Lecture 9

Compactness, I

Let’s review what we’ve done in the last week:

1. We’ve defined the notion of a topology T on a set X. The definition

was a bit abstract; a topology is a collection of subsets of X, which
we will call open, satisfying some conditions. When a set is given a
topology, we call it a topological space.

2. We also saw that a given set may have many different topologies, so we
often have to specify which topology we’re talking about. (For example,
when X = Rn , we could give X the discrete, the trivial, or the standard
topology. Our favorite should be the standard topology.)

3. We also saw that any poset admits a topology, called the Alexandroff
topology.

4. We defined a function between topological spaces to be continuous if

preimages of open subsets are open.

5. We then gave a way to turn subsets of spaces into spaces, by giving

subsets the subspace topology. So for example, things like simplices,
spheres, and disks are all of a sudden topological spaces!

Today, we’re going to learn about a very important property that some
topological spaces have, called compactness.
In a highly sophisticated way, a space being “compact” will be very much
like a set being “finite.” Compactness is some way that we can talk about a
space as being “small enough” to comprehend using very nice tools.

91
92 LECTURE 9. COMPACTNESS, I

This doesn’t mean that non-compact spaces are inaccessible. (Even for
sets, we can understand infinite sets like Z just fine.) But there are more
tools available for studying compact spaces.

9.1 Continuous functions compose

Before we go any further, let me state an incredibly important property of
continuous functions.

Proposition 9.1.1. Let X, Y, Z be topological spaces. Fix a continuous

function f : X æ Y and another continuous function g : Y æ Z. Then the
composition
g¶f :X æZ
is also continuous.

Try to prove this as an exercise before looking at the proof:

Proof. Let W µ Z be an open subset. We must prove that (g ¶ f )≠1 (W ) is

an open subset of X. Observe:

(g ¶ f )≠1 (W ) = {x œ X | (g ¶ f )(x) œ W }
= {x œ X | g(f (x)) œ W }
= {x œ X | f (x) œ g ≠1 (W )}
= f ≠1 (g ≠1 (W )).

But since g is assumed continuous, we know that g ≠1 (W ) is open. Hence

f ≠1 (g ≠1 (W )) is open (because f is also assumed continuous).
This completes the proof.

9.2 Open covers and subcovers

Do you remember when we proved that Rn could be written as a union of
open balls? There were many, many different ways that we could do this—we
could take a union of open balls all centered at the origin, or we could take
a union of open balls with all kinds of different centers, too.
Regardless, either collection is an example of a cover.
9.2. OPEN COVERS AND SUBCOVERS 93

Definition 9.2.1. Let X be a set. A cover of X is a collection of subsets

{U– }–œA such that €
U– = X.
–œA

Oftentimes, instead of working with the generality of a function A æ P(X),

we will assume that a cover is specified by a subset of P(X) (that is, we may
assume that A æ P(X) is an injection).
In this case, we may denote an open cover by a fancy U , so

U = {U– }–œA .

The word “cover” is used because the collection “covers” all of X. (Every
element of x is inside one of the U– .)
Definition 9.2.2. Choose a subset B µ A of the indexing set A. If the
collection {U— }—œB is a cover of X, we call it a subcover of the original cover.
Example 9.2.3. Let A = Rn ◊ R>0 , and for all (x, r) œ A, define U(x,r) =
Ball(x, r). We saw long ago that this collection {U(x,r) }(x,r)œA is a cover of
Rn .
Now choose the subset B µ A consisting of those (x, r) for which x is the
origin of Rn , and r is a positive integer. Then the collection {U— }—œB is also
a cover of Rn . Hence {U— }—œB is a subcover of {U– }–œA .
Definition 9.2.4. Let X be a topological space. A cover {U– }–œA is called
an open cover if, for every – œ A, the set U– µ X is open.
In other words, an open cover is a cover consisting of open subsets.
Example 9.2.5. Both the covers in Example 9.2.3 are open covers of Rn .
The following exercise shows that covers of X induce covers of subspaces;
moreover, open covers of X induce open covers of A, simply by intersecting
elements of the cover with A:
Exercise 9.2.6. Let {U– }–œA be an open cover of a topological space X.
Fix a subset A µ X, and endow it with the subspace topology.
Then the collection {V– }–œA , where

V– := U– fl A

is an open cover of A.
94 LECTURE 9. COMPACTNESS, I

Proof. We must verify two facts: That {V– }–œA is a cover of A, and that
each V– is open (in A).
Each V– is open by definition of subspace topology.
t
Clearly, the union –œA V– is a subset of A because each V– is a subset
of A.
So we must only show that A is a subset of this union. So fix a œ A.
Then a œ X because A is a subset of X. In particular, because {U– }–œA is
a cover of X, there is some – œ A for which a œ U– . Because a œ A by
assumption, we conclude that a œ U– fl A = V– . This finishes the proof.

