100%(2)Il 100% ha trovato utile questo documento (2 voti)

446 visualizzazioni180 pagineApr 22, 2017

© © All Rights Reserved

PDF, TXT o leggi online da Scribd

© All Rights Reserved

100%(2)Il 100% ha trovato utile questo documento (2 voti)

446 visualizzazioni180 pagine© All Rights Reserved

Sei sulla pagina 1di 180

Set theory is the mathematics of infinity and part of the core curriculum for

mathematics majors. This book blends theory and connections with other parts of

mathematics so that readers can understand the place of set theory within the wider

context. Beginning with the theoretical fundamentals, the author proceeds to illustrate

applications to topology, analysis and combinatorics, as well as to pure set theory.

Concepts such as Boolean algebras, trees, games, dense linear orderings, ideals, filters

and club and stationary sets are also developed.

Pitched specifically at undergraduate students, the approach is neither esoteric nor

encyclopedic. The author, an experienced instructor, includes motivating examples and

over 100 exercises designed for homework assignments, reviews and exams. It is

appropriate for undergraduates as a course textbook or for self-study. Graduate

students and researchers will also find it useful as a refresher or to solidify their

understanding of basic set theory.

Mellon University, Pennsylvania.

A Course on Set Theory

ERNEST SCHIMMERLING

Carnegie Mellon University, Pennsylvania

cambridge university press

Cambridge, New York, Melbourne, Madrid, Cape Town,

Singapore, Sao Paulo, Delhi, Tokyo, Mexico City

Cambridge University Press

The Edinburgh Building, Cambridge CB2 8RU, UK

Published in the United States of America by Cambridge University Press, New York

www.cambridge.org

Information on this title: www.cambridge.org/9781107008175

C E. Schimmerling 2011

and to the provisions of relevant collective licensing agreements,

no reproduction of any part may take place without the written

permission of Cambridge University Press.

A catalogue record for this publication is available from the British Library

ISBN 978-1-107-40048-1 Paperback

accuracy of URLs for external or third-party internet websites referred to in

this publication, and does not guarantee that any content on such websites is,

or will remain, accurate or appropriate.

Contents

Acknowledgements x

1 Preliminaries 1

2 ZFC 7

3 Order 22

3.1 Wellorderings 23

3.2 Ordinal numbers 28

3.3 Ordinal arithmetic 41

4 Cardinality 53

4.1 Cardinal numbers 53

4.2 Cardinal arithmetic 60

4.3 Conality 67

5 Trees 82

5.1 Topology fundamentals 82

5.2 The Baire space 85

5.3 Illfounded and wellfounded trees 96

5.4 Innite games 105

5.5 Ramsey theory 115

5.6 Trees of uncountable height 119

6 Dense linear orderings 125

6.1 Denitions and examples 125

6.2 Rational numbers 128

6.3 Real numbers 131

7 Filters and ideals 141

7.1 Motivation and denitions 141

7.2 Club and stationary sets 153

vi Contents

Appendix Summary of exercises on Boolean algebra 163

Selected further reading 164

Bibliography 166

Index 167

Note to the instructor

which is taught at Carnegie Mellon University every spring. It

is aimed at serious students who have taken at least one proof-

based mathematics course in any area. Most are mathematics or

computer science majors, or both, but life and physical science,

engineering, economics and philosophy students have also done

well in the course. Other students have used this book to learn

the material on their own or as a refresher. Mastering this book

and learning a bit of mathematical logic, which is not included,

would prepare the student for a rst-year graduate level set theory

course in the future. The book also contains the minimum amount

of set theory that everyone planning to go on in math should know.

I have included slightly more than the maximum amount of

material that I have covered in a fteen-week semester. But I do

not reach the maximum every time; in fact, only once. For a slower

pace or shorter academic term, one of several options would be to

skip Sections 5.6 and 7.2, which are more advanced.

There are over one hundred exercises, more than enough for

eight homework assignments, two midterm exams, a nal exam

and review problems before each exam. Exercises are located at

the ends of Chapters 1, 2, 3, 4 and 6. They are also dispersed

throughout Chapters 5 and 7. This slight lack of uniformity is

tied to the presentation and ultimately makes sense.

In roughly the rst half of the book, through Chapter 4, I de-

velop ordinal and cardinal arithmetic starting from the axioms of

ZermeloFraenkel Set Theory with the Axiom of Choice (ZFC). In

other words, this is not a book on what some call naive set theory.

There is one minor way in which the presentation is not entirely

viii Note to the instructor

rigorous. Namely, in listing the axioms of ZFC, I use the imprecise

word property instead of the formal expression rst-order formula

because mathematical logic is not a prerequisite for the course.

Some other textbooks develop the theory of cardinality for as

long as possible without using the Axiom of Choice (AC). I do

not take this approach because it adds technicalities, which are

not used later in the course, and gives students the misleading

impression that AC is controversial. By assuming AC from the

start, I am able to streamline the theory of cardinality. I may note

how AC has been used in a proof but I do not belabor the point.

Once, when an alternate proof without AC exists, it is outlined in

an exercise.

The second half of the book is designed to give students a sense

of the place of set theory within mathematics. Where I draw con-

nections to other elds, I include all the necessary background

material. Some of the other areas that come up in Chapter 5 are

topology, metric spaces, trees, games and Ramsey theory. The

real numbers are constructed using Dedekind cuts in Chapter 6.

Chapter 7 introduces the student to lters and ideals, and takes up

the combinatorics of uncountable sets. There is no section specif-

ically on Boolean algebra but it is one of the recurring themes in

the exercises throughout the book. For the readers convenience,

I have briey summarized the results on Boolean algebra in the

Appendix. All of this material is self-contained.

As I mentioned, before starting this book, students should have

at least one semesters worth of experience reading and writing

proofs in any area of mathematics; it does not matter which area.

They should be comfortable with sets, relations and functions,

having seen and used them at a basic level earlier. They should

know the dierence between integers, rational numbers and real

numbers, even if they have not seen them explicitly constructed.

And they should have experience with recursive denitions along

the integers and proofs by induction on the integers. These no-

tions come up again here but in more sophisticated ways than in

a rst theoretical mathematics course. There are no other pre-

requisites. However, because of the emphasis on connections to

other elds, students who have taken courses on logic, analysis,

algebra, or discrete mathematics will enjoy seeing how set theory

and these other subjects t together. The unifying perspective of

Note to the instructor ix

set theory will give students signicant advantages in their future

mathematics courses.

Acknowledgements

Herbert Enderton and Set theory: an introduction to independence

proofs by Kenneth Kunen. When I started teaching undergraduate

set theory, I recommended Introduction to set theory by Karel

Hrbacek and Thomas Jech to my students. The reader who knows

these other textbooks will be aware of their positive inuence.

This book began as a series of handouts for undergraduate stu-

dents at Carnegie Mellon University. Over the years, they found

typographical errors and indicated what needed more explana-

tion, for which I am grateful. I also thank Michael Klipper for

proofreading a draft of the book in Spring 2008, when he was a

graduate student in the CMU Doctor of Philosophy program.

During the writing of this book, I was partially supported by

National Science Foundation Grant DMS-0700047.

1

Preliminaries

tools of mathematics. After millennia of doing mathematics, math-

ematicians started trying to write down the rules of the game.

Since mathematics had already fanned out into many subareas,

each with its own terminology and concerns, the rst task was

to nd a reasonable common language. It turns out that every-

thing mathematicians do can be reduced to statements about sets,

equality and membership. These three concepts are so fundamen-

tal that we cannot dene them; we can only describe them. About

equality alone, there is little to say other than two things are

equal if and only if they are the same thing. Describing sets and

membership has been trickier. After several decades and some

false starts, mathematicians came up with a system of laws that

reected their intuition about sets, equality and membership, at

least the intuition that they had built up so far. Most importantly,

all of the theorems of mathematics that were known at the time

could be derived from just these laws. In this context, it is com-

mon to refer to laws as axioms, and to this particular system as

ZermeloFraenkel Set Theory with the Axiom of Choice, or ZFC.

In the rst unit of the course, through Chapter 4, we examine this

system and get some practice using it to build up the theory of

innite numbers.

In another sense, set theory is a part of mathematics like any

other, rich in ideas, techniques and connections to other areas.

This perspective is emphasized more than the foundational aspects

of set theory throughout the course but especially in the second

half, Chapters 57. There, our choice of topics within set theory is

2 Preliminaries

designed to give the reader an impression of the depth and breadth

of the subject and where it ts within the whole of mathematics.

To get started, we review some basic notation and terminology.

We expect that the reader is familiar with the following notions

but perhaps has not seen them expressed in exactly the same way.

Ordered pairs are used everywhere in mathematics, for example,

to refer to points on the plane in geometry. The precise meaning

of (x, y) is left to the imagination in most other courses but we

need to be more specic.

Denition 1.1 (x, y) = {{x}, {x, y}} is the ordered pair with

rst coordinate x and second coordinate y.

It is convenient that (x, y) is dened in terms of sets. After all,

this is set theory, so everything should be a set! The main point

of the denition is that from looking at {{x}, {x, y}} we can tell

which is the rst coordinate and which is the second coordinate.

Namely, if {{x}, {x, y}} has exactly two elements, then the rst

coordinate is

x = the unique z such that {z} {{x}, {x, y}}

and the second coordinate is

y = the unique z = x such that {x, z} {{x}, {x, y}}.

And, if {{x}, {x, y}} has just one element, which can only happen

if x = y, then the rst and second coordinates are both

x = the unique z such that {z} {{x}}.

To understand this formula, keep in mind that

{x, y} = {y, x}

and

{x, x} = {x}.

In particular,

{{x}, {x, x}} = {{x}, {x}} = {{x}}

and {x} is the only element of {{x}}.

Denition 1.2 A B = {(x, y) | x A and y B} is the

Cartesian product of A and B.

Preliminaries 3

Denition 1.3 R is a relation from A to B i R is a subset of

A B, that is

R A B.

Sometimes, if we know that R is a relation, then we write xRy

instead of (x, y) R. For example, we write

2<

not

( 2, ) <

because the latter is confusing.

Denition 1.4 Let R be a relation from A to B and S A.

1. The domain of R is

dom(R) = {x A | there exists y such that xRy}.

R[S] = {y B | there exists x S such that xRy}.

3. The range of R is

ran(R) = {y B | there exists x such that xRy}.

Denition 1.5 f is a function from A to B i f is a relation

from A to B and, for every x A, there exists a unique y such

that (x, y) f .

If we happen to know that f is a function, then we write

f (x) = y

instead of (x, y) f . When we write f : A B, it is implicit that

f is a function from A to B. In certain situations, we refer to a

function f by writing x f (x) or f (x) | x A. There are times

when we write fx instead of f (x); this is when we are thinking of

elements x of A as indices and fx | x A as an indexed family.

If the domain of f consists of ordered pairs, then it is common

to write f (x, x ) instead of f ((x, x )). Functions are also called

4 Preliminaries

operations and maps. Some people distinguish between a function

f : A B and its graph,

graph(f ) = {(x, f (x)) | x A},

but we do not. To us they are the same, that is, f = graph(f ), as

we see from Denition 1.5.

Denition 1.6 If f : A B is a function and S A, then the

restriction of f to S is

f S = {(x, f (x)) | x S}.

2. f is a surjection i for every y B, there exists x A such

that f (x) = y.

3. f is a bijection i f is both an injection and a surjection.

Injections are also called one-to-one functions. Surjections from

A to B are also called functions from A onto B. Bijections are

also called one-to-one correspondences.

Denition 1.8 If f is an injection from A to B, then we write

f 1 for the unique injection g : f [A] A with the property that

g(f (x)) = x for every x A. In other words,

f 1 = {(f (x), x) | x A}.

Finally, we assume that the reader has good intuition about the

set of integers,

Z = {. . . , 2, 1, 0, 1, 2, . . . },

the set of rational numbers,

Q = {m/n | m, n Z and n = 0}

and the set of real numbers, R. One thing we will do in this course

is dene all these kinds of numbers, starting from the natural

numbers 0, 1, 2, 3, 4, etc. Each natural number will be the set of

natural numbers that precedes it. Thus 0 = , where is the set

with no members. After that, 1 = {0}, 2 = {0, 1}, 3 = {0, 1, 2},

Preliminaries 5

4 = {0, 1, 2, 3}, etc. This happens to be very convenient because

then

m < n m n.

In other words, the usual ordering on the natural numbers coin-

cides with membership.

We use natural numbers to denote cardinality, for example,

when we say, Lance Armstrong won the Tour de France seven

times. And we use natural numbers to denote order, for example,

when we say, the attorney general is seventh in the presidential

line of succession. Another thing we will do in this course is ex-

tend the notions of cardinality and order into the innite. Finite

cardinal and ordinal numbers are basically the same thing; one

could say that the dierence between seven and seventh is

just grammatical. However, the dierence between innite cardi-

nal and ordinal numbers is more profound, as we will explain in

Chapters 3 and 4.

Exercises

Exercise 1.1 If R is a relation, then we dene

R1 = {(y, x) | xRy}.

Give an example where R is a function but R1 is not.

Exercise 1.2 How many functions whose domain is the empty

set are there? In other words, given a set B, how many functions

f : B are there?

Exercise 1.3 Explain why (x, y, z) = (x, (y, z)) is a reasonable

denition of an ordered triple.

this book. We assume that the reader has studied them before

but this exercise reviews all the necessary denitions and facts.

Let A be a set and R be a relation on A, that is, R A A.

Then:

R is a reexive relation on A i for every x A, xRx.

R is a symmetric relation on A i for all x, y A, if xRy, then

yRx.

6 Preliminaries

R is a transitive relation on A i for all x, y, z A, if xRy and

yRz, then xRz.1

R is an equivalence relation on A i R is a reexive, symmetric

and transitive relation on A.

Assuming that R is an equivalence relation on A, for every x A,

we dene the equivalence class of x to be

[x]R = {y A | xRy}.

It is also standard to write

A/R = {[x]R | x A}.

A partition of A is a family F of non-empty subsets of A such that

A is the union of F, that is,

A= F = {x | there exists X F such that x X}

and

the elements of F are pairwise disjoint, that is, for all X, Y F,

if X = Y , then X Y = .

Now here are the exercises:

1. Let R be an equivalence relation on A. Prove that A/R is a

partition of A.

2. Let F be a partition of A. Prove that there exists a unique

equivalence relation R such that F = A/R.

1 Later in the book we will dene transitive set, which is dierent from transitive

relation. Unfortunately, it will be important to pay attention to this subtle

dierence in terminology.

2

ZFC

ory, we have in mind a collection of axioms in a certain language.

The language of set theory has two symbols, = and , although

sometimes we add symbols that are dened in terms of these two

to make things easier to read. For example, we write A B when

we mean that, for every x, if x A, then x B.

ZermeloFraenkel Set Theory with the Axiom of Choice, or ZFC

for short, is a certain theory in the language of set theory that we

will describe in this chapter. There are innitely many axioms of

ZFC, each of which says something rather intuitive about sets,

equality and membership. In our list below, some axioms of ZFC

are presented individually whereas others are presented as schemes

for generating innitely many axioms. One last comment about

terminology before we begin: throughout the course,

and

member = element.

x belongs to A,

x is an element of A and

x is a member of A,

8 ZFC

Empty Set Axiom

This axiom says that there is a unique set without members. For-

mally, it is written

!A x (x A) .

In plain English, this says:

There exists a unique A such that, for every x,

x is not an element of A.

The unique set without elements is written .

Extensionality Axiom

This axiom says that two sets are equal if they have the same

members. Formally, it is written

A B [ x (x A x B) = A = B ] .

Because we dened

A B x (x A = x B) ,

another way to write the Extensionality Axiom is

A B [ (A B and B A) = A = B ] .

In other words, two sets are equal if each is a subset of the other.

By logic alone, if A = B, then A and B have the same members.

Combining this fact with the Extensionality Axiom, we have that

A B [ x (x A x B) A = B ] .

Equivalently,

A B [ (A B and B A) A = B ] .

Pairing Axiom

This axiom allows us to form singletons and unordered pairs. Its

formal statement is

x y !A z [z A (z = x or z = y)] .

If x = y, then we write {x, y} for the unique set whose only

members are x and y and call it an unordered pair. We always

ZFC 9

write {x} instead of {x, x} and call it a singleton. At this point,

it makes sense to dene the rst three natural numbers 0 = ,

1 = {0} and 2 = {0, 1}. We can also justify dening ordered pairs

by setting

(x, y) = {{x} , {x, y}}

whenever we are given x and y as we did in Denition 1.1. As a

reminder, when x = y, what we really have is

(x, x) = {{x}} .

Notice that, based on this denition, when we write (x, y), we can

tell that x is the rst coordinate and y is the second coordinate.

Formally, this means we can prove that for all x, y, x and y ,

(x, y) = (x , y ) (x = x and y = y ).

Union Axiom

This axiom allows us to form unions. Its formal statement is

F !A x [x A Y F (x Y )] .

We write F for the unique set whose members are exactly the

members of the members of F. In other words,

F = {x | there exists Y F such that x Y }.

It is important to note that, in the Union Axiom, the family F is

allowed to be innite. We often use dierent notation when F is

nite. For example, we dene

A B = {A, B}

and

ABC = {A, B, C}.

At this point, we can dene the remaining natural numbers

3 = 2 {2} = {0, 1, 2},

5 = 4 {4} = {0, 1, 2, 3, 4}

10 ZFC

and, in general,

n + 1 = n {n} = {0, . . . , n}.

This axiom allows us to form the set of all subsets of a given set.

Its formal statement is

A !F X (X F X A).

We write P(A) for the unique set of subsets of A. In other words,

P(A) = {X | X A}.

We call P(A) the power set of A. As an example, let us see what

happens when we start with the empty set and take power sets

over and over. Dene

V0 = ,

V1 = P(V0 ) = {},

V2 = P(V1 ) = {, {}},

and, in general,

Vn+1 = P(Vn ).

The sets Vn will come up again later.

Comprehension Scheme

This axiom scheme gives us a way to form specic subsets of a

given set. It says the following.

For each property P (x), the following is an axiom:

A !B x [x B (x A and P (x))] .

Notice that the word property appears in quotes. There are

innitely many properties, which is why ZFC has innitely many

axioms. We will not give a formal denition of property because

it involves rst-order logic, which is not a prerequisite. It is enough

ZFC 11

for students in this course to depend on their intuition about the

meaning of property. Given a property P (x), we write

{x A | P (x)}

for the set of elements x of A for which P (x) is true. For example,

{x 10 | x is even} = {0, 2, 4, 6, 8}

and

{x 10 | x is odd} = {1, 3, 5, 7, 9}.

It is important to note that the Comprehension Scheme does not,

in general, permit us to dene sets by writing {x | P (x)}. In fact,

for some P (x), what we think of as {x | P (x)} is not a set. For

example {x | x = x} is not a set by the result in Exercise 2.6.

At this point, we are justied in making many familiar deni-

tions. For example, Denition 1.2 says that, given sets A and B,

we have the Cartesian product

A B = {(x, y) | x A and y B} .

In order to see that this is a legitimate denition, note that if

x A and y B, then

(x, y) = {{x} , {x, y}} P (P (A B)) .

Thus A B =

{z P (P (A B)) | z = (x, y) for some x A and y B} ,

which means the denition of A B is justied by a combination

of the axioms we have listed so far. See Exercise 2.4.

Innity Axiom

In an indirect way, this axiom tells us that the set of natural

numbers exists. Its formal statement is

I [ I and x (x I = x {x} I) ].

We say that a set I is inductive if I is a witness to the Innity

Axiom. In other words,

I is inductive [ I and x (x I = x {x} I) ].

12 ZFC

Proposition 2.1 There is a unique inductive set I such that

I J for every inductive set J.

The proof of Proposition 2.1 is broken up into smaller steps in

Exercise 2.5. The set of natural numbers is dened to be the unique

inductive set that is a subset of every other inductive set. We write

(lower case Greek omega) for the set of natural numbers. In

other words,

= {0, 1, 2, 3, . . . }.

It is also common for mathematicians to write N instead of

although we will not. Not only does the Innity Axiom allow us

to dene , it implies that we can prove statements by induction

on n and make recursive denitions for n just as you have

done in your other mathematics courses.

To see the relationship with induction, suppose we wish to prove

that a given property P (n) holds for every natural number n. By

Proposition 2.1 and the denition of , it would be enough to

show that {n | P (n)} is an inductive set. In other words,

show that P (0) holds and if P (n) holds, then so does P (n + 1).

Replacement Scheme

This is a scheme for generating innitely more axioms.

For each property P (x, y), the following is an axiom:

The same comments we made about meaning of property when

we discussed the Comprehension Scheme apply here too.

Because we will not emphasize how the Replacement Scheme is

used later in the book, let us give a concrete example and some

intuition here. Suppose we want to dene

V = {Vn | n } .

That is, we want to let V be the union of the innite family

{Vn | n }.

This family is supposed to be the range of the innite sequence

Vn | n .

ZFC 13

But why does this innite sequence exist? In other words, why is

it a set? Given a particular natural number, say 5, we can prove

from the other axioms that the nite sequence

Vm | m < 5 = V0 , V1 , V2 , V3 , V4

exists. The Replacement Scheme can be used to make the leap

from innitely many nite sequences to one innite sequence. To

see what we mean, let P (x, y) be the following property: x is a

natural number and there exists a function f with domain

dom(f ) = x + 1 = {0, . . . , x}

such that

f (0) = ,

for every n < x, f (n + 1) is the power set of f (n), and

f (x) = y.

Then P (n, Vn ) holds for every n . By the Replacement Scheme,

there is a set B such that, for every n , Vn B. Now use the

Comprehension Scheme to dene

Vn | n < = {(x, y) B | P (x, y)}.

Finally, use the Union Axiom to dene V as we originally wanted

to do.

A slogan that captures the intuition behind the Replacement

Scheme is: If an assignment looks like a function and its domain

is a set, then its range is also a set, so it really is a function. In

the example above, we used the Replacement Scheme to see that

the assignment n Vn really is a function with domain and

range {Vn | n }.

Foundation Axiom

This axiom says that if S is a non-empty set, then there exists

x S such that, for every y S, y x. In symbols,

S (S = = x S y S (y x)).

For example, should S = 2, the only witness to the Foundation

Axiom would be x = 0 because 2 = {0, 1} and 0 {0} = 1 but

1 = 0. On the other hand, when S = {0, {1}}, both elements

14 ZFC

of S satisfy the requirement of the Foundation Axiom because

0 {1} and {1} 0.

The Foundation Axiom turns out to be equivalent to the state-

ment that there is no sequence xn | n such that

xn+1 xn x1 x0 .

In particular, it implies that no set is an element of itself. For

otherwise, if x x, then

x x x x,

which contradicts the Foundation Axiom. We will make additional

comments about this axiom later but it is not a focus of the book.

Axiom of Choice

This axiom says that every family of sets has a choice function.

F function c A F (A = = c(A) A) .

The function c is called a choice function for F because it chooses

an element c(A) out of every non-empty A that belongs to F. From

experience, given nitely many non-empty sets, you can pick one

element from each. The Axiom of Choice says this is possible even

if you start with innitely many non-empty sets.

The rst time the Axiom of Choice is used is in Chapter 4.

There, it is essential for being able to assign a numerical cardinal-

ity (size) to each set, which is one of the most important things

we do in this book. As we go along, we will point out where and

how the Axiom of Choice is used.

Exercises

In the following exercises, you only need to be attentive to the

axioms of ZFC when so instructed, namely, in Exercises 2.1, 2.4,

2.6, the second part of 2.8 and the rst part of 2.11. Otherwise,

use the same style of mathematical argumentation as in your other

proof-based mathematics courses without reference to ZFC.

ZFC 15

Exercise 2.1 Prove that the following theories are equivalent.

1. Empty Set Axiom + Extensionality Axiom.

2. A x (x A) + Extensionality Axiom.

Hint: Obviously 1 implies 2. In the other direction, uniqueness is

the issue.

Exercise 2.2 Recall that 0 = and n + 1 = {0, . . . , n} for every

n . Recall also that V0 = and Vn+1 = P(Vn ) for every n .

