Topology Course Notes

Fall 2013 to Spring 2015

Gary Gruenhage Mark Guest

2019

Auburn University
 c 2019 by Gary Gruenhage & Mark Guest. This work may be
downloaded for personal use. Publishing it elsewhere or selling it is
strictly forbidden.

The 350 theorems herein are theorems stated by Gruenhage in

IBL style graduate topology courses over a two year period; most
topics covered the first year are standard, while more specialized topics
in set-theoretic topology were covered during the second year. The
organization of these theorems, together with hints and proofs, into
this book is due to Guest. A large percentage of the proofs are due to
Guest, with other proofs due to other students in the classes.
Contents
1 The Basics 1
1.1 Some Set Theory We Should Know . . . . . . . . . . . . 1
1.1.1 Cardinality & Cardinal Numbers . . . . . . . . . 1
1.1.2 Ordinal Numbers . . . . . . . . . . . . . . . . . . 2
1.1.3 Transfinite Induction . . . . . . . . . . . . . . . . 4
1.1.4 Axiom of Choice & Equivalents . . . . . . . . . . 6
1.2 Topological Spaces . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Closed Sets . . . . . . . . . . . . . . . . . . . . . 7
1.2.2 Interiors & Boundaries . . . . . . . . . . . . . . . 8
1.2.3 Bases . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.4 Subspaces . . . . . . . . . . . . . . . . . . . . . . 10
1.2.5 Continuous Functions . . . . . . . . . . . . . . . 10
1.2.6 Homeomorphisms . . . . . . . . . . . . . . . . . . 12
1.2.7 Convergent Sequences . . . . . . . . . . . . . . . 12
1.3 Some Types of Spaces . . . . . . . . . . . . . . . . . . . 12
1.3.1 Metric Spaces . . . . . . . . . . . . . . . . . . . . 12
1.3.2 Product Spaces . . . . . . . . . . . . . . . . . . . 13
1.3.3 Quotient Spaces . . . . . . . . . . . . . . . . . . 15

2 Topological Properties 17
2.1 Separation Axioms . . . . . . . . . . . . . . . . . . . . . 17
2.1.1 T0 , T1 , and T2 . . . . . . . . . . . . . . . . . . . 17
2.1.2 T3 , T3.5 , and T4 . . . . . . . . . . . . . . . . . . . 17
2.1.3 Urysohn’s Lemma . . . . . . . . . . . . . . . . . 18
2.1.4 Completely Normal . . . . . . . . . . . . . . . . 19
2.1.5 Perfectly Normal . . . . . . . . . . . . . . . . . . 19
2.2 Covering Properties & Countability Axioms . . . . . . . 20
2.2.1 Compact and Lindelöf . . . . . . . . . . . . . . . 20
2.2.2 Filters & Tychonoff’s Theorem . . . . . . . . . . 22

iii
Contents

2.2.3 Countably Compact . . . . . . . . . . . . . . . . 23

2.2.4 First-Countable . . . . . . . . . . . . . . . . . . . 24
2.2.5 Separable & Second-Countable . . . . . . . . . . 24
2.2.6 Sequentially Compact . . . . . . . . . . . . . . . 26
2.2.7 Locally Compact & Baire Spaces . . . . . . . . . 26
2.3 Connectedness . . . . . . . . . . . . . . . . . . . . . . . 28
2.3.1 Connected & Path Connected . . . . . . . . . . . 28
2.3.2 Components & Path Components . . . . . . . . . 29
2.3.3 Locally Connected & Locally Path Connected . . 30
2.3.4 Quasicomponents . . . . . . . . . . . . . . . . . . 30
2.3.5 Some Theorems . . . . . . . . . . . . . . . . . . . 31
2.4 Paracompactness . . . . . . . . . . . . . . . . . . . . . . 32
2.4.1 Point-Finite & Locally Finite Collections . . . . 32
2.4.2 Point-Finite & Locally Finite Refinements . . . . 32
2.4.3 Partitions of Unity . . . . . . . . . . . . . . . . . 33
2.5 Modifications of Some Properties . . . . . . . . . . . . . 33
2.5.1 The Countable Chain Condition . . . . . . . . . 33
2.5.2 Collectionwise Normal . . . . . . . . . . . . . . . 34
2.5.3 Monotonically Normal . . . . . . . . . . . . . . . 35

3 More About Metrizable Spaces 37

3.1 Metrization Theorems . . . . . . . . . . . . . . . . . . . 37
3.1.1 Urysohn’s Metrization Theorem . . . . . . . . . . 37
3.1.2 Metrizable Spaces Are Paracompact . . . . . . . 38
3.1.3 Nagata-Smirnov Metrization Theorem . . . . . . 39
3.1.4 Gδ -Diagonals . . . . . . . . . . . . . . . . . . . . 39
3.1.5 Networks . . . . . . . . . . . . . . . . . . . . . . 40
3.1.6 Perfect Images . . . . . . . . . . . . . . . . . . . 40
3.2 Completeness . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2.1 Complete Metrics . . . . . . . . . . . . . . . . . . 41
3.2.2 C ∗ (X) . . . . . . . . . . . . . . . . . . . . . . . . 42
3.2.3 Isometries . . . . . . . . . . . . . . . . . . . . . . 43

4 More About Linearly Ordered Spaces 45

4.1 The Cantor Set . . . . . . . . . . . . . . . . . . . . . . . 45
4.1.1 The Construction . . . . . . . . . . . . . . . . . . 45

iv
 Contents

4.1.2 A Space-Filling Curve . . . . . . . . . . . . . . . 46

4.2 Characterizing Some Ordered Spaces . . . . . . . . . . . 46
4.2.1 R & [0, 1] . . . . . . . . . . . . . . . . . . . . . . 46
4.2.2 Q . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2.3 Compact Spaces . . . . . . . . . . . . . . . . . . 47
4.3 The Ordinal ω1 . . . . . . . . . . . . . . . . . . . . . . . 48
4.3.1 Closed & Unbounded Subsets . . . . . . . . . . . 48
4.3.2 Stationary Subsets . . . . . . . . . . . . . . . . . 48
4.4 Characterizing Paracompact Ordered Spaces . . . . . . 49
4.4.1 Cofinality of Limit Ordinals . . . . . . . . . . . . 50
4.4.2 Club & Stationary Subsets of Cardinals . . . . . 50
4.4.3 The Characterization . . . . . . . . . . . . . . . 51
4.5 Hereditarily Lindelöf but not Separable . . . . . . . . . 51
4.5.1 ccc Linearly Ordered Spaces . . . . . . . . . . . . 51
4.5.2 Suslin Lines . . . . . . . . . . . . . . . . . . . . . 52

5 Some Set-Theoretic Topology 53

5.1 Trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.1.1 The Cantor Tree . . . . . . . . . . . . . . . . . . 53
5.1.2 The Irrationals . . . . . . . . . . . . . . . . . . . 54
5.1.3 Aronszajn and Suslin Trees . . . . . . . . . . . . 55
5.2 Martin’s Axiom . . . . . . . . . . . . . . . . . . . . . . . 56
5.2.1 The Statement . . . . . . . . . . . . . . . . . . . 56
5.2.2 Working with MA(κ) . . . . . . . . . . . . . . . . 57
5.2.3 First Category Subsets of R . . . . . . . . . . . . 59
5.3 Small Cardinals . . . . . . . . . . . . . . . . . . . . . . . 59
5.3.1 Unbounded and Dominating Families in ω ω . . . 59
5.3.2 Subsets of the Irrationals . . . . . . . . . . . . . 60
5.4 Normal vs. Collectionwise Normal . . . . . . . . . . . . 61
5.4.1 Tangent Disc Spaces . . . . . . . . . . . . . . . . 61
5.4.2 Q-Sets . . . . . . . . . . . . . . . . . . . . . . . . 62
5.5 Countably Compact vs. Pseudocompact . . . . . . . . . 63
5.5.1 Almost Disjoint Families . . . . . . . . . . . . . . 63
5.5.2 Mrówka Spaces . . . . . . . . . . . . . . . . . . . 63
5.6 Hereditarily Separable vs. Hereditarily Lindelöf . . . . . 64
5.6.1 Some History . . . . . . . . . . . . . . . . . . . . 64

v
Contents

5.6.2 The Kunen Line . . . . . . . . . . . . . . . . . . 64

6 The Stone-Čech Compactification 67

6.1 Hausdorff Compactifications . . . . . . . . . . . . . . . . 67
6.1.1 One-Point Compactifications . . . . . . . . . . . 67
6.1.2 Compactification by Adding Points . . . . . . . . 68
κ
6.1.3 Compactification by Embedding in [0, 1] . . . . 68
6.1.4 The Stone-Čech Compactification . . . . . . . . . 69
6.1.5 Four More Results . . . . . . . . . . . . . . . . . 70
6.2 The Topology of βω . . . . . . . . . . . . . . . . . . . . 71
6.2.1 ω ∗ is not separable . . . . . . . . . . . . . . . . . 73
6.2.2 A strange property of βω . . . . . . . . . . . . . 73
6.2.3 Closed subsets of βω . . . . . . . . . . . . . . . . 74
6.2.4 P -points . . . . . . . . . . . . . . . . . . . . . . . 74
6.2.5 ω ∗ is not homogeneous . . . . . . . . . . . . . . . 75
6.2.6 Products of countably compact spaces . . . . . . 75
6.2.7 Products of sequentially compact spaces . . . . . 76
6.2.8 Characterizing βX . . . . . . . . . . . . . . . . . 77
6.2.9 Continua . . . . . . . . . . . . . . . . . . . . . . 79

7 Topological Games 81
7.1 Completeness & the Baire Property . . . . . . . . . . . 81
7.1.1 Complete metric spaces . . . . . . . . . . . . . . 81
7.1.2 Čech complete . . . . . . . . . . . . . . . . . . . 82
7.1.3 The Baire property . . . . . . . . . . . . . . . . . 83
7.2 The Choquet game . . . . . . . . . . . . . . . . . . . . . 85
7.2.1 Strategies . . . . . . . . . . . . . . . . . . . . . . 85
7.2.2 Choquet Spaces . . . . . . . . . . . . . . . . . . . 86
7.2.3 Products of Baire Spaces . . . . . . . . . . . . . 87
7.3 W -Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 88
7.3.1 The W -Game . . . . . . . . . . . . . . . . . . . . 88
7.3.2 w-spaces . . . . . . . . . . . . . . . . . . . . . . . 90
7.3.3 w but not W . . . . . . . . . . . . . . . . . . . . 91

8 Hints 93

9 Proofs 101

vi
1 The Basics
1.1 Some Set Theory We Should Know
1.1.1 Cardinality & Cardinal Numbers
Two sets A and B are said to have the same cardinality, and we write
|A| = |B|, if there exists a bijection f : A → B. We also say |A| 6 |B| if
there exists an injection f : A → B. The Schröder-Bernstein Theorem
says: If there are one-to-one maps f : A → B and g : B → A, then
|A| = |B|.
A set is called countable if it is either finite or has the same cardinality
as the set N of positive integers. To specify countable but not finite, we
say countably infinite.

Theorem 1.
(a) A countable union of countable sets is countable.

(b) If A1 , A2 , . . . , An are countable, so is A1 × · · · × An .

(c) If A is countable, so is the set of all finite subsets of A, as well

as the set of all finite sequences of elements of A.

(d) The set Q of all rational numbers is countable.

Theorem 2. The following sets have the same cardinality as the set R
of real numbers:
(a) the set P(N) of all subsets of the natural numbers N;

(b) the set of all functions f : N → {0, 1};

(c) the set of all infinite sequences of 0’s and 1’s;

1
 1 The Basics

(d) the set of all infinite sequences of real numbers.

The cardinality of N (and any countable infinite set) is denoted by

ℵ0 . ℵ1 denotes the next infinite cardinal, ℵ2 the next, etc. Cantor’s
Continuum Hypothesis (CH) says that the cardinality of R is ℵ1 ; it
turns out that CH is undecidable from the usual axioms (ZFC) of set
theory. The next theorem shows that for any cardinal, there is a bigger
one.

F Theorem 3. For any set X, we have |X| < |P(X)|, where P(X) is the
set of all subsets of X.

If the set X has cardinality κ, then the cardinal 2κ is defined to be

the cardinality of the set P(X)—which has the same cardinality as the
set of all functions from X into {0, 1}, by the same argument that gets
Theorem 2(b). So 2ℵ0 is the cardinality of R, and CH can be stated as
2 ℵ0 = ℵ 1 .

1.1.2 Ordinal Numbers

A relation R on a set X is a subset of X × X; we write xRy to mean
hx, yi ∈ R. A (strict) partial order 1 on X is a relation < satisfying, for
every x, y, z ∈ X:
(1) x < y and y < x are not both true, so < is antireflexive; and

(2) x < y and y < z implies x < z, so < is transitive.

If < is a relation, we may write x 6 y to mean “x < y or x = y.” We
say that < is a linear order if it satisfies
(3) the trichotomy property: for any x, y ∈ X, exactly one of x < y,
y < x, or x = y holds.
And finally, a linear order < is a well-order if
(4) every non-empty subset of X has a <-least element.
1A non-strict partial order on X is a relation 6 on X which is reflexive (so x 6 x
for all x), antisymmetric (so x 6 y and y 6 x if and only if x = y), and transitive.
We will just say that a set is partially ordered and let context indicate if the
relation is strict or not.

2
 1.1 Some Set Theory We Should Know

If the well-order is understood, we may simply say a least element.

Two linearly ordered sets hA, <A i and hB, <B i are order-isomorphic
if there is an order-preserving bijection h : A → B; that is, x <A y if
and only if h(x) <B h(y).
We will assume the following:

Theorem 4. There is a class Ord and a linear ordering < on Ord

with the following properties:
(a) Every nonempty subset of Ord has a least element;
(b) Given any well-ordered set hX, ≺i, there exists a unique α ∈ Ord
such that hX, ≺i is order-isomorphic to hPα , <i, where Pα is the
set {β ∈ Ord : β < α} of predecessors of α.

Elements of Ord are called ordinals. Intuitively, an ordinal number

denotes position in a well-ordered sequence. The least ordinal is denoted
by 0, the next by 1, the next by 2, etc. The least ordinal with infinitely
many predecessors, that is, the least ordinal greater than all the finite
ordinals 0, 1, 2, . . ., is denoted by ω. Then comes ω + 1, ω + 2, . . . . The
least ordinal bigger than ω + n for every n < ω is denoted by ω · 2. One
can similarly define ω · 3, ω · 4, . . . . The least ordinal bigger than ω · n
for all n < ω is denoted by ω · ω = ω 2 . Etc.
Note that all ordinals mentioned in the above paragraph have only
countably many predecessors. Such ordinals are called countable ordi-
nals. The least ordinal with uncountably many predecessors, that is,
the least ordinal greater than all of the countable ordinals, is denoted
by ω1 . A better definition of ℵ1 than the one given previously is that
ℵ1 denotes the cardinality of the set of predecessors of ω1 . ℵ2 , ℵ3 , . . .
may be defined similarly. See below where ℵα is defined by transfinite
induction for any ordinal α. Denote by Card the class of all cardinals
ℵ0 , ℵ1 , ℵ2 , . . ..
The position in modern set theory, and the position we shall take
advantage of, is that an ordinal is the set of its predecessors. Thus
0 = ∅, 1 = {0}, 2 = {0, 1}, . . . , ω = {0, 1, 2, . . .}, ω + 1 = ω ∪ {ω}, etc.
Moreover, for α, β ∈ Ord we have α < β if and only if α ∈ β. This
viewpoint gives us the following, though our limited definitions may
make it difficult to prove without the hint.

3
 1 The Basics

S
F Theorem 5. If A is a set of ordinals, then sup A = A is the least
ordinal greater than every member of A.

Corollary 6. If C is any countable subset of ω1 , then there is a countable

ordinal α such that α > β for all β ∈ C.

The following notation will be used throughout. If A and B are sets,

α is an ordinal, and κ is a cardinal, then
κ
(1) [A] = {X ⊂ A : |X| = κ};
<κ S λ
(2) [A] = {[A] : λ < κ};
6κ <κ κ
(3) [A] = [A] ∪ [A] ;

(4) B A is the set of all functions from A to B, so Aα is the set of all

functions from α to A (which is also the set of all sequences of
length α in A);

(5) A<α = {Aβ : β < α}; and

S

(6) A6α = A<α ∪ Aα .

1.1.3 Transfinite Induction

Theorem 7, Principle of Transfinite Induction. Let κ be an ordinal.
Suppose we have for each α < κ a statement Sα such that:
(1) S0 is true;

(2) If β < κ and Sα is true for every α < β, then Sβ is true.

Then Sγ is true for every γ < κ.

F Theorem 8, Principle of Definition by Transfinite Induction. Let κ be

an ordinal, B a set, and F the set of all functions (including the empty
function) into B whose domain is the set {β ∈ Ord : β < α} for some
α < κ. Let R : F → B. Then there is a unique function G with domain
{α ∈ Ord : α < κ} such that G(α) = R(G  {β ∈ Ord : β < α}) for
all α < κ.

4
 1.1 Some Set Theory We Should Know

Interpret this as follows: Let f ∈ F. Then f is a function whose

domain is the predecessors of some ordinal α. Think of f as a possibility
for what you have done in the induction prior to α. Then R is a “recipe”
for what you then do at step α: you do R(f ). The conclusion of the
principle says that there is a unique function G whose domain is the
predecessors of κ (which is the ordinal you are inducting on; κ = ω in
ordinary induction) such that G(α) is always what the recipe tells you
to do at step α, given G(β) for all β < α.
What do you do at the first step in the induction? Well, you haven’t
done anything yet, so you do R(∅). And note that
G(0) = R(G  {β ∈ Ord : β < 0}) = R(G  ∅) = R(∅).
So given R, G(0) is uniquely defined! What next? Note that we want
G(1) = R(G  {β : β < 1}) = R(G  {0}), which we see now is uniquely
defined since G(0) is uniquely defined. And so on.
The Transfinite Induction Principle and the Principle of Definition
by Transfinite Induction also hold if in their statements we replace κ
by the class Ord of all ordinals.
Now the Principle of Definition by Transfinite Induction justifies
defining ωα and ℵα for any ordinal α as follows: ℵ0 is as defined previ-
ously, and ω0 = ω. Suppose α is an ordinal and ωβ and ℵβ have been
defined for every β < α.
(1) If α has an immediate predecessor γ (so α = γ + 1), then let ωα
be the least ordinal with more than ℵγ -many predecessors, and
let ℵα denote the cardinality of ωα .
(2) If α has no immediate predecessor, in which case α is called a
limit ordinal , then let ωα be the least ordinal greater than ωβ for
every β < α. Again, let ℵα denote that cardinality of the set of
predecessors of ωα .

Exercise 9. For α < ω1 , define collections Bα of subsets of a set X as

follows: B0 is an arbitrary collection of subsets ofTX. If Bα has been
6ℵ
defined, then let Bα+1 = Bα ∪ {X \ A : AS∈ Bα } ∪ { A : A ∈ [Bα ] 0 }.
S is a limit ordinal let Bα = β<α Bβ .
Finally, if α
Let B = α<ω1 Bα . Prove that B is closed under complements, count-
able intersections, and countable unions.

5
1 The Basics

If X is a topological space and B0 is the collection of all open sets

of X, then B is precisely the collection of all Borel subsets of X. A
standard definition of this collection is that it is the smallest σ-algebra
containing the open sets, where a collection B of subsets of a set X is
a σ-algebra if it is closed under complements, countable intersections,
and countable unions.

1.1.4 Axiom of Choice & Equivalents

The Axiom of Choice (AC) says: givenS any collection A of nonempty
sets, there is a function f : A → A such that f (A) ∈ A for every
A ∈ A; such an f is called a choice function for A.
Zorn’s Lemma (ZL) works as follows. Suppose hP, <i is a partially
ordered set with the following property: If L is any subset of P linearly
ordered by <, then there is p ∈ P such that ` 6 p for any ` ∈ L.
Then there is an element q in P such that no member of P is strictly
greater than q. Colloquially: if every chain in the partial order hP, <i
is bounded above in P , then P has a maximal element.

Theorem 10. The following are equivalent:

(a) The Axiom of Choice;

(b) Zorn’s Lemma;

(c) Every set can be well-ordered.

1.2 Topological Spaces

A topological space is an ordered pair hX, T i, where X is a set and T
is a collection of subsets of X such that
(1) ∅ ∈ T and X ∈ T ;

(2) U ∩ V ∈ T whenever U, V ∈ T ; and

S
(3) U ∈ T whenever U ⊂ T .

6
 1.2 Topological Spaces

T is called a topology on the set X; elements of T are called open

sets. To put it in other words, (2) says that pairwise (and hence finite)
intersections of open sets are open, and (3) says that any union of open
sets is open.
We will often write X is a topological space or just X is a space to
mean that X is a set with some fixed but unspecified topology; if we
say U is open in X, we mean that U is a member of whatever topology
on X we happen to be working with.

1.2.1 Closed Sets

If X is a space, and A ⊂ X, then A is closed if and only if X \ A is
open.

Lemma 11, De Morgan’s Laws. If X is any set and A is any collec-

tion of subsets of X, then
S T
(a) X \ A = {X \ A : A ∈ A} and
T S
(b) X \ A = {X \ A : A ∈ A}.

Corollary 12. In any space X, the sets ∅ and X are closed, the union
of finitely many closed sets is closed, and the intersection of any number
of closed sets is closed.

Let X be a space and A ⊂ X. The closure of A, denoted A or cl A,

is the intersection of all closed sets containing A; note A is closed by
Corollary 12, and A is closed if and only if A = A. We can think of the
closure of A as being the “smallest” closed set containing A.
If x ∈ X and every open set containing x contains a point of A other
than x, we say that x is a limit point of A; the set of limit points of A
is denoted by Ad or lim A.

Theorem 13. For any subset A of a space X we have A = A ∪ Ad .

Theorem 14. If A and B are subsets of a space X, then

(a) ∅ = ∅,

7
1 The Basics

(b) A ⊂ B → A ⊂ B,

(c) A = A, and

(d) A ∪ B = A ∪ B.

Theorem 15. If {Aα : α ∈ Λ} is any collection of subsets of a space X,

then
T T
(a) α∈Λ Aα ⊂ α∈Λ Aα and
S S
(b) α∈Λ Aα ⊃ α∈Λ Aα

1.2.2 Interiors & Boundaries

The interior of A, denoted A◦ or int A, is the union of all open sets
contained in A; we can think of the interior of A as the “largest” open
set contained in A. If x ∈ A◦ we say that A is a neighborhood of x, so
A consists of those points whose every neighborhood meets A.

Theorem 16. For any subset A of a space X:

(a) A◦ = X \ X \ A and
◦
(b) A = X \ (X \ A) .

The boundary of A, denoted ∂A or bd A, consists of those points

whose every neighborhood meets both A and X \ A.

Theorem 17. For any subset A of a space X:

(a) ∂∂A ⊂ ∂A;

(b) ∂A = A ∩ X \ A = A \ A◦ , whence ∂A is always closed;

(c) A is closed if and only if ∂A ⊂ A;

(d) ∂A ∩ A◦ = ∅, and A is open if and only if ∂A ∩ A = ∅;

(e) A is closed-and-open if and only if ∂A = ∅

8
 1.2 Topological Spaces

The isolated points of A are those x ∈ A having a neighborhood U

for which A ∩ U = {x}; if every point of A is isolated we say A is a
discrete subset of X, and if every point of X is isolated then X has the
discrete topology.

Theorem 18. Let A be a subset of a space X, and let A0 consist of the

isolated points of A. Then
(a) Ad ∩ A0 = ∅,

(b) A = Ad ∪ A0 , and

(c) the isolated points of A are contained in A0 .

1.2.3 Bases
Let hX, T i be a topological space, and let B be a subcollection of T .
We say B is a base, and refer to its members as basic open sets, for the
topological space hX, T i if and only if every nonempty member of T is
a union of members of B.
A subcollection S of T is a subbase for hX, T i if the set of finite
intersections in S is a base for hX, T i.

Theorem 19. Let X be a set and let B be a collection of subsets of X.

Then there is a (unique) topology T on X such that B is a base for
hX, T i if and only if
(a) Each x ∈ X is in some member of B;

(b) Whenever B1 , B2 ∈ B and x ∈ B1 ∩ B2 , there exists B ∈ B with

x ∈ B ⊂ B1 ∩ B2 .

Collections B1 and B2 of subsets of a set X are equivalent bases if

and only if there is a topology T on X such that B1 and B2 are both
bases for hX, T i.

Theorem 20. Let hX, T i be a topological space, let B1 be a base for

hX, T i, and let B2 be a collection of subsets of X. If
(1) x ∈ B1 ∈ B1 → ∃B2 ∈ B2 (x ∈ B2 ⊂ B1 ), and

9
1 The Basics

(2) x ∈ B2 ∈ B2 → ∃B1 ∈ B1 (x ∈ B1 ⊂ B2 ),
then B1 and B2 are equivalent bases.

Theorem 21. Let B1 and B2 be bases for the topologies T1 and T2 ,

respectively, on a set X. Then T2 is finer than T1 if and only if every
member of B1 is a union of members of B2 .

1.2.4 Subspaces
A subspace of a topological space hX, T i is a pair hA, TA i, where A ⊂ X
and TA = {U ∩ A : U ∈ T }; TA is called the subspace topology on A,
the relative topology on A, or the topology on A induced by T .

Theorem 22. Let A be a subspace of a space X.

(a) H is closed in A if and only if H = A ∩ K for some K closed
in X.

(b) If U is open in A and A is open in X, then U is open is X.

(c) If K is closed in A and A is closed in X, then K is closed in

X.

Theorem 23. If hX, T i is a topological space, B a base for T , and

A ⊂ X, then the collection {B ∩ A : B ∈ B} is a base for hA, TA i.

Theorem 24. Let A be a subspace of a space X, and let H ⊂ A. Then

clA H = clX H ∩ A, where clA H indicates we are taking the closure of
H in the subspace A.

1.2.5 Continuous Functions

The following notation will help us differentiate between applying a
function to a point which is a set and applying a function to each point
of a set: if f : X → Y , A ⊂ X, and B ⊂ Y , then f [A] = {f (a) : a ∈ A}
and f −1 [B] = {x ∈ X : f (x) ∈ B}.
If f : X → Y , where X and Y are spaces, and x0 ∈ X, we say f is
continuous at x0 if, given any open set V in Y with f (x0 ) ∈ V , there is

10
 1.2 Topological Spaces

an open set U in X with x0 ∈ U and f [U ] ⊂ V . Also, f is continuous

if f is continuous at every x0 ∈ X.

Theorem 25. If f : X → Y , where X and Y are spaces, then the fol-

lowing are equivalent:
(a) f is continuous;

(b) f −1 [V ] is open in X whenever V is subbasic open in Y ;

(c) f −1 [V ] is open in X whenever V is basic open in Y ;

(d) f −1 [V ] is open in X whenever V is open in Y ;

(e) f −1 [K] is closed in X whenever K is closed in Y ; and

(f) whenever A ⊂ X and x ∈ A, then f (x) ∈ f [A] (in other words,

f [A] ⊂ f [A]).

Theorem 26. Let A be a subspace of X.

(a) The inclusion map j : A → X is continuous.

(b) If f : X → Y is continuous, then the restriction of f to A,

denoted f  A, is a continuous function from A to Y .

Theorem 27. Constant functions are continuous, and the composition

of continuous functions is continuous.

Theorem 28.
S
(a) Let f : X → Y , where X = α∈Λ Uα with each Uα open. If
f  Uα is continuous for each α ∈ Λ, then f is continuous.

(b) Let X = H ∪ K with H and K closed, and let f : H → Y

and g : K → Y be continuous. If f  (H ∩ K) = g  (H ∩ K),
then h : X → Y defined so that h  H = f and h  K = g is
continuous.

11
1 The Basics

1.2.6 Homeomorphisms
A function f : X → Y is closed (resp., open) if the image of every closed
set (resp., open set) in X is closed (resp., open) in Y .
A function h : X → Y is a homeomorphism if h is one-to-one, onto,
and both h and h−1 are continuous. If such a homeomorphism exists
then the spaces X and Y are said to be homeomorphic, denoted by
X∼=Y .

Theorem 29. If h : X → Y is one-to-one and onto, then the following

are equivalent:
(a) h is a homeomorphism;
(b) h is open and continuous; and
(c) h is closed and continuous.

Exercise 30. Find spaces X and Y and a mapping f : X → Y which

is one-to-one, onto, and continuous, but not a homeomorphism.

1.2.7 Convergent Sequences

Let X be a topological space, x ∈ X, and hxn in∈N a sequence of points
in X. We say hxn i converges to x, and write hxn i → x, if every (open)
neighborhood of x contains xn for sufficiently large n ∈ N; that is, for
any (open) neighborhood U of x, there is k ∈ N such that xn ∈ U for all
n > k. Alternatively, we may say that each neighborhood of x contains
a tail of the sequence.

Theorem 31. If f : X → Y is continuous and hxn in∈N → x in X, then

hf (xn )i → f (x) in Y .

1.3 Some Types of Spaces

1.3.1 Metric Spaces
Let X be a set. A function d : X × X → R is called a metric on X if,
for each x, y, z ∈ X, d satisfies:

12
 1.3 Some Types of Spaces

(1) d(x, y) > 0 and d(x, y) = 0 if and only if x = y;

(2) d(x, y) = d(y, x); and

(3) d(x, z) 6 d(x, y) + d(y, z).

Conditions (2) and (3) say that d is symmetric and satisfies the triangle
inequality, respectively.
A metric space is a pair hX, di, where X is a set and d is a metric on
X. If ε > 0, then Bd (x, ε) = {y ∈ X : d(x, y) < ε} is the ε-ball about x
(with respect to d); if it is understood what metric we are talking about,
we may omit the subscript d and write B(x, ε).

Lemma 32. Let hX, di be a metric space, and let y ∈ Bd (x, ε). Then
there exists δ > 0 such that Bd (y, δ) ⊂ Bd (x, ε).

Lemma 33. Let hX, di be a metric space. Then the collection of all
ε-balls is a base for a topology on X.

Let hX, di be a metric space; the topology on X given in Lemma 33

is called the metric topology (generated by the metric d). If d and d0
are two metrics on the same set X which generate the same topology,
then d and d0 are called equivalent metrics. A topological space hX, T i
is said to be metrizable if there is a metric on X which generates T .

Corollary 34. Let hX, di be a metric space. Then a subset U of X is

open in the metric topology if and only if, for every x ∈ U , there is
ε > 0 such that Bd (x, ε) ⊂ U .

Theorem 35. Let hX, di be a metric space, and let Y ⊂ X. Let dY be

the metric d restricted to Y ×Y . Then hY, dY i is a metric space, and the
metric topology on Y is the same as the subspace topology with respect
to the metric topology on X.

1.3.2 Product Spaces

Q
Let {hXα , Tα i : α ∈ Λ} be a collection of spaces; put X = α∈Λ Xα .
A point of X can be thought of as a sequence ~x = hxα iα∈Λ , where

13
1 The Basics

S
xα ∈ Xα for all α, or as a function f : Λ → α∈Λ Xα with f (α) ∈ Xα
for all α.
Q
Denote by B be the collection of all sets of the form α∈Λ Uα , where
Uα ∈ Tα for all α, and Uα = Xα for cofinitely2 many α. Then B is a
base for a topology on X; this topology is called the product topology,
or the Tychonoff (product) topology, and X with this topology is called
the product space of {hXα , Tα i : α ∈ Λ}. Each Xα is called a coordinate
space of X.
If F is a finite subset of Λ,
Q we may write
Q a typical basic open set in
the product space as B = α∈F Uα × α∈Λ\F Xα , where Uα is open
in Xα . The finite set F is sometimes called the support of the basic
open set B.
Observe: if for Q each α, Bα is a base for Xα , then the collection of all
sets of the form α∈Λ Bα , where Bα = Xα for cofinitely many α and
Bα ∈ Bα if Bα 6= Xα , is an equivalent base for the product topology.
For each α ∈ Λ, the function πα : X → Xα defined by ~x 7→ xα
is called theQprojection map onto Xα . Note that if Uα ⊂ Xα , then
πα−1 [Uα ] = β∈Λ Vβ , where Vα = Uα and Vβ = Xβ if β 6= α. Thus
a
T typical basic open set in the product topology can be denoted by
−1
π
i6n αi [U αi ], where αi ∈ Λ and Uαi is open in Xαi for each i 6 n.
So, the collection

{πα−1 [Uα ] : α ∈ Λ ∧ Uα ∈ Tα }

is a subbase for the product topology.

Theorem 36. Projection maps are always continuous and open.

Q
Theorem 37. Let X = α∈Λ Xα be a product space, and let ~x ∈ X.
Fix γ ∈ Λ. Then the “cross-section” {~y ∈ X : ∀α 6= γ(yα = xα )} is
homeomorphic to Xγ .
Q
Theorem 38. Let X = α∈Λ Xα be a product space, and let Aα ⊂ Xα
Q Q
for all α ∈ Λ. Then α∈Λ Aα = α∈Λ Aα .

2 Or, all but finitely.

14
 1.3 Some Types of Spaces

Lemma 39. Let hX, di be a metric space and let M > 0. Define d0 by
d0 (x, y) = M if d(x, y) > M and d0 (x, y) = d(x, y) otherwise. Then d0
is an equivalent metric on X.

Theorem 40. Let {hXn , dn i : n ∈ N} be metric spaces such that, for

n
Q n ∈ N, dn (x, y) 6 1/2 for every x, y ∈ Xn . For ~x, ~y ∈ X =
each
x, ~y ) = max {dn (xn , yn ) : n ∈ N}. Then d is a met-
n∈N Xn , define d(~
ric on X and the topology generated by d is the same as the product
topology.

Corollary 41. A countable product of metrizable spaces is metrizable.

1.3.3 Quotient Spaces

Let X and Y be spaces, and q : X → Y a surjective map. Then q is
called a quotient map provided U is open in Y if and only if q −1 [U ] is
open in X. Note that quotient maps are continuous.

Theorem 42. Open continuous surjections and closed continuous sur-

jections are quotient maps.

Theorem 43. Let q : X → Y be a quotient map, and f : Y → Z arbi-

trary. Then f is continuous if and only if f ◦ q : X → Z is continuous.

Let q : X → Y be a quotient map. A subset U of X is called saturated

if U = q −1 [q[U ]]; equivalently, U meets a fiber q −1 (y) if and only if U
contains it.

Theorem 44. Let q : X → Y be a quotient map. Then the collection of

open sets of Y is equal to {q[U ] : U is a saturated open subset of X}.

Let hX, T i be a space, Y a set, and f : X → Y a surjection. Define

Tf = {U ⊂ Y : f −1 [U ] ∈ T }. Then Tf is easily seen to be a topology on
Y , under which f is a quotient map. Tf is called the quotient topology
induced by f .
Let X be a space, and let D be a partition of X (that is, D is a
pairwise-disjoint cover of X). Define p : X → D so that p(x) = D if

15
1 The Basics

and only if x ∈ D. Then D with the quotient topology induced by p is

called a quotient space of X. We say that D is the space obtained from
X by identifying points x, y ∈ X if they are in the same member of D.
If D has only one element A with more than one point, the resulting
quotient space is also denoted by X/A.
Let X be a space and let ∼ be an equivalence relation on X. Then
X/∼ denotes the quotient space of X where D is the collection of
∼-equivalence classes.
For example, let X = [0, 1] × [0, 1].
(1) If we identify hx, 0i with hx, 1i for each x ∈ [0, 1], the resulting
quotient space is a cylinder.

(2) If we also identify h0, yi with h1, yi for each y, the resulting quo-
tient space is a torus.

(3) If we identify as in (1), and also identify h0, yi with h1, 1 − yi,
the result is a Klein bottle.

(4) If we identify hx, 0i with h1 − x, 1i, the resulting quotient space

is a Möbius strip.

(5) If we identify as in (4), and also identify h0, yi with h1, 1 − yi, the
result is a real projective plane. We also get a projective plane if
we identify antipodal points of (the surface of) a sphere.

16
2 Topological Properties

2.1 Separation Axioms

2.1.1 T0 , T1 , and T2
A space X is a T0 -space if whenever x, y ∈ X with x 6= y, there is an
open set containing one of these points but not the other. If there is
always an open set containing x and missing y, then X is a T1 -space.
If there are always disjoint open sets containing x and y, then X is a
T2 -space or a Hausdorff space.

Theorem 45. A space X is a T1 -space if and only if every point of X

is a closed set.

2.1.2 T3 , T3.5 , and T4

A space X is a T3 -space, or regular space, if X is a T1 -space, and
whenever x ∈ X and H is a closed set not containing x, then there are
disjoint open sets containing x and H, respectively.

Theorem 46. A T1 -space X is regular if and only if

(a) whenever x ∈ X and H is a closed subset of X not containing
x, there exists an open set U containing x whose closure misses
H; or,

(b) whenever x is in an open V , there is an open set U containing

x such that U ⊂ V .

A space X is a T3.5 -space, or completely regular space, if X is a T1 -

space, and whenever x ∈ X and H is a closed set not containing x, there

17
 2 Topological Properties

is a continuous function f : X → [0, 1] with f (x) = 0 and f (y) = 1 for

all y ∈ H.
X is a T4 -space, or normal space, if X is a T1 -space and any two
disjoint closed sets are contained in disjoint open sets.

Theorem 47. A T1 -space X is normal if and only if

(a) whenever H and K are disjoint closed subsets of X, there exists
an open set U such that U ⊃ H and U ∩ K = ∅; or,

(b) whenever a closed set H is contained in an open set V , there is

an open set U containing H such that U ⊂ V .

Some authors don’t include T1 in the definitions of regular or normal,

and define T3 to be T1 plus regular, and T4 to be T1 plus normal. Still
others do as in the previous sentence but with T3 and regular swapped,
and T4 and normal swapped, in which case T3 and T4 don’t imply T1 !

Theorem 48. T4 → T3 → T2 → T1 → T0 and T3.5 → T3 .

Theorem 49. Metrizable spaces are normal.

Q
Theorem 50. Let X = α∈Λ Xα be a product space. If each Xα is a
Ti -space, where i = 0, 1, 2, 3 or 3.5, then so is X.

2.1.3 Urysohn’s Lemma

F Theorem 51. Suppose H and K are disjoint closed subsets of a normal
space X. Then one can assign to each rational number r with 0 < r < 1
an open set Ur such that
(a) H ⊂ Ur and U r ∩ K = ∅;

(b) r < r0 → U r ⊂ Ur0 .

Theorem 52, Urysohn’s Lemma. The following are equivalent for a

T1 -space X:
(a) X is normal;

18
 2.1 Separation Axioms

(b) Whenever H and K are disjoint closed subsets of X, there exists

a continuous function f : X → [0, 1] with f (x) = 0 for all x ∈ H
and f (x) = 1 for all x ∈ K.

Corollary 53. T4 → T3.5 ; that is, every normal space is completely

regular.

Theorem 54, Tietze Extension Theorem. Suppose H is a closed sub-

set of a normal space X. Then any continuous function f : H → R
can be extended to a continuous function fˆ: X → R. Furthermore, the
range of fˆ can be made to be contained in the convex hull of the range
of f .

2.1.4 Completely Normal

Two subsets A and B of a space X are separated if A ∩ B = ∅ and
B ∩ A = ∅; that is, no point of any one of the sets is in the closure of
the other set. A T1 -space X is said to be completely normal if given
any two separated sets A and B, there are disjoint open sets containing
A and B, respectively.

Theorem 55. A space X is completely normal if and only if every

subspace of X is normal.

Corollary 56. Metrizable spaces are completely normal.

If P is a property of topological spaces, we say that a space X is

hereditarily P if and only if every subspace of X has property P . For
example, “completely normal” is equivalent to “hereditarily normal”
(and the property is sometimes called that).

2.1.5 Perfectly Normal

Lemma 57. Suppose A and B are subsets of a space X, and that there
are open sets U1 , U2 , . . . and V1 , V2 , . . . such that:
S S
(a) A ⊂ n∈N Un and B ⊂ n∈N Vn ; and

19
2 Topological Properties

(b) for each n ∈ N, U n ∩ B = ∅ = V n ∩ A.

Then there are disjoint open sets U and V containing A and B, respec-
tively.

X is said to be perfectly normal if X is normal and every closed set

is Gδ (a countable intersection of open sets).

Theorem 58. Metrizable spaces are perfectly normal.

Theorem 59, the Vedenissoff Theorem. The following are equivalent

for a T1 -space X:
(a) X is perfectly normal;

T there is aTcollection {Un : n ∈ N}

(b) For any closed subset H of X,
of open sets such that H = n∈N Un = n∈N U n ;
(c) For any closed subset H of X, there exists a continuous function
f : X → [0, 1] such that H = f −1 (0);
(d) Whenever H and K are disjoint closed subsets of X, there exists
a continuous function f : X → [0, 1] such that H = f −1 (0) and
K = f −1 (1).

Theorem 60. Every subspace of a perfectly normal space is perfectly

normal, hence perfectly normal spaces are completely normal.

2.2 Covering Properties & Countability

Axioms
2.2.1 Compact and Lindelöf
A collection U of subsets of a space X is said to S be a cover of X if
every point of X is in some member of U, so X = U. If U is a cover
of X, then a subcollection V of U is a subcover if V is also a cover. A
cover U is an open cover if every member of U is open.
A space X is compact (resp., Lindelöf ) if every open cover of X has
a finite (resp., countable) subcover.

20
 2.2 Covering Properties & Countability Axioms

Theorem 61.
(a) Every continuous image of a compact (resp., Lindelöf ) space is
compact (resp., Lindelöf ).

(b) Every closed subset of a compact (resp., Lindelöf ) space is com-

pact (resp., Lindelöf ).

Theorem 62. Every compact subspace of a Hausdorff space is closed.

Exercise 63. Let α ∈ Ord. Then the set [0, α] = {β ∈ Ord : 0 6 β 6 α}

with the order topology (using the usual ordering on the ordinals) is a
compact Hausdorff space.

Note that [0, α] = α + 1 in our view of the ordinals.

Theorem 64. Suppose f : X → Y is continuous, one-to-one, and onto.

If X is compact and Y is Hausdorff, then f is a homeomorphism.

Exercise 65. Let X = [0, 2π], and let A = {0, 2π}. Show that X/A is
homeomorphic to the unit circle

S 1 = {hx, yi ∈ R2 : x2 + y 2 = 1}.

Similarly, the space obtained from the unit disk

B 2 = {hx, yi ∈ R2 : x2 + y 2 6 1}

by identifying its boundary S 1 to a single point is homeomorphic to the

2-sphere
S 2 = {hx, y, zi ∈ R3 : x2 + y 2 + z 2 = 1}.

Theorem 66. Suppose f : X → R is continuous, where X is a compact

space. Then there exists x0 ∈ X such that f (x0 ) > f (x) for every
x ∈ X.

Theorem 67.
(a) Every compact Hausdorff space is regular.

21
 2 Topological Properties

(b) Every compact Hausdorff space is normal.

F Theorem 68. Every regular Lindelöf space is normal.

F Example 69. The Sorgenfrey line S is a regular Lindelöf space, and is

completely normal, but S2 is neither Lindelöf nor normal.

Theorem 70. Let X be a compact metric space with metric d. If U is

an open cover of X, then there exists λ > 0 such that every ball of
radius λ is contained in some member of U.

Sometimes the conclusion of the previous theorem is stated in the

following way: “every set of diameter less than λ is contained in some
member of U.” The number λ is called a “Lebesgue number” of the
open cover.

Lemma 71, the Tube Lemma. Let X be a space, and let Y be compact.
Fix x0 ∈ X. SThen for any collection U of open sets in X × Y with
{x0 } × Y ⊂ U, there is a finite subcollection
S V of U and an open
neighborhood W of x0 such that W × Y ⊂ V.

Theorem 72. If X and Y are compact, so is X × Y .

2.2.2 Filters & Tychonoff’s Theorem

A collection F of subsets of a set X is said to have the finite intersection
property if the intersection of any finite subcollection of F is nonempty.

Theorem 73. A space X is compact if and only if forTevery collection

F of closed sets with the finite intersection property, F 6= ∅.

A collection F of subsets of a set X is called a filter on X if:

(1) F1 ∩ F2 ∈ F whenever F1 , F2 ∈ F;

(2) G ∈ F whenever X ⊃ G ⊃ F for some F ∈ F;

(3) ∅ ∈
/ F.

22
 2.2 Covering Properties & Countability Axioms

In other words, a filter is a collection of nonempty sets which is closed

under supersets and finite intersections. A filter F on X is called an
ultrafilter if F is not properly contained in any other filter on X.

Theorem 74. Let F be a filter on X. Then the following are equivalent:

(a) F is an ultrafilter;

(b) If G ⊂ X and G ∩ F 6= ∅ for every F ∈ F, then G ∈ F;

(c) For every G ⊂ X, either G ∈ F or X \ G ∈ F.

Theorem 75. Suppose F is a collection of subsets of X having the finite

intersection property. Then F is contained in some maximal collection
G of subsets of X with the finite intersection property; furthermore, any
such G is an ultrafilter on X.

Let X be a space, p ∈ X, and F a filter on X. We say that F

clusters at p if p ∈ F for every F ∈ F, and we say F converges to p if
every neighborhood of p contains a member of F (equivalently, every
neighborhood of p is in F).

Theorem 76. The following are equivalent for a space X:

(a) X is compact;

(b) Every filter on X clusters at some point;

(c) Every ultrafilter on X converges to some point.

Theorem 77, Tychonoff’s Theorem. Any product of compact spaces

is compact.

2.2.3 Countably Compact

X is countably compact if every countable open cover of X has a finite
subcover.

Theorem 78. A space X is compact if and only if X is both countably

compact and Lindelöf.

23
2 Topological Properties

Theorem 79.
(a) If X is countably compact, then every infinite subset of X has
a limit point in X.

(b) If X is a T1 -space, then X is countably compact if and only if

every infinite subset of X has a limit point in X.

2.2.4 First-Countable
A local base at a point x of a space X is a collection Bx of open
neighborhoods of x such that, whenever x ∈ U where U is open, there
is some B ∈ Bx with x ∈ B ⊂ U . A space X is first-countable if every
point of X has a countable local base.
T
If {Un : n ∈ N} is a local base at x and we put Vn = {Ui : i 6 n},
then hVn in∈N is another countable local base at x which is decreasing
(in the sense that i < j implies Vi ⊃ Vj ). So a point has a countable
local base if and only if it has a countable decreasing local base.

Theorem 80. Let X be a first-countable space. A point x is in the

closure of a subset A of X if and only if there is a sequence han in∈N in
A such that han i → x.

Theorem 81. If X is first-countable, then f : X → Y is continuous if

and only if hf (xn )i → f (x) in Y whenever hxn in∈N → x in X.

Theorem 82. Every metric space is first-countable.

Exercise 83. Let X = N × [0, 1] with its topology as a subspace of the

plane. Note that each {n}×[0, 1] is open and closed in X; we may say X
is a topological sum of countably many copies of [0, 1]. Let A = N×{0}.
Describe the topology of X/A (that is, what do the neighborhoods of
points of X/A look like?) Is X/A metrizable?

2.2.5 Separable & Second-Countable

A subset D of a topological space X is dense in X if X = D; equivalently,
every non-empty open set contains a point of D. X is separable if it

24
 2.2 Covering Properties & Countability Axioms

has a countable dense subset. A space X is second-countable if it has a

countable base.

Theorem 84. Every subspace of a second-countable space is second-

countable, separable, and Lindelöf.

Theorem 85. The following are equivalent for a metrizable space X:

(a) X is second-countable,

(b) X is Lindelöf,

(c) X is separable,

(d) X is hereditarily separable, and

(e) X is hereditarily Lindelöf.

Theorem 86. If Xn is a first-countable

Q (resp., second-countable) space
for each n ∈ N, then X = n∈N Xn is first-countable (resp., second-
countable).

Example 87. Let 2 denote a two-point discrete space (with elements 0

and 1, say). Let Λ be any uncountable index set. Then 2Λ is neither
first- nor second-countable.
Q
Theorem 88. If Xn is separable for each n ∈ N, then n∈N Xn is
separable.

Let A be a subset of a space X and let x ∈ X. If every neighborhood

of x contains uncountably many points of A, then we call x a complete
accumulation point of A.

Theorem 89.
(a) Every infinite subset of a compact space X has a limit point in
X.

(b) Every uncountable subset A of a Lindelöf space X has a complete

accumulation point in X.

25
2 Topological Properties

Corollary 90. For every uncountable subset A of a separable metric

space, there is a complete accumulation point a of A such that a ∈ A.

2.2.6 Sequentially Compact

A space X is sequentially compact if every infinite sequence in X has
an infinite convergent subsequence.

Theorem 91.
(a) Sequentially compact spaces are countably compact.

(b) If X is first-countable, then X is countably compact if and only

if X is sequentially compact.

Theorem 92. The following are equivalent for a separable metric space
X:
(a) X is compact;

(b) X is countably compact;

(c) X is sequentially compact.

The previous theorem is in fact true for any metric space, separable
or not, but the proof is more difficult.

Exercise 93. Give the set ω1 of all countable ordinals the order topology
(using the usual order on the ordinals). Then ω1 with this topology is
sequentially compact (and so also countably compact) but not compact.

2.2.7 Locally Compact & Baire Spaces

A space X is locally compact if every point has a compact neighborhood;
that is, for each x ∈ X, there is a compact subset N of X such that
x ∈ N ◦.

Theorem 94. Let hX, T i be a locally compact non-compact Hausdorff

space. Then there is a compact Hausdorff space hX ∗ , T ∗ i satisfying:

26
 2.2 Covering Properties & Countability Axioms

(a) X ∗ = X ∪ {∞}, where ∞ ∈

/ X;

(b) TX∗ = T (that is, the relative topology on X as a subspace of X ∗

is the same as T ); and

(c) U is an open neighborhood of ∞ in X ∗ if and only if X ∗ \ U is

a compact subset of X.

The space hX ∗ , T ∗ i in the previous theorem is called the one-point

compactification of the locally compact Hausdorff space hX, T i.

Theorem 95. Every locally compact Hausdorff space is completely reg-

ular.

Lemma 96. Let X be a locally compact Hausdorff space. Then the col-
lection of compact neighborhoods of a point x in X is a local neighbor-
hood base at x (that is, every open set containing x contains a compact
neighborhood of x).

Lemma 97. Suppose hHn in∈N is a decreasing sequence of non-empty

T
closed sets in a space X, and H1 is (countably) compact. Then n∈N Hn
is non-empty.

Theorem 98. If {Gn : n ∈ N} Tare dense open subsets of a locally com-

pact Hausdorff space X, then n∈N Gn is dense.

A space X with the property of Theorem 98 is called a Baire space.

So all locally compact Hausdorff spaces, in particular Rn for all n ∈ N,
are Baire spaces.

Lemma 99. Every open subspace of a Baire space is Baire.

S
Theorem 100. Suppose X is a Baire space, and X = n∈N Xn . Then
for some n ∈ N, int (cl Xn ) =
6 S ∅; that is, some Xn is dense in some
open subset of X. Moreover, n∈N int (cl Xn ) is dense in X.

27
2 Topological Properties

2.3 Connectedness
2.3.1 Connected & Path Connected
A space X is connected if and only if X cannot be written as the union
of two disjoint nonempty open sets; equivalently, X has no proper
nonempty subset which is both open and closed.

Theorem 101. Let hX, T i be a topological space, and let C ⊂ X. Then

the following are equivalent:
(a) hC, TC i is connected;

(b) C is not the union of two nonempty separated sets in X.

It follows from Theorem 101 (by taking C = X) that a space X is

connected if and only if X is not the union of two nonempty separated
sets. In fact, this is often taken as the definition of connected.

Theorem 102. Let X be a space, A ⊂ X, and suppose A ⊂ B ⊂ A. If

A is connected, so is B. In particular, the closure of a connected set is
connected.

Theorem 103. A subspace X of the real line R is connected if and only

if X is convex (that is, a < b ∈ X implies [a, b] ⊂ X).

It is easily seen that a subset of R is convex if and only if it is an

interval (infinite intervals such as (a, ∞) or (−∞, b) allowed).
A space X is path connected if and only if for any two points a, b in
X, there is a continuous function f : [0, 1] → X such that f (0) = a and
f (1) = b. The function f is called a path from a to b.

Theorem 104. The continuous image of a connected (resp., path con-

nected) space is connected (resp., path connected).

Theorem 105. Suppose {Aα : α ∈ Λ} is a collection of connected (resp.,

path connected)
S subspaces of a space X, and for any α, β ∈ Λ, Aα ∩Aβ 6=
∅. Then α∈Λ Aα is connected (resp., path connected).

28
 2.3 Connectedness

Theorem 106.
(a) If X and Y are connected spaces, so is X × Y .
(b) Any product of connected spaces is connected.

Theorem 107. Every path connected space is connected.

Q
Theorem 108. Let X = α∈Λ Xα be a product space, and let Y be any
space. A function f : Y → X is continuous if and only if πα ◦f : Y → Xα
is continuous for each α ∈ Λ.

Theorem 109. Any product of path connected spaces is path connected.

Lemma 110. Let X and Y be topological spaces, and let f : X → Y be

continuous. Let G = {hx, f (x)i : x ∈ X} ⊂ X × Y be the “graph” of f .
Then G is homeomorphic to X.

Example 111. The topologist’s sine curve

{hx, sin 1/xi : 0 < x 6 1} ∪ ({0} × [−1, 1]) ⊂ R2
is connected but not path connected.

Note that the topologist’s sine curve is also a closed and bounded
subset of the plane, hence compact. A compact, connected metric space
is often called a continuum. Continuum theory is an important subfield
of general topology.

2.3.2 Components & Path Components

Let X be a space, and x ∈ X. The component C(x) of x is the union
of all connected subsets of X which contain x. Similarly, the path com-
ponent P (x) of x is the union of all path connected subsets of X which
contain x.

Theorem 112.
(a) The component (resp., path component) of a point x is the largest
connected (resp., path connected) subset of X which contains x.

29
2 Topological Properties

(b) The collection of components (resp., path components) of a space

X forms a partition of X (that is, they are pairwise disjoint and
their union is X) into connected (resp., path connected) subsets
of X.

(c) Each component of X is closed in X.

Unlike components, path components need not be closed: it is easy

to see that the topologist’s sine curve has a path component which is
not closed.

2.3.3 Locally Connected & Locally Path Connected

A space X is locally connected (resp., locally path connected ) if whenever
x ∈ U where U is open in X, there is a connected (resp., path connected)
subset N of U such that x ∈ N ◦ .
The topologist’s sine curve is an example of a connected space which
is not locally connected.

Lemma 113. The components (resp., path components) of a locally

connected (resp., locally path connected) space are open.

Theorem 114. A space X is locally connected (resp., locally path con-

nected) if and only if X has a base consisting of connected (resp., path
connected) subsets.

Theorem 115. If X is locally path connected, then the components and

path components of X are the same. In particular, a connected space
which is locally path connected is path connected.

2.3.4 Quasicomponents
Let X be a space and x ∈ X. The quasicomponent Q(x) of x is the
intersection of all closed-and-open, or clopen, sets containing x.

Theorem 116. For any space X and x ∈ X, the component C(x) of x

is contained in the quasicomponent Q(x) of x.

30
 2.3 Connectedness

Lemma 117. Let X be a compact

T space, let H be a collection of closed
subsets of X, and let H = H. If H ⊂ U , whereTU is open, then there
is a finite subcollection H0 of H such that H ⊂ H0 ⊂ U .

Theorem 118. If X is a compact Hausdorff space, then the components

of X are the same as the quasicomponents of X.

Let X be the following subspace of the plane:

[
{h0, 0i, h0, 1i} ∪ ({1/n} × [0, 1]).
n∈N

Then the component in X of the point h0, 0i is itself, but the quasicom-
ponent of h0, 0i is {h0, 0i, h0, 1i}.

2.3.5 Some Theorems

Theorem 119. If hKn in∈N is a decreasing sequence
T of compact, con-
nected subsets of a Hausdorff space X, then K = n∈N Kn is compact
and connected.

Theorem 120, “To the boundary” Theorem. Let H be a proper closed

subspace of a compact connected Hausdorff space X. Then any compo-
nent of the subspace H meets the boundary of H.

Lemma 121. Let X be a compact connected Hausdorff space. Suppose

K = {Kn : n ∈ N} is a collection of disjoint closed sets covering X,
at least two of which are nonempty. Then for each k ∈ N, there is a
compact connected subspace Ck of X which meets at least two members
of K.

Theorem 122. Let X be a compact connected Hausdorff space. Then

every partition K of X into disjoint closed sets is either uncountable or
contains only one element.

31
2 Topological Properties

2.4 Paracompactness
2.4.1 Point-Finite & Locally Finite Collections
A collection U of subsets of a space X is said to be point-finite (resp.,
locally finite) if, for each x ∈ X, the set {U ∈ U : x ∈ U } is finite (resp.,
there is a neighborhood Ox of x such that the set {U ∈ U : Ox ∩ U 6= ∅}
is finite).
A collection U S of subsets of X is S
called closure-preserving if, for any
U 0 ⊂ U, we have {U : U ∈ U 0 } = {U : U ∈ U 0 }.

Lemma 123. Suppose U is a locally finite collection of subsets of X.

Then:
(a) U is closure-preserving;

(b) If X is compact, then U is finite.

Exercise 124.
(a) Find a locally finite cover of the real line R by open intervals.

(b) Find a point-finite cover of R which is not locally finite.

(c) Find a point-finite open cover of [0, 1] which is not closure-

preserving.

(d) Find an infinite collection of subsets of [0, 1] which is closure-

preserving.

2.4.2 Point-Finite & Locally Finite Refinements

Let U be a cover of a space X. A cover V of X is called a refinement
of U if for every V ∈ V, there is some U ∈ U with V ⊂ U . A space X
is said to be paracompact (resp., metacompact) if every open cover U
of X has a locally finite (resp., point-finite) open refinement V.

Theorem 125.
(a) Every paracompact T2 -space is regular.

32
 2.5 Modifications of Some Properties

(b) Every paracompact T2 -space is normal.

Theorem 126. Every regular Lindelöf space is paracompact.

Theorem 127. Let X be a T1 -space. Then X is compact if and only if

X is paracompact and countably compact.

