Topologı́a

Gabriel Andre Asmat Medina

Universidad Nacional de Ingenierı́a, 2021

Topology Atlas
Toronto, Ontario, Canada
http://at.yorku.ca/topology/
[email protected]

Aisling McCluskey
National University of Ireland, Galway
Ollscoil na hÉireann, Gaillimh
[email protected]

T.B.M. McMaster
Queen’s University Belfast
[email protected]

Cataloguing in Publication Data

Topology course lecture notes / by Aisling McCluskey and T.B.M. Mc-
Master.
iv, 44 p.
ISBN 0-9730867-4-?
1. Topology I. McCluskey, Aisling II. McMaster, T.B.M. III. Title.
Dewey QA303
MSC (2000) 54-01

Copyright c 1997 Aisling McCluskey and T.B.M. McMaster. All rights

reserved. Individual users of this publication are permitted to make
fair use of the material in teaching, research and reviewing. No part
of this publication may be used for commercial redistribution without
the prior permission of the publisher or the copyright holder.

ISBN 0-9730867-3-?
Published August 1997. Preliminary versions of this document were
distributed on the World Wide Web. This document is available in
several formats for printing, online browsing and viewing; Please con-
tact the publisher for details.
 Contents

Chapter 1. Fundamental Concepts 1

1.1. Describing Topological Spaces 1
1.2. Closed sets and Closure 5
1.3. Continuity and Homeomorphism 7
1.4. Additional Observations 10
Chapter 2. Topological Properties 11
2.1. Compactness 11
2.2. Other Covering Conditions 17
2.3. Connectedness 17
2.4. Separability 21
Chapter 3. Convergence 23
3.1. The Failure of Sequences 23
3.2. Nets—A Kind of ‘Super-Sequence’ 24
3.3. First Countable Spaces—Where Sequences Suffice 27
Chapter 4. Product Spaces 31
4.1. Constructing Products 31
4.2. Products and Topological Properties 34
Chapter 5. Separation Axioms 37
5.1. T1 Spaces 37
5.2. T2 (Hausdorff) Spaces 38
5.3. T3 Spaces 40
5.4. T3 1 Spaces 40
2
5.5. T4 Spaces 41

i
 CHAPTER 1

Fundamental Concepts

In the study of metric spaces, we observed that:

(i) many of the concepts can be described purely in terms of open
sets,
(ii) open-set descriptions are sometimes simpler than metric de-
scriptions, e.g., continuity,
(iii) many results about these concepts can be proved using only
the basic properties of open sets (namely, that both the empty
set and the underlying set X are open, that the intersection
of any two open sets is again open and that the union of arbi-
trarily many open sets is open).
This prompts the question: How far would we get if we started with a
collection of subsets possessing these above-mentioned properties and
proceeded to define everything in terms of them?

1.1. Describing Topological Spaces

We noted above that many important results in metric spaces can
be proved using only the basic properties of open sets that
• the empty set and underlying set X are both open,
• the intersection of any two open sets is open, and
• unions of arbitrarily many open sets are open.
We will call any collection of sets on X satisfying these properties a
topology. In the following section, we also seek to give alternative ways
of describing this important collection of sets.

1.1.1. Defining Topological Spaces.

Definition 1.1. A topological space is a pair (X, T ) consisting
of a set X and a family T of subsets of X satisfying the following
conditions:
(T1) ∅ ∈ T and X ∈ T ;
(T2) T is closed under arbitrary union;
(T3) T is closed under finite intersection.
1
2 1. FUNDAMENTAL CONCEPTS

The set X is called a space, the elements of X are called points of

the space and the subsets of X belonging to T are called open in the
space; the family T of open subsets of X is also called a topology for
X.
Examples.
(i). Any metric space (X, d) is a topological space where Td , the
topology for X induced by the metric d, is defined by agreeing that G
shall be declared as open whenever each x in G is contained in an open
ball entirely in G, i.e.,
∅ ⊂ G ⊆ X is open in (X, Td )
⇔ ∀x ∈ G, ∃rx > 0 such that x ∈ Brx (x) ⊆ G.
(ii). The following is a special case of (i), above. Let R be the set
of real numbers and let I be the usual (metric) topology defined by
agreeing that
∅ ⊂ G ⊆ X is open in (R, I) (alternatively, I-open)
⇔ ∀x ∈ G, ∃rx > 0 such that (x − rx , x + rx ) ⊂ G.
(iii). Define T0 = {∅, X} for any set X—known as the trivial or
anti-discrete topology.
(iv). Define D = {G ⊆ X : G ⊆ X}—known as the discrete topol-
ogy.
(v). For any non-empty set X, the family C = {G ⊆ X : G =
∅ or X \ G is finite} is a topology for X called the cofinite topology.
(vi). For any non-empty set X, the family L = {G ⊆ X : G =
∅ or X \ G is countable} is a topology for X called the cocountable
topology.
1.1.2. Neighbourhoods. Occasionally, arguments can be simpli-
fied when the sets involved are not “over-described”. In particular, it
is sometimes suffices to use sets which contain open sets but are not
necessarily open. We call such sets neighbourhoods.
Definition 1.2. Given a topological space (X, T ) with x ∈ X,
then N ⊆ X is said to be a (T )-neighbourhood of x ⇔ there exist
an open set G with x ∈ G ⊆ N .
It follows then that a set U ⊆ X is open iff for every x ∈ U , there
exists a neighbourhood Nx of x contained in U . (Check this!)
Lemma 1.3. Let (X, T ) be a topological space and, for each x ∈ X,
let N (x) be the family of neighbourhoods of x. Then
(i) U ∈ N (x) ⇒ x ∈ U .
 1.1. DESCRIBING TOPOLOGICAL SPACES 3

(ii) N (x) is closed under finite intersections.

(iii) U ∈ N (x) and U ⊆ V ⇒ V ∈ N (x).
(iv) U ∈ N (x) ⇒ there exists W ∈ N (x) such that W ⊆ U and
W ∈ N (y) for each y ∈ W .
Proof. Exercise! 
Examples.
(i). Let x ∈ X, and define Tx = {∅, {x}, X}. Then Tx is a topology
for X and V ⊆ X is a neighbourhood of x iff x ∈ V . However, the
only neighbourhood of y ∈ X where y 6= x is X itself.
(ii). Let x ∈ X and define a topology I(x) for X as follows:
I(x) = {G ⊆ X : x ∈ G} ∪ {∅}.
Note here that every neighbourhood of a point in X is open.
(iii). Let x ∈ X and define a topology E(x) for X as follows:
E(x) = {G ⊆ X : x 6∈ G} ∪ {X}.
Note here that {y} is open for every y 6= x in X, that {x, y} is not
open, is not a neighbourhood of x yet is a neighbourhood of y. In fact,
the only neighbourhood of x is X.
1.1.3. Bases and Subbases. It often happens that the open sets
of a space can be very complicated and yet they can all be described
using a selection of fairly simple special ones. When this happens, the
set of simple open sets is called a base or subbase (depending on how
the description is to be done). In addition, it is fortunate that many
topological concepts can be characterized in terms of these simpler base
or subbase elements.
Definition 1.4. Let (X, T ) be a topological space. A family B ⊆
T is called a base (X, T ) if and only if every non-empty open subset
of X can be represented as a union of a subfamily of B.
It is easily verified that B ⊆ T is a base for (X, T ) if and only if
whenever x ∈ G ∈ T , there exists B ∈ B such that x ∈ B ⊆ G.
Clearly, a topological space can have many bases.
Lemma 1.5. If B is a family of subsets of a set X such that
(B1) for any B1 , B2 ∈ B and every point x ∈ B1 ∩ B2 , there exists
B3 ∈ B with x ∈ B3 ⊆ B1 ∩ B2 , and
(B2) for every x ∈ X, there exists B ∈ B such that x ∈ B,
then B is a base for a unique topology on X. Conversely, any base B
for a topological space (X, T ) satisfies (B1) and (B2).
Proof. Exercise! 
4 1. FUNDAMENTAL CONCEPTS

Definition 1.6. Let (X, T ) be a topological space. A family S ⊆

T is called a subbase for (X, T ) if and only if the family of all finite
intersections ki=1 Ui , where Ui ∈ S for i = 1, 2, . . . , k is a base for
T
(X, T ).

Examples.
(i). In any metric space (X, d), {Br (x) : x ∈ X, r > 0} forms a base
for the induced metric topology Td on X.
(ii). For the real line R with its usual (Euclidean) topology, the
family {(a, b) : a, b ∈ Q, a < b} is a base.
(iii). For an arbitrary set X, the family {{x} : x ∈ X} is a base for
(X, D).
(iv). The family of all ‘semi-infinite’ open intervals (a, ∞) and (−∞, b)
in R is a subbase for (R, I).

1.1.4. Generating Topologies. From the above examples, it fol-

lows that for a set X one can select in many different ways a family T
such that (X, T ) is a topological space. If T1 and T2 are two topologies
for X and T2 ⊆ T1 , then we say that the topology T1 is finer than the
topology T2 , or that T2 is coarser than the topology T1 . The discrete
topology for X is the finest one; the trivial topology is the coarsest.
If X is an arbitrary infinite set with distinct points x and y, then one
can readily verify that the topologies I(x) and I(y) are incomparable,
i.e., neither is finer than the other.
By generating a topology for X, we mean selecting a family T of
subsets of X which satisfies conditions (T1)–(T3). Often it is more
convenient not to describe the family T of open sets directly. The
concept of a base offers an alternative method of generating topologies.
Examples.
[Sorgenfrey line]. Given the real numbers R, let B be the family of
all intervals [x, r) where x, r ∈ R, x < r and r is rational. One can
readily check that B has properties (B1)–(B2). The space Rs , generated
by B, is called the Sorgenfrey line and has many interesting properties.
Note that the Sorgenfrey topology is finer than the Euclidean topology
on R. (Check!)
[Niemytzki plane]. Let L denote the closed upper half-plane. We
define a topology for L by declaring the basic open sets to be the
following:
(I) the (Euclidean) open discs in the upper half-plane;
(II) the (Euclidean) open discs tangent to the ‘edge’ of the L, to-
gether with the point of tangency.
 1.2. CLOSED SETS AND CLOSURE 5


6
qi


i
q
  -

Figure 1. Niemytzki plane

Note. If yn → y in L, then
(i) y not on ‘edge’: same as Euclidean convergence.
(ii) y on the ‘edge’: same as Euclidean, but yn must approach y
from ‘inside’. Thus, for example, yn = ( n1 , 0) 6→ (0, 0)!

1.1.5. New Spaces from Old. A subset of a topological space

inherits a topology of its own, in an obvious way:
Definition 1.7. Given a topological space (X, T ) with A ⊆ X,
then the family TA = {A ∩ G : G ∈ T } is a topology for A, called the
subspace (or relative or induced) topology for A. (A, TA ) is called
a subspace of (X, T ).
Example. The interval I = [0, 1] with its natural (Euclidean) topol-
ogy is a (closed) subspace of (R, I).
Warning. Although this definition, and several of the results which
flow from it, may suggest that subspaces in general topology are going
to be ‘easy’ in the sense that a lot of the structure just gets traced onto
the subset, there is unfortunately a rich source of mistakes here also:
because we are handling two topologies at once. When we inspect a
subset B of A, and refer to it as ’open’ (or ’closed’, or a ’neighbour-
hood’ of some point p . . . ) we must be exceedingly careful as to which
topology is intended. For instance, in the previous example, [0, 1] itself
is open in the subspace topology on I but, of course, not in the ’back-
ground’ topology of R. In such circumstances, it is advisable to specify
the topology being used each time by saying T -open, TA -open, and so
on.

