You are on page 1of 14

Department of CSE

2012311871
Jaemin Shin
Real Analysis HW # 3
Chapter 10
Exercise 2. A measure space (S, , ) is said to be complete if contains all subsets of sets with measure zero;
i.e., (S, , ) is complete if Y whenever Y Z, Z , and (Z) = 0. In this case, show that if
f is measurable and g = f a.e., then g is also measurable. Is this true if (S, , ) is not complete?
proof ) (1) Let N = {x : f(x) = g(x)}, then we have that N is measurable since g = f a.e. Let N be
the algebra associated with the range of f and g. Let E N. Then we have,
g
1
(E) = f
1
(E) \ N {x N : g(x) E}.
Then we see that g
1
(E) M since {x N : g(x) E} is measurable as it is a subset of N.
Note that weve proven this theorem for a general function f : X Y where X is equipped
with algebra M and Y is equipped with N. For most theorems we need to assume that
Y = R or C.

(2) (i) Let is complete measure. Suppose f is measurable and f = g a.e.


Let N = {x : f(x) = g(x)}. Then for a set E we have
f
1
(E) N
c
g
1
(E) f
1
(E) N.
Since is complete, every subset of N is measurable, so if E is measurable, f
1
(E) is
measurable and both f
1
(E) M
c
and f
1
(E) M are measurable for any M N. In
particular, g
1
(E) is measurable.
(ii) Conversely, if f is measurable and f = g a.e. implies g is measurable, then let N be
a subset of null set. Let E
1
, E
2
be disjoint sets in N, then dene f(x) = e
1
E
1
for all
x X, and let
g(x) =
_
e
1
, when x X \ N
e
2
, when x X
Then f is measurable, f = g a.e., so g is measurable. This gives that g
1
(E
2
) = N is
measurable in X, hence is complete.

1
Exercise 4. If (S, , ) is a measure space, and if f and {f
k
} are measurable and nite a.e. in a measurable set
E, then {f
k
} is said to converge in -measure on E to limit f if
lim
k
{x E : |f(x) f
k
(x)| > } = 0, > 0.
Formulate and prove analogues of theorems (4.21) (4.23).
Theorem 1 (4.21). Let f and f
k
, k = 1, 2, . . . , be measurable and nite a.e. in E. If f
k
f a.e. on
E and |E| < +, then f
k

f on E.
proof ) Observe that for all > 0
{|f
k
f| > } { |f
k
f| }.
An application of the Markove inequality yields
{|f
k
f| > } { |f
k
f| }

E
|f
k
f|d
=
1

( |f
k
f|)1
E
d.
Since (E) < +, the function 1
E
is integrable, dominates the integrand ( |f
k
f|)1
E
, and
Lebesques dominated convergence theorem implies that lim
k

E
( |f
k
f|)d = 0.

Theorem 2 (4.22). If f
k

f on E, there is a subsequence f
k
j
such that f
k
j
f a.e. in E.
proof ) Since f
k

f, for given g = 1, 2, . . . there exists k


k
such that
{|f f
k
| >
1
j
} <
1
2
j
, for k k
j
.
We may assume that k
j
. Let E
j
= {|f f
k
j
| >
1
j
} and H
m
=

j=m
E
j
.
Then, (E
j
) < 2
j
, (H
m
)

j=m
2
j
= 2
m+1
, and |f f
k
j
|
1
j
in E E
j
.
Thus, if j m, |f f
k
j
|
1
j
in E H
m
. So that f
k
j
f in E H
m
, since (H
m
) 0.
It follows that f
k
j
f a.e. in E.

