Mathematics 205A

Introduction to Topology — I

Course Notes — Part One

Revised, Fall 2005, Fall 2008, Fall 2014

Department of Mathematics

University of California, Riverside

 Table of Contents

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii

I. Foundational material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1. Basic set theory(1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2. Products, relations and functions (1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
3. Cardinal numbers(1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
4. The real number system(1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

II. Metric and topological spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1. Metrics and topologies(2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2. Closed sets and limit points(2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18
3. Continuous functions(2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4. Cartesian products(2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

III. Spaces with special properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

1. Compact spaces — I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2. Complete metric spaces(1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3. Implications of completeness(1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4. Connected spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5. Variants of connectedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

IV. Function spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .71

1. General properties(0) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2. Adjoint equivalences(0) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

V. Constructions on spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
1. Quotient spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2. Sums and cutting and pasting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

VI. Spaces with additional properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

1. Second countable spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
2. Compact spaces – II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3. Separation axioms(2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4. Local compactness and compactifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5. Metrization theorems(0) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

Appendices:

A. Topological groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

B. Stereographic projection and inverse geometry . . . . . . . . . . . . . . . . . . . . . . 129
C. Homogeneity of open sets in Euclidean spaces . . . . . . . . . . . . . . . . . . . . . . 133
D. Normal forms for orthogonal transformations . . . . . . . . . . . . . . . . . . . . . . . 137

References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .140
Summary of Files in the Course Directory . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
 Introduction

These first entry level graduate course in topology and geometry splits naturally into two parts:
1. Topics from point set (also known as general) topology.
2. Fundamental groups and covering spaces; only part of this subject is covered in the first
course, and the remainder is covered in the second course.
These notes cover material for the first part of the course. The corresponding notes for the
second part of the course are in the document fundgp-notes.pdf.
The main text for both parts of the course is the following classic book on the subject:
J. R. Munkres. Topology (Second Edition), Prentice-Hall, Saddle River NJ, 2000.
ISBN: 0–13–181629–2.

The choice of topics to be covered

Most beginning graduate students have seen at least some material from point set topology in
an undergraduate real variables course, and some have seen even more material in undergraduate
topology courses, but the topics vary depending upon the institution and instructor. Therefore
these notes develop the subject from the beginning for the sake of completeness, but the course
itself will spend little if any time on some topics widely covered in undergraduate real variables
courses, and coverage of a few other topics will focus on the generalizations of other key concepts
and results to abstract topological spaces. Sections of the notes with limited or no course coverage
are flagged in the notes with numerical superscripts which have the following meanings:
Section name(3) Some material from undergraduate real variables courses that will be
skipped in the course itself. However, everything in this section is part of the material
covered in course and qualifying examinations.
Section name(2) Material probably seen in previous courses, only covered lightly to
provide a basis for material in the course itself; everything in this section is also part of
the material covered in course and qualifying examinations.
Section name(1) Prerequisite material from set theory or real variables courses, not
covered in the course or qualifying examinations. Section III.3 is an exception; the material
up to, but not including, the subheading Baire spaces is part of the material covered in
course and qualifying examinations.
Section name(0) Not covered in the course or qualifying examinations, included mainly
for reference purposes. Section VI.5 is an exception to this rule; the statements of the
main theorems in this section should be understood, but there is no need to know the
proofs.

iii
 Numbering conventions for results

Notation like “Theorem X.11.22” will indicate Theorem 22 in Section X.11 of the notes (but
the specific citation in this sentence does not correspond to anything in the notes — there is no
Unit X).

Some alternate and additional references

The online directory

http://math.ucr.edu/∼res/math145A-2014/
contains extensive material for a point set topology course at the undergraduate level and may be
useful as a source of information on basic topics which are either covered lightly or not at all in the
present course. In particular, the file
http://math.ucr.edu/∼res/math145A-2014/geom-background.pdf
includes links to undergraduate mathematics courses whose contents led to the development of
material in this course.
The books by Dugundji and Kelley in the bibliography are excellent graduate level point set
topology texts, and each of these views the subject somewhat differently from the perspective in
Munkres and these notes. The following more recent text is also a very good alternate reference
for much of the material in this course:
T. Lawson. Topology: A Geometric Approach. Oxford University Press, New York–
etc., 2003. ISBN: 0-19-920248-6.
Finally, we should also mention the undergraduate level textbook around which the files in the
directory http://math.ucr.edu/∼res/math145A-2014/ are organized.
W. A. Sutherland. Introduction to Metric and Topological Spaces. (Second Ed.)
Oxford University Press, New York–etc., 2009. ISBN: 019956308X.
There is also a very useful companion website for this book:
http://www.oup.com/uk/booksites/content/9780199563081/
Among other things, this site includes topics and details not in Sutherland’s book, and there are
also solutions to half of the exercises.
NOTE ON WEB LINKS. Many of the links to the Word Wide Web in these notes are so long
and complicated that they can be frustrating to enter from a keyboard, but unfortunately some
links are not clickable due to software constraints; if such cases, there is a clickable link in the file
math205Awww2014.pdf.

References to Sutherland for these notes

For several reasons such as review or slightly different approaches, some students might find
it helpful to use Sutherland’s book as a supplementary reference for this course. Therefore we
are including a table which gives the chapters in Sutherland (and the notes on these chapters in
the files http://math.ucr.edu/∼res/math145A-2014/math145Anotes∗.pdf where the numbers
in the wild card string ∗ refer to one or more chapters in Sutherland) which correspond to the
appropriate sections of these notes; sections with little or no corresponding material in Sutherland
are not listed in the table.

iv
 Section in these notes Chapter(s) in Sutherland

Introduction 1 (and Prefaces)

I.1 2
I.2 2–3
I.4 4

II.1 5 – 8, 11, 16
II.2 6–9
II.3 5–8
II.4 5, 10

III.1 13
III.2 16 – 17
III.3 17
III.4 12
III.5 12

V.1 15

VI.1 8
VI.2 13 – 14
VI.3 7, 11 – 12
VI.5 11

Secondary course goals

We have already noted that all students in an entry level graduate course have seen simplified
versions of many (if not most) concepts and results from the first part of Mathematics 205A in pre-
requisite courses. Some of the reasons for the organization of the course go beyond the introduction
of new material and the coverage of many topics in greater detail than in the prerequisites. These
include (1) the presentation of an abstract, unified approach to some basic topics in mathemat-
ics, (2) illustrations that increasing formality and abstraction sometimes yield simpler and more
conceptual understandings of basic facts, (3) improvement of skills in reading and writing proofs.
The enhancement of proof writing skills is particularly important, and it is indispensable for
studying mathematics at the graduate level. Many proofs at this level are significantly more difficult
than their counterparts at the undergraduate level. In particular, the level of abstraction is usually
higher, the arguments are frequently much longer or more complicated, simple steps in arguments
are sometimes omitted or mentioned only briefly, and often the approaches are far less direct, in
many cases with crucial steps relying on points which at first are easily overlooked.
A review of mathematical proofs at the undergraduate level is given in the course directory file
mathproofs.pdf, and a few additional suggestions are given in the file math205Asolutions00.pdf.
Of course, many other articles on writing mathematical proofs can be found by searching for phrases
like writing proofs and/or writing proofs topology using Google or a similar search engine. General
comments about such searches appear in the file aabInternetresources.pdf.

v
 Remarks on priorities

In order to write mathematical proofs at the graduate level, it is necessary to understand

most proofs in the texts and course notes so thoroughly that they can be explained convincingly to
someone else who has adequate background knowledge. However, there are proofs for which a more
passive understanding of the proof is enough; in other words, the reader should be understand the
assertions well enough to conclude whether or not they are valid, but for a various reasons there is
no need to have an active understanding as described in the previous sentence.
Such proofs already arise in prerequisite courses, and one example involves the existence and
uniqueness (up to order-preserving algebraic isomorphism) of a system satisfying the standard
axioms for the real number system. Without these existence and uniqueness results, most of
mathematics would not have an acceptable logical foundation. However, in each case there are
reasons why it is not that important to dwell too much upon the proofs. There are two standard
proofs of existence, one due to R. Dedekind and the other due to G. Cantor. In each case, there is
a formal construction based upon a reasonable idea (for the Dedekind approach it is the idea that a
real number is specified by the rational numbers which are strictly less than it, and for the Cantor
approach it is the idea that a real number is the limit of a sequence of rational numbers), but the
verifications are generally tedious and the methods of proof are not useful for much else. One might
compare the details of verification to the scaffolding in a large construction project: The structure
is absolutely necessary, but it can be set aside once everything is completed. The following quote
from the Sherlock Holmes story, A Study in Scarlet by A. C. Doyle (1859–1930), reflects the need
to leave some details in the background.
· · · My [Watson’s] surprise reached a climax, however, when I found incidentally that
he [Holmes] was ignorant of the Copernican Theory and of the composition of the Solar
System. That any civilized human being in this nineteenth century should not be aware
that the earth traveled round the sun appeared to be to me such an extraordinary fact
that I could hardly realize it.
“You appear to be astonished,” he said, smiling at my expression of surprise. “Now that
I do know it I shall do my best to forget it.”
“To forget it!”
“You see,” he explained, “I consider that a man’s brain originally is like a little empty
attic, and you have to stock it with such furniture as you choose. A fool takes in all the
lumber of every sort that he comes across, so that the knowledge which might be useful
to him gets crowded out, or at best is jumbled up with a lot of other things so that he has
a difficulty in laying his hands upon it. Now the skilful workman is very careful indeed as
to what he takes into his brain-attic. He will have nothing but the tools which may help
him in doing his work, but of these he has a large assortment, and all in the most perfect
order. It is a mistake to think that that little room has elastic walls and can distend to
any extent. Depend upon it there comes a time when for every addition of knowledge you
forget something that you knew before. It is of the highest importance, therefore, not to
have useless facts elbowing out the useful ones.”
“But the Solar System!” I protested”
“What the deuce is it to me?” he interrupted impatiently; “you say that we go round the
sun. If we went round the moon it would not make a pennyworth of difference to me or
to my work.”
We should add an important qualification to this analogy. For the sorts of mathematical proofs
we describe, it is highly desirable to know enough about the proof that one can summarize the

vi
main steps in the argument to someone with adequate background. In the case of the existence
and uniqueness proofs for the real numbers, this may be done as follows:
Existence.
The Cantor or Dedekind construction of objects which correspond to our intuitive idea of
how real numbers can be specified.
Define reasonable candidates for addition, multiplication and ordering of the constructed
objects.
Show that these defined notions satisfy the axioms for a real number system.
Uniqueness.
Define the map from one system to another starting by sending the zero and unit to one
into the zero and unit of the other.
Recursively define a map ϕ from the “positive integers” in one system to the “positive
integers” in the other by sending n + 1 times the unit in the first system to ϕ(n) + unit
in the second.
Prove that ϕ is 1–1 onto, order preserving, and also preserves sums and products.
Extend ϕ to a map from the “rational numbers” in one system to the “rational numbers”
in the other, and prove that this extension is also 1–1 onto, order preserving, and also
preserves sums and products.
Extend ϕ to a map from the entire first system to the entire system using the fact that
an element of a complete ordered field is determined by the “rational numbers” which are
strictly less than it.
Verify that this extension is 1–1 onto, order preserving, and also preserves sums and
products.
The preceding should also apply to several basic results (e.g., Tychonoff’s Theorem in Section III.1
and several results in Section VI.5) which are stated without proof in the notes.

vii
 I. Foundational Material

Since this is a graduate course in mathematics, it is appropriate to assume some background

material as well as some experience in working with it. In particular, we shall presume that the
reader has a working knowledge of basic set theory; however, familiarity with a fully developed
axiomatic approach to that subject will not be needed. For the sake of completeness, the following
files on foundational material have been placed into the course directory:
foundations1.pdf : This is a review of mathematical logic and simple set-theoretic
algebra.
foundations2.pdf : This discusses some material in set theory that either deviates
slightly from the treatment in Munkres or is not covered there. Also included are dis-
cussions of the Axiom of Choice and the Generalized Continuum Hypothesis.
realnumbers.pdf : This gives a list of axioms for the real numbers and some important
properties, including those that play a major role in this course.
uniqreals.pdf : This proves that the axioms for the real numbers completely characterize
the latter up to a structure preserving one-to-one correspondence (i.e,, an isomorphism).
One needs to prove a result of this sort in order to talk about THE real number system. The
uniqueness proof is not particularly difficult but it is rather tedious; given the importance
of uniqueness, it is necessary to look through this argument carefully at least once and to
understand it at least passively.
numberexpansions.pdf : Starting with the standard characterization of the real numbers
as a complete ordered field (see Section 4 of Munkres), this document gives proofs for the
standard elementary facts about base N expansions of integers and decimal expansions of
real numbers that are used constantly in computational work. These proofs may provide
useful examples to illustrate how mathematics uses logical arguments to justify its basic
rules and relationships. Once again, at least a passive understanding of the arguments is
worthwhile.
categories.pdf : Category theory will not be used explicitly in this course, but given
its fundamental nature a discussion of the basic ideas and some examples is included for
reference.
In addition, a fairly complete set of lecture notes for an upper level undergraduate course in set
theory (Mathematics 144) along with several other documents will be available for downloading in
the online directory http://math.ucr.edu/∼res/math144 . This directory contains lecture notes
for Mathematics 144 in the file(s) setsnotesn.pdf where 1 ≤ n ≤ 8, and there are several related
supplementary files, most notably the following:
ordertypecount.pdf
productposet.pdf
setsnotes3add1.pdf
subrationals.pdf
transcendentals.pdf
Several files of exercises in set theory are also available in the directory as
math144exercisesn.pdf

1
and solutions are given in files named math144solutionsn.pdf (in the first case the variable n
corresponds to the unit numbers from the notes, but in the second it does not).
NOTES ON INTERNET REFERENCES. At several points there will be references to Internet
sites. The file aabInternetresources.pdf discusses reliability issues concerning these (and other
Internet) sources. Clickable links to the Internet references in these notes are given in the file
math205Awww2014.pdf in the course directory.

Coverage of topics in this unit

Most of the discussion of foundational material in the notes below will concentrate on points
that are not worked out in detail in Munkres but are important for the course.

I.1 : Basic set theory

(Munkres, §§ 1, 2, 3, 9)

As indicated on page 3 of Munkres, our approach to questions in logic and set theory is
to assume some familiarity with the most elementary ideas (the so-called “naı̈ve” approach) and
to discuss what else is needed without spending so much time on these subjects that important
material in topology must be omitted.
Munkres mentions that an overly casual approach to set theory can lead to logical paradoxes.
For example, this happens if we try to consider the “set of all sets.” During the early part of the
twentieth century mathematicians realized that problems with such things could be avoided by
stipulating that sets cannot be “too large,” and effective safeguards to eliminate such difficulties
were built into the formal axioms for set theory. One simple and reliable way of avoiding such
problems with the informal approach to set theory in this course is to assume that all constructions
take place in some extremely large set that is viewed as universal. This is consistent with the
Gödel-Bernays-von Neumann axiomatic approach to set theory, in which one handles the problem
by considering two types of collections of objects: The CLASSES can be fairly arbitrary, but
the SETS are constrained by a simple logical condition (specifically, they need to belong to some
other class). Further information can be found in the online set theory notes mentioned above. —
Viewing everything as contained in some very large set is a strongly recommended option for this
course if difficulties ever arise.

The Axiom of Choice

Since this topic is not always covered in undergraduate set theory courses but plays a crucial
role in mathematics at higher levels, it is important to read and understand the treatment given
in Sections 9 and 10 of Munkres. We shall generally use the Axiom of Choice only at the level of
the discussion in these sections of the text. Section 11 of Munkres contains additional results on
this topic which are used frequently in modern mathematics; still further information is given in
the Supplementary Exercises following Section 11 and in the online references for set theory that
were mentioned previously.
In addition to the finite Axiom of Choice, one also sees a countable Axiom of Choice, which
states that if {An } is a sequence of nonempty sets (indexed by the nonnegative integers), then one
can find a sequence of points {an } such that an ∈ An for all n. This version is also widely accepted,
even in many contexts where the general Axiom of Choice is not.

2
 I.2 : Products, relations and functions

(Munkres, §§ 5, 6, 8)

The main differences between this course and Munkres are outlined in the first few pages of
the file foundations2.pdf that was mentioned previously.

Definitions of relations and functions

Since functions play such an important role in topology, it is appropriate to stress the differ-
ences between the definition of functions in this course and the definitions that appear in many
mathematical textbooks.
Our definition of a function is equivalent to that of Munkres, but the crucial difference between
the definitions of relations and functions in Munkres and this course is that the source set (or
domain) and the target set (or codomain) are explicit pieces of data in the definition of a function.
Generally the source or domain is unambiguous because usually (but not always!) it is the set on
which the function is defined. However, the target or codomain is often ignored. The target must
contain the image of the function, but for each set containing the image there is a different function
whose target is the given set. For example, this means that we distinguish between the functions
f : R → R and g : R → [0, 1] defined by the same formula:

1
f (x) = g(x) =
x2 +1

In our setting these functions are related by the composition identity f = j o g, where j : [0, 1] → R
is the inclusion mapping sending an element of the unit interval into itself. One particular context
in which it is necessary to make such distinctions is the construction of the fundamental group as
in Chapter 9 of Munkres.
We should also note that the need to specify codomains is important in certain standard
applications of the mathematical sciences; for example, when one defines a numerical function in
many computer languages, it is often necessary to specify whether the values should be stored as
whole numbers or rounded off decimal expressions.

Comments on terminology

Mathematicians have two alternate lists of words for discussing functions that are 1–1 and
onto, and for set theory and many other branches of mathematics these are used interchangeably.
One of these lists is given on page 18 of Munkres: A map if injective if it is 1–1, surjective if it
is onto, and bijective if it is both (i.e., a 1–1 correspondence). The terms monomorphism and
epimorphism from category theory are often also used in set theory as synonyms for 1–1 and onto.
However, some care is needed when using these category-theoretic terms for functions in discussing
morphisms of topological objects, so we shall (try to) avoid using such terminology here.

3
 Images and inverse images of subsets

Given a function f from a set X to a set Y , it is often necessary to consider the image of a
subset A ⊂ X or the inverse image of a set B ⊂ Y with respect to f . These are defined on page
19 of Munkres, and there are some comments about ambiguities in the notation if f is 1–1 and
onto. However, there is also another potential ambiguity worth noting. Namely, if we have a set
A such that both A ⊂ X and A ∈ X are true, then the functional value f (A) ∈ Y and the image
of A under F , which is a subset of Y , need not be the same, even though the notation of Munkres
uses f (A) for both. In practice there are usually no such problems with possible ambiguities, but
strictly speaking one should should be able to distinguishing these two objects. We shall (attempt
to) follow the convention in Kelley’s General Topology and denoted the images and inverse images
as above by f [A] and f −1 [B] respectively.
While on the topic of notational differences, we should also note that increasing numbers of
mathematical writings are using A\B instead of A − B to denote the set-theoretic difference of all
points in A but not in B (for example, the first notaton is used in Sutherland’s book). We shall
not attempt to discuss the relative advantages or disadvantages of either option.

Disjoint unions

We shall also use an elementary set-theoretic construction that does not appear in Munkres;
namely, given two sets A and B ` we need to have a disjoint union (or direct sum, free union or
coproduct), written A t B or A B, which is a union of two disjoint subsets that are essentially
identical copies of A and B. Formally this set is defined by the identity
a

A B = A × {1}) ∪ B × {2} ⊂ A ∪ B × {1, 2}

and the file disjoint-union.pdf contains a drawing to illustrate the construction. One use of
disjoint unions will appear in the discussion of cardinal numbers; other applications will be discussed
in the main body of the course (in particular, see the section, “Sums and cutting and pasting”).

I.3 : Cardinal numbers

(Munkres, §§ 4, 7, 9)

The theory of cardinal numbers for infinite sets illustrates that set theory is not just a more
sophisticated approach to well-known mathematical concepts and that it is also a powerful tool for
obtaining important new insights into mathematics.
Definition. Two sets A and B have the same cardinality (or cardinal number) if there is a 1–1
onto map (or 1–1 correspondence) f : A → B. Frequently we write |A| = |B| in this case; it follows
immediately that this relation is reflexive, symmetric and transitive.
We begin with two basic properties of infinite sets:
PROPOSITION 1. Every infinite set contains a subset that is in 1 − 1 correspondence with
the positive integers.

4
PROPOSITION 2. (Galileo’s Paradox) A set A is infinite if and only if there is a 1 − 1
correspondence between A and a proper subset of A.

Partial ordering of cardinalities

Definition. If A and B are sets, we write |A| ≤ |B| if there is a 1–1 map from A to B.
It follows immediately that this relation is transitive and reflexive, but the proof that it is
symmetric is decidedly nontrivial:
THEOREM 3. (Schröder-Bernstein Theorem) If A and B are sets such that there are 1 − 1
maps A → B and B → A, then |A| = |B|.
Sketch of proof. This is the classic argument from Birkhoff and MacLane’s Survey of Modern
Algebra (see page 340 in the Third Edition).
Let f : A → B and g : B → A be the 1–1 mappings. Each a ∈ A is the image of at most one
parent element b ∈ B; in turn, the latter (if it exists) has at most one parent element in A, and so
on. The idea is to trace back the ancestry of each element as far as possible. For each point in A
or B there are exactly three possibilities; namely, the ancestral chain may go back forever, it may
end in A, or it may end in B.
Split A and B into three pieces corresponding to these cases, and call the pieces A 1 , A2 , A3
and B1 , B2 , B3 (the possibilities are ordered as in the previous paragraph).
The map f defines a 1–1 correspondence between A 1 and B1 (and likewise for g). Furthermore,
g defines a 1–1 correspondence from B 2 to A2 , and f defines a 1–1 correspondence from B 3 to A3 .
If we combine these 1–1 correspondences A 1 ←→ B1 , A2 ←→ B2 and A3 ←→ B3 , we obtain a 1–1
correspondence between all of A and all of B.
If we are only considering subsets of some large set, then we can define the ”cardinal number”
of a subset to the the equivalence class of that set under the relation A ∼ B ⇐⇒ |A| = |B|. If one
assumes the Axiom of Choice it is possible to give an equivalent definition of cardinal number that
extends to all sets, but for the time being we shall not worry about this point.
The Schröder-Bernstein Theorem and the preceding observations imply that if A is an infinite
subset of the positive integers N+ , then |A| = |N+ |. Since we also know that |A| ≤ |N+ | for every
infinite set A, it follows that |N+ | can be viewed as the unique smallest infinite cardinal number
and that it is ≤ every other infinite cardinal number. Following Cantor’s notation this cardinal
number is generally denoted by ℵ0 (aleph-null).
In contrast, Cantor’s Diagonal Process technique implies that in effect there is no largest
infinite cardinal number.
THEOREM 4. If A is a set and P(A) is the set of all subsets of A then |A| < |P(A)|.
Sketch of proof. The first step is to notice that P(A) is in 1–1 correspondence with the set of
all set-theoretic functions A → {0, 1}. Given a subset B let χ B be the characteristic function of
B that is 1 on B and 0 on A − B. Conversely, given a function j of the type described, if we set
B = j −1 [ {1} ] then j = χB .
The map A → P(A) sending a to {a} is a 1–1 mapping, and therefore |A| ≤ |P(A)|. Therefore
it is only necessary to prove that the cardinalities are unequal. The Diagonal Process argument

5
is worked out in the proof of Theorem 7.7 on pages 49–50 of Munkres in the special case where
A = N+ , and the same argument works for an arbitrary nonempty set.
Notational conventions. If one assumes the Axiom of Choice then one can show that the
class of all cardinal numbers is linearly ordered, and in fact it is well-ordered (every nonempty
subcollection has a least element). Of course, the minimum element of the class of infinite cardinals
is ℵ0 , and one proceeds similarly to define ℵ 1 to be the least infinite cardinal in the complement
of {ℵ0 }, ℵ2 to be the least infinite cardinal in the complement of { ℵ 0 , ℵ1 }, and similarly one can
define ℵn recursively for each positive integer n.
We now claim that there is a set S whose cardinality is greater than ℵ n for all n. Choose An
such that |An | = ℵn and consider S = ∪n An . By construction we must have |S| > |A k | for all k.
Therefore one can then define ℵω to be the first infinite cardinal not in { ℵ 0 , ℵ1 , ℵ2 , · · · }. Similar
considerations show that for any SET of cardinal numbers one can always find a larger one.

Cardinal arithmetic

One can perform a limited number of arithmetic operations with cardinal numbers, but it
is necessary to realize that these do not enjoy all the familiar properties of the corresponding
operations on positive integers.
Definition. If A and B are sets, then
(i) the sum |A| + |B| is equal to |A t B|,
(ii) the product |A| · |B| is equal to |A × B|,
(iii) the exponentiation |A||B| is equal to |F(B, A)|, the set of all set-theoretic functions from
B to A.
These definitions are motivated by the following basic result:
PROPOSITION 5. If A and B are finite sets with |A| and |B| elements, then (i) − (iii) are true
if sums, products and exponential powers are defined in the usual manner for nonnegative integers
If A and B are finite, the first two are true by simple counting arguments, and the third is
true because the set of functions is in 1–1 correspondence with a product of |B| copies of A.
Note that the Diagonal Process argument above shows that |P(A)| = 2 |A| .
The following simple result illustrates the difference between finite and infinite cardinals:
PROPOSITION 6. If A is finite and B is infinite, then |A| + |B| = |B|.
Sketch of proof. Let C ⊂ B be a subset in 1–1 correspondence with N + . Clearly

|A t C| = |A t N+ | = |N+ | = |C|

and one can use this to construct a 1–1 correspondence between B and A t B.
The following standard identities involving ℵ 0 were first noted by Galileo and Cantor respec-
tively.
THEOREM 7. We have ℵ0 + ℵ0 = ℵ0 and ℵ0 · ℵ0 = ℵ0 .

6
 The first of these is easily seen by splitting N + into the even and odd positive integers, while
the traditional diagonal argument for proving the latter is worked out in Exercise 2 on page 45 of
Munkres.
We now have the following standard consequences.
COROLLARY 8. If Z and Q are the integers and rationals respectively then |Z| = |Q| = ℵ 0 .
These lead directly to the following fundamental result.
THEOREM 9. 

|R| = P N+  = 2ℵ0

Sketch of proof. Note that Munkres does not prove this result explicitly although a proof is
indicated in the exercises and on page 177 the real numbers are shown to have cardinality greater
than ℵ0 .
Usually this is derived using decimal expansions of real numbers, but we shall give a proof that
does not involve

decimals (although the idea is similar). The idea is to construct 1–1 maps from R
to P N+ and vice versa and then to apply the Schröder-Bernstein Theorem.
Let D : R → P(Q) be the Dedekind cut map sending a real number r to the set of all rational
numbers less than r. Since there is always a rational number between any two distinct real numbers,


it follows that this map is 1–1. Since |Q| = ℵ 0 there is a 1–1 correspondence from P(Q) to P N+ ,


and the composite of D with this map gives the desired 1–1 map from R to P N+ .



Let P∞ N+ denote the set of all infinite subsets of N+ , and define a function from P∞ N+
to R as follows: Given an infinite subset B let χ B be its characteristic function and consider the
infinite series P X
B = χB (k) · 2−k .
k

This
P −k series always converges because its terms are nonnegative and less than 1 (and the series
k 2 converges), and different infinite subsets yield different values (look at the firstPvalue P
of k
that is in one subset but not in the other; if, say, k lies in A but not in B then we have A > B ;
verifying this is left to the P
reader as an exercise — one must use the P assumption that both A and
B are infinite). Note that C ∈ [0, 1] for all infinite subsets C since k 2−k = 1.
If A is a finite subset, consider the finite sum
X
ΘA = 2 + χA (k) · 2−k .
k

Once again it follows that different finite subsets determine different real (in fact, rational) numbers.
Furthermore, since the value associated to a finite set lies in the interval [2, 3] it is clear that a
finite set and an infinite set

cannot go to the same real number. Therefore we have constructed a
1–1 function from P N+ to R.
Since we have constructed 1–1 mappings both ways, we can use the Schröder-Bernstein Theo-
rem to complete the proof.
Natural question. The Continuum Hypothesis states that |R| = ℵ 1 ; important results of P.
Cohen show that one can construct models of set theory for which this statement is true and other
models for which it is false. A concise description of the methods and results is given in a recent

7
survey article by T. Jech (What is ... Forcing?, Notices of the American Mathematical Society 55
(2008), 692–693).
One can ask which cardinal numbers are possible values for |R|. Results on this and more
general questions of the same type can also be settled using the methods introduced by Cohen. In
particular, it turns out that |R| can be equal to ℵ n for every positive integer n but it cannot be
equal to ℵω (all these are defined above). A proof of the last assertion appears in the exercises on
page 66 of the book, Set Theory and Metric Spaces, by I. Kaplansky.
Finally, we prove the another fundamental and well known result about the cardinality of R:
PROPOSITION 10. For all positive integers n we have |R n | = |R|.
Using the Axiom of Choice one can show that |A t A| = |A| and |A n | = |A| for every infinite
set A and positive integer n, but we shall outline a direct and relatively standard argument.
Sketch of proof. There is a generalization of a familiar law of exponents for cardinal numbers

γ α+β = γα · γβ

that follows from the 1–1 correspondence

F(A t B, C) −→ F(A, C) × F(B, C)

which sends a function f to

f o iA , f o iB
(recall that iA and iB are the injections from A and B to A t B).
An inductive argument shows that it suffices to prove the result when n = 2. In this case the
proof becomes a completely formal exercise involving the exponential law described above:

|R2 | = |R| × |R| = 2ℵ0 · 2ℵ0 = 2ℵ0 +ℵ0 = 2 ℵ0

COROLLARY 11. We also have 2ℵ0 + 2ℵ0 = 2ℵ0 and ℵ0 · 2ℵ0 = 2ℵ0 .
Proof. These are consequences of the following chain of inequalities:

2 ℵ0 ≤ 2 ℵ0 + 2 ℵ0 ≤ ℵ 0 · 2 ℵ0 ≤

2 ℵ0 · 2 ℵ0 = 2 ℵ0

Remarks. 1. The following generalizations of the usual laws of exponents also hold for
cardinal numbers:
β
α
γ αβ = γα , β·γ = βα · γα
The proofs require some lengthy (but relatively elementary) digressions. A variant of the first
identity is established in Section IV.2, and the proof of the first identity is an exercise for that
section. However, aside from these instances, the generalized laws of exponents will not be used
subsequently in these notes.
2. Another natural question about cardinal arithmetic is whether 2 α = 2β implies α = β as is
the case for nonnegative integers. If the Generalized Continuum Hypothesis is true, then the answer
is yes. On the other hand, this condition is not strong enough to imply the Generalized Continuum
Hypothesis, and one can also construct models of set theory for which α < β but 2 α = 2β . More

8
generally, very strong results on the possible sequences of cardinal numbers that can be written as
2α for some α are given by results of W. B. Easton that expands upon the key ideas in Cohen’s
work. Easton’s result essentially states that a few straightforward necessary conditions on such
sequences are also sufficient. These results first appeared in the following paper by Easton: Powers
of regular cardinals, Ann. Math Logic 1 (1970), 139–178. A more recent paper by T. Jech covers
subsequent work on this problem: Singular cardinals and the PCF theory, Bull. Symbolic Logic 1
(1995), 408–424.

I.4 : The real number system

(Munkres, § 4)

The first Chapter of Rudin, Principles of Mathematical Analysis (Third Edition), contains
detailed arguments for many of the results about real numbers from the files in the course direc-
tory. Aside from the usual identities involving addition, subtraction, multiplication, division and
inequalities, the most crucial properties of the real numbers throughout the course are the following:
(1) Completeness. Every nonempty subset of R that has an upper bound has a least
upper bound (or l.u.b), and every nonempty subset of R that has a lower bound has a
greatest lower bound (or g.l.b). Alternate sup and inf terminology for these concepts is
described on page 27 of Munkres.
(2) Density of rational numbers. If we are given two real numbers a and b such that
a < b, then there is a rational number q such that a < q < b.
1
(3) If ε > 0 then there is a positive integer n such that< ε.
n
The document maximality.pdf discusses a more simply stated condition which is equivalent
to (1) and (2); namely, R is a maximal ordered field which contains the rational numbers Q as a
dense ordered subfield.
Online references for further information about the real numbers are given at the beginning of
this unit.

The Cantor Set. At various points of the course it is convenient to consider a subset of the real
numbers known as the Cantor set, which also arises in undergraduate real analysis courses; there
is a complete formal definition of this set in Exercise 27.6 on page 178 of Munkres. The Cantor set
is a subset is closed under taking limits of sequences, its length (or measure) is equal to zero, but
it has the same cardinality as R. A proof of these facts appears in Problems 37–38 on page 46 of
Royden, Real Analysis, Third Edition. The online document cantor-set.pdf contains a drawing
to illustrate the construction of the Cantor set and some additional online references.

9
 II . Metric and topological spaces

The discussion of the main topics in this course begins here.

As noted in a classic text on general topology by W. Franz, “the word ‘topology’ is derived
from the Greek word τ óπoς which means ‘place,’ ‘position,’ or ‘space’ ... it is a subdiscipline of
geometry” (see pp. 1–3 of the book by Franz; complete bibliographic information is given at the
end of these notes).

Although a detailed discussion of the history of point set topology is beyond the scope of these
notes, it is useful to mention two important points that motivated the original development of the
subject.
(1) Several important theorems about continuous functions of real valued functions of a (sin-
gle) real variable have analogs in other contexts, and the most effective way to work with
such analogs is to develop a unified approach.
(2) When one considers functions of several real variables, more thought must be given to
the sets on which functions are defined. For one variable, the emphasis is on functions
defined over some interval, which may or may not have end points. On the other hand,
for functions of two or more real variables there is an overwhelming variety of shapes to
consider (e.g., round, square, triangular, hexagonal, octagonal, with and without some or
all boundary points, with or without holes inside, U-shaped, X-shaped, Y-shaped, ... ).
Clearly there are far too many to enumerate in a relatively simple manner. Therefore,
when one wants to discuss concepts like partial differentiation, it is better to begin by
considering a reasonable class of regions or domains to work with. Mathematically,
these conditions are given by the definition of an open set.
Some of these points are discussed in more detail (accompanied by a few illustrations) in the file
pstop-motiv.pdf.
The concepts of point set topology have proven to be useful — in fact indispensable — for a
wide range of mathematical contexts where it is meaningful to talk about two objects being close to
each other in some algebraic, analytic or geometric sense. In particular, these concepts have played
a major and foundational role in the application of geometric ideas to solve analytical questions, and
the interactions between the two subjects have stimulated each one to a great extent (interactions
with algebra have also been mutually beneficial). A graduate course in point set topology should
take the important links with algebra and analysis into account, but it also seems important to
retain as much of the geometric nature of the subject as possible, particularly for a course that is the
first third of a full year sequence, and striking a decent balance from a contemporary perspective
is one goal of these notes.
Here are some further references on the history of topology and related topics:
www.math.uncc.edu/∼droyster/courses/fall99/math4181/classnotes/notes1.pdf
www-gap.dcs.st-and.ac.uk/∼history/Hist/Topics/Topology in mathematics.html
en.wikipedia.org/wiki/Topology
uob-community.ballarat.edu.au/∼smorris/topbook.pdf
History of Topology, I. M. James (ed.)

10
Full information on the book by James and other books cited in these notes appears in the bibli-
ography at the end.

II.1 : Metrics and topologies

(Munkres, §§ 12–14, 16, 20)

For both historical and logical reasons one can view the most basic aspects of point set theory
as natural generalizations of important properties of certain subsets of the real numbers. One
approach to doing this is to base the discussion on an abstract notion of DISTANCE that generalizes
the usual notion of distance between two numbers in the obvious fashion. Another approach is to
take the concept of an OPEN SUBSET (or something logically equivalent) as the fundamental
abstract structure. It is not difficult to formulate the concept of an open set if one has a notion
of distance, so there is a natural progression of abstraction from the real numbers to metric spaces
(sets with a suitable notion of distance) to topological spaces (sets with a suitable notion of open
subsets).
Eventually a course in point set topology needs to cover both types of structures, but there is
no universal agreement on which should come first and when the other should be introduced. The
approach in these notes will be to introduce metric spaces first and topological spaces immediately
afterwards. This will allow us to take advantage of the strengths of both approaches throughout
the course.

The basic definitions and a few examples

The notion of distance between two points in Rn is fundamentally important in multivariable

calculus and some aspects of linear algebra. It turns out that an extremely short list of properties for
distances are enough to prove abstract versions of many important results from advanced calculus
and real variables courses.
Definition. A metric space is a pair (X, d) consisting of a set X and a function d : X × X → R
(sometimes called the metric or distance function, with d(x, y) being called the distance from
x to y, or between x and y) such that the following properties hold:
(MS1) d(x, y) ≥ 0 for all x, y ∈ X.
(MS2) d(x, y) = 0 if and only if x = y.
(MS3) d(x, y) = d(y, x) ≥ 0 for all x, y ∈ X.
(MS4) d(x, z) ≤ d(x, y) + d(y, z) for all x, y, z ∈ X.
The last property is often called the triangle inequality because it generalizes the usual triangle
inequality from classical Euclidean geometry.

EXAMPLES. 1. The most important examples are the ordinary coordinate or Euclidean
spaces Rn for which d(x, y) = |x−y|. The four basic properties for an abstract metric are established
in undergraduate courses which discuss linear or vector algebra.

11
 2. If (X, d) is a metric space and A is a subset of X, then one can make A into a metric
space using the subspace metric given by the restriction d (A × A); less formally, this means that
the distances between points of A are the same as their distances in X itself.
3. If S is an arbitrary set, then one can make S into a metric space with the so-called
discrete metric, for which d(s, t) = 1 if s 6= t and 0 if s = t. It is a routine exercise to verify that
this defines a metric on x.
4. One can form an abstract generalization of the first example a follows: If one defines a
norm on a real vector space V to be a function sending each v ∈ V to a nonnegative real number
|v| such that
(a) |v| = 0 if and only if v = 0,
(b) |cv| = |c||v| for all c ∈ R and v ∈ V ,
(c) |v + w| ≤ |v| + |w| for all v, w ∈ V ,
then the formula d(v, w) = |v − w| defines a metric on V . It is again an elementary exercise to
verify that this satisfies the four conditions required for a metric.
5. We need to give some additional examples in order to show that the preceding construc-
tion yields something beyond ordinary Euclidean spaces.
(5A) Actually, this is two examples. Take V = R n , write P a typical vector x in coordinates as
(x1 , · · · , xn ), and consider the functions |x|1 = i |xi | and |x|∞ = maxi { |xi | }. It is
again elementary to check that each of these define norms. If one wants to distinguish
the previous norm from these it is customary to write the latter as |x| 2 , which reflects
the quadratic nature of the latter (as in measure theory, one can interpolate a continuous
series of norms |x|p for 1 ≤ p ≤ ∞, but we shall not need these examples here).
(5B) This is a much larger example. Let S be a set, and let C(S) be the space of all bounded
real valued functions on S. Then a norm is defined by the formula |f | = sup x∈S { |f (x)| }.
It is also possible to construct a vast array of other norms on vector spaces at this point, but we
shall not do so in order to avoid straying too far from the central themes of the course.
The examples above show that one can construct many different metrics on a given set. How-
ever, it clearly becomes very cumbersome to write (X, d) every time we are referring to a metric
space, so in order to simplify the exposition we shall often simply write X if the metric is clear
from the context.

Open sets

The basic definition extends the one for Euclidean spaces.

Definition. Let (X, d) be a metric space. A subset U ⊂ X is said to be open if for each x ∈ U
there is a positive real number ε (depending upon x) such that d(x, y) < ε =⇒ y ∈ U .
For each r > 0 and x ∈ X, the set

Nr (y) = { y ∈ X | d(x, y) < r }

is called the open ball (or disk or neighborhood) of radius r centered at x. One can rewrite the
definition of open set to say that for all x ∈ U there exists an ε > 0 such that N ε (x) ⊂ U .

12
 The most important properties of open sets in metric spaces are summarized in the following
result:

THEOREM 1. Let U be the family of all open subsets of X. The the following hold:
(i) Each subset of the form Nε (x) is open in X.
(ii) The empty set and X itself are both open in X.
(iii) If for each α ∈ A the set Uα is open in X then ∪α Uα is also open in X.
(iv) If U1 and U2 are open in X then U1 ∩ U2 is also open in X.
One can combine (iv) and finite induction to prove that the intersection of any finite collection
of open subsets of X is also open in X.
Proof. (i) Let y ∈ Nε (x), and let s = d(x, y) so that ε − s > 0. We claim that N ε−s (y) ⊂ Nε (x).
It may be worthwhile to draw a two-dimensional picture at this point in order to make this assertion
plausible; the formal proof of the assertion proceeds as follows. Suppose that z ∈ N ε−s (y), so that
d(y, z) < ε − s. We need to prove that d(x, z) < ε. Applying the triangle inequality we have

d(x, z) ≤ d(x, y) + d(y, z) = s + d(y, z) < s + (ε − s) < ε

as required.
(ii) We shall first consider the case of the empty set. It has no points so the condition on all
points in it will automatically be true because it is a statement about nothing (one often says such
statements are vacuously true). The openness of X follows because N ε (x) = X for all x and ε.
(iii) Suppose that x ∈ ∪α Uα , and choose β ∈ A so that x ∈ Uβ . Then one can find ε > 0 so
that Nε (x) ⊂ Uβ , and since Uβ ⊂ ∪α Uα it also follows that Nε (x) ⊂ ∪α Uα .
(iv) Let i = 1 or 2, and let x ∈ U1 ∩ U2 . Then one has εi > 0 so that Nεi (x) ⊂ Ui for i = 1, 2.
If we take ε to be the smaller of ε1 and ε2 then Nε (x) ⊂ U1 ∩ U2 . Once again, it might be helpful
to draw a picture as an aid to understanding this proof.
It is useful to look at the meaning of open subset for one of the examples described above;
namely, the discrete metric on a set. In this case N 1 (x) is merely the one point set {x}. Thus every
one point subset of S is open with respect to the discrete metric. But if W ⊂ S, then clearly we
have [
W = {x}
x∈W

so by the preceding theorem we see that every subset of a metric space with a discrete metric is an
open subset. Of course, for examples like Euclidean spaces there are many examples of subsets
that are not open. In particular, one point subsets are NEVER open in Euclidean spaces (unless
one adopts the convention R0 = {0}, in which case this object must be excluded).

Topological spaces

The preceding theorem provides the motivation for the central concept of a course in point set
topology:
Definition. A topological space is a pair (X, T) consisting of a set X and a collection T of subsets
of X satisfying the following conditions:

13
 (TS1) The empty set and X itself both belong to T.
(TS2) If for each α ∈ A the set Uα ⊂ X belongs to T, then ∪α Uα also belongs to T.
(TS3) If U1 and U2 belong to T, then U1 ∩ U2 also belongs to T.
We often say that T is a topology on X or that T is the associated family of open subsets of
X, and we say that U ⊂ X is open if U ∈ T. As before, if the topology on a set is clear from the
context we shall often use the set by itself to denote a topological space.
Note. One can combine (TS3) and finite induction to prove that the intersection of every
finite collection of subsets in T is also open in T.
If (X, d) is a metric space and T denotes the family of open subsets of X, then by the preceding
theorem (X, T) is automatically a topological space. We often call this the metric topology
(associated to d).
In particular, if S is an arbitrary set and we put the discrete metric on S, then we have seen
that all subsets of S are open in the corresponding metric topology. More generally, a topological
space is said to be discrete if the every subset is open (equivalently, every one point subset is open);
by previous observations, this is just the metric topology associated to the discrete metric.
On the other hand, there are many examples of topological spaces that do not come from
metric spaces.

EXAMPLES. 1. Given a set X, the indiscrete topology on X is the family T consisting only of
the empty set and X itself. It is elementary to verify that this defines a topology on X. However, if
X contains at least two points then this cannot come from a metric space because if X is a metric
space and p ∈ X then X − {p} is open for all p ∈ X (we shall prove this below).
2. Given a set X, the finitary topology or cofinite topology on X is the family T consisting
of the empty set and all subsets of the form X − A where A is finite. The verification that T is
a topology can be found in Example 3 on page 77 of Munkres. If X is finite this is equal to the
metric topology for the discrete metric. On the other hand, if X is infinite, then if u and v are
distinct points of X and U and V , then U ∩ V is always infinite (its complement X − U ∪ X − V
is finite!), and by the Hausdorff separation property below it follows that T does not come from a
metric on X if the latter is infinite.

Here are the results that we need to show that these examples do not come from metrics:
PROPOSITION 2. (T1 Separation Property) If X is a metric space and p ∈ X, then X − {p}
is open.
Proof. Let q ∈ X − {p}, so that q 6= p and r = d(q, p) > 0. Then clearly N r (q) ⊂ X − {p}, and
hence the latter is open.
PROPOSITION 3. (T2 or Hausdorff Separation Property) If X is a metric space and u, v ∈ X
are distinct points, then there exist disjoint open subsets U and V containing u and v respectively.
Proof. Let 2ε = d(x, y) > 0, and take U and V to be N ε (u) and Nε (v) respectively. To see
that these are disjoint, suppose that they do have some point z in common. Then by the Triangle
Inequality and z ∈ Nε (u) ∩ Nε (v) we have

2ε = d(u, v) ≤ d(u, z) + d(z, v) < ε + ε

14
which is a contradiction. Therefore the intersection must be empty. Once again, it might be helpful
to draw a picture as an aid to understanding this proof.
The following result is often useful in working with open sets:
LEMMA 4. If U is an open subset of a metric space, then one can find numbers ε(y) > 0 for all
y ∈ U such that U = ∪y Nε(y) (y).
Proof. If ε(y) > 0 such that Nε(y) (y) ⊂ U , then we have the chain of inclusions
[ [
U = {y} ⊂ Nε(y) (y) ⊂ U
y y

shows that U = ∪y Nε(y) (y).