1. Prove by induction that if n , then P(n) has 2n elements.

2. List the elements of V4 .

3. Make a conjecture regarding the size of Vn . Then prove your

conjecture by induction on n .

Exercise 2.3 Dene z to be a transitive set i for all x and y,

if x y and y z, then x z. This is equivalent to saying that,

for every y, if y z, then y z.

1. Prove that Vn is a transitive set for every n .

2. Prove that Vn Vn+1 for every n .

3. Prove that Vn = n for every n .

Exercise 2.4 Give a detailed explanation of how the denitions

of A B, dom(R) and R[S] given in Denitions 1.2 and 1.4 are

justied by the axioms of ZFC.

Exercise 2.5 Prove Proposition 2.1 by showing the following.

1. If I and J are inductive sets, then I J is an inductive set.

2. Suppose that K is an inductive set. Let

F = {J P(K) | J is an inductive set}

and

I = {x K | J F ( x J )} .

Prove the following statements.

(a) I is an inductive set.

(b) For every J, if J is an inductive set, then I J.

(c) Suppose that I is an inductive set and, for every inductive

set J,

I J.

Then I = I.

16 ZFC

Exercise 2.6 Prove that there is no set V such that x V for

every set x. (This shows that {x | x = x} is not a set.) If you

use the Foundation Axiom in your proof, then nd a second proof

that does not use the Foundation Axiom.

Exercise 2.7 In general, dene the intersection of a non-empty

family G to be

G = {a | X G ( a X )}.

S F = {S X | X F}

and

S F= {S X | X F}.

A

B = {f | f is a function from A to B}.

Consider a function of the form (a, b) S(a,b) with domain AB.

In other words, consider an indexed family

S(a,b) | (a, b) A B.

1. Prove that

S(a,b) = S(a,f (a)) .

aA bB f A B aA

S(a,b) S(a,b ) =

whenever a A and b, b B but b = b . Now prove the same

equation as in part 1 but without using the Axiom of Choice.

Exercise 2.9 By denition, the symmetric dierence of A and

B is

A B = (A B) (B A).

Verify the distributive law

A (B C) = (A B) (A C).

Remark: This is one step in showing that if S is a set, then P(S)

ZFC 17

is a ring if addition is taken to be symmetric dierence and mul-

tiplication is taken to be intersection. We only mention this for

readers who happen to know some abstract algebra.

formula

(x, y) E x y is nite.

is,

P () /E = {[x]E | x P()}.

B0 = A

and

Bn+1 = Bn .

Let

C= {Bn | n < }.

2. Prove that C is a transitive set. (See Exercise 2.3 for the de-

nition of transitive set.)

3. Prove that if D is a transitive set and A D, then C D.

18 ZFC

Beginning exercises on Boolean algebra

To understand Exercises 2.12 and 2.13 below, and many exercises

later in the book,1 you must rst read the following denitions

and example.

A Boolean algebra is a 6-tuple of the form

B = (B, , , , , )

where B is a set, and are binary operations on B, is a unary

operation on B, and are distinct elements of B, and for all

X, Y, Z B, the following ten laws hold.

Associativity

X (Y Z) = (X Y ) Z

X (Y Z) = (X Y ) Z

Commutativity

X Y =Y X

X Y =Y X

Distributivity

X (Y Z) = (X Y ) (X Z)

X (Y Z) = (X Y ) (X Z)

Identity

X =X

X =X

Complementation

X X =

X X =

Each Boolean algebra, B = (B, , , , , ), has an associated

Boolean algebra relation, , which is dened by

X Y X = X Y

for all X, Y B. Here is some special terminology for Boolean

algebras that will be used in the exercises. If A B, then we say

1 The Appendix lists the exercises to which we are referring.

ZFC 19

that A is an atom i A = and, for every X B, if X A, then

X = or X = A. We say that B is nite i B is nite.

Given Boolean algebras

B = (B, B , B , B , B , B )

and

C = (C, C , C , C , C , C ),

an isomorphism from B to C is dened to be a function f such

that

f is a bijection from B to C and

for all X, Y B,

f (X B Y ) = f (X) C f (Y ),

f (X B Y ) = f (X) C f (Y ),

f (B X) = C f (X),

f (B ) = C

and

f (B ) = C .

Example You should work out the details of the following asser-

tions before attempting the exercises that follow. Let S = and

put

B(S) = (P(S), , , , , S)

where X means S X in this context. Then B(S) is a Boolean

algebra. It is called the Boolean algebra of subsets of S. If S is

nite and has n elements, then P(S) has 2n elements hence B(S)

is nite. For B(S), the Boolean algebra relation amounts to

X Y X = X Y X Y.

The atoms of B(S) are exactly the singletons {a} for a S. The

function X S X is an isomorphism from B(S) to the Boolean

algebra

(P(S), , , , S, ).

Notice that union and intersection are exchanged, as are and S.

20 ZFC

Exercise 2.12 Consider an arbitrary nite Boolean algebra

B = (B, , , , , ).

Let S be the set of atoms of B.

1. Prove that if X B and X = , then there exists A S such

that A X.

2. Let X B. Suppose that X = and

{A S | A X} = {A1 , . . . , Am }.

Let

Y = A1 Am .

Prove

X = Y.

Hint: Certain basic facts about sets generalize to Boolean al-

gebras. For example, for sets we know that

if X Y and Y X, then X = Y ,

and for Boolean algebras we have that

if X Y and Y X, then X = Y .

The reason is that if X Y and Y X, then

X =X Y =Y X =Y

by the commutativity law for and the denition of . The

moral is that you should base your intuition about Boolean

algebras on what you already know about Boolean algebras of

sets. Of course, ultimately, you need to prove your intuition

is correct using just the laws of Boolean algebras. It should

also be said that the solution to this exercise is relatively long

so organizing your answer into a well-chosen series of lemmas

would be very helpful.

3. Let f : B P(S) be dened by

f (X) = {A S | A X}.

Prove that f is an isomorphism from B to B(S). Remark: This

explains why intuition coming from Boolean algebras of sets

really is valuable intuition about nite Boolean algebras.

ZFC 21

Exercise 2.13 As in Exercise 2.10, let E be the equivalence

relation on P() dened by

x E y x y is nite.

1. Prove that the following table of equations determines a Boolean

algebra B = (B, , , , , ).

B = P()/E

[x]E [y]E = [x y]E

[x]E [y]E = [x y]E

[x]E = [ x]E

= []E

= []E

Before proving the laws of Boolean algebras, you must show

that the operations , and are well-dened by the equations

listed above. So part of what you must show is that if x E x

and y E y , then

(x y) E (x y ),

(x y) E (x y )

and

( x) E ( x ).

Remark: This is an example of a quotient Boolean algebra. In

the literature, it is referred to as P()/Finite.

2. Prove that the Boolean algebra B that was dened in part 1

has no atoms.

3

Order

that illustrated the dierence between ordinal and cardinal num-

bers. Let us expand on our example of ordinal numbers, which

involved the presidential line of succession:

2nd Speaker of the House

3rd President pro tempore of the Senate

4th Secretary of State

5th Secretary of the Treasury

6th Secretary of Defense

7th Attorney General

8th Secretary of the Interior

9th Secretary of Agriculture

10th Secretary of Commerce

11th Secretary of Labor

12th Secretary of Health and Human Services

13th Secretary of Housing and Urban Development

14th Secretary of Transportation

15th Secretary of Energy

3.1 Wellorderings 23

For fun, here are a few more:

17th Secretary of Veterans Aairs

18th Secretary of Homeland Security

1st, 2nd, 3rd, etc. However, sometimes it makes sense to count 0th,

1st, 2nd, 3rd, etc. For example, at the top of the presidential line

of succession, we really have:

0th President

the sequence of length 6,

x0 , x1 , x2 , x3 , x4 , x5 = 11, 6, 18, 9, 72, 31

the 0th number is x0 = 11, the 1st number is x1 = 6, etc., and

the 5th and nal number is x5 = 31. Keep in mind that, in plain

English, it would be very strange to say that the fth item is the

last in a list of six items!

In set theory, we also continue counting ordinal numbers past

all the nite ordinal numbers. For example, is the rst innite

ordinal number, followed by + 1, + 2, etc. It takes a bit of

theoretical work to make concrete sense of this idea, so this is

what we do rst.

3.1 Wellorderings

Denition 3.1 Let A be a set and be a relation on A.

1. (A, ) is transitive i for all x, y, z A, if x y and y z,

then x z.

2. (A, ) is irreexive i for every x A, x x.

3. (A, ) is total i for all x, y A, either x y or x = y or

y x.

24 Order

4. (A, ) is a strict linear ordering i it is transitive, irreexive

and total.

that

x y (x y or x = y)

for all x, y A. Then (A, ) is a linear ordering i (A, ) is a

strict linear ordering.

The denition tells us how to pass from a strict linear ordering

to its associated linear ordering. In the other direction, we have

the fact that if (A, ) is a linear ordering and

x y (x y and x = y)

for all x, y A, then (A, ) is a strict linear ordering and (A, )

is the linear ordering associated to (A, ). Occasionally, we might

drop the word strict when it is clear from the choice of symbols

or context which we mean.

Here is another remark on notation. If we are given that (A, )

is a strict linear ordering, then we might write without bothering

to explain that it is the linear ordering associated to (A, ) even

though ocially we should explain. Suppose, instead, that we are

told that (A, R) is a strict linear ordering. It is unlikely that we

would write R for the associated linear ordering because it looks

so strange. If, for some reason, we wrote R, then we could not

assume that the reader knows what we mean, so we would have to

explain. Ocially, and are two completely dierent symbols

even though looks like a combination of and =.

Denition 3.3 Let (A, ) be a strict linear ordering, S A

and x S. Then x is the -least element of S i for every y S,

x y.

ordering and, for every non-empty S A, S has a -least element.

The extra property that turns a strict linear ordering into a

wellordering is called wellfoundedness. Every nite linear ordering

is a wellordering. Also, the usual linear ordering of the natural

3.1 Wellorderings 25

numbers is a wellordering. On the other hand, the usual linear

ordering on the set of integers,

Z = { 2, 1, 0, 1, 2, . . . },

is obviously not a wellordering because there is no least integer.

However, if we dene a brand new relation on Z by

0 1 1 2 2 3 3 ,

then (Z, ) is a wellordering.

Lemma 3.5 Let (A, ) be a strict linear ordering. Then (A, )

is a wellordering i there is no sequence xn | n such that,

for every n , xn+1 xn .

Proof First suppose that (A, ) is a wellordering. Consider an

arbitrary sequence xn | n . Let S = {xn | n }. Let y be

the -least element of S. Say y = xn . Then xn xn+1 . Hence

xn+1 xn .

Second, suppose that (A, ) is not a wellordering. Therefore,

there exists S A such that S = and S does not have a -

least element. Dene a sequence xn by recursion on n as

follows. For the base case, let x0 S if possible. Otherwise, leave

x0 undened. For the successor case, let xn+1 S with xn+1 xn

if possible. Otherwise, leave xn+1 undened.

We claim that, for every n , xn is dened, xn S and,

if n 1, then xn xn1 . We prove this claim by induction on

n . The base case is n = 0. Since S = , it is possible to

let x0 S. So that is what we did. The successor case has the

induction hypothesis that xn is dened and xn S. Since S does

not have a -least element, it is possible to let xn+1 S with

xn+1 xn . So that is what we did. Note that x(n+1)1 = xn .

In the second part of the previous proof, we made a denition by

recursion on n and proved a statement by induction on n .

Our ability to do this is tied to the fact that the usual ordering of

is a wellordering. The following fundamental theorems put the

phenomena of induction and recursion in the general context of

wellorderings.

Theorem 3.6 (Proofs by induction) Let (A, ) be a wellorder-

ing. Let P (x) be a statement about a variable x. Suppose that, for

26 Order

every y A,

(x y P (x) holds) = P (y) holds.

Then, for every y A, P (y) holds.

Proof Let S = {x A | P (x) does not hold}. For contradiction,

suppose that S = . Let y be the -least element of S. Then,

y S and, for every x y, x S. Hence, for every x y, P (x)

holds. From the hypothesis of the theorem, P (y) holds. That is,

y S. This is a contradiction.

exists A A such that f : A B.

ing. Let

F :AP B

be a function where B is a set and P is the set of partial functions

from A to B. Then there is a unique function G : A B such

that

G(y) = F (y, G {x | x y})

for every y A.

The point is that the equation in Theorem 3.8 determines the

function G. We describe this as a recursive denition of G because

the equation is a recipe for nding G(y) based on the earlier values

G(x) for x y. Specically, the function F outputs G(y) when

we input y and the restriction of G to {x | x y}.

The set P in the Theorem 3.8 might seem a bit mysterious at

rst. Not all partial functions are relevant to the statement and

proof of the theorem. Really, we only use that, for every y A,

G {x | x y} P.

But there would be no advantage in shrinking down P to just the

partial functions we need.

Proof of Theorem 3.8 Let us say that g is a z-approximation i

z A,

g is a partial function from A to B,

3.1 Wellorderings 27

dom(g) = {y | y z} and

for every y z,

g(y) = F (y, g {x | x y}) .

The reason we call it a z-approximation is that g approximates G

up to and including z.

Claim 3.8.1 For every z A, there exists at most one z-

approximation.

Proof of claim Fix z A. Let g and h be z-approximations. We

prove that g(y) = h(y) by induction on y z using Theorem 3.6.

Let y z. Our induction hypothesis is that, for every x y,

g(x) = h(x).

In other words,

g {x | x y} = h {x | x y}.

Therefore,

g(y) = F (y, g {x | x y}) = F (y, h {x | x y}) = h(y),

which completes the proof of the claim.

Claim 3.8.2 For every z A, there exists a z-approximation.

Proof of claim We argue by induction using Theorem 3.6. Let

z A. Our induction hypothesis is that, for every y z, there

exists a y-approximation; call it gy . By Claim 3.8.1, gy is the only

y-approximation. Observe that if x y z, then

gy {w | w x}

is an x-approximation. Since there is only one x-approximation,

gx = gy {w | w x}

whenever x y z. Let

h= gy .

yz

dom(h) = {x | x y} = {x | x z}

yz

28 Order

and, for every y z,

gy = h {x | x y}.

Now let

gz = h {(z, F (z, h)}.

In other words,

h(y) if y z

gz (y) =

F (z, h) if y = z.

By the two claims, for each z A, there is a unique z-approximation,

which we call gz . As we already calculated, if y z, then

gy = gz {x | x y}.

Dene

G= gz .

zA

dom(G) = {y | y z} = A

zA

G(z) = gz (z) = F (z, gz {y | y z}) = F (z, G {y | y z}) .

Thus G witnesses the conclusion of Theorem 3.8. To see that G is

the unique witness to the theorem, argue just like in the proof of

Claim 3.8.1.

We start by repeating a denition from Exercise 2.3.

and y A, then x A.

3.2 Ordinal numbers 29

Equivalently, A is a transitive set i for every y A, y A.

The importance of transitivity will not really begin to be apparent

until Denition 3.15. First, here are three basic facts about how

certain operations preserve transitivity.

Lemma 3.10 Let A be a transitive set. Then A {A} is also a

transitive set.

Proof Let y A{A} and x y. We must show that x A{A}.

In fact, we will show that x A. Either y A or y {A}. If

y A, then x A since A is transitive. If y {A}, then y = A,

so x A.

Lemma 3.11 If A is a transitive set, then so is P(A).

Proof Assume that A is transitive, z P(A) and y z. We

must show that y P(A). By our assumption, z A and y z, so

y A. Since A is transitive, y A. Thus, y P(A) as desired.

Lemma 3.12 If F is a family of transitive sets, then F is a

transitive set too.

Proof Let y F. Pick A F such that y A. Since A is

transitive, y A. Hence y F as desired.

As motivation for the next denition, observe that membership

is not a relation. This is because {(x, y) | x y} is not a set. For

if it were a set, then it would be a relation and its domain would

also be a set. But its domain would be {x | x = x}, which is not

a set by Exercise 2.6. On the other hand, if A is a set, then so is

{(x, y) A A | x y}

by the fact that A A is a set and the Comprehension Scheme.

Denition 3.13 When we write (A, ) we really mean

(A, {(x, y) A A | x y}) .

Similarly, when we say that (A, ) is transitive, we really mean

that the restriction of to A is a transitive relation on A.

Do not get confused between A being transitive and (A, ) being

transitive. The meaning is dierent. For example, if A = {{x}},

then is a transitive relation when restricted to A because A only

has one element, but A is not a transitive set because {x} A

and {x} A. The two notions of transitivity are related though.

30 Order

Lemma 3.14 Suppose that A is a transitive set. Then (A, ) is

transitive i for every z A, z is a transitive set.

Proof Given that A is a transitive set, the following six state-

ments are equivalent.

1. (A, ) is transitive.

2. x, y, z A ((x y and y z) = x z).

3. z A y z A x y A (x z).

4. z A y z x y A (x z).

5. z A y z x y (x z).

6. z A (z is a transitive set).

The rst equivalence (1 2) is by denition. The second

(2 3) is just logic. The third (3 4) is because if z A,

then z A since A is a transitive set, so z A = z. The fourth

(4 5) is similar: if z A and y z, then y A (since A is a

transitive set), hence y A (again since A is a transitive set), so

y A = y. The fth (5 6) is by denition.

Now we come to one of the most important denitions of the

course.

Denition 3.15 A set is an ordinal i is a transitive set

and (, ) is a wellordering.

By the Foundation Axiom, is an ordinal i is a transitive

set and (, ) is a strict linear ordering. Also, by the Foundation

Axiom, (, ) is always irreexive. Combining these observations

with Lemma 3.14 we get the following useful characterization of

when a set is an ordinal.

Lemma 3.16 A set is an ordinal i is a transitive set, every

element of is a transitive set and (, ) is total.

In fact, we have seen some ordinals. If n , then n is an

ordinal. Also, is an ordinal. We have also seen transitive sets

that are not ordinals. For example, for every n , if n > 2, then

Vn is a transitive set that is not an ordinal.

It is important to have a reasonably good picture of where we

are headed before plunging into technical facts about ordinals. As

you read this paragraph, beware of the signicant work required

to justify this description of the ordinals, work that is captured

by Lemmas 3.18, 3.19, 3.20, 3.21 and 3.22, and results on ordinal

3.2 Ordinal numbers 31

addition in the next section. Here is the picture you should have

in mind. Starting from the empty set, we use the operation

{}

at successor stages and take unions at limit stages to generate all

the ordinals beginning with the natural numbers 0 = , 1 = {0},

2 = {0, 1}, 3 = {0, 1, 2}, etc. The next ordinal after all the natural

numbers is the set of natural numbers,

= {0, 1, 2, . . . }.

After comes the innite sequence of ordinals

+ 1 = {0, 1, 2, . . . , }

+ 2 = {0, 1, 2, . . . , , + 1}

+ 3 = {0, 1, 2, . . . , , + 1, + 2}

..

.

followed by innitely more ordinals

+ = {0, 1, 2, . . . , , + 1, + 2, . . . }

+ + 1 = {0, 1, 2, . . . , , + 1, + 2, . . . , + }

+ + 2 = {0, 1, 2, . . . , , + 1, + 2, . . . , + , + + 1}

..

.

after which comes the ordinal

+ + .

Skipping ahead, we eventually get to

+++

and, somewhat later, to

= + + + .

The list of ordinals never ends. Notice that, for natural numbers,

the membership relation coincides with the usual strict linear

ordering < on the natural numbers, and that this pattern contin-

ues through the ordinals we have listed above. Namely,

0 1 2 3 + 1 + 2 + 3 .

32 Order

Returning now to a rigorous exposition, we make the following

general notational rules.

Denition 3.17 If and are ordinals, then we may write

<

and

( < or = ).

This convention will allow us to write

0 < 1 < 2 < 3 < < < + 1 < + 2 < + 3 <

with the same meaning as

0 1 2 3 + 1 + 2 + 3

once we nish developing the theory of ordinals.

Next are some important basic principles about ordinals to

which we have alluded above. The proofs might seem confusing

the rst time through because they involve all three notions: , <

and . Read slowly and attentively. Remember, in general, once

we establish that and are ordinals (but not before) we are free

to write < instead of .

Lemma 3.18 Let be an ordinal. Then every element of is

an ordinal. Thus

= { | is an ordinal and < }.

Proof Let . By the forward direction of Lemma 3.14, every

member of is a transitive set. Thus is a transitive set. Because

is a transitive set, . Since (, ) is a wellordering, so

is (, ). (Every subset of a wellordered set is also wellordered.)

Therefore, is an ordinal by Denition 3.15.

Lemma 3.19 Let and be ordinals. Then

.

Proof First we prove the forward (left to right) direction. Assume

. If = , we are done. So assume < . In other words,

. Since is a transitive set, . So, again, we are done.

Now, for the proof of the reverse (right to left) direction, assume

that . If = , then we are done, so assume that = .

3.2 Ordinal numbers 33

Then (set dierence) is a non-empty subset of and (, <)

is a wellordering, so we may let be the <-least element of .

Claim 3.19.1 = .

Proof of claim By the Extensionality Axiom, Lemma 3.18 and

the fact that both and are subsets of , the claim is equivalent

to saying that, for every < ,

< < .

Let < be given. If < , then < by the denition of . In

order to nish the proof of the claim, we assume that < but

< and work towards a contradiction. Since and are both

elements of and (, <) is a total ordering, saying that < is

equivalent to saying that . Combining facts, we have that

< .

This is the same as saying that either

and

or else

= and .

Either way, since is a transitive set, we conclude that .

But by the denition of . This contradiction proves the

claim.

Now we are done proving the reverse direction of Lemma 3.19

because = < .

Lemma 3.20 If and are ordinals, then either < or

= or > .

Proof Assume that . By Lemma 3.19, . Because

(, <) is a wellordering, we may let be the <-least element of

(set dierence). Then but . Hence, by

Lemma 3.19 but < . Thus = < .

Lemma 3.21 Let be an ordinal and = {}. Then is

an ordinal and < . Moreover, if is an ordinal and < ,

then .

34 Order

Proof By Lemma 3.10, is a transitive set. Obviously, every ele-

ment of is also a transitive set and (, ) is total. By Lemma 3.16,

is an ordinal. Clearly, < . Suppose that is an ordinal and

< . Then since is transitive. Hence = {} .

By Lemma 3.19, .

Because of Lemma 3.21, it is natural to write + 1 = {}

for ordinals . We call + 1 a successor ordinal. Non-zero ordinals

that are not successor ordinals are called limit ordinals.

Lemma 3.22 Let A be a set of ordinals and = A. Then

is an ordinal and, for every A, . Moreover, if is an

ordinal and for every A, then .

The proof of Lemma 3.22 is left as a practice problem; it builds

on the chain of lemmas starting from Lemma 3.12. Recall that

supremum is another way to say least upper bound. Lemma 3.22

says that if A is a set of ordinals, then sup(A) = A. If A is a set

of ordinals and sup(A) A, then sup(A) is the maximumelement

of A, in which case we may write max(A) = sup(A) = A. But

not every set of ordinals has a maximum element. For example,

sup ({5, 6, 7, . . . }) = {5, 6, 7, . . . } =

but

{5, 6, 7, . . . } ,

so {5, 6, 7, . . . } does not have a maximum element. On the other

hand,

sup ({5, 6, 7, . . . , }) = {5, 6, 7, . . . , } =

and

{5, 6, 7, . . . , } ,

so

max ({5, 6, 7, . . . , }) = .

If A is a set of ordinals and A = , then A has a <-least ele-

ment called the minimum of A and denoted min(A). To justify

the denition of min(A) use the fact that A sup(A) + 1 and

(sup(A) + 1, <) is a wellordering.

Since they are important, let us state versions of Theorems 3.6

and 3.8 that relate to ordinals.

3.2 Ordinal numbers 35

Theorem 3.23 (Proofs by induction) Let P () be a statement

about a variable . Assume that, for every ordinal ,

( < P () holds) = P () holds.

Then, for every ordinal , P () holds.

