Math 1540

Analysis II
Spring, 2011
Notes #1
These notes are supplementary to the text, and do not replace it. I will post
them on the website, at irregular intervals. Not all material in the text, or in the
lectures, will be discussed in the notes. Ordinarily, you should read the text rst,
then the relevant notes, if any. This rst set is concerned with the material in
Sections 5.5 and 5.6 of the text.
1 The space of continuous functions
While you have had rather abstract denitions of such concepts as metric spaces and
normed vector spaces, most of 1530, and also 1540, are about the spaces 1
n
. This
is what is sometimes called classical analysis, about nite dimensional spaces,
and provides the essential background to graduate analysis courses. More and
more, however, students are being introduced to some innite dimensional spaces
earlier. This section is about one of the most important of these spaces, the space
of continuous functions from some subset of a metric space ` to some normed
vector space `. The text gives a careful denition, calling the space C (. `).
The simplest case is when ` = 1 (= 1
1
). Let be a subset of 1. Then let
C (. 1) = , : 1 [ , is continuous .
In the most common applications is a compact interval. Thus C ([0. 1] . 1) is the
space of all continuous , : [0. 1] 1.
When 1 and ` = 1. C (. 1) is often shortened to C () . However any
careful writer makes it clear what is meant by this notation. More generally, could
be a subset of any metric space `. and , could map ` to 1 or to 1
m
.
Denition 1 If `, then
C (. 1
m
) = , : 1
m
[ , is continuous .
It is easy to turn C (. 1
m
) into a vector space, by dening addition and scalar
multiplication in the usual way: (, + q) (r) := , (r) + q (r), (c,) (r) = c (, (r)) .
You probably ran into this even in linear algebra. (I use := to denote an equation
which denes the quantity on the left.)
1
However it is also very useful to dene a norm, so that C (. 1
m
) is a normed
vector space. This is a little trickier. We shall only do so when is a compact
set. In that case, Theorem 4.2.2 (in the text) implies that each , C (. 1
m
) is
bounded. (It is a good thought exercise to look at Theorem 4.2.2, observe that the
word bounded does not appear, and explain to yourself why it implies that in the
case we are considering, , is bounded.) This allows us to dene a norm. With an
eye toward the application we will give shortly, we take : = 1.
Denition 2 If 1 is compact, and , C (. 1) . let
[[,[[ = max
x2A
[, (r)[ .
Note that in this denition I am assuming that the maximum of , is dened.
This is justied by the Maximum-Minimum theorem, Theorem 4.4.1. If I say that
the maximum of a function , is c. this means that there is a particular r in the
domain of , such that , (r) = c. and furthermore, there is no r with , (r) c. Be
sure you are straight on the dierence between a maximum and supremum.
The text gives the denition required for more general and `. As an example,
consider C (1. 1) where 1 is the interval (. ). In that case , : 1 1 can be
continuous and yet not have a maximum. You should be able to think of an example
which is a bounded function. More signicantly, , might not be bounded.
In that case, C (1. 1) is not turned into a normed space in a natural way. (There
may be some peculiar way to do it. ) Instead, for general . the text considers
C
b
(. 1) := , C (. 1) [ , is bounded .
Note that this is a vector space. Then for , C
b
(. 1) we let
[[,[[ = sup
x2A
[, (r)[ .
(What theorem tells us this exists?) Some homework at the end of the section is
concerned with this.
Theorem 5.5.1 implies that if is compact then C (. 1) is a normed linear
space, as is C
b
(. 1) for general . We will mainly be concerned with part (ii) of
this theorem, since I think that in all our examples ` will be a normed vector space.
Once we dene a norm, we can dene convergence of sequences.
Denition 3 A sequence ,
k
C
b
(. 1) converges to , C
b
(. 1) if
lim
k!1
[[,
k
,[[ = 0.
2
As the text points out, the following assertion is then obvious from the denitions
of convergence of a sequence in C
b
(. 1) and uniform convergence as it was discussed
in section 5.1.
Proposition 4 A sequence ,
k
C
b
(. 1) converges to , C
b
(. 1) if and only
if
lim
k!1
,
k
(r) = , (r)
uniformly on .
Why? and does it make any dierence if is compact? If is compact, what
is the relation between C (. 1) and C
b
(. 1)? What can you say about these two
spaces if you dont know that is compact (it might be, or not).
2 Examples
1. : = 1. = [0. 1]. The space C ([0. 1] . 1) is often denoted by C ([0. 1]) . the image
space 1 being understood. Clearly, , (r) = r
2
+ cos r + c
x
is in C (. 1) . but
q (r) =
_
1
x
if 0 < r _ 1
0 if r = 0
is not.
2. : = 1. = (0. 1) . Then q (r) =
1
x
for 0 < r < 1 is in C (. 1) . Is q in
C
b
(. 1)?
3. : = 2. = [0. 1] [0. ). Is r
y
in C
b
(. 1)? What about
x
?
3 The Ascoli-Arzela theorem
This is one of the most important theorems in analysis. As one application, it tells
us what subsets of C (. 1) are compact.
First lets recall how to decide if a subset o of 1
n
is compact. Since 1
n
is a
metric space, the Bolzano-Weierstrass theorem tells us that o is compact if and only
if every sequence of points in o has a convergent subsequence. Then the Heine-Borel
theorem says that a set 1
n
is compact if and only if it is closed and bounded.
We saw that C ([0. 1]) is a normed linear space. Let
o = , C ([0. 1]) [ [[,[[ _ 1 .
3
Obviously o is bounded. To see that o is closed, suppose that ,
n
o and ,
n
q
in C ([0. 1]) . (This was dened in the previous section.) I claim that [[q[[ _ 1.
Suppose not, and [[q[[ 1. Let o = [[q[[ 1. or
[[q[[ = o + 1 (1)
Then there is an ` such that for : _ `. [[,
n
q[[ <

