AN INTRODUCTION TO SET THEORY

Professor William A. R. Weiss

September 6, 2006
2
Contents

0 Introduction 7

1 LOST 11

2 FOUND 19

3 The Axioms of Set Theory 23

4 The Natural Numbers 31

5 The Ordinal Numbers 41

6 Orderings 53

7 Cardinality 63

8 There Is Nothing Real About The Real Numbers 69

9 The Universe 77

3
4 CONTENTS

10 Reflection 83

11 Elementary Submodels 93

12 Constructibility 105

13 Appendices 121

.1 The Axioms of ZFC . . . . . . . . . . . . . . . . . . . . . . . . 121

.2 Tentative Axioms . . . . . . . . . . . . . . . . . . . . . . . . . 122

CONTENTS 5

Preface

These notes for a graduate course in set theory are on their way to be-
coming a book. They originated as handwritten notes in a course at the
University of Toronto given by Prof. William Weiss. Cynthia Church pro-
duced the first electronic copy in December 2002. James Talmage Adams
produced the copy here in February 2005. Chapters 1 to 9 are close to fi-
nal form. Chapters 10, 11, and 12 are quite readable, but should not be
considered as a final draft. One more chapter will be added.
6 CONTENTS
Chapter 0

Introduction

Set Theory is the true study of infinity. This alone assures the subject of a
place prominent in human culture. But even more, Set Theory is the milieu
in which mathematics takes place today. As such, it is expected to provide
a firm foundation for the rest of mathematics. And it does—up to a point;
we will prove theorems shedding light on this issue.

Because the fundamentals of Set Theory are known to all mathemati-

cians, basic problems in the subject seem elementary. Here are three simple
statements about sets and functions. They look like they could appear on a
homework assignment in an undergraduate course.

1. For any two sets X and Y , either there is a one-to-one function from
X into Y or a one-to-one function from Y into X.
2. If there is a one-to-one function from X into Y and also a one-to-one
function from Y into X, then there is a one-to-one function from X
onto Y .
3. If X is a subset of the real numbers, then either there is a one-to-one
function from the set of real numbers into X or there is a one-to-one
function from X into the set of rational numbers.

They won’t appear on an assignment, however, because they are quite dif-

7
8 CHAPTER 0. INTRODUCTION

ficult to prove. Statement (2) is true; it is called the Schroder-Bernstein

Theorem. The proof, if you haven’t seen it before, is quite tricky but never-
theless uses only standard ideas from the nineteenth century. Statement (1)
is also true, but its proof needed a new concept from the twentieth century,
a new axiom called the Axiom of Choice.

Statement (3) actually was on a homework assignment of sorts. It was

the first problem in a tremendously influential list of twenty-three problems
posed by David Hilbert to the 1900 meeting of the International Congress of
Mathematicians. Statement (3) is a reformulation of the famous Continuum
Hypothesis. We don’t know if it is true or not, but there is hope that the
twenty-first century will bring a solution. We do know, however, that another
new axiom will be needed here. All these statements will be discussed later
in the book.

Although Elementary Set Theory is well-known and straightforward, the

modern subject, Axiomatic Set Theory, is both conceptually more difficult
and more interesting. Complex issues arise in Set Theory more than any
other area of pure mathematics; in particular, Mathematical Logic is used in
a fundamental way. Although the necessary logic is presented in this book,
it would be beneficial for the reader to have taken a prior course in logic
under the auspices of mathematics, computer science or philosophy. In fact,
it would be beneficial for everyone to have had a course in logic, but most
people seem to make their way in the world without one.

In order to introduce one of the thorny issues, let’s consider the set of
all those numbers which can be easily described, say in fewer then twenty
English words. This leads to something called Richard’s Paradox. The set

{x : x is a number which can be described

in fewer than twenty English words}
must be finite since there are only finitely many English words. Now, there
are infinitely many counting numbers (i.e., the natural numbers) and so there
must be some counting number (in fact infinitely many of them) which are not
in our set. So there is a smallest counting number which is not in the set. This
number can be uniquely described as “the smallest counting number which
cannot be described in fewer than twenty English words”. Count them—14
words. So the number must be in the set. But it can’t be in the set. That’s
 9

a contradiction. What is wrong here?

Our naive intuition about sets is wrong here. Not every collection of
numbers with a description is a set. In fact it would be better to stay away
from using languages like English to describe sets. Our first task will be to
build a new language for describing sets, one in which such contradictions
cannot arise.

We also need to clarify exactly what is meant by “set”. What is a set?

We do not know the complete answer to this question. Many problems are
still unsolved simply because we do not know whether or not certain objects
constitute a set or not. Most of the proposed new axioms for Set Theory are
of this nature. Nevertheless, there is much that we do know about sets and
this book is the beginning of the story.
10 CHAPTER 0. INTRODUCTION
Chapter 1

LOST

We construct a language suitable for describing sets.

The symbols:

variables v 0 , v1 , v2 , . . .
equality symbols =
membership symbol ∈
logical connectives ∧, ∨, ¬, →, ↔
quantifiers ∀, ∃
parentheses (, )

The atomic formulas are strings of symbols of the form:

(vi ∈ vj ) or (vi = vj )

The collection of formulas of set theory is defined as follows:

1. An atomic formula is a formula.

2. If Φ is any formula, then (¬Φ) is also a formula.
3. If Φ and Ψ are formulas, then (Φ ∧ Ψ) is also a formula.

11
12 CHAPTER 1. LOST

4. If Φ and Ψ are formulas, then (Φ ∨ Ψ) is also a formula.

5. If Φ and Ψ are formulas, then (Φ → Ψ) is also a formula.

6. If Φ and Ψ are formulas, then (Φ ↔ Ψ) is also a formula.

7. If Φ is a formula and vi is a variable, then (∀vi )Φ is also a formula.

8. If Φ is a formula and vi is a variable, then (∃vi )Φ is also a formula.

Furthermore, any formula is built up this way from atomic formulas and a
finite number of applications of the inferences 1 through 8.

Now that we have specified a language of set theory, we could specify

a proof system. We will not do this here—see n different logic books for
n different proof systems. However, these are essentially all the same—
satisfying the completeness theorem (due to K. Gödel) which essentially says
that any formula either has a proof or it has an interpretation in which it
is false (but not both!). In all these proof systems we have the usual logical
equivalences which are common to everyday mathematics. For example:

For any formulas Φ and Ψ:

(¬(¬(Φ))) is equivalent to Φ;
(Φ ∧ Ψ) is equivalent to ¬((¬Φ) ∨ (¬Ψ));
(Φ → Ψ) is equivalent to ((¬Φ) ∨ Ψ);
(Φ ↔ Ψ) is equivalent to ((Φ → Ψ) ∧ (Ψ → Φ));
((∃vi )Φ is equivalent to (¬(∀vi )(¬Φ)); and,
(Φ ↔ Ψ) is equivalent to (Ψ ↔ Φ).

The complete collection of subformulas of a formula Φ is defined as fol-

lows:

1. Φ is a subformula of Φ;

2. If (¬Ψ) is a subformula of Φ, then so is Ψ;

3. If (Θ ∧ Ψ) is a subformula of Φ, then so are Θ and Ψ;

 13

4. If (Θ ∨ Ψ) is a subformula of Φ, then so are Θ and Ψ;

5. If (Θ → Ψ) is a subformula of Φ, then so are Θ and Ψ;

6. If (Θ ↔ Ψ) is a subformula of Φ, then so are Θ and Ψ;

7. If (∀vi )Ψ is a subformula of Φ and vi is a variable, then Ψ is a subformula

of Φ; and,

8. If (∃vi )Ψ is a subformula of Φ and vi is a variable, then Ψ is a subformula

of Φ.

To say that a variable vi occurs bound in a formula Φ means one of the

following two conditions holds:

1. For some subformula Ψ of Φ, (∀vi )Ψ is a subformula of Φ; or,

2. For some subformula Ψ of Φ, (∃vi )Ψ is a subformula of Φ.

The result, Φ∗ , of substituting the variable vj for each bound occurrence

of the variable vi in the formula Φ is defined by constructing a Ψ∗ for each
subformula Ψ of Φ as follows:

1. If Ψ is atomic, then Ψ∗ is Ψ;

2. If Ψ is (¬Θ) for some formula Θ, then Ψ∗ is (¬Θ∗ );

3. If Ψ is (Γ ∧ Θ) for some formula Θ, then Ψ∗ is (Γ∗ ∧ Θ∗ );

4. If Ψ is (Γ ∨ Θ) for some formula Θ, then Ψ∗ is (Γ∗ ∨ Θ∗ );

5. If Ψ is (Γ → Θ) for some formula Θ, then Ψ∗ is (Γ∗ → Θ∗ );

6. If Ψ is (Γ ↔ Θ) for some formula Θ, then Ψ∗ is (Γ∗ ↔ Θ∗ );

7. If Ψ is (∀vk )Θ for some formula Θ then Ψ∗ is just (∀vk )Θ∗ if k 6= i, but

if k = i then Ψ∗ is (∀vj )Γ where Γ is the result of substituting vj for
each occurrence of vi in Θ; and,
14 CHAPTER 1. LOST

8. If Ψ is (∃vk )Θ for some formula Θ then Ψ∗ is just (∃vk )Θ∗ ifk 6= i, but
if k = i then Ψ∗ is (∃vj )Γ where Γ is the result of substituting vj for
each occurrence of vi in Θ.

That a variable vi occurs free in a formula Φ means that at least one of

the following is true:

1. Φ is an atomic formula and vi occurs in Φ;

2. Φ is (¬Ψ), Ψ is a formula and vi occurs free in Ψ;

3. (Θ ∧ Ψ), Θ and Ψ are formulas and vi occurs free in Θ or occurs free

in Ψ;

4. Φ is (Θ ∨ Ψ), Θ and Ψ are formulas and vi occurs free in Θ or occurs

free in Ψ;

5. Φ is (Θ → Ψ), Θ and Ψ are formulas and vi occurs free in Θ or occurs

free in Ψ;

6. Φ is (Θ ↔ Ψ), Θ and Ψ are formulas and vi occurs free in Θ or occurs

free in Ψ;

7. Φ is (∀vj )Ψ and Ψ is a formula and vi occurs free in Ψ and i 6= j; or,

8. Φ is (∃vj )Ψ and Ψ is a formula and vi occurs free in Ψ and i 6= j.

As in the example below, a variable can occur both free and bound in a
formula. However, notice that if a variable occurs in a formula at all it must
occur either free, or bound, or both (but not at the same occurrence).

We define the important notion of the substitution of a variable vj for

each free occurrence of the variable vi in the formula Φ. This procedure is as
follows.

1. Substitute a new variable vl for all bound occurrences of vi in Φ.

2. Substitute another new variable vk for all bound occurrences of vj in

the result of (1).
 15

3. Directly substitute vj for each occurrence of vi in the result of (2).

Example. Let us substitute v2 for all free occurrences of v1 in the formula

((∀v1 )((v1 = v2 ) → (v1 ∈ v0 )) ∧ (∃v2 )(v2 ∈ v1 ))

The steps are as follows.

1. ((∀v1 )((v1 = v2 ) → (v1 ∈ v0 )) ∧ (∃v2 )(v2 ∈ v1 ))

2. ((∀v3 )((v3 = v2 ) → (v3 ∈ v0 )) ∧ (∃v2 )(v2 ∈ v1 ))

3. ((∀v3 )((v3 = v2 ) → (v3 ∈ v0 )) ∧ (∃v4 )(v4 ∈ v1 ))

4. ((∀v3 )((v3 = v2 ) → (v3 ∈ v0 )) ∧ (∃v4 )(v4 ∈ v2 ))

For the reader who is new to this abstract game of formal logic, step (2)
in the substitution proceedure may appear to be unnecessary. It is indeed
necessary, but the reason is not obvious until we look again at the example
to see what would happen if step (2) were omitted. This step essentially
changes (∃v2 )(v2 ∈ v1 ) to (∃v4 )(v4 ∈ v1 ). We can agree that each of these
means the same thing, namely, “v1 is non-empty”. However, when v2 is
directly substituted into each we get something different: (∃v2 )(v2 ∈ v2 ) and
(∃v4 )(v4 ∈ v2 ). The latter says that “v2 is non-empty” and this is, of course
what we would hope would be the result of substituting v2 for v1 in “v1 is
non-empty”. But the former statement, (∃v2 )(v2 ∈ v2 ), seems quite different,
making the strange assertion that “v2 is an element of itself”, and this is not
what we have in mind. What caused this problem? An occurrence of the
variable v2 became bound as a result of being substituted for v1 . We will not
allow this to happen. When we substitute v2 for the free v1 we must ensure
that this freedom is preserved for v2 .

For a formula Φ and variables vi and vj , let Φ(vi |vj ) denote the formula
which results from substituting vj for each free occurance of vi . In order
to make Φ(vi |vj ) well defined, we insist that in steps (1) and (2) of the
substitution process, the first new variable available is used. Of course, the
use of any other new variable gives an equivalent formula. In the example, if
Φ is the formula on the first line, then Φ(v1 |v2 ) is the formula on the fourth
line.
16 CHAPTER 1. LOST

As a simple application we can show how to express “there exists a unique

element”. For any formula Φ of the language of set theory we denote by
(∃!vj )Φ the formula
((∃vj )Φ ∧ (∀vj )(∀vl )((Φ ∧ Φ(vj |vl )) → (vj = vl )))
where vl is the first available variable which does not occur free in Φ. The
expression (∃!vj ) can be considered as an abbreviation in the language of
set theory, that is, an expression which is not actually part of the language.
However, whenever we have a formula containing this expression, we can
quickly convert it to a proper formula of the language of set theory.

A class is just a string of symbols of the form {vi : Φ} where vi is a

variable and Φ is a formula. Two important and well-known examples are:
{v0 : (¬(v0 = v0 ))}
which is called the empty set and is usually denoted by ∅, and
{v0 : (v0 = v0 )}
which is called the universe and is usually denoted by V.

A term is defined to be either a class or a variable. Terms are the names

for what the language of set theory talks about. A grammatical analogy is
that terms correspond to nouns and pronouns—classes to nouns and variables
to pronouns. Continuing the analogy, the predicates, or verbs, are = and ∈.
The atomic formulas are the basic relationships among the predicates and
the variables.

We can incorporate classes into the language of set theory by showing

how the predicates relate to them. Let Ψ and Θ be formulas of the language
of set theory and let vj , vk and vl be variables. We write:
vk ∈ {vj : Ψ} instead of Ψ(vj |vk )
vk = {vj : Ψ} instead of (∀vl )((vl ∈ vk ) ↔ Ψ(vj |vl ))
{vj : Ψ} = vk instead of (∀vl )(Ψ(vj |vl ) ↔ (vl ∈ vk ))
{vj : Ψ} = {vk : Θ} instead of (∀vl )(Ψ(vj |vl ) ↔ Θ(vk |vl ))
{vj : Ψ} ∈ vk instead of (∃vl )((vl ∈ vk ) ∧ (∀vj )((vj ∈ vl ) ↔ Ψ))
{vj : Ψ} ∈ {vk : Θ} instead of (∃vl )(Θ(vk |vl ) ∧ (∀vj )((vj ∈ vl ) ↔ Ψ))
 17

whenever vl is neither vj nor vk and occurs in neither Ψ nor Θ.

We can now show how to express, as a proper formula of set theory,

the substitution of a term t for each free occurrence of the variable vi in the
formula Φ. We denote the resulting formula of set theory by Φ(vi |t). The
case when t is a variable vj has already been discussed. Now we turn our
attention to the case when t is a class {vj : Ψ} and carry out a proceedure
similar to the variable case.

1. Substitute the first available new variable for all bound occurrences of
vi in Φ.

2. In the result of (1), substitute, in turn, the first available new variable
for all bound occurrences of each variable which occurs free in Ψ.

3. In the result of (2), directly substitute {vj : Ψ} for vi into each atomic
subformula in turn, using the table above.

For example, the atomic subformula (vi ∈ vk ) is replaced by the new subfor-
mula
(∃vl )((vl ∈ vk ) ∧ (∀vj )((vj ∈ vl ) ↔ Ψ))
where vl is the first available new variable. Likewise, the atomic subformula
(vi = vi ) is replaced by the new subformula

(∀vl )(Ψ(vj |vl ) ↔ Ψ(vj |vl ))

where vl is the first available new variable (although it is not important to

change from vj to vl in this particular instance).
18 CHAPTER 1. LOST
Chapter 2

FOUND

The language of set theory is very precise, but it is extremely difficult for us
to read mathematical formulas in that language. We need to find a way to
make these formula more intelligible.

In order to avoid the inconsistencies associated with Richard’s paradox,

we must ensure that the formula Φ in the class {vj : Φ} is indeed a proper
formula of the language of set theory—or, at least, can be converted to a
proper formula once the abbreviations are eliminated. It is not so important
that we actually write classes using proper formulas, but what is important is
that whatever formula we write down can be converted into a proper formula
by eliminating abbreviations and slang.

We can now relax our formalism if we keep the previous paragraph in

mind. Let’s adopt these conventions.

1. We can use any letters that we like for variables, not just v0 , v1 , v2 , . . . .

2. We can freely omit parentheses and sometimes use brackets ] and [

instead.