Comparing and constructing topologies on a set

We have seen that a given set may have several different metrics and several different topologies.
Some additional examples are mentioned in the following document:
http://math.ucr.edu/∼res/math145A-2014/intro2topA-07.pdf
There are also a few general aspects of the set of all topologies on a space that are worth examining
in some more detail at this time.
Since topologies on a set are just families of subspaces, it is meaningful to ask if one is contained
in the other. Every topology must contain the indiscrete topology, and since the topology of a
discrete metric contains every subset it is clear that every other topology is contained in this one.
Frequently one sees statements that one topology T on a space is stronger or weaker than another
topology S if one contains the other, and the terms coarser or finer are also used in such contexts.
Unfortunately, this notation can be hopelessly confusing because, say, there is no consistency about
whether a stronger topology has more open sets than a weaker one or vice versa. We shall avoid
this by simply saying that one topology is larger or smaller than the other (compare the first few
lines on page 78 of Munkres).
The following observation is elementary to verify:
FACT 5. An intersection of a nonempty family of topologies on a given set X is also a topology
on X.
In contrast, the union of two topologies on X is not necessarily a topology on X (see Exercise
II.1.6 for a specific example).
Another abstract feature of topologies is that given a family A of subsets of X, there is a
unique minimal topology T(A) on X that contains A.
In view of Fact 5, the topology T(A) can be described as the intersection of all topologies that
contain A; this family is nonempty because the discrete topology is a specific example of a topology
containing A. However, for many purposes it is necessary to have a more explicit description of this
topology. Define A∗ to be the set of arbitrary unions of sets having the form A 1 ∩ · · · ∩ Ak for some
finite family of subsets {A1 , · · · , Ak } in A together with X and the empty set. To show that A ∗ is
a topology it is necessary to verify that it is closed under arbitrary unions and finite intersections.
It will be convenient to let B denote the set of finite intersections of sets in A; then every element

15
of A∗ except possibly X and the empty set will be a union of subsets in B. Since a union of unions
of subsets in B is again a union of subsets in B, it follows that A ∗ is closed under taking arbitrary
unions. Suppose now that U and V lie in A ∗ . Write these sets as ∪β Uβ and ∪γ Vγ respectively;
then [
U ∩V = Uβ ∩ V γ
β,γ

where each summand Uβ ∩ Vγ is a finite intersection of subsets in A, and this implies that U ∩V ∈ A ∗
as required.
Definition. A family A of subsets of X is called a subbase for the topology T on X if A ∗ = T.

Basic open subsets for a topology

There is a special type of subbase known as a base (or basis sometimes) that is often useful
for constructing and studying topologies.
Definition. A family B of subsets of X is called a base for the topology T on X provided
(B0) B ∗ = T,
(B1) each x ∈ X belongs to at least one B ∈ B,
(B2) if x ∈ B1 ∩ B2 where B1 , B2 ∈ B, then there is a B3 ∈ B such that x ∈ B3 and
B3 ⊂ B 1 ∩ B 2 .
If B is a base for T we often refer to the sets in B as basic open subsets.
The following result is elementary to prove:
PROPOSITION 6. If B is a family of subsets of X satisfying (B1) and (B2) above, then the
smallest topology containing B is the set of all unions of sets in B together with the empty set.
Lemma 13.2 on page 80 of Munkres is another useful result on bases for topologies.
Example — The Order Topology. If (X, ≤) is a linearly ordered set (so that X is the
set and ≤ is the linear ordering relation), then one can construct a topology that is analogous to
the standard topology on R which is generated by unbounded open intervals of the form

(u, +∞) = { x ∈ X | u < x } (−∞, v) = { x ∈ X | x < v }

where u, v ∈ X. It follows immediately that if u < v in X then the bounded “open interval”

(u, v) = { x ∈ X | u < x < v }

is also open with respect to this topology.

PROPOSITION 7. If X is the real numbers with the usual linear ordering, then the order
topology is equal to the metric topology.
Proof. The order topology contains the metric topology because the sets N δ (x) are intersections
(−∞, x + δ) ∩ (x − δ, +∞), and the metric topology contains the order topology because we have
[ [
(u, +∞) = Nt−u (t) , (−∞, v) = Nv−t (t)
t>u t<v

16
for all u, v ∈ X.

Subspace topologies

We have already noted that a subset of a metric space can be viewed in a natural way as a
metric space in its own right by restricting the metric. There is a parallel way of viewing a subset
of a topological space as a topological space in its own right, and it turns out that if X is a metric
space, the topologies that one obtains on A in both fashions are identical.
Definition. If (X, T) is a topological space and A is a subset of X, then the subspace topology
on A is the family T|A of all intersections U ∩ A where U ∈ T. — It is an elementary set-theoretic
exercise to verify that this defines a topology on A.
The following result relates the metric and subspace topologies on a subset of a metric space.
PROPOSITION 8. If X is a metric space and A ⊂ X, then the metric topology on A is
identical to the subspace topology on A.
Proof. It will be convenient to distinguish the open disks in A and X by N A and N X respectively.
By construction we have that N A = A ∩ N X .
Let T|A denote the subspace topology and let M A denote the metric topology. Every set in
T|A has the form U ∩ A where U ∈ T, where T denotes the metric topology on X. By definition
of the metric topology on X, for every point y ∈ A ∩ U there is an r > 0 such that N rX (y) ⊂ U ,
and therefore we have
NrA (y) = A ∩ NrX (y) ⊂ A ∩ U
which shows that every set in T|A belongs to M A . Conversely, every open set W in the metric
A
topology is a union of open disks having the form N r(y) (y) for y ∈ W and suitably chosen r(y) > 0
(use the lemma stated above), and therefore we have
!
[ [   [
A X A
W = Nr(y) (y) = A ∩ Nr(y) (y) = A∩ Nr(y) (y)
y y y

which shows that W is an intersection of A with an open subset of X.

Neighborhoods

In a metric space the sets Nδ (x) are often called δ-neighborhoods; for general topological spaces
one also uses the term neighborhood of a point (say) p to denote an open set containing p (compare
the definition on page 96 of Munkres); however, sometimes the term “neighborhood” has a more
general meaning of a set N such that N contains some open subset U which in turn contains the
point p. One should be aware of this possible difference when reading other books or papers.
The following online site contains further discussions and some useful illustrations:
http://en.wikipedia.org/wiki/Topological neighbourhood

17
 II.2 : Closed sets and limit points

(Munkres, § 17)

The usual discussion of limits for sequences for real numbers extends directly to metric spaces.
Given a metric space X and a sequence {a n } in X, we say that limn→∞ an = a if for all ε > 0 there
is a positive integer M such that n > M implies d(a n , a) < ε. As in real variables, a sequence has
at most one limit, and the proof in the general case is essentially the same.
Closed subsets of the real line are precisely the subsets that are “closed under taking limits of
convergent sequences,” so it is clear that one should be able to discuss closed subsets of an arbitrary
metric space. At first glance it is less obvious that one can also discuss closed sets for arbitrary
topological spaces, but this is indeed the case, and one major objective here to justify this.

Limit points and limits of sequences

We begin with the following definition:

Definition. Let X be a topological space, and let A ⊂ X be a subset of X. A point y ∈ X
is called a limit point of A if for all open sets U containing y the intersection A ∩ (U − {y}) is
nonempty. The set of all limit points of A in X is written L(A; X), and when the ambient space
X is clear from the context we shall often write L(A).
We shall begin with an elementary but useful fact.
PROPOSITION 1. If A is a subset of B in a topological space X, then L(A) is a subset of
L(B).
Proof. If y ∈ L(A), then for every open set U containing y we have (U − {y}) ∩ A is nonempty.
On the other hand if A ⊂ B, then

(U − {y}) ∩ A ⊂ (U − {y}) ∩ B

and since the first of these is nonempty so is the second. Therefore we must also have y ∈ L(B).
The motivation for the name limit point is implicit in the following result:
PROPOSITION 2. The following are equivalent for a metric space X, a point y ∈ X and a
subset A ⊂ X:
(i) y ∈ L(A).
(ii) There is a sequence of points {a n } in A such that an 6= y for all n but limn→∞ an = y.


Proof. (i) =⇒ (ii) Let n be a positive integer, and consider the open set N 1/n (y). By the
definition of L(A) there is a point an ∈ A such that an 6= y but an ∈ N1/n (y). Then

1
d(an , y) <
n

for all n and therefore limn→∞ an = y.

18


(ii) =⇒ (i) Let the sequence {an } be given as in the statement of (ii), let U be an open
subset containing y, let ε > 0 be such that N ε (y) ⊂ U , and choose M such that n > M implies
d(an , a) < ε. Then we have

aM +1 ∈ A ∩ (Nε (y) − {y}) ⊂ A ∩ (U − {y})

so that set on the right hand side is nonempty, and therefore we have y ∈ L(A).
The nest result is the key to defining closed subsets in arbitrary topological spaces.
THEOREM 3. If A is a subset of a topological space X, then X − A is open if and only if
L(A) ⊂ A.
Proof. (=⇒) Suppose that X −A is open and the set L(A) is not contained in A. Let y ∈ L(A)−A;
clearly y ∈ X − A. By the definition of the set of limit points, it follows that there is a point
x ∈ X − A that is also in A, which is a contradiction. Thus L(A) ⊂ A if X − A is open.
(⇐=) Suppose that L(A) ⊂ A and let y ∈ X − A. By hypothesis y is not a limit point of A and
therefore there is some open set Uy containing y such that Uy − {y} and A are disjoint (Note: You
should check that the conclusion is the negation of the condition in the definition of limit point!).
Since we also know that y 6∈ A it follows that U y and A are also disjoint, so that Uy ⊂ X − A.
Therefore we have the string of inclusions
[ [
X −A = {y} ⊂ Uy ⊂ X − A
y6∈A y6∈A

which shows that X − A = ∪y Uy ; since the right hand side is a union of open sets, it follows that
the sets on both sides of the equation are open.
This leads us to a purely topological definition of closed set that is compatible with the notion
of “closure under taking limits of convergent sequences” for metric spaces.
Definition. A subset F of a topological space X is closed if and only if its relative complement
X − A is open.
Note. In contrast with the usual usage for the terms “open” and “closed,” a subset of a
topological space may be open but not closed, closed but not open, neither open nor closed, or
both open and closed. Over the real line these are illustrated by the subsets (0, 1), [0, 1], [0, 1) and
R itself (you should verify this for each example).
Closed subsets have the following properties that correspond to the fundamental properties of
open subsets.
PROPOSITION 4. The family of closed subsets of a topological space X has the following
properties:
(i) The empty set and X itself are both closed in X.
(ii) If for each α ∈ A the set Fα is closed in X then ∩α Fα is also closed in X.
(iii) If F1 and F2 are closed in X then F1 ∪ F2 is also closed in X.
Note. One can combine (iii) and finite induction to prove that the union of any finite
collection of closed subsets in X is also closed in X.
Proof. (i) The empty set and X are complements of each other, so since each is open their
complements — which are merely the empty set and X itself — are closed.

19
 (ii) This follows immediately from the complementation formula
\ [
X− Fα = (X − Fα )
α α

and the fact that unions of open subsets are open.

(iii) This follows immediately from the complementation formula

X − (F1 ∪ F2 ) = (X − F1 ) ∩ (X − F2 )

and the fact that the intersection of two open subsets is open.
Clearly one could define mathematical systems equivalent to topological spaces by specifying
families of closed subsets satisfying the three properties in the preceding proposition. In fact, there
are also many other equivalent ways of describing topological spaces, but we shall not say very
much about them here.
For metric spaces one has the following important fact regarding closed subsets.
PROPOSITION 5. (Equivalent form of the T1 separation property) If X is a metric space and
x ∈ X, then the one point set {x} is closed in X.
This follows immediately from an earlier observation that X −{x} is open in X if X is a metric
space. As noted previously, the indiscrete topology on a set with at least two elements does not
have the corresponding property.

Closures and interiors of subsets

In many mathematical contexts it is useful and enlightening to have constructions that “fill
the gaps (or empty spaces)” in a mathematical object. For example, the real numbers are a way
of filling the gaps in the rational numbers. Given a subset of, say, the real line, one often wants to
expand this set so that it contains all limits of sequences that are defined on that set. This is done
by considering the closure of the set, and the concept can be formulated in a manner that applies
to all topological spaces.
Definition. Given a topological space X and a subset A ⊂ X, the closure of A is the set
A = A ∪ L(A).
The terminology suggests that A should be the smallest closed subset of X that contains A.
Verifying this will take a little work.
PROPOSITION 6. The set A is the intersection of all closed subsets containing A, and
consequently it is the smallest closed subset containing A.
Proof. Let F be the intersection described in the statement of the proposition. We need to prove
the two inclusions A ⊂ F and F ⊂ A.


Proof that A ⊂ F : Since A ⊂ F it is only necessary to show that L(A) ⊂ F . It follows from
an earlier result that A ⊂ F implies L(A) ⊂ L(F ). But since F is closed we know that L(F ) ⊂ F ,
and therefore we have that L(A) ⊂ F as required.


Proof that F ⊂ A : This will follow if we can show that A is closed in X, or equivalently
that X − A is open in X. So suppose that y ∈ X − A. By definition this means that y 6∈ A and

20
y 6∈ L(A). The latter in turn means that there is an open set U y ⊂ X such that A ∩ (Uy − {y})
is empty. But we also know that y 6∈ A,so we can strengthen the latter to say that A ∩ U y is
empty. This immediately implies that U y lies in the complement of A, but it ALSO implies that
Uy lies in the complement of L(A), for if z ∈ U y then Uy is an open subset containing z such that
(Uy − {z}) ∩ A is empty. Thus we have Uy ⊂ X − A for all y, yielding the usual chain of inclusions
[ [
X −A = {y} ⊂ Uy ⊂ X −A
y6∈A y6∈A

which shows that X − A = ∪y Uy and therefore is open; but this means that A is closed.
There is a complementary concept of the interior of a set A, which is the largest open subset
U contained in A. Formally, one can define the interior by the formula

Int (A) = X − X − A

and the proof that this is the union of all open subsets contained in A reduces to an exercise in set
theory.
For the sake of completeness, here are the details: One can rewrite the defining equation
as X − Int (A) = X − A and since the latter contains X − A, by taking complements we
have that Int(A) is an open set that is contained in A. Suppose now that U is any open
subset contained in A. Then X − U is a closed set that contains X − A and thus also
X − A = X − Int (A); taking complements once again we see that U is contained in Int(A).
Warning. In some topological contexts the term “interior” has an entirely different meaning,
but in this course the term will always have the meaning given above.
The following result provides an extremely useful relation between the notions of closure and
passage to subspaces.
PROPOSITION 7. Given a topological space X and subspaces A, Y such that A ⊂ Y ⊂ X, let
Closure Y (A) denote the closure of A with respect to the subspace topology on Y . Then

Closure Y (A) = A ∩ Y .

Proof. Once again we have to prove the inclusions in both directions.

Proof that Closure Y (A) ⊂ A ∩ Y : Note first that the close subsets of Y have the form
Y ∩ F where F is closed in X (Proof: E is closed in Y ⇔ Y − E is open in Y ⇔ Y − E = X ∩ U
for some U open in X ⇔ E = Y − Y ∩ U for some U open in X ⇔ E = Y ∩ (X − U ) for some U
open in X ⇔ E = Y ∩ F for some F closed in X). — It follows that the right hand side is a closed
subset of Y and therefore contains the set Closure Y (A).


Proof that A ∩Y ⊂ Closure Y (A) : The right hand side is a closed subset of Y , and therefore
by the preceding paragraph it has the form B ∩ Y where B is closed in X. By construction B ⊃ A,
so B must also contain A. But this means that

A∩Y ⊂ B∩Y = Closure Y (A)

which yields the desired inclusion.

21
 Convergence in general topological spaces

In general one cannot work with limits of sequences in abstract topological spaces as easily
and effectively as one can work with them in metric spaces. The crucial property of metric spaces
that allows one to work with sequences is the following:
PROPOSITION 8. (First Countability Property) If X is a metric space and x ∈ X then there
is a sequence of decreasing open subsets U k such that every open subset contains some U k .
1
In fact, we can take Uk to be the open disk of radius k centered at x (i.e., the set N1/k (x)).
There is a somewhat more complicated concept of net that serve a similar purpose to sequences
for arbitrary topological spaces. Nets for topological spaces are not as important or useful as
sequences for metric spaces, but there are some situations, particularly in analysis, where it is
convenient to have them. A concise but readable introduction to nets appears on pages 187–188 of
Munkres.

II.3 : Continuous functions

(Munkres, §§ 18, 21)

The standard definitions for continuous and uniformly continuous functions generalize imme-
diately to metric spaces.
Definition. Let (X, dX ) and (Y, dY ) be metric spaces, and let a ∈ X. A set-theoretic function
f : X → Y is said to be continuous at a if for each ε > 0 there is a δ = δ(ε) > 0 such that
dX (x, a) < δ implies dY (f (x), f (a)) < ε. The function f is said to be continuous (on all of X) if
it is continuous at every point of X.
As in the case of functions of a real variable, the numbers δ(ε) depend upon the point a.
Definition. Let (X, dX ) and (Y, dY ) be metric spaces, and let a ∈ X. A set-theoretic function
f : X → Y is said to be uniformly continuous if for each ε > 0 there is a δ = δ(ε) > 0 such that for
all u, v ∈ X, we have that dX (u, v) < δ implies dY (f (u), f (v)) < ε.
As in real variables, the difference between continuity and uniform continuity is that δ depends
upon ε and a for continuity but it depends only upon ε for uniform continuity. The results of Section
III.1 show that every continuous real valued function on a closed interval is uniformly continuous,
but the function f (x) = 1/x from the positive real numbers to themselves is a function which is
continuous but not uniformly continuous.
The following characterization of continuity yields a definition that is meaningful for functions
on arbitrary topological spaces:
THEOREM 1. Let (X, dX ) and (Y, dY ) be metric spaces, and let f : X → Y be a function.
Then f is continuous if and only if for each open set V ⊂ Y , the inverse image f −1 [V ] is open in
X.
Proof. (=⇒) Choose y ∈ Y and x ∈ X so that y = f (x). There is an ε > 0 so that N ε (y) ⊂ V ,
and by continuity there is a δ > 0 such that f maps N δ (x) into Nε (y). It follows as in many
previous arguments that [
f −1 [V ] = Nδ (x)
x

22
(check this out!) and therefore the left hand side is an open subset of X.
(⇐=) Choose y ∈ Y and x ∈ X so that y = f (x), and let ε > 0 be given. By the hypothesis
we know that the set
W = f −1 [Nε (y)]
is an open subset of X containing x. If we choose δ > 0 so that N δ (x) ⊂ W , then it follows that f
maps Nδ (x) into Nε (y).
In view of the above, if (X, TX ) and (Y, TY ) are topological spaces we may DEFINE a set-
theoretic map f : X → Y to be continuous if and only if for each open set V ⊂ Y , the inverse
image f −1 [V ] is open in X.
Several equivalent formulations of continuity are established in Theorem 18.1 on pages 104–105
of Munkres and Lemma 21.3 on page 130 of Munkres. Here is an overlapping list of equivalences:
THEOREM 2. (Characterizations of continuity) Let X and Y be topological spaces and let
f : X → Y be a set-theoretic map. Then the following are equivalent:
(1) The mapping f is continuous.
(2) For every closed subset F ⊂ Y the inverse image f −1 [F ] is closed.
 
(3) For all A ⊂ X we have f A ⊂ f [A].
 
(4) For all B ⊂ Y we have f −1 [B] ⊂ f −1 B .
(5) For all A ⊂ X we have Int (f [A]) ⊂ f [Int(A)].

(6) For all B ⊂ Y we have f −1 [Int(B)] ⊂ Int f −1 [B] .
If X and Y are metric spaces then the following is also equivalent to the preceding conditions:
(7) For all sequences { xn } in x such that limn→∞ xn = a we have limn→∞ f (xn ) = f (a).

The statements and proofs of the results in Munkres should be read and understood. Verifi-
cation of the statements not proven in Munkres is left to the reader as an exercise.
It is not possible to discuss uniform continuity in a topological space unless some extra structure
is added; one reference for an abstract treatment of such uniform structures (or uniformities) is
Kelley, General Topology, and a the book by Isbell is a more specialized reference. Topological
spaces with uniform structures are often known as uniform spaces. A class of spaces known as
topological groups have particularly important examples of the uniform structures that exist on
uniform spaces. An introduction to the theory of topological groups appears in Appendix A of
these notes.

Examples. 1. A real variables textbook (and even a calculus or precalculus textbook) contains
many examples of continuous functions from subsets of the real numbers to the real numbers.
2. If A is a subset of a topological space with the subspace topology, then the inclusion
map i : A → X is continuous because i−1 [U ] = U ∩ A for all open subsets U . In fact, if X is a
metric space and A has the subspace metric, then the inclusion map is uniformly continuous; for
each ε > 0 we can take δ = ε.
3. An important special case of the preceding example occurs when A = X, and in this
case the inclusion is the identity map on X.

23
 4. Let X be an arbitrary metric space, and let A be a nonempty subset of X. For each
point x ∈ X the distance from x to A is defined by the formula

d(x, A) = g.l.b.a∈A d(x, a)

where the greatest lower bound exists and is nonnegative because all distances are nonnegative.
We claim that the function d(−, A) is uniformly continuous. — Here is a Proof: By the Triangle
Inequality we have that d(x, a) ≤ d(x, y) + d(y, a) for all x, y ∈ X and a ∈ A. Therefore it
follows that d(x, A) ≤ d(x, y) + d(y, a). Subtract d(x, y) from each side. This yields the inequality
d(x, A) − d(x, y) ≤ d(y, a), which in turn implies that the left hand side is ≤ d(y, A). We can
now rewrite this in the form d(x, A) − d(y, A) ≤ d(x, y). If we reverse the roles of x and y in this
argument we get the complementary inequality d(y, A) − d(x, A) ≤ d(x, y). Combining these, we
obtain the inequality  
d(y, A) − d(x, A) ≤ d(x, y)

which shows that the function in question is in fact uniformly continuous because for each ε > 0
we can take δ = ε.
5. We shall end this list of examples with one that is very simple but quite important.
Suppose that X and Y are any topological spaces and that y ∈ Y . Then there is a constant map
Cy : X → Y which sends every point of X to y, and for all X and y the map C y is continuous. To
see this, let V ⊂ Y be open, and consider f −1 (V ). If y ∈ V then the inverse image is all of X but
if y 6∈ V then the inverse image is the empty set. In either case the inverse image is open.
In analysis there are theorems stating that sums, products and composites of continuous func-
tions are continuous. Metric and topological spaces usually do not have the algebraic structure
needed to construct sums and products. However, one does have the following version of continuity
for composite functions.
PROPOSITION 3. If X, Y, Z are topological spaces and f : X → Y and g : Y → Z are
continuous, then so is the composite g o f : X → Z.
Proof. Suppose that W is open in Z; then by continuity it follows that V = g −1 [W ] is open in
Y and U = f −1 [V ] is open in X. However, we also have
 
u = f −1 g −1 (W ) = (g o f )−1 [W ]

and therefore it follows that g o f is also continuous.

COROLLARY 4. If f : X → Y is continuous and A ⊂ X is equipped with the subspace
topology, then the restriction F |A : A → Y is continuous.
This is true because the restriction is the composite of f and the inclusion map for A ⊂ X;
we have already noted that the latter is continuous.
In addition to the preceding way of constructing continuous functions by restricting the domain,
it is also possible to construct new continuous functions by shrinking the codomain if the image of
the function is a proper subset.
PROPOSITION 5. Let f : X → Y be a continuous function, let B ⊂ Y be equipped with the
subspace topology, let j : B → Y denote the inclusion map, and suppose that f [X] ⊂ B. Then
there is a unique continuous map g : X → B such that j o g = f .
Proof. On the set-theoretic level one simply defines g by the rule g(x) = f (x). We need to verify
that this map is continuous.

24
 Suppose that V is open in B. Then B = W ∩ B where W is open in Y . Given an arbitrary
subset A ⊂ Y , elementary set-theoretic considerations imply that

f −1 [A] = f −1 [A ∩ B] = g −1 [A ∩ B]

with the first equation holding because f (X) ⊂ B and the second holding because f (x) = g(x) for
all x. Therefore if V is open in B and V = W ∩ B (where W is open in Y ), then

g −1 [V ] = f −1 [W ∩ B] = f −1 [W ] .

Since f is continuous the set on the right hand side of the equation is open in X; therefore the set
on the left hand side is also open and the map g is continuous.

Homeomorphisms and other special mappings

We begin with a natural question:

Continuity of inverses. Suppose that f : X → Y is a continuous map of topological spaces
that is a 1 − 1 correspondence. Is the inverse map f −1 also continuous?
There are many examples to show that the answer to the question is negative. One purely
formal approach is to take a X = Y with f = id X and the topologies on the domain and codomain
equal to the discrete and indiscrete topologies respectively. Then f is continuous (every map into
a space with the indiscrete topology is continuous!). What can we say about the continuity of
the inverse? By construction the inverse is just the identity map from a space with the indiscrete
topology to a space with the discrete topology. If X has more than one element and A is a
nonempty proper subset, then A is open in the discrete topology but not in the indiscrete topology,
and therefore the inverse map is not continuous.
Here is a more tangible example. Let S 1 be the unit circle in the cartesian plane, and let
f : [0, 1) → S 1 send t to (cos 2πt, sin 2πt). Then f is clearly continuous and 1–1 onto (it might
be helpful to draw a picture of this). However, f −1 is not continuous at the point (1, 0) ∈ R 2 .
Specifically, the
set [0, 12 ) is open in [0, 1), but its inverse image in the circle under f −1 — which is
simply f [0, 21 ) (why?) — is not open in the circle. To see this, note that every open subset of
the circle containing (1, 0) must contain some points whose second coordinates are negative.
We are thus led to the following:
Fundamental Definition. A continuous 1–1 onto map f : X → Y of topological spaces is a
homeomorphism if f −1 is also continuous.
It follows immediately from the definition that
for every topological space X the identity map id X is continuous (it is understood that
X has the same topology whether it is viewed as the domain or the codomain),
the inverse of a homeomorphism is a homeomorphism, and
the composite of two (composable) homeomorphisms is a homeomorphism.
Here is an alternate characterization of homeomorphisms that may be enlightening:
PROPOSITION 6. Let (X, TX ) and (Y, TY ) be topological spaces, and let f : X → Y be a
set-theoretic map that is 1 − 1 and onto. Then f is a homeomorphism if and only if for each subset
A ⊂ X, we have that A is open in X if and only if f [A] is open in Y .

25
Proof. (=⇒) Let ⊂ X. If f [A] is open in Y , then by continuity of f we have that
 
A = f −1 f [A]

is open in X. Similarly, if A is open in X, then by continuity of f −1 we have that

−1
f [A] = f −1 [A]

is open in Y .
(⇐=) The hypotheses on f and the first set-theoretic identity in the previous paragraph
imply that f is continuous, and the hypotheses together with the second set-theoretic identity in
the previous paragraph imply that f −1 is continuous.
One can state and prove a similar theorem in which “open” is replaced by “closed”:
PROPOSITION 7. Let (X, TX ) and (Y, TY ) be topological spaces, and let f : X → Y be a
set-theoretic map that is 1 − 1 and onto. Then f is a homeomorphism if and only if for each subset
A ⊂ X, we have that A is closed in X if and only if f [A] is closed in Y .
Proof. The argument is the same as the preceding one with “open” replacing “closed”.
The preceding results lead to some other important classes of mappings on topological spaces.
Definitions. A set-theoretic map f : X → Y of topological spaces is open if for each open set
U ⊂ X, the image f [U ] is open in Y . Similarly, a set-theoretic map f : X → Y of topological
spaces is closed if for each closed set A ⊂ X, the image f [A] is closed in Y .
Here are some instructive examples:
1. The identity map from a set with the indiscrete topology to a set with the discrete topology
is both open and closed but not continuous.
2. The previously constructed map from [0, 1) to S 1 is continuous but neither open nor closed.
Its inverse is open and closed but not continuous.
3. The map from R2 to R sending (x, y) to x is continuous and open but not closed. [Hints:
The proof that the map is open reduces to showing that the image of an open δ-disk is always open.
Why is the image of the open δ-disk about (x, y) equal to the open interval {t | x − δ < t < x + δ }?
To show the map is not closed consider the graph of 1/x and its image under the given map.]
4. The map from R to itself sending x to its absolute value is continuous and closed but not
open. [Hints: To show the map is not open, consider the image of the whole space. To show it is
closed, explain why every closed subset F can be written as a union F = F + ∪ F− where F± is a
closed set consisting of all points in F that are respectively nonnegative or nonpositive. Why does
f take F± to a closed set, and how does this show that f is closed?]
The following result is elementary:
PROPOSITION 8. The composite of two open mappings is open, and the composite of two
closed mappings is closed. Identity maps are always open and closed.
For metric spaces there is an extremely special type of homeomorphism:
Definition. Let (X, dX ) and (Y, dY ) be metric spaces. A set-theoretic

function f : X → Y is
said to be an isometry if it is onto and d X (u, v) = dY f (u), f (v) for all u, v ∈ X. — Such a map
is automatically 1–1 because


u 6= v =⇒ dY f (u), f (v) = dX (u, v) > 0 .

26
By construction such a map is also uniformly continuous and has a uniformly continuous inverse
(which is also an isometry). Note that identity maps are always isometries, the composites of
isometries are isometries, and inverses to isometries are isometries.
There is an extensive further discussion of homeomorphisms, their geometrical properties, and
related mappings (with illustrations) in the course directory file homeomorphisms.pdf.

Different metrics determining the same topology

The discussion of diameters and bounded metric spaces on pages 121–122 of Munkres should
be read at this point. Theorem 20.1 on page 121 is an important case of an extremely general
phenomenon:
If (X, d) is a metric space, then in general there are many different metrics e such that the
identity map from (X, d) to (X, e) is a homeomorphism. Furthermore, in general there
are many examples for which the identity map is also uniformly continuous, and in fact
one can even find large classes of examples for which the identity map in the opposite
direction is also uniformly continuous.
Important examples of this will arise later in the course.

Metric spaces of functions

We have already noted that the set BF(X) of bounded functions on a set X has a metric space
structure with  
d(f, g) = sup f (x) − g(x) .
x

In such a space the limit of a sequence of functions corresponds to uniform convergence: We have
limn→∞ fn = f if and only if for all ε > 0 we can find M so that n > M implies |f n (x) − f (x)| < ε
for all x ∈ X (this is not quite trivial because the latter inequality only implies |f n − f | ≤ ε, but
if the condition holds we can also find M 0 so that n > M 0 implies |fn (x) − f (x)| < ε/2 and the
latter certainly implies |fn − f | ≤ ε/2 < ε), A large amount of the theory of uniform convergence
for functions of real variables carries over to this general setting. In particular, pages 147–151 of
Rudin, Principles of Mathematical Analysis (Third Edition), go through with only minor changes.
In particular, at certain places in a point set topology course it is necessary to use the following
result, which is proved on page 132 of Munkres or pages 149–150 of the book by Rudin mentioned
above.
THEOREM 9. Let X be a topological space, let BF(X) be defined as above, and let BC(X)
denote the set of all continuous functions in BF(X). Suppose that lim n→∞ fn = f in BF(X) where
each of the functions fn is continuous. Then f is also continuous.
The spaces BC(X) also have a great deal of algebraic structure (for example, addition and
multiplication of functions) that one has for continuous functions on, say, the unit interval. This is
all discussed in Section 21 of Munkres (pp. 129–133).

27
 Piecing together continuous functions

The previous material provides one powerful method for constructing continuous functions on
metric and topological spaces. This is essentially an analytic method for constructing functions. It
is often important to have a similarly useful geometric method for obtaining continuous functions.
Geometric piecing problem. Suppose that we are given topological spaces X and Y , a family
of subsets {Aα } of X and a continuous function fα : Aα → Y for each α. Is it possible to form a
continuous function g : ∪α Aα → Y such that g(x) = fα (x) if x ∈ Aα ?
Note. Given a function h : A → B and C ⊂ B, the composite of h with the inclusion C → A
is the restriction of h to C, which we are denoting by h|C. It follows that if h is a continuous map
of topological spaces and C has the subspace topology then h|C is automatically continuous (in
fact, if A is a metric space then the inclusion map is uniformly continuous).
There is an obvious set-theoretic condition that is necessary if a function g as above actually
exists. Namely, for all y ∈ Aα ∩ Aβ we need the consistency condition f α (y) = fβ (y). In terms of
the restriction notation this can be rewritten formally as

fα | A α ∩ A β = fβ | Aα ∩ Aβ .

For certain families of subspaces this turns out to be the only condition needed to piece together a
continuous function defined on an entire space.
THEOREM 10. Let X and Y be topological spaces, let A = {A α } be a family of subsets of X
such that X = ∪α Aα , and for each let fα : Aα → Y be a continuous function. Assume that these
functions satisfy the consistency condition f α | Aα ∩ Aβ = fβ | Aα ∩ Aβ . If either
(i) A is a family of open subsets, OR
(ii) A is a finite family of closed subsets,
then there is a unique continuous function f : X → Y such that f |A α = fα for all α.
To see that a similar result does not hold for arbitrary families of closed subsets, consider the
family A of one point sets {x}. Given an arbitrary set-theoretic function g from a metric space
X to a topological space Y , the restrictions g|{x} are all continuous and the consistency condition
follows because the sets in the family are pairwise disjoint. Thus any discontinuous function g
satisfies the conditions for the family of closed subsets A, and for most choices of X there are many
discontinuous functions to choose from.
Proof of Theorem. First (open covering) case. The consistency condition ensures that there
is a well-defined set-theoretic function f : X → Y with the desired properties, so the real issue is
to prove this function is continuous. Let V be an open subset of Y . Then we have
[

f −1 [V ] = Aα ∩ f −1 [V ]
α

and since (f |Aα )−1 [V ] = Aα ∩ f −1 [V ] the right hand side of the displayed expression is simply the
−1
union of the sets (f |Aα ) [V ] = fα−1 [V ]. By the continuity of the functions f α the sets on the right
hand side of this expression are open, and therefore the union of these sets, which is just f −1 [V ],
is also open, proving that f is continuous.

28
 Second (finite closed covering) case. Many of the steps in the argument are the same so we
shall concentrate on the differences. First of all, we need to replace the open subset V with a closed
subset F . The same argument then shows that
[
f −1 [F ] = fα−1 [F ]
α

where each summand on the right hand side is closed by continuity. Since the union on the right
hand side is finite, the union on the right hand side is again a closed subset, and this implies that
f −1 [F ] is closed in X.

II.4 : Cartesian products

(Munkres, §§ 15, 19)

Product constructions are useful in mathematics both as a means of describing more compli-
cated objects in simpler terms (for example, expressing vectors in terms of magnitude and direction
or resolution into x, y and z components) and also as the basis for considering quantities (formally,
functions) whose values depend upon several variables.

Topological structures on finite products

Since product structures are less ambiguously defined for topological spaces as opposed to
metric spaces, we shall begin with the former.
There are two ways of viewing Cartesian products with finitely many factors. Clearly one
wants the product of the sets A1 , · · · , An to be the set of all ordered n-tuples (or lists of length n)
having the form (a1 , · · · , an ) where ai ∈ Ai for all i. Formally these can be described as functions
α from {1, · · · , n} to ∪i Ai such that α(i) ∈ Ai for all i. Alternatively, one can view finite products
as objects constructed inductively from 2-fold products by the recursive formula

A1 × · · · × An+1 = (A1 × · · · × An ) × An+1

for all n ≥ 2. It is an elementary exercise to see that the two formulations both result in equivalent
concepts of ordered n-tuples with the property

(a1 , · · · , an ) = (b1 , · · · , bn ) ⇐⇒ a i = bi , ∀i.

An abstract approach to this equivalence is described in the file characterizations.pdf.

Definition. Let n ≥ 2 be an integer, and for each integer i between 1 and n let (X i , Ti ) be a
topological space. The product topology on X 1 × · · · × Xn is the topology generated
Q by all sets
of the form U1 × · · · × Un , where
Q U i ∈ T i for all i. Frequently we shall write i X i to denote the
product of the sets Xi and i (Xi , Ti ) to denote the product topology; if the topologies Q on the
factors are clear from the context and it is also clear that we want the product topology on i Xi
we shall frequently use the latter to denote the product space.

29
 Before proceeding further we make a simple but useful observation.
PROPOSITION 1. Every open subset in the product topology is a union of open subsets of
the form U1 × · · · × Un where Ui ∈ Ti for all i.
Proof. The topology generated by a family F of subsets consists of arbitrary unions of finite
intersections of sets in F, so it suffices to show that the latter is closed under finite intersections;
by associativityQand induction
Q it suffices to check this for the intersections of pairs of subsets. But
if we are given i Vi and i Wi where Vi and Wi are open in Xi for all i, then se have
Y \ Y Y
Vi Wi = (Vi ∩ Wi )
i i i

so that the family of products of open subsets is closed under finite intersections.
COROLLARY 2. Let Bi be a base for the topology on Xi . Then every open subset in the
product topology is a union of open subsets of the form V 1 × · · · × Vn where Vi ∈ Bi for all i.
Example. The online file nonrectangular.pdf contains a proof that for each r > 0 the
open disk in R2 defined by x2 + y 2 < r 2 is NOT a union of finitely many product open sets of the
form Ui × Vi , where Ui and Vi are open subsets of R.
Proof of the corollary. Since a union of unions is a union, it suffices to show this for open
subsets of the form U1 × · · · × Un , where Ui ∈ Ti for all i. Since each Bi is a base for Ti , we can
express each Ui as a union ∪α[i] Vα[i] , and it follows that U1 × · · · × Un is equal to
   
[ [ [ Y
 Vα[1]  × · · · ×  Vα[n]  = Vα[i]
α[1] α[n] (α[1], ... ,α[n]) i

and hence U1 × · · · × Un is a union of products of the prescribed type.

The following result provides some motivation for the definition:
THEOREM 3. For each n ≥ 2 the topology on Rn with respect to the Euclidean metric is equal
to the product topology associated to the family (X i , Ti ), where each space is the real line with
the usual topology.
Proof. FIRST STEP — Open sets in the product topology are open in the metric topology. By
the preceding corollary, every open subset in the product topology is a union of sets of the form
U1 × · · · × Un where each Ui is an open interval in R, and since arbitrary unions ofQmetrically
open sets are metrically open, it suffices to show that each product of open intervals i (ai , bi ) is
open in the metric topology. Let x = (x 1 , · · · , xn ) be a point in this product, and let ε > 0 be
such that ε < (xi − ai ), (bi − xi ) for all i. If y = (y1 , · · · , yn ) satisfies d(y, x) < ε with respect
to the standard Euclidean metric then we have

|yi − xi | ≤ d(y, x) < ε

for all i. It is an elementary exercise toQcheck that this displayed inequality and the previous one
imply yi ∈ (ai , bi ) for all i. Therefore i (ai , bi ) is open in the metric topology because N ε (x) is
contained in this product.
SECOND STEP — Open sets in the metric topology are open in the product topology. It
suffices to show that for each x ∈ Rn and ε > 0 there is a δ > 0 so that the open rectangular region

30
Q
− δ, xi + δ) is contained in Nε (x). To see this, note that given a metrically open subset U
i (xi
we may write it as [
Nε(x) (x)
x∈U

for suitable positive real numbers ε(x). The latter union then contains the union
[ Y

xi − δ(x, ε), xi + δ(x, ε)
x∈U i

which in turn contains ∪x {x} = U . Thus U is a union of sets having the form
Y
(xi − δ, xi + δ)
i

and hence is open in the product topology.

We must now verify the statement about N ε (x) at the beginning of the previous paragraph.
The proof is best understood using a simple picture in the plane. Consider the open disk in the
uv-plane consisting of all points for which u 2 + v 2 < 1. How large of an open square centered at the
origin can one fit inside this open disk? In particular, one can ask this for a square whose sides are
parallel
√ to√the coordinate axes. It turns out that the square in question has one vertex of the form
(1/ 2, 1/ 2) and the other three vertices given by multiplying either or both coordinates of the
latter by −1. Now suppose we are looking inside the open unit disk in coordinate 3-space.√ What
is the largest cube in that case? The coordinates of the vertices turn out to be ±1/ 3. One can
then form an educated guess regarding the vertices for a maximal n-dimensional hypercube inside
the unit n-dimensional hyperdisk.
Formally, proceed as follows. Given a fixed n and an arbitrary ε > 0, let
ε
δ = √
n
Q
and consider the set i (xi − δ, xi + δ). If y belongs to this set then |yi − xi | < δ for all i and
therefore !1/2 !1/2
X X ε2
d(x, y) = |xi − yi |2 < = ε
i i
n

as required.

General properties of (finite) product topologies

Given a sequence of sets X1 , · · · , Xn and an integer j between 1 and n, there is a map

Y
pj : Xi −→ Xj
i

called projection onto the ith coordinate defined by the formula

pj (x1 , · · · , xn ) = xj .

31
The following result characterizes the product topology in terms of these projections:
Q
PROPOSITION 4. Let (Xi , Ti ) Q Q space for 1 ≤ i ≤ n, and let i Ti denote
be a topological
the product topology on the product i Xi . Then i Ti is the unique smallest topology such that
each projection map pj is continuous.
Proof. Continuity of projections. Let W be open in X j . Then
Y
p−1
j [W ] = Wi
i

where Wi = Xi if i 6= j and Wj = W . This product set is open in the product topology and
therefore pj is continuous.
Q
Minimality
Q property. Suppose that T is a topology on i Xi such that each pj is continuous.
Let U = i Ui where Ui is open in Xi ; we need to show that U is open with respect to T. By the
continuity of the projections we know that each set p −1
j [Uj ] is open with respect to T, and therefore
the finite intersection \ Y
p−1
i [Ui ] = Ui
i i

is open
Q with respect to T. Since every open set in the product topology is a union of sets of the
form i Ui it follows that every open set in the product topology is also open with respect to T.
COROLLARY 5. Let (Xi , Ti ) be a topological space for 1 ≤ i ≤ n, let (Y, Q W) be a topological
space, for each i let fi : Y → Xi be a set-theoretic function, and let f : Y
→ i Xi be the unique
Q f (y) = f1 (y), · · · , fn (y) . Then f is continuous
function such that f o pi = fi for all i so that
(with respect to the product topology on i Xi ) if and only if each function fi is continuous.
Proof. If f is continuous then the continuity of the projections p i and the continuity of composites
imply that each fi is continuous because fi = pi o f .
Now suppose
Q that each fi is continuous. If we can show that the inverse image of each basic
open subset i Ui is under f is open, then since inverse images preserve unions it will follow that
the inverse image of every open set under f is open and hence that f is continuous. As before we
know that Y \
Ui = p−1
i [Ui ]
i i

and if we take inverse images (and use the fact that inverse images preserve intersections) then we
have " #
Y \ \
f −1 Ui = f −1 o p−1
i [U i ] = fi−1 [Ui ]
i i i

and the latter is open because each f i is continuous.

By construction a product of open subsets is open in the product topology (where we are only
dealing with finite products). The analogous statement for closed subsets is also true:
PROPOSITION 6. Let (Xi , Ti ) be aQtopological space for 1 ≤ i ≤ Qn, and for each i suppose
that Fi is a closed subset of Xi . Then i Fi is a closed subset of i Xi with respect to the
product topology.
Proof. This follows from the set-theoretic equation
\ Y
p−1
i [Fi ] = Fi
i i

32
the continuity of the projections p i and the fact that inverse images of closed subsets with respect
to a continuous function are closed.
COROLLARY 7. Let (Xi , Ti ) be a topological space for 1 ≤ i ≤ n, and for each i suppose that
Ai is a subset of Xi . Then
Y Y
Ai = Ai .
i i

Proof. The first set in the display contains the second because the first is a closed set containing
the product of the Ai and the second is the smallest such closed subset. To see that the first is
contained
Q in the second, let b be a point in the product
Q of the closure,Qand let U be an open subset
of i Xi that contains b; we need to prove that U ∩ i Ai 6= ∅. Let i Vi be a basic open subset
that contains b and is contained in U . Since the coordinates
Q b satisfy b j ∈ Aj for allQj, it follows
of Q
that Vj ∩ Aj 6= ∅ Qfor all j, and from this we have that i Vi ∩ i Ai 6= ∅; Qsince U ⊃ i Vi it also
follows that U ∩ i Ai 6= ∅. But this means that b lies in the closure of i Ai .
Projection maps also have the following important property.
Q
PROPOSITION 8. (Openness of projections) The coordinate projection maps p j : i Xi → X j
are open.
Proof. The set-theoretic equality
h[ i [
g Wα = g[Wα ]
α

showsQthat it suffices to prove pj [W ] is open if W is a basic open subset. But such a set has the
form i Ui where each Ui is open in Xi , and the image of this set under pj is simply Wj .
We have already given an example to show that coordinate projections are not necessarily
closed; namely projection onto either coordinate is a continuous and open map from R 2 to R, but
the image of the closed set of points satisfying the equation xy = 1 (geometrically a hyperbola
whose asymptotes are the x- and y-axes) is R − {0}, which is not a closed subset of the real line.

Products and morphisms

If we are given a sequence of set-theoretic functions f i : Xi → Yi , then there one can define
the Cartesian product of morphisms
Q Y Q
F = i fi : Xi −→ i Yi
i

by the formula

F (x1 , · · · , xn ) = f1 (x1 ), · · · , fn (xn )
or alternatively by the conditions
πiY o F = fi o πiX
Q Q
where πiX and πiY denote the ith coordinate projections for i Xi and i Yi respectively. Maps
of this sort arise very frequently when one constructs new continuous functions out of old ones. If
n = 2 one often describes such product maps using notation of the form

f1 × f 2 : X 1 × X 2 −→ Y1 × Y2

33
and similar notation is often used for other small values of n. Here are some properties of the product
construction that are extremely elementary but also extremely important in many situations:
PROPOSITION 9. (i) In the preceding notation, if each X i and Yi are topological spaces the
function F is continuous with respect to the product topologies if and only if each f i is continuous.
(ii) If each fi is an identity map, then so is F .
Suppose we are also given sets Z i and (set-theoretic) maps gi : Yi → Zi , and we set G
(iii) Q
equal to i gi . Then Q
G oF = i (gi fi ) .
o

The verifications of these statements are left to the reader as exercises.

Another important class of morphisms involving products are the maps that permute coordi-
nates. We shall only discuss the simplest example here. Given two topological spaces X 1 and X2
the twist map or transposition map

τ (X1 , X2 ) : X1 × X2 −→ X2 × X1

is the map sending (a, b) ∈ X1 × X2 to (b, a) ∈ X2 × X1 . These maps have the following elementary
but important properties:
PROPOSITION 10. If X1 and X2 are topological spaces, then τ (X1 , X2 ) is continuous with re-
spect to the product topologies. In fact, it is a homeomorphism whose inverse is given by τ (X 2 , X1 ).
Proof. The second statement is purely set-theoretic, and it is elementary to verify. To check the
continuity of the twist map, let p1 and p2 be the coordinate projections for the domain and let q 1
and q2 be the coordinate projections for the codomain. We then have the identities

q1 o τ (X1 , X2 ) = p2 , q2 o τ (X1 , X2 ) = p1

and the continuity of τ (X1 , X2 ) follows immediately from these.

Products and metric spaces

Q
If (Xi , di ) are metric spaces for 1 ≤ i ≤ n, then it is possible to put metrics on i Xi whose
underlying topologies are the product topology. In fact, there are three particularly important
product Qmetrics. We shall describe three specific examples that are particularly significant. Let
x, y ∈ i Xi and express them in terms of coordinates as (x 1 , · · · , xn ) and (y1 , · · · , yn ) respec-
tively. Then the following formulas define metrics on the product:
dh∞i (x, y) = maxi { di (xi , yi ) }
P

2 1/2
dh2i (x, y) = i di (xi , yi )
P
dh1i (x, y) = i di (xi , yi )

The verification that each formula defines a metric is left to the reader as an exercise. We then
have the following result:
PROPOSITION 11. The topology Q determined byQthe metric d h∞i is the product topology.
Furthermore, the identity map from ( i Xi , d ) to ( i Xi , dhβi ) is uniformly continuous for all
hαi

choices of α, β ∈ {1, 2, ∞}.

