General Topology

Jesper M. Mller
Matematisk Institut, Universitetsparken 5, DK2100 Kbenhavn
E-mail address: [email protected]
URL: http://www.math.ku.dk/
~
moller
Contents
Chapter 0. Introduction 5
Chapter 1. Sets and maps 7
1. Sets, functions and relations 7
2. The integers and the real numbers 11
3. Products and coproducts 13
4. Finite and innite sets 14
5. Countable and uncountable sets 16
6. Well-ordered sets 18
7. Partially ordered sets, The Maximum Principle and Zorns lemma 19
Chapter 2. Topological spaces and continuous maps 23
1. Topological spaces 23
2. Order topologies 25
3. The product topology 25
4. The subspace topology 26
5. Closed sets and limit points 29
6. Continuous functions 32
7. The quotient topology 36
8. Metric topologies 42
9. Connected spaces 45
10. Compact spaces 51
11. Locally compact spaces and the Alexandro compactication 57
Chapter 3. Regular and normal spaces 61
1. Countability Axioms 61
2. Separation Axioms 62
3. Normal spaces 64
4. Second countable regular spaces and the Urysohn metrization theorem 66
5. Completely regular spaces and the Stone

Cech compactication 69
6. Manifolds 71
Chapter 4. Relations between topological spaces 73
Bibliography 75
3
CHAPTER 0
Introduction
These notes are intended as an to introduction general topology. They should be sucient for further
studies in geometry or algebraic topology.
Comments from readers are welcome. Thanks to Micha l Jab lonowski and Antonio Daz Ramos for
pointing out misprinst and errors in earlier versions of these notes.
5
CHAPTER 1
Sets and maps
This chapter is concerned with set theory which is the basis of all mathematics. Maybe it even can
be said that mathematics is the science of sets. We really dont know what a set is but neither do the
biologists know what life is and that doesnt stop them from investigating it.
1. Sets, functions and relations
1.1. Sets. A set is a collection of mathematical objects. We write a S if the set S contains the
object a.
1.1. Example. The natural numbers 1, 2, 3, . . . can be collected to form the set Z
+
= 1, 2, 3, . . ..
This nave form of set theory unfortunately leads to paradoxes. Russels paradox
1
concerns the
formula S , S. First note that it may well happen that a set is a member of itself. The set of all innite
sets is an example. The Russel set
R = S [ S , S
is the set of all sets that are not a member of itself. Is R R or is R , R?
How can we remove this contradiction?
1.2. Definition. The universe of mathematical objects is stratied. Level 0 of the universe consists
of (possibly) some atomic objects. Level i > 0 consists of collections of objects from lower levels. A set is
a mathematical object that is not atomic.
No object of the universe can satisfy S S for atoms do not have elements and a set and an element
from that set can not be in the same level. Thus R consists of everything in the universe. Since the
elements of R occupy all levels of the universe there is no level left for R to be in. Therefore R is outside
the universe, R is not a set. The contradiction has evaporated!
Axiomatic set theory is an attempt to make this precise formulating a theory based on axioms, the
ZFC-axioms, for set theory. (Z stands for Zermelo, F for Fraenkel, and C for Axiom of Choice.) It is not
possible to prove or disprove the statement ZFC is consistent within ZFC that is within mathematics
[12].
If A and B are sets then
A B = x [ x A and x B A B = x [ x A or x B
AB = (x, y) [ x A and y B AHB = (1, a) [ a A (2, b) [ b B
and
AB = x [ x A and x , B
are also sets. These operations satisfy
A (B C) = (A B) (A C) A (B C) = (A B) (A C)
A(B C) = (AB) (AC) A(B C) = (AB) (AC)
as well as several other rules.
We say that A is a subset of B, or B a superset of A, if all elements of A are elements of B. The
sets A and B are equal if A and B have the same elements. In mathematical symbols,
A B x A: x B
A = B (x A: x B and x B: x A) A B and B A
The power set of A,
T(A) = B [ B A
is the set of all subsets of A.
1
If a person says I am lying is he lying?
7
8 1. SETS AND MAPS
1.2. Functions. Functions or maps are fundamental to all of mathematics. So what is a function?
1.3. Definition. A function from A to B is a subset f of A B such that for all a in A there is
exactly one b in B such that (a, b) f.
We write f : A B for the function f A B and think of f as a rule that to any element a A
associates a unique object f(a) B. The set A is the domain of f, the set B is the codomain of f;
dom(f) = A, cod(f) = B.
The function f is
injective or one-to-one if distinct elements of A have distinct images in B,
surjective or onto if all elements in B are images of elements in A,
bijective if both injective and surjective, if any element of B is the image of precisely one element
of A.
In other words, the map f is injective, surjective, bijective i the equation f(a) = b has at most one
solution, at least one solution precisely one solution, for all b B.
If f : A B and g : B C are maps such that cod(f) = dom(g), then the composition is the map
g f : A C dened by g f(a) = g(f(a)).
1.4. Proposition. Let A and B be two sets.
(1) Let f : A B be any map. Then
f is injective f has a left inverse
f is surjective
AC
f has a right inverse
f is bijective f has an inverse
(2) There exists a surjective map A B
AC
There exits an injective map B A
Two of the statements in Proposition 1.4 require the Axiom of Choice (1.27).
Any left inverse is surjective and any right inverse is injective.
If f : A B is bijective then the inverse f
1
: B A is the map that to b B associates the unique
solution to the equation f(a) = b, ie
a = f
1
(b) f(a) = b
for all a A, b B.
Let map(A, B) denote the set of all maps from A to B. Then
map(X, AB) = map(X, A) map(X, B), map(AHB, X) = map(A, X) map(B, X)
for all sets X, A, and B. Some people like to rewrite this as
map(X, AB) = map(X, (A, B)), map(AHB, X) = map((A, B), X)
Here, (A, B) is a pair of spaces and maps (f, g): (X, Y ) (A, B) between pairs of spaces are dened to
be pairs of maps f : X A, g : Y B. The diagonal, X = (X, X), takes a space X to the pair (X, X).
These people say that the product is right adjoint to the diagonal and the coproduct is left adjoint to
the diagonal.
1.5. Relations. There are many types of relations. We shall here concentrate on equivalence rela-
tions and order relations.
1.6. Definition. A relation R on the set A is a subset R AA.
1.7. Example. We may dene a relation D on Z
+
by aDb if a divides b. The relation D Z
+
Z
+
has the properties that aDa for all a and aDb and bDc = aDc for all a, b, c. We say that D is reexive
and transitive.
1.5.1. Equivalence relations. Equality is a typical equivalence relation. Here is the general denition.
1.9. Definition. An equivalence relation on a set A is a relation AA that is
Reexive: a a for all a A
Symmetric: a b = b a for all a, b A
Transitive: a b and b c = a c for all a, b, c A
1. SETS, FUNCTIONS AND RELATIONS 9
The equivalence class containing a A is the subset
[a] = b A [ a b
of all elements of A that are equivalent to a. There is a canonical map [ ] : A A/ onto the set
A/= [a] [ a A T(A)
of equivalence classes that takes the element a A to the equivalence class [a] A/ containing a.
A map f : A B is said to respect the equivalence relation if a
1
a
2
= f(a
1
) = f(a
2
) for
all a
1
, a
2
A (f is constant on each equivalence class). The canonical map [ ] : A A/ respects the
equivalence relation and it is the universal example of such a map: Any map f : A B that respects the
equivalence relation factors uniquely through A/ in the sense that there is a unique map f such that
the diagram
A
f

[ ]

A
A
A
A
A
A
A
B
A/
!f

}
}
}
}
}
}
}
commutes. How would you dene f?
1.10. Example. (1) Equality is an equivalence relation. The equivalence class [a] = a contains
just one element.
(2) a mod b mod n is an equivalence relation on Z. The equivalence class [a] = a + nZ consists of all
integers congruent to a modn and the set of equivalence classes is Z/nZ = [0], [1], . . . , [n 1].
(3) x y
def
[x[ = [y[ is an equivalence relation in the plane R
2
. The equivalence class [x] is a circle
centered at the origin and R
2
/ is the collection of all circles centered at the origin. The canonical map
R
2
R
2
/ takes a point to the circle on which it lies.
(4) If f : A B is any function, a
1
a
2
def
f(a
1
) = f(a
2
) is an equivalence relation on A. The
equivalence class [a] = f
1
(f(a)) A is the bre over f(a) B. we write A/f for the set of equivalence
classes. The canonical map A A/f takes a point to the bre in which it lies. Any map f : A B can
be factored
A
f

[ ]

B
B
B
B
B
B
B
B
B
A/f
f

{
{
{
{
{
{
{
as the composition of a surjection followed by an injection. The corestriction f : A/f f(A) of f is a
bijection between the set of bres A/f and the image f(A).
(5) [Ex 3.2] (Restriction) Let X be a set and A X a subset. Declare any two elements of A to be
equivalent and any element outside A to be equivalent only to itself. This is an equivalence relation. The
equivalence classes are A and x for x X A. One writes X/A for the set of equivalence classes.
(6) [Ex 3.5] (Equivalence relation generated by a relation) The intersection of any family of equivalence
relations is an equivalence relation. The intersection of all equivalence relations containing a given relation
R is called the equivalence relation generated by R.
1.11. Lemma. Let be an equivalence relation on a set A. Then
(1) a [a]
(2) [a] = [b] a b
(3) If [a] [b] ,= then [a] = [b]
Proof. (1) is reexivity, (2) is symmetry, (3) is transitivity: If c [a] [b], then a c b so a b
and [a] = [b] by (2).
This lemma implies that the set A/ T(A) is a partition of A, a set of nonempty, disjoint subsets
of A whose union is all of A. Conversely, given any partition of A we dene an equivalence relation by
declaring a and b to be equivalent if they lie in the same subset of the partition. We conclude that an
equivalence relation is essentially the same thing as a partition.
10 1. SETS AND MAPS
1.5.2. Linear Orders. The usual order relation < on Z or R is an example of a linear order. Here is
the general denition.
1.12. Definition. A linear order on the set A is a relation < AA that is
Comparable: If a ,= b then a < b or b < a for all a, b A
Nonreexive: a < a for no a A
Transitive: a < b and b < c = a < c for all a, b, c A
What are the right maps between ordered sets?
1.13. Definition. Let (A, <) and (B, <) be linearly ordered sets. An order preserving map is a map
f : A B such that a
1
< a
2
= f(a
1
) < f(a
2
) for all a
1
, a
2
A. An order isomorphism is a bijective
order preserving map.
An order preserving map f : A B is always injective. If there exists an order isomorphismf : A B,
then we say that (A, <) and (B, <) have the same order type.
How can we make new ordered sets out of old ordered sets? Well, any subset of a linearly ordered set
is a linearly ordered set in the obvious way using the restriction of the order relation. Also the product
of two linearly ordered set is a linearly ordered set.
1.14. Definition. Let (A, <) and (B, <) be linearly ordered sets. The dictionary order on AB is
the linear order given by
(a
1
, b
1
) < (a
2
, b
2
)
def
(a
1
< a
2
) or (a
1
= a
2
and b
1
< b
2
)
The restriction of a dictionary order to a product subspace is the dictionary order of the restricted
linear orders. (Hey, what did that sentence mean?)
What about orders on AHB, A B, map(A, B) or T(A)?
What are the invariant properties of ordered sets? In a linearly ordered set (A, <) it makes sense to
dene intervals such as
(a, b) = x A [ a < x < b, (, b] = x A [ x b
and similarly for other types of intervals, [a, b], (a, b], (, b] etc.
If (a, b) = then a is the immediate predecessor of b, and b the immediate successor of a.
Let (A, <) be an ordered set and B A a subset.
M is a largest element of B if M B and b M for all b B. The element m is a smallest
element of B if m B and m b for all b B. We denote the largest element (if it exists) by
max B and the smallest element (if it exists) by min B.
M is an upper bound for B if M A and b M for all b B. The element m is a lower bound
for B if m A and m b for all b B. The set of upper bounds is

bB
[b, ) and the set of
lower bounds is

bB
(, b].
If the set of upper bounds has a smallest element, min

bB
[b, ), it is called the least upper
bound for B and denoted sup B. If the set of lower bounds has a largest element, max

bB
(, b],
it is called the greatest lower bound for B and denoted inf B.
1.15. Definition. An ordered set (A, <) has the least upper bound property if any nonempty subset
of A that has an upper bound has a least upper bound. If also (x, y) ,= for all x < y, then (A, <) is a
linear continuum.
1.16. Example. (1) R and (0, 1) have the same order type. [0, 1) and (0, 1) have distinct order
types for [0, 1) has a smallest element and (0, 1) doesnt. 1 (0, 1) and [0, 1) have the same order
type as we all can nd an explicit order isomorphism between them.
(2) R R has a linear dictionary order. What are the intervals (1 2, 1 3), [1 2, 3 2] and
(1 2, 3 4]? Is RR a linear continuum? Is [0, 1] [0, 1]?
(3) We now consider two subsets of RR. The dictionary order on Z
+
[0, 1) has the same order type
as [1, ) so it is a linear continuum. In the dictionary order on [0, 1) Z
+
each element (a, n) has
(a, n+1) as its immediate successor so it is not a linear continuum. Thus Z
+
[0, 1) and [0, 1)Z
+
do
not have the same order type. (So, in general, (A, <) (B, <) and (B, <) (A, <) represent dierent
order types. This is no surprise since the dictionary order is not symmetric in the two variables.)
(4) (R, <) is a linear continuum as we all learn in kindergarten. The sub-ordered set (Z
+
, <) has the
least upper bound property but it is not a linear continuum as (1, 2) = .
2. THE INTEGERS AND THE REAL NUMBERS 11
(5) (1, 1) has the least upper bound property: Let B be any bounded from above subset of (1, 1)
and let M (1, 1) be an upper bound. Then B is also bounded from above in R, of course, so there
is a least upper bound, sup B, in R. Now sup B is the smallest upper bound so that sup B M < 1.
We conclude that sup B lies in (1, 1) and so it is also a least upper bound in (1, 1). In fact, any
convex subset of a linear continuum is a linear continuum.
(6) R 0 does not have the least upper bound property as the subset B = 1,
1
2
,
1
3
, . . . is
bounded from above (by say 100) but the set of upper bounds (0, ) has no smallest element.
2. The integers and the real numbers
We shall assume that the real numbers R exists with all the usual properties: (R, +, ) is a eld,
(R, +, , <) is an ordered eld, (R, <) is a linear continuum (1.15).
What about Z
+
?
1.17. Definition. A subset A R is inductive if 1 A and a A = a + 1 A.
There are inductive subsets of R, for instance R itself and [1, ).
1.18. Definition. Z
+
is the intersection of all inductive subsets of R.
We have that 1 Z
+
and Z
+
[1, ) because [1, ) is inductive so 1 = min Z
+
is the smallest
element of Z
+
.
1.19. Theorem. (Induction Principle) Let J be a subset of Z
+
such that
1 J and n Z
+
: n J = n + 1 J
Then J = Z
+
.
Proof. J is inductive so J contains the smallest inductive set, Z
+
.
1.20. Theorem. Any nonempty subset of Z
+
has a smallest element.
Before the proof, we need a lemma.
For each n Z
+
, write
S
n
= x Z
+
[ x < n
for the set of positive integers smaller than n (the section below n). Note that S
1
= and S
n+1
= S
n
n.
1.21. Lemma. For any n Z
+
, any nonempty subset of S
n
has a smallest element.
Proof. Let J Z
+
be the set of integers for which the lemma is true. It is enough (1.19) to show
that J is inductive. 1 J for the trivial reason that there are no nonempty subsets of S
1
= . Suppose
that n J. Consider a nonempty subset A of S
n+1
. If A consists of n alone, then n = min A is the
smallest element of A. If not, A contains integers < n, and then min(A S
n
) is the smallest element of
A. Thus n + 1 J.
Proof of Theorem 1.20. Let A Z
+
be any nonempty subset. The intersection A S
n
is
nonempty for some n, so it has a smallest element (1.21). This is also the smallest element of A.
1.22. Theorem (General Induction Principle). Let J be a subset of Z
+
such that
n Z
+
: S
n
J = n J
Then J = Z
+
.
Proof. We show the contrapositive. Let J be a proper subset of Z
+
. Consider the smallest element
n = min(Z
+
J) outside J. Then n , J and S
n
J (for n is the smallest element not in J meaning
that all elements smaller than n are in J). Thus J does not satisfy the hypothesis of the theorem.
1.23. Theorem (Archimedean Principle). Z
+
has no upper bound in R: For any real number there
is a natural number which is greater.
Proof. We assume the opposite and derive a contradiction. Suppose that Z
+
is bounded from
above. Let b = sup Z
+
be the least upper bound (R has the least upper bound property). Since b 1 is
not an upper bound (it is smaller than the least upper bound), there is a positive integer n Z
+
such
that n > b 1. Then n + 1 is also an integer (Z
+
is inductive) and n + 1 > b. This contradicts that b is
an upper bound for Z
+
.
12 1. SETS AND MAPS
1.24. Theorem (Principle of Recursive Denitions). For any set B and any function
:

map(S
n
, B) [ n Z
+
B
there exists a unique function h: Z
+
B such that h(n) = (h[S
n
) for all n Z
+
.
Proof. See [8, Ex 8.8].
This follows from the Induction Principle, but we shall not go into details. It is usually considered
bad taste to dene h in terms of h but the Principle of Recursive Denition is a permit to do exactly
that in certain situations. Here is an example of a recursive denition from computer programing
fibo := func< n | n le 2 select 1 else Self(n-1) + Self(n-2) >;
of the Fibonacci function. Mathematicians (sometimes) prefer instead to apply the Principle of Recursive
Denitions to the map
(S
n
f
Z
+
) =
_
1 n < 2
f(n 1) +f(n 2) n > 2
Recursive functions can be computed by Turing machines.
3. PRODUCTS AND COPRODUCTS 13

jJ
A
j
J

jJ
A
j
-
J

jJ
A
j
6
Figure 1. The coproduct
3. Products and coproducts
1.25. Definition. An indexed family of sets consists of a set / of sets, an index set J, and a
surjective function f : J /.
We often denote the set f(j) by A
j
and the whole indexed family by A
j

jJ
. Any set / of sets can
be made into an indexed family of sets by using the identity map / / as the indexing function.
We dene the union, the intersection, the product, and the coproduct of the indexed family as

jj
A
j
= a [ a A
j
for all j J,
_
jj
A
j
= a [ a A
j
for at least one j J

jJ
A
j
= x map(J,
_
A
j
) [ j J : x(j) A
j

jJ
A
j
=
_
jJ
(j, a) J
_
jJ
A
j
[ a A
j

There are natural maps

A
j
A
j
j

J
Q
A
j

j
:

jJ
A
j
A
j
(projection)
j
: A
j

jJ
A
j
(injection)
given by
j
(x) = x(j) and
j
(a) = (j, a) for all j J. These maps are used in establishing the identities
map(X,

jJ
A
j
) =

jJ
map(X, A
j
), map(

jJ
A
j
, Y ) =

jJ
map(A
j
, Y )
for any sets X and Y . This gives in particular maps
:

jJ
A
j

jJ
A
j
(diagonal), :

jJ
A
j

_
jJ
A
j
(codiagonal)
If the index set J = S
n+1
= 1, . . . , n then we also write
A
1
A
n
, A
1
A
n
, A
1
A
n
A
1
H HA
n
for

jS
n+1
A
j
,

jS
n+1
A
j
,

jS
n+1
A
j

jS
n+1
A
j
, respectively. If also and A
j
= A for all j S
n+1
we write A
n
for the product

jS
n+1
A. The elements of A
n
are all n-tuples (a
1
, . . . , a
n
) of elements
from A.
If the index set J = Z
+
then we also write
A
1
A
n
, A
1
A
n
, A
1
A
n
A
1
H HA
n

for

jZ
+
A
j
,

jZ
+
A
j
,

jZ
+
A
j
,

jZ
+
A
j
, respectively. If also A
j
= A for all j we write A

for the
product

jZ
+
A, the set of all functions x: Z
+
A, i.e. all sequences (x
1
, . . . , x
n
, . . .) of elements from
A.
2
1.26. Example. (1) S
1
S
2
S
n
=

nZ
+
S
n
= Z
+
.
(2) If the set / = A consists of just one set A then

jJ
A = A =

jJ
A,

jJ
A = map(J, A),
and

jJ
A = J A.
(3) There is a bijection (which one?) between 0, 1

= map(Z
+
, 0, 1) and T(Z
+
). More generally,
there is a bijection (which one) between the product

jJ
0, 1 = map(J, 0, 1) and the power set
T(J).
Even though we shall not specify our (ZF) axioms for set theory, let us mention just one axiom which
has a kind of contended status since some of its consequences are counter-intuitive.

`
_

_
2
is the formal set within set theory corresponding to the nave set Z
+
[12, V.1.5]
14 1. SETS AND MAPS
1.27. Axiom (Axiom of Choice (AC)). For any nonempty set of nonempty disjoint sets, /, there
exists a set C

AA
A such that C A contains exactly one element for all A /.
If the ZF axioms of set theory are consistent, then both ZF+AC (Godel 1938) and ZF+AC (Fraenkel
and Mostowsski, Cohen) are consistent theories [12, IV.2.8]. You may take or leave AC without penalty.
(Just like you may take or leave Euclids axiom on parallels depending on what kind of geometry you like
to do.) We shall here include AC and work within ZFC (ZF + AC).
Unlike the other axioms of set theory, the AC does not determine the set C uniquely.
1.28. Theorem. [3, Thm B.18] The following statements are equivalent:
(1) The Axiom of Choice
(2) Any surjective map has a right inverse.
(3) For any nonempty indexed family of (not necessarily disjoint) nonempty sets, A
j

jJ
, there
exists a function c: J

jJ
A
j
(a choice function) such that c(j) A
j
for all j J.
(4)

jJ
A
j
,= for any nonempty indexed family of nonempty sets.
Proof. (1) = (2): Let f : A B be a surjective map. Dene the right inverse g : B A by g(b) =
C f
1
(b) where C A =

bB
f
1
(b) is a set such that C f
1
(b) contains exactly one point for each
b B.
(2) = (3): Dene c to be J

A
j

A
j
where the rst map is a right inverse to the function

jJ
A
j
J taking A
j
to j for all j J.
(3) (4): By denition, the product is the set of choice functions.
(3) = (1): Let / be a nonempty set of nonempty sets. Put C = c(/) where c: /

AA
A is a choice
function.
c: J

jJ
A
j
-
J

jJ
A
j
6

Figure 2. A choice function

Here is a special, but often used, case. Let A be any nonempty set and T

(A) = T(A) the set of

nonempty subsets of A. Then there exists (1.28.(3)) a choice function c: T

(A) A such that c(B) B

for any nonempty B A. (The choice function selects an element in each nonempty subset of A.) If A
is the set R of real numbers, then the Axiom of Choice is a permit to work with a function that to every
nonempty subset of R associates an element of that subset. How would you write down such a function?
4. Finite and innite sets
1.29. Definition. A set A is nite if S
n+1
A for some n Z
+
. A set is innite if it is not nite.
We write X Y if there is a bijection between the two sets X and Y .
1.30. Lemma. Let n Z
+
and let B be a proper subset of S
n+1
.
(1) It is impossible to map B onto S
n+1
.
(2) S
m+1
B for some m < n.
Proof. Both statements are proved by induction.
(1) If n = 1, then S
2
= 1 and B = so the assertion is true in this case. Assume it is true for some
n Z
+
. Consider a proper subset B of S
n+1+1
. Suppose that there exists a surjection f : B S
n+1+1
.
By permuting the elements of B and S
n+1+1
if necessary, we can assume that f(n + 1) = n + 1. Then
B f
1
(n + 1) is a proper subset of S
n+1
which is mapped onto S
n+1
by f. But that is impossible by
induction hypothesis.
(2) If n = 1, then S
2
= 1 and B = so S
1
B. Assume the assertion is true for some n Z
+
.
Consider a proper subset B of S
n+1+1
. By permuting the elements of S
n+1+1
if necessary, we can assume
that n + 1 , B so that B is a subset of S
n+1
. If B = S
n+1
, then B S
n+1
, of course. Otherwise, B is a
proper subset of S
n+1
and then S
m+1
B for some m < n < n + 1 by induction hypothesis.
4. FINITE AND INFINITE SETS 15
1.31. Corollary. Let A be a nite set.
(1) If S
m+1
A S
n+1
, then m = n.
(2) Any subset of A is nite.
Proof. (1) Suppose that m ,= n. We may then assume that m < n. But then S
m+1
is a proper
subset of S
n+1
which can be mapped onto S
n+1
. That is not possible.
(2) Since this is true for the model nite set S
n+1
, it is true for all nite sets.
We have just learned that if A is nite then S
n+1
A for a unique n Z
+
. This n is called the
cardinality of A and it is denoted cardA or [A[. We also learned that if B A then B is nite, [B[ [A[,
and
(1.32) [B[ < [A[ B A
which is often called the pidgeon-hole principle.
1.33. Theorem (Characterization of nite sets). Let A be a set. The following statements are
equivalent
(1) A is nite
(2) There exists a surjection S
n+1
A for some n Z
+
(3) There exists an injection A S
n+1
for some n Z
+
Proof. (1) = (2): There even exists a bijection S
n+1
A.
(2) (3): 1.4.(2)
(3) = (1): If there exists an injection A S
n+1
, then there exists a bijection between A and a subset
of S
n+1
. But we have just seen that all subsets nite sets are nite.
1.34. Corollary (Hereditary properties of nite sets).
(1) Subsets of nite sets are nite.
(2) Images of nite sets are nite.
(3) Finite unions of nite sets are nite.
(4) Finite Cartesian products of nite sets are nite.
Proof. (1) Proved already in 1.31.
(2) S
n+1
A B.
(3) To see that the union of two nite sets is nite, it is enough to show S
m+1
H S
n+1
is nite (for the
union of any two nite sets is an image of this set). But it is immediate that S
m+n+1
S
m+1
H S
n+1
.
Induction now shows that A
1
A
n
is nite when A
1
, . . . , A
n
are nitely many nite sets.
(4) Let A and B be nite. Since A B =

aA
B is the union of nitely many nite sets, it is nite.
Induction now shows that A
1
A
n
is nite when A
1
, . . . , A
n
are nitely many nite sets.
Are all sets nite? No!
1.35. Corollary. Z
+
is innite.
Proof. There is a surjective map of the proper subset Z
+
1 onto Z
+
.
1.36. Theorem (Characterization of innite sets). Let A be a set. The following are equivalent:
(1) A is innite
(2) There exists an injective map Z
+
A
(3) There exists a surjective map A Z
+
(4) A is in bijection with a proper subset of itself
Proof. (1) = (2): Let c: T

