Real Analysis

Chapter 6 Solutions

Jonathan Conder

3. Since Lp and Lr are subspaces of CX , their intersection is a vector space. It is clear that k · k is a norm (this
p
r
follows directly from the fact that k · kp and k · kr are norms). Let hfn i∞
n=1 be a Cauchy sequence in L ∩ L . Since
kfm − fn kp ≤ kfm − fn k and kfm − fn kr ≤ kfm − fn k for all m, n ∈ N, it is clear that hfn i∞
n=1 is a Cauchy sequence
p
r
p
r
in both L and L . Let gp ∈ L and gr ∈ L be the respective limits of this sequence. Given ε ∈ (0, ∞), there exists
N ∈ N such that kfn − gp kp < ε(p+1)/p for all n ∈ N with n ≥ N. If n ∈ N and n ≥ N
Z
Z
Z
p
p
p
|fn − gp | ≤ |fn − gp |p < ε(p+1)p/p = εp+1 ,
ε ≤
µ(E)ε =
E

E

so µ(E) < ε, where E := {x ∈ X | εp ≤ |fn (x) − gp (x)|p } = {x ∈ X | ε ≤ |fn (x) − gp (x)|}. This shows that hfn i∞
n=1
converges in measure to gp . If r < ∞, a similar argument shows that hfn i∞
converges
in
measure
to
g
.
Otherwise,
r
n=1
some subsequence of hfn i∞
converges
to
g
almost
everywhere,
and
this
subsequence also converges to gr almost
p
n=1

everywhere (since hfn in=1 converges uniformly to gr off a set of measure 0). In either case gp = gr , so gp ∈ Lp ∩ Lr .
p
r

p
r
Since hfn i∞
n=1 converges to gp in L and L , it easily follows that hfn in=1 converges to gp in L ∩ L . This shows that
Lp ∩ Lr is a Banach space.
Now let q ∈ (p, r). There exists λ ∈ (0, 1) such that kf kq ≤ kf kλp kf kr1−λ for all f ∈ Lp ∩Lr . In particular, if f ∈ Lp ∩Lr
and kf k = 1, then kf kp ≤ 1 and kf kr ≤ 1 and hence kf kq ≤ 1λ 11−λ = 1. This shows that the linear map Lp ∩Lr ,→ Lq
is continuous.
4. Since Lp and Lr are subspaces of CX , their sum is a vector space. Clearly kf k ≥ 0 for all f ∈ Lp + Lr . Let f ∈ Lp + Lr
and a ∈ C. If a = 0 then kaf k = k0k = 0 = |a|kf k because
0 = k0kp + k0kr ∈ {kgkp + khkr | g ∈ Lp , h ∈ Lr and f = g + h}.
Otherwise kaf k ≤ kagkp +kahkr = |a|(kgkp +khkr ) for all g ∈ Lp and h ∈ Lr with f = g+h, and hence kaf k/|a| ≤ kf k.
This implies that |a|kf k = ka−1 af k/|a−1 | ≤ kaf k, which shows that kaf k = |a|kf k. Now let f1 , f2 ∈ Lp + Lr and
ε ∈ (0, ∞). There exist g1 , g2 ∈ Lp and h1 , h2 ∈ Lr such that f1 = g1 + h1 , f2 = g2 + h2 , kg1 kp + kh1 kr < kf1 k + ε and
kg2 kp + kh2 kr < kf2 k + ε. It follows that
kf1 + f2 k ≤ kg1 + g2 kp + kh1 + h2 kr ≤ kg1 kp + kg2 kp + kh1 kr + kh2 kr < kf1 k + kf2 k + 2ε,
which shows that kf1 + f2 k ≤ kf1 k + kf2 k. Finally, let f ∈ Lp + Lr , and suppose that kf k = 0. Given ε ∈ (0, 1),
there exist g ∈ Lp and h ∈ Lr such that f = g + h and kgkp + khkr < ε(p+1)/p . It follows that µ(G) < ε, where
G := {x ∈ X | ε ≤ |g(x)|}, because
Z
Z
Z
p
p
p
µ(G)ε =
ε ≤
|g| ≤ |g|p < ε(p+1)p/p = εp+1 .
G

G

Similarly µ(H) < ε where H := {x ∈ X | ε ≤ |h(x)|} (if r = ∞ then µ(H) = 0 because ε(p+1)/p < ε). Define
F := {x ∈ X | 2ε ≤ |f (x)|}, so that F ⊆ G ∪ H. Then µ(F ) ≤ µ(G ∪ H) < 2ε. It follows that

−1
{x ∈ X | 0 < |f (x)|} = ∪∞
≤ |f (x)|},
n=1 ∩k=n {x ∈ X | 2k

has measure 0, so f = 0 almost everywhere. This shows that k · k is a norm.
P
p
r
p
r
Let ∞
n=1 fn be an absolutelty convergent series in L + L . For each n ∈ N there exist gn ∈ L and hn ∈ L such
P
P

that fn = gn + hn and kgn kp + khn kr < kfn k + 2−n . It follows that ∞
n=1 gn and
n=1 hn are absolutely convergent
p
r
p
r
series in L and L respectively, so they have limits g ∈ L and h ∈ L respectively. This convergence also holds in
1