9.3 The definition of compactness

Here is one of the most important definitions in topology; it is one of the
most confusing as well, but it is incredibly powerful.

Definition 9.3.1. Let X be a topological space. We say that X is compact

if every open cover of X admits a finite subcover.

In other words, X is called compact if the following holds. For every open
cover {U– }–œA , there exists some finite subset B µ A so that the collection
{U— }—œB is a cover of X.

9.4 Straightforward examples

Whenever you are given a new definition, it is good to ask for examples and
non-examples. Here are a few:

Example 9.4.1. Let X be a finite set, and let T be a topology on the finite
set. (So that we may consider X to be a topological space.)
I claim that X is compact.
Here is a proof: Any topology T µ P(X) has only finitely many elements,
so X has only finitely many open subsets to begin with. In particular, fix an
open cover {U– }–œA . It may be that A itself is an infinite set, we can choose
some finite subset B µ A so that for every – œ A, there is some — œ B for
which U– = U— . Then {U— }—œB is a finite subcover.

Example 9.4.2. Let X = Rn with the standard topology. This is not

compact.
9.4. STRAIGHTFORWARD EXAMPLES 95

To prove that something is not compact, we need only find one example
of an open cover that does not admit a finite subcover.
So let’s choose the following open cover: Let A = Z>0 (so that A is the
set of positive integers), and define, for all n œ A,
Un := Ball(0, n)
to be the open ball of radius n about the origin of Rn . We have seen that
{Un }nœZ>0 is an open cover of Rn .
Now, if B µ A is any finite subset (so that B is some finite collection
of positive integers), there is a maximal element of B. Call this maximal
element N . Then €
U— = UN = Ball(0, N ).
—œB

In particular, the collection {U— }—œB is not a cover of Rn because any element
x œ Rn having distance larger than N from the origin is not inside the union
t
—œB U— .
Thus, the open cover {Un }nœZ>0 admits no finite subcover. This shows
that Rn is not compact.
Example 9.4.3. Let X be an infinite set, and let Tdisc be the discrete topol-
ogy on X. (This means that any subset of X is declared open.)
I claim that (X, Tdisc ) is not a compact space.
Let A = X, and for all x œ A, declare Ux to be the one-element set
{x} µ X. By definition of the discrete topology, Ux is open in X. Hence
{Ux }xœA is an open cover of X.
On the other hand, this does not admit a finite subcover. For a finite
subset of A is a finite collection {x1 , . . . , xn } of some points in X. The union
of the sets {xi }i=1,...,n is clearly the set {x1 , . . . , xn }, which is not all of X
because X is assumed infinite.
This shows that any infinite set, when equipped with the discrete topol-
ogy, is not compact.
Here is another example, which we state as a proposition:
Proposition 9.4.4. Let X be a compact space. Suppose Y is a space home-
omorphic to X. Then Y is also compact.
This proposition is a verification that “compactness” is a notion that de-
pends only on the topology of space—after all, homeomorphisms are equiv-
alences of topological spaces.
96 LECTURE 9. COMPACTNESS, I

Proof. Let V be an open cover of Y . We must prove that V admits a finite

subcover.
Fix a continuous function „ : X æ Y . Then for all V œ V, the pre-image
„≠1 (V ) is an open subset of X. Let U be the collection of open subsets given
by
U := {U µ X | U = „≠1 (V ) for some V œ V.
In other words, U œ U if and only if U arises as the preimage of some V œ V.
Then U is an open cover of X, because x œ X =∆ f (x) œ V for some
V œ V. (This last claim uses that V is a cover of Y .)
Because X is compact, we may choose a finite subcover of U. So let
t
U1 , . . . , Un be the finite collection of open sets of X for which i=1,...,n Ui = X
and for which Ui œ U for all i.
Now suppose that „ is further a homeomorphism.1 Then „ is a bijec-
tion, so if the collection {U1 , . . . , Un } is a cover of X, then the collection
{„(U1 ), . . . , „(Un )} is a cover of Y . Moreover, again because „ is a bijection,
the fact that Ui = „≠1 (Vi ) means „(Ui ) = Vi . So the collection {V1 , . . . , Vn }
is a finite subcover of V.
We have shown that any open cover of Y admits a finite subcover, com-
pleting the proof.