(A) A be a choice function. Dene h: Z

+
A recursively by
h(1) = c(A)
h(i) = c(Ah(1), . . . , h(i 1)), i > 1
Then h is injective (if i < j then h(j) Ah(1), . . . , h(i), . . . , h(j 1) so h(i) ,= h(j)).
(2) (3): 1.4.(2)
(2) = (4): We view Z
+
as a subset of A. Then A = (A Z
+
) Z
+
is in bijection with the proper
subset A1 = (AZ
+
) (Z
+
1).
(4) = (1): This is 1.30.
Here we applied the Principle of Recursive Denitions (1.24) to (S
n
f
Z
+
) = c(Af(S
n
)).
16 1. SETS AND MAPS
5. Countable and uncountable sets
1.37. Definition. A set C is countably innite if Z
+
C. It is countable if it is nite or countably
innite. It is uncountable if it is not countable.
1.38. Lemma. Any subset of Z
+
is either nite or countably innite (in bijection with Z
+
).
Proof. Let C Z
+
be an innite set of positive integers. We show (the stronger statement) that
C has the order type of Z
+
. Dene a function h: Z
+
C recursively (1.24) by
h(1) = min C
h(i) = min(C h(1), . . . , h(i 1)), i > 1
using 1.20. Note that C h(1), . . . , h(i 1) is nonempty since C is innite (1.33). We claim that h is
bijective.
h is order preserving: If i < j, then
h(i) = min(C h(1), . . . , h(i 1)) < min(C h(1), . . . , h(i 1), . . . , h(j 1)) = h(j)
because C h(1), . . . , h(i 1) C h(1), . . . , h(i 1), . . . , h(j 1).
h is surjective: Let c C. We must nd a positive integer m such that c = h(m). Our only hope is
m = minn Z
+
[ h(n) c
(Note that this has a meaning since the set n Z
+
[ h(n) c is nonempty as we can not inject the
innite set Z
+
into the nite set 1, . . . , c 1 = S
c
(1.31). Note also that again we use 1.20.) By
denition of m,
h(m) c and h(n) c n m
The last of these two properties is equivalent to n < m h(n) < c, so c , h(1), . . . , h(m 1), or
c C h(1), . . . , h(m1), and therefore
h(m) = min(C h(1), . . . , h(m1)) c
by denition of h. Thus h(m) = c.
Here we applied the Principle of Recursive Denitions (1.24) to (S
n
f
C) = min(C f(S
n
)).
1.39. Theorem (Characterization of countable sets). Let A be a set. The following statements are
equivalent
(1) A is countable
(2) There exists a surjection Z
+
A
(3) There exists an injection A Z
+
Proof. If A is nite, the theorem is true, so we assume that A is countably innite.
(1) = (2): Clear.
(2) (3): 1.4.(2)
(3) = (1): We may as well assume that A Z
+
. Since A is assumed to innite, A is countably innite
by Lemma 1.38.
1.40. Example. Z
+
is obviously innitely countable. The map f : Z
+
Z
+
Z
+
given by f(m, n) =
2
m
3
n
is injective by uniqueness of prime factorizations. The map g : Z
+
Z
+
Q
+
given by g(m, n) =
m
n
, is surjective. Thus Z
+
Z
+
and Q
+
are innitely countable.
1.41. Corollary (Hereditary properties of countable sets).
(1) A subset of a countable set is countable
(2) The image of a countable set is countable.
(3) A countable union of countable sets is countable (assuming AC).
3
(4) A nite product of countable sets is countable.
Proof. (1) B A Z
+
.
(2) Z
+
A B.
(3) Let A
j

jJ
be an indexed family of sets where J is countable and each set A
j
is countable. It is
enough to show that

A
j
is countable. We leave the case where the index set J is nite as an exercise
3
In set theory without AC, R is a countable union of countable sets [12, p 228]
5. COUNTABLE AND UNCOUNTABLE SETS 17
and consider only the case where J is innite. Then we may as well assume that J = Z
+
. Choose (!) for
each n Z
+
an injective map f
n
: A
n
Z
+
. Then we have injective maps

A
n

f
n

Z
+
= Z
+
Z
+
(1.40)
Z
+
so

A
n
is countable.
(4) If A and B are countable, so is A B =

aA
B as we have just seen. Now use induction to show
that if A
1
, . . . , A
n
are countable, so is A
1
A
n
.
You may think that a countable product of countable sets is countable or indeed that all sets are
nite or countable but thats false.
1.42. Theorem. Let A be any set.
(1) There is no injective map T(A) A
(2) There is no surjective map A T(A)
Proof. (Cantors diagonal argument.) It is a general fact (1.4.(2)) that (1) (2). Thus it
suces to show (2). Let g : A T(A) be any function. Then
a A [ a , g(a) T(A)
is not in the image of g. Because if this set were of the form g(b) for some b A, then wed have
b g(b) b , g(b)

1.43. Corollary. The set T(Z
+
) = map(Z
+
, 0, 1) =

nZ
+
0, 1 = 0, 1

is uncountable.
Russels paradox also exploits Cantors diagonal argument.
We have seen (1.38) that any subset of Z
+
is either nite or in bijection with Z
+
. What about
subsets of R?
1.44. Conjecture (Cantors Continuum Hypothesis, CH). Any subset of R is either countable or
in bijection with R.
CH is independent of the ZFC axioms for set theory in that if ZFC is consistent then both ZFC+CH
(Godel 1950) and ZFC+CH (Cohen 1963) are consistent theories [12, VII.4.26] [4]. Our axioms are not
adequate to settle the CH.
Look up the generalized continuum hypothesis (GCH) [Ex 11.8] (due to Hausdor) somewhere [15,
16]. It is not customary to assume the GCH; if you do, the AC becomes a theorem.
18 1. SETS AND MAPS
6. Well-ordered sets
We have seen that all nonempty subsets of (Z
+
, <) have a smallest element and we have used this
property in quite a few places so there is reason to suspect that this is an important property in general.
This is the reason for the following denition. You may think of well-ordered sets as some kind of
generalized versions of Z
+
.
1.45. Definition. A set A with a linear order < is well-ordered if any nonempty subset has a smallest
element.
Any well-ordered set has a smallest element. Any element a A (but the largest element, if there is
one) in a well-ordered set has an immediate successor a
+
, the smallest successor. (And any element (but
the smallest) has an immediate predecessor?) A well-ordered set can not contain an innite descending
chain x
1
> x
2
> , in fact, a linearly ordered set is well-ordered if and only if it does not contain a
copy of the negative integers Z

[Ex 10.4].
Let (A, <) be a well-ordered set and an element of A. The subset of predecessors of ,
S

(A) = S

= (, ) = a A [ a <
is called the section of A by .
The induction principle and the principle of recursive denitions apply not only to Z
+
but to any
well-ordered set.
1.46. Theorem (Principle of Transnite Induction). (Cf 1.22) Let (A, <) be a well-ordered set and
J A a subset such that
A: S

J = J
Then J = A.
Proof. Formally identical to the proof of 1.22.
1.47. Theorem (Principle of Transnite Recursive Denitions). Let (A, <) be a well-ordered set. For
any set B and any function
:

map(S

, B) [ A B
there exists a unique function h: A B such that h() = (h[S

) for all A.
1.48. Proposition (Hereditary properties of well-ordered sets).
(1) A subset of a well-ordered set is well-ordered.
(2) The coproduct of any well-ordered family of well-ordered sets is well-ordered [Ex 10.8].
(3) The product of any nite family of well-ordered sets is well-ordered.
Proof. (1) Clear.
(2) Let J be a well-ordered set and A
j

jJ
a family of well-ordered sets indexed by J. For i, j J
and x A
i
, y A
j
, dene
(i, x) < (j, y)
def
i < j or (i = j and x < y)
and convince yourself that this is a well-ordering.
(3) If (A, <) and (B, <) well-ordered then A B =

aA
B is well-ordered. Now use induction to
show that the product A
1
A
n
of nitely many well-ordered sets A
1
, . . . , A
n
is well-ordered.

If C a nonempty subset of A B then min C = (c

1
, min
2
(C
1
1
(c
1
))) where c
1
= min
1
(C) is
the smallest element of C.
1.49. Example. (1) The positive integers (Z
+
, <) is a well-ordered set.
(2) Z and R are not well-ordered sets.
(3) Any section S
n+1
= 1, 2, . . . , n of Z
+
is well-ordered (1.48.(1)).
(4) The product S
n+1
Z
+
is well-ordered (1.48.(3)). The nite products Z
n
+
= Z
+
Z
+
Z
+
are
well-ordered (1.48.(3)).
(5) The innite product 0, 1

is not well-ordered for it contains the innite descending chain (1, 0, 0, 0, . . .) >
(0, 1, 0, 0, . . .) > (0, 0, 1, 0, . . .) > .
7. PARTIALLY ORDERED SETS, THE MAXIMUM PRINCIPLE AND ZORNS LEMMA 19
(6) The set S

= [1, ] = Z
+
H is well-ordered (1.48.(2)). It has as its largest element. The
section S

= [1, ) = Z
+
is countably innite but any other section is nite. Any nite subset A of
[1, ) has an upper bound because the set of non-upper bounds
x [1, ) [ a A: x < a =
_
aA
S
a
is nite (1.34.(3)) but [1, ) is innite. S

has the same order type as the interval [1 1, 2 1] in

Z
+
Z
+
.
Which of these well-ordereds have the same order type [Ex 10.3]? Draw pictures of examples of
well-ordered sets.
We can classify completely all nite well-ordered sets.
1.50. Theorem (Finite order types). [Ex 6.4] Any nite linearly ordered set A of cardinality n has
the order type of (S
n+1
, <); in particular, it is well-ordered and it has a largest element.
Proof. Dene h: S
n+1
A recursively by h(1) = min A and
h(i) = min(Ah(1), . . . , h(i 1), i > 1
Then h is order preserving. In particular, h is injective and hence bijective (by the pigeon hole principle
(1.32)) since the two sets have the same cardinality.
Can you nd an explicit order preserving bijection S
m+1
S
n+1
S
m+n+1
?
So there is just one order type of a given nite cardinality n. There are many countably innite
well-ordered sets (1.49). Is there an uncountable well-ordered set? Our examples, R and 0, 1

, of
uncountable sets are not well-ordered (1.49.(2), 1.49.(5)).
1.51. Theorem (Well-ordering theorem). (Zermelo 1904) Any set can be well-ordered.
We focus on the minimal criminal, the minimal uncountable well-ordered set. (It may help to look at
1.49.(6) again.)
1.52. Lemma. There exists a well-ordered set S

= [0, ] with a smallest element, 0, and a largest

element, , such that:
(1) The section S

= [0, ) by is uncountable but any other section, S

= [0, ) for < , is

countable.
(2) Any countable subset of S

= [0, ) has an upper bound in S

= [0, )
Proof. (Cf 1.49.(6)) Take any uncountable well-ordered set A (1.51). Append a greatest element to
A. Call the result A again. Now A has at least one uncountable section. Let be the smallest element
of A such that the section by this element is uncountable, that is = min A [ S

is uncountable.
Put S

= [0, ] where 0 is the smallest element of A. This well-ordered set satises (1) and (2). Let C
be a countable subset of S

= [0, ). We want to show that it has an upper bound. We consider the set
of elements of S

that are not upper bounds, i.e

x S

[ x is not an upper bound for C = x S

[ c C: x < c =
countable
..
_
cC
S
c

uncountable
..
S

This set of not upper bounds is countable for it is a countable union of countable sets (1.41.(4)). But S

is uncountable, so the set of not upper bounds is a proper subset.

See [SupplExI : 8] for an explicit construction of a well-ordered uncountable set. Z
+
is the well-
ordered set of all nite (nonzero) order types and S

is the well-ordered set of all countable (nonzero)

order types. See [Ex 10.6] for further properties of S

.
Recall that the ordered set Z
+
[0, 1) is a linear continuum of the same order type as [1, ) R.
What happens if we replace Z
+
by S

[Ex 24.6, 24.12]?

7. Partially ordered sets, The Maximum Principle and Zorns lemma
If we do not insist on comparability in our order relation (Denition 1.12) we obtain a partial order:
1.53. Definition. A strict partial order on a set A is a non-reexive and transitive relation AA:
(1) a a holds for no a A
(2) a b and b c implies a c.
20 1. SETS AND MAPS
A partially ordered set (a poset, for short) is a set with a strict partial order. We do not require
that any two elements can be compared. For instance, power sets are strictly partially ordered by strict
inclusion (in fact, it is good idea to read as is contained in). Here is an example of a nite poset:
A11
A12 A21
A22
A23
A24
A31
A32
A33
A34
A41
A42
A43
A44
A51
A52

O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O

o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o

o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o

O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
























o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o

O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O

o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o

O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O

O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O

o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o

O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O

?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?

?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?

O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O

o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o























O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O

o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o

o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o

O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
1.54. Definition. Let (A, ) be a set with a strict partial order and m an element of A.
m is maximal if it has no successors, a A: m _ a = m = a
m is an upper bound for B A if all elements of B precedes or equals m, b B: b _ m.
In the above example, the elements A51 and A52 are maximal because they have no successors. The
element A51, but not the element A452, is an upper bound for the set A33, A42
1.55. Theorem (Hausdors Maximum Principle). Any linearly ordered subset of a poset is contained
in a maximal linearly ordered subset.
Proof. We shall only consider the case where the linearly ordered subset is empty where the state-
ment is that any poset contains maximal linearly ordered subsets. As a special case, suppose that the
poset is innitely countable. We may as well assume that the poset is Z
+
with some partial order .
Dene h: Z
+
0, 1 recursively by h(1) = 0 and
h(i) =
_
0 j < i[h(j) = 0 i is linearly ordered wrt
1 otherwise
for i > 0. Then H = h
1
(0) is a maximal linearly ordered subset.
For the proof in the general case, let (A, <) be a poset, well-order A (1.51), apply the Principle of
Transnite Recursion (1.47) to the function
(S

f
0, 1) =
_
0 f
1
(0) is a linearly ordered subset of A
1 otherwise
and get a function h: A 0, 1.
1.56. Theorem (Zorns lemma). Let (A, <) be a poset. Suppose that all linearly ordered subsets of
A have upper bounds. Then any linearly ordered subset is bounded from above by a maximal element. In
particular, for any element a of A there is a maximal element m of A such that a _ m.
Proof. Let H A be a maximal linearly ordered subset. According to Hausdors Maximum
Principle, we may assume that H is maximal. By hypothesis, H has an upper bound m, i.e. x _ m for
all x H. By maximality of H, m must be in H, and there can be no element in A greater than m.
(Suppose that m d for some d. Then x _ m d for all elements of H so H d is linearly ordered,
contradicting maximality of H.)
We shall later use Zorns lemma to prove Tychonos theorem (2.149) that a product of compact
spaces is compact. In fact, The Axiom of Choice, Zermelos Well-ordering theorem, Hausdors maximum
principle, Zorns lemma, and Tychonos theorem are equivalent.
Here are two typical applications. Recall that a basis for a vector space over a eld is a maximal
independent subset and that a maximal ideal in a ring is a maximal proper ideal.
1.57. Theorem. Any linearly independent subset of a vector space is contained in a basis.
Proof. Apply Zorns lemma to the poset of independent subsets of the vector space. Any linearly
ordered set of independent subsets has an upper bound, namely its union.
1.58. Theorem. Any proper ideal of a ring is contained in a maximal ideal.
7. PARTIALLY ORDERED SETS, THE MAXIMUM PRINCIPLE AND ZORNS LEMMA 21
Proof. Apply Zorns lemma to the poset of proper ideals. Any linearly ordered set of proper ideals
has an upper bound, namely its union.
As a corollary of (1.57) we see that R and R
2
are isomorphic as vector spaces over Q.
Other authors prefer to work with partial orders instead of strict partial orders.
1.59. Definition. Let A be a set. A relation _ on A is said to be a partial order precisely when
it is symmetric (that is a _ a for all a in A), transitive (that is a _ b and b _ c implies a _ c), and
anti-symmetric (that is a _ b and b _ a implies a = b).
> SubgroupLattice(AlternatingGroup(5));
Partially ordered set of subgroup classes
----------------------------------------------
[1] Order 1 Length 1 Maximal Subgroups:
---
[2] Order 2 Length 15 Maximal Subgroups: 1
[3] Order 3 Length 10 Maximal Subgroups: 1
[4] Order 5 Length 6 Maximal Subgroups: 1
---
[5] Order 4 Length 5 Maximal Subgroups: 2
[6] Order 6 Length 10 Maximal Subgroups: 2 3
[7] Order 10 Length 6 Maximal Subgroups: 2 4
---
[8] Order 12 Length 5 Maximal Subgroups: 3 5
---
[9] Order 60 Length 1 Maximal Subgroups: 6 7 8
Table 1. The poset of subgroups of the alternating group A
5
CHAPTER 2
Topological spaces and continuous maps
1. Topological spaces
What does it mean that a map f : X Y between two sets is continuous? To answer this question,
and much more, we equip sets with topologies.
2.1. Definition. Let X be a set. A topology on X is a set T T(X) of subsets of X, called open
sets, such that
(1) and X are open
(2) The intersection of nitely many open sets is open
(3) Any union of open sets is open
A topological space is a set X together with a topology T on X.
2.2. Example (Examples of topologies). (1) In the trivial topology T = , X, only two subsets
are open.
(2) In the discrete topology T = T(X), all subsets are open.
(3) In the particular point topology, the open sets are , X and all subsets containing a particular point
x X. For instance, the Sierpinski space is the set X = 0, 1 with the particular point topology for
the point 0. The open sets are T = , 0, X.
(4) In the nite complement topology (or conite topology), the open sets are and X and all subsets
with a nite complement.
(5) The standard topology on the real line R is T = unions of open intervals.
(6) More generally, suppose that (X, d) is a metric space. The open r-ball centered at x X is the
set B
d
(x, r) = y X [ d(x, y) < r of points within distance r > 0 from x. The metric topology on
X is the collection T
d
= unions of open balls. The open sets of the topological space (X, T
d
) and
the open sets in the metric space (X, d) are the same. (See .8 for more on metric topologies.)
The Sierpinski topology and the nite complement topology on an innite set are not metric topolo-
gies.
Topologies on X are partially ordered by inclusion. For instance, the nite complement topology
(2.2.(4)) on R is contained in the standard topology (2.2.(5), and the indiscrete topology (2.2.(1)) on
0, 1 is contained in the Sierpinski topology (2.2.(3)) is contained in the discrete topology (2.2.(2)).
2.3. Definition (Comparison of topologies). Let T and T

be two topologies on X.
T is ner than T

is coarser than T
_
def
T T

The nest topology is the topology with the most opens sets, the coarsest topology is the one with
fewest open sets. (Think of sandpaper!) The discrete topology is ner and the indiscrete topology coarser
than any other topology: T(X) T , X. Of course, two topologies may also be incomparable.
2.4. Definition. A neighborhood of a point x X is an open set containing x. A neighborhood of
a set A X is an open set containing A.
2.5. Subbasis and basis for a topology. In a metric topology (2.2.(6)) the open sets are unions
of balls. This is a special case of a topology basis. Remember that, in a metric topology, any point is
contained in an open ball and the intersection of two open balls is a union of open balls.
2.6. Definition (Basis and subbasis). A topology basis is a set B T(X) of subsets of X, called
basis sets, such that
(1) Any point of X lies in a basis set, ie X =

B.
(2) The intersection of any two basis sets is a union of basis sets.
A topology subbasis is a set o T(X) of subsets of X, called subbasis sets, such that
23
24 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
(1) Any point of X lies in a subasis set, ie X =

o
If B is a topology basis then the set of subsets of X
T
B
= unions of basis sets
is called the topology generated by B. Note that T
B
is a topology, the coarsests topology in which the
basis sets are open.
If o is a topology subbasis then the set of subsets of X
T
S
= unions of nite intersections of subbasis sets
is called the topology generated by o. Note that T
S
is a topology, the coarsests topology in which the
subbasis sets are open. (Use the distributive laws [1] for and .) The set
(2.7) B
S
= Finite intersections of o-sets
is a topology basis generating the same topology as the subbasis: T
B
S
= T
S
.
A topology is a topology basis is a topology subbasis.
Given a topology T and a topology basis B, we say that B a basis for T if B generates T , ie T
B
= T .
Given a topology T and a topology subbasis o, we say that o a subbasis for T if o generates T , ie
T
S
= T .
2.8. Proposition (Finding a (sub)basis for a given topology). The topology basis B is a basis for
the topology T if and only if
(1) the basis sets are open, and,
(2) all open sets are unions of basis sets
The subbasis o is a subbasis for the topology T if and only if
(1) the subbasis sets are open, and,
(2) all open sets are unions of nite intersections of subbasis sets
2.9. Example. (1) The set of all open rays o = (, b) [ b R (a, +) [ a R is a
subbasis and the set B = (a, b) [ a, b R, a < b of all open intervals is a basis for the standard
topology on R (2.2.(5)).
(2) The set of all open balls is a basis for the metric topology on a metric space (2.2.(6)).
How can we compare topologies given by bases? How can we tell if two bases, or a subbasis and a
basis, generate the same topology? (Two topologies, bases or subbases are said to be equivalent if they
generate the same topology.)
2.10. Lemma (Comparison). Let B and B

be two bases and o a subbasis.

(1)
T
B
T
B
B T
B
All B-sets are open in T
B

(2)
T
B
= T
B

_
All B-sets are open in T
B

All B

-sets are open in T

B
_
(3)
T
B
= T
S

_
All B-sets are open in T
S
All o-sets are open in T
B
_
Proof. (1) is obvious since T
B
is the coarsest topology containing B. Item (2) is immediate from (1).
Item (3) is proved in the same way since T
B
= T
S
i T
B
T
S
and T
B
T
S
i B T
S
and T
B
o.
2.11. Example. [Topologies on R] We consider three topologies on R:
R: The standard topology with basis the open intervals (a, b).
R

: The right half-open interval topology with basis the right half-open intervals [a, b).
R
K
: The K-topology with basis (a, b) (a, b) K where K = 1,
1
2
,
1
3
,
1
4
, . . ..
The right half-open interval topology is strictly ner than the standard topology because any open interval
is a union of half-open intervals but not conversely (2.10) ((a, b) =

a<x<b
[x, b), and [0, 1) is open in R

but not in R; an open interval containing 0 is not a subset of [0, 1)). The K-topology is strictly ner
than the standard topology because its basis contains the standard basis and RK is open in R
K
but
not in R (an open interval containing 0 is not a subset of R K). The topologies R

and R
K
are not
comparable [Ex 13.6].
3. THE PRODUCT TOPOLOGY 25
2.12. Example. (1) In a metric space, the set B = B(x, r) [ x X, r > 0 of open balls is (by
denition) a basis for the metric topology T
d
. The collection of open balls of radius
1
n
, n Z
+
, is an
equivalent topology basis for T
d
.
(2) The collection of rectangular regions (a
1
, b
1
)(a
2
, b
2
) in the plane R
2
is a topology basis equivalent
to the standard basis of open balls B(a, r) = x R
2
[ [x a[ < r. you can always put a ball inside
a rectangle and a rectangle inside a ball.
(3) Let f : X Y be any map. If T is a topology on Y with basis B or subbasis o , then the pull-back
f
1
(T ) is a topology, the initial topology for f, on X with basis f
1
(B) and subbasis f
1
(o).
(4) More generally, let X be a set, Y
j
a collection of topological spaces, and f
j
: X Y
j
, j J, a set
of maps. Let T
j
be the topology on Y
j
, B
j
a basis and o
j
a subbasis for all j J. Then

f
1
j
(T
j
),

f
1
j
(B
j
),

f
1
j
(o
j
) are equivalent subbases on X. The topology they generate is called the initial
topology for the maps f
j
, j J.
2. Order topologies
We associate a topological space to any linearly ordered set and obtain a large supply of examples
of topological spaces. You may view the topological space as a means to study the ordered set, to nd
invariants, or you may view this construction as a provider of interesting examples of topological spaces.
Let (X, <) be a linearly ordered set containing at least two points. The open rays in X are the
subsets
(, b) = x X [ x < b, (a, +) = x X [ a < x
of X. The set o
<
of all open rays is clearly a subbasis (2.8) for a topology on X (just as in 2.9).
2.13. Definition. The order topology T
<
on the linearly ordered set X is the topology generated by
all open rays. A linearly ordered space is a linearly ordered set with the order topology.
The open intervals in X are the subsets of the form
(a, b) = (, b) (a, +) = x X [ a < x < b, a, b X, a < b.
2.14. Lemma (Basis for the order topology). Let (X, <) be a linearly ordered set.
(1) The union of all open rays and all open intervals is a basis for the order topology T
<
.
(2) If X has no smallest and no largest element, then the set (a, b) [ a, b X, a < b of all open
intervals is a basis for the order topology.
Proof. We noted in (2.7) that B
S
<
= Finite intersections of o-sets = o (a, b) [ a, b X, a < b
is a basis for the topology generated by the subbasis o
<
.
If X has a smallest element a
0
then (, b) = [a
0
, b) is open. If X has no smallest element, then
the open ray (, b) =

a<c
(a, c) is a union of open intervals and we do not need this open ray in the
basis. Similar remarks apply to the greatest element when it exists.
2.15. Example. (1) The order topology on the ordered set (R, <) is the standard topology (2.9).
(2) The order topology on the ordered set R
2
has as basis the collection of all open intervals (a
1

a
2
, b
1
b
2
). An equivalent basis (2.10) consists of the open intervals (a b
1
, a b
2
). The order
topology R
2
<
is strictly ner than the metric topology R
2
d
.
(3) The order topology on Z
+
is the discrete topology because (, n) (, n + 1) = n is open.
(4) The order topology on Z
+
Z
+
is not discrete. Any open set that contains the element 2 1 also
contains elements from 1 Z
+
. Thus the set 1 2 is not open.
(5) The order topology on Z Z is discrete.
(6) Is the order topology on S

discrete?
(7) I
2
= [0, 1]
2
with the order topology is denoted I
2
o
and called the ordered square. The open sets
containing the point x y I
2
o
look quite dierent depending on whether y 0, 1 or 0 < y < 1.
3. The product topology
Let (X
j
)
jJ
be an indexed family of topological spaces. Let
k
:

jJ
X
j
X
k
be the projection
map. An open cylinder is a subset of the product space of the form

1
k
(U
k
), U
k
X
k
open, k J.
The set
1
k
(U
k
) consists of the points (x
j
)

X
j
with kth coordinate in U
k
. Alternatively,
1
k
(U
k
)
consists of all choice functions c: J

jJ
U
j
such that c(k) U
k
(Figure 2).
26 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
2.16. Definition. The product topology on

jJ
X
j
is the topology with subbasis
o
Q
=
_
jJ

1
j
(U
j
) [ U
j
X
j
open
consisting of all open cylinders or, equivalently, with basis (2.7)
B
Q
=

jJ
U
j
[ U
j
X
j
open and U
j
= X
j
for all but nitely many j J
The product topology is the coarsest topology making all the projection maps
j
:

X
j
X
j
, j J,
continuous.
This becomes particularly simple when we consider nite products.
2.17. Lemma. Let X = X
1
X
2
X
k
be a nite Cartesian product. The collection
B = U
1
U
2
U
k
[ U
1
open in X
1
, U
2
open in X
2
, . . . , U
k
open in X
k

of all products of open sets is a basis for the product topology.

