Académique Documents
Professionnel Documents
Culture Documents
so that a point B in the product consists of coordinates B chosen from the \ 's. But what exactly
does a symbol like B B mean if there are uncountably many coordinates?
We can get an idea by first thinking about a countable product. For sets \" \# \8 ,... we can
informally define the product set as a certain set of sequences: \
8" \8 B8 B8 \8 .
But if we want to be careful about set theory, then a legal definition of \ should have the form
\ B8 Y B8 \8 . From what pre-existing set Y will the sequences in \ be chosen? The
answer is easy: we write
\
8" \8 B 8" \8 B8 B8 \8
239
A point B in \ E is, by definition, a function that chooses a coordinate B from each set
in the collection \ E. To say that such a choice function must exist if all the \ 's are
nonempty is precisely the Axiom of Choice. (See the discussion following Theorem I.6.8.)
Theorem 2.2 The Axiom of Choice (AC) is equivalent to the statement that every product of
nonempty sets is nonempty.
Note: In ZF set theory, certain special products can be shown to be nonempty without using AC.
1) If E g, then \ E E \ E B \ g
(a set of functions)
But there is an obvious way to identify these two sets: the ordered pair B" B# corresponds to the
function B " B" # B# \" \# "#
4) If E , then \8 8
8" \8 B 8" \8 B8 \8
B" B# B8 B8 \8 the set of all sequences B8 where B8 \8
240
Definition 2.4 Let \ be a set. For each E, suppose \ g is a topological space and that
0 \ \ . The weak topology on \ generated by the collection Y 0 E is the smallest
topology on \ that makes all the 0 's continuous.
Certainly, there is at least one topology on \ that makes all the 0 's continuous: the discrete topology.
Since he intersection of a collection of topologies on \ is a topology (why? ), the weak topology
exists we can describe it from the top down as {g : g is a topology on \ making all the 0 's
continuous
However, this slick description of the weak topology doesn't give a useful description of what sets are
open. Usually it is more useful to describe the weak topology on \ from the bottom up. to make all
the 0 's continuous it is necessary and sufficient that
for each E and for each open set Y \ , the set 0" Y must be open.
Therefore the weak topology g is the smallest topology that contains all such sets 0" Y and that is
the topology for which 0" Y E Y open in \ is a subbase. (See Example III.8.6.)
Therefore a base for the weak topology consists of all finite intersections of sets from . A typical
basic open set has form 0"
Y" ] 0"
Y# ... 0"
Y8 where each 3 E and each Y3 is
"
#
8
open in \3 . To cut down on symbols, we will use a special notation for these subbasic and basic open
sets:
We will write Y for 0" Y
A typical basic open set is then Y 0"
Y" ] 0"
Y# ... 0"
Y8
"
#
8
which we further abbreviate as Y Y" Y# Y8
So B Y Y" Y# Y8 iff 03 B Y3 for each 3 " 8
This notation is not standard but it should be because it's very handy. You should verify that to get a
base for the weak topology g on \ , it is sufficient to use only the sets Y" Y# Y8 where
each Y3 is a basic (or even subbasic) open set in \3 .
Example 2.5 Suppose \ g is any topological space and E \ Let 3 E \ be the inclusion
map 3+ + Then the subspace topology on E is the same as the weak topology generated by
Y 3 To see this, just note that a base for the weak topology is 3" Y Y open in \ and that
3" Y Y E which is a typical open set in the subspace topology.
The following theorem tells us that a map 0 into a space \ with the weak topology is continuous iff
each composition 0 0 is continuous.
Theorem 2.6 For E, suppose 0 \ \ g and that \ has the weak topology generated by
the functions 0 . Let ^ be a topological space and 0 ^ \ . Then 0 is continuous iff
0 0 ^ \ is continuous for every .
Proof If 0 is continuous, then each composition 0 0 is continuous. Conversely, suppose each
0 0 is continuous. To show that 0 is continuous, it is sufficient to show that 0 " Z is open in ^
whenever Z is a subbasic open set in \ (why? ). So let Z Y with Y open in \ . Then
0 " Z 0 " 0" Y 0 0 " Y which is open since 0 0 is continuous.
241
Definition 2.7 For each E let \ g be a topological space. The product topology g on the
set \ is the weak topology generated by the collection of projection maps Y 1 E
The product topology is sometimes called the Tychonoff topology. We always assume that a
product space \ has the product topology unless something else is explicitly stated.
Since the product topology is a weak topology, a subbase for the product topology consists of all sets
of form Y 1" Y , where E and Y is open in \ A base consists of all finite
intersections of such sets
Y" Y8 1"" Y" ] ... 1"8 Y8
B \ B3 Y3 for each 3 " 8
A basic open set in \ depends on only finitely many coordinates in the following sense:
B Y" Y# Y8 iff B satisfies the finitely many restrictions B3 Y3 3 " 8.
Since (basic) open sets containing B are the standard for measuring closeness to B, we can
say, roughly, that in the product topology closeness depends on only finitely many
coordinates.
For a finite index set E " # 8, the product topology on E \ is just the one for which the
basic open sets are the open boxes E Y :
E Y 83" Y3 Y" Y# Y8 Y" Y# Y8
Then condition (*) that Y \ for all but finitely many ' s is satisfied automatically so, for finite
products the product topology is exactly as we defined it in Chapter III.
Definition 2.7 is probably not what was expected. Someone's first guess would more likely be to
use all boxes E Y (Y open in \ as a base in defining the product topology. But in Definition
2.7, a box of the form E Y might not be open because (*) may not hold. It is perfectly possible,
of course, to define a different topology on the set E \ using all boxes of the form E Y as a
base. This alternate topology on E \ is called the box topology. But we will see that our
definition of the product topology is the right definition to use. (Of course, the box topology and the
product topology coincide for finite products.)
Proof By definition, the product topology is one that makes all the 1 's continuous. and clearly, each
1 is onto if the product is nonempty. To show that 1 is open, it is sufficient to show that the image
of a basic open set Y Y" Y# Y8 is open. This is clearly true if Y g.
242
Y 3
\
if 3
..
if " 8
Example 2.9
1) A subbasic open set in has form 1"" Y Y or 1#" Z Z , where Y and
Z are open in (We get a subbase even if we restrict ourselves just to using basic open intervals
Y Z in ). So basic open sets have form Y Z Y Z . Therefore the product
topology on is the usual topology on # .
The function 0 given by 0 > ># sin ># is continuous because the
compositions 1" 0 > ># and 1# 0 > sin ># are both continuous functions from to .
2) Let \ i! the product of countably many copies of . A base for the product
topology consists of all sets Y Y8 where finitely many Y8 's are singletons and all the others are
equal to Each Y is infinite (in fact lY l - why?), so every nonempty open set is infinite. In
particular, \ has no isolated points More generally, an infinite product of (nonempty) discrete spaces
is not discrete.
The box topology on \ is the discrete topology. For each point B 5" 5# 58 , ... \ ,
the set B
8" 58 is open in the box topology. (For a finite product, the box and product
topologies care the same: a finite product of discrete spaces is discrete.)
!
"
j
,
where
j# has its usual metric, ..
#
8"
8
#
8
Suppose B C L and let % !. What condition on C will guarantee that .C B % ?