Theorem 3 (4.23). A necessary and sucient condition that {f


k
} convergence in measure on E is that for
each > 0,
lim
k,l0
{x E : |f
k
(x) f
l
(x)| > } = 0.
proof ) (i) The necessity follows from the formula
{x E : |f
k
f
l
| > } {x E : |f
k
f| >

2
} {x E : |f
l
f| >

2
}
and the fact that the measures of the sets on the right tend to zero as k, l if f
k

f.
(ii) For every integer j we can nd N
j
such that for k, l N
j
,
{x E : |f
k
f
l
|
1
2
j
} <
1
2
j
,
and we may assume that for each j, N
j+1
> N
j
.
Let
E
j
= {x E : |f
N
j
f
N
j+1
|
1
2
j
}.
Then if x /

_
j=l
E
j
, we have for r > s l
|f
N
r
f
N
s
|
r

i=s+1
|f
N
i
f
N
i1
| <

i=s+1
1
2
i
=
1
2
s
. (1)
2
So f
N
j
is a Cauchy sequence for each x / limsup E
j
=

l=1

_
j=l
E
j
. But, for all l,
(limsup E
j
)
_

_
j=l
E
j
_

j=l
1
2
j
=
1
2
j1
.
So f
N
j
converges a.e. to some measurable function f. Also from (1) we have that f
N
j

f
uniformly outside

_
j=l
E
j
, and hence, for every positive ,
{x E : |f
N
j
f| >

2
} 0 as j . (2)
But
{x E : |f
k
f| > } {x E : |f
k
f
N
j
| >

2
} {x E : |f f
N
j
| >

2
}.
If k and N
j
are suciently large, the measure of the rst set on the right is arbitrarily small,
as f
k
is Cauchy in measure. But the second set has been shown to have arbitrarily small
measure by (2) and the result follows.

3
Exercise 6. (a) If f
1
, f
2
L(d) and

E
f
1
d =

E
f
2
d for all measurable E, show that f
1
= f
2
a.e. ().
proof ) Since f
1
and f
2
are d-measurable functions, we have E {f
1
f
2
}, E {f
1
< f
2
} d for
all E d. Thus

E
|f
1
f
2
|d =

E{f
1
f
2
}
(f
1
f
2
)d +

E{f
1
<f
2
}
(f
2
f
1
)d,
while

E{f
1
f
2
}
(f
1
f
2
)d =
_

E{f
1
f
2
}
f
1
d

E{f
1
f
2
}
f
2
d
_
= 0
by the linearity of the integral and because of the assumption. The other term on the right-
hand side can be treated similarly, and the conclusion follows from

|u|d = 0 |u| = 0 a.e. ({u = 0}) = 0.

(b) Prove the uniqueness of f and in (10.40).


proof ) If (E) < , by Lebesque decomposition we have (A) =

A
fd +(A).
Assume we have another decomposition (A) =

A
f
1
d +
1
(A). Then,

A
(f f
1
)d = (A)
1
(A).
The left side is absolutely continuous and the right side is singular, so they must be same.
Hence f and is unique.
If (E) = 0, then we have E =

E
j
, and (E
j
) < . So similarly we can get f and as
unique.

(c) Let be -nite, and let f


1
, f
2
L
p

(d), 1/p + 1/p

= 1, 1 p . If

f
1
gd =

f
2
gd
for all g L
p
(d), show hat f
1
= f
2
a.e.
proof ) Because f
1
, f
2
L
p

(d), g L
p
(d), so

f
1
g
_

f
p

1
dx
_
1/p
_

g
p
dx
_
1/p
< .
Therefore, f
1
g L(d).
According to (1) we know that f
1
g = f
2
g a.e. (), while (1) is true for any g L
p
(d).
Thus,
f
1
= f
2
, a.e. ().

4
Exercise 8. Show that for 1 p < , the class of simple functions vanishing outside sets of nite measure is dense in
L
p
(d).(i.e. for every u L
p
(d) one can nd a sequence (f
j
)
jN
such that lim
j
f
j
u
p
= 0.)
proof ) Assume rst that u L
p
(d) is positive. Then we can nd an increasing sequence (f
j
)
jN
of
positive simple functions with sup
jN
f
j
= u.
Since 0 f
j
u, we have f
j
L
p
(d) as well as sup
jN

|f
j
|
p
d =

|u|
p
d. We can now apply
Riesz theorem and deduce that lim
j
f
j
u
p
= 0.
For a general u L
p
(d), we consider its positive and negative parts u

and construct, as before,


sequences g
j
, h
j
L
p
(d) with g
j
u
+
and h
j
u

in L
p
(d).
But g
j
h
j
L
p
(d), and
u (g
j
h
j
)
p
u
+
g
j

p
+u

h
j

p
j
0
nishes the proof.