3. We can write out “and” for “∧”, “or” for “∨”, “implies” for “→” and
use the “if...then...” format as well as other common English expres-
sions for the logical connectives and quantifiers.

19
20 CHAPTER 2. FOUND

4. We will use the notation Φ(x, y, w1 , . . . , wk ) to indicate that all free

variables of Φ lie among x, y, w1 , . . . , wk . When the context is clear we
use the notation Φ(x, t, w1 , . . . , wk ) for the result of substituting the
term t for each free occurrence of the variable y in Φ, i.e., Φ(y|t).

5. We can write out formulas, including statements of theorems, in any

way easily seen to be convertible to a proper formula in the language
of set theory.

For any terms r, s, and t, we make the following abbreviations of formulas.

(∀x ∈ t)Φ for (∀x)(x ∈ t → Φ)

(∃x ∈ t)Φ for (∃x)(x ∈ t ∧ Φ)
s∈/t for ¬(s ∈ t)
s 6= t for ¬(s = t)
s⊆t for (∀x)(x ∈ s → x ∈ t)

Whenever we have a finite number of terms t1 , t2 , . . . , tn the notation

{t1 , t2 , . . . , tn } is used as an abbreviation for the class:

{x : x = t1 ∨ x = t2 ∨ · · · ∨ x = tn }.

Furthermore, {t : Φ} will stand for {x : x = t ∧ Φ}, while {x ∈ t : Φ} will

represent {x : x ∈ t ∧ Φ}.

We also abbreviate the following important classes.

Union s ∪ t for {x : x ∈ s ∨ x ∈ t}
Intersection s ∩ t for {x : x ∈ s ∧ x ∈ t}
Difference s\t for {x : x ∈ s ∧ x ∈ / t}
Symmetric Difference s4t for (s \ t) ∪ (t \ s)
Ordered Pair hs, ti for {{s}, {s, t}}
Cartesian Product s × t for {p : ∃x ∃y (x ∈ s ∧ y ∈ t ∧ p = hx, yi)}
Domain dom(f ) for {x : ∃y hx, yi ∈ f }
Range rng(f ) for {y : ∃x hx, yi ∈ f }
 21

Image f 00 A for {y : ∃x ∈ A hx, yi ∈ f }

Inverse Image f ←B for {x : ∃y ∈ B hx, yi ∈ f }
Restriction f |A for {p : p ∈ f ∧ ∃x ∈ A ∃y p = hx, yi}
Inverse f −1 for {p : ∃x ∃y hx, yi ∈ f ∧ hy, xi = p}
These latter abbreviations are most often used when f is a function. We
write
f is a function
for
∀p ∈ f ∃x ∃y p = hx, yi ∧ (∀x)(∃y hx, yi ∈ f → ∃!y hx, yi ∈ f )
and whenever f is a function we write
f : X → Y for f is a function ∧ dom(f ) = X ∧ rng(f ) ⊆ Y
f is one − to − one for ∀y ∈ rng(f ) ∃!x hx, yi ∈ f
f is onto Y for Y = rng(f )

We also use the terms injection (for a one-to-one function), surjection (for
an onto function), and bijection (for both properties together).

Russell’s Paradox

The following is a theorem.

¬∃z z = {x : x ∈
/ x}.
The proof of this is simple. Just ask whether or not z ∈ z.

The paradox is only for the naive, not for us. {x : x ∈/ x} is a class—just
a description in the language of set theory. There is no reason why what it
describes should exist. In everyday life we describe many things which don’t
exist, fictional characters for example. Bertrand Russell did exist and Peter
Pan did not, although they each have descriptions in English. Although
Peter Pan does not exist, we still find it worthwhile to speak about him. The
same is true in mathematics.

Upon reflection, you would say that in fact, nothing is an element of itself
so that
{x : x ∈
/ x} = {x : x = x} = V
22 CHAPTER 2. FOUND

and so Russell’s paradox leads to:

¬∃z z = V.

It seems we have proved that the universe does not exists. A pity!

The mathematical universe fails to have a mathematical existence in the

same way that the physical universe fails to have a physical existence. The
things that have a physical existence are exactly the things in the universe,
but the universe itself is not an object in the universe. This does bring up
an important point—do any of the usual mathematical objects exist? What
about the other things we described as classes? What about ∅? Can we
prove that ∅ exists?

Actually, we can’t; at least not yet. You can’t prove very much, if you
don’t assume something to start. We could prove Russell’s Paradox because,
amazingly, it only required the basic rules of logic and required nothing
mathematical—that is, nothing about the “real meaning” of ∈. Continuing
from Russell’s Paradox to “¬∃z z = V” required us to assume that “∀x x ∈ /
x”—not an unreasonable assumption by any means, but a mathematical
assumption none the less. The existence of the empty set “∃z z = ∅” may
well be another necessary assumption.

Generally set theorists, and indeed all mathematicians, are quite willing
to assume anything which is obviously true. It is, after all, only the things
which are not obviously true which need some form of proof. The problem,
of course, is that we must somehow know what is “obviously true”. Naively,
“∃z z = V” would seem to be true, but it is not and if it or any other
false statement is assumed, all our proofs become infected with the virus of
inconsistency and all of our theorems become suspect.

Historically, considerable thought has been given to the construction of

the basic assumptions for set theory. All of mathematics is based on these
assumptions; they are the foundation upon which everything else is built.
These assumptions are called axioms and this system is called the ZFC Axiom
System. We will begin to study it in the next chapter.
Chapter 3

The Axioms of Set Theory

We will explore the ZFC Axiom System. Each axiom should be “obviously
true” in the context of those things that we desire to call sets. Because we
cannot give a mathematical proof of a basic assumption, we must rely on
intuition to determine truth, even if this feels uncomfortable. Beyond the
issue of truth is the question of consistency. Since we are unable to prove
that our assumptions are true, can we at least show that together they will
not lead to a contradiction? Unfortunately, we cannot even do this—it is
ruled out by the famous incompleteness theorems of K. Gödel. Intuition is
our only guide. We begin.

We have the following axioms:

The Axiom of Equality ∀x ∀y [x = y → ∀z (x ∈ z ↔ y ∈ z)]

The Axiom of Extensionality ∀x ∀y [x = y ↔ ∀u (u ∈ x ↔ u ∈ y)]
The Axiom of Existence ∃z z = ∅
The Axiom of Pairing ∀x ∀y ∃z z = {x, y}

Different authors give slightly different formulations of the ZFC axioms.

All formulations are equivalent. Some authors omit the Axiom of Equality
and Axiom of Existence because they are consequences of the usual logical
background to all mathematics. We include them for emphasis. Redundancy
is not a bad thing and there is considerable redundancy in this system.

23
24 CHAPTER 3. THE AXIOMS OF SET THEORY

The following theorem gives some results that we would be quite willing
to assume, were they not to follow from the axioms. The first three parts are
immediate consequences of the Axiom of Extentionality.
Theorem 1.
1. ∀x x = x.

2. ∀x ∀y x = y → y = x.

3. ∀x ∀y ∀z [(x = y ∧ y = z) → x = z].

4. ∀x ∀y ∃z z = hx, yi.

5. ∀u ∀v ∀x ∀y [hu, vi = hx, yi ↔ (u = x ∧ v = y)].

Exercise 1. Prove parts (4) and (5) of Theorem 1

We now assert the existence of unions and intersections. No doubt the

reader has experienced a symmetry between these two concepts. Here how-
ever, while the Union Axiom is used extensively, the Intersection Axiom is
redundant and is omitted in most developments of the subject. We include
it here because it has some educational value (see Exercise 4).

The Union Axiom ∀x [x 6= ∅ → ∃z z = {w : (∃y ∈ x)(w ∈ y)}]

S
The class {w : (∃y ∈ x)(w ∈ y)} is abbreviated as x and called the “big
union”.

The Intersection Axiom ∀x [x 6= ∅ → ∃z z = {w : (∀y ∈ x)(w ∈ y)}

T
The class {w : (∀y ∈ x)(w ∈ y)} is abbreviated as x and called the “big
intersection”.

The Axiom of Foundation ∀x [x 6= ∅ → (∃y ∈ x)(x ∩ y = ∅)]

This axiom, while it may be “obviously true”, is not certainly obvious. Let’s
investigate what it says: suppose there were a non-empty x such that (∀y ∈
x)(x ∩ y 6= ∅). For any y ∈ x we would be able to get z1 ∈ y ∩ x. Since z1 ∈ x
we would be able to get z2 ∈ z1 ∩ x. The process continues forever:

· · · ∈ z3 ∈ z2 ∈ z1 ∈ y ∈ x
 25

We wish to rule out such an infinite regress. We simply choose not to have
them; we want our sets to be founded: each such sequence should eventu-
ally end with the empty set. Hence the name of the axiom, which is also
known as the Axiom of Regularity. It is nevertheless best understood by its
consequences.

Theorem 2.

1. ∀x ∀y ∃z z = x ∪ y.

2. ∀x ∀y ∃z z = x ∩ y.

3. ∀u u ∈
/ u.

4. ∀u ∀v u ∈ v → v ∈
/ u.

Exercise 2. Prove Theorem 2.

S
Let f (x) denote the class {y : hx, yi ∈ f }.

Exercise 3. Suppose f is a function and x ∈ dom(f ). Prove that

hx, yi ∈ f iff y = f (x).

Suppose that x is a set and that there is some way of removing each
element u ∈ x and replacing u with some element v. Would the result be
a set? Well, of course—provided there are no tricks here. That is, there
should be a well defined replacement procedure which ensures that each u is
replaced by only one v. This well defined procedure should be described by
a formula, Φ, in the language of set theory. We can guarantee that each u is
replaced by exactly one v by insisting that ∀u ∈ x ∃!v Φ(x, u, v).

We would like to obtain an axiom, written in the language of set theory

stating that for each set x and each such formula Φ we get a set z. However,
this is impossible. We cannot express “for each formula” in the language of
set theory—in fact this formal language was designed for the express purpose
of avoiding such expressions which bring us perilously close to Richard’s
Paradox.
26 CHAPTER 3. THE AXIOMS OF SET THEORY

The answer to this conundrum is to utilise not just one axiom, but in-
finitely many—one axiom for each formula of the language of set theory.
Such a system is called an axiom scheme.

The Replacement Axiom Scheme

For each formula Φ(x, u, v, w1 , . . . , wn ) of the language of set theory, we have

the axiom:
∀w1 . . . ∀wn ∀x [∀u ∈ x ∃!v Φ → ∃z z = {v : ∃u ∈ x Φ}]

Note that we have allowed Φ to have w1 , . . . , wn as parameters, that is,

free variables which may be used to specify various procedures in various
contexts within a mathematical proof. This is illustrated by the following
theorem.
Theorem 3. ∀x ∀y ∃z z = x × y.

Proof. From Theorem 1 parts (4) and (5), for all t ∈ y we get
∀u ∈ x ∃!v v = hu, ti.
We now use Replacement with the formula “Φ(x, u, v, t)” as “v = hu, ti”; t
is a parameter. We obtain, for each t ∈ y:
∃q q = {v : ∃u ∈ x v = hu, ti}.
By Extensionality, in fact ∀t ∈ y ∃!q q = {v : ∃u ∈ x v = hu, ti}.

We again use Replacement, this time with the formula Φ(y, t, q, x) as

“q = {v : ∃u ∈ x v = hu, ti}”; here x is a parameter. We obtain:
∃r r = {q : ∃t ∈ y q = {v : ∃u ∈ x v = hu, ti}}
S
By the Union Axiom ∃z z = r and so we have:
z = {p : ∃q [q ∈ r ∧ p ∈ q]}
= {p : ∃q [(∃t ∈ y) q = {v : ∃u ∈ x v = hu, ti} ∧ p ∈ q]}
= {p : (∃t ∈ y)(∃q)[q = {v : ∃u ∈ x v = hu, ti} ∧ p ∈ q]}
= {p : (∃t ∈ y)p ∈ {v : ∃u ∈ x v = hu, ti}}
= {p : (∃t ∈ y)(∃u ∈ x) p = hu, ti}
=x×y
 27

Exercise 4. Show that the Intersection Axiom is indeed redundant.

It is natural to believe that for any set x, the collection of those elements
y ∈ x which satisfy some particular property should also be a set. Again, no
tricks—the property should be specified by a formula of the language of set
theory. Since this should hold for any formula, we are again led to a scheme.

The Comprehension Scheme

For each formula Φ(x, y, w1 , . . . , wn ) of the language of set theory, we have

the statement:

∀w1 . . . ∀wn ∀x ∃z z = {y : y ∈ x ∧ Φ(x, y, w1 , . . . , wn )}

This scheme could be another axiom scheme (and often is treated as such).
However, this would be unnecessary, since the Comprehension Scheme follows
from what we have already assumed. It is, in fact, a theorem scheme—that is,
infinitely many theorems, one for each formula of the language of set theory.
Of course we cannot write down infinitely many proofs, so how can we prove
this theorem scheme?

We give a uniform method for proving each instance of the scheme. So to

be certain that any given instance of the theorem scheme is true, we consider
the uniform method applied to that particular instance. We give this general
method below.

For each formula Φ(x, u, w1 , . . . , wn ) of the language of set theory we have:

Theorem 4. Φ

∀w1 . . . ∀wn ∀x ∃z z = {u : u ∈ x ∧ Φ}.

Proof. Apply Replacement with the formula Ψ(x, u, v, w1 , . . . , wn ) given by:

(Φ(x, u, w1 , . . . , wn ) → v = {u}) ∧ (¬Φ(x, u, w1 , . . . , wn ) → v = ∅)

28 CHAPTER 3. THE AXIOMS OF SET THEORY

to obtain:

∃y y = {v : (∃u ∈ x)[(Φ → v = {u}) ∧ (¬Φ → v = ∅)]}.

Note that {{u} : Φ(x, u,Sw1 , . . . , wn )} ⊆ y and the only other possible element
of y is ∅. Now let z = y to finish the proof.

Theorem 4 Φ can be thought of as infinitely many theorems, one for each

Φ. The proof of any one of those theorems can be done in a finite number
of steps, which invoke only a finite number f theorems or axioms. A proof
cannot have infinite length, nor invoke infinitely many axioms or lemmas.

We state the last of the “set behavior” axioms.

The Axiom of Choice

∀X [(∀x ∈ X ∀y ∈ X (x = y ↔ x ∩ y 6= ∅)) → ∃z (∀x ∈ X ∃!y y ∈ x ∩ z)]

In human language, the Axiom of Choice says that if you have a collection
X of pairwise disjoint non-empty sets, then you get a set z which contains
one element from each set in the collection. Although the axiom gives the
existence of some “choice set” z, there is no mention of uniqueness—there
are quite likely many possible sets z which satisfy the axiom and we are given
no formula which would single out any one particular z.

The Axiom of Choice can be viewed as a kind of replacement, in which

each set in the collection is replaced by one of its elements. This leads to the
following useful reformulation which will be used in Theorem 22.

Lemma. There is a choice function on any set of non-empty sets; i.e.,

[
∀X [∅ ∈
/ X → (∃f )(f : X → X ∧ (∀x ∈ X)(f (x) ∈ x))].

Proof. Given such an X, by Replacement there is a set

Y = {{x} × x : x ∈ X}
 29

which satisfies the hypothesis

S S of Choice. So, ∃z ∀y ∈ Y ∃!p p ∈
of the Axiom
y ∩ z. Let f = z ∩ ( Y ). Then f : X → X and each f (x) ∈ x.

We state the last of the “set creation” axioms.

The Power Set Axiom ∀x ∃z z = {y : y ⊆ x}

We denote {y : y ⊆ x} by P(x), called the power set of x. For reasons to

be understood later, it is important to know explicitly when the Power Set
Axiom is used. This completes the list of the ZFC Axiom System with one
exception to come later—higher analogues of the Axiom of Existence.
30 CHAPTER 3. THE AXIOMS OF SET THEORY
Chapter 4

The Natural Numbers

We now construct the natural numbers. That is, we will represent the natural
numbers in our universe of set theory. We will construct a number system
which behaves mathematically exactly like the natural numbers, with exactly
the same arithmetic and order properties. We will not claim that what we
construct are the actual natural numbers—whatever they are made of. But
we will take the liberty of calling our constructs “the natural numbers”. We
begin by taking 0 as the empty set ∅. We write
1 for {0}
2 for {0, 1}
3 for {0, 1, 2}
succ(x) for x ∪ {x}

We write “n is a natural number” for

[n = ∅ ∨ (∃l ∈ n)(n = succ(l))] ∧ (∀m ∈ n)[m = ∅ ∨ (∃l ∈ n)(m = succ(l))]
and write:
N for {n : n is a natural number}

The reader can gain some familiarity with these definitions by checking
that succ(n) ∈ N for all n ∈ N.

31
32 CHAPTER 4. THE NATURAL NUMBERS

We now begin to develop the basic properties of the natural numbers

by introducing an important concept. We say that a term t is transitive
whenever we have (∀x ∈ t)(x ⊆ t).
Theorem 5.
1. Each natural number is transitive.
2. N is transitive; i.e., every element of a natural number is a natural
number.

Proof. Suppose that (1) were false; i.e., some n ∈ N is not transitive, so that:

{k : k ∈ n and ¬(k ⊆ n)} =

6 ∅.

