Professional Documents
Culture Documents
PreliminarySolutions Ucdavis Analysis
PreliminarySolutions Ucdavis Analysis
n
(x)[ 1 for all
n, x. Show that if g(x) is a function such that
lim
x
f
n
(x) = g(x)
then g(x) is a continuous function.
Proof:
Let x
0
R. By the mean value theorem and the bound on f
n
we see
[f
n
(x) f
n
(x
0
)[ sup
x
[f
n
(x)[ [x x
0
[ [x x
0
[
Now let > 0 and =
3
. For any x satisfying [x x
0
[ we know [f
n
(x)
f
n
(x
0
)[
3
from the above. The pointwise convergence of f
n
to g implies that
there exist N N such that
[f
n
(x) g(x)[
3
[f
n
(x
0
) g(x
0
)[
3
for all n N. By adding and subtracting terms and using the triangle inequality
we nd
[g(x) g(x
0
)[ [g(x) f
N
(x)[ +[f
N
(x) f
N
(x
0
)[ +[f
N
(x
0
) g(x
0
)[
which shows that g is continuous. 2
19
3.0.12 Problem 7
Problem 7
(a) State the Stone-Weierstrass theorem in the context of C(X, R), where X
is a compact Hausdor space.
(b) State the Radon-Nikodym theorem, as it applis to a pair of -nite mea-
sures and dened on a measurable space (X, /).
(c) State the denitions of the terms normal topological space and absolutely
continuous function.
Proof:
Obtained by opening the right book and turning to the right page.
2
20
3.0.13 Problem 8
Problem 8
Let f : X Y be a mapping between topological spaces X and Y . Let c be a
base for the topology of Y . Show that if f
1
(E) is open in X for each E c,
then f is continuous.
Proof:
Let V Y be open. We must show that f
1
(V ) X is open. Since c is a base
for the topology of Y we can write
V =
_
_
=
_
f
1
(E
)
By hypothesis every f
1
(E
) := lim
xa
F(x)].
Proof:
To show the former consider the sequence of nested sets
(a 1, a] (a
1
2
, a] (a
1
3
, a]
Since is a measure and (a 1, a] < we have
(a) =
_
n=1
(a
1
n
, a]
_
= lim
n
(a
1
n
, a]
= lim
n
F(a) F(a
1
n
)
= F(a) F(a
)
which is what we wanted. Note that the latter limit F(a
) exists since F is
increasing and F(a) < .
To show the latter we observe that [a, b] = a
)
2
22
3.0.15 Problem 10
Problem 10
Let f be a B
[0,1]
2-measurable real-valued function such that the partial deriva-
tive
f
t
(x, t) exists for each (x, t) [0, 1]
2
and M := sup
f
t
(x, t)
< . Prove
that
f
t
(x, t) is measurable, and that for all t [0, 1],
d
dt
_
1
0
f(x, t) dx =
_
1
0
f
t
(x, t) dx
Proof:
First we show that
f
t
(x, t) is measurable as a function of x. By denition,
f
t
(x, t) = lim
t
t
f(x, t
) f(x, t)
t
t
Since f is B
[0,1]
2-measurable its restriction f
t
(x) = f(x, t) is B
[0,1]
-measurable.
The sum and product of measurable functions is a measurable function so the
quotients
q
t
(x) =
f(x, t
) f(x, t)
t
t
=
f
t
(x) f
t
(x)
t
t
are measurable. Furthermore the pointwise limit of measurable functions is
measurable so
f
t
(x, t) = lim
t
t
q
t
(x)
is B
[0,1]
-measurable as a function of x. To show the equality we rst rewrite the
left hand side as a limit
d
dt
_
1
0
f(x, t) dx = lim
n
_
1
0
n
_
f
_
x, t +
1
n
_
f(x, t)
_
dx
By the mean value theorem we have
n
_
f
_
x, t +
1
n
_
f(x, t)
_
n
_
f
t
(x, t
)
1
n
_
M
where t < t
< t +
1
n
. The last inequality follows from the hypothesized bound
on
f
t
. Note that the constant function g(x) M is in L
1
[0, 1]. By applying
Lebesgues dominated convergence theorem and using the mean value theorem
again we conclude
d
dt
_
1
0
f(x, t) dx =
_
1
0
lim
n
n
_
f
_
x, t +
1
n
_
f(x, t)
_
dx
=
_
1
0
f
t
(x, t) dx
2
23
3.0.16 Problem 11
Problem 11
Let m denote the Lebesgue measure on R, x f L
1
(R, B
R
, m), and dene a
function G : R R by the formula
G(t) =
_
R
f(x +t) dm(x)
Prove that G is a continuous function.
Proof:
First note that since f L
1
the integral
_
R
f(x) dm(x)
is well-dened and nite. The Lebesgue measure m is translation invariant,
which implies that
_
R
f(x +t) dm(x) =
_
R
f(x) dm(x)
for all t R. Therefore G(t) is constant, and thus continuous.
2
24
Chapter 4
Winter 2002
Author: Adam Sorkin
4.0.17 Problem 1
Statement: Problem Number 1
Show that the
2
norm is indeed a norm.
Proof:
Recall that
2
(N) is the space of all square summable sequences of complex
numbers. Letting a
i
denote one such sequence, let us temporarily denote
f(a
i
) =
i
[a
i
[
2
.
To show that f is indeed a norm, we must have that f is positive denite,
homogeneous, and satises the triangle inequality. Positive deniteness follows
immediately from the denition, and similarly homogeneity. As usual, only the
third takes any real work. To prove that f(a
i
+ b
i
) f(a
i
) + f(b
i
), we begin
with the simple observation that
0 ([a
i
b
j
[ [a
j
b
i
[)
2
.
Expanding this out and summing over i ,= j gives
i=j
[a
i
a
j
b
i
b
j
[
i=j
[a
i
[
2
[b
j
[
2
If we add to both sides the quantity
i
[a
i
b
i
[
2
and rearrange slightly, we get
_
i
[a
i
b
i
[
_
2
i
[a
i
[
2
_
_
_
j
[b
j
[
2
_
_
= f(a
i
)
2
f(b
i
)
2
.
25
Taking roots gives
i
[a
i
b
i
[ f(a
i
)f(b
i
), and using the triangle inequality on
complex numbers gives
f(a
i
+b
i
)
2
=
i
[a
i
+b
i
[
2
i
_
[a
i
[
2
+[b
i
[
2
+ 2[a
i
b
i
[
_
f(a
i
)
2
+f(b
i
)
2
+ 2f(a
i
)f(b
i
)
(f(a
i
) +f(b
i
))
2
Taking roots gives the desired inequality. 2
26
4.0.18 Problem 2
Statement: Problem Number 2
Prove that C([0, 1]), the space of continuous functions on [0, 1], is not complete
in the L
1
metric: (f, g) =
_
[f(x) g(x)[dx.
Proof:
Consider the sequence of functions f
n
: [0, 1] R dened by f
n
(x) = x
n
. Clearly
these functions are continuous and integrable, hence in C([0, 1]). Moreover, f
n
is Cauchy, for we can compute |f
n
| = 1/(n + 1).