34
 The online file http://math.ucr.edu/∼res/math145A-2014/product-metrics.pdf contains
a drawing which compares the neighborhoods of radius r for the product metrics d hpi for p = 1, 2, ∞.
As noted in that file, one in fact has a continuous family of product metrics d hpi for 1 ≤ p ≤ ∞.
In the companion files
http://math.ucr.edu/∼res/math145A-2014/product-metrics2.pdf
http://math.ucr.edu/∼res/math145A-2014/product-metrics3.pdf
there are drawings of the disks of fixed radius r centered at 0 ∈ R 2 with respect to such product
metrics, and certain inequalities andQlimit relationships Q
are also derived. The latter imply that the
proposition generalizes to all pairs ( i Xi , dhαi ) and ( i Xi , dhβi ) such that 1 ≤ α, β ≤ ∞.
Proof. To verify the assertion about d h∞i note that

dh∞i (x, y) < ε ⇐⇒ yi ∈ Nε (xi ), ∀i .

Thus the ε-neighborhood of x with respect to the d h∞i metric is just

Y
Nε (xi ) .
i

Q
By previous results, a base for the product topology on i Xi is given by open sets of the form
Y
Nε(i) (xi )
i

and this implies that the product

Q topology contains the metric topology. On the other hand, for
each j the projection map pj : i Xi → Xj is uniformly continuous because


dj pj (x), pj (y) = dj (xj , yj ) ≤ dh∞i (x, y)

implies we can take δ = ε in the criterion for uniform continuity. This means that the metric
topology contains the product topology, and therefore by the previous observations we see that the
topologies are equal.
The uniform continuity statements are direct consequences of the following inequalities for
nonnegative real numbers αi for 1 ≤ i ≤ n:
!1/2
X X
maxi { αi } ≤ α2i ≤ αi ≤ n · maxi { αi }
i i

The middle inequality is perhaps the least trivial, and it can be verified by squaring both sides and
noting that the corresponding inequality holds for the squares. These inequalities imply that the
identity maps
! ! ! !
Y Y Y Y
h∞i h1i h2i h∞i
Xi , d −→ Xi , d −→ Xi , d −→ Xi , d
i i i i

are uniformly continuous (and in fact the δ corresponding to a given ε can be read off explicitly
from the inequalities!), and of course the composites of any two or three consecutive maps from
this diagram are also uniformly continuous.

35
 The following basic result on products and metric spaces is also worth mentioning:
PROPOSITION 12. If X is a metric space then the distance function d : X × X → R is
uniformly continuous (where R has the usual topology).
Proof. Let (x, y) ∈ X × X, and view the product topology as coming from the maximum metric
by the preceding discussion. Given ε > 0 suppose that (u, v) ∈ X × X satisfies

ε
max d(x, u), d(y, v) < .
2

Then several applications of the Triangle Inequality show that d(u, v) − d(x, y) < ε and therefore
d is uniformly continuous.

Products and the Hausdorff Separation Property

Fundamental Definition. We shall say that a topological space X has the Hausdorff Separation
Property (or more simply, it is a Hausdorff space or it is Hausdorff )) if for each pair of distinct
points u, v ∈ X there are disjoint open subsets U, V ⊂ X such that u ∈ U and v ∈ V . As noted
in Section II.1, metric spaces have this property but it does not necessarily hold for an arbitrary
topological space (specifically, this is true for the indiscrete topology on a space containing more
than one point or a finitary topology on a space with infinitely many points).
PROPOSITION 13. In a Hausdorff space every one point subset is closed.
Proof. Given p ∈ X we shall show that X − {p} is open if X is Hausdorff. Suppose that
y ∈ X − {p}. Then there are disjoint open subsets U y and Vy such that p ∈ Uy and y ∈ Vy .
Therefore we have [ [
X − {p} = {y} ⊂ Vy ⊂ X − {p}
y6=p y6=p

which implies that the last two subsets are equal, and thus X − {p} is open because it is a union
of open subsets.
We now come to a result that has appeared on countless examinations:
THEOREM 14. Given a set X, let the diagonal δ X denote the set of all points (u, v) ∈ X × X
such that u = v. Then X is Hausdorff if and only if ∆ X is closed in X × X with respect to the
product topology.
Proof. This follows because each of the statements listed below is equivalent to the adjacent
one(s):
(1) X is Hausdorff.

v) ∈ X × X − ∆X there are open subsets U, V ⊂ X such that u ∈ U , v ∈ V and

(2) Given (u,
U × V ∩ ∆X = ∅.
(3) X × X − ∆X is open in X × X with respect to the product topology.
(4) ∆X is closed in X × X with respect to the product topology.
Taken together, these prove the result.
The theorem has an extremely important consequence:

36
PROPOSITION 15. Let Y be a Hausdorff space, and let f and g be continuous functions from
a topological space X to Y . Then the equalizer set

E = { x ∈ X | f (x) = g(x) }

is closed in X.
Proof. If H : X → Y × Y is the function defined by


H(x) = f (x), g(x)

then we have already noted that H is continuous if f and g are continuous. It follows immediately
that E = H −1 [∆Y ]. Since Y is Hausdorff we know that ∆Y is closed in Y × Y and therefore its
inverse image under H, which is simply E, is closed in X.
SPECIAL CASE. If f and g are continuous real valued functions on the unit interval [0, 1] and
f (x) = g(x) for all rational points of [0, 1], then f = g.
Proof. The proposition shows that if f and g are continuous functions from the same space X
into a Hausdorff space Y and f |A = g|A then f |A = g|A. In this case A is the set of all rational
points in X = [0, 1] and A = X. More generally this argument shows that if Y is a Hausdorff space,
X is any space and A ⊂ X is a subspace such that A = X and f |A = g|A, then f = g.
It is easy to construct counterexamples to the conclusion of the proposition if the codomain is
not Hausdorff. Suppose that X and Y both have the associated indiscrete topologies where both
sets have at least two elements. Then every function from X to Y is continuous, and every nonempty
subset A ⊂ X is dense (i.e., A = X). Large families of counterexamples can be constructed in this
manner; details are left to the reader as an exercise.

Infinite products

In earlier decades infinite products of topological spaces received a great deal of attention in
point set topology. We shall not deal with such objects extensively here, but it seems worthwhile
to say a little about them for the sake of completeness and to avoid some natural possibilities for
misunderstandings.
Given an indexed family of sets Xα with indexing set A, the set-theoretic cartesian product
Y
Xα
α∈A

may be defined formally as the set of all set-theoretic functions x from A to ∪ α Xα such that
x(α) ∈ Xα for all α. This captures the intuitive ideas that the elements of the cartesian product
are given by the coordinates xα and that two elements are equal if and only if all their coordinates
are equal; in particular, if A = {1, · · · , n} then there is a canonical 1–1 correspondence between
this set and the earlier definition of Cartesian product.
The Axiom of Choice in set theory is equivalent
Q to the statement that if each of the sets X α
is nonempty, then so is their cartesian product α Xα .
Q
As in the case of finite products there are projection maps p β : α Xα → Xβ defined by
pβ (x) = x(β).

37
 Assume now that we have an indexed
Q family of topological spaces (X α , Tα ). The crucial
property of the product topology for α Xα will be that it is the unique smallest topology such
that every projection map pβ is continuous.
The preceding condition implies that the product topology should be generated by all sets of
the form p−1β [Uβ ] where Uβ is open in Xβ , and thus the product topology will be arbitrary unions
of finite intersections of such sets.
PROPOSITION
Q 16. A base for the product topology is given by all open subsets of the form
α Uα where each Uα is open in Xα AND Uα = Xα for all but finitely many α.
Proof. The subsets described in the proposition are finite intersections of sets having the form
p−1
β [Uβ ]; specifically, if Γ is a finite subset of A then the subsets in the proposition have the form

\
p−1
γ [Uγ ] .
γ∈Γ

Another
Q topology on the product is the so-called box topology generated by all subsets of the
form α Uα where Uα is an arbitrary open subset of Xα . For finite products these yield the same
topology, but this is not true for infinite products. A fairly detailed discussion of the differences
appears in Section 19 of Munkres.
Final remark. Theorem 19.6 on page 117 of Munkres gives a fundamentally important
property of the product topology (in both the finite and infinite cases).
Finally, here are some facts about finite products that carry over to infinite products. The
proofs are essentially the same.
PROPOSITION 17. (Openness of projections) For each β in the indexing set the coordinate
projection maps Y
pβ : Xα −→ Xβ
α
are open.
PROPOSITION 18. (Products of closed subsets) Let (X α , Tα ) be
Q a topological space for
α ∈QA, and for each α suppose that Fα is a closed subset of Xα . Then α Fα is a closed subset
of α Xα with respect to the product topology.
In the preceding result, one should note that, in contrast to Proposition 15 above, the statement
remains true even if infinitely many of the closed subsets F α are proper subsets of the corresponding
spaces Xα .
COROLLARY 19. Let (Xα , Tα ) be a topological space for α ∈ A, and for each α suppose that
Aα is a subset of Xα . Then
Y Y
Aα = Aα .
α α

THEOREM 20. (Mappings into products) Let A be a set, and for each α ∈ A let f α : Xα → Yα
be a set-theoretic map. Then there is a unique map
Q Y Q
F = α fi : Xα −→ α Yα
α

38
defined by the conditions
παY o F = fα o παX
Q Q
where παX and παY denote the αth coordinate projections for α Xα and α Yα respectively. This
map is continuous if and only if each f α is continuous, and it is the identity map if each f α is an
identity map. Finally,
Q if we are also given sets Z α and (set-theoretic) maps gα : Yα → Zα , and we
set G equal to α gα , then Q
GoF = α (gα fα ) .
o

Finally we mention one more that is an exercise in Munkres (Theorem 19.4, page 116; see also
Exercise 3 on page 118). One proof (probably not the best one!) for products of two spaces appears
in Section III.1 of these notes.
THEOREM 21. (Products and the Hausdorff Separation Property) Let A beQa nonempty set,
and supposed that Xα is a Hausdorff topological space for each α ∈ A. Then α Xα is also a
Hausdorff space.

39
 III . Spaces with special properties

We have seen that one can derive a relatively sizable amount of information simply from the
axioms for metric and topological spaces. However, it should not be surprising that one needs to
impose further conditions on spaces in order to prove more substantial results, including abstract
versions of the Maximum Value Theorem and Intermediate Value Theorem from single variable
calculus. It turns out that these two results relies on a separate basic properties of the open subsets
in a closed interval. The underlying concepts are known as compactness and connectedness, and
they are treated in this unit.
In calculus it is also important to know that certain infinite series have meaningful sums,
and indeed one reasons that mathematicians tightened their standards of logical rigor in the 19 th
century was to analyze the validity of certain strange and unanticipated results that arose from
casual manipulations with infinite series; some of the results were justified, but others were not (this
was expected because some of the formulas contradicted each other). One abstract version of the
basic condition guaranteeing convergence of reasonable infinite series is called completeness, and
it is also discussed in this unit along with some important geometrical an analytical implications
(however, the applications to analysis go far beyond the scope of this course).

III.1 : Compact spaces – I

(Munkres, §§ 26, 27)

One of the most fundamental properties of continuous functions on closed intervals is that they
have maximum and minimum values. In contrast, a continuous function on an open or half open
interval does not necessarily have this property. In most real variables courses, the existence of
maximum and minimum values is established with the help of the Heine-Borel-Lebesgue Theorem
(sometimes the third name is dropped when referring to this result, and sometimes the first name
is dropped). The conclusion of this result is so important that it has become incorporated into a
definition. However, before proceeding to the main result we need a preliminary concept.
Definition. If X is a topological space and U = {U α } is a family of open subsets of X, we say
that U is an open covering of X if ∪α Uα = X. A subfamily V ⊂ U is said to be a subcovering if
∪β Vβ = X, where V = {Vβ }.

Definition. A topological space is said to be compact if every open covering has a finite
subcovering.
The main point of the Heine-Borel-Lebesgue Theorem is that closed intervals in the real line
are compact. An abstract version of this result is established as Theorem 27.1 on pages 172–173
of Munkres (see Corollary 27.2 on the second of these pages for the case of interest to us here). In
fact, one has the following characterization of compact subsets of the real line.
Characterization of compact subsets. A subset K of the real line is compact if and only if
it is closed and bounded.

40
 Several portions of the proof are true under much more general conditions, so we shall establish
these first.
THEOREM 1. If A is a compact subset of a Hausdorff space, then A is closed in X.
Proof. We use the Hausdorff Separation Property to show that X − A is open.
Let y ∈ X − A, then for each a ∈ A we have y 6= a, and therefore by the Hausdorff Separation
Property there are open sets U(a,y) and V(a,y) (in X) containing y and a respectively such that
U(a,y) and V(a,y) are disjoint. The family of subsets {A ∩ V (a,y) } is an open covering of A and thus
has a finite subcovering
A ∩ V(a1 ,y) , · · · , A ∩ V(ak ,y) .
By construction we have A ⊂ V(a1 ,y) ∪ · · · ∪ V(ak ,y) , and thus if we take

Uy = U(a1 ,y) ∩ · · · ∩ U(ak ,y)

then Uy is an open subset containing y and Uy ∩ A = ∅. A (by now) familiar argument shows that
X − A = ∪y Uy and hence that X − A is open.
PROPOSITION 2. If A is a closed subset of a compact topological space X, then A is compact.
Proof. Let U = {Uα } be an open covering of A; choose open sets V α in X so that Uα = A ∩ Vα ,
and let V be the open covering of X given by the sets V α together with X − A. By compactness
of X there is a finite subcovering, which we may as well assume contains X − A as well as open
subsets Vα(1) , · · · , Vα(k) . It then follows that the corresponding subsets U α(1) , · · · , Uα(k) for a
finite subcovering of A.
THEOREM 3. If f : X → Y is continuous and X is compact, then its image f [X] is also
compact.
Proof. Let {Uα } be an open covering of f (X), and choose open subsets V α in Y so that
Uα = f [X] ∩ Vα . Then the sets

Wα = f −1 [Uα ] = f −1 [Vα ]

form an open covering of X, so there is a finite subcovering of X having the form W 1 , · · · , Wk . But
 
f [Wj ] = f f −1 [Uj ] = Uj

and therefore the sets U1 , · · · , Uk for a finite (open) subcovering of f [X].

Proof of characterization of compact subsets of the real line. (=⇒) By the first result
above a compact subset of the real line is closed. To see that it is bounded, consider the open
covering given by the intersections of A with the open intervals (−n, n) where n runs through the
positive integers. If A were not bounded, this open covering would not have a finite subcovering,
so A must be bounded as claimed.
(⇐=) If A is bounded then A is a subset of some closed interval [−M, M ]. Since A is closed
in R, it is also closed in the compact set [−M, M ], and therefore A is compact by the second of the
results above.
With a little additional effort one can modify the proof of the Heine-Borel-Lebesgue Theorem
to show that every box-shaped subset of R k of the form

[a1 , b1 ] × · · · × [ak , bk ]

41
is compact (see Theorem 2.40 on page 39 of Rudin, Principles of Mathematical Analysis).
We shall also give an alternate and more abstract proof of this result later in the course. This
generalization yields an extension of the characterization of compact subsets from R to R k for all
positive integers k.
COROLLARY 4. If k is a positive integer, then a subset A of R k is compact if and only if it is
closed and bounded.
Sketch of proof. (=⇒) The subset A is closed for the same reasons as before. If f j denotes the
restriction of the j th coordinate function to A, then fj (A) is a compact and hence bounded subset
of R. If we choose M > 0 so that ∪j fj [A] ⊂ [−M, M ], then a ∈ A ⇒ a = (a1 , · · · ak ) where
|aj | ≤ M for all j. Hence A is bounded.
(⇐=) If A is closed and bounded then for some M > 0 we know that A is a closed subset of
the rectangular compact set
[−M, M ] × · · · × [−M, M ]
and therefore A is compact.
The following consequence of the preceding results is a far-reaching generalization of a funda-
mental result from calculus.
COROLLARY 5. If X is compact and f : X → R is continuous, then f attains maximum and
minimum values on X.
Proof. This reduces to showing the following: If A ⊂ R is compact, so that it is closed and
bounded, then both the least upper bound and greatest lower bound of A belong to A. We shall
only verify the statement regarding the least upper bound; the other statement follows by reversing
the directions of all inequalities.
Let M be the least upper bound of A (which exists because A is bounded). The for every
positive integer n we can find a point a n ∈ A such that
1
M− < an ≤ M
n
where the second inequality is true because M is an upper bound for A. It follows immediately
that M = limn→∞ an , and since A is closed it follows that M ∈ A.

The Finite Intersection Property

There is a characterization of compactness in terms of closed subsets. Given a family A = {F α }

of closed subsets of a topological space, we shall say that A has the finite intersection property if

Fα(1) ∩ · · · ∩ Fα(k) 6= ∅

for all finite subcollections

{ Fα(1) , · · · Fα(k) } ⊂ A .

THEOREM 6. A topological space X is compact if and only if for every family of closed subsets
A = {Fα } with the finite intersection property we have ∩ α Fα 6= ∅.
A proof of this result and some further remarks appear on page 170 of Munkres.

42
 Compactness and continuous mappings

We have already noted that continuous map that is 1–1 and onto is not necessarily a homeo-
morphism. However, if one puts suitable hypotheses on the domain or codomain it is sometimes
possible to prove that a 1–1 onto continuous map is a homeomorphism without checking the con-
tinuity of the inverse directly. In particular, this holds for compact metric spaces.
PROPOSITION 7. Suppose that f : X → Y is a continuous map from a compact topological
space to a Hausdorff space. Then f is a closed mapping.
Proof. Suppose that A is closed in X. Then A is compact, and therefore f [A] is also compact in
Y . But since Y is a Hausdorff space this implies that f [A] is closed in Y .
The preceding result yields an extremely simple and useful criterion for showing that certain
continuous and 1–1 onto maps are homeomorphisms (i.e., the inverses are continuous).
THEOREM 8. If f : X → Y is a continuous and 1 − 1 onto map from a compact topological
space to a Hausdorff space, then f is a homeomorphism.
Example. A continuous map from a compact topological space onto a Hausdorff space need not
be open. Let X ⊂ R be the union of the intervals [1, 2] ∪ [3, 4], let Y = [1, 3], and let f : X → Y be
given by f (x) = x if x ∈ [1, 2] and f (x) = x − 1 if x ∈ [3, 4]. Clearly f is continuous, but the set
[1, 2] is open in X while its image — which is also [1, 2] is not open in Y .

Products and compactness

The following sort of question arises frequently in mathematics:

PROPERTIES OF PRODUCTS. If X and Y are systems that have some property P and
there is a reasonable notion of direct product X × Y , does this product also have property P?
Here are some examples involving topological spaces for which there is a positive answer:
1. Suppose that X and Y are discrete spaces. Then X × Y is also discrete. (Proof: If
(x, y) ∈ X × Y then {x} is open in X and {y} is open in Y . Therefore

{(x, y)} = {x} × {y}

is open in X × Y , and since x and y are arbitrary this means that every subset of X × Y is open.)
2. Suppose that X and Y are spaces in which one point subsets are closed. Then the same
is true for X × Y ; the proof is analogous to the previous one.
3. If X and Y are finite, then the same is true for X × Y .
4. If X and Y are homeomorphic to metric spaces, then the same is true for X × Y . In fact,
we have given three ways of constructing a metric on the product.
5. If X and Y are Hausdorff spaces, then X × Y is also Hausdorff. (Proof: One way of
doing this is to use the characterization of a Hausdorff space W in terms of the diagonal ∆ W in
W × W being closed. Let Shuff be the “middle four shuffle map”

X ×X ×Y ×Y −→ X × Y × X × Y

43
that sends (x1 , x2 , y1 , y2 ) to (x1 , y1 , x2 , y2 ). This map is continuous because its projections onto the
four factors are continuous, and the same is true for the inverse map which sends (x 1 , y1 , x2 , y2 ) to
(x1 , x2 , y1 , y2 ). Since Shuff is a homeomorphism, it follows that
 
∆X×Y = Shuff ∆X × ∆Y

is closed in (X × Y ) × (X × Y ). This completes the argument.)

In contrast, here is one example where there is a negative answer:
6. If X and Y are homeomorphic to subsets of the real line, the product X × Y is not
necessarily homeomorphic to a subset of the real line. An easy counterexample is given by taking
X = Y = R. We shall prove this when we discuss connectedness later in the course.
THEOREM 9. The product of finitely many compact spaces is compact.
Using the canonical homeomorphism

(X × Y ) × Z ∼
= X ×Y ×Z

and finite induction we can reduce the proof to the case of a product of two compact spaces. The
proof depends upon the following result which is also useful in other contexts.
PROPOSITION 10. (Tube Lemma. ) Let X and Y be topological spaces such that X is
compact, let y ∈ Y , and let W = { Wα } be a family of open subsets of X × Y such that X × {y}
is contained in ∪α Wα . Then there is a finite open covering U(y) = { U i } of X and an open subset
V (y) of Y containing y such that each product set U i × V (y) is contained in some Wα .
Proof of the Tube Lemma. First of all, we claim that X × {y} is homeomorphic to X and
therefore is compact. To see this, consider the map h y : X → X×{y} defined by hy (x) = (x, y). The
projections onto the factors are the identity and the constant map, and therefore h y is continuous.
In fact, hy is a homeomorphism because projection onto the X factor yields a continuous inverse
to h. Maps of this form are often called slice inclusions.
For each x ∈ X let W (x) be an open subset in W such that x ∈ W (x) × {y}. Let U x and Vx
be open subsets of X and Y respectively such that

(x, y) ∈ Ux × Vx ⊂ W (x) .

Then U = { Ux } is an open covering of X and hence there is a finite subcovering

U(y) = { Ux1 , · · · , Uxn } .

If V (y) = ∩i Vxi , it follows that

Uxi × V (y) ⊂ Uxi × Vxi ⊂ W (xi )

which proves the lemma.

A picture illustrating this proof is given in the files tubelemma.pdf in the course directory.
Proof of Theorem 9. Let W = { Wα } be an open covering of X × Y , and for each y ∈ Y let
W(y) ⊂ W be a family that covers X × {y}.

44
 Given y ∈ Y , let U(y) and V (y) be associated to W(y) as in the Tube Lemma. The sets V (y)
form an open covering of Y and therefore there is a finite subcovering { V (y 1 ), · · · , V (ym ) } .
Then the finite family of sets

U = { Uβ × V (yj ) | Uβ ∈ U(yj ) }

is a finite open covering of X × Y and for each set in the family there is some W γ(β,j) in W such
that
Uβ × V (yj ) ⊂ Wγ(β,j) .
The finite collection of sets Wγ(β,j) is the desired finite subcovering of W.

Compactness and infinite products

The preceding result on compactness of products extends to infinite products provided one
assumes the Axiom of Choice (in fact, the statement of the theorem is equivalent to the latter). This
was established by A. N. Tychonoff and is known as Tychonoff ’s Theorem. Proofs of Tychonoff’s
Theorem and a crucial preliminary result appear on pages 233–235 of Munkres. This result has
fundamental applications in several mathematical contexts, and perhaps the most important involve
functional analysis. In geometric topology, probably the most important applications involve the
Hilbert cube, which is a countably infinite product of copies of the closed unit interval [0, 1]; in this
case Exercise 1 on page 280 gives an alternate proof of compactness for the space in question.
One major application of Tychonoff’s Theorem to analysis is the Banach-Alaoglu Theorem
(the names are pronounced BAHN-akh and ah-la-OH-gloo; see pages 68–69 of Rudin, Functional
Analysis, for a statement and proof of the theorem).

Compact metric spaces

If a compact topological space is determined by a metric, then many additional statements

can be made. For example, we have the following generalization of the boundedness property:
PROPOSITION 11. If X is a compact metric space then there is a constant K > 0 such that
d(x, y) ≤ K for all x, y ∈ X.
Proof. By Example 4 in the list of examples of continuous functions in the previous note, for a
fixed z ∈ X the function f (x) = d(x, z) is (uniformly) continuous. Let M be the maximum value
of this function. Given two points x, y ∈ X the Triangle Inequality now implies that

d(x, y) ≤ d(x, z) + d(y, z) ≤ M + M = 2M

and therefore we can take K = 2M .

Another important and much deeper property of a closed interval in the real line is that every
infinite sequence in the interval has a convergent subsequence (the Bolzano-Weierstrass Theorem).
This property also holds for compact metric spaces.
THEOREM 12. If X is a compact metric space, then every infinite sequence in X has a
convergent subsequence.

45
 Later in this course we shall prove a converse to this result (see Section VI.2).
Proof. Let {an } be an infinite sequence in X, and suppose it has no convergent subsequence.
If the sequence takes only finitely many values, then at least one of them occurs infinitely many
times, and thus one can find a convergent subsequence, so we may as well assume that the sequence
takes infinitely many distinct values. Let A ⊂ X be the set of all these values.
We claim that L(A) = ∅; suppose that b ∈ L(A). One can then recursively construct a
subsequence that converges to b as follows. Suppose that the first r terms of the subsequence a n(k)
have been defined so that d(b, an(j) ) < k1 . Let Ur+1 be the set containing b obtained by taking the
open disk N1/(r+1) (b) and removing all elements a` of the original sequence for ` ≤ n(k) that are
not equal to b. Since one point (and hence finite) subsets of a metric space are closed, it follows
that Ur+1 is open. Therefore, by the definition of a limit point there is some a ∈ A such that
a 6= b and a ∈ Ur+1 . By construction this point has the form a m for some m > n(r), and we set
m = n(r + 1). This yields a subsequence whose limit is b.
Since L(A) is empty it follows that it is contained in A and therefore A is closed. Since X is
compact, so is A.
We shall obtain a contradiction by showing that the infinite set A is not compact. If a ∈ A,
then since a 6∈ L(A) we can find an open subset U a in X such that A ∩ Ua = {a}. It follows that
every one point subset of A is open in the subspace topology and hence that every subset is open
in the subspace topology. Since A is infinite, it follows that the open covering of A by one point
subsets does not have a finite subcovering, which shows that A is not compact.
The contradiction means that our original assumption — the existence of an infinite sequence
in X with no convergent subsequence — is incorrect, and therefore it follows that every infinite
sequence in X has a convergent subsequence.
The next result plays an important role in several analytic and geometric considerations. In
particular, we shall use it to show that a continuous map from a compact metric space to another
metric space is uniformly continuous. The result also has fundamentally important applications in
algebraic and geometric topology.
THEOREM 13. (The Lebesgue Covering Lemma; the name is pronounced le-BAYG) Let X be
a compact metric space, and let U be an open covering of X. Then there is a number η > 0 such
that for every pair of points x, y ∈ X such that d(x, y) < η there is an open set V in U such that
x, y ∈ V .
Proof. For each p ∈ X there is an ε(x) > 0 such that N 2ε(x) (x) is contained in some element of
U. Let Wx = Nε(x) (x).
The family W = {Wx } is an open covering of X, so there is a finite subcovering of the form

{Wx1 , · · · , Wxk } .

Let εj = ε(xj ) > 0, and let η be the minimum of the positive numbers ε 1 , · · · , εk .
Suppose now that d(x, y) < η. Choose i so that x ∈ W xi . Then by the Triangle Inequality we
have
d(y, xi ) ≤ d(y, x) + d(x, xi ) < η + εi ≤ εi + εi = 2εi
which shows that y ∈ N2ε(xi ) (xi ). By the hypotheses, the latter set is contained in some open set
V from the family U, and therefore we have x, y ∈ N 2ε(xi ) (xi ) ⊂ V , where V ∈ U.

46
 A number η satisfying the conditions of the conclusion of the preceding result is called a
Lebesgue number for the open covering. It is easy to see that Theorem 13 fails for noncompact
metric spaces. For example, consider the open covering of the set of the open unit interval (0, 1)
given by the open subintervals  
1 1
,
2k+1 2k−1
where k runs through the positive integers.
The uniform continuity property is an immediate consequence of the Lebesgue Covering
Lemma.
THEOREM 14. (Heine–Cantor Theorem) Let X and Y be metric spaces where X is compact,
and let f : X → Y be continuous. Then f is uniformly continuous.
Proof.  and for each y ∈ Y consider the open set N ε/2 (y) ⊂ Y . By
Let ε > 0 be arbitrary,
−1
continuity the sets f Nε/2 (y) form an open covering of X, and by the compactness of X this
open covering has a Lebesgue number η. Suppose now that u, v ∈ X satisfy d(u, v) < η. Then
there is some y ∈ Y such that u, v ∈ f −1 Nε/2 (y) . It follows that f (u), f (v) ∈ Nε/2 (y), and by
the Triangle Inequality we have
ε ε
d (f (u), f (v)) ≤ d (f (u), y) + d (y, f (u)) < + = ε
2 2
so that
d(u, v) < η =⇒ d (f (u), f (v)) < ε
for all u, v ∈ X.

III.2 : Complete metric spaces

(Munkres, §§43, 45)

Infinite series play an extremely important role in the theory and applications of the real
number system; this is particularly apparent in the computational view of real numbers in terms
of infinite decimal expansions and the use of power series to work with large families of functions
in calculus. However, some care is needed in working with infinite series to ensure the reliability
of any calculations done with them; in particular, it is necessary to know whether or not a series
actually produces a meaningful real number (in other words, it converges). Many of the important
criteria for convergence of infinite series in calculus rely on the property of real numbers known as
completeness. The definition of this concept requires an important preliminary notion.
Definition. Let X be a metric space. A sequence { a n } in X is a Cauchy sequence (or a
fundamental sequence) if for every ε > 0 there is a positive integer N such that m, n > M implies
d(am , an ) < ε.
PROPOSITION 1. Every convergent sequence is a Cauchy sequence.
Proof. Suppose that limn→∞ an = L. Given a positive real number ε, choose M such that n > M
implies d(L, an ) < ε/2. The Triangle Inequality then implies that

d(am , an ) ≤ d(am , L) + d(L, an ) < ε/2 + ε/2 = ε

47
and therefore { an } is a Cauchy sequence.
It is easy to find examples of Cauchy sequences in metric spaces that do not have limits. For
example, take X to be the open interval (0, 1) and consider the sequence a n = n1 . Of course this
sequence does have a limit if one expands X to the closed unit interval. This is a special case of
the following basic property of the real numbers:
THEOREM 2. Every Cauchy sequence in R k converges for all k ≥ 1.
References for the proof of this fact include Theorem 43.2 on page 265 of Munkres and Theorem
3.11 on pages 53–54 of Rudin. The relevance of this theorem to infinite series is explained in a
mathematically rigorous fashion on pages 58–78 of Rudin (see also Exercise 15 on page 81).
Definition. A metric space X is complete if every Cauchy sequence in X converges.
One of the main objective of this section is to show that every Cauchy sequence in a metric
space X converges in some larger metric space Y containing X (isometrically) as a subspace.

Properties of complete metric spaces

Complete metric spaces behave like compact topological spaces in several respects. Of course,
there are also some major differences; for example, the real numbers are complete but not compact,
and the real numbers are also homeomorphic to the noncomplete subspace (−1, 1), say by the map
f : (−1, 1) → R defined using the formula
x
f (x) =
1 − |x|

but one can begin the analogies with the following result:
PROPOSITION 3. A compact metric space is complete.
Proof. Let { an } be a Cauchy sequence in X. By previous results we know this sequence has
a convergent subsequence { an(k) }. Let limk→∞ an(k) = L; we claim that limn→∞ an = L. Given
ε > 0 choose M1 so that k > M1 implies d(L, an(k) ) < ε/2, and choose M2 so that m, n > M2
implies d(am , an ) < ε/2. Take M to be the larger of M2 and M1 . Then the Triangle Inequality
implies that d(L, an ) < ε if n > M (we have used this sort of argument many times already; at
this point the reader should try to fill in the details as an exercise).
One also has the following useful result about closed subsets and complete metric spaces:
PROPOSITION 4. Let X be a metric space and let A ⊂ X. If A is complete in the subspace
metric, then A is closed in X. Conversely, if A is closed in X and X is complete, then A is complete
in the subspace metric.
Proof. Suppose first that A is complete with respect to the subspace metric and that { a n }
is a sequence in A with a limit L ∈ X. By a previous result the sequence is a Cauchy sequence,
and therefore it has a limit in A. Since the limit of a convergent sequence is unique, this limit in
A must be the point L.
Now suppose that A is closed in X where X is complete. If { a n } is a Cauchy sequence in
A then the completeness of X implies that the sequence has a limit L ∈ X. Since A is a closed
subspace, this means that L ∈ A, so that the Cauchy sequence in A has a limit in A. Therefore A
is complete.

48
 The following result on completeness and products reflects another similarity with compact
topological spaces.
PROPOSITION 5. If X and Y are complete metric spaces, then so is X × Y with respect to
each of the three product metrics.
Proof. Let dhqi be a product metric where q = 1, 2 or ∞. By construction each of the
projection maps

X × Y, dhqi −→ (X, dX ) , (Y, dY )
sends points whose distance in X × Y is ε to points in X and Y with the same property.
It follows that if { (xn , yn ) } is a Cauchy sequence in X × Y then { x n } and { yn } are Cauchy
sequences in X and Y respectively. By the completeness of X and Y each of these sequences has
a limit, and we shall call these limits x and y respectively. To complete the proof we need to show
that
lim (xn , yn ) = (x, y)
n→∞

with respect to each of the three product metrics.

The previously established inequalities

dh∞i ≤ dh2i ≤ dh1i

show that it suffices to prove the limit statement for d h1i . Given ε > 0, choose M so that n > M
implies that both dX (x, xn ) and dY (y, yn ) are less than ε/2; then the standard arguments (supply
them!) show that n > M implies
 
dh1i (x, y) , (xn , yn ) < ε

and this proves the statement(s) about limits.

Function spaces

The next result provides some important examples of complete metric spaces.
PROPOSITION 6. If X is a set then BF(X) is a complete metric space with respect to the
norm described previously; furthermore, if X is a topological space then BC(X) is a closed and
hence complete subset of BF(X).
Proof. We shall first prove that BF(X) is complete. If { f n } is a Cauchy sequence in BF(X),
then by definition we know that { fn (x) } is a Cauchy sequence of real numbers for each x ∈ X.
Since we know that R is complete, it follows that for each x ∈ X there is a real number f (x) to
which { fn (x) } converges. We need to prove two things. First, we need to show that f is bounded.
Next we have to show that limn→∞ fn = f in BF(X).
To verify that { fn (x) } is bounded, choose M so that m, n > M implies |f m − fn | < 1. Then
for all n and x we know that

|f (x)| ≤ max{ |fi (x)|, i < M ; |fM (x)| + 1 }

49
and therefore we also have

|f (x)| ≤ max{ |fi |, i < M ; |fM | + 1 }

so that the left hand side is bounded by the right hand side for all x ∈ X and therefore f ∈ BF(X).
To show that { fn } converges to f in BF(X), let x ∈ X be arbitrary, and given ε > 0 choose
M so that m, n > M implies |fm − fn | < ε/2. Since { fn (x) } converges to f (x), there is some
Kx > 0 such that m > Kx implies |fm (x) − f (x)| < ε/4. Therefore if n > M and m > M + K x we
have
3ε
|f (x) − fn (x)| ≤ |f (x) − fm (x)| + |fm (x) − fn (x)| <
4
which means that
3ε
|f − fn | = sup{ |f (x) − fn (x)| } ≤ < ε
4
and thus that limn→∞ fn = f in BF(X).
We next need to show that f is continuous if each f n is continuous. Given ε > 0 and x ∈ X
we shall find an open set U such that x ∈ U and for all y ∈ U we have |f (y) − f (x)| < ε.
First of all, choose M such that n > M implies |f n − f | < ε/3. Next choose an open set U
containing x such that y ∈ U implies |f M +1 (y) − fM +1 (x)| < ε/3. The Triangle Inequality for real
numbers then implies that

|f (y) − f (x)| ≤ |f (y) − fM +1 (y)| + |fM +1 (y) − fM +1 (x)| + |f (x) − fM +1 (x)| < ε

and therefore f is continuous at x; since x was arbitrary, this means that f ∈ BC(X).
The examples in the proposition are special cases of the following important mathematical
structure:
Definition. If (V, | · · · |) is a normed vector space, then V is said to be a Banach space if it is
complete with respect to the associated metric.

Intersections of nested closed sets

We had previously noted that compact metric spaces are characterized by the fact that families
of closed subspaces with the finite intersection property have nonempty intersections. An important
special case involves nested sequences of closed subsets { A n } that are nonempty and satisfy An+1 ⊂
An for all n. In this case compactness implies that the intersection ∩ n An is nonempty. There is
an analog of this property for compact metric spaces. Recall that the diameter of a (subset of a)
metric space is given by
diam (A) = sup d(u, v)
u,v∈A

if the set of distances has an upper bound and by +∞ if no such upper bound exists.
PROPOSITION 7. (Cantor’s Nested Intersection Property) Let X be a complete metric space,
and let { An } be a nested sequence of nonempty closed subsets of X such that lim n→∞ diam (An ) =
0. Then ∩n An consists of one point.
A proof that the intersection is nonempty is given in Lemma 48.3 on page 297 of Munkres.
Suppose that y and z lie in the intersection. Then y, z ∈ A n for all n, and therefore d(y, z) ≤

50
diam (An ) for all n. Since the right hand side goes to zero as n → ∞, it follows that the left hand
side is ≤ 0; since the left hand side is nonnegative by construction, it must be zero, so that y = z
and the intersection contains exactly one point.

Completions of metric spaces

We have already mentioned that one recurrent theme in theoretical mathematics is the desire
to see whether empty spaces in mathematical structures can be filled in some sense, giving the real
numbers as an example. In fact, one can view the real numbers as a system obtained from the
rational numbers by filling in gaps so that every Cauchy sequence converges. In particular, the
finite decimal fraction approximations to a real number form a set of rational numbers converging
to the given real number, and as such they are Cauchy sequences that usually do not converge to
rational points. Our objective here is to show that every metric space can be expanded to a larger
one in which every Cauchy sequence converges. Often this is done by a brute force construction
that starts with the set of all Cauchy sequences in the metric space (for example, see Munkres,
Exercise 9, page 271). We shall construct this completion by a method that takes just about the
same amount of work but also yields some illuminating insights of independent interest.
PROPOSITION 8. If (X, d) is a metric space, then there is a 1 − 1 isometry from X into a
Banach space.
Proof. Before starting the proof we recall that a distance function satisfies

|d(u, w) − d(v, w)| ≤ d(u, v)

for all u, v, w ∈ X. This follows from two applications of the Triangle Inequality.
Let BC(x) be the Banach space of bounded continuous functions on X, and choose some point
a ∈ X. Define a map ϕ : X → BC(X) by the formula

[ϕ(x)](y) = d(y, x) − d(y, a) .

The right hand side is a continuous function of y because

   
[ϕ(x)](y) − [ϕ(x)](z) = (d(y, x) − d(y, a)) − (d(z, x) − d(z, a))  ≤
   
d(y, x) − d(y, x) + d(z, a) − d(y, a) = 2 d(y, z)

for all y, z ∈ X, and it is a bounded function of y because

 
 
d(y, x) − d(y, a) ≤ d(x, a)

for all y ∈ X if a and x are held fixed.

The estimates of the previous paragraph also show that
 
ϕ(x) − ϕ(y) ≤ d(x, y)

and in fact equality holds because

   
   
[ϕ(x)](y) − [ϕ(y)](y) = (d(x, y) − d(a, y)) − (d(y, y) − d(y, a))  = d(x, y)

51
so that ϕ is an isometry (hence uniformly continuous).
Definition. If X is a metric space, then a completion of X is a pair (f, Y ) consisting of a complete
metric space and an isometry f : X → Y such that f [X] = Y .
The preceding result implies that completions exist because the closure of ϕ[X] in BC(X) is
a closed subset of a complete metric spaced and therefore is complete. We shall also prove that up
to an isometry there is only one way of completing a metric space. Here is a formal statement:
THEOREM 9. (Uniqueness of completions of metric spaces) Let X be a metric space, and let
(f, Y ) and (g, Z) be completions of X. Then there is a unique 1 − 1 onto isometry h : Y → Z such
that h o f = g.
This result is a consequence of the following more general statement:
THEOREM 10. Let X be a metric space, let (f, Y ) be a completion of X, let W be a complete
metric space, and let h : X → W be a uniformly continuous function. Then there is a unique
uniformly continuous function H : Y → W such that H o f = h. Furthermore, if h is an isometry
then so is H.
Proof. The basic idea of the proof is simple. Since f [X] = Y it follows that for each y ∈ Y there
is a sequence of points { xn } in X such that limn→∞ f (xn ) = y. The only way we can extend h is
if we take H(y) = limn→∞ h(xn ). We need to show this actually works. The first step is to verify
that the definition of H(y) makes sense (in particular, the sequence { h(x n ) } actually converges)
and does not depend upon the choice of sequence in f [X] converging to y. Next, we have to show
that the function is uniformly continuous. Finally we have to show that H is an isometry if h is.
FIRST STEP. How do we know that the sequence converges? The hypotheses on h and W
suggest a couple of ideas. Since W is complete the sequence { h(x n ) } will converge if it is a Cauchy
sequence, and thus one might hope that the uniform continuity of h and the convergence of the
sequence { f (xn ) } imply the convergence of { h(xn ) }. Thus it suffices to show that a uniformly
continuous map of metric spaces from X to Y takes Cauchy sequences in X to Cauchy sequences
in Y . To see this, let h : Y → W be a uniformly continuous map of metric spaces, and assume
that { yn } is a Cauchy sequence in Y . Given ε > 0 there is a δ > 0 such that d(u, v) < δ implies
d(h(u), h(v)) < ε. Since we have a Cauchy sequence it follows that there is an M such that
m, n > M implies d(ym , yn ) < δ. It follows that m, n > M implies d(h(y m ), h(yn )) < ε. — This
implies that there is some point w0 ∈ W such that w0 = limn→∞ h(xn ).
SECOND STEP. If we have two sequences { u n } and { vn } in X such that

lim f (un ) = y = lim f (vn )

n→∞ n→∞

we need to show that

lim h(un ) = lim h(vn )
n→∞ n→∞

in order to know that H is well-defined. Denote the limits of these two sequences by α and β
respectively; it will suffice to show that d(α, β) < ε for all ε > 0.
Given ε > 0 one can find a positive integer M such that for all n > M we have d( h(u n ), α ) <
ε/3 and d( h(vn ), β ) < ε/3. Now choose δ > 0 so that d(s, t) < δ implies d( h(s), h(t) ) < ε/3, and
choose P so that p ≥ P implies d( f (up ), y ) < δ/2 and d( f (vp ), y ) < δ/2; the latter imply that
d( f (up ), f (vp ) ) < δ, so that d( f (up ), f (vp ) ) < ε/3. If we choose q > M + P then we have

d(α, β) ≤ d( h(uq ), α ) + d( h(uq ), h(vq ) ) + d( h(vq ), β )

52
and since q > M the right hand side is less than d( h(u q ), h(vq ) ) + 2ε/3. Finally, since q > P we
also know that d( h(uq ), h(vq ) ) < ε/3, and therefore d(α, β) < ε as required.
THIRD STEP. We need to show that the function H is uniformly continuous. Since h is
uniformly continuous, for each ε > 0 there is a δ > 0 such that d(u, v) < δ implies d( h(u), h(v) ) <
ε/3. Given a, b ∈ Y , let { un } and { vn } be sequences in X such that limn→∞ f (un ) = a and
limn→∞ f (vn ) = b, and suppose that d(a, b) < δ. Choose M so large that n > M implies

δ − d(a, b)
d(un , a), d(vn , b) <
2
and also
ε
d ( h(un ), H(a) ) , d ( h(vn ), H(b) ) < .
3
We then have that d(un , vn ) < δ and that
ε
d ( H(a), H(b) ) ≤ d ( h(un ), H(a) ) + d( h(un ), h(vn ) ) + d ( h(vn ), H(b) ) < 3 · = ε
3
which shows that H is uniformly continuous.
FOURTH STEP. We need to show that H is an isometry if h is an isometry. This will follow
from a more general fact:
Suppose that { un } and { vn } are convergent sequences in a metric space E with lim n→∞ un =
u and limn→∞ vn = v. Then limn→∞ d(un , vn ) = d(u, v).
Using this we may complete the fourth step as follows: Express u, v ∈ Y as limits of sequences
{ f (un ) } and { f (un ) } respectively. Then the assertion implies that

d(u, v) = lim d( f (un ), f (vn ) ) = lim d(un , vn )

n→∞ n→∞

where the last equation holds because f is an isometry. On the other hand we also have

d( H(u), H(v) ) = lim d( h(un ), h(vn ) ) = lim d(un , vn )

n→∞ n→∞

because h is also an isometry. Since limits of sequences are unique it follows that d(u, v) =
d( H(u), H(v) ) and therefore H is an isometry.
We must now verify the general assertion about limits of distances. Consider the inequality
     
d(un , vn ) − d(u, v) ≤ d(un , vn ) − d(u, vn ) + d(u, vn ) − d(u, v) ≤ d(un , u) + d(vn , v) .

If we choose M so that n > M implies d(u n , u), d(vn , v) < ε/2 the inequalities imply that
 
d(un , vn ) − d(u, v) < ε

which proves the assertion about limits.

Proof of the Uniqueness Statement. Let X be a metric space, and let (f, Y ) and (g, Z) be
completions of X. By the preceding theorem there are unique isometries G : Y → Z and F : Z → X
such that G o f = g and F o g = f . These in turn imply that F o G o f = f and G o F o g = g for the
maps F G : Y → Y and GF : Z → Z. Since idY and idZ also satisfy idY o f = f and idZ o g = g
it follows that G o F = idZ and F o G = idY , showing that the isometries F and G are 1–1 and

53
onto. This completes the proof that any two completions of a given metric space are isometric by
an isometry compatible with the isometric inclusions of X.

III.3 : Implications of completeness

(Munkres, § 48)

There are two fundamentally important properties of complete metric spaces that arise in
numerous analytic and geometric contexts. One of these (Baire’s Theorem) can be viewed informally
as saying that a complete metric space cannot be decomposed into “thin” pieces. The other (the
Banach Contraction Lemma) is a powerful method for finding solutions to various sorts of equations
in a wide range of contexts. We shall give both topological and analytic examples in this course.

Nowhere dense and meager subspaces

Definition. A subset A of a topological space X is said to be nowhere dense in X if Int X (A) = ∅.