2.4.3 Partitions of Unity

If φ is a function from X into [0, 1], then the support of φ, denoted
supp φ, is defined to be the closure of the set {x ∈ X : φ(x) 6= 0}. Now
let {Uα : α ∈ J} be an indexed open cover of X. We say that an indexed
family {φα : α ∈ J} of functions from X into [0, 1] is a partition of unity
on X dominated by, or subordinated to, {Uα : α ∈ J} if:
(1) supp φα ⊂ Uα for all α ∈ J;

(2) the collection {supp φα : α ∈ J} is locally finite; and

P
(3) α∈J φα (x) = 1 for all x ∈ X.

Theorem 128. Let X be a paracompact Hausdorff space. Then every

indexed open cover {Uα : α ∈ J} has a partition of unity subordinated
to it.

A partition of unity can sometimes be used to patch together func-

tions defined locally on a space to get a function defined on the whole
space.

2.5 Modifications of Some Properties

2.5.1 The Countable Chain Condition
A space X is said to have the countable chain condition (ccc) if every
pairwise-disjoint collection of open subsets of X is countable.

Theorem 129. Every separable space has the ccc.

33
 2 Topological Properties

[0,1]
Theorem 130. [0, 1] is separable.

Let c = |[0, 1]|. By a similar argument, any product of c (or fewer)

separable spaces is separable. Let c+ denote the least cardinal greater
than c.
c+
Theorem 131. [0, 1] is not separable.

F Theorem 132, ∆-system Lemma. Let F be an uncountable collection

of finite sets. Then there is an uncountable subcollection G of F and a
set R such that G1 ∩ G2 = R for any distinct G1 , G2 ∈ G.

In Theorem 132, the collection G is called a ∆-system and the set R

is called the root of the ∆-system.

Theorem
Q 133. Let {Xα : α < κ} be a collection of ccc spaces. Then
α<κ Xα has the ccc if and only if every finite subproduct
Q has the ccc,
where a finite subproduct is a product of the form α∈F Xα for some
finite subset F of κ.

c+
Corollary 134. [0, 1] has the ccc but is not separable.

2.5.2 Collectionwise Normal

A collection H of subsets of a space X is said to be discrete in X if
every point of X has a neighborhood meeting at most one member
of H. A T1 -space X is said to be collectionwise normal if, given any
discrete collection H of closed sets, there is a pairwise-disjoint collection
{UH : H ∈ H} of open sets with H ⊂ UH for every H ∈ H.
For example, let S be the Sorgenfrey line and let H = {hx, −xi : x ∈ S}.
Then H is a discrete collection of closed subsets of S2 . Note that there
is no pairwise-disjoint collection of open sets separating the members
of H. So S2 is not collectionwise normal. (We already know it’s not
even normal, but not collectionwise normal is easier to see.)

Lemma 135. If H is S a discrete collection of closed sets, then H is

pairwise-disjoint, and H0 is closed for every subcollection H0 of H.

34
 2.5 Modifications of Some Properties

Theorem 136. Every paracompact T2 -space is collectionwise normal.

Corollary 137. Every metrizable space is collectionwise normal.

F Example 138, Bing’s G. Let A be an uncountable set, and let P(A) be

the set of all subsets of A. For each α ∈ A, define eα : P(A) → {0, 1}
by eα (B) = 1 if α ∈ B and eα (B) = 0 if α ∈ / B. Let E = {eα : α ∈ A}.
Note that E can be considered to be a subset of 2P(A) .
Let X be the set 2P(A) with the topology defined by declaring every
point of X \ E to be isolated, while each eα has its usual product neigh-
borhoods in 2P(A) . Then X is normal but not collectionwise normal.

2.5.3 Monotonically Normal

A space X is said to be monotonically normal if to each pair hH, Ki
of disjoint closed sets, one can assign an open set U (H, K) such that
(a) H ⊂ U (H, K) ⊂ U (H, K) ⊂ X \ K;

(b) If H ⊂ H 0 and K ⊃ K 0 , then U (H, K) ⊂ U (H 0 , K 0 ).

An operator U (H, K) satisfying the above conditions is called a mono-
tone normality operator for X.

Theorem 139. Metrizable spaces are monotonically normal.

Lemma 140. If X is monotonically normal, then there is a monotone

normality operator U (H, K) for X satisfying U (H, K) ∩ U (K, H) = ∅
for any pair hH, Ki of disjoint closed sets.

Theorem 141. Monotonically normal spaces are collectionwise normal.

Theorem 142. The following are equivalent for a T1 -space X:

(a) X is monotonically normal;

(b) To each x ∈ X and open neighborhood U of x, one can assign

an open neighborhood Ux of x satisfying:

Ux ∩ Vy 6= ∅ → x ∈ V or y ∈ U ;

35
2 Topological Properties

(c) Same as 142(b), but with the neighborhoods U restricted to mem-

bers of a given base B.

Theorem 143. Every subspace of a monotonically normal T1 -space is

monotonically normal.

Theorem 144. Every linearly ordered space is monotonically normal.

M.E. Rudin proved in 2001 that the class of compact Hausdorff

monotonically normal spaces is exactly the class of continuous images
of compact linearly ordered spaces. The reverse direction is relatively
easy: it’s not hard to show that the closed image of a monotonically
normal space is monotonically normal. But the forward direction is a
deep and very difficult result.

36
3 More About Metrizable
Spaces

3.1 Metrization Theorems

3.1.1 Urysohn’s Metrization Theorem
Let F be a family of continuous functions from a space X to the unit
interval [0, 1]. We say F separates points if, given x1 6= x2 ∈ X, there
exists f ∈ F such that f (x1 ) 6= f (x2 ). We say F separates points from
closed sets if, given x ∈ X and any closed set H with x ∈ / H, there
exists f ∈ F with f (x) ∈/ f [H]. The evaluation function determined by
F
F is the function eF : X → [0, 1] defined by x 7→ hf (x)if ∈F .

Theorem 145. Let F be a collection of continuous functions from X

into [0, 1]. Then:
(a) eF is continuous;

(b) If F separates points, then eF is one-to-one;

(c) If X is a T1 -space and F separates points from closed sets, then

F
eF : X → eF [X] is a homeomorphic embedding of X into [0, 1] .

Theorem 146, Urysohn’s Metrization Theorem. The following are equiv-

alent for a space X:
(a) X is a separable metrizable space;

(b) X is a regular second-countable space;

N
(c) X is homeomorphic to a subspace of the Hilbert cube [0, 1] .

37
 3 More About Metrizable Spaces

Theorem 147. A space X is completely regular if and only if X is

κ
homeomorphic to a subspace of [0, 1] for some cardinal κ.

3.1.2 Metrizable Spaces Are Paracompact

F Theorem 148. The following are equivalent for a regular space X:
(a) X is paracompact;

(b) Every open cover of X has a locally finite refinement;

(c) Every open cover of X has a locally finite closed refinement.

F Theorem 149. Let X beSa regular space. If every open cover of X has
an open refinement V = n∈N Vn , where each Vn is locally finite, then
X is paracompact.

A collection that is the union of countably many collections with

property P may be called a σ-P collection. So, V in Theorem 149 is
σ-locally finite.

Lemma 150. Let X be metrizable with metric d. Let U = {Uα : α < κ}

be an open cover of X indexed by the ordinal κ. For each α < κ and
n ∈ N, let
[
Hα,n = {x ∈ Uα \ Uβ : d(x, X \ Uα ) > 1/n},
β<α

where d(x, Y ) = sup {d(x, y) : y ∈ Y }, and let

[
Vα,n = Bd (x, 1/4n).
x∈Hα,n

S
Then for each n, Vn = {Vα,n : α < κ} is locally finite, and V = n∈N Vn
is a refinement of U.

In fact, in Lemma 150 each point of X has a neighborhood meeting

at most one member of Vn . Note the use of AC in the form of the
well-ordering principle in Corollary 151.

38
 3.1 Metrization Theorems

Corollary 151. Every metrizable space is paracompact.

It is immediate from Corollary 151 and Theorem 127 that countably

compact metrizable spaces are compact. Thus Theorem 92—that com-
pact, countably compact, and sequentially compact are equivalent in
metric spaces—becomes easy.

3.1.3 Nagata-Smirnov Metrization Theorem

Lemma 152. If X is regular and has a σ-locally finite base, then X is
perfectly normal.
S
Lemma 153. Suppose X is regular and has a base B = {Bn : n ∈ N},
where each Bn is locally finite. For each n ∈ N and each B ∈ Bn , let
fn,B : X → [0, 1/n] be continuous such that fn,B (x) = 0 if and only if
x∈/ B, and let
F = {fn,B : n ∈ N ∧ B ∈ Bn }.
Now define d : X × X → R by d(x, y) = sup {|f (x) − f (y)| : f ∈ F}.
Then d is a metric on X which generates the topology of X.

Theorem 154, The Nagata-Smirnov Metrization Theorem. A space

X is metrizable if and only if X is regular and has a σ-locally finite
base.

Theorem 155. If a paracompact T2 -space is locally metrizable—that is,

if every point has a metrizable neighborhood—then it is metrizable.

3.1.4 Gδ -Diagonals
F Lemma 156. Let X be a compact Hausdorff space and suppose there is
a countable collection U of open sets of X such that, whenever x 6= y ∈
X, there is some U ∈ U with x ∈ U and y ∈ / U . Then X is metrizable.

A space X is said to have a Gδ -diagonal if the diagonal, ∆ =

{hx, xi ∈ X 2 : x ∈ X}, is a Gδ set T
in X 2 (that is, there are open subsets
2
Un , n < ω, of X such that ∆ = n<ω Un ).

39
 3 More About Metrizable Spaces

Theorem 157. A compact Hausdorff space X is metrizable if and only

if X has a Gδ diagonal.

3.1.5 Networks
A collection N of subsets of a space X is a network for X if whenever
x ∈ U , where U is open in X, there is some N ∈ N with x ∈ N ⊂ U .

Theorem 158. If X has a countable network, then so does every con-

tinuous image of X.

F Theorem 159. The following are equivalent for a T1 -space X:

(a) X has a countable network;

(b) X is the continuous image of a separable metric space.

Because of Theorem 159, spaces having a countable network are

sometimes called cosmic spaces.

Theorem 160. If X is regular and has a countable network, then X

has a Gδ -diagonal.

Theorem 161. If X is a compact metrizable space, then so is every

Hausdorff continuous image of X.

3.1.6 Perfect Images

Theorem 162. Let f : X → Y be a continuous surjection. Then the
following are equivalent:
(a) f is closed;

(b) Whenever y ∈ Y and U is an open set in X containing f −1 (y),

there is an open set V in Y containing y with f −1 [V ] ⊂ U ;

(c) For each open set U in X, the set f ∗ [U ] = {y ∈ Y : f −1 (y) ⊂ U }

is open in Y .

40
 3.2 Completeness

A continuous surjection f : X → Y is said to be perfect if f is closed

and f −1 (y) is compact for each y ∈ Y .

Theorem 163. The perfect image of a separable metrizable space is

separable metrizable.

Indeed the perfect image of any metrizable space—separable or not—

is metrizable, but the argument is more complicated.

3.2 Completeness
3.2.1 Complete Metrics
Let X be a metric space with metric d. A sequence hxn iN is d-Cauchy
if for each ε > 0 there is k ∈ N such that d(xi , xj ) 6 ε whenever
i, j > k. The metric d is complete if every d-Cauchy sequence converges.
A metrizable space X is completely metrizable if there is a complete
metric d for X which generates the topology on X.
The usual metric on Rn is complete. The usual metric on the space of
irrationals is not complete, but the irrationals are completely metrizable
(this fact will follow from subsequent results).

Theorem 164. The following are equivalent for a metric d:

(a) d is complete;

(b) Whenever A1 , A2 , . . . is a decreasing

T sequence of nonempty closed
sets with diam hAn i → 0, then n∈N An 6= ∅, where diam A =
sup {d(x, y) : x, y ∈ A}.

Corollary 165. If X is compact and metrizable, then every metric which

generates the topology is complete.

Theorem 166. A completely metrizable space is a Baire space; that is,

the intersection of countably many dense open sets is always dense.

41
3 More About Metrizable Spaces

Theorem 167. If d is a complete metric for a space X and A ⊂ X,

then d  A × A is a complete metric for the subspace A if and only if
A is closed in X.

Theorem 168. A countable product of completely metrizable spaces is

completely metrizable.

Lemma 169.
(a) Let U be an open subset of a metrizable space X with metric d.
Then the map defined by x 7→ hx, 1/d(x, X \ U )i is a homeomor-
phism of U onto a closed subset of X × R.

(b) Let G be a Gδ -subset of a metrizable space X. Then G is home-

omorphic to a closed subset of X × RN .

Corollary 170. A subspace Y of a completely metrizable space X is

completely metrizable if and only if Y is Gδ in X.

Lemma 171. If a locally compact space Y is a dense subspace of a

Hausdorff space X, then Y is open in X.

Corollary 172. Every locally compact metrizable space is completely

metrizable.

3.2.2 C ∗ (X)
Given a space X, define C ∗ (X) to be the set of all bounded continuous
real-valued functions on X. For f, g ∈ C ∗ (X), define

d(f, g) = sup |f (x) − g(x)|.

x∈X

Theorem 173. The function d : C ∗ (X) × C ∗ (X) → R defined above is

a complete metric on C ∗ (X).

This doesn’t work for C(X), the set of all (not necessarily bounded)
continuous real-valued functions on X, because in this case the “metric”
d as defined for C ∗ (X) wouldn’t necessarily be defined for all f, g ∈

42
 3.2 Completeness

C(X)—the sup could be infinite. However, note that it does work if X

is compact, for then C(X) and C ∗ (X) are the same.
C ∗ (X) is a vector space over R under the usual operations of addition
of functions and scalar multiplication. Also C ∗ (X) has a “norm” k·k
defined by
kf k = sup |f (x)|.
x∈X

That is, k·k satisfies:

(1) kf k > 0 and kf k = 0 if and only if f = 0;

(2) kf + gk 6 kf k + kgk;

(3) kcf k = |c| · kf k for all c ∈ R.

So, C ∗ (X) is what is called a normed linear space.
Given a norm, one can define a metric d by declaring d(f, g) =
kf − gk. For the norm kf k = supx∈X |f (x)| on C ∗ (X), the associated
metric is precisely the metric on C ∗ (X) defined earlier.
A Banach space is a complete normed linear space; that is, complete
in the metric defined by the norm. So by Theorem 173, C ∗ (X) is a
Banach space.

3.2.3 Isometries
Let hX, di and hX 0 , d0 i be metric spaces. A mapping j : X → X 0 is
called an isometry if for each x, y ∈ X, d0 (j(x), j(y)) = d(x, y); that is,
j preserves distances.

Theorem 174. Let hX, di be a metric space and let a ∈ X; for each
x ∈ X, let fx ∈ C ∗ (X) be defined by fx (z) = d(z, x) − d(z, a). Then
the map j : X → C ∗ (X) defined by j(x) = fx is an isometry under the
usual distance on C ∗ (X).

If the metric d in the above theorem were bounded, defining fx (z) =

d(z, x) would suffice.

Corollary 175. Every metric space hX, di is isometrically embeddable

in a complete metric space.

43
 3 More About Metrizable Spaces

F Theorem 176. For every metric space hX, di there is a unique (up to
˜ with X a dense subspace of X̃
isometry) complete metric space hX̃, di
and d˜  X × X = d.

The metric space hX̃, di ˜ given by Theorem 176 is called the com-
pletion of hX, di. It can also be defined (by a method analogous to a
classical way of defining the reals from the rationals) by “adding a point”
to X for each Cauchy sequence in X which does not converge in X,
calling two Cauchy sequences hxn iN , hyn iN equivalent if d(xn , yn ) → 0.
Each “point” of X̃ \X, then, is an equivalence class of Cauchy sequences
in X. The distance d˜ between two equivalence class with representatives
˜ n i, hyn i) = limn→∞ d(xn , yn ).
hxn i and hyn i is defined by d(hx

44
4 More About Linearly Ordered
Spaces
4.1 The Cantor Set
Recall that we may represent a finite sequence of length n < ω in
a set X as a function σ : n → X, so that σ = hσ(0), . . . , σ(n − 1)i =
hσ0 , . . . , σn−1 i. If x ∈ X, then τ = σax is a new sequence τ : n+1 → X
such that τ  n = σ and τ (n) = x; that is, we have appended x to the
end of our sequence σ.
Also recall that a function is a set of ordered pairs, so |σ| gives the
length of σ.

4.1.1 The Construction

For each finite sequence σ of 0’s and 1’s (including the empty sequence
∅), we define a closed subinterval Iσ of [0, 1] as follows. Start by setting
I∅ = [0, 1]. Then if Iσ has been defined, let Iσa0 and Iσa1 be the left and
right thirds, respectively, of Iσ . Thus I0 = [0, 1/3], I1 =
S [2/3, 1], I00 =
[0, 1/9], I01 = [2/9, 1/3], etc. For each n < ω, let Cn = T{Iσ : |σ| = n}.
So C0 = [0, 1], C1 = [0, 1/3] ∪ [2/3, 1], etc. Finally, C = n<ω Cn is the
Cantor set.

Theorem 177. The Cantor set C as defined above is (as a subspace of

the real line R) compact, metrizable, has no isolated points, and has
a countable base of open and closed sets. It is an uncountable closed
subset of R with empty interior in R.

A space which has a base of open and closed sets is sometimes called
zero-dimensional , or more precisely, is said to have small inductive

45
 4 More About Linearly Ordered Spaces

dimension zero. There are several concepts of dimension.

F Theorem 178. Suppose X is a compact Hausdorff metrizable space

with no isolated points, and has a countable base of clopen sets. Then
X is homeomorphic to the Cantor set.

Corollary 179. The following spaces are homeomorphic to the Cantor

set C:
N
(a) {0, 1} ;
Q
(b) Any product space n∈N Xn , where each Xn is a finite discrete
space with at least two points;

(c) C×F for any nonempty finite discrete space F , so the topological
sum of a finite number of copies of C is homeomorphic to C;

(d) C × C;

(e) CN .

4.1.2 A Space-Filling Curve

P∞
Theorem 180. The function f : 2N → [0, 1] defined by ~x 7→ n=1 xn /2n
is continuous and onto.
2
F Theorem 181. There is a continuous function from [0, 1] onto [0, 1] —
N
or even onto the Hilbert cube [0, 1] .

4.2 Characterizing Some Ordered Spaces

4.2.1 R & [0, 1]
Theorem 182. Suppose hX, <i and hY, ≺i are countable linearly or-
dered sets with no first or last point, and both are densely ordered (that
is, between any two points there is another point). Then there is an
order-preserving bijection f : X → Y .

Theorem 183. A linearly ordered space X is connected if and only if

46
 4.2 Characterizing Some Ordered Spaces

(a) X is densely ordered, and

(b) every bounded above subset of X has a least upper bound.

Theorem 184. Suppose X is a connected linearly ordered space with

no first or last point, and is separable. Then X is order-isomorphic to
the real line R with the usual order.

Corollary 185. Suppose X is a separable compact connected linearly

ordered space. Then X is homeomorphic to the unit interval [0, 1].

4.2.2 Q
Lemma 186. If X is completely regular and |X| < |R|, then X has a
base of clopen sets.

F Lemma 187. Suppose κ is an infinite cardinal. If κ is the least cardinal

of a base for a space X, then for every base B, there is a base C ⊂ B
such that |C| = κ.

The least cardinal of a base for a space X is called the weight of X

and is denoted by w(X).

Theorem 188. Suppose X is countable, regular, first-countable, and

has no isolated points. Then X is homeomorphic to the rationals Q.

Corollary 189. The following are homeomorphic to the space Q of ra-

tionals.
(a) Qn for every positive integer n;
(b) Any countable dense subset of Rn for any n ∈ N, or of the
Cantor set C;
(c) Q with the right half-open interval topology.

4.2.3 Compact Spaces

Theorem 190. A linearly ordered space X is compact if and only if
every subset of X has a least upper bound and a greatest lower bound.

47
 4 More About Linearly Ordered Spaces

F Theorem 191. Every linearly ordered space X is a dense subset of a

compact linearly ordered space X̂.

4.3 The Ordinal ω1

We now discuss paracompactness of ordered spaces, with the eventual
goal of the characterization theorem of the next section. The following
result is a corollary of a couple of results from previous chapters.

Theorem 192. The space ω1 of countable ordinals is not paracompact.

4.3.1 Closed & Unbounded Subsets

Theorem 193.
(a) Let C and D be closed (in the order topology) and unbounded
subsets of ω1 . Then C ∩ D is unbounded.

(b) Let
T Cn , n < ω, be club—closed and unbounded—in ω1 . Then
n<ω Cn is club.

F Theorem 194. ω1 is not perfectly normal.

Theorem 195. Let C be club in ω1 . Then C is homeomorphic to ω1 .

Theorem 196. If f : ω1 → ω1 , then C = {α : ∀β < α(f (β) < α)} is

club.

4.3.2 Stationary Subsets

A subset S of ω1 is said to be stationary in ω1 if S ∩ C 6= ∅ for every
closed unbounded set C.

Theorem 197, Pressing Down Lemma. Let S be a stationary set in

ω1 . If for each ordinal α in S, α > 0, we choose an ordinal βα < α,
then there is some β < ω1 such that β = βα for uncountably many
α ∈ S. (Such a map α 7→ βα is said to be “pressing down.”)

48
 4.4 Characterizing Paracompact Ordered Spaces

In fact, there is a β < ω1 such that f −1 (β) is not just uncountable

but stationary. It is possible to use this stronger version of the pressing
down lemma to show that there is a collection of ω1 -many disjoint
stationary sets. We can, however, show the following.

Proposition A. There are disjoint stationary sets.

Theorem 198. Let S be a stationary subset of ω1 . Then S is not para-

compact.

Theorem 199. Let C be a closed subset of a linearly ordered space X.

Then X \ C is the union of a disjoint collection of convex open sets.

Theorem 200. The following are equivalent for a subspace S of ω1 :

(a) S is metrizable,

(b) S is paracompact, and

(c) S is non-stationary.

F Example 201. The space ω1 ×(ω1 +1) is not normal. Hence, the product
of a normal space and a compact Hausdorff space need not be normal.

4.4 Characterizing Paracompact Ordered

Spaces
It will be convenient to define an ordinal κ to be a cardinal if there
is no function from an ordinal α < κ onto κ. With this notation,
ω and ω1 denote the least infinite cardinal and the least uncountably
infinite cardinal, respectively (as well as the least ordinals with infinitely
many and uncountably many predecessors, respectively). A cardinal
κ is called a successor cardinal if there is a cardinal λ < κ such that
κ is the least cardinal greater than λ. In this case, κ is often denoted
by λ+ . A cardinal κ which is not a successor cardinal is called a limit
cardinal. For example ω1 , ω2 , . . . are successor cardinals while ω and
ωω = sup {ωn : n < ω} are limit cardinals.

49
 4 More About Linearly Ordered Spaces

4.4.1 Cofinality of Limit Ordinals

Let λ be a limit ordinal. We define the cofinality of λ, denoted cf λ, to
be the least cardinal κ such that there is a function f : κ → λ with
{f (α) : α < κ} unbounded in λ. For example, if λ is any countable limit
ordinal, then cf λ = ω. Similarly, cf ωω = ω.
An infinite cardinal κ is said to be regular if cf κ = κ. Otherwise, κ
is called singular. Clearly, ωω is singular.

Proposition B. ω1 is a regular cardinal.

F Lemma 202.
(a) If κ is any infinite cardinal, then the union of 6 κ-many sets,
each of cardinality 6 κ, has cardinality 6 κ;

(b) if κ is a successor cardinal, then κ is regular;

(c) For any limit ordinal λ, cf λ is a regular cardinal.

Theorem 203. The following are equivalent for an infinite cardinal κ:

(a) κ is regular;

(b) For any A ⊂ κ, if |A| < κ, then sup A < κ;

(c) The union of < κ-many sets, each with cardinality < κ, has
cardinality < κ.

4.4.2 Club & Stationary Subsets of Cardinals

Closed unbounded and stationary sets are defined for any uncountable
regular cardinal κ in the same way they were defined for ω1 , and the
analogues of Theorem 193, Theorem 195, and the Pressing Down Lemma
hold by similar arguments.

Theorem 204. Let κ be an uncountable regular cardinal.

T
(a) If C is a collection of < κ-many club subsets of κ, then C is
club;

50
 4.5 Hereditarily Lindelöf but not Separable

(b) If f : κ → κ, then C = {α < κ : ∀β < α(f (β) < α)} is club;

(c) If S ⊂ κ is stationary, and to each α ∈ S with α > 0 we choose

βα < α, then there is some β < κ such that βα = β for κ-many
α ∈ S.

Theorem 205. Let S be a stationary subset of a regular cardinal κ.

Then S is not paracompact.

4.4.3 The Characterization

Let U be a collection of sets and let U, V ∈ U. A finite linked chain in
U from U to V is a sequence U1 , U2 , . . . , Un of members of U such that
U1 = U , Un = V , and Ui ∩ Ui+1 6= ∅ for any i = 1, 2, . . . , n − 1. U is
said to be connected if there is a finite linked chain in U between any
two members of U.
The next result has nothing to do with ordered spaces, but is good
to know.

Theorem 206. A space X is connected if and only if every cover of X

by nonempty sets is connected.

F Theorem 207. A linearly ordered space X is paracompact if and only

if X does not contain a closed subspace homeomorphic to a stationary
subset of a regular uncountable cardinal.

4.5 Hereditarily Lindelöf but not Separable

4.5.1 ccc Linearly Ordered Spaces
Lemma 208. The following are equivalent for a space X:
(a) X is hereditarily Lindelöf;

(b) Every open subspace of X is Lindelöf;

(c) For any collectionSU of S

open subsets of X, there is a countable
V ⊂ U such that V = U;

51
 4 More About Linearly Ordered Spaces

(d) There is no subset {xα : α < ω1 } of X with the property that,

for each α < ω1 , xα ∈
/ {xβ : β > α}.

A set {xα : α < ω1 } satisfying the condition in Lemma 208(d) is said

to be right-separated in type ω1 . Left-separated in type ω1 is defined
analogously, and the analogous result is that a space is hereditarily
separable if and only if it does not contain a subspace which is left-
separated in type ω1 .

F Theorem 209. Supose X is a ccc linearly ordered space. Then X is

hereditarily Lindelöf.

4.5.2 Suslin Lines

A linearly ordered space X is a Suslin line if X is ccc, connected, has
no first or last point, and is not homeomorphic to the real line. By
Theorem 184, a Suslin line cannot be separable.

F Theorem 210. If there is a ccc nonseparable linearly ordered space,

then:
(a) there is one which is densely ordered and such that no nonempty
open interval is separable;

(b) there is a Suslin line.

Sometimes a Suslin line is defined to be a nonseparable ccc linearly

ordered space. By Theorem 210, this is equivalent to our definition (in
the sense that one exists if and only if the other exists).

52
5 Some Set-Theoretic Topology
5.1 Trees
A partially ordered set hT, 6i is a tree if the set Pt = {s ∈ T : s < t} is
well-ordered by 6 for all t ∈ T . The elements of T are called the nodes
of T . A maximal linearly ordered subset of T is called a branch of T .
Recall that any well-ordered set is order-isomorphic to (the prede-
cessors of) an ordinal. If hT, 6i is a tree, then for any ordinal α, the
set
Lα = {t ∈ T : Pt is isomorphic to α}
is called the αth level of T . The least α such that Lα is empty is called
the height of T .

5.1.1 The Cantor Tree

Let T be the set of all finite sequences of 0’s and 1’s (including the empty
sequence). We can equivalently describe T as the set of all functions σ
from some natural number n into 2, where here we think of n as the
set {0, 1, . . . , n − 1} and 2 as the set {0, 1}. (Or, σ is a sequence in 2 of
length n.) If σ, τ ∈ T , define σ < τ if and only if σ is an initial segment
of τ ; for example, 110 < 1100, 01 < 0111, etc. Then hT, 6i is called
the Cantor tree. The set 2<ω , ordered by extension, is another way to
describe the Cantor tree.
Note that the branches of the Cantor tree can be identified with 2ω ,
ω
the set of all functions f : ω → 2. S Given f ∈ 2 , {f  n : n < ω} is
a branch,
S and given a branch b, b is a function from ω to 2. Note
b = { b  n : n < ω}.
The Cantor tree hT, 6i is completely described (up to isomorphism)
as follows:

53
 5 Some Set-Theoretic Topology

(1) The least level L0 of T has exactly one node;

(2) Each node σ has exactly two (immediate) successors; and

(3) T has height ω.

Theorem 211. Recall our construction of the Cantor set C in Sec-

tion 4.1. If B = {Iσ ∩ C : σ ∈ 2<ω }, then B is a countable base of
clopen sets in C such that hB, ⊃i is isomorphic to the Cantor tree.

Theorem 212. A space X is homeomorphic to the Cantor set C if and

only if X has a base B consisting of clopen sets such that hB, ⊃i is
isomorphic to the Cantor tree and the intersection of each branch of
this tree is a single point.

F Theorem 213. Let hT, 6i be the Cantor tree. Let X S be the set of all
branches of T . Note that if b is a branch of T , then b ∈ 2ω . Define a
S ≺ on X S
linear order as follows: if b, b0 ∈ X and n is the least
S integer
such Sthat b(n) 6= b0 (n), then b ≺ b0 if and only if b(n) = 0
and b0 (n) = 1. Then X with the topology induced by this order is
homeomorphic to C.

F Theorem 214. The following are continuous images of C:

(a) the unit interval [0, 1];
ω
(b) the Hilbert cube [0, 1] ;

(c) any closed subset of C; and

(d) any compact metric space.

5.1.2 The Irrationals

n
For each n < ω, S let ω n denote the <ωset of all functions from n into ω,
<ω
and let ω = n<ω ω . If σ, τ ∈ ω , let σ < τ if and only if σ is an
initial segment of τ . Then hω <ω , <i is a tree of height ω with one node
at the least level (the empty sequence) and such that every node has a
countable infinite number of immediate successors.

54
 5.1 Trees

Exercise 215. Let P denote the irrationals (as a subspace of R). Show
that P has a base B of clopen sets such that hB, ⊃i is a tree isomorphic
to hω <ω , 6i, and such that the intersection of each branch is a singleton.

A space X is said to be nowhere-locally-compact if no point of X has

a compact neighborhood.

Theorem 216. The following are equivalent for a space X:

(a) X is homeomorphic to the space of irrationals (as a subspace of
the real line with the usual Euclidean topology);

(b) X has a base B consisting of clopen sets such that hB, ⊃i is

isomorphic to the tree hω <ω , 6i, and the intersection of each
branch of this tree is a single point; and

(c) X is a nowhere-locally-compact complete separable metric space

and has a countable base of clopen sets.

Corollary 217. Let P denote the space of irrationals. Then P is home-

omorphic to P2 , Pω , and ω ω (where ω is given the discrete topology).

A subset A of the real line R is said to be analytic if there is a

continuous surjection f : P → A. It is known that every Borel set is
analytic, every analytic set is Lebesgue measurable, and that there
are analytic sets that are not Borel (a highly nontrivial result!) and
measurable sets that are not analytic. Analytic sets play a major role
in the field of “descriptive set theory.”

5.1.3 Aronszajn and Suslin Trees

Theorem 218. If T is a tree of height ω, and every level of T is finite,
then T has an infinite branch.

An Aronszajn tree is a tree of height ω1 such that every branch and

every level is countable. A Suslin tree is a tree of height ω1 such that
every branch and every antichain is countable. Note that every Suslin
tree is Aronszajn.

55
 5 Some Set-Theoretic Topology

Theorem 219. There is an Aronszajn tree.

Theorem 219 shows that the natural analogue of Theorem 218 for
trees of height ω1 is false: every level can be countable yet there is no
branch going all the way to the top of the tree. If, however, you require
every level to be finite, then there is such a branch.

F Theorem 220. If there is a Suslin line, then there is a Suslin tree.

F Theorem 221. If there is a Suslin tree, then there is a Suslin line.

F Theorem 222. If S is a Suslin line, then S 2 does not have the ccc.

F Theorem 223. If there is a Suslin line, there is one such that no non-
degenerate interval is separable. Such a Suslin line is the union of
ω1 -many nowhere-dense sets.

5.2 Martin’s Axiom

Like the Continuum Hypothesis (CH), Martin’s Axiom (MA) is an
axiom of set theory that is known to be consistent with and independent
of ZFC. Martin’s Axiom has had many applications in certain parts of
general topology and real analysis. It’s most powerful when conjuncted
with the negation of the Continuum Hypothesis. Informally, assuming
MA + ¬CH one can bump up to ω1 or higher some results that are true
in ZFC for ω. For example, in ZFC you know how to prove that the real
line is not the union of countably many nowhere dense sets. Assuming
MA + ¬CH, R is not the union of ω1 -many nowhere-dense sets. We’ll
also see that MA + ¬CH kills Suslin lines.

5.2.1 The Statement

Let hP, 6i be a partially ordered set. We say D ⊂ P is dense in P if
for any p ∈ P there is q 6 p with q ∈ D. A subset G of P is called a
filter in P if
(a) For each p, q ∈ G, there is r ∈ G with r 6 p and r 6 q;

56
 5.2 Martin’s Axiom

(b) For each p ∈ G, if p 6 q then q ∈ G.

For example, let X be any set and let P(X) \ {∅} be the collection of
all nonempty subsets of X. For p, q ∈ P(X) \ {∅}, define p 6 q if and
only if p ⊂ q. Then G is a filter on P(X) \ {∅} if and only if G is a
filter of subsets of X in the usual sense.
We say that two elements p, q of P are comparable if p 6 q or q 6 p,
compatible (abbreviated p 6⊥ q) if there is r ∈ P with r 6 p and r 6 q,
and incompatible (abbreviated p ⊥ q) if they are not compatible. In the
above example of P(X) \ {∅}, p and q are comparable if and only if one
is a subset of the other, compatible if and only if they have nonempty
intersection, and thus incompatible if and only if they are disjoint.
An antichain in P is a subset A of P such that every two elements of
A are incompatible. We say that hP, 6i has the ccc if every antichain
is countable, or equivalently, every uncountable subset of P has a pair
of compatible elements.
Note that P(X) \ {∅} will not have the ccc if X is uncountable. But
now let X be a topological space, let O(X) be the collection of all
nonempty open sets, and define p 6 q if and only if p ⊂ q. Then the
partial order on O(X) has the ccc if and only if the space X has the ccc.
Note that a subset D of O(X) is dense in this partial order if and only
if D is a collection of nonempty open sets such that every nonempty
open set contains a member of D. (Such a collection D is sometimes
called a π-base for the space X.)
Finally, we now define Martin’s Axiom (MA): Let κ be a cardinal.
MA(κ) is the following statement: Whenever hP, 6i is a ccc partial
order, and D is a family of 6 κ-many dense sets, there is a filter G in
P such that G ∩ D 6= ∅ for every D ∈ D.
MA is the statement that MA(κ) holds for every κ < 2ω .

5.2.2 Working with MA(κ)

Theorem 224. MA(ω) is true, hence CH → MA.

F Theorem 225. Assume MA(κ). Let X be a compact Hausdorff T ccc

space. If U is a collection of 6 κ-many dense open sets, then U 6= ∅.

57
 5 Some Set-Theoretic Topology

Corollary 226. MA(2ω ) is false.

Corollary 227. Assume MA(ω1 ). Then there are no Suslin lines.

The topological statement in Theorem 225 is actually equivalent to

MA(κ). So Martin’s Axiom is equivalent to the statement that no com-
pact Hausdorff ccc space is the union of fewer than 2ω -many nowhere-
dense sets. In particular, MA implies that the real line is not the union
of few than 2ω -many nowhere-dense sets. One can also show that MA
implies the real line is not the union of fewer than 2ω -many Lebesgue
measure zero sets.

F Lemma 228. Assume MA(ω1 ). Suppose X is ccc and {Uα : α < ω1 } is

a collection of nonempty open subsets of X. Then there is an uncount-
able subset A of ω1 such that {Uα : α ∈ A} has the finite intersection
property.

A space X is said to have property K if every uncountable collection

U of nonempty open sets contains an uncountable subcollection V such
that V1 ∩ V2 6= ∅ for every V1 , V2 ∈ V.

Corollary 229. Assume MA(ω1 ). Then every ccc space has property K.

Theorem 230. Assume MA(ω1 ). If X and Y are ccc, so is X × Y .

Hence any product of ccc spaces is ccc.

Note that the above theorem also yields Corollary 227.

Lemma 231.
(a) A regular Lindelöf space X is hereditarily Lindelöf if and only
if X is perfectly normal.
(b) If X is compact Hausdorff, then X is first-countable if and only
if every point of X is a Gδ -set. Consequently, compact Hausdorff
perfectly normal spaces are first-countable.

F Theorem 232. Assume MA(ω1 ). If X is a compact Hausdorff heredi-

tarily Lindelöf space, then X is hereditarily separable.

58
 5.3 Small Cardinals

F Lemma 233. Assume MA(κ). Let {Uα : α < κ} be a collection of dense

open subsets
T of the real line R. Then there is a dense Gδ -set G such
that G ⊂ α<κ Uα .

5.2.3 First Category Subsets of R

A subset X of R is said to be of first category—some texts call sets
of first category meagre, and their complements comeagre—in R if X
is contained in the union of countably many nowhere-dense subsets
of R. Since the closure of a nowhere-dense sets is nowhere-dense, it is
equivalent to say that X is first category if and only if the complement
of X contains a dense Gδ -set.

Theorem 234. Assume MA. Then the union of < 2ω -many first cat-
egory subsets of R is first category. In particular, any subset of R of
cardinality < 2ω is first category.

It is also true that assuming MA, the union of < 2ω -many Lebesgue
measure zero subsets of R has measure zero, and in particular, any
subset of R of cardinality < 2ω has measure zero.

5.3 Small Cardinals

5.3.1 Unbounded and Dominating Families in ω ω
Let f, g ∈ ω ω . Define f <∗ g if and only if there is k < ω such that
f (n) < g(n) for all n > k. Call a subset F of ω ω
(1) <∗ -unbounded if there is no g ∈ ω ω such that f <∗ g for all
f ∈ F, and

(2) <∗ -dominating if for every g ∈ ω ω there is f ∈ F such that

g <∗ f .
For convenience we may say a family is simply unbounded or dominating.
Define b (resp., d) to be the least cardinal of a <∗ -unbounded (resp.,
∗
< -dominating) family in ω ω .

59
 5 Some Set-Theoretic Topology

Theorem 235. ω1 6 b 6 d 6 c, so CH → b = d = c.

Theorem 236. MA → b = d = c.

Theorem 237. There exists a collection {fα : α < b} ⊂ ω ω which is

unbounded and <∗ -increasing.

Since every cofinal subset of an unbounded <∗ -increasing family is

also unbounded, it is easy to see that b is a regular cardinal; d need
not be.

Theorem 238. If b = d, then there exists {fα : α < d} ⊂ ω ω which is

dominating and <∗ -increasing.

A dominating <∗ -increasing family in ω ω may be called a scale.

5.3.2 Subsets of the Irrationals

The cardinal d has important connections to the structure of compact
subsets of the irrationals. Recall P∼
=ω ω . Call a collection A of compact
subsets of ω dominating if every compact subset of ω ω is contained
ω

in some A ∈ A.

F Theorem 239. Let C be the collection of compact subsets of ω ω , A the

subcollections of C which cover ω ω , and B the subcollections of C which
are dominating. Then d = min {|A| : A ∈ A} = min {|B| : B ∈ B}.

Next we have a topological characterization involving b. Call a subset

A of the irrationals (think in the real line now) concentrated about the
rationals Q if for any open U ⊃ Q, the set A \ U is countable.

Proposition C. The set of all irrationals is not concentrated about Q.

Proposition D. Define a subset A ⊂ P to be (< κ)-concentrated about

Q if |A ∩ C| < κ for every C ⊂ P closed in the reals. Then
κ
b = min {κ > ω : ∃A ∈ [P] (A is (< κ)-concentrated about Q)}.

60
 5.4 Normal vs. Collectionwise Normal

Theorem 240. b = ω1 if and only if there is an uncountable subset A

of the irrationals concentrated about Q.

Theorem 240 takes added interest due to the following result.

Example 241. Let A be an uncountable subset of the irrationals con-

centrated about Q. Let X be the space A ∪ Q, where points of A are
isolated and points of Q have their usual Euclidean neighborhoods (in
the subspace A ∪ Q of R). Then X is a regular Lindelöf space but its
product with the irrationals is not Lindelöf.

So b = ω1 , and in particular CH, implies that there is a regular

Lindelöf space whose product with the irrationals is not Lindelöf. But
the following question of E.A. Michael has remained open for over 40
years: Is there in ZFC a regular Lindelöf space whose product with the
irrationals is not Lindelöf?
A recent result on this problem says you do not have to worry about
“regular;” any example, even not T0 , implies that there is a regular one.

5.4 Normal vs. Collectionwise Normal

5.4.1 Tangent Disc Spaces
We begin by defining a modification of the Tangent Disc Space. Let A
be a subset of the real line R. Let X(A) be the space whose set is
(A × {0}) ∪ {hq, ri : q, r ∈ Q ∧ r > 0}.
That is, X(A) consists of the points on the x-axis corresponding to
a ∈ A, together will all points in the upper half-plane with rational
coordinates.
Let the points in the upper half-plane be isolated, and for each a ∈ A
and n > 0, let
p
D(a, n) = {ha, 0i} ∪ {hq, ri ∈ Q2 : (q − a)2 + (r − 1/n)2 < 1/n}
be a basic neighborhood of ha, 0i;that is, D(a, n) consists of the point
ha, 0i together with all points with rational coordinated in the interior
of a disk of radius 1/n tangent to the x-axis at ha, 0i.

61
 5 Some Set-Theoretic Topology

Note that A × {0} is closed in X(A), and discrete as a subspace. It

follows that the collection of singletons {{ha, 0i} : a ∈ A} is a discrete
collection of closed sets, and hence the union of any subcollection, that
is, B × {0} for any B ⊂ A, is closed in X(A).

F Theorem 242. Let A ⊂ R, and let X(A) be as defined above. Then:

(a) If A = R, then X(A) is not normal;

(b) If A is uncountable, then X(A) is not collectionwise normal;

(c) If every subset of A is a Gδ -set in A, then X(A) is normal.

5.4.2 Q-Sets
An uncountable subset A of R whose every subset is Gδ in A is called
a Q-set.

F Theorem 243. If 2ω < 2ω1 , then there are no Q-sets.

F Theorem 244. Let κ be an uncountable cardinal, and assume MA(κ).

Then every subset of R of cardinality κ is a Q-set.

Corollary 245. Assume MA(ω1 ). Then there is a Q-set, and hence there
is a normal first-countable separable non-collectionwise normal space.

Corollary 246. Assume MA. Then 2κ = 2ω for every infinite cardinal

κ < 2ω .

Bob Heath showed that if there is a normal first-countable separable

non-collectionwise normal space, then there is a Q-set. So, by Theo-
rem 242(c), the existence of such a space is equivalent to the existence
of a Q-set.

Lemma 247, Jones’ Lemma. Let X be a separable space, and let D be

a closed discrete subset of X. If 2|D| > 2ω , then X is not normal.
ω
Since 22 > 2ω , Lemma 247 gives another way to show Theorem 242(a),
the square of the Sorgenfrey line is not normal, and the like.

62
 5.5 Countably Compact vs. Pseudocompact

F Corollary 248. If 2ω < 2ω1 , then every normal separable space is col-
lectionwise normal.

5.5 Countably Compact vs. Pseudocompact

5.5.1 Almost Disjoint Families
A collection A of infinite subsets of ω is said to be almost-disjoint if the
intersection of any two distinct members of A is finite. By a standard
Zorn’s Lemma argument, every almost-disjoint family A is contained
is a maximal almost-disjoint family A0 .

Theorem 249. There is an almost-disjoint family of subsets of ω of

cardinality 2ω .

Proposition E. Any finite partition of ω into infinite sets is a maximal

almost-disjoint family.

Lemma 250. No countably infinite almost-disjoint family of subsets of

ω is maximal.

Theorem 251. Assume MA. Then every infinite maximal almost-disjoint

family of subsets of ω has cardinality 2ω .

5.5.2 Mrówka Spaces

Let A be an almost-disjoint family of subsets of ω. Define a space ψ(A)
as follows. The underlying set for ψ(A) is ω ∪ {xA : A ∈ A}, where
{xA : A ∈ A} is a set of distinct points not in ω. Define the topology
by declaring the points of ω to be isolated, and the nth member of a
local base at xA to be b(xA , n) = {xA } ∪ (A \ n).

Theorem 252. ψ(A) is locally compact Hausdorff, and {xA : A ∈ A}

is a closed discrete subset of ψ(A). Thus, if A is infinite then ψ(A) is
not countably compact.

63
5 Some Set-Theoretic Topology

A space X is said to be pseudocompact if every continuous f : X → R

has a bounded range.

Theorem 253. Every countably compact space is pseudocompact.

Theorem 254. If A is an infinite maximal almost-disjoint family of

subsets of ω, then ψ(A) is pseudocompact (but not countably compact).

5.6 Hereditarily Separable vs. Hereditarily

Lindelöf
5.6.1 Some History
An S-space is a regular hereditarily separable space which is not heredi-
tarily Lindelöf; an L-space is a regular hereditarily Lindelöf space which
is not hereditarily separable. By Theorem 232, assuming MA(ω1 ) there
are no compact L-spaces. MA(ω1 ) also implies there are no compact
S-spaces. On the other hand, a Suslin line (compactified by adding a
first and last point) is a compact L-space. In this section, we show the
Continuum Hypothesis can be used to construct a compact S-space.
The results in these notes related to S- and L-spaces were proven by
the late 1970’s. In the early 1980’s, S. Todorcevic proved that under the
Proper Forcing Axiom (PFA)—an axiom stronger than MA(ω1 )—there
are no S-spaces. Whether or not there are L-spaces in ZFC remained
unsettled until 2006 when J. Moore surprised everyone by constructing
an L-space without assuming any special axioms of set theory.

5.6.2 The Kunen Line

Throughout this section, 2ω = ω1 is assumed. Let R = {xα : α < ω1 },
where Q = {xn : n < ω}.

Lemma 255. Let A = {C ⊂ R : |C| = ω, |C| > ω}. It is possible to enu-

merate A as {Aα : ω 6 α < ω1 } such that ∀α > ω(xβ ∈ Aα → β < α).

Let X = R, and for each α 6 ω1 let Xα = {xβ : β < α}. Let A be

64
 5.6 Hereditarily Separable vs. Hereditarily Lindelöf

indexed as in Lemma 255. We inductively define a topology Tα on Xα .

Let Tω be the discrete topology on Xω . Note that Aω ⊂ Xω . Choose
a sequence n0 < n1 < · · · such that hxni i → xω in the Euclidean
topology and such that xni ∈ Aω if xω ∈ Aω . Let

Bω = {{xni }i>k ∪ {xω } : k ∈ ω},

and let Tω+1 be the topology on Xω+1 generated by Tω ∪ Bω .

Note that Tω+1 is locally compact Hausdorff, has a countable base of
clopen sets, and every subset of Xω+1 which is open in the Euclidean
topology of Xω+1 is open in Tω+1 (so Tω+1 is finer than the Euclidean
topology). Also, xω ∈ Aω in Tω+1 if this is true in R.

F Lemma 256. Suppose α < ω1 and for all β < α we have defined a
topology Tβ on Xβ satisfying:
(a) hXβ , Tβ i is a locally compact Hausdorff space with a countable
base of clopen sets, and Tβ is finer than the Euclidean topology
on Xβ ;

(b) γ < β → Tγ ⊂ Tβ and Tγ = Tβ  Xγ = {U ∩ Xγ : U ∈ Tβ };

(c) If γ < δ < β and xδ is in the Euclidean closure of Aγ , then xδ

is in clTδ+1 Aγ .
Then there is a topology Tα on Xα satisfying 256(a), 256(b), and 256(c);
hence this defines Tα for all ω 6 α < ω1 .

Theorem 257. Let Xα and Tα , α < ω1 , beSas in Lemma 256. Let Tω1
be the topology on X = R generated by ω6α<ω1 Tα . Then hR, Tω1 i
is locally compact Hausdorff, hereditarily separable, and non-Lindelöf,
hence its one-point compactification is a compact S-space.

65
6 The Stone-Čech
Compactification
6.1 Hausdorff Compactifications
A compactification of a completely regular space X is a compact T2 -
space αX and a homeomorphic embedding α : X → αX such that α[X]
is dense in αX.
To explain the separation assumptions in the above definition: we will
only be concerned with compactifications which are Hausdorff. Note
that only completely regular spaces can have a T2 -compactification.
Hence in this section, unless otherwise stated, all spaces are assumed
to be completely regular.
A few of the concepts and results below are from previous chapters
but are included here for completeness.

6.1.1 One-Point Compactifications

Let X be a locally compact non-compact Hausdorff space. The one-
point compactification ωX of X is the space X ∪ {∞}, where points of
X have their usual neighborhoods, and a neighborhood of ∞ has the
form {∞} ∪ X \ K where K is a compact subset of X. In the framework
of the previous definition, the map ω : X → ωX is the identity map on
X.
Of course, X is open in its one-point compactification. We’ll see that
this is the case for any compactification of a locally compact X.

Lemma 258. If X is a dense subset of a compact T2 -space Z, and X

is locally compact, then X is open in Z.

67
 6 The Stone-Čech Compactification

Corollary 259. If αX is a compactification of a locally compact space

X, then α[X] is open in αX.

If αX and γX are compactifications of X, we say αX > γX if and

only if there exists a continuous function f : αX → γX such that
f ◦ α = γ. If such f exists which is a homeomorphism, we say that αX
and γX are equivalent compactifications, and write αX ≈ γX.

Theorem 260. If X is locally compact, ωX is the one-point compacti-

fication, and αX is any compactification, then αX > ωX.

So ωX is the “smallest” compactification of X, as it intuitively ought

to be.

6.1.2 Compactification by Adding Points

We often think of obtaining a compactification αX of X by adding
some points to X to make it compact; that is, X is a dense subset of
αX (instead of merely homeomorphic to one) and the map α : X → αX
is the identity on X. The next result says we don’t lose anything by
thinking of compactifications in this way.

F Theorem 261. If αX is any compactification of X, then there is an

equivalent compactification γX such that X ⊂ γX and the map γ : X →
γX is the identity map.

Lemma 262. Let f and g be continuous mappings from a space X into

a Hausdorff space Y . If f and g agree on a dense subset of X, then
f = g.

Theorem 263. αX ≈ γX if and only if αX > γX and γX > αX.

6.1.3 Compactification by Embedding in [0, 1]κ

Let F be a family of continuous functions from a space X to the unit
interval [0, 1]. We say F separates points if, given x1 =6 x2 ∈ X, there
exists f ∈ F such that f (x1 ) 6= f (x2 ). We say F separates points from

68
 6.1 Hausdorff Compactifications

closed sets if, given x ∈ X and any closed set H with x ∈ / H, there
exists f ∈ F with f (x) ∈/ f [H]. The evaluation function determined by
F
F is the function eF : X → [0, 1] defined by x 7→ hf (x)if ∈F .

Theorem 264. Let F be a collection of continuous functions from X

into the unit interval. Then:
(a) eF is continuous;
(b) If F separates points, then eF is one-to-one;
(c) If X is a T1 -space and F separates points from closed sets, then
F
eF : X → eF [X] is a homeomorphic embedding of X into [0, 1] .

Corollary 265. A space X is completely regular if and only if X is

κ
homeomorphic to a subspace of [0, 1] for some cardinal κ.

Corollary 266. Any completely regular space X has a Hausdorff com-

pactification.

Let C(X, [0, 1]) be the collection of all continuous functions from a
completely regular space X into the unit interval [0, 1], and let F be any
subfamily of C(X, [0, 1]) which separates points from closed sets. Let eF
F
be the embedding of X into [0, 1] defined above, and let eF X = eF [X],
F
where the closure is taken in [0, 1] . It follows from Theorem 264(c)
that eF X is a compactification of X.

Theorem 267. If αX is any compactification of X, there exists a sub-

family F of C(X, [0, 1]) for which eF X ≈ αX.

6.1.4 The Stone-Čech Compactification

The Stone-Čech compactification of a completely regular space X is
defined to be eF X, where F = C(X, [0, 1]). The Stone-Čech compactifi-
cation of X is denoted by βX, and the embedding eC(X,[0,1]) is denoted
by βX , or just β if X is understood.

Theorem 268. Suppose h : X → Y is a continuous map from X into

C(X,[0,1])
Y (both T3.5 ). Then there exists a continuous H : [0, 1] →

69
 6 The Stone-Čech Compactification

C(Y,[0,1])
[0, 1] such that H ◦ βX = βY ◦ h. In particular, H restricted to
βX maps βX into βY .

Theorem 269. For every compact T2 -space Y and each continuous map
f : X → Y , there exists a continuous F : βX → Y such that f = F ◦βX .

If, as often is done, we think of X as a subset of βX, and dispense with

the map βX : X → βX, then Theorem 269 says that any continuous
map from X to a compact space can be extended to βX.

Corollary 270. If αX is any compactification of X, then βX > αX.

Also, any compactification of X which satisfies the condition of Theo-
rem 269 is equivalent to βX.

Theorem 271. [0, ω1 ] is the only (up to equivalence) Hausdorff com-

pactification of the space [0, ω1 ) of countable ordinals. Hence β[0, ω1 ) =
[0, ω1 ].

Theorem 272. Let ω be the discrete space of natural numbers. Then

βω maps continuously onto any separable compact Hausdorff space.

Corollary 273. |βω| > 2c .

Lemma 274. If X is a separable T2 -space, then |X| 6 2c .

Corollary 275. |βω| = 2c .

F Theorem 276. Let Z be a compact Hausdorff space, and let X be a

dense subspace of Z (so Z is a compactification of X). If every contin-
uous f : X → I extends some continuous f ∗ : Z → I, then Z ≈ βX.

6.1.5 Four More Results

In the next 4 results assume, as we may, that X ⊂ βX and that the
mapping βX : X → βX is the identity on X.

Theorem 277. If X is normal, then any two disjoint closed subsets of

X have disjoint closures in βX.

70
 6.2 The Topology of βω

Theorem 278. Let X be a normal space, and let H ⊂ X be closed.

Then clβX H ≈ βH.

Corollary 279. βω is homeomorphic to a subspace of β[0, 1); also of

βR.

Theorem 280. β[0, 1) \ [0, 1) is connected, and βR \ R has exactly two

components.

6.2 The Topology of βω

Now we will investigate the space βω. We have already seen that it is
surprisingly large. Next we’ll see that this space can be conveniently
described as the space of all ultrafilters on ω with a certain topology.
Recall, a collection F of subsets of a set X is called a filter on X if:
Tn
(1) Whenever F1 , F2 , . . . , Fn ∈ F, then i=1 Fi ∈ F;

(2) Whenever F ∈ F and F ⊂ G ⊂ X, then G ∈ F;

(3) ∅ ∈
/ F.
In other words, a filter on X is a collection of nonempty subsets of
X which is closed under supersets and finite intersections.
Also, a filter F on X is called an ultrafilter if F is not properly
contained in any other filter on X. A trivial example of an ultrafilter
on X is the collection of all subsets of X containing a fixed element
T like this are called fixed ultrafilters. A filter F is
x0 in X. Ultrafilters
said to be free if F = ∅. The Axiom of Choice is needed to show
the existence of free ultrafilter’s. A standard Zorn’s Lemma argument
(or a nearly trivial Tukey’s Lemma argument) shows that any filter on
X is contained in an ultrafilter, so applying this to any free filter, for
example, to the filter of co-finite subsets of an infinite set X, gets you
a free ultrafilter on X.

Theorem 281. Let F be a filter on X. Then the following are equiva-

lent:
(a) F is an ultrafilter;

71
6 The Stone-Čech Compactification

(b) If G ⊂ X and G ∩ F 6= ∅ for every F ∈ F, then G ∈ F;

(c) For every G ⊂ X, either G ∈ F or X \ G ∈ F.

Theorem 282. Let p ∈ βω \ ω, and let Fp = {U ∩ ω : p ∈ U ◦ }. Then:

(a) Fp is a free ultrafilter on ω;

(b) p 6= q → Fp 6= Fq ;

(c) If F is a free ultrafilter on ω, then there is a unique p ∈ βω \ ω

with Fp = F.

Theorem 283.
(a) If A ⊂ ω, then clβω A = A ∪ {p ∈ βω \ ω : A ∈ Fp };

(b) For each A ⊂ ω, clβω A is clopen in βω;

(c) The collection {clβω A : A ⊂ ω} forms a base for βω.

Now define a space as follows. The set for the space is

ω ∪ {F : F is a free ultrafilter on ω}.

It is easy to check that, for any A, B ⊂ ω, we have A ∩ B = A ∩ B. It

follows that {A : A ⊂ ω} is a base for a topology on the space.

Theorem 284. The space as defined above is homeomorphic to βω.

We now explore more topological properties of βω. By compactness,

every infinite subset of ω has a limit point in βω, and said limit point
must of course be in βω \ ω, but we’ll see we do not have sequential
convergence:

Theorem 285. No nontrivial—that is, not eventually constant—sequence

in ω converges to a point of βω.

Of course, the “interesting” points of βω are the points of the remain-

der βω \ ω, which is often denoted by ω ∗ . Note that ω ∗ , being closed in
βω, is also a compact space. For each infinite A ⊂ ω, let A∗ = clβω A∩ω ∗ .

72
 6.2 The Topology of βω

Then by Theorem 283, the collection {A∗ : A ⊂ ω ∧ |A| = ℵ0 } is a

clopen base for ω ∗ .

Lemma 286. Let A, B ⊂ ω. Then:

(a) A∗ ∩ B ∗ = (A ∩ B)∗ ;
(b) A∗ ∩ B ∗ = ∅ if and only if A ∩ B is finite;
(c) A∗ ⊂ B ∗ if and only if A \ B is finite.

6.2.1 ω ∗ is not separable

For A, B ⊂ ω, we say that:
(a) A and B are almost disjoint if A ∩ B is finite;
(b) A is almost included in B (denoted A ⊂∗ B) if A \ B is finite.
Of course, any disjoint family of subsets of ω is countable, but as we
saw last chapter. . .

