You are on page 1of 41

1 Motivation

Consider the real numbers R. From calculus (or analysis), we know what it
means for a sequence of points to converge to a point in R, and we know
what it means for a function from R to R to be continuous. These ideas
have to do with nearness: continuous maps send pairs of nearby points to
pairs of nearby points and convergent sequences eventually get closer and
closer to their limits.
Part of the motivation behind topology is to formalize and distill the
idea of nearness. Explaining this aspect of the structure of R will allow
us to generalize this notion of nearness to other mathematical objects
which do not have the extra structure (geometric structure, order structure,
algebraic structure) that R has.
Once we have an abstract model for what non-geometric nearness in
space means, we can use this to understand many spaces other than R,
including graphs, Klein bottles, Cantor sets, spaces of functions, and others.
We can also then understand operations on these spaces, such as taking
products, cutting, and gluing.
In analysis, when we want to talk about points that are near a point p,
we pick an > 0 and consider the set of all points whose distance to p is less
than . This is intrinsically geometric. The usual approach in topology is to
keep track of all the possible sets of points that we might want to think of
as being the set of points near p, however, to forget which this set would
correspond to.
A neighborhood S of a point p in R is a subset of R, containing p, such
that S contains all the points in R that are close enough to p. An open
set U is a subset of R that is a neighborhood of every point it contains. By
keeping track of either open sets or neighborhoods, we can have a notion of
nearness without using any geometry.
Example 1.1. In the real numbers R, the neighborhoods of a point p in
R are those subsets of R which contain an open interval containing p. The
open sets in R are the subsets of R that can be written as unions of open
interval, indexed over some possibly innite index set I:
U =

iI
(a
i
, b
i
).
The empty set is also open.
We dont consider closed intervals to be open sets because a closed inter-
val is not a neighborhood of its endpoints: there are points in R arbitrarily
1
close to the endpoint that are not in the closed interval. Having dened
open sets and neighborhoods, it is straightforward to give denitions for
continuity and convergence for R that make no mention of any > 0.
2 Basics about spaces
2.1 Axioms and denitions
The rst goal is to provide a general denition of a topological space, such
that the real numbers will be an example. We will work with the most
common formulation of the denition, which uses open sets.
Denition 2.1. A topological space (X, T ) is a set X together with a set
T of subsets of X. The elements of T are called open sets and satisfy the
following axioms:
1. Arbitrary unions of open sets are open.
2. Finite intersections of open sets are open.
3. The set X and the empty set are both open.
Given a set X, a topology on X is a topological space structure (X, T ).
Denition 2.2. Having dened topological spaces, we can dene the fol-
lowing vocabulary:
A point in a topological space (X, T ) is simply an element of X.
When it is necessary to refer to X as a set (and not as a space), the
set X is called the point set of (X, T ).
A subset A X is closed if its complement XA is open in (X, T ).
A neighborhood of a point p in a space (X, T ) is a set S, p S X,
such that there is an open set U with p U S. An open neighborhood
of p is simply an open set containing p.
Having dened neighborhoods, we can now characterize open sets in
terms of neighborhoods.
Proposition 2.3. A set U is open if and only if for each p U, U is a
neighborhood of p.
2
Proof. If U is open, then for each p in U, U is a (nonproper) superset of an
open set containing p (that is, U is the open set containing p). So U is a
neighborhood of each p in U.
Conversely, suppose that U is a neighborhood of each point p in U. Then
for each p in U, there is an open set V
p
with p V
p
U. Let V =

pU
V
p
.
Clearly U V , since any p U is in V
p
and therefore in V . However, if
p V then p is in some V
q
for some q U; each V
q
is a subset of U, so this
p is in U. So V U and therefore U = V . Then U is a union of open sets
and is therefore open.
This proposition is the link between open sets and neighborhoods. It can
be used to prove a characterization of topological spaces with neighborhoods
as a primitive notion. This is left as an exercise.
Now we show that the standard topology on R ts our denition.
Proposition 2.4. The set of open sets for R specied in Example 1.1 gives
a topology on R.
Proof. Well show this set of open sets satises the axioms for open sets from
Denition 2.1. We have dened an open set as a union of open intervals.
Since a union of unions of open intervals is a bigger union of open intervals,
a union of open sets is open.
Now suppose U =

iI
(a
i
, b
i
) and V =

jJ
(c
i
, d
i
) are open sets. Then
U V = (

i
(a
i
, b
i
)) (

i
(c
i
, d
i
)) =

iI,jJ
((a
i
, b
i
) (c
j
, d
j
)).
Since each (a
i
, b
i
) (c
j
, d
j
) is either an open interval or is empty, we have
written U V as a union of open intervals and it is open. Inductively any
nite intersection of open sets is open.
Since R =
iN
(i, i) and weve declared the empty set to be open, the
third axiom is also satised.
Example 2.5. A few more examples of topological spaces:
Let X be any nonempty set. The discrete topology on X is the topo-
logical structure where all subsets of X are open.
Let X be any nonempty set. The indiscrete topology on X is the
structure where only X and are open.
3
Let X be any innite set. Declare the closed subsets of X to be
all the nite sets (so the open subsets of X are the sets with nite
complement). This denes a topology, called the nite complement
topology on X.
Note that we have dened four dierent topologies on R. Provided a
set X has more than one point, there are many dierent topologies on X.
Therefore it is necessary to specify the full topological structure when intro-
ducing a space. However, once weve specied the structure, it is rare that
we specically refer to the set of open sets T ; instead we simply use the
word open to specify its elements. For this reason, its common practice
to refer to a space (X, T ) using only the symbol for its point set X.
2.2 Subspaces
Any subset of a topological space inherits the structure of a space.
Denition 2.6. Let (X, T ) be a space and let S be a subset of X. The
subspace topology on S is the one whose open sets are exactly the intersec-
tions of open subsets of X with S. Precisely, let o = U S[U T . Then
(S, o) is the subspace topology on S.
This topology on S is also called the induced topology.
Proposition 2.7. Let X be a space and let S be a subset of X. The sub-
space topology on S is a topology on S (that is, it satises the axioms for a
topological space).
The proof is straightfoward.
Example 2.8. Some examples of subspaces of R with the standard topol-
ogy:
The various kinds of intervals [0, 1], [0, 1), (0, 1], (0, 1).
The set 0 1/n[n N.
The rational numbers Q.
The irrational numbers RQ.
4
2.3 Subbases and bases
If you want give someone the denition of a particular topological space, you
must specify the point set and the set of all open sets. One way to specify
the set of all open sets is to let the axioms do the work.
Proposition 2.9. Suppose X is a set and ( is a set of subsets of S. Let
B
0
be the set of all intersections of nitely many elements of (, and let
B = B
0
X, . Let T be the set of all unions of elements of B. Then T
is the set of open sets for a topology on X.
Proof. Clearly unions of elements of T are in T . Let U, V T , with U =

iI
U
i
and V =

jJ
V
j
for some U
i
, V
j
B. Then
U V = (

iI
U
i
) (

jJ
V
j
) =

iI,jJ
(U
i
V
j
).
Since each U
i
V
j
is clearly in B, it follows that U V is in T and T is
closed under nite intersections. Of course, since X and are in B, they
are in T as well, and T is the set of open sets for a topology on X.
Denition 2.10. Let (X, T ) be a topological space. A set B of open sets
in X is a basis for (X, T ) if the set of all unions of elements of B is T . A
set ( of open sets in X is a subbasis for (X, T ) if the set of all intersections
of nitely many elements of X, together with X and , form a basis for
(X, T ).
So Proposition 2.9 asserts that any set of subsets of a set X is a sub-
basis for some topology on X. In particular, one can specify a topology by
specifying a subbasis.
Proposition 2.11. Let X be a set and suppose B is a set of subsets of X.
Suppose for any U, V B, the intersection U V can be written as a union
of members of B. Further suppose that each point in X is in some U in B.
Then B is a basis for some topology on X.
Proof. Dene T to be the set of all possible unions of sets of members of
B. Clearly T satises the axiom about unions. For U, V T , we write
U =

iI
U
i
and V =

jJ
V
j
where all the U
i
and V
j
are in B. Then U V
is

iI,jJ
(U
i
V
j
). Since each U
i
V
j
can be written as a union of members
of B, this U V can be written as such a union and inductively, T satises
the axiom about unions. Since each p X is in some set in B, the union
of all members of B is T , and therefore X is in T . Since is the empty
union, is in T . So T satises the axioms for a topology, and B is a basis
for T .
5
Example 2.12. Bases and subbases can be used to describe the spaces we
have already seen:
The set of open intervals is a basis for the standard topology on R.
The set of singleton subsets is a basis for the discrete topology on any
set X.
The topology on a set X with the subbasis
Xp[p X
is the nite complement topology.
If S is a subset of a space X and B is a basis for X, then SU[U B
is a basis for S with the subspace topology.
3 Continuity and homeomorphisms
Often in mathematics, one is interested not only in studying mathematical
objects, but also in studying functions between mathematical objects that
preserve some of the structure. In topology, we mainly consider continuous
functions.
Denition 3.1. Suppose X and Y are topological spaces. A function
f : X Y is continuous if for every open subset U of Y , the preimage
f
1
(U) is an open subset of X.
This denition seems a little mysterious at rst. In the denition
for the continuity of a function f : R R, we say f is continuous if for
every point p, it pulls back every small open interval around f(p) to a set
containing a small open interval around p. This easily translates to saying
that a continuous function f pulls back every neighborhood of f(p) to a
neighborhood of p. Since an open set is a neighborhood of any of the points
it contains, continuity should mean that a function pulls back open sets to
open sets.
Continuity means nearby points get sent to nearby points. However close
together I want points to end up, if they start out close enough together, the
function sends them to points that close together. however small an open
set I want to hit with my function, the set of points that hit it are still an
open set.
Bases make it more convenient to check that a function is continuous.
6
Proposition 3.2. Suppose X and Y are topological spaces and Y has a
basis B. Let f : X Y be a function and suppose f
1
(U) is open for each
basis element U B. Then f is continuous.
Proof. Let V Y be open. Write V as a union of elements U
i