By Comprehension ∃x x = {k ∈ n : ¬(k ⊆ n)} and so by Foundation there

is y ∈ x such that y ∩ x = ∅. Note that since ∅ ∈
/ x and y ∈ n we have that
y = succ(l) for some l ∈ n. But since l ∈ y, l ∈ / x and so l ⊆ n. Hence
y = l ∪ {l} ⊆ n, contradicting that y ∈ x.

We also prove (2) indirectly; suppose n ∈ N with

{m : m ∈ n and m ∈ 6 ∅.
/ N} =

By Comprehension ∃x x = {m ∈ n : m ∈ / N} and so Foundation gives y ∈ x

such that y ∩ x = ∅. Since y ∈ n, we have y = succ(l) for some l ∈ n.
Since l ∈ y and y ∩ x = ∅ we must have l ∈ N. But then y = succ(l) ∈ N,
contradicting that y ∈ x.

Theorem 6. (Trichotomy of Natural Numbers)

Let m, n ∈ N. Exactly one of three situations occurs:

m ∈ n, n ∈ m, m = n.

Proof. That at most one occurs follows from Theorem 2. That at least one
occurs follows from this lemma.
 33

Lemma. Let m, n ∈ N.

1. If m ⊆ n, then either m = n or m ∈ n.

2. If m ∈
/ n, then n ⊆ m.

Proof. We begin the proof of (1) by letting S denote

{x ∈ N : (∃y ∈ N)(y ⊆ x and y 6= x and y ∈

/ x)}.

It will suffice to prove that S = ∅. We use an indirect proof—pick some

n1 ∈ S. If n1 ∩ S 6= ∅, Foundation gives us n2 ∈ n1 ∩ S with n2 ∩ (n1 ∩ S) = ∅.
By transitivity, n2 ⊆ n1 so that n2 ∩ S = ∅. Thus, we always have some
n ∈ S such that n ∩ S = ∅.

For just such an n, choose m ∈ N with m ⊆ n, m 6= n, and m ∈/ n. Using

Foundation, choose l ∈ n \ m such that l ∩ (n \ m) = ∅. Transitivity gives
l ⊆ n, so we must have l ⊆ m. We have l 6= m since l ∈ n and m ∈ / n.
Therefore we conclude that m \ l 6= ∅.

Using Foundation, pick k ∈ m \ l such that k ∩ (m \ l) = ∅. Transitivity

of m gives k ⊆ m and so we have k ⊆ l. Now, because l ∈ n we have l ∈ N
and l ∈
/ S so that either k = l or k ∈ l. However, k = l contradicts l ∈/ m
and k ∈ l contradicts k ∈ m \ l.

We prove the contrapositive of (2). Suppose that n is not a subset of m;

using Foundation pick l ∈ n \ m such that l ∩ (n \ m) = ∅. By transitivity,
l ⊆ n and hence l ⊆ m. Now by (1) applied to l and m, we conclude that
l = m. Hence m ∈ n.

These theorems show that “∈” behaves on N just like the usual ordering
“<” on the natural numbers. In fact, we often use “<” for “∈” when writing
about the natural numbers. We also use the relation symbols ≤, >, and ≥
in their usual sense.
34 CHAPTER 4. THE NATURAL NUMBERS

The next theorem scheme justifies ordinary mathematical induction. For

brevity let us write w
~ for w1 , . . . , wn .

For each formula Φ(v, w)

~ of the language of set theory we have:
Theorem 7. Φ

For all w,
~ if
∀n ∈ N [(∀m ∈ n Φ(m, w))
~ → Φ(n, w)]
~
then
∀n ∈ N Φ(n, w).
~

Proof. We will assume that the theorem is false and derive a contradiction.
~ and a fixed l ∈ N such that ¬Φ(l, w).
We have w ~

Let t be any transitive subset of N containing l (e.g., t = l ∪ {l}). By

Comprehension, ∃s s = {n ∈ t : ¬Φ(n, w)}. ~ By Foundation, we get y ∈ s
such that y ∩ s = ∅. Transitivity of t guarantees that (∀n ∈ y) Φ(n, w).
~ This,
in turn, contradicts that y ∈ s.

The statement ∀m ∈ n Φ(m, w)

~ in Theorem 7 Φ is usually called the
inductive hypothesis.
Exercise 5. Prove or disprove that for each formula Φ(v, w)
~ we have

∀w ~ → Φ(n, w))
~ [(∀m > n Φ(m, w)) ~ → ∀n ∈ N Φ(n, w)].
~

Recursion on N is a way of defining new terms (in particular, functions

with domain N). Roughly speaking, values of a function F at larger numbers
are defined in terms of the values of F at smaller numbers.

We begin with the example of a function F , where we set F (0) = 3 and

F (succ(n)) = succ(F (n)) for each natural number n. We have set out a short
recursive procedure which gives a way to calculate F (n) for any n ∈ N. The
reader may carry out this procedure a few steps and recognise this function
F as F (n) = 3 + n. However, all this is a little vague. What exactly is F ?
 35

In particular, is there a formula for calculating F ? How do we verify that F

behaves like we think it should?

In order to give some answers to these questions, let us analyse the ex-
ample. There is an implicit formula for the calculation of y = F (x) which
is
[x = 0 → y = 3] ∧ (∀n ∈ N)[x = succ(n) → y = succ(F (n))]
However the formula involves F , the very thing that we are trying to describe.
Is this a vicious circle? No—the formula only involves the value of F at a
number n less than x, not F (x) itself. In fact, you might say that the formula
doesn’t really involve F at all; it just involves F |x. Let’s rewrite the formula
as
[x = 0 → y = 3] ∧ (∀n)[x = succ(n) → y = succ(f (n))]
and denote it by Φ(x, f, y). Our recursive procedure is then described by

Φ(x, F |x, F (x)).

In order to describe F we use functions f which approximate F on initial

parts of its domain, for example f = {h0, 3i}, f = {h0, 3i, h1, 4i} or

f = {h0, 3i, h1, 4i, h2, 5i},

where each such f satisfies Φ(x, f |x, f (x)) for the appropriate x’s. We will
obtain F as the amalgamation of all these little f ’s. F is

{hx, yi : (∃n ∈ N)(∃f )[f : n → V ∧ f (x) = y ∧ ∀m ∈ n Φ(m, f |m, f (m))]}.

But in order to justify this we will need to notice that

(∀x ∈ N)(∀f )[(f : x → V) → ∃!y Φ(x, f, y)],

which simply states that we have a well defined procedure given by Φ.

Let us now go to the general context in which the above example will be
a special case. For any formula Φ(x, f, y, w)
~ of the language of set theory, we
denote by REC(Φ, N) the class

{hx, yi : (∃n ∈ N)(∃f )[f : n → V ∧ f (x) = y ∧ ∀m ∈ n Φ(m, f |m, f (m), w)]}.

~
36 CHAPTER 4. THE NATURAL NUMBERS

We will show, under the appropriate hypothesis, that REC(Φ, N) is the

unique function on N which satisfies the procedure given by Φ. This requires
a theorem scheme.

For each formula Φ(x, f, y, w)

~ of the language of set theory we have:

Theorem 8. Φ

For all w,
~ suppose that we have

(∀x ∈ N)(∀f )[(f : x → V) → ∃!y Φ(x, f, y, w)].

~

Then, letting F denote REC(Φ, N), we have:

1. F : N → V;

2. ∀m ∈ N Φ(m, F |m, F (m), w);

~
and, furthermore, for any n ∈ N and any function H with n ∈ dom(H),
we have:

~ for all m ∈ n ∪ {n}, then H(n) = F (n).

3. If Φ(m, H|m, H(m), w)

Proof. We first prove the following claim.

Claim.

(∀x ∈ N)(∀y1 )(∀y2 )[(hx, y1 i ∈ REC(Φ, N)∧hx, y2 i ∈ REC(Φ, N)) → y1 = y2 ]

Proof of Claim. By definition of REC(Φ, N) we have, for i = 1, 2, functions

fi with domain ni ∈ N such that fi (x) = yi and

(∀m ∈ ni ) Φ(m, fi |m, fi (m), w).

~

It suffices to prove that

(∀m ∈ N)(m ∈ x ∪ {x} → f1 (m) = f2 (m)),

which we do by induction on m ∈ N. To this end, we assume that m ∈ N

and
(∀j ∈ m)(j ∈ x ∪ {x} → f1 (j) = f2 (j))
 37

with intent to show that

m ∈ x ∪ {x} → f1 (m) = f2 (m).

To do this suppose m ∈ x ∪ {x}. Since x ∈ n1 ∩ n2 we have m ∈ n1 ∩ n2 so

that we have both

Φ(m, f1 |m, f1 (m), w)

~ and Φ(m, f2 |m, f2 (m), w).
~

By transitivity j ∈ x ∪ {x} for all j ∈ m and so by the inductive hypothesis

f1 |m = f2 |m. Now by the hypothesis of this theorem with f = f1 |m = f2 |m
we deduce that f1 (m) = f2 (m). This concludes the proof of the claim.

In order to verify (1), it suffices to show that

(∀x ∈ N)(∃y) [hx, yi ∈ REC(Φ, N)]

by induction. To this end, we assume that

(∀j ∈ x)(∃y) [hj, yi ∈ REC(Φ, N)]

with intent to show that ∃y hx, yi ∈ REC(Φ, N). For each j ∈ x there is
nj ∈ N and fj : nj → V such that

(∀m ∈ nj ) Φ(m, fj |m, fj (m), w).

~

If x ∈ nj for some j, then hx, fj (x)i S∈ REC(Φ, N) and we are done; so

assume that nj ≤ x for all j. Let g = {fj : j ∈ x}. By the claim, the fj ’s
agree on their common domains, so that g is a function with domain x and

(∀m ∈ x) Φ(m, g|m, g(m), w).

~

By the hypothesis of the theorem applied to g there is a unique y such that

~ Define f to be the function f = g ∪ {hx, yi}. It is straightfor-
Φ(x, g, y, w).
ward to verify that f witnesses that hx, yi ∈ REC(Φ, N).

To prove (2), note that, by (1), for each x ∈ N there is n ∈ N and

f : n → V such that F (x) = f (x) and, in fact, F |n = f . Hence,

(∀m ∈ n) Φ(m, f |m, f (m), w).

~
38 CHAPTER 4. THE NATURAL NUMBERS

We prove (3) by induction. Assume that

(∀m ∈ n) H(m) = F (m)

with intent to show that H(n) = F (n). We assume Φ(n, H|n, H(n), w) ~ and
by (2) we have Φ(n, F |n, F (n), w).
~ By the hypothesis of the theorem applied
to H|n = F |n we get H(n) = F (n).

By applying this theorem to our specific example we see that REC(Φ, N)

does indeed give us a function F . Since F is defined by recursion on N, we
use induction on N to verify the properties of F . For example, it is easy to
use induction to check that F (n) ∈ N for all n ∈ N.

We do not often explicitly state the formula Φ in a definition by recursion.

The definition of F would be more often given by:

F (0) = 3
F (succ(n)) = succ(F (n))

This is just how the example started; nevertheless, this allows us to construct
the formula Φ immediately, should we wish. Of course, in this particular
example we can use the plus symbol and give the definition by recursion by
the following formulas.

3+0=3
3 + succ(n) = succ(3 + n)

Now, let’s use definition by recursion in other examples. We can define

general addition on N by the formulas

a+0=a
a + succ(b) = succ(a + b)

for each a ∈ N. The same trick can be used for multiplicaton:

a·0=0
a · (succ(b)) = a · b + a
 39

for each a ∈ N, using the previously defined notion of addition. In each

example there are two cases to specify—the zero case and the successor case.
Exponentiation is defined similarly:

a0 = 1
asucc(b) = ab · a

The reader is invited to construct, in each case, the appropriate formula Φ,

with a as a parameter, and to check that the hypothesis of the previous
theorem is satisfied.

G. Peano developed the properties of the natural numbers from zero, the
successor operation and induction on N. You may like to see for yourself
some of what this entails.

Exercise 6. Let a and b be natural numbers. Show that

1. a + b = b + a; and,

2. a · b = b · a.

Do not believe this next result:

Proposition. All natural numbers are equal.

Proof. It is sufficient to show by induction on n ∈ N that if a ∈ N and b ∈ N

and max (a, b) = n, then a = b. If n = 0 then a = 0 = b. Assume the
inductive hypothesis for n and let a ∈ N and b ∈ N be such that

max (a, b) = n + 1.

Then max (a − 1, b − 1) = n and so a − 1 = b − 1 and consequently a = b.

40 CHAPTER 4. THE NATURAL NUMBERS
Chapter 5

The Ordinal Numbers

The natural number system can be extended to the system of ordinal num-
bers.

An ordinal is a transitive set of transitive sets. More formally: for any

term t, “t is an ordinal” is an abbreviation for

(t is transitive) ∧ (∀x ∈ t)(x is transitive).

We often use lower case Greek letters to denote ordinals. We denote

{α : α is an ordinal} by ON.

From Theorem 5 we see immediately that N ⊆ ON.

Theorem 9.

1. ON is transitive.

2. ¬(∃z)(z = ON).

Proof.

1. Let α ∈ ON; we must prove that α ⊆ ON. Let x ∈ α; we must prove

that

41
42 CHAPTER 5. THE ORDINAL NUMBERS

(a) x is transitive; and,

(b) (∀y ∈ x)(y is transitive).

Clearly (a) follows from the definition of ordinal. To prove (b), let
y ∈ x; by transitivity of α we have y ∈ α; hence y is transitive.

2. Assume (∃z)(z = ON). From (1) we have that ON is a transitive

set of transitive sets, i.e., an ordinal. This leads to the contradiction
ON ∈ ON.

Theorem 10. (Trichotomy of Ordinals)

(∀α ∈ ON)(∀β ∈ ON)(α ∈ β ∨ β ∈ α ∨ α = β).

Proof. The reader may check that a proof of this theorem can be obtained
by replacing “N” with “ON” in the proof of Theorem 6.

Because of this theorem, when α and β are ordinals, we often write α < β
for α ∈ β.

Since N ⊆ ON, it is natural to wonder whether N = ON. In fact, we

believe that “N = ON” can be neither proved nor disproved from the axioms
that we have stated (provided, of course, that those axioms are actually
consistent). We find ourselves at a crossroads in Set Theory. We can either
add “N = ON” to our axiom system, or we can add “N 6= ON”.

The axiom “N = ON” essentially says that there are no infinite sets and
the axiom “N 6= ON” essentially says that there are indeed infinite sets. Of
course, we go for the infinite!

The Axiom of Infinity N 6= ON

 43

As a consequence, there is a set of all natural numbers; in fact, N ∈ ON.

Theorem 11. (∃z)(z ∈ ON ∧ z = N).

Proof. Since N ⊆ ON and N 6= ON, pick α ∈ ON \ N. We claim that for each

n ∈ N we have n ∈ α; in fact, this follows immediately from the trichotomy
of ordinals and the transitivity of N. Thus N = {x ∈ α : x ∈ N} and by
Comprehension ∃z z = {x ∈ α : x ∈ N}. The fact that N ∈ ON now follows
immediately from Theorem 5.

The lower case Greek letter ω is reserved for the set N considered as an
ordinal; i.e., ω = N. Theorems 5 and 11 now show that the natural numbers
are the smallest ordinals, which are immediately succeeded by ω, after which
the rest follow. The other ordinals are generated by two processes illustrated
by the next lemma.

Lemma.

1. ∀α ∈ ON ∃β ∈ ON β = succ(α).
S
2. ∀S [S ⊆ ON → ∃β ∈ ON β = S].

Exercise 7. Prove this lemma.

For S ⊆ ON we write sup S for the least element of

{β ∈ ON : (∀α ∈ S)(α ≤ β)}

if such an element exists.

S
Lemma. ∀S [S ⊆ ON → S = sup S]

Exercise 8. Prove this lemma.

An ordinal α is called a successor ordinal whenever ∃β ∈ ON α = succ(β).

If α = sup α, then α is called a limit ordinal.
44 CHAPTER 5. THE ORDINAL NUMBERS

Lemma. Each ordinal is either a successor ordinal or a limit ordinal, but

not both.
Exercise 9. Prove this lemma.

We can perform induction on the ordinals via a process called

transfinite induction. In order to justify transfinite induction we need a the-
orem scheme.

For each formula Φ(v, w)

~ of the language of set theory we have:
Theorem 12. Φ

For all w,
~ if
∀n ∈ ON [(∀m ∈ n Φ(m, w))
~ → Φ(n, w)]
~
then
∀n ∈ ON Φ(n, w).
~

Proof. The reader may check that a proof of this theorem scheme can be
obtained by replacing “N” with “ON” in the proof of Theorem Scheme 7.

We can also carry out recursive definitions on ON. This process is called
transfinite recursion. For any formula Φ(x, f, y, w)
~ of the language of set
theory, we denote by REC(Φ, ON) the class

{hx, yi : (∃n ∈ ON)(∃f )[f : n → V∧f (x) = y∧∀m ∈ n Φ(m, f |m, f (m), w)]}.
~

Transfinite recursion is justified by the following theorem scheme.

For each formula Φ(x, f, y, w)

~ of the language of set theory we have:
Theorem 13. Φ

For all w,
~ suppose that we have

(∀x ∈ ON)(∀f )[(f : x → V) → ∃!y Φ(x, f, y, w)].

