You are on page 1of 24

Real Analysis Chapter 2 Solutions Jonathan Conder

1. Suppose f is measurable. Then f −1 ({−∞}) ∈ M and f −1 ({∞}) ∈ M, because {−∞} and {∞} are Borel sets. If
B ⊆ R is Borel then f −1 (B) ∈ M, and hence f −1 (B) ∩ Y ∈ M (since R is also Borel). Thus f is measurable on Y.
Conversely, suppose that f −1 ({−∞}) ∈ M, f −1 ({∞}) ∈ M and f is measurable on Y. Let B ⊆ R be Borel. Then
f −1 (B) ∩ Y ∈ M, and f −1 (B) = (f −1 (B) ∩ Y ) ∪ (f −1 (B) \ Y ). Clearly f −1 (B) \ Y = f −1 (B ∩ {−∞, ∞}), which is
measurable because it is either ∅, f −1 ({−∞}), f −1 ({−∞}) or f −1 ({−∞}) ∪ f −1 ({−∞}). This implies that f −1 (B) ∈
M, so f is measurable.

3. If fn : X → R for all n ∈ N, then g := lim inf n→∞ fn and h := lim supn→∞ fn are measurable. Therefore f := h − g is
measurable (setting f (x) = 1 whenever g(x) = h(x) ∈ {−∞, ∞}), and hence

/ {−∞, ∞}} = f −1 ({0}) = f −1 ([−∞, 0]) ∈ M.


{x ∈ X | lim fn (x) exists} = {x ∈ X | lim inf fn (x) = lim sup fn (x) ∈
n→∞ n→∞ n→∞

If fn : X → C for all n ∈ N then (fn )n∈N converges at x ∈ X iff (Re(fn ))n∈N and (Im(fn ))n∈N converge at x. So

{x ∈ X | lim fn (x) exists} = {x ∈ X | lim Re(fn (x)) exists} ∩ {x ∈ X | lim Im(fn (x)) exists} ∈ M
n→∞ n→∞ n→∞

4. If a ∈ R then there is a sequence (an )n∈N in Q ∩ (a, ∞) converging to a, and f −1 ((a, ∞]) = ∪n∈N f −1 ((an , ∞]) ∈ M.
Since BR is generated by such intervals (a, ∞], it follows that f is measurable.

5. Suppose that f is measurable, and let E be a measurable set from the codomain of f. Then f −1 (E) ∈ M, so
f −1 (E) ∩ A, f −1 (E) ∩ B ∈ M. Therefore f is measurable on A and on B.
Conversely, suppose that f is measurable on A and on B. Again let E be a measurable set from the codomain of f.
Then f −1 (E) ∩ A, f −1 (E) ∩ B ∈ M, so f −1 (E) = (f −1 (E) ∩ A) ∪ (f −1 (E) ∩ B) ∈ M and f is measurable.

6. For example set X := R and M := L. There exists a non-measurable set A ⊆ X, and for each a ∈ A the set {a} is
measurable. Hence {χ{a} }a∈A is a family of measurable functions, but its supremum is χA , which is not measurable
because χ−1
A ([1, ∞]) = A.

8. Since f is measurable iff −f is measurable, we may assume that f is increasing. Let a ∈ R and x ∈ f −1 ([a, ∞)). If
y ∈ [x, ∞) then f (y) ≥ f (x) ≥ a and hence y ∈ f −1 ([a, ∞)). This shows that f −1 ([a, ∞)) is an interval, so it is Borel
measurable and hence f is Borel measurable.

9. (a) If x, y ∈ [0, 1] and x < y, then g(x) = f (x) + x ≤ f (y) + x < f (y) + y = g(y) and hence g is injective. Since
g(0) = f (0) = 0 and g(1) = f (1) + 1 = 2, the intermediate value theorem implies that g maps [0, 1] onto [0, 2].
Let x ∈ [0, 2] and ε ∈ (0, ∞). Then there exists x0 ∈ [0, 1] such that g(x0 ) = x. Define x1 := max{x0 − 21 ε, 0}
and x2 := min{x0 + 12 ε, 1}. Then g(x1 ) ≤ x ≤ g(x2 ), and at least one of the inequalities is strict. Define
δ := min({x − g(x1 ), g(x2 ) − x} \ {0}). Given y ∈ [0, 2] and |y − x| < δ, it is straightforward to check that
g(x1 ) ≤ y ≤ g(x2 ). Indeed, if x = g(x1 ) then g(x1 ) = g(0) = 0, and otherwise g(x1 ) ≤ x − δ. Similarly x2 = 1 or
x + δ ≤ g(x2 ). Since g is increasing, it is clear that x1 ≤ h(y) ≤ x2 . This implies that x0 − 21 ε ≤ h(y) ≤ x0 + 12 ε,
so |h(y) − h(x)| = |h(y) − x0 | < ε. Therefore h is continuous on [0, 2].
(b) Note that C = { n∈N 3−n an | (an )n∈N is a sequence in {0, 2}}, which implies that
P

( ! )
X X
−n −n
g(C) = f 3 an + 3 an (an )n∈N is a sequence in {0, 2}


n∈N n∈N

1
Real Analysis Chapter 2 Solutions Jonathan Conder

( )
X a
−n n
X
−n

= 2 + 3 an (an )n∈N is a sequence in {0, 2}

2
n∈N n∈N
( )
X
= (2−n−1 + 3−n )an (an )n∈N is a sequence in {0, 2} .


n∈N

Set C0 := [0, 2], and for each n ∈ N construct Cn from Cn−1 by removing an open interval of length 3−n from
the middle of each interval comprising Cn . This works because Cn−1 is the union of 2n−1 intervals of length
21−n + 31−n > 3−n (indeed, 20 + 30 = 2 and 21 (21−n + 31−n − 3n ) = 2−n + 3−n ). Set C 0 := ∩n∈N Cn , so that

2
m(C 0 ) = lim m(Cn ) = lim 2n (2−n + 3−n ) = 1 + lim ( )n = 1.
n→∞ n→∞ n→∞ 3

Let x ∈ g(C) and N ∈ N. There exists a sequence (an )n∈N in {0, 2} such that x = n∈N (2−n−1 + 3−n )an . Clearly
P

N ∞
X X 2−N −1 3−N −1
0≤x− (2−n−1 + 3−n )an ≤ 2(2−n−1 + 3−n ) = + 2 = 2−N + 3−N .
1 − 2−1 1 − 3−1
n=1 n=N +1

By induction on N it can be shown that N −n−1 + 3−n )a is the left endpoint of an interval from C ,
P
n=1 (2 n N
th
because the N term in the series is either 0 or 2 −N −N
+ 2 · 3 , the latter of which is the sum of length of the
intervals in CN and the length of the gaps between them. The above calculation therefore implies that x ∈ CN .
It follows that x ∈ C 0 , which shows that g(C) ⊆ C 0 .
Conversely, let x ∈ C 0 , so that x ∈ Cn for all n ∈ N. For each n ∈ N define an ∈ {0, 2} depending on whether
the interval x belongs to in Cn is the left or right child of its parent in Cn−1 . Then x and N −n−1 + 3−n )a
P
n=1 (2 n
are from the same interval in CN , for all N ∈ N. In particular

N
X
lim x − (2−n−1 + 3−n )an ≤ lim (2−N + 3−N ) = 0,

N →∞ N →∞
n=1
P∞ −n−1
which implies that x = n=1 (2 + 3−n )an ∈ g(C). Therefore C 0 ⊆ g(C), and hence m(g(C)) = m(C 0 ) = 1.
(c) Since A ⊆ g(C), it is clear that B ⊆ C. Therefore m∗ (B) ≤ m∗ (C) = 0, so B is Lebesgue measurable. If B was
Borel measurable, then h−1 (B) would be as well, because h is continuous. However h−1 (B) = A, which is not
Borel. Hence B is not Borel measurable.
(d) Set F := χB and G := h. Then F is Lebesgue measurable because B ∈ L, and G is continuous by part (a). But

(F ◦ G)−1 ([1, ∞)) = {x ∈ [0, 2] | χB (h(x)) ∈ [1, ∞)} = {x ∈ [0, 2] | h(x) ∈ B} = h−1 (B) = A ∈
/ L,

so F ◦ G is not Lebesgue measurable.

11. If n ∈ N and i ∈ Z then fn is clearly Borel measurable on [ai , ai+1 ], because fn |[ai ,ai+1 ] is the sum of products of Borel
measurable functions. By an obvious generalisation of exercise 5, it follows that each fn is Borel measurable. Let
(x, y) ∈ R × Rk and ε ∈ (0, ∞). Then there exists δ ∈ (0, ∞) such that |f (x0 , y) − f (x, y)| < ε for all x0 ∈ (x − δ, x + δ).
Moreover, there exists N ∈ N such that N1 < δ. Let n ∈ N with n ≥ N, and choose i ∈ Z so that x ∈ [ai , ai+1 ]. Since
1
n < δ it is clear that ai , ai+1 ∈ (x − δ, x + δ). Therefore

f (ai+1 , y)(x − ai ) − f (ai , y)(x − ai+1 ) − f (x, y)(ai+1 − ai )
|fn (x, y) − f (x, y)| =

ai+1 − ai

2
Real Analysis Chapter 2 Solutions Jonathan Conder


f (ai+1 , y)(x − ai ) − f (ai , y)(x − ai+1 ) − f (x, y)(x − ai ) + f (x, y)(x − ai+1 )
=
ai+1 − ai

(f (ai+1 , y) − f (x, y))(x − ai ) − (f (ai , y) − f (x, y))(x − ai+1 )
=
ai+1 − ai

(f (ai+1 , y) − f (x, y))(x − ai ) (f (ai , y) − f (x, y))(x − ai+1 )
≤ +
ai+1 − ai ai+1 − ai
x − ai ai+1 − x
= |f (ai+1 , y) − f (x, y)| · + |f (ai , y) − f (x, y)| ·
ai+1 − ai ai+1 − ai
x − ai ai+1 − x
≤ε· +ε·
ai+1 − ai ai+1 − ai
= ε.

This implies that (fn )n∈N converges to f pointwise, so f is Borel measurable. Clearly every function on R that is
continuous in each variable is Borel measurable. Let k ∈ N, and suppose that every function on Rk that is continuous
in each variable is Borel measurable. Also let g be a function on Rk+1 that is continuous in each variable. Then g(x, ·)
is a function on Rk that is continuous in each variable, and hence Borel measurable, for each x ∈ R. From above,
it follows that g is Borel measurable. By induction, for each k ∈ N every function on Rk that is continuous in each
variable is Borel measurable.