1.2. Closed sets and Closure

Just as many concepts in metric spaces were described in terms of
basic open sets, yet others were characterized in terms of closed sets.
In this section we
6 1. FUNDAMENTAL CONCEPTS

• define closed sets in a general topological space and

• examine the related notion of the closure of a given set.
1.2.1. Closed Sets.
Definition 1.8. Given a topological space (X, T ) with F ⊆ X,
then F is said to be T -closed iff its complement X \ F is T -open.
From De Morgan’s Laws and properties (T1)–(T3) of open sets,
we infer that the family F of closed sets of a space has the following
properties:
(F1) X ∈ F and ∅ ∈ F;
(F2) F is closed under finite union;
(F3) F is closed under arbitrary intersection.
Sets which are simultaneously closed and open in a topological space
are sometimes referred to as clopen sets. For example, members of
the base B = {[x, r) : x, r ∈ R, x < r, r rational} for the Sorgenfrey
line are clopen with respect to the topology generated by B. Indeed,
for the discrete space (X, D), every subset is clopen.
1.2.2. Closure of Sets.
Definition 1.9. If (X, T ) is a topological space and A ⊆ X, then
T \
A = {F ⊆ X : A ⊆ F and F is closed}
is called the T -closure of A.
T
Evidently, A (or A when there is no danger of ambiguity) is the
smallest closed subset of X which contains A. Note that A is closed
⇔ A = A.
Lemma 1.10. If (X, T ) is a topological space with A, B ⊆ X, then
(i) ∅ = ∅;
(ii) A ⊆ A;
(iii) A = A;
(iv) A ∪ B = A ∪ B.
Proof. Exercise! 
Theorem 1.11. Given a topological space with A ⊆ X, then x ∈ A
iff for each neighbourhood U of x, U ∩ A =
6 ∅.
Proof. ⇒: Let x ∈ A and let U be a neighbourhood of x; then
there exists open G with x ∈ G ⊆ U . If U ∩ A = ∅, then G ∩ A = ∅ and
so A ⊆ X \ G ⇒ A ⊆ X \ G whence x ∈ X \ G, thereby contradicting
the assumption that U ∩ A = ∅.
 1.3. CONTINUITY AND HOMEOMORPHISM 7

⇐: If x ∈ X \ A, then X \ A is an open neighbourhood of x so that,

by hypothesis, (X \ A) ∩ A 6= ∅, which is a contradiction (i.e., a false
statement). 
Examples.
(i). For an arbitrary infinite set X with the cofinite topology C, the
closed sets are just the finite ones together with X. So for any A ⊆ X,
(
A if A is finite,
A=
X if A is infinite.
Note that any two non-empty open subsets of X have non-empty in-
tersection.
(ii). For an arbitrary uncountable set X with the cocountable topol-
ogy L, the closed sets are the countable ones and X itself. Note that
if we let X = R, then [0, 1] = R! (In the usual Euclidean topology,
[0, 1] = [0, 1].)
(iii). For the space (X, Tx ) defined earlier, if ∅ ⊂ A ⊆ X, then
(
X if x ∈ A,
A=
X \ {x} if x 6∈ A.
(iv). For (X, I(x)) with A ⊆ X,
(
A if x 6∈ A,
A=
X if x ∈ A.
(v). For (X, E(x)) with ∅ ⊂ A ⊆ X,
(
A if x ∈ A,
A=
A ∪ {x} if x ∈6 A,
(vi). In (X, D), every subset equals its own closure.

1.3. Continuity and Homeomorphism

The central notion of continuity of functions is extended in this
section to general topological spaces. The useful characterization of
continuous functions in metric spaces as those functions where the in-
verse image of every open set is open is used as a definition in the
general setting.
Because many properties of spaces are preserved by continuous
functions, spaces related by a bijection (one-to-one and onto function)
which is continuous in both directions will have many properties in com-
mon. These properties are identified as topological properties. Spaces
so related are called homeomorphic.
8 1. FUNDAMENTAL CONCEPTS

1.3.1. Continuity. The primitive intuition of a continuous pro-

cess is that of one in which small changes in the input produce small,
’non-catastrophic’ changes in the corresponding output. This idea for-
malizes easily and naturally for mappings from one metric space to
another: f is continuous at a point p in such a setting whenever we
can force the distance between f (x) and f (p) to be as small as is
desired, merely by taking the distance between x and p to be small
enough. That form of definition is useless in the absence of a properly
defined ’distance’ function but, fortunately, it is equivalent to the de-
mand that the preimage of each open subset of the target metric space
shall be open in the domain. Thus expressed, the idea is immediately
transferrable to general topology:
Definition 1.12. Let (X, T ) and (Y, S) be topological spaces; a
mapping f : X → Y is called continuous iff f −1 (U ) ∈ T for each
U ∈ S, i.e., the inverse image of any open subset of Y is open in X.
Examples.
(i). If (X, D) is discrete and (Y, S) is an arbitrary topological space,
then any function f : X → Y is continuous!
Again, if (X, T ) is an arbitrary topological space and (Y, T0 ) is
trivial, any mapping g : X → Y is continuous.
(ii). If (X, T ), (Y, S) are arbitrary topological spaces and f : X →
Y is a constant map, then f is continuous.
(iii). Let X be an arbitrary set having more than two elements, with
x ∈ X. Let T = I(x), S = Tx in the definition of continuity; then the
identity map idX : X → X is continuous. However, if we interchange
T with S so that T = Tx and S = I(x), then idX : X → X is not
continuous! Note that idX : (X, T1 ) → (X, T2 ) is continuous if and only
if T1 is finer than T2 .
Theorem 1.13. If (X1 , T1 ), (X2 , T2 ) and (X3 , T3 ) are topological
spaces and h : X1 → X2 and g : X2 → X3 are continuous, then g ◦
h : X1 → X3 is continuous.
Proof. Immediate. 
There are several different ways to ’recognise’ continuity for a map-
ping between topological spaces, of which the next theorem indicates
two of the most useful apart from the definition itself:
Theorem 1.14. Let f be a mapping from a topological space (X1 , T1 )
to a topological space (X2 , T2 ). The following statements are equivalent:
(i) f is continuous,
 1.3. CONTINUITY AND HOMEOMORPHISM 9

(ii) the preimage under f of each closed subset of X2 is closed in

X1 ,
(iii) for every subset A of X1 , f (A) ⊆ f (A).
Proof. It is easy to see that (i) implies (ii). Assuming that (ii)
holds, apply it to the closed set f (A) and (iii) readily follows. Now if
(iii) is assumed and G is a given open subset of X2 , use (iii) on the
set A = X1 \ f −1 (G) and verify that it follows that f −1 (G) must be
open. 
1.3.2. Homeomorphism.
Definition 1.15. Let (X, T ), (Y, S) be topological spaces and let
h : X → Y be bijective. Then h is a homeomorphism iff h is con-
tinuous and h−1 is continuous. If such a map exists, (X, T ) and (Y, S)
are called homeomorphic.
Such a map has the property that
G ∈ T ⇔ f (G) ∈ S.
It follows that any statement about a topological space which is ul-
timately expressible solely in terms of the open sets (together with
set-theoretic relations and operations) will be true for both (X, T ) and
(Y, S) if it is true for either. In other words, (X, T ) and (Y, S) are in-
distinguishable as topological spaces. The reader who has had abstract
algebra will note that homeomorphism is the analogy in the setting of
topological spaces and continuous functions to the notion of isomor-
phism in the setting of groups (or rings) and homomorphisms, and to
that of linear isomorphism in the context of vector spaces and linear
maps.
Example. For every space (X, T ), the identity mapping idX : X →
X is a homeomorphism.
A property of topological spaces which when possessed by a space
is also possessed by every space homeomorphic to it is called a topo-
logical invariant. We shall meet some examples of such properties
later.
One can readily verify that if f is a homeomorphism, then the in-
verse mapping f −1 is also a homeomorphism and that the composition
g◦f of two homeomorphisms f and g is again a homeomorphism. Thus,
the relation ‘X and Y are homeomorphic’ is an equivalence relation.
In general, it may be quite difficult to demonstrate that two spaces
are homeomorphic (unless a homeomorphism is obvious or can easily
be discovered). For example, to verify that (R, I) is homeomorphic to
(0, 1) with its induced metric topology, it is necessary to demonstrate,
10 1. FUNDAMENTAL CONCEPTS

2x−1
for instance, that h : (0, 1) → R where h(x) = x(x−1) is a homeomor-
phism.
It is often easier to show that two spaces are not homeomorphic:
simply exhibit an invariant which is possessed by one space and not
the other.
Example. The spaces (X, I(x)) and (X, E(x)) are not homeomor-
phic since, for example, (X, I(x)) has the topological invariant ‘each
neighbourhood is open’ while (X, E(x)) does not.
1.4. Additional Observations
Definition 1.16. A sequence (xn ) in a topological space (X, T ) is
said to converge to a point x ∈ X iff (xn ) eventually belongs to every
neighbourhood of x, i.e., iff for every neighbourhood U of x, there exists
n0 ∈ N such that xn ∈ U for all n ≥ n0 .
Caution. We learnt that, for metric spaces, sequential convergence
was adequate to describe the topology of such spaces (in the sense that
the basic primitives of ‘open set’, ‘neighbourhood’, ‘closure’ etc. could
be fully characterised in terms of sequential convergence). However, for
general topological spaces, sequential convergence fails. We illustrate:
(i). Limits are not always unique. For example, in (X, T0 ), each
sequence (xn ) converges to every x ∈ X.
(ii). In R with the cocountable topology L, [0, 1] is not closed and
so G = (−∞, 0) ∪ (1, ∞) is not open—yet if xn → x where x ∈ G,
then Assignment 1 shows that xn ∈ G for all sufficiently large n. Fur-
L
ther, 2 ∈ [0, 1] , yet no sequence in [0, 1] can approach 2. So another
characterisation fails to carry over from metric space theory.
Finally, every L-convergent sequence of points in [0, 1] must have
its limit in [0, 1]—but [0, 1] is not closed (in L)!
Hence, to discuss topological convergence thoroughly, we need to
develop a new basic set-theoretic tool which generalises the notion of
sequence. It is called a net—we shall return to this later.
Definition 1.17. A topological space (X, T ) is called metrizable
iff there exists a metric d on X such that the topology Td induced by
d coincides with the original topology T on X.
The investigations above show that (X, T0 ) and (R, L) are exam-
ples of non-metrizable spaces. However, the discrete space (X, D) is
metrizable, being induced by the discrete metric
(
1 if x 6= y,
d(x, y) =
0 if x = y.
 CHAPTER 2