2.18. Corollary. Suppose that B
j
is a basis for the topology on X
j
, j = 1, . . . , k. Then B
1
B
k
is a basis for the product topology on X
1
X
k
.
Proof. Note that B
1
B
k
is indeed a basis. Compare it to the basis of 2.17 using 2.10.
2.19. Products of linearly ordered spaces. When (X, <) and (Y, <) are linearly ordered sets, we
now have two topologies on the Cartesian product XY : The product topology of the order topologies,
(X, T
<
) (Y, T
<
), and the order topology of the product dictionary order, (X Y, T
<
). These two
topologies are in general not identical or even comparable. It is dicult to imagine any general relation
between them since X
<
Y
<
is essentially symmetric in X and Y whereas the dictionary order has no
such symmetry. But even when X = Y there does not seem to be a general pattern: The order topology
(Z
+
Z
+
, T
<
) is coarser than the product topology (Z
+
, T
<
)(Z
+
, T
<
) (which is discrete) (2.15.(4)(5)).
On the other hand, the order topology (RR, T
<
) is ner than the product topology (R, T
<
) (R, T
<
)
which is the standard topology, see 2.15.(1)(2) and [Ex 16.9].
2.20. Corollary (Cf [Ex 16.9]). Let (X, <) and (Y, <) be two linearly ordered sets. Suppose that Y
does not have a largest or a smallest element. Then the order topology (X Y )
<
= X
d
Y
<
where X
d
is X with the discrete topology. This topology is ner than X
<
Y
<
.
Proof. The equations
(, a b) =
_
x<a
x Y a (, b), (a b, +) = a (b, +)
_
x>a
x Y
show that the order topology is coarser than the product of the discrete and the order topology. On the
other hand, the sets x (c, d) is a basis for = X
d
Y
<
(2.14.(2), 2.18) and x (c, d) = (x c, x d)
is open in the order topology since it is an open interval.
The corollary shows that the order topology on XY does not yield anything new in case the second
factor is unbounded in both directions. As an example where this is not the case we have already seen
Z
+
Z
+
and I
2
o
(2.15.(4), 2.15.(7)).
2.21. The coproduct topology. The coproduct topology on the coproduct

X
j
(Chapter 1.3)
is the nest topology making all the inclusion maps
j
: X
j

X
j
, j J, continuous.
This means that
U

jJ
X
j
is open
1
j
(U) is open for all j J
Alternatively, the open sets of the coproduct

X
j
are the coproducts

U
j
of open set U
j
X
j
.
4. The subspace topology
Let (X, T ) be a topological space and Y X a subset. We make Y into a topological space.
2.22. Definition. The subspace topology on Y is the topology T

= Y T = Y U [ U T . A
subset V Y is
_
open
closed
_
relative to Y if it is
_
open
closed
_
in T

.
It is immediate that T

is indeed a topology.
4. THE SUBSPACE TOPOLOGY 27
2.23. Lemma. If B is a basis for T , then Y B = Y U [ U B is a basis for the subspace topology
Y T . If o is a subbasis for T , then Y o = Y U [ U o is a subbasis for the subspace topology
Y T .
Proof. This is 2.12.(3) applied to the inclusion map A X.
2.24. Lemma. Assume that A Y X. Then
(1) A is
_
open
closed
_
in Y A = Y U for some
_
open
closed
_
set U in X
(2) If Y is
_
open
closed
_
then: A is
_
open
closed
_
in Y A is
_
open
closed
_
in X
Proof. (1) This is the denition of the subspace topology.
(2) Suppose that Y is open and that A Y . Then
A open in Y A = Y U for some open U X A open in X
in that A = A Y . Similarly, if Y is closed.
The lemma says that an open subset of an open subset is open and that a closed subset of a closed
subset is closed.
The next theorem says that the subspace and the product space operations commute.
2.25. Theorem. Let Y
j
X
j
, j J. The subspace topology that

Y
j
inherits from

X
j
is the
product topology of the subspace topologies on Y
j
.
Proof. The subspace topology on

Y
j
has subbasis

Y
j
o
Q
X
j
=

Y
j

_
kj

1
k
(U
k
) =
_
kJ

Y
j

1
k
(U
k
), U
k
X
k
open,
and the product topology on

Y
j
has subbasis
o
Q
Y
j
=
_
kJ

1
k
(Y
k
U
k
), U
k
X
k
open
These two subbases are identical for
1
k
(Y
k
U
k
)
[8, Ex 2.2]
=
1
k
(Y
k
)
1
k
(U
k
) = (

Y
j
)
1
k
(U
k
).
2.26. Subspaces of linearly ordered spaces. When (X, <) is a linearly ordered set and Y X
a subset we now have two topologies on Y . We can view Y as a subspace of the topological space X
<
with the order topology or we can view Y as a sub-ordered set of X and give Y the order topology. Note
that these two topologies are not the same when X = R and Y = [0, 1) 2 R: In the subspace
topology 2 is open in Y but 2 is not open in the order topology because Y has the order type of
[0, 1]. See 2.28 for more examples. The point is that
Any open ray in Y is the intersection of Y with an open ray in X
The intersection of Y with an open ray in X need not be an open ray in Y
However, if Y happens to be convex then open rays in Y are precisely Y intersected with open rays in
X.
2.27. Lemma (Y
<
Y X
<
). Let (X, <) be a linearly ordered set and Y X a subset. The order
topology on Y is coarser than the subspace topology Y in general. If Y is convex, the two topologies on
Y are identical.
Proof. The order topology on Y has subbasis
o
<
= Y (, b) [ b Y Y (a, ) [ a Y
and the subspace topology on Y has subbasis
o

= Y (, b) [ b X Y (a, ) [ a X
Clearly, o
<
o

, so the order topology on Y is coarser than the subspace topology in general.

If Y is convex and b X Y then b is either a lower or an upper bound for Y as we cannot have
y
1
< b < y
2
for two points y
1
and y
2
of Y . If b is a lower bound for Y , then Y (, b) = , and if b is
an upper bound for Y , then Y (, b) = Y . Therefore also o

T
<
so that in fact T
<
= T

2.28. Example. (1) S

21
= 1 Z
+
2 1 = [1 1, 2 1] is a convex subset of Z
+
Z
+
. The
subspace topology (2.15.(4)) is the same as the order topology.
28 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
(2) The subset Z
+
Z
+
is not convex in ZZ so we expect the subspace topology to be strictly ner
than the order topology. Indeed, the subspace topology that Z
+
Z
+
inherits from the discrete space
Z Z is discrete but the order topology is not discrete (2.15.(4)(5)).
(3) Consider X = R
2
and Y = [0, 1]
2
with order topologies. Y is not convex so we expect the subspace
topology to be strictly ner than the order topology. Indeed, the subspace topology on [0, 1]
2
, which
is [0, 1]
d
[0, 1], is strictly ner than I
2
o
(2.15.(7)): The set
[0,
1
2
] [0, 1] = ([0, 1] [0, 1]) (,
1
2
2)
is open in the subspace topology on Y but it is not open in the order topology on Y as any basis open
set (2.14) containing
1
2
1 also contains points with rst coordinate >
1
2
.
(4) Consider X = R
2
and Y = (0, 1)
2
with order topologies. Y is not convex so we expect the subspace
topology to be strictly ner than the order topology. But it isnt! The reason is that (0, 1) does not
have a greatest nor a smallest element (2.20).
(5) The subset Q of the linearly ordered set R is not convex but nevertheless the subspace topology
inherited from R is the order topology. Again, the reason seems to be that Q does not have a greatest
nor a smallest element.
5. CLOSED SETS AND LIMIT POINTS 29
5. Closed sets and limit points
2.29. Definition. A subset of a topological space is closed if its complement is open.
2.30. Lemma. The closed subsets of a topological space X have these properties:
(1) and X are closed
(2) The union of nitely many closed sets is closed
(3) Any intersection of closed sets is closed.
Conversely, any set of subsets of X with these properties, are the closed sets for a topology.
2.31. Example. (1) [a, b] is closed in R, [a, b) is neither closed nor open, R is both closed and
open.
(2) Let X = [0, 1] (2, 3). The subsets [0, 1] and (2, 3) of X are both closed and open (they are clopen)
as subsets of X.
(3) K = 1,
1
2
,
1
3
, . . . is not closed in R but it is closed in R
K
(2.11).
(4) R

is (2.11) the set of real numbers equipped with the topology generated by the basis sets [a, b)
of right half-open intervals. All sets that are (open) closed in the standard topology R are also
(open) closed in the ner topology R

. Sets of the form (, a) =

x<a
[x, a) = R [a, ),
[a, b) = (, b) [a, ), and [a, ) =

a<x
[a, x) = R (, a) are both open and closed. Sets of
the form (, b] or [a, b] are closed (since they are closed in the standard topology) and not open
since they are not unions of basis sets. Sets of the form (a, ) are open (since they are open in
standard topology) and not closed. Sets of the form (a, b] are neither open nor closed.
(5) Let X be a well-ordered set (6). In the order topology (2), sets of the form (a, ) = [a
+
, ) =
X (, a], (, b] = (, b
+
) = X (b, ) and (a, b] = (a, ) (, b] are closed and open.
(Here, b
+
denotes b if b is the largest element and the immediate successor of b if b is not the largest
element.)
(6) Let R be a ring and Spec(R) the set of prime ideals of R. (We adopt the convention that R itself
is not a prime ideal. The zero ideal is a prime ideal if R is a domain.) For any subset S of R let
V (S) = a Spec(R) [ a S
be the set of prime ideals containing S. Then = V (1), Spec(R) = V (0), V (S
1
) V (S
2
) = V (S
1
S
2
),
and

V (S
i
) = V (

S
i
). The Zariski topology on Spec(R) is the topology where the closed sets are the
sets V (S) for S R. Thus the closed sets are intersections of the closed sets V (r) = a Spec(R) [
a r, r R. The closed sets of Spec(Z) = 0 p [ p > 0 is prime are intersections of the closed
sets V (0) = Spec(Z) and V (n) = p [ p[n, 0 ,= n Z, consisting of the prime divisors of n. The closed
sets of Spec(C[X]) = (0) (X a) [ a C are intersections of the closed sets V (0) = Spec(Z)
and V (P) = a C [ P(a) = 0 consisting of the roots of the complex polynomial 0 ,= P C[X].
The subspace topology on C Spec(C[X]) is the nite complement topology (2.2.(4)).
2.32. Closure, interior, and boundary. We consider the largest open set contained in A and the
smallest closed set containing A.
2.33. Definition. The interior of A is the union of all open sets contained in A,
Int A =
_
U A [ U open = A

,
the closure of A is the intersection of all closed sets containing A,
Cl A =

C A [ C closed = A,
and the boundary of A is A = AA

.
2.34. Proposition (Properties of interior and closure). Let A, B X and let x X. Then
(1) x A

There exists a neighborhood U of x such that U A

(2) X A = (X A)

and X A

= X A
(3) x A U A ,= for all neighborhoods U of x
(4) x A U A ,= and U (X A) ,= for all neighborhoods U of x
(5) A

A A
(6) A

is open, it is the largest open set contained in A, and A is open i A

= A
(7) A is closed, it is the smallest closed set containing A, and A is closed i A = A
(8) (A B)

= A

, A B = A B, (A B)

, A B A B,
(9) A B =
_
Int A Int B
Cl A Cl B
30 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
(10) X = A

A (X A) is a disjoint union.
Proof. The rst assertion is clear as A

is the union of the open sets contained in A. The second

claim follows from:
X A = X

C A [ C closed =
_
X C X A [ X C open
=
_
U X A [ U open = (X A)

or one can simply say that X A is the largest open subset of X A since A is the smallest closed
superset of A. The third assertion is just a reformulation of the rst one,
x , A x X A x (X A)

x has a neighborhood disjoint from A,

since X A = (X A)

. The remaining facts are easily proved.

2.35. Proposition (Closure with respect to subspace). Let A Y X. Then
(1) A is closed in Y A = Y C for some closed set C in X
(2) Cl
Y
(A) = Y A.
(3) If Y is closed, then Cl
Y
(A) = A.
Proof. (1) This is 2.24.1.
(2) The set Cl
Y
(A) is the intersection of all relatively closed sets containing A. The relatively closed sets
are the sets of the form Y C where C is closed in X. This means that
Cl
Y
(A) =

Y C [ C A, C closed = Y

C [ C A, C closed = Y A
by a direct computation.
(3) As Y is closed and A Y , also A Y so that Cl
Y
(A) = Y A = A.
If A Y X, Y Int(A) Int
Y
(A) for Y Int(A) is a relatively open set contained in A. These
two sets are equal if Y is open but they are distinct in general (consider A = [0, 1) [0, 1] = Y ).
2.36. Definition. A subset A X is said to be dense if A = X, or, equivalently, if every open
subset of X contains a point of A.
2.37. Proposition. Let A be a dense and U an open subset of X. Then A U = U.
Proof. The inclusion A U U is general. For the other inclusion, consider a point x U. Let V
be any neighborhood of x. Then V (AU) = (V U) A is not empty since V U is a neighborhood
of x and A is dense. But this says that x is in the closure of A U.
2.38. Limit points and isolated points. Let X be a topological space and A a subset of X.
2.39. Definition (Limit points, isolated points). A point x X is a limit point of A if U(Ax) ,=
for all neighborhoods U of x. The set of limit points
1
of A is denoted A

. A point a A is an isolated
point if a has a neighborhood that intersects A only in a.
Equivalently, x X is a limit point of A i x Ax, and a A is an isolated point of A i a
is open in A. These two concepts are almost each others opposite:
x is not a limit point of A x has a neighborhood U such that U (A x) = x
x is an isolated point of A x
2.40. Proposition. Let A be a subset of X and A

the set of limit points of A. Then A A

= A
and A A

= a A [ a is not an isolated point of A so that

A A

A is closed
A A

A has no isolated points

A A

= A is discrete
A

= A is closed and discrete

If B A then B

.
1
The set of limit points of A is sometimes denotes A
d
and called the derived set of A
5. CLOSED SETS AND LIMIT POINTS 31
Proof. It is clear that all limit points of A are in A, so that A A

A, and that all points in the

closure of A that are not in A are limit points,
AA = x X A [ all neighborhoods of x meet A A

A,
so that A = A (A A) A A

. This shows that A A

= A. From the above discussion we have

that the set of limit points in A, A A

= A (A A

), is the set of non-isolated points of A. If

AA

= then all all points of A are isolated so that the subspace A has the discrete topology. We have
A

= A A

, A A

= A is closed and discrete.

2.41. Example. (1) Q = R = RQ (RQ contains (Q0)

2). Cl Int Q = , Int Cl Q = R.

(2) Let C
r
R
2
be the circle with center (0, r) and radius r > 0. Then
_
nZ
+
C
1/n
=
_
nZ
+
C
1/n
,
_
nZ
+
C
n
= (R0)
_
nZ
+
C
n
The rst set of decreasing circles (known as the Hawaiian Earring [8, Exmp 1 p 436]) is closed because
it is also the intersection of a collection (which collection?) of closed sets.
(3) The set of limit points of K =
1
n
[ n Z
+
is 0 in R and R

and in R
K
.
(4) Let X be a linearly ordered space (2.13). Closed intervals [a, b] are closed because their complements
X [a, b] = (, a) (b, ) are open. Therefore the closure of an interval of the form [a, b) is either
[a, b) or [a, b]. If b has an immediate predecessor b

then [a, b) = [a, b

] is closed so that [a, b) = [a, b).

Otherwise, [a, b) = [a, b] because any neighborhood of b contains an interval of the form (c, b] for some
c < b and [a, b) (c, b], which equals [a, b] or (c, b], is not empty.
2.42. Convergence, the Hausdor property, and the T
1
-axiom. Let x
n
, n Z
+
, be a se-
quence of points in X, ie a map Z
+
X.
2.43. Definition. The sequence x
n
converges to the point x X if for any neighborhood U of x
there exists some N such that x
n
U for all n > N.
If a sequence converges in some topology on X it also converges in any coarser topology but not
necessarily in a ner topology.
2.44. Example. In the Sierpinski space X = 0, 1 (2.2.(3)), the sequence 0, 0, converges to 0
and to 1. In R with the nite complement topology (2.2.(4)), the sequence 1, 2, 3, converges to any
point. In R
K
, the sequence
1
n
does not converge; in R and R

it converges to 0 (and to no other point).

The sequence
1
n
converges to 0 in R and R
K
, but does not converge in R

. In the ordered space

[1, ] = Z
+
H (1.49.(6)), the sequence n converges to (and to no other point). Can you nd a
sequence in the ordered space [0, ] (1.52) that converges to ?
2.45. Definition (Separation Axioms). A topological space is T
1
-space if points are closed: For any
two distinct points x
1
,= x
2
in X there exists an open set U such that x
1
U and x
2
, U.
A topological space X is a T
2
-space (or a Hausdor space) if there are enough open sets to separate
points: For any two distinct points x
1
,= x
2
in X there exist disjoint open sets, U
1
and U
2
, such that
x
1
U
1
and x
2
U
2
.
All Hausdor spaces are T
1
. X is T
1
i all nite subsets are closed. Conite topologies are T
1
by
construction and not T
2
when the space has innitely many points. Particular point topologies are not
T
1
(on spaces with more than one point).
All linearly ordered [Ex 17.10] or metric spaces are Hausdor; in particular, R is Hausdor. R

and
R
K
are Hausdor because the topologies are ner than the standard Hausdor topology R.
2.46. Theorem. A sequence in a Hausdor space can not converge to two distinct points.
A property of a topological space is said to be (weakly) hereditary if any (closed) subspace of a space
with the property also has the property. Hausdorness is hereditary and also passes to product spaces.
2.47. Theorem (Hereditary properties of Hausdor spaces). [Ex 17.11, 17.12] Any subset of a Haus-
dor space is Hausdor. Any product of Hausdor spaces is Hausdor.
2.48. Theorem. Suppose that X is T
1
. Let A be a subset of and x a point in X. Then
x is a limit point All neighborhoods of x intersect A in innitely many points
32 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
Proof. =: If all neighborhoods of x intersect A in innitely many points, then, clearly, they also
intersect A in a point that is not x.
=: Assume that x is a limit point and let U be a neighborhood of x. Then U contains a point a
1
of
A dierent from x. Remove this point from U. Since points are closed, U a
1
is a new neighborhood
of x. This new neighborhood of x contains a point a
2
of A dierent from x and a
1
. In this way we
recursively nd a whole sequence of distinct points in U A.
6. Continuous functions
Let f : X Y be a map between two topological spaces.
2.49. Definition. The map f : X Y is continuous if all open subsets of Y have open preimages
in X: V open in Y = f
1
(V ) open in X
If T
X
is the topology on X and T
Y
is the topology on Y then
(2.50) f is continuous f
1
(T
Y
) T
X
f
1
(B
Y
) T
X
f
1
(o
Y
) T
X
where B
Y
is a basis and o
Y
a subbasis for T
Y
. The ner the topology in Y and the coarser the topology
in X are, the more dicult is it for f : X Y to be continuous.
2.51. Theorem. Let f : X Y be a map between two topological spaces. The following are equivalent:
(1) f is continuous
(2) The preimage of any open set in Y is open in X: V open in Y = f
1
(V ) open in X
(3) The preimage of any closed set in Y is closed in X: C closed in Y = f
1
(C) closed in X
(4) f
1
(B

) (f
1
(B))

for any B Y
(5) f
1
(B) f
1
(B) for any B Y
(6) f(A) f(A) for any A X
(7) For any point x X and any neighborhood V Y of f(x) there is a neighborhood U X of x
such that f(U) V .
Proof. It is easy to see that (7),(1),(2), and (3) are equivalent.
(2) = (4):
f
1
(B

) f
1
(B)
f
1
(B

) open
_
= f
1
(B

) (f
1
(B))

.
(4) = (2): Let B be any open set in Y . Then f
1
(B)
B=B

= f
1
(B

)
(4)
(f
1
(B))

f
1
(B) so f
1
(B)
is open since it equals its own interior.
(3) (5): Similar to (2) (4).
(3) = (6):
A f
1
(f(A)) f
1
(f(A))
f
1
(f(A)) is closed
_
= A f
1
(f(A)) f(A) f(A).
(6) = (5): f(f
1
(B))
(6)
f(f
1
(B))
f(f
1
(B))B
B
2.52. Example (Examples of continuous maps). (1) If X and Y and Y have metric topologies
(2.2.(6)) then f : X Y is continuous if and only if for all x X and all > 0 there is a > 0 so
that fB(x, ) B(f(x), ).
(2) If X has the discrete topology (2.2.(2)) or Y has the trivial topology (2.2.(1)) then any map
f : X Y is continuous.
(3) If X and Y have the particular point topologies (2.2.(3)) then f : X Y is continuous if and only
if f preserves the particular points.
(4) If X and Y have nite complement topologies (2.2.(4)) then f : X Y is continuous if and only
if f has nite bres.
(5) The function f(x) = x is continuous R R, but not continuous R
K
R
K
(K is closed
in R
K
but f
1
(K) is not closed (2.44)) and not continuous R

([0, ) is open in R

but
f
1
([0, )) = (, 0] is not open (2.31.(4))).
(6) Since [a, b) is closed and open in R

(2.31.(4)) the map 1

[a,b)
: R

0, 1, with value 1 on [a, b)

and value 0 outside [a, b), is continuous.
(7) Any ring homomorphism R S induces a map Spec(S) Spec(R) taking any prime ideal x S
to the prime ideal f
1
x R. This map is continuous because it takes Zariski closed set to Zariski
closed sets (2.31.(6)).
2.53. Theorem. (1) The identity function is continuous.
(2) The composition of two continuous functions is continuous.
6. CONTINUOUS FUNCTIONS 33
(3) Let B be a subspace of Y , A a subspace of X, and f : X Y a map taking values in B. Then
f : X Y is continuous = f[A: A Y is continuous (restriction)
f : X Y is continuous B[f : X B is continuous (corestriction)
(4) Let f
j
: X
j
Y
j
, j J, be an indexed set of maps. Then

f
j
:

X
j

Y
j
is continuous f
j
: X
j
Y
j
is continuous for all j J
(5) (Glueing lemma) Let B be a covering of X. Suppose either that X is covered by the interiors of
the sets in B or that B is a locally nite closed covering. (The covering B is locally nite if any
point in X has a neighborhood that intersects only nitely many of the sets from B.) Then
f[B: B Y is continuous for all B in B = f : X Y is continuous
for any map f : X Y .
Proof. Most of these observations are easy to check. The =-part of (4) uses 2.64 below. In (5),
let us consider the case where B is a locally nite closed covering (or just a closed covering so that any
point in X has a neighborhood that is contained in a nite union of sets from B. Suppose that A B is
open in B for all B B. We claim that A is open. Let a be a point in A. Choose a neighborhood U of a
and nitely many sets B
1
, . . . , B
m
B such that a U B
1
B
m
. We may assume that a B
i
for
all i = 1, . . . , m for otherwise we just replace U by U B
i
. Since A B
i
is open in B
i
there is an open
set U
i
such that A B
i
= U
i
B
i
for all i = 1, . . . , m. Now U U
1
U
m
is an open neighborhood
of a and
U U
1
U
m
(B
1
B
m
) (U
1
U
m
) (B
1
U
1
) (B
m
U
m
) A
This shows that A is open.
2.54. Homeomorphisms and embeddings. One of the central problems in topology is to decide
if two given spaces are homeomorphic.
2.55. Definition (Homeomorphism). A bijective continuous map f : X Y is a homeomorphism
if its inverse is continuous.
A bijection f : X Y induces a bijection between subsets of X and subsets of Y , and it is a home-
omorphism if and only if this bijection restricts to a bijection
_
Open (or closed) subsets of X
_

Uf(U)
f
1
(V )V

_
Open (or closed) subsets of Y
_
between open (or closed) subsets of X and open (or closed) subsets of Y .
We now extend the subspace topology (2.22) to a slightly more general situation.
2.56. Definition (Embedding topology). Let X be a set, Y a topological space, and f : X Y an
injective map. The embedding topology on X (for the map f) is the collection
f
1
(T
Y
) = f
1
(V ) [ V Y open
of subsets of X.
The subspace topology for A X is the embedding topology for the inclusion map A X.
2.57. Proposition (Characterization of the embedding topology). Suppose that X has the embedding
topology for the map f : X Y . Then
(1) X Y is continuous and,
(2) for any map A X into X,
A X is continuous A X
f
Y is continuous
The embedding topology is the only topology on X with these two properties. The embedding topology is
the coarsest topology on X such that f : X Y is continuous.
Proof. This is because
A
g
X is continuous
(2.50)
g
1
(T
X
) T
A
g
1
(f
1
T
Y
) T
A

(fg)
1
(T
Y
) T
A
A
g
X
f
Y is continuous
by denition of the embedding topology. The identity map of X is a homeomorphism whenever X is
equipped with a topology with these two properties.
34 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
2.58. Definition (Embedding). An injective continuous map f : X Y is an embedding if the
topology on X is the embedding topology for f, ie T
X
= f
1
T
Y
.
Any injective map f : X Y induces a bijection between subsets of X and subsets of f(X), and it
is an embedding if and only if this bijection restricts to a bijection
(2.59)
_
Open (or closed) subsets of X
_

Uf(U)
f
1
(V )V

_
Open (or closed) subsets of f(X)
_
between open (or closed) subsets of X and open (or closed) subsets of f(X).
Alternatively, the injective map f : X Y is an embedding if and only if the bijective corestriction
f(X)[f : X f(X) is a homeomorphism. An embedding is a homeomorphism followed by an inclusion.
The inclusion A X of a subspace is an embedding. Any open (closed) continuous injective map is an
embedding.
2.60. Example. (1) The map f(x) = 3x + 1 is a homeomorphism R R.
(2) The identity map R

R is bijective and continuous but not a homeomorphism.

