0% found this document useful (0 votes)
451 views13 pages

Folland 6

This summary provides the key information from the document in 3 sentences: The document discusses properties of functions in Lp spaces and their limits. It examines whether functions converge weakly or in measure as p approaches 0 or infinity. While some functions converge weakly, counterexamples show that weak convergence is not always guaranteed, even if the functions converge almost everywhere or in measure. The properties depend on the specific functions and measures involved in each Lp space.

Uploaded by

memex1
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
451 views13 pages

Folland 6

This summary provides the key information from the document in 3 sentences: The document discusses properties of functions in Lp spaces and their limits. It examines whether functions converge weakly or in measure as p approaches 0 or infinity. While some functions converge weakly, counterexamples show that weak convergence is not always guaranteed, even if the functions converge almost everywhere or in measure. The properties depend on the specific functions and measures involved in each Lp space.

Uploaded by

memex1
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 13

Folland: Real Analysis, Chapter 6

Sebastien Picard

Problem 6.8
Suppose (X) = 1R and f Lp for some p > 0, so that f Lq for 0 < q < p.
a. log
in 3.5, with
F (t) = et ).
R ||fq||q log |f |. (Use Exercise
R 42d
R
b. ( |f | 1)/q logR||f ||q , and ( |f |q 1)/q log |f | as q 0.
c. limq0 ||f ||q = exp( log |f |).
Solution:
(a) Define g = q log |f |.
R
Case 1: Suppose g
/ L1 . Then | log |f |q | = . We can split up the integral
Z
Z
Z
Z
Z
q
q
q
q
| log |f | | = log |f | |f |>1 log |f | |f |<1 |f | log |f |q |f |<1 .
R
R
q
q
Since
f

L
,
we
must
have
log
|f
|

=
.
Therefore,
since
log |f |q |f |>1 is finite,
|f
|<1
R
q log |f | = and the inequality holds trivially.
Case 2: Suppose g L1 . Let F (t) = et . We know that F (t) is convex on the real line since
the exponential function is absolutely continuous on every compact interval and F (t) = F (t) > 0.
Thus we can apply Jensens inequality:
Z
Z
1
log |f | =
log |f |q
q
Z
1
=
g
q
R
1
= log e g
q
Z
1
log eg
q
Z
1
= log |f |q = log ||f ||q .
q
(b)
The first inequality follows directly from the fact that log(x) x 1.
Z
Z
1
1
q
log ||f ||q = log |f | ( |f |q 1).
q
q
In order to prove the limit, we will need the fact that the map h : (0, ) R defined by
h(x) = (ax 1)/x is a monotone increasing function for any a 0. If a = 0, this is immediate. If
a > 0, we prove monotonicity by showing that h (x) 0. Indeed,
1

ax log ax ax + 1
h (x) =
,
x2

hence the result will follow if we show that w : (0, ) R, w(x) = x log x x is greater than or
equal to 1. Since w only vanishes at x = 1, and w (1) > 0, we see that the minimum of w occurs
at x = 1 and hence w 1.
We now prove the limit. By lHopitals rule, pointwise we have
log |f | |f |q
|f |q 1
= lim
= log |f |.
q0
q0
q
1
lim

As shown before, (|f |q 1)/q is monotone increasing, hence the limit as q 0 is monotone
decreasing. By the Monotone Convergence Theorem, we have
Z
Z
Z
1
1
q
q
(|f | 1) = lim (|f | 1) = log |f |.
lim
q0 q
q0
q