Proof Consider an arbitrary ordinal and let = + 1. We

prove that P () holds for every < by induction along the

wellordering (, <) using Theorem 3.6. But there is really nothing

to do because the assumption here is at least as strong as the

assumption of Theorem 3.6. To see the connection more clearly,

substitute (A, ) = (, <), y = and x = in the statement of

Theorem 3.6.

Notice that Theorem 3.23 gives us a method for showing that

a property holds for every ordinal, not just every ordinal up to a

given ordinal. Often, the verication of the hypothesis of Theo-

rem 3.23 is broken up into three cases: = 0, = + 1 and

a limit ordinal. Similarly, in applications of Theorem 3.24 below,

sometimes the denition of F (, ) is broken up into three cases:

= 0, = + 1 and a limit ordinal.

Theorem 3.24 (Recursive denitions) Suppose that is an or-

dinal. Let

F :P B

be a function where B is a set and P is the set of partial functions

from to B. Then there is a unique function G : B such that

G() = F (, G )

for every < .

Notice that Theorem 3.24 has a dierent avor than Theo-

rem 3.23 in that it only tells us how to make recursive denitions

up to a given ordinal . Although it is possible to state a theorem

on recursive denitions through all the ordinals, the statement

would be overly technical, so we prefer to show how such deni-

tions can be made and justied by giving an example.

Here is an illustration of a recursive denition followed by a

proof by induction on the ordinals. If is an ordinal, then using

Theorem 3.24, we may dene a function V with domain

36 Order

by saying that

V0 = ,

V+1 = P (V )

and

V = {V | < }

if is a limit ordinal. If we really want to match up this denition

of V with Theorem 3.24, then we end up with G : V starting

with

if = 0

F (, g) = P(g()) if = + 1

{g() | < } if is a limit ordinal.

Notice that the denition V does not depend on so really what

we have is a way of assigning a set V to each ordinal . The

assignment V is not a function because what should be its

domain is not a set by the following lemma.

Lemma 3.25 There is no set of all ordinals.

Proof Suppose for contradiction that

= { | is an ordinal}

is a set. It is easy to see that is transitive and (, ) is a

wellordering. But then is an ordinal. In other words, .

But then

is an innite descending sequence of members of . This contra-

dicts that (, ) is a wellordering.

Notice that, once we realized that , we could have used the

Foundation Axiom to nish the proof of Lemma 3.25 slightly more

quickly. The Foundation Axiom automatically holds for ordinals,

which is essentially how we got away without using it above.

We have been discussing recursion; now let us see an example

of induction.

Lemma 3.26 Let be an ordinal and < . Then the following

hold.

3.2 Ordinal numbers 37

(1) V is a transitive set.

(2) For every < , V V .

hypothesis is that the lemma holds for every < .

Base case = 0.

In this case, (1)0 holds because V0 = is a transitive set,

whereas (2)0 holds because there is nothing to check.

Successor case = + 1.

In this case, V = P (V ). By Lemma 3.11, since (1) holds,

(1) holds too. Also, since (1) holds, if y V , then y V , so

y V . This shows that V V . This conclusion together with

(2) implies that (2) holds.

Limit case is a limit ordinal.

In this case, V = {V | < }. The fact that (1) holds is

immediate from Lemma 3.12 and the assumption that (1) for

< . Also, (2) follows from (2) for < .

Here is an interesting fact that plays no role in the rest of the

book. It turns out that the Foundation Axiom is equivalent to the

statement

x ( is an ordinal and x V ).

That is, every set belongs to some V . The proof can be found in

various graduate level textbooks on set theory.

Now we continue developing the theory of ordinal numbers.

Denition 3.27 Let R A A and S B B. Suppose that

: A B. Then is an isomorphism from (A, R) to (B, S) i

is a bijection from A to B and, for every x, y A,

xRy (x)S(y).

We write : (A, R) (B, S) in this case.

is summarized by the following theorem, which is a special case of

a more general theorem by Mostowski.

38 Order

Theorem 3.28 (Mostowski collapse) Let (A, ) be a wellorder-

ing. Then there exists an ordinal and an isomorphism

: (A, ) (, <).

Moreover, and are unique in the sense that if is an ordinal

and

: (A, ) ( , <)

is an isomorphism, then = and = .

The isomorphism in the conclusion of Theorem 3.28 is called

the Mostowski collapse of (A, ). Here is a specic example. Let

A = {0, 1, 4}

and

m n (m, n A and m < n) .

Then the Mostowski collapse : (A, ) (, <) is determined by

the following table.

m (m)

2 0

3 1

5 2

6 3

7 4

.. ..

. .

In general, if x is the -least element of A, then (x) = 0. If,

in addition, y is the -least element of A {x}, then (y) = 1.

And so on. Notice that the list on the left has gaps (0, 1 and 4

are missing) but the list on the right has no gaps. Intuitively, the

function is called the collapse because it gets rid of the gaps.

Another way to think about the Mostowski collapse is as follows.

Suppose we list A as a0 , a1 , . . . in increasing order according to

using ordinals as indices. In the example above, we would have

a0 = 2, a1 = 3, a2 = 5, a3 = 6, etc. What we would end up with

is, for every b A,

b = a(b) .

In other words, the Mostowski collapse tells us the index.

3.2 Ordinal numbers 39

Proof of Theorem 3.28 We are given a wellordering (A, ) and

we are looking for an ordinal and an isomorphism

: (A, ) (, <).

Let us work backwards to see what the denitions of and

must be. Suppose we already have and as above. Then, for all

x, y A,

x y (x) < (y) (x) (y).

Therefore, for every y A,

(y) = {(x) | x y}.

Notice that this is a recursive equation because the value of (y) is

determined from {x | x y}. By Theorem 3.8, there is at most

one function that satises this recursive equation. Moreover,

once we know , we also know because = [A]. This explains

why and are unique. In other words, it proves the moreover

part of Theorem 3.28. It also gives us the clue we need to prove

the existence part of Theorem 3.28, which is what we do next.

Without assuming that is an ordinal and is an isomorphism,

recursively dene : A B by setting

(y) = {(x) | x y}.

This is not entirely legitimate, at least not if we wish to implement

Theorem 3.8, because we are required to name the set B before we

dene by recursion. We will leave out the argument that we can

choose B in advance since it involves reasoning from the axioms

of ZFC that is overly technical from the perspective of this book.

Claim 3.28.1 is an isomorphism from (A, ) to ([A], ).

Proof First we show that is an injection. Consider arbitrary

x, y A such that x = y. Then either x y or y x. Suppose

that x y. Then (x) (y) by the denition of . On the other

hand, (x) (x) by the Foundation Axiom. Hence (x) = (y).

The proof is similar if y x.

Every injection is a bijection with its range. It remains to see

that is order preserving. We already noted that if x y, then

(x) (y). For the converse, suppose that (x) (y). Then

there exists x y such that (x ) = (x). Since is an injection,

x = x. Hence x y as required.

40 Order

Claim 3.28.2 [A] is an ordinal.

Proof If u v and v [A], then there are x, y A such that

u = (x) and v = (y), so u [A]. This shows that [A] is a

transitive set. By the previous claim, ([A], ) is an isomorphic

copy of (A, ). Since (A, ) is a wellordering, so is ([A], ).

Setting = [A] concludes the proof of Theorem 3.28.

The previous theorem justies the following denition.

Denition 3.29 Let (A, A ) be a wellordering. Then

type(A, A )

is the unique ordinal isomorphic to (A, A ).

Suppose that A is a set of ordinals. Say A . Then (, <) is a

wellordering and hence so is (A, <). This motivates the following

notation.

Denition 3.30 Let A be a set of ordinals. Then

type(A) = type(A, <).

We end this section with a technical lemma about the Mostowski

collapse of a set of ordinals. It will be used in Chapter 4.

Lemma 3.31 Let A be a set of ordinals and suppose that A .

Let

= type(A)

and

: (A, <) (, <)

be the Mostowski collapse of (A, <). Then, for every A,

() .

Moreover, .

Proof First we prove that () by induction on A. Recall

that if and are ordinals, then

.

This is by Lemma 3.19. Therefore, what we must show is equiva-

lent to

()

3.3 Ordinal arithmetic 41

for every A. By the denition of the Mostowski collapse,

() = {() | A }.

By the induction hypothesis, for every A ,

()

so

() < + 1 .

Therefore,

() { + 1 | A } .

that if A, then

() < .

Since = [A], this implies that , hence .

In this section, we extend the usual notions of addition, multipli-

cation and exponentiation of natural numbers to all ordinals.

There are various ways to join a pair of sets. For example, given

A and B, we can form their union A B. But sometimes we want

to take a disjoint union instead. This means that we take the

union of disjoint copies of A and B. The advantage is that, given

a point in the disjoint union, it comes either from A or from B

but not both. A convenient way to dene the disjoint union is

({0} A) ({1} B). Building on this idea, sometimes we are

given two wellorderings and we want to put one after the other to

form a new wellordering. The following denition tells us how.

Denition 3.32 Let (A, A ) and (B, B ) be wellorderings. Then

their concatenation is

(A, A )

(B, B ) = (C, C )

where

C = ({0} A) ({1} B)

42 Order

and, for all (i, x), (j, y) C,

(i = j = 0 and x A y)

or

(i, x) C (j, y)

(i = 0 and j = 1) .

or

(i = j = 1 and x B y)

Lemma 3.33 Let (A, A ) and (B, B ) be wellorderings, and

(C, C ) = (A, A )

(B, B ).

Then (C, C ) is a wellordering.

Proof It is straightforward to verify that (C, C ) is a strict linear

ordering. For contradiction, suppose that (in , xn ) | n < is an

innite descending sequence from C. Since (B, B ) is a wellorder-

ing, there is some m < such that, for every n < , if m < n,

then in = 0. But then xn | n < m is an innite descending

sequence from A, which is a contradiction.

+ = type ((, <)

(, <)) .

In other words, + is the unique ordinal isomorphic to

(, <)

(, <).

It is the combination of Lemma 3.33 and Theorem 3.28 that justi-

es Denition 3.34. We also remark that, earlier, we dened + 1

to mean {} whereas here we used a dierent denition of

+ 1. The two denitions coincide because

(, <)

(1, <) ( {}, <).

Make sure you understand what is being asserted here and why it

is true!

Example 3 + 2 = 5. This is because if (C, C ) = (3, <)

(2, <),

then

(0, 0) C (0, 1) C (0, 2) C (1, 0) C (1, 1)

and so we see that (C, C ) (5, <).

3.3 Ordinal arithmetic 43

Example 2 + 3 = 5. This is because if (C, C ) = (2, <)

(3, <),

then

(0, 0) C (0, 1) C (1, 0) C (1, 1) C (1, 2)

and so we see that (C, C ) (5, <).

Example 3 + = . This is because (C, C ) = (3, <)

(, <)

consists of the initial segment

(0, 0) C (0, 1) C (0, 2)

followed by the innite tail

(1, 0) C (1, 1) C (1, 2) C (1, 3) C (1, 4) C (1, 5) C

from which we see that

(C, C ) (, <)

according to the isomorphism

n if i = 0

(i, n)

3+n if i = 1.

Notice that 3 + = = + 3, so ordinal addition is not

commutative! However, ordinal addition is associative.

Lemma 3.35 For all ordinals , and ,

( + ) + = + ( + ).

Proof Let

(C, C ) = (, <)

(, <)

and

D = (C, C )

(, <).

Let

(E, E ) = (, <)

(, <)

and

F = (, <)

(E, E ).

It is enough to see that there is an isomorphism

: (D, D ) (F, F ).

44 Order

Dene by cases according to the following list of equations.

((0, (0, ))) = (0, )

((0, (1, ))) = (1, (0, ))

((1, )) = (1, (1, ))

It is clear that this works.

Addition can also be dened recursively in terms of the assign-

ment that takes an ordinal to its successor + 1. The following

lemma shows this.

function f with domain such that, for every < ,

if = 0

f () = f () + 1 if = + 1

sup ({f () | < }) if is a limit ordinal,

f () = + .

addition satises the three conditions we specied for f . Namely,

+ 0 = ,

+ ( + 1) = ( + ) + 1 and

if is a limit ordinal, then + = sup ({ + | < }).

Here is an entertaining false argument. We saw that

3 + = = 0 + .

Subtracting from both sides of this equation, we see that 3 = 0.

Do you see why this is nonsense? There is no inverse operation for

ordinal addition. The following lemma is as close as we come to

ordinal subtraction.

ordinal such that + = .

3.3 Ordinal arithmetic 45

Sketch of proof Let D = { | < }. It is clear that

(, <)

(D, <) (, <).

By Theorem 3.28, there is a unique ordinal such that

(D, <) (, <),

namely,

= type(D).

It is clear that

(, <)

(, <) (, <).

This is equivalent to saying that

+ = .

It remains to see that is the unique solution to this equation.

Suppose that

+ = .

Then

(, <)

( , <) (, <),

from which one can argue that

(D, <) ( , <).

By the uniqueness clause of Theorem 3.28, we see that = . The

remaining details form Exercise 3.12.

Just before Lemma 3.37, we saw that + = 3 + does not

imply = 3. In other words, we cannot cancel on the right. It is a

consequence of the uniqueness clause in Lemma 3.37 that we can

cancel on the left. For example, if + = + 3, then = 3.

Now we work towards dening multiplication. For this, we use

another way of putting together two wellorderings.

Denition 3.38 Let (A, A ) and (B, B ) be wellorderings. Then

the lexicographic ordering on A B is the relation such that,

for all (x, y), (x , y ) A B,

(x A x )

(x, y) (x , y ) or .

(x = x and y B y )

46 Order

Lemma 3.39 Let (A, A ) and (B, B ) be wellorderings. Then

(C, C ) is a wellordering where C = A B and C is the lexico-

graphic ordering on C.

The proof of Lemma 3.39 is Exercise 3.7.

Denition 3.40 Let and be ordinals. Let be the unique

ordinal isomorphic to the lexicographic ordering on . Then

the product is

= .

This is not a misprint; by tradition, ordinal multiplication is

read from right to left. By , we mean many copies of ,

not the other way around, and sometimes it matters for innite

ordinals.

Example 3 2 = 6. This is because, if C = 2 3 and C is the

lexicographic order on C, then

(0, 0) C (0, 1) C (0, 2) C (1, 0) C (1, 1) (1, 2)

and so we see that (C, C ) (6, ).

Example 2 3 = 6. This is because, if C = 3 2 and C is the

lexicographic order on C, then

(0, 0) C (0, 1) C (1, 0) C (1, 1) C (2, 0) (2, 1)

and so we see that (C, C ) (6, ).

Example 3 = . This is because, if C = 3 and C is the

lexicographic order on C, then (C, C ) looks like

(0, 0) C (0, 1) C (0, 2)

followed by

(1, 0) C (1, 1) C (1, 2)

then

(2, 0) C (2, 1) C (2, 2)

then

(3, 0) C (3, 1) C (3, 2)

then

(4, 0) C (4, 1) C (4, 2)

3.3 Ordinal arithmetic 47

and so on. From this, we see that

(C, C ) (, <)

according to the isomorphism

(i, n) 3 i + n.

Example 3 = + + . This is because, if C = 3 and C

is the lexicographic order on C, then (C, C ) looks like

(0, 0) C (0, 1) C (0, 2) C (0, 3) C (0, 4) C

followed by

(1, 0) C (1, 1) C (1, 2) C (1, 3) C (1, 4) C

then

(2, 0) C (2, 1) C (2, 2) C (2, 3) C (2, 4) C .

From this we see that

(C, C ) ( + + , <)

according to the isomorphism

n if i = 0

(i, n) + n if i = 1

+ + n if i = 2,

which we could also write as

(i, n) i + n.

Notice that

3 = = 3,

so ordinal multiplication is not commutative! However, ordinal

multiplication is associative as the following lemma shows. The

proof is Exercise 3.6(5).

Lemma 3.41 For all ordinals , and ,

( ) = ( ) .

Ordinal multiplication distributes over ordinal addition in the

following way. The proof is the rst part of Exercise 3.8.

48 Order

Lemma 3.42 For all ordinals , and ,

( + ) = ( ) + ( ).

Ordinal multiplication can also be dened recursively in terms

of ordinal addition as the following lemma shows. The proof is

Exercise 3.9.

Lemma 3.43 Let and be ordinals. Then there is a unique

function f with domain such that, for every < ,

0 if = 0

f () = f () + if = + 1

sup ({f () | < }) if is a limit ordinal,

namely, the function given by

f () = .

For ordinal addition and ordinal multiplication, we started with

denitions in terms of wellorderings, then stated lemmas giving

equivalent recursive denitions. For variety and because it is eas-

ier in this case, we give the recursive denition of ordinal expo-

nentiation and leave the interpretation in terms of wellorderings

to the reader as a rather tricky exercise. (Exercise 3.15 will get

the reader started.)

Denition 3.44 For every ordinal ,

0 = 1,

for every ordinal ,

+1 =

and

for every limit ordinal ,

= sup | 0 < < .

Example If 2 , then + = . To see this, rst note that

2 = = (1 + ) = ( 1) + ( ) = + 2 .

3.3 Ordinal arithmetic 49

Now suppose that 2 . By Lemma 3.37, there is an ordinal

such that = 2 + . Thus,

= 2 + = ( + 2 ) + = + ( 2 + ) = + .

Exercises

Exercise 3.1 Let (A, A ) be a wellordering such that A = .

For each y A, dene

pred(A,A ) (y) = {x A | x A y}.

Suppose that S A and, for all x, y A, if y S and x A y,

then x S. Prove that there exists y A such that

S = pred(A,A ) (y).

Exercise 3.2 Let (A, A ) and (B, B ) be wellorderings. Use

Theorem 3.28 to prove that exactly one of the following conditions

holds:

(A, A ) (B, B ).

There exists y B,

(A, A ) pred(B,B ) (y) ordered by B .

There exists y A,

(B, B ) pred(A,A ) (y) ordered by A .

{ V | is an ordinal} = .

As a hint, see the last part of Exercise 2.3.

Exercise 3.4 Prove that the following facts about ordinal addi-

tion hold for all ordinals , and .

1. 0 + = = + 0.

2. + .

3. If < , then + < + .

4. If , then + + .

Exercise 3.5 Complete the proof of Lemma 3.36.

50 Order

Exercise 3.6 Prove that the following facts about ordinal mul-

tiplication hold for all ordinals , and .

1. 0 = 0 = 0.

2. 1 = 1 .

3. If 0 < and < , then < .

4. If , then .

5. ( ) = ( ). (That is, prove Lemma 3.41.)

Exercise 3.7 Prove Lemma 3.39.

Exercise 3.8 This exercise is on distributive laws for ordinal

addition and multiplication.

1. Prove Lemma 3.42.

2. Give an example of ordinals , and such that

( + ) = ( ) + ( ).

Exercise 3.9 Prove Lemma 3.43.

Exercise 3.10 Prove that the following facts about ordinal ex-

ponentiation hold for all ordinals , and .

1. If = 0, then 0 = 0.

2. 1 = 1.

3. If 1 < and < , then < .

4. If , then .

5. If 1 < , then .

Exercise 3.11 Prove that the following facts about ordinal arith-

metic hold for all ordinals , and .

1. + = .

2. ( ) = .

Exercise 3.12 Complete the proof of Lemma 3.37.

Exercise 3.13 Let and be ordinals. Prove that if > 0,

then there are unique ordinals and such that < and

= ( ) + .

3.3 Ordinal arithmetic 51

Exercise 3.14 Let be an ordinal such that = 0. Prove that

there are unique n, 1 , . . . , n , 1 , . . . , n such that

1 n < ,

1 > > n ,

1 i < for every i = 1, . . . n, and

= 1 1 + + n n .

This is called Cantor normal form.

Exercise 3.15 For each function x : , dene the support

of x to be the set

{n < | x(n) = 0}.

Recall that

= {x | x is a function from to }.

Let

A = {x | x has nite support}.

Given x, y A such that x = y, there exists a largest n < such

that x(n) = y(n) and we dene

x A y x(n) < y(n).

Prove that (A, A ) is a wellordering and

type(A, A ) = (ordinal exponentiation).

Exercise 3.16 Find two functions

f : +

and

g :+ ++

such that

sup(f []) = +

and

sup(g[ + ]) = + +

but if h = g f is the composition, then

sup(h[]) < + + .

52 Order

Exercise 3.17 Let < < be three limit ordinals and

f :

and

g:

be two functions such that

sup(f []) =

and

sup(g[]) = .

Assume that g is non-decreasing in the sense that if < ,

then g() g(). Let h = g f be the composition. Prove that

sup(h[]) = .

4

Cardinality

is a fancy word for size. Cardinal numbers will be dened to be

certain kinds of ordinal numbers. Not every ordinal is a cardinal

though. The theory here builds on that of the previous chapter.

Denition 4.1 We say that A and B have the same cardinality

and write A B i there is a bijection from A to B.

Granted, it is strange to say that two sets have the same cardi-

nality without having said what cardinality means. But we need

a lemma before giving that denition.

Lemma 4.2 For every set A, there exists an ordinal such that

A.

f (1) be an element of A other than f (0), etc. We keep going until

we list all the elements of A as f () for some < . Now we have

to make rigorous mathematical sense of this idea. Let

F = {X A | X = } .

such that, for every X F,

c(X) X.

54 Cardinality

Dene f () by recursion on ordinals as follows. If

A f [] F,

then let

f () = c (A f []) .

On the other hand, if

A f [] F,

then leave f () undened for every .

Observe that if < and both f () and f () are dened,

then f () f [] and f () A f [], hence f () = f (). This

calculation almost shows that f is an injection; what is missing is

a proof that f has a set domain.

First suppose there is an ordinal such that f () is undened.

Let be the least such ordinal. Then f is an injection with =

dom(f ) and ran(f ) A. Also,

A f [] F,

which means exactly that A f [] = . Equivalently, it means

that f [] = A. Therefore, f is a bijection from to A as desired.

It remains to see that there is an ordinal such that f () is

undened. Suppose otherwise. Then, for every ordinal , f () is

dened. Let

S = {a A | there exists such that f () = a}.

Then, for every a S, there exists a unique such that f () = a.

Apply the Replacement Scheme to this property to conclude that

there is a set and a function g : S such that, for every

a S, f (g(a)) = a. Because f () is dened for every ordinal ,

and because f () = f () whenever = , it must be that is

the set of all ordinals. But we proved earlier, in Lemma 3.25, that

there is no set of all ordinals.

We remark that the use of the Axiom of Choice cannot be elim-

inated from the proof of Lemma 4.2. Elaborating on the meaning

of this remark: If you remove the Axiom of Choice from ZFC the

result is known as ZF. It turns out that ZF + Lemma 4.2 implies

the Axiom of Choice. Prove this as a practice problem!

At last, we dene cardinal numbers and the cardinality of sets.

4.1 Cardinal numbers 55

Denition 4.3 is a cardinal i is an ordinal and, for every

< ,

Denition 4.4 |A| is the least ordinal such that A .

You should convince yourself of the following facts, whose proofs

amount to composing bijections.

Lemma 4.5 |A| is a cardinal.

Lemma 4.6 A B |A| = |B|.

Every natural number is both an ordinal and a cardinal. Also,

is a cardinal. However, the ordinals

+ 1, + 2, + 3, . . .

are all countably innite, which is to say that their cardinality is .

(For us, countable means nite or countably innite. Uncountable

means not countable.) In particular, the ordinals displayed above

are not cardinals. Moreover, the ordinals

2, 3, 4, . . .

are not cardinals. Nor are the ordinals

2 , 3 , 4 , . . . .

The ordinals

, , ...

are not cardinals either. All of these are ordinals are countable, as

you should try to verify. We need another idea to reach uncount-

able sets.

Theorem 4.7 (Cantor) There is no surjection from A to P(A).

Proof Consider an arbitrary function f : A P(A). Let

C = {x A | x f (x)}.

For every x A,

x C x f (x)

hence

C = f (x).

56 Cardinality

In particular,

C f [A].

Therefore f : A P(A) is not a surjection.

The proof we just gave is an example of a diagonal argument.

This is an imprecise term that you will see used in more and more

general ways throughout the book. Here, the idea is that if you

visualize the graph of the relation

{(x, y) A A | x f (y)},

then what we call the diagonal is

D = {x | x f (x)}.