it converges to f in measure. . j=1 j=1 j=1 j=1 p 2 .→ Lp + Lr is continuous. p p ∞ 10. where Fj is the union of these rectangles. m) and ε ∈ (0. Otherwise replace x−1/p1 by | log(x)|. . Then |g|p ≤ |g|q ≤ |f |q and |h|r ≤ |h|q ≤ |f |q . 6. namely g + h.∞) (x). hence subsequential convergence almost everywhere). It follows that h|fnkj | ij=1 converges to |f | almost everywhere. Fix j ∈ {1. We may shrink Fj to give a finite collection of rectangles with sides that are intervals ε with rational endpoints. whose union Gj satisfies kbj χFj − bj χGj kp < 4r .Real Analysis Chapter 6 Solutions Jonathan Conder PN PN PN P Lp + Lr . By Proposition 6. Otherwise. in which case hkfnkj − f kp i∞ j=1 also ∞ converges to 0. define f (x) := x−1/p1 χ(0. so by the dominated convergence theorem |f |p ∈ L1 . set f (x) := x−1/p1 | log(x)|−2/p1 χ(0. In exercise 3 we showed that convergence in Lp implies convergence in measure (which in turn implies Cauchyness in measure. h|fn − f |p i∞ n=1 and h2 (|fn | + |f | )in=1 converge to 0 and 2 everywhere. Moreover. Conversely. Most of the hard work has been done as part of exercise 2. 1 ) (x) + x−1/p0 χ(2. Let f ∈ Lp (Rn . let hfn i∞ n=1 be a sequence of measurable X functions in C which converges to f : X → C in measure. ∞). which implies that Z 1/p Z 1/r q q kf k = kg + hk ≤ kgkp + khkr ≤ |f | + |f | = 11/p + 11/p = 2. ε There exists bj ∈ Q(i) such that kaj χEj − bj χEj kp = |aj − bj |m(Ej )1/p < 4r . Since |fn − f |p ≤ 2p (|fn |p + |f |p ). r) and let f ∈ Lq with kf kq = 1. in which case hkfn −f kp in=1 also converges to 0. by Theorem 2. Suppose that hkfn kp in=1 converges p p p ∞ p+1 |f |p almost to kf kp . p ∞ the dominated convergence theorem implies that hkfnkj − f kp ij=1 converges to 0. Define E := {x ∈ X | 1 < |f (x)|}. and g p ∈ L1 ). It follows that r r r r X X X X f − bj χGj ≤ kf − gk + kai χEi − bi χEi kp + kbi χEi − bi χFi kp + kbi χFi − bi χGi kp < ε. so g ∈ Lp and h ∈ Lr . 9. Let hfn i∞ n=1 be a sequence in L which converges to f ∈ L almost everywhere. because k N n=1 hn − hkr for all n=1 hn − hk ≤ k0kp + k n=1 gn − gkp + k0kr and k n=1 gn − gk ≤ k P∞ P∞ p r N ∈ N. whence f ∈ Lp (note that |fn |p ≤ g p for all n ∈ N. 2 (c) Set p1 := p0 in part (b). The converse follows from the reverse triangle inequality. Suppose there exists g ∈ Lp such that |fn | ≤ g for ∞ all n ∈ N. Therefore n=1 fn = n=1 (gn + hn ) has a limit in L + L . so it has a further subsequence p ∞ p hfnkj i∞ j=1 which converges to f almost everywhere. 1 ) + x−1/p0 | log(x)|−2/p0 χ(2. so hkfn − f kp i∞ n=1 itself converges to 0.7 there is a simple function g = j=1 aj χEj such that ε m(∪rj=1 Ej ) < ∞ and kf − gkp < 4 . Now let q ∈ (p. Pr 13.40(c) there is a ε finite collection of rectangles whose sides are intervals such that kbj χEj − bj χFj kp = |bj |m(Ej 4Fj )1/p < 4r . If hfnk i∞ k=1 is a subsequence of hfn in=1 . 2 (b) If p1 < ∞. (a) If p1 < ∞. p p p p Furthermore h|fnkj − f |p i∞ j=1 converges to 0 almost everywhere. n→∞ n→∞ n→∞ ∞ the generalised dominated convergence theorem implies that hkfn −f kpp i∞ n=1 converges to 0. and Z Z  p lim 2p (|fn |p + |f |p ) = 2p lim kfn kpp + 2p kf kpp = 2p lim kfn kp + 2p kf kpp = 2p kf kpp + 2p kf kpp = 2p+1 |f |p . and since |fnkj − f | ≤ 2 (g + |f | ) for all j ∈ N. r} and assume without loss of generality that m(Ej ) > 0. . This shows that the linear map Lq . . omit the first term. This shows that every subsequence of hkfn − f kp in=1 has a subsequence which converges to 0.∞) . This shows that Lp + Lr is a Banach space. g := f χE and h := f χE c .64.