Now f
n
has no limit in C([0, 1]), for any L
1
-limit of f
n
must be a pointwise limit
of f
n
. But f
n
converges pointwise to a discontinuous function; specically its
pointwise limit is zero on [0, 1) and 1 at 1. hence we see 2
27
4.0.19 Problem 3
Statement: Problem Number 3
Let C([0, 1]) be the space of continuous function son the nit interval with the
uniform norm and let R[x] be the subspace of polynomials. Give an example
of an unbounded linear transformation T : R[x] R.
Proof:
Let T : R[x] R be dened by evaluation at r, for some r > 1. That is, for
some p R[x], T(p) = p(r). It is clear the evaluation mapping is linear, and
we can bound its norm below using the polynomial x
n
. Notice x
n
has norm 1
in the uniform norm, and so |T| |Tx| = r
n
. Taking n large shows T has
unbounded norm. 2
28
4.0.20 Problem 4
Statement: Problem Number 4
Let X be a metric space. Prove or disprove the following:
(a) If X is compact, then X is complete.
(b) If X is complete, then X is compact.
Proof:
The rst statement is true; the second false. To see compactness implies com-
plete, we use the sequential characterization of compactness. Now let x
n
be a
Cauchy sequence in X. Then by compactness, x
n
contains a subsequence x
k(n)
which converges to some x
0
in X. Hence x
n
x
0
, for
[x
n
x
0
[ [x
n
x
k(n)
[ +[x
k(n)
x
0
[.
Therefore x
n
has a limit in X, and so X is complete. Now complete does
not imply compact in a metric space. A familiar counterexample is the real
numbers, which are certainly not compact, and are complete (though we do not
prove this). 2
29
4.0.21 Problem 5
Statement: Problem Number 5
Let H be a Hilbert space and V a linear subspace. Show that V
= V .
Proof:
Recall that V
= (V )
= V V
V
= V
. 2
30
4.0.22 Problem 6
Statement: Problem Number 6
State Jensens inequality. Show that the function : x log(1/x) is convex.
Suppose that a
i
0 and p
i
> 0 such that the p
i
sum to 1. Prove that
a
p
1
1
a
p
n
n
a
1
p
1
+ +a
n
p
n
.
Proof:
Jensens inequality is as follows. Let (X, ) be a nite measure space, with
=
_
X
1 < . Let f : X R be integrable, and : R R a convex
function. Then
(
1
_
X
fd)
1
_
X
fd
To see that : (0, ) R as dened above is convex, notice it is twice
dierentiable. Then as
= 1/x
2
> 0, it is a convex function.
Finally, consider the measure space X = p
i
with measure
_
X
fd =
f(p
i
)p
i
. Let be as dened above, and notice the function f : X R
dened by f(p
i
) = a
i
is integrable. From Jensens inequality we get
log(
i
a
i
p
i
)
i
p
i
log a
i
.
Exponentiating both sides gives the desired inequality. 2
31
Chapter 5
Winter 2005
Author: Jerey Anderson
5.0.23 Problem 1
Statement: Problem 1
Show that the mapping T : R R dened by
T(x) =
2
+x arctan(x)
has no xed points in R and that
[T(x) T(y)[ [x y[ for all distinct x, y R
Why does this example not contradict the contraction mapping theorem?
Proof:
Suppose hoping for contradiction that there exists a xed point for the map
T : R R. Then, by denition of xed point, there is some x R such that
T(x) = x
T(x) =
2
+x arctan(x) = x
2
arctan(x) = x x
arctan(x) =
2
Then, under our assumption, we have that their exists some x R such that
tan(
2
) = x
This is not possible, since tan is not dened at
2
.
Now, let us consider
[T(x) T(y)[
32
for any two distinct x, y R. Without loss of generality assume that x < y.
We know by the mean value theorem, we have there exists some z (x, y) such
that
|T(x) T(y)| = T
(z)[x y[
Then, we also have that for any x R
T
(x) =
d
dx
_
2
+x arctan(x)
_
= 0 + 1
1
1 +x
2
We can then bound T
(z)[x y[
< 1[x y[
= [x y[
This proves the second part of our claim.
To substantiate the last claim of this problem, consider the statement of
the Contraction mapping theorem. In the mapping dened above, there is no
c (0, 1) such that
[T(x) T(y)[ < c[x y[
for all x, y R.
Assume, hoping for contradiction, that the exists a c (0, 1) such that
[T(x) T(y)[ < c[x y[
(ie assume that the antecedents of the contraction mapping theorem hold). We
know by the Mean Value Theorem taught in Math 21A here at UC Davis, for
all x, y R, there exists some R such that
[T(x) T(y)[ = T
()[x y[
= 1
1
1 +
2
. .
T
()
[x y[
We will explicitly exhibit a pair (x, y) which dees this inequality, thus leading
to a contradiction and proving that no such c can exist. Let y = 0. Then we
have
[T(x) T(0)[ =
T(x)
2
2
+x arctan(x)
2
= [x arctan(x)[
33
We note that for x > 0 we have x arctan(x) > x
2
by the properties of
arctan(x). Now choose
x =
2(1 c)
We note that
2(1c)
> 0/2. Then we have
[T(x) T(0)[ = [x arctan(x)[
> x
2
=
2(1 c)
2
=
c)
2(1 c)
= cx
Then, given we have a c, we have found a pair (x, y) of points that dees the
bound. This contradicts the assumption that our function T above satises the
contraction mapping theorem. With this, we have shown each of the three parts
of this problem. 2
34
5.0.24 Problem 2
Statement: Problem 2
Prove that the vector space C([a, b]) is seperable. Here and below, C([a, b])
is the vector space of continuous functions f : [a, b] R with the supremum
norm.
Proof:
This is exactly what we wanted to show. 2
35
5.0.25 Problem 3
Statement: Problem 3
Suppose that f
n
C([a, b]) is a sequence of functions converging uniformly to
a function f. Show that
lim
n
b
_
a
f
n
(x)dx =
b
_
a
f(x)dx
Give a counterexample to show that the pointwise convergence of continuous
functions f
n
to a continuous function f does not imply convergence of the
corresponding integrals.
Proof:
Let f
n
C([a, b]) be a sequence of functions converging uniformly to a function
f. Let > 0. Since f
n
converges uniformly, we have that there is an N N
such that for all n N
[f
n
(x) f(x)[ <
b a
for all x [a, b]. Assuming n N, we have
b
_
a
f
n
(x)dx
b
_
a
f(x)dx
b
_
a
(f
n
(x) f(x))dx
b
_
a
[f
n
(x) f(x)[dx bringing absolute value inside integral
<
b
_
a
(b a)dx by assumption on n
=
b a
b
_
a
1dx
=
b a
(b a)
=
This proves the rst part of our problem. For the second part of this problem,
consider the sequence f
n
C([0, 1]) which linearly interpolates the following
secquence of points
for x (0, 1/2
n
), f
n
is dened by the line through the points (0, 0) and
(1/2
n
, 2
n
)
for x (1/2
n
, 1/2
n1
), f
n
is dened by the line through the points
(1/2
n
, 2
n
) and (1/2
n1
, 0)
36
for x (1/2
n1
, 0), f
n
is dened by the constant function 0.