To come up with a set C missing from the range of f , we take the

complement of the diagonal,

C = A D.

The reason C = f (x) is that one of C and f (x) has x as an element,

and the other does not. We can express the previous sentence in

terms of the symmetric dierence:

x C f (x)

hence

C f (x) =

thus

C = f (x).

Corollary 4.8 P() is uncountable.

Proof Clearly P() is innite. By Theorem 4.7, there is no sur-

jection from to P(). Hence there is no bijection.

Bijections are used in the denition of cardinality but sometimes

only surjections or injections are easily available. Here are some

basic facts that relate these notions.

Lemma 4.9 If < and there is a surjection f : , then

is not a cardinal.

4.1 Cardinal numbers 57

Proof Suppose that < and f : is a surjection. Let

S = { < | f () = f () for every < }.

Let g = f S. Then g is a bijection from S to . Let = type(S)

and : (S, <) be the Mostowski collapse of (S, <). By

Lemma 3.31, since S , . So < . Let

h = g 1 .

Then h : is a bijection. In other words . Thus is not

a cardinal.

Lemma 4.10 Let and be cardinals. Then

there is an injection f : .

Proof If , then the identity function is an injection from

to . For the reverse direction, suppose that f : is an

injection. Let

S = f [] = {f () | < }

and : (S, <) be the Mostowski collapse. Then the composi-

tion

f :

is a bijection. Thus || = ||. Since is a cardinal, || = . Because

S , by Lemma 3.31, . Putting these facts together we

have that

= || = || .

such that > .

Proof Let = |P()|. First note that the identity function is an

injection from to P(), which implies that by Lemma

4.10. But Theorem 4.7 implies that = , so < .

Denition 4.12 + is the least cardinal strictly greater than .

The proof of Corollary 4.11 shows that + |P()|. You should

not assume that equality holds. The discussion after Corollary 4.27

explains why.

Sometimes, the following theorem is covered in courses other

58 Cardinality

than set theory but with a very dierent proof. (See Exercise 4.14.)

The proof here is extremely short because it builds on the theory

of ordinals and cardinals, which we have at hand.

Theorem 4.13 (CantorBernsteinSchroeder) Suppose that

there is an injection from A to B, and

there is an injection from B to A.

Then A B.

Proof First recall that A |A| and B |B|. By Lemma 4.10

and the hypothesis of the theorem, |A| |B| and |B| |A|. So

|A| = |B|. Hence A B.

By Corollary 4.11, there is no largest cardinal. The next result

is a kind of continuity for cardinal numbers.

Lemma 4.14 If A is a set of cardinals, then sup(A) is a cardinal.

Proof We may assume that A does not have a maximum element,

as otherwise

sup(A) = max(A) A.

For contradiction, suppose that sup(A) is not a cardinal. Let <

sup(A) and f : sup(A) be a surjection. Since A does not have

a maximum element, there exists A such that < . Let

S = { < | f () }

and g = f S. Then g : S is a surjection. Let = type(S)

and

: (S, <)

be the Mostowski collapse. Because S , by Lemma 3.31, .

So < . Let h = g 1 . Then h : is a surjection. Together

with Lemma 4.9, this shows that is a not a cardinal. But we

assumed that every element of A is a cardinal.

Next we list the innite cardinals in increasing order using or-

dinals as indices:

0 , 1 , 2 , . . . , , +1 , +2 , . . .

The letter is read aleph and is the rst letter of the Hebrew al-

phabet. Here is the formal recursive denition of our list of innite

cardinals.

4.1 Cardinal numbers 59

Denition 4.15 Let

0 = .

By recursion on > 0, dene to be the least cardinal greater

than for all < .

It is tempting and correct to write

= min | is a cardinal and > for all <

but keep in mind that what we have inside min() is not a set.

Corollary 4.16 We have that

0 = ,

+1 = ( )+ for every ordinal , and

= sup ({ | < }) for every limit ordinal .

Proof The rst two clauses are obvious. The last clause follows

from Lemma 4.14.

is a cardinal such that = + . If = 0 and is not a successor

cardinal, then we say that is a limit cardinal.

It is important to note that the only successor ordinals that are

cardinals are the natural numbers. Every innite cardinal (includ-

ing every innite successor cardinal) is a limit ordinal. Corollary

4.16 implies the following result, which spells out how these con-

cepts are related.

Corollary 4.18 For every ordinal ,

is a limit cardinal i either = 0 or is a limit ordinal,

and

is a successor cardinal i is a successor ordinal.

Corollary 4.18 covers all innite cardinals by the following fact.

Lemma 4.19 Let be an innite cardinal. Then there is an

ordinal such that = .

Proof By induction on innite cardinals .

60 Cardinality

Base case = .

Then = 0 and 0 < = .

Successor case = + .

By the induction hypothesis, there is an such that = .

Then

= + = ( )+ = +1

by Corollary 4.16, and

+ 1 + 1 < + = .

Limit case is a limit cardinal.

Let

= sup({ | < }).

By the case hypothesis and the induction hypothesis,

sup({ | < })

= sup({ < | is a cardinal})

=

and is a limit ordinal. Similarly,

= sup({ < | is a cardinal})

= sup({ | < })

= sup({ | < })

= .

The last line is by Corollary 4.16.

Every cardinal is an ordinal but cardinal arithmetic is completely

dierent from ordinal arithmetic when it comes to innite cardi-

nals. It is important to keep track of which kind of arithmetic you

are doing. Usually, it is clear from the context.

4.2 Cardinal arithmetic 61

Denition 4.20 For all cardinals and ,

= |({0} ) ({1} )|

and

= | | .

Unlike ordinal addition and multiplication, cardinal addition

and multiplication are commutative. The main point in seeing

that

=

is that if A and B are sets, then (x, y) (y, x) is a bijection from

A B to B A, hence |A B| = |B A|.

The next result says that, for natural numbers, cardinal addition

and multiplication coincide with ordinal addition and multiplica-

tion. The reader should work out the proofs as an exercise.

Lemma 4.21 If m, n < 0 , then mn = m+n and mn = mn.

Remember that ordinal addition and multiplication for innite

ordinals were interesting operations with subtle properties. By

contrast, cardinal addition and multiplication for innite cardinals

turn out to be trivial to calculate by the following two results.

Lemma 4.22 Let be an innite cardinal. Then = .

Proof By induction on . The induction hypothesis is that

=

whenever is a cardinal such that 0 < . We will use two

dierent orderings of the Cartesian product . First dene

, <lex (, )

i either

<

or

= and < .

Then dene

(, ) (, )

62 Cardinality

i either

max , < max (, )

or

max , = max (, ) and , <lex (, ) .

Figure 4.1 is a picture of ordered by . For a given < ,

the order increases across the horizontal arrow (1) leaving out

(, ), then increases up the vertical arrow (2) until it reaches

(, ). Next, increases across (3) leaving out ( + 1, + 1), then

up (4) including ( + 1, + 1). And so on.

We claim that is a wellordering of . It is obviously a strict

linear ordering. Towards seeing that is wellfounded, consider an

arbitrary S such that S = . Let

= min ({max(, ) | (, ) S}) .

Let

= min ({ | there is with (, ) S and max (, ) = })

and

= min ({ | (, ) S}) .

..

.

(0, + 1) _ _ _ _ _ _ _ _ _ _ _ _/ ( + 1, + 1)

(3 )

O

(0, ) _ _ _ _ _ _ _ _/ (, )

(1 )

O

(4 )

.. (2 )

.

(0, 0) (, 0) ( + 1, 0)

4.2 Cardinal arithmetic 63

It is easy to check that (, ) is the -least element of S. Now that

we know is a wellordering, we can talk about its order type.

Claim 4.22.1 type ( , ) = .

Proof First note that type ( , ) . This is because

(, 0) (, 0)

whenever < < . So it is enough to see that

type ( , ) .

For this, it is enough to see that, for every (, ) ,

type (pred (, ) , ) <

where, by denition,

pred (, ) = , | , (, ) .

If = 0 , then and are natural numbers and pred (, ) is

nite, hence

type (pred (, ) , ) < 0 =

as desired. So we may assume that 1 . Let

= max(, , 0 ) + 1

and

= ||.

Then is a cardinal and 0 < , so by the induction hypoth-

esis,

= .

Therefore

type (pred (, ) , ) type ( , ) < +

as desired.

From Claim 4.22.1, it follows that

( , ) (, <).

Since isomorphisms are bijections,

| | = || = .

This completes the proof of Lemma 4.22.

64 Cardinality

With regard to the proof of Lemma 4.22, we remark that if is

an ordinal, then the lexigraphic ordering on is a wellordering

of type (ordinal product). In other words,

type( , <lex ) = .

This is by the denition of ordinal multiplication, Denition 3.40.

In particular, if is an innite ordinal, then

|| < < ||+ .

In particular, is not a cardinal.

Lemma 4.22 is a special case of the following general result.

Theorem 4.23 If 0 < are cardinals and 0 , then

= = .

Proof The theorem can be veried easily if = 1. If 2, then

by Lemma 4.22,

=2=

so equality holds throughout.

Theorem 4.23 tells us that if at least one of and is innite,

then

= = max(, ).

So cardinal addition and multiplication really are easy to calculate!

Now we dene cardinal exponentiation. The notation is the

same as for ordinal exponentiation but the meaning is dierent.

For this denition, recall from Exercise 2.8 that if A and B are

sets, then

A

B = {f | f is a function from A to B}.

= | |

We repeat our warning that writing is ambiguous. Do you

mean ordinal exponentiation or cardinal exponentiation? Always

make sure it is clear which, either from the context, or by saying

so explicitly.

Our rst fact about cardinal exponentiation is that it is the same

4.2 Cardinal arithmetic 65

as ordinal exponentiation when restricted to the natural numbers.

The reader should work out the proof as an exercise.

Lemma 4.25 If m, n < 0 , then nm is the same whether com-

puted as ordinal exponentiation or cardinal exponentiation.

Cardinal exponentiation becomes quite interesting when we look

at innite powers.

Lemma 4.26 |P(A)| = 2|A|

Proof Dene a function

char : P(A) A 2

by setting

0 if a X

char(X)(a) =

1 if a X

for every X A and a A. Note that char is a function whose

outputs are themselves functions. The function

char(X) : A 2

is called the characteristic function of X in A.1 To see that char

is an injection, note that if X, Y A and X = Y , then, for any

a XY ,

char(X) (a) = char(Y ) (a) ,

so

char(X) = char(Y ).

To see that char is a surjection, note that if f A 2, then

char ({a A | f (a) = 1}) = f.

We have shown that P(A) A 2. From this, Lemma 4.26 is clear.

a consequence of Lemma 4.26.

Corollary 4.27 If is a cardinal, then 2 + .

1 Elsewhere, the characteristic function of X is written X , where is the lower

case Greek letter chi, and X is a subscript. But this is hard to read.

66 Cardinality

This brings up an important question, namely, what is the value

of 2 ? Focusing on the most basic case, what is the value of 20 ?

Is 20 = 1 ? Perhaps 20 = 2 ? Could it be that 20 = 7+4 ?

We know that, for some ordinal 1,

20 =

but it turns out that the value of cannot be determined using

only the axioms of ZFC because of deep theorems of Kurt G odel

and Paul Cohen.2 This is interesting because

|R| = 20 ,

so really we are asking how many real numbers there are. This

problem was posed by Georg Cantor in the late 1800s, who asked:

Is there is an uncountable A R such that A R?

It was also rst on the most famous list of open problems, which

David Hilbert compiled at the start of the twentieth century. The

answer no to Cantors question is known as the Continuum Hy-

pothesis, or CH, which says 20 = 1 . The answer yes says that

20 2 . Cantor conjectured CH is true, but an informal poll sug-

gests that most set theorists today who have an opinion believe

CH is counterintuitive. Few have strong feelings about what the

actual value of 20 should be although some feel it should be 2 .

As we mentioned already, there are theorems due to Godel and Co-

hen which together say roughly that ZFC is not powerful enough

to answer Cantors question, so it is unknown what methodology

could lead to an answer. Additional explanation would be beyond

the scope of this book; it should be the topic of your next set

theory course!

Remember that, for nite numbers, (m )k = m k . Do not con-

k k

fuse this with the standard convention m = m( ) . The rst

equation generalizes to all cardinal numbers as the following lemma

shows. Other basic facts about cardinals can be found in the ex-

ercises.

2 The hypothesis of these Go del and Cohen theorems is that ZFC is a consistent

theory, meaning there is no proof that 0 = 1 using only the axioms of ZFC.

Go del proved that ZFC is consistent with CH in 1940. Cohen proved that ZFC is

consistent with the negation of CH in 1963. The combination of these results

says CH is independent of ZFC. These results can be found in many graduate

level textbooks on set theory.

4.3 Conality 67

Lemma 4.28 Let , and be cardinals. Then

( ) = .

Proof It is easy to see that

=

and

= .

There is a bijection

F :

dened by

F (g)(, ) = g()().

Putting together these observations, we are done.

Denition 4.29 (Strange notation) = for ordinals 1.

We write when we want to emphasize that it is an ordinal.

We write when we want to emphasize that it is a cardinal.

However, it is dicult to keep the notation consistent with such

intentions when we simultaneously consider cardinal and ordinal

properties of = . Ultimately, whether we write or

reduces to a matter of style.

Note that 0 is not dened; we always write either or 0 .

4.3 Conality

Recall that is the least cardinal greater than every n for n < .

In this sense, feels rather large. On the other hand, the function

n n

maps to and has range unbounded in , i.e.,

= sup n .

n<

reached from below in this way makes it feel somewhat smaller

68 Cardinality

than before. Here, we examine this phenomenon and tie it up

with cardinal arithmetic.

Denition 4.30 If is a limit ordinal, then the conality of ,

cf(),

is the least ordinal such that there exists a function f :

with

sup(f []) = .

We say is singular if cf() < . Otherwise, we say is regular.

The notions above are dened only for limit ordinals, not for 0 or

successor ordinals + 1. Keep in mind that every innite cardinal

is a limit ordinal. (Recall the reason is that |A {A}| = |A| for

every innite set A.) Here is a list of examples of conalities to

think about now and as you read.

cf (0 ) = 0

cf (1 ) = 1

cf (2 ) = 2

cf ( ) = 0

cf (+1 ) = +1

cf (+2 ) = +2

cf (+ ) = 0

cf (1 ) = 1

Here is a second list with more examples to think about.

cf ( + ) = 0

cf ( ) = 0

cf ( ) = 0 (ordinal exponentiation)

cf (1 + ) = 0

cf (1 + 1 ) = 1

cf (1 + 1 + ) = 0

4.3 Conality 69

equations listed above. Currently, you should see that, for every

limit ordinal ,

cf() || .

This is because every surjection onto has range unbounded in .

It follows from these inequalities that if cf() = , then || = .

In other words, if is a regular limit ordinal, then is a cardinal.

It is worth doing Exercises 3.16 and 3.17 before reading the

proof of the next result.

Lemma 4.31 cf() is a regular cardinal.

Proof Let f : cf() be a function whose range is unbounded

in . We know there is such a function f by the denition of

conality.

The following claim records slightly more information than we

need for the rest of the proof but the extra information is useful

elsewhere.

Claim 4.31.1 There is a function g : cf() such that

ran(g) is unbounded in ,

g is non-decreasing in the sense that if < cf(), then

g() g(),

g is continuous in the sense that for every limit ordinal <

cf(),

g() = sup g().

<

g() = sup f ().

<

is,

g : cf() .

To see that g is continuous, observe that, for every limit ordinal

< cf(),

g() = sup f () = sup sup f () = sup g().

< < < <

70 Cardinality

Clearly, cf() is a limit ordinal. Using this, we see that the range

of g is unbounded in because

sup g() = sup sup f () = sup f () = .

<cf() <cf() < <cf()

as in Claim 4.31.1. Since every regular limit ordinal is a cardinal,

in order to nish proving the lemma, it suces to show that cf()

is a regular ordinal. That is, given an ordinal < cf() and a

function h : cf(), we must conclude that the range of h is

bounded in cf(). Let S = h[]. For contradiction, assume that

sup(S) = cf().

Using this assumption and the fact that g is non-decreasing and

unbounded, we see that

sup g(h()) = sup g() = sup g() = .

< S <cf()

gh:

has range unbounded in . Because < cf(), this contradicts the

denition of conality.

The following result implies that cf(+1 ) = +1 for every

ordinal .

Lemma 4.32 Let be an innite cardinal and = + . Then

is a regular cardinal.

Proof We must show that cf() = . Let and f :

be a function. For each < , f () < = + , so |f ()| .

Thus

|sup(f [])| f () = < .

<

obviously cf() for every limit ordinal . Hence cf() = ,

which means is regular.

4.3 Conality 71

Now that we understand successor cardinals, let us look at a

few examples of singular cardinals.

Example Using the fact that

= sup n

n<

we see that

cf( ) = 0 .

Example Using the fact that

+ = sup +n ,

n<

we see that

cf(+ ) = 0 .

Example Using the fact that

1 = sup ,

< 1

we see that

cf(1 ) 1 .

We claim that

cf(1 ) = 0 .

Consider an arbitrary function

f : 1 .

Dene g : 1 by letting

g(n) = the least < 1 such that f (n) < .

Then, for every n < ,

|f (n)| f (n) < g(n) < 1 .

Since 1 is a regular cardinal, there exists < 1 such that, for

every n < , g(n) < . Hence f (n) < for every n < . So

sup f (n) < 1 .

n<

cf(1 ) = 1 .

72 Cardinality

The solution to Exercise 4.13 involves calculations similar to

those in our examples above.

The following theorem is an extension of Theorem 4.7. Its proof

is a more elaborate diagonal argument.

Theorem 4.33 (G. K

onig) If is an innite cardinal, then

cf() > .

Proof Let = cf (). Fix f : such that the range of f is

unbounded in . Consider an arbitrary function G : . It is

enough to see that G is not a surjection. For each < , let

A = {G()() | < f ()}.

Then, for every < ,

A

and

|A | f () < .

In particular, for every < ,

A = .

For each < , let h() be the least element of A . Then, for

every < and every < f (),

h() = G()().

Recall that, for every < , there exists < such that < f ().

Therefore, for every < ,

h = G().

This shows that G is not a surjection.

Corollary 4.34 If is an innite cardinal, then

cf (2 ) > .

Proof Apply the previous theorem with = 2 to see that

)

(2 )cf(2 > 2 .

But, if , then

(2 ) = 2 = 2

by Lemmas 4.28 and Theorem 4.23. Corollary 4.34 follows.

4.3 Conality 73

An interesting special case of Corollary 4.34 is the fact that

cf(20 ) > 0 .

Notice, also, that we recover Theorem 4.7 from Corollary 4.34

because

2 cf (2 ) > .

Exercises

Exercise 4.1 Let

< n

2= 2.

n<

2. Let

F = {{x n | n < } | x 2} .

Prove that |F| = 20 .

3. Prove that there exists a family G P() such that

|G| = 20

and for all A, B G, if A = B, then A B is nite.

Hint: Observe that F P(< 2).

Exercise 4.2 Let

< n

=

n<

and

0 <0 = < .

0

0

= 0 .

Exercise 4.3 Prove the following equations.

1. 0 0 = 20

2. 1 0 = 20

3. 0 1 = 21

4. 1 1 = 21

74 Cardinality

Exercise 4.4 Let

<

=

<

and

< = <

+)

whenever and are innite cardinals. Show that (< = 2 .

Exercise 4.5 Prove that if are innite cardinals, then

|{X | |X| = }| = .

Then explain why

|{X 2 | |X| = 0 }| = max(2 , 20 ),

|{X 2 | |X| = 1 }| = 21

and

|{X 2 | |X| = 2 }| = 22 .

Exercise 4.6 Prove that, for every ordinal , there is a cardinal

> such that

cf() =

and

= .

Hint: First recall that for every ordinal by Lemma 4.19.

Now consider the sequence of cardinals n | n < dened by

induction according to 0 = +1 and n+1 = n .

Exercise 4.7 Express the cardinality of the sets below in the

form

, 2 , 22 , ...

and explain your calculations. For your solutions, you may use

facts about Q, R, C and continuous functions from calculus courses.

1. Q = {x | x is a rational number}

2. R = {x | x is a real number}

3. R Q = {x R | x is irrational}

4. {x R | 0 < x < 1}

5. {x C | x is a root of a polynomial with rational coecients}

4.3 Conality 75

6. R R = {f | f is a function from R to R}

7. Q R = {f | f is a function from Q to R}

8. R Q = {f | f is a function from R to Q}

9. {f | f is a continuous function from R to R}

Exercise 4.8 Let <R be the usual ordering of R. For every

A R, we may abuse notation by writing

(A, <R )

when we really mean

(A, {(x, y) A A | x <R y}).

You may use facts from calculus courses in your solutions to:

1. Prove that, for every A R, if (A, <R ) is a wellordering, then

type(A, <R ) < 1 .

2. Prove that, for every < 1 , there exists A R such that

type(A , <R ) = .

Exercise 4.9 Let f : B B be a function and X B. Prove

that there exists A B such that X A, f [A] A and

|A| |X| 0 .

Exercise 4.10 If Sn | n < is a sequence of sets, then dene

f is a function with dom(f ) =

Sn = f .

and f (n) Sn for all n <

n<

1. Show that

|P( )| P(n ) .

n<

2. Show that

|P( )| P(n ) .

n<

A A n | n <

76 Cardinality

We remark that, from Exercise 4.10, it is immediate that

2 = 2n ,

n<

2 = 2n .

n<

(n )0 = max(n , (0 )0 ).

Hint: It is obvious that the left side is at least as large as the right

side. To prove the other direction, use induction on n < and the

fact that n is a regular cardinal.

Exercise 4.12 This exercise is about ordinal exponentiation.

2. Show that 1

= 1 .

3. Show that < 1 | = is uncountable.

Exercise 4.13 Prove that, for every limit ordinal ,

cf( ) = cf().

we gave of Theorem 4.13 used the Axiom of Choice because it used

the fact that every set has a cardinality. Complete the following

outline of a proof that avoids the Axiom of Choice. Let

f :AB

and

g:BA

be injections. By recursion, dene A0 = A, B0 = B,

An+1 = g[Bn ]

and

Bn+1 = f [An ].

4.3 Conality 77

Let

Aeven = (A2n A2n+1 ) ,

n<

Aodd = (A2n+1 A2n+2 )

n<

and

A = An .

n<

Dene

f (x) if x Aeven

1

h(x) = g (x) if x Aodd

f (x) if x A .

Prove that h is well-dened and h is a bijection from A to B.

alence relation on P() dened by

xEy x y is nite.

Prove that P()/E has cardinality 20 .

By denition this means that is a relation on P that is reexive

and transitive. A subset C P is called a chain i (C, ) is a

linear ordering. Since we already know that is reexive and

transitive, if C P , then C is a chain i for all x, y C, either

x y or y x. Assume that every chain has an upper bound in

(P, ). In other words, assume that, for every chain C, there exists

y P such that, for every x C, x y. Prove that (P, ) has

a maximal element. In other words, prove that there exists y P

such that, for every x P , y x.

Hint: Suppose otherwise. Let = |P |. By recursion on < ,

build a chain C = {x | < } such that C does not have an

upper bound to get a contradiction.

Remark: This proof of Zorns lemma uses the Axiom of Choice to

know that the partial ordering has a cardinality. It also turns out

that ZF together with Zorns lemma implies AC. Therefore, Zorns

lemma and AC are equivalent. This is another good exercise!

78 Cardinality

Exercise 4.17 (Boolean algebras of truth tables) For each pos-

itive n < , dene

Tn = (Tn , n , n , n , 0, 1)

as follows.

that, for all a, b 2, if a n = b n, then f (a) = f (b).

If f, g Tn and a 2, then

and

(n f )(a) = 1 f (a) = 0.

You may take it for granted that each Tn is a Boolean algebra.