13. Let E ∈ M. By Fatou’s lemma


Z Z Z Z Z
f = f χE = lim inf fn χE ≤ lim inf fn χE = lim inf fn
E n→∞ n→∞ n→∞ E
R R
and similarly E c f ≤ lim inf n→∞ E c fn . But f χE + f χE c = f and fn χE + fn χE c = fn for all n ∈ N, which implies
R R R R R R
that E c f = f − E f and (for sufficiently large n ∈ N) E c fn = fn − E fn . Therefore
Z Z Z Z Z Z  Z Z
f− f= f ≤ lim inf fn = lim inf fn − fn = f − lim sup fn ,
E Ec n→∞ Ec n→∞ E n→∞ E
R R R R R
so lim supn→∞ E fn ≤ E f ≤ lim inf n→∞ E fn and hence E f = limn→∞ E fn .
Define F := (−∞, 0) and for each n ∈ N set Fn := F ∪ [n, n + 1). Then χF and each χFn are in L+ , the sequence
R R R R
(χFn )n∈N converges to χF pointwise and χF = ∞ = limn→∞ χFn . However, [0,∞) χF = 0 6= 1 = limn→∞ [0,∞) χFn .

14. Clearly λ(E) ≥ 0 for all E ∈ M. Moreover, λ(∅) = ∅ f dµ = f χ∅ dµ = 0 dµ = 0. If {En }n∈N is a pairwise disjoint
R R R

subcollection of M then (f χ∪N En )N ∈N is a sequence of measurable functions increasing to f χ∪n∈N En , so


n=1
Z
λ(∪n∈N En ) = f dµ
∪n∈N En
Z
= f χ∪n∈N En dµ
Z
= lim f χ∪N En dµ
N →∞ n=1

Z N
X
= lim f χEn dµ
N →∞
n=1
N Z
X
= lim f χEn dµ
N →∞
n=1

3
Real Analysis Chapter 2 Solutions Jonathan Conder

N Z
X
= lim f dµ
N →∞
n=1 En
∞ Z
X
= f dµ
n=1 En
X∞
= λ(En )
n=1

by the monotone convergence theorem. Therefore λ is a measure. Now let g ∈ L+ . If g is simple with standard
representation N
P
n=1 an χEn , then
Z N
X N
X Z N
X Z N
Z X Z
g dλ = an λ(En ) = an f dµ = an f χEn dµ = an f χEn dµ = f g dµ.
n=1 n=1 En n=1 n=1

Otherwise, there exists an increasing sequence (gn )n∈N of simple functions in L+ which converges pointwise to g, so
that (f gn )n∈N increases pointwise to f g and hence
Z Z Z Z
g dλ = lim gn dλ = lim f gn dµ = f g dµ,
n→∞ n→∞

by two applications of the monotone convergence theorem.

15. Since f1 < ∞, the functions {fn }n∈N and f can be adjusted on a set of measure zero (namely f1−1 ({∞})) so that
R

they map into [0, ∞). This does not affect their integrals. Clearly (f1 − fn )n∈N increases pointwise to f1 − f. Moreover
f1 − fn ∈ L+ for all n ∈ N. By the monotone convergence theorem (f1 − f ) = limn→∞ (f1 − fn ). Therefore
R R
Z Z Z Z
f = f + (f1 − f ) − (f1 − f )
Z Z
= f1 − lim (f1 − fn )
n→∞
Z Z 
= lim f1 − (f1 − fn )
n→∞
Z Z Z 
= lim fn + (f1 − fn ) − (f1 − fn )
n→∞
Z
= lim fn ,
n→∞
R R R
since (f1 − f ) ≤ f1 < ∞, and similarly (f1 − fn ) < ∞ for all n ∈ N.

16. For each n ∈ N define En := {x ∈ X | f (x) > n−1 }. Clearly (f χEn )n∈N increases pointwise to f, so by the monotone
R R
convergence theorem ( En f )n∈N increases to f. In particular, given ε ∈ (0, ∞) there exists n ∈ N such that
R R R
En f > f − ε. Since f < ∞ it is clear that µ(En ) < ∞.

17. Let (fn )n∈N be an increasing sequence in L+ , and set f := limn→∞ f. Then f ∈ L+ , and by Fatou’s lemma
Z Z Z
f = lim inf fn ≤ lim inf fn .
n→∞ n→∞
R R R R
Since fn ≤ f and hence fn ≤ f for all n ∈ N, it is clear that lim supn→∞ fn ≤ f. Therefore
Z Z Z
lim sup fn = lim inf fn = lim fn ,
n→∞ n→∞ n→∞
R R
so f = limn→∞ fn .

4
Real Analysis Chapter 2 Solutions Jonathan Conder

18. Let g ∈ L+ ∩ L1 , and (fn : X → R)n∈N be a sequence of measurable functions such that fn ≥ −g for all n ∈ N. Define
h := lim inf n→∞ fn . Clearly h ≥ −g, so h− (x) = max{−h(x), 0} ≤ g(x) for all x ∈ X. It follows that h− ∈ L1 and
g − h− ∈ L+ . Similarly fn− ∈ L1 and g − fn− , fn + g ∈ L+ for all n ∈ N. Therefore, by Fatou’s lemma
Z Z Z Z Z
+ −
h+ g = h − h + g
Z Z
= h + (g − h− )
+

Z
= (h + g)
Z
= lim inf (fn + g)
n→∞
Z
≤ lim inf (fn + g)
n→∞
Z Z 
= lim inf fn+ + (g − fn− )
n→∞
Z Z Z 
+ −
= lim inf fn + g − fn
n→∞
Z Z
= lim inf fn + g.
n→∞
R R R R
Since g < ∞, it follows that lim inf n→∞ fn = h ≤ lim inf n→∞ fn .
Let (fn : X → R)n∈N be a sequence of nonpositive measurable functions. Define h := lim supn→∞ fn . Then h ≤ 0 and
(−fn )n∈N is a sequence in L+ , so by Fatou’s lemma
Z Z Z Z Z Z Z Z Z
− + −
− h = h − h = −h = lim inf −fn ≤ lim inf −fn = lim inf fn = lim inf − fn = − lim sup fn .
n→∞ n→∞ n→∞ n→∞ n→∞
R R R
Therefore lim supn→∞ fn ≤ h= lim supn→∞ fn .

19. (a) There exists N ∈ N such that |f (x) − fn (x)| ≤ 1 for all x ∈ X and n ∈ N with n ≥ N. In particular

|f | = |f − fN + fN | ≤ |f − fN | + |fN | ≤ 1 + |fN |,

and hence |f | dµ ≤ 1+|fN | dµ = µ(X)+ |fN | dµ < ∞. This implies that f ∈ L1 (µ). Similarly 1+|f | ∈ L1 (µ).
R R R

Since |fn | ≤ 1 + |f | for all n ∈ N with n ≥ N, the dominated convergence theorem implies that
Z Z Z Z
lim fn dµ = lim fN +n dµ = lim fN +n dµ = f dµ.
n→∞ n→∞ n→∞

(b) For each n ∈ N define fn := 2−n χ[−2n ,2n ] . Clearly (fn )n∈N converges uniformly to 0, but for each n ∈ N
Z Z
−n n n
fn dµ = 2 µ([−2 , 2 ]) = 2 6= 0 = 0 dµ,

which implies that fn ∈ L1 (µ) (where µ is the Lebesgue measure) and limn→∞ fn dµ 6= 0 dµ.
R R

R R R R
20. It suffices to show that limn→∞ Re(fn ) = Re(f ) and limn→∞ Im(fn ) = Im(f ). Since limn→∞ Re(fn ) = Re(f )
and limn→∞ Im(fn ) = Im(f ) pointwise almost everywhere, while | Re(fn )| ≤ |fn | and | Im(fn )| ≤ |fn | for all n ∈ N,
we may assume without loss of generality that f and each fn are real-valued. If N, n ∈ N such that n ≥ N, then
Z Z ∞ Z Z Z ∞ Z
inf gm + fm ≤ gn + fn ≤ sup gm + fn ,
m=N m=N

5
Real Analysis Chapter 2 Solutions Jonathan Conder

which implies that


Z Z ∞  Z ∞ Z ∞ Z ∞ Z ∞
inf gm + fm ≤ inf sup gm + fn = sup gm + inf fm .
m=N m=N n=N m=N m=N

Therefore Z Z  Z Z Z Z
lim inf gn + fn ≤ lim sup gn + lim inf fn = g + lim inf fn .
n→∞ n→∞ n→∞ n→∞

Similarly Z Z  Z Z Z Z
lim inf gn − fn ≤ g + lim inf − fn = g − lim sup fn .
n→∞ n→∞ n→∞
Since gn + fn , gn − fn ∈ L+ for all n ∈ N, Fatou’s lemma implies that
Z Z Z Z Z Z Z  Z Z
g + f = (g + f ) = lim inf (gn + fn ) ≤ lim inf (gn + fn ) = lim inf gn + fn ≤ g + lim inf fn
n→∞ n→∞ n→∞ n→∞

and
Z Z Z Z Z Z Z  Z Z
g− f= (g − f ) = lim inf (gn − fn ) ≤ lim inf (gn − fn ) = lim inf gn − fn ≤ g − lim sup fn .
n→∞ n→∞ n→∞ n→∞
R R R R R R
Since g < ∞, it follows that lim supn→∞ fn ≤ f ≤ lim inf n→∞ fn , and hence f = limn→∞ fn .
R
21. Suppose that limn→∞ |fn − f | = 0. For every ε ∈ (0, ∞), there exists N ∈ N such that
Z Z Z Z Z Z

|fn | − |f | = (|fn | − |f |) ≤ |fn | − |f | ≤ |fn − f | = |fn − f | − 0 < ε

R R
for all n ∈ N with n ≥ N (by the reverse triangle inequality). Therefore limn→∞ |fn | = |f |.
Conversely, suppose that limn→∞ |fn | = |f |. For each n ∈ N it is clear that |fn | + |f | ∈ L1 and |fn − f | ≤ |fn | + |f |,
R R

so that |fn − f | ∈ L1 . Moreover (|fn − f |)n∈N converges to 0 ∈ L1 pointwise almost everywhere. Also (|fn | + |f |)n∈N
converges to 2|f | ∈ L1 pointwise almost everywhere, and
Z Z Z Z Z
lim (|fn | + |f |) = lim |fn | + |f | = 2 |f | = 2|f |.
n→∞ n→∞
R R
Therefore, by the previous exercise, limn→∞ |fn − f | = 0 = 0.