5
Exercise 10. If is a set function whose Jordan decomposition is = V V , dene

E
fd =

E
fdV

E
fdV ,
provided not both integrals on the right are innite with the same sign. If V is total variational of
on E, and if |f| M, prove that |

E
fd| MV .
proof ) If V is total variational of on E, then
V = || = V +V .
Thus

E
fd

E
fdV

E
fdV

E
fdV

E
fdV

E
|f|dV +

E
|f|dV
M
_
E
dV +

E
dV
_
M|(E)|
= MV.

6
Exercise 12. Give an example of a pair of measures and such that is absolutely continuous with respect to ,
but given > 0, there is no > 0 such that (A) < for every A with (A) < . [Thus, the analogue
for measures of (10.34) may fail.]
Prove the analogue of (10.35) for mutually singular measures and .
proof ) Choose =

n=n
0
(a
1
n
, a), =

_
n=n
0
(a +
1
n
, b), where a < b.
It is easy to see that for any A, if (A) = 0, then (A) = 0.
And for , when n increases then decreases. If n
0
, that is (A) < , for any but for ,
we can choose (A) > .
Theorem 4 (10.35). An additive set function is singular on E with respect to if and only if given
> 0, there is a measurable subset E
0
of E such that (E
0
) < and V (E E
0
; ) < .
Two measures and are mutually singular if and only if given > 0, there are disjoint measurable
subsets E
1
and E
2
with E = E
1

E
2
and (E
1
) < , (E
2
) < .
proof ) () Because and are mutually singular, so there exists E
1
such that (E
1
) = 0 and
(E E
1
) = 0. So (E
1
) < , (E
2
) = (E E
1
) < .
() Choose for each k = 1, 2 . . . , a measurable E
k
< E with (E
k
) < 2
k
and (EE
k
)2
k
.
Let E
1
= limsup E
k
, since z

k=m
E
k
for every m. It follows that (E
1
) = 0.
Moreover,
(E E
1
) =
_
E limsup E
k
_
=
_
liminf (E E
k
)
_
liminf (E E
k
) = 0.
So, and are mutually singular.

7
Exercise 14. Complete the proof of (10.44) for p = 1.
proof ) For p = 1, we can get g

from (10.43). So we need to prove g

for > 0.
Let A = {x : |g| > g

}. Then 0 < (A) , and there exists an increasing sequence


of sets A
n
of nite measure whose union is A such that (A
n
) (A). So we can nd a subset
B A, such that 0 < (B) < .
Let f = sign(g)

B
(B)
. Then f L
1
with f
L
1 = 1.
Therefore,
(f) =

E
fg d =
1
(B)

B
|g| d g

.
Thus, g

8
Exercise 16. Consider a convolution operator Tf(x) =

R
n
f(y)K(x y)dy with K 0. If Tf
p
Mf
p
, 1
p , for f L
p
(R
n
, dx), show that Tf
p
Mf
p
, 1/p + 1/p

= 1.
[Use Exercise 15 to write Tf
p
= sup
g
p
1
|

R
n
(Tf)g dx|, and note that

R
n
(Tf)(x)g(x)dx =

R
n
(T g)(y)f(y)dy,
where g(x) = g(x).]
proof )
(1) Since

R
n
(Tf)(x)g(x) dx =

R
n
(T g)(y)f(y) dy, we can get
Tf
p
= sup
g
p
1

R
n
(Tf)(x)g(x) dx

= sup
g
p
1

R
n
(T g)(y)f(y) dy

sup
g
p
1
_
R
n
(T g)
p
(y)dy
_1
p
_
f
p

(y)dy
_ 1
p

= sup
g
p
1
_
R
n
(T g)
p
(y) dy
_1
p
f
p

sup
g
p
1
M g
p
f
p
.
Since, g
p
= g
p
1, then
Tf
p
Mf
p
.