2
. In this case, by the triangle
inequality,
[[q[[ = [[q ,
n
+ ,
n
[[ _ [[q ,
n
[[ +[[,
n
[[ _
o
2
+ 1.
which contradicts (1) .
Hence, o is closed and bounded. Let us see if o is compact. Consider the sequence
,
n
where ,
n
(r) = r
n
for 0 _ r _ 1. Then ,
n
o for each :. Also.
lim
n!1
,
n
(r) =
_
0 if 0 _ r < 1
1 if r = 1.
Hence,,
n
does not converge to a continuous function. Further, no subsequence of
,
n
converges to a continuous function. (Do you think this needs further proof? If
so, ll in the details!)
This shows that o is not compact. The Heine-Borel theorem does not hold in the
innite dimensional normed linear space C ([0. 1]). Some other testable condition is
needed to guarantee that a set o is compact.
Denition 5 A sequence ,
n
of functions in C ([0. 1]) is uniformly bounded if there
is an ` 0 such that [[,
n
[[ _ ` for : = 1. 2. 3. .
Denition 6 A sequence o of functions in C ([0. 1]) is equicontinuous if for each
0 there is a o 0 such that if r and are in [0. 1] with [r [ < o, and , o.
then
[, (r) , ()[ < . (2)
In other words, in the denition of continuity of a function, the o does not depend
on r. or the function , chosen from o. (Compare this with the denition of uniform
continuity of a function.)
One example of an equicontinuous set of functions is a set o of functions in
C ([0. 1]) such that every function , in o satises a Lipschitz condition in [0. 1] and
the Lipschitz constant 1 can be chosen to be the same for every , in o. In this case,
if 0 is given then choose o =
"
L
and check that (2) holds. The o does not depend
on ,.
4
For instance, we can let o be the sequence ,
n
of functions given by ,
n
(r) =
x
n
n
.
Then ,
0
n
(r) = r
n1
. so [,
0
n
(r)[ _ 1 for every : and every r [0. 1] . By the mean
value theorem we can choose 1 = 1.
On the other hand, we will show that r
n
is a uniformly bounded sequence
(on [0. 1]), but not an equicontinuous one. Uniform boundedness is obvious. For
equicontinuity, suppose that =
1
2
and that there is a o such that (2) holds for
every : and every r and in [0. 1] with [r [ < o. Choose = 1 and r = 1

