Real Analysis qual study guide

James C. Hateley

1. Measure Theory
Exercise 1.1. If A R and  > 0 show open sets O R such that m (O) m (A) + .
[
Proof: Let {In } be a countable cover for A, then A
In . Since m (O) m (A) + . This implies
n=1

that
(

m (O)  m (A) where m (A) =

inf
S

In

)
l(In )

n=1

If l(Ik ) = for some k then there is nothing to show, so suppose (an , bn ) = In then l(In ) < , n.
Let On = (an + 2n , bn ) then we have
X

l(On ) = bn an 2n 
X
X
l(On ) =
bn an
2n  =
[
m ( On )  m (A)

l(In )
X
bn an 

So let O =

On , then m (O)  m (A) O R st m (O) m (A) + 

Exercise 1.2. If A, B R, m (A) = 0, then m (A B) = m (B)

Proof: m (A B) m (A) + m (B), and m (B) m (A B), hence we have
m (B) m (A B) m (A) + m (B) = m (B)
m (A B)

= m (B)

Exercise 1.3. Prove E M iff  > 0, O R open, such that E O and m (O\E) < 
Proof: () O\E = E c O implies that m (O\E) = m (E c O), but we have
m (O) = m (E c O) + m (E O)

So suppose m (E) < m (E c O) = m (O) m

S (E O). Let In be a countable cover for E, so
n
In = (an , bn ). Let On = (an , bn + 2 ) and let O = On . Then
X
X
X
m (O) =
l(On ) =
2n  + bn an =  +
bn an , and m (E O) = m (E)

since E O. So we have
=

m (E O)

m (E O) = m (E)

l(In )

bn an

X
l(On )
l(In )
X
X
+
bn an
bn an = 

O R open, st E O and m (O\E) 

() Conversely, suppose  > 0, O R, such that E O and m (O\E) <  and that O M. Then
m (O) = m (E c O) + m (E O), but m (E c O) = m (O\E) < 
This implies that
m (O) = m (E O) + 

m (O) = m (E) + 
1

EM

Exercise 1.4. Prove E M iff  > 0 F R closed, such that F E and m (E\F ) < 
Proof: () E\F = F c E this implies that m (E\F ) = m (F c E), but we have
m (F ) = m (F c E) + m (E F )
So suppose m (E) < m (F c E) = m (F ) m (FS E). Let In be a countable cover for E,
where In = (an , bn ). Let Fn = [an , bn 2n ] and let F = Fn . Then we have
X
X
X
m (F ) =
l(Fn ) =
bn an 2n  =
bn an ,
and m (E F ) = m (F ), since F E. So

X
X
m (E F ) = m (F )
l(In ) =
bn an
X
X
X
X
m (E F )
l(In )
l(Fn ) =
bn an
bn an +  = 

F R Closed, st F E and m (E\F ) 

() Conversely, suppose  > 0, F R, such that F E and m (E\F ) <  and that F M. Then
m (E) = m (F c E) + m (E F ),
but m (F c E) = m (E\F ) < . This implies that
m (E) m (F E) + 

m (E) m (F ) + 

EM

Vitali Let E be a set of finite outer measure and a collection of intervals that cover E in the sence
of Vitali. Then, given  > 0 there is a finite disjoint collection {IN } of intervals in such that
!
N
[

In < 
E\
n=1

Exercise 1.5. Does there exists a Lebesgue measurable subset A of R such that for every interval (a, b)
we have (A (a, b)) = (b a)/2?
Proof: First suppose that there is such a mesurable set A such that 0 6= (A (a, b)) = (b a)/2.
Then there exsits an open set O such that A O and (O\A) < , so let  = /2. Now O is open, so
there are disjoint intervals (xk , yk ) such that O is a countable union of these intervals. So

[
[
[(xk , yk ) (a, b)] = (ckl , dkl ).
O (a, b) =
k=1

Hence (O (a, b)) =

l dkl ckl , and we have

A O (a, b) = A (a, b) =

[A (ckl , dkl )]

Now
= (A (a, b)) =

1X
(dkl ckl )
2
l

but
X
(dkl ckl )

(O (a, b))

((O\A) (a, b)) + (A (a, b))

1X
(O\A) +
(dkl ckl )
2
l
1X
< +
(dkl ckl )
2
=

But this implies that

/2 = 

1X
(dkl ckl )
2
l

So (A) = 0. which implies that (Ac ) = . Now if there were to exsits such a set A we have (Ac ) = 0,
and so
1
b a = ((a, b)) = (A (a, b)) + (Ac (a, b)) = (Ac (a, b)) = (b a)
2
So there cannot exist such a set  .

Exercise 1.6. Assume that E [0, 1] is measurable and for any (a, b) [0, 1] we have
1
(E [a, b]) (b a)
2
Show that (E) = 1.
Proof: By the previous problem, using the same proof, we know that (E c ) = 0. So the result is
shown.
Exercise 1.7. Let E1 , . . . , En be measurable subsets of [0, 1]. Suppose almost every x [0, 1] belongs
to at least k of these subsets. Prove that atleast one of the E1 , . . . , En has measure of at least k/n.
Proof: Suppose not, then for each i we have (Ei ) < k/n. Define a function f (x) as follows.
n
X
Ei
f (x) =
i=1

where Ei denotes the characteristic function of Ei . Now since all most all x [0, 1] are in at least k
of the Ei we have f (x) k almost everywhere in [0, 1]. Now
Z
Z
n
n Z
X
X
Ei
Ei dx =
k=
k dx
f (x) dx =
[0,1]

[0,1]

i=1

[0,1]

i=1

But this implies that

n
X

Ei <

i=1

n
X
k
=k
n
i=1
k
n .

Which is a contradiction, hence at least one Ei has (Ei )

Exercise 1.8. Consider a measure space (X , A, ) and a sequences of measurable sets En , n N, such
that

X
(En ) <
n=1

Show that almost every x X is an element of at most finitely many En0 s.

Proof: It suffices to show that (x : x Enk ) = 0. So consider the following
!
m
\
lim x : x
Enk
m

k=1

If we have shown the above limit is zero, then were done. To see this look at the following sum,
!

X
\
X
x:x
Enk <
(En ) <
N =1

n=1

k=1

and hence
lim x : x

m
\

!
E nk

=0

k=1

Therefore almost every x X is an element of at most finitely many En0 s

.

Exercise 1.9. Consider a measure space (X , A, ) with (X ) < , and a sequences fn : X R of

measurable functions such that lim fn (x) = f (x) for all x X . Show that for every  > 0 there exists
n

a set E of measure (E)  such that fn converges uniformly to f outside the set E.
Proof: This is Ergoroffs theorem. See below.
Theorem (Egoroff s) If fn is a sequence of measurable functions that converge to a real-valued
function f a.e. on a measurable set E of finite measure, then given > 0, there is a subset A of E with
(A) < such that fn converges to f uniformly on E\A
Proof: Let > 0, then for each n, there exists a set An E with An < 2n , and there is an
Nn such that for all x
/ An and k An we have |fk (x) f (x)| < 1/n. Let A = An , then by
construction A E and A < . Choose n0 such that 1/n0 < . Now if x
/ A and k Nn0 then
|fk (x) f (x)| < 1/n0 < . Therefore fn converges uniformly on E\A.

Exercise 1.10. Let g be an absolutely continuous monotone function on [0, 1]. Prove that if E [0, 1]
is a set of Lebesgue measure zero, then the set g(E) = {g(x) : x E} R is also a set of Lebesgue
measure zero.
Proof: Let E [0, 1] with zero measure, then for any epsilon  > 0, there exists an open cover O for
E, such that (O\E) < . Now O being open in [0, 1] implies that O = (an , bn ), where (an , bn ) are
disjoint. Now by absolutely continuity of g(x) we have

X
X
> 0 s.t.
(In ) <

|g(In [0, 1])| <

n=1

n=1

Now g(E) |g(In [0, 1])| which implies that (g(E)) < , so given an there exists a > 0 such
that the above hold, then let = . Since is arbitrary we have (g(E)) = 0 
Remark: The above problem (1.10) is commonly refered to as Lusins N condition.

Exercise 1.11. Suppose f is Lipschitz continuous in [0, 1]. Show that

(a) (f (E)) = 0 if (E) = 0.
(b) If E is measurable, then f (E) is also measurable.
Proof: For part (a) if f is Lipschitz continuous then it is absolutely continuous, and so if (E) = 0,
then (f (E)) = 0 (see above proof).
For part (b) Let E be a measurable set and let  > 0. Now there exists an open set O such that
(O\E) < , where O is a disjoint union of intervals In = (an , bn ). Now since f is absolutely continiuous,
it can be approximated by simple functions, namely In . Choose these functions such that





X


cn In  < 
f


n=1

Now (In ) = bn an > 0, so it is measurable. Let R, then the f (E) is measurable if {x : f (x) }
is a measurable set for any R. but we have now
{x : f (x) } {x : In +  }
We know simple functions are measurable, and our choice of simple functions approximates f (x),
therefore f is measurable  .
Theorem (Lusins) Let f be a measurable real-valued function on an interval [a, b]. Then given
> 0, there is a continuous function on [a, b] such that {x : f (x) 6= (x)} <

Proof: Let f (x) be measurable on [a, b] and let > 0. For each n, there is a continuous function hn
on [a, b] such that
{x : |hn (x) f (x)| 2n2 } < 2n2
Denote these sets as En . Then by construction we have
|hn (x) f (x)| < 2n2 , for x [a, b]\En
Let E = En , then E < /4 and {hn } is a sequence of continuous, thus measurable, functions that
converges to f on [a, b]\E. By Egoroffs theorem, there is a subset A [a, b]\E such that A < /4 and
hn converges uniformly to f on [a, b]\(EA). Thus f is continuous on [a, b]\(EA) with (EA) < /2.
Now there is an open set O such that (E A) O and (O\(E A)) < /2. Then we have f is continuous on [a, b]\O, which is closed. Hence there exists a that is continuous on (, ) such that
f = on [a, b]\O, where {x : f (x) 6= (x)} (O) <

Exercise 1.12. Prove the following statement. Supoose that F is a sub--algebra of the Borel -algebra
on the real line. If f (x) and g(x) are F -measurable and if
Z
Z
f dx =
g dx, A F
A

Then f (x) = g(x) almost everywhere.

