General/Point-Set Topology Lecture Notes

Thomas Baird
Winter 2020

Contents
1 Introduction 2

2 Sets and Relations 4

2.1 Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.1 Order Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.2 Equivalence Relations . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Cartesian Products and Coproducts . . . . . . . . . . . . . . . . . . . . . . 8
2.2.1 Coproducts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2.2 Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3 Topological Spaces 11
3.1 The Subspace Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.2 Closed Sets and Limit Points . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.3 Convergence and the Hausdorff Condition . . . . . . . . . . . . . . . . . . 17
3.4 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.5 Patching Continuous Functions Together . . . . . . . . . . . . . . . . . . . 19
3.6 The Initial Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.7 Product Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.8 The Final Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.9 Quotient Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.10 Coproducts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4 Metric Topologies 26
4.1 Function Spaces and Uniform Convergence . . . . . . . . . . . . . . . . . . 29
4.2 Metric Spaces and Continuity . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.3 Complete Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.4 Completing Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

5 Connected Spaces 38
5.1 Path-Connected Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

1
6 Compactness 44
6.1 Compact Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
6.2 Compactness and Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
6.3 The One-Point Compactification . . . . . . . . . . . . . . . . . . . . . . . . 51
6.4 Compact-Open Topology (skipped in 2017) . . . . . . . . . . . . . . . . . . 54

7 Advanced Material 55
7.1 Separation Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
7.2 Urysohn Lemma and Metrization Theorem . . . . . . . . . . . . . . . . . . 56
7.3 Tychonoff’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

1 Introduction
Topology is the study of continuity. We begin with a familiar definition of continuity.
Definition 1. A function f : R → R is continuous if ∀x ∈ R and ∀ > 0, ∃ δ > 0 such
that f (Bδ (x)) ⊆ B (f (x)).
Here we use notation, Bδ (x) = {x0 ∈ R||x − x0 | < δ}, called the open interval (or ball)
of radius δ centred at x.

The quantity |x − x0 | should be thought of as the “distance” between x and x0 . This

definition of continuity generalizes to sets equipped with an appropriate notion of distance.
Definition 2. Let X be a set. A metric on X is a function
d : X × X → R≥0
called the distance or metric function, satisfying the following axioms for x, x0 , x00 ∈ X
1. d(x, x0 ) = 0 ⇔ x = x0 (d separates points)
2. d(x, x0 ) = d(x0 , x) (d is symmetric)
3. d(x, x00 ) ≤ d(x, x0 ) + d(x0 , x00 ) (the triangle inequality)

A set X equipped with a metric d is called a metric space.

2
Example 1. X = Rn with the Euclidean metric d(~a, ~b) :=
pPn
2
i=1 (ai − bi ) .

Example 2. X = Rn with the taxi-cab metric d(~a, ~b) := ni=1 |ai − bi |.

P

Example 3. X = Cont([0, 1], R) the set of real valued continuous functions on the unit
interval, with Lp -metric Z 1
d(f, g) := ( |f − g|p )1/p ,
0
where p ≥ 1 is a fixed real number.

Example 4. Given p ∈ {2, 3, 5, 7, 11} a prime number, let X = Z with p-adic metric

d(m, n) := p−k

where pk is the largest power of p dividing m − n ( if m = n, then d(m, n) = p−∞ = 0 ).

Definition 3 (version I). Let (X, dX ) and (Y, dY ) be metric spaces. A map f : X → Y
is continuous if ∀x ∈ X, ∀ > 0, ∃δ > 0 such that

f (Bδ (x)) ⊂ B (f (x)).

where Bδ (x) = {x0 ∈ X|dX (x0 , x) < δ} and B (y) = {y 0 ∈ Y |dY (y 0 , y) < }.

This is where topology stood before the advent of topological spaces.1

Definition 4. Let (X, d) be a metric space. A subset U ⊂ X is called open if ∀x ∈ U ,

∃ > 0 such that B (x) ⊂ U .

Example 5. In Rn the Euclidean and taxi-cab metrics give rise to the same class of open
sets.

Definition 5 (version II). A map f : X → Y between metric spaces is continuous if and

only if the pre-image of all open sets in Y are open in X, i.e.

If U ⊆ Y is open, then f −1 (U ) := {x ∈ X|f (x) ∈ U } ⊆ X is open.

Exercise 1. Show that these two definitions of continuity are equivalent.

Observe that the second definition of continuity makes no explicit mention of the
metric. Once we have determined which sets are open, the metric can be discarded. Thus
this definition makes sense for the more general class of topological spaces defined below.
1
The modern point-set definition of a topological space was introduced by Felix Hausdorff in a very
influential book “The Principles of Set Theory” in 1914.

3
Definition 6. A topological space (X, τ ) is a set X and a collection τ of subsets of X,
called the open sets, satisfying the following conditions:

i) ∅ and X are open,

ii) Any union of open sets is open,
iii) Any finite intersection of open sets is open.
Exercise 2. Show that the open sets of a metric space determine a topology. This is
called the metric topology.
What do we gain from this abstraction?

I. Different metrics may induce the same topology

• Can choose a metric suited to particular purpose
• Metrics may be complicated, while the topology may be simple
• Can study families of metrics on a fixed topological space
II. Many interesting topologies do not come from metrics
• Zariski topology on algebraic varieties and schemes (algebra and geometry)
• The weak and weak* topologies on a normed vector space (analysis)
• Any interesting topology on a finite set (combinatorics)

2 Sets and Relations

We adopt a naive point of view on set theory, and assume that what is meant by a set is
intuitively clear.
Let X be a set. A subset A ⊆ X is a set whose elements all belong to X.
Let I be an index set (which may be finite, infinite or uncountable). For each i ∈ I,
let Ai ⊆ X be a subset. Define the union to be
[
Ai := {x ∈ X|x ∈ Ai for some i ∈ I}
i∈I

and the intersection \

Ai := {x ∈ X|x ∈ Ai for all i ∈ I}
i∈I

Proposition 2.1. Let I and Ai be as above, and let B ⊆ X. Then intersections distribute
over unions and vice-versa. I.e.
[
[
i) B∩ Ai = (B ∩ Ai )
i∈I i∈I
\
\
ii) B∪ Ai = (B ∪ Ai )
i∈I i∈I

4
Proof. i)
[
[

p∈B∩ Ai ⇔ p ∈ B and p ∈ Ai
i∈I i∈I
⇔ p ∈ B and p ∈ Ai , for some i ∈ I
⇔ p ∈ B ∩ Ai , for some i ∈ I
[
⇔ p ∈ (B ∩ Ai )
i∈I

ii) is similar.
The complement A ⊂ X, denoted Ac is defined

Ac := X \ A = {p ∈ X|p ∈
/ A}.
Observe that (Ac )c = A.
Proposition 2.2 (De Morgan’s Laws). Let I and Ai be as above. The following identities
hold. [
c \
i) Ai = Aci
i∈I i∈I
\
c [
ii) Ai = Aci
i∈I i∈I

Proof. i)
[
c [
p∈ Ai ⇔ p 6∈ Ai
i∈I i∈I
⇔ p 6∈ Ai for all i ∈ I
⇔ p ∈ Aci for all i ∈ I
\
⇔ p∈ Aci .
i∈I

ii) is equivalent to i) (Prove this).

Let f : X → Y be a map of sets. The image (or forward image) of a set A ⊆ X is
defined

f (A) = {f (x) ∈ Y |x ∈ A} ⊆ Y.
The preimage of a set B ⊆ Y is defined

f −1 (B) := {x ∈ X|f (x) ∈ B} ⊆ X.

Proposition 2.3. Preimages commute with unions, intersections, and complements. That
is, given subsets Bi ⊆ Y

f −1 i∈I Bi = i∈I f −1 (Bi )

S
S
(i)

5
 f −1 = i∈I f −1 (Bi ),
T
T
(ii) i∈I Bi

(iii) f −1 (B)c = f −1 (B c )

Proof. i)

[ [
p ∈ f −1


Bi ⇔ f (p) ∈ Bi
i∈I i∈I
⇔ f (p) ∈ Bi for some i ∈ I
⇔ p ∈ f −1 (Bi ) for some i ∈ I
[
⇔ p∈ f −1 (Bi )
i∈I

ii) and iii) are similar.

We remark that the same property does not hold true for forward images. For example
consider the case where X = {x1 , x2 } consists of two points, Y = {y} is a single point and
f : X → Y is the (unique) map. If A1 = {x1 } and A2 = {x2 }, then f (A1 ∩ A2 ) = f (∅) = ∅
but f (A1 ) ∩ f (A2 ) = Y .

2.1 Relations
A relation on a set X a subset of R ⊂ X × X. We use notation xRy to denote (x, y) ∈ R
and x 6 Ry (this is meant to be a bar through R) to denote (x, y) 6∈ R. We consider two
general types of relations: order relations and equivalence relations.

2.1.1 Order Relations

Definition 7. A partial order on a set X is a relation ≤ satisfying the following axioms
for all x, y, z ∈ X

i) x ≤ x (reflexivity)

ii) x ≤ y and y ≤ x implies x = y (anti-symmetry)

iii) x ≤ y and y ≤ z implies x ≤ z (transitivity).

Example 6. The real numbers R and the integers Z with the usual ≤.
Example 7. For any set X, the power set ℘(X) is the set of all subsets of X. The subset
relation A ⊆ B is a partial order on ℘(X).
Example 8. Any subset of a partially ordered set under the restriction of ≤.
Definition 8. We say that x ∈ X is maximal (respectively, minimal ) with respect to a
partial order ≤ if , for all y ∈ X, x ≤ y implies x = y (respectively, y ≤ x implies x = y ).

6
Example 9. X = {linearly independent subsets of Rn } ordered by inclusion. Then the
maximal elements of X are bases of Rn while the only minimal element is the empty set.

Definition 9. A simple or linear order is a partial order satisfying the additional condi-
tion that for all x, y ∈ X either x ≤ y or y ≤ x.

Example 10. (R, ≤) and all of its subsets.

Definition 10. A linear order (X, ≤) is called well-ordered if every non-empty subset of
X contains a minimal element.

Example 11. Z≥0 = 0, 1, 2, 3, ... is well-ordered.

Example 12. Z, R, R≥0 , [0, 1] are not well-ordered.

Well-ordering Principle (Zermelo) Every set can be well-ordered.

The proof of this theorem is long, but not hard. It uses and is equivalent to,

The Axiom of Choice: Let {Ai }i∈I be a collection of disjoint, nonempty sets. Then
there exists a set B containing exactly one element from each Ai .

The axiom of choice is a part of the most commonly accepted axiomatic foundation
of set theory known as ZFC after Zermelo, Fraenkel and choice. The Axiom of Choice
seems innocuous, but leads to some bizarre consequences.
Both of these are equivalent to a third axiom called Zorn’s Lemma. Before stating
Zorn’s Lemma we need another definition.

Definition 11. Let (X, ≤) be a partially ordered set and let A ⊆ X. An upper bound
(respectively lower bound ) of A in X is an element x ∈ X such that for all a ∈ A, a ≤ x
(resp. x ≤ a).

Zorn’s Lemma: Let (X, ≤) be partially ordered. Suppose that every linearly ordered
subset of X has an upper bound in X. Then X contains a maximal element.

Theorem 2.4. Every vector space V has a basis.

Proof. Let X be the set of all linearly independent subsets of V , ordered by inclusion. A
basis of V is simply a maximal element in X. By Zorn’s Lemma, it is enough to show:
Claim:
S Suppose that {Si }i∈I ⊆ X is linearly ordered by inclusion. Then the union
S := i∈I Si is in X.
Suppose that v1 , ..., vn ∈ A. Then each vk ∈ Sik for some ik . Since {Si1 , ..., Sin } is finite
and linearly ordered, one of the Sik must contain all the of others. Thus v1 , ..., vn ∈ Sik
must be linearly independent and we conclude that S ∈ X.

7
2.1.2 Equivalence Relations
Definition 12. An equivalence relation on a set X is a relation ∼ satisfying, for all
x, y, z ∈ X

i) x ∼ x (reflexivity)

ii) x ∼ y implies y ∼ x (symmetry)

iii) x ∼ y and y ∼ z implies x ∼ z (transitivity)

Given x ∈ X, the equivalence class of x is

[x] := {y ∈ X|x ∼ y}
Notice that [x] = [y] if and only if x ∼ y. The equivalence classes determine a partition
of X into a union of disjoint sets (prove this).
Let E := {[x] | x ∈ X} be the set of equivalence classes. There is a canonical map

Q : X → E, x 7→ [x]

called the quotient map.

Given any relation R on X, it is possible to generate an equivalence relation ∼R by
the rule x ∼R y if and only if there exists a finite sequence {xi ∈ X}ni=0 for n ≥ 0 satisfying

(i) x0 = x,
(ii) xn = y and,
(iii) xi Rxi−1 or xi−1 Rxi for all i = 1, ..., n.

Exercise 3. Prove that ∼R is an equivalence relation.

2.2 Cartesian Products and Coproducts

Given a collection of sets {Xα }α∈I , one may form the product set or Cartesian product:
Y
Xα = {(xα )|xα ∈ Xα for all α ∈ I}
α∈I

whose elements are I-tuples (xα ). This is a pretty clear concept when I is finite, but what
about when I is infinite? or uncountable?
It is helpful to begin with a related concept.

2.2.1 Coproducts
Definition 13. Given a collection of sets {Xα }α∈I , the coproduct (or disjoint union) is
the set a a
Xα = Xα := {(x, α)| α ∈ I, and x ∈ Xα }.
α∈I

8
The coproduct of a finite collecton of sets {X1 , X2 , ..., Xn } may also be written as X1 q
X2 q ... q Xn .
Example 13. The coproduct {a, b} and {b, c} consists of four elements

{a, b} q {b, c} = {(a, 1), (b, 1), (b, 2), (c, 2)}.

Example 14. The coproduct R and R consists of two copies of the real line

R q R = {(x, i)|x ∈ R, i = 1, 2}.

Given β ∈ I, the map a

iβ : Xβ ,→ Xα
defined by iβ (x) = (x, β) is called an inclusion map. It is quite common to informally
think of the Xβ as subsets of the coproduct which form a partition.
The coproduct satisfies (and can be defined by) the following universal property.
Proposition 2.5. Suppose that Y is a set and let {gα : Xα → Y }α∈I be a collection of
functions. Then there is a unique function
a
g: Xα → Y

such that for all α ∈ I, gα = g ◦ iα . Equivalently, the following diagram commutes for all
β∈I
iβ
/
`
Xβ Xα
g
gβ
' 
Y
This property characterizes the coproduct up to a uniquely determined isomorphism.
`
Proof. The map g : Xα → Y is simply defined by

g((x, α)) = gα (x).

It is pretty clear that g ◦ iα = gα and that this determines g uniquely.

Now suppose that Z is a set equipped with maps jα : Xα → Z possessing the universal
property. Then there exist uniquely determined maps
a
i:Z→ Xα

and a
j: Xα → Z
such that for every β ∈ I
jβ = j ◦ iβ
and
iβ = i ◦ jβ .

9
Composing, we have iβ = i ◦ j ◦ iβ so by the uniqueness property
i ◦ j = Id` Xα .
Similarly jβ = j ◦ i ◦ jβ so
j ◦ i = IdZ .
Thus i and j are inverse isomorphisms.
Since the coproduct is uniquely determined by its universal property, the universal
property provides a kind of definition of what the coproduct is. This universal property
defining the coproduct is categorical in nature (speaks only of equalities and compositions
of morphisms between objects) so it can be used to define coproducts of topological spaces,
groups, modules, etc..
The argument we used to show that the a universal property determines the coprod-
uct up to uniquely determined isomorphism, is called proof by abstract nonsense (this
expression is often attributed to Saunders MacLane). This argument remains valid for
coproducts in other categories as well.

