Professional Documents
Culture Documents
OF
TOPOLOGY
C.O. Christenson
~ W.L. Voxman
5B Associates
w, Idaho, U.S.A.
Digitized by the Internet Archive
in 2022 with funding from
Kahle/Austin Foundation
https://archive.org/details/aspectsoftopologOO0O0O0chri_j6l7
Second Edition
ASPECTS OF
1IQOPOLOGY
CHARLES O. CHRISTENSON
WILLIAM
L. VOXMAN
Department of Mathematics
College of Letters and Science
University of Idaho
Moscow, Idaho
Second Edition
Bryan A. Smith, Editor
Copyright ©1998 by Charles O. Christenson and
William L. Voxman; published by BCS Associates, Moscow, Idaho, USA.
p. cm.
QA611.C436 1998
514—de2 98-45593
10987654321
Winston S. Churchill
Preface to Second Edition
Over twenty years have passed since the original edition was published. Dur-
ing those years our students and colleagues have from time to time forcefully
brought to our attention various shortcomings. In this reincarnation of the
book we have attempted to minimize confusion by rewriting and/or rephras-
ing many of the proofs and explanatory passages. No doubt, however, in the
process new ambiguities have arisen. These should be viewed as our way of
ensuring that topology instructors will continue to be needed in the future.
Substantive changes have occurred in chapters 8, 13, 15 and 18. The
old chapter 18 has been partially tucked into chapter 8 and pruned. We
have added a new chapter 18 that gives a brief introduction to the theory
of dimension. The last two of sections of chapter 13 have been reorganized
significantly. We have rewritten the section of chapter 8 dealing with CW-
complexes, and the of chapter 15 dealing with the fundamental group of the
complement of a graph to make them more readable.
We have also tried to make the index more complete and usable, and
all of the many figures have been redrawn.
Many of the problems at the end of chapters have been reformulated,
added, and/or shuffled. We view these problems and the students agonizing
over them to be essential to any course based on this book. In the first
edition, we marked problems we thought would give students difficulty by
*. However, over the years it became clear to us that what is difficult and
what is easy is often a highly individual matter, and the *’s were serving
no purpose other than to scare some students from attempting problems.
None of the *’s have been retained. Nevertheless we can say with assurance
that students will find some problems easy, and some nearly impossible. Our
advice to the student: Do not be discouraged when you run into a problem
you can not solve—but keep thinking about it!
Vv
There is more material in the text than any instructor should cover in a
year’s course. For what it is worth, we generally cover chapters 1,2,3,4,6,7,8,
12,13, and depending on the interest and ability of the class (and the whims
of the instructor) an additional two to four chapters.
We are grateful to our students over the years who have directly and
indirectly pointed out ambiguities and other sources of confusion in the first
edition. We particularly thank K. Meerdink, C. Steenberg, and S. Stockett.
We also thank our colleagues John Cobb and Mark Nielsen for their help and
stimulating discussions. Finally we thank Bryan Smith, who not only helped
with the original edition, but has functioned as editor for this edition.
C.Christenson
W.Voxman
Moscow, Idaho, August 1998
vl
CONTENTS
Peeinces 44, Sian gk. Pir sc kg ek pk cae eal
Chapter OF 'Prelmmaries 90". . . "2, attietiisegedt
en’)
af). ae
A. Basic Notation
B. Cartesian Products and Relations
C. Functions
D. The Axiom of Choice and Some of its Variants
E. Cardinal Numbers and Ordinal Numbers
F. Minkowski’s Inequality
Chapter I: The Basic:Constructe wnu ao, Se, ote? sere 8 11
. Topological Spaces and Continuity
. Further Examples of Topological Spaces
. Metric Spaces: A Preview
. Building New Spaces from Old Ones
. A Potpourri of Fundamental Concepts
. Continuity
Bs
coi
Coe
Sey . Homeomorphisms
Problems
Chapter 2: Connectedness and Compactness ...........47
. Connectedness: General Results
. Connectedness: Slightly Deeper Results
. Path Connectedness
. Components
. Locally Connected Spaces
. Simple Chains
. Compactness
= . Two Characterizations of Compactness
[o2)
@)
ke)
[esl
23)
(ep)
an)
I. Compactness: Local, Countable, and Sequential
J. The One Point Compactification
Problems
Chapter 3: Metric Spaces . Pk
A. Compactness in a Metric Setting
B. Complete Metric Spaces
C. Convex and Hausdorff Metrics
Problems
vii
viii Contents
PRELIMINARICS
7 a ae Ee amet
' ; a
i 7
A HAS & Lr vi .
e 7 26@e ss ’ a
e » e
ih)7 s
“ad -: i wise _
‘
—a s -
> a ? -
Le é > a" Td =
— : ,
- = Pan ihe. af
é th) eenmw O¢
-! an a ae @ 7.
acy orhee aahta
¢ ps Phual ale i : a” =
Pe Adee a
hp-<enile’ vas on 6
4, Tie Oe
ts Yim * :
\ taped ae a Tan) T agus fe -
QD Dreet Lust whim
HNO. weed ;
Chapter 0
PRELIMINARIES
In this chapter we introduce much of the notation used throughout the book.
In addition, we give a sketch of the elements of set theory that is needed in
the ensuing chapters. No attempt is made to give a full presentation of this
material; for more complete expositions of set theory, see the books by Monk
[1969], Halmos [1960], as well as Hrbacek and Jech [1984].
The proofs of most propositions found in this chapter are left to the
reader, presumably as review exercises. Our recommendation for the use of
this chapter is to initially scan it, and then refer back as necessary while
reading the rest of the book.
A. BASIC NOTATION
The following symbols are used frequently:
is a subset of
is a proper subset of
is an element of
is not an element of
the set of integers (also considered as an additive group)
the set of positive integers
+-
NNRWAWN
the set of natural numbers
1 the set of real numbers
the set of rational numbers
the interval {x € R! : 0<a< 1}
n
the set {(21,22,.-., 2a): 41 E R'}
ee
ee the empty set
P(X) the set of subsets of X
1
2 0. Preliminaries
C. FUNCTIONS
(0.C.1) Definition. Suppose that X and Y are sets. A function from X
into Y, denoted by f : X > Y, is a subset f of X x Y with the property
that for each x € X, there is a unique y € Y such that (z,y) € f. Usually
f(z) = y is written instead of (x,y) € f. The set X is called the domain of
f and {y€Y : f(x) =y for some z € X} is the image of f; the image of f
will often be denoted by Im f or f(X).
(0.C.2) Definition. Suppose f : X + Y and A C X. The image of A
under f, denoted by f(A), is defined to be {f(z) : 2 € A}.
(0.C.3) Definition. Suppose f : X — Y is a function. Then f is onto Y
(or is a surjection) if and only if f(X) =Y.
(0.C.4) Definition. Suppose that f : X > Y is a function. Then f is one
to one (or an injection), denoted by 1-1, if and only if whenever 21, 272, € X
and 2; # Zo, then f(r1) # f(x2). If f is both 1-1 and onto, f is called a
byection.
V4
x —_“_—-y
xX ¥
Pp g
WwW Z
6 0. Preliminaries
F. MINKOWSKI’S INEQUALITY
The following inequality (Minkowski’s inequality) will frequently be useful:
Minkowski’s Inequality
nm
Sy abi <
a=1
n n n
Let
Then we have
n n 2
Qoins 2idi)
0<y Dial aibi)” 4
— heaps) = aibi)” Eh yeha
iz a ee
a pas
Sia Dia OF seal Ba.
or
lA
Proof. Let aj = x2; — yi, b: = yi — 2%, and apply the previous two
propositions. O
Chapter 1
ae te ee f(e)
a c b w —_—_——— 2
f((a,6))
The reader should verify that the two definitions are indeed equivalent. The
advantage of the latter definition is that if we can establish the notion of
“open interval” or, more generally, “open subset” of a given set, then our
definition of continuity can easily be generalized.
1
12 1. The Basic Constructs
Recall that an open set in R is any set that can be written as the
union of a family of open intervals (in particular, this will make the empty
set open, since it is the union over an empty family of open intervals). It is
easy to verify that
i) the union of any y collection of open
p sets is open, and
(ii) the intersection of any finite number of open sets is open.
We can now say that a function f : R! + R! is continuous at a point
c if and only if for each open set U containing f(c), there is an open set V
containing c such that f(V) C U. The reader should check that this defi-
nition of continuity is compatible with the previous ones. The appropriate
designation of open subsets of a set X determines what is called a topological
structure for X. Much of the beauty of topology stems from the fact that
properties (i) and (77) above are all that is needed to determine whether or
not a subfamily U of P(X) (the collection of all subsets of X) is a topolog-
ical structure, and that the idea of continuity is readily defined using this
structure.
(1.A.9) Definition. Bases that generate the same topology on a set X are
said to be equivalent.
It follows from the previous exercise that B = {(x,y) : x,y are rational}
and B = {(z,y) : 2,y are irrational} are equivalent bases for the usual
topology on R!.
Continuity may also be expressed in terms of basis elements.
(1.4.10) Theorem. Suppose (X,U/) and (Y,V) are topological spaces and
B is a basis for U and B is a basis for V. A function f : X - Y is continuous
at a point c if and only if whenever f(c) € B € B, there is a B € B such that
c€ Band f(B) CB.
Proof. Suppose f is continuous and f(c) € B © B. Since B is also in V
and f is continuous, there are sets U € U and B € B such that ce BCU
and f(c) € f(B) Ccf(U) CB.
For the converse, suppose f(c) € V € V. Then there is a B € B such
that f(c) € B C V, and by hypothesis, there isa B € B CU such that c € B
and f(B)cCBcV. Oa
Further Examples of Topological Spaces 15
(1.A.11) Exercise. Suppose (X,U/) and (Y,V) are topological spaces and
Bisa basis for Y. Show that a function f : X — Y is continuous if and only
if f—!(B) is open in X for each B € B.
(1.B.3) Exercise. Determine which of the above spaces 1-10 are Hausdorff.
(1.B.4) Definition. A topological space (X,U/) is second countable if and
only if U has a basis consisting of a countable number of sets.
(1.B.5) Exercise. Determine which of the above spaces 1-10 are second
countable.
(1.B.6) Definition. If (X,U/) is a topological space and x € X, then an
(open) basis for U at x (often called an (open) neighborhood basis for U at
x) is a subcollection B, C U with the properties that
(i) ifc € U EU, there is a V € B, such that c € V CU, and
(it) = E (\yep, V-
(1.B.7) Definition. A topological space (X,U) is first countable if and only
if there is a countable neighborhood basis at each point x € X.
Note that €” is first countable, since for each z € €”, the set {51 /;(z) :
i € Zt} is a countable neighborhood basis at z. In fact, by Problem 1 of
section B every second countable space is also first countable.
(1.B.8) Exercise. Find a space that is first countable but is not second
countable.
The function d is called the distance function or the metric for the metric
space (X,d).
The first condition implies that the distance from a point to itself is
0 and that the distance between distinct points is positive. Condition (ii)
states that if you walk from x to y, you go the same distance as if you walk
from y to x. The third condition is as a generalization of the idea that the
shortest distance between two points is a straight line, (or that the length of
one side of a triangle is less than the sum of the lengths of the other sides).
(1.C.2) Notation. Suppose that (X,d) is a metric space, € X, ande > 0.
We denote the set {y € X : d(z,y) < €} by S4(z). When the metric is clear
from context, d is frequently omitted and we simply write S,(z).
The distance function d can be used to define a topology for X. In the
next theorem we show that if (X,d) is a metric space, then B = {S¢(x) : x €
X, € > 0} is a basis for a topology on X. This topology is called the metric
topology or the topology induced by the metric d. In the future, whenever
we speak of a metric space we will assume that it has the metric topology
(unless we specifically impose a different topology).
(1.C.3) Theorem. If (X,d) is a metric space, then B = {S,(x) : © €
X, € > 0} is a basis for a topology on X.
Proof. By (1.A.6) it suffices to show: ifp € S.(z)N S,(y), then there
is a positive number v such that S,(p) C S.(z)N Sy(y). Let A1 = d(z,p)
and A2 = d(y,p). Then if v is any positive number less than both e — A, and
ps — Ag, it is easy to see that S,(p) is the required basis element containing
p.
We list some examples of metric spaces. The reader should verify that
each example satisfies properties (i), (ii), and (72) of (1.C.1).
(1.C.4) Examples.
1. (The usual metric for R'.) Let X be R’ and define d(z,y) = |x —y].
2. (The usual metric for R?.) Let X be R? and define d(z,y) =
(z1 — yi)? + (2 — y2)? where & = (21,22) and y = (yi, y2). Use the
Minkowski inequality, (0.F.2), to show that d satisfies the triangle inequality.
18 1. The Basic Constructs
du,o) =
Hence U EU.
Similarly, to show that V € U it will suffice to show that for a typical
basis element of the product topology, B = U, x U2 x --- x Un, and for
z = (%1,22,..-,2n) € B, there is an € > 0 such that x € S,(x) C B.
Hence, suppose that z € B. Then for each z;, there is an €; > 0 such that
{ye €1 : jn; —y| < ex} CU;. Let e = min{e, : i =1,2,...,n}. We show
thatsSe@ec Uy x Ua eX Uns Ve Sle 225 104 Fn) S.(z), then
n
\x5 = zi| < » (ei = za)? <<16R<G,
11
Therefore z; € U; foreachi. O
(1.D.11) Definition. For each i = 1,2,...,n define the ith projection map,
Der Meee Aa x Xn) ea) = ae
piesc
> Xi DY Pi(@ig
topology and Y CU is any topology on the set Piet X; for which each p; is
continuous, then V =U.
Proof. Suppose U; is an open set in X;. Then p; (Ui) =A Mg eX
U; x Xia1 X +++ X Xn is clearly a member of U, and therefore p; is continuous
(1.4.4).
Suppose now VY C U and each p; is continuous with respect to V. To
demonstrate that V = U it suffices to show that each basis set U = U; x
Uy x --- x Uy, in U belongs to V (why?). Since p; is continuous for each 2, it
follows that p~!(U;) is in V. But U = U, x Uz x --- x Un = (V1,
|(Ui),
andhenceUeEV. QO
(1.D.13) Exercise. Suppose for 1 = 1,2,...,n, fj : X; — Yj; is continuous.
Show that the map [J fi : []j_, Xi: > []jL, Yi defined by
(1.D.14) Exercise. Show that the family U/ defined above forms a topology.
The topological space obtained in the manner just described is custom-
arily called the free union of the spaces X,, and the corresponding topology
is frequently referred to as the disjoint union topology. Requiring that the
spaces in question be pairwise disjoint can be avoided, since any collection of
sets may be replaced by a disjoint collection of sets as follows. Suppose that
{X, : a € A} is a family of sets. For each a € A, define X, = Xq x {a}.
Then clearly for a ~ B, XgN Xz = . Now suppose {(Xq,Ua) : a € A} is
a collection of topological spaces. If a € A, let Xq = Xq x {a} and define
a topology Ug for Xq by declaring a subset Ug x {a} C Xq x {a} open if
and only if U, is open in Xqg. Then the free union of the X, is defined to
be the free union of the disjoint spaces (oxhae) described above.
Although the concepts introduced in this section are relatively simple, the
reader should give them careful consideration. A firm grasp of these ideas is
needed before one can successfully proceed to the more intriguing aspects of
topology.
Closed Sets
Proof. (i) Trivial. (27) Recall that the intersection of closed sets is closed.
(iii) Since A is contained in each set of the intersection, and A (being closed)
is one of those sets, then A C A. On the other hand, A is always contained
in A. (iv) Since B is closed, we have A C B. (v) Trivial. (vi) Since the
finite union of closed sets is closed, we have by (iv) that AUB C AUB. On
the other hand, A C (AUB), and hence by (v), A C AUB. Similarly, B is
contained in AU B and therefore AUB C AUB. (vii) Since ANBCA
and ANB C B it follows from (v) that ANB CANB. a
gg So SO May Mn ee Ae
(1.E.9) Exercise. Show that a space X is T, if and only if for each pair
of distinct points z,y € X, there is an open set containing xz that does not
contain y and an open set containing y that does not contain z.
A Potpourri of Fundamental Concepts 25
Accumulation Points
Although any of the three sets int A, ext A, and Fr A may be empty, it
should be obvious that int A is the largest open set contained in A and ext A
is the largest open set contained in X \ A.
F. CONTINUITY
In this section and its problem set we examine the concept of continuity in
light of our newly developed topological ideas. It is reasonable to suppose
that continuity can be expressed in terms of each concept that can be used
to determine a topology. We first establish a few useful generalizations of
some familiar results.
(1.F.4) Theorem. Suppose (X,U) and (Y,V) are topological spaces and
f:X —-Y. Then the following conditions are equivalent:
(i) f is continuous;
(ii) for each V € V, f~1(V) is open in X;
(iit) for each closed set D C Y, f~!(D) is closed in X.
Proof.
(i) = (it) This is just (1.A.4).
(it) = (tit) Suppose D is closed in Y and let
V = Y \ D. Then V is
open, and hence f~!(V) is open. But f~!(D) = X \ f~1(V), which proves
that f—!(D) is closed in X.
(iit) = (ti) Suppose that V € V. Then Y \ V is closed, and hence
=1(Y \ V) is closed in X. But f-(Y \V) = X \ f-*(V) and therefore
f-(V) isopenin X. O
We next give a criterion for continuity that is expressed in terms of
the closure operator. Similar results for the interior operator, the derivation
operator, and the frontier operator can be found in the problems.
(1.F.5) Theorem. Suppose (X,U) and (Y, V) are topological spaces and
f:X —Y. Then f is continuous if and only if f(A) C f(A) for each subset
A of X.
Proof. Suppose f is continuous, and A is a subset of X. Since A C
f-1(f(A)) C f7!(f(A)) and the latter set is closed by (ii) above, we obtain
Ac f71(f(A)). It follows that f(A) c f(A).
Conversely, suppose B is a closed set in Y. Consider A == f-\(B ).
By hypothesis [Male )) ¢ f(f-(B)) c B = B. Hence, ee
Hh e
f-f(f-1(B)) C f-!(B), so f-!(B) is closed. O
The next result will see continual service during succeeding chapters.
Continuity 29
F(z) ife@eA
Maye
g(a) if pe B
is continuous.
f(x). [Note: It is easy to see that this theorem also holds for first countable
spaces X and Y.]
Proof. Suppose that f is continuous at xz, and let V be any open set
containing f(z). Since f is continuous, there is a neighborhood U of x such
that f(U) C V. The convergence of {z;} implies that there is some integer
N such that z; € U whenever i > N. But then for each i > N we have that
f (zi) € V and hence {f(x;)} converges to f(z).
To prove the converse, suppose f is not continuous at « € X. Then
there must be an « > O such that no neighborhood of x is mapped by f
into S.(f(x)). For each positive integer n select a point tp € Sj/n(z) for
which f(rn) ¢ S-(f(x)). Clearly, the sequence {z,} converges to z, but the
corresponding sequence {f(xrp)} fails to converge to f(z). O
The preceding theorem gives some rigor to the concept of continuity
as often taught in elementary calculus: a function is continuous at a point
xz € €! if and only if whenever points in €! approach z, then their images
under f approach f(z). Note that half of the theorem holds for arbitrary
topological spaces. The other half does not hold in general. The concept
of net, defined in Chapter 11, was devised in order to obtain an analogous
theorem that holds for arbitrary topological spaces.
(1.F.11) Exercise. Let (X,d) be a metric space. Show that the metric
function d: X x X — [0,00) is continuous.
(1.F.12) Exercise. Suppose (X,d) is a metric space, and A C X. Show
that the function V : X — [0,00), defined by U(x) = d({x}, A), is continu-
ous.
G. HOMEOMORPHISMS
Let f : X — Y be a continuous function from one topological space onto
another. In spite of the continuity of f, the space X may differ considerably
from Y: a flagrant example of this occurs when X is arbitrary and Y consists
of a single point. Not only may Y be a good deal simpler than X, but,
surprisingly, as we shall see later, the reverse situation can happen as well.
In Chapter 9, a continuous function is constructed that maps the unit interval
I = [0,1] onto the unit square J x I. In fact, maps can be found that carry
I onto cubes, as well as onto a host of far more exotic spaces.
Homeomorphisms Sil
In these cases, the continuous function in question will not be 1-1. Nev-
ertheless, even with this further imposition, the spaces X and Y may still
have little similarity. To see this, suppose X is an infinite set with the discrete
topology and Y is any space with the same cardinality as X. Then every
1-1, onto map from X to Y is continuous, even though X has uncountably
many open sets and Y could have as few as two.
Suppose now, however, that f : X — Y is a continuous bijection such
that f—! (which is a function from Y to X since f is a bijection) is also
continuous. The reader should verify that not only is there a 1-1 correspon-
dence between the points of X and Y, but also between the open sets of the
respective topologies. Thus, the topological structure of X has an almost
exact counterpart in Y. Bijective, continuous mappings with continuous in-
verses are called homeomorphisms, and the corresponding spaces are said to
be homeomorphic or topologically equivalent. An embedding of X into Y isa
continuous 1-1 map f : X > Y such that f-! : f(X) > X is continuous. At
least a germinal feel for topology should arise from the following discussion
of homeomorphisms.
(1.G.1) Definition. A function f : X — Y is open (respectively, closed) if
and only if f maps open (respectively, closed) sets onto open (respectively,
closed) sets.
(1.G.2) Exercise. Show that homeomorphisms are both open and closed.
(1.G.3) Exercise. Show that a continuous open (or closed) bijection is a
homeomorphism.
(1.G.4) Exercise. (a) Show that if X; x--- x Xn has the product topology,
then the projection maps, p; are open, but need not be closed. (b) Show
that i: [0,1] + €? defined by i(x) = (x,0) is an embedding.
(1.G.5) Exercise. Show that two metrics d and d for a set X are equivalent
if and only if the identity map id: (X,d) — (X,d) is a homeomorphism.
For a topologist, homeomorphic spaces are essentially identical. Thus
little distinction (if any) is made between the following pair of spaces (con-
sidered as subspaces of €7), since one can be mapped homeomorphically onto
the other.
32 1. The Basic Constructs
In a like vein, since a doughnut can be transformed into a coffee cup via
a homeomorphism (see figure), a topologist has classically been defined as
an otherwise enlightened individual who fails to note any difference between
these two objects.
oe ee
(1.G.6) Exercise. Show that (—1,1) and €! are homeomorphic via the
map h : €} — (1,1) defined by h(x) = x/(1 + |z)).
The problem of establishing the existence or nonexistence of a home-
omorphism between two topological spaces is by no means a trivial one.
Readers might enjoy discovering this for themselves by attempting to show
that €1 and €? are not homeomorphic. In order to attack this problem
in general, we shall investigate in the ensuing chapters certain topological
invariants, i.e., properties of topological spaces that are preserved by home-
omorphisms. If a space X possesses a certain property that is topologically
invariant, and the space Y does not, then clearly X and Y cannot be home-
omorphic. A couple of obvious invariants (which are not particularly useful)
are the cardinality of the sets in question and the cardinality of their re-
spective topologies. Note that neither of those invariants helps in deciding
whether or not €! and €? are homeomorphic.
Metrizability is a prime example of a topological invariant.
(b) 51 (0);
(c) S2(0)
wo
Points x in X other that (0,—1) and (0,1) are given neighborhood bases
consisting of (small) open intervals centered at z. Base neighborhoods of
(0,1) are of the form {(z,y) : 6 < 2 < Oandy =O}U{(ey) 7 0< 2 <
a and y = 1}, and base neighborhoods of (0,—1) are of the form {(z,y) :
b<x<Oandy=O0}U{(z,y) : 0<2<aand y= -—1}. Show that X with
the resulting topology is separable, 7), first countable, and locally Euclidean
(ie., show that each point has a neighborhood homeomorphic to €'), but
that X is not T> and, hence, is not a 1-manifold.
PROBLEMS
Section A
6. Suppose U; and U2 are topologies for a set X and id: (X,U;) > (X,U2)
is the identity map. Show that id is continuous if and only if Uy Cc U4.
7. Suppose R! has the usual topology. Show that a function f:R'3=R})
is continuous if and only if f~*(—oo,a) and f~1(a, 00) are open for each
aé Rl.
8. Show that a subset A of R’ is open in the usual topology if and only if A
can be written as the countable union of pairwise disjoint open intervals
(intervals of the form (—o0o,a), (6,00), (—00, 00) are allowed).
9. Suppose (X,U/) and (Y, V) are topological spaces. A function f : X 3 Y
is open if and only if f(U) € V whenever U € U. Find an open function
that is not continuous.
10. Find an example of bases B and B for topological spaces (X,U) and
(Y, V), and a continuous function f : X > Y such that there is a basis
element B € B for which f~1(B) ¢ B.
Section B
linear ordering and determine if X (with the order topology) is T2, first
countable, or second countable.
8. For each x € Z and eachn EN, let B® = {y : y=rn+a, r € Z}.
Show that B= {B? : c € Z,n € N} is a basis for a topology on Z.
9. Suppose X is first countable (respectively, second countable), Y is a
topological space, and f : X — Y is continuous, open (i.e., f takes open
sets to open sets), and onto. Show Y is first countable (respectively,
second countable).
10. True or false? Suppose X is second countable. Then every nest of
distinct open sets in X is countable. |
Section C
te Let X = R?, and for points z = (21,22) and y = (y;, y2) define
Show that d is a metric and that the sets shown in the figure possess
different “area” if d is used to measure the length of sides.
ee
Describe the topology induced by this metric.
Suppose V is a vector space over R!. A function ¢ : V — [0,00) is
called a norm for V if and only if for all x,y € V we have
(i) (x) = 0 if and only if c = 0,
(ii) o(a +) < d(x) + 419),
(iti) d(ax) = |al(xz) where a is a real number.
The pair (V,¢) is called a normed vector space and ¢(z) is frequently
denoted by ||z||.
Suppose (V,@) is a normed vector space and d : V x V — (0,00) is
defined by d(z,y) = ||z — y||. Show that d is a metric for V.
French railway metric. Let X = R?. Suppose z and y are points in X.
If x and y are on a line passing through the origin, define their distance
to be the usual Euclidean distance. If x and y do not lie on such a line,
then define their distance as in the post office metric (problem 5). Show
that the French railway metric is a metric and (informally) describe the
induced topology.
Section D
Suppose (Xj, di), (X2,d2),..., (Xn, dn) are metric spaces, and suppose
Y =X, x X_ X--: x Xn. Define
p: Y x Y — (0, 00) by setting
iis
Xy = Xj x X92 is X29
Yi q1 Vi x Y5 q2 Y>
Suppose X,, X2,..., Xn and Z are topological spaces and Y = |]j_, Xi.
Let g: Z > Y be a function. Show that g is continuous if and only if
pig is continuous for each projection map p;.
(a) Is the free union of a family of first countable spaces first countable?
(b) Is the free union of a family of second countable spaces second
countable?
10. Show that the free union of metrizable spaces is metrizable.
dil. Suppose X and Y are topological spaces and V C X x Y is open. Show
that V[z] = {y : (z,y) € V} and V[y] = {x : (z,y) € V} are open in
Y and X respectively. Is the converse true: if AC X x Y and A[z] and
Aly] are open for each z € X and y € Y, must A be open in X x Y?
12. Are there any open sets in the product topology of X x Y that are not
of the form U x V where U is open in X and V is open in Y?
Problems 41
Section E
(1) c(0) = 9;
(it) AC c(A) for each A € P(X);
(iit) e(c(A)) = c(A) for each A € P(X);
(iv) c(AU B) =c(A) Uc(B) for each A and B in P(X).
Let U = {X\c(A) : A€ P(X)}. Show that U is a topology for X with
the property that for each A € P(X), A =c(A). A function with the
properties of c is called a closure operator.
23. Let X be a set and I: P(X) + P(X) have the properties that:
(i) I(X) =;
(it) I(I(A)) =I(A) for each A € P(X);
(iit) I(A) C A for each A € P(X);
(iv) I(AN B) =1(A)NI(B) for each A and B in P(X).
Let U = {I(A) : A € P(X)}. Show that UW is a topology for X, and
that for each A € P(X), J(A) = int A. A function with the properties
of I is called an interior operator.
24. Formulate and prove results similar to those in problems 22 and 23 above
for
(a) a frontier operator, and
(b) a derivation operator.
25. Prove that in a T; space, a point x is an accumulation point of a set A
if and only if every neighborhood of z contains infinitely many points
of A.
26. Suppose A and B are disjoint closed subsets of a metric space. Show that
there are disjoint open sets U and V containing A and B respectively.
7 Suppose A is an open subset of a topological space X. Prove or disprove
that int(A) = A.
Problems 43
Section F
Section G
4, Fill in the blank: Let id : (X,U) > (X,U'). Then id is open if and only
if U Pee
Suppose X and X are homeomorphic and Y and Y are homeomorphic.
Show that X x Y and X x Y are homeomorphic.
Find a mapping: (a) that is open but not closed; (b) that is closed but
not open; (c) that is continuous but neither open nor closed.
Let A be the set of irrationals in €' with the relative topology. Is A
separable?
Suppose A is a dense subset of X and G is a dense open subset of X.
Show that ANG is dense in X.
Suppose A is dense in X and B is dense in Y. Show that Ax B is dense
in An ey’,
10. A map f : X > Y isa local homeomorphism if and only if for each point
x € X, there is an open set U containing x that is mapped homeomorphi-
cally by f onto an open subset of Y. Show that local homeomorphisms
are continuous and open.
11. Prove the following converse of (1.G.12). Suppose Y has the property
that for any space X, any dense subset D of X, and any two continuous
functions f,g: X — Y which agree on D, then f = g. Then Y is T».
12. A metric space X is totally bounded if and only if for each « > 0, there
dEC pOints 2 4i05,.05ee ouch that HX =A) SAe,)) tednsy 1 2509.57:
Show that a totally bounded metric space is separable.
13. Find an example of a nonseparable metric space that does not have the
discrete topology.
14. Suppose (X,d) is a separable metric space and that A C X. Show that
(A, d|,) is separable. In general, is it true that a subspace of a separable
space is separable?
15. Suppose R! has the half-open interval topology U, and A = (0,1) is
given the relative topology (with respect to U). Show that R! and A
are homeomorphic.
16. True or False: If U is an open subset of X and h: U > X is an
embedding, then h(U) is open in X.
Li. Assume the statement in problem 16 is true whenever X = €”, and
prove that it is true for n-manifolds.
18. A subset A of a space X is semiopen if and only if there is an open set
U in X such that U CAC U.
(a) Show that A is semiopen if and only if A = AND for each dense
subset D of X.
46 1. The Basic Constructs
CONNECTEDNESS AND
COMPACTNESS
47
48 2. Connectedness and Compactness
Disconnected Spaces.
1. (0,1) U (3, 4) (with the relative topology from &').
2. (0,1) U (1, 4] (with the relative topology from €').
3. The rational numbers (with the relative topology).
4. Any set (containing two or more points) with the discrete topology.
Although it seems quite reasonable from a visual standpoint, the “cri-
terion” that a space is connected if it consists of solely “one piece” is often
impossible to apply in practice. For instance, while it seems that €? is con-
nected, what about €? \ {(z,y) : x,y € Q}? Is this subspace in “one piece”
in spite of all the holes? In €! no such complications arise, as we see from
the following theorem. But first we recall that an interval is a subset Z of R
with the property: ifz,y,z€ R,z,ye€TZ,andr<z<y, then ze TZ.
