Topology Course Lecture Notes

Aisling McCluskey and Brian McMaster

August 1997
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 descriptions,
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 un-
derlying set X are open, that the intersection of any two open sets is
again open and that the union of arbitrarily many open sets is open).
This prompts the question: How far would we get if we started with a collec-
tion of subsets possessing these above-mentioned properties and proceeded
to dene 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.
1
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 Dening Topological Spaces
Denition 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 nite intersection.
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 T
d
, the topology
for X induced by the metric d, is dened 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, T
d
)
x G, r
x
> 0 such that x B
r
x
(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 dened by agreeing
that
G X is open in (R, I) (alternatively, I-open)
x G, r
x
> 0 such that (x r
x
, x +r
x
) G.
(iii) Dene T
0
= {, X} for any set X known as the trivial or anti-discrete topology.
(iv) Dene D = {G X : G X} known as the discrete topology.
2
(v) For any non-empty set X, the family C = {G X : G = or X \ G is
nite} is a topology for X called the conite 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 simplied when the sets involved are not
over-described. In particular, it is sometimes suces to use sets which
contain open sets but are not necessarily open. We call such sets neighbor-
hoods.
Denition 1.2 Given a topological space (X, T ) with x X, then N X
is said to be a (T )-neighbourhood of x open set G with x G N.
It follows then that a set U X is open i for every x U, there exists a
neighbourhood N
x
of x contained in U. (Check this!)
Lemma 1.1 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.
(ii) N(x) is closed under nite intersections.
(iii) U N(x) and U V V N(x).
(iv) U N(x) W N(x) such that W U and W N(y) for each
y W.
Proof Exercise!
Examples
(i) Let x X, and dene T
x
= {, {x}, X}. Then T
x
is a topology for X
and V X is a neighbourhood of x i x V . However, the only nhd
of y X where y = x is X itself
3
(ii) Let x X and dene a topology I(x) for X as follows:
I(x) = {G X : x G} {}.
Note here that every nhd of a point in X is open.
(iii) Let x X and dene a topology E(x) for X as follows:
E(x) = {G X : x G} {X}.
Note here that {y} is open for every y = x in X, that {x, y} is not
open, is not a nhd of x yet is a nhd of y.
In fact, the only nhd 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.
Denition 1.3 Let (X, T ) be a topological space. A family B T is called
a base for (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 veried that B T is a base for (X, T ) if and only if whenever
x G T , B B such that x B G.
Clearly, a topological space can have many bases.
Lemma 1.2 If B is a family of subsets of a set X such that
(B1) for any B
1
, B
2
B and every point x B
1
B
2
, there exists B
3
B
with x B
3
B
1
B
2
, 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 ) satises (B1) and (B2).
Proof (Exercise!)
Denition 1.4 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 nite intersections

k
i=1
U
i
, where U
i
S for i = 1, 2, . . . , k is a base for (X, T ).
4
Examples
(i) In any metric space (X, d), {B
r
(x) : x X, r > 0} forms a base for
the induced metric topology T
d
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-innite open intervals (a, ) and (, b) in R
is a subbase for (R, I).
1.1.4 Generating Topologies
From the above examples, it follows that for a set X one can select in many
dierent ways a family T such that (X, T ) is a topological space. If T
1
and
T
2
are two topologies for X and T
2
T
1
, then we say that the topology T
1
is
ner than the topology T
2
, or that T
2
is coarser than the topology T
1
. The
discrete topology for X is the nest one; the trivial topology is the coarsest.
If X is an arbitrary innite 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 ner than the other.
By generating a topology for X, we mean selecting a family T of subsets of
X which satises conditions (T1)(T3). Often it is more convenient not to
describe the family T of open sets directly. The concept of a base oers 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 R
s
, generated by B,
is called the Sorgenfrey line and has many interesting properties. Note
that the Sorgenfrey topology is ner than the Euclidean topology on
R. (Check!)
[Niemytzki plane] Let L denote the closed upper half-plane. We dene
a topology for L by declaring the basic open sets to be the following:
5
(I) the (Euclidean) open discs in the upper half-plane;
(II) the (Euclidean) open discs tangent to the edge of the L, together
with the point of tangency.
Note If y
n
y in L, then
(i) y not on edge: same as Euclidean convergence.
(ii) y on the edge: same as Euclidean, but y
n
must approach y from
inside. Thus, for example, y
n
= (
1
n
, 0) (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:
Denition 1.5 Given a topological space (X, T ) with A X, then the fam-
ily T
A
= {A G : G T } is a topology for A, called the subspace (or
relative or induced) topology for A. (A, T
A
) is called a subspace of (X, T ).
Example
The interval I = [0, 1] with its natural (Euclidean) topology is a (closed)
subspace of (R, I).
Warning: Although this denition, and several of the results which ow 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 neighbourhood 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 background topology of R. In such circumstances, it is
advisable to specify the topology being used each time by saying T -open,
T
A
-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
dene closed sets in a general topological space and
examine the related notion of the closure of a given set.
1.2.1 Closed Sets
Denition 1.6 Given a topological space (X, T ) with F X, then F is said
to be T -closed i its complement X \ F is T -open.
From De Morgans 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 nite 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
Denition 1.7 If (X, T ) is a topological space and A X, then
A
T
= {F X : A F and F is closed}
is called the T -closure of A.
Evidently, A
T
(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 1.3 If (X, T ) is a topological space with A, B X, then
(i)

=
(ii) A

A
(iii)

A =

A
7
(iv) A B =

A

B.
Proof Exercise!
Theorem 1.1 Given a topological space with A X, then x

A i for
each nhd U of x, U A = .
Proof
: Let x

A and let U be a nhd of x; then there exists open G with
x G U. If UA = , then GA = and so A X\G

A X\G
whence x X\G, thereby contradicting the assumption that UA = .
: If x X \

A, then X \

A is an open nhd of x so that, by hypothesis,
(X \

A) A = , which is a contradiction (i.e., a false statement).
Examples
(i) For an arbitrary innite set X with the conite topology C, the closed
sets are just the nite ones together with X. So for any A X,

A =
_
A if A is nite
X if A is innite
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 topology 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, T
x
) dened earlier, if A X, then

