10

Lecture 3: Compactness.

Definitions and Basic Properties.

Definition 1. An open cover of a metric space X is a collection (countable

or uncountable) of open sets {Uα } such that X ⊆ ∪α Uα . A metric space X is
compact if every open cover of X has a finite subcover. Specifically, if {Uα } is an
open cover of X , then there is a finite set {α1 , . . . , αN } such that X ⊆ ∪N
n=1 Uαn .
A collection of subsets {Aα } of a metric space X has the finite intersection
property (FIP) if for every finite set of indices, {α1 , . . . , αN }, the intersection
∩N
n=1 Aαn is non-empty.

Theorem 2. (Theorem 1, p. 93, Kolomogorov) A metric space X is compact if

and only if every collection of closed sets with the FIP has nonempty intersection.

Proof: (=⇒) Suppose that X is compact and that {Aα } has the FIP
and has empty intersection. Consider the collection of open sets {(Aα )c }
consisting of the complements of the Aα . Since ∩α Aα = ∅ then ∪α (Aα )c =
X so that {(Aα )c } is an open cover of X . Since X is compact, there exists
a finite set {α1 , . . . , αN } such that ∪N c
n=1 (Aαn ) = X . But this means that
N
∩n=1 Aαn = ∅ which contradicts the assumption that {Aα } has the FIP.
(⇐=) Let {Uα } be an open cover of X . Then ∩α (Uα )c = ∅. Therefore the
collection of closed sets {(Uα )c } cannot have the FIP. Therefore there is a
finite collection {α1 , . . . , αN } such that ∩N c
n=1 (Uαn ) = ∅. But this means
N
that ∩n=1 Uαn = X and {Uα } has a finite subcover.

Theorem 3.

(a) (Theorem 2, p. 93, K) Every closed subset of a compact metric space is

compact.

(b) (Theorem 3, p. 93, K) If K is a compact subset of a metric space X , then

K is closed

Proof: (a) Let M be a closed subspace of the compact space X , and let
{Uα } be an open cover of M, since M is closed, the collection {Uα }∪{Mc }
is an open cover of X which contains a finite subcover since X is compact.
Therefore, there is a collection {α1 , . . . , αN } such that {Uαn }N c
n=1 ∪ {M }
N
covers X . Hence {Uαn }n=1 must cover M. Therefore, M is compact.
(b) Let K be a compact subset of X and let y ∈ / K. It will suffice to show
that y ∈ / [K] as this would imply that [K] ⊆ K. For each x ∈ K, let
²x = d(x, y)/2. Then B(x; ²x ) ∩ B(y; ²x ) = ∅. The collection {B(x; ²x )}x∈K
forms an open cover of K and since K is compact there is a finite set
{x1 , . . . , xN } such that K ⊆ ∪N
n=1 B(x; ²x ). Let ² = min{²x1 , . . . , ²xN }.
 11

Then B(y; ²) ∩ K = ∅ since if x ∈ K then x ∈ B(xn ; ²xn ) for some n, and

hence is not in B(y; ²xn ) for that n. Since B(y; ²) ⊆ B(y; ²xn ), x cannot be
in B(y; ²) either. Therefore y ∈/ [K].
Continuous Functions on Compact Spaces

Lemma 4. A function f from a metric space X into a metric space Y is

continuous at each point of X if and only if for every open set U ⊆ Y, f −1 (U) is
open.

Proof: (=⇒) Suppose that f is continuous on X , that U ⊆ Y is open, and

let x ∈ f −1 (U). Since f (x) ∈ U there is an ² > 0 such that B(f (x); ²) ⊆ U
and since f is continuous at x a δ > 0 such that if y ∈ B(x; δ) then
f (y) ∈ B(f (x); ²) ⊆ U . Therefore, B(x; δ) ⊆ f −1 (U) and f −1 (U) is open.
(⇐=) Let x ∈ X , ² > 0, and assume that the inverse image of under f of
an open set is open. Then f −1 (B(f (x); ²)) is open and contains x. Hence
there is a δ > 0 such that B(x; δ) ⊆ f −1 (B(f (x); ²)). But this is precisely
the definition of continuity at x.

Theorem 5.
(a) (Theorem 5, p. 94, K) The continuous image of a compact space is compact.

(b) (Theorem 6, p. 94, K) A continuous injection of a compact space X onto

a metric space Y (which also must be compact) is a homeomorphism, that
is, the inverse function is also continuous.