23. (a) Let x ∈ [a, b], and suppose that H(x) = h(x). Fix ε ∈ (0, ∞). Since

lim (sup f ([a, b] ∩ [x − δ, x + δ]) − inf f ([a, b] ∩ [x − δ, x + δ])) = H(x) − h(x) = 0,


δ→0+

there exists η ∈ (0, ∞) such that | sup f ([a, b] ∩ [x − δ, x + δ]) − inf f ([a, b] ∩ [x − δ, x + δ])| < ε for all δ ∈ (0, η),
in particular for δ := η2 . If y ∈ [a, b] and |x − y| < δ then y ∈ [a, b] ∩ [x − δ, x + δ], so

f (x) − f (y) ≤ sup f ([a, b] ∩ [x − δ, x + δ]) − inf f ([a, b] ∩ [x − δ, x + δ]) < ε

and similarly f (y) − f (x) < ε, so |f (x) − f (y)| < ε. This shows that f is continuous at x.
ε
Conversely, suppose that f is continuous at x. Let ε ∈ (0, ∞). There exists η ∈ (0, ∞) such that |f (x) − f (y)| < 3
for all y ∈ [a, b] with |x − y| < η. Therefore f (x) − 3ε < f (y) < f (x) + 3ε for all y ∈ [a, b] ∩ (x − η, x + η), so
ε ε
sup f ([a, b] ∩ [x − δ, x + δ]) − inf f ([a, b] ∩ [x − δ, x + δ]) ≤ f (x) + − f (x) + < ε
3 3
for all δ ∈ (0, η). This shows that

H(x) − h(x) = lim (sup f ([a, b] ∩ [x − δ, x + δ]) − inf f ([a, b] ∩ [x − δ, x + δ])) = 0.


δ→0+

6
Real Analysis Chapter 2 Solutions Jonathan Conder

b
(b) Choose a nested sequence (Pn )n∈N of partitions of [a, b] such that (SPn f )n∈N converges to I a (f ). For each n ∈ N
let En be the set of endpoints of the intervals comprising Pn , so that m(En ) = 0. Define E := ∪n∈N En , so that
m(E) = 0. Let x ∈ [a, b] \ E, and choose δ ∈ (0, ∞) such that
ε
sup f ([a, b] ∩ [x − δ, x + δ]) < H(x) + .
2
b
There exists N ∈ N such that SPn f < I a (f ) + εδ 2 for all n ∈ N with n ≥ N. Fix n ∈ N with n ≥ N. There is an
0 0 0 0 / En ). If [a0 , b0 ] ⊆ [x − δ, x + δ] then
interval [a , b ] in Pn such that x ∈ (a , b ) (because x ∈
ε
GPn (x) = sup f ([a0 , b0 ]) ≤ sup f ([a, b] ∩ [x − δ, x + δ]) < H(x) + < H(x) + ε.
2
Otherwise a0 < x − δ or x + δ < b0 , so that [x − δ, x] or [x, x + δ] is contained in (a0 , b0 ). Construct a new partition
Pn0 of [a, b] from Pn by inserting x and one of x − δ or x + δ between a0 and b0 . In the former case

SPn0 f − SPn f = sup f ([a0 , x − δ])(x − δ − a0 ) + sup f ([x − δ, x])δ + sup f ([x, b0 ])(b0 − x) − GPn (x)(b0 − a0 )
 ε
< GPn (x)(x − δ − a0 ) + H(x) + δ + GPn (x)(b0 − x) − GPn (x)(b0 − a0 )
2  ε
= GPn (x)(x − δ − a0 + b0 − x − b0 + a0 ) + H(x) + δ
2
 ε 
= GPn (x)(−δ) + H(x) + δ
2
 ε 
= H(x) − GPn (x) + δ,
2
which still holds for the latter case, by a similar calculation. It follows that
 
1 ε 1 b εδ ε
GPn (x) < (SPn f − SPn0 f ) + H(x) + < I a (f ) + − SPn0 f + H(x) + ≤ H(x) + ε.
δ 2 δ 2 2

Since x ∈ (a0 , b0 ), there exists η ∈ (0, ∞) such that [x − η, x + η] ⊆ (a0 , b0 ). This implies that

H(x) = inf sup f ([a, b] ∩ [x − ζ, x + ζ]) ≤ sup f ([a, b] ∩ [x − η, x + η]) ≤ sup f ([a0 , b0 ]) = GPn (x).
ζ∈(0,∞)

Therefore |GPn (x) − H(x)| < ε, so (GPn (x))n∈N converges to H(x) and hence (GPn )n∈N converges to H pointwise
almost everywhere. Since f is bounded and m([a, b]) < ∞, the dominated convergence theorem implies that
Z Z
b
H dm = lim GPn dm = lim SPn f = I a (f ).
[a,b] n→∞ n→∞

A similar argument implies that (gPn )n∈N converges to h pointwise almost everywhere, for all nested sequences
(Pn )n∈N of partitions of [a, b] such that (sPn f )n∈N converges to I ba (f ). Therefore [a,b] h dm = I ba (f ).
R

25. (a) By the monotone convergence theorem and Theorem 2.28,


Z Z 1 1
f = lim x−1/2 dx = lim 2x1/2 = lim (2 − 2n−1/2 ) = 2. (1)

n→∞ 1/n n→∞ 1/n n→∞

R P∞ −n R P∞ −n R
Therefore |g| = n=1 2 f (x − rn ) dx = n=1 2 f = 2, by the monotone convergence theorem. It
1
follows that g ∈ L (m), and g < ∞ almost everywhere by Proposition 2.20.

7
Real Analysis Chapter 2 Solutions Jonathan Conder

(b) Let E ⊆ R be a null set and suppose that h ∈ L1 (m) is equal to g on E c . If I ⊆ R is an interval with at least
two points, there exists n ∈ N such that rn is an interior point of I. For each k ∈ N note that (rn , rn + k −1 ) ∩ I
has positive measure, so there exists xk ∈ ((rn , rn + k −1 ) ∩ I) \ E. Clearly limk→∞ xk = rn , in which case
limk→∞ 2−n f (xk − rn ) = 2−n limk→∞ (xk − rn )−1/2 = ∞. But 2−n f (xk − rn ) ≤ g(xk ) = h(xk ) for all k ∈ N, which
implies that h is unbounded on I. This shows that h is unbounded on every interval, so it is clearly everywhere
discontinuous.
(c) By part (a) g 2 < ∞ almost everywhere. If I ⊆ R is an interval with at least two points, there exists n ∈ N such
that rn is an interior point of I. There exists δ ∈ (0, 1) such that (rn , rn + δ) ⊆ I, and
Z Z rn +δ Z rn +δ Z δ Z δ
g2 ≥ g2 ≥ 2−2n f (x − rn )2 dx = 2−2n f 2 = 2−2n x−1 dx = ∞,
I rn rn 0 0

where the last step follows from an argument similar to (1) (and is even in the undergraduate calculus textbooks).

26. Let x ∈ R and ε ∈ (0, ∞). For each n ∈ N define fn := |f |χ[x−2−n ,x+2−n ] , so that (fn )n∈N is a sequence in L1 (m)
which is dominated by |f | ∈ L1 (m). Moreover (fn )n∈N converges to 0 pointwise almost everywhere Therefore
Z Z
lim fn = 0 = 0,
n→∞

by the dominated convergence theorem. Choose n ∈ N such that fn < ε, and let y ∈ (x − 2−n , x + 2−n ). Then
R
Z x Z y

|F (x) − F (y)| =
f (t) dt − f (t) dt
Z−∞ Z
−∞


= f χ(−∞,x] − f χ(−∞,y]

Z

= f · (χ(−∞,x] − χ(−∞,y] )

Z
≤ |f | · |χ(−∞,x] − χ(−∞,y] |
Z
= |f | · χ[min{x,y},max{x,y}]
Z
≤ |f | · χ[x−2−n ,x+2−n ]
Z
= fn
Z

= fn

< ε.

This shows that then F is continuous at x, and hence F is continuous on R.

28. (a) Fix x ∈ [0, ∞) and define f : [1, ∞) → (0, 1] by f (n) := (1 + nx )−n . Then
 x  n(−xn−2 ) x  x
(log ◦f )0 (n) = − log 1 + − = − log 1 +
n 1 + nx n+x n
for all n ∈ [1, ∞). Note that
  X ∞  k ∞  k x x
x 1 x X x x
exp = ≤1+ = 1 + n+xx = 1 + n+x
n =1+
n+x k! n + x n+x 1 − n+x n+x n
k=0 k=1

8
Real Analysis Chapter 2 Solutions Jonathan Conder

x
≤ log 1 + nx for all n ∈ [1, ∞), so log ◦f is a decreasing function. Therefore f = exp ◦ log ◦f

and hence n+x
is decreasing, so the sequence (fn )n∈N is decreasing as well, where each fn : [0, ∞) → (0, 1] is defined by
fn (x) := (1 + nx )−n . In particular fn ≤ f2 for all n ∈ N with n ≥ 2. By the monotone convergence theorem
Z ∞ Z ∞ Z n Z n   
x −2 d x −1
f2 = lim f2 χ[0,n] = lim 1+ dx = lim −2 1 + dx,
0 n→∞ 0 n→∞ 0 2 n→∞ 0 dx 2
so by the fundamental theorem of calculus
Z ∞  
−1
 n −1 2
f2 = lim 2(1 + 0) − 2 1 + = 2 − lim n = 2.
0 n→∞ 2 n→∞ 1+ 2

Therefore f2 ∈ L1 , so (since | sin | ≤ 1) by the dominated convergence theorem


Z ∞ Z ∞ Z ∞
x −n x  x −n x
lim 1+ sin dx = lim 1 + sin dx = 0 dx = 0.
n→∞ 0 n n 0 n→∞ n n 0

Indeed, if x ∈ [0, ∞) then limn→∞ sin( nx ) = sin(limn→∞ nx ) = sin(0) = 0, and


 x −n    x 
lim 1+ = exp lim −n log 1 +
n→∞ n n→∞ n !!
x

log 1 + n
= exp − lim 1
n→∞
n
!!
x

log 1 + n − log(1)
= exp −x lim x
n→∞
n

= exp −x log0 (1)




= e−x .

(b) If x ∈ [0, 1] and n ∈ N with n ≥ 1 then 0 ≤ (1 + nx2 )(1 + x2 )−n ≤ 1 because


n   n  
2 n
X n 2 k 2
X n
(1 + x ) = (x ) = 1 + nx + (x2 )k ≥ 1 + nx2 .
k k
k=0 k=2
R1
Moreover 0 1 dx = 1. Hence, by the dominated convergence theorem
Z 1 Z 1 Z 1
2 2 −n 2 2 −n
lim (1 + nx )(1 + x ) dx = lim (1 + nx )(1 + x ) dx = 0 dx = 0.
n→∞ 0 0 n→∞ 0
n
Indeed, (1 + nx2 )(1 + x2 )−n ≤ (1 + nx2 )(1 + nx2 + x4 )−1 for all x ∈ [0, 1] and n ∈ N with n ≥ 2, and

2

1 + nx2 1 + nx2 n−2 + n−1 x2 0+0


lim n = lim = lim n−1 = =0
0 + 0 + 12 x4

2
n→∞ 1 + nx + 4 n(n−1) −2 −1 2 4
2 x x4 n→∞ n + n x + 2n x
n→∞ 1 + nx2 +
2

for all x ∈ (0, 1] (and hence almost all x ∈ [0, 1]).