(c) From
part (a), after composing both sides with the exponential function, we see that ||f ||q
R
exp( log |f |) for all 0 < q < p. From part (b), we see that
Z
Z

q
|f | 1 /q log |f |.
lim sup log ||f ||q lim sup
q0

q0

R
Therefore,
lim
sup
||f
||

exp(
log |f |). Therefore, the limit exists, and limq0 ||f ||p =
q
q0
R
exp( log |f |).
Problem 6.10
Suppose 1 p < . If fn , f Lp and fn f a.e., then ||fn f ||p 0 iff ||fn ||p ||f ||p. (Use
Exercise 20 in 2.3.)
Solution:
Suppose ||fn f ||p 0. Then by the reverse triangle inequality,
| ||fn ||p ||f ||p | ||fn f ||p 0
as n .
On the other hand, suppose ||fn ||p ||f ||p. Notice
|fn f |p 2p (|f |p + |fn |p ).
We can define gn = 2p (|f |p + |fn |p ) and g = 2p+1|f |p L1 . Then gn g a.e, and furthermore,
2

lim

gn = 2

|f | + lim 2

|fn |

p+1

|f | =

g.

By Exercise 20 in 2.3,
lim

|fn f | =

lim |fn f |p = 0.

Problem 6.20
Suppose supn ||fn ||p < and fn f a.e.
a. If 1 < p < , then fn Rf weakly in Lp . (Given g Lq , where q is conjugate to p, and > 0,
there
exists (i) > 0 such that E |g|q < whenever (E) < , (ii) A X such that (A) < and
R
|g|q < , and (iii) B A such that (A\B) < and fn f uniformly on B.)
X\A
b. The result of (a) is false in general for p = 1. (Find counterexamples in L1 (R, m) and l1 .)
It is, however, true for p = if is finite and weak convergence is replaced by weak* convergence.
Solution:
a. Claim (i) follows from Corollary 3.6, and claim (iii) is Egoroffs theorem (2.33). We first prove
claim (ii). By Proposition 2.20, P = {x : |g|q > 0} is finite. Hence we can assume that P =
i Pi
where (Pi ) < and Pi are disjoint. Therefore,
Z

|g| =

|g| =

Z
X
i=1

Pi

|g|q < .

Since the sequence converges, there exists an N N such that


A = N
i=1 Pi has the required properties, which proves (ii).

P R
i=N

Pi

|g|q < . Therefore,

We now prove that fn f weakly in Lp . Take g Lq and any > 0. Since supn ||fn ||p C0 < ,
we have that f Lp by Fatous lemma:
Z
Z
p
|f | lim inf |fn |p C0p < .
By (i), choose > 0 such that
Z

|g| <

6C0

q

when (E) < . By (ii), choose A X such that (A) < and

q
Z

q
|g| <
.
6C0
X\A
By (iii), choose B A and N N such that (A\B) < and

|fn (x) f (x)| <

3(B)1/p ||g||

,
q

for all x B and n N.


With this setup, when n N, we apply Holders inequality to obtain
Z
Z
Z


fn g

f
g


X

X\A

|fn f | |g| +

||fn f ||p

Z

X\A

|g|

Z

|fn f | |g| +

A\B
1/q
q
q

+ ||fn f ||p

1/q Z

|g|
3(B)1/p ||g||q
B
1/q
Z
Z
q
2C0
|g|
+ 2C0
+

X\A

A\B

|fn f | |g|

Z

A\B

|g|

1/q

1/p

|g|

1/q

< /3 + /3 + /3 = .

b. First, a counterexample in L1 (R, m). Consider


fn = n (0,1/n) .
We
R have ||fn ||1 R= 1 for all positive integers n, and fn f a.e., where1 f = 0. However,
lim fn 1 = 1 but f 1 = 0. Therefore fn does not converge weakly to 0 in L .
Next, a counterexample in l1 . Consider
fn (i) =

0
1

if i 6= n,
if i = n.

We
R see that ||fn ||l1 =
R 1 for all positive integers n, and fn f a.e., where f = 0. But
lim fn 1 d = 1 and f 1 d = 0, where is the counting measure. Therefore fn does not
converge to 0 weakly in l1 .
Now we show that if supn ||fn || = C0 < , fn f a.e., and is finite, then fn f
in the weak* sense on L . The first thing we show is that ||f || < . Indeed, for a.e. x, there exists
a positive integer n such that |fn (x) f (x)| < 1. Then
|f (x)| |f (x) fn (x)| + |fn (x)| 1 + C0 .
Hence ||f || 1 + C0 .
4

By the finite condition, L = (L1 ) . Let g L1 and > 0.