243
Pick R so that
3R "
"
3#
%#
#
If B3 C3 #
%#
#R
.B C B3 C3 # B3 C3 # "# R
R
3"
3R "
%#
#R
%# "#
#
metric topology on the product L , and we see here that we can make C close to B by
requiring closeness in just finitely many coordinates " R This is just what the product
topology does and, in fact, the product topology on L turns out to be the topology g. .
In ! " we can see a similar phenomenon quite clearly: for two points B !B" B8 and
C !C" C8 C will be close to B if C and B agree in, say, the first 8 decimal places.
Roughly speaking, closeness depends on only finitely many coordinates.
A handy rule of thumb that has proved true every time I've used it is that if a topology on a
product set makes closeness depend on only a finite number of coordinates, then that
topology is the product topology.
3) The bottom line: a mathematical definition justifies itself by the fruit it bears. The
definition of the product topology will lead to some beautiful theorems. For example, we will
see that compact Hausdorff spaces are (topologically) nothing other than the closed subspaces
of cubes ! "7 with the product topology (7 may be infinite). For the time being, you will
need to accept that things work out nicely down the road, and that by contrast, the box
topology turns out to be rather ill-behaved. (See Exercise E11.)
As a simple example of nice behavior, the following theorem is exactly what one would hope for and
the proof depends on having the correct definition for the product topology. The theorem says that
convergence of sequences in a product is coordinatewise convergence : that is, in a product,
B8 B iff for all , the >2 coordinate of B8 converges (in \ to the >2 coordinate of B. The
product topology is sometimes called the topology of coordinatewise convergence.
In the proof, R is the max of a finite set. If \ has the box topology, the basic open set Y Y
might involve infinitely many open sets Y \ . For each such , we could pick R , just as in the
proof. But the set of all R 's might not have a max R so the proof would collapse. Can you create a
specific example with the box topology where this happens?
Example 2.12 Consider \< < , where \< . Each point 0 in the product is a
function 0 Suppose that 08 is a sequence of points in . By Theorem 2.11, 08 0
iff 08 < 0 < for each < . With the product topology, convergence of a sequence of functions
in is called (in analysis) pointwise convergence.
244
The following theorem is stated loosely. You can easily create variations. Any reasonable version
of the statement is probably true.
Theorem 2.13 Topological products are associative in any reasonable sense: for example, if
E F G where F and G are disjoint indexing sets, then
\ E \" " F \# # G
0 is a mapping into a product Y ^ , so 0 is continuous iff 1" 0 and 1# 0 are both continuous.
But 1" 0 \ E ] \" F is also a map into a product so 1" 0 is
continuous iff 1" 1" 0 \ E \" is continuous for all " F
But 1" 1" 0 1" \ E \" which is continuous.
The proof that 1# 0 is continuous is completely similar.
245
Exercise 2.14 Recall that for a space \ and cardinal 7, \ 7 denotes the product of 7 copies of the
space \ . Prove that \ 7 8 is homeomorphic to \ 78 . (Hint: Look at the proof of Theorem I.14.7;
the bijection 9 given there is a homeomorphism.)
Notice that cancellation properties may not be true. For example, ! and ! " are
homeomorphic (both are countable discrete spaces) but topologically you can't cancel the ! is
not homeomorphic to 0,1 !
Here are a few results which are quite simple but very handy to remember. The first states that
singleton factors are topologically irrelevant in a product.
Lemma 2.15
E \ " F :" E \
Proof " F :" is itself a one-point space :}, so we only need to prove that
E \ : E \ . The map 0 B : B is clearly a homeomorphism
Lemma 2.16 Suppose E \ g. For any F E, F \ is homeomorphic to a subspace ^
of E \ that is, F \ can be embedded in E \ In fact, if all the \ 's are X" -spaces,
then F \ is homeomorphic to a closed subspace ^ of E \ .
Proof Pick a point : : E \ . Then by Lemma 2.15,
F \ F \ EF : ^ E \
Now suppose all the \ 's are X" . If C E \ ^ , then for some # E F C# :# Since
\# is a X" space, there is an open set Y# in \# that contains C# but not :# . Then C Y# and
Y# F \ EF : g.
Therefore ^ is closed in E \ .
Note: 1) Assume all the \ 's . Go through the preceding proof step-by-step when E " 5
and when E and
2) In the case F ! , Lemma 2.16 says that each factor \! is homeomorphic to subspace
of E \ (a closed subspace if all the \ 's are X" ).
3) But Lemma 2.16 does not say that if all the \ 's are X" , then every embedded copy of
\! in E \ is closed: only that there exists a closed homeomorphic copy. (It is very easy to show
a copy of embedded in # that is not closed in # , for example ... ?)
Lemma 2.17 Suppose \ \ E g. Then \ is a X" space (or, X# space) iff every
factor \ is X" (or, X# ).
246
Proof for Hausdorff Suppose all the \ 's are Hausdorff. If B C \ , then B! C! for some
! . Pick disjoint open sets Y! and Z! in \! containing B! and C! .
Then
Theorem 2.19 The product of countably many 2-point discrete spaces is homeomorphic to the Cantor
set G .
247
Exercises
E1. Does it ever happen that \ ! open in \ ? If so, what is a necessary and sufficient
condition on \ for this to happen?
E2. a) Suppose \ and ] are topological spaces and that E \ F ] Prove that
int\] (E F int\ E int] F: that is, the interior of the product is the product of the interiors.
(By induction, the same result holds for any finite product.) Give an example to show that the
statement may be false for infinite products.
b) Suppose E \ for all E. Prove that in the product \ \ ,
cl (E cl E .
Note: When the E 's are closed, this shows that E is closed: so any product of closed sets is
closed. Can you see any plausible reason why products of closures are better behaved than products
of interiors?
248
E6.
a) Suppose a34 ! # for all 3 4 . Prove that there exists a sequence n5 in such that,
for each 3, lim a3,85 exists.
5
Hint: picture the +34 in an infinite matrix. For each fixed 4, the j-th column" of the matrix is a
point in the Cantor set G ! #i! , and G is a compact metric space.
b) Let +8 be an absolutely convergent sequence of real numbers. A series +8w where each
8"
Hint: Absolute convergence just guarantees that every subseries converges. Each subseries +8w
8"
8"
can be associated in a natural way with a point B ! "i! Consider the mapping 0 ! "i!
given by 0 B +8w Must 0 be a homeomorphism?
8"
b) Prove 7 is separable (Hint: When 7 is infinite, consider the simplest possible points in
the product.
Part b) implies that an arbitrarily large product of X" spaces with more than one point can
be separable. According to Theorem 3.8, this statement is false for X# spaces.
249
E8. In any set \ , we can define a topology by choosing a nonempty family of subsets Y and defining
closed sets to be all sets which can be written as an intersection of finite unions of sets from Y .
Y is called a subbase for the closed sets of \ . (This construction is complementary to generating a
topology on \ by using a collection of sets as a subbase for the open sets.)
a) Verify that this procedure does give a topology on \ .
b) Suppose \ is a topological space. Give \ the topology for which collection of closed
boxes Y J J closed in \ is a subbase for the closed sets. Is this topology the product
topology?
250
3. Productive Properties
We want to consider how some familiar topological properties behave with respect to products.
Definition 3.1 Suppose that, for each E, the space \ has a certain property T . We say that
productive
if \ must have property T
if \ must have property T when E is countable
countably
productive
the property T is
finitely productive
if \ must have property T when E is finite
For example, Lemma 2.17 shows that the X" and X# properties are productive.