It is helpful to think of Tn as the Boolean algebra of truth tables in

n variables. For example, a typical element f of T2 can be thought

of as the truth table

0 0 f (0, 0, . . . )

0 1 f (0, 1, . . . )

1 0 f (1, 0, . . . )

1 1 f (1, 1, . . . )

4.3 Conality 79

of as the simpler truth table

0 f (0, . . . )

1 f (1, . . . )

2. Find a nite Boolean algebra B such that, for every n < ,

B Tn .

3. The Boolean algebra relation for Tn is given by

f n g f n g = f.

Give a more practical description of n in terms of entries in

truth tables.

4. Figure 4.2 shows the elements of T2 organized into levels with

some arrows drawn between some truth tables. What is the

signicance of the arrows? Which arrows are missing? Copy

the gure and add all the missing arrows between truth tables

on neighboring levels.

5. List all the atoms of T1 using truth table notation. Where are

they on Figure 4.2? List all the atoms of T2 . Where are they

on Figure 4.2? How many atoms does T3 have?

6. Dene

T = (T , , , , 0, 1)

by setting

T = Tn ,

n<

= n ,

n<

= n

n<

and

= n .

n<

80 Cardinality

To make sure you understand the denition of T , convince

yourself that if f Tm and g Tn where m < n < , then

f Tn and f g = f n g.

(a) It is a fact that T is a Boolean algebra. Pick any three of

the ten dening equations for Boolean algebras and show

that they hold for T . You may use the fact that Tn is a

Boolean algebra for n < .

(b) Prove that T has no atoms.

(c) Explain why T is countable.

(d) Give a specic example of a function f : 2 that does

not belong to T .

0 0 1

0 1 1

1 0 1

1 1 1

0 0 0 0 0 1 0 0 1 0 0 1

0 1 1 0 1 0 0 1 1 0 1 1

1 0 1 1 0 1 1 0 0 1 0 1

1 1 1 l5 1 1 1 1 1 1 1 1 0

x< lllll

x ll

xx lll

xxx lllll

x lll

0 0 0 ll 0 0 0 0 0 1 0 0 0 0 0 1 0 0 1

0 1 0 0 1 1 0 1 0 0 1 1 0 1 0 0 1 1

1 0 1 1 0 0 1 0 0 1 0 1 1 0 1 1 0 0

1 1 1 iRRR 1 1 1 1 1 1 1 1 0 5 1 1 0 1 1 0

RRR x< lll

RRR x l llll

RRR xx lll

RRR

RRR xx lll

RRR xx lllll

0 0 0 0 0 0 ll 0 0 0 0 0 1

0 1 0 0 1 0 0 1 1 0 1 0

1 0 0 1 0 1 1 0 0 1 0 0

1 1 1 1 1 0 1 1 0 1 1 0

0 0 0

0 1 0

1 0 0

1 1 0

Figure 4.2 What is the signicance of this picture? See Exercise 4.17.

5

Trees

of mathematics. Most of this chapter is concerned with trees of

height at most but the last section goes into trees of height

1 . We will look at trees in various contexts: topology, analysis,

combinatorics and games.

To get started, we go over some elementary denitions and facts

about topological spaces and metric spaces.

Denition 5.1 A topological space is a pair (S, T ) such that

T P(S),

S T,

for every non-empty nite F T ,

F T,

for every F T ,

F T.

We also say that T is a topology on S.

The most important example of a topological space (S, T ) has

S=R

and

T = {U | U is an open subset of R}

5.1 Topology fundamentals 83

where U is an open subset of R i U is a union of open intervals.

For the record, open intervals of R are sets of the form

{x R | a < x < b}

where a < b are real numbers ordered in the usual way. In this

section, we will use the notation (a, b) for the open interval from

a to b even though it conicts with our notation for ordered pairs.

Other basic notation from calculus may also be used here. Some

examples of open subsets of R are

R = {(n, n) | n = 1, 2, 3, . . . },

= ,

(0, 1) = {(0, 1)},

(0, ) = {(0, n) | n = 1, 2, 3, . . . },

(, 0) = {(n, 0) | n = 1, 2, 3, . . . },

R {0} = (, 0) (0, )

and

RZ= {(n, n + 1) | n Z}.

The following fact is left as an exercise; we will give a similar proof

in the next section.

Lemma 5.2 The family of open subsets of R is a topology on R.

Topological spaces are related to metric spaces, which we dene

next.

Denition 5.3 A metric space is a pair (S, d) where

d : S S [0, ) = {x R | 0 x}

is a function from S S to the set of non-negative real numbers

such that, for all x, y, z S,

d(x, y) = d(y, x),

d(x, y) = 0 x = y

84 Trees

and

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

The most important example of a metric space (S, d) has

S=R

and

d = |x y|.

between x and y. This is the usual distance function for R. Here

are some well-known facts:

|x y| 0,

|x y| = |y x|,

|x y| = 0 x = y

and

|x z| |x y| + |y z|.

R.

{x R | |x c| < r}

take c = (b + a)/2 and r = (b a)/2 to see this. In this sense,

the topology of R comes from the metric on R. One says that the

topology and the metric are compatible when they are related in

this manner. Not every topological space has a compatible metric

but many interesting ones do.

5.2 The Baire space 85

>> x

>>

>>

>>

>>

>>

>>

>>

s

In this section, we endow with a topology and a metric, which

turn out to be compatible. Throughout this and subsequent sec-

tions, it is very important to keep in mind the distinction between

< (the set of nite sequences of natural numbers) and (the

Denition 5.5 If n < and s n , then

Ns = {x | x n = s}.

These are the basic open subsets of .

Figure 5.1 is an attempt to illustrate the basic open set Ns ,

which consists of all the innite branches x that pass through s.

The following easy observation is often useful.

Lemma 5.6 If s, t < , then

if s t, then Nt Ns ,

if t s, then Ns Nt , and

otherwise, Ns Nt = .

Denition 5.7 U is an open subset of i there is a family F

of basic open subsets of such that

U= F

86 Trees

Note that, because < is countable, U is an open subset of

i there is a sequence sn | n < from < such that

U= Ns n .

n<

.

Proof Everything is obvious except the fact that the intersection

of nitely many open sets is open. It is enough to show that the

intersection of two open sets is open. The general statement follows

by induction because we can add parentheses as follows:

U0 U1 Un = U0 (U1 Un ) .

Say

A= F

and

B= G,

where F and G are families of basic open sets. We must show that

A B is an open set. Let H be the collection of basic open sets

Nt for which there are r, s < such that

Nr F and Ns G,

r s or s r, and

t = r s.

The last two clauses say that either

r = s dom(r)

or

s = r dom(s)

(we say r and s are comparable in this case), and t is the longer

of the two (which is the union because they are comparable). We

will be done when we show that

AB = H.

5.2 The Baire space 87

First suppose that x A B. Then there are r, s < such that

x Nr F

and

x Ns G.

Then the nite sequences r and s are comparable because they

are both restrictions of the same innite sequence x. That is,

r = x dom(r)

and

s = x dom(s).

Let t = r s be the longer of the two. Then

x Nt H,

so

x H.

This shows that

AB H.

We leave the easier reverse inclusion to the reader.

Denition 5.9 C is a closed subset of i C is an open

subset of .

It is time for some examples. Consider an arbitrary x . The

singleton {x} is closed since its complement is open:

{x} = {Ns | s < but s x}.

Ns {x}

because Ns has innitely many elements while {x} has just one.

Thus {x} is closed but not open. It follows easily that {x}

is open but not closed.

Most sets are neither open nor closed. One way to see this is to

observe that

{Ns | s < } = < = 0 ,

88 Trees

and

|{C | C is a closed subset of }| = |{U | U is an open subset of }|

= <

= 0 0

= 20 ,

which is strictly smaller than

0

|P( )| = 22 .

Since there are strictly more subsets of the Baire space than there

are open or closed subsets, there must be subsets which are neither

open nor closed.

It is also easy to come up with specic examples of sets which

are neither open nor closed. Given n < , s n and x , let

s

x = s {(n + k, x(k)) | k < }.

A less precise but somehow clearer way to write this is

s

x = s(0), . . . , s(n 1), x(0), x(1), x(2), . . .

where if n = 0, then s

x = x. Suppose that U is open but not

closed and C is closed but not open. (We already gave examples

of such sets.) Let

A = {0

x | x U } {1

y | y C}.

We will prove that A is not open and leave the verication that A

is not closed to the reader. As C is not open, there exists x C

such that, for every n < ,

Nxn C.

Let

y = 1

x.

Then, for every n < ,

Nyn A.

This implies that A is not open.

Denition 5.10 Clopen means both closed and open.

5.2 The Baire space 89

If (S, T ) is a topological

space, then S T by denition and

T because = is the union of the empty family of open sets.

Thus both and S are always clopen. In the standard topology

on R, the only clopen sets are and R. In the Baire space, there

are clopen sets other than and . For example, if n < and

s n , then Ns is clopen since Ns is obviously open and

Ns = {Nt | t n but t = s}

is also open. We get more examples by noting that a union of

nitely many clopen sets is also clopen. For example,

N0 N1 = {x | x(0) = 0 or x(0) = 1}

is clopen.

Denition 5.11 For x, y , the distance between x and y is

1/2n if n is least such that x(n) = y(n)

d(x, y) =

0 if x = y.

For example,

d (0, 7, 4, 3, . . . , 0, 7, 9, 9, . . . ) = 1/22 = 1/4.

The proofs of the following lemmas are left to the reader.

Lemma 5.12 d is a metric on .

1. There exists s < such that A = Ns .

2. There exist c and a real number r > 0 such that

A = {x | d(x, c) < r}.

The previous lemma says that the topology on is compatible

with the metric d. It is worth observing that d is not the only

metric with this property. For example, if we dene e so that

1/(n + 1) if n is least such that x(n) = y(n)

e(x, y) =

0 if x = y,

then e is also a metric compatible with the Baire topology on .

It is also worth observing that, although we wrote c for center

and r for radius in Lemma 5.13, the Baire metric space is dierent

90 Trees

from the real line in that the center and radius of a basic open set

are not unique. For example,

{x | d(x, c) < r} = N =

for every c and real number r > 1. We are using the notation

to denote the empty sequence. Technically, = = 0, so we

have three names for the same thing. Another example is

{x | d(x, c) < r} = N0

for every c with c(0) = 0 and 1/2 < r 1.

Next we explain what this has to do with trees, the title of this

chapter. The following denition of tree is not the most general

but it suces for all but the last section of this chapter.

Denition 5.14 Let be a set. Then T is a tree on i

T <

and, for all m, n and s n , if s T and m < n, then

s m T.

We will focus on the cases = and = 2. As an example,

T = { , 2, 7, 2, 8, 2, 9, 7, 1, 7, 5, 7, 7,

2, 8, 1, 2, 8, 1, 1, 2, 8, 1, 5}

is a tree on . As part of checking that T is a tree, note that

2, 8, 1, 5 T and so are all of its restrictions: 2, 8, 1, 2, 8, 2

and .

Denition 5.15 If T is a tree on , then the set of innite

branches of T is

[T ] = {x | x n T for every n < }.

Example [< ] = .

Example [< 2] = 2.

Example If s < , then [{r < | r s or s r}] = Ns .

Example If x , then [{x n | n < }] = {x}.

It is easy to see that these examples of sets of the form [T ]

are closed subsets of the Baire space. This is no accident as the

following result explains.

5.2 The Baire space 91

Lemma 5.16 Let C . Then C is a closed subset of i

there is a tree T on such that C = [T ].

Proof First we prove the reverse direction. Assume T is a tree

on and C = [T ]. Let U = C. To see that C is closed we

show that U is open. For this, simply observe that

U = [T ]

= {x | there exists n < such that x n T }

= {Ns | s < T }

is a union of basic open sets.

For the forward direction of Lemma 5.16, consider an arbitrary

closed subset C of . Put

T = {x n | n < and x C}.

Clearly C [T ]. We nish by showing that [T ] C. For contra-

diction, suppose

y [T ] C.

Let U = C. Then U is open and y U . So there exists n <

such that

Nyn U.

In other words,

Nyn C = .

But, since y [T ], there exists x C such that x n = y n.

Thus

x Nyn C.

This contradiction completes the proof.

Exercises

Exercise 5.1 Let

I = {x | x is an injection from to }

and

S = {x | x is a surjection from to }.

Answer the following questions and prove your answer is correct.

92 Trees

1. Is I open?

2. Is I closed?

3. Is S open?

4. Is S closed?

T = {x n | n < and x A}.

Prove that [T ] is the closure of A in the Baire space. By this we

mean that [T ] is closed and, for every closed set C, if A C, then

[T ] C.

of

space i for every F T , if

S= F,

S= G.

i for every F T , if

S= F,

S= G.

2. The Cantor space is the topological space on 2 whose open

sets are exactly those of the form 2 U where U is an open

subset of the Baire space. You could say that the Cantor space

topology is inherited from the Baire space.

Prove that the Cantor space is compact.

Hint: Let F be a family of open subsets of 2. Assume there

is no nite G F such that

G = 2.

5.2 The Baire space 93

Prove that there is an x 2 such that

x F.

sion, maintain that there is no nite G F such that

G Nxn 2.

D is said to be dense i for every non-empty U T ,

D U = .

A topological (S, T ) space is said to be separable i it has a count-

able dense subset. Show that the Baire space is separable.

Exercise 5.6 Let D be the set of x such that, for every

m < , there exists n < such that m < n and x(n) = 0.

2. Prove that D is not open.

3. Prove that D is not closed.

4. Find a sequence Un | n < of subsets of such that

D= Un

n<

Exercise 5.7 We need three denitions before stating the exer-

cise. Consider an arbitrary metric space (S, d).

We say that xi | i < converges to y and write

lim xi = y

i

for every j < , if j > i, then d(xj , y) < r.

We call xi | i < a Cauchy sequence i for every r R, if

r > 0, then there exists i < such that, for all j, k < , if

j, k > i, then d(xj , xk ) < r.

94 Trees

We say that (S, d) is complete i for every Cauchy sequence

xi | i < from S, there is y S such that

lim xi = y.

i

The following exercises are about the Baire space with the metric

dened by

d(x, y) = 1/2n (x n = y n but x(n) = y(n))

and

d(x, y) = 0 x = y

but we remark that parts 1 and 2 hold in every metric space.

1. Let C . Prove that C is closed i for every sequence

xi | i <

from C and every y S, if

lim xi = y,

i

then y C. This says that a set is closed i it has all its limit

points.

2. Prove that if limi xi = y, then xi | i < is a Cauchy

sequence.

3. Prove that the Baire space is complete.

sequence of subsets of . Assume that, for every n < , Dn is

both open and dense in the Baire space. Let

E= Dn .

n<

there are s, t T such that r s, r t, s t and t s. Prove

that if T is a non-empty perfect tree on , then T has 20 many

branches, that is,

|[T ]| = 20 .

5.2 The Baire space 95

Exercise 5.10 Let Dn | n < be a sequence of subsets of .

Assume that, for every n < , Dn is both open and dense in the

Baire space. Let

E= Dn .

n<

Prove that

|E| = 20 .

Hint: By Exercise 5.9, it is enough to show that there is a perfect

tree T such that [T ] E. Construct T using ideas similar to the

solution to Exercise 5.8.

subset of the Baire space and T be a tree on such that C = [T ].

The CantorBendixon derivative of T is dened to be

T = {s T | Ns [T ] has at least two elements}.

By recursion, dene

T 0 = T,

T +1 = (T )

whenever is an ordinal, and

T = T .

<

1. By induction on all ordinals , prove that T is a tree on

and, for every < ,

T T .

2. Prove that there exists < 1 such that T +1 = T .

3. Let be least such that

T +1 = T .

(This is the CantorBendixon rank of T .)

(a) Prove that T is a perfect tree and [T ] C.

(b) Prove that if T = , then |C| 0 .

96 Trees

Notice that the combination of Exercises 5.9 and 5.11 shows that

closed subsets of the Baire space are either countable or have

cardinality 20 .

is a tree T on whose CantorBendixon rank is and (T ) = .

Hint: Obviously, T0 = and T1 = < {0} work. Next dene T2 and

T3 . Once you see a pattern for natural numbers, try T . Then you

will be on your way to constructing T by recursion.

for every non-empty perfect tree T , neither [T ] A nor [T ]

A. Hint: Let

T | < 20

x | < 20

and

y | < 20

y [T ] {x | < }

and

x [T ] {y | }.

Remark: Exercise 5.13 is another example of a diagonal argu-

ment. Intuitively, we diagonalize over all non-empty perfect trees

to make sure none of them work.

The rst result of this section is that if T is a tree on with

innite height and nite levels, then T has an innite branch.

5.3 Illfounded and wellfounded trees 97

Theorem 5.17 (D. K

onig) Let T be a tree on . Assume that,

for every n < ,

T n =

and

|T n | < 0 .

Then

[T ] = .

Corollary 5.18 Let T be a tree on 2. Assume that, for every

n < ,

T n 2 = .

Then

[T ] = .

Corollary 5.18 is an immediate consequence of Theorem 5.17,

which we will prove after some discussion and an example. We

already used the words level and height informally. Now let us

ocially dene them.

Denition 5.19 Let T be a tree on . Then, for every n < ,

leveln (T ) = T n

and

height(T ) = {n < | leveln (T ) = }.

Notice that if T is a tree on , then height(T ) is an ordinal and

height(T ) .

This is because trees are closed downward: if s leveln (T ) and

m < n, then s m levelm (T ). As height(T ) is a transitive set of

natural numbers, it is itself an ordinal .

In Theorem 5.17, we cannot drop the hypothesis that all levels

of T are nite. Consider the tree depicted in Figure 5.2. It consists

of all restrictions sn m of sequences

sn = n, 0, . . . , 0

where, in the displayed sequence, n < and there are n zeros.

Then T has innite height but no innite branch.

98 Trees

.

..

3, 0, 0, 0

2, 0, 0 3, 0, 0

DD

DD vvv jjjjjjj

v j

DD vv jjjj

DD vvvjjjjjj

jvjj

r < , let

Tr = {s T | r s or s r}.

Notice that Tr T , Tr is a tree on and Tr has nite levels.

Dene a function x : by recursion as follows. Assume

that x n has been dened so that x n T and Txn has innite

height. Since Txn has nite levels, we can write

Txn = Ts i

i<j

si n+1 Txn .

From the equation above and the fact that Txn has innite height,

it follows that there is at least one i < j such that Tsi has innite

height. Dene x (n + 1) = si for the least such i. In other words,

put x(n) = si (n) for this i. This completes the denition of x.

Clearly, x [T ], which proves the theorem.

5.3 Illfounded and wellfounded trees 99

Recall that the Cantor space is compact by Exercise 5.4. It is

possible to derive Corollary 5.18 directly from the fact that the

Cantor space is compact. This alternative proof uses Lemma 5.16,

which, in the case of the Cantor space, tells us that C is closed

subset of 2 i there is a tree T on 2 such that C = [T ]. Consider

this a hint for Exercise 5.15.

Theorem 5.17 gives conditions that imply a tree has innite

branches. Now we want to understand when a tree does not have

innite branches. It may help to picture trees growing downward

instead of upward for this discussion. Not having innite branches

makes a tree wellfounded according to the following denition.

[T ] = . Otherwise, T is illfounded. We call s a terminal node of

T i s T and there is no t T with t s.

downward because if x [T ], then

x 2 x 1 x 0.

Theorem 5.17 says that an innite tree with nite levels is ill-

founded. We will characterize wellfounded trees in terms of rank

functions.

function f with domain T such that, for all s, t T ,

if s t, then f (s) > f (t).

then T is wellfounded.

founded tree on , then there is a natural way to dene a rank

function for T , which is what the following theorem explains.

100 Trees

Theorem 5.23 Let T be a wellfounded tree on . Then there is

a unique function

T : T 1

if s is not a terminal node of T , then

T (s) = sup ({T (t) + 1 | t s}) .

determining T in Theorem 5.23 look like a recursive denition

but it is not immediately clear which wellordering underlies the

recursion. This is sorted out in Exercises 5.18 and 5.19.

T has a rank function.

we let the rank of T be

rank(T ) = T ( ).

most . But only wellfounded trees have ranks, and these ranks

are sometimes strictly greater than .

{s | dom(s) n and s(m) = 1 for every m < dom(s)}

and

both have rank n.

rank . Figure 5.3 shows the same tree with the nodes labeled

according to their rank values.

5.3 Illfounded and wellfounded trees 101

.

..

0 1

0 1 2

0? 1 2 o3

??? ooooo

?? o

?? ooo

oooo

U = {0

s | s T },

then

rank(U ) = + 1.

See Figure 5.4 for a picture of U with its nodes labeled according

to their rank values. We should explain the notation we are using

in the denition of U . For r m and s n , let

r

s = r(0), . . . , r(m 1), s(0), . . . , s(n 1) m+n .

Put another way,

r(i) if i < m

(r

s) (i) =

s(i m) if m i < m + n.

The examples above beg the question: which ordinals are the

ranks of trees on ? The answer is exactly the countable ordinals

by Theorem 5.23 and Exercise 5.20.

102 Trees

.

..

0 1

0 1 2

0 EE 1 2 m3

EE yyy mmmmm

EE y m

EE yy mm

E yymymmmm

m

+1

Exercises

Exercise 5.14 Let T be a tree on . Assume that [T ] = . Prove

that there exists a unique x [T ] such that, for all y [T ] and

n < , if y n = x n, then x(n) y(n). We call x the left-most

branch of T .

Exercise 5.15 You should notice that the proof of Theorem 5.17

is similar to the solution to Exercise 5.4(2). This exercise explains

why.

1. Use Lemma 5.16 and Corollary 5.18 to derive the fact that the

Cantor space is compact.

2. Use Lemma 5.16 and the fact that the Cantor space is compact

to derive Corollary 5.18.

5.3 Illfounded and wellfounded trees 103

Exercise 5.16 Recall that clopen means closed and open.

1. Prove that, for every C 2, if C is a clopen subset of the

Cantor space, then C is a union of nitely many basic open

subsets of the Cantor space. In other words, there is a nite set

{s0 , . . . , sn1 } <

such that

C= Nsi 2.

i<n

the Baire space but neither C nor C is a nite union of

basic open sets. Explain why your example has this property.

3. Prove that the only two clopen subsets of R are and R.

Exercise 5.17 Let T be the wellfounded tree consisting of de-

scending sequences of natural numbers. In other words,

T = {s < | for all m, n dom(s), if m < n, then s(m) > s(n)}.

Calculate rank(T ).

on < by declaring that, for all s, t < ,

t <KB s

i either t s or there exists n < such that t n = s n but

t(n) < s(n).

1. Prove that <KB is a strict linear ordering of < .

2. Let T be a tree on .

(a) Prove that if the restriction of <KB to T is a wellordering,

then T is wellfounded.

(b) Prove that if T is wellfounded, then the restriction of <KB

to T is a wellordering.

Exercise 5.19 Prove Theorem 5.23 in the following two steps.

1. Explain why the properties of T listed in the statement of

Theorem 5.23 form a legitimate denition by recursion on the

restriction of <KB to T .

2. Explain why the range of T is a countable ordinal.

104 Trees

Exercise 5.20 Prove by induction on < 1 that there exists

a wellfounded tree T on with rank(T ) = .

Exercise 5.21 Let T be a non-empty wellfounded tree on 2.

Prove that rank(T ) < .

of the Cantor space. That is,

B = {X | X is a clopen subset of 2},

X B Y = X Y,

X B Y = X Y,

B X = 2 X,

B =

and

B = 2.

Prove B is a countable atomless Boolean algebra.

Exercise 5.23 Let B be the Boolean algebra of clopen subsets

of the Baire space. That is,

B = {X | X is a clopen subset of },

X B Y = X Y,

X B Y = X Y,

B X = X,

B =

and

B = .

Prove B is an atomless Boolean algebra of cardinality 20 .

5.4 Innite games 105

5.4 Innite games

Let A . We describe a game, which is called GA . The game

has two players, I and II, who take turns playing natural numbers

x0 , x1 , etc. A run of the game GA looks as follows.

I x0 x2 x4

II x1 x3 x5

If x = xn | n < is a run of GA , then player I wins the run i

x A. Otherwise, player II wins the run.

This is a very general sort of game. Notice that nite games also

t this scheme because we may ignore moves after a winner has

been declared. (There are really two kinds of nite length games.