Theorem 287. There is a pairwise almost disjoint family A of subsets

of ω of cardinality c.

Corollary 288. There is a family of c-many disjoint open subsets of ω ∗ ;

in particular, ω ∗ is not separable.

6.2.2 A strange property of βω

Now we want to see a very strange property of βω: it contains no
nontrivial convergent sequences at all. First, recall that a subset D of
a space X is relatively discrete if the subspace topology on D is the
discrete topology. Equivalently, D being relatively discrete means that
no point of D is a limit point of D; this should be compared with the
concept of D being closed discrete, which means no point of the whole
space X is a limit point of D.

Lemma 289. If D is a countably infinite relatively discrete subset of a

regular space, then there is a family of disjoint open sets {Ud : d ∈ D}
with d ∈ Ud for each d ∈ D.

73
 6 The Stone-Čech Compactification

Now our “no convergent sequences” in βω result follows quite easily

from:

Theorem 290. If D ⊂ βω is countably infinite and relatively discrete,

then D ≈ βD (which of course is homeomorphic to βω).

Corollary 291. βω contains no nontrivial convergent sequence.

Lemma 292. Every infinite Hausdorff space contains an infinite rela-

tively discrete subset.

6.2.3 Closed subsets of βω

The following says that closed subsets of βω are either small (finite) or
very big (cardinality 2c ).

Theorem 293. Every infinite closed subset of βω (or of ω ∗ ) contains

a copy of βω.

Lemma 294. If A0 , A1 , A2 , . . . are infinite subsets of ω, and An+1 ⊂∗

An for all n < ω, then there is an infinite A ⊂ ω such that A ⊂∗ An
for all n < ω.

Such a set A as in Lemma 294 is sometimes called a pseudo-intersection

of the An .

Theorem 295. Every non-empty Gδ set in ω ∗ has non-empty interior.

6.2.4 P -points
A point p in a space X is called a P -point if every Gδ set containing p
is a neighborhood of p (equivalently, the intersection of countably many
neighborhoods of p is again a neighborhood of p).

F Theorem 296. Assume CH. Then ω ∗ has a P -point.

Lemma 297. Every infinite compact T2 -space has a non-P -point.

74
 6.2 The Topology of βω

6.2.5 ω ∗ is not homogeneous

A space X is said to be homogeneous if for any x, y ∈ X, there is a
homeomorphism h : X → X such that h(x) = y.

Corollary 298. Assume CH. Then ω ∗ is not homogeneous.

With a significantly more difficult argument, in the late 1960’s, Z. Fro-

lik proved that ω ∗ is not homogeneous without using any special (that
is, beyond ZFC) axioms of set theory.
As we’ve seen, ω ∗ is the space of free ultrafilters on ω with a certain
topology. When one first encounters the notion of free ultrafilters, they
may seem rather elusive since you can’t define one without the help
of the Axiom of Choice. So Corollary 298 may be a little surprising,
because it shows there are some free ultrafilters that really are different
from other free ultrafilters.

6.2.6 Products of countably compact spaces

Let X be a space and let A be an infinite subset of X. Call a point p
an infinite accumulation point of A if every neighborhood of p contains
infinitely many points of A.

Theorem 299. A topological space X is countably compact if and only

if every infinite subset of X has an infinite accumulation point.

In a T1 -space, a point p is a limit point of a set A if and only if it is

an infinite accumulation point of A. So we have the following corollary.

Corollary 300. If X is a T1 -space, then X is countably compact if and

only if X does not contain an infinite subset with no limit point (which
would be an infinite closed discrete set).

Lemma 301. Suppose X, Y ⊂ βω and X ∩ Y = ω. Then X × Y is not

countably compact.

Theorem 302.

75
 6 The Stone-Čech Compactification

(a) There is a subset X of ω ∗ such that X and ω ∗ \ X meet every

infinite closed subset of ω ∗ ;
(b) If X is as in 302(a), then ω ∪ X is countably compact.

Corollary 303. There are countably compact spaces X, Y ⊂ βω such

that X ∩ Y = ω and hence X × Y is not countably compact.

6.2.7 Products of sequentially compact spaces

In contrast to countable compactness, sequential compactness is count-
ably productive.
Q
F Theorem 304. If Xn , n < ω, are sequentially compact then so is n<ω Xn .

But not every product of sequentially compact spaces is sequentially

compact.
[0,1]
F Theorem 305. [0, 1] is not sequentially compact.

Let A (∗ B denote that A ⊂∗ B but A 6=∗ B; that is, A \ B is

finite but B \ A is infinite. Let α be an ordinal. A sequence Aβ , β < α,
of infinite subsets of ω is called a (decreasing) tower of length α if
β < γ < α implies Aγ (∗ Aβ . If an infinite set A has the property that
A ⊂∗ Aβ for every β < α, we call A a pseudo-intersection of the tower
{Aβ : β < α}.
Given any decreasing tower, if it has an infinite pseudo-intersection
we can extend it. If we inductively keep extending until we can extend
no more, we arrive at a decreasing tower which does have an infinite
pseudo-intersection. We define t to the least cardinal κ such that there
is a tower of length κ with no infinite pseudo-intersection.
Obviously, t 6 c. It follows from Lemma 294 that t > ω1 .

F Theorem 306. t 6 b.

Theorem 307. Assume MA. Then t = c.

The next theorem is a strengthening of Theorem 304.

76
 6.2 The Topology of βω

Theorem 308. A product of fewer than t-many sequentially compact

spaces is sequentially compact, and the product of t-many sequentially
compact spaces is countably compact.

In particular, since t > ω1 , the product of ω1 -many sequentially

compact spaces is always countably compact. The following related
problem is a well-known unsolved problem (the “Scarborough-Stone
problem”):

Question. Must the product of any family of regular sequentially com-

pact spaces be countably compact?
The regular assumption is in there because there is known to be a
collection of Hausdorff sequentially compact spaces whose product is not
countably compact. In 1984, E.K. van Douwen proved that assuming
b = c (in particular, CH or MA) that there is a family of 2c -many
sequentially compact regular spaces whose product is not countably
compact (that 2c = |βω| is not a coincidence: he used βω to assist him
in the construction).
It is still not known if one can construct without assuming any special
axioms of set theory a family of regular sequentially compact spaces
whose product is not countably compact.

6.2.8 Characterizing βX
Now we want to get a characterization of βX in general, in the same
spirit as out characterization of βω in terms of ultrafilters. Roughly
speaking, the characterization is obtained by replacing ultrafilters on ω
with ultrafilters of zero-sets in X. We’ll apply our characterization to ob-
tain an interesting property of β[0, ∞) \ [0, ∞) (which by Corollary 259
is compact and by Theorem 280 is connected).
A subset F of a space X is called a zero-set in X if F = g −1 (0)
for some continuous g : X → R. It is equivalent to say, instead of
F = g −1 (0), that F = g −1 (C) for some closed C ⊂ R. The complement
of a zero-set is called a cozero-set.

Lemma 309.

77
6 The Stone-Čech Compactification

(a) Each point of a completely regular space has a neighborhood base

of zero-sets and an open neighborhood base of cozero-sets.

(b) Every zero-set F in X is closed and a regular Gδ Tset; that is,

there
T are open sets Un , n < ω, such that F = n<ω Un =
n<ω U n .

(c) In a normal space, the zero-sets are precisely the Gδ sets; so in

a perfectly normal space, every closed set is a zero-set.

In the following results, when relevant, we always assume X ⊂ βX.

Lemma 310.
(a) If F1 , F2 are zero-sets, so is F1 ∪ F2 .
T
(b) If Fn is a zero-set for each n < ω, then F = n<ω Fn is a
zero-set.

(c) If F is a zero-set in X, and Y ⊂ X, then F ∩ Y is a zero-set in

Y.

(d) If F1 and F2 are disjoint zero-sets, then there exists a continuous

f : X → [0, 1] with F1 = f −1 (0) and F2 = f −1 (1).

(e) Disjoint zero-sets in X have disjoint closures in βX.

A z-filter in X is a collection F of zero-sets satisfying:

(1) ∅ ∈
/ F;

(2) F1 , F2 ∈ F → F1 ∩ F2 ∈ F;

(3) If F ∈ F and G is a zero-set containing F , then G ∈ F.

A maximal z-filter is called a z-ultrafilter. It is straightforward to show
that a z-filter F is a z-ultrafilter if and only if F has the following
property: For each zero-set Z, if Z ∩ F 6= ∅ for each F ∈ F, then
Z ∈ F.

78
 6.2 The Topology of βω

Theorem 311. Let X be a completely regular space, and for each p ∈

βX \ X let

Fp = {F ⊂ X : F is a zero-set in X and p ∈ clβX F }.

Then:
(a) Fp is a z-ultrafilter in X.

(b) For every zero-set F in X, clβX F = F ∪{p ∈ βX \ X : F ∈ Fp }.

(c) The collection {βX \ clβX F : F is a zero-set in X} is a base

for the topology of βX.

(d) For each cozero-set U in X, let

Ex U = U ∪ {p ∈ βX \ X : ∃F ∈ Fp (F ⊂ U )}.

Then Ex U = βX \ clβX (X \ U ), and {Ex U : U ⊂ X, U cozero}

forms a base for βX.

(e) For a cozero subset U of X, Ex U is the largest open subset of

βX whose intersection with X equals U .

6.2.9 Continua
A continuum is a compact connected Hausdorff space, though many
topologists include the adjective “metrizable” in the definition. A con-
tinuum is indecomposable if it cannot be written as the union of two
proper subcontinua. The Knaster continuum, or buckethandle contin-
uum, constructed from the Cantor set by joining certain pairs of points
with semicircles, is an example of an indecomposable continuum in the
plane.

F Theorem 312. A continuum X is indecomposable if and only if every

proper subcontinuum is nowhere-dense in X.

Let H be the half-line [0, ∞). By Theorem 280, H∗ = βH \ H is a

continuum. Our goal now is to prove it is indecomposable.

79
6 The Stone-Čech Compactification

Lemma 313. Let O be open in H such that 0 ∈ / O and H \ O is un-

there are an , bn ∈ H with a0 < b0 < a1 < b1 < a2 < . . .
bounded. Then S
such that O ⊂ n<ω (an , bn ).

Lemma 314. Let K be a proper subcontinuum of H∗ . Then there are

an , bn ∈ H with an < bn < an+1 for all n such that:
S
(a) If O = n<ω (an , bn ), then K ⊂ Ex O;
S
(b) If u = {A ⊂ ω : K ⊂ clβH ( n∈A [an , bn ])}, then u is an ultrafil-
ter on ω.

Proposition F. If p ∈ Ex O \ X, then there exists F ∈ Fp where F ⊆ O.

Since F ∈ Fp , p ∈ clβX F ⊂ clβX O.

Theorem 315. H∗ is indecomposable.

80
7 Topological Games
7.1 Completeness & the Baire Property
In this section we will introduce Čech completeness, which for metrizable
spaces is equivalent to completely metrizable. It makes sense to study
this after our chapter on compactifications, because the definition of
Čech complete is in terms of the Stone-Čech compactification. This
study will also lead naturally into the topic of topological games.
In this section, when we talk about a compactification αX of X we
will assume that X ⊂ αX and that the map α : X → αX is the identity
on X.

7.1.1 Complete metric spaces

First, recall that a metric space hX, di is complete if every d-Cauchy
sequence converges. A space hX, T i is completely metrizable if X has
a complete metric compatible with T . We have also proved:
(1) The following are equivalent for a complete metric d on a space
X:
(a) d is complete;
(b) If A0 , A1 , . . . is a decreasing
T sequence of closed sets and
diam An → 0, then n<ω An 6= ∅.

(2) If hX, di is complete and Y ⊂ X, then hY, d  Y 2 i is complete if

and only if Y is closed in X, but Y is completely metrizable if
and only if Y is a Gδ set in X.

(3) Any metric space hX, di has a completion; that is, a unique (up
˜ such that X is dense
to isometry) complete metric space hX̃, di
in X̃ and d˜  X × X = d.

81
7 Topological Games

(4) Every completely metrizable space is a Baire space; that is, the
intersection of countably many dense open sets is dense.
For example, since the subset P of irrationals in the real line R is
not closed in R, the Euclidean metric on P is not a complete metric;
however, P is completely metrizable since R is and P is Gδ in R. One
can obtain a complete metric on P using P∼ =ω ω and for f = 6 g in ω ω
n
defining d(f, g) = 1/2 if and only if n is least such that f (n) 6= g(n).
On the other hand, the space Q of rationals is not Gδ in R so Q is
not completely metrizable and thus there is not a complete metric on
Q which generates the usual topology of Q.

7.1.2 Čech complete

A completely regular space X is Čech complete if X is a Gδ set in some
compactification αX of X (equivalently, αX \ X is σ-compact).
Note that every locally compact completely regular space is Čech
complete, since locally compact spaces are open in any compactification.
The irrationals are Gδ in the one point compactification of the reals, so
they are Čech complete. We will soon see that in fact any completely
metrizable space is Čech complete.

Lemma 316. If αX and γX are compactifications of X, and f : αX →

γX is a continuous function with f  X = idX , then f −1 [γX \ X] =
αX \ X.

Theorem 317. The following are equivalent for a completely regular

space X:
(a) X is Čech complete;
(b) X is a Gδ set in βX;
(c) X is a Gδ set in every compactification of X;
(d) X is a Gδ set in every completely regular space in which it is
densely embedded.

Theorem 318. The following are equivalent for a completely regular

space X:

82
 7.1 Completeness & the Baire Property

(a) X is Čech complete;

(b) There is a sequence U0 , U1 , U2 , . . . of open covers of X such that:

whenever F is a filter of closed sets such that for each n < ω
there
T is some Fn ∈ F with Fn ⊂ U for some U ∈ Un , then
F 6= ∅.

Exercise 319. True or False:

(a) The Sorgenfrey line is Čech complete.

(b) The tangent disk space is Čech complete.

Theorem 320. The following are equivalent for a metrizable space X:

(a) X is Čech complete;

(b) X is completely metrizable.

Corollary 321. Let X be completely metrizable, and A ⊂ X. Then A

is completely metrizable if and only if A is Gδ in X.

Theorem 322.
(a) If X is a closed or Gδ subspace of a Čech complete space Y ,
then X is Čech complete.
Q
(b) If Xn is Čech complete for each n < ω, then n<ω Xn is Čech
complete.

F Example 323. ω ω1 is not Čech complete.

7.1.3 The Baire property

Theorem 324. Every Čech complete space is a Baire space.

There is no decreasing sequence version (as there is for complete

metric spaces) of Theorem 318(b) (by this we mean replacing the filter
F with a decreasing sequence of closed sets hFn in<ω ) that is equivalent
to Čech completeness. To see this, not that if there were, then any

83
 7 Topological Games

completely regular countably compact space would be Čech complete

(since in a countably compact space, every decreasing sequence of closed
sets has non-empty intersection). But the countably compact space we
encountered in Theorem 302 is not Čech complete:

Example 325. Let X be a subspace of ω ∗ such that X and ω ∗ \ X

meet every infinite closed subset of ω ∗ . Then X is complete regular and
countably compact but not Čech complete.

Regular countably compact spaces are, however, Baire:

Theorem 326. Every regular countably compact space is a Baire space.

There are separable metrizable spaces that are Baire but not Čech
complete; equivalently, not completely metrizable. A subset B of the
real line is Bernstein if both B and R\B meet every uncountable closed
subset of R.

Theorem 327. Let B be a Bernstein subset of the real line. Then B is

a Baire space but is not completely metrizable.

Theorem 328.
(a) If X is an open subspace, or a dense Gδ subspace, of a Baire
space, then X is Baire.

(b) If X has a dense Baire subspace, then X is Baire.

F Example 329. There is a subspace X of the real line which is Baire

but has a closed (and hence Gδ too) non-Baire subspace.

F Theorem 330. If X is Baire and Y is Baire and second-countable, then

X × Y is Baire.

Theorem 331. If X is Baire, and Y is regular and countably compact

or Čech complete, then X × Y is Baire.

84
 7.2 The Choquet game

7.2 The Choquet game

Let X be a space. Define a two-person infinite game on X, called
the Choquet game, or the Banach-Mazur game, and denoted Ch(X),
as follows. There are two players, Empty (E) and Nonempty (NE).
In the first round, E chooses a nonempty open set U0 , and NE then
chooses a nonempty open subset V0 of U0 . In the nth round, E chooses a
nonempty open subset Un of Vn−1 , and NE responds with a nonempty
open Vn ⊂ Un . There is a round for every n < ω.
A play of the game produces a decreasing sequence

U0 ⊃ V0 ⊃ U1 ⊃ V1 ⊃ U2 ⊃ . . .
T
of nonempty
T openTsets. We say NE wins the game if n<ω Un =6 ∅ (note
that n<ω Un = n<ω Vn ,Tsince the sequence is decreasing). Otherwise
E wins; that is, E wins if n<ω Un = ∅.

7.2.1 Strategies
A strategy for player E is a function σ from finite decreasing sequences
of nonempty open sets (including the empty sequence) of the form

hU0 , V0 , U1 , V1 , . . . , Un , Vn i

to the collection of all nonempty open sets such that, for each n < ω,

σ(hU0 , V0 , U1 , V1 , . . . , Un , Vn i) ⊂ Vn .

Such a strategy σ is called a winning strategy for E if whenever

U0 , V0 , U1 , V1 , U2 , . . .

is an infinite decreasing sequence of open sets such that U0 = σ(∅) and

Un+1 = σ(hU0 , V0 , U1 , V1 , . . . , Un , Vn i)
T
for each n < ω, then n<ω Un = ∅. A strategy and winning strategy
from NE is defined analogously.

85
7 Topological Games

Obviously, E and NE cannot both have a winning strategy. But

it is possible that neither does; in such a case we say the game is
undetermined .
When considering so-called “perfect information” winning strategies
as above, we may consider a winning strategy for E or NE to be a
function σ defined on finite sequences of the opponents moves only,
and such that if σ(U0 , . . . , Un ) is the player’s move given the opponent
has played U0 , . . . , Un , then that player is guaranteed to win. This can
simplify notation for arguments involving such strategies.

Theorem 332. A space X is a Baire space if and only if E has no

winning strategy in the game Ch(X).

Theorem 333. If X is either Čech complete, or regular and countably

compact, then NE has a winning strategy in Ch(X).

Theorem 334. Let B be a Bernstein subset of the real line. Then Ch(B)
is undetermined.

It is sometimes useful to know that one may assume a player is

choosing from a certain base B for X. Let Ch(X, B) denote the game
in which both players always choose a member of B. The next theorem
implies that the game Ch(X, B) is equivalent to the original game
Ch(X) in the sense that a player has a winning strategy in one game
if and only if they have one in the other.

Theorem 335. Let B and C be bases for a space X. Then E (resp., NE)
has a winning strategy in Ch(X, B) if and only if E (resp., NE) has a
winning strategy in Ch(X, C).

7.2.2 Choquet Spaces

A space X is said to be Choquet if NE has a winning strategy in the
game Ch(X); clearly, Choquet spaces are Baire.
When the game is called the Banach-Mazur game, the players E
and NE are denoted β and α, respectively (and yes, β goes first!), and
the property “Choquet” as defined here is called weakly α-favorable.

86
 7.2 The Choquet game

α-favorable means NE has a winning strategy depending only on E’s

last move (sometimes called a stationary winning strategy, or a tactic).

Theorem 336. Any product of Choquet spaces is Choquet.

Corollary 337. Any product of Čech complete spaces is Choquet, hence

Baire.

Note that by Theorem 322 and Example 323, while a countable prod-
uct of Čech complete spaces is always Čech complete, an uncountable
product need not be Čech complete. But the above corollary says the
product will still be Baire; in particular, ω ω1 is Baire but not Čech
complete, and any product of real lines is Baire.

7.2.3 Products of Baire Spaces

It’s time we saw an example of Baire spaces X and Y whose product
is not Baire. By various results above, neither X nor Y can be Čech
complete, regular countably compact, or second-countable Baire. In
particular, if we want metrizable such X and Y , then neither factor can
be separable. But we will presently see that there are metrizable Baire
spaces X and Y of weight ω1 such that X ×Y is not Baire. The example
is due to Fleissner and Kunen; examples were previously known, but
their example is simpler, and a nice application of stationary sets.
First let’s fix some notation. The spaces are going to be subspaces
of ω1ω , where the set ω1 of countable ordinals is given the discrete
topology. So points in ω1ω are functions from ω to ω1 , or equivalently,
infinite sequences (in order type ω) of countable ordinals.
Let ω1<ω denote the set of all finite sequences of countable ordinals.
If σ ∈ ω1<ω , let [σ] = {f ∈ ω1ω : σ ⊂ f }. Note that σ ⊂ f simply means
that the infinite sequence f extends the finite sequence σ. It is easy to
check that B = {[σ] : σ ∈ ω1<ω } is a base for ω1ω . Also, for each f ∈ ω1ω ,
the collection {[f  n] : n < ω} is a local base at the point f .
For f ∈ ω1ω , let f ∗ = sup {f (n) : n < ω}. Of course, f ∗ is a countable
ordinal. For A ⊂ ω1 , let A∗ = {f ∈ ω1ω : f ∗ ∈ A}.

F Lemma 338.

87
7 Topological Games

(a) If A ⊂ ω1 is uncountable, then A∗ is dense in ω1ω .

(b) If A and B are uncountable disjoint subsets of ω1 , then A∗ × B ∗

is not Baire.

By the previous lemma, if we can find two disjoint uncountable sub-

sets A and B of ω1 such that A∗ and B ∗ are Baire, we’ll have our
example. It turns out that disjoint stationary sets (we saw last chapter
that they exist) will do the job. First a technical lemma to be used as
a tool in the proof of the theorem.

Lemma 339. Let U be dense open in ω1ω . For each σ ∈ ω1<ω , choose
an extension σU of σ such that [σU ] ⊂ U . Then

CU = {δ ∈ ω1 : ∀σ ∈ δ <ω (σU ∈ δ <ω )}

is club in ω1 .

Theorem 340. Let A be a stationary subset of ω1 . Then A∗ is Baire.

Corollary 341. There are metrizable Baire spaces X and Y such that
X × Y is not Baire.

7.3 W -Spaces
7.3.1 The W -Game
Consider the following topological game, denoted GO,P (X, x), played
in a space X at a point x ∈ X: In round n, Player O (for “Open”) plays
an open neighborhood Un of x, and Player P (for “Point”) selects a
point xn ∈ Un . Player O wins the game if hxn i → x; otherwise P wins.
Let us call this game the W -game (on X, at x) for short. If O has a
winning strategy in the W -game at every x ∈ X, then X is called a
W -space.
A space X is said to be a Fréchet space if whenever x ∈ A, there are
xn ∈ A with hxn i → x.

88
 7.3 W -Spaces

Theorem 342. Every first-countable space is a W -space and every W -

space is a Fréchet space.

The one-point compactification of an uncountable discrete space is

an example of a W -space which is not first-countable: all O needs to
do is to make P choose a different point each time.
Here’s an example of a Fréchet space which is not a W -space (in fact,
P has a winning strategy). It is called the sequential fan, and is denoted
by Sω . It can be described as the quotient space of the topological
sum of countably many convergent sequences obtained by identifying
the limit points of the sequences to a single point ∞. This makes ∞
the only non-isolated point. The space can also be described as the
set (ω × ω) ∪ {∞}, where points of ω × ω are isolated, and for each
function f : ω → ω, a basic neighborhood U (f ) of ∞ is defined by
U (f ) = {∞} ∪ {hm, ni ∈ ω × ω : n > f (m)}.

Example 343. The sequential fan Sω is Fréchet, but P has a winning

strategy in the W -game at ∞.

Theorem 344. Every countable W -space is first-countable, as is every

regular separable W -space.

Theorem 345.
(a) Every subspace of a W -space is a W -space; in particular, every
countable subspace of a W -space is first-countable;
(b) The product of countably many W -spaces is a W -space.

Something more general than Theorem 345(b) above holds. For each
α < κ pick a point pα in a space Xα . Then the set

Σα<κ Xα = {~x ∈ Πα<κ Xα : |{α : xα 6= pα }| 6 ω},

with the topology as a subspace of the usual product topology, is a

called a Σ-product of the Xα ’s (with base point p~ = hpα iα<κ ).

Theorem 346. Let {Xα : α < κ} be a collection of W -spaces. Then

Σα<κ Xα is also a W -space.

89
 7 Topological Games

7.3.2 w-spaces
If the point picker does not have a winning strategy in the W -game at
x in a space X, then x is called a w-point and X is a w-space if every
point is a w-point. So, if X is a w-space which is not a W -space, then
at some point in X the game is undetermined.
Though weaker than W , w-spaces still have strong convergence prop-
erties. A space X is said to be strongly Fréchet, or countably bi-sequential,
if whenever A0 , A1 , . . . is a decreasing sequence of subsets of X with
x ∈ An for all n < ω, then there are xn ∈ An with hxn i → x. A space
X is said to have countable tightness if whenever x ∈ A, then there is
a countable subset B of A with x ∈ B.
Some of the implications in the next theorem are trivial or already
done; they are listed for the sake of comparison and future reference. By
347(c)↔347(d) below, the w-space property is equivalent to a slightly
more general form of strongly Fréchet.

F Theorem 347. Each statement below implies the next, and 347(d) im-
plies 347(c).
(a) X is a W -space;

(b) X has countable tightness and every countable subspace is first-

countable;

(c) Whenever x ∈ An for every n < ω, then there are xn ∈ An with

xn → x;

(d) X is a w-space;

(e) X is strongly Fréchet;

(f) X is Fréchet;

(g) X has countable tightness.

We saw that the sequential fan Sω is Fréchet; it is easily seen not to

be strongly Fréchet. If p is a free ultrafilter on ω, then the subspace
ω ∪ {p} of βω has countable tightness (trivially), but is not Fréchet.

90
 7.3 W -Spaces

Finding a w-space that is not a W -space is harder; indeed, Gruen-

hage left it as an open question in his original paper on W -spaces.
The following example is due to Istvan Juhasz. If T is a tree, the
interval topology on T is the topology with basis sets of the form
(s, t] = {x ∈ T : s < x 6 t}. Since the predecessors of any element of T
is well-ordered, this is a locally compact topology, and so has a one-
point compactification T ∪ {∞}. Let t↓ = {s ∈ S T : s 6 t}. It is easy to
see that K ⊂ T is compact if and only if K ⊂ t∈F t↓ for some finite
F ⊂ T . Note that a countably infinite subset S of T converges to ∞
if and only if t↓ ∩ S is finite for each t ∈ T . Examples of such S are
antichains and cofinal subsets of some branch.

Example 348. Let T be an Aronszajn tree with the interval topology,

and T ∪ {∞} its one-point compactification. Then T ∪ {∞} is not a
W -space, but it has countable tightness and every countable subset is
first-countable (hence it is a w-space).

7.3.3 w but not W

Example 348 leaves open the question of whether or not there could be a
countable w-space which is not W . It turns out there are such examples,
one such being obtained from something called a “Hausdorff gap.” A
Hausdorff gap is a pair hhaα iω1 , hbα iω1 i of ω1 -sequences of subsets of ω
satisfying:
(1) aα ⊂∗ αβ ⊂∗ bβ ⊂∗ bα for each α < β < ω1 ;
(2) There is no c ⊂ ω such that aα ⊂∗ c ⊂∗ bα for each α < ω1 .
So the sets aα on the “left side” of the gap are ⊂∗ -increasing, the
right side is ⊂∗ -decreasing, every set on the right contains mod-finite
every set on the left, and there is no set in between. To get a feel for
this, try the following exercise.

Theorem 349. Let han i and hbn i be sequences of subsets of ω with

am ⊂∗ an ⊂∗ bn ⊂∗ bm for each m < n < ω. Then there is a subset c
of ω such that an ⊂∗ c ⊂∗ bn for each n < ω.

The existence of a Hausdorff gap, sometimes called an (ω1 , ω1∗ )-gap,

91
7 Topological Games

in ZFC (which was proven by Hausdorff himself!) might be surprising

in view of Theorem 307, the proof of which shows that every decreasing
tower of length ω1 has an infinite pseudo-intersection assuming MA(ω1 ).
So its existence can be thought of as illustrating an absolute property
of the cardinal ω1 , which can’t be negated by MA or anything else.
Here’s why we are interested now in Hausdorff gaps. Given such, let
I = {x ⊂ ω : x ∩ aα is finite for every α < ω1 }. Note that I is closed
under finite unions, and that besides finite sets, for any α the set ω \ bα
is in I.
Now let X = ω ∪ {∞}, where the points of ω are isolated, and a
neighborhood of ∞ has the form X \ I, where I ∈ I. There are three
easy obervations about X:
(1) every aα converges to ∞;

(2) {∞} ∪ bα is open for each α; and

(3) if ∞ ∈ A, then A must meet some aα in an infinite set.

Example 350. The space X = ω ∪ {∞} defined above is a w-space but

not a W -space.

92
8 Hints
Hint for Theorem 3
Given f : X → P(X), consider the set Y = {x ∈ X : x ∈
/ f (x)}.

Hint for Theorem 5

A transitive subset of Ord is a member of Ord. (A set X is called
transitive if Y ⊂ X whenever Y ∈ X.)

Hint for Theorem 8

Note that G(0) and G(1) are uniquely defined, given R. Let γ < κ.
If there is a function g with domain δ, where γ < δ, such that for
any α < δ, g(α) = R(g  α), then define G(γ) to be g(γ). Prove by
transfinite induction that G is uniquely defined for every α < κ and
satisfies the condition G(α) = R(G  α) for any α < κ.

Hint for Theorem 51

Well-order Q ∩ [0, 1], handling the endpoints carefully, and induct.

Hint for Example 69

Consider Y = {hx, −xi : x ∈ S} as a subspace of S2 .

Hint for Theorem 68

Recall Lemma 57.

Hint for Theorem 132

Without loss of generality, every member of F has the same cardinality
k. Induct on k.

93
8 Hints

Hint for Example 138

For non-collectionwise normal, show that {{eα } : α ∈ A} is a discrete
collection of singleton sets and use the fact that the usual product
topology on 2P(A) is ccc.

Hint for Theorem 148

For (c → a): Let U be an open cover of X. By assumption, there is a
locally finite closed refinement A of U. For each A ∈ A, choose UA ∈ U
with A ⊂ UA . Since A is locally finite, the collection V of all open sets
which meet only finitely many elements of A is an open cover of X.
Let B be aSlocally finite closed refinement of V. For each A ∈ A, let
WA = X \ {B ∈ B : B ∩ A = ∅}. Finally, let OA = WA ∩ UA . Show
that O = {OA : A ∈ A} is a locally finite open refinement of U.

Hint for Theorem 149

Use Theorem 148(b).

Hint for Lemma 156

S <ω <ω
B = {X \ V : V ∈ [U] }, where [U] is the collection of all finite
subsets of U, is a countable base for X.

Hint for Theorem 159

For (a → b): If N is a countable network for X, show that the topology
on X obtained by taking N ∪ {X \ N : N ∈ N } as a subbase is a
separable metrizable topology finer than the original topology on X.

Hint for Theorem 176

˜ exists (consider
It follows readily from Corollary 175 that such an hX̃, di
the closure of the image of X under the isometric embedding). You
need to show uniqueness.

Hint for Theorem 178

Construct what is often called a “Cantor tree” of clopen sets, as follows.
Let B1 , B2 , . . . enumerate a countable base of nonempty clopen sets.
Divide X = X∅ into two nonempty clopen pieces X0 and X1 such that,

94
unless B1 is the whole space, then B1 is one of the pieces. Then divide
each of X0 and X1 into two pieces, say X00 , X01 , X10 , and X11 such
that each piece is either contained in B2 or completely misses B2 . And
so on. Show that given any infinite sequence ~e = e1 e2 e3 . . . of 0’s and
1’s, the intersection of the “branch” X∅ , Xe1 , Xe1 e2 , . . . is a single point
p, and the nodes of this branch, that is, {X~en : n ∈ N}, form a base at
p.

Hint for Theorem 181

Get such a function whose domain is C and then extend linearly to
[0, 1].

Hint for Lemma 187

For every infinite cardinal κ, the set of all finite subsets of κ also has
cardinality κ.

Hint for Theorem 191

Given X, call a subset A of X left-closed if A is closed, and a ∈ A and b <
a implies b ∈ A. For example, for each x ∈ X, the set {a ∈ X : a 6 x}
is left-closed. (Other left-closed sets are often called “gaps.”) Let X̂ be
all left-closed sets, ordered by ⊂. Include the empty set in X̂ if and
only if X has no least element.

Hint for Theorem 194

Show that the set H of all limit ordinals is a closed set, and any open
superset of H contains all but countably many points of the space.

Hint for Example 201

Let H = {hα, αi : α < ω1 } and K = [0, ω1 ) × {ω1 }. Show H and K are
disjoint closed sets which can’t be separated.

Hint for Lemma 202

Note that 202(a) is essentially equivalent to κ = |κ × κ|.

95
8 Hints

Hint for Theorem 207

For the “if” direction, first show
S that it suffices to construct a locally
finite open refinement of U on U, where U is a connected collection
of open intervals.

Hint for Theorem 209

Since open subspaces of ccc linearly ordered space are also ccc linearly
ordered, it suffices to show X is Lindelöf. Let U be an open cover of
X. Define x ∼ y if and only if [x, y] is covered by some countable
subcollection of U. Show that each equivalence class E is open and is
covered by some countable subcollection of U.

Hint for Theorem 210

For 210(a), define x ∼ y if and only if the interval from x to y is
separable. Note that equivalence classes are convex. Show that the
set of equivalence classes with the natural order satisfies the desired
conditions.

Hint for Theorem 213

Show that 2ω with the topology induced by the lexicographic order is
the same as the usual Tychonoff product topology.

Hint for Theorem 214

For 214(a), define f : 2ω → [0, 1] so that f (~x) = n<ω xn /2n+1 , where
P
xn is the nth -coordinate of ~x. For 214(c), if K is a closed subset of C
note that K × C is homeomorphic to C.

Hint for Theorem 220

Let X be a Suslin line. Define closed subsets Cα , α < ω1 , of X as
follows. Let C0 = ∅. S
If α is a limit ordinal and Cβ has been defined for
all β < α, let Cα = β<α Cβ . If α = γ + 1 and Cγ has been defined,
let Lγ be the convex components of X \ Cγ . For each I ∈ Lγ , choose
a countable sequence of points converging to each endpoint of I. Let
Cγ0 be the collection of these chosen points for all I ∈ Lγ . Then let
Cα = Cγ ∪ Cγ0 .
S Now let Lα be the convex components of X \ Cα , and let T =
α<ω1 Lα . Show that T ordered by ⊃ is a Suslin tree.

96
Hint for Theorem 221
Let T be a Suslin tree, and let ≺ be an arbitrary linear order on T .
Let X be the set of all branches of T . For any branch b, let b(α) be
the member of b in level α of the tree. Given b1 , b2 ∈ X, let α be
minimal such that b1 (α) 6= b2 (α), and then define b1 < b2 if and only
if b1 (α) ≺ b2 (α). Show that X with this order is ccc and nonseparable.

Hint for S Theorem 222

Let T = α<ω1 Lα be as in the hint for Theorem 220. For each α, choose
Iα ∈ Lα . There are disjoint Iα0 , Iα1 ∈ Lα+1 contained in Iα . Show that
{Iα0 × Iα1 : α < ω1 } is pairwise-disjoint.

Hint for Theorem 223

For the second part, study the hint for Theorem 220.

Hint for Theorem 225

Use the partial order O(X) defined on page 57.

Hint for Lemma 228 S S

Show there exists α0 < ω1 such that S β>α Uβ = β>α0 Uβ for all
α > α0 . Let P = {O ∈ O(X) : O ⊂ β>α0 Uβ } and apply MA, where
O(X) is the partial order defined on page 57.

Hint for Theorem 232

Suppose not; there are points xα , α < ω1 , with xα ∈
/ {xβ : β < α}. Let
Y = {xα : α < ω1 }. It follows from first-countability that every point
of Y is in {xβ : β < α} for some α < ω1 . Use compactness of Y and
Lemma 228 applied to Y to get a contradiction.

Hint for Lemma 233

Let B be a countable base for R. Let P be all pairs of the form hC, ~ F i,
~
where C is a finite sequence hC0 , . . . , Cn i of members of B and F is a
~ 0 , F 0 i 6 hC,
finite subset of κ. Define a partial order 6 on P so that hC ~ Fi
if and only if

97
8 Hints

~ 0 extends C,
(1) C ~

(2) F 0 ⊃ F , and
~ 0 \ dom C
(3) for each i ∈ dom C ~ we have Ci ⊂ T
Uα .
α∈F
For each B ∈ B, k < ω, and α < κ put

~ F i ∈ P : ∃i > k(Ci ⊂ B)}

DB,k = {hC,

~ F i ∈ P : α ∈ F }. Now what can we do with MA(κ)?

and Eα = {hC,

Hint for Theorem 239

Note that if f ∈ ω ω , then the set Kf = {g ∈ ω ω : ∀n[g(n) 6 f (n)]} is
compact.

Hint for Theorem 242

For 242(c) recall when working with Bing’s G—Example 138—we noted
that to prove normality for a space consisting of a closed discrete set
plus isolated points, we need only show that any subset of the closed
discrete set, and its complement in the closed discrete set, can be put
into disjoint open sets.

Hint for Theorem 243

How many Gδ sets can there be in a space with a countable base?

Hint for Theorem 244

Let A ⊂ R have cardinality κ, and let B ⊂ A. Use MA(κ) to show that
there is a sequence hC0 , C1 , . . .i of basic open sets of A such that every
x ∈ B is in infinitely many Ci ’s, while
T everyS y ∈ A \ B is in only finitely
many Ci ’s. Then note that B = n<ω ( i>n Ci ).

Hint for Corollary 248

In a normal space, any countable discrete collection of closed sets can
be separated by disjoint open sets.

98
Hint for Lemma 256 S
For limit α, simply let Tα be the topology generated by β<α Tβ . If α
is a successor, say α = β + 1, then Xα = Xβ ∪ {xβ }, and the task is
to define a neighborhood base Bβ at xβ so that the topology generated
by Tβ ∪ Bβ satisfies 256(a), 256(b), and 256(c).

Hint for Theorem 261

Let γX = X ∪ (αX \ α[X]). Define a natural one-to-one and onto
function h : γX → αX and use h to “pull back” the topology on αX
to define a topology γX.

Hint for Theorem 276

Let f = idX : X → Z and apply Theorem 269; show that the resulting
F : βX → Z is a homeomorphism.

Hint for Theorem 296

Construct an ω1 -sequence hAα iω1 of infinite subsets of ω such that
α < β → Aβ ⊂∗ Aα and u = {B ⊂ ω : ∃α < ω1 (Aα ⊂∗ B)} is an
ultrafilter.

Hint for Theorem 304

Apply Lemma 294.

Hint for Theorem 305

Show that II has a closed subspace which is not sequentially compact.

Hint for Theorem 306

For an infinite subset A of ω, let fA denote its increasing enumeration.
Observe that if B (∗ A, then fA <∗ fB .

Hint for Theorem 312

Make use of Theorem 120.

Hint for Example 323

Show that ω ω1 is not Gδ in (ω + 1)ω1 .

99
8 Hints

Hint for Example 329

Look for an X that is Baire because it contains most of R, but has a
closed subspace homeomorphic to Q.

Hint for Theorem 330

Note that if O is dense open in X × Y , and B is open in Y , then
πX [O ∩ (X × B)] is dense open in X.

Hint for Lemma 338

For 338(b), let Ui be the collection of all hf, gi ∈ ω ω1 × ω ω1 such that
∗ ∗
min {f
T , g } > max {f (i), g(i)}. Then Ui is dense open in ω ω1 × ω ω1
and i∈ω Ui ∩ (A × B ∗ ) = ∅.
∗

Hint for Theorem 347

For (c → d), observe that 347(c) implies Fréchet. Suppose σ is a strategy
for P in the W -game at p ∈ X, where X satisfies 347(c). We need to
show that σ can be beat. Let

P∅ = {σ(U ) : U open, p ∈ U }.

Then p ∈ P ∅ , so there is a countable subset A∅ of P∅ with p ∈ A∅ (A∅

could be taken to be a convergent sequence, but this doesn’t matter for
this argument). Then for each y0 ∈ A∅ , let Uy0 be such that y0 = σ(Uy0 )
and let
Py0 = {σ(Uy0 , U ) : U open, p ∈ U }.
Let Ay0 be a countable subset of Py0 with p ∈ Ay0 . Now for each
y1 ∈ Ay0 , let Uy0 y1 be such that y1 = σ(Uy0 , Uy0 y1 ) and let

Py0 y1 = {σ(Uy0 , Uy0 y1 , U ) : U open, p ∈ U }.

And so on. Apply 347(c) to the resulting Ay0 y1 ...yn ’s.

100
9 Proofs
Proof of Theorem 1
Let pn denote the nth prime number; so p1 = 2, p2 = 3, etc.

(a) Note {pkn : n, k ∈ N} = n∈N {pkn : k ∈ N}. Certainly the right-

S
hand side is a countable union of countable sets. The left-hand side
is countable by the map defined as follows: 1 7→ p11 , 2 7→ p21 , 3 7→ p12 ,
4 7→ p31 , 5 7→ p22 , 6 7→ p13 , 7 7→ p41 , 8 7→ p32 , 9 7→ p23 , etc.
S
(b) A × B = {{a} × B : a ∈ A}, and

hha1 , . . . , an−1 i, an i 7→ ha1 , . . . , an i

is a bijection between (A1 × · · · × An−1 ) × An and A1 × · · · × An .

(c) We may assume A = {pkn : n, k ∈ N}; there is a one-to-one corre-

spondence between the finite subsets of A and N given by ∅ 7→ 1 and
{pkn11 , . . . , pknm
m
} 7→ pkn11 · · · pknm
m
.
The map ha1 , . . . , an i 7→ {ha1 , 1i, . . . , han , ni} gives an injection from
the set of finite sequences in A into A × N.

(d) Certainly |N| 6 |Q| and |Q+ | 6 |N × N|, where Q+ are the positive
rational numbers. 

Proof of Theorem 2
(a) Certainly r 7→ {q < r : q ∈ Q} witnesses |R| 6 |P(Q)| = |P(N)|.
For each XP∈ P(N), let X(n) = 2 if n ∈ X and X(n) = 0 otherwise;
∞
then X 7→ n=1 X(n)/3n is an injection from P(N) to R.

(b) The map f 7→ f −1 (1) gives a bijection between the functions from
N to {0, 1} and P(N).

101
9 Proofs

(c) An infinite sequence of 0’s and 1’s is a function from N to {0, 1}.

(d) Let F be a bijection between R and the set of functions from N to

{0, 1}. Then

hr1 , r2 , . . .i 7→
−1
F (hF (r1 )(1), F (r1 )(2), F (r2 )(1), F (r1 )(3), F (r2 )(2), F (r3 )(1), . . .i)

is a bijection between R and the infinite sequences in R. 

Proof of Theorem 3
Let f : X → P(X) and put Y = {x ∈ X : x ∈
/ f (x)}; then

x∈Y →x∈
/ f (x) → f (x) 6= Y

and
x∈
/ Y → x ∈ f (x) → f (x) 6= Y,

so Y ∈
/ ran f . 

Proof of Theorem 4
See any book about ZFC, such as Kunen’s Set Theory. 

Proof of Theorem 5 S
Let A ⊂ Ord and put a = A. Certainly a ⊂ Ord, so we need to show
a is transitive. Let Y ∈ a; there exists α ∈ A such that Y < α. If β < Y
we necessarily have β < α, so any predecessor of Y is a predecessor of
α. This yields Y ⊂ α (whence ordinals are transitive sets). But α ⊂ a
by construction, so we must have Y ⊂ a. Thus a is transitive and an
ordinal by the hint.
That a is an upper bound of A is immediate. For any β < a there
is α ∈ A with β < α, so no predecessor of a is an upper bound of A.
Thus, a is the least upper bound of A. 

Proof of Corollary
S 6
Observe α = C is an ordinal with countably many predecessors. 

102
Proof of Theorem 7
Suppose on the contrary that the set A of those α < κ for which Sα is
false is nonempty; let β = min A. By 7(1), 0 < β. Since β is minimal,
we must have that Sα is true for each α < β. Then Sβ is true by 7(2)
so β ∈/ A, a contradiction. 

Proof of Theorem 8

Proof of Exercise 9
Given A ∈ B, there exists α < ω1 with A ∈ Bα . Then X \ A ∈ Bα+1 ,
so B is closed under complements.
Let A = {An : n < ω} ⊂ B; by Corollary 6 there exists T α < ω1
such that An ∈ Bα for each n < ω. By construction, A ∈ Bα+1
S B is closed
so T under countable intersections. Note that we may write
A = X \ n<ω (X \ An ), so B is closed under countable unions as
well. 
Proof of Theorem 10
(a → b) Let hP, <i be a partially ordered set with no maximal element
and let f : P(P ) \ {∅} → P be a choice function. For each a ∈ P put
Pa = {x ∈ P : a < x} and observe each Pa must be nonempty.
Fix a0 = f (P ). Suppose aα has been defined for
Tall α < β. If β = α+1,
put aβ T = f (Paα ). If β is a limit ordinal and α<β Paα 6= ∅, define
aβ = f ( α<β Paα ) and continue.
Note that this process
T must terminate; that is, there exists a limit
ordinal γ such that α<γ Paα = ∅. It follows that haα iα<γ is a chain
in hP, <i which has no upper bound in P .

(b → c) Let X be a (nonempty) set and let P be the set of all pairs

hA, ≺A i such that A ⊂ X and ≺A is a well-order on A. Define a partial
order < on P so that hA, ≺A i < hA0 , ≺A0 i if and only if hA0 , ≺A0 i
extends hA, ≺A i; that is, hA, ≺A i is a proper initial segment of hA0 , ≺A0 i.
It is easy to check that hP, <i is a partially ordered set whose every
chain has an upper bound in P .
By ZL there is a maximal element hM, ≺M i in hP, <i. It must be
the case that M = X, else we could extend hM, ≺M i. Hence, ≺M is a

103
9 Proofs

well-order on X.

(c → a) Let A be a collection of nonempty sets; well-order each A ∈ A.

Then A 7→ min A defines a choice function on A. 

Proof of Lemma 11
Certainly
[ [
x∈X\ A↔x∈
/ A ↔ ∀A ∈ A(x ∈
/ A)
\
↔ ∀A ∈ A(x ∈ X \ A) ↔ x ∈ {X \ A : A ∈ A}

and
\ \
x∈X\ A↔x∈
/ A ↔ ∃A ∈ A(x ∈/ A)
[
↔ ∃A ∈ A(x ∈ X \ A) ↔ x ∈ {X \ A : A ∈ A}. 

Proof of Corollary 12
By definition we need to show the complement of each is open, using
Lemma 11 where necessary. Certainly X \ ∅ = X and X \ X = ∅ are
open. If F1 , . . . , Fn are closed in X, then

X \ (F1 ∪ · · · ∪ Fn ) = (X \ F1 ) ∩ · · · ∩ (X \ Fn )

is open as a finite intersection of open sets. If F is a collection of closed

subsets of X, then
\ [
X\ F= {X \ F : F ∈ F}

is open as a union of open sets. 

Proof of Theorem 13
Note that any open set containing x but missing A corresponds to a
closed set containing A but missing x, and every open set containing a
point of A ∪ Ad meets A. 
Proof of Theorem 14
(a) We know ∅ is closed.

104
(b) Certainly B is a closed set containing A.

(c) Note that A is already closed.

(d) Note that A ∪ B ⊂ A ∪ B and A, B ⊂ A ∪ B; use 14(b). 

ProofTof Theorem 15 S
Note {Aα : α ∈ Λ} ⊂ Aβ and {Aα : α ∈ Λ} ⊃ Aβ for each β ∈ Λ.

Proof of Theorem 16
(a) Note X \ A ⊂ K ⊂ X \ A◦ implies A◦ ⊂ X \ K ⊂ X \ X \ A, so
there is no closed set strictly between X \ A and X \ A◦ .
◦
(b) By 16(a), X\(X \ A) = X\[X \ X \ (X \ A)] = X \ (X \ A) = A.


Proof of Theorem 17
(a) Let U be an open neighborhood of x ∈ ∂∂A. Then U is an open
neighborhood of each point of U ∩ ∂A, which must be nonempty. It
follows that U meets A and X \ A, so x ∈ ∂A.

(b) Note x ∈ A if and only if every neighborhood of x meets A, x ∈

/ A◦
X \ A if and only if every neighborhood of x meets X \ A, and x ∈
if no neighborhood of x lies entirely inside A.

(c) A is closed if and only if every point outside of A has a neighborhood

missing A, if and only if every point outside of A is not in the boundary
of A.

(d) Observe that A◦ is a neighborhood of each of its points missing

X \ A, so no point in the interior of A is a point in the boundary of A.
If A is open then A = A◦ , so ∂A ∩ A = ∂A ∩ A◦ = ∅. If ∂A ∩ A = ∅
then every point of A has a neighborhood lying entirely inside A, so A
is open.

(e) By 17(c) and 17(d), A is closed-and-open if and only if ∂A is a

105
9 Proofs

subset of A which misses A. 

Proof of Theorem 18
(a) By definition, an isolated point x of A has a neighborhood which
does not meet A at a point other than x.

(b) Note that x ∈ A\Ad if and only if x has a neighborhood U such that
A ∩ U = {x}. By Theorem 13, A = A ∪ Ad = (A \ Ad ) ∪ Ad = A0 ∪ Ad .

(c) If x is an isolated point of A, there is a neighborhood U of x such

that A ∩ U = {x}. But this implies A ∩ U ⊂ {x}, so x is not a limit
point of A. By 18(a) and 18(b), we must have x ∈ A0 . 

Proof of Theorem 19
→ X, being open, may be written as a union of members of B, so each
point of X is in some member of B.
The intersection of two members of B is open, so it may be written as
a (possibly empty) union of members of B; points in such an intersection
are necessarily in some member of B.
S
← Put T = { U : U ⊂ B}; certainly T is closed under arbitrary
unions and ∅, X ∈ T . Fix U, V ∈ T . If x ∈ U ∩V , there exist Ux , Vx ∈ B
such that x ∈ Ux ⊂ U and x ∈ Vx ⊂ V . By assumption there exists
B ∈ B such that x ∈ B ⊂ Ux ∩ Vx , so U ∩ V ∈ T . Thus T is a topology
on X for which B is a base.
SupposeST 0 is another topology on X generated by B. If U ∈ T 0 ,
then U = U for some U ⊂ B. But this implies U ∈ T , so T 0 ⊂ T . By
the same argument T ⊂ T 0 . 

Proof of Theorem 20
Certainly every point of X is contained in some member of B2 . Take
B1 , B2 ∈ B2 with x ∈ B1 ∩ B2 . Then there exist B10 , B20 ∈ B1 such that
B10 ⊂ B1 , B20 ⊂ B2 , and x ∈ B10 ∩ B20 . Since B1 is a base, let B 0 ∈ B
such that x ∈ B 0 ⊂ B10 ∩ B20 . Then there exists B ∈ B2 such that
x ∈ B ⊂ B 0 ⊂ B10 ∩ B20 ⊂ B1 ∩ B2 . By Theorem 19, B2 is a base for a
topology T 0 on X.

106
S Observe that for each U ∈SB1 there
S exists BU ⊂ B2 such that U =
BU . So if U ⊂ B1 we have U = {BU : U ∈ U}, implying T ⊂ T 0 .
A similar argument shows T 0 ⊂ T , whence B1 and B2 are equivalent
bases. 

Proof of Theorem 21
If T1 ⊂ T2 and B ∈ B1 , then B is open in T2 and may be written as a
union of members of B2 .
Now suppose every member of B1 is a union of members S of B2 ; fix
U ∈ T1 . There is a subcollection U of B1 such that U = U, and S for
each B ∈ U there
S S is a subcollection U B of B 2 such that B = UB .
Then U = { UB : B ∈ U}, so U is open in T2 . 

Proof of Theorem 22
(a) If H is closed in A, then A \ H = U ∩ A for some U open in X.
This means (X \ U ) ∩ A = H, with X \ U closed in X.
If K is closed in X and H = A ∩ K, then (X \ K) ∩ A = A \ H is
open in A.

(b) There exists U 0 open in X such that U = U 0 ∩ A. Then U is a

finite intersection of open subsets of X, so U is open in X.

(c) There exists U 0 open in X such that A \ K = U 0 ∩ A. Certainly

X \ A is open in X, so X \ (U 0 ∪ (X \ A)) = K is closed in X. 

Proof of Theorem 23
Let U 0 be open in A ⊂ X. Then U 0 = U ∩ A for someSU open in X
0
and there exists a subcollection BS of B such that U = B 0 . It follows
that U = U ∩ A = ( B ) ∩ A = {B ∩ A : B ∈ B 0 }.
0
S 0


Proof of Theorem 24
⊂ Note clX H ∩ A is closed in A and contains H.

⊃ Let x ∈ A \ clA H. There exists U open in X such that x ∈ U ∩ A

and (U ∩ A) ∩ H = U ∩ H = ∅. This implies x ∈/ clX H. 

107
9 Proofs

Proof of Theorem 25
(a → b) We may write f −1 [V ] = {Ux : x ∈ f −1 [V ]}, where Ux is
S
an open neighborhood of x with f [Ux ] ⊂ V .

(b → c) Observe f −1 [V1 ∩ V2 ] = f −1 [V1 ] ∩ f −1 [V2 ].

(c → d) Observe f −1 [ Λ Vα ] = Λ f −1 [Uα ].
S S

(d → e) Observe f −1 [K] = X \ f −1 [Y \ K].

(e → f ) Observe that f −1 [f [A]] is closed containing A.

(f → a) Fix x ∈ X and V open in Y with f (x) ∈ V ; put A =

f −1 [Y \ V ]. Then f [A] ⊂ f [f −1 [Y \ V ]] ⊂ Y \ V , so x ∈ X \ A and
f [X \ A] ⊂ V . 

Proof of Theorem 26
(a) Given U open in X we have j −1 [U ] = {x ∈ A : x ∈ U } = A ∩ U ,
which is open in A by definition.

(b) Given V ⊂ Y we have (f  A)−1 [V ] = A ∩ f −1 [V ]. 

Proof of Theorem 27
If f : X → Y is constant then f −1 [V ] ∈ {∅, X}, so preimages (of open
sets) are open.
−1
If f : X → Y and g : Y → Z, then (g ◦ f ) [W ] = f −1 [g −1 [W ]]. 

Proof of Theorem 28 S
−1
(a) Observe f −1 [V ] = Λ (f −1 [V ] ∩ Uα ) = Λ (f  Uα ) [V ]. By The-
S
−1
orem 22(b), (f  Uα ) [V ] is open in X for each α ∈ Λ whenever V is
open in Y .

(b) Observe h−1 [A] = f −1 [A] ∪ g −1 [A]. By Theorem 22(c), f −1 [A] and
g −1 [A] are closed in X whenever A is closed in Y . 

108
Proof of Theorem 29
(a → c) If K is closed in X, then h[K] = (h−1 )−1 [K] is closed in Y .

(c → b) If U is open in X, then h[U ] = Y \ h[X \ U ] is open in Y .

(b → a) If U is open in X, then (h−1 )−1 [U ] = h[U ] is open in Y . 

Solution of Exercise 30
Let X have more than one point; let T1 and T2 be the discrete and
indiscrete topologies on X, respectively. Since T2 ⊂ T1 , the identity
map j : hX, T1 i → hX, T2 i is a continuous bijection. That j −1 is not
continuous follows from T1 6⊂ T2 . 
Proof of Theorem 31
Let V ⊂ Y be open containing f (x). Then f −1 [V ] is an open neighbor-
hood of x, so it contains a tail of hxn i. It follows that V contains the
corresponding tail of hf (xn )i, so hf (xn )i → f (x). 

Proof of Lemma 32
Put δ = ε − d(x, y) and let z ∈ B(y, δ). Then

d(z, x) 6 d(z, y) + d(y, x) < δ + d(y, x) = ε − d(x, y) + d(y, x) = ε,

so z ∈ B(x, ε). 

Proof of Lemma 33
Certainly each x ∈ X is in some member of B.
Let B1 , B2 ∈ B and suppose x ∈ B1 ∩ B2 . By Lemma 32, there
exist δ1 , δ2 > 0 such that B(x, δ1 ) ⊂ B1 and B(x, δ2 ) ⊂ B2 ; define
δ = min {δ1 , δ2 }. Then B(x, δ) ⊂ B1 ∩ B2 and B(x, δ) ∈ B, so B is a
base by Theorem 19. 
Proof of Corollary 34
By Lemma 33, any open set may be written as a union of ε-balls and
any union of ε-balls is open. 
Proof of Theorem 35
Let x, y, z ∈ Y . Certainly
(1) dY (x, y) = d(x, y) > 0 ∧ dY (x, y) = d(x, y) = 0 ↔ x = y,

109
9 Proofs

(2) dY (x, y) = d(x, y) = d(y, x) = dY (y, x), and

(3) dY (x, z) = d(x, z) 6 d(x, y) + d(y, z) = dY (x, y) + dY (y, z),
so dY is a metric on Y .
Let B1 = {d(x, ε) ∩ Y : x ∈ X, ε > 0} be a base for the subspace
topology on Y , and let B2 = {dY (y, ε) : y ∈ Y, ε > 0}. Certainly B2 ⊂
B1 . Let x ∈ X and ε > 0 such that Bd (x, ε) ∩ Y 6= ∅ and let y ∈
Bd (x, ε) ∩ Y . By Lemma 32, there exists δ > 0 such that Bd (y, δ) ⊂
Bd (x, ε). Then BdY (y, δ) = Bd (y, δ) ∩ Y ⊂ Bd (x, ε) ∩ Y , so B1 and B2
are equivalent bases by Theorem 20. By Lemma 33 B2 is a base for
the metric topology on hY, dY i, so the subspace and metric topologies
must be the same. 
Proof of Theorem
Q 36
Let X = Λ Xα ; fix β ∈ Λ and consider the projection map πβ . If
Uβ ⊂ Xβ is open then πβ−1 [Uβ ] = Uβ × Λ\{β} Xα is a basic open set
Q
of X, so πβ is continuous.
Now let U ⊂ X be open and take xβ ∈Qπβ [U ]; fix ~x ∈ U such that
πβ (~x) = xβ . There is a basic open set B = Λ Uα such that ~x ∈ B ⊂ U .
It follows that xβ ∈ πβ [B] = Uβ ⊂ πβ [U ], so πβ is open. 