(2.A.11) Exercises.
(a) Why is €' connected?
(b) Suppose that {Aq : a € A} is a collection of connected subspaces
of a space X, and that B C X is also connected. Show that if
BNA, # 9 for each a € A, then BU(U{Aq : a € A}) is connected.
The following theorem, quite useful in spite of its trivial proof, gives
a convenient characterization of connectedness. For the remainder of this
chapter we let S denote the set {0,1} with the discrete topology.
(2.A.12) Theorem. A space X is connected if and only if there is no
continuous function f : X — S which is onto.
Proof. If f : X -> S is onto, then the sets f~'(0) and f~'(1) form
a separation of X. On the other hand, if X is not connected there is a
separation (U,V) of X. Define f : X — S by f(x) = 0 if e € U and
f(x) =1ifaeéV. Then f is clearly a continuous map of X ontoS. O
Theorems involving the fixed point property are not only aesthetically
pleasing in themselves, but also have wide ranging applications to other areas
of mathematics. Our first fixed point result is a special case of the Brouwer
fixed point theorem, which states that the space obtained by taking the
Connectedness: Slightly Deeper Results 51
product of J = [0,1] with itself a finite number of times has the fixed point
property.
(2.B.7) Exercises.
(a) Show that the fixed point property is a topological invariant. That
is, show that if h : X —+ Y is a homeomorphism and X has the fixed point
property, then so does Y.
(b) Show that [0,1] and S' are not homeomorphic.
The topological invariant, connectedness, may be exploited to show that
E! is not homeomorphic with €?, even though both spaces are connected.
(2.B.8) Theorem. The space €! is not homeomorphic to €?.
Proof. Suppose a homeomorphism h : €? + &! exists. Let c € €? and
note that E? \ {c} is connected by (2.B.6). However h(E? \ {c}) = €1 — {h(c)}
is not connected, which contradicts (2.4.14). O
(2.B.9) Exercise. Show that (0,1) does not have the fixed point property,
and hence (0,1) is not homeomorphic with (0, 1].
(2.B.10) Exercise. Use the topological invariant, connectedness, to give
another proof that (0,1) and [0, 1] are not homeomorphic.
We conclude this section with a rather strange, but intuitive, result that
will see frequent use in succeeding chapters.
C. PATH CONNECTEDNESS
In this section we introduce a different concept of connectedness—one that
will find repeated applications in later chapters. A layman’s notion of con-
nectedness would probably center on the idea that a space X is connected if
one can move from one point in X to another without ever leaving X. This
idea is conceptualized mathematically by the following two definitions.
f (2t) if0<t<$;
A(t) =
gat=1) fe sre.
That h is continuous follows from the map gluing theorem (1.F.6), as well
as (1.F.3) and (1.F.1). Clearly h(0)=aandh(1)=c. O
(2.C.4) Exercises.
(a) Show that if there is a path from z to y, then there is a path from
y to x.
Although path connected spaces are connected, the next example shows
there are connected spaces that are not path connected.
D. COMPONENTS
We can view spaces that are not connected as consisting of a (possibly infi-
nite) number of connected pieces. This leads to the following notion.
(2.D.1) Definition. Suppose X is a topological space and A is a subspace
of X. Then A is a component of X if and only if
(i) A is connected, and
(ii) if B is a connected subspace of X containing A, then B = A.
In other words, components are maximal connected subspaces.
56 2. Connectedness and Compactness
(2.D.3) Exercise. Find the components of the rational numbers with the
relative topology. Do the same for the irrational numbers.
Parts (17) and (277) of the above theorem can also be expressed as saying
that the components form a partition of X.
While it is clear that a space with the discrete topology is totally dis-
connected, there are other examples as well. By (2.D.3) the set of rationals
(or irrationals) with the relative topology is a totally disconnected space.
Locally Connected Spaces 57.
(2.D.8) Exercise. Show that for open subsets of €”, path components and
components coincide. Is the modifier “open” necessary?
Ai
The components of X = Ao U (UP2., An) are precisely the sets Ao, Ai,....
We know from (2.D.4) that Ao must be closed. Since it is clear that Apo is
not open in X we see that components need not be open. (Another example
of this behavior was given in (2.D.3).) For certain types of spaces, how-
ever, components will be both open and closed. These are known as locally
connected spaces, and they include the n-manifolds. The formal definition
of such spaces is couched in somewhat different terms (emphasizing the lo-
cal nature of the concept), but the fact that the two notions are essentially
equivalent is readily established in (2.E.2).
(2.E.1) Definition. A space X is locally connected if and only if for each
point z € X and each open set U containing 2, there is a connected open
set V such that s € V CU.
Note that the definition above is equivalent to saying that there is a
neighborhood base at each point consisting of connected open sets. It is
clear that local connectedness is a topological property.
58 2. Connectedness and Compactness
F. SIMPLE CHAINS
We introduced paths as a means of mathematically traveling from one point
to another in a topological space. A geometrically pleasing concept, vaguely
akin to path, is that of a chain.
It should be observed that the links of a simple chain need not be con-
nected; thus the accompanying figure illustrates a perfectly acceptable simple
chain. This makes the characterization of connectedness given in problem
F’.1 seem even more spectacular.
CS pres Re
2 : ees A2
A3
Frequently, simple chains are forged from links that are open sets; a
common procedure for their construction consists of starting with an arbi-
trary entanglement of open sets and then extracting a suitable simple chain
from the confusion. That this may be done is a result of the following elegant
and useful theorem.
G. COMPACTNESS
The notion of compactness, to which we now turn, is the generalization of
a second property of some intervals—closed and bounded—promised in the
introduction to the chapter. In calculus, great use is made of the fact that a
continuous function defined on a closed interval is bounded, and attains it’s
maximum and minimum on the interval. On the other hand some functions,
such as f(z) = 1/z, defined on (0,1] fail to be bounded. The property
of being a closed set is well defined in a general topological space, but the
concept of boundedness is tied to the notion of a metric. Hence in order
to generalize, we must couch the definition in terms other than closed and
bounded. The astute reader will notice that definition (2.G.2) draws on the
Heine-Borel Theorem of analysis for inspiration.
The notion of cover provides topologists with one of their most successful
tools. Covers play a prominent role in the study of such diverse areas as
metrizability, dimension theory, and Cech homology, in addition to giving us
a means for defining compactness.
Any finite space is compact, no matter what its topology and, indeed,
any space with only a finite number of open sets is compact.
(2.G.4) Example. Suppose X = R! with the open ray topology, and let
A = [0,1]. Then A is a compact subset of X, but A is not closed (in the
open ray topology).
{ifs (Oda> 1 =A; 2).25,n} fromayial (Omen res CP wiSincer fais "ontoyrthe
collection {O1,O2,...,On} is a finite subcover of Y. O
(2.G.10) Corollary. Compactness is a topological (and a continuous) in-
variant. O
The following exercise is an important result. Among other things, it
says: many 1-1, onto, continuous functions are automatically homeomor-
phisms.
The next theorem, useful in itself, also has a proof that is typical of
many theorems having compactness as a hypothesis.
sets {V* : a € A} each of which contains B. Note that U?xV? C W for each
a € A. Once more, we use compactness (this time of A) to obtain a finite
subcover {U%!,U%,...,U°™} of A. Set U = Wi US and V1) eV" to
complete the proof. O
(2.H.4) Exercise.
(a) Let A be a partially ordered set with the property that if a, 6 € A,
then there is a \ € A suchthat A > a and A > B. Suppose that {A, : a € A}
is a collection of compact subsets of a T> space such that Ag C Ag if and
only if a > B. Show that if {Aa : a € A} C U, where U is open, then
there is a yp € A such that A, C U for each a > p.
(b) Suppose that {A, : a € A} is a family of closed compact sets such
that (\{Aq : a € A} is a subset of an open set U. Show that there is a finite
set {Q1,Q2,...,Qn} C A such that (\_, A; CU.
(c) Suppose in (2.H.3) one also assumes that each A; # 0. Show that
(d) Can one make a similar improvement in parts (a) and (b)?
Our second characterization of compactness, based on nests of closed
subsets, is somewhat subtler than its predecessor. Some equivalent of the
axiom of choice is apparently needed for its proof.
(2.1.4) Exercise. Suppose that X1, X2,...,Xn are locally compact spaces.
Show that [];"_, X; is locally compact.
(2.1.5) Exercise. Show that local compactness is a topological invariant.
Is it a continuous invariant?
(2.1.6) Definition. Suppose that f : Zt — X is a sequence. A sequence
g: Z* + X is a subsequence of f if and only if there is a strictly increasing
function h: Z+ — Z+ such that fh = g. We denote f(n) by zp and h(2) by
n;. The subsequence g is defined by g(t) = rn, and is generally denoted by
{Zn,}-
(2.1.7) Definition. A topological space X is sequentially compact if and
only if every (infinite) sequence in X has a convergent subsequence.
In the next chapter, we show that the notions of compactness and se-
quential compactness coincide in metric spaces. However, in general there
66 2. Connectedness and Compactness
Note that it follows from (1.G.11) that separable metric spaces are Lin-
delof.
(2.1.12) Exercise. Show that a metric space X is Lindel6f if and only if X
is second countable. [Hint: For each positive integer n, cover X with sets of
the form S,/,(z).]
One Point Compactifications 67
; Ey x y a? +y?
(c= l+a?+y?’ l+a2?+y?? 14+a?4+y?
(x,y)
PROBLEMS
Sections A and B
(ie AG GD
(ii) ANC#6, AND, and
(iti) (CND)U(CND)=6
Problems 69
Section C
1. Show that path connectivity is a topological invariant (in fact, a contin-
uous invariant).
2. Suppose Xj, X2,...,X, are topological spaces. Show that likes X; is
path connected if and only if each X; is path connected.
3. Prove or disprove: If A is a path connected subset of X, then A is path
connected.
4. Prove or disprove: If A and B are path connected subsets of a space X
and ANB # G, then AU B is path connected.
5. <A topological space X is said to be contractible if and only if there is
a continuous map H : X x I > X and a point z* € X such that
H(a,0) = weand H(#,1) = a* for each € xX. Prove-thatiévery
contractible space is path connected.
Section D
1. Define an equivalence relation on a space X by setting x ~ y if and only
if for each separation (U,V) of X, x and y are both in U or both in V.
(a) Show that ~ is an equivalence relation. The equivalence classes are
called quasicomponents.
(b) Show that quasicomponents are closed.
(c) Show that each component of X is contained in some quasicompo-
nent.
(d) Describe the quasicomponents of the following subset of €?:
Section E
TS Show that open subsets of locally connected spaces are locally connected.
Is the same true for arbitrary subspaces of locally connected spaces?
Suppose X = AUB, where A and B are locally connected. (a) Must X
be locally connected? (b) If A and B are also closed, must X be locally
connected?
Show that if x and y are points belonging to different components of a
locally connected topological space X, then there is a separation (U, V)
of X such that r€ U andy eV.
Does the conclusion of problem 3 hold if the hypothesis “locally con-
nected” is dropped?
Show that in a locally connected T> space, components and quasicom-
ponents coincide.
Prove or disprove: If a space X is locally path connected, then X is
locally connected. What about the converse?
Suppose X is connected and locally connected and f : X — I is contin-
uous and bijective. Show that X and J are homeomorphic.
Section F
I Show that a space X is connected if and only if for each x,y € X and
for each family U of open sets whose union is X, UY contains a simple
chain from z to y. [Warning: You may not assume that the links are
connected]]
Show that every two points in a connected n-manifold can be joined by
a simple chain whose links are homeomorphic to €”.
Suppose X is a locally connected T; space and p, x and y are points
in X. Show that x and y are in distinct components of X \ {p} if and
only if every simple chain of connected open sets from zx to y has a link
containing p.
A metric space K is chainable if and only if for each € > 0, there is a
simple chain of open sets covering K such that each link has diameter
less than e. Are the following spaces chainable?
Problems 73
(a) I
(b)" lx T
(c) The unit circle
(d) The topologist’s sine curve
Section G and H
Section |
18. Show that a TJ; space X is countably compact if and only if every infinite
open cover of X has a proper subcover.
Section J
1. Let X be the rationals with the relative topology and let X* denote the
one point compactification of X. Show directly that X* is not T>.
2. To what is the one point compactification of (0, 1] homeomorphic?
3. Show that the one point compactification of Zt U {0} is homeomorphic
toOLOLU Tn = neta}. ;
4. Suppose X is a locally compact T> space and f : X + €' is continuous.
Show that there is a continuous function f : X* — €! such that f(z) =
f(z) for each x € X if and only if for each € > 0, there is a compact
subset K, of X such that |f(x) — f(y)| < € whenever x,y ¢ K..
5. Suppose X is a Ty space and X™* is first countable. Is X necessarily
locally compact?
6. Let X = (0,1) U (1,2) U (2,3) U (9,10) be given the relative topology
(with respect to €!). Describe the one point compactification of X.
Chapter 3
METRIC SPACES
To consolidate and extend some of the ideas introduced in the previous two
chapters, we now focus our attention on metric spaces. We considera variety
of metric spaces beginning with those that are compact.
The following example, the Porcupine space, shows that in a more gen-
eral setting it is possible for a sequence to have a cluster point that is not
the limit of any subsequence.
(3.4.3) Example and Exercise. Let ¢;,¢2,... be a sequence of disjoint
closed rays in the plane emanating from o = (0,0). Let X = U7, 4. It is
fet
78 3. Metric Spaces
convenient to adopt the following notation: If a,b € £; and d(o,a) < d(o,b),
where d is the usual metric for €?, we write a = b and define
Nu=A(a,
0) be: O <<G-—.0).
(3.4.18) Theorem. (Levine [1960]). Suppose (X,d) and (Y,p) are metric
spaces and f : X + Y is a continuous function. Then there is a metric d*
on X, equivalent to d, that makes f uniformly continuous.
82 3. Metric Spaces
(3.B.2) Exercises.
(a) Give an example of a metric space that. is not complete.
(b) Show that completeness is not a topological property. [See the next
theorem.]
(c) Show that completeness is preserved by homeomorphisms that are
also isometries. [If (X,d) and (Y,p) are metric spaces, a function
f : X — ¥ is an isometry if and only if d(z,y) = p(f(z), f(y)) for
all z,y € X.]
Complete Metric Spaces 83
Proof. For each positive integer i select a point 2; € A;. Since the
diameters of the A;’s converge to 0 and the A;’s are nested, it is clear that
{x;} is a Cauchy sequence and hence converges to a point z € X. We
show that ()$2, Ai = {x}. If ¢ ¢ ()72, Ai, let A; be the set with smallest
index that fails to contain z. By (3.A.14), d(z,A;) =r > 0. However, this
is impossible, since it implies d(z,A;) > r for 1 > j, which prevents the
sequence from converging to x. Therefore, we have x € (3 A;. That z is
the only point in the intersection follows readily from the observation that
if y is any other point in X, then d(z,y) = s > 0; however, for sufficiently
large i, diam A;<s. O
(3.B.9) Corollary. Suppose (X,d) is a complete metric space and let
D,,D2,... be a sequence of open dense subsets of X. Then (}%<, D; is
dense in X.
Proof. Suppose U is a nonempty open set in X. It suffices to show
that UN (lex D;) # 9. Since D; is open and dense, D; NU must be
nonempty and open. Hence, there is a positive number €; < 1 and a point
xz, € X such that S.,(z1) C Dj NU. The open set S,,(r1) has nonempty
intersection with D2, and consequently there is a point x2 and a positive
number €2 < 1/2 such that S,,(%2) C D2 N S_,(1). Similarly, S,,(%2) N D3
is open and nonempty, and therefore there is an e3 < 1/3, etc. Repeated
application of this procedure leads to a decreasing sequence of closed sets
with diameters converging to 0. By the previous theorem, there is a point z
common to each S,,(x;). Hence, x is contained in both U and (2, Di. O
Spaces other than complete metric spaces may also have the property
described in the foregoing corollary; such spaces are known as Baire spaces.
This use of the word “category” should not be confused with category
used in the context of functors and categories (see Chapter 12).
(3.B.15) Examples.
1. The rationals are of the first category in €!.
2. The irrationals are of the second category in €}.
If the irrationals could be written as a countable union of nowhere dense
sets, then with the addition of the individual rational points as nowhere dense
subsets, €! itself could be represented as a countable union of nowhere dense
subsets and would be of the first category (in itself). This contradicts the
following classical theorem.
(3.B.16) Theorem (Baire Category Theorem). Any complete metric
space is second category in itself. That is, a complete metric space can not
be written as a countable union of nowhere dense sets.
Proof. Suppose X is a complete metric space and X is of the first
category. Then X can be written as a countable union of nowhere dense
sets A,, Ag,. ... Since each A; is nowhere dense, X ‘ A; is both open and
dense. From the completeness of X, we have by (3.B.9) that ();,(X \ Ai) is
nonempty. But this contradicts (\f2,(X \ Ai) = X \ (Uf, 4i) = 0. Hence
it follows that X must be of the second category. O
(3.B.17) Definition. A point x of a space X is isolated if and only if x has
a neighborhood that contains no other point of X, i.e., {x} is open in X.
(3.B.18) Corollary. Every countable complete metric space has an isolated
point. O
Note that every point of a discrete topological space is isolated. At the
other end of the spectrum are the perfect spaces.
If X is not connected, then there are open sets U and V in €2 such that
X CUUV and XNUNV = 9. We assume thatp € U. For each t € (0, 1/2),
let Ky = {c € K : L(c) contains a boundary point (x,t) of U}. We claim
that for each ¢, Ky; is nowhere dense in K. Suppose not; then there are
points c1,c2 € K such that 0 4 (c1,c2)NK C Ky. Lete € EN (c1, C2).
Then the boundary point (z-, t) of U has an irrational y coordinate, i.e., t is
irrational. (If t were rational then (z,,t) would be in X and hence in either
U or V, both alternatives being impossible.) Now suppose t € F'N (ci, c2);
we conclude in a similar way that t is rational. Hence, K; must be nowhere
dense in K.
Let Ko = {c € K : L(c) has no boundary points in U}. Since p € U,
we have that for any c € Ko, L(c) is contained in U, and consequently,
Ko # K (otherwise, VM X would be empty). Thus, there are points c; and
co in K such that (c1,c2)N KN Ky = 9. Therefore, one may assume that
Ko = 9, since otherwise we could replace the space X with a “new” X that
sits over a part of the interval (c1,c2) (the “new” X is a subset of X that
is homeomorphic to X). With this minor point out of the way, we observe
that K = (U{K; : tis rational}) U E, (why?), or, in other words, K is a
countable union of nowhere dense sets, which contradicts (3.B.16). Hence,
X must be connected.
It is easy to establish that the removal of point p totally disconnects X.
Each component of X \ {p} must be contained in L(c) for some c (any subset
of X that contains points in two L(c)’s can be separated by a line running
between the L(c)’s, and through the missing point p). But clearly the points
of X lying in each L(c) are totally disconnected, because either points with
rational second coordinates or points with irrational second coordinates are
missing. Thus components consist of individual points.
As another vivid illustration of the Baire Category Theorem in action,
we prove the following rather delicate and unexpected result.
(3.B.23) Theorem. Suppose (X,d) is a connected, locally connected, com-
plete metric space. Then X cannot be written as a countable union of
nonempty pairwise disjoint closed subsets.
Proof. Suppose the contrary, X = 72, Ci where the C;’s are nonempty,
closed and pairwise disjoint. Let D = X \ (Uj2, int Ci) = Uf2, FrCi. Then
D is a closed subset of X, and hence, by (3.B.5) and (3.B.16), D is second
category (in itself).
Claim. If U is an open set in X such that UN FrC; # ) for some 7, then
UnN(D\FrC;) # 0.
To establish the claim, first observe that by hypothesis we may assume
U is connected. Since U intersects FrC;, we have that UM (X \ Cj) is
nonempty, and hence, there is a positive integer 7 #7 such that UNC; # 0.
88 3. Metric Spaces
If U fails to intersect FrC;, then C; NU = (int Cj) NU, which implies that
UC; is both open and closed (in U), contradicting the connectedness of
U. Therefore, UN Fr C; # 0, and the claim is proven.
Since D is second category, not all of the FrC; can be nowhere dense
(in D). By the connectedness of X, Fr C; # 0 whenever C; # 0. Thus, there
is an open set V in X and an integer j such that 9 AVNDC PEC; = bY C,.
Therefore, we have that VN FrC; # @ and VN (D \ FrC;) = 6, which
contradicts the claim. O
h(C)
PROBLEMS
Section A
Show that a metric space is compact if and only if every infinite subset
has an accumulation point (cf. chapter 2, problem I-13).
Suppose X is a first countable space, {x;} is a sequence in X, and z is
a cluster point of {x;}. Show that z is a limit point of a subsequence of
{ai}:
12. Find a subspace of €! and sets A and B that satisfy the hypothesis of
(3.4.14) and for which there is no point b € B with d(A,b) = d(A, B).
Section B
Eb: Show that a metric space is compact if and only if it is bounded in every
equivalent metric.
12: Show that every subset of a first category set is first category.
13. Prove or disprove: every uncountable space is second category (in itself).
14. Show that if A is a nowhere dense subset of a topological space X, then
X \ A is dense in X. Is the converse true?
15. Suppose X and Y are metric spaces and that A C X and BC Y. Show
that if A x B is nowhere dense in X x Y, then either A is nowhere dense
in X or B is nowhere dense in Y.
16. Suppose (X,d) is a metric space and that A is a countable subset of X.
Show that A is of the first category in X if and only if A has no isolated
point in X.
17: Prove or disprove: every open subset of a second category space is of
the second category.
18. Show that the countable union of first category sets is of the first cate-
gory.
19. Show that every countably infinite complete metric space contains an
infinite number of isolated points.
20. The Cantor set may be used as follows to describe a map f : I > €! that
is open but not continuous. On each component C, of the complement
of the Cantor set in J, let fa : Cag > €! be any strictly increasing
continuous function that has 0 in its image. On the Cantor set itself,
define f to be identically 0. Show that f is open but not continuous.
a1, Show that if U is an open subset of a space X, then FrU is nowhere
dense.
22 Prove the following ‘converse’ of (3.B.8). Suppose (X,d) is a metric
space and the intersection of each decreasing sequence of nonempty
closed balls whose radii converge to zero is nonempty. Then (X,d)
is complete.
2o- Another subset of the plane having an explosion point may be con-
structed as follows. As before, start with the “cone” over the Cantor
set. Let S denote the family of all half-closed segments each connecting
a point of the Cantor set with p = (1/2,1/2) but excluding p, and let
@ : [0,2] + S be 1-1 and onto. Let A be the family of all compact
subsets of €% that intersect an uncountable number of members of S
(but miss p). Let w : [0,2] + A be 1-1 and onto (why is this possible’).
For each a € (0,2), let d(a) = Sq and ~(a) = Ag. Let ao be the
smallest element of [0,2) such that Ap N Sa, # @, and select a point
Ca, in this intersection. Let a; be the smallest element in [0,) (and
94 3. Metric Spaces
Section C
Show that the conclusion of (3.C.2) need not hold if A has empty inte-
rior.
Suppose p,q,r € (X,d). Then q is said to be between p and r if and only
if d(p,r) = d(p,q) + d(q,r). A subset A C (X,d) is linear if and only if
there is an isometry (into) ¥: A > &}.
(a) Suppose p,g,r € X. Show that {p,q,r} is linear if and only if one
of the points p, q, or r is between the other two.
(b) Give an example of a metric space consisting of four points that is
not linear but for which every proper subset is linear.
An arc a contained in (X,d) is a segment if and only if Ima is linear.
Suppose (X,d) is a strongly convex metric space without ramifications.
Let pg and pr be segments in X such that pq pr \ {p} 4 9. Show that
either pg C pr or pr C pg. [Here, pq denotes the segment with endpoints
p and q.]
If (X,d) is strongly convex and p,q,r € X, show that q is between p
and r if and only if q € pr. [See problem 3 for notation.]
Suppose Uj, U2,...,Un are open subsets ‘of a topological space X, and
define (Ui,...,Un) ={AC X : ACU, Uj and ANU; # O for each ¢}.
Show that sets of this form determine a basis for a topology for the set
V={ACX : Ais closed and A # 9}. This topology is called the
Vietoris topology.
Problems 95
6. Show that for a compact metric space X, the Vietoris topology (see
problem 5) and the topology generated by the Hausdorff metric coincide
on the set H = {C : C compact in X}.
Suppose f : X — Y is continuous, where X and Y are metric spaces
and Y is also compact and connected. Let Hx = {A Cc X;AF
Q and is connected and compact}, and suppose Hy is defined similarly.
For each A € Hx, let f(A) = {f(a) : a € A}. Show that:
(a) f : Hx > Hy defined by A 4 f(A) is a well defined function
(called the induced map).
(b) f is continuous
(c) f need not be onto even if f is onto.
The function f is said to be weakly confluent if and only if for each
B € Hy there is a component A of f~!(B) such that f(A) = B.
(d) Assume that f is onto. Then f is onto if and only if f is weakly
confluent.
Let X = IJ, and let H be the family of all subsets of J that consist of one
or two points. Show that H with the Hausdorff metric is homeomorphic
to a (solid) triangle.
Let X = I and let H be the family of all subsets of J that consist of
one, two, or three points. Show that H with the Hausdorff metric is
homeomorphic to a tetrahedron.
10. Let X be a metric space and H be the family of all compact subsets
of X with the Hausdorff metric p. Suppose that C),C2,C3 € H and
Ci; C C2. Show that p(C3, C3 U C;) = p(Cs, C2).
0 ont Py Che
a le Get
(A.o2(
ent —Siang
aft 1 Oh Shae tres
ie eed sp 6, ae
2 > by 2 @ ter ekg 1;
2 ‘ . Se ORS Ne Ue et
=
‘i i'tegy Ocede Ss
=
Chapter 4
SEPARATION PROPERTIES
In this chapter we examine a variety of spaces that have some, but in general
not all, of the attributes of metric spaces. Normal spaces may lack a distance
function, but they nevertheless have sufficient topological structure to yield
some of topology’s major theorems. We also introduce a number of other
types of spaces that are of interest precisely because they lack some of the
properties inherent to metric spaces.
A. NORMAL SPACES
We have previously seen that in €*, the distance between closed subsets A
and B may be zero even if A and B are disjoint.
UsAU:
- +,
V=BU'.’.’.
Nevertheless, note as illustrated above that in the case shown, we can find
disjoint open sets U and V that serve to keep A and B apart. In fact, in
metric spaces, all that is needed for such a “separation” to exist is that the
closure of each of the sets does not intersect the other. This is the content
of the next theorem.
o7
98 4 Separation Properties
(4.A.4) Exercise. Show that a space X is normal if and only if for each
closed subset A and for each open set U with A C U, there is an open set V
such that ACV CV CU.
(4.4.5) Exercise. Show that a space X is completely normal if and only
if every subspace of X is normal. [Hint: Consider X \ (AN B).]
Metric spaces are, of course, normal. However, in view of the Tihonov
plank cited above, there are normal spaces that are not metrizable. A number
of further examples may be derived from the following result, whose proof is
an almost classic illustration of the use of compactness.
The inductive step should now be apparent. Suppose the sets Uy (satisfying
the lemma) have been defined for all membersof D with denominators less
than or equal to 2*. Once again, by (4.4.4), there are open sets U;_1/2k+1,
j =1,2,...,2*, with the property
AC Uj /(ae+1) C U1 (241) SU pa ee ee
The Uq’s are now defined for all d’s of the form p/2**1, and this completes
the induction, and the proof of the lemma. O
(ii) if |f(a)| < c for each a € A, then F may be chosen so that |F'(x)| < c
for each z € X. }
Proof. The proof is based on repeated applications of (4.B.3). Three
cases are considered.
Case 1. Suppose |f(a)| < c for each a € A. By (4.B.3), there is a map
go : X — €} such that |f(a) — go(a)| < (2/3)c for each a € A and |go(z)| <
(1/3)c, for each x € X. Application of (4.B.3) to f — go yields a function
gi : X — €, where |gi(z)| < (1/3)(2/3)c for z € X and |f(a) — go(a) —
gi(a)| < (2/3)(2/3)c for a € A. Continue inductively and obtain a sequence
of functions go, 91, 92,... such that for each n € N, |gn(x)| < (1/3)(2/3)"c on
X and |f(a) — go(a) —--: — gn(a)| < (2/3)(2/3)"c on A. Define F: X + E}
by
F(a\= a Oil z).
For each n, let sn(x) = )7j.5 9i(z). Note that {s,} converges uniformly
to F(z) since |S 4 gn(x)| < 07-9 (1/3)(2/3)"c = c. Hence, there is an
integer N such that for n > N, |sp(x) — F(x)| < € for each x € X. We give
a common advanced calculus argument to show that F is continuous. Let
zo € X be given. Choose N so |s,(xz) — F(x)| < €/3 when n > N. Since
Sn(x) is continuous, there is a neighborhood U of zo such that if x € U,
then |sy(z) — sn(o)| < €/3. Hence |F(x) — F(ao)| < |F(x) — sn(z)| +
|s~(2) — Sn(Zo)| + |Sn~ (20) — F'(zo)| < € whenever x € U, which establishes
the continuity of F. That F is an extension of f and |F(x)| < c for each
xz € X may be readily verified by the reader.
Case 2. Suppose |f(a)| < c for a € A. By Case 1 there is a continuous
extension F' of f such that |F(z)| <c. Let B = {x : |F(x)| = c}. Note that
B is closed and disjoint from A. Apply (4.B.1) to obtain a function h: X >
E' such that h(A) = 1 and h(B) = 0. The desired extension G of f is defined
by pointwise multiplying the functions F and h, i.e., G(x) = h(x) - F(z).
Case 3. Suppose f is not bounded. Take any homeomorphism h :
€} - (-1,1). By Case 2, the map hf : A — (—1,1) may be extended to
G: xX - (-1,1); then F = h“!G extends f to allof X. O
Euclidean 1-space €' is not the only range space for which the foregoing
theorem is valid. In fact, quite a number of spaces may be used in place of
€'. Such spaces are called Absolute Retracts.
This use of the word “retract” will make more sense shortly, but first
let us prove a basic result concerning AR’s.
(4.B.10) Theorem. If the spaces Yi, Y2,...,Yp_ are AR’s, and if the product
space |];_, Yi is normal, then the product space is an AR.
Proof. Suppose X is a normal space and A is a closed subset of X. Let f
be a continuous map from A into Flzas Y;. For each i, the map p;f : A > Y;
may be extended to a map F; : X + Yj. Define F : X > []/_,Y; by
Fry = (F(a) Pols)... crac): DO
(4.B.11) Corollary. Euclidean space €” isan AR. O
(4.B.12) Exercise. Show that J x J x--- x J is an AR.
(4.B.13) Exercise. Show that the property of being an AR is a topological
invariant.
A id A
(b) Show that a subset A of X is a retract of X if and only if, for each
topological space Z, each continuous map f : A + Z can be extended to a
continuous function from X to Z.
(4.B.16) Examples.