To explicitly formulate this sequence of functions, we rely on the following:
Given two points (x
1
, y
1
) and (x
2
, y
2
) we can dene the interpolating line by
computing
m =
y
2
y
1
x
2
x
1
(y y
1
) = m(x x
1
)
Then, this sequence of function is dened by
f
n
(x) =
_
_
_
2
2n
x if 0 x < 2
n
2
2n
(x 1/2
n
) + 2
n
if 2
n
x < 2
(n1)
0 if 2
(n1)
x 1
We notice that this sequence of functions converges pointwise to zero, yet the
integral of each f
n
is one. This demonstrates exactly what we would like. 2
37
5.0.26 Problem 4
Statement: Problem 4
Let l
2
(Z) denote the complex Hilbert space of sequences x
n
C, n Z such
that
n=
[x
n
[
2
<
Dene the shift operator S : l
2
(Z) l
2
(Z) by
S((x
n
)) = (x
n+1
)
Show that S has no eigenvalues.
Proof:
Let x
n
2
(Z) as described above. Assume, hoping for contradiction, that
the shift operator does have an eigenvalue C. By denition we have the
following equalities:
S((x
0
)) = x
0+1
= x
1
= x
0
S((x
1
)) = x
1+1
= x
2
= x
1
=
2
x
0
S((x
2
)) = x
2+1
= x
3
= x
2
=
3
x
0
With this initial observation, we see that the assumption that C is an eigen-
value immediately mandates a very unique structure to the sequence x
n
n=
.
Specically, assume that n is a positive integer. Then
x
n
=
n
x
0
Similarly if m is a negative integer, then
x
m
=
m
x
0
Without loss of generality, we can assume [x
0
[ = 1 (if not, we know that the
sequence convergence and hence we can use the constant multiple rule for con-
verging sequences taught in Math 21C here at UC Davis and normalize to enforce
this condition).
Then we have
[x
n
[
2
=
[[
2n
The sequence x
n
n=
illustrates that = 0 is can never be an eigenvalue.
Similarly, if 0 < [[ < 1 we have that the sum
0
[[
2n
38
diverges. Similarly, if [[ > 1, we know that
0
[[
2n
diverges because it fails the test for convergence. Last, if = 1, we have that the
sum of all ones diverges because it fails the convergence test. We have reached
a contradiction for any C. It must be that S has no eigenvalues.s 2
39
5.0.27 Problem 5
Statement: Problem 5
Consider the initial value problem
u
(t) = [u(t)[
, u(0) = 0
Show that the solution to this problem is unique if > 1 and not unique if
0 < 1
Proof (exercise 2.13 in Applied Analysis):
Suppose that > 1. In this case we reference theorem 2.26 of Applied Analysis.
We must check that the antecedents of this theorem are preserved:
In this case, let us dene
f(t, u) = u
(t) = [u(t)[
n=1
H and
x 1 such that
i. there exists M > 0 such that |x
n
| M for all n
ii. lim
n
x
n
, y = x, y for all y 1
0
Prove that x
n
n=1
converges in the weak topology of 1
Proof: Theorem 8.40 in Applied Analysis
Let all the assumptions in the problem statement hold. Let y 1. Let > 0.
We want to show that for that
lim
n
x
n
, y = x, y
We can rewrite this in the form
lim
n
x
n
x, y = 0
using the denition of the inner product. We will prove our statement in this
form.
Since y 1 and 1
0
is dense in 1, we have that there is some 1
0
such that | y| < (this is an immediate consequence of the denition of
density). Since 1
0
, we know by our assumptions above
lim
n
x
n
, = x,
We can rearrange this using the same ideas as above to see
lim
n
x
n
x, = 0 (5.1)
We see that there is some postive integer N such that
[ x
n
x, [ <
for all n > N by the denition of weak convergence.
Now let us consider
[x
n
x, y[ = [x
n
x, y +[
= [x
n
x, y [ +[x
n
x, [
[x
n
x, y [ + using the assumption (5.1)
|x
n
x|| y| + applying Cauchy Schwarz Inequality
(|x
n
| +|x|)| y| + by the triangle inequaltiy
(M +|x|)| y| + since x
n
is bounded
(M +|x|) + by choice of
This is exactly what we wanted to show. 2
41
Chapter 6
Fall 2007
6.0.29 Problem 1
Author: David Renfrew
Statement: Problem 1
Suppose that f : [0, 1] R is continuous. Prove that
lim
n
_
1
0
f(x
n
)dx
exists and evaluate the limit. Does the limit alway exist if f is only assumed
to be integrable?
Proof:
Let g(x) := sup
x
f(x), because f is continuous this exist and g is integrable
with f(x) g(x) so we can apply the LDCT:
lim
n
_
1
0
f(x
n
)dx =
_
1
0
lim
n
f(x
n
)dx =
_
1
0
f( lim
n
x
n
)dx = f(0)
The limit does not necessiarly exist if f is only integrable.
Consider f(x) = x
1/2
which is integrable but
_
1
0
f(x
n
)dx =
_
1
0
x
n/2
dx =
_
1
1 n/2
_
x
1n/2
1
0
This integral does not converge for n > 1.
This establishes our claim 2
42
6.0.30 Problem 2
Author: David Renfrew
6.0.31 Problem 1
Author: David Renfrew
Statement: Problem 2
Suppose that for each n Z, we are given a real number
n
. For each t R,
dene a linear operator T(t) on 2-periodic function by
T(t)
_
nZ
f
n
e
inx
_
=
nZ
e
i
n
t
f
n
e
inx
,
where f(x) =
nZ
f
n
e
inx
with f
n
C.
(a) Show that T(t) : L
2
(T) L
2
(T) is a unitary map.
(b) Show that T(s)T(t) = T(s +t) for all s, t R.
(c) Prove that if f C
(T).
Proof:
(a) First observe by the density of the trig polynomials T(t) is onto.
Then let f, g L
2
(T).
Then compute the inner product using the Fourier coecients.
T(t)f, T(t)g =
nZ
_
e
i
n
t
f
n
_
_
e
i
n
t
g
n
_
=
nZ
f
n
g
n
= f, g
(b) By direct computation:
T(s)T(t)f = T(s)
nZ
e
i
n
t
f
n
e
inx
=
nZ
e
i
n
s
e
i
n
t
f
n
e
inx
=
nZ
e
i
n
s+t
f
n
e
inx
= T(s +t)f
(c) This follows by Sobolev embedding.
43
6.0.32 Problem 3
Author: David Renfrew
Statement: Problem 3
Let (X, | |
X
), (Y, | |
Y
), (Z, | |
Z
) be Banach spaces, with X compactly
imbedded in Y , and Y continuously imbedded in Z (meaning: X Y Z;
bounded sets in (X, | |
X
) are precompact in (Y, | |
Y
); and there is a constant
M such that |x|
Z
M|x|
Y
for every x Y ). Prove that for every > 0
there exists a constant C() such that
|x|
Y
|x|
X
+C()|x|
Z
for every x X.
Proof:
Assume for contradiction that there is an for which such a C() does not exist.
Then there exist a sequence x
n
with X-norm 1 such that
|x
n
|
Y
> +n|x
n
|
Z
By the compact embedding the x
n
s have a convergent subsequence in the Y
norm, let us pass to this subsequence.
This means |x
n
|
Y
will converge to a constant, but this means the above in-
equality will only be preserved if |x
n
|
Z
goes to zero.