iI
of B:
V =

iI
U
i
.
Then:
f
1
(V ) = f
1
(

iI
U
i
) =

iI
f
1
(U
i
),
which is open.
Some near-trivial observations about continuity:
Proposition 3.3. For any space X, the identity map from X to itself
is continuous.
Compositions of continuous maps are continuous: given continuous
maps between spaces f : X Y and g : Y Z, the composition g
f : X Z is continuous.
If S is a subspace of a space X, then the inclusion S X is continu-
ous.
If X has the discrete topology, then any function from X to any space
Y is continuous; the discrete topology on X is the only topology on X
with this property.
If X has the indiscrete topology, then any function from any space Y
to X is continuous; the indiscrete topology on X is the only topology
on X with this property.
Denition 3.4. Suppose X and Y are topological spaces. A homeomor-
phism from X to Y is an invertible, continuous function with continuous
inverse. The spaces X and Y are homoemorphic if a homeomorphism exists
between them.
Recall from set theory that a function is invertible if and only if it is one-
to-one and onto (a bijection). Warning: A continuous, invertible function
does not necessarily have continuous inverse. For example, consider any
bijection from a space with the discrete topology to a space that does not
have the discrete topology.
7
The relation of being homeomorphic is an equivalence relation on spaces
that is very strongin some sense, homeomorphism is the right notion of
sameness for spaces. We will proceed to justify this remark.
Suppose (X, T ) is a space, Y is a set, and f : X Y is a map of sets
that is one-to-one and onto. Dene f(T ) to be the set f(U)[U T . Call
(Y, f(T )) the relabeling of (X, T ) by f.
Proposition 3.5. The relabeling (Y, f(T )) is a topological space and the
relabeling map f : (X, T ) (Y, f(T )) is a homeomorphism.
Proof. For any U
i

iI
elements of T , we have
f(

iI
U
i
) =

iI
f(U
i
).
This implies that f(T ) satises the axiom about unions is satised. For any
U, V T , we have
f(U V ) = f(U) f(V )
since f is one-to-one. So f(T ) satises the axiom about intersections. Fi-
nally, since f is onto, f(X) = Y and Y f(T ). So (Y, f(T )) is a space.
The map f is continuous since f
1
(f(U)) = U for any U in T . For
U T , the pullback by f
1
is f(U), which is clearly in f(T ). So f is a
homeomorphism.
When we study a topological space, we are not usually interested in
specic points, or in the names for specic points. The map f is giving a
new name to each point of X, but is not otherwise changing the space.
In fact, there is a converse to the observation the relabeling is a homeo-
morphism:
Proposition 3.6. Let f : (X, T ) (Y, o) be a homeomorphism. Then
o = f(T ). In particular, (Y, o) is equal to the relabeling of (X, T ) by f.
Proof. Suppose U o. Then since f is continuous, f
1
(U) T . Then
U = f(f
1
(U)) is in f(T ).
Now suppose U f(T ). Then U = f(V ) for V T . Then since f
1
is continuous with respect to o, f(V ) (the pullback by f
1
(V )) is in o. So
o = f(T ).
To show that two spaces are homeomorphic, usually one has to construct
a homeomorphism between them. To show that spaces are not homeomor-
phic, one has to show that the spaces dier with respect to some property
8
that is preserved under homeomorphism. Since a homeomorphism is a rela-
beling of the points in a space, any property that is dened using only the
topological structure of a space will be preserved under homeomorphism.
Properties not preserved under homeomorphisms are not part of the topo-
logical structure of a space, but come from some kind of additional structure
that we may have imposed separately: geometric structure, algebraic struc-
ture, order structure, names of special points.
Simple examples of topological properties of spaces include:
Existence of nite closed sets.
Existence of innite closed sets.
Existence of nite open sets.
Much of this course will be concerned with dening topological prop-
erties, proving relations between them, and showing that certain examples
exhibit or fail to exhibit these properties.
4 Closures, limit points and boundaries
4.1 Closed sets
Recall that a subset A of a space X is closed if and only if its complement
XA is open.
There is a convenient characterization of the set of closed subsets of a
topological space.
Proposition 4.1. The closed sets of a topological space (X, T ) satisfy the
following axioms:
1. Arbitrary intersections of closed sets are closed.
2. Finite unions of closed sets are closed.
3. The set X and the empty set are both closed.
Conversely, if ( is a set of subsets of X satisfying these axioms, then ( is
the set of closed sets for some topology on X.
The proof is an exercise using DeMorgans laws on set operations.
Closed sets can be used to detect continuity.
Proposition 4.2. A function between spaces f : X Y is continuous if
and only if for every closed A Y , the preimage f
1
(A) X is closed.
9
Proof. The preimages of complementary sets are complementary: if Y =
A U, then X = f
1
(A) f
1
(U). So f pulls back closed sets to closed
sets if and only if f pulls back open sets to open sets.
Since innite intersections of closed sets are closed, it is useful to make
the following denition.
Denition 4.3. Let X be a topological space and S a subset of X. The
closure of S, denoted

S, is the intersection of all closed sets containing S.
Example 4.4. In the standard topology on R, the closure of a bounded
interval is the closed interval with the same endpoints.
Some near-trivial observations about closed sets:
Proposition 4.5. Let X be a space and S any subset of X.
The closure

S is closed.
S is closed if and only if S =

S.
The closure of the closure of S (write it S

) is equal to

S.
Denition 4.6. A subset S of a space X is dense if its closure is the whole
space:

S = X.
Example 4.7. In the standard topology on R, the rationals and the irra-
tionals are both dense.
Existence of proper, dense subsets of dierent sizes (nite or innite) is
an example of a topological property.
4.2 Sequences and limit points
From calculus and analysis, we are used to probing the topological proper-
ties of R using sequences. For completeness, we recall the denition of a
sequence.
Denition 4.8. A sequence of points in a space X is simply a function
from the natural numbers N to X. By convention, the function inputs are
denoted as subscripts, so that we write a sequence as p
i

iN
for p
i
X.
Naturally we must redene convergence in a purely topological context.
10
Denition 4.9. A sequence p
i

iN
in a space X converges to a point p X
if for each open neighborhood U of p, there is a number N N (depending
on U) such that for all i > N, we have p
i
U.
A linguistic convenience: we say a set U contains a tail of the sequence
p
i

iN
if there is some N N with all p
i
in U for i > N. So a sequence
converges to a point if every open neighborhood of the point contains a tail
of the sequence.
These denitions make the following important fact nearly a trivialty.
Proposition 4.10. Suppose f : X Y is continuous and p
i

iN
is a
sequence in X converging to p X. Then f(p
i
) converges to f(p).
Proof. Let U be an open neighborhood of f(p). Then f
1
(U) is an open
neighborhood of p. Then f
1
(U) contains a tail of the sequence p
i

iN
.
Then U contains a tail of the sequence f(p
i
)
iN
.
The topological space structure lets us tell when a point is near a set;
such a point is called a limit point.
Denition 4.11. Let X be a topological space and let S be a subset of
X. A point p in X is a limit point of S if every open neighborhood U of p
contains a point of S other than p:
U Sp ,= .
Example 4.12. In R with the usual topology, the limit points of an interval
are the endpoints of the interval. Note that it does not matter whether these
endpoints are contained in the interval or not.
Proposition 4.13. For any set S in any topological space X, the closure
of S is equal to the union of S with all the limit points of S.
Proof. For a point p X, p is not in

S if and only if there is a closed set A
with S A and p / A. There is such an A if and only if there is an open
neighborhood U of p with U S = , since we can take U = XA and vice
versa. There is such a U if and only if p is not in S and is not a limit point
of S.
Sequences cannot escape the closure of a set:
Proposition 4.14. Suppose S is a subset of a space X and p
i

iN
is a
sequence of points in S converging to a point q X. Then q is in

S.
11
Proof. If p
i

iN
is eventually constant, then q is equal to p
i
for large enough
i and q S. If p
i

iN
is not eventually constant, innitely many of the p
i
are not equal to q. Let U be an open neighborhood of q. Then U contains
a tail of p
i

iN
. Then U contains some point in S (in fact, innitely many
points in S), and q is a limit point of S.
Later we will see that the converse to this theorem is not true in all
spaces: there are examples of a limit point q of a set S in a space X, but
where no sequence of elements in S converges to q.
Corollary 4.15. If A is a closed subset of a space X and a sequence of
points in A converges to a point q X, then q A.
This is immediate since

A = A.
4.3 Topologically derived sets
In addition to the closure and the set of limit points (sometimes called the
derived set), there are several ways of using the topological structure to
derive new subsets of a space from old ones.
Denition 4.16. Let X be a topological space and S a subset of X.
The interior of S, denoted S

, is the union of all open subsets of S.