1. Any point in a space X is a retract of X.
2. Any closed interval of €1 is a retract of €'.
3. If x € €” and € > 0, then the closed ball S.(x) in €” is a retract of
Gon
(4.B.17) Exercise. Show that if X has the fixed point property and A is
a retract of X, then A has the fixed point property.
h(X2) \ h(X1) G U2, U, NU, = 0, and (U; U U2) M (h(X1) M h(X2)) = 0. For
i = 1,2 define sets P; = [U \(U; UU2)] UA(X;) and functions r; : P; > h(X;)
by
fe: ifz €U \ (U; UU2)
Clearly r; is continuous.
| Now let Ey = (YUU; UU2) \ U2 and Ey = (UUU,; UU2) \ U;. Then for
i = 1,2 we have P; = [(U\(U, UU2)]UA(X;) = (E;\(U; UU2)) UA(X;). Hence
Further Results Concerning Normal Spaces 109
(4.B.29) Remark. Once we have (10.C.3), it will follow that (4.B.28) may
be restated so that if X is a compact metric space, then X = X; U XQ is an
ANR»m.
(it) for each a € A, the family {Vy : \ < a} together with the family
{U5 21d 2a} cover, X.
We then show that the family {V, : a € A} is an open cover of X.
For purposes of motivation we observe that X \U,5.,, Ua is closed and
is contained in U,,. Hence by (4.4.4), there is an open set Va, such that
COON se Ca) Varo Vay C Uay- In order to replace Ua, by a similar
construction, we make use of the fact that {Va,,Ua,,.--} is an open cover
Of
Now assume for each 3 < a, we have constructed an open set Vg such
that Vg C Ug and that {Vg : 8 < a} and {U, : y > a} form a cover of X.
We show that this permits the construction of an open set V, such that
(i) Va C Ug, and
(ii) the families {Vg : @ < a} and {U, : y > a} form a cover of X.
Note that X \ (U{Vg : B < a}UU{U, : y > a}) C U, and X normal
implies that there is an open set V, such that
Then the combined collections {Vg : 8B < a} and {U, : 7 < a} cover
X, and this completes the inductive step. To see that the family {Vg :
B € A} covers X, suppose z € X and let U,,,U,,,...,U,,, be the sets
of UY that contain x. Assume that 7, > yj for 1 = 1,2,...,n —1. Then
{Ver si«vieig Vere «3 9| yn Unagas
oe} covers: X 5.whichsimplies thater y@ Maal
Vee, a Cl
is normal by (4.A.6).
1 2 3 4 w
(4.C.3) Exercise. Show that [0,w] and [0,9] are compact. [Hint: Consider
A = {a € [0,w] : [0,a) is compact } and show that if A = sup A, then
A€ A, and A =w.]
(4.D.2) Exercise. Show that a space X is regular if and only if for each
point « € X and for each open set U containing x, there is an open set V
such that rE V CV CU.
Note the similarity between definition (4.D.1) and the definition for
normality. In fact, it should be clear that normal T; spaces are regular.
Requiring that points be closed is obviously not a particularly strong
imposition on a topological space, but a still weaker condition defines an
even lower form of topological life. A space X is To if and only if whenever
x,y € X, an open set may be found that either contains x but not y or
contains y but not z; there is no guarantee, however, as to whether the open
set will contain specifically x or y.
A hierarchy of conditions (with increasingly restrictive structure) may
be set up as follows:
To space—defined above.
T, space—defined in Chapter 1.
T2 space—defined in Chapter 1.
T3 space—a regular T) space.
T4 space—a normal T) space.
Ts space—a completely normal T} space.
Metric spaces.
The T;’s are frequently referred to as separation axioms; basically, they
reflect the degree to which points or sets may be kept apart by open sets.
Strangely enough, an inordinate amount of mathematical energy has been
expended in defining separation conditions that lie somewhere between the
ones listed above (15, 22, etc.); however, with a few notable exceptions, this
sort of exercise would appear to be about as significant as it is interesting.
(4.D.3) Exercise. Show that metric > T; > T, > T3 > T7 > T; > To.
In the remainder of this chapter we see that none of the above implica-
tions is reversible.
(4.D.4) Examples.
1. Any set (consisting of more than one point) with the indiscrete topol-
ogy fails to be Tp.
2. The set R! with the open ray topology is Ty but not 7}.
3. The set R' with the finite complement topology is T; but not Th.
4. Let X = R'. For each point x € R! and each €!-open interval U
containing z, set U = {y € U : y is rational} U {z}. The family of all such
sets form a basis for a topology that is Tz but not regular.
5. (Niemytzki’s Tangent disk, or bubble topology). Define sets X and L
by X = {(z,y) € R* : y > 0} and L = {(z,y) € R? : y = 0}.To forma
topology for X let B be the set of all U C X such that either
The Separation Axioms 113
Then B is a basis for a topology for X that is T; but not T,. We use the Baire
Category Theorem to show that X is not normal. Let A = {(z,0) € L :
zx is rational} and B = {(z,0) € L : z is irrational}. Certainly A and B are
disjoint closed subsets of X (any subset of L is closed in X). Suppose that
U and V are open subsets of X containing A and B, respectively. We show
that U and V must intersect. For each (z,0) € B, let Dz be a disk of radius
rz that lies entirely in V and is tangent to L at (z,0). Forn = 1,2,..., set
Wr = {(x,0) € B : rz >1/n}. Then L = AU(U?2, W,,). Since L (with the
usual topology for £1) is a complete metric space, it follows from (3.B.16)
that some W ,,contains an open interval. However, since every open interval
in L contains a rational point, U and V must intersect.
(4.D.5) Exercise. Show that the space X in Example 5 is regular, T;, and
separable. Show that L with the relative topology is not separable.
We mention one other type of space that has special interest in analysis.
Its definition is reminiscent of Uryson’s lemma.
(4.D.8) Definition. A T, space is a Tihonov space (or T31) if and only
if for each closed subset C' of X and for each point x € X \ C, there is a
continuous function f : X > I such that f(z) = 0 and f(C) =1.
If the requirement that X be T; is removed in (4.D.8), then X is said
to be completely regular.
(4.D.12) Example (Half-Open Interval Topology). Let X = R' and let the
topology for X be determined by a basis consisting of all half-open intervals
of the form [a, b) where a < b.
We show that X is 75. It is clear that X is T;. Suppose that A and B
are subsets of X such that ANB = @ and BNA = 9. For each z € A there is
a half-open interval [x, e,) contained entirely in X \ B, and similarly for each
y € B there is a half-open interval [y,e,) C X \ A. It is readily seen that the
sets U = U{[z,ec) : x € A} and V = U{[y,e,) : y € B} are disjoint open
sets containing A and B, respectively.
PROBLEMS
Section A
Section B
1. Show that subspaces of perfectly normal spaces are perfectly normal.
2. Show that [0,1) is an AR.
3. Prove Uryson’s lemma using Tietze’s extension theorem.
4. Suppose X is a normal space and A and B are disjoint closed subsets
of X. Show that there is a continuous map f : X — [a,b] such that
f(A) =azand fils) =.
5. Let S = {0,1}. Show that a space X is path connected if and only if
each continuous map f : S + X can be extended to (0, 1].
6. Suppose (X,d) is a metric space. Show that (X,d) is perfectly normal,
i.e., fill in the details for the remarks above (4.B.5).
Show that a retract of a T> space is closed.
Show that if a (finite) product X of ANR’s is normal, then X as an
ANR.
9. Assume the (n + 1)-dimensional Brouwer fixed point theorem. Use
(4.B.17) to show that S” is not an AR.
10. Find a closed subset of a normal space that is not a Gs set.
11. Show that a pseudometric space is perfectly normal.
12. Suppose X and Y are topological spaces, A is a retract of X and B is
a retract of Y. Show that A x B is a retract of X x Y.
13. Suppose X is T, and an AR. If A C X is a retract of X then A is an
AR.
14. Show that an open normal subset of an ANR is an ANR.
15. Do the following closed subspaces of the plane €? have the fixed point
property?
16. Does there exist a countably infinite connected normal space that is T;?
17. Use Tietze’s Theorem to give an elegant proof of problem 7 from Section
A.
Problems ilaly/
Section C
Prove a converse of (4.C.2): if every point finite open cover has an open
shrinkage, then X is normal.
Show that closed subsets of a normal space are normal.
Show that a space X is normal if and only if for each finite open cover
{U,,U2,...,Un} ofX there is an open cover of X, {V, V2,..., Vn}, with
the property that V; C U; fori = 1,2,...,n. Avoid using the axiom of
choice in your proof.
Suppose X is a normal space and Aj, A2,..., A, are closed subsets of
X such that ()i_, Ai = 0. Show that there are open sets Vi, V2,...,Vn
such that A; C V; and ()i_, Vi = 0.
Let X, be any uncountable set and X2 be any infinite set. Suppose
that X, and X>2 have discrete topologies and let Xf = X1 U {c0;} and
XZ = Xz U {2} be the corresponding one-point compactifications.
Show that Xf x X} is normal, but (Xf x X3) \ {(co1, o02)} is not.
Section D
} . - - .
, (nN. ae Dat
pag :
> on 7 mney oe ee ae =—
ameeithin
grr :
te rsanianioniny hen theelis raniahadlaed ;
mi
‘5
ye
? ‘wa “a peony
ee oe
7 “bee At
ve
ojieee fie
Paeoer a ct
Weriltionrat
tm 16) weit a wid a-ak Pe: 7
coe Sst | 0. ebaiented
ngtv a eereiynt
ne 7
ss > > eabbpis joeihoviiap> One tho Srnceeesg Site deca Th
wach <0 gar 8 @.T,; Tad cai Cte eniieee Oe i
a shes “© tte 22 open aoe ll en T mb ©
: D2 Fer av © eos PN Oe ae
) 2 eet ere) 7
a
’ (ta ee)
Ss th ae
= = Gia, gtruaee fd
th Vite h.
et if Tow a 4
yi # -
ad Vina D: is
1-MANIFOLDS AND
SOME PLANE THEOREMS
A. 1-MANIFOLDS
We defined an n-manifold M in chapter 1 as a separable metric space with
the property that each point of M is contained in an open set homeomor-
phic to €". Our goal in this section is to classify connected 1-manifolds
(both of them). In Chapter 16, we perform the somewhat more complicated
task of classifying compact 2-manifolds. Curiously, it may be shown that a
reasonable classification of 4-manifolds is impossible (Markov [1960]).
(5.A.1) Definition. A triode is a space homeomorphic to
{(t,y)
€ E? : toe peu ee 7 r=, (sey I,
121
122 5. 1-Manifolds and Some Plane Theorems
Initially we consider a special type of simple closed curve, one that con-
sists entirely of vertical and horizontal line segments. These will be referred
to as rectilinear simple closed curves. The first theorem is a simple exercise
in analytic geometry.
Our next goal is to generalize Theorem 5.C.6 to obtain the Jordan curve
theorem for rectilinear simple closed curves. We begin by “simplifying” a
given simple closed curve.
V1 v2
Us5 U3
U4
fe
& ly —~*,
a,
Ses
k
(5.C.15) Notation. If t> 0, R; will denote {(z,y) € E? : |z| <t, |y| < t}.
(5.C.16) Theorem (Rectilinear Annulus Theorem). If R and R’ are
rectilinear simple closed curves with R' C I(R), then T = (RUI(R))N(R'U
E(R')) is homeomorphic to the annulus S!' x I. In fact, there is a space
homeomorphism that carries T onto FR \ int Ry.
Proof. Construct a small rectangle connecting R and R’ as indicated in
the figure. Note that AFDC and BF EC are rectilinear simple closed curves
whose intersection consists of the line segments F’ and C.
The Jordan Curve Theorem 131
A ait
A= U4U5 U1
B= U1 U4
V1
B C = 1403
U5
D= U2VU6G U3
V4
E= U2U3
fee U1 V2
ps
Use the map gluing theorem to combine h; and hz into a homeomor-
phism h mapping (RUI(R))N(R' UE(R’)) onto the annular region between
A'B' and C'D'. The homeomorphism h may be extended to I(DE), (5.C.11).
In order to extend h to E(AB), let g : €? > E* be an extension of h|p given
by (5.C.11). Then we have g : E(AB) - E(A'B'). Now define h : €? + €
by
‘ h(a). ilove AUTH),
huey =
ee if2e RUE(R). Oo
mesh so that if a brick intersects S, then the brick lies entirely in U. Let M
be the union of all of the partition elements that intersect S$. Let R be the
component of the frontier that contains S in its interior (why does such a
component exist?). By the previous theorem, R is a rectilinear simple closed
curve. If J(R) is contained in U, we are done. If not, then there are a finite
number of components of Fr M (other than R) that contain points of €? \U
in their interiors. We shall call such components holes. The holes satisfy the
following conditions:
(7) each hole is bounded by a rectilinear simple closed curve that lies
in I(R), and
(ti) each hole contains points of €? \ U.
Pitta ies
retraction. Let p be any point in €? that does not lie in S and connect each
point of S with p.
however, the proof is by no means trivial. In fact, similar theorems for higher
dimensions are false without additional hypotheses. For instance, in Chapter
15, we construct an example of a 2-sphere that cannot be carried by a space
homeomorphism onto the standard 2-sphere. The Alexander horned sphere
shown below is one of the most famous such examples.
GE **ieeenee ace
Choose 0 < Az < €/2 so that A, is less than the distance between C,
and S\C',. Note that if y € C, and z is a point of S such that d(y, z) < Az,
then z € C, and at most one component of S \ {y, z} has diameter greater
than e. Cover S with the sets {C, : x € X}, and select a finite subcover
{Cz,,Cz5,..-,Cz,}. Let 6; be a Lebesgue number for this cover, and let
6 = min{d;,z,,-.-, Az,}. Then it is easy to verify that 5 is the required
positive number. O
Tro
4. Show that the rectilinear simple close curve running around the
outside of the T;’s is the desired one.
142 5. 1-Manifolds and Some Plane Theorems
Si
Let D, be the component of S,/2 \ (5 \ (az, bz)) that contains (az, bz),
and select points gz and 7, such that g, € (az,z) and fz € (a,b). Let
Az = min{d(gz, x),d(7z,z)} and let 6, be a necking number for Az.
Now pick a point pp € D, MN E(S) and let Q; and R, be rectilinear
arcs in E(S)M D, with initial point p, and terminal points in S5, (gz) and
Ss, (72), respectively. From the respective terminal points, send out straight
line segments toward g, and f,, and let g, andr, be the first points in S that
these segments encounter. Replace these segments by rectilinear ones. Now
devise a method to obtain arcs [pz,qz] and [pz,rz] that have only the point
Pz in common and that lie (with the exception of g; and rz) in Dz N E(S).
Qazi Tons Ix) Toy) zr3) Tao A243 T 235 Qz5; Vo4) CERO RLS | Urn ee Ae) dx
For each 2, let T; denote the simple closed curve composed of the arcs
[Pei ) dai], (az; ’ Tz;] and [rz Px; )-
3. We say that T; and T; (i # j) are adjacent if and only if |i—j| = 1, or
1=1and j =n, or j = 1 andi = n; otherwise, they are nonadjacent. Now
alter the T;’s to meet the conditions. All repairs are to be done so that the
diameters of T; remain less than € and that each T; still intersects S only in
The Schénflies Theorem 143
the arc [qz;,72;]. To illustrate the basic procedure, we indicate how T, and
T3 may be pulled apart.
First, note that we may assume that 7; and T3 intersect only a finite
number of times, since slight adjustments of say T3 will alleviate the situation
indicated in the next figure.
oo00000000000000000000
0
The arc (gz, Tx,]is mapped onto [tn, t1], the arc [rz,, dz,] is mapped onto
[t1, ta], the arc [gz,,7z3] is mapped onto [t2,t3], etc. By the construction it
is clear that this homeomorphism moves no point more thane. O
Consider, first, the following subsets of S' (the points are specified by in-
dicating their radian measure): A; = {0,7}; Ap = {0,7/2,7,37/2}; A3 =
{0, 7/4, 2/2, 37/4, 7, 5/4, 32/2, 77/4}; etc.
The Schonflies Theorem 145
m/2
‘Ay ‘A2
37/2
31/2
figure. Then use (5.C.11) to further extend h to the two 2-cells bounded
by Si, S2, Bi and B} so that they are mapped homeomorphically onto the
corresponding cells in the disk bounded by T;, T2, A} and A}. (Clearly
(5.C.11) applies, even though the T;’s are not rectilinear. The skeptical
reader may See replace the T;’s and S! by eS in this proof.)
The points a = 73(0)jpas.= ran), as = cst /2), 2 = 13(37/2),
a3 = r3(7/4), a8 = r3(3n/4), a? = 13(57/4) and a3 = r3(Tm/4) are now
mapped by h~! onto points b?, 7 = 1,2,...,8 in S3. Arcs of diameter less
than 1/2 are found and mapped onto the obvious candidates in the disk.
The four pieces of the annular region between Sz and S3 are carried by an
extension of h to the appropriate wedges in the interior of S?.
Continue for each positive integer 7, and the end result is a homeomor-
phism h from I(S) onto I(S!). Let h = g~! on S, and we have constructed
a function from SUI(S) onto S'UI(S*) that is clearly 1-1, onto, and is con-
tinuous except possibly at points of S. To remove any doubts concerning the
behavior of h at a point 2 € S, we show that if {x;} is a sequence of points in
I(S) that converges to x, then {h(a;)} converges to h(x). It may be assumed
that for each 7, x; belongs to the annulus bounded by S; and S;4,. If x; is not
in S;, select a point c; € S; (c; isin the same rectilinear region as z;) such that
d(x;,c;) < 1/i. Consider d(h(x),h(xi)) < d(h(x), h(ci)) + d(h(c;), h(x;)).
Clearly, for large i, one need only worry about d(h(x),h(c;)). Let «€> 0
be given. Corresponding to € there is a 6 > 0 such that if s;,s2 € S and
d(s1,82) < 6, then d(g~!(s1),9~+(s2)) < €/2. Furthermore, there is a posi-
tive integer N such that for i > N, d(h;*(y),y) < 6/4 for each y € S; and
d(r;(x),x) < €/2 for each z € S.
Now, h(c;) equals r;g~!hj*(c;). For i > N we have d(h;'(ci),ci) < 6/4
and d(c;,z) < d(ci,zi) + d(aj,z) < 6/4+ 6/4. Hence, d(h7(c),2) <
d(h; *(ci), ci) + d(c,z) < 35/4, and thus d(g-1h;*(c:),g71(a)) < €/2. Fi-
nally, from the triangle nea we have d(rjh~1g;'(ci),g7!(z)) is less
than d(rig~1h; '(ci), 971A;*(ci) +d(g7h;* (ci),g~(a) which is less than
e. The continuity of h follows. (Why is it legitimate to use the sequence {c;}
instead of the original sequence {z;}?)
An analogous procedure may be used to obtain both a sequence {S;} of
rectilinear simple closed curves that converges to S from the exterior of S,
and an analogous homeomorphism. 0
of domain, and it is not difficult to construct examples that show that not
all topological spaces satisfy this condition. Invariance of domain turns out
to be one of the most powerful features of €”, and its proof is appropriately
sophisticated. The Schénflies theorem may be used to show that invariance
of domain holds in €?. That the invariance of domain property also holds in
€” for n > 2 is proven in Chapter 15.
(5.E.3) Exercise. Suppose S and S are simple closed curves in €?. Let U
be an open subset of €? such that U is homeomorphic to €? and SUS CU.
Show that there is a homeomorphism h : €? — £€? such that h(S) = S and
hlea\u
= id.
(5.E.4) Exercise. Suppose S and S are simple closed curves in €2 such
that S Cc I(S). Show that there is a homeomorphism h : €? - E€? such that
h(S) = S! and h(S) is a circle of radius 1/2 centered at the origin.
(5.E.5) Exercise (Annulus Theorem). Suppose S and S are simple
closed curves in €? such that $ C I(S). Show that J(S) MN E(S) is an annulus.
PROBLEMS
Section A
(c) L is connected.
(d) L is not compact (in fact, not Lindeléf).
(e) L is countably compact.
(f) L is not separable.
Section B
Sections C and D
Dy UW By. Cz D>.
(b) Show that £ does not belong to the bounded complementary domain
of either A, UD; UB, or Az UC, U Do.
(c) Deduce from (1) and (2) that f must be 1-1.
(d) Show that f is a homeomorphism.
150 5. 1-Manifolds and Some Plane Theorems
Show that there are only a countable number of mutually disjoint triodes
in the plane (Pittman [1970]) [Hint: Let D= {D; : i € Z+} be a basis
for €! consisting of a countable number of open disks. If T is a triode,
let Dr, € D such that 0 € Dry and Dri N{X,Y,Z} =9.
xX
a
ZL
151
152 6. The Product Topology and Inverse Systems
In the first exercise that follows, the reader is asked to confirm that the
product topology is indeed the smallest topology for which continuity of the
projection functions is maintained. The exercises, theorems and problems of
this section give further justification for this choice for the topology.
(6.A.3) Exercise. Suppose {Xq : a € A} is a collection of topological
spaces and [],¢, Xa has the product topology U. Show: if V is another
topology for [],¢, Xa such that V GU, then not all of the projection maps
are continuous.
(6.4.4) Exercise. Show that the projection maps are open for the product
topology U and the box topology V.
The procedure used to obtain a basis for the product topology from finite
intersections of subsets is utilized frequently enough to merit special consid-
eration.
(6.A.7) Exercise. Show that the topology obtained from a subbasis is the
smallest topology that contains the subbasic sets as open sets.
(6.A.8) Exercise. Show that sets of the form (a,0oo) and (—oo,b), where
a,b € R, form a subbasis for €!. Formulate and prove a similar result for
the order topology.
X \ (Mea Vi) = U(X \ Vi) C Uli Uji Mi; However, that is absurd,
for we would then have
from the fact that pg(V) = Xg for all but a finite number of 8 € A—even
though V is not necessarily a basic open set.
For the converse, let K C A be a finite subset such that a € A\ K
implies Xq is connected. Suppose z € V C hve a Xa Where V is open. We
may assume (why?) V is of the form Vy, x Va, X-:: X Va, X []Xy, where
7 € A\{a1,a2,...,an}. Now, fori = 1,2,...,n, let Uz; be a connected open
set such that tq; € Ua; C Va;. For each B; € K, let Ug; be a connected open
set in X that contains g,. Then U = Ug, x:::xUa, xUg, x---xUz,, XT] X47;
where y # aj, 3; is open and connected. Since x € U C JV, it follows that
Tlaca Xa is locally connected. O
The proof of the following theorem is left to the hyperactive reader.
where x = (%1,22,...) and y = (y1, y2,...). The details needed to show that
p is a metric are omitted, and we concentrate on showing that the metric
topology induced by p coincides with the product topology for iez+ XK
Let M denote the topology induced by p and let P be the product topology.
We first show that P_C M. Supposex = (2),22,...) is a point in
Iliez+ Xi and let U = S21(a1) x. --- x S(¢,) x T],,, Xi be a typical
basic P-open set containing x. Let « = min{e,/2,...,€,/2"}. We show
S?(z) C U. Suppose y = (y1, y2,---) € S?(x). Then p(x,y) < € and hence
>a Gi(2i, yi)/2* < €. Therefore, d;(2;, yi) Si, 168 tasked, ae, ond
consequently S?(a) CU.
We now show M C P. Let U = S?(x) be a typical basic M-open set
containing x. Choose N large enough so that 57°, 1/2' < €/2. Then one
can easily check that Sein (a1) X +++ x Sern (Ln) x Ilisnai elles a0)
Inverse Systems: The Preliminaries 157
(6.B.5) Examples.
1. Let D =N and for each n € N let X, = €!. Define each bonding
map f, to be the identity function.
fi fa fs fa
2. Let D = N and for each n € N let X,, = Z with the discrete topology.
Define f, by setting f,(p) = p — 1, for each p € Z.
Inverse Systems: The Preliminaries 159
. fi fe g fs M fa
fi fa fs
Xo
(6.B.6) Exercise. Guess the inverse limit in each of the preceding examples.
(Hint: In the last example, try the Cantor set.|
Since Xoo is given the relative topology inherited from |], -p Xa, a basis
for X,. may be defined by taking intersections of members of the basis for
Heep Xa with X... However, an amazingly simple description of this basis is
obtained in the next theorem. If (Xa, fag, D) is an inverse system and X qq is
the inverse limit, then the restriction of the projection pg : Bees Die ee
to Xo is denoted by pa. Theorem 6.B.8 says that the analog of a subbasis
for the product topology is actually a basis for the inverse limit.
(6.B.7) Lemma. If (Xa, fog, D) is an inverse system, then po = fapps
whenever 0 > a.
160 6. The Product Topology and Inverse Systems
Hence BEB. O
cannot be empty if the factor spaces of the inverse system are compact and
To.
(6.B.11) Theorem. Suppose (Xq, fag, D) is an inverse system, where each
Xq is compact and T2. Then X. is nonempty, compact and T».
Proof. We use the finite intersection property characterization of com-
pactness to find a point in X.. Let A = {Agg : a < B}. Then JA is a collec-
tion of closed subsets of the compact space [],¢p Xa (6.B.10). We show that
A has the finite intersection property. Suppose Aq,g,,---,Aa,, is any finite
subcollection of A. Since D is directed, there is a 1 € D such that » > ;
for? = 1,2,...,n. Pick an arbitrary point 7, € X) and fori = 1,2,...,n
let ta; = fa;(Z,) and zg, = fg,,(). Define a point {ye} € [xen Xa
by Settingsya; = wa and wei rgnior/ += 1,2, ..., nvyt=iey; and let
ting Yq be arbitrary for the remaining coordinates. Then {ya} € ()j_, Aa:a;
since fais: (Ya;) = Faia; (p;) = faa: fa:r(£a) = foir(Z,) = La; = Yai:
Thus, A has the finite intersection property. It follows from (2.H.2) that
Na<p 40g # 0. However, by (6.B.10), we have (1),
<gAag = Xoo. Further-
more, Xo is compact, since it is a closed subspace of the compact space
Haep Xa. That X is Tz comes from problem 13 of Sections Band C. O
(6.B.12) Exercise. Show that if (Xa, fag,D) is an inverse system such
that each fag is 1-1 and onto, then py is 1-1 and onto.
(6.B.13) Exercise. Show that if (Xn, fn, N) is an inverse system such that
fn is onto for each n, then py is onto for each n.
Yu
a Jap
Ye
The family of maps H = {hq : a € D} is called a map between A and B.
ho oil he h3
Yo Yi Y2 ee Oo
than 1 and which also satisfy conditions (i) and (ii) above. Lemma (6.C.10)
allows us to assume that if Up € Up, then Up and ho(Uo) both contain the
same number of elements from U; and U4, respectively. Hence, there is a
1-1 function h; from U4, onto U/; with the property that if Uy; € U,, Uo € Uo,
and U; C Uo, then hi(U1) C ho(Uo). This procedure can be repeated in the
obvious inductive fashion to obtain two inverse systems and a map between
them, as indicated in the following diagram. Each U,, is assumed to have the
discrete topology and each f; and fi is defined as in (6.C.5).
Uo U; Up Us
ho hy ho hg
4 fe 2 fs 2
It is clear, (6.C.8) and (6.C.9), that hg is a homeomorphism. By
(6.C.5), X is homeomorphic to Uz and Y is homeomorphic to WU. O
We now prove one of the most startling and least intuitive results in all
of topology.
Xo Xi Xo X3+~—
n 92 93
Denote the respective inverse limits by Vi. and Xo, and the limit of
the hn by hoo. Clearly, hoo is onto, Voo is a compact metric space and X is
homeomorphic with X... We now investigate properties of Voo.
We first show V,, is totally disconnected. Suppose x = (Zo, %1,...) and
y = (Yo, ¥1,---) are distinct points of Vo. We are looking for an open and
closed set in V, that contains z but not y. Since x and y are distinct, there
is a first integer n such that rp # yn. Suppose Zo # Yo, where rp = u x {i}
and yo = t x {i}. If u # G, then an integer n may be chosen large enough
so that any two members of U, that contain z and y respectively must be
disjoint. Suppose u € Uny and t € Ung. Then the corresponding sets Vji;...p
and Vpjij...¢ are disjoint and both open and closed. Furthermore, Vij...) is
pulled back by p;' to an open and closed set in V.. which contains x but
not y; thus x and y have been separated in Vo.
If u = t, then 2 ~ ¢. Therefore, we have zo € Vo; and yo € Voz;
furthermore Vo; and Vo; are disjoint, open, and closed. Again, p—'(Vo;) is an
open and closed subset of V. containing x but not y.
If zo = yo, let m be the first integer such that zm 4 Ym. Suppose, for
instance that m = 2. Then since the case u # % can be handled as above,
we have z2 = u x {i} x {j} x {k} and yo = ux {i} x {j} x {k}, where
k Ak. Thus, to € Voij, and also yo € Vaijh> where Voi;, and Voisk are
disjoint, open, and closed. Then p~*(V2;;,) is an open and closed subset of
Vo. containing x but not y. Hence, Vx is totally disconnected.
Although V,. may not be perfect, it follows from (3.B.20) that Vi. x K
is perfect. Therefore V.. x K is homeomorphic to K (6.C.11, 6.C.13), and
hence the following maps exist: a homeomorphism H : K + (V. x K), a
continuous surjection q : (Voo XK’) > Voo defined by q(v, y) = v, a continuous
surjection hg : Voo —- Xoo, and a homeomorphism Y : X, ~ X. Then
f = WVhe¢d maps K-ontock.
topology. It can be shown that every compact T> space is the continuous
image of a closed subset of some Cantor space (see Aleksandrov and Uryson
[1929], for example).
(6.D.2) Exercise. Show that the inverse limit Tos ot (1, fas) is home-
omorphic to D = (\2, 7".
We next investigate an inverse system, whose limit corresponds to the
solenoid D = (\%2, T’, that proves to be more useful than the one above.
168 6. The Product Topology and Inverse Systems
Suppose that (Cn, 9n,N) is the inverse system where, for each n, Ope
is the unit circle represented in polar coordinates by {(r,@) : r = 1}, and
Gn 2 Cn —> Cn-1 is defined by g,(1,0) = (1,28).
61 (p)
We construct a map between the systems (T”, fn,N) and (Cn, gn,N) as
follows. Let p € T° and let Q be the plane containing the z axis and p.
Define 09(p) to be the angle measured counterclockwise in radians between
the positive « axis and Q?. We assume the range of 4 is [0,27). Now
suppose p € T!. Let Q) be the plane passing through the z axis and p.
Define 6,(p) to be the angle measured counterclockwise in radians formed
between the x axis and Qh subject to the following conditions:
(i) if 6;(p) = 0, then so does 69(p);
(ii) the range of 0, is [0,47), and 6, is continuous on T! \ 6;1(0);
(iii) 61(p) = 1 (H) if and only if Q; = Qj and p and # belong to the
same component of QT" (see the preceding figure).
If p € T?, then let Of be the plane containing the z axis and p, and
define 92(p) as the angle measured counterclockwise in radians between the
x axis and Ope where 62 satisfies the conditions:
(i) if 62(p) = 0, then so does 6; (p);
(ii) the range of 2 is [0, 8m), and 62 is continuous on T? \ 651 (0);
(iit) O2(p) = 42(p) if and only if Q? = Q3 and p and # belong to the
same component of Q2 NT”.
Similar functions are defined for each positive integer n. Note that 6n(p)
either equals 6,_1(p) or equals 6,_1(p) + 2"7. Define a map h, : T" + C,
by hn(p) = (1, 6n(p)/2”). Then hp is clearly continuous, and hence we have
the following diagram.