By fact (i), choose > 0 such that
Z

|g| <

3(2C0 + 1)

when (E) < . By fact (ii), choose A X such that (A) < and
Z

.
|g| <
3(2C0 + 1)
X\A
By fact (iii), choose B A and N N such that (A\B) < and
|fn (x) f (x)| <

,
3||g||1

for all x B and n N. With this setup, when n N we have


Z
Z
Z


fn g
f g

X

|fn f | |g| +
|fn f | |g| +
|fn f | |g|
A\B
B
Z
Z
Z

|g|
||fn f ||
|g| + ||fn f ||
|g| +
3||g||1 B
X\A
A\B

< (2C0 + 1)
||g||1 = .
+ (2C0 + 1)
+
3(2C0 + 1)
3(2C0 + 1) 3||g||1
X\A

Problem 6.22
Let X = [0, 1], with Lebesgue measure.
a. Let fn (x) = cos 2nx. Then fn 0 weakly in L2 (see Exercise 63 in 5.5), but fn 9 0 a.e. or
in measure.
b. Let fn (x) = n(0,1/n) . Then fn 0 a.e. and in measure, but fn 9 0 weakly in Lp for any p.
Solution:

a. Let un (x) = 2 cos 2nx. We note the following two identities, where n 6= m.
Z 1
Z 1
Z 1
1 1
2
2
un = 2
(cos 2nx) dx = 2
( + cos 4nx)dx = 1.
2
0
0
0 2
Z

un um = 2

(cos 2nx)(cos 2mx)dx =

1
0

(cos(2x(n m)) + cos(2x(n + m)))dx = 0.

Therefore {un } is orthonormal


in the Hilbert space L2 . By Exercise 63(a), {un } converges weakly

to 0 in L2 . Therefore, {(1/ 2)un } = {fn } converges weakly to 0 in L2 .

We show that fn 9 0 a.e. by contradiction. If fn 0 a.e., then (fn fn ) 0 a.e. Since


1 is dominated by an integrable function on [0, 1], by the Dominated Convergence Theorem we
have
Z
Z
2
lim (cos 2nx) dx = lim(cos 2nx)2 dx = 0.

fn2

This is clearly nonsense since

R1
0

cos(2nx) = 1/2.

Finally, we show fn 9 0 in measure. We explicitely calculate the measure of the set of all
x [0, 1] such that | cos 2nx| 1/2. Denote
Ek = 2k,






5
3
7
+ 2k
+ 2k, + 2k + 2k,
+ 2k
+ 2k, 2(k + 1) .
4
4
4
4

Then | cos 2nx| 1/2 iff 2nx Ek for some k N. Therefore, using the 2 periodicity of
cosine, we see that | cos 2nx| 1/2 when x [0, 1] lies in one of the 4n disjoint intervals of length
1/8n. Hence
1
{x [0, 1] : | cos 2nx| 1/2} = .
2
Since the measure is constant at 1/2 for all n, fn 9 0 in measure.

(b) First we show that fn (x) = n(0,1/n) 0 a.e. Fix x (0, 1]. Then fn (x) = 0 for all n > 1/x.
Next, we show that fn (x) = n(0,1/n) 0 in measure. For all > 0, we have
{x [0, 1] : |fn (x)| } (0, 1/n) = 1/n 0
as n . Lastly, we show fn 9 0 weakly in Lp for any p. Since 1 Lq ([0, 1]), let be the bounded
R1
linear functional on Lp defined by (f ) = 0 f . Then
Z 1
1
lim (fn ) = lim
fn (x)dx = lim n = 1.
n
0
Therefore, fn does not converge weakly to zero in Lp .