A =
_
X if x A
X \ {x} if x A
(iv) For (X, I(x)) with A X,

A =
_
A if x A
X if x A
8
(v) For (X, E(x)) with A X,

A =
_
A if x A
A {x} if x 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 func-
tions in metric spaces as those functions where the inverse image of every
open set is open is used as a denition 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 con-
tinuous in both directions will have many properties in common. These
properties are identied as topological properties. Spaces so related are called
homeomorphic.
1.3.1 Continuity
The primitive intuition of a continuous process is that of one in which small
changes in the input produce small, non-catastrophic changes in the cor-
responding output. This idea formalizes 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 denition is useless in the absence of a properly
dened distance function but, fortunately, it is equivalent to the demand
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:
Denition 1.8 Let (X, T ) and (Y, S) be topological spaces; a mapping f :
X Y is called continuous i f
1
(U) T for each U S i.e. the inverse
image of any open subset of Y is open in X.
9
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, T
0
) 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 = T
x
in the denition of continuity; then the identity
map id
X
: X X is continuous. However, if we interchange T with
S so that T = T
x
and S = I(x), then id
X
: X X is not continuous!
Note that id
X
: (X, T
1
) (X, T
2
) is continuous if and only if T
1
is ner
than T
2
.
Theorem 1.2 If (X
1
, T
1
), (X
2
, T
2
) and (X
3
, T
3
) are topological spaces and
h : X
1
X
2
and g : X
2
X
3
are continuous, then g h : X
1
X
3
is
continuous.
Proof Immediate.
There are several dierent ways to recognise continuity for a mapping be-
tween topological spaces, of which the next theorem indicates two of the most
useful apart from the denition itself:
Theorem 1.3 Let f be a mapping from a topological space (X
1
, T
1
) to a
topological space (X
2
, T
2
). The following statements are equivalent:
(i) f is continuous,
(ii) the preimage under f of each closed subset of X
2
is closed in X
1
,
(iii) for every subset A of X
1
, 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 X
2
, use (iii) on the set A = X
1
\ f
1
(G) and verify
that it follows that f
1
(G) must be open.
10
1.3.2 Homeomorphism
Denition 1.9 Let (X, T ), (Y, S) be topological spaces and let h : X Y
be bijective. Then h is a homeomorphism i h is continuous and h
1
is
continuous. If such a map exists, (X, T ) and (Y, S) are called homeomor-
phic.
Such a map has the property that
G T f(G) S.
It follows that any statement about a topological space which is ultimately
expressible solely in terms of the open sets (together with set-theoretic rela-
tions 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 indistinguishable as topological
spaces. The reader who has had abstract algebra will note that homeo-
morphism is the analogy in the setting of topological spaces and continuous
functions to the notion of isomorphism 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 id
X
: X X is a homeomor-
phism.
A property of topological spaces which when possessed by a space is also pos-
sessed by every space homeomorphic to it is called a topological invariant.
We shall meet some examples of such properties later.
One can readily verify that if f is a homeomorphism, then the inverse map-
ping 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 dicult to demonstrate that two spaces are home-
omorphic (unless a homeomorphism is obvious or can easily be discovered).
For example, to verify that (R, I) is homeomorphic to (0, 1) with its in-
duced metric topology, it is necessary to demonstrate, for instance, that
h : (0, 1) R where h(x) =
2x1
x(x1)
is a homeomorphism.
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 homeomorphic since, for example,
11
(X, I(x)) has the topological invariant each nhd is open while (X, E(x)) does
not.
1.4 Additional Observations
Denition 1.10 A sequence (x
n
) in a topological space (X, T ) is said to
converge to a point x X i (x
n
) eventually belongs to every nhd of x i.e.
i for every nhd U of x, there exists n
0
N such that x
n
U for all n n
0
.
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, T
0
), each sequence
(x
n
) 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 x
n
x where x G,
then Assignment 1 shows that x
n
G for all suciently large n. Fur-
ther, 2 [0, 1]
L
, 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.
Denition 1.11 A topological space (X, T ) is called metrizable i there
exists a metric d on X such that the topology T
d
induced by d coincides with
the original topology T on X.
The investigations above show that (X, T
0
) and (R, L) are examples of non-
metrizable spaces. However, the discrete space (X, D) is metrizable, being
induced by the discrete metric
d(x, y) =
_
1 if x = y
0 if x = y
12
Chapter 2
Topological Properties
We explained in the previous chapter what a topological property (homeo-
morphic 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 theorem of Heine-Borel-
Lebesgue asserts that every covering of such an interval by open sets has a
nite subcover. In this section, we use this feature of closed and bounded
subsets to dene the corresponding notion, compactness, in a general topo-
logical space. In addition, we consider important variants of this notion:
sequential compactness and local compactness.
2.1.1 Compactness Dened
Given a set X with A X, a cover for A is a family of subsets U = {U
i
: I}
of X such that A
iI
U
i
. A subcover of a given cover U for A is a subfamily
V U which still forms a cover for A.
13
If A is a subspace of a space (X, T ), U is an open cover for A i 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 nite subcover. This theorem has
extraordinarily profound consequences and like most good theorems, its con-
clusion has become a denition.
Denition 2.1 (X, T ) is said to be compact i every open cover of X has
a nite subcover.
Theorem 2.1 (Alexanders Subbase Theorem) Let S be any subbase
for (X, T ). If every open cover of X by members of S has a nite subcover,
then X is compact.
The proof of this deep result is an application of Zorns lemma, and is not
an exercise for the faint-hearted!
Examples
(i) (R, I) is not compact, for consider U = {(n, n) : n N}. Similarly,
(C, T
usual
) is not compact.
(ii) (0, 1) is not compact, for consider U = {(
1
n
, 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 nite.
(v) T nite for any X (X, T ) compact.
(vi) X nite, T any topology for X (X, T ) compact.
14
(vii) X innite (X, D) not compact.
(viii) Given (X, T ), if (x
n
) is a sequence in X convergent to x, then {x
n
:
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, T
A
) is
a compact space. Its a nuisance to have to look at T
A
in order to decide on
this. It would be easier to use the original T . Thankfully, we can!
Lemma 2.1 A is a compact subset of (X, T ) i every T -open cover of A
has a nite subcover.
Proof Exercise.
Lemma 2.2 Compactness is closed-hereditary and preserved by continuous
maps.
Proof Exercise.
Example
The unit circle in R
2
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 x
0
M,
K B(x
0
, 1) B(x
0
, 2) B(x
0
, 3)
K
j
i=1
B(x
0
, n
i
)
where we can arrange n
1
< n
2
< . . . < n
j
. Thus K B(x
0
, n
j
) and so any
two points of K lie within n
j
of x
0
and hence within 2n
j
of each other i.e. K
is bounded.
K is closed, since if x

K and x K, then for each y K, d
y
=
1
2
d(x, y) > 0
so we may form the (open) cover of K as follows: {B(y, d
y
) : y K}
which reduces to a nite subcover {B(y
i
, d
y
i
) : y
i
K, i = 1, . . . , n}. The
corresponding neighbourhoods of x, namely B(x, d
y
i
), i = 1, . . . , n, may be
intersected giving a neighbourhood of x which misses K contradiction!)
Neither half is valid in all topological spaces;
15
compact bounded doesnt 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
conite topology!)
Further, in a metric space, a closed bounded subset neednt be compact (e.g.
consider M with the discrete metric and let A M be innite; 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
R
n
; the following is a special case of the theorem:
Theorem 2.2 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 nite subfamily of U}. Note that if x K and a y x, then
y K. Clearly, K = since a K. Moreover, given x K, there exists

x
> 0 such that [x, x +
x
) K (since x some open U chosen nite
subcover of U). Since K is bounded, k

= sup K exists.
(i) 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.
(ii) k

= b : If k

< b, choose U U with k

U and note that

[k

, k

+) U for some > 0 contradiction!