Some topological properties behave very badly with respect to products. For example, the Lindelof
It is sufficient to show that a collection h of basic open sets covering \ has a countable
subcover. Given such a cover h , Let i + , + , h and define E i . For a
moment, we can think of E as a subspace of with its usual topology. Then E is Lindelof
(why?), and i is a covering of E by usual open sets, so there is a countable subfamily i w with
i w E.
Replace the left endpoints of the intervals in i w to get h w + , + , i w h . If h w
covers \ we are done, so suppose \ h w g. For each B \ h w , pick a set + , in
h that contains B. In fact, B must be the left endpoint of + , for if B + , h and
B +, then B i i w h w So, for each B h w we can pick a set [B, ,B ) h .
If B and C are distinct points not in h w , then B, ,B ) and C, ,C ) must be disjoint (why?). But
there can be at most countably many disjoint intervals B ,B So h w B, ,B ) B h w } is
a countable subcollection of h that covers W .
However the Sorgenfrey plane W W is not Lindelof.
If it were, then its closed subspace
H B C B C " would also be Lindelof
(Theorem III.7.10). But that is impossible
since H is uncountable and discrete in the subspace topology. (See the figure on the following
page.)
251
Fortunately, many other topological properties interact more nicely with products. Here are several
topological properties T to which the same theorem applies. We combine these into one large
theorem for efficiency.
Theorem 3.3 Suppose \ \ E g. Let T be any one of the properties first
countable, second countable, metrizable, or completely metrizable. Then \ has property T iff
1) all the \ 's have property T , and
2) there are only countably \ 's that do not have the trivial topology.
This theorem essentially is a statement about countable products because:
1) The nontrivial \ 's are the interesting factors, and 2) says there are only countably
many of them. In practice, one hardly ever works with trivial spaces; and if we completely exclude
trivial spaces from the discussion, then the theorem just states that \ has property T iff \ is a
countable product of spaces with property T .
2) A nonempty X" space \ has the trivial topology iff l\ l " So, if we deal
only with X" spaces (as is most often the case) the theorem says that \ has property T iff all the \ 's
have property T and all but countably many of the \ 's are topologically irrelevant singletons.
Of course, in the case of metrizability (or complete metrizability), the X" condition is automatically
satisfied.
Proof Throughout, let F the set of interesting indices E \ is a nontrivial space.
We begin with the case T first countable.
Suppose 1) and 2) hold and B \. We need to produce a countable neighborhood base at B.
For each F , let Y8 8 be a countable open neighborhood base at B \ Let
UB Y Y is a finite intersection of sets of the form Y8 F 8
Since F is countable, UB is a countable collection of open sets containing B and we claim that
UB is a neighborhood base at B \. To see this, suppose Z Z" Z5 is a basic
open set containing B. (We may assume that all 3 's are in F why?). For each 3 " 5
8
pick Y 833 so that B3 Y 833 Z3 Then Y Y8"" Y8## Yk5 UB and B Y Z
252
Choose
if " F
if F
B
C"
if "
if "
253
8"
.8 B8 C8
#8
Then . is a metric on \ (check ), and we claim that g. is the product topology g . Because \
is a countable product of first countable spaces, g is first countable, so g can be described
using sequences: so it is sufficient to show that D5 D in \ g iff D5 D in \ g.
But D5 D in \ g iff the D5 converges coordinatewise. Therefore it is sufficient to
show that:
(D5 D in \ g.
iff
a8 D5 8 D8 in \8 .8 , that is,
a8 .8 D5 8 D8 !
8"
.8 D5 8 D8
#8
%
#8! .
Therefore, if 5 O
.8 D5 8! D8!
certainly ! #8!
#%8! , and therefore so .8! D5 8! D8! %.
So .8! D5 8! D8! !.
ii) On the other hand, suppose .8 D5 8 D8 ! for every 8. Let % !. Choose
R so that
8R "
"
#8
%
#
.R D5 R DR #R
8"
254
%
#R
.8 D5 8 D8
#8
8"
.8 D5 8 D8
#8
8R "
.8 D5 8 D8
#8
%
#R
%
#
Therefore .D5 D !.
Suppose T completely metrizable
What is the correct formulation and proof of the theorem for T pseudometrizable ?
We might wonder why T separable is not included with Theorem 3.3. Since separability is a
countability property, we might hope that separability is preserved in countable products although
our experience Lindelof
spaces could make us hesitate. The reason for the omission is that separability
is better behaved for products than the other properties. (You should try to prove directly that a
countable product of separable spaces is separable remembering that in the product topology,
closeness depends on finitely many coordinates. If necessary, first look at finite products.) But
surprisingly, the product of as many as - separable spaces must be separable, and the product of more
than - nontrivial separable spaces can sometimes be separable.
We begin the treatment of separability and products with a simple lemma which is really nothing but
set theory.
Lemma 3.4 Suppose lEl - There exists a countable collection e of subsets of E with the
following property: given distinct " , # , ... , 8 E, there are disjoint sets E" E# E8 e such
that 3 E3 for each 3.
Proof (Think about how you would prove the theorem when E The general case is just a
carry over on the same idea.)
Since lEl - , there is a one-to-one map 9 E . Let e 9" + , + ,
For distinct " , # ,..., 8 E, 9" , 9# , ..., 98 are distinct real numbers; we can choose
+3 , ,3 so that 93 +3 ,3 and so that the intervals +3 ,3 are pairwise disjoint. Then the sets
E3 9" +3 ,3 e are the ones we need.
Theorem 3.5
Part 2) of Theorem 3.5 is attributed (independently) to several people. In a slightly more general
version, it is sometimes called the Hewitt-Marczewski-Pondiczery Theorem. Here is an amusing
255
sidelight, written by topologist Melvin Henriksen online in the Topology Atlas. A few words have
been modified to conform with our notation:
Most topologists are familiar with the Hewitt-Marczewski-Pondiczery theorem. It states that if
7 is an infinite cardinal, then a product of #7 topological spaces, each of which has a dense
set of cardinality 7 also has a dense set with 7 points. In particular, the product of separable spaces is separable (where - is the cardinal number of the continuum). Hewitt's
proof appeared in [Bull. Amer. Math. Soc. 52 (1946), 641-643], Marczewski's proof in [Fund.
Math. 34 (1947), 127-143], and Pondiczery's in [Duke Math. 11 (1944), 835-837]. A proof
and a few historical remarks appear in Chapter 2 of Engelking's General Topology. The
spread in the publication dates is due to dislocations caused by the Second World War; there
is no doubt that these discoveries were made independently.
Hewitt and Marczewski are well-known as contributors to general topology, but who was (or
is) Pondiczery? The answer may be found in Lion Hunting & Other Mathematical Pursuits
edited by G. Alexanderson and D. Mugler, Mathematical Association of America, 1995. It is a
collection of memorabilia about Ralph P. Boas Jr. (1912-1992), whose accomplishments
included writing many papers in mathematical analysis as well as several books, making a lot
of expository contributions to the American Mathematical Monthly, being an accomplished
administrator (e.g., he was the first editor of Mathematical Reviews (MR) who set the tone for
this vitally important publication, and was the chairman for the Mathematics Department at
Northwestern University for many years and helped to improve its already high quality), and
helping us all to see that there is a lot of humor in what we do. He wrote many humorous
articles under pseudonyms, sometimes jointly with others. The most famous is A Contribution
to the Mathematical Theory of Big Game Hunting by H. Petard that appeared in the Monthly
in 1938. This book is a delight to read.
