Real Analysis Chapter 8 Solutions Jonathan Conder

3. (a) Note that η (0) (t) = 1 · e−1/t for all t ∈ (0, ∞), where 1 is a polynomial of degree 0. Given k ∈ N, suppose
that η (k−1) (t) = Pk−1 (1/t)e−1/t for all t ∈ (0, ∞), where Pk−1 (x) is some polynomial of degree 2(k − 1). By
0
the product rule and the chain rule, η (k) (t) = Pk−1 (1/t)(−t−2 )e−1/t + Pk−1 (1/t)e−1/t t−2 = Pk (1/t)e−1/t for all
t ∈ (0, ∞), where Pk (x) := x2 (Pk−1 (x) − Pk−10 (x)) is a polynomial of degree 2 + 2(k − 1) = 2k. It follows by
induction that this result holds for all k ∈ N ∪ {0}.
(b) By definition η (0) (0) = 0. Given k ∈ N, suppose that η (k−1) (0) = 0. Note that
1
(n + 1)e−(n+1) 1+ n 1
lim = lim = < 1,
n→∞ ne−n n→∞ e e
P∞ −n
so that n=1 ne converges by the ratio test. This implies that limn→∞ ne−n = 0, and hence

η (k−1) (t) − η (k−1) (0) Pk−1 (t)e−1/t e−1/t

lim = lim = Pk−1 (0) lim = 0.
t&0 t t&0 t t&0 t

The left-hand derivative of η (k−1) at 0 is clearly 0, so η (k) (0) = 0. By induction, η (k) (0) = 0 for all k ∈ N.

4. This question is surprisingly subtle, and one of the few instances where it is important to keep track of null sets
explicitly. It can be shown that Ar f is uniformly continuous for arbitrary f ∈ L∞ and r ∈ (0, ∞), but the δ
constructed depends on r, which will cause issues later. Instead, let ε ∈ (0, ∞) and take δ ∈ (0, ∞) such that
kτy f − f k∞ < 2ε for all y ∈ Bδ (0). If r ∈ (0, ∞) and x, y ∈ Rn satisfy |x − y| < δ, then
Z Z
1 1
|Ar f (x) − Ar f (y)| = f (z) dz − f (z) dz
m(Br (x)) Br (x) m(Br (y)) Br (y)
Z Z
1 1
= f (z + x − y) dz − f (z) dz
m(Br (y)) Br (y) m(Br (y)) Br (y)
Z
1
≤ |τy−x f (z) − f (z)| dz
m(Br (y)) Br (y)
≤ kτy−x f − f k∞
ε
< .
2
In particular, Ar f is uniformly continuous. If x ∈ Rn and r, s ∈ (0, δ), then the above inequality gives
Z Z
1 1
|Ar f (x) − As f (x)| = Ar f (x) − As f (y) dy + As f (y) dy − As f (x)
m(Br (x)) Br (x) m(Br (x)) Br (x)
Z Z
1 1
≤ (f (y) − As f (y)) dy + |As f (y) − As f (x)| dy
m(Br (x)) Br (x) m(Br (x)) Br (x)
Z Z ! Z
1 1 1 ε
< f (y) − f (z) dz dy + dy
m(Br (x)) Br (x) m(Bs (y)) Bs (y) m(Br (x)) Br (x) 2
Z Z
1 ε
≤ |f (y) − f (z)| dz dy +
m(Br (x))m(Bs (y)) Br (x) Bs (y) 2
Z Z
1 ε
= |f (y) − τ−z f (y)| dz dy +
m(Br (x))m(Bs (y)) Br (x) Bs (0) 2
Z Z
1 ε
= |f (y) − τ−z f (y)| dy dz +
m(Br (x))m(Bs (y)) Bs (0) Br (x) 2

1
 Real Analysis Chapter 8 Solutions Jonathan Conder

Z Z
1 ε
≤ kτ−z f − f k∞ dy dz +
m(Br (x))m(Bs (y)) Bs (0) Br (x) 2
< ε.

Therefore (A1/n f )∞
n=1 is uniformly Cauchy, so it converges uniformly to a function g which is uniformly continuous
(by a standard argument). Theorem 3.18 implies that f = g almost everywhere.

7. Choose R ∈ (0, ∞) so that g is supported on BR (0). The partial derivatives of g (up to order k) are also supported on
BR (0), and hence they are bounded. If x ∈ Rn and y ∈ B1 (x) then f ∗ g(y) = f (z)g(y − z) dz = (f χBR+1 (x) ) ∗ g(y),
R

so (f ∗ g)|B1 (x) ∈ C k (B1 (x)) by Proposition 8.10. It follows that f ∗ g ∈ C k (Rn ).

8. If x ∈ Rn , ej is the j th standard basis vector in Rn and t ∈ R is non-zero then

(f ∗ g)(x + tej ) − (f ∗ g)(x)

0≤ − (∂j f ) ∗ g
t
Z Z
−1
= t (f (x + tej − y) − f (x − y))g(y) dy − (∂j f )(x − y)g(y) dy
Z
= (t−1 ((τ−tej f )(x − y) − f (x − y)) − (∂j f )(x − y))g(y) dy
Z
≤ |t−1 ((τ−tej f )(x − y) − f (x − y)) − (∂j f )(x − y)||g(y)| dy
Z 1/p
−1 p
≤ |t ((τ−tej f )(x − y) − f (x − y)) − (∂j f )(x − y)| dy kgkq
Z 1/p
−1 p
= |t ((τ−tej f )(y) − f (y)) − (∂j f )(y)| dy kgkq

= kt−1 (τ−tej f − f ) − ∂j f kp kgkq

by Hölder’s inequality, so by the squeeze theorem ∂j (f ∗ g) = (∂j f ) ∗ g.