2.2.2 Products
Definition 14. Given a collection of sets {Xα }α∈I , an I-tuple is a function
a
x:I→ Xα
α∈I
Q Q
such that x(α) ∈ Xα . The product set, denoted α∈I Xα or Xα , is the set of I-
tuples. The product of a finite collecton of sets {X1 , X2 , ..., Xn } may also be written as
X1 × X2 × .... × Xn . The product of a set with itself n times may be written X n .
When working with I-tuples, it is more common to write xα instead of x(α), and to
represent an I-tuple x by a list of it’s values (xα ).
For every β ∈ I, define the projection map
Y
πβ : Xα → Xβ , πβ (x) = xβ
α∈I

The product set satisfies (and can be defined by) the following universal property.
Proposition 2.6. Suppose Y is a set and let {fα : Y → Xα }α∈I be a collection of maps.
Then there exists a unique map Y
f :Y → Xα
α∈I
satisfying fα = πα ◦ f . Equivalently, the following diagram commutes for all β ∈ I:

f
/
Q
Y Xα
πα
fβ
' 
Xβ
Q
This property characterizes α∈I Xα up to uniquely determined isomorphism.

10
 Q
Proof. The map f : Y → α∈I Xα is defined by f (y)α = fα (y). That this is the unique
function satisfying fα = πα ◦ f is pretty clear. That the product set is uniquely charac-
terized by this property follows by abstract nonsense.
We note that the universal property possessed by the product is categorical in nature
(speaks only of equalities and compositions of objects and morphisms) so it translates to
define products of topological spaces, groups, modules, etc..

3 Topological Spaces
A topological space or simply a space consists of a set X and a collection τ of subsets of
X, called the open sets, such that

1. ∅ and X are open,

2. Any union of open sets is open,

3. Any finite intersection of open sets is open.

It is conventional to denote a topological space (X, τ ) simply by X when there is unlikely

to be confusion about the topological structure (this will not be the case today).
The following topologies exist on any set X.
Example 15. Trivial topology: τ = {∅, X}.
Example 16. Discrete topology: τ = ℘(X) (the power set).
Example 17. Finite complement topology: U ∈ τ ⇔ X \ U is a finite set or U = ∅.
Definition 15. Suppose that τ and τ 0 are two topologies on a set X. If τ ⊆ τ 0 then we
say τ is coarser than τ 0 , and τ 0 is finer than τ .
These relations determine a partial order on the set of all topologies on an underlying
set X.
Example 18. The coarsest topology on X is the trivial topology. The finest topology on
X is the discrete topology.
It is usually inconvenient to define a topology by specifying each and every open set.
Instead, we usually “generate” the topology from a special collection of open sets.
Definition 16. A subbase (for a topology) on a set X is a collection of subsets S ⊆ ℘(X)
whose union is X. 2
Proposition 3.1. Let S be a subbase for a topology. Define τ ⊆ ℘(X) to be the collection
of all unions of finite intersections of sets in S. Then τ is a topology, called the topology
generated by S.
Given an arbitrary collection S 0 of subsets of X, we can make a subbase S = S 0 ∪ {X}. If S 0 was
2

already a subbase then S 0 and S generate the same topology.

11
Proof. We must verify that τ satisfies the axioms.

1) The intersection of the empty collection is empty and X ∈ τ because it is the union of
sets in S.
2) A union of unions of finite intersections is a union of finite intersections.
3)It is enough to prove that the intersection of any two elements U, V ∈ τ S lies in τ . If
eitherSU or V equals X then there is noting to prove. Otherwise, U := i∈I Ui and
V := j∈J Vj where Ui and Vj are finite intersections of sets in S for all i ∈ I and j ∈ J.
Then
[
U ∩ V = ( Ui ) ∩ V
i∈I
[
= (Ui ∩ V )
i∈I
[ [
= (Ui ∩ ( Vj ))
i∈I j∈J
[[
= (Ui ∩ Vj )
i∈I j∈J
[
= (Ui ∩ Vj )
(i,j)∈I×J

is a union of finite intersections, hence is an element of τ .

Remark 1. Observe that the topology generated by S is the coarsest topology on X for
which every element of S is open.

The fact that any subbase generates a topology is useful theoretically, but it can be
difficult in practice to determine whether a given set is open or not by the above definition.
Thus we often use this intermediate approach.

Definition 17. A base for a topology on X is a collection β of subsets of X such that

(i) β is a subbase (i.e the union of elements in β is X)

(ii) For any pair of sets U1 , U2 ∈ β, and any point p ∈ U1 ∩ U2 , there exists a set U3 ∈ β
such that p ∈ U3 ⊆ U1 ∩ U2 .

Definition 18. The topology generated by the base β is the collection τ of subsets U ⊆ X
satisfying any of the following equivalent conditions.

(i) U is a union of finite intersections of elements of β (same as for subbase).

12
(ii) U is a union of elements of β.

(iii) For all p ∈ U there exists V ∈ β such that p ∈ V ⊆ U .

Exercise 4. Show that (i), (ii), (iii) are equivalent for β a base.

The value in considering topologies generated by a base, rather than an arbitrary

subbase, is that now we can use any of conditions (i), (ii) and (iii) to check if a set is
open.

Example 19. Let X = Rn and let β := {B (p)|p ∈ X,  > 0}. Then β is the base for a
topology, called the Euclidean topology on X. More generally, for a metric space, the set
of open balls forms a base for the metric topology. We will prove this in a when we return
to metric spaces later in the course.

A sufficient condition for a subbase to be a base is that it be closed under finite

intersections, i.e. if U1 , U2 ∈ S then U1 ∩ U2 ∈ S. This is true in the following examples.

Example 20. Given a subbase S, define β to be the set of finite intersections of elements
in S. Then β is a base for the topology.

Example 21. Let (X, ≤) be a totally ordered set. Given a, b ∈ X, define

(a, b) := {x ∈ X | a ≤ x ≤ b and x 6= a, b}.

The set β := {(a, b) | a, b ∈ X} ∪ {X} is a base. The topology it generates is called the
order topology on X.

Proof. Since X ∈ β, the union of elements of β is X. Given (a, b), (a0 , b0 ) ∈ X, then
either (a, b) ∩ (a0 , b0 ) = ∅ or (a, b) ∩ (a0 , b0 ) = (a00 , b00 ) where a00 = max{a, a0 } and b00 =
min{b, b0 }.

Example 22. Let X = Cn and let C[x1 , ..., xn ] denote the set of polynomial functions
on Cn . Given f ∈ C[x1 , ..., xn ], let Uf := {p ∈ Cn |f (p) 6= 0}. Then the set β := {Uf |f ∈
C[x1 , ..., xn ]} is a base. The topology it generates is called the Zariski topology. 3

Proof. For complex valued functions f, g we have f (p)g(p) = 0 if and only if f (p) = 0 or
g(p) = 0. Thus Uf ∩ Ug = Uf g and β is actually closed under intersection.

Remark 2. When n = 1, the Zariski topology on C coincides with the finite complement
topology (Prove this).
3
Similarly, we can define the Zariski topology on k n for any field k.

13
3.1 The Subspace Topology
Let (X, τ ) be a topological space, and A ⊆ X a subset. Then A inherits a topological
structure τA from (X, τ ) defined by

τA := {U ∩ A|U ∈ τ }.

We call τA the subspace topology.

Proposition 3.2. The subspace topology τA is a topology on A.
Proof. 1) ∅ = ∅ ∩ A and A = X ∩ A are elements of τA .

Now let {Uα ∩ A}α∈I be an arbitrary collection of elements of τA , where Uα are open
sets in X. Then
S S
2) α∈I (Uα ∩ A) = ( α∈I Uα ) ∩ A ∈ τA because intersections distribute over unions.
T T
3) If I is finite, then α∈I (Uα ∩ A) = ( α∈I Uα ) ∩ A ∈ τA because intersections com-
mute with intersections.
Example 23. Any subset of Rn inherits a Euclidean topology.
Example 24. Any subset of Cn inherits a Zariski topology.
The subspace topology can also be defined using a universal property. Let i : A ,→ X
be the inclusion map of sets.
Proposition 3.3. The subspace topology on A is the coarsest topology for which the
inclusion map is continuous.
Proof. Recall that a map is continuous if the pre-image of open sets is open. Let τA0 be
a topology on A and suppose that i : (A, τA0 ) ,→ (X, τ ) is continuous. Then for any open
set U ⊆ X, the preimage i−1 (U ) = U ∩ A must be continuous. So by the definition of the
subspace topology τA , we have τA ⊆ τA0 .
Proposition 3.4. Suppose that X is a space with topology generated by a subbase S ⊆
℘(X) and let A ⊆ X. The subspace topology on A is generated by the subbase SA :=
{U ∩ A | U ∈ S}. If S is a base, then SA is also a base.
Proof. Clearly SA is a subbase and the elements of SA are open in A, so the topology
generated by SA is coarser than subspace topology. It remains to show that all open sets
in A are unions of finite intersections of elements of SA .
The general open set in X has the form ∪α∈I ∩nj=1 α
Bα,j where Bα,j ∈ S. Thus the
general open set in A has the form

A ∩ (∪α∈I ∩nj=1
α
Bα,j ) = ∪α∈I ∩nj=1
α
(A ∩ Bα,j ),
where (A ∩ Bα,j ) is in SA .
The proof that if S is a base then SA is a base is left as an exercise.

14
Definition 19. A subset A ⊆ X is called a discrete subset if A the subspace topology
equals the discrete topology.

Definition 20. A continuous function f : X → Y is called an embedding if it is injective

and restricts to a homeomorphism from X to the subspace f (X) ⊆ Y .

3.2 Closed Sets and Limit Points

Definition 21. A subset A ⊆ X of a topological space is called closed if the complement
Ac = X \ A is open.

Remark 3. Closed is not the same as “not open”. Some subsets are both open and
closed (e.g. the empty set) and some sets are neither neither open, nor closed (e.g. the
half open interval in R in the Euclidean topology).

Proposition 3.5. Let X be a topological space. Then

(i) ∅ and X are closed.

(ii) Arbitrary intersections of closed sets are closed.

(iii) Finite unions of closed sets are closed.

Proof. (i) (∅)c = X and X c = ∅ are open.

The remaining cases follow by de Morgan’s laws

(ii) If Aα is closed then \
c [

Aα = (Aα )c
α∈I α∈I

is open.

(iii) If Ai is closed then

n n
[
c \
Ai = (Ai )c
i=1 i=1

is open.
Of course, if one knows the closed sets of a topological space, one can recover the open
sets by taking complements. Thus we may equivalently define a topology on X to be a
class of closed sets satisfying properties i), ii) and iii) of Proposition 3.5. This point of
view is sometimes more convenient.
For A ⊆ X a subset of a topological space, the closure of A is the subset of X
\
Ā := C
A⊆C, C closed

15
 Observe that Ā is closed because it is the intersection of closed sets. Thus we may
also describe Ā as the smallest closed set containing A. Note that A is closed if and only
if A = Ā.
Similarly, one defines the interior of A to be the largest open set contained in A.
[
int(A) := U = (Ac )c
U ⊆A, U open

Example 25. Let (a, b] ⊂ R with the Euclidean topology. Then int((a, b]) = (a, b) and
(a, b] = [a, b].

Example 26. Let (0, 1] ⊂ (0, 2] with the Euclidean subspace topology. Then (0, 1] = (0, 1]
because (0, 1] is already closed as a subset of (0, 1]. To see this observe that (0, 1] =
[−1, 1] ∩ (0, 2] and that [−1, 1] is closed in R.

Example 27. The subset of integers Z ⊂ C has closure Z = C in the Zariski/finite-

complement topology because the only infinite closed set is C.

An (open) neighbourhood (often shortened open nbhd ) of a point p ∈ X is an open set

U containing p.

Definition 22. Let A be a subset of a topological space X. An element p ∈ X is called

a limit point of A if for all open nbhds p ∈ U we have

(U \ {p}) ∩ A 6= ∅

Example 28. If A = (a, b] ⊂ R, then the set of limit points is [a, b]

Example 29. If A = {1/n|n = 1, 2, 3, ...} ⊂ R, then the only limit point is 0.

Example 30. Suppose that X has the trivial topology and A = {p} is a one element
subset. Then every point *except* p is a limit point for A (prove this).

Theorem 3.6. Let A be a subset of a topological space X and let A0 be the set of limit
points of A. Then
A0 ∪ A = Ā

Corollary 3.7. A is closed if and only if it contains all of its limit points.

Proof of Theorem. Suppose p 6∈ Ā. Then (Ā)c is a nbhd of p and (Ā)c ∩ A = ∅ so p 6∈ A0

and consequently p 6∈ A ∪ A0 .
Now suppose that p 6∈ A ∪ A0 . Then since p 6∈ A0 , there exists a nbhd U of p with
(U \ {p}) ∩ A = ∅. Since p 6∈ A, it follows that U ∩ A = ∅, so U c ⊇ A. Since U c is closed
we have U c ⊇ Ā so p 6∈ Ā.

16
3.3 Convergence and the Hausdorff Condition
A sequence (xn )∞ n=1 of points in a topological space X is said to converge to a limit L ∈ X
if for every open nbhd U of L the tail of the sequence lies in U . That is, for every U there
exists an integer N such the xn ∈ U for n > N .
It is possible for a sequence to converge to more than one point. For example, in the
trivial topology every sequence converges to every point!
Definition 23. A topological space X is called Hausdorff if for any two distinct points
p1 , p2 ∈ X, there exists open nbhds U1 , U2 of p1 , p2 respectively, such that U1 ∩ U2 = ∅.
We say that a Hausdorff topology separates points.
Proposition 3.8. A sequence (xn )∞
n=1 in a Hausdorff space can converge to at most one
point.
Proof. Let (xn )∞ 0
n=1 converge to L, and choose L 6= L. Then there exist disjoint open
nbhds U and U 0 of L and L0 respectively. Since the tail of the sequence lies in U it cannot
lie in U 0 .
Example 31. The Euclidean topology on Rn is Hausdorff, because p1 , p2 ∈ Rn with
d(p1 , p2 ) =  > 0, then B/3 (p1 ) and B/3 (p2 ) are disjoint open nbhds.
Example 32. The Zariski topology on Cn is not Hausdorff. In fact any two non-empty
open sets intersect. To see this, it is enough to show that any two basic open sets intersect.
The product of non-zero polynomials is non-zero, so Uf1 ∩ Uf2 = Uf1 f2 is empty if and
only if f1 f2 = 0 if and only if f1 = 0 or f2 = 0.
Proposition 3.9. Subspaces of Hausdorff spaces are Hausdorff.
Proof. Let A ⊂ X and X is Hausdorff. If p1 , p2 ∈ A are distinct points, then there exist
open nbhds U1 , U2 ⊂ X which separate p1 and p2 . Then U1 ∩ A and U2 ∩ A are open
nbhds in A separating p1 and p2 .
Proposition 3.10. Finite subsets of Hausdorff spaces are closed.
Proof. It is enough to prove that one element sets are closed (because finite unions of
closed sets are closed). Let p ∈ X. For every q ∈ X distinct from p, there exists an open
set Uq disjoint from a nbhd of p, so p 6∈ Uq . Thus
[
X \ {p} = Uq
q∈X\{p}

is open, so {p} is closed.