In this book, Ralph Boas confesses that he concocted the name from Pondicheree (a place in
India fought over by the Dutch, English and French), changed the spelling to make it sound
Slavic, and added the initials E.S. because he contemplated writing spoofs on extra-sensory
perception under the name E.S. Pondiczery. Instead, Pondiczery wrote notes in the Monthly,
reviews for MR, and the paper that is the subject of this article. It is the only one reviewed in
MR credited to this pseudonymous author.
One mystery remains. Did Ralph Boas have a collaborator in writing this paper? He certainly
had the talent to write it himself, but facts cannot be established by deduction alone. His son
Harold (also a mathematician) does not know the answer to this question...
Proof 1) Let H be a countable dense set in \ . For each 1 H is countable and dense in \
(because \ 1 \ 1 cl H cl 1 H ). Therefore each \ is separable.
2) Choose a family e as in Lemma 3.4 and for each , let H B1 , B2 B8 be a
countable dense set in \ . Define a countable set f by
f E" E8 6" 68 8 63 , E3 e with the E3 's pairwise disjoint
Pick a point : \ for each . For each #8-tuple = E" E8 6" 68 f , define a point
B= \ with coordinates
256
B=
B63
:
if E3
if E3
Let H B= = f . H is countable and we claim that H is dense in \ . To see this, consider any
nonempty basic open set Y Y" Y5 we will show that Y H g
For Y Y" Y5 g,
i) Choose disjoint sets E" E5 in e so that " E" 5 E5
ii) For each 3 " 5 H3 is dense in \3 and we can pick a point B833 in H3 Y3
B"3 , B#3 B83 Y3
Let = E" E3 E5 8" 83 85 f Because 3 E3 , we have B= 3 B833 Y3 .
Therefore B= Y H.
Example 3.6 The rather abstract construction of a dense set H in the proof of Theorem 3.5 can be
nicely illustrated with a concrete example. Consider < \< , where each \< Choose
e to be the collection of all open intervals + , with rational endpoints, and make a list these intervals
as E" E8 . In each \< choose H< ; " ; # ; 8 (Since all the H< 's are identical,
we can omit the subscript < on the points; but just to stay consistent with the notation i8 the proof,
we still index the ; 's using superscripts .) For each <, (arbitrarily) pick :< ! \<
One example of a 6-tuple in the collection f is = E' E# E& # ( %, where E" E# E$ are disjoint
open intervals with rational endpoints. The corresponding point B= is the function B=
with
#
;(
;
B= < %
;
!
The dense set H consists of all step functions (such as B= ) that are ! outside a finite union
E8" E8# E85 of disjoint open intervals with rational endpoints and which have a constant
rational value on each E3 .
Caution: In Example 2.12 we saw that the product topology on is the topology of pointwise
convergence that is, 08 0 in iff 08 < 0 < for each < . But is not first
countable (why?) so we cannot say that sequences are sufficient to describe the topology. In particular,
if 0 then 0 cl H but we cannot say that there must be sequence of step functions from H that
converges pointwise to 0
Since is not first countable, is an example of a separable space that is neither second countable
nor metrizable.
In contrast to the properties discussed in Theorem 3.3, an arbitrarily large product of nontrivial
separable spaces can sometimes turn out to be separable, as the next example shows. However,
257
Theorem 3.8 shows that for Hausdorff spaces, a nonempty product with more than - factors cannot be
separable.
Example 3.7 For each E, let \ be a set with l\ l " Choose : \ and let
g S \ : S g The singleton set : is dense in \ g , so \ is separable.
If : : \ E \ , then singleton set : is dense (why? look at a nonempty basic
open set Y .) So \ is separable, no matter how large the index set E is.
The \ 's in this example are not X" . But Exercise E7 shows that an arbitrarily large
product of separable X" spaces can turn out to be separable.
Theorem 3.8 Suppose \ E \ g where each \ is a X# space with more than one point.
If \ is separable, then lEl -
Proof For each , we can pick a pair of disjoint, nonempty open sets Y and Z in \ Let H be a
countable dense set in \ and let H Y H for each . If " E there is a point
: Y Z" H because H is dense. Then : H but : H" Y" H since
B" Y" therefore H H" . Therefore the map 9 E c H given by 9 H is one-toone, so lEl lc Hl #i! -
We saw in Corollary V.2.19 that a finite product of connected spaces is connected. In fact, the
following theorem shows that connectedness behaves very nicely with respect to any product. The
proof of the theorem is interesting because, unlike our previous proofs about products, this proof uses
the theorem about finite products to prove the general case.
Theorem 3.9 Suppose \ E \ g. \ is connected iff each \ is connected.
Just for reference, we state one more theorem here. We will not prove Theorem 3.10 until Chapter IX,
but we may use it in examples. (Of course, the proof in Chapter IX will not depend on any of these
258
examples! )
Theorem 3.10 (Tychonoff Product Theorem) Suppose \ E \ g. Then \ is compact iff
each \ is compact.
The proof in Tychonoff's Theorem is quite easy (why?). And the easy theorem that a finite
product of compact spaces is compact was in Exercise IV.E.26. .
259
Exercises
E10. Let \ ! "!" have the product topology.
a) Prove that the set of all functions in \ with finite range (sometimes called step functions) is
dense in \ .
b) By Theorem 3.5, \ is separable. Describe a countable set of step functions which is dense
in \ .
c) Let E B \ 0 is the characteristic function of a singleton set <
Prove that E, with the subspace topology, is discrete and not separable. Is E closed?
d) Prove that E has exactly one limit point, D , in \ and that if R is a neighborhood of D ,
then E R is finite.
E11. Boxes of the form Y E, where Y is open in \ , are a base for the box topology
on E \ {\ E. Throughout this problem, we assume that products have the box
topology rather than the usual product topology.
a) Show that the diagonal map 0 i! given by 0 B B B B is not continuous,
but that its composition with each projection map is continuous.
b) Show that ! "i! is not compact.
Hint: let E! ! " and E" ! ". Consider the collection h of all sets of the form
E%" E%# ... E%8 , where %" %# %8 ! "i! .
Note: [ ith the product topology, in contrast, ! "7 is compact for any cardinal 7 by the
Tychonoff Product Theorem (3.10) which we will prove later.
c) Show that i! is not connected by showing that the set E B i! B is an unbounded
sequence in is clopen.
d) Suppose \ . and \ . E are metric spaces. Prove that a function
0 \ \ (with the box topology) is continuous iff each coordinate function 0 1 0 is
continuous and each B \ has a neighborhood on which all but a finite number of the 0 's are
constant.
E12. State and prove a theorem that gives a necessary and sufficient condition for a product of spaces
to be path connected.
Suppose \ E \ g, and that, for each E, there exist disjoint nonempty open sets
Y and Z in \ . If \ is separable, then lEl -
260
It is often possible to embed a space \ in a product \ Such embeddings will give us some nice
theorems for example, we will see that there is a separable metric space \ that contains all other
separable metric spaces topologically a universal separable metric space.
To illustrate the general method we will use, consider two maps 0" ! " and 0# ! "
given by 0" B B# and 0# B /B Using these maps we can define / ! " # by
using 0" and 0# as coordinate functions: /B 0" B 0# B B# /B This map / is called the
evaluation map defined by the set of functions 0" , 0# . In this example, / is an embedding that is,
/ is a homeomorphism of ! " into # so that ! " ran/ # (see the figure).