The exterior of S is the union of a open subsets that are disjoint from
S.
The boundary of S, denoted S, is the set of points whose every neigh-
borhood contains elements of S and elements not in S:
S =
p X[ for all open U X with p U, U S ,= and US ,= .
Proposition 4.17. For any subset S of any space X, the exterior, interior,
and boundary of S are disjoint, and their union is X.
Proof. If p S

, then some open neighborhood of p is a subset of S. So


clearly p is not in S or in the exterior of S. Similarly, if p is in the exterior
of S, then some open neighborhood of p is disjoint from S, and p cannot be
in S.
Now suppose p X and p is not in S. Then some open neighborhood
of p is either disjoint from S or a subset of S. In the rst case, p is in the
exterior of S, and in the second case, it is in S

.
12
Proposition 4.18. For any subset S of any space X, we have

S = S S

and S =

SS

.
Proof. A point p is in the exterior of S if and only if there is an open set U
with p U and U S = . This is true if and only if there is a closed set
A = XU with p / A and S A. This is true if and only if p is not in

S.
So therefore

S = S S

, the complement of the exterior of S.


Then

SS

is S.
Proposition 4.19. The following are true:
A subset S of X is open if and only if S

= S.
A subset S of X is closed if and only if S S.
A subset S of X is open if and only if S S = .
The proofs are straightforward.
There is a rich interplay between these set operations. In fact, just using
the operations of set complementation and set closure, one can derive as
many as 14 dierent sets from a subset of a space. We will explore this in
the second homework assignment.
4.4 Comparing topologies on the same point set
Denition 4.20. Fix a set X and let o, T be two dierent topologies on
X. Then (X, o) is coarser than (X, T ) if o T . In the same situation,
(X, T ) is ner than (X, o).
To restate: for (X, T ) to be ner than (X, o), we need that every open
set in o is also open in T , but possibly some other sets are open as well.
The relation coarser is reexive, transitive relation on the set of all
topologies on a point set. The discrete topology is the nest topology on
a set, and the indiscrete topology is the coarsest topology on a set. The
discrete and indiscrete topologies can both be compared to any topology on
the set. As the following example shows, not all topologies are comparable.
Example 4.21. Let X be a set and let p X. The particular point topology
with respect to p is the topology on X where a set is open if and only if
it contains p. Let p ,= q X and let T (p) and T (q) be the respective
particular point topologies. Then T (p) , T (q) and T (q) , T (p); neither
topology is coarser than the other.
13
Many topologies on a space can be characterized as the coarsest topology
with a particular property.
Example 4.22. Let X be a space and let S X. Then the subspace
topology is the coarsest topology for which the inclusion S X is
continuous.
The topology of nite complements is the coarsest topology in which
points are closed.
The topology generated by a given subbasis B is the coarsest topology
in which every set in B is open.
5 Topologies from other structure
5.1 Order topologies
The standard topology on R is a special case of the following general con-
struction:
Denition 5.1. A strict total ordering on a set X is a relation < satisfying:
for all x, y X with x ,= y, we have x < y or y < x,
for all x, y, z X with x < y and y < z, we have x < z, and
for all x X, we have x ,< x.
A set with a strict total ordering is an ordered set.
Just like in the real line, we have intervals in any ordered set. Dene
(x, y) = z X[x < z < y, and dene closed and half-closed intervals to be
the union of the open interval with the indicated endpoints. We also have
the intervals (, x) and (x, ) (these are sometimes called open rays).
If X has a strict total ordering <, then the order topology on X is the
topology with a basis given by open intervals (x, y) for x, y X, along with
half-innite open intervals (, x) and (x, ) for x X.
Since any nite intersection of intervals is either empty or an interval,
this set is actually a basis for a topology and not just a subbasis.
14
5.2 Countable and uncountable sets
Recall that a set is countable if it is nite or can be put in bijection with N.
A set is uncountable if it is innite and cannot be put in bijection with N.
Intuitively, countability of sets relates to the amount of information
needed to specify a member of a set. If there is a universal bound on the
number of bits needed to specify a member of the set S, then S is nite.
If there is no universal bound on the number of bits needed to specify a
member of the set S, but every member of S can be specied with a nite
number of bits, then S is countable. If most members of the set S require
an innite number of bits to specify, then S is uncountable. (The preceding
remarks are not so much rigorous results as they are intuitive guidelines.)
We put together some results from set theory and analysis:
Proposition 5.2. Subsets and quotients of countable sets are count-
able.
Unions of countable sets, indexed over a countable set, are countable.
The Cartesian product of a family of nitely many countable sets is
countable.
Power sets of innite sets are uncountable.
The rational numbers Q are countable, but the real numbers R are
uncountable.
5.3 A pathological example
Denition 5.3. A well ordering on a set X is a strict total order such
that every nonempty subset of X contains a minimal element. An initial
segment of a well ordered set is a subset that contains all the predecessors
of any element it contains.
The following theorem in set theory depends on the axiom of choice. We
quote it without proof. Halmoss Naive Set Theory is an excellent book and
contains this theorem.
Theorem 5.4 (Well-ordering theorem). Every set is the domain of a well
ordering.
From this can deduce the following:
Proposition 5.5. There is an uncountable, well ordered set (X, <), such
that every proper initial segment of X is countable.
15
Proof. Suppose (Y, <) is uncountable and well ordered and let 0 denote
the minimal element of Y . Let S be the set of y Y such that [0, y]
is uncountable. Since Y is well ordered, S has a least element . Let
X = [0, ] with the induced ordering. Then X is the desired set.
Example 5.6. Let X = [0, ] be the uncountable, well ordered set, with
every proper initial segment countable, from the previous proposition. The
set S = [0, ) has closure X, so that is a limit point of S. Suppose p
i

iN
is a sequence in S. Let T be the union of the intervals [0, p
i
] for i N. Then
T is a countable set. Let q be the next element in X after T (that is, the
least element of the set of elements of X greater than every element of T).
Then [0, q] = T q is countable, so q ,= . In particular, (q, ) is an open
neighborhood of that does not contain any of the p
i
. So no sequence of
points in S converges , even though

S.
Intuitively, the reason that no sequence in S converges to is that
sequences are too short.
5.4 Metric Spaces
Denition 5.7. A metric space is a set X with a distance function d: X
X R satisfying the following axioms:
d(x, y) = d(y, x) for any x, y X (symmetry),
d(x, z) d(x, y) +d(y, z) for any x, y, z X (triangle inequality), and
d(x, y) 0 for any x, y X, with d(x, y) = 0 if and only if x = y.
Example 5.8. For any n 1, the n-dimensional Cartesian power R
n
of the
real line is a metric space with the Euclidean (Pythagorean) metric:
d((x
1
, . . . , x
n
), (y
1
, . . . , y
n
)) =