Note
An alternative proof [Willard, Page 116] is to invoke the connected nature of
[a, b] by showing K is clopen in [a, b].
Theorem 2.3 Any continuous map from a compact space into a metric space
is bounded.
Proof Immediate.
Corollary 2.1 If (X, T ) is compact and f : X R is continuous, 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 f(X). Since
16
f(X) = f(X), then there exists > 0 such that (m , m + ) f(X) =
i.e. for all x X, f(x) m . . .contra!
Similarly, if l 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
Denition 2.2 A topological space (X, T ) is said to be sequentially com-
pact if and only if every sequence in X has a convergent subsequence.
Recall from Chapter 1 the denition of convergence of sequences in topolog-
ical spaces and the cautionary remarks accompanying it. There we noted
that, contrary to the metric space situation, sequences in topology can have
several dierent limits! Consider, for example, (X, T
0
) and (R, L). In the
latter space, if x
n
l, then x
n
= l for all n some n
0
. Thus the sequence
1,
1
2
,
1
4
,
1
8
, . . . does not converge in (R, L)!
Lemma 2.3 Sequential compactness is closed-hereditary and preserved by
continuous maps.
Proof Exercise.
We shall prove in the next section that in metric spaces, sequential compact-
ness and compactness are equivalent!
Denition 2.3 Given a topological space (X, T ), a subset A of X and x
X, x is said to be an accumulation point of A i every neighbourhood of
x contains innitely many points of A.
Lemma 2.4 Given a compact space (X, T ) with an innite subset A of X,
then A has an accumulation point.
Proof Suppose not; then for each x X, there exists a neighbourhood N
x
of
x such that N
x
A is (at most) nite; the family {N
x
: x X} is an open
cover of X and so has a nite subcover {N
x
i
: i = 1, . . . , n}. But A X and
A is innite, whence
A = A X = A (N
x
i
) =
n
i=1
(A N
x
i
)
is nite!
17
Lemma 2.5 Given a sequentially compact metric space (M, d) and > 0,
there is a nite number of open balls, radius , which cover M.
Proof Suppose not and that for some > 0, there exists no nite family of
open balls, radius , covering M. We derive a contradiction by constructing
a sequence (x
n
) inductively such that d(x
m
, x
n
) for all n, m (n = m),
whence no subsequence is even Cauchy!
Let x
1
M and suppose inductively that x
1
, . . . , x
k
have been chosen in M
such that d(x
i
, x
j
) for all i, j k, i = j.By hypothesis, {B(x
i
, ) : i =
1, . . . , k} is not an (open) cover of M and so there exists x
k+1
M such that
d(x
k+1
, x
i
) for 1 i k. We thus construct the required sequence (x
n
),
which clearly has no convergent subsequence.
Theorem 2.4 A metric space is compact i it is sequentially compact.
Proof
: Suppose (M, d) is compact. Given any sequence (x
n
) in M, either A =
{x
1
, x
2
, . . .} is nite or it is innite. If A is nite, there must be at least
one point l in A which occurs innitely often in the sequence and its
occurrences form a subsequence converging to l. If A is innite, then by
the previous lemma there exists x X such that every neighbourhood
of x contains innitely many points of A.
For each k , B(x,
1
k
) contains innitely many x
n
s: select one, call
it x
n
k
, making sure that n
k
> n
k1
> n
k2
. . .. We have a subsequence
(x
n
1
, x
n
2
, . . . , x
n
k
, . . .) so that d(x, x
n
k
) <
1
k
0 i.e. x
n
k
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 {G
i
: i I} of M having no nite
subcover. By Lemma 2.5, with =
1
n
(n ), we can cover M by a
nite number of balls of radius
1
n
. For each n, there has to be one of
these, say B(x
n
,
1
n
), which cannot be covered by any nite number of
the sets G
i
. The sequence (x
n
) must have a convergent subsequence
(x
n
k
) which converges to a limit l. Yet {G
i
: i I} covers M, so
l some G
i
0
, say.
As k , x
n
k
l; but also
1
n
k
0 and 1/n
k
is the radius of the ball
centred on x
n
k
. So eventually B(x
n
k
,
1
n
k
) is inside G
i
0
, contradictory to
18
their choice! (More rigorously, there exists m such that B(l,
2
m
)
G
i
0
. Now B(l,
1
m
) contains x
n
k
for all k k
0
say, so choose k k
0
such that n
k
m. Then B(x
n
k
,
1
n
k
) B(l,
2
m
) G
i
0
.) Hence, M is
compact.
2.1.5 Compactness and Uniform Continuity
Recall that a map f : (X
1
, d
1
) (X
2
, d
2
), where (X
i
, d
i
) is a metric space
for each i, is uniformly continuous on X
i
if given any > 0, > 0 such that
d
1
(x, y) < for x, y X
1
d
2
(f(x), f(y)) < .
Ordinary continuity of f is a local property, while uniform continuity is a
global property since it says something about the behaviour of f over the
whole space X
1
. Since compactness allows us to pass from the local to the
global, the next result is not surprising:
Theorem 2.5 If (X, d) is a compact metric space and f : X R is contin-
uous, 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,
x
> 0 such that
d(x, y) < 2
x
|f(x) f(y)| <

2
. The family {B

x
(x) : x X} is an open
cover of X and so has a nite subcover {B

x
i
(x
i
) : i = 1, . . . , n} of X. Let
= min{
x
i
: i = 1, . . . , n}; then, given x, y X such that d(x, y) < , it
follows that |f(x) f(y)| <
(for x B
x
i
(x
i
) for some i, whence d(x, x
i
) <
x
i
and so d(y, x
i
) d(y, x) +
d(x, x
i
) < +
x
i
2
x
i
|f(y) f(x
i
)| <

2
.
Thus |f(x) f(y)| |f(x) f(x
i
)| +|f(x
i
) 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
Denition 2.4 A topological space (X, T ) is locally compact i each point
of X has a compact neighbourhood.
Clearly, every compact space is locally compact. However, the converse is
not true.
Examples
19
(i) With X innite, the discrete space (X, D) is clearly locally compact (for
each x X, {x} is a compact neighbourhood of x!) but not compact.
(ii) With X innite 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 neigh-
bourhood 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 neighbour-
hood C in Q so we can choose > 0 such that J = Q [, ] C.
Now J is closed in (compact) C and is therefore compact in R. Thus,
J must be closed in Rbut

J
R
= [, ]!
Lemma 2.6 (i) Local compactness is closed-hereditary.
(ii) Local compactness is preserved by continuous open maps it is not
preserved by continuous maps in general. Consider id
Q
: (Q, D)
(Q, I
Q
) which is continuous and onto; (Q, D) is locally compact while
(Q, I
Q
) isnt!
Proof Exercise.
2.2 Other Covering Conditions
Denition 2.5 A topological space (X, T ) is said to be
(i) Lindelof i every open cover of X has a countable subcover
(ii) countably compact i every countable open cover of X has a nite
subcover.
Thus, a space is compact precisely when it is both Lindelof and countably
compact. Further, every sequentially compact space is countably 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 condi-
tions compact, countably compact and sequentially compact are equivalent.
Note Second countable separable; separable + metrizable second count-
able . . . and so in metrizable spaces, second countability and separability are
equivalent.
20
2.3 Connectedness
It is not terribly hard to know when a set on the real line is connected, or
of just one piece. This notion is extended to general topological 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 rst term calculus.
2.3.1 Denition 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 denition of
partition is not aected by replacing the term open by closed.
Denition 2.6 A connected space (X, T ) is one which has no partition.
(Otherwise, (X, T ) is said to be disconnected.)
If = A (X, T ), we call A a connected set in X whenever (A, T
A
) is a
connected space.
Lemma 2.7 (X, T ) is connected i X and are the only subsets which are
clopen.
Examples
(i) (X, T
0
) is connected.
(ii) (X, D) cannot be connected unless |X| = 1. (Indeed the only connected
subsets are the singletons!)
(iii) The Sorgenfrey line R
s
is disconnected (for [x, ) is clopen!).
(iv) The subspace Q of (R, I) is not connected because
Q [