The argument above used an idea that is worth stating explicitly.
Proposition 3.11. Let X be a topological space. A subset V ⊆ X is open if and only if
for all p ∈ V , there exists an open set Up ⊆ X such that p ∈ Up ⊆ U .
Proof. Exercise.

17
3.4 Continuity
Definition 24. Let X, Y be topological spaces. A map f : X → Y is continuous if for
all open sets U ⊆ Y the pre-image f −1 (U ) is open in X. Equivalently, f is continuous if
for all closed sets C ⊂ Y the pre-image f −1 (C) is closed.

Example 33. Let τ and τ 0 be two different topologies on a set X. The identity function
Id : X → X is continuous as a map from (X, τ ) to (X, τ 0 ) if and only if τ is finer than τ 0 .

Proposition 3.12. If the topology of Y is generated by a subbase S then f : X → Y is

continuous if and only if f −1 (B) is open for all B ∈ S.

Proof. A general open set in Y is a union of finite intersections of sets in S. That is of

the form nα
[ \ 
U= Bα,i
α∈I i=1

where each Bα,i ∈ S. Taking the preimage commutes with unions and intersections so:
nα
[ \ nα

 [ \ 
f −1 (U ) = f −1 Bα,i = f −1 (Bα,i )
α∈I i=1 α∈I i=1
−1
which is open if f (Bα,i ) is open.

Exercise 5. Show that a map f : Rm → Rn is continuous in the Euclidean topology iff

it is continuous in the  − δ sense.

Definition 25. A continuous map f : X → Y is called a homeomorphism 4 if it is bijective

and the inverse map f −1 : Y → X is also continuous. In this case we say X and Y are
homeomorphic (this is the notion of isomorphism in the category of topological spaces).

Equivalently, a homeomorphism is a bijection f : X → Y with the property that

U ⊆ X is open if and only if f (U ) ⊆ Y is open.

Example 34. Let f : (−1, 1) → R sending f (x) = tan(πx/2). Then f is a homeomor-

phism, with inverse f −1 (y) = (2/π)tan−1 (y). (Proof: Both f , f −1 send open intervals to
open invervals).

4
The word homeomorphism comes from the Greek words homoios = similar and morph = shape or
form.

18
Example 35. Let τ and τ 0 be two different topologies on a set X. The identity map is
a homeomorphism between (X, τ ) and (X, τ 0 ) if and only if τ = τ 0 .

Theorem 3.13. Let X, Y, Z be topological spaces.

a) Constant maps: are continuous.

b) Inclusion maps: If A ⊆ X is a subspace, then the inclusion map is continuous

c) Composition: If f : X → Y and g : Y → Z are continuous, then g ◦ f is con-

tinuous.

d) Restriction of domain: If A ⊆ X is a subspace, and f : X → Y is continu-

ous, then the restriction f |A : A → Y is continuous.

e) Restriction of codomain: If B ⊆ Y is a subspace and f : X → Y is contin-

uous with f (X) ⊆ B then the restricted map g : X → B defined by f (x) = g(x) is
continuous.

Proof. a) If f : X → Y is constant, and U ⊂ Y then f −1 (U ) = ∅ or X, so the preimage

of any set is open.

b) Done previously.

c) Suppose U ⊂ Z is open. Since g is continuous, g −1 (U ) ⊂ Y is open. Since f is

continuous, (g ◦ f )−1 (U ) = f −1 (g −1 (U )) ⊂ X is open.

d) This follows from b) and c), because f |A = f ◦ i is the composition of f with the
inclusion map i : A ,→ X.

e) The general open set in B is of the form U ∩ B for U open in Y , and g −1 (U ∩ B) =

f −1 (U ∩ B) = f −1 (U ) is open.

3.5 Patching Continuous Functions Together

Lemma 3.14. Let X be a topological space, B ⊆ A ⊆ X a chain of subspaces.
(i) If A is open in X, then B is open in A iff B is open in X.
(ii) If A is closed in X, then B is closed in A iff B is closed in X.

To see why this condition on A is necessary, consider the case X = R with Euclidean
topology, A = [0, 1], A0 = (1/2, 3/2) and B = A ∩ A0 = (1/2, 1]. Then B is open as a
subset of A, is closed as a subset of A0 but is neither open nor closed as a subset of R.
Proof. We prove (i) only, because (ii) is similar. Certainly if B is open in X, then
B ∩ A = B is open in A by definition of the subspace topology.

19
 Now suppose A is open in X. If B is open in A then by the definition of the subspace
topology, B = U ∩ A for some U open in X. But then B is an intersection of open sets
in X, hence open in X.
Theorem 3.15. Let X be a topological space and let {Uα }α∈I be an open cover (i.e. the
Uα are open sets and their union is X). A map f : X → Y is continuous if and only if
its restrictions fα := f |Uα : Uα → Y are continuous for all α ∈ I.
Proof. We proved already that if f is continuous, then its restrictions fα are continuous.
Now assume the fα are all continuous. Let V ⊆ Y be an open set. Then

[
f −1 (V ) = f −1 (V ) ∩ ( Uα )
α∈I
[
= (f −1 (V ) ∩ Uα )
α∈I
[
= fα−1 (V ).
α∈I

The fα−1 (V ) are open in Uα hence open in X by Lemma 3.14. Thus f −1 (V ) is a union of
open sets, hence open.
Theorem 3.15 allows us to define continuous maps locally. That is, if we have a
collection of continuous maps fα defined on each Uα which agree on overlaps, then we can
patch them together to define a map on X. A similar results is valid for closed covers:
Exercise 6. Let X be a topological space and {Ci }ni=1 be a finite closed cover. Show that
f : X → Y is continuous if and only if f |Ci is continuous for all i.

3.6 The Initial Topology

Definition 26. Let X be a set and let {fα : X → Yα }α∈I a collection of functions, where
the Yα are topological spaces. The initial topology on X is the coarsest topology for which
the fα are all continuous.
Explicitly, the initial topology on X is the topology generated by the subbase

S := {fα−1 (Uα ) | Uα ⊆ Yα is open}

Example 36. The subspace topology of A ⊆ X is the initial topology with respect to
inclusion i : A ,→ X.
Example 37. Let V be normed vector space over F = R or F = C (e.g. a Hilbert space),
and let {L : V → F } be the set of bounded linear functionals. The initial topology is
called the weak topology on V (this is coarser than the norm (metric) topology).
Example 38. Let X = Cn and {f : Cn → C} be the set of polynomial functions, and
give C the finite complement topology. Then the initial topology with respect to these
maps is the Zariski topology.

20
3.7 Product Topology
Definition 27. Let {Xα } be a collection of topological spaces. The product space, denoted
Y
Xα
α∈I

is the topological space with underlying set the cartesian product set, equipped with the
initial topology with respect to the projection maps
Y
πβ : Xα → Xβ ,
α∈I

for all β ∈ I.
The product topology is generated by the subbase of open sets of the form:
Y Y
Uα ⊆ Xα (1)
α∈I α∈I

where Uα is open in Xα and Uα = Xα for all but one of the factors. It is generated by a
base of open sets of the form (1) where Uα is open in Xα and Uα = Xα for all but finitely
many α ∈ I (prove this).
Example 39. In the Euclidean topologies, Rn ∼
Qn
= i=1 R (prove this).
Example 40. In the Zariski topology, Cn ∼ 6= ni=1 C (prove this).
Q

Example 41. The product topology on R∞ = ∞

Q
i=1 R contains (and is generated by)
open sets of the form (a1 , b1 ) ×
Q... × (an , bn ) × R × R × ... where n is an arbitrary positive
∞
integer. Warning: the subset i=1 (0, 1) is not an open set in the product topology.
The product topology satisfies (and is determined up to unique isomorphism by) the
following universal property:
Proposition 3.16. Let Z be a topological space and let {fα : Z → Xα }α∈I be a collection
of continuous maps. Then there exists a unique continuous map
Y
f :Z→ Xα
α∈I

such that fα = πα ◦ f . The fα are called the coordinate functions of f .

Proof. Homework Exercise.
Proposition 3.17 (Subspaces commute with products). Q Let {Xα }α∈I be a collection of
spaces and let Aα ⊆ Xα be a subset. The product α∈I Aα has the same topology when
regarded
Q as the product of subspaces as when regarded as the subspace of the product space
α∈I X α .
Proof. Homework exercise.

21
3.8 The Final Topology
Let X be set and let {fα : Yα → X}α∈I be a collection of maps of sets, where the Yα are
topological spaces. The final topology on X is the finest topology on X such that fα is
continuous for every α.
Proposition 3.18. The open sets in the final topology are equal to the collection τ :=
{V ∈ X|f −1 (V ) is open in Yα for all α ∈ I}.
Proof. It is clear that if τ is a topology, it must be the finest topology for which the fα
are all continuous. Thus it suffices to check that τ satisfies the axioms of a topology.
For any α ∈ I:
• fα−1 (∅) = ∅ and f −1 (X) = Yα are both open in Yα , so ∅ and X lie in τ .

• If {Vj }j∈J ⊆ τ , then fα−1 ( j∈J Vj ) = j∈J fα−1 (Vj ) is open in Yα , so j∈J Vj ∈ τ .
S S S

• Finite intersections is similar to unions.

Exercise 7. Show that C ⊂ X is closed in the final topology exactly when fα−1 (C) is
closed for all α ∈ I.

3.9 Quotient Topology

Let ∼ be an equivalence relation on a topological space X and denote by X∼ the set of
equivalence classes ( I denoted this by E before), and let

Q : X → X∼

be the quotient map Q(x) = [x].

Definition 28. The quotient topology on X∼ is the final topology on X∼ with respect
to the map Q. In particular, U ⊆ X∼ is open if and only if Q−1 (U ) is open. We call
X the quotient space.
∼
Example 42. X = [0, 1] with Euclidean topology and ∼ is the equivalence relation
generated by 0 ∼ 1. Then the quotient space

is homeomorphic to the circle S 1 := {(x, y) ∈ R2 |x2 + y 2 = 1}.

22
Example 43. X = R with Euclidean topology and ∼ is the equivalence relation generated
by x ∼ x + 1. Then the quotient space is again diffeomorphic to S 1 .

Example 44. A torus. Let X := [0, 1] × [0, 1] (with Euclidean topology). Let ∼ be the
equivalence relation generated by (x, 1) ∼ (x, 0) and (0, y) ∼ (1, y) for all x, y ∈ [0, 1].
Then the quotient space is homeomorphic to the torus S 1 × S 1 .

Example 45. X = S 2 is the unit sphere in R3 with the Euclidean geometry, which with
think of as the surface of the globe. Choose an equivalence relation x ∼ y if they have
the same latitude. Then the quotient space X/ ∼ is homeomorphic to an interval (say
[−90, 90]).

By some abstract nonsense, the following universal property characterizes the quotient
space up to unique isomorphism.

Proposition 3.19. Let X is a topological space with equivalence relation ∼ and quotient
space Q : X → X∼. Let f : X → Y be a continuous map with the property that for
a, b ∈ X,
a ∼ b ⇒ f (a) = f (b).
Then there exists a unique continuous map g : X∼ → Y for which the diagram

f
X /
=Y
Q g
(2)
!
X
∼
commutes. That is to say, f = g ◦ Q. We say that f descends to g.

Proof. It is clear at the level of sets that commutativity of the diagram forces g([x]) =
f (x). Since
[x] = [y] ⇔ x ∼ y ⇒ f (x) = f (y) ⇒ g([x]) = g([y]),
we see that g is well defined as a map of sets.

23
 To prove that g is continuous, for any U ⊂ Y open we must show g −1 (U ) is open. We
know that
Q−1 (g −1 (U )) = (g ◦ Q)−1 (U ) = f −1 (U )
which is open because f is continuous. It follows from the definition of the quotient
topology that g −1 (U ) is open.

3.10 Coproducts
`
Given a collection {Xα }α∈I of topological spaces,
` the coproduct space, denoted α∈I Xα
is constructed as follows. The set underlying α∈I Xα is

{(x, α)|α ∈ I and x ∈ Xα }

the “disjoint union” of the Xα . The topology is the final topology with respect to the
inclusion maps iβ , a
iβ : Xβ ,→ Xα , iβ (x) = (x, β).
α∈I

Informally,
` we think of the Xα as disjoint subspaces of the coproduct and declare a
subset U ⊆ α∈I Xα to be open if U ∩ Xα is open in Xα .
Example 46. Let [0, 1] be the unit interval with Euclidean topology. Then [0, 1] q [0, 1]
is homeomorphic to a disjoint union of closed intervals, say [0, 1] ∪ [2, 3].
The coproduct is determined, up to unique isomorphism, by the following universal
property.
Proposition 3.20. The coproduct satisfies the following universal property. Given a
topological space Y and a collection
` of continuous maps {fα : Xα → Y }α∈I , there exists a
unique continuous map F : α∈I Xα → Y such that the following diagram commutes for
every β ∈ I:
fβ
Xβ / Y
;
iβ F (3)
`$
α∈I Xα

Proof. Commutativity of the diagram requires that F ((x, β)) = fβ (x) and this is clearly
well-defined as a map of sets. To see that F is continuous, choose U ⊂ Y open. Then for
all β
iβ−1 (F −1 (U )) = fβ−1 (U )
is open, so by definition of the final topology, F −1 (U ) is open and we conclude that F is
continuous.
Remark 4. The proofs of Propositions 3.19 and 3.20 are very similar. In fact, both are
special cases of a more general construction: that of a colimit. We will not explore this
more sophisticated notion. Every colimit can be produced by first taking a coproduct,
and then taking a quotient, and we will follow this approach.

24
Example 47. Consider the complex plane C (with either the Euclidean or Zariski topol-
ogy). Consider the coproduct C q C and use notation (z, 1) for an element in the first
copy of C and (z, 2) for an element the second copy. Let ∼ be the equivalence relation
generated by
(z, 1) ∼ (1/z, 2) for all z 6= 0.
If C ∼= R2 is given the Euclidean topology, the quotient (C q C)/ ∼ is homeomorphic to
S 2 in the Euclidean topology and is called the Riemann sphere. If C carries the Zariski
topology, (C q C)/ ∼ carries the finite-complement topology, and is called the complex
projective line.

Example 48. The space Dn := {~x ∈ Rn ||~x| ≤ 1} is called the n-disk or n-cell (we adopt
the convention that D0 is a point). It contains the (n−1)-sphere S n−1 := {~x ∈ Rn ||~x| = 1}
as a subspace. Let X be a topological space and let f : S n−1 → X be a continuous map.
We form a new space
X̂ := (X q Dn )/ ∼
where ∼ is the equivalence relation generated by ~x ∼ f (~x), for all ~x ∈ S n−1 ⊂ Dn . We
say that X̂ is obtained from X by gluing on an n-cell (or adjoining an n-cell).

Observe that as a set, X̂ is the disjoint union of X with the interior of Dn , but the
topology is not the coproduct of these subspaces.

Definition 29. We define n-dimensional cell-complexes (or CW-complexes) inductively.

A 0-dimensional cell-complex is a set with the discrete topology, X0 (i.e. a coproduct
of 0-cells).
An n-dimensional cell-complex Xn is a space constructed by gluing n-cells to a (n−1)-
dimensional cell-complex using maps f : S n−1 → Xn−1 .

Example 49. S n is a n-dimensional cell-complex. Simply glue Dn to D0 by the constant

map S n−1 → D0 .