s ∈ (0. n ∈ N with m ≥ n ≥ N.2. m). If m. This shows that hfn i∞ n=1 −(p+1)/p 1/p is Cauchy in measure. ∞) set fr := χBr (0) . ∞). n ∈ N and m ≥ n ≥ N then Z Z Z p p p µ(E)ε = ε ≤ |fm − fn | ≤ |fm − fn |p = kfm − fn kpp < (εδ)(p+1)p/p = εp δ. There exists N ∈ N such that kfm − fn kp < εδ m. In particular D is uncountable. ∞) → D. so there is an injective function (0. δ ∈ (0. ∞). If D ⊆ L∞ (Rn . Since {|fn | }n=1 is a finite subset of L . E E so µ(E) < δ. Now let ε ∈ (0.Real Analysis Chapter 6 Solutions Jonathan Conder so the collection of rational linear combinations of characteristic functions of finite unions of rectangles with sides that are intervals with rational endpoints (which is countable) is dense in Lp (Rn . m) is dense then B1/2 (fr ) ∩ D 6= ∅ for all r ∈ (0. p 1/p for all 15. There exists N ∈ N such that kfm − fn kp < 2 ε for all m. If n ∈ N and E ∈ M such that n > N and µ(E) < δ. R Hence. then . where E := {x ∈ X | ε ≤ |fm (x) − fn (x)|} = {x ∈ X | εp ≤ |fm (x) − fn (x)|p }. n ∈ N p N 1 with m ≥ n ≥ N. ∞) such that | E |fn |p | < 2−(p+1) ε < ε for all n ∈ N and E ∈ M with n ≤ N and µ(E) < δ. Let hfn i∞ n=1 be a Cauchy sequence in L . ∞). there exists δ ∈ (0. ∞) such that r 6= s (find some open rectangle inside Br (0)4Bs (0)). For each r ∈ (0. it is uniformly integrable by exercise 11 in section 3. Let ε. and note that kfr − fs k∞ = 1 for all r.

Z .

Z .

Z .

Z .

.

.

.

p p p p p p p.

p.

.

|fn |p .

= |fn | ≤ 2 (|fn − fN | + |fN | ) ≤ 2 kfn − fN kp + 2 .

|fN | .

. < 2−1 ε + 2−1 ε = ε.

.

let hfn i∞ n=1 be a sequence in L which satisfies properties (i)-(iii). Clearly E c |fn | ≤ En n ∈ N and n > N then Z Z Z p p p p p p p |fn | ≤ 2 (|fn − fN | + |fN | ) ≤ 2 kfn − fN kp + 2 |fN |p < 2−1 ε + 2−1 ε = ε. m→∞ F c m m→∞ m→∞ F m Fm Hence. where E := ∪N c |fn | < 2 n=1 En . so Z  Z Z Z Z Z Z p p p p p p lim |fn | = lim |fn | − |fn | = |fn | − lim |fn | = |fn | − |fn |p = 0. and note that µ(En ) < ∞ because µ(En )m−1 = Z m−1 ≤ Z En |fn |p ≤ Z |fn |p < ∞. En R p p −(p+1) ε < ε for all n ∈ N with n ≤ N. 3 . n ∈ N. R Ec Ec Ec This shows that hfn i∞ n=1 has property (iii). For each m. E E E E p This shows that {|fn |p }∞ n=1 is uniformly integrable. there exists m ∈ N such that R c Fm F |fn |p < 2−(p+1) ε. Otherwise. Then hFm i∞ m=1 is increasing and ∪m=1 Fm = F. ∞) such that F |fn |p = | F |fn |p | < 2−(p+3) εp for all n ∈ N and F ∈ M with µ(F ) < δ. set F := {x ∈ X | 0 < |fn (x)| } and define ∞ Fm := {x ∈ X | m−1 ≤ |fn (x)|p } for each m ∈ N. take δ ∈ (0. If Therefore µ(E) < ∞. p Conversely. ∞) and choose E ∈ M such R that µ(E) < ∞ and E c |fn |p < 2−(p+3) εp for all n ∈ N. Now fix n ∈ N. n ∈ N define Amn := {x ∈ X | ε(2µ(E))−1/p ≤ |fm (x) − fn (x)|}. If µ(E) = 0 then Z Z Z Z Z εp p p p p p p p p < εp |fm − fn | = |fm − fn | ≤ 2 (|fm | + |fn | ) = 2 |fm | + 2 |fn |p < 2p+1−(p+3) εp = 4 Ec Ec Ec Ec R R and hence kfm − fn kp < ε for all m. Let ε ∈ (0. Define En := Fm .

Real Analysis so that Chapter 6 Solutions Z |fm − fn |p ≤ Z εp (2µ(E))−1 ≤ Z εp (2µ(E))−1 = E E\Amn E\Amn Jonathan Conder εp . in which case Z Z Z Z εp |fn |p < 2p+1−(p+3) εp = . it is clear that φ dλ = φ dµ + φ dν. This shows that hfn i∞ n=1 is Cauchy in L . n ∈ N with m ≥ n ≥ N. 4 2 4 p and hence kfm − fn kp < ε. (a) Let f ∈ L2 (λ) and suppose that kf k2 = 1. ∞]. 18. n ∈ N with m ≥ n ≥ N. |fm |p + 2p |fm − fn |p ≤ 2p (|fm |p + |fn |p ) ≤ 2p 4 Amn Amn Amn Amn It follows that Z p Z |fm − fn | ≤ p Z |fm − fn | + Ec p Z |fm − fn | + E\Amn |fm − fn |p < Amn εp εp ε p + + = εp . for all m. 2 There exists N ∈ N such that µ(Amn ) < δ for all m. Therefore . For every simple function φ : X → [0. hence every φ ∈ L+ and in R R R fact every φ ∈ L1 (λ).

Z .

Z Z Z Z p .

.

.

f dν .

≤ |f | dν ≤ |f | dµ + |f | dν = |f | dλ ≤ kf k2 k1k2 = λ(X) .

.