2,

2]
. .
closed in Q
= Q (

2,

2)
. .
open in Q
21
is clopen and is neither universal nor empty.
(v) (X, C) is connected except when X is nite; indeed, every innite sub-
set of X is connected.
(vi) (X, L) is connected except when X is countable; indeed, every uncount-
able subset of X is connected.
(vii) In (R, I), A R is connected i 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 2.8 A (X, T ) is not connected i there exist T -open sets G,
H such that A GH, AG = , AH = and AGH = . (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 2.9 In R, if [a, b] = F
1
F
2
where F
1
, F
2
are both closed and a F
1
,
b F
2
then F
1
F
2
= .
Proof Exercise.
Theorem 2.6 Let I (R, I). Then I is connected i I is an interval.
Proof
: If I is not an interval, then there exist a < b < c with a I, b I and
c I. Take A = I (, b) and B = I (b, ). Then A B = I,
A B = , A = , B = , 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.
22
: Suppose I is not connected and that I is an interval. By the closed
version of Lemma 2.8, there exist closed subsets K
1
, K
2
of R such
that I K
1
K
2
, I K
1
= , I K
2
= and I K
1
K
2
= .
Select a I K
1
, b I K
2
; without loss of generality, a < b.
Then [a, b] I so that [a, b] = ([a, b] K
1
) ([a, b] K
2
), whence by
Lemma 2.9, = [a, b] K
1
K
2
I K
1
K
2
= !
2.3.3 Connectedness and Continuous Maps
Lemma 2.10 Connectedness is preserved by continuous maps.
Proof Exercise.
Corollary 2.2 (Intermediate Value Theorem) If f : [a, b] R is con-
tinuous and f(a) < y < f(b), then y must be a value of f.
Proof Exercise.
Corollary 2.3 (Fixed point theorem for [0, 1]) If f : [0, 1] [0, 1] is
continuous, then it has a xed 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 continuous. 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 xed point
since [a, b]

= [0, 1] and every continuous function has a xed point is a
homeomorphic invariant.
Lemma 2.11 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 = = 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 2.12 If the family {A
i
: i I} of connected subsets of a space
(X, T ) has a non-empty intersection, then its union
iI
A
i
is connected.
23
Proof Suppose not and that there exists a non-empty proper clopen subset Y
of
iI
A
i
. Then for each i I, either A
i
Y or A
i

iI
A
i
\ Y . However
if for some j, A
j
Y , then A
i
Y for each i I (since
iI
A
i
= ) which
implies that
iI
A
i
Y !
Similarly, if for some k I, A
k

iI
A
i
\ Y , then
iI
A
i

iI
A
i
\ Y !
Corollary 2.4 Given a family {C
i
: i I} of connected subsets of a space
(X, T ), if B X is also connected and B C
i
= for all i I, then
B (
iI
C
i
) is connected.
Proof Take A
i
= B C
i
in Lemma 2.12.
Lemma 2.13 If A is a connected subset of a space (X, T ) and A B

A
T
,
then B is a connected subset.
Proof If B is not connected, then there exists Y B which is clopen
in B. By Lemma 2.11, either A Y or A B \ Y . Suppose A Y (a
similar argument suces for A B \ Y ); then

A
T


Y
T
and so B \ Y =
B (B \ Y ) =

A
T
B
(B \ Y )

Y
T
B
(B \ Y ) = Y (B \ Y ) = a
contradiction!
Denition 2.7 Let (X, T ) be a topological space with x X; we dene the
component of x, C
x
, in (X, T ) to be the union of all connected subsets of
X which contain x i.e.
C
x
= {A X : x A and A is connected}.
For each x X, it follows from Lemma 2.12 that C
x
is the maximum con-
nected subset of X which contains x. Also it is clear that if x, y X, either
C
x
= C
y
or C
x
C
y
= (for if z C
x
C
y
, then C
x
C
y
C
z
C
x
C
y
whence C
x
= C
y
(= C
z
)). Thus we may speak of the components of a space
(X, T ) (without reference to specic points of X): they partition the space
into connected closed subsets (by Lemma 2.13) 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 singletons. (Thus,
components need not be open.)
24
Denition 2.8 A space (X, T ) is totally disconnected i the only con-
nected subsets of X are the singletons (equivalently, the components of (X, T )
are the singletons).
Thus, by the previous examples, we see that the space Q of rationals, the
space R\ Q of irrationals and any discrete space are all totally disconnected.
Further, the Sorgenfrey line R
s
is totally disconnected.
2.3.4 Pathwise Connectedness
Denition 2.9 A topological space (X, T ) is pathwise connected i 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 2.7 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 p
x
: [0, 1] X from a to x. Then, for each x X, p
x
([0, 1])
is connected; moreover, p
a
(0) = a
xX
p
x
([0, 1]) so that by Lemma 2.12,
X =
xX
p
x
([0, 1]) is connected.
Note well The converse is false. Consider the following example, the topolo-
gists sine curve:
V = {(x, 0) : x 0} {(x, sin
1
x
) : x > 0}
is a connected space, but no path can be found from (0, 0) to any point
(x, sin
1
x
) with x > 0
(for suppose, w.l.o.g., there exists a path p : [0, 1] X with p(0) = (
1

, 0)
and p(1) = (0, 0). Then
1
p, being continuous, must take all values between
0 and
1

, in particular
1
(2n+
1
2
)
for each n i.e. there exists t
n
[0, 1] such that

1
p(t
n
) =
1
(2n+
1
2
)
for each n. Thus, p(t
n
) = (
1
(2n+
1
2
)
, 1) (0, 1) as n .
Now t
n
[0, 1] for all n which implies that there exists a subsequence (t
n
k
)
in [0, 1] with t
n
k
. Then p(t
n
k
) p() and so
1
p(t
n
k
) 0. Thus
p() = (0, y) for some y, whence y = 0 (since p() X)!)
2.4 Separability
Denition 2.10 A topological space is said to be
25
(i) separable i it has a countable dense subset.
(ii) completely separable (equivalently,second countable) i it has a
countable base.
Examples
(i) (R, I) is separable (since

Q = R).
(ii) (X, C) is separable for any X.
(iii) (R, L) is not separable.
Theorem 2.8 (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 = {x
1
, x
2
, . . .} be
dense. Consider B = {B(x
i
, q) : i , q Q, q > 0}, a countable collection
of open sets. One can show that B is a base for T
d
. . . over to you!
Theorem 2.9 (i) Complete separability is hereditary.
(ii) Separability is not hereditary. (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.)
Example
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 = 0, {0, x} is an open neighbourhood of x
and so there exists B
x
B such that x B
x
{0, x}.Thus B
x
= {0, x} i.e.
B is uncountable . . . contradiction!
Theorem 2.10 Separability is preserved by continuous maps.
Proof Exercise.
Note Complete separability is not preserved by continuous maps.
26
Chapter 3
Convergence
In Chapter 1, we dened limits of sequences in a topological space (X, T ) so
as to assimilate the metric denition. 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 diculty:
develop a kind of super-sequence or net which does for general topol-
ogy 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 includes the
limit of every convergent sequence of points of A.
Theorem 3.3 Given f : (X, T ) (Y, T