The Dyadic Solenoid 169
ho h1 h2 hg
However,
hn fa(p)= (1,SEP?)
and since scurrying around the unit circle 27 radians lands you back at the
same point, the commutativity of the diagram is established.
We show h is 1-1 and onto: Suppose t = (p,p...) and t = (p,p,...) are
distinct points in T,. Then clearly ho(p) 4 ho(p), and hence hoo(t) # hoo(t).
Therefore h is 1-1. In order to show hq is onto, let c = (c9,c1,...) € Coo,
where ¢, = (1)%;,) foreach nm. Note that r, =7p,=1/2 of Tr = (tye /2) + 9,
For each, tet o;,-= 2". hen eithera,, = f_2q00l Sp yoy ee e. Let
Hy, = 6751(an), i.e., Hn is the appropriate component of the cross section of
T,, determined by the plane that contains the z axis and forms an angle of
Zn radians with the z axis. Readers can convince themselves that the point
t = (p,p,...), where {p} = ()7—_, Hp, is mapped by hoo onto c.
Recall that a topological space X is homogeneous if and only if, given
any two points z and y in X, there is a homeomorphism h : X — X that
carries z to y. Note how the use of inverse limits yields an easy proof of the
following theorem:
(6.D.3) Theorem. The dyadic solenoid is homogeneous.
Proof. We use the above notation, where the solenoid is considered to
be Css. Let x = (49,21,...) and y = (yo,41,...) be points in C,,, where
Zn = (1,Qn) and yn, = (1, Bn). Consider the mapping system
170 6. The Product Topology and Inverse Systems
Co = C1 = C2 = Co
ko ki ko k3
Co sangs C1 more E C3
where k,(1,9) = (1,8 + Bn — Qn). It is a routine matter to check that the
diagram commutes and that each k, is a homeomorphism. Thus, by (6.C.8),
(6.C.9), and (2.G.11), koo is a homeomorphism, and a direct computation
shows that k..(z)=y. O
More results involving inverse limit systems will be obtained in Chapter
9. At this point, the following articles should be well within the grasp of
the rara avis that is interested in some curious developments in inverse limit
theory. Brown [1960], Jolley and Rogers [1970], Nadler [1970], and Kresimar
and MardeSi¢é [1968].
PROBLEMS
Sections A
1. Show that the projection maps of [[,<, Xa are open, but need not be
closed.
2. Show that [],<, Xa has each of the following properties if and only if
each X, does:
(a) To,
(b) Ty,
(c) T,
(d) regularity,
(e) complete regularity
3. Let {Xq : a € A} be a family of topological spaces, and for each a let
B. be a basis for Xq. Describe a basis for [[,<, Xa in terms of the
Bas:
4. Show that [].,<, Xa is first (second) countable if and only if each Xq is
first (second) countable and all but a countable number of the X, have
the indiscrete topology.
5. Let P be a topological property and X a nonempty set. Then a topology
U on X is maximal with respect to P if and only if whenever V is a
topology for X and U G V, then V does not have property P; UW is
minimal with respect to P if and only if whenever V is a topology for
X and VY G U, then V does not have property P. Show that: (a)
Problems il
14, Show that the Tihonov theorem does not Aes if []acq Xa is given the
box topology.
15. Suppose {Xq : a € A} is a family of topological spaces. Show that
Toca Xa i8 separable if and only if each Xq is separable and all but at
most 2%° spaces are a point (2° is the number of functions from Z into
the set {0,1}). [Hint: See Comfort [1969].
16. Show that II I, is compact but not sequentially compact.
ael
Sections B and C
11. Find examples which show that the hypotheses of (6.C.9) are necessary.
12. Consider the set (no topology) X = ][ {0.1}. Show that it is not
Tel
possible to impose metric topology on X so that X is compact.
13. Suppose (X., faZ, D) is an inverse system and X,. # @. If each XQ is
To, respectively T), respectively T>, respectively T3, respectively T3312,
then so is Xq.
Section D
1. Consider the Cantor set as an inverse limit and show that it is homoge-
neous.
*. ma a
» = ~~ € oa -® co=—ay Mi 7 < - a
= a _ a1 - oe
7 ad - -
= LY 2
ice A ' ‘
-»
a fies :
= a
on -
.
an o
a q
.
-~ . ‘
—_
t=
Chapter 7
175
176 7. Function Spaces, Weak Topologies, and Hilbert Space
(7.A.1) Exercise. Show that subbasic sets for the product topology on
Hom(X,Y) are of the form S(z,U) ={f:X + Y : f(x) € U}, where U is
an open subset of Y and ze X.
The sup topology (or equivalently the compact-open topology on C(X, Y))
when X is compact and Y is metric) is frequently referred to as the topology
of uniform convergence. The following exercise gives the motivation for this
terminology.
(7.B.12) Exercise. Suppose X is a compact space and (Y,d) is a metric
space. Show: if {fn} converges to f in C(X,Y) (with the sup topology),
then {fn} converges to f uniformly, i.e., given « > 0, there is an N € Zt
such that if n > N, then d(fn(x), f(x)) < € for alla € X.
(7.B.13) Definition. Suppose X and Y are topological spaces. The func-
tion e : Hom(X,Y) xX — Y defined by e(f,x) = f(z) is called the evaluation
function associated with Hom(X,Y).
Note that e(f,z) = e:(f), where e, is the evaluation map discussed
previously. Our principal result involving e is the following.
x ls Y
é é!
|
Iyer I5—j—[I
ge 1s
[Hint: If {zs} € []pexJy, let the g-th coordinate of #({zs}) be the (gh)-th
coordinate of {zs}.] Show also that #(é(X)) Cc é(Y).
The Weak Topology: II 181
0 ifs Ee XnqiNA
fn(z)=¢1 if 6 win 8
fn(c) if2e Xn
E. HILBERT SPACE
One of the more interesting spaces in topology, both because of its versatility
and because of its unpredictability, is the classical Hilbert space. Hilbert
space serves as an important bridge in spanning the hiatus between analysis
and topology; we, however, will focus our attention on some of its topological
properties. Although it has been shown that Hilbert space is homeomorphic
to the countable product of €! with itself (Anderson [1966]), we begin with
the classical description of Hilbert space.
Let H denote the set of all infinite sequences {2,, 72,...} in R’ that are
square summable (i.e., )>;~; z? < co). We frequently denote such sequences
by (an) or (x1, 22,...). A metric p is defined on H that is completely anal-
ogous to the usual Euclidean metric for €”. If @ = (tn) and y = (yn) are
elements of H, then define
184 7. Function Spaces, Weak Topologies, and Hilbert Space
Consequently, we have
From the definition of Hilbert space and from the foregoing theorems,
one might be led to believe that (H, p) has the same properties as Euclidean
space. The next two theorems should dispel any such notions.
186 7. Function Spaces, Weak Topologies, and Hilbert Space
(7.E.4) Theorem. Hilbert space (H, p) is not locally compact. In fact, for
each x € H and each € > 0, S,(x) fails to be compact.
Proof. Suppose x = (11,22,...) € H and e > 0. For n = 1,2,... let
Yn, = (21, 22,---)En—-1,En + 6, £n41,--+). Note that for each,7y,°¢ See),
but for i # j, p(yi,
Yj) = /2e. Thus, the sequence {y,,} has no convergent
subsequence. Hence S,(x) is not sequentially compact, so by (3.A.11) can
not be compact. O
Invariance of domain fails badly in (H, p), i.e., open sets in H may be
mapped homeomorphically onto non-open subsets of H. In fact, H itself can
be embedded (even isometrically) into itself as a nowhere dense subset.
(7.E.5) Theorem. Suppose A = {(21,%2,...) € H : 2 = O} and define
W:H A by W((21,22...)) = (0,21, 22,...,). Then w is an isometry and
int A= 0.
Proof. Clearly w is an isometry. To see that A is nowhere dense in H,
let x = (%1,22,...,) € H and let € > 0; we show that S,(x) is not contained
in A and hence A has empty interior. Choose b # 0 so that |x; — b| < €/2
and let y = (b, £2, 73,...). The proof will be complete once we establish: if
6 = min{|b|,€/2}, then Ss(y) C S.(x) and Ss(y)N A= 0.
Suppose z € S;(y). Then we have p(x, z) < p(x, y)+p(y, z) < €/2+€/2,
and consequently, Ss(y) C S.(x). If w = (wi, we,...) € Ss(y)N A, then
w, = 0, and hence p(w,y) > |b| > 6 > p(w, y)—an unusual occurrence at
best. O
Next we show that any separable metric space can be embedded in He.
(7.E.10) Theorem. The Hilbert cube, Hc, has the fixed point property.
Proof. Consider Hg as the product space [0, 1] x [0, 1/2] x [0, 1/27] x ---
(why is this possible?). Let f be a continuous map from Ho into Hc,
188 7. Function Spaces, Weak Topologies, and Hilbert Space
and let € > 0 be given. Choose N large enough so that >, 1/n? < ,
and let A = [0,1] x [0,1/2] x --- x [0,1/2%]. Define a map fi AoA
by f(21,22,-..,;2n) = (y1,Y2,---,YN), wheres BN ONO...) =
(15 °09,,4045
(41, Y2s-++>YNyYN+1)YN+2,---). Then f has afixed point a = (a), @2,...,an)
by the Brouwer fixed point theorem (14.C.22). Let a = (a1, a2,...,an,0,0,...).
Then p(f(a),a) < €, and consequently, by (7.E.9), Hc has the fixed point
property. O
PROBLEMS
Sections A and B
Section C
1. Let X = 0,00) have the relative topology from €1. Define f : R' > X
by f(x) = x”. Determine the weak topology for R’ induced by f.
2. Suppose Y is a topological space and f : X — Y. Let W be the weak
topology for X induced by f. Show that A C X is closed (with respect
to the weak topology) if and only if A = f~!(B), where B is closed in
YE
3. Generalize in some way the result of problem 2, where the weak topology
for X is induced by the family of functions {f, :X + Yo : a € A}.
4. Suppose (X,U) is a T3 1 space. Show that U coincides with the weak
topology generated by {f : X — €! : f is continuous}.
5. Suppose S is an arbitrary set and f : S — T is an onto function, where T'
is a connected topological space. Show that if S has the weak topology
induced by f, then S is connected.
6. Repeat problem 5, replacing “connected” with “compact” and “count-
ably compact.”
7. Suppose A is a subset of a space X. Show that the relative topology
for A coincides with the weak topology for A generated by the inclusion
functioni: A> X.
8. Suppose (X,U) is a topological space. Let {Ya : a € A} be a family of
spaces and {f,:X — Ya : a € A} a corresponding family of functions
with the property that for each space Z and for each function f : Z > X,
f is continuous if and only if fy f is continuous for each a € A. Show
that U/ is the weak topology for X generated by {fg : X + Yo : a € A}.
Section D
1. Let X have the weak topology with respect to {Ag : a € A}. Show
that C is closed in X if and only if CM Ag is closed in A, for each a.
2. Suppose X is a topological space and {Ag : a € A} is an open cover of
X. Show that the weak topology induced by the cover {A,} coincides
with the given topology on X.
3. Suppose (X,//) is a topological space and A = {A : a € A} is a closed
cover of X. Let W be the weak topology on X induced by A. Show:
a) UCW.
Problems 191
d, (x,y) rye Ay
d(z,y)= d2(x,y) zy € Ag
inf {d(z,z) +do(z,y)} 2 € Ay \ Ao and y € A \ Aj.
zE€AiNA2
Section E
Show that the topology induced by d coincides with the product topol-
ogy.
Show that []?°, €} with the metric given in problem 4 contains all sep-
arable metric spaces X via the function f(x) = (d(z,z1),d(z,22)...),
where x € X and {z;} is a countable dense subset of X.
Find an example of an infinite set of points A in H with the property
' that p(x,y) = 2 for each x,y € A.
Show there is a point x and a closed set C in (H, p) with p(x, C) = € > 0,
but p(xz,c) > € for each cE C.
Prove or disprove: If X, Y, and Z are topological spaces with X x Y
homeomorphic to X x Z, then Y is homeomorphic to Z.
Let X = []7°, €},
a
where €} = €' for each i. Define a metric
co
has Ge as
Le, teil
(2,y) > i! (1 + [zi — ysl)
Show that (X,d) is a complete, separable, metric space which is not
totally bounded.
10. Show that (X,d) in problem 9 is homeomorphic to a subspace of Hilbert
space, and Hilbert space is homeomorphic to a subspace of (X, d).
Chapter 8
QUOTIENT SPACES
193
194 8. Quotient Spaces
xX 2 “a
P Dy
X/Gs
B. IDENTIFICATIONS
Partitions of a space X are frequently formed by “identifying” certain points
in X with others. An equivalence relation ~ on X leads quite naturally
to such a partition, where the members of the partition are the equivalence
classes under ~. In this case, we denote the resulting decomposition or
quotient space by X/~. For example, if X = [0,27], we can induce an
equivalence relation ~ on X by declaring 0 ~ 27, and x ~ y if and only if
x = y for z,y € (0,27). The equivalence classes consist precisely of single
points in the open interval (0,27) together with the set {0,27}. What does
the corresponding quotient space look like? Intuitively, the end points of
the interval have been “glued” together, so it would seem that the resulting
space should be a circle; that this is indeed the case may be established by
considering the following commutative diagram, where f(x) = (cos z, sin z).
x A S!
X/G;= X/~
Since [0,27] is compact, (8.4.7) tells us @y is a homeomorphism. Conse-
quently, the quotient space is topologically a circle. The only trick here was
to find a continuous onto map for which Gy is ~.
Many interesting topological spaces may be constructed in a similar
fashion. For example, although the torus is usually defined to be the product
space S! x S!, an alternate description can be obtained using quotient spaces.
Start with the square X in the zy plane (€7) whose vertices are the points
(0,0), (0,2), (2,0) and (2,2). Identify each point in X of the form (0,y) with
the corresponding point (2, y) on the opposite side of the square. Intuitively,
that has the effect of rolling up the square into a tube. Now glue the ends
of the tube together by identifying points of the form (z,0) with (z,2) and
196 8. Quotient Spaces
XG i S1 x S!
P Dy
XIGy
More exotic spaces may also be constructed when starting with a square.
For instance, the Klein bottle can be defined as the space obtained by forming
a tube and then identifying points of the form (z,0) with (2—z,2). This has
the effect, intuitively, of twisting the tube before gluing. It can be shown that
the resulting space is not embeddable in €%, although it can be embedded in
&4
(2—z,2) (2,2)
SA
ES stor
(8.B.2) Exercise. Suppose X is a subspace of €” C €"+!. Sometimes the
Cone over X is defined to be the subspace of €"* consisting of all points
on the line segments from (0,...,0,1) to (2,0) where x € X. Problems 14
and 15 show that the Cone of X is not necessarily homeomorphic with the
cone of X.
a) If X Cc €” andi: cone(X) — Cone(X) is the obvious map, show
that 7 is continuous.
b) What reasonable conditions can be imposed on X C €” to guarantee
that i is a homeomorphism.
AN
se
Cones and suspensions may also be derived as special cases of an im-
portant construction that involves “sewing spaces together” with the aid of
a given continuous function.
Recall from Chapter 1 that if X and Y are disjoint spaces, then the free
union of X and Y is the topological space Z = X UY, where aset W C Z
is open if and only if WN X and WNY are open in X and Y, respectively.
Suppose X and Y are disjoint spaces and A is a closed subset of X. Let
f :4—Y bea continuous function. Define an equivalence relation in X UY
by identifying each point z € f~*(y) with y. Denote the resulting quotient
space by X Uy Y. Thus A has been “glued” to its image f(A), and in the
198 8. Quotient Spaces
process, X and Y have been sewn together. The space X Uy Y is called the
adjunction of X andY by f. (Of course, if X and Y are not initially disjoint,
then the usual procedure may be utilized to “disjointify” them.)
DGC 9 6
C. IDENTIFICATION MAPS
(8.C.1) Definition. Let X and Y be topological spaces and let f : X > Y
be onto. The map f is an identification map if and only if the topology for
Y coincides with U = {U CY : f7'(U) is open in X}.
Clearly every identification map is continuous. For examples: the quo-
tient map P : X — X/G is an identification map, as well as continuous and
open or closed mappings.
A key property of identification maps is given in the next theorem.
(8.D.4) Remarks.
1. If (X,U) is a topological space, then U/ is always a subset of the
k-topology for X associated with U.
2. Suppose X is a k-space and Y is an arbitrary space. A function
f : X + Y is continuous if and only if fl¢ : C > Y is continuous
for each compact subset C' of X (8.D.2).
(8.D.5) Theorem. Suppose that (X,U) is a topological space and that
C={C : Cis acompact subset of (X,U)}. Then (X,U) is a k-space if and
only if a subset A of (X,U/) is closed whenever ANC is closed in C' for each
G.aC,
Proof. Suppose (X,U) is a k-space and A C X. Then by hypothesis,
the set A is U/-closed if and only if ig'(A) = ANC is closed for each C € C.
Conversely, suppose a subset A C X is U-closed whenever ANC’ is
closed in C for each C € C. We need to show that U = K, where K is the
k-topology on X. By (8.D.4 1.), UC K. We use (8.D.1) to establish the
reverse inclusion. Let A be K-closed. Then ig'(A) = ANC is closed for
each C’ € C. Hence, by hypothesis, A is U-closed. Thereforek CU. QO
(8.D.6) Exercise. Show that “open” can be substituted for “closed” in the
hypothesis of the preceding theorem.
The next theorem gives us a rich source of k-spaces
(8.D.7) Theorem. (i) Locally compact spaces are k-spaces. (ii) First
countable spaces are k-spaces. (In particular, every metric space is a k-
space.)
202 8. Quotient Spaces
E. CW-COMPLEXES
The standard open n-dimensional cell, also called an open n-cell or simply
an n-cell, is
IB) = (ata
ety by ree ajit--- +22 <ol}:
CW-Complexes 203
See eh,
Pos Ene Cas etre 1)
In addition, Bo = 1(B°) = {0} € €1 and S-! = . An (open) n-cell,
respectively, n-sphere in a topological space X is a subspace homeomorphic
to I(B”), respectively, S”.
The CW-complexes offer an excellent illustration of how different con-
structions (in this case, the quotient topology and the weak topology induced
by a family of spaces) may be successfully intertwined to produce spaces
of considerably more interest than their progenitors. The notion of CW-
complex was introduced by Whitehead [1949]. Essentially, CW-complexes
are formed by gluing together closed n-cells. For instance, to a collection of
0-cells (points) one might carefully glue a number of closed 1-cells, and then
to the resulting space a number of closed 2-cells might be adjoined, etc.
(8.E.2) Example.
There are two natural points of view to take when discussing how a
topological space is built out of cells. The first, is to have our building blocks
in hand, and then to construct a space “dimension by dimension.” First, put
the 0-cells in place, next glue in the closed 1-cells, etc. This is perhaps the
most intuitive way to proceed. However, there is some awkwardness in the
formulation since one can not describe the gluing maps for the closed n-cells
until one has set the closed 0- through (n — 1)-cells in place.
The second approach is to start with a topological space, break it up
into open cells, and specify how we want these open cells and their closures
to relate to each other and to the topology of X. We start with this latter
approach.
(8.E.3) Definition. Let X be a topological space. A cell decomposition of
X is a set € of subspaces of X such that: each element c € € is an (open)
cell, and (as a set) X is the free union of these cells. The n-skeleton of X is
the subspace X" = U{c € € : dim(c) < n}. Clearly
and (as a set) X =|) X”. For each cell c let ¢ C X be the closure of c in X
and defie the boundary of c by bd(c) = é\c.
(8.E.4) Definition. Let X be a topological space with a cell decomposition
€ and let c € € be an n-cell. A continuous function F : B” + X where
F(S"-1) ¢ X™7! and F|;(g=) is a homeomorphism onto c is called a char-
acteristic map of c, and f = F|gn-1 : S""=! X"1, is called a sewing map
for c.
(8.E.5) Theorem. If X is T2, € is a cell decomposition for X, and
F : B” > X is a characteristic function for a cell c € €, then
Gear (BY) exXeakvec
Gixbd(c) (Say epee
(iii) @ and bd(c) are compact and F : B” — € is an identification map.
Proof. (i) and (ii): Since F is continuous, F(B") C F(I(B")) C ©,
(1.F.5). Furthermore, F'(B”) is compact, hence closed since X is T>. Since
e C F(6"); we have ¢-G.iF(B"); Hence F(B") = ¢ =e bd(c).).Since
RUB") )\-= & clearly F(SP—*) Dbd(c); Furthermore; F(S8a" nc Ai
by the definition of characteristic map. Then we have F(S"~!) C bd(c),
since cU X"~! = O. (iii): By (2.G.11), F is a closed map, and hence an
identification by the remarks after (8.C.1). O
(8.E.6) Definition. Suppose X is a T> space with a cell decomposition €,
and each cell in the decomposition has a characteristic map. We call X a
CW-complez if and only if the following hold:
(C) (closure finite) For each cell c € €, @ intersects only finitely many
cells.
(W) (weak topology) If A C X and AMZ is closed for each cell c, then
A is closed in X.
A CW-complex is called finite if it contains only a finite number of cells,
otherwise it is infinite. A CW-complex is said to be n-dimensional if it
contains an n-cell but no cells of higher dimensions. If it is not n-dimensional
for any n we Say it is co-dimensional.
Condition (W) implies that the given topology on X coincides with the
weak topology induced by the closure of the cells. This avoids pathologies
as in example (a) below. Condition (C) circumvents the infelicity that can
be observed in example (f) below. In all the examples that follow we merely
describe the cell decomposition and leave it for the reader to decide that
characteristic maps exist.
(8.E.7) Examples.
(a) Let X = [0,1] with the usual topology. For each zx € [0,1] let {z} bea
0-cell. This cell decomposition does not yield X as a CW-complex since the
CW-Complexes 205
the only 0-cells, the open line segment from 2 to x2 be a 1-cell, and S? \ {t}
be a 2-cell co. Then X with this cell decomposition is a CW-complex. Note:
while t =bd(c2) C X?, it is not a union of cells.
(e) Any cell decomposition of a closed subspace of E” into a finite number
of cells is a CW-complex.
There is one 2-cell (S? minus the closed arc from zo to xz,). There are a
countable number of 0-cells, namely, ro, £1, £2, 73,°--. There are a countable
number of 1-cells, (the open arc from 2; to 2:4; for each i > 1. The 2-sphere
X with this cell decomposition has the weak topology with respect to the
closure of its cells, but is not closure finite. Note that X! with the weak
topology is not a subspace of X ({zo} is open in the weak topology but
not the relative topology on X!). Hence X! (with the relative topology) is
not a CW-complex. Closure finiteness prevents this bizarre occurrence, see
(8.E.11) iv.
(8.E.8) Lemma. Suppose X is a CW-complex with cell decomposition €,
€zp C €, and B= U. ce, ©, and suppose further c* CB for all cE Ez. If
c € €, then there is a finite subset €7, C € such that:
(i) cE Cp,
(ii) ife € Cpr, then € C F, where F, = U-ce,, ©, and
(iii) ifc € €g then F, C B.
Proof. We induct on the dimension of the cell. If c is a 0-cell, then
€r, = {c} works. Assume true for all cells of dimension < n and let c be an
n-cell. By condition (C), bd(e) is the union of finitely many cells c; each of
dimension <n. Define €p, to be {c}U U.ebd(c) fFe+ 4
(8.E.9) Theorem. Let X be a CW-complex with cell decomposition €.
Suppose €4 C X and A= Unce, ¢c. The following are equivalent:
(i) A is a CW-complex.
(ii) A is closed in X.
206 8. Quotient Spaces
(8.E.17) Examples.
1. Any CW-complex that is locally compact is locally finite (see the
next theorem).
2. As an example of a CW-complex that fails to be locally finite, let
I,, = I for each n € Z* and let X be the space obtained by identifying all
{0} € In together. Then X with the obvious cell structure is not locally
finite.
Locally finite CW-complexes are easily characterized.
In recent years, there has been considerable interest in quotient spaces result-
ing from what are called upper semicontinuous decompositions. For readers
Upper Semicontinuous Decompositions: an Introduction 209
(2, f(z))
(2,0)
The collection of segments connecting points (z, 0) with their image (2, f(x)),
together with individual points of €? that do not lie in any of these segments
yields a decomposition that is upper semicontinuous.
Geometrically, the preceding example of an upper semicontinuous de-
composition illustrates what is happening: little sets in the decomposition
may approach big ones, but the converse is not allowed (this will be made
more precise by theorem (8.F.10)). That such decompositions are by no
means rare is be seen from the following exercise.
(8.F.5) Examples—Exercises.
(a) Let X = €?. For each x € &! let g, = {(z,y) € E? : —co < y < oo},
and G = {g, : z € €'}. Show that G is usc and X/G is homeomorphic to
Ex:
(b) Suppose that X = €! and G is the usc decomposition of €1 whose
nondegenerate elements (i.e. those that are not singletons) consist of closures
of components of the complement in J of the Cantor set. Show that &} is
homeomorphic to €1/G.
(c) Suppose X = S! and the only nondegenerate element of a decom-
position G consists of the three vertices in S! of an equilateral triangle.
Describe X/G.
(d) Suppose X = S! and G is the collection of all triples s;, 82,83 € S!
that form the vertices of an equilateral triangle. Describe X/G.
(e) Suppose X = €° and the only nondegenerate element of G is the
unit circle. Are €° and X/G homeomorphic?
(f) Suppose X = €? and G consists of the origin, and all 2-spheres of
the form {(z,y,z) : 2? +y?+z* =r?} where 0 <r < oo. Describe X/G.
The notions of lim inf and lim sup are quite useful in the study of usc
decompositions.
7G
By (8.4.6), ®¢ is a homeomorphism. If A is a closed subset of X, then
P(A) = 67° f(A), which is closed in X/G. Hence P is a closed map and G
1S isc, © 4E)
The following rather amazing result shatters any notions that the reader
might have that monotone usc decompositions are a panacea for the problem
of inheritability.
PROBLEMS
Sections A and B
14. Show that the following spaces are not homeomorphic: Let A be the
cone over {(n,0) : n € Z*}; let B be the-union of {(n,0) : n € Zt}
and all points in the plane lying on line segments connecting points of
{(n,0) : n € Zt} with (0, 1), (with the relative topology from £7) [Hint:
Use first countability.]
15. Let X = (0,1) x [0, 1] and identify all points of the form (x, 1) with each
other. Show that the resulting quotient space is not homeomorphic to
Y, where Y is the union of the interior of the triangle with vertices
(0,0), (1,0), and (0,1), the open line segment {(z,0) : 0< a < 1}, and
the point (0,1) (with the relative topology from €”).
16. Identify the boundary points of a square so that the resulting quotient
space is not a 2-manifold.
Section C
Section D
Section E
Section F
CONTINUA
A. THE ARC
Our study of continua begins with an investigation of arcs. Recall that an
arc is defined to be a homeomorphism from the closed interval J into a space
X. However, here, as is often the case in mathematics, the definition is
not always easy to apply. The primary goal of this section is to obtain a
characterization arcs that is sometimes more convenient than the definition.
An arc is obviously a metric continuum, but then so are a host of other
spaces that are not arcs (as usual, the arc will be confused with its image
when it is convenient to do so). Intuitively, it seems that an arc is as “skinny”
as a continuum can be. However, circles, trees, etc., are just as slender. The
question, then, is how do these spaces differ from an arc? Notice that if any
point is removed from a circle, the resulting space is still connected. Further-
more, various points can be extracted from a tree without disconnecting it.
On the other hand, if any point is plucked from an arc (with the exception
of the two endpoints), the resulting space is not connected. Trivial as this
observation may be, it is the keystone of the arc. We shall work toward
221
222 9. Continua
proving that any metric continuum with exactly two “non-cut points” is an
arc. :
a tree
(9.A.5) Theorem. The separation order < is a linear order on S(a, b).
Proof. The proof is in two steps.
1. The separation order < is a partial order.
We prove only the transitivity of <. Antisymmetry is established by
Exercise (9.4.6). Suppose s; < sg and sz < 83. Only the trivial cases are
omitted if we assume 51, S2,s3 € S(a,b) \ {a,b} and s; # so # 83. By the
definition, there are separations U;,, Vs, of X \{s;} with a € Us, and b € V,,.
We want s3 € V;,. But by (9.A.3) V;, C Vs,. Hence s3 € V,, C Vay.
2. The separation order < is a linear order.
Suppose s1,52 € S(a,b). We must show either s; < sg or 82 < 3}.
Again, we ignore the trivial cases by assuming s1, 82 € S(a,b) \ {a,b}, and
Ss; # S2. Let (U,V) be a separation of X \ {si}, with a € U and be V.
Either s2 € V (in which case s; < s2 by the definition and we are finished),
or $2 € U (in which case corollary (9.A.4) gives so <s,). O
(9.A.6) Exercise. Show: if s1 < s2 and sg < $1, then s; = sg.
Since the separation order is a linear order, there are two natural topolo-
gies for S(a,b): the order topology and the relative topology from X. In
general, these topologies need not be the same, but if we start with a T; con-
tinuum having exactly two non-cut points, then the two topologies coincide.
(ii) p € Ve, MVz, so the connected set Vz, U {ra} is contained in V;,
and hence, 2q € V;.
Consequently, U, U {z} is contained in Uz, UVz,, and since z € U;z,, it
follows that U, is a proper subset of Uz, for each tq. This contradicts
the maximality of the nest W. Thus (){Uz,, : a € A} = 9, and furthermore
{Uz,,U{ta} : a € A} = 0 (why?). Therefore, it follows that {Vz, : a € A}
is an open cover of the compact set X that has no finite subcover (why’),
which is impossible. Hence, we conclude that U must have a non-cut point.
The argument for V is similar. O
Proof of 9.4.7. It is obvious that X = S(a,b), since if p € X \ {a,b},
then p is a cut point and thus there is a separation (U, V) of X \{p}. Lemma
(9.4.8) (together with the hypothesis of the theorem) guarantees that a and
b are separated by p, and hence p € S(a,)).
Let 7 denote the given topology for X, and let F denote the order
topology. To show that F C 7, it is sufficient to prove that sets of the form
West aca <p} and W = {x : @ > p} are in JT. Suppose p € X \ {a, db}.
Then there is a separation (Up, Vp) of X \ {p} such that a € U, and b € Vp.
We show that W = Up, and a similar argument may be used to show that
W = Vp.
Suppose z € W. If x € Vp, then (by the definition of the separation
order) z > p. This contradicts c € W; hence we have W C Up.
Now suppose z € U,. By (9.A.3) this implies that p € V, which by
definition says z < p. Hence x € W, and U, = W.
We now show 7 C F. Let O be an element of the given topology T
and x € O. It suffices to find an (<)-interval (c,d) such that x € (c,d) C O.
If no such interval exists, then the collection {[p,q]N (X \O) : x € (p,q)}
has the finite intersection property. Since X is compact, the intersection
of all the members of this family is nonempty and is contained in X \ O.
(Note: [p,q] is closed, since F C 7.) However, it is easy to show that
(\[p,q] : « € (p,q)} = {x} C O). This should strike the reader as being
a bit unusual. Therefore, the desired open interval exists, and consequently
T CF. (The pedant should check the cases t =a, andzr=b.) O
Now that the candidate for arc status has been successfully ordered, the
task remains of defining a homeomorphism from K onto J. The next lemma
gives us a way to start such a map.