(3) The map [0, 1) S
1
: t (cos(2t), sin(2t)) is continuous and bijective but not a homemorphism.
The image of the open set [0,
1
2
) is not open in S
1
.
(4) Find an example of an injective continuous map R R
2
that is not an embedding.
(5) The obvious bijection [0, 1) 2 [0, 1] is continuous but not a homeomorphism (the domain
has an isolated point, the codomain has no isolated points). There does not exist any continuous
surjection in the other direction.
(6) The spaces [1 1, 2 1] Z
+
Z
+
and K =
1
n
[ n Z
+
R are homeomorphic.
(7) The map R R R: t (t, t) is an embedding. For any continuous map f : X Y , the map
X X Y : x (x, f(x)) is an embedding; see (3.25) for a generalization.
(8) R
n
embeds into S
n
via stereographic projection.
(9) R
2
embeds in R
3
. Does R
3
imbed in R
2
? (See notes on algebraic topology for the answer.)
(10) Are the spaces

C
n
and

C
1/n
of 2.41.(2) homeomorphic?
(11) A knot is in embedding of S
1
in R
3
(or S
3
). Two knots, K
0
: S
1
R
3
and K
1
: S
1
R
3
, are
equivalent if there exists a homeomorphism h of R
3
such that h(K
0
) = K
1
. The fundamental problem
of knot theory [1] is to classify knots up to equivalence.
2.61. Lemma. If f : X Y is a homeomorphism (embedding) then the corestriction of the restriction
f(A)[f[A: A f(A) (B[f[A: A B) is a homeomorphism (embedding) for any subset A of X (and any
subset B of Y containing f(A)). If the maps f
j
: X
j
Y
j
are homeomorphisms (embeddings) then the
product map

f
j
:

X
j
Y
j
is a homeomorphism (embedding).
Proof. In case of homeomorphisms there is a continuous inverse in both cases. In case of embed-
dings, use that an embedding is a homeomorphism followed by an inclusion map.
2.62. Lemma (Composition of embeddings). Let X
f

Y
g

Z be continuous maps. Then

f and g are embeddings = g f is an embedding = f is an embedding
Proof. To prove he second implication, note rst that f is an injective continuous map. Let U X
be open. Since g f is an embedding, U = (g f)
1
W for some open W Z. But (g f)
1
= f
1
g
1
W
where g
1
W is open in Y since g is continuous. This shows that f is an embedding.
2.63. Maps into products. There is an easy test for when a map into a product space is contin-
uous.
2.64. Theorem (Characterization of the product topology). Give

Y
j
the product topology. Then
(1) the projections
j
:

Y
j
Y
j
are continuous, and,
(2) for any map f : X

jJ
Y
j
into the product space we have
X
f

jJ
Y
j
is continuous j J : X
f

jJ
Y
j

j
Y
j
is continuous
The product topology is the only topology on the product set with these two properties.
6. CONTINUOUS FUNCTIONS 35
Proof. Let T
X
be the topology on X and T
j
the topology on Y
j
. Then o
Q
=

jJ

1
j
(T
j
) is a
subbasis for the product topology on

jJ
Y
j
(2.16). Therefore
f : X

jJ
Y
j
is continuous
(2.50)
f
1
(
_
jJ

1
j
(T
j
)) T
X

_
jJ
f
1
(
1
j
(T
j
)) T
X
j J : (
j
f)
1
(T
j
) T
X
j J :
j
f is continuous
by denition of continuity (2.50).
We now show that the product topology is the unique topology with these properties. Take two
copies of the product set

jJ
X
j
. Equip one copy with the product topology and the other copy with
some topology that has the two properties of the theorem. Then the identity map between these two
copies is a homeomorphism.
The reason for the great similarity between 2.64 and 2.57 is that in both cases we use an initial
topology.
2.65. Example. Suppose that J and K are sets and that (X
j
)
jJ
and (Y
k
)
kK
are indexed families
of topological spaces. Given a map g : J K between index sets and an indexed family of continuous
maps (f
j
: Y
g(j)
X
j
)
jJ
. Then there is a unique map between product spaces such that

kK
Y
k

g(j)

jJ
X
j

Y
g(j)
f
j

X
j
commutes and this map of product spaces is continuous by 2.64.
2.66. Theorem. Let (X
j
)
jJ
be an indexed family of topological spaces with subspaces A
j
X
j
.
Then

jJ
A
j
is a subspace of

jJ
X
j
.
(1)

A
j
=

A
j
.
(2)
_
A
j
_

j
and equality holds if A
j
= X
j
for all but nitely many j J.
Proof. (1) Let (x
j
) be a point of

X
j
. Since o
Q
=

jJ

1
j
(T
j
) is a subbasis for the product topology
on

X
j
(2.16) we have:
(x
j
)

A
j
k J :
1
k
(U
k
)

A
j
,= for all neighborhoods U
k
of x
k
k J : U
k
A
k
,= for all neighborhoods U
k
of x
k
k J : x
k
A
k
(x
j
)

A
j
(2)
_
A
j
_

j
because
j
is an open map (2.71) so that
j
_
_
A
j
_

_
A

j
for all j J. If
A
j
= X
j
for all but nitely many j J then

A

j

_
A
j
_

because

A

j
is open and contained in

A
j
.
It follows that a product of closed sets is closed. (Whereas a product of open sets need not be open
in the product topology.)
2.67. Maps out of coproducts.
2.68. Theorem. Let f :

jJ
X
j
Y be a map out of a coproduct space. Then
f :

jJ
X
j
Y is continuous f
j
: X
j
Y is continuous for all j J
where
j
: X
j

jJ
X
j
is the inclusion map.
Let f
j
: X
j
Y
j
, j J, be an indexed set of maps. Then

f
j
:

X
j

Y
j
is continuous f
j
: X
j
Y
j
is continuous for all j J
36 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
7. The quotient topology
In this section we will look at the quotient space construction. But rst we consider open and closed
maps.
2.1. Open and closed maps. [2, I.5] Let X and Y be topological spaces and f : X Y a map.
2.69. Definition. The map f : X Y is
_
open
closed
_
if for all U X we have:
U
_
open
closed
_
in X = f(U)
_
open
closed
_
in Y
The restriction (2.80) of an open (or closed) map to an arbitrary subspace need not be open (closed).
However,
2.70. Proposition. [8, Ex 22.5] The restriction of an open (or closed) map f : X Y to an open
(closed) subspace A X is an open (closed) map f(A)[f[A: A f(A) or f[A: A Y .
Proof. Suppose that f : X Y is an open map and A X an open subset. Let U be an open
subset of A. The implications
U is open in A (A is open in X)
= U open in X (f is open)
= f(U) open in Y
= f(U) open in f(A)
show that f[A: A f(A) is open.
2.71. Proposition (Projections are open). The projection map
j
:

X
j
X
j
is open.
Proof. The map
j
takes the basis B
Q
(2.16) for the product topology into the topology on X
j
.
2.72. Lemma (Characterization of open or closed continuous maps). Let f : X Y be a continuous
map.
(1) f is open if and only if f
1
(B

) = f
1
(B)

for all B Y .
(2) f is closed if and only if f(A) = f(A) for all A X
Proof. See Solution June 04 (Problem 1) and Solution Jan 05 (Problem 1).
A bijective continuous map that is open or closed is a homeomorphism.
2.73. Example. (1) The projection
1
: X Y X is continuous and open (2.71). It is closed
if Y is compact. (Use 2.143 to see this.) The projection map
1
: RR R is not closed for
H = (x, y) RR [ xy = 1 is closed but
1
(H) = R0 is not closed. Thus the product of two
closed maps (the identity map of R and the constant map R ) need not be closed.
(2) The map f : [1, 2] [0, 1] given by
f(x) =
_

_
0 1 x 0
x 0 x 1
1 1 x 2
is continuous and closed (2.141.(1)). It is not open for f([1, 1/2)) = 0 is not open.
2.2. Quotient topologies and quotient maps. Quotient maps are continuous surjective maps
that generalize both continuous, open surjective maps and continuous, closed surjective maps.
2.74. Definition (Quotient topology). Let X be a topological space, Y a set, and p: X Y a
surjective map. The quotient topology on Y is the collection
V Y [ p
1
(V ) is open in X
of subsets of Y .
The quotient topology is sometimes called the nal topology [2, I.2.4] with respect to the map p.
2.75. Lemma (Characterization of the quotient topology). Suppose that Y has the quotient topology
with respect to the map p: X Y . Then
(1) p: X Y is continuous, and,
7. THE QUOTIENT TOPOLOGY 37
(2) for any map g : Y Z out of Y
Y
g
Z is continuous X
p
Y
g
Z is continuous
The quotient topology is the only topology on Y with these two properties. The quotient topology is the
nest topology on Y such that p: X Y is continuous.
Proof. This is because
Y
g
Z is continuous g
1
(T
Z
) T
Y
p
1
g
1
(T
Z
) T
X
(gp)
1
(T
Z
) T
X
X
p
Y
g
Z is continuous
by denition of the quotient topology.
If we give Y some topology with these two properties then the identity map between the two topologies
is a homeomorphism.
2.76. Definition (Quotient map). A surjective continuous map p: X Y is a quotient map if the
topology on Y is the quotient topology.
This means that the surjective map p: X Y is a quotient map if and only if for all V Y :
p
1
(V ) is open in X V is open in Y
Quotient maps and embeddings (2.58) are dual concepts.
Subsets A of X of the form A = p
1
(B) =

yB
p
1
(y) for some subset B of Y are called saturated
subsets of X. They are the subsets that are unions of bres f
1
(y), y Y . The saturation of A X is
the union f
1
f(A) =

yf(A)
f
1
(y) of all bres that meet A. A is saturated if and only if A = f
1
f(A).
2.77. Proposition. For a surjective map p: X Y the following are equivalent:
(1) p: X Y is a quotient map
(2) For all V Y we have: p
1
(V ) is
_
open
closed
_
in X V is
_
open
closed
_
in Y .
(3) p: X Y is continuous and maps saturated
_
open
closed
_
sets to
_
open
closed
_
open sets
Proof. Condition (1) and condition (2) with the word open are clearly equivalent. Suppose now
that: p
1
(V ) is open V is open. Then we get
p
1
(C) is closed X p
1
(C) is open p
1
(Y C) is open
Y C is open C is closed
for all C Y . This shows that the two conditions of (2) are equivalent. The content of (3) is just a
reformulation of (2).
A surjective map p: X Y induces a bijection between subsets of Y and saturated subsets of X,
and it is a quotient map if and only if this bijection restricts to a bijection
_
Saturated open (closed) subsets
of X
_

Up(U)
p
1
(V )V

_
Open (closed) subsets
of Y
_
between open (or closed) subsets of Y and open (or closed) saturated subsets of X.
2.78. Corollary. (1) Any
_
open
closed
_
continuous surjective map is a quotient map.
(2) A quotient map f : X Y is
_
open
closed
_
if and only if all
_
open
closed
_
sets A X have
_
open
closed
_
saturations f
1
f(A).
(3) A bijective continuous map is a quotient map if and only if it is a homeomorphism.
Proof. (1) Let f : X Y be an open map. Then
f
1
(V ) is open
f open
= ff
1
(V ) is open V is open
f is continuous
= f
1
(V ) is open
for all V Y . This shows that f is quotient.
38 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
(2) Suppose that f : X Y is quotient. Then
f is open
(2.69)
f(A) is open in Y for all open sets A X
(2.76)
f
1
f(A) is open in X for all open sets A X
which is the claim.
(3) If f : X Y is a bijective, any subset of X is saturated, and if f is also quotient, then it determines a
bijection between the open (saturated) subsets of X and open subsets of Y . So f is a homeomorphism.
There are quotient maps that are neither open nor closed [8, Ex 22.3] [2, Ex 10 p 135] and there are
(non-identity) quotient maps that are both open and closed (2.83) [2, Ex 3 p 128].
2.79. Corollary (Composition of quotient maps). Let X
f

Y
g

Z be continuous maps.
Then
f and g are quotient = g f is quotient = g is quotient
Proof. The rst assertion is a tautology: Assume that f and g are quotient. Then
(g f)
1
(V ) open in X f
1
g
1
(V ) open in X
f is quotient
= g
1
(V ) open in Y
g is quotient
= V open in Z
for any set V Y . Next, suppose that X
f
Y
g
Z is a quotient map. Then the last map g is surjective
and
g
1
(V ) is open in Y
f continuous
= (g f)
1
(V ) is open in X
g f quotient
= V is open in Z
for any set V Z.
2.80. Example. (1) The projection map
1
: RR R is (2.73.(1)) open, continuous, and sur-
jective so it is a quotient map. The restriction
1
[H (0, 0) is continuous and surjective, even
bijective, but it is not a quotient map (2.77) for it is not a homeomorphism: (0, 0) is open and
saturated in H (0, 0) but
1
((0, 0)) = 0 is not open.
Thus the restriction of a quotient map need not be a quotient map in general. On the positive side
we have
2.81. Proposition. The restriction-corestriction of a quotient map p: X Y to an open (or closed)
saturated subspace A X is a quotient map p(A)[p[A: A p(A).
Proof. (Similar to the proof of 2.70.) Let p: X Y be a quotient map and B Y an open set.
(The case where B is closed is similar.) The claim is that B[p[p
1
(B): p
1
(B) B is quotient. For any
U B the implications
p
1
(U) open in p
1
(B) (p
1
(B) is open)
= p
1
(U) open in X (p is quotient)
= U is open in Y
= U is open in B
show that B[p[p
1
(B): p
1
(B) B is quotient.
A typical situation is when R is an equivalence relation on the space X and X X/R is the map
that takes a point to its equivalence class. We call X/R with the quotient topology for the quotient space
of the equivalence relation R. A set of equivalence classes is an open subset of X/R if and only if the
union of equivalence classes is an open subset of X: V X/R is open

[x]V
[x] X is open. We
shall often say that X/R is the space obtained by identifying equivalent points of X.
A continuous map f : X Y respects the equivalence relation R if equivalent points have identical
images, that is if x
1
Rx
2
= f(x
1
) = f(x
2
). The quotient map X X/R respects the equivalence
relation R and it is the universal example of such a map.
2.82. Theorem (The universal property of quotient spaces). Let R be an equivalence relation on the
space X and let f : X Y be a continuous map.
(1) The map p: X X/R respects the equivalence relation R.
7. THE QUOTIENT TOPOLOGY 39
(2) If the continuous map f : X Y respects R then there exists a unique continuous map f : X/R Y
such that
X
f

A
A
A
A
A
A
A
A
Y
X/R
f

}
}
}
}
}
}
}
commutes. (We say that f factors uniquely through X/R.) Conversely, if f factors through
X/R then f respects R.
(3) If f exists then: f is quotient f is quotient
Proof. If f exists then clearly f respects R. Conversely, if f respects R then we can dene f[x] =
f(x) and this map is continuous by 2.75 and it is the only possibility. The rest follows from 2.79:
f is quotient pf is quotient f is quotient.
The theorem says that there is a bijective correspondence
_
Continuous maps X Y
that respect R
_

ff
gpg

_
Continuous maps
X/R Y
_
taking quotient maps to quotient maps.
2.83. Example (Orbit spaces for group actions). (1) Real projective n-space RP
n
is the quotient
space of S
n
by the equivalence relation with equivalence classes x, x S
n
. The quotient map
p: S
n
RP
n
is both open and closed since (2.78) the saturation U of an open (closed) set U S
n
is
open (closed) because x x is a homeomorphism. Elements of RP
n
can be thought of as lines through
the origin of R
n+1
. A set of lines is open if the set of intersection points with the unit sphere is open.
(2) More generally, let G X X be the action of a discrete group G on a space X. Give the orbit
space GX the quotient topology and let p
G
: X GX be the quotient map. The points in the orbit
space are orbits of points in X and the open subsets are orbits of open subsets of X. The saturation of
any subset A of X is the orbit GA =

gG
gA of A. If A is open, GA is open as a union of open sets; if
A is closed and G is nite, GA is closed. Thus the quotient map p
G
is always open (2.78), and if G is
nite, it is also closed.
(3) The n-dimensional Mobius band is the orbit space MB
n
= (1, 1))(S
n1
R) for the action
(x, t) (x, t) of the group with two elements. (Take n = 2 to get the standard Mobius band.)
The (n 1)-dimensional real projective space RP
n1
is a retract of MB
n
as there are continuous maps
RP
n1

MB
n
induced by the the maps S
n1

S
n1
R . The homeomorphism S
n1
R
S
n
N, S between the cylinder over S
n1
and S
n
with two points removed induces a homeomorphism
between MB
n
and RP
n
with one point removed.
2.84. Example. (1) Let f : X Y be any surjective continuous map. Consider the equivalence
relation corresponding to the partition X =

yY
f
1
(y) of X into bres f
1
(y), y Y . Let X/f
denote the quotient space. Thus X/f is the set of bres equipped with the quotient topology. By
constuction, the map f respects this equivalence relation so there is a unique continuous map f such
that the diagram
X

C
C
C
C
C
C
C
C
f

Y
X/f
f

{
{
{
{
{
{
{
{
commutes. Note that f is bijective. The bijective continuous map f : X/f Y is a homeomorphism if
and only if f is quotient (2.78.(3), 2.82). In particular, all quotient maps have (up to homeomorphism)
the form X X/R for some equivalence relation R on X.
(2) Let f : X Y be any surjective continuous map. The induced map f : X/f Y is a continuous
bijection but in general not a homeomorphism. Instead, the topology on the quotient space X/f
is ner than the topology on Y because the quotient topology is the maximal topology so that the
projection map is continuous. This can sometimes be used to show that X/f is Hausdor if Y is
Hausdor.
(3) Let f : X Y be any continuous map. Then f has a canonical decomposition
X
p
X/f
f
f(X)

Y
40 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
where p is a quotient map, f is a continuous bijection, and is an inclusion map.
(4) Let X be a topological space and A
1
, A
2
, . . . , A
k
a nite collection of closed subsets. Consider the
equivalence relation where the equivalence classes are the sets A
1
, A
2
, . . . , A
k
together with the sets
x for x , A
1
A
2
A
k
. The quotient space X/(A
1
, . . . , A
k
) is obtained from X by identifying
each of the sets A
i
to the point p(A
i
). The quotient map p: X X/(A
1
, . . . , A
k
) is closed because
(2.78) closed sets A X have closed saturations A

A
i
A=
A
i
. A continuous map f : X Y factors
through the quotient space X/(A
1
, . . . , A
k
) if and only if it sends each of the sets A
i
X to a point in Y
(2.82). The restriction p[X(A
1
A
k
): X(A
1
A
k
) X/(A
1
, . . . , A
k
)p(A
1
), . . . , p(A
k
)
to the complement of A
1
A
k
is a homeomorphism (2.70). In case of just one closed subspace
A X, the quotient space is denoted X/A.
(5) The standard map f : [0, 1] S
1
that takes t [0, 1] to (cos(2t), sin(2t)) is quotient because it
is continuous and closed. (If you cant see this now, we will prove it later (2.140.(1)).) The induced
map [0, 1]/0, 1 S
1
is a homeomorphism. More generally, the standard map D
n
/S
n1
S
n
is a
homeomorphism where D
n
R
n
, the unit disc, is the set of vectors of length 1.
(6) Let R be the equivalence relation zero or not zero on R. The quotient space R/R is homeomor-
phic to Sierpinski space 0, 1.
(7) There is an obvious continuous surjective map f :

Z
+
S
1
= Z
+
S
1

C
n
(2.41.(2)) that takes
Z
+
1 to the point common to all the circles. This map is continuous because its restriction to each of
the open sets nS
1
is continuous (2.53.(5). However, f is not a quotient map (2.77) for the image of
the closed saturated set consisting of the points n(cos(

2
), sin(

2
)) is not closed as it does not contain
all its limit points. The induced bijective continuous map f : Z
+
S
1
/Z
+
1

C
n
is therefore not
a homeomorphism. There is an obvious continuous surjective map g :

Z
+
S
1
= Z
+
S
1

C
1/n
(2.41.(2)) that takes Z
+
1 to the point common to all the circles. This map is continuous because
its restriction to each of the open sets n S
1
is continuous (2.53.(5)). However, g is not a quotient
map for the image of the closed saturated set Z
+
1 is not closed as it does not contain all its
limit points. The induced bijective continuous map g : Z
+
S
1
/Z
+
1

C
1/n
is therefore not
a homeomorphism either. (Actually (2.98.(7)), the quotient space Z
+
S
1
/Z
+
1, known as the
countable wedge of circles
_
nZ
+
S
1
[8, Lemma 71.4], is not homeomorphic to any subspace of the
plane.)
(8) [8, p 451] Let P
4g
be a regular 4g-gon and with edges labeled a
1
, b
1
, a
1
, b
1
, . . . , a
g
, b
g
, a
g
, b
g
in
counter-clockwise direction. The closed orientable surface M
g
of genus g 1 is (homeomorphic
to) the quotient space P
4g
/R where R is the equivalence relation that makes the identications
a
1
b
1
a
1
1
b
1
1
a
g
b
g
a
1
g
b
1
g
on the perimeter and no identications in the interior of the polygon. See
[8, p 452] for the case g = 2. Are any of these surfaces homeomorphic to each other? [8, Thm 77.5]
(9) [8, p 452] Let P
2g
be a regular 2g-gon with edges labeled a
1
, a
1
, . . . , a
g
, a
g
in counter-clockwise
direction. The closed non-orientable surface N
g
of genus g 1 is the quotient space P
2g
/R where R is
the equivalence relation that makes the identications a
2
1
a
2
g
on the perimeter and no identications
in the interior of the polygon. For g = 1 we get the projective plane RP
2
and for g = 2 we get the
Klein Bottle [8, Ex 74.3].
2.85. Example. (The adjunction space) [8, Ex 35.8] [5, p 93] [10, Chp 1, Exercise B p 56] Consider
the set-up X A
?
_ i

Y consisting of a space X and a continuous map f : A Y dened on the

closed subspace A X. Let R be the smallest equivalence relation on X H Y such that aRf(a) for all
a A; the equivalence classes of R are a H f(a) for a A, x for x X A, and f
1
(y) H y for
y Y . The adjunction space is the quotient space
X
f
Y = X HY/R
for the equivalence relation R. Let f : X X
f
Y be the map X X H Y X
f
Y and let
i : Y X
f
Y be the map Y X HY X
f
Y . These two continuous maps agree on A in the sense
that f i = i f and the adjunction space is the universal space with this property. For any other space
Z receiving maps X Z Y that agree on A there exists a unique continuous map X
f
Y Z such
7. THE QUOTIENT TOPOLOGY 41
that the diagram
A
_

i

Y
i

X
f

X
f
Y
!

