Sei sulla pagina 1di 180

A Course on Set Theory

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.

e r n e s t s c h i m m e r l i n g is a Professor of Mathematical Sciences at Carnegie


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

This publication is in copyright. Subject to statutory exception


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.

First published 2011

Printed in the United Kingdom at the University Press, Cambridge

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

Library of Congress Cataloguing in Publication data

ISBN 978-1-107-00817-5 Hardback


ISBN 978-1-107-40048-1 Paperback

Cambridge University Press has no responsibility for the persistence or


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

Note to the instructor page vii


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

This book was written for an undergraduate set theory course,


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

As an undergraduate, I studied from Elements of set theory by


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

In one sense, set theory is the study of mathematics using the


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}.

2. The image of S under R is


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}.

Notice that ran(R) = R[dom(R)].


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}.

Denition 1.7 Let f : A B be a function.

1. f is an injection i for all x, x A, if x = x , then f (x) = f (x ).


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.

Exercise 1.4 Equivalence relations play an important role in


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

In the most general terms, when we talk about a mathematical the-


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,

set = collection = family

and
member = element.

Also, the three phrases,

x belongs to A,
x is an element of A and
x is a member of A,

all mean the same thing, namely x 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},

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

5 = 4 {4} = {0, 1, 2, 3, 4}
10 ZFC
and, in general,
n + 1 = n {n} = {0, . . . , n}.

Power Set Axiom


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 ) = {, {}},

V3 = P(V2 ) = {, {}, {{}}, {, {}}}


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:

A [(x A y P (x, y)) = (B x A y B P (x, y))] .


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 )}.

Let S be a set and F P(S). Assume that F = . Prove that


 
S F = {S X | X F}
and
 
S F= {S X | X F}.

Exercise 2.8 In general, dene


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

2. Assume in addition that


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.

Exercise 2.10 Dene a relation E on P() according to the


formula
(x, y) E x  y is nite.

(See Exercise 2.9 for the denition of x  y.)

1. Prove that E is an equivalence relation on P ().

2. We write [x]E for the equivalence class of x, in other words,

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

Prove that for every E-equivalence class [x]E is innite.

3. By P () /E we mean the family of E-equivalence classes, that


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

Prove that P () /E is innite.

Exercise 2.11 Let A be a set. By recursion on n , dene

B0 = A

and

Bn+1 = Bn .

Let

C= {Bn | n < }.

1. Which axioms of ZFC are used to see that C is a set?


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.

We call C the transitive closure of A.


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

At the end of Chapter 1, we gave examples of sentences in English


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:

1st Vice President


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:

16th Secretary of Education


17th Secretary of Veterans Aairs
18th Secretary of Homeland Security

What is the point? In English, we usually count ordinal numbers


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

In set theory, we start counting ordinals from 0. For example, in


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.

Denition 3.2 Let and  be two relations on A. Suppose


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.

Denition 3.4 (A, ) is a wellordering i it is a strict linear


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.

Denition 3.7 f is a partial function from A to B i there


exists A A such that f : A B.

Theorem 3.8 (Recursive denitions) Let (A, ) be a wellorder-


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

Then h is a partial function from A to B with



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.

It is easy to see that gz is a z-approximation.


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

Then G is a function with



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

and, for every z A,


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.

3.2 Ordinal numbers


We start by repeating a denition from Exercise 2.3.

Denition 3.9 A is a transitive set i for all x and y, if x y


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 .

Proof We prove the lemma by induction on . The induction


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.

The relationship between ordinals and arbitrary wellorderings


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 } .

This completes the proof by induction. In particular, we have seen


that if A, then
() < .
Since = [A], this implies that , hence .

3.3 Ordinal arithmetic


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.

Denition 3.34 Let and be ordinals. Then their sum is


+ = 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.

Lemma 3.36 Let and be ordinals. Then there is a unique


function f with domain such that, for every < ,


if = 0
f () = f () + 1 if = + 1


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

namely, the function given by


f () = + .

Sketch of proof Use induction on < to see that that ordinal


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

+ 0 = ,
+ ( + 1) = ( + ) + 1 and
if is a limit ordinal, then + = sup ({ + | < }).