(2) Setting z = x/y, and let q be the conjugate exponent to p, we have


0
|K(x, y)f(x)|dx =


0
|K(yz, y)f(yz)|ydz =


0
|K(z, 1)f
z
(y)|dz
where f
z
(y) = f(yz); moreover,
f
z

p
=
_

0
|f(yz)|
p
dy
_
1/p
=
_

0
|f(x)|
p
z
1
dx
_
1/p
= z
1/p
f
p
.
Therefore, by Minkowskis inequality for integrals, Tf exists a.e. and
Tf
p


0
|K(z, 1)|f
z

p
dz = f
p


0
|K(z, 1)|z
1/p
dz = Cf
p
.
Finally, setting u = y
1
, we have


0
|K(1, y)|y
1/q
dy =


0
|K(y
1
, 1)|y
1(1/q)
dy
=


0
|K(u, 1)|u
1/p
du = C,
so the same reasoning shows that Sg is dened a.e. and that Sg
q
Cg
q
.

9
Exercise 18. Give an example to show that (L

= L
1
.
_
Consider L

[1, 1] with Lebesque measure, and let C be the subspace of continuous functions on
[1, 1] with the sup norm. Dene l(f) = f(0) for f C. Then l is a bounded linear functional on C. so
by the Hahn-Banach theorem (see, e.g., p.62 of N. Dunford and J. T. Schawrtz, Linear Operators, Part
1, 4th printing, Interscience, New York, 1967.), l has an extension

l (L

. If there were a function


g L
1
[1, 1] such that

l(f) =

1
1
fgdx for f L

[1, 1], then we would have f(0) =

1
1
fgdx
for f C. Show that this implies that g = 0 a.e., so that l 0. The functional l is called the
-function.
_
proof ) Because C
1
[1, 1] is dense in L
1
[1, 1], so for g L
1
there exists f
n
C
1
[1, 1] such that f
n
g.
Then,
f
n
(0) =

1
1
f
n
g dx =

1
1
f
2
n
= (f
n
).
_
f
n
C
1
[1, 1] so f
n
is bounded, f
n
g =

1
1
f
n

1
1
g
_
Since is a linear functional, and (f
n
) =

1
1
f
2
n
dx, so f
n
= 0, a.e.
Therefore, g = 0, a.e.

10
Exercise 20. Under the hypothesis of (10.49), prove that
lim
h0
1
(Q
x
(h))

Q
x
(h)
|f(y) f(x)|d(y) = 0 a.e. ().
proof ) Let {r
k
}

k=1
be the set of rational numbers for any k, the function |f(y) r
k
| is clearly integrable
on R
n
. Therefore, we have
lim
h0
1
(Q
x
(h))

Q
x
(h)
|f(y) r
k
|d(y) = |f(x) r
k
|, a.e. x R
n
. (3)
Let z
k
be the set such that (3) is not valid. And |z
k
| = 0, set z =

k=1
z
k
, |z| = 0 if x / z, then
by the inequality
1
(Q
x
(h))

Q
x
(h)
|f(y) f(x)|d(y)
1
(Q
x
(h))

Q
x
(h)
_
|f(y) r
k
| +|r
k
f(x)|
_
d(y)

c
(Q
x
(h))

Q
x
(h)
|f(y) r
k
|d(y) +
c
(Q
x
(h))

Q
x
(h)
|r
k
f(x)|d(y)
=
c
(Q
x
(h))

Q
x
(h)
|f(y) r
k
|d(y) +c|r
k
f(x)|.
Then if we take lim
h0
on both sides,
lim
h0
1
(Q
x
(h))

Q
x
(h)
|f(y) f(x)|d(y) 2c|f(x) r
k
|.
By choosing r
k
approximating f(x), we obtain
lim
h0
1
(Q
x
(h))

Q
x
(h)
|f(y) f(x)|d(y) = 0, for all x / z.