Note that φn = φn−1 χ{1. there is an increasing −1 −1 sequence of simple functions hφn i∞ n=1 which converges to (1−g) χA ν-almost everywhere. This shows that νa  µ.. R R In fact E f (1 − g) dν = E f g dµ for all f ∈ L2 (λ) and E ∈ M (as f χE ∈ L2 (λ)).n} and hence kφn k = kn−1 χ{1. 19.. then φ(χ{k} ) = g(k) and hence (φn )∞ d {k} (Bε/2 (g(k))). It follows that g(k) < ε. φn ∈ L2 (λ) for all n ∈ N.. Therefore (φn )∞ cN −1 (B1 (0)).. Moreover Z Z Z Z µ(E) = χE dµ = χE · g dµ = χE (1 − g) dν = 0 dν = 0 for all E ∈ M with E ⊆ B. λ({x ∈ X | g(x) ≤ −n−1 }) = 0 for all n ∈ N. Therefore νs ⊥ µ.. −1 If k ∈ N and ε ∈ (0. Similarly 1 − g ≥ 0. This n=1 is frequently in χ shows that φ 6= φg for all g ∈ l1 .29 that (φn )∞ n=1 has a weak* cluster point φ ∈ (l ) . In particular 0 ≤ g ≤ 1 µ-almost everywhere and ν-almost everywhere. so there exists R R R R R R g ∈ L2 (λ) such that f dν = f g dλ = f g dµ + f g dν. so g = 0. by Proposition 6. Since (1 − g)−1 χA ≥ 0 ν-almost everywhere. ∞).. Therefore Z Z Z Z χA (1 − g) dν = lim φn (1 − g) dν = lim φn g dµ = g(1 − g)−1 χA dµ νa (E) = ν(A ∩ E) = n→∞ E n→∞ E E E 1−g for all E ∈ M. n=1 is frequently in χ 1 ε 1 there exists n ∈ N with n < 2 and n ≥ k such that |g(k) − n | = |g(k) − φn (χ{k} )| < 2ε . Moreover. This shows that the linear functional f 7→ f dν on L2 (λ) is bounded. which implies that g ≥ 0 λ-almost everywhere. In particular. and hence f (1 − g) dν = f g dµ. by the monotone convergence theorem. It follows by Alaoglu’s ∞ ∗ 1 theorem and Theorem 4. for all f ∈ L2 (λ).. R by H¨older’s inequality.. which is impossible because φn (χN ) = 1 for all n ∈ N. In particular E Im(g) dλ = 0 for all E ∈ M. R R R R R R (b) Note that E g dλ = χE g dλ = χE dν = ν(E) ≥ 0 and E (1 − g) dλ = E dλ − E g dλ = λ(E) − ν(E) ≥ 0 R for all E ∈ M. (c) Clearly νs (E) = ν(∅) = 0 for all E ∈ M with E ⊆ B c . λ-almost everywhere. Suppose that φ = φg for some g ∈ l . Since (1−g) χA < ∞ pointwise.13. 4 .n} k1 = 1. whence 0 ≤ g ≤ 1. which implies that Im(g) = 0 λ-almost everywhere.

It follows from H¨older’s inequality that Z !1/q Z q |(fn − f )g| ≤ kfn − f kp (A\B)c |g| (A\B)c ≤ (kfn kp + kf kp ) ε ε ≤ 4M 2 for all n ∈ N. If n ∈ N and n ≥ N. it R R suffices to show that h fn gi∞ f g. There exists g ∈ Lq . µ(B) < δ and hfn |A\B i∞ n=1 converges R R R q q q q uniformly to f |A\B . then Z !1/p Z |(fn − f )g| ≤ A\B |fn − f | A\B   ε εp 1/p kgkq < kgkq ≤ µ(A \ B) p 2 L 2 and hence . In particular (A\B)c |g| = Ac |g| + B |g| < (ε/4M ) . Set L := µ(A)kgkpq + 1 and choose N ∈ N such that |fn (x) − f (x)| < εp /2p L for all x ∈ A \ B and n ∈ N with n ≥ N. Now define F := {x ∈ X | 0 < |g(x)|q } ∞ and set Fm := {x ∈ X | m−1 ≤ |g(x)|q } for each m ∈ N. Note that µ(A) < ∞ because Z Z Z q −1 −1 |g| ≤ |g|q < ∞. n→∞ so f ∈ Lp and kf kp ≤ M. where A := Fm . where q := p/(p − 1). Then hFm i∞ m=1 is increasing and ∪m=1 Fm = F. Hence. ∞). By Fatou’s lemma Z Z p |f | ≤ lim inf n→∞ |fn |p = lim inf kfn kpp ≤ M p . Set M := supn∈N kfn kp + 1.Real Analysis Chapter 6 Solutions Jonathan Conder R 20. so  Z Z Z Z Z Z Z q q q q q q |g|q = 0. m ≤ µ(A)m = Hence. let ε ∈ (0. |g| = |g| − lim |g| = lim |g| = |g| − lim |g| − m→∞ F c m m→∞ m→∞ F m Fm F |g|q < 21 (ε/4M )q . such that g(h) = hg for all h ∈ Lp . there exists m ∈ N such that R Ac A A By Egoroff’s theorem. Since n=1 converges to R q E 7→ E |g| is a finite measure on M which is absolutely continuous with respect to µ. there exists δ ∈ (0. it follows that there exists B ∈ M such that B ⊆ A. To this end. ∞) R R such that E |g|q = | E |g|q | < 12 (ε/4M )q for all E ∈ M with µ(E) < δ. (a) Let g ∈ (Lp )∗ .