i=1
(x
i
y
i
)
2
.
Denition 5.9. In a metric space (X, d), the open ball of radius r R
centered at x X, is the set
B
r
(x) = y X[d(x, y) < r.
Example 5.10. In R
n
, these are the interiors of round spheres of the given
radius, centered at the given point.
16
Proposition 5.11. The set of all open metric balls in (X, d) form a basis
for a topology on X.
Proof. Let x, y X and let r, s > 0. If B
r
(x)B
s
(y) = , there is nothing to
show. So consider any z B
r
(x) B
s
(y). Of course, this means d(x, z) < r
and d(y, z) < s. Select t > 0 such that:
t < minr d(x, z), s d(y, z).
Suppose that w B
t
(z). Then
d(w, x) d(w, z) + d(z, x) < t + d(x, z) < r d(x, z) + d(x, z) = r,
by the triangle inequality and the choice of t. So w is in B
r
(x). Similarly, any
w B
t
(z) is also in B
s
(y), and therefore B
t
(z) B
r
(x) B
s
(y). So for any
z B
r
(x) B
s
(y), there is a t(z) > 0 such that B
t(z)
(z) B
r
(x) B
s
(y).
In particular, B
r
(x) B
s
(y) is a neighborhood of each z it contains. So
B
r
(x) B
s
(y) is open.
Denition 5.12. For any metric space (X, d), the topology with basis given
by the set of all open metric balls is the metric topology on (X, d).
The metric topology on R is the same as the order topology. Note that
we have nally dened a reasonable topology on R
n
for n > 1.
Example 5.13. Any set X is a metric space with the distance function
d(x, y) = 1. This metric space topology is the discrete topology.
5.5 The topology on a nite product
The following is the standard way to use the topology on a pair of spaces to
put a topology on the Cartesian product of their point sets.
Denition 5.14. Let (X, T ) and (Y, o) be spaces. The product topology on
X Y is the one with basis
U V [U T , V |.
Inductively we can dene a topology on R
n
by R
n
= R
n1
R, where R
has the standard topology. Luckily, this topology turns out to be the same
as the metric topology. This will probably be a homework exercise.
Proposition 5.15. Some observations:
17
For any y Y , the inclusion X XY by x (x, y) is continuous.
Similarly inclusions from Y to X Y are continuous.
There are canonical coordinate projections p
1
: XY X (by (x, y)
y) and p
2
: X Y Y ; these are continuous.
The product topology on X Y is the coarsest topology in which the
coordinate projections are continuous.
The product topology is characterized by the following property: For
any space Z and any continuous maps f : Z X and g : Z Y , there
is a unique continuous map f g : Z XY such that p
1
(f g) = f
and p
2
(f g) = g.
5.6 Countability properties
As we saw in the example of an uncountable well ordering, there are un-
countable spaces that behave very badly. On the other hand, there are un-
countable spaces, like R with the standard topology, in which we still have
a lot of control. The following denitions help distinguish these situations.
Denition 5.16. Let X be a space.
X is separable if X contains a countable dense subset.
X is second countable if X has a countable basis.
X is rst countable if for each p X, there is a countable neigh-
borhood basis: a set A
p
of open neighborhoods of p, such that every
neighborhood of p contains some member of A
p
.
Proposition 5.17. Any second countable space is both rst countable and
separable.
Proof. Let B be a countable basis for X. For each p X, the set U
B[p U is a countable neighborhood basis. Let S consist of one point from
each member of B. Then S is a countable dense subset of X.
Proposition 5.18. Any subspace of a rst countable space is rst countable,
and any subspace of a second countable space is second countable.
Proof idea: We nd our new basis or neighborhood bases by intersecting the
old ones with the subset.
Proposition 5.19. Finite products of separable spaces are separable.
18
Proof idea: The product of the countable dense subsets is a countable dense
subset.
Slightly less trivial:
Proposition 5.20. Any separable metric space is second countable.
Proof. Let X be a metric space with metric d and countable dense subset S.
We will show that the set B of open balls with rational radii around points in
S form a countable basis for X. Consider the basic open set B
r
(x), for some
x X and r > 0. Let y B
r
(x). There is an s > 0 such that B
s
(y) B
r
(x).
Let z S B
s
3
(y) and let t (
s
3
,
2s
3
) Q. Then y B
t
(z), and it follows
from the triangle inequality that B
t
(z) B
s
(y) B
r
(x). So we have shown
that every point in B
r
(x) is inside a set in B that is contained in B
r
(x).
Therefore B
r
(x) is a union of members of B, and therefore B is a basis.
Since B is in one-to-one correspondance with the product S ((0, ) Q)
of countable sets, B is countable.
Proposition 5.21. If X is rst countable, then if x X is a limit point of
a set S X, then there is a sequence in S whose limit is x.
Proof. Let U
i

iN
be a countable neighborhood basis for x. We can assume
that U
n+1
U
n
: if not, we simply replace U
n
with the nite intersection

n
i=0
U
i
. Dene a sequence by letting p
n
be any point in U
n
Sx (this
is nonempty since x is a limit point of S). Since the U
i
are nested, this
sequence converges to x.
Example 5.22. Countability properties of spaces we have already seen:
R
n
with the metric topology is second countable.
The order topology on an uncountable well ordering is not separable
or rst countable.
A discrete topology is second countable if and only if it is separable if
and only if the point-set is countable. Any discrete topology is trivially
rst countable. Note that since discrete spaces are metric spaces, there
are metric spaces that are not separable.
Any indiscrete topology is second countable.
An uncountable particular point topology X with particular point p
has an uncountable (closed) discrete subspace Xp. Therefore X is
not second countable. However, X is separable (with countable dense
19
subset p) even though Xp is not. This X is rst countable: p
is a neighborhood basis for p and p, q is a neighborhood basis for
q ,= p.
Example 5.23. Less trivially: Let

be the set of bounded sequences in


R, with metric
d(f
i
, g
i
) = sup
i
[f
i
g
i
[.
Give

the metric topology. Then

not separable. Proof: Let S be a


countable set and let (f
j
)
i
be an enumeration of S, where (f
j
)
1
, . . . , (f
j
)
n
, . . .
is one of the sequences in S. Dene a sequence g
i
by:
g
i
=

0 if [f
i
[ > 1
2 if [f
i
[ 1
Then for each j, we have d(g
i
, (f
j
)
i
) 1, so that the open ball of radius
1/2 around g
i
does not contain any of element of S. The

S ,=

.
Zariski topologies
Denition 5.24. Let F be a eld and let n N, n 1. Let F[x
1
, . . . , x
n
]
be the ring of polynomials over F in n variables. Let X = F
n
. The Zariski
topology on X is the topology where the closed sets are exactly the nullsets
of sets of polynomials : A X is closed i there is S F[x
1
, . . . , x
n
] with
A = (a
1
, . . . , a
n
) X[f(a
1
, . . . , a
n
) = 0 for all f S.
Further, if S is a subset of X, then the subspace topology on S from the
Zariski topology is usually also referred to as the Zariski topology. Example:
The set of invertible n n matrices over F is a subset of F
n
2
and inherits
a Zariski topology from it.
Proposition 5.25. The Zariski topology is a topology.
Proof. Suppose A
i
is the closed set corresponding to the set of polynomials
S
i
, for i in some index set I. Then A
1
A
2
is the nullset of the set of
all pairwise products of polynomials in S
1
with polynomials in S
2
. The
intersection

iI
A
i
is the nullset of

iI
S
i
.
Proposition 5.26. Some observations about the Zariski topology:
If F is nite then the Zariski topology is discrete.
If n = 1 then the Zariski topology is the topology of nite complements.
20
Proposition 5.27. If F is R then every nonempty open set in the Zariski
topology is dense.
Proof. Suppose U is a nonempty open set. Let A = XU. If

U ,= X, then
A

,= . Then there is nonempty open V A. Let B = XV . Then A


and B are closed proper subsets of X and A B = X. Then A and B
correspond to two sets of polynomials S and T in R[x
1
, . . . , x
n
]. Since A
and B are proper, there are f S and g T that are nonzero. Then the
product fg is a nonzero polynomial. Taking limits of x
1
, . . . , x
n
as they go
to innity, the value of fg(x
1
, . . . , x
n
) will go to . So there is a point
where fg ,= 0, so A B ,= X.
Example 5.28. There are subsets of R
n
that are closed in the standard
topology but are dense in the Zariski topology. For example, Z
n
is such a
set. However, Z is not Zariski dense in R
2
. (What is its closure?)
Often when studying structures dened over R or C (such as groups of
matrices), it is convenient to work simultaneously with the standard topol-
ogy and with the Zariski topology. Therefore one often encounters the term
Zariski dense in mathematical literature; of course, it simply means dense
in the Zariski topology.
Next time, well talk about separation properties. I briey mention two
of them:
Denition.
A space is T
1
if points are closed.
A space X is Hausdor if for x, y X, x ,= y, there are disjoint open sets
U, V with x U and y V .
The Zariski topology on R
n
is an important example of a topological
space that is T
1
but not Hausdor.
Separation properties
One measure of the quality of a space is how well the open sets separate
subsets of the space from each other. A series of properties called separation
properties serve this purpose.
Denition 5.29. Two subsets S, T of a space X are separated if

S T =
S

T = .
Denition 5.30. Let X be a space.
21
X is T
0
if for every pair x, y X, x ,= y, there is an open set U in X
with exactly one of x, y in U.
X is T
1
if for every pair x, y X, x ,= y, there are open sets U, V with
x U, y / U, y V , y / V .
X is T
2
if for every pair x, y X, x ,= y, there are disjoint open sets
U, V with x U, y V .
X is T
3
if for every closed set A and every x XA, there are disjoint
open sets U, V with A U and x V .
X is T
4
if for every pair of disjoint closed sets A and B, there are
disjoint open sets U, V with A U and B V .
X is T
5
if for every pair of separated sets S and T, there are disjoint
open sets U, V with S U and T V .
X is regular if X is T
0
and T
3
.
X is normal if X is T
1
and T
4
.
X is completely normal if X is T
1
and T
5
.
Hausdor is a synonym for T
2
.
This is a lot of denitions... I will expect students to know the denitions
of Hausdor, T
1
, normal and completely normal, but not the others.
The next proposition collects some near-trivial statements.
Proposition 5.31. X is T
1
if and only if points are closed.
Hausdor implies T
1
and T
1
implies T
0
.
T
5
implies T
4
, and completely normal implies normal.
Normal implies T
3
and Hausdor.
So completely normal implies any of the other separation properties.
Proposition 5.32. Metric spaces are completely normal.
Proof. Let S and T be separated subsets of a metric space (X, d). Then no
x S is a limit point of T, so for each x S we can nd a real number
r(x) such that B
r(x)
(x) is disjoint from T. Of course we can also nd a real
number s(y) for each y T such that B
s(y)
(y) is disjoint from S. Then
by the triangle inequality,