~

Then, letting F denote REC(Φ, ON), we have:

 45

1. F : ON → V;

2. ∀x ∈ ON Φ(x, F |x, F (x), w);

~ and, furthermore, for any n ∈ ON and
any function H with n ∈ dom(H) we have:

~ for all x ∈ n ∪ {n} then H(n) = F (n).

3. If Φ(x, H|x, H(x), w)

Proof. The reader may check that a proof of this theorem scheme can be
obtained by replacing “N” with “ON” in the proof of Theorem Scheme 8.

When applying transfinite recursion on ON we often have three sepa-

rate cases to specify, rather than just two as with recursion on N. This is
illustrated by the recursive definitions of the arithmetic operations on ON.

α + 0 = α;
Addition: α + succ(β) = succ(α + β);
α + δ = sup {α + η : η ∈ δ}, for each limit ordinal δ.

α · 0 = 0;
Multiplication: α · succ(β) = (α · β) + α;
α · δ = sup {α · η : η ∈ δ}, for each limit ordinal δ.

α0 = 1;
Exponentiation: αsucc(β) = (αβ ) · α;
αδ = sup {αη : η ∈ δ}, for each limit ordinal δ.

Note that, in each case, we are extending the operation from N to all of
ON. The following theorem shows that these operations behave similarly on
N and ON.
Theorem 14. Let α, β, and δ be ordinals and S be a non-empty set of
ordinals. We have,
46 CHAPTER 5. THE ORDINAL NUMBERS

1. 0 + α = α;

2. If β < δ then α + β < α + δ;

3. α + sup S = sup {α + η : η ∈ S};

4. α + (β + δ) = (α + β) + δ;

5. If α < β then α + δ ≤ β + δ;

6. 0 · α = 0;

7. 1 · α = α;

8. If 0 < α and β < δ then α · β < α · δ;

9. α · sup S = sup {α · η : η ∈ S};

10. α · (β + δ) = (α · β) + (α · δ);

11. α · (β · δ) = (α · β) · δ;

12. If α < β then α · δ ≤ β · δ;

13. 1α = 1;

14. If 1 < α and β < δ then αβ < αδ ;

15. αsup S = sup {αη : η ∈ S};

16. α(β+δ) = αβ · αδ ;

17. (αβ )δ = αβ·δ ;

18. If α < β then αδ ≤ β δ ; and

19. If 0 < α then ω α is a limit ordinal.

Exercise 10. Build your transfinite induction skills by proving two parts of
this theorem.

However, ordinal addition and multiplication are not commutative. This

is illustrated by the following examples, which are easy to verify from the
basic definitions.
 47

Examples.

1. 1 + ω = 2 + ω

2. 1 + ω 6= ω + 1

3. 1 · ω = 2 · ω

4. 2 · ω 6= ω · 2

5. 2ω = 4ω

6. (2 · 2)ω 6= 2ω · 2ω

Commonly, any function f with dom(f ) ⊆ ω is called a sequence. If

dom(f ) ⊆ n + 1 for some n ∈ ω, we say that f is a finite sequence; otherwise
f is an infinite sequence. As usual, we denote the sequence f by {fn }, where
each fn = f (n).

Theorem 15. There is no infinite descending sequence of ordinals.

Proof. Let’s use an indirect proof. Suppose x ⊆ ω is infinite and f : x → ON

such that if n < m then f (n) > f (m). Let X = {f (n) : n ∈ x}. By
Foundation there is y ∈ X such that y ∩ X = ∅; i.e., there is n ∈ x such that
f (n) ∩ X = ∅. However, if m ∈ x and m > n then f (m) ∈ f (n), which is a
contradiction.

Lemma. If α is a non-zero ordinal, then there is a largest ordinal β such

that ω β ≤ α.

Exercise 11. Prove this lemma. Show that the β ≤ α and that there are
cases in which β = α. Such ordinals β are called epsilon numbers (The
smallest such ordinal α = ω α is called 0 .)

Lemma. ∀α ∈ ON ∀β ∈ α ∃!γ α = β + γ.

Exercise 12. Prove this lemma.

48 CHAPTER 5. THE ORDINAL NUMBERS

If n ∈ ω and s : (n + 1) → ON is a finite sequence of ordinals, then the

Xn
sum s(i) is defined by recursion as follows.
i=0

0
X
s(i) = s(0); and,
i=0
m+1
X m
X
s(i) = s(i) + s(m + 1), for m < n.
i=0 i=0

This shows that statements like the following theorem can be written
precisely in the language of set theory.
Theorem 16. (Cantor Normal Form)

For each non-zero ordinal α there is a unique n ∈ ω and finite sequences

m0 , . . . , mn of natural numbers and β0 , . . . , βn of ordinals which satisfy β0 >
β1 > · · · > βn such that
α = ω β0 m 0 + ω β1 m 1 + · · · + ω βn m n .

Proof. Using the penultimate lemma, let

β0 = max {β : ω β ≤ α}
and then let
m0 = max {m ∈ ω : ω β0 m ≤ α}
which must exist since ω β m ≤ α for all m ∈ ω would imply that ω β0 +1 ≤ α.

By the previous lemma, there is some α0 ∈ ON such that

α = ω β0 m0 + α0
where the maximality of m0 ensures that α0 < ω β0 . Now let
β1 = max {β : ω β ≤ α0 }
so that β1 < β0 . Proceed to get
m1 = max {m ∈ ω : ω β1 m ≤ α0 }
 49

and α1 < ω β1 such that α0 = ω β1 m1 + α1 . We continue in this manner as

long as possible. We must have to stop after a finite number of steps or else
β0 > β1 > β2 > . . . would be an infinite decreasing sequence of ordinals. The
only way we could stop would be if some αn = 0. This proves the existence
of the sum. Uniqueness follows by induction on α ∈ ON.

Exercise 13. Verify the last statement of this proof.

Lemma.
1. If 0 < m < ω and δ is a limit ordinal, then m · δ = δ.

2. If k ∈ ω, and m0 , . . . , mk < ω, and α0 , . . . , αk < β, then

m 0 · ω α0 + · · · + m k · ω αk < ω β .
Exercise 14. Prove this lemma.

There is an interesting application of ordinal arithmetic to Number The-

ory. Pick a number—say x = 54. We have 54 = 25 + 24 + 22 + 2 when it
is written as the simplest sum of powers of 2. In fact, we can write out 54
using only the the arithmetic operations and the numbers 1 and 2. This will
be the first step in a recursively defined sequence of natural numbers, {xn }.
It begins with n = 2 and is constructed as follows.
2 +1) 2
x2 = 54 = 2(2 + 22 + 22 + 2.

Subtract 1.
2 +1) 2
x2 − 1 = 2(2 + 22 + 22 + 1.
Change all 2’s to 3’s, leaving the 1’s alone.
3 +1) 3
x3 = 3(3 + 33 + 33 + 1.

Subtract 1.
3 +1) 3
x3 − 1 = 3(3 + 33 + 33 .
Change all 3’s to 4’s, leaving any 1’s or 2’s alone.
4 +1) 4
x4 = 4(4 + 44 + 44 .
50 CHAPTER 5. THE ORDINAL NUMBERS

Subtract 1.
4 +1) 4
x4 − 1 = 4(4 + 44 + 3 · 43 + 3 · 42 + 3 · 4 + 3.

Change all 4’s to 5’s, leaving any 1’s, 2’s or 3’s alone.
5 +1) 5
x5 = 5(5 + 55 + 3 · 53 + 3 · 52 + 3 · 5 + 3.

Subtract 1 and continue, changing 5’s to 6’s, subtracting 1, changing 6’s to

7’s and so on. One may ask the value of the limit

lim xn .
n→∞

What is your guess? The answer is surprising.

Theorem 17. (Goodstein)

For any initial choice of x there is some n such that xn = 0.

Proof. We use an indirect proof; suppose x ∈ N and for all n ≥ 2 we have

xn 6= 0. From this sequence, we construct another sequence. For each n ≥ 2
we let gn be the result of replacing each occurrence of n in xn by ω. So, in
the example above we would get:
ω +1) ω)
g2 = ω (ω + ω (ω + ω ω + ω,
ω +1) ω)
g3 = ω (ω + ω (ω + ω ω + 1,
ω +1) ω)
g4 = ω (ω + ω (ω + ωω ,
ω +1) ω)
g5 = ω (ω + ω (ω + 3 · ω 3 + 3 · ω 2 + 3 · ω + 3,
ω +1) ω)
g6 = ω (ω + ω (ω + 3 · ω 3 + 3 · ω 2 + 3 · ω + 2,

etc. The previous lemma can now be used to show that {gn } would be an
infinite decreasing sequence of ordinals.

It is interesting that, although the statement of the theorem does not

mention infinity in any way, we used the Axiom of Infinity in its proof. We
do not need the Axiom of Infinity in order to verify the theorem for any one
 51

particular value of x—we just need to carry out the arithmetic. The reader
can do this for x = 4; however, finishing our example x = 54 would be tedious.
Moreover, the calculations are somewhat different for different values of x.
Mathematical logicians have proved that, in fact, there is no uniform method
of finitary calculations which will give a proof of the theorem for all x. The
Axiom of Infinity is necessary for the proof.
52 CHAPTER 5. THE ORDINAL NUMBERS
Chapter 6

Orderings

In the following definitions, R and C are terms.

1. We say R is a relation on C whenever R ⊆ C × C.

2. We say a relation R is irreflexive on C whenever ∀x ∈ C hx, xi ∈

/ R.

3. We say a relation R is transitive on C whenever

∀x ∀y ∀z [(hx, yi ∈ R ∧ hy, zi ∈ R) → hx, zi ∈ R.]

4. We say a relation R is well founded on C whenever

∀X [(X ⊆ C ∧ X 6= ∅) → (∃x ∈ X ∀y ∈ X hy, xi ∈

/ R)].

Such an x is called minimal for X.

5. We say a relation R is total on C whenever

∀x ∈ C ∀y ∈ C [hx, yi ∈ R ∨ hy, xi ∈ R ∨ x = y].

6. We say R is extensional on C whenever

∀x ∈ C ∀y ∈ C [x = y ↔ ∀z ∈ C (hz, xi ∈ R ↔ hz, yi ∈ R)].

53
54 CHAPTER 6. ORDERINGS

7. We say a relation R is a partial ordering on C whenever it is both

irreflexive and transitive.

8. We say R is a linear ordering on C whenever it is a partial ordering

which is total on C.

9. We say R is a well ordering indexordering!wellon C whenever it is a

well founded linear ordering.

An example of a partial ordering R on the power set P(X) is given by:

hx, yi ∈ R iff (x ⊆ y ∧ x 6= y).

An example of a well ordering R on an ordinal α is given by the membership

relation:
hx, yi ∈ R iff x ∈ y.
Every linear ordering is an extensional relation. The membership relation on
any set is extensional.
Remark. In fact, we can characterise ordinals as transitive sets well ordered
by the membership relation.

For partial orderings (a.k.a. partial orders) we usually use < instead of
R and we write x < y for hx, yi ∈ R.

Whenever ∃z z = hX, <i and < is a partial ordering on X, we say that

hX, <i is a partially ordered set.

Whenever ∃z z = hX, <i and < is well ordering on X, we say that hX, <i
is a well ordered set.

The study of partially ordered sets continues to be a major theme in

contemporary Set Theory and the construction of elaborate partial orders is
of great technical importance. In contrast, well orders have been thoroughly
analysed and we will classify all the well ordered sets in this chapter.

From the definition of inverse image we see that <← {z} is equal to

{x : ∃y ∈ {z} hx, yi ∈ <} = {x : x < z},

 55

the class of predecessors of z in the ordering <.

That a subset Y of X is an initial segment of hX, <i means that

∀y ∈ Y <← {y} ⊆ Y.

If, in addition, Y 6= X, we say Y is a proper initial segment.

Lemma. Let hX, <i be a well ordered set. Then Y is a proper initial segment
of hX, <i iff ∃x ∈ X Y = <← {x}.
Exercise 15. Prove this lemma.

Let hX, <X i and hY, <Y i be a partially ordered sets. We say f : X → Y
is order preserving whenever

(∀u ∈ X)(∀v ∈ X) [u <X v ↔ f (u) <Y f (v)].

More generally, this terminology is used even when <X and <Y are relations
but non necessarily partial orderings.

An isomorphism is an order preserving bijection. Whenever there is an

isomorphism from hX, <X i to hY, <Y i we say that hX, <X i is isomorphic to
hY, <Y i and write hX, <X i ∼
= hY, <Y i or just X ∼
= Y.
Lemma. If X and Y are well ordered sets and X ∼ = Y , then there is a unique
isomorphism from X to Y . It maps each initial segment of X onto an initial
segment of Y .
Exercise 16. Prove this lemma.
Exercise 17. Prove that any isomorphism between transitive sets is the
identity. Here the ordering is the one given by the membership relation.
Theorem 18. No well ordered set is isomorphic to a proper initial segment
of itself. Moreover, there is no order preserving function from a well ordered
set hA, <i into one of its proper initial segments.

Proof. Using the penultimate lemma, it is sufficient to show that we obtain

a contradiction from the existence of an a ∈ A and f such that

f : A → (<← {a})
56 CHAPTER 6. ORDERINGS

is order preserving.

Let X = {x ∈ A : f (x) < x}; X 6= ∅ because f (a) < a. Let z be the

least element of X; f (z) < z. But f (f (z)) < f (z) since f is order preserving,
which contradicts the minimality of z.

Theorem 19. Any two well ordered sets are either isomorphic or one is
isomorphic to an initial segment of the other.

Proof. Let hA, <A i and hB, <B i be well ordered sets. We need to prove that
one of the following three conditions holds:

1. hA, <A i ∼
= hB, <B i;
2. ∃b ∈ B hA, <A i ∼
= h<←
B {b}, <B i; or,

3. ∃a ∈ A h<← ∼
A {a}, <A i = hB, <B i.

Let
f = {hx, yi ∈ A × B :<← ∼ ←
A {x} = <B {y}}.

Claim 1. f is a function.
Proof of Claim 1. Otherwise there exists x and y1 6= y2 , with hx, y1 i ∈ f
and hx, y2 i ∈ f . Hence <← ∼ ←
B {y1 } = <B {y2 }, which contradicts the previous
theorem.
Claim 2. f is order preserving (and hence also one-to-one).
Proof of Claim 2. If {x1 , x2 } ⊆ dom(f ) with x1 <A x2 then

<← ∼ ←
A {x2 } = <B {f (x2 )}.

So f (x1 ) ∈ <←B {f (x2 )}; that is, f (x1 ) <B f (x2 ).

Claim 3. dom(f ) is an initial segment of A.
Proof of Claim 3. We show that if x ∈ dom(f ) then <← A {x} ⊆ dom(f ).
← ∼ ←
Let f (x) = y, so <A {x} = <B {y}. If z <A x, then for some w ∈ B we
have <← ∼ ←
A {z} = <B {w} by the lemmas of Exercises 15 and 16. Therefore
hz, wi ∈ f , and hence z ∈ dom(f ).
 57

Claim 4. rng(f ) is an initial segment of B.

Exercise 18. Prove this claim.
Claim 5. dom(f ) and rng(f ) cannot both be proper initial segments.
Proof of Claim 5. Suppose otherwise; then we have dom(f ) =<← {a} for
some a ∈ A, and rng(f ) =<← {b} for some b ∈ B. But this will mean
ha, bi ∈ f since dom(f ) ∼
= rng(f ).

This complete the proof.

Induction can be carried out on any well founded relation on any set. This
is called ∈ −induction. The verification of this, of course, must be expressed
as a theorem scheme.

For each formula Φ(v, w)

~ of the language of set theory we have:

Theorem 20. Φ

Suppose R is a well founded relation on a set X. For all w,

~ if:

∀x ∈ X [(∀y ∈ (R← {x}) Φ(y, w))

~ → Φ(x, w)]
~

then
∀x ∈ X Φ(x, w).
~

~ and some x ∈ X
Proof. By way of indirect proof, suppose that for some w
we have ¬Φ(x, w).
~ By Comprehension,

∃z z = {y ∈ X : ¬Φ(y, w)}.
~

Our assumption is that z 6= ∅, so there is some u ∈ z such that hy, ui ∈

/ R
←
for all y ∈ z. Then ∀y ∈ R {u} Φ(y, w). ~ The inductive hypothesis gives
~ which contradicts that u ∈ z.
Φ(u, w),
58 CHAPTER 6. ORDERINGS

As an application, we prove the following theorem. (When dealing with

ordinals, it is understood that the ordering is the well order generated by the
membership relation ∈.
Theorem 21. Each well ordered set is isomorphic to a unique ordinal.

Proof. Let hX, <i be a well ordered set. Pick p ∈

/ X and extend the well
∗
ordering on X to X = X ∪ {p} by making x < p for each x ∈ X. It suffices
to prove that
∀x ∈ X ∗ ∃!α ∈ ON (<← {x}) ∼
= α.
because the theorem follows when x = p.

We prove this statement by induction on the well ordered set hX ∗ , <i. It

is necessary to prove only that for all x ∈ X ∗ ,
[∀y ∈ (<← {x}) ∃!α ∈ ON (<← {y}) ∼
= α]
→ [∃!β ∈ ON (< {x}) ∼
←
= β].