Proof: Let denote the Lebesgue measure on the Borel sets. Now since both f and g are F -measurable,
for any n 1, the sets
An = {x : f (x) g(x) 1/n},

Bn = {x : g(x) f (x) 1/n}

are both measurable and contained in F . Now we also have

\
\
Bn
An , B = {x : g(x) f (x) > 0} =
A = {x : f (x) g(x) > 0} =
n=1

n=1

contained in F since F is a -algebra. Now using the convenetion that = 0, we have

Z
f g dx = 0
A
Z
If (A) > 0 then as f g > 0 implies by that
f g dx > 0, which is a contradiction. Hence we have
A

(A) = 0. By the same argument also have

Z
g f dx = 0

(B) = 0.

Now A B = and A B is the set of points where f (x) 6= g(x), hence f = g almost everywhere

.

Exercise 1.13. Let E R. Let E 2 = {e2 : e E}

(a) Show that if (E) = 0, then (E 2 ) = 0
(b) Suppose (E) < , it it true that (E 2 ) <
Proof: For part (a) consider the intervales In = [n, n+1] for in Z. Now consider the function f (x) = x2 .
If pn = (ak , bk ) is an open subset of In such that for < 0
(pn ) <

n
X
k=1

|f (bk ) f (ak )| =

N
X

|b2k a2k | (2|n| + 1)

k=1

Hence f (x) is absolutely continuous on In . Now a function is absolutely continuous on an interval I if

and only if the following are satisfies:
f is continuous on I

f is of bounded variation on I
f satisfies Lusins (N ) condition, or for every subset E of I such that (E) = 0, (f (E)) = 0.
Remark: The above condition for absolute continuity is the Banach-Zarecki Theorem.
Now define En = E In , then En In and hence by Lusins (N ) condition (f (En )) = 0. Now the set
f (En ) is given by
f (En ) = {e2 : e E In }
Now
E2 =

{e2 : e E In } =

nZ

f (En )

nZ

and so
(E 2 )

(En ) =

nZ

(En ) = 0

nZ

For part (b), the statement is not always true. For each n N, let En = [n, n + n3/2 ), then for each
(En ) = n3/2 . Now if E = En , then
(E) =

(En ) =

n=1

X
1
3/2
n
n=1

Now En2 = [n2 , n2 + 2n1/2 + n3 ), and so (En2 ) = 2n1/2 + n3 n1/2 . Also E 2 = En2 , and the
sets En2 are mutually disjoint. Hence
2

(E ) =

(En2 )

n=1

X
1

=
1/2
n
n=1

Exercise 1.14. Suppose a measure is defined on a -algebra M of subset of X , and is the

corresponding outer measure. Suppose A, B X . Then A B if (AB) = 0. Prove that is an
equivalence relation.
Proof: For symmetry we have, by definition, AB = (A B)\(A B) = (B A)\(B A) = BA,
and so if (AB) = 0, then (BA) = 0. Hence A B if and only if B A.
For reflexivity, we have (AA) = A\A = , hence A A.
For transitivity, let A, B, C X . First notice, by element chasing, AC (AB) (BC), and so
we have
0 (AB) = ((AB) (BC)) (AB) + (BC)
Now if A B and B C, then (AB) = (BC) = 0, and so (AB) = 0, hence A C.
Therefore is an equivalence relation on X  .

Exercise 1.15. Let (X , M, ) be a measure space.

(a) Suppose (X ) < . If f and fn are measurable functions with fn f almost everywhere,

[
prove that there exists sets H, Ek M such that X = H
Ek , where (H) = 0 and fn f
k=1

uniformly on each Ek
(b) Is the result of (a) still true if (X , M, ) is -finte?
Proof: For part (a), since (X ) < and fn f almost everywhere, by Egoroffs theorem, for any

k N, there is Hk M such that (Hk ) < 1/k and fn f uniformly on Ek = Hkc . Now define
H =
k=1 Hk , then H Hk , and so 0 (H) 1/k for all k, hence (H) = 0. Now
!c

[
[
\
c
Ek =
Hk =
Hk
= Hc
k=1

k=1

k=1

and so

X =H

!
Ek

k=1

where fk converges uniformly to f on any Ek

.

For part (b), the statement is true. Since X is -finite, we can write X as a disjoint union of finite sets,
i.e.

[
X =
Xn where (Xn ) < n Xi Xj = for i 6= j
n=1

Now for each Xn apply part (a). Then we have

Xn = Hn

Ek,n with (Hn ) = 0

k=1

Let H =
n=1 Hn , then (H) =

(Hn ) = 0. So we have

n=1

X =

Xn

Hn

n=1

n=1

!!
Ek,n

k=1

!
Hn

n=1

= H

Ek

n,k=1

Ek,n

n,k=1

Now H has measure zero and {Ek,n }

n,k=1 is a countable collection of open sets for which fn f
uniformly  .

Exercise 1.16. Suppose fn is a sequence of measurable functions on [0, 1]. For x [0, 1] define
h(x) = #{n : fn (x) = 0} (the number of indicies n for which fn (x) = 0. Assuming that h <
everywhere, prove that the function h is measurable.
Proof: First consider the measure space ([0, 1], [0, 1], ), where is the Lebesgue measure. Since fn
is measurable for all n we know that the set {x : fn (x) = } is measurable, for R. In particular,
the set {x : fn (x) = 0} is measurable. Now we have

{x : fn (x) = 0}

n=1

is measurable with respect to , since it is the countable union of measurable sets. Now consider the
measure space (N, (N), ) where is the counting measure. Now we know that
h(x) = {n : fn (x) = 0} <

So consider the following:

(
{x : h(x) = }



)

[


x : #  fn (x) = 0 =

n

(
=

x:

n=1

x:

)
{n : fn (x) = 0} =
)
{n : fn (x) = 0} <

n=1

[0, 1]
Hence the function h(x) is measurable

.

2. Lebesgue Integration
Exercise 2.1. Consider the Lebesgue measure space (R, M, ). Let f be an extended real-valued M measurable function on R. For x R and r > 0 let Br (x) = {y R : |y x| < r}. With r > 0 fixed,
define a function g on R by setting
Z
g(x) =
f (y)(dy) for
xR
Br (x)

(a) Suppose f is locally -integrable on R. Show that g is a real-valued continuous function on R.

(b) Show that if f is -integrable on R then g is uniformly continuous on R.
Proof: If we show part (b), then part (a) follows by the same argument. Let x R. Now if f is
integrable on R2 so is |f |. Hence if  > 0, there is > 0 such that if (A) < , then we have
Z

|f | dy .
2
A
Now as B(x, r) and B(y, r) are open balls with area r2 with centers offset by |y x|, we have that
(B(x, r)\B(y, r)) = (B(y, r)\B(x, r)) 0 as y x
Hence given > 0, there is an > 0 such that if |y x| < , then
(B(x, r)\B(y, r)) = (B(y, r)\B(x, r)) <
So for |y x| < , we have
Z

|g(x) g(y)|

|f | d +
B(y,r)\B(x,r)

That is, g(x) is uniformly continuous on R

|f | d <
B(x,r)\B(y,r)



+ < .
2 2

.

Theorem (Jensens Inequality) If is a convex function on R and f an integrable function on

[0, 1].
Z

Z
(f (t)) dt
f (t) dt .
Proof: Let
Z
=

f (t) dt,

y = m(x ) + (a)

Then y is the equation of a supporting line at . Now we have

Z
(f (t)) m(f (t) ) + ()
(f (t)) dt () dt

Theorem (Bounded Convergence) Let fn be a sequence of measurable functions defined on a set

E of finite measure, and suppose that there is a real number M such that |fn | M for all N and all
x. If f (x) = lim fn (x) pointwise in E, then
Z
Z
f = lim
fn .
E

Proof: Let  > 0, thn there is an N and a measurable set A E with A <

n N and x E\A we have |fn (x) f (x)| < 2(E)
. Now,

Z
Z
Z 






 fn
f  =  fn f 

E
E
ZE

|fn f |
ZE
Z
=
|fn f | +
|fn f |
E\A

<
Z

such that for all



+ =
2 2

Z
fn

Therefore we have


4M

.

Exercise 2.2. Suppose fn is a sequence of measurable functions such that fn converges to f almost
everywhere. If for each  > 0, there is a C such that
Z
|fn | dx < .
|fn |>C

Show that f is integrable on [0, 2]

Proof: First the interval [0, 2], is not important. The result can be shown for any finite interval. Fix
 > 0, now if f is to be integrable, then so is |f |. Let C be such in the hypothesis, by Fatous lemma
we have
Z 2
Z 2
|f | dx lim inf
|fn | dx
0
0
!
Z
Z
=

|fn | dx +

lim inf
[0,2]{|fn |>C}

|fn | dx
[0,2]{|fn |C}

 + C(0, 2)
Z

|f | dx is bounded and hence f is integrable

Therefore

.