An evaluation map does not always give an embedding: for example, the evaluation map
/ ! " # defined by the family cos #1B sin #1B is not a homeomorphism between ! " and
ran/ # (why? what is ran/? )
We want to generalize the idea of an evaluation map / into a product and to find some conditions under
which / will be an embedding.
Definition 4.1 Suppose \ and \ ( E are topological spaces and that 0 \ \ for each
The evaluation map defined by the family 0 E is the function / \ E \ given by
/B 0 B
Then /B is the point in the product \ whose th coordinate function is 0 B:\. In more informal
coordinate notation, /B 0 B
Exercise 4.2 Suppose \ E \ For each there is a projection map 1 \ \ . What is
the evaluation map defined by the family 1 E ?
261
Definition 4.3 Suppose \ and \ ( E are topological spaces and that 0 \ \ We say that
the family 0 E separates points if, for each pair of points B C \ , there exists an E
for which 0 B 0 C.
Clearly, the evaluation map / \ E \ is one-to-one for all B C \ /B /C
for all B C \ there is an E for which /B 0 B 0 C /C
the family 0 E separates points.
Theorem 4.4 Suppose \ has the weak topology generated by the maps 0 \ \ g and that
the family 0 E separates points. Then / is an embedding that is, / a homeomorphism
between \ and /\ \ .
Proof Since 0 separates points, / is one-to-one and / is continuous because each composition
1 / 0 is continuous.
/ preserves unions and (since / is one-to-one) intersections. So to check that / is an open map
from \ to /\, it is sufficient to show that / takes subbasic open sets in \ to open sets in /\.
Because \ has the weak topology, a subbasic open set has the form Y 0" Z , where Z is open in
\ . But then /Y /0" Z 1" Z /\ is an open set in /\
(Note: /Y might not be open in \ , but that is irrelevant. See the earlier example where
/B B# /B )
The converse of Theorem 4.4 is also true: if e is an embedding, then the 0 's separate points and \
has the weak topology generated by the 0 's. However, we do not need this fact and will omit the proof
(which is not very hard).
Example 4.5
Let \ . be a separable metric space. We can assume, without loss of generality, that . " \ is
second countable so there is a countable base U Y" Y8 for the open sets. For each 8, let
08 B .B \ Y8 Then 08 \ ! " is continuous and, since \ Y8 is closed, we have
08 B ! iff B Y8 If B C \ , there is an 8 such that B Y8 and C Y8 Then
08 C ! 08 B, so 08 's separate points.
We claim that the topology g. on \ is actually just the weak topology gA generated by the 08 's.
Because the functions 08 are continuous if \ has the topology g. , we know that g. gA To show
g. gA , suppose B Y g. For some 8, we have B Y8 Y and therefore 08 B - ! But
Z8 08" #- " is a subbasic open set in the weak topology and B Z8 Y8 Y Therefore
Y gA .
By Theorem 4.4, / \ ! "i! is an embedding, so \ / \ ! "i! We sometimes write this
top
as \ ! "i! . From Theorems 3.3 and 3.5, we know that ! "i! is itself a separable metrizable
space and therefore all its subspaces are separable and metrizable. Putting all this together, we get
that topologically, the separable metrizable spaces are nothing more and nothing less than the
subspaces of ! "i! .
262
i) separable metric spaces must not be very complicated since topologically they are nothing
more than the subspaces of a single very nice space: the cube ! "i!
ii) separable metric spaces can get quite complicated, so the subspaces of a cube ! "i!
are more complicated than we imagined.
"
Since ! "i!
8" ! 8 L (the Hilbert cube) j# , we can also say that topologically
the separable metrizable spaces are precisely the subspaces of j# . This is particularly amusing
because of the almost Euclidean metric . on j# :
"
.B C B8 C8 # #
8"
In some sense, this elegant Euclidean-like metric is adequate to describe the topology of any
separable metric space.
We can summarize by saying that each of L and j# is a universal separable metric space (But they
are not homeomorphic: L is compact but j# is not.)
Example 4.6
Suppose \ . is a metric space, not necessarily separable, and that Y E is a base for the
topology g. , where 7 lEl An argument completely analogous to the reasoning in Example 4.5
top
(just replace 8 everywhere with ) shows that \ ! "7 . Therefore topologically every metric
space is a subspace of some sufficiently large cube. Of course, if 7 i! , the cube ! "7 is not
itself metrizable (why? ); in general this cube will have many subspaces that are nonmetrizable. So the
result is not quite as dramatic as in the separable case.
The weight A\ of a topological space \ g is defined as minlU l U is a base for g i!
We are assuming here that the min in the definition exists: see Example 5.22 in Chapter VIII.
J or some very simple spaces, the min could be finite in which case the i! guarantees
that A\ i! convenient for purely technical reasons that don't matter in these notes).
The density $ \ is defined as minlHl H is dense in \ g i! For a metrizable space, it is
not hard to prove that A\ $ \. The proof is just like our earlier proof (in Theorem III.6.5) that
separability and second countability are equivalent in metrizable spaces. Therefore we have that for
any metric space \ .
top
\ ! "A\ ! "$ \
(*)
Notice that, for a given space \ ., the exponents in this statement are not necessarily the smallest
top
possible. For example, (*) says that ! "A ! "i! , but in fact we can do much better than
the exponent i! ! " ! " ! "" !
We add one additional comment, without proof: For a given infinite cardinal 7, it is possible to define
a metrizable space L7 with weight 7 such that every metric space \ with weight 7 can be
i!
embedded in Hi7! In other words, L7
is a metrizable space which is universal for all metric spaces
with A\ 7. The price of metrizability, here, is that we need to replace ! " by a more
complicated space L7 .
263
Without going into all the details, you can think of L7 as a star with 7 different copies of
! " (rays) all placed with ! at the center of the star. For two points B C on the same ray
of the star, .B C lB Cl; if B C are on different rays, the distance between them is
measured via the origin .B C lBl lCl
The condition in Theorem 4.4 that \ has the weak topology generated by a collection of maps
0 \ \ is sometimes not so easy to check. The following definition and theorem can
sometimes help.
Definition 4.7 Suppose \ and \ ( E are topological spaces and that, for each , 0 \ \
We say that the collection Y 0 E separates points from closed sets if whenever J is a
closed set in \ and B J , there is an such that 0 B cl 0 J
Example 4.8 Let \ and Y G. Suppose J is a closed set in and < J . There is an
open interval + , for which < + , J . Define 0 G with a graph like the one shown
in the figure:
In particular, i) ii) imply that U is a subbase for the topology on \ so that \ has the weak
topology generated by the 0 's.
264
Note: the more open (closed) sets there are in \ , the harder it is for a given family 0 E to
succeed in separating points and closed sets. In fact the lemma shows that a family of continuous
functions 0 E succeeds in separating points and closed sets only if g is the smallest
topology that makes the 0 's continuous.
Proof Suppose U is a base. Then the sets 0" Z in U are open, so the 0 's are continuous and
i) holds.
To prove ii), suppose J is closed in \ and B J . For some and some Z open in \ , we
have B 0" Z \ J Then 0 B Z and Z 0 J g since 0" Z J g.
Therefore 0 B cl 0 J so ii) also holds.