(c) Define f : [0, ∞) → R by f (x) := sin(x) − x. Since f (0) = 0 and f 0 (x) = cos(x) − 1 ≤ 0 for all x ∈ [0, ∞),
this function is nonpositive and hence sin(x) ≤ x for all x ∈ [0, ∞). Moreover sin(x) ≥ 0 for all x ∈ [0, 1], so
−x ≤ sin(x) for all x ∈ [0, ∞). By the monotone convergence theorem and the fundamental theorem of calculus,
Z ∞ Z ∞ Z n
2 −1 2 −1 d π
(1+x ) dx = lim (1+x ) χ[0,n] (x) dx = lim tan−1 (x) dx = lim (tan−1 (n)−tan−1 (0)) = .
0 n→∞ 0 n→∞ 0 dx n→∞ 2

9
Real Analysis Chapter 2 Solutions Jonathan Conder

Since | sin( nx )/ nx | ≤ 1 for all x ∈ [0, ∞) and n ∈ N, the dominated convergence theorem implies that
∞ ∞ sin( nx )
Z x Z
−1
lim n sin (x(1 + x )) 2
dx = lim x (1 + x2 )−1 dx
n→∞ 0 n n→∞ 0
n
sin( nx )
Z ∞
= lim x (1 + x2 )−1 dx
0 n→∞
n
∞ sin( nx ) − sin(0)
Z
= lim x (1 + x2 )−1 dx
0 n→∞
n
Z ∞
= sin0 (0)(1 + x2 )−1 dx
Z0 ∞
= cos(0)(1 + x2 )−1 dx
Z0 ∞
= (1 + x2 )−1 dx
0
π
= .
2
(d) Fix n ∈ N. By the monotone convergence theorem and the fundamental theorem of calculus,
Z ∞ Z m
2 2 −1 d π
n(1 + n x ) dx = lim tan−1 (nx) dx = lim tan−1 (nm) − tan−1 (na) = − tan−1 (na).
a m→∞ a dx m→∞ 2
Therefore 
Z ∞


 0, a > 0

2 2 −1
lim n(1 + n x ) dx = π2 , a = 0
n→∞ a 

π, a < 0.

R∞
This implies that there is no f ∈ L1 such that limn→∞ a n(1 + n2 x2 )−1 dx = f, by a similar argument to
R

exercise 26. So the usual convergence theorem approach would not have helped to solve this exercise..

29. If t ∈ (0, ∞), then (by the monotone convergence theorem and Theorem 2.28)

k
e−tx
k  −tk
e0
Z Z 
−tx −tx e 1
e dx = lim e dx = lim = lim − + = .
0 k→∞ 0 k→∞ −t 0 k→∞ t t t
R∞
Given n ∈ N, define fn : [0, ∞) × [ 12 , 2] → [0, ∞) by fn (x, t) := xn e−tx . We claim that 0 fn (x, t) dx = n! t−(n+1) for
R∞
each t ∈ [ 21 , 2]. By induction, we may assume that 0 fn−1 (x, t) dx = (n − 1)! t−n . Note that

fn−1 (x, t) = xn−1 e−tx ≤ 3n−1 (n − 1)! ex/3 e−tx = 3n−1 (n − 1)! e−(t−1/3)x ≤ 3n−1 (n − 1)! e−(1/6)x

for all x ∈ [0, ∞), so fn−1 (−, t) ∈ L1 ([0, ∞)). Moreover




fn−1 (x, t) = |−xn e−tx | = fn (x, t) ≤ 3n n! e−(1/6)x

∂t

for all x ∈ [0, ∞), so by Theorem 2.27


Z ∞ Z ∞
∂ ∂
fn (x, t) dx = − fn−1 (x, t) dx = − (n − 1)! t−n = n(n − 1)! t−(n+1) = n! t−(n+1) ,
0 ∂t 0 ∂t
R ∞ n −x
as claimed. In particular 0 x e dx = n!.

10
Real Analysis Chapter 2 Solutions Jonathan Conder

R∞ 2
If t ∈ (0, ∞), then −∞ e−tx dx = π/t. The easiest proof of this requires a theorem from later in the course, but
p

if you are curious you can look up an alternative proof on Wikipedia. Given n ∈ N, define fn : R × [ 21 , 2] → [0, ∞)
2 R∞ √ −(2n+1)/2
by fn (x, t) := x2n e−tx . We claim that −∞ fn (x, t) dx = (2n)!
4n n! πt for each t ∈ [ 12 , 2]. By induction, we may
R∞ (2n−2)! √
assume that −∞ fn−1 (x, t) dx = 4n−1 (n−1)!
πt−(2n−1)/2 . Note that
2 2 /3 2 2 2
fn−1 (x, t) = x2n−2 e−tx ≤ 3n−1 (n − 1)! ex e−tx = 3n−1 (n − 1)! e−(t−1/3)x ≤ 3n−1 (n − 1)! e−(1/6)x

for all x ∈ R, so fn−1 (−, t) ∈ L1 (R). Moreover




fn−1 (x, t) = |−x2n e−tx2 | = fn (x, t) ≤ 3n n! e−(1/6)x2

∂t
for all x ∈ R, so by Theorem 2.27
Z ∞ Z ∞

fn (x, t) dx = − fn−1 (x, t) dx
−∞ ∂t −∞
∂ (2n − 2)! √ −(2n−1)/2
=− πt
∂t 4n−1 (n − 1)!
(2n − 2)!(2n − 1) √ −(2n+1)/2
= πt
2 · 4n−1 (n − 1)!
(2n)! √ −(2n+1)/2
= πt
4n · 4n−1 (n − 1)!
(2n)! √ −(2n+1)/2
= n πt ,
4 n!
R∞ 2 √
as claimed. In particular −∞ x2n e−x dx = (2n)!
4n n! π.

30. For each k ∈ N, we claim that (1 − k −1 x)k ≤ e−x for all x ∈ (0, k). If so, xn (1 − k −1 x)k χ(0,k) (x) ≤ xn e−x for all k ∈ N
and x ∈ [0, ∞), and by the previous exercise we may apply the dominated convergence theorem to show that
Z k Z ∞ Z k
lim xn (1 − k −1 x)k dx = lim xn (1 − k −1 x)k χ(0,k) dx = xn e−x dx = n!
k→∞ 0 0 k→∞ 0

(to prove that limk→∞ (1 − k −1 x)k


= e−x ,
take logarithms and apply l’Hôpital’s rule). Now we prove the claim. It
−1
suffices to show that k log(1 − k x) + x ≤ 0 for all x ∈ (0, k). This is certainly true for x = 0. Moreover,
d k(−k −1 ) 1 −k −1 x
(k log(1 − k −1 x) + x) = + 1 = 1 − = <0
dx 1 − k −1 x (1 − k −1 x) 1 − k −1 x
for all x ∈ (0, k). By the mean value theorem k log(1 − k −1 x) + x = (k log(1 − k −1 x) + x) − (k log(1 − 0) + 0) < 0 for
all x ∈ (0, k), which proves the claim.
R R R R
33. There clearly exists a subsequence ( fnk )k∈N of ( fn )n∈N such that limk→∞ fnk = lim inf n→∞ fn . Moreover
(fnk )k∈N converges to f in measure, because for every ε ∈ (0, ∞)

lim µ({x ∈ X | |fnk (x) − f (x)| ≥ ε}) = lim µ({x ∈ X | |fn (x) − f (x)| ≥ ε}) = 0.
k→∞ n→∞

In particular (fnk )k∈N is Cauchy in measure, so it has a subsequence (fnki )i∈N which converges to a measurable function
g pointwise almost everywhere. Clearly (fnki )i∈N also converges to g + pointwise almost everywhere. Moreover, f = g +
almost everywhere because (fnki )i∈N converges in measure to both f and g + (thus µ({x ∈ X | |f (x)−g + (x)| ≥ ε}) < δ
for all , δ ∈ (0, ∞)). Therefore, by Fatou’s lemma
Z Z Z Z Z Z
+
f = g ≤ lim inf fnki = lim fnki = lim fnk = lim inf fn .
i→∞ i→∞ k→∞ n→∞

11
Real Analysis Chapter 2 Solutions Jonathan Conder

R R R R
34. (a) It suffices to show that limn→∞ Re(fn ) = Re(f ) and limn→∞ Im(fn ) = Im(f ). Since

{x ∈ X | | Re(fn )(x) − Re(f )(x)| ≥ ε} ∪ {x ∈ X | | Im(fn )(x) − Im(f )(x)| ≥ ε} ⊆ {x ∈ X | |fn (x) − f (x)| ≥ ε}

for all n ∈ N and ε ∈ (0, ∞), while | Re(fn )| ≤ |fn | and | Im(fn )| ≤ |fn | for all n ∈ N, we may assume
without loss of generality that f and each fn are real-valued. Note that (fn )n∈N is Cauchy in measure, so it has
a subsequence which converges pointwise almost everywhere to a measurable function which equals f almost
everywhere. Therefore f ∈ L1 . Since (g + fn )n∈N and (g − fn )n∈N are sequences of non-negative measurable
functions which converge in measure to g + f and g − f respectively, the previous exercise implies that
Z Z Z Z Z Z  Z Z
g + f = (g + f ) ≤ lim inf (g + fn ) = lim inf g + fn = g + lim inf fn
n→∞ n→∞ n→∞

and Z Z Z Z Z Z  Z Z
g− f= (g − f ) ≤ lim inf (g − fn ) = lim inf g− fn = g − lim sup fn
n→∞ n→∞ n→∞
R R R R R R
Since g < ∞, it follows that lim supn→∞ fn ≤ f ≤ lim inf n→∞ fn , and hence f = limn→∞ fn .
(b) Note that (|fn − f |)n∈N converges to 0 in measure, because

{x ∈ X | |fn (x) − f (x)| − 0 ≥ ε} = {x ∈ X | |fn (x) − f (x)| ≥ ε}

for all n ∈ N and ε ∈ (0, ∞). Moreover |fn − f | ≤ |fn | + |f | ≤ 2g ∈ L1 for all n ∈ N. Therefore, by part (a),
Z Z
lim kfn − f k1 = lim |fn − f | = 0 = 0.
n→∞ n→∞

This implies that (fn )n∈N converges to f in L1 .

35. Suppose that (fn )n∈N converges to f in measure. For every ε ∈ (0, ∞), limn→∞ µ({x ∈ X | |fn (x) − f | ≥ ε}) = 0. In
particular, for every ε ∈ (0, ∞) there exists N ∈ N such that

0 − ε < µ({x ∈ X | |fn (x) − f | ≥ ε}) < 0 + ε = ε

for all n ∈ N with n ≥ N.