Theorem (Fatous Lemma) If fn is a sequence of nonnegative measurable functions and fn (x)

f (x) almost everywhere on a set E, then
Z
Z
f lim inf
fn .
E

Proof: Since the integral over a set of measure zero is zero, (WLOG) we can assume that the converges
is everywhere. Let h be a bounded measurable fuction which is not greater that f and which vanishes
outside a set A E of finite measure. Define a function hn , by
hn (x) = min{h(x), fn (x)}.
Then hn is bounded by the bound for h and vanishes outside A. Now hn h pointwise in A, hence
we have by the bounded convergence theorem
Z
Z
Z
Z
h=
h = lim
hn lim inf
fn .
E

10

Taking supremum over h gives us the result

.

Theorem (Monotone Convergence) Let fn be an increasing sequence of nonnegative measurable

functions, and let f = lim f a.e. Then
Z
Z
f = lim fn .
Proof: By Fatous lemma we have
Z

Z
f lim inf

fn .
E

Now for each n, since f is monotone, we have fn f , and so

Z
Z
Z
Z
f
f

lim sup f
f

Z
f lim

fn

.

Remark: Let the positive part of f be denoted by f + (x) = max{f (x), 0}, and the negative part be denoted by f (x) = max{f (x), 0}. If f is measurable then so are f + and f . Futhermore f = f + f
and |f | = f + + f .

Exercise
R 2.3. Let f be a real-valued continuous function on [0, ) such that the improper Riemann
integral 0 f (x) dx converges. Is f Lebesgue integrable on [0, )?
Proof: f does not have to be Lebesgue integrable. Let n 0 and define a function fn as follows

1
4

n+1 x x [2n, 2n + 2 ]
4
1
fn (x) = n+1 x x [2n + 2 , 2n + 32 ]

4
3
n+1 x x [2n + 2 , 2n + 2]
Now fn is continuous on [0, ) and when considering Riemann integration, we have
Z 2n+1
Z 2n+2
Z
1
and
fn (x) dx =
fn (x) dx = 0
fn dx = 0
n+1
0
0
0
for each fixed n. Now define

X
fn (x)
f (x) =
n=0

Then since fn has disjoint support for any N N and 2N < y < 2N + 2, we have
Z y
Z y
f (x) dx =
f (x) dx,
0

2N

and so the Riemann integral of f (x) converges to 0 on [0, ). Now if a measurable function f is
Lebesgue integrablem then so is |f |. But,
Z

X
1
= .
|f | dx = 2
n+1
0
n=1
Therefore f is Riemann integrable but not Lebesgue integrable

.

Exercise 2.4. Consider the real valued function f (x, t), where x Rn and t I = (a, b). Suppose the
following hold.
(1) f (x, ) is integrable over I for all x E
(2) There exists an integrable function g(t) on I such that |f (x, t)| g(t), x E, t I.
(3) For some x0 E Zthen function f (, t) is continuous on I
Then the function F (x) =

f (x, t) dt is continuous at x0
I

Proof: Let xn be any sequence in E such that xn x0 . Define a sequence of functions as fn (t) =

11

f (xn , t). Then by hypothesis we have fn (t) g(t), for t I almost everywhere. Let f (t) = f (x0 , t),
now since f (x, t) is continuous at x0 , we have fn f . So by the Lebesgue Dominated Convergence
theorem we have
Z
lim
|fn (t) f (t)| dt = 0
n

Hence we have
 Z
Z



|F (xn ) F (x0 )| =  fn (t) f (t) dt |fn (t) f (t)| dt 0
I

Or F (x) is continuous at x0

Theorem (Lebesgue Dominated Convergence) Let g be integrable over E and let fn be a sequence
of measurable functions such that |fn | g on E and for almost all x E we have f (x) = lim fn (x).
Then
Z
Z
f = lim
fn .
E

Proof: Assuming the hypothesis, the function g fn is nonnegative, so by Fatous lemma we have
Z
Z
(g f ) lim inf (g fn )
E

Now since |f | g, f is integrable and we have

Z
Z
Z
Z
g
f
g lim sup
fn
E

Hence we have

Z
f lim sup

fn

Considering g + fn , we have the result

Z

Z
f lim inf

and so the result follows

fn
E

.

Exercise 2.5. Show that the Lebesgue Dominated Convergence theorem holds if almost everywhere
convergence is replaced by convergence in measure.
Proof: Suppose that fn f in measure, and there is an integrable function g such that fn g almost
everywhere. Now |fn f | is integrable for each n, and |fn f |[k,k] converges to |fn f |. By the
Lebesgue Dominated Convergence theorem we have
Z k
Z
|fn f |
|fn f |
k

Let  > 0, then there exsits an N0 such that

Z
|fn f | <
|x|>N0


3

also for each n, given  > 0, there exists > 0 such that for any set A with (A) < we have
Z

|fn f | <
3
A
Let A = {|fn f | }. Then there exists an N1 , such that for all n N1 , we have A = {|fn f |
} < . Let N = max{N0 , N1 }
Z
Z
Z
Z


|fn f | =
|fn f | +
|fn f | +
|fn f | < + + 2N < 
3 3
X
|x|>N
[N,N ]A
[N,N ]Ac
Z

Let =
, therefore we have
|fn f | 0, as n  .
6N
X

12

Exercise 2.6. Show that an extended real valued integrable function is finite almost everywhere.
Proof: Consider the measur space (X , M, ). Let E = {x C : |f | = }. Now since f is integrable,
it is measurable hence the set E is measurable. Now suppose (E) > 0, then as |f | > 0 on E we have
Z
Z
>
|f | d
|f | d =
X

This contradicts to the integrability of f , thus (E) = 0. Therefore f is finite almost everywhere

.

Exercise 2.7. If fn is a sequence of measurable functions such that

Z
X
|fn | <
n=1

Show that

fn converges almost everywhere to an integrable function f and that

n=1

Z
f=

Z
X

fn <

n=1

Proof: Define gN to be the partial sums of |fn |. Then gN is measurable since each fn , and hence |fn |
is measurable. Let g = lim gn , then g is measurable as it is the limit of measurable functions. Now
Z
Z X

Z
X
f=
|fn | =
|fn | <
n=1

n=1

So g is integrable, and hence g is finite almost everywhere. Define f (x) as follows

(P
R
if |g(x)| <
n=1 fn
f (x) =
0
otherwise
Then gN f as N almost everywhere. We also have
Z 
Z


 f
|f |



Z  X




=
fn 



n=1
Z X

|fn |
n=1

Z
=

g<

We also have that

N

N

X  X
X


|gN | = 
fn 
|fn |
|fn | = g


n=1

n=1

n=1

almost everywhere. Now by the Lebesgue Dominated Convergence theorem, we have

Z

Z
f=

Z
lim gN = lim

gN = lim

N Z
X
n=1

fn =

Z
X
n=1

fn

13

Exercise 2.8. Let (X , M, ) be a measure space, and let fn be a sequences of nonnegative extended
real-valued M-measurable functions on X . Suppose lim fn = f exists almost everywhere on X and
fn f almost everywhere. For n N, show that
Z

Z
f d = lim

Proof: First if

f dx = , applying Fatous lemma we have

Z

lim inf fn d lim inf

X n

fn d lim

fn d .
X

And so lim

fn d =

fn d
X

f dx = .

R
Now if f dx < , since fn f almost everywhere, we have |fn | |f | almost everywhere, and we
have lim fn = f exists almost everywhere, we have by the Lebesgue Dominated Convergence theorem
Z

Z
| |fn | |f || d

|fn f | d = 0

Z
fn d =

lim

f d
X

Exercise 2.9. Let f be a nonnegative Lebesgue measurable function on [0, 1]. Suppose f is bounded
R1
above by 1 and 0 f dx = 1. Show that f = 1 almost everywhere on [0, 1]
Proof: let 1 >  > 0 and define the set E as
E = {x [0, 1] : 0 f 1 }
Now we have
Z
1 =

Z
f dx

f dx +
ZE

f dx
ZE
1  dx

f dx +
Ec

(E c ) + (E) (E)
=

1 (E)

Hence since this holds for any  (0, 1), we must have (E) = 0. Therefore f = 1 almost everywhere
on [0, 1]  .

Exercise 2.10. Let f be a real-valued Lebesgue measurable function on [0, ) such that:
(1) f is locally integrable
(2) lim f = c
x
Z
1 a
f dx = c.
Show that lim
a a 0

14

Proof: Let  > 0, then there is an M > 0 such that if x > M , then |f (x) c| < . Let a > M , now


 Z a
Z


1
1  a


f c dx
f dx c =
a

a 0
0
Z
1 a

|f c| dx
a 0
Z
Z
1 M
1 a
=
|f c| dx +
|f c| dx
a 0
a M
Z
1 M
1
<
|f c| dx +  (a M )
a 0
a
Z
M
1 M
|f c| dx + (1
=
)
a 0
a
Now since M is fixed, and by the integrability of f , we have
 Z a

1


f dx c < 
a
0

and since this is for any  > 0 and all a > M , we have
 Z a

1


lim 
f dx c = 0
a a
0

1
a a

lim

f dx = c

.

Exercise 2.11. Let f be a real-valued uniformly continuous function on [0, ). Show that if f is
Lebesgue integrable on [0, ), then lim f (x) = 0.
x

Proof: First if f is Lebesgue integrable, then so is |f |. Now decompose the integral as follows
Z
Z k+1
Z k+1
X
>
|f (x)| dx =
|f (x)| dx, denote ak =
|f (x)| dx.
0

k=1

Now ak > 0, and since the integral is convergent this implies that ak 0 as k , which inturn
implies that ak is Cauchy. So we have


Z
m
X



 > 0, N s.t. 
ak  < , n, m > N

|f (x)| dx < 


N +1
k=n

Since |f (x)| is positive and  is arbitrary this implies that f (x) 0 as N

.