Either the length is xed in advance or else the length depends

on exactly how the players move. Both kinds of nite games can

be modeled with our innite games.) One dierence between our

games and some familiar games like chess is that we do not allow a

run of the game to end in a draw. This is because either x A or

x A. An arbitrary way to get around this objection is to declare

that draws go to player II. See Exercise 5.24 for more about chess.

Many properties in mathematics can be expressed in terms of

games, so general theorems about games can be quite useful. Along

these lines, a series of exercises at the end of this section illustrate

one of many ways in which games and mathematical analysis are

related.

Naturally, we are more interested in winning strategies than we

are in the player who wins a particular run of a particular game.

So let us continue making denitions associated to the game GA .

A strategy is a function : < . If is a strategy and

b , then b is the run that results if I uses and II plays b.

Formally, b is dened by recursion according to the equations

( b)2n+1 = bn

and

( b)2n = (( b) 2n).

See Figure 5.5 for another way of depicting the run b. We call

a winning strategy for player I i for every b ,

b A.

I ( ) (( ), b0 ) (( ), b0 , (( ), b0 ), b1 )

II b0 b1 b2

I a0 a1 a2 a3

II (a0 ) (a0 , (a0 ), a1 ) (a0 , (a0 ), a1 , (a0 , (a0 ), a1 ), a2 )

108 Trees

Similarly, if is a strategy and a , then a is the run

that results if II uses and I plays a. Formally, a is dened by

recursion according to the equations

(a )2n = an

and

(a )2n+1 = ((a ) (2n + 1)).

Figure 5.6 depicts the run a another way. We call a winning

strategy for player II i for every a ,

a A.

We say that A is determined i either player I has a winning

strategy or player II has a winning strategy. Otherwise, A is un-

determined. Obviously, it is not possible for both players to have

winning strategies because otherwise we could play the two strate-

gies against each other to get a contradiction.

Example Let A be the set of surjections from to . Let be

the strategy such that

(xi | i < 2n) = n

and

(xi | i < 2n + 1) = 0.

Then, for every b ,

b A.

Thus is a winning strategy for player I in GA . So A is determined.

Observe that is not the only winning strategy for player I in GA ;

there are innitely many others.

This section includes two theorems about the determinacy of

games. The rst, Theorem 5.26, says that some games are unde-

termined. It will be apparent from our proof that there are 20

many strategies and 22 0 many games. In particular, there are

strictly more games than strategies. By itself, this is not an argu-

ment that there are undetermined games because some strategies

win in more than one game. For example, if is a winning strategy

for player I in GA and A B, then is also a winning strategy

for player I in GB . So more work than mere counting is needed to

see that there are undetermined games.

5.4 Innite games 109

Theorem 5.26 (GaleStewart) There is an undetermined subset

of .

Proof First we claim that if is a strategy, then the two functions

b b

and

a a

are injections from to itself. This claim is clear because b is

the sequence of odd values of b and a is the sequence of even

values of a .

Next observe that

| | = 0 0 = 20

and

< <

( 0 )

|{ | is a strategy}| = ( ) = 0 0 = 0 0 = 20 .

Say

{ | is a strategy} = | < 20 .

that a and b have been selected for < . Let

A = {a | < }

and

B = { b | < }.

Notice that

|A | < 20 .

Because

b b

is an injection, we can pick b such that

b A .

This determines

B+1 = { b | }.

110 Trees

Notice that

|B+1 | + 1 < 20 .

Because

a a

is an injection, we can pick a such that

a B+1 .

That completes the denition of a | < 20 and b | < 20 .

Let

A = {a | < 20 }

and

B = { b | < 20 }.

From the recursive denition, it is clear that

A B = .

Let be an arbitrary strategy. Say = . Then is not a winning

strategy for player II in GA because

a A.

And is not a winning strategy for player I in GA because

b B.

Therefore, A is undetermined.

It is worth noting how the Axiom of Choice was used in the

previous proof. Without it, we would not necessarily be able to

index all the strategies with ordinals at the start. The proof was

yet another example of a diagonal argument. Intuitively, we diag-

onalized over all strategies to make sure that none of them work,

handling the th strategy at stage .

We have seen that some games are undetermined. But many

of the games that come up in practice, open games for example,

turn out to be determined. This is more important because it has

mathematical applications.

Theorem 5.27 (GaleStewart) Every open subset of is de-

termined.

5.4 Innite games 111

Proof Let U be an open subset of . Suppose that player I does

not have a winning strategy in GU . We must show that player II

has a winning strategy in GU . Let

C = U.

Then C is closed, so by Lemma 5.16, there is a tree T on such

that C = [T ]. For s < , let

Us = {x | s

x U }.

Notice that

U = U.

Let W be the set

{s < | dom(s) is even and I has a winning strategy in GUs }.

We refer to the elements of W as winning positions for player I in

GU . The following conditions are obviously true.

1. W .

2. Let s < such that dom(s) is even but s W . Then:

(a) For every k < , there is < such that s

k, W .

(b) There exists x C such that x dom(s) = s.

From conditions (1) and (2a) we can read o a certain strategy

: <

such that, for all a and n < ,

(a ) 2n W.

Just to be specic, given s < such that dom(s) is even but

s W , for every k < , dene

(s

k) = the least < such that s

k, W .

The other values of are irrelevant, so make them zero. Some-

times this is called a non-losing strategy for player II because it

avoids winning positions for player I. We claim that is a winning

strategy for player II in GU . To see this, let a and y = a .

We must show that y U . By condition (2b), for every n < ,

there exists xn C such that

xn 2n = y 2n.

112 Trees

This can be expressed by the inequality

d(xn , y) 1/22n .

Hence

lim xn = y.

n

Since C is closed,

y C = U.

A more direct way to argue this last part is to note that, for every

n < ,

y 2n = xn 2n T

hence

y [T ] = C = U.

Exercises

Exercise 5.24 Use Theorem 5.27 to explain why, in chess, either

White has a winning strategy, or Black has a strategy to avoid

losing.

that B is determined. Hint: By Theorem 5.26, there is an

undetermined set A. Do not worry about how A was constructed.

Rather, take A as given and dene B from A in a way that takes

advantage of the asymmetry that player I goes rst.

is determined. Hint: One approach is to model the proof on that

of Theorem 5.27. An easier approach is to use the statement of

Theorem 5.27 and derive closed determinacy as a corollary. But

the proof is not completely trivial because, by Exercise 5.25, there

are determined sets whose complements are not determined.

5.4 Innite games 113

For the following series of exercises, we dene another kind of

game, called a perfect set game for reasons that will become ap-

parent. For A 2, let GA be the game whose runs have the

following pattern.

I s0 s1 s2

II n0 n1 n2

At stage 2i, player I must play si < 2 or else he loses. At stage

2i + 1, player II must play ni < 2 or else he loses. At the end of

the run, player I wins if

s0

n0

s1

n1

s2

n2 A.

Otherwise, player II wins the run.

Exercise 5.27 Provide formal denitions of the following ter-

minology.

1. is a strategy for player I in GA .

2. is a winning strategy for player I in GA .

3. is a strategy for player II in GA .

4. is a winning strategy for player II in GA .

Exercise 5.28 Let A 2. Suppose that there exists a perfect

tree T on 2 such that [T ] A. Prove that player I has a winning

strategy in GA .

Exercise 5.29 Let A 2. Suppose that player I has a winning

strategy in GA . Prove that there exists a perfect tree T on 2 such

that [T ] A.

Exercise 5.30 Let A 2. Suppose that A is countable. Prove

that player II has a winning strategy in GA .

Exercise 5.31 This exercise is harder than the previous three

but it is the most important. Let A 2. Suppose that player

II has a winning strategy in GA . Prove that A is countable by

completing the following outline.

Let be a winning strategy for player II in GA . Suppose that

p is a position of even length 2j. If 2j = 0, then

p = ,

whereas if 2j > 0, then we may specify that

p = s0 , n0 , . . . , sj1 , nj1 .

114 Trees

Saying that p has even length is the same as saying that, starting

from p, it is player Is turn to move. Assume, in addition, that p

is consistent with , by which we mean that

n0 = (s0 ),

n1 = (s0 , n0 , s1 ),

n1 = (s0 , n0 , s1 , n1 , s2 ),

and so on for every ni with i < j. Now dene

p = s0

n0

. . .

sj1

nj1 .

Notice that p < 2 and the domain of p is its nite cardinality

|p | = j + |si |.

i<j

p x and

if q is a position such that

q extends p,

q has even length (so it is player Is turn to move after q),

and

q is consistent with ,

then q x.

1. Prove that each p as above rejects exactly one x 2. Hint:

Dene x(m) by recursion on m < . Start by setting x(m) =

p (m) for all m < |p |. Now suppose that m |p |. There is a

unique s < 2 such that

x m = p

s.

Let

n = (p

s).

If n = 0, then put x(m) = 1. Otherwise n = 1, in which case

put x(m) = 0. Show that p rejects x and, if p rejects y, then

y = x.

2. Prove that if x A, then there exists p as above that rejects x.

Hint: Suppose otherwise and contradict the assumption that

is a winning strategy for player II.

5.5 Ramsey theory 115

3. Use the previous results 1 and 2 to conclude that A is countable.

Hint: Count positions.

Exercise 5.32 Let C 2 and assume that C is closed. Sketch

a proof that GC is determined. In other words, prove that either

player I has a winning strategy in GC , or else player II does.

Hint: All of the ideas are contained in Theorem 5.27 and Exercise

5.26 but writing up a complete proof is challenging because the

notation is complicated.

Exercise 5.33 Use Exercises 5.29, 5.31 and 5.32 to prove that

if C is a closed subset of 2, then either C is countable or C has

a perfect subset. Notice this also follows from the Cantor perfect

set theorem, which was the subject of Exercise 5.11, but the two

proofs are very dierent.

Ramsey theory is often introduced with the following scenario.

Imagine that you are hosting a party and you would like to invite

enough people so that either there is a trio of guests who have

met the other two in the trio, or there is a trio of guests who

have met neither of the other two in the trio. This can be modeled

mathematically as follows. Let I represent the set of invited guests

and

[I]2 = {{a, b} | a, b I and a = b}

be the set of pairs of guests. Let the function

F : [I]2 2

be given by

0 a and b have not met each other

F ({a, b}) =

1 a and b have met each other.

Such a function is referred to as a coloring of pairs from I by

two colors. The two colors are 0 and 1. So 2 = {0, 1} is the set of

colors. A subset H I is called homogeneous for F i there exists

k 2 such that, for all a, b H, if a = b, then

F ({a, b}) = k.

116 Trees

Your goal as host is to choose I large enough so that, for every F

as above, there is an H as above with |H| = 3. By the following

two exercises, you should invite at least six guests.

Exercise 5.34 Prove that for every function

F : [6]2 2,

there exists H 6 and k 2 such that |H| = 3 and, for all

a, b H, if a = b, then

F ({a, b}) = k.

Exercise 5.35 Find a function

F : [5]2 2

such that, for every H 5 and k 2, if |H| = 3, then there exists

a, b H such that a = b and

F ({a, b}) = k.

The phenomenon described above has many important exten-

sions. Given m < and a set I, dene

[I]m = {p I | |p| = m}.

The nite Ramsey theorem says that, for all positive , m, n < ,

there exists r < such that, given

a set I such that |I| r, and

a function F : [I]m ,

there exist

a subset H I such that |H| n, and

a number k <

with the property that, for every p [H]m ,

F (p) = k.

Exercises 5.34 and 5.35 show that, with m = 2 (colorings of pairs),

= 2 (two colors) and n = 3 (homogeneous set with three ele-

ments), the least witness to the nite Ramsey theorem is r = 6.

The nite Ramsey theorem is typically proved in a basic course

on discrete mathematics; we will not prove it here. The follow-

ing innite Ramsey theorem and results like it are of signicant

5.5 Ramsey theory 117

importance in set theory and its applications. The way we have

organized the proof explains why it sits in our chapter on trees.

Theorem 5.28 (Ramsey) Let 0 < < and F be a function of

the form

F : []2 .

Then there exists k < and an innite H such that, for every

p [H]2 ,

F (p) = k.

Proof Let T be the set of strictly increasing s < such that,

for every m < dom(s), there exists k < such that, for every

n < dom(s), if m < n, then

F ({s(m), s(n)}) = k.

Then T is a tree on . In this context, strictly increasing means

that, for all m < n < dom(s),

s(m) < s(n).

We will prove that T has an innite branch. But rst let us show

why the existence of such a branch suces to prove the theorem.

Suppose that x is an innite branch of T . For each m < , let km

be the unique k < such that, for every n < , if m < n, then

F ({x(m), x(n)}) = k.

Since {km | m < } is nite, there exists an innite S

and k < such that, for every m S,

km = k.

Let

H = {x(m) | m S}.

Since S is innite and x is increasing, H is innite. Moreover, for

every a, b H,

F ({a, b}) = k.

Thus H witnesses the conclusion of the theorem.

Now we construct an innite branch x through T . By recursion

118 Trees

on n < , we dene x(n) and, simultaneously, kn < and an

innite In . Start by dening

I0 =

and

x(0) = 0.

Now assume we are given In and it is an innite subset of . Let

x(n) = min(In ).

This is consistent with our having set x(0) = 0. For each k < ,

let

Jk = {i In | i > x(n) and F ({x(n), i}) = k}.

Then

In {x(n)} = J0 J 1 .

Since In {x(n)} is innite, there exists k < such that Jk is

innite. Let kn be the least such k and

In+1 = Jkn .

That completes the recursive construction. By induction on n < ,

it is obvious that

x(n) = min(In )

and, for every m < n,

In Im+1 Im {x(m)},

and

F ({x(m), x(n)}) = km .

Therefore, x is an innite branch through T .

Exercise 5.36 Let 0 < m < . Show that Theorem 5.28 remains

true if 2 is replaced by m in its statement. Hint: Use induction

on m. The case m = 1 follows from the pigeonhole principle. (If

you partition an innite set into nitely many pieces, then one of

the pieces must be innite.) Think of the proof of Theorem 5.28 as

showing that the case m = 1 implies the case m = 2. Generalize

this to see that case m implies case m + 1.

5.6 Trees of uncountable height 119

Exercise 5.37 Prove that Theorem 5.28 becomes false if

is replaced by 1 and innite is replaced by uncountable

using the following example. Fix an injection

g : 1 R.

In order to avoid possible confusion, we write <R for the usual

order of R here. Dene

F : [1 ]2 2

by

0 if < and g() >R g()

F ({, }) =

1 if < and g() <R g().

F . In other words, prove that if H is an uncountable subset of 1

and k < 2, then there exists p [H]2 such that F (p) = k. Hint:

Argue by contradiction and use the fact that between any two real

numbers there is a rational number.

So far, all the trees we have looked at have at most many levels.

This section is on trees of height 1 . We might expect that our

results about trees of height would lift to theorems about trees

of height 1 . For example, recall that Theorem 5.17 says that if T

is a subtree of < and, for every n < ,

0 < |T n | < 0 ,

then T has an -branch, i.e., there exists b : such that, for

every n < ,

b n T.

Does this statement remain true if we replace by 1 and 0 by

1 ? It turns out that the answer is no; counterexamples are called

Aronszajn trees and the point of this section is to construct one.

This is a big topic; in order to keep this section manageable, we

still do not give the most general denition of tree.

120 Trees

Denition 5.29 A subtree of <1 is a subset T < 1 such

that, for all s T and < dom(s),

s T.

An 1 -branch of T is a function b : 1 such that, for every

< 1 ,

b T.

such that

0 < |T | < 1

for every < 1 but T has no 1 -branch.

Proof Let

I = {s <1 | s is an injection}.

The good news is that I is obviously a subtree of <1 and I has

no 1 -branch because there is no injection from 1 to . The bad

news is that, whenever < 1 ,

|I | = |{s | s is an injection}| = 20 1 ,

which is too large. We will nd an appropriate subtree T I. For

< 1 and s, t , dene

s t |{ < | s() = t()}| < 0 .

It is easy to see that is an equivalence relation on .

Claim 5.30.1 There is a sequence s | < 1 so that, for

every < 1 ,

s I

and, for every < ,

s s .

Assuming Claim 5.30.1, if we dene

T = {t I | t s }

< 1

observe that T is a subtree of <1 . The point is that T is closed

5.6 Trees of uncountable height 121

downward under restriction. This is true because if < < 1

and

t T ,

then

t s ,

hence

t s s ,

so

t T .

Second, observe that the levels of T are countable. This is because,

for every < 1 and

t T ,

there are n < and 0 < < n1 < such that t() = s ()

for every < except if = m for some m < n. Since there are

at most countably many ways to make these sorts of nite changes

to s , we have that, for every < 1 ,

1 |T | 0 .

It remains to prove Claim 5.30.1. We dene s by recursion on

< 1 . In addition to maintaining the two requirements of the

claim, we also maintain the extra property that, for every < 1 ,

| ran(s )| = 0 .

This is so that the recursion does not run out of steam, as you will

see. We have no choice but to set s0 = . If s has been dened,

then pick k ran(s ) and dene

s+1 = s {(, k)}.

Now suppose that < 1 is a limit ordinal and we have

s | <

satisfying the three requirements. The denition of s is somewhat

tricky in this case. Since cf() = , there are ordinals

0 < 1 < i < < = sup i .

i<

122 Trees

By recursion on i < , dene a sequence ti | i < from I such

that, for every j < ,

tj j sj

and, for every i < j,

ti = tj i .

Start with

t0 = s0 .

Suppose we are given ti with

ti i si .

Let us say that a pair (, ) is bad i

< i < i+1

and

ti () = si + 1 ( ).

The reason we call them bad is because their existence prevents

us from setting

ti () if < i

ti+1 () =

si + 1 () if i+1 i .

Remember that ti+1 is supposed to be an injection! However, we

claim that there are only nitely many bad pairs. First note that

if (, ) is a bad pair, then

ti () = si + 1 ().

This is because si + 1 is an injection, so it cannot take on the same

value, ti (), twice. From this and the fact that

ti i si + 1 i ,

it follows that there are only nitely many rst coordinates of bad

pairs. Finally, observe that if (, ) and (, ) are bad pairs,

then

si + 1 ( ) = ti () = si + 1 ( ),

so = , again because si + 1 is an injection. This shows that

there are only nitely many bad pairs. Say the number of bad

5.6 Trees of uncountable height 123

) for m < n.

pairs is n where n < . List the bad pairs as (m , m

Then pick distinct

km (ran(ti ) ran(si + 1 ))

for m < n. This is possible by the extra property on si and si + 1 ,

and the fact that ti i si i si + 1 i . Our solution to the

problem of bad pairs is to dene

ti () if < i

for every m < n

ti+1 () = si + 1 () if i+1 i and = m

.

km if = m

Obviously, this satisies our requirements for ti+1 . Now, having

completed the denition of ti | i < , we put

t= ti .

i<

The good news is that t meets the rst and second requirements

for s of Claim 5.30.1, namely that

t I

and, for every < ,

s t .

The problem is that it might not satisfy the extra property that

| ran(t)| = 0 .

To x up this potential problem, put

t(2i ) if = i for some i <

s () =

t() if {i | i < }.

Then all three requirements on s are met.

Exercises

Exercise 5.38 Let T be the tree constructed in the proof of

Theorem 5.30. Suppose that < < 1 and u T . Explain

why there exists v T such that u = v . We say that u

has extensions to every level of T .

124 Trees

Exercise 5.39 Let T be the tree constructed in the proof of

Theorem 5.30. Suppose that < 1 and u T . Explain why

there exist > and v, w T such that

u=v=w

but

v = w.

We say that u splits in T .

Theorem 5.30. Find a sequence u | < 1 of members of T

such that, for all < < 1 ,

u u

and

u u .

In this case, we say that u | < 1 is an antichain of T

because u and u are incomparable (cannot be compared using

) whenever = .

Exercise 5.41 Let T be a subtree of < 1 with an 1 -branch

b : 1 .

Assume that every u T splits in T . Prove that T has an un-

countable antichain.

6

Dense linear orderings

This chapter is mainly about two theorems, one due to Cantor, the

other to Dedekind, which are characterizations of the rationals, Q,

and the reals, R, in terms of their respective orderings.

Recall that (A, ) is a strict linear ordering i is a transitive,

irreexive and total relation on A. As usual, we write x y for

x y or x = y.

Denition 6.1 (A, ) is a dense linear ordering i (A, ) is a

strict linear order with at least two elements and, for all x, y A,

if x y, then there exists z A such that x z y.

We required A to have at least two elements because, otherwise,

(0, <) and (1, <) would be dense linear orderings, which would be

counterintuitive. It follows easily from Denition 6.1 that every

dense linear ordering is innite.

Denition 6.2 If (A, ) is a strict linear ordering and L, R A,

then we say:

L is a left endpoint of (A, ) i L x for every x A.

R is a right endpoint of (A, ) i x R for every x A.

In this chapter we are mainly interested in dense linear orderings

without endpoints. Here are several examples, each referring to the

usual ordering of real numbers.

Q = {m/n | m, n Z and n = 0}

126 Dense linear orderings

R = the set of real numbers

the open interval (0, 1)

(0, 1) Q

The reader has not yet seen denitions of Q and R in this book

but has intuition about these based on doing mathematics since

childhood. Temporarily, we rely only on that intuition.

Continuing with our introduction, we make the following key

denition, which is really just a repetition of Denition 3.27.

Denition 6.3 We say that f is an isomorphism from (A, A )

to (B, B ) and write

f : (A, A ) (B, B )

i f is a bijection from A to B and, for all x, y A,

x A y f (x) B f (y).

We say that (A, A ) is isomorphic to (B, B ) and write

(A, A ) (B, B )

i there is an isomorphism f : (A, A ) (B, B ).

For example, with the usual ordering on the real line, the open

interval (/2, /2) is isomorphic to R as witnessed by the func-

tion

x tan(x).

So,

(/2, /2) R.

It also turns out that

(/2, /2) Q Q

but since arctan(1) = /4 Q, a function other than tangent is

needed. We will see why there is such an isomorphism later.

As an example of non-isomorphism, observe that Q R because

|Q| = 0 = 20 = |R|,

so there is not even a bijection between Q and R. A more subtle

example is the fact that

R

Q R.

6.1 Denitions and examples 127

(f (z ), )

o f (z) /

aCC

CC

CC

CC

o / _o _ _ _ _ _ _ _ z _ _ _ _ _ _ _ _ _/

(0 , ) Q

Figure 6.1 R

Q R

Here, R

Q is the concatenation of R followed by Q. It is easy

to see that R

Q is a dense linear ordering without endpoints.

Suppose for contradiction that

f : R

Q R.

Q.1 Then, f (z) R

and

(0, ) Q (f (z), ).

The set on the left has cardinality 0 and the set on the right has

cardinality 20 , so there is no bijection between them. Figure 6.1

illustrates the point.

This brings up an interesting question: what about Q versus

Q

Q? In the next section, we will state and prove Theorem 6.5,

which implies that

Q Q

Q.

(/2, /2) Q Q,

Q is

({0} R) ({1} Q) and z is the ordered pair (1, 0).

128 Dense linear orderings

6.2 Rational numbers

This section has two theorems, the rst of which should come as

no surprise.

Theorem 6.4 There is a countable dense linear ordering without

endpoints.

Exercise 6.1 outlines a proof of Theorem 6.4 based only on what

we already know about natural numbers, their ordering and their

arithmetic. Obviously, (, <) is not dense and has a left endpoint,

so some work is needed. It would be tempting to blurt out that

Q with its ordering already witnesses Theorem 6.4 but this would

be cheating since the point here is to say what Q is up to isomor-

phism.

The next theorem tells us that in a certain sense it does not

matter which countable dense linear ordering without endpoints

we work with since they are all the same up to isomorphism. It

says we can take our pick and make it Q. The technique used

in the proof, a back-and-forth construction, is also important in

other parts of mathematics. An application of this technique to

the theory of Boolean algebras is the topic of Exercise 6.6.

Theorem 6.5 (Cantor) Let (A, A ) and (B, B ) be countable

dense linear orderings without endpoints. Then

(A, A ) (B, B ).