Proof: (a) Let X be a compact space and f a continuous function from

X into a metric space Y. Let {Uα } be an open cover of f (X ). Since f is
continuous and by Lemma 4, each of the sets f −1 (Uα ) is open and the
collection {f −1 (Uα )} forms an open cover of X (for given x ∈ X , there is
an αx such that f (x) ∈ Uαx , and hence x ∈ f −1 (Uαx )). Since X is compact
there is a finite set {α1 , . . . , αN } such that X ⊆ ∪N
n=1 f
−1
(Uαn ). But this
N
means that f (X ) ⊆ ∪n=1 Uαn , and so f (X ) is compact.
(b) Since we are trying to show that f −1 is continuous, it is enough to
show that (f −1 )−1 = f maps open sets to open sets. Let U ⊆ X be open.
Then its complement U c is closed and hence compact since X is compact by
Theorem 3(a). By part (a), the f (U c ) is also compact and by Theorem 3(b)
it is closed. Since f (U c ) = f (U)c (Hint: This is because f is injective and
onto) f (U )c is also closed, hence f (U ) is open.

Compactness and Completeness

Theorem 6. (Theorem 7, p. 94, K) If a metric space X is compact then every

infinite subset of X has a limit point.
12

Proof: Suppose X is compact and let M be an infinite subset of X . We

can extract from M a sequence of distinct points {xn }∞ n=1 . Let An =
{xn , xn+1 , . . .} Then {[An ]} is a sequence of closed sets with the FIP.
Since X is compact, there is an x ∈ ∩∞ n=1 An .
To see that x is a limit point of M, let ² > 0 and consider B(x; ²). Since
x ∈ [An ] for all n, and since An is closed, x is a closure point for each An .
If x were an isolated point for some An , then x would equal xm for some
m. In this case, x could not be an isolated point for any An with n > m
since all of the xn are distinct. This means that for all sufficiently large n,
B(x; ²) ∩ {xn , xn+1 , . . .} 6= ∅. Hence B(x; ²) ∩ M \ {x} 6= ∅ and x is a limit
point of M.

Definition 7. A subset M of a metric space X is totally bounded if for

every ² > 0 there is a finite set of points {xn }N N
n=1 such that M ⊆ ∪n=1 B(xn ; ²).
Another way of putting this is that every point of M is within a distance ² of at
least one of the xn . The set {xn } is called a finite ²–net of M.

Remark 8. (a) A subset M of a metric space X is bounded if for some x ∈ X

and r > 0, M ⊆ B(x; r).
(b) It is always true that a totally bounded set is bounded. (Exercise)

Example 9. (a) Every bounded set in Rn is totally bounded. To see this, let
δ > 0 and consider the lattice of points δZn . Then every point of Rn is within
√
a distance of at most δ n/2 from a point of δZn . Therefore, the collection of
√
balls {B(x; δ n/2)}x∈δZn covers all of Rn . For each R > 0, let BR = {x =
(x1 , . . . , xn ) ∈ Rn : |xi | ≤ R}. That is, BR is a box centered at the origin whose
sides are of length 2R. Note that for any R > 0, only finitely many balls from
√
the set {B(x; δ n/2)}x∈δZn are required to cover BR . Specifically, if R < N δ for
some N ∈ N then at most N n balls are required to cover BR .
Now, if M ⊆ Rn is bounded and ² > 0 is given, choose δ so small that
√
δ n/2 < ² and N ∈ N so large that M is contained in the box BN δ . Then M
can be covered by finitely many balls of radius ².
(b) The unit sphere in `2 is bounded but not totally bounded. This implies
that total boundedness and boundedness are not equivalent properties in infinite
dimensional spaces.
To see why this is true, note that by definition the unit sphere is bounded.
Now let en be the sequence in `2 consisting of all zeros except for a 1 in the nth
√
position. Then all of the en are on the unit sphere and if n 6= m, d(en , em ) = 2.
√
Therefore, if ² < 2/2, then no two en can sit in the same ² ball and we will
require infinitely many of them to cover the unit sphere, which therefore cannot
be totally bounded.

Theorem 10. Let X be a totally bounded metric space. Then the following
are equivalent.
 13

(a) X is compact.

(b) Every countable open cover of X admits a finite subcover.

(c) Every countable collection of closed sets with the FIP has nonempty in-
tersection.

(d) Every infinite subset of X has a limit point.

Proof: (a)=⇒(b) Follows from the definition of compactness.