The proofs of the following results are left to the reader as exercises.
PROPOSITION 1. If A is a subset of X then A is nowhere dense in X if and only if X − A is
dense in X.
PROPOSITION 2. Suppose that A ⊂ B ⊂ C ⊂ X and that B is nowhere dense in C. Then A
is nowhere dense in C and B is nowhere dense in X.
PROPOSITION 3. Suppose that A and B are nowhere dense subsets of X. Then A ∪ B is
nowhere dense in X.
It is particularly useful to understand when a one point subset of X is nowhere dense. Since
one point subsets are closed, a one point subset { x } is nowhere dense in X if and only if it is not
open; i.e., the point x is not isolated in X. The file
http://math.ucr.edu/∼res/math145A-2014/zariski-topology.pdf
proves that a large class of closed subsets in R n are nowhere dense; namely, if p(t1 , · · · , tn ) is a
nonzero polynomial in n variables (or indeterminates), then the set V (p) of all points a ∈ R n such
that p(a1 , · · · , an ) = 0 is a (closed and ) nowhere dense subset of R n .
Definition. A subset B ⊂ X is said to be meager or of the first (Baire) category in X if it can be
written as a countable union B = ∪n An where each An is closed and nowhere dense. A subset C
is said to be nonmeager or of the second (Baire) category in X if it is not meager, and in this case
the complement X − C is said to be residual or co-meager.
Examples.
1. A closed nowhere dense subset is always meager.
2. The rationals determine a meager subset of the real numbers, but this subset is definitely
not nowhere dense because the rationals are dense in the reals).

54
 3. Here is a more complicated but still important example. Let R ∞ denote the inner product
space consisting of all sequences (x 1 , x2 , · · · ) such that xk = 0 for all but finitely many k, with
the inner product given by the convergent (in fact, finite) series
X
hx, yi = xj yj .
j

If Bn is the set of all points for which xj = 0 for all j > n, then Bn is closed and nowhere dense in
Bn+1 and hence Bn is also closed and nowhere dense in X (why?). Since R ∞ = ∪n Bn it follows
that R∞ is meager in itself.
Note. The concepts of first and second category were so named before category theory was
invented; there is absolutely no mathematically sustantive connection between the two uses
of the word “category.” In order to avoid confusion many authors have made conscious efforts to
avoid terms like first and second category (compare the comments at the bottom of page 295 of
Munkres), but this terminology is still very widely used by mathematicians and others. One way of
dealing with this ambiguity is to use the term Baire category when referring to concepts involving
first and second category as definedabove.

Baire spaces

Definition. A topological space X is said to be a Baire space if every open subset is nonmeager.
Equivalent definitions are given on pages 295–296 of Munkres (in particular, see Lemma 48.1 on
page 296).
The following result provides many examples of Baire spaces:
Theorem 4. (Baire Category Theorem) If X is a complete metric space, then X is a Baire space.
A proof of this result for X itself is given in Theorem 48.2 on page 296 of Munkres (note that
this proof uses Lemma 48.3 on the following page).
The Baire Category Theorem has many extremely important and useful consequences. We
shall begin with one that only involves point set theory.
PROPOSITION 5. If X is a countable compact metric space, then X has at least one isolated
point.
Proof. Let X = { x1 , x2 , ... }, and for each i let Fi = { xi }. If X has no isolated points, then each
Fi is closed and nowhere dense, and therefore X is meager in itself. By Baire’s Category Theorem
this is impossible.
COROLLARY 6. If X is a compact metric space such that every point of X is a limit point,
then X is uncountable.
This is essentially the contrapositive of the theorem.
On pages 41–42 of Rudin’s Principles of Mathematical Analysis, this corollary is applied to
show that the Cantor set, which was mentioned in Section I.4, is uncountable. One can also prove
that the Cantor set has the same cardinality as the real numbers by a different argument (see
Problems 37–38 on page 46 of Royden, Real Analysis, Third Edition, as well as Exercise 6 on page
179 of Munkres).

55
 In Section 49 of Munkres the Baire Category Theorem is used to prove the existence of con-
tinuous functions on the unit interval that are not differentiable at any point of the interval. The
theorem also plays a crucial role in the foundations of the theory of Banach spaces and functional
analysis; in particular, it is the key ingredient of the proofs of the Uniform Boundedness Principle
and Open Mapping Theorem (e.g., see Rudin, Functional Analysis, for more information on this).

The Contraction Lemma

The following result is an extremely useful tool for finding solutions to a very wide range of
nonlinear equations.
THEOREM 7. (The Contraction Lemma) Let X be a complete metric space, and let T : X → X
be a map such that d(T (x), T (y)) ≤ α · d(x, y) for some fixed α ∈ (0, 1) and all x, y ∈ X (in
particular, T is uniformly continuous). Then there is a unique z ∈ X such that T (z) = z (in other
words, a unique fixed point for T ).
To see the need for completeness, consider the open interval (0, 1) and let T be multiplication
1
by 2.

Proof. The idea is beautifully simple. One starts with an arbitrary point x ∈ X and considers
the sequence of points x, T (x), T 2 (x), · · · . This sequence is shown to be a Cauchy sequence, and
the limit z of this sequence turns out to be the unique fixed point.
More formally, we begin by noting that T has at most one fixed point. If z, w ∈ X satisfy
T (z) = z and T (w) = w, then we have


0 ≤ d(z, w) = d T (z), T (w) ≤ α · d(z, w)

and since 0 < α < 1 this can only happen if d(z, w) = 0; i.e., if z = w.
We now follow the idea described in the first paragraph of the proof. By induction on n we
have


d T n (x), T n+1 (x) ≤ αn · d x, T (x)
and therefore by the Triangle Inequality for m > n we also have
m
X
n m



d T (x), T (x) ≤ αi d x, T (x) =
i=n+1

αn+1 (1 − αm−n )
αn+1

· d x, T (x) ≤ · d x, T (x)
1−α 1−α
which implies that the sequence { T n (x) } is a Cauchy sequence. By the completeness of X there
is a point z such that z = limn→∞ T n (x).
By Theorem 23.1 on page 130 of Munkres we have


T (z) = lim T T n (x) = lim T n+1 (x)
n→∞ n→∞

and by a change of variable (specifically, take k = n+1) the right hand side is equal to lim k→∞ T k (x),
which by construction is z. Therefore we have T (z) = z.

56
 Applications of the Contraction Lemma to solving functional equations are discussed in an
addendum to this section (cubicroots.pdf in the course directory). For the time being we
shall merely use the Contraction Lemma to prove the basic existence and uniqueness theorem for
solutions of first order differential equations in one variable.
THEOREM 8. (Picard Successive Approximation Method for the solutions of differential
equations) Let F (x, y) be a real valued function of two variables on an open set U such that F has
continuous partial derivatives on U . Then for each (a, b) ∈ U there is a positive real number δ > 0
such that there is a unique solution of the differential equation

dy
= F (x, y)
dx
on the interval (a − δ, a + δ) satisfying the initial condition y(a) = b.
Proof. To motivate the proof, note first that a function f is a solution of the differential equation
with the given initial value condition if and only if
Z x
f (x) = b + F (t, f (t)) dt
a

where as usual the integral is zero if x = a, while if x < a the integral from a to x is defined to be
the negative of the integral from x to a.
The idea of the proof is to use the right hand side to define a map of bounded continuous
functions and then to apply the Contraction Lemma. However, one needs to be a bit careful in
order to specify exactly which sorts of functions form the space upon which the mapping is defined
and in order to ensure that the map has the contraction property.
Choose h, k > 0 so that

S = [a − h, a + h] × [b − k, b + k] ⊂ U

so that F and its (first) partial derivatives are bounded on S. Let L be an upper bound for F . By
the Mean Value Theorem we have that
  
   ∂F 
F (x, y1 ) − F (x, y2 ) ≤ max(u,v)∈S  (u, v)  · |y1 − y2 |
 ∂y 

for all x, y1 , y2 . Let A be the maximum value of the absolute value of the second partial derivative
on S, so that we have  
F (x, y1 ) − F (x, y2 ) ≤ A · |y1 − y2 |

for all x, y1 and y2 . Inequalities of this sort are generally said to be Lipschitz conditions.
Choose δ > 0 so that δ ≤ h, L δ < k and A δ < 1. Define M to be the metric space of all
bounded continuous functions g on (a − δ, a + δ) for which |g − b| ≤ L δ, where as usual we identify
a real number with the constant function whose value is that number.
For every metric space Z, every z ∈ Z and every positive real number B, the set of points w
with d(z, w) ≤ B is closed (why?), and therefore M is a complete metric space. We need to show
that the map Z x
[T (g)](x) = b + F (t, g(t)) dt
a

57
is defined for all g ∈ X, it maps


X ⊂ BC (a − δ, a + δ)

into itself, and it satisfies the hypothesis of the Contraction Lemma on M .

First of all, it follows immediately that T (g) is continuous whenever g is continuous (fill in the
details here). Next, by the boundedness of F on the closed solid rectangle S we have
Z x  Z x 
   
|T (g) − b| =  F (t, g(t)) dt ≤ L ·  dt ≤ L δ
a a

so that g ∈ M implies T (g) ∈ M .

Finally, let g1 , g2 ∈ X and consider |T (g1 ) − T (g2 )|. By definition the latter is equal to the
least upper bound of the numbers
Z x 
 
 (F (t, g1 (t)) − (t, g2 (t))) dt ≤

a
Z x 
F (t, g1 (t)) − (t, g2 (t)) dt ≤ A δ · |g1 − g2 | .
a

Since A δ < 1, all the hypotheses of the Contraction Lemma apply so that there is a unique fixed
point, and as noted above this unique fixed point must be the (necessarily unique) solution of the
original differential equation with the prescribed boundary condition.
Note. The theorem and its proof remain valid if we weaken the assumption on F and only
assume that it satisfies the Lipschitz condition
 
F (x, y1 ) − F (x, y2 ) ≤ A · |y1 − y2 |

which is derived in the course of the argument. In fact, one can even prove an existence theorem
with the much weaker hypothesis that F is continuous (compare Exercise 25 on pages 170–171
of Rudin’s book), but uniqueness does not follow. For example, if F (x, y) = y 1/2 , then the zero
function and x2 /4 are both solutions to the differential equation with initial condition y(0) = 0.

III.4 : Connected spaces

(Munkres, §§ 23 – 25)

One of the most basic results in single variable calculus is the Intermediate Value Theorem,
which states that if a real valued function f is continuous on an interval J ⊂ R and a, b ∈ J are
points such that f (a) 6= f (b), then for each real number y between f (a) and f (b) there is a real
number x between a and b such that f (x) = y.
There are many other situations where one has such a conclusion, and it is useful to have a
systematic understanding of when one has such intermediate value results. This requires one to
find the abstract concept which underlies the Intermediate Value Theorem, and this notion is called
connectedness.

58
Definition. A separation of topological space X is a pair of disjoint closed proper subspaces A
and B whose union is X. A space is said to be connected if it does not have any separations. A
space is said to be disconnected if it is not connected.
Of course we want intervals in the real line to be connected, but before addressing this point
we give two equivalent formulations of the concept of separation.
PROPOSITION 1. If X is a topological space, the the following are equivalent:
(i) One can write X = A ∪ B where A and B are nonempty disjoint closed subsets.
(ii) One can write X = A ∪ B where A and B are nonempty disjoint open subsets.
(iii) There is a nonempty proper subset A ⊂ X that is both open and closed (sometimes one
says that such a subset is clopen).
Proof. (i) =⇒ (ii) By construction we have B = X − A and A = X − B, so that the subsets A
and B are also open in X.
(ii) =⇒ (iii) The set A is nonempty and it is a proper subset because B = X −A is nonempty.
By hypothesis A is closed, and since A = X − B we know that A is also open.
(iii) =⇒ (i) If B = X − A, then B is closed since A is open and B is open since A is closed.
By hypothesis, A is nonempty, and since it is a proper subset we also know that B is nonempty.
The conditions A ∩ B = ∅ and A ∪ B = X follow immediately from the definition of B.

Connectedness and the real line

There are some immediate examples of spaces that are connected and spaces that are not
connected. Every space with at most one point is connected because there are no nonempty proper
subsets, and hence there are no subsets for which condition (iii) is meaningful. Every set with
an indiscrete topology is connected because there are no nonempty proper subsets that are either
open or closed. On the other hand, if a set S has at least two elements and is given the discrete
topology, then every subset of S is open (hence by complementation every subset is also closed!),
and therefore every nonempty proper subset of S will be open and closed.
In particular, the preceding discussion implies that a topological space with the discrete topol-
ogy is connected if and only if it contains at most one element.
Without further discussion we proceed to the single most important family of examples of
connected sets.
THEOREM 3. Let A be a subset of R with at least two elements. Then A is connected if and
only if for each a, b ∈ A such that a < b the entire interval [a, b] is contained in A.
Proof. ( =⇒ ) Suppose that the conclusion is false, so that there is some c satisfying a < c < b
and c 6∈ A. Let
B = A ∩ (c, ∞) = A ∩ [c, ∞)
where the second equality holds because c 6∈ A. We know that b ∈ B so that the latter is nonempty,
and the two descriptions of B in the displayed formula show it is both open and closed in R. Finally,
since a ∈ A − B we know that B is a proper subset, and therefore A is not connected. We have
thus shown the contrapositive of what we wanted to prove (and this proves the latter).

59
 ( ⇐= ) Suppose that C is a nonempty open and closed subset of A. Without loss of generality
we may as well assume that a ∈ C; if this is false, then a lies in the nonempty open closed subset
A − C and we can go through the same argument reversing the roles of C and A − C at each point.
By hypothesis we know that [a, b] ⊂ A. Since C is open in A there is a δ > 0 such that

[a, a + δ) ⊂ (a − δ, a + δ) ∩ C ∩ [a, b] ⊂ C

and thus the set 

K = { y ∈ (a, b]  [a, y] ⊂ C }
is nonempty with an upper bound (specifically, b), Let y ∗ be the least upper bound of K; we claim
that y ∗ = b. In order to do this we need to show that y ∗ < b is impossible.
By the definition of least upper bound, for each n > 0 there is some y n ∈ K such that

1
y∗ − < yn ≤ y ∗
n
and for this sequence we have y ∗ = limn→∞ yn . Since each of the points of the sequence lies in C
and the latter is closed, it follows that y ∗ ∈ K. Suppose now that y ∗ < b. On the other hand, since
C is open there is some η > 0 such that y ∗ + η < b and

(y ∗ − η, y ∗ + η) ⊂ C .

The latter in turn implies h ηi

a, y ∗ +
⊂ C
2
which means that y ∗ is NOT an upper bound for K. This contradiction forces the conclusion y ∗ = b.
The preceding argument shows that if C is a nonempty open and closed subset of A containing
a point a and b is another point in A such that a < b, then b ∈ C. — To prove that C = A
and hence that A is connected, it suffices to verify that the corresponding statement holds for all
b ∈ A such that b < a. The only way this could fail would be if b ∈ A − C. But in this case our
argument would imply that a ∈ A − C, which is false. Therefore we have shown that a nonempty
open and closed subset of A must be all of A if the latter has the intermediate point property
described in the statement of the theorem.
COROLLARY 4. The connected subsets of R are all given by the following list:
(1) Closed intervals [a, b] where a, b ∈ R.
(2) Open intervals (a, b) where a ∈ R ∪ {−∞} and b ∈ R ∪ {+∞}.
(3) Half open intervals [a, b) or (a, b] where a ∈ R and b ∈ R ∪ {+∞} in the first case and
a ∈ R ∪ {−∞} and b ∈ R in the second.
It follows that the cardinality of the set of all connected subspaces of R is equal to the car-
dinality of R itself. The verifications of this statement and the corollary are left to the reader as
exercises.
The proof of the Intermediate Value Theorem is essentially a combination of the characteri-
zation of connected subsets of the real line and the following abstract result, which establishes the
conclusion of the Intermediate Value Theorem for arbitrary continuous real valued functions on
connected topological spaces:

60
PROPOSITION 5. If f : X → Y is continuous and X is connected, then f [X] is also connected.
Proof. Let C ⊂ f [X] be a nonempty subset that is both open and closed. Then C = U ∩ f [X]
where U is open in Y and C = E ∩ f [X] where E is closed in Y . We then have that
 
f −1 [C] = f −1 U ∩ f (X) = f −1 [U ]

is open in X and  
f −1 [C] = f −1 E ∩ f (X) = f −1 [E]
is closed in X. Since C is a nonempty subset of f [X] it follows that f −1 [C] is nonempty and
therefore by the connectedness of X we have f −1 [C] = X. In particular, for all x ∈ X this means
that f (x) ∈ C, which in turn means that C ⊃ f [X]; by assumption the reverse inequality holds so
that C = f [X]. Therefore we have shown that the only nonempty subset of f [X] that is open and
closed is f [X] itself, which means that f [X] is connected.
For the sake of completeness we shall state and prove the following result, whic provides a
simple criterion for showing that certain real valued functions of a real variable have continuous
inverses (among other things, this result is applied in homeomorphisms.pdf).
THEOREM 6. (An Inverse Function Theorem) Let a and b be real numbers or ± ∞ such that
a < b, and let f : (a, b) → R be a continuous function which is strictly increasing. Then the image
of f is an open interval (c, d), where c and d are real numbers or ± ∞, and there is an inverse
function g : (c, d) → R such that g(f (x)) = x for all x ∈ (a, b) and f (g(y)) = y for all y ∈ (c, d); in
other words, x = g(y) if and only if y = f (x).
There is a similar result for continuous functions which are strictly decreasing, and it is an
immediate consequence of the theorem by the following argument:
If we are given a continuous function f : (a, b) → R which is strictly decreasing, then
F = −f is increasing, so by the theorem it has an inverse G, and we obtain the inverse g
to f by setting g equal to −G.
Proof of Theorem 6. Since (a, b) is a connected subset of R, its image must also be a connected
subset of R, which means that the image is some interval. In fact, the image must be an open
interval, for (a, b) has no minimal or maximal element, and since f is strictly increasing the same
must be true for its image. Therefore the image must be an interval of the form (c, d), where c and
d are real numbers or ± ∞.
Since a strictly increasing function is 1–1 (because x < x 0 implies f (x) < f (x0 ) ), we know that
there is some set-theoretic inverse function g : (c, d) → (a, b). This inverse is also strictly increasing,
for if y < y 0 and we have y = f (x) and y 0 = f (x0 ), then we must have g(y) = x < x0 = g(y 0 ); if the
latter did not hold, then we would have xz ≥ x 0 , which would imply y ≥ y 0 because f is strictly
increasing. To complete the proof of the theorem, we need to verify that g is continuous.
The continuity of g will follow if we can find a base B for the topology of (a, b) such that for
each basic open set V ∈ B the set g −1 [V ] is open in (c, d). We shall take the base consisting of all
open intervals (u, v) ⊂ (a, b), where a < u < v < b, and we claim that g −1 [ (u, v) ] = ( f (u), f (v) ).
Since g(y) ∈ (u, v) implies that y = f o g(y) ∈ ( f (u), f (v) ) because f is strictly increasing, it
follows immediately that g −1 [ (u, v) ] ⊂ ( f (u), f (v) ), so at this point we only need to prove that
the reverse inclusion also holds. But if f (u) < y < f (v) then we can use the strictly increasing
behavior of the inverse function g to conclude that u = g o f (u) < g(y) < g o f (v) = v which yields
g(y) ∈ (u, v) and hence implies that g −1 [ (u, v) ] ⊃ ( f (u), f (v) ). By the previous remarks, this
concludes the proof that g is continuous.

61
 We also have the following generalization of Theorem 6:
COMPLEMENT 6A. If J ⊂ R is a connected subset and f : J → R is continuous and strictly
increasing, then f maps J homeomorphicaly onto an interval of the same type (open, closed, half-
open with a left hand endpoint or half-open with a right hand endpoint).
Proof. By Theorem 6 we need only consider intervals which are not open. One quick way
of generalizing Theorem 6 to such intervals is to extend the original function f to a larger open
interval so that the new function F is still continuous and strictly increasing. Specifically, we can
construct the extension F as follows: Let J be an interval of the form [a, b), (a, b] or [a, b], where
we allow a = −∞ or b = +∞ if a or b (respectively) is not contained in J.
(1) If a ∈ J, extend f to (a − 1, b) by the linear formula F (x) = x − a + f (a) on (a − 1, a].
The combination of these maps is continuous because F (a) = f (a).
(2) If b ∈ J, extend f to (a, b + 1) by the linear formula F (x) = x − b + f (b) on [b, b + 1). The
combination of these maps is continuous because F (b) = f (b).
(3) If a, b ∈ J, extend f to (a − 1, b + 1) as in (1) and (2). For the same reasons as above, the
combination of the maps defined on the subintervals (a − 1, a], [a, b] and [b, b + 1) will be
continuous.
In all cases the extended function F has a continuous, strictly increasing inverse by the theorem,
and we shall denote this continuous, strictly increasing inverse by G. We can then retrieve the
inverse g to f by taking the restriction of G to (f (a), f (b)], [f (a), f (b)) and [f (a), f (b)] in the
respective three cases, and in each instance the continuity of g follows from the continuity of G.

As before, there are also similar results for continuous and strictly decreasing functions
defined on closed or half-open intervals.
The file inverse-fcn-thms.pdf discusses some additional results known as Inverse Function
Theorems (see also Theorem 1.8 in this unit).

Finding connected (sub)sets

The Intermediate Value Theorem for connected spaces is a very powerful statement on the
existence of solutions to equations of the form y = f (x), and therefore it is important to recognize
when it applies, particularly to subspaces of the plane or other Euclidean spaces. One expects
that the theorem will apply to open and closed rectangles that are products of two open or closed
intervals respectively. These facts will follow quickly from our abstract discussion below.
PROPOSITION 7. Suppose that X is a topological space, that C is an open and closed subset
of X, and that A is a connected subset of X. Then either A ⊂ C or A ⊂ X − C.
Proof. The intersection A × C is an open and closed subset of A, so by connectedness of the
latter it must either be all of A or empty. In the first case we have either A ⊂ C, and in the second
we have A ⊂ X − C.
PROPOSITION 8. Suppose that X is a topological space and that A and B are connected
subsets of X with a nonempty intersection. Then A ∪ B is connected.
Proof. Suppose that C is a nonempty open and closed subset of A∪B. Without loss of generality
we may as well assume that some point x 0 ∈ A ⊂ B belongs to C; if instead we have x 0 ∈ A∪B −C
then we can switch the roles of C and A ∪ B − C in the proof of the first case.

62
 Since A is a connected subset of A ∪ B we must have A ⊂ C or A ⊂ A ∪ B − C. Since x 0 ∈ C
the latter is impossible, and hence A ⊂ C. If we interchange the roles of A and B in this argument
we also conclude that B ⊂ C, so that

C ⊂ A∪B ⊂ C

which implies that the two sets are equal and hence that A ∪ B must be connected.
Remark. If A and B are connected subsets of a topological space X it does NOT follow that
A ∩ B is connected. Here is a counterexample when X = R 2 :
Let A and B be the semicircles in the unit circle (with equation x 2 + y 2 = 1) whose second
coordinates are positive and negative respectively. Each subset is connected because the semicircles
are the continuous images of the interval [−1, 1] under the continuous mappings
p

γ± (t) = t, ± 1 − t2

but the intersection is the pair of points with coordinates (±1, 0) and this set is not connected (it
has the discrete topology).
Definition. Given a topological space X and a, b ∈ X, define a binary relation ∼ [CONN] by
a ∼[CONN] b (for a, b ∈ X) if and only if there is a connected subset of X that contains both a and
b.
ELEMENTARY FACT 9. The preceding binary relation is an equivalence relation, and its
equivalence classes are called the (connected) components of X.
PROPOSITION 10. The connected components of X are maximal connected subsets of X.
Proof. Let A be a connected component of X, and let C be a nonempty open and closed subset
of A. Take a to be a point in C ∩ A.
If y ∈ A, then by definition there is a connected subset A y of X that contains both a and y.
The definition of the equivalence relation implies that A y ⊂ A. By an earlier result we know that
either Ay ⊂ C or Ay ⊂ A − C. The latter is impossible because a ∈ A y ∩ C, and therefore Ay ⊂ C
for all y ∈ A. Since y ∈ Ay for all y, this means that A ⊂ C so that C = A and therefore A must
be connected.
To verify maximality, suppose that B is a connected set such that B ⊃ A. Then for each b ∈ B
the set B itself is a connected subset of X that contains a and b, and therefore all points of B are
in the same component as a, which is merely A. Therefore B ⊂ A and A is maximal.
In a discrete space the connected components are just the one point subsets and as such they
are open and closed. One can ask whether connected components in arbitrary spaces have similar
properties. It turns out that these subsets are always closed but not necessarily open. The first of
these will be an immediate consequence of the following result.
PROPOSITION 11. If X is a topological space and A ⊂ X is nonempty and connected, then
its closure A is also connected.
Since a component is a maximal connected subset, the preceding result shows that a component
must be equal to its own closure and therefore must be closed.
Proof of Proposition 11. Let C be a nonempty open and closed subset of A. It follows that
there is some point y ∈ A and since C is open it also follows that either A ∩ C 6= ∅. Since A is
connected it follows that A ⊂ C, and since C is closed in A it also follows that A ⊂ C.

63
COROLLARY 12. In the notation of the previous proposition, if B ⊂ X satisfies A ⊂ B ⊂ A,
then B is connected.
Proof. The proposition implies that Closure (A, B) is connected, and the latter is just the set
A ∩ B = B.
Example. There are many examples to show that connected components are not necessarily
open subsets. In particular, the rational numbers with the subspace topology inherited from the
Q
real numbers have this property. — Let a ∈ Q and consider the open sets N √2/2n (a). We claim
these sets are both open and closed in Q; openness is immediate,
√ and they are closed because a
rational number b lies in such a set if and only if |b − a| ≤ 2/2n because
√
2
a ±
2n
is never a rational number. Therefore it follows that if A is the connected component of a in Q
Q
we must have that A ⊂ N√2/2n (a) for all n > 0. But this forces A to be equal to {a}, which is
definitely not open in Q (each ε-neighborhood contains infinitely may other rational numbers).

Products and connectedness

The next result provides an important tool for recognizing connected subset in Euclidean
spaces.
THEOREM 13. If X and Y are connected spaces then so is their product X × Y .
Proof. The result is trivial if either X or Y is empty, so assume that both are nonempty. Let
(x0 , y0 ) ∈ X ×Y , and let C be the connected component of (x 0 , y0 ). We shall show that C = X ×Y .
For each (u, v) ∈ X × Y we have slice homeomorphisms from X and Y to X × {v} and
{u} × Y respectively, and therefore the latter subspaces are all connected. This means that for each
(x, y) ∈ X × Y the points (x, y) and (x0 , y) lie in the same connected component, and similarly
the points (x0 , y) and (x0 , y0 ) also lie in the same connected component. By the transitivity of the
relation of belonging to the same component it follows that (x, y) and (x 0 , y0 ) also lie in the same
connected component, and hence that C = X × Y .
COMPLEMENT 14. The same conclusion holds for arbitrary finite products.
In fact, an arbitrary product of connected spaces is connected. An outline of the proof appears
in Exercise 10 on page 152 of Munkres.
Proof. This follows from the theorem by induction and the canonical homeomorphism
n+1
Y


X1 × · · · × Xn × Xn+1 ∼
= Xi .
i=1

n
COROLLARY 15. For each positive integer n the space Q R is connected, and for each sequence
of closed intervals [ai , bi ] (where 1 ≤ i ≤ n) the product i [ai , bi ] is connected.
The results proven thus far have the following noteworthy consequence: The cardinality of
the set of connected subsets of R2 is the same as the cardinality of the set of all subsets of R 2 (or

64
equivalently the cardinality of the set of all subsets of R). — To see this, begin by noting that the
open rectangular region (0, 1)2 is connected by the theorem as is the closed rectangle [0, 1] 2 , and
the closed rectangle is the closure of the open rectangle (it is easy to find infinite sequences in the
open rectangle that converge to an arbitrary point of the closed rectangle; alternatively, one can
use the general rule
A×B = A×B
to show this). Given a subset S of (0, 1) ∼
= R. consider the set

CS = [0, 1) × [0, 1] ∪ {1} × S .

By the previous results on closures of connected sets it follows that each set C S is connected and
by construction CS 6= CT if S 6= T . Therefore there are at least as many connected subsets of R 2
as we have claimed. On the other hand, there are at most as many of these as there are subsets of
R2 , and therefore by the Schröder-Bernstein Theorem it follows that the cardinalities are the same.
The preceding yields an example for an assertion made earlier: R 2 is not homeomorphic to a
subset of R. If A is an arbitrary subset of R, the characterization of connected subsets of the real
line shows that the cardinality of the set of connected subsets of A is at most the cardinality of the
real numbers themselves, but we know that the cardinality of the set of connected subsets of R 2 is
greater than this.

Distinguishing homeomorphism types

Connectedness provides an effective means for showing that certain pairs of spaces are not
homeomorphic to each other.
PROPOSITION 16. No two sets in the following list are homeomorphic:
(i) The closed unit interval [0, 1].
(ii) The open unit interval (0, 1).
(iii) The half-open unit interval (0, 1].
(iv) The circle S 1 ⊂ R2 defined by the equation x2 + y 2 = 1.
Proof. The first and last sets are compact while the second and third are not, so it suffices to
show that S 1 is not homeomorphic to [0, 1] and (0, 1) is not homeomorphic to (0, 1] (Would the
result remain true if we added the half-open interval [0, 1) to the list? Why or why not?).
Our reasoning relies on the following observation: If f : X → Y is a homeomorphism and
A is a finite subset, then f |X − A maps X − A homeomorphically to Y − f (A); in particular,
X − A is connected if and only if Y − f (A) is connected. If one removes two points from S 1
the resulting space is disconnected (supply the details!), but if one removes the endpoints of [0, 1]
the resulting space is still connected. Therefore the observation shows that S 1 and [0, 1] cannot be
homeomorphic.
Similarly, if one removes the endpoint from [0, 1) then the resulting space is connected, but if
one removes any point from (0, 1) the resulting space is disconnected.
Further discussion along these lines yields complete topological characterizations of 1-dimen-
sional objects like the unit circle or a closed interval. Textbook discussions of this appear in books

65
by Hocking and Young (Topology, Section 2–5 on pages 52–55, with background material in the
preceding section) and Christensen and Voxman (Aspects of Topology (First Edition), Section 9.A
on pages 227–232 with accompanying exercises on page 251, and closely related material in Section
5.A on pages 127–128).

III.5 : Variants of connectedness

(Munkres, §§ 23, 24, 25)

If U is an open subset of some Euclidean space, then U is connected if and only if each pair
of points in U can be joined by a broken line curve that lies entirely in U . This fact, which we
shall prove at the end of the present section, reflects two important refinements of the concept of
connectedness.

Locally connected spaces

Definition. A topological space X is said to be locally connected if for each x ∈ X and each open
set U ⊂ X such that x ∈ U there is a connected open set V such that x ∈ V ⊂ U .
Example. If U is open in some Euclidean space then U is locally connected. Suppose that
W ⊂ U is an open set and that x ∈ W . Choose δ > 0 so that N δ (x) ⊂ W . To see that Nδ (x) is
connected, given y in the latter consider the image J y of the curve

γ(t) = x + t(y − x)

for 0 ≤ t ≤ 1. The set Jy is a connected set containing x and y, and therefore y lies in the same
connected component of Nδ (x) as x, and since y was arbitrary this implies that the set in question
is connected.
If we take a disconnected open subset of some Euclidean space, we have an example of a
locally connected space that is not connected. Finding examples of spaces that are connected but
not locally connected takes more work. Before doing this we state a basic characterization of local
connectedness that is established in Theorem 25.3 on page 161 of Munkres.
PROPOSITION 1. A topological space X is locally connected if and only if for each open
subset U , the components of U are open.
Example. A connected space that is not locally connected. Let A be the graph of the
function f (x) = sin(1/x) for x > 0, and let B = A. It follows that B is the union of A with the
set {0} × [−1, 1] and that B is connected because A is connected. We claim that B is not locally
connected. Let W be the set of all points in B for which the second coordinate lines in (−1, 1); then
W is open and therefore it suffices to find a component of W that is not open. By construction
W does not contain any points whose first coordinates have the form 2/mπ where m is an odd
positive integer. Therefore if C is the connected component in W containing the connected set
{0} × (−1, 1), it follows that C cannot have any points whose first coordinates are ≥ 2/mπ for all
m and thus that the first components of points in C must be zero. In other words, we must have
C = {0} × (−1, 1). This set is not open. In particular, for δ > 0 the set N δ ([origin]) contains

66
 

points of W that are not in C — specifically, all points of the form 1/kπ, 0 where k is sufficiently
large.
The proof of the following result is left to the reader as an exercise:
PROPOSITION 2. A finite product of locally connected spaces is locally connected.
There is an analog for infinite products with a curious extra condition. Namely, an arbitrary
product of locally connected spaces is locally connected if and only if all but at most finitely
Q many
of the factors are also connected. To see the need for this condition, note that if x ∈ α Xα and
U is an open set containing x, then pα [U ] = Xα for all but finitely many α (filling in the details
and the proof of the original assertion are left to the reader as exercises).

Path or arcwise connectedness

The connectedness of the real interval leads to an important and useful criterion for recognizing
connected spaces.
Definition. A topological space X is said to be path connected or arcwise connected if for each
pair of points x, y ∈ X there is a continuous function (or curve or path) γ : [a, b] → X such that
γ(a) = x and γ(b) = y.
PROPOSITION 3. An arcwise connected space is connected.
Proof. Let x ∈ X, and let C be the connected component of x in X. Given y ∈ Y , let γ be as in
the definition, and let Jy be the image of γ. Then Jy is a connected set containing x and y, and
consequently we must have y ∈ C. Since y ∈ X was arbitrary, this means that X = C.
Example. The previous example B constructed from the graph of f (x) = sin(1/x) is an
example of a connected space that is not arcwise connected. — To see this, suppose that γ is a
continuous curve in B defined on a closed interval [a, b] that joins a point of the form (0, y 0 ) to a
point (x1 , y1 ) with x1 > 0. Since {0} × [−1, 1] is compact, it follows that the closed set
 
γ −1 {0} × [−1, 1] ⊂ [a, b)

is a closed subset of [a, b] and thus has a maximum point c < b. At least one of the open sets
   
V− = B ∩ R × (−1, 1] , V+ = B ∩ R × [−1, 1)

contains γ(c) depending upon whether the last coordinate is ±1 or neither so choose V ε to be such
an open subset. Then there is a δ > 0 such that δ < b − c and 0 ≤ δ implies that γ(c + t) ∈ V ε . It
follows that for all but finitely many positive integers n there are points t n ∈ (0, δ) such that the
first coordinate of γ(tn ) is equal to 2/nπ; specifically, pick any value t 0 ∈ (0, δ), so that the first
coordinate η0 of γ(t0 ) will be positive, and then notice that for all sufficiently large n we have

2
< t0 .
nπ
However, by construction V− does not have any points whose first coordinates have the form 2/nπ
where n is an integer of the form 4k + 3 (the values of x for which sin(1/x) = −1), and V + does not
have any points whose first coordinates have the form 2/nπ where n is an integer of the form 4k + 1
(the values of x for which sin(1/x) = +1). Therefore it is not possible to construct a continuous

67
curve in B joining a point with zero first coordinate to a point with positive first coordinate, and
therefore the connected set B is not arcwise connected.

Analogies with connectedness

There is a concept of path component or arc(wise) component that is analogous to the concept
of connected component.
Definition. Given a topological space X and x, y ∈ X, define a binary relation ∼ [ARC] by
x ∼[ARC] y (for a, b ∈ X) if and only if there is a continuous function γ : [a, b] → X such that
γ(a) = x and γ(b) = y. We often say that γ is a curve joining x and y if this condition holds.
It is obvious that this relation is reflexive (use the constant curve) and symmetric (consider the
curve δ defined on [−b, −a] with δ(t) = γ(−t). To see that the relation is transitive it is convenient
to introduce a concept that arises frequently in mathematics.
Definition. Let X be a topological space, and suppose that γ : [a, b] → X and δ : [c, d] → X are
continuous curves such that γ(b) = δ(c). The sum or concatenation (stringing together) γ + δ is
the continuous curve
σ : [a, b + d − c] −→ X
defined by σ(t) = γ(t) if t ∈ [a, b] and σ(t) = δ(t + b − c) if t ∈ [b, b + d − c]. An illustration of this
appears in the course directory in the file concat.pdf.
By construction the sum of curves is an associative operation whenever it is defined.
Transitivity of ∼[ARC] follows immediately from this definition, for if γ joins x and y and δ
joins y and z, then γ + δ joins x and z.
Definition. The equivalence classes of ∼ [ARC] are called path components or arc components of
X. It follows that every arc component is arcwise connected, but the preceding example shows
that arc components need not be closed or open in the ambient space X.
The statements of the next two results are parallel to those for connected spaces, but the proofs
are entirely different.
PROPOSITION 4. If X is an arcwise connected space and f : X → Y is continuous, then f [X]
is arcwise connected.
Proof. Given a, b ∈ f [X] write a = f (c) and b = f (d) for c, d ∈ X. Since X is arcwise connected
there is a continuous curve γ in X joining c to d, and the composite f o γ is a continuous curve in
f [X] joining a to b.
PROPOSITION 5. A (finite) product of arcwise connected spaces is arcwise connected.
In fact, the finiteness condition is completely unnecessary in the statement and proof of this
result.
Q
Proof. Let X1 , · · · , Xn be the spaces in question, let u, v ∈ i Xi and for each j between 1
and n let uj and vj be the j th coordinates of u and v respectively. Then for each j one can join
uj to vj by a continuous curve γj . Suppose that γj is defined on the closed interval [aj , bj ] and let
Lj : [0, 1] → [aj , bj ] be the unique linearQfunction that sends 0 to a j and 1 to bj . Then there is a
unique continuous function α : [0, 1] → i Xi whose projection onto the j th coordinate is γj o Lj
for each j, and by construction α joins a to b.

68
 There is also a corresponding notion of local path or arcwise connectedness (see Munkres, page
161), and as noted in Theorem 25.5 on that page (the proof continues to page 162), if a space is
locally arcwise connected then its components and path components are identical.

Strong arcwise connectedness

In some references the term “arcwise connected” is reserved for spaces which satisfy the fol-
lowing strengthening of the condition in these notes.
Definition. A topological space X is said to be strongly arcwise connected if given two disticnt
points x, y ∈ X there is a continuous curve γ : [0, 1] → X such that γ(0) = x, γ(1) = y, and γ
maps [0, 1] homeomorphically onto its image.
If X is an indiscrete space with more than one point, then X is arcwise connected in the sense
of these notes, but X is not strongly arcwise connected; this is true because every subspace of X is
also indiscrete and hence is not Hausdorff if it contains more than one point. On the other hand,
for Hausdorff spaces one can prove that a space is arcwise connected if and only if it is strongly
arcwise connected. One reference is Sections 27 and 31 of the book by Willard in the bibliography.

Open subsets of Euclidean spaces

The discussion of local connectedness for open sets in Euclidean spaces actually proves that
such sets are locally arcwise connected and hence their components, which are open, are the same
as their arc components. In particular, an open subset of a Euclidean space is connected if and
only if it is arcwise connected, and the discussion at the beginning of this section asserts that one
can choose the curves in question to be of a special type. In order to prove this we need to give a
formal definition of a broken line curve.
Definition. A closed line segment curve in R n is a continuous curve γ defined on [0, 1] by an
equation of the form
γ(t) = a + t · (b − a)
for some a, b ∈ Rn . A broken line curve is a finite iterated concatenation of closed line segment
curves.
PROPOSITION 6. Let U be open in Rn . Then U is connected if and only if every pair of
points in U can be joined by a broken line curve that lies entirely in U .
Proof. If the conclusion is true then U is arcwise connected (and hence connected). To prove the
(=⇒) implication, define a binary relation ∼ by u ∼ v if and only if there is a broken line curve in U
joining u to v. This is an equivalence relation, and in fact the equivalence classes are open subsets
(if x ∈ U and Nδ (x) ⊂ U , then every point in Nδ (x) can be joined to x by a closed line segment
curve). It follows that the union of any family of equivalence classes is also open, and in particular,
if W is an equivalence class this means that U − W , which is the union of all the equivalence classes
except W , is also open. The latter implies that W is closed, and since U is connected it follows that
there can be only one equivalence class for the equivalence relation described above; this proves
that each pair of points in U can be joined by a broken line curve in U .
Given an open connected subset of Rn one can ask many different questions about the contin-
uous curves joining two arbitrary points in U , including the following: Given two points in U , can

69
they be joined by a curve whose coordinate functions are infinitely differentiable? If so, can one
find such a function such that the tangent vector at every point is nonzero?
The answer to both questions is yes, but the proofs are more complicated than the ones given
above. Such results can be established using techniques from an introductory graduate course on
smooth manifolds (for example, Mathematics 205C).

Space-filling curves and related issues

As noted in the online file http://math.ucr.edu/∼res/math145A-2014/intro2topA-13.pdf

and Section 44 of Munkres, if γ is a an arbitrary parametrized curve γ defined on an interval J ⊂ R,
one must not assume that its image resembles the intuitive notion of a curve as a union of subsets
which are homeomorphic to intervals. There are some simple assumptions which guarantee the
image has this form (for example, if the curve takes values in R n and has nonzero tangent vectors
everywhere), but in general the image γ[J] can be quite unpredictable. In particular, the cited
document discusses examples where J = [0, 1] and the image of γ is the solid square [0, 1] × [0, 1].
The cited file also states a theorem which completely characterizes all possible images γ[J] up to
homeomorphism (the Hahn-Mazurkiewicz Theorem) when J is a closed interval.

70
 IV . Function spaces

An extremely important class of topological spaces is given by placing suitable topologies on

sets of continuous functions from one topological space to another. Such objects have important
uses in analysis and also in topology itself. We shall concentrate on examples where the spaces
in question are particularly easy to work with. There are two reasons for this. First of all, such
examples are totally adequate for many purposes in analysis and topology. Second, many of the
proofs simplify dramatically in the cases we shall consider. We shall discuss the generalizations
treated in Munkres and give references to the corresponding discussions in that source.

IV.1 : General properties

(Munkres, §§ 45–47)

If Z is a compact topological space, then we have already seen that the formula

d(x, y) = max t∈X |f (t) − g(t)|

defines a metric on the set C(X, R) of real valued continuous functions on X. A similar definition
works if R is replaced by an arbitrary metric space Y .
Definition. Let X and Y be metric spaces, and assume that X is compact. Given two continuous
functions f, g : X → Y there is some x ∈ X such that the distance from f (x) to g(x) is maximal.
Following our previous definition when Y is a normed

vector space, we define the distance from f
to g to be this maximum value of d f (x), g(x) .
We have seen that this definition makes C(X, Y ) into a metric space if Y is the real numbers.
In fact, this definition also yields a metric space for arbitrary choices of Y . All of the properties
except the Triangle Inequality follow immediately, so we shall limit the discussion to this property:
Given continuous functions

f , g and h from X to Y , we know there is some x ∈ X such that
d(f, h) = d f (x), h(x) . Applying the Triangle Inequality for Y to the right hand side, we obtain


d(f, h) = d f (x), h(x) ≤



d f (x), g(x) + d g(x), h(x) ≤ d(f, g) + d(g, h)
and therefore the Triangle Inequality holds for C(X, Y ).
The metric we have constructed is usually called the uniform metric and the topology is usually
called the uniform topology because convergence of a sequence in this metric is the usual notion of
uniform convergence.

Formal properties of function spaces

Function space constructions in mathematics usually satisfy some basic formal conditions; we
shall describe these for the topologies on function spaces that were defined above. Such properties
are often useful for a variety of purposes.

71
 We shall begin very generally. Suppose that we have some class C of mathematical objects and
a procedure for associating to each object X in C a new object T (X) in some possibly different class
C 0 . For many reasons it is often useful to have an associated construction for “good” morphisms
in the class C. Describing this notion more precisely in a general setting would require a nontrivial
digression, but if C is the class of topological spaces then a natural choice for “good” morphisms is
given by continuous functions.
To be more specific, given objects X and Y with a morphism f from X to Y , one would
like to extend the construction T to obtain a morphism T (f ) relating T (X) and T (Y ). For some
constructions we want a covariance property for which the map T (f ) goes from T (X) to T (Y ),
but for other constructions we want a contravariance property for which the map T (f ) goes in
the opposite direction.
There are two ways of making constructions for the class C of topological spaces that are of
interest to us here:
(i) One can fix a space B and take U (X) to be the continuous functions from X to B.
(ii) One can fix a space A and take V (X) to be the continuous functions from A to X.
In the first case one obtains a contravariant construction as follows: Given a continuous function
f : Y → X, define

U (f ) : U (X) = C(X, B) −→ C(Y, B) = U (Y )

sending g : X → B to the composite g o f .

PROPOSITION 1. If U is defined as above then the following hold:
(i) For all X, the map U (id X ) is the identity.
(ii) For all continuous maps h : Z → Y we have U (f o h) = U (h) o U (f ).
Proof. The first follows because g o idX = g for all g, and the second follows because

[U (f o h)] (g) = g o (f o h) = (g o f ) o h = [U (h)](g o f ) =

 
[U (h)] [U (f )](g) = [U (h) o U (f )](g)

for all g.
Notation. The two identities in the proposition are often summarized in the statement that U
is a contravariant functor from topological spaces to sets.
In the second case one proceeds similarly to obtain a covariant construction as follows: Given
a continuous function f : Y → X, define

V (f ) : V (X) = C(A, X) −→ C(A, Y ) = V (Y )

sending g : A → X to the composite f o g.

PROPOSITION 2. If V is defined as above then the following hold:
(i) For all X, the map V (id X ) is the identity.
(ii) For all continuous maps h : Y → Z we have V (h o f ) = V (h) o V (f ).
Proof. The first follows because idX o g = g for all g, and the second follows because

72
 [V (h o f )] (g) = (h o f ) o g = h o (f o g) = [V (h)](f o g) =
[V (h)] ( [V (f )](g) ) = [V (h) o V (f )](g)
for all g.
Notation. The two identities in the proposition are often summarized in the statement that V
is a covariant functor from topological spaces to sets.
In the preceding constructions the uniform metric makes the sets U (W ) and V (W ) into topo-
logical spaces if the domain is a compact metric space and the codomain is a metric space. Therefore
the following questions arise immediately:
QUESTION 1. Suppose that B, X and Y are metric spaces such that X and Y are compact,
and that f : Y → X is continuous. Is the associated function U (f ) a continuous mapping with
respect to the uniform metrics on the function spaces?
QUESTION 2. Suppose that A, X and Y are metric spaces such that A is compact, and that
f : Y → X is continuous. Is the associated function V (f ) a continuous mapping with respect to
the uniform metrics on the function spaces?
The following two related questions also arise naturally.
QUESTION 3. Suppose that X and Y are metric spaces, and assume that X is compact. Let

e : C(X, Y ) × X −→ Y

be the evaluation mapping defined by e(f, x) = f (x). Is e a continuous mapping?