): if f is continuous, then f pre-

serves limits of sequences i.e. whenever x
n
l in X, then f(x
n
) f(l) in
Y .
27
In each case above, it is routine to prove the statement true in a general
topological space as asserted. We illustrate by proving Theorem 3.3:
Let f be continuous and x
n
l in X. We must show that f(x
n
) 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
x
n
f
1
(N)n n
0
say. Thus f(x
n
) Nn n
0
, whence f(x
n
) 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.
Example
In (R, L), (0, 1) = R. So, for example, 5 (0, 1) and yet the only way a
sequence (x
n
) converges to a limit l is for x
n
= 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 3.2s
converse is false.
Further, id
R
: (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 do-
main. The set of positive integers, of course, possesses a particularly simple
ordering; there is a rst member, second member, third member, etc. Not all
sets are naturally endowed with so simple an ordering. For example, dictio-
nary (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:
dene a net (in analogy with sequence),
identify an associated notion of convergence,
show that net convergence is sucient to characterize closure of sets,
28
and that compactness can be characterized in terms of convergence of
subnets.
Note that these last two items generalize the role of sequences in a metric
space.
3.2.1 Denition of Net
Denition 3.1 A binary relation on a set P is said to be a pre-order i
(i) p p p P
(ii) p q and q r imply p r 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.
If it is also true that for p, q P,
(iii) p q and q p imply p = q, P is said to be a partially ordered set
(or poset).
Denition 3.2 A pre-ordered set P is said to be directed (or updirected)
i 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.)
Denition 3.3 Let (P, ) be a poset. Then if x, y P with x y and
y x, we write x y and say that x and y are incomparable;
If E P, then E is said to be totally unordered (or diverse) i x, y E
implies x = y or x y.
If C P, then C is said to be linear (or a chain or a total order) i
x, y C implies x < y, x = y or y < x.
(P, ) is said to be a lattice i each pair of members of P has a greatest
lower bound and a least upper bound.
A lattice (P, ) is said to be complete i 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) i v x
(x v), x P v = x.
Denition 3.4 A net in a (non-empty set) X is any function x : A X
whose domain A is a directed set.
29
In imitation of the familiar notation in sequences, we usually write the net
value x() as x

. A typical net x : A X will usually appear as (x

, A)
or (x

)
A
or some such notation.
Examples of Nets
(i) N, Z, NN are all directed sets, where suitable pre-orders are respec-
tively the usual magnitude ordering for N and Z, and (i, j) (m, n)
i 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
1
x
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 x
N
from each neigh-
bourhood N of x; then (x
N
)
NN
x
is a net in X ( since it denes a
mapping from (N
x
, ) into X). Recall that N
x
is ordered by inverse
set inclusion!
3.2.2 Net Convergence
Denition 3.5 A net (x

)
A
in (X, T ) converges to a limit l if for each
neighbourhood N of l, there exists some
N
A such that x

N for all

N
.
In such a case, we sometimes say that the net (x

)
A
eventuates N. Clearly, this denition incorporates the old denition of limit
of 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 x
N
are chosen
. . . prove!
3.2.3 Net Convergence and Closure
Our claim is that nets fully describe the structure of a topological space. Our
rst 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 3.4 Given (X, T ), A X, p X: p

A i there exists a net in
A converging to p.
30
Proof If some net of points of A converges to p, then every neighbourhood
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 a
N
of A that belongs also to N. The
net which these choices constitute converges to p, as required.
Theorem 3.5 Given (X, T ), A X, A is closed i it contains every limit
of every (convergent) net of its own points.
Proof This is really just a corollary of the preceding theorem.
Theorem 3.6 Given f : (X, T ) (Y, T

), f is continuous i f preserves
net convergence.
Proof Exercise.
3.2.4 Nets and Compactness
Denition 3.6 Let (x

)
A
be any net and let
0
A. The
th
0
tail of the
net is the set {x

:
0
} = x([
0
, )). We denote it by x(
0
).
Denition 3.7 Let (x

)
A
and (y

)
B
be any two nets. We call (y

)
B
a subnet of (x

)
A
provided that every tail of (x

) contains a tail of (y

)
i.e. provided:

0
A
0
B such that x(
0
) y(
0
).
We expected a denition like subsequence to turn up here and we are dis-
appointed 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 well try to avoid this whenever
possible! There is however one result certainly worth knowing:
Theorem 3.7 (X, T ) is compact i in X, every net has (at least one) con-
vergent subnet.
(So, for example, (n) is a net in R with no convergent subnet.)
Proof Not required.
Corollary 3.1 Compactness is closed-hereditary
31
Proof (for if (x

) is a net in a closed set F X, then it has a convergent

subnet (y

) in X. Thus there exists a subnet (z

) of (y

) in F which converges
in X, whence its limit is in F).
Corollary 3.2 Compactness is preserved by continuous maps
Proof (for if X is compact and f continuous, let (y

)
A
be a net in f(X).
Then for each A, y

= f(x

) for some x

X. The net (x

)
A
has a convergent subnet (z

)
B
, say z

l, whence f(z

) f(l). Then
(f(z

))
B
is a convergent subnet of (y

)
A
).
Example
If (x
n
k
) is a subsequence of a sequence (x
n
), then it is a subnet of it also;
because the i
th
0
tail of the sequence (x
n
) is
{x
i
0
, x
i
0
+1
, x
i
0
+2
, . . .} ()
while the i
th
0
tail of the subsequence (x
n
k
) is:
{x
n
i
0
, x
n
i
0
+1
, x
i
n
0
+2
, . . .} ()
and we see that () () merely because n
i
0
i
0
.
Lemma 3.1 If a net (x

) converges to a limit l, then so do all its subnets.

Proof Let (y

) be a subnet of (x

); let N be a neighbourhood of l. Then

there exists
0
such that x

N for all
0
. Further, there exists
0
such
that {y

:
0
} {x

:
0
} and so y

N for all
0
.
3.3 First Countable Spaces - Where Sequences
Suce
Why do sequences suce to describe structure in R, C and other metric
spaces but not in many other topological spaces? The key here is recognizing
that many proofs regarding convergence in metric spaces 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 suce to describe the topological structure
of spaces with this characteristic.
32
3.3.1 First Countable Spaces
So what characteristic common to R, C and other metric spaces makes se-
quences so good at describing their structure?
Denition 3.8 Let x (X, T ). A countable neighbourhood base at
x means: a sequence N
1
, N
2
, N
3
, . . . of particular neighbourhoods of x such
that every neighbourhood of x shall contain one of the N
i
s.
Note that we may assume that N
1
N
2
N
3
because, if not, then we
can work with N
1
, N
1
N
2
, N
1
N
2
N
3
, . . .
Denition 3.9 We call (X, T ) rst-countable when every point in X has
a countable neighbourhood base.
Example The classic example of a rst-countable space is any metric (or
metrizable) space because if x (M, d), then B(x, 1), B(x,
1
2
), B(x,
1
3
), . . . is
a countable neighbourhood base at x.
Theorem 3.8 First-countability is hereditary and preserved by continuous
open onto maps.
Proof Left to the reader.
Theorem 3.9 (i) Complete separability implies rst 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 rst countable ( since metrizable), yet
is not completely separable.
(iii) Suppose X countable and (X, T ) rst 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.
33
Example
The Arens-Fort space (see, for example, Steen and Seebach, Counterexamples
in Topology is not rst-countable because otherwise it would be completely
separable which is false!
3.3.2 Power of Sequences in First Countable Spaces
The following three results illustrate that sequences suce for rst-countable
spaces in the sense that we dont need to use nets to describe their structure.
This is why sequences are suciently general to describe, fully, metric and
metrizable spaces.
Theorem 3.10 Given a rst-countable space (X, T )
(i) p X, A X, then p