Proof of Theorem 37 Q
Set Y = {~y ∈ X : ∀α 6= γ(yα = xα )} = Λ Yα , where Yα = {xα } for
α 6= γ and Yγ = Xγ . By Theorem 36, the projection map πγ : Y → Yγ
is open and continuous. Clearly πγ is one-to-one and onto, thus a
homeomorphism by Theorem 29. 
Proof of Theorem 38 Q Q
Let ~x = hxα iΛ ∈ X, A = Λ Aα , L = A, and R = Λ Aα .
If ~x ∈
/ R, there exists β ∈ Λ such that xβ ∈
/ Aβ ; let Uβ ⊂ Xβ be open
containing xβ and missing Aβ . Then πβ−1 [Uβ ] misses Q A, so ~x ∈/ L.
If ~x ∈
/ L, there is a basic open neighborhood U = Λ Uα of ~x missing
A; take β ∈ Λ such that Uβ misses Aβ . Then xβ ∈ / Aβ , so ~x ∈
/ R. 

Proof of Lemma 39
Fix x, y, z ∈ X.
(1) That d0 (x, y) > 0 and d0 (x, y) = 0 if and only if x = y is imme-
diate from the definition of d0 and the fact that d has the same

110
 properties.

(2) If d(x, y) > M then d(y, x) > M as well, so d0 (x, y) = M and

d0 (y, x) = M . If d(x, y) < M then d(y, x) < M as well, so
d0 (x, y) = d(x, y) = d(y, x) = d0 (y, x).

(3) First note that d0 (x, y) + d0 (y, z) is always the minimum of

D = {d(x, y) + d(y, z), M + d(y, z), d(x, y) + M, 2M }.

If M 6 d(x, z) 6 d(x, y) + d(y, z), then

d0 (x, z) = M 6 min D = d0 (x, y) + d0 (y, z).

If d(x, z) < M we have

d0 (x, z) = d(x, z) 6 min D = d0 (x, y) + d0 (y, z),

since d(x, z) 6 d(x, y) + d(y, z).

It follows that d0 is a metric.
Fix x ∈ X and ε > 0. With δ = min {ε/2, M }, observe

Bd (x, δ/2) ⊂ Bd0 (x, δ) = Bd (x, δ) ⊂ Bd (x, ε).

Then a union of d-balls may be written as a union of d0 -balls and vice

versa, so d and d0 must generate the same topology. 

Proof of Theorem 40
Fix ~x, ~y , ~z ∈ X.
(1) Certainly d(~x, ~y ) > 0 since dn (xn , yn ) > 0 for all n ∈ N. Also,
d(~x, ~y ) = 0 if and only if dn (xn , yn ) = 0 for all n ∈ N if and only
if xn = yn for all n ∈ N if and only if ~x = ~y .

(2) Since each dn is symmetric,

d(~x, ~y ) = max {dn (xn , yn ) : n ∈ N}

= max {dn (yn , xn ) : n ∈ N}
= d(~x, ~y ).

111
9 Proofs

(3) Each dn satisfies the triangle inequality, so

d(~x, ~z) = max {dn (xn , zn ) : n ∈ N}

6 max {dn (xn , yn ) + dn (yn , zn ) : n ∈ N}
6 max {dn (xn , yn ) : n ∈ N} + max {dn (yn , zn ) : n ∈ N}
= d(~x, ~y ) + d(~y , ~z).

It follows that
Q d is a metric.
Let B = N Un be basic open with support F ; fix ~x = hxn iN ∈
B. For each i ∈ F , let εi > 0 such that Bdi (xi , εi ) ⊂ Ui . Put ε =
min {εi : i ∈ F } and take ~y = hyn iN ∈ Bd (~x, ε). For each i ∈ F we have

di (xi , yi ) 6 d(~x, ~y ) < ε 6 εi ,

implying yi ∈ Bdi (xi , εi ) ⊂ Ui . For each i ∈ N \ F we have Ui = Xi ,

which of course contains yi . It follows that ~y ∈ B.
Now, fix ~x ∈ X and ε > 0. Let n ∈ N such that 1/2n < ε and put
n−1
Y ∞
Y
B= Bdi (xi , ε) × Xi .
i=1 i=n

Certainly ~x ∈ B; take ~y ∈ B. For 1 6 i < n we have di (xi , yi ) < ε, and

for all i > n we have di (xi , yi ) 6 1/2i 6 1/2n < ε. Thus

d(~x, ~y ) = max {di (xi , yi ) : i ∈ N} < ε,

so ~y ∈ B(~x, ε).
By Theorem 20, the topology generated by d is equivalent to the
product topology on X. 
Proof of Corollary 41
Use Lemma 39 to get the bounded metrics necessary to apply Theo-
rem 40. 
Proof of Theorem 42
Let q : X → Y be a continuous surjection. By definition q −1 [V ] is open
in X whenever V is open in Y . Take V ⊂ Y such that q −1 [V ] is open
in X.

112
 If q is open, then q[q −1 [V ]] = V is open in Y . If q is closed, then
q[X \ q −1 [V ]] = Y \ q[q −1 [V ]] = Y \ V is closed in Y . In either case, V
being open makes q a quotient map. 

Proof of Theorem 43
→ Quotient maps are continuous, so this follows from Theorem 27.

−1
← Let W ⊂ Z be open; observe (f ◦ q) [W ] = q −1 [f −1 [W ]] is open
in X. Since q is a quotient map we have f −1 [W ] open in Y . 

Proof of Theorem 44
If V is open in Y we have q −1 [q[q −1 [V ]]] = q −1 [V ], so q −1 [V ] is an
open saturated subset of X. If U ⊂ X is open and saturated then
q −1 [q[U ]] = U is open in X, so q[U ] is open in Y by definition. 

Proof of Theorem 45
→ Fix x ∈ X; each y ∈ X \ {x} has a neighborhood missing x, so
X \ {x} is open.

← Given x 6= y ∈ X, X \ {y} is open containing x but not y. 

Proof of Theorem 46

Proof of Theorem 47

Proof of Theorem 48
That T4 → T3 → T2 → T1 → T0 is straightforward.
Let X be a T31/2 -space, let x ∈ X, and let H ⊂ X be closed not
containing x. Then there exists a continuous f : X → [0, 1] such that
f (x) = 0 and f [H] = {1}. Certainly f −1 [[0, 1/3)] and f −1 [(2/3, 1]] are
disjoint open sets containing x and H, respectively. 

Proof of Theorem 49
Let hX, di be a metric space and let A and B be disjoint closed subsets

113
9 Proofs

of X. For each a ∈ A choose εa > 0 such that B(a, εa ) ∩ B = ∅.

S for each b ∈ B chooseSεb > 0 such that B(b, εb ) ∩ A = ∅. Put
Similarly,
U = a∈A B(a, εa /2) and V = b∈B B(b, εb /2); certainly U and V are
open sets containing A and B, respectively.
Suppose there exists z ∈ U ∩ V . Fix a ∈ A and b ∈ B such that
z ∈ B(a, εa /2) ∩ B(b, εb /2). Without loss of generality we may assume
εa 6 εb , whence d(a, b) 6 d(a, z) + d(z, b) < εa /2 + εb /2 6 εb . It follows
that a ∈ B(b, εb ), a contradiction. 

Proof of Theorem 50
Suppose Xα is a T0 -space for each α ∈ Λ. Let ~x 6= ~y ∈ X. Then there
exists α ∈ Λ such that xα 6= yα , so there exists some Uα open in Xα
containing one of xα , yα but not the other. It follows that πα−1 (Uα ) is
open in X and contains one of ~x, ~y but not the other. The argument
for T1 and T2 -spaces is similar.
Now suppose Xα is T3 for Qeach α ∈ Λ.
QLet ~x ∈ U , with U ⊂ X open.
There is a basic open set α∈F Uα × α∈Λ\F Xα ⊂ U containing ~x,
where F ⊂ Λ is finite and each Uα is open in T Xα . For each α ∈ F , take
Vα open Xα such that V α ⊂ Uα . Put V = α∈F πα−1 (Uα ). Certainly
~x ∈ V and
Y Y Y Y
V = Vα× Xα ⊂ Uα × Xα ⊂ U.
α∈F α∈Λ\F α∈F α∈Λ\F

Hence, X is T3 .
Finally suppose Xα is completely regular for each α ∈ Λ. Let A ⊂ X
be closed
Tn and take ~b ∈ X \ A. Then there exists a basic open set
B = i=1 παi (Uαi ) such that ~b ∈ B and B ∩A = ∅. For each 1 6 i 6 n
−1

there exists a continuous fi : Xαi → [0, 1] such that fi (bαi ) = 0 and

fi  (Xαi \ Uαi ) = 1.
For each i = 1, . . . , n, the map that sends ~x ∈ X to fi (xαi ) is con-
tinuous since fi (xαi ) = fi ◦ παi (~x). Hence if we define f : X → [0, 1]
by f (~x) = max {fi (xαi ) : 1 6 i 6 n}, then f is continuous. Note that
f (~b) = 0. Take ~a ∈ A. Then ~a ∈ / B, implying there exists some i
such that aαi ∈ / Uαi . It follows that f (aαi ) = 1, so f (~a) = 1. Thus
f  A = 1. 

114
Proof of Theorem 51
Let X be normal and let H and K be disjoint closed subsets thereof.
Let hrn : n < ωi be a well-ordering of Q ∩ [0, 1] with r0 = 1 and r1 = 0.
By normality, there exists an open set U of X such that H ⊂ U and
U ∩ K = ∅. Put Ur0 = U . There also exists V ⊂ X open such that
H ⊂ V and V ⊂ Ur0 . Put Ur1 = V .
Suppose for some n > 1 that Ur0 , . . . , Urn−1 satisfy (a) and (b). There
exist 0 6 i, j 6 n − 1 such that

ri = max {rk : 0 6 k 6 n − 1 ∧ rk < rn }

and
rj = min {rk : 0 6 k 6 n − 1 ∧ rk > rn }.
Since ri < rj we have U ri ⊂ Urj , so there exists an open set W with
U ri ⊂ W and W ⊂ Urj . Put Urn = W ; then Ur0 , . . . , Urn satisfy (a)
and (b). By induction, define appropriate Ur for all rational r ∈ (0, 1).

Proof of Theorem 52
(a → b) Let H and K be disjoint closed subsets of X and let Ur ,
r ∈ Q∩(0, 1), be defined as guaranteed by Theorem 51; put U0 = ∅ and
U1 = X. For each x ∈ X let Qx = {r ∈ Q ∩ [0, 1] : x ∈ Ur }. Certainly
0 6 inf Qx 6 1.
Define f : X → [0, 1] by x 7→ inf Qx . If x ∈ H then x ∈ Ur for all
r ∈ Q ∩ (0, 1), so f (x) = 0. If x ∈ K then x ∈ Ur ↔ r = 1, so f (x) = 1.
Let 0 < a < 1 and put U = f −1 [[0, a)] and V = f −1 [(a, 0]]. If U
and V are open then the preimage of any subbasic open set is open,
implying f is continuous. (Recall that preimages preserve unions and
intersections, and an open set may be written as a union of finite
intersections of subbasic open sets.)
Let x ∈ U ; then inf Qx < a, so there exists rx ∈ Q ∩ (0, a) such that
x ∈ USrx . But y ∈ Urx → inf Qy < a → y ∈ U , so Urx ⊂ U . Thus,
U = x∈U Urx is open.
Let x ∈ V ; then a < inf Qx so there exists rx ∈ Q ∩ (a, 1) such that
rx < inf Qx . Certainly x ∈ X \ U rx . Moreover,

y ∈ X \ U rx → rx 6 inf Qy → a < inf Qy → y ∈ V.

S
Thus, V = x∈V Urx is open.

115
9 Proofs

(b → a) Note f −1 [[0, 1/3)] and f −1 [(2/3, 1]] are disjoint open sets
containing H and K, respectively. 

Proof of Corollary 53
Apply Theorem 52, noting singletons are closed in T4 spaces. 

Proof of Theorem 54

Proof of Theorem 55
→ Let A, B ⊂ Y ⊂ X, with A and B disjoint and closed (in Y ). Note
that A ∩ B = A ∩ (Y ∩ B) = (A ∩ Y ) ∩ B = A ∩ B = ∅. Similarly
A ∩ B = ∅, so A and B are separated in X. Then there exist disjoint
open (in X) sets U and V containing A and B, respectively. It follows
that U ∩ Y and V ∩ Y are disjoint open subsets of Y containing A and
B, respectively.

← Let A and B be separated subsets of X and put Y = X \ (A ∩ B);

then A, B ⊂ Y with Y ⊂ X open. Note that
Y Y
A ∩B = (A ∩ Y ) ∩ (B ∩ Y ) = A ∩ B ∩ Y = ∅,
Y
so there exist disjoint open (in Y ) sets U and V containing A and
Y
B , respectively. But Y is open, so U and V are open in X containing
Y
A ⊂ A and B ⊂ B Y , respectively. 

Proof of Corollary 56
Just combine Theorems 35, 49 and 55. 

Proof of Lemma 57
Fix a ∈ A. Then there exists na ∈ N such that a ∈ Una . For each
1 6 i 6 na , choose Wi open in X containing a such that Wi ∩ Vi = ∅.
Put WSa = W1 ∩ · · · ∩ Wna . Construct such Wa for each a ∈ A and let
U = a∈A (Una ∩ Wa ). Certainly
T nb A ⊂ U.
0
Similarly,
S take W b = i=1 W i , where Wi0 is open and Wi0 ∩ Ui = ∅.
Put V = b∈B (Vnb ∩ Wb ). Certainly B ⊂ V .

116
 Suppose on the contrary that U ∩ V 6= ∅ and let x ∈ U ∩ V . Then
there exist a ∈ A and b ∈ B such that x ∈ Una ∩ Wi for all 1 6 i 6 na
and x ∈ Vnb ∩ Wi0 for all 1 6 i 6 nb . Without loss of generality we may
assume na 6 nb . Since x ∈ Wi0 for all i 6 nb , x ∈ Wn0 a ⊂ X \ U na . Thus
x∈/ Una , a contradiction. 

Proof of Theorem 58
Let (X, d) be a metric space. We know X is normal. Let A ⊂ X be
closed. For each n ∈ N,
Tlet Un = ∪{B(a, 1/n) : a ∈ A}. Certainly Un is
open. Moreover, A ⊂ n∈N Un .
T
Let x ∈ n∈N Un . Then for each n ∈ N there exists an ∈ A such that
limn→∞ d(an , x) = 0. So (an )∞
x ∈ B(an , 1/n). This implies thatT n=1 →
x and x ∈ A = A. It follows that n∈N Un ⊂ A, whence A is Gδ . 

Proof of Theorem 59
(c → d) Let H and K be disjoint closed subsets of X and let f, g :
X → I be continuous functions such that H = f −1 (0) and K = g −1 (0).
Since H ∩ K = ∅, f (x) + g(x) > 0 for all x ∈ X. Consider the function
h : X → I defined by

f (x)
h(x) = .
f (x) + g(x)

Certainly h is continuous. Moreover, h(x) = 0 iff f (x) = 0 iff x ∈ H, so

H = h−1 (0). Additionally, h(x) = 1 iff f (x) = f (x) + g(x) iff g(x) = 0
iff x ∈ K, so K = h−1 (1).

(d → b) Let H be a closed subset of X. If H = X, take Un = X for

each n ∈ N. Otherwise, pick x ∈ X \H and let f : X → I be continuous
such that H = f −1 (0) and {x} = f −1 (1).
For each n ∈ N, put Un = f −1 [[0, 1/n)]. Certainly each Un is open.
Moreover,
\
H = f −1 (0) = f −1 [ [0, 1/n)]
n∈N

\ \
= f −1 [[0, 1/n)] = Un
n∈N n∈N

117
9 Proofs

and U n+1 ⊂ f −1 [[0, 1/n + 1]] ⊂ Un . It follows that

\ \
Un ⊂ Un ,
n∈N n∈N

so \ \
H= Un = U n.
n∈N n∈N

(b → a) Let A and T B be disjointTclosed sets. Let W1 , W2 , . . . be open

sets such
S that A = n∈N Wn = n∈N W n . Put Vn = X \ W n . Then
B ⊂ n∈N Vn and S V n ∩ A = ∅. Similarly, there are open sets Un
such that A ⊂ n∈N Un and U n ∩ B = ∅. By Lemma 57 A and B
can be separated by disjoint open sets, so X is normal. Certainly our
assumption implies closed sets are Gδ , so X is perfectly normal.

(a →Tc) Let H ⊂ X be closed. Then there exist open sets Un such that
H = b∈N Un . By normality, for each n ∈ N there exists a continuous
function fn : X → I such that fn (x) P∞= 0 for all nx ∈ H and fn (x) = 1
for all x ∈ X \ Un . Put f (x) = n=1 fn (x)/2 . Since the range of
each fn is [0, 1], this is an absolutely convergent series of functions and
therefore continuous. Moreover f (x) = 0 iff fn (x) = 0 for all n ∈ N,
implying x ∈ H. 

Proof of Theorem 60
Let X be perfectly normal and let Y ⊂ X. Let H be closed in Y ;
Y X
then H = H = H ∩ Y . By Theorem 59(b), there are open (in X)
X T T X
sets Un , n < ω, such that H = n<ω Un = n<ω U n . Then H =
X T T X T
H ∩Y = ( n<ω Un )∩Y = ( n<ω U n )∩Y . Now H = n<ω (Un ∩Y ) ⊂
T Y T X T X X
n<ω Un ∩ Y = n<ω (Un ∩ Y ∩ Y ) ⊂ ( n<ω U n )∩Y = H ∩Y =
H, so all are equal. Thus Y is perfectly normal by Theorem 59(b). 

Proof of Theorem 61
(a) Let X be a compact space and Y a space. Let f : X → Y be
continuous and let V = {Vα : α ∈ Λ} be an open cover of f [X]. Then
{f −1 [Vα ] : α ∈ Λ} is an open cover of X by continuity. Since X is
compact there is a finite subset F of Λ such that {f −1 [Vα ] : α ∈ F }

118
covers X. Thus
[ [ [
f [X] = f [ f −1 [Vα ]] = f [f −1 [Vα ]] ⊂ Vα .
α∈F α∈F α∈F

It follows that {Vα : α ∈ F } is a finite subcover of V, so f [X] is compact.

(b) Let A be a closed subset of a compact space X; let U be an

open cover of A by open subsets of X. Then U ∪ {X \ A} is an open
cover of X, so there exists a finite subcollection {U1 , . . . , Un } of U such
that {U1 , . . . , Un } ∪ {X \ A} covers X. It follows that A is covered by
{U1 , . . . , Un }, so A is compact.
Similar proofs handle the Lindelöf cases. 

Proof of Theorem 62
Let X be Hausdorff and let Y ⊂ X be compact. Take x ∈ X \ Y . For
each y ∈ Y let Uy and Vy be disjoint open neighborhoods of x and y,
respectively.
Sn Sicne Y is compact,
Tn there exist y1 , . . . , yn ∈ Y such that
Y ⊂ i=1 Vyi . Then U = i=1 Uyi is an open neighborhood of x which
does not meet Y . So X \ Y is open, whence Y is closed. 

Proof of Exercise 63
If β < γ ∈ [0, α], then [0, β + 1) and (β, α] are disjoint open sets
containing β and γ, respectively.
Certainly [0, 0] = {0} is compact. Suppose [0, β] is compact for all
β < α. If U is an open cover of [0, α], there exists γ < α such that (γ, α]
is contained in some member of U; we need only finitely more members
of U to cover [0, γ]. 

Proof of Theorem 64
Let X be compact, Y Hausdorff, and f : X → Y a continuous bijec-
tion. Let H ⊂ X be closed. Since X is compact, H is compact. Thus
f [H] is compact, hence closed. So f is a closed map and therefore a
homeomorphism. 

Proof of Exercise 65
X/A = {{x} : x ∈ (0, 2π)} ∪ {A} with quotient map p : X → X/A
defined by x 7→ {x} iff x ∈ (0, 2π) and x 7→ A iff x ∈ A. The topology

119
9 Proofs

of X/A is
Tp = {U ⊂ X/A : p−1 [U ] is open in X}.
Define f : X/A → S 1 by

f (θ) = (cos θ, sin θ) for θ ∈ X/A \ {A}

and f [A] = (1, 0). Certainly f is well-defined.

To show injectivity let θ, ψ ∈ X/A. Then

f (θ) = f (ψ) → (cos θ, sin θ) = (cos ψ, sin ψ) → θ = ψ.

For surjectivity, let hx, yi ∈ S 1 ; then x = cos θ and y = sin θ for some
θ ∈ [0, 2π].
It remains to show f is continuous. By Theorem 43, we need only show
g = f ◦ p : [0, 2π, →]S 1 is continuous. Observe that g(x) = hcos x, sin xi
for all x ∈ [0, 2π], hence is continuous. So f is continuous too.
Finally, note that X/A is compact since it is the quotient—hence,
continuous—image of the compact space X. Since X/A is compact and
S 1 is T2 , f is a homeomorphism by Theorem 64. 
Proof of Theorem 66
Suppose there is no such x0 . Then for each x ∈ X there is εx > 0 and
yx ∈ X such that f (yx ) > f (x) + εx . The collection

{f −1 [(−∞, f (x) + εx )] : x ∈ X}

is an open cover of X, so there exist x1 , . . . , xn ∈ X such that

n
[
X⊂ f −1 [(−∞, f (xi ) + εxi )].
i=1

Let j be such that f (xj ) + εxj = max {f (xi ) + εxi : i = 1, . . . , n}. Now
each x ∈ X is in f −1 [(−∞, f (xi ) + εxi )] for some i, whence f (x) <
f (xi ) + εxi 6 f (xj ) + εxj . But f (yxj ) > f (xj ) + εxj , a contradiction.

Proof of Theorem 67
(a) Let X be compact Hausdorff. Let x ∈ X and H ⊂ X closed not
containing x. For each y ∈ H there exist disjoint open neighborhoods

120
Uy and Vy of xSand y, respectively.
n Sny1 , . . . , yn ∈ H
Tn Then there exist
such that H ⊂ i=1 Vyi . Put U = i=1 Uyi and V = i=1 Vyi . Then U
and V are disjoint open neighborhoods of x and H, respectively, whence
X is regular.

(b) Now let H and K be disjoint closed subsets of X. For each y ∈

K there exist disjoint open neighborhoods Uy and Vy of H and y,
n
respectively. Sny1 , . . . , yn ∈ K such that K ⊂ ∪i=1 Vyi .
Tn Then there exist
Put U = i=1 Uyi and V = i=1 Vyi . Then U and V are disjoint open
sets containing H and K, respectively, whence X is normal. 

Proof of Theorem 68
Let X be a regular Lindelöf space and let A and B be disjoint closed
subsets of X. For each a ∈ A there exists an open neighborhood Ua of
a such that U a ∩ B = ∅. Similarly, for each b ∈ B there exists an open
neighborhood Vb of b such that V b ∩ A = ∅.
Certainly {Ua : a ∈ A} and {Vb : b ∈ B} are open covers of A and B,
respectively. Since A and B are Lindelöf, there exist countable subcovers
{un : n ∈ N} and {Vn : n ∈ N} thereof, respectively.
S S
Now A ⊂ n∈N Un and B ⊂ n∈N Vn . Moreover, U n ∩ B = ∅ =
V n ∩ A for all n ∈ N. By Lemma 57 A and B may be separated, whence
X is normal. 
Proof of Example 69
Recall {(a, b] : a, b ∈ R ∧ a < b} is a base (of clopen sets) for S.

(a) To see that S is regular, let x ∈ S and let H ⊂ S be closed not

containing x. There is a basic open set U containing x and missing H;
U and S \ U are disjoint open sets containing x and H, respectively.
that S is Lindelöf, let U = {(aα , bα ] : α ∈ Λ} be a cover of S;
To see S
put A = {(aα , bα ) : α ∈ Λ}. Since R is hereditarily Lindelöf, there is
a countable subcollection U 0 of U which covers A. Let Λ0 ⊂ Λ such that
S \ A = {bα : α ∈ Λ0 }. Observe

∀α, β ∈ Λ0 (bα 6= bβ → (aα , bα ] ∩ (aβ , bβ ] = ∅).

Since (aα , bα ] meets Q for each α ∈ Λ0 , S \ A is countable.

121
9 Proofs

We show that S is perfectly normal; first note that it is normal by

Theorem 68. Let U be open in S and let V be the interior of U in
R. Since R is perfectly normal, V is Fσ in R and also S (observe that
the Sorgenfrey topology is finer than the Euclidean topology). By the
argument above U \ V is countable, whence U is Fσ .

(b) Note that Y = {hx, −xi : x ∈ S} is an uncountable closed dis-

crete subspace of S2 . By Theorem 61(b), S2 cannot be Lindelöf. By
Lemma 247, S2 is not normal since Q × Q is dense in S2 and 2c > c. 

Proof of Theorem 70
Let U be an open cover of X. For all x ∈ X, there exists Ux ∈ U and
εx > 0 such that B(x, εx ) ⊂ Ux . Take B = {B(x, εx /2) : x ∈ X} and let
B 0 be a finite subcover thereof. Put λ = min {εx /2 : B(x, εx /2) ∈ B 0 }.
B 0 is finite, so λ > 0.
Now let y ∈ X. We want to show that B(y, λ) is contained in some
member of U. Since B 0 is a cover, there exists x ∈ X such that y ∈
B(x, εx /2). Suppose z ∈ B(y, λ). Then d(x, z) 6 d(x, y) + d(y, z) <
εx /2 + εx /2 = εx . Thus B(y, λ) ⊂ B(x, εx ) ⊂ Ux ∈ U. 

Proof of Lemma 71
Let U be a collection of open sets in X × Y covering {x0 } × Y . Note
that {x0 } × Y is compact, being homeomorphic to Y . For each y ∈ Y ,
let Wy ∈ U such that (x0 , y) ∈ Wy .
For each y ∈ Y , choose a basis element of X × Y , say Uy × Vy ,
such that (x0 , y) ∈ Uy × Vy ⊂ Wy . There there is a finite collection
TN
{Uyi × Vyi : 1 6 i 6 N } convering {x0 } × Y . Put O = i=1 Uyi and
V = {Wyi : i = 1, . . . , N }. Then
N
[ N
[
{x0 } × Y ⊂ O × Y ⊂ (Uyi × Vyi ) ⊂ Wyi = ∪V.
i=1 i=1 

Proof of Theorem 72
Let X and Y be compact and let U be an open cover of X × Y . Let
x ∈ X. Since {x} × Y ⊂ ∪U, there is a finite subcollection Vx =
{Uix : 1 6 i 6 m(x)} of U and an open neighborhood Ox of x such

122
that Ox × Y ⊂ ∪Vx . Since X is compact, {Ox : x ∈ X} has a finite
x
subcover {Oxi : 1 6 i 6 n}. Put V = {Ui j : 1 6 i 6 m(xj ), 1 6 j 6 n}.
Certainly V is a finite subcollection of U.
Take (x, y) ∈ X × Y . Because {Oxj : 1 6 j 6 n} covers X, x ∈ Oxj
Sm(x ) x
for some j. Moreover, (x, y) ∈ Oxj × Y ⊂ i=1 j Ui j , so there exists i
x
such that (x, y) ∈ Ui j ∈ V. So V is a cover, whence X ×Y is compact.

Proof of Theorem 73
→ Let X be a space and let F be a T collection of closed sets with the
finite intersection property such that F = ∅. Then {X \ F : F ∈ F}
is an open cover of X with no finite subcover, whence X is not compact.

← Let X be a space in which every collection of closed sets with the

finite intersection property has a nonempty intersection. Let U be an
open cover of X and observe that {X \ U : U ∈ U} cannot have the
finite intersection property. There exists a finite subcollection U 0 of U
such that {X \ U : U ∈ U 0 } = ∅. It follows that U 0 is a finite subcover
T
of U. 

Proof of Theorem 74
(a → b) Let G ⊂ X such that G ∩ F 6= ∅ for all F ∈ F; put

F 0 = {F 0 ⊂ X : ∃F ∈ F(G ∩ F ⊂ F 0 )}.

It is straightforward to check that F 0 is a filter with F ⊂ F 0 and G ∈ F 0 .

Since F is an ultrafilter, we must have F = F 0 .

(b → c) Let G ⊂ X such that G ∩ F = ∅ for some F ∈ F. Then

F ⊂ X \ G, so X \ G ∈ F.

(c → a) Let F 0 be a filter on X such that F ⊂ F 0 and fix A ∈ F 0 .

By assumption, either A ∈ F or X \ A ∈ F. If X \ A ∈ F then
A ∩ (X \ A) = ∅ ∈ F 0 , a contradiction. Thus F 0 ⊂ F. 

Proof of Theorem 75
Let P be the set of all collections G ⊂ P(X) having the finite inter-
section property such that F ⊂ G; note that P is partially ordered

123
9 Proofs

S
by inclusion. Let H be a chain in hP, ⊂i and put H = H. Certainly
F ⊂ H. Moreover, H has the finite intersection property since any finite
subcollection thereof must be contained in some member of H. Thus
H ∈ P, so H is bounded above in hP, ⊂i. By Zorn’s Lemma, there is a
maximal G ∈ P.
We now show that G is a filter.
(1) Let F1 , F2 ∈ G. Since G has the finite intersection property, we
have F ∩ (F1 ∩ F2 ) 6= ∅ for all F ∈ G. Thus G ∪ {F1 ∩ F2 } has the
finite intersection property and contains G. By the maximality
of G we must have F1 ∩ F2 ∈ G.

(2) Let F ∈ G and G ⊂ X scuh that F ⊂ G ⊂ X. For any F 0 ∈ G we

have ∅ 6= F 0 ∩ F ⊂ F 0 ∩ G, so G ∪ {G} has the finite intersection
property and contains G. By the maximality of G we must have
G ∈ G.

(3) That ∅ ∈
/ G follows from G having the finite intersection property.
To see that G is an ultrafilter, let F ⊂ X such that F ∩ G =
6 ∅ for all
G ∈ G. Then G ∪ {F } has the finite intersection property and contains
G, so F ∈ G. By Theorem 74(b), G is an ultrafilter. 

Proof of Theorem 76
(a → b) Let F be a filter on X and put G = {F : F ∈ F}. Certainly
G has
T the finite intersection property; by Theorem 73, there exists
x ∈ G. Then x ∈ F for all F ∈ F, so F clusters at x.

(b → c) Let F be an ultrafilter on X and let x be a cluster point of

F. If U is a neighborhood of x then U ∩ F = ∅ for all F ∈ F, so U ∈ F
by Theorem 74(b). It follows that every neighborhood of x belongs to
F, so F converges to x.

(c → a) Suppose X is not compact and let {Uα : α ∈ Λ} be an open

cover of X with no finite subcover. Then {X \ Uα : α ∈ Λ} has the finite
intersection property and is contained in an ultrafilter F by Theorem 75.
Fix x ∈ X; there exists α ∈ Λ such that x ∈ Uα . But X \ Uα ∈ F
implies Uα ∈/ F, so F does not converge to x. 

124
Proof of Theorem 77 Q
Let Xα be compact for all α ∈ Λ. Put X = α∈Λ Xα and let F be an
ultrafilter on X. We want to show that Fα = {πα [F ] : F ∈ F} is an
ultrafilter on Xα for all α ∈ Λ.
For convenience fix α ∈ Λ, let π = πα , and put G = Fα .
(1) Let G1 , G2 ∈ G; there exist F1 , F2 ∈ F such that G1 = π[F1 ] and
G2 = π[F2 ]. Thus G1 ∩ G2 = π[F1 ] ∩ π[F2 ] ⊃ π[F1 ∩ F2 ], so F1 ∩
F2 ⊂ π −1 [G1 ∩ G2 ]. Since F1 ∩ F2 ∈ F, we have π −1 [G1 ∩ G2 ] ∈
F. It follows that G1 ∩ G2 = π[π −1 [G1 ∩ G2 ]] ∈ G.

(2) Let G ∈ G and G0 ⊂ Xα such that G ⊂ G0 ; there exists F ∈ F

such that G = π[F ]. Let F 0 = π −1 [G0 ] and note F ⊂ F 0 (so
F 0 ∈ F). It follows that G0 = π[F 0 ] ∈ G.

(3) That ∅ ∈
/ G follows from ∅ ∈
/ F.
To see that G is an ultrafilter, let G ⊂ Xα ; by Theorem 74(c) either
F = π −1 [G] ∈ F or X \ F ∈ F. It follows that G = π[F ] ∈ G or
Xα \ G = π[X \ F ] ∈ G.
For each α ∈ Λ, there exists xα ∈ Xα such that Fα converges to
Q . We want to show that F converges to ~x = hxα : α ∈ Λi. Let U =
x α
α∈Λ Uα be a basic open neighborhood of ~ x; let A ⊂ Λ be the set of
α ∈ Λ with Uα 6= Xα . For all α ∈ A, Fα converges to xα so Uα ∈ Fα .
Hence, for all α ∈ A there is some
T Fα ∈ F such Tthat πα [Fα ] = Uα . This
implies πα−1 [Fα ] ⊃ Fα , so U ⊃ α∈A Fα . But α∈A Fα ∈ F, we have
U ∈ F. 
Proof of Theorem 78
→ Any open cover (countable or not) has a finite subcover, hence a
countable subcover.

← Any open cover has a countable subcover, which in turn has a finite
subcover. 

Proof of Theorem 79
(a) Let A be a countably infinite subset of X countably compact.
Suppose on the contrary that A has not limit in X. For each a ∈ A
pick Ua open such that Ua ∩ A = {a}. Since A is closed {Ua : a ∈ A} ∪

125
9 Proofs

{X \ A} is a countable open cover of X with no finite subcover, a

contradiction.

(b) → This is 79(a).

← Let X be a T1 -space which is not countably compact. Let
{Un : n < ω} be a countable open cover with no finite subcover.
Without loss of generality we may assume thatSUn ( Un+1 for
each n < ω. For eacn n < ω pick xn ∈ Un \ i<n Ui and put
A = hxn : n < ωi. Take x ∈ X and n < ω such that x ∈ Un .
Observe that Un ∩ A ⊂ {x0 , . . . , xn }. Since singletons are closed,
(Un \ {x0 , . . . , xn }) ∪ {x} is an open neighborhood of x containing
no point of A other than perhaps x itself. It follows that A is an
infinite subset of X without a limit point. 

Proof of Theorem 80
→ Let hBn : n < ωi be a decreasing local base at x and for each n < ω
pick an ∈ A ∩ Bn ; certainly han i → x.

← If there is a sequence han i in A converging to x then every neigh-

borhood of x meets A, whence x ∈ A. 

Proof of Theorem 81
→ This is Theorem 31.

← Let A ⊂ X and x ∈ A; by Theorem 80 there is a sequence han i in

A converging to x. By assumption hf (xn )i → f (x), so f (x) ∈ f [A]. By
Theorem 25(f), f is continuous. 

Proof of Theorem 82
Let hX, di be a metric space and fix x ∈ X. Set Bx = {B(x, 1/n) : n ∈ N}
and let U ⊂ X be open containing x. By Corollary 34, there exists
ε > 0 such that B(x, ε) ⊂ U . Pick n ∈ N such that 1/n < ε. Then
B(x, 1/n) ⊂ B(x, ε) ⊂ U , whence Bx is a local base at x. 

Proof of Exercise 83
Note X/A = {A} ∪ {(i, y) : i ∈ N ∧ 0 < y 6 1}. Points of the form hi, yi

126
for y > 0 have their usual neighborhoods in the interval {i} × (0, 1],
and a basic open neighborhood of A has the form
[
{A} ∪ ({i} × (0, εi )),
i∈N

where 0 < εi 6 1 for all i.

We wish to show that X/A is not first-countable. Indeed, suppose
{Ui : i ∈ N} is a countable base at A. If q is the quotient map, then
q −1 [Ui ] contains (i, 0) and is open in X, so for all i there is some yi > 0
such that xi = (i, yi ) ∈ Ui . Then X \ {xi : i ∈ N} is a saturated open
set in X, hence its image is an open neighborhood of A which does not
contain any xi , hence not any Ui , so we get a contradiction.
Since X/A is not first-countable, it cannot be metrizable. 
Proof of Theorem 84
Let X be second-countable and let B = {Bn : n < ω} be a countable
base for X.
(a) Note that if Y ⊂ X then {Bn ∩ Y : n < ω} is a countable base for
the subspace topology on Y by Theorem 23.

(b) It suffices to show that X is separable. For each n < ω pick xn ∈ Bn

and let D = {xn : n < ω}. Let U ⊂ X be a nonempty open set. There
exists n < ω such that Bn ⊂ U , whence xn ⊂ U .

(c) It suffices to show X is Lindelöf. Let U be an open cover of X. For

each n < ω, if there is U ∈ U with Bn ⊂ U pick one and call it Vn . If
there is no such U , leave Vn undefined. Let V be the collection of those
Vn which are defined. Let x ∈ X; then x ∈ U for some U ∈ U. Since B
is a base, there exists Bn ∈ B with x ∈ Bn ⊂ U . It follows that Vn is
defined and x ∈ Bn ⊂ Vn . 

Proof of Theorem 85
Let hX, di be a metric space.
(a → d) By Theorem 84.

(d → c) By definition.

127
9 Proofs

(c → a) Let D be a countable dense subset of X. Put

B = {B(x, 1/n) : x ∈ D ∧ n ∈ N};
certainly B is countable. Let U ⊂ X be open and x ∈ U . Then there
exists ε > 0 such that B(x, ε) ⊂ U . Pick n ∈ N such that 2/n < ε and
let y ∈ B(x, 1/n) ∩ D. Certainly x ∈ B(y, 1/n). Take z ∈ B(y, 1/n).
Then
d(x, z) 6 d(x, y) + d(y, z) < 1/n + 1/n = 2/n < ε,
so z ∈ B(x, ε). It follows that x ∈ B(y, 1/n) ⊂ B(x, ε) ⊂ U , so B is a
base.

(a → e) By Theorem 84.

(e → b) By definition.

(b → a) Let Vn be aScountable subcover of {B(x, 1/n) : x ∈ X} for

each n ∈ N. Put B = n∈N Vn ; certainly B is countable. Let W ⊂ X
be open and x ∈ W . There exists ε > 0 such that B(x, ε) ⊂ W .
Choose n ∈ N such that 2/n < ε. Then there exists y ∈ X such that
x ∈ B(y, 1/n) ∈ Vn . Take z ∈ B(y, 1/n). Then
d(x, z) 6 d(x, y) + d(y, z) < 1/n + 1/n = 2/n < ε,
so z ∈ B(x, ε). It follows that x ∈ B(y, 1/n) ⊂ B(x, ε) ⊂ W , so B is a
base. 

Proof of Theorem
Q 86
Let X = n∈N Xn , where each Xn is first-countable. Take ~x ∈ X;
~x = (xn )n∈N . For each n ∈ N, let Bn be a countable local base at xn .
Put
Ym Y∞
B = { Bi × Xi : Bi ∈ Bi , m ∈ N}.
i=1 i=m+1
Certainly BQis countable.
Q∞
m
Let U = i=1 Ui × i=m+1 Xi be a basic open neighborhood of ~x.
~ m
Q
For
Q∞ each i 6 m, take B i ∈ B i such that B i ⊂ Ui . Then ∈ i=1 i ×
B
X
i=m+1 i ⊂ U , so B is a local base at ~
x .
The proof for each Xn second-countable is similar. 

128
Proof of Example 87
Let {Un : n < ω} be collection of open neighborhoods of ~0; we show
this cannot be a local base at ~0. For all n Q
< ω, there exists a basic open
set On such that ~0 ∈ On ⊂ Un ; say On = {On,α : α ∈ Λ}. Then On,α
is open in 2 and On,α = 2 for all but finitely many α.
Note there exists β ∈ Λ such that On,β = 2 for all n < ω; put
V = πβ−1 (0). Let ~x ∈ 2Λ such that xβ = 1 and xα = 0 for all α 6= β.
Then ~x ∈ On ⊂ Un for each n < ω but ~x ∈ / V . Since Un 6⊂ V for all
n < ω, there is no countable local base at ~0. 
Proof of Q
Theorem 88
Let X = n∈N Xn , with each Xn separable. For each n ∈ N let Dn be
a countable dense subset of Xn . Fix ~x = (xn )n∈N ∈ X and put
<ω
D = {(dn )n∈F × (xn )n∈F
/ : F ∈ [N] , dn ∈ Dn }.

Certainly DQis countable.

Q
Let U = i∈A Ui × i∈A / Xi be a basic open set, so A is finite. Pick
di ∈ Di ∩ Ui for each i ∈ A. Put ~y = (yi )i∈A × (xi )i∈A y ∈ D ∩U,
/ . Then ~
so D is dense. 
Proof of Theorem 89
(a) Let A be an infinite subset of X compact. Suppose on the contrary
that A has no limit point in X. Then for each x ∈ X there exists an
open neighborhood Ux of x such that Ux ∩ (A \ {x}) = ∅. Note that
U = {Ux : x ∈ X} is an open cover of X and let V be a finite subcover
thereof. But each V ∈ V contains at most one point of A implying A
is finite, a contradiction.

(b) Let A be an uncountable subset of a Lindelöf space X. Suppose A

has no complete accumulation point in X. Then for each x ∈ X there
exists an open neighborhood Ux of x such that Ux ∩A is countable. Note
that U = {Ux : x ∈ X} is an open cover of
S X and let V be a countable
subcover thereof. But this implies A = V ∈V (V ∩ A) is countable, a
contradiction. 

Proof of Corollary 90
Recall that separable metric spaces are hereditarily Lindelöf and apply

129
9 Proofs

Theorem 89(b). 

Proof of Theorem 91
(a) Let X be a space which is not countably compact. Let {Un : n < ω}
be a countable open cover of X with no finite subcover; without loss
of generality we may assume that Un ( Un+1 for each n < ω. Take a
sequence hxn in<ω in X with xn ∈ Un+1 \ Un . Let x ∈ X and pick k < ω
such that x ∈ Uk . Then Uk is an open neighborhood of x containing
only finitely many terms in the sequence. It follows that hxn i has no
infinite convergent subsequence.

(b) → Let X be first-countable and countably compact. Let hxn in<ω

be an infinite sequence in X and suppose on the contrary there
is no x ∈ X whose every neighborhood contains infinitely many
of the xn . Put An = {xm : m > n}; fix x ∈ X. Then there is
a nieghborhood U of x and n < ω such that {k : xk ∈ U } ⊂ n,
implying x ∈ / An . Thus {X \ An : n < ω} is a countable open cover
of X with no finite subcover, a contradiction.
So, let x ∈ X such that every neighborhood of x contains infi-
nite many terms of the sequence. Let hVn : n < ωi be a countable
decreasing local base at x. Pick a subsequence hxni i such that
xni ∈ Vi for all i < ω and i < j → ni < nj . Certainly hxni i → x.
← This is 91(a). 

Proof of Theorem 92
Let hX, di be a metric space, not necessarily separable.

(a → c) By Theorems 91(b), 78 and 82.

(c → b) By Theorem 91(a).

(b → a) We claim that {B(x, 1/n) : x ∈ X} has a finite subcover for

each n ∈ N. Suppose not. Take x0 ∈ X. Then B(x0 , 1/n) 6= X, so
pickSx1 ∈ X \ B(x0 , 1/n). If x0 , . . . , xk have been chosen, pick xk+1 ∈
X \ i6k B(xi , 1/n). Put A = hxn : n < ωi; for each p < q < ω we have
d(xp , xq ) > 1/n. Thus A has no limit point, a contradiction.

130
 For each n ∈ N let Fn ⊂ X be finite such that {B(x, 1/n) : x ∈ Fn }
covers X. Note that for each x ∈ X S there exists y ∈ Fn such that
d(x, y) < 1/n. It follows that D = n∈N Fn is a countable dense
subset of X. Thus X is separable, hence Lindelöf by Theorem 85. By
Theorem 78, X is compact. 

Proof of Exercise 93
Let hαn i be a sequence in ω1 ; by Corollary 6 there exists α < ω1
such that αn ∈ [0, α] for all n < ω. Each β ∈ [0, α] has countably
many predecessors, so [0, α] must be first-countable. By Theorem 52(b),
[0, a] is sequentially compact. Thus hαn i has an infinite convergent
subsequence in [0, α], hence in ω1 .
To see that ω1 is not compact, consider {[0, α] : α < ω1 }. 

Proof of Theorem 94
Let X ∗ = X ∪ {∞} and B = T ∪ {X ∗ \ K : K ⊂ X ∧ K compact}; let
T ∗ be the topology generated by B.
Let U be an open cover of X ∗ . Then there exists U∞ ∈ U such
that ∞ ∈ U∞ , which implies that X ∗ \ U∞ is compact. So there exists
V = {U1 , . . . , Un } ⊂ U which covers X ∗ \ U∞ . Hence V ∪ {U∞ } is a
finite subcover of U.
Let x ∈ X. Since X is locally compact, there exists a compact neigh-
borhood Ux of x. It follows that X \ Ux is an open neighborhood of
∞. Then Ux◦ and X \ Ux are disjoint open sets containing x and ∞,
respectively. Also, if x, y ∈ X, these two points can be separated by
disjoint open sets since X is Hausdorff. Thus X ∗ is Hausdorff. 

Proof of Theorem 95
Claim A. Every subspace of a completely regular space is completely
regular.

Proof of Claim A. Let X be a completely regular space and Y ⊂ X. Let

C be closed in Y and y ∈ Y \ C. Then y ∈ / C = clY C = clX C ∩ Y ,
so y ∈
/ clX C. Since X is completely regular, there exists f : X → [0, 1]
continuous such that f (y) = 0 and f [clX C] = {1}. Define g : Y → [0, 1]
by y 7→ (f  Y )(y). Then g is continuous, g(y) = 0, and g[C] = {1}.
Thus Y is completely regular. 

131
9 Proofs

Since compact Hausdorff spaces are completely regular, this follows

from the one-point compactification of a locally compact Hausdorff
space and Proof A. 

Proof of Lemma 96
Let x ∈ U with U open in X. We need to show there is a compact
neighborhood K of x contained in U .
Since X is locally compact, there is a compact neighborhood N of
x. Then V = N ◦ ∩ U is open in X and contains x. Also, N \ V is
closed in N , hence in X. By Theorem 95, X is completely regular. Let
f : X → [0, 1] be continuous such that f (x) = 0 and f [N \ V ] = {1}.
Let K = f −1 [[0, 1/2]] ∩ N . Then K is closed in N , hence compact. Also,
x ∈ f −1 [[0, 1/2]] ∩ V which is open in X and, since V ⊂ N , is containd
in K. So K is a compact neighborhood of x. It remains to prove that
K ⊂ U . Let y ∈ K. Then y ∈ N . Since f (y) ∈ [0, 1/2], y ∈ / N \ V . But
y ∈ N , so y ∈ V ⊂ U . 

Proof of Lemma
T 97
Suppose that n∈N Hn is empty. Then {H1 \ Hn : n ∈ N} is a (count-
able) open cover of H1 . Let {H1 \ Hn : n ∈ F }, with F ⊂ N finite, be
a subcover thereof. If k = max F , we have H1 \ Hk = H1 . But this
implies Hk = ∅, a contradiction. 

Proof of Theorem
T 98
Put D = n∈N Gn and let U ⊂ X be nonempty and open. Since G1
is dense, pick x1 ∈ G1 ∩ U . Let H1 be a compact neighborhood of x1
contained in G1 ∩ U . Since G2 is dense, pick x2 ∈ G2 ∩ H1◦ . Let H2
be a compact neighborhood of x2 contained in G2 ∩ H1◦ . Continuing
in this way, construct for each n ∈ N a nonempty
T compact Hn such
◦
T1 ⊂ U and Hn ⊂ Gn ∩ Hn−1 . Put K = n∈N Hn . Then K ⊂
that H
U ∩ ( n∈N Gn ) = U ∩ D. By Theorem 98, K 6= ∅ so U ∩ D 6= ∅. 

Proof of Lemma 99
Let X be a Baire space and let O ⊂ X be open. Let {Un : n ∈ N} be
O
a collection of open sets in O with U n = O for each n ∈ N. We show
T O
that n∈N Un = O. Since O is open in X, Un is open in X for each
◦
n ∈ N. For each n ∈ N, set Vn = Un ∪ X \ O . Thus Vn is open in X

132
for every n ∈ N.
We claim each Vn is dense in X. Fix n ∈ N and let V be a nonempty
open set in X. If V ∩ O =
6 ∅, then V ∩ Un 6= ∅ since Un is dense in O. If
◦
V ∩O = ∅, then V ⊂ X \O, so V ⊂ X \ O implying V ⊂TVn . Therefore
V ∩ Vn 6= ∅, so Vn is dense. Since X is Baire, we have Tn∈N Vn = X.
Let V ⊂ O be a nonempty open set. T We have V ∩T( n∈N Vn ) 6= ∅.
◦
But V ∩ X \ O = ∅ so V must meet n∈N Un . Thus n∈N Un is dense
in O, so O is Baire. 
Proof of Theorem 100 S
Let X be a Baire space with X = n∈N Xn . Suppose on the contrary
◦
T all n ∈ N. Then X \ X n is dense and open in X for
that X n = ∅ for
each n ∈ N, so n∈N X \ X n is dense (and nonempty) in X. But
\ \ [
X \ Xn ⊂ X \ Xn = X \ Xn = ∅,
n∈N n∈N n∈N

a contradiction. S
Now let U be open in X, so U = n∈N (U ∩ Xn ). By Lemma 99, U
is Baire. By the previous argument, there exists n ∈ N such that
U ◦
∅ 6= intU (U ∩ Xn ) = intU (X n ∩ U ) ⊂ X n ∩ U.
S ◦
Thus, every open set meets n∈N X n . 

Proof of Theorem 101

(a → b) Suppose C = A ∪ B, where A and B are nonempty separated
sets in X. Put U = X \ clX B and V = X \ clX A. Then C ∩ U = A
and C ∩ V = B are disjoint open subsets of C, so C is not connected.

(b → a) Now suppose C is not connected. Then C = A ∪ B, where

A and B are disjoint nonempty open subsets of C. Let U and V be
open in X such that A = U ∩ C and B = V ∩ C. Since A ⊂ X \ V ,
clX A ∩ B = ∅. Similarly, clX B ∩ A = ∅. Thus C is the union of two
nonempty separated sets in X. 

Proof of Theorem 102

Let A be connected and suppose on the contrary that B is not connected.
Then B = H ∪ K, where H and K are nonempty separated sets in X.

133
9 Proofs

Furthermore, H ∩ A and K ∩ A are separated sets whose union is A.

Since A is connected, we may assume without loss of generality that
K ∩ A = ∅. Then A ⊂ H so B ⊂ A ⊂ H, contradicting the fact that
K ∩ H = ∅. 
Proof of Theorem 103
→ Suppose X is not convex; there exist a < b in X such that (a, b) 6⊂ X.
Fix x ∈ (a, b) such that x ∈
/ X. Then we may write X = A ∪ B, where
A = X ∩ (−∞, x) and B = (x, ∞) are disjoint nonempty open sets in
X.

← Suppose X is not connected; then X = A ∪ B with A and B

nonempty separated sets in R. Without loss of generality, there exist
a ∈ A and b ∈ B such that a < b. Let x = sup(A ∩ [a, b]). Note x ∈ A,
so x ∈
/ B.
If x ∈
/ A, then x ∈
/ X. If x ∈ A, then x ∈
/ B so there exists y ∈ (x, b)
with y ∈
/ B. But (x, b) misses A, so y ∈
/ X. 

Proof of Theorem 104

Let f : X → Y be a continuous surjection. Suppose Y = U ∪ V , where
U and V are nonempty disjoint open sets. Then X = f −1 (U ) ∪ f −1 (V ),
so X is not connected.
Now suppose X is path connected and take x, y ∈ Y . Choose a ∈
f −1 (p) and b ∈ f −1 (q). Let g be a path in X from a to b. Then f ◦ g
is a path in Y from x to y, whence Y is path connected. 

Proof of Theorem 105

Let {Aα : α ∈ Λ} be a collection of connected subspaces of a space X
such that Aα ∩ Aβ 6= ∅ for any α, β ∈ Λ. Put A = ∪{Aα : α ∈ Λ}
and suppose on the contrary that A is not connected. Then there exist
nonempty separated subsets U and V of X such that A = U ∪ V . For
each α ∈ Λ, put Uα = U ∩ Aα and Vα = V ∩ Aα . Then Aα = Uα ∪ Vα
for each α ∈ Λ with Uα and Vα separated. But each Aα is connected,
so either Uα = ∅ or Vα = ∅. Without loss of generality we may assume
Uα = ∅ for some α ∈ Λ, so Aα = Vα . For each β ∈ Λ we have

Aα ∩ Aβ = Vα ∩ Aβ ⊂ V ∩ Aβ = Vβ ,

134
so Vβ 6= ∅. It follows that Uβ = ∅ for each β ∈ Λ so U = ∅, a
contradiction.
Now let {Aα : α ∈ Λ} be a collection of path connected subspaces
of a space X such that Aα ∩ Aβ 6= ∅ for any α, β ∈ Λ. Put A =
∪{Aα : α ∈ Λ} and take x, y ∈ A. Choose α, β ∈ Λ such that x ∈ Aα
and y ∈ Aβ , and pick z ∈ Aα ∩ Aβ . Let fα : I → Aα be a path from x
to z and fβ : I → Aβ a path from z to y. Then f : I → A defined by
(
fα (2t) if 0 6 t 6 1/2
f (t) =
fβ (2t − 1) if 1/2 6 t 6 1

is a path from x to y. 

Proof of Theorem 106

(a) For any x ∈ X and y ∈ Y we know {x} × Y and X × {y} are
connected. For each x ∈ X and y ∈ Y , let P (x, y) = ({x} × Y ) ∪ (X ×
{y}). Since ({x} × Y ) ∩ (X × {y}) is nonempty, P (x, y) is connected
for each (x, y).
Note that ∪{P (x, y) : (x, y) ∈ X × Y } = X × Y . If x, x0 ∈ X and
y, y 0 ∈ Y , then P (x, y) ∩ P (x0 , y 0 ) ⊃ {(x0 , y)}. Thus X × Y is connected
by Theorem 105.
Q
(b) Let X = α∈Λ Xα , with each Xα connected. Suppose on the
contrary that X is not connected. Then there exist disjoint nonempty
open
Q sets UQand V such that X = U ∪ V . Take ~x ∈ U . Let B =
α∈F Uα × / Xα be a Q
α∈F Qwith ~x ∈ B ⊂ U .
basic open set
Fix ~y ∈ X. Then C = α∈F Xα × α∈F / {yα } is connected and
C ∩ B 6= ∅. Moreover U ∩ C 6= ∅, so y ∈ C ⊂ U . But this implies
V = ∅, a contradiction. 

Proof of Theorem 107

Let X be a path connected space. Suppose on the contrary that X
is not connected. Then there are disjoint nonempty open sets U and
V such that X = U ∪ V . Take a ∈ U and b ∈ V and let f : I → X
be a path from a to b. Put K = f (I). Then U ∩ K and V ∩ K are
disjoint nonempty open subsets of K whose union is K, whence K is

135
9 Proofs

not connected. But K is the image of a connected space, so we have a

contradiction. 
Proof of Theorem 108
→ If f is continuous, then πα ◦ f is continuous as a composition of
continuous functions.

←
Q Now suppose
Q that πα ◦ f is continuous for all α ∈ Λ. Let U =
α∈F Uα × / Xα be a basic open set, so F is finite. Then
α∈F

−1
\ \
f −1 [U ] = f −1 [ πα−1 [Uα ]] = (πα ◦ f ) [Uα ]
α∈F α∈F

is open as a finite intersection of open sets. 

Proof of QTheorem 109

Let X = α∈Λ Xα , with each Xα path connected, and take ~a, ~b ∈ X.
For each α ∈ Λ, let fα : I → Xα be a path from aα to bα . Define f : I →
X by f (t) = (fα (t))α ∈ Λ. Then f is continuous because πα ◦ f = fα
is continuous for each α ∈ Λ. Moreover f (0) = (fα (0))α∈Λ = ~a and
f (1) = (fα (1))α∈Λ = ~b, so f is a path from ~a to ~b. 

Proof of Lemma 110

Let f : X → Y be continuous and let G = {hx, f (x)i : x ∈ X}. Define
h : G → X by (x, f (x)) 7→ x. Certainly h is a bijection. Furthermore
h = π1  G, so h is continuous. Note that h−1 : X → G ⊂ X × Y by
x 7→ (x, f (x)), so (π1 ◦ h−1 ) = idX and (π2 ◦ h−1 ) = f are continuous.
It follows that h−1 is continuous. 
Proof of Example 111
Let G = {(x, sin 1/x) : 0 < x 6 1}. Then G ∼= (0, 1], so G is connected.
It follows that the topologist’s sine curve G = ({0} × [−1, 1]) ∪ G is
connected.
Now suppose on the contrary that S is path connected. Let f : I →
S be a path from (0, 0) to (1/π, 0). Then f (t) = (x(t), y(t)), where
x = π1 ◦ f and y = π2 ◦ f are continuous. Note that x[I] must be a
connected, hence convex, subset of R. Since x(0) = 0 and x(1) = 1/π,
there exists 0 < t1 < 1 such that x(t1 ) = 2/3π. Similarly there exists

136
0 < t2 < t1 such that x(t2 ) = 2/5π. Continuing this way, we define
a sequence (tn )n∈N satisfying 0 < · · · < tn < · · · < t1 < 1 such that
x(tn ) = 2/(2n + 1)π. Note y(tn ) = (−1)n .
Since the sequence (tn )n∈N is decreasing and bounded below we
have tn → t0 for some t0 . Since y is continuous, y(tn ) → y(t0 ). But
limn→∞ (−1)n does not exist so y(tn ) cannot converge, a contradic-
tion. 

Proof of Theorem 112

(a) Trivial.

(b) Certainly, for if two (path) components met their union would be
a larger (path) connected set.

(c) The closure of a connected set is connected. 