Problem 6.27 (Hilberts


R Inequality)
1
The operator
R 1/p T f (x) 1= 0 (x + y) f (y)dy satisfies ||T f ||p Cp ||f ||p for 1 < p < , where
Cp = 0 x
(x+ 1) dx. (For those who know about contour integrals: Show that Cp = csc(/p).)
Solution:
6

Define a function K on (0, ) (0, ) by K(x, y) = (x + y)1 . Then K is a Lebesgue measurable


function such that K(x, y) = 1 K(x, y) for all > 0. In order to apply Theorem 6.20, we must
R
R1
verify that 0 |K(x, R1)|x1/p dx is finite. This follows from p > 1, and the facts that 0 x <

when 0 < < 1 and 1 x < when 1 < .


Z

1/p

|K(x, 1)|x

dx =

1
0

1
+
x1/p (x + 1)

x1/p (x + 1)

1
x1/p

1
x1/p x

< .

Hence the conditions of Theorem 6.20 are satisfied, and the desired inequality follows directly from
the theorem.

Problem 6.29
Suppose that 1 p < , r > 0, and h is a nonnegative measurable function on (0, ). Then:
Z x
p
 p Z
Z
p
r1
x
h(y)dy dx
xpr1 h(x)p dx,
r
0
0
0
Z
p
 p Z
Z
p
xp+r1 h(x)p dx.
xr1
h(y)dy dx
r
0
0
x
(Apply Theorem 6.20 with K(x, y) = x1 y (0,) (y x), f (x) = x h(x), and g(x) = x h(x) for
suitable , , .)
Solution:
We start by showing the first inequality. Consider K(x, y) = x1 y (0,) (y x) and f (x) =
x h(x), where = (r + 1)/p and = 1 (r + 1)/p. If f
/ Lp (0, ), the right-hand side of the
inequality is infinite and the inequality holds trivially. Hence we assume that f Lp (0, ). It is clear
that K is a Lebesgue measurable function on (0, ) (0, ) such that K(x, y) = 1 K(x, y) for
all > 0. Using the notation from Theorem 6.20, we evaluate C:
Z
Z
Z 1
p
1/p
1
1/p
C=
|K(x, 1)|x
dx =
x (0,) (1 x)x
dx =
x1+(r/p) dx = .
r
0
0
0
By Theorem 6.20, ||T f ||p C||f ||p, where
Z
Z
1

(r+1)/p
Tf =
x y (0,) (y x)x h(x)dx = y
0

h(x)dx.

Then we have the following inequalities:


Z
p
(r+1)/p y

p
y
h(x)dx p C p x1(r+1)/p h(x) p .
Z

(r+1)

Z

0
y

p
 p Z
p
xp(r+1) h(x)p dx.
h(x)dx dy
r
0

and f. Consider
To show the second inequality, we reapply the theorem with different functions K

K(x,
y) = x1 y (0,) (x y) and f(x) = x h(x), where = (1 r)/p and = 1 + (r 1)/p. If
f
/ Lp (0, ), the right-hand side of the inequality is infinite and the inequality holds trivially. Hence
we assume that f Lp (0, ).
We evaluate C:
Z
Z
1/p

C=
|K(x, 1)|x
dx =
0

1/p

(0,) (x 1)x

dx =

By Theorem 6.20, ||T f||p C||f||p , where


Z
Z
1

(1+r)/p
T f =
x y (0,) (x y)x h(x)dx = y
0

p
x1(r/p) dx = .
r

h(x)dx.
y

Then we have the following inequalities:


Z
p
(1+r)/p

p
y
h(x)dx p C p x1+(r1)/p h(x) p .
y

r1

Z

p
 p Z
p
xp+r1 h(x)p dx.
h(x)dx dy
r
0

Problem 6.31 (A Generalized


Inequality)
Pn Holders
Qn
1
1
pj
r
Suppose
that
1

and
p
=
r
=
1.
If
f

L
for
j
=
1,
.
.
.
,
n,
then
j
j
j
1
1 fj L and
Qn
Qn
|| 1 fj ||r 1 ||fj ||pj . (First do the case n=2.)
Solution:
We do the case n = 2 and then proceed by induction. When n = 2, we apply Holders inequality
with conjugate exponents p1 /r and p2 /r.
Z
r
||f1 f2 ||r = |f1 f2 |r || |f1 |r ||p1 /r || |f2 |r ||p2 /r = ||f1 ||rp1 ||f2||rp2 .
The result follows by taking rth roots of both sides.
Suppose theP
result is true for (n 1). First we Holders inequality with conjugate exponents
1
pn /r and (1/r)( 1n1 p1
j ) .
n
r
Y

fj r =
1

|fn |

n1
Y
j=1

|fj |r

Y
n1



r


|fj |r (1/r)(Pn1 p1 )1
|fn | pn /r
1

j=1

Y r
r n1
= fn
fj Pn1
pn

j=1

1
p1
j )