Z
commutes. The map i : Y X
f
Y is closed, for closed sets B Y X H Y have closed saturations
f
1
(B)HB. Since i is injective it is an embedding; its image is a closed subspace of X
f
Y homeomorphic
to Y . The map f[X A: X A X
f
Y is open, for open sets U X A X H Y have open
saturations U H . Since f[X A is also injective, it is an embedding; its image is an open subspace of
X
f
Y homeomorphic to X A. The quotient map X H Y X
f
Y is closed if the map f is closed
for then also closed subsets B X X H Y have closed saturations B f
1
f(B A) H f(A B). We
shall later see [8, Ex 35.8] that X
f
Y is normal when X and Y are normal.
Only few topological properties are preserved by quotient maps. The reason is that surjective open
maps and surjective closed maps are quotient maps so that any property invariant under quotient maps
must also be invariant under both open and closed maps.
As we saw in 2.84.(6) the quotient space of a Hausdor space need not be Hausdor, not even T
1
. In
general, the quotient space X/R is T
1
if and only if all equivalence classes are closed sets. (For instance,
if X and Y are T
1
then also the adjunction space X
f
Y is T
1
.) The quotient space X/R is Hausdor if
and only if any two distinct equivalence classes are contained in disjoint open saturated sets. We record
an easy criterion for Hausdorfness even though you may not yet know the meaning of all the terms.
2.86. Proposition. If X is regular and A X is closed then the quotient space X/A is Hausdor.
The product of two quotient maps need not be a quotient map [2, I.5.3] in general but here is an
important case where it actually is the case.
2.87. Theorem. Let p: A B and q : C D be quotient maps. If B and C are locally compact
Hausdor spaces (2.170) then p q : AC B D is a quotient map.
Proof. Using Lemma 2.88 below we see that the map p q is the composition
AC
p1

B C
1q

B D
of two quotient maps and therefore itself a quotient map.
2.88. Lemma (Whitehead Theorem). [5, 3.3.17] Let p: X Y be a quotient map and Z a locally
compact space. Then
p 1: X Z Y Z
is a quotient map.
Proof. Let A X Z. We must show: (p 1)
1
(A) is open = A is open. This means that for
any point (x, y) (p 1)
1
(A) we must nd a saturated neighborhood U of x and a neighborhood V of
y such that U V (p 1)
1
(A).
Since (p1)
1
(A) is open in the product topology there is a neighborhood U
1
of x and a neighborhood
V of y such that U
1
V (p 1)
1
(A). Since Y is locally compact Hausdor we may assume (2.170)
that V is compact and U
1
V (p1)
1
(A). Note that also p
1
(pU
1
) V is contained in (p1)
1
(A).
The tube lemma 2.143 says that each point of p
1
(pU
1
) has a neighborhood such that the product of
this neighborhood with V is contained in the open set (p 1)
1
(A). Let U
2
be the union of these
neighborhoods. Then p
1
(pU
1
) U
2
and U
2
V (p 1)
1
(A). Continuing inductively we nd open
sets U
1
U
2
U
i
U
i+1
such that p
1
(pU
i
) U
i+1
and U
i+1
V (p 1)
1
(A). The
open set U =

U
i
is saturated because U p
1
(pU) =

p
1
(pU
i
)

U
i+1
= U. Thus also U V is
saturated and U V

U
i
V (p 1)
1
(A).
For instance, if p: X Z is a quotient map, then also p 1: X [0, 1] Z [0, 1] is a quotient
map. This is important for homotopy theory.
42 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
8. Metric topologies
If X is a set with a metric d: X X [0, ) the collection B
d
(x, ) [ x X, > 0 of balls
B
d
(x, ) = y X [ d(x, y) < is a basis for the metric topology T
d
induced by d.
2.89. Definition. A metric space is the topological space associated to a metric set. A topological
space is metrizable if the topology is induced by some metric on X.
Hausdor dimension, fractals, or chaos are examples of metric, rather than topological, concepts.
2.90. Theorem (Continuity in the metric world). Let f : X Y be a map between metric spaces
with metrics d
X
and d
Y
, respectively. The following conditions are equivalent:
(1) f is continuous
(2) x X > 0 > 0y X: d
X
(x, y) < d
Y
(f(x), f(y)) <
(3) x X > 0 > 0: f(B
X
(x, )) B
Y
(f(x), )
Proof. Essentially [8, Ex 18.1].
2.91. Proposition (Comparison of metric topologies). Let d and d

be metrics on X and T
d
, T
d
the
associated metric topologies. Then
T
d
T
d
x X > 0 > 0: B
d
(x, ) B
d
(x, )
Proof. T
d
T
d
if and only if the identity map (X, T
d
) (X, T
d
) is continuous [8, Ex 18.3] .
2.92. Lemma (Standard bounded metric). Let d be a metric on X. Then d

(x, y) = mind(x, y), 1

is a bounded metric on X (called the standard bounded metric corresponding to d) that induce the same
topology on X as d.
Proof. Either use the above proposition or use that for any metric the collection of balls of radius
< 1 is a basis for the metric topology. These bases are the same for the two metrics.
2.93. Theorem (Hereditary properties of metrizable spaces). Any subspace of a metrizable space is
a metrizable. Any countable product of metrizable spaces is metrizable.
Proof. See [8, Ex 21.1] for the rst assertion. To prove the second assertion, let X
n
, n Z
+
, be a
countable collection of metric spaces. We may assume that each X
n
has diameter at most 1 (2.92). Put
d((x
n
), (y
n
)) = sup
1
n
d
n
(x
n
, y
n
) [ n Z
+

for points (x
n
) and (y
n
) of

X
n
and convince yourself that d is a metric. The idea here is that
1
n
d
n
(x
n
, y
n
)
1
n
becomes small when n becomes large. For any > 0
d((x
n
), (y
n
)) n N:
1
n
d
n
(x
n
, y
n
)
where N is such that N > 1.
The claim is that the metric toplogy coincides with the product topology on

X
n
. We need to
show that the metric topology enjoys the two properties that characterizes the product topology (2.64).
First, the projection maps
n
:

X
n
X
n
are continuous because d(x, y) < = d
n
(x, y) < n (2.90).
Second, let f : X

X
n
be a map such that X
f

X
n

n
X
n
is continuous for all n. Given x X
and > 0, there exist neighborhoods U
n
of x such that d
n
(
n
f(x),
n
f(y)) < n for all y U
n
. Then
d(f(x), f(y)) < for all y U
1
. . . U
N
where N > 1 (remember that all the spaces X
n
have diameter
at most 1). This shows that f : X

X
n
is continuous.
2.94. The rst countability axiom. Which topological spaces are metrizable? To address this
question we need to build up an arsenal of metrizable and non-metrizable spaces and to identify properties
that are common to all metrizable spaces. Here are the rst such properties: All metric spaces are
Hausdor and rst countable.
2.95. Definition (Neighborhood basis). A neighborhood basis at x X is a collection of neighbor-
hoods of x such that any neighborhood of x contains a member of the collection.
2.96. Definition (First countable spaces). Let X be a space and x a point in X. We say that X has
a countable basis at x if there is a countable neighborhood basis at x. X is rst countable if all points of
X have a countable neighborhood basis.
All metrizable spaces are rst countable since B(x, 1/n) [ n Z
+
is a countable neighborhood
basis at x.
8. METRIC TOPOLOGIES 43
2.97. Proposition (Hereditary properties of rst countable spaces). [8, Thm 30.2] Any subspace of
a rst countable space is rst countable. Any countable product of rst countable spaces is rst countable.
Proof. The rst assertion is immediate. Let

X
n
be a countable product of rst countable spaces.
Let (x
n
) be a point of

X
n
. Let B
n
be countable basis at x
n
X
n
. The collection of all products

U
n
where U
n
B
n
for nitely many n and U
n
= X
n
for all other n is then a countable (1.41) basis at
(x
n
).
2.98. Example. (1) R is rst countable because it is a metric space.
(2) R

is rst countable. The collection of half-open intervals [a, b) where b > a is rational is a
countable basis of neighborhoods at the point a. Is R

metrizable? (3.3) [8, Ex 30.6]

(3) Is R
K
rst countable? Is it metrizable?
(4) The ordered space S

= [0, ] (1.52) is not rst countable at but it is rst countable at any

other point. Let U
n
be any countable collection of neighborhoods of . Choose a
n
< such that
(a
n
, ] U
n
and chose a such that a
n
< a < for all n. This is possible because the countable set
a
n
[0, ) of left end-points has an upper bound in [0, ) (1.52.(2)). Then (a, ] is a neighborhood
of that does not contain any of the U
n
because it does not even contain any of the intervals (a
n
, ].
The section S

, however, is rst countable: 1 = [1, 2) is open so S

has a nite basis at the rst

element 1. For any other element, > 1, we can use the countable collection of neighborhoods of the
form (, ] for < (2.31.(5)). Is S

metrizable? (2.164)[8, Exmp 3 p 181, Ex 30.7].

(5) R
J
is not rst countable at any point when J is uncountable: Let U
n

nZ
+
be any countable
collection of neighborhoods of, say, the point (0)
jJ
of R
J
. I claim that there is a neighborhood that
does not contain any of the U
n
. This is because there is an index j
0
J such that
j
0
(U
n
) = R for
all n. Indeed, the set of js for which this is not true
j J [ n Z
+
:
j
(U
n
) ,= R =
_
nZ
+
j J [
j
(U
n
) ,= R
is a countable union of nite sets, hence countable (1.41.(3)). Then the neighborhood
1
j
0
(R1)
of (0)
jJ
does not contain any of the neighborhoods U
n
in the countable collection. (R
J
does not
even satisfy the sequence lemma (2.100.(1)) [8, Exmp 2 p 133].)
(6) The closed (2.84.(4)) quotient map R R/Z takes the rst countable space R to a space that is
not rst countable [5, 1.4.17] at the point corresponding to Z. This can be seen by a kind of diagonal
argument: Let U
n

nZ
be any countable collection of open neighborhoods of Z R. Let U be the
open neighborhood of Z such that U (n, n+1), n Z, equals U
n
(n, n+1) with one point deleted.
Then U does not contain any of the U
n
.
(7) The quotient map p:

nZ
+
S
1

_
nZ
+
S
1
is closed (2.84.(4)). The domain

nZ
+
S
1
= Z
+

S
1
R R
2
is rst countable (2.97) but the image
_
nZ
+
S
1
(2.84.(7)) is not: Let U
n

nZ
+
be
any collection of saturated neighborhoods of Z
+
1. The saturated neighborhood U which at level
n equals U
n
with one point deleted (cf Cantors diagonal argument (1.43)) does not contain any of
the U
n
. It follows (2.97) that
_
nZ
+
S
1
does not embed in R
2
nor in any other rst countable space.
For instance, the universal property of quotient spaces (2.82) gives a factorization
(2.99)

nZ
+
S
1
f

K
K
K
K
K
K
K
K
K
K

nZ
+
S
1
_
nZ
+
S
1
f

s
s
s
s
s
s
s
s
s
s
of the continuous map f such that
m
(f[n S
1
) is the identity function when m = n and the
constant function when m ,= n. The induced map f is an injective continuous map. It can not be an
embedding for the countable product

S
1
of circles is rst countable (2.97). The topology on
_
S
1
is ner than the subspace topology inherited from

S
1
.
These examples show that the uncountable product of rst countable (even metric) spaces and the
quotient of a rst countable space may fail to be rst countable (2.98.(5), 2.98.(7)). Some linearly ordered
spaces are rst countable, some are not (2.98.(1), 2.98.(4)).
In a rst countable (eg metric) space X, the points of the closure of any A X can be approached
by sequences from A in the sense that they are precisely the limit points of convergent sequences in A.
This is not true in general (2.98.(4)).
2.100. Lemma. Let X be a topological space.
44 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
(1) (The sequence lemma) Let A X be a subspace and x a point of X. Then
x is the limit of a sequence of points from A = x A
The converse holds if X is rst countable.
(2) (Continuous map preserve convergent sequences) Let f : X Y be a map of X into a space Y .
Then
f is continuous = f(x
n
) f(x) whenever x
n
x for any sequence x
n
in X
The converse holds if X is rst countable.
Proof. (1) The direction = is clear. Conversely, suppose that x A. Let U
n
be a countable basis
at x. We may assume that U
1
U
2
U
n
U
n+1
as we may replace U
n
by U
1
. . . U
n
.
For each n choose a point x
n
A U
n
. We claim that the sequence (x
n
) converges to x. Let U be any
neighborhood of x. Then U
n
U for some n so that x
m
U
m
U
n
U for m n.
(2) The direction = is clear. Conversely, suppose that f(x
n
) f(x) whenever x
n
x. We want to
show that f is continuous, ie (2.51.(6)) that f(A) f(A) for any A X. Let x A. Since X is rst
countable, there is by (1) a sequence of points a
n
A converging to x. By hypothesis, the sequence
f(a
n
) f(A) converges to f(x). Thus f(x) f(A) by (1) again.
We say that X satises the sequence lemma
1
if for any A X and for any x A there is a sequence
of points in A converging to x.
To summarize:
X is metrizable = X is 1st countable = X satises the sequence lemma (is Frechet)
Examples show that neither of these arrows reverse.
The largest element is a limit point of [0, ) but it is not the limit of any sequence in [0, ) as any
such sequence has an upper bound in [0, ) (1.52.(2)). Thus [0, ] does not satisfy the sequence lemma.
Hence it is not 1st countable and not metrizable.
2.101. The uniform metric. Let Y be a metric space with a bounded metric d and let J be a set.
We shall discuss uniform convergence which is a metric, not a topological, concept.
2.102. Definition. [8, pp 124, 266] The uniform metric on Y
J
=

jJ
Y = map(J, Y ) is the metric
given by d(f, g) = supj J [ d(f(j), g(j)).
It is easy to see that this is indeed a metric.
2.103. Theorem. On R
J
we have: product topology uniform topology box topology
Proof. Omitted.
The elements of Y
J
are functions f : J Y . Note that a sequence of functions f
n
: J Y converges
in the uniform metric to the function f : J Y if and only if
> 0N > 0j Jn Z
+
: n > N = d(f
n
(j), f(j)) <
We say that the sequence of functions f
n
: J Y converges uniformly to the function f : J Y .
2.104. Theorem (Uniform limit theorem). (Cf [8, Thm 43.6]) Suppose that J is a topological space
and that Y is a metric space. The uniform limit of any sequence (f
n
) of continuous functions f
n
: J Y
is continuous.
Proof. Well-known.
1
Aka a Frechet space
9. CONNECTED SPACES 45
9. Connected spaces
2.105. Definition. The topological space X is connected if it is not the union X = X
0
X
1
of two
disjoint open non-empty subsets X
0
and X
1
.
Two subsets A and B of a space X are separated if AB = = AB. This means that the two sets
are disjoint and neither contains a limit point of the other. Two disjoint open (closed) sets are separated.
If C A and D B and A and B are separated, then C and D are separated. A separation of X
consists of two separated non-empty subsets A and B with union X = A B.
2.106. Theorem. The following are equivalent:
(1) X is connected
(2) The only clopen (closed and open) subsets of X are and X
(3) X has no separations
(4) Every continuous map X 0, 1 to the discrete space 0, 1 is constant.
Proof. We show that the negated statements are equivalent.
(1) = (2): Suppose that X = U
1
U
2
where U
1
and U
1
are disjoint, open, and non-empty. Then U
1
is an open, closed, non-empty, proper subset of X.
(2) = (3): If C is a closed, open, nonempty, proper subset of X then X = C(XC) is a separation
of X.
(3) = (4): Suppose that X = A B where A and B are separated. Then A A, for A does not
meet B, so A is closed. The map f : X 0, 1 given by f(A) = 0 and f(B) = 1 is continuous since all
closed subsets of the co-domain have closed pre-images.
(4) = (1): Let f : X 0, 1 be a surjective continuous map. Then X = f
1
(0) f
1
(1) is the
union of two disjoint non-empty open subsets.
2.107. Theorem. If X is connected, then f(X) is connected for any continuous map f : X Y .
Proof. We use the equivalence of (1) and (4) in Theorem 2.106. Let f : X Y be a continuous
map. If f(X) is not connected, there is a non-constant continuous map f(X) 0, 1 and hence a
non-constant continuous map X 0, 1. So X is not connected.
2.108. Example. (1) R0 = R

R
+
is not connected for it is the union of two disjoint open
non-empty subsets.
(2) We shall later prove that R is connected (2.119) and that the connected subsets of R are precisely
the intervals, rays, R, and .
(3) R

is not connected for [a, b) is a closed and open subset whenever a < b (2.31.(4)). In fact, any
subset Y of R

containing at least two points a < b is disconnected as Y [a, b) is closed and open
but not equal to or Y . (R

is totally disconnected.)
(4) R
K
is connected [8, Ex 27.3].
(5) Q is totally disconnected (and not discrete): Let Y be any subspace of Q containing at least two
points a < b. Choose an irrational number t between a and b. Then Y (t, ) = Y [t, ) is an
open, closed, non-empty, proper subset of Y .
(6) Particular point topologies (2.2.(3)) are connected.
A subspace of X is said to be connected if it is connected in the subspace topology. A subspace of a
connected space need obviously not be connected. So how can we tell if a subspace is connected?
2.109. Lemma. Let Y X be a subspace. Then
Y is connected Y is not the union of two separated non-empty subsets of X
Proof. Suppose that Y = Y
1
Y
2
is the union of two subspaces. Observe that
(2.110) Y
1
and Y
2
are separated in Y Y
1
and Y
2
are separated in X
because Cl
Y
(Y
1
) Y
2
2.35.(2)
= Y
1
Y Y
2
= Y
1
Y
2
. The lemma now follows immediately from 2.106, the
equivalence of (1) and (3).
2.111. Corollary. Suppose that Y X is a connected subspace. For every pair A and B of
separated subsets of X such that Y A B we have either Y A or Y B.
Proof. The subsets Y A and Y B are separated (since the bigger sets A and B are separated)
with union Y . By 2.109, one of them must be empty, Y A = , say, so that Y B.
46 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
2.112. Corollary. The closure of any connected subspace is connected. Indeed, if C Y C and
C is connected, then Y is connected.
Proof. Let f : Y 0, 1 be a continuous map. Then f(C) is a single point since C is connected.
Moereover, f(y) = f(C) for all y Y since otherwise we could separate y and C by disjoint open sets.
Thus f is constant.
2.113. Theorem. Let Y
j
[ j J be a set of connected subspaces of the space X. Suppose that there
is an index j
0
J such that Y
j
and Y
j
0
are not separated for any j J. Then the union

jJ
Y
j
is
connected.
Proof. Let f :

jJ
Y
j
0, 1 be a continuous map. The image f(Y
j
) of Y
j
is a single point for
each j J since Y
j
is connected. In fact, f(Y
j
) = f(Y
j
0
) as otherwise we could separate Y
j
and Y
j
0
by
disjoint open sets. Thus f is constant.
2.114. Corollary. The union of a collection of connected subspaces with a point in common is
connected.
Proof. Apply 2.113 with any of the subspaces as Y
j
0
.
2.115. Corollary. Suppose that for any two points in X there is a connected subspace containing
both of them. Then X is connected.
Proof. Let x
0
be some xed point of X. For each point x X, let C
x
be a connected subspace
containing x
0
and x. Then X =

C
x
is connected as

C
x
,= (2.114).
2.116. Theorem. Products of connected spaces are connected.
Proof. We prove rst that the product X Y of two connected spaces X and Y is connected. In
fact, for any two points (x
1
, y
1
) and (x
2
, y
2
) the subspace X y
1
x
2
Y contains the two points
and this subspace is connected since (2.114) it is the union of two connected subspaces with a point in
common. Thus X Y is connected by Corollary 2.115.
Next, induction shows that the product of nitely many connected spaces is connected.
Finally [8, Ex 23.10], consider an arbitrary product

X
j
of connected spaces X
j
, j J. Choose
(1.28) a point x
j
in each of the spaces X
j
(assuming that all the spaces of the product are non-empty).
For every nite subset F of J let C
F


X
j
be the product of the subspaces X
j
if j F and x
j

if j , F. Since C
F
is connected for each nite subset F and these subsets have the point (x
j
)
jJ
in
common, the union

FF
C
F
, where T is the collection of all nite subsets of J, is connected (2.114).
This union is not all of

X
j
but its closure is, so

FF
C
F
=

jJ
X
j
is connected (2.112).
2.117. Connected subspaces of linearly ordered spaces. We determine the connected subsets
of R, or, more generally, of any linear continuum.
Recall that a subset C of a linearly ordered set X is convex if a, b C = [a, b] C. Connected
subspaces are convex. But are convex subspaces connected? Not always: The convex subset [0, 1] is not
connected in Z but it is connected in R.
2.118. Lemma. Suppose that X is a linear continuum (1.15). Then
connected subsets of X = convex subsets of X
The inclusion holds in any linearly ordered space.
Proof. Suppose rst that X is any linearly ordered space and let C X a subspace that is not
convex. Then there exist points a < x < b in X such that a and b are in C and x is outside C. Since
C (, x) (x, +) is contained in the union of two separated subsets and meet both of them, C is
not connected (2.111).
Let now X be a linear continuum and C a convex subset of X. We claim that C is connected. Pick
a xed point a C and note that C is a union of closed intervals with a as one of their end-points.
Therefore it suces (2.114) to prove the claim in case C = [a, b] is a closed interval. Suppose that
[a, b] = AB is the union of two disjoint relative open nonempty subsets A. We may assume that b B.
The sets A and B are closed and open in [a, b]. As the closed interval [a, b] is closed in X (2.13), A and
B are also closed in X (2.35). The nonempty bounded set A has a least upper bound, c = sup A. Now,
c b since b is an upper bound and c A since A is closed in X. (The least upper bound of any bounded
set always belongs to the closure of the set since otherwise it wouldnt be the least upper bound.) So
c < b (for b , A) and A [a, b). But A is also open in [a, b] and in the subspace [a, b) (which has the
9. CONNECTED SPACES 47
subspace topology which is the order topology (2.27)). This contradicts c A for no element of an open
subset A of [a, b) can be an upper bound for A: For any point d A there exists an open interval (x, y)
around d such that [a, b) (x, y) A and since [d, y) ,= , d is not an upper bound for A. (In short, c A
since A is closed in X and c , A since A is open in [a, b).) This is a contradiction and therefore A must
be empty.
The inclusion does not hold for all linearly ordered spaces as for instance Z
+
is convex but not
connected (there are gaps).
In particular, the connected subsets of R are precisely the convex subsets which are , R, and all
intervals (bounded or unbounded, closed, open or halfopen).
We shall now identify the linearly ordered spaces that are connected.
2.119. Theorem. [2, Ex 7 p 382, Prop 1 p 336] Let X be a linearly ordered space. Then
X is connected in the order topology X is a linear continuum
The connected subsets of a linear continuum X are: X, , intervals and rays.
Proof. =: [8, Ex 24.4] [5, Problem 6.3.2] [2, Ex 7 p 382]. Suppose that X is a linearly ordered
set that is not a linear continuum. Then there are nonempty, proper, clopen subsets of X:
If (x, y) = for some points x < y then (, x] = (, y) is clopen and ,= , X.
If A X is a nonempty subset bounded from above which has no least upper bound then the
set of upper bounds B =

aA
[a, ) =

bB
(b, ) is clopen and ,= , X.
Therefore X is not connected (2.106).
=: Assume X is a linear continuum. From 2.118 we know that the the connected and the convex
subsets of X are the same. In particular, the linear continuum X, certainly convex in itself, is connected
in the order topology. Let C be a nonempty convex subset of X. We look at two cases:
C is neither bounded from above nor below. Let x be any point of X. Since x is neither a lower
nor an upper bound for C there exist a, b C so that a < x < b. Then x C by convexity.
Thus C = X.
C is bounded from above but not from below. Let c = sup C be its least upper bound. Then
C (, c]. Let x < c be any point. Since x is neither a lower nor an upper bound for C
there exist a, b C so that a < x < b. Then x C by convexity. Thus (, c) C (, c]
and C is either (, c) or (, c].
The arguments are similar for the other cases. Recall that X also has the greatest lower bound property
[8, Ex 3.13].
The real line R, the ordered square I
2
o
(2.15.(7)), the (ordinary) (half) line [1, ) = Z
+
[0, 1)
(1.49.(6)), and the long (half) line [0, ) [0, 1) (1.52) [8, Ex 24.6, 24.12] are examples of linear continua.
2.120. Theorem. [Intermediate Value Theorem] Let f : X Y be a continuous map of a connected
topological space X to a linearly ordered space Y . Then f(X) is convex. If Y is a linear continuum,
f(X) is an interval (bounded or unbounded, closed, half-open, or open)
Proof. X connected
(2.107)
= f(X) connected
(2.119)
= f(X) convex. For subsets of a linear continuum
we know (2.119) that connected = convex = interval.
Any linearly ordered space containing two consecutive points, two points a and b with (a, b) = is
not connected as X = (, b) (a, +) is the union of two disjoint open sets.
Any well-ordered set X containing at least two points is totally disconnected in the order topology.
For if C X contains a < b then a , C (a, b] b is closed and open in C since (a, b] is closed and open
in X.
2.121. Path connected spaces. Path connectedness is a stronger property than connectedness.
2.122. Definition. The topological space X is path connected if for any two points x
0
and x
1
in X
there is a continuous map (a path) f : [0, 1] X with f(0) = x
0
and f(1) = x
1
.
The image under a continuous map of a path connected space is path connected, cf 2.107. Any
product of path connected spaces is path connected [8, Ex 24.8], cf 2.113.
2.123. Example. The punctured euclidian plane R
n
0 is path connected when n 2 since any
two points can be joined by a path of broken lines. Thus also the (n1)-sphere S
n1
, which is the image
of R
n
0 under the continuous map x x/[x[, is path connected for (n 1) 1.
48 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
Since the unit interval I is connected (2.119), all paths f(I) are also connected (2.107) so all path
connected spaces are, as unions of paths emanating from one xed point, connected (2.115). The converse
is not true, not all connected spaces are path connected.
2.124. Example. (1) (Topologists sine curve) Let S be the graph of the function sin(1/x), 0 <
x 1, considered as a subspace of the plane R
2
. The (closed) topologists sine curve is the subspace
S = S (0 [1, 1]) of the plane. It follows from 2.107 and 2.112 that S is connected. We shall
soon see that S is not path connected (2.133.(2)).
(2) The ordered square I
2
o
(2.15.(7)) is connected since it is a linear continuum [8, Ex 3.15] but it is
not path connected. For suppose that f : [0, 1] I
2
o
is a path from the smallest element (the lower
left corner) f(0) = 0 0 to the largest element (the upper right corner) f(1) = 1 1. Then f is
surjective for the image contains (2.120) the interval [0 0, 1 1] = I
2
o
. But this is impossible since
the ordered square I
2
o
contains uncountably many open disjoint subsets (eg x(0, 1) = (x0, x1),
0 x 1) but [0, 1] does not contain uncountably many open disjoint subsets (choose a rational
number in each of them).
(3) R
K
(2.11) is connected but not path-connected space [8, Ex 27.3].
(4) The Stone

Cech compactication R of R is connected but not path-connected [8, Ex 38.9].

2.125. Components and path components. The relations on X dened by
x y There is a connected subset C X such that x C and y C
x
p
y There is a path in X between x and y
are equivalence relations. Check transitivity! (You may need the Glueing lemma (2.53.(5)).)
2.126. Definition. The components of X are the equivalence classes of the rst equivalence relation.
The path components of X are the equivalence classes of the second equivalence relation
The component containing the point x X is, by denition, the union of all connected subspaces
containing x. The path component of the point x X consists, by denition, of all points with a path
to x.
2.127. Theorem. (1) The components are connected closed disjoint subsets with union X. Any
connected subset of X is contained in precisely one of the components.
(2) If X =

, where the C

are connected subspaces such that for all ,= there is a separation

A B of X with C

A and C

B, then the C

are the components of X.