9. Suppose that f is absolutely continuous on every bounded interval and that f 0 ∈ Lp . If t, x ∈ R and t 6= 0 then

(t−1 (τ−t f − f ) − f 0 )(x) = t−1 ((τ−t f )(x) − f (x)) − f 0 (x)

= t−1 (f (x + t) − f (x)) − f 0 (x)
Z x+t
= t−1 f 0 (y) dy − f 0 (x)
x
Z t
= t−1 (f 0 (x + y) − f 0 (x)) dy
Z0
= |t|−1 (f 0 (x + y) − f 0 (x)) dy
[t,0]∪[0,t]

and hence, by Minkowski’s inequality for integrals

Z Z 1/p
−1 0 −1 0 0 p
kt (τ−t f − f ) − f kp ≤ |t (f (x + y) − f (x))| dx dy
[t,0]∪[0,t]
Z Z 1/p
−1 0 0 p
= |t| |(τ−y f )(x) − f (x)| dx dy
[t,0]∪[0,t]

2
 Real Analysis Chapter 8 Solutions Jonathan Conder

Z
−1
= |t| kτ−y f 0 − f 0 kp dy.
[t,0]∪[0,t]

Given ε ∈ (0, ∞), since f 0 ∈ Lp there exists δ ∈ (0, ∞) such that kτ−y f 0 − f 0 kp < ε for all y ∈ (−δ, δ). In particular,
if |t| < δ then [t,0]∪[0,t] kτ−y f 0 − f 0 kp dy ≤ [t,0]∪[0,t] ε dy = ε|t|, in which case kt−1 (τ−t f − f ) − f 0 kp ≤ ε. This shows
R R

that f 0 is the Lp derivative of f.

Conversely, suppose that f has an Lp derivative h. Define φ ∈ Cc (X) by φ(x) := (1 − |x|)χ[−1,1] (x). Clearly 0 ≤
φ ≤ 4(1 + |x|)−2 and φ = 1, so hf ∗ φ1/n i∞
R
n=1 converges to f pointwise on Lf . If a, b ∈ Lf and a < b then
h ∈ L ([a, b]) ⊆ L ([a, b]), and the previous exercise implies that h(f ∗ φ1/n )0 i∞
p 1 ∞
n=1 = hh ∗ φ1/n in=1 . This sequence
converges to h in L1 ([a, b]). Moreover f ∗ φ1/n is absolutely continuous for all n ∈ N, because
Z Z 1/n
0
|(f ∗ φ1/n ) (x)| = |(h ∗ φ1/n )(x)| ≤ |h(x − y)φ1/n (y)| dy = |h(x − y)|nφ(ny) dy ≤ nkhkL1 [a−1/n,b+1/n]
−1/n

for all x ∈ [a, b] (so its derivative is bounded on [a, b]). It follows that
Z x Z x
f (x) − f (a) = lim ((f ∗ φ1/n )(x) − (f ∗ φ1/n )(a)) = lim h ∗ φ1/n = h
n→∞ n→∞ a a

for all x ∈ Lf ∩ [a, b]. By redefining f on a null set, it follows that f ∈ AC([a, b]) and f 0 = h almost everywhere
on [a, b]. Given an interval with endpoints not in Lf , there is a larger interval whose endpoints do lie in Lf , so f is
absolutely continuous, with f 0 = h, on every bounded interval. This implies that f 0 = h almost everywhere, in which
case f 0 ∈ Lp .

10. Let x ∈ Rn and t ∈ (0, ∞). Set E0 := Bt (0) and for each k ∈ N define Ek := B2k t (0) \ B2k−1 t (0). Then
Z Z
|f (x − y)φt (y)| dy = |f (x − y)|t−n |φ(t−1 y)| dy
E0 E0
Z
1
≤ n |f (x − y)|C(1 + |t−1 y|)−n−ε dy
t E0
C · m(B1 (0))
Z
≤ |f (x − y)| dy
m(Bt (0)) E0
C · m(B1 (0))
Z
= |f (y)| dy
m(Bt (x)) Bt (x)

≤ C · m(B1 (0)) · Hf (x).

Moreover, if k ∈ N then
Z Z
|f (x − y)φt (y)| dy = |f (x − y)|t−n |φ(t−1 y)| dy
Ek Ek
Z
1
≤ n |f (x − y)|C(1 + |t−1 y|)−n−ε dy
t Ek
Z
C
≤ n |f (x − y)|(1 + 2k−1 )−n−ε dy
t Ek
C · 2kn m(B1 (0))
Z
≤ (k−1)(n+ε) |f (x − y)| dy
2 · m(B2k t (0)) B2k t (0)
C · m(B1 (0))
Z
= (k−1)ε−n |f (y)| dy
2 · m(B2k t (x)) B2k t (x)

3
 Real Analysis Chapter 8 Solutions Jonathan Conder

C · m(B1 (0))
≤ Hf (x).
2(k−1)ε−n
It follows that
∞
!
2n
Z X  
n−(k−1)ε
|f ∗ φt (x)| ≤ |f (x − y)φt (y)| dy ≤ C · m(B1 (0)) 1 + 2 Hf (x) = C · m(B1 (0)) 1 + Hf (x)
1 − 2−ε
k=1
2n
and hence Mφ f ≤ C · m(B1 (0))(1 + )Hf.
1−2−ε
R1
13. (a) Clearly fb(0) = 0 ( 12 − x) dx = 0. If κ ∈ Z \ {0} then we integrate by parts:
 −2πiκx 1 Z 1 −2πiκx 1
e−2πiκx
Z 1  
1 −2πiκx 1 e e 1+1 1
f (κ) =
b −x e dx = −x + dx = + = .
0 2 2 −2πiκ 0 0 −2πiκ 4πiκ −4π 2 κ2 0 2πiκ

(b) By Plancherel’s theorem

∞ ∞ Z 1 2 3 1
2π 2 1 π2

X 1 2
X
b(k)|2 = 2π 2
X
b(κ)|2 = 2π 2 kf k2 = 2π 2 1
= 4π |f |f − x dx = − − x = .
k2 0 2 3 2 6
k=1 k=1 κ∈Z 0