Topological Properties

We explained in the previous chapter what a topological property

(homeomorphic invariant) is but gave few good examples. We now
explore some of the most important ones. Recurring themes will be:
• When do subspaces inherit the property?
• How do continuous maps relate to the property?
• Does the property behave specially in metric spaces?
2.1. Compactness
We all recall the important and useful theorem from calculus, that
functions which are continuous on a closed and bounded interval take
on a maximum and minimum value on that interval. The classic the-
orem of Heine-Borel-Lebesgue asserts that every covering of such an
interval by open sets has a finite subcover. In this section, we use
this feature of closed and bounded subsets to define the correspond-
ing notion, compactness, in a general topological space. In addition,
we consider important variants of this notion: sequential compactness
and local compactness.
2.1.1. Compactness Defined. Given a set X with A ⊆ X, a
coverSfor A is a family of subsets U = {Ui : i ∈ I} of X such that
A ⊆ i∈I Ui . A subcover of a given cover U for A is a subfamily V ⊂ U
which still forms a cover for A.
If A is a subspace of a space (X, T ), U is an open cover for A iff U
is a cover for A such that each member of U is open in X.
The classic theorem of Heine-Borel-Lebesgue asserts that, in R,
every open cover of a closed bounded subset has a finite subcover.
This theorem has extraordinarily profound consequences and like most
good theorems, its conclusion has become a definition.
Definition 2.1. (X, T ) is said to be compact iff every open cover
of X has a finite subcover.
Theorem 2.2 (Alexander’s Subbase Theorem). Let S be any sub-
base for (X, T ). If every open cover of X by members of S has a finite
subcover, then X is compact.
11
12 2. TOPOLOGICAL PROPERTIES

The proof of this deep result is an application of Zorn’s lemma, and

is not an exercise for the faint-hearted!
Examples.
(i). (R, I) is not compact, for consider U = {(−n, n) : n ∈ N}.
(ii). (0, 1) is not compact, for consider U = {( n1 , 1) : n ≥ 2}.
(iii). (X, C) is compact, for any X.
(iv). Given x ∈ X, (X, E(x)) is compact; (X, I(x)) is not compact
unless X is finite.
(v). T finite for any X ⇒ (X, T ) compact.
(vi). X finite, T any topology for X ⇒ (X, T ) compact.
(vii). X infinite ⇒ (X, D) not compact.
(viii). Given (X, T ), if (xn ) is a sequence in X convergent to x, then
{xn : n ∈ N} ∪ {x} is compact.
2.1.2. Compactness for Subspaces. We call a subset A of (X, T )
a compact subset when the subspace (A, TA ) is a compact space. It’s a
nuisance to have to look at TA in order to decide on this. It would be
easier to use the original T . Thankfully, we can!
Lemma 2.3. A is a compact subset of (X, T ) iff every T -open cover
of A has a finite subcover.
Proof. Exercise. 
Lemma 2.4. Compactness is closed-hereditary and preserved by con-
tinuous maps.
Proof. Exercise. 
Example. The unit circle in R2 is compact; indeed, paths in any
space are compact.
2.1.3. Compactness in Metric Spaces. In any metric space
(M, d), every compact subset K is closed and bounded:
Bounded, since given any x0 ∈ M ,
j
[
K ⊆ B(x0 , 1) ∪ B(x0 , 2) ∪ B(x0 , 3) ∪ · · · ⇒ K ⊆ B(x0 , ni )
i=1

where we can arrange n1 < n2 < . . . < nj . Thus K ⊆ B(x0 , nj ) and so

any two points of K lie within nj of x0 and hence within 2nj of each
other, i.e., K is bounded.
K is closed, since if x ∈ K and x 6∈ K, then for each y ∈ K,
dy = 21 d(x, y) > 0 so we may form the (open) cover of K as follows:
{B(y, dy ) : y ∈ K} which reduces to a finite subcover {B(yi , dyi ) :
yi ∈ K, i = 1, . . . , n}. The corresponding neighbourhoods of x, namely
 2.1. COMPACTNESS 13

B(x, dyi ), i = 1, . . . , n, may be intersected giving a neighbourhood of x

which misses K —contradiction!
Neither half is valid in all topological spaces;
• ‘compact ⇒ bounded’ doesn’t even make sense since ‘bounded’
depends on the metric.
• ‘compact ⇒ closed’ makes sense but is not always true. For
example, in (R, C), (0, 1) is not closed yet it is compact (since
its topology is the cofinite topology!)
Further, in a metric space, a closed bounded subset needn’t be
compact (e.g., consider M with the discrete metric and let A ⊆ M be
infinite; then A is closed, bounded (since A ⊆ B(x, 2) = M for any
x ∈ M ), yet it is certainly not compact! Alternatively, the subspace
(0, 1) is closed (in itself), bounded, but not compact.)
However, the Heine-Borel theorem asserts that such is the case for
R and Rn ; the following is a special case of the theorem:
Theorem 2.5. Every closed, bounded interval [a, b] in R is compact.
Proof. Let U be any open cover of [a, b] and let K = {x ∈ [a, b] :
[a, x] is covered by a finite subfamily of U}. Note that if x ∈ K and
a ≤ y ≤ x, then y ∈ K. Clearly, K 6= ∅ since a ∈ K. Moreover,
given x ∈ K, there exists δx > 0 such that [x, x + δx ) ⊆ K (since x
∈ some open U ∈ chosen finite subcover of U). Since K is bounded,
k ∗ = sup K exists.
Case k ∗ ∈ K. Choose U ∈ U such that k ∗ ∈ U ; then there exists
 > 0 such that (k ∗ − , k ∗ ] ⊆ U . Since there exists x ∈ K such that
k ∗ −  < x < k ∗ , k ∗ ∈ K.
Case k ∗ = b. If k ∗ < b, choose U ∈ U with k ∗ ∈ U and note that
[k , k ∗ + δ) ⊆ U for some δ > 0—contradiction!
∗

Note. An alternative proof is to invoke the connected nature of [a, b]
by showing K is clopen in [a, b].
Theorem 2.6. Any continuous map from a compact space into a
metric space is bounded.
Proof. Immediate. 
Corollary 2.7. If (X, T ) is compact and f : X → R is continu-
ous, then f is bounded and attains its bounds.
Proof. Clearly, f is bounded. Let m = sup f (X) and l = inf f (X);
we must prove that m ∈ f (X) and l ∈ f (X). Suppose that m 6∈ f (X).
Since f (X) = f (X), then there exists  > 0 such that (m − , m + ) ∩
f (X) = ∅, i.e., for all x ∈ X, f (x) ≤ m − —contradiction!
14 2. TOPOLOGICAL PROPERTIES

Similarly, if l 6∈ f (X), then there exists  > 0 such that [l, l + ) ∩

f (X) = ∅ whence l +  is a lower bound for f (X)! 
2.1.4. Sequential Compactness.
Definition 2.8. A topological space (X, T ) is said to be sequen-
tially compact if and only if every sequence in X has a convergent
subsequence.
Recall from Chapter 1 the definition of convergence of sequences
in topological spaces and the cautionary remarks accompanying it.
There we noted that, contrary to the metric space situation, sequences
in topology can have several different limits! Consider, for example,
(X, T0 ) and (R, L). In the latter space, if xn → l, then xn = l for all
n ≥ some n0 . Thus the sequence 1, 21 , 14 , 18 , . . . does not converge in
(R, L)!
Lemma 2.9. Sequential compactness is closed-hereditary and pre-
served by continuous maps.
Proof. Exercise. 
We shall prove in the next section that in metric spaces, sequential
compactness and compactness are equivalent!
Definition 2.10. Given a topological space (X, T ), a subset A of
X and x ∈ X, x is said to be an accumulation point of A iff every
neighbourhood of x contains infinitely many points of A.
Lemma 2.11. Given a compact space (X, T ) with an infinite subset
A of X, then A has an accumulation point.
Proof. Suppose not; then for each x ∈ X, there exists a neigh-
bourhood Nx of x such that Nx ∩ A is (at most) finite; the family
{Nx : x ∈ X} is an open cover of X and so has a finite subcover
{Nxi : i = 1, . . . , n}. But A ⊆ X and A is infinite, whence
[ [n
A = A ∩ X = A ∩ ( Nxi ) = (A ∩ Nxi )
i=1
is finite! 
Lemma 2.12. Given a sequentially compact metric space (M, d) and
 > 0, there is a finite number of open balls, radius , which cover M .
Proof. Suppose not and that for some  > 0, there exists no finite
family of open balls, radius , covering M . We derive a contradiction
by constructing a sequence (xn ) inductively such that d(xm , xn ) ≥  for
all n, m (n 6= m), whence no subsequence is even Cauchy!
 2.1. COMPACTNESS 15

Let x1 ∈ M and suppose inductively that x1 , . . . , xk have been cho-

sen in M such that d(xi , xj ) ≥  for all i, j ≤ k, i 6= j. By hypothesis,
{B(xi , ) : i = 1, . . . , k} is not an (open) cover of M and so there
exists xk+1 ∈ M such that d(xk+1 , xi ) ≥  for 1 ≤ i ≤ k. We thus
construct the required sequence (xn ), which clearly has no convergent
subsequence. 
Theorem 2.13. A metric space is compact iff it is sequentially
compact.
Proof. ⇒. Suppose (M, d) is compact. Given any sequence (xn )
in M , either A = {x1 , x2 , . . .} is finite or it is infinite. If A is finite,
there must be at least one point l in A which occurs infinitely often in
the sequence and its occurrences form a subsequence converging to l.
If A is infinite, then by the previous lemma there exists x ∈ X such
that every neighbourhood of x contains infinitely many points of A.
For each k ∈ ω, B(x, k1 ) contains infinitely many xn ’s: select one,
call it xnk , making sure that nk > nk−1 > nk−2 . . .. We have a subse-
quence (xn1 , xn2 , . . . , xnk , . . .) so that d(x, xnk ) < k1 → 0, i.e., xnk → x.
Thus in either case there exists a convergent subsequence and so (M, d)
is sequentially compact.
⇐. Conversely, suppose (M, d) is sequentially compact and not
compact. Then there exists some open cover {Gi : i ∈ I} of M having
no finite subcover. By Lemma 2.12, with  = n1 (n ∈ ω), we can cover
M by a finite number of balls of radius n1 . For each n, there has to
be one of these, say B(xn , n1 ), which cannot be covered by any finite
number of the sets Gi . The sequence (xn ) must have a convergent
subsequence (xnk ) which converges to a limit l. Yet {Gi : i ∈ I} covers
M , so l ∈ some Gi0 , say.
As k → ∞, xnk → l; but also n1k → 0 and 1/nk is the radius of the
ball centred on xnk . So eventually B(xnk , n1k ) is inside Gi0 , contradic-
tory to their choice! (More rigorously, there exists m ∈ ω such that
B(l, m2 ) ⊆ Gi0 . Now B(l, m1 ) contains xnk for all k ≥ k0 say, so choose
k ≥ k0 such that nk ≥ m. Then B(xnk , n1k ) ⊆ B(l, m2 ) ⊆ Gi0 .) Hence,
M is compact. 