Exercise 2.12. Let f L1 (R). With h > 0 fixed, define a function h on R by setting
Z x+h
1
f (t) (dt), for x R
h (x) =
2h xh
(a) Show that h is measurable on R.
(b) Show that h L1 (R) and kh k1 kf k1 .
For part (a) since f is integrable, then f is measurable. So the integral of a measurable function is
measurable, thus h (x) is measurable.
For part (b) First apply the change of variable y = x t, then we have
Z h
Z
Z x+h
Z h
f (t) (dt) =
f (x y) (dy) =
f (x y) (dy) =
xh

f (x y)[h,h] (y) (dy)

Where [h,h] (y) is the charactistic function on [h, h]. So we have

h (x) =

1
f [h,h]
2h

kh (x)k1 =

1
1
kf [h,h] k1
kf k1 k[h,h] k1 = kf k1
2h
2h

.

15

Exercise 2.13. Let f be a Lebesgue integrable function of the real line. Prove that
Z
lim
f (x) sin(nx) dx = 0.
n

Proof: If f is intergrable, then there exists a sequences of step function n such that
Z

 > 0 N s.t. |f n | <
2
Now we have

Z


 f (x) sin(nx) dx


R

Z
|f (x) sin(nx)| dx
ZR

|(f (x) n (x)) sin(nx)| dx +

Z

+ |n (x) sin(nx)| dx
2
R
R

<

|n (x) sin(nx)| dx
R

Now
S n being a step function we have it as the sum of simple functions over disjoint interval In , where
n=1 In = R, i.e.
n =

ak,n Ik,n

k=1

and so we have
Z

Z
|n (x) sin(nx)| dx

|ak,n |

k,n


sin(nx) dx

Z
|ak,n |

|sin(nx)| dx 0 as n
Ik,n

k=1

Hence for some N large enough and all n > N we have


Z
Z




 f (x) sin(nx) dx <  +
|n (x) sin(nx)| < + = 
 2

2
2
R
R

3. Convergence
Exercise 3.1. Consider the Lebesgue measure space (R, M, ) on R. Let f be a -integrable extended
real-valued M-measurable function on R. Show that
Z
|f (x + h) f (x)| (dx) = 0.
lim
h0

Proof: First since f (x) is integrable, we have

Z
Z
f (x + h) (dx) =
f (x) (dx)
R

h R

Also since f is integrable, there exists a sequence of continuous function n , such that
Z

|f (x) n (x)| (dx) <
3
Now |n (x + h) n (x)| < 3 if |h| < . Let N be large enough, then
Z
Z
  
|f f (x+h)| (dx) |f n (x)|+|n (x+h)f (x+h)|+|n (x+h)n (x)|(dx) < + + = 
3 3 3
Therefore we have
Z
|f (x + h) f (x)| (dx) = 0

lim

h0

.

16

Exercise 3.2. Let (X , M, ) be a measure space. Let fn and f be an extended real-valued M- measurable fuctions on a set E X such that lim fn = f on E. Then for every R we have
n

{E : f > } lim inf {E : fn } and {E : f < } lim inf {E : fn }

n

Proof: I will only show the first inequality since the proofs are identical. Let A = {x E : f (x) },
A,n = {x E : fn (x) } and let A denote the characteristic function of A. First we need to show
An A in measure. Let  > 0, denote the set F,n by
F,n = {x E : |A,n A | }.
Now we want to show that the measure of this set is small. First notice that
c
F,n
{x A : |f fn | < }

Let x be in this subset, then this implies two thing. First if f (x) > > , and fn (x) > f (x) > .
So we must have
F,n {x A : |f fn | }
Now since fn converges to f almost everywhere in E, it converges in measure, and hence the measure
of the set (F,n ) < . This implies that A,n converges to A in measure. Now Fatous lemma
holds for a sequence of functions converging in measure, so we have
Z
Z
A d lim inf
A,n d {E : f > } lim inf {E : fn } 
E

Exercise 3.3. Let g(x) be a real-valued function of bounded variation on an interval [a, b]. Suppose
that f is a real-valued decreasing function on [a, b]. Show that g(f (x)) is also of bounded variation. If
f is just a bounded continuous function is g(f (x)) still of bounded variation.
Proof: Since g is of bounded variation we have, let P be all the possible partitions of [a, b]
Vab (g) = sup

n
X

|g(xi ) g(xi1 )|

{xi }P i=1

Now fix  > 0 and pick an {xi } such that

Vab (g) <

N
X

|g(xi ) g(xi1 )| + 

i=1

Now since f is decreasing on we have that f (xi+1 ) < f (xi ). Now call yi = fxi , {yi } {a, b} then is a
partition of [a, b], and so we have
N
X
i=1

|g(yi ) g(yi1 )| < sup

n
X

{xi }P i=1

|g(xi ) g(xi1 )| = Vab (g)

This can be done for any partition of [a, b].Therefore g(f (x)) is also of bounded variation.
For the second part, no. Consider the function
(
x sin
f (x) =
1

1
x

x 6= 0
x=0

and let g(x) = x. Now g(x) is a function of bounded variation on [1, 1] and f (x) is a bounded and
continuous on [1, 1], but g(f (x)) = f (x), which not a function of bounded variation on [1, 1].

Exercise 3.4. Let (X , M, ) be a measure space. Let fn and f be an extended real-valued M- measurable fuctions
Z on a set E X with (E) < . Show that fn converges to 0 in measure on E if and
|fn |
only if lim
d = 0
n E 1 + |fn |

17

Proof: () If fn converges to 0 in measure then we have

{x E : |fn | } < .
Call this set A . Now
Z
E

|fn |
d =
1 + |fn |

Z
A

|fn |
d +
1 + |fn |

Z
Ac

|fn |
d
1 + |fn |

1
1+x , x

0 is monotone, and uniformly continuous on any bounded interval. Now

Now the function
(A ) < , and so there is a such that So we have
Z
Z
|fn |
|fn |
d +
d < (A ) + (E) < (1 + (E))
A 1 + |fn |
Ac 1 + |fn |
Hence we have

Z
E

|fn |
d 0 as  0
1 + |fn |

() Now suppose that

|fn |
d 0 as  0
E 1 + |fn |
and suppose that there exists and 0 such that {x E : |fn | 0 } 0 . Then we have
Z
Z
Z
|fn |
20
|fn |
|fn |
d +
d
+
d
2
1 + 0
Ac 1 + |fn |
Ac 1 + |fn |
A0 1 + |fn |
Z

20
1 + 20

But this implies that

20

1 + 20
let  =

20
2(1 + 20 )

Z
A0

|fn |
d +
1 + |fn |

, then we have a contradiction

Z
Ac

|fn |
<
1 + |fn |

.

Exercise 3.5. Suppose (E) < and fn converges to f in measure on E and gn converges to measure
on E. Prove that fn gn converges to f g in measure on E.
Proof: Let hn = fn gn and let h = f g. Now h and hn are measurable since fn and gn are. For each
> 0 define
An () = {x : |hn (x) h(x)| }
and let an () = (An ()). Now because fn and gn converge in measure, for any subsequences fnk , gnk
there are subsequences fnkj and gnkj , such that both fnkj and gnkj converge almost everywhere to f
and g respectively. Hence we have hnkj = fnkj hnkj , which converges to h = f g almost everywhere
on E. Now since hnkj converges almost everywhere and (E) is finite we have that hnk converges in
measure. Now
lim |h hn | lim sup |h hnk | 0
n

k n

Hence lim an () = lim ank () = 0, or hn converges in measure

n

.

(Convergence in measure) A sequences fn of measurable functions is said to converge to f in

measure if, given  > 0, there is an N such that for all n N we have
{x : |f (x) fn (x)| } 
Remark: Let an be a sequence of real numbers. If there is an a R, such that for every subsequence
ank , there is a subsequences for which ankl a, then an a.

18

Exercise 3.6. If fn , f L2 and fn f almost everywhere, then kfn f k2 0 if and only if

kfn k2 kf k2 .
Proof: () Suppose kfn f k2 0, now
kfn f k22

Holders inequality

fn2 2f fn + f 2
Z
2
kf k2 2 |fn f | + kf k22

kf k22 2kfn k2 kf k2 + kf k22

|kf k2 kfn k2 |2

Therefore as kfn f k22 0 we have kfn k2 kf k2 .

() Now suppose kfn k2 kf k2 and fn f almost everywhere. Now for p 1, and for finite a,b, we
have
|a + b|p 2p (|a|p + |b|p )
For each n, let
gn = 4(|fn |2 + |f |2 ) |fn f |2 .
Now gn 0 almost everywhere. Since fn and f are finite almost everywhere, by Fatous lemma we
have
Z
Z
lim inf gn lim inf gn
2
Now since fn
Z f almost everywhere we have lim inf gn = 8|f | almost everywhere. So we have
8kf k22 lim inf gn . Now
Z
Z
Z
Z
lim inf gn = 4 lim inf |fn |2 + 4 lim inf |f |2 = lim sup |fn f |2

4 lim inf kfn k22 + 4kf k22 lim sup kfn f k22

8kfn k22 lim sup kfn f k22

so we have 0 lim sup kfn f k22 , hence 0 lim sup kfn f k22 0. Therefore we have
lim sup kfn f k2 = lim inf kfn f k2 = 0

kfn f k2 0

.

Remark: A sequences of functions fn converges in measure to f if and only if for every sequences fnk ,
there is a subsequence fnkj that converges almost everywhere to f .