Z .

Z Z Z Z .

.

.

fn g − f g .

n . ≤ |(fn − f )g| = |(f − f )g| + |(fn − f )g| < ε.

.

A\B (A\B)c R R This shows that h fn gi∞ f g. as required. 2n + 1]. if n is odd X xn (m)g(m) = g(n) = 1. Then supn∈N kxn k1 = supn∈N 1 = 1 and hxn i∞ n=1 P 1 )∗ . Clearly |g| ≤ 1 and f 7→ f g dm is in L1 (R. However R ∞ h fn g dmi∞ n=1 does not converge. if n is even. where E := ∪n=1 [2n. n=1 converges to (b) For each n ∈ N define fn := χ[n.n+1] . if n is even. and hence hfn in=1 does not converge weakly. Then supn∈N kfn k1 = supn∈N 1 = 1 and hfn i∞ n=1 converges to 0 pointwise. so that |g| ≤ 1 and hence x 7→ ∞ x(m)g(m) is in (l m=1  ∞ 0.n+1]∩E dm = 1. Set g := χ2N . m=1 5 . However converges to 0 pointwise. Similarly. because  Z Z 0. for each n ∈ N define xn : N → C by xn := χ{n} . if n is odd fn g dm = χ[n. m)∗ . R ∞ Define g := χE .

To this end. Since so h C. and hence hxn in=1 does not converge weakly. We aim to show that h gfn i∞ n=1 converges to gf for all g let g ∈ L1 and set M := supn∈N kfn k∞ . It follows that |g(fn − f )| ≤ |g|(|fn | + |f |) ≤ 2|g|M R almost everywhere for all n ∈ N. ∪∞ n=1 En has measure zero. so by the dominated convergence theorem h |g(fn − f )|i∞ n=1 converges to R 0 = 0. ∞ which converges almost everywhere to f : X → Now let hfn i∞ n=1 be a sequence in L R R supn∈N kfn k∞ < ∞ and µ is σ-finite. it is easily shown that |f | ≤ M almost everywhere. Suppose that ∈ L1 .Real Analysis Chapter 6 Solutions Jonathan Conder P∞ ∞ ∞ m=1 xn (m)g(m)in=1 does not converge. For each n ∈ N set En := {x ∈ X | fn (x) > M }. Since .

Z .

Z Z .

.

.

0 ≤ .

gfn − gf .

.

fn i = cos(2πmx) cos(2πnx) dx 0 Z 1 1 (cos(2π(m − n)x) + cos(2π(m + n)x)) dx = 2 0  . n=1 converges to p p/(p−1) defines a functional φ ∈ lp (A)∗ . ≤ |g(fn − f )| R R for all n ∈ N. 22. so 21. (a) If m. The converse is a special case of exercise 20(a). Suppose that (fn )∞ a n=1 converges weakly to f in l (A). so supn∈N kfn kp = supn∈N kfbn k < ∞ by the uniform boundedness principle. If φ ∈ lp (A)∗ then supn∈N |fbn (φ)| = supn∈N |φ(fn )| < ∞. n ∈ N and m 6= n then Z 1 hfm . as required. If a ∈ A then χ{a} ∈ l ∞ ∞ ∞ (fn (a))n=1 = (φa (fn ))n=1 converges to φa (f ) = f (a). Therefore (fn )n=1 converges pointwise to f. the squeeze theorem implies that h gfn i∞ gf.

1 sin(2π(m − n)x) sin(2π(m + n)x) .

.

1 = + .

2 2π(m − n) 2π(m + n) 0     1 sin(2π(m − n)) sin(2π(m + n)) 1 sin(0) sin(0) = + − + 2 2π(m − n) 2π(m + n) 2 2π(m − n) 2π(m + n) = 0. Moreover. if n ∈ N then Z hfn . fn i = 0 1 1 cos(2πnx) dx = 2 2 Z 0 1 .

1 1 sin(4πnx) .

.

1 1 (cos(0) + cos(4πnx)) dx = + = . 2 2 4πn .

it follows that | cos(2πnx) − 1| = | cos(2πnx) − cos(2πm + π/2)| < 1/2 and hence ∞ cos(2πnx) > 1/2.0 2 √ 2 This shows that h 2fn i∞ n=1 is an orthonormal sequence in L . {x ∈ X | 1/ 2 ≤ cos(2πnx)} = ∪m=0 [(8m + 1)/8n. Let x ∈ (0. (8m + 3)/8n]. Thus. for each n ∈ N. 1/4π) and N ∈ N. Let n ∈ N be the smallest number such that nx − 1/4 > m (which exists because R is Archimedean). it is not the case that hfn i∞ n=1 converges to 0 almost everywhere. By the mean value theorem and the fact that | sin | ≤ 1. 6 . Therefore hfn i∞ n=1 converges weakly to 0. 1/4π). (8m + 3)/8] and √ n−1 hence. √ Note that {x ∈ R | 1/ 2 ≤ cos(2πx)} = ∪m∈Z [m + 1/8. This implies that hcos(2πnx)i∞ n=1 does not converge to 0. so hfn in=1 does not converge to 0 on (0. m + 3/8] = ∪m∈Z [(8m + 1)/8. in which case 0 < nx − m − 1/4 ≤ x < 1/4π. which ∞ shows that hfn in=1 does not converge to 0 in measure. so it converges weakly to 0 by exercise 63 of the previous homework. It follows that √ √ P m({x ∈ X | 1/ 2 ≤ |fn (x) − 0|}) ≥ m({x ∈ X | 1/ 2 ≤ cos(2πnx)}) = n−1 m=0 2/8n = 2/8 for all n ∈ N. Then n > N ≥ 1 and hence (n − 1)x − 1/4 ≤ m. Choose m ∈ N with N x − 1/4 ≤ m. Therefore |2πnx − 2πm − π/2| < 1/2.