2.1.5. Compactness and Uniform Continuity. Recall that a

map f : (X1 , d1 ) → (X2 , d2 ), where (Xi , di ) is a metric space for each i,
is uniformly continuous on Xi if given any  > 0, there is a δ > 0 such
that for all x, y ∈ X1 , d1 (x, y) < δ ⇒ d2 (f (x), f (y)) < .
Ordinary continuity of f is a local property, while uniform continu-
ity is a global property since it says something about the behaviour of
16 2. TOPOLOGICAL PROPERTIES

f over the whole space X1 . Since compactness allows us to pass from

the local to the global, the next result is not surprising:
Theorem 2.14. If (X, d) is a compact metric space and f : X → R
is continuous, then f is uniformly continuous on X.
Note. Result holds for any metric space codomain.
Proof. Let  > 0; since f is continuous, for each x ∈ X, there
exists δx > 0 such that d(x, y) < 2δx ⇒ |f (x) − f (y)| < 2 . The family
{Bδx (x) : x ∈ X} is an open cover of X and so has a finite subcover
{Bδxi (xi ) : i = 1, . . . , n} of X. Let δ = min{δxi : i = 1, . . . , n}. Then,
given x, y ∈ X such that d(x, y) < δ, it follows that |f (x) − f (y)| < .
For x ∈ Bδxi (xi ) for some i, whence d(x, xi ) < δxi and so
d(y, xi ) ≤ d(y, x) + d(x, xi ) < δ + δxi ≤ 2δxi ⇒ |f (y) − f (xi )| < 2 .
 
Thus |f (x) − f (y)| ≤ |f (x) − f (xi )| + |f (xi ) − f (y)| < 2
+ 2
= . 
Note. Compactness is not a necessary condition on the domain
for uniform continuity. For example, for any metric space (X, d), let
f : X → X be the identity map. Then f is easily seen to be uniformly
continuous on X.
2.1.6. Local Compactness.
Definition 2.15. A topological space (X, T ) is locally compact
iff each point of X has a compact neighbourhood.
Clearly, every compact space is locally compact. However, the con-
verse is not true.
Examples.
(i). With X infinite, the discrete space (X, D) is clearly locally com-
pact (for each x ∈ X, {x} is a compact neighbourhood of x!) but not
compact.
(ii). With X infinite and x ∈ X, (X, I(x)) is locally compact (but
not compact).
(iii). (R, I) is locally compact (x ∈ R ⇒ [x − 1, x + 1] is a compact
neighbourhood of x).
(iv). The set of rational numbers Q with its usual topology is not
a locally compact space, for suppose otherwise; then 0 has a compact
neighbourhood C in Q so we can choose  > 0 such that J = Q ∩
[−, ] ⊆ C. Now J is closed in (compact) C and is therefore compact
R
in R. Thus, J must be closed in R—but J = [−, ]!
Lemma 2.16.
(i) Local compactness is closed-hereditary.
 2.3. CONNECTEDNESS 17

(ii) Local compactness is preserved by continuous open maps.

Proof. Exercise. 
Note. Local compactness is not preserved by continuous maps in
general. Consider idQ : (Q, D) → (Q, IQ ) which is continuous and onto;
(Q, D) is locally compact while (Q, IQ ) isn’t!

2.2. Other Covering Conditions

Definition 2.17. A topological space (X, T ) is said to be
(i) Lindelöf iff every open cover of X has a countable subcover;
(ii) countably compact iff every countable open cover of X has
a finite subcover.
Thus, a space is compact precisely when it is both Lindelöf and
countably compact. Further, every sequentially compact space is count-
ably compact, although the converse is not true. Moreover, sequential
compactness neither implies nor is implied by compactness.
However, for metric spaces, or more generally, metrizable spaces,
the conditions compact, countably compact and sequentially compact
are equivalent.
Note. Second countable ⇒ separable; separable + metrizable ⇒
second countable . . . and so in metrizable spaces, second countability
and separability are equivalent.

2.3. Connectedness
It is not terribly hard to know when a set on the real line is con-
nected, or ‘of just one piece.’ This notion is extended to general topo-
logical spaces in this section and alternative characterizations of the
notion are given. In addition the relationship between continuous maps
and and connectedness is given. This provides an elegant restatement
of the familiar Intermediate Value Theorem from first term calculus.
2.3.1. Definition of Connectedness. A partition of (X, T ) means
a pair of disjoint, non-empty, T -open subsets whose union is X. Notice
that, since these sets are complements of one another, they are both
closed as well as both open. Indeed, the definition of ’partition’ is not
affected by replacing the term ’open’ by ’closed’.
Definition 2.18. A connected space (X, T ) is one which has no
partition. (Otherwise, (X, T ) is said to be disconnected.)
If ∅ =
6 A ⊆ (X, T ), we call A a connected set in X whenever (A, TA )
is a connected space.
18 2. TOPOLOGICAL PROPERTIES

Lemma 2.19. (X, T ) is connected iff X and ∅ are the only subsets
which are clopen.
Examples.
(i). (X, T0 ) is connected.
(ii). (X, D) cannot be connected unless |X| = 1. (Indeed the only
connected subsets are the singletons!)
(iii). The Sorgenfrey line Rs is disconnected (for [x, ∞) is clopen!).
(iv). The subspace Q of (R, I) is not connected because
√ √ √ √
Q ∩ [− 2, 2] = Q ∩ (− 2, 2)
| {z } | {z }
closed in Q open in Q

is clopen and is neither universal nor empty.

(v). (X, C) is connected except when X is finite; indeed, every infi-
nite subset of X is connected.
(vi). (X, L) is connected except when X is countable; indeed, every
uncountable subset of X is connected.
(vii). In (R, I), A ⊆ R is connected iff A is an interval. (Thus,
subspaces of connected spaces are not usually connected—examples
abound in (R, I).)
2.3.2. Characterizations of Connectedness.
Lemma 2.20. ∅ ⊂ A ⊆ (X, T ) is not connected iff there exist T -
open sets G, H such that A ⊆ G ∪ H, A ∩ G 6= ∅, A ∩ H 6= ∅ and
A ∩ G ∩ H = ∅. (Again, we can replace ’open’ by ’closed’ here.)
Proof. Exercise. 
Note. By an interval in R, we mean any subset I such that whenever
a < b < c and whenever a ∈ I and c ∈ I then b ∈ I. It is routine to
check that the only ones are (a, b), [a, b], [a, b), (a, b], [a, ∞), (a, ∞),
(−∞, b), (−∞, b], (−∞, ∞) = R and {a} for real a, b, a < b where
appropriate.
It turns out that these are exactly the connected subsets of (R, I):
Lemma 2.21. In R, if [a, b] = F1 ∪ F2 where F1 , F2 are both closed
and a ∈ F1 , b ∈ F2 then F1 ∩ F2 6= ∅.
Proof. Exercise. 
Theorem 2.22. Let ∅ ⊂ I ⊆ (R, I). Then I is connected iff I is
an interval.
Proof. ⇒. If I is not an interval, then there exist a < b < c with
a ∈ I, b 6∈ I and c ∈ I. Take A = I ∩ (−∞, b) and B = I ∩ (b, ∞).
 2.3. CONNECTEDNESS 19

Then A ∪ B = I, A ∩ B = ∅, A 6= ∅, B 6= ∅, A ⊂ I, B ⊂ I and A, B
are both open in I, i.e., A and B partition I and so I is not connected.
⇐. Suppose I is not connected and that I is an interval. By the
‘closed’ version of Lemma 2.20, there exist closed subsets K1 , K2 of R
such that I ⊆ K1 ∪ K2 , I ∩ K1 6= ∅, I ∩ K2 6= ∅ and I ∩ K1 ∩ K2 = ∅.
Select a ∈ I ∩ K1 , b ∈ I ∩ K2 ; without loss of generality, a < b.
Then [a, b] ⊆ I so that [a, b] = ([a, b] ∩ K1 ) ∪ ([a, b] ∩ K2 ), whence by
Lemma 2.21, ∅ = 6 [a, b] ∩ K1 ∩ K2 ⊆ I ∩ K1 ∩ K2 = ∅! 

2.3.3. Connectedness and Continuous Maps.

Lemma 2.23. Connectedness is preserved by continuous maps.
Proof. Exercise. 
Corollary 2.24 (Intermediate Value Theorem). If f : [a, b] → R
is continuous and f (a) < y < f (b), then y must be a value of f .
Proof. Exercise. 
Corollary 2.25 (Fixed point theorem for [0, 1]). If f : [0, 1] →
[0, 1] is continuous, then it has a ‘fixed point’, i.e., there exists some
x ∈ [0, 1] such that f (x) = x.
Proof. Consider g(x) = f (x) − x. Then g : [0, 1] → R is contin-
uous. Further, g(0) = f (0) ≥ 0 and g(1) = f (1) − 1 ≤ 0 so that 0 is
intermediate between g(0) and g(1). Thus, by the Intermediate Value
Theorem, there exists x ∈ [0, 1] such that 0 = g(x) = f (x) − x, i.e.,
such that f (x) = x. 
Note. Given continuous h : [a, b] → [a, b], it follows that h has a
fixed point since [a, b] ∼
= [0, 1] and ‘every continuous function has a
fixed point’ is a homeomorphic invariant.
Lemma 2.26. Let (X, T ) be disconnected with ∅ ⊂ Y ⊂ X, Y
clopen. If A is any connected subset of X, then A ⊆ Y or A ⊆ X \ Y .
Proof. If A ∩ Y 6= ∅ =6 A ∩ (X \ Y ), then ∅ ⊂ A ∩ Y ⊂ A and
A ∩ Y is clopen in A. Thus, A is not connected! It follows that either
A ∩ Y = ∅ or A ∩ X \ Y = ∅, i.e., either A ⊆ X \ Y or A ⊆ Y . 
Lemma 2.27. If the family {Ai : i ∈ I} of connected subsets
S of a
space (X, T ) has a non-empty intersection, then its union i∈I Ai is
connected.
Proof. SupposeSnot and that there exists a non-empty proper
clopen subset Y of i∈I Ai . Then for each i ∈ I, either Ai ⊆ Y or
20 2. TOPOLOGICAL PROPERTIES
S
Ai ⊆ i∈I Ai \ Y T. However if for some j, Aj ⊆ S Y , then Ai ⊆ Y for
each i ∈ I (since i∈I Ai 6= ∅) which implies
S that i∈I Ai ⊆ S
Y!
S Similarly, if for some k ∈ I, Ak ⊆ i∈I Ai \ Y , then i∈I Ai ⊆
i∈I Ai \ Y ! 
Corollary 2.28. Given a family {Ci : i ∈ I} of connected subsets
of a space (X, T ),
S if B ⊆ X is also connected and B ∩ Ci 6= ∅ for all
i ∈ I, then B ∪ ( i∈I Ci ) is connected.
Proof. Take Ai = B ∪ Ci in Lemma 2.27. 
Lemma 2.29. If A is a connected subset of a space (X, T ) and
T
A ⊆ B ⊆ A , then B is a connected subset.
Proof. If B is not connected, then there exists ∅ ⊂ Y ⊂ B which
is clopen in B. By Lemma 2.26, either A ⊆ Y or A ⊆ B \ Y . Suppose
T T
A ⊆ Y (a similar argument suffices for A ⊆ B \ Y ); then A ⊆ Y
and so
TB TB
B \Y = B ∩(B \Y ) = A ∩(B \Y ) ⊆ Y ∩(B \Y ) = Y ∩(B \Y ) = ∅
—a contradiction! 
Definition 2.30. Let (X, T ) be a topological space with x ∈ X;
we define the component of x, Cx , in (X, T ) to be the union of all
connected subsets of X which contain x, i.e.,
[
Cx = {A ⊆ X : x ∈ A and A is connected}.
For each x ∈ X, it follows from Lemma 2.27 that Cx is the maximum
connected subset of X which contains x. Also it is clear that if x,
y ∈ X, either Cx = Cy or Cx ∩ Cy = ∅ (for if z ∈ Cx ∩ Cy , then
Cx ∪ Cy ⊆ Cz ⊆ Cx ∩ Cy whence Cx = Cy (= Cz )). Thus we may
speak of the components of a space (X, T ) (without reference to specific
points of X): they partition the space into connected closed subsets
(by Lemma 2.29) and are precisely the maximal connected subsets of
X.
Examples.
(i). If (X, T ) is connected, (X, T ) has only one component, namely
X!
(ii). For any discrete space, the components are the singletons.
(iii). In Q (with its usual topology), the components are the single-
tons. (Thus, components need not be open.)
Definition 2.31. A space (X, T ) is totally disconnected iff the
only connected subsets of X are the singletons (equivalently, the com-
ponents of (X, T ) are the singletons).
 2.4. SEPARABILITY 21

Thus, by the previous examples, we see that the space Q of ratio-

nals, the space R \ Q of irrationals and any discrete space are all totally
disconnected. Further, the Sorgenfrey line Rs is totally disconnected.