15. (a) Let a ∈ (0, ∞), and note that χ[−a,a] ∈ L1 . If ξ ∈ R \ {0} then
x=a
e−2πiξx e2πiξa − e−2πiξa
Z
2ia sin(2πξa)
b[−a,a] (ξ) = e−2πiξx χ[−a,a] (x) dx =
χ = = = 2a sinc(2ξa),
−2πiξ x=−a 2πiξ 2πiξa

b[−a,a] (0) = χ[−a,a] = 2a = 2a sinc(2a · 0). It follows that χ∨

R
and it is clear that χ [−a,a] (ξ) = χb[−a,a] (−ξ) =
2a sinc(−2ξa) = 2a sinc(2ξa) for all ξ ∈ R.
(b) Since the Fourier transform is linear, it is clear that Ha is a subspace of L2 . If hfn i∞ n=1 is a sequence in Ha
2 ∞ 2
which converges to f ∈ L , then hfn in=1 converges to f in L (because the Fourier transform is unitary), and
c b
some subsequence of hfc ∞
n in=1 converges to f pointwise almost everywhere (because it converges in measure). In
b
particular fb(ξ) = 0 for almost all ξ ∈ R \ [−a, a], which implies that f ∈ Ha . This shows that Ha is a closed
subspace of L2 , so it is a Hilbert space.
Define φ := (2a)−1/2 χ[−a,a] and f := φ. b If ψ ∈ S then F2 ψ is the reflection of (ψ)
b ∨ = ψ about 0. Therefore F2
agrees with the reflection operator on a dense subset of L2 , so the two operators are equal and fb is the reflection
of φ about 0. This shows that fb = φ, so f ∈ Ha . For each k ∈ Z define fk := τk/2a f, so that fbk (ξ) = e−πiξk/a fb(ξ)
for all ξ ∈ R, and in particular fk ∈ Ha . If j, k ∈ Z then
Z Z Z a
−πiξj/a b πiξk/a b πiξ(k−j)/a 2 1
hfj , fk i = hfj , fk i = e
b b f (ξ)e f (ξ) dξ = e |φ(ξ)| dξ = eπiξ(k−j)/a dξ,
2a −a
1 1
Ra
and if j 6= k it follows that hfj , fk i = 2πi(k−j) (eπi(k−j) − eπi(j−k) ) = 0. Moreover, hfk , fk i = 2a −a 1 = 1. This
shows that {fk }k∈Z is an orthonormal set in Ha . If g ∈ Ha and g ⊥ fk for all k ∈ Z, then
Z a √ Z √ √
gb(ξ)eπiξk/a dξ = 2a gb(ξ)eπiξk/a φ(ξ) dξ = 2ahb g , fbk i = 2ahg, fk i = 0
−a

for all k ∈ Z. This implies that gb|[−a,a] ∈ M⊥ , where M is the closed span of the collection of functions of the
form ξ 7→ e−πiξk/a (where k ∈ Z). But M = L2 ([−a, a]) by the Stone-Weierstraß theorem and the fact that the
inclusion C([−a, a]) ,→ L2 ([−a, a]) is a bounded linear map with dense range (this is essentially Theorem 8.20).
Therefore gb|[−a,a] = 0 almost everywhere, so gb = 0 almost everywhere and hence g = 0 almost everywhere. This
shows that {fk }k∈Z is an orthonormal basis for Ha .

4
 Real Analysis Chapter 8 Solutions Jonathan Conder

(c) Given f ∈ Ha , the series k∈Z hf, fk ifk = k∈Z hfb, fbk ifk converges to f in Ha (thus in L2 ). If k ∈ Z then
P P

F2 f (−k/2a)
Z Z a πixk/a Z
e 1 f (k/2a)
hfb, fbk i = fb(x)eπixk/a φ(x) dx = √ fb(x) dx = √ eπixk/a fb(x) dx = √ = √ .
−a 2a 2a 2a 2a
Let Z ⊆ Z. By Hölder’s inequality
X sX X sX X
|hf, fk ifk (ξ)| ≤ |hf, fk i|2 2
|fk (ξ)| ≤ |hf, fk i|2 2a| sinc(2aξ − k)|2
k∈Z k∈Z k∈Z k∈Z k∈Z

for all ξ ∈ R. Parseval’s identity shows that the first sum k∈Z |hf, fk i|2 can be made small by choosing Z
P
P
appropriately, so the series k∈Z hf, fk ifk will be uniformly Cauchy provided that the second sum is uniformly
1
bounded. It suffices to show this for ξ ∈ [0, 2a ) because
X X X
2a| sinc(2a(ξ + i/2a) − k)|2 = 2a| sinc(2aξ + i − k)|2 = 2a| sinc(2aξ − j)|2
k∈Z k∈Z j∈Z
1
for all ξ ∈ R and i ∈ Z. Given ξ ∈ [0, 2a ) define Z := Z \ {−1, 0, 1}, so that (since | sinc | ≤ 1)
X X | sin(2πaξ)|2
2a| sinc(2aξ − k)|2 ≤ 6a + 2a
|2πaξ − πk|2
k∈Z k∈Z
X 2a
≤ 6a +
|2πaξ − πk|2
k∈Z
X 2a
≤ 6a +
(π|k| − 2πaξ)2
k∈Z
X 2a
≤ 6a +
(π|k| − π)2
k∈Z
X 2a
≤ 6a +
(π|k|/2)2
k∈Z
X 8a
= 6a +
π 2 |k|2
k∈Z
∞
16a X 1
= 6a +
π2 k2
k=2

< ∞.
P
This shows that k∈Z f (k/2a) sinc(2aξ − k) is uniformly Cauchy for all ξ ∈ R. It is clear that sinc is continuous,
and in fact sinc ∈ C0 because | sinc(ξ)| ≤ |πξ|−1 for all ξ ∈ R \ {0}. Hence, the above series converges uniformly
to some g ∈ C0 . A subsequence of its partial sums converges pointwise almost everywhere to f (because the
series converges to f in measure). This implies that f = g almost everywhere.