xA
B
r(x)/2
(x) and

yB
B
r(y)/2
(y) are disjoint
22
open sets containing S and T respectively. This proves that metric spaces
are T
5
.
If x and y are points in X, then the open balls of radius d(x, y)/2 around
each are disjoint open neighborhoods, showing X is Hausdor and therefore
T
1
.
Example 5.33. Since they are metric spaces, R
n
with the standard
topology, and any discrete space, will satisfy all separation properties.
Indiscrete spaces are not T
0
, T
1
, or Hausdor, but are vacuously T
3
,
T
4
, and T
5
.
The Zariski topology on R
n
is T
1
but not Hausdor.
Particular point topologies are T
0
but not T
1
.
Let Y be the indiscrete topology on two points. For any space X, the
product X Y will not be T
0
, T
1
or Hausdor, but will be T
i
if and
only if X is for i = 3, 4, 5.
Recall that a topological property is hereditary if it is preserved when
taking subspaces.
Proposition 5.34. The following properties are hereditary: being T
0
, being
T
1
, being Hausdor, being T
5
(and therefore being completely normal). A
closed subspace of a normal space is normal.
No proof will be given. It is tricky to give an example of a normal space
with a non-normal subspace, but we may do it later in the quarter.
The following is perhaps the rst nontrivial theorem in point-set topol-
ogy.
Theorem 5.35 (Urysohns lemma). A space X is T
4
if and only if for
each pair of disjoint closed sets A and B, there is a continuous function
f : X [0, 1] R (with the standard topology, such that f(x) = 0 for all
x A, and f(y) = 1 for all y B.
A dyadic rational number is a rational number whose denominator (in
reduced form) is a power of two.
Proof. The if direction is easy and is omitted.
Let S be the set of dyadic rationals in (0, 1). We will start by inductively
dening two families of open sets U(r) and V (r) for all r S, satisfying
23
U(r) V (r) = for all r S,
B U(r)

U(r) XA for all r S, and
for r < r

S, we have (XU(r)) (XV (r

)) = .
Since X is T
4
, we can select U(1/2), V (1/2) disjoint open sets with A
V (1/2) and B U(1/2).
Now let n > 0 and assume we have dened all U(r), V (r) for r S with
denominator 2
n
or less. Let k N with 0 k < 2
n
, and set s = k/2
n
,
t = (k + 1)/2
n
, and r = (s + t)/2 = (2k + 1)/2
n+1
. If s = 0, we let U(r)
and V (r) be disjoint open sets such that A V (r) and XV (t) U(r). If
t = 1, we let U(r) and V (r) be disjoint open sets such that XU(s) V (r)
and B U(r). Otherwise, we let U(r) and V (r) be disjoint open sets such
that XU(s) V (r) and XV (t) U(r). This is possible since X is T
4
,
and having made these denitions all sets dened will satisfy the conditions.
Having dened these sets, we now dene our function f : X [0, 1].
Note that the conditions imply that for r r

, we have U(r

) U(r). Set
f(x) = sup0r S[x U(r). Let (a, b) be an open subinterval of (0, 1)
and let x f
1
((a, b)). There are a

, b

S with f(x) (a

, b

) (a, b) and
a < a

(since S is dense in [0, 1]). Then f(U(a

U(b

)) (a, b), U(a

U(b

)
is open and contains x. (If y U(a

) then f(y) a

; if y /

U(b

) then
y / U(b

) and y b

). In particular, every point in f


1
((a, b)) is contained
in an open subset of f
1
((a, b)), so f
1
((a, b)) is open. Similarly, if x
f
1
((a, 1]), then there is a

S with f(x) (a

, 1] (a, 1] and a < a

,
and x U(a

) f
1
((a, 1]). And of course, if x f
1
([0, b)), then there
is b

S with f(x) [0, b

) [0, b), and x X

U(b

) f
1
([0, b)). This
proves f is continuous. Since B U(r) for all r, we have f(B) 1; since
A X

U(r) for all r, we have f(A) 0.


From Urysohns lemma, we can deduce that for normal spaces, continu-
ous functions to [0, 1] separate points.
Proposition 5.36. In a normal space X, for any two points x, y X,
x ,= y, we can nd continuous f : X [0, 1] with f(x) ,= f(y).
Urysohns lemma makes it possible to nd a function on a space that
roughly approximates a function on a subset.
Lemma 5.37. Let X be normal and let A X be closed. Suppose f : A
[1, 1] is continuous. Then there is a continuous function g : X [1/3, 1/3]
such that
sup
xA
[f(x) g(x)[
2
3
.
24
Proof. Let B = f
1
([1/3, 1]), and let C = f
1
([1, 1/3]). By Urysohns
lemma, there is a function on X that is 1 on B and 0 on C; by rescaling
this function we get the desired function g.
This lemma is key to the following theorem:
Theorem 5.38 (Tietze extension theorem). A space X is T
4
if and only if
for every closed subset A and every continuous function f : A R, there is
a function g : X R extending f, meaning g[
A
= f.
Proof sketch. A rigorous proof of this theorem requires quoting some basic
results from analysis, so we give a sketch instead. The if direction is
easy and is omitted. For the only if direction, we compose f with a
homeomorphism R (1, 1) to get a bounded function. Assuming f is
bounded, we approximate f by a function f
1
dened on X, using the lemma.
The dierence is presumably nonzero, so we approximate the dierence by
another function f
2
and then f
1
+ f
2
is a function on X that restricts to
a better approximation of f. Repeating this process innitely many times,
the error of the approximations goes to zero exponentially fast, and the
approximations converge to the desired function g (this uses the fact from
analysis that uniform limits of continuous functions are continuous).
More on metric spaces
Some last comments about separation and countability properties as they
pertain to metric spaces. As we mentioned before, metric spaces are sepa-
rable if and only if they are second countable. However, taking the balls of
rational radius around each point gives us the following:
Proposition 5.39. Every metric space is rst countable.
This means that we can characterize limit points and closures (and there-
fore closed sets) in metric spaces in terms of sequences. Further, since metric
spaces are Hausdor, we can use sequences to uniquely specify points:
Proposition 5.40. In a Hausdor space, any sequence converges to at most
one point.
The proof is an exercise using the denition. In particular, it follows from
these remarks that any point in a separable metric space can be specied
uniquely by a sequence of points in a xed countable dense subset (like Q
in R).
25
6 Connectedness properties
The disjoint union
Denition 6.1. The disjoint union of spaces (X, T ), (Y, o) is denoted X .
Y ; the point set is the disjoint union X.Y of the pointsets, and the topology
has the basis
U V [U T , V o,
where U and V are considered as subsets of the disjoint copies of X and Y
in X . Y .
Proposition 6.2. Some observations:
There are canonical inclusions i
1
: X X . Y and i
2
: Y X . Y ;
these are continuous.
The topology on X . Y is the nest one that makes the canonical
inclusions continuous.
The disjoint union X . Y is characterized by the following property:
For any space Z and any pair of continuous maps f : X Z and
g : Y Z, there is a unique continuous function f . g : X . Y Z
that with (f . g) i
1
= f and (f . g) i
2
= g.
Connectedness properties
Denition 6.3. A separation of a space X is a pair of disjoint open nonempty
subsets U, V with U V = X.
A space is connected if X has no separation. Otherwise it is disconnected.
A subset S of a space X is connected if it is when considered as a subspace.
Proposition 6.4. R with the standard topology is connected.
Proof. If R = U V were a separation, then without loss of generality there
would be some x RU with U (, x) ,= . Then by replacing U with
U (, x) and V with V (U (x, ), we may assume that U is bounded
above. Let y be the least upper bound for U. Certainly y is not in U (if so,
an open interval around y is contained in U and y is not an upper bound
for U). If y V , then an open interval around y is contained in V , and
V U ,= . Then v / U V , a contradiction.
Example 6.5. Any disjoint union of nonempty spaces is disconnected.
If X is a space and S, T X are nonempty separated sets, then the
subspace topology on S T is disconnected.
26
Note that the only properties of R that we used in proving connectedness
were the least upper bound property, and the fact that an open interval
around any point p contains points strictly less than and strictly greater
than p. An ordered set with these properties is called a linear continuum,
and it follows that all linear continua are connected in the order topology.
Proposition 6.6. If S X is connected and T X with S T