Proof Say A = {a0 , a1 , . . . } and B = {b0 , b1 , . . . }. Warning! It is

denitely not true that i < j ai A aj because (, <) is a

wellordering whereas (A, A ) is an illfounded relation.

By recursion on n < , we will dene nite bijections

fn : dom(fn ) ran(fn )

such that

{ai | i < n} dom(fn ) A

and

{bi | i < n} ran(fn ) B

and

fn : (dom(fn ), A ) (ran(fn ), B ).

6.2 Rational numbers 129

We also maintain that, for all m < n < ,

fm fn .

In other words,

fn dom(fm ) = fm .

Mere success in this construction is enough because if

f= fn ,

n<

f : (A, A ) (B, B ).

For example, to verify that f dened this way is order preserving,

note that if

i, j < n < ,

then

ai A aj fn (ai ) B fn (aj ) f (ai ) B f (aj )

because ai , aj dom(fn ), fn is order preserving and

fn = f dom(fn ).

Here is the recursive denition of fn . For the base step, set

f0 = . Now assume that fn has been dened. The denition of

fn+1 is a two-step process: back and forth.

Forth step We dene a nite isomorphism

g : (dom(g), A ) (ran(g), B ).

with

dom(g) = fn {an }.

Case 1 an dom(fn ).

Set g = fn .

Case 2 an dom(fn ).

We claim that there exists i < such that if

g = fn {(an , bi )},

130 Dense linear orderings

then

g : (dom(g), A ) (ran(g), B ).

The proof of the claim breaks up into three subcases.

Subcase 1 an A x for every x dom(fn ).

Since (B, B ) does not have a left endpoint, we may pick i <

such that bi B y for every y ran(fn ).

Subcase 2 x A an for every x dom(fn ).

Since (B, B ) does not have a right endpoint, we may pick i <

such that y B bi for every y ran(fn ).

Subcase 3 Otherwise.

Since dom(fn ) is nite, we can pick , r dom(fn ) such that

A an A r

and, for every a dom(fn ), either

a A

or

r A a.

Then,

fn ( ) B fn (r)

and, for every b ran(fn ), either

b B fn ( )

or

fn (r) B b.

Because (B, B ) is a dense linear ordering, we can pick i < such

that

fn ( ) B bi B fn (r).

In each of the three subcases, it is clear that the prescribed

choice of i works, so the claim is proved, case 2 has been handled

and the forth step is complete.

6.3 Real numbers 131

Back step We dene a nite isomorphism

h : (dom(h), A ) (ran(h), B ).

with

ran(h) = ran(g) {bn }.

The process for dening h from g is like the process of dening

g from fn except that we reverse the roles of A and B.

Finally, having completed both the back and forth steps, let

fn+1 = h.

Having characterized Q with its ordering up to isomorphism, we

turn our attention to R. What is R and which criteria characterize

R with its ordering up to isomorphism? As motivation, consider

the set

S = {x Q | x2 < 2}.

Notice that S is bounded above by the rational number 3/2. This

is because, if x is a rational number and x2 < 2, then x < 3/2.

(If x 3/2, then x2 9/4 > 2.) Based on the mathematics you

knew

before starting to read this book, youwould have said that

2 is also an upper bound for S, in fact, 2 is the least upper

bound for S, which we write

lub(S) = 2.

2 Q and we have not dened

This is still correct except that

R so it is not fair to mention 2 yet. The moral of this paragraph

is that, in passing from Q to R, we want to include least upper

bounds for all bounded sets.

Denition 6.6 Let (A, A ) be a strict linear ordering. If S A

and y A, then y is an upper bound for S i x A y for every

x S.

To be clear, x A y means that either x A y or x = y.

Denition 6.7 Let (A, ) be a strict linear ordering. If S A

and y A, then y is a least upper bound for S i y is an upper

bound for S and, for every upper bound z for S, y A z.

132 Dense linear orderings

It is easy to see that S has at most one least upper bound in

(A, A ). For if it had two, y and z, then y A z and z A y so

y = z.

Now we arrive at the property that distinguishes R from Q.

Denition 6.8 A strict linear ordering (A, A ) has the least

upper bound property i for every non-empty S A, if S has

an upper bound in (A, A ), then S has a least upper bound in

(A, A ).

We indicated above that Q with its usual ordering does not have

the least upper bound property since

lub({x Q | x2 < 2}) Q.

Of course, this is not the only example of a subset of Q without

a least upper bound in Q. By contrast, the least upper bound

property is one of the key properties that characterize R with its

usual ordering.

Another essential property about the usual ordering of R is

that Q R and between any two real numbers there is a rational

number. In other words, Q is dense in R.

Denition 6.9 If (B, B ) is a strict linear ordering and A B,

then A is dense in (B, B ) i for all x, y B, there exists z A

such that x B z B y.

Theorem 6.5 tells us that in a certain sense it does not matter

which countable dense linear ordering without endpoints we take

to be Q since they are all isomorphic. The next theorem tells us

that, once we settle on a choice of Q, there is a way to extend

Q to obtain an appropriate choice for R, and this choice for R is

unique up to isomorphisms that x Q.

Theorem 6.10 (Dedekind) Let (A, A ) be a countable dense

linear ordering without endpoints. Then there exists (B, B ) such

that:

(B, B ) is a dense linear ordering without endpoints;

(A, A ) is a subordering of (B, B ), that is, A B and, for all

x, y A,

x A y x B y;

A is dense in (B, B );

6.3 Real numbers 133

(B, B ) has the least upper bound property.

then there is an isomorphism

f : (B, B ) (B , B )

f (x) = x.

that, for every y B,

and setting

Now we prove the rst part of Theorem 6.10. The commutative

diagram in Figure 6.2 summarizes a big part of our plan. First, we

will dene (D, D ) and prove it is a dense linear ordering without

endpoints that has the least upper bound property. Then, we will

nd an isomorphic copy (C, C ) of (A, A ) sitting densely inside

of (D, D ). Finally, we will nd (B, B ) sitting above (A, A )

the way that (D, D ) sits above (C, C ). What exactly the linear

orderings and isomorphisms in the diagram are will be explained

soon.

The proof uses the following key denition. We say that a set

L is a left-cut i

L A,

L = ,

L = A,

for every x A and y L, if x A y, then x L, and

for every x L, there exists y L such that x A y.

The next to last condition says that left-cuts are closed to the left.

The last condition says that left-cuts do not have a right endpoint.

isomorphism

(B, B ) o_ _ _ _ _ _ g _ _ _ _ _ _ (D, D )

O O

countable dense subordering countable dense subordering

xL x

(A, A ) _ _ _ _ _ _ _ _ _ _ _ _/ (C, C )

isomorphism

6.3 Real numbers 135

Before continuing, we give a couple of examples of left-cuts in

the most interesting case where (A, A ) = (Q, <Q ). Both

{x Q | x < 2}

and

{x Q | x < 0 or x2 < 2}

are left-cuts of (Q, <Q ). The rst example has a least upper bound

in Q, namely

2 = lub({x Q | x < 2}).

But the second

example does not have a least upper bound in Q

because 2 Q. It is worth keeping these examples in mind in

what follows.

Returning to the proof of the theorem, dene

D = {L A | L is a left-cut}

and

L D M L M.

We claim that (D, D ) is a strict linear ordering. It is obviously

transitive and irreexive, so all that remains is to see that it is

total. Suppose that L and M are left-cuts. We must show that

M L or L M . For this, we assume M L and prove L M .

Pick y M such that y L. Consider an arbitrary x L. Since

(A, A ) is a strict linear ordering, either x A y or y A x. If

x A y, then x M since y M and M is closed to the left. If

y A x, then y L since x L and L is closed to the left, but this

is a contradiction since y L. Therefore x M . Since x was an

arbitrary element of L, we have seen that L M . Clearly, L M

since y M L. This shows that (D, D ) is total.

Now we verify that (D, D ) has no endpoints. Let M D. Since

M is a left-cut, M = and M = A. Pick y M and z A M .

Let L = {x A | x A y} and N = {x A | x A z}. Routine

checking shows that L and N are left-cuts and L M N .

Since (A, A ) has no right endpoint, there exists z A such that

z A z . Let N = {x A | x A z }. It is easy to see that

L M N . Therefore, M is not an endpoint of (D, D ).

Before proving that (D, D ) has the least upper bound prop-

erty, we pause again to give a motivating example in the case

136 Dense linear orderings

(A, A ) = (Q, <Q ). Let

S = {x Q | x < y} | y Q and y 2 < 2 .

Then S is a non-empty family of left-cuts of (Q, <Q ). Also, S is

bounded in the sense that every element of S is contained in the

left-cut {z Q | z < 3/2}. Finally, notice that

S = {x Q | x < 0 and x2 < 2}.

Back to the proof of the theorem, we are ready to show that

(D, D ) has the least upper bound property. Suppose that S is

a non-empty subset of D and S has an upper bound in (D, D ).

We must prove that S has a least upper bound in (D, D ). Let

M = S. It suces to show that M D and M = lub(S).

Leaving some of the details to the reader, here are the main ideas

for showing that M is a left-cut and an upper bound for S.

If L S, then L A. Thus M A.

S= and, if L S, then L = . Thus M = .

If L S, then L has no right endpoint. Therefore M has no

right endpoint.

If L S, then L is closed to the left. Therefore M is closed to

the left.

If N is an upper bound for S, then L N for every L S, so

M N . Because N = A, also M = A.

Now we explain why M is the least upper bound of S. Consider

any upper bound N for S. We must show that M N . Suppose

N M since both are left-cuts. Pick y M N .

otherwise. Then

Since M = S, there exists L S such that y L. Then L N

because N is an upper bound for S. Thus y N , which is a

contradiction.

Technically, (A, A ) is not a subordering of (D, D ) as we can-

not expect A D. However, we can nd a subordering (C, C )

of (D, D ) such that

(A, A ) (C, C ).

To see this, for each y A, let

Ly = {x A | x A y}.

6.3 Real numbers 137

Then let

C = {Ly | y A}.

It is easy to see that the map

y Ly

is an order-preserving injection from (A, A ) to (D, D ) whose

range is C.

We must prove that C is dense in (D, D ). Suppose that K D

M . Recall this means that K M . Pick y M K. Since M is a

left-cut, it does not have a largest element, so we may pick z M

such that y A z. Clearly,

K D Ly D Lz D M.

Observe that Lz is strictly between K and M . This shows that C

is dense in (D, D ).

Towards dening (B, B ) as in the statement of Theorem 6.10,

rst dene an injection g with domain D as follows.

If x A, then g(Lx ) = x.

If M D C, then g(M ) = (A, M ).

In the rst case, notice that if Lx = Ly , then x = y, so g really

is a function. In the second case, one point is that (A, M ) A

because otherwise

A {A} {{A}, {A, M }} = (A, M ) A,

which contradicts the Foundation Axiom. The other point is that

(A, M ) = (A, N ) M = N.

The two points combined are used to see that g really is an injec-

tion. Finally, dene

B = g[D]

and dene B by

g(L) B g(M ) L D M.

Clearly,

g : (D, D ) (B, B )

and

g C : (C, C ) (A, A ),

138 Dense linear orderings

which completes the proof of Theorem 6.10.

Theorem 6.10, we say that (B, B ) is a Dedekind completion of

(A, A ). We may dene R to be a Dedekind completion of Q. By

the moreover part of Theorem 6.10, it is not particularly important

which Dedekind completion we choose.

Throughout this chapter, we have been thinking about Q and

R as certain kinds of linear orderings. But there is much more

to numbers than how they are ordered! The rational numbers

also come equipped with an arithmetic structure, by which we

mean addition, subtraction, multiplication, division, exponentia-

tion, etc. Moreover, the arithmetic structure of the rational num-

bers lifts nicely to the familiar arithmetic structure for the real

numbers. It is possible to explain how this lifting is achieved in

terms of left-cuts and Dedekind completions. We already saw a

hint of this in our discussion of 2 earlier in the chapter. Curious

readers might enjoy working out the formal development as an

independent project.

Exercises

Exercise 6.1 The point of this exercise is to dene Q. Therefore,

you may not assume anything about Q, not even that Q exists,

in your solution. However, you may use standard properties of

with its order, addition and multiplication. For example, writing

3 6

2 = 4 is unacceptable at this point because we have not yet dened

fractions but writing 3 4 = 12 = 2 6 is acceptable because it only

refers to natural numbers and multiplication.

Let

S = ( 1).

6.3 Real numbers 139

1. Prove that E is an equivalence relation on S.

2. Write [(a, b)]E for the E-equivalence class of (a, b). That is,

[(a, b)]E = (a , b ) | (a , b )E(a, b) .

Let B = S/E. That is

B = {[(a, b)]E | (a, b) S} .

Explain why the formula

[(a, b)]E B [(c, d)]E ad < bc

denes a relation B on B. Hint: If you think there is nothing

to check, think again.

3. Prove that (B, B ) is a countable dense linear ordering with

no right endpoint and whose left endpoint is [(0, 1)]E .

4. Let

A = B {[(0, 1)]E }

and A be the restriction of B to A. Prove that (A, A ) is a

countable dense linear ordering without endpoints.

5. Prove that (A, #A ) is a countable dense linear ordering without

endpoints. (Notice the order is reversed.)

6. Dene

(Q, <Q ) = (A, #A )

(B, B ).

Prove that (Q, <Q ) is a countable dense linear ordering without

endpoints.

Exercise 6.2 Let

Z = {. . . , 2, 1, 0, 1, 2, . . . }

have the usual order on the integers. Prove that Z .

Exercise 6.3 Find a family F such that

every element of F is a countable dense linear ordering, and

for every countable dense linear ordering (A, A ), there exists

a unique (B, B ) F such that (A, A ) (B, B ).

Exercise 6.4 Complete the proof of the moreover part of The-

orem 6.10. See the hint given there.

Exercise 6.5 Prove that R

R R.

140 Dense linear orderings

Exercise 6.6 If

B = (B, B , B , B , B , B )

is a Boolean algebra, then we call B countable i B is countably

innite. Prove that if A and B are countable Boolean algebras

with no atoms, then A and B are isomorphic. Hint: The solution

is quite involved and breaks up into two main components:

Use a back-and-forth style argument to build an order isomor-

phism

f : (A, A ) (B, B ).

Prove that every order isomorphism between Boolean algebras

is also a Boolean algebra isomorphism. Hence,

f : A B.

This second part is a general fact about Boolean algebras, so

your proof should not use the countable and atomless assump-

tions.

Here are a couple of useful lemmas you should prove for the back-

and-forth part of the argument:

If f is a nite partial Boolean algebra isomorphism from A to B,

then there exists a nite partial Boolean algebra isomorphism

g from A to B such that the domain and range of g are nite

Boolean algebras. To see this, take dom(g) to be the Boolean

subalgebra of A generated by dom(f ), and ran(g) to be the

Boolean subalgebra of B generated by ran(f ). Then extend f

in the obvious way to dene g.

The ordering of an atomless Boolean algebra is dense in the

sense that if x z, then there exists y such that x y z.

Keep in mind that the ordering is not linear!

We remark that two examples of a countable atomless Boolean

algebras were given in Exercises 4.17 and 5.22. By this exercise,

they are isomorphic.

7

Filters and ideals

mathematics. After some preliminaries, in Section 7.1, we prove

Tarskis ultralter existence theorem and touch on the theory of

ultraproducts. Filters and ideals are particularly important in ad-

vanced set theory, where the lter generated by the closed un-

bounded subsets of an uncountable regular cardinal plays a major

role. We give the reader a taste of this sort of innitary combina-

torics in Section 7.2. The main results there are Fodors theorem

and an interesting special case of Solovays splitting theorem.

There are many mathematical contexts in which we are given a

set X and we talk about large subsets of X and small subsets of

X. This is so common that it is worth writing down what these

situations share.

F is a lter over X i the following conditions hold:

F.

X F.

For all A, B X, if A F and A B, then B F.

For all A, B X, if A, B F, then A B F.

dening conditions as follows.

142 Filters and ideals

The empty set is not large.

X is large.

If A is large and A B, then B is also large.

If A and B are large, then so is A B.

If you are not convinced by the last clause, replace large by

almost everything to make it even more believable.

Ideals are to small sets what lters are to large sets.

Denition 7.2 Let X be a non-empty set and I P(X). Then

I is an ideal over X i the following conditions hold:

I.

X I.

For all A, B X, if B I and A B, then A I.

For all A, B X, if A, B I, then A B I.

In reading the list above, where you see a set belongs to I, you

can say out loud that it is small (or, maybe better, almost nothing)

to understand the motivation for the condition.

The next two results explain how lters and ideals are related.

Lemma 7.3 If F is a lter over X, then

{X A | A F}

is an ideal over X.

Lemma 7.4 If I is an ideal over X, then

{X A | A I}

is a lter over X.

So the operation A X A takes lters over X to ideals over

X and vice-versa as described by the lemmas. This is an example

of what we call duality in mathematics.

Is every subset A of X either large or small? It depends on the

situation. This thought leads to ultralters and prime ideals.

Denition 7.5 Let F be a lter over X. Then F is an ultralter

over X i for every A X, either A F or X A F.

ideal over X i for every A X, either A I or X A I.

7.1 Motivation and denitions 143

It is about time we introduced some examples!

Example If p X, then {A X | p A} is a principal ultralter

over X. This is the least interesting kind of ultralter.

Example {A | A is nite} is the Frechet lter over . It

is not an ultralter over . For example, neither

Even = {2n | n < }

nor

Odd = {2n + 1 | n < }

are members of the Frechet lter over .

Example For A , dene

|A n |

density(A) = lim

n n

if the limit exists. Then

{A | density(A) = 0}

is the density ideal over .

Example Let I = {x R | 0 x 1}. For the reader who knows

about Lebesgue measure, we mention that an important topic in

analysis and probability is the ideal of null sets,

{A I | A has Lebesgue measure 0},

and its dual lter,

{A I | A has Lebesgue measure 1}.

This is not an ultralter because there are subsets of I whose

Lebesgue measure is strictly between 0 and 1. For example, the

interval

{x R | 0 x 1/2}

has Lebesgue measure 1/2.

Exercise 7.1 Let P and X be non-empty sets with P X, and

F = {A X | P A}.

1. Prove that F is a lter over X.

2. Prove that the following are equivalent:

144 Filters and ideals

(a) F is an ultralter over X.

(b) P is a singleton.

(c) F is a principal ultralter over X.

Exercise 7.2 Let F be the Frechet lter over . Suppose that

G is an ultralter over such that G F. Prove that G is not

principal.

Exercise 7.3 Let X be a non-empty set and F be an ultralter

over X. Prove that, for every n < and sequence Ai | i < n of

subsets of X, if

Ai F,

i<n

The rst important result on this topic is that every lter ex-

tends to an ultralter. You should notice that the rst sentence

of the proof uses the Axiom of Choice to know that P(X) has a

cardinality.

Theorem 7.7 (Tarski) Let F be a lter over X. Then there

exists an ultralter G over X such that F G.

Proof Let = |P(X)| and A | < be a sequence such that

P(X) = {A | < }.

We will dene a sequence G | < by recursion. After each

case in our denition of G , we will verify that if < , then:

G is a lter,

A G or X A G , and

G G .

Base case = 0.

Dene G0 = F.

Successor case = + 1.

We break up this case into three subcases. The rst subcase is,

for every B G ,

B A = .

7.1 Motivation and denitions 145

In the rst subcase, let

G+1 = {C X | there exists B G such that B A C}.

Obvserve that G {A } G+1 because:

G G+1 since, if B G , then B A B, so B G+1 ,

and

A G+1 since X G , so X A G+1 .

Observe that G+1 is a lter because:

G+1 by the rst subcase hypothesis,

X G since X A X,

G+1 is clearly closed upward under , and

G+1 is closed under pairwise intersections since if B, C G ,

then

(B A ) (C A ) = (B C) A G+1 .

The second subcase is that the rst subcase fails and, for every

B G ,

B (X A ) = .

In the second subcase, let

G+1 = {C X | there exists B G such that B(XA ) C}.

Much like in the rst subcase, one shows that

G {X A } G+1

and G+1 is a lter. The reader should complete the verication.

The third subcase is that the rst and second subcases fail. We

will show this does not happen. For contradiction, suppose

B, C G ,

B A =

and

C (X A ) = .

Let D = B C. Then

D G ,

D A =

146 Filters and ideals

and

D (X A ) = .

Therefore D = . But G since G is a lter.

Limit case is a limit ordinal.

Dene

G = G .

<

-increasing sequence of lters. Also, if < , then either

A G+1 G

or

(X A ) G+1 G .

That completes the recursive denition of the sequence

G | <

and the verication that it has the desired properties. Now let

G= G .

<

Finally, G is an ultralter because, for every < , either

A G+1 G

or

(X A ) G+1 G.

X = and F is the Frechet lter over . Notice that in the proof

= |P()| = 20 .

Let G | < 20 be the sequence of lters extending F that was

recursively constructed in the proof of Theorem 7.7.

1. Prove by induction on < 20 that |G | < 20 and G is not

an ultralter over .

7.1 Motivation and denitions 147

2. Use ideas similar to the proof of Theorem 7.7 to show that there

are 22 0 many non-principal ultralters over . Hint: This is a

slightly challenging exercise. View G | < 20 as a branch

through a certain kind of tree with 20 many levels. Argue

that the tree has 22 0 many distinct branches each of which

corresponds to a dierent ultralter over .

Exercise 7.5 below introduces the reader to a certain construc-

tion, which is known as taking an ultrapower by an ultralter.

Intuitively, the idea is to start with a sequence of structures (in

the exercise, the structures are linear orderings) and an ultralter,

F, and to form a new structure by averaging out according to F.

Our meaning will become clear when the reader does the exercise

and reads the discussion that follows. Before starting, recall how

products of sets were dened in Exercise 4.10. In particular, given

a sequence An | n < , we dene n< An to be the set of

functions f such that dom(f ) = and, for every n < ,

f (n) An .

In the special case where all the An s are the same, say An = B,

we end up with

n< B = {f | f is a function from to B} = B.

n< An =

signicance of the ultrapower construction.

Exercise 7.5 Let An | n < be a sequence and

P = n< An .

Let F be an ultralter over .

1. Dene a relation on P by

f g {n < | f (n) = g(n)} F.

Prove that is an equivalence relation on P .

2. For f P , let

[f ] = {g P | f g} .

Also, let

A = {[f ] | f P } .

148 Filters and ideals

Assume that, for each n < , we are given a relation

Rn An An .

Prove that we may dene a relation

RAA

by setting

[f ] R [g] {n < | f (n) Rn g(n)} F.

In other words, prove that the denition of R does not depend

on the choice of representatives for the equivalence classes [f ]

and [g].

3. Assume that (An , Rn ) is a strict linear ordering for every n < .

Prove that (A, R) is also a strict linear ordering.

4. Now assume that, for every n < ,

(An , Rn ) = (, <)

where < is the usual order on the natural numbers. Suppose

that F is a non-principal ultralter over . Prove that (A, R)

is not a wellordering. Hint: Notice that, in this case, P = .

For c Z, consider the function

fc :

dened by

n + c if n + c 0

fc (n) =

0 otherwise.

Prove that

R [f3 ] R [f2 ] R [f1 ] R [f0 ].

5. Again assume that, for every n < ,

(An , Rn ) = (, <).

But suppose instead that F is a principal ultralter over .

Prove that

(A, R) (, <).

Hint: Most but not all of the details are contained in the dis-

cussion after this exercise, so read it rst if you get stuck.

7.1 Motivation and denitions 149

That completes the instructions for Exercise 7.5 but there is

much more we should tell the reader about the construction done

there. The pair (A, R) is called the ultraproduct of the sequence

(An , Rn ) | n < by F. A popular way to express the denition

of R is

[f ] R [g] f (n) Rn g(n) for F-almost every n < .

Often, it helps to think about ultraproducts using this alternative

language. This language makes it clearer what we meant by aver-

aging out in the paragraph preceding Exericse 7.5. In this exercise,

we took ultraproducts of linear orderings but it is also possible to

take ultraproducts of other kinds of structures. For example, the

reader should see how the ultraproduct of a sequence of Boolean

algebras would be dened.