16. (a) If k ∈ N and x ∈ R then


0,

 if x ∈ (−∞, −k − 1] ∪ [k + 1, ∞)


if x ∈ [1 − k, k − 1]
Z Z 
2,
fk (x) = χ[−1,1] (x − y)χ[−k,k] (y) dy = χ[x−1,x+1]∩[−k,k] =


k+1−x if x ∈ [k − 1, k + 1]


k + 1 + x if x ∈ [−k − 1, 1 − k].


It clearly follows that kfk ku = 2.

5
 Real Analysis Chapter 8 Solutions Jonathan Conder

(b) If k ∈ N and ξ ∈ R then

4k
fk∨ (ξ) = fbk (−ξ) = χ χ[−k,k] (−ξ) = 2 sinc(−2ξ) · 2k sinc(−2ξk) =
b[−1,1] (−ξ)b sin(2πξ) sin(2πkξ).
(πξ)2
Therefore
48k 2 | sin(πζ/6k) sin(πζ/6)|
Z Z Z
4k| sin(2πξ) sin(2πkξ)| 4| sin(πζ/6k) sin(πζ/6)|
kfk∨ k1 = dξ = dζ = dζ.
(πξ)2 12(πζ/12k)2 π2ζ 2

Note that | sin(πζ/6)| ≥ 12 for all ζ ∈ ∪i∈Z ([1, 5] + 6i) and (provided that k ≥ 6) at least m := d k2 e of these
intervals are contained in [k, 5k]. Indeed, at least one of them begins in [k, k + 5], after which there is an available
length of 4k − 5 ≥ 6m − 2, which is enough room for m such intervals. Hence, there exists a ∈ 6Z such that
∪m−1
i=0 [ai , bi ] ⊆ [k, 5k], where ai := 1 + a + 6i and bi := ai + 4 for all i ∈ {0, . . . , m − 1}. Therefore

m−1
X Z bi m−1  m−1
X (bi − ai ) m−1
12k 2 X 12k 2  1 1 X 4k 2 4m 2k
kfk∨ k1 ≥ 2 2
dζ = 2
− ≥ k 2
≥ 2
= ≥ ,
ai π ζ π ai bi ai bi (5k) 25 25
i=0 i=0 i=0 i=0

which implies that limk→∞ kfk∨ k1 = ∞.

(c) If f ∈ L1 and fb = 0 then fb ∈ L1 and hence f = (fb)∨ = 0 almost everywhere. In particular F|L1 is injective (it
has trivial kernel). Moreover F|L1 ∈ L(L1 , C0 ) because |fb(ξ)| = | e−2πiξx f (x) dx| ≤ kf k1 for all f ∈ L1 and
R

ξ ∈ R, by definition. If F(L1 ) = C0 , then F|L1 has an inverse G ∈ L(C0 , L1 ) by the open mapping theorem (since
L1 and C0 are Banach spaces). This contradicts the previous parts, because kGk ≥ kGf k k1 k
kfk ku ≥ 25 for all k ∈ N
with k ≥ 6.

17. (a) Note that

Z Z ∞ Z ∞ Z ∞
−2πx a−1 2π −2πx 1 Γ(a)
|f (x)| dx = e x dx = e (2πx) a−1
dx = e−t ta−1 dt =
0 (2π)a 0 (2π)a 0 (2π)a

and hence f ∈ L1 . Similarly, if a > 12 then f ∈ L2 because

Z ∞ Z ∞ Z ∞
(4π)2 Γ(2a − 1)
Z
−4πx 2a−2 −4πx 4π
2
|f (x)| dx = e x dx = 2a
e (4πx) 2a−2
dx = 2a
e−t t2a−2 dt = .
0 (4π) 0 (4π) 0 (4π)2a−1

(b) Given ξ ∈ R, define γ : R → C by γ(x) := 2π(iξ + 1)x. Note that

Z Z ∞ Z ∞
−2πiξx −2π(iξ+1)x a−1 2π(iξ + 1)
f (ξ) = e
b f (x) dx = e x dx = a
e−2π(iξ+1)x (2π(iξ + 1)x)a−1 dx.
0 (2π(iξ + 1)) 0

By part (a) x 7→ e−2πiξx f (x) is in L1 , so the dominated convergence theorem implies that
Z ∞ Z N
e−2π(iξ+1)x (2π(iξ + 1)x)a−1 dx = lim e−2π(iξ+1)x (2π(iξ + 1)x)a−1 dx
0 N →∞ 0 Z
1
= lim e−z z a−1 dz.
2π(iξ + 1) N →∞ γ|[0,N ]

Given N ∈ N, define γ1 : [0, 2πN ] → C by γ1 (x) := x and γ2 : [0 : 2πN ] by γ2 (x) := 2πN + iξx. Clearly γ1
followed by γ2 is homotopic to γ|[0,N ] , so by Cauchy’s theorem
Z Z Z
−z a−1 −z a−1
e z dz = e z dz + e−z z a−1 dz.
γ|[0,N ] γ1 γ2

6
 Real Analysis Chapter 8 Solutions Jonathan Conder

R 2πN −x a−1
Note that limN →∞ γ1 e−z z a−1 dz = limN →∞ 0
R
e x dx = Γ(a), whereas
Z Z 2πN Z 2πN
−z a−1 −(2πN +iξx) a−1 −2πN
e z dz ≤ |e (2πN + iξx) iξ| dx = e |ξ| |2πN + iξx|a−1 dx.
γ2 0 0

If a ≥ 1 then |2πN + iξx|a−1 ≤ (2πN + |ξ|x)a−1 ≤ (2πN (1 + |ξ|))a−1 for all x ∈ [0, 2πN ]. Otherwise |2πN +
iξx|a−1 = (4π 2 N 2 + |ξ|2 x2 )(a−1)/2 ≤ (2πN )a−1 for all x ∈ [0, 2πN ]. In either case, the above integral exhibits
polynomial growth as N → ∞ (which is slower than exponential decay), so
Z Z Z
−z a−1 −z a−1
lim e z dz = lim e z dz + lim e−z z a−1 dz = Γ(a) + 0 = Γ(a).
N →∞ γ| N →∞ γ1 N →∞ γ2
[0,N ]