QUESTION 4. Suppose that X, Y and Z are metric spaces such that X and Y are compact.
Let
ϕ : C(X, Y ) × C(Y, Z) −→ C(X, Z)
be the composition mapping ϕ(f, g) = g o f . Is ϕ a continuous mapping?
Our first observation is that Question 4 is particularly fundamental.
THEOREM 3. If Question 4 has a positive answer, then Questions 1 and 3 also have positive
answers, and Question 2 has a positive answer at least if X is compact.
Proof. We begin by showing that a positive answer to Question 4 implies the same for Question
3. By a special case of the former, the composition map

C({0}, X) × C(X, Y ) −→ C({0}, Y )

is continuous. Now for every metric space W there is a canonical 1–1 correspondence from W to
C({0}, W ) sending each point w ∈ W to the unique function such that f (0) = w. Furthermore,
this map is an isometry with respect to the uniform metric on the function space. If θ W : W →
C({0}, W ) denotes the canonical 1 - 1 correspondence, then it is an elementary exercise to verify
that
e = θY−1 o ϕ o (θX × id)
and thus the continuity of ϕ implies the continuity of e.
We next show that a positive answer to Question 4 implies the same for Question 1. In this
case we have
[U (f )](g) = ϕ(f, g)

73
so the continuity of the two variable function ϕ implies the continuity of the single variable function
U (f ) obtained by holding f fixed.
Finally, we show that a positive answer to Question 4 implies the same for Question 2 if X is
compact. In this case we have
[V (f )](g) = ϕ(g, f )
so the continuity of the two variable function ϕ implies the continuity of the single variable function
V (f ) obtained by holding f fixed.
We are now left with two things to prove; namely, that Question 2 has a positive answer even
if X is not necessarily compact, and that Question 4 has a positive answer. We begin by showing
that Question 2 has a positive answer in all cases.

PROPOSITION 4. Let X, Y, Z be metric spaces such that X is compact. Then the map
V (f ) : C(X, Y ) → C(X, Z), defined by sending a function g to the function f o g, is continuous.
Proof. Let g : X → Y be an arbitrary continuous function; we shall prove that V (f ) is continuous
at g. Let ε > 0 be arbitrary.
For each x ∈ X there is some δ(x) > 0 such that y ∈ N δ (x)( g(x) ) implies that

1
d( f (y), f (g(x)) ) < 3ε .

By compactness of g[X] there is a finite set of points x i ∈ X such that the disks of radius 14 δ(xi )
cover the compact set g[X]. Let δ > 0 be the minimum of the numbers 14 δ(xi ), and suppose that
d(g, h) is less than the minimum of δ and 13 ε. It will suffice to show that

d(f o g(x), f o h(x) ) < ε

for all x ∈ X.
1
For each x ∈ X choose i such that x lies in the neighborhood of radius 4 δ(xi ), and suppose
that d(g, h) < δ. Then we have





d f o g(x), f o h(x) ≤ d f o g(x), f o g(xi ) + d f o g(xi ), f o h(xi ) + d f o h(xi ), f o h(x) .

The first of these is less that 13 ε because we are assuming that the distance between g(x) and g(x i )
is less than 14 δ(xi ), and the second summand is less than 13 ε by the hypothesis on d(g, h). Proving
the third summand is less than 31 ε requires some additional work. By construction, the distances
between g(x) and g(xi ), g(xi ) and h(xi ), and g(x) and h(x) are all less than than 14 δ(xi ), so by the
Triangle Inequality the distance between h(x) and g(x) is less than 43 δ(xi ). The latter implies that
the distances between the images of these
points under f is less than 31 ε, and this completes the
last step in proving that d f o g(x), f o h(x) < ε.
We conclude the discussion proving the continuity of the composition map.
THEOREM 5. In the setting of Question 4, the composition mapping

ϕ : C(X, Y ) × C(Y, Z) −→ C(X, Z)

is continuous with respect to the uniform metric(s) and the product topology of the latter on the
domain.

74
Proof. We shall prove that ϕ is continuous at an arbitrary point (g, f ) in the domain. Let ε > 0
as usual.
The function f is uniformly continuous because Y is compact. Hence we can find some δ > 0
such that d(u, v) < δ implies d( f (u), f (v) ) < 12 ε. Therefore if (g 0 , f 0 ) satisfies d(g 0 , g) < δ and
d(f 0 , f ) < 21 ε, then it follows that




d f o g, g 0 o f 0 ≤ d g o f, f o g 0 + d f o g 0 , f 0 o g 0

where the first summand on the right hand side is less than 21 ε by the uniform continuity consid-
erations and the second is less than 12 ε because of our assumption on f and f 0 .

The compact-open topology

We have already noted that our approach differs from Munkres, so our purpose in this sub-
section is to relate his approach to ours. One motivation for the approach is that one would like
to have a way of defining the topology of the uniform metric on C(X, Y ) entirely in terms of the
topologies on X and Y (and hence that this topology does not depend upon the choices of metrics).
Definition. Let X and Y be topological spaces, and let C(X, Y ) be the set of continuous
functions from X to Y . The compact-open topology on C(X, Y ) is the topology generated by the
sets 
W(K, U ) = f ∈ C(X, Y ) | f [K] ⊂ U
where K is compact in X and U is open in Y .
Theorem 46.8 on pages 285–286 of Munkres then contains the following conclusion as a special
case:
THEOREM 6. (Equivalence theorem for function space topologies) If X and Y are metric
spaces and X is compact, then the compact-open topology is equal to the topology determined by
the uniform metric.
Remark 1. All the results for uniform metric topologies in this section extend to compact-open
topologies (at least in many cases). In particular, the continuity of the evaluation map is proven in
Theorem 46.10 on page 286 of Munkres, and the continuity of composition is established in Exercise
7 on page 289 of Munkres.
Remark 2. One can also prove that the compact-open topology is metrizable under more general
conditions than the compactness of X and the metrizability of Y . As noted in Exercise 7 for
this section, for arbitrary spaces X and Y the compact-open topology on C(X, Y ) can only be
metrizable if Y is metrizable (the constant functions form a subset homeomorphic to Y ), so the
latter is a necessary assumption. One standard condition on X is that it should be a countable
union of open subspaces with compact closures. If we express X = ∪ n Kn where Kn is compact,
then one can use the methods of Exercise VI.5.1 to construct a metric on C(X, Y ) such that a
sequence fn of functions converges to a limit function f if and only if for each of the compact
subsets Km the functions fn |Km converge uniformly to f |Km .

Compactness in function spaces

There are many natural questions about the topological properties of function spaces that can
be stated formally, and some have extremely far-reaching implications to questions of independent
interest. One particularly important example is the following:

75
Compactness question. Let X and Y be metric spaces such that X is compact, and let A be a
family of functions in C(X, Y ). Under what conditions on the functions in A does the latter have
compact closure in C(X, Y ) with respect to the uniform metric?
An extremely useful answer to this question is given by the following result, which generalizes
a classical theorem due to G. Ascoli and C. Arzelà.
THEOREM 7. In the above setting, a subset A ⊂ C(X, Y ) has compact closure if and only if
it is bounded and equicontinuous; i.e., for every

ε > 0 there is a δ > 0 such that for each f ∈ A
we have that d(u, v) < δ implies d f (u), f (v) < ε.
Example. Let X be the unit interval, let Y be the real line, and let A be the set of all continuous
functions expressible in the form Z x
f (x) = g(y) dy
0

for some arbitrary continuous real valued function g on [0, 1] such that |g| ≤ 1. If u < v we then
have Z v 
 
|f (u) − f (v)| =  g(y) dy  ≤ |u − v|
u

so that |f (x)| ≤ 1 for all f and x, and for each ε we can in fact take δ = ε.
The classical case of this result is stated and proved as Theorem 45.4 on pages 278–279 of
Munkres, and the more general result is given by Theorem 47.1 on pages 290–292 of Munkres.
AN APPLICATION OF EQUICONTINUITY AND COMPACTNESS. In Exercise 25 on pages
170–171 of Rudin, Principles of Mathematical Analysis, the relation between compactness and
equicontinuity is applied to prove that if F (x, y) is a continuous function of two variables, then
the differential equation y 0 = F (x, y) can be solved with a given initial condition. This is a partial
extension of the Picard result for functions F satisfying Lipschitz conditions that we previously
established using the Contraction Lemma; as noted in Unit III, one major difference with Picard’s
result is that the solution to the differential equation is not necessarily unique if F is merely
continuous. For example, if F (x, y) = y 1/2 , then both the zero function and y = 14 x2 solve the
differential equation and satisfy y(0) = 0.

76
 IV.2 : The adjoint property

(Munkres, §§ 46–47)

Product constructions are useful in mathematics both as The following result is a special case
of more general theorems (compare Munkres, Theorem 46.11, p. 287). We are only stating a simple
case because we want to present the important central idea without introducing a great deal of
additional material at this point.
THEOREM 1. Let K, X and Y be metric spaces, and assume Kand X are compact. Let
C(X, Y ) be the space of continuous functions from X to Y with the uniform metric. Then the
following hold:
(i) Given a continuous function F : K × X → Y , there is a unique continuous function
F : K → C(X, Y ) such that for all k ∈ K and x ∈ X we have [F [ (k)](x) = F (k, x).
[

(ii) Conversely, if g : K → C(X, Y ) is continuous, then there is a unique continuous F :

K × X → Y such that g = F [ .


(iii) Finally, if A : C(K × X, Y ) → C K, C(X, Y ) is the 1 − 1 correspondence given by (i)
and (ii) above, then A is an isometry with respect to the uniform metrics. In particular, A is a
homeomorphism of function spaces.
A concise way of expressing the conclusion of this theorem is by the formula


C(X × Y, Z) ∼
= C X, C(Y, Z)

which means that in some formal sense the construction sending X to X × Y is adjoint to the
construction sending Z to C(Y, Z). Here is an online reference with more background information
on this choice of terminology:
http://en.wikipedia.org/wiki/Adjoint functor
Proof. Since K × X is compact, the function F is uniformly continuous, and hence for every
ε > 0 there is some δ > 0 such that d(x, x 0 ) < δ and d(k, k 0 ) < δ imply


d F (k, x), F (k 0 , x0 ) < ε.

But this clearly implies that d(F [ (k), F [ (k 0 ) ) is also less than ε because the maximum value for
the latter is attained for some x ∈ X by compactness of X.
Suppose now that we have two continuous mappings F and G from K × X to Y such that
F [ = G[ . Then for all k we have F [ (k) = G[ (k). The latter means that for all x ∈ X we have

F (k, x) = [F [ (k)](x) = [G[ (k)](x) = G(k, x)

and this proves the uniqueness assertion.

To complete the proof of (ii), suppose we are given a continuous function g : K → C(X, Y ).
Then there is a set-theoretic function F : K ×X → Y defined by F (k, y) = [g(k)](y), so that F [ = g
follows immediately from the construction. It remains to show that F is continuous. Let (k, x) ∈
K × X be arbitrary. By the compactness of K we know that g is uniformly continuous.

1 Therefore
0 0
for every ε > 0 there is a δ1 > 0 such that d(k, k ) < δ implies d g(k), g(k ) < 2 ε. Furthermore,

77


by continuity there is a δ2 > 0 such that d(x, x0 ) < δ2 implies d [g(k)](x), [g(k 0 )](x) < 21 ε.
Therefore if we let δ be the smaller of δ 1 and δ2 , it will follow that d(k, k 0 ) < δ and d(k, k 0 ) < δ
imply
d F (k, x), F (k 0 , x0 ) ) = d [g(k)](x), [g(k 0 )](x0 ) ) ≤
ε ε
d [g(k)](x), [g(k)](x0 ) ) + d [g(k)](x0 ), [g(k 0 )](x0 ) ) ≤ + = ε
2 2
and hence F is continuous at the arbitrary point (k, x) ∈ K × X.
Finally, we need to check that A is an isometry. Given continuous functions F
and G in
the domain of A, the distance between F [ and G[ is the maximum of d F [ (z), G[ (z) . Let z0
[ [


be the point at which
this maximum is realized. Then d F (z0 ), G (z0 ) is the maximum of
d F (z0 , x), G(z0 , x) , which is less than or equal to d(F, G). Therefore A has the property that
d(F [ , G[ ) ≤ d(F, G).


To see that equality holds, choose (z 1 , x1 ) ∈ K × X such that d F (z1 , x1 ), G(z1 , x1 ) is
maximal. It then follows that


d(F, G) ≤ d F [ (z1 ), G[ (z1 ) ≤ d(F [ , G[ ) .

If we combine this with the conclusion of he previous paragraph, we conclude that A is an isometry.
Important special case. In topology the special case of this result with K = [0, 1] is particularly
important. A continuous map h from [0, 1] to C(X, Y ) then represents a curve joining two functions.
By the theorem we know that h = H [ for some unique function H : X × [0, 1] → Y , and by
construction we know that the restrictions of H to X × {0} and X × {1} are basically the original
functions h(0) and h(1). The map H is called a homotopy from h(0) to h(1). This concept plays
a fundamental role in modern topology. We shall simply make one observation:
PROPOSITION 2. Suppose that X is a compact metric space and W is open in R n for some
n. Let f be a continuous map from X to W . Then there is some δ > 0 such that if g : X → W is
a continuous map satisfying d(f, g) < δ, then g is homotopic to f .
Proof. Let K = f (X), so that K is a compact subset of W . Then the continuous function
h(x) = d(x, R
n − W ) is positive on K and hence has a minimum value, say δ > 0. It follows that
Nδ f (x); Rn ⊂ W for all x ∈ X. Therefore if g satisfies d(f, g) < δ, then for all t ∈ [0, 1] the
image of the function
ht (x) = t · g(x) + (1 − t) · f (x)
is contained in W , and thus the right hand side defines a homotopy H : X × [0, 1] → W such that
H [ (0) = f and H [ (1) = g.

78
 V. Constructions on Spaces

This material in this unit concerns two basic themes that run throughout geometry (and
topology) — all the way from children’s toys to the frontiers of research. One of these is the
construction of new objects from old ones by gluing certain subsets together. For example, one can
form a circle (actually, a space homeomorphic to a circle) from a closed interval by gluing the two
endpoints together. A second theme is the formation of new objects from old ones by first creating
several disjoint copies of the original objects and then gluing pieces together in an appropriate
fashion. For example, one can form a cube from six pairwise disjoint solid squares with sides of
equal length by gluing the latter together in a suitable way along the edges.
Two widely known examples of such constructions are the Möbius strip and the Klein bottle,
and we shall indicate how they can be formed using the ideas presented here. The latter can also
be used to show that one can construct a Klein bottle from two Möbius strips by gluing the latter
together along their (homeomorphic) edges.

V.1 : Quotient spaces

(Munkres, § 22)

We shall adopt a somewhat different approach from the one appearing in Munkres. In mathe-
matics it is often useful to create a quotient object from a mathematical structure and a well-behaved
equivalence relation on such an object. For example, if n is a positive integer greater than 1, then
one can construct the ring Zn of integers modulo n using the equivalence classes of the relation

a ≡ b (n) ⇐⇒ a − b = kn, some k ∈ Z

and the projection from Z to the set of equivalence classes Z n is compatible with the addition
and multiplication on both systems. One often says that Z n is a quotient ring of Z, and further
constructions of this sort are indispensable in most of abstract algebra.
More generally, quotient constructions arise naturally for many types of mathematical systems,
so the following question about topological spaces arises naturally at least from a formal perspective:
Question. Let X be a topological space and let R be an equivalence relation on X. Is there a
reasonable definition of a topology on the set of equivalence classes X/R?
One obvious requirement is that the projection map

πR : X −→ X/R

taking a point x to its R-equivalence class [x] R (frequently abbreviated to [x]) is continuous. One
trivial way of achieving this is to take the indiscrete topology on X/R, but something this easy
should seem too good to be true and probably too simple to be useful (and it is!). For example, if
X is Hausdorff we would like the topology on X/R to be Hausdorff, at least as often as possible
(there are fundamental examples to show that one cannot always have a Hausdorff topology).

79
 It turns out that the right topology to take for X/R is the unique maximal topology for which
the equivalence class projection π R is continuous, and it is useful to formulate things somewhat
more generally.
Definition. Let (X, T) be a topological space, let Y be a set, and let f : X → Y be a map of
sets. The quotient topology f∗ T on Y is defined by the condition
V ∈ f∗ T if and only if f −1 [V ] ∈ T.
Before proceeding, we need to check that the construction above yields a topology on X/R.
— The inverse image of the empty set is the empty set and the inverse image of X/R is X, so f ∗ T
contains the empty set and X/R. If Uα lies in X/R for all α, then f −1 [∪α Uα ] = ∪α f −1 [Uα ] where
each term on the right hand side lies in T by the definition of f ∗ T; since the union of open sets
in X is again an open subset, it follows that f −1 [∪α Uα ] is open in X which in turn implies that
∪α Uα belongs to f∗ T. Likewise, if U1 and U2 are in f∗ T, then f −1 [U1 ∩ U2 ] = f −1 [U1 ] ∩ f −1 [U2 ],
and each term of the right hand side lies in T; since the latter is a topology for x it follows that
the right hand side also lies in T, and therefore it follows that U 1 ∩ U2 lies in f∗ T.
The following are immediate consequences of the definition and the preceding paragraph.
PROPOSITION 1. In the setting above, we have the following:
(i) f defines a continuous map from (X, T) to (Y, f ∗ T),
(ii) f∗ T contains every topology U for which f : (X, T) → (Y, U) is continuous.
(iii) A subset B ⊂ Y is closed with respect to f ∗ T if and only if its inverse image f −1 (B) is
closed with respect to T.
Proof. The first statement follows because a set V ⊂ Y is open if and only if f −1 [V ] is open in
X. The second is verified by noting that if U is given as above and W ∈ U, then f −1 [W ] ⊂ T by
continuity, and hence U ⊂ T. The third statement holds because B is closed with respect to the
quotient topology if and only if Y − B is open, which is true if and only if f −1 [Y − B] = X − f −1 [B]
is open in X, which in turn is true if and only if f −1 [B] is closed in X.

Quotients and morphisms

It is helpful to deal first with some aspects of equivalence class projections that are entirely
set-theoretic and relate the discussion here to the approach in Munkres.
If X is a set and R is an equivalence relation on X, then the equivalence class projection π R
is onto. In fact, every onto map can be viewed as an equivalence class projection as follows: If
f : X → Y is an arbitrary onto map of sets, and then one can define an equivalence relation R f on X
by u ∼ v if and only if f (u) = f (v). There is a canonical 1–1 correspondence h : Y → X/R f which
sends the equivalence class [x] to f (x). It is an elementary exercise to show that h is well-defined,
1–1 and onto (verify this!).
The following set-theoretic observation describes an fundamental property of quotient con-
structions with respect to functions.
PROPOSITION 2. Let X and Y be sets, let f : X → Y be a function, let R be an equivalence
relation on X, and let p : X → X/R be the map sending an element to its equivalence class.
Suppose that whenever u ∼R v in X we have f (u) = f (v). Then there is a unique function
g : X/R → Y such that g o p = f .

80
 The point is that a well-defined function is obtained from the formula g([x]) = f (x). Analogous
results hold for a wide range of mathematical structures; a version for topological spaces is given
immediately after the proof below.
Proof. Suppose that [x] = [y]; then by definition of the equivalence relation we have f (x) = f (y),
and therefore the formula defining g yields

g([x]) = f (x) = f (y) = g([y])

which shows that g is indeed well-defined.

COROLLARY 3. Let f, X, Y, R be as in the proposition, suppose they satisfy the conditions
given there, suppose there are topologies on X and Y such that f is continuous, and put the
quotient topology on X/R. Then the unique map g is continuous.
−1 −1
Suppose that W is open in Y . Then g −1 [W ] is open in X/R if and only if πR (g [W ] is open
−1 −1 −1
in X. But g πR = f , and hence πR (g [W ] = f [W ]. Since f is continuous, the latter is in fact
o

open in X. Therefore g −1 [W ] is indeed open, so that g is continuous as required.

The following relates our approach to that of Munkres.
Definition. Let f : X → Y be continuous and onto, let R f be the equivalence relation described
above, and let h : X/R → Y be the standard 1–1 onto map described above. By construction,
h is continuous if X/R is given the quotient topology. We say that f is a quotient map if h is a
homeomorphism.
By the definitions, f is a quotient map if and only if for every B ⊂ Y we have that B is open
in Y if and only if f −1 [B] is open in X. — The statement remains true if one replaces “open” by
“closed” everywhere.
Remark. In many books the quotient topology is only defined for continuous maps that are
onto, so we shall comment on what happens if f is not onto. In this case, if y 6∈ f [X] then {y} is an
open and closed subset because f −1 [ {y} ] = ∅ and ∅ is open and closed in X; more generally, every
subset of Y − f [X] is open and closed for the same reason. It also follows that f [X] is open and
closed because its inverse image is the open and closed subset X. The quotient topology on the
entire space Y is given by the quotient topology on f [X] with respect to the onto map g : X → f [X]
determined by f and the discrete topology on Y − f [X].

Recognizing quotient maps

The following result provides extremely useful criteria for concluding that a continuous onto
map is a quotient map.
PROPOSITION 4. Suppose that f : X → Y is continuous and onto, and also assume that f
is either an open mapping or a closed mapping. Then f is a quotient map.
Proof. We shall only do the case where f is open; the other case follows by replacing “open”
with “closed” everywhere in the argument.
We need to show that V is open in Y if and only if f −1 [V ] is open in X. The ( =⇒ ) implication
is true by continuity, and the other implication follows from the hypothesis that f is open because
if f −1 [V ] is open in X then  
V = f f −1 [V ]

81
must be open in Y .
Exercise 3 on page 145 of Munkres gives an example of a quotient map that is neither open
nor closed. We have already given examples of continuous open onto maps that are not closed and
continuous closed onto maps that are not open.
COROLLARY 5. If X is compact, Y is Hausdorff and f : X → Y is continuous, then the
quotient space X/Rf is homeomorphic to the subspace f [X] (and hence the quotient is Hausdorff).
Proof. The preceding arguments yield a continuous 1–1 onto map h from the space X/R f , which
is compact, to the space f [X], which is Hausdorff. Earlier results imply that h is a closed mapping
and therefore a homeomorphism.

Important examples

Throughout this discussion D n will denote the set of all points x in R n satisfying |x| ≤ 1 (the
unit n-disk) and S n−1 will denote the subset of all point for which |x| = 1 (the unit (n − 1)-sphere).
We start with the first example at the top of this unit; namely, the circle S 1 is homeomorphic
to the quotient of [0, 1] modulo the equivalence relation R whose equivalence classes are the one
point sets {t} for t ∈ (0, 1) and the two point set {0, 1}. The construction of a homeomorphism is
fairly typical; one constructs a continuous onto map from [0, 1] to S 1 for which the inverse images
of points are the equivalence classes of R. Specifically, let f be the map [0, 1] → S 1 defined by
f (t) = exp(2πit).
Non-Hausdorff quotients. We have already mentioned that one can find equivalence relations
on Hausdorff spaces for which the quotient spaces are not Hausdorff. The example we shall consider
is one with only finitely many equivalence classes: Take the equivalence relation A on the real line
R whose equivalence classes are all positive real, all negative reals and zero (one verbal description
of this relation is that two real numbers are A-related if and only if one is a positive real multiple
of another. Then there are three equivalence classes that we shall call +, − and 0, and the closed
subsets are precisely the following:

∅, R/A, {0}, {+, 0}, {−, 0}

Since the one point subsets {±} are not closed in this topology, it is not Hausdorff. Since the
quotient topology is the largest topology such that the projection map is continuous, it follows that
in this case there is NO Hausdorff topology on R/A for which the projection

πA : R → R/A

is continuous (if U is a Hausdorff topology for which the map is continuous, then it is contained in
the quotient topology and hence the latter must be Hausdorff.
Exercise 6 on page 145 of Munkres gives an example of a quotient map on a Hausdorff space
where one point subsets in the quotient space are always closed but the quotient is not Hausdorff.
We now come to the constructions of the Möbius strip and Klein bottle. Our description of
the former will be designed to reflect the usual construction by gluing together the two short ends
of a rectangle whose length is much greater than its width. Let K > 0 be a real number, and
consider the equivalence relation M K on [−K, K] × [−1, 1] whose equivalence classes are the one
point sets { (s, t) } for |s| < K and the two point sets { (−K, −t), (K, t) } for each t ∈ [−1, 1].

82
Mathematically this corresponds to gluing {K} × [−1, 1] to {−K} × [−1, 1] with a twist. A Möbius
strip may be viewed as the associated quotient space; note that any two models constructed above
are homeomorphic (one can shrink or stretch the first coordinate; details are left to the reader).
— Similarly, one can construct a Klein bottle from the cylinder [−K, K] × S 1 by means of the
equivalence relation whose equivalence classes are the one point sets { (s, z) } for |s| < K and the
two point sets { (−K, z), (K, z) } for each z ∈ S 1 . Mathematically this corresponds to gluing
{K} × S 1 to {−K} × S 1 by the reflection map that interchanges the upper and lower arcs with
endpoints ±1. As in the previous case, the quotient spaces obtained for different values of K are
all homeomorphic to each other.
It is physically clear that one can construct the Möbius strip in R 3 , and although one cannot
find a space homeomorphic to the Klein bottle in R 3 (one needs some algebraic topology to prove
this!), some thought strongly suggests that the Klein bottle should be homeomorphic to a subset of
R4 (this has been exploited by numerous science fiction authors). For the sake of completeness (and
to prove that the spaces constructed are Hausdorff) we shall prove these realization statements.

The first step is a simple geometric observation:

LEMMA 6. For all positive integers p and q the products S p × S q and S p × D q+1 are homeo-
morphic to subsets of Rp+q+1 .
Proof. We know that S p ×R is homeomorphic to the nonzero vectors in Rp+1 by the map sending
(x, t) to tx because

P (v) = |v|−1 v, |v|
is the inverse. Taking products with R q shows that S p × Rq+1 is homeomorphic to a subset of
Rp+q+1 , and the lemma follows because the former clearly contains S p × S q and S p × D q+1
The next step is to observe that one can write the spaces in question as quotients of [a, b] × X
for a < b ∈ R and X = [0, 1] or S 1 depending upon whether we are constructing the Möbius strip
or Klein bottle; it is only necessary to consider the increasing linear homeomorphism from [−K, K]
to [a, b] and substitute a and b for −K and K in the description of the equivalence relations.
The final step is to construct continuous maps from the Möbius strip and Klein bottle to
S × D 2 and S 1 × S 2 respectively such that the inverse images of points in the codomains are
1

merely the equivalence classes of the defining relations for the quotient spaces. Since maps from
compact spaces into Hausdorff spaces are closed, this means that the images are homeomorphic to
the quotient spaces of the domains. For the Möbius strip the map f : [0, 1] × [−1, 1] → S 1 × D 2 is
given by

f (u, v) = exp(2πiu), v exp(πiu)
and for the Klein bottle the map g : [0, 1] × S 1 → S 1 × S 2 is given by


g(u, v) = exp(2πiu), Au (J(v))


where J : S 1 → S 2 is the standard inclusion of S 1 = S 2 ∩ R2 × {0} and Au is the orthogonal
rotation matrix  
1 0 0
 0 cos πu − sin πu 
0 sin πu cos πu
for which A1 (J(v)) = J (v). Verifications that these maps have all the desired properties are left
to the reader as an exercise.
Final remark. A very extensive treatment of quotient topologies is given in Chapter VI of
the text, Topology, by J. Dugundji.

83
 V.2 : Sums and cutting and pasting

Most texts and courses on set theory and point set topology do not say much about disjoint
union constructions, one reason being that everything is fairly elementary when one finally has the
right definitions (two references in print are Sections I.3 and III.4–III.7 of Jänich, Topology, and
Section 8.7 of Royden, Real Analysis). However, these objects arise immediately in a wide range
of geometrical and topological constructions of the sort described at the beginning of this unit, in-
cluding some fundamental examples from later courses in this sequence. A brief but comprehensive
treatment seems worthwhile to make everything more precise and to eliminate the need to address
the underlying issues in contexts that also involve more sophisticated concepts.

Disjoint union topologies

We have already defined the disjoin union (or set-theoretic sum) of two sets A and B to be
the set a [
A B = (A × {1}) (B × {2}) ⊂ (A ∪ B) × {1, 2}
` `
with injection maps iA : A → A B and iB : A → A B given by iA (a) = (a, 1) and iB (b) = (b, ` 2).
The images of these injections are disjoint copies of A and B, and the union of the images is A B.
Definition. If X and Y are topological spaces, the disjoint union
` topology or (direct) sum topology
or coproduct topology consists of all subsets having the form U V , where U is open in X and V
is open in Y .
`
We claim that this construction defines a topology on X Y , and the latter is a union of
disjoint homeomorphic copies of X and Y such that each of the copies is an open and closed subset.
Formally, all this is expressed as follows:
PROPOSITION 1. (Elementary
` properties of disjoint unions) The family of subsets described
above is a topology for X Y such that the injection maps iX and iY are homeomorphisms onto
their respective`images. These images are pairwise disjoint, and they are also open and closed
subspaces of X Y . Each injection map is continuous, open and closed.
Sketch of proof. This is all pretty elementary, but we include it because the properties are so
fundamental and the details are not readily available in the standard texts.
`
` X and Y are`open in themselves and ∅ is open `
Since in both, it follows that X Y and
∅ = ∅ ∅ are open in X Y . Given a family of subsets { Uα Vα } in the so-called disjoint union
topology, then the identity
! !
[ a  [ a [
Uα Vα = Uα Vα
α α α

shows
` that the so-called
` disjoint union topology is indeed closed under unions, and similarly the if
U1 V1 and U2 V2 belong to the so-called disjoint union topology, then the identity
   
\  a  \ a \
Ui Vi =  Ui   Vi 
i=1,2 i=1,2 i=1,2

84
shows that the so-called disjoint union topology is also closed under finite intersections. In partic-
ular, we are justified in calling this family a topology.
By construction U is open in X if and only if i X [U ] is open in iX [X], and V is open in Y if
and only if iY [V ] is open in iY [Y ]; these
` prove the assertions that i X and iY are homeomorphisms
onto their images. Since iX [X] = X ∅, it follows that the image of iX is open, and of course
similar considerations apply to the image of i Y . Also, the identity
 a 
iX [X] = X Y − iY [Y ]

shows that the image of iX is closed, and similar considerations apply to the image of i Y .
` `
The continuity of iX follows because every open set in X Y has the form U V where U
and V are open in X and Y respectively and
h a i
i−1
X U V = U

with similar
` conditions valid for i Y . The openness of iX follows immediately from the identity
iX [U ] = U ∅ and again similar considerations apply to i Y . Finally, to prove that iX is closed, let
F ⊂ X be closed. Then X − F is open in X and the identity
a  a   a 
iX [F ] = F ∅ = X Y − (X − F ) Y
`
shows that iX [F ] is closed in X
Y ; once more, similar considerations apply to i Y .
`
COROLLARY ` 2. The closed subsets of X Y with the disjoint union topology are the sets of
the form E F where E and F are closed in X and Y respectively.

`If the topologies on X and Y are clear from the context, we shall generally assume that the
X Y is furnished with the disjoint union topology unless there is an explicit statement to the
contrary.
Since the disjoint union topology is not covered in many texts, we shall go into more detail
than usual in describing their elementary properties.
PROPOSITION
` 3. (Further elementary properties) (i) If X and Y are discrete, then so is
X Y.
`
(ii) If X and Y are Hausdorff, then so is X Y.
`
(iii) If X and Y are homeomorphic to metric spaces, then so is X Y.
(iv) If f : X → W and g`: Y → W are continuous maps into some space W , then there is a
unique continuous map h : X Y → W such that h o iX = f and h o iY = g.
` `
(v) The spaces X Y and Y X are homeomorphic for all X and Y . Furthermore, if Z is a
third topological space then there is an “associativity” homeomorphism
 a a a a 
X Y Z ∼
= X Y Z

(in other words, the disjoint sum construction is commutative and associative up to a canonical
homeomorphism).

85
 `
Sketches of proofs. (i) A` space is discrete if every subset is open. Suppose that E ⊂ X Y .
Then E may be written as A B where A ⊂ X and B ⊂ Y . Since X ` and Y are discrete
`it follows
that A and B are open in X and Y respectively, and therefore E = A B is open in X Y . Since
E was arbitrary, this means that the disjoint union is discrete.
(ii) If one of the points p, q lies in the image of X and the other lies in the image of Y , then the
images of X and Y are disjoint open subsets containing p and q respectively. On the other hand,
if both lie in either X or Y , let V and W be disjoint`open subsets containing the preimages of p
and q in X or Y . Then the images of V and W in X Y are disjoint open subsets that contain p
and q respectively.
(iii) As noted in Theorem 20.1 on page 121 of Munkres, if the topologies on X and Y come
from metrics, one can choose the metrics so that the distances between two points are ≤ 1. Let d X
and dY be metrics of this type.
`
Define a metric d∗ on X Y by dX or dY for ordered pairs of points (p, q) such that both
lie in the image of iX or iY respectively, and set d∗ (p, q) = 2 if one of p, q lies in the image of i X
and the other lines in the image of iY . It follows immediately that d∗ is nonnegative, is zero if and
only if p = q and is symmetric in p and q. All that remains to check is the Triangle Inequality:

d∗ (p, r) ≤ d∗ (p, q) + d∗ (q, r)

The verification breaks down into cases depending upon which points lie in the image of one
injection and which lie in the image of another. If all three of p, q, r lie in the image of one of the
injection maps, then the Triangle Inequality for these three points is an immediate consequence of
the corresponding properties for d X and dY . Suppose now that p and r lie in the image of one
injection and q lies in the image of the other. Then we have d ∗ (p, r) ≤ 1 and

d∗ (p, q) + d∗ (q, r) = 2 + 2 = 4

so the Triangle Inequality holds in these cases too. Finally, if p and r lie in the images of different
injections, then either p and q lie in the images of different injections or else q and r lie in the images
of different injections. This means that d ∗ (p, r) = 2 and d∗ (p, q) + d∗ (q, r) ≥ 2, and consequently
the Triangle Inequality holds for all ordered pairs (p, r).
(iv) Define h(x, 1) = f (x) and h(y, 2) = g(y) for all x ∈ X and y ∈ Y . By construction
h o iX = f and h o iY = g, so it remains to show that h is continuous and there is no other continuous
map satisfying the functional equations. The latter is true for set theoretic reasons; the equations
specify the behavior of h on the union of the images of the injections, but this image is the entire
disjoint union. To see that` h is continuous, let U be an open subset `of X, and consider the inverse
image U ∗ = h−1 (U ) in X Y . This subset has the form U ∗ = V W for some subsets V ⊂ X
and W ⊂ Y . But by construction we have

V = i−1 ∗ −1 o −1
X (U ) = iX h (U ) = f −1 (U )

and the set on the right is open because f is continuous. Similarly,

W = i−1 ∗ −1 o −1
Y (U ) = iY h (U ) = g −1 (U )
`
so that the set on the right is also open. Therefore`U ∗ = V W where V and W are open in X
and Y respectively, and therefore U ∗ is open in X Y , which is exactly what we needed to prove
the continuity of h.

86
 (v) We shall merely indicate the main steps in proving`these assertions
` and leave the details to
the reader as an exercise. The homeomorphism τ from X Y to Y X is given by sending (x, 1)
to (x, 2) and (y, 2) to (y, 1); one needs to check this map is 1–1, onto, continuous and open (in fact,
if τXY is the map

described above,
then its inverse
is τ Y X ). The “associativity

homeomorphism”
sends (x, 1), 1 to (x, 1), (y, 2), 1 to (y, 1), 2 , and (z, 2) to (z, 2), 2 . Once again, one needs
to check this map is 1–1, onto, continuous and open.
COMPLEMENT 3. There is an analog of Property (iv) for untopologized sets.
Perhaps the fastest way to see this is to make the sets into topological spaces with the discrete
topologies and then to apply (i) and (iv).
Property (iv) is dual to a fundamental property of product spaces. Specifically, ordered pairs
of maps from a fixed object A to objects B and C correspond to maps from A into B × C, while
ordered pairs of `
maps going TO a fixed object A and coming FROM ` objects B and C correspond
to maps from B C into A. For this reason one often refers to B C as the coproduct of B and
C (either as`sets or as topological spaces); this is also
Q the reason for denoting disjoint unions by
the symbol , which is merely the product symbol turned upside down.

Copy, cut and paste constructions

Frequently the construction of spaces out of pieces proceeds by a series of steps where one
takes two spaces, say A and B, makes disjoint copies of them, finds closed subspaces C and D
that are homeomorphic by some homeomorphism h, and finally glues A and B together using
this homeomorphism. For example, one can think of a rectangle as being formed from two right
triangles by gluing the latter along the hypotenuse. Of course, there are also many more complicated
examples of this sort.
`
Formally speaking, we can try to model this process by forming the disjoint union A B and
then factoring out by the equivalence relation

x ∼ y ⇐⇒ x = y or

x = iA (a), y = iB (h(a)) for some a ∈ A or

y = iA (a), x = iB (h(a)) for some a ∈ A.
It is an elementary but tedious exercise in bookkeeping to to verify that this defines an equivalence
relation (the details are left to the reader; it is probably best simply to note that there are three
types of equivalence classes — one point sets in A − C, one point sets in C − D, and two point sets
of the form {a, h(a)} for a ∈ C).
The resulting quotient space will be denoted by
[
A B.
h:C≡D

As a test of how well this approach works, consider the following question:
Scissors and Paste Problem. Suppose we are given a topological space X and closed subspaces
A and B such that X = A∪B. If we take C = D = A∩B and let h be the identity homeomorphism,
does this construction yield the original space X?

87
 One would expect that the answer is yes, and here is the proof:
`
Retrieving the original space.` Let Y be the quotient space of A B with respect to the
equivalence relation, and let p : A B →` Y be the quotient map. By the preceding observations,
there is a unique continuous map f : A B → X such that f o iA and f o iB are the inclusions
A ⊂ X and B ⊂ X respectively. By construction, if u ∼ v with respect to the equivalence relation
described above, then f (u) = f (v), and therefore there is a unique continuous map h : Y → X such
that f = h o p. We claim that h is a homeomorphism. First of all, h is onto because the identities
h o p o iA = inclusionA and h o p o iB = inclusionB imply that the image contains A ∪ B, which is
all of X. Next, h is 1–1. Suppose that h(u) = h(v) but u 6= v, and write u = p(u 0 ), v = p(v 0 ).
The preceding identities imply that h is 1-1 on both A and B, and therefore one of u 0 , v 0 must lie
in A and the other in B. By construction, it follows that the inclusion maps send u 0 and v 0 to
the same point in X. But this means that u 0 and v 0 correspond to the same point in A ∩ B so
that u = p(u0 ) = p(v 0 ) = v. Therefore the map h is 1–1. To prove that h is a homeomorphism,
it suffices to show that h takes closed subsets`to closed subsets. Let F be a closed subset of Y .
Then the inverse image p−1 [F ] is closed in A B. However, if we write write h[F ] ∩ A = P and
h[F ] ∩ B = Q, then it follows that p−1 [F ] = iA [P ] ∪ iB [Q]. Thus iA [P ] = p−1 [F ] ∩ ` iA [A] and
iB [Q] = p−1 [F ] ∩ iB [B], and consequently the subsets i A [P ] and iB [Q] are closed in A B. But
this means that P and Q are closed in A and B respectively, so that P ∪Q is closed in X. Therefore
it suffices to verify that h[F ] = P ∪ Q. But if x ∈ F , then the surjectivity of p implies that x = p(y)
for some y ∈ p−1 [F ] = iA [P ] ∪ iB [Q]; if y ∈ iA [P ] then we have

h(x) = h(p(y)) = f (y) = f o iA (y) = y

for some y ∈ P , while if y ∈ iB (Q) the same sorts of considerations show that h(x) = y for some
y ∈ Q. Hence h(F ) is contained in P ∪ Q. On the other hand, if y ∈ P or y ∈ Q then the preceding
equations for P and their analogs for Q show that y = h(p(y)) and p(y) ∈ F for y ∈ P ∪ Q, so that
P ∪ Q is contained in h[F ] as required.
One can formulate an analog of the scissors and paste problem if A and B are open rather
than closed subset of X, and once again the answer is that one does retrieve the original space.
The argument is similar to the closed case and is left to the reader as an exercise.
Examples. Many examples for the scissors and paste theorem can be created involving
subsets of Euclidean 3-space. For example, as noted before one can view the surface of a cube as
being constructed by a sequence of such operations in which one adds a solid square homeomorphic
to [0, 1]2 to the space constructed at the previous step. Our focus here will involve examples of
objects in 4-dimensional space that can be constructed by a single scissors and paste construction
involving objects in 3-dimensional space.
1. The hypersphere S 3 ⊂ R4 is the set of all points (x, y, z, w) whose coordinates satisfy the
equation
x2 + y 2 + z 2 + w 2 = 1
and it can be constructed from two 3-dimensional disks by gluing them together along the boundary
spheres. An explicit homeomorphism
[
D3 D 3 −→ S 3
id(S 2 )

can be constructed using the maps

 p 
f± (x, y, z) = x, y, z, 1 − x2 − y 2 − z 2

88
on the two copies of D 3 . The resulting map is well defined because the restrictions of f ± to S 2 are
equal.
2. We shall also show that the Klein bottle can be constructed by gluing together two Möbius
strips along the simple
√ closed curves

on their edges. Let g ± : [−1, 1] → S 1 be the continuous 1–1
2
map sending t to ± 1 − t , t . It then follows that the images F± of the maps id[0,1] × [−1, 1]
satisfy F + ∪F− = [0, 1] × S 1 and F+ ∩ [0, 1] × {−1, 1}. If ϕ : [0, 1] × S 1 → K is the quotient
projection to the Klein bottle, then it is relatively elementary to verify that each of the sets ϕ[F ± ]
is homeomorphic to the Möbius strip (look at the equivalence relation given by identifying two
points if they have the same images under ϕ o g± ) and the intersection turns out to be the set
ϕ[F+ ] ∩ ϕ[F− ], which is homeomorphic to the edge curve for either of these Möbius strips.

Disjoint unions of families of sets

As in the case of products, one can form disjoint unions of arbitrary finite collections of sets
or spaces recursively using the construction for a pair of sets. However, there are also cases where
one wants to form disjoint unions of infinite collections, so we shall sketch how this can be done,
leaving the proofs to the reader as exercises.
Definition. If A is a set and { Xα | α ∈ A } is a family of sets indexed by A, the disjoint union
(or set-theoretic sum) a
Xα
α∈A

is the subset of all !

[
(x, α) ∈ Xα ×A
α∈S

such that x ∈ Xα .
This is a direct generalization of the preceding construction, which may be viewed as the
special case where A = {1, 2}. For each β ∈ A one has an injection map
a
iβ : X β −→ Xα
α∈A

the images of i β and iγ are disjoint if β 6= γ and the union of the

sending x to (x, β); as before, `
images of the maps iα is all of α Xα .
Notation.
P In the setting above, suppose that
` each X α is a topological space with topology T α .
Let α Tα be the set of all disjoint unions α Uα where Uα is open in Xα for each α.
`
As in the previous discussion, this defines a topology on α Xα , and the basic properties can
be listed as follows:
P `
THEOREM 4. (i) The family of subsets α Tα defines a topology for α Xα such that the
injection maps`iα are homeomorphisms onto their respective images. The latter are open and closed
subspaces of α Xα , and each injection is continuous, open and closed.
`
` (ii) The closed subsets of Xα with the disjoint union topology are the sets of the form
Fα where Fα is closed in Xα for each α.

89
 `
(iii) If each Xα is discrete then so is α Xα .
`
(iv) If each Xα is Hausdorff then so is α Xα .
`
(v) If each Xα is homeomorphic to a metric space, then so is α Xα .
(vi) If for each α we are given a continuous
` function f : X α → W into some fixed space W ,
then there is a unique continuous map h : α Xα → W such that h o iα = fα for all α.
The verifications of these properties are direct extensions of the earlier arguments, and the
details are left to the reader.
In linear algebra one frequently encounters vector spaces that are isomorphic to direct sums
of other spaces but not explicitly presented in this way, and it is important to have simple criteria
for recognizing situations of this type. Similarly, in working with topological spaces one frequently
encounters spaces that are homeomorphic to disjoint unions but not presented in this way, and in
this context it is also convenient to have a simple criterion for recognizing such objects.
THEOREM 5. (Recognition principle) Suppose that a space Y is a union of pairwise` disjoint
subspaces Xα , each of which is open and closed in Y . Then Y is homeomorphic to α Xα .
Proof. For each α ∈ A let jα : Xα → Y be the inclusion map. By (vi) above there is a unique
continuous function a
J: Xα −→ Y
α

such that J o iα = jα for all α. We claim that J is a homeomorphism; in other words, we need to
show that J is 1–1 onto and open. Suppose that we have (x α , α) ∈ iα [Xα ] and (zβ , β) ∈ iβ [Xβ ] such
that J(xα , α) = J(zβ , β). By the definition of J this implies i α (xα ) = iβ (zβ ). Since the images of
iα and iβ are pairwise disjoint, this means that α = β. Since i α is an inclusion map, it is 1–1, and
therefore we have xα = zβ . The proof that J is onto drops out of the identities
! " #
a [ [ [
J Xα =J iα [Xα ] = J [ iα [Xα ] ] = jα [Xα ] = Y .
α α α α

Finally, to prove that J is open let W be open in the disjoint union, so that we have
a
W = Uα
α

where each Uα is open in the corresponding Xα . It then follows that J[W ] = ∪α Uα . But for each
α we know that Uα is open in Xα and the latter is open in Y , so it follows that each U α is open in
Y and hence that J[W ] is open.

90
 VI . Spaces with additional properties

This unit is essentially a continuation of Unit III, and it deals with two main issues:

(1) Topological properties of spaces that are not compact but still have some important prop-
erties in common with compact spaces (open subsets of R n are particularly important
examples in this connection).
(2) Recognition of topological spaces that come from metric spaces. We shall concentrate on
questions involving spaces constructed from reasonable pieces and merely state the general
results with references to Munkres for the proofs.

In two thirds of a ten week beginning graduate course it is not possible to cover everything
about topological spaces that is useful in a broad range of mathematical contexts and/or for further
courses in geometry and topology. Two particularly worthwhile topics of this sort are paracom-
pact spaces (Munkres, § 41) and the compact-open topologies on spaces of continuous functions
(Munkres, § 46). Another interesting topic, known as dimension theory (Munkres, § 50), deals with
the following natural question: How can one use topology to define the dimension of a topological
space (as an integer ≥ −1 or ∞) such that the topological dimension of R n is precisely n? It
seems reasonable to expect that Rm and Rn are not homeomorphic if m 6= n, and such a definition
would yield this as a simple corollary. Proper mappings are another important topic that definitely
would be worth discussing; these mappings are discussed at a number of points in Munkres (where
they are called perfect maps), and the file proper.pdf in the course directory contains further
information.
An extremely comprehensive listing of properties of topological spaces, along with theorems
and examples to describe the logical interrelationships between the concepts, is contained in the
book, Counterexamples in Topology, by L. A. Steen and J. A. Seebach; the reference charts at the
end are particularly helpful for obtaining a good overview of this area. Another book containing
a very substantial amount of information on different properties of topological spaces is the text,
Topology, by J. Dugundji.
In analysis one considers a large variety of topological vector spaces (each one point subset
is closed, and both addition and scalar multiplication are continuous), and questions about the
metrizability of these spaces (and the properties of such metrics) arise naturally. This topic is
discussed specifically in Chapter 1 of Rudin, Functional Analysis.