Conversely, suppose i) and ii) hold. If B S and S is open in \ , we need to find a set
0" Z U such that B 0" Z S Since B J \ S, condition ii) gives us an for which
0 B cl 0 J . Then 0 B Z \ cl 0 J Then B 0" Z and we claim 0" Z S:
Suppose D 0" Z so 0 D Z \ cl 0 J
But if D J , then 0 D 0 J 0 cl J cl 0 J
Therefore D \ J S
Proof Since the 0 's are continuous, Lemma 4.9 gives us that \ has the weak topology. Then
Theorem 4.4 implies that / is an embedding.
If the space \ we are trying to embed is a X" -space (as is most often the case), then
i) ii) in Theorem 4.10 , so we have the simpler statement given in the following corollary.
Corollary 4.11 Suppose 0 \ \ is continuous for each E. If \ is a X" -space and
0 E separates points and closed sets, then evaluation map / \ \ is an embedding.
Proof By Theorem 4.10, it is sufficient to show that the 0 's separate points, so suppose B C \ .
Since B the closed set C, there is an for which 0 B cl 0 C. Therefore 0 B 0 C so
0 E separates points.
265
Exercises
E14. A space \ g is called a X! space if whenever B C \ , then aB aC (equivalently, either
there is an open set Y containing B but not C, or vice-versa). Notice that the X! condition is weaker
than X" (see example III.2.6.4). Clearly, a subspace of a X! -space is X! .
a) Prove that a nonempty product \ \ E is X! iff each \ is X! .
b) Let W be Sierpinski space that is, W ! " with the topology g g 1 ! ". Use
the embedding theorems to prove that a space \ is X! iff \ is homeomorphic to a subspace of W 7 for
some cardinal 7. (Nearly all interesting spaces are X! , so nearly all interesting spaces are
(topologically) just subspaces of W 7 for some cardinal 7.)
Hint : for each open set Y in \ , let ;Y be the characteristic function of Y Use an embedding
theorem.
E15. a) Let \ . be a metric space. Prove that G\ ( the collection of continuous real-valued
functions on \ ) separates points from closed sets (Since \ is X" , G\ therefore also separates
points).
b) Suppose \ is any X! topological space for which G\ separates points and closed sets.
Prove that \ can be embedded in a product of copies of .
E16. A space \ satisfies the countable chain condition (CCC) if every collection of nonempty
pairwise disjoint open sets must be countable. (For example, every separable space satisfies CCC.)
Suppose that \ is separable for each E . Prove that \ \ : E satisfies CCC.
(There isn't much to prove when | A | c; why?)
Hint: Let U> > X be any such collection. We can assume all the U> 's are basic open sets (why?).
Prove that if W T and | W | c, then | W | i! and hence X must be countable.)
E17. There are several ways to define dimension for topological spaces. One classical method is the
following inductive definition.
Define dim g 1.
For : \ , we say that \ has dimension 0 at p if there is a neighborhood base at p
consisting of sets with 1 dimensional (that is, empty) frontiers. Since a set has empty frontier iff
it is clopen, \ has dimension 0 at : iff : has a neighborhood base consisting of clopen sets.
We say \ has dimension 8 at : if : has a neighborhood base consisting of sets with n 1
dimensional frontiers.
We say \ has dimension n, and write dim\ n, if \ has dimension 8 at p for each
p X and that dim(X) n if dim(\ ) 8 but dim\
8 ". We say dim\ if
dim\ 8 is false for every 8
While this definition of dim\ makes sense for any topological space \ , it turns out that
dim produces a nicely behaved dimension theory only for separable metric spaces. The
dimension function dim is sometimes called small inductive dimension to distinguish it from
other more general definitions of dimension. The classic discussion of small inductive dimension
is in Dimension Theory (Hurewicz and Wallman).
266
E18. Suppose \ and ] are X! -spaces. Part a) outlines a sufficient condition enabling ] to be
top
embedded in a product of \ 's, i.e., that ] \ 7 for some cardinal m. Parts b) and c) look at some
corollaries.
a) Theorem Let \ and ] be X! spaces. Then ] can be topologically embedded in \ 7 for
some cardinal 7 if, for every closed set J ] and every point C J , there exists a
continuous function 0 ] \ 8 such that f (y) cl f [F] (here, n is finite and depends on f ).
Proof Let X > C J J is closed in ] and C J and, for each such pair
> C 0 , let 0> be the function given in the hypothesis. Let \> \ 8 (the space
containing the values of 0> ). Then clearly \> > X is homeomorphic to \ 7 for some
7, so it suffices to show ] can be embedded in \> > X .
Let
2 ] \> > X be defined as follows: for C ] , 2> has for its >-th coordinate
0> C i.e., 2C> 0> C
i) Show 2 is continuous.
ii) Show 2 is one-to-one.
iii) Show that 2 is a closed mapping onto its range 2] to complete the proof that 2 is
+ homeomorphism between ] and 2] \> > X . (Note: the converse of the
theorem is also true. Both the theorem and converse are due to S. Mrowka.)
b) Let F denote Sierpinski space {{0,1},{g,{0},{0,1}}. Use the theorem to show that every T!
space Y can be embedded in F 7 for some m.
c) Let D denote the discrete space 0,1. Use the theorem to show that every X" -space ]
satisfying dim] ! (see Exercise E14) can be embedded in D7 for some m.
Parts b) and c) are due to Alexandroff.
Mrowka also proved that there is no T" -space \ such that every T" -space ] can be
embedded in X 7 for some m.
E19 Let \ ! " and consider the product space \ 7 , where 7 is an infinite cardinal.
267
E20. A space \ is called totally disconnected every connected subset E satisfies lEl " Prove that
a totally disconnected compact Hausdorff space is homeomorphic to a closed subspace of ! "7 for
some 7. (Hint: see Lemma V.5.6).
E21. Suppose \ is a countable space that does not have a countable neighborhood base at the point
+ \ (For instance, let + ! ! in the space P, Example III.9.8.)
i!
Let E4 B \ i! B3 + for 3 4 and E
4! E4 \ . Prove that no point in the
(countable) space E has a countable neighborhood base. (Note: it is not necessary that \ be
countable. That condition simply forces E to be countable and makes the example more dramatic.)
268
Theorem 5.2 Suppose ] has the strong topology generated by a collection of maps 1 E If
^ is a topological space and 0 ] ^ , then 0 is continuous iff 0 1 \ ^ is continuous for
each E.
1
0
Proof For each E, we have \ ] ^ , and the 1 's are continuous since ] has the strong
topology.
If 0 is continuous, so is each composition 0 1
Suppose each 0 1 is continuous and that Y is open in ^ . We want to show that 0 " Y
is open in ] . But ] has the strong topology, so 0 " Y is open in ] iff each 1" 0 " Y
is open in \ But 1" 0 " Y 0 1 " Y which is open because 0 1 is
continuous.
We introduced the idea of the strong topology as a parallel to the definition of weak topology.
However, we are going to use the strong topology only in a special case: when there is only one map
1 1 and 1 is onto.
Definition 5.3 Suppose \ g is a topological space and 1 \ ] is onto. The strong topology on
] generated by 1 is also called the quotient topology on ] . If ] has the quotient topology from 1, we
269
say that 1 \ ] is a quotient mapping and we say ] is a quotient of \ . We also say the ] is a
quotient space of \ and sometimes as ] \1.