Combining the above calculations, it follows that fb(ξ) = (2π(iξ + 1))−a Γ(a).
(c) Define g : R → R in a similar manner to f but with b instead of a. By part (a) both f and g are in L2 , and
Z Z
−a −b
(1 − ix) (1 + ix) dx = (2π) a+b
(2π(1 + ix))−a (2π(1 + ix))−b dx
Z
= (2π) a+b
fb(x)Γ(a)−1 gb(x)Γ(b)−1 dx

(2π)a+b
= (b
g , fb)
Γ(a)Γ(b)
by part (b). Since the Fourier transform is unitary (from L2 to L2 ),
Z Z ∞ Z ∞ Z
2πx a−1 2πx b−1 2πx 2 a+b−2
g , f ) = (g, f ) = f (x)g(x) dx =
(b b e x e x dx = (e ) x dx = |h(x)|2 dx
0 0

where h : R → R is defined in a similar manner to f but with a+b

2 instead of a. By part (a), it follows that

(2π)a+b Γ(a + b − 1) 22−a−b πΓ(a + b − 1)

Z
(1 − ix)−a (1 + ix)−b dx = = .
(4π)a+b−1 Γ(a)Γ(b) Γ(a)Γ(b)

19. Since {x ∈ Rn | f (x) 6= 0} has finite measure, f ∈ L1 and hence

Z
f (ξ) = e−2πiξ·x f (x) dx
b

for all ξ ∈ Rn . Given a measurable set E ⊆ Rn , the Minkowski and Hölder inequalities imply that
Z Z Z 2
|fb|2 = e−2πiξ·x f (x) dx dξ
E E
Z Z 2
e−2πiξ·x f (x)
p
= χE (ξ) dx dξ

Z sZ !2
≤ |e−2πiξ·x f (x)χE (ξ)|2 dξ dx

Z sZ !2
= |f (x)| χE (ξ) dξ dx
 Z 2
p
= m(E) |f (x)| dx
S
≤ m(E)(kf k2 kχS k2 )2
= kf k22 m(S)m(E).

7
 Real Analysis Chapter 8 Solutions Jonathan Conder

22. (a) Let ξ, ζ ∈ Rn and suppose that |ξ| = |ζ|. If |ξ| = 0 then ξ = 0 = ζ and hence J(ξ) = J(ζ). Otherwise, set
α := |ξ|−1 , and note that {αξ} and {αζ} can be extended to orthonormal bases for Rn . Let R ∈ O(n) be
the change of basis matrix between these bases, and note that R(αξ) = αζ, implying that Rξ = ζ. It follows
−1
that J(ζ) = J(Rξ) = S eix·(Rξ) dσ(x) = S ei(R x)·ξ dσ(x) = | det(R−1 )|−1 S eix·ξ dσ(x) = J(ξ). Therefore J(ξ)
R R R

depends only on |ξ|. By Theorem 2.49

Z
Fb(ξ) = e−2πiξ·x f (|x|) dx
Z ∞Z
= e−2πiξ·(rx) f (|rx|)rn−1 dσ(x) dr
Z0 ∞ ZS
= ei(2πrξ)·(−x) dσ(x)f (r)rn−1 dr
0 S
Z ∞
= | − 1|−1 J(2πrξ)f (r)rn−1 dr
0
Z ∞
= j(2πr|ξ|)f (r)rn−1 dr,
0
R∞ n−1 dr for all s ∈ [0, ∞).
so (Fb is radial and) g(s) = 0 j(2πrs)f (r)r
(b) If k ∈ {1, . . . , n} then ∂ξ∂k eix·ξ = ixk eix·ξ and hence | ∂ξ∂k eix·ξ | = |xk | ≤ 1 for all x ∈ S and ξ ∈ Rn . Moreover
∂ 2 ix·ξ ∂ 2 ix·ξ
= (ixk )2 eix·ξ = −x2k eix·ξ , so | ∂ξ | = |xk |2 ≤ 1 for all x ∈ S and ξ ∈ Rn . Since 1 dσ = σ(S) < ∞,
R
∂ξk2
e 2e
k
the dominated convergence theorem implies that
∂ 2 ix·ξ
Z Z Z
∂ ∂ ix·ξ
(∂k2 J)(ξ) = e dσ(x) = 2 e dσ(x) = −x2k eix·ξ dσ(x)
∂ξk S ∂ξk S ∂ξ k S

for all ξ ∈ Rn , in which case nk=1 ∂k2 J + J = 0. Indeed, if ξ ∈ Rn then

P

n n n
Z ! Z ! Z
X X X
2 2 ix·ξ
(∂k J)(ξ) = −xk e dσ(x) = − 2
xk e dσ(x) = − |x|2 eix·ξ dσ(x) = −J(ξ).
ix·ξ

k=1 S k=1 S k=1 S

(c) If k ∈ {1, . . . , k} and ξ ∈ Rn is non-zero then ∂

∂ξk |ξ| = ∂
∂ξk (ξ · ξ)1/2 = 12 (ξ · ξ)−1/2 (2ξk ) = ξk |ξ|−1 , so that

∂ ξk
(∂k J)(ξ) = j(|ξ|) = j 0 (|ξ|)
∂ξk |ξ|
and hence
ξ2 |ξ| − ξk2 |ξ|−1
 
∂ 0 ξk
(∂k2 J)(ξ) = j (|ξ|) = j 00 (|ξ|) k2 + j 0 (|ξ|) .
∂ξk |ξ| |ξ| |ξ|2
This implies that
n n n n
X
00
X ξk2 0
X |ξ| 0
X ξk2
∂k2 J(ξ) + J(ξ) = j (|ξ|) + j (|ξ|) − j (|ξ|) + j(|ξ|)
|ξ|2 |ξ|2 |ξ|3
k=1 k=1 k=1 k=1
 
00 0 n 1
= j (|ξ|) + j (|ξ|) − + j(|ξ|)
|ξ| |ξ|
= j 00 (|ξ|) + (n − 1)j 0 (|ξ|)|ξ|−1 + j(|ξ|),

which is zero by part (b), in which case |ξ|j 00 (|ξ|) + (n − 1)j 0 (|ξ|) + |ξ|j(|ξ|) = 0, for all ξ ∈ Rn \ {0}. Moreover
(∂1 J)(0, . . . , 0) = j 0 (0) = −(∂1 J)(0, . . . , 0), since J(r, 0, . . . , 0) = j(r) = J(−r, 0, . . . , 0) for all r ∈ [0, ∞), so
j 0 (0) = 0 and hence rj 00 (r) + (n − 1)j 0 (r) + rj(r) = 0 for all r ∈ [0, ∞).