1]. if f ∈ L2 ((0. x. where q is the conjugate exponent to p. it follows by Tonelli’s theorem that p Z Z Z Z 2 −1/q −1 p/q xp/q K(xy)f (x)p dx dy K(xy)f (x) dx dy ≤ y φ(p ) ZZ 2 −1 p/q = φ(p ) y −1/q xp/q K(xy)f (x)p dy dx Z Z −1 p/q p/q 2 p = φ(p ) x f (x) K(xy)y −1/q dy dx Z 2 −1 p/q xp/q f (x)p x−1/p φ(p−1 ) dx = φ(p ) Z 2 2 2 −1 (p+q)/q = φ(p ) x(p −q )/(pq ) f (x)p dx Z −1 p = φ(p ) x(p−q)/q f (x)p dx Z = φ(p−1 )p xp−2 f (x)p dx. 1)|x dx = Cp = 1+1/p 1/p 1/p 1+1/p x +x x 0 0 x 1 0 It follows by Theorem 6. ∞) then m({x ∈ X | ε ≤ |fn (x) − 0|}) ≤ m({x ∈ X | 0 < |fn (x)|}) = m((0. Thus. Z p Z p/q Z 2 −1/q K(xy)f (x) dx ≤ K(xy)x dx xp/q K(xy)f (x)p dx Z 2 −1/q −1 p/q xp/q K(xy)f (x)p dx. ∞) then Z Z Z −1/q −1/q −1 1/q−1 K(xy)x dx = K(z)(z/y) y dz = y K(z)z 1/p−1 dz = y −1/p φ(p−1 ) 2 2 and hence. 27.20 that kT f kp ≤ Cp kf kp for all f ∈ Lp . in which case fn (x) = 0 for all n ∈ N with n ≥ N. Z ∞ Z 1 Z ∞ Z ∞ 1 1 1 −1/p dx ≤ dx + dx < ∞. This shows that hfn i∞ n=1 converges to 0 on (0. If ε ∈ (0. which shows that hfn i∞ n=1 converges R ∞ p p to 0 in measure. λy) = λ−1 K(x. p it follows that hfn i∞ n=1 does not converge to 0 weakly in L . 1] there exists N ∈ N such that 1/N < x. (b) From the long calculation Z Z 2 |T f (y)| dy = in part (a). ∞] it is clear that hfn in=1 is a sequence in L . by H¨older’s inequality (and Tonelli again)  Z 1/p ZZ Z Z −1 p−2 p K(xy)f (x)g(y) dx dy = K(xy)f (x) dx g(y) dy ≤ φ(p ) x f (x) dx kgkq . |K(x. Moreover. Define a measurable function K : (0. y) for all λ. Note that K(λx. (a) If y ∈ (0. 30. Given p ∈ [1. y ∈ (0. ∞)2 → C by K(x. y) := (x + y)−1 .Real Analysis Chapter 6 Solutions Jonathan Conder (b) If x ∈ (0. and f 7→ f is in (L )∗ by H¨ older’s R q inequality and the fact that χX ∈ L . ∞). 1/n)) = 1/n for all n ∈ N. =y φ(p ) Since p + q = pq and hence p − q = pq − 2q. using H¨ older’s inequality and the fact that K(xy)f (x) = K(xy)1/q x−1/q x1/q K(xy)1/p f (x). ∞)) then . hence almost everywhere. Since fn = 1 6= 0 for all n ∈ N.

Z .

2 2 Z Z Z .

.

1 2 .

K(xy)f (x) dx.

dy ≤ K(xy)|f (x)| dx dy ≤ φ( 2 ) x0 |f (x)|2 dx .

.

7 .

|T f (x)| dµ(x) = K(·. by H¨older’s inequality Z Z y kK f (y)k2 dν(y) = kK y k2 |f (y)| dν(y) = kkf k1 ≤ kkk2 kf k2 = kKk2 kf k2 . so by Minkowski’s inequality for integrals 2 Z Z 2 Z 2 kK(·. Since Lq ((0. y)f (y) dν(y) ≤ 2 This shows that T f ∈ L2 (µ) and kT f k2 ≤ kKk2 kf k2 . Therefore T : L2 ((0. so T f is continuous. Fubini’s theorem implies that K y ∈ L2 (µ) and hence K y f (y) ∈ L2 (µ). y)f (y)| dν(y) ≤ kKx k2 kf k2 . x)) for all x ∈ (0. Therefore Kx ∈ L2 (ν) for R almost all x ∈ X. and it is clearly linear. By H¨ older’s inequality |K(x. ∞)) to C(0. Then Z ZZ ZZ kkk22 = |k(y)|2 dν(y) = |K y (x)|2 dµ(x) dν(y) = |K(x. ∞) there R R∞ exists n ∈ N such that n |f |q < ( 2ε )q . for almost all x ∈ X. Rx If f ∈ Lq ((0. This implies that y 7→ kK y f (y)k2 is in L1 (ν). ∞)). Given ε ∈ (0. y)|2 dµ(x) dν(y) = kKk22 and hence. Define k : Y → [0. Since |K|2 ∈ L1 (µ × ν). Moreover kT k ≤ φ( 12 ). and hence T f (x) converges absolutely. ∞)) → L2 ((0.∞) . ∞)) is well-defined. ∞) by k(y) := kK y k2 . it is clear that T is a well-defined linear map from Lq ((0. ∞). ∞). Now . Fubini’s theorem implies that |Kx |2 ∈ L1 (ν) for almost all x ∈ X. for almost all y ∈ Y. ∞)) with kT f k2 ≤ φ( 21 )kf k2 . the map x 7→ 0 f is absolutely continuous on (0. 33. y)f (y)k2 dν(y) ≤ (kKk2 kf k2 )2 . x)) ⊆ L1 ((0. 32.Real Analysis Chapter 6 Solutions Jonathan Conder and hence T f ∈ L2 ((0. by continuity of the finite measure E 7→ E |f |q .