A i there exists a sequence of points of A
converging to p.
(ii) A X is closed i A contains every limit of every convergent sequence
of its own points.
(iii) f : (X, T ) (Y, T

) is continuous i 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
N
1
N
2
N
3
, each of which must intersect A. So choose
x
j
N
j
A for all j 1. Then (x
j
) is a sequence in A and, given any
neighbourhood H of p, H must contain one of the N
j
s i.e. H N
j
0

N
j
0
+1
so that x
j
H for all j j
0
. That is, x
j
p.
(ii) Corollary of (i).
(iii) f continuous implies that it must preserve limits of sequences (by The-
orem 3.3). Conversely, if f is not continuous, there exists A X such
that f(

A) f(A). Thus, there exists p f(

A) \ f(A) so p = f(x),
some x

A. So there exists a sequence (x
n
) in A with x
n
x.
34
Yet, if f(x
n
) 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.
35
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 topological spaces?
What topological properties are preserved by this construction?
4.1 Constructing Products
The process of constructing a product falls naturally into two stages.
The rst stage, which is entirely set-theoretic, consists in describing 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.
36
4.1.1 Set-Theoretic Construction
Suppose throughout that we are given a family of topological spaces {(X
i
, T
i
) :
i I} where I is some non-empty labelling or index set.
Our rst task is to get a clear mental picture of what we mean by the product
of the sets X
i
. Look again at the nite case where I = {1, 2, . . . , n}. Here,
the product set
X = X
1
X
2
X
3
. . . X
n
=
n

i=1
X
i
= {(p
1
, p
2
, . . . , p
n
) : p
i
X
i
, i I}.
i.e. the elements of X are the functions x : I
n
i=1
X
i
such that x(1) X
1
,
x(2) X
2
, . . . , x(n) X
n
i.e. x(i) X
i
i where, for convenience, we usually
write x
i
instead of x(i). In this form, the denition extends immediately to
any I, nite or innite i.e. if {X
i
: i I} is any family of sets, then their
product is
{x : I
iI
X
i
for which x(i) X
i
i I}
except that we normally write x
i
rather than x(i).
Then a typical element of X =

X
i
will look like: (x
i
)
iI
or just (x
i
). We
will still call x
i
the i
th
coordinate of (x
i
)
iI
. ( Note that the Axiom of Choice
assures us that

X
i
is non-empty provided none of the X
i
s are empty.)
4.1.2 Topologizing the Product
Of the many possible topologies that could be imposed on X =

X
i
, we
describe the most useful. This topology is just right in the sense that it
is barely ne enough to guarantee the continuity of the coordinate projec-
tion functions while being just course enough allow the important result of
Theorem 4.1.
Denition 4.1 For each i I, the i
th
projection is the map
i
:

X
i

X
i
which selects the i
th
coordinate i.e.
i
((x
i
)
iI
) = x
i
.
An open cylinder means the inverse projection of some non-empty T
i
-open
set i.e.
1
i
(G
i
) where i I, G
i
= , G
i
T
i
.
An open box is the intersection of nitely many open cylinders
n
j=1

1
i
j
(G
i
j
).
The only drawable case I = {1, 2} may help explain:
(Here will be, eventually, a picture!)
37
We use these open cylinders and boxes to generate a topology with just
enough open sets to guarantee that projection maps will be continuous.
Note that the open cylinders form a subbase for a certain topology T on
X =

X
i
and therefore the open boxes form a base for T ; T is called the
[Tychono]product topology and (X, T ) is the product of the given fam-
ily of spaces. We write (X, T ) =

{(X
i
, T
i
) : i I} =

iI
(X
i
, T
i
) or even
T =

iI
T
i
.
Notice that if
n
j=1

1
i
j
(G
i
j
) is any open box, then without loss of generality
we can assume i
1
, i
2
, . . . i
n
all dierent because if there were repetitions like
. . .
1
i
k
(G)
1
i
k
(H) . . .
we can replace each by
. . .
1
i
k
(G H) . . .
and thus eliminate all repetitions.
It is routine to check that if T
n
is the usual topology on R
n
, and T the usual
topology on R, then
(R, T ) (R, T ) . . . (R, T ) = (R
n
, T
n
)
as one would hope!
Lemma 4.1 In a product space (X, T ), N is a neighbourhood of p X i
there exists some open box B such that p B N.
Lemma 4.2 For each i I,
(i)
i
is continuous
(ii)
i
is an open mapping.
Proof
(i) Immediate.
(ii) Given open G X, then G is a union of basic open sets {B
k
: k K}
in X, whence
i
(G) is a union of open subsets {B
i
k
: k K} of X
i
and
is therefore open. (The notation here is intended to convey that B
i
k
is
the component along the i-th coordinate axis of the open box B
k
.)
38
Theorem 4.1 A map into a product space is continuous i its composite
with each projection is continuous.
Proof Since the projections are continuous, so must be their composites with
any continuous map. To establish the converse, rst 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 ).
Solution
Dene 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 suce to show that both are contin-
uous.

1
f =

2
;
2
f =

1
. Now

i
is continuous for i = 1, 2 and so f is
continuous! Similarly, g is continuous.
Worked example Show that the product of innitely many copies of (N, D)
is not locally compact.
Solution
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. Pick an open box B such that p B G C. B looks like

n
j=1

1
i
j
(G
i
j
). Choose i
n+1
I \ {i
1
, i
2
, . . . , i
n
}; then
i
n+1
(C) is compact
(since compactness is preserved by continuous maps).
Thus, p
i
n+1

i
n+1
(B) = X
i
n+1

i
n+1
(C) X
i
n+1
= (N, D). Thus,

i
n+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 several theorems
which assert that certain topological properties are productive i.e. are
possessed by the product if enjoyed by each factor. Several of these theorems
are given below.
39
4.2.1 Products and Connectedness
Theorem 4.2 Any product of connected spaces must be connected.
Proof is left to the reader.
4.2.2 Products and Compactness
Theorem 4.3 (Tychonos theorem) Any product of compact spaces is com-
pact i.e. compactness is productive.
Proof It suces to prove that any covering of X by open cylinders has a
nite subcover. Suppose not and let C be a family of open cylinders which
covers X but for which no nite subcover exists. For each i I, consider
{G
i
j
: G
i
j
X
i
and
1
i
(G
i
j
) C}.
This cannot cover X
i
(otherwise, X
i
, being compact, would be covered by
nitely many, say X
i
= G
i
1
G
i
2
. . . G
in
, whence
X =
1
i
(X
i
) =
1
i
(G
i
1
. . .
1
i
(G
i
n
).
. .
all in C, contrary to the choice of C
Select, therefore, z
i
X
i
\ { those G
i
j
s}; consider z = (z
i
)
iI
X. Since
C covered X, z some C C. Now C =
1
k
(G
k
) for some k I and so

