (s0 − s))−1 F (s0 ) − 0 F (t0 )
t −s t −s
0
 
t−s t−s s −s
= 0 F (t0 ) − 0 0
F (s0 ) − 0 F (t0 )
t −s t −s t −s
t − s s0 − s
 
t−s t−s
= 0 + 0 F (t0 ) − 0 F (s0 )
t − s t − s0 t0 − s t − s0
t0 − s0 + s0 − s t−s
= (t − s) 0 0 0
F (t0 ) − 0 F (s0 )
(t − s)(t − s ) t − s0
t−s
= 0 (F (t0 ) − F (s0 ))
t − s0
and hence
F (t) − F (s) F (t0 ) − F (s0 )
≤ . (3)
t−s t0 − s0
Conversely, suppose that (3) holds for all s, t, s0 , t0 ∈ (a, b) with s ≤ s0 < t0 and s < t ≤ t0 . Let x, y ∈ (a, b) and
λ ∈ (0, 1). Also, set z := λy + (1 − λ)x. If x < y, then x < z < y so

F (z) − F (x) F (y) − F (z)


≤ ,
z−x y−z
which implies that
F (z) F (z) F (y) F (x)
+ ≤ +
z−x y−z y−z z−x
and hence
 −1  
1 1 F (y) F (x)
F (z) ≤ + +
z−x y−z y−z z−x
y − z + z − x −1 F (y)
   
F (x)
= +
(z − x)(y − z) y−z z−x
1
= ((z − x)F (y) + (y − z)F (x))
y−x
1
= ((λy − λx)F (y) + ((1 − λ)y − (1 − λ)x)F (x))
y−x
= λF (y) + (1 − λ)F (x).

We obtain the same result for the case x > y by swapping x with y and replacing λ with 1 − λ. Finally, if x = y
then F (z) = F (x) = λF (x) + (1 − λ)F (x) = λF (y) + (1 − λ)F (x). Therefore F is convex.
(b) Suppose that F is convex, and let [t, s0 ] ⊆ (a, b). There exists s ∈ (a, t) and t0 ∈ (s0 , b). Let ε ∈ (0, ∞) and set

F (t0 ) − F (s0 ) F (s) − F (t)


 
M := max , .
t0 − s0 t−s

If (a1 , b1 ), (a2 , b2 ), . . . , (an , bn ) ⊆ [t, s0 ] is a finite collection of disjoint intervals with nk=1 (bk − ak ) < M
ε
P
then

F (t) − F (s) F (bk ) − F (ak ) F (t0 ) − F (s0 )


−M ≤ ≤ ≤ ≤M
t−s bk − ak t0 − s0

14
Real Analysis Chapter 3 Solutions Jonathan Conder

for all k ∈ {1, 2, . . . , n} and hence


n n n
X X X ε
|F (bk ) − F (ak )| ≤ M |bk − ak | = M (bk − ak ) < M = ε.
M
k=1 k=1 k=1

Thus F is absolutely continuous on [t, s0 ], and differentiable on some E ⊆ (a, b). If x0 , y0 ∈ E and x0 < y0 , then

F (x) − F (x0 ) F (y) − F (y0 )



x − x0 y − y0
for all x ∈ (x0 , y0 ) and y ∈ (y0 , b), so

F (x) − F (x0 ) F (y) − F (y0 )


F 0 (x0 ) = lim ≤
x↓x0 x − x0 y − y0

for all y ∈ (y0 , b) and hence


F (y) − F (y0 )
F 0 (x0 ) ≤ lim = F 0 (y0 ).
y↓y0 y − y0
This implies that F0 is increasing on E.
Conversely, suppose that F is absolutely continuous and F 0 is increasing on the set E ⊆ (a, b) where F is
differentiable. Since F is absolutely continuous, m((a, b) \ E) = 0. Let x, y ∈ (a, b), λ ∈ (0, 1) and suppose that
x < y. Define z := λy + (1 − λ)x and T : [x, y] → [x, z] by T (t) := λt + (1 − λ)x. Note that F ◦ T is absolutely
continuous, because if (a1 , b1 ), (a2 , b2 ), . . . , (an , bn ) ⊆ [x, y] is a finite collection of disjoint intervals, then so is
(T (a1 ), T (b1 )), (T (a2 ), T (b2 )), . . . , (T (an ), T (bn )) ⊆ [x, z], and
n
X n
X n
X n
X
(T (bk ) − T (ak )) = (λbk − λak ) = λ (bk − ak ) < (bk − ak ).
k=1 k=1 k=1 k=1

Since T (t) ≤ λt + (1 − λ)t = t for all t ∈ [x, y] and F 0 is increasing on E, it follows that
Z y Z y Z y Z y
0 0 0 0
F (T (y))−F (T (x)) = (F ◦T ) dm = F (T (t))T (t) dt = λ F (T (t)) dt ≤ λ F 0 (t) dt = λ(F (y)−F (x)).
x x x x

Therefore

F (z) = F (T (y)) − F (T (x)) + F (x) ≤ λ(F (y) − F (x)) + F (x) = λF (y) + (1 − λ)F (x),

which implies that F is convex by the same argument used in part (a) for the cases x > y and x = y.
(c) Fix t0 ∈ (a, b) and let t1 , t2 , t3 , t4 ∈ (a, b) such that t1 ≤ t2 < t0 < t3 ≤ t4 . From part (a)

F (t0 ) − F (t1 ) F (t0 ) − F (t2 ) F (t3 ) − F (t0 ) F (t4 ) − F (t0 )


≤ ≤ ≤ ,
t0 − t1 t0 − t2 t3 − t0 t4 − t0
F (t0 )−F (t) F (t)−F (t0 )
which implies that limt↑t0 t0 −t and limt↓t0 t−t0 exist (by monotone convergence) and

F (t0 ) − F (t0 ) F (t0 ) − F (t) F (t) − F (t0 ) F (t00 ) − F (t0 )


≤ lim ≤ lim ≤
t0 − t0 t↑t0 t0 − t t↓t0 t − t0 t00 − t0

(by taking one limit at a time) for all t0 ∈ (a, t0 ) and t00 ∈ (t0 , b). Set β := limt↑t0 F (t0t0)−F
−t
(t)
and let t ∈ (a, b). If
t > t0 then F (t) − F (t0 ) ≥ β(t − t0 ), and if t = t0 then F (t) − F (t0 ) = 0 = β(t − t0 ). Otherwise t < t0 , in which
case F (t0 ) − F (t) ≤ β(t0 − t) and hence F (t) − F (t0 ) ≥ β(t − t0 ).

15
Real Analysis Chapter 3 Solutions Jonathan Conder

R R R R
(d) Since b − g > 0 everywhere, (b − g) dµ > 0 and hence g dµ < b dµ = b. Similarly a < g dµ, so there exists
R R
β ∈ R such that F (t) − F ( g dµ) ≥ β(t − g dµ) for all t ∈ (a, b). In particular
Z Z  Z   Z 
F ◦ g dµ ≥ F g dµ + β g − g dµ dµ
Z Z Z Z Z
=F g dµ 1 dµ + β g dµ − β g dµ 1 dµ
Z  Z Z
=F g dµ + β g dµ − β g dµ
Z 
=F g dµ .

16

You might also like