Proof of Lemma 113

Let X be a locally (path) connected space and let C be a (path) compo-
nent of X. For each x ∈ C let Nx be a (path) connected neighborhood
thereof. Certainly each Nx ⊂ C, whence C is open. 

Proof of Theorem 114

Let X be locally (path) connected and let B be the collection of all
open and (path) connected subsets of X. Let W ⊂ X be open and take
x ∈ W . Let CW (x) be the (path) component of x in W . We want to
see that CW (x) is open in W ; by Lemma 113 it suffices to show that
W is locally (path) connected.
Let y ∈ W and U ⊂ W open containing y. As W is open in X, U
is also open in X. Since X is locally (path) connected, there exists
Ny ⊂ U (path) connected such that y ∈ Ny◦ . Thus, W is locally (path)
connected.
Now suppose X has a base consisting of (path) connected subsets. If
W ⊂ X is open and x ∈ X, there exists a (path) connected basic open
set B such that x ∈ B ⊂ W . It follows immediately that X is locally
(path) connected. 

137
9 Proofs

Proof of Theorem 115

Suppose X is locally path connected and fix x ∈ X. Let C(x) be the
component of x in X and P (x) its path component. Then P (x) ⊂ C(x)
since P (x) is connected and contains x. P (x) is open by Lemma 113.
Likewise, X \ P (x) is open as a union of path components. It follows
that P (x) is a nonempty clopen subset of C(x), whence P (x) = C(x).

Proof of Theorem 116

If H is clopen and H ∩ C(x) 6= ∅, then C(x) ⊂ H since X = H ∪ X \ H
and C(x) is connected. Thus C(x) ⊂ Q(x). 

Proof of Lemma 117

Let U ⊂ S X be open containing H. Since H ⊂ U , we have X \ U ⊂
X \ H = K∈K X \ K. Being closed, X \ U is compact so there exists a
finite subcollection K0 of K such that X \ U ⊂ K∈K0 X \ K. It follows
S
that H = ∩K ⊂ ∩K0 ⊂ U . 
Proof of Theorem 118
Fix x ∈ X; by Theorems 112(a) and 116, it suffices to show Q(x) is
connected. Suppose on the contrary that Q(x) is not connected. Let A
and B be disjoint nonempty clopen subsets of Q(x) with Q(x) = A ∪ B.
As Q(x) is closed in X, both A and B are closed in X. Since X is
normal, being compact Hausdorff, there exist disjoint open sets U and
V in X containing A and B, respectively. By Lemma 117, there exist
clopen (in X) sets F1 , . . . , Fn such that Q(x) ⊂ F1 ∩ · · · ∩ Fn ⊂ U ∪ V ;
put F = F1 ∩ · · · ∩ Fn , U 0 = U ∩ F , and V 0 = V ∩ F . Since F is clopen
in X and F = U 0 ∪ V 0 , we have that U 0 and V 0 are clopen in X.
Without loss of generality, suppose x ∈ A. Then x ∈ U 0 , so Q(x) ⊂ U 0 .
Thus Q(x) ∩ B = ∅, a contradiction. 

Proof of Theorem 119

Let hKi be a decreasing sequence ofT compact, connected subsets of a
Hausdorff space X and put K = n<ω Kn . Certainly K is a closed
subset of K0 , so K is compact.
Suppose on the contrary that K is not connected; let A and B be
disjoint nonempty closed subsets of K such that K = A ∪ B. Note
that A and B are closed in K0 . Since K0 is normal, being compact
Hausdorff, there exist disjoint open sets U and V (in K0 ) containing

138
A and B, respectively. By Lemma 117 there exists n < ω such that
Kn ⊂ U ∪ V . We may write Kn = U 0 ∪ V 0 , where U 0 = U ∩ Kn and
V 0 ∩ Kn are disjoint nonempty open (in Kn ) sets. Thus Kn is not
connected, a contradiction. 
Proof of Theorem 120
Suppose on the contrary there exists h ∈ H such that C(h) ∩ ∂H = ∅.
Note that ∂H is closed, and by Theorem 118 we have C(h) = Q(h). By
Lemma 117 there is a finite collection, whose intersection we denote by
F , of clopen subsets of H such that
C(h) = Q(h) ⊂ F ⊂ H \ ∂H = H ◦ .
Observe that F is a nonempty clopen subset of X, being a closed subset
of H and an open subset of H ◦ . Thus X = F ⊂ H, a contradiction. 
Proof of Lemma 121
Fix k < ω. If Kk = ∅ we may simply take Ck = X, so assume Kk is
nonempty. Choose j < ω such that j 6= i and Kj is nonempty. Since
X is normal, being compact Hausdorff, there exist disjoint open sets
U and V containing Kj and Kk , respectively. Let x ∈ Kj and let Ck
be the component of x in U . Note that Ck is compact and connected,
Ck ∩Kk = ∅, and Ck ∩Kk 6= ∅. By Theorem 120, we have Ck ∩∂U 6= ∅;
let y ∈ Ck ∩ ∂U . Since K covers X, there exists ` < ω with y ∈ K` .
Now ` 6= j since Kj ⊂ U and ` 6= k since U ∩ V = ∅. 
Proof of Theorem 122
Suppose on the contrary that K is a partition of X into closed subsets
with 1 < |K| < ℵ1 . Immediately we must have |K| = ℵ0 , for if K is finite
we may write X as a union of disjoint clopen subsets. Let {Kn : n < ω}
be an enumeration of K without repetition.
Applying Lemma 121 with k = 0, we get a compact connected sub-
space C0 of X missing K0 and meeting at least two members of K.
Put K0 = {C0 ∩ Kn : n < ω}. By Lemma 121 there exists a compact
connected subspace C1 of C0 which misses K1 and meets at least two
members of K0 . Continuing in this way, for each n < ω we get a com-
pact connected Cn T missing Kn such that Cn ⊃ Cn+1 . By Lemma 117,
we have that C = n<ω Cn is nonempty. But C ∩ Kn = ∅ for each
n < ω, a contradiction. 

139
9 Proofs

Proof of Lemma 123

Let U = {Uα : α ∈ Λ}, and for each x ∈ X let Vx be an open neighbor-
hood of x such that Fx = {α ∈ Λ : V ∩ Uα 6= ∅} is finite.

(a) If I ⊂ Λ and x ∈ {Uα : α ∈ I}, then

[ [ [
x∈ {Uα : α ∈ I ∩ Fx } = {U α : α ∈ I ∩ Fx } ⊂ {U α : α ∈ I}.

(b) ThereSis a finite subset F of X such that {Vx : x ∈ F } covers X.

Then Λ = {Fx : x ∈ F }, which is finite. 

Solution of Exercise 124

(a) Take U = {(n, n + 2) : n ∈ Z}.

(b) Take U = {(0, 1/n) : n ∈ N} ∪ {R}.

(c) Take U = {(1/n + 1, 1/n : b ∈ N} ∪ {[0, 1]}.

(d) Take U = {[0, 1/n] : n ∈ N}. 

Proof of Theorem 125

Let X be a paracompact T2 -space.

(a) Fix x ∈ X and H ⊂ X closed not containing x. For each y ∈ H

let Oy and Vy be open containing x and y, respectively. Note that
U = {X \ H} ∪ {Vy : y ∈ H} is an open over of X; let V be a locally
finite open refinement thereof. Take W = {V ∈ VS: V ∩ H 6= ∅}. By
Lemma 123(a) W is closure-preserving; put W = W and note W is
open containing H.
Fix A ∈ W and note A ⊂ Vy for some y ∈ H. Since S x ∈ Oy and
Oy ∩ Vy = ∅, we have x ∈ / A. It follows that x ∈
/ W = {A : A ∈ W}.

(b) Let H and K be disjoint closed sets. By 125(a), for each y ∈ K there
exist disjoint open sets Oy and Vy containing H and y, respectively. Let
V be a locally
S finite open refinement of U = {X \ K} ∪ {Vy : y ∈ K}.
Put W = {A ∈ V : A ∩ K 6= ∅} and note that X \ W and W are
disjoint open sets containing H and K, respectively. 

140
Proof of Theorem 126
Let X be regular Lindelöf and let A be an open cover of X. For each
x ∈ X there exists Ax ∈ A and Wx open in X such that x ∈ Wx ⊂
W x ⊂ Ax . Let W = {Wxn : xn ∈ X ∧ n < ω} be a countable
S subcover
of {Wx : x ∈ X}. For each n < ω put Vn = Axn \ ( i<n W xi ); take
V = {Vn : n < ω}.
Fix y ∈ X and put m = min {n < ω : y ∈ Wxn }. Note y ∈ Axm and
y∈/ W xi for i < m, so y ∈ Vm . To see that V is locally finite, observe
that Vn ∩ Wxm = ∅ when n > m. 

Proof of Theorem 127

→ Certainly a finite subcover of an open cover is a locally finite open
refinement of the same.

← It suffices to show that any locally finite collection of subsets of X

is finite. (To do so we only need the fact that X is countably compact.)
Let U = {Un ⊂ X : n < ω} be locally finite and put
[ [
Cn = {U k : k > n} = {Uk : k > n}.

Observe that C1 ⊃ C2 ⊃ · · · and that each Cn is closed and nonempty.



Proof of Theorem 128

Proof of Theorem 129

Proof of Theorem 130

Let D be the set of all functions f : I → I of the form


q1 if t ∈ [0, p1 )

q2 if t ∈ [p1 , p2 )

f (t) = . ,

..


if t ∈ [pn−1 , 1]

qn

141
9 Proofs

where pi , qi ∈ Q ∩ I, p1 < p2 < · · · < pn−1 , and n ∈ N. Certainly this

is a countable dense subset of I I . 
Proof of Theorem 131
+
Let D be a dense subset of I c . For each α ∈ c+ , put Dα = πα−1 [[0, 1/2)]∩
6 β ∈ c+ then πα−1 [(1/2, 1]] ∩ Dβ 6= ∅, so Dα 6= Dβ . It follows
D. If α =
that |D| > c+ . 

Proof of Theorem 132

Suppose S is an uncountable collection of finite sets. Since
[
S= {A ∈ S : |A| = n},
n<ω

there exists n ∈ N and an uncountable S 0 ⊂ S such that |A| = n for all

A ∈ S 0 . We prove the following via induction on n:

(∗) If S 0 is an uncountable collection of sets and n ∈ N such that

|A| = n for all A ∈ S 0 , then there exists an uncountable S 000 ⊂ S 0
and a set r such that A ∩ B = r for all A 6= B ∈ S 000 .

Well, if n = 1 then all members of S 0 are distinct, so let S 000 = S 0 and

r = ∅.
Suppose k ∈ N and (∗) holds whenever n = k; we show (∗) holds
when n = k + 1. Suppose S 0 is uncountable and |A| = k + 1 for all
A ∈ S 0 . For each A ∈ S 0 , enumerate A = {A(1), . . . , A(k + 1)} and let
A6k = {A(1), . . . , A(k)}.
(1) If {A6k : A ∈ S 0 } is countable, there exists an uncountable S 000 ⊂
S 0 such that A6k = B6k and A(k + 1) 6= B(k + 1) for all A 6=
B ∈ S 000 . Let r = A6k for some (any) A ∈ S 000 .

(2) If {A6k : A ∈ S 0 } is uncountable, then by the induction hypoth-

esis there exists an uncountable S 00 ⊂ S 0 and a set r such that
A6k ∩ B6k = r for all A 6= B ∈ S 00 .
If {A(k + 1) : A ∈ S 00 } is countable, then there is an uncountable
S 000 ⊂ S 00 and x such that A(k + 1) = x for all A ∈ S 000 ; then S 000
and r ∪ {x} work. If {A(k + 1) : A ∈ S 00 } is uncountable, then
there is an uncountable S 000 ⊂ cS 00 such that the A(k + 1)’s are

142
 distinct as A ranges over S 000 . Note that A(k + 1) ∈
/ r for any
A ∈ S 000 . Hence A ∩ B = r for any A 6= B ∈ S 000 . 

Proof of QTheorem 133 Q

Let X = α<κ Xα . Suppose for some finite F ⊂ κ that XF = α∈F Xα
does not have the ccc. Then there exists an uncountable collection
{Ui : i ∈ Λ} Q
of pairwise-disjoint basic open subsets of XF . If for each
i ∈ Λ, Ui = α∈F Uαi with Uαi open in Xα , then
Y Y
{ Uαi × Xα : i ∈ Λ}
α∈F α∈κ\F

is an uncountable collection of pairwise-disjoint open subsets of X.

Thus, X does not have the ccc.
Now, suppose X does not have the ccc. Then there exists an uncount-
able collection {Ui : i ∈ Λ} of pairwise-disjoint basic open subsets of
X. For each i ∈ Λ, let Fi ⊂ κ be the support of Ui . By Theorem 132,
there exists an uncountable subcollection Λ0 of Λ and R ⊂ κ such that
Fi ∩ Fj = R for all distinct i, j ∈ Λ0 . Observe that R 6= ∅, else the Ui
0
would not be pairwise-disjoint. If i 6= j ∈ Λ Q, there exists α ∈ 0R such
that πα [Ui ] ∩ πα [Uj ] = ∅. It follows that { α∈R πα [Ui ] : i ∈QΛ } is an
uncountable collection of pairwise-disjoint open subsets of α∈R Xα ,
whence not every finite subproduct has the ccc. 

Proof of Corollary 134

Proof of Lemma 135

Suppose H is a discrete collection of closed sets. If two members of H
0
S 0 then obviously H can’t be discrete. Let H ⊂ H, and suppose
intersect,
p∈/ H . Let U be an open neighborhood of P meeting at most S one
member, say H0 , of H. If p ∈ H0 , then H0 ∈ / H0 , so U misses H0 . If
H0 ∈ H0 , then
Sp ∈
/ H0 and then V = U \ H0 is an S open neighborhood
0
of p missing
S 0 H . Thus p is not a limit point of H0 , and it follows
that H is closed. 

143
9 Proofs

Proof of Theorem 136

Suppose X is paracompact Hausdorff. Let H be a discrete collection
of closed subsets of X. For each x ∈ X, let Ux be a neighborhood of x
intersection at most one member of H. Let Ux0 be an open set containing
x whose closure misses
(1) {H 0 ∈ H : H 0 6= H} if there exists H ∈ H such that Ux ∩ H 6=
S
∅, or
S
(2) H if there is no such H.
Let V be a locally finite refinement of {Ux ∩ Ux0 : x ∈ X}. Note that
the closure of each member of V meets at most one member of H.
Define, for H ∈ H,
[
WH = X \ {V ∈ V : V ∩ H = ∅}
[
= X \ {V : V ∈ V ∧ V ∩ H = ∅}.

WH is an open set containing H. Suppose H 0 6= H and x ∈ WH ∩ WH 0 .

Let V ∈ V contain x; then V meets at most one member of H. Therefore,
either V ∩H = ∅ or V ∩H 0 = ∅. Therefore, either x ∈
/ WH or x ∈
/ WH 0 ,
a contradiction. So WH ∩ WH 0 = ∅. 
Proof of Corollary 137

Proof of Example 138

We first prove that X is not collectionwise normal. Note that {{eα } : α ∈ A}
is a discrete collection of singleton sets if and only if E is a closed set, and
its subspace topology is discrete. If x ∈ / E, then {x} is an open set miss-
ing E; thus E is closed. Now fix eα ∈ E. Let Uα = {f ∈ X : f [{α}] = 1}.
Uα is an open set containing {eα } and missing {eβ } for each β 6= α.
Thus E is discrete.
Now suppose eα ∈ Vα for all α ∈ A, where Vα is open in X. There is
an open set Oα in the product topology of 2P(A) such that eα ∈ Oα ⊂
Vα . The product topology is ccc, so the collection {Oα : α ∈ A} is not
pairwise disjoint, hence {Vα : αA} is not pairwise disjoint. So X is not
collectionwise normal.

144
 We now show that X is normal. Since X \ E consists of isolated
points, it is not difficult to see that if any two disjoint subsets of E
can be separated by disjoint open sets, then any two disjoint closed
subsets of X can be separated by disjoint open sets. So let H and
K be disjoint subsets of E. Let AH = {α ∈ A : eα ∈ H}. Let U =
{f ∈ X : f [AH ] = 1} and V = {f ∈ X : f [AH ] = 0}. Then

eα ∈ H → α ∈ AH → eα [AH ] = 1 → eα ∈ U,

and
eα ∈ K → α ∈
/ AH → eα [AH ] = 0 → eα ∈ V.
So U and V are disjoint open sets containing H and K. 
Proof of Theorem 139
Let hX, di be a metric space. For each x ∈ X and K ⊂ X closed not
containing x, put d(x, K) = inf {d(x, y) : y ∈ K}. Some neighborhood
of x misses K, so d(x, K) > S 0. For each pair H, K of disjoint closed
subsets of X, let U (H, K) = x∈H B(x, d(x, K)/2). Since U (H, K) ∩
U (K, H) = ∅, H ⊂ U (H, K) ⊂ U (H, K) ⊂ X \ K. If H ⊂ H 0 and
K 0 ⊂ K, with H 0 , K 0 disjoint closed subsets of X, observe
[
B(x, d(x, K 0 )/2) ⊂ U (H 0 , K 0 ).
x∈H

But d(x, K 0 ) > d(x, K), so

[
U (H, K) ⊂ B(x, d(x, K 0 )/2) ⊂ U (H 0 , K 0 ).
x∈H 

Proof of Lemma 140

Assume X is monotonically normal. Let H, K be disjoint closed subsets
of X. By assumption, there exists a monotone normality operator U
for X. Define U 0 (H, K) = U (H, K) \ U (K, H). Clearly, U 0 inherits the
properties of a monotone normality operator from U . Furthermore,

U 0 (H, K) ∩ U 0 (K, H) =

[U (H, K) \ U (K, H)] ∩ [U (K, H) \ U (H, K)] = ∅. 

145
9 Proofs

Proof of Theorem 141

Let X be a monotonically normal space and let U be the monotone
normality operator guaranteed by Lemma 140. Let H be a discrete
collection of closed subsets of X. Note by Lemma 135 that H is pairwise-
disjoint and H0 is closed for any subcollection H0 of H.
S
We want to find a collection U of pairwise-disjoint
S open sets UH such
that UH ⊃ H for each H ∈ H. Put UH = U (H, H \ {H}) for each
H ∈ H.
Let K 6= H ∈ H. Then
[ [
UH ∩ UK = U (H, H \ {H}) ∩ U (K, H \ {K})

⊂ U (H, K) ∩ U (K, H) = ∅,
so U = {UH : H ∈ H} is the desired collection. 
Proof of Theorem 142
Suppose X is T1 .
(a → b) Note singletons in X are closed and let U be a monotone
normality operator satisfying the condition of Lemma 140. For each
x ∈ X and open neighborhood W of x, let Wx = U ({x}, X \ W ). Then
Wx is an open neighborhood of x. Suppose x, y ∈ X and W, V are open
neighborhoods of x and y, respectively, with x ∈
/ V and y ∈
/ W . Then
X \ W ⊃ {y} and {x} ⊂ X \ V . So
Wx ∩Vy = U ({x}, X\W )∩U ({y}, X\V ) ⊂ U ({x}, {y})∩U ({y}, {x}) = ∅.

(b → c) Trivial.

(c → a) For disjoint closed H and K, define

[
U (H, K) = {Ux : x ∈ H ∧ U ∩ K = ∅}.

For each x ∈ H there exists a (basic) open neighborhood thereof which

misses K, so that H ⊂ U (H, K) ⊂ U (H, K). We show U (H, K) ⊂ X\K.
Suppose p ∈ K. Let V be an open set containing p which misses H.
We may assume Vp ⊂ V . If x ∈ H and U ∩ K = ∅, then we can have
neither x ∈ Vp nor y ∈ Ux , so Vp ∩ Ux = ∅. So Vp ∩ U (H, K) = ∅, so
p∈/ U (H, K). 

146
Proof of Theorem 143
Let X be a monotonically normal T1 -space. For each x ∈ X and each
open neighborhood U 0 of x, let Ux0 be defined as in Theorem 142(b).
Let Y ⊂ X. For each open U ⊂ Y , there exists U 0 open in X such that
U = U 0 ∩ Y . For each y ∈ Y and each open neighborhood U of y, put
Uy = Uy0 ∩ Y .
6 ∅, then Y ∩ (Ux0 ∩ Vy0 ) 6= ∅, so Ux0 ∩ Vy0 =
Let x, y ∈ Y . If Ux ∩ Vy = 6 ∅.
It follows that x ∈ V or y ∈ U 0 . But x ∈ V 0 implies x ∈ V and y ∈ U 0
0

implies y ∈ U , so Ux ∩Vy 6= ∅ implies x ∈ V or y ∈ U . By Theorem 142,

Y is monotonically normal. 
Proof of Theorem 144
Let X be linearly ordered by <. We will show X is monotonically
normal by way of Theorem 142. Let B be the base of all open intervals
(a, b) with a, b ∈ X ∪ {−∞, ∞}. Let ≺ be any well-ordering of X.
For a < x < b, define ax as a if (a, x) = ∅, otherwise let ax be the
≺-least element of (a, x). Similarly, define bx to be x if (x, b) = ∅, and
the ≺-least element of (x, b) otherwise. Now, let x ∈ (a, b) ∈ B and
y ∈ (c, d) ∈ B. Put (a, b)x = (ax , bx ) and (c, d)y = (cy , dy ).
Assume (a, b)x ∩ (c, d)y = 6 ∅. Without loss of generality we may
assume d > c. Suppose that x ∈ / (c, d) and y 6 (a, b). Then x 6 c and
y > b. Also, (a, b)x ∩ (c, d)y = (ax , bx ) ∩ (cy , dy ) 6= ∅ implies cy < bx .
So we have x 6 c < cy < bx < b 6 y. But bx is the ≺-least element
of (x, b) so bx ≺ cy , and cy is the ≺-least element of (c, y) so cy ≺ bx ,
a contradiction. Thus y ∈ (a, b) or x ∈ (c, d), so X is monotonically
normal. 
Proof of Theorem 145
(a) Note πf ◦ eF = f for all f ∈ F.

(b) let x, y ∈ X such that x 6= y. Since F separates points, there exists

f ∈ F such that f (x) 6= f (y). It follows that

eF (x) = hf (x) : f ∈ Fi =
6 hf (y) : f ∈ Fi = eF (y).

(c) Note F separates points; it suffices to show eF is closed. Let H ⊂ X

be closed and suppose on the contrary that eF [H] is not closed. Fix
y ∈ eF [H] \ eF [H] and x ∈ X \ H such that eF (x) = y. Since x ∈ /H

147
9 Proofs

and F separates points from closed sets, theereQexists f ∈ F such that

f (x) ∈
/ f [H]. But this means y = eF (x) ∈ / f ∈F f [H] ⊃ eF [H], a
contradiction. 

Proof of Theorem 146

(a → b) X is normal by Theorem 49 and second-countable by Theo-
rem 85.

(b → c) X is hereditarily Lindelöf by Theorem 84 and normal by

Theorem 68; we first show it is perfectly normal.
Let H be a closed subset of X. Then for each x ∈ X \ H, there is
an open neighborhood Ux of x such that U x ∩ H = ∅. S Let U be a
countable subcover
T of {U x : x ∈ X \ H}. Since X \ H = {U : U ∈ U},
we have H = {X \ U : U ∈ U}.
Now let B = {Bn : n < ω} be a base for X. By Theorem 59(c), for
each n < ω there is a map fn : X → [0, 1] such that fn−1 (0) = X \ Bn ;
put F = {fn : n < ω}. If H ⊂ X is closed and x ∈ X \ H, there
exists n < ω such that x ∈ Bn and H ⊂ X \ Bn . It follows that
fn (x) ∈
/ f [H], so F separates points from closed sets. By Theorem 145,
X is homeomorphic to a subspace of I N .

(c → a) Note I N is separable by Theorem 88 and metrizable by Corol-

lary 41, so any subspace of I N is separable and metrizable. 

Proof of Theorem 147

→ For any H closed in X and any x ∈ X \H, there is a map fx,H : X →
[0, 1] which separates x from H; let F be the collection of these maps.
|F |
By Theorem 145, X is homeomorphic to a subspace of [0, 1] .

← I κ is completely regular by Theorem 50, so any subspace is also. 

Proof of Theorem 148

Let X be a regular space.

(a → b) By definition.

148
(b → c) Let U be an open cover of X. For each x ∈ X, pick Ux ∈ U
and Wx ⊂ X open such that x ∈ Wx ⊂ W x ⊂ Ux . Let V be a locally
finite refinement of {Wx : x ∈ X}; put B = {V : V ∈ V}. That B is a
closed refinement of U is immediate.
Fix x ∈ X and let O ⊂ X be an open neighborhood of x meeting
finitely many members of V. It follows that {V ∈ V : O ∩ V 6= ∅} is
finite, so B is locally finite.

(a → c) Let U be an open cover of X, and let A be a locally finite

closed refinement thereof. Let V be those open subsets of X which meet
only finitely many members of A; note that every point in X has a
neighborhood meeting only finitely many members of A, so V is an
open cover X. Let B be a locally finite closed refinement of V. For each
A ∈ A pick UA ∈ U such that A ⊂ UA ; define
[
WA = X \ {B ∈ B : B ∩ A = ∅}.

Put OA = WA ∩ UA and let O = {OA : A ∈ A}; certainly O is an open

refinement of U.
Fix x ∈ X and let O be an open neighborhood of x meeting only
finitely many members of B, say B1 , B2 , . . . , Bn . Let I be those members
of A meeting at least one of B1 , B2 , . . . , Bn ; note I is finite since B
refines V. Note O ⊂ B1 ∪ B2 ∪ · · · ∪ Bn ; if A ∈ A \ I then A misses
B1 , B2 , . . . , Bn , so O misses WA . It follows that O misses OA for each
A ∈ A \ I, whence O can meet only finitely many members of O. 

Proof of Theorem 149 S

Let U be an open cover of X and let V = n∈N Vn be S a σ-locally finite
open refinement thereof.SFor each n ∈ N let Wn = Vn , and for each
V ∈ Vn let Sn (V ) = S
V \ i<n Wi . Let Cn = {Sn (V ) : V ∈ Vn } for each
n ∈ N, and put C = n∈N Cn .
Fix x ∈ X and let nx = min {n ∈ N : x ∈ Wn }. Choose Vx ∈ Vnx
such that x ∈ Vx . Then x ∈ Snx (Vx ) ∈ C, so C refines U.
For each i 6 nx , choose an open neighborhood S Oi of x meeting
nx
only finitely many members of Vi . Note that O = i=1 (Vx ∩ Oi ) is
an open neighborhood of x meeting only finitely many members of

149
9 Proofs

C1 , C2 , . . . , Cnx . But Vx misses Sn (V ) whenever n > nx , so O misses

every member of Cn whenever n > nx . It follows that C is locally finite.
By Theorem 148, X is paracompact. 

Proof of Lemma 150

For each x ∈ X, let B(x) = Bd (x, 1/4n).
Fix n ∈ N and x ∈ X; we show B(x) meets at most one element of
Vn . Suppose there are α1 < α2 < κ such that B(x) ∩ Vαi ,n 6= ∅ for
i = 1, 2. For i = 1, 2 there is yi ∈ Hαi ,n such that B(x) ∩ B(yi ) 6= ∅.
Choose zi ∈ B(x) ∩ B(yi ) for i = 1, 2. Then

d(y1 , y2 ) 6 d(y1 , x) + d(x, y2 )

6 d(y1 , z1 ) + d(z1 , x) + d(x, z2 ) + d(z2 , y2 ) < 1/n.

S y2 ∈ Bd (y1 , 1/n) ⊂ Uα1 . This is a contradiction, since

It follows that
y2 ∈ Uα2 \ α<α2 Uα .
We nowSshow that V covers X. Let αx = min {α : x ∈ Uα }. Then
x ∈ Uαx \ α<αx Uα and distd (x, X \ Uαx ) > 0, where

distd (x, A) = inf {d(x, a) : a ∈ A}.

Choose m ∈ N such that distd (x, X \ Uαx ) > 1/m. Then

x ∈ Hαx ,m ⊂ Vαx ,m ∈ Vm ⊂ V.

By construction Vα,n ⊂ Uα for each α < κ and n ∈ N. Thus V is a

σ-locally finite open refinement of U. 

Proof of Corollary 151

Combine Theorem 48, Theorem 49, Theorem 149, and Lemma 150. 

Proof of S
Lemma 152
Let B = n∈N Bn be a σ-locally finite base of a regular space X. Let
H ⊂ X be closed. For all n ∈ N, put
[
Un = {B : B ∈ Bn ∧ B ⊂ X \ H}.

Note that U n ⊂ X \ H.

150
 T
Put On = X \ U n ; note H ⊂ On for all n ∈ N. Let x ∈ n∈N On and
suppose x ∈
/ H; there exists U ⊂ X open such that x ∈ U ⊂ U ⊂ X \H.
Pick n ∈ N and B ∈ Bn such that x ∈ B ⊂ Ux . It follows that
B ⊂ X \H, so x ∈ Un . But On ⊂ X \Un implies x ∈/ On , a contradiction.
By Theorem 59(b), X is perfectly normal. 
Proof of Lemma 153
We first show that d is a metric on X. For each n ∈ N and B ∈ Bn ,
|fn,B (x) − fn,B (y)| 6 1/n 6 1. So supf ∈F {|f (x) − f (y)|} 6 1, whence
d is well-defined.
Certainly d(x, y) = d(y, x) and x = y implies d(x, y) = 0. Suppose
x 6= y. Then there exists n ∈ N and B ∈ Bn such that x ∈ B and y ∈ / B,
implying fn,B (y) = 0 and fn,B (x) > 0; we have

0 < fn,B (x) 6 sup {|f (x) − f (y)|} = d(x, y).

f ∈F

Let x, y, z ∈ X. For each f ∈ F we have

|f (x) − f (z)| 6 |f (x) − f (y)| + |f (y) − f (z)| 6 d(x, y) + d(y, z),

so d(x, z) 6 d(x, y) + d(y, z).

Let T be the given topology on X, and let Td be the topology on X
generated by d; we first show Td ⊂ T . Fix x ∈ X and ε > 0; we find
O ∈ T with x ∈ O ⊂ Bd (x, ε). Choose n0 ∈ N such that 1/n0 < ε/2.
Claim. For all 1 6 n 6 n0 , there exists On ∈ T such that x ∈ On
implies supB∈Bn |fn,B (x) − fn,B (y)| 6 ε/2 for all y ∈ On .
Proof of Claim. Let 1 6 n 6 n0 ; then there is an open neighborhood U
of x meeting only finitely many members of Bn , say B1 , . . . , Bm . Using
fn,Bi , choose an open neighborhood Ui of x such that

|fn,Bi (x) − fn,bi (y)| < ε/2

Tm
for all y ∈ Ui . Put On = i=1 (U ∩ Ui ); for all y ∈ On we have
|fn,B (x) − fn,B (y)| < ε/2 for all B ∈ Bn . It follows that

sup {|fn,B (x) − fn,B (y)|} 6 ε/2

B∈Bn

for all y ∈ On . 

151
9 Proofs

T n0
Put O = n=1 ; certainly O is an open neighborhood of x. Let y ∈ O.
For n > n0 , |fn,B (x) − fn,B (y)| < 1/n < 1/n0 < ε/2 for all B ∈ Bn .
For n 6 n0 , O ⊂ On gives |fn,B (x) − fn,B (y)| 6 ε/2 for all B ∈ Bn .
Thus
d(x, y) = sup {|f (x) − f (y)|} 6 ε/2 < ε
f ∈F

for all y ∈ O, so O ⊂ Bd (x, ε).

It remains to show T ⊂ Td . Let B ∈ B and x ∈ B; there exists n ∈ N
such that fn,B (x) > 0. Let ε = fn,B (x). Note that y ∈ Bd (x, ε) implies
|fn,B (x) − fn,B (y)| < ε. It follows that fn,B (y) 6= 0, so y ∈ B. Hence,
Bd (x, ε) ⊂ B. 
Proof of Theorem 154
→ Let X be metrizable; recall that metrizable spaces are regular and
paracompact. For each n ∈ N define Un = {B(x, 1/n) : x ∈SX} and
let Vn be a locally finite open refinement thereof. Let V = n∈N Vn ;
certainly V is σ-locally finite.
Fix x ∈ X and ε > 0. Choose m ∈ N such that 2/m < ε. Pick
V ∈ Vm containing x and fix y ∈ V . Recall that Vm refines Um , so
V ⊂ B(z, 1/m) for some z ∈ X. It follows that
d(x, y) 6 d(x, z) + d(z, y) 6 1/m + 1/m = 2/m < ε,
so y ∈ B(x, ε).

← This is Lemma 153. 

Proof of Theorem 155

Let X be a locally metrizable paracompact T2 -space; by Theorem 125,
X is regular. For each x ∈ X let Nx be a metrizable neighborhood of x;
put U = {Nx◦ : x ∈ X}. Let V be a locally finite open refinement of U.
For each V ∈ V, let dV be a metric generating the topology on V . Each
V ∈ V is open in X, so any ball BdV (x, ε) ⊂ V with x ∈ V and ε > 0 will
be open in X. For each n ∈ N let Bn = {BdV (x, 1/n) : V ∈ V ∧ Sx ∈ V }
and let Bn0 be a locally finite open refinement thereof. Put B = n∈N Bn0 ;
certainly B is σ-locally finite.
Now let U ⊂ X be open with x ∈ U . We know finitely many members
of V contain x, say V1 , . . . , Vk . For each 1 6 i 6 k U ∩ Vi is an open

152
neighborhood of x in Vi , so there exists ε > 0 such that BdVi (x, ε) ⊂
U ∩Vi for all 1 6 i 6 k. Choose n ∈ N such that 1/n < ε/2. Recall Bn0 is
a cover, so there exists B ∈ Bn0 ⊂ B containing x. But Bn0 is a refinement
of Bn , so there is V ∈ V and y ∈ V such that x ∈ B ⊂ BdV (y, 1/n).
Since x ∈ V , V = Vi for some 1 6 i 6 k. It follows that

x ∈ B ⊂ BdV (y, 1/n) = BdVi (y, 1/n) ⊂ BdVi (x, ε) ⊂ U,

hence B is a basis. By Theorem 154, X is metrizable. 

Proof of Lemma 156
Since X is compact Hausdorff, X is regular. Let
[
B = {X \ V : V ⊂ U is finite}.

Clearly B is countable.
Let U ⊂ X be open and let x ∈ U . For all y ∈ X \ U , there exists
Uy ∈ U with y ∈ Uy and x ∈ / Uy . Observe that the Uy cover X \ U ,
so there exist finitely many, say U1 , . . . , Un , which cover X \ U . Then
Tn Tn Tn
x ∈ i=1 (X \ U i ) ⊂ U . But i=1 (X \ U i ) = X \ i=1 Ui ∈ B, so
B is a base. It follows that X is second-countable, hence separable
metrizable. 
Proof of Theorem 157
Suppose X is compact Hausdorff.

← Assume X has a Gδ diagonal. T There exist open subsets Un , n <

ω, of X × X such that ∆ = n<ω Un . We show X is metrizable by
Lemma 156. Because X is compact Hausdorff, X is regular. So for each
n < ω and x ∈ X there exists an open Bn (x) ⊂ X containing x, with
Bn (x) × Bn (x) ⊂ Un . For each
S n < ω let Vn be a finite subcover of
{Bn (x) : x ∈ X}. Let B = n<ω Vn . Clearly B is countable. And if
x 6= y ∈ X then there exists n < ω such that hx, yi ∈/ Un . So there
exists B ∈ Vn such that hx, xi ⊂ B × B ⊂ Un and so y ∈/ B. 

Proof of Theorem 158

Suppose X has a countable network {Ni : i < ω} and f : X → Y is a
continuous surjection. Claim {f [Ni ] : i < ω} is a countable network for

153
9 Proofs

Y . Well suppose V is open in Y and y ∈ V . We find i < ω such that

y ∈ f [Ni ] ⊂ V . Since f is surjective, there exists x ∈ f −1 (y). Since
{Ni : i < ω} is a network for X and f −1 [V ] is an open subset of X
containing x, there exists i < ω such that x ∈ Ni ⊂ f −1 [V ]. Then
y = f (x) ∈ f [Ni ] ⊂ f [f −1 [V ]] ⊂ V . 

Proof of Theorem 159

(a → b) Let hX, T i be a T1 -space with T a countable network N . Put
S = N ∪ {X \ N : N ∈ N } and B = { F : F ⊂ S is finite}. Let TN
be the topology on X generated by B.
Certainly hX, TN i is second-countable. Let x ∈ X and H ⊂ X closed
in hX, TN i not containing
S x. Pick U ∈ B such that x ∈ U ⊂ X \ H.
Recall that U = F for some finite F ⊂ S. Since each F ∈ F is
clopen, U is clopen. It follows that U and X \ U are disjoint open
sets containing x and H, respectively. Thus hX, TN i is regular, hence
separable metrizable. S
Let U ∈ T . Then U = N 0 for some N 0 ⊂ N . Since N 0 ⊂ B, U ∈ TN .
Therefore, f : hX, TN i → hX, T i defined by x 7→ x is continuous.

(b → a) Recall that separable metric spaces are second-countable.

Since any countable base is necessarily a countable network, X has
a countable network by Theorem 158. 

Proof of Theorem 160

Suppose X is regular and has a countable network N . We show X × X
has a Gδ diagonal ∆ by showing (X × X) \ ∆ is the countable union
of closed sets.
Clearly L = {N × M : N, M ∈ N ∧ N × M ⊂ (X × X) \ ∆} is a
countable
S collection of closed subsets of X ×X. We claim (X ×X)\∆ =
L. Well, suppose hx, yi ∈ (X × X) \ ∆. Since X × X is regular there
exists a basic open set B1 × B2 with hx, yi ∈ B 1 × B 2 ⊂ (X × X) \ ∆.
There exists N, M ∈ N such that x ∈ N ⊂ B1 and y ∈ M ⊂ B2 , so
that hx, yi ∈ N × M ⊂ B 1 × B 2 ⊂ (X × X) \ ∆. 

Proof of Theorem 161

Suppose X is compact metrizable and Y is a Hausdorff continuous
image of X. Every compact metrizable space has a countable basis.

154
In particular, X has a countable network. By Theorem 158, Y has a
countable network. As the continuous image of a compact space, Y is
compact. Since Y is also Hausdorff by assumption, we have Y is regular.
By Theorem 160, Y has a Gδ diagonal. This, together with the fact that
Y is compact Hausdorff, implies Y is metrizable by Theorem 157. 

Proof of Theorem 162

Let f : X → Y be a continuous surjection.

(a → b) Suppose y ∈ Y . Let U ⊂ X be an open set containing f −1 (y).

Then V = Y \ f [X \ U ] is open in Y containing y. Since V ∩ f [X \ U ] =
∅, we have f −1 [V ] ⊂ U .

(b → c) Let U ⊂ X be open and let y ∈ f ∗ [U ]. There exists an open

V ⊂ Y such that y ∈ V and f −1 [V ] ⊂ U . So f −1 (y 0 ) ⊂ U for all y 0 ∈ V .
So V ⊂ f ∗ [U ], implying f ∗ [U ].

(c → a) Suppose H ⊂ X is closed and y is a limit point of f [H]. By

assumption, f ∗ [X \ H] is open; it clearly misses f [H], so y ∈
/ f ∗ [X \ H].
Hence exists x ∈ H with f (x) = y. So y ∈ f [H]. This proves f [H] is
closed. 

Proof of Theorem 163

Let f : X → Y be a perfect surjection, where X is separable metrizable.
Then X has a countable base B. Without loss of generality, we may
assume B is closed under finite unions. Now let C = {f ∗ [B] : B ∈ B},
where f ∗ is as defined in Theorem 162(c). So C is a countable collection
of open sets.
We prove that C is a base for Y . Let y ∈ U , where U is open in Y .
Then f −1 (y) is a compact subset of the open set f −1 [U ]. Since B is
closed under finite unions, there is some B ∈ B with f −1 (y) ⊂ B ⊂
f −1 [U ]. Then y ∈ f ∗ [B] ⊂ U . Hence C is a countable base for Y .
Since a regular space with a countable base is separable and metriz-
able, it remains to prove that Y is regular. Each point of X is closed,
and f is a closed map, so it follows that each point of Y is closed; so Y
is T1 . Now suppose y ∈ Y and H is a closed subset of Y not containing
y. Since X is normal, there are disjoint open sets U and V containing

155
9 Proofs

f −1 (y) and f −1 [H], respectively. Then f ∗ [U ] and f ∗ [V ] are disjoint

open sets containing y and H, respectively. Thus Y is regular. 
Proof of Theorem 164
(a → b) Let A1 , A2 , . . . be a decreasing sequence of nonempty closed
sets with diam hAn i → 0; fix xn ∈ An for all n ∈ N. Note hxn i is a
Cauchy sequence. By completeness, hxn i → x for some x ∈ X. Since
hxn in>k → x for each k ∈ N, x is a limit point of {xn : n > k} ⊂ Ak .
Since Ak is closed, x ∈ Ak for all k.

(b → a) Let hxn i be a Cauchy sequence, and for each n set An =

{xi : i > n}; certainly A1 , A2 , . . . is a decreasing sequence of nonempty
closed sets. Fix ε > 0; there is some k ∈ N such that d(xi , xj ) 6 ε
whenever i, j > k. It follows that diam Ak = diam {xi : i > k} 6 ε, so
diam hAn i → T 0.
Pick x ∈ n∈N An and let U be a neighborhood of x. There is
ε > 0 such that B(x, ε) ⊂ U . Pick An where diam An < ε. Then
x ∈ An ⊂ B(x, ε). But An contains all but finitely many members of
hxn i, so B(x, ε) does also. Thus, hxn i → x. 

Proof of Corollary 165

Let d be a metric on the compact space X. By Theorem 73, any de-
creasing sequence of nonempty closed sets has nonempty intersection.
By Theorem 164, d is complete. 
Proof of Theorem 166
Let d be a complete metric on the space X. Let O = {On : n ∈ N} be a
countable
T collection of dense open subsets of X; we want to show that
O is dense.
Let V be a nonempty open set. Since O1 is dense, V ∩ O1 6= ∅. Pick
x1 ∈ V ∩ O1 and 0 < ε1 < 1 such that B(x1 , ε1 ) ⊂ V ∩ O1 . Since
O2 is dense, B(x1 , ε1 ) ∩ O2 6= ∅. Again, pick x2 ∈ B(x1 , ε1 ) ∩ O2 and
0 < ε2 < 1/2 such that B(x2 , ε2 ) ⊂ B(x1 , ε1 ) ∩ O2 . Inductively, for
each n ∈ N we may pick xn ∈ B(xn−1 , εn−1 ) ∩ On and 0 < εn < 1/n
such that B(xn , εn ) ⊂ B(xn−1
T , εn−1 ) ∩ On .
By Theorem 164, fix x ∈ n∈N B(xn , εn ). Since x ∈ On for all n ∈ N
T
we have x ∈ O. Moreover, B(x1 , ε1 ) ⊂ V ∩ O1 , so x ∈ V . 

156
Proof of Theorem 167
→ Assume d  A × A is complete and let A ⊂ X. Fix x ∈ A. There
exists hxn i → x where xn ∈ A for all n ∈ N; certainly hxn i is Cauchy
in A. It follows that hxn i → y for some y ∈ A. Since X is Hausdorff we
must have y = x, whence A is closed.

← Let d be a complete metric on X and let A ⊂ X be closed. Let hxn i

be a Cauchy sequence of A; certainly hxn i is Cauchy in X. Since d is
complete, there exists x ∈ X such that hxn i → x in X. Certainly x is
a limit point of A, so x ∈ A = A. 

Proof of Theorem 168

First a useful fact. Let hX, di be a complete metric space and define
d0 : X × X → R by
(
M if d(x, y) > M,
d0 (x, y) =
d(x, y) otherwise.

By Lemma 39, d0 is equivalent to d; we want to show d0 is complete.

Let hxn i be a d0 -Cauchy sequence. Then for each ε > 0 there exists
k ∈ N such that d0 (xi , xj ) 6 ε whenever i, j > k. If ε < M , then
d(xi , xj ) 6 ε for all i, j > k. If ε > M , set ε0 = M/2; there exists k 0 ∈ N
such that d0 (xi , xj ) 6 ε0 for i, j > k 0 . It follows that d(xi , xj ) 6 ε0 < ε
for all i, j > k 0 . So, hxn i is d-Cauchy and hence converges. It follows
that d0 is complete.
Let hXn i be a collection of completely metrizable spaces. Then for
each n ∈ N there exists a complete metric dn which generates the
topology on Xn . Define for each n ∈ N an equivalent metric d0n such that
d0n (x, y) = 1/2n if dn (x, y) > 1/2n and d0n (x, y) = dn (x, y) otherwise.
By Theorem 40, d(~x, ~y ) = maxn∈N {d0n (xn , yn )} is a metric on the
product of the Xn .
Let h~xn i be a d-Cauchy sequence, where ~xn = h~xn (1), ~xn (2), . . .i. Fix
∞
n ∈ N. We show that h~xn (i)ii=1 is a d0n -Cauchy sequence. Fix ε > 0;
since h~xn i is d-Cauchy, there exists k ∈ N such that d(~xi , ~xj ) 6 ε for all
i, j > k. It follows that max {d0m (~xi (m), ~xj (m)) : m ∈ N} = d(~xi , ~xj ) 6
ε for i, j > k; in particular, d0n (~xi (n), ~xj (n)) 6 ε for i, j > k.

157
9 Proofs

∞
Thus for each n ∈ N. h~xi (n)ii=1 → xn for some xn ∈ Xn . Hence
h~xi i → hxn i = ~x. 

Proof of Lemma 169

(a) Let U ⊂ X be open in X metrizable. Define f : U → f [U ] ⊂
X × R so that x 7→ (x, 1/d(x, X \ U )). Since X \ U is closed, we have
d(x, X \ U ) = 0 iff x ∈ X \ U . So d(x, X \ U ) > 0 for all x ∈ U . It
follows that f is well-defined.
(1) Clearly f is one-to-one and onto its range.
(2) f is continuous as each coordinate function is (noting d is con-
tinuous). Furthermore f −1 (x, y) = x is also continuous, so f is
a homeomorphism.
(3) Let hf (xn )i ⊂ f [U ] be a converging sequence to some y ∈ X × R.
We show y ∈ f [U ]. We have f (xn ) = (xn , 1/d(xn , X \ U )) →
y = (y1 , y2 ), so xn → y1 and 1/d(xn , X \ U ) → y2 . By continuity
of the distance function: xn → y1 implies 1/d(xn , X \ U ) →
1/d(y1 , X \ U ), thus we should have y1 ∈ U and 1/d(y1 , X \ U ) =
y2 since R is Hausdorff. Therefore

y = (y1 , y2 ) = (y1 , 1/d(y1 , X \ U )) ∈ f [U ],

so f [U ] is closed.
T
(b) Now suppose G is Gδ , so G = n>1 Gi with each Gi open in X
metrizable. Define h : G → h[G] ⊂ X × RN so that

x 7→ hx, 1/d(x, X \ G1 ), 1/d(x, X \ G2 ), . . .i.

A similar argument to 169(a) shows that h is a homeomorphism. Now

let hh(xn )i ⊂ h[G] be a sequence converging to some y ∈ X × RN . So

h(xn ) = hxn , 1/d(xn , X \ G1 ), 1/d(xn , X \ G2 ), . . .i

→ y = hy0 , y1 , y2 , . . .i.

Then xn → y0 and 1/d(xn , X \ Gi ) → yi for all i > 1. Now xn → y0

implies
1/d(xn , X \ Gi ) → 1/d(y0 , X \ Gi ),

158
so y0 ∈ Gi and 1/d(y0 , X \ Gi ) = yi for all i > 1. Hence y0 ∈ G and

y = hy0 , 1/d(y0 , X \ G1 ), 1/d(y0 , X \ G2 ), . . .i ∈ h[G],

therefore h[G] is closed. 

Proof of Corollary 170

Let X be a Gδ -subset of the completely metrizable space Y . By Lemma 169(b),
X is homeomorphic to a closed subset of Y × RN . But Y × RN is com-
pletely metrizable by Theorem 168, so X is completely metrizable by
Theorem 167. 
Proof of Lemma 171
Let y ∈ Y . We need to show there is an open-in-X set containing y and
contained in Y . By local compactness, there is an open-in-Y subset U
Y
of Y with y ∈ U and U = K compact. There is an open-in-X subset
U 0 of X such that U 0 ∩ Y = U . Since Y is dense in X, U is dense
in U 0 . Now U = U 0 ∩ Y ⊂ K and K, begin compact, is closed in the
X
Hausdorff space X. Thus U 0 ⊂ U ⊂ K ⊂ Y , so U 0 ⊂ Y , and y is in
the open-in-X set U 0 . Hence Y is open in X. 
Proof of Corollary 172
Let X be locally compact and metrizable, and let X̃ be its completion.
By Lemma 171 X is open, hence Gδ , in X̃. By Corollary 170, X is
completely metrizable. 
Proof of Theorem 173
That d(f, g) = d(g, f ), d(f, g) > 0, and d(f, g) = 0 iff f = g are
straightforward. Let f, g, h ∈ C ∗ (X). For each x ∈ X, |f (x) − h(x)| 6
|f (x) − g(x)| + |g(x) − h(x)|. Hence

d(f, h) = sup |f (x) − h(x)| 6 sup (|f (x) − g(x)| + |g(x) − h(x)|)
x∈X x∈X

6 sup |f (x) − g(x)| + sup |g(x) − h(x)| = d(f, g) + d(g, h).

x∈X x∈X

It follows that d is a metric on C ∗ (X).

Now we prove d is complete. Suppose hfn i is d-Cauchy. Let x ∈ X
and consider hfn (x)i. If ε > 0, there is N ∈ N such that d(fi , fj ) 6 ε

159
9 Proofs

for all i, j > N . But then |fi (x) − fj (x)| 6 ε too, so hfn (x)i is Cauchy,
whence it converges to some g(x) ∈ R; let this define the function g.
Note that

(∀x ∈ X)(∀i > N )[|fi (x) − g(x)| = lim |fi (x) − fj (x)| 6 ε].
j∈N

It follows that g is bounded since fi is, and that hfn i converges uniformly
to g, so g is continuous and hence g ∈ C ∗ (X). Finally it also follows
that d(fn , g) → 0, so hfn i → g in the metric topology. 

Proof of Theorem 174

Let d0 be the metric on C ∗ (X). Then

d0 (j(x), j(y)) = d0 (fx , fy ) = sup |fx (z) − fy (z)|

z∈X

= sup |(d(z, x) − d(z, a)) − (d(z, y) − d(z, a))|

z∈X

= sup |d(z, x) − d(z, y)|.

z∈X

Since d(z, x) 6 d(z, y) + d(x, y), we have d(z, x) − d(z, y) 6 d(x, y). Sim-
ilarly, −(d(z, x) − d(z, y)) 6 d(x, y). Hence supz∈X |d(z, x) − d(z, y)| 6
d(x, y). Now, if z = y then d(z, x) − d(z, y) = d(x, y). Hence

d0 (j(x), j(y)) = d(x, y),

so j is an isometry. 
Proof of Corollary 175

Proof of Theorem 176

Let hX, di be a metric space and let hX̃1 , d˜1 i and hX̃2 , d˜2 i be complete
metric spaces such that
(1) X is dense in X̃1 and X̃2 , and
(2) d˜1  X × X = d˜2  X × X = d.
We want to show that hX̃1 , d˜1 i and hX̃2 , d˜2 i are isometric.
Define a function h : X̃1 → X̃2 as follows:

160
 (1) If x ∈ X ⊂ X̃1 , take h(x) = x.

(2) If x ∈ X̃1 \ X, let hxn i be a sequence in X converging to x (when

considered as a sequence in X̃1 .; note that such a sequence exists
because X̃1 is complete metric and X is dense therein).
Note that hxn i is d˜2 -Cauchy, so it coverges to some y ∈ X̃2 . Put
h(x) = y.
Certainly h is well-defined for all x ∈ X. To see this for x ∈ X̃1 \ X,
suppose hxn i and hyn i are two sequences in X converging to x ∈ X̃1 .
Since both coverge to x, it must be the case that d(xn , yn ) → 0. This
is true in X̃2 , so hxn i and hyn i must coverge to the same point in X2
too.
Now let x, y ∈ X̃1 and let hxn i and hyn i be sequences in X converging
to x and y, respectively. Then

d˜1 (x, y) = lim d˜1 (xn , yn ) = lim d(xn , yn )

n→∞ n→∞

= lim d˜2 (xn , yn ) = d˜2 (h(x), h(y)),

n→∞

so h is an isometry. 

Proof of Theorem 177

For each n > 1, we have Cn is a union of 2n disjoint compact intervals.
Hence Cn is compact and nonempty for each n > 1. Futhermore, we
have Cn+1S⊂ Cn for each n > 1. Hence, by Lemma 97, we conclude
that C = n∈N Cn is a compact nonempty set. It is clear that C is
metrizable as a subspace of a metrizable space.
Let us consider the collection B = {C ∩ Iσ : ∃n ∈ N(σ ∈ 2n )}. Let
n > 1 and σ ∈ 2n be given. We have that C ∩ Iσ is closed as an
intersection of closed sets. Furthermore, we have Cn \ Iσ is closed,
hence C \ (C ∩ Iσ ) = C ∩ (Cn \ Iσ ) is closed. Then C ∩ Iσ is also open
in C, so B is a countable collection of clopen sets in C.
Let U be any open set in C, and let x ∈ U . There is ε > 0 such that
(x − ε, x + ε) ∩ C ⊂ U . Choose n > 1 such that 1/3n < ε. Since x ∈ Cn ,
there is σ ∈ 2n such that x ∈ Iσ . Observe that diam Iσ = 1/3n . Hence
Iσ ⊂ (x−ε, x+ε), which implies that x ∈ C∩Iσ ⊂ C∩(x−ε, x+ε) ⊂ U .

161
9 Proofs

It follows that B is a base for C, whence C has a countable base of

clopen sets.
Let n > 1 and σ ∈ 2n be given. Let a, b ∈ R such that Iσ = [a, b]. By
construction we have a ∈ Iσ0 ⊂ Cm and b ∈ Iσ1 ⊂ Cm for all m > 1,
where
σ0 = σa |00 {z
. . . 0} and σ1 = σa |11 {z
. . . 1} .
m times m times

Hence, {a, b} ⊂ C. This implies that {a, b} ⊂ C ∩ Iσ . Thus |C ∩ Iσ | > 2.

This shows that any element of B has at least two points. Since B is
a base for C, we conclude that any nonempty open set in C has more
than one point. Thus, C has no isolated points.
Let x ∈ C and ε > 0 be fixed. From above, we know that there is
σ ∈ 2n , for some n > 1, such that x ∈ Iσ ⊂ (x − ε, x + ε). let K denote
the middle third open interval such that

Iσ = Iσa0 ∪ K ∪ Iσa1 .

We have that K ⊂ I \ Cn+1 . Hence K ⊂ (x − ε, x + ε) ∩ (I \ C). Thus

(x − ε, x + ε) 6⊂ C, so C has empty interior in R.
For each n > 1, we have
[
C = Cn ∩ C = C ∩ Iσ .
σ∈2n

Since for each n > 1 and σ 6= σ 0 ∈ 2n we have Iσ ∩ Iσ0 = ∅, for each

x ∈ C there is a unique σnx ∈ 2n such that x ∈ Iσnx . For each x ∈ C
and n > 1, let us set σnx = x1 . . . xn . Observe that if m < n, then σnx
x
extends σm . Thus xn is uniquely determined by any σjx for j > n. Now
define
f : C → 2N so that x 7→ hxn i.
Certainly f is well-defined. Let hxn i ∈ 2N be given and consider the
sequence of compact sets hIx1 ...xn i. By definition, we have

Ix1 ...xn+1 ⊂ Ix1 ...xn and diam Ix1 ...xn = 1/3n

T
for all n > 1. T
Thus by Lemma 97 we have n∈N Ix1 ...xn = 6 T ∅. Clearly
we have that n∈N Ix1 ...xn ⊂ C. Since diam Ix1 ...xn → 0, n∈N Ix1 ...xn

162
T a single point. That is, there is a unique x ∈ C such that {x} =
is
n∈N Ix1 ...xn . Hence f (x) = hxn i. Since x is uniquely determined for
each given sequence hxn i ∈ 2N , we conclude that f is bijective. Thus C
is uncountable. 
Proof of Theorem 178
Let B = {Bn : n > 1} be a base of nonempty clopen sets in X.

Claim 1 Given any nonempty clopen set A ∈ X and any n > 1, there
is An nonempty clopen in X such that A\An 6= ∅ and A = An ∪(A\An ).
Furthermore, either each of An , A \ An is contained in Bn or completely
misses Bn .
∅ ( Bn ∩ A ( A: Then A = A ∩ Bn ∪ A \ Bn with A ∩ Bn nonempty
clopen in X and A \ Bn 6= ∅.
A ∩ Bn = ∅ or A ⊂ Bn : Since X has no isolated points and A is
nonempty open in X, there is x, y ∈ A such that x = 6 y. Choose Bm ∈ B
such that x ∈ Bm , y ∈ A \ Bm . Thus A = Bm ∪ (A \ Bm ).

Construction of Cantor Tree of clopen sets Divide X = X∅ into

two clopen nonempty pieces X0 and X1 such that, unless B1 is the
whole space, then B1 is one of the pieces. Then divide each of X0 , X1
into two pieces—say X00 , X01 , X10 , X11 —such that each piece is either
contained in B2 or completely misses B2 . And so on.

Claim 3 Given an infinite sequence σ = hσ1 , σ2 , . . .i ∈ 2N , the inter-

section of the branch X∅ , Xσ1 , Xσ1 σ2 , . . . is a single point.
Let σ ∈ 2N be given. Since X∅ ⊃ Xσ1 ⊃ Xσ1 σ2 ⊃ · · · are compact
nonempty sets in X, then their intersection A is nonempty. Let p ∈ A
and y ∈ X such that y = 6 p. Then there exist n, m ∈ N such that
p ∈ Bn , y ∈ Bm , and Bn ∩ Bm = ∅. Thus p ∈ Xσ1 ...σn ∩ Bn and by
part 1, Xσ1 ...σn ⊂ Bn . There y ∈
/ Xσ1 ...σn and so y ∈ / A, thus A{p}.