k
(z) = z
k
G
k
, contradicting the choice of the z
i
s.
To prove the above without Alexanders Subbase Theorem is very dicult in
general, but it is fairly simple in the special case where I is nite. Several fur-
ther results show that various topological properties are nitely productive
in this sense.
Theorem 4.4 If (X
1
, T
1
), (X
2
, T
2
), . . . , (X
n
, T
n
) are nitely many sequen-
tially compact spaces, then their product is sequentially compact.
Proof
Take any sequence (x
n
) X. The sequence (
1
(x
n
))
n1
in sequentially
compact X
1
has a convergent subsequence
1
(x
n
k
) l
1
X
1
. The se-
quence (
2
(x
n
k
))
k1
in sequentially compact X
2
has a convergent subsequence
(
2
(x
n
k
j
))
j1
l
2
X
2
and
1
(x
n
k
j
) l
1
also.
Do this n times! We get a subsequence (y
p
)
p1
of the original sequence such
that
i
(y
p
) l
i
for i = 1, 2, . . . , n. Its easy to check that y
p
(l
1
, l
2
, . . . , l
n
)
so that X is sequentially compact, as required.
40
Lemma 4.3 The product of subspaces is a subspace of the product.
Proof
Let (X, T ) =

iI
(X
i
, T
i
); let Y
i
X
i
for each i I. There appear to
be two dierent ways to topologise

Y
i
:
either (i) give it the subspace topology induced by

T
i
or (ii) give it the product of all the individual subspace topologies (T
i
)
Y
i
.
The point is that these topologies coincideif G

i
0
is open in (T
i
0
)
Y
i
0
where
i
0
I i.e. G

i
0
= Y
i
0
G
i
0
for some G
i
0
T
i
0
, a typical subbasic open set for
(ii) is
{(y
i
)

Y
i
: y
i
0
G

i
0
}
which equals

Y
i
{(x
i
)

X
i
: x
i
0
G
i
0
T
i
0
, i
0
I}
=

Y
i
{ a typical open cylinder in

X
i
}
which is a typical subbasic open set in (i). Hence, (i) = (ii).
Theorem 4.5 Local compactness is nitely productive.
Proof
Given x = (x
1
, x
2
, . . . , x
n
) (X, T ) =

n
i=1
(X
i
, T
i
), we must show that x
has a compact neighbourhood. Now, for all i = 1, . . . , n, x
i
has a compact
neighbourhood C
i
in (X
i
, T
i
) so we choose T
i
-open set G
i
such that x
i
G
i

C
i
. Then
x G
1
G
2
. . . G
n
. .

n
1

1
i
(G
i
)
C
1
C
2
. . . C
n
. .
compact subset of

X
i
i.e. x has C
1
C
2
. . . C
n
as a compact neighbourhood. (Note that the
previous lemma is used here, to allow us to apply Tychonos theorem to
the product of the compact subspaces C
i
, and then to view this object as a
subspace of the full product!) Thus, X is locally compact.
Lemma 4.4

Y
i
T
=

Y
i
T
i
( in notation of previous lemma).
Proof Do it yourself! (The closure of a product is a product of the closures.)
41
4.2.3 Products and Separability
Theorem 4.6 Separability is nitely productive.
Proof
For 1 i n, choose countable D
i
X
i
where

D
i
T
i
= X
i
. Consider
D = D
1
D
2
. . . D
n
=

n
1
D
i
, again countable. Then

D =

D
i
T
=

D
i
T
i
=

X
i
= X.
Notice that the converses of all such theorems are easily true. For example,
Theorem 4.7 If (X, T ) =

iI
(X
i
, T
i
) is
(i) compact
(ii) sequentially compact
(iii) locally compact
(iv) connected
(v) separable
(vi) completely separable
then so is every factor space (X
i
, T
i
).
Proof For each i I, the projection mapping
i
: X X
i
is continuous,
open and onto. Thus, by previous results, the result follows.
42
Chapter 5
Separation Axioms
We have observed instances of topological statements which, although 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 = y, it is possible to nd 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
=
1
2
d(x, y)) but fails in, for example, a trivial space (X, T
0
).
What we do now is to see how demanding certain minimum levels-of-supply
of open sets gradually eliminates the more pathological topologies, leaving
us with those which behave like metric spaces to a greater or lesser extent.
5.1 T
1
Spaces
Denition 5.1 A topological space (X, T ) is T
1
if, for each x in X, {x} is
closed.
Comment 5.1 (i) Every metrizable space is T
1
(ii) (X, T
0
) isnt T
1
unless |X| = 1
Theorem 5.1 (i) T
1
is hereditary
43
(ii) T
1
is productive
(iii) T
1
every nite set is closed. More precisely, (X, T ) is T
1
i T C,
i.e. C is the weakest of all the T
1
topologies that can be dened on X.
Proof is left to the reader.
The respects in which T
1
-spaces are nicer than others are mostly concerned
with cluster point of a set (an idea we have avoided!). We show the equiv-
alence, in T
1
spaces, of the two forms of its denition used in analysis.
Theorem 5.2 Given a T
1
space (X, T ), p X and A X, the following
are equivalent:
(i) Every neighbourhood of p contains innitely many points of A
(ii) Every neighbourhood of p contains at least one point of A dierent from
p.
Proof Obviously, (i) (ii); conversely, suppose (i) fails; so there exists a
neighbourhood N of p such that N A is nite. Consider H = [X \ (N
A)]{p}; it is conite and is thus an (open) neighbourhood of p. Hence NH
is a neighbourhood of p which contains no points of A, except possibly p itself.
Thus, (ii) fails also.
Hence, (i) (ii).
5.2 T
2
(Hausdor) Spaces
Denition 5.2 A topological space (X, T ) is T
2
(or Hausdor) i given
x = y in X, disjoint neighbourhoods of x and y.
Comment 5.2 (i) Every metrizable space is T
2
(ii) T
2
T
1
(i.e. any T
2
space is T
1
, for if x, y T
2
X and y {x}, then
every neighbourhood of y contains x, whence x = y.)
(iii) (X, C), with X innite, cannot be T
2
Theorem 5.3 (i) T
2
is hereditary
(ii) T
2
is productive.
44
Proof
i The proof is left to the reader.
ii Let (X, T ) =

iI
(X
i
, T
i
) be any product of T
2
spaces. Let x = (x
i
)
iI
and y = (y
i
)
iI
be distinct elements of X. Then there exists i
0
I
such that x
i
0
= y
i
0
in X
i
0
. Choose disjoint open sets G, H in (X
i
0
, T
i
0
)
so that x
i
0
G, y
i
0
H. Then x
1
i
0
(G) T , y
1
i
0
(H) T and
since G H = ,
1
i
0
(G)
1
i
0
(H) = . Hence result.
The T
2
axiom is particularly valuable when exploring compactness. Part of
the reason is that T
2
implies that points and compact sets can be separated
o by open sets and even implies that compact sets can be separated o
from other compact sets in the same way.
Theorem 5.4 In a T
2
-space (X, T ), if C is a compact set and x C, then
there exist T -open sets G and H so that x G, C H and G H = .
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 nite subcover
by appealing to compactness . . .
Corollary 5.1 In a T
2
-space, any compact set is closed.
Corollary 5.2 In a T
2
-space, if C and K are non-empty compact and dis-
joint, 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 although
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 Corollary 5.2 that, amongst compact
T
2
spaces, this cannot happen.
Theorem 5.5 Let f : (X
1
, T
1
) (X
2
, T
2
) be one-one, onto and continuous,
where X
1
is compact and X
2
is T
2
. Then f is a homeomorphism.
Proof It suces to prove that f is closed. Given closed K X
1
, then K is
compact whence f(K) is compact and so f(K) is closed. Thus f is a closed
map.
45
Theorem 5.6 (X, T ) is T
2
i no net in X has more than one limit.
Proof
(i) (ii): Let x = 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.
(ii) (i): Suppose that (X, T ) is not Hausdor and that x = y are points in X
for which every neighbourhood of x intersects every neighbourhood of
y. Let N
x
(N
y
) be the neighbourhood systems at x (y) respectively.
Then both N
x
and N
y
are directed by reverse inclusion. We order the
Cartesian product N
x
N
y
by agreeing that
(U
x
, U
y
) (V
x
, V
y
) U
x
V
x
and U
y
V
y
.
Evidently, this order is directed. For each (U
x
, U
y
) N
x
N
y
, U
x