Proof. Let A = {a,,@2,...} and define h(a,) = 1/2. Let n; be the first
integer such that an, < a; and nz be the first integer such that a; < dn,.
Define h(an,) to be 1/2? and h(an,) to be 3/2?. Et cetera. O
(9.A.10) Theorem. Suppose X is a metric continuum with exactly two
non-cut points a and b. Then X is homeomorphic to J.
Proof. By (2.1.13), there is a countable dense subset A of X\{a, b}. Since
K is connected, it follows that A (ordered by the separation order) satisfies
the two conditions of the hypothesis of the preceding lemma. Consequently,
there is a function h mapping A onto D that preserves order. Let h(a) = 0
and h(b) = 1. Suppose p € X \ (AU {a,b}); we must define h(p). Since p is
a cut point, there is a separation (Up, V,) of X \ {p}, such that a € U, and
bE V,. Note that 0 < sup{h(x) : x € ANU,} < inf{h(z) : c € ANV,} <1,
and an elementary argument yields sup{h(r) : x € ANU,} = inf{h(z) :
xz € ANV,}. Define h(p) to be this common value. It is not difficult to check
that h is the desired homeomorphism. 0
B. PEANO CONTINUA
We now turn our attention to a major class of continua, the Peano continua.
A Peano continuum is a locally connected metric continuum. Arcs are Peano
226 9. Continua
continua, but hardly representative of them. The result toward which we are
heading is the venerable (1913) Hahn-Mazurkiewicz theorem, which states
that every Peano continuum is a continuous image of the arc J. Thus there is
a continuous map from J onto the n-dimensional cube, or even more amusing,
there is a continuous function mapping J onto the Hilbert cube. The reader
whose intuition concerning continuous functions has suddenly taken a jolt is
in good company. Peano’s amazing discovery in 1890 that J could be mapped
continuously onto the unit square created havoc in the mathematical world
(but scarcely anywhere else, needless to say). Among the major casualties
brought about by Peano’s result was the then existing concept of dimension.
As a first step toward proving the Hahn-Mazurkiewicz theorem, we will
show that Peano continua are arc connected, an interesting result in itself.
The proof is based on the notion of simple chains. Recall from chapter 2: if
a and b are points in a topological space X, then a simple chain of sets from
a to 6 is a finite collection of subsets {U;,U2,...,Un} of X, where a € Uj,
bE U, and U;NU; # O if and only
if |i — j| < 1.
(9.B.1) Theorem. If X is a Peano continuum, then X is arc connected.
Proof. Let a and b be distinct points in X (if we can not find distinct
points in X, it is arc connected by default). It is necessary to construct an
arc from a to b.
We will construct a sequence of simple chains from a to b, Ci,C2,... each
composed of connected open links such that
(2) HU;,. 1s adlink of C,, then diam U;,;< 1/2".
(it) If Un, is a link of Cy, then ee CUneig 10F Some Unt © tnt
(Hence, U,,; can not intersect Un_1¢ if€<j—loré>j+1)]
(iit) If Un C Un-1,j, then no U,, for k > 7 has a point in common
with Un_-1i,m form < j — 2, and no U,,, for k < i has a point in
common with Un_-1m for m > 7 +2.
Suppose such a family of simple chains has been constructed. For n > 1
define
Ch = U Uni
j-2
Dn = (J Uni and En = (J Uns.
A=1 i>j+2
Now divide each of the nine squares into nine more squares and partition I
into 81 pieces. Define a map f2 from J into I x J by mapping each segment
in succession onto a diagonal of the appropriate square.
230 9. Continua
A pattern has begun to emerge which the reader should be able to continue.
(9.B.5) Exercise. Show that the sequence of functions f,, fo,... converges
uniformly to a (continuous) function f : ] + X. Note: The pointwise limits
{fn(z)} converge for each z, and {f,} is a Cauchy sequence in B(J,I x I)
with the sup metric.
The second step consists of the removal of the interiors of eight more
squares, and the third step 64 such interiors are eliminated, etc. That which
Chainable Continua 231
remains after the removal of all these squares is called Sierpinski’s universal
plane curve.
If instead of removing squares from the unit square, one removes rect-
angular prisms (in a somewhat different fashion, Engelking [1978]) from the
unit cube, then the resulting space is called the universal curve. The uni-
versal curve is a l-dimensional Peano continuum with the property that
any 1-dimensional continuum can be embedded in it. Interestingly, it can
be shown that simple closed curves and the universal curve are the only
1-dimensional Peano continua that are homogeneous, Anderson [1958].
C. CHAINABLE CONTINUA
We have already seen the wide applicability of arguments using simple chains.
Our study now focuses on spaces that may be defined in terms of simple
chains.
(9.C.1) Definition. A simple chain C of open sets in a metric space is
called an e-chain if and only if each link of C has diameter less than e.
(9.C.3) Examples.
1. The unit interval I is chainable from 0 to 1;
2. {0,1} with the relative topology is not chainable;
3. (0,1) (with the usual metric) is chainable, but not from a to b for
any a,b € (0,1);
4. €' with the usual metric is not chainable, even though it is homeo-
morphic to (0, 1).
5. The topologist’s sine curve is chainable.
6. The rationals in [0, 1] are chainable from 0 to 1.
In spite of examples 2 and 3, chainability does have some hereditary
traits.
232 9. Continua
The converse of the preceding exercise is false, as may be seen from our
semiuniversal example, the topologist’s sine curve X.
234 9. Continua
The space X is irreducible from a to b but not chainable between these points
(although X is chainable between a and c).
The next theorem shows that irreducible continua are quite common.
It is easy to see that arcs are unicoherent and circles are not. It is not
so obvious what the status of €”, disks, spheres, etc., should be. The next
result throws little light on this problem, although it does yield a number of
examples of unicoherent spaces.
union of two disjoint closed sets, either FrC* C FrD; or FrC* C Fr Da, say
FrC* c FrD,. However, z € Dy) NC* and C Cc Di N(X \ C*). Therefore,
D,NFrC* # O and it follows that D, N FrD; # 9, which contradicts the
openness of D;. O
In the problem section, the reader is asked to prove the converse of this
theorem.
a
eeOO
e Oxc es
WAT
t
AY CBO) rier
ee
Bs,
a te esow
unt
Pt
oe Ast
Loman
IAI
CRTC ATU Bee WY Va.NUR DANGER DXi>)
a nn as *
240 9. Continua
(9.D.15) Exercise. Show that if p;(A) C #;(B) for infinitely many i, then
pi(A) C p;(B) for all i.
(9.D.16) Exercise. Finish the proof of the previous theorem by showing
that AC Bor BCA.
The Pseudo-Arc 241
E. THE PSEUDO-ARC
aS
CORAL AORN
= KOO BOOOOOOR OOO
(2COOP MOS OOM OOIOOP SOO).
=) b
CBSE
RRRRRIED (a)
HOTTIN
& (=) 5
LPN OTTO TN
COODIOROOOOOOMOOOIOOOR
OOS
& =SS
iS)
—
COO SOOTHING)
Cr C3 C3 C4 C3
A broken line
PROBLEMS
‘
Section A
Section B
Find lim sup and lim inf of the following sequence of subsets of the plane
AioAs Ag
Section C
Sections D and E
8. Let K be the Cantor set. Let So be the union of all semicircles in the
upper half-plane with both endpoints in K and such that the endpoints
are symmetric with respect to the line = 1/2. For i = 1,2,..., let
S; be the union of all semicircles in the lower half-plane whose ends are
on K and are symmetric with respect to the line z = 5/(3'- 2). Let
248 9. Continua
So
S3
S2
Sy
Show that the plane contains uncountably may pairwise disjoint non-
degenerate continua, none of which contains an arc. [Hint: Use the
pseudo-arc.]
10. Construct the first two stages of a pseudo-arc starting with six links in
the initial chain.
11. Show that every topological space X that is a nested intersection, i.e.,
co
PARACOMPACTNESS AND
AND METRIZABILITY
A. PARACOMPACTNESS
(10.A.1) Definition. A family C of subsets of a space X is locally finite if
and only if each xz € X lies in a neighborhood that intersects at most a finite
number of members of C.
As an easy but important application of this definition, we claim that
the union of a locally finite family of closed sets is closed.
249
250 10. Paracompactness and Metrizability
(10.A.2) Exercise.
(a) Suppose {Aq : a € A} is a locally finite collection of closed subsets
of a space X. Show that U,¢, Aa is closed.
(b) Suppose {Aq : a € A} is a locally finite collection of subsets of a
space X. Show that {Aq : a € A} is locally finite.
The simplest type of cover refinement is defined next.
For each & € X, let F, = U{Va : x ¢ Va}, and note once again
that by (10.A.2) F, is closed. Finally, set Z, equal to W, \ F,. Then
Z={Z, : c € X} isa barycentric refinement of YU. O
We note in passing that the converse of this theorem also holds, but do
not belabor (or prove) the point.
The following inequality might well be more than a frivolous conjecture:
card(known mathematicians) < card(known varieties of compactness).
Certainly, €” is o-compact for all n, and equally obvious is the fact that
an uncountable discrete metric space is not o-compact. Two basic properties
of o-compact spaces are given next.
B. PARTITIONS OF UNITY
We show that Vj, V2,... is a development for X and satisfies the hy-
pothesis of the Aleksandrov-Uryson Theorem. Suppose x € X and i € Zt.
To establish that the V;’s form a development, an integer k € Z+ must be
found such that St(z,V,) C V;7. Let k = aZ,, and suppose that z € V,’.
Since ay > k = az,,, we have V,’ = ny C Vys_,- Therefore, x € Uae,, ©
Uge = Vi? (a7; = n(x, a7)). Thus, Vi, V2,... is a eau @ X.
Now suppose Vi, V,4., # 0, or equivalently, OFC aaz) ae a¥) au:
Then if n(z,a?) < n(y,a¥), we have U?niga) Oe a?) 4 0, which by hy-
pothesis implies V,., = U%,z GUSaay Uzaon Ve: An analogous ar-
gument results in a similar Poreieon ifn(y,a/) < n(z,a?). Hence, by the
Aleksandrov-Uryson theorem, X is metrizable. The converse is straightfor-
ward and left as a problem (see Problem C-4). O
The following question is of considerable interest. Suppose X is metriz-
able and f : X > Y is continuous and onto. Is Y metrizable? Stone used
Frink’s result to obtain a particularly elegant solution to this problem.
(10.C.11) Remark. It follows from the proof of (10.A.14) that any para-
compact space is fully normal. Stone [1959] showed that T2 fully normal
spaces are paracompact.
D. MOORE SPACES
There is continuing interest in investigating an old concept: Moore spaces. In
fact one can argue that Moore spaces have helped give rise to a whole branch
of topology—set theoretic topology. Moore spaces are a generalization of
metric spaces.
— metrizable
(10.C.12)
Jones [1937] showed that every separable normal Moore space is metrizable,
however the general conjecture has deep set theory ramifications. It is nev-
ertheless in some sense settled. Quoting from van Mill and Reed [1990], “We
now know that the normal Moore space conjecture is in the hands of the
gods and large cardinals (see Tall [1984] and Fleissner [1984]).” We next
give an example of a Moore space that is not metrizable.
The space to be constructed is connected and locally Euclidean (if it
were metrizable, it would be a 2-manifold). In €? start with a Cantor set on
the x axis of ee and let Aj; Ai, Aj; Aiii, Aj12, Aj21,; Aj22;3 ... denote the
“middle thirds.”
We lay out a network of roads from the point (3, 5 ), all of which are
directed toward the Cantor set in such a way that every point of the Cantor
set is a limit of precisely one route and each route eventually “arrives” at
precisely one point of the Cantor set.
Moore Spaces 261
From each point of the Cantor set, extend a ray in the negative y di-
rection parallel to the y axis, as indicated in the preceding figure. Let M be
the union of these rays and the road system, and let N = M x {-1, 1].
To obtain a space that is locally €?, something must be done about the
branch points. Consider a cylinder about a typical branch point p. This
“neighborhood” of p is replaced by a hexagon. The sides @), £2, £3 are glued
to £), £4 and £5 respectively, and thus the “fat Y’s” around a branch point are
replaced by a twisted two-dimensional hexagon. This procedure is repeated
for each branch point. Let X be the space derived from N in this manner.
What is the topology for X? For points outside K x [-—1, 1], the usual relative
262 10. Paracompactness and Metrizability
plane topology is used, i.e., disks form a basis at each point. Suppose w €
K x [-1,1], and say w = (z,0). A typical neighborhood for w will have the
following form. Let € > 0 be given. Starting at a point a distance € from the
Cantor set, we select the route in X that leads to w: Near a point (which is
not a branch point), we have the following situation (a)
(a)
It is easy to see that the two halves form a whole disk (without boundary).
(10.D.2) Exercise. Show that X with the topology described above is con-
nected (even arc connected), regular, T2, and each point has a neighborhood
homeomorphic to €?.
Completion of Metric Spaces 263
Step 6. Note that 0(X) is dense in ae since if [{s,,}] € GC. then for large
k we have that 6(s;) is “close to” {sp}.
Proof. The trick is exactly the same as that used in the previous theo-
rem: the set A is embedded as a closed subset of a complete metric space.
Since A is a Gs subset of X, A may be written as an intersection of open
subsets of X, say A = ee U,. By the previous theorem, each U,, is
topologically complete, and hence by (6.A.17) the product ie U, is a
complete metric space. Embed A in []*°_, Un by »: A > [°° Un, where
w(a) = (a,a,...).
Again, it is not unreasonable to ask the reader to check
that A and (A) are homeomorphic and 7(A) is closed in []°°., Un. O
If one can complete metric spaces, can one “complete” Moore spaces?
The question is meaningless unless a reasonable definition can be given for
“complete Moore space.” Theorem (3.B.8) is used to motivate the following
definitions.
Problems 265
PROBLEMS
Section A
Section B
1. Show that a T> space X is normal if for each locally finite cover U/, there
is a partition of unity subordinate to U.
2. Suppose X is paracompact andU = {U, : a € A} isa locally finite open
cover of X. Show: if w : A — (0,00) is a function, then there is a map
f : X — (0,00) such that for each « € X, f(x) < sup{(a) : c € Ug}.
3. If {U,,U2,...,Un} is finite open cover of a normal space X, show there
are continuous functions fj, fo,..., fn such that:
(alesis X= [0,1 doreach tHal,.2;
5 am
(bi 4a: fy se Ohew@ for esen a= Lite gad
(©) Shy AG) = 1.
(Hint: Use (4.C.2) and (4.B.1).]
4. Use Problem 3 to show that every compact n-manifold M can be embed-
ded in €™ for some m € Z*. [Hint: Find an open cover {U;,U2,...,U:}
of M such that for each 7, there is a homeomorphism g; : U; > €”. Find
continuous functions fj, fo,..., f¢ as in Problem 3, and fori = 1,2,...,t
define
Section C
Section D
plane one unit below the midpoint of A;. Draw segments emanating
from a; that go half way toward the centers of Ai; and Ajo. Let ay
268 10. Paracompactness and Metrizability
a1
\v
Section E
lt mone @
Chapter 11
Let f : Zt > X \ {(0,0)} be any onto map, and set 2, = f(n) for each
n € Zt. It is clear that (0,0) is a cluster point of the sequence {z,}. We
show: no subsequence of {x,,} converges to (0,0).
271
272 11. Nets and Filters
(11.A.6) Exercise. Show that the space X of the above example is T> and
Lindel6of.
B. NETS
The following example points out that sequences are special cases of
nets.
Nets 273
As was the case with sequences, the most important property associated
with nets is convergence.
C. FILTERS
The concept of filter, while seemingly far removed from the idea of se-
quence, nevertheless provides an important and useful theory of convergence.
We have seen that neighborhoods of a point are critical in determining con-
vergence at the point. Filters may be considered as an abstraction of the
following two properties of these neighborhoods:
(i) the intersection of two neighborhoods is again a neighborhood, and
(ii) if G is a neighborhood of z, then any set containing G is also a
neighborhood.
(11.C.3) Examples.
1. A neighborhood base at a point x € X is a filter, called the neigh-
borhood filter at zx.
2. Let N: D— X beanetin X. Then F={ACX : N is eventually
in A} is a filter.
3. The family B = {(a,0o) : a € R'} is a base for a filter.
4. If X is an infinite set, then the family of all sets with finite comple-
ments is a filter in X.
(11.C.4) Definition. Suppose F is a filter in a set X. We say F is even-
tually in a subset A C X if and only if A € F. A filter F in a topological
space X converges to a point x € X if and only if F is eventually in each
neighborhood of z.
Thus one might picture such a filter as funneling down towards z.
Suppose F is a filter in a set X and f : X — Y is a function from X
into an arbitrary set Y. Although f(F) ={f(F) : F € F} is not necessarily
a filter (why?), it is at least a base for a filter {G CY : f(F) CG for some
F € F}. Abusing notation slightly, we also denote this filter by f(F).
276 11. Nets and Filters
The next theorem gives us one of the more peculiar properties of ultra-
filters.
Nets were first introduced by E.H.Moore and H.L.Smith [1922], and refine-
ments of the theory were added by Kelley [1950a]. The French, exhibiting
their customary enthusiasm for American inspirations, have demonstrated a
marked predilection for filters. The concept of the filter found its origin in
the work of the noted French mathematician Cartan [1937]. The purpose of
this section is to underscore the futility of any filter vs. net contest by show-
ing that nets arise from filters in a most natural way, and, no less naturally,
278 11. Nets and Filters
filters may be derived from nets. Our personal feeling is that nets are “more
intuitive,” but filters are easier to work with, often yielding elegant proofs.
Suppose N = {zra}eep is a net in a set X. For each a € D, let
Tx = {tg : B > a}. Then it is easy to show that {T, : a € D} has the
finite intersection property and therefore is a base for a unique filter, which
we denote by Fy.
(11.D.1) Theorem. A net N is eventually in a set A C X if and only
if Fn is eventually in A. Consequently, if X is a topological space, and N
converges to € X, so must Fy.
Proof. If N is eventually in A, there is an a € D such that for 6 > a,
zg € A. Thus, T, C A, and hence A € Fy.
If Fy is eventually in A, then A € Fy, which implies there is an a such
that T, C A. Hence, for all G > a, we have rg € Ty C A, ie., {xq} is
eventually in A. O
PROBLEMS
Sections A and B
Section C
Section D
aa oe enel near
= . =, D4 Fed coe
D ik ,
7 P ag ihe otet ee
-_ > Fe 6. «4 @ierce 7, ae
; 7
iv = = an £¢ a
DPA ot WSN 0%
a - way st _
ie vo tani eonly¢ wren
y 04 “ —
ot
.
THE ALGEBRAIZATION OF
TOPOLOGY
283
284 12. The Algebraization of Topology
ve
Fy (A)=f(A)=9(A)
F\()=9(z)
Fo(#)=f(z)
2 for0
(f *g9)(t) = e : : t
g(2t—1) for 2
_ f f(2t) if0<t<}3
enO= er font <1
g(1)
fits) faig(*y)
Ee, Gig Saye aoteg)
1)=g9(0
(12.A.11) Exercise. State and prove the analog of (12.A.9) for matching
paths.
Yo if0<t<—2s+1
H(s,t) = oe
Ti ) iif —2 Paid. ie
Sitsol.
: f
(1/2,0) :
Note that Ho = c,, +f, Hi = f, and H(0,t) = H(1,t) = yo for all ¢ € I.
A similar homotopy demonstrates that ioe Pens GG
(12.A4.14) Theorem. Suppose f,g,h € £L. Then [f]((9][h]) = ([F][g]) [A],
and thus (G,-) is associative.
Proof. It suffices to show that f * (g * ee (f *g)*h. Define
H(s,t) = 4 g(4s
—-2 +1) for
4s — t --
(FA) cs ae
Me
The Fundamental Group (The First Homotopy Group) 289
t
(1/4,1) (1/2,1)
s
(1/2,0) (3/4,0)
(12.A.15) Exercise. State and prove analogs of the previous two theorems
when f, g, and h are paths that match properly.
Finally, inverses for elements of (G,-) must be found. This is done with
the aid of the well-known principle that if you walk down a path in one
direction and subsequently retrace your steps, then you might as well not
have left in the first place.
t
Yo TOU S08 Sa,
t 1
1a f (2s — t) fot SiS
(s,t) = 1 a
{' (28+%t—1), for =< s< ——
2 2
2-t
Yo for igo
290 12. The Algebraization of Topology
f (2s) f’ (2s—1) ;
Note that Ho = (f * f") and H; = c,,. A similar figure and homotopy can
be given to establish that f” * FO) Cv0- Oo
Proof (outline). For each loop in X x Y with base point (xo, yo), corre-
sponding loops must be located in X and Y. This is accomplished via the
projection maps px and py, which send X x Y onto X and Y, respectively.
If f : I + X xY is a loop in X x Y, then pxf and pyf are loops in X
and Y. Define a function w : ™(X x Y,(z0,yo)) ™(X,20) x ™(Y, yo)
by #([f]) = ([pxf], [py f]). It is not difficult to verify that this map is well-
defined.
292 12. The Algebraization of Topology
O0<t<i
k(t) = rete.
(xo, fo(2t-—1)) $<t<l
is mapped onto ((f1],[fo]). 0
The fundamental groups of some rather simple topological spaces are
calculated in the next section. Before succumbing to such formalities, how-
ever, we attempt to discover in an intuitive fashion the fundamental group of
S!. (Two complete but distinct proofs of the fact that 71(S1,
sq) is infinite
cyclic are given in Chapter 13 and 14.)
A fairly obvious candidate for a generator of 1(S1, sq) is the equivalence
class [a] determined by the loop shown in the next figure, where a wraps
once around S! in the direction indicated.
(12.C.3) Examples.
1. Homeomorphic spaces are clearly homotopically equivalent.
2. Suppose D is the unit disk minus its center and E = S'. Let
f:E—- D be the inclusion map, and g: D > E be the obvious projection
map. Then it is easy to see that gf = idg and fg ~ idp; hence, D and E
are homotopically equivalent, and f and g are homotopy equivalencies.
We are now able to prove a result that implies that homeomorphic spaces
have isomorphic fundamental groups, or equivalently, if the fundamental
groups of two spaces differ, the spaces cannot be homeomorphic.
(12.C.4) Theorem. If pointed spaces (X, xo) and (Y, yo) are homotopically
equivalent, then 7(X,z0) and 71(Y, yo) are isomorphic (cf. Problem C 17).
Proof. Suppose f : (X,20) > (Y,yo) and g : (Y,yo) > (X,2o) are
homotopy inverses. Then from (12.C.1) parts (c), (d) and (e) we obtain
294 12. The Algebraization of Topology
PCP = (fg)« = idx s(Y,yo) and Oxf = (of)« = idx, (X,x0)- Consequently,
In (12.C.7), the contraction homotopy does not move Zo. This is not
always the case as may be seen in the next example.
(12.C.9) Theorem.
(i) Contractible spaces are path connected.
(ii) If X is contractible, Y is a topological space, and f,g: Y ~ X are
continuous, then f and g are homotopic.
(tit) A space X is contractible if and only if X is homotopically equiva-
lent to a point.
Proof. (i) Suppose that X is contractible to zo, and that 2; € X. Let
H : X x I + X be a homotopy such that Hp = idx and H, = c,, where
Ce is defined by cz,(z) = 2. Then f : I + X defined by f(t) = H(21,t)
is a path from x; to zo. It now follows easily from (2.C.3) that X is path
connected.
(ii) Let H : X x I -+ X be the homotopy given by part (i). Define
w:YxI-+X by wly,t) = A(f(y),t). Then ~ is a homotopy between f
and c,,. All maps from Y into X are homotopic to the constant function
Cz,, and hence to each other.
(177) Suppose that X is contractible to x9 € X. Let g: {ao} + X be
the inclusion map and let f : X — {zo} be the constant function. Then
gf ~ idx and fg =id;,,}. Thus, X and {x9} are homotopically equivalent.
Continuous Functions and Homomorphisms 295
B?
(12.C.20) Examples.
1. The fundamental group of an annulus is isomorphic to that of a circle.
2. The fundamental group of a Moebius strip is isomorphic to that of a
circle.
Categories and functors form the essential vocabulary for the study of
homology theory (which is treated in courses in algebraic topology). We will
find this terminology useful in Section F as well as in Chapter 14.
In Section B, we saw that the homotopy functor sends finite products
of topological spaces to direct products of groups. In Section F, we shall
see that direct limits are also fairly well behaved. It will also be shown that
wedge products of topological spaces (disjoint unions of pointed spaces with
the base points identified) are sent to free products of groups.
One of the advantages of studying categories and functors is that one
can recognize the similarity of structures in the different categories, e.g.,
products of topological spaces vs. direct products of groups, wedge products
of topological spaces vs. free products of groups, etc. Once the corresponding
structures are identified, then it becomes natural to ask which structures are
preserved by a given functor. Thus, category theory may be employed to
help bring into focus a broad range of natural questions concerning functors
between pairs of categories.
Before reading this section, the reader should be (or become) familiar with
Sections A and B of the Appendix. Difficulties encountered in calculating
fundamental groups soon converge on the insurmountable. However, some
relief is afforded by the Seifert-van Kampen Theorem, two versions of which
are given below. These results were established independently by Seifert
[1931] and van Kampen [1933] and were generalized by Olum [1958]. The
main emphasis of this section is to make the statement of the theorem plau-
sible, and to show how the theorem may be applied. The proof itself is
omitted.
300 12. The Algebraization of Topology
a B
How are loops in X accounted for that do not lie completely in either
X, of X2: for instance, the loop a in the next figure?
The basic procedure followed is to break a into paths a;, each of which
lies entirely in Xo, X, or Xo.
With all this in mind, it should seem plausible that G is the “amalga-
mated” product of G; and Go, i.e., the free product of G; and G2 modulo
the relations mentioned previously. More precisely, suppose that {A; Ra}
and {B; Rg} are presentations of G; and G2, respectively. Then we have
G = {A,B;Ra, Re, 91([a])02([o])~* where [a] € Go}. The following is the
classic formulation of the Seifert-van Kampen theorem.
302 12. The Algebraization of Topology
Xy X2
where G = 7(X, 20), the G;’s are the respective fundamental groups of the
X;’s, and the indicated homomorphisms are those induced by the inclusion
mappings.
Then
(i) wi (G1) Uwe(G2) generates G, and
(ii) If H is an arbitrary group and y; : G; ~ H, i = 0, 1, 2 are homo-
morphisms satisfying W = ~16, = W262, then there is a unique homomor-
phism A: G > H such that w = Aw;, 1 = 0, 1, 2.
There are a number of corollaries that follow rather easily from the main
result.
Q@
Q@
As usual, the first step consists in finding suitable open subsets. Let c denote
the center of the disk and let xo be any other point in the interior of the disk.
Let X; = P2 \ {c} and let X2 be the interior of the disk (considered as a
subset of P2). Note that after identification has taken place, a is a circle, and
furthermore there is a deformation retract from X, onto a. Therefore, G; =
™(X 1,20) = Z, and G; is generated by the equivalence class represented by
y = B’ xaxB. The group 7 (X9, xo) is trivial and 7(X1N X2, 20) is infinite
cyclic with generator [5]. Note that @,([5]) = [7]?. Therefore, by (12.E.5),
™(P2,%0) = ™1(X1,20)/K, where K is the normal subgroup generated by
[y]?. Thus 7 (P2, 20) is cyclic of order 2.
Direct Limits 305
F. DIRECT LIMITS
In this section, direct systems and direct limits of topological spaces are
defined and characterized by a “universal mapping property.” The homotopy
functor will transform a direct system of pointed topological spaces into a
direct system of groups (see Section D of the Appendix). Amazingly, (with
some reasonable conditions on the topological spaces) the fundamental group
of the direct limit of the topological spaces is isomorphic to the direct limit of
the fundamental groups. This result will be indispensable when we consider
the examples in section E of chapter 15.
xX@&
Then
(i) fa is continuous,
(ii) for each a, 8 € D witha < £ the following diagram is commutative,
xX”
f
fi
Xo faa Xp
ow
co
pen Ga Y
fap ha.
Xe
xo E Y
DEE
14
Xo
a fay Xy
bas
Xa ;Jar XB
Suppose further for each topological space Z, for each family of continuous
functions {hg : X_q + Z : a € D} with the property that the following
diagrams are commutative
es
Xo fap Xp
DEES
h Y oxo h
Direct Limits 309
The reader will want to reread the entire section in order to observe that
nothing goes awry if one assumes that all topological spaces are endowed
with a base point, and that all continuous functions preserve base points. To .
nudge the less than diligent reader, we present the following exercise.
(12.F.9) Exercise. Restate and prove the analog of (12.F.7) for the cate-
gory of pointed topological spaces.
The reader should now check the Appendix and observe that the analog
of each of the preceding theorems also holds in the category of groups and
homomorphisms.
Next we examine what is produced when the homotopy functor operates
on a direct system.
ne
(ga)» (fp)«
us OG ro)
(fa) (fa).
Thus, if we can show ™(X, 20) C Ugen(fa)*(™1 (Xa; %0)), then there will
be an epimorphism h : iim (m7 (Xq,20)) 4 ™m (XS, ro) by Theorem (Ap.D.2).
Let [€] € m(X™,20). Since {fa(Xa) : @ € D} forms an open cover
of €(I), there are a finite number of f.(Xq)’s that cover (J). Let 6 be
an index larger than all of the a@’s in the finite cover. Then we have that
€(I) C fg(Xg). Since the fag’s are inclusion maps, fg is one to one. Hence,
we may define a path m = fz 1¢. Then (fg)«[m] equals [é].
We now show h is one to one. Let x € iim(m (Xq,2o)) and A(x) = e;
we show that x = e. Since fim(m7 (Xa, 20)) C Usep(Ga)x
(71 (Xa, Zo)), there
isanaé€ Danday € ™(Xqa,2Zo) such that (ga).(y) = 2.
The diagram
iim (m4 (Xa, 20) > ™ (X®, 20)
(Ga)« (fa)
(Xe, Zo)
is commutative, and therefore (fa).(y) = e. Now let wu: J + Xq be a loop
based at xo such that [y~] = y. Since (fa).(y) = e, there is a homotopy
H:IxTI - (X®,%9) such that Ho = fi. and, = ¢,. Let Bybe
sufficiently large so that Im H C fg(Xg) (and a < 8). As before, fg is one
to one, and hence fz 'H is a homotopy between fag’ and Cz). Therefore,
(fas)«(w) =e. By the commutativity of the following diagram,
PROBLEMS
Sections A and B
1. Show: If f,g: X - Y are homotopic and A C_X, then f\4 and g|, are
homotopic.
2. Find an example of a space X containing points xo and 2; such that
7™(X,29) and 71(X,21) are not isomorphic.
Problems 311
Let X and Y be topological spaces and suppose C(X, Y) has the compact-
open topology.
Section C
™(Y,
Go)
(See the proof of (12.B.2) for the definition of ky.)
10. Show: If A is a deformation retract of X and B is a deformation retract
of A, then B is a deformation retract of X.
11. Show that if A, B, and AN B are nonempty, open, path connected
subsets of AUB, and if A and B are simply connected, then so is AUB.
12. Use problem 11 to show 7™(S?) = {e}.
13. Let n and s be the north and south poles of an n-sphere. Prove {n, s}
is not a retract of S”.