25
Example 50. Dn is an n-dimensional cell-complex. Glue Dn to S n−1 by the identity
map S n−1 → S n−1 .

Example 51. The torus is a 2-dimensional cell-complex. Recall the definition of the
torus as a quotient. This gives the same space as the cell-complex with one 0-cell and two
1-cells and the one 2-cell. The 2-cell is glued according to the map S 1 → (S 1 q S 1 )/ ∼
which winds around the first loop, then the second loop, then backwards around the first
and backwards around the second.

4 Metric Topologies
Definition 30. A metric on a set X is a function

d:X ×X →R
satisfying: ∀a, b, c ∈ X

• d(a, b) ≥ 0 and d(a, b) = 0 ⇔ a = b.

• d(a, b) = d(b, a).

• d(a, b) + d(b, c) ≥ d(a, c).

The function d is called the distance or metric function.

26
 The third condition above is called the triangle inequality. This terminology comes
from the following helpful diagram:

Definition 31. The metric topology on a metric space (X, d) is generated by the subbase,

B := {B (a)|a ∈ X,  > 0}

where B (a) = {x ∈ X|d(x, a) < } is the epsilon-ball centered at a.

Proposition 4.1. The subbase B above is actually a base. This means in particular that
U ⊂ X is open in the metric topology if for every a ∈ U there exists  > 0 such that
B (a) ⊂ U .

Proof. Let Br1 (a) and Br2 (b) be two basic open sets, and suppose c ∈ Br1 (a) ∩ Br2 (b). We
must show that there exists a basic open set containing c and contained in Br1 (a)∩Br2 (b).
By definition then, d(a, c) < r1 and d(b, c) < r2 . It follows that

r := min(r1 − d(a, c), r2 − d(b, c)) > 0.

Claim: Br (c) ⊆ Br1 (a) ∩ Br2 (b).

If p ∈ Br (c), then

d(a, p) ≤ d(a, c) + d(c, p) < d(a, c) + r ≤ r1 ,

so p ∈ Br1 (a). Similarly,

d(b, p) ≤ d(b, c) + d(c, p) < d(a, c) + r ≤ r2 ,

so p ∈ Br2 (b).

Definition 32. A topological space (X, τ ) is called metrizable if there exists a metric on
X whose metric topology agrees with τ .

When is a topology metrizable? We will obtain a partial answer to this question later
in the course with the Urysohn Metrization Theorem.

Proposition 4.2. Let A ⊂ X be a subset of a metric space (X, d). Then A is a metric
space under the restriction d|A of d and the metric topology agrees with the subspace
topology on A.

27
Proof. The axioms of a metric are clearly preserved under restriction.
d|
Also B A (a) = Bd (a) ∩ A, so a basic open set in metric topology is the restriction
of a open in the subspace topology, so the metric topology is coarser than the subspace
topology.
On the other hand, a general open set in the subspace topology has the form U ∩ A
where U is open in X. Given p ∈ A ∩ U , we may find  > 0 such that Bd (p) ⊂ U
d|
and thus B A (p) = Bd (p) ∩ A ⊆ U ∩ A, so U ∩ A is open in the metric topology, so
the metric topology is also finer than the subspace topology and the two topologies must
coincide.

Lemma 4.3. Let d1 and d2 be two metrics on the same set X. Suppose that for every
a ∈ X,  > 0, there exists a δ > 0 such that

Bd1 (a) ⊇ Bδd2 (a).

Then the metric topology of d2 is finer than the metric topology of d1 (in fact this is if
and only if ).

Proof. Let U be a open with respect to d1 . Then for all a ∈ U , there exists  > 0 such
that Bd1 (a) ⊆ U . Thus there exists δ > 0 such that Bδd2 (a) ⊆ U , so U is also open with
respect to d2 .

Example 52. Let X = Rn . The Euclidean metric

de (~a, ~b) = (a1 − b1 )2 + ... + (an − bn )2

p

and the taxi-cab metric n

X
dt (~a, ~b) = |ai − bi |
i=1

both determine the Euclidean topology on Rn , because

n
X Xn
de (~a, ~b)2 = 2
|ai − bi | ≤ ( |ai − bi |)2 = dt (~a, ~b)2
i=1 i=1

while on the other hand,

n
X n
X
dt (~a, ~b) = |ai − bi | ≤ de (~a, ~b) = nde (~a, ~b)
i=1 i=1
n
o for any ~a ∈ R , and  > 0,
de
B/n (~a) ⊆ Bdt (~a) ⊆ Bde (~a).

The following proposition allows us to replace any metric by a bounded one - this is
useful in constructions.

28
Proposition 4.4. Let d be a metric on a set X, and define d¯ : X × X → R by
¯ b) = min{d(a, b), 1}.
d(a,

Then d¯ is a metric and it determines the same topology as d. We call d¯ the standard
bounded metric associated to d.

Proof. We begin by showing that d¯ is a metric. Clearly d(a,¯ b) ≥ 0, d(a,

¯ b) = 0 ⇔ a = b
¯ ¯
and d(a, b) = d(b, a). It remains to verify the triangle inequality:
?
¯ b) + d(b,
d(a, ¯ c) ≥ d(a,
¯ c)
¯ b) = 1 or d(b,
If d(a, ¯ c) = 1 then the LHS ≥ 1 ≥ RHS so we are okay. Otherwise, we
have:

¯ b) + d(b,
d(a, ¯ c) = d(a, b) + d(b, c) ≥ d(a, c) ≥ d(a,
¯ c)

completing the proof that d¯ is a metric. To prove that d¯ determines the same topology
¯
as d, first observe that for δ < 1, Bδd (a) = Bδd (a). Thus for any a ∈ X and any  > 0,
¯ ¯
Bδd (a) = Bδd (a) ⊆ Bd (a) ⊆ Bd (a)
where δ = min{, 1/2}. Applying Lemma 4.3 completes the proof.

4.1 Function Spaces and Uniform Convergence

For a pair of sets I and X, denote by

X I := F unct(I, X)

the set of maps of sets from I to X. There is an isomorphism of sets

XI ∼
Y
= X, f 7→ (f (i))i∈I ,
i∈I

so it is possible to think an element of X I as either a function or as an I-tuple. We use

notation
f (i) = fi
to emphasize this correspondence. Let d be a metric on X and let d¯ = min{d, 1} be
standard bounded metric associated to d.

Proposition 4.5. The function ρ : X I × X I → R defined by

¯ i , gi ) | i ∈ I}
ρ(f, g) = sup{ d(f

is a metric on X I called the uniform metric.

29
 ¯ i , gi )|i ∈ I} is bounded above by 1, so the supremum exists
Proof. First observe that {d(f
and ρ is well-defined (this is why we worked with the bounded metric).

(1) (Positivity) ρ(f, g) ≥ 0 because it is the supremum of a nonnegative set and

ρ(f, g) = 0 ⇔ d(fi , gi ) = 0, ∀i ∈ I
⇔ fi = gi , ∀i ∈ I
⇔ f = g.
¯ i , gi |i ∈ I} = sup{d(g
(2) (Symmetry) ρ(f, g) = sup{d(f ¯ i , fi )|i ∈ I} = ρ(g, f ).

(3) (Triangle inequality) By the definition of the supremum, for all  > 0, there exists
i ∈ I such that

¯ i , hi ) + 
ρ(f, h) ≤ d(f
¯ i , gi ) + d(g
≤ d(f ¯ i , hi ) + 
≤ ρ(f, g) + ρ(g, h) + 

and because this is true for all positive , we conclude

ρ(f, h) ≤ ρ(f, g) + ρ(g, h)

Observe that an open ball in the uniform metric has the form
Y ¯
Bρ (f ) = Bd (fi ).
i∈I

Example 53. Consider RI where R has Euclidean metric. If 0 represents the constant
zero function, then Bρ (0) is the set of functions f : I → R whose range lies in the interval
(−, ).
In particular (0, 1)N ⊆ RN is open in the uniform metric topology, though it is not
open in the product topology.

Example 54. If Y is a topological space and (X, d) is a metric space, then set of contin-
uous maps Cont(Y, X) inherits a uniform metric by restriction.

Proposition 4.6. The uniform metric topology ( or simplyQuniform topology) on X I

is finer than the product topology under the bijection X I ∼
= ∈I X. If I is finite, then the
uniform and product topologies agree.

Proof of Proposition 4.6. To show that the uniform topology is finer than the product
topology, it suffices to prove that every sub-basic open set in the product topology is also
open in the uniform topology.

30
 Recall that a sub-basic open set for the product topology is one of the form
Y
U= Ui
i∈I

¯ and such that

where each Ui is open in Xi = X (equipped with the metric topology of d)
Ui = X for all but one element α ∈ I. Let (xi ) ∈ U be an arbitrary element. Choose
 > 0 so that
¯
Bd (xα ) ⊆ Uα .
The I-tuple (xi ) corresponds to the function f ∈ X I is defined by fi = xi for all i. Then
the -ball around f satisfies
Y ¯ Y
(xi ) ∈ Bρ (f ) ⊆ Bd (xi ) ⊆ Ui = U
i∈I i∈I

so U is also open in the uniform topology.

Conversely, when I is finite the open balls in the uniform metric have the form
Y ¯
Bρ (f ) = Bd (xi )
i∈I

so are basic open sets in the product topology. Thus in this case the two topologies
agree.
Definition 33. Let (fn )∞ I
n=1 be a sequence of functions in X . We say that (fn )n=1
∞

converges uniformly (with respect to a metric on X) if the sequence converges in the

uniform topology. We say that (fn )∞
n=1 converges pointwise if it converges in the product
topology.
An immediate corollary of Proposition 4.6 is:
Corollary 4.7. Uniform convergence in X I implies pointwise convergence. The converse
holds if I is finite.
Theorem 4.8 (Uniform Limit Theorem). Let Y be a topological space, (X, d) a metric
space, (fn : Y → X)∞ n=1 a sequence of continuous functions and suppose that the fn
converge uniformly to f ∈ X Y . Then the limit f is continuous.
Proof. For notational simplicity assume for simplicity that d = d.¯ Given U ⊂ X, we must
show that f −1 (U ) is open in Y . It is enough to construct for every y0 ∈ f −1 (U ) an open
neighbourhood V such that y0 ∈ V ⊆ f −1 (U ).
Choose  > 0 small enough so that

Bd (f (y0 )) ⊂ U.

Choose N large enough so that

ρ(fN , f ) < /2,
or equivalently d(f (y), fN (y)) < /2 for all y ∈ Y . Set

31
 V := fN−1 (B/2
d
(f (y0 ))) ⊆ Y.
Then y0 ∈ V because d(f (y0 ), fN (y0 )) ≤ ρ(f, fN ) < /2, and V is open because fN is
continuous.
For any other point y ∈ V we have

d(f (y), f (y0 )) ≤ d(f (y), fN (y)) + d(fN (y), f (y0 ))

≤ ρ(f, fN ) + /2
< /2 + /2 = 

Thus f (V ) ⊆ U so V ⊆ f −1 (U ). We conclude that f −1 (U ) is open and that f is continu-

ous.

Example 55. On the other hand, the sequence of continuous functions xn : [0, 1] → R
converges in the product topology (i.e. converges pointwise) but converges to a non-
continuous function.

4.2 Metric Spaces and Continuity

For metric spaces we also have an  − δ definition of continuity.

Theorem 4.9. Let f : X → Y be a map between metric spaces (X, dX ) and (Y, dY ).
Then f is continuous if and only if ∀x ∈ X and ∀ > 0, ∃δ > 0 such that

f (BδdX (x)) ⊆ BdY (f (x))

Proof. Suppose f is continuous. Then for any x and , the preimage

U = f −1 (BdY (f (x)))

is open in X. In particular, since x ∈ U , there exists an open ball BδdX (x) ⊂ U . Then by
definition of the preimage, f (BδdX (x)) ⊆ BdY (f (x)).
Conversely, suppose that the  − δ condition holds. Let V ⊂ Y be open. We want to
that f −1 (V ) is open. Let x ∈ f −1 (V ) so f (x) ∈ V . V is open so there exists  > 0 with
BdY (f (x)) ⊂ V . By  − δ there exists BδdX (x) such that

f (BδdX (x)) ⊆ BdY (f (x)) ⊆ V

thus BδdX (x) ⊂ f −1 (V ), so f −1 (V ) is open.
It is also possible to define continuity for metrizable spaces using convergent sequences.
We begin with a lemma.

Lemma 4.10. Let X be a topological space, and A ⊆ X. If there exists a sequence

{xn }∞
n=1 ⊂ A converging to x ∈ X then x lies in the closure Ā. The converse holds if X
is metrizable.

32
Proof. Suppose that xn → x (i.e. xn converges to x). Suppose that x 6∈ Ā. Then x ∈ (Ā)c
is an open neighbourhood. By the definition of convergence, there exists xn ∈ (Ā)c for n
sufficiently large. But xn ∈ A ⊆ Ā, so this is a contradiction.
Conversely, suppose that X is metrizable and x ∈ Ā. For each n ∈ Z+ choose
xn ∈ A ∩ B1/n (x) 6= ∅. For any open set x ∈ U , we have xn ∈ B1/n (x) ∩ A ⊂ U ∩ A for n
sufficiently large, thus xn → x converges.

Corollary 4.11. Let X be a topological space and Y a metric space. Then the set of
continuous function Cont(X, Y ) is a closed subset of the set of all functions Y X in the
uniform topology. This is not necessarily true in the product topology.

Proof. The uniform limit theorem says that convergent subsequences contained in Cont(X, Y )
converge to an element of Cont(X, Y ) in the uniform metric topology on Y X , so Cont(X, Y )
must be closed by Lemma 4.10.

Theorem 4.12. If f : X → Y is a continuous map, then for every convergent sequence

xn → x in X, the sequence (f (xn ))∞
n=1 converges to f (x) in Y . The converse holds if X
is metrizable.

Proof. Suppose f is continuous and xn → x in X. For any open neighbourhood V ⊂ Y

of f (x), we have

f −1 (V ) is open nbhd of x ∈ X ⇒ xn ∈ f −1 (V ) for n large

⇒ f (xn ) ∈ V for n large
⇒ f (xn ) → f (x) converges.

Now suppose that X is metrizable and f (xn ) → f (x) converges whenever xn → x

converges. To show that f is continuous we suppose that C ⊂ Y is closed and prove that
f −1 (C) is closed.
Suppose x ∈ f −1 (C). Then

∃(xn )∞
n=1 ⊂ f
−1
(C), such that xn → x ⇒ f (xn ) → f (x)
⇒ f (x) ∈ C
⇒ x ∈ f −1 (C)

so we conclude that f −1 (C) = f −1 (C) is closed.

4.3 Complete Metric Spaces

Definition 34. Let (X, d) be a metric space. A sequence (xn )∞ n=1 ⊂ X is called a Cauchy
sequence if for all  > 0, there exists N such that ∀m, n > N , d(xm , xn ) < .

Proposition 4.13. Every convergent sequence is Cauchy.

33
Proof. Suppose that the sequence (xn )∞ n=1 converges to a point L. Then by definition, for
all  > 0, there exists N such that for all n > N , xn ∈ B/2 (L). Consequently, if m, n > N
then
d(xm , xn ) ≤ d(xm , L) + d(xn , L) < /2 + /2 = 
as required.

Definition 35. A metric space is called complete if every Cauchy sequence converges.

Example 56. R \ {0} is not complete in the Euclidean metric.

Example 57. The subspace Q ⊂ R of rational numbers is not complete in the Euclidean
metric. For instance 3, 3.1, 3.14, 3.141, ... converges to π 6∈ Q.

Example 58. An open interval (0, 1) ⊂ R is not complete in the Euclidean metric.

Recall that topologically (0, 1) ∼

= R. The next proposition demonstrates that com-
pleteness is a metric property rather than a topological property.