(b)=⇒(a) Let {Uα } be an open cover (countable or uncountable) of X . Since
X is totally bounded, we can define a doubly infinite sequence of points in X
as follows. For each k = 1, 2, 3, . . . let {xn,k }N
n=1 be a finite collection with
k

Nk
the property that X ⊆ ∪n=1 B(xn,k , 1/k). For each x ∈ X , choose n and k so
that x ∈ B(xn,k , 1/k) and B(xn,k , 1/k) ⊆ Uα for some α. This can be done
by noting that x ∈ Uα for some α so that for some ² > 0, B(x; ²) ⊆ Uα . Now
choose k so large that 1/k < ²/2 and let n be such that x ∈ B(xn,k , 1/k).
Then if y ∈ B(xn,k , 1/k), d(x, y) ≤ d(x, xn,k ) + d(xn,k , y) < 1/k + 1/k < ².
Hence B(xn,k , 1/k) ⊆ Uα . The collection of all such xn,k chosen this way
is a countable set and the associated Uα must be countable as well. Since
each x ∈ X is accounted for by some xn,k , this countable collection of open
sets covers X .
Hence by (b) this countable subcover of X admits a finite subcover.
Hence X is compact.
(b)⇐⇒(c) This is the same argument as in Theorem 2.
(a)=⇒(d) This is Theorem 6.
(d)=⇒(c) Let {Fn }∞ n=1 be a countable collection of closed sets satisfying
the FIP. Define AN = ∩N ∞
n=1 Fn . Then the sequence {AN }N +1 satisfies (1)
AN is nonempty and closed for each N , (2) AN +1 ⊆ AN for all N , (3)
{AN }∞ ∞ ∞
N +1 has the FIP, and (4) ∩n=1 Fn = ∩N +1 AN .
There are now two possibilities: First, for some N0 , AN0 = AN0 +1 = · · ·
in which case ∩∞ N =1 AN = AN0 and (c) is satisfied. The other possibility is
that the AN are distinct for infinitely many N . By renumbering the {AN }
if necessary, we can assume that the AN are distinct for all N . In this
case, we can choose an infinite sequence of distinct points {xn } such that
xN ∈ AN for all N . By (d), this sequence must have a limit point, call
it x. Since the sequence {xN , xN +1 , . . .} ⊆ AN for each N , x is a limit
point for each set AN . Since each AN is closed, x ∈ AN for each N . Hence
x ∈ ∩∞ ∞
N +1 AN and by property (4) above, ∩n=1 Fn is not empty. Therefore
(c) holds in this case also.

Theorem 11. (Theorem 2, p. 100, K) A metric space X is compact if and

only if it is totally bounded and complete.
14

Proof: (=⇒) Suppose X is compact. Given ² > 0, we can cover X with

the collection of open balls {B(x; ²)}x∈X and extract a finite subcover of
the form {B(xn ; ²)}N n=1 . Hence X is totally bounded. Now suppose that
{xn } is a Cauchy sequence in X . We can assume that {xn } is an infinite
set because any Cauchy sequence with only finitely many distinct elements
must have all of the xn the same for all n sufficiently large. By Theorem 6,
xn has a limit point and hence {xn } must converge to that limit point and
X is complete.
(⇐=) Suppose that X is totally bounded and complete. To show it is com-
pact, we will use Theorem 10 and show that every infinite subset of X has
a limit point. To this end, assume that M is an infinite subset of X and
extract from M a sequence {xn }∞ n=1 of distinct points.
We will define a subsequence of this sequence as follows. Since X is totally
bounded, there is a finite collection of balls of radius 1 covering X . At least
one of these balls must contain infinitely many of the xn . Choosing such a
ball allows us to define a subsequence {x1,n }∞ n=1 contained in one such ball.
Using total boundedness again, we know that X is covered by finitely many
balls of radius 1/2, at least one of which contains infinitely many of the
x1,n . Choosing one such ball allows us to extract a subsequence of {x1,n }
which we will call {x2,n }∞ n=1 . Continuing in this fashion, we have a doubly
infinite sequence of points {xk,n } with the property that (1) {xk+1,n }∞ n=1 ⊆
{xk,n }∞n=1 , (2) d(x k,n , x k,m ) < 1/k for all n, m ∈ N. For each k, choose
an element of the sequence {xk,n } which has not previously been chosen.
In this way, define a sequence of distinct points {yk }∞ k=1 with the property
that yk ∈ {xk,n }∞ n=1 .
Since for every m > 0, d(yn , yn+m ) < 2/n, {yk } is a Cauchy sequence in
X and hence converges (say to y) by the completeness of X . To see that y
is a limit point of M, note that for every ² > 0, B(y; ²) contains infinitely
many of the yk and hence infinitely many points in M.