Exercise 3.7. If fn 0 and fn (x) f (x), in measure then

Z
Z
f (x) dx lim inf fn (x) dx
Proof: Let fnk be any subsequence of fn . then there exists an fnkj such that fnkj converges to f
almost everywhere. By Fatous Lemma we have
Z
Z
Z
Z
f lim inf fnkj = lim fnk lim inf fn
Exercise 3.8. Suppose fn converges to two functions f and g in measure on D. Show that f = g
almost everywhere on D
Proof: Define the set E as E = {x D : |fn (x) f (x)| > 0}. Then if En = {x D : |fn (x) f (x)|
1
1
and |fn (x) g(x)| < 2m
,
1/m}, we have E = lim En . Now if for some n we have |fn (x) f (x)| < 2m
then we have
1
|f g| |fn f | + |fn g| <
m

19

And so

x : |fn (x) f (x)| <

1
2m



1
x : |fn (x) g(x)| <
2m

x : |f (x) g(x)| <

1
2m

which implies that

 




1
1
1
x : |fn (x) g(x)|

x : |f (x) g(x)|
x : |fn (x) f (x)|
2m
2m
2m


1
2
2
This implies that x : |f (x) g(x)|
0. Hence
<
. Now as n , we have m
2m
m
{x : |f (x) g(x)| > 0} = 0  .

Exercise 3.9. Let fn f in Lp (X , M, ), with 1 p < , and let gn be a seqences of measurable

functions such that |gn | M < for all n, and gn g almost everywhere. Prove that gn fn gf in
Lp (X , M, )
Proof: Since fn f in Lp , since Lp is complete we have f Lp . Also since |gn | M , for all n this
implies that |g| M . Now
Z
Z
kfn gn gn f kpp = (fn gn gn f )p M p |fn f |p
M p kfn f kpp
So we have kfn gn gn f kp M kfn f kp , and so
kfn gn gf kp M kfn f kp 0 as n
Therefore gn fn gf in Lp (X , M, )

.

Exercise 3.10. Suppose f is differentiable everywhere on (a, b). Prove that f 0 is a Borel measurable
function on (a, b)
Proof: f 0 is Borel measurable if {x : f 0 (x) } is a Borel set. So
 


1
f 0 (x) lim n f x +
f (x)
n
n


 

1
f (x) 0
lim f x +
n
n
n
 


1

1
for all but finitely many n
f x+
f (x)
m
n
n
m




1

1
x lim inf x : f x +
f (x)
m
n
n
m



[ \
1

1
x
x:f x+
f (x)
m
k
k
m
n1 kn



\ [ \
1

1
x
x:f x+
f (x)
k
k
m
m1 n1 kn

Now since f (x) is differentiable almost everywhere, it is continuous almost everywhere and so the f (x),
and f (x + 1\k) are measurable. Any linear combination of them is measurable, and so the set




1

1
x:f x+
f (x)
k
k
m
is measurable. Now the collection of all such sets form a -algebra, and hence the countable union and
intersection of these sets are measurable. Therefore f 0 (x) is measurable  .

20

Exercise 3.11. Let cn,i be an array of nonnegative exteneded real numbers for n, i N.
(a) Show that
lim inf

cn,i

iN

X
iN

lim inf cn,i

(b) If cn,i is an increasing sequences for each i N then

X
X
lim
cn,i =
lim cn,i
n

iN

iN

Proof: For part (a) first let denote the counting measure. Now if M = P(N), then (N, M, )
forms a measure space. Now let an be sequences with cn [0, ]. Then the function a(n) = an is
M-measurable, and so
Z
X
c d =
cn
nN

Then by Fatous lemma we have

Z
Z
lim inf cn lim inf
cn
N n

nN

X
iN

lim inf cn,i lim inf

cn,i

iN

For part (b) using the same measure space (N, M, ), we know that cn (i) cn (i+1), so by the Monotone
convergence theorem we have
Z
Z
X
X
lim cn = lim
cn

lim cn,i lim

cn,i 
N n

iN

iN

Theorem (Ascoli-Arzela) Let F be an equicontinuous family of functions from a separable space

X to a metric space Y . Let fn be a sequence in F such that for each x X the closure of the set
{fn (x) : 0 n < } is compact. Then there is a subsequence fnk that converges pointwise to a
continuous function f , and the convergence is uniform on each compact subset X.
Exercise 3.12. Let {qk } be all the rational numbers in [0, 1]. Show that

X
1
1
p
converges a.e. in [0, 1]
k 2 |x qk |

k=1

Proof: Fix 0 > 0, consider the two sets

1
1
E1 = p

0
|x qk |

1
1
and E2 = p
>
0
|x qk |

Now for each fixed x [0, 1]\Q we can enumerate the rationals however we want (Zorns Lemma).
Choose such an ordering so that
1
x E2
p
< k 10
|x qk |
That is the closer qk gets to x, the large the index. Now let be the counting measure, then we have
Z
Z

X
1
1
1
1
1
1
p
p
p
=
d
+
d
2
2
2
k
|x qk |
|x qk |
|x qk |
E1 k
E2 k
k=1
Z
Z
1 1
1
<
+
d
2 
1+0
k
k
0
E
E
Z 1
Z 2
1 1
1
<
+
d <
2 
1+0
k
k
0
N
N
This can be done for all x [0, 1]\Q. Therefore the series converges almost everywhere in [0, 1]

.

21

Exercise 3.13. Let (X , M, ) be a finite measure space. Let fn be an arbitrary sequence of real-valued
measurable functions on X . Show that for every  > 0 there exists E M with (E) <  and a sequence
of positive real numbers an such that an fn 0 for x X \E
Proof: First denote the set Em = {x : m 1 |fn | < m}, then the sets Em are disjoint and cover X .
Now define as such

X
1
=
k2
k=1

Since (X ) < , if  > 0, there is an Mn such that

X

(En ) = {x : |fn | Mn }
>
n2
mMn

Now choose these Mn such that Mn > Mn1 for all n. Define the sets Fn = {x : |fn | Mn }, then we
have (Fn ) < n 2 . Now if E = En , then
(E)

(En ) <

n=1

 X 1
=
n=1 n2

Let an = 1/Mn3 , then if x X \E, then we have

an |fn (x)| <
And so we have

1
1
Mn = 2
Mn3
Mn

 X
X
X
X
1
1


|an fn (x)|
an fn (x)

<

2


Mn
n2
n=1
n=1
n=1
n=1

Therefore we must have an fn (x) 0 on X \E

.

Exercise 3.14. Prove that the gamma function

Z
(x) =

tx1 et

is well defined and continuous for x > 0

Proof: Let let f (t, x) = tx+1 et , and x > 0 and decompose the integral into two integrals (0, 1] and
(1, ). For the first we have
Z 1
Z 1
tx x
x1 t
t
e dt
tx1 dt = 0 <
x
0
0
Now f (t, x) is continuous on (1, ), and also t2 f (t, x) 0 as t , so there is an M such that M
bounds t2 f (t, x) on (1, ). Now
Z
Z 1
Z
1
tx1 et dt =
tx+1 et t2 dt = M
dt = M
2
t
1
0
0
And so (x) is well defined on (0, ).
To show continuity, let xn , be a cauchy sequence, and define fn (t) = f (t, xn ). Now by continuity of
f (t, x) on (0, ) (0, ), we have that for each x, fn f on t (0, ). now f (t, x) is bounded on
(1, ), call this bound M > 1. Define a function g(t) by
(
tx1
0<t1
g(t) = M t
t e
1<t
Now fn , f g on (0, ), so by the Lebesgue Dominated Convergence theorem we have
Z
|fn f | 0 as n
0

22

and so we have
Z

|(xn ) (x)| = 

 Z

fn (t) f (t, x) dt

fn (t) f (t, x)| dt 0 as n

This holds for any sequence such that xn x (0, ), therefore (x) is continuous on (0, )

.

4. Lp spaces
Exercise 4.1. Let 1 p < q < . Which of the following statements are true and which are false?
(a)
(b)
(c)
(d)

Lp (R) Lq (R)
Lq (R) Lp (R)
Lp ([2, 5]) Lq ([2, 5])
Lq ([2, 5]) Lp ([2, 5])

Proof: Only part (d) is true. This can easily be shown for any finite interval, let I = [a, b] Let
f Lq (I). Then |f |p Lq/p (I). Now by Holders inequality we have
Z
|f |p = k|f |p kq/p k1kr
I

where r is conjugate to pq . Now

kf kpp k|f |p kq/p k1kr =
Hence we have kf kp kf kq (I)

qp
qp

Z

(|f |p )q/p

p/q
(I)

qp
q

= kf kpq (I)

qp
q

, therefore f Lp (I)

.

For a counterexample to part (c) consider the function f (x) = (x 2)1/2 , and let p = 1 and q = 2,
then f Lp ([2, 5]), but f
/ Lq ([2, 5]).
For a counterexample to part (b) consider the function f (x) = (1 + x2 )1/2 , and let p = 1, q = 2, then
f Lq (R) but f
/ Lp (R).
For a counterexample to part (a) consider the counterexample to part (c) with the zero extension.
Theorem (Holder Inequality) If p and q are nonnegative extended real numbers such that
1 1
+ =1
p q
and if f Lp and g Lq then f g L1 and
Z
|f g| kf kp kgkq
Proof: Assume 1 < p < , and suppose that f , g 0. Let h = g q1 , then g = hp1 . Now
ptf (x)g(x) = ptf (x)hp1 (h(x) + tf (x))p h(x)p
so we have

Z
pt

Z
fg

and we also have

Z
pt

|h + tf |p

hp = kh + tf kpp khkpp

f g khkpp + ktf kpp khkpp

now differentiating bothsides with respect to t at t = 0 , we have

Z
p f g pkf kp khkp1
= pkf kp kgkq
p

.