To prove this, fix x ∈ X ∗ and for each y < x denote by αy and fy ,

respectively, the unique ordinal and isomorphism such that
fy : (<← {y}) → αy .
Claim. If y < z are elements of <← {x}, then
fy = fz |<← {y} .
Proof of Claim. To prove this claim, first note that fz maps <← {y} onto
an initial segment S of αz . It is easy to check that any initial segment of
an ordinal is an ordinal. So by uniqueness S = αy and by uniqueness of the
isomorphism we get the claim.
S
Let f = {fy : y < x}; that f exists is proved using Replacement. There
are now two cases.

Case 1: <← {x} has no largest element.

Let β = sup {αy : y < x}. Using the claim, we see that
f : (<← {x}) → β
is an isomorphism.
 59

Case 2: <← {x} has a largest element—say z. Then fz ∪ {hz, αz i} is our

isomorphism with our β = αz + 1.

In either case, uniqueness follows since no two ordinals are isomorphic,

one being an initial segment of the other. This completes the proof of the
claim.

The order type, typehX, <i, of a well ordered set hX, <i is defined to
be the unique ordinal to which hX, <i is isomorphic. Well orders are all
classified by their order type.

We now wish to prove the fundamental theorem of Set Theory, due to

E. Zermelo, that every set can be well ordered. To do this, we need to call
up the Axiom of Choice and the Power Set Axiom. In modern set theory,
the Axiom of Choice is mainly used for the proof of Zermelo’s theorem. The
theorem is also called the Well Ordering Principle.
Theorem 22. (Zermelo)

(∀X)(∃ <) [hX, <i is a well ordered set].

Proof. Let f be a choice function on P(X) \ {∅}. By the Power Set Axiom
and Comprehension we have,
∃W W = {hA, <A i : A ∈ P(X)
∧ (<A is a well ordering of A)
∧ (∀x ∈ A) f (X \ {a ∈ A : a <A x}) = x}.
Now let [
X∗ = {A : ∃ <A hA, <A i ∈ W }
and [
<∗ = {<A : ∃A hA, <A i ∈ W }.
We now make three claims which will immediately give the theorem.
Claim 1. If hA, <A i and hB, <B i are in W , then one is an initial segment of
the other.
60 CHAPTER 6. ORDERINGS

Proof of Claim 1. Without loss of generality, by Theorem 19, there is an

initial segment C of B and an isomorphism h : A → C.

By induction on the well ordered set hA, <A i, we prove that A is an initial
segment of B by showing that h(x) = x for all x ∈ A.

Let x ∈ A and assume that ∀y <A x h(y) = y; we need to show that

h(x) = x.

Now {a ∈ A : a <A x} = {h(a) ∈ B : h(a) <B h(x)} since h is order-

preserving. So {a ∈ A : a <A x} = {b ∈ B : b <B h(x)} since rng(h) is an
initial segment of B.

So x = f (X \ {a ∈ A : a <A x}) = f (X \ {b ∈ B : b <B h(x)}) = h(x).

Claim 2. hX ∗ , <∗ i is a well ordered set.
Proof of Claim 2.
1. <∗ is irreflexive:
This is because for no x and A do we have hx, xi ∈<A .
2. <∗ is transitive:
Let {x, y, z} ⊆ X ∗ with x <∗ y and y <∗ z. By the first claim there is
hA, <A i ∈ W such that {x, y, z} ⊆ A, x <A y, and y <A z. So x <A z
and hence x <∗ z.
3. <∗ is total:
Let {x, y} ⊆ X ∗ . By the first claim, there is hA, <A i ∈ W with {x, y} ∈
A, so that one of the three conditions x <A y, y <A x, and x = y holds;
hence x <∗ y, y <∗ x or x = y.
4. <∗ is well founded:
6 Y ⊆ X ∗ . There is some hA, <A i ∈ W such that A ∩ Y 6= ∅;
Let ∅ =
let y ∈ A ∩ Y be <A −minimal. We will prove, in fact, that y is
<∗ −minimal in Y . If x ∈ Y \ {y} then by the first claim there is
hB, <B i ∈ W such that x ∈ B and A is an initial segment of B, so that
y <B X and hence y <∗ x.
Claim 3. X ∗ = X.
Proof of Claim 3. Clearly X ∗ ⊆ X. If X ∗ 6= X then ∃x ∈ X \ X ∗ such that
x = f (X \ X ∗ ). We can now let < be <∗ ∪{ha, xi : a ∈ X ∗ } and note that
 61

< is a well ordering of X ∗ ∪ {x}. By the first claim, for any y ∈ X ∗ there is
hA, <A i ∈ W such that y ∈ A and

{a ∈ A : a <A y} = {a ∈ A : a <∗ y} = {a ∈ X ∗ ∪ {x} : a < y}.

So

y = f (X \ {a ∈ A : a <A y}) = f (X \ {a ∈ X ∗ ∪ {x} : a < y}).

Also
x = f (X \ {a ∈ X ∗ ∪ {x} : a < x})
since
X ∗ = {a ∈ X ∗ ∪ {x} : a < x}.
Hence hX ∗ ∪ {x}, <i ∈ W , which is a contradiction.
62 CHAPTER 6. ORDERINGS
Chapter 7

Cardinality

In this chapter, we investigate a concept which aims to translate our intuitive

notion of size into formal language. By Zermelo’s Well Ordering Principle
(Theorem 22) every set can be well ordered. By Theorem 18, every well
ordered set is isomorphic to an ordinal. Therefore, for any set x there is
some ordinal κ ∈ ON and a bijection f : x → κ.

We define the cardinality of x, |x|, to be the least κ ∈ ON such that there

is some bijection f : x → κ. Every set has a cardinality.

Those ordinals which are |x| for some x are called cardinals.

Exercise 19. Prove that each n ∈ ω is a cardinal and that ω is a cardinal.

Show that ω + 1 is not a cardinal and that, in fact, each other cardinal is a
limit ordinal.

Theorem 23. The following are equivalent.

1. κ is a cardinal.

2. (∀α < κ)(¬∃ bijection f : κ → α); i.e., |κ| = κ.

3. (∀α < κ)(¬∃ injection f : κ → α).

63
64 CHAPTER 7. CARDINALITY

Proof. We prove the negations of each are equivalent:

¬(1) (∀x)[(∃ bijection g : x → κ) → (∃α < κ)(∃ bijection h : x → α)]
¬(2) ∃α < κ ∃ bijection f : κ → α
¬(3) ∃α < κ ∃ injection f : κ → α

¬(2) ⇒ ¬(1) Just take h = f ◦ g.

¬(1) ⇒ ¬(3) Just consider x = κ.
¬(3) ⇒ ¬(2) Suppose α < κ and f : κ → α is an injection. By Theorem
21, there is an isomorphism g : β → f 00 κ for some β ∈ ON. Now
g −1 ◦ f : κ → β is a bijection. It suffices to prove that β ≤ α and, since
g is order preserving, this follows from Theorem 18.

The following exercises are applications of the theorem. The first two
statements relate to the questions raised in the introduction.
Exercise 20. Prove the following:
1. |x| = |y| iff ∃ bijection f : x → y.
2. |x| ≤ |y| iff ∃ injection f : x → y.
3. |x| ≥ |y| iff ∃ surjection f : x → y.
Exercise 21. Prove that the supremum of a set of cardinals is a cardinal.
Theorem 24. (G. Cantor)

∀x |x| < |P(x)|.

Proof. First note that if |x| ≥ |P(x)|, then there would be a surjection
g : x → P(x).
But this cannot happen, since {a ∈ x : a ∈ / g 00 (x).
/ g(a)} ∈
 65

For any ordinal α, we denote by α+ the least cardinal greater than α.

This is well defined by Theorem 24.
Exercise 22. Prove that ¬∃z z = {κ : κ is a cardinal}.
Theorem 25. For any infinite cardinal κ, |κ × κ| = κ.

Proof. Let κ be an infinite cardinal. The formulas |κ| = |κ × {0}| and

|κ × {0}| ≤ |κ × κ| imply that κ ≤ |κ × κ|. We now show that |κ × κ| ≤ κ.
We use induction and assume that |λ × λ| = λ for each infinite cardinal
λ < κ.

We define an ordering on κ × κ by:


max {α0 , β0 } < max {α1 , β1 };

hα0 , β0 i < hα1 , β1 i iff max {α0 , β0 } = max {α1 , β1 } ∧ α0 < α1 ; or,

max {α0 , β0 } = max {α1 , β1 } ∧ α0 = α1 ∧ β0 < β1 .


It is easy to check that < well orders κ × κ.

Let θ = type hκ × κ, <i. It suffices to show that θ ≤ κ. And for this it

suffices to show that for all hα, βi ∈ κ × κ,

type h<← {hα, βi}, <i < κ.

To this end, pick hα, βi ∈ κ × κ. Let δ ∈ κ be such that <← {hα, βi} ⊆ δ × δ.
This is possible since κ is a limit ordinal. It now suffices to prove that

type h<← {hδ, δi}, <i < κ.

By Theorem 23 it suffices to prove that |<← {hδ, δi}| < κ and so it suffices
to prove that |δ × δ| < κ.

Since |δ × δ| = ||δ| × |δ||, it suffices to prove that |λ × λ| < κ for all

cardinals λ < κ. If λ is infinite, this is true by inductive hypothesis. If λ is
finite, then |λ × λ| < ω ≤ κ.

Corollary. If X is infinite, then |X × Y | = max {|X|, |Y |}.

66 CHAPTER 7. CARDINALITY
S
Theorem 26. For any X we have | X| ≤ max {|X|, sup {|a| : a ∈ X}},
provided that at least one element of X ∪ {X} is infinite.

Proof. Let κ = sup {|a| : a ∈ X}. Using Exercise 20, for S

each a ∈ X there is
a surjection fa : κ → a. Define a surjection f : X × κ → X by

f (ha, αi) = fa (α).

Using Exercise 20 and the previous corollary, the result follows.

Define B A as {f : f : B → A} and [A]κ as {x : x ⊆ A ∧ |x| = κ}.

Lemma. If κ is an infinite cardinal, then |κ κ| = |[κ]κ | = |P(κ)|.

Proof. We have,
κ κ
2⊆ κ ⊆ [κ × κ]κ ⊆ P(κ × κ).
Using characteristic functions it is easily proved that |κ 2| = |P(κ)|. Since
|κ × κ| = κ we have the result.

Lemma. If κ ≤ λ and λ is an infinite cardinal, then |κ λ| = |[λ]κ |.

Proof. For each x ∈ [λ]κ there is a bijection fx : κ → x. Since fx ∈ κ

λ, we
have an injection from [λ]κ into κ λ, so |[λ]κ | ≤ |κ λ|.

Now κ λ ⊆ [κ × λ]κ . Thus, |κ λ| ≤ |[λ]κ |.

For any ordinal κ, define κ+ to be the least cardinal greater than κ.

A function f : δ → κ is said to be cofinal in κ whenever sup rng(f ) = κ.

A set S ⊆ κ is said to be cofinal whenever sup S = κ. The cofinality, cf (κ),
 67

of a cardinal κ is the least δ ∈ ON such that there is a cofinal f : δ → κ.

This is also the least size of an S ⊆ κ cofinal in κ.

A cardinal κ is said to be regular whenever cf (κ) = κ. Otherwise, κ

is said to be singular. These notions are of interest when κ is an infinite
cardinal. In particular, ω is regular.
Lemma.
1. For each infinite κ, cf (κ) is a regular cardinal.
2. For each infinite κ, κ+ is a regular cardinal.
3. Each infinite singular cardinal contains a cofinal subset of regular car-
dinals.
Exercise 23. Prove this lemma.
Theorem 27. (König’s Theorem)

For each infinite cardinal κ, |cf (κ) κ| > κ.

Proof. We show that there is no surjection g : κ → δ κ, where δ = cf (κ).

Let f : δ → κ witness that cf (κ) = δ. Define h : δ → κ such that each
h(α) ∈
/ {g(β)(α) : β < f (α)}. Then h ∈/ g 00 (κ), since otherwise h = g(β) for
some β < κ; pick α ∈ δ such that f (α) > β.

Corollary. For each infinite cardinal κ, cf (|P(κ)|) > κ.

Proof. Let λ = |P(κ)|. Suppose cf (λ) ≤ κ. Then

λ = |P(κ)| = |κ 2| = |(κ×κ) 2| = |κ (κ 2)| = |κ λ| ≥ |cf (λ) λ| > λ.

Cantor’s Theorem guarantees that for each ordinal α there is a set, P(α),
which has cardinality greater than α. However, it does not imply, for exam-
ple, that ω + = |P(ω)|. This statement is called the Continuum Hypothesis,
and is equivalent to the third question in the introduction.
68 CHAPTER 7. CARDINALITY

The aleph function ℵ : ON → ON is defined as follows:

ℵ( 0) = ω
ℵ(α) = sup {ℵ(β)+ : β ∈ α}.
We write ℵα for ℵ(α). We also sometimes write ωα for ℵ(α).

The beth function i : ON → ON is defined as follows:

i(0) = ω
i(α) = sup {|P(i(β))| : β ∈ α}.
We write iβ for i(β).

It is apparent that ℵ1 ≤ i1 . The continuum hypothesis is the statement

ℵ1 = i1 ; this is abbreviated as CH. The generalised continuum hypothesis,
GCH, is the statement ∀α ∈ ON ℵα = iα .
Exercise 24. Prove that
∀κ [κ is an infinite cardinal → (∃α ∈ ON )(κ = ℵα )].

A cardinal κ is said to be inaccessible whenever both κ is regular and

∀λ < κ |P(λ)| < κ.
An inaccessible cardinal is sometimes said to be strongly inaccessible, and
the term weakly inaccesible is given to a regular cardinal κ such that
∀λ < κ λ+ < κ.
Under the GCH these two notions are equivalent.
Axiom of Inaccessibles ∃κ κ > ω and κ is an inaccessible cardinal
This axiom is a stronger version of the Axiom of Infinity, but the mathemat-
ical community is not quite ready to replace the Axiom of Infinity with it
just yet. In fact, the Axiom of Inacessibles is not included in the basic ZFC
axiom system and is therefore always explicitly stated whenever it is used.
Exercise 25. Are the following two statements true?

1. κ is weakly inaccessible iff κ = ℵκ .

2. κ is strongly inaccesssible iff κ = iκ .
Chapter 8

There Is Nothing Real About

The Real Numbers

We now formulate three familiar number systems in the language of set the-
ory. From the natural numbers we shall construct the integers; from the
integers we shall construct the decimal numbers and the real numbers.

For each n ∈ N let −n = {{m} : m ∈ n}. The integers, denoted by Z,

are defined as
Z = N ∪ {−n : n ∈ N}.
For each n ∈ N \ {∅} we denote {n} by −n.

We can extend the ordering < on N to Z by letting x < y iff one of the
following holds:

1. x ∈ N ∧ y ∈ N ∧ x < y;
2. x ∈
/ N ∧ y ∈ N; or,
S S
3. x ∈
/ N∧y ∈ / N ∧ y < x.

To form the reals, first let

ω
F = {f : f ∈ Z}.

69
70CHAPTER 8. THERE IS NOTHING REAL ABOUT THE REAL NUMBERS

We pose a few restrictions on such functions as follows. Let us write:

A(f ) for (∀n > 0)(−9 ≤ f (n) ≤ 9);

B(f ) for (∀n ∈ ω)(f (n) ≥ 0) ∨ (∀n ∈ ω)(f (n) ≤ 0);
C(f ) for (∀m ∈ ω)(∃n ∈ ω \ m)(f (n) ∈/ {9, −9}); and,
D(f ) for (∃m ∈ ω)(∀n ∈ ω \ m)(f (n) = 0).

Finally, let
R = {f : f ∈ F and A(f ) and B(f ) ∧ C(f )}
to obtain the real numbers and let

D = {f : f ∈ R and D(f )}

to obtain the decimal numbers.

We now order R as follows: let f < g iff

(∃n ∈ ω) [f (n) < g(n) ∧ (∀m ∈ n)(f (m) = g(m))].

This ordering clearly extends our ordering on N and Z, and restricts to D.

In light of these definitions, the operations of addition, multiplication and

exponentiation defined in Chapter 4 can be formally extended from N to Z,
D and R in a natural—if cumbersome—fashion.
Lemma.
1. |D| = ℵ0 .

2. R is uncountable.

3. < is a linear order on D.

4. (∀p ∈ R)(∀q ∈ R) [p < q → ∃d ∈ D p < d < q].

I.e., D is a countable dense subset of R.

5. < is complete; i.e., bounded subsets have suprema and infima.

Exercise 26. Prove this lemma.
Theorem 28. Any two countable dense linear orders without endpoints are
isomorphic.
 71

Proof. This method of proof, the back-and-forth argument, is due to G.Cantor.

The idea is to define an isomorphism recursively in ω steps, such that at each
step we have an order-preserving finite function; at even steps f (xi ) is defined
and at odd steps f −1 (yj ) is defined.