Claim 4 {Xσn : n ∈ N} is a base at p.

Let U be open in X such that p ∈ U . Then there is n ∈ N such that
p ∈ Bn ⊂ U . Thus p ∈ Xσn ∩ Bn and by the construction, Xσn ⊂ Bn .

163
9 Proofs

Consider the map f : 2N → X defined by

\
f (σ) = Xσ0 ∩ Xσn .
n∈N

From Claim 3, f is well-defined.

Claim 5 f is one-to-one.
Let σ, τ ∈ 2N such that σ 6= τ . Then there is i ∈ N such that σi 6= τi .
Thus Xσ(i+1) ∩ Xσ(i+1) = ∅. It follows that f (σ) 6= f (τ ).

Claim 6 f is onto.
Let p ∈ X. We construct σ ∈ 2N as follows: σ1 = 0 if p ∈ X0 and
σ1 = 1 if p ∈ X1 . If σ1 , . . . , σn have been defined, take σn+1 = 0 if
p ∈ Xσ1 ...σn a0 and σn+1 = 1 if p ∈ Xσ1 ...σn a1 . Then f (σ) = p.

Claim 7 f is continuous.
Let p ∈ X and σ ∈ 2N such that f (σ) = p. It suffices to show that
−1
f [Xσn ] is open in 2N for each n ∈ N. Since

f −1 [Xσn ] = {σ1 } × · · · × {σn } × {2} × {2} × · · ·

is basic open, f is continuous.

Claim 8 f is a homeomorphism.
Since 2N is compact and X is Hausdorff, f is a closed map; closed
continuous bijections are homeomorphisms per Theorem 29.

Claim 9 X is homeomorphic to the Cantor ternary set C.

By Theorem 177, C is compact Hausdorff without isolated points and
has a countable base of nonempty clopen sets. Thus C is homeomorphic
to 2N . But X also is homeomorphic to 2N , so X ∼
= C. 

Proof of Corollary 179

164
Proof of Theorem 180
Certainly f [2N ] ⊂ [0, 1] and f is well-defined.
To see that f is continuous, let h~xn i → ~x in 2N . Let ε > 0 and
consider B(f (~x), ε). Pick N ∈ N such that 1/2N < ε/2, and pick
m ∈ N such that πi (~xn ) = πi (~x) for all i 6 N and for all n > m.
Then d(f (~x), f (~xn )) 6 1/2N +1 + 1/2N +1 + · · · 6 1/2N < ε/2 and so
f (~x) ∈ B(f (~x), ε).
To show f is onto, let D = {a/2n : a ∈ Z ∧ n ∈ N} ∩ [0, 1] be the set
of all dyadic rationals in the unit interval. We will show D is dense in
[0, 1], f is onto D, and f is onto [0, 1].
(1) Let ε > 0 and let B = B(x, ε) ⊂ [0, 1]. Note we may pick
n ∈ N such that 1/2n < ε. Furthermore pick a ∈ N such that
a/2n 6 x < a + 1/2n , so that d(x, a/2n ) < 1/2n < ε. It follows
that a/2n ∈ B, so D is dense.

(2) Let a/2n ∈ D. Let a = a1 a2 . . . an be the binary representation

n−1 n−2
of
Pna, thus ia = a1 /2 n + a2 /2 + · · · + an /20 . Then a/2n =
i=1 ai /2 and a/2 = f ((a1 , a2 , . . . , an , 0, 0, . . .)). Thus D ⊂
f [2N ].

(3) By Corollary 179 2N ∼ = C, which is compact by Theorem 177.

Every continuous image of a compact space is again compact—
Theorem 61—so f [2N ] is a compact subspace of the Hausdorff
space [0, 1], hence closed by Theorem 62. Thus [0, 1] = D ⊂ f [2N ].


Proof of Theorem 181

By Theorem 180 there exists a continuous surjection h : C → [0, 1];
2
so h̄ : C2 → [0, 1] defined by hs, ti 7→ hh(s), h(t)i is also a continuous
surjection. Using Corollary 179 we may get a continuous surjection
2
g : C → [0, 1] .
2
Define f : [0, 1] → [0, 1] as follows:
(1) If x ∈ C, take f (x) = g(x).

(2) If x ∈
/ C, there exists a maximal (a, b) ⊂ [0, 1] \ C such that
x ∈ (a, b). Put f (x) = (1 − t)g(a) + tg(b), where 0 < t < 1 and
x = (1 − t)a + tb.

165
9 Proofs

Certainly f is surjective.
Let x ∈ [0, 1]; if x ∈
/ C, then f is continuous at x by linearity. If
x ∈ C, let hxn i be a sequence in [0, 1] converging to x. We have the
following cases:
∞
(1) If hxn in=k ⊂ C for some k ∈ N then
∞ ∞
hf (xn )in=k = hg(xn )in=k → g(x) = f (x),
whence hf (xn )i → f (x).
(2) If hxn i does not have a tail in C, then for each n ∈ N with xn ∈
/C
let (an , bn ) ⊂ [0, 1] \ C be maximal containing xn .
(a) If diam (an , bn ) → 0, define a sequence hx0n i as follows:
(1) If xn ∈ C, take x0n = xn .
/ C, let x0n be the endpoint of (an , bn ) to
(2) If xn ∈
which xn is closest. In a tie, say x0n = bn .
Certainly hx0n i → x and
hf (x0n )i = hg(x0n )i = g(x) = f (x).
Since diam (an , bn ) → 0 we must have d(g(an ), g(bn )) → 0
as well. But this implies d(f (xn ), f (x0n )) → 0, whence
hf (xn )i → f (x).
(b) If diam (an , bn ) 6→ 0, then there is a maximal (a, b) ⊂
[0, 1] \ C containing infinitely many terms of hxn i. If (a, b)
contains a tail of hxn i, then hf (xn )i → f (x) by linearity.
If (a, b) does not contain a tail, we may assume without a
loss of generality that x = b. Then infinitely many terms
converge from the left and right. Let hyn i be the subse-
quence of hxn i consisting of those terms strictly less than
x. Note that hyn i → x, so hf (yn )i → f (x) by linearity.
Let hzn i be the subsequence of hxn i containing those
terms not in hyn i. Note that hzn i → x and satisfies ei-
ther (1) or Case (2a). Thus hf (zn )i → f (x), otherwise x
would be an isolated point of C.
It follows that hf (xn )i → f (x). 

166
Proof of Theorem 182
Let hX, <i = {xn : n < ω} and hY, ≺i = {yn : n < ω} be densely and
linearly ordered sets with no first or last point. Put X0 = {x0 } and Y0 =
{y0 }. Let f0 : X0 → Y0 . Suppose for some n < ω that fn : Xn → Yn is an
order-preserving bijection with {x0 , . . . , xn } ⊂ Xn , {y0 , . . . , yn } ⊂ Yn ,
and both Xn and Yn finite.
0 0 0
If xn+1 ∈ Xn put Xn+1 = Xn , Yn+1 = Yn , and fn+1 = fn . Otherwise,
let A = {x ∈ Xn : x < xn+1 } and B = {x ∈ Xn : xn+1 < x}. If A =
Xn , pick y ∈ Y \ Yn such that y 0 ≺ y for all y 0 ∈ Yn . If B = Xn , pick
y ∈ Y \ Yn such that y precy 0 for all y 0 ∈ Yn . Otherwise, put a = max A
0
and b = min B and pick y ∈ (fn (a), fn (b)). Set Xn+1 = Xn ∪ {xn+1 }
0 0 0 0 0
and Yn+1 = Yn ∪ {y}, and define fn+1 : Xn+1 → Yn+1 so that fn+1 
0
Xn = fn and fn+1 (xn+1 ) = y.
0 0 0 0
If yn+1 ∈ Yn+1 put Xn+1 = Xn+1 , Yn+1 = Yn+1 , and fn+1 = fn+1 .
0 0
Otherwise, let A = {y ∈ Yn+1 : y ≺ yn+1 } and B = {y ∈ Yn+1 : yn+1 ≺ y}.
0 0
If A = Yn+1 , pick x ∈ X \ Xn+1 such that x0 < x for all x0 ∈
0 0 0
Xn+1 . If B = Yn+1 , pick x ∈ X \ Xn+1 such that x < x0 for all
0 0
x ∈ Xn+1 . Otherwise, let a = max A and b = min B and pick x ∈
0
((fn+1 )−1 (a), (fn+1 0
)−1 (b)). Set Xn+1 = Xn+1 0
∪{x} and Yn+1 = Yn+1 0
∪
0 0
{yn+1 }, and define fn+1 : Xn+1 → Yn+1 so that fn+1  Xn+1 = fn+1
and fn+1 (x) = yn+1 .
Note that fn+1 : Xn+1 → Yn+1 is an order-preserving bijection with
{x0 , . . . , xn+1 } ⊂ S Xn+1 , {y0 , . . . , yn+1 } ⊂ Yn+1 , and both Xn+1 and
Yn+1 finite. Then n<ω fn is an order-preserving bijection from X to
Y. 
Proof of Theorem 183
Let X be a connected linearly ordered space. Let x, y ∈ X with x 6= y.
Without loss of generality we may assume that x < y. If (x, y) = ∅,
then X = (−∞, y) ∪ (x, ∞), implying X is not connected. It follows
that (x, y) 6= ∅, whence X is densely ordered.
Now, let A be a bounded set and suppose on the contrary that A
has no least upper bound. Then for all x ∈ A there exists some y ∈ A
with x < y. Also, for each upper bound b ofSA there exists another
upperSbound b0 such that b0 < b. Take U = {(−∞, y) : y ∈ A} and
V = {(b, ∞) : b is an upper bound of A}. Then X = U ∪ V with
U, V open and disjoint, a contradiction. It follows that every bounded

167
9 Proofs

subset of X has a least upper bound.

For the other direction, let X be a linearly ordered non-connected
space in which every bounded subset has a least upper bound. Let
U, V ⊂ X be disjoint, nonempty, and open such that X = U ∪ V .
Without loss of generality we may assume there exists x ∈ U and
y ∈ V such that x < y. Put U 0 = (−∞, y) ∩ U and let b be a least
upper bound of U 0 . Note that either b ∈ U 0 or b ∈ V .
If b ∈ U 0 , then there exists (a, c) ⊂ X with b ∈ (a, c) ⊂ U 0 . Since
(b, c) ∩ V = ∅ and (b, c) ∩ U = ∅, it must be the case that (b, c) = ∅.
Similarly, if b ∈ V there exists (a, c) ⊂ X with b ∈ (a, c) ⊂ V . Now
(a, b) ∩ U 0 = ∅ and (a, b) ∩ V = ∅, since b is a least upper bound, so
(a, b) = ∅. It follows that X is not densely ordered. 

Proof of Theorem 184

Let X be a separable, connected, and linearly ordered space with no
first or last point. Let D be a countable dense subset of X. Observe that
D is densely ordered and has no first or last point. By Theorem 182,
there exists an order-preserving bijection g : D → Q.
For each x ∈ X put Lx = {d ∈ D : d < x}. Define f : X → R so that
x 7→ sup g[Lx ]. Since g[Lx ] is bounded above, sup g[Lx ] exists, so f is
well-defined. Let x1 , x2 ∈ X with x1 6= x2 . Without loss of generality
we may assume that x1 < x2 . Since X is densely ordered and D is
dense, there exist d1 , d2 ∈ D such that x1 < d1 < d2 < x2 . Then

sup g[Lx1 ] 6 g(d1 ) < g(d2 ) 6 sup g[Lx2 ],

so f (x1 ) < f (x2 ). Thus f is one-to-one and order-preserving.

Let y ∈ R and put L = {q ∈ Q : q < y}. Then g −1 [L] ⊂ D is bounded
above, so x = sup g −1 [L] ∈ X exists. But Lx = g −1 [L], so f (x) = y. It
follows that f is onto. 

Proof of Corollary 185

Let X be a separable compact connected linearly ordered space. First
we show that X has a first and last point. Well, if X has neither then
X ∼= R by Theorem 184, so that X is not compact, a contradiction.
Without loss of generality, suppose x0 is a first point of X. If X has no
last point, then the collection {[x0 , x) : x ∈ X} is an open cover of X,

168
so it has a finite subcover. So there exists x ∈ X such that [x0 , x) = X,
a contradiction. Thus X has a last point x1 .
Now X\{x0 , x1 } has no first or last point. (For instance, if it has a first
point x2 then (x0 , x2 ) = ∅. But this would contradict Theorem 183.)
By Theorem 184, (x0 , x1 ) ∼ =R∼ = (0, 1); hence X ∼ = [0, 1]. 

Proof of Lemma 186 Q

Since X is T31/2 , it may be embedded into some product i∈I Ri
of real lines. For instance, the map from X to RC(X) given by x 7→
hf (x)i
Q f ∈C(X) is an embedding. We therefore consider X as a subspace
of i∈I Ri . Let
\ <ω
B={ πi−1 [(ci , di )] ∩ X : F ∈ [I] ∧ ∀i ∈ F (ci < di ∈ Ri \ πi [X])}.
i∈F

If F ⊂ I and ci < di ∈ Ri \ πi [X] for each i ∈ F then

\ \
πi−1 [(ci , di )] ∩ X = πi−1 [[ci , di ]] ∩ X,
i∈F i∈F

so that the sets of B are clopen in X.

We now show B is a base for X. Suppose U is an open subset in
X and hxi i ∈ U ; note that hxi i here is some point ~x in the product
space, and not a sequence of points in U . We find a clopenQsubset of X
containing hxi i, contained in U . There exists an open U 0 ⊂ T i∈I Ri such
that U 0 ∩ X = U . There exists a basic product open set i∈F πi−1 [Ui0 ]
containing hxi i and contained in U (here F is some finite subset of I
and each Ui0 is open in Ri ). For each i ∈ F there exists ai , bi ∈ Ri such
that xi ∈ (ai , bi ) ⊂ Ui0 .
Since |X| < |R|, πi [X] does not contain an interval for any i ∈ I. So
for any i ∈ I and a < b ∈ Ri , there exists r ∈ (a, b) ∩ Ri \ πi [X]. In
particular, for each i ∈ F there exist ci < di ∈ (ai , bi ) ∩ Ri \ πi [X] with
xi ∈ (ci , di ). We have
\ \
hxi i ∈ πi−1 [(ci , di )] ∩ X ⊂ πi−1 [(ai , bi )] ∩ X
i∈F i∈F
\
⊂ πi−1 [Ui0 ] ∩ X ⊂ U 0 ∩ X = U.
i∈F 

169
9 Proofs

Proof of Lemma 187

Let X be a space and let A and B be bases thereof with |A| = κ. Fix
A ∈ A. Let B(A) = {B ∈ B : B ⊂ A} and let

A(A) = {A0 ∈ A : ∃B ∈ B(A)[A0 ⊂ B]}.

S
Since A is open and B is a base, we must have B(A) = A. Similarly,
A(A) = A. Finally, for each A0 ∈ A(A), let B(A0 ) ∈ B(A) such that
S
A ⊂ B(A0 ). Put C(A) = {B(A0 ) : A0 ∈ A(A)}. Since
0

[ [ [
A(A) ⊂ C(A) ⊂ B(A),
S
we have C(A)S = A. Moreover, |C(A)| 6 κ.
Let C = A∈A C(A). Certainly C ⊂ B is a base. By Lemma 202(a),
|C| 6 κ.
If κ is the least cardinal of a base for X, naturally |C| = κ. 

Proof of Theorem 188

Let X be a countable, regular and first-countable space with no iso-
lated points. Observe that X is second-countable, hence metrizable and
completely regular. By Lemma 186 and Lemma 187, X has a countable
base B = {Bn : n < ω} of clopen sets.

Claim A. X embeds into 2ω .

Claim B. X is homeomorphic to a dense subset of C.

Claim C. Let f ∈ 2ω . Then h : 2ω → 2ω defined so that g 7→ f + g is a

homeomorphism.

Claim D. Let E ⊂ C be the collection of endpoints of all Iσ , where

σ ∈ 2<ω . Then X is homeomorphic to a subset of C disjoint from E.

Claim E. Let D ⊂ C be dense and disjoint from E. Then D is densely

ordered and has no first or last point.

By the above claims, let X be densely embedded in C such that

X ∩ E = ∅. By Theorem 182 there exists an order-preserving bijection
from X to Q, so X ∼= Q.

170
Proof of Claim A. For each n < ω, define fn : X → 2 so that fn [Bn ] =
{0} and fn [X \ Bn ] = {1}. Certainly fn is continuous for each n < ω.
Put F = {fn : n < ω}. and define eF : X → 2ω so that x 7→ hfn (x)i.
Note that πn ◦ eF = fn for each n < ω, so eF is continuous.
Let x ∈ X and H ⊂ X closed not containing x. Then there exists
n < ω such that x ∈ Bn ⊂ X \ H. It follows that fn (x) = 0 and
fn [H] = {1}, so F separates points from closed sets. By Theorem 145,
eF is a homeomorphic embedding into 2ω .
Note that we could have defined each fn to map X onto the endpoints
of [0, 1] to more closely follow Theorem 145. 

Proof of Claim B. Since 2ω ∼

= C, there exists an embedding f : X → C
by Proof C. Observe that f [X] ⊂ C is compact Hausdorff and has a
countable base of clopen sets. Since f [X] has no isolated points, neither
does f [X]. By Theorem 178, there is a homeomorphism h : f [X] → C.
Since f [X] is dense in f [X], we have h[f [X]] dense in C. 

Proof of Claim C. Certainly h is a bijection. Let U ⊂ 2ω be a basic

open set. Then
Y Y Y
U= {0} × {1} × 2,
i∈F0 i∈F1 i∈F
/ 0 ∪F1

where F0 , F1 ⊂ ω are finite and disjoint. Thus

Y Y Y Y Y
h[U ] = {0} × {1} × {1} × {0} × 2,
i∈F00 i∈F000 i∈F10 i∈F100 i∈F
/ 0 ∪F1

where Fj0 = {i ∈ Fj : f (i) = 0} and Fj00 = {i ∈ Fj : f (i) = 1} for j =

0, 1. It follows that h is open; by the same argument h−1 is open. 

Proof of Claim D. Let X be (densely) embedded in C and let h wit-

ness C =∼ 2ω . Pick a ∈ 2ω \ {f + g : f ∈ h[X] ∧ g ∈ h[E]}. Put Xa =
{f + a : f ∈ h[X]}. Then h[X] ∼ = Xa and h[E] ∩ Xa = ∅. It follows
that X ∼= h−1
[Xa ] and E ∩ h−1
[Xa ] = ∅. 

Proof of Claim E. Suppose on the contrary that D has a first point, say
d. Since D ∩ E = ∅, we must have d > 0; thus (−∞, d) is a nonempty

171
9 Proofs

open set of C missing D, a contradiction. By a similar argument, D

has no last point.
Now let d1 , d2 ∈ D with d1 < d2 . Then there exists σ ∈ 2<ω such
that d1 ∈ Iσ but d2 ∈ / Iσ . It follows that (d1 , d2 ) contains the right
endpoint of Iσ . Since (d1 , d2 ) is nonempty, (d1 , d2 ) ∩ D 6= ∅, whence D
is densely ordered. 

Proof of Corollary 189

Proof of Theorem 190

First, let X be a compact linearly ordered space. Suppose on the con-
trary that there exists some Y ⊂ X without a least upper bound. Put
U = {(−∞, y) : y ∈ Y } and V = {(x, ∞) : x ∈ X ∧ ∀y ∈ Y (y < x)}.
Certainly U ∪ V is an open cover of X, so there exist finite subcol-
lections U 0 and V 0 of U and V, respectively, such that U 0 ∪ V 0 covers
X.
If Y 6= ∅ then U 0 6= ∅, so we may assume

U 0 = {(−∞, y1 ), . . . , (−∞, yn )}

with y1 < · · · < yn . It follows that Y ⊂ (−∞, yn ) which implies yn ∈

/ Y,
a contradiction. On the other hand, if Y = ∅ we may assume V 0 =
{(x1 , ∞), . . . , (xm , ∞)} with x1 < · · · < xm . Then X ⊂ (x1 , ∞) so
x1 ∈/ X, a contradiction. It follows that every Y ⊂ X must have a least
upper bound and, by a similar argument, a greatest lower bound.
Now let X be a linearly ordered space in which every subset has a
least upper bound and a greatest lower bound. Observe that X has a
least element, say 0, and a greatest element, say 1.
Let U be an open cover of X and let A be the collection of all x ∈ X
such that [0, x] can be covered by finitely many members of U. Put
a = sup A. We claim that a ∈ A and a = 1.
That a ∈ A is immediate, for there must be some x ∈ A and U ∈ U
such that (x, a] ⊂ U . Suppose on the contrary that a < 1. If a has no
immediate successor, there exists U ∈ U such that (a, y] ⊂ U for some
y > a. But this means [0, y] can be covered by finitely many members

172
of U, so a must have an immediate successor, say b. But some member
of U must contain b, so [0, b] can be covered by finitely many members
of U, contradicting the definition of a. It follows that a = 1, whence X
is compact. 
Proof of Theorem 191
Let X̂ = h{A ⊂ X : A is left-closed}, ⊂i. By Theorem 190, it suffices
to show that every subset of X̂ has a least upper bound and a greatest
lower bound. Let A ⊂ X̂.
S
Claim. A is the least upper bound of A.
S
Proof of Claim. Clearly A is an upper bound of A. NoteS b < a ∈
0
S
S A → b < a ∈ A for some A ∈ A, which implies b ∈ A ⊂ A. Thus
A is left-closed
S and a member of X̂.
Let B ( A be left-closed. We show that B is not an upper bound
of A. Suppose S it were. Then B ⊃ A for every A ∈ A. But B is closed,
so then B ⊃ A, a contradiction. 
T
Claim. A is the greatest lower bound of A.
T
Proof of Claim. If b < a ∈ A, then b < a ∈ A for all A ∈ A, so b ∈ A
T T
for all A ∈TA and b ∈ A; thus A ∈ X̂. Certainly A is a lower bound
for A. If ATis a proper subset of a left-closed set B, then there exists
b ∈ B with A < b. If B T were a lower bound for A, then b ∈ B ⊂ A
for every A, whence b ∈ A, contradiction. 

Claim. X is dense in X̂.

Proof of Claim. Let (A, B) ⊂ X̂ be nonempty. There is C ∈ X̂ such

that A ⊂ C ⊂ B. There exists x ∈ C with A < x such that A ⊂
{a ∈ X : a 6 x} ⊂ C ⊂ B. 

Proof of Theorem 192

TODO: Find the corollaries that prove this. 

Proof of Theorem 193

(a) Since C and D are unbounded, there exists an increasing sequence
hxi i with x2i ∈ C and x2i+1 ∈ D for each i < ω. Then x = sup hxi i

173
9 Proofs

is a limit point of both C and D. Since C and D are closed, we have

x ∈ C ∩ D.

(b) Using the unboundedness of the sets Cn , we construct a strictly

increasing sequence hxi i of points in ω1 as follows. Let x0 ∈ C0 and
x1 ∈ C1 . Let x2 ∈ C0 , x3 ∈ C1 , and x4 ∈ C2 . Let x5 ∈ C0 , x6 ∈ C1 ,
x7 ∈ C2 , and x8 ∈ C3 . Continue in this manner. Let x = sup hxi i. Then
for each n < ω, hxi i has a subsequence T in Cn which limits to x. By
closedness of the sets Cn , we have x ∈ n<ω Cn . 

Proof of Theorem 194

Let H ⊂ ω1 be the of limit ordinals in ω1 . If α ∈ ω1 \ H then α = 0 or
α = β + 1 for some β < ω1 . So α is isolated. Claim any open superset
of H contains all but ω-many points of ω1 . Well, let O ⊂ ω1 be open
containing H. Then H ∩ (ω1 \ O) = ∅ and ω1 \ O is closed, so it can’t
be unbounded and miss H. Thus ω1 \ O is countable. Let {On : n < ω}
be a collection of open sets, each containing H. ForT each n < ω there
exists αn < ω1 such thatTOn ⊃ [αn , ω1 ). Then n<ω On ⊃ [α, ω1 )
where α = supn<ω αn . So n<ω On = 6 H. So H is not Gδ and hence ω1
is not perfectly normal. 
Proof of Theorem 195
Let C be a closed unbounded set in ω1 . Because C is well-ordered, there
is a least c0 ∈ C. Send c0 to 0. Similarly, there is a least c1 ∈ C \ {c0 };
send c1 to 1. Now, if cβ has been defined for all β < α, where α < ω1 ,
put cα = min (C \ {cβ : β < α}) and send cα to α.
Let f be the map just constructed. Certainly f is bijective. Let
(β, α] ⊂ ω1 be basic open. We claim that f −1 [(β, α]] = (cβ , cα ] ∩
C is basic open in C. Note that if α is a limit ordinal, then cα =
min (C \ {cβ : β < α}) = sup {cβ : β < α} is a limit of C ∩ (0, cα ). If α
is not a limit, (cα−1 , cα ] ∩ C = {cα } = f −1 [{α}]. The claim follows. 

Proof of Theorem 196

Suppose α is a limit of C not in C. Suppose β < α. Then (β, α]∩C =6 ∅.
There is γ ∈ (β, α) ∩ C. Then f (β) < γ < α. Therefore α ∈ C, a
contradiction, so C is closed.
Let γ < ω1 . We need to show there is an α ∈ C such that α > γ. Pick

174
α0 > γ. Pick α1 > sup {f (β) : β < α0 }. Pick α2 > sup {f (β) : β < α1 }.
Having already picked αn , pick αn+1 = {f (β) : β < αn }. Let α =
sup {αn : n < ω}. Suppose β < α. Then β < αn for some n, hence
f (β) < αn+1 < α. So α ∈ C. 

Proof of Theorem 197

Let f be pressing down. Suppose |f −1 (α)| S 6 ω for all α. Define
g : ω1 → ω1 such that for all α, g(α) > sup β6α f −1 (β). Let C be as
in Theorem 196: C = {α : ∀β < α(g(β) < α)}. There is an α ∈ C ∩ S.
Let β = f (α). Then β < α, but g(β) > α, a contradiction to α ∈ C. 

Proof of Proposition A
Suppose not. For every S ⊂ ω1 , either S or ω1 \ S is nonstationary,
implying either S or ω1 \ S contains a club. We will show why this is
not possible.
Identify ω1 with a subset of [0, 1]. Either [0, 1/2] or [1/2, 1] contains a
club. Without loss of generality we may assume [1/2, 1] contains a club.
Consider [1/2, 3/4] and [3/4, 1]. One of them must contain a club: say
[3/4, 1] is nonstationary. Then [0, 1/2] ∪ [3/4, 1] is nonstationary, and
[1/2, 3/4] contains a club. Similarly, [1/2, 5/8] or [5/8, 3/4] contains a
club. Inductively, get I1 ⊃ I2 ⊃ I3 ⊃ · · · suchTthat for all n, In contains
aTclub with diam In = 1/2n−1 . Therefore In contains a club, but
| In | = 1. 

Proof of Theorem 198

Let U = {[0, α] ∩ S : α < ω1 } be an open cover of S and let V be an
open refinement thereof. Let C be the set of all points α in ω1 that are
limit points of S. Then C is closed unbounded. Let S 0 = C ∩ S; it is
easily check that S 0 is stationary.
For each α ∈ S 0 , there exists V ∈ V and f (α) ∈ S such that f (α) < α
and [f (α), α]∩S ⊂ V . By the Pressing Down Lemma, there exists β ∈ S
such that f (α) = β for uncountably many α ∈ S 0 .
It suffices to show that β is a member of more than finitely many
members of V. To that end, suppose on the contrary that β is a member
of only finitelySmany members of V, say Vα1 , . . . , Vαn . Let γ < ω1 such
n
that γ > sup i=1 Vαi . Then there exists δ ∈ S 0 such that δ > γ and
f (δ) = β. It follows that [β, δ] ∩ S 6⊂ Vαi for i = 1, . . . , n. Thus β ∈ Vδ

175
9 Proofs

with Vδ 6= Vαi for all 1 6 i 6 n, a contradiction.

Since β is contained in more than finitely many members of V, V is
not point-finite. It follows that U has no locally-finite open refinement,
whence S is not paracompact.
The same proof shows that there is no point-finite or even point-
countable open refinement of U, and hence a stationary set S is not
even metacompact or metalindelöf. 

Proof of Theorem 199

Proof of Theorem 200

Let S ⊂ ω1 . Metrizability of S implies its paracompactness, and by
Theorem 198 paracompactness of S implies S is nonstationary. It re-
mains to prove that if S is nonstationary, then it is metrizable. If S is
nonstationary, thereSexists a club C such that C ∩ S = ∅. By Theo-
rem 199, ω1 \ C = α∈I Mα with the Mα convex open disjoint. As a
countable subspace of a regular first-countable space, each Mα is regular
and second-countable. So each Mα is metrizable by a metric dα with
max 1. It is easily check that the function d given by d(x, y) = dα (x, y)
if x, y ∈ Mα , d(x, y) = 2 if x inMα and y ∈ Mβ with α 6= β, is a metric
on ω1 \ C which induces the subspace topology on ω1 \ C. 

Proof of Example 201

Proof of Lemma 202

(a) This is essentially equivalent to κ = |κ × κ|. Suppose it is true for
all cardinals less than κ. Define a well-order ≺ on κ × κ as follows. Let
hα, βi and hγ, δi be distinct points in κ × κ. Define hα, βi ≺ hγ, δi if and
only if max {α, β} < max {γ, δ}, or max α, β = max {γ, δ} and hα, βi is
less than hγ, δi is the lexicographic order. It is not difficult to check that
this is a well-ordering. Let hγ, δi ∈ κ × κ and suppose max {γ, δ} = µ.
The set of predecessors of hγ, δi has cardinality at most |µ × µ| = |µ|
by the induction hypothesis, which is less than κ. But κ is the only
ordinal such that cardinality of every predecessor is less than κ, while

176
the cardinality of the whole set is not. Hence |κ × κ| = κ.
For each αS < κ, let |Aα | 6 κ. Let fα : κ → Aα be S onto. Then
F : κ × κ → α<κ Aα by F (α, β) = fα (β) is onto. Hence | α<κ Aα | 6
|κ × κ| 6 κ.

(b) Suppose κ = λ+ , but that µ = cf κ < κ. Note that µ 6 λ. Let

f : µ → κ be unbounded in κ. Then {f (α) : α < µ} is a collection of
6 λ-many sets each of cardinality 6 λ, so its union has cardinality 6 λ.
But its union is κ, contradiction.

(c) Suppose µ = cf λ is not regular. Then ν = cf µ < µ. Let f : ν → µ

be cofinal and nondecreasing, and g : µ → λ cofinal and nondecreasing.
It is easy to check that g ◦ f : ν → λ is cofinal, so cf λ 6 ν, contradiction.


Proof of Theorem 203

Let κ be an infinite cardinal.
(a → b) Suppose A ⊂ κ with |A| < κ. Let λ = |A| and enumer-
ate A = {aγ : γ < λ}. Define f : λ → κ by γ 7→ αγ . Then sup A =
supγ<λ f (γ) < κ.

(b → c) Suppose |Aγ | < κ for γ < λ < κ. Without loss of gener-

ality, we may assume Aγ ⊂ κ for each γ < λ and that the Aγ are
pairwise-disjoint. By assumption, sup Aγ < κ for each γ < λ. Then
supγ<λ (sup Aγ ) < κ, so
[
| Aγ | 6 |sup(sup Aγ )| 6 sup(sup Aγ ) < κ.
γ<λ γ<λ
γ<λ

(c → a) Suppose λ < κ and let f : λ → κ. For γ < λ, defineSAγ = f (γ).

(Recall that an ordinal is the set of its predecessors.) Then | γ<λ Aγ | <
κ, implying f is bounded in κ. 

Proof of Theorem 204

177
9 Proofs

Proof of Theorem 205

Suppose κ is regular and S ⊂ κ is stationary. Let U = {[0, α] ∩ S : α ∈ S},
and suppose V is a locally finite open refinement of U. For each α ∈ S
there exists βα < α such that [βα , α] ∩ S ⊂ Vα ∈ V. By Theorem 204(c)
there exists β with βα = β for κ-many α ∈ S. On the other hand, only
finitely many members S of V contain β. There exists α ∈ S such that
βα = β and α > sup {V ∈ V : β ∈ V }.SBut there exists V 0 ∈ V such
that [β, α] ⊂ V 0 , contradicting α > sup {V ∈ V : β ∈ V }. 

Proof of Theorem 206

Suppose X is connected and U is an open cover of X. Let U(U ) denote
those V ∈ U for which there is a finite linked chain between U and V .
S
Claim. O = U(U ) is connected.

Proof of Claim. Clearly O is open. We first show it is also closed. Sup-

pose x is a limit point of O. Since U is an open cover, there exists
Ux ∈ U such that x ∈ Ux . Then Ux ∩ O 6= ∅, so there exists V ∈ U(U )
such taht Ux ∩ V 6= ∅. Thus we may extend a finite linked chain be-
tween U and V to a finite linked chain between U and Ux , so that
x ∈ Ux ⊂ O. 

Since X is connected and O is clopen, we must have O = X. Let

U1 , U2 ∈ U. There exist V1 , V2 ∈ U(U )—note that U(U ) covers X—such
that V1 ∩ U1 and V2 ∩ U2 are both nonempty. Extend chains between
V1 and U , and V2 and U , to a chain between U1 and U2 .
If U and V are nonempty disjoint clopen sets U ∪ V = X, then
U = {U, V } is an open cover of X which is not connected. 

Proof of Theorem 207

Proof of Lemma 208

(a → b) Trivial.

(b → c) Trivial.

(c → a) Trivial.

178
(c → d) Let {xα : α < ω1 } ⊂ X such that xα ∈ / {xβ : β > α} for all
α ∈ ω1 . Then for each α ∈ ω1 there exists an open neighborhood Uα
of xα such that Uα ∩S{xβ : β > α} = ∅. Take U = {Uα : α < ω1 }. If
V ⊂ U is countable,S V mayS contain only countably many members
of {xα : α < ω}, so V 6= U.
S
S → c) Let U be a collection of open subsets of X such that V 6=
(d
U for all countable V ⊂ U. Pick U0 ∈ U and x0 ∈ U0 . Let 0 < α < ω1
S suppose Uβ ∈ U and xβ ∈ Uβ have
and S been defined for all β < α. Since
{UβS: β < α} is a proper subset of U, there
S exists Uα ∈ U such that
Uα \ {Uβ : β < α} = 6 ∅. Pick xα ∈ Uα \ {Uβ : β < α} and observe
that for each β < α, Uβ is a neighborhood of xβ not containing xα . It
follows that {xα : α < ω1 } is a subset of X such that xα ∈
/ {xβ : β > α}
for all α ∈ ω1 . 

Proof of Theorem 209

Proof of Theorem 210

(a) Let X be a ccc nonseparable linearly ordered space. Define x ∼ y
if and only if the interval from x to y is separable. Note that equiva-
lence classes are convex. We claim that the set S = {[x] : x ∈ X} of
equivalence classes with the natural order, denoted by ≺, satisfies the
desired conditions.
It follows easily that S is ccc because X is. To see that S is densely
ordered, suppose ([x], [y]) is empty. Then (x, y) is separable and [x] =
[y], a contradiction.
To complete the proof, we need to show that S has no separable open
intervals. First we show that every equivalence class [x] is separable.
It follows from the ccc of X that cf [x] is countable, and similarly
so is the coinitiality of [x]. Let us assume [x] has a greatest point
but no least point; other cases can be handled similarly. Let hyn i be
a decreasing
S sequence of points of [x] which is coinitial in [x]. Then
[x] = n<ω (yn , z], where z is the greatest point of [x]. Since each (yn , z)
is separable, so is [x]. Now suppose ([x], [y]) is separable, say with dense
set D = {[zi ] : i < ω}. For each i, let Ei be a dense (in X) subset of

179
9 Proofs

S
[zi ]. It follows that (x, y) ∩ [ i<ω Ei ] is dense in (x, y), hence [x] = [y],
a contradiction.

(b) Let X be a ccc nonseparable linearly ordered space. By 210(a),

we may also assume X is densely ordered and that no nonempty open
interval in X is separable. Let X̂ be the compactification of X as in
Theorem 191.
We claim that X̂ is densely ordered. Suppose not. Let a, b ∈ X̂
with (a, b) = ∅. If a ∈ X̂ \ X and b ∈ X, then (−∞, a) ∩ X is a left-
closed subset of X. Note that in X, every neighborhood of b contains
an interval (c, b], where c < b. But then c ∈ (−∞, a) and since X is
densely ordered, (c, b) ∩ (= ∞, a) 6= ∅. So b is a limit point of (−∞, a),
a contradiction to it being left-closed.
If a, b ∈ X̂ \ X, then (−∞, a) ∩ X = (−∞, b) ∩ X, so these subsets
are identified with the same element of X̂, forcing a = b. If a ∈ X and
b ∈ X̂ \ X, then (−∞, a] ∩ X = (−∞, b) ∩ X, giving a = b as above. So
X̂ is densely ordered.
By Theorem 190, every subset of X̂ has a least upper bound. Because
X̂ is densely ordered and every bounded subset of X̂ has a least upper
bound, X̂ is connected by Theorem 183. Throw out the first and last
points of X̂ to get a space X 0 that is connected, ccc, and has a subspace
that is not separable. It follows that X 0 is not homeomorphic to R and
is thus a Suslin line. 

Proof of Theorem
T 211
Since C = n∈N Cn and each Cn is a finite union of closed sets, then
Cn is closed and thus C is closed. C is compact since it is a closed
subset of the compact space I. C is metrizable since it is a subspace of
the metrizable space R.

Claim 1 : Iσ ∩ C is clopen. Let B = {Iσ ∩ C : σ ∈ 2<ω }. For a given

σ, Iσ ∩ C is closed since it is the intersection of two closed subsets. Also,
Cn \ Iσ is closed since Cn is a finite union of disjoint closed intervals
one of which is Iσ , so C \ (C ∩ Iσ ) = C ∩ (Cn \ Iσ ) is closed, which gives
us C ∩ Iσ is open.

180
Claim 2: B is a base for C, and C has empty interior in R. Let
U be any open set in C and let x ∈ U . There some ε > 0 such that
(x − ε, x + ε) ∩ C ⊂ U . Pick some n > 1 with 1/3n < ε. Since x ∈ Cn ,
x ∈ Iσ for some σ of length n. Notice diam Iσ = 1/3n . So x ∈ C ∩ Iσ ⊂
C ∩ (x − ε, x + ε) ⊂ U .
Now if U were open in R and contained in C, then as above we would
have Iσ ⊂ U ⊂ C for some σ. But the middle third of Iσ is disjoint
from C, contradiction. Thus C has empty interior in R.

Claim 3: C is uncountable. For each f : ω → 2, there is exactly one

point xf in the intersection of the branch coded by f ; that is, {xf } =
ω
T
n<ω If n ∩C. Suppose f, g ∈ 2 and f =6 g. Let n be minimal such that
f (n) 6= g(n). Then xf ∈ If n+1 , xg ∈ Ign+1 , and If n+1 ∩ Ign+1 = ∅.
Hence xf 6= xg . Thus the map f 7→ xf is one-to-one, and so C is
uncountable (in fact, it has the same cardinality as the real line).

Claim 4: C has no isolated points. Let σ ∈ 2n . If f and g are distinct

members of 2ω such that f  n = g  n = σ, then xf and xg are distinct
points in Iσ ∩ C. So every base element has at least two points, hence
C has no isolated points. 

Proof of Theorem 212

Let X be a space with a base B = {Bσ : σ ∈ 2<ω } of clopen sets
isomorphic to the Cantor tree such that the intersection of any branch
ω
is a single point.
T For each c ∈ C, there T is a unique σc ∈ 2 such
that {c} = n<ω Iσc n . By assumption n<ω Bσc n contains a single
element, say xc . Define h : C → X by c 7→ xc .

Claim 1: h is surjective. Suppose not. Then there exists x ∈ X

and σ ∈ 2<ω such that x ∈ Bσ but x ∈ / Bσa0 ∪ Bσa1 . Then Bσ \
(Bσa0 ∪ Bσa1 ) is an open neighborhood of x not containing any base
elements, a contradiction.

181
9 Proofs

Claim 2: h is injective. Suppose note. Then there exists σ, τ ∈ 2ω

and x ∈ X such that σ 6= τ but
\ \
Bσn = {x} = Bτ n .
n<ω n<ω

Let k < ω such that σ  k = τ  k but σ(k) 6= τ (k); put υ = σ  k. Then

x ∈ Bυa0 ∩ Bυa1 , so there exists u ∈ 2<ω such that x ∈ Bu ⊂ Bυa0 ∩
Bυ∩1 . Thus υa0 and υa1 are initial segments of u, a contradiction.
Proof of Theorem 213

Proof of Theorem 214

(a) Define a Cantor tree of closed subintervals of [0, 1] as follows: J∅ =
[0, 1], and if Jσ = [`, r] has been defined for σ ∈ 2n , let Jσa0 and Jσa1
be [`, m] and [m, r], respectively, where m is the midpoint of [`, r]. Note
that each branch of this tree is a single point, and every point of [0, 1]
is the intersection of some branch. The map that sends the intersection
of a branch of the Cantor tree to the intersection of the corresponding
branch of this tree for [0, 1] is easily seen to be continuous and onto.
It is not a bijection—it can’t be, else it’d be a homeomorphism—but
it is two-to-one on a countable set and one-to-one otherwise.

(b) Let f be a continuous function from C onto [0, 1]. For ~x ∈ Cω , let
ω
f (~x) = hf (xn )i ∈ [0, 1] . Then f is clearly continuous and onto. Since
ω ∼
C = C, we are done.

(c) If K is a closed subset of C, it is easily seen that K × C is compact,

has no isolated points, and has a countable base of clopen sets. Hence
K × C is homeomorphic to C. The projection onto the first coordinate
is a continuous map onto K.

(d) Let M be any compact metric space. Then M is homeomorphic to a

ω
(compact, hence closed) subset N of the Hilbert cube. Let f : C → [0, 1]
−1
be continuous and onto. Let K = f [N ]. Then K is a closed subset
of C, so by 214(c), there is a continuous surjection g : C → K. Then
f ◦ g : C → N is continuous and onto. Since N ∼ = M we are done. 

182
Proof of Exercise 215
Let q1 , q2 , . . . be an enumeration of the rationals with q1 = 0. We define
clopen subsets {Bσ : σ ∈ ω <ω } of P such that
(1) B∅ = P;

(2) for each σ, the collection {Bσan : n < ω} is a clopen partition of

Bσ ; and

(3) if σ has length n, then diam Bσ 6 1/n.

If we do the above, the resulting Bσ ’s will form a base which under ⊃
is a tree isomorphic to hω <ω , 6i. There is one extra thing we need to
be concerned about: how do we know the intersectino of each branch
is a single point? (It could be empty.)
To start, of course we set B∅ = P. The collection {(n, n + 1) ∩ P : n ∈ Z}
is a countable clopen partition of P, so we may let {Bn : n < ω} index
it. As we continue, each Bσ will have the form (a, b) ∩ P for some ra-
tionals a, b. We want to make sure that each rational is included as an
endpoint at some stage. Note that since q1 = 0, q1 is taken care of. We
will make sure qn gets taken care of at or preceding the construction
of the nth level of the tree.
So, suppose Bσ has been defined for all σ of length 6 n. Let σ have
length n. We show how to construct Bσan for n < ω. We have assumed
Bσ = (a, b)∩P for some rationals a, b. Let hak i be a decreasing sequence
of rationals in (a, b) converging to a, and hbk i an increasing sequence
of rationals in (a, b) converging to b. We may assume we make the
choice such that a0 < b0 , and that the intervals (a0 , b0 ), (an+1 , an ), and
(bn , bn+1 ) for n < ω have diameter 6 1/n + 1. We may also assume
that if the nth qn lies between a and b, then qn is one of the an ’s or bn ’s.
Note that if qn was not already an endpoint of some chosen interval,
then it will be taken care of at this stage. Finally, let {Bσan : n < ω}
be a listing of

{(a0 , b0 ) ∩ P} ∪ {(an+1 , an ) ∩ P : n < ω} ∪ {(bn , bn+1 ) ∩ P : n < ω}.

Clearly the Bσ ’s satisfy our requirements above. We need to see

that the intersection of each branch is a singleton. Note that each
branch corresponds to a decreasing sequence I0 ∩ P, I1 ∩ P, . . . where

183
9 Proofs

the endpoints of T
the In ’s are rational and the diameters go to 0. In the
whole real line, n<ω In is a single point, say x. If x is irrational, we
are done. Suppose x is rational. At some stage, say n, x becomes an
endpoint of one of the intervals. So it is either an endpoint
T of In , or an
endpoint of some disjoint interval. In either case, x ∈/ n<ω In , and so
x must be irrational. 

Proof of Theorem 216

(a → b) By Exercise 215.

(b → c) Suppose X has a base B such that hB, ⊃i ∼ = hω <ω , 6i and

the intersection of each branch of hB, ⊃i is a single point. We show X
(1) has a countable base of clopen sets, implying it is separable;
(2) is nowhere locally compact; and
(3) is completely metrizable.

(1) ω <ω is countable.

(2) Let B = {Bσ : σ ∈ ω <ω } be an enumeration of B such that Bσ ⊃
Bτ if and only if τ extends σ. Note that for each σ ∈ ω <ω , {Bσaj : j < ω}
partitions Bσ : each Bσaj is contained in Bσ since σaj extends σ. We
now show {Bσaj : j < ω} covers Bσ . Suppose x ∈ Bσ . If i 6= j < ω, then
the subsets Bσai and Bσaj of Bσ are incomparable under ⊃. Thus Bσ
contains more than one point. Let y ∈ Bσ with y 6= x. Since B is a base,
there exists B ∈ B with x ∈ B and y ∈ / B and there exists Bτ ⊂ B ∩ Bσ
with x ∈ Bτ . Bτ is a proper subset of Bσ containing x, so τ properly
extends σ and x ∈ Bτ n+1 ∈ {Bσaj : j < ω}. Thus {Bσaj : j < ω} cov-
ers Bσ . Finally, if i 6= j < ω we have Bσai ∩ Bσaj = ∅. Otherwise
there exists Bτ ⊂ Bσai ∩ Bσaj . Then τ extends both σai and σaj,
which is impossible.
Now we are ready to show X is nowhere locally compact. Suppose
U is a neighborhood. Then U has nonempty interior. There exists
Bσ ⊂ U ◦ ⊂ U . If U is compact then since Bσ is closed, we have Bσ
is compact. But {Bσaj : j < ω} is an open cover of Bσ with no finite
subcover.
(3) For x, y ∈ X let Lx,y = {i ∈ ω : ∃σ ∈ ω i (x, y ∈ Bσ )}. Define d : X×

184
X → [0, ∞) by

1/max Lx,y
 if Lx,y is bounded and max Lx,y > 0,
d(x, y) = 1 if Lx,y is bounded and max Lx,y = 0,

0 if Lx,y is unbounded.


d is a metric. If x = y, then x and y may not be separated by sets

in B, so Lx,y is unbounded, so d(x, y) = 0. Conversely, if x 6= y then
there exists σ, τ ∈ ω <ω with x ∈ Bσ , y ∈ Bτ , and Bσ ∩ Bτ = ∅. Then
x and y are separated at level max {dom σ, dom τ }, so Lx,y is bounded.
Hence d(x, y) 6= 0.
Trivially, d(x, y) = d(y, x).
The triangle inequality is trivial to see if d(x, y) = 0, so suppose
d(x, y) = 1/n. Then x and z or y and z are contained in disjoint basic
sets at level n + 1 (say x and z). So there is a max level m 6 n at
which x and z are together. Thus d(x, y) = 1/n 6 1/m = d(x, z) 6
d(x, z) + d(y, z).
Now we show that the topology induced by d coincides with the
topology induced by B (the original topology on X.) It is easy to see
that B(x, 1/n) = Bτ , where τ ∈ ω n+1 is such that x ∈ Bτ . So a metric
ball contains a set in B. Conversely, if ∅ 6= σ ∈ ω <ω and x ∈ Bσ , then
B(x, 1/dom σ) = Bσai ⊂ Bσ , with i such that x ∈ Bσai . If σ = ∅ then
Bσ = X so any metric ball is contained in it.

d is complete. Suppose hxn i ∈ X ω is a Cauchy sequence with respect

to d. There exists an increasing sequence hmn i ∈ ω ω such that for each
n, k > mn impies d(xmn , xk ) < 1/2n+1 . The balls B(xmn , 1/2n ), n < ω,
are nested (decreasing). Indeed, if n < ω and x ∈ B(xmn+1 , 1/2n+1 )
then we have

d(x, xmn ) 6 d(x, xmn+1 ) + d(xmn , xmn+1 ) < 1/2n+1 + 1/2n+1 = 1/2n ,

whence x ∈ B(xmn , 1/2n ). Thus the balls B(xmn , 1/2n ) correspond to

a brance in the tree hB, ⊃i, which intersects to a point x. Let ε > 0
and let n ∈ N such that 2/2n < ε. For k > mn we have

d(x, xk ) 6 d(xmn , x) + d(xmn , xk ) < 1/2n + 1/2n+1 < 2/2n < ε,

185
9 Proofs

so x is the limit of hxn i.

(c → a) Suppose X is nowhere locally compact, completely metrizable,

and has a countable base B of clopen sets.

Claim. For every nonempty open U ⊂ X and n ∈ N there exists a

partition of U into ω-many nonempty clopen sets, each with diameter
less than 1/n.

Let U ⊂ X nonempty open and n ∈ N. Using the fact that X is nowhere

locally compact, Un = {B ∈ B : diam B < 1/n ∧ B ⊂ U } is an open
cover of U with no finite subcover. Enumerate Un = S{B0 , B1 , . . .}. After
throwing away the empty elements in Un∗ = {Bn \ i<n Bi : n < ω}, we
have the desired partition.
Using the claim, recursively construct a tree of clopen sets isomorphic
to hω <ω , 6i. Each branch of this tree intersects to at most one point
since its nodes have diameters shrinking to 0. The intersection of each
branch is nonempty: create a Cauchy sequence by selecting a sequence
up the branch. It converges by completeness, to a point which must be
in the intersection of the branch. Construct a homeomorphism between
X and the space of irrationals by mapping branches to branches, as in
Proof 213. 

Proof of Corollary 217

Proof of Theorem 218

Let T be a tree of height ω such that Lα is finite for each α < ω. For
each t ∈ T , put St = {s ∈ T : t < s}. Since T has height ω, there must
be infinitely many nodes in T .
Pick b0 ∈ L0 so that Sb0 is infinite. This is possible since L0 is finite.
Suppose for some n < ω that b0 < · · · < bn have been defined so that
bi ∈ Li and Sbn is infinite. Pick bn+1 ∈ Sbn ∩ Ln+1 such that Sbn+1 is
infinite. This is possible since Sbn ∩ Ln+1 is finite.
Certainly b = {bn : n < ω} is an infinite branch of T . 

186
Proof of Theorem 219
We construct the tree T by induction, along with a function θ : T → Q
which is increasing in the sense that x < y ∈ T implies θ(x) < θ(y).
This function will guarantee that there are no branches all the way
through the tree, and will also tell us some things about the structure
of T . (See the comment after the proof.)
For each α < ω1 , we let Tα denote the set of nodes of T of height < α.
To start, let Tω = ω <ω . Let θ(∅) = 0. The first level L1 of Tω —that is,
the set of immediate successors of ∅—is countable infinite, so we can
let θ  L1 be a bijection from L1 to the positive rationals. Given t ∈ L1
with θ(t) = qt , let θ send the immediate successors of t one-to-one and
onto the set of rationals > qt . And so on.
From here on, we denote θ(t) by qt .
Next we define Tω+ω as follows. For each s ∈ Tω and q ∈ Q with q >
qs , choose a branch b(s, q) of Tω with s ∈ b(s, q) and q = sup {qt : t ∈ b(s, q)}.
This is possible by the construction of θ on Tω . Now let {T (s, q) : s ∈ Tω , q > qs }
be a collection of copies of ω <ω disjoint from each other and from
T
Sω ; we’ll put T (s, q) above the nodes of b(s, q). let Tω+ω = Tω ∪
<ω
s∈Tω ,q>qs T (s, q). Give the copies of ω their usual order, and if
x ∈ Tω and y ∈ T (s, q), define x < y if and only if x ∈ b(s, q). To help
see the picture, note that if ∅s,q is the copy in T (s, q) of the empty node
of ω <ω , then the set of predecessors of ∅s,q in Tω+ω is precisely b(s, q).
So ∅s,q stands at the top of bs,q at level ω, and T (s, q) extends above
it. Define θ(∅s,q ) = q, and extend θ to T (s, q) level by level simlar to
the way it was defined on Tω .
Now suppose α is a limit ordinal greater than ω + ω, and Tβ and
θ  Tβ has been defined for all limit ordinals β < α such that the
following holds:
(∗) If γ < δ < β, s is in the γ th level of Tβ and q ∈ Q is such that
q > qs , then there is a successor t of s in the δ th level of Tβ with
qt = q.
It is easily checked that (∗) holds for Tω and Tω+ω . S
If α is Sa limit of limit ordinals, we simply let Tα = β<α Tβ and
θ  Tα = β<α θ  Tβ . It is easy to check that (∗) holds with β = α.
It remains to construct Tα for the limit ordinal α when there is a
maximal limit ordinal β < α, so α = β + ω. The construction in this

187
9 Proofs

case is similar to the construction of Tω+ω from Tω . First we show that

for each s ∈ Tβ and q > qs , there is a branch b(s, q) of Tβ which meets
every level of Tβ below β and such that q = sup {qt : t ∈ b(s, q)}. let
hγn i be an increasing sequence of ordinals with supremum β, such that
γ0 is the level of s0 = s. Let hqn i be a sequence of rationals with q0 = qs
and q = sup {qn : n < ω}. By (∗), there is a successor s1 of s0 at level
γ1 with qs1 = q1 , then a successor s2 of s1 at level γ2 with qs2 = q2 , and
so on. Let b(s, q) be the branch determined by the sequence hsn i. Let
{T (s, q) : s ∈ Tβ ∧ q > qs } be a collection of copies
S of ω <ω disjoint from
each other and from Tβ and let Tα = Tβ ∪ s∈Tβ ,q>qs T (s, q). Define
the order on Tα and θ to Tα in the same way this was done for Tω+ω
vis-à-vis Tω .
S
This defines Tα for all α < ω1 . Let T = α<ω1 Tα and for each
t ∈ T let θ(t) = qt ∈ Q be as defined in the construction. Then it is
immediate from the construction that T has height ω1 , every level of
T is countable, and as noted earlier, since s < t → qs < qt there are no
uncountable branches. 

Comment. Let θ(t) = qt be as in the proof, and for each q ∈ Q

let AS q = {t ∈ T : qt = q}. Note that each Aq is an antichain of T , so
T = q∈Q Aq is the union of countably many antichains, at least one of
which must be uncountable. So this tree is definitely not Suslin. Indeed,
any Aronszajn tree which is the union of countably many antichains is
called a special Aronszajn tree.
It is consistent with the axioms of set theory not only that there are
no Suslin trees, but that every Aronszajn tree is special; for example,
this holds under the axiom MA(ω1 ).
Proof of Theorem 220
Let X be a Suslin line. Define separable closed subsets Cα , α < ω1 ,
of X as follows. Let C0 = ∅. If αSis a limit ordinal and Cβ has been
defined for all β < α, let Cα = β<α Cβ . If α = γ + 1 and Cγ has
been defined, let Lγ be the collection of convex components of X \ Cγ .
For each I ∈ Lγ , choose a countable sequence of points converging to
each endpoint of I. Let Cγ0 be the collection of these chosen points for
all I ∈ Lγ . Then let Cα = Cγ ∪ Cγ0 . Since the closure of the union of

188
countably many separable sets is separable, and Cγ0 is countable, we
see that each Cα is separable.
NowS let Lα be the set of all convex components of X \ Cα , and let
T = α<ω1 Lα . We claim that T ordered by ⊃ is a Suslin tree.
Let I ∈ T . Then there is a unique α such that I is a convex component
of X \ Cα . Since the Cα ’s get bigger with α, if I ⊂ J ∈ T , then J is
a convex component of X \ Cβ for some β < α. Furthermore, for each
β < α there is a unique such J, call it Jβ . Then {Jβ : β < α} is the set
of predecessors of I and has order type α, so I is a member of level α
of T . Since each Cα is separable, X \ Cα 6= ∅, so T has a node at every
level α < ω1 , and hence T has height ω1 .
Let Lα denote the αth level of T . Suppose I ∈ Lα , J ∈ Lβ , α 6 β,
and I and J are incomparable. There is a unique J 0 ∈ Lα with J ⊂ J 0 .
Since I and J are incomparable, I = 6 J 0 , but this means that I ∩ J 0 = ∅,
and I ∩J = ∅. Thus an antichain in T corresponds to a pairwise-disjoint
collection of open sets, so every antichain of T is countable.
Note that by construction, each I in T has at least two immediate
successors. We need to show that T has no uncountable chain. We show
something more general: if T is a tree with no uncountable antichains,
and each node of T has at least two immediate successors, then T has no
uncountable chain. Suppose it did. Then there is a chain {tα : α < ω1 }
with tα ∈ Lα . Let sα be an immediate successor of tα+1 which is not
equal to tα+1 . It is easy to check that {sα : α < ω1 } is an uncountable
antichain, a contradiction. 

Proof of Theorem 221

Let T be a Suslin tree and let ≺ be some linear order on T . Let X be
the set of all branches of T . For each b ∈ X let b(α) be the member
of b at level α of the tree. Given b1 6= b2 ∈ X, let α be least such that
b1 (α) 6= b2 (α) and define b1 < b2 if and only if b1 (α) ≺ b2 (α). Then
clearly < linearly orders X.

Claim. X is ccc.

Proof of Claim. Let U = {(ai , bi ) : i ∈ I} be a collection of nonempty

pairwise disjoint open intervals in X. For each i ∈ I let ci ∈ (ai , bi ), the
let α be minimal such that ci (α) 6= ai (α), bi (α) and let ti = ci (α + 1).

189
9 Proofs

Then {ti : i ∈ I} is an antichain in T : if i = 6 j ∈ I and c is a branch

containing ti and tj then c ∈ (ai , bi ) ∩ (aj , bj ), a contradiction. So I is
countable by the ccc in T . 
Claim. X is nonseparable.