VI.1 : Second countable spaces

(Munkres, § 30)

We have already noted that continuous functions from the unit interval to a Hausdorff space
are completely determined by their restrictions to the rational points of the interval. In fact, this
property holds for all subsets of Euclidean spaces. The proof of this depends upon the existence of
a countable dense subset and the fact that the topology comes from a metric. There are two useful

91
equivalent characterizations of such metric spaces, and one of the conditions implies the other two
for arbitrary topological spaces.
Definitions. If X is a topological space then X is said to be
(i) separable if it has a countable dense subset,
(ii) second countable (or to satisfy the second countability axiom) if there is a countable base
for the topology (i.e., a countable family B of open sets such that every open set is a union
of sets in B,
(iii) Lindelöf (or to have the Lindelöf property) if every open covering has a countable sub-
covering.

The logical relations between these concepts are given as follows:

THEOREM 1. A topological space that is second countable is also separable and Lindelöf, and
a metric space that is either separable or Lindelöf is also second countable.
In particular, all three concepts are equivalent for metric spaces. We shall prove the various
implications separately below.
At the end of this section we shall give examples to show that separable or Lindelöf spaces
need not be second countable, and for arbitrary topological spaces there is no relation between
separability and the Lindelöf property (i.e., there are examples where each is true and the other is
false).
The book, Counterexamples in Topology, by Steen and Seebach, is a standard reference for
examples of topological spaces which have one property but not another.

Implications of second countability

Proof of Theorem 1. We shall begin by showing that second countability implies the other
properties.
SEPARABILITY. Let B = {U1 , U2 , · · · } be a countable base for the topology, and form the
countable subset A ⊂ X by picking a point a i ∈ Ui for each i. To show that A is dense in X we
need to show that every open subset of X contains a point in A. Since every open set W ⊂ X is
a union of sets in B there is at least one U j that is contained in W . We then have a j ∈ Uj ⊂ W .
Therefore A = X.
LINDELÖF PROPERTY. Once again let B = {U1 , U2 , · · · } be a countable base for the topology.
Given an open covering U = { Wα } of X, let

B0 = {V1 , V2 , · · · }

be the (countable) family of all basic open sets that are contained in some element of W. It follows
that B0 is an open covering of X (because X is a union of the W α and each Wα is a union of sets
in B0 ). If for each j we pick α(j) such that V j ⊂ Wα(j) it follows that

WO = {Wα(1) , Wα(2) , · · · }

is a countable subcovering of W.

92
 For the time being we shall simply note that R n is an example of a separable metric space;
a countable dense subset is given by the subset Q n of points whose coordinates are all rational
numbers. The results below will show that Rn and its subspaces also have the other two properties.
Second countability behaves well with respect to some standard operations on topological
spaces:
PROPOSITION 2. A subspace of a second countable space is second countable, and the product
of two (hence finitely many) second countable spaces is second countable.
This is stated and proved as Theorem 30.2 on page 191 of Munkres.

The reverse implications for metric spaces

We shall show that each of the other two properties implies second countability for metric
spaces. The first of these implications will prove that R n is a separable metric space.
SEPARABLE METRIC SPACES ARE SECOND COUNTABLE. Let A = {a 1 , a2 , · · · } be a countable
dense subset and consider the countable family of open sets W m,n = N1/m (an ). Given an open set
U in X and a point p ∈ X, let ε > 0 be chosen so that N ε (x) ⊂ U . Choose m and n such that
1/2m < ε and d(x, an ) < 1/2m. If we set W (x) = Wm,n it then follows that x ∈ W (x) ⊂ U , and
consequently we also have U = ∪x W (x), which shows that the countable family W = { W m,n } is
a base for the topology.
COROLLARY 3. The space Rn is a second countable space.

The preceding implications have the following useful consequence.

PROPOSITION 4. If X is a separable metric space and A ⊂ X, then A is also separable.
This follows because separable metric implies second countable and the latter implies separable.
One can construct examples of separable topological spaces that have nonseparable subspaces.
COROLLARY 5. Every subset S of Rn has a countable dense subset.
Note that the subset S might not contain any points at all from some arbitrary countable
dense subset D ⊂ Rn .
PROOF THAT lINDELÖF METRIC SPACES ARE SECOND COUNTABLE. For each positive integer
n let Un be the family of all sets N1/n (x) where x runs through all the points of X. Then U n is an
open covering of X and consequently has a countable subcovering W n . We claim that W = ∪n Wn
is a base for the metric topology.
Let V be an open subset of X and let x ∈ V . Then there is some positive integer M such
that N1/n (x) ⊂ V for all n > M . Choose an open set N 0 (x) from W2n such that x ∈ N 0 . It then
follows that N 0 (x) ⊂ N1/n (x) ⊂ V , and therefore we have ∪x N 0 (x) = V , which shows that W is a
countable base for the metric topology on X.
A compact space is automatically Lindelöf, and therefore we have the following consequence
for compact metric spaces:
PROPOSITION 6. A compact metric space is second countable and has a countable dense
subset.

93
 Examples involving nonmetrizable spaces

We conclude this section with examples of spaces that are separable but not Lindelöf, Lindelöf
but not separable, and both separable and Lindelöf but not second countable. By the previous
results of this section it follows that examples of the first two types are neither second countable
nor homeomorphic to metric spaces, and examples of the third type are likewise not homeomorphic
to metric spaces.
SEPARABILITY DOES NOT IMPLY THE LINDEL ÖF PROPERTY. In one of the exercises there is an
example of a separable Hausdorff space X which has a closed subspace A that is both uncountable
and discrete. We claim this space cannot be Lindelöf. If it were, then the closed subspace A would
also be Lindelöf by another of the exercises. Since the open covering of A by one point subsets does
not have a countable subcovering, it follows that A and hence X cannot be Lindelöf.
THE LINDELÖF PROPERTY DOES NOT IMPLY SEPARABILITY. Since compact spaces are Lin-
delöf, it suffices to find a compact space that is not separable. We can do this using Tychonoff’s
Theorem (see page 42 of these notes for more on this result). Specifically, let X = {0, 1} A be an
uncountable product of spaces homeomorphic to {0, 1} with the discrete topology, and let A be the
indexing set for the family of spaces that are factors of X. It will be convenient to assume that
|A| > 2ℵ0 . Then X is compact (by Tychonoff’s Theorem) and Hausdorff (since this is true for all
products of Hausdorff spaces). We claim that X is not separable.
The first thing to note is that if Y is a separable space then the cardinality of the set of
continuous real valued functions on Y is precisely 2 ℵ0 ; the constant functions are a subset with this
cardinality, and we may show there are at most 2 ℵ0 such functions as follows: If D is a countable
dense subset of Y , then a continuous real valued function is completely determined by its restriction
to D, and thus an upper bound for the cardinality of the set of continuous real valued functions
on Y is given by the cardinality of the set of all set-theoretic real valued functions on D. But the
latter cardinality is

ℵ0
2 ℵ0 = 2 ℵ0 × ℵ 0 = 2 ℵ0 .
Therefore, if we can show that the cardinality of X = {0, 1} A is greater than 2ℵ0 it will follow that
X is not separable.
Given α ∈ A let fα be given by projection onto the α factor of {0, 1} A followed by the inclusion
of {0, 1} in R. Then α 6= β implies fα 6= fβ because their zero sets are different, and therefore we
have a subset of continuous real valued functions on {0, 1} A that is in 1–1 correspondence with A.
Since |A| > 2ℵ0 it follows that X cannot have a countable dense subset.
A SEPARABLE HAUSDORFF SPACE THAT ALSO SATISFIES THE LINDEL ÖF PROPERTY IS NOT
NECESSARILY SECOND COUNTABLE. Every countable space is automatically separable and Lin-
delöf (the space is its own countable dense subset, and given an open covering one can extract a
countable subcovering indexed by the space itself — for each x let U x be an open subset in the
covering that contains x). We also know that second countable implies first countable. Therefore
it will suffice to construct a countable Hausdorff space that is not first countable.
Define an equivalence relation on the space Q2 of points in R2 with rational coordinates such
that the equivalence classes are all one point sets {(x, y)}, where x and y are rational with y 6= 0,
together with the rational points on the x-axis. This space P P (Q) is sometimes known as the
rational pinched plane, and by construction it is countable. We need to show that this space is
Hausdorff but not first countable.

94
 To verify that the space is Hausdorff, suppose we have two distinct points in P P (Q). It follows
immediately that quotient space projection defines a homeomorphism from Q × (Q − {0}) to the
open subset P P (Q) − {L}, where L is the equivalence class given by the x-axis. Therefore the
open subset P P (Q) − {L} is Hausdorff, and there are disjoint open neighborhoods for each pair
of distinct points in P P (Q) − {L}. It remains to consider the case where one distinct point is L
and the other comes from (x, y) where y 6= 0. By the definition of the quotient topology, these
two points have disjoint open neighborhoods if and only if there are disjoint open subsets of Q 2
containing (x, y) and Q × {0}. But this is true because Q 2 is metrizable and hence T3 .
Finally we need to show that P P (Q) is not first countable; in fact, we claim that there is
no sequence of open sets { Un } in P P (Q) such that each one contains L and every open subset
W containing L contains some Un . Let π be the quotient space projection onto P P (Q); by the
definition of the quotient topology the open neighborhoods of L in P P (Q) are precisely the sets of
the form π[V ] where V is an open subset of Q2 containing the x-axis. Therefore it will suffice to
show there is an open subset W0 ⊂ Q2 containing the x-axis such that π[W 0 ] does not contain any
of the open sets Un . Since Un is the image of π −1 [Un ] under π, it will suffice to show that one can
find an open subset W0 such that W0 does not contain any of the open sets π −1 [Un ] ⊂ Q2 . Note
that each set π −1 [Un ] is an open subset containing the rational points of the x-axis.


Given a positive integer n there is an ε n > 0 such that Nεn (n, 0) is contained in π −1 [Un ].
If we define W0 to be the open set
!
[ \
(−n − 1, n + 1) × (−εn /2, +εn /2) Q2
n

then W0 contains the rational points of the x-axis but for each n we know that W does not contain
the open segment {n} × (−εn , +εn ), which is contained in π −1 [Un ]; it might be useful to draw a
picture as an aid to understanding this. Therefore W 0 cannot contain any of the open sets π −1 [Un ],
and as noted above this implies that P P (Q) is not first (or second) countable.

95
 VI.2 : Compact spaces – II

(Munkres, §§ 26, 27, 28)

This unit contains some additional results on compact spaces that are useful in many contexts.

Sequential compactness for metric spaces

We have shown that infinite sequences in compact metric spaces always have convergent sub-
sequences and noted that the converse is also true. Here is a proof of that converse:
THEOREM 1. If X is a metric space such that every infinite sequence has a convergent
subsequence, then X is compact.
Proof. The idea is to show first that X is separable, then to use the results on second countability
to show that X is Lindelöf, and finally to extract a finite subcovering from a countable subcovering.
Proof that a metric space is separable if every infinite sequence has a convergent subsequence.
Let ε > 0 be given. We claim that there is a finite collection of points Y (ε) such that the finite
family
{ Nε (y) | y ∈ Y (ε) }
is an open covering of X. — Suppose that no such finite set exists. Then one can recursively
construct a sequence { xn } in X such that
[
xn 6∈ Nε (xi ) .
i<n

By construction we then have that d(x p , xq ) ≥ ε if p 6= q, and therefore { xn } has no Cauchy

(hence no convergent) subsequence.
If A = ∪n Y (1/n) then A is countable and for every δ > 0 and x ∈ X there is a point of A
whose distance to x is less than δ, and therefore A is dense in X.
Proof that open coverings have finite subcoverings. Since X is separable and metric, it is
second countable, and therefore it also has the Lindelöf property. Therefore given an open covering
W of X we can fine a countable subcovering U = {U 1 , U2 , · · · }. We need to extract a finite
subcovering from U.
For each positive integer n let
[ 
En = X− Un .
i≤n

Then each En is closed and En ⊃ En+1 for all n. If some En is empty then the first n sets in U
form a finite subcovering; in fact, if some E n is finite, then there still is a finite subcovering (take
the first n sets in U and add one more set from U for each point in the intersection). Therefore the
proof reduces to finding a contradiction if one assumes that each E n is infinite.
If each En is infinite, then one can find a sequence of distinct points y n such that yj ∈ Ej for
each j. The assumption on X implies that the infinite sequence { y j } has a convergent subsequence,

96
say { yk(j) }. Let y ∗ be the limit of this subsequence. By construction, y m ∈ En if m ≥ n, and
because each Ei is closed it follows that y ∗ ∈ Ek(j) for all j. Since the sequence of closed sets { E n }
is decreasing, it follows that \ \
En = Ek(j)
n j
∗
and that y belongs to this intersection. But by construction the set ∩ n En is empty because U is a
countable open covering for X, and therefore we have a contradiction; it follows that U must have
a finite subcovering, and therefore X must be compact.

Wallace’s Theorem

The following result is related to the proof that a product of two compact spaces is compact,
and it turns out to be extremely useful in many contexts.
THEOREM 2. (Wallace’s Theorem) Let X and Y be topological spaces, let A and B be compact
subsets of X and Y , and let W be an open subset of X × Y that contains A and B. Then there
are open subsets U and V of X and Y respectively such that A ⊂ U , B ⊂ V and

A×B ⊂ U ×V ⊂ W .

Proof. Given p = (x, y) ∈ A × B one can find open sets U p and Vp in X and Y respectively such
that x ∈ Up , y ∈ Vp and Up × Vp ⊂ W .
For a fixed b ∈ B the open sets Up × Vp define an open covering of the compact subset A × {b}
and hence there is a finite subcovering associated to the family of sets

Ab = {Up(1) × Vp(1) , · · · , Up(M (b)) × Vp(M (b)) } .

If we take Vb# to be the intersection of the sets Vp(j) and Ub# to be the union of the sets Up(j) , it
follows that
A × {b} ⊂ Ub# × Vb# ⊂ W
for each b ∈ B.
The family of open subsets { Vb# } in Y defines an open covering of the compact subspace b,
and therefore there is a finite subcovering associated to some family of sets
# # # #
F = {Ub(1) × Vb(1) , · · · , Ub(N ) × Vb(N ) } .

# #
If we now take U to be the intersection of the sets U b(i) and V to be the union of the sets Vb(i) ,
it follows that
A×B ⊂ U ×V ⊂ W
as required.

97
 VI.3 : Separation axioms

(Munkres, §§ 31, 32, 33, 35)

In some sense the definitions for topological spaces and metric spaces present an interesting
contrast. While it is clear that one does not need the full force of the properties of a metric space
to prove many basic results in point set theory, it is also apparent that one needs something more
than the austere structure of a topological space to go beyond a certain point. Thus far we have
introduced several conditions like the Hausdorff Separation Property which suffice for proving a
number of basic results. This property is just one of a list of increasingly stronger properties that lie
between a topological space with no further conditions at all and a topological space that comes from
a metric space. There are many important examples of topological spaces in topology, geometry
and analysis that are not homeomorphic to metric spaces (e.g., many infinite-dimensional objects in
algebraic topology and the so-called weak topologies on Banach spaces), and such objects are one
motivation for introducing concepts somewhere between metric spaces and arbitrary topological
spaces. A second motivation, which can be viewed as interesting for its own sake as well as its
usefulness in certain situations, is the following:
METRIZABILITY PROBLEM. What sorts of topological conditions are necessary or suffi-
cient for a topological space to be homeomorphic to a metric space?
We shall consider this problem in some detail as the final topic of the course.

The Ti conditions

The traditional way of organizing the separation properties that may or may not hold in a
topological space involves a list of statements T i where the subscript is some rational number and
the strength of the condition increases with the index (so if i > j then T i =⇒ Tj ). We shall only
deal with the statements for i = 0, 1, 2, 3, 3 12 and 4. Definitions for certain other values of i (and
a great deal more) may be found online in the web sites, the paper and the reference to Munkres
listed below:
en.wikipedia.org/wiki/Separation axiom — This list is pretty comprehensive.
at.yorku.ca/i/d/e/b75.htm — This points to electronic copies of a paper, “Definition
bank” in general topology, by G. V. Nagalagi, which can be downloaded in pdf format.
The listing of properties Ti for i < 1 is particularly extensive.
The paper, Espaces T1 21 , by Carlos A. Infantozzi [Proceedings of the International Sym-
posium on Topology and its Applications (Budva, 1972), pp. 116–122; Savez Društava
Mat. Fiz. i Astronom., Belgrade, 1973], deals with the case i = 1 12 .
Exercise 6(b) on page 213 of Munkres describes a natural candidate for T 6 .
To answer obvious questions about the choice of terminology, the symbolism T i comes from
the German word Trennung, which means separation (the terms were introduced in the classic book
of Alexandroff and Hopf on topology, which was published in the nineteen thirties and written in
German). In any case, here are the most important of the separation properties for this course:
Definitions. (T0) A topological space X is said to be a T 0 space (or a Kolmogorov space) if for
each pair of points x, y ∈ X there is an open set containing one but not the other.

98
 (T1) A topological space X is said to be a T 1 space (or a Fréchet space) if for each x ∈ X
the one point set {x} is closed in X.
Warning. The name “Fréchet space” is more frequently used for another type of object
which arises in functional analysis (see Rudin’s book on that topic for further information).
(T2) A topological space X is said to be a T 2 space if it has the Hausdorff Separation Property.
(T3) A topological space X is said to be a T 3 space if it is a T1 space and is also regular:
Given a point x ∈ X and an open set U containing x, there is an open subset V such that

x∈ V ⊂ V ⊂ U

or equivalently that if x ∈ X and F is a closed subset not containing X, then there are disjoint
open subsets V and W such that x ∈ V and F ⊂ W .
(T3 21 ) A topological space X is said to be a T 3 12 space (or a Tychonoff space) if it is a T 1
space and is also completely regular: Given a point x ∈ X and a closed subset F ⊂ X not containing
x, there is a continuous function f : X → [0, 1] such that f (x) = 0 and f = 1 on F .
(T4) A topological space X is said to be a T 4 space if it is a T1 space and it is also normal:
Given a closed subset E ⊂ X and an open set U containing x, there is an open subset V such that

E⊂ V ⊂ V ⊂U

or equivalently that if E and F are disjoint closed subsets of X, then there are disjoint open subsets
V and W such that E ⊂ V and F ⊂ W .
Most of the implications
if i > j , then Ti =⇒ Tj
are clear or have already been established, the main exceptions involving the case i = 3 21 . To see
that T3 21 implies T3 , given x and F let f be the continuous function and take V and W to be
   
f −1 [0, 21 ) and f −1 ( 12 , 1] respectively. The proof that T4 implies T3 12 is considerably more
difficult and in fact relies on the following deep result:
THEOREM 1. (Urysohn’s Lemma) If X is a T 1 space, then X is T4 if and only if for each
pair of disjoint closed subspaces E, F ⊂ X there is a continuous function f : X → [0, 1] such that
f = 0 on E and f = 1 on F .
We shall not need the result in this generality (a simple proof for metric spaces is given below);
Section 33 of Munkres (pages 207–212) gives a detailed proof.
We can now prove that T4 implies T3 12 as follows: If X is T5 , let F be a closed subset of
X, and suppose that x ∈ F . Then {x} is closed in X, and hence by Urysohn’s Lemma there is a
continuous real valued function f on X such that f = 0 on {x} ( equivalently, f (x) = 0) and f = 1
on F , showing that X is T3 12 .
There are two things one would like for the preceding list of separation properties. First of all,
every metric space should be a T4 space. Second, to avoid redundancies one would like to know
that if i > j then Ti and Tj are not logically equivalent; i.e., there is an example of a topological
space that is a Tj space but not a Ti space. Some examples appear in Munkres, and there are
many other examples of this sort in the book by Steen and Seebach. The proof that metric spaces
are T4 turns out to be fairly straightforward.

99
PROPOSITION 2. Every metric space is a T 4 space.
Proof. We have already shown that metric spaces are T 1 (and even T2 ). To prove the normality
condition of Urysohn’s Lemma involving continuous functions, consider the function

d(x, E)
f (x) = .
d(x, E) + d(x, F )

We know that the distance functions in the formula are continuous, so the formula will define a
continuous real valued function if the denominator is nonzero. But since E and F are disjoint
it follows that for each x ∈ X either x 6∈ E or x 6∈ F is true (and maybe both are true). This
means that at least one of the numbers d(x, E), d(x, F ) is positive and hence their sum is always
positive. Since the numerator is nonnegative and less than or equal to the denominator, it follows
that f (x) ∈ [0, 1] for all x ∈ X. If x ∈ E then d(x, E) = 0 and therefore f (x) = 0, while if x ∈ F
then d(x, F ) = 0 and therefore f (x) = 1.

Compactness and separation axioms

The following result has appeared on many examinations. When combined with Urysohn’s
Lemma it provides a powerful means for constructing continuous real valued functions on compact
Hausdorff spaces with far-reaching consequences, particularly in functional analysis. Stronger forms
of this result exist for spaces that are T 2 and paracompact (a condition implied by compactness
defined in Section 6; see Section 41 on pages 252–260 of Munkres, and particularly see Theorem
41.1 on pages 253–254 for an analog of the theorem below). The result for paracompactness also
has far-reaching consequences in topology and differential geometry.
THEOREM 3. If a topological space is compact and T 2 , it is also T4 .
Proof. We shall give a quick proof that uses Wallace’s Theorem. The more traditional proof is
given by combining Lemma 26.4 on page 166 of Munkres (with the reasoning given on the previous
page), which shows that compact and T 2 implies T3 , with Theorem 32.3 on page 202 of Munkres,
which shows that compact and T3 implies T4 .
Recall that a space X is Hausdorff if and only if the diagonal ∆ X is closed in X ×X. If A and B
are disjoint subsets of a set S, it is immediate that A∩B = ∅ is true if and only if (A×B)∩∆ X = ∅.
Combining these observations, we see that if E and F are disjoint closed subspaces of a Hausdorff
space X then E × F is contained in the open set X × X − ∆ X .
If X is compact, then so are E and F , and therefore by Wallace’s Theorem it follows that
there are open subsets U, V ⊂ X such that

E×F ⊂ U ×V ⊂ X × X − ∆X .

This means that U and V are disjoint open subsets containing E and F respectively.

References for examples

As noted before, if i < j it is not always easy to find examples of topological spaces that are T i
but not Tj , and the examples in Munkres are spread out over Sections 31 through 33. Therefore
we shall give an index to those examples here.

100
 When analyzing the logical relations among separation properties it is often useful to consider
the following related questions:
(1) If A is a subspace of a Ti space, is A also a Ti space?
(2) If X and Y are Ti spaces, is their product X × Y also a T i space?
In particular, if i < j and either (1) or (2) is true for T i spaces but not for Tj spaces, then it
follows that Ti is strictly weaker than Tj . Specifically, the following answers to (1) and (2) show
that Ti is strictly weaker than T4 if i < 4:
FACT 4. The answers to both (1) and (2) are positive for T i spaces if i < 4 and negative for
T4 spaces.
Given that the proofs for i < 4 are fairly direct, the failures of these results to hold for T 4 are
a bit surprising at first. However, they become less surprising in light of Urysohn’s Lemma, which
has many far-reaching implications (compare the first paragraph of Section 32 in Munkres).
It is relatively straightforward to show that the answers to both (1) and (2) are positive for
T0 and T1 , and the proofs are left to the reader as exercises. It is also straightforward to produce
examples to show that T0 does not imply T1 and T1 does not imply T2 . In the first case, one can
use the Sierpiński space whose underlying set is {0, 1} and whose open subsets are the empty set,
the set itself and {0}. In the second case one can use the finite complement topology on an infinite
set. We shall now list the references to Munkres for the remaining cases.
FACT 5. Subspaces and products of Ti spaces are Ti if i = 1, 2, 3, 3 12 .
The first two cases are treated in Theorem 31,2 on pages 196–197 of Munkres, and the other
case is treated in Theorem 33.2 on pages 211-212 of Munkres.
FACT 6. Subspaces and products of T4 spaces are not necessarily T4 . — The reference
here is a combination of Theorem 32.1 and Example 2 on pages 202–204 of Munkres.
FACT 7. T2 does not imply T3 . — The reference is Example 1 on pages 197–198 of
Munkres.
FACT 8. T3 does not imply T3 21 . — The reference is Exercise 11 on pages 214 of Munkres.
FACT 9. T3 21 does not imply T4 . — One can extract this from Example 2 on pages
203–204 either by taking a subspace of a T 4 space that is not T4 or by taking a product of T4
spaces that is not T4 . Since T4 implies T3 12 and this property is preserved under taking subspaces
and products, it follows that the spaces in the given example are T 3 21 but not T4 .
As we have already noted, the book by Steen and Seebach is a comprehensive summary of
many further results and examples for the logical interrelationships of various special properties of
topological spaces.
SEPARATION PROPERTIES AND QUOTIENT SPACES. We have already noted that quotients
of Hausdorff spaces need not be Hausdorff. Some conditions under which separation properties
are preserved under taking quotients are given in Section 37 of Munkres, Elements of Algebraic
Topology. Another excellent reference is Section 5 in Appendix A to Massey, Algebraic Topology:
An Introduction (CAUTION: There is a somewhat different book by the same author with the very
similar title, A Basic Course in Algebraic Topology).

101
 Non-Hausdorff topologies

As noted in Munkres and numerous other references, the addition of the Hausdorff Separation
Property to the axioms for a topological space yields a class of objects that are more general than
metric spaces but are relatively closed to one’s geometric intuition in many ways. One additional
motivation is that most of the spaces that are important to mathematicians satisfy the Hausdorff
Separation Property. In some respects the examples of non-Hausdorff spaces that one sees in an
introductory topology course may be viewed as instructive, showing that non-Hausdorff spaces
often have very strange properties and pointing out that in some cases this property does not
automatically hold, even if one makes simple constructions starting with Hausdorff spaces.
However, there are mathematical situations in which non-Hausdorff spaces arise, and in some
branches of mathematics these examples turn out to be extremely important. One particularly
noteworthy class of examples is given by the Zariski topologies from algebraic geometry. Here
is the basic idea in the most fundamental cases: Let k be an algebraically closed field (every
nonconstant polynomial factors completely into a product of linear polynomials, as in the complex
numbers). A set A ⊂ kn is said to be Zariski closed if A is the set of solutions for some finite
system or polynomial equations in n variables (where the coefficients lie in k). It is an easy exercise
in algebra to show that the family of all such subsets satisfies the conditions for closed subsets of
a topological space, and the details are worked out in the following online document:
http://math.ucr.edu/∼res/math145A-2014/zariski-topology.pdf
Strictly speaking, the cited document consideres on the cases where k is the real or complex
numbers, but the same argument works for an arbitrary field (however, at this level of generality
we usually do not have a second topology defined by a metric on the field). The resulting topological
structure on kn is called the Zariski topology. Other objects in algebraic geometry admit similar
notions of Zariski topologies, but the latter quickly reach well beyond the scope of this course. If
n = 1 then the Zariski topology on k is the finite complement topology; since algebraically closed
fields are always infinite (this follows immediately from the theory of finite fields), the Zariski
topology on k is not Hausdorff although it is T 1 . Similarly, most other Zariski topologies are not
Hausdorff (and not necessarily even T 1 ). Two of the exercises for earlier sections (one on irreducible
spaces, one on noetherian spaces) contain elementary results that arise naturally when one works
with Zariski topologies; in fact, k n with the Zariski topology is both irreducible and noetherian.
The introduction of these topological structures in the nineteen forties was an elementary but
far-reaching step in formulating the present day mathematical foundations for algebraic geometry.
Dieudonné’s book on the history of algebraic geometry provides some further information on these
points.
During the past 25 to 30 years, non-Hausdorff topological spaces have also been used in certain
areas of theoretical computer science. Although many of the basic ideas in such studies come from
topology and branches of the “foundations of mathematics,” the basic structures of the work and
its goals differ substantially from those of the traditional core of mathematics, and the relevant
spaces are much less geometrically intuitive than the Zariski topologies.
Two introductory references for this material are Section 3.4 of the Book, Practical Foundations
of Mathematics, by P. Taylor, and a survey article by M. W. Mislove, Topology, Domain Theory
and Theoretical Computer Science (Topology Atlas Preprint #181), which is available online at
http://at.yorku.ca/p/a/a/z/15.htm∼. A survey article on this topic is cited in the following
online document:
http://math.ucr.edu/∼res/math145A-2014/intro2topA-11a.pdf

102
The book Domains and Lambda Calculi, by R. Amadio and P.-L. Curien, presents this material
specifically in connection with its applications to topics in computer science. More elementary
discussions of finite non-Hausdorff topological spaces are given in the following online references:
http://math.uchicago.edu/∼may/MISC/FiniteSpaces.pdf
http://en.wikipedia.org/wiki/Finite topological space

Counting topologies on finite sets.

The online files

http://math.ucr.edu/∼res/math145A-2014/intro2topA-07a.pdf
http://math.ucr.edu/∼res/math145A-2014/intro2topA-12b.pdf
enumerate the topologies on sets with two and three elements, and they also look at two refine-
ments of this question; specifically, these include the enumeration of homeomorphism types of such
topologies (so that two topologies on {1, · · · , n} are identified if one is obtained from the other by
a permutation of {1, · · · , n}), and enumeration of those topologies or homeomorphism types given
by connected topological spaces (every finite space is homeomorphic to a disjoint union of finitely
many open, closed and connected subsets).
We shall conclude this section by noting (without references) that a crude upper bound for
2
the number of topologies on the set {1, · · · , n} is given by 2 n −n .

103
 VI.4 : Local compactness and compactifications

(Munkres, §§ 29, 37, 38)

As in the case of connectedness, it is often useful to have a variant of compactness that reflects
a basic property of open subsets in Euclidean spaces: Specifically, if U is such a set and x ∈ U ,
then one can find an ε > 0 such that the closure of N ε (x) is compact; in fact if we choose δ > 0
such that Nδ (x) ⊂ U , then the compact closure property will hold for all ε such that ε < δ.
Definitions. A topological space X is said to be locally compact in the weak sense if for each
x ∈ X there is a compact subset N such that x ∈ Int(N ) ⊂ N .
A topological space X is said to be locally compact in the strong sense if for each x ∈ X and
each open subset U containing x there is a compact subset N such that x ∈ Int(N ) ⊂ N ⊂ U .
Clearly the second condition implies the first, but in many important cases these concepts are
equivalent:
PROPOSITION 1. If X is a Hausdorff space, then X is locally compact in the weak sense if
and only if X is locally compact in the strong sense. Furthermore, in this case if x ∈ X and U is an
open set containing x, then there is an open set W such that x ∈ W ⊂ W ⊂ U and W is compact.
Proof. It is only necessary to show that the weak sense implies the strong sense and that the
additional condition holds in this case. Suppose that X is locally compact in the weak sense and that
x ∈ U where U is open in X. Let N be the compact set described above, and let V = Int(N ) ∩ U .
Then x ∈ V where V is open in N . Since the latter is regular (it is compact Hausdorff), there is
an open subset W ⊂ V such that

x∈W ⊂ Closure(W, N ) ⊂ V .

The set W is in fact open in X (because W = N ∩W 0 where W 0 is open in X and W = W ∩V = V ∩

W 0 since W ⊂ V ⊂ N ), Furthermore, since N is closed in X it follows that Closure(W, N ) = W ∩N
must be equal to W .
Not all subspaces of a locally compact Hausdorff space are locally compact. For example, the
set of all rational numbers in the real line is not locally compact. (Proof: Suppose that a ∈ Q and
that B ⊂ Q is an open subset in the subspace topology such that B ∩ Q is compact. Without loss of
generality we may assume that B is an open interval centered at a. The compactness assumption
on the closure implies that B ∩ Q is in fact a compact, hence closed and bounded, subset of the
real line. This is impossible since there are many irrational numbers that are limit points of B.).
However, a large number of interesting subspaces are locally compact.
PROPOSITION 2. If X is a locally compact Hausdorff space in the strong sense and Y is
either an open or a closed subset of X, then Y is locally compact (and Hausdorff).
Proof. The proof for open subsets follows because if Y is open in X and U is open in Y , then U
is open in X; one can then use the strong form of local compactness to prove the existence of an
open subset of U with the required properties.
Suppose now that Y is closed, let y ∈ Y , and let U be an open subset of Y containing y. Write
U = Y ∩ U1 where U1 is open in X. Then there is an open set W 1 in X such that

x ∈ W 1 ⊂ W1 ⊂ U1

104
and W1 is compact. Let W = W1 ∩ Y ; then

x ∈ W ⊂ W1 ∩ Y ⊂ U

where W1 ∩ Y is compact because it is closed in X and contained in the compact subspace W 1 .

Since
Closure(W, Y ) = W ∩ Y ⊂ W1 ∩ Y
it follows that Closure(W, Y ) is compact and that

x ∈ W ⊂ Closure(W, Y ) ⊂ W1 ∩ Y ⊂U

and therefore Y is locally compact in the strong sense.

COROLLARY 3. If X is locally compact Hausdorff and B = U ∩ F , where U ⊂ X is open and
F ⊂ X is closed, then B is locally compact.
Proof. By the proposition we know that F is locally compact Hausdorff. Since B = U ∩ F is
open in F , the proposition then implies the same conclusion for B.
COROLLARY 4. If U is open in a compact Hausdorff space, then U is locally compact.
Proof. A compact Hausdorff space is clearly locally compact in the weak sense and hence locally
compact in the strong sense. Therefore U must also be locally compact in the strong sense.
COROLLARY 5. If X is a locally compact Hausdorff space, then X is T 3 .
The defining conditions for locally compact Hausdorff in the strong sense imply that such a
space is regular. In fact, one can go further using Urysohn’s Lemma to prove that a locally compact
Hausdorff space is in fact completely regular (hence it is T 3 12 ).
Note. A locally compact Hausdorff space is not necessarily T 4 . In particular, Example 2 on
pages 203–204 of Munkres actually describes an OPEN subset of a compact Hausdorff space that
is not T4 , and by the proposition above this subset is locally compact Hausdorff.

Compactifications of noncompact spaces

Frequently in mathematics it is helpful to add points at infinity to a mathematical system to

deal with exceptional cases. For example, when dealing with limits in single variable calculus this
can be done using an extended real number system that consists of the real line together with two
additional points called ±∞. In some other cases, it is preferable to add only a single point at
infinity; for example, this is necessary if one wants to have something equal to

1
lim
x→0 x
and in the theory of functions of a complex variable it is also natural to have only one point at
infinity.
In other situations it is desirable to add many different points at infinity. Projective geometry
is perhaps the most basic example. In this subject one wants to add a point at infinity to each line
in such a way that two lines are parallel if and only if their associated extended lines contain the
same point at infinity. This turns out to be useful for many reasons; in particular, it allows one to

105
state certain results in a uniform manner without detailed and often lengthy lists of special cases
(however, you do not actually get something for nothing — one must first invest effort into the
construction of points at infinity in order to obtain simplified arguments and conclusions).
In all these cases and many others, one basic property of the enriched spaces with added
points at infinity is that these enriched spaces are compact and contain the original spaces as dense
subspaces. In order to simplify the discussion but include the most interesting examples, we shall
only consider (original and enriched) spaces that are Hausdorff.
Definition. If X is a topological space, then a compactification of X is a pair (Y, f ) where Y
is compact and f : X → Y is a continuous map that is a homeomorphism onto a dense subspace.
Two compactifications (Y, f ) and (Z, g) are equivalent if there is a homeomorphism h : Y → Z such
that h o f = g.
The compactification (Y, f ) is said to dominate the compactification (Z, g) if there is a contin-
uous map h : Y → Z such that h o f = g, and in this case we write (Y, f ) ≥ (Z, g).
There is also a corresponding notion of abstract closure in which there is no compactness
assumption on Y , and one can define equivalence and domination in a parallel manner. Given an
abstract closure (Y, f ) of X, its residual set is the subset Y − f [X].
Here are some basic properties of Hausdorff compactifications and abstract closures.
PROPOSITION 6. (i) If (Y, f ) and (Z, g) are Hausdorff compactifications or abstract closures
of the same Hausdorff space X, then there is at most one h : Y → Z such that h o f = g.
(ii) If (Y, f ) and (Z, g) are Hausdorff compactifications such that (Y, f ) ≥ (Z, g) and h is the
continuous map defining the domination, then h is onto.
(iii) There is a set of equivalence classes of Hausdorff compactifications or abstract closures of a
Hausdorff spaces such that every Hausdorff compactification of X is equivalent to a compactification
in that set.
(iv) The relation of domination makes the equivalence classes of Hausdorff compactifications
or abstract closures into a partially ordered set.
Proof. Proof of (i). Let h : Y → Z be a homeomorphism such that h o f = g. If h0 is another
such map then the restrictions of h and h 0 to f [X] are equal. Since f [X] is dense and the set of
points where two functions into a Hausdorff space are equal is a closed subset of the domain, it
follows that this subset is all of Y and thus h = h 0 everywhere.
Proof of (ii). By construction the image of h contains the dense subspace g[X], and since
h[Y ] is compact and Z is Hausdorff it also follows that h[Y ] is closed. Therefore we must have
h[Y ] = Z.
Note that the analogous result for abstract closures is false; by definition the identity map on
a Hausdorff space X is an abstract closure in the sense of the definition, and in general there are
many abstract closures (Y, f ) such that f [X] 6= Y .
Proof of (iii). We need to introduce some set-theoretic notation. Given a set S and a
nonempty family of subsets M ⊂ P(S), a family A ⊂ P(S) is called a filter provided
[a] if B ∈ A, B ⊂ C and C ∈ M then C ∈ A,
[b] if B ∈ A and C ⊂ A then B ∩ C ∈ A.
Formally, this concept is dual to the concept of an ideal in a Boolean algebra where union
and intersection are interpreted as addition and multiplication, but for our purposes the important

106
point is that in a topological space X the set N x of all open subsets containing a given point x ∈ X
is a filter.
Given a family of subsets B ⊂ M that is closed under finite intersections, the smallest filter
B ∗ containing B (or the filter generated by B) consists of B together with all subsets containing
some element of B.
If X is a Hausdorff topological space then F(X) will denote the set of all filters of open subsets
in X. For Hausdorff spaces it is immediate that p 6= q implies N p 6= Nq . Given a 1–1 continuous
map g from a Hausdorff space X to another Hausdorff space Y , there is an associated map f ∗ from
Y to F(X) that sends y to the filter generated by f ∗−1 [Ny ].
We claim that f ∗ is 1–1 provided f is 1–1, Y is Hausdorff and f [X] is dense in Y . Suppose
that u, v ∈ Y . Then there are disjoint open subset U, V ∈ Y such that u ∈ U and v ∈ V It suffices
to show that f −1 [U ] ∈ f ∗ (u) − f ∗ (v) and f −1 [V ] ∈ f ∗ (v) − f ∗ (u). In fact, if we can prove the first,
then we can obtain a proof of the second by reversing the roles of u and v and U and V throughout
the argument.
By construction we have f −1 [U ] ∈ f ∗ (u) so verification of the claim reduces to showing that
f [U ] 6∈ f ∗ (v). Assume that we do have f −1 [U ] ∈ f ∗ (v). Then there is some open set W in Y
−1

containing v such that f −1 [W ] ⊂ f −1 [U ], and it follows that we also have f −1 [W ∩ V ] ⊂ f −1 [U ].

Since f −1 [U ] ∩ f −1 [V ] = f −1 [U ∩ V ] = ∅ this means that f −1 [W ∩ V ] must be empty. To see that
this is impossible, note that f [X] ∩ W ∩ V 6= ∅ because v ∈ W ∩ V and f [X] is dense in Y .
It follows that Y can be identified with a subset of F(X). By construction the latter is
a
subset of P(X) and therefore the number of points in Y is at most the cardinality of P P(X) .
Since each topology on Y is a family of subsets of Y it follows that there is a specific bound on the
cardinality of equivalence classes of Hausdorff spaces that contain a homeomorphic copy of X as a
dense subspace.
Proof of (iv). Reflexivity is trivial (take the identity map on the compact space) and tran-
sitivity is also trivial (take the composite of the mappings on the compact spaces). Suppose now
that (Y, f ) ≥ (Z, g) and vice versa. Then there are continuous maps h : Y → Z and k : Z → Y
such that g = h o f and f = k o g. Therefore we have f = k o h o f and g = h o k o g, so that k o h and
idY agree on the dense subset f [X] and h o k and idZ agree on the dense subset g[X]. These imply
that h and k are inverses to each other and hence that the two compactifications are equivalent.

Examples of compactifications

In general there are many compactifications of a Hausdorff space. For example, there are many
closed bounded subsets of the plane that contain open dense subsets homeomorphic to R 2 ∼ = (0, 1)2
(these include solid rectangles with an arbitrary finite number of open holes removed; see the
file swisscheese.pdf for more about this). Furthermore, a Hausdorff compactification does not
necessarily inherit certain “good” properties of the original space; in particular, Example 3 on page
238 of Munkres shows that a compactification of the real line is not necessarily locally connected
or path connected.
In view of the examples in the previous paragraph and many others, it seems advisable to
begin with simple questions about the structure of the partially ordered set of equivalence classes
of Hausdorff compactifications. Two obvious questions are whether this set has maximal or minimal

107
elements. It turns out that every T3 12 space has a maximal Hausdorff compactification (the Stone-
Čech compactification) that is unique up to equivalence (see stone-disamb.pdf for comments on
mathematicians named Stone; the reference here is to M. H. Stone). This object is constructed
directly in Section 38 of Munkres, and several of its important properties are also described there.
Note. Since a compact T2 space is T4 and every subspace of a T3 12 space is again T3 21 , it
follows that a topological space has a Hausdorff compactification if and only if it is T 3 12 .

One difficulty with the Stone-Čech compactification is that for most noncompact spaces its
structure is not well understood. In particular, this applies to the Stone- Čech compactification of
the nonnegative integers N with the discrete topology, which has been the object of considerable
study. Some further information appears in the book by Walker and the Wikipedia reference given
below (see the file aabInternetresources.pdf for general remarks about Wikipedia references on
mathematical topics).
http://en.wikipedia.org/Stone%C4%8Cech compactification
A closely related compactification, which exists for arbitrary T 1 spaces, is the Wallman (or
Wallman-Shanin) compactification. Further information on this construction can be found in older
texts such as Kelley or Dugundji, the paper by Shah and Das listed in the bibliography, of the article
Banaschewski compactification on pp. 83–84 of the encyclopedia volume edited by M. Hazewinkel.

The Alexandroff one point compactification

We shall now consider the other extreme and consider minimal types of compactifications
f : X → Y where the added (or residual) set Y − f (X) consists of a single point. If X = R n there
is a standard and important visualization of this compactification; the space Y turns out to be
homeomorphic to the n-dimensional unit sphere in R n+1 .
Note that if a space X has a Hausdorff compactification (Y, f ) such that Y − f (X) is a single
point (or more generally a closed subset!) then by a corollary stated above the space X must be
locally compact. This observation will explain our assumption in the basic construction below.
Definition. Let X be a locally compact Hausdorff topological space that is not compact, and let
∞ be some point not in X (for example, in the standard axiomatic model for set theory one has
X 6∈ X so we can take ∞ = {X}). The one point or Alexandroff compactification of X is the set
X • = X ∪ {∞} with open sets given as follows:
(1) Bounded open sets that are open subsets of X itself.
(2) Open neighborhoods of ∞ that are unions {∞} ∪ X − K where K is a compact subset of
X.
It is necessary to verify that this family of sets defines a topology on X. The empty set is
open in X • because it is open in X, and X • is open because it is equal to {∞} ∪ X − ∅ and
∅ is compact. Suppose that we have a family of open sets in X • , and split it into the bounded
open sets Uα and the open neighborhoods of infinity {∞} ∪ X − K β where Kβ is compact. Some
elementary set-theoretic manipulation shows that the union of this family is either the bounded
open set ∩α Uα if there are no open neighborhoods of infinity in the family or else it is
[

{∞} X − K−U

108
where U = ∪α Uα and K = ∩β Kβ (note that an arbitrary intersection of compact subsets in a
Hausdorff space is compact). The details of the set-theoretic algebra are described on the top of
page 184 of Munkres. The latter also gives the arguments to verify that the family of open subsets
defined above is closed under (finite) intersection.
CLAIM 7. The inclusion of X in X • is a compactification.
Indication of proof. We need to show that the inclusion of X in X • is 1–1 continuous and open
(this will show it is a homeomorphism onto its image), that the image of X is dense in X • and that
X • is compact Hausdorff. A proof that X • is compact Hausdorff appears near the bottom of page
184 of Munkres.
To see that X is dense in X • , it suffices to verify that ∞ is a limit point of X. Let
[

{∞} X −K

(with K compact) be an open set containing the point at infinity. Since X is noncompact we must
have X − K 6= ∅, and this proves the limit point assertion.
By construction the inclusion map from X to X • is 1–1 and open; we need to show it is also
continuous. This follows because the inverse image of a finite open set is just the set itself, and the
latter is open in X by construction, while the inverse image of a open neighborhood of infinity has
the form X − K where K is compact; since compact subsets are closed it follows that the inverse
image X − K is open in X.
The uniqueness of this one point compactification is given by the following result:
PROPOSITION 8. If X is locally compact Hausdorff and (Y, f ) is a Hausdorff compactification
such that Y − f [X] is a single point, then there is a unique homeomorphism h : X • → Y such that
h|X = f .
The proof of this is given as Step 1 on the bottom of page 183 in Munkres.

One point compactifications of Euclidean spaces

Since the spaces Rn are some of the most fundamental examples of locally compact spaces
that are not compact, it is natural to ask if some additional insight into the nature of the one point
compactification can be obtained in these cases. The following result shows that the one point
compactification of Rn is homeomorphic to the n-dimensional sphere

S n = { x ∈ Rn+1 | |x|2 = 1 } .