2.3.4. Pathwise Connectedness.

Definition 2.32. A topological space (X, T ) is pathwise con-
nected iff for any x, y ∈ X, there exists a continuous function f : [0, 1] →
X such that f (0) = x and f (1) = y. Such a function f is called a path
from x to y.
Theorem 2.33. Every pathwise connected space is connected.
Proof. Let (X, T ) be pathwise connected and let a ∈ X; for every
x ∈ X, there exists a path px : [0, 1] → X from a to x.
T Then, for each
x ∈ X, px ([0, 1]) is connected;
S moreover, pa (0) = a ∈ x∈X px ([0, 1]) so
that by Lemma 2.27, X = x∈X px ([0, 1]) is connected. 
Note well. The converse is false. Consider the following example,
the topologist’s sine curve:
V = {(x, 0) : x ≤ 0} ∪ {(x, sin x1 ) : x > 0}
is a connected space, but no path can be found from (0, 0) to any point
(x, sin x1 ) with x > 0
For suppose, w.l.o.g., there exists a path p : [0, 1] → X with p(0) =
1
( π , 0) and p(1) = (0, 0). Then π1 ◦ p, being continuous, must take all
values between 0 and π1 , in particular (2n+1 1 )π for each n, i.e., there
2
1
exists tn ∈ [0, 1] such that π1 ◦ p(tn ) = (2n+ 12 )π
for each n. Thus,
p(tn ) =( (2n+1 1 )π , 1)
→ (0, 1) as n → ∞. Now tn ∈ [0, 1] for all n which
2
implies that there exists a subsequence (tnk ) in [0, 1] with tnk → λ.
Then p(tnk ) → p(λ) and so π1 ◦ p(tnk ) → 0. Thus p(λ) = (0, y) for
some y, whence y = 0 (since p(λ) ∈ X)!

2.4. Separability
Definition 2.34. A topological space is said to be
(i) separable iff it has a countable dense subset;
(ii) completely separable (equivalently, second countable) iff
it has a countable base.
Examples.
(i). (R, I) is separable (since Q = R).
(ii). (X, C) is separable for any X.
22 2. TOPOLOGICAL PROPERTIES

(iii). (R, L) is not separable.

Theorem 2.35.
(i) Complete separability implies separability.
(ii) The converse is true in metric spaces.
Proof. We prove only (ii). In metric space (M, d), let D =
{x1 , x2 , . . .} be dense. Consider B = {B(xi , q) : i ∈ ω, q ∈ Q, q > 0}, a
countable collection of open sets. One can show that B is a base for Td
. . . over to you! 
Theorem 2.36.
(i) Complete separability is hereditary.
(ii) Separability is not hereditary.
Examples.
(i). Consider the ‘included point’ topology I(0) on R; then (R, I(0))
is separable, since {0} = R. However, R \ {0} is not separable because
it is discrete.
(ii). Separability does not imply complete separability since, for
example, (R, I(0)) is separable but not completely separable. (Suppose
there exists a countable base B for its topology. Given x 6= 0, {0, x} is
an open neighbourhood of x and so there exists Bx ∈ B such that x ∈
Bx ⊆ {0, x}. Thus Bx = {0, x}, i.e., B is uncountable—contradiction!
Theorem 2.37. Separability is preserved by continuous maps.
Proof. Exercise. 
Note. Complete separability is not preserved by continuous maps.
 CHAPTER 3

Convergence

In Chapter 1, we defined limits of sequences in a topological space

(X, T ) so as to assimilate the metric definition. We noted, however,
that not everything we knew about this idea in metric spaces is valid
in topological spaces.
We will examine two main ways around this difficulty:
• develop a kind of ‘super-sequence’ or net which does for general
topology what ordinary sequences do for metric spaces.
• identify the class of topological spaces in which the old idea of
sequential limit is good enough.

3.1. The Failure of Sequences

The following important results are probably familiar to us in the
context of metric spaces, or at least in the setting of the real line, R.
Theorem 3.1. Given (X, T ), A ⊆ X, p ∈ X: if there exists some
sequence of points of A tending to p, then p ∈ A.
Theorem 3.2. Given (X, T ), A ⊆ X: if A is closed, then A in-
cludes the limit of every convergent sequence of points of A.
0
Theorem 3.3. Given f : (X, T ) → (Y, T ): if f is continuous,
then f ‘preserves limits of sequences’, i.e., whenever xn → l in X,
then f (xn ) → f (l) in Y .
In each case above, it is routine to prove the statement true in a
general topological space as asserted. We illustrate by proving Theo-
rem 3.3:
Let f be continuous and xn → l in X. We must show that f (xn ) →
f (l). Given a neighbourhood N of f (l), there exists open G such
that f (l) ∈ G ⊆ N . Then l ∈ f −1 (G) ⊆ f −1 (N ), i.e., f −1 (N ) is a
neighbourhood of l and so xn ∈ f −1 (N ), for all n ≥ n0 say. Thus
f (xn ) ∈ N , for all n ≥ n0 , whence f (xn ) → f (l).
In metric spaces, the converses of these results are also true but our
main point here is that in general topology, the converses are not valid.
23
24 3. CONVERGENCE

Example. In (R, L), (0, 1) = R. So, for example, 5 ∈ (0, 1) and yet
the only way a sequence (xn ) converges to a limit l is for xn = l from
some stage on. So no sequence in (0, 1) can converge to 5 proving that
the converse of Theorem 3.1 is false.
Continuing, the limit of any convergent sequence in (0, 1) must
belong to (0, 1) for the same reason and yet (0, 1) is not closed. Thus,
Theorem 3.2’s converse is false.
Further, idR : (R, L) → (R, I) is not continuous and yet it does
preserve limits of sequences.
Now this is a great nuisance! Sequences are of immense usefulness
in real analysis and in metric spaces and elsewhere—and their failure
to describe general topology adequately is a technical handicap. What
to do?
3.2. Nets—A Kind of ‘Super-Sequence’
Recall that a sequence is just a function having the positive integers
as domain. The set of positive integers, of course, possesses a partic-
ularly simple ordering; there is a first member, second member, third
member, etc. Not all sets are naturally endowed with so simple an
ordering. For example, dictionary (lexographical) ordering of words is
more complex (though still relative nice as orderings go). By replacing
the domain of positive integers with a set having a more complicated
ordering we will:
• define a net (in analogy with sequence),
• identify an associated notion of convergence,
• show that net convergence is sufficient to characterize closure
of sets,
• and that compactness can be characterized in terms of conver-
gence of subnets.
Note that these last two items generalize the role of sequences in a
metric space.
3.2.1. Definition of Net.
Definition 3.4. A binary relation ≤ on a set P is said to be a
pre-order iff
(i) p ≤ p, for all p ∈ P ;
(ii) p ≤ q and q ≤ r imply p ≤ r, for all p, q, r ∈ P .
We often refer to P as being a pre-ordered set when it is understood
that ≤ is the pre-order in question.
P is said to be a partially ordered set (or poset) if it is also true
that
 3.2. NETS—A KIND OF ‘SUPER-SEQUENCE’ 25

(iii) p ≤ q and q ≤ p imply p = q, for all p, q ∈ P .

Definition 3.5. A pre-ordered set P is said to be directed (or
updirected) iff each pair of members of P has an upperbound. (i.e., if
p, q ∈ P , then there exists s ∈ P such that p ≤ s, q ≤ s.)
Definition 3.6. Let (P, ≤) be a poset. Then if x, y ∈ P with x 6≤ y
and y 6≤ x, we write x k y and say that x and y are incomparable;
If E ⊆ P , then E is said to be totally unordered (or diverse) iff
x, y ∈ E implies x = y or x k y.
If C ⊆ P , then C is said to be linear (or a chain or a total order)
iff x, y ∈ C implies x < y, x = y or y < x.
(P, ≤) is said to be a lattice iff each pair of members of P has a
greatest lower bound and a least upper bound.
A lattice (P, ≤) is said to be complete iff every non-empty subset
Y of P has a greatest lower bound (∧Y ) and a least upper bound (∨Y ).
An element v of a poset (P, ≤) is said to be maximal (minimal) iff
v ≤ x (x ≤ v), x ∈ P ⇒ v = x.
Definition 3.7. A net in a (non-empty set) X is any function
x : A → X whose domain A is a directed set.
In imitation of the familiar notation in sequences, we usually write
the net value x(α) as xα . A typical net x : A → X will usually appear
as (xα , α ∈ A) or (xα )α∈A or some such notation.
Examples.
(i). N, Z, N × N are all directed sets, where suitable pre-orders are
respectively the usual magnitude ordering for N and Z, and (i, j) ≤
(m, n) iff i ≤ m and j ≤ n, in N × N. Thus, for example, a sequence is
an example of a net.
(ii). The real function f : R \ {0} → R given by f (x) = 3 − x1 is a
net, since its domain is a chain. Any real function is a net.
(iii). Given x ∈ (X, T ), select in any fashion an element xN from
each neighbourhood N of x; then (xN )N ∈Nx is a net in X (since it
defines a mapping from (Nx , ≤) into X). Recall that Nx is ordered by
inverse set inclusion!
3.2.2. Net Convergence.
Definition 3.8. A net (xα )α∈A in (X, T ) converges to a limit l
if for each neighbourhood N of l, there exists some αN ∈ A such that
xα ∈ N for all α ≥ αN .
In such a case, we sometimes say that the net (xα )α∈A eventuates
N . Clearly, this definition incorporates the old definition of ‘limit of
26 3. CONVERGENCE

a sequence’. The limit of the net f described in (ii) above is 3. In

(iii), the net described converges to x no matter how the values xN are
chosen—prove!