This implies x
n
0 in all spaces. This contradicts the assumption that
|x
n
|
Y
> , so the desired inequality is proved.
44
6.0.33 Problem 4
Author: David Renfrew
Statement: Problem 4
Let 1 be the weighted L
2
-space
1 =
_
f : R C
_
R
e
|x|
[f(x)[
2
dx <
_
with inner product
f, g =
_
R
e
|x|
f(x)g(x)dx.
Let T : 1 1 be the translation operator
(Tf)(x) = f(x + 1).
Compute the adjoint T
n=1
is a bounded seqeunce in H
1
() for R
3
open,
bounded, and smooth. Show that there exists an f H
1
() such that for a
subsequence f
n
l
l=1
,
f
n
l
Df
n
l
fDf weakly in L
2
(),
where D =
_
x
1
,
x
2
,
x
3
_
denotes the weak gradient operator.
Proof:
(a) Let R
n
be an open, bounded Lipschitz domain. Let 1 p n and
p
:=
np
np
.
Then W
1,p
() is compacted embedded in L
q
() for all 1 q < p
.
(b)
Since f
n
is bounded in H
1
so f
n
and Df
n
are bounded in L
2
.
Bounded sets are weakly precompact so after passing to a subsequence there
exist an f and g such that f
n
f and Df
n
g. Additionally g = Df.
f
n
Df
n
= (f
n
f)Df
n
+fDf
n
The Sobolev conjugate p =
np
np
=
32
32
= 6 > 2, so we can apply Rellichs
Theorem, after passing to a subseqence, to say that f
n
l
f strongly in L
2
.
Since Df
n
are bounded (f
n
f)Df
n
converges to 0 in L
2
. So we have:
lim
l
f
n
l
Df
n
l
= lim
l
(f
n
l
f)Df
n
l
+fDf
n
l
0 +fDf
As desired.
46
Author: David Renfrew
Statement: Problem 6
Let := B(0,
1
2
) R
2
denote the open ball of radius
1
2
. For x = (x
1
, x
2
) ,
let
u(x
1
, x
2
) = x
1
x
2
[log ([ log([x[)[) log log 2] where [x[ =
_
x
2
1
+x
2
2
.
(a) Show that u C
1
().
(b) Show that
2
u
x
2
j
C(). for j = 1, 2, but that u , C
2
().
(c) Using the elliptic regularity theorem for the Dirichlet problem on the disc,
show that u H
2
().
Proof:
First we compute the derivaties:
x
1
u(x
1
, x
2
) = x
2
[log ([ log([x[)[) log log 2] +
x
1
x
2
[ log([x[)[
log([x[)
[ log([x[)[
1
[x[
x
1
[x[
x
2
u(x
1
, x
2
) = x
1
[log ([ log([x[)[) log log 2] +
x
1
x
2
[ log([x[)[
log([x[)
[ log([x[)[
1
[x[
x
2
[x[
47
Chapter 7
Winter 2008
7.0.35 Problem 1
Author: Jerey Anderson
Statement: Problem Number
Dene f
n
: [0, 1] R by
f
n
(x) = (1)
n
x
n
(1 x)
a. Show that
n=0
f
n
converges uniformly on [0, 1]
b. Show that
n=0
[f
n
[ converges pointwise on [0, 1] but not uniformly
Proof:
48
7.0.36 Problem 2
Author: Jerey Anderson
49
7.0.37 Problem 3
Author: Jerey Anderson
Statement: Problem 3
Suppose that / is a (nonzero) closed linear subspace of a Hilbert space 1 and
: / C is a bounded linear functional on /. Prove that there is a unique
extension of to a bounded linear function on 1 with the same norm.
Proof:
The rst thing to notice is that this problem is known as the Hahn-Banach
Theorem for Closed linear subspaces. With this in mind, the way to establish
this result is:
construct the norm preserving extension explicitly to the larger subspace
Invoke the axiom of choice to establish that the construction above gives
our desired extension
Then, let 1 be a Hilbert space, let / be a closed linear subspace of 1, let
: / C be a bounded linear functional on /. We want to show
i. there exists a norm preserving extension of
ii. this extension is unique
First we will show that an extension exists. Let x
0
1 such that x
0
/ /,
Let
Y
1
= x
0
, / = spanx
0
, /
We know that for any y Y
1
, we have y = x + x
0
for x M, C. Let us
dene a mapping : Y
1
C by
(y) = (x +x
0
)
= (x) +
for some C.
We can check that this mapping is linear:
(y +z) = ((x +x
0
) + (w +w
0
))
= (x +w) + ( +)
= (x) + +(w) +
= (y) + (z)
The trick of this problem is to choose so that the extended funcitonal has
norm 1.
Recall by denition
||
op
= sup
y=1
(y)|
H
= sup
x+x
0
=1
|(x) +|
C
50
We can bound ||
op
above by one using the inequalities
|(y)|
op
= |(x) +|
C
|x +x
0
|
H
= |y|
H
Now, to nd the lower bound, take x = x /. Then
[(x) +[ = [ (x) +[
| x +x
0
|
H
This is true if and only if
[[[(x) [ [[|x x
0
|
H
[(x) [ |x x
0
|
H
From here, we would like to solve for such that
|x x
0
|
H
(x) |x x
0
|
H
(x) |x x
0
|
H
(x) +|x x
0
|
H
[(x) |x x
0
|, (x) |x x
0
|]
We introduce the notation
Ax = (x) |x x
0
|
Bx = (x) +|x x
0
|
We also let = [Ax, Bx] R. We note that ,= Ax Bx for all
x, y /. We note that
(x) (y) = (x y)
= |x y|
= |x +x
0
x
0
y|
|x x
0
| +|y x
0
|
Then we have
(x) |x x
0
| (y) +|y x
0
|
With this we see there is an extension that perserves the norm. We must now
establish uniqueness.
51
7.0.38 Problem 4
Author: Mohammad Omar
52
7.0.39 Problem 5
Author: Jerey Anderson
Statement: Problem 5
Let 1 p < and let I = (1, 1) denote the open interval in R. Find the
values of as a function of p for which the function [x[
W
1,p
(I)
Proof:
We want to nd the values of as a function of p such that ([x[)
W
1,p
(I).
By the denition of the space W
1,p
(I), we know we must solve three related
problems. These problems include
(a) Find the values of in terms of p such that ([x[)
L
p
(I)
(b) Find the values of in terms of p such that
d
dx
([x[)
L
p
(I) (where we
are taking the weak derivative here)
(c) Compare these values of to nd the correct interval we desire
First, by the denition of the L
p
-norm, we have
|([x[)
|
p
=
_
_
1
_
1
[([x[)
[
p
dm
_
_
1/p
=
_
_
1
_
0
x
p
dm
_
_
1/p
by symmetry of [x[ around x = 0
We know by a theorem discussed in Math 21B here at UC Davis
_
1
x
dm < (1, )
From here we notice that there is a relationship between the integrability of
x
1/
on (0, 1) and the integrability of x
=
_
_
_
(x)
if x < 0
0 if x = 0
x
if x > 0
53
This function is classically dierentiable everywhere except x = 0. The classic
derivative of this function using methods taught in Math 21A here at UC Davis
is given by
d
dx
([x[)
=
_
(x)
1
if x < 0
x
1
if x > 0
Now we take the L
p
norm of the weak derivative ([x[)
:
|([x[)
|
p
=
_
_
1
_
1
[([x[)
1
[
p
dm
_
_
1/p
= 2
_
_
1
_
0
(x)
p(1)
dm
_
_
1/p
We conclude
|([x[)
|
p
< p( 1) > 1
> 1
1
p
Last, we conclude that the only way that [x[
W
1,p
(I) is if > 1
1
p
54
7.0.40 Problem 6
Author: Mohammad Omar
55
Chapter 8
Fall 2008
8.0.41 Problem 1
Author: Bailey Meeker
Statement: Problem 1
Prove that the dual space of c
0
is
1
, where c
0
=
_
x
n
n=1
[ lim
n
x
n
= 0
_
Proof:
We can restate this problem as such: Let T c
0
. Then there is a a
n
n=1
1
such that T(x
n
n=1
) =
x
n
a
n
for all x
n
n=1
1
.