We can
now take rth roots of both sides, and apply the induction hypothesis with r such that
P
n1
(r )1 = 1 p1
j .
n
n
Y
Y
n1

Y











||fj ||pj .
fj r
fj r fn pn
1

j=1

j=1

Problem P
6.38
kp
k
p
f L iff
2 f (2 ) < .
Solution:
Suppose f Lp . Using that f is decreasing, and Theorem 2.15 to exchange an infinite sum and
an integral, we compute the following:
Z 2k

X
X
p
k
k kp
f (2 )
f (2 )2 =
p1 d
p
12
2k1

Z 2k

X
p

f ()p1d
p
12
2k1

Z
p
f ()p1d
=
p
12
Z0
1
=
|f |p <
p
12
where the last step follows from Proposition 6.24. On the other hand, if
have the following:
Z
Z
p
|f | = p
f ()p1d

2kp f (2k ) < , we

=p

Z
X

2k+1

f ()p1d

2k
k

f (2 )

= (2 1)

2k+1

p1 d

2k

f (2k )2kp <

Problem 6.39
If f Lp , then lim0 p f () = lim p f () = 0. (First suppose f is simple.)
Solution:
P
First, suppose f = n1 aj Ej , where the Ej are disjoint P
and (Ej ) < for all j. Let amax =
max{aj } and amin = min{aj }. Then when < amin , f () = (Ej ) < . Hence
9

lim p f () = lim p


(Ej ) = 0.

On the other hand, when > amax , then f () = 0. Therefore, lim p f () = 0.


P
We now take an arbitrary f Lp . We show that there exists a simple function = n1 aj Ej ,
where Ej are disjoint and (Ej ) < for all j, such that ||f ||pp < for every > 0. By Theorem 2.10,
there is a sequence {
|n | |f |. RThen every n Lp ,
Pand
Pnn} of simple functions such that n f a.e.
n
and assuming n = 1 aj Ej where the Ej are disjoint, we see 1 |aj |p (Ej ) = |n |p ||f ||pp < .
Hence (Ej ) < for all j. Since
|f n |p 2p (|f |p + |n |p ) 2p+1 |f |p,
we have that |f n |p is dominated by an L1 function, hence by dominated convergence,
Z
Z
p
lim |f n | = lim |f n |p = 0.
P
Therefore, we can choose = n1 aj Ej , where (Ej ) < for all j, such that ||f ||pp < /2p+1 .
By the above result, for all small enough, we have p () < /2p+1. Now, we can write
f = (f ) + and apply Proposition 6.22(d) and Chebyshevs Inequality:
1
1
lim sup p f () lim sup { p f ( ) + p ( ) }
2
2
0+
0+
p
p
p p
lim sup { 2 ||f ||p + 2 () }
0+

< /2 + /2 = .
Therefore, lim0 p f () = 0. Similarly, for all large enough, we have p () < /2p+1 . The
same calculation shows:
1
1
lim sup p f () lim sup { p f ( ) + p ( ) }
2
2

p
p
p p
lim sup { 2 ||f ||p + 2 () }

< /2 + /2 = .