.

Z Z ∞ .

−1/p ∞ .

−1/p .

n) )(x)| = . |T f (x) − T (f χ(0.

x f −x f χ(0.n) .

.

∞) | ≤ x−1/p kχ(0.∞) kq ε < 2 for all x ∈ (0.n) | 0 Z = x−1/p χ(0. ∞)). ∞). so T f ∈ C0 ((0. Moreover.x) kp kf χ[n. thus compact (because |T f | is continuous). by H¨older’s inequality . 0 Z 0∞ ≤ x−1/p |f − f χ(0.n) k1 )p . then Z n ε −1/p ≤ |T (f χ(0. In particular. ∞) and |T f (x)| ≥ ε.x) |f χ[n. by H¨ older’s inequality.n) )(x)| ≤ x |f | 2 0 and hence x ≤ ( 2ε kf χ(0. This implies that {x ∈ X | ε ≤ |T f (x)|} is bounded. if x ∈ (0.

.

Z Z .

.

−1/p −1/p .

.

|T f (x)| = .

x f χ(0.x) .

x) kp kf kq = kf kq for all x ∈ (0. so T is a bounded linear map from Lq ((0. This shows that kT f ku ≤ kf kq . so that p = (p − 1)q > q and hence q/p < 1. Suppose p > 2. ≤ x |f χ(0. ∞)) to C0 ((0. ∞).x) | ≤ x−1/p kχ(0. Set q := p/(p − 1). By H¨older’s inequality Z 1 0 Z |f | ≤ 0 1 1/p Z x|f (x)| dx 0 p 0 0 8 1 x −q/p 1/q dx <∞ . 34. ∞)).

Define g : [0.∞) := f. 21 ) and δ ∈ (. 1] → C by g(0) := f (1) − 0 f 0 and g|(0. 1). 1]). 1]) because g 0 = f 0 almost everywhere. 1}.  ∈ (0. If p = 2. 1]. and g 0 ∈ L1 ([0. then H¨older’s inequality implies that s s s Z 2 Z δ Z 1 Z 1 Z 1 0 −1 0 2 0 2 x dx = |f | ≤ x|f (x)| dx · x|f (x)| dx · (log(2) − log()) = x|f 0 (x)|2 dx · log(2)   0 and hence (where M := 0≤ 0 0 qR 1 |f ()| | log()|1/2 x|f 0 (x)|2 dx · log(2)) . 2). so g is absolutely continuous on [0. This clearly also holds for  ∈ {0. Moreover Z  Z 1 Z 1 Z 1 f0 f0 = f0 + f0 = − g() − g(0) = f () − f (1) + 0  0 0 for all  ∈ (0. 1] and hence limx→0 f (x) = limx→0 g(x) = g(0).Real Analysis Chapter 6 Solutions Jonathan Conder R1 and hence f 0 ∈ L1 ([0. Then g is differentiable almost everywhere on [0.

R δ .

.

.

  Z δ .

f (δ) −  f 0 .

p). By the monotone convergence theorem Z Z X ∞ ∞ ∞ ∞ X X X |f |q ≤ (2n )q χEn = 2nq µ(En ) ≤ µ(E0 ) + 2nq λf (2n−1 ) ≤ µ(E0 ) + 2nq+p−np [f ]pp < ∞.2n ] n∈Z 0 n∈Z 9 . By Proposition 6. suppose that f ∈ Lp . It is obvious that f ∈ Lq for the case q = ∞. f (α) ≤ [f ]pp n=1 for all α ∈ (0. This shows that f ∈ Lq . For each n ∈ Z define En := {x ∈ X | 2n < |f (x)| ≤ 2n+1 }. Define E0 := {x ∈ X | 0 < |f (x)| ≤ 1}. Then ! Z Z ∞ X q q 1−n q kf k∞ χE0 + |f | ≤ (2 ) χEn n=1 = kf kq∞ µ(E0 ) + ∞ X 2q−nq µ(En ) n=1 ≤ kf kq∞ λf (1) + ∞ X 2q−nq λf (2−n ) n=1 ≤ kf kq∞ [f ]pp + ∞ X 2q−nq+np [f ]pp n=1 and hence f ∈ Lq . Conversely. 0 −1/2 |f | ≤ | log()|−1/2 (|f (δ)| + M ) |f (δ)| + = ≤ | log()| 1/2 | log()|  0 36. Fix q ∈ (0. ∞) (note that P∞ nq−np n=1 2 is a convergent Now suppose that f ∈ (weak Lp ) ∩ L∞ . By the monotone convergence theorem Z Z X X X p |f | ≤ (2n+1 )p χEn = 2p 2np µ(En ) ≤ 2p 2np λf (2n ) < ∞ n∈Z n∈Z n∈Z and hence f ∈ Lp . ∞). n=0 n=0 n=1 αp λ because E0 ⊆ {x ∈ X | 0 6= f (x)} and geometric series). and for each n ∈ N set En := {x ∈ X | 2n−1 < |f (x)| ≤ 2n }. P 38. Suppose f ∈ weak Lp and µ({x ∈ X | 0 6= f (x)}) < ∞. Let f : X → C be a measurable function and suppose that n∈Z 2np λf (2n ) < ∞. Define E0 := {x ∈ X | 1 < |f (x)|} and for each n ∈ N set En := {x ∈ X | 2−n < |f (x)| ≤ 21−n }.24 Z ∞X X np n 2 λf (2 ) = 2np−n+1 λf (2n )χ(2n−1 . and fix q ∈ (p.

