Chapter 4.

1, Page 117
Problem 10: A topological space X is called disconnected if there exist
nonempty open sets U, V such that U V = and U V = X; otherwise
X is connected. When we speak of connected or disconnected subsets of
X, we refer to the relative topology on them.
(a) X is connected i and X are the only subsets of X that are both
open and closed.
(b) If E

A
is a collection of connected subsets of X such that
A
E

,=
, then
A
E

is connected.
(c) If A X is connected, then

A is connected.=
(d) Every point x X is contained in a unique maximal connected subset
of X, and this subset is closed. (It is called the connected compo-
nent of x.)
Solution.
(a) If X is disconnected, there exist nonempty disjoint open sets U and
V with U V = X. Then U and V are both closed as well since they
are complements of open sets, so there exist clopen sets (both U and
V ) other than and X. Conversely, suppose F is a clopen set with
F ,= and F ,= X. Let U = F and V = X F. Then V is nonempty
since F ,= X and open because F is closed, so X is disconnected.
(b) Let X =

and suppose X is disconnected, so X = U V where U

and V are disjoint, open, and nonempty. For each E

, we must have
either E

U or E

V since otherwise E

= (E

U) (E

V )
is a disconnection of E

. Because U and V are both nonempty, there

exist E
1
U and E
2
V . But then E
1
E
2
= , contradicting the
fact that

is nonempty. Hence X must be connected.

(c) Suppose

A is disconnected. Then disjoint open U, V X with

A U V and

A U and

A V nonempty. Then A U must be
nonempty since otherwise X U would be a closed set containing A
so that

A U would be empty. Similarly, A V must be nonempty.
Then A = (A U) (A V ) is a disconnection of A.
(d) Given x X, let
C
x
=
_
xU
U connected
U.
Then C
x
is connected by part (b), since x U. Since every connected
set containing x is a subset of C
x
, it is the maximal connected subset
containing x. Now suppose y C

x
. By the maximality of C
x
, C
x
y
is disconnected, so it is contained in disjoint open sets U and V . These
cannot both intersect C
x
since this would give a disconnection of C
x
,
so there exist U and V open and disjoint with C
x
U and y V .
But then y V C

x
, so C

x
is open, so C
x
is closed.

Probem 13: If X is a topological space, U is open in X, and A is dense in

X, then

U = U A.
1
2
Solution. Obviously U A

U since U A U. Conversely, since A =
(A U) (A U

) where U

= X U, we have
X =

A = A U A U

.
Now U

is a closed set containing A U

, so A U

. Therefore
U A U. But this implies

U A U.
Chapter 4.2, Page 123
Problem 26: Let X and Y be topological spaces.
(a) If X is connected and f C(X, Y ), then f(X) is connected.
(b) X is called arcwise connected if for all x
0
, x
1
X there exists f
C([0, 1], X) with f(0) = x
0
and f(1) = x
1
. Every arcwise connected
space is connected.
(c) Let X = (s, t) R
2
: t = sin(s
1
) (0, 0), with the relative topol-
ogy induced from R
2
. Then X is connected but not arcwise connected.
Solution.
(a) Suppose f(X) = U V where U, V are disjoint, nonempty, and open.
Then X = f
1
(f(X)) = f
1
(U) f
1
(V ) which are clearly disjoint
and nonempty, and are open because f is continuous.
(b) Let X be arcwise connected, and suppose X is disconnected, so X =
U V where U, V are disjoint, nonempty, and open. Let x
1
U and
x
2
V . Let C be a curve from x
1
to x
2
. Then C = (C U) (C V )
and both are nonempty because x
1
C U and x
2
C V . This
contradicts the fact that C is connected by part (a) because it is the
image of the connected set [0, 1] under a continuous function.
(c) First, we show that X is connected: Suppose U, V is a disconnection of
X. Suppose WLOG that (0, 0) U. Since V is nonempty it intersects
either (s, sin(s
1
)) : s > 0 or (s, sin(s
1
)) : s < 0, WLOG the
former, which we call X

. But U intersects X

as well since it contains a

neighborhood of the origin. Then (X

U)(X

V ) is a disconnection
of X

. But X

is the image of the connected set (0, ) R under

the continuous function x (x, sin(x
1
)) and hence connected, a
contradiction. Therefore X must be connected.
To show that X is not path connected, suppose : [0, 1] X is a
path with (0) = (0, 0) and (1) = (
1
2
, 0). Then I =
x
((t)) is
a connected subset of the real line, where
x
((a, b)) = a, since
x
and are both continuous. Since I contains 0 and
1
2
, it contains
everything in between. This implies that the image of includes the
points p
n
= (
1
2n+/2
, 1) for n = 1, 2, . . . . But p
n
(0, 1) which is not
in X and hence not in the image of . Thus, the image of is not closed
in R
2
; but in fact it must be compact and therefore closed because is
continuous and [0, 1] is compact. This contradiction proves that there
is no path from (0, 0) to (
1
2
, 0), so X is not path connected.

Problem 27: If X

is connected for each A, then X =

A
X

is
connected.
Solution. We rst solve the problem for nite products:
3
Lemma 1. The product of nitely many connected spaces is connected.
Proof. By induction, it is sucient to prove this for the product of two
spaces, since

n
j=1
X
j
= (

n1
j=1
X
j
)X
n
. Suppose X and Y are connected.
If XY is disconnected, let U and V be disjoint open nonempty sets with
U V = X Y . Let (x
0
, y
0
) U and (x
1
, y
1
) V . If (x
0
, y
1
) U,
let Z = (x, y
0
) : x X, which is homeomorphic to X by Exercise 18.
Then Z U and Z V are both nonempty, since (x
0
, y
1
) U Z and
(x
1
, y
1
) V Z. But then Z = (U Z) (V Z) is a disconnection of
Z, a contradiction because X is connected. Similarly, if (x
0
, y
1
) V , let
Z = (x
0
, y) : y Y which is homeomorphic to Y by Exercise 18, and we
obtain a disconnection of Y by the same argument. Hence X Y must be
connected.
Now let X =

be an arbitrary product of connected spaces. Choose

any x X. For any nite set
1
, . . . ,
n
of indices, let
U
1,...,n
= y X :

(y) =

(x) for all ,=

1
, . . . ,
n
.
Then U
1,...,n
is homeomorphic to X
1
X
n
by Exercise 18, and
therefore connected by the lemma. Since each U
1,...,n
contains x, their
union, which we call U, is connected by Exercise 10b. Therefore

U is also
connected by Exercise 10c. Finally, we will show that

U = X. Let V X
be any open set. Using the basis for the product topology, V must contain
a set of the form

V = U
1
U
m

=1,...,m
X

for some open sets U

i
X
i
. Let z U
1
U
m
be any point, and
dene a point z X by

(z) =
_

i
( z) =
1
, . . . ,
m

(x) else.
Then z U
1,...,m
U since it equals x in all but nitely many coordi-
nates; but z

V V as well. Hence every open set V X contains a
point of U, so

U = X which must be connected.
Chapter 4.4, Page 130
Problem 39: Every sequentially compact space is countably compact.
Solution. Let X be a sequentially compact space and U
1
, U
2
, . . . a count-
able open cover. Suppose there is no nite subcover. Construct a sequence
x
n
as follows: Let j
1
= 1 and x
1
be any point of U
j1
= U
1
. Next, let
j
2
be the smallest n 2 such that U
j2
U
1
,= , and let x
2
U
j2
U
1
.
Then let x
3
U
j3
(U
j1
U
j2
) where j
3
is the smallest possible value for
which the latter set is nonempty. This process cannot terminate because
that would imply that U
j1
, . . . , U
jm
would be a nite subcover. Note that
x
n
has the property that x
n
/ U
i
for any i < j
n
. Now let x be the limit
of some subsequence x
n
k
. Let U
M
be any open set containing x. Then
U
M
contains all but nitely many of the x
n
k
. But x
n
k
cannot be in U
i
for
4
i < n
k
, a contradiction. Hence the process must terminate, i.e. there is a
nite subcover.
Problem 43: For x [0, 1), let