2
.
Since lim
n!1
_
1

2
_
n
= 0. we can choose : so large that ,
n
(1) ,
n
_
1

2
_
=
1
_
1

2
_
n

1
2
. contradicting (2). Hence, o is not equicontinuous. As we saw, o
does not have a convergent subsequence.
Theorem 7 (Ascoli-Arzela) If o = ,
n
is a uniformly bounded and equicontinuous
sequence of functions in C ([0. 1]) . then o has a subsequence which converges in
C ([0. 1]) .
Remark 8 It is important that the limit function is also in C ([0. 1]) . But this is
what we mean when we say that a subsequence converges in C ([0. 1]).
The proof (for a more general case) is in the text, and will be given in class.
The practical application of this theorem is often easier than the example above
might suggest. If , is dierentiable on [0. 1] . and for some . [,
0
(r)[ _ 1 for every r
[0. 1] . then 1 is a Lipschitz constant for ,. This follows from the mean value theorem.
Hence, for a sequence ,
n
, if each ,
n
is dierentiable on [0. 1] . and [,
0
n
(r)[ _ 1 for
every : and every r [0. 1] . then ,
n
is equicontinuous. If in addition, there is
a 1 such that [[,
n
[[ _ 1 for every :. then ,
n
is equicontinuous and uniformly
bounded, and some subsequence of ,
n
converges in C ([0. 1]).
Turning to compactness, if o C ([0. 1]) . every , o is dierentiable on [0. 1] .
and there are numbers 1 and 1 such that [[,[[ _ 1 and [,
0
(r)[ _ 1 for every , o
and every r [0. 1] . then o is a compact subset of C ([0.1]).
As an example of a compact set we can consider
o =
_
q [ q (r) =
_
x
0
, (:) d: for some , C ([0. 1]) with [[,[[ _ 1
_
.
You should use the Ascoli-Arzela theorem to show that o is compact.
4 Contraction mapping theorem
This is another result that can be used to show that sequences converge. It is easier
to prove than Ascoli-Arzela and gives a stronger result. However there are many
5
examples where Ascoli-Arzela can be applied but the contraction mapping theorem
cannot. The statement below is weaker than the one in the text, but is adequate for
many applications, including the one given in the next section. You should compare
it with the one in the text (Theorem 5.7.1) and gure out what makes the text
version stronger. Also determine why the proof given below would not be a valid
proof for Theorem 5.7.1.
Theorem 9 Let 1 be a Banach space, ` a closed subset of 1. and a mapping
from ` to ` such that for some / [0. 1).
[[(r) ()[[ _ / [[r [[
for any two points r and in `. Then there is a unique point . in ` such that
(.) = ..
Denition 10 Such a point . is called a xed point for .
Proof. Choose some r
0
in `. Dene a sequence r
i
by setting r
i+1
= (r
i
) . for
i = 0. 1. 2. . Then for any :.
r
n
= r
0
+ (r
1
r
0
) + (r
2
r
1
) + + (r
n
r
n1
) .
Also, for i _ 1.
[[r
i+1
r
i
[[ _ / [[r
i
r
i1
[[ .
and by induction we easily show that
[[r
i+1
r
i
[[ _ /
i
[[r
1
r
0
[[ .
Because [/[ < 1.
1
i=1
/
i
converges, which implies that
1
i=1
[[r
i+1
r
i
[[ converges.
(Why?) By the Weirerstrass M test (Theorem 5.2.2),
1
i=1
(r
i+1
r
i
) converges in
1. and hence lim
n!1
r
n
exists. Let . be this limit. Since ` is closed and r
n
`
for each :, . `. Also, r
n+1
= (r
n
) . and so (from the denition of limit)
lim
n!1
(r
n
) = lim
n!1
r
n+1
= ..
Further, for any :
|(.) .| = [[(.) (r
n
) + (r
n
) .[[
_ / [[. r
n
[[ +[[(r
n
) .[[ .
6
Since the limit of the right side as : is zero, and the left side is independent of
:. it follows that the left side is zero for every :. and so . is a xed point for .
To prove uniqueness, suppose that there are two xed points, say r and . Then
, (r) = r. , () =
and so
[r [ = [, (r) , ()[ _ / [r [
where 0 < / < 1. This is only possible if r = .
5 Application
There is an easy application of the contraction mapping theorem to dierential equa-
tions. We consider an initial value problem for an ode, of the form
r
0
= , (t. r) (3)
r (0) = 0. (4)
Here, r is one dimensional. It is also possible to consider systems of odes, in which
r and , (t. r) are :-dimensional, but we will stick with one dimensional case.
Theorem 11 Suppose that , is continuous in a rectangle given by 1 _ t _ 1.
A _ r _ A. Suppose also that in the function , satises a Lipschitz condition
of the form
[, (t. r) , (t. )[ _ 1[r [ .
Then there is some o 0 such that the initial value problem (3)-(4) has a unique
solution c(t) on the interval [o. o] such that c(0) = 0.
Remark 12 Notice that I have used c to denote a solution to the ode (3). This
is sometimes more convenient than denoting the solution by r. To say that c is a
solution to (3) is simply to say that c
0
(t) = , (t. c(t)).
Remark 13 It is important also to note that the interval [o. o] may be smaller than
[1. 1]. In class I will give an example showing that this is a necessary restriction
in the theorem. The solution may not exist on [1. 1] .
7
Proof. We will use an integral equation which is equivalent to the initial value
problem (3)-(4). If is a continuous function on the interval 1 := [1. 1] . with
[ (t)[ _ A on 1. and , is continuous on , then Theorem 4.3.1 implies that
, (:. (:)) is a continuous function of : for : 1. Then Theorem 4.8.4 implies that
_
t
0
, (:. (:)) d: is dened for : 1. and the fundamental theorem of calculus tells
us that this function is dierentiable and that
d
dt
__
t
0
, (:. (:)) d:
_
= , (t. (t)) .
Let
1 = max
(t;x)2
[, (t. r)[
and set
o = min
_
1.
A
1
.
1
21
_
.
Then let 1 = C ([o. o]) and ` = 1 [ [[[[ _ A . where [[[[ = max
t
[ (t)[ .
It is easy to show that ` is a closed subset of 1. On ` dene a mapping : ` 1
by
() (t) =
_
t
0
, (:. (:)) d:.
If `. then for each t with [t[ _ o we use the denition of o to get
[[()[[ _