U
y
= and hence we may select a point z
(U
x
,U
y
)
U
x
U
y
. If W
x
is
any neighbourhood of x, W
y
any neighbourhood of y and (U
x
, U
y
)
(W
x
, W
y
), then
z
(U
x
,U
y
)
U
x
U
y
W
x
W
y
.
That is, the net {z
(U
x
,U
y
)
, (U
x
, U
y
) N
x
N
y
} eventually belongs to
both W
x
and W
y
and consequently converges to both x and y!
Corollary 5.3 Let f : (X
1
, T
1
) (X
2
, T
2
), g : (X
1
, T
1
) (X
2
, T
2
) be con-
tinuous where X
2
is T
2
. Then their agreement set is closed i.e. A = {x :
f(x) = g(x)} is closed.
5.3 T
3
Spaces
Denition 5.3 A space (X, T ) is called T
3
or regular provided :-
(i) it is T
1
, and
(ii) given x closed F, there exist disjoint open sets G and H so that
x G, F H.
Comment 5.3 (i) Every metrizable space is T
3
; for it is certainly T
1
and
given x 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.
46
(ii) Obviously T
3
T
2
.
(iii) One can devise examples of T
2
spaces which are not T
3
.
(iv) Its fairly routine to check that T
3
is productive and hereditary.
(v) Warning: Some books take T
3
to mean Denition 5.3(ii) alone, and
regular to mean Denition 5.3(i) and (ii); others do exactly the oppo-
site!
5.4 T
3
1
2
Spaces
Denition 5.4 A space (X, T ) is T
3
1
2
or completely regular or Tychono
i
(i) it is T
1
, and
(ii) given x X, closed non-empty F X such that x F, there exists
continuous f : X [0, 1] such that f(F) = {0} and f(x) = 1.
Comment 5.4 (i) Every metrizable space is T
3
1
2
(ii) Every T
3
1
2
space is T
3
( such a space is certainly T
1
and given x
closed F, choose f as in the denition; dene G = f
1
([0,
1
3
)), H =
f
1
((
2
3
, 1]) and observe that T
3
follows.)
(iii) Examples are known of T
3
spaces which fail to be Tychono
(iv) T
3
1
2
is productive and hereditary.
5.5 T
4
Spaces
Denition 5.5 A space (X, T ) is T
4
or normal if
(i) it is T
1
, 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 5.7 Every metrizable space (X, T ) is T
4
.
47
Proof Certainly, X is T
1
; choose a metric d on X such that T is T
d
. The
distance of a point p from a non-empty set A can be dened 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, GH = . Also, each is open (if x G and =
1
2
{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 x
n
A such that d(p, x
n
) <
1
n
. So d(p, x
n
) 0
i.e. x
n
p, whence p

A. Thus for each x A, x B =

B so that
d(x, B) > 0 = d(x, A) i.e. x G. Hence A G. Similarly B H.
Its true that T
4
T
3
1
2
but not very obvious. First note that if G
0
, G
1
are
open in a T
4
space with

G
0
G
1
, then there exists open G1
2
with

G
0
G1
2
and

G1
2
G
1
(because the given

G
0
and X \ G
1
are disjoint closed sets so
that there exist disjoint open sets G1
2
, H such that

G
0
G1
2
, X \ G
1
H
i.e. G
1
(closed) X \ H G1
2
).
Lemma 5.1 (Urysohns Lemma) Let F
1
, F
2
be disjoint non-empty closed
subsets of a T
4
space; then there exists a continuous function f : X [0, 1]
such that f(F
1
) = {0}, f(F
2
) = {1}.
Proof Given disjoint closed F
1
and F
2
, choose disjoint open G
0
and H
0
so that
F
1
G
0
, F
2
H
0
. Dene G
1
= X\F
2
(open). Since G
0
(closed) X\H
0

X \ F
2
= G
1
, we have

G
0
G
1
.
By the previous remark, we can now construct:
(i) G1
2
T :

G
0
G1
2
,

G1
2
G
1
.
(ii) G1
4
, G3
4
T :

G
0
G1
4
,

G1
4
G1
2
,

G1
2
G3
4
,

G3
4
G
1
.
(iii) . . . and so on!
Thus we get an indexed family of open sets
{G
r
: r =
m
2
n
, 0 m 2
n
, n 1}
48
such that r
1
r
2


G
r
1
G
r
2
.
Observe that the index set is dense in [0, 1]: if s < t in [0, 1], there exists
some
m
2
n
such that s <
m
2
n
< t. Dene
f(x) =
_
inf{r : x G
r
} x F
2
1 x F
2
.
Certainly f : X [0, 1], f(F
2
) = {1}, f(F
1
) = {0}. To show f continuous,
it suces to show that f
1
([0, )) and f
1
((, 1]) are open for 0 < < 1.
Well, f(x) < i there exists some r =
m
2
n
such that f(x) < r < . It
follows that f
1
([0, )) =
r<
G
r
, a union of open sets.
Again, f(x) > i there exist r
1
, r
2
such that < r
1
< r
2
< f(x), implying
that x G
r
2
whence x

G
r
1
. It follows that f
1
((, 1]) =
r
1
>
(X \

G
r
1
),
which is again open.
Corollary 5.4 Every T
4
space is T
3
1
2
.
Proof Immediate from Lemma 5.1. (Note that there exist spaces which are
T
3
1
2
but not T
4
.)
Theorem 5.8 Any compact T
2
space is T
4
.
Proof Use Corollary 5.2 to Theorem 5.3.
Note Unlike the previous axioms, T
4
is neither hereditary nor productive.
The global view of the hierarchy can now be lled in as an exercise from
data supplied above:-
Metrizable Hereditary? Productive?
T
4
T
3
1
2
T
3
T
2
T
1
The following is presented as an indication of how close we are to having
come full circle.
Theorem 5.9 Any completely separable T
4
space is metrizable!
Sketch Proof
Choose a countable base; list as {(G
n
, H
n
) : n 1} those pairs of elements
49
of the base for which

G
n
H
n
. For each n, use Lemma 5.1 to get continuous
f
n
: X [0, 1] such that f
n
(

G
n
) = {0}, f
n
(X \ H
n
) = {1}. Dene
d(x, y) =

n1
{
f
n
(x) f
n
(y)
2
n
}
2
.
One conrms that d is a metric, and induces the original topology.
50