3.2.3. Net Convergence and Closure. Our claim is that nets

‘fully describe’ the structure of a topological space. Our first piece
of evidence to support this is that with nets, instead of sequences,
Theorems 3.1, 3.2 and 3.3 have workable converses:
Theorem 3.9. Given (X, T ), A ⊆ X, p ∈ X: p ∈ A iff there exists
a net in A converging to p.
Proof. If some net of points of A converges to p, then every neigh-
bourhood of p contains points of A (namely, values of the net) and so
we get p ∈ A. Conversely, if p is a closure point of A then, for each
neighbourhood N of p, it will be possible to choose an element aN
of A that belongs also to N . The net which these choices constitute
converges to p, as required. 
Theorem 3.10. Given (X, T ), A ⊆ X, A is closed iff it contains
every limit of every (convergent) net of its own points.
Proof. This is really just a corollary of the preceding theorem. 
0
Theorem 3.11. Given f : (X, T ) → (Y, T ), f is continuous iff f
preserves net convergence.
Proof. Exercise. 

3.2.4. Nets and Compactness.

Definition 3.12. Let (xα )α∈A be any net and let α0 ∈ A. The α0th
tail of the net is the set {xα : α ≥ α0 } = x([α0 , )). We denote it by
x(α0 →).
Definition 3.13. Let (xα )α∈A and (yβ )β∈B be any two nets. We
call (yβ )β∈B a subnet of (xα )α∈A provided that every tail of (xα ) con-
tains a tail of (yβ ), i.e., provided:
∀α0 ∈ A ∃β0 ∈ B such that x(α0 →) ⊇ y(β0 →).
We expected a definition like ‘subsequence’ to turn up here and we
are disappointed that it has to be so complicated.
Net theory ceases to be a straightforward generalisation of sequence
theory precisely when we have to take a subnet . . . so we’ll try to avoid
this whenever possible! There is however one result certainly worth
knowing:
 3.3. FIRST COUNTABLE SPACES—WHERE SEQUENCES SUFFICE 27

Theorem 3.14. (X, T ) is compact iff in X, every net has (at least
one) convergent subnet.
(So, for example, (n) is a net in R with no convergent subnet.)
Proof. Not required. 
Corollary 3.15. Compactness is closed-hereditary.
Proof. For if (xα ) is a net in a closed set F ⊆ X, then it has a
convergent subnet (yβ ) in X. Thus there exists a subnet (zγ ) of (yβ )
in F which converges in X, whence its limit is in F . 
Corollary 3.16. Compactness is preserved by continuous maps.
Proof. For if X is compact and f continuous, let (yα )α∈A be a
net in f (X). Then for each α ∈ A, yα = f (xα ) for some xα ∈ X.
The net (xα )α∈A has a convergent subnet (zβ )β∈B , say zβ → l, whence
f (zβ ) → f (l). Then (f (zβ ))β∈B is a convergent subnet of (yα )α∈A . 
Example. If (xnk ) is a subsequence of a sequence (xn ), then it is a
subnet of it also; because the ith
0 tail of the sequence (xn ) is
(∗) {xi0 , xi0 +1 , xi0 +2 , . . .}
while the ith
0 tail of the subsequence (xnk ) is:
(∗∗) {xni0 , xni0 +1 , xin0 +2 , . . .}
and we see that (∗∗) ⊆ (∗) merely because ni0 ≥ i0 .
Lemma 3.17. If a net (xα ) converges to a limit l, then so do all its
subnets.
Proof. Let (yβ ) be a subnet of (xα ); let N be a neighbourhood of
l. Then there exists α0 such that xα ∈ N for all α ≥ α0 . Further, there
exists β0 such that {yβ : β ≥ β0 } ⊆ {xα : α ≥ α0 } and so yβ ∈ N for
all β ≥ β0 . 
3.3. First Countable Spaces—Where Sequences Suffice
Why do sequences suffice to describe structure in R and other metric
spaces but not in many other topological spaces? The key here is recog-
nizing that many proofs regarding convergence in metric space involve
constructing sequences of nested open sets about a point. Sometimes
these describe the topological structure near the point and other times
not. In what follows we
• identify the local characteristic of topological space that makes
these proofs work,
• and prove that sequences suffice to describe the topological
structure of spaces with this characteristic.
28 3. CONVERGENCE

3.3.1. First Countable Spaces. So what characteristic common

to R and other metric spaces makes sequences so ‘good’ at describing
their structure?
Definition 3.18. Let x ∈ (X, T ). A countable neighbourhood
base at x means: a sequence N1 , N2 , N3 , . . . of particular neighbour-
hoods of x such that every neighbourhood of x shall contain one of the
Ni ’s.
Note that we may assume that N1 ⊇ N2 ⊇ N3 ⊇ · · · because, if
not, then we can work with N1 , N1 ∩ N2 , N1 ∩ N2 ∩ N3 , . . .
Definition 3.19. We call (X, T ) first-countable when every point
in X has a countable neighbourhood base.
Example. The classic example of a first-countable space is any met-
ric (or metrizable) space because if x ∈ (M, d), then B(x, 1), B(x, 21 ),
B(x, 13 ), . . . is a countable neighbourhood base at x.
Theorem 3.20. First-countability is hereditary and preserved by
continuous open onto maps.
Proof. Left to the reader. 
Theorem 3.21.
(i) Complete separability implies first countability.
(ii) Converse not always true.
(iii) Converse valid on a countable underlying set.
Proof. (i). If B is a countable base for (X, T ) and p ∈ X, consider
{B ∈ B : p ∈ B} which is a countable family of neighbourhoods of p.
Moreover, they form a neighbourhood base at p.
(ii). An uncountable discrete space is first countable (since metriz-
able), yet is not completely separable.
(iii). Suppose X countable and (X, T ) first countable. For each
x ∈ X, choose a countable neighbourhood base: N (x, 1), N (x, 2),
N (x, 3), . . . . Each is a neighbourhood of x and so contains an open
neighbourhood of x: G(x, 1), G(x, 2), G(x, 3), . . . . Then B = {G(x, n) :
n ∈ N, x ∈ X} is a countable family of open sets and is a base for
(X, T ). Thus, (X, T ) is completely separable. 
Example. The Arens-Fort space X = N2 ∪ {∞} is a countable set
with a topology that is not first countable, hence not completely sep-
arable either: each point (m, n) is isolated, and U is neighbourhood
of ∞ iff there is an N ∈ N and a function f : N → N such that
{(m, n) : m > N, n > f (m)} ⊂ U .
 3.3. FIRST COUNTABLE SPACES—WHERE SEQUENCES SUFFICE 29

3.3.2. Power of Sequences in First Countable Spaces. The

following three results illustrate that ‘sequences suffice for first-countable
spaces’ in the sense that we don’t need to use nets to describe their
structure. This is why sequences are sufficiently general to describe,
fully, metric and metrizable spaces.
Theorem 3.22. Given a first-countable space (X, T ):
(i) p ∈ X, A ⊆ X, then p ∈ A iff there exists a sequence of points
of A converging to p.
(ii) A ⊆ X is closed iff A contains every limit of every convergent
sequence of its own points.
(iii) f : (X, T ) → (Y, T 0 ) is continuous iff it preserves limits of
(convergent) sequences.
Proof. (i). Theorem 3.1 said that if there exists a sequence in A
converging to some p ∈ X, then p ∈ A.
Conversely, if p ∈ A, then p has a countable base of neighbourhoods
N1 ⊇ N2 ⊇ N3 ⊇ · · · , each of which must intersect A. So choose
xj ∈ Nj ∩ A for all j ≥ 1. Then (xj ) is a sequence in A and, given
any neighbourhood H of p, H must contain one of the Nj ’s, i.e., H ⊇
Nj0 ⊇ Nj0 +1 ⊇ · · · so that xj ∈ H for all j ≥ j0 . That is, xj → p.
(ii). Corollary of (i).
(iii). f continuous implies that it must preserve limits of sequences
(by Theorem 3.3). Conversely, if f is not continuous, there exists A ⊆
X such that f (A) 6⊆ f (A). Thus, there exists p ∈ f (A) \ f (A) so
p = f (x), some x ∈ A. So there exists a sequence (xn ) in A with
xn → x.
Yet, if f (xn ) → f (x) (= p), p would be the limit of a sequence
in f (A) so that p ∈ f (A) —contradiction! Thus f fails to preserve
convergence of this sequence. 
 CHAPTER 4

Product Spaces

A common task in topology is to construct new topological spaces

from other spaces. One way of doing this is by taking products. All
are familiar with identifying the plane or 3-dimensional Euclidean space
with ordered pairs or triples of numbers each of which is a member of
the real line. Fewer are probably familar with realizing the torus as
ordered pairs of complex numbers of modulus one. In this chaper we
answer two questions:
• How do the above product constructions generalize to topo-
logical spaces?
• What topological properties are preserved by this construc-
tion?

4.1. Constructing Products

The process of constructing a product falls naturally into two stages.
• The first stage, which is entirely set-theoretic, consists in de-
scribing an element of the underlying set of the product. This
task is primarily one of generalizing the notion of ordered pair
or triple.
• The second stage is describing what open sets look like. This
will be done by describing a subbasis for the topology. The
guiding goal is to provide just enough opens sets to guarantee
the continuity of certain important functions.

4.1.1. Set-Theoretic Construction. Suppose throughout that

we are given a family of topological spaces {(Xi , Ti ) : i ∈ I} where I is
some non-empty ‘labelling’ or index set.
Our first task is to get a clear mental picture of what we mean
by the product of the sets Xi . Look again at the finite case where
I = {1, 2, . . . , n}. Here, the product set
n
Y
X = X1 ×X2 ×X3 ×. . .×Xn = Xi = {(p1 , p2 , . . . , pn ) : pi ∈ Xi , i ∈ I}.
i=1
31
32 4. PRODUCT SPACES

i.e., the elements of X are the functions x : I → ni=1 Xi such that

S
x(1) ∈ X1 , x(2) ∈ X2 , . . . , x(n) ∈ Xn , i.e., x(i) ∈ Xi for all i where,
for convenience, we usually write xi instead of x(i). In this form, the
definition extends immediately to any I, finite or infinite, i.e., if {Xi :
i ∈ I} is any family of sets, then their product is
[
{x : I → Xi for which x(i) ∈ Xi for all i ∈ I}
i∈I

except that we normally write xi ratherQ than x(i).

Then a typical element of X = Xi will look like: (xi )i∈I or just
(xi ). We will still call xi the ith coordinate
Q of (xi )i∈I . (Note that the
Axiom of Choice assures us that Xi is non-empty provided none of
the Xi ’s are empty.)