1
a
n
(x)2
n
(a
n
(x) = 0 or 1) be the base-2
decimal expansion of x. (If x is a dyadic rational, choose the expansion
such that a
n
(x) = 0 for n large.) Then the sequence a
n
) in 0, 1
[0,1)
has
no pointwise convergent subsequence. (Hence 0, 1
[0,1)
with the product
topology arising from the discrete topology on 0, 1 is not sequentially
compact. It is, however, compact.)
Solution. Suppose some subsequence a
nj
converges to a function a 0, 1
[0,1)
.
Let z [0, 1) be a number such that a
nj
(z) is 0 for j even and 1 for j odd.
Then b 0, 1
[0,1)
: b(z) = a(z) is a neighborhood of a in 0, 1
[0,1)
.
However, it does not contain all but nitely many of the a
nj
; it only con-
tains a
nj
for j even if a(z) = 0, and for j odd if a(z) = 1. Hence a
nj
does
not converge to a, a contradiction.
Chapter 4.5, Page 134
Problem 49: Let X be a compact Hausdor space and E X.
(a) If E is open, then E is locally compact in the relative topology.
(b) If E is dense in X and locally compact in the relative topology, then
E is open. (Use Exercise 13.)
(c) E is locally compact in the relative topology i E is relatively open in

E.
Solution.
(a) Let x E. By Proposition 4.30, x N E for some compact
neighborhood N of x. Then N is also a compact neighborhood in the
relative topology, so E is locally compact.
(b) For any x E, we have x

U

K E where

U is open in E and

K compact in E in the relative topology. This means that

U = U E
for some U open in X. Moreover,

K must be compact as a subset of
X: If U

is any open cover, U

E is an open cover for

K in the
relative topology, which has a nite subcover U
1
E, . . . , U
n
E,
from which it follows that U
1
, . . . , U
n
is a nite subcover. Then
U

U = U E =

K =

K E,
where all the closures are taken in X, by Exercise 13. So x is contained
in an open set contained in E. Therefore E is open.
(c) Suppose E is locally compact in the relative topology. Now

E is a
compact Hausdor space in which E is dense, so E is open in

E by part
(b). Conversely, suppose E is open in

E. For any x E, Proposition
4.30 applied to the LCH space

E implies that there is a compact
neighborhood N E

E, so E is locally compact.

Problem 50: Let U be an open subset of a compact Hausdor space X and

U

its one-point compactication. If : X U

is dened by (x) = x if
x U and (x) = if x U
c
, then is continuous.
5
Solution. Let V U

be any open set. We wish to show that

1
(V ) is
open in X. Suppose rst that V is an open subset of U. (V is open in the
relative topology, but since U is open this implies it is also open as a subset
of X.) Then
1
(V ) = V since is the identity map on U and everything
else is sent to / V . Thus
1
(V ) is open in X. Now consider the other
case, where V = (U K) where K U is compact. Now

1
(V ) =
1
()
1
(U K) = (X U) (U K) = X K
which is open in X. Hence is continuous.
Problem 56: Dene : [0, ] [0, 1] by (t) = t/(t +1) for t [0, ) and
() = 1.
(a) is strictly increasing and (t +s) (t) + (s).
(b) If (Y, ) is a metric space, then is a bounded metric on Y that
denes the same topology as .
(c) If X is a topological space, the function (f, g) = (sup
xX
[f(x)
g(x)[) is a metric on C
X
whose associated topology is the topology of
uniform convergence.
(d) If X is a -compact LCH space and U
n

1
is as in Proposition 4.39,
the function
(f, g) =

1
2
n

_
sup
x

Un
[f(x) g(x)[
_
is a metric on C
X
whose associated topology is the topology of uniform
convergence on compact sets.
Solution.
(a) That is strictly decreasing follows from the fact that
1 (t) =
_
1
t+1
0 t <
0 t =
is strictly decreasing, which is obvious since
1
t+1
is strictly positive and
decreasing on [0, ). The inequality (s +t) (s) +(t) is obvious
if either s or t is , because if e.g. s = the LHS is 1 and the RHS
is 1 plus some nonnegative number. If neither is ,
(s + 1)(t + 1) 1 +s +t

s +t + 2
(s + 1)(t + 1)

2 +s +t
1 +s +t

1
s + 1
+
1
t + 1
1 +
1
s +t + 1
1
1
s +t + 1
1
1
s + 1
+ 1
1
t + 1
(s +t) (s) + (t).
(b) Clearly is nonnegative and symmetric. Moreover, since (t) = 0
i t = 0, we have
((x, y)) = 0 (x, y) = 0 x = y.
6
Finally,
((x, y)) ((x, z) +(z, y)) (since is increasing)
((x, z)) + ((z, y)) (since (t +s) (t) + (s))
so satises the triangle inequality. Thus, it is a metric, and
is obviously bounded since its values are in [0, 1]. The collection of
positive-diameter balls is the same in both metrics, since (x, x
0
) <
((x, x
0
)) <

1+
and conversely ((x, x
0
)) < (x, x
0
) <

1
. Since balls form the basis of the metric topology, the topologies
induced by the two metrics are identical.
(c) This follows immediately from part (b) with as the uniform metric
on C
X
.
(d) That this is a metric follows from the fact that any nite-valued sum
of metrics is another metric (the axioms are trivial to check). Now
suppose that f
n
f in this topology, i.e. (f
n
, f) 0. This is true
i (sup
x

Um
[f
n
(x) f(x)[) 0 for all m, which in turn is true i
sup
x

Um
[f
n
(x) f(x)[ 0 for all m. By Proposition 4.40 this is true
i f
n
f uniformly on compact sets. Hence the topology dened by
is precisely the topology of uniform convergence on compact sets.

Chapter 4.7, Page 142

Problem 68: Let X and Y be compact Hausdor spaces. The algebra gen-
erated by functions of the form f(x, y) = g(x)h(y), where g C(X) and
h C(Y ), is dense in C(X Y ).
Solution. This is an almost trivial application of the Stone-Weierstrass the-
orem, since separable functions g(x)h(y) include the constant functions,
and they can easily be seen to separate points: Given (x
1
, y
1
) and (x
2
, y
2
),
suppose WLOG that x
1
,= x
2
(since the points must dier in at least one
coordinate). Let h(y) = 1 and g(x) be any continuous function which takes
dierent values at x
1
and x
2
; such g must exist by Urysohns lemma. Then
g(x)h(y) takes dierent values at (x
1
, y
1
) and (x
2
, y
2
).
Chapter 4.8, Page 146
Problem 74: Consider N (with the discrete topology) as a subset of its Stone-
Cech compactication N.
(a) If A and B are disjoint subsets of N, their closures in N are disjoint.
(Hint.)
(b) No sequence in N converges in N unless it is eventually constant (so
N is emphatically not sequentially compact).
Problem 77: Let X
n
,
n

1
be a countable family of metric spaces whose
metrics take values in [0, 1]. (The latter restriction can always be satised;
see Exercise 56b.) Let X =

1
X
n
. If x, y X, say x = (x
1
, x
2
, . . . ) and
y = (y
1
, y
2
, . . . ), dene (x, y) =

1
2
n

n
(x
n
, y
n
). Then is a metric
that denes the product topology on X.
7
Solution. That is a metric follows from the fact that it is a nite-valued
sum of metrics. To see that it denes the product topology, let
U = U
i1
U
in

i=i1,...,in
R
be a basis element for the product topology. Let x U and choose neigh-
borhoods (x
ij

j
, x
ij
+
j
) U
ij
for j = 1, . . . , n. Then z X : (z, x) <
min(
j
2
i
j
) is contained in U, so the metric topology is ner than the product
topology. Conversely, let
V = z X : (z, x) <
be a basis element for the metric topology. Choose N such that
1
2
N
<

2
.
Then the set
N

i=1
(x
i


2
, x
i
+

2
)

i>N
R
is contained in V , so that the product topology is ner than the metric
topology.
Chapter 5.1, Page 154
Problem 6: Suppose that X is a nite-dimensional vector space. Let e
1
, . . . , e
n
be a basis for X, and dene |

n
1
a
j
e
j
|
1
=

n
1
[a
j
[.
(a) | |
1
is a norm on X.
(b) The map (a
1
, . . . , a
n
)

n
1
a
j
e
j
is continuous from K
n
with the usual
Euclidean topology to X with the topology dened by | |
1
.
(c) x X : |x|
1
= 1 is compact in the topology dened by | |
1
.
(d) All norms on X are equivalent. (Compare any norm to | |
1
.)
Solution.
(a)
Obviously |x|
1
0 for all x.
|x|
1
= 0 a
j
= 0 for all j x = 0.
|c(

a
j
e
j
)|
1
= |

ca
j
e
j
|
1
=

[ca
j
[ = [c[

[a
j
[ = [c[|

a
j
e
j
|
1
Finally,
_
_
_
_

a
j
e
j
_
+
_

b
j
e
j
__
_
_
1
=
_
_
_

(a
j
+b
j
)e
j
_
_
_
1
=

[a
j
+b
j
[

([a
j
[ +[b
j
[)
=

[a
j
[ +

[b
j
[
=
_
_
_

a
j
e
j
_
_
_
1
+
_
_
_

b
j
e
j
_
_
_
1
.
(b) Since K
n
is a Banach space, we need only check continuity at 0. Let
> 0 and choose p K
n
with |p| < /n. Then

n
1
[p
i
[
2
<
2
/n
2
so
[p
i
[ < /n for all i. Then |f(p
i
)|
1
=

[p
i
[ < .
(c) The set (k
1
, . . . , k
n
) K
n
:

[k
i
[ = 1 is compact in K
n
by the
Heine-Borel theorem, since it is closed and bounded. Since the set x
X : |x|
1
= 1 is the image of this set under the map (a
1
, . . . , a
n
)

a
i
e
i
, part (b) implies that it is compact as well.
8
(d) Let | | be any other norm on X. Then for x =

a
i
e
i
,
|x|

[a
i
[|e
i
| M

[a
i
[ = M|x|
1
where M = max(|e
i
|). This implies that | | is continuous with
respect to | |
1
. Now since the function x |x| is a continuous
function on (X, | |
1
) and the set x X : |x|
1
= 1 is compact,
the function |x| attains a minimum m on this set. Since |x| , = 0
for x ,= 0, this minimum is nonzero. By scaling properties of norms,
|x| m|x|
1
for all x X. Hence | |
1
is continuous with respect to
| |.

Problem 8: Let (X, /) be a measurable space, and let M(X) be the space
of complex measures on (X, /). Then || = [[(X) is a norm on M(X)
that makes M(X) into a Banach space. (Use Theorem 5.1).
Solution. Obviously M(X) is a vector space over C. By Proposition 3.14,
the total variation is a norm on this vector space. (The norm properties
other than the triangle inequality are completely obvious.)
To show that M(X) is complete and therefore a Banach space, we need
to show that every absolutely convergent sequence converges. Let
i
be
complex measures such that

|
i
| < . Let be a xed positive -nite
measure on X. Then we can write d
i
= f
i
d by the Radon-Nikodym
theorem; by hypothesis we have

i=1
|
i
| =

i=1
_
X
d[
i
[ =

i=1
_
X
[f
i
[d =
_
X

i=1
[f
i
[d <
so that f =

i=1
f
i
L
1
() by the triangle inequality. Dene by d =
fd. Then
|| =
_
X
[f[d
_
X

[f
i
[d <
so M(X). Moreover, if
N
=

N
i=1

i
, then d
N
=

i=N+1
f
i
d,
and
|
N
| =
_
X

i=N+1
f
i

d
_
X

i=N+1
[f
i
[d =

i=N+1
|
i
| 0
since the tails of the convergent series

|
i
| tend to 0.
Problem 11: If 0 < 1, let

([0, 1]) be the space of Holder continuous

functions of exponent on [0, 1]. That is, f

([0, 1]) i |f|

< ,
where
|f|

= [f(0)[ + sup
x,y[0,1], x=y
[f(x) f(y)[
[x y[

.
(a) | |

is a norm that makes

([0, 1]) into a Banach space.

(b) Let

([0, 1]) be the set of all f

([0, 1]) such that

[f(x) f(y)[
[x y[

0 as x y, for all y [0, 1].

9
If < 1,

([0, 1]) is an innite-dimensional closed subspace of

([0, 1]).
If = 1,

([0, 1]) contains only constant functions.

Solution.
(a) If f ,= 0, then either [f(0)[ , = 0, in which case |f|

> 0, or f(0) = 0
and y [0, 1] with f(y) ,= 0, in which case
|f|

[f(y) f(0)[
[y 0[

=
[f(y)[
[y[

> 0.
On the other hand, if f = 0 then clearly |f|

= 0.
Also,
|cf|

= [cf(0)[ + sup
[cf(x) cf(y)[
[x y[

= [c[[f(0)[ + sup
[c[[f(x) f(y)[
[x y[

= [c[[f(0)[ +[c[ sup

[f(x) f(y)[
[x y[

= [c[|f|

.
Finally,
|f +g|

= [f(0) +g(0)[ + sup

[f(x) +g(x) f(y) g(y)[
[x y[

[f(0)[ +[g(0)[ + sup

[f(x) f(y)[ +[g(x) g(y)[
[x y[

[f(0)[ +[g(0)[ + sup

[f(x) f(y)[
[x y[

+ sup
[g(x) g(y)[
[x y[

= |f|

+|g|

.
Hence, | |

is a norm. To show that this normed linear space is com-

plete, suppose f
i

([0, 1]) with

|f
i
|

< . Then |

f
i
|

<
by the triangle inequality, so if we dene f =

f
i
then f

([0, 1]).
Let g
N
= f

N
i=1
f
i
=

i=N+1
f
i
. Then |g
N
|

i=N+1
|f
i
|
0. Since every absolutely convergent sequence converges,

([0, 1]) is
Banach.
(b) Since
[af(x) +bg(x) af(y) bg(y)[
[x y[

[a[
[f(x) f(y)[
[x y[

+[b[
[g(x) g(y)[
[x y[

,
if f, g

then af +bg

. To show this subspace is closed, suppose

f
n
f in the | |

norm. Since
|f
n
f|

= [f
n
(0) f(0)[ + sup
[f
n
(x) f(x) f
n
(x) +f(y)[
[x y[

,
this implies
sup
[f
n
(x) f(x) f
n
(x) +f(y)[
[x y[

0.
10
Given > 0, choose N such that this supremum is less than

2
for
n = N. For xed y, we can then choose > 0 such that
[f
N
(x) f
N
(y)[
[x y[

<

2
for 0 < [x y[ < . Then for 0 < [x y[ < ,
[f(x) f(y)[
[x y[

[f
N
(x) f
N
(y)[
[x y[

+
[f(x) f
N
(x) f(y) +f
N
(y)[
[x y[

<

2
+

2
= .
Hence, f

as well. Finally, note that for = 1 the dierence

quotient tending to zero implies that f is dierentiable everywhere
with derivative zero, so that f is constant. On the other hand, for
0 < < 1 the functions f
n
(x) = x
n
: n = 0, 1, . . . are all in

because
[f
n
(x) f
n
(y)[
[x y[

=
[x
n
y
n
[
[x y[

= [x y[
1
[x
n1
+x
n2
y + +xy
n2
+y
n1
[
n[x y[
1
0
as x y. Hence the space

is innite-dimensional.

Problem 15: Suppose that X and Y are normed vector spaces and T
L(X, Y ). Let N(T) = x X : Tx = 0.
(a) N(T) is a closed subspace of X.
(b) There is a unique S L(X/N(T), Y ) such that T = S where
: X X/M is the projection. Moreover, |S| = |T|.
Solution.
(a) N(T) is a subspace by the linearity of T: If x, y N(T) then T(ax +
by) = aT(x) +bT(y) = 0 so ax +by N(T). It is closed because T is
continuous: if x
n
N(T) and x
n
x, 0 = T(x
n
) T(x).
(b) The equation S = T is equivalent to S(x + N(T)) = T(x) for all
x X. For x+M X/N(T), we dene S(x+M) to be T(x). Clearly
this map is unique if it works; it does because if x
1
+ M = x
2
+ M,
then x
2
x
1
N(T) so T(x
2
) = T(x
1
) +T(x
2
x
1
) = T(x
1
). So S is
well-dened. Because the projection map has norm 1 by Exercise 12c,
|T| |S||| = |S|. On the other hand, suppose |x + N(T)| = 1.
By denition this means that for any given there exists y N(T)
with |x +y| < 1 +. Then
|S(x +N(T))| = |T(x)| = |T(x +y)| |T||x +y| < (1 +)|T|.
This is true for all , so |S(x + N(T))| |T|. Hence |S| |T|
(which also implies S L(X/N(T), Y )).