Let
m,n
(x) =
_
1 : n = m
0 : n ,= m
. Let a
n
= T(
n
) where
n
= (0, 0, . . . , 0, 1, 0, . . .)
with the 1 appearing in the nth coordinate. Let x
n
n=1
c
0
. Finally, dene
y
n
n=1
by the folowing: y
n
=
n
k=1
x
k
k
.
Step 1: Show that T can be identied with some sequence a
n
.
By the denition above,
T(y
n
) = T
_
n
k=1
x
k
k
_
=
n
k=1
T(x
k
k
) =
n
k=1
x
k
T(
k
) =
n
k=1
x
k
a
k
.
Note that
n
k=1
T(x
k
k
) =
n
k=1
x
k
T(
k
) since T is a bounded linear functional
on c
0
, and thus scalar multiples come through. Consider
y
n
x
n
n=1
=
n
k=1
x
k
k
x
n
n=1
= (x
1
, x
2
, . . . , x
n
, 0, . . .) x
n
n=1
= (x
1
, x
2
, . . . , x
n
, 0, . . .) x
n
n=1
= (0, . . . , 0, x
n+1
, x
n+2
, . . .).
Then consider the c
0
norm of this dierence:
|y
n
x
n
n=1
|
c
0
= lub [y
n
x
n
n=1
[ = lub[x
k
[ [k n.
Then if we take the limit as n we see that
lim
n
|y
n
x
n
n=1
|
c
0
= lim
n
lub[x
k
[ [k n = limsup
n
x
n
n=1
= lim
n
x
n
= 0.
56
Therefore lim
n
y
n
= x
n
n=1
in c
0
. We assumed that T is a bounded linear
functional, therefore T is continuous. Then
T(x
n
n=1
) = T(lim
n
y
n
) = lim
n
T(y
n
) = lim
n
n
k=1
x
k
a
k
=
k=1
x
k
a
k
.
Step 2: Show that the sequence a
n
n=1
1
.
To prove this, we will use the lowing sequence:
m,n
m=1
where
m,n
=
_
_
_
1 m n and a
m
0
0 m n and a
m
< 0
0 m > n
Consider
T(
m,n
m=1
) =
n
k=1
k,n
a
k
=
n
k=1
[a
k
[
by construction. We know that
m,n
m=1
c
0
since
m,n
= 0 for all n > m.
Specically we note that |
m,n
|
c
0
= 1 by construction.
2
57
8.0.42 Problem 2
Author: Mohamed Omar
Statement: Problem 2
Let f
n
n=1
be a sequence of dierentiable functions on a nite interval [a, b]
such that the function themselves and their derivatives are uniformly bounded
on [a, b]. Prove that f
n
n=1
has a uniformly converging subsequence.
Proof:
Since [a, b] is a compact subset of R, the Arzela-Ascoli Theorem implies that
f
n
C([a, b]) has a uniformly convergent subsequence if we can prove it
is both uniformly bounded and equicontinuous. We are given that f
n
is
uniformly bounded, so it suces to show it is equicontinuous. We prove the
stronger result that the f
n
is uniformly equicontinuous.
Since f
n
are uniformly bounded, there is a constant C > 0 such that
[f
n
(x)[ C for all n and for all x [a, b]. Now let > 0, f
n
and x, y [a, b] be
arbitrary and such that [xy[ <
C
. By the Mean Value Theorem, there exists
c (x, y) such that f
n
(c) =
f
n
(x) f
n
(y)
x y
. From this we have
[f
n
(x) f
n
(y)[ = [f
n
(c)[[x y[ < C
_
C
_
= .
This argument is independent of our choice of x, y and n, so we conclude f
n
is uniformly equicontinuous. 2
58
8.0.43 Problem 3
Author: Mohamed Omar
Statement: Problem 3
Let f L
1
(R) and V
f
be the closed subspace generatred by the translates of
f : f( y)[y R. Suppose
f(
0
) = 0 for some
0
. Show that
h(
0
) = 0 for
all h V
f
. Show that if V
f
= L
1
(R), then
f never vanishes.
Proof:
Recall that, upto scaling,
f(
0
) =
_
R
f(x)e
2ix
0
dx.
Now suppose
f(
0
) = 0. Then for any h = f( y) for y R, we have
h(
0
) =
_
R
f(x y)e
2ix
0
dx
=
_
R
f(x)e
2i(x+y)
0
dx
= e
2iy
0
_
R
f(x)e
2ix
0
dx
= e
2iy
0
f(
0
)
= 0.
Let W
f
be the linear space spanned by functions h = f( y) for y R. Since
the Fourier Transform is linear, it follows
h(
0
) = 0 for all h W
f
. Now any
function h V
f
is either in W
f
(in which case we just showed
h(
0
) = 0),
or is the L
1
limit of functions
h
n
W
f
, in which case
h() = 0 because L
1
convergence implies pointwise convergence. 2
59
8.0.44 Problem 4
Author: Mohamed Omar
Statement: Problem 4
a. State the Stone-Weierstrass theorem for a compact Hausdor space X.
b. Prove that the algevra generated by function so the form
f(x, y) = g(x)h(y)
where g, h C(X) is dense in C(X X).
Proof:
[a.] Let X be a compact Hausdor space, and A be a subalgebra of C(X)
which contains a non-zero constant function, and separates points (i.e. for all
x
1
, x
2
X, there exists f C(X) such that f(x
1
) ,= f(x
2
)). Then A is dense
in C(X).
[b.] First note that if X is a compact Hausdor space, then so is XX. Let
A be the algebra generated by functions of the form g(x)h(y) where g, h C(X).
Since for any pair g, h C(X) we have g(x)h(y) C(XX), A is a subalgebra
of C(X X). The constant function g(x, y) = 1 for all (x, y) X X is
a member of A since it is the product of the constant function g(x) = 1 and
itself. By the Stone-Weierstrass Theorem it suces to show that A separates
points. Let (x
1
, y
1
), (x
2
, y
2
) X X be dierent points. Then at least one of
the following is true: x
1
,= x
2
or y
1
,= y
2
. Assume without loss of generality
that x
1
,= x
2
. Since X is Hausdor, there is an open set U containing x
1
that
does not contain x
2
. By Urysohns Lemma, there is a function g C(X) that
is 1 on U and 0 otherwise. This g, together with the constant function h = 1 on
X, gives the function gh C(X X) that separates (x
1
, y
1
) and (x
2
, y
2
) (since
gh(x
1
, y
1
) = 1 and gh(x
2
, y
2
) = 0). 2
60
8.0.45 Problem 5
Author: Mohamed Omar
Statement: Problem 5
For r > 0, dene the dilation d
r
f : R R of the function f : R R by
d
r
f(x) = f(rx) and the dialation d
r
T of the distribution R T
(R) by
d
r
T, =
1
r
T, d
1/r
_
for all test functions T(R).