9.5 Some compact posets

To see the simplest examples of compact topological spaces that have in-
finitely many elements, we turn to posets.
Let Z = Z>0 fi {Œ}. In other words, as a set, Z is obtained by adding
one new element to the set of all positive integers. We call this new element
Œ. You’ll see why in a moment.
Note that Z has countably infinite cardinality.
We can give Z a poset structure as follows. If a, b œ Z>0 µ Z, we declare
aleqb if a is less than or equal to b in the usual sense (of integers being less
than or equal to each other). If a œ Z>0 and b = Œ, we declare a Æ b. We
also declare that Œ Æ Œ to ensure reflexivity.
I will leave it to you to see that this relation is transitive and antisym-
metric.
1
This does NOT mean that every function from X to Y H is a homeomorphism; but
this is a round-about way of saying that we can choose some homeomorphism from X to
Y , and we are calling it „ for no good reason.
9.6. CLOSED INTERVALS 97

Then let us endow Z with the Alexandroff topology.

Proposition 9.5.1. Z is compact.

Proof. Let U µ Z be an open subset. By definition of Alexandroff topology,

we know that for every a œ U and every b œ Z satisfying a Æ b, we know
that b œ U . Thus, whenever U is non-empty, we know that Œ œ U .
Well, if U is an open cover of Z, there is some element U œ U containing
1 œ Z. Then by definition of Alexandrov topology, U contains every element
b satisfying b > 1. In other words, U = Z.
Thus, any open cover U must satisfy Z œ U, and in particular, any open
cover admits a subcover consisting of only one open subset: Z.

If you study the above proof, you realize that the only property of Z we
used is that there is some element (namely, 1) that is less than or equal to
any other element. Such an element doesn’t always exist in a poset. (For
example, the poset Z doesn’t have such an element.) But here is another
example of a compact poset:

Proposition 9.5.2. Let A be a set. (Infinite or otherwise.) Let P(A) be the

power set, considered as a poset via the relation µ. Then P(A), given the
Alexandroff topology, is compact.

Proof. Any open cover U of P(A) must have some U œ U for which ÿ œ U .
But by definition of Alexandroff topology, any B œ P(A) satisfying ÿ µ B
must be contained in U ; meaning U contains every element of P(A).
In other words, any open cover U must satisfy P(A) œ U; so U admits
a finite subcover—in fact, a subcover consisting of a single element called
P(A).

9.6 Closed intervals

The following is the most important example of a compact space. Most other
examples of compact subspaces of Euclidean space are in one way or another
constructed out of this one:

Theorem 9.6.1. Let A = [0, 1] µ R be the closed interval from 0 to 1. Then

A, given the subspace topology, is compact.
98 LECTURE 9. COMPACTNESS, I

You may use this theorem freely from now on. It will soon be superseded
by the Heine-Borel theorem, but the proof of the Heine-Borel theorem will
actually depend on Theorem 9.6.1.
In fact, there is nothing special about [0, 1]. Any closed and bounded
interval is compact as a consequence of the above theorem.
Corollary 9.6.2. Fix two real numbers a, b satisfying a < b. Then the
closed interval [a, b] (endowed with the subspace topology inherited from R)
is compact.
Proof. By Proposition 9.4.4, we are finished if we can exhibit a homeomor-
phism between [a, b] and [0, 1].
Consider the function
f : R æ R, x ‘æ (b ≠ a)x + a.
f is continuous, as we know from calculus. (This is relying on the Theorem
from previous lectures that continuity in the sense of topology is equivalent
to continuity in the sense of calculus.)
We know that, because [0, 1] is given the subspace topology, the inclusion
function i[0,1] : [0, 1] æ R is also continuous. Moreover, by Proposition 8.3.2,
the composition of conitnuous functions is still continuous, so we see that
the function
j = f ¶ iA [0, 1] : [0, 1] æ R, x ‘æ (b ≠ a)x + a
is continuous. What is the image of j? It is precisely the interval [a, b]. (Note
that j(0) = a and j(1) = b; I’ll let you fill in the rest of the details.) Thus,
by the universal property of the subspace topology of [a, b] µ R, the function
j Õ : [0, 1] æ [a, b], x ‘æ (b ≠ a)x + a
is continuous. It is straightforward to verify that j Õ is both an injection and
a surjection, hence a bijection.
In fact, you can write an inverse to j Õ as follows:
1
g : [a, b] æ [0, 1] x ‘æ (x ≠ a).
b≠a
A similar argument to the demonstration that j Õ is continuous shows that g
is also continuous.
Because j Õ is a continuous bijection whose inverse is also continuous, j Õ is
a homeomorphism.
9.7. PROOF THAT [0, 1] IS COMPACT 99