Proposition 4.14. R with the Euclidean metric is complete.

In fact, R is by definition the “smallest” complete metric space containing Q. We

sketch this idea in the next section.

Proposition 4.15. Rk is complete with respect to the Euclidean metric d.

Proof. Suppose (~xn )∞ k

n=1 is a Cauchy sequence in R , where ~ xn = (x1n , ...xkn ). Then since
i i i ∞
d(~xm , ~xn ) ≥ |xm − xn | for any i, we also have (xn )n=1 is Cauchy in R for each i ∈ {1, ..., k}
and thus much converge to a limit Li ∈ R. But then the sequence (~xn )∞ n=1 converges to
(L1 , ..., Lk ) pointwise, which is the same as converging in the product topology. Since we
know the product topology equals the Euclidean topology, this proves that Rn is complete
in the Euclidean metric.

Theorem 4.16. Let (X, d) be a complete metric space. A subset A ⊂ X is complete with
respect to the restricted metric if and only if A is closed.

Proof. A is closed ⇔ Every convergent sequence contained in A converges to a point in

A (see Lemma 4.10) ⇔ Every Cauchy sequence in A converges in A.

Corollary 4.17. Subspaces of Rn are complete in the Euclidean metric if and only if they
are closed.

Theorem 4.18. Let (X, d) be a complete metric space, I a set. The function space X I
is complete in the uniform metric.

34
Proof. Assume without loss of generality that d = inf{d, 1} is bounded. Let

ρ(f, g) = sup{d(f (α), g(α))|α ∈ I}

denote the uniform metric for f, g ∈ X I .

Let (fn : I → X)∞ I
n=1 be a Cauchy sequence in X . Then

ρ(fm , fn ) ≥ d(fm (α), fn (α)), ∀α ∈ I

So for every α ∈ I, the sequence (fn (α))∞

n=1 is Cauchy in X, thus convergent. Define
f ∈ X I by
f (α) = lim fn (α).
n→∞

We will show that fn → f uniformly.

Given  > 0, choose N large enough so that for m, n > N ,

ρ(fm , fn ) < /2.

For any α ∈ I and m > N we have:

d(fm (α), f (α)) ≤ lim sup d(fm (α), fn (α)) + d(fn (α), f (α))
n→∞


≤ lim sup ρ(fm , fn )) + lim (d(fn (α), f (α)
n→∞ n→∞

≤ /2 + 0

Therefore ρ(fm , f ) ≤ /2 < .

Example 59. For any set I, the space RI is complete in the uniform Euclidean metric.

Corollary 4.19. Let X be a topological space and Y a complete metric space. Then
Cont(X, Y ) is complete in the uniform metric.

Proof. Since Cont(X, Y ) is closed in the complete metric space Y X , it is also complete.

4.4 Completing Metric Spaces

Let (X, dX ) and (Y, dY ) be metric spaces. A map f : X → Y is called an metric embedding
if for all a, b ∈ X,
dX (a, b) = dY (f (a), f (b)).

Proposition 4.20. Metric embeddings are injective.

Proof. Suppose that f (a) = f (b). Then dY (f (a), f (b)) = dX (a, b) = 0, so a = b.

Consequently, an metric embedding f restricts to an isomorphism with a metric sub-
space of Y
X∼= f (X) ⊆ Y.

35
Definition 36. A metric embedding f : (X, dX ) ,→ (Y, dY ) is called a completion if
(Y, dY ) is complete and the closure f (X) = Y ( i.e. f (X) is dense in Y ).
Theorem 4.21. For any metric space (X, d), a completion exists and is unique (up to
unique bijective isometry). We denote by (X, d) the completion of (X, d).
Proof. Let Cau(X) be the set of Cauchy sequences in (X, d). Given a completion f :
X ,→ Y , there is an induced map
F : Cau(X) → Y, F ((xn )∞
n=1 ) = lim f (xn ).
n→∞

Observe that this map is well defined, because Y is complete and it is surjective because
f (X) is dense so every point of Y is the limit of some Cauchy sequence in f (X) (see
Lemma 4.10).
Idea: Construct X = Cau(X)/ ∼ for an equivalence relation ∼ such that F descends
to an isomorphism
(Cau(X)/ ∼) ∼ = Y.
It is helpful in the proof to introduce a new definition. A premetric on a set S is a
map D : S × S → R that satisfies all the conditions of a metric, except that D(a, b) = 0
does not imply that a = b( i.e. D ≥ 0, and satisfies reflexivity, symmetry, and the triangle
equality).
Lemma 4.22. Let (S, D) be a premetric space. The relation
a ∼D b ⇔ D(a, b) = 0
is an equivalence relation on S and the premetric descends to metric on the quotient set
S/ ∼D . That is
D : (S/ ∼D ) × (S/ ∼D ) → R
defined by D([a], [b]) = D(a, b) is a well-defined metric.
Proof. First we show that ∼D is an equivalence relation. The only thing that is not
immediate is transitivity. Suppose that a ∼D b ∼D c so D(a, b) = D(b, c) = 0. Then by
the triangle inequality
D(a, c) ≤ D(a, b) + D(b, c) = 0 + 0 = 0,
and a ∼D c and transitivity holds.
Next, we show that D is well defined. Suppose that a ∼D b so that [a] = [b] ∈ S/ ∼.
Then
D(a, c) ≤ D(a, b) + D(b, c) = D(b, c).
Similarly we get D(b, c) ≤ D(a, c), so we conclude that D(a, c) = D(b, c). Thus the
definition D([a], [c]) = D([b], [c]) is independent of the choice of representative of [a] = [b].
That D possesses all the properties of a metric now follows immediately from the fact
that D is a premetric. For example, the triangle inequality:
D([a], [c]) = D(a, c) ≤ D(a, b) + D(b, c) = D([a], [b]) + D([b], [c])

36
 Returning to Cau(X), we claim that the map D : Cau(X) × Cau(X) → R by
D((xn ), (yn )) = limn→∞ d(xn , yn ) is a premetric on Cau(X).
First must show that D is well-defined. Since (xn ) and (yn ) are Cauchy sequences,
there exists N so that if m, n > N we have d(xm , xn ) <  and d(yn , ym ) < . Therefore
by the triangle inequality,

d(xm , ym ) ≤ d(xm , xn ) + d(xn , yn ) + d(yn , ym ) <  + d(xn , yn ),

so d(xm , ym ) − d(xn , yn ) < . Similarly, d(xn , yn ) − d(xm , ym ) <  so

|d(xm , ym ) − d(xn , yn )| < .

Thus(d(xn , yn ))∞
n=1 is Cauchy in R and must converge to D((xn ), (yn )).
Now we verify that D is a premetric. Certainly D ≥ 0, D((xn ), (xn )) = lim(0) = 0
and is symmetric. To prove triangle inequality,
 
D((xn ), (yn )) = lim d(xn , yn ) ≤ lim d(xn , zn )+d(zn , yn ) = D((xn ), (zn ))+D((zn ), (yn )).
n→∞ n→∞

Set X := (Cau(X)/ ∼D ). By the Lemma, D descends to a metric D on X.

The map i : X ,→ X sending x to the constant sequence (x)∞
n=1 is a metric embedding.
The image i(X) is dense in X because any point [(xn )] ∈ X is the limit of a sequence
(i(xn ))∞
n=1 in i(X).
It only remains to prove that X is complete. It is simpler to prove a more general
result.
Lemma 4.23. Let (Y, d) be a metric space and let A ⊂ Y be a dense subspace (i.e. the
closure of A is Y ). If every Cauchy sequence in A converges in Y , then Y is complete.
Proof. Choose (yn )∞
n=1 a Cauchy sequence in Y . Since A is dense, we can construct
another sequence (an )∞
n=1 in A such that d(yn , an ) < 1/n. For any  > 0 choose N so that
if m, n > N ,
d(ym , yn ) < /2, and 1/N < /4
Then
d(am , an ) ≤ d(am , ym ) + d(ym , yn ) + d(yn , an ) < /4 + /2 + /4 = 
so (an ) is Cauchy, hence it converges to a limit L ∈ Y . Finally

d(L, yn ) ≤ d(L, an ) + d(an , yn ) → 0

as n → ∞ so (yn )∞
n=1 converges to L.

Uniqueness can be proven using abstract nonsense.

Example 60. If X is any subset of Rn with the Euclidean metric, then the completion
may be canonically identified with the closure X ⊂ Rn (prove this).
Example 61. Let Q ,→ R inclusion of the rational numbers into the reals is the com-
pletion of Q under the Euclidean metric. In fact, this is one way to construct the real
numbers from the rationals.

37
Example 62. p-adic numbers. This is a completion of Q using a different metric. Let
p ∈ Z be a prime number. Given any element r ∈ Q, there is a unique integer N such
that
a
r = pN
b
where a, b ∈ Z are relatively prime and not divisible by p. The p-adic norm is a map
Q → R defined by
1
||r||p = N .
p
The p-adic metric on Q is defined by

dp (r, s) = ||r − s||p .

I leave it as an exercise to prove this is a metric. Observe that two rational numbers r, s
are close together in this metric if their difference is highly divisible by p (or rather has
many factors of p in the numerator compared to the denominator). The completion of Q
with respect to dp is the set Qp of p-adic numbers. The p-adic integers Zp are the closure
of Z as a subset of Qp .

Example 63. Let [0, 1] be the unit interval and consider the set X := Cont([0, 1], R) of
continuous functions in the Euclidean topologies. For any real number p with 1 ≤ p < ∞,
the Lp -norm is defined:
Z 1

1/p
||.||p : X → R , ||f ||p = |f (t)|p dt
0

The Lp -metric is
dp (f, g) = ||f − g||p
The completion is the space Lp ([0, 1]) of Lp -functions.

Aside: There is a notational convention that when p = ∞, d∞ (f, g) = ||f −g||∞ equals
the uniform metric. Since the uniform metric on Cont([0, 1], R) is complete (Corollary
4.19), this means that L∞ ([0, 1]) = Cont([0, 1], R). It interesting to ponder why this is a
good convention.

5 Connected Spaces
Definition 37. Let X be a topological space. A separation of X is a pair of non-empty,
open sets U, V ⊂ X satisfying

(i) U ∩ V = ∅

(ii) U ∪ V = X.

In other words, U, V are disjoint and cover X. We say X is connected if no separation of

X exists. We say a subset A ⊂ X is connected is it is connected in the subspace topology.

38
Remark 5. Observe that for any separation U, V of X, U = V c and V = U c are also
both closed. Thus we may also define a separation to be a pair of non-empty closed sets
U, V satisfying U ∩ V = ∅ and U ∪ V = X.

Example 64. The subspace X := [0, 1] ∪ [2, 3] ⊂ R is not connected in the Euclidean
topology, because the pair U = [0, 1] and V = [2, 3] is a separation of X.

Proposition 5.1. A space X is connected if and only if the only subsets of X which are
both open and closed are ∅ and X.

Proof. A subset U ⊆ X is both open and closed if and only if U and U c are open. Thus
is U 6= ∅, X then U, U c separate X.
Conversely, if U, V separate X then U is both open and closed.

Definition 38. A subspace X ⊂ R is called convex if for all a, b ∈ X the interval

[a, b] ⊆ X.

It is not hard to show that the convex subsets of R are the open, closed and half-open
intervals, the open and closed rays, and R itself.

Proposition 5.2. Let X ⊂ R with the Euclidean topology. X is convex if and only if X
is connected.

Proof. First, assume that X ⊆ R is not convex. Then for some pair a, b ∈ X there exists
r ∈ [a, b] ∩ X c . Then a ∈ U = (−∞, r) ∩ X is open and b ∈ U c = (r, ∞) ∩ X is also open
so U, U c separate X.
Now suppose X is not connected and let A, B ⊂ X be a separation. Choose points
a ∈ A and b ∈ B and assume without loss of generality that a < b. We aim to show that
[a, b] 6⊆ X.
Assume that [a, b] ⊆ X. Then the sets A0 = A ∩ [a, b] and B0 = B ∩ [a, b] form a
separation of [a, b]. Since both A0 and B0 are closed in [a, b] and [a, b] is also closed in R,
it follows that A0 and B0 are closed in R (a closed subspace of a closed subspace is closed
by Lemma 3.14).
In particular the supremum
c = sup(A0 )
must be an element of A0 . We conclude that c 6= b ∈ B0 . The non-empty set (c, b] is
disjoint from A0 so it must be contained in B0 . But since B0 is closed, it follows that
c ∈ B0 which is a contradiction because A0 , B0 are disjoint. Thus

[a, b] 6⊆ X
so X is not convex.

The most important property of connectedness is the following.

Theorem 5.3. Let f : X → Y be a continuous map, with X connected. Then the image
f (X) is connected.

39
Proof. Suppose that U and V separate f (X). Then f −1 (U ) and f −1 (V ) are non-empty,
open and satisfy

f −1 (U ) ∩ f −1 (V ) = f −1 (U ∩ V ) = f −1 (∅) = ∅
and
f −1 (U ) ∪ f −1 (V ) = f −1 (U ∪ V ) = f −1 (f (X)) = X
so they separate X.

Example 65. The circle S 1 can be constructed as a quotient space [0, 1]/{0 ∼ 1}. The
quotient map Q : [0, 1] → S 1 is surjective, so S 1 is connected.

By the same reasoning we get the following corollary.

Corollary 5.4. Let X be a connected space and let ∼ be an equivalence relation on X.

The quotient space X∼ is connected.

Corollary 5.5 (Intermediate Value Theorem). Let f : X → R be continuous with X

connected. For a, b ∈ X let r ∈ [f (a), f (b)]. Then there exists c ∈ X for which f (c) = r.

Proof. X is connected so f (X) is a connected subset of R thus it is convex, so r ∈

f (X).

Lemma 5.6. Let X be a topological space and let {Aα }α∈I be a collection of connected
subspaces,
S each containing a common point p ∈ Aα for all α ∈ I. Then the union
A := α∈I Aα is connected.

Proof. Suppose for the sake of contradiction that U and V separate A and assume without
loss of generality that p ∈ U . Then since Aα is connected, one of U ∩ Aα and V ∩ Aα must
be empty (otherwise they would form a separation of Aα ). Since p ∈ U ∩ Aα it must be
that V ∩ Aα = ∅. But this is true for all α so V ∩ A = ∅ also, which is a contradiction.

Proposition 5.7. Given a topological space X define a relation ∼ by x1 ∼ x2 if there

exists a connected subspace A ⊂ X containing both x1 and x2 . Then ∼ is an equivalence
relation. The equivalence classes of ∼ are the maximal connected subspaces of X and are
called the connected components of X.

Proof. Clearly ∼ possesses both the identity and symmetry properties. It remains prove
transitivity. Suppose x ∼ y and y ∼ z. Then there exist connected subsets A, B ⊆ X
such that x, y ∈ A and y, z ∈ B. Since y is contained in both A and B, it follows by
Lemma 5.6 that A ∪ B is connected. Since A ∪ B contains both x and z, it follows that
x ∼ z.
To see that the equivalence classes are connected, observe that [x] equals the union of
all connected subsets A ⊂ X containing x, so by Lemma 5.6 [x] is connected.

Proposition
Qn 5.8. Let {Xi }ni=1 be a finite collection of connected spaces. Then the product
i=1 Xi is connected.