Problems 313
Section D
1. Find (somewhere in mathematics) an example of a contravariant functor.
2. (The Higher Homotopy Groups) Suppose X is a topological space. Let
OI” = {(a1,...,2n) € I” : for somei, 2; = 0 or 1}, and let Qn(X, 20) =
C((I", 01”), (X, z0)). If T58 € Qn(X, xo), define
i (2riya55. ve ay) Ty
Che Oar. e.' iia) =
g(22, mam Ml fei oie an)
NIFTA
© A. Ty IA
IA Nile
a
(c) Show that for each n > 1, the construction above gives rise to
a functor from Zp to the category of abelian groups and homo-
morphisms. (It should be no surprise that this is called the n-th
homotopy functor.)
(d) Show that ifn > 1, then t,(X xY, (xo, yo)) & t(X, Lo) Xtn(Y, yo).
(e) Show that if (X,2o) and (Y, yo) are homotopically equivalent, then
T(X, Zo) and 7,(Y, yo) are isomorphic.
Remark: In general, it is exceedingly difficult to compute the higher homo-
topy groups. In fact, their computation even for n-spheres is still not fully
done. Some of the known results leave one gasping, e.g., 715(S°) = Z x Z120
and 716(S®) = Z x Z2 x Zo X Ze (Zn is the group of integers mod n).
3. Suppose (€, #) is a category, f € Home(A, B), and g € Home(B, A).
Then f is a right inverse of g if and only if go f = 7zd,, in which case g
is a called a left inverse of f.
(a) Show that in the category of sets (hence also in the category of
topological spaces and continuous functions), if f has a left inverse,
then f is 1-1, and if f has a right inverse, then f is onto.
A map f is called a €-equivalence if and only if f has both a left and a right
inverse.
(b) Show that if f has a left inverse g and a right inverse h, then g = h.
(c) Show that a homotopy equivalence is a €-equivalence, where € is
the homotopy category.
4. Define a category as follows: Let 9 = Zt be the class of objects. If
m,n € O, define Homs(m,n) = {(m,a,n) : a is a common divisor of
m and n}. Define (n,b,q) o(m,a,n) = (m,d,q), where d is the greatest
common divisor of a and b. Show that (0,8) is a category, where
§ = {Homp(m,n) : n,m Ee OD}.
Section E
ce
Calculate the fundamental group of the five-petaled rose:
Section F
COVERING SPACES
The beautiful and surprisingly close interplay between algebra and topology
is well illustrated in the theory of covering spaces. This theory is very rem-
iniscent of Galois theory and has numerous applications in both topology
and geometry. In particular, some of the results may be used to obtain an
elegant computation of the fundamental group of S'. Throughout this chap-
ter the reader is especially advised to construct numerous figures illustrating
the definitions as well as the hypotheses and proofs of theorems.
317
318 13. Covering Spaces
(13.A.7) Exercise. Find three distinct covering spaces of the torus (one of
which is the plane).
The following theorem is one of the most useful results in covering space
theory. It states that a path a lying in the base space X may be “lifted in
a unique fashion” to a path in the covering space. Here, “lifted” means that
the projection of the lifted path coincides with a, and “unique” means that
if pa = pa and a(0) = a(0), then a = a.
(13.4.8) Theorem (Unique Path Lifting Theorem). If (X,p) is a
covering space of X, a: I + X is a path in X with initial point zo, and
#9 € p—'(ao), then there is a unique path @: I > X with initial point Zo
and pa@ =a.
Proof. The idea of the proof is quite simple. We subdivide J finely
enough so that each piece of the subdivision is mapped by a into a canonical
neighborhood, and then use the local homeomorphism p to lift the pieces.
Some care is taken in order to ensure that the appropriate endpoint of the
(7 + 1)-st piece matches up properly with a suitable endpoint of the (i)-th
piece.
Cover X with a family U = {U, : x € X} of canonical neighborhoods.
Then a~!(U/) is an open cover of I. Since I is compact, there is a Lebesgue
number y for this cover. Let 0 = to < ti,:-: < tn = 1 be a partition of
I of mesh less than y. The following assertion is proven by induction: For
The Lifting Theorems 319
each 7, 0 <7 < n, there is a unique continuous map 4; : [0, t;] X such
that @;(0) = Zo and pd; = aljo.,;. Obviously, for i = 0, the map Qp exists.
Assume then we have constructed such a map for some i, where i < n. By
the choice of the partition, we know that a((t;,t:41]) is contained in some
U €U. Let S; be the component of p~'(U) that contains 4;(t;). Since
Pj = ps, is an embedding, it follows that p;' : U + S; is a continuous
function. Consequently, the map 4441 ((0, tisi1] > x given by
Gi(t) if0<t<t;
dizi (t) =
a peagy ties tat
is well defined. Continuity of @;4; follows from the map gluing theorem. It
should be clear that G;+4; has all of the requisite properties. The function
Qn is the desired lifting. We leave the uniqueness for the reader. O
The next exercise states an amazing result: no matter which point you
take in the base space, you always have the same number of points lying
above it. This result makes possible many of the beautiful theorems that
follow.
B. 7,(S's)=Z (A DIVERSION)
The previous two theorems are employed to confirm that 7,(S!,s) = Z. In
the earlier discussion, we claimed that the integer assigned to a member of
m(S',s) would somehow indicate the number of times a path was wrapped
around S!. Let s = (1,0). Consider a typical loop a based at s. By (13.A.3),
(E1,p) is a covering space of S!, where p(t) = (cos 2rt, sin 27t). By (13.A.8),
o may be lifted uniquely to a path G starting at 0 € €' and for 0 € J,
po(0) = o(0) = s. Furthermore, since pé(1) = o(1) = s, it follows that é(1)
must be an integer. The reader should recognize that this integer reflects
the net number of circuits that the loop o makes around S! as o runs from
0 to 1. A loop o that starts around S! in one direction, but from time to
m7 (S',s)= Z (A Diversion) 321
time reverses itself, is characterized by zigzags traced out by & in €!. This
leads to the following definition.
(i) Ho = 9;
(ii) Ay =7;
(iii) H(0,t) = H(1,t) =s for eachte I.
By (13.A.9), there is a unique lifted homotopy H : I x I > €! with the
property that pH = H and Ho = 6. By (iii), H(0,t) and H(1,t) € p-*(s)
for each t. However, p~+(s) consists of the integers, which is a discrete space.
Therefore, since H(1,0) = &(1), it follows that H(1,t) = &(1) for all ¢ (since
H({1} x I) is connected). Similarly, H(0,t) = (0) = (0) = 0. Then, since
pHo = Ho = 0 and pH; = H; = 7, it follows from (13.4.9) and (13.A.8)
that Ho = 6 and H, =f. Finally, we have é(1) = H(1,0) = H(1,1) = (1),
so @ is well defined. O
The next theorem and exercise show that xo and Zo are not special.
Map Liftings 323
D. MAP LIFTINGS
Proof. Half of the proof is trivial, for if such an ‘4exists, then f = Ppfiand
hence f, = px fx. This latter equations yields f,(77(Y, yo)) C px (11 (X, £o)).
Now the remarkable part: the creation of a function f mapping (Y, yo)
into (X, £0). Suppose y € Y, and let a be any path from yo to y. Then fa
324 13. Covering Spaces
Ta (1)=f(y)
Ta
By (13.A.8) there is a path 7g,gr that lifts f(a* 8"). We first show that
Taxgr is a loop. Since f(a x 8") is a loop, by hypothesis there is a loop h
based at £9 such that [ph] = [f(a x B")]. Let H: I x I > X be a homotopy
such that Ho = ph, Hy = flax 0), and HCl} x1) = HAO} x1) = 34.
Observe that ph = Ho and hence by the Homotopy Lifting Theorem, H can
be lifted to a homotopy H:IxI-—X such that Ho =hand H, = Tor
(pH, = f(axG")). Furthermore, H(1,t) € p~! (ao), {1} x J is connected, and
H(1,0) = @. Then H(1,1) = @o since p ‘(zo) is discrete. Consequently
Tax” 18 a loop based at Zo.
Map Liftings 325
oe ees 0 t
The terminal point of Tag is in W, and since f(z) = Taxa(1), the result
follows. QO
326 13. Covering Spaces
A given space may have a number of distinct coverings (13.A.7). In this sec-
tion, we investigate the possibility of whether there is a connection between
these covering spaces.
(13.E.1) Definition. Suppose (X1,p1) and (Xo,po) are covering spaces of
X. A covering morphism is a continuous function ¢: MenmeXeithat yields
the following commutative diagram.
io Seek
XxX
It is clear that the set of all covering translations form a group, where
the group operation is composition. This group which we denote by D(X ,P);
is called the Covering translation group, or Deckbewegungs group.
Given a regular covering space (Xx, Pp), there are a variety of groups that
may be associated with it. Two such groups include 7 (X, to) /ps(™1(X, £))
(which superficially does not seem to say very much geometrically) and
Dix ,p). This latter group appears to be somewhat removed from any homo-
topy considerations. The next theorem asserts that an amazing coincidence
occurs: these seemingly unrelated groups are isomorphic. This is one of the
most aesthetically pleasing results in mathematics.
Hence,
$(gh) = [p(B * 9a)|p.(m (X, #)) = [pB * pgalp,(m1(X, €))
= [pB][polp.(m(X,#)) (since pg = p)
= [pB
lps(m1 (X, 4) [palps(m (X, €))
= $(9)(h).
328 13. Covering Spaces
Finally we show ¢ is onto: Consider {a]p,(71(X, £)) where [a] € 1 (X, 20).
Let 7 be the lifting of a that starts at ¢, and set ; = 7(1). We wish to
find an h € D(X,p) such that h(@) = @,. Actually, we show: whenever
&, € p~!(zo), there is an h € D(X,p) such that h(@) = #1.
(X, Zo)
bs Si
Zh
SRS TD
Since ™(X,@o) = {e}, we have p,(m(X,#0)) C Bx(m(X,#0)) and
hence, by (13.D.1), there is a lifting ¢ of p such that the diagram is commu-
tative. Thus, ¢ is a covering morphism. O
It follows from the previous theorem that any simply connected space
serves as its own universal covering space (where the projection map is the
identity). However, we will presently obtain more substantial results.
Proof of (13.F.5). For clarity we subdivide the proof into the following
parts.
On the other hand, if 2 = (a) € p-1(U) then p(Z) = a(1) € U and hence
2 = (a) = (ax B) € Uz € {Uz : p(£) € U} where @ is the constant path at
a(1).
Finally, we see that B is a basis for a topology on Xg. Suppose that
2 € Us, Ve., where Z = (y), #1 = (v1), and £2 = (y2). Then 2 = (y1 * #1)
and Z = (72 * G2) where 6,(7) C U and (o(I) C V. Consequently, we have
Yo 1% and y > yx Se. Let W € S, where y(1)€ W andW CUNY.
We show that 2 € Wz C Uz, NVz,. If § € Wz, we have that 7 = (yx) where
BI) CW CUNY. Therefore, § = (71 *(81*8)) = (72*(B2*B)) € Us, AVE.
Thus BG is a basis. We endow Xq with the topology induced by B.
(D) Members of the family {U : U € S} serve as canonical neighborhoods.
The continuity of p follows from the properties of B, which may be easily
verified by the reader. We prove the following assertion: if U € S, = (a),
and Uz; C bP then Plu, : Uz; > U is a homeomorphism.
It is easy to show that Plu; is onto. Suppose y € U, and let 6 be a path
in U from p(£) = a(1) to y. Then p((ax B)) = y.
We show that ply; is one to one: Suppose p((a*8) = p((ax*T)); we show
that (a*3) = (ax7). Since p((a*)) = p((axr)), we have B(1) = (1), and
consequently 8x7" is a loop at x in U. We now establish that [(a*8)x (a*
T)"] € G. This is done by showing that (a « 8) x (ax*T)” = (ax8) x(r" xa")
is homotopically equivalent to the constant path c,, : J — X, which implies
that [(a
* B) x (a*T)"]}=e€G.
Since 8x7” is a loop in U, there is a homotopy H : I x I + X such
that:
(i) H(s;0)= (6 #7" )(s) for each s © I.
(ii) Ai(sy l= eyior each sé I.
(ip) (0; t= (1,t) =aidor each 1.6 I,
By reparameterizing, stacking homotopies and using the map gluing theorem,
we get the desired homotopy (as indicated in the figure below) between
ax Bxt" xa” and c,,. Thus p is an injection.
332 13. Covering Spaces
We now turn our attention to three things. The covering space D4 (and
of course the covering map), the base space X, and the group of covering
translations. In some sense, any two of these determines the third.
(1) We saw (13.E.5) that if (X,p) is a regular covering space over X, then
D(X,p)= ™(X)/pam (X).
(2) In the previous theorem (13.F.5) we saw that given a suitably nice space
X and a given normal subgroup G of m(X) (which determines D rather
indirectly), then we can construct X (and p) so that p,(m(X)) =G p]
and D(X,p) ~ ™(X)/pam1(X).
(3) We next show that given a suitably nice space X and an appropriate
family H of homeomorphisms on X, then we can construct a base space
X and a covering map p, so that D(X,p) = H.
(13.F.7) Definition. Suppose X isa topological space. Suppose further
that H C4 XxX =X sie homeomorphism} and H is a group under the
operation of composition. The group H is said to be properly discontinuous
if and only if for each « € X there is an open neighborhood U of x such that
{h(U) : h € H} is a pairwise disjoint family of open sets.
Universal Covering Spaces 333
pn : (S',
89) + (S",
80) defined by (cosa, sin a) — (cos na, sinna)
334 13. Covering Spaces
works. Why is this the only such covering space (up to isomorphism)?
Suppose another, (KX,P) exists. Since X is an n-sheeted covering space,
Pom (x ,£o) is a subgroup of order n in {a; Ne but there is only one such
subgroup, and so P,7(X) = pem(S,). Theorem (13.D.1) gives us covering
morphisms ¢ : (X, #9) — (S!,
so) and w : (S!,
so) 3 (X,@o). The reader can
supply the recurring argument which shows ¢w and ~w¢ are identity maps.
of the three /4 and the three /_ in a one to one fashion, and do the same for
the r;’s and r_’s. However, we must impose one additional condition: the
resulting snarl must be connected. Below we show one such realization. We
take p to be the “projection” map (even though you might have to carry out
the construction in €* to avoid the illusory crossings).
as =
od i
rn
Find Dix ,p). Problem 3 from section C may be applied to see that the
example pictured is a regular covering space. ,
(13.F.15) Exercise. Show that the Klein bottle also appears as the base
space of €?, where H is the group generated by h ft (defined by hy:(xr,y) =
(x +1, —y)) and the hi,s, where hy is defined by hn(z,y) = (x,y +n). Note:
This gives €? as the universal covering space of the Klein bottle.
PROBLEMS
Sections A and B
ay.
Xy X94
v6
is commutative and es 1, Pi) and (Oe2, P2) are covering spaces of X, then
(X1,p) is a covering space of X2.
336 13. Covering Spaces
Section C
Section D
Sections E and F
Problems 5-10 yield the Borsuk-Ulam theorem (problem 8) and some of its
consequences.
338 13. Covering Spaces
P? Si
; ot pitt oneal
6A a
7 Piso itaeast
- wh aAe 1A 7
wt 11 :
O femct 0s >Re =)(Sah
; é Cet “tout
3 - on =
: >a8 aa ite
ta : a (eng ox
.
: 2a if}a i‘|- teeta
‘
Chapter 14
SOME ELEMENTS OF
SIMPLICIAL THEORY
When the Jordan curve theorem was established in Chapter 5, we first proved
the result for simple closed curves composed of straight line segments, and
then generalized to arbitrary simple closed curves. This illustrates one of the
basic techniques in topology: theorems are proven for spaces with some kind
of linear structure, and then efforts are made to extend these results to more
general spaces. Frequently, however, this technique leads to more questions
than answers. In this chapter, we will work almost exclusively with spaces
that have or may be given a local linear structure. The advantages inherent
in dealing with such spaces will become apparent as the theory is developed.
g= {Sonn ee a yaa
341
342 14. Some Elements of Simplicial Theory
k k
{Soameet 20, Sarma}
i=0 i=0
A 2-complex A 3-complex
to a carrier is the relative topology with respect to €”. Ifo is a simplex, the
boundary of o, denoted by G, is the carrier of the proper faces of o.
(14.A.14) Example.
[ODS
In Chapter 16, a more general notion of simplicial complex and triangu-
lation is employed. The following definitions show in what sense noncompact
spaces may be triangulated.
U3 W2
This conjecture and variations of it was settled in the late 1960’s. The 2-
dimensional case was proven by Papakyriapoulos [1943]. The 3-manifold
case was proven by Bing [1954] and Moise [1954]. Milnor [1961] showed fail-
ure for 7-manifolds. The general situation is exposed in the papers of Cas-
son [1967], Sullivan [1967], Lashof, Rothenberg [1968], Armstrong, Cooke,
Rourke [1968], and Kirby, Seibenmann [1969]. Copies of some of those pa-
pers (and others) as well as an extensive introduction can be found in Ranicki
[1996].
This section concludes with a theorem whose proof follows immediately
from (4.B.28).
B. BASIC CONSTRUCTIONS
star(o,K)
link(o,K)
(14.B.6) Example.
eQ9
\ VY
Note that K is a subdivision of L if and only if |K| = |L| and each
simplex in L is the union of simplexes in K.
Basic Constructions 349
(14.B.12) Example.
GO)
(14.B.15) Example.
(14.B.19) Example.
Basic Constructions 351
(14.B.24) Example.
(14.B.28) Example.
k
1
d(bz,0) = ||be — vo|| =
(0) ~% i=0
1 k
1
rei (de ‘ »)]
1=0
=
k+1
diam o.
In this ||v — ul] represents the Euclidean norm, i.e., the distance from v — u
to the origin. O
Simplicial Approximation Theorems 355
(14.C.3) Theorem. Vertices vo, v1,. ) Uk from a complex K are the ver-
tices of a simplex in K if and only ifnes star(v;) # @ (or equivalently,
Nizo Istar(vi)| 4 0).
Proof. Suppose vo,v1,...,Ug are the Rertices of a simplex o. Then
clearly o € star(v;) for each i,me hence o € ae_o Star(v;) and furthermore,
lo] C Mg [star(s
Now suppose z € |star(v;)| for i = 0,1,...,k. Let o be the unique
simplex of kK that contains z. Since o C |star(v;)| for each i, we have that
o € star(v;), and hence for each i, vu; € o. Since any subset of the vertices of
a simplex spans a face of the simplex, it follows that v9, v1,..., vg determines
asimplex of K. O
K| - > Ec
V1 Wi w2
Here, the function f maps K onto the perimeter of the triangle w;w2w 3 and
sends both v; and v2 to w;. The problem is that K does not have enough
vertices for f to be simplicially approximated.
The previous theorem and (14.C.9) combine to yield the following result.
(14.C.12) Theorem (Simplicial Approximation Theorem II). If K
and L are complexes and f : |K| — |L| is continuous, then there are an
integer n and a simplicial approximation ¢ to f : |K‘")| — |L| such that
@ ~ f (mod A), where A= {x € |K|: f(z) =¢(z)}. O
The next corollary follows easily from (14.C.12), (14.C.2) and (14.C.4).
(14.C.13) Corollary. Suppose K and L are simplicial complexes, f : |K| >
|L| is continuous, and e > 0. Then there are integers m and n and a simplicial
map ¢:|K‘™| — |L(™| that is homotopic to f (mod.A) and moves points a
distance less than « (where A= {x € |K| : f(z) =¢(x)}). O
(14.C.14) Corollary. Suppose K and L are complexes. Then every ho-
motopy equivalence class of Hom(|K’|,|Z|) can be represented by a simplicial
map. In particular, if f : |K| — |L| is a homeomorphism, then there is a
simplicial approximation ¢ of f such that the isomorphism f, is equal to ¢,
(of course ¢ need not be a simplicial homeomorphism). 0
(14.C.15) Corollary. The fundamental group of S” is trivial for n > 2
(cf. the discussion between (12.E.1) and (12.E.2)).
Proof. Suppose S” is represented by the boundary of an n + 1 simplex,
say o. Let vo be a vertex, and suppose that f : J + || be a loop based
at vo. If |o| \ f(2) 4 9, then it is an exercise in analytic geometry to show
that f ~ cy, mod {0,1}). If f() covers all of |a|, we proceed as follows. By
(14.C.11), I may be subdivided so there is a simplicial map 7¥ : [| — |o|
that is homotopic to f. Furthermore, since f(0) = f(1) = vo, we have by
(14.C.7) that 7(0) = 7(1) = vo. Since (|J™]|) is in the 1-skeleton of |o|, it
follows that |o| \ y(\J°™|) # 0. Another exercise in analytic geometry will
yield y ~ cy, (mod {0,1}). O
(14.C.16) Corollary. For n > 2, every map f : S" + S' is null homotopic.
Proof. Since 7(S") = {e}, f can be lifted to the universal covering
space €! of S! (13.D.1). O
(14.C.17) Exercise. Show that every continuous map f » SF + S" (k <n)
is null homotopic.
cated in the previous figure how this may be done. Successive subdivisions
are denoted by (K,M), (K,M)®),....
Simplicial Approximation Theorems 359
Il. If @ = upd, +++ Uj-1UiVi41 +++ Un and {vj-1, Vi, Viti} spans a simplex
in K, then a is replaced by vov, --- Uj—1Uj41 °°" Un-
II-!. If a = vp, - + Uj—-1Vi41 +++ Un and {v;-1v;i41} spans a simplex in
K, then a is replaced by vovq - ++ Uj—1UjVi41 °°*Un-
It is easy to verify that these four operations lead to an equivalence
relation on the edge paths of K.
(14.D.2) Examples.
vo U1 =V4 U3 vo V1 U3
V6 Ug V10 U6 U8 Vv10
(14.D.3) Exercise. Suppose @ and f are equivalent edge paths with end
points v and 0, and & and @ are equivalent edges with end points 0 and uw.
Show that ad ~ Bf.
362 14. Some Elements of Simplicial Theory
Even though the preceding exercises are trivial, they are indispensable
in allowing one to define a group structure based on edge loops.
Suppose @ = U0U;, Viz *** Ui, Vo is an edge loop in K based at vp. Define
w([a]) = toi, ti:i. -+- ti,0- The remainder of the proof is straightforward, but
the reader should verify: (i) 7 is a well defined homomorphism, and (ii) dy
and ~w@ are the identity isomorphisms of e(K,vo) and G, respectively. O
(14.D.13) Notation. With the notation as above, we henceforth denote
the generators $(s;;) of e(K, vo) by gi;.
To illustrate the relative ease of calculating edge path groups with the
aid of the preceding theorem, we start with the unit circle S!. Since S' is
topologically the same as a triangle, a contractible subcomplex L containing
all vertices may be obtained as indicated in the next figure.
V1
gol gi2 L g
G~{so2;}~{902;}
vO 902 U2
Thus, G is a group with one generator, go2, and no relations, i.e., the infinite
cyclic group Z. Before proceeding further, let us establish that the foregoing
result is no coincidence, i.e., for any simplicial complex K, we have e(K, vo)
is isomorphic to 7 (|K], vo).
364 14. Some Elements of Simplicial Theory
c b c b
ao ay a2 a3 a4 a5 a 5
c b
a0
z 3 5
The last equivalence is due to the fact that aga, ---a,%4bcag is equivalent to
do and G is simplicial. Therefore, dis 1-1. O
Observe that it follows from the proof of the foregoing theorem that the
fundamental group 7 (||, vo) depends only on the carrier of the 2-skeleton
| 2| of K.
Despite the apparent ease of calculating fundamental groups via the edge
path group, the reader should realize that all is not well. Many triangulable
spaces require an inordinate number of 2-simplexes, (the torus, for example
requires at least twenty-one 1l-simplexes, fourteen 2-simplexes, and seven
0-simplexes). Thus the number of generators and relations may become
astronomical, even for relatively simple spaces. Computation of the edge
path group is often greatly facilitated by the following theorem, which is due
to Seifert and van Kampen. It is the polyhedral analog of the Seifert-van
Kampen theorem stated in Chapter 12.
Suppose G is a group with presentation {g1,...,9n : T1,--.;Tm} and
H is a group with presentation {h1,...,hp : $1...,8;}. Recall from group
366 14. Some Elements of Simplicial Theory
(14.D.17) Corollary. If K = L; UL», |K|, |Li|, |L2|, and |N| = |L1 N Lo|
are connected, and |Lg| is simply connected, then e(K, vo) may be obtained
from e(L1, v9) by adding the relation (71),(gi;) =e for each g;; € N \ Ty.
Proof. The free product of e(L1,v9) and e(L2,v9) is isomorphic to
e(L1,v0). By use of (14.D.15), we obtain additional relations of the form
(11) (giz) ((t2)x(giz))~* = e for each gi; € e(N, v0), where g;; corresponds to
an ordered 1-simplex in N. Since (i2).(gi;) = e, we have ((i2).(gi3))7! =e,
and hence also relations of the form (71)4(gi;) =e. O
Edge Path Groups: How to Compute 71(|K|,
vo) for Any Complex 367
Note that under this identification, the sides of the square form the wedge
product of S! with itself, and a 2-cell (the original square) has been attached
to the product. Alternatively, this may be described as indicated in the next
figure. A 2-cell D is attached to S' V S! by gluing ab onto vstv, be onto
vs't'v, cd onto vtsv, and da onto vt's'v.
acb—stus
bda—stus
t U
368 14. Some Elements of Simplicial Theory
Ake
WA ABA
/
U7 /
Vo U5
ANT AASK
7
Let s be the 2-simplex with vertices vj, v, and b, and |s9| be the closure
of s. Let b’ be the barycenter of the 2-simplex (bujv}). Radial projection from
b yields a deformation retraction of |K|\|s9| onto |K| that leaves points of ||
pointwise fixed. Of course, the projection is originally done in |P| and then
transferred over to |K | by f; furthermore, the projection is not piecewise
linear. By (12.C.19), m(|K| \ |s9|,v0) is isomorphic to ™(|K|,vo). Since
m™(|s9|) = {e}, (14.D.17) may be applied, where Lz = s9 and L; = K \s, to
see that e(K ,Uo) may be obtained from e(K \ 8, Uo) by adding the relations
arising from generators in L; M Lz. Note that any ordered 1-simplexe in
L, M Lg serves as a generator for e(L1 N Lz, vo).
In essence, what has occurred is that if a cell is glued to a loop in K,
then the group element determined by this loop is obliterated (reduced to
the identity). O
deformable to S' V S!. Hence, since e(C) = {e} and e(S! V.Sa) = gig> ss},
we have e(T’) = {91, 92; 91929‘99;'}-
92
There are two rather startling results that may with some imagination
be extracted from our work thus far. The enterprising reader might wish
to consider the following two assertions, the first of which is not difficult to
establish.
(i) For each finitely presented group G, there is a two-dimensional
polyhedron whose fundamental group is isomorphic to G.
(ii) For each finitely presented group G, there is a 4-manifold that has
G as its fundamental group.
The latter result coupled with the work of Rabin [1958] leads one to
conclude that there is no countable algorithm that classifies 4-manifolds.
PROBLEMS
Section A
Section B
Lis The house with two rooms (see next figure) consists of the top labeled T;
the partition, P; the bottom, B; the walls W;, W2,W3, and W4; the two
curtains, K,, K2; and the two tunnels, C;,C2. One enters the bottom
room by crawling down from the roof through tunnel C;. Similarly, one
enters the top room by crawling up from under the floor B through
tunnel C2. Show that the house with two rooms (which is triangulable)
can be obtained by collapsing a 3-cell; hence, the house with two rooms
is a retract of a 3-cell, and thus is contractible.
Show that the house with two rooms is not collapsible to a point, but if
curtain Ky is thickened into a 3-cell as shown in the second figure, then
the resulting complex is collapsible.
372 14. Some Elements of Simplicial Theory
Show that the product of the house with two rooms and J is collapsible.
Section C
1. Suppose K and L are simplicial complexes and f,, fo: K — L are sim-
plicial maps. The maps f; and f2 are contiguous if and only if for each
(vov1 --- Uz) € K, there is a simplex r € L such that fi (vo),..., fi(ve)
and fo(vo),..-, fo(v~) are vertices of T.
(a) Suppose K and L are simplicial complexes and f,, fo : K > L are
simplicial approximations to a continuous map ¢ : |K| > |L|. Show
that f; and f2 are contiguous.
Problems S713
(b) Show: if K and L are simplicial complexes, and fi, fo: K 3 L are
contiguous simplicial maps, then f; and f2 are homotopic.
2. Show: if K and L are finite simplicial complexes, then there are only a
countable number of homotopy classes of mappings from |K| into |L].
3. Suppose A is a closed subset of S” and f : A> S"~! is continuous. Let
{Cj : a € A} be the family of components of S” \ A. For each a € A,
remove a point cq from Cg. Show that there is a continuous extension
off, £25” \ (icy + ae Ai SS".
4. Suppose K, L, and M are simplicial complexes, f : |K| > |L| and
g : |L| + |M| are continuous maps, and ¢f, ¢, are simplicial approxi-
mations of f and g, respectively. Show that ¢,¢y is a simplicial approx-
imation of gf.
5. Suppose K and L are simplicial complexes and f : |K| > |L| is con-
tinuous. Show: there is a simplicial approximation to f if and only if
the carrier of each simplex o € K is contained in f~!(star(v)) for some
vertex uv € L.
Section D
2. Use the edge path group to calculate the fundamental group of the
projective plane P2.
5. The two holed torus T, may be considered as the quotient space obtained
by identifying edges of the octagon as indicated in the figure.
FURTHER APPLICATIONS
OF HOMOTOPY
Thus far, we have considered two major areas of topology where the concept
of homotopy was indispensable: the creation of the fundamental group, and
the development of covering spaces. These do not represent the only contexts
in which the notion of homotopy is relevant. This chapter is devoted to
exploring a few additional problems where homotopy theory is crucial.
375
376 15. Further Applications of Homotopy
The reader may feel that the foregoing definition is a typical mathemat-
ical ruse to get around the problem, i.e., avoid it by restricting attention only
to those spaces that. behave themselves. While this may indeed be the case,
there are nevertheless a number of topological pairs that yield the AHEP,
e.g., polyhedral pairs.
< |o9|xI
ax{0}
The Jordan curve theorem states in part that a simple closed curve lying in
E? separates the plane. Obvious as this result seems from a visual stand-
point, it was no trivial matter to verify. Hence, one might suspect that in
general it could prove difficult to determine whether or not a given subset
of €” separates the whole space. The goal in this section is to establish one
important criterion for such a separation to take place. Homotopy theory
once again proves to be an unwitting accomplice.
We make use of certain maps first described by Borsuk, which commonly
bear his name. These maps are constructed as follows. For each p € &",
define Y, : (E” \ {p}) + S"~} by ¥,(x) = (x — p) \ ||z — p||, where z and p
are considered as vectors, and ||x — p|| represents the usual norm (distance
The Separation Problem 379
ea ifs EeUN(CUC*),
F(z)=4 .
Peerit eee (CAN Oia rt,
380 15. Further Applications of Homotopy
Note that this corollary yields a portion of the Jordan curve theorem,
although it should be observed that neither (15.B.1) nor the corollary gives
the number of components that a separation may cause, nor if this number
is a topological invariant. Results of this nature may be found in Dugundji
[1966].