Show that the dialation of a regular distribution T
f
, given by
T
f
, =
_
f(x)(x)dx
agrees with the dilation of the corresponding function f
A distribution is homogeneous of degree n if d
r
T = r
n
T. Show that the
distribution is homogeneous of degree 1
IF T is a homogeneous distribution of degree n, prove that the derivative
T
=
1
r
_
f(x)
_
x
r
_
dx
=
_
f(rx)(x) dx
= T
rf
,
Since is arbitrary, we conclude d
r
T
f
= T
rf
.
[b.] Let D(R) be arbitrary. We have
d
r
, =
1
r
, d
1/r
=
1
r
(r 0) (by denition of )
=
1
r
(0)
=
1
r
,
Since is arbitrary, we conclude d
r
=
1
r
and hence is homogeneous of degree
-1.
61
[c.] Let T
, =
1
r
T
, d
1/r
=
1
r
T, (d
1/r
)
=
1
r
T,
1
r
d
1/r
(
)
=
1
r
2
T, d
1/r
(
)
=
1
r
2
T, d
1/r
=
1
r
d
r
T,
=
1
r
r
n
T,
)
= r
n1
T
, .
Since is arbitrary, it follows d
r
T
= r
n1
T and hence T
is homogeneous of
degree n 1.
62
8.0.46 Problem 6
Author: Mohamed Omar
Statement: Problem 6
Calculus of variations.
Proof:
63
Chapter 9
Winter 2009
Author: Ricky Kwok
9.0.47 Problem 1
Statement: Problem 1
Let 1 < p < 2.
(a) Given an example of a function f L
1
(R) such that f / L
p
(R) and a
function g L
2
(R) such that g / L
p
(R).
(b) If f L
1
(R) L
2
(R), prove that f L
p
(R).
Proof:
Let f(x) = x
1/p
(0,1)
. Then
[[f[[
1
= lim
0
_
1
x
1/p
dx =
x
11/p
1 1/p
1
0
=
p
p 1
< .
However,
[[f[[
p
p
= lim
0
_
1
x
1
dx = lim
0
ln(x)
= lim
0
log() = .
Hence f L
1
, but f / L
p
.
Let g(x) = x
1/p
(1,)
. Since p < 2, we have 1/p > 1/2, or 2/p > 1, so that
1 2/p < 0.
[[g[[
2
=
_
1
x
2/p
dx =
x
12/p
1 2/p
1
= lim
x
x
12/p
1 2/p
1
1 2/p
=
p
p 2
.
However,
[[g[[
p
p
=
_
1
x
1
dx = .
64
Thus, g L
2
, but g / L
p
.
Now suppose f L
1
L
2
. Dene
A = x R : [f(x)[ > 1.
Then we have
[[f[[
p
p
=
_
R
[f(x)[
p
dx
=
_
A
[f(x)[
p
dx +
_
A
c
[f(x)[
p
dx
_
A
[f(x)[
2
dx +
_
A
c
[f(x)[
1
dx [[f[[
2
2
+[[f[[
1
1
Therefore, if f L
1
and f L
2
, then f L
p
(R). 2
65
9.0.48 Problem 2
Statement: Problem 2
(a) State the Weierstrass approximation theorem.
(b) Suppose that f : [0, 1] R is continuous and
_
1
0
x
n
f(x) dx = 0
for all nonnegative integers n. Prove that f = 0.
Proof:
The Weierstrass approximation theorem states that for a closed interval [a, b],
the space of continuous functions on this interval C([a, b]) can be uniformly
approximated with polynomials, i.e. there exists a sequence (f
m
) ([a, b])
such that
lim
m
[[f
m
f[[
unif
= lim
m
sup
x[a,b]
[f
m
(x) f(x)[ = 0,
where ([a, b]) is the space of polynomials with domain [a, b].
Since f is continuous over [0, 1], we can take a sequence of polynomials
(f
m
) : [0, 1] R to approximate f uniformly. Then we can pass limit under
the integral sign (see Exercise 2.2).
lim
m
_
1
0
[f
m
(x)f(x)f(x)f(x)[ dx
_
1
0
lim
m
[f
m
(x)f(x)[[f(x)[ dx lim
m
[[f
m
f[[
unif
_
1
0
[f(x)[ dx.
This shows
_
1
0
f(x)
2
dx = 0. Since (f(x))
2
0 for all x, and f is continuous,
f(x) = 0 for all x, i.e. f 0. 2
66
9.0.49 Problem 3
Statement: Problem 3
(a) Dene strong convergence, x
n
x, and weak convergence, x
n
x, of a
sequence (x
n
) in a Hilbert space 1.
(b) If x
n
x weakly in 1 and [[x
n
[[ [[x[[, prove that x
n
x strongly.
(c) Give an example of a Hilbert space 1 and a sequence (x
n
) in 1 such that
x
n
x weakly and
[[x[[ < liminf
n
[[x
n
[[.
Proof:
(a) Let [[ [[ denote the norm in the Hilbert space. Strong convergence is
lim
n
[[x
n
x[[ = 0.
Let (, ) denote the inner product in the Hilbert space. Weak convergence
is
lim
n
(x
n
, y) = (x, y) y 1.
(b) Suppose x
n
x and [[x
n
[[ [[x[[. Then [[x
n
[[
2
[[x[[ and (x
n
, x)
(x, x) = [[x[[
2
and (x, x
n
) = (x
n
, x) (x, x) = [[x[[
2
since x 1,
[[x
n
x[[
2
= (x
n
x, x
n
x) = (x
n
, x
n
) (x
n
, x) (x, x
n
) + (x, x)
[[x[[
2
[[x[[
2
[[x[[
2
+[[x[[
2
= 0.
Hence, [[x
n
x[[
2
0, therefore [[x
n
x[[ 0.
(c) Let 1 = L
2
(R), be the set of all square-integrable functions on the real
line. Dene the sequence g
n
=
(n,n+1)
denote the characteristic function
on the intervals (n, n + 1). Then we have for every n N,
[[g
n
[[ =
_
n+1
n
1
2
dx = 1.
I claim g
n
0. To see this, let f L
2
(R). Then
[[f[[
2
2
=
_
[f(x)[ dx =
n=
_
n+1
n
[f(x)[
2
dx.
Now, Bessels inequality states that
67
lim
n
_
n+1
n
[f(x)[
2
dx = 0
Then the inner product of g
n
with f is given by
(g
n
, f) =
_
n+1
n
f(x) dx =: y
n
.
Notice that by Jensens inequality,
_
n+1
n
[f(x)[
2
dx
__
n+1
n
f(x) dx
_2
.
So we have
_
n+1
n
f(x) dx 0 as n . Therefore,
(g
n
, f) 0 g
n
0.