(3) The path components of X are path connected disjoint subsets with union X. Any path connected
subset is contained in precisely one of the path components.
(4) The path components are connected. The components are unions of path components.
Proof. (1) The components form a partition of X since they are equivalence classes (1.11). The
components are connected by the denition of the relation and 2.115. All the points of a connected
subset are equivalent so they are contained in the same equivalence class. In particular (2.115), C C
for any component C so C is closed.
(2) Since each C

is connected, it is contained in precisely one component (2.127.(1)). Could there be

two of the C

s, say C

and C

, inside one component, say C? No: Choose a separation X = AB such

that C

A and C

B. Then C X = AB and C meets both A and B which is impossible (2.111).

(3) The path components are path-connected by their very denition. All path-connected spaces are
connected, so each path-component is contained in precisely one component. Thus the components are
unions of path-components.
If X has only nitely many components the components are closed and open. If X = C
1
C
n
where the nitely many subspaces C
i
are connected and separated then the C
i
are the components of X.
The path components can be closed, open, closed and open, or neither closed nor open. However, for
locally connected spaces (2.129) components and path components coincide (2.131).
2.128. Locally connected and locally path connected spaces.
2.129. Definition. The space X is locally (path) connected at the point x if every neighborhood of
x contains a (path) connected neighborhood of x. The space X is locally (path) connected if it is locally
(path) connected at each of its points.
Thus a space is locally (path) connected i it has a basis of (path) connected subsets.
Consider this table
9. CONNECTED SPACES 49
R R0 S Q
Connected YES NO YES NO
Locally connected YES YES NO NO
before you draw any conclusions about the relationship between connected and locally connected spaces.
Note in particular that a space can be connected (even path connected (2.133)) and not locally connected.
On the positive side, note that open subsets of locally (path) connected spaces are locally (path)
connected and that quotient spaces of locally connected spaces are locally connected (2.132).
2.130. Proposition. The space X is locally (path) connected if and only if open subsets have open
(path) components (open in X or, what is the same (2.24), open in the open set). In a locally (path-)
connected space the (path-) components are clopen.
Proof. Assume that X is locally connected and let U be an open subset. Consider a component,
C, of U. The claim is that C is open. Let x be a point in C. Choose a connected neighborhood V of x
such that V U. Since V is connected and intersects C, V C (2.127). This shows that C is open. In
particular, the components (of the open set X) are open and, since they form a partition, also closed.
Conversely, assume that open subsets have open components. Let x be a point of X and U a neigh-
borhood of x. Let C be the component of U containing x. Then C is an open connected neighborhood
of x contained in U.
It follows for instance that the open subsets of R are unions of at most countably many open
intervals. For any open set is the union of its components which are open, since R is locally connected,
and connected, so they are open intervals. Since each of the open intervals contains a rational number,
there are at most countably many components. Any closed subset of R is the complement to a union of
at most countably many open intervals.
2.131. Theorem. In a locally path-connected space the path-components and the components are the
same.
Proof. Suppose that X is locally path-connected. Each path-component P is contained (2.127) in
a unique component, C. Since C is connected and P is clopen (2.130), P = C (2.106).
For instance, the components and the path components of a locally Euclidean space, such as a
manifold, are the same. A locally path-connected space is path-connected if and only if it is connected.
2.132. Proposition. [8, Ex 25.8] [2, I.11.6] Locally (path) connected spaces have locally (path)
connected quotient spaces.
Proof. Let X be a locally connected space and p: X Y a quotient map. Let V be a subset
of Y and C a component of V . Then p
1
(C) is a union of components of p
1
(V ) because continuous
maps preserve connectedness (2.107). If V is open in Y , p
1
(V ) is open in X, and since X is locally
connected it follows (2.130) that p
1
(C) is open in X. This means (2.74) that C is open in Y . Since thus
open subsets of Y have open components, Y is locally connected (2.130). (The same proof applies with
connected replaced by path connected.)
2.133. Example. (1) The comb space X = ([0, 1] 0) ((0 1/n [ n Z
+
) [0, 1]) R
2
is clearly path connected but it is not locally connected for (2.131) the open set U = X [0, 1] 0
has a component (use (2.127.(2)) to identify the components) that is not open. (The same is true for
any (small) neighborhood of (0, 1/2).) See [8, Ex 25.5] for a similar example.
(2) The closed topologists sine curve S is not locally connected: Let U be a small neighborhood
(well, not too big) around (0, 1/2). Then U has a component that is not open. For instance, U =
S ([0, 1] 0 has a component (2.127.(2)), 0(0, 1] that is not open (no neighborhood of (0,
1
2
)
is disjoint from S). It follows (2.132, 2.141.(2)) that S is not path-connected for it can not be a
quotient space of the locally connected compact space [0, 1]. In fact, S has two path-components:
S = S (R
+
R), which is open and not closed for its closure is S, and 0 [0, 1], which is closed
and not open for S is not closed. Since there are two path components and one component, S is not
locally path connected (2.131). (The Warsaw circle, the union of S with an arc from one end of S to
[0, 1]) is an example of a path connected not locally connected space, so is the next example.)
(3) The space X =

C
n
(2.41.(2)) is even path connected but still not locally connected for the
component R
+
0 of the open set X (0, 0) (use (2.127.(2)) to identify the components) is not
open (any neighborhood of (1, 0) contains points from the circles). Alternatively, X is not locally
connected since any small neighborhood of (1, 0) has a separation.
50 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
(4) Any linear continuum is locally connected since the basis for the topology consists of connected
sets (2.119). It need not be locally path connected [8, Ex 25.3].
(5) The (generalized) Jordan curve theorem says that R
n
h(S
n1
) has two path-components for any
embedding h: S
n1
R
n
of an (n1)-sphere into R
n
, n 1. This is usually proved using algebraic
topology.
Figure 1. The Warsaw circle
10. COMPACT SPACES 51
10. Compact spaces
2.135. Definition. A topological space X is compact if it satises one of the following two equivalent
conditions:
(1) Every collection of open subsets whose union is X contains a nite subcollection whose union
is X.
(2) Every collection of closed subsets whose intersection is contains a nite subcollection whose
intersection is .
A collection of open subsets whose union is the whole space is called an open covering. Thus a space
is compact if any open covering contains a nite open covering.
Any nite space is compact as there are only nitely many open subsets. Any compact discrete
space is nite. The real line R is not compact as the open covering (n, n) [ n Z
+
has no nite
subcovering.
How can we recognize the compact subspaces of a given topological space?
2.136. Proposition. Let Y be a subspace of X. The following are equivalent:
(1) Y is compact
(2) Every collection of open subsets of X whose union contains Y contains a nite subcollection
whose union contains Y
(3) Every collection of closed subsets of X whose intersection is disjoint from Y contains a nite
subcollection whose intersection is disjoint from Y
Proof. (1) = (2): Let U
j
[ j J be a collection of open subsets of X such that Y

U
j
.
Then Y U
j
[ j J is an open covering of Y . Since Y is compact, Y =

jJ

Y U
j
for some nite
index set J

J. This means that Y

jJ

U
j
.
(2) = (1): Let V
j
[ j J be an open covering of Y . Then V
j
= Y U
j
for some open set
U
j
X and Y

jJ
U
j
. By assumption, Y

jJ

U
j
for some nite index set J

J. Therefore
Y = Y

jJ

U
j
=

jJ

Y U
j
=

jJ

V
j
. This shows that Y is compact.
(2) (3): Clear from DeMorgans laws.
2.137. Theorem. Closed subspaces of compact spaces are compact.
Proof. Suppose that X is a compact space and Y X a closed subset. Let F
j
, j J, be a
collection of closed subsets of X such that Y

jJ
F
j
=

jJ
Y F
j
= . By compactness of X,
=

jJ

Y F
j
= Y

jJ

F
j
for some nite index set J

J. Thus Y is compact by 2.136.

2.138. Theorem. If X is compact, then f(X) is compact for any continuous map f : X Y .
Proof. Let U
j
, j j, be a collection of open subsets of Y that cover the image f(X). Then
X =

jJ
f
1
(U
j
) is an open covering of X. Since X is compact, X =

jJ

f
1
(U
j
) for some nite
index set J

J. Then f(X) = f
_
jJ

f
1
(U
j
)
_
=

jJ

ff
1
(U
j
)

jJ

U
j
. This shows that f(X)
is compact (2.135).
Compact subspaces of Hausdor spaces behave to some extent like points.
2.139. Theorem. (1) Compact subspaces of Hausdor spaces are closed.
(2) Any two disjoint compact subspaces of a Hausdor space can be separated by disjoint open sets.
Proof. Let L and K be two disjoint compact subspaces of X. Consider rst the special case where
L = x
0
is a point. For each point x K, the Hausdor property implies that there are disjoint open
sets U
x
, V
x
such that and x U
x
and x
0
V
x
. Since K is compact (2.136), K U
x
1
. . . U
x
t
for
nitely many points x
1
, . . . , x
t
K. Set U = U
x
1
. . . U
x
t
and V = V
x
1
. . . V
x
t
. The existence of
V alone says that K is closed.
Assume next that L is any compact subspace of X. We have just seen that for each point y L
there are disjoint open sets U
y
K and V
y
y. By compactness, L is covered by nitely many of the
V
y
. Then K is contained in the intersection of the corresponding nitely many U
y
.
2.140. Corollary. Let X be a compact Hausdor space.
(1) compact subspaces of X = closed subspaces of X
(2) If A and B are disjoint closed sets in X then there exist disjoint open sets U, V such that A U
and B V .
52 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
Proof. (1) is 2.139 and is 2.137.
(2) Let A and B be disjoint closed subsets of X. Then A and B are compact as just shown. Now apply
2.139.
2.141. Lemma (Closed Map Lemma). Suppose that X is a compact space, Y is a Hausdor space,
and f : X Y is a continuous map.
(1) f is closed (2.69).
(2) If f is surjective, it is a closed quotient map (2.76).
(3) If f is injective, it is an embedding (2.58).
(4) If f is bijective, it is a homeomorphism (2.55).
Proof. Let f : X Y be a continuous map of a compact space X into a Hausdor space Y .
(1) We have
C is closed in X
2.137
= C is compact
2.138
= f(C) is compact
2.139
= f(C) is closed in Y
which shows that f is a closed map.
(2) Every closed continuous surjective map is a closed quotient map (2.78).
(3) Every closed continuous injective map is an embedding (2.59).
(4) Every closed bijective continuous map is a homeomorphism (2.78.(3)).

2.142. Corollary. If f : X Y is continuous and X is compact and locally connected and Y is

Hausdor, then f(X) is compact and locally connected.
Proof. We know (2.141) that f : X f(X) is a (closed) quotient map. The quotient image of a
locally connected space is locally connected (2.132).
2.143. Lemma (The Tube Lemma). Let X and Y be topological spaces where Y is compact. Let x
0
be a point of X. For any neighborhood N X Y of the slice x
0
Y there is a neighborhood U of x
0
such that x
0
Y U Y N.
Proof. For each point y Y there is a product neighborhood such that x
0
y U
y
V
y
N.
Since Y is compact, there are nitely many points y
1
, . . . , y
k
Y such that Y = V
1
. . . V
k
where
V
i
= V
y
i
. Put U = U
1
. . . U
k
.
2.144. Theorem. (Cf 2.149) The product of nitely many compact spaces is compact.
Proof. We show that the product XY of two compact spaces X and Y is compact. The theorem
in its general form follows by induction. Suppose that A
j
, j J, is an open covering of X Y . We
show that there is a nite subcovering. For each point x X, the slice x Y , homeomorphic to Y ,
is compact and therefore x Y

jJ
x
A
j
for some nite index set J
x
J (2.136). Thanks to the
tube lemma (2.143) we know that the open set

jJ
x
A
j
actually contains a whole tube U
x
Y for some
neighborhood U
x
of x. By compactness, X = U
x
1
. . . U
x
k
can be covered by nitely many of the
neighborhoods U
x
. Now
X Y =
_
_
1ik
U
x
i
_
Y =
_
1ik
(U
x
i
Y ) =
_
1ik
_
jJ
x
i
A
j
=
_
j
S
1ik
J
x
i
A
j
where

1ik
J
x
i
J is nite (1.34).
Here are two small lemmas that are used quite often.
2.145. Lemma (Criterion for noncompactness). If X contains an innite closed discrete subspace,
then X is not compact.
Proof. Any closed subspace of a compact space is compact (2.137). A discrete and compact space
is nite.
2.146. Lemma (Intersection of a nested sequence of compact sets). [8, Ex 28.5] [5, 3.10.2] Let C
1

C
2
C
n
be a descending sequence of closed nonempty subsets of a compact space. Then

C
n
,= .
Proof. If

C
n
= then C
n
= for some n Z
+
by 2.135.(2).
2.147. Theorem. [Cf [8, Ex 27.5]] A nonempty compact Hausdor space without isolated points
(2.39) is uncountable.
10. COMPACT SPACES 53
Proof. Let x
n
be a sequence of points in X. It suces (1.39) to show that x
n
[ n Z
+
, = X.
Since X is Hausdor and has no isolated points there is a descending sequence
V
1
V
2
V
n1
V
n

of nonempty open sets such that x
n
, V
n
for all n. These are constructed recursively. Put V
0
= X.
Suppose that V
n1
has been constructed for some n Z
+
. Since x
n
is not isolated, x
n
, = V
n1
. Choose
a point y
n
V
n1
distinct from x
n
and choose disjoint (separated) open sets U
n
X, V
n
V
n1
such
that x
n
U
n
and y
n
V
n
. Then U
n
V
n
= so x
n
, V
n
.
By construction, the intersection

V
n
contains none of the points x
n
of the sequence and by
compactness (2.146),

V
n
,= . Thus X contains a point that is not in the sequence.
The Alexandro compactication (2.167) Z
+
= K = K0, where K is as in 2.11, is a countable
compact Hausdor space with isolated points.
Is it true that a connected Hausdor space is uncountable?
2.148. Example. (1) The (unreduced) suspension of the space X is the quotient space
SX = [0, 1] X/(0 X, 1 X)
What is the suspension of the n-sphere S
n
? Dene f : [0, 1] S
n
S
n+1
to be the (continuous)
map that takes (t, x), t [0, 1], to the geodesic path from the north pole (0, 1) S
n+1
, through the
equatorial point (x, 0) S
n
S
n+1
, to the south pole (0, 1) S
n+1
. (Coordinates in R
n+2
=
R
n+1
R.) By the universal property of quotient spaces there is an induced continuous and bijective
map f : SS
n
S
n+1
. Since SS
n
is compact (as a quotient of a product of two compact spaces) and
S
n+1
is Hausdor (as a subspace of a Hausdor space), f is a homeomorphism (2.141). We often
write SS
n
= S
n+1
, n 0, where the equality sign stands for is homeomorphic to.
(2) Every injective continuous map S
1
R
2
is an embedding. Can you nd an injective continuous
map R
1
R
2
that is not an embedding?
(3) Alexanders horned sphere [7, Example 2B.2] is a wild embedding of the 2-sphere S
2
in R
3
such
that the unbounded component of the complement R
3
S
2
(2.133.(5)) contains non-contractible
loops. It may be easier instead to consider Alexanders horned disc [3, p 232]. (The horned sphere is
obtained by cutting out a disc of the standard sphere and replacing it a horned disc.)
(4) Let X be a Hausdor space and
X
0
X
1
X
n1
X
n
X
an ascending sequence of closed subspaces. Assume that the topology on X is coherent with this
ltration in the sense that
A is closed A X
n
is closed for all n
holds for all subsets A of X. Then any compact subspace C of X is contained in a nite stage of the
ltration. To see this, choose a point t
n
C (X
n
X
n1
) for all n Z
+
for which this intersection
is nonempty. Let T = t
n
be the set of these points. We want to prove that T is nite. Certainly,
T X
n
is nite, hence closed in the Hausdor space X
n
for all n 0. Therefore T is closed. In
fact, any subset of T is closed by the same argument and thus T is discrete. But any closed discrete
subspace of the compact space C is nite (2.145).
2.149. Theorem (Tychono theorem). The product

jJ
X
j
of any collection (X
j
)
jJ
of compact
spaces is compact.
Proof. Put X =

jJ
X
j
. Let us say that a collection of subsets of X is an FIP-collection (nite
intersection property) if any nite subcollection has nonempty intersection. We must (2.135.(2)) show
/ is FIP =

AA
A ,=
holds for any collection / of subsets of X.
So let / be a FIP-collection of subsets of X. Somehow we must nd a point x in

AA
A.
Step 1. The FIP-collection / is contained in a maximal FIP-collection.
The set (supercollection?) A of FIP-collections containing / is a strictly partially ordered by strict
inclusion. Any linearly ordered subset (subsupercollection?) B A has an upper bound, namely the
FIP-collection

B =

BB
B containing /. Now Zorns lemma (1.56) says that A has maximal elements.
We can therefore assume that / is a maximal FIP-collection.
54 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
Step 2. Maximality of / implies
(2.150) A
1
, . . . , A
k
/ = A
1
A
k
/
and for any subset A
0
X we have
(2.151) A /: A
0
A ,= = A
0
/
To prove (2.150), note that the collection / A
1
A
k
is FIP, hence equals / by maximality. To
prove (2.151), note that the collection /A
0
is FIP: Let A
1
, . . . , A
k
/. By (2.150), A
1
A
k
/
so A
1
A
k
A ,= by assumption.
Step 3.
j
(A) [ A / is FIP for all j J.

j
(A
1
)
j
(A
k
)
j
(A
1
A
j
) ,= for any nite collection of sets A
1
, . . . , A
k
/.
Since X
j
is compact and
j
(A) [ A / is an FIP collection of closed subsets of X
j
, the intersection

AA

j
(A) X
j
is nonempty (2.135.(2)). Choose a point x
j

j
(A) for each j J and put
x = (x
j
) X.
Step 3. x

AA
A.
The point is that
1
j
(U
j
) / for any neighborhood U
j
of x
j
. This follows from (2.151) since
1
j
(U
j
)
A ,= , or equivalently, U
j

j
(A) ,= , for any A / as x
j

j
(A). Since / has the FIP it follows that

1
j
1
(U
j
1
)
1
j
k
(U
j
k
) A ,= whenever U
j
i
are neighborhoods of the nitely many points x
j
i
X
j
i
and A /. Thus x A for all A in the collection /; in other words x

AA
A.
2.152. Compact subspaces of linearly ordered spaces. We show rst that any compact sub-
space of a linearly ordered space is contained in a closed interval.
2.153. Lemma. Let X be a linearly ordered space and , = C X a nonempty compact subspace.
Then C [m, M] for some elements m, M C.
Proof. The claim is that C has a smallest and a largest element. The proof is by contradiction.
Assume that C has no largest element. Then C

cC
(, c) as there for any a C is a c C such
that a < c. By compactness
C (, c
1
) (, c
k
) (, c)
where c = maxc
1
, . . . , c
k
is the largest (1.50) of the nitely many elements c
1
, . . . , c
k
C. But this is
a contradiction as c C and c , (, c).
We now show that the eg the unit interval [0, 1] is compact in R. Note that the unit interval [0, 1]
is not compact in R
1
2
. (The intersection all the closed subspaces [
1
n
+
1
2
,
1
n
+
1
2
] [0, 1], n 2, is
empty but the intersection of nitely many of them is not empty.) The reason for this dierence is that
R, but not R
1
2
(1.16.(6)), has the least upper bound property (1.15).
2.154. Theorem. Let X be a linearly ordered space with the least upper bound property. Then every
closed interval [a, b] in X is compact.
Proof. Let [a, b] X be a closed interval and / and open covering of [a, b] (with the subspace
topology which is the order topology (2.27)). We must show that [a, b] is covered by nitely many of the
sets from the collection /. The set
C = x [a, b] [ [a, x] can be covered by nitely many members of /
is nonempty (a C) and bounded from above (by b). Let c = sup C be the least upper bound of C.
Then a c b. We would like to show that c = b.
Step 1 If x C and x < b then C (x, b] ,= .
Proof of Step 1. Suppose rst that x has an immediate successor y > x. We can not have x < b < y
for then y would not be an immediate successor. So x < y b. Clearly [a, y] = [a, x] y can be
covered by nitely many members of /. Suppose next that x has no immediate successor. Choose an
open set A / containing x. Since A is open in [a, b] and contains x < b, A contains an interval of the
form [x, d) for some d b. Since d is not an immediate successor of x there is a point y (x, d). Now
[a, y] [a, x] [x, d) [a, x] A can be covered by nitely sets from /.
Step 2 c C.
Proof of Step 2. The claim is that [a, c] can be covered by nitely many members of /. From Step 1 we
have that C contains elements > a so the upper bound c is also > a. Choose A / such that c A.
Since A is open in [a, b] and a < c, A contains an interval of the form (d, c] for some d where a d.
Since d is not an uppe bound for C, there are points from C in (d, c]. Let y be such a point. Now
[a, c] = [a, y] (d, c] [a, y] A can be covered by nitely many sets from /.
10. COMPACT SPACES 55
Step 3 c = b.
Proof of Step 3. By Step 2, c C. But then c = b for Step 1 says that if c < b, then c can not be an
upper bound.
For instance, the ordered square I
2
o
= [00, 11], Z
+
= [1, ], and S

= [1, ] are compact linearly

ordered spaces [8, Ex 10.1]. More importantly, since the linearly ordered space R has the least upper
bound property, we get that all closed intervals in R are compact.
See [8, Ex 27.1] for the converse.
2.155. Corollary. Let X be a linearly ordered space and , = C X a nonempty subspace. Then
C is compact and connected = C is a closed interval
The converse holds when X is a linear continuum.
Proof. Since C is compact, C [m, M] for some m, M C (2.153) and [m, M] C because C is
connected, hence convex (2.118). In a a linear continuum every closed interval is compact (2.154) and
connected (2.118).
In particular, the compact and connected subsets of the linear continuum R are precisely the closed
intervals.
2.156. Theorem (HeineBorel). A subset C of the euclidean space R
n
is compact if and only if it is
closed and bounded in the euclidean metric.
Proof. Let C R
n
be compact. Since C is compact and (R, R)
n
[ R > 0 an open covering of
R
n
, there is an R > 0 such that C [R, R]
n
. Thus C is bounded. Compact subspaces of Hausdor
spaces are closed (2.139), so C is closed.
Conversely, if C is closed and bounded then C is a closed subset of [R, R]
n
for some R > 0.
But [R, R]
n
is compact (2.155, 2.144) and closed subsets of compact spaces are compact (2.137); in
particular, C is compact.
Can you now answer question 2.60.(10)?
2.157. Theorem (Extreme Value theorem). Let f : X Y be a continuous map of a nonempty
topological space X into a linearly ordered space Y .
(1) If X is compact then there exist m, M X such that f(X) [f(m), f(M)].
(2) If X is compact and connected then there exist m, M X such that f(X) = [f(m), f(M)].
Proof. (1) Since the image f(X) is compact (2.138) we can apply (2.153).
(2) Since the image f(X) is compact (2.138) and connected (2.107) we can apply (2.155).
2.158. Compactness in metric spaces. Since any open covering of compact space contains a
nite subcovering we see that any open covering of a compact metric space contains a subcovering of
open sets of size bounded from below. This observation is expressed in the Lebesgue lemma.
2.159. Lemma (Lebesgue lemma). Let X be a compact metric space. For any open covering of X
there exists a positive number > 0 (the Lebesgue number of the open covering) such that any subset of
X with diameter < is contained in one of the open sets in the covering.
Proof. Cover X by nitely many open balls, X = B(x
1
,
1
) B(x
2
,
2
) B(x
k
,
k
), such that
each double radius ball B(x
i
, 2
i
) is contained in one of the open sets of the covering. Let be any
positive number smaller than all the
i
. Then any subset of diameter < is contained in one of the
balls B(x
i
, 2
i
) and therefore in one of the open sets from the covering. (Let A X be a subset with
diam(A) < . Choose any point a in A. Then a is in one of the open balls B(x
i
,
i
). For any point
b A, the distance d(b, x
i
) d(b, a) + d(a, x
i
) < +
i
< 2
i
which means that the point b is contained
in B(x
i
, 2
i
).)
The open covering of R consisting of the open intervals (1, 1) and (x1/[x[, x+1/[x[) for all x R
with [x[ > 1 does not have a Lebesgue number.
2.160. Definition. A map f : X Y between metric spaces is uniformly continuous if for all > 0
there is a > 0 such that d
Y
(f(x
1
), f(x
2
)) < whenever d
X
(x
1
, x
2
) < :
> 0 > 0x
1
, x
2
X: d
X
(x
1
, x
2
) < d
Y
(f(x
1
), f(x
2
)) <
2.161. Theorem (Uniform continuity theorem). Let f : X Y be a continuous map between metric
spaces. If X is compact, then f is uniformly continuous.
56 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
Proof. Given > 0. Let be the Lebesgue number of the open covering f
1
B(y,
1
2
) [ y Y .
Then we have for all x
1
, x
2
X
d
X
(x
1
, x
2
) < = diamx
1
, x
2
<
= y Y : x
1
, x
2
f
1
B(y,
1
2
)
= y Y : f(x
1
), f(x
2
) B(y,
1
2
)
= d
Y
(f(x
1
), f(x
2
)) <

2.159. Limit point compactness and sequential compactness.

2.162. Definition. A space X is
(1) limit point compact if any innite subset of X has a limit point
(2) sequentially compact if any sequence in X has a convergent subsequence.
In other words, a space is limit point compact if it contains no innite closed discrete subspaces
(2.40).
We shall not say much about these other forms of compactness (see [5, 3.10] for a more thorough
discussion).
2.163. Theorem. (Cf [8, Ex 28.4, Ex 35.3]) For any topological space X we have
X is compact = X is limit point compact
1st countable + T
1
= X is sequentially compact
All three forms of compactness are equivalent for a metrizable space.
Proof. X is compact = X is limit point compact: A subset with no limit points is closed and
discrete (2.40), hence nite (2.145).
X is limit point compact, 1st countable and T
1
= X is sequentially compact: Let (x
n
) be a sequence
in X. Consider the set A = x
n
[ n Z
+
of points in the sequenec. If A is nite there is a constant
subsequence (1.34.(3)). If A is innite, A has a limit point x by hypothesis. If X is 1st countable, there
is a countable nested countable basis, U
1
U
2
, at x just as in the proof of (2.100). Since x is a
limit point, there is a sequence element x
n
1
U
1
. Suppose that we have found n
1
< < n
k
such that
x
n
i
U
i
. Since x is a limit point and X is T
1
, there are innitely many points from A in U
k+1
(2.48). In
particular, there is n
k+1
< n
k
such that x
n
k+1
is in U
k+1
. The subsequence x
n
k
converges to x.
X is sequentially compact and metrizable = X compact: This is more complicated (but should also be
well-known from your experience with metric spaces) so we skip the proof.
2.164. Example. (1) The well-ordered space S

of all countable ordinals is limit point compact

and sequentially compact but it is not compact. S

is not compact (2.153) for it is a linearly ordered

space with no greatest element [8, Ex 10.6]. On the other hand, any countably innite subset of S

is contained in a compact subset (1.52.(2), 2.154). Therefore (2.163) any countably innite subset,
indeed any innite subset (2.40), has a limit point and any sequence has a convergent subsequence.
(Alternatively, use that S

is rst countable (2.98.(4)).) It follows (2.163) that S

is not metrizable.
(2) The Stone

Cech compactication (3.37) Z

+
of the positive integers is compact but it is not limit
point compact as the sequence (n)
nZ
+
has no convergent subsequence. Indeed, no point of the
remainder Z
+
Z
+
is the limit of a sequence in Z
+
[8, Ex 38.9]. (It follows that Z
+
is not rst
countable and not metrizable).
11. LOCALLY COMPACT SPACES AND THE ALEXANDROFF COMPACTIFICATION 57
11. Locally compact spaces and the Alexandro compactication
2.165. Definition. A space X is locally compact at the point x X if x lies in the interior of some
compact subset of X. A space is locally compact if it is locally compact at each of its points.
This means that X is locally compact at the point x if x U C where U is an open and C a
compact subset of X.
Compact spaces are locally compact (U = X = C). The real line R (more generally, R
n
) is locally
compact but not compact. All linearly ordered spaces (2.13) with the least upper bound property are
locally compact. The space Q R of rational numbers is not locally compact [8, Ex 29.1].
2.166. Definition. A compactication of the space X is an embedding c: X cX of X into a
compact Hausdor space cX such that c(X) is dense in cX.
The only compactication of a compact space X is X itself as X is closed and dense in cX.
Let X be a locally compact Hausdor space. Let X = X denote the union of X with a set
consisting of a single point . The collection
T = U [ U X open X C [ C X compact
is a topology on X: It is easy to see that nite intersections and arbitrary unions of open sets of the
rst (second) kind are again open of the rst (second) kind. It follows that T is closed under arbitrary
unions. It is also closed under nite intersections since U (XC) = U C is open in X (2.139). And
X = X and are open.
If X is compact, then X is X with an added isolated point. The next theorem describes X in the
more interesting case when X is not compact.
2.167. Theorem (Alexandro). Let X be a locally compact but not compact Hausdor space.
(1) The map X X is a compactication missing one point.
(2) If c: X cX is another compactication of X then there is a unique closed quotient map
c: cX X = cX/(cX X) such that the diagram
X
c
|
|
|
|
|
|
|
|