Chapter 5.2, Page 159

Problem 18: Let X be a normed vector space.
(a) If M is a closed subspace and x X M then M +Cx is closed. (Use
Theorem 5.8a.)
11
(b) Every nite-dimensional subspace of X is closed.
Solution.
(a) Let z
n
= m
n
+
n
x z where m
n
M and
n
C. We wish to show
z M +Cx. By Theorem 5.8a, there exists X

with (x) = ,= 0
and (M) = 0. Then

n
= (m
n
+
n
x) = (z
n
) (z)
by continuity. This implies that the sequence
n
is convergent to some
number C. Let z

= z x. Then
(z
n
z) + (
n
)x = m
n
z

since both sides equal m

n
+xz. But both terms on the left converge
to 0, so m
n
z

. Since M is closed, this implies z

M, so z =
z

+x M +Cx.
(b) Let e
1
, . . . , e
n
be a basis for a nite-dimensional space F X. Since
0 is closed, it follows from part (a) and induction that span(e
1
, . . . , e
k
)
is closed for k = 1, . . . , n.

Problem 24: Suppose that A is a Banach space.

(a) Let

A, (A

be the natural images of A, A

in A

, A

, and
let

A
0
= F A

: F[

A = 0. Then (A

A
0
= 0 and
(A

+

A
0
= A

.
(b) A is reexive i A

is reexive.
Solution.
(a) (A

A
0
is the (image of the) stu in X

that is 0 on all of A. This

is just the 0 vector. On the other hand, given anything in X

, its
restriction to

A yields a bounded linear functional on X; subtracting
this yields something which is 0 on

A, which proves

A
0
+ (A

=
A

. Note that these two together allow us to write X

= (A

A)
0
.
(b) First we note that reexivity of A is equivalent to the condition that

A
0
= 0. It is clear that reexivity implies this condition; for the
reverse, if A is not reexive, let v A

A, and use Theorem 5.8a
to get a nonzero element of A

which is 0 on

A. Using this and the
direct sum from part (a), we have
A

reexive (A

= A

A
0
= 0
A reexive.

Problem 25: If X is a Banach space and X

is separable, then X is separable.

Solution. Let
n
be a countable dense subset of X

. Choose unit vectors

x
n
X with |
n
(x
n
)|
1
2
|
n
|. Let V be the set of all nite rational linear
combinations of the x
i
. Then V is countable; we will show that it is dense
in X. First, we note that

V is the same as the closure of span(x
n
) since
12
any linear combination may be approximated by a rational combination.
In particular,

V is a linear space. Now suppose there exists y /

V . Then
the Hahn-Banach theorem implies a functional X

with norm 1 such

that (y) = 1 and (v) = 0 for all v

V . Since the
n
are dense, there is a
sequence
n
k
. Since |x
n
k
| = 1 this implies
|
n
k
(x
n
k
)| = |
n
k
(x
n
k
) (x
n
k
)| |
n
k
||x
n
k
| 0.
But |
n
k
(x
n
k
)|
1
2
|
n
k
|, so |
n
k
| 0. This contradicts the fact that

n
k
and || = 1. Hence there is no y /

V , so V is dense in X.
Problem 26: Let X be a real vector space and let P be a subset of X such
that
(i) If x, y P, then x +y P;
(ii) If x P and 0, then x P;
(iii) If x P and x P, then x = 0.
(Example: If X is a space of real-valued functions, P can be the set of
nonnegative functions in X.)
(a) The relation dened by x y i y x P is a partial ordering on
X.
(b) (Krein Extension Theorem) Suppose that M is a subspace of X
such that for each x X there exists y M with x y. If f is a
linear functional on M such that f(x) 0 for x M P, there is a
linear functional F on X such that F(x) 0 for x P and F[M = f.
(Hint.)
Solution.
(a) Suppose x y and y x. By denition, this means x y P and
y x P; by the third property, this implies x y = 0, i.e. x = y.
Now suppose x y and y z. Then y x P and z y P; by the
rst property this implies z x P, so x z. Hence is a partial
ordering.
(b) Dene p : X R by
p(x) = inff(y) : y M, y x.
Then p(x) is a sublinear functional:
First, p is sublinear because
p(x
1
+x
2
) = inff(y) : y M, y x
1
x
2
P
inff(y
1
) : y
1
M, y
1
x
1
P + inff(y
2
) : y
2
M, y
2
x
2
P
= p(x
1
) +p(x
2
)
since any such y
1
and y
2
generate a corresponding y = y
1
+y
2
.
Suppose 0. If = 0 then p(x) = 0 for any x since f(y) 0
for all y with y = y 0 P and this minimum is attained at
y = 0. If > 0 then
p(x) = inff(y) : y M, y x P
= inff(u) : u M, u x P = p(x)
= inff(u) : u M, u x P = p(x)
13
where u =
1

y.
Now for x M, if y M such that y x P, then
f(y) = f(y x) +f(x) f(x)
since y x P. Hence f(x) = p(x) on M. By the Hahn-Banach
theorem, there exists an extension F : X R with F[
M
= f and
F(x) p(x) for all x. To show this implies F(x) 0 for x P,
suppose to the contrary that x P and F(x) < 0. Then F(x) > 0;
however, p(x) = 0 because 0 M and 0 (x) P. This contradicts
the construction of F as satisfying F(x) p(x). So F(x) 0 for
x P.

Chapter 5.3, Page 164

Problem 28: The Baire category theorem remains true if X is assumed to be
an LCH space rather than a complete metric space. (The proof is similar;
the substitute for completeness is Proposition 4.21.)
Solution. Let X be LCH and U
n
open and dense. Let W X be open.
We wish to show W intersects U
n
. Since U
1
is dense, U
1
W is nonempty,
and of course its open. By Proposition 4.30, every open set in an LCH space
contains a compact set which contains another open set. Let V
1
K
1

(U
1
W) with V
1
open and K
1
compact. Now U
2
is dense, so it intersects
V
1
. Let V
2
K
2
(V
1
U
2
) with V
2
open and K
2
compact. Continuing,
we get a nested sequence K
n
of nonempty compact sets. By Proposition
4.21, K
n
is nonempty; but K
n
U
n
W, so W (U
n
) is nonempty.
Problem 29: Let Y/L
1
() where is counting measure on N, and let X =
f Y :

1
n[f(n)[ < , equipped with the L
1
norm.
(a) X is a proper dense subspace of Y ; hence X is not complete.
(b) Dene T : X Y by Tf(n) = nf(n). Then T is closed but not
bounded.
(c) Let S = T
1
. Then S : Y X is bounded and surjective but not
open.
Solution.
(a) X is a subspace because

n[f(n)[ <

n[cf(n)[ = [c[

n[f(n)[ <
and

n[f(n)[,

n[g(n)[ <

n[f(n) +g(n)[

n([f(n)[ +
[g(n)[) < . It is proper because it does not contain the function
f(n) =
1
n
. It is dense because it contains all sequences with nitely
many nonzero terms, which are dense because every series is the limit
of its partial sums.
(b) To show T is closed, suppose (f
k
, Tf
k
) converges to a point (f, g) in
the product space X Y . By denition of the product metric, this
implies f
k
f in X and Tf
k
g in Y . The former condition means

[f
k
(n) f(n)[ 0 as k . But
1
convergence implies point-
wise convergence because counting measure has no sets with measures
between 0 and 1, so f
k
(n) f(n) for all n. Similarly, nf
k
(n) g(n)
for all n. So g(n) = nlim
k
f
k
(n) = nf(n). Hence (f, g) is on the
14
graph, so the graph is closed. However, T is not bounded: Consider
the functions
f
k
(n) =
kn
=
_
1 n = k
0 else.
Then |f
k
| = 1 but |Tf
k
| = n. Hence T is not bounded.
(c) Since

n[f(n)[

[f(n)[ for any f

1
, |S| 1. (In fact the norm
is 1, as can be seen from the unit sequence 1,0,0, . . . .) S is clearly
surjective since for any f X, S(Tf) = f. However, the image of
the open unit ball in
1
does not contain the functions g
k
(n) =
f
k
(n)
k
because g
k
= Sf
k
. But |g
k
| =
1
k
0, so the image of the unit ball is
not open.

Chapter 5.4, Page 170

Problem 45: The space C

(R) of all innitely dierentiable functions on R

has a Frechet space topology with respect to which f
n
f i f
(k)
n
f
(k)
uniformly on compact sets for all k 0.
Solution. Let

j,k
(f) = sup
x

Uj
[f
(k)
(x)[.
where the U
j
are as dened in Proposition 4.39. By Proposition 4.40,
f
(k)
n
f
(k)
uniformly on compact sets i f
(k)
n
f
(k)
uniformly on

U
j
for all
j. Hence f
(k)
n
f
(k)
uniformly on compact sets for all k i
j,k
(f
n
f) 0
for all j and k. C

(R) is Hausdor in this topology by Proposition 5.16a

because every nonzero function is nonzero at some point. To show it is
complete, suppose f
n
is a Cauchy sequence, so for any j and k,
j,k
(f
n

f
m
) 0 as m, n . Let f be the pointwise limit of f
n
, which must
exist and be continuous since Cauchyness in this space implies Cauchyness
in the metric space C([0, 1]). It follows by induction that f
(k)
n
f
(k)
for all
k; here we use the theorem that if the derivatives of a pointwise convergent
sequence of functions converge uniformly, then the limit of the derivatives
is the derivative of the limit. (This theorem can be found, for example, on
page 356 of Langs Real and Functional Analysis; it may also be in Folland
but I dont know where.) Hence C

is a Frechet space in this topology.