Conversely, suppose that, for every ε ∈ (0, ∞), there exists N ∈ N such that µ({x ∈ X | |fn (x) − f | ≥ ε}) < ε
for all n ∈ N with n ≥ N. Let ε ∈ (0, ∞) and δ ∈ (0, ∞). Define η := min{ε, δ}. There exists N ∈ N such that
µ({x ∈ X | |fn (x) − f | ≥ η}) < η for all n ∈ N with n ≥ N. Therefore

µ({x ∈ X | |fn (x) − f | ≥ ε}) ≤ µ({x ∈ X | |fn (x) − f | ≥ η}) < η ≤ δ

for all n ∈ N with n ≥ N, which implies that limn→∞ µ({x ∈ X | |fn (x) − f | ≥ ε}) = 0. This shows that (fn )n∈N
converges to f in measure.

37. (a) Let x ∈ X be a point where (fn )n∈N converges to f. Then

lim φ(fn (x)) = φ( lim fn (x)) = φ(f (x)),


n→∞ n→∞

so (φ ◦ fn )n∈N converges to φ ◦ f on the same set that (fn )n∈N converges to f.

12
Real Analysis Chapter 2 Solutions Jonathan Conder

(b) Suppose that (fn )n∈N converges to f uniformly, and let ε ∈ (0, ∞). Since φ is uniformly continuous, there exists
δ ∈ (0, ∞) such that |φ(w) − φ(z)| < ε for all w, z ∈ C with |w − z| < δ. Moreover, there exists N ∈ N such that
|fn (x) − f (x)| < δ for all x ∈ X and n ∈ N with n ≥ N. Therefore |φ(fn (x)) − φ(f (x))| < ε for all x ∈ X and
n ∈ N with n ≥ N. This shows that (φ ◦ fn )n∈N converges to φ ◦ f uniformly.
Now suppose that (fn )n∈N converges to f almost uniformly. For every ε ∈ (0, ∞) there exists E ∈ M such that
µ(E) < ε and (fn )n∈N converges to f uniformly on E c , and hence (φ ◦ fn )n∈N converges to φ ◦ f uniformly on
E c by the previous argument. This shows that (φ ◦ fn )n∈N converges to φ ◦ f almost uniformly.
Finally, suppose that (fn )n∈N converges to f in measure. Let ε ∈ (0, ∞). There exists δ ∈ (0, ∞) such that
|φ(w) − φ(z)| < ε for all w, z ∈ C with |w − z| < δ. Moreover

lim µ({x ∈ X | |fn (x) − f (x)| ≥ δ}) = 0.


n→∞

Clearly {x ∈ X | |φ(fn (x)) − φ(f (x))| ≥ ε} ⊆ {x ∈ X | |fn (x) − f (x)| ≥ δ} for all n ∈ N, so

lim µ({x ∈ X | |φ(fn (x)) − φ(f (x))| ≥ ε}) = 0


n→∞

and hence (φ ◦ fn )n∈N converges to φ ◦ f in measure.


(c) For each n ∈ N define a measurable function fn : R → C by fn (x) := 2−n . Also define φ : C → C by

1, Re(z) > 0
φ(z) :=
−1, Re(z) ≤ 0.

For all x ∈ R (fn (x))n∈N converges to 0, but (φ(fn (x)))n∈N = (1)n∈N does not converge to to φ(0) = −1.
Now define f : R → C by f (x) := x, and for each n ∈ N define fn : R → C by fn (x) := x + 2−n . Clearly (fn )n∈N
is a sequence of measurable functions converging uniformly, almost uniformly and in measure to the measurable
function f. Define φ : C → C by φ(z) := z 2 . Let E ⊆ R and suppose that (φ ◦ fn )n∈N converges to φ ◦ f uniformly
on E. Then there exists N ∈ N such that |φ(fn (x)) − φ(f (x))| < 1 for all x ∈ E and n ∈ N with n ≥ N Since

|φ(fN (x)) − φ(f (x))| = |(x + 2−N )2 − x2 | = |21−N x + 2−2N |

for all x ∈ E, it follows that E ⊆ (−2N −1 − 2−N −1 , 2N −1 − 2−N −1 ). In particular µ(E c ) = ∞, so (φ ◦ fn )n∈N
does not converge to φ ◦ f uniformly or almost uniformly. If ε ∈ (0, ∞) and n ∈ N, then

[2n−1 ε, ∞) ⊆ {x ∈ R | |φ(fn (x)) − φ(f (x))| ≥ ε}

because |φ(fn (x)) − φ(f (x))| = 21−n x + 2−2n ≥ ε + 2−2n for all x ∈ [2n−1 ε, ∞). Therefore

lim µ({x ∈ R | |φ(fn (x)) − φ(f (x))| ≥ ε}) = ∞,


n→∞

so (φ ◦ fn )n∈N does not converge to φ ◦ f in measure.

39. Let (fn )n∈N be a sequence of functions which converges to f almost uniformly. For each n ∈ N there exists En ∈ M
such that µ(En ) < 2−n and (fn )n∈N converges to f uniformly (hence pointwise) on Enc . Define E := ∩n∈N ∪∞ k=n Ek .
∞ ∞
Then µ(E) = limn→∞ µ(∪k=n Ek ) = 0, since µ(∪k=n Ek ) ≤ 2 1−n c
for all n ∈ N. Moreover, if x ∈ E there exists n ∈ N
c
such that x ∈ En and hence limk→∞ fk (x) = f (x). Therefore (fn )n∈N converges to f pointwise almost everywhere.
Let ε ∈ (0, ∞) and take E ∈ M such that µ(E) < ε and (fn )n∈N converges to f uniformly on E c . There exists N ∈ N
such that |fn (x) − f (x)| < ε for all x ∈ E c and n ∈ N with n ≥ N. It follows that {x ∈ X | |fn (x) − f (x)| ≥ ε} ⊆ E,
and hence µ({x ∈ X | |fn (x) − f (x)| ≥ ε}) ≤ µ(E) < ε. By exercise 35, (fn )n∈N converges to f in measure.

13
Real Analysis Chapter 2 Solutions Jonathan Conder

40. Let (fn )n∈N be a sequence of complex-valued measurable functions that converge pointwise to some f : X → C on a
set A ⊆ X with µ(Ac ) = 0. Suppose there exists g ∈ L1 (µ) such that |fn | ≤ g for all n ∈ N. Then |f (x)| ≤ g(x) for
all x ∈ A. Fix k ∈ N, and for each n ∈ N define Ek,n := ∪∞ −1
m=n {x ∈ A | |fm (x) − f (x)| ≥ 2k }. If x ∈ Ek,1 there
exists m ∈ N such that |fm (x) − f (x)| ≥ 2k −1 and hence 2g(x) ≥ |fm (x)| + |f (x)| ≥ 2k −1 . Therefore k −1 χEk,1 ≤ g, so
k −1 χEk,1 ∈ L1 and hence µ(Ek,1 ) < ∞. Since ∩n∈N Ek,n ⊆ A ∩ Ac = ∅, it follows that limn→∞ µ(En,k ) = 0. Now let
ε ∈ (0, ∞) and for each k ∈ N choose nk ∈ N so that µ(Enk ,k ) < 2−k ε. Define E := (∪k∈N Enk ,k ) ∪ Ac . Then µ(E) < ε,
and for each δ ∈ (0, ∞) there exists k ∈ N such that 2k −1 < δ, whence |fm (x) − f (x)| < 2k −1 < δ for all x ∈ E c and
m ∈ N with m ≥ nk . This implies that (fn )n∈N converges to f uniformly on E c .

42. Suppose that (fn )n∈N converges to f in measure. Given ε ∈ (0, ∞), there exists N ∈ N such that

µ({x ∈ N | ε ≤ |fn (x) − f (x)|}) < 1

for all n ∈ N with n ≥ N. This implies that {x ∈ N | ε ≤ |fn (x) − f (x)|} = ∅, and hence kfn − f ku ≤ ε, for all n ∈ N
with n ≥ N. Therefore (fn )n∈N converges uniformly to f.
Now suppose that (fn )n∈N converges uniformly to f. If ε ∈ (0, ∞), there exists N ∈ N such that |fn (x) − f (x)| < ε
for all x, n ∈ N with n ≥ N. In particular µ({x ∈ N | ε ≤ |fn (x) − f (x)|}) = 0 for all n ∈ N with n ≥ N, which implies
that (fn )n∈N converges to f in measure.

44. For each n ∈ N define En := f −1 (Bn (0)) = {x ∈ [a, b] | |f (x)| < n}. Then limn→∞ µ(En ) = µ(∪n∈N En ) = µ([a, b]), so
there exists m ∈ N such that µ([a, b]) − µ(Em ) < 3ε . Define g : R → C by

f (x), x ∈ E
m
g(x) :=
0, x ∈ Ec .
m

Then |g| ≤ mχEm ≤ mχ[a,b] , so g ∈ L1 (µ). Hence for each n ∈ N there exists a compactly supported continuous
function gn : R → C such that kgn − gk1 < n−1 . Clearly (gn )n∈N converges to g in measure, so there exists a
subsequence (gnk )k∈N which converges to g pointwise almost everywhere. After restricting these functions to [a, b],
Egoroff’s theorem implies that there exists F ⊆ [a, b] such that µ(F ) < 3ε and (gnk )k∈N converges to g uniformly on
[a, b] \ F. By inner regularity there exists a compact set E ⊆ Em \ F such that µ(E) > µ(Em \ F ) − 3ε and hence
ε ε ε
µ([a, b] \ E) = µ([a, b]) − µ(E) < µ([a, b]) − µ(Em \ F ) + ≤ µ([a, b]) + µ(F ) − µ(Em ) + < 3 · = ε.
3 3 3
Moreover (gnk )k∈N converges to g uniformly on E, so f |E = g|E is continuous.
R
46. Fix y ∈ Y. Then χD (x, y) = χ{y} (x) for all x ∈ X, so χD (x, y) dµ(x) = µ({y}) = 0. This implies that
ZZ Z
χD (x, y) dµ(x) dν(y) = 0 dν(y) = 0.
R
Now fix x ∈ X. Clearly χD (x, y) = χ{x} (y) for all y ∈ Y, so χD (x, y) dν(y) = ν({x}) = 1. It follows that
ZZ Z
χD (x, y) dν(y) dµ(x) = 1 dµ(x) = µ(X) = 1.
R
By definition χD d(µ × ν) = (µ × ν)(D), and hence
Z (∞ )
X
χD d(µ × ν) = inf (µ × ν)(En ) (En )∞
n=1 is a sequence of finite disjoint unions of rectangles covering D


n=1

14
Real Analysis Chapter 2 Solutions Jonathan Conder


( )
X

= inf µ(An )ν(Bn ) (An × Bn )n=1 is a sequence of rectangles covering D


n=1

If (An × Bn )∞ ∞
n=1 is a sequence of rectangles covering D, then (An ∩ Bn )n=1 covers X. Clearly this implies that
µ∗ (An ∩ Bn ) > 0 for some n ∈ N. In particular µ(An ) > 0 and ν(Bn ) = ∞, because the Lebesgue outer measure of a
finite set is 0. Therefore ∞
P R
n=1 µ(An )ν(Bn ) = ∞, so χD d(µ × ν) = inf{∞} = ∞.