23

Exercise 4.2. Let f L3/2 ([0, 5]). Prove that

t

lim

t0+ t1/3

f (s) ds = 0.
0

Proof: Applying Holders inequality we have



Z t
 1



f
(s)
ds
 t1/3


|f (s)| ds

t1/3

Z

2/3 Z
|f (s)| ds

1/3
ds

t1/3
0
0
1
1/3 kf (s)k3/2 t1/3
t
2/3
Z t
|f (s)|3/2 ds
0 as

t0+

Exercise 4.3. Suppose f C 1 [0, 1], f (0) = f (1), and f > f 0 everywhere.
(1) Prove that f > 0 everywhere.
(2) Prove that
Z 1
Z 1
f2
d

f d
0
0 f f
0
Proof: For (1), if f (x) = R+ then everything holds. So suppose that, there exists an c (a, b)
(0, 1) such that f 0 (c) = 0. (WLOG) suppose that this c is not a saddle point for f (x), also suppose that
f (c) < 0. Now if there is a > 0 such that f (c) > f (x), for all x B(c, ), then we have f 0 (x) > 0 for
x (c , c). This implies that f 0 (x) > f (x) for x (c , c). If there is a > 0 such that f (c) < f (x),
for all x B(c, ), then we have f 0 (x) > 0 for x (c, c + ), which implies that f 0 (x) > f (x) for
x (c, c + ). For both cases we have a contradiction. Therefore f (x) > 0 for all x (0, 1). Now if
f (0) = 0, f cannot be constant since 0 6 0. this implies that, for some > 0, f 0 (x) > 0 for x [0, ),
which is a contradiction. Therefore f (x) > 0 for all x [0, 1].

For (2) since f > f 0 we have that f f 0 is well defined on [0, 1]. So,
Z

2

2
p
f
0 d
f

f
f f0
0
Z 1
Z 1
f2

d
f f 0 d
0
f

f
0
0
Z 1
Z 1
f2
d
f d

0
0
0 f f
Z

d =

H
olders inequality

The last line holds since f f 0 > 0. This implies that:

Z

Z
f d

f2
d
f f0

Exercise 4.4. If f (x) Lp L for some p < . Show that

(a) f (x) Lq for q > p.
(b) limq kf kq = kf k .
Proof: For part (a) Let 0 < p < q < and let f Lp L . Then if =

q
p

and if =

q
qp ,

then we

24

have

= 1. Now applying Holders inequality we have

Z
kf kqq =
|f |q
Z
1
1
=
|f |q( + )
Z
=
|f |p |f |qp
Z
=
|f |p |f |qp

k|f |p k1 k|f |qp k

Now since |f | kf k almost everywhere and q p > 0 we have |f |qp k|f |qp k almost everywhere,
and so k|f |qp k < . Also since f is monotone increasing, we have k|f |qp k = kf kqp
. We also
have k|f |p k1 = kf kpp < . Therefore f Lq  .
For part (b), first suppose that kf k = 0. This implies that f = 0 almost everywhere and hence
kf kq = 0 for all q. Hence limq kf kq kf k trivially.
Now suppose that f Lp L and kf k =
6 0. From part (a) we have

1/q
1 p
kf kq kf kpp
(kf k ) q
Now let  > 0, then on a set E of nonzero measure, |f | > kf k . If (E) = , shoose a subset of E
with finite measure. Then we have
Z
q
kf kq =
|f |q d
ZE

(kf k )q d
E

= (E)|kf k |q .
Now this is for all q > p. Let qn be a sequence of numbers greater than p that converges to . Then
1

lim (E) qn |kf k |

lim inf kf kqn

lim sup kf kqn

lim sup kf kpp

q1

1 qpn

(kf k )

and so
|kf k | lim inf kf kqn lim sup kf kqn kf k
n

Since this holds for all  > 0 we have limn kf kqn = kf k . Now since this is for any sequence qn , we
have limq kf kq = kf k .

Exercise 4.5. Suppose that f Lp ([0, 1]) for some p > 2. Prove that g(x) = f (x2 ) L1 ([0, 1])
Proof: f Lp ([0, 1]) implies that kf kp < . In particular this implies that kgkp = kf (x2 )kp < .
Now
Z 1
Z 1
|g(x)| dx =
|f (x2 )| dx
0

change of variables (y = x2 )

h
olders inequality

1
|f (y) | dy
2 y
0



1
1

kf kp

y p
2
p1

25

Now f Lp ([0, 1]) and since p > 2 we have 1y

p
p1

< , therefore g(x) L1 ([0, 1])

.

Exercise 4.6. Let f Lp (X ) Lq (X ) with 1 p < q < . Prove that f Lr (X ) for all p r q.
Proof: Let E1 = {x : 0 |f (x)| 1}, and E2 = {x : 1 > |f (x)|}, then E1 , E2 are a Hahn
decomposition for X . Now suppose f Lp Lq . Now
Z
Z
r
r
kf kr =
|f | +
|f |r
E1
E2
Z
Z
p

|f | +
|f |q
E1
E2
Z
Z

|f |p +
|f |q
X

= kf kpp + kf kqq

f Lr (X )

Exercise 4.7. Suppose f and g are real-valued -measurable functions on R, such that
(1) f is -integrable.
(2) g C0 (R).
For c > 0 define gc (t) = g(ct). Prove that:
Z
(a) lim
f gc d = 0,
c R
Z
Z
(b) lim
f gc d = g(0) f d.
c0

Proof: For part (a) define hn (x) = f (x)gn (x). Now since f L1 (R) we know that f (x) < a.e., and
since g C0 (R) we know that
gn (x) 0 as n .
For a fixed x such that f (x) < we have
hn (x) 0 as n
Hence hn 0a.e.. Also since g C0 (R) we have that there is some M such that |g(x)| < M . So we
have
 Z
Z
Z


 hn (x) d
|f (x)gn (x)| d M
|f (x)| d <


R

since f L1 (R). Hence by the Lebesgue Dominated Convergence theorem we have

Z
Z
Z
lim
f gn d = lim
hn d =
lim hn d = 0
n

R n

Proof: For part (b) we know that for all n > 0, f gn L1 (R). Define hn (x) = |f (x)g(xn1 )|, again
since g C0 (R) we have that there is some M such that |g(x)| < M . So
Z
 Z
Z


 hn (x) d
|f (x)gn (x)| d M
|f (x)| d <


R

Hence by the Lebesgue Dominated Convergence theorem we have

Z
Z
lim
f g1/n d = lim
hn d
n R
n R
Z
=
lim hn d
n
ZR
=
lim f g1/n d
R n
Z
= g(0) f d
R

26

Exercise 4.8. Let E be a measurable subset of the real line. Prove that L (E) is complete.
Proof: Let fn be a Cauchy sequence of measurable functions in L . Then there exists and k N such
that if m, n Nk , then
1
1
kfn fm k < , n, m > Nk
|fn fm | < a.e.
k
k
Now define the sets En,m,k by


1
En,m,k = x E : |fn (x) fm (x)|
k
then for each n, m > Nk , the set En,m,k is empty. Let F be defined by
[
F =
En,m,k
km,n,Nk

Now F is a countable union of empty sets, and therefore is empty. Now for any x E\F we have
1
|fn (x) fm (x)| <
k
and so fn (x) is a Cauchy sequence in R. Now
1
|fm (x)| |fm (x) fn (x)| + |fn (x)| < + |fn (x)|.
k
Taking m , we have
1
1
|f (x)| + |fn (x)| < + kfn (x)k a.e.
k
k
Hence for each n we have |f | k1 + kfn k almost everywhere so f L . Therefore L is complete
.
Theorem (Riesz-Fischer) The Lp (E) spaces are complete.
Proof: For 1 p < , let fn be a Cauchy sequence on Lp .
 > 0 N s.t. kfm fn kp <  n, m > N
Now let nk = N 2k , then the subsequence fnk , satisfies
kfnk+1 fnk kp <

1
2k

Define the function f by

f (x) = fn1 +

(fnk+1 fnk )

for x E

k=1

Now the partial sums SN (f ) is just

SN (f ) = fn1 +

(fnk+1 fnk ) = fnN

k=N

Define the function g(x) by,

f (x) = |fn1 | +

|fnk+1 fnk |

for x E

k=1

Now by Minikowskis inequality we have

kSN (g)kp kfn1 kp + k

N
1
X

|fnk+1 fnk |kp kfn1 kp +

k=1

N
1
X
k=1

1
2k

So the increasing sequences of partial sums kSn (g)kp is bounded above by kfn1 k + 1. Hence we have
Z
Z
Z
p
p
g <
|f | <
fp <
E

27

This implies that the series fnk converges almost everywhere. Now
N

X



|f fnN | = |S (f ) SN 1 (f )| = 
fnk+1 fnk  g


k=1

Hence by the Lebesgue dominated convergence thoerem we have

Z
lim kf fnk kpp =
lim (f (x) fnk )p = 0
k

E k

Hence fnk converges to f in Lp (E). Now fn is itself Cauchy, hence fn converges to f is in Lp (E).

f (x)g(x) dx is finite for any f (x) L2 . Prove

Exercise 4.9. Let g(x) be measurable and suppose

a

that g(x) L2 .
Z

Z
gdx < which implies that g L1 [a, b]. Let F =

Proof: If f = 1, then f L2 ([a, b]) so

a

then F is a bounded linear functional from L2 ([a, b]) to R. So there exists an M such that
(Z
)

gdx,
a

kF (f )k = sup
kf k2 =1

fg

< M,

f L2 ([0, 1])

Then by the Reisz Representation Theorem g must be in L2 ([0, 1])

.

Theorem (Riesz Representation) Let F be a bounded linear functional on Lp for 1 p < .