4.1.2. Topologizing the Product. Q Of the many possible topolo-

gies that could be imposed on X = Xi , we describe the most useful.
This topology is ’just right’ in the sense that it is barely fine enough to
guarantee the continuity of the coordinate projection functions while
being just course enough allow the important result of Theorem 4.4.
Definition 4.1. For each i ∈ I, the ith projection is the map
Xi → Xi which ‘selects the ith coordinate’, i.e., πi ((xi )i∈I ) = xi .
Q
πi :
An open cylinder means the inverse projection of some non-empty
Ti -open set, i.e., πi−1 (Gi ) where i ∈ I, Gi 6= ∅, Gi ∈ Ti .
Tn An −1open box is the intersection of finitely many open cylinders
j=1 πij (Gij ).
We use these open cylinders and boxes to generate a topology with
just enough open sets to guarantee that projection maps will be con-
tinuous. Note that Q the open cylinders form a subbase for a certain
topology T on X = Xi and therefore the open boxes form a base for
T ; T is called the [Tychonoff] product topology and (X, Q T ) is the
product of Qthe given family of spaces. Q We write (X, T ) = {(Xi , Ti ) :
i ∈ I} = i∈I (Xi , Ti ) or even T = i∈I Ti .
Notice that if nj=1 πi−1
T
j
(Gij ) is any open box, then without loss of
generality we can assume i1 , i2 , . . . in all different because if there were
repetitions like
. . . ∩ πi−1
k
(G) ∩ πi−1
k
(H) . . .
we can replace each by
. . . ∩ πi−1
k
(G ∩ H) ∩ . . .
and thus eliminate all repetitions.
 4.1. CONSTRUCTING PRODUCTS 33

It is routine to check that if Tn is the usual topology on Rn , and T

the usual topology on R, then
(R, T ) × (R, T ) × . . . × (R, T ) = (Rn , Tn )
as one would hope!
Lemma 4.2. In a product space (X, T ), N is a neighbourhood of
p ∈ X iff there exists some open box B such that p ∈ B ⊆ N .
Lemma 4.3. For each i ∈ I,
(i) πi is continuous, and
(ii) πi is an open mapping.
Proof. (i). Immediate.
(ii). Given open G ⊆ X, then G is a union of basic open sets {Bk :
k ∈ K} in X, whence πi (G) is a union of open subsets {Bki : k ∈ K}
of Xi and is therefore open. (The notation here is intended to convey
that Bki is the ’component along the i-th coordinate axis’ of the open
box Bk .) 
Theorem 4.4. A map into a product space is continuous iff its
composite with each projection is continuous.
Proof. Since the projections are continuous, so must be their com-
posites with any continuous map. To establish the converse, first show
that if S is a subbase for the codomain (target) of a mapping f , then
f will be continuous provided that the preimage of every member of S
is open; now use the fact that the open cylinders constitute a subbase
for the product topology. 
Worked example. Show that (X, T ) × (Y, S) is homeomorphic to
(Y, S) × (X, T ).
Proof. Define f : X × Y → Y × X, g : Y × X → X × Y by
f (x, y) = (y, x), g(y, x) = (x, y). Clearly these are one-one, onto and
mutually inverse. It will suffice to show that both are continuous.
0 0 0
π1 ◦ f = π2 ; π2 ◦ f = π1 . Now πi is continuous for i = 1, 2 and so f
is continuous! Similarly, g is continuous. 
Worked example. Show that the product of infinitely many copies
of (N, D) is not locally compact.
Proof. We claim that no point has a compact neighbourhood.
Suppose otherwise; then there exists p ∈ X, C ⊆ X and G ⊆ X
with C compact, G open and p ∈ G ⊆ C.T Pick an open box B
such that p ∈ B ⊆ G ⊆ C. B looks like nj=1 πi−1
j
(Gij ). Choose
34 4. PRODUCT SPACES

in+1 ∈ I \ {i1 , i2 , . . . , in }; then πin+1 (C) is compact (since compactness

is preserved by continuous maps).
Thus, pin+1 ∈ πin+1 (B) = Xin+1 ⊆ πin+1 (C) ⊆ Xin+1 = (N, D).
Thus, πin+1 (C) = (N, D)—which is not compact! 

4.2. Products and Topological Properties

The topological properties possessed by a product depends, of course,
on the properties possessed by the individual factors. There are sev-
eral theorems which assert that certain topological properties are pro-
ductive, i.e., are possessed by the product if enjoyed by each factor.
Several of these theorems are given below.
4.2.1. Products and Connectedness.
Theorem 4.5. Any product of connected spaces must be connected.
Proof. Proof is left to the reader. 
4.2.2. Products and Compactness.
Theorem 4.6 (Tychonoff’s theorem). Any product of compact spaces
is compact, i.e., compactness is productive.
Proof. It suffices to prove that any covering of X by open cylin-
ders has a finite subcover. Suppose not and let C be a family of open
cylinders which covers X but for which no finite subcover exists. For
each i ∈ I, consider
{Gij : Gij ⊆ Xi and πi−1 (Gij ) ∈ C}.
This cannot cover Xi (otherwise, Xi , being compact, would be covered
by finitely many, say Xi = Gi1 ∪ Gi2 ∪ . . . ∪ Gin , whence
X = πi−1 (Xi ) = πi−1 (Gi1 ) ∪ . . . ∪ πi−1 (Gin ).
| {z }
all in C, contrary to the choice of C
S
Select, therefore, zi ∈ Xi \ {those Gij ’s}; consider z = (zi )i∈I ∈ X.
Since C covered X, z ∈ some C ∈ C. Now C = πk−1 (Gk ) for some k ∈ I
and so πk (z) = zk ∈ Gk , contradicting the choice of the zi ’s. 
To prove the above without Alexander’s Subbase Theorem is very
difficult in general, but it is fairly simple in the special case where I is
finite. Several further results show that various topological properties
are ’finitely productive’ in this sense.
Theorem 4.7. If (X1 , T1 ), (X2 , T2 ), . . . , (Xn , Tn ) are finitely many
sequentially compact spaces, then their product is sequentially compact.
 4.2. PRODUCTS AND TOPOLOGICAL PROPERTIES 35

Proof. Take any sequence (xn ) ∈ X. The sequence (π1 (xn ))n≥1
in sequentially compact X1 has a convergent subsequence π1 (xnk ) →
l1 ∈ X1 . The sequence (π2 (xnk ))k≥1 in sequentially compact X2 has
a convergent subsequence (π2 (xnkj ))j≥1 → l2 ∈ X2 and π1 (xnkj ) → l1
also.
Do this n times! We get a subsequence (yp )p≥1 of the original se-
quence such that πi (yp ) → li for i = 1, 2, . . . , n. It’s easy to check that
yp → (l1 , l2 , . . . , ln ) so that X is sequentially compact, as required. 
Lemma 4.8. The product of subspaces is a subspace of the product.
Q
Proof. Let (X, T ) = i∈I (Xi , Ti ); let ∅ ⊂ Yi ⊆ Q Xi for each i ∈ I.
There appear to be two different ways to topologise Yi : either
Q
(i) give it the subspace topology induced by Ti , or
(ii) give it the product of all the individual subspace topologies
(Ti )Yi .
The point is that these topologies coincide—if G∗i0 is open in (Ti0 )Yi0
where i0 ∈ I, i.e., G∗i0 = Yi0 ∩ Gi0 for some Gi0 ∈ Ti0 , a typical subbasic
open set for (ii) is Y
{(yi ) ∈ Yi : yi0 ∈ G∗i0 }
which equals
Y Y
Yi ∩ {(xi ) ∈ Xi : xi0 ∈ Gi0 ∈ Ti0 , i0 ∈ I}
Y Y
= Yi ∩ {a typical open cylinder in Xi }
which is a typical subbasic open set in (i). Hence, (i) = (ii). 
Theorem 4.9. Local compactness is finitely productive.
Proof. Given x = (x1 , x2 , . . . , xn ) ∈ (X, T ) = ni=1 (Xi , Ti ), we
Q
must show that x has a compact neighbourhood. Now, for all i =
1, . . . , n, xi has a compact neighbourhood Ci in (Xi , Ti ) so we choose
Ti -open set Gi such that xi ∈ Gi ⊆ Ci . Then
x ∈ G1 × G2 × . . . × Gn ⊆ C1 × C2 × . . . × Cn ,
| {z } | {z }
πi−1 (Gi )
Tn Q
1
compact subset of Xi

i.e., x has C1 ×C2 ×. . .×Cn as a compact neighbourhood. (Note that the

previous lemma is used here, to allow us to apply Tychonoff’s theorem
to the product of the compact subspaces Ci , and then to view this
object as a subspace of the full product!) Thus, X is locally compact.

Q T Q Ti
Lemma 4.10. Yi = Yi (in notation of previous lemma).
36 4. PRODUCT SPACES

Proof. Do it yourself! (The closure of a product is a product of

the closures.) 
4.2.3. Products and Separability.
Theorem 4.11. Separability is finitely productive.
Ti
Proof. For 1 ≤ i ≤ n, choose countable
Q Di ⊆ Xi where Di = Xi .
Consider D = D1 × D2 × . . . × Dn = n1 Di , again countable. Then
Q T Q Ti Q
D = Di = Di = Xi = X. 
Notice that the converses of all such theorems are easily true. For
example,
Q
Theorem 4.12. If (X, T ) = i∈I (Xi , Ti ) is
(i) compact,
(ii) sequentially compact,
(iii) locally compact,
(iv) connected,
(v) separable,
(vi) completely separable
then so is every ‘factor space’ (Xi , Ti ).
Proof. For each i ∈ I, the projection mapping πi : X → Xi is
continuous, open and onto. Thus, by previous results, the result fol-
lows. 
 CHAPTER 5

Separation Axioms

We have observed instances of topological statements which, al-

though true for all metric (and metrizable) spaces, fail for some other
topological spaces. Frequently, the cause of failure can be traced to
there being ‘not enough open sets’ (in senses to be made precise). For
instance, in any metric space, compact subsets are always closed; but
not in every topological space, for the proof ultimately depends on the
observation
‘given x 6= y, it is possible to find disjoint open sets
G and H with x ∈ G and y ∈ H’
which is true in a metric space (e.g., put G = B(x, ), H = B(y, )
where  = 12 d(x, y)) but fails in, for example, a trivial space (X, T0 ).
What we do now is to see how ‘demanding certain minimum levels-
of-supply of open sets’ gradually eliminates the more pathological topolo-
gies, leaving us with those which behave like metric spaces to a greater
or lesser extent.

5.1. T1 Spaces
Definition 5.1. A topological space (X, T ) is T1 if, for each x in
X, {x} is closed.
Comments.
(i). Every metrizable space is T1 .
(ii). (X, T0 ) isn’t T1 unless |X| = 1.
Theorem 5.2.
(i) T1 is hereditary.
(ii) T1 is productive.
(iii) T1 ⇒ every finite set is closed. More precisely, (X, T ) is T1 iff
T ⊇ C, i.e.. C is the weakest of all the T1 topologies that can
be defined on X.
Proof. Proof is left to the reader. 
The respects in which T1 -spaces are ‘nicer’ than others are mostly
concerned with ‘cluster point of a set’ (an idea we have avoided!). We
37
38 5. SEPARATION AXIOMS

show the equivalence, in T1 spaces, of the two forms of its definition

used in analysis.
Theorem 5.3. Given a T1 space (X, T ), p ∈ X and A ⊆ X, the
following are equivalent:
(i) Every neighbourhood of p contains infinitely many points of A.
(ii) Every neighbourhood of p contains at least one point of A dif-
ferent from p.
Proof. Obviously, (i) ⇒ (ii).
Conversely, suppose (i) fails; so there exists a neighbourhood N of
p such that N ∩ A is finite. Consider H = [X \ (N ∩ A)] ∪ {p}; it is
cofinite and is thus an (open) neighbourhood of p. Hence N ∩ H is a
neighbourhood of p which contains no points of A, except possibly p
itself. Thus, (ii) fails also.
Hence, (i) ⇔ (ii). 