_
t
0
1d:

= [1t[ _ 1o _ A.
Hence, : ` `. Also, if
1
and
2
are in `. then for each t [o. o] .
[(
1
) (t) (
2
) (t)[ _
_
t
0
[, (:.
1
(:)) , (:.
2
(:))[ d: _
_

0
[, (:.
1
(:)) , (:.
2
(:))[ d:
_
_

0
1[
1
(:)
2
(:)[ d: _ 1o [[
1

2
[[ .
Hence [[(
1
) (
2
)[[ _ 1o [[
1

2
[[ . Since 1o < 1. is a contraction on `.
and so by the contraction mapping theorem, has a unique xed point, which we
denote by c. Then for [t[ _ o,
c(t) =
_
t
0
, (:. c(:)) d:. (5)
Obviously c(0) = 0. and by dierentiating (5) we see that c satises (1). This proves
Theorem 11.
8
6 Homework (Due Wednesday, January 12 at the
beginning of class. No late homework accepted.
Problems will be discussed at the beginning of
the class where they are due.)
When the answer is in the back of the book, you still have to prove that it is correct.
1. pg. 272, # 1
2. pg. 272, # 4
3. pg. 275, #5.
4. pg. 282, # 4.
5. pg. 322, # 47 (rst part). (You can take : = : = 1. )
9