Precisely, if X = {xi : i ∈ ω} and Y = {yj : j ∈ ω} are two countable

dense linear orders we define f : X → Y by the formulas

f0 = {hx0 , y0 i}
fn+1 = fn ∪ {hxi , yj i}

where

1. if n is even, i = min {k ∈ ω : xk ∈ / dom(fn )} and j is chosen so that

fn ∪ {hxi , yj i} is order-preserving; and,

2. if n is odd, j = min {k ∈ ω : yk ∈ / rng(fn )} and i is chosen so that

fn ∪ {hxi , yj i} is order-preserving.

for each n ∈ ω, there is indeed a choice of j in (1) and i

We then check that S
in (2) and that f = {fn : n ∈ ω} is an isomorphism.

Any complete dense linear order without endpoints and with a countable
dense subset is isomorphic to hR, <i. Can “with a countable dense subset”
be replaced by “in which every collection of disjoint intervals is countable”?.
The affirmation of this is called the Suslin Hypothesis. It is the second most
important problem in Set Theory. It too requires new axioms for its solution.

We have begun with N, extended to Z, then extended again to R. We

now extend once more to ∗ R. This is the set of hyperreals. In order to do
this, we introduce the important notion of an ultrafilter.

A collection of subsets U ⊆ P(S) is a filter provided that it satisfies the

first three of the following conditions:

1. S ∈ U and ∅ ∈
/ U.
72CHAPTER 8. THERE IS NOTHING REAL ABOUT THE REAL NUMBERS

2. If A ∈ U and B ∈ U, then A ∩ B ∈ U.

3. If A ∈ U and A ⊆ B, then B ∈ U.

4. A ∈ U or B ∈ U whenever A ∪ B = S.

5. ∀x ∈ S {x} ∈
/ U.

If a filter U obeys condition (4) it is called an ultrafilter, and if U satisfies

conditions (1) through (5) it is said to be a free or non-principal ultrafilter.

An ultrafilter is a maximal filter under inclusion. Every filter can be

extended to an ultrafilter. (We can recursively define it.) There is a free
ultrafilter over ω.

Theorem 29. (Ramsey)

If P : [ω]2 → {0, 1}, then there is H ∈ [ω]ω such that |P 00 [H]2 | = 1.

Proof. Let U be a free ultrafilter over ω. Either:

1. {α ∈ ω : {β ∈ ω : P ({α, β}) = 0} ∈ U} ∈ U; or,

2. {α ∈ ω : {β ∈ ω : P ({α, β}) = 1} ∈ U} ∈ U.

As such, the proof breaks into two similar cases. We address case (1).

Let S = {α ∈ ω : {β ∈ ω : P ({α, β}) = 0} ∈ U}. Pick α0 ∈ S and let

S0 = {β ∈ ω : P ({α0 , β}) = 0}.

Pick α1 ∈ S ∩ S0 and let

S1 = {β ∈ ω : P ({α1 , β}) = 0}.

In general, recursively choose {αn : n < ω} such that for each n

αn+1 ∈ S ∩ S0 ∩ · · · ∩ Sn ,
 73

where
Sn = {β ∈ ω : P ({αn , β}) = 0}.
Then H = {αn : n < ω} exhibits the desired property.

Theorem 30. (Sierpinski)

There is a function P : [ω1 ]2 → {0, 1} such that there is no H ∈ [ω1 ]ω1

with |P 00 [H]2 | = 1.

Proof. Let f : ω1 → R be an injection. Define P as follows: for α < β, let

(
0, if f (α) < f (β);
P ({α, β}) = (8.1)
1, if f (α) > f (β).

The following exercise finishes the proof.

Exercise 27. There is no subset of R with order type ω1 .

Let U be a free ultrafilter over ω. Form an equivalence relation ∼ on ω R

by the rule:
f ∼ g whenever {n ∈ ω : f (n) = g(n)} ∈ U.
The equivalence class of f is denoted by
ω
[f ] = {g ∈ R : g ∼ f }.

The set of equivalence classes of ∼ is called the ultrapower of R with respect

to U. The elements of ω R are often called the hyperreal numbers and denoted
∗
R.

There is a natural embedding of R into ∗ R given by

x 7→ [fx ]

where fx : ω → R is the constant function; i.e., fx (n) = x for all n ∈ ω; we

identify R with its image under the natural embedding.
74CHAPTER 8. THERE IS NOTHING REAL ABOUT THE REAL NUMBERS

We can define an ordering ∗ < on ∗ R by the rule:

a <∗ b whenever ∃f ∈ a ∃g ∈ b {n ∈ ω : f (n) < g(n)} ∈ U.

∗
Lemma. < is a linear ordering on ∗ R which extends the usual ordering of
R.
Exercise 28. Prove this lemma.

We usually omit the asterisk, writing < for ∗ <.

Exercise 29. There is a subset of ∗ R of order type ω1 .

For each function F : Rn → R there is a natural extension

∗
F : (∗ R)n → ∗ R

given by
∗
F (~a) = [F ◦ ~s]
where ~a = ha0 , . . . , an−1 i ∈ (∗ R)n and ~s : ω → Rn such that for each j ∈ ω

~s(j) = hs0 (j), . . . , sn−1 (j)i

and si ∈ ai for each i.

Theorem 31. (The Leibniz Transfer Principle)

Suppose F : Rn → R and G : Rn → R.

1. ∀~a ∈ Rn F (~a) = G(~a) iff ∀~a ∈ (∗ R)n ∗

F (~a) = ∗
G(~a).

2. ∀~a ∈ Rn F (~a) < G(~a) iff ∀~a ∈ (∗ R)n ∗

F (~a) < ∗
G(~a).
Exercise 30. Prove the above theorem. This includes verifying that ∗ F is
a function.

As a consequence of this theorem, we can extend + and × to ∗ R. For

example, a + b = c means that

∃f ∈ a ∃g ∈ b ∃h ∈ c {n ∈ ω : f (n) + g(n) = h(n)} ∈ U.

 75

Indeed, the natural embedding embeds R as an ordered subfield of ∗ R.

In order to do elementary calculus we consider the infinitesimal elements

of R. Note that R 6= ∗ R; consider
∗

α = [h1, 1/2, 1/3, . . . , 1/n, . . . i].

Then α > 0 but α < r for each positive r ∈ R; a member of ∗ R with this
property is called a positive infinitesimal. There are negative infinitesimals;
0 is an infinitesimal. Since ∗ R is a field 1/α exists and 1/α > r for any real
number r; it is an example of a positive infinite number.

A hyperreal number a is said to be finite whenever |a| < r for some real
r. Two hyperreal numbers a and b are said to be infinitely close whenever
a − b is infinitesimal. We write a ≈ b.

Lemma. Each finite hyperreal number is infinitely close to a unique real

number.

Proof. Let a be finite. Let s = sup {r ∈ R : r < a}. Then a ≈ s. If we also

have another real t such that a ≈ t, then we have s ≈ t and so s = t.

If a is finite, the standard part of a, st(a), is defined to be the unique real

number which is infinitely close to a.

It is easy to check that for finite a and b,

st(a + b) = st(a) + st(b); and,

st(a × b) = st(a) × st(b).

For a function F : R → R we define the derivative, F 0 (x), of F (x) to be

 
F (x + 4x) − F (x)
st
4x

provided this exists and is the same for each non-zero infinitesimal 4x.
76CHAPTER 8. THERE IS NOTHING REAL ABOUT THE REAL NUMBERS

The fact that F 0 (x) exists implies that for each infinitesimal 4x there is
an infinitesimal  such that F (x + 4x) = F (x) + F 0 (x)4x + 4x. That is,
for any 4x ≈ 0 there is some  ≈ 0 such that

4y = F 0 (x)4x + 4x

where 4y = F (x + 4x) − F (x).

Theorem 32. (The Chain Rule)

Suppose y = F (x) and x = G(t) are differentiable functions.

Then y = F (G(t)) is differentiable and has derivative F 0 (G(t)) · G0 (t).

Proof. Let 4t be any non-zero infinitesimal. Let 4x = G(t + 4t) − G(t).

Since G0 (x) exists, 4x is infinitesimal. Let 4y = F (x + 4x) − F (x). We
4y
wish to calculate st( 4x ), which will be the derivative of y = F (G(t)). We
consider two cases.

Case 0: 4x = 0
4y = 0, st( 4y
4t
) = 0, and G0 (t) = st( 4x
4t
) = 0.
4y 0 0
So st( 4t ) = F (G(t)) · G (t).

Case 1: 4x 6= 0
4y 4y 4x
4t
= 4x · 4t . So st( 4y
4t
4y
) = st( 4x ) · st( 4x
4t
), and again,
4y 0 0 0 0
st( 4t ) = F (x) · G (t) = F (G(t)) · G (t).
Chapter 9

The Universe

In this chapter we shall discuss two methods of measuring the complexity

of a set, as well as their corresponding gradations of the universe. For this
discussion it will be helpful to develop both a new induction and a new
recursion procedure, this time on the whole universe. Each of these will
depend upon the fact that every set is contained in a transitive set, which
we now prove.

By recursion on N, we define:
[
0
X = X; and,
[ [[
n+1
X = ( n X).

We define the transitive closure of X as,

[ [
trcl(X) = { n X : n ∈ ω}.

Theorem 33.

1. ∀X trcl(X) is the smallest transitive set containing X.

S
2. ∀x trcl(x) = x ∪ {trcl(y) : y ∈ x}.

S
Proof. Note that Y is transitive iff Y ⊆ Y , so trcl(X) is transitive.

77
78 CHAPTER 9. THE UNIVERSE

1. If Y is transitive and X ⊆ Y then for each n ∈ N, n X ⊆ n Y ⊆ Y .

S S
So trcl(X) ⊆ Y .
S
2. To prove that trcl(x)S ⊇ x ∪ {trcl(y) : y ∈ S nx}, notice
S n+1 first that
since ∀y ∈ x y ⊆ x, we have that ∀n ∈ ω y ⊆ x. Hence
trcl(y) ⊆ trcl(x).
Sn For theSopposite containment, we prove by induction
that ∀n ∈ ω x ⊆ x ∪ {trcl(y) : y ∈ x}.

As with N and ON, we can perform induction on the universe, called

∈ −induction, as illustrated by the following theorem scheme.

For each formula Φ(v, w)

~ of the language of set theory we have:
Theorem 34. Φ

For all w,
~ if
∀n [(∀m ∈ n Φ(m, w))
~ → Φ(n, w)]
~
then
∀n Φ(n, w).
~

Proof. We shall assume that the theorem is false and derive a contradiction.
~ and a fixed l such that ¬Φ(l, w).
We have w ~

Let t be any transitive set containing l. Thanks to the previous theorem,

we can let t = trcl(l ∪ {l}). The proof now proceeds verbatim as the proofs
of Theorems 7 and 12.

We can also carry out recursive definitions on V; this is called ∈ −recursion.

For any formula Φ(x, f, y, w)
~ of the language of set theory, we denote by
REC(Φ, V) the class

{hx, yi : (∃n)(∃f ) [f : n → V ∧ f (x) = y ∧ ∀m ∈ n Φ(m, f |m, f (m), w)]}.

~

∈ −recursion is justified by the next theorem scheme.

 79

For each formula Φ(x, f, y, w)

~ of the language of set theory we have:

Theorem 35. Φ

For all w,
~ suppose that we have

(∀x)(∀f ) [(f : x → V) → ∃!y Φ(x, f, y, w)].

~

Then, letting F denote REC(Φ, V), we have:

1. F : V → V;

2. ∀x Φ(x, F |x, F (x), w);

~ and furthermore for any n and any function H
with n ∈ dom(H) we have,

~ for all x ∈ n ∪ {n} then H(n) = F (n).

3. Φ(x, H|x, H(x), w)

Proof. The proof is similar to that of Theorems 8 and 13.

Our first new measure of the size of a set is given by the rank function.
This associates, to each set x, an ordinal rank(x) by the following rule:

rank(x) = sup {rank(y) + 1 : y ∈ x}

Observe that ∀α ∈ ON rank(α) = α.

By recursion on ON we define the cumulative hierarchy, an ordinal-gradation

on V, as follows.

R(0) = ∅;
R(α + 1) = P(R(α)); and,
[
R(δ) = {R(α) : α < δ} if δ is a limit ordinal.

Sometimes we write Rα or Vα for R(α). The next theorem connects and lists
some useful properties of the cumulative hierarchy and the rank function.

Theorem 36.
80 CHAPTER 9. THE UNIVERSE

1. ∀α ∈ ON R(α) is transitive.

2. ∀α ∈ ON ∀β ∈ ON (β < α → R(β) ⊆ R(α).

3. ∀x ∀α ∈ ON (x ∈ R(α) ↔ ∃β ∈ α x ⊆ R(β)).

4. ∀x ∀α ∈ ON (x ∈ R(α + 1) \ R(α) ↔ rank(x) = α).

S
5. ∀x ∃α ∈ ON x ∈ R(α); i.e., V = {R(α) : α ∈ ON}.

Proof.

1. This is an easy induction on α ∈ ON.

2. Apply induction on α, using (1).

3. (→) Note that if γ is least ordinal such that x ∈ R(γ), then γ is a

successor ordinal, so choose β such that γ = β + 1;
(←) This uses (2).

4. First show by induction on α that rank(x) < α implies x ∈ R(α).

5. This follows from ∀x ∃α ∈ ON rank(x) = α.

We have discussed cardinality as a way of measuring the size of a set.

However, if our set is not transitive, the cardinality function does not tell the
whole story since it cannot distinguish the elements of the set. For example,
although N ∈ {N}, |N| = ℵ0 while |{N}| = 1. Intuitively, we think of {N} as
no smaller than N. As such, we define the hereditary cardinality, hcard(x),
of a set x, as the cardinality of its transitive closure:

hcard(x) = |trcl(x)|

The corresponding cardinal-gradation is defined as follows.

For each cardinal κ,

H(κ) = {x : |trcl(x) < κ}.

 81

The members of H(ω) are called the hereditarily finite sets and the members
of H(ω1 ) are called the hereditarily countable sets. The next theorem lists
some important properties of H.

Theorem 37.

1. For any infinite cardinal κ, H(κ) is transitive.

2. For any infinite cardinal κ, ∀x hcard(x) < κ → rank(x) < κ.

3. For any infinite cardinal κ, H(κ) ⊆ R(κ).

4. For any infinite cardinal κ, ∃z z = H(κ).

5. ∀x ∃κ
S (κ is a cardinal and x ∈ H(κ)); i.e,
V = {H(κ) : κ is a cardinal}.

Proof.

1. Apply part (2) of Theorem 33.

2. Case 1: κ is regular.
The proof is by ∈-induction on V, noting that if rank(y) < κ for
each y ∈ x and |x| < κ, then rank(x) < κ.
Case 2: κ is singular.
There is a regular cardinal λ such that hcard(x) < λ < κ; now
use Case 1.

3. This follows from (2) and Theorem 36.

4. Apply (3) and Comprehension.

5. This follows since ∀x ∃κ hcard(x) = κ.

Theorem 38. If κ is an inaccessible cardinal, then H(κ) = R(κ).

82 CHAPTER 9. THE UNIVERSE

Proof. From Theorem 37 we have H(κ) ⊆ R(κ).

For the reverse containment, let x ∈ R(κ). By Theorem 36, ∃α < κ

such that x ∈ R(α + 1) \ R(α). So x ⊆ R(α). Since R(α) is transitive,
trcl(x) ⊆ R(α).

It suffices to prove by induction that ∀α < κ |R(α)| < κ. For successor

α = β + 1, we note that |P(λ)| < κ, where λ = |R(β)|; for limit α we apply
Theorem 36, observing that κ is regular.
Chapter 10

Reflection

There is a generalisation of the Equality Axiom, called the Equality Principle,

which states that for any formula Φ we have x = y implying that Φ holds at
x iff Φ holds at y. The proof requires a new technique, called induction on
complexity of the formula.

We make this precise. For each formula Φ of set theory all of whose free
variables lie among v0 , . . . , vn we write Φ(v0 , . . . , vn ) and for each i and j with
0 ≤ i ≤ n, we denote by Φ(v0 , . . . , vi /vj , . . . , vn ) the result of substituting vj
for each free occurance of vi .

For each formula Φ(v0 , . . . , vn ) and each i and j with 0 ≤ i ≤ n we have:

Theorem 39. Φ, i, j

∀v0 . . . ∀vi . . . ∀vn ∀vj [vi = vj → (Φ(v0 , . . . , vn ) ↔ (Φ(v0 , . . . , vi /vj , . . . , vn ))]

This is a scheme of theorems, one for each appropriate Φ, i, j.

Proof. The proof will come in two steps.

1. We first prove this for atomic formulas Φ.

83
84 CHAPTER 10. REFLECTION

Case 1 Φ is vi ∈ vk where i 6= k
This is true by Axiom of Equality.
Case 2 Φ is vk ∈ vi where i 6= k
This is true by Axiom of Extensionality.
Case 3 Φ is vi ∈ vi
This is true since by Theorem 2 both vi ∈ vj and vj ∈ vi are false.
Case 4 Φ is vk ∈ vk where i 6= k
This is true since both Φ(v0 , . . . , vn ) and Φ(v0 , . . . , vi /vj , . . . , vn )
are the same.
Case 5 Φ is vi = vk where i 6= k
This is true by Theorem 1.
Case 6 Φ is vk = vi where i 6= k
This is similar to case 5.
Case 7 Φ is vi = vi
This is true since by Theorem 1 both vi = vi and vj = vj are true.
Case 8 Φ is vk = vk where i 6= k
This is similar to case 4.
2. We now show that for any subformula Ω of Φ, if the theorem is true
for all proper subformulas of Ω then it is true for Ω. Here Θ and Ψ
are subformulas of Ω and Ω is expressed in parentheses for each case
according to how Ω is built.

