Topological Spaces: Compactness in Metric Spaces
Definition
Let be a metric space and any positive number. Then a set
is said to be an -net for a set
if,
for every there is at least one point such that
.
Definition
Given a metric space and a subset , suppose has a
finite -net for every . Then is said to be
totally bounded.
Note
Example 5 shows that the Hilbert cube (or fundamental parallelepiped) is an
infinite-dimensional totally bounded set.
Theorem 1
Every countably compact metric space is totally bounded.
Proof
Suppose is not totally bounded, then there is an such that
has no finite -net. Choose any point . Then
contains at least one point, say , such that , since otherwise would be an -net for .
Moreover, contains a point such that ,
, since otherwise the pair
would be an -net for . More generally, once having found the
points , we choose such that
. This
construction gives an infinite sequence of distinct points
with no limit points, since
if
. But then cannot be countably compact.
Corollary 1
Every countably compact metric space has a countable everywhere dense subset
and a countable base.
Proof
Since is totally bounded, by Theorem 1, has a finite -net for
every . The union of all these nets is then a countably
everywhere dense subset of . It follows from Theorem 5
Topological Spaces: Basic Concepts that has a countable base.
Corollary 2
Every countably compact metric space is compact.
Proof
Remark
Total boundedness is not sufficient for a metric space to be compact. The set
of rational points in the interval forms a metric space which is
totally bounded but not compact.
Theorem 2
A metric space is compact if and only if it is totally bounded and
complete.
Proof
To see that compactness of implies completeness of , we need only note
that if has a Cauchy sequence with no limit, then has
no limit point in . Hence if is compact, is complete. By Theorem 1,
if is compact, is totally bounded.
Conversely, suppose is totally bounded and complete, and let be
any infinite sequence of distinct points in . Let be a finite -net
for , and construct a closed sphere of radius 1 about every point of .
Since these spheres cover and there are finitely many of them, at least
one of the spheres, say , contains an infinite subsequence
of the sequence . Let be a finite -net for , and
construct a closed sphere of for every point of . Then at least one
of these spheres, say , contains an infinite subsequence
of the sequence . Continue this construction indefinitely,
where has radius . Let be the closed sphere with the
same centre as but with a radius twice as large (i.e., equal to
). Then clearly
and moreover as . Since is
complete, it follows from the nested sphere theorem (Theorem 2, Complete Metric
Spaces) that
In fact, there is a point such that
(recall Problem 3, Complete Metric Spaces). Clearly is a limit point of
the original sequence , since every neighbourhood of contains
some sphere and hence some infinite subsequence .
Therefore every infinite sequence of distinct points of has a
limit point of . It follows that is countably compact and hence compact,
by Corollary 2.
Theorem 3
A subset of a complete metric space is relatively compact if and only
if it is totally bounded.
Proof
An immediate consequence of Theorem 2 and the fact that a closed subset of a
complete metric space is itself complete.
Remark
The utility of Theorem 3 stems from the fact it is usually easier to prove that
a set is totally bounded than to give a direct proof of its relative
compactness.
Definition
A family of functions defined on a closed interval is
said to be uniformly bounded if there exists a number such that
for all and all .
Definition
A family of functions defined a closed interval is said
to be equicontinuous if, given any , there exists a number
such that imples
for all and all
.
Theorem 4 (Arzelà)
A necessary and sufficient condition for a family of continuous
functions defined on a closed interval to be relatively compact
in is that be uniformly bounded and equicontinuous.
Proof
We first prove necessity and then sufficiency.
(Necessity.) Suppose is relatively compact in . Then by
Theorem 3, it is totally bounded, so given any , there is a
finite -net in (see Problem 1).
Being a continuous function defined on a closed interval, each is
bounded:
Let . By the definition of an
-net, given any , there is at least one
such that
Therefore , i.e. is uniformly bounded. Moreover, each function in the
-net is continuous, and hence uniformly continuous, on .
Hence, given any , there is a such that
whenever .
Let . Then, given any
and choosing such that , we have
whenever . This proves the equicontinuity of .