Then apply Theorem 3.24. The details comprise Exercise 3.5.


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.

Lemma 3.37 Let be ordinals. Then there is a unique


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 3 = sup ({3n | n < }) = < 3 .


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 .

Exercise 3.3 Prove that, for every ordinal ,


{ 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

We now turn from the study of order to that of cardinality, which


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.

4.1 Cardinal numbers


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.

Proof The rough idea is to let f (0) be an element of A, then let


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 = } .

By the Axiom of Choice, there exists a choice function c : F A


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
= || = || .

Corollary 4.11 For every cardinal , there exists a cardinal


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.

Denition 4.17 We say that is a successor cardinal i there


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.

4.2 Cardinal arithmetic


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)

Figure 4.1 ordered by 


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}.

Denition 4.24 For all cardinals and ,


= | |
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.

Earlier, we established that + |P()|. Thus, the following is


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<

Let us point out that = 0 < . The fact that can be


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

By the end of this chapter, you should be able to explain the


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().
<

Proof of claim Dene g with dom(g) = cf() by


g() = sup f ().
<

By the denition of conality, if < cf(), then g() < . That


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()

Continuing with the proof of Lemma 4.31, let g : cf() be


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()

We have shown that the composition


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 () = < .
< 

In particular, sup(f []) < . This shows that cf() . But


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<

This proves our claim. We conclude that


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<

1. Prove that < 2 is countable.


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 = <  .

Show that <


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< 

Hint: Consider the function


A  A n | n < 
76 Cardinality
We remark that, from Exercise 4.10, it is immediate that
 
 
 
2 =  2n  ,
n< 

which is usually abbreviated



2 = 2n .
n<

Exercise 4.11 Prove that, for every 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.

1. Show that if < 1 , then < 1 .


2. Show that 1
 = 1 . 
3. Show that < 1 | = is uncountable.
Exercise 4.13 Prove that, for every limit ordinal ,
cf( ) = cf().

Exercise 4.14 (CantorBernsteinSchroeder theorem) The proof


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.

Exercise 4.15 As in Exercises 2.10 and 2.13, let E be the equiv-


alence relation on P() dened by
xEy x  y is nite.
Prove that P()/E has cardinality 20 .

Exercise 4.16 (Zorns lemma) Let (P, ) be a partial ordering.


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.

Tn is the set of all functions f from 2 to 2 with the property


that, for all a, b 2, if a  n = b  n, then f (a) = f (b).
If f, g Tn and a 2, then

(f n g)(a) = 1 f (a) = 1 or g(a) = 1,

(f n g)(a) = 1 f (a) = 1 and g(a) = 1

and
(n f )(a) = 1 f (a) = 0.

For every a 2, 0(a) = 0 and 1(a) = 1.

Notice that 0, 1 Tn for every n < .


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, . . . )

and, if f happens to be an element of T1 , then f can be thought


4.3 Conality 79
of as the simpler truth table

0 f (0, . . . )

1 f (1, . . . )

1. How many elements does Tn have? Explain.


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

As you might expect, trees play important roles in many parts


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.

5.1 Topology fundamentals


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).

We also say that d is a metric on S.

The last clause in the denition is called the triangle inequality.


The most important example of a metric space (S, d) has

S=R

and
d = |x y|.

In this context, |x y| means the absolute value of the dierence


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|.

The following lemma is immediate from these facts.

Lemma 5.4 The usual distance function for R is a metric on


R.

Notice that each open interval (a, b) of R has the form

{x R | |x c| < r}

for some c R (the center) and positive r R (the radius). Just


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

Figure 5.1 A basic open neighborhood Ns and x Ns

5.2 The Baire space


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

set of innite sequences of natural numbers).


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<

Lemma 5.8 {U | U is an open subset of } is a topology on


.

This is the Baire topological space.


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}.

However, {x} is not open since, for every s < ,


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 .

Lemma 5.13 For every A , the following are equivalent.


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?

Exercise 5.2 Let A . Put


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.