Problem 46: If X is a vector space, Y a normed linear space, T the weak
topology on X generated by a family of linear maps T

: X Y , and
T

the topology dened by the seminorms x |T

x|, then T = T

.
Solution. The basis for the weak topology generated by the T

is sets of
the form
U
y
= T
1

(B

(y)) = x : |T

(x) y| < .
On the other hand, the basis for the topology generated by the seminorms
is sets
V
x0
= x : |T

(x) T

(x
0
)| < .
15
Obviously, every member of this generating set is a member of the weak
generating set. On the other hand, let U
y
be a generating set for the
weak topology. If it is nonempty, let x U
y
, and let |T

(x) y| = < .
Then V
x()
= z : |T

(z) T

(x)| < is contained in U

y
. So
every member of this generating set contains a member of the generating
set for the seminorm-induced topology.
Chapter 6.1, Page 186
Problem 7: If f L
p
L

for some p < , so that f L

q
for all q > p,
then |f|

= lim
q
|f|
q
.
Solution. By replacing f with
f
f
, we may assume WLOG that |f|

=
1. (Unless, of course, f 0, in which case the problem is trivial.) For any
> 0, let A = x : [f(x)[ > 1 . Then 0 < (A) < because |f|

= 1
and because f L
p
. Now f > (1 )
A
, so |f|
q
((A))
1/q
(1 ).
Because x
1/q
1 as q for every positive number x, this implies that
|f|
q
is eventually greater than 1 2. On the other hand, [f[
q
[f[
p
a.e.
for all q > p, so |f|
q
|f|
p/q
p
which tends to 1 as q , so |f|
q
is
eventually less than 1 +.
Problem 8: Suppose (X) = 1 and f L
p
for some p > 0, so that f L
q
for 0 < q < p.
(a) log |f|
q

_
log [f[.
(b) (
_
[f[
q
1)/q log |f|
q
, and (
_
[f[
q
1)/q
_
log [f[ as q 0.
(c) lim
q0
|f|
q
= exp(
_
log [f[).
Solution.
(a) Because e
x
is convex, Jensens inequality implies
e

log |f|
q

_
e
log |f|
q
(e

log |f|
)
q

_
[f[
q
e

log |f|
|f|
q

_
log [f[ log |f|
q
.
(b) Because y 1 log y for all positive real numbers y (this is easily
checked by noting their equality at y = 1 and comparing derivatives),
_
[f[
q
1 log
_
[f[
q
= log |f|
q
q
= q log |f|
q
.
By LHopitals Rule,
lim
q0
_
[f[
q
1
q
= lim
q0
_
[f[
q
log [f[ =
_
log [f[
where the dierentiation under the integral is justied by Theorem
2.27. The question then becomes how to justify the nal equality. If
[f[ = 0 on a set of positive measure then we dont even need LHopitals
Rule because the numerator
_
[f[
q
1 approaches a nonzero negative
number, so

|f|
q
1
q
=
_
log [f[. Hence, we assume [f[ , = 0
a.e. and write X =

j=0
E
j
where E
0
= [f[ 1 and E
j
=
1
j+1

16
[f[ <
1
j
for j 1. Now on E
0
, we can choose any p

< p; then
[f[
p

log [f[ = O([f[

p
) L
1
and [f[
q
log [f[ < [f[
p

log [f[ for q < p

, so
_
E0
[f[
q
log [f[
_
E0
log [f[ by the Dominated Convergence Theorem.
For the rest of the domain, if
_
X\E0
log [f[ =

j=1
_
Ej
log [f[
diverges, then for any M there exists k such that

k
j=1
_
Ej
log [f[ <
M 1. By choosing q
0
small enough that [f[
q
log [f[ < log [f[
M
M+1
on E
1
, . . . , E
k
, we guarantee that
_
X\E0
[f[
q
log [f[
k

j=1
_
[f[
q
log [f[ M
for q < q
0
, so this integral goes to as well. On the other hand, if
_
log [f[ converges, choose k so that

j=k+1
log [f[ has absolute value
less than . Then [f[
q
log [f[ log [f[ uniformly on each E
1
, . . . , E
k
,
so
_
E1E
k
[f[
q
log [f[
_
E1...E
k
log [f[.
Moreover, [
_
Ej
[f[
q
log [f[[ [
_
Ej
log [f[[ for j > k, so

j=k+1
Ej
[f[
q
log [f[

j=k+1

_
Ej
log [f[

< .
This implies that
_
X\E0
[f[
q
log [f[ and
_
X\E0
log [f[ are eventually
within 2.
(c) By parts (a) and (b) and the Sandwich Theorem, log |f|
q

_
log [f[.
Exponentiating both sides gives the result.

Problem 11: If f is a measurable function on X, dene the essential range

R
f
of f to be the set of all z C such that x : [f(x) z[ < has positive
measure for all > 0.
(a) R
f
is closed.
(b) If f L

, then R
f
is compact and |f|

= max[z[ : z R
f
.
Solution.
(a) Let z
n
z where z
n
R
f
. Given > 0, choose z
N
with [z
N
z[ <

2
.
Then x : [f(x)z
N
[ <

2
x : [f(x)z[ < , and since the former
set has positive measure, the latter does as well. Hence z R
f
.
(b) Let M = |f|

. Then R
f
B
M
(0) so its bounded as well as closed,
hence compact by the Heine-Borel theorem.
Clearly max
zR
f
[z[ M. If it were strictly less than M, say ,
choose

with <

< M. Then every point z in the closed annulus

A = z :

[z[ M has a neighborhood N

z
such that x : f(x)
N
z
has measure zero. Since this annulus is compact, there is a nite
subcover, which implies that x : f(x) A has measure zero as well.
But then (x : [f(x)[ > ) = 0, contradicting the fact that |f| > .
17

Problem 13: L
p
(R
r
, m) is separable for 1 p < . However, L

(R
n
, m) is
not separable.
Solution. By Math 245A, step functions (i.e. linear combinations of charac-
teristic functions of rectangles) are dense in L
p
. But every rectangle can be
approximated arbitrarily well by rational rectangles, so rational linear com-
binations of rational rectangles are dense in L
p
. There are only countably
many rational rectangles, so there are only countably many Q-linear com-
binations of them. Hence L
p
is separable for 1 p < . However, L

(R
n
)
contains the uncountable family of functions
[0,x]
n : x R which are all
distance 1 away from each other, so it cannot be separable.
Problem 14: If g L

, the operator T dened by Tf = fg is bounded on

L
p
for 1 p . Its operator norm is at most |g|

, with equality if is
seminite.
Solution. Let M = |g|

. Then for 1 p < ,

|Tf|
p
p
=
_
[f[
p
[g[
p
d
_
[f[
p
M
p
d = M
p
|f|
p
p
so |T| M. For p = , |Tf| = sup[fg[ sup[f[ sup[g[ = M supf
where the suprema are essential, so |T| M again. If is seminite, let
E

= x : [g(x)[ < M and choose F

with 0 < (F

< . Let
f =
F
which is in L
p
because F

has nite measure. Then

|Tf|
p
p
=
_
F
[g[
p
(M )
p
(F

) = (M )
p
|f|
p
p
so |Tf| M .
Chapter 6.2, Page 191
Problem 20: Suppose sup
n
|f
n
|
p
< and f
n
f a.e.
(a) If 1 < p < , then f
n
f weakly in L
p
.
(b) The result of (a) is false in general for p = 1. It is, however, true for
p = if is -nite and weak convergence is replaced by weak-*
convergence.
Solution.
(a) Since (L
p
)