11
Exercise 22. Let be a nite measure on A, let f be measurable and bounded on A, and let be convex in an
interval containing the range of f. Prove that

A
fd

A
d
_

A
(f)d

A
d
.
[This is Jensens inequality for measures.]
proof ) If the right-hand side of above inequality is innite, there is nothing to show. Therefore we may
assume

A
(f)d < . Since (x) is convex, we nd for any (x) := x + (x) that

A
(f)d

A
d

A
(f +)d

A
d
=

A
fd

A
d
+ =
_

A
fd

A
d
_
and the inequality follows from a property of convex function
(x) = sup{(x) : (z) = z + (z), z (a, b)}
if we pass to the sup over all linear functions satisfying .

12
Exercise 24. Let (S, , ) be a nite measure space, and let f be -meausrable and integrable over S. Let
0
be a -algebra satisfying
0
. Of course, f may not be
0
-measurable. Show that there is a unique
function f
0
which is
0
-measurable such that

fgd =

f
0
gd for every
0
-measurable g for which
the integrals are nite. The function f
0
is called conditional expectation of f with respect to
0
, denoted
f
0
= E(f|
0
). [Apply the Radon-Nikodym theorem to the set function (E) =

E
fd, E
0
.]
proof ) If f is a characteristic function then

E
fgd = (E) =

E
f
0
gd.
Since E
0
where f
0
is a characteristic function.
By linearity, it is true for all simple functions by setting f
0
= f|
E
then for general nonnegative
functions.
Setting f
0
= f|
E
then sup

E
=

E
fgd =

E
f
0
gd and we can expand it to f L
1
by applying
it to positive negative parts for real and imaginary parts.

13
Exercise 26. (Hardys inequality) Let f 0 on (0, ), 1 p < , d(x) = x

dx and d(x) = x
+p
dx on (0, ).
Prove there exists a constant c independent of f such that
(i)

0
(

x
0
f(t)dt)
p
d(x) c

0
f
p
(x)d(x) ( < 1)
(ii)

0
(

x
f(t)dt)
p
d(x) c

0
f
p
(x)d(x) ( > 1)
[ For (i), (

x
0
f(t)dt)
p
cx
p1

0
f(t)t

dt by Holders inequality, provided p 1 > 0. Multiply


both sides by x

, integrate over (0, ), change the order of integration, and observe that an appropriate
exists since < 1. ]
proof ) Let K(x, y) = x
1
y

y>x
. Then K(x, y) =
1
K(x, y). With h(x) = x

f(x) and g(y) =


y

f(y), we have
Th(x) =


0
K(x, y)h(x)dx =


0
x
+1
y

y>x
f(x)dx = y

y
0
x
+1
f(x)dx
and
Sg(y) =


0
K(x, y)g(y)dy =


0
x
1
y

y>x
f(y)dy = x
1


x
y

f(y)dy.
By Exercise 16.,


0
y
p
_
y
0
x
+1
f(x)dx
_
p
dy =


0
|Th|
p
dy = Th
p
p
C
p
h
p
p
= C
p


0
x
p
f(x)
p
dx
and


0
x
q(1)
_

x
y

f(y)dy
_
q
dx =


0
|Sg|
p
dx = Sg
q
q
C
q
g
q
q
= C
q


0
y
q
f(y)
q
dy
where
C =


0
|K(x, 1)|x
1/p
dx =


0
x
1

1>x
x
1/p
dx =

1
0
x
11/p
dx =
1
1/p
.
Letting = 1 and =
r+1
p
in the rst, and letting = , = 1 +
r1
p
with the role of p and
q interchanged in the second, yields the desired equalities.

14

You might also like