C
C
C
C
C
C
C
C
cX
c

X
commutes.
Proof. We must verify the following points:
The subspace topology on X is the topology on X: The subspace topology X T is clearly the original
topology on X.
X is dense in X: The intersection of X and some neighborhood of has the form X C where C is
compact. Since X is assumed non-compact, X C is not empty.
X is Hausdor: Let x
1
and x
2
be two distinct points in X. If both points are in X then there are
disjoint open sets U
1
, U
2
X X containing x
1
and x
2
, respectively. If x
1
X and x
2
= , choose an
open set U and a compact set C in X such that x U C. Then U x and X C are disjoint
open sets in X.
X is compact: Let U
j

jJ
be any open covering of X. At least one of these open sets contains . If
U
k
, k J, then U
k
= (X C) for some compact set C X. There is a nite set K J such
that U
j

jK
covers C. Then U
j

jK{k}
is a nite open covering of X = X .
Let now c: X cX be another compactication X. Dene c: cX X by c(x) = x for all x X
and c(cX X) = . We check that c is continuous. For any open set U X X, c
1
(U) is
open in X and hence (2.24.(2)) in cX since X is open in cX (2.168). For any compact set C X,
c
1
(X C) = cX C is open in cX since the compact set C is closed in the Hausdor space cX
(2.140.(1)). This shows that c is continuous. By construction, c is surjective and hence (2.141) a closed
quotient map by the Closed Map Lemma (2.141).
The theorem says that X = cX/(cX X) where cX is any compactication of X. In particular, if
cX consists of X and one extra point then the map c is a bijective quotient map, ie a homeomorphism
(3). The space X is called the Alexandro compactication or the one-point compactication of X.
2.168. Lemma. Any locally compact and dense subspace of a Hausdor space is open.
58 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
Proof. Suppose that Y is Hausdor and that the subspace X Y is locally compact and dense.
Let x X. Since X is locally compact Hausdor the point x has a neighborhood X U, where U is
open in Y , such that its relative closure
Cl
X
(X U)
2.35
= X X U
2.37
= X U
is compact and hence closed in the Hausdor space Y (2.139). But no part of U can stick outside X U
for since U (X U) is open
U (X U) ,=
X dense
= (X U) (X U) ,=
which is absurd. Thus we must have U X U, in particular, U X. This shows that X is open.
2.169. Example. (1) The n-sphere S
n
= R
n
is the Alexandro compactication of the locally
compact Hausdor space R
n
for there is (2.60.(8)) a homeomorphism of R
n
onto the complement of
a point in S
n
.
(2) The real projective plane is the Alexandro compactication of the Mobius band. More generally,
real projective n-space RP
n
= MB
n
is the Alexandro compactication of the n-dimensional Mobius
band MB
n
(2.83) which is a locally compact Hausdor space (indeed a manifold (3.40)).
(3) Let X be a linearly ordered space with the least upper bound property and let [a, b) be a half-
open interval in the locally compact space X. Then [a, b) = [a, b]. For instance the Alexandro
compactications of the half-open intervals [0, 1) R, Z
+
= [1, ) Z
+
Z
+
, and S

= [1, ) S

are [0, 1], [1, ], [1, ] = S

, respectively.
(4) The Alexandro compactication of a countable union of disjoint copies of the real line

_

nZ
+
R
_
= (Z
+
R)
2.172.(4)
= Z
+
S
1
=
_
nZ
+
C
1/n
is the Hawaiian Earring (2.41.(2)) (which is compact by the Heine-Borel theorem (2.156)).
(5) The Warsaw circle W (which compact by the Heine-Borel theorem (2.156)) (2.133.(2)) is a com-
pactication of R and the quotient space W/(W R) = R = S
1
.
2.170. Corollary (Characterization of locally compact Hausdor spaces). For a Hausdor space
X the following conditions are equivalent:
(1) X is locally compact
(2) X is homeomorphic to an open subset of a compact space
(3) For any point x X and any neighborhood U of x there is an open set V such that x V
V U and V is compact.
Proof. (1) = (2): The locally compact Hausdor space X is homeomorphic to the open subspace
X of the compact Hausdor space X (2.167).
(2) = (3): Suppose that Y is a compact Hausdor space and that X Y is an open subset. X is
Hausdor since subspaces of Hausdor spaces are Hausdor (2.47). Let x be a point of X and U X
a neighborhood of x in X. Then U is open also in Y (2.24) so that the complement Y U is closed
and compact (2.137) in the compact space Y . In the Hausdor space Y we can separate (2.140.(2)) the
disjoint closed subspaces x and Y U by disjoint open sets, V x and W Y U or Y W U. As
V is disjoint from W it is contained in the closed subset Y W and then also V Y W U. Here,
V is compact as a closed subspace of the compact space Y (2.137). Note (2.35) that the closure of V in
X equals the closure of V in Y because V , the closure of V in Y , is contained in U and X.
(3) = (1): Let x be a point of X. By property (3) with U = X there is an open set V such that
x V V X and V is compact. Thus X is locally compact at x.
2.171. Corollary. (1) Closed subsets of locally compact spaces are locally compact.
(2) Open or closed subsets of locally compact Hausdor spaces are locally compact Hausdor.
Proof. (1) Let X be a locally compact space and A X a closed subspace. We use the denition
(2.165) directly to show that A is locally compact. Let a A. There are subsets a U C X such
that U is open and C is compact. Then A U A C where A U is a neighborhood in A and C A
is compact as a closed subset of the compact set C (2.137).
(2) If X is locally compact Hausdor and A X open, it is immediate from 2.170.(2) that A is locally
compact Hausdor.
11. LOCALLY COMPACT SPACES AND THE ALEXANDROFF COMPACTIFICATION 59
An arbitrary subspace of a locally compact Hausdor space need not be locally compact (2.172.(1)).
The product of nitely many locally compact spaces is locally compact but an arbitrary product of locally
compact spaces need not be locally compact [8, Ex 29.2]; for instance Z

+
is not locally compact. The
image of a locally compact space under an open continuous [8, 29.3] or a perfect map [8, Ex 31.7] [5,
3.7.21] is locally compact but the image under a general continuous map of a locally compact space need
not be locally compact; indeed, the quotient of a locally compact space need not be locally compact
(2.172.(2)) [5, 3.3.16].
2.172. Example. (1)

nZ
+
C
n
R
2
(2.41.(2)) is not locally compact at the origin: Any neigh-
borhood of 0 0 contains a countably innite closed discrete subspace so it can not be contained in
any compact subspace (2.145).
(2) The quotient space R/Z (2.98.(6)) is not locally compact at the point corresponding to Z: Any
neighborhood of this point contains an innite closed discrete subspace so it can not be contained in
any compact subspace (2.145).
(3) In diagram (2.99), the space

S
1
is locally compact but not compact,

S
1
is compact, and
_
S
1
is not locally compact (at the one point common to all the circles).
(4) (Wedge sums and smash products) A pointed space is a topological space together with one of its
points, called the base point. The wedge sum of two pointed disjoint spaces (X, x
0
) and (Y, y
0
) is the
quotient space
X Y = (X HY )/(x
0
y
0
)
obtained from the disjoint union of X and Y (2.21) by identifying the two base points. Let f : X Y X Y
be the continuous map given by f(x) = (x, y
0
), x X, and f(y) = (x
0
, y), y Y . The image
f(X Y ) = X y
0
x
0
Y =
1
2
(y
0
)
1
1
(x
0
) is closed when X and Y are T
1
spaces.
In fact, f is a closed map for if C X is closed then f(C) = (x, y
0
) [ x C =
1
1
(C)
1
2
(y
0
)
is also closed. Thus f is a quotient map onto its image and its factorization f as in the commutative
diagram (2.82.(2))
X Y

I
I
I
I
I
I
I
I
I
f

X Y
X Y
f

u
u
u
u
u
u
u
u
u
is an embedding. We can therefore identify X Y and X y
0
x
0
Y . The smash product of
the pointed spaces (X, x
0
) and (Y, y
0
) is dened to be
X Y = (X Y )/(X Y ) = X Y/(X y
0
x
0
Y )
the quotient of the product space X Y by the closed subspace X Y ( = ).
If A X and B Y are closed subspaces, the universal property of quotient maps (2.82.(2))
produces a continuous bijection g such that the diagram
X Y

R
R
R
R
R
R
R
R
R
R
R
R
R
R
g
X
g
Y

X/AY/B
g
X/AY/B

X/A Y/B
(X Y )/(X B AY )
g

k
k
k
k
k
k
k
k
k
k
k
k
k
k
k
commutes. However, g may not be a homeomorphism since the product g
X
g
Y
of the two closed
quotient maps g
X
: X X/A and g
Y
: Y Y/B may not be a quotient map. But if X and Y are
locally compact Hausdor spaces then g
X
g
Y
is quotient (2.87) so that g is a homeomorphism (2.79,
2.82.(3), 2.78.(3)) and
(2.173) (X Y )/(X B AY ) = X/A Y/B
in this case.
(5) The torus S
1
S
1
is a compactication of RR with remainder S
1
S
1
.
(6) (The Alexandro compactication of a product space) If X and Y are locally compact Hausdor
spaces the map X Y X Y X Y is an embedding. This follows from (2.81, 2.78.(3))
when we note that the rst map embeds X Y into an open saturated subset of X Y (2.61).
The universal property of the Alexandro compactication (2.167) implies that
(2.174) (X Y ) = X Y
60 2. TOPOLOGICAL SPACES AND CONTINUOUS MAPS
for any two locally compact Hausdor spaces X and Y . In particular,
S
m
S
n
= R
m
R
n
(2.174)
= (R
m
R
n
) = R
m+n
= S
m+n
when we view the spheres as Alexandro compactications of euclidean spaces.
(7) The nsphere S
n
is (homeomorphic to) the quotient space I/I. Hence
S
n
= S
1
. . . S
1
= I/I . . . I/I = I
n
/I
n
by (2.173) and the formula (Exam June 2003) for the boundary of the product of two sets.
(8) (The p-adic integers.) Let p be a prime number. The product ring

n=1
Z/p
n
Z of all the residue
(discrete topological) rings Z/p
n
Z, 0 < n < , is compact according to the Tychono theorem
(2.149). The subring, called the p-adic integers,
Z
p
= (a
n
) [ n: a
n
a
n+1
mod p
n

n=1
Z/p
n
Z
is closed and therefore also compact (2.137). Indeed, the sets (a
n
)

n=1
[ a
i
a
j
mod p
i
are closed
(and open) whenever 0 < i < j because the projection maps are continuous (2.64). The diagonal ring
homomorphism
Z

Z
p

n=1
Z/p
n
Z, (a) = (a mod p
n
)

n=1
embeds the ring of integers into Z
p
. By the denition of the product topology (2.16), Z is dense
in Z
p
in that any neighborhood of a point in Z
p
contains a point from Z. Thus Z
p
= Z is a
compact topological ring containing the integers as a dense subring and the map : Z Z
p
is a
compactication (2.166) of the discrete space Z. As a ring, Z
p
is quite dierent from Z. For instance,
the integer 1p is invertible in Z
p
with inverse (1p)
1
= (1, 1+p, 1+p+p
2
, . . .) =

n=0
p
n
. Actually,
any x = (x
1
, x
2
, . . .) Z
p
with nonzero reduction mod p is invertible because any x
n
Z/p
n
Z with
nonzero reduction mod p is invertible [9, Prop 2, p 12].
Some mathematicians prefer to include Hausdorness in the denition of (local) compactness. What
we call a (locally) compact Hausdor space they simply call a (locally) compact space; what we call a
(locally) compact space they call a (locally) quasicompact space.
CHAPTER 3
Regular and normal spaces
1. Countability Axioms
Let us recall the basic question: Which topological spaces are metrizable? In order to further analyze
this question we shall look at a few more properties of topological spaces.
We have already encountered the rst countability axiom. There are three more axioms using the
term countable. Here are the four countability axioms.
3.1. Definition. A topological space
that has a countable neighborhood basis at each of its points is called a rst countable space
that has a countable basis is called a second countable space
that contains a countable dense subset is said to contain a countable dense subset
1
in which every open covering has a countable subcovering is called a Lindelof space
A subset A X is dense if A = X, that is if every nonempty open subset contains a point from A.
Any second countable space is rst countable. Any subspace of a (rst) second countable space is
(rst) second countable. A countable product of (rst) second countable spaces is (rst) second countable
(2.97). Quotient spaces of rst coountable spaces need not be rst countable (2.98.(7)).
The real line R is second countable for the open intervals (a, b) with rational end-points form a
countable basis for the topology. R

is rst but not second countable (3.8).

Metric spaces are rst countable but not necessarily second countable [8, Example 2 p 190].
Any compact space is Lindelof. Any closed subset of a Lindelof space is Lindelof [8, Ex 30.9] (with
a proof that is similar to that of 2.137). A product of two Lindelof spaces need not be Lindelof (3.9).
The next result implies that R, in fact any subset of the 2nd countable space R, is Lindelof.
3.2. Theorem. Let X be a topological space. Then
X has a countable dense subset X is 2nd countable
.

X is Lindelof
X is 1st countable
If X is metrizable, the three conditions of the top line equivalent.
Proof. Suppose rst that X is second countable and let B be a countable basis for the topology.
X has a countable dense subset: Pick a point b
B
B in each basis set B B. Then b
B
[ B B is
countable (1.41.(2)) and dense.
X is Lindelof: Let | be an open covering of X. For each basis set B B which is contained in some open
set from the collection |, pick any U
B
| such that B U
B
. Then the at most countable collection
U
B
of these open sets from | is an open covering: Let x be any point in X. Since x is contained in a
member U of | and every open set is a union of basis sets, we have x B U for some basis set B B.
But then x B U
B
.
Any metric space with a countable dense subset is 2nd countable: Let X be a metric space with a count-
able dense subset A X. Then the collection B(a, r) [ a A, r Q
+
of balls centered at
points in A and with a rational radius is a countable (1.41.(4)) basis for the topology: It suces
to show that for any open ball B(x, ) in X and any y B(x, ) there are a A and r Q
+
such that y B(a, r) B(x, ). Let r be a positive rational number such that 2r + d(x, y) <
and let a A B(y, r). Then y B(a, r), of course, and B(a, r) B(x, ) for if d(a, z) < r then
d(x, z) d(x, y) +d(y, z) d(x, y) +d(y, a) +d(a, z) < d(x, y) + 2r < .
Any metric Lindelof space is 2nd countable: Let X be a metric Lindelof space. For each positive rational
number r, let A
r
be a countable subset of X such that X =

aA
r
B(a, r). Then A =

rQ
+
A
r
is a
1
Or to be separable
61
62 3. REGULAR AND NORMAL SPACES
dense countable (1.41.(3)) subset: For any open ball B(x, ) and any positive rational r < there is an
a A
r
such that x B(a, r). Then a B(x, r) B(x, ).
3.3. Example. The ordered square I
2
o
is compact (2.154) and therefore Lindelof but it is not second
countable since it contains uncountably many disjoint open sets (x 0, x 1), x I. Thus I
2
o
is not
metrizable [8, Ex 30.6].
2. Separation Axioms
3.4. Definition. A space X is called a
T
1
-space if points x are closed in X
T
2
-space or a Hausdor space if for any pair of distinct points x, y X there exist disjoint open
sets U, V X such that x U and y V
T
3
-space or a regular space if points are closed and for any point x X and any closed set
B X not containing x there exist disjoint open sets U, V X such that x U and B V
T
4
-space or a normal space if points are closed and for every par of disjoint closed sets A, B X
there exist disjoint open sets U, V X such that A U and B V .
We have the following sequence of implications
X is normal = X is regular = X is Hausdor = X is T
1
where none of the arrows reverse (3.9, 3.10, [8, Ex 22.6]).
3.5. Lemma. Let X be a T
1
-space. Then
(1) X is regular For every point x X and every neighborhood U of x there exists an open
set V such that x V V U.
(2) X is normal For every closed set A and every neighborhood U of A there is an open set
V such that A V V U.
Proof. Let B = X U.
3.6. Theorem. (Cf 2.47) Any subspace of a regular space is regular. Any product of regular spaces
is regular.
Proof. Let X be a regular space and Y X a subset. Then Y is Hausdor (2.47)). Consider a
point y Y and a and a closed set B X such that y , Y B. Then y , B and since Y is regular there
exist disjoint open sets U and V such that y U and B V . The relatively open sets U Y and V Y
are disjoint and they contain y and B Y , respectively.
Let X =

X
j
be the Cartesian product of regular space X
j
, j J. Then X is Hausdor (2.47)). We
use 3.5.(1) to show that X is regular. Let x = (x
j
) be a point in X and U =

U
j
a basis neighborhood
of x. Put V
j
= X
j
whenever U
j
= X
j
. Otherwise, choose V
j
such that x
j
V
j
V
j
U
j
. Then
V =

V
j
is a neighborhood of x in the product topology and (2.66) V =

V
j

U
j
= U. Thus X is
regular.
3.7. Theorem. [8, Ex 32.1] A closed subspace of a normal space is normal.
Proof. Quite similar to the proof (3.6) that a subspace of a regular space is regular.
An arbitraty subspace of a normal space need not be normal (3.32) and the product of two normal
spaces need not be normal ((3.9),[8, Example 2 p 203], [5, 2.3.36]).
3.8. Example (Sorgenfreys half-open interval topology). The half-open intervals [a, b) form a basis
for the space R

(2.11, 2.31.(4), 2.98.(2), 2.108.(3)).

R

is 1st countable: At the point x the collection of open sets of the form [x, b) where b > x is
rational, is a countable local basis at x.
R

is not 2nd countable: Let B be any basis for the topology. For each point x choose a member
B
x
of B such that x B
x
[x, x + 1). The map R B: x B
x
is injective for if B
x
= B
y
then x = inf B
x
= inf B
y
= y.
R

has a countable dense subset: Q is dense since any (basis) open set in R

contains rational
points.
2. SEPARATION AXIOMS 63
R

is Lindelof: It suces (!) to show that any open covering by basis open sets contains a
countable subcovering, ie that if R =

jJ
[a
j
, b
j
) is covered by a collection of right half-open
intervals [a
j
, b
j
) then R is actually already covered by countably many of these intervals. (Note
that this is true had the intervals been open as R is Lindelof.) Write
R =
_
jJ
(a
j
, b
j
)
_
R
_
jJ
(a
j
, b
j
)
_
as the (disjoint) union of the corresponding open intervals and the complement of this union.
The rst set can be covered by countably many of the intervals (a
j
, b
j
) for any subset of R (with
the standard topology) is 2nd countable and hence Lindelof (3.2). Also the second set can be
covered by countably many of the intervals (a
j
, b
j
) simply because it is countable: The second
set
R
_
jJ
(a
j
, b
j
) =
_
jJ
[a
j
, b
j
)
_
jJ
(a
j
, b
j
) = a
k
[ j J : a
k
, (a
j
, b
j
)
consists of some of the left end-points of the intervals. The open intervals (a
k
, b
k
) are disjoint for
a
k
in this set. But there is only room for countably many open disjoint intervals in R (choose
a rational point in each of them) so there are at most countaly many left end-dpoints a
k
in the
second set.
R

is normal: Let A, B R

be disjoint closed sets. For each point a A R B there

is an x
a
R such that [a, x
a
) R B. Let U
A
=

aA
[a, x
a
). Dene U
B
=

bB
[b, x
b
)
similarly. Then U
A
and U
B
are open sets containing A and B, respectively. If U
A
U
B
,=
then [a, x
a
) [b, x
b
) ,= for some a A, b B. Say a < b; then b [a, x
a
) RB which is a
contradiction So U
A
and U
B
are disjoint. (R

is even completely normal [8, Ex 32.6] by this

argument.)
Since R

has a dense countable subset and is not second countable (3.8) it is not metrizable (3.2).
3.9. Example (Sorgenfreys half-open square topology). The half-open rectangles [a, b) [c, d) form
a basis (2.18) for the product topology R

. The anti-diagonal L = (x, x) [ x R is a closed

(clear!) discrete (L [x, ) [x, ) = (x, x)) subspace of the same cardinality as R. Q Q is a
countable dense subspace.
R

is not Lindelof: A Lindelof space can not contain an uncountable closed discrete sub-
space (cf 2.145, [8, Ex 30.9]).
R

is not normal: A normal space with a countable dense subset can not contain a closed
discrete subspace of the same cardinality as R [5, 2.1.10]. (Let X be a space with a countable
dense subset. Since any continuous map of X into a Hausdor space is determined by its
values on a dense subspace [8, Ex 18.13], the set of continuous maps X R has at most the
cardinality of R
Z
=

Z
R. Let X be any normal space and L a closed discrete subspace. The
Tietze extension theorem (3.15) says that any map L R extends to continuous map X R.
Thus the set of continuous maps X R has at least the cardinality of R
L
. If L and R have the
same cardinality, R
L
= R
R
=

R
R which is greater (1.42) than the cardinality of

Z
R.) See
[8, Ex 31.9] for a concrete example of two disjoint closed subspaces that can not be separated
by open sets.
R

is (completely) regular: since it is the product (3.30) of two (compeletely) regular

(3.28) (even normal (3.8)) spaces.
Example 3.8 shows that the arrows of Theorem 3.2 do not reverse. Example 3.9 shows that the
product of two normal spaces need not be normal, that the product of two Lindelof spaces need not be
Lindelof, and provides an example of a (completely) regular space that is not normal.
3.10. Example (A Hausdor space that is not regular). The open intervals plus the sets (a, b) K
where K =
1
n
[ n Z
+
form a basis for the topology R
K
(2.11, 2.124.(3)). This is a Hausdor topology
on R since it is ner than the standard topology. But R
K
is not regular. The set K is closed and 0 , K.
Suppose that U 0 and V K are disjoint open sets. We may choose U to be a basis open set. Then U
must be of the form U = (a, b) K for the other basis sets containing 0 intersect K. Let k be a point in
(a, b) K (which is nonempty). Since k is in the open set V , there is a basis open set containing k and
contained in V ; it must be of the form (c, d). But U V ((a, b) K) (c, d) and ((a, b) K) (c, d) ,=
for cardinality reasons. This is a contradiction
64 3. REGULAR AND NORMAL SPACES
3. Normal spaces
Normal spaces can be characterized in two ways.
3.11. Corollary. Let X be a T
1
-space. Then the following conditions are equivalent:
(1) X is normal: For any pair of disjoint closed sets A, B X there exist disjoint open sets
U, V X such that A U and B V .
(2) (Urysohns characterization of normality) For any pair of disjoint closed sets A, B X there
exists a continuous function f : X [0, 1] such that f(A) = 0 and f(B) = 1.
(3) (Tietzes characterization of normality) For any closed subset A of X and any continuous func-
tion f : A [0, 1] there exists a continuous function F : X [0, 1] such that F[A = f.
The proof relies on Urysohns lemma (3.12) and Tietzes extension theorem (3.15).
3.12. Theorem (Urysohn lemma). Let X be a normal space and let A and B be disjoint closed subsets
of X. Then there exists a continuous function (a Urysohn function) f : X [0, 1] such that f(a) = 0 for
a A and f(b) = 1 for b B.
Proof. We shall recursively dene open sets U
r
X for all r Q[0, 1] such that (make a drawing!)
(3.13) r < s = A U
0
U
r
U
r
U
s
U
s
U
1
X B
To begin, let U
1
be any open set such that A U
1
X B, for instance U
1
= X B. By normality
(3.5.(2)) there is an open set U
0
such that A U
0
U
0
U
1
. To proceed, arrange the elements of
Q [0, 1] into a sequence
Q [0, 1] = r
0
= 0, r
1
= 1, r
2
, r
3
, r
4
, . . .
such that the rst two elements are r
0
= 0 and r
1
= 1. Assume that the open sets U
r
i
satisfying condition
(3.13) have been dened for i n, where n 1. We shall now dene U
r
n+1
. Suppose that if we put the
numbers r
0
, r
1
, . . . , r
n
, r
n+1
in order
0 = r
0
< < r

< r
n+1
< r
m
< < r
1
= 1
the immediate predecessor of r
n+1
is r

and r
m
is the immediate successor. Then U
r

U
r
m
by (3.13).
By normality (3.5.(2)) there is an open set U
r
n+1
such that U
r

U
r
n+1
U
r
n+1
U
r
m
. The sets U
r
i
,
i n + 1, still satisfy (3.13).
We are now ready to dene the function. Consider the function f : X [0, 1] given by
f(x) =
_
infr Q [0, 1] [ x U
r
x U
1
1 x X U
1
Then f(B) = 1 by denition and f(A) = 0 since A U
0
. But why is f continuous? It suces (2.50) to
show that the subbasis intervals (2.13) [0, a), a > 0, and (b, 1], b < 1, have open preimages. Since
f(x) < a r < a: x U
r
x
_
r<a
U
r
f(x) > b r

> b: x , U
r
r > b: x , U
r
x
_
r>b
(X U
r
)
the sets
f
1
([0, a)) =
_
r<a
U
r
and f
1
((b, 1]) =
_
r>b
(X U
r
)
are open.
3.14. Example. Let X = R and let A = (, 1] and B = [2, ). If we let U
r
= (, r) for
r Q [0, 1] then
f(x) =
_