The next result contains an important consequence of (15.B.1). This
result identifies one of the most powerful characteristics of E”: invariance of
domain. The theorem states that embeddings map open subsets of €” into
open subsets of €”. This property also holds for n-manifolds (see problem
B.10).
(15.B.4) Theorem (Invariance of Domain). If U is an open subset of
E€” and h: U > €” is an embedding, then h(U) is open in E”.
Proof. Suppose z = h(u) € A(U). Since U is open, there is an € > 0
such that if V = S.(u), then V C U. Note, since V does not separate €”,
it follows from the previous theorem that h(V) also fails to separate €”.
Furthermore, h(V) \ h(Fr V) is connected, since h(V \ FrV) = h(V)\A(FrV)
and V \ Fr V is clearly connected. Hence, €” \ h(Fr V) may be written as the
disjoint union of the connected sets €” \ h(V) and h(V) \ A(FrV).
Since these two sets are connected, they must be components of the
open set €” \ h(FrV). However, components of open subsets of a locally
connected space are open; hence, in particular, W = h(V) \ FrV) is open,
and since z € W C A(U), it follows that h(U) is an open subset of €*. O
C. UNICOHERENCE
We use a pair of lemmas to prove the theorem. The first lemma follows
readily from covering space theory, and the proof of the second is left as an
easy exercise.
(15.C.3) Lemma. Suppose f and g are maps from X into E1. Define f
and @ by f(x) = (cos f(z),sin f(x)) and g(x) = (cosg(z),sing(x)). If we
have f(x) = g(x) for each x € X, then f(x) = g(x) + 2nm for some integer
nandforeachre€ X. O
Proof of (15.C.1). Suppose X is not unicoherent. Then there are closed
connected subsets A and B such that X = AUB and ANB is not con-
nected. Write 4 B as the disjoint union of nonempty closed subsets C;
and C2. For each z € X, let gg = d(x, Ci)/[d(x, C1) + d(x, Cz)]. Define func-
tions f4 : A > S! and fg: B > S! by letting fa(x) = (cos7qe,sin7qz)
and fp(x) = (cos7@z,—sintgz) = (cos(—7qgz),sin(—7gz)). Observe that
382 15. Further Applications of Homotopy
fal|ans = falans, and consequently the map gluing theorem ensures that
the function
ba fa(z) ifa@gA
os esi ifzeB
is both well defined and continuous. By hypothesis f is inessential; by
(15.C.2), there is a map g: X — €! such that f(x) = (cosg(z),sin g(z)).
Lemma (15.C.3) gives integers m and n such that g(x) = 7q,+2m7 ifee A
and g(x) = —7q, + 2na if x € B.
However, if x € Cy C ANB, then g(x) = 2mm = 2nz, hence m = n.
On the other hand, if r € C2 C ANB, then g(x) = 7 + 2ma = —7 + 2nz,
and we have shown that 7 = —7. Thus, X must be unicoherent. O
(15.C.4) Corollary. The spaces €”, I” and S"*? are unicoherent for each
mGeNe ge
Our goal is to convince the reader of the existence of an algorithm that may .
be used to compute the fundamental group of the complement of a knot, or
more generally of the complement of a finite graph lying in €°. A careful
blend of mathematics and sleight of hand is employed. We hope that the
reader will be convinced that the hand waves can be done mathematically.
It is assumed that only the more naive, or masochistic, reader will attempt
to do more than determine that the constructions can be carried out.
(15.D.1) Exercise. Show that (E? \ {p}) = {e}.
(15.D.2) Definition. A graph in €° is the carrier of a one dimensional
simplicial complex in €?.
ser
In the remainder of this section, we assume that all graphs have been
simplified in this way. We make a further observation—subdivision of G
(considered as a complex) does not change 7;(E? \ G).
(15.D.4) Definition. A branch point v of a graph G is a vertex of G that
is the face of three or more l-simplexes of G. The graph just pictured has
three branch points.
(15.D.5) Definition. Let p : €? + E? C E* be the usual projection map
onto the zy plane, i.e., p(z,y,z)= (z,y, 0), and let p = plc.
The map p projects G normally onto the zy plane (or p(G) is a normal
projection) if and only if
(i) p is 1-1 except at a finite number of points of G;
(ii) for each g € G, card(p~'p(g)) < 2;
(iii) if v is a vertex of G, then card(p~'p(v)) = 1.
Observe that (iii) rules out the following configurations.
pa
| _
384 15. Further Applications of Homotopy
The line segment contains —> > The line segment contains
an overcrossing an undercrossing
Now, at each branch point add (if nena a nearby vertex on each
arc coming to the branch point and then push the branch point through the
xy-plane. After this stage the graph appears as:
Pick as base point p = (0,0,1). Let A be the slab between z = —e/2 and
z = —3€/2 with p;'(the part of G that lies in z = —e) removed. Add an arc
that starts in A, misses G and goes to p. Clearly 7(A) ~ {ri,r2,.-.,7%m;3 }
where r; is represented by a loop from p to the plane z = —e then to and
counterclockwise around the i-th hole and back to p.
(15.D.9) Exercise. Use the Seifert-van Kampen theorem to show for any
eS Sy
7™(AUB) & ph emee Laie cena e
m1 (E* \ G), for a Polygonal Graph G in €3 387
where the circumflex means the rj-th generator has been removed. [Any
loop encircling r; can be slid over AU B and expressed in terms of the other
generators. |
Next, purely for purposes of accounting, we make some more arbitrary
choices. We remove the branch points from G and obtain a number of arcs
and circles. Choose an arbirary direction in each of these and direct the arcs
of G lying in z > —e accordingly.
We now choose the “right hand rule” for determining the exponent of a
loop from the base point that passes once under the i-th arc of G\ R. Stand
on the arc facing the direction of the arrow. If the loop passes under you from
right to left then it gets an exponent of +1, otherwise, -1. [Alternatively,
point the thumb of your right hand in the direction of the i-th arc and curl
your fingers under the thumb. If the fingers are going in the same direction
as the loop when it passes under the i-th arc, then it gets an exponent of
+1, otherwise, -1.]
Li
ih e
For each arc a; (i = 1,...,k) in G\ R pick a loop z; from the base point
that passes under a; once and has an exponent of +1. [We only picture a
small part of x; below.]
388 15. Further Applications of Homotopy
In the future we will not draw the z;’s; a label of x; next to an arc will imply
that the loop running under the adjacent arc is in a positive direction. The
circles in the drawing represent a birds-eye view of the r,’s.
(15.D.10) Exercise. a) Show that an open 3-ball with the z-axis removed
has a 1-sphere as a deformation retract.
b) Let H represent the set of points in €? with z > —e. Use part a)
and several applications of the Seifert-van Kampen theorem to show that
7™(H \ G) = Aas ek als
(15.D.13) Example. For the graph pictured at the beginning of this sec-
tion:
mi (E? \ G) = {21,--. 510; 23 po23T7, XeT\o
Ly -T3, T3T]
Ly Te,
©2025 '21, ©1242] Tp", Ly 2522, T9Z10r4}
m(E°\K)={21; J=Z
2. Trefoil knot.
r3= —1 -1,-1
m(E3\K)2{01,02,03,04j) 2, 030125 B30, © 21}
390 15. Further Applications of Homotopy
r5
3 ; ah wee i Po
m1 (E \G)={21,02,03,24,25 ; 232224 »25 ©4%3,%5T1L, Lo I;
: St oti eo)
HEI ce een Lo, %3, L1%3X%172, T1VM3IQV3, L2L3L22) }
—_—_——”
TQ=T17%321
ay
{Pis035 27, Tz, £10301 110321, V1 U3L1XL3%1
73, ©123X1 L311 132121}
= ae: 2 a
= (21, 2a; hq, 3, (1123) i:
now be checked by brute force. The quotient group A/(x?, 23,23) that we
have obtained is the dihedral group of order 6, (which also happens to be
the symmetric group S3), and is neither cyclic nor abelian. Hence, A is not
isomorphic to Z.
In all of these examples, one proves wildness by showing that the funda-
mental group of a complementary domain is nontrivial. There are, however,
examples of wild arcs and spheres whose complementary domains have triv-
ial fundamental groups (see Fox and Artin [1948]). The proofs of wildness in
these latter cases involve far more subtle properties of 3-cells than we shall
consider.
(15.E.7) Example (A Wild Arc) (Fox and Artin [1948]). The construc-
tion of this arc involves a certain amount of tedious description. The reader
is cautioned to read the next paragraphs with the figures below in sight.
Let C’ be the right circular cylinder in the figure. The basic idea is to
embed C in each of the regions D; in such a manner that the end point of the
arcs on the left-hand boundary of each D; match up with the appropriate
points on the right hand side. On the base B_, mark three collinear points
t_, m_, and b_. On the other base B,, mark collinear points t,, m4, and
b,. Construct in C’ three disjoint polygonal arcs a, a2, and a3 whose images
are A;, Ag, and A3 with the properties that A; joins t_ to m4, Ae joins by
to t,, and A3 joins m_ to b_. These arcs are arranged as indicated in the
figure (it is possible to describe the arcs analytically, but by now the reader
should be convinced of the sterility of such exercises). Let A = A; UA2UA3.
Suppose S is the solid, double spear point obtained by rotating about
the x-axis the area bounded by z = 0, z = (x/2) +1, and z = (—2/2) +
1. Subdivide S into a countable number of pieces by the family of planes
z = +(2-—1/2™~'), where m = 0,1,.... Let Dn be the segment of S
with x coordinates satisfying (2 — 2?-") <‘x < (2 —2!-") for n > 0 or
(—2+2-") <2 < (-2+2'™) ifm < 0. Let p= (=2;0,0)handq ='(2;,0)0):
For each n, let f, be a homeomorphism from C onto D,, with the fol-
lowing properties:
Wild Sets do Exist 393
(i) The simple closed curves bounding the bases of C are mapped to the
simple closed curves bounding the bases of D,. Furthermore, the
indicated orientation of the simple closed curves is to be preserved.
(ii) The bases B_ and B, are mapped respectively to the left and the
right faces of Dn.
(iii) The points fn(t+), fn(m+), and f,(b4) lie in descending order
on a vertical line through the x axis and coincide with the points
fn+i(t_), fav (m_), and fn4i(b_) respectively.
(iv) The set f,(A) projects into the xz plane as
NS where NX
indicates that the point on the solid line has greater y coordinate
than the corresponding point on the broken line.
(15.E.8) Exercise. Convince yourself that the homeomorphism f,, exists
and X = {p}U (Uz fn(A)) U {q} is the image of the arc a in the following
sketch.
In order to facilitate the use of the algorithm that gives the description
of the fundamental group of the complement of a, we redraw that figure and
label subarcs that extend from undercrossing to undercrossing as follows:
nor
394 15. Further Applications of Homotopy
Since f,+ is the inclusion map, it now follows from the geometry above,
(12.F.11) and (Ap.D.9) that 7,(€? \ X) is isomorphic to
aaaare Z ; =
lima (O°
\XA; (a; ber, @e Z); ap cncer,
b=
a =i
Cit 14iCi44, Ci-1 = bibi41;
Ay ey
; b; ajc; =e
;
(i Se Z)}.
DApSiagoby
ox aaab: 8)
= aybrag pee ety (Koved
(ennaegs) aed Gee
= qee c) ic;Seal
Cry
Note that in both cases 1 and 2 we have that cjc;_1Ci41C; = Ci41CiCj_-1.-
Thus, it follows from (Ap.C.9) and (Ap.C.13) that
is mapped onto the D,’s. Then Y = {p}U (Ufi(T;)) U {gq} is the desired
2-sphere.
(15.E.10) Exercise. Convince yourself that the closure of one of the com-
plementary domains of Y is a 3-cell.
(15.E.11) Remark. It follows from (15.E.4) that both complementary
domains of a tame 2-sphere S in €3 have trivial fundamental groups.
Let D be the complementary domain in the previous exercise whose
closure is a cell. Since it is clear that S* \ D is a deformation retract of
S? \ X, we have that 1(S? \ D) = m™(S? \ X) # {e}; consequently, S3 \ D
is not a cell, and by Remark (15.E.11), Y is a wild 2-sphere.
(15.E.12) Example (A 2-sphere in S* for which the closure of nei-
ther complementary domain is a cell).
In this example, push a bubble on the 2-sphere into the complementary
domain that is a cell. Now cut off the bubble and graft in its place another
copy of the wild sphere (minus a disk) to obtain the following:
PROBLEMS
Section A
Section B
Section C
1. Show: if the union of two continua C; and C2 is S" (n > 1), then CyNC2
is connected.
Show that any contractible, locally path connected space is unicoherent.
Is the space described in (12.C.8) unicoherent?
Suppose X is a compact metric space. Give Z = C(X, S') the compact-
open topology. Show: if Z is path connected, then X is unicoherent.
5. Show: if U; and U2 are connected open subsets of S* and FrU; N Fr U2
is connected, then U; M U2 is connected.
Section D
1. Find a presentation for the group of the following knot, and prove the
group is trivial.
LES
Problems 399
2. Find a presentation for the group of the following knot, and show the
group is nontrivial.
EIS
3. Find a presentation for the group of the following knot, and show the
knot group is nontrivial.
4. Show that the groups obtained in problems 2 and 3 are not isomorphic.
5. Find a presentation for the group of the following knot, and show the
group is nontrivial.
Ss
6. Show that the fundamental group of the complement of the following
graph is {a,b; }.
400 15. Further Applications of Homotopy
7. Find a presentation fof the group of the following know and show that
it is differest from the group of the trefoil.
8.
OOOO
Show: if K is a knot, G is the group of K, and [G, G] is the commutator
subgroup of G, then G'/[G, G] is infinite cyclic.
Chapter 16
2-MANIFOLDS
401
402 16. 2-Manifolds
Hy (a) if x € I(Yn)
Ay+1(£) = {
9H nsi(z) if 2 € Tent \I(%n)
Then h: M -+ €? defined by h(x) = H,(x) whenever x € I(yn) is the
desired homeomorphism.
The next goal is to show that all 2-manifolds are triangulable. Since
compactness will not be assumed here, the more general definition of tri-
angulation (14.4.16) is applicable. The conditions imposed on the disks in
the next lemma are sometimes abbreviated by saying the disks are in “gen-
eral position.” The triangulation of 2-manifolds is an early example of the
exploitation of general position. A few additional lemmas are convenient.
Claim. There are arcs a; and a2 from p to q that lie in Cy and C2,
respectively, and which intersect h(F;, MN E,) in at most a finite number of
points. Note that if such arcs can be found, then the union of these arcs
forms a simple closed curve and Dny; = h~!(I(a1 Uaz)) will satisfy the
inductive requirement.
Proof of claim. We will say that an arc in M is admissible if and only if
its intersection with F’, is finite. A point in M will be called a crossing point
if and only if it is on the boundary of more than one D;. Observe that if
P1,P2,--+,Pq is a collection of points in M and if for each i = 1,2,...,q—1,
there is an admissible arc with end points p; and p;,,, then there is an
admissible arc with end points p; and p,. Furthermore, it is easy to see that
every noncrossing point has a neighborhood in which any two points lie on
an admissible arc. (Use (5.D.9), for example.)
In C, construct two small admissible arcs 3; and G2 with end points
{p, zi} and {q, z2}, respectively, such that (31 \ {p}) U (Ge \ {q}) C int C).
Since each point x € int C) is a noncrossing point, there is an arc connected
neighborhood N, of x lying in int C; such that any two points of Nz may
be joined by an admissible arc. By (2.F.2), there is a simple chain of the
N,’s between z; and zg. It should be obvious how to obtain the admissible
arc a; from p to q. Of course, an identical procedure may be used to join
p and q in C2 with an admissible arc ag. As indicated previously, this not
only completes the proof of the claim but also of the lemma. O
oe
2
Q
If the total number of 2-simplices in S is ered Ba 2, then the quadri-
lateral Q feanee from the identification of oj, with oj, (by Ri,;,) will poe
at least one face o}in 3available for appropriate identification with a face of OF,
of a third simplex a? in S. Otherwise M would not be connected! Let Rj,;,
be the relation responsible for this identification.
1
Tiss
_—
1
Tio
2
os
408 16. 2-Manifolds
U7 U6 U5
and the sequence aa will represent the projective plane. All other compact
connected 2-manifolds will have at least two distinct letters in their repre-
sentation.
and then pasted together again along b to yield the next figure.
a
Note that in either case, after cutting and reglueing, there is now an
additional vertex vo in Vo, and the two v;’s have become one.
Of course, under the maps indicated above, a configuration acc! may
have been introduced; however, by (16.B.3), such configurations can be re-
moved without increasing the number of elements in any of the equivalence
classes. Repetition of the foregoing procedure will eventually place all of the
vertices in the equivalence class containing vp. O
From this point on, we assume that if P is a polygon representing a
2-manifold M, than all the vertices of P belong to the same equivalence
class.
(16.B.5) Lemma. If M = ac@c, then M = aaaf-!.
Proof.
a ; af oe a? af es
ib
fy
Integrating the preceding five lemmas with the exercise, we have the
principle result.
(16.B.9) Theorem. Every representation of a compact connected 2-manifold
has a normal form of
(i) aa~? or aa, or
(ii) aabbccdd..., or
(iii)wabas?
b=*cdem* das. «. Oo
Thus, if a compact connected 2-manifold is not a 2-sphere or a projec-
tive plane, it may be described either as a collection of projective planes
glued together, or as an n-holed torus. The forms just described are called
normal forms. It remains to show that distinct normal forms are associated
with distinct (i.e., nonhomeomorphic) manifolds. We accomplish this by
demonstrating that manifolds with distinct normal forms have nonisomor-
phic fundamental groups. é
(16.B.10) Exercise. Use the polyhedral Seifert-van Kampen theorem to
establish that the fundamental group of a 2-manifold, represented by a 2n
The Classification of Compact 2-Manifolds 413
Lid Wy 5Ai9)
id igh eX}
= (iy Byte
an ea ete ya ye
by the Tietze equivalence IJ (Ap.C.13). When we consider A/[A; A], the first
relation a is equivalent to x”, and the second relation can be used to eliminate
the generator z, (Tietze equivalence J (Ap.C.8)). Hence, abelianization
yields Z@Z@---@ZO Zo, where there are n — 1 factors of Z. Note that if
n = 1, we obtain Zo.
If
hE @1 220, a2 'a3r403 24° + pein th
C. A CHARACTERIZATION OF &£
Then G extends g. It follows from (4.B.21) and (15.A.1) that g may also be
extended to B? (also see problem 13 of section B, and C); hence, N is simply
connected. Therefore, by (16.B.9) and (16.B.10) N is homeomorphic to S?,
and since N is a proper subset of N, N can be embedded in €?.. O
PROBLEMS
Section A
v = number of vertices in T
e = number of 1-simplices in T
¢ = number of 2-simplices in T’
Define €(M) = v—e+t. The number €(M) is called the Euler char-
acteristic of M. It can be shown that €(M) is a topological invariant
(it does not depend on the triangulation used). Assume that to be the
case and find the Euler characteristic of the torus, the Klein bottle, and
the projective plane.
7. Suppose M is a compact connected 2-manifold and T' is a triangulation
of M. With the notation of the preceding problem, show that
(a) ‘St=e;
ibe = ee — €(M));
(c) vu> 4(7 + \/49—24€(M
Sections B and C
that the connected sum does not depend on the disks removed or on the
homeomorphism used. Find the connected sum of two 2-spheres.
Suppose M and N are connected compact 2-manifolds. Let S be the
connected sum of M and N. Show that S is a compact 2-manifold.
Show that the connected sum of a torus and a projective plane is home-
omorphic to the connected sum of three projective planes.
Show that the connected sum of two projective planes is homeomorphic
to a Klein bottle.
Show that the connected sum of a Klein bottle and a projective plane is
homeomorphic to the connected sum of a torus and a projective plane.
Show that the connected sum of the Mobius strip and a torus is home-
omorphic to the connected sum of a Mobius strip and a Klein bottle.
Find the Euler characteristic of the connected sum of a projective plane
and n tori.
Show: The Euler characteristic of the connected sum of two 2-manifolds
is equal to the sum of their Euler characteristics.
What is the universal covering space of S??
Show that the universal covering space of a compact 2-manifold is either
Se One.
Show: if M is a compact 2-manifold and M is neither S? nor the pro-
jective plane, then the universal covering space of M is €?.
12. Let S be the boundary of a 2-simplex. Suppose f is a continuous map
from S into a triangulated 2-manifold with boundary and let «€ > 0 be
given. Show that there is a simplicial map g: S > M that is homotopic
to f such that
(i) g is a local homeomorphism, and
(ii) the homotopy between f and g does not move points more than e.
13. Suppose X is a topological space and f, ie : S' + X are homotopic.
Show: if f has a continuous extension g : B? + X, then f may also be
extended to B?. (Use no cannons!)
Chapter 17
AN INTRODUCTION TO
nN—-MANIFOLDS
A. SOME PRELIMINARIES
In Chapter 1, an n-manifold M was defined as a separable metric space with
the property that every point in M has a neighborhood homeomorphic to
E”. This definition eliminates such common spaces as J, half-open inter-
vals, disks, etc. However, these spaces do satisfy the following more general
concept.
419
420 17. An Introduction to n-Manifolds
Observe that in the preceding example, Int C and BdC (in the manifold
context) do not agree with the interior and frontier of C when C is considered
as a subspace of €°.
The following theorem states that for connected n-manifolds, the sepa-
rability requirement in (17.A.1) is redundant.
(17.A.4) Theorem. If X is a connected metric space such that each point
has a neighborhood homeomorphic to B”, then X is separable.
Proof. Since X is a metric space, X is paracompact (10.A.8), and
hence, by (10.A.18) may be written as the disjoint union of o-compact spaces.
However, X is connected; consequently, X is itself a o-compact metric space.
By (10.A.17), X is Lindeléf, and therefore separable by (2.1.13). O
As was the case with connected manifolds, connected manifolds with
boundary are arc connected.
The simplest n-manifolds with boundary are €", S”, and B”. Homeo-
morphic copies of these latter two spaces are of special importance.
We have already shown that any compact convex subset of E€” is an n-cell
(3.C.2). The following exercise yields another way that cells may arise.
(17.A.10) Exercise. Suppose C is an n-cell, where C c €” C E€"*!, and
that p € €"*! \ E". Show that the cone over C from p is an (n + 1)-cell.
(17.A.11) Exercise. Show that the suspension of an (n — 1)-sphere is an
n-sphere.
The following extension property of n-cells is frequently useful.
ee {InlfGep fz #0
0 a 2 4),
Pelz) ifz#e
nas
0 it a0
Pe)
B. n-ANNULI
(17.B.4) Theorem. If C and D are tame n-cells in E€” such that CC Int D.
Then Int [C, D] is an open n-annulus. [Denote Int [C, D] by (C, D).]
Proof. An open n-annulus is homeomorphic with €" \ B”; hence it
suffices to show (C, D) and €”" \ B” are homeomorphic. Since D is tame,
there is a homeomorphism h : €” + €” which maps D onto B". Since h(C)
is a closed subset of the interior of B”, there is by (17.4.14) a homeomorphism
f :€" > Int B" that is the identity on h(C). Since h(C) = f~*h(C) isa tame
n-cell, there is a homeomorphism g : €” + €” such that gf~'h(C) = B”.
The desired homeomorphism is gf~*h|ic,p). 0
Although we are no position to prove the annulus theorem, we do es-
tablish a somewhat weakened version.
424 17. An Introduction to n-Manifolds
C. CELLULAR SETS
Ci
Cellular Sets 425
On the other hand, it can be shown that not all cells are cellular. For
example, the wild cell constructed previously (15.E.13) is not a cellular set,
as we prove presently. On the third hand, tame cells are cellular.
Proof. Let D be a tame n-cell that lies in Int C), and let h : Cp > B”
be a homeomorphism. By the invariance of domain, h(D) has nonempty
interior. We may assume that h(D) contains 0. Choose € > 0 such that B?
is contained in Int h(D), and select r, 0 < r < 1, such that h(C1) c BP.
C2
eo
Then C = f~1(D) is tame, and C' Cc Int C2. To see that C; C Int C, we just
observe that C; C h-(IntB™) = h-1g-1(Int B%) C h-*g*(Int h(D)) =
Int h-1g—4h(D) = Int f-1(D) =IntC. 0
(17.C.4) Exercise. Show [D, C] is an n-annulus, where C and D are defined
as in the foregoing proof.
426 17. An Introduction to n-Manifolds
Note that (17.C.13) does not assert that the closure of each comple-
mentary domain is an n-cell. In fact, we have seen (15.E.12) that ifn = 3,
then there is a 2-sphere in €° for which the closure of neither complementary
domain is a 3-cell. However, for n = 2 the following result holds.
In the next theorem, we see that single inverse sets are cellular.
(17.C.17) Theorem. Suppose C is an n-cell and f : C + S” is a continuous
map with a unique inverse set K. If K C Int C, then K is cellular.
it £ if@w@ek
ane rete ee
Then g is an embedding, and hence g(C) is an n-cell contained in U and
containing K in its interior. Since U was arbitrary, K must be cellular. O
(17.C.20) Exercise. Show that the homeomorphism h mentioned above
exists (by returning to €” if necessary) and find a decreasing sequence of
cells whose intersection is K.
Each member of a pair of inverse sets is also cellular.
al Oagh ty
Sa ti Gate
£ ifxe Ka
(@) hae
Nis ae
Note that Ky is the only inverse set of g, and hence by (17.C.17) we have
Ko is cellular. A similar argument shows K;, is cellular. O
a ifze A,
f(z) = 456 if z € B,
qh “(a Wait re hi(Sh an),
Compact n-Manifolds 431
SUA ST Oo
E. COMPACT n-MANIFOLDS
subarc of a; that lies outside of D and joins 2; to some point yi; contained
in A(B?, 572).
Proof of (17.E.1). For each positive k, let Uy, be a finite open cover of
M by open sets of diameter less than 1/2*. Let X;, be a finite collection of
points that contain at least one point from each U € Uy. By (17.A.17), there
is an n-cell D, in M that contains X,; furthermore, we may assume that
D, is collared (why?). Let hy : B” — D; be any homeomorphism. By the
previous theorem, corresponding to e = 1/2* and 6 = 1/2, there is an n-cell
Dz, and a homeomorphism hz : B” > D2 such that
(i) AgUD, C Ds,
(ii) De = halos. gs and
PROBLEMS
Section A
15. Suppose X and Y are manifold factors (i.e., there are spaces W; and
W. such that X x W, and Y x W2 are manifolds). Show that X x Y is
a manifold factor and Bd(X x Y) = (X x BdY) U(BdX x Y).
436 17. An Introduction to n-Manifolds
(b) Let S denote the solid torus. Can the homeomorphism of part (a)
be extended to a continuous function f : S > S?
L ts Assume the following result of Hanner[1951]: If X is a metric space, and
U,,U2,... is an open cover of X, where each U; is an ANRy, then X
is an ANRy. Show that every n-manifold is an ANRyy.
18. (a) Suppose M is a compact connected n-manifold, and N C M isa
compact n-manifold. Show that N = M.
(b) A topological space Y dominates a topological space X if and only if
there are continuous maps f and g such that the following diagram
is commutative
AC
Y id is
Suppose M and N are compact connected manifolds, M dominates N,
and N dominates M. Show that N and M are homeomorphic.
19. Find an example showing the conclusion of 18b above can fail if the
hypothesis “manifold” is dropped.
Sections B and C
In (17.B.5) show: not only may we assume that S"~! C Int A, but this
actually must be the case.
Suppose C’ C €” is connected and €” \ C is homeomorphic to €” \ {p}.
Show that C is compact.
Let Ci, C2, and C3 be n-cells with C, C IntC2 C C2 C Int C3. Show:
if [C},C2] and [C2,C3] are n-annuli, then there is a homeomorphism
h:C3.—> BF such that for i = 1,2,3, h(C;). = BP.
Suppose D C &” is an n-cell lying in Int B” such that [Bd D, Bd B”) is
a half-open annulus. Show that D is tame.
Problems 437
Sections D and E
DIMENSION THEORY
439
440 18. Dimension Theory
The following example shows that taking the union of subspaces can
raise dimension.
metric spaces are completely normal (4.4.1), there are open sets U;, U2 of
X such that C; U U; Cc U;, and (U,,U2) is a separation of U; UU2. Let
P= X \(U, UU2). Then P is a partition of X between C; and C2 and
YPriP=Y MOUs) GY VU} =P. O
Note that Example (18.A.6) shows us the following lemma can not be
improved to ind(X) <n—-1.
(18.A.14) Lemma. Suppose X = Y U Z is a separable metric space with
ind(Y) <n-—1 and ind(Z) < 0, then ind(X) <n.
Proof. Let € X and let O be an open set containing z. Since {zr}
and (X \ O) are disjoint closed sets, (18.A.10) yields open sets O1,O2 € X
with z € O,, X \O C Og and [X \ (O; UO2)|
NZ = @. Hence x € O, CO.
Finally, ind(FrO,) <n—1 since FrO; Cc[X \(O, UO2)) CX\ZCY. O
metric space where, for each i, X; is closed in X ae) ind(X;) <n, then
ind(X) <n.
Proof. If n= —1 the theorem it trivial, and for n = 0 it is just (18.A.13).
Assume the theorem holds for all n < j, where 7 > 1, and suppose
a Eee X; where each X; is closed in X and satisfies ind(X;) < j. For
each i, by (18.A.2c), (1.G.11), and problem (1.B.5), we may find a countable
base B; for X; such that ind(Frx,B) < j —1 for all B € B;.
Ley = U Frx,B. By assumption, ind(Y) < 7 — 1. Hence for each
BEUB;
i, (18.A.12) yields ind(Z;) < 0 where Z; C X;\ Y C Xj. Now, by (18.A.13),
ind Z < 0 where Z = UZ;. (Why is each Z; closed in Z?). The theorem now
follows from (18.A.14). O
444 18. Dimension Theory
(18.A.21) Lemma. Let X be a separable metric space with ind(X) < n—1.
Suppose for eachi = 1,...,n that C;, D; form a pair of disjoint closed subsets
of X. Then there are n closed sets P,,...,P, such that P; is a partition
We end this section by showing that Euclidian n-space has little induc-
tive dimension n. It is an immediate consequence of (18.A.3) and (18.A.19)
that ind(E”) < n.
446 18. Dimension Theory
The primary goal of this section is to show that big inductive dimension and
covering dimension agree on the class of metric spaces.
The definition of big inductive dimension is obviously inspired by the
definition of little inductive dimension—just replace “point” by “closed set.”
(18.B.1) Definition. Let M be the class of all metric spaces. Define a
function, big inductive dimension, Ind: M > N U {—1, oo} by
(i) Ind(X) = —1 if and only if X = 0.
(it) Ind(X) < n where n € N if and only if for each closed set C and
each open set U containing C, there is an open set V such that
CcCUCV and Ind(Fr(U)) <n-1.
(iit) Ind(X) = n if and only if Ind(X) < n and Ind(X) <n-1.
(iv) Ind(X) = oo if and only if Ind(X) ¢ n for alln EN
The definition of covering dimension stems from the observation that
for every open cover of €! there has to be some point that is in (at least)
2 elements of the cover; for every open cover of E? there is some point that
is in (at least) 3 elements of the cover; etc. Before formulating this it is
convenient to have the next definition.
(«*) for each i, {C; NUin, are Os NUim} has order <n,
AS f)Ung:
SEC JES
Uso ,0 =3 () UB,0
B<Bo
and define Uz, = {Uj, ataea- We claim that U,, is an open cover of X.