8
 Real Analysis Chapter 8 Solutions Jonathan Conder

(d) Define f : [0, ∞) → C by f (r) := rj(r), so that f 0 (r) = j(r) + rj 0 (r) and hence f 00 (r) = 2j 0 (r) + rj 00 (r) for
all r ∈ [0, ∞). It follows from part (c) that f 00 + f = 0, which is a second order ODE satisfied by the function
g : [0, ∞) → R defined by g(x) := σ(S) sin(x). Since f 0 (0) = j(0) = S 1 dσ = σ(S) = g 0 (0) and f (0) = 0 = g(0),
R

it follows that f = g, in which case j(r) = r−1 f (r) = σ(S)r−1 sin(r) = 4πr−1 sin(r) for all r ∈ (0, ∞). Since
j(0) = σ(S) = 4π, this statement can be interpreted for r = 0 by taking limits.

23. (a) Define linear operators P, Q on S(R) by P f (x) := f 0 (x) and Qf (x) := xf (x). If f, g ∈ S(R) then
Z Z Z Z
(Qf )g = xf (x)g(x) dx = f (x)xg(x) dx = f (Qg).

Moreover, (f g)0 = f 0 g + f g 0 , and (by the monotone convergence theorem applied separately to the positive and
negative real and imaginary parts of f 0 g, followed by the fundamental theorem of calculus)
Z Z N
0
(f g) = lim (f g)0 = lim (f (N )g(N ) − f (−N )g(−N )) = 0.
N →∞ −N N →∞

It follows that Z Z Z Z Z
0 0
(P f )g = fg= (f g) − f g0 =− f (P g),

in which case
√ Z Z Z Z Z Z Z √ Z
2 (T f )g = (Qf − P f )g = (Qf )g − (P f )g = f (Qg) + f (P g) = f (Qg + P g) = 2 f (T ∗ g).

f (T ∗ g). If x ∈ R then (P Qf )(x) = (Qf )0 (x) = f (x) + xf 0 (x) = f (x) + (QP f )(x), so
R R
Thus (T f )g =

2[T ∗ , T ] = [Q + P, Q − P ] = [Q, Q] − [Q, P ] + [P, Q] − [P, P ] = 2[P, Q] = 2I.

This implies that [T ∗ , T ] = I = T 0 . If k ∈ N with k > 1 and [T ∗ , T k−1 ] = (k − 1)T k−2 , then

T ∗ T k = T ∗ T k−1 T = [T ∗ , T k−1 ]T + T k−1 T ∗ T = (k − 1)T k−1 + T k−1 [T ∗ , T ] + T k−1 T T ∗ = kT k−1 + T k T ∗ ,

in which case [T ∗ , T k ] = kT k−1 . By induction, this shows that [T ∗ , T k ] = kT k−1 . for all k ∈ N.
√ √
(b) If k ∈ N then T hk = (k!)−1/2 T k+1 h0 = k + 1((k + 1)!)−1/2 T k+1 h0 = k + 1hk+1 , and since
√ 2 /2 2 /2
2(T ∗ h0 )(x) = xh0 (x) + h00 (x) = π −1/4 xe−x + π −1/4 (−x)e−x = 0,
√
part (a) implies that T ∗ hk = (k!)−1/2 T ∗ T k h0 = (k!)−1/2 ([T ∗ , T k ] + T k T ∗ )h0 = (k!)−1/2 kT k−1 h0 = khk−1 .
These formulae also hold for k = 0 if we let h−1 be an arbitrary function, and it clearly follows that T T ∗ hk =
√ √ √ √
T ( khk−1 ) = kT hk−1 = k khk = khk for all k ∈ N ∪ {0}.
(c) Note that S = 2T T ∗ + I = (Q − P )(Q + P ) + I = Q2 + [Q, P ] − P 2 + [P, Q] = Q2 − P 2 . Hence, if f ∈ S(R) then
Sf (x) = Q2 f (x) − P 2 f (x) = x2 f (x) − f 00 (x) for all x ∈ R. Moreover if k ∈ N ∪ {0} then

Shk = 2T T ∗ hk + hk = 2khk + hk = (2k + 1)hk .

−x 2 √
(d) Note that kh0 k22 = |h0 |2 = e√ √π
R R
π
dx = π
= 1. If k ∈ N and khk−1 k2 = 1, then
Z Z Z Z √ √ Z
−1 ∗ −1 ∗ −1
khk k22 = hk hk = k (T T hk )hk = k (T hk )(T ∗ h k) =k khk−1 khk−1 = |hk−1 |2 = 1,

9
Real Analysis Chapter 8 Solutions Jonathan Conder

so by induction khk k = 1 for all k ∈ N ∪ {0}. If j, k ∈ N ∪ {0} and j > k then

s
Z
−1
Z
∗ −1
Z
∗ −1
Z p √ k
(hj , hk ) = hj hk = j (T T hj )hk = j (T hj )(T ∗ hk ) = j jhj−1 ( khk−1 ) = (hj−1 , hk−1 )
j
q
k(k−1)···0
so by induction (hj , hk ) = j(j−1)···(j−k) (hj−k−1 , h−1 ) = 0. This shows that {hk }∞
k=0 is orthonormal.