The example of an ultraproduct (A, R) given in the last two

parts of Exercise 7.5 is called the ultrapower of (, <) by F. Instead

of saying ultrapower, we use the term ultraproduct in this case

because all of the pairs (An , Rn ) are the same. The main point

of part 4 of Exercise 7.5 is that the ultraproduct of wellorderings

need not be a wellordering. Figure 7.1 is a rough picture of what

(A, R) looks like in this case. The initial segment of (A, R) that is

isomorphic to is really:

[n 0] R [n 1] R [n 2] R

Keep in mind that, for a xed c < , the function n c is the

function with the constant value c. The chain of relations that we

just displayed says that, for every constant c < ,

[n c] R [n c + 1].

This is because

{n < | c < c + 1} = F.

But we are also claiming that there is no equivalence class strictly

between each [n c] and [n c + 1]. To see this, let b = c + 1

and suppose f : is a function such that

[f ] R [n b].

Then

{n < | f (n) < b} F.

150 Filters and ideals

/ (isomorphic to )

..

.

o / (isomorphic to Z)

..

.

o / (isomorphic to Z)

..

.

Clearly,

{n < | f (n) < b} = {n < | f (n) = a}.

a<b

{n < | f (n) = a} F.

[f ] = [n a].

Looking again at Figure 7.1, after the initial segment of (A, R)

that is isomorphic to , there are innitely many pieces each of

which is isomorphic to Z. We call these Z-chains. In terms of the

functions fc dened in part 4 of Exercise 7.5, here is one example

7.1 Motivation and denitions 151

of a Z-chain:

R [f2 ] R [f1 ] R [f0 ] R [f1 ] R [f2 ] R .

Using the ideas from the previous paragraph, the reader should

prove that [fc ] R [fc+1 ] and there are no equivalence classes of

functions strictly between [fc ] and [fc+1 ]. In other words, that

this really is an example of a Z-chain.

Now dene

3n + c if 3n + c 0

hc (n) =

0 otherwise.

Here is a second example of a Z-chain:

R [h2 ] R [h1 ] R [h0 ] R [h1 ] R [h2 ] R

The reader should verify that this is indeed a Z-chain. We claim

that this Z-chain lies entirely after our rst example of a Z-chain.

In other words, for all a, c Z,

[fa ] R [hc ].

By the denition of R, this just says that

{n < | fa (n) < hc (n)} F.

The main observation needed to see why this is true is that

{n < | n + a < 3n + c} = {n < | (a c)/2 < n}

= {n < | n (a c)/2}

F

because it is the complement of a nite set and F is non-principal.

We give a third example of a Z-chain. For b Z, dene

2n + b if 2n + b 0

gb (n) =

0 otherwise.

Our third Z-chain

R [g2 ] R [g1 ] R [g0 ] R [g1 ] R [g2 ] R

lies strictly between the other two as the reader should verify.

Building on these observations, one sees that the Z-chains them-

selves form a dense linear ordering without endpoints. We leave it

152 Filters and ideals

to the reader to ll in the details and continue this investigation

as an extremely worthwhile project.

Ultraproducts are used in many branches of mathematics, not

just set theory. A famous and intriguing example is Abraham

Robinsons theory of non-standard analysis, which rehabilitated

Gottfried Wilhelm Leibnizs seventeenth-century innitesimal cal-

culus. Innitesimals are supposed to be numbers > 0 such that

< x for every positive number x R. Of course, there are no such

R. Nevertheless, without really knowing what he meant by in-

nitesimals, Leibniz developed recipes for working with them that

yielded correct answers to questions about geometry and physics.

While this represented tremendous intuition, Leibnizs theory was

considered controversial and was eventually abandoned in favor of

the rigorous development of calculus provided by Augustin-Louis

Cauchy in the eighteenth-century. Much later, in the 1960s, Robin-

son vindicated Leibniz by saying what innitesimals really are and

explaining why it was legitimate to derive calculus formulas using

them. To get the idea, let F be a non-principal ultralter over

and take the ultrapower of (R, <) by F. Call this ultrapower

(R , < ). Pretty much like in our elaboration on Exercise 7.5, we

see that, for all constants a < b in R,

[n a] < [n b].

So there is a copy of (R, <) sitting inside of (R , < ). But there are

new points to the left, to the right and in between. For example,

for every c R,

[n c] < [n n],

which shows there are new positive innite members of R entirely

to the right of our copy of R. Similarly, for every c R,

[n n] < [n c],

so there are new negative innite members of R entirely to the

left of our copy of R. Even more interesting is the fact that, for

every positive c R,

[n 0] < [n 1/n] < [n c].

In other words, [n 1/n] is greater than our copy of 0 and less

than our copy of c for every positive real number c. For this reason,

it is reasonable to say that the equivalence class [n 1/n] is an

7.2 Club and stationary sets 153

example of an innitesimal member of R . (Technically, when we

write n 1/n here, we really mean

1/n if n = 0

n

0 if n = 0

because the domain must be but we cannot divide by zero.)

There is more to understanding why one can reason about (R , < )

and come to certain correct conclusions about (R, <) but that

requires a basic background in mathematical logic, which we do

not presume. Our intent was merely to introduce the reader to

this fascinating and historically signicant subject.

This section builds on the previous section and Chapters 3 and 4.

Given a limit ordinal , there is a interesting and very useful lter

over called the club lter. Its dual is called the non-stationary

ideal. Before we can say what these are, we need some denitions.

Denition 7.8 Let be a limit ordinal and C be a set. Then:

C is unbounded in i sup(C ) = .

C is closed in i for every < , if C = , then

sup(C ) C.

C is club in i C is closed and unbounded in .

Here are some examples with = 1 .

Example If < 1 , then { < | < } is club in 1 .

Example { < 1 | is a limit ordinal} is club in 1 . It is closed

because a limit of limit ordinals is also a limit ordinal. It is un-

bounded because if is a countable ordinal, then + is a count-

able limit ordinal.

Example { < 1 | is a successor ordinal} is not club in 1 .

While it is unbounded, it is not closed. For example,

= sup (n + 1)

n<

154 Filters and ideals

Example { < 1 | = } is club in 1 . Here is ordinal

exponentiation. The proof is a little more than what you were

asked to show in Exercise 4.12.

Denition 7.8 applies to all limit ordinals but most often we

apply it specically to uncountable regular cardinals such as 1 .

Remember that is regular i cf() = . And remember that all

innite successor cardinals are regular.

Exercise 7.6 Let be a regular cardinal and C . Prove C

is unbounded in i |C| = .

Now we say what this has to do with lters.

Lemma 7.9 Assume that cf() > . Let

F = {A | there exists a club C in such that C A}.

Then F is a lter over .

Proof The only condition in the denition of lter that is not

obvious is closure under intersections. It is enough to show that if

C and D are club in , then so is C D.

First we show that C D is closed in . Let < and assume

that C D = . Let

= sup(C D ).

We must show that C D. Easily, we see that = sup(C )

and = sup(D ). Since C and D are closed, C and D.

To nish we show that C D is unbounded in . Let < .

We must nd C D such that < . By recursion on n < ,

dene ordinals n , n < as follows. Pick 0 C with 0 > .

Given n , pick n D with n > n . Given n , pick n+1 C

with n+1 > n . We can do all this picking because C and D are

unbounded in . Now let be the supremum of either sequence;

it is the same because of the interleaving. That is,

= sup n = sup n .

n< n<

= sup(C ) C

and

= sup(D ) D,

7.2 Club and stationary sets 155

hence C D as desired.

The lter in Lemma 7.9 is called the club lter over . The

main point of the proof is that the intersection of two clubs is

club if has uncountable conality. This is not necessarily true if

has countable conality. For instance, Even = {2n | n < } and

Odd = {2n + 1 | n < } are disjoint unbounded subsets of , and

these sets are closed for trivial reasons.

f :

be a function. Prove that

{ < | f [] }

is club in . Remark: This implies that { < 1 | = } is club

in 1 in the special case = 1 and f : . The special

case was the subject of Exercise 4.12 and the general argument is

similar.

that if < and C | < is a sequence of club subsets of ,

then the set

{C | < }

= +1, is immediate from the induction hypothesis and the case

= 2, which was handled in the proof of Lemma 7.9. Suppose is

a limit ordinal. In this case, the proof of closure is straightforward

(similar to the case = 2). For the proof of unboundedness, given

0 < , dene an increasing sequence | < such that, for

every < ,

{C | < }.

sequence C | < such that, for every < , C is club in

but

{C | < } = .

156 Filters and ideals

Let F be the club lter over . Recall that

F = {S | there exists a club C in such that C S}.

As we indicated before, we intuitively think of members of F as

large subsets of . Let I be the ideal dual to F. Then

I = {S | S F}.

We think of members of I as small subsets of . Observe that if

S , then

S I there is a club C in such that S C = .

We are also interested in subsets of that are not small. Notice

that if S , then

S I for every club C in , S C = .

Intuitively, this says that a subset of is not small i it meets

every large subset of . We give such sets the following name.

Denition 7.10 Let be a limit ordinal and S . Then S is

stationary in i for every club C in ,

S C = .

Notice that if C is club, then C is stationary. This is because if

D is club, then C D is club, in particular, C D = . Intuitively,

this says that if a set is large, then it is not small.

You can get additional intuition for Denition 7.10 if you hap-

pen to know about Lebesgue measure on the unit interval

{x R | 0 x 1}.

The relevant analogies are:

contains a club stationary not stationary

= =

measure 1 positive measure measure 0

The following exercise gives an important example with = 2 .

Exercise 7.10 Let

C = { < 2 | is a limit ordinal},

E = { C | cf() = }

and

C E = { C | cf() = 1 }.

7.2 Club and stationary sets 157

1. Prove that C is club in 2 . (This is pretty obvious.)

2. Prove that E is stationary in 2 .

3. Prove that C E is stationary in 2 .

4. Use parts 1, 2 and 3 to prove that the club lter over 2 is not

an ultralter.

Now we come to one of the most fundamental tools for studying

club and stationary sets.

Theorem 7.11 (Fodor) Let be uncountable regular cardinal

and f : be a function such that

{ < | f () < }

is stationary in . Then there exists < such that

{ < | f () = }

is stationary in .

We will derive Theorem 7.11 from Lemma 7.12, which is inter-

esting in its own right. Exercise 7.9 tells us that the intersection

of many club sets might not be club. Lemma 7.12 says that the

diagonal intersection of many club sets is club.

Lemma 7.12 Let be an uncountable regular cardinal. Suppose

that

C | <

is a sequence of club subsets of . Let

D = { < | C for every < }.

Then D is club in .

We call D the diagonal intersection of C | < . Most

commonly, you will see it written < C .

Proof of Lemma 7.12 First we show that D is closed in . Let

< . Assume that D = and let = sup(D ). We

must show that D. For contradiction, suppose that D. It

follows easily that is a limit ordinal and D = D . Hence

= sup(D ).

By the denition of D, there exists < such that C . Since

C is closed, there are two cases:

158 Filters and ideals

1. C = .

2. C = and sup(C ) C .

In the second case, sup(C ) < because C . In either

case, we may pick D such that < < . In the second

case, we can also make sure that sup(C ) < . Then, in both

cases, D and < but C . This directly contradicts the

denition of D.

Now we show that D is unbounded in . By recursion on <

dene as follows. Let 0 = 1. Given < , pick +1 >

such that

+1 C .

<

= sup .

<

tion on < , one sees that, for every < ,

<

and if < , then

C .

Suppose that < is a limit ordinal. Then, for every < ,

{ | + 1 < < } C

and

= sup({ | + 1 < < }) = sup(C )

so

C

since C is closed. We have seen that

{ | < and is a limit ordinal} D.

The set on the left is unbounded in and hence so is D.

Proof of Theorem 7.11 Let

S = { < | f () < }.

7.2 Club and stationary sets 159

Our assumption is that S is stationary in . For each < , let

T = { S | f () = }.

For contradiction, suppose that no T is stationary in . For each

< , pick C club in such that T C = . Let

D = { < | C for every < }.

By Lemma 7.12, D is club in . Pick D S. Then, for every

< , f () = . In other words, f () . But f () < since

S.

In Exercise 7.10, we saw that the club lter over 2 is not an ul-

tralter. The proof outlined there generalizes to regular cardinals

2 . To see this, note that, for every regular cardinal < ,

{ < | cf() = }

is stationary in . Moreover, these sets are disjoint for dierent

. But Exercise 7.10 does not generalize to = 1 because

is the only regular cardinal less than 1 . However, the following

theorem implies that the club lter is not an ultralter over 1 . It

is a special case of a more powerful result known as the Solovay

splitting theorem.

Theorem 7.13 There is S 1 such that S and 1 S are

stationary in 1 .

Proof Suppose Theorem 7.13 is false. For each positive < 1 ,

pick a surjection

f : .

This is obviously possible because is countable. Intuitively, we

will reach a contradiction by using the club lter over 1 to average

out the sequence f | < 1 and obtain a new surjection from

onto 1 . This will be a contradiction because 1 is uncountable

by denition.

Claim 7.13.1 For every n < , there exists < 1 such that

{ < 1 | f (n) = }

is stationary in 1 .

160 Filters and ideals

Proof Fix n < . Consider the function

g : f (n).

Then g() < whenever 0 < < 1 . By Fodors theorem, there

exists < 1 such that

{ < 1 | g() = }

is stationary in 1 .

Claim 7.13.2 For every n < , there is at most one < 1

such that

{ < 1 | f (n) = }

is stationary in 1 .

Proof Fix n < . For < 1 , let

S = { < 1 | f (n) = }.

Clearly, if < < 1 , then

S S = .

Suppose < < 1 and both S and S are stationary. Then

1 S is also stationary because

S 1 S .

This means the statement of Theorem 7.13 holds with S = S . But

we assumed that Theorem 7.13 is false, so we have a contradiction.

by setting g(n) equal to the unique < 1 such that

{ < 1 | f (n) = }

is stationary in 1 .

Claim 7.13.3 g is a surjection from to 1 .

Proof Let < 1 . For contradiction, suppose that ran(g).

Then, for every n < ,

{ < 1 | f (n) = }

7.2 Club and stationary sets 161

is non-stationary in 1 . For each n < , pick a club Cn in 1 such

that

Cn { < 1 | f (n) = } = .

In other words, for every n and every Cn ,

f (n) = .

Let

D= {Cn | n < }.

Then D is club in 1 and, for every n < and every D,

f (n) = .

Since D is unbounded in 1 , there exists D such that < .

Since f is a surjection from to , there exists n < such that

f (n) = .

This is a contradiction.

Claim 7.13.3 contradicts the fact that 1 is uncountable.

Exercise 7.11 Use Theorem 7.13 to prove that the club lter

over 1 is not an ultralter.

Solovays splitting theorem says that if is a regular uncount-

able cardinal and S is a stationary in , then there is a sequence

S | < of stationary subsets of S such that, for all < < ,

S S = .

In other words, S can be split into many disjoint stationary

pieces. This is more powerful than Theorem 7.13, which says that

1 can be split into two disjoint stationary pieces.

Appendix

Summary of exercises on Boolean algebra

exercise gave a characterization of nite Boolean algebras up to

isomorphism, namely, they all look like P(S) for some nite set

S. The proof involved looking at atoms.

An example of an innite atomless Boolean algebra, P()/Finite,

was given in Exercise 2.13. By Exercise 4.15, P()/Finite is un-

countable. Another example of an uncountable atomless Boolean

algebra, the family of clopen subsets of the Baire space, was the

subject of Exercise 5.23.

The nite Boolean algebras of truth tables, Tn for n < , and

the innite Boolean algebra T where discussed in Exercise 4.17.

We saw that T is a countable atomless Boolean algebra in part 6

of that exercise. Another example of a countable atomless Boolean

algebra was given in Exercise 5.22. This was the family of clopen

subsets of the Cantor space. The fact that all countable atomless

Boolean algebras are isomorphic was the topic of Exercise 6.6.

This is an important theorem whose proof uses a back-and-forth

construction.

Selected further reading

For example, some of the material that we covered in our course

can also be found in Enderton (1977) and Hrbacek and Jech

(1999).1 These books are dierent from each other and from ours,

which certainly benets the reader.

Those who would like to go on to more advanced set theory

should rst learn basic mathematical logic. Again, there are many

options. To name two, Enderton (2001) is an excellent starting

point, while Goldstern and Judah (1998) is a bit more advanced.

This course and an elementary background in logic prepare the

reader for graduate level set theory. Two indispensable texts are

Kunen (1983) and Jech (2003). Our reader who enjoyed ordinal

and cardinal arithmetic and innitary combinatorics, especially

Sections 4.3, 5.5, 5.6 and 7.2, and would like to learn the rela-

tive consistency results of Godel and Cohen on the Continuum

Hypothesis, which were mentioned in Section 4.2, will be particu-

larly drawn to these wonderful classics.

The material covered in Sections 5.1 through 5.4 is part of a

broad subject called descriptive set theory, which is a certain com-

bination of set theory, analysis and logic. To continue in this direc-

tion, the reader would want to know the fundamentals of analysis.

At the advanced undergraduate level, two analysis textbooks to

consider are Rudin (1976) and Royden (1988). For descriptive set

theory, two beginning graduate level texts are Kechris (1995) and

Moschovakis (2009). These have very dierent emphases. Roughly,

1 Here and below, we cite only the most recent edition available.

Selected further reading 165

the former ties set theory to analysis more than logic, while for

the latter it is the other way around.

Yet another subject that is intertwined with set theory and logic

is model theory. In Chapter 6, we saw examples of classication up

to isomorphism. This idea is important throughout mathematics

but especially in model theory. We also touched on ultraproducts

in Section 7.1. This is a model-theoretic construction that has

applications in many elds, particularly in set theory. A classic

beginning graduate model theory text is Chang and Keisler (1990).

Set theory is a vast topic of current mathematical research.

The enormous Handbook of set theory, edited by Foreman and

Kanamori (2010), comes in three volumes with a total of twenty-

four chapters by various authors. It suces to give the reader an

accurate impression of the many directions the subject has taken

in recent decades.

Bibliography

Chang, C. C., and Keisler, H. J. 1990. Model theory. Third edn. Studies in

Logic and the Foundations of Mathematics, vol. 73. Amsterdam: North-

Holland.

Enderton, Herbert B. 1977. Elements of set theory. New York: Academic

Press [Harcourt Brace Jovanovich Publishers].

Enderton, Herbert B. 2001. A mathematical introduction to logic. Second edn.

Harcourt/Academic Press, Burlington, MA.

Foreman, Matthew, and Kanamori, Akihiro (eds). 2010. Handbook of set the-

ory. New York: Springer-Verlag. In three volumes.

Goldstern, Martin, and Judah, Haim. 1998. The incompleteness phenomenon.

Natick, MA: A. K. Peters Ltd. Reprint of the 1995 original.

Hrbacek, Karel, and Jech, Thomas. 1999. Introduction to set theory. Third

edn. Monographs and Textbooks in Pure and Applied Mathematics, vol.

220. New York: Marcel Dekker Inc.

Jech, Thomas. 2003. Set theory. Springer Monographs in Mathematics. Berlin:

Springer-Verlag. The third millennium edition, revised and expanded.

Kechris, Alexander S. 1995. Classical descriptive set theory. Graduate Texts

in Mathematics, vol. 156. New York: Springer-Verlag.

Kunen, Kenneth. 1983. Set theory. Studies in Logic and the Foundations of

Mathematics, vol. 102. Amsterdam: North-Holland. Reprint of the 1980

original.

Moschovakis, Yiannis N. 2009. Descriptive set theory. Second edn. Mathe-

matical Surveys and Monographs, vol. 155. Providence, RI: American

Mathematical Society.

Royden, H. L. 1988. Real analysis. Third edn. New York: Macmillan.

Rudin, Walter. 1976. Principles of mathematical analysis. Third edn. New

York: McGraw-Hill. International Series in Pure and Applied Mathe-

matics.

Index

aleph (), 58 complete metric space, 94

antichain, 124 concatenation of wellorderings, 41

Aronszajn tree, 120 Continuum Hypothesis (CH), 66, 164

atom, 1820, 78, 104, 139 convergent sequence, 93

countable, 55

back-and-forth construction, 128, 129, countably innite, 55

140, 163

Baire category theorem, 94 Dedekind completion, 138

Baire space, 84 DeMorgan laws, 16

bijection, 4 density ideal, 143

Boolean algebra, 1820, 77, 78, 104, determined game, 105

139, 163 diagonal argument, 56, 72, 96, 110

diagonal intersection, 157

Cantor normal form, 51 domain, 3

Cantor perfect set theorem, 95, 113, 115 duality, 142

Cantor space, 92

Cantor theorem, 55 equivalence relation, 5, 17, 20, 77, 120,

CantorBendixon derivative, 95 139

CantorBernsteinSchroeder theorem, family, 7

58, 76 lter, 141

cardinal arithmetic Fodors theorem, 157

addition, 61 Frechet lter, 143, 146

conality, 68 function, 3

exponentiation, 64 GaleStewart theorems, 108, 110

multiplication, 61 Godel, 66, 164

Cartesian product, 2, 11, 15 graph of a function, 4

Cauchy, 152

Cauchy sequence, 93 Hilbert, 66

chain in a partial ordering, 77 homogeneous set, 115

characteristic function, 65 ideal, 142

choice function, 14 image, 3

clopen, 88, 103 indexed family, 3

closed set of ordinals, 153 induction, viii, 25, 34

closure, 92 inductive set, 11, 15

club lter, 154 inherited topology, 92

club set, 153 injection, 4

conality, 68 inverse function, 4

Cohen, 66, 164 irreexive relation, 23

collection, 7 isomorphism

coloring, 115 of Boolean algebras, 19, 139

168 Index

of relations, 37, 126 supremum, 34

KleeneBrouwer ordering, 103 surjection, 4

Konig innity lemma, 97 symmetric dierence, 16

Konig lemma, 72 Tarski ultralter existence theorem, 144

language of set theory, 7 topological space, 82

Lebesgue measure, 143, 156 total relation, 23

left-most branch of a tree, 102 transitive closure, 17

Leibniz, 152 transitive relation, 23

transitive set, 15, 28

limit cardinal, 59

tree on , 90

limit ordinal, 34

truth table, 78

Lindelof space, 92

type of a wellordering, 40

linear ordering, 24

ultralter, 142

map, 3

ultrapower, 149

maximum, 34

ultraproduct, 149

metric compatible with a topology, 84,

unbounded set of ordinals, 153

89

uncountable, 55

metric space, 83

Mostowski collapse, 37, 40 V hierarchy, 15, 30, 36, 37, 49

non-standard analysis, 152 wellfoundedness, 24

wellordering, 24

one-to-one correspondence, 4

one-to-one function, 4 ZF, 54, 77

onto function, 4 ZFC

operation, 3 Axiom of Choice, 14, 16, 53, 54, 57,

ordered pair, 2 76, 77, 110, 144

ordinal, 30 Comprehension Scheme, 10

ordinal arithmetic Empty Set Axiom, 7

addition, 42 Extensionality Axiom, 8

exponentiation, 48 Foundation Axiom, 13, 16, 30, 36, 37,

multiplication, 46 39, 137

Innity Axiom, 11

partial function, 26 Pairing Axiom, 8

partial ordering, 77 Power Set Axiom, 10

partition, 5 Replacement Scheme, 12, 54

perfect set game, 113 Union Axiom, 9

perfect tree, 94 Zorn lemma, 77

pigeonhole principle, 118

prime ideal, 142

principal ultralter, 143

Ramsey theorem, 117

range, 3

recursion, viii, 26, 35

regular cardinal, 68

relation, 2

restriction of a function, 4

Robinson, 152

Russell paradox, 16

separable topological space, 93

singular cardinal, 68

Solovay splitting theorem, 159, 161

stationary set, 156

strategy, 105

strict linear ordering, 24

subset, 7

successor cardinal, 59

successor ordinal, 34

## Molto più che documenti.

Scopri tutto ciò che Scribd ha da offrire, inclusi libri e audiolibri dei maggiori editori.

Annulla in qualsiasi momento.