PROPOSITION 9. Let en+1 ∈ Rn+1 be the unit vector whose last coordinate is 1 (and
whose other coordinates are zero). Then there is a canonical homeomorphism from the subspace
S n − { en+1 } to Rn .
Sketch of proof. The homeomorphism is defined by stereographic projection, whose physical
realization is the polar projection map of the earth centered at the south pole. Mathematically this
is given as follows: View Rn as the linear subspace spanned by the first n unit vectors, and given

109
a point v ∈ S n such that v 6= en+1 let w be the unique point of Rn such that w − en+1 lies on the
straight line joining en+1 to v. The explicit formula for this map is

2

f (v) = 2 en+1 + · v − en+1
1 − hv, en+1 i

and illustrations of this appear in the files stereopic2.∗ in the course directory.
The stereographic projection map f is continuous by the formula given above, and elementary
considerations from Euclidean geometry show that this map defines a 1–1 correspondence between
S n − { en+1 } and Rn . In order to give a rigorous proof that f is a homeomorphism, it suffices to
verify that the map
g : Rn −→ S n − { en+1 }
defined by the formula

4

g(w) = en+1 = · w − 2 en+1
|w|2 +4

is an inverse to f ; i.e., we have g(f (v)) = v and f (g(w)) = w for all v ∈ S n − { en+1 } and w ∈ Rn .
In principle the verification of these formulas is entirely elementary, but the details are tedious and
therefore omitted.
The following important geometrical property of stereographic projection was essentially first
discovered long ago by Hipparchus of Rhodes (c. 190 B.C.E. — c. 120 B.C.E.):
THEOREM 10. (Conformal mapping property) Let α, β : [0, 1] → R n be differentiable curves
with α(0) = β(0) and α0 (0), β 0 (0) 6= 0. Then the image curves g o α and g o β in S n satisfy the
conditions g o α(0) = g o β(0) and [g o α]0 (0), [g o β]0 (0) 6= 0, and



angle α0 (0), β 0 (0) = angle [g o α]0 (0), [g o β]0 (0) .

This result will be established in Appendix B.

Locally connected compactifications

Although topological spaces exist in great variety and can exhibit strikingly orig-
inal properties, the main concern of topology has generally been the study of
spaces which are relatively well-behaved.
RS, Some recent results on topological manifolds, Amer. Math. Monthly 78
(1971), 941–952.

Most (but definitely not all) of the topological spaces which arise in geometry or analysis are
either locally connected or explicitly realizable as subsets of locally connected spaces. Therefore
it is natural to expect that there are useful examples of compactifications for which both the
original space and its compactification are locally connected. The following online site lists several
fundamentally important constructions of this type:

110
 http://en.wikipedia.org/wiki/Compactification (mathematics)
We shall only discuss one example of this sort here; namely, the (Freudenthal) end point
compactification. Probably the best way to motivate this construction is to begin with a basic
difference between the noncompactness properties of the real line and the coordinate plane.
Let n ≥ 2 be an integer, and let K ⊂ Rn be a compact subset. Then there is a compact
subset L ⊃ K such that Rn − L is connected. On the other hand, if K ⊂ R is compact,
then for all compact sets L ⊃ K the complement R − L is disconnected, but one can
choose L so that the complement has exactly two connected components.
This corresponds to a fundamental difference in “limits at infinity” between R and R n for
n ≥ 2. Namely, one cannot discuss limit expressions involving +∞ or −∞ for functions if the
relevant variables are not 1-dimensional, but it is still meaningful to discuss limits involving ∞
with no sign attached.
There are also natural examples where one can go to infinity in any number of different ways.
For example, if x denotes the union of the coordinate axes in R n , then there are 2n distinct ways
of doing so. The end point compactification is essentially a compactification in which each such
method of going to infinity corresponds to a different point. More precisely, if we are given a space
which is connected and locally connected, then the end point compactification is a compact space
with the same properties such that the residual set is totally disconnected (the topology has a base
given by sets that are both open and closed). The precise definitions are given at the following
online site:
http:/en.wikipedia.org/wiki/End (topology)
Further information on this construction appears in Chapter VI of the book by Isbell and the
papers by F. Raymond and L. Siebenmann which are listed in the bibliography.

111
 VI.5 : Metrization theorems

(Munkres, §§ 39, 40, 41, 42)

As noted before, it is natural to ask for necessary and sufficient conditions that a topology on
a space comes from a metric. Most point set topology texts, including Munkres, cover this material
in considerable detail.
Our approach here is somewhat different; namely, we want to show that compact Hausdorff
spaces built out of compact subsets of finite-dimensional Euclidean spaces are also homeomorphic to
subsets of such spaces. Examples of such objects arise in many geometric and topological contexts.
Our results deal with subsets of finite-dimensional Euclidean spaces and the general metrization
results involve finding homeomorphic copies of a space in various infinite-dimensional spaces, so the
results given here are not actually contained in the more general ones that are stated and proved
in Munkres.
Since a metric is by definition a real valued continuous function on X × X, it is not surprising
that the proofs of metrization theorems rely heavily on constructing continuous real valued functions
on a space. Therefore it is necessary to begin with results about continuous functions on compact
metric spaces.

Constructions for continuous functions

The basis for all these constructions is Urysohn’s Lemma, which is true for arbitrary T 4 spaces
and was verified earlier in the notes for metric spaces: Given two nonempty disjoint closed subsets
E and F in X, there is a continuous function f : X → [0, 1] such that f = 0 on E and f = 1 on
F.
Notation. A Hausdorff space will be called a UL-space if Urysohn’s Lemma is true. We have
already noted that this is equivalent to the T 4 condition and that every metric space is a UL-space.
The conclusion of the following basic result is in fact logically equivalent to Urysohn’s Lemma:
THEOREM 1. (Tietze Extension Theorem) Let X be a UL-space, let A be a closed subset of
X and let f : A → [0, 1] be a continuous function. Then f extends to a continuous function from
all of X to [0, 1].
It is easy to prove a converse result that X is a UL-space if for every closed subset A ⊂ X and
every continuous function A → [0, 1] there is an extension to X, for if E and F are closed subsets
of X, then the function that is 0 on E and 1 on F is continuous and defined on a closed subset;
the extension to X shows that the UL-space condition is fulfilled.
Proof. Since the closed intervals [0, 1] and [−1, 1] are homeomorphic, we may as well replace the
former by the latter in the proof. The main idea of the proof is to construct a sequence of functions
{ ϕn } such that
lim ϕn |A = f
n→∞

in BC(A), and the basis for the recursive step in the construction is the following:
LEMMA  2. Let r > 0 and let h : A → [−r, r] be continuous. Then there is a continuous function
g : X → − 3 r, 13 r such that k g|A − h k≤ 32 r.
1

112
  1
  1  
Proof of Lemma 2. Let B = h−1 −r, − r
 1 1 
3 and C = h −1
3 r, r , so that B and C are
disjoint closed subsets of A. Take g : X → − 3 r, 3 r to be a continuous function such that g = − 31 r
on B and g = 13 r on C. The verification that

2
k g|A − h k ≤ r
3
separates into three cases depending upon whether a point a ∈ A lies in B, C or the set
h i
h−1 − 31 r, 31 r

(at least one of these must hold). In each case one can show directly that |g(a) − h(a)| ≤ 23 r.
Proof of Theorem 1 continued. Start off with f : A → [−1, 1], and apply the lemma to get
a function g1 : X → [− 31 , 13 ] with the properties stated in the lemma. Consider now the function
f1 = f −(g1 |A), which is a continuous function that takes values in [− 32 , 32 ]. Let g2 be the continuous
function on X associated to f1 as in the lemma, define f2 to be f1 − (g2 |A), and note that
 2 2
k f2 k≤ .
3
We can now continue recursively to define sequences of continuous real valued functions g n on X
and fn = fn−1 − (gn |A) such that
   n−1  n−1
1 2 2
k gn k ≤ · , k fn k ≤ .
3 3 3

Note that we have

n
X
fn = f − gn |A
i=1

for all positive integers n.

We want to define a continuous function G(x) by an infinite series
X
gn
n

and this will be possible if X

k gn k
n

converges. But the latter sum is dominated by the convergent series

 n−1
1 X 2
·
3 n
3

so there is no
Pproblem with constructing the continuous function G. To see that G|A = f , note
that G|A = n gn |A and for each n we have that
n
X  n
2
kf− (gi |A) k = k fn k ≤
i=1
3

113
 P
so that we also have f = n gn |A. Finally, G maps X into [−1, 1] because
 n−1
1 X 2
kGk ≤ · = 1.
3 n 3

COROLLARY 3. Let X be a UL-space, let A be a closed subset of X and let f : A → (−1, 1)

be a continuous function. Then f extends to a continuous function from all of X to (−1, 1).
Proof. By the theorem we have a continuous function G : X → [−1, 1] such that G = f on A.
We need to modify this function to something that still extends f but only takes values in (−1, 1).
Let D ⊂ X be G−1 [ {−1, 1} ]. By construction D and A are disjoint closed subsets, so there
is a continuous function k : X → [0, 1] that is 0 on D and 1 on A. If we set F equal to the product
of G and k, then it follows that F takes values in (−1, 1) and F |A = f .
COROLLARY 4. Let X be a UL-space, let A be a closed subset of X and let f : A → R n be
a continuous function. Then f extends to a continuous function from all of X to R n .
Proof. If n = 1 this follows from the previous corollary because (−1, 1) is homeomorphic to R
(specifically, take the map h(x) = x/(1 − |x|) ). If n ≥ 2 let f 1 , · · · , fn be the coordinate functions
of f , and let Fi be a continuous extension of fi for each i. If F is defined by the formula


F (x) = F1 (x), · · · , Fn (x)

then F |A = f .

Piecewise metrizable spaces

The following result provides a useful criterion for recognizing that certain compact Hausdorff
spaces that are homeomorphic to subsets of Euclidean spaces.
PROPOSITION 5. Let X be a compact Hausdorff space such that X is a union of closed
subsets A ∪ B, where A and B are homeomorphic to subsets of some finite-dimensional Euclidean
space(s). Then X is also homeomorphic to a subset of some finite-dimensional Euclidean space.
Proof. We may as well assume that both A and B are homeomorphic to subsets of the same Eu-
clidean space Rn (take the larger of the dimensions of the spaces containing A and B respectively).
Let f : A → Rn and g : B → Rn be 1–1 continuous mappings (hence homeomorphisms onto their
images).
Let F0 : B → Rn and G : A → Rn be continuous functions that extend f |A ∩ B and g|A ∩ B
respectively. We can then define F, G : X → R n by piecing together f and F0 on A and B in the
first case and by piecing together G0 and g on A and B in the second.
We shall also need two more continuous functions to construct a continuous embedding (a 1–1
continuous map that is a homeomorphism onto its image) on X = A ∪ B. Let α : X → R be
defined by dB (x, A ∩ B) on B and by 0 on A; this function is well defined and continuous because
the two definitions agree on A ∩ B. Note also that α(x) = 0 if and only if x ∈ A. Similarly, let
β : X → R be defined by dA (x, A ∩ B) on A and by 0 on B; as before, this function is well defined,
and furthermore β(x) = 0 if and only if x ∈ B.

114
 Define a continuous function

h:X −→ Rn × Rn × R × R ∼
= R2n+2

by the formula  
h(x) = F (x), G(x), α(x), β(x) .

By construction h is continuous, and since X is compact the map h will be an embedding if and
only if h is 1–1. Suppose that h(y) = h(z). If y ∈ A then we know that α(y) = 0 and therefore we
must also have α(z) = 0 so that y and z both belong to A. Likewise, if h(y) = h(z) and y ∈ B
then we must also have z ∈ B. We then also have F (y) = F (z) and G(y) = G(z). If y ∈ A, then
the fact that z also lies in A combines with the first equation to show that y = z, while if y ∈ B,
then the fact that z also lies in B combines with the first equation to show that y = z. In either
case we have that h(y) = h(z) implies y = z.
COROLLARY 6. If X is a compact Hausdorff space that is a finite union of the closed metrizable
subspaces Ai that are homeomorphic to subsets of some finite-dimensional Euclidean space, then X
is also homeomorphic to subsets of some finite-dimensional Euclidean space and hence metrizable.
In particular, this gives an alternate proof of Theorem 36.2 on pages 226–227 of Munkres (the
details are left to the reader, but here is a hint — show that the space in the theorem is a finite
union of subspaces homeomorphic to closed disks in R n ).
The next result is also useful for showing that certain compact Hausdorff spaces are homeo-
morphic to subsets of Euclidean spaces. Some preliminaries are needed.
Definition. If X and Y are topological spaces, and f : X → Y is continuous, then the mapping
cylinder Mf of f is the quotient of Y q (X × [0, 1]) modulo the equivalence relation generated by
the conditions

(x, 1) ∈ X × {1} ∼ f (x) ∈ Y for all x ∈X .

The equivalence classes of this relation are the one point sets { (x, t) } for t < 1 and the hybrid sets
 
{y} q f −1 {y} × {1} .

This space is a quotient of a compact space (the disjoint union of two compact spaces) and therefore
is compact.
PROPOSITION 7. If X and Y are homeomorphic to subspaces of R n for some n, then the
mapping cylinder Mf is also homeomorphic to a subspace of some Euclidean space.
Outline of proof. The details are left to the reader as an exercise, but the underlying idea is as
follows. Start out with embeddings α and β of X and Y in R n and construct a map

H : Mf −→ Rn × R × Rn ∼
= R2n+1


that is equal to 0, 0, β(y) on Y and given by
 
(1 − t)α(x), 1 − t, t · β(f (x))

on X × [0, 1]. This yields a well defined map on M f because it is consistent with the equivalence
relation, and the proof that h is a homeomorphism onto its image reduces to showing that h is 1–1;
the latter is an elementary exercise.

115
 Example. In algebraic topology one often encounters the following construction called
adjoining or attaching a k-cell: Given a space A and a continuous map f : S k−1 → A, we define

B = A ∪ f ek

to be the disjoint union of A and the disk D k modulo the equivalence relation generated by x ∈ S k−1
with f (x) ∈ A for all x. Let E ⊂ B be the set of all points that come from A ∪ { x ∈ D k | |x| ≥ 12 }
and let F be the image of 12 D k in B. Then E is homeomorphic to Mf , and then one can apply
both results above to show that B is homeomorphic to a subset of some Euclidean space if A is
homeomorphic to a compact subset of some Euclidean space.
A finite cell complex (also called a finite CW complex) is a compact Hausdorff space that is
obtained from a one point space by a finite number of attachments of k i -cells for varying values
of i. We allow the case where i = 0 with the conventions that S −1 = ∅ and adjoining a 0-cell is
simply the disjoint union of the original space with a one point space. The preceding results imply
that each finite cell complex is homeomorphic to a subset of some Euclidean space.

An alternate approach

The metrizability proofs given above are direct and complete, and they show that the examples
are in fact homeomorphic to subsets of ordinary finite dimensional Euclidean spaces. Another
approach to proving the metrizability of quotients of compact metrizable spaces is by means of the
Hausdorff metric on the family of closed subsets. The precise definitions and formal properties of
the Hausdorff metric are given in Exercises 7 and 8 on pages 280–281 of Munkres. As noted in part
(a) of Exercise 7, this metric makes the set of all closed subsets of a metric space X into another
metric space that Munkres denotes by H.
PROPOSITION 8. Let X be a compact metric space, and let Y be a T 1 quotient space of X.
Then Y is metrizable.
The two classes of examples above are homeomorphic to quotient spaces of compact metric
spaces (presented as disjoint unions of other compact metric spaces), and in each case every equiv-
alence class is a closed subset of the disjoint union. It follows that the quotient spaces are T 1
in this case and therefore the proposition implies metrizability of the space constructed from the
pieces (although it does not yield the embeddability in some Euclidean space if each piece is so
embeddable). The need for the T1 condition is illustrated by a variant of an earlier example: Take
X = [−1, 1] and consider the equivalence relation x ∼ y if and only if x and y are positive real
multiples of each other; as in Section V.1, the quotient space is a non-Hausdorff space consisting
of three points (in fact, this quotient space is not T 1 because the equivalence classes of −1 and +1
are not closed subsets).
Proof of Proposition 8. If Y is T1 then every equivalence class in X is a closed subset,
and therefore we have a map F : X → H sending x to f −1 [ {f (x)} ]. If we can show that F is
continuous, then the metrizability of Y may be established as follows: Let π : X → Y be the
quotient projection. Since π(u) = π(v) implies F (u) = F (v) there is a unique continuous map
g : Y → H such that F = g o π. By construction g is a 1–1 continuous map from the compact
space Y (recall it is the image of a compact space) to the metric space H, and therefore g is a
homeomorphism onto its image.
To verify that F is continuous,
let ε > 0, let D denote the Hausdorff metric on H. What does
it mean to say that D F (u), F (v) < ε? If one defines U (A, ε) ⊂ X as in Munkres to be the set

116
of all points whose distance from a subset A is less than ε, then the condition on the Hausdorff
metric is that F (u) ⊂ U ( F (v), ε) and F (v) ⊂ U ( F (u), ε). Suppose now that d(u, v) < ε where
as usual d denotes the original metric on X. Then the distance from u to F (v) is less than ε and
likewise the distance from v to F (u) is less than ε. Therefore we have F (u) ⊂ U ( F (v), ε) and
F (v) ⊂ U ( F (u), ε), and as noted above this implies D( F (u), F (v) ) < ε so that F is uniformly
continuous. We can now use the argument of the first paragraph of the proof to show that Y is
metrizable.

General metrization theorems

We begin with an early and powerful result on metrization. The proof is given in Theorem 34.1
on pages 215–218 of Munkres, with important preliminary material appearing in Theorem 20.5 on
pages 125–126 and Theorem 32.1 on pages 200–201.
THEOREM 9. (Urysohn Metrization Theorem) Let X be a second countable space. Then X is
homeomorphic to a metric space if and only if X is T 3 .
In fact, the argument shows that X is homeomorphic to a subspace of a compact metric space
if and only if X is T3 and second countable, for one constructs an embedding into a countable
product of copies of [0, 1] and the latter is compact by Tychonoff’s Theorem (as noted before, one
can also use Exercise 1 on page 280 of Munkres to prove compactness of the product). Some of the
key points in the embedding theorem are discussed in embeddings.pdf
The necessary and sufficient conditions for the metrizability of arbitrary topological spaces
require some additional concepts.
Definition. A family of subsets A = { A α } in a topological space X is said to be locally finite if
for each x ∈ X there is an open neighborhood U such that U ∩ A α 6= ∅ for only finitely many Aα .
Examples. Aside from finite families, perhaps the most basic examples of locally finite families
are given by the following families of subsets of R n .
(1) For each positive integer n let A n be the closed annulus consisting of all points x such
that n − 1 ≤ |x| ≤ n. Then for each y ∈ Rn the set N1/2 (y) only contains points from at
most two closed sets in the family (verify this!).
(2) For each positive integer n let V n be the open annulus consisting of all points x such that
n − 2 < |x| < n + 1. The details for this example are left to the reader as an exercise.
Locally finite families are useful in many contexts. For example, we have the following result:
PROPOSITION 10. If X is a topological space and A = { A α } is a locally finite family of
closed subsets (not necessarily finite), then ∪ α Aα is also closed.
Proofs of this and other basic results on locally finite families of subsets appear on pages 112
and 244–245 of Munkres.
Here is the ultimate result on metrization.
THEOREM 11. (Nagata-Smirnov Metrization Theorem) A topological space is metrizable if
and only if it is T3 and there is a base that is a countable union of locally finite families (also
known as a σ-locally finite base).
This is Theorem 40.3 on page 250 of Munkres, and the a proof including preliminary observa-
tions is contained in Section 40 on pages 248–252. Note that the Urysohn Metrization Theorem is

117
an immediate consequence of this result because a countable base is a countable union of families
such that each has exactly one element.

Paracompactness and metrization

A somewhat different metrization theorem due to YU. M. Smirnov is in some sense the ultimate
result on finding a metric on spaces built out of metrizable pieces. The statement and background
for this theorem involve a concept called paracompactness.
Definition. Let X be a topological space, and let U be an open covering of X. A second open
covering V is said to be a refinement of U if for each V ∈ V there is some U ∈ U such that V ⊂ U .
— Note that if V is a subcovering of U then V is a refinement of U, but there are refinements which
are not subcoverings. For example, if U MIN consists only of the set X, then every open covering V
is a refinement of UMIN .
Examples. If (X, d) is a metric space, then every open covering has a refinement consisting
of sets having the form Nδ(β) (xβ ) for suitable choices of the points x β and the radii δ(β). More
generally, if A is a base for the topology of X then every open covering has a refinement consisting
of open sets in A.
Definition. If X is a topological space, then X is said to be paracompact if every open covering
of X has an open locally finite refinement. — Note that compact spaces are paracompact because
subcoverings are refinements.
We have already noted that these notes do not discuss paracompact spaces in much detail, but
they are discussed at some length in Section 41 of Munkres. Our main interests concern their role
in metrization, and the first result along these lines is the following:
THEOREM 12. (A. H. Stone) If (X, d) is a metric space, then X is paracompact.
A proof of this result is given on page 257 of Munkres; the hard work in the proof is concentrated
in two earlier results; namely, Lemma 39.2 (see pages 246–247) and Lemma 41.3 (see pages 254–
257).
The file stone-disamb.pdf distinguishes between two mathematicians named Stone who made
contributions cited in Munkres.
Smirnov’s is a partial converse to A. H. Stone’s Theorem (alternatively, we can view A. H.
Stone’s Theorem as a converse to Smirnov’s theorem).
THEOREM 13. (Smirnov) A Hausdorff space is metrizable if it is paracompact and locally
metrizable.
A proof of this result appears in the course directory file smirnov.pdf. Roughly speaking,
paracompactness provides a means for constructing a globally defined metric out of locally defined
metrics.

118
 Appendix A : Topological groups

(Munkres, Supplementary exercises following § 22)

As the name suggests, a topological group is a mathematical structure that is both a topological
space and a group, with some sort of compatibility between the topological and algebraic structures.
Such objects lie at the point where two different areas of pure mathematics meet.
One motivation for looking at such combination structures is pure intellectual curiosity. Sci-
entists are always interested in learning what happens if you combine A with B. Sometimes the
results are not particularly useful (or do not seem so at the time), but very often these combinations
can lead to important new insights into our knowledge of the original structures and to powerful
new methods for analyzing questions that had previously been relatively difficult to study.
It turns out that topological groups form a rich family of interesting, relatively accessible
and fundamentally important topological spaces. One reason for this is that the group structure
turns out to impose some severe restrictions on the topology of the underlying space. In the other
direction, the structure theory of an important special case of topological groups — namely, the
compact connected Lie (pronounced LEE) groups — foreshadowed the classification of finite simple
groups that was completed during the second half of the twentieth century (Note: A finite group
is said to be SIMPLE if it is nonabelian and the only normal subgroups are the trivial subgroup
and the group itself; the alternating groups on n ≥ 5 letters are the most basic examples, there is a
very large class of such groups that are related to Lie groups, and there is a list of 26 other groups
that are called sporadic).
Another reason for the importance of topological groups is that such structures arise in a
wide range of mathematical contexts. In particular, many of the most important objects studied
in mathematical analysis come from topological groups (usually with some additional structure).
Topological groups also arise play crucial roles in many areas of geometry, topology and algebra, and
they are quite useful in application of mathematics to the sciences as well; physics is an especially
prominent example.).
Having discussed the role of topological groups, the next step is to describe them formally.
Definition. A topological group is a quadruple

(G, T, m, inv)

consisting of a nonempty topological space (G, T) and a group (G, m, inv) such that the multipli-
cation map m : G × G → G and the inverse map inv : G → G are continuous.

Examples. 1. The real numbers R form a topological group with respect to the addition
operation, and the nonzero real numbers form a topological group with respect to multiplication.
Similar statements hold for the complex numbers C. In fact, one can the real and complex numbers
with the addition and multiplication operations as examples of a suitably defined topological field.
2. The unit circle S 1 is a topological group because it is a subgroup of the multiplicative
group of nonzero complex numbers.
3. If (G, · · · ) and (H, · · · ) are topological groups then one can make their product G × H
into a topological space and a group, and it turns out that the product topology and the product

119
group operations define a topological group structure on G × H. If we specialize this to the case
G = H = S 1 we obtain the group structure of the 2-dimensional torus T 2 . By induction the k-fold
product of S 1 with itself, which is equivalent to T k−1 × S 1 is also a topological group which is
called the k-torus and denoted by T k .
4. Let F = R or C, and let GL(n, F) be the group of invertible n × n matrices over F.
Suppose first that F = R. Then GL(n, R) is an open subset of the set of all n × n matrices (with
2
the topology inherited from Rn ), and the coordinates of the multiplication and inverse maps are
respectively polynomial and rational functions of the entries of the matrices (multiplication) or
matrix (inversion). Therefore the conditions for a topological group are satisfied. Suppose now
that F = C. Then GL(n, C) consists of all complex matrices whose determinants are nonzero. The
natural topology for the space of n × n matrices over C is given by identifying the matrices with
2
points of R2n , and from this viewpoint the real and imaginary parts of the complex determinant are
polynomial functions of the real and imaginary parts of the matrix entries. Therefore we see that
2
GL(n, C) corresponds to an open subset of R2n , and as in the real case it follows that multiplication
and inversion are polynomial and rational functions so that GL(n, C) is also a topological group.
5. The subgroups of orthogonal and unitary matrices are important compact subgroups of
GL(n, R) and GL(n, C) respectively. The verification that they are subgroups is carried out in
linear algebra courses, Why are they compact? Recall that orthogonal and unitary matrices are
characterized by the fact that their rows (equivalently, columns) form an orthonormal set. Since
the rows (or columns) are all unit vectors, it follows that the groups O n and Un of orthogonal and
unitary matrices are subsets of the compact set
Qn 2
S αn−1 ⊂ Rαn

where α = 1 for R and α = 2 for C. Since the orthonormality conditions reduce to equations
involving certain polynomials in the real and imaginary parts of the matrix entries, it follows that
the orthogonal and unitary groups are closed subsets of the products of spheres displayed above,
and therefore these groups are compact.
6. In Exercises I.1.1 and VI.2.4 we considered a metric d p on the integers for each prime
p, and we noted that the completion Zcp was a compact metric space. In fact, one can extend the
usual addition and multiplication maps on Z to continuous maps

cp × Z
b :Z
µ cp cp
−→ Z

cp × Z
b :Z
α cp cp
−→ Z
that make the space in question into a compact topological commutative ring with unit. The existence
of these extensions follows directly from the results on extending uniformly continuous functions
on metric spaces to the completions of the latter and the uniform continuity of ordinary addition
and multiplication on Z with respect to the metrics d p (verify this!). The systems obtained in this
manner are known as the p-adic integers.

Especially in topology, whenever some type of mathematical structure is defined, one should
also define the mappings or morphisms from one such object to another. For topological groups, the
notion of continuous homomorphism is an obvious choice. Formally, these are continuous functions
ϕ : G → H such that ϕ(ab) = ϕ(a) · ϕ(b) or alternatively satisfy the morphism identity

mH o (ϕ × ϕ) = ϕ o mG : G × G −→ H

120
where mG and mH are the multiplication maps for G and H respectively.
One particularly important continuous homomorphism is the exponential map from the additive
topological group of real or complex numbers to the multiplicative topological group of nonzero
real or complex numbers. Another example on GL(n, F) is the map sending an invertible matrix
to its transposed inverse. Over the complex numbers one also has the conjugation map on both the
additive and nonzero multiplicative groups of complex numbers and the conjugate of the transposed
inverse for invertible complex matrices.
All the nonexponential examples in the preceding paragraph are in fact topological automor-
phisms of the groups in question; i.e., they are continuous homomorphisms that have continuous
inverses. In fact, for all these examples the map is equal to its own inverse.

Properties of topological groups

We had previously stated that the group structure on a topological group (and especially its
continuity!) implies strong restrictions on the topology of the underlying space. We shall discuss
the most basic properties here.

HOMOGENEITY OF TOPOLOGICAL GROUPS. A topological space X is said to be homogeneous

if for each pair of points u v ∈ X there is a homeomorphism h : X → X such that h(u) = v. It
is necessary to be somewhat careful when using this term, because the expression “homogeneous
space” has a special meaning that is described below. There are many examples of spaces that are
homogeneous, and there are many spaces that do not satisfy this condition.

EXAMPLES. 1. Every normed vector space V is homogeneous. Given two vectors u, v ∈ V

the translation map T (x) = x + (v − u) is an isometry and sends u to v.
2. The closed unit interval [0, 1] is NOT homogeneous; more precisely, it is impossible to
construct a homeomorphism taking an end point to a point in the open interval (0, 1). If such a
homeomorphism h existed then for all open sets U containing 1 there would be homeomorphisms
between U − {1} and h[U ] − {h(1)}. On the other hand, sets of the latter type are disconnected if
h(1) ∈ (0, 1) while U − {1} is the connected set [0, 1) if U = [0, 1].
3. Since the open interval (0, 1) is homeomorphic to R, it follows from the first example
that (0, 1) is homogeneous. In fact, if U is an open connected subset of R n , then U is homogeneous.
The proof is relatively elementary, but since the argument is a bit lengthy it will be given separately
in Appendix C.
The homogeneity of topological groups is essentially a generalization of the first example given
above:
PROPOSITION 1. If G is a topological group, let L a : G → G be the continuous map
La (g) = a · g (“left multiplication”), and let R a : G → G be the continuous map Ra (g) = g · a
(“right multiplication”). Then La and Ra are homeomorphisms.
Sketch of proof. Let b = a−1 . Then the associativity and continuity of multiplication imply
that Lb and Rb are inverses to La and Ra respectively.
COROLLARY 2. A topological group is homogeneous.
Proof. Given distinct points a , b ∈ G let c = b a −1 and note that Lc (a) = b.

121
SEPARATION PROPERTIES OF TOPOLOGICAL GROUPS. All of the examples of topological
groups that we have explicitly described are Hausdorff. It turns out that weaker separation proper-
ties imply the Hausdorff Separation Property, and the Hausdorff Separation Property itself implies
even stronger separation properties. We shall begin by proving the first of these statements.
PROPOSITION 3. If a topological group is a T 0 space, then it is a Hausdorff (or T2 ) space.
Proof. The proof splits into two parts, one showing the implication T 0 =⇒ T1 and the other
showing the implication T1 =⇒ T2 .
( T0 =⇒ T1 ). It suffices to prove that the one point set {1} consisting only of the
identity element is closed in G; for every other element g, one can apply the homeomorphism L g to
see that {g} = Lg [ {1} ] is also closed in G. Therefore we need only show that G − {1} is open. If x
belongs to the latter, then the T0 condition implies that there is an open subset W that contains
either 1 or x but not both. If x ∈ Wx let Ux = Wx . Suppose now that 1 ∈ Wx but x 6∈ Wx ; if
we can find a homeomorphism h : G → G such that h switches 1 and x, then h(W x ) is an open
set containing x but not 1, and if we set U x = h(Wx ) this will express G − {1} as a union of open
sets and thus prove that G − {1} is open. The desired homeomorphism is merely the composite
Lx o inv.
( T1 =⇒ T2 ). It will suffice to show that the diagonal is closed in G × G. Let D :
G × G → G be the composite m o (idG × inv). As noted in Exercise 1 on page 145 of Munkres,
this map is continuous. But ∆G = D −1 [{1} ] and since {1} is closed in G by assumption, it follows
that ∆G is closed in G × G so that G is Hausdorff, which is the same as saying that G is T 2 .
If a topological group is Hausdorff, then one can prove that it satisfies even stronger separation
properties. In particular, Exercise 7 on page 146 of Munkres shows that a Hausdorff (equivalently,
a T0 ) topological group is T3 , and Exercise 10 on pages 213–214 of Munkres shows that such a
topological group is also T3 12 .

CLOSED SUBGROUPS OF TOPOLOGICAL GROUPS. In the theory of topological groups it

is important to know whether or not a subgroup is closed, and if a subgroup is not closed it is
important to have the following additional information about its closure:
PROPOSITION 4. If G is topological group and H is a subgroup, then its closure H is also a
subgroup. Furthermore, if H is normal then so is H.
Before proceeding further, here are some examples when G = R (with addition as the group
operation). The rational numbers are clearly not a closed subgroup (since they are dense in the
real line), but the integers are, and one way of seeing this is to note that the complement of
integers is equal to the union of the open intervals (n, n + 1) where n ranges over all the elements
of Z. Over the complex numbers, the additive groups of real and purely imaginary numbers are
closed subgroups (why?) and in the groups GL(n, F) the orthogonal and unitary groups are closed
subgroups because the latter are compact.
Proof of Proposition 4. If H is a subgroup then we know that m[H ×H] = H and inv[H] = H.
Recall that a function f is continuous if and only if it satisfies the condition

f [A] ⊂ f [A] .

If we apply this to multiplication and the inverse map we find that

   
m H × H = m H × H ⊂ m[H × H] = H

122
and similarly  
inv H ⊂ H
so that H is a subgroup of G.
Suppose now that H is normal in G; in other words, for all a ∈ G the inner automorphism
Ia (x) = axa−1 maps H to itself. This mapping is continuous because it is the composite L a o Ra−1 ;
by associativity it can also be written as R a−1 o La . As in the
 previous paragraph the continuity of
this map and the hypothesis Ia [H] ⊂ H imply that Ia H ⊂ H.
In group theory, if one has a group G and a subgroup H, then it is possible to form the set
G/H of cosets, and it H is a normal subgroup then one can construct a quotient group structure on
G/H. For topological groups one can carry out the same constructions and topologize G/H using
the quotient topology; this quotient is T 1 if and only if H is a closed subgroup. Some additional
information on this can be found in Exercises 5 and 6 on page 146 of Munkres (see also Exercise 7(d)
on that page). We shall also note that the standard isomorphism theorems known from ordinary
group theory have topologized analogs for topological groups.
Example. Although closed subgroups play an extremely important role in the theory of
topological groups, the non-closed subgroups are too important to be ignored, so we shall give
one more example involving T 2 : Given an irrational real number a > 0, consider the continuous
homomorphism
ϕ : R −→ T 2
defined by  
ϕ(t) = exp(2πit), exp(2πiat)
and let H be the image of ϕ. To see that H is not closed, consider its intersection with {1} × S 1 ,
and call this intersection

L. By definition it follows that L consists of all elements of the form
1, exp(2πian) for some n ∈ Z. If H is a closed subgroup of T 2 then so is L. By the previous
description L is countable; we claim it is also infinite; if not, then there would be two integers
p 6= q such that exp(2πiap) = exp(2πiaq), and the latter would imply that a(p − q) is an integer,
contradicting the irrationality of a. Since T 2 is a compact metric space and L is a closed subset,
it follows that L is complete with respect to the subspace metric. Therefore a corollary of Baire’s
Theorem implies that L has an isolated point. But L is a topological group and therefore every
point is isolated by homogeneity. This and compactness force L to be finite, contradicting our
previous conclusion that L was infinite. The contradiction arose from the assumption that the
subgroup H was closed, and therefore we see that H cannot be a closed subgroup.

CONNECTED SUBGROUPS OF TOPOLOGICAL GROUPS. It is also important to know whether

a subgroup of a topological group is open, but the reasons for this are fundamentally different, and
the following result indicates why this is the case:
PROPOSITION 5. An open subgroup of a topological group is also closed.
In contrast, a closed subgroup of a topological group is not necessarily open, and the simplest
example is probably the inclusion of R × {0} in R 2 .
Proof. Let H be an open subgroup of G. Then there exist elements g α ∈ G such that the cosets
Hgα are pairwise disjoint and their union is G; we may as well suppose that g β = 1 if gβ is the
unique element such that H = Hgβ . Since the right translation maps R gα are homeomorphisms it
follows that each coset is open. Therefore
[
H = G − Hgα
gα 6=1

123
expresses H as the complement of a union of open subsets, and it follows that H is closed.
Here are a few consequences of this relatively elementary but far-reaching observation.
PROPOSITION 6. (i) The smallest subgroup which contains a fixed open subset of a topological
group is both open and closed.
(ii) The connected component of the identity in a topological group is a normal closed sub-
group.
(iii) If a topological group is connected, then it is generated by every open neighborhood of
the identity.
Proof. (i) Let W be an open neighborhood of the identity; by taking the union of W with its
inverse we may assume that inv[W ] = W . Consider the sequence of sets W n defined inductively
so that W1 = W and Wn+1 = Wn · W , where the raised dot denotes group multiplication. By
construction Wn is mapped to itself by the inverse mapping; we claim that it is also open. To see
this, note that if U is an open subset of G and A is an arbitrary set then the identity A · U =
∪a La (U ) shows that A · U is open. Consider the set W ∞ = ∪n Wn . If H is an arbitrary subgroup
of G containing W , then H certainly contains W ∞ . We claim that W∞ itself is a subgroup; at
this point we only need to verify that W ∞ is closed under multiplication, but this is an elementary
exercise (work out the details!). Thus W ∞ is the smallest subgroup containing W , and it is an
open subset. By our previous result it is also a closed subset.
Proof of (ii). As usual, write a ∼ b if a and b lie in the same connected component of G.
Since left multiplication is continuous we know that a ∼ b implies ga ∼ gb for all g ∈ G. Therefore
if a ∼ 1 and b ∼ 1 we obtain the relation ab ∼ a, so that ab ∼ 1 by transitivity. Therefore the
component of the identity is closed under multiplication. Similarly, if a ∼ 1 then left multiplication
by a−1 yields 1 ∼ a−1 and hence that the component of the identity is also closed under taking
inverses. Thus this component G◦ is a subgroup, and it is closed because connected components
are closed.
Finally to see that G◦ is normal, let x ∈ G and let Ix denote conjugation by x. Since Ix is a
homeomorphism and takes 1 to itself, it follows that I x must also take the connected component
of 1, which is G◦ , to itself, and therefore the latter must be a normal subgroup.
Proof of (iii). If U is a neighborhood of the identity we have seen that the smallest subgroup
H containing U is open and closed. Since 1 ∈ H, the set is also nonempty, and therefore by
connectedness we must have H = G.
Note that the additive groups of p-adic integers and the rational numbers are examples for
which the component of the identity is not an open subgroup.
At this point we shall only state and prove one more result which illustrates the strong prop-
erties of topological groups.
PROPOSITION 7. If G is a connected topological group and D is a discrete normal subgroup
of G, then D is central.
The normality is crucial in this result. The group GL(n, C) is connected if n ≥ 2 (see the
exercises for hints on proving this) but it contains many discrete non-central subgroups. One simple
example is the subgroup with two elements given by the identity and the diagonal matrix with a
−1 in the upper left corner and +1 in every other diagonal entry.
Proof. Let J : G × D → D be the continuous map sending (g, x) to gxg −1 ; the image lies in D
because the latter is a normal subgroup. What is the image of G × {x}? We know it is connected

124
and that it contains x. Since {x} is the connected component of x in D because the latter is
discrete, it follows that J maps G × {x} onto x. The latter means that gxg −1 = x for all g ∈ G,
which can be rewritten in the form gx = xg for all g ∈ G. Therefore x lies in the center of G. Since
x was arbitrary it follows that all of D lies in the center.

Analysis on topological groups

This subsection discusses how some important concepts from real variables (uniform continuity
and a good theory of integration) can be developed for topological groups. A reader who would
prefer to continue with the topological and geometrical discussion may skip this section without
missing any topological or geometrical material.
As noted in Section I.3 of the notes, topological groups have a uniform structure that allows
one to formulate a useful notion of uniform continuity. In principle, the idea is to consider a
family of symmetric neighborhoods of the identity (i.e., neighborhoods mapped to themselves
under inversion) and to formulate a uniform concept of closeness using these neighborhoods. For
example, if we are given two functions f and g from a set X to the topological group G and a
symmetric open neighborhood V of the identity, one can say that f and g are uniformly within V
of each other if f (x) · g(x)−1 ∈ V for all x ∈ X
In fact, every topological group can be viewed as a uniform space in two ways; the left unifor-
mity turns all left multiplications into uniformly continuous maps while the right uniformity turns
all right multiplications into uniformly continuous maps. If G is not abelian, then these two struc-
tures need not coincide. These uniform structures allow to talk about notions such as completeness,
uniform continuity and uniform convergence on topological groups.

INTEGRATION ON LOCALLY COMPACT TOPOLOGICAL GROUPS. In mathematical analysis,

the locally compact groups are of particular importance because they admit a natural notion of
measure and integral that was introduced by Alfréd Haar in the nineteen thirties. The idea is to
assign a “translation invariant volume” to subsets of a locally compact Hausdorff topological group
(e.g. if one takes a reasonable subset S and considers its left translate L g (S) for some g ∈ G, then
the volumes of the two sets are equal) and subsequently to define an integral for functions on those
groups.
If G is a locally compact Hausdorff topological group, one considers the algebra M of subsets
generated by all compact subsets of G and including all countable unions, countable intersections
and complements. If g is an element of G and S is a set in M, then the set L a (S) is also in M. It
turns out that there is, up to a positive multiplicative constant, only one left-translation-invariant
measure on M which is finite on all compact sets, and this is the Haar measure on G (Note:
There is also an essentially unique right-translation-invariant measure on M, but the two measures
need not coincide; the difference between these two measures is completely understood). Using the
general Lebesgue integration approach, one can then define an integral for all measurable functions
f : G → R (or C).

125
 The Internet site
http://www.wikipedia.org/wiki/Haar measure
contains further information about integration on locally compact groups and its uses in mathe-
matics (and physics!).

Remarks on Lie groups

We have already mentioned the particular importance of Lie groups in mathematics. It is

beyond the scope of these notes to describe the objects fully (this is material for the third course in
the geometry/topology sequence), but we have developed enough ideas to discuss a few important
points.
We begin with a discussion of the exponential map for n × n matrices over the real numbers.
Our results on norms for finite-dimensional real vector spaces imply that R k is complete with
2
respect to every norm. Take k = n2 and view Rn as the space of all n × n matrices over R. The
exponential mapping
exp : M(n; R) −→ GL(n, R)
is the map defined by the familiar infinite series
X 1
exp(A) = An
n!
n≥0

and this series converges absolutely because

X 1 X 1
k An k ≤ kAkn
n! n!
n≥0 n≥0

and the right hand side converges to exp(k A k).

THEOREM 8. The exponential map defines a homeomorphism from an open neighborhood of
0 ∈ M(n; R) to an open neighborhood of I ∈ GL(n, R).
Proof. The key to this is the Inverse Function Theorem. By construction we know that

exp(A) − exp(0) = exp(A) − I = A + A · θ(A)

where limkAk→0 θ(A) = 0. Therefore it follows that D exp(0) = I. If exp is a C 1 mapping then we
can apply the Inverse Function Theorem to complete the proof.
One direct way to verify that exp is C 1 is to use power series. If A is an n × n matrix and B
is another matrix of the same size that is close to zero, then we may write
X 1
exp(A + B) − exp(A) = [(A + B)n − An ]
n>0
n!

where each of the terms in brackets is a sum of degree n monomials in the noncommuting matrices
A and B, and each monomial in the summand has at least one factor equal to B. The n th bracketed
term can be written as a sum fn (A, B) + gn (A, B) where each term in fn (A, B) has exactly one B

126
factor and each term in gn (A, B) has at least two B factors. The first of these may be written in
the form
Xn
fn (A, B) = Ar B An−r−1
r=1

and one can check that the series

X 1
Φ(A, B) = fn (A, B)
n>1
n!

converges absolutely because k fn (A, B) k≤ n kAkn−1 · kBk. Furthermore, it follows that Φ is

continuous in A (by uniform convergence of the series) and is linear in B. Therefore we will have
that D exp(A)(B) = Φ(A, B) and exp is C 1 if and only if for B 6= 0 we can show that

1 X 1
lim gn (A, B) = 0
B→0 kBk n>0 n!

An upper estimate for the norm of the right hand side is given by

1 X 1
k gn (A, B) k
kBk n>0 n!

where each term kgn (A, B)k is bounded from above by

n  
X n
kBkr kAkn−r .
r=2
r

The standard considerations then show that the series

1 X 1
Θ(A, B) = gn (A, B)
kBk n>0 n!

converges absolutely for all A and B and that

lim Θ(A, B) = 0
B→0

if A is held fixed. Thus we have shown directly that the matrix exponential is a C 1 function.
This result yields an important structural property.
PROPOSITION 9. There is a neighborhood U of the identity in GL(n, R) such that the only
subgroup contained in U is the trivial subgroup.
The conclusion is often abbreviated to say that GL(n, R) is a topological group with no small
subgroups (NSS).
Proof. Choose δ > 0 so that exp maps the disk of radius 2 δ about the origin homeomorphically to
an neighborhood W of the identity in GL(n, R). Suppose that U is the image of the neighborhood
of radius δ in GL(n, R), and let H be a subgroup that lies entirely in U . Suppose further that H
is nontrivial, and choose a matrix A so that kAk < δ and 1 6= exp(A) ∈ U .

127
 Consider the C∞ map ϕA : R −→ GL(n, R) sending t to exp(tA). Since exp(P + Q) =
exp(P ) exp(Q) if P and Q commute, it follows that ϕ A is a continuous homomorphism. The
sequence { knAk = nkAk } is unbounded and therefore there is a first n such that knAk = nkAk > δ.
Since kAk < δ it follows that nkAk = knAk < 2 δ. But this means that

exp(nA) = exp(A)n ∈ W −U ;

on the other hand, since exp(A) was supposed to belong to a subgroup H ⊂ U , we also have that
exp(A)n = exp(nA) ∈ U . The contradiction arises from the assumption that there was a nontrivial
subgroup H contained in U , and therefore no such subgroup exists.
COROLLARY 10. If H is a subgroup of GL(n, R), then H has no small subgroups.
Results of A. Gleason and (jointly and independently) D. Montgomery and L. Zippin provide
a completely topological characterization of Lie groups as locally compact Hausdorff groups with
no small subgroups. These results are proved in the book by Montgomery and Zippin as well as
Kaplansky’s Lie Algebras and Locally Compact Groups.
Note that a locally compact Hausdorff topological group that is locally connected and second
countable does not necessarily satisfy the “no small subgroups” condition. A countably infinite
product of copies of S 1 is a relatively simple counterexample.

Local topological groups

We have seen that if G is a connected topological group, then G is generated by each open
neighborhood of the identity, and it follows that much if not all of the group’s structure is retriev-
able from such arbitrarily small neighborhoods. In particular, this principle plays a fundamentally
important role in the study of Lie groups, and beginning with Lie’s research there have been several
reasons for studying potential fragments of topological groups which behave formally like neighbor-
hoods of the identity. These potential fragments are called local topological groups. Similarly, a
homomorphism of connected topological groups is determined by how it maps a small neighborhood
of the identity in one group into a small neighborhood of the identity in another group, so there are
also reasons for studying potential fragments of homomorphisms defined on these potential neigh-
borhoods of the identity; as one might expect, such mappings are called local homomorphisms.
Further information about local homomorphisms and local groups can be found at the following
site:
http://www.encyclopediaofmath.org/index.php/Local topological group
In the theory of finite-dimensional Lie groups, it turns out that every suitably defined local Lie
group actually comes from a Lie group, and likewise for continuous homomorphisms. Further
information can be found in the book by C. Chevalley in the bibliography.

128
 Appendix B : Stereographic projection and inverse geometry

The conformal property of stereographic projections can be established fairly efficiently using
the concepts and methods of inverse geometry. This topic is relatively elementary, and it has
important connections to complex variables and hyperbolic (= Bolyai-Lobachevsky noneuclidean)
geometry.
Definition. Let r > 0 be a real number, let y ∈ Rn , and let S(r; y) be the set of all points
n
x ∈ R such that |x − y| = r. The inversion map with respect to the sphere S(r; y) is the map T
on Rn − {0} defined by the formula

r2

T (x) = y + · x−y .
|x − y|

Altternatively, T (x) is defined so that T (x) − y is the unique positive scalar multiple of x − y such
that   

T (x) − y  · x − y = r 2 .