= L
q
, we need to show that
_
f
n
g
_
fg for all g L
q
.
For any such g, let > 0. Since [g[
q
L
1
, the absolute continuity
of integration implies the existence of > 0 such that
_
E
[g[
q
< if
(E) < . Now if we let E
0
= [g[
q
1 and E
n
=
1
n+1
[g[
q
<
1
n

for n = 1, 2, . . . , then each E

n
must have nite measure in order for
_
[g[
q
to be nite. Now
_
X
[f[
q
d =
_
E0
[g[
q
d +

n=1
_
En
[g[
q
d,
and since the sum is convergent, N > 0 such that

n=N+1
_
[g[
q
d <
. Let A =
N
n=0
E
n
. Then (A) =

N
n=0
(E
n
) < and
_
X\A
[g[
q
<
18
. By Lusins theorem, there exists a subset B A with (A B) <
and f
n
f uniformly on B. Then
_
[f
n
f[g =
_
A\B
[f
n
f[g +
_
B
[f
n
f[g +
_
X\A
[f
n
f[g.
We use the notation |h|
r,E
= |h
E
|
r
= (
_
E
[h[
r
)
1/r
; note that then
|h|
r,E
|h|
r
which is the usual L
r
norm. Then the rst integral is
at most |f
n
f|
p,A\B
|g|
q,A\B
by Holders inequality. This in turn
is at most 2M
1/q
where M = sup|f
n
|
p
< . The second integral
is at most |f
n
f|
p,A\B
|g|
q,A\B
, which tends to 0 as n since
f
n
f uniformly on B. Finally, the third integral is at most |f
n

f|
p,X\A
|g|
q,X\A
2M
1/q
. Putting the pieces together, we see that
_
[f
n
f[g 0 as n . Hence f
n
f weakly.
(b) Consider f
n
=
[n,n+1]
. Then f
n
0 pointwise, and |f
n
|
p
= 1 is
bounded. However, f
n
, 0 weakly: Consider the function g(x) = 1,
which is in L

= (L
1
)

. Then
_
f
n
g = 1 , 0 =
_
0g. Similarly,
in
1
= L
1
(Z) we can let f
n
=
n
be the nth unit sequence; then
f
n
0 pointwise and |f
n
| = 1 is bounded, but the constant function
g = 1 (
1
)

shows that f
n
, 0 weakly.
Now consider the case p = with -nite. Since L

= (L
1
)

,
weak-* convergence means that
_
gf
n

_
gf for all g L
1
. But

_
[f
n
f[g

_
[f
n
f[[g[ 0
by the Dominated Convergence Theorem, since [f
n
f[[g[ 2M[g[
L
1
where M = sup|f
n
|

< .

Problem 21: If 1 < p < , f

n
f weakly in
p
(A) i sup
n
|f
n
|
p
< and
f
n
f pointwise.
Solution. If sup
n
|f
n
|
p
< and f
n
f pointwise, then f
n
f weakly in

p
(A) by Problem 20. For the converse, suppose f
n
f weakly. WLOG we
may assume f = 0. To show f
n
0 pointwise, let g
y
(x) =
xy
for x, y A.
Then |g|
q
= 1 so g
q
(A), and

x
g
y
(x)f
n
(x) = f
n
(y); by hypothesis,
this tends to 0 as n with y xed. To show |f
n
|
p
is bounded, suppose
it is not. By taking a subsequence if necessary, we may assume |f
n
|
p

. Let f
n1
be such that |f
n1
|
p
> 2, and choose a nite subset A
1
A
such that

xA1
[f
n1
(x)[
p
= M
1
> 1 and

xA\A1
[f
n1
(x)[
p
< . Now
because f
n
0 pointwise and A
1
is nite, f
n
0 uniformly on A
1
,
so

xA1
[f
n
(x)[
p
<

2
for all suciently large n. Take n
2
large enough
that this obtains and that |f
n2
|
p
= M
2
> 4. Let A
2
(A A
1
) be a
nite set with

xA2
[f
n2
(x)[
p
= M
2
> 2 and

xA\A2
[f
n2
(x)[
p
< . We
then take n
3
large enough that

xA1A2
[f
n3
(x)[
p
<

2
and |f
n3
|
p
> 8;
let A
3
(A A
1
A
2
) be nite with

xA3
[f
n3
(x)[
p
= M
3
> 4 and

xA\A3
[f
n3
(x)[
p
< , etc. We now dene g : A C by
g(x) =
_
|fn
k
(x)|
p2
fn
k
(x)
M
k
x A
k
0 x / A
k
for any k
.
19
Then
|g|
q
q
=

k=1

xA
k
[g(x)[
q
=

k=1

xA
k
[f
n
k
(x)[
q(p1)
M
q
k
=

k=1
1
M
q
k

xA
k
[f
n
k
(x)[
p
=

k=1
1
M
q
k
M
k
=

k=1
1
M
q1
k

k=1
1
2
k(q1)
<
since the latter series is geometric with common ratio
1
2
q1
. Hence g

q
(A). Now

xA
f
n
k
(x)g(x) =

xA
k
f
n
k
(x)g(x) +

xA\A
k
f
n
k
(x)g(x)
=

xA
k
[f
n
k
(x)[
p
M
k
+

xA\A
k
f
n
k
(x)g(x)
= 1 +

xA\A
k
f
n
k
(x)g(x).
But

xA\A
k
f
n
k
(x)g(x)

xA\A
k
[f
n
k
(x)g(x)[

_
_

xA\A
k
[f
n
k
(x)[
p
_
_
1/p
_
_

xA\A
k
[g(x)[
q
_
_
1/q

1/p
|g|
q
by Holders inequality. If we choose suciently small that
1/p
<
1
2
, then

xA\A
k
f
n
k
(x)g(x)

<
1
2
, so

xA
f
n
k
(x)g(x)

>
1
2
. This contradicts
the fact that f
n
0 weakly in
p
. Hence |f
n
|
p
must be bounded.
Problem 22: Let X = [0, 1] with Lebesgue measure.
(a) Let f
n
(x) = cos 2nx. Then f
n
0 weakly in L
2
, but f
n
, 0 a.e. or
in measure.
(b) Let f
n
(x) = n
(0,1/n)
. Then f
n
0 a.e. and in measure, but f
n
, 0
weakly in L
p
for any p.
Solution.
20
(a) Since cos 2nx =
e
2inx
+e
2inx
2
, it suces to prove that

f(n) =
_
1
0
f(x)e
2inx
dx 0
as [n[ for f L
2
([0, 1]). But

f
2
by Plancherels theorem, so
certainly

f(n) 0 as [n[ . However, since nx| is dense in [0, 1]
for irrational x, it is in y : cos 2y >
1
2
for innitely many n. Hence
cos 2nx , 0 for irrational x, so it certainly does not converge to 0 a.e.
Moreover, by symmetry, x : cos 2nx
1
2
consists of 4n intervals
of length
1
2n
arccos(
1
2
) (some of these intervals are adjacent to each
other, so they could actually be viewed as 2n 1 intervals of length
1
n
arccos
1
2
and two intervals of length
1
2n
arccos
1
2
, but this doesnt
matter.) Hence (x : [ cos 2nx >
1
2
) is a constant and therefore
does not tend to 0, so cos 2nx , 0 in measure.
(b) Clearly f
n
(x) 0 for all x ,= 0, which is certainly a.e., and (x :
[f
n
(x) ) =
1
n
0 for all > 0, so f
n
0 in measure. However,
let g(x) =
[0,1]
which is in L
p
for all 1 p . Then
_
f
n
(x)g(x) =
1 , 0, which proves that f
n
, 0 weakly in L
p
for 1 p < . In fact
it proves it for p = as well, since g L
1
(L

Chapter 6.3, Page 196

Problem 27 (Hilberts Inequality): The operator Tf(x) =
_

0
(x+y)
1
f(y)dy
satises |Tf|
p
C
p
|f|
p
for 1 < p < , where C
p
=
_

0
x
1/p
(x+1)
1
dx.
Solution. This is an immediate consequence of Theorem 6.20 with K(x, y) =
1
x+y
, which obviously satises K(x, y) =
1
K(x, y). Just for kicks, lets
calculate C
p
with a contour integral: We use the keyhole contour slit along
the positive real axis, with an internal circle of radius and exterior circle
of radius R. Then
_
C
R
dz
z
1/p
(z+1)

2R
R
1/p
(R1)
= O(R
1/p
0 whereas
_
C
dz
z
1/p
(z+1)

2

1/p
(1)
= O(
11/p
) 0 as R and 0, both by
the ML estimate. On the top branch, z
1/p
= x
1/p
so the integral is
I =
_

0
dx
x
1/p
(x + 1)
;
on the bottom branch, z
1/p
= x
1/p
e
2i/p
and the integral is
_

0
dx
e
2i/p
x
1/p
(x + 1)
dx = e
2i/p
I.
The only pole inside the contour is at z = 1 with residue (1)
1/p
=
e
i/p
. By the Residue Theorem,
2ie
i/p
= (1 e
2i/p
)I.
So
I = 2i
e
i/p
(1 e
2i/p
)
=
2i
e
i/p
e
i/p
=

sin/p
.