40
Proof. By induction, it suffices to consider the case n = 2. Let X and Y be connected
spaces and choose a fixed base point (a, b) ∈ X × Y . The subspace X × {b} ∼ = X is
∼
connected and for each x ∈ X, the space {x} × Y = Y is connected. Applying Lemma
5.6, the union
Tx := X × {b} ∪ {x} × Y
is connected because they share the common point (x, b). Since Tx contains (a, b) for
every value of x, we apply Lemma 5.6 again to see that
[
X ×Y = Tx
x∈X

is connected.
A more general result holds.

Theorem
Q 5.9. Let {Xα }α∈I be a collection of connected spaces. Then the product space
α∈I Xα is connected.

Before we prove this, we need a lemma.

Lemma 5.10. Let X be a topological space, and let A ⊂ X be connected and dense (i.e.
the closure Ā = X). Then X is connected.

Proof. Suppose that U, V ⊆ X separate X. Then either U ∩ A or V ∩ A is empty. Say

U ∩ A = ∅. Then A ⊆ V . Because V is closed in X and A is dense, this implies that
X = Ā ⊆ V . This is a contradiction.
Q
Proof of Theorem 5.9. Choose a fixed basepoint (aα ) ∈ α∈I Xα . Given a finite subset
K ⊂ I define Y
XK := {(xα ) ∈ Xα | xα = aα if α 6∈ K }
α∈I
Q
The subset XK is homeomorphic to the finite product α∈K Xα , so it is connected. All
XK contain the basepoint, so applying Lemma 5.6 we learn that the subspace
[
Y := XK
finite K⊂I
Q
is connected. By Lemma 5.10 it suffices to prove that Y is dense in α∈I Xα . Q
Recall that the basic open sets in the product space have the form B = α∈I Uα ,
where the Uα are nonempty, open and Uα = Xα for all α outside of finite subset K ⊂ I.
But then
Y ∩ B ⊇ XK ∩ B 6= ∅,
so Y Qintersects every basic open set B, and hence every non-empty open set. Thus
Y = α∈I Xα completing the proof.

Example 66. RI and [0, 1]I are connected in the product topology for any I. Quotients
of these spaces are connected.

41
5.1 Path-Connected Spaces
Definition 39. Let X be a topological space. A path in X is a continuous map f :
[0, 1] → X. We say that the path f joins a, b ∈ X if f (0) = a and f (1) = b. A space X
is called path-connected if every two points a, b ∈ X can be joined by a path.
Proposition 5.11. The relation

a∼b if a can be joined to b

is an equivalence relation. The equivalence classes of ∼ are called the path-components

of X and are equal to the maximal path-connected subspaces of X.
Proof. Clearly a ∼ a for any a ∈ X using the constant path. If f (t) : [0, 1] → X joins a
to b then f (1 − t) joins b to a, so ∼ is symmetric. Now suppose that f1 joins a to b and
f2 joins b to c. Define the concatenation
(
f1 (2t) if t ∈ [0, 1/2]
g(t) :=
f2 (2t − 1) if t ∈ [1/2, 1]
Then g restricts to a continuous function on both [0, 1/2] and [1/2, 1]. These form a finite
closed cover of [0, 1] so by Exercise 6, g is continuous and joins g(0) = a to g(1) = c. It
follows that ∼ satisfies transitivity.
That the path-components are equal to the maximal path-connected subsets follows
immediately from ∼ being an equivalence relation.
Theorem 5.12. If a space X is path-connected, then it is connected.
Proof. Suppose that X is not connected. Then there exists a separation U, V of X ( so
U, V are open, nonempty, disjoint and cover X). Choose a ∈ U and b ∈ V and suppose
there exists a path f : [0, 1] → X, with f (0) = a and f (1) = b. Then f −1 (U ) and f −1 (V )
must separate [0, 1]. But this is a contradiction, because [0, 1] is connected.
This is useful because it is often easier to verify that the space is path-connected than
to prove it is connected. For example:
Example 67. A subset A ⊂ RI is convex if every pair a, b ∈ A is joined by the straight
line path f (t) := (1 − t)a + tb. Clearly every convex set is path connected, so it is also
connected. This in particular includes, the n-disk Dn := {~x ∈ Rn |||~x|| ≤ 1}.
The converse of Theorem 5.12 is not true however. Some connected spaces are not
path-connected.
Example 68. The topologist’s sine curve. Consider the continuous function

f : (0, 1] → R, f (x) = cos(π/x).

The graph G of f is the image in R2 the continuous map x 7→ (x, f (x)).

42
Since (0, 1] is connected, it follows from Theorem 5.3 that G is also connected. By Lemma
5.10, the closure
G = G ∪ ({0} × [−1, 1])
is also connected. However G is not path-connected.
Proof. Suppose there is a path f : [0, 1] → G joining (0, 0) to (1, −1). We may write f in
coordinates f (t) = (x(t), y(t)) where x(t), y(t) are both continuous real-valued functions.
Let
c = sup{t ∈ [0, 1] | x(t) = 0}.
Then x(c) = 0 (thus c < 1) and x(t) > 0 for all t ∈ (c, 1]. For every δ > 0, there is a
sufficiently large integer N such that
1 1
, ∈ [0, x(c + δ)) = [x(c), x(c + δ)).
N N +1
Consequently, for every δ > 0

y([c, c + δ)) ⊇ {cos(N π), cos((N + 1)π)} = {±1}

so y(t) is not continuous at t = c. Contradiction.
Observe that some results about path-connected spaces are much easier to prove than
for connected spaces.
Theorem 5.13. The product of any collection {Xα }α∈I of path-connected spaces is path-
connected.
Q
Proof. Suppose ā := (aα ) and b̄ := (bα ) are two points in α∈I Xα . For each α ∈ I, there
exists a path fα : [0, 1] → Xα joining aα to bα . Then the product map
Y
(fα )α∈I : [0, 1] → Xα

joins ā to b̄.
Theorem 5.14. The continuous image of a path connected space is path connected.
Proof. Suppose φ : X → Y is continuous and X is path connected. For any pair of points
φ(a) and φ(b) in φ(X), choose a path f : [0, 1] → X joining a to b. Then the composition
φ ◦ f joins φ(a) and φ(b).

43
Example 69. Suppose that X is a (path-)connected space. If X̂ is constructed by gluing
an n-cell onto X, for n ≥ 1, then X̂ is also (path-)connected.
Proof. The space X̂ is constructed as a quotient space
a
X̂ := (X Dn )/ ∼,

where ∼ is the equivalence relation generated by φ(p) ∼ = p, where φ : S n−1 → X is a

n−1 n
gluing map and` Sn ⊂ D is the boundary sphere.
If Q : X D → X̂ is the quotient map, then the image sets Q(X) and Q(Dn )
continuous images of (path-)connected space, and thus are (path-)connected. For any
point p ∈ S n−1 , Q(p) lies in the intersection of Q(X) and Q(Dn ), so

X̂ = Q(X) ∪ Q(Dn )

is a union of (path-)connected spaces sharing a common base point, thus is (path-

)connected.

6 Compactness
Definition 40. A collection A of subsets of a space X is called a cover or covering if
[
U = X.
U ∈A
S
More generally, if Y ⊂ X, then we say A is a cover of Y if Y ⊆ U ∈A U . An open cover
is a covering by open sets.

Definition 41. A topological space X is compact if every open covering of A contains

a finite subcollection {U1 , ..., Un } ⊆ A that also covers X. We call {U1 , ..., Un } as finite
subcover.

Example 70. Any topology on a finite set is compact, because every open cover is already
finite.

Example 71. The trivial topology on any set is compact, because there are only two
open sets so every open cover is already finite.

Example 72. Suppose that (xn )∞ n=1 is a convergent sequence in a space X, with limit
xn → L. Then the subspace {xn |n ∈ Z+ } ∪ {L} is a compact subset of X, because any
open set containing L contains all but finitely many elements in the sequence.

Example 73. R is not compact in the Euclidean topology, because A := {(n, n + 2)|n ∈
Z} is an open cover with no finite subcover.

Example 74. The half open interval (0, 1] ⊂ R is not compact, because A := {( n1 , 1]|n ∈
Z+ } is an open cover with no finite subcover.

44
Theorem 6.1. Every closed and bounded interval [a, b] ⊂ R is compact.
Proof. Let A be an open cover of [a, b]. Define a subset

S := {λ ∈ [a, b] for which [a, λ] has a finite subcover in A }

Our aim is to prove that b ∈ S. First observe that a ∈ S, because [a, a] = {a} is a finite
set. Thus S is nonempty, and is bounded above, so c := sup(S) exists.

Claim: c ∈ S.
Proof. If c = a then this is clear. Now suppose a < c. Choose U ∈ A such that c ∈ U .
Thus for some  > 0, (c − , c] ⊆ U . Because c is the supremum of S, there exists

d ∈ (c − , c] ∩ S 6= ∅,

and it follows that [a, d] has a finite subcover {U1 , ..., Un } ⊂ A. But then {A1 , ..., An , U }
is a finite cover of [a, d] ∪ (c − , c] = [a, c].
Claim: c = b.
Proof. Suppose c < b. If {U1 , ..., Un } covers [a, c], it also covers [a, c + ] for  > 0 small,
thus c +  ∈ S which is a contradiction.

Theorem 6.2. Every closed subset of a compact space is compact.

Proof. Let X be compact, Y ⊆ X closed and let A be an open covering of Y . By definition
of the subspace topology, every open covering of Y has the form

A := {U ∩ Y |U ∈ A0 }

where A0 is a collection of open sets in X covering Y . Then A0 ∪ {X \ Y } is an open

covering of X, so it contains a finite subcovering {U1 , ..., Un }, and {Ui ∩ Y }ni=1 forms a
finite cover of Y contained in A.
Example 75. Every closed and bounded subspace of R is compact, because is also a
closed subset of the compact space [−N, N ] for some large N .
Theorem 6.3. Every compact subspace of a Hausdorff space is closed.
Proof. Let X be Hausdorff and Y ⊆ X compact. If Y = X then Y is closed and the
result is immediate.
So suppose Y is a proper subset of X. Fix a point p ∈ X \ Y . For each y ∈ Y , there
exist disjoint open neighbourhoods Uy , Vy containing y and p respectively. Construct in
this fashion an open cover {Uy |y ∈ Y } of Y . Since Y is compact, there exists a finite
subcover {Uy1 , ..., Uyn }, such that
[ n
Uyi ⊇ Y.
i=1

45
From this we conclude that the finite intersection
n
\
Vyi
i=1

is an open neighbourhood of p contained in Y c . Such a neighbourhood exists for all p ∈ Y c

so Y c is open, and Y is closed.
Theorem 6.4. The product of finitely many compact spaces is compact
Before proving this we must establish a lemma.
Lemma 6.5 (The Tube Lemma). Let X and Y be spaces, with Y compact. Suppose that
an open subset N ⊆ X × Y contains a “slice” {x0 } × Y . Then there exists an open nbhd
x0 ∈ W ⊆ X such that W × Y ⊂ N .

Proof. Recall that the basic open sets in X × Y have the form U × V for open sets U ⊆ X
and V ⊆ Y . Let
A := {U × V basic |U × V ⊆ N }.
This forms an open cover of N hence also of {x0 } × Y . Since {x0 } × Y is compact, there
subcover {Ui × Vi }ni=1 (we insist that each of these actually intersects the slice).
is a finite T
Set W = ni=1 Ui . Then W is an open neighbourhood of x0 and
n
[ n
[
W ×Y = W × Vi ⊆ Ui × Vi ⊆ N
i=1 i=1

proving the lemma.

Proof of Theorem. It is enough to show that the product of two compact sets is compact.
Let X, Y be compact and let A be an open cover of X × Y . For each x ∈ X, there exists
a finite subcover Ax ⊆ A of the slice {x} × Y . By the tube lemma, there exists an open
neighbourhood x ∈ Wx ⊆ X such that
[
Wx × Y ⊆ U.
U ∈Ax

The collection {Wx |x ∈ X} is an open cover of X so it contains a finite subcover {Wxi }ni=1 .
Then,

46
 n
[ n  [
[ 
X ×Y = W xi × Y ⊆ U
i=1 i=1 U ∈Axi

so, Ax1 ∪ ... ∪ Axn ⊂ A is a finite subcovering of X × Y , so X × Y is compact.

Theorem 6.6 (Heine-Borel Theorem). Let X be a subspace of Rn with the Euclidean
topology. Then X is compact if and only if X is closed and bounded.
Proof. Exercise on problem set 5.
Now that we have constructed a large class of compact sets: What are they good for?
One of the most useful properties of compact sets is the following.
Theorem 6.7. The image of a compact space under a continuous map is compact.
Proof. Suppose X is compact and f : X → Y is a continuous map. Let A be an open
cover of f (X) ⊆ Y . Then the collection
f −1 A := {f −1 (U )|U ∈ A}
is an open cover of X. Since X is compact, there is a finite subcover {f −1 (Ui )}ni=1 of
X, but this implies that {Ui }ni=1 ⊂ A is a finite subcovering of f (X), so f (X) is also
compact.
Corollary 6.8. A quotient space of a compact space is compact.
Corollary 6.9 (Extreme Value Theorem). Let f : X → R be a continuous map with X
compact and non-empty. Then there exists xi , xs ∈ X such that
f (xi ) = inf(f (X)), f (xs ) = sup(f (X))
Proof. The image f (X) ⊂ R is compact, so it is closed and bounded. It follows that
inf(f (X)) and sup(f (X)) both exist and are contained in f (X), so the result follows.
Theorem 6.10. Let f : X → Y be bijective and continuous, where X is compact and Y
is Hausdorff. Then f is a homeomophism.
Proof. In remains to show that the inverse map f −1 : Y → X is continuous. We do this
by proving that if C ⊆ X is closed, then (f −1 )−1 (C) = f (C) is closed in Y .
Let C ⊆ X be closed. Since X is compact, it follows that C is compact by Theorem
6.2. Since f is continuous, by Theorem 6.7 we know that f (C) is compact. By Theorem
6.3 it follows that f (C) is closed.
Example 76. Consider the continuous map f : [0, 2π] → S 1 defined by
f (t) = (cos(t), sin(t)).

This determines a continuous bijection from the quotient space f¯ : [0,2π] ∼

→ S 1 , under
relation 0 ∼ 2π. Since S is Hausdorff (it is a subspace of the Hausdorff space R2 ) and
1
[0,2π]
∼
is compact (it is the quotient of a closed and bound interval), it follows that f¯ is a
homeomorphism.

47
6.1 Compact Metric Spaces
Definition 42. Let (X, d) be a metric space, and let A ⊆ X with A 6= ∅. For p ∈ X
define the distance from p to A by

d(p, A) = inf{d(p, a)|a ∈ A}

Proposition 6.11. For fixed, nonempty A ⊆ X, the map p 7→ d(p, A) is a continuous
map from X to R.
Proof. Both R and X are metric spaces, so we may use an  − δ proof. We will prove that
for any p, q ∈ X, |d(p, A) − d(q, A)| ≤ d(p, q), so setting  = δ will suffice.
By the triangle inequality, for all a ∈ A,

d(p, A) ≤ d(p, a) ≤ d(p, q) + d(q, a)

Taking infimum over a ∈ A, we have

d(p, A) ≤ d(p, q) + d(q, A) ⇒ d(p, A) − d(q, A) ≤ d(p, q)

Reversing the roles of p and q gives d(q, A) − d(p, A) ≤ d(p, q), so |d(p, A) − d(q, A)| ≤
d(p, q) as desired.

