1

Chapter 10: Compact Metric Spaces

10.1 Definition. A collection of open sets {Ui : i I} in X is an open
cover of Y X if Y iI Ui . A subcover of {Ui : i I} is a subcollection
{Uj : j J} for some J I that still covers Y . It is a finite subcover if J
is finite.
10.2 Definition.
1. A metric space X is compact if every open cover of X has a finite
subcover.
2. A metric space X is sequentially compact if every sequence of points
in X has a convergent subsequence converging to a point in X.
10.3 Examples.
1. (0, 1] is not sequentially compact (using the Heine-Borel theorem) and
not compact. To show that (0, 1] is not compact, it is sufficient find
an open cover of (0, 1] that has no finite subcover. But a moments
consideration of the cover consisting exactly of the sets Un := ( n1 , 2)
shows that this is just such a cover. That is, nN Un = (0, 2) (0, 1].
But if F is any finite subset of {Un : n N}, then F contains an
element Uk such that k i for each Ui F . But this means that
Uk Ui for each Ui F and hence F = Uk = ( k1 , 2) ( (0, 1].
2. [0, 1] is sequentially compact (applying Heine-Borel). In fact, [0, 1] is
also compact (as we will see shortly).
3.

is neither compact nor sequentially compact. That it is not sequentially compact follows from the fact that R is unbounded and
Heine-Borel. To see that it is not compact, simply notice that the
open cover consisting exactly of the sets Un = (n, n) can have no
finite subcover. Using reasoning similar to that of example 1, if F is
a finite subset of {Un : n N} then F contains an element Uk such
that k i for each Ui F . But then F = Uk = (k, k) ( R, so F
cannot be an open cover for R.

4. X = C[0, 1] is not compact. Denote by Br (f ) the open ball of radius r

under the sup-norm centered at the function f C[0, 1], and consider

2
the set U = {B 1 (f ) : f C[0, 1]}. Clearly, U is an open cover of
4
C[0, 1]. Now define the sequence (fn )nN as

2n(n + 1)(x 1 )
n+1
fn (x) =
1

2n(n
+
1)(x

1
if 0 x n+1
1
2n+1
if n+1
< x 2n(n+1)

if
if

2n+1
<x
2n(n+1)
1
<x1
n

1
n

The graphs of f1 , f2 , f3 , and f4 are plotted below.

Notice that k fm fn k = 1 whenever m 6= n. Hence, each element

of U can contain at most one function from (fn )nN , and therefore
every finite subset of U fails to cover C[0, 1].
5. (Cantors Intersection Theorem.) If C1 C2 C3 . . . is a decreasing
sequence of nonempty sequentially compact subsets of Rn , then k1 Ck
is non-empty.
To see this, choose the sequence (an )nN so that an Cn for every
n. Clearly, (an ) is a sequence in C1 . The compactness of C1 tells us
that (an ) has a convergent subsequence (ank )kN . Say that ank a
as k . Then a is the limit of a sequence in Ci for each i, which
means that a Ci for each i, and the result follows.
10.4 Definition. A metric space X is totally bounded if, for every  > 0,
there exist x1 , x2 , . . . , xk X, with k finite, so that {B (xi ) : 1 i k} is
an open cover of X.

3
10.5 Examples.
1. Example 4 of 10.3 shows that the closed unit ball in C[0, 1] is not
totally bounded.
2. (0, 1] is totally bounded since for any  > 0, {(i, (i + 1)) : 0 i
1
, i Z} is an open cover. However, (0, 1] is not compact. (We will

see shortly that the ingredient missing from (0, 1] and essential for
compactness is in fact completeness.)
10.6 Definition. A collection of closed sets {Ci : i I} has the finite
intersection property if every finite subcollection has nonempty intersection.
10.7 Theorem. (The Borel-Lebesgue Theorem.) For a metric space
(X, ), the following are equivalent:
1. X is compact.
2. Every collection of closed subsets of X with the finite intersection
property has non-empty intersection.
3. X is sequentially compact.
4. X is complete and totally bounded.
Proof. Well first show (1) implies (2). Consider a collection of closed
subsets {Ci : i I} of X having the finite intersection property, and assume
that iI Ci = . Put Ui := Cic for each i, and notice that each Ui is
open. We have iI Ui = iI Cic = (iI Ci )c = X, so that {Ui : i I}
is an open cover of X. Since X is compact, there exists a finite subcover
{Un1 , Un2 , . . . , Unk } of X. Hence,
X = Un1 Un2 . . . Unk
= (Unc1 Unc2 . . . Unck )c
= (Cn1 Cn2 . . . Cnk )c ,
which means that Cn1 Cn2 . . . Cnk = , in contradiction with the finite
intersection property.
The argument from (2) to (1) runs as follows: Suppose that {Ui : i