48. Clearly |f | d(µ × ν) = (µ × ν)(∪∞ ∞


R
n=1 {(n, n), (n + 1, n)}). If (An × Bn )n=1 is a sequence of rectangles covering

∪∞ ∞
P
n=1 {(n, n), (n + 1, n)}, then (An ∩ Bn )n=1 covers N and hence n=1 µ(An ∩ Bn ) = ∞. This implies that


X ∞
X ∞
X ∞
X
2
µ(An )ν(Bn ) = µ(An )µ(Bn ) ≥ µ(An ∩ Bn ) ≥ µ(An ∩ Bn ) = ∞,
n=1 n=1 n=1 n=1
R
since µ(An ∩ Bn ) ∈ {0} ∪ [1, ∞] for all n ∈ N. Therefore |f | d(µ × ν) = inf{∞} = ∞. Fix n ∈ Y. Then f (m, n) =
R
χ{n} (m) − χ{n+1} (m) for all m ∈ X, and hence f (m, n) dµ(m) = µ({n}) − µ({n + 1}) = 0. This implies that
ZZ Z
f (m, n) dµ(m) dν(n) = 0 dν(n) = 0.
R
Now fix m ∈ X \{1}. Then f (m, n) = χ{m} (n)−χ{m−1} (n) for all n ∈ Y, so f (m, n) dν(n) = ν({m})−ν({m−1}) = 0.
R R
Moreover, f (1, n) dν(n) = χ{1} dν = ν({1}) = 1. It follows that
ZZ Z
f (m, n) dν(n) dµ(m) = χ{1} dµ = µ({1}) = 1.

49. (a) Since µ and ν are σ-finite,


Z Z
ν(Ex ) dµ(x) = µ(E y ) dν(y) = (µ × ν)(E) = 0.

This implies that ν(Ex ) = µ(E y ) = 0 for almost every x ∈ X and y ∈ Y.


(b) Let E ⊆ X × Y be a null set such that f (x, y) = 0 for all x ∈ X and y ∈ Y such that (x, y) ∈/ E. If x ∈ X, then
R
fx (y) = 0 for all y ∈ Y such that y ∈
/ Ex . Hence fx = 0 almost everywhere, so fx is integrable with fx dν = 0,
for almost all x ∈ X (by the previous exercise). Similarly f y is integrable and f y dµ = 0 for almost every
R

y ∈ Y.
Now let f be L-measurable. There exists an (M ⊗ N)-measurable function g such that f = g λ-almost everywhere.
Moreover gx is N-measurable and g y is M-measurable for all x ∈ X and y ∈ Y. If f ≥ 0 then g ≥ 0 without loss of
generality, so by Tonelli’s theorem x 7→ gx dν and y 7→ g y dµ are non-negative and (M ⊗ N)-measurable, while
R R

Z ZZ ZZ
g dλ = g(x, y) dµ(x) dν(y) = g(x, y) dν(y) dµ(x). (2)

Since |g| = |f | λ-almost everywhere, |g| d(µ × ν) = |g| dλ = |f | dλ and hence g ∈ L1 (µ × ν) whenever f ∈ L1 (λ).
R R R

By Fubini’s theorem, this implies that gx ∈ L1 (ν) and g y ∈ L1 (µ) for almost all x ∈ X and y ∈ Y, while x 7→ gx dν
R

and y 7→ g y dµ are in L1 (µ) and L1 (ν) respectively. Also (2) holds in this case. The corresponding statements about
R

f follow by applying part (b) of this exercise to f −g. In particular, fx −gx = 0 almost everywhere for almost all x ∈ X,
so fx is N-measurable for almost all x ∈ X. Similarly f y is M-measurable for almost all y ∈ Y. Since fx − gx ∈ L1 (ν)
and f y − g y ∈ L1 (µ) for almost all x ∈ X and y ∈ Y, it is clear that fx ∈ L1 (ν) and f y ∈ L1 (µ) for almost all x ∈ X

15
Real Analysis Chapter 2 Solutions Jonathan Conder

and y ∈ Y, provided that f ∈ L1 (λ). In either of the two cases gx dν = (fx − gx ) dν + gx dν = fx dν for almost
R R R R
R
all x ∈ X, so x 7→ fx dν is measurable and in the second case, integrable (for the first case, assume without loss of
generality that g ≤ f ). The same clearly holds for y 7→ f y dµ, so (because f = g almost everywhere)
R

Z ZZ ZZ ZZ ZZ
f dλ = g(x, y) dµ(x) dν(y) = f (x, y) dµ(x) dν(y) = g(x, y) dν(y) dµ(x) = f (x, y) dν(y) dµ(x).

50. Subtraction is a continuous map from [0, ∞]×[0, ∞) → (−∞, ∞], since it is constant on the closed set {∞}×[0, ∞). In
particular, the preimage of [0, ∞) (or (0, ∞)) under subtraction is an open subset of [0, ∞]×[0, ∞), so it is a countable
union of rectangles (An × Bn )∞ n=1 . Hence, the preimage of [0, ∞) (or (0, ∞)) under the map (x, y) 7→ f (x) − y is

E := {(x, y) ∈ X × [0, ∞) | (f (x), y) ∈ An × Bn for some n ∈ N}


= ∪∞
n=1 {(x, y) ∈ X × [0, ∞) | x ∈ f
−1
(An ) and y ∈ Bn }
= ∪∞
n=1 (f
−1
(An ) × Bn ).

Clearly E is (M ⊗ BR )-measurable, and hence Gf = E ∪ (f −1 ({∞}) × {∞}) is also (M ⊗ BR )-measurable (the same
holds for the redefinition of Gf , because in that case Gf = E if we take (0, ∞) instead of [0, ∞) above). For the
second part, assume that m({∞}) = 0 and m|[0,∞) agrees with the Lebesgue measure. By Tonelli’s theorem
Z ZZ
(µ × m)(Gf ) = (µ × m)(E) = χE = χE (x, y) dm(y) dµ(x).

R R f (x)
If x ∈ X then (x, y) ∈ E iff y ∈ [0, ∞) and f (x) − y ≥ 0 (or > 0), so χE (x, y) dm(y) = 0 1 dm(y) = f (x).
R
Therefore (µ × m)(Gf ) = f dµ as required.

51. (a) Define F, G : X × Y → C by F (x, y) := f (x) and G(x, y) := g(y). Then F −1 (A) = f −1 (A) × Y and G−1 (A) =
X × g −1 (A) for all A ⊆ C, so F and G are (M ⊗ N)-measurable. Therefore h = F G is (M ⊗ N)-measurable.
(b) Suppose f ≥ 0 and g ≥ 0. There exist increasing sequences (φn )∞ ∞
n=1 and (ψn )n=1 of non-negative simple functions
which converge pointwise to f and g respectively. For each n ∈ N define Φn , Ψn : X × Y → [0, ∞] as in part (a),
Pk Pl
so that (Φn Ψn )∞ n=1 converges pointwise to h. Fix n ∈ N, and write φn = i=1 ai χAi and φn = j=1 bj χBj for
some a1 , a2 , . . . , ak , b1 , b2 , . . . , bl ∈ [0, ∞] and measurable sets A1 , A2 , . . . , Ak , B1 , B2 , . . . , Bl . Clearly

k
! l 
k X l k X l
X X X X
Φn Ψn = ai χ(Ai ×Y )  bj χ(X×Bj ) =
 ai χ(Ai ×Y ) bj χ(X×Bj ) = ai bj χ(Ai ×Bj ) ,
i=1 j=1 i=1 j=1 i=1 j=1

and hence
Z l
k X k X
l k
! l

Z Z
X X X X
Φn Ψn = ai bj (µ×ν)(Ai ×Bj ) = ai µ(Ai )bj ν(Bj ) = ai µ(Ai )  bj ν(Bj ) = φn · ψn .
i=1 j=1 i=1 j=1 i=1 j=1

By the monotone convergence theorem, it follows that


Z Z Z Z Z Z Z Z
h = lim Φn Ψn = lim φn · ψn = lim φn · lim ψn = f · g.
n→∞ n→∞ n→∞ n→∞

|f | · |g| < ∞, so h ∈ L1 (µ × ν). If f (X) ⊆ R and g(Y ) ⊆ R, then


R R R
Hence, in general |h| =
Z Z Z
h = h − h−+

16
Real Analysis Chapter 2 Solutions Jonathan Conder

Z Z Z Z
= F G + F G − F G − F − G+
+ + − − + −

Z Z Z Z Z Z Z Z
− − − −
= f · g + f · g − f · g − f · g+
+ + +

Z Z Z  Z Z Z 
= f+ g+ − g− + f − g− − g+
Z Z  Z Z 
+ − + −
= f − f g − g
Z Z
= f · g.

Since F G = (Re(F )+i Im(F ))(Re(G)+i Im(G)) = Re(F ) Re(G)−Im(F ) Im(G)+i(Re(F ) Im(G)+Im(F ) Re(G)),
Z Z Z
h = Re(h) + i Im(h)
Z Z Z Z
= Re(F ) Re(G) − Im(F ) Im(G) + i Re(F ) Im(G) + i Im(F ) Re(G)
Z Z Z Z Z Z Z Z
= Re(f ) · Re(g) − Im(f ) · Im(g) + i Re(f ) · Im(g) + i Im(f ) · Re(g)
Z Z Z  Z Z Z 
= Re(f ) Re(g) + i Im(g) − Im(f ) Im(g) − i Re(g)
Z Z Z  Z Z Z 
= Re(f ) Re(g) + i Im(g) + i Im(f ) Re(g) + i Im(g)
Z Z  Z Z 
= Re(f ) + i Im(f ) Re(g) + i Im(g)
Z Z
= f · g.