Then [[0[[ = 0 < 1 = liminf
n
[[g
n
[[. Therefore g
n
=
(n,n+1)
satises
the weak convergence to 0 and [[0[[ < liminf
n
[[g
n
[[ = 1. 2
68
9.0.50 Problem 4
Statement: Problem 4
Suppose that T : 1 1 is a bounded linear operator on a complex Hilbert
space 1 such that
T
= T, T
2
= I
and T ,= iI. Dene
P =
1
2
(I +iT), Q =
1
2
(I iT).
(a) Prove that P, Q are orthogonal projections on 1.
(b) Determine the spectrum of T, and classify it.
Proof:
(a) Notice
P
2
=
1
4
(I +iT)(I +iT) =
1
4
(I +iT +iT +i
2
T
2
) =
1
4
(2I + 2iT) = P,
and similarly Q
2
= Q. Then for all elements y 1,
(Px, y) =
_
x
2
+
iTx
2
, y
_
=
1
2
(x, y) +
i
2
(Tx, y)
=
_
x,
y
2
_
+
i
2
(x, T
y) =
_
x,
y
2
+
i
2
T
y
_
=
_
x,
_
1
2
I +
1
2
iT
_
y
_
= (x, Py),
and similarly (Qx, y) = (x, Qy). Therefore P and Q are orthogonal pro-
jections (projections and self-adjoint).
(b) To nd the spectrum of T, (T), we rst nd the resolvent of T, (T) and
then (T) = C (T).
(T) = C : I T : 1 1, is one-to-one and onto .
We use the fact that an operator is one-to-one and onto if and only if it
is invertible. So, we must nd all values such that (I T)
1
is well-
dened. Using T
2
= I, we take a guess that (I T)
1
= c(I +T) for
some c C. To solve for c,
I = (I T)c(I +T) = c(
2
I T
2
) = c(
2
+ 1)I.
69
We can see that c =
1
2
+1
. Hence,
(I T)
1
=
I +T
1 +
2
.
This operator is well-dened for all values C i. However, since
T ,= iI, i / (T), the resolvent set is consists of all complex numbers
(T) = C. Therefore, (T) = . 2
70
9.0.51 Problem 5
Statement: Problem 5
Let S(R) be the Schwartz space of smooth, rapidly decreasing functions f :
R C. Dene an operator H : S(R) L
2
(R) by
(
Hf)() = isgn()
f() =
_
i
f() if > 0
i
f() if < 0,
where
f denotes the Fourier transform of f.
(a) Why is Hf L
2
(R) for any f S(R)?
(b) If f (R) and Hf L
1
(R), show that
_
R
f(x) dx = 0.
[Hint: you may want to use the Riemann-Lebesgue Lemma.]
Proof:
(a) The function f is in any L
p
space since f is in the Schwartz space. In
particular, f L
1
L
2
, implying the Fourier transform of f is in L
2
and
Plancherels theorem holds:
[[
f[[
2
= [[f[[
2
.
Notice that the L
2
norm of
Hf is equal to the L
2
norm of
f because
Hf()
f()
R.
Combining these two, we have
[[Hf[[
2
= [[
Hf[[
2
= [[
f[[
2
= [[f[[
2
< .
Therefore, Hf L
2
.
(b) Since Hf L
1
, then the Riemann-Lebesgue Lemma states that
Hf is
continuous and
Hf() 0 as [[ . In particular
Hf is continuous at
0, so
i lim
0
+
f() = lim
0
+
Hf() = lim
0
Hf() = -i lim
0
f().
Since the left hand side is equal to the negative of the right hand side,
these two must both equal zero. Hence,
f(0) = 0. But by the denition of
the Fourier transform,
71
0 =
f(0) =
_
R
e
2i0x
f(x) dx =
_
R
f(x) dx.
Therefore f has zero integral. 2
72
9.0.52 Problem 6
Statement: Problem 6
Let denote the Laplace operator in R
3
.
(a) Prove that
lim
0
_
B
c
1
[x[
f(x) dx = 4f(0), f S(R
3
)
where B
c
u in R
2
.
Prove that
|Du
|
L
2 C|u|
L
2,
where the constant C depends on the mollier, not on u.
Proof:
By denition, we have that
u
(x) =
u =
_
R
2
(x y)u(y)dy.
Taking the partial derivative with respect to x
i
, and utilizing the Lebesgue
Dominated Convergence theorem yields
x
i
u
(x) =
x
i
_
R
2
(x y)u(y)dy =
_
R
2
x
i
(x y)u(y)dy.
Now, by Youngs inequality for convolutions (letting p = 2, q = 1 and r = 2)
we have
|
x
i
u
|
L
2 |
x
i
|
L
1|u|
L
2. (10.1)
Now, examining |
x
i
|
L
1, we have
|
x
i
|
L
1 =
_
R
2
[
x
i
(x)[dx
=
_
R
2
x
i
_
x
dx.
74
Making the change of variables y =
x
, we have that
x
i
=
1
x
1
and dx =
2
dy.
Therefore the above expression becomes
1
_
R
2
[
y
i
(y)[ dy.
Because C
c
(R
2
), there exists a constant C that bounds the above expres-
sion. Thus we have
|
x
i
u
|
L
2
C
.
Combining this with (10.1), and noting that i is arbitrary gives the result
|Du
|
L
2 C|u|
L
2.
This is exactly what we wanted to show. 2
75
10.0.54 Problem 2
Author: Owen Lewis
Statement: Problem Number
Let B(0, 1) R
3
denote the unit ball [x[ < 1. Prove that u = log [x[
H
1
(B(0, 1)).
Proof:
We begin by showing that u L
2
(B(0, 1)). We rst note that for [x[ < 1, then
[log ([x[)[
1
|x|
. Therefore, utilizing polar coordinates we have
|u|
L
2
(B(0,1))
=
_
B(0,1)
u
2
dx
=
_
B(0,1)
log
2
([x[)dx
_
B(0,1)
1
[x[
2
dx
= C
_
1
0
1
r
2
r
2
dr
= C.
Therefore, u L
2
(B(0, 1)). Now a quick calculation shows that
x
i
u =
1
[x[
x
i
[x[
,
and thus
Du =
x
[x[
2
.
By denition,
|Du|
L
2
L
2
(B(0,1))
=
_
B(0,1)
Du Dudx
=
_
B(0,1)
x x
[x[
4
dx
= C
_
1
0
r
2
r
4
r
2
dr
= C.
Therefore Du L
2
(B(0, 1)). Combined with the previous result we have that
u H
1
(B(0, 1)).
This is exactly what we wanted to show. 2
76
10.0.55 Problem 6
Author: Owen Lewis
Statement: Problem Number
Prove that a normed linear space is complete if and only if every absolutely
summable sequence is summable.
Proof:
First, we assume that our space is complete. Let x
n
be an absolutely summable
sequence. Now dene y
m
=
m
n=1
x
n
. We have that
|y
m
y
k
| = |
k
n=m
x
n
|
k
n=m
|x
n
|.
Because x
n
is absolutely summable, we know that
n=1
|x
n
| = M, and thus
for any > 0, we can choose m suciently large such that
m
n=1
x
n
< .
Combined with the previous inequality, this shows that the sequence y
m
is
Cauchy. By completeness, it has a limit x. Therefore we have
n=1
x
n
= lim
m
m
n=1
x
n
= lim
m
y
m
= x.