1 − x] r ∈ (1 − x. . . 1)) = x + r    1 Since r 7→ x 2r + 1 2 r ∈ (0. Clearly 1 Hχ(0. Similarly r ∈ (0. we may choose φ ∈ Lp such that kf − φkp < 0 ≤ αp λf (α) ≤ αp λf −φ α 2 + α p λφ α 2 ε .Real Analysis Chapter 6 Solutions p ∞ Z ≤2 0 ≤ 2p 0 = 2p p 2(n−1)(p−1) λf (α)χ(2n−1 . 2p+1 α→0 By Chebyshev’s inequality ≤ 2p kf − φkpp + 2p  α p 2 λφ α 2 < ε ε + =ε 2 2 provided that α ∈ (0. Pn 39. ∞).x+r) If x ∈ (−∞.2n ] (α) dα n∈Z ∞ Z X Jonathan Conder X αp−1 λf (α)χ(2n−1 . . . 1)) . x + r) ∩ (0. 1)) = x + r    1 Given x ∈ (0.    2r   m((x − r. ∞) is sufficiently small (large). x] r ∈ (x. 1 − x]. 1 − x] and 1 ≤ 2r for all r ∈ (1 − x. so lim αp λφ (α) = lim 0 = 0 α→∞ α→∞ lim αp λφ (α) = lim αp λφ (0) = 0. . We may assume that ai 6= 0 for all i ∈ {1. |an |} and λφ (α) = λφ (0) if α < min{|a1 |.1) (x) = 1. and α→0 Given ε ∈ (0.1) (x) = sup r∈(0. . If φ = i=1 ai χEi is simple with λφ (0) < ∞ and α ∈ (0. 12 ] r ∈ (0. Therefore limα→0 αp λf (α) = 0 (limα→∞ αp λf (α) = 0). it follows that Hχ(0. 0] then    0   m((x − r. Since x + r ≤ 2r for all r ∈ [x. . . ∞).∞) Z |χ(0. ∞). ∞). 43.1) (y)| dy = sup (x−r. is increasing on [−x. x + r)) m((x − r. x + r) ∩ (0. . ∞) . −x) r ∈ [−x.∞) m((x − r. then λφ (α) = 0 if α ≥ max{|a1 |. 1 − x] r ∈ (1 − x. 1)) = 1 − x + r    1 10 1 2−2x . x) r ∈ [x. ∞). |an |}. . . 1 − x) r ∈ [1 − x. x + r) ∩ (0.∞) (α) dα Z0 ∞ =2 0 p −1 =2 p αp−1 λf (α) dα kf kpp < ∞. x + r) ∩ (0.1) (x) =    2r   m((x − r. n}. . 2r r∈(0. it follows that Hχ(0.2n ] (α) dα n∈Z ∞ Z αp−1 λf (α)χ(0.

∞) dx .1) ∈ weak L1 because x 7→ x1 is in weak L1 . and hence Hχ(0. is increasing on [x − 1.1) (x) = Z p Z 0 |Hχ(0.1) | ≤ 1 so Hχ(0. x]. ∞)    0   m((x − r. Continuing from above Z .1) ∈ Lp for all p ∈ (1. but not for p = 1. ∞). Moreover Hχ(0. if p ∈ [1. ∞].1) (x) = 1. Given x ∈ [1. Z dx + 0 1 ∞ Hence. (2x)p which is finite iff p 6= 1. x] r ∈ (x. 1).1) | = −∞ dx + (2 − 2x)p Z 1 1 2x . it follows that Hχ(0. x + r) ∩ (0.Real Analysis Chapter 6 Solutions Jonathan Conder whenever x ∈ ( 12 . Also |Hχ(0. 1)) = 1 − x + r    1 Since r 7→ 1−x 2r + 1 2 r ∈ (0. x − 1) r ∈ [x − 1.

0 .

∞ .

.

1 1 1 1 .

.

1) | = . = + 1 + +1− |Hχ(0.

.

11 .1) kp = 1 + (p − 1)−1 21−p for all p ∈ (1. p−1 p−1 p−1 −2(1 − p)(2 − 2x) 2(1 − p)(2x) 2(p − 1)2 2(1 − p)2p−1 −∞ 1 p and hence kHχ(0.1) kp behaves like (p − 1)−1 as p → 1.1) kp = 1 for all p ∈ [1. ∞). It is obvious that kχ(0. ∞]. so kHχ(0.