Proof of Claim. Let C ⊂ X be countable. Let κ = sup {ht b : b ∈ C} + 1.

Then there exists t ∈ T at level κ and three branches a < b < c ∈ X,
each containing t. Then a(α) = b(α) for each α < κ so any d ∈ C lies
on the same side of a and b, hence d ∈/ (a, b). So (a, b) is a nonempty
open set missing C, so C is not dense in X. 

Proof of Theorem 222 S

Let S be a Suslin line and let T = α<ω1 Lα be as in the hint for
Theorem 220. For each α, pick Iα ∈ Lα and let Iα0 , Iα1 ∈ Lα+1 be
disjoint subsets of Iα .
Let α, β < ω1 with α < β. Certainly Iβ has to miss either Iα0 or Iα1 .
Thus either Iβ0 misses Iα0 or Iβ1 misses Iα1 , so (Iα0 × Iα1 ) ∩ (Iβ0 × Iβ1 ) = ∅.
It follows that {Iα0 × Iα1 : α < ω1 } is a pairwise-disjoint collection. 
Proof of Theorem 223
If there is a Suslin line, then by Theorem 210(a), there is a densely
ordered ccc linearly ordered space X such that no nonempty open
interval is separable. In the proof of Theorem 210(b), a Suslin line Y is
constructed from such a linearly ordered space such that X is dense in Y .
It follows that Y has no nonempty separable open intervals either. Now
let C = {Cα : α < ω1 } be the strictly increasing sequence of separable
closed subsets of Y as constructed in the proof of Theorem 220. For
each α < ω1 , Cα is closed so if it failed to be nowhere dense, it would
contain an open interval of Y , and this interval would be separable
because Cα is. This is aScontradiction, so Cα is nowhere dense.
S We claim that Y = α<ω1 Cα . Suppose otherwise, and let p ∈ Y \
α<ω1 Cα . Then for each α, p ∈ / Cα , so there is a convex component Iα
of Y \ Cα containing p. In the proof of Theorem 220, it was shown that
any pair I, J of convex components of a pair of Cα ’s is either disjoint
or comparable. So {Iα : α < ω1 } is an uncountable chain in the Suslin
tree constructed there, which is a contradiction. Thus Y is a Suslin line
which is the union of ω1 -many nowhere dense sets. 

190
Proof of Theorem 224
Let D0 , D1 , . . . be countably many dense subsets of a partial order
P . Choose d0 ∈ D0 . Since D1 is dense, there is d1 ∈ D1 with d1 6
d0 . Similarly, there is d2 ∈ D2 with d2 6 d1 , and so on. Let F =
{p ∈ P : ∃n < ω(dn 6 p)}. It is easy to check that F is a filter in P . By
construction, F ∩ Dn 6= ∅ for all n. 
Proof of Theorem 225
Let U be a collection of 6 κ-many dense open sets in the compact
Hausdorff ccc-space X. Let O(X) be the collection of nonempty open
subsets of X, ordered by ⊂. Note that incompatible elements of O(X)
are disjoint, so since X has the ccc, so does O(X).
For each U ∈ U, let DU = {V ∈ O(X) : V ⊂ U }. Since X is regular,
it is easy to check that each DU is dense in the partial order O(X).
By MA(κ), there is a filter G is O(X) which meets each DU . Let gU ∈
G ∩ DU . Since G is a filter, whenever U1 , . . . , Un are in U, there is
h ∈ O(X) with h ⊂ gUi , for each 1 6 i 6 n. It follows that {gU : U ∈ U}
has the finiteTintersection property, and so does {g U : U ∈ U}.
T Since X
is compact, U ∈U g U 6= ∅. Since gU ∈ DU , g U ⊂ U , and so U 6= ∅.

Proof of Corollary 226

Let X = [0, 1], and for each x ∈ [0, 1], let Ux ∈ X \ {x}. Then U =
{Ux : x ∈ [0, 1]} is a collection
T of 2ω -many dense open subsets of the
interval, and of course U = ∅. 
Proof of Corollary 227
By Theorem 223, if there is a Suslin line, then there is a Suslin line
X which is a union of ω1 -many nowhere dense sets, say {Aα : α < ω1 }.
Then {X \ Aα : α < ω1 } is a collection of ω1 -many dense open sets with
empty intersection. 
Proof of Lemma 228
Suppose MA(ω1 ), X is a ccc space, and {Uα : α < ω1 } is an uncountable
collection of (distinct) nonempty open subsets S of X. S
We claim there exists α0 < ω1 such that β>α0 Uβ ⊂ β>α Uβ for
S S
each α > α0 . If not, there exists α0 > 0 such that β>0 Uβ \ β>α0 Uβ 6=
∅. If δ < ω1 and αγ has been defined for all γ < δ, pick αδ = supγ<δ αγ
S S
if δ is a limit, and pick αδ > αδ−1 such that β>αδ−1 Uβ \ β>αδ Uβ 6= ∅

191
9 Proofs

if δ is a successor. Then
[ [
{ Uβ \ Uβ : δ < ω1 }
β>αδ β>αδ+1

violates the ccc in X. S

Let P = {O ∈ O(X) : O ⊂ β>α0 Uβ }. For each α > α0 let

Dα = {O ∈ P : ∃β > α(O ⊂ Uβ )}.

S S
Then Dα is dense in P; let O ∈ P. Then O ⊂ β>α0 Uβ ⊂ β>α Uβ
so there exists β > α such that O ∩ Uβ 6= ∅. Then O ∩ Uβ ∈ Dα and
O ∩ Uβ ⊂ O as desired.
Note that X is ccc implies O(X) is ccc implies P is ccc. By MA(ω1 )
there exists a filter G on P such that G ∩ Dα 6= ∅ for all α > α0 .
Since G is closed upward, for each α > α0 there exists β > α such
that Uβ ∈ G. Let γ0 > α0 + 1 such that Uγ0 ∈ G. If δ < ω1 and an
increasing sequence hγβ iβ<δ has been defined so that Uγβ ∈ G for each
β < δ, let γα > supβ<δ γβ such that Uγδ ∈ G. Then {Uγδ : δ < ω1 } is an
uncountable subcollection of {Uα : α < ω1 } with the finite intersection
property. 
Proof of Corollary 229
Immediate from Lemma 228. 

Proof of Theorem 230

Suppose X, Y have the ccc and {Uα : α ∈ Λ} is an uncountable col-
lection of pairwise disjoint nonempty open subsets of X × Y . Assume
the Uα ’s are basic; Uα = Vα × Wα . By Corollary 229, X has prop-
erty K. There exists an uncountable A ⊂ Λ such that {Vα : α ∈ A}
is linked. Then {Wα : α ∈ A} is uncountable, so there exists an un-
countable B ⊂ A such that {WB : β ∈ B} is linked. Let β = 6 γ ∈ B.
Then Uβ ∩ Uγ = (Vβ × Wβ ) ∩ (Vγ × Wγ ) 6= ∅ since Vβ ∩ Vγ 6= ∅
and Wβ ∩ Wγ 6= ∅, a contradiction. The theorem follows from Theo-
rem 132. 
Proof of Lemma 231
(a) Let hX, T i be a regular Lindelöf space and suppose X is hereditarily
Lindelöf. Certainly X is normal. Let H ⊂ X be closed and for each

192
x ∈ X \ H let Ux ∈ T such that x ∈ Ux ⊂ U x ⊂ X \ H. Since X \ H
1 , . . . ∈ X such that {Uxi : i < ω} covers
is Lindelöf, there exist x0 , xT
X \ H. It follows that H = i<ω X \ U xi .
Now suppose XSis perfectly normal. Let U ⊂ T be uncountable T
and put H = X \ U. Let {Vi : i < ω} ⊂ T such that H = i<ω Vi .
Observe that each X \ Vi is closed, hence Lindelöf. S For each i < ω, let
Vi ⊂ U be a countable
S cover
S of X \ V i . Put V = i<ω Vi . Then V ⊂ U
is countable with V = U, whence X is hereditarily Lindelöf.

(b) Let hX, T i be a compact Hausdorff space and suppose X is first-

countable. Let x ∈ X and let {Ui : i < ω} be a local base at x. T If y ∈
X \{x} there must exist some i < ω such that y ∈ / Ui , so {x} = i<ω Ui .
Now suppose T every point of X is Gδ . Let x ∈ X and {Vi : i < ω} ⊂
T with {x} = i<ω Vi . Put U0 = V0 . If n < ω and Un has been
defined, let Un+1 ∈ T such that x ∈ Un+1 T ⊂ U n+1T ⊂ Un ∩ Vn+1 .
Then U0 ⊃ U 1 ⊃ U1 ⊃ · · · and {x} = i<ω Ui = i<ω U i . Let U
be an open neighborhood of x. Since X \ U is compact and covered
by {X \ U i : i < ω}, there exists k < ω such that X \ U ⊂ X \ U k . It
follows that Uk ⊂ U , whence {Ui : i < ω} is a local base at x. 

Proof of Theorem 232

Assume MA(ω1 ). Suppose X is compact Hausdorff hereditarily Lindelöf
but not hereditarily separable. Then there exists a subspace S of X
which is not separable. Then S is not separable: suppose {xn : n < ω}
is dense in S. By Lemma 231, X is first-countable. For each n < ω
let {Uin : i < ω} be a local base at xn . For each n, i < ω there exists
sni ∈ S ∩ Uin . Then {sni : n, i < ω} is countable and dense in S.

Claim 1: There are points xα ∈ S, for all α < ω1 , such that xα is not
a member of clS {xβ : β < α} = {xβ : β < α}. Let x0 ∈ S. Assuming
α < ω1 and xβ , β < α, have been chosen appropriately, {xβ : β < α}
is countable and thus there exists xα ∈ S \ clS {xβ : β < α}.
Let Y = {xα : α < ω1 }.

Claim 2: For each y ∈ Y there exists α < ω1 such that y ∈

{xβ : β < α}. If y ∈ Y then, letting {Un : n < ω} be a local base

193
9 Proofs

at y and xαn ∈ Un for each n, we have y ∈ {xβ : β < supn<ω αn }.

Y is compact Hausdorff and therefore Y is regular, so by Claim 1 for
each α < ω1 there exists an open Uα ⊂ Y such that xα ∈ Uα ⊂ U α ⊂
Y \ {xβ : β < α}. Note that Y is ccc since it is hereditarily Lindelöf. By
Lemma 228 there exists a cofinal subset A of ω1 such that {Uα : α ∈ A}
has the finite intersection property. Then {U α : α ∈ A} T has the finite
intersection property. Since Y is compact we have α∈A U α 6= ∅.
This contradicts Claim 2, as a point in this intersection cannot be
in {xβ : β < α} for any α < ω1 . 

Proof of Lemma 233

Let B be a countable base for R. Let P be all pairs of the form hC, ~ F i,
~
where C is a finite sequence of members of B, and F is a finite subset
~ 0 , F 0 i, hC,
of κ. For hC ~ F i ∈ P , define hC
~ 0 , F 0 i 6 hC,
~ F i if and only
~ extends C,
if C 0 ~ F ⊃ F , and for each i ∈ dom C
0 ~ \ dom C,
0 ~ we have
T
Ci ⊂ α∈F Uα .

Claim 1: hP, 6i is a partially ordered set. We need to prove tran-

sitivity. Suppose hC~ 00 , F 00 i 6 hC ~ 0 , F 0 i 6 hC,
~ F i. Then C~ 00 extends C
~0
extends C ~ and F ⊃ F ⊃ F , so C
00 0 ~ extends C
00 ~ and F ⊃ F . If
00

i ∈ dom ~ 00 ~ ~ 00 ~0
T C \ dom C,Tthen either i ∈ dom C~ 0 \ dom C~ in which case
Ci ⊂ α∈F 0 Uα ⊂ α∈F Uα , or i ∈ dom C \ dom C in which case
T
Ci ⊂ α∈F Uα . Thus hC ~ 00 , F 00 i 6 hC, ~ F i.

Claim 2: hP, 6i has the ccc. Suppose hC ~ α , Fα i is in P for each α < ω1 .

We need to show that there is α 6= β such that hC ~ α , Fα i and hC~ β , Fβ i are
compatible. Since B is countable, so is the collection of finite sequences
from B, so there are α 6= β such that C ~α = C ~ β = C. ~ Then it is easy
to see that hC,~ Fα ∪ Fβ i is less than or equal to both hC ~ α , Fα i and
~
hCβ , Fβ i, and hence they are compatible.
Now we define some dense sets. For each B ∈ B and k < ω, let

~ F i ∈ P : ∃i > k(Ci ⊂ B)}.

DB,k = {hC,

The DB,k ’s serve two purposes. One, they will make sure that the
generic filter G contains elements whose first coordinate is a sequence

194
of arbitrarily long length, and thus G will determine an infinite
S sequence
hC0G , C1G , . . .i of members of B. Two, they make sure that i>n CiG is
dense in R for each n.
Also, for each α < κ, let

~ F i ∈ P : α ∈ F }.
Eα = {hC,

The
S Eα ’s will make sure that for each α, there is some n so that
i>n Ci ⊂ Uα .
Let us show that the DB,k ’s and the Eα ’s are indeed dense in P .
~ F i ∈ P . Then hC,
Let hC, ~ F ∪ {α}i 6 hC, ~ F i, so Eα is dense. Now fix
~
B ∈ B and k < ω. Suppose C = hC0 , C1 , . . . , Cn i. Let T m > max {n, k}
and chooseTCi for i = n + 1, . . . , m such that T C i ⊂ α∈F Uα , and also
Cm ⊂ B ∩ α∈F Uα . This is possible since α∈F Uα is dense open. Let
~ 0 = hC0 , C1 , . . . , Cm i. Then hC
C ~ 0 , F i ∈ DB,k and hC ~ 0 , F i 6 hC,
~ F i. So
DB,k is dense.
Let G be a filter in P meeting all Eα ’s, α < κ, and all DB,k ’s, B ∈ B
and k < ω.

Claim 3: If hC, ~ F i and hC ~ 0 , F 0 i are in G, then the sequence C ~ extends

~ 0 ~ ~ 0 0
C or vice-versa. Well, if hC, F i and hC , F i are in G, then there is
some hC ~ 00 , F 00 i in G such that hC ~ 00 , F 00 i is less than or equal to both
~ ~ 0 0
hC, F i and hC , F i, and hence the sequence C ~ 00 extends both C~ and
C~ . The claim follows.
0

Now, since G meets all DB,k ’s, there are arbitrarily long sequences
appearing as first coordinates of members of G, so we can define an
infinite sequence C ~ G = hC G , C G , . . .i of members of B by defining C G
0 1 i
th
to be the i term of C ~ for some hC, ~ F i ∈ G with i ∈ dom C. ~ It follows
from Claim 3 that it doesn’t matter which member of G we use to define
CiG . Note that C ~ G = S {C ~ : ∃F (hC, ~ F i ∈ G)}, where C ~ is viewed as a
set of ordered pairs.

Claim 4: For each n < ω, i>n CiG is dense in R. If hC,

S ~ Fi ∈
G ∩ DB,n , then we have Ci ⊂ B for some i > n. It follows that every
G
S B ∈ BG contains Ci for some i > n, from which it easily
basic open set
follows that i>n Ci is dense in R.

195
9 Proofs

Claim 5: For each α < κ, there is some n < ω such that i>n CiG ⊂
S

Uα . Fix α < κ, and let hC, ~ F i ∈ G ∩ Eα . Then α ∈ F . Let C ~ =

~ 0 0
hC0 , C1 , . . . , Cn i. Suppose i > n, and let hC , F i ∈ G such that i ∈
dom C ~ 0 . There is some hC ~ 00 , F 00 i ∈ G with hC
~ 00 , F 00 i less than or equal
~ 0 0 ~
to both hC , F i and hC, F i. Then Ci = Ci00 and Ci00 ⊂ Uα because
G
~ 00 , F 00 i 6 hC,
hC ~ F i and α ∈ F .
G
T S
Finally, from Claims 4 and 5, it follows that T n<ω ( i>n Ci ) is a
dense Gδ subset of R which is contained in α<κ Uα . 

Proof of Theorem 234

Assume MA, and let F be a collection of < 2ω -many first category
subsets of R. Let κ = |F|. For each F ∈ F, let N (F) be a countable
collection of nowhere-dense sets covering F . Since the closure of a
nowhere-dense set is nowhere-dense, we may assume each member of
N (F) is closed. Let

U = {R \ N : ∃F ∈ F[N ∈ N (F)]}.

Then U is a collection of
T κ-many dense open sets. By Lemma 233, there
is a dense Gδ -set G ⊂ U. Then S R \ G is the union
S of countably many
nowhere dense sets and contains F. Hence F is first category. 

Proof of Theorem 235

Since the set of all functions in ω ω is both <∗ -dominating and <∗ -
unbounded, certainly both b and d are 6 c. Also, every <∗ -dominating
family is obviously <∗ -unbounded, so b 6 d. So it remains to prove
ω1 6 b. Let {fn : n < ω} be a countable subset of ω ω . Define g(n) =
1 + max {fi (n) : i 6 n}. Then g(n) > fk (n) for all n > k. This shows
that every countable subset of ω ω is <∗ -bounded, and it follows that
ω1 6 b. 

Proof of Theorem 236

Let κ < c, and consider a collection of functions {fα ∈ ω ω : α < κ}. Let
<ω
P = hω <ω , [κ] i be ordered by 4, where hg, Gi 4 hh, Hi if and only if
(1) g ⊃ h,

(2) G ⊃ H, and

196
 (3) for n ∈ |h| \ |h| and α ∈ H, g(n) > fα (n).
To check that 4 is a partial order on P, let hg, Gi 4 hh, Hi 4 hi, Ii.
One may easily see that g ⊃ h ⊃ i and G ⊃ H ⊃ I. For n ∈ |h| \ |i|, if
n ∈ |g| \ |h|, then g(n) > fα (n) for all α ∈ H ⊃ I; otherwise n ∈ |h| \ |i|,
and thus g(n) > fα (n) for all α ∈ I.
>ω
To check that hP, 4i has the ccc, let A ∈ [bP ] . Since ω <ω is
countable, fix hg, G0 i, hg, G1 i ∈ A such taht G0 6= G1 . It’s then easy to
verify that hg, G0 ∪ G1 i extends both, so A is not an antichain.
Let Dα,n = {hg, Gi : |g| > n ∧ α ∈ G}. For each hh, Hi ∈ P, hh, Hi <
hh0 , H ∪ {α}i ∈ Dα,n such that h0 ⊃ h, |h0 | > n and for each i ∈ |h0 |\|h|
and β ∈ H, h0 (i) > fβ (i). Thus each Dα,n is dense.
Apply MA → MA(κ)Sto obtain a filter L intersecting each Dα,n . It
then follows that f = {`(0) : ` ∈ L} is a function in ω ω due to the
compatibility of each `(0) and the fact that L intersects D0,n for each
n < ω.
Let α <Sκ, and choose `n ∈ L ∩ Dα,n for each n < ω. It follows
that f = {`n (0) : n < ω} ∈ ω ω . Also, for i < |`0 (0)|, choose ` ∈ L
such that ` < `0 , `i . Then since i ∈ |`i | \ |`0 | ⊂ |`| \ |`0 |, it follows that
f (i) = `(0)(i) > fα (i).
Thus {fα : α < κ} is not <∗ −unbounded, and b > κ. Therefore
b > c, and the result follows. 
Proof of Theorem 237
Let {gα ∈ ω ω : α < b} be <∗ -unbounded, and define f0 = g0 . For each
limit ordinal γ < b note that {fα : α < γ} is not <∗ -unbounded; define
fγ such that fα <∗ fγ for each α < γ and gγ <∗ fγ . For a successor
ordinal γ + 1, define fγ+1 such that fγ+1 (n) = fγ (n) + gγ+1 (n) + 1 for
each n < ω, noting fγ , gγ+1 <∗ fγ+1 .
By construction, {fα : α < b} is <∗ -increasing. Let f ∈ ω ω . Since
{gα : α < b} is <∗ -unbounded, pick α such that gα 6<∗ f . Since gα 6∗
fα , it follows that fα 6<∗ f , and thus {fα : α < b} is also <∗ -unbounded.

Proof of Theorem 238

Assume b = d. Let {gα : α < d} be <∗ -dominating; put f0 = g0 . For
each limit ordinal γ < d = b, the collection {fα : α < γ} is not <∗ -
unbounded, so define fγ such that fα <∗ fγ for each α < γ, and
gγ <∗ fγ . For each successor ordinal γ + 1, define fγ+1 such that

197
9 Proofs

fγ+1 (n) = fγ (n)+gγ+1 (n)+1 for each n < ω, and note that fγ <∗ fγ+1
and gγ+1 <∗ fγ+1 .
Certainly {fα : α < d} is <∗ -increasing. To show this collection is <∗ -
dominating, let g ∈ ω ω . Because {gα : α < d} is <∗ -dominating, there
exists α < d such that g <∗ gα . Thus g <∗ gα <∗ fα , so {fα : α < d}
is <∗ -dominating. 
Proof of Theorem 239
Let λ = min A and µ = min B.
d 6 λ Let {Kα : α < λ} be a collection of λ-many compact sets cover-
ing ω ω . Then πn [Kα ] is compact, and thus finite, for each α < λ and
n < ω. For each α < λ define fα ∈ ω ω by fα (n) = 1 + max πn [Kα ]. Cer-
tainly |{fα : α < λ}| 6 λ. To see that {fα : α < λ} is <∗ -dominating,
note that g ∈ Kα implies g < fα .

λ 6 µ Every dominating family A of compact subsets of ω ω covers ω ω :

if g ∈ ω ω then certainly {g} is compact, so there exists A ∈ A such
that {g} ⊂ A (that is, g ∈ A).

µ 6 d Let F ⊂ ω ω such that F is <∗ dominating and |F| = d. For

each f ∈ F and k, ` ∈ ω define

Kfk,` = {g ∈ ω ω : n > k → g(n) 6 f (n) ∧ n < k → g(n) 6 `}.

Note Kfk,` is compact. Let K = {Kfk,` : f ∈ F ∧ k, ` ∈ ω}. Then |K| 6

|F| = d. Let K ⊂ ω ω be compact. Define g ∈ ω ω by g(n) = max πn [K].
There exists f ∈ F and k ∈ ω such that g(n) < f (n) when n > k. Let
` = max {g(n) : n < k}. Then g ∈ Kfk,` , K ⊂ Kfk,l . 

Proof of Proposition C
Note that an equivalent definition follows: A ⊂ P is concentrated about
the rationals if and only if for any closed subset C in the real line which
misses the rationals, A ∩ C is countable.
There exists a closed uncountable subset of the irrationals. If Q =
hqn : n < ωi, the collection U = {(qn − 1/2n , qn + 1/2n ) : n < ω} does
not cover R, so R\ U is an example. Alternately, consider 2ω as a subset
S
of ω ω —it must be compact and thus closed in every T2 superspace. 

198
Proof of Proposition D
Let fα ∈ ω ω for α < b such that {fα : α < b} is <∗ -unbounded and
<∗ -increasing, and let θ : ω ω → I ∩ P be a homeomorphism. Let A =
b
θ[{fα : α < b}] ∈ [P] .
Let C ⊂ P be closed in the reals. Since C ∩ I is compact, it follows
that θ−1 [C] = θ−1 [C ∩ I] is compact in ω ω , so there exists g ∈ ω ω
satisfying g >∗ f for each f ∈ θ−1 [C]. Then there exists β < b such
that g 6>∗ fα for all β 6 α < b. Therefore |θ−1 [C]| 6 |{fα : α < β}| < b,
and thus |A ∩ C| < b. Thus A is (< b)-concentrated about Q.
κ
Now suppose that κ > ω, and that A ∈ [P] is (< κ)-concentrated
about Q. Let θ : ω ω → P be a homeomorphism. Let g ∈ ω ω , and note
that

Kg,i,j = {f ∈ ω ω : n < i → f (n) < j ∧ n > i → f (n) < g(n)}

is compact, so θ[Kg,i,j ] is compact

S and thus closed in the reals. As a re-
sult, |A ∩ θ[Kg,i,j ]| < κ. Since {Kg,i,j : i, j < ω} = {f ∈ ω ω : f <∗ g},
it follows that |A ∩ θ[{f ∈ ω ω : f <∗ g}]| < κ × ω = κ so there exists
x ∈ A such that θ−1 (x) 6<∗ g. Therefore θ−1 [A] is <∗ -unbounded, so
κ > b. 

Proof of Theorem 240

Immediate from Proposition D. When b = ω1 , then there exists A ∈
ω
[P] 1 which is (< ω1 )-concentrated (therefore concentrated) about Q.
And if there exists some uncountable A ⊂ P concentrated about Q,
ω
then any A0 ∈ [A] 1 is concentrated (therefore (< ω1 )-concentrated
about Q, showing b = ω1 . 

Proof of Example 241

Let everything be defined as in the statement.
To see that X is Lindelöf, let U beSan open cover of X; let V ⊂ U
be countable covering Q. Since S X \ V is countable, there exists a
countable W ⊂ U covering X \ V. Then V ∪W is a countable subcover
of U.
To see that X is regular let x ∈ X and let H ⊂ X be closed not
containing X. If x ∈ A, then {x} and X \ {x} separate x and H. If
x ∈ Q, there is a neighborhood (a, b) ∩ X of x missing H with a, b ∈ P.

199
9 Proofs

Then X \ [(a, b) ∩ X] is an open set containing H, so x and H may be

separated.
Finally, consider K = {(a, a) : a ∈ A} ⊂ X × P. For (x, y) ∈ (X ×
P) \ K, there exist disjoint Euclidean neighborhoods U and V of x and
y, respectively. So (U ∩ X) × (V ∩ P) is an open neighborhood of (x, y)
missing K. Thus K is closed.
Furthermore, each a ∈ A is isolated in X, so {{a} × P : a ∈ A} is
an open cover of K which is uncountable and pairwise disjoint, so K
is discrete and uncountable and not Lindelöf. Therefore X × P is not
Lindelöf. 
Proof of Theorem 242
(a) Let H = Q × {0} and K = P × {0}. Then H and K are closed
and disjoint. Suppose U and V are open and disjoint with H ⊂ U and
K ⊂ V . For each p ∈ P there exists np ∈ N suchSthat D(p, np ) ⊂ V .
For each n ∈ N let An = {p ∈ P : np = n}. Then n∈N An = P. Since
P is a Baire space there exists n ∈ N such that An is dense in some
interval (a, b) of R. Let q ∈ (a, b) ∩ Q and hpi i an increasing sequence of
irrationals in (a, b) ∩ An convering to q. Then any basic neighborhood
of hq, 0i must meet a basic open neighborhood D(pi , n) for some i. So
U ∩ V 6= ∅.

(b) H = {{ha, 0i} : a ∈ A} is a discrete collection of closed sets. Sup-

pose U = {UH : H ∈ H} is a disjoint collection of open sets separating
H. Then we may identify each UH with a unique hqH , rH i ∈ UH where
qH , rH ∈ Q.

(c) Let A ⊂ R such that every subset of A is Gδ in A. Let H ⊂ A and

put KS= A \ H. Let {Hn : n < ω} witness H is Fσ . For each n < ω, set
Un = {D(x, 1) : x ∈ Hn }. S
Certainly H × {0} ⊂ n<ω Un . Fix n < ω and let y ∈ K. Let
(a, b) ⊂ R be an interval containing y and missing Hn and pick k ∈ N
such that D(y, k) ∩ [D(a, 1) ∩ D(b, 1)] = ∅. If x ∈ Hn , then x 6 a or
x > b, so D(y, k) ∩ D(x, 1) = ∅ as well. It follows that hy, 0i 6 U n , so
U n ∩ (K × {0}) = ∅ for each n < ω.
Similarly there
S is a collection {Vn : n < ω} of open sets such that
K × {0} ⊂ n<ω Vn and V n ∩ (H × {0}) = ∅ for all n < ω. By

200
Lemma 57, H × {0} and K × {0} may be separated, so X(A) is normal.


Proof of Theorem 243

Suppose A is an uncountable subspace of reals and 2ω < 2ω1 . Then A
ha a countable base {Bn : n < ω}. Define f : τA → 2ω by
(
1 if Bn ⊂ U ,
f [U ](n) =
0 else.

Suppose U 6= V ∈ τA . Then {n < ω : Bn ⊂ U } = 6 {n < ω : Bn ⊂ V }.

Without loss of generality we may assume n ∈ {n < ω : Bn ⊂ U } \
{n < ω : Bn ⊂ V }. Then f [U ](n) = 1 6= 0 = f [V ](n), so {f }U =
6 f [V ].
So f is an injection.
Let Gδ be the family of Gδ -subsets of A. For each T G ∈ Gδ , choose
a sequence of open-in-τA sets hUn i such that G = n<ω Un . Clearly
ω
G 7→ hf [Un ]i is an injection from Gδ into (2ω ) .
ω
We have |Gδ | 6 (2ω ) = 2ω 6 2ω1 6 P(A), so not every subset of A
is Gδ . 

Proof of Theorem 244

Let A ⊂ R have cardinality κ < c and let B ⊂ A. Let C be a countable
base for A. The plan is to use MA(κ) to show that there is a sequence
hC0 , C1 , . . .i of members of C such that every x ∈ B is in infinitely many
Ci ’s, while every y ∈ T A \ B isS in only finitely many Ci ’s. If we do this,
then note that B = n<ω ( i>n Ci ), and hence B is a Gδ -set in the
space A.
To this end, let P be all pairs of the form hC, ~ F i where C ~ is a
finite sequence of members of C and F is a finite subset of A \ B. For
~ Gi, hC,
hD, ~ F i ∈ P , define hD,
~ Gi 6 hC,~ F i if and only if D
~ extends C,~
~ ~
G ⊃ F , and for each i ∈ dom D \ dom C we have Ci ∩ F = ∅. That
hP, 6i is a partially ordered set, and has the ccc, follows from essentially
the same argument as in Proof 234.
Now we define some dense sets. For each b ∈ B and k < ω, let

~ F i ∈ P : ∃i > k(b ∈ Ci )}.

Db,k = {hC,

201
9 Proofs

Also, for each a ∈ A \ B, let

~ F i ∈ P : a ∈ F }.
Ea = {hC,

That these sets are dense in P follows as easily as in Proof 234.

Let G be a filter in P meeting all Ea ’s, a ∈ A \ B, and all Db,k ’s,
b ∈ B and k < ω. Since G meets all Db,k ’s, there are arbitrarily long
sequences appearing as first coordinates of member of G, so we can
define an infinite sequence C ~ G = hC G , C G , . . .i of members of C by
0 1
~ for some hC,
defining CiG to be the ith of C ~ F i ∈ G such that i ∈ dom C. ~
Note that
[
~G =
C ~ : ∃F (hC,
{C ~ F i ∈ G)},

where C ~ is viewed as a set of ordered pairs.

Fix b ∈ B. Since G meets Db,k for eack k < ω, it follows that b ∈ CiG
for infinitely many i. Now fix a ∈ A \ B. There is some hC, ~ F i ∈ G ∩ Ea .
Then a ∈ F . Let C~ = hC0 , C1 , . . . , Cn i. Suppose i > n and let hD,~ Gi ∈
G such that i ∈ dom D. ~ There is some hE, ~ Hi ∈ G with hJ, ~ Hi less
than or equal to both hD, ~ Gi and hC, ~ F i. Then C G = Ji and a ∈ / Ji
i
because hJ, ~ Hi 6 hC,
~ F i and a ∈ F . It follows that a ∈ / i>n CiG , and
S

hence a is in only finitely many terms of C ~ G . Now the result follows as

indicated in the first paragraph. 

Proof of Corollary 245

Proof of Corollary 246

Assume MA, and let κ < 2ω be an infinite cardinal. There is a subset
A of the real line of cardinality κ. By MA and Theorem 244, every
subset of A is Gδ in A. There is a countable base for A and every open
subset of A is a union of members of this base, so there are no more
than 2ω -many open sets in A. Each Gδ -set is by definition a countable
ω
intersection of open sets, so there are no more than (2ω ) = 2ω×ω = 2ω
many Gδ -sets in A. So the cardinality of the power set of A, which is
2κ , is no more than 2ω . And of course it is at least that, so 2κ = 2ω .

202
Proof of Lemma 247
Let X be a normal space, let D ⊂ X be closed discrete, and let G ⊂ X
be dense. For H ⊂ D, let UH and VH be disjoint open sets containing
H and D \ H, respectively. Define f : P(D) → P(G) by f [H] = UH ∩ G.
Let H, K ⊂ D with H 6= K. Without loss of generality, we may
assume H \ K is nonempty. Then G ∩ (UH ∩ VK ) is a nonempty subset
of f [H] missing f [K], so f [H] 6= f [K]. Since f is one-to-one, 2|D| 6 2|G| .
This proves a more general version of the lemma. Apply this with G
a countable dense set to obtain the lemma as stated. 

Proof of Corollary 248

Suppose 2ω < 2ω1 and let X be a normal, separable space. Let H be a
discrete collection of closed sets in X. For each H ∈ H, choose xH ∈ H
and put D = {xH : H ∈ H}. Then D is closed discrete, so 2|D| 6 2ω ,
implying |D| 6 ω and |H| 6 ω.
SEnumerate H = {Hi : i < ω}. For each i < ω, observe that Hi and
( H) \ Hi are disjoint closed sets. Since X is normal, for each
S i<ω
there exist disjoint open sets Ui and Vi containing
T H i and ( H) \ Hi ,
respectively. For each i < ω, put Wi = Ui ∩ ( j<i Vj ). Certainly Hi ⊂
Wi for each i < ω. Moreover, if i < j < ω then Wi ∩ Wj = ∅ since
Wi ⊂ Ui and Wj ⊂ Vi . 

Proof of Theorem 249

Certainly it suffices to show that there is an almost-disjoint family of
subsets of some countable set, such as Q, of cardinality 2ω . For each
x ∈ R, let Sx be the terms of a sequence of rationals converging to x.
Then x 6= y → Sx ∩ Sy is finite, so {Sx : x ∈ R} is an almost-disjoint
family of subsets of Q of cardinality 2ω . 

Proof of Proposition E

Proof of Lemma 250

Suppose {An : n < ω} is
S an almost-disjoint family of subsets of ω. Note
that each finite union i6n Ai has infinite complement, for otherwise
An+1 would have to intersect some Ai , i 6 n, in an infinite set. Now

203
9 Proofs

for each n choose

[
xn ∈ ω \ ({xi : i < n} ∪ Ai ).
i6n

Let X = {xn : n < ω}. Then X is infinite but X ∩ An is finite for each
n. Hence {An : n < ω} is not maximal. 

Proof of Theorem 251

Let A = {Aα : α < κ} be an almost-disjoint family (of infinite subsets
of ω) of cardinality κ, where ω 6 κ < 2ω . Let P be all pairs of the form
h~x, F i, where ~x is a finite sequence of members of ω and F is a finite
subset of κ. For h~y , Gi and h~x, F i in P , define h~y , Gi 6 h~x, F i if and
only S if ~y extends ~x, G ⊃ F , and for each i ∈ dom ~y \ dom ~x we have
xi ∈ / α∈F Aα .
To see that hP, leqi is a poset we need to show transitivity. Suppose
h~z, Hi 6 h~y , Gi 6 h~x, F i. Then ~z extends ~y extends ~x and H ⊃ G ⊃ F ,
so ~z extends ~x and H ⊃ F . If i ∈ dom ~z \ dom ~x, then either
S S
(1) i ∈ dom ~z \ ~y in which case xi ∈ / α∈F Aα since xi ∈ / α∈G Aα
and F ⊂ G, or
S
(2) i ∈ dom ~y \ ~x in which case xi ∈ / α∈F Aα .
Thus h~z, Hi 6 h~x, F i.
To see that hP, 6i has the ccc, suppose h~xα , Fα i ∈ P for each α < ω1 .
There are only countably many finite sequences in ω, so there exist
α 6= β < ω1 with ~xα = ~xβ = ~x. Certainly h~x, Fα ∪ Fβ i 6 h~xα , Fα i and
h~x, Fα ∪ Fβ i 6 h~xβ , Fβ i, so h~xα , Fα i and h~xβ , Fβ i are compatible.
For each k < ω, put Dk = {h~x, F i ∈ P : |{xi : i ∈ dom ~x}| > k}. For
each α < κ, put Eα = {h~x, F i ∈ P : α ∈ F }. Let h~x, F i ∈ P . Then
h~x, F ∪ {α}i 6 h~x, F i, so Eα is dense in P for each α < κ. Fix k < ω
and suppose ~x = hx0 , x1 , . . . , xn i. Take
[
{xn+1 , . . . , xn+k+1 } ⊂ ω \ ( Aα ∪ {xi : i ∈ dom ~x})
α∈F

6 xj whenever n+1 6 i < j 6 n+k+1. This is possible since

so that xi =
any finite subcollection of A must leave infinitely many members of ω

204
 uncovered. Put ~y = hx0 , . . . , xn , xn+1 , . . . , xn+k+1 i. Then h~y , F i ∈ Dk
and h~y , F i 6 h~x, F i, so Dk is dense in P .
Let G be a filter in P meeting Eα and Dk for each α < κ and each
k < ω. If h~x, F i, h~y , Gi ∈ G then there exists h~z, Hi ∈ G less than or
equal to both of them, whence ~z extends both ~y and ~x. It follows that
~y extends ~x, or vise-versa.
Let ~xG = hxG0 , xG1 , . . .i, where xGi is the ith term of ~x for some h~x, F i ∈
G with i ∈ dom ~x. Put A = {xGi : i < ω}. Since G meets every Dk , we
know |A| = ω.
Finally, fix α < κ and let h~x, F i ∈ G ∩ Eα , so α ∈ F . Let ~x =
hx0 , x1 , . . . , xn i. Suppose i > n and let h~y , Gi ∈ G such that i ∈ dom ~y .
Take h~z, Hi ∈ G less than or equal to both h~y , Gi and h~x, F i. Then
xGi = zi 6 Aα , so |A ∩ Aα | < ω. It follows that A ∪ {A} is an almost-
disjoint family, hence A is not maximal. 
Proof of Theorem 252
Consider the basic open neighborhood b(xA , 0) = {xA } ∪ A of xA .
Since every open set U containing xA contains b(xA , n) for some n, U
therefore contains all but at most finitely many points of b(xA , 0). Thus
b(xA , 0) is compact. Since all points other than the xA ’s are isolated,
we have that ψ(A) is locally compact.
To see Hausdorff, suppose A 6= B ∈ A. Then A ∩ B is finite, so there
is some n < ω such that b(xA , n) ∩ b(xB , n) = ∅. Since all “non-xA ’s”
are isolated, it easily follows that ψ(A) is Hausdorff (this is the only
place in the proof where “almost-disjoint” is needed).
Let D = {xA : A ∈ A}. Every point outside of D is isolated, so D is
closed. Also, b(xA , 0) ∩ D = {xA }, so D is discrete. Thus D is closed
discrete. 
Proof of Theorem 253
Let X be countably compact, and suppose f : X → R is continuous.
Then f [X] is a countably compact subset of R. But countable com-
pactness and compactness are equivalent in metric spaces, so f [X] is
compact, hence bounded. Thus X is pseudocompact. 
Proof of Theorem 254
Let A be an infinite MAD family of infinite subsets of ω. Let f : ψ(A) →
R be continuous and suppose on the contrary that f has unbounded

205
9 Proofs

range. Without loss of generality, we may assume there exists an in-

creasing sequence hxn i in ω such that hf (xn )i is increasing and un-
bounded. Let X = {xi : i < ω}. Since A is MAD, there exists A ∈ A
with |X ∩ A| = ω. Let han i be the order-preserving indexing of A. Then
han i → xA but hf (an )i 6→ f (xA ), a contradiction. 

Proof of Lemma 255

By CH, A has cardinality ω1 , so we can let {A0α : α < ω1 } be a list-
ing of A in type ω1 . Let Aω = A0α where α is least such that A0α ⊂
{xn : n < ω}; since {xn : n < ω} is the rationals, there will always be
such an A0α (indeed, there are ω1 -many α which satisfy this).
Now suppose Aβ has been defined for each β < α, where α < ω1 . Let
Aα = A0δ , where δ is at least such that A0δ ⊂ {xβ : β < α} and A0δ 6= Aβ
for any β < α. This defined Aα for all α < ω1 .
Certainly the condition holds by construction; it remains to prove
that {Aα : ω 6 α < ω1 } = A, which will be the case if for every γ < ω1
there is some α < ω1 such that Aα = A0γ . To this end, fix γ < ω1 .
There will be some ρ < ω1 such that A0γ ⊂ {xβ : β < ρ}. Then at any
step α > ρ, Aα = A0γ if γ is least such that A0γ has not already been
chosen. Since the induction goes on for ω1 -many steps and there are
only countably many ordinals µ < γ, A0γ must be chosen at some stage.
Hence {Aα : ω 6 α < ω1 } = A. 

Proof of Lemma 256 S

For limit α, let τα be the topology generated by β<α τβ . It is straight-
forward to check that 256(a) to 256(c) hold with β = α.
If α = β + 1, then Xα = Xβ ∪ {xβ }, and the task is to define a
neighborhood base Bβ at xβ so that the topology generated by τβ ∪ Bβ
satisfy 256(a) to 256(c). Enumerate {γ < α : xβ ∈ Aγ } as {γi : i < ω}.
Let x0,0 ∈ B(xβ , 1) ∩ Aγ0 and let C0 = {x0,0 }; here, B(x, ε) is the ball
about x of radius ε with respect to the Euclidean metric. If Ci and xj,i
have been defined for i 6 n and j 6 i, for each j 6 n + 1 let xj,n+1 ∈
B(xβ , 1/2n+1 ) ∩ Aγj . Then let Cn+1 = Cn ∪ {xj,n+1 : j 6 n + 1}.
Let C = {xj,n : n < ω ∧ j 6 n}. Every ball about xβ contains xj,n for
all sufficiently large n, and hence contains all but finitely many members
of C. So C = {ci : i < ω} is a sequence which converges to xβ in the
Euclidean topology. Let Ki be a compact open set in Xβ containing

206
ci with diam Ki 6 1/i + 1; this is possible since the topology of Xβ
is a locally compact 0-dimensional topology
S finer than the Euclidean
topology. Then let sets Uβ,m = {xβ } ∪ i>m Ki for m < ω be a local
base Bβ at xβ in Xα .
Let τα be the topology on Xα generated by τβ ∪ Bβ . 
Proof of Theorem 257
That hR, τω1 i is locally compact Hausdorff follows from Lemmas 256(a)
and 256(b). Certainly {Xα : α < ω1 } is an open cover with no countable
subcover, so hR, τω1 i is not Lindelöf.
Let Y ⊂ R be uncountable and let D be a countable dense (in
the Euclidean topology) subset thereof. Since |D| > ω—with D the
Euclidean closure—there exists γ < ω1 such that D = Aγ . Let δ ∈
(γ, ω1 ). By Lemma 256(c),
xδ ∈ Aγ = Y → xδ ∈ clτδ1 Aδ → xδ ∈ clτω1 Aγ .
It follows that |Y \ clτω1 Aγ | 6 ω, so D ∪ (Y \ clτω1 Aγ ) is a countable
dense (in τω1 ) subset of Y . Thus, hR, τω1 i is hereditarily separable. 
Proof of Lemma 258
Let X be a locally compact, dense subset of a T2 -space Z, let x ∈ X,
and let Nx be a compact neighborhood of x in X. Let Ux be an open
subset of Z satisfying Nx◦ = X ∩ Ux . We have Nx◦ is dense in Ux so that
Ux ⊂ U x = Nx◦ ⊂ N x = Nx ,
where closures are in Z. Because each point of X is contained within a
set open in Z contained within X, X is open in Z. 
Proof of Corollary 259
Assume αX is a compactification of a locally compact space X. By
definition, αX is a T2 -space in which homeomorphism α : X → αX
embeds X as a dense subset α[X] of αX. This implies α[X] is also
locally compact, so that α[X] is open in αX by Lemma 258. 
Proof of Theorem 260
Define f : αX → ωX by p 7→ (ω ◦ α−1 )(p) if p ∈ α[X] and p 7→ ∞ if
p ∈ αX \ α[X]. For each x ∈ X we have
(f ◦ α)(x) = f (α(x)) = (ω ◦ α−1 )(α(x)) = ω(x),

207
9 Proofs

so that f ◦ α = ω. Now we prove f is continuous. To that end, let U

be open in ωX. We show f −1 [U ] is open in αX.
If ∞ ∈/ U then U is an open subset of ω[X], so
−1
f −1 [U ] = (ω ◦ α−1 ) [U ]
is open in α[X]. By Corollary 259, f −1 [U ] is also open in αX.
If ∞ ∈ U , then

f −1 [U ] = αX \ f −1 [ωX \ U ]
−1
= αX \ f −1 [ω[X] \ U ] = αX \ (ω ◦ α−1 ) [ω[X] \ U ].
By definition of the one-point compactification ωX, we know that
−1
ω[X]\U is compact. Note that the preimage (ω ◦ α−1 ) is the image of
−1
the continuous inverse of ω ◦ α−1 . So (ω ◦ α−1 ) [ω[X] \ U ] is compact,
thus it is closed in αX. Again we see that f −1 [U ] is open. 
Proof of Theorem 261
Let γX = X ∪ (αX \ α[X]). Define θ : γX → αX such that

α(x) : x ∈ X
θ(x) =
x : x ∈ αX \ α[X]
and declare U to be open in γX if and only if θ[U ] is open in αX. Since
θ is a bijection, it follows that θ is a homeomorphism by definition,
and X is dense in γX. Finally, if γ : X → γX is the inclusion map,
θ ◦ α = γ trivially, so αX ≈ γX. 
Proof of Lemma 262
Suppose f and g agree on the dense subset D of X, but f 6= g. Pick p ∈
X such that f (p) 6= g(p). Since Y is Hausdorff, we can take disjoint open
sets O1 and O2 containing f (p) and g(p), respectively. By continuity,
there are open sets Uf and Ug contiaining p such that f (Uf ) ⊂ O1 and
g(Ug ) ⊂ O2 . Since D is dense, there is a point d ∈ D ∩ Uf ∩ Ug . But
then f (d) ∈ O1 and g(d) ∈ O2 , so f (d) 6= g(d), contradiction. 
Proof of Theorem 263
Let αX and γX be compactifications of a space X such that α : X →
αX and γ : X → γX are identity maps. The forwards implication is
immediate.

208
 Suppose αX > γX and γX > αX. Let f : αX → γX and g : γX →
αX be continuous such that f ◦ α = γ and g ◦ γ = α. If x ∈ X then
f (g(x)) = f (g(γ(x))) = f (α(x)) = γ(x) = x.
Thus f ◦ g and iγX agree on a dense subset of γX, so f ◦ g = iγX . By
a similar argument g ◦ f = iαX , so f and g are inverses. It follows that
f is a homeomorphism, whence αX ≈ γX 
Proof of Theorem 264
(a) For each subbasic open set πf−1 [U ] of I F where U is open in I,
note f −1 [U ] is open in X by the continuity of f . For each x ∈ f −1 (U ),
eF (x)(f ) = f (x) ∈ U , and thus eF (x) ∈ πf−1 [U ]. Thus eF is continuous.

(b) Let x = 6 y. Since F separates points, choose f ∈ F such that

f (x) 6= f (y). It then follows from
eF (x)(f ) = f (x) 6= f (y) = eF (y)(f )
that eF (x) 6= eF (y).

(c) By (a), eF is continuous. As X is T1 , points are closed, and thus

F separates points, showing eF is one-to-one by (b). And of course, eF
is onto its range eF [X].
We conclude the proof by showing that eF is a closed map. Let H
be closed, and x0 6∈ eF [H]. As eF [H] is a bijection, choose x 6∈ H with
eF (x) = x0 .
Since F separates points from closed sets, choose f ∈ F such that
f (x) 6∈ f [H]. So let U be an open set separating f (x) from f [H].
x0 (f ) = eF (x)(f ) = f (x) ∈ U , so x0 ∈ πf−1 [U ]. Furthermore for each
h0 ∈ eF [H] where h0 = eF [h], h0 (f ) = f (h) 6∈ U , so h0 6∈ πf−1 [U ]. Thus
x0 is not a limit point of eF [H], showing eF [H] is closed. 

Proof of Corollary 265

Proof of Corollary 266

209
9 Proofs

Proof of Theorem 267

Let αX be a compactification of X. Without loss of generality, assume
α is the identity on X. Define F = {f  X|f ∈ C(αX, I)}, and ϕ :
αX → eF X by ϕ(a) = eC(αX,I) (a) = hf (a)if ∈C(αX,I) . 

Proof of Theorem 268

Define H : I C(X,I) → I C(Y,I) by πg H = πg◦h . Then H is continuous
and for any x ∈ X and g ∈ C(Y, I) we have

πg (H ◦ βX )(x) = πg H(hf (x)i) = (g ◦ h)(x)

= g(h(x)) = πg βY (h(x)) = πg (βY ◦ h)(x),

so that H ◦ βX = βY ◦ h. By continuity of H we have

H[βX] = H[clI C(X,I) βX (X)] ⊆ clI C(Y,I) H[βX (X)]

= clI C(Y,I) βY [h(X)] ⊆ βY. 

Proof of Theorem 269

By the previous theorem, there exists a continuous F 0 : I C(X,I) →
I C(Y,I) such that F 0 ◦ βX = βY ◦ f and F 0  βX X : βX X → βY Y . Since
Y is already compact, βY−1 is a homeomorphism from βY Y to Y .
Let F = βY−1 ◦ (F 0  βX X). Then for each x ∈ X,

F (βX (x)) = βY−1 (F 0 (βX (x))) = βY−1 (βY (f (x))) = f (x) 

Proof of Corollary 270

Proof of Theorem 271

Let f : ω1 → I κ be continuous. For each α < κ there exist βα < ω1 and
xα ∈ I such that (πα ◦f )(β) = xα for all β > βα . Define F : ω1 +1 → I κ
so that (
(πα ◦ f )(β) if β 6= ω1
(πα ◦ F )(β) =
xa if β = ω1
for all α < κ and β < ω1 + 1.

210
 Certainly F  ω1 = f . Let U ⊂ I be open. If xα ∈ U , then (πα ◦
F )−1 (U ) = (πα ◦ f )−1 (U ). Otherwise, (πα ◦ F )−1 (U ) = (πα ◦ f )−1 (U ) ∪
(βα , ω1 ]. Thus F is continuous, so ω1 + 1 ≈ βω1 .
Alternatively... Let p 6= q be in the remainder αX \ X. Then there
exist open sets U, V containing p, q whose closures are disjoint. But
then
[U ∩ ω1 ] ∩ [V ∩ ω1 ] = ∅
are closed unbounded sets, and should intersect. Contradiction. 
Proof of Theorem 272
Let X be a separable compact Hausdorff space with countable dense
subset D. There exists a bijection f : ω → D; f is continuous since ω
is a discrete space. Using theorem 19, let F : βω → X be a continuous
extension of f . Clearly, F (βω) is dense in X (for F (βω) ⊇ F (ω) =
f (ω) = D). Furthermore, F (βω) is closed in X since βω is compact, F
is continuous, and X is Hausdorff. So we have F (βω) = F (βω) = X,
illustrating βω maps continuously onto separable compact Hausdorff
space X. 
Proof of Corollary 273

Proof of Lemma 274

Define f : X → 2P(ω) by
(
1 if x ∈ {di : i ∈ A}
f (x)(A) = .
0 else

We show f is injective, proving that |X| 6 2c . To that end, suppose

x 6= y ∈ X. Since X is Hausdorff, there exists an open set U containing
x whose closure misses y. Let A = {i ∈ ω : di ∈ U }. Then x ∈
{di : i ∈ A} but y ∈
/ {di : i ∈ A}. So f (x)(A) = 1 and f (y)(A) = 0. So
f (x) 6= f (y). 

Proof of Corollary 275

211
9 Proofs

Proof of Theorem 276

Let ι : X → Z be the inclusion map, so Z = ιX. Recall β = eC(X,I) :
X → I C(X,I) . Define g : ιX → βX by g(z)(f ) = f ∗ (z). Note g is
continuous as f ∗ is continuous, and g ◦ ι = β as for each x ∈ X and
f ∈ C(X, I),

(g ◦ ι)(x)(f ) = g(ι(x))(f ) = f ∗ (ι(x)) = f ∗ (x)

= f (x) = eC(X,I) (x)(f ) = β(x)(f ).

Thus ιX = Z > βX, so Z ≈ βX. 
Proof of Theorem 277
Let H and K be disjoint closed subsets of a normal space X. By theorem
7500.25, there exists a continuous function f : X → I with f (H) = {0}
and f (K) = {1}. By theorem 19, there is a continuous extension F :
βX → I of f . Then F −1 ({0}) and F −1 ({1}) are disjoint closed subsets
of βX containing H and K, respectively, so clβX H ∩ clβX K = ∅. 

Proof of Theorem 278

Let f : H → I be a continuous function. Because H is a closed subset
of normal X, f can be extended continuously to fˆ : X → I using the
Tietze Extension Theorem. By theorem 19 (and the remark following it),
fˆ can be extended continuously to F : βX → I. Define f ∗ : clβX H → I
to be the restriction of F to clβX H. Because clβX H is compact and
Hausdorff, H is dense in clβX H, and f extends continuously to f ∗ ,
clβX H ≈ βH by theorem 26. 

Proof of Corollary 279

Proof of Theorem 280

It suffices to show
\
β[0, 1) \ [0, 1) = clβ[0,1) [1 − 1/n, 1),
n∈N

since the intersection of a nested family of compact connected spaces

is connected.

212
 (⊆): For a contradiction suppose p ∈ β[0, 1) \ [0, 1) but n ∈ N such
that p ∈/ clβ[0,1) [1 − 1/n, 1). There exists an open U in β[0, 1) with
p ∈ U and U ∩ clβ[0,1) [1 − 1/n, 1) = ∅. Then U ∩ [0, 1) ⊆ [0, 1 − 1/n].
Since p ∈/ [0, 1) and [0, 1 − 1/n] is closed in β[0, 1), there exists an open
set V in β[0, 1) with p ∈ V and V ∩ [0, 1 − 1/n] = ∅. But U ∩ V is a
nonempty open set in β[0, 1) which misses [0, 1), a contradiction.
(⊇): Suppose p ∈ [0, 1). There exists n ∈ N such that p < 1 − 1/n.
Then {p} and [1 − 1/n, 1) are disjoint closed sets in [0, 1), so they have
disjoint closures in β[0, 1), i.e., p ∈
/ clβ[0,1) [1 − 1/n, 1).
Letting (−∞, −n] ∪ [n, ∞) play the role of [1 − 1/n, 1),
\ \
βR\R = clβR ((−∞, −n]∪[n, ∞)) = clβR (−∞, −n]∪clβR [n, ∞) =
n∈N n∈N

\ \
clβR (−∞, −n] ∪ clβR [n, ∞),
n∈N n∈N

so that βR \ R is the union of two connected components. 

Proof of Theorem 281

¬(c)→ ¬(a): Let G ⊆ X satisfy G 6∈ F and X \ G 6∈ F. If there does
not exist F ∈ F such that F ∩ G = ∅, then let F 0 = {F ∩ G : F ∈ F}
and observe F 0 is a filter and F 0 ⊇ F. Otherwise there exists H ∈ F
where H ∩ G = ∅, but then H \ G = H and for each F ∈ F, ∅ = 6
00
F ∩ H = F ∩ (H \ G) ⊆ F \ G. Thus F = {F \ G : F ∈ F} is a filter
and F 00 ⊇ F.
¬(a)→ ¬(c): Let F 0 ) F be a filter, and G ∈ F 0 \F. Then X \G 6∈ F 0 ,
and thus X \ G 6∈ F.
¬(b)→ ¬(a): F ∪{G} may be extended to a filter properly containing
F.
¬(a)→ ¬(b): Let F 0 ) F. Then there is a G ∈ F 0 which intersects
every member of F; therefore, G and X \ G are not in F. 

Proof of Theorem 282

213
9 Proofs

Proof of Theorem 283

βω
Let A ⊂ ω. To see (i), first note that if x ∈ ω then x ∈ A iff x ∈ A.
βω
Moreover, if x ∈ βω \ ω then x ∈ A iff every neighborhood of x meets
βω
A iff A ∈ Fx . Thus, A = A ∪ {p ∈ βω \ ω : A ∈ Fp }.
βω βω βω βω
To see (ii), observe that βω = A ∪ω \ A with A ∩ω \ A = ∅.
For (iii), let U ⊂ βω be open and let x ∈ U . By regularity, there
βω
exists an open V ⊂ βω with x ∈ V ⊂ V ⊂ U . Put A = V ∩ ω.
If x ∈ ω, then x ∈ A. If x ∈ βω \ ω, then A ∈ Fx . In either case,
βω βω
x∈A ⊂V ⊂ U. 
Proof of Theorem 284
Let X be the space defined above. Define ϕ : X → βω by ϕ(x) = x for
x ∈ ω and ϕ(F) = p for free ultrafilter F on ω and p ∈ βω \ ω satisfying
Fp = F. By theorem 32, ϕ is a well-defined bijection. Let A ⊂ ω. By
βω
theorem 33, {A : A ⊂ ω} forms a base for βω. We have
βω
ϕ−1 (A ) = ϕ−1 (A∪{p ∈ βω \ω : A ∈ Fp }) = A∪{F : A ∈ F} = A

thus demonstrating that the pre-image of a base element in βω is a

βω
base element in X. By a similar argument, ϕ(A) = A . We conclude
ϕ is a homeomorphism, and X ≈ βω. 

Proof of Theorem 285

Let (an ) be a nontrivial (not eventually constant) sequence of points in
ω. Then (an ) converges to no point in ω. Now suppose for a contradic-
tion that (an ) converges to p ∈ βω \ ω. Then (an ) is unbounded. Let
(ank ) be a strictly increasing subsequence of (an ). Then p ∈ clβω {an2k :
k ∈ ω} ∩ clβω {an2k+1 : k ∈ ω}, a contradiction. 

Proof of Lemma 286

Proof of Theorem 287

For each A ⊆ ω, let CA = {A ∩ n ⊆ ω : n < ω}.
Then for A 6= B, assume a ∈ A\B. Then as a ∈ A∩i for all i > a, but
a 6∈ B ∩ i for all i < ω, we conclude that CA , CB are almost disjoint. 