21
Problem 29: Suppose that 1 p < , r > 0, and h is a nonnegative
measurable function on (0, ). Then
_

0
x
r1
__
x
0
h(y)dy
_
p
dx
_
p
r
_
p
_

0
x
pr1
h(x)
p
dx,
_

0
x
r1
__

x
h(y)dy
_
p
dx
_
p
r
_
p
_

0
x
p+r1
h(x)
p
dx.
Solution. Let K(x, y) = x
1
y

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

h(x) and g(y) = y

h(y), we have
Tf(y) =
_

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

0
x
+1
y

y>x
h(x)dx = y

_
y
0
x
+1
h(x)dx
and
Sg(x) =
_

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

0
x
1
y

y>x
h(y)dy = x
1
_

x
y

h(y)dy.
By Theorem 6.20,
_

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

0
[Tf[
p
p
dy C
p
|f|
p
p
= C
p
_

0
x
p
h(x)
p
dx
and
_

0
x
q(1)
__

x
y

h(y)dy
_
q
dx = |Sg|
q
q
C
q
|g|
q
q
= C
q
_

0
y
q
h(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.
Problem 31 (A Generalized Holder Inequality): Suppose that 1 p
j

and

n
1
p
1
j
= r
1
1. If f
j
L
pj
for j = 1, 2, . . . , n, then

n
1
f
j
L
r
and |

n
1
f
j
|
r

n
1
|f
j
|
pj
.
Solution. By induction on n, it is sucient to consider the case n = 2.
Let g
1
= [f
1
[
r
and g
2
= [f
2
[
r
. Then g
1
L
p1/r
and g
2
L
p2/r
. Since
1
p1/r
+
1
p2/r
= 1, Holders inequality says that g
1
g
2
L
1
and |g
1
g
2
|
1

|g
1
|
p1/r
|g
2
|
p2/r
. Translating these back into the corresponding statements
about f
1
and f
2
, we have f
1
f
2
L
r
and |f
1
f
2
|
r
|f
1
|
p1
|f
2
|
p2
.
Chapter 6.4, Page 199
Problem 38: f L
p
i

2
kp

f
(2
k
) < .
Solution. Let
E
n
= x : 2
n
< [f(x)[ 2
n+1
.
22
Then
f
(2
k
) =

n=k
(E
n
). Now

n=
2
n

En
[f[

n=
2
n+1

En
.
Raising everything to the pth power and integrating,

n=
2
np
(E
n
) |f|
p
p

n=
2
(n+1)p
(E
n
).
The left and right sums dier by a constant factor of 2
p
, so we see that
|f|
p
is nite i

n=
(E
n
)2
(n+1)p
=

n=
(E
n
)
n

k=
2
kp
=

k=
2
kp

n=k
(E
n
)
=

k=
2
kp

f
(2
k
)
is.
Chapter 7.1, Page 215
Problem 4: Let X be an LCH space.
(a) If f C
C
(X, [0, )), then f
1
([a, )) is a compact G

set for all

a > 0.
(b) If K X is a compact G

set, there exists f C

C
(X, [0, 1]) such that
K = f
1
(1).
Solution.
(a) Since f is continuous and [0, ) is closed, f
1
([a, )) is closed. More-
over,
f
1
([a, )) =

_
n=1
f
1
__
a
1
n
,
__
so it is also G

.
(b) Let K =

n=1
G
n
where G
n
is open. By Urysohns lemma, there exist
continuous f
n
: X [0, 1] with f
n
= 1 on K and f
n
= 0 outside G
n
.
Let
f =

n=1
f
n
2
n
.
Then f is continuous, 0 f 1, and f
1
(1) = K. However,
f may not be compactly supported. To x this, let F K be a
compact neighborhood of K, which exists by Proposition 4.31. Let
g : X [0, 1] be continuous with g = 1 on K and g = 0 outside F.
23
Then the product fg : X [0, 1] is continuous, (fg)
1
(1) = K, and
supp(fg) F is compact.

Chapter 7.2, Page 220

Problem 11: Suppose that is a Radon measure on X such that (x) = 0
for all x X, and A B
X
satises 0 < (A) < . Then for any such
that 0 < < (A) there is a Borel set B A such that (B) = .
Solution. Let > 0. By inner regularity, there exists compact K
0
A
such that < (K) (A). Now for each x K, (x) = 0, so by
outer regularity on sets of nite measure, A G
x
x open such that
(G
x
) < . (We may take G
x
to be open in the relative topology, since
the restriction of to A is still Radon.) By compactness, there is a nite
subcover G
x1
, . . . , G
xn
of K. For each j = 1, . . . , n, let G
(j)
= G
x1

G
xj
. Then for some j, < (G
(j)
) < +, because (G
(j)
) < (G
(j1)
)+
and (G
(n)
) (K
0
) > . Let G
0
be equal to the G
(j)
with this property.
Thus G
0
A and < (G
0
) < + . By inner regularity, K
1
G
0
compact with (K
1
) > . Then we can cover K
1
with nitely many open
sets of measure at most /2, which we can also require to be subsets of G
0
;
an appropriate union of these yields G
1
G
0
with < (G
1
) < + /2.
Continuing, we have a nested sequence G
0
G
1
. . . with < (G
j
) <
+

2
j
. Then E = G
j
is a subset of A with measure .
Problem 12: Let X = RR
d
, where R
d
denotes R with the discrete topology.
If f is a function on X, let f
y
(x) = f(x, y); and if E X, let E
y
= x :
(x, y) E.
(a) f C
C
(X) i f
y
C
C
(R) for all y and f
y
= 0 for all but nitely
many y.
(b) Dene a positive linear functional on C
C
(X) by I(f) =

yR
_
f(x, y)dx,
and let be the associated Radon measure on X. Then (E) =
for any E such that E
y
,= for uncountably many y.
(c) Let E = 0 R
d
. Then (E) = but (K) = 0 for all compact
K E.
Solution.
(a) Suppose f C
C
(X). Then for any xed y, f
y
= ef where e : R X
is the continuous embedding x (x, y). So f
y
is continuous. More-
over, supp(f
y
) (supp(f)) where : X R is dened by (x, y) =
x. So supp(f
y
) is a closed subset of the continuous image of a compact
set, hence compact. Thus, f
y
C
C
(R) for all y. Suppose there are
inninitely many pairs (x

, y

) with y

distinct and f(x

, y

) ,= 0.
Consider the open cover R y : y R
d
of supp(f). Any nite
subcover will only include nitely many of the (x

, y

) and hence can-

not cover the whole set, contradicting its compactness. Hence f
y
= 0
for all but nitely many y.
24
Conversely, suppose f
y
= 0 for all but nitely many y and f
y
C
C
(R)
for all y. Then
supp(f) =
_
{y:f
y
=0}
supp(f
y
) y
is a nite union of compact sets, hence compact. Also, for any open
subset U of the range,
f
1
(U) =
_
yR
d
(f
y
)
1
(U)
is a union of open sets and therefore open. Thus, f C
C
(X).
(b) Let E be a set such that E
y
,= for uncountably many y, and let G
be any open set with E G. Then G contains an interval I
y
y for
each y with E
y
,= , since these intervals are a basis for the topology of
RR
d
. Since there are uncountably many intervals, there must exist
N such that uncountably many of the intervals have length greater
than
1
N
. For each such interval I
y
, there exists a function f I
y
with
_
Iy
f >
1
2N
(e.g. take a subinterval of length at least
1
2N
and use
Urysohns lemma to construct a function which is 1 on this subinterval
and 0 outside I
y
), so
(G) =

yR
_
f(x, y)dx

1
2N
= .
By regularity, since (G) = for any open set G containing E, then
(E) = .
(c) By part (b), (E) = . However, if K E is compact, then K =
0 y
1
, . . . , y
n
for some nite subset y
1
, . . . , y
n
R
d
. Then, given
> 0, let f

(x) : R R be a continuous function with values in [0, 1]

which is 1 at the origin and 0 for [x[ > . Then
g

(x, y) =
_
f

(x) y = y
1
, . . . , y
n
0 else
is in C
C
(X), g

K
, and
I(g

) =
n

i=1
_
f

(x)dx 2n.
Thus,
(K) = infI(g) : g C
C
(X), g
K
= 0.