Additional Problem from Class


R
Let f : X C be measurable, and let E = {p : X |f |p = ||f ||pp < }. It follows from a result in
Folland that the set E is connected.
i) Prove that the function (p) = log(||f ||pp) is convex in the interior of E, and that is continuous
on E.
ii) Can E consist of a single point? Can it be any connected subset of (0, )?
10

iii) Let ||f ||r < for some r < . Prove that limr ||f ||r = ||f || .
Solution:
(i) Let t (0, 1), and p, q E. By Holders Inequality, we know that
Z
|f |pt|f |q(1t) || |f |pt ||1/t || |f |q(1t) ||1/(1t) .
Therefore, we can take the logarithm and obtain
Z

|f |tp+(1t)q
(tp + (1 t)q) = log
 Z

Z


p t
q 1t
|f |
|f |
log
Z
Z


p
|f |q
|f | + (1 t) log
= t log
= t(p) + (1 t)(q).

Next, we show continuity of by showing sequential continuity. Suppose {qn } is a sequence in E


converging to a point p E. If p is in the interior of E, there exists a > 0 such that (p, p+) E
and qn (p , p + ) for all n > N for some N N. Then we have for all n > N:
|f |qn = |f |qn |f |1 + |f |qn |f |1 |f |p+ |f |1 + |f |p |f |1 |f |p+ + |f |p .
R
R
By the Dominated Convergence Theorem, lim |f |qn = |f |p . Therefore, by continuity of the
logarithm, lim (qn ) = (p) and is continuous in the interior of E.
The case when p is at an endpoint of E is similar. First we do the case when qn p (ie p is
a right endpoint of E). Then there exists a > 0 such that (p , p] E and qn (p , p] for all
n > N for some N N. Then we have for all n > N:
|f |qn = |f |qn |f |1 + |f |qn |f |1 |f |p |f |1 + |f |p |f |1 |f |p + |f |p .
By the Dominated Convergence Theorem, lim (qn ) = (p).
When p is a left endpoint of E, then qn p, and there exists a > 0 such that [p, p + ) E and
qn [p, p + ) for all n > N for some N N. Then we have for all n > N:
|f |qn = |f |qn |f |1 + |f |qn |f |1 |f |p+ |f |1 + |f |p |f |1 |f |p+ + |f |p .
By the Dominated Convergence Theorem, lim (qn ) = (p).

(ii) We construct a function f : (0, ) C such that E = {1}. We first use the fact that for > 0,
we have
11

1
(0,1) <
x

if and only if < 1. We then consider the function

X
1
an
f=
1 (0,1) ,
n
(1+1/n)
2
x
n=1

where

an =

1
x(1+1/n)1


(0,1) 1 .

Then f is a series of functions in L1 that converges absolutely, so by completeness, f L1 . However, f


/ Lp for any p > 1.
Next, we use the fact that for > 0, we have
Z
1
(1,) <
x
0
if and only if > 1. We then consider the function
g=

X
bn
1
(1,) ,
n x(1+1/n)
2
n=1

where

bn =

1
x(1+1/n)


(1,) 1 .

Then g is a series of functions in L1 that converges absolutely, so by completeness, g L1 . However, g


/ Lp for any p < 1.
Therefore, the function f + g is in L1 , but not in Lp for any p (0, 1) (1, ).
We now show how to get any interval E = (a, b), again using X = (0, ). Let f (x) = x1/b
when 0 < x < 1 and identically zero otherwise. Let g(x) = x1/a when x > 1 and identically zero
otherwise. Then f + g is such that E = (a, b).

12

(iii) First, suppose f L . Then


lim ||f ||q = lim

Z
Z

|f |qr |f |r

1/q

1/q
qr
||f ||
|f |r
q
Z
1/q
(qr)/q
|f |r
= lim ||f ||

lim

= ||f ||.

Therefore, if the limit exists we have limq ||f ||q ||f ||. Now, for all > 0, there is a set
E X of positive measure such that |f (x)| ||f || for all x E. Then when q > r we have
Z
1/q
|f |q
(||f || )((E))1/q .
||f ||q
E

Now (E) is finite, so taking the limit as q we get limq ||f ||q ||f || . Since this
holds for all > 0, we conclude that limq ||f ||q = ||f ||.
Next, suppose ||f || = . I have a marvelous proof of this case, which this margin is too
narrow to contain.

13

You might also like