(Sufficiency.) Suppose is uniformly bounded and equicontinuous. Accordin
to Theorem 3, to prove that is relatively compact in , we
need only show that is totally bounded, i.e., that given any , there exists a finite -net for in . Since
is uniformly bounded, for all , and by
equicontinuity, let be such that
for all whenever . Divide the interval along the x-axis into subintervals of length less than
, by introducing points of subdivision such
that
and then draw a vertical line through each of these points. Similarly, divide
the interval along the y-axis into subintervals of length
less than , by introducing points of subdivision
such that
and then draw a horizontal line through each of these points. In this way, the
rectangle , is divided into cells of horizontal
side length less than and vertical side length less than .
We now associate with each function a polygonal line
which has vertices at points of the form and differs
from the function by less than at every point (the
reader should draw a figure and convince themself on the existence of such a
function).
Since
by construction, we have
Moreover,
since is linear between the point and . Let be any
point in and the point of subdivision nearest to on the
left. Then
i.e. the set of polygonal lines forms an -net for .
But there are obviously only finitely many such lines. Therefore is
totally bounded.
Theorem 5 (Peano)
Let be defined and continuous on a plane domain . Then at least
one integral curve of the differential equation
passes through each point of .
Proof
By the continuity of , we have in some domain
containing the point . Draw the lines with slopes
and through the point . Then draw vertical lines
and which together with the first two lines form two
isosceles triangles contained in with common vertex .
This gives a closed interval , which will figure in the
rest of the proof.
The next step is to construct a family of polygonal lines, called
Euler lines, associated with the differential equation
.
We begin by drawing the line with slope through the point
. Next, choosing a point on the first line, we draw
the line through the point . Then, choosing a point
on the second line, we draw the line with slope
through the point , and so on indefinitely. Suppose we construct a
whole sequence of such Euler lines going
through the point , with the property that the length of the
longest line segment making up approaches as .
Let be the function with graph . Then this gives a family of
functions , all defined on the interval
, which is easily seen to be uniformly bounded and equicontinuous (as
they are bounded by and are straight lines). It follows from Arzelà’s
theorem that the sequence is relatively compact and therefore
contains a uniformly convergent subsequence
Let . Then clearly
, so that the curve passes through the point
. We now show that the satisfies the differential
equation in the open interval . This means
showing that, given any and any points , we
have
whenever is sufficiently small, or equivalently that
whenever is sufficiently large and is sufficiently small. Let
. Then, by continuity of , given any , there is
a number such that
whenever , .
The set of points satisfying these inequalities is a rectange which we
denote by . Let be so large that for all , the length of the
longest segment making up is less than and moreover
Then all the Euler lines with live inside the rectange
(why?). Suppose has vertices
,
where .
[Note we assume that . The case is treated similarly.]
Then
Hence, if ,
Adding these inequalities, we get
if , which is equivalent to
Problems
Let be a totally bounded subset of a metric space .
Prove that the -nets figuring in the definition of total
boundedness of can always be chosen to consist of points of rather
than .
Hint. Given an -net for consisting of points
, all with of some point of ,
replace each point by a point such that
.
Prove that every totally bounded metric space is separable.
Hint. Construct a finite -net for every . Then take
the union of these nets.
Let be a bounded subset of the space . Prove that the set of
all functions with compact.
Given two metric compacta and , let be the set of all
continuous mappings of into . Let distance be defined in by
the formula . Prove that
is a metric space. Let be the set of all mappings of
into , with the same metric. Prove that is closed in .
Hint. Use the method of Problem 1, page 65 to prove that the limit of a
uniformly convergent subsequence of continuous mappings is itself a
continuous mapping.
Let , , and be the same as in the preceding problem. Prove
the following generalisation of Arzelà’s theorem: A necessary and sufficient
condition for a set to be relatively compact is that
be an equicontinuous family of functions, in the sense that given any
, there exists a number such that
implies for all
and all .
Hint. To prove the sufficiency, show that is relatively compact in
(defined in the preceding problem) and hence in , since
is closed in . To prove the relative compactness of in
, first represent as a union of finitely many pairwise disjoint
sets such that implies . For
example, let be a -net for , and let
Then let , be an -net in , and let be the
set of all functions taking the values on the sets . Given any
and any , let be such that and let
be such that . Show that ,
thereby proving that is a finite -net for in .