S, then
T is connected.
Proof. Suppose T = U V is a separation of T. Any point in T is in S
or a limit point of S. If x U is a limit point of S, then since U is open,
US ,= . Similarly, V S ,= . So S = (US)(V S) is a separation.
Proposition 6.7. The image of a connected space under a continuous func-
tion is connected.
Proof. Suppose f : X Y is a continuous function. If f(X) is disconnected
with a separation f(X) = UV , then X = f
1
(U)f
1
(V ) is a separation.
Example 6.8. Since any open interval is homeomorphic to R, any open
interval is connected. Since any interval is a subset of the closure of an
open interval, any interval is connected. Alternatively, we can see that any
interval is connected because any interval is a continuous image of R.
Since intervals are connected, we can test for connectedness using maps
of intervals.
Denition 6.9. In a space X with x, y X, a path from x to y is a
continuous map f : [0, 1] X with f(0) = x and f(1) = y. A path from x
to y is an arc if it is one-to-one.
A space is path-connected if any two points can be joined by a path. A
space is arc-connected if any two distinct points can be joined by an arc.
No proof is given for the following.
Proposition 6.10. The image of a path-connected space under a continuous
map is path-connected, and the image of an arc-connected space under an
injective, continuous map is arc-connected.
Proposition 6.11. Arc-connected implies path-connected and path-connected
implies connected.
27
Proof. Every arc is a path, so arc-connected implies path-connected.
Suppose X is a space and X = U V is a separation. Let is a path
from a point in U to a point in V and let A = ([0, 1]). Then A is the image
of a connected space and is connected. But A = (A U) (A V ) is a
separation; this is a contradiction.
Example 6.12. The indiscrete topology on a countable set is path-connected
but not arc-connected.
The topologists sine curve is connected but not path-connected. Let S
be the graph of f : (0, 1] R, by f(x) = sin(1/x), a subset of R
2
. The
topologists sine curve is the closure of S, which is S 0 [1, 1]. S is a
continuous image of a path-connected space and is therefore connected, and

S is therefore connected. However, there is no path from a point in S to a


point in 0[1, 1]. Suppose were such a path. Without loss of generality,
([0, 1)) S. Since the projection of to the x-axis is continuous, hits all
y-values in [1, 1] on a suciently small interval (a, 1]. Let s [1, 1] with
(0, s) ,= (1). We can nd a sequence p
i

i
in [0, 1] with (p
i
)
i
converging
to (0, s). Then is not continuous, a contradiction.
Types of components
Denition 6.13. Let X be a space. Dene a relation
c
on X by a
c
b if
there is some connected S X with a, b S. The equivalence classes of X
under
c
are the connected components (or just components) of X.
Dene a relation
q
on X by a
q
b if a and b are on the same side of
every separation: for every separation X = U V with a U, we have b
also in U. The equivalence classes of X under
q
are the quasicomponents
of X.
The existence of paths from a to b denes an equivalence relation; the
path components of X are the equivalence classes under this relation.
Theres a problem with dening arc components, specically, the ex-
istence of an arc between two points is not in general a transitive relation.
Example 6.14. Let Y be the two-point space 0, 1 with the indiscrete topol-
ogy. Let X
0
= Y [0, ), and let X = (1, 0) 0 [0, ). Then X
is [0, ) with the standard topology, but with 0 doubled. There is an arc
from (0, 1) to any point in X, but no arc from (1, 0) to (0, 0).
Then X is path connected but not arc connected, and the notion of arc
component doesnt make sense for X.
28
Proposition 6.15. Let X be a space and x X. Then the path compo-
nent of x is a subset of the connected component of x, and the connected
component of x is a subset of the quasicomponent of x.
Proof. If there is a path from x to y, then the image of this path is a
connected subspace of X containing x and y. This shows the rst inclusion.
If S is a connected subspace of X containing x and y, then any separation
X = U V will have to have S on one side, say S U. Otherwise the
separation will induce a separation of S. So x and y are both in U.
Proposition 6.16. If X has only one quasicomponent, then X is connected.
The proof is easy and is omitted.
Proposition 6.17. The quasicomponent of x is the intersection of all sets
containing x that are both closed and open.
Proof. Suppose y is in the quasicomponent of x. Then for each separation
X = U V with x U, we have y U. If U is closed and open with x U,
then either U = X or X = U (XU) is a separation. In either case, y U,
so y is in the intersection of all such sets.
Suppose y is not in the quasicomponent of x. Then there is some sepa-
ration X = U V with x U and y V . Then U is closed and open and
y / U, so y is not in the intersection of all such sets.
Example 6.18. For n N, n > 0, let R
n
be the rectangle in R
2
with
corners (1 1/n, n), (1 1/n, n), (1 + 1/n, n) and (1 + 1/n, n), let
L
1
= 1R and L
2
= 1R. Let S = L
1
L
2

n
R
n
. Let S = U V
be a separation of S. Suppose L
1
U. Then for large n, R
n
U, and
therefore L
2
U. However, any given R
n
is can be separated from L
1
L
2
.
So L
1
L
2
is a quasicomponent of S. But clearly, L
1
and L
2
are dierent
connected components of S.
Proposition 6.19. Quasicomponents are closed and connected components
are closed.
The proofs are easy and are omitted.
Example 6.20. The path components of the topologists sine curve

S above
are S and 0 [1, 1], but

S is connected. Note that S is not closed; so
generally, path components are not necessarily closed.
29
Local connectedness properties and disconnectedness
Denition 6.21. A space X is locally connected if for every point x X and
every open neighborhood U of x, there is a connected open neighborhood
V of x with V U.
A space X is locally path connected if for every point x X and every
open neighborhood U of x, there is a path-connected open neighborhood V
of x with V U.
The following needs no proof.
Proposition 6.22. A locally path connected space is locally connected.
Example 6.23. The topologists sine curve

S is connected but not locally
path connected: every suciently small neighborhood of (0, 0) has innitely
many path components.
You should be able to nd an example of a locally connected space that
is not connected.
Lemma 6.24. In a space X, suppose S and T are connected sets with
nonempty intersection. Then S T is connected. Similarly, if S and T are
path connected sets with nonempty intersection then ST is path connected.
Proof is easy and is omitted.
Proposition 6.25. If X is locally connected, then each connected compo-
nent of X is open; therefore the quasicomponents of X coincide with the
connected components.
Proof. If S is a connected component of X and S is not open, then there is
a point x S that is a limit point of XS. If X is locally connected, then
X has a connected open neighborhood U. Since S and U are connected and
S U is nonempty, S U is connected, and therefore U S. but then x is
not a limit point of XS. So S is open.
Then S is open and closed, but since S is connected no proper subset of
S is both open and closed. So S is a quasicomponent.
Proposition 6.26. If X is locally path connected, then the quasicomponents
of X coincide with the path components.
Proof. Suppose S is a path component of X. Suppose x is in the closure of
S. Since X is locally path connected, x has a path connected neighborhood
U. Then S U is nonempty, and therefore S U is path connected, and
U S. So

S S and S is closed. Further, such an x cannot be a limit
30
point of XS, and therefore S is open. Since S is connected and open and
closed, S is a quasicomponent of X.
Denition 6.27. A space is totally disconnected if its connected compo-
nents are all points.
Example 6.28. A space is both totally disconnected and locally connected
if and only if it is discrete.
The subspaces Q and RQ of R with the standard topology are totally
disconnected.
Any subspace of a totally disconnected space is totally disconnected.
7 Quotients
If f : X S is a surjective function from a space X to a set S, then we can
use f to dene a topology on S.
Denition 7.1. Let X be a space, S a set, and f : X S a surjective
function. The quotient topology on S (with respect to f : X S) is the
topology where the open sets of S are exactly the sets U S such that the
preimage f
1
(U) is open.
It is routine to verify that this is a topology. Note that if V X is open,
this denition does not imply that f(V ) is open.
Example 7.2. Let X = [0, 1] with the standard topology and let S = [0, 1).
Dene f : X S by f(x) = x for x [0, 1) and f(1) = 0. The quotient
topology on S has a basis of sets of the form (a, b) [0, 1) and [0, b)(a, 1)
[0, 1]. This topology makes S homeomorphic to the unit circle in R
2
in the
standard topology.
Since we are dening a topology by collapsing together points of the
space, to dene the topology as generally as possible we want lose as little
information as possible when forming the quotient. Losing as little informa-
tion as possible corresponds to taking as ne a topology as possible.
Proposition 7.3. The quotient topology on S with respect to f : X S is
the nest topology on S such that f : X S is continuous.
Proof. It is immediate from the denition that f is continuous with respect
to the quotient topology. Now let o be another topology on S that makes f
continuous. Suppose U o. Then f
1
(U) is open in X, and by denition
U is also open in the quotient topology. So o is coarser than the quotient
topology.
31
For every quotient space S with respect to f : X S as above, the
function f is a continuous surjection with respect to the quotient topol-
ogy. However, not every continuous surjection f : X Y of spaces can be
realized this way.
Denition 7.4. Let X and Y be spaces. A quotient map f : X Y is a
continuous surjection such that the topology on Y is equal to the quotient
topology on Y with respect to the set map f : X Y . Equivalently, a
continuous surjection f : X Y is a quotient map if and only if the topology
on Y is the nest topology making f continuous.
Example 7.5. Let f : X Y be a bijection and let X have the discrete
topology and let Y have the indiscrete topology. Then f is continuous but
not a quotient map.
There are a couple of easy criteria to recognize quotient maps.
Denition 7.6. Let X and Y be spaces. A function f : X Y is open if
for every open U X, we have f(U) open in Y . Similarly, f is closed if for
every closed A X, we have f(A) closed in Y .
Note that a continuous bijection is a homeomorphism if and only if it is
closed if and only if it is open.
Proposition 7.7. Every open continuous surjection is a quotient map, and
every closed continuous surjection is a quotient map.
Proof. Suppose f : X Y is an open continuous surjection. Suppose U
Y . Because f is continuous, f
1
(U) is open if U is. Because f is open,
U = f(f
1
(U)) is open if f
1
(U) is. Then the topology on Y is the quotient
topology, and f is a quotient map.
If f is closed and U Y , then U = Y f(Xf
1
(U)) is open if U is. So
for the same reasons as above, a closed continuous surjection is a quotient
map.
The real utility of quotient spaces is that we can dene surjective set
maps that identify any sets of points that we wish. The quotient space is
then the closest space to the original space in which these sets of points are
identied. Identifying sets of points corresponds to the intuitive notions of
gluing and collapsing.
Denition 7.8. Let X be a space and let be an equivalence relation on
S. The quotient of X with respect to is the quotient (in the sense of
32
functions) with respect to the surjective function X X/, where X/ is
the set of -equivalence classes. Often the quotient space is also denoted
X/.
Example 7.9. Consider the interval [0, 1] in the subspace topology. We get
a space homeomorphic to the circle (from the previous example) by taking
the quotient with respect to the equivalence relation with x y if and only
if x = y or x, y = 0, 1.
Perhaps less trivially, the quotient of R by the equivalence relation x
x + k for all k Z is also homeomorphic to the circle.
Example 7.10. Let X = [0, 1] [0, 1] in the standard topology. Dene
an equivalence relation by (a, b) (c, d) if and only if (a, b) = (c, d) or (
a = c and b, d = 0, 1 ) or (b = d and a, c = 0, 1) or (a, b), (c, d)
(0, 0), (0, 1), (1, 0), (1, 1). Then X is a torus; X is homeomorphic to the
subspace of R
3
in the standard topology given by surface of revolution of a
circle not intersecting the axis of revolution (the surface of a donut).
Example 7.11. In general, the quotients of X with respect to the dening
relations for quasicomponents, connected components, and path components
can be interesting spaces. If X is locally connected, then the connected
component and quasicomponent quotients are discrete. If X is locally path
connected, all these quotients are discrete.
Example 7.12. The quotient of the topologists sine curve

