Anda di halaman 1dari 2

UNFINISHED BUSINESS: COMPACTNESS IN METRIC

SPACES II
MATH 431 OCT. 16

Well start by giving an analogue to the Cantor intersection theorem.


Theorem (Generalized Cantor Intersection Theorem). The metric space
(X, ) is complete if and only if every decreasing sequence A1 A2 of
nonempty, closed sets with diam(An ) 0 has a nontrivial intersection.
Its not exactly a generalization, since we make the extra assumption that
diam(An ) 0.
Proof. If (X, ) is complete then from the nested, closed, nonempty sets we
can choose a sequence xj Aj . The sequence (xj ) is Cauchy, since the tail
Tn = {xj : j n} is contained in An and diam(Tn ) diam(An ) implies
that diam(Tn ) 0 (this is an application of problem 3d from HW 3). This
implies that (xj ) is a Cauchy sequence and therefore has a limit in X. But
each Aj is closed and therefore contains x.
Conversely, consider the closures of the tails of a Cauchy sequence (xj )
satisfy T1 T2 . . . , and diam(Tn )T= diam(Tn ) 0 (this is by problem
3b from HW 3). It follows that x Tj is a limit of (xj ), since (xj , x)

diam(Tj ) 0 (again, by problem 3d from HW 3).
Theorem (Generalized Heine-Borel). For a metric space (X, ), X is compact iff X is complete and totally bounded.
Proof. () It was shown that compactness implies completeness in Theorem
4.101.
Compactness
implies total boundedness because for every
 > 0, the open
S
S
cover xX B(x, ) can be refined to a finite subcover nj=1 B(xj , ).
()The opposite direction is more interesting. We prove this by contradiction. To this end, suppose that O = {Oi | i I} is an open cover of X
containing no finite subcover.
By total boundedness, any set in X can be covered by finitely many balls
of a given radius, hence if each of these balls were finitely coverable, then
so would be the entire set. Thus, by assumption, at least one of the balls
cannot be covered by a finite sub cover of O. We proceed inductively.
1The proof of this fact is rather simple: let (x ) be a Cauchy sequence in X. By
n
sequential compactness (every sequence in X has a convergent subsequence), there is x
so that x Tn for each n. Because (xn ) is Cauchy, we have (xn , x) diam(Tn ) =
diam(Tn ) 0.
1

MATH 431 OCT. 16

For the initial case, cover X by finitely many balls of radius 1: X =


SN 0
(0)
j=1 B(xj , 1). At least one such ball cannot be covered by finitely many
elements of O. Call this Z1 and observe that it has diameter 2.
For the general case, assume we have Zk , an open set with diameter
1/k which cannot be covered by finitely many elements of O. Since Zk is
totally bounded (as a subset of totally bounded set X), we can cover Zk
S k
(k)
1
with finitely many balls Zk N
j=1 B(xj , k+1 ). At least one of the sets
(k)

1
B(xj , k+1
) Zk cannot be covered by finitely many elements of O. Call
this Zk+1 and observe that it is nonempty, open and has diameter at most
2
k+1 .
We have constructed of a decreasing sequence X Z1 Z2 . . . of sets
with limk diam(Zk ) = 0 and none is empty. Because X is complete, we
can apply the generalized Cantor intersection theorem
T to the nested family
of sets Z1 , Z2 , . . . to ensure that there is some z k=1 Zk .
Because O is an open cover, there is some O0 O containing z. Since z
is an interior point of O0 , there is r > 0 so that B(z, r) O0 . It follows that
Zk B(z, r) O0 as soon as diam(Zk ) = k2 < r. This is a contradiction,
since it shows that each of the Zk (for k > 2/r) are coverable by a finite
number of elements of O (indeed, all but finitely many of them can be
covered by O0 ).


This allows us to prove the converse of Theorem 3.38 ii) in the book.
Corollary. If (X, ) is sequentially compact then (X, ) is compact.
Proof. By contradiction. Assume X is not compact. Then X is not totally
bounded or X is not complete.
If X is not complete, then it contains a Cauchy sequence (xn ) without
a limit. The image of the sequence T = {xn | n N} does not have an
accumulation point (if p were an accumulation point of T then we could
find a subsequence (xn(k) ) with xn(k) p, which would imply that xn p).
So X is not sequentially compact.
If X is not totally bounded, then, for some r > 0, no matter how we
choose the finite set {u1 , . . . , un } X, X fails to be covered by {B(uj , r) |
j = 1 . . . n}. This permits
S choice of a sequence {u1 , u2 , ...} in X in that order
n
so that un+1 X \
j=1 B(uj , r) . A consequence of this construction is
that (un , uj ) r for all n 6= j. Hence the infinite set {un | n N} has no
accumulation point.


Anda mungkin juga menyukai