(e) If k ∈ N and
dk−1 −x2 /2 2 /2
T k−1 f (x) = (−1)k−1 2(1−k)/2 ex (e f (x)),
dxk−1
for all x ∈ R (which is clearly true when k = 1), the product rule implies that
k−1 k−1 k
 
k k−1 −k/2 x2 /2 d −x2 /2 x2 /2 d −x2 /2 x2 /2 d −x2 /2
T f (x) = (−1) 2 xe (e f (x)) − xe (e f (x)) − e (e f (x))
dxk−1 dxk−1 dxk
2 dk 2
= (−1)k 2−k/2 ex /2 k (e−x /2 f (x)).
dx
2 k 2
for all x ∈ R. Therefore T k f (x) = (−1)k 2−k/2 ex /2 dx
d
k (e
−x /2 f (x)) for all x ∈ R and k ∈ N ∪ {0}, by induction.

If k ∈ N ∪ {0} then hk = k −1/2 T hk−1 = · · · = (k!)−1/2 T k h0 , so for all x ∈ R

2 /2 dk −x2 /2 (−1)k x2 /2 d
k
2
hk (x) = (k!)−1/2 (−1)k 2−k/2 ex (e h0 (x)) = √ e e−x .
dxk 1/2 k
π 2 k! dx k

k 2 2
d
(f) Given k ∈ N ∪ {0}, it is easily shown by induction and the product rule that dx −x = P (x)e−x for all x ∈ R,
ke k
where Pk (x) is some polynomial of degree k. The formula for hk from part (e) implies that

(−1)k 2 2 (−1)k
Hk (x) = √ ex Pk (x)e−x = √ Pk (x),
π 1/2 2k k! π 1/2 2k k!
so Hk (x) is also a polynomial of degree k. In particular H0 (x) is a non-zero constant, so all the constant
k−1
polynomials are in span{H0 (x)}. If the polynomials of degree less than k are in span{Hj (x)}j=0 and c ∈ R \ {0}
−1 k−1
is the leading term of Hk (x), then x −c Hk (x) ∈ span{Hj (x)}j=0 , which implies that x ∈ span{Hj (x)}kj=0 and
n n

hence every polynomial of degree at most k is in span{Hj (x)}kj=0 . By induction, this shows that span{Hj (x)}kj=0
is the set of polynomials of degree at most k, for all k ∈ N ∪ {0}.
2 /2
(g) Let f ∈ L2 and suppose that f ⊥ hk for all k ∈ N ∪ {0}. Define g : R → C by g(x) := f (x)e−x , so that g ∈ L1
(by Hölder’s inequality). If ξ, x ∈ R and N ∈ N then
N N
X (−2πiξx)k X |2πξx|k 2 /2 2 /2
g(x) ≤ |f (x)|e−x ≤ e2π|ξx|−x |f (x)|.
k! k!
k=0 k=0
2 /2 2 /2
If ξ ∈ R then x 7→ e2π|ξx|−x is clearly in L2 , so x 7→ e2π|ξx|−x |f (x)| is in L1 . Moreover
∞ N
(−2πiξx)k (−2πiξx)k
Z Z X Z X
−2πiξx
gb(ξ) = e g(x) dx = g(x) dx = lim g(x) dx
k! N →∞ k!
k=0 k=0

(2πiξx)k
by the dominated convergence theorem. If N ∈ N then N ∈ spanC {Hk (x)}N
P
k=0 k! k=0 , and since Hk g = f hk
1
for all k ∈ N∪{0}, it follows that gb(ξ) = limN →∞ 0 = 0. In particular gb ∈ L , so by the Fourier inversion theorem
2
g = (bg )∨ = 0 almost everywhere. Since e−x /2 > 0 for all x ∈ R, this implies that f = 0 in L2 . Therefore {hk }∞k=0
is an orthonormal basis for L2 .

10
 Real Analysis Chapter 8 Solutions Jonathan Conder

(h) Clearly A is linear and bijective (its inverse is given by A−1 f (x) := (2π)−1/4 f ((2π)−1/2 x)). If f ∈ L2 then
Z Z √ √ Z √
2 2 2 1
kAf k2 = |Af (x)| dx = 2π|f (x 2π)| dx = √ 2π|f (t)|2 dt = kf k22 ,
2π

which shows that A is unitary (by the polarisation identity). Moreover, if ξ ∈ R then (assuming f ∈ L1 )
Z
−2πiξx 1/4
√ 1/4
Z √ √
− 2πiξx 2π
√ 1
Z √
− 2πiξt
Af (ξ) = e
c (2π) f (x 2π) dx = (2π) e f (x 2π) dx = e f (t) dt,
(2π)1/4

in which case
−1/4 Z √ Z
c (ξ) = (2π)
fe(ξ) = A−1 Af e − 2πi(2π)−1/2 ξt
f (t) dt = √
1
e−iξt f (t) dt.
(2π)1/4 2π

If f ∈ S then clearly fe ∈ S, and it follows that

√ Z
−iξt
Z Z
(teiξt + iξeiξt ) 1
Z
2π Tff (ξ) = e T f (t) dt = T f (t)eiξt dt = f (t) √ dt = √ f (t)(t − iξ)e−iξt dt.
2 2
On the other hand
√
 Z Z 
i −iξt d −iξt
− 2πiT (fe)(ξ) = − √ ξ e f (t) dt − e f (t) dt .
2 dξ
d −iξt
Since | dξ e f (t)| = | − ite−iξt f (t)| = |tf (t)| for all t ∈ R, and t 7→ tf (t) is in L1 (as f ∈ S), it follows that

√
 Z Z 
i −iξt d −iξt
− 2πiT (f )(ξ) = − √
e ξ e f (t) dt − e f (t) dt
2 dξ
Z Z 
i −iξt −iξt
= −√ ξe f (t) dt + ite f (t) dt
2
Z Z 
1 −iξt −iξt
=√ te f (t) dt − iξe f (t) dt
2
Z
1
=√ (t − iξ)e−iξt f (t) dt
2
√
= 2π Tff (ξ).
√ 2
In particular Tff = −iT (fe). Note that Ah0 (x) = (2π)1/4 h0 ( 2πx) = 21/4 e−πx for all x ∈ R and hence Ah
d0 (ξ) =
1/4 −πξ 2
2 e for all ξ ∈ R (by Proposition 8.24). It follows that

d0 (ξ) = (2π)−1/4 21/4 e−π(2π)−1 ξ2 = π −1/4 eξ2 /2 = h0 (ξ)

h0 (ξ) = A−1 Ah
e

for all ξ ∈ R. If k ∈ N then hk = (k!)−1/2 T k h0 , so for all ξ ∈ R

hk (ξ) = (k!)−1/2 T]
e k h (ξ) = (k!)−1/2 (−i)T (T^
0
k−1 h )(ξ) = · · · = (k!)−1/2 (−i)k T k e
0 h0 (ξ) = (−i)k hk (ξ).