If Go is an ordinal of the form y + 1 for some ordinal y this is easy since
U3, = Uy, and hence Uz, = U, is an open cover by assumption. Thus we
assume (3 is a limit ordinal. That is G9 has no immediate predecessor.
Let x € X. Since C is locally finite, there is an open set O containing x
and an ordinal number y < ( such that OM C, = 0 whenever y < T < po.
Since UW, is a cover of X, there is an a € A such that x € U,.. By (iit),
t € Ug.q for y < B < fo; hence x € UZ,,. Consequently U4,, is a cover of
X. We now show that each Uz,, is open. Let r € Uj,,,O be an open set
containing x, and y < fp an ordinal such that UNC, = 0 for y < T < Bo.
Now pick a set U,,q that contains x. By (117) and the definition of Uj,. we
have r € ONU4,q C Ug,.. Hence WU, is an open cover of X.
Now consider {Cg, 1 Ug, ataca- This is an open cover of Cg,. By
the hypothesis on C, only a finite number of elements are nonempty. Hence
the cover has an open shrinkage {Va}aca of order < n. Now define Ug, by
Up, = (UB, .0 \Cao) UVa. The reader may verify that Ug, has the requisite
properties, and that U/p is the desired cover V. O
Big Inductive and Covering Dimension 451
The next, rather tedious lemma is needed to obtain one more charac-
terization of cov. We suggest a peek at (10.4.10) might be appropriate for
any readers not having a photographic memory.
Ses 4
ie = | ae fy dg — U; for 1 <i
and define f;,; to be idy,. Clearly, for each 1 < j and for each U € Uj,
Ure FU). J $ ‘
Suppose O = {Q,,...,O¢} is a finite open cover of X. Define an open
cover O = {O1,Oz,...} by
The O;’s are clearly open and they form a cover since, by hypothesis, the
stars form a base. Now for each k = 1,2,... define subfamilies Vk and W,
of Uk by ,
Furthermore, for each V € VY; let i(V) be the largest integer < k for which
ficv),e(V) € Vicv). This exists since f1,4(V) M (Uj; Xj) = 9 and fan(V)N
On=V ThOps 0.
For each W € W; Cc W let
We now work to show that i(V) and W are well defined on W = UW;. By the
definition of O;, and since WN O; = 0, there is aU € Ui; so that WNU £9,
and an i(W) < £@ for which W C St(U,U;) C Oyw). Since U C fi3(U)
whenever 7 < j, we have W Cc W; hence We Oww): Now when 7 # j we
have Wi W; = 0. Thus if W € UW; both W and i(W) are well defined
and W is a refinement of 0.
We next show W = {W : W € W} is actually cover of X. Let x € X.
Pick k so that z € Ox \Uj<, Oj, and U € Us so that x € U. Now, we have
ere UN, Og -C (fuoyn)) € W. Hence W is a cover of X.
Finally, we show ord(W) <n. Suppose c € Wi}. NW2N---MW» where
Wi € Wm; and W; #4 W; when i ¥ j By the definition of W,, we have
mi <k, (i =1,2,...,m), where k is the integer for which «€ Ox \(Uj <4 Oj):
By the definition of W, there are sets V; € Vy; such that fm, %;(Vi) = Wi,
i(Vi) =m, and z € V;N Ox,. Since x € Ox;, we have k < k;. The sets U; all
contain z, where U;= fx.x;(Vi) € Ve for i =1,...,m. Since ord(Uy,) < n, it
suffices to show that U; # U; when i # j. ewes that each U; is in Vy and
that 1(U;) = i(Vi) =m;. Hence U; 4 U; when m; # m;. If mj; = mj; we have
U; # Uj since fm,,4(Ui) = fi:,ei(Vi)=Wi # Wy = fj,h; (Vj) = fmy,e (U5).
O
Since the U;’s are locally finite, it follows from the claim and the def-
initions of G;, H;, that Gin O;-1 = 0 = H;NO,-1 for all i. Hence
Oa te H, = int (O;) and ©;1, 0eCEM \ G; = int O;. Furthermore
O;NO; = since G; U H; = X. Consequently,
Oe Jo. and O = We
i=1 i=1
are open, disjoint and contain C and D, respectively. For each 7 define
Lao \ (O; U O;) = G; U H;. Then
The sets O, O, and P are as desired, except it is not clear that cov(P) < n—1.
To prove this last detail; for each 7 we define an open cover of P by
Y; ={U NP : U €uand Cito # 0}. We show that V; is a cover of
PCH,. Ifx € P, then x € P C P; C G; and there is at least one U € Ui;
such that UN O;-1 = 0. If U € Uj41 and UN O; £0, then if V € U; where
U CV, we have Vn O; # (, and hence V is not in G;. Thus an Ons # 0),
which implies VN P € V;. Therefore V;+ is a refinement of V;. By (18.B.20)
cov(P) < n-—1, since diam(V;) < 1/i for allV EV; O
The first goal of this section is to show that the three notions of dimension
we have considered agree on the class of separable metric spaces. Then we
prove a some additional basic theorems for cov = Ind which hold in general
metric spaces.
Ui = X\({B : Be Ui,Bj,BNC
= 9}.
Clearly each U; is closed. Also each U; is open since each B € B is open and
closed and U5_1B; is locally finite. Hence U;,U2,... is a sequence of open
and closed sets such that
We let the reader show that for each i, {Fr(Ba) : Ba € B;} is a locally
finite closed collection. Since Ind(Fr(B)) < n—1 for each B € B, we
apply (18.B.21) and (18.B.13) to obtain Ind(C) < n—1. Define BN D by
BOD={{BND : B eB}. Since D and BND satisfy the hypothesis for
(18.C.7), we have Ind(D) < 0. Now an application of (18.C.8) gives us that
Ind(X) =Ind(CUD)<(n-1)+1l=n. O
(18.C.10) Corollary (Monotonicity). Suppose X is a metric space and
Y CX. Then dim(Y) < dim(X).
Proof. It is an exercise to establish this for Ind. O
(18.C.11) Theorem. Let X be a metric space. Then, dim(X) < n if and
only if X = Uh! X; where dim(X;) < 0 for each i.
Proof. (=) This is a corollary of the proof of (18.C.9).
(<=) This is a corollary of (18.C.8). O
(18.C.12) Corollary. Suppose X; and X2 are subspaces of the metric
space X, then
PROBLEMS
Sections A
Section B and C
Prove the sum theorem for Ind without making use of (18.B.21)
Prove the product theorem for cov without making use of (18.B.21)
Prove (18.B.10) for Ind without making use of (18.B.21)
ie
NS
eeeLet X be a compact metric space show: dim(X) < n implies that for
each metric p on X and for each € > 0 there is a finite open cover U
such that mesh(/) < € and ord(U) < n.
5. Suppose X is a metric space and dim(X) < n > 1. Show that for each
k =0,1,2,...,n —1 there is a closed subspace Y with dim(Y) = k.
APPENDIX
A. FREE GROUPS
(Ap.A.1) Definition. Suppose A is a set. A syllable on A is an element of
Ax Z. We denote (a,n) by a”. A word on A is a finite sequence of syllables.
The word with no syllables is denoted by 1.
Let W(A) denote the set of all words on A. Define a binary operation
on W(A) , i.e., a function -, that maps W(A) x W(A) into W(A) by
The operation - converts W(A) into a semigroup (the word semigroup on A),
i.e., the binary operation - is associative, and W(A) has an identity element,
i
(Ap.A.2) Definition. Let W(A) be the word semigroup on the set A. Sup-
pose: a € A, wi, we € W(A), w = wia°we, u = wi 2, t = wya"a™ we, and
s=w a"t™ we. We say u is obtained from w by an elementary contraction
of type I and w is obtained from u by an elementary expansion of type I.
Also, the element s is obtained from t by an elementary contraction of type
II, and t is obtained from s by an elementary expansion of type II.
460 Appendix
(Ap.A.13) Remark. The reader should note that a free group with a
free basis containing more than one element is not abelian. Hence, free
groups should not be confused with free abelian groups (which are free in
the category of abelian groups and homomorphisms, but not in the category
of groups and homomorphisms).
Group Presentations 461
B. GROUP PRESENTATIONS
{A; R} 7 G
(Ap.B.4) Exercise. Show: every group has a presentation. {Hint: Use
(Ap.A.15).]
(Ap.B.5) Examples.
1. {x; } and {z,y;2} are presentations for the infinite cyclic group.
2. {x,y;cyxz~‘y~'} is a presentation of a free abelian group on two
generators (note, this is not a free group).
3. {x;x7} is a presentation for Zz (the group of integers mod 2).
4. {x,y;cyz~1y—}, x”, y®} and {xz; 2°} are presentations for Ze.
5. {z,y;2z7, y”, (cy)"} is a presentation for the dihedral group of order
2n.
(Ap.B.8) Remark. Finiteness does not work the way that it should. There
are examples of finitely presented groups which have subgroups that are
not finitely presented. The enterprising reader might seek an example of a
nonfinitely generated subgroup of {z, y; }.
(Ap.B.9) Definition. Given two disjoint groups G and H, the free product
of G and H, denoted by G x H, is the group generated by all the elements
of G and all the elements of H, subject to the relations g:g293 1 for all
91,92,93 € G with gig2 = g3, and hyhgh;* for all hi,ho,h3 € H with
hy he => hz.
Homomorphisms and Tietze's Theorem 463
Sie CaplisaU
ns ls a aePars nesses
F[X]/(R)—— FIY/(S)
Proof. Note that f, = p,fpx' is a well defined map and is the desired
homomorphism. O
ery
P Pp Pp
TGA i ify
2 al 2g GX
yields a presentation map II’: {Y ;S} + {X;R}. The mappings JI and II'
constitute Tietze equivalencies of types II and II', respectively.
te jt a (SMG
M2) = {
eet =v
(Ap.D.2) Exercise. Show that for each group H and for each family
of homomorphisms {hg : Ga > H : a € D} such that ha(ga)he(ge) =
he(ga)he(ga) for all a,8 € D and for all gg € Gag and gg € Gg, there
is a unique homomorphism h : X.¢pGa — H such that for each a € D,
hig = ha, where iq is the natural injection from G, into the direct product.
(Ap.D.3) Definition. Let K be the smallest normal subgroup of x.¢pGa
generated by {ia(g)~‘igfas(g) : a,8 € Dia < B, g € Ga}. The direct
limit of the system (Ga, fag,D) is (XaenGa) /K and is denoted by lim Gg.
(Ap.D.4) Theorem. Suppose (Ga, fag, D) is a direct system of groups and
lim G', is the corresponding direct limit. For each a, define a homomorphism
fa : Gy + limG, by fa = pia, where p is the quotient homomorphism
(we denote p(g) by [g] or [g]x) from x.epGq to iimGa = (NeenGe l/h:
Then whenever a < £ the following diagram is commutative, and limG, is
generated by U..,, Im fa:
iim Ge
an
Ga—G5—> Ga
(Ap.D.5) Examples.
1. Let D=N. For eachn €N, let G, = Z and set f,;(z) = z for all
moi. Then iim G, is isomorphic to Z.
2. Let D=N. For each n €N, let Gp = {m2" : m € Z} with +
as the group operation, and set f,;(z) = 2*-"z for n < i. Then iim G,, is
isomorphic to Z.
3. Let D=N. For each n €N, let G, = Z and set fyi(z) = 2'-"z for
n<i. Find limG,.
(Ap.D.6) Exercise. Suppose that A, B, and C are groups, and f : A > B,
g : AC are homomorphisms such that f is onto. There is a homomor-
phism h: B > C such that hf = g if and only if ker f C kerg.
: eae
Then there is a unique homomorphism h from lim G, onto H such that for
each a < #3, the following tetrahedral diagram is commutative.
im
fim Ge
Re
Ga ——> H
fap we
Ge
Ve
fim Ge
Jal
oes
Ga fap Ge
aes
Ga fap Ge
Suppose further whenever G is a group and {ho Gay Gora € Dh} isa
family of homomorphisms such that the following diagram is commutative
for each a < £,
7X
G
Go 23 Ge
Direct Limits of Groups 469
—> Sire ; .
7 + ey be erouled
bo
8» heed: Pie ke VP
DFID nde aM et © =I
<5 ; /.
Bibliography.
ALEKSANDROV, P. [1927]: Simpliziale Approximationen in der allgemeinen
Topologie, Math. Ann. Vol. 96, pp.489-511
ALEKSANDROV, P. [1939]: Bikompackte Erweiterungen topologishe Raume,
Mat. Sb. Vol. 5, pp.403-423
ALEKSANDROV, P. and Uryson, P. [1923]: Une condition nécessaire et suff-
isante pour qu’ne classe (L) soit une classe (D), C. R. Acad Sci. Paris
Vol. 177, pp.1274-1277
ALEKSANDROV, P. and Uryson, P. [1929]: Memoire sur les espaces topologi-
ques compactes, Verh. Ak. Wet. Amst. Vol. 14, pp.1-96
ALEXANDER, J. W. [1924] : On the subdivision of 3-space by a polyhedron,
Proc. Nat. Acad. Sci. Vol. 10, pp.6-8
ALEXANDER, J. W. [1924]: An example of a simply connected surface bound-
ing a region with is not simply connected, Proc. Nat. Acad. Sci. Vol. 10,
pp.8-10
ANDERSON, R. D. [1958]: One-dimensional continuous curves and a homo-
geneity theorem, Ann. of Math. Vol. 68, pp.1-16
ANDERSON, R.D. [1966]: Hilbert space is homeomorphic to the countable
infinite product of lines, Bull. Amer. Math. Soc. Vol. 72, pp.515-519
ANDERSON, R. D. and BING, R. H. [1968]: A complete elementary proof that
Hilbert space is homeomorphic to the countable infinite product of lines,
Bull. Amer. Math. Soc. Vol. 74, pp.771—792
ANDERSON, R. D. and CHOQUET, G. [1959]: A plane continuum no two
of whose nondegenerate subcontinua are homeomorphic: an application of
inverse limits, Proc. Am. Math. Soc. Vol. 10, pp.347-353
ANTOINE, L. [1921]: Sur ’homeomorphie de deux figures et de leurs voisinages,
J. Math. pures et appl. Vol. 8, pp.221-325
ARENS, R. and DuGUNDJI, J. [1951]: Topologies for function spaces, Pac. J.
Math. Vol. 1, pp.5-31
ARMENTROUT, S. [1966]: Monotone decompositions of E*, Topology Semi-
nar, Wisconsin, 1965. Annals of Mathematical Studies, No. 60, Princeton
University Press, Princeton. pp.1—-27
ARMENTROUT, S. [1967]: Topology Conference, Arizona State Univ., Tempe,
Arizona, pp.22-35
ARMENTROUT, S. [1968]: Concerning cellular decompositions of 3-manifolds
that yield 3-manifolds, Trans. Am. Math. Soc. Vol. 133, pp.307-332
ARMENTROUT, S. [1970]: A decomposition of FE? into straight arcs and single-
tons, Diss. Math. Rozp. Mat. Vol. 68, p.46
ARMENTROUT, S. and PRICE, T. [1969]: Decompositions into compact sets
with UV-properties, Trans. Am. Math. Soc. Vol. 141, pp.433-442
ARMSTRONG,M.A., RouRKE, C.P., and CooKE, G.E. [1968]: The Princeton
notes on the Hauptvermutung, Princeton Press, Princeton
BALL, B.J. [1984]: Arcwise connectedness and the persistence of errors, Am.
Math. Monthly. Vol. 91, pp.431-433
471
472
Brouwer, L.E.J. [1912]: Beweis des Jordanschen Siitz fiir den n-dimension-
alen Raum, , Math. Ann. Vol. 69, pp.314-319
Brouwer, L.E.J. [1912]: Beweis des Jordanschen Satz fiir den n-dimension-
alen Gebietes, , Math. Ann. Vol. 71, pp.55-56
Brown, M. [1960a]: Some applications of an approximation theorem for in-
verse limits, Proc. Am. Math. Soc. Vol. 11, pp.478-483
Brown, M. [1960b]: A proof of the generalized Schoenflies theorem. Bull.
Am. Math. Soc. Vol. 66, pp.74-76
Brown, M. [1961]: The monotone union of open n-cells is an open n-cell,
Proc. Am. Math. Soc. Vol. 12, pp.812-814
Brown, M. [1962a]: Locally flat embeddings of topological manifolds, Topol-
ogy of 3-manifolds and Related Topics, pp.83-91. Prentice-Hall, Englewood
Cliffs, N.J.
Brown, M. [1962b]: A mapping theorem for untriangulated manifolds, Topol-
ogy of 3-mantfolds and Related Topics, pp.92-94. Prentice-Hall, Englewood
Cliffs, N.J.
Brown, R.F. [1974]: Elementary consequences of the noncontractibility of the
circle, Am. Math. Monthly Vol. 81, pp.247-252
BURGESS, C.E. [1959]: Chainable continua and indecomposibility, Pac. J.
Math. Vol. 9, pp.653-659
BurRGEss, C.E. [1961]: Homogeneous continua which are almost chainable,
Can. J. Math. Vol. 18, pp.519-528
BurRGEss, C.E. and CANNON, J.W. [1971]: Embeddings of surfaces in E?,
Rocky Mt. J. Math. Vol. 1, pp.259--344
Cantor, G. [1883]: Uber unendliche lineare Punktmannigfaltigkeiten, Math.
Ann. Vol. 21, pp.545-591
CAPEL, C.E. [1954]: Inverse limit spaces, Duke Math. J. Vol. 21, pp.233-245
CARTAN, H. [1937a]: Théorie des Filtres, C. R. Acad. Sci. Paris Vol. 205,
pp.595—-598
CaRTAN, H. [1937b]: Filtres et Ultrafiltres, C. R. Acad. Sci. Paris Vol. 205,
777-779
CASLER, B.G. [1965]: An imbedding theorem for connected 3-manifolds with
boundary, Proc Am. Math. Soc. Vol. 16, pp.559-566
Casson, A.J. [1967]: Generalizations and applications of block bundles, Fel-
lowship dissertation Trinity College Library, Cambridge
CaTLIN, D.E. [1968]: A short proof of the nest characterization of compactness,
Am. Math. Monthly Vol. 75, p.751
CHERNOFF, P.R. [1992]: A simple proof of Tychonoff’s theorem via nets,
Amer. Math. Monthly Vol. 99, pp.932-934
CHITTENDEN, E.W. [1927]: On the metrization problem and related problems
in the theory of abstract sets, Bull. Am. Math. Soc. Vol. 33, pp;13-34
CHRISTENSON, C.O. and OsBoRNE, R.P. [1968]: Pointlike subsets of a man-
ifold, Pac. J. Math. Vol. 24, pp;431-435
Coss, J.I. and VoxMAN, W.L. [1972]: Some fixed point results for UV de-
compositions of compact metric spaces, Proc. Am. Math. Soc. Vol. 38,
pp.156-160
474
Coss, J.I. and VOXMAN, W.L. [1980]: Dispersion points and fixed points,
Am. Math. Monthly, Vol. 87, pp.278-281
CouEN, D.E. [1954]: Spaces with weak topology. Quart. J. Math. Ozford,
Vol. 5, pp.77-80
ComFort, W.W. [1969]: A short proof of Marczewski’s separability theorem.
Am. Math. Monthly, Vol. 76, pp.1041—1042
CROWELL, R.H. and Fox, R.H. [1963]: Introduction to Knot Theory, Ginn
and Company, Boston
DAVERMAN, R.J. [1986]: Decompositions of Manifolds, Academic Press, Or-
lando
DEBNATH, L. and MIKUSINSKI, P. [1990]: Introduction to Hilbert spaces with
Applications, Academic Press, San Diego
DIEUDONNE J. [1944]: Uné généralization des espaces compacts, J. Math. pures
et appl. Vol. 23, pp.65—76
Do p, A. [1972]: Lectures on Algebraic Topology, Springer Verlag, Berlin
Do Lp, A. [1963]: Partitions of unity in the theory of fibrations, Ann. of Math.
Vol. 78, pp.223-255
DuGUNDJI, J. [1966]: Topology, Allyn and Bacon, Boston
Dyer, E. [1956]: A fixed point theorem, Proc. Am. Math. Soc. Vol. 7,
pp.662-672
Dyer, E. [1956]: Certain transformations which lower dimension, Ann. of
Math. Vol. 63, pp.15-19
EATON, W.T. and PIXLEy, C. [1975]: S! cross and UV© decompositions
of S* yileds S! x S1, Geometric Topology 1974, Lecture Notes Vol. 438,
Springer-Verlag, Berlin pp.166—194
EpGAR. G.A. [1990]: Measure, Topology and Fractal Geometry, Springer-
Verlag, New York
ENGELKING, R. [1977]: Dimension Theory, North Holland, Amsterdam
FALCONER, K. [1990]: Fractal Geometry: Mathematical Foundations and Ap-
plications,John Wiley & Sons, Chichester
FLEISSNER, W.G. [1984]: The normal Moore space conjecture and large cardi-
nals, Handbook of Set Theoretic Topology, (ed. Kunen and Vaughan), North
Holland, Amsterdam pp.733-760
Fox, R.H. [1943]: On homotopy type and deformation retracts, Ann. of Math.
Vol. 44, pp.40-50
Fox, R.H. [1945]: On topologies for function spaces, Bull. Am. Math. Soc.
Vol. 51, pp.429-432
Fox, R.H. [1953]: Free differential calculus I, Derivation in the free group ring,
Ann. of Math. Vol. 57, pp.547-560
Fox, R.H. and Artin, E. [1948]: Some wild cells and spheres in three-
dimensional space, Ann. of Math. Vol. 49, pp.979-990
FRASER, R.B. Jr. [1972]: A new characterization of Peano continua, Prace
Mat. Vol. 16, pp.247-248
FRINK, A.H. [1937]: Distance functions and the metrization problem, Bull.
Am. Math. Soc. Vol. 48, pp.133 -142
475
JOLLEY, R.F. and RocsErs, J.T. JR. [1970]: Inverse limit spaces defined by
only finitely many distinct bonding maps, Fund. Math. Vol. 68, pp.117-120
JONES, F.B. [1937]: Concerning normal and completely normal spaces, Bull.
Am. Math. Soc. Vol. 438, pp.671-677
JONES, F.B. [1951]: Certain homogeneous unicoherent indecomposible con-
tinua, Proc. Am. Math. Soc. Vol. 2, pp.855-859
KAMPEN, E.R. VAN [1933]: On the connection between the fundamental group
of some related spaces, Am. J. Math. Vol. 55, pp.261-267
KATETOV, M. [1952]: On the dimension of non-separable spaces I., Czech.
Math. J. Vol. 2, pp.333-368
KELLEY, L.J. [1950a]: Convergence in Topology, Duke Math. J. Vol. 17,
pp.277-283
KELLEY, L.J. [1950b]: The Tychonoff product theorem implies the axiom of
choice, Fund. Math. Vol. 37, pp.75-76
KELLEY, F.J. [1955]: General Topology, Van Nostrand, New York
KIRBY, R.C. and SIEBENMANN, L.C. [1969]: On the triangulation of manifolds
and the Hauptvermutung, Bull. Am. Math. Soc. Vol. 75, pp.742—749
KirBy, R.C., SEIBENMANN, L.C., and Wall, C.T.C. [1969]: The annulus
conjecture and triangulation, Not. Am. Math. Soc. Vol. 16, p.432
KNASTER, B. and KURATOWSKI, C. [1921]: Sur les ensembles connexes, Fund.
Math. Vol. 2, pp.206—255
KNIGHT, C.J. [1964]: Box topologies, Quart. J. Math. Ozford Vol. 15, pp.41-
54
KRESIMAR, D. and MARDESI¢, S. [1968]: A necessary and sufficient condition
for the n-dimensionality of inverse limits, Proceedings of the International
Symposium of Topology and its Applications, Belgrade, pp.124—129
LASHOF, R.K. and ROTHENBERG, M.G. [1968]: Hauptvermutung for man-
ifolds, Proc. 1967 Conference on Top. of Man., Prindle Weber Schmidt,
pp.81-105
LEVINE, N. [1960]: Remarks on uniform continuity in metric spaces, Am.
Math. Monthly Vol. 67, pp.562-563
MAGNus, W., KARRASS, A. and SOLITAR, D. [1966]: Combinatorial Group
Theory, Wiley, New York
Markov, A.A. [1960]: Unsolvability of the homeomorphism problem, Proc.
Int. Cong. Math., 1958, Cambridge Univ. Press, Cambridge
MassgEy, W.S. [1967]: Algebraic Topology, An Introduction, Harcourt, Brace,
and World, New York
MAUNDER, C.R.F. [1970]: Algebraic Topology, Van Nostrand Reinhold, Lon-
don
MAZURKIEWICZ, S. [1920]: Sur les lignes de Jordan, Fund. Math. Vol. 1,
pp.166—209
McAULEY, L.F. [1962]: Upper semicontinuous decompositions of E? into E3
and generalizations to metric spaces, Topology of 3-manifolds and Related
Topics, Prentice-Hall, Engrlewood Cliffs N.J., pp.21-36
MENGER, K. [1923]: Uber die Dimensionalitat von Punktmenger I, Monat.
Fiir Math. und Phys. Vol. 33, pp.148-160
477
MILL, J. VAN and Reed, G.M. [1990]: Open Problems in Topology, North
Holland, Amsterdam
Minor, J. [1961]: Two complexes which are homeomorphic but combinato-
rially distinct, Ann. of Math. Vol. 74, pp.575-590
MolsE, E.E. [1948]: An indecomposable plane continuum which is homeo-
morphic to each of its nondegenerate subcontinua, Trans. Am. Math. Soc.
Vol. 63, pp.581-594
Molsg, E.E. [1949]: Grille decomposition and convexification theorems for
compact metric locally connected continua, Bull. Am. Math. Soc. Vol. 55,
pp.1111-1121
MoIsE, E.E. [1952a]: Affine structures in 3-manifolds II. Positional properites
of 2-spheres, Ann. of Math. Vol. 55, pp.172-176
MoIsE, E.E. [1952b]: Affine structures in 3-manifolds V. The triangulation
theorem and Hauptvermutung, Ann. of Math. Vol. 56, pp.96-114
Moise, E.E. [1954]: Affine structures in 3-manifolds VIII; invariance of knot
types, local tame embeddings, Ann. of Math. Vol. 59, pp.159-170
Monk, J.D. [1969]: Introduction to Set Theory, McGraw-Hill, New York
Moore, R.L. [1925]: Concerning upper semi-continuous collections of con-
tinua, Trans. Am. Math. Soc. Vol. 27, pp.416—428
Moore, E.H. and SMITH, H.L. [1922]: A general theory of limits, Am. J.
Math. Vol. 44, pp.102-121
Morita, K. [1954]: Normal families and dimension theory for metric spaces,
Math. Ann. Vol. 128, pp.350-312
MUNKRES, J.R. [1975]: A First Course in Topology, Prentice-Hall, Englewoods
Cliffs N.J.
NADLER, S.B. JR. [1970]: A note on inverse limits of finite spaces, Can. Math.
Bull. Vol. 13, pp.69-70
NADLER, S.B. JR. [1973]: The indecomposibility of the dyadic solenoid, Am.
Math. Monthly Vol. 80, pp.677-679
NADLER, S.B. JR. [1978]: Hyperspaces of Sets, Marcel Dekker, New York
NADLER, S.B. JR. [1992]: Continuum theory, an introduction, Marcel Dekker,
New York
NAGAMI, K. [1970]: Dimension Theory, Academic Press, New York
NaGaTA, J. [1950]: On a necessary and sufficient condition of metrizability, J.
Inst. Poly. Osaka City Univ. Vol. 1, pp.93-100
NEWMANN, M.H.A. [1964]: Elements of the topology of plane sets of points,
Cambridge Univ. Press, London
NEWMANN, M.H.A. [1966]: The engulfing theorem for topological manifolds,
Ann. of Math. Vol. 84, pp.555-571
O.uM, P. [1958]: Nonabelian cohomology and van Kampen’s theorem, Ann.
of Math. Vol. 68, pp.658-668
PAPAKYRIAKOPOULOS, C.D. [1943]: A new proof of the invariance of the ho-
mology groups of a complex, Bull. Soc. Math. Gréce Vol. 22, pp.1-54
PITTMAN, C.R. [1979]: An elementary proof of the triod theorem, Proc. Am.
Math. Soc. Vol. 25, p.919
478
Annulus: neighborhood; 16
2-; 130 Bd; 406, 419, 421
n-; 423 Big inductive dimension; 446
half-open; 423 Bing’s metrization theorem; 259
open; 423 Bolzano-Weierstrass theorem; 80
Annulus conjecture; 423 Bonding maps; 158
Annulus theorem; 147, 423 Borsuk map; 379
ANR; (see Absolute neighborhood re- Borsuk’s top hat theorem; 375
tract) Borsuk-Ulam theorem; 337-338
ANRy;; (see Metric absolute neighbor- Bound:
hood retract), greatest lower; 4,
AR, ar; (see (A)absolute retract) least upper; 4
Arc. 56 lower; 4
polygonal-; 59 upper; 4
wild; 392 Boundary; 406, 407, 419
Arc-connected space; (see Connected Branch point; 383
space) Brick partition; 132
482 Index
F(A]; 460
E1; 48, 50, 52, 61, 69-71, 121 F, subset; 103, 247
E15 251,55,.62,/65 face; 342, 351
Edge loop; 361 Factor space; 158
Index 485
[Metric] ING I
Hausdorff; 90 Nagata;Smirnov metrization theorem;
normal; 115 259, 267
post office; 38 Necking number; 138
pseudo-; 38 Neighborhood; 27
starfish; 266 Neighborhood basis; (see Basis)
sup; 178 Nerve; 370
taxicab; 18 Nest; 6, 64
topology; 17 . Net; 30, 272
usual; 17 universal; 279
Metric Absolute Neighborhood Retract No retraction theorem; 125, 296, 359
(ANRm); 107, 108, 256, 346, 371 Non-cut point; 222
Metrically isolated; 91 non-degenerate element; 210
Metric space; 16, 225 Norm; 39
bounded; 19 Normal form; 412
chainable; 72, 231 Normal metric; (see Metric)
chainable from a to 6; 231 Normal projection; 393
complete; 82, 157, 184, 189 Normal space; (see Topological space)
convex; 89 Normed vector space; 39
[Sphere] an arc; 55
m= 11061137, 2038391, 421 a path; 53
bicollared; 430 Theta curve; 150
collared; 430 Thread; 158
tame; 137, 423 Tietze equivalence; 464
wild; 137, 391, 423 Tietze’s extension theorem; 103
Spine; 414 Tietze theorem; 464
Standard n-ball; (see B”) Tihonov plank; 98, 110
Standard n-sphere; (see S”) Tihonov space; (see Topological space)
Star: Tihonov’s theorem; 154, 172, 277, 279
convex; 294 Topological invariant; 33
of a simplex; 347 Topological pair; 285
with respect to a cover; 252, 451 Topological space; 12
Stone’s metrization theorem; 213, 257 arc connected; (see Connected space)
Strongly continuous; (see Function) Baire; 84
Subbasis; 152 binormal; 375, 397
Subcomplex: collectionwise normal; 259
of a CW-complex; 206 compact; (see Compact space)
of a simplicial complex; 343 completely normal; 98
Subcover; 60 completely regular; 114