5.2. T2 (Hausdorff ) Spaces

Definition 5.4. A topological space (X, T ) is T2 (or Hausdorff)
iff given x 6= y in X, there exist disjoint neighbourhoods of x and y.
Comments.
(i). Every metrizable space is T2 .
(ii). T2 ⇒ T1 , i.e., any T2 space is T1 , for if x, y ∈ X, X is T2 , and
y ∈ {x}, then every neighbourhood of y contains x, whence x = y.
(iii). (X, C), with X infinite, cannot be T2 .
Theorem 5.5.
(i) T2 is hereditary.
(ii) T2 is productive.
Proof. (i). The proofQ is left to the reader.
(ii). Let (X, T ) = i∈I (Xi , Ti ) be any product of T2 spaces. Let
x = (xi )i∈I and y = (yi )i∈I be distinct elements of X. Then there
exists i0 ∈ I such that xi0 6= yi0 in Xi0 . Choose disjoint open sets
G, H in (Xi0 , Ti0 ) so that xi0 ∈ G, yi0 ∈ H. Then x ∈ πi−1 0
(G) ∈ T ,
y ∈ πi−10
(H) ∈ T and since G ∩ H = ∅, π −1
i0 (G) ∩ π −1
i0 (H) = ∅. Hence
result. 
The T2 axiom is particularly valuable when exploring compactness.
Part of the reason is that T2 implies that points and compact sets can
be ‘separated off’ by open sets and even implies that compact sets can
be ‘separated off’ from other compact sets in the same way.
 5.2. T2 (HAUSDORFF) SPACES 39

Theorem 5.6. In a T2 -space (X, T ), if C is a compact set and

x 6∈ C, then there exist T -open sets G and H so that x ∈ G, C ⊆ H
and G ∩ H = ∅.
Sketch of proof. A valuable exercise: separate each point of C
from x using disjoint open sets, note that the open neighbourhoods of
the various elements of C, thus obtained, make up an open covering of
C, reduce it to a finite subcover by appealing to compactness . . . 
Corollary 5.7. In a T2 -space, any compact set is closed.
Corollary 5.8. In a T2 -space, if C and K are non-empty compact
and disjoint, then there exist open G, H such that C ⊆ G, K ⊆ H and
G ∩ H = ∅.
A basic formal distinction between algebra and topology is that al-
though the inverse of a one-one, onto group homomorphism [etc!] is
automatically a homomorphism again, the inverse of a one-one, onto
continuous map can fail to be continuous. It is a consequence of Corol-
lary 5.8 (and Lemma 2.4) that, amongst compact T2 spaces, this cannot
happen.
Theorem 5.9. Let f : (X1 , T1 ) → (X2 , T2 ) be one-one, onto and
continuous, where X1 is compact and X2 is T2 . Then f is a homeo-
morphism.
Proof. It suffices to prove that f is closed. Given closed K ⊆ X1 ,
then K is compact whence f (K) is compact and so f (K) is closed.
Thus f is a closed map. 
Theorem 5.10. (X, T ) is T2 iff no net in X has more than one
limit.
Proof. ⇒. Let x 6= y in X; by hypothesis, there exist disjoint
neighbourhoods U of x, V of y. Since a net cannot eventually belong
to each of two disjoint sets, it is clear that no net in X can converge
to both x and y.
⇐. Suppose that (X, T ) is not Hausdorff and that x 6= y are
points in X for which every neighbourhood of x intersects every neigh-
bourhood of y. Let Nx , Ny be the neighbourhood systems at x, y
respectively. Then both Nx and Ny are directed by reverse inclusion.
We order the Cartesian product Nx × Ny by agreeing that
(Ux , Uy ) ≥ (Vx , Vy ) ⇔ Ux ⊆ Vx and Uy ⊆ Vy .
Evidently, this order is directed. For each (Ux , Uy ) ∈ Nx × Ny , Ux ∩
Uy 6= ∅ and hence we may select a point z(Ux ,Uy ) ∈ Ux ∩ Uy . If Wx is
40 5. SEPARATION AXIOMS

any neighbourhood of x, Wy any neighbourhood of y and (Ux , Uy ) ≥

(Wx , Wy ), then
z(Ux ,Uy ) ∈ Ux ∩ Uy ⊆ Wx ∩ Wy .
That is, the net (z(Ux ,Uy ) , (Ux , Uy ) ∈ Nx × Ny ) eventually belongs to
both Wx and Wy and consequently converges to both x and y! 
Corollary 5.11. Let f : (X1 , T1 ) → (X2 , T2 ), g : (X1 , T1 ) → (X2 , T2 )
be continuous where X2 is T2 . Then their ‘agreement set’ is closed, i.e.,
A = {x : f (x) = g(x)} is closed.
5.3. T3 Spaces
Definition 5.12. A space (X, T ) is called T3 or regular provided:
(i) it is T1 , and
(ii) given x 6∈ closed F , there exist disjoint open sets G and H so
that x ∈ G, F ⊆ H.
Comments.
(i). Every metrizable space is T3 ; for it is certainly T1 and given x
6∈ closed F , we have x ∈ open X \ F so there exists  > 0 so that
x ∈ B(x, ) ⊆ X \ F . Put G = B(x, 2 ) and H = {y : d(x, y) > 2 }; the
result now follows.
(ii). Obviously T3 ⇒ T2 .
(iii). One can devise examples of T2 spaces which are not T3 .
(iv). It’s fairly routine to check that T3 is productive and hereditary.
Warning. Some books take T3 to mean Definition 5.12(ii) alone,
and regular to mean Definition 5.12(i) and (ii); others do exactly the
opposite!
5.4. T3 1 Spaces
2

Definition 5.13. A space (X, T ) is T3 1 or completely regular

2
or Tychonoff iff
(i) it is T1 , and
(ii) given x ∈ X, closed non-empty F ⊆ X such that x 6∈ F , there
exists continuous f : X → [0, 1] such that f (F ) = {0} and
f (x) = 1.
Comments.
(i). Every metrizable space is T3 1 .
2
(ii). Every T3 1 space is T3 . Such a space is certainly T1 and given
2
x 6∈ closed F , choose f as in the definition; define G = f −1 ([0, 13 )),
H = f −1 (( 32 , 1]) and observe that T3 follows.
(iii). Examples are known of T3 spaces which fail to be Tychonoff.
 5.5. T4 SPACES 41

(iv). T3 1 is productive and hereditary.

2

5.5. T4 Spaces
Definition 5.14. A space (X, T ) is T4 or normal if
(i) it is T1 , and
(ii) given disjoint non-empty closed subsets A, B of X, there exist
disjoint open sets G, H such that A ⊆ G, B ⊆ H.
Theorem 5.15. Every metrizable space (X, T ) is T4 .
Proof. Certainly, X is T1 ; choose a metric d on X such that T is
Td . The distance of a point p from a non-empty set A can be defined
thus:
d(p, A) = inf{d(p, a) : a ∈ A}.
Given disjoint non-empty closed sets A, B, let
G = {x : d(x, A) < d(x, B)},
H = {x : d(x, B) < d(x, A)}.
Clearly, G ∩ H = ∅. Also, each is open (if x ∈ G and  = 21 [d(x, B) −
d(x, A)], then B(x, ) ⊆ G, by the triangle inequality.)
Now, if d(p, A) = 0, then for all n ∈ N, there exists xn ∈ A such
that d(p, xn ) < n1 . So d(p, xn ) → 0, i.e., xn → p, whence p ∈ A. Thus
for each x ∈ A, x 6∈ B = B so that d(x, B) > 0 = d(x, A), i.e., x ∈ G.
Hence A ⊆ G. Similarly B ⊆ H. 
It’s true that T4 ⇒ T3 1 but not very obvious. First note that if G0 ,
2
G1 are open in a T4 space with G0 ⊆ G1 , then there exists open G 1
2
with G0 ⊆ G 1 and G 1 ⊆ G1 (because the given G0 and X \ G1 are
2 2
disjoint closed sets so that there exist disjoint open sets G 1 , H such
2
that G0 ⊆ G 1 , X \ G1 ⊆ H, i.e., G1 ⊇ closed X \ H ⊇ G 1 ).
2 2

Lemma 5.16 (Urysohn’s Lemma). Let F1 , F2 be disjoint non-empty

closed subsets of a T4 space; then there exists a continuous function
f : X → [0, 1] such that f (F1 ) = {0}, f (F2 ) = {1}.
Proof. Given disjoint closed F1 and F2 , choose disjoint open G0
and H0 so that F1 ⊆ G0 , F2 ⊆ H0 . Define G1 = X \ F2 (open). Since
G0 ⊆ (closed) X \ H0 ⊆ X \ F2 = G1 , we have G0 ⊆ G1 .
By the previous remark, we can now construct:
(i) G 1 ∈ T : G0 ⊆ G 1 , G 1 ⊆ G1 .
2 2 2
(ii) G 1 , G 3 ∈ T : G0 ⊆ G 1 , G 1 ⊆ G 1 , G 1 ⊆ G 3 , G 3 ⊆ G1 .
4 4 4 4 2 2 4 4
(iii) . . . and so on!
42 5. SEPARATION AXIOMS

Thus we get an indexed family of open sets

{Gr : r = m
2n
,0 ≤ m ≤ 2n , n ≥ 1}
such that r1 ≤ r2 ⇒ Gr1 ⊆ Gr2 .
Observe that the index set is dense in [0, 1]: if s < t in [0, 1], there
exists some 2mn such that s < 2mn < t. Define
(
inf{r : x ∈ Gr } x 6∈ F2
f (x) =
1 x ∈ F2 .

Certainly f : X → [0, 1], f (F2 ) = {1}, f (F1 ) = {0}. To show f contin-

uous, it suffices to show that f −1 ([0, α)) and f −1 ((α, 1]) are open for
0 < α < 1.
Well, f (x) < α iff there exists some r = 2mn such that f (x) < r < α.
It follows that f −1 ([0, α)) = r<α Gr , a union of open sets.
S
Again, f (x) > α iff there exist r1 , r2 such that α < r1 < r2 < f (x),
implying
S that x 6∈ Gr2 whence x 6∈ Gr1 . It follows that f −1 ((α, 1]) =
r1 >α (X \ Gr1 ), which is again open. 
Corollary 5.17. Every T4 space is T3 1 .
2

Proof. Immediate from Lemma 5.16. 

Note that there exist spaces which are T3 1 but not T4 .
2

Theorem 5.18. Any compact T2 space is T4 .

Proof. Use Corollary 5.8 to Theorem 5.5. 
Note. Unlike the previous axioms, T4 is neither hereditary nor pro-
ductive. The global view of the hierarchy can now be filled in as an
exercise from data supplied above:

Metrizable T4 T3 1 T3 T2 T1
2
Hereditary?
Productive?

The following is presented as an indication of how close we are to

having ‘come full circle’.
Theorem 5.19. Any completely separable T4 space is metrizable!
Sketch of proof. Choose a countable base; list as {(Gn , Hn ) :
n ≥ 1} those pairs of elements of the base for which Gn ⊆ Hn . For
 5.5. T4 SPACES 43

each n, use Lemma 5.16 to get continuous fn : X → [0, 1] such that

fn (Gn ) = {0}, fn (X \ Hn ) = {1}. Define
v
uX  2
u fn (x) − fn (y)
d(x, y) = t .
n≥1
2n

One confirms that d is a metric, and induces the original topology.