4
I} is an open cover of X and put Ci := Uic for each i. Suppose further
that no finite subset of {Ui : i I} covers X. Then if a subcollection
{Cn1 , Cn2 , . . . , Cnk } of {Ci : i I} satisfies Cn1 Cn2 . . . Cnk = , we
would have
Un1 Un2 . . . Unk
= (Unc1 Unc2 . . . Unck )c
= (Cn1 Cn2 . . . Cnk )c
= X,
a contradiction with the assumption that no finite subset of {Ui : i I}
covers X. Thus, {Ci : i I} has the finite intersection property. By (2),
iI Ci 6= , so iI Ui 6= X, meaning that {Ui : i I} is not an open cover
for X.
Well now show that (3) follows from (1). Assume we have a sequence
(xn )nN in X with no convergent subsequence. Since no term in the sequence
can occur infinitely many times (otherwise we would have a convergent subsequence), we can assume without loss of generality that xi 6= xj whenever
i 6= j. Notice that each term of the sequence (xn ) is an isolated point of
{xn : n N}, since otherwise, (xn ) would have a convergent subsequence.
Hence, for each i there exists an open ball, call it Ui , centred at xi with the
property that xj
/ Ui for all i 6= j. Now put U0 := X\{xn : n N}. U0 is
open since its complement consists only of isolated points, and so is closed.
Then {U0 } {Un : n N} is an open cover for X. But this open cover has
no finite subcover, since any finite subcollection of these sets would fail to
include infinitely many terms from the sequence (xn ) in its union.
To see that (3) implies (4), assume that (xn )nN is a Cauchy sequence
in a sequentially compact space X. Say that (xnk )kN is a convergent subsequence of (xn ) and that xnk x. Let  > 0 be given, and choose N
so that (xi , xj ) < /2 whenever i, j N . Next, choose nk > N so that
(xnk , x) < /2. Then we have
(x, xN ) (x, xnk ) + (xnk , xN ) < /2 + /2 = .
Thus, xn x as n , showing that X is complete. Next, well show
that X is totally bounded.
Assume that X is not totally bounded, and take  > 0 such that X
cannot be covered by a collection consisting of only finitely many -balls.

5
Choose x1 X, x2 X\B (x1 ), then x3 X\B (x1 )\B (x2 ), and so on.
We thus have a sequence (xn ) which cannot contain a convergent subsequence since (xi , xj )  for all i 6= j.
We can also obtain (3) from (4). Consider a sequence (xn )nN in X.
Since X is totally bounded, we have, for every n N, a finite set of
(n)
(n)
(n) (n)
(n)
points {y1 , y2 , . . . , yr(n) } such that X B 1 (y1 ) . . . B 1 (yr(n) ). Let
n
n
(Sn )nN be the sequence of finite subsets of X obtained by putting Sn :=
(n)
(n)
{y1 , . . . , yr(n) )}. We can find a convergent subsequence (zn )nN of (xn )
(1)

using the following procedure: Since S1 is finite, there is a yn(1) S1

(1)

such that B1 (yn(1) ) contains infinitely many points from (xn ). Select z1
(2)

(1)

from B1 (yn(1) ). Now, since S2 is finite, there is a yn(2) S2 such that

(1)

(2)

B1 (yn(1) ) B 1 (yn(2) ) contains infinitely many points from (xn ). Choose z2

2

(1)

(2)

from B1 (yn(1) ) B 1 (yn(2) ). Now continue this procedure for each k > 1, se2

(k)

(i)

lecting yn(k) from Sk such that ki=1 B 1 (yn(i) ) contains infinitely many points
i

(i)

from (xn ), and then selecting zk from ki=1 B 1 (yn(i) ). (zn ) is clearly Cauchy,
i
and by the completeness of X, (zn ) converges to a point in X.
Finally, we show that (3) implies (1). Well need the following preliminary result: If (X, ) is a sequentially compact metric space, and {Ui : i I}
an open cover for X. Then there is an r > 0 such that for each x X,
Br (x) Ui for some i I. The proof is by contradiction. Assume that
for some r > 0, there is an x X, possibly depending on r, such that for
each i I, Br (x) * Ui . Now choose the sequence (xn )nN in X so that
B 1 (xn ) * Ui for all i I.
n
Since X is sequentially compact, (xn ) has a convergent subsequence,
(xnk )kN . Say that xnk x as k , where x X. There must be
some i0 such that x Ui0 , and, since Ui0 is open, an r0 > 0 such that
Br0 (x) Ui0 . So choose N such that (x, xN ) < 12 r0 and N1 < 21 r0 . Now if
y B 1 (xN ), then
N

(x, y) (x, xN ) + (xN , y)

1
1
< r0 + r 0
2
2
< r0 ,
and hence y Br0 (x) Ui0 . It follows that B 1 (xN ) Br0 (x) Ui0 , a
N
contradiction, and the preliminary result is proven.

6
Returning to the main argument, let {Ui : i I} be an open cover of
X. By the preliminary result, there exists an r > 0 such that for each
x X, Br (x) Ui for some i I. We know that (3) implies (4), thus X
is totally bounded. That is, X Br (y1 ) Br (y2 ) . . . Br (yk ) for some
points y1 , y2 , . . . , yk X, with k N. However, for each i I, we have
that Br (yi ) Uj(i) for some j(i) I. Thus {Uj(1) , Uj(2) , . . . , Uj(k) } is a finite
subcover for X, and were done. 
10.8 Theorem. Let (X, ) and (Y, ) be metric spaces and f : X Y
be a continuous function. Then for each compact subset C X, f (C) Y
is compact.
Proof. Let {Ui : i I} be an open cover of f (C), and for each i
I, define Vi to be the pre-image of Ui under f . Notice that since f is
continuous, each Vi is open. Thus, {Vi : i I} is an open cover of C. But
since C is compact, there exists a finite subcover {Vi(1) , Vi(2) , . . . , Vi(k) } for
C, and hence {Ui(1) , Ui(2) , . . . , Ui(k) } is a finite subcover of f (C). So, f (C)
is compact.