9.7 Proof that [0, 1] is compact

I will write the proof here, but you may take the above theorem for granted.
You will prove it in an Analysis class. The most important part of the proof
is an understanding of the construction of R.
First, let us recall the most “consequential”2 property of R:

Proposition 9.7.1. Let a1 , a2 , . . . be a bounded, increasing sequence of ra-

tional numbers. This means that ai Æ ai+1 for all i, and that there is some
real number A so that ai < A for all i.
Then the sequence a1 , a2 , . . . converges to some b œ R.

I state this as a proposition, but it is a direct consequence of the con-

struction of the real line as a Cauchy complete entity. I will not get much
more into this here. Indeed, the fact one can construct a set satisfying the
above Proposition is the theorem that allows us to construct the real line.

Proof of Theorem 9.6.1. Reduction I: Let us first assume that U is an open

cover of [0, 1] by open balls. More accurately, suppose that for every U œ U,
there is an x œ R and r > 0 œ R such that U = Ball(x, r) fl [0, 1]. We will
explain at the end why this reduction allows us to conclude the theorem.
Given such an open cover {U– }–œA of [0, 1], we now show that there is
a finite subcover. For every – œ A, we choose the numbers (x– , r– ) so that
U– = Ball(x– , r– ).
To begin our proof, let A1 µ A consist of those – for which 0 œ Ball(x– , r– ),
and let
a1 := sup (x– + r– ).
–œA1

If a1 > 1, we are done, for then there is some (x– , r– ) for which Ball(x– , r– )
contains both 0 and 1, hence the entire interval [0, 1]. (This exhibits an open
subcover with a single element—[0, 1] itself.)
So suppose a1 Æ 1. We let A2 µ A consist of those – for which a1 œ
Ball(x– , r– ). ANd as before, we define

a2 := sup (x– + r– ).
–œA2

2
This is a bad joke.
100 LECTURE 9. COMPACTNESS, I

Inductively, if it turns out that an < 1, we may define An+1 µ A to consist

of those – for which an œ Ball(x– , r– ), and set

an+1

to equal the lesser of 1 and sup–œAn+1 x– + r– . (So an+1 Æ 1.) Note that

an Æ an+1 . (9.7.0.1)

Why? By definition of open cover, there is some element U = Ball(x, r) in

U that contains an , so by the recentering lemma, there is some ‘ > 0 so that
Ball(an , ‘) is fully contained in U . In particular, an+1 Ø x + r > an + ‘, so
an+1 > an + ‘. In particular, (9.7.0.1) follows.
Then we have an increasing sequence a1 , a2 , . . . which is bounded (as
an Æ 1 for all i). Moreover, by choosing rational numbers rn so that
an < rn < an+1 , we obtain a real number b to which r1 , r2 , . . . converge by
Proposition 9.7.1. I promise it is straightforward to check that the sequence
a1 , a2 , . . . also converges to b.
On the other hand, [0, 1] is closed. (It is also straightforward to see
that the complement is open.) Hence b œ [0, 1]. But by definition of our
open cover, there is some (x, r) for which Ball(x, r) contains b. Further by
definition of convergence, this means that for every n large enough, an œ
Ball(x, r), contradicting the definition of an+1 unless an+1 = 1. In other
words, for n large enough, all the an equals 1. So there is only a finite
collection of open numbers 0 = a0 , a1 , a2 , . . . , an = 1 picked out by the above
process.
Now choose, for each i, a pair (xi , ri ) so that Ball(xi , ri ) contains ai
and Ball(xi , ri ) œ U. By definition of the ai , we can choose these so that
Ball(xi , ri ) intersects Ball(xi+1 , ri+1 ) for each i. In particular, these form a
finite subcover.
Now let us explain why Reduction One is enough to complete the proof.
Given an arbitrary open cover V = {V— }—œB , we may consider a much bigger
open cover U = {U– }–œA consisting of only open balls, satisfying the property
that for every open ball U– œ U, there exists some — such that U– µ V— . (For
example, A could consist of triplets (x, r, —) for which Ball(x, r) µ V— .) Then
Reduction One allows us to choose a finite collection –1 , . . . , –n œ A for which
{U–1 , . . . , U–n } forms a subcover. Then, for each i, choosing a —i such that
V—i ∏ U–i , we see that {V—1 , . . . , V—n } forms a subcover of V.