Since {hk }∞ 2 ∞ 2
k=1 is an orthonormal basis for L , its unitary image {φk }k=0 is also an orthonormal basis for L .
dk = AA−1 Ah
Moreover, for each k ∈ N ∪ {0} it is clear that φbk = Ah dk = Ae hk = A(−i)k hk = (−i)k φk .

25. (a) Let f ∈ AC(R) be periodic and suppose that f 0 ∈ Lp (T). If x, y ∈ T and x > y then
Z x Z x
0
|f (x) − f (y)| = f ≤ |f 0 | ≤ kf 0 kp kχ[y,x] kq = kf 0 kp (x − y)1/q ,
y y

11
Real Analysis Chapter 8 Solutions Jonathan Conder

in which case |f (x) − f (y)| ≤ kf 0 kp |x − y|1/q for all x, y ∈ R, because f is periodic. Therefore f ∈ Λ1/q (T).
Given α ∈ ( 1q , 1), choose β ∈ ( 1q , α) and define g : [0, 21 ] → R by g(x) := βxβ−1 . Note that g ∈ Lp ([0, 12 ]) because
(β − 1)p > ( 1q − 1)p = −1. Now define f : R → R by
Z xX
f (x) := ((τk g)χ[k.k+ 1 ) − (τk+ 1 g)χ[k+ 1 ,k+1) ),
2 2 2
0 k∈Z

so that, by definition, f is periodic and absolutely continuous on every bounded interval. Moreover
Z x
|f (x) − f (0)| 1 xβ − 0β
α
= α
g= = xβ−α
|x − 0| x 0 xα

for all x ∈ (0, 12 ), and limx&0 xβ−α = ∞, so f ∈

/ Λα (T).
(b) Let α ∈ (0, 1) and define h : [0, 1] → [0, 1] by h(t) := sin(πt). Since α < 1 the series ∞ −1/α converges. Let
P
k=1 k
S ∈ (0, ∞) be its sum and for each n ∈ N set tn := S −1 n−1 −1/α . Now define f : [0, 1] → [0, 1] by
P
k=1 k
∞
χ[tn ,tn+1 ] (t)
 
X x − tn
f (t) := h .
n tn+1 − tn
n=1

If x ∈ [0, 1] then x ∈ [tn , tn+1 ] for some n ∈ N. Given y ∈ [tn , tn+1 ], the mean value theorem implies that
   
1 x − tn y − tn
|f (x) − f (y)| = h −h
n tn+1 − tn tn+1 − tn
1 x − tn y − tn
≤ −
n tn+1 − tn tn+1 − tn
|x − y|
=
n|tn+1 − tn |
|x − y|α |tn+1 − tn |1−α
≤
n|tn+1 − tn |
|x − y|α
=
n|tn+1 − tn |α
|x − y|α
=
n(S −1 n−1/α )α
= S α |x − y|α .

Given y ∈ [tn+1 , tn+2 ], set z := tn+1 and note that

|f (x) − f (y)|1/α ≤ 21/α (|f (x) − f (z)|1/α + |f (z) − f (y)|1/α ) ≤ 21/α (S|x − z| + S|z − y|),

so |f (x) − f (y)| ≤ 2S α (|x − z| + |z − y|)α = 2S α (z − x + y − z)α = 2S α |x − y|α . If y ∈ [tn+2 , 1] then

1 1 |f (x)| + |f (y)| |f (x) − f (y)|
S α |x − y|α = (S(y − x))α ≥ (S(tn+2 − tn−1 ))α = ≥ ≥ ≥ .
n+1 2n 4 4
This shows that |f (x) − f (y)| ≤ 4S α |x − y|α for all x, y ∈ [0, 1], in which case f ∈ Λα (T). However
N   N N
X tn + tn+1 X 1 X 1
f − f (tn ) = −0 =
2 n n
n=1 n=1 n=1

for all N ∈ N, so f is not of bounded variation.

12
 Real Analysis Chapter 8 Solutions Jonathan Conder

2
30. Define Φ : Rn → R by Φ(ξ) := e−π|ξ| , so that Φ(0) = 1 and Φ∨ = Φ ∈ L1 . Note that
Z 1+|ξ| Z 1+|ξ|
dt
2n log(1 + |ξ|) ≤ 2|ξ| ≤ 2|ξ| dt = 2|ξ|2 ≤ π|ξ|2 + 2n log(1 + n)
1 t 1

for all ξ ∈ (−∞, −n) ∪ (n, ∞). Moreover 2n log(1 + |ξ|) ≤ 2n log(1 + n) ≤ π|ξ|2 + 2n log(1 + n) for all ξ ∈ [−n, n]. It
2
follows that (1 + |ξ|)2n ≤ Ceπ|ξ| , where C := (1 + n)2n , and hence |Φ(ξ)| = Φ(ξ) ≤ C(1 + |ξ|)−2n , for all ξ ∈ R. Since
0 ∈ Lf , Theorem 8.35 implies that
Z Z
−πt2 |ξ|2
lim f (ξ)e
b dξ = lim fb(ξ)Φ(tξ)e2πiξ·0 dξ = f (0),
t&0 t&0

in which case, by Fatou’s lemma,

Z Z
2 |ξ|2
kfbk1 = fb(ξ) dξ = lim fb(ξ)e−πt dξ ≤ f (0) < ∞.
t&0

13