S by path
components is homeomorphic to the Sierpinski space 0, 1 with open sets
, 0, 0, 1.
Example 7.13. The quotient of the space X of nested rectangles by quasi-
components is homeomorphic to the subspace 0 1/(n + 1)[n N R
in the standard topology. The quasicomponent of the two vertical lines
maps to 0 in any such homeomorphism. The quotient of X by connected
components is the same space but with the point 0 doubled.
The properties preserved under taking quotients (in general) are not
much better than the properties preserved under the images of general con-
tinuous surjections. Quotients of connected spaces are connected. Separa-
tion properties are not necessarily preserved.
Example 7.14. Let be any irrational number. Dene an equivalence
relation on R by declaring x x + k + m for all x R and k, m Z. Let
X be R/ where R has the standard topology, and let p: R X denote
the quotient map. Note that since each equivalence class is countable, X
33
is uncountable (otherwise would decompose R into a countable union of
countable sets). Suppose x, y R with p(x) ,= p(y). Let U X be an open
neighborhood of p(x). Then f
1
(U) is an open set. For some a, b R, we
have x (a, b) f
1
(U). Since is irrational, the set k + m[k, m Z
is dense in R and we can nd k, m Z with y +k +m (a, b). Of course,
this means y f
1
(U), and therefore that p(y) U. We have shown that
any nontrivial open subset of X is all of X. Therefore we have exhibited
the uncountable indiscrete topology as a quotient of the standard topology
on R. In particular, without further hypotheses on the equivalence relation
or the map, a quotient of a completely normal space need not even be T
0
.
8 Compactness and related properties
Open covers and the Lindel of property
Denition 8.1. Let X be a space and let S be a subset of X. An open
cover of S is a collection U
i

iI
(for some index set I) of open subsets of
X, such that S

iI
U
i
. A subcover of U
i

iI
is U
j

jJ
for some subset
J of the index set I, such that we still have S

jJ
U
j
. An open cover
V
j

jJ
is a renement of the open cover U
i

iI
if for every j J, there
is some i I with V
j
U
i
.
By the cardinality of a cover, we mean the cardinality of its index set.
One of the properties of spaces related to covers is also related the count-
ability properties we discussed earlier.
Denition 8.2. A space X is Lindelof if every open cover of X has a
countable subcover.
Proposition 8.3. Every second countable space is Lindelof.
Proof. Suppose U
i

iI
is an open cover of the second countable space X.
Let V
n

nN
be a countable basis for X. Let S N be the subset of such
that n S if and only if there is some i I with V
n
U
i
. Each U
i
can be
expressed as a union of elements of V
n

nS
, so V
n

nS
is an open cover
of X and a countable renement of U
i

iI
. For each n S, select some
i(n) I such that V
n
U
i(n)
. Then U
i(n)

nS
is a countable subcover.
Proposition 8.4. A metric space is Lindelof if and only if it is separable.
Proof. Since we know separable metric spaces are second countable and
therefore Lindelof, we only need to show that a Lindelof metric space X is
34
separable. For each n N0, the set of all open balls of radius 1/n is an
open cover of X; let x(n, m) X be points such that B
1/n
(x(n, m))
mN
is
a countable subcover. Then the set x(n, m)[m, n N, n ,= 0 is a countable
dense subset: if B
r
(x) is an open ball, let n be large enough that 1/n < r/2;
if no x(n, m) is in B
r
(x), then x is not in any B
1/n
(x(n, m)) and the set of
such balls could not be a cover.
Proposition 8.5. The property of being Lindelof is preserved under taking
continuous images and closed subspaces.
The proofs are straightforward.
Compactness
Compactness is an important property of certain spaces that is dicult to
describe intuitively. It is a sort of intrinsic boundedness of spacesa way
of characterizing a space as bounded without any reference to distance.
Denition 8.6. A subset S of a space X is compact if every open cover of
S has a nite subcover. The space X is compact if every open cover of X
has a nite subcover.
Proposition 8.7. A subset S of a space X is compact if and only if S is a
compact space in the subspace topology.
The proof is straightforward. Note that any nite space is compact, and
a discrete space is compact if and only if it is nite. We need a nontrivial
example of a compact space.
Proposition 8.8. Any closed interval is a compact subset of R in the stan-
dard topology.
Proof. We prove the proposition for [0, 1] and the general statement follows
easily. Suppose ( is an open cover of [0, 1] in R. Let S [0, 1] be the set
with x S if and only if there is a nite subcollection of ( covering [0, x].
Of course, 0 S, so S is nonempty. Let y be the least upper bound of S.
If y / S, there there is U ( with y U. There is y

U with y

< y and
(y

, y) U, and there is a nite subcollection (

of ( covering [0, y

]. But
then (

U is a nite cover of [0, y], a contradiction. If y S and y < 1,


then there is a nite subset (

of ( and a U (

with y U. Since U is an
open set, there is y

U with y

> y, contradicting the supposition that y


is the least upper bound of S. So we must have y = 1, meaning that ( has
a nite subcover covering [0, 1].
35
The previous proposition and the following observation give many other
examples of compact spaces.
Proposition 8.9. Closed subspaces of compact spaces are compact.
The proof is straightforward.
Compact spaces are much better behaved when we add the hypothesis
that they be Hausdor.
Proposition 8.10. Compact subspaces of Hausdor spaces are closed.
Proof. Let S be a compact subset of a Hausdor space X. Let x XS.
For each y S, there are open disjoint open sets U
y
and V
y
with y U
y
and x V
y
. Since U
y

yS
is an open cover of S, we have a nite subcover
U
y
1
, . . . , U
y
n
. Then

n
i=1
V
y
i
is an open neighborhood of x that is disjoint
from S. This shows XS is open and S is closed.
At this point we can easily prove the Heine-Borel theorem.
Theorem 8.11. A subset S of R in the standard topology is compact if and
only if it is closed and bounded.
Proof. Suppose S is closed and bounded. Then S is a subset of a closed
interval [a, b]. Since [a, b] is compact and S is a closed subset of [a, b], S is
compact.
Now suppose S is compact. Then since R is Hausdor, S is closed. The
open cover (k, k +2)
kZ
of S has a nite subcover, from which we deduce
that S is bounded.
Another benet of assuming spaces are Hausdor:
Proposition 8.12. Every compact Hausdor space is normal.
Proof. Suppose A and B are closed subsets of a compact Hausdor space X.
Then A and B are compact. Fix a A. For b B, nd disjoint open U
b
and
V
b
with a U
b
and v V
b
. Of course, a nite subcollection V
b
1
, . . . , V
b
n

covers B, and taking the union of these V


b
i
and the intersection of these U
b
i
gives us disjoint open sets U
a
and V
a
with a U
a
and B U
b
. Then a nite
subcollection U
a
1
, . . . , U
a
n
of the U
a
covers A; and the union of these U
a
i
and the intersection of these V
a
i
give disjoint open sets that respectively
contain A and B.
The following proposition has a straightforward proof that is omitted.
36
Proposition 8.13. Continuous images of compact spaces are compact.
This then implies one of the most useful facts about compact spaces:
Proposition 8.14. Let X be a compact space and f : X R a continuous
function. Then f is bounded and there are points x, y X realizing the
extreme values of f: for all z X, we have f(x) f(z) f(y).
Proof. We know f(X) is a compact subset of R and is therefore closed and
bounded. Let a be the least upper bound of f(X). Of course a is in the
closure