From our earlier discussion we know that if \ is a topological space and 1 \ ] , then the quotient
topology on ] is g Y 1" Y is open in \. Thus Y g if and only if 1" Y is open in \ :
notice in this description that
only if
if
Quotients of \ are used to create new spaces ] by pasting together (identifying) several points of
\ to become a single new point. Here are two intuitive examples:
i) Begin with \ ! " and identify ! with " (that is, paste ! and " together to become a
single point). The result is a circle, W " This identification is exactly what the map 1 ! " W "
given by 1B cos #1B sin #1B accomplishes. It turns out (see below) that the usual topology on
W " is the same as quotient topology generated by the map 1. Therefore we can say that 1 is a quotient
map and that W " is a quotient of ! "
ii) If we take the space \ W " and use a mapping 1 to identify the north and south poles
together, the result is a figure-eight space ] . The usual topology on ] (from # ) turns out to be the
same as quotient topology generated by 1 (see below). Therefore we can say that 1 is a quotient
mapping and the the figure-eight is a quotient of W " .
Suppose we are given an onto map 1 \ g ] g w . How can we tell whether 1 is a quotient
map that is, how can we tell whether g w is the quotient topology? By definition, we must check that
Y g w iff 1" Y g Sometimes it is fairly straightforward to do this. But the following theorem
can sometimes make things much easier.
Theorem 5.4 Suppose 1 \ g ] g w is continuous and onto. If 1 is open (or closed), then
g w is the quotient topology, so 1 is a quotient map. In particular, if \ is compact and ] is Hausdorff,
1 is a quotient mapping.
Note: Whether the map 1 \ ] is continuous depends, of course, on the topology g w , but if g w
makes 1 continuous, then so would any smaller topology on ] . The theorem tells us that if 1 is
both continuous and open (or closed), then g w is completely determined by 1: g w is the largest
topology making 1 continuous.
Proof Suppose Y ] W/ must show Y g w iff 1" Y g If Y g w , then 1" Y g
because 1 is continuous. On the other hand, suppose 1" Y g . Since 1 is onto, 11" Y Y and
so, if 1 is open, Y 11" Y g w
For J ] 1" ] J \ 1" J It follows easily that J is closed in the quotient
space if and only if 1" J is closed in \ With that observation, the proof that a continuous, closed,
onto map 1 is a quotient map is exactly parallel to the proof when 1 is open.
If \ is compact and ] is X# , then 1 must be closed, so 1 is a quotient mapping.
Note: Theorem 5.4 implies that the map 1B cos #1B sin #1B from ! " to W " is a quotient
map, but 1 is not open. The same formula 1 can be used to define a quotient map 1 W "
which is not closed (why?). Exercise E25 gives examples of quotient maps 1 that are neither
270
Example 5.5 For +, , , define + , iff , + is even. There are two equivalence classes
! % # ! # % and " $ " " $ so ! "
Define 1 by 1+ + and give the quotient topology. A set Y is
open in iff 1" Y is open in , and that is true for every Y because is discrete.
Therefore the quotient is a two point discrete space.
Example 5.6 Let \ . be a pseudometric space and define a relation in \ by B D iff
.B D !. Let ] \ , define 1 \ ] by 1B B. Give ] the quotient topology
Points at distance ! in \ have been identified with each other to become one point (the equivalence
class) in ]
For B D ] , define . w B D .B D. In order to see that . w is well-defined,
we need to check that the definition is independent of the representatives chosen from the equivalence
classes:
If B w B and D w D, then .B B w ! and .D D w ! Therefore
.B D .B B w .B w D w .D w D .B w D w , and similarly
.B w D w .B D. Thus .B w D w .B D so . w B w D w . w B D
It is easy to check that . w is a pseudometric on ] . In fact, . w is a metric: if . w B D !, then
.B D !, which means that B D and B D
We now have two definitions for topologies on ] : the quotient topology g and the metric topology
g. w . We claim that g g. w To see this, first notice that
w
Therefore 1" F%. B F%. B and 1 F%. B F%. B. But then we have
Y g. w
iff
iff
iff
iff
Y is a union of . w -balls
1" Y is a union of . -balls
1" Y is open in \
Y g
271
The metric space ] . w is called the metric identification of the pseudometric space \ .. In effect,
we turn the pseudometric space into a metric space by agreeing that points in \ at distance ! are
identified together and treated as one point.
Note: In this particular example, it is easy to verify that the quotient mapping 1 \ ] is open,
so 1 would be a homeomorphism if only 1 were " ". If the original pseudometric . is actually a
metric, then 1 is " " and a homeomorphism: the metric identification of a metric space \ . is
itself.
Example 5.7 Exactly what does it mean if we say identify the endpoints of ! " and get a circle?
Of course, one could simply take this to be the definition of a (topological) circle. Or, it could mean
that we already know what a circle is and are claiming that a certain quotient space is homeomorphic to
a circle. We take the latter point of view.
Let 1 ! " W " be given by 1B cos #1B sin #1B) This map is onto and " " except
that 1! 1", so 1 corresponds to the equivalence relation on \ for which ! " (and there are no
other equivalences except that B B for every B We can think of the equivalence classes as
corresponding in a natural way to the points of W " .
Here W " has its usual topology and 1 is continuous. However, since \ is compact and W " is Hausdorff,
Theorem 5.4 gives that the usual topology on W " is the quotient topology and 1 is a quotient map.
When it seems apparent that the result of making certain identifications produces some familiar
space ] , we need to check that the familiar topology on ] is actually the quotient topology. Example
5.7 is reassuring: if we believed, intuitively, that the result of identifying the endpoints of ! " should
be W " but then found that the quotient topology on the set W " turned out to be different from the usual
topology, we would be inclined to think that we had made the wrong definition for a quotient.
Example 5.8 Suppose we take a square ! "# and identify points on top and bottom edges using the
equivalence relation B ! B " We can schematically picture this identification as
272
The arrows indicate that the edges are to be identified as we move along the top and bottom edges in
the same direction. We have an obvious map 1 from ! "# to a cylinder in $ which identifies points
in just this way, and we can think of the equivalence classes as corresponding in a natural way to the
points of the cylinder.
The cylinder has its usual topology from $ and the map 1 is (clearly) continuous and onto. Again,
Theorem 5.4 gives that the usual topology on the cylinder is, in fact, the quotient topology.
273
Example 5.9 Similarly, we can show that a torus (the surface of a doughnut) is the result of the
following identifications in ! "# : B ! B " and ! C " C
Thinking in two steps, we see that the identification of the two vertical edges produces a cylinder; the
circular ends of the cylinder are then identified (in the same direction) to produce the torus.
The two circles darkly shaded on the surface represent the identified edges.
We can identify the equivalence classes naturally with the points of this torus in $ and just as before
we see that the usual topology on the torus is in fact the quotient topology.
274
Example 5.10 Define an equivalence relation in ! "# by setting B ! " B " Intuitively,
the idea is to identify the points on the top and bottom edges with each other as we move along the
edges in opposite directions. We can picture this schematically as
Physically, we can think of a strip of paper and glue the top and bottom edges together after making a
half-twist. The quotient space \ is called a Mobius
strip.
strip Q in $ and define a map 1 ! "# Q that accomplishes the identification we have in mind.
In that case there is a natural way to associate the equivalence classes to the points of the torus in $
and again Theorem 5.4 guarantees that the usual topology on the Mobius
strip is the quotient topology.