_
0 x 0
x 0 < x < 1
1 1 x
is the Urysohn function with f(A) = 0 and f(B) = 1.
3.15. Theorem (Tietze extension theorem). Every continuous map from a closed subspace A of a
normal space X into (0, 1), [0, 1) or [0, 1] can be extended to X.
3. NORMAL SPACES 65
3.16. Lemma. Let X be a normal space and A X a closed subspace. For r > 0 and any continuous
map f
0
: A [r, r] there exists a continuous map g : X R such that
(3.17) x X: [g(x)[
1
3
r and a A: [f
0
(a) g(a)[
2
3
r
Proof. Since the sets f
1
0
([r,
1
3
r]) A and f
1
0
([
1
3
r, r]) A are closed and disjoint in A they are
also closed and disjoint in X. Choose (3.12) a Urysohn function g : X [
1
3
r,
1
3
r] such that g(x) =
1
3
r
on f
1
0
([r,
1
3
r]) and g(x) =
1
3
r on f
1
0
([
1
3
r, r]). From
f
1
0
([r, r]) = f
1
0
([r,
1
3
r])
. .
g=
1
3
r
f
1
0
([
1
3
r,
1
3
r])
. .
|g|
1
3
r
f
1
0
([
1
3
r, r])
. .
g=
1
3
r
we see that the second inequality of (3.17) is satised.
Proof of 3.15. We shall rst prove the theorem for functions from A to [0, 1] or rather to the
homeomorphic interval [1, 1] which is more convenient for reasons of notation.
Given a continuous function f : A [1, 1]. We have just seen (3.16) that there exists a continuous
real function g
1
: X R such that
x X: [g
1
(x)[
1
3
and a A: [f(a) g
1
(a)[
2
3
Now Lemma 3.16 applied to the function f g
1
on A says that there exists a continuous real function
g
2
: X R such that
x X: [g
2
(x)[
1
3
2
3
and a A: [f(a) (g
1
(a) +g
2
(a))[
_
2
3
_
2
Proceeding this way we recursively (1.24) dene a sequence g
1
, g
2
, of continuous real functions on X
such that
(3.18) x X: [g
n
(x)[
1
3
_
2
3
_
n1
and a A: [f(a)
n

i=1
g
i
(a)[
_
2
3
_
n
By the rst inequality in (3.18), the series

n=1
g
n
(x) converges uniformly and the sum function F =

n=1
g
n
is continuous by the uniform limit theorem (2.104). By the rst inequality in (3.18),
[F(x)[
1
3

n=1
_
2
3
_
n1
=
1
3
3 = 1,
and by the second inequality in (3.18), F(a) = f(a) for all a A.
Assume now that f : A (1, 1) maps A into the open interval between 1 and 1. We know that
we can extend f to a continuous function F
1
: X [1, 1] into the closed interval between 1 and 1. We
want to modify F
1
so that it does not take the values 1 and 1 and stays the same on A. The closed sets
F
1
1
(1) and A are disjoint. There exists (3.12) a Urysohn function U : X [0, 1] such that U = 0 on
F
1
1
(1) and U = 1 on A. Then F = U F
1
is an extension of f that maps X into (1, 1).
A similar procedure applies to functions f : A [1, 1) into the half-open interval.
Proof of Corollary 3.11. The Urysohn lemma (3.12) says that (1) = (2) and the converse
is clear. The Tietze extension theorem (3.15) says that (1) = (3). Only eg (3) = (2) remains.
Assume that X is a T
1
-space with property (3). Let A, B be two disjoint closed subsets. The function
f : A B [0, 1] given by f(A) = 0 and f(B) = 1 is continuous (2.53.(5)). Let f : X [0, 1] be a
continuous extension of f. Then f is a Urysohn function for A and B.
Many familiar classes of topological spaces are normal.
3.19. Theorem. Compact Hausdor spaces are normal.
Proof. See 2.140.(2).
In fact, every regular Lindelof space is normal [8, Ex 32.4] [5, 3.8.2].
3.20. Theorem. Metrizable spaces are normal.
66 3. REGULAR AND NORMAL SPACES
Proof. Let X be a metric space with metric d and let A, B be disjoint closed sets. The continuous
function f : X [0, 1] given by
f(x) =
d(x, A)
d(x, A) +d(x, B)
is a Urysohn function with f(A) = 0 and f(B) = 1. This shows that X is normal (3.11).
3.21. Theorem. [5, Problem 1.7.4]. Linearly ordered spaces are normal.
Proof. We shall only prove the special case that every well-ordered space is normal. The half-open
intervals (a, b], a < b, are (closed and) open (2.31.(5)). Let A and B be two disjoint closed subsets and
let a
0
denote the smallest element of X. Suppose that neither A nor B contain a
0
. For any point a A
there exists a point x
a
< a such that (x
a
, a] is disjoint from B. Similarly, for any point b B there
exists a point x
b
< b such that (x
b
, b] is disjoint from A. The proof now proceeds as the proof (3.8) for
normality of R

. Suppose next that a

0
A B, say a
0
A. The one-point set a
0
= [a
0
, a
+
0
) is open
and closed (as X is Hausdor). By the above, we can nd disjoint open sets U, V such that Aa
0
U
and B V . Then A U a
0
and B V a
0
where the open sets U a
0
and V a
0
are
disjoint.
In particular, the well-ordered spaces S

and S

are normal (the latter space is even compact

Hausdor).
4. Second countable regular spaces and the Urysohn metrization theorem
We investigate closer the class of regular spaces. We start with an embedding theorem that is used
in other contexts as well.
3.22. An embedding theorem. We discuss embeddings into product spaces.
3.23. Definition. A set f
j
: X Y
j
[ j J of continuous functions is said to separate points and
closed sets if for any point x X and any closed subset C X we have
x , C = j J : f
j
(x) , f
j
(C)
3.24. Lemma. Let f : X Y be a map that separates points and closed sets. Assume also that f is
injective (eg that X is T
1
). Then f is an embedding.
Proof. If X is T
1
, points are closed so that f separates points, ie f is injective. For any point x X
and any closed set C X we have that
f(x) f(C)
f separates
= x C = f(x) f(C)
so we get that f(X) f(C) = f(C). But this equality says that f(C) is closed in f(X). Hence the
bijective continuous map f : X f(X) is closed, so it is a homeomorphism.
3.25. Theorem (Diagonal embedding theorem). Let f
j
: X Y
j
[ j J be a family of continuous
functions that separates points and closed sets. Assume that X is T
1
or that at least one of the functions
f
j
is injective. Then the diagonal map f = (f
j
): X

jJ
Y
j
is an embedding.
Proof. Also the map f separates points and closed sets because
f(x) f(C) f(x)

f
j
(C)
2.66
f(x)

f
j
(C)
j J : f
j
(x) f
j
(C)
(f
j
) separates
= x C
for all points x X and all closed subsets C X. The theorem now follows from 3.24.
In particular, if one of the functions f
j
is injective and separates points and closed sets then f = (f
j
)
is an embedding. For instance, the graph X XY : x (x, g(x)) is an embedding for any continuous
map g : X Y .
4. SECOND COUNTABLE REGULAR SPACES AND THE URYSOHN METRIZATION THEOREM 67
3.26. A universal second countable regular space. We say that a space X is universal for
some property if X has this property and any space that has this property embeds into X.
3.27. Theorem (Urysohn metrization theorem). The following conditions are equivalent for a second
countable space X:
(1) X is regular
(2) X is normal
(3) X is homeomorphic to a subspace of [0, 1]

(4) X is metrizable
The Hilbert cube [0, 1]

is a universal second countable metrizable (or normal or regular) space.

Proof. (1) = (2): Let X be a regular space with a countable basis B. We claim that X is normal.
Let A and B be disjoint closed sets in X. By regularity (3.5.(1)), each point a A X B has a basis
neighborhood U
a
B such that a U
a
U
a
X B. Let U
1
, U
2
, . . . be the elements in the image of
the map A B: a U
a
. Then
A

_
n=1
U
n
and U
n
B = for n = 1, 2 . . .
Similarly, there is a sequence V
1
, V
2
, . . . of basis open sets such that
B

_
n=1
V
n
and V
n
A = for n = 1, 2 . . .
The open sets

n=1
U
n
and

n=1
V
n
may not be disjoint. Consider instead the open sets
U

1
= U
1
V
1
V

1
= V
1
U
1
U

2
= U
2
(V
1
V
2
) V

2
= V
2
(U
1
U
2
)
.
.
.
.
.
.
U

n
= U
n
(V
1
V
n
) V

n
= V
n
(U
1
U
n
)
Even though we have removed part of U
n
we have removed no points from A from U
n
, and we have
removed no points from B from V
n
. Therefore the open sets U

n
still cover A and the open sets V

n
still
cover B:
A

_
n=1
U

n
and B

_
n=1
V

n
and in fact these sets are disjoint:

_
n=1
U

_
n=1
V

n
=
_
_
mn
U

m
V

n
_

_
_
m>n
U

m
V

n
_
=
because U

m
U
m
does not intersect V

n
XU
m
if m n and U

m
XV
n
does not intersect V

n
V
n
if m > n. (Make a drawing of U
1
, U
2
and V
1
, V
2
.)
(2) = (3): Let X be a normal space with a countable basis B. We show that there is a countable set
f
UV
of continuous functions X [0, 1] that separate points and closed sets. Namely, for each pair
U, V of basis open sets U, V B such that U V , choose a Urysohn function f
UV
: X [0, 1] (3.12)
such that f
UV
(U) = 0 and f
UV
(X V ) = 1. Then f
UV
separates points and closed sets: For any
closed subset C and any point x , C, or x X C, there are (3.5) basis open sets U, V such that
x U U V X C (choose V rst). Then f
UV
(x) = 0 and f
UV
(C) = 1 so that f
UV
(x) , f
UV
(C).
According to the Diagonal embedding theorem (3.25) there is an embedding X [0, 1]

with the f
UV
as coordinate functions.
(3) = (4): [0, 1]

is metrizable and any subspace of a metrizable space is metrizable. (2.93).

(4) = (1): Any metrizable space is regular, even normal (3.20).
The point is that in a second countable regular hence normal space there is a countable set of
(Urysohn) functions that separate points and closed sets. Therefore such a space embeds in the Hilbert
cube.
68 3. REGULAR AND NORMAL SPACES
We have (3.27, 3.2) the following identities
_
2nd countable
regular
_
=
_
2nd countable
normal
_
=
_
2nd countable
metrizable
_
=
_
Countable dense subset
metrizable
_
=
_
Lindelof
metrizable
_
=
_
Subspace of
Hilbert cube
_
between classes of topological spaces.
5. COMPLETELY REGULAR SPACES AND THE STONE

CECH COMPACTIFICATION 69
5. Completely regular spaces and the Stone

Cech compactication
There is no version of the Tietze extension theorem (3.15) for regular spaces, ie it is in general not
true that continuous functions separate closed sets and points in regular spaces. Instead we have a new
class of spaces where this is true.
3.28. Definition. A space X is completely regular if points are closed in X and for any closed
subset C of X and any point x , C there exists a continuous function f : X [0, 1] such that f(x) = 0
and f(C) = 1.
Clearly
normal = completely regular = regular = Hausdor = T
1
and none of these arrows reverse (3.9, [8, Ex 33.11]).
In a completely regular space there are enough continuous functions X [0, 1] to separate points
and closed sets.
It is easy to see that any subspace of a completely regular space is completely regular [8, 33.2].
3.29. Theorem. The following conditions are equivalent for a topological space X:
(1) X is completely regular
(2) X is homeomorphic to a subspace of [0, 1]
J
for some set J
(3) X is homeomorphic to a subspace of a compact Hausdor space
Proof. (1) = (2): If X is completely regular then the set ((X) of continuous maps X [0, 1]
separates points and closed sets. The evaluation map
: X

jC(X)
[0, 1],
j
((x)) = j(x), j ((X), x X,
is therefore an embedding (3.25).
(2) = (3): [0, 1]
J
is compact Hausdor by the Tychono theorem (2.149).
(3) = (1): A compact Hausdor space is normal (3.19), hence completely regular and subspaces of
completely regular spaces are completely regular.
Closed subspaces of compact Hausdor spaces are compact Hausdor (2.140), open subspaces are
locally compact Hausdor (2.170), and arbitrary subspaces are completely regular (3.29).
3.30. Corollary. (Cf 2.47, 3.6) Any subspace of a completely regular space is completely regular.
Any product of completely regular spaces is completely regular.
Proof. The rst part is easily proved [8, 33.2] (and we already used it above). The second part
follows from (3.29) because (2.61) the product of embeddings is an embedding.
3.31. Corollary. Any locally compact Hausdor space is completely regular.
Proof. Locally compact Hausdor spaces are open subspaces of compact spaces (2.170). Completely
regular spaces are subspaces of compact spaces (3.29).
3.32. Example (A normal space with a non-normal subspace). Take any completely regular but not
normal space (for instance (3.9) R

) and embed it into [0, 1]

J
for some set J (3.29). Then you have
an example of a normal (even compact Hausdor) space with a non-normal subspace.
3.33. The Stone

Cech construction. For any topological space X let ((X) denote the set of
continuous maps j : X I of X to the unit interval I = [0, 1] and let
: X

jC(X)
I = map(((X), I)
be the continuous evaluation map given by (x)(j) = j(x) or
j
= j for all j ((X). (The space
map(((X), I) is a kind of double-dual of X). This construction is natural: For any continuous map
f : X Y of X into a space Y , there is an induced map ((X) ((Y ): f

of sets and yet an induced

continuous map (2.65)
(3.34) map(((X), I) =

jC(X)
I
f

kC(Y )
I = map(((Y ), I)
70 3. REGULAR AND NORMAL SPACES
such that the diagram
(3.35) X
f

jC(X)
I
f

k
=
kf

L
L
L
L
L
L
L
L
L
L
L

kC(Y )
I

k
.r
r
r
r
r
r
r
r
r
r
r
I
commutes: The lower triangle commutes by the denition of the map f

(2.65) and then the upper

square commutes because
k
f

X
=
kf

X
= kf =
k

Y
f. Put X = (X) and dene : X X
to be the corestriction of : X

jC(X)
I to X. Then X is a compact Hausdor space (it is a
closed subspace of the compact (2.149) Hausdor space

I) and is, by design, a continuous map with

a dense image X in X. This construction is natural: For any continuous map f : X Y , the induced
map f

(3.34) takes X into Y for

f

(X) = f

(
X
X) f

X
X
(3.35)
=
Y
f(X)
Y
Y = Y
and thus we obtain from (3.35) a new commutative diagram of continuous maps
(3.36) X
f

X
f

Y
where f : X Y is the corestriction to Y of the restriction of f

to X. (This makes into a

functor with a natural transformation from the identity functor from the category of topological spaces
to the category of compact Hausdor spaces.)
The next result says that the map X X is the universal map from X to a compact Hausdor
space.
3.37. Theorem (Stone

Cech compactication). Let X be a topological space.

(1) The map X X is a continuous map of X into a compact Hausdor space.
(2) For any continuous map f : X Y of X into a compact Hausdor space Y there exists a unique
continuous map f : X Y such that the diagram
(3.38) X

B
B
B
B
B
B
B
B
f

Y
X
f

|
|
|
|
|
|
|
|
commutes. (We say that f : X Y factors uniquely through X.)
(3) Suppose X X is a map of X to a compact Hausdor space X such that any map of X to
a compact Hausdor space factors uniquely through X. Then there exists a homeomorphism
X X such that
X
|
|
|
|
|
|
|
|

B
B
B
B
B
B
B
B
X


X
commutes.
(4) If X is completely regular then X X is an embedding. If X is compact Hausdor then
X X is a homeomorphism.
Proof. (4): Since the corestriction of an embedding is an embedding (2.61) and X

I is an
embedding when X is completely regular (3.29), also X X is an embedding. If X is compact
Hausdor, X is normal (3.19), hence completely regular, so we have just seen that : X X is an
embedding. The image of this embedding is closed (2.141.(1)) and dense. Thus the embedding is bijective
so it is a homeomorphism.
6. MANIFOLDS 71
(1) and (2): If Y is compact Hausdor then by item (4)
Y
is a homeomorphism in diagram (3.36) and
hence f =
1
Y
f is a possibility in (3.38). It is the only possibility [8, Ex 18.13], for the image of X
is dense in X.
(3) Let X X be a map to a compact Hausdor space satisfying the above universal property. Then
there exist continuous maps X

X that, by uniqueness, are inverse to each other. (All universal
constructions are essentially unique.)
If X is completely regular then X X is a compactication (3.37.(4)) and it is called the Stone

Cech compactication of X. Conversely, if X has a compactication then X is homeomorphic to a

subspace of a compact space and therefore (3.29) completely regular. We state these observations in
3.39. Lemma. X has a compactication (2.166) if and only if X is completely regular.
The Stone

Cech compactication X of a completely regular space X is the maximal compactication

in the sense that for any other compactication X cX there is a closed quotient map X cX map
such that
X
|
|
|
|
|
|
|
|

B
B
B
B
B
B
B
B
X

cX
commutes. The map X cX is closed by 2.141 and surjective because its image is closed and dense.
The Alexandro compactication X of a noncompact locally compact Hausdor space X is the
minimal compactication in the sense that X = cX/(cXX) for any other compactication X cX.
Indeed we saw in 2.167 that there is a closed quotient map cX X, taking cXX to , such that
X
|
|
|
|
|
|
|
|

C
C
C
C
C
C
C
C
cX

X
commutes. (What is the minimal compactication of a compact Hausdor space?)
For instance, there are quotient maps
[0, 1]

Q
Q
Q
Q
Q
Q
Q
R

p
p
p
p
p
p

O
O
O
O
O
O
O
R = S
1
C

l
l
l
l
l
l
l
l
of compactications of R. Here
C = 0 [1, 1] (x, sin
1
x
) [ 0 < x
1
L
is the Warsaw circle, a compactication of R with remainder C R = [1, 1], which is obtained by
closing up the closed topologists sine curve S by (a piece-wise linear) arc from (0, 0) to (
1
, 0). In
particular, C/[1, 1] = S
1
. More generally, there are quotient maps R
n
[0, 1]
n
R
n
= S
n
of
compactications of euclidean n-space R
n
.
Investigations of the Stone

Cech compactication of the integers Z raise several issues of a very

fundamental nature [13, 14, 11] whose answers depend on your chosen model of set theory.
6. Manifolds
3.40. Definition. A manifold is a locally euclidean second countable Hausdor space.
Manifolds are locally compact and locally path connected since they are locally euclidean. All locally
compact Hausdor spaces are (completely) regular. Thus manifolds are second countable regular spaces
and hence they are normal, metrizable, Lindelofand they have countable dense subsets.
The line with two zeroes is an example of a locally euclidean space that is not Hausdor. The long
line [8, Ex 24.12] is a locally euclidean space that is not 2nd countable.
The nsphere S
n
and the real projective nspace RP
n
(2.83) are examples of manifolds. S
n
lies
embedded in R
n+1
from birth but what about RP
n
? Does RP
n
embed in some euclidean space?
3.41. Theorem (Embeddings of compact manifolds). Any compact manifold embeds in R
N
for some
N.
72 3. REGULAR AND NORMAL SPACES
Proof. We may assume that M is connected. Let n be the dimension of M (we assume that this
is well-dened). Let B = x R
n
[ [x[ 1 be the unit ball in R
n
. Cover M by nitely many closed
sets B
1
, . . . , B
k
such that each B
i
is homeomorphic to B and such that the interiors of the B
i
cover M.
Let f
i
: M S
n
be the continuous map obtained by collapsing the complement to the interior of B
i
to
a point. The map f = (f
1
, . . . , f
k
): M S
n
S
n
is injective: If x lies in the interior of B
i
then
f
i
(x) ,= f
i
(y) for all y ,= x. Since M is compact and (S
n
)
k
Hausdor, f is an embedding. Finally, each
sphere S
n
embeds into R
n+1
so that (S
n
)
k
embeds into (R
n+1
)
k
.
In particular RP
2
(2.83) embeds into some euclidean space. Which one?
There is up to homeomorphism just one compact manifold of dimension 1, the circle. Compact
manifolds of dimension 2 are described by a number, the genus, and orientability. In particular, any simply
connected 2-dimensional compact manifold is homeomorphic to S
2
. We do not know (December 2003)
if there are any simply connected compact 3-dimensional manifolds besides S
3
(Poincare conjecture).
Classication of 4-dimensional compact manifolds is logically impossible.
Any manifold admits a partition of unity. We shall prove the existence in case of compact manifolds.
3.42. Theorem (Partition of unity). Let X = U
1
U
k
be a nite open covering of the normal
space X. Then there exist continuous functions
i
: X [0, 1], 1 i k, such that
(1)
i
> 0 U
i
(2)

k
i=1

i
(x) = 1 for all x X.
Proof. We show rst that there is an open covering V
i
of X such that V
i
U
i
(we shrink the
sets of the covering). Since the closed set X (U
2
U
k
) is contained in the open set U
1
there is an
open set V
1
such that
X (U
2
U
k
) V
1
V
1
U
1
by normality (3.5). Now V
1
U
2
U
k
is an open covering of X. Apply this procedure once again to
nd an open set V
2
such that V
2
U
2
and V
1
V
2
U
3
U
k
is still an open covering of X. After
nitely many steps we have an open covering V
i
such that V
i
U
i
for all i.
Do this one more time to obtain an open covering W
i
such that W
i
W
i
V
i
V
i
U
i
for
all i. Now choose a Urysohn function (3.12)
i
: X [0, 1] such that
i
(W
i
) = 1 and
i
(X V
i
) = 0.
Then x [
i
(x) > 0 V
i
U
i
and (x) =

i
(x) > 1 for any x X since x W
i
for some i.
Hence
i
=

i

is a well-dened continuous function on M taking values in the unit interval such that

k
i=1

i
(x) =

k
i=1

i
(x)
(x)
=
1
(x)

k
i=1

i
(x) =
(x)
(x)
= 1.
CHAPTER 4
Relations between topological spaces
s
e
q
u
e
n
t
i
a
l
l
y
c
o
m
p
a
c
t
+
m
e
t
r
i
z
a
b
l
e

c
o
u
n
t
a
b
l
e
d
e
n
s
e
s
u
b
s
e
t
+
m
e
t
r
i
z
a
b
l
e

l
i
m
i
t
p
o
i
n
t
c
o
m
p
a
c
t
+
m
e
t
r
i
z
a
b
l
e

+
1
s
t
c
o
u
n
t
a
b
l
e
+
T
1

2
n
d
c
o
u
n
t
a
b
l
e

T
h
m
3
0
.
3

T
h
m
3
0
.
3

L
i
n
d
e
l
o
f
+
m
e
t
r
i
z
a
b
l
e
.
c
o
m
p
a
c
t

.
T
h
m
2
8
.
1

l
o
c
a
l
l
y
c
o
m
p
a
c
t
1
s
t
c
o
u
n
t
a
b
l
e
T
h
m
3
0
.
1

S
e
q
u
e
n
c
e
l
m
a
c
o
m
p
a
c
t
H
a
u
s
d
o
r

l
o
c
a
l
l
y
c
o
m
p
a
c
t
H
a
u
s
d
o
r

E
x
2
2
.
7

m
e
t
r
i
z
a
b
l
e

E
x
3
2
.
7

c
o
m
p
l
e
t
e
l
y
n
o
r
m
a
l

n
o
r
m
a
l
T
h
m
3
3
.
1

c
o
m
p
l
e
t
e
l
y
r
e
g
u
l
a
r

t
o
p
g
r
p
E
x
3
3
.
1
0
.
r
e
g
u
l
a
r
2
n
d
c
o
u
n
t
a
b
l
e
T
h
m
3
4
.
1

w
e
l
l
-
o
r
d
e
r
e
d
t
o
p
o
l
o
g
y

E
x
2
3
.
5

o
r
d
e
r
t
o
p
o
l
o
g
y
T
h
m
3
2
.
4

r
e
g
u
l
a
r

m
a
n
i
f
o
l
d
E
x
3
6
.
1

J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
J
t
o
t
a
l
l
y
d
i
s
c
o
n
n
e
c
t
e
d
l
i
n
e
a
r
c
o
n
t
i
n
u
u
m

T
h
m
2
4
.
1

H
a
u
s
d
o
r

l
o
c
a
l
l
y
c
o
n
n
e
c
t
e
d
c
o
n
n
e
c
t
e
d
T
1
l
o
c
a
l
l
y
p
a
t
h
c
o
n
n
e
c
t
e
d

p
a
t
h
c
o
n
n
e
c
t
e
d
p
.
1
5
5

73
Bibliography
[1] Colin C. Adams, The knot book, W. H. Freeman and Company, New York, 1994, An elementary introduction to the
mathematical theory of knots. MR 94m:57007
[2] Nicolas Bourbaki, General topology. Chapters 14, Elements of Mathematics, Springer-Verlag, Berlin, 1998, Translated
from the French, Reprint of the 1989 English translation. MR 2000h:54001a
[3] Glen E. Bredon, Topology and geometry, Graduate Texts in Mathematics, vol. 139, Springer-Verlag, New York, 1993.
MR 94d:55001
[4] John P. Burgess, Book review, Notre Dame J. Formal Logic 44 (2003), no. 3, 227251. MR MR675916 (84a:03007)
[5] Ryszard Engelking, General topology, second ed., Sigma Series in Pure Mathematics, vol. 6, Heldermann Verlag, Berlin,
1989, Translated from the Polish by the author. MR 91c:54001
[6] Ryszard Engelking and Karol Sieklucki, Topology: a geometric approach, Sigma Series in Pure Mathematics, vol. 4,
Heldermann Verlag, Berlin, 1992, Translated from the Polish original by Adam Ostaszewski. MR 94d:54001
[7] Allen Hatcher, Algebraic topology, Cambridge University Press, Cambridge, 2002. MR 2002k:55001
[8] James R. Munkres, Topology. Second edition, Prentice-Hall Inc., Englewood Clis, N.J., 2000. MR 57 #4063
[9] J.-P. Serre, A course in arithmetic, Springer-Verlag, New York, 1973, Translated from the French, Graduate Texts in
Mathematics, No. 7. MR MR0344216 (49 #8956)
[10] Edwin H. Spanier, Algebraic topology, Springer-Verlag, New York, 1981, Corrected reprint. MR 83i:55001
[11] Juris Steprans, The autohomeomorphism group of the

Cech-Stone compactication of the integers, Trans. Amer. Math.
Soc. 355 (2003), no. 10, 42234240 (electronic). MR 1 990 584
[12] George Tourlakis, Lectures in logic and set theory. Vol. 2, Cambridge Studies in Advanced Mathematics, vol. 83,
Cambridge University Press, Cambridge, 2003, Set theory. MR 2004a:03003
[13] Jan van Mill, An introduction to , Handbook of set-theoretic topology, North-Holland, Amsterdam, 1984, pp. 503
567. MR 86f:54027
[14] Russell C. Walker, The Stone-

Cech compactication, Springer-Verlag, New York, 1974, Ergebnisse der Mathematik

und ihrer Grenzgebiete, Band 83. MR 52 #1595
[15] W. Hugh Woodin, The continuum hypothesis, Part I, Notices of the American Math. Soc. 48 (2001), no. 6, 567576.
[16] , The continuum hypothesis, Part II, Notices of the American Math. Soc. 48 (2001), no. 7, 681690.
75