Lemma 6.12 (Lebesgue Number Lemma). Let A be an open covering of a compact metric
space X. There exists δ > 0, called the Lebesgue number, such that for all p ∈ X, the
open ball Bδ (p) is contained in some U ∈ A.
Proof. If X ∈ A then any value of δ > 0 will do. So we assume that X 6∈ A. Because X
is compact, we may choose a finite subcover {U1 , ..., Un } ⊆ A. Let Ci = X \ Ui 6= ∅ and
define a function
n
1X
f : X → R, f (p) = d(p, Ci )
n i=1
thus f (p) is the average distance from p to the sets Ci . Note that f is continuous because
it is a linear combination of continuous real valued functions (using Proposition 6.11 and
a homework problem).
Observe that for any p ∈ X, there must be some i such that d(p, Ci ) ≥ f (p), which
implies that Bf (p) (p) ⊆ Ui . Thus to prove the Lemma it is enough to find a positive lower
bound for f .
Claim: f (p) > 0 for all p ∈ X.
Proof. Choose Ui containing p. Since Ui is open, B (p) ⊂ Ui for some  > 0. Therefore
d(p, Ci ) ≥ , so f (p) ≥ /n > 0.
By the extreme value theorem, there exists p0 ∈ X such that inf{f (p)|p ∈ X} =
f (p0 ) > 0. Setting δ = f (p0 ) completes the argument.

48
 1
Example 77. Consider the open cover of the open interval (0, 1) by A := {( 2n+2 , 21n )|n ∈
Z≥0 }. This cover admits no Lebesgue number, because each  ∈ (0, 1) is only contained
in any intervals of length less than or equal to 3.
Definition 43. A map f : X → Y between metric spaces is uniformly continuous, if
for all  > 0, there exists δ > 0 such that for any p, q ∈ X,
dX (p, q) < δ ⇒ dY (f (p), f (q)) < .
Example 78. The map f : R → R, f (x) = x2 is not uniformly continuous with respect
to the Euclidean metric. For any fixed  > 0, the required value of δ > 0 gets smaller and
smaller as x → ∞ and the slope of the graph gets steeper.

Corollary 6.13. If X is compact and X, Y are metric spaces, then any continuous map
f : X → Y is uniformly continuous.
Proof. Given  > 0, define an open cover A := {f −1 (B/2 (y))|y ∈ Y }. Let δ > 0 be the
Lebesgue number for A. Then dX (p, q) < δ implies that p, q ∈ Bδ (p) ⊂ f −1 (B/2 (y)) for
some y ∈ Y . Thus
dY (f (p), f (q)) ≤ dY (f (p), y) + dY (y, f (q)) < /2 + /2 = 

6.2 Compactness and Limits

Definition 44. A space X is called limit point compact if every infinite subset A ⊆ X
has a limit point.
Theorem 6.14. If X is compact, then it is limit point compact.
Proof. Suppose that X is compact and A ⊆ X has no limit points. We will show that A
is finite.
Since A has no limit points, we know that it contains all of its (non-existent) limit
points so by Corollary 3.7 we know that A = Ā is closed. Closed subspaces of compact
spaces are compact, so A is compact. Since A has no limit points, for each p ∈ A, there
exists an open set Up ⊂ X, such that Up ∩ {p} = {p}, so the singleton subsets of A are
open. The collection of singleton sets A := {{p}|p ∈ A} then forms an open cover, with
no proper sub-covers. Since A must admit a finite subcover, we conclude that A must be
finite.

49
 The converse of Theorem 6.14 does not hold.
Example 79. Let X = {a, b} be a two element set equipped with the trivial topology
(only {∅, X} are open). Now let Z have the discrete topology and consider the product
space Z × X. In this topology, any subset containing (n, a) has (n, b) as a limit point, and
vice-versa. Thus every non-empty subset of Z×X contains a limit point, so Z×X is limit
point compact. However, it is not compact because the open cover A := {{n} × X|n ∈ Z}
contains no finite subcovers.
Observe also that the projection map Z×X → Z is continuous and surjective, but Z is
not limit point compact (Z is discrete so no subset has any limit points). Thus continuous
images of limit point compact spaces are not limit point compact in general.
The preceding example explains why compactness is a more useful concept in general
topology than limit point compactness. However, for metric spaces they are the same
thing.
Theorem 6.15. A metric space (X, d) is compact if and only if it is limit point compact.
Proof. One direction has already been proven in Theorem 6.14, and remains to prove the
other.

Let (X, d) be a limit point compact space.

Claim 1: Every infinite sequence in (X, d) contains a convergent subsequence.

Proof. Let S := {x1 , ...xn , ...} be the set of values of the sequence. If S is finite, then some
value L ∈ S occurs infinitely often, so we can choose a constant subsequence which obvi-
ously converges. If S is infinite, then it has a limit point L (by limit point compactness).
Construct a subsequence (xn(i) )∞ i=1 converging to L recursively by

xn(i) ∈ B 1 (L) and n(i) > n(i − 1).

i

Claim 2: Every open covering of X admits a Lebesgue number.

Proof. Let A be an open cover of X. Assume for the sake of contradiction that there is
no Lebesgue number for A. That is, for every δ > 0, there exists p ∈ X such that Bδ (p)
is not contained in any element of A.
Construct a sequence, (xn )∞n=1 such that B1/n (xn ) is not contained in any element of
A. By the preceding lemma, we know that some subsequence (xn(i) )∞ i=1 converges to a
limit L.
Choose U ∈ A containing L. Then B (p) ⊂ U for some  > 0. If i is chosen so that
1
n(i)
< /2 and d(L, xn(i) ) < /2, then

B 1 (xn(i) ) ⊆ B/2 (xn(i) ) ⊆ B (L) ⊆ U

n(i)

which is a contradiction.

50
 Claim 3: For any  > 0, there exists a finite covering of X by -balls.
Proof. Assume such a cover does not Sexist for some  > 0. Construct a sequence by
x1 ∈ X, x2 ∈ X \ B (x1 ),...., xn ∈ X \ i<n B (xi ), .... For every m 6= n, d(xm , xn ) ≥  so
no subsequence is Cauchy, so no subsequence converges. Contradiction.
Finally, given an open covering A of X with Lebesgue number δ > 0, choose a finite
covering by δ-balls Bδ (x1 ), ...Bδ (xn ). Each Bδ (xi ) is contained inside an open set Ui ∈ A,
so {U1 , ..., Un } is a finite subcover and X is compact.

6.3 The One-Point Compactification

Because compactness is such a useful property, mathematicians often like to compactify
non-compact topological spaces.

Definition 45. A compactification of a topological space X is a compact topological

space Y and a continuous embedding i : X ,→ Y with dense image, i(X) = Y .

Example 80 (The stupid compactification). Let (X, τX ) be an arbitrary topological

space. Define a compactification,
Y = X ∪ {∗}
to be the disjoint union of X with a single point ∗, and with topology

τ = {U ⊆ X ⊂ Y | U is open in X} ∪ {Y }.

The obvious inclusion X ,→ Y is a dense embedding, and Y is compact because any open
covering of Y must contain Y .

The stupid compactification is stupid, because it doesn’t really tell us anything in-
teresting about X. For example, the only continuous real valued functions on Y are the
constant functions (prove this), even if X admits many interesting real-valued functions.
To get a nicer compactification, it helps to place a few mild conditions on X.

Definition 46. A topological space X is called locally compact if for every p ∈ X,

there exists an open neighbourhood p ∈ U ⊆ X such that the closure U is compact in the
subspace topology.

Example 81. Every compact space is locally compact (set U = X).

Example 82. Rn is locally compact in the Euclidean topology: Every point is contained in
an open neighbourhood of the form (a1 , b1 )×...×(an , bn ), whose closure [a1 , b1 ]×....×[an , bn ]
is compact.

Example 83. R∞ = RZ+ is not locally compact. The basic open sets have the form
(a1 , b1 )×...×(an , bn )×R×R×...., which have non-compact closure, [a1 , b1 ]×....×R×R......

51
Example 84. The subspace {(x, y) ∈ R2 | y < 0} ∪ {(0, 0)} ⊂ R2 is not locally compact
at (0, 0).

Theorem 6.16. Let X be a locally compact, Hausdorff space. There exists a compact
Hausdorff space X∞ such that,

• X ⊂ X∞ is a subspace.

• The complement X∞ − X is a single point, usually denoted ∞, called the point at

infinity.

This space X∞ is called the one-point compactification, and determined up to unique

isomorphism by these properties.

Remark 6. The one-point compactification is technically only a compactification if X

isn’t already compact to begin with. If X is compact, then X∞ ∼
`
= X {∞} (prove this).
Proof. As a set X∞ := X ∪ {∞}. The point ∞ is called the point at infinity. The open
sets of X∞ come in two types:

(1) U ⊆ X ⊂ X∞ is open in X.

(2) X∞ − K, where K ⊂ X is compact.

This is clearly a embedding, because every open set in X is of the form X ∩ U for
a set U of type (1), and X ∩ (X∞ − K) = X − K is open in X ( K is compact in a
Hausdorff space, hence closed). Every open neighbourhood of ∞ is of type (2), and if X
is not compact, then every such neighbourhood intersects X, so X is dense in X∞ .

Claim: X∞ is a topological space.

Proof. We must verify the axioms.

(i) ∅ is open of type (1) and X∞ is open of type (2).

(ii) We show that the intersection of two open sets is open. Let U, V be open sets of
type (1) and X∞ − K, X∞ − C open sets of type (2).

– U ∩ V is open in X so it is open of type (1).

– U ∩ (X∞ − K) = U ∩ (X − K) is open of type (1), because K is a compact
subset of a Hausdorff space, hence closed in X, so X − K is open in X.

52
 – (X∞ − K) ∩ (X∞ − C) = X∞ − (K ∪ C) is open of type (2) because K ∪ C is
a finite union of compact sets, hence compact (Homework exercise).

(iii) First observe that an arbitrary union of open sets of type (1) is open of type (1),
since it must be open in X. Now consider an arbitrary union of type (2) open sets
[ \
(X∞ − Kα ) = X∞ − ( Kα ).
α∈I α∈I

T
The Kα are closed in X, so α∈I Kα is S closed in X, hence compact because it is a closed
subset of a compact set Kα . Therefore α∈I (X∞ − Kα ) is open on type (2).
An arbitrary union of open sets may be expressed as the union of a union of type (1)
sets with a union of type (2) sets. By what we have shown, it is enough to show that if
U is type (1) and X∞ − K is type (2) then

U ∪ (X∞ − K) = X∞ − (K ∩ (X − U ))
is open of type (2), because (X − U ) is closed in X, so K ∩ (X − U ) is a closed subset of
a compact space, hence compact.
Claim: X∞ is compact.
Proof. Let A be an open covering of X∞ . Choose an open set V ∈ A containing ∞ must
be of type (2). Consequently K := X∞ − V is a compact set. Choose a finite subcovering
{U1 , ..., Un } of K, then {U1 , ..., Un , V } is a finite covering of X∞ .
Claim: X∞ is Hausdorff.
Proof. Certainly any two points in X are separated by open sets of type (1), because
X was Hausdorff to begin with. It remains to show that ∞ can be separated from any
other point p ∈ X. Because X is locally compact, there exists some open neighbourhood
p ∈ U ⊆ X, such that Ū is compact. Thus U and X∞ − Ū separate p and ∞.
We leave the verification that X∞ is unique up to isomorphism as an exercise.

Example 85. (Rn )∞ ∼

= S n.
Example 86. The cone satisfying x2 + y 2 = z 2 in R3 has compactification: ....
Proposition 6.17. Let A ⊆ X be a subspace of a locally compact Hausdorff space X. If
A is open or closed, then A is locally compact Hausdorff.
Proof. Any subspace of a Hausdorff space is Hausdorff (simply take the separating sets
in X and intersect them with A).
We first consider the case that A is open. Then A is also open in X∞ so X∞ − A
is closed, hence compact. Choose a point a ∈ A. For every point p ∈ X∞ − A, choose
separating open sets a ∈ Up and p ∈ Vp with Up ∩ Vp = ∅. The collection

A := {Vp |p ∈ X∞ − A}

53
is an open cover of the compact set X∞ − A, so we may choose a finite open cover
{Vp1 , ..., Vpn }. The intersection U := ∩ni=1 Upi ⊆ A is open in X∞ , hence also in A. The
closure Ū in X∞ is compact and satisfies

Ū ⊆ ∩ni=1 Ūi ⊆ A
so Ū ∩ A = Ū is also the closure of U as a subset of A, thus A is locally compact.
Now consider the case that A is closed. Then A ∪ {∞} is closed in X∞ , hence is
compact Hausdorff. Then A ⊂ A ∪ {∞} is an open subset of a compact Hausdorff space,
so we conclude that it must be locally compact.

Corollary 6.18. If a subset X ⊂ Rn is either open or closed, then it is locally compact

Hausdorff in the subspace topology. In particular, X admits a one-point compactification.

6.4 Compact-Open Topology (skipped in 2017)

Let X and Y be topological spaces, and let Cont(X, Y ) be the set of continuous maps
f : X → Y . Given subsets K ⊆ X and U ⊂ Y , define

S(K, U ) := {f ∈ Cont(X, Y )|f (K) ⊆ U }

Definition 47. The compact-open topology on Cont(X, Y ) is the topology generated by

the subbase {S(K, U )| K is compact and U is open}

Example 87. Let X = ∗ be the one-element space. Then there is a natural bijection

Cont(∗, Y ) ∼
= Y.

This bijection is a homeomorphism if Cont(∗, Y ) is endowed with the compact-open topol-

ogy.

Example 88. More generally, if I is any set endowed with the discrete topology, then
every map from I to Y is continuous, so there is a canonical bijection

Cont(I, Y ) ∼
= YI ∼
Y
= Y.
i∈I

This bijection is a homeomorphism between the compact-open topology on Cont(I, Y )

and the product topology on Y (Exercise).

Proposition 6.19. For general topological spaces, the compact-open topology on Cont(X, Y )
is (non-strictly) finer than the product topology (a.k.a. the pointwise convergence topol-
ogy). If Y is metrizable, then the compact-open topology is (non-strictly) coarser than the
uniform topology.

54
Proof. Let τp , τco , τu be the product, compact-open and uniform topologies respectively.
(1) τp ⊆ τco :
Recall that τp is the subspace topology for the inclusion

Cont(X, Y ) ⊆ Y X ∼
Y
= Y.
p∈X

The sub-base
Q of open sets for the product topology consist of preimages πp−1 (U ), where
πp : p∈X Y → Y is projection onto the pth factor and U ⊂ Y is open. In our new
notation, πp−1 (U ) = S({p}, U ), which is open in τco . So all the sub-base open sets in τp
are open in τco , so τp ⊂ τco .
(2) τco ⊆ τu :
It suffices to show that for compact K ⊆ X and open U ⊆ Y , that S(K, U ) ∈ τu .
Suppose for notational simplicity that the metric on Y , d = min{d, 1} is already bounded.
Let ρ be the uniform metric on Cont(X, Y ):

ρ(f, g) = sup{d(f (x), g(x))| x ∈ X}

Choose a element f ∈ S(K, U ) and let C := U c = Y −U . Define a function µ : K → R
by µ(x) := d(f (x), C). Then µ is continuous, because it is to composition of f with the
continuous function d(−, C) : Y → R. Since f (K) ⊆ U = C c , we know that µ(x) > 0 for
all x in K. Since K is compact, we deduce from the extreme value theorem that there
exists δ > 0, such that µ(x) ≥ δ for all x ∈ K. Consequently, if g ∈ Cont(X, Y ) satisfies
ρ(f, g) < δ, then for all x ∈ K we have d(g(x), C) ≥ d(f (x), C)−d(f (x), g(x)) > δ−δ = 0.
Thus g(K) ⊂ U and
Bδρ (f ) ⊆ S(K, U )
so S(K, U ) is open in τu .