Then there exists a function g Lq such that
Z
F (f ) = f g.
We also have kF k = kgkq .
Proof: Just considering the finite dimensional case. Let be of finite measure. Then every bounded
measurable function is in Lp (). Define a set function on the measurable sets by (E) = F (E ). If
E is the union of a sequence En of disjoint measurable sets, define a sequence n = sgn F En and set
X
f=
n En
Then F is bounded and we have

X
|(En )| = F (f ) < ,
n=1

(En ) = F (f ) = (E)

n=1

Hence is a signed measure, and by construction it is absolutely continuous with respect to . By the
Radon-Nikodym Theorem, there is a measurable function g such that for each measurable set E we
have
Z
(E) =

g d
E

Since is always finite implies that g integrable. Now if is a simple function, the linearity of F and
of the integral imply that
Z
F () = g d
Since the left-hand side is bounded by kF kkkp we have g Lq . Now let G be the bounded linear
functional defined on Lp by
Z
G(f ) = f g d

28

Then G F is a bounded linear function which vanishes on the subspace of simple functions, which are
dense in Lp . Hence we must have G F = 0 in Lp . So for all f Lp , we have
Z
F (f ) = f g d
and by construction kF k = kGk = kgkq

.

Exercise 4.10. Let (X , M, ) be a measure space and let f be an extended real-valued M- measurable
function on X such that
Z
|f |p d < for p (0, ).
X
p

Show that lim {x : |f (x)| } = 0

Proof: First define the set E = {x X : f (x) }. Now notice that E E if > , also because
f Lp we have (E ) < 1 if is large enough, in particular (E ) = 0. Now
Z
Z
p {x : |f (x)| } = p
d
|f |p d
E

Hence we have
lim p {x : |f (x) }

|f |p d = 0

Since f Lp , then |f | L1 (X ), (E ) = 0 and the integral of an Lebesgue integrable function over

a set of measure zero is zero  .
5. Signed Measures
Remark: If (X , M) is a measure space, and if , are two measure defined on (X , M). and are
said to be mutually singular (), if there are disjoint stes A and B, in M such that X = A B
and (A) = (B) = 0. A measure is said to be absolutely continuous with respect to the measure ,
( << ), if (A) = 0 for each set A for which (A) = 0.

Exercise 5.1. Let be a measure and let , 1 , 2 be signed measure on the measurable space (X , A).
Prove:
(a) If and << , then = 0
(b) If 1 and 2 , then, if we set = c1 1 +c2 2 with c1 , c2 R such that is a signed measure,
thwn we have .
(c) If 1 << and 2 << , then, if we set = c1 1 + c2 2 with c1 , c2 R such that is a signed
measure, thwn we have << .
Proof: For part (a), if is a signed measure such that and << . There are disjoint measurable
sets A and B such that X = A B and ||(B) = ||(A) = 0. Then |(A)| = 0 so ||(X) = ||(A) +
||(B) = 0. Hence we have + = = 0 i.e. = 0.
For part (b), there are disjoint measurable sets Ai and Bi such that X = Ai Bi and (Bi ) = i (Ai ) = 0,
for i = 1, 2. Now X = (A1 A2 ) (B1 B2 ) and (A1 A2 ) (B1 B2 ) = . Now we have
(c1 1 + c2 2 )(A1 A2 ) = (B1 B2 ) = 0

(c1 1 + c2 2 )

For part (c), suppose 1 << and 2 << . If (E) = 0, then 1 (E) = 2 (E) = 0. Hence
(c1 1 + c2 2 )(E) = 0

(c1 1 + c2 2 ) <<

29

Exercise 5.2. Let be a positive measure and be a finite positive measure on a measurable space
(X , M). Show that if << , then for every  > 0 there is a > 0, such that for every E M with
(E) < , we have (E) < .
Proof: Suppose not, Then there is an  > 0 such that for every > 0, there is E M, such that
(E ) < , and (E ) . In particular, for every n 1, there is an En such that (En ) < n12 and
(En ) . Now we have

X
X
1
(En ) =
<
2
n
n=1
n=1
Let E = lim sup En , then (E) = 0. Now since << , we have (E) = 0. Now
(E) = (lim sup En ) lim sup (En ) 
But this implies that (En )  > 0, and hence (E) > 0, which is a contradiction. Therefore given
 > 0 there is a > 0, such that for every E M with (E) < , we have (E) <   .
Theorem (Hahn Decomposition) Let be a signed measure on the measurable space (X, M).
Then there is a positive set A and a negative set B such that X = A B and A B = .
Theorem (Jordan Decomposition) Let be a signed measure on the measurable space (X, M).
Then there are two mutually singular measure + and on (X, M) such that = + . Moreover,
there is only one such pair of mutually singular measures.

Exercise 5.3. Suppose (X , M) is a measurable space, and Y is the set of all signed measure on M
for which (E) < , whenevery E M. For 1 , 2 Y , define
d(1 , 2 ) = sup |1 (E) 2 (E)|
EM

Show that d is a metric on Y and that Y equipped with d is a complete metric space.
Proof: Since i are choosen such that i (E) < , then for any 1 , 2 Y and E M, we have
|1 (E) 2 (E)| < . So we have d : Y Y [0, ). Now to show d is a metric on Y we need to show
symmetry, positive definiteness and the triangle inequality. Clearly d(1 , 2 ) = d(2 , 1 ) by definition
of d. For the triangle inequality we have
d(, )

sup |(E) (E)|

EM

sup {|(E) (E)| + |(E) (E)|}



sup {|(E) (E)|} + sup |(F ) (F )|

EM

EM

F M

= d(, ) + d(, )
Now to show definiteness, if = , then |(E) (E)| = 0 for any E M, and so d(, ) = 0.
On the other hand if d(, ) = 0, then we have |(E) (E)| = 0. Let (A1 , B1 ), (A2 , B2 ) be Hahn
decompostions of , and respectively.
Case 1: If E A1 A2 , then (E) = + (E), and (Y ) = + (Y ), hence |(E)(E)| = |+ (E) + (E)|.
So we have + = + on A1 A2 .
Case 2, 3: If E A1 B2 , then we have (E) = (E) and (E) = + (E), hence
0 = |(E) (E)| = | (E) + (E)| = (E) + + (E)
Hence = + = 0 on E A1 B2 . If E A2 B1 , by the same proof we have the result + = = 0
on E A2 B1

30

Case 3: If E B1 B2 , then (E) = (E) and (E) = (E). So

0 = |(E) (E)| = | (E) + (E)|
and so = = 0 on E B1 B2 . So definiteness holds, therefore d is a metric on Y .
Now to show the metric space is complete. Let n be a Cauchy sequence. Then for any  > 0, there is
an N such that if m, n > N , we have
sup |vn (E) vm (E)| < 
EM

If particular, for a fixed set E, we have n is a Cauchy sequence in R. Hence there exists some (E) R,
such that n . By the uniform boundedness pricipal we know that is bounded, and hence
n in the metric d

Remark: The measure || is defined from the Jordan decomposition by, ||(E) = + E + E.
Theorem (Radon-Nikodym) let (X , M, ) be a -finite measure space, and let be a measure
defined on M which is absolutely continuous with respect to . Then there is a nonnegative measurable
function f such that for each set E on M we have
Z
(E) =
f d
E

The function f is unique in the sense that if g is any measurable function with this property then g = f
almost everywhere.
Proof: Only the finite case is considered. Let be finite then is a signed measure for each
rational number . Let (A , B ) be a Hahn decomposition for , and take A0 = X and B0 = .
Now B B = B A . So we have
( )(B B ) 0

( )(B B ) 0

hence we must have (B B ) = 0. Now there exists a measurable function f such that for each
rational we have f almost everywhere on A and f almost everywhere on B . Since B0 =
be an arbitrary set in M, and set
Ek = E (B(k+1)/N Bk/N )
Then E =

Ek , and this union is disjoint modulo null sets. Hence we have

k=1

(E) = (E ) +

(Ek ).

k=0

k+1
k
f
on Ek , and so
N
N
Z
k
k+1
f d
(Ek ).
(Ek )
N
N
Ek

Since Ek B(k+1)/N Ak/N , we have

Now since

k
k+1
(Ek ) (Ek )
(Ek ), we have
N
N
Z
1
1
(Ek ) (Ek )
f d (Ek ) + (Ek ).
N
N
Ek

) On E we have f = almost everywhere. If (E ) > 0, we must have (E ) > 0, since ( )(E

is positive for each . If (E ) = 0, we have (E ) = 0. Since << , for either case we have
Z
(E ) =
f d.
E

Hence we have
(E)

1
(E)
N

Z
f d (E) +
E

1
(E).
N

31

Z
Since (E) is finite and N arbitrary, we must have (E) =

f d.
E

The function f =

d
d

above is called the Radon-Nikodym derivative of with respect to .

Exercise 5.4. Suppose and are -finite measures on a measurable space (X , A), such that << ,
and << . Prove that


d

xX :
=1
= 0.
d
Proof: First notice that if E X , such that ( )E = 0, then we have (E) = (E). But we have
<< , hence if ( )E = 0, then (E) = 0 = (E). Conversely if (E) = (E) and << ,
then (E)
n (E) = 0, and
o so = 0 thus (E) = 0. So if (E) = (E), then (E) = (E) = 0. Now
let E = x X :

but

d
d

d
d

= 1 and consider (E). By the Radon-Nikodym theorem we have

Z
Z
d
(E) =
d =
d
d
E
E

= 1 on E, and so
Z

Hence (E) =

n

xX :

d
d

(E) =
o
=1
=0

d
d =
d

Z
d = (E)
E

.
Exercise 5.5. Let and be two measure on the same measurable space, such that is -finite and
is absolutely continuous with respect to .
(a) If f is a nonngeative measurable function, show that
Z
Z  
d
f d = f
d
d
(b) If f is a measurable function, prove that f is integrable with respect to , if and only if f

d
d

is

integralble with respect to , and in this case, part (a) still holds.
Proof: For part (a), let E be a measurable set and let f = E . Suppose that h =
Z
Z
Z
Z
Z
f d = E d = (E) =
h d = hE d = f h d.

d
d

exists. Then

So the equality holds for charactersitc functions. Let f = be a simple function, then by the above we
have
Z
Z
d = h d.
Now let f be a nonnegative measurable function. There there exists a monotone sequence of simple
functions n such that 0 n f and n f almost everywhere. Applying the Monotone Covergence
theorem, we have
Z
Z
Z
Z
f d = lim

n d = lim
n h d = f h d  .
n
R +
R
For part (b), f is -integrable if and only if f d f d is finite. Now by part (a) we have
Z
Z
Z
Z
f + d = f + h d and
f d = f h d.
n

So we have f is -integrable if and only if f is -integrable

.