Case 1 (¬Θ)
This is true since if

Θ(v0 , . . . , vn ) ↔ Θ(v0 , . . . , vi /vj , . . . , vn )

then
¬Θ(v0 , . . . , vn ) ↔ ¬Θ(v0 , . . . , vi /vj , . . . , vn )
Case 2 (Θ ∧ Ψ)
From the hypothesis that

Θ(v0 , . . . , vn ) ↔ Θ(v0 , . . . , vi /vj , . . . , vn )

and
Ψ(v0 , . . . , vn ) ↔ Ψ(v0 , . . . , vi /vj , . . . , vn )
 85

we obain
Θ(v0 , . . . , vn ) ∧ Ψ(v0 , . . . , vn )
iff
Θ(v0 , . . . , vi /vj , . . . , vn ) ∧ Ψ(v0 , . . . , vi /vj , . . . , vn ).

Cases 3 through 5 result from Cases 1 and 2.

Case 3 (Θ ∨ Ψ)
Case 4 (Θ → Ψ)
Case 5 (Θ ↔ Ψ)
Case 6 (∀vk Θ) and i 6= k
We have

∀v0 . . . ∀vn ∀vj [vi = vj → Θ(v0 , . . . , vn )

↔ Θ(v0 , . . . , vi /vj , . . . , vn )].

If vk is not free in Θ, then Θ ↔ ∀vk Θ and we are done. If vk is

free in Θ, then since 0 ≤ k ≤ n we have

∀v0 . . . ∀vn ∀vj ∀vk [vi = vj → Θ(v0 , . . . , vn )

↔ Θ(v0 , . . . , vi /vj , . . . , vn )]

and so

∀v0 . . . ∀vn ∀vj [vi = vj → (∀vk )(Θ(v0 , . . . , vn )

↔ Θ(v0 , . . . , vi /vj , . . . , vn ))]

and so

∀v0 . . . ∀vn ∀vj [vi = vj → (∀vk )Θ(v0 , . . . , vn )

↔ (∀vk )Θ(v0 , . . . , vi /vj , . . . , vn ))].

Case 7 (∃vk ) Θ) and i 6= k

This follows from Cases 1 and 6.
Case 8 (∀vi Θ)
This is true since vi is not free in Φ, hence Φ(v0 , . . . , vi /vj , . . . , vn )
is Φ(v0 , . . . , vn ).
86 CHAPTER 10. REFLECTION

Case 9 (∃vi ) Θ)
This is similar to case 8.

We conclude that the theorem scheme holds for all appropriate Φ, i, j.

Let M be a term and Φ any formula of the language of set theory. We

define the relativisation of Φ to M , denoted by ΦM , as follows:

1. If Φ is atomic then ΦM is Φ;

2. If Φ is ¬Ψ then ΦM is ¬ΨM ;

3. If Φ is (Ψ1 ∧ Ψ2 ) then ΦM is (ΨM M

1 ∧ Ψ2 );

4. If Φ is (Ψ1 ∨ Ψ2 ) then ΦM is (ΨM M

1 ∨ Ψ2 );

5. If Φ is (Ψ1 → Ψ2 ) then ΦM is (ΨM M

1 → Ψ2 );

6. If Φ is (Ψ1 ↔ Ψ2 ); then ΦM is (ΨM M

1 ↔ Ψ2 );

7. If Φ is (∀vi )Ψ then ΦM is (∀vi ∈ M )ΨM ; and,

8. If Φ is (∃vi )Ψ then ΦM is (∃vi ∈ M )ΨM .

We write M |= Φ for ΦM and moreover whenever Φ has no free variables we

say that M is a model of Φ.

We denote by ZFC the collection of axioms which include: Equality,

Extensionality, Existence, Pairing, Foundation, Union, Intersection, the Re-
placement Scheme, Power Set, Choice and Infinity.

For each axiom Φ of ZFC, except for the Axiom of Infinity, we have:

Lemma. Φ

R(ω) |= Φ.
 87

Lemma. If M is transitive, then M models Equality, Extensionality, Exis-

tence and Foundation.

For each axiom Φ of ZFC, except for those in the Replacement Scheme,
we have:

Theorem 40. For each uncountable κ, R(κ) |= Φ.

Exercise 31. Prove the above theorem scheme. Use the fact that V |= Φ.

For each axiom Φ of ZFC, except for Power Set, we have:

Theorem 41. For each uncountable regular cardinal κ, H(κ) |= Φ.

Exercise 32. Prove the above theorem scheme.

If κ is an inaccessible cardinal, then R(κ) = H(κ) |= Φ, for each axiom

Φ of ZFC.

Lemma. If κ is the least inaccessible cardinal, then

H(κ) |= ¬∃ an inaccessible cardinal.

From this lemma we can infer that there is no proof, from ZFC, that
there is an inaccessible cardinal. Suppose Θ is the conjuction of all the
(finitely many) axioms used in such a proof. Then from Θ we can derive
(∃λ)(λ is an inaccessible). Let κ be the least inaccessible cardinal. Then
H(κ) |= Θ so
H(κ) |= (∃λ)(λ is an inaccessible).
This assumes that our proof system is sound; i.e., if from Θ1 we can derive Θ2
and M |= Θ1 , then M |= Θ2 . We conclude that ZFC plus “¬∃ an inaccessible”
is consistent.

We can also infer that ZFC minus “Infinity” plus “¬(∃z)(z = N)” is
consistent. Suppose not. Suppose Θ is the conjuction of the finitely many
axioms of ZFC − Inf needed to prove (∃z)(z = N ). Then H(ω) |= Θ, so
H(ω) |= (∃z)(z = N), which is a contradiction.
88 CHAPTER 10. REFLECTION

In fact, given any collection of formulas without free variables and a class
M such that for each Φ in the collection M |= Φ, we can then conclude that
the collection is consistent.

For example, ZFC minus “Power Set” plus “(∀x)(x is countable)” is con-
sistent:

Lemma. H(ω1 ) |= (∀x)(x is countable).

If Φ(v0 , . . . , vk ) is a formula of the language of set theory and

M = {x : χM (x, ~v )} and C = {x : χC (x, ~v )} are classes, then we say Φ
is absolute between M and C whenever

(∀v0 ∈ M ∩ C) . . . (∀vk ∈ M ∩ C) [ΦM ↔ ΦC ].

This concept is most often used when M ⊆ C. When C = V, we say that

Φ is absolute for M .

For classes M = {x : χM (x, ~v )} and C = {x : χC (x, ~v )} with M ⊆ C and

a list Φ0 , . . . , Φm of formulas of set theory such that for each i ≤ m every
subformula of Φi is contained in the list, we have:

Lemma. χM , χC , Φ1 , . . . , Φm

The following are equivalent:

1. Each of Φ1 , . . . , Φm are absolute between M and C.

2. Whenever Φi is ∃x Φj (x, ~v ) for i, j ≤ m we have

(∀v1 ∈ M . . . ∀vk ∈ M )(∃x ∈ C ΦC v ) → ∃x ∈ M ΦC

j (x, ~ j (x, ~
v )).

The latter statement is called the Tarski-Vaught Condition.

Proof. ((1) ⇒ (2))

 89

Let v0 ∈ M, . . . , vk ∈ M and suppose ∃x ∈ C ΦC j (x, v0 , . . . , vk ). Then

C M
Φi (v0 , . . . , vk ) holds. By absoluteness Φi (v0 , . . . , vk ) holds; i.e.,
M
∃x ∈ M Φj (x, v0 , . . . , vk ), so by absoluteness of Φj (x, v0 , . . . , vk ) for x ∈ M
we have ∃x ∈ M ΦC j (x, v0 , . . . , vk ).

(2) ⇒ (1)

This is proved by induction on complexity of Φi , noting that each sub-

formula appears in the list. There is no problem with the atomic formula
step since atomic formulas are always absolute. Similarly, the negation and
connective steps are easy. Now if Φi is ∃x Φj and each of v1 , . . . , vk is in M
we have

ΦM M
i (v1 , . . . , vk ) ⇔ ∃x ∈ M Φj (x, v1 , . . . , vk )
⇔ ∃x ∈ M ΦC
j (x, v1 , . . . , vk )
⇔ ∃x ∈ C ΦC
j (x, v1 , . . . , vk )
⇔ ΦC
i (v1 , . . . , vk )

where the second implication is due to the inductive hypothesis and the third
implication is by part (2).

The next theorem scheme is called the Levy Reflection Principle.

For each formula Φ of the language of set theory, we have:

Theorem 42. Φ

∀α ∈ ON ∃β ∈ ON [β > α and Φ is absolute for R(β)].

If, in addition, Φ has no free variables then Φ implies R(β) |= Φ. This is

interpreted as the truth of Φ being reflected to R(β).

Proof. Form a collection Φ1 , . . . , Φm of all the subformulas of Φ. We will

use the Tarski-Vaught Condition for Φ1 , . . . , Φm to get absoluteness between
R(β) and V; but first we must find β.
90 CHAPTER 10. REFLECTION

For each i = 1, . . . , m such that Φi is ∃x Ψ for some formula Ψ we define

fi such that fi : V → ON by setting
fi (hy1 , . . . , yli i) = min {γ : γ ∈ ON ∧ ∃x ∈ R(γ) Ψ(x, y1 , . . . , yli )}
if such an ordinal exists, and fi (hy1 , . . . , yli i) = α otherwise.

Now define h : ω → ON by recursion by the formulas

h(0) = α
h(n + 1) = sup {fi (hy1 , . . . , yli i) : 1 ≤ i < m and each yj ∈ R(h(n))}
and then let β = sup {h(n) : n ∈ ω}. This β works.

The analogous theorem scheme can be proven for the H(κ) hierarchy as
well.

We can now argue that ZFC cannot be finitely axiomatised. That is,
there is no one formula without free variables which implies all axioms of
ZFC and is, in turn, implied by ZFC. Suppose such a Φ exists. By the
Levy Reflection Principle, choose the least β ∈ ON such that ΦR(β) . We
have R(β) |= Φ. Hence R(β) |= ∃α ∈ ON ΦR(α) , since this instance of the
theorem follows from ZFC. Thus,
(∃α ∈ ON ΦR(α) )R(β) .
That is,
∃α ∈ (ON ∩ R(β)) ΦR(α)∩R(β) .
But ON ∩ R(β) = β, the ordinals of rank < β. So, α < β and we have
∃α < β ΦR(α) , contradicting the minimality of β.

For any formula Φ of the language of set theory with no free variables
and classes M = {v : χM (v)}, C = {v : χC (v)}, and F = {v : χF (v)} we
have:
Lemma. Φ, χM , χC , χF

If F : M → C is an isomorphism then Φ is absolute between M and C.

That is, M |= Φ iff C |= Φ.
 91

Proof. It is easy to show by induction on the complexity of Φ that

Ψ(x1 , . . . , xk ) ⇔ Ψ(F (x1 ), . . . , F (xk ))

for each subformula Ψ of Φ.

A bounded formula (also called a 40 formula) is one which is built up as

usual with respect to atomic formulas and connectives, but where each ∃x Φ
clause is replaced by ∃x ∈ y Φ, and the ∀x Φ clause is replaced by ∀x ∈ y Φ.

40 formulas are absolute for transitive models. That is, for each 40
~ and for each class M = {x : χM (x, w)}
formula Φ(v0 , . . . , vk , w) ~ we have:
Theorem 43. Φ, χM

∀w
~ if M is transitive, then

(∀v0 ∈ M ) . . . (∀vk ∈ M ) [ΦM (v0 , . . . , vk , w)

~ ↔ Φ(v0 , . . . , vk , w)]
~

Proof. We use induction on the complexity of Φ. We only show the ∃ step.

Suppose [(∀v0 ∈ M ) . . . (∀vk ∈ M ) ΦM (v0 , . . . , vk , w)]
~ ↔ Φ(v0 , . . . , vk , w).
~
We wish to consider (∃vi ∈ vj ) Φ(v0 , . . . , vk , w).
~

Fix any v0 , . . . , vk ∈ M , but not vi ; however, vj is fixed in M . Now

~ M → (∃vi ∈ vj Φ(v0 , . . . , vk , w))

(∃vi ∈ vj Φ(v0 , . . . , vk , w)) ~

since ΦM (~v , w)
~ → Φ(~v , w).
~ Also,

(∃vi ∈ vj ) Φ(~v , w)
~
→(∃vi ∈ vj ) ΦM (~v , w)
~ since Φ(~v , w)
~
→ΦM (~v , w)
~
→∃vi ∈ (vj ∩ M ) ΦM (~v , w)
~ since vj ∈ M implies vj ⊆ M
~ M
→((∃vi ∈ vj ) Φ(~v , w))
92 CHAPTER 10. REFLECTION

A formula Φ(w)~ is said to be 4T1 , where T is a subcollection of ZFC,

whenever there are 40 formulas Φ1 (~v , w)
~ and Φ2 (~v , w)
~ such that using only
the axioms from T we can prove that both:

(∀w)(Φ
~ ~ ↔ ∃v0 . . . ∃vk Φ1 (~v , w))
1 (w) ~
(∀w)(Φ
~ ~ ↔ ∀v0 . . . ∀vk Φ2 (~v , w)).
2 (w) ~

We can now use the above theorem to show that if Φ(w) ~ is a 4T1 formula
and M |= Θ for each Θ in T , then Φ(w) ~ is absolute for M , whenever M is
transitive. We do so as follows, letting Ψ play the role of Ψ1 above:

~ ⇔ ∃v0 . . . ∃vk Ψ(~v , w)

Φ(w) ~
⇔ (∃v0 ∈ M ) . . . (∃vk ∈ M ) [Ψ(~v , w)]
~
~ M
⇔ ∃v0 . . . ∃vk Ψ(~v , w))
~ M
⇔ Φ(w)

~ being 4T1 ; the second from

The first and third implications accrue from Φ(w)
the fact that Ψ is 40 and M is transitive and models the axioms of T .

This is often used with T as ZFC without Power Set and M = H(κ).
Chapter 11

Elementary Submodels

In this chapter we shall first introduce a collection of set operations proposed

by Kurt Gödel which are used to build sets. We shall then discuss the new
concept of elementary submodel.

We now define the ordered n−tuple with the following infinitely many
formulas, thereby extending the notion of ordered pair.

hxi = x
hx, yi = {{x}, {x, y}}
hx, y, zi = hhx, yi, zi
hx1 , . . . , xn i = hhx1 , . . . , xn−1 i, xn i

The following operations shuffle the components of such tuples in a set

S.

F0 (S) = {hhu, vi, wi : hu, hv, wii ∈ S}

F1 (S) = {hu, hv, wii : hhu, vi, wi ∈ S}
F2 (S) = {hv, ui : hu, vi ∈ S}
F3 (S) = {hv, u, wi : hu, v, wi ∈ S}
F4 (S) = {ht, v, u, wi : ht, u, v, wi ∈ S}

Lemma. (Shuffle Lemma Scheme)

93
94 CHAPTER 11. ELEMENTARY SUBMODELS

For any m ∈ N and any permutation σ : m → m, there is a composition of

the operations F0 , F1 , F2 , F3 , F4 such that for any S,

Fσ (S) = {hxσ(0) , . . . , xσ(m−1) i : hx0 , . . . , xm−1 i ∈ S}.

Proof. Since binary exchanges generate the symmetric group, noting that
the identity permutation is given by F2 ◦ F2 , it suffices to consider only σ
such that for some l ≤ m,

i + 1, if i = l;

σ(i) = i − 1, if i = l + 1;

i, otherwise.


For convenience, let Fin denote the n−fold composition of Fi . There are
several cases.

Case 1: m = 2
Fσ = F2

Case 2: m = 3, l = 1
Fσ = F3

Case 3: m = 3, l = 2
Fσ = F0 ◦ F2 ◦ F3 ◦ F2 ◦ F1

Case 4: m ≥ 4, l = 1
Fσ = F0m−3 ◦ F3 ◦ F1m−3

Case 5: m ≥ 4, 1 ≤ l ≤ m − 1
Fσ = F0m−l−2 ◦ F4 ◦ F1m−l−2

Case 6: m ≥ 4, l = m − 1
F σ = F0 ◦ F2 ◦ F3 ◦ F2 ◦ F1

The Gödel Operations are as follows:

 95

G1 (X, Y ) = {X, Y }
G2 (X, Y ) = X \ Y
G3 (X, Y ) = {hu, vi : u ∈ X ∧ v ∈ Y }; i.e., X × Y
G4 (X, Y ) = {hu, vi : u ∈ X ∧ v ∈ Y ∧ u = v}
G5 (X, Y ) = {hu, vi : u ∈ X ∧ v ∈ Y ∧ u ∈ v}
G6 (X, Y ) = {hu, vi : u ∈ X ∧ v ∈ Y ∧ v ∈ u}
G7 (X, R) = {u : ∃x ∈ X hu, xi ∈ R}
G8 (X, R) = {u : ∀x ∈ X hu, xi ∈ R}
G9 (X, R) = F0 (R)
G10 (X, R) = F1 (R)
G11 (X, R) = F2 (R)
G12 (X, R) = F3 (R)
G13 (X, R) = F4 (R)

G9 through G13 are defined as binary operations for conformity.