55. (a) Fix y ∈ (0, 1], and define F : [0, 1] → R by F (x) := x(x2 + y 2 )−1 . By the quotient rule

(x2 + y 2 ) − x(2x) y 2 − x2
F 0 (x) = = = −f y (x)
(x2 + y 2 )2 (x2 + y 2 )2

for all x ∈ [0, 1]. This implies that


Z 1 Z y
y 1
(f y )− = −f y = F (y) − F (0) = F (y) = =
0 0 2y 2 2y
R1 R1 1 1
and similarly 0 (f y )+ = y f y = −F (1) + F (y) = 2y − 1+y 2 . By the Tonelli and monotone convergence theorems

1Z 1 1 1 − log( n1 )
Z Z Z Z
− − 1 1
f = f (x, y) dx dy = dy = lim dy = lim = ∞,
E 0 0 0 2y n→∞ 1 2y n→∞ 2
n

which implies that Z Z 1Z 1 Z 1 


+ + 1 1
f = f (x, y) dx dy = − dy = ∞
E 0 0 0 2y 1 + y 2
R1 1
R1
because 0 1+y 2 dy ≤ 0 1 dy < ∞ and
Z 1  Z 1 Z 1
1 1 1 1
− dy + dy = dy = ∞
0 2y 1 + y 2 0 1 + y2 0 2y

17
Real Analysis Chapter 2 Solutions Jonathan Conder

R
This shows that E f is not defined. However,
Z 1Z 1 Z 1 Z 1
1 π
f (x, y) dx dy = (−F (1) + F (0)) dy = − 2
dy = −(tan−1 (1) − tan−1 (0)) = − .
0 0 0 0 1+y 4

Since f (x, y) = −f (y, x) for all x, y ∈ (0, 1], it follows that


Z 1Z 1
π
f (x, y) dy dx = .
0 0 4

(b) Since f (x, y) ≥ 0 for all (x, y) ∈ E \ {(1, 1)}, all three integrals exist and Tonelli’s theorem implies that they are
equal.
(c) By Tonelli’s theorem, the fundamental theorem of calculus and the monotone convergence theorem,
Z Z 1Z 1
+
f = f + (x, y) dx dy
E 0 0
1
1
1 −3
Z Z  
2
= x− dx dy
0 1
+y 2
2
Z 1  −2 !
2 1
= (−2)−1 − y −2 dy
0 2
Z 1
1 2 −2
= (y − 4) dy
2 0
Z 1
1 2
= lim (y −2 − 4) dy
n→∞ 2 1
n+4
 −1  −1 !
1 1 1 4 4
= lim − + − +
n→∞ 2 2 n+4 2 n+4
n
= lim
n→∞ 2

= ∞.

Similarly
Z Z 1Z 1

f = f − (x, y) dx dy
E 0 0
1 1 −3
Z Z −y 
2 2 1
= −x dx dy
0 0 2
Z 1  −2 !
2 1
= 2−1 −2
y − dy
0 2
Z 1
1 2
= (y −2 − 4) dy
2 0
= ∞.
R
Therefore E f does not exist. However, the above working implies that
1
Z 1Z 1 Z
2
Z 1
f (x, y) dx dy = f (x, y) dx dy
0 0 0 0

18
Real Analysis Chapter 2 Solutions Jonathan Conder

1
Z
2
Z 1 Z 1 
+ −
= f (x, y) dx − f (x, y) dx dy
0 0 0
Z 1  
2 1 −2 1
= (y − 4) − (y −2 − 4) dy
0 2 2
= 0.

If x ∈ [0, 21 ], then fx ≤ 0 and hence


1 −3 −2
Z 1 Z 1 Z −x  

2 1 1
f (x, y) dy = − f (x, y) dy = − −x dy = − −x .
0 0 0 2 2

Similarly, if x ∈ [ 12 , 1] then

x− 12
1 1
1 −3 1 −2
Z Z Z    
+
f (x, y) dy = f (x, y) dy = x− dy = x − .
0 0 0 2 2

By the monotone convergence theorem and the fundamental theorem of calculus, this implies that
Z 1 Z 1 + Z 1 −2
1
f (x, y) dy dx = x− dx
0 0 1 2
2
1
1 −2
Z  
= lim x− dx
n→∞ 1 1
+ n+2 2
2
 −1  −1 !
1 1
= lim − +
n→∞ 2 n+2
= lim n
n→∞

= ∞.

Similarly
Z 1 Z 1 − Z 1 −2

1 2
f (x, y) dy dx = −x dx
0 0 0 2
Z 1− 1  −2
2 n+2 1
= lim −x dx
n→∞ 0 2
 −1  −1 !
1 1
= lim −
n→∞ n+2 2
= lim n
n→∞

= ∞.
R1R1
This shows that 0 0 f (x, y) dy dx does not exist.

56. Define a measurable function h : (0, a)2 → C by h(t, x) := t−1 f (t)χE (t, x), where E := {(t, x) ∈ (0, a)2 | x < t} is
Ra
open, hence measurable. Then g(x) = 0 t−1 f (t)χ(x,a) (t) dt = hx for all x ∈ (0, a). By Tonelli’s theorem
R

Z Z aZ a Z aZ t Z a
|h| = t−1 |f (t)|χE (t, x) dx dt = t−1 |f (t)| dx dt = |f (t)| dt < ∞.
0 0 0 0 0

19
Real Analysis Chapter 2 Solutions Jonathan Conder

Hence h is integrable, so g is integrable by Fubini’s theorem, and


Z a Z aZ a Z Z aZ a Z aZ t Z a
−1 −1
g= hx (t) dt dx = h = t f (t)χE (t, x) dx dt = t f (t) dx dt = f (t) dt.
0 0 0 0 0 0 0 0

58. Let s ∈ (0, ∞) and define f : [0, ∞) × [0, 1] → [0, 1] by f (x, y) := e−sx sin(2xy). Clearly |f (x, y)| ≤ e−sx for all
x, y ∈ [0, ∞) × [0, 1], so f ∈ L1 . Since
∞Z 1 ∞   1 ∞ ∞
− cos(2x)
Z Z Z Z
−sx −sx cos(2xy) −sx 1
e−sx x−1 sin2 (x) dx,

e sin(2xy) dy dx = −e dx = e dx =
0 0 0 2x
0 0 2x 0

Fubini’s theorem implies that


Z ∞ Z 1Z ∞ Z 1
1
−sx −1 −sx 2y 1 1
2
dy = log(4 s + y ) = log(1 + 4s−2 )
−1 2 2

e x sin (x) dx = e sin(2xy) dx dy = 2 2
0 0 0 0 s + 4y 4 0 4

(the middle step can be done using integration by parts or by expressing sin as a difference of complex exponentials).

59. (a) Let n ∈ N. If x ∈ [(n + 61 )π, (n + 56 )π] then | sin(x)| ≥ 21 , and x−1 ≥ (n + 1)−1 π −1 . Therefore

∞ ∞ ∞ 4 ∞

Z Z X Z
1 X 1 X
|f | ≥ χ[(n+ 1 )π,(n+ 5 )π] = χ[(n+ 1 )π,(n+ 5 )π] = = ∞,
0 2(n + 1)π 6 6 2(n + 1)π 6 6 2(n + 1)π
n=1 n=1 n=1

by the monotone convergence theorem.


(b) Fix b ∈ (0, ∞), and define f : (0, b)2 → R by f (x, y) := e−xy sin(x). Clearly |f | ≤ 1, so |f | ≤ b2 < ∞. Hence,
R

by Fubini’s theorem and the fundamental theorem of calculus


Z bZ b Z bZ b
f (x, y) dx dy = e−xy sin(x) dy dx
0 0 0 0
Z b  −xb
sin(x) e0 sin(x)

e
= − dx
0 −x −x
sin(x) e−xb sin(x)
Z b 
= − dx (3)
0 x x

Since | sin(x)| ≤ x for all x ∈ (0, ∞) it is clear that


2 2
e−xb sin(x)
Z b −xb
b
e−b 1 − e−b
Z b
e0
Z
e sin(x) −xb
dx ≤
dx ≤ e dx = − = . (4)

0 x 0
x 0 −b −b b

Integrating f by parts twice with respect to x suggests we define a function F : (0, b)2 → R by

y sin(x) + cos(x)
F (x, y) := −e−xy ,
y2 + 1
so that
∂ y sin(x) + cos(x) y cos(x) − sin(x)
F (x, y) = ye−xy 2
− e−xy
∂x y +1 y2 + 1
y 2 sin(x) + y cos(x) − y cos(x) + sin(x)
= e−xy
y2 + 1
= f (x, y)

20
Real Analysis Chapter 2 Solutions Jonathan Conder

and hence
Z bZ b Z b
f (x, y) dx dy = (F (b, y) − F (0, y)) dy
0 0 0
Z b 
0 y sin(0) + cos(0) −by y sin(b) + cos(b)
= e −e dy
0 y2 + 1 y2 + 1
Z b 
1 −by y sin(b) + cos(b)
= −e dy. (5)
0 y2 + 1 y2 + 1
Either y ≤ 1 or y ≤ y 2 for all y ∈ (0, ∞), so
2
1 − e−b
Z b Z b   Z b
−by y sin(b) + cos(b)
−by y| sin(b)| | cos(b)| −by
e dy ≤
e + 2 dy ≤ 2e dy = 2 . (6)

0 y2 + 1 0 y2 + 1 y +1 0 b
Together (3), (4), (5) and (6) imply that
Z b Z bZ b Z b
sin(x) 1 π
lim dx + 0 = lim f (x, y) dx dy = lim 2
dy + 0 = lim (tan−1 (b) − tan−1 (0)) = .
b→∞ 0 x b→∞ 0 0 b→∞ 0 y + 1 b→∞ 2

60. If x, y ∈ (0, ∞) then, by Exercise 51


Z ∞ Z ∞ Z ∞Z ∞
x−1 −s y−1 −t
Γ(x)Γ(y) = s e ds t e dt = sx−1 ty−1 e−(s+t) ds dt.
0 0 0 0

Define G : (0, ∞) × (0, 1) → (0, ∞)2 by G(u, v) := (uv, u(1 − v)), and check that G is a C 1 -diffeomorphism with
Jacobian determinant −u at the point (u, v). By Theorem 2.47, Tonelli’s theorem and Exercise 51, Γ(x)Γ(y) is
Z 1Z ∞ Z 1 Z ∞ Z 1
(uv)x−1 (u(1 − v))y−1 e−u u du dv = v x−1 (1 − v)y−1 dv ux+y−1 e−u du = Γ(x + y) tx−1 (1 − t)y−1 dt.
0 0 0 0 0

61. (a) Let α, β ∈ (0, ∞) and x ∈ [0, ∞). Note that


Z x
1
Iα (Iβ f )(x) = (x − t)α−1 Iβ f (t) dt
Γ(α) 0
Z x Z t
1 α−1 1
= (x − t) (t − s)β−1 f (s) ds dt
Γ(α) 0 Γ(β) 0
Z xZ t
1
= (x − t)α−1 (t − s)β−1 f (s) ds dt.
Γ(α)Γ(β) 0 0
R1
Since f is bounded on [0, x] and 0 tγ dt < ∞ for all γ ∈ (−1, ∞), the Fubini-Tonelli theorem implies that
Z xZ x
1
Iα (Iβ f )(x) = (x − t)α−1 (t − s)β−1 f (s) dt ds.
Γ(α)Γ(β) 0 s
For each s we apply the substitution u := (t − s)/(x − s) to the inner integral, and obtain
Z xZ 1
1
Iα (Iβ f )(x) = (x − s − u(x − s))α−1 (u(x − s))β−1 f (s)(x − s) du ds
Γ(α)Γ(β) 0 0
Z xZ 1
1
= (x − s)α−1 (1 − u)α−1 uβ−1 (x − s)β−1 f (s)(x − s) du ds
Γ(α)Γ(β) 0 0
Z x Z 1
1 α+β−1
= (x − s) f (s) ds (1 − u)α−1 uβ−1 du
Γ(α)Γ(β) 0 0
Z x
1 α+β−1
= (x − s) f (s) ds
Γ(α + β) 0
= Iα+β f (x).