Exercise 5.3 A topological space (S, T ) is said to be a Lindel


of
space i for every F T , if

S= F,

then there is a countable G F such that



S= G.

Prove that the Baire space is a Lindelof space.

Exercise 5.4 A topological space (S, T ) is said to be compact


i for every F T , if

S= F,

then there is a nite G F such that



S= G.

1. Prove that the Baire space is not compact.


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.

Use recursion to dene x(n) in terms of x  n. Along the recur-


sion, maintain that there is no nite G F such that

G Nxn 2.

Exercise 5.5 If (S, T ) is a topological space and D S, then


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.

1. Prove that D is dense.


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<

and, for every n < , Un is open and dense.


Exercise 5.7 We need three denitions before stating the exer-
cise. Consider an arbitrary metric space (S, d).

Let xi | i <  be a sequence of elements of S and y S.


We say that xi | i <  converges to y and write
lim xi = y
i

i for every r R, if r > 0, then there exists i < such that,


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.

Exercise 5.8 (Baire category theorem) 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 is dense in the Baire space.

Exercise 5.9 A tree T on is called perfect i for every r T ,


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.

Exercise 5.11 (Cantor perfect set theorem) Let C be a closed


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 .
<

whenever is a limit ordinal.


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 .

Exercise 5.12 Prove by induction that, for every < 1 , there


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.

Exercise 5.13 Prove that there exists a set A such that,


for every non-empty perfect tree T , neither [T ] A nor [T ]
A. Hint: Let

T | < 20 

enumerate the non-empty perfect trees. Recursively dene

x | < 20 

and
y | < 20 

such that, for every < 20 ,

y [T ] {x | < }

and
x [T ] {y | }.

Then let A = {x | < 20 }.


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.

5.3 Illfounded and wellfounded trees


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

1, 0 2, 0 3, 0

0 1 2 j 3


DD
DD vvv jjjjjjj
v j
DD vv jjjj
DD vvvjjjjjj

jvjj

Figure 5.2 An innite tree with no innite branches

Proof of Theorem 5.17 We will need the following notation. Given


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

where j < and, for every 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.

Denition 5.20 If T is a tree on , then T is wellfounded i


[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.

This terminology makes sense if you think of the tree as growing


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.

Denition 5.21 Let T be a tree on . A rank function for T is


function f with domain T such that, for all s, t T ,

f (s) is an ordinal and


if s  t, then f (s) > f (t).

Lemma 5.22 If T is a tree on and T has a rank function,


then T is wellfounded.

Proof For contradiction, suppose that [T ] = . Let x [T ]. Then

< f (x  2) < f (x  1) < f (x  0)

is an innite descending sequence of ordinals.

The converse of Lemma 5.22 is also true. In fact, if T is a well-


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

such that, for every s T ,

if s is a terminal node of T , then T (s) = 0, and


if s is not a terminal node of T , then
T (s) = sup ({T (t) + 1 | t  s}) .

In particular, T is a rank function for T .

We call T the rank function associated to T . The two clauses


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.

Corollary 5.24 Let T be a tree on . Then T is wellfounded i


T has a rank function.

Denition 5.25 If T is a non-empty wellfounded tree on , then


we let the rank of T be
rank(T ) = T ( ).

Do not confuse height with rank. Every tree on has height at


most . But only wellfounded trees have ranks, and these ranks
are sometimes strictly greater than .

Example For every n < , the trees


{s | dom(s) n and s(m) = 1 for every m < dom(s)}

and

{s | dom(s) n and s(m) < for every m < dom(s)}


both have rank n.

Example Figure 5.2 shows an example of a wellfounded tree of


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

Figure 5.3 A wellfounded tree of rank (labels are ranks)

Example If we let T be the tree in Figure 5.2 and dene


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

Figure 5.4 A wellfounded tree of rank + 1 (labels are ranks)

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

2. Find an example of a set C such that C is a clopen subset of


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 ).

Exercise 5.18 (KleeneBrouwer ordering) Dene a relation <KB


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 ) < .