f(X) (either a f(x) or else a is certainly a limit point of f(x)), so
since f(X) is closed, a f(X). Similarly f(X) contains its greatest lower
bound, and therefore points in x map to both of these extremal values. Of
course, this means that there are points in X mapping to these values.
Another special property of maps from compact spaces:
Proposition 8.15. Let X be compact and let Y be Hausdor. Then any
continuous map f : X Y is closed. Further, f : X f(X) is a quotient
map. Finally, if f : X f(X) is injective, it is a homeomorphism.
Proof. Let A X be closed. Then A is compact. Then the image f(A)
is compact. Since Y is Hausdor, f(A) is closed. The map f : X f(X)
is a closed surjection, so it is a quotient map. Bijective quotient maps are
homeomorphisms, so the last statement also follows.
Denition 8.16. A surjective, closed, continuous map of spaces f : X Y
is perfect if for each y Y , the set f
1
(y) is compact in X.
Lemma 8.17. Suppose f : X Y is a perfect map. Let y Y and let U
be an open subset of X containing f
1
(y). Then there is an open subset
V of Y with f
1
(y) f
1
(V ) U.
Proof. The complement XU is a closed set that does not intersect f
1
(y).
Then since f is closed, f(XU) is closed and does not contain y. Let
V = Y f(XU). Then y V , so f
1
(y) f
1
(V ). Of course f
1
(V ) is
open, and it is not hard to see that f
1
(V ) U.
Proposition 8.18. Suppose f : X Y is a perfect map and Y is compact.
Then X is compact.
Proof. Suppose ( is an open cover of Y . For each y Y , the set f
1
(y)
is compact, and so there is a nite subset (
y
of ( that covers f
1
(y). Let
U
y
=

(
y
. By the lemma, for each y, there is an open set V
y
Y with
37
f
1
(y) f
1
(V
y
) U
y
. Then the collection V
y

yY
is an open cover of
Y and has a nite subcover V
y
1
, . . . , V
y
1
. Then the collection (
y
1
(
y
n
is a nite subcover of ( covering Y .
Corollary 8.19. A product of two compact spaces is compact.
Proof. Let X and Y be compact spaces and consider the coordinate projec-
tion p: X Y Y . Of course p is continuous and surjective, and for any
y Y the preimage p
1
(y) is homeomorphic to X and is therefore compact.
So to show that p is a perfect map, we just need to show that it is closed.
Let A X Y be closed. Suppose y Y is a limit point of p(A) and sup-
pose for contradiction that y / p(A). Then X Y A is open and contains
p
1
(y) as a subset. Since X Y A is open, we can write it as a union of
basis elements of the form U
i
V
i
for U
i
open in X and V
i
open in Y . This
collection of basis elements then covers p
1
(y); since p
1
(y) is compact, we
can nd a nite subcover U
1
V
1
, . . . , U
n
V
n
. We can assume each U
i
V
i
intersects p
1
(Y ), so that V = V
1
V
n
is an open neighborhood of y in
Y . Then p
1
(V ) contains p
1
(y) but is a subset of X Y A. This implies
that y V but V p(A) = , contradicting our hypothese that y was a
limit point of p(A). So p is closed and therefore perfect, and therefore XY
is compact.
The following needs no further proof.
Corollary 8.20. A subset of R
n
is compact if and only if it is closed and
bounded.
There is also a characterization of compactness in terms of intersections
of closed sets.
Denition 8.21. A collection of sets ( has the nite intersection property
if for every nite subcollection (

, we have

,= .
Proposition 8.22. A space X is compact if and only if for every collection
( of closed sets with the nite intersection property, we have that the total
intesection

( is nonempty.
Proof. Given a collection of closed sets (, we can dene a collection of open
sets by taking the complements: B = XA[A (. Note that

( =
if and only if

B = X. If a space X is compact and ( is a set of closed


sets with

( = , then its collection of complements B covers X and has a


nite subcover B

. The collection of complements (

of B

is a nite subset of
38
( with

= . So if ( has the nite intersection property, then certainly

( ,= .
Now suppose that every collection of closed sets with the nite inter-
section property has nonempty intersection. Let B be an open cover of X.
Then the collection of complements ( has empty intersection, and therefore
has a nite subcollection (

with empty intersection. Then the collection of


complements of (

is a nite subcover of B.
Local compactness
Denition 8.23. A space X is locally compact at a point p X if p has a
compact neighborhood, and X is locally compact if it is locally compact at
every point.
Denition 8.24. Given spaces X and Y , a continuous map f : X Y is
proper if for every compact A Y , we have f
1
(A) compact in X.
Proposition 8.25. If X is a space, Y is a locally compact Hausdor space
and f : X Y is proper and injective, then f is a homeomorphism onto its
image (meaning f : X f(X) is a homeomorphism).
Proof. Let f : X Y be as above and let A X be closed. Suppose y Y
is a limit point of f(A). Let B Y be a compact neighborhood of y. Let
C = Af
1
(B) X. Then y is a limit point of f(C): if U Y is an open
neighborhood of Y , it contains a smaller open neighborhood V with V B,
and V contains a point of f(A)y which is therefore in f(C). Since f
is proper, f
1
(B) is compact, and since A is closed, C = A f
1
(B) is
compact. Since Y is Hausdor, f(C) is compact, and therefore closed. So
y f(C), and therefore y f(A), and f(A) is closed. Since f sends closed
subsets of X to closed subsets of f(X), f is closed. Since it is injective,
f : X f(X) is a closed bijection and is therefore a homeomorphism.
Denition 8.26. Let X be a Hausdor space. The one-point compactica-
tion of X is the space Y = X y
0
, where the open sets are:
U X Y for any open U X, and
y
0
XA for any compact subset A of X.
Proposition 8.27. Let X be a Hausdor space and let Y be its one-point
compactication. Then Y is a compact space, and is Hausdor if and only
if X is locally compact.
39
Proof. Let y
0
denote the point in Y X. First we show that Y is indeed a
space. Of course the intersection of two open subsets of X is open in Y . If
U = y
0
XA and V U are open, then U V = UA, which is an open
subset of X and is therefore open in Y . If instead V = y
0
XB, then
U V = y
0
X(A B), which is open because A B is compact. If
A
i

iI
is a collection of compact subsets of X, then since X is Hausdor it
is closed, and

iI
A
i
is closed and is compact because it is a closed subset of
the compact set A
j
for some xed j I. Then the union of y
0
XA
i

iI
is y
0
X(

iI
A
i
), which is open. If we take a union of a collection of
open sets, some of which contain y
0
and some of which do not, then we
can take the union by separately taking the unions of those containing y
0
and then those that do not, and then taking the union of these two sets. Of
course, the two sub-unions are both open. If U is open in X and y
0
XA
is open in Y , then their union is y
0
X(AU). Since AU is a closed
subset of a compact subset A of X, it is compact, and unions of open sets
in Y are open. Of course is open in Y . The entire space Y is open in Y
because is a compact subset of X.
Next we show that Y is compact. Suppose ( is an open covering of Y .
Then there is some U ( with y
0
U, and some compact A X with
U = y
0
XA. Then (U is an open covering of A, and has a nite
subcover (

. Then U (

is a nite subcover of ( covering Y .


If x, y X, then there are disjoint open neighborhoods of x and y in
Y because X is Hausdor. If x X and then there are disjoint open
neighborhoods of x and y
0
in Y if and only if X is locally compact: since X
is locally compact, we have some A, U X with x U A with U open
and A compact; then U and y
0
XA are disjoint open neighborhoods
respectively containing x and y
0
. Conversely, if x and y
0
have disjoint open
neighborhoods U and V , then V = y
0
XA for some compact A and
disjointness implies U A (so that X is locally compact at X).
Proposition 8.28. Suppose X and Y are locally compact Hausdor spaces
with respective one-point compactications

X and

Y , with

XX = x
0
and

Y Y = y
0
. Then a continuous map f : X Y extends to a continuous
map

f :

X

Y with

f(x
0
) = y
0
if and only if it is proper.
Proof. First we suppose that f is the restriction of

f. Let A Y be compact.
Then U =

Y A is open in

Y . Then

f
1
(U) is open in

X. Since

f(x
0
) = y
0
,
we have x
0


f
1
(U). Note that

f
1
(U) = x
0
Xf
1
(A). In particular,
this implies f
1
(A) is compact and therefore f is proper.
Now suppose f is proper and extend it to

f as described. Let U

Y
be open. If U Y , then

f
1
(U) = f
1
(U) is open. If y
0
U, then
40
A =

Y U is a compact subset of Y . Then f
1
(A) is compact since f is
proper, and therefore

f
1
(U) = x
0
Xf
1
(A) is open in

X. Therefore

f is continuous.
41

You might also like