214
Proof of Corollary 288
Let A be as in the previous theorem. Then A∗ = {A∗ : A ∈ A} is a
pairwise disjoint collection (Lem 36) of clopen sets (Thm 33) of size
2ω . 
Proof of Lemma 289
Let D = {dn : n < ω} be a relatively discrete subset of a regular space
X. For each n < ω let Wn ⊂ X be open so that Wn ∩ D = {dn }. By
regularity, for each n < ω there exists open Vn ⊂ X with dn ∈ Vn ⊂
Vn ⊂ Wn . For each n < ω, put Un = Vn \ (∪i<n Vi ). Then {Un : n < ω}
is a pairwise disjoint family of open sets with dn ∈ Un for each n < ω.
Proof of Theorem 290
Let D ⊂ βω be countably infinite and relatively discrete. Let f : D → I
be continuous. By Lemma 39, there is a family of disjoint open sets
U = {Ud : d ∈ D} with d ∈ Ud for each d ∈ D. Define f ∗ : ω → I
by f ∗ (x) = f (d) if x ∈ Ud for some d ∈ D, and f ∗ (x) = 0 otherwise.
Because U is a family of disjoint sets, f ∗ is well-defined. Extend f ∗
continuously to F : βω → I. Since F (x) = f ∗ (x) = f (d) for every
x ∈ Ud ∩ ω, and Ud ∩ ω is dense in Ud , it follows that F (d) = f (d) for
every d ∈ D. Thus F  D is a continuous extension of f to D, hence
D ≈ βD. 
Proof of Corollary 291
The closure of any nontrivial convergent sequence D = {di : i ∈ ω}
is the singleton {d} consisting of the point d to which the sequence
converges, and {d} 6≈ βD. 

Proof of Lemma 292

Let X be an infinite Hausdorff space. If X is discrete, we are done.
Otherwise, X has a limit point x∗ . Let x0 ∈ X \ {x∗ }. Let U0 be open
in X with x∗ ∈ U0 and x0 ∈ / U 0 . There exists x1 ∈ U0 \ {x∗ }. Let
U1 ⊆ U0 be open in X with x∗ ∈ U1 and x1 ∈ / U 1 . Continue in this
manner. Then for any n ∈ ω, (X \ U n ) ∩ Un−1 is an open set containing
xn but no xi with i 6= n. So {xn : n ∈ ω} is an infinite discret subspace
of X. 
Proof of Theorem 293
Let X be an infinite closed subset of βω. Let A be a countably infinite

215
9 Proofs

discrete subspace of X. Clearly A is also discrete as a subspace of βω.

By Theorem 40, we have βω ' clβω A ⊆ X 

Proof of Lemma 294

Observe that ⊆∗ is a partial order: Let A ⊆∗ B and B ⊆∗ C. For
a ∈ A \ C, then either a ∈ B \ C, or a 6∈ B \ C. In the latter case, as
a 6∈ C, we have that a 6∈ B, and thus a ∈ A \ B. Therefore |A \ C| 6
|A \ B| + |B \ C| < ω.
T
It follows then that | i<n Ai | = ω; as this is true for n = 0, assume
it is true for n. Then as An ⊆∗ Ai for i < n, we have that
\ [
|An \ Ai | = | (An \ Ai )| < ω
i<n i<n
T
Since An is infinite, the induction hypothesis follows: |An ∩ i<n Ai | =
ω.
T
Finally, pick am ∈ i<m Ai such that am > ai for all i < m, and
define A = {am : m < ω}. Then for n < ω, am ∈ An for all m > n, and
thus |A \ An | < ω. 

Proof of Theorem 295

Let G be a non-empty Gδ -set in ω ∗ . Then there exists p ∈ G and
U0 , U1 , . . . open in ω ∗ such that G = ∩n<ω Un . Let A0 ⊂ ω so that
p ∈ A∗0 ⊂ U0 . If An ⊂ ω has been defined for some n < ω, let An+1 ⊂ ω
so that p ∈ A∗n+1 ⊂ A∗n ∩ Un+1 . Then A∗0 ⊃ A∗1 ⊃ . . . and G ⊃ ∩n<ω A∗n .
It follows that A0 ⊃∗ A1 ⊃∗ . . ., so there exists an infinite A ⊂ ω
such that A ⊂∗ An for each n < ω. Thus A∗ ⊂ A∗n for each n < ω, so
A∗ ⊂ G◦ . 
Proof of Theorem 296
For each n ∈ ω let An = ω ∩ [n, ∞). Enumerate P(ω) = {Xα : α < ω1 }
so that Xn = An for each n < ω. Suppose ω 6 α < ω1 and {Aβ : β < α}
has been defined so that
(i) Aγ ⊆∗ Aβ for β < γ < α, and
(ii) Aβ ⊆∗ Xβ for each β < α.
The sets Aβ ∩ [n, ∞), β < α, n < ω, have the finite intersection
property, so at least one of Xα and ω \ Xα intersects every one of them.
Choose one of the two that does, say Xα . Then |Xα ∩ Aβ | = ω for each

216
β < α and Xα ∩ Aγ ⊆∗ Xα ∩ Aβ for β < γ < α. By Lemma 44 there
exists an infinite Aα ⊆ ω such that Aα ⊆∗ Xα ∩ Aβ for each β < α.
Claim the set u in the hint is an ultrafilter. Well clearly it has the
finite intersection property, so it generates a filter even if it isn’t one
already. By construction every subset of ω or its complement is in u,
∗
so that u is an ultrafilter. It
Tis in ω since it includes the sets An .
Given any Gδ -set G = n∈ω Un containing u, for each n ∈ ω let
βn < ω1 such that Aβn ⊆ Un and let α = sup{βn : n < ω} + 1. Then
u ∈ A∗α ⊆ G. 

Proof of Lemma 297

Let X be an infinite compact T2 -space. By lemma 42, X contains an
infinite relatively discrete subset Y = {yi : i < ω}. Because X is
T2 (and hence T1 ), {yi } is closed in X for each i < ω. This implies
X \ Y = ∩i<ω X \ {yi } is Gδ .
Now, the closure Y of Y in compact X is also compact, so that Y 6= Y
(otherwise, {{yi } : i < ω} would be an open cover of Y for which there
is no finite subcover). Let x ∈ Y \ Y ⊆ X \ Y . Any open set containing
x meets Y , so x is not an element of the interior of X \ Y . Therefore,
X \ Y is not a neighborhood of x, and x is not a P -point. 

Proof of Corollary 298

Proof of Theorem 299

Suppose X is not countably compact. S Then let U = {Un : n < ω} be
an open cover such that xn ∈ Un \ i<n Ui . Then for each x ∈ Un , note
that Un misses xm for m > n, so x is not an infinite accumulation point
of {xn : n < ω}.
If X is countably compact, choose xn ∈ X. For n < ω, let Cn =
{xi : i > n}. Note that since X \ Cn ⊆ X \ Cn+1 , and X \ Cn ( X,
there is no finite subcollection of {X \ Cn : n < ω} covering X. Thus
as X is countably T compact, {X \ Cn : n < ω} is not an open cover.
So choose x ∈ n<ω Cn . For any open neighborhood U of x and
n < ω, U intersects {xi : i > n}. Thus U intersects infinitely many xn ,
making x an infinite accumulation point. 

217
9 Proofs

Proof of Corollary 300

Proof of Lemma 301

Let X, Y ⊂ βω with X ∩ Y = ω and put D = {(x, x) : x ∈ ω}.
Certainly D ⊂ X × Y is discrete. If (p, q) ∈
/ D then there exist disjoint
neighborhoods (in βω) U and V of p and q, respectively. Then (U ∩
X) × (V ∩ Y ) is a neighborhood of (p, q) missing D, so D is closed. 

Proof of Theorem 302

(a) There are at most 2c (infinite) closed subsets of ω ∗ since it has a
basis of size c (note that there are actually 2c by Corollary 38). Let
{Cα : α < κ} be an enumeration of the infinite closed subsets of ω ∗ ,
where κ 6 2c is a cardinal. By Theorem 43, each Cα has cardinality
2c . Recursively define X = {xα : α < κ} and {yα : α < κ} so that
xα ∈ Cα \ {yβ : β < α} and yα ∈ Cα \ {xβ : β 6 α} for each α < κ.

(b) Let X be a subset of ω ∗ such that X and ω ∗ \X meet every infinite

closed subset of ω ∗ . Let Y be an infinite subset of ω ∪ X. By lemma 42,
Y contains a countably infinite relatively discrete subset Z. The closure
Z of Z in βω contains a copy of βω by theorem 43, and is uncountable.
Let x ∈ (ω ∪ X) ∩ (Z \ Z). As a limit point of Z, x is also a limit point
of Y . So ω ∪ X is countably compact by corollary 50. 

Proof of Corollary 303

Proof of QTheorem
304
ω
Let x ∈ n<ω Xn . Since X0 is sequentially compact, it follows that
there exists a subsequence x0 of x such that π0 (x0 ) converges in X0 .
If xn is a subsequence of x, then as Xn+1 is sequentially compact,
there exists a subsequence xn+1 of xn such that πn+1 (xn+1 ) converges
in Xn+1 . (In fact, πi (xi ) converges in Xi for all i 6 n + 1.)
Define fn : ω → ω such that xn (i) = x(fn (i)) and ran(fn ) ⊇
ran(fn+1 ). It follows by REFERENCE that there exists an increas-
ing f : ω → ω such that ran(f ) ⊆∗ ran(fn ) for all n < ω.

218
 Q
ω
Let y ∈ n<ω Xn such that y(n) = x(f (n)).QIt follows that πn (y)
converges in Xn for all n < ω, so y converges in n<ω Xn . 

Proof of Theorem 305

Recall βω = eF (ω) ⊂ I F , with F = C(ω, I). Since

|F| = |C(ω, I)| = (2ω )ω = 2ω·ω = 2ω = |I|,

βω is homeomorphic to a closed subset of I I . As βω is not sequentially

compact, neither is I I . 

Proof of Theorem 306

We show that if κ < t then every subset of ω ω of size κ is <∗ -bounded,
so that b 6< t.
Let κ < t and let {fα : α < κ} ⊆ ω ω . Let g0 ∈ ω ω be increasing with
f0 < g0 . Suppose α 6 κ and an increasing gβ has been defined for each
β < α so that fγ <∗ gβ and ran gβ ⊆∗ ran gγ for γ 6 β. There exists
an infinite X ⊆ ω such that X ⊆∗ ran gβ for each β < α. Let gα be
increasing with range ⊆ X skipping infinitely many members of X and
fα < gα . Using the hint we have fγ < gγ <∗ gα for γ 6 α. Continuing
in the manner gκ will be a <∗ -upper bound for {fα : α < κ}. 

Proof of Theorem 307

Let κ ∈ c. Let {Aα : α < κ} be a tower of length κ. We will show
this tower has an infinite pseudo-intersection. Let P be the collection
~ F ), where C
of all pairs of the form (C, ~ is a finite sequence of distinct
elements in ω, and F is a finite subset of κ. For (C, ~ 0, F 0) ∈ P ,
~ F ), (C
0 0
define (C~ , F ) < (C,
~ F ) iff

~ 0 extends C,
• C ~
0
• F ⊇ F , and
~ 0 ) \ dom(C),
• for each i ∈ dom(C ~ we have C 0 ∈ T
i α∈F Aα .

(<, partial order on P ?) Reflexivity and antisymmetry of < are triv-

ially satisfied. For transitivity, let (C, ~ 0 , F 0 ), (C
~ F ), (C ~ ” , F ” ) ∈ P such
0 0
that (C~ , F ) < (C
” ” ~ , F ) < (C,
~ F ). Clearly, C~ extends C
” ~ and F ” ⊇ F .
0 00
Let i ∈ dom(C ~ )\dom(C).
” ~ If i ∈ dom(C ~ ), then C ∈ T
Aα because
i α∈F

219
9 Proofs

~ 0 , F 0 ) < (C, ~ F ). If i 6∈ dom(C ~ 0 ), then C 00 ∈ T T

(C i α∈F 0 Aα ⊆ α∈F Aα ]
0 0
~ ” ” ~
because (C , F ) < (C , F ). So < is a partial order on P .
((P, <) has the ccc?) Let X = {(C ~ α , Fα ) : α ∈ Λ} be an uncountable
collection of elements of P . There are only a countable number of finite
sequences of distinct elements in ω, so there exist distinct α1 , α2 ∈ Λ
~α = C ~ α and Fα = ~ α , Fα ∪Fα ) ∈ P such
such that C 1 2 1 6 Fα2 . We have (C 1 1 2
that (C ~ α , Fα ∪Fα ) < (C ~ α , Fα ) and (C ~ α , Fα ∪Fα ) < (C ~ α , Fα ), so
1 1 2 1 1 1 1 2 2 2
that X is not an antichain. Therefore, every antichain of P is countable,
and (P, <) has the ccc.
Now, for each i ∈ ω, define Di = {(C, ~ F ) ∈ P : i ∈ dom(C)}. ~ For
each α ∈ κ, define Eα = {(C, F ) ∈ P : α ∈ F }. ~
(Eα dense for each α ∈ κ?) Let (C, ~ F ) ∈ P and α ∈ κ. (C, ~ F ∪{α}) ∈
~ ~
Eα and (C, F ∪ {a}) < (C, F ), so that Eα is dense in P .
(Di dense for each i ∈ ω?) Let (C, ~ F ) ∈ P and i ∈ ω. Wlog, assume
i 6∈ dom(C). ~ Because {Aα : α < κ} is a tower, the set T
α∈F Aα is
0
infinite. So we may (and do) extend ~ ~
C to C for which i ∈ dom(C ~ 0)
T
by selecting distinct elements of α∈F Aα not already appearing in C. ~
0 0
(C~ , F ) < (C, ~ F ) and (C ~ , F ) ∈ Di , so that Di is dense in P .
The collection D = {Eα : α ∈ κ} ∪ {Di : i ∈ ω} is a collection of
6 κ-many dense sets, so by Martin’s Axiom, there exists a filter G
in P such that G ∩ D = 6 ∅ for every D ∈ D. If (C, ~ F ), (C~ 0 , F 0 ) ∈ G,
then there is some (C ~ ” , F ” ) ∈ G such that (C ~ ” , F ” ) < (C,~ F ) and
0 0
(C~ , F ) < (C
” ” ~ , F ), and hence the sequence C ~ extends both C
” ~ and
C~ 0 . So either C ~ extends C ~ 0 or vice-versa. Define C ~ and F so that
~ F ) ∈ Di ∩ G for each i ∈ ω.
(C,
({Ci : i ∈ ω}, pseudo-intersection?) Let α ∈ κ. Let (C, ~ F ) ∈ Eα ∩ G.
0 0
~
Suppose i 6∈ dom(C). There is (C , F ) ∈ G ∩ Di , and (C ~ ~ ”, F ”) ∈ G
0 0
~ ” ” ~ ~ ” ” ~ ~”
with (C , F ) < (C , F ) and (C , F ) < (C,
~ ~ ” ” ~ ”
T F ). So i ∈ dom(C and
i 6∈ dom(C). Since (C , F ) < (C, F ), Ci ∈ β∈F Aβ ⊆ Aα . 

Proof of Theorem 308 Q

Let κ < t and let X = α<κ Xα with each Xα sequentially compact.
Let (xn )n<ω be a sequence in X. Since X0 is sequentially compact,
take A0 ⊂ ω and y0 ∈ X0 so that (xn (0))n∈A0 → y0 . Suppose for some
β < κ that for each α < β we have defined Aα ⊂ ω and yα ∈ Xα so
that (xn (α))n∈Aα → yα . If β = γ + 1, let Aβ ⊂ Aγ and yβ ∈ Xβ so that

220
(xn (β))n∈Aβ → yβ . If β is a limit ordinal, let A0β be an infinite pseudo-
intersection of the tower {Aα : α < β}. Take Aβ ⊂ A0β and yβ ∈ Xβ
so that (xn (β))n∈Aβ → yβ . Let A be an infinite pseudo-intersection
of {Aα : α < κ}. Then (xn )n∈A → (yα )α<κ , so X is sequentially
compact. 
Proof of Lemma 309
For (a), let x ∈ X and U be an open neighborhood of x. Since X is
completely regular, there exists a continuous function fU : X → I such
that fU (x) = 0 and fU (y) = 1 for all y 6∈ U . So let FU = {z : fU (z) 6
0.5} ⊆ U ; note that this is a zero-set neighborhood of x. Likewise,
let GU = {z : fU (z) < 0.5} ⊆ U ; note that this is a cozero-set open
neighborhood of x.
For (b), note that if F is a zero-set, then F = f −1 [{0}] for some
continuous f : X → R, so F is closed. Furthermore, letting Un = {x :
f (x) < 2−n } witnesses that F is a regular Gδ set.
T For (c), assume
T the Un are decreasing. If X is normal and F =
−n
U
n∈ω n = n∈ω n , then let fn : X → [0, 2
U ] be continuous and
−n
P all x ∈ Un+1 and fn (y) = 2
satisfy fn (x) = 0 for for all x 6∈ Un . It
follows that f = n<ω fn is a continuous function from X to [0, 2]
such that f −1 [{0}] = F . 
Proof of Lemma 310

Proof of Theorem 311

(a) Let F1 , F2 ∈ Fp . By Lemma 310(b), F1 ∩ F2 is a zero-set. Suppose
βω
p 6∈ F1 ∩ F2 . Using Lemma 309(a), let G be a zero-set neighborhood
βX
of p satisfying G ∩ (F1 ∩ F2 ) 6= ∅. By Lemma 310(c), G ∩ X is a zero-
set in X and, again by Lemma 310(b), G ∩ F1 and G ∩ F2 are zero-sets
βX βX
in X. We have p ∈ (G ∩ F1 ) ∩ (G ∩ F2 ) but (G ∩ X) ∩ F1 ∩ F2 =
βX
(G ∩ F1 ) ∩ (G ∩ F2 ). So p ∈ F1 ∩ F2 , and F1 ∩ F2 ∈ Fp . It is easy to
check the other requirements that Fp be a z-filter.
Let F be a filter with Fp ⊂ F, and let G ∈ F. Let U ⊂ βX be open
and contain p. Using Lemma 309(a), let F be a zero-set neighborhood
βX
of p contained in U . F ∩ X is a zero-set in X, and F ∩ X ⊃ {p},

221
9 Proofs

βX
so F ∈ Fp . U ∩ G ⊃ F ∩ G =
6 ∅, thus p ∈ G , g ∈ Fp , and Fp is
maximal.
βω βω
(b) Let F be a zero-set in X. F = F ∪ {p ∈ βX \ X : p ∈ F }=
F ∪ {p ∈ βX \ X : F ∈ Fp }.

(c) Let x ∈ βX and U be an open neighborhood of x. There exists a

cozero-set V such that x ∈ V ⊆ U . Then X \ V is a zero-set in X, and
X \ V = βX \ V . Thus x ∈ βX \ X \ V ⊆ U .

(d) Let U be cozero in X. Note X \ U = X \U ∪{p ∈ βX \X : X \U ∈

Fp }. As Fp is a z-ultrafilter, we have

βX \ X \ U = U ∪ {p ∈ βX \ X : X \ U 6∈ Fp }

= U ∪ {p ∈ βX \ X : ∃F ⊆ U (F ∈ Fp )}
If x ∈ βX \ F for some zero-set F in X, then U = X \ F is a cozero-set
in F where x ∈ βX \ F = βX \ X \ U = Ex U , so the Ex U ’s form a
base.

(e) Let U be cozero in X. Suppose p 6∈ Ex U = βX \ (X \ U ). Then

p ∈ X \ U , so any open neighborhood of p hits the complement of U ,
so p cannot be in an open set whose intersection lays inside U . 

Proof of Theorem 312

Let X be a continuum. The reverse implication is immediate, for if
H and K are proper subcontinua of X with X = H ∪ K we have
X \ H ⊂ K ◦.
For the forwards direction, suppose H is a proper subcontinuum of
X with H ◦ 6= ∅ and put Y = X \ H ◦ . If Y is connected, then Y
is a proper subcontinuum and X = H ∪ Y . Otherwise, let C be the
collection of components of Y . Pick K ( Y clopen (in Y ) and observe
that K = ∪{C ∈ C : K ∩C 6= ∅}. Since each C ∈ C meets ∂Y , and thus
H, H ∪ K is a proper subcontinuum of X. Similarly, H ∪ (Y \ K) is a
proper subcontinuum of X. It follows that X is not indecomposable.

222
Proof of Lemma 313
Wlog, assume O is unbounded. Put a0 = 0, and let x0 ∈ O such that
x0 > a0 . Choose h0 ∈ H \ O such that h0 > max{x0 , 1}. Because
h0 ∈ H \ O, there is an open interval (b0 , a1 ) containing h0 missing O.
For each n ∈ ω, select xn > an , hn ∈ H \ O with hn > max{xn , n + 1},
and bn , an+1 ∈ H \ O with cn ∈ (bn , an+1 ) ⊂ H \ O. 

Proof of Lemma 314

(a) Let U ⊆ βH be open with K ⊆ U and H∗ ∩ clβH U ( H∗ . Let
O = U ∩ H. Assume 0 ∈ / O. Since H \ O is unbounded,
S by Lemma 63
there areSnumbers an < bn < an+1 so that O ⊆ n∈ω (an , bn ). Then
K ⊆ Ex n∈ω (an , bn ).

(b) Since clβH (L1 ∩ L2 ) = clβH L1 ∩ clβH L2 whenever L1 and L2 are

S subsets of H, it is clear that u is a filter. By part (a), K ⊆
closed
clβH Sn∈ω [an , bn ]. Let A ⊆Sω. If neither A ∈ u nor ω \ A ∈ u, then
clβH n∈A [an , bn ] and clβH n∈ω\A [an , bn ] disconnect K. 

Proof of Proposition F
If p ∈ Ex O \ X, then there exists F ∈ Fp where F ⊆ O. Since F ∈ Fp ,
βX βX
p∈F ⊆O . 
Proof of Theorem 315 S
Let K, L be a proper subcontinuum, so let UA = n∈A (an , bn ) for
A ⊆ ω and uK be the ultrafilter of subsets of ω such that

K ⊆ UA

for each A ∈ uK as in the previous lemma.

We proceed by showing that H∗ \ K is dense. Consider an arbitrary
open basic set Ex V . Then consider

A = {n < ω : V ∩ (an , bn ) 6= ∅}.

Then either A is finite, or there exists an infinite subset of A which

is not in uK . Either way, let A0 ⊆ A such that A0 6∈ uK . Then

V ∩ UA0

223
9 Proofs

has a distinct closure from Uω\A0 ⊇ K.

Thus Ex(V ∩ UA0 ) hits a point distinct from K, showing that H∗ \ K
is dense. 
Proof of Lemma 316
Let p ∈ αX \ X and x0 ∈ X. Let U, V be open disjoint neighborhoods
of p, x0 respectively in αX. Note p ∈ U ∩ X. Then by the continuity of
f , f (p) ∈ f (U ∩ X) ⊆ f (U ∩ X) = U ∩ X, so f (p) 6= x0 . 

Proof of Theorem 317

Note all the backwards implications are immediate.
For (a)→(b), let X be a Gδ subset of αX, and then let f : βX →
αX such that f  X is the identity map. Then there T exist open in
αX neighborhoods U n for n < ω such that X = n<ω Un . Then as
X = f −1 [ n<ω Un ] = n<ω f −1 [Un ], we see X is Gδ in βX.
T T

T For (b)→(c), let αX be any compactification

∗
S of X. Then as X =
n<ω nU for U n open in βX, we have X = n<ω KS n for Kn com-
∗
S
plements of Un in βX. Since f (X ) = f ( n<ω Kn ) = n<ω f (Kn ) =
αX \ X, it follows that αX \ X is Fσ , so X is Gδ in αX.
For (c)→(d), if X is Gδ in αX, and X is a dense subset of Z, then
X is also a dense subset of βZ, which is a compactification of X, so X
is Gδ in βZ, and therefore also in Z. 

Proof of Theorem 318

Let X be Čech complete. Using theorem 67, let αX be a compactifica-
tion in which X = ∩i∈ω Ui for open Ui ⊂ αX, i ∈ ω. For each i ∈ ω, let
αX
Ui be an open cover of X satisfying U ⊂ Ui for every U ∈ Ui .
Let F be a filter of closed sets such that for each n there is some
Fn ∈ F with Fn ⊂ U for some U ∈ Un . Because αX is compact,
T αX T αX
F ∈F F 6= ∅. Let x ∈ F ∈F F . For some Vn ∈ Un , there is
αX αX T αX
an Fn ∈ F with x ∈ F n ⊂ V n ⊂ Un so that x ∈ n∈ω F n ⊂
T
n∈ω Un = X.
Let U0 , U1 , U2 , . . . be a sequence of open covers of X with the desired
property. For eachS n < ω let Vn = T {V ⊂ βX : V open, V ∩ X ∈ Un }
and put Vn = Vn . Certainly X ⊂ n<ω Vn .
Let x ∈ n<ω Vn and put F 0 = {X ∩O : x ∈ O, O open in βX}. Note
T
that if O1 and O2 are open neighborhoods of x we have X ∩(O1 ∩O2 ) ⊃

224
X ∩O1 ∩ O2 6= ∅. Let F be the filter of closed sets generated by F 0 . Fix
n < ω and pick V ∈ Vn such that x ∈ V . Let W ⊂ βX be open with
x ∈TW ⊂ W ⊂ V . Then X ∩ W ∈ F, X ∩ V ∈ Un and X ∩ W ⊂ X ∩ V ,
so F 6= ∅. It follows that
\ \
X ∩ {O : x ∈ O, O open in βX} = F 0 6= ∅.
T T
Since {O : x ∈ O, O open in βX} = {x} we have X = n<ω Vn , so
X is Gδ in βX. 
Proof of Exercise 319
False.
Let S = [0, 1) have the Sorgenfrey topology (where S \ {0} is a dense
copy of the Sorgenfrey line), and R = (0, 1] have the reverse Sorgenfrey
topology. These spaces embed in S × {0} ∪ R × {1}, where for each
hx, 0i,
[x, x + ε) × {0} ∪ (x, x + ε) × {1}
is a basic open set, and for each hx, 1i,

(x − ε, x] × {1} ∪ (x − ε, x) × {0}

is a basic open set.

A modification of the usual Heine-Borel theorem for the real line
shows that this space is compact. Furthermore, any open set containing
S × {0} misses at most countably many points of R × {1}, so S × {0}
is not Gδ .
True. Let X be the tangent disk space. If x is on the x-axis, let
D(x, ε) be the tangent disk neighborhood of x of radius ε, and if x is in
the upper half plane, let B(x, ε) be the standard Euclidean open ball
about x of radius ε. For each n ∈ ω, let

Un = {D(x, 1/2n ) : x ∈ R × {0}} ∪ {B((x1 , x2 ), 1/2n+1 ) : x2 > 1/2n }.

Let F be a filter of closed sets such that for each n there is some
Fn ∈ F with Fn ⊂ Un for some Un ∈ Un . T
Case 1. Some F ∈ F is compact. Then clearly F 6= ∅.
Case 2. No F ∈ F is compact. Then for each n, Un = D(xn , 1/2n )
for some point xn on the x-axis. 

225
9 Proofs

Proof of Theorem 320

(a)⇒(b): Suppose X is metrizable and Čech-complete. Then by The-
orem 67 X is Gδ in its metric completion. By Completeness Fact (b),
X is completely metrizable.
(b)⇒(a): Let d be a complete metric on X. For each n ∈ ω let
Un = {Bd (x, 1/n) : x ∈ X}. Now suppose F is a filter of closed subsets
T n ∈ ω there is Fn ∈ F and U ∈ Un such that
of X such that for each
Fn ⊆ U . Let Fn0 = k6n Fk . Then the sets Fn0 are nonempty closed
decreasing
T 0 with diameters going to 0. Since d is complete, we have
Fn 6= ∅. Clearly there can be only one point x in this intersection.
T F0 ∈ F. Again using completeness of dTand0 the fact that F is a filter,
Let
T(Fn ∩ F ) 6= ∅, so it must be that x ∈ (Fn ∩ F ). Thus x witnesses
F 6= ∅. By Theorem 68(b), X is Čech-complete. 

Proof of Corollary 321

Proof of Theorem 322 T

As Y is Čech-complete, let Y = n<ω Vn witness that Y is Gδ in some
compactification Z.
If X = n<ω Un for Un open in Y , then let Un0 be open in Z with
T
Z
Un = Un0 ∩ Y . Then X = n<ω Un0 ∩ Vn ∩ X , showing that X is Gδ
T
Z
in its compactification X .
T
If X is closed in Y , then X = K ∩ Y = K ∩ n<ω Vn ⊆ K ∩
T Z
n<ω Vn for some closed (compact) subset K of Z. Thus X is Gδ in
T Z Z
its compactification K ∩ n<ω Vn = X .
Proof of (b) missing. 

Proof of Example 323

Let U0 , U1 , . . . be open in (ω + 1)ω1 such that ω ω1 ⊂ n<ω Un . For each
T

n < ω, let Vn ⊂ Un be a basic open set containing ~ Fn ⊂ ω1 be

Qβ0 and let Qω1
the support of V n . Put β = sup(∪n<ω F n ). Then α=0 {0}× α=β+1 ω+
1 ⊂ n<ω Vn ⊂ n<ω Un , so ω ω1 6= n<ω Un .
T T T


Proof of Theorem 324

Let X be a Čech complete space. Let {Dn : n ∈ ω} be a countable

226
collection of dense open sets in T X. Let {Un : n ∈ ω} be a collection
of
T open sets in βX with X =
T n∈ω Un . Since βX is a Baire space,
n∈ω (Ex Dn ∩ Un ) 6
= ∅. So n∈ω Dn 6= ∅. 

Proof of Example 325

βω
Consider any basic open set A ∩ ω ∗ . It follows that A is infinite, so
βω βω
therefore A ∩ ω ∗ is closed and infinite. Thus X intersects A ∩ ω ∗ ,
showing that X is dense in ω ∗ .
Let Un be open in ω ∗ with X ⊆ Un for n < ω. Then ω ∗ \ Un ⊆ ω ∗ \ X.
Note then that ω ∗ \ Un cannot be infinite, so Un is co-finite. Then
∗
T
Tco-countable, but since X,ω \X are uncountable, it follows
n<ω Un is
that X 6= n<ω Un . 

Proof of Theorem 326

Let X be a regular countably compact space. Let {Un }n∈ω be a col-
lection of dense open sets. Let V ⊆ X be open. Using regularity, let
V0 =6 ∅ be open with V0 ⊆ V0 ⊆ (U0 ∩ V ). For n > 0, let Vn 6= ∅ be
open with Vn ⊆ Vn ⊆ (U0 ∩ Vn−1 ). V0 ⊇ V1 ⊇ . . . is a decreasing
T chain
of nonempty closed sets in a countably compact
T space,
T so n∈ω n 6= ∅.
V
For each
T n ∈ ω, V ⊆
nT U n and Vn ⊆ V , so V
n∈ω n ⊆ n∈ω (Un ∩V
T ), and
V ∩ ( n∈ω Un ) = n∈ω (Un ∩ V ) is nonempty. We conclude n∈ω Un
is dense, and X is a Baire space. 

Proof of Theorem 327

Suppose on the contrary that B is completely metrizable. Then T B is a
Gδ -set in R, so there exist open sets Un ⊂ R, n < ω, with B = n<ω Un .
Observe that X \ Un is a closed set missing B, so X S \ Un must be
countable for each n < ω. It follows that X \ B = n<ω X \ Un is
countable, a contradiction as X \ B must be Bernstein also.
To see that B is Baire, let G0n be a dense open subset of B for each
n < ω and let U 0 ⊂ B be open. For each n < ω let Gn be open in R
such that G0n = Gn ∩ B, and let U be open in R such that U 0 = U ∩ B.
Observe that each Gn is dense in R.
Since U ∩ G0 is open and nonempty, there is a closed interval I∅
contained therein with diam I∅ < 1. Suppose for some n > 0 that
{Iσ : σ ∈ 2n−1 } is a pairwise-disjoint
Tn−1collection of closed intervals with
diam Iσ < 1/2n−1 and Iσ ⊂ U ∩ i=0 Gi . For each σ ∈ 2n−1 , Iσ◦ ∩ Gn

227
9 Proofs

is nonempty and open, so pick disjoint closed intervals Iσa0 and Iσa1
contained therein with diameters less than 1/2n .
Put C = n<ω ∪{Iσ : σ ∈ 2n }. If σ ∈ 2ω then diam Iσn
T
T → 0, so
∩{Iσn : n < ω} is a single point. It follows thatTC ⊂ U ∩ n<ω Gn is
a Cantor set. As B must meet C we have U 0 ∩ n<ω G0n = 6 ∅, whence
B is Baire. 
Proof of Theorem 328
(a): Suppose Y is Baire and X is a subspace of Y . For each T n ∈ ω let
Dn be dense open in X. In each case below we shall show that n∈ω Dn
is dense in X.
(a)(i): Suppose X is open in Y . Let U be nonempty open in X. Since
T open in Y and each Dn ∪ Y \ X is dense open in
U is also (nonempty)
Y , we
T have U ∩ n∈ω (Dn ∪ Y \ X) 6= ∅. Since U ⊆ X we must have
U ∩ n∈ω Dn 6= ∅.
T X is a dense Gδ in Y . There exist open sets Xn ⊆ Y
(a)(ii): Suppose
such that X = n∈ω Xn . For each n ∈ ω there exists an open Dn0 ⊆ Y
such that Dn = Dn0 ∩ X. Note for each n ∈ ω we have Dn0 ∩ Xn dense
open in Y . Let U be nonempty open in X. There Texists an open U 0 ⊆ Y
0 0 0
suchTthat U = U ∩ X. Since Y is Baire, U ∩ n∈ω Dn ∩ Xn 6= ∅. So
U ∩ Dn 6= ∅.
(b): Suppose Y is a dense Baire subspace of X.TLet {Dn : n ∈ ω} be a
collection of dense open subsets of X. We show n∈ω Dn is dense in X.
Claim: Dn ∩ Y is dense (open) in Y , for each n ∈ ω. Let V be nonempty
open in Y . Then V = V 0 ∩ Y for some V 0 (nonempty) open in X. Then
Dn ∩ V 0 is nonempty open in X, so Dn ∩ V = (Dn ∩ V 0 ) ∩ Y 6= ∅
by density of Y in X. Done with claim. Let U be nonempty open in
X. Then U ∩ Y is nonempty open in Y . Since T Y is Baire and each
Dn ∩TY is dense open in Y , we have (U ∩ Y ) n∈ω (Dn ∩ Y ) 6= ∅. So
U ∩ n∈ω Dn 6= ∅. 

Proof of Example 329

Let X = R \ (C \ E), where C is the Cantor set, and E is the set of
endpoints of the Cantor set. Note that R \ C is a dense Baire subspace,
so X is Baire. However, E is a countable subspace of metrizable X
with no isolated points, and E = X ∩ C, so E is a closed copy of the
rationals. 

228
Proof of Theorem 330
Let X be Baire, and let Y be Baire and second-countable. Let {Bn }n∈ω
be a base for Y , and let {On }n∈ω be a collection of dense open subsets
of X × Y .
Define Uij = πX (Oi ∩ (X × Bj )). Clearly, Uij is open in X × Y . Let
W ⊆ X be open and nonempty. Since W × Bj is open in X × Y and
Oi is dense, Oi ∩ (W × Bj ) is nonempty. Thus, for each i, j ∈ ω, Uij is
dense and open in X.
Now, let U × VTbe a basic open subset of X × Y . Because X is Baire,
there is x ∈ U ∩ i,j∈ω Uij . Define Vi = {y ∈ Y : (x, y) ∈ Oi } for each
i ∈ ω. Vi is open in Y since for each y ∈ Vi , there exists a basic open set
Uy × Vy ⊆ X × Y satisfying (x, y) ∈ Uy × Vy ⊆ Oi , so that y ∈ Vy ⊆ Vi .
To show Vi is dense in Y , let Bj be a basic open subset of Y . Because
x ∈ Uij = πX (Oi ∩ (X × Bj )), there exists (x, y) ∈ Oi ∩ (X × Bj ) so
that y ∈ Vi ∩ Bj .
T
T Because Y is Baire, there is y ∈ V ∩ i∈ω Vi . So (x, y) ∈ (U × V ) ∩
i∈ω Oi , and X × Y is Baire. 

Proof of Theorem 331

Let On be open in X × Y for n < ω. Consider T any basic open U × V
in X × Y . We need to prove (U × V ) ∩ n∈ω On 6= ∅.
Let {Uα }α be a maximal disjoint collection of open subsets of U
such that for each α there exist Vα open in Y such that Uα × Vα ⊆
O0 ∩ (U × V ).
S
Claim. W1 = α Uα is dense in U . Suppose N is any nonempty
open subset of U . Since O0 is dense open in X × Y , O0 ∩ (N × V ) 6= ∅.
So there exists a basic open set A × B ⊂ O0 ∩ (N × V ). By regularity
of Y we may assume A × B ⊂ O0 ∩ (N × V ). A can’t be disjoint from
W1 , as that would contradict maximality of {Uα }α . Since A ⊂ N , we
have N ∩ W1 6= ∅, proving W1 dense in U .
Now for each α, let {Uαβ }β be a maximal disjoint collection of open
subsets of Uα such that for each β there exist Vαβ open in Y such that
Sαβ × Vαβ ⊆ O1 ∩ (Uα × Vα ). By a proof similar
U S to that of the Claim,
β U αβ is dense in U α . It follows that W 2 = α,β Uαβ is dense in U .
Continue in this way, defining W3 , W T 4 , ... dense open in U . Since X is
Baire, so is U , hence there exists x ∈ n∈N Wn . Since {Uα }α is pairwise
disjoint, there exists a unique α0 with x ∈ Uα0 . Similarly, there exists a

229
9 Proofs

unique α1Twith x ∈ Uα0 α1 . And so on. So we obtain a sequence α0 , α1 , ...

with x ∈ n∈ω Uα0 α1 ...αn . By construction, Vα0 ⊃ Vα0 α1 ⊃ ....
T
If Y is countably
T compact, there
T exists y ∈ n∈ω V α0 α1 ...αn . Then
(x, y) ∈ (U × V ) ∩ n∈ω On , so n∈ω On is dense. Hence X × Y is Baire.
If Y is Čech-complete, let Cn , n ∈ ω, be a sequence of open covers
satisfying the conditions of Theorem 318(b). In the above construction,
we may choose T V α0 α1 ...αn ⊂ C for some C ∈ Cn , guaranteeing the
existence of y ∈ n∈ω V α0 α1 ...αn , and then the rest of the proof proceeds
as before. 
Proof of Theorem 332
Suppose X is Baire and let σ be a strategy for E. Let U = σ(∅) be E’s
first move. Let {Uα }α be a maximal disjoint collection of open subsets
S that, for each α, there is Vα ⊂ U such that Uα = σ(Vα ). Then
of U such
W1 = α Uα is dense open in U . For each α, let {Uαβ }β be a maximal
disjoint collection of open subsets of Uα such that, for
S each β, there is
Vαβ ⊂ Uα such that Uαβ = σ(Vα , Vαβ ). Then W2 = α,β Uαβ is dense
open in U . Define W3 , W4 , . . . dense
T open in U by continuing in this
way. U is Baire, so there exists x ∈ n∈N Wn . Since {Uα }α is pairwise-
disjoint, there exists a unique α0 with x ∈ Uα0 . Similarly, there exists
a unique
T α1 with x ∈ Uα0 α1 . We obtain a sequence α0 , α1 , . . . with
x ∈ n<ω Uα0 ,α1 ,...,αn . By construction, Vα0 ⊃ Vα0 α1 ⊃ . . .. Thus NE
wins by playing Vα0 α1 ...αn in round n, so σ is not a winning strategy
for E.
Now suppose X is not Baire. S Without loss of generality we may
assume X is meagre, so X = n∈N Xn with each Xn nowhere-dense.
Let U0 = X be E’s first move. If NE plays Vn in round n, let E play
Un+1 = Vn \ X n+1 in round n + 1. Certainly this is a winning strategy
for E. 
Proof of Theorem 333
If X is regular and countably compact, then NE just needs to pick the
sequence of Vn ’s so that V n ⊆ Un for each n ∈ ω.
T
Now suppose X is Čech complete. Then X = n∈ω Wn for some
decreasing sequence of open subsets Wn of βX. Suppose n ∈ ω and
the game has been played properly through stage n − 1, and E has
selected a legal Un . Player NE can let Vn be nonempty open in X with

230
clβX Vn ⊆ Wn ∩ Ex Un . If the game continues in this manner, then
\ \ \
∅ 6= clβX Vn ⊆ X ∩ clβX Vn ⊆ Un ,
n∈ω n∈ω n∈ω

so that NE wins. Alternatively, let U0 , U1 , ... be a sequence of open

covers of X given by Theorem 68. If Un has been chosen, then there
exists W ∈ Un such that W ∩ Un 6= ∅. Let Vn be nonempty open in
X with closure contained in W ∩ Un . Continuing in this manner, the
closures of Vn ’s form a filter with the desired property. AnTelement in
the intersection of this filter corresponds to an element in n∈ω Un . 

Proof of Theorem 334

Note that E 6↑ Ch(B) and E 6↑ Ch(R \ B) as Bernstein sets are Baire.
If σ is a strategy for player N E in Ch(B), then define a strategy τ
for E in Ch(R \ B) such that τ (∅) = R and

τ (sahU i) ⊆ τ (s) ∩ σ(sahU i)

such that the diameter of τ (sahU i) is less than 2−|s| for all sequences
s of open sets. Since E lacks a winning strategy in Ch(R T \ B), there
exists a counterattack hV0 , V1 , . . .i such that (R \ B) ∩ n<ω Vn 6= ∅.
Since the diameters of T
the sets played according to τ converge to 0, it
follows that (R \ B) ∩ n<ω Vn = {x}. Therefore since

Vn ⊆ τ (hV0 , . . . , Vn−1 i) ⊆ σ(hV0 , . . . , Vn−1 i)

it follows that hV0 , V1 , . . .i is a valid counterattack to σ such that B ∩

T
n<ω Vn = ∅. Thus σ is not a winning strategy.
Therefore E 6↑ Ch(B) and N E 6↑ Ch(B), so Ch(B) is undetermined.

Proof of Theorem 335

Proof of QTheorem 336

Let X = α<κ Xα , with each Xα Choquet. We show NE has a winning
strategy in Ch(X, B), where B is the collection of basic open sets. For
each α < κ let σα be a winning strategy for NE in Xα . Let Un =

231
9 Proofs

Q
α<κ Uα,n denote E’s play in round n, and let Fn be the support of
Un . For each α < κ and n < ω, put
(
σα (Uα,iα , . . . , Uα,n ) if α ∈ Fn
Vα,n = ,
Xα if α ∈
/ Fn
Q
where iα = min{i : α ∈ Fi }. For each n < ω let Vn = α<κ Vα,n . Then
\ \ Y Y \
Vn = Vα,n = Vα,n 6= ∅,
n<ω n<ω α<κ α<κ n<ω

so NE has a winning strategy in Ch(X, B). 

Proof of Corollary 337

Proof of Lemma 338

(a): Let A ⊆ ω1 uncountable. Let σ ∈ ω1<ω . A is cofinal in ω1 , so there
exists α ∈ A with α > ran σ. Let k ∈ ω \ dom σ. Define f ∈ ω1ω by
f  dom σ = σ, f (k) = α, and f (n) = 0 for all n ∈ ω \ (dom σ ∪ {k}).
Then f ∈ [σ] ∩ A∗ .
(b): Claim Ui is open in ω1ω × ω1ω . Let (f, g) ∈ Ui . There exist
k1 , k2 ∈ ω such that min{f (k1 ), g(k2 )} > max{f (i), g(i)}. Let σ = f 
{i, k1 } and τ = g  {i, k2 }. Then (f, g) ∈ [σ] × [τ ] ⊆ Ui . Claim Ui is
dense in ω1ω × ω1ω . Let σ, τ ∈ ω1<ω . Assume i ∈ dom σ ∩ dom τ . Let
k ∈ ω \ (dom σ ∪ dom τ ). Let f, g ∈ ω1ω so that f extends σ, g extends
T (f, g) ∈ Ui ∩ ([σ] × [τ ]).
τ , and min{f (k), g(k)} > max{σ(i), τ (i)}. Then
Since A and B are disjoint, it is clear that i∈ω Ui ∩ (A∗ × B ∗ ) = ∅.
Note that Ui ∩ (A∗ × B ∗ ) is open dense in A∗ × B ∗ (we could modify the
density argument above to get (f, g) ∈ A∗ × B ∗ by choosing f (k) ∈ A
and g(k) ∈ B with min{g(k), f (k)} > ran σ ∪ran τ and setting f (n) = 0
for all n ∈ ω \ (dom σ ∪ {k})), so this proves A∗ × B ∗ is not Baire. 

Proof of Lemma 339

First, we show CU is closed. Let β be a limit point of CU , let σ ∈ β <ω ,
and let α = max{σ(n) : n ∈ dom(σ)}. Because β is a limit point of CU
and (α, β] is open, there is γ ∈ CU ∩ (α, β], so that σ ∈ γ <ω , γ ∈ CU ,
and σU ∈ γ <ω ⊂ β <ω . We conclude β ∈ CU , and CU is closed.

232
 Next, we show CU is unbounded by showing it has an unbounded
subset. Define f : ω1 → ω1 by f (α) = sup{max{σ(n) : n ∈ dom σU } :
σ ∈ α<ω }. By MH7550.44, C = {α : ∀β < α, f (β) < α} is closed and
0
unbounded so that C = {α ∈ C : α is a limit ordinal} is unbounded.
0 0
To show C ⊂ CU , let α ∈ C and σ ∈ α<ω . Because α is a limit ordinal,
max{σ(n) : n ∈ dom σ} = β < β + 1 < α, so that f (β + 1) < α and
σ ∈ (β + 1)<ω . This means max{σU (n) : n ∈ dom σU } 6 f (β + 1) < α,
0
implying σU ∈ α<ω and α ∈ CU . So C ⊂ CU , and CU is unbounded.
Proof of Theorem 340
Let G00 , G01 , . . . be dense open subsets of A∗ . Then there exist dense
open subsets G0 , G1 , . . . of ω1ω such that G0n = Gn ∩ A∗ . For each n < ω
and σ ∈ ω1ω , let σ Gn be an extension of σ such that [σ Gn ] ⊂ Gn and
put Cn = {δ < ω1 : ∀σ ∈T δ <ω (σ Gn ∈ δ <ω }.
ω
Fix σ ∈ ω1 . Put C = n<ω Cn and note C is club. Pick δ ∈ C ∩ A
such that σ ∈ δ <ω and let (δn )n<ω be cofinal in δ. Put τ0 = σ. If
τi ∈ δ <ω hasS been defined for some i < ω, set τi+1 = τiGi aδi .
Let f = n<ω τn . Certainly f = δ, so f ∈ A∗ . Since σ ⊂ f , f ∈ [σ].
∗

Also f extends each τnGn , so f ∈ Gn for all n < ω. Thus

\ \
f ∗ ∈ A∗ ∩ [σ] ∩ Gn = (A∗ ∩ [σ]) ∩ G0n .
n<ω n<ω 

Proof of Corollary 341

Proof of Theorem 342

Suppose X is first-countable. Let x ∈ X, and let B0 , B1 , B2 , ... be a
decreasing local base at x. Then a winning strategy for O is to play Bn
in the nth round.
Suppose X is a W -space, x ∈ X, and x ∈ A. Consider a strategy
for P that chooses a point in A in every round. If O uses her winning
strategy, the result will be a sequence of points in A which converge to
x. 
Proof of Example 343
Let X = (ω × ω) ∪ {∞} be the sequential fan. Note that each column
{n} × ω is a sequence converging to ∞. So in the nth round, P can

233
9 Proofs

always choose (n, kn ) for some kn ∈ ω. But then if f (n) = kn + 1, then

B(f ) misses all of P’s chosen points. So this is a winning strategy for
P.
Now suppose ∞ ∈ A. If A ∩ ({n} × ω) is finite for each n, then
as above we may construct a function f such that B(f ) ∩ A = ∅, a
contradiction. So A ∩ ({n} × ω) is infinite for some n, and this infinite
set converges to ∞. Thus X is Fréchet. 

Proof of Theorem 344

Let x ∈ X, and let σ be a strategy for O in GO,P (X, x).
If X is countable but not first-countable at x, then let B = {σ(s) :
s ∈ X <ω }. B is obviously countable. Since it cannot be a base, there
exists an open neighborhood U of x such that no σ(s) is a subset of U .
Then choose
pn ∈ σ(hp0 , . . . , pn−1 i) \ U
and observe that the pn are a legal counterattack defeating σ. Thus X
is not a W -space.
Likewise, if D is a countable dense subspace of regular X, but X is
not first-countable at x, then let B = {σ(s) : s ∈ D<ω }. B is obviously
countable. Since it cannot be a base, there exists an open neighborhood
U of x such that no σ(s) is a subset of U . Then choose


pn ∈ D ∩ σ(hp0 , . . . , pn−1 i) \ U

and observe that the pn are a legal counterattack defeating σ. Thus X

is not a W -space. 

Proof of Theorem 345

Suppose X is a W -space and A ⊆ X. Let a ∈ A. Let σ be a winning
strategy for player O in the game GO,P (X, a). Let ρ : A<ω → τ (A)
be defined by ρ(ha0 , ..., an−1 i) = σ(ha0 , ..., an−1 i) ∩ A. Clearly ρ is a
winning strategy for player O in the game GO,P (A, a).
Q
Suppose X = n∈ω Xn where each Xn is a W -space. Let (xn ) ∈ X.
For each n ∈ ω let σn be a winning strategy for player O in the game
GO,P (Xn , xn ). Let σ(∅) = X. Suppose i ∈ ω and player P has chosen
(pin ) ∈ σ(h(p0n ), ..., (pi−1 0 i
Q
n )i). Let σ(h(p n ), ..., (pn )i) = n n where Un =
U
σn (hp0n , ..., pi−1
n i) if n 6 i and U n = X n otherwise. Continuing the game

234
in this manner we get pin → xn for each n ∈ ω, so that (pin ) → (xn ).
Thus σ is a winning strategy for player O in the game GO,P (X, (xn )).

Proof of Theorem
P 346
Let X = α<κ α with each Xα a W -space and base point ~
X x =
(xα )α<κ . Let σα be a winning strategy for O in GO,P (Xα , xα ). We
construct a winning strategy for O in GO,P (X, ~x).
Put σ(∅) = X. If P responds with ~x(0), where supp(~x(0)) = S0 =
{α0,i : i < ω}, put σ(h~x(0)i) = πα−1
0,0
(σα0,0 (x(0)α0,0 ). If P plays ~x(n) at
the end of round n, where supp(~x(n)) = Sn = {αn,i : i < ω}, put

σ(h~x(0), . . . , ~x(n)i) =
\
{πα−1 (σα (F )) : F ∈ Fn (α), α = αj,i , 0 6 i, j 6 n},

where Fn (α) is the collection of all subsequences of hx(0)α , . . . , x(n)α i.

We claim x(n)α → xα for each α < κ. If α ∈ / Sn for all n < ω,
then x(n)α = xα . Otherwise α = αk,` . If n0 = max{k, `} and n > n0 ,
then x(n + 1)α ∈ σα (hx(n0 )α , . . . , x(n)α i). Thus x(n)α → xα , so σ is a
winning strategy. 
Proof of Theorem 347
Assume (a). Every countable subspace is first-countable by 94(a). Let
x ∈ A, for some A ⊂ X. By 94(a), A is a W -space. Let σ be a winning
strategy in GO,P (A, X) for O. Put U0 = σ(∅) and x0 ∈ U0 ∩ A. Having
defined Ui and xi for i < n, put Un = σ(x0 , . . . , xn−1 ) and xn ∈ Un ∩ A.
Finally, put B = {xn : n ∈ ω}. B is countable and x ∈ B, so X is
countably tight.
Now assume (b). Let x ∈ An for all n ∈ ω. Because X is countably
S all
n ∈ ω there exists countable Bn ⊂ An with x ∈ Bn . Be-
tight, for
cause n∈ω ∪{x} is countable, it is also first countable by assumption.
Let {Cn : n ∈ ω} be a countable base at x with Cn ⊃ Cn+1 for each
n ∈ ω. Let xn ∈ Cn ∩ Bn . For any open neighborhood N of x, we have
N ⊃ Cm ⊃ {xm , xm+1 , . . .} for some m ∈ ω, so xn → x.
Note first that X is Fréchet by setting An = A in (c).
Let Tx be the collection of open neighborhoods of x ∈ X. Assume
that for s ∈ ω <ω and t 6 s, Ut ∈ Tx . Then let
Xs = {σ(hUs1 , . . . , Us , U i) : U ∈ Tx }.

235
9 Proofs

Observe that for each neighborhood U of x, σ(hUs1 , . . . , Us , U i) ∈

Xs ∩ U , so x ∈ Xs . As X is Fréchet, let

As = {σ(hUs1 , . . . , Us , Usahni i) : n < ω} ⊆ Xs

satisfy x ∈ As .
For any f ∈ ω ω and n < ω, observe that x ∈ Af n ; therefore there is
xf n ∈ Af n where xf n → x.
Finally, pick f ∈ ω ω such that

xf n = σ(hUf 1 , . . . , Uf (n+1) i)

and observe that σ is defeated by the counterattack hUf 1 , Uf 2 , . . .i as

xf n → x.
(d) → (c): Suppose there are subsets An of X and a point x with
x ∈ An for each n ∈ ω, but there is no sequence (xn ) with xn ∈ An
for each n ∈ ω and xn → x. If Un is the latest move of player O in
GO,P (X, x), then player P may choose pn ∈ Un ∩ An . Then pn 6→ x, so
this outlines a winning strategy for player P in GO,P (X, x). Thus X is
not a w-space.
(c) → (e): Trivial.
(e) → (f ) → (g): Trivial. 

Proof of Example 348

For each countable subset C of T ∪ {∞}, it follows that C ⊆ Tα ∪ {∞}
for some α < ω1 where Tα is everything below the αth level of T .
Since T is an open first-countable subset of T ∪ {∞}, we need only
show ∞ has a countable base when ∞ ∈ C. Note {{C \ t↓ : t ∈ Tα+1 }
is countable. Each C \ t↓ is a neighborhood of ∞. We claim we have
a subbase. A subbasic open set of the entire space is C \ s↓ for s ∈ T .
If s ∈ Tα+1 we are done. Otherwise, note s  α ∈ Tα+1 and C \ s↓ =
C \ (s  α)↓ .
We now will show that T ∪ {∞} is countably tight. Consider A ⊆ T ,
and note that for each t ∈ A ∩ T , t ∈ A ∩ t↓ . If A contains an infinite
antichain B, then choose a countably infinite subset B 0 ⊆ B and note
∞ ∈ B 0 . If A contains a infinite subset B of comparable elements with
no upper bound in T , then B is countable and ∞ ∈ B.

236
 Otherwise, consider the minimal elements of A. There must be finitely
many since A lacks an infinite antichain. Then for each minimal element,
consider the branches containing that element and a distinct element
from other such branches; again there must be finitely many since A
lacks an infinite antichain. Finally, since A lacks an infinite subset of
comparable elements, there must be an element of T in each such branch
which is above every element of A in that branch. Thus ∞ 6∈ A.
Finally, let σ be a strategy for O in GO,P (T ∪ {∞}, ∞). For each
open neighborhood U of ∞, let F (U ) ∈ [T ]<ω satisfy T \ F (U )↓ ⊆ U .
Well order the limit levels L(T ) of T with ≺ such that |s| < |t| →
s ≺ t. Since levels of T are countable, it follows that hT, ≺i is order-
isomorphic to ω1 . Let
C = {t ∈ L(t) : q ∈ {s : s ≺ t}<ω → F (σ(q)) ∈ [{s : s ≺ t}]<ω }
which is closed and unbounded. Choose tn ∈ C with |tn | < |tn+1 |
converging to t ∈ C, then note that playing t  |tn | during round n is
a legal counterattack to σ which converges to t, not ∞, so σ is not a
winning strategy. 
Proof of Theorem 349 S S
For each n < ω let cn = an \ ( m6nSω \ bm ). Put c = n<ω cn and fix
n < ω. Note an ⊂∗ cn since |an ∩ ( m6n ω \ bm )| < ω, so an ⊂∗ c. If
m < n we S have cm ⊂∗ am ⊂∗ bn , and if m > n we have cm ⊂ bn . Thus
|c \ bn | 6 | m6n cm \ bn | < ω, so c ⊂∗ bn . 

Proof of Example 350

We show X satisfies property (c) in Theorem 96. Suppose x ∈ An for
each n ∈ ω. If x ∈ ω then x ∈ An for each n ∈ ω, so we may set
xn = x for each n ∈ ω. Now suppose x = ∞. For each n ∈ ω there
exists αn < ω1 such that |An ∩ aαn | = ω. Let α = supn∈ω αn . Then
|An ∩ aα | = ω for each n ∈ ω. Let x0 ∈ A0 ∩ aα . Assuming xk has been
defined, choose xk+1 ∈ Ak+1 ∩ aα \ {xj : j 6 k}. Since aα converges to
∞ (every open nbhd of ∞ contains all but finitely many points of aα ),
xn → x.
By a previous theorem, X is a W -space implies X is first-countable.
We will show X is not first-countable by contradiction. Suppose X is
first-countable, and let U0 ⊃ U1 ⊃ U2 ⊃ . . . be a decreasing open base

237
9 Proofs

at ∞. For each α ∈ ω1 , there is nα ∈ ω such that bα ⊃ Unα \ {∞}.

So, there is a k ∈ ω such that A = {α : nα = k} is uncountable, and
Uk \ {∞} ⊂ bα for each α ∈ A. Therefore, Uk \ {∞} ⊂∗ bα for each
α ∈ ω1 . Since every aα converges to ∞, aα ⊂∗ Uk \{∞} for each α ∈ ω1 .
So aα ⊂∗ Uk \ {∞} ⊂∗ bα for all α ∈ ω1 , a contradiction. We conclude
X is not first-countable, and therefore not a W -space. 

238