Exercise 5.22 Let B be the Boolean algebra of clopen subsets


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

Figure 5.5 The run b


I a0 a1 a2 a3

II (a0 ) (a0 , (a0 ), a1 ) (a0 , (a0 ), a1 , (a0 , (a0 ), a1 ), a2 )

Figure 5.6 The run a


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 .

Now choose a and b by recursion on < 20 as follows. Assume


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.

Exercise 5.25 Prove there is an undetermined set B such


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.

Exercise 5.26 Let C be a closed subset of . Prove that C


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

For x 2, we say that p rejects x i


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.

5.5 Ramsey theory


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)},

x(m) < x(n)


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().

Prove that there is no uncountable set that is homogeneous for


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.

5.6 Trees of uncountable height


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.

Theorem 5.30 (Aronszajn) There exists a subtree T of < 1

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

then T witnesses the statement of the theorem. To see this, rst


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 .

Exercise 5.40 Let T be the tree constructed in the proof of


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.

6.1 Denitions and examples


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.

Let z be the zero of Q in the concatenation 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.

Theorem 6.5 also implies that

(/2, /2) Q  Q,

which we already mentioned.

1 Formally, the underlying set of the strict linear ordering R


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<

then it is easy to verify that


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.

6.3 Real numbers


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.

Moreover, if (B  , B  ) has the same four properties as (B, B ),


then there is an isomorphism

f : (B, B )  (B  , B  )

such that, for every x A,

f (x) = x.

Proof The moreover part of Theorem 6.10 is proved by observing


that, for every y B,

y = the lub in (B, B ) of {x A | x B y}

and setting

f (y) = the lub in (B  , B  ) of {x A | x B y}.

The details are left to the reader. See Exercise 6.4.


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
 
 
x L x
(A, A ) _ _ _ _ _ _ _ _ _ _ _ _/ (C, C )
isomorphism

Figure 6.2 Plan for the proof of Theorem 6.10


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}.

This example gives a big hint for what is going on.


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.

To describe the relationship between (A, A ) and (B, B ) in


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).

Here 1 = {0} = {1, 2, 3, . . . }. Dene a relation E on S by

(a, b) E (c, d) ad = bc.


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

Filters and ideals come up in just about every area of modern


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.

7.1 Motivation and denitions


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.

Denition 7.1 Let X be a non-empty set and F P(X). Then


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.

To understand the motivation for lters, one can paraphrase the


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.

Denition 7.6 Let I be an ideal over X. Then I is a prime


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

then there exists i < n such that Ai F.


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 .
<

It is easy to check that G is a lter because it is the union of a


-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 .
<

As in the limit case, we see that G is a lter. Clearly, F = G0 G.


Finally, G is an ultralter because, for every < , either
A G+1 G
or
(X A ) G+1 G.

Exercise 7.4 Consider the special case of Theorem 7.7 in which


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 =

Immediately after Exercise 7.5, there is a long discussion of the


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)

..
.

Figure 7.1 An ultrapower of (, <) by a non-principal ultralter

Clearly,

{n < | f (n) < b} = {n < | f (n) = a}.
a<b

By Exercise 7.3, there exists a < b such that

{n < | f (n) = a} F.

By denition, this just says that

[f ] = [n  a].

Finally, observe that a c.


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.

7.2 Club and stationary sets


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<

but is not a successor ordinal.


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<

Then < because cf() > . Since C and D are closed,


= 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.

Exercise 7.7 Let be an uncountable regular cardinal and


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.

Exercise 7.8 Let be an uncountable regular cardinal. Prove


that if < and C | <  is a sequence of club subsets of ,
then the set

{C | < }

is club in . Hint: Use induction on < . The successor case,


= +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 | < }.

Exercise 7.9 Let be a regular cardinal. Give an example of a


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 .
<

This is possible by Exercise 7.8. If is a limit ordinal, then let


= sup .
<

That completes the recursive denition of  | < . By induc-


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.

Claims 7.13.1 and 7.13.2 allow us to dene a function g : 1


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

Boolean algebras were dened just before Exercises 2.12. That


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

There are several other undergraduate textbooks on set theory.


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

-least element, 24 compact topological space, 92, 102


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