21
Real Analysis Chapter 2 Solutions Jonathan Conder

(b) Clearly I1 f is an antiderivative of f. Given n ∈ N, we aim to show that In f is an nth-order antiderivative of f.


By induction, we may assume that n > 1 and that In−1 (I1 f ) is an (n − 1)th-order antiderivative of I1 f. Hence

(In f )(n) = ((In−1 (I1 f ))(n−1) )0 = (I1 f )0 = f.

62. Let E ⊆ S n−1 be measurable and T ∈ SO(n) a rotation. We want to show that σ(T (E)) = σ(E). Note that
 
−1
Φ ((0, 1) × E) = x ∈ Rn \ {0} |x| ∈ (0, 1) and x ∈ E ,

|x|
while
 
−1
n x
Φ ((0, 1) × T (E)) = x ∈ R \ {0} |x| ∈ (0, 1) and
∈ T (E)
|x|
T −1 x
 
n
−1
= x ∈ R \ {0} |T x| ∈ (0, 1) and −1 ∈ E

|T x|
= {x ∈ Rn \ {0} | T −1 x ∈ Φ−1 ((0, 1) × E)}
= T (Φ−1 ((0, 1) × E)).

Therefore

ρ((0, 1))σ(E) = m∗ ((0, 1)×E) = m(Φ−1 ((0, 1)×E)) = m(T (Φ−1 ((0, 1)×E))) = m∗ ((0, 1)×T (E)) = ρ((0, 1))σ(T (E)).
R1
Since ρ((0, 1)) = 0 rn−1 dr = n−1 > 0, it follows that σ(T (E)) = σ(E).

64. Let a, b ∈ R and set α := a + n − 1. By Corollary 2.51 it suffices to determine when


Z 1/2 Z ∞
α b
r | log(r)| dr and rα | log(r)|b dr
0 2

are finite. We claim that, given ε ∈ (0, ∞), there exists δ ∈ (0, 21 ) such that | log(r)| ≤ 2r−ε for all r ∈ (0, δ). To prove
this, choose n ∈ N such that n−1 < ε and define δ := log(n!)−n . If r ∈ (0, δ) then
−1/n
| log(r)| = log(r−1 ) = log((r−1/n )n ) ≤ log(n! er ) = log(n!) + r−1/n = δ −1/n + r−1/n ≤ 2r−1/n ≤ 2r−ε ,

1
as claimed. If α > −1 and b > 0, set ε := 2b (α + 1), take the corresponding δ and note that
Z δ Z δ Z δ
α b α b −(α+1)/2 b
r | log(r)| dr ≤ r 2r dr = 2 r(α−1)/2 dr < ∞,
0 0 0

as 21 (α − 1) > −1. Since rα | log(r)|b is bounded for r ∈ (δ, 12 ), it follows that the first integral is finite. If α > −1 and
b ≤ 0 then | log(r)|b = log(r−1 )b ≤ log(2)b for all r ∈ (0, 12 ), in which case the first integral is finite. Similarly, it is
1
infinite if α < −1 and b > 0. If α < −1 and b < 0, set ε := − 2b (α + 1), take the corresponding δ and note that
Z δ Z δ Z δ
rα | log(r)|b dr ≥ rα 2b r(α+1)/2 dr = 2b r(3α+1)/2 dr = ∞,
0 0 0

as 12 (3α + 1) < −1. Thus, the first integral is finite. Finally, set α := −1. By the monotone convergence theorem

1/2 1/2
1/2
(− log(r))b+1
Z Z
α b b
r | log(r)| dr = − lim (− log(r)) d(− log(r)) = − lim ,
0 t→0 t t→0 b+1 t

22
Real Analysis Chapter 2 Solutions Jonathan Conder

R 1/2
which is finite iff b < −1 (the case b = −1 should really be handled separately). In summary, 0 rα | log(r)|b dr
is finite iff a > −n or (a = −n and b < −1). The second integral can be treated similarly; alternatively we can
substitute s := r−1 and note that
Z ∞ Z 1/2
rα | log(r)|b dr = s−α−2 | log(s)|b ds,
2 0

which is finite iff −α − 2 > −1 or (−α − 2 = −1 and b < −1); in other words a < −n or (a = −n and b < −1).

65. (a) This is clear if n = 2. If n ≥ 3, let F : Rn−1 → Rn−1 be the map corresponding to G. Also let π : Rn−1 → Rn−2
and ρ : Rn−1 → R be projections on to the first n − 2 and last coordinate(s), respectively. Clearly

G(r, φ1 , . . . , φn−2 , θ) = (π(F (r, φ1 , . . . , φn−2 )), ρ(F (r, φ1 , . . . , φn−2 )) cos(θ), ρ(F (r, φ1 , . . . , φn−2 )) sin(θ)).

If x ∈ Rn then (by induction) we may assume that (x1 , . . . , xn−2 , |(xn−1 , xn )|) = F (r, φ1 , . . . , φn−2 ) for some
r, φ1 , . . . , φn−2 ∈ R, in which case it is clear that x = G(r, φ1 , . . . , φn−2 , θ) for some θ ∈ R. Moreover, if
r, φ1 , . . . , φn−2 , θ ∈ R then (assuming, by induction, that |F (r, φ1 , . . . , φn−2 )| = |r|)
q
|G(r, φ1 , . . . , φn−2 , θ)| = |π(F (r, φ1 , . . . , φn−2 ))|2 + ρ(F (r, φ1 , . . . , φn−2 ))2 cos2 (θ) + ρ(F (r, φ1 , . . . , φn−2 ))2 sin2 (θ)
q
= |r|2 − ρ(F (r, φ1 , . . . , φn−2 ))2 + ρ(F (r, φ1 , . . . , φn−2 ))2 (cos2 (θ) + sin2 (θ))

= r2
= |r|.

(b) Denote the component functions of F by F 1 , . . . , F n−1 . The Jacobian of G at a point (r, φ1 , . . . , φn−2 , θ) ∈ Rn is
 
Fr1 Fφ11 ··· Fφ1n−2 0
 .. .. .. .. .. 

 . . . . . 

 Frn−2 n−2 n−2 ,
· · ·
 
 F φ 1
Fφ n−2
0 
F n−1 cos(θ) F n−1 cos(θ) · · · F n−1 cos(θ) −F n−1 sin(θ)
 r φ1 φn−2 
Frn−1 sin(θ) Fφn−1
1
sin(θ) · · · Fφn−1
n−2
sin(θ) F n−1 cos(θ)

so its determinant is F n−1 sin2 (θ) det(DF ) + F n−1 cos2 (θ) det(DF ) = r sin(φ1 ) . . . sin(φn−2 ) det(DF ). It easily
follows by induction that det(DG) has the required form (the case n = 2 is trivial).
(c) This is well-known for n = 2, so we may assume that n ≥ 3 and F |(0,∞)×(0,π)n−3 ×(0,2π) is injective. We may refine
our argument from part (a) to show that G(Ω) contains the points of Rn whose coordinates are all nonzero, in
which case Rn \ G(Ω) is clearly a null set. If (r, φ1 , . . . , φn−2 , θ) ∈ Ω and (R, Φ1 , . . . , Φn−2 , Θ) ∈ Ω map to the
same point under G, then
π(F (r, φ1 , . . . , φn−2 )) = π(F (R, Φ1 , . . . , Φn−2 ))
and ρ(F (r, φ1 , . . . , φn−2 ))2 = ρ(F (R, Φ1 , . . . , Φn−2 ))2 (because cos2 (θ) + sin2 (θ) = cos2 (Θ) + sin2 (Θ)). By defini-
tion ρ(F (r, φ1 , . . . , φn−2 )) = r sin(φ1 ) . . . sin(φn−2 ), which is positive by the definition of Ω. Therefore

ρ(F (r, φ1 , . . . , φn−2 )) = ρ(F (R, Φ1 , . . . , Φn−2 )),

and hence (r, φ1 , . . . , φn−2 ) = (R, Φ1 , . . . , Φn−2 ). It clearly follows that θ = Θ. This shows that G|Ω is injective,
so it has an inverse defined on G(Ω); the inverse function theorem (cf. part (b)) implies that the inverse is smooth
and G(Ω) is open, in which case G|Ω is a diffeomorphism.

23
Real Analysis Chapter 2 Solutions Jonathan Conder

(d) If we view S n−1 as a smooth manifold it is straightforward to show that (F |Ω0 )−1 is a diffeomorphism, but
that’s outside the scope of this course. Given an integrable function f : S 1 → C, define g : Rn → C by
x
g(x) := f ( |x| )χB1 (0) (x). By Theorem 2.49 and Exercise 51
Z Z ∞Z Z 1Z Z
n−1 n−1 1
g= g(rx)r dσ(x) dr = f (x)r dσ(x) dr = f.
Rn 0 S n−1 0 S n−1 n S n−1

On the other hand,


Z
f (F (φ1 , . . . , φn−2 , θ)) sinn−2 (φ1 ) . . . sin(φn−2 ) dφ1 · · · dφn−2 dθ
Ω0
Z 1 Z
=n rn−1 dr f (F (φ1 , . . . , φn−2 , θ))| sinn−2 (φ1 ) . . . sin(φn−2 )| dφ1 · · · dφn−2 dθ
0
Z0 ∞ Z Ω

=n g(rF (φ1 , . . . , φn−2 , θ))rn−1 | sinn−2 (φ1 ) . . . sin(φn−2 )| dφ1 · · · dφn−2 dθ dr


0
Z0 Ω

= n g(G(r, φ1 , . . . , φn−2 , θ))| det(D(r,φ1 ,...,φn−2 ) G)| dφ1 · · · dφn−2 dθ dr


ZΩ
=n g
G(Ω)
Z
= f.
S n−1

24

You might also like