Another way of saying this is that inversion interchanges the exterior points to S(r; y) and the
interior points with the center deleted. If r = 1 and y = 0 then inversion simply takes a nonzero
vector x and sends it to the nonzero vector pointing in the same direction with length equal to the
reciprocal of |x| (this should explain the term “inversion”).
If n = 2 then inversion corresponds to the conjugate of a complex analytic function. Specifically,
if a is the center of the circle and r is the radius, then inversion is given in complex numbers by
the formula

−1
−1
T (z) = r 2 · z − a = r2 · z − a
where the last equation holds by the basic properties of complex conjugation. If r = 1 and a = 0
then inversion is just the conjugate of the analytic map sending z to z −1 .
The geometric properties of the analytic inverse map on the complex plane are frequently
discussed in complex variables textbooks. In particular, this map has nonzero derivative wherever
the function is defined, and accordingly the map is conformal. Furthermore, the map z → z −1 is an
involution (its composite with itself is the identity), and it sends circles not containing 0 to circles
of the same type. In addition, it sends lines not containing the origin into circles containing the
origin and vice versa (Note: This means that 0 lies on the circle itself and NOT that 0 is the center
of the circle!). Since complex conjugation sends lines and circles to lines and circles, preserves the
angles at which curves intersect and also sends 0 to itself, it follows that the inversion map with
respect to the unit circle centered at 0 also has all these properties).
It turns out that all inversion maps have similar properties. In particular, they send the every
point of sphere S(r; y) to itself and interchange the exterior points of that sphere with all of the
interior points except y (where the inversion map is not defined), they preserve the angles at which
curves intersect, they are involutions, they send hyperspheres not containing the central point y to
circles of the same type, and they send hyperplanes not containing {y} into hyperspheres containing
{y} and vice versa. We shall limit our proofs to the properties that we need to study stereographic
projections; the reader is encouraged to work out the proofs of the other assertions.

129
 Relating stereographic projections and inversions

We begin by recalling(?) some simple observations involving isometries and similarity trans-
formations from Rn to itself. Proofs or hints for proofs in the cases of isometries can be found in
many standard linear algebra texts.
FACT 1. If b ∈ Rn and F is translation by b (formally, F (x) = x + b), then F is an isometry from
Rn to itself and F sends the straight line curve from x to y defined by

α(t) = ty + (1 − t)x

to the straight line curve from F (x) to F (y) defined by

α(t) = tF (y) + (1 − t)F (x) .

Definition. If r > 0 then a similarity transformation with ratio of
similitude r on a metric space
X is a 1–1 correspondence f from X to itself such that d f (x), f (y) = r · d(x, y) for all x, y ∈ X.
Note that every isometry (including every identity map) is a similarity transformation with ra-
tion of similitude 1 and conversely, the inverse of a similarity transformation with ratio of similitude
r is a similarity transformation with ratio of similitude r −1 , and the composite of two similarity
transformations with ratios of similitude r and s is a similarity transformation with ratio of simil-
itude rs. Of course, if r > 0 then the invertible linear transformation rI on R n is a similarity
transformation with ratio of similitude r.
FACT 2. In Rn every similarity transformation with ratio of similitude r satisfying F (0) = 0 has
the form F (x) = rA(x) where A is given by an n × n orthogonal matrix.
FACT 3. In Rn every similarity transformation F is conformal; specifically, if α and β are
differentiable curves in Rn that are defined on a neighborhood of 0 ∈ R such that α(0) = β(0) such
that both α0 (0) and β 0 (0) are nonzero, then the angle between α 0 (0) and β 0 (0) is equal to the angle
between [F o α]0 (0) and [F o β]0 (0).
The verification of the third property uses Facts 1 and 2 together with the additional observa-
tion that if A is given by an orthogonal transformation then A preserves inner products and hence
the cosines of angles between vectors.
The key to relating inversions and stereographic projections is the following result:
PROPOSITION 4. Let e ∈ Rn be a unit vector, and let T be inversion with respect to the
sphere S(1; 0). Then T interchanges the
hyperplane defined by the equation hx, ei = −1 with the
nonzero points of the sphere S 12 ; − 12 e .
Proof. By definition we have
1
T (x) = ·x
hx, xi
and therefore the proof amounts to finding all x such that
 
 T (x) + 1 e  = 1 .
2 2

This equation is equivalent to

 2
1
4 =  T (x) + 21 e  = hT (x) + 12 e, T (x) + 12 ei

130
and the last expression may be rewritten in the form

1 hx, xi hx, ei 1
hT (x), T (x)i + hT (x), ei + = 2
+ +
4 hx, xi hx, xi 4

which simplifies to
1 hx, ei 1
+ + .
hx, xi hx, xi 4
Our objective was to determine when this expression is equal to 41 , and it follows immediately that
the latter is true if and only if 1 + hx, ei = 0; i.e., it holds if and only if hx, ei = −1.
An llustration of the preceding result appears in the file stereopic1.pdf in the course direc-
tory.
COROLLARY 5. Let e be as above, and let W be the (n − 1)-dimensional subspace of vectors
that are perpendicular to e. Then the stereographic projection map from S(1; 0) − {e} to W is
given by the restriction of the composite
G ot oH
to S(1; 0) − {e}, where H is the similarity transformation H(u) = 21 (u − e) and G is the translation
isometry G(v) = v + e.
Proof. It will be convenient to talk about the closed ray starting at a vector a and passing
through a vector b; this is the image of the parametrized curve

γ(t) = (1 − t)a + tb = a + t(b − a)

where t ≥ 0.
First note that the composite T o H sends the ray starting at the point e and passing through a
point x with hx, ei < 1 to the ray starting at e and passing through the point H(x), and the latter
satisfies hH(x), ei < 0. This is true for the mapping H by Fact 1 and the equation H(e) = 0, and
it is true for T o H because T is inversion with respect
to a sphere centered at 0. Furthermore, by
construction H sends the sphere S(1; 0) to S 12 ; − 21 e , and it also sends the hyperplane P defined
by hx, ei = −1 to itself.
By construction, stereographic construction sends the point y ∈ S(1; 0) − {e} to the point
z ∈ W such that z − e is the unique point at which the ray starting at
e and passing through y
meets the hyperplane P , and inversion sends the point η ∈ S 21 ; − 21 e − {0} to the unique point
α at which the ray starting at 0 and passing through η meets the hyperplane P .
Combining these, we see that T o H maps y to the unique point where the ray passing through
0 and y meets the hyperplane P . This point may be written uniquely in the form w − e where
w ∈ W , and in fact we have w = G o T o H(y). On the other hand, by the preceding paragraph we
also know that w is given by the stereographic projection.

The conformal property for inversions

The following is an immediate consequence of the Chain Rule and Fact 3 stated above:
PROPOSITION 6. Let U be open in Rn , let x ∈ U and let f : U → Rn be a C1 mapping.
Then f is conformal at x if Df (x) is a nonzero scalar multiple of an orthogonal transformation.

131
 The key observation behind the proposition is that if γ : (−δ, δ) → U is a differentiable curve
with γ(0) = x then
[f o γ]0 (0) = Df (x)[γ 0 (0)]
by the Chain Rule.
We are now ready to prove the result that we wanted to establish.
THEOREM 7. Every inversion map is conformal.
Proof. It is convenient to reduce everything to the case where the sphere is S(1; 0). Given an
arbitrary sphere S(r; y) there is a similarity transformation sending S(r; y) to S(1; 0) that is defined
by the formula F (x) = r −1 (x − y), and if T 0 and T are the associated inversions then we have

T 0 = F −1 o T o F .

By construction DF (x) = r −1 I for all x and therefore it follows that DT 0 = DT (as usual, “D ”
denotes the derivative of a function). Therefore, by the proposition it will suffice to show that DT
is always a scalar multiple of an orthogonal map.
Let x 6= 0 and write an arbitrary vector v ∈ Rn as a sum v = cx+u where c ∈ R and hx, ui = 0.
We have already noted that T (x) = ρ(x) −2 · x where ρ(x) = |x|, and we wish to use this in order to
compute the value of DT (x) at some vector h ∈ R n . The appropriate generalization of the Leibniz
Rule for products and elementary multivariable calculus show that

DT (x)[h] = ρ(x)−2 h + h∇[ρ(x)]−2 , hi · x

where
∇[ρ(x)]−2 = −[ρ(x)]−4 · ∇[ρ(x)]2 = −2[ρ(x)]−4 x .
If hx, hi = 0 this shows that DF (x)[h] = [ρ(x)] −2 h, while if h = x it follows that DF (x)[x] =
−[ρ(x)]−2 x. In particular, this shows that there is an orthonormal basis for R n consisting of
eigenvectors for DT (x) with associated eigenvalues ±ρ(x) −2 , and therefore it follows that DT (x)
is a positive scalar multiple of an orthogonal map, which in turn implies that T is conformal at x.
Since x was arbitrary, this proves the theorem.

132
 Appendix C : Homogeneity of open sets in Euclidean spaces

The purpose of this section is to prove a result that was stated in Appendix A:
THEOREM 1. (Homogeneity Theorem) If U is an open connected subset of R n and a and b
are distinct points of U , then there is a homeomorphism h : U → U such that h(a) = b.
There are two major steps in the proof; the first is to show that connected open sets are
locally homogeneous, and the second is to use local homogeneity and connectedness to prove global
homogeneity.

A local construction

The following result provides one way to verify local homogeneity:

PROPOSITION 2. Let D n be the solid unit disk in Rn , and let v ∈ D n be an interior point
with |v| < 1. Then there is a homeomorphism f : D n → D n such that f is the identity on S n−1
and f (0) = v.
Sketch of Proof. The geometric motivation is simple. Every point on D n lies on a closed
segment joining the origin to a point on S n−1 . One maps such a segment linearly to the segment
joining v to the same point on S n−1 . As is often the case with such geometrical ideas, it takes a
fair amount of algebraic manipulation to show that this actually works. An illustration of the basic
idea is given in the file radialproj.pdf in the course directory.
The following result will be useful in the course of the proof:
LEMMA 3. D n is homeomorphic to the quotient space of S n−1 × [0, 1] whose equivalence
classes are the one point subsets for all points with positive first coordinates together with the set
S n−1 × {0}.
Proof of Lemma 3. Consider the continuous map g : S n−1 × [0, 1] → D n sending (x, t) to
tx. The inverse images of points are precisely the classes described above, and since g is a closed
and surjective mapping it follows that the quotient space with the given equivalence classes is
homeomorphic to the image, which is D n .
Proof of Proposition 2 continued. Define F : S n−1 × [0, 1] → D n by the formula

F (u, t) = tu + (1 − t)v .

If t = 0 then F (u, t) = F (u, 0) = v and therefore the lemma implies that F passes to a continuous
mapping f : D n → D n . This map corresponds to the geometric idea proposed in the first paragraph
of the proof. We need to prove that f is 1–1 and onto.
By construction we have f (0) = v; it will be useful to start by proving that no other point
maps to v. But suppose that we have f (tu) = v where t ∈ (0, 1] and |u| = 1. By construction this
means that
v = tu + (1 − t) v
which is equivalent to the equation t(u − v) = 0. Since |v| < 1 = |u| we clearly have u 6= v so this
forces the conclusion that t = 0, which contradicts our original assumption that t > 0.

133
 Suppose now that we have w ∈ D n − {v}. Consider first the case where w ∈ S n−1 . Then by
construction w = f (w), and there is no other point w ∗ on S n−1 such that w ∗ = f (w). On the other
hand, if |ξ| < 1 we claim that f (ξ) 6∈ S n−1 because
 
 
|f (ξ)| =  |ξ|u + (1 − |ξ|)v  ≤ |ξ| + (1 − |ξ|) · |v| < |ξ| + (1 − |ξ|) = 1 .

Therefore we only need to show that if |w| < 1 then there is a unique point ξ such that |ξ| < 1
and f (ξ) = w. This in turn reduces to showing that there is a unique point u ∈ S n−1 such that w
lies on the open segment joining v and u; the latter is equivalent to showing that there is a unique
real number t such that t > 1 and
 
v + t(w − v) = 1 .

Rather than prove this by working out an explicit but complicated formula for t in terms of v
and w, we shall approach the assertion by analyzing the behavior of the quadratic function
 2
f (t) = v + t(w − v)

which represents the square of the distance between the point v + t(w − v) and the origin. The
coefficient of t2 for this function is the positive number |w − v| 2 and therefore limt→±∞ f (t) = +∞.
Furthermore, f (0) = |v|2 < 1 and f (1) = |w|2 < 1, and therefore there are exactly two values of t
such that f (t) = 1, one of which is greater than 1 and one of less is less than 0.
If s is chosen so that s > 1 and f (s) = 1 in the notation of the preceding paragraphs and we
let u = v + s(v − w), then f (x) = w if and only if
 
1−s
x = w + v.
s

Local homogeneity

The local homogeneity of an open subset of R n is an direct consequence of the proposition.

Note first that the proposition above remains true for every closed disk in R n of the form

D(y; r) = { y ∈ Rn | |y − a| = r } .

PROPOSITION 4. (Local homogeneity property) If U is open in R n and a ∈ U , then there

is an open neighborhood v of x such that V ⊂ U and for all x ∈ V there is a homeomorphism
h : U → U sending a to x.
Proof. Note first that the proposition above remains valid for every closed disk in R n of the form

D(y; r) = { y ∈ Rn | |y − a| ≤ r } .

To see this, observe that this disk is the image of D n under the self-homeomorphism h of R n sending
x to rx + a. Thus if b ∈ D(y; r) with |b − a| < r we can construct a homeomorphism sending a to
b and fixing the boundary sphere by the formula



g(y) = h f h−1 (y)

134
where h is the homeomorphism in the proposition sending 0 to h −1 (b). Given an open set U ⊂ Rn
and a point a ∈ U , one can find some r > 0 so that D(y; r) ⊂ U . For each b in the interior of this
disk we can construct a homeomorphism of the closed disk sending a to b such that the restriction
of the homeomorphism to the boundary is the identity. Given two points bλ, b 0 in the interior of
this disk one can use such homeomorphisms and their inverses to construct a homeomorphism of
the disk that is again fixed on the boundary sphere and sends b to b 0 . This homeomorphism can be
extended to the entire open set U by defining it to be the identity on the complement of the disk
(why does this work?). This proves the desired property.

A global homogeneity theorem

Suppose now that U ⊂ Rn is open and connected. The following result allows us to extend
the local homogeneity property to a global result:
PROPOSITION 5. Let X be a connected topological space, and for each x ∈ X suppose
that there is an open neighborhood V of X such that for each v ∈ V there is a homeomorphism
h : X → X such that h(x) = v. Then X is homogeneous; in other words, for each pair of distinct
points x, y ∈ X there is a homeomorphism h : X → X such that h(x) = y.
Proof. Consider the binary relation on X given by a ∼ b if and only if there is a homeomorphism
h : X → X such that h(a) = b. It follows immediately that this is an equivalence relation.
Furthermore, if C is an equivalence class of ∼ and x ∈ C, then there is an open neighborhood V
of x such that V ⊂ C. In particular, it follows that C is open. Since this is true for all equivalence
classes it follows that the latter decompose X into pairwise disjoint open subsets. Therefore the
union of all equivalence classes except C is also open, and hence C is closed. But X is connected,
and therefore we must have C = X, so that all points of X are equivalent under ∼ . By the
definition of the latter, it follows that X is homogeneous.

Some related examples

In Appendix A we noted that a closed interval [a, b] ⊂ R is not homogeneous; on the other
hand, an open interval (a, b) is homogeneous (since an open interval is homeomorphic to the real
line, one does not need the preceding material to prove this fact). More generally, a product of
n open intervals is also homogeneous, and it is natural to ask if the solid hypercube [−1, 1] n (a
product of n copies of the interval [−1, 1] with itself) is also not homogeneous, where n can be an
arbitrary positive number or ∞. It turns out that such a product is not homogeneous if n is finite,
but it is homogeneous if n = ∞.
The key to proving the result for finite values of n is the following basic result, which is stated
and proved in Munkres, Elements of Algebraic Topology (see Theorem 36.5 on page 207):
THEOREM 6. (Invariance of Domain, L. E. J. Brouwer) Let U be an open subset of R n , and let
f : U → Rn be a continuous 1 − 1 mapping. Then f is open, and hence it defines a homeomorphism
from U to h[U ], which is also open in R n .
COROLLARY 7. Let f : [−1, 1]n → [−1, 1]n be a homeomorphism, and let W be the subspace
(−1, 1)n . Then f maps W onto itself.

135
 The complement of W is the set of all points for which at least one coordinate is equal to ±1,
and it is the set-theoretic frontier of [−1, 1] n in Rn . This means that W can also be described as
the interior of [−1, 1]n in Rn .
Proof of Corollary 7. By Invariance of Domain we know that f maps W onto an open subset
of [−1, 1]n , and since W is the interior of the latter it follows that f maps W into itself. Of course,
the same is true for f −1 . To see that f [W ] = W , let x ∈ W and observe that by the preceding
sentence we must have y = f −1 (x) ∈ W . Since x = f (y) it follows that x ∈ f [W ], which is what
we wanted to prove.
One can push this further as follows: Given an arbitrary topological space and x, y ∈ X, we can
define an equivalence relation such that x ∼ y if and only if there is a homeomorphism h : X → X
such that h(x) = y. The preceding result implies that interior and frontier points of [−1, 1] n lie in
different equivalence classes.
CLAIM 8. Under the given equivalence relation, there are exactly two equivalence classes on
n
[−1, 1] ; namely, one class which corresponds to interior points and another which corresponds to
frontier points.
Sketch of proof. The methods used to prove local homogeneity also show that every pair
of interior points is equivalent, and in fact one can take the homeomorphism h such that it is the
identity on the set of frontier points. Thus it is only necessary to check that if x and y are two
frontier points then there is a homeomorphism which sends one to the other. One quick way to do
for Section II.4 which shows that [−1, 1] n is homeomorphic to the
this is to start with an exercise P
n
disk D of all points satisfying x2i = 1 such that the frontier of the hypercube corresponds to
the boundary sphere of the disk. If A is an n × n orthogonal matrix, then f A (x) = A · x defines a
homeomorphism from D n to itself such that the boundary sphere is sent to itself. Given two unit
vectors there is always an orthogonal matrix sending one to the other, and therefore it follows that
every pair of frontier points on the hypercube is equivalent under the given relation.
In view of the preceding, the following result may seem surprising:
THEOREM 9. (O. H. Keller) If Q is the Hilbert cube [−1, 1] ∞ and x and y are two points of
Q, then there is a homeomorphism h from Q to itself such that h(x) = y.
The reference to Keller’s original paper appears in the bibliography; the book by Chapman
discusses the theorem as part of a more general picture.

136
 Appendix D : Normal forms for orthogonal transformations

The following online reference contains background material from linear algebra which is related
to the discussion below:
http://math.ucr.edu/∼res/math132/linalgnotes.pdf
The Spectral Theorem in linear algebra implies that a normal linear transformation on a
complex inner product space (one that commutes with its adjoint) has an orthonormal basis of
eigenvectors. In particular, since the adjoint of a unitary transformation is its inverse, the result
implies that every unitary transformation has an orthonormal basis of eigenvectors.
It is clear that one cannot have a direct generalization of the preceding result to orthogonal
transformations on real inner product spaces. In particular, plane rotations given by matrices of
the form  
cos θ − sin θ
sin θ cos θ

do not have such a basis over the reals except in the relatively trivial cases when θ is an integral
multiple of π and the matrices reduce to (−I) k for some integer k. However, if one takes these into
account it is possible to prove the following strong result on the existence of a “good” orthonormal
basis for a given orthogonal transformation.
THEOREM 1. (Normal form) Let V be a finite-dimensional real inner product space, and
let T : V → V be an orthogonal transformation of V . Then there is an orthonormal direct sum
decomposition of V into T -invariant subspaces W i such that the dimension of each Wi is either 1
or 2.
In particular, this result implies that there is an ordered orthonormal basis for V such that
the matrix of T with respect to this ordered orthonormal basis is a block sum of 2 × 2 and 1 × 1
orthogonal matrices.
It is beyond the scope of these notes to go into detail about the results from a standard
linear algebra course that we use in the proof of the result on normal forms. Virtually all of the
background information can be found in nearly any linear algebra text that includes a proof of the
Spectral Theorem (e.g., the book by Fraleigh and Beauregard cited in the bibliography).

Small dimensional orthogonal transformations

Since orthogonal transformations preserve the lengths of vectors it is clear that a 1-dimensional
orthogonal transformation is just multiplication by ± 1. It is also not difficult to describe 2-
dimensional orthogonal transformations completely using the fact that their columns must be
orthonormal. In particular, it follows that the matrices representing 2-dimensional orthogonal
transformations of R2 with respect to the standard inner product have the form
 
cos θ ∓ sin θ
sin θ ± cos θ

137
for some real number θ. We have already noted that one class of cases corresponds to rotation
through θ, and it is an exercise in linear algebra to check that each of the remaining matrices
 
cos θ sin θ
sin θ − cos θ
has an orthonormal basis of eigenvectors, with one vector in the basis having eigenvalue −1 and
the other having eigenvalue +1. The details of verifying this are left to the reader as an exercise
(Hint: First verify that the characteristic polynomial is t 2 − 1).
One can use this to prove a geometrically sharper version of the result on normal form.
THEOREM 2. (Strong normal form) Let V be a finite-dimensional real inner product space,
and let T : V → V be an orthogonal transformation of V . Then there is an orthonormal direct sum
decomposition of V into T -invariant subspaces W i such that the dimension of each Wi is either 1
or 2 and T operates by a plane rotation on each 2-dimensional summand.
This is an immediate consequence of the preceding discussion, for if T operates on an invariant
2-dimensional subspace with by a map that is not a rotation, then we can split the subspace into
two 1-dimensional eigenspaces.

Complexification

If V = Rn and T is represented by the orthogonal matrix A, it is clear how one can extend
T to a unitary transformation on Cn . We shall need a version of this principle that works for an
arbitrary real inner product space V . The idea is simple but a little inelegant; it can be done better
if one uses tensor products, but we want to prove the result without introducing them.
One defines the complexification of V formally to be very much as one defines the complex
numbers. The underlying set of complex vectors V C is given by V × V , addition is defined in a
coordinatewise fashion, multiplication by a complex scalar a + bi is given by the formula
[a + bi](v, w) = (av − bw, bv + aw)
and the complexified inner product is defined by
h (v, w), (v 0 , w0 ) icomplex =


(hv, v 0 i + hw, w0 i) + i · hw, v 0 i − hv, w0 i .
Both intuitively and formally the pair (v, w) can be viewed as v + i w if one identifies a vector v in
the original space with (v, 0) in the complexification. Verification that the structure defined above
is actually a complex inner product space is essentially an exercise in bookkeeping and will not be
carried out here.
If we are given a linear transformation T : V → W of real inner product spaces, then
T ×T :V ×V −→ W × W
defines a complex linear transformation T C on the complexification. Moreover, this construction is
compatible with taking adjoints and composites:

∗
(T ∗ )C = TC
(S o T )C = SC o TC
In particular, if T is orthogonal then T C is unitary.

138
 Derivation of the normal form

The main ideas behind the proof are

(1) to extract as much information as possible using the Spectral Theorem for the unitary
transformation TC ,
(2) prove the result by induction on the dimension of V .
In order to do the latter we need the following observation that mirrors one step in the proof of
the Spectral Theorem.
PROPOSITION 3. Let T be as above, and assume that W ⊂ V is T -invariant. Then the
orthogonal complement W ⊥ is also T -invariant.
In the Spectral Theorem the induction proof begins by noting that there is an invariant 1-
dimensional subspace corresponding to an eigenvector for T . One then obtains a transformation
on the orthogonal complement of this subspace and can apply the inductive hypothesis to the
associated transformation on this invariant subspace. We would like to do something similar here,
and in order to begin the induction we need the following result.
LEMMA 4. If V is a finite dimensional real inner product space and T : V → V is an orthogonal
transformation, then there is a subspace W ⊂ V of dimension 1 or 2 that is T -invariant.
If we have this subspace, then we can proceed as before, using the induction hypothesis to
split W ⊥ into an orthogonal direct sum of T -invariant subspaces of dimension 1 or 2, and this will
complete the derivation of the normal form.
Proof of Lemma 4. Suppose that λ is an eigenvalue of T C and x is a nonzero eigenvector; since
TC is unitary we know that |λ| = 1. Express both λ and x in terms of their real and imaginary
components, using the fact that |λ| = 1:

λ = cos θ + i sin θ, x = (v, w) = v + i w

Then the eigenvalue equation TC (x) = λ x may be rewritten in the form

T (v) + i T (w) = TC (x) = λ x = (cos θ + i sin θ) · (v + i w) =

   
cos θ v − sin θ w + i sin θ v + cos θ w

which yields the following pair of equations:

T (v) = cos θ v − sin θ w

T (w) = sin θ v + cos θ w

It follows that the subspace W spanned by v and w is a T -invariant subspace, and since it is
spanned by two vectors its dimension is at most 2. On the other hand, since x 6= 0 we also know
that at least one of v and w is nonzero and therefore the dimension of W is at least 1.

139
 References

The following book is the text for this part of the course:

J. R. Munkres. Topology. (Second Edition), Prentice-Hall, Saddle River NJ, 2000.

ISBN: 0–13–181629–2.

Further references

Some books are to be tasted, others to be swallowed, and some few to be chewed
and digested; that is, some books are to be read only in parts; others to be
read, but not curiously; and some few to be read wholly, and with diligence and
attention.

Francis Bacon (1561-1626), Of Studies

Alexandroff, Pavel S.; Hopf, Heinz. Topologie, Band 1. (German. Grundlehren der mathe-
matischen Wissenschaften etc. Nr. 45.) Julius Springer, Berlin, 1935.
Amadio, Roberto M.; Curien, Pierre-Louis. Domains and Lambda Calculi. (Cambridge
Tracts in Theoretical Computer Science, 46.) Cambridge University Press, Cambridge, 1998. ISBN:
0-521-62277-8.
Birkhoff, Garrett; MacLane, Saunders. A Survey of Modern Algebra. (Reprint of the Third
1968 Edition.) Chelsea, New York NY, 1988. ISBN: 0-023-74310-7.
Birkhoff, Garrett; MacLane, Saunders. A Survey of Modern Algebra. (Fourth Edition).
Prentice-Hall, Englewood Cliffs NJ, 1977. ISBN: 0-023-10070-2.
Bourbaki, Nicolas. General Topology, Chapters 1–4. (Translated from the French. Reprint of
the 1989 English translation. Elements of Mathematics.) Springer-Verlag, Berlin-Heidelberg-New
York, etc., 1998. ISBN: 3-540-64241-2.
Bourbaki, Nicolas. General Topology, Chapters 5–10. (Translated from the French. Reprint of
the 1989 English translation. Elements of Mathematics.) Springer-Verlag, Berlin-Heidelberg-New
York, etc., 1998. ISBN: 3-540-64563-2.
Chapman, Thomas A. Lectures on Hilbert Cube manifolds. (CBMS Regional Conference Series
in Mathematics, No. 28.) American Math. Soc., Providence RI, 1976. ISBN: 0-821-81678-0.
Chevalley, Claude. Theory of Lie groups – I (15 th printing, Princeton Mathematical Series
Volume 8, Princeton Landmarks in Mathematics). Princeton University Press, Princeton, NJ,
1999. ISBN: 0-691-04990-4.
Christenson, Charles O.; Voxman, William L. Aspects of Topology. (Pure and applied
Mathematics, Vol. 39.) Marcel Dekker, New York-Basel, 1977.

140
Christenson, Charles O.; Voxman, William L. Aspects of Topology. (Second Edition.) BCS
Associates, Moscow ID, 1998. ISBN: 0-914351-07-9; 0-914351-08-7.
Dieudonné, Jean A. History of algebraic geometry. An outline of the history and development of
algebraic geometry. (Translated from French by Judith D. Sally. Wadsworth Mathematics Series.)
Wadsworth International, Belmont, CA, 1985. ISBN: 0-534-03723-2.
Dugundji, James. Topology. (Reprint of the 1966 Edition. Allyn and Bacon Series in Advanced
Mathematics. Allyn and Bacon, Boston MA-London (U.K.)-Sydney (Austr.), 1978. ISBN: 0-205-
00271-4.
Edwards, C. H., Jr. Advanced Calculus of Several Variables. (Corrected Reprint of 1973
Edition). Dover, Mineola NY, 1994. ISBN: 0–496–68336–2.
Fraleigh, John B.; Beauregard, Raymond A. Linear Algebra (Third Edition. Historical
notes by Victor J. Katz) Addison-Wesley, Reading MA etc., 1995. ISBN: 0-201-52675-1.
Franz, Wolfgang. General Topology. (Translated from the German by Leo F. Boron.) Frederick
Ungar, New York NY, 1965.
Goffman, Casper; Pedrick, George. First Course in Functional Analysis. (Second Edition.)
Chelsea, New York NY, 1983. ISBN: 0-828-40319-8.
Hazewinkel, Michiel (ed.). Encyclopædia of Mathematics: Supplement 3. Springer-Verlag,
Berlin-(etc.), 2002. ISBN: 0-792-34709-9.
Hewitt, Edwin; Stromberg, Karl. Real and Abstract Analysis. A Modern Treatment of the
Theory of Functions of a Real Variable. (Graduate Texts in Mathematics Vol. 25.) Springer-Verlag,
Berlin-Heidelberg-New York-etc., 1997. ISBN: 0-387-90138-8.
Hocking, John G.; Young, Gail S. Topology. (Second edition.) Dover, New York NY, 1988.
ISBN: 0-486-65676-4.
Isbell, John. R. Uniform Spaces. (Third printing with corrections.) Mathematical Surveys and
Monographs, Vol. 12. American Math. Soc., Providence RI, 1991. ISBN: 0-444-82375-1.
James, I. M. (ed.). History of topology. North-Holland, Amsterdam (NL), 1999. ISBN:
0-444-82375-1.
Jänich, Klaus. Topology. (With a chapter by Theodor Bröcker. Translated from the German by
Silvio Levy. Undergraduate Texts in Mathematics.) Springer-Verlag, Berlin-Heidelberg-New York,
1984. ISBN: 0-387-90892-7.
Kaplansky, Irving. Set Theory and Metric Spaces. (Second edition.) Chelsea, New York, 1977.
ISBN 0-828-40298-1.
Kaplansky, Irving. Lie Algebras and Locally Compact Groups. (Chicago Lectures in Mathe-
matics Series.) University of Chicago Press, Chicago IL, 1995. ISBN: 0-226-42453-7.
Kasriel, Robert H. Undergraduate Topology. (Reprint of the 1971 Edition.) Robert E. Krieger,
Huntington NY, 1977. ISBN: 0-721-65298-0
Keller, O. H. Die Homöomorphie der kompakten knovexen Mengen im Hilbertschen Raum.
Math. Annalen 105 (1931), 748–758.
Kelley, John L. General Topology. (Reprint of the 1955 Edition. Graduate Texts in Mathematics,
No. 27.) Springer-Verlag, Berlin-Heidelberg-New York, 1975. ISBN: 0-387-90125-6.

141
Krantz, Steven G. Real Analysis and Foundations. (Studies in Advanced Mathematics.) CRC
Press, Boca Raton FL, 1991. ISBN: 0-8493-7156-2.
McCarty George S. Topology : An Introduction with Application to Topological Groups.
(Second Edition. ) Dover, New York NY, 1988. ISBN: 0-486-65633-0.
Montgomery, Deane; Zippin, Leo. Topological Transformation Groups. (Reprint of the 1955
Edition.) Robert E. Krieger, Huntington NY, 1974. ISBN: 0-88275-169-7
Munkres, James R. Elements of Algebraic Topology. Westview Press, Boulder CO, 1993. ISBN:
0-201-62728-2.
Pontryagin, Lev S. Topological Groups. (Translated from the Second Russian Edition with a
preface by Arlen Brown, with additional material translated by P. S. V. Naidu. Classics of Soviet
Mathematics) Taylor and Francis, London (U. K.), 1987. ISBN: 2-881-24133-6
Raymond, Frank A. The end point compactification of manifolds. Pacific Journal of Math. 10
(1960), 947–963.
Rickart, Charles E. General Theory of Banach Algebras. (Reprinted with Corrections.) Robert
E. Krieger, Huntington NY, 1974. ISBN: 0-88275-091-7.
Royden, Halsey L. Real Analysis. (Third Edition.) Macmillan, New York NY, 1988. ISBN:
0-02-404151-3.
Rudin, Mary Ellen. Lectures on Set-Theoretic Topology. (CBMS Regional Conference Series
in Mathematics No. 23, reprinted with corrections.) American Mathematical Society, Providence
RI, 1977. ISBN: 0-8218-1673-X.
Rudin, Walter. Principles of Mathematical Analysis. (Third Edition. International Series
in Pure and Applied Mathematics.) McGraw-Hill, New York-Auckland-Düsseldorf, 1976. ISBN:
0-07-054235-X.
Rudin, Walter. Functional Analysis. (Second Edition. International Series in Pure and Applied
Mathematics.) McGraw-Hill, New York NY, etc., 1991. ISBN: 0-07-054236-8.
Siebenmann, Laurence C. The obstruction to finding a boundary for an open manifold of
dimension greater than five. Ph. D. Thesis, Princeton University, 1965. — Available online at the
following sites:
http://www.maths.ed.ac.uk./∼aar/sieben.pdf
http://www.maths.ed.ac.uk./∼aar/thesis.pdf
Shah, Sejal; Das, T. K. On nearly Hausdorff compactifications. Applied General Topology 7
(2006), 125–130.
Steen, Lynn Arthur; Seebach, J. Arthur, Jr. Counterexamples in Topology. (Reprint of
the Second (1978) Edition.) Dover, Mineola NY, 1995. ISBN: 0-486-68735-X.
Taylor, Paul. Practical Foundations of Mathematics. (Cambridge Studies in Advanced Mathe-
matics 59.) Cambridge University Press, Cambridge, 1998. ISBN: 0-521-63107-6.
Walker, Russell C. The Stone-Čech Compactification. (Ergeb. Math. 2. Folge, Bd. 83.
Springer-Verlag, NBerlin-Heidelberg-New York, 1974. ISBN: 3-540-06699-3.
Willard, Stephen. General Topology. Dover Publications, Mineola NY, 2004. ISBN: 0-4864-
3479-6.
Wolf, Robert S. Proof, Logic, and Conjecture: The Mathematician’s Toolbox. W. H. Freeman,
New York NY, 1998. ISBN: 0-7167-3050-2.

142
 Summary of Files in the Course Directory

The course directory is ∼res/math205A-2014 on the math.ucr.edu network. This summary

only lists files which are relevant to the first part of the course (the material covered in these notes).

aabInformation-2014.pdf
This is the handout from the first day of class.
aabInternetresources.pdf
General statements on the reliability of online Internet sources, with special attention to
Wikipedia articles which are mentioned in these notes and Google searches.
affine+convex.pdf
Some basic properties of a class of homeomorphisms from R n to itself which are known as
affine (AFF-fine) transformations; these include all isometries and similarities of R n (see also
metgeom.pdf).
affine+measure.pdf
A note on affine transformations of Rn which preserve the standard (Riemann or Lebegue)
measure; one can combine point set topology and measure theory to study more general questions
about the measure-preserving homeomorphisms, but such material quickly goes beyond the scope
of this course.
beyond205A.pdf
Remarks about subsequent courses in the sequence and some areas of mathematics which are
closeley related to this course, including information on the continuation of the material from Part
II of Munkres at the beginning of Mathematics 205B.
braintest1.pdf
This is definitely not a serious piece of course material, but it does illustrate the importance
of staying focused on the main points when learning or doing mathematics.
braintest2.pdf
A sequel to the preceding document.
braintest3.pdf
Another sequel to the preceding documents.
braintest4.pdf
Yet another sequel to the preceding documents.
brief0.pdf
General statement about light or brief coverage of topics at the beginning of the course.
brief1.pdf
Sketch for light or brief coverage of selected topics from Unit I.
brief2∗.pdf
Sketches for light or brief coverage of selected topics from Unit II, with ∗ equal to a (Sections
II.1–2), b (Section II.3) or c (Section II.4).

143
 brief3.pdf
Sketch for light or brief coverage of selected topics at the beginning of Section III.1, at which
point light or brief coverage is discontinued.
cantor-disamb.pdf
Disambiguation for two mathematicians named Cantor who made important contributions to
mathematics in the late 19th and early 20th centuries.
cantor-set.pdf
Some additional discussion of the Cantor set, with a drawing and some online references.
cardinals.pdf
Background material on the cardinal numbers, summarizing everything about them that is
needed to use them in this course.
categories2014.pdf
A brief survey of category theory, not needed for this part of the course but included as
background.
characterizations.pdf
A description of the axiomatic characterizations for topological products and sums or dis-
joint unions by means of continuous functions (these involve projections onto the coordinates for
products, and injections into the summands for sums or disjoint unions).
concat.pdf
A picture illustrating the stringing together, or concatenation, of two curves, where the ending
point of the first is the starting point of the second.
cubicroots.pdf
This note discusses the application of the Contraction Lemma to finding the roots of cubic
polynomials with real coefficients.
cutpoints.pdf
This note discusses the points in a topological space for which the complements are discon-
nected; since homeomorphic spaces have homeomorphic sets of cut points, the latter sometimes
provide a criterion for concluding that two spaces cannot be homeomorphic.
disjoint-union.pdf
A drawing which illustrates the disjoint union construction on a pair of sets A, B.
dpmetrics.pdf
The three metrics dp (where p = 1, 2, ∞) on a product of metric spaces are showm to be part
of a continuous family of such metrics for 1 ≤ p ≤ ∞, and some results comparing these metrics are
also established (for example, in each case their open sets are given by the product topology). This
document involves some analytic input which sometimes appears in multivariable calculus texts or
courses. The files dpunitdisks.pdf and product-metricsn.pdf — where n = 1, 2, 3 — contain
additional material on this topic. Aside from the results for the three cases p = 1, 2 or ∞, none of
this material is needed elsewhere in the course.

144
 dpunitdisks.pdf
This file contains drawings for the unit disks in R 2 for certain dp metrics as defined in the
preceding document. The files product-metricsn.pdf — where n = 1, 2, 3 — contain additional
drawings and information on these metrics. Aside from the results for the three cases p = 1, 2 or
∞, none of this material is needed elsewhere in the course.
dualSBtheorem.pdf
This is an analog of the Schröder-Bernstein Theorem in which 1–1 maps from a set A to a set
B are replaced by onto maps from B to A, with the same conclusion regarding cardinalities (see
also the file topschroederbernstein.pdf below).
elltangents.pdf
An example of a geoemtrical problem involving ellipses which can be handle very efficiently
using an affine transformation which sends the ellipse to a circle. More generally, one can often use
self-homeomorphisms of a space to reduce the proof of a statement to a special case which is easier
to analyze.
embeddings.pdf
This file gives a proof for a result on embedding one topological space into a product of other
spaces (in many important cases, the factors are all homeomorphic to [0, 1]). The result is needed
to prove two results in Unit VI of these notes.
flattening.pdf
Informal, illustrated description of a homeomorphism from R 2 to itself which takes an angle
and flattens it into a straight line. This is another example to illustrate how a homeomorphism
from the plane to itself can deform subsets of the plane.
foundations1.pdf
A brief discussion of basic logic and some very elementary set theory. This is meant as review
and/or background.
foundations2.pdf
This discusses topics on relations and functions from the perspective of this course to the extent
that it differs from the main text; it also includes comments on axioms for the positive integers
and discussions of the roles of the Axiom of Choice and the Generalized Continuum Hypothesis in
set theory. The latter discussions are not needed for the course but may be helpful for obtaining a
broader perspective of the mathematical foundations underlying the course.
gentopexercises2014.pdf
The current version of the exercises for the course.
gentopnotes2014.pdf
This document, which is the first part of the course notes.
gentopsubheadings2014.pdf
A more detailed table of contents which lists all the subheadings in the preceding document.

145
 homeomorphisms.pdf
A discussion, with illustrations, concerning the geometric behavior of homeomorphisms be-
tween subsets of Rn , including homeomorphisms from such subsets to themselves.
inv-stereo.pdf
This document proves that the inverse of the stereographic projection map, which sends R n to
the complement of a point in the sphere S n , is conformal in the sense that it preserves the angles
at which regular smooth curves intersect.
lambert-fcn.pdf
This is a discussion of the inverse function to z = w exp(w); this example has an inverse, but
the inverse cannot be written explicitly in terms of the usual functions from first year calculus (the
file nonelementary-integrals.pdf contains some background information).
linalgnotes.pdf
Notes for a second undergraduate course in linear algebra. These are included partly for
reference purposes, but also because the results on classifying quadrics in Section V.2 have the
following topological implication: If Σ ⊂ R n is defined by a nontrivial quadratic polynomial in n
variables, then there are only finitely many possibilities for the homeomorphism type of Σ because
one can find a homeomorphism from Rn to itself which sends Σ to an example in a (relatively short)
finite list.
lipschitz1.pdf
Remarks on mappings of metric spaces which satisfy a Lipschitz condition.
lipschitz2.pdf
Further results on functions which satisfy Lipschitz conditions.
math205Asolutionsn.pdf
Files with solutions to the exercises, with 0 ≤ n ≤ 9. The file math205Asolutions00.pdf
contains suggestions for approaches to solving the exercises in the subsequent files. Files with the
names math205Asolutionsna.pdf contain illustrations and related hints for solving a few problems
where drawings seem particularly useful (however, drawings are potentially useful for solving most
of the problems).
math205Awww2014.pdf
A clickable listing of World Wide Web references which are cited in the course notes and other
directory documents.
mathproofs.pdf
Undergraduate level background discussion of mathematical proofs.
maximality.pdf
Proof that the real numbers can be described as a maximal ordered field (see also the file
uniqreals.pdf).
metgeom.pdf
This an appendix to the course notes which discusses isometries and similarity maps of Eu-
clidean spaces, including a proof that partial isometries and similarities on subsets of Euclidean

146
spaces extend to global isometries and similarities of the appropriate Euclidean spaces. The file
affine+convex.pdf contains further information on some of the topics discussed here.
nicecurves.pdf
This an appendix to the course notes, and it proves an assertion from Section III.5 about
joining two points in a connected open subset of Euclidean space by a curve that is infinitely
differentiable and has nonzero tangent vectors at every point.
no-max-or-min.pdf Examples of continuous real valued functions on noncompact sets with
no maximum or minimum values.
nonlemenary-integrals.pdf Remarks about functions like e−x 2 which have antiderivatives
which are not given by closed formulas involving the standard functions in single variable calculus
courses (this is related to lambert-fcn.pdf).
nonrectangular.pdf A proof that the open unit disk in the plane (with respect to the usual
Euclidean metric) cannot be written as a finite union of basic open sets in the product topology
for R2 .
numberexpansions.pdf
Proofs that the axioms for the real numbers in Section 4 of Munkres yield the usual decimal
and other expansions for real numbers.
ordinals.pdf
Background material on the ordinal numbers. The latter may be viewed as equivalence classes
of well-ordered sets, and the class of all ordinals determines a natural indexing for the class of all
cardinal numbers. This summary, which was written by N. Strickland of the University of Sheffield,
is available online from
http://www.shef.ac.uk/∼pm1nps/courses/topology/ordinals.∗
and is included in the course directory only for the convenience of students enrolled in this course.
The site also includes several other interesting and informative documents related to this course.
outline205A1.pdf
An outline of sections in the class notes which will be covered in course and qualifying exami-
nations.
polya.pdf
A one page summary of the advice for solving problems in the classic book, How to Solve It,
by G. Pólya (with an extra piece of advice added at the end).
product-metricsn.pdf
Information on the previously mentioned d p metrics on a product of metric spaces, with n
ranging from 1 to 3. The first file contains an illustrated proof that the metrics for p = 1, 2, ∞ on a
product metric all define the product topology, the second contains a color illustration comparing
the metrics for several values of p and links to other drawings, and the third contains additional
results comparing these metrics and describing their limit properties. There is considerable overlap
with the previously mentioned files dpmetrics.pdf and dpunitdisks.pdf. Aside from the results
for the three cases p = 1, 2 or ∞, none of this material is needed elsewhere in the course.

147
 projspaces.pdf
This document states and proves results about the topological properties of real and complex
projective spaces.
proper.pdf
A discussion of the basic facts about proper maps, with emphasis on their properties that are
relevant to algebraic geometry. This is supplementary material not covered in the course, and the
level of exposition is slightly higher than in the course notes.
pstop-motiv.pdf
A summary of factors leading to the development of point set topology, particularly as they
arise in undergraduate mathematics courses.
quotmap.pdf
Remarks on an example, from Munkres, of a quotient map which is neither an open nor a
closed mapping.
radproj.pdf
A picture illustrating the radial projection map that is used to prove a result in Appendix C.
realnumbers.pdf
A summary of the basic properties of the real numbers.
reflections.pdf
Derivation of the standard formula for the geometric transformatiion of R 2 which sends each
point to its mirror image with respect to some line; the same methods also apply to reflections
about hyperplanes in Rn .
rpn-in-rk.pdf
A proof that real projective spaces are homeomorphic to subsets of Euclidean (= Cartesian)
spaces.
stone-disamb.pdf
Disambiguation for the two mathematicians named Stone who made contributions to the sub-
ject matter of this course.
smirnov.pdf
A proof of Smirnov’s Theorem on constructing global metrics for topological spaces whose
topologies locally come from metric spaces. This is supplementary material not covered in the
course, and the level of exposition is slightly higher than in the course notes.
stereopic1.pdf
A picture showing the relationship between stereographic projections and inversions with re-
spect to spheres.
stereopic2.pdf
More pictures involving stereographic projections.

148
 swisscheese.pdf
A picture illustrating how a square with two holes can be viewed as a compactification of the
Euclidean plane.
synthetic-geom.pdf
A discussion of how modernized axiomatic settings for classical synthetic geometry (the ap-
proach in Euclid’s Elements) can be used to define topological structures on systems which satisfy
appropriate sets of axioms.
topschroederbernstein.pdf
Examples which show that the dual Schröder-Bernstein Theorem (see dualSBtheorem.pdf
above) is systematically false for topological spaces and continuous mappings.
tubelemma.pdf
Drawing for the proof of the Tube Lemma in Section III.1.
uniform-complete.pdf
This document proves a basic result on uniformly continuous homeomorphisms of complete
metric spaces and applies it to conclude that all of the d p product metrics are complete if the
metrics on the factors are complete.
uniqreals.pdf Proof that the usual axioms for the real number system (a complete ordered
field) uniquely characterize the real numbers up to isomorphism. (see also maximality.pdf).
wallace-disamb.pdf
Disambiguation for the two mathematicians named A. Wallace who made highly significant
contributions to topology.
zerogradient.pdf
An application of connectedness to show that a function on an open subset of R n is constant
on a connected open region in Rn if it has vanishing partial derivatives on that set; this is stated
in most multivariable calculus texts, but frequently these books only prove the result for special
classes of open sets such as open disks or rectangular open subsets.

149