This shows that every absolutely summable sequence is summable.
Now assume that every absolutely summable sequence is summable. Let
x
n
be a Cauchy sequence. By passing to a subsequence if necessary, we can
claim that |x
k
x
k1
| 2
k
. Now dene y
1
= x
1
, and y
k
= x
k
x
k1
for
k > 1. By denition,
m
k=1
y
k
= x
m
.
Furthermore, by construction
k=1
|y
k
| =
k=1
|x
k
x
k1
|
k=1
2
k
< .
Thus the sequence y
k
is absolutely summable. By assumption it is therefore
summable, with limit x. Therefore,
lim
k
x
k
=
k=1
y
k
= x,
which is to say that the subsequence x
k
is convergent. However, if any Cauchy
sequence has a convergent subsequence, then the entire sequence is convergent.
Thus, our original Cauchy sequence is convergent, and x
n
x. This shows that
the space is complete, and completes the proof.
77
Chapter 11
Winter 2010
Author: Nathan Hannon
11.0.56 Problem 1
Statement: Problem 1
Let (X, d) be a complete metric space, x X and r > 0. Set D := x X :
d(x, x) r, and let f : D X satisfying
d(f(x), f(y)) k d(x, y)
for any x, y D, where k (0, 1) is a constant.
Prove that if d( x, f( x) r(1 k) then f admits a unique xed point.
(Guidelines: Assume the Banach xed point theorem, also known as the con-
traction mapping theorem.)
Proof:
Let x D. Then
d(f(x), x) d(f(x), f( x)) +d(f( x), x)
k d(x, x) +r(1 k)
rk +r(1 k)
= r.
Hence f(x) D. Since x was arbitrary, we may write f : D D, and f satises
the hypotheses of the Banach xed point theorem. We conclude that f admits
a unique xed point. 2
78
11.0.57 Problem 2
Statement: Problem 2
Give an example of two normed vector spaces, X and Y , and of a sequence
of operators, T
n
n=0
, T
n
L(X, Y ) (L(X, Y ) is the space of the continuous
operators from X to Y , with the topology induced by the operator norm) such
that T
n
n=0
is a Cauchy sequence but it does not converge in L(X, Y ).
(Notice that Y cannot be a Banach space otherwise L(X, Y ) is complete.)
Proof:
Let X = R and let Y L
1
([1, 1]) be the subspace consisting of continuous
functions. We note that any operator T L(X, Y ) is determined completely by
T(1) and that |T| = |T(1)|. It therefore suces to nd a Cauchy sequence f
n
in Y that does not converge in Y .
Let f
n
= x
1
2n+1
Y . Then f
n
sgn x in L
1
([1, 1]). Hence (f
n
) is Cauchy
in L
1
([1, 1]) and hence in Y . However, because sgn x , Y , (f
n
) does not
converge in Y . 2
79
11.0.58 Problem 3
Statement: Problem 3
Let (a
n
) be a sequence of positive numbers such that
n=1
a
3
n
converges. Show that
n=1
a
n
n
also converges.
Proof:
We note that (a
n
)
3
(N). Furthermore, since
n=1
1
n
3
2
< ,
we have
_
1
n
_
3
2
(N). By Hlders inequality,
n=1
a
n
n
|(a
n
)|
3
_
_
_
_
_
1
n
__
_
_
_
3
2
< .
This is exactly what we wanted to show. 2
80
11.0.59 Problem 4
Statement: Problem 4
Suppose that h : [0, 1]
2
[0, 1]
2
is a continuously dierentiable funcion from
the square to the square with a continuously dierentiable inverse h
1
. Dene
an operator T on the Hilbert space L
2
([0, 1]
2
) by the formula T(f) = f h.
Prove that T is a well-dened bounded operator on this Hilbert space.
Proof:
Let u = h(x) and let J(u) denote the Jacobian of x with respect to u. Since h
and h
1
are continuously dierentiable, J is bounded. Let
M = sup
u[0,1]
2
[J(u)[.
We also note that h maps [0, 1]
2
to itself. We can use a change of variables to
evaluate |T(f)|
2
:
|T(f)|
2
2
=
_
[0,1]
2
[(f h)(x)[
2
dx
=
_
[0,1]
2
[J(u)[ [f(u)[
2
du
_
[0,1]
2
M[f(u)[
2
du
= M|f|
2
2
.
Hence
|T(f)|
2
M|f|
2
;
i. e., T is well-dened and bounded. 2
81
11.0.60 Problem 5
Statement: Problem 5
Let H
s
(R) denote the Sobolev space of order s on the real line R, and let
|u|
s
=
__
R
_
1 +[[
2
_
s
[ u()[
2
d
_1
2
denote the norm on H
s
(R), where u() =
1
2
_
R
u(x)e
ix
dx denotes the
Fourier transform of u.
Suppose that r < s < t, all real, and > 0 is given. Show that there exists a
constant C > 0 such that
|u|
s
|u|
t
+C|u|
r
u H
t
(R).
Proof:
Let
g() =
_
1 +[[
2
_1
2
.
We note that g is strictly positive.
Suppose that u H
t
(R) and dene a measure on R by
d = [ u()[
2
d.
We note that
|g
r
|
L
2
(R,)
=
__
R
_
1 +[[
2
_
s
[ u()[
2
d
_1
2
= |u|
r
.
Similarly, |g
s
|
L
2
(R,)
= |u|
s
and |g
t
|
L
2
(R,)
= |u|
t
.
Let
C =
sr
st
.
We note that C does not depend on the choice of u.
Suppose that g() = y. If y
1
st
, then
y
s
= y
st
y
t
y
t
y
t
+Cy
r
.
Otherwise, y <
1
st
, and
y
s
= y
sr
y
r
< Cy
r
y
t
+Cy
r
.
It follows that
g()
s
g()
t
+Cg()
r
R.
82
Using Minkowskis inequality,
|g
s
|
L
2
(R,)
|g
t
+Cg
r
|
L
2
(R,)
|g
t
|
L
2
(R,)
+C|g
r
|
L
2
(R,)
|u|
s
|u|
t
+C|u|
r
.
This is exactly what we wanted to show. 2
83
11.0.61 Problem 6
Statement: Problem 6
Let f : [0, 1] R. Show that f is continuous if and only if the graph of f is
compact in R
2
.
Proof:
Denote by G the graph of f in R
2
. Suppose that f is continuous. Since [0, 1] is
compact, f must be bounded. Hence G is bounded.
Now suppose that (x
n
, y
n
) G and (x
n
, y
n
) (x, y). Then x
n
x and
y
n
y. However, because f is continuous, y
n
= f(x
n
) f(x). It follows that
y = f(x) and that (x, y) G. Since this holds for all convergent sequences
(x
n
, y
n
), G is closed.
By the Heine-Borel theorem, G is compact.
Conversely, suppose that G is compact. By the Heine-Borel theorem, G
is closed and bounded. Suppose that x
n
x and that f(x
n
) = y
n
. Let
y = liminf y
n
and y
= limsup y
n
. Since G is bounded,
< y < y
k
such that y
n
k
y and
y
n
k
y
.
Then (x
n
k
, y
n
k
) (x, y) and (x
n
k
, y
n
k
) (x, y
). Since G is closed,
(x, y), (x, y