Chapter 7.3, Page 224

Problem 20: Some examples of nonreexivity of C
0
(X):
(a) If M(X), let () =

xX
(x). This sum is well dened,
and M(X)

. If there exists a nonzero M(X) such that

(x) = 0 for all x X, then is not in the image of C
0
(x) in
M(X)

C
0
(X)

.
25
(b) At the other extreme, let X = N with the discrete topology; then
C
0
(X)

1
and (
1
)

.
Solution.
(a) The sum () is well-dened because it is absolutely convergent: If
is a nite positive measure,

(x) (X) < because all nite

sums are bounded by (X) : (x
1
)+ +(x
n
) = (x
1
, . . . , x
n
)
(X). For general , the sum is absolutely convergent because the
sums of the positive real terms, negative real terms, positive imagi-
nary terms, and negative imaginary terms are all nite. The linearity
of is cleary, so M(X)

. Now suppose there exist as specied

in the problem. Given f C
0
(X), let

f denote its image in C
0
(X)

.
If
_
fd ,= 0 then clearly

f ,= since () = 0. If
_
fd = 0, let
x
f
X be a point at which [f(x
f
)[ < 1. (Such a point must exist
since f 0 at .) Let
f
= +
x
f
where
x
f
(E) = 1 if x
f
E and
0 otherwise. Then

f(
f
) =
_
fd
f
=
_
fd +
_
fd
x
f
= f(x
f
)
whereas
(
f
) =

(x) +

x
f
(x) = 1
and these are not equal because [f(x
f
)[ < 1. Hence

f ,= for any
f C
0
(X).
(b) In this case all functions (sequences) are continuous and all compact
sets are nite, so C
0
(X) is the set of sequences which tend to zero.
Now C
0
(X)

is the set of nite complex measures on N; because N

is countable, a measure on Z is dened by its values on each point,
so every such measure corresponds to a sequence dened by a
n
=
(n). The measure is then nite i

[(n)[ =

[a
n
[ < ,
i.e. a
n

1
. Thus, C
0
(X)

1
. To show that (
1
)

, we
need to show that the inclusion

(
1
)

given by f
f
, where

f
(g) =

n
f(n)g(n), is bijective and norm-preserving. If |f|

= M,
then for any g
1
with |g| = 1,
[
f
(g)[ =

n=1
f(n)g(n)

n=1
[f(n)[[g(n)[

n=1
M[g(n)[ = M|g|
1
= M.
On the other hand, given > 0, choose N such that [f(N)[ > M ,
and let
N
be the sequence given by
N
(n) = 1 for n = N and 0
else. Then |
N
|
1
= 1 and [
f
(
N
)[ = [f(N)[ > M . Together,
these imply that |
f
| = |f|. This in turn implies that f
f
is
injective. To show its surjective, let T :
1
C be a continuous linear
functional. Because the (nite) linear combinations of the
n
are dense
in
1
(every sequence is the limit of its truncations), T is dened by its
values on
n
. This gives a correspondence between T and the sequence
a
n
= T(
n
). This sequence is in

because |
n
|
1
= 1 so [T(
n
)[ is
bounded. Hence (
1
)

. Obviously there are bounded sequences

which do not tend to zero, so C
0
(N) is not reexive.

26
Problem 22: A sequence f
n
in C
0
(X) converges weakly to f C
0
(X) i
sup|f
n
|
u
< and f
n
f pointwise.
Solution. Suppose sup|f
n
| = M < and f
n
f pointwise. Let > 0.
By Egorovs theorem, E X with (E) <

4M
and f
n
f uniformly on
X E. Then for any C
0
(X)

= M(X),
_
X
(f
n
f)d =
_
E
(f
n
f)d +
_
X\E
f
n
fd
where the rst integral is at most /2 since |f
n
f|

2M and (E) <

4M
, and the second integral tends to zero as n because the conver-
gence is uniform and (X) < . Hence, [
_
X
(f
n
f)d[ is eventually less
than for all , so it tends to zero, i.e.
_
f
n
d
_
fd. Thus f
n
f
weakly. Conversely, suppose f
n
f weakly. Then f
n
f pointwise; if
not, let x be a point at which f
n
(x) , f(x), and let
x
be the unit point
measure centered at x. This is a nite Radon measure, as is easy to verify.
Then
_
f
n
d
x
= f
n
(x) , f(x) =
_
fd
x
, a contradiction. To show that
sup|f
n
|

< , we use the uniform boundedness principle: C

0
(X) is a Ba-
nach space, and since f
n
f pointwise, [f
n
(x) f(x)[ is bounded for each
x. The UBP then implies that sup|f
n
f| < sup|f
n
| < .
Chapter 8.2, Page 245
Problem 8: Suppose that f L
p
(R). If there exists h L
p
(R) such that
lim
y0
|y
1
(t
y
f f) h|
p
= 0,
we call h the (strong) L
p
derivative of f. If f L
p
(R
n
), L
p
partial
derivatives of f are dened similarly. Suppose that p and q are conjugate
exponents, f L
p
, g L
q
, and the L
p
derivative
j
f exists. Then
j
(f g)
exists (in the ordinary sense) and equals (
j
f) g.
Solution. By Proposition 8.8,
_
_
_
_
_
1
y
_

yej
f f
_
h
_
g
_
_
_
_
u

_
_
_
_
1
y
_

yej
f f
_
h
_
_
_
_
p
|g|
q
0
as y 0. (Here e
j
is the jth unit coordinate vector in R
n
.) Thus,

j
(f g)(x) = lim
y0
f g(x +ye
j
) f g(x)
y
= lim
y0
_

yej
f f
y
g
_
(x) = hg(x).

Chapter 9.1, Page 289

Problem 6: If f is absolutely continuous on compact subsets of an interval
U R, the distribution derivative f

(U) coincides with the pointwise

(a.e.-dened) derivative of f.
Solution. Since the product of absolutely continuous functions is absolutely
continuous (this was an exercise in 245A), f is absolutely continuous on
compact subsets of U for any C

C
(U). Then, because supp() U,
0 = f

U
=
_
U
d
dx
((x)f(x)) dx =
_
U

(x)f(x)dx +
_
U
f

(x)(x)dx.
27
Thus,
f

, ) =
_
U
f

(x)(x)dx =
_
U
f(x)

(x)dx,
that is, the distribution derivative coincides with the ordinary derivative.

Problem 7: Suppose f L
1
loc
(R). Then the distribution derivative f

is a
complex measure on R i f agrees a.e. with a function F NBV , in which
case f

, ) =
_
dF.
Solution. Suppose f = F NBV a.e. Using Theorem 3.36,
f

, ) =
_
f

dx =
_
F

dx =
_
dF
so the distribution derivative of f is the complex measure dF. On the other
hand, suppose there is a complex measure such that
_
f

=
_
d.
By Theorem 3.29, d = dF for a function F NBV , and by Theorem 3.36
again,

_
F

dx =
_
F

dx =
_
dF =
_
d =
_
f

dx
from which it follows that f

= F a.e.
Problem 8: Suppose f L
p
(R
n
). If the strong L
p
derivatives
j
f exists
in the sense of Exercise 8 in Chapter 8.2, they coincide with the partial
derivatives of f in the sense of distributions.
Solution. The partial derivative in the sense of distributions is

_
f
j
=
_
f lim
y0

yej

y
= lim
y0
_
f

yej

y
= lim
y0
_
(
yej
f) f
y
= lim
u0
_

uej
f f
u
.
Here e
j
is the jth unit coordinate vector in R
n
. We have made the substi-
tution u = y in the last line. The equality
_
f
yej
=
_
(
yej
f) follows
from the translation invariance of Lebesgue measure. Also note that the
limit can be moved outside the integral on the second line by the Domi-
nated Convergence Theorem because
ye
j
(x)(x)
y
=
j
(x

) for some x

between x and x+y, by the Mean Value Theorem; but

j
is bounded and
compactly supported, so
ye
j
(x)(x)
y
L
q
and
_
_
_
_
f

yej

y
_
_
_
_
1
|f|
p
|

yej
(x) (x)
y
|
q
< .
Now
ue
j
ff
u
h in L
p
= (L
q
)

, and L
q
, so
_
ue
j
ff
u

_
h and

_
f
j
= lim
u0
_

uej
f f
u
=
_
h = h, ),
28
that is, the distribution derivatives coincide with the strong L
p
derivatives.