7 Advanced Material
7.1 Separation Axioms
Let X be a topological set, and A, B ⊆ X a pair of disjoint subsets. We say that A, B
can be separated in X if there exist open sets, U, V ⊆ X such that A ⊆ U , B ⊆ V and
U ∩ V = ∅.

Definition 48. Let X be a topological space for which every singleton set is closed ( we
say all the points of X are closed).

• X is called Hausdorff if every pair of distinct of points p, q ∈ X can be separated.

• X is called Regular if every closed set A can be separated from any point p 6∈ A.

• X is called Normal if every pair of disjoint closed sets A, B ⊆ X can be separated.

55
 It is clear that each of these conditions is stronger than the last. That is

X is normal ⇒ X is regular ⇒ X is Hausdorff

We will be most interested in the normal condition in what follows. Many familiar spaces
are normal.

Proposition 7.1. Metric spaces are always normal.

Proof. Let A, B ⊆ X be a disjoint pair of closed sets in a metric space (X, d). For each
point a ∈ A ⊆ B c , there exists an (a) > 0 such that B(a) (a) is disjoint from B. Similarly,
for each b ∈ B, there exists (b) > 0 such that B(b) (b) is disjoint from A. Define
[ [
U := B(a)/3 (a), V := B(b)/3 (b).
a∈A b∈B

Certainly U and V are open and contain A and B respectively. To see that they are
disjoint, assume there exists p ∈ U ∩ V . Then for some a ∈ A and some b ∈ B, we have
p ∈ B(a)/3 (a) and p ∈ B(b)/3 (b). By the triangle inequality, this implies that

2
d(a, b) < ((a) + (b))/3 ≤ sup{(a), (b)}
3
which is a contradiction.

Proposition 7.2. Compact Hausdorff spaces are always normal.

Proof. Exercise.
The following equivalent condition will prove useful.

Proposition 7.3. A space X is normal if and only if for every pair of sets C ⊆ U ⊆ X
with C closed and U open, there exists another open set V with

C ⊆ V ⊆ V ⊆ U. (4)

Proof. We have C ⊆ U as in the hypothesis if and only if C and U c are disjoint closed
sets. If V satisfies equation (4), then V and (V )c separate C and U c . Conversely, if V, W
form a separation of C and U c , then C ⊆ V ⊆ V ⊆ W c ⊆ U .

7.2 Urysohn Lemma and Metrization Theorem

The following result is of basic importance to the study of normal spaces.

Lemma 7.4 (Urysohn Lemma). Let X be a normal space and let A and B be disjoint
closed subsets. Let [a, b] be a closed interval in R. There exists a continuous map f : X →
[a, b] such that f (A) = a and f (B) = b.

56
Remark 7. If such a function exists for every pair A, B, then X is normal because
f −1 ([a, a + ]) and f −1 ((b − , b]) separates A and B. Thus we may regard the Urysohn
Lemma as providing a third equivalent definition of Normality: that every pair of closed
spaces be “separated” by a continuous real valued function ( an analogous statement holds
for Hausdorff and regular spaces).

Proof. For simplicity we work with [0, 1] = [a, b] (no loss of generality, since [0, 1] ∼
= [a, b]).
−1
Our basic strategy is to define f by choosing open pre-images Ur := f ([0, r)) for
every r ∈ Q ∩ [0, 1] and then extending continuously.

Step 1: Construct a family of open sets {Ur | r ∈ Q ∩ [0, 1]} with the properties (*)

• A ⊆ Ur and Ur ∩ B = ∅ for all r.

• r < s ⇒ U r ⊆ Us .

We begin by setting U1 = B c . Choose U0 so that A ⊆ U0 ⊆ U 0 ⊆ U1 (this is possible by

Proposition 7.3).
Now observe that Q ∩ [0, 1] is countable so it can be ordered into a sequence {ri }∞
i=0 .
We choose r0 = 0 and r1 = 1. We construct the remaining Urn for n ≥ 2 inductively as
follows:
Let p = max{ri |i < n and ri < rn } and q = min{ri | i < n and ri > rn }. Then select
Urn so that

U p ⊆ Urn ⊆ U rn ⊆ Uq .
We now extend this collection to {Ur |r ∈ Q} by setting Ur = ∅ if r < 0 and Ur = X
is r > 1. This still satisfies properties (*).
Step 2: Define a function f : X → R by

f (x) = inf{r ∈ Q|x ∈ Ur }.

Observe that the image f (X) is contained in [0, 1] because Ur = ∅ for r < 0 and
Ur = X for r > 1. Also, f (A) = 0, because A ⊆ U0 and f (B) = 1 because U1 ∩ B = ∅.
It only remains to prove that f is continuous.
Claim: For s ∈ Q, then the following implications are true:

(i) f (x) < s ⇒ x ∈ Us ⇒ f (x) ≤ s.

(ii) f (x) > s ⇒ x 6∈ Us ⇒ f (x) ≥ s.

57
Proof. Note that the first line is logically equivalent to the second (contrapositives) so it
suffices to prove the first.
If f (x) < s, then by definition there exists a rational number r < s such that x ∈ Ur .
But then Ur ⊆ Us , so x ∈ Us .
If x ∈ Us , then f (x) is the infimum of a set containing s so f (x) ≤ s.
Let (a, b) ⊆ R be an open interval in R. Since such intervals form a base for the
Euclidean topology on R it will suffice to show that f −1 ((a, b)) is open.
Let x ∈ f −1 ((a, b)) and choose rational numbers p, p0 , q so that

a < p < p0 < f (x) < q < b.

Define the open set V = Uq ∩ (Ūp )c .

For any y ∈ V ⊆ Uq ∩ Upc we have f (y) ∈ [p, q] ⊂ (a, b) so V ⊆ f −1 ((a, b)) and
c
x ∈ Uq ∩ Upc0 ⊆ Uq ∩ U p = V.
Thus f −1 ((a, b)) is open and f is continuous.

Definition 49. A topological space X is called second countable if can be generated by

a countable base of open sets.

Example 89. Rn is second countable because it is generated by B := {B (x̄) | x̄ ∈ Qn ,  ∈

Q>0 }. Also every subspace of Rn is second countable, by restricting B.

Theorem 7.5 (Urysohn’s Metrization Theorem). Every normal, second countable space
X is metrizable.

Proof. Consider the set [0, 1]Z+ equipped with the uniform metric topology,

ρ((xi )∞ ∞
i=1 , (yi )i=1 ) = sup{|xi − yi | | i ∈ Z+ }.

We will construct an embedding

F : X ,→ [0, 1]Z+

so that X ∼
= F (X) acquires a metric by restriction.

Step 1: There exists a countable collection of continuous maps

{fn : X → [0, 1]}∞

n=1

such that

(a) fn (X) ⊆ [0, 1/n]

(b) For all open U ⊆ X and p ∈ U , there exists fn such that fn (p) > 0 and f (U c ) = 0.

58
Proof. Since X is second countable, there exists a countable base B := {Bk }∞
k=1 . For each
pair of k, l ∈ Z+ such that B k ⊆ Bl , choose a function

gk,l : X → [0, 1]

such that gk,l (B k ) = 1 and gk,l (Blc ) = 0 (gk,l exists because X is normal).
By the definition of a base, for every open U and p ∈ U , there exists a basic open
set with p ∈ Bl ⊆ U . Because X is normal, there exists another open set which we may
choose to be basic, such that
p ∈ Bk ⊆ B k ⊆ Bl
(remember points are closed) so that gk,l (p) = 1 and gk,l (U c ) = 0.
The collection {gk,l } is countable, so we may relabel {gk,l } = {gn }∞
n=1 . Setting fn :=
1
g completes the construction.
n n

We use the fn as coordinate functions to define

F = (fn )∞
n=1 : X → [0, 1] .
Z+

Note that F is by definition continuous for the product topology on [0, 1]Z+ but is not
automatically continuous with respect to the uniform topology.

Step 2: F is injective.
Suppose that x, y ∈ X and x 6= y. Since X is normal, it is also Hausdorff, so there
exists an open set U with x ∈ U , y 6∈ U . Simply choose fn so that fn (x) > 0 and
fn (U c ) = fn (y) = 0. It follows that F (x) 6= F (y).

Step 3: F is continuous.
Let p ∈ X. We must show that for any  > 0, there exists an open neighbourhood
p ∈ U ⊆ X open, such that F (U ) ⊆ B (F (p)) or equivalently, such that

x ∈ U ⇒ ρ(F (x), F (p)) < .

Choose M ∈ Z+ large enough so that 1/M < /2. Because the fn are all continuous, we
may choose for each n ≤ M an open set p ∈ Un such that fn (Un ) ⊆ B/2 (fn (p)). Set
U = U1 ∩ .... ∩ UM . If x ∈ U , then

(
if n ≤ M because x ∈ U ⊆ Un
|fn (x) − fn (p)| < /2
if n > M because f (X) ⊂ [0, 1/M ] ⊂ [0, /2].

Consequently ρ(F (x), F (p)) = sup{|fn (x) − fn (p)|} ≤ /2 < .

Step 4: F −1 : F (X) → X is continuous.

Proof. Given U ⊆ X open, we must show (F −1 )−1 (U ) = F (U ) is open in F (X).
It suffices to show that for each U ⊂ X open and p ∈ U there exists an open set
V ⊂ [0, 1]Z+ such that
F (p) ∈ V ∩ F (X) ⊆ F (U ).

59
Choose n ∈ Z+ so that fn (p) > 0 and fn (U c ) = 0. Set

V := πn−1 ((0, 1]) ⊆ [0, 1]Z+

where πn : [0, 1]Z+ → [0, 1] is the nth projection map. Then V is open in the product
topology, hence also open in the uniform topology. Furthermore

p ∈ F −1 (V ) = fn−1 ((0, 1]) ⊆ U

so f (p) ∈ V ∩ F (X) ⊆ F (U )
We have shown that X embeds as a subspace of a metric space, hence is metrizable.

7.3 Tychonoff ’s Theorem

Our goal in this section is to prove

Theorem 7.6 (Tychonoff’s Theorem). Every product of compact spaces is compact.

The proof requires some groundwork. First, we introduce a new characterization of

compactness.

Definition 50. A collection A of subsets of X has the finite intersection property or

f.i.p. if every finite subcollection {A1 , ..., An } ∈ A satisfies

A1 ∩ ... ∩ An 6= ∅.

Proposition 7.7. A space X is compact if and only ifTfor every collection A of closed
subsets possessing the f.i.p., the intersection of all sets A∈A A is nonempty.

Proof. Let X be a compact set and A a collection of closed sets. Let à = {Ac |A ∈ A}
be the complementary collection of open sets. Using de Morgan’s law

(∩n An )c = ∪n Acn

we see that A has f.i.p. if and only if à contains no finite subcovers. Since X is compact,
this implies that à is not a covering, which again by de Morgan’s law implies that the
intersection of all elements in A is non-empty. The converse follows by similar reasoning.

Lemma 7.8. Let A be a collection of subsets of X (not necessarily closed) possessing the
f.i.p.. Then there exists a larger collection D ⊇ A possessing the f.i.p. which is maximal
in the sense that any strictly larger collection does not have f.i.p..

Example 90. Let X be any set, let p ∈ X, and set A = {{p}}. Then D := {A ⊆ X|p ∈
A} is the unique maximal collection possessing f.i.p. and containing A.

60
Proof of Lemma. Consider the set

Λ := {E | E is a collection of subsets of X possessing f.i.p. and E ⊇ A}

partially ordered by inclusion. We are seeking a maximal element of Λ and will use Zorn’s
Lemma to prove its existence.
It remains to show that Λ satisfies the hypotheses of Zorn’s Lemma. Let S ⊂ Λ be a
linearly ordered subset, i.e. if E1 , E2 ∈ S then either E1 ⊆ E2 or E2 ⊆ E1 .
S
Claim: E = Ei ∈S Ei is an upperbound of S in Λ.
Proof. Clearly as a collection of sets E contains every collection in S. We need only show
that E possesses the f.i.p.. Choose any finite collection of sets A1 , ..., An ∈ E. For each
1 ≤ k ≤ n there exists Ek ∈ S such that Ak ∈ Ek . Amongst these there is one, say Ei
which contains all the other Ek (any finite linearly ordered set contains an upperbound).
Thus A1 , ..., An ∈ Ei and since Ei possesses f.i.p., we get A1 ∩...∩An 6= ∅. Thus E possesses
f.i.p..

Corollary 7.9. X is compact if and only if every maximal collection of subsets possessing
the f.i.p. D satisfies \
Ā 6= ∅.
A∈D

Proof.
T If X is compact and D possesses f.i.p. then {Ā|A ∈ D} also possesses f.i.p. so
A∈D Ā 6= ∅ by Proposition 7.7.
Conversely, if A is a collection of closed sets possessing f.i.p. and D ⊇ A is a maximal
collection possessing f.i.p. then
\ \ \
A= Ā ⊇ B̄ 6= ∅
A∈A A∈A B∈D

implies that X is compact.

Lemma 7.10. Let D be a collection of subsets of X which is maximal with respect to the
f.i.p.. Then
(a) If A ∈ D and A ⊆ B, then B ∈ D.

(b) D is closed under finite intersections.

(c) If B ∩ A 6= ∅ for every A ∈ D, then B ∈ D.

Proof. (a)This is clear.
(b) Let {A1 , ..., An } ⊆ D and let A := A1 ∩ ... ∩ An . If B1 , ..., Bm ∈ D then

A ∩ B1 ∩ ... ∩ Bm = A1 ∩ ... ∩ An ∩ B1 ... ∩ Bm 6= ∅

so D ∪ {A} possesses f.i.p.. By maximality, A ∈ D.

61
 (c) Let A1 , ..., An ∈ D. Then

B ∩ A1 ∩ ... ∩ An = B ∩ A 6= ∅,

so B ∈ D by maximality.
Proof of Tychonoff ’s Theorem. Let {Xα }α∈I be a collection of compact spaces. We want
to show that the product space Y
X := Xα
α∈I

is compact. Let T D be a maximal collection of subsets of X possessing the f.i.p.. Our goal
is to show that A∈D Ā 6= ∅.
Denote by πα : X → Xα the projection map. Consider the collection {πα (A) | A ∈ D}
of subsets of Xα . This collection possesses the finite intersection property because for any
{A1 , ..., An } ⊂ D,
n
\ n
\
πα (Ai ) ⊇ πα ( Ai ) 6= ∅
i=1 i=1

Since Xα is compact, there exists a point

\
pα ∈ πα (A) 6= ∅.
A∈D

Claim: The point p = (pα )α∈I ∈ Ā for all A ∈ D.

Proof. The claim holds if and only if every open neighbourhood of p ∈ U ⊆ X intersects
every A ∈ D, and by Lemma 7.10 this holds if and only if every open neighbourhood
of p is contained in D. It suffices to prove this for sub-basic open neighbourhoods of p,
because D is closed under finite intersections and unions by Lemma 7.10.
Recall that the sub-basic open sets in X are those of the form πα−1 (Uα ) where Uα is
open in Xα . Now π −1 (Uα ) contains p if and only if pα ∈ Uα . For every A ∈ D, we have
pα ∈ πα (A) so we conclude that

Uα ∩ πα (A) 6= ∅ ⇒ Uα ∩ πα (A) 6= ∅ (since Uα is open) ⇒ π −1 (Uα ) ∩ A 6= ∅.

which implies by part (b) Lemma 7.10 that π −1 (Uα ) ∈ D, completing the proof.

62