Theorem (Lebesgue Decomposition) Let (X , M, ) be a -finite measure space and a -finite

32

measure defined on M. Then we can find a measure 0 , singular with respect to and a measure 1
absolutely continuous with respect to , such that = 0 + 1 . Furthermore, the measures 0 and 1
are unique.
6. Topological and Product Measure Spaces
Exercise 6.1. Let L be a normed space. Then every weakly bounded set X is bounded.
Proof: Let : L L , by (x)(f ) = f (x), where x L, f L . Now X is a Banach space and
(X) is a family of bounded linear functionals on X , and for each f L we have
sup{(x)(f ) : x X} = sup{f (x) : x X} <
Then from the uniform boundness principle we have
sup{kxk : x X} = sup{k(x)k : x X} <
Therefore every weakly bounded nonempty set of a normed space is bounded

.

Exercise 6.2. Suppose that A is a subset in R2 . Define for each x R2 , p(x) = inf{|y x| : y A}.
Show that Br = {x : p(x) r} is a closed set for each nonnegative r. Is the measure of B0 equal to the
outer measure of A?
Proof: Let z (Br ), and let  > 0. Then there is x Br such that |x z| < . So we have
p(z)

inf{|z y| : y A}

inf{|z x| + |x y| : y A}
 + inf{|x y| : y A}
 + r.
This is for all  > 0, therefore p(z) r which implies z Br thus Br is closed. Now B0 = A A.
First by definition of p(x) we have for any x A, p(x) = 0. Hence x B0 , Now suppose that x A,
then for any  > 0, there is a y A such that |x y| < . Therefore we have
p(z) = inf{|z y| : y A} = 0

x B0 ,

and so A B0 . Now suppose x B0 . Then inf{|x y| : y A} = 0, so for every  > 0 there is a y A

such that |x y| < . So x A, therefore we have B0 = A = A A. Now
(A) (B0 ) = (A A) = (A ) + (A) = (A) + (A)
Since A is open and A is measurable. Therefore (A) = (B0 ), if and only if (A) = 0

.

Exercise 6.3. Prove that an algebraic basis in any infinite-dimensional Banach space must be uncountable.
Proof: Let V be an infinite-dimensional Banach space over F, and suppose {xn }nN is a countable
Hamel basis. Then v V if any only if there exists ai F such that
v=

k
X

ai xi

i=1

for some xi {xn }. Now let hxi i denote the span of xi , then we have
[
V =
h{xn }kn=1 i
kN

But this implies that V is a countable union of proper subspace of finite dimension. Which implies
that V would be of first category, since every finite dimensional proper subspace of a normed space is
nowhere dense. Which is a contradiction to the Baire Category Theorem. Therefore any basis for an

33

infinite-dimensional Banach space must be uncountable

Theorem (Hahn-Banach) Let p be a real-valued function defined on the vector space X satisfying
p(x + y) p(x) + p(y) and p(x) = p(x) for each 0. Suppose that f is a linear functional defined
on a subspace S and that f (s) p(s) for all s S. Then there is a linear function F defined on X
such that F (x) p(x) for all x, and F (s) = f (s) for all s S.

Exercise 6.4. Let be a finite Borel measure on the real line, and set F (x) = {(, x]}. Prove
that is absolutely continuous with respect to the Lebesgue measure if and only if F is an absolutely
continuous function. In this case show that its Radon-Nikodym derivative is the derivative of F , almost
everywhere.
Proof: () First suppose that << . Let (E), then there exists an open set O, such that E O
and (O) < . Now O being open, there are disjoint intervals (xk , yk ), such that
[
X
O=
(xk , yk ),
(O) =
(yk xk ) < 
k=1

k=1

Since << , there exists a delta such that if (O) < , then (O) < . So we have
X
X
|F (yk ) F (xk )| =
(xk , yk ) <
k=1

k=1

So F (x) is an absolutely continuous function.

() Suppose that F (x) is absolutely continuous. Then we have
X
X
 > 0 > 0 s.t.
|yk xk | <

|F (yk ) F (xk )| < 

k=1

k=1

Choose such disjoint intervals (xk , yk ) and call the union of these intervals O, then we have (O) < .
Now by definition of F (x), we have
X
(O) =
|F (yk ) F (xk )| <
k=1

and so << .
To see that F is Radon-Nikodym derivative, we know that since F is absolutely continuous we have
that F 0 (t) exists almost everywhere so
Z x
(, x] = F (x) =
F 0 (t) d(t)

We also have that

x

Z
(, x] =

d =

d
d

which implies that

Z

(, x] = F (x) =

F 0 (t) d(t) =

d
d


d


d
Hence by the Radon-Nikodym theorem we know that F =
almost everywhere.
d
0

Theorem (Tonnelis) Suppose (X Y, (A B), ) is the product space of two -finite measure
spaces, and f : X Y [0, ] and f (x, y) be a nonnegative measurable function in the product
measure, then
Z
F1 (x) =
f (x, ) d is A measurable of x X
Y

34

Z
f (, y) d is B measurable of x Y

F2 (y) =
X

and
Z

Z
f d() =

X Y

Z
F1 d =

F2 d
Y

i.e., the iterated integrals is equal to the the integral in the product space


Z Z
Z Z
Z
f (x, y) d d =
f (, y) d d =
f (x, y) d( ).
X

X Y

Theorem (Fubinis) Suppose(X Y, (A B), ) is the product space of two -finite measure
spaces, and f : X Y [0, ] and f (x, y) be an integrable function in the product space, then


Z Z
Z Z
Z
f (x, y) d d =
f (, y) d d =
f (x, y) d( ).
X

X Y

Theorem (Fubini-Tonelli) Suppose (X Y, (A B), ) is the product space of two -finite

measure spaces. Let f be an extended real-valued (A B) measurable function on X Y. If either


Z Z
Z Z
|f | d d < or
|f | d d <
X

then f is -integrable, furthermore the iterated integrals are equal to the product integral.

Exercise 6.5. Let f be a real valued measurable function on the finite measure space (X , M, ). Prove
that the function F (x, y) = f (x) 5f (y) + 4 is measurable in the product measure space (X X , (M
M), ), and that F is integrable if and only if f is integrable.
Proof: First since f (x) is measurable, we have both sections F (x0 , y), and F (x, y0 ) as measurable
for each fixed x0 , y0 . Now for x X we have F (x, x) = 4(f (x) 1), which is measurable. Having
F (x, y) being measurable on each section, and the diagonal is enough for F (x, y) to be measurable in
the product space.
Z
Now let f be integrable, hence |f | is integrable, so let M =
|f (x)| dx, now we have
X

Z Z

Z
|f (x) 5f (y) + 4| dxdy

M + 5|f (y)|(X ) + 4(X ) dy

X

M (X ) + 5M (X ) + 4(X )2

4(X )(M + (X )) <

Z Z

|f (x) 5f (y) + 4| dydx <

=

by the same computation

Then by Fubini-Tonelli theorem F (x, y) is integrable. Now suppose that F (x, y) is integrable, then by
Fubinis theorem we have that the iterations are equal, but this is true if and only if f (x) is integrable
.
Theorem (Stone-Weierstrass) Let X be a compacct space and A an algebra of continuous realvalued functions on X that separates the points of X and contains the constant functions. Then given
any continuous real-valued function f on X and any  > 0 there is a function g A such that for all
x X we have |g(x) f (x)| < . In other words, A is a dense subset of C(X).
Theorem (Closed Graph) Let A be a linear transformation on a Banach space X to a Banach space
Y . Suppose that A has the property that, whenever xn is a sequence in X that converges to some point
x and Axn converges in Y to a point y, then y = Ax. Then A is continuous.

35

Exercise 6.6. Let (X , A, ) and (Y, B, ) be the measure spaces given by

X = Y = [0, 1]
A = B = ([0, 1])
be the Lebesgue measure on R, and the counting measure.
Consider the product measure space (X Y, (A B)), and its subset E = {(x, y) X Y : x = y}
(1) Show that E
R R (A B)
R R
(2) Show that X Y E d d 6= Y X E d d.
(3) Explain why Tonellis theorem is not applicable.
Proof: For (1) First notice that the following sets
 


k1 k
k1 k
,

,
Ak =
n
n
n
n
are measurable. Now let define En as follows
n
[
En =
Ak ,
k=1

Then the sets En are measurable as they are countable union of measurable sets. Then the set E is
given by

\
En = {(x, y) X Y : x = y}
E=
n=1

is measurable since is a countable intersection of measurable sets.

For (2) by a direct computation we have
Z Z
Z
E d d =
X

Z
E d d =

(E)d =

d = 1
0

Z
(E) d =

and
Z Z

0 d = 0
0

Tonellis theorem is not applicable because the measure space (Y, B, ) is not -finite

.