275
Example 5.11 If we identify the vertical edges of ! "# (to get a cylinder) and then identify its
circular ends with a half-twist (reversing orientation): ! C " C and B ! " B ". We get
a quotient space which is called a Klein bottle.
It turns out that a Klein bottle cannot be embedded in $ the physical construction would require a
self-intersection (additional points identified) which is not allowed. A pseudo-picture looks like
In these pictures, the thin neck of the bottle actually intersects the main body in order to re-emerge
from the inside in a real Klein bottle (in % , say), the slef-intersection would not happen.
In fact, you can imagine the Klein bottle as a subset of % using color as a 4>2 dimension. To each
point B C D on the bottle pictured above, add a 4th coordinate to get B C D < Now color the
points on the bottle in varying shades of red and let < be a number measuring the intensity of red
coloration at a point. Do the coloring in such a way that the surface blushes as it intersects itself
so that the points of intersection seen above in $ will be different (in their 4th coordinates).
Alternately, you can think of the Klein bottle as a parametrized surface traced out by a moving point
T B C D > where B C D depend on time > and > is recorded as a 4th -coordinate. A point on the
surface then has coordinates of form B C D >. At a point where we see a self-intersection in $
there are really two different points (with different time coordinates >).
276
Example 5.12
1) In W " , identify antipodal points that is, in vector notation, T T for each T W " .
Convince yourself that the quotient W " is W " .
2) Let H# be the unit disk T # lT l ". Identify antipodal points on the boundary of
H : that is, T T if T W " H# The quotient H is called the projective plane, a space
which, like the Klein bottle, cannot be embedded in $ .
#
3) For any space \ , we can form the product \ ! " and let B " C " for all
B C \ . The quotient \ ! " is called the cone over \ . (Why?)
4) For any space \ , we can form the product \ " " and define B " C " and
B " C " for all B C \ . The quotient \ " " is called the suspension of \ .
(Why?)
There is one other very simple construction for combining topological spaces. It is often used in
conjunction with quotients.
Definition 5.13 For each E, let \ g be a topological space, and assume that the sets \ are
pairwise disjoint. The topological sum (or free sum) of the \ 's is the space E \ g where
g S E \ S \ is open in \ for every E}. We denote the topological sum by
\ . In the case lEl #, we use the simpler notation \" \#
In \ , each \ is a clopen subspace. Any set open (or closed) in \ is open (or closed) in the sum.
The topological sum \ can be pictured as a union of the disjoint pieces \ , all far apart from
E
1) If the \ 's are nonempty and separable, then \ is separable iff lEl i! .
2) If the \ 's are nonempty and second countable, then \ is second countable iff lEl i! .
E
277
. B C if B C \
.B C
"
otherwise
so \ is metrizable if all the \ 's are metrizable. You should be able to find similar statements for
E
"
bounds it. Of course, F W . Let 0 E F be a homeomorphism. The we can use 0 to join the
spaces by sewing the edge of the Mobius
strip to the edge of the hole in W # . The result is a sphere
with a crosscap.
There is a very nice theorem, which we will not prove here, which uses all these ideas. It is a
classification theorem for certain surfaces.
Definition 5.18 A Hausdorff space \ is a 2-manifold if each B \ has an open neighborhood Y that
is homeomorphic to # . Thus, a 2-manifold looks locally just like the Euclidean plane. A surface is
a Hausdorff 2-manifold.
Theorem 5.19 Let \ be a compact, connected surface. The \ is homeomorphic to a sphere W # or to
W # with a finite number of handles and crosscaps attached.
You can read more about this theorem and its proof in Algebraic Topology: An Introduction (William
Massey).
278
Exercises
E22. a) Let be the equivalence relation on # given by B" C" B# C# iff C" C# . Prove that
# is homeomorphic to .
b) Find a counterexample to the following assertion: if is an equivalence relation on a
space \ and each equivalence class is homeomorphic to the same space ] , then \ ] is
homeomorphic to \ .
Why might someone ever wonder whether this assertion might be true? In part a), we have
\ , each equivalence class is homeomorphic to and (\ \ . In this
example, you divide out equivalence classes that all look like , then multiply by , and you're
back where you started.
c) Let 1 # be given by 1B C B# C# .
homeomorphic to what familiar space?
d) On # , define an equivalence relation (B" C" B# C# iff B" C"# B# C## . Prove
that / is homeomorphic to some familiar space.
#
E23. For B C ! ", define B C iff B C is rational. Prove that the corresponding quotient
space ! " is trivial.
E25. a) Let ]" ! " with its usual topology and ]# ! " with the discrete topology. Define
1 ]" ]# ] by letting 1 be the identity map on both ]" and ]# Prove that 1 is a quotient map
that is neither open nor closed.
b) Let # have the topology g for which a subbase consists of all the usual open sets together
with the singleton set ! !. Let have the usual topology and let 0 # be the projection
0 B C B. Prove that 0 is a quotient map which is neither open nor closed.
E28. Let ]" ! " and let ]# ]" ! ". Prove ]" is not homeomorphic to ]# but
279
E29. Show that no continuous image of can be represented as a topological sum \ ] , where
\ ] g. How can this be result be strengthened?
E30. Suppose \= (s W ) and ]> (> X ) are pairwise disjoint spaces. Prove that \= ]> is
homeomorphic to
\= ]>
>X
=W
=W>X
E31. This problem outlines a proof (due to Ira Rosenholtz) that every nonempty compact metric space
\ is a continuous image of the Cantor set G . From Example 4.5, we know that \ is homeomorphic to
a subspace of ! "i!
a) Prove that the Cantor set G consists of all reals of the form
4!
+4
$4
where
each +4 ! or 2.
b) Prove that ! " is a continuous image of G . Hint: Define 1
4!
+4
$4
4!
+4
#4
4!
w
'
280
Chapter VI Review
Explain why each statement is true, or provide a counterexample.
14. If \
8" E8 , where the E8 's are disjoint clopen sets in \ , then \ z
8" E8
topological sum of the E8 's).
( the
15. Let \8 ! " and ]8 with their usual topologies. Then \8 is homeomorphic to ]8
8"
8"
8" ]8
281
18. Suppose E8 is a connected subset of \8 g and that
8" E8 is dense in
8" \8 Then each
\8 is connected.
19. In , every neighborhood of the function sin contains a step function (that is, a function with
finite range).
20. Let X be an uncountable set with the cocountable topology. Then {B B B \ is a closed
subset of the product \ \.
21. Let E 8" 8 , and let 7 be any cardinal number. E5 is a closed set in 5
22. If G is the Cantor set, then G G G G is homeomorphic to G G .
33. W " W " is homeomorphic to the infinity symbol:
34. 31. Let T be the set of all real polynomials in one variable, with domain restricted to ! ", for
which ranT ! " Then T is dense in ! "!"
35. Every metric space is a quotient of a pseudometric space.
37. A separable metric space with a basis of clopen sets is homeomorphic to a subspace of the Cantor
set.
38. Let {\ E} be a nonempty product space. Each \ is a quotient of the product space
\ E
-
39. Suppose \ does not have the trivial topology. Then \ # cannot be separable.
40. Every countable space \ is a quotient of
41. is homeomorphic to the sum of i! disjoint copies of .
43. The unit circle, W " , is homeomorphic to a product E \ , where each \ ! "
(i.e., W " can be factored into a product of subspaces of ! " ).
44. i! is homeomorphic to .
282