We now construct a function which, when coupled with recursion on ON,

will enumerate all possible compositions of Gödel operations. G : N × V → V
is given by the following rule. For each k ∈ ω and each ~s : k → V, we define
G|N×~s by recursion on N as follows:

~s(i),
 if n = 17i , 0 ≤ i ≤ k;
G(n, ~s) = Gm (G(i, ~s), G(j, ~s)), if n = m · 19i+1 23j+1 , 1 ≤ m ≤ 13;

∅, otherwise.


Now G enumerates all compositions of Gödel operations.

We wish to prove that sets defined by 40 formulas can be obtained

through a composition of Gödel operations. First we need a lemma.

Lemma. Each of the following sets is equal to a composition of Gödel oper-

ations on X, w.
~
96 CHAPTER 11. ELEMENTARY SUBMODELS

1. {hx1 , . . . , xm i : x1 ∈ X ∧ · · · ∧ xm ∈ X ∧ wi = wj }

2. {hx1 , . . . , xm i : x1 ∈ X ∧ · · · ∧ xm ∈ X ∧ wi ∈ wj }

3. {hx1 , . . . , xm i : x1 ∈ X ∧ · · · ∧ xm ∈ X ∧ wi = xj }

4. {hx1 , . . . , xm i : x1 ∈ X ∧ · · · ∧ xm ∈ X ∧ wi ∈ xj }

5. {hx1 , . . . , xm i : x1 ∈ X ∧ · · · ∧ xm ∈ X ∧ xi ∈ wj }

6. {hx1 , . . . , xm i : x1 ∈ X ∧ · · · ∧ xm ∈ X ∧ xi = xj }

7. {hx1 , . . . , xm i : x1 ∈ X ∧ · · · ∧ xm ∈ X ∧ xi ∈ xj }

Proof. First note that

{hx1 , . . . , xm i : x1 ∈ X ∧ · · · ∧ xm ∈ X} = G3 (G3 . . . (G3 (X, X), . . . X), X)

where G3 is composed (m − 1)−fold; i.e., (. . . ((X × X) × X) × · · · × X).

Now call this set Pm (X), the mth power of X. Let us now examine each of
the seven cases individually.

1. This is either Pm (X) or ∅ depending upon whether or not wi = wj .

2. This is either Pm (X) or ∅ depending upon whether or not wi ∈ wj .

3. If m = 1, then this set is hwi i = wi . For m > 1, we may, thanks to the

Shuffle Lemma, without loss of generality assume that j = m. The set
is equal to
G3 (Pm−1 (X), wi )
if wi ∈ X and ∅ otherwise.

4. Again, without loss of generality, we assume j = m. This set is equal

to
G3 (Pm−1 (X), G7 (X, G6 (X, {wi }))).

5. Again, assume j = m. This set is given by

G3 (Pm−1 (X), G7 (X, G5 (X, {wi }))).

 97

6. This time, assume i = m − 1 and j = m. This set is

G3 (Pm−2 (X), G4 (X, X)).

7. We assume again that i = m − 1 and j = m. This set becomes

G3 (Pm−2 (X), G5 (X, X)).

For each formula Φ(x, w)

~ of the language of set theory we have:

Theorem 44. Φ

(∀w)(∀X)(∃n
~ ∈ ω) [{x ∈ X : ΦX (x, w)}
~ = G(n, ~s)]
where ~s(i) = wi for i < k and s(k) = X.

Proof. We prove by induction on the complexity of Φ that for all m ∈ ω and

Φ

(∀w)(∃n
~ ∈ ω)[{hx1 , . . . , xm i : x1 ∈ X ∧ · · · ∧ xm ∈ X
and ΦX (x1 , . . . , xm , w)}
~ = G(n, ~s)]

The proof of the theorem for any given Φ will assume the corresponding
result for a finite number of simpler formulas, the proper subformulas of Φ.

We begin by looking at atomic formulas Φ. This step is covered by the

previous lemma.

Now we proceed by induction on complexity. Suppose that

{hx1 , . . . , xm i : x1 ∈ X ∧ · · · ∧ xm ∈ X and ΦX
1 (x1 , . . . , xm , w)}
~ = G(n1 , ~s)

and

{hx1 , . . . , xm i : x1 ∈ X ∧ · · · ∧ xm ∈ X and ΦX
2 (x1 , . . . , xm , w)}
~ = G(n2 , ~s).
98 CHAPTER 11. ELEMENTARY SUBMODELS

Then

{hx1 , . . . , xm i : x1 ∈ X . . . xm ∈ X and ¬ΦX

1 (x1 , . . . , xm , w)}
~
= Pm (X) \ G(n1 , ~s) = G2 (Pm (X), G(n1 , ~s))

and

{hx1 , . . . , xm i : x1 ∈ X . . . xm ∈ X and ΦX X
1 ∧ Φ2 (x1 , . . . , xm , w)}
~
= G(n1 , ~s) ∩ G(n2 , ~s).

The other connectives can be formed from ¬ and ∧; as such, ∀xl is ¬∃xl ,
so it only remains to do the Φ = ∃xl Φ1 step. Thanks again to the last
lemma, we may assume that l = m. Then Φ is ∃xm Φ1 and

{hx1 . . . xm−1 i : x1 ∈ X ∧ · · · ∧ xm−1 ∈ X and ΦX (x1 , . . . , xm−1 , w)}

~
= G7 (G(n1 , ~s))

Let’s write G(n, X, ~y ) for G(n, ~s), where ~s(0) = X and ~s(k + 1) = ~y (k)
for all k ∈ dom(~y ) = dom(~s) − 1.

M is said to be an elementary submodel of N whenever

1. M ⊆ N ; and,
k
2. ∀k ∈ ω ∀~y ∈ M ∀n ∈ ω G(n, N, ~y ) ∩ N 6= ∅ ⇔ G(n, N, ~y ) ∩ M 6= ∅.

We write M ≺ N .

Justification of the terminology comes from the following theorem scheme.

For each formula Φ of the language of set theory we have:

Theorem 45. Φ

Suppose M ≺ N . Then Φ is absolute between M and N .

 99

Proof. We will use the Tarski-Vaught criterion. Let Φ0 , . . . , Φm enumer-

ate Φ and each of its subformulas. Suppose Φi is ∃x Φj (x, y0 , . . . , yk ) with
y0 , . . . , yk ∈ M in a sequence ~y .

Let n ∈ ω so that by Theorem 44 we can find n ∈ ω such that

G(n, N, ~y ) = {x ∈ N : ΦN
j (x, y0 , . . . , yk )}.

Then

∃x ∈ N ΦN
j (x, y0 , . . . , yk ) ↔ G(n, N, ~
y ) ∩ N 6= ∅
↔ G(n, N, ~y ) ∩ M 6= ∅
↔ ∃x ∈ M ΦN
j (x, y0 , . . . , yk ).

The following is sometimes called the Lowenheim-Skolem-Tarski theorem.

Theorem 46. Suppose X ⊆ N . Then there is an M such that

1. M ≺ N ;

2. X ⊆ M ; and,

3. |M | ≤ max{ω, |X|}.

Proof. Define F : ω × {k N : k ∈ ω} → N by choice:

S

(
some element of G(n, N, ~s) if G(n, N, ~s) 6= ∅
F (n, ~s) =
any element of N otherwise

Now define {Xn }n∈ω by recursion on N as follows:

X0 = X
[
Xm+1 = Xm ∪ F 00 (ω × {k (Xm ) : k ∈ ω})
100 CHAPTER 11. ELEMENTARY SUBMODELS
S
Let M = m∈ω Xm . As such, (2) and (3) are clearly satisfied. To check
(1), let ~y ∈ k (Xm ). If G(n, N, ~y ) ∩ N 6= ∅, then F (n, ~y ) ∈ Xm+1 ⊆ M and
G(n, N, ~y ) ∩ M 6= ∅.

The use of elementary submodels of the H(θ) can be illustrated.

Theorem 47. (Pressing Down Lemma)

Let f : ω1 \ {0} → ω1 be regressive; i.e., f (α) < α for all α.

Then ∃β ∈ ω1 such that f ← {β} is uncountable.

Theorem 48. (Delta System Lemma)

Let A be an uncountable collection of finite sets.

Then ∃D ⊆ A ∃R such that

1. D is uncountable, and

2. ∀D1 , D2 ∈ D D1 ∩ D2 = R.

We need some lemmas. Assume M ≺ H(θ) where θ is an uncountable

regular cardinal. For each 40 formula Φ(v0 , . . . , vk ) we have:

Lemma. Φ

(∀y0 ∈ M ) . . . (∀yk ∈ M ) [M |= Φ(y0 , . . . , yk ) ⇔ Φ(y0 , . . . , yk )].

Proof.

M |= Φ(y0 , . . . , yk ) ⇔ H(θ) |= Φ(y0 , . . . , yk ) by elementarity,

⇔ Φ(y0 , . . . , yk ) since H(θ) is transitive.
 101

Remark. The same is true for 4T1 formulas where T is ZFC without Power
Set.

For any formula Φ(v0 , . . . , vk ) of LOST, we have:

Lemma. Φ

∀y0 ∈ M ∀y2 ∈ M . . . ∀yk ∈ M ∀x ∈ H(θ)

[H(θ) |= z = {x : Φ(x, y0 , . . . , yk )} → z ∈ M ].

Proof. Let y0 , . . . , yk ∈ M and z ∈ H(θ) be given such that

H(θ) |= z = {x : Φ(x, y0 , . . . , yk )}.

Then,

H(θ) |= ∃u u = {x : Φ(x, y0 , . . . , yk )}
⇒M |= ∃u u = {x : Φ(x, y0 , . . . , yk )}
⇒∃p ∈ M [M |= p = {x : Φ(x, y0 , . . . , yk )}]
⇒H(θ) |= p = {x : Φ(x, y0 , . . . , yk )}
⇒H(θ) |= p = z.

H(θ) is transitive; therefore, p = z and hence z ∈ M .

Corollaries. 1. If M ≺ H(θ), then

(a) ∅ ∈ M ;
(b) ω ∈ M ; and,
(c) ω ⊆ M .

2. If also θ > ω1 , then ω1 ∈ M .

Proof. ∅ and ω are direct. For ω ⊆ M show that y ∈ M ⇒ y ∪ {y} ∈ M .

102 CHAPTER 11. ELEMENTARY SUBMODELS

Lemma. Suppose M ≺ H(θ) where θ is regular and uncountable. Suppose

p is countable and p ∈ M . Then p ⊆ M .

Proof. Let q ∈ p; we must show that q ∈ M . Let f0 : ω → p be a surjection.

Since {ω, p} ⊂ H(θ) we must have f0 ∈ H(θ). Since the formula “f : ω →
p and p is surjective” is a 40 formula and {f0 , ω, p} ⊂ H(θ), we have H(θ) |=
(f0 : ω → p and p is surjective). So

H(θ) |= (∃f )(f : ω → and p is surjective).

Since {ω, p} ⊂ M we have,

M |= (∃f )(f : ω → p and p is surjective).

That is, (∃fp ∈ M )(fp : ω → p and p is surjective).

Pick n ∈ ω such that fp (n) = q, and again use the first lemma as follows.
Since {p, fp , n} ⊂ M and (∃!x)(x ∈ p and fp (n) = x) is a 40 formula

M |= (∃!x)(x ∈ p and fp (n) = x).

That is, (∃!x)(x ∈ p ∩ M and fp (n) = x). Since x is unique, x = q and thus
q ∈ M.

Corollary. ω1 ∩ M ∈ ω1 .

Proof. It is enough to show that ω1 ∩ M is a countable initial segment of ω1 .

If α ∈ ω1 ∩ M , then by the above lemma, α ⊆ M .

Proof of Pressing Down Lemma

Let M ≺ H(ω2 ) such that M is countable and f ∈ M . Let δ = ω1 ∩ M

and let β = f (δ) < δ. Then,

(∀α < δ)(∃x ∈ ω1 ) [x > α ∧ f (x) = β].

 103

So ∀α < δ H(ω2 ) |= (∃x ∈ ω1 )(x > α ∧ f (x) = β), since everything relevant
is in H(ω2 ). Hence,
∀α < δ M |= (∃x ∈ ω1 )(x > α ∧ f (α) = β)
since {α, β, ω1 , f } ⊂ M . Now, since δ = ω1 ∩ M we have,
M |= (∀α ∈ ω1 )(∃x ∈ ω1 ) [x > α ∧ f (α) = β].
So H(ω2 ) |= (∀α ∈ ω1 )(∃x ∈ ω1 ) [x > α ∧ f (α) = β]. Thus we have
H(ω2 ) |= f ← {β} is uncountable.
Again, since everything relevant is in H(ω2 ) we conclude that f ← {β} is
uncountable.

Proof of the Delta System Lemma

Let A be as given. We may, without loss of generosity, let

A = {a(α) : α < ω1 }
where a : ω1 → V. We may also assume that a : ω1 → P(ω1 ).

Let M be countable with {A, a} ⊆ M and M ≺ H(ω2 ). Let δ = ω1 ∩ M .

Let R = a(δ) ∩ δ. Since R ⊆ M , we know R ∈ M by the second lemma. So,
∀α < δ ∃β > α a(β) ∩ β = R
⇒H(ω2 ) |= (∀α < δ)(∃β > α) [a(β) ∩ β = R]
⇒(∀α < δ) [H(ω2 ) |= (∃β > α)(a(β) ∩ β = R)]
⇒(∀α < δ) [M |= (∃β > α)(a(β) ∩ β = R)]
⇒M |= (∀α < ω1 )(∃β > α) [a(β) ∩ β = R]
⇒(∀α < ω1 )(∃β > α) [a(β) ∩ β = R].
Now recursively define D : ω1 → A as follows:
D(α) = a(0);
D(γ) = a(β)
104 CHAPTER 11. ELEMENTARY SUBMODELS

where β is the least ordinal such that

β > sup {D(γ) : γ < α} and a(β) ∩ β = R.

Now if γ1 < γ2 < ω1 , then D(γ1 ) ⊆ γ2 . So,

R ⊆ D(γ1 ) ∩ D(γ2 ) ⊆ γ2 ∩ D(γ2 ) = R.

Thus we let D = {D(α) : α < ω1 }.

Theorem 49. (Elementary Chain Theorem)

Suppose that δ is a limit ordinal and {Mα : α < δ} is a set of elementary

submodels of H(θ) such that

∀α ∀α0 (α < α0 < δ → Mα ⊂ Mα0 ).

Let [
Mδ = {Mα : α < δ}.
Then Mδ ≺ H(θ).

k
Proof. Let k ∈ ω, let ~y ∈ Mδ , and let n ∈ ω. We need to show that

G(n, H(θ), ~y ) ∩ H(θ) 6= ∅ ⇒ G(n, H(θ), ~y ) ∩ Mδ 6= ∅.

k
But this is easy since ~y ∈ Mα for some α < δ.
Chapter 12

Constructibility

The Gödel closure of a set X is denoted by

k
cl(X) = {X ∩ G(n, ~y ) : n ∈ ω and ∃k ∈ ω ~y ∈ (X)}.

The constructible sets are obtained by first defining a function

L : ON → V
by recursion as follows:
L(0) = ∅
L(α + 1) = cl(L(α) ∪ {L(α)})
[
L(δ) = {L(α) : α < δ} for a limit ordinal δ

S
We denote by L the class {L(α) : α ∈ ON}. Sets in L are said to be
constructible.
Lemma. For each ordinal α, L(α) ⊆ R(α).

Proof. This is proved by induction. L(0) = ∅ = R(0) and for each α ∈ ON

we have, by definition,
L(α + 1) ⊆ P(L(α))
⊆ R(α + 1)

105
106 CHAPTER 12. CONSTRUCTIBILITY

Lemma.

1. ∀X X ⊆ cl(X).

2. If X is transitive, then cl(X) is transitive.

3. For each ordinal α, L(α) is transitive.

Proof.

1. For any w ∈ X, w = G(1, ~s), where ~s(0) = w.

2. Now, if z ∈ cl(X) then z ⊆ X so z ⊆ cl(X).

3. This follows from (1) by induction on ON.

Lemma.

1. For all ordinals α < β, L(α) ∈ L(β).

2. For all ordinals α < β, L(α) ⊆ L(β).

Proof.

1. For each α, L(α) ∈ L(α + 1) by Part (1) of the previous lemma. We

then apply induction on β.

2. This follows from (1) by transitivity of L(β).

Lemma.

1. For each ordinal β, β ∈

/ L(β).

2. For each ordinal β, β ∈ L(β + 1).

 107

Proof.

1. This is proved by induction on β. The case β = 0 is easy. If β = α + 1

then β ∈ L(α + 1) would imply that β ⊆ L(α) and hence

α ∈ β ⊆ L(α)

contradicting the inductive hypothesis. If β is a limit ordinal and β ∈

L(β) then β ∈ L(α) for some α ∈ β and hence α ∈ β ∈ L(α), again a
contradiction.

2. We employ induction on β. The β = 0 case is given by 0 ∈ {0}. We

do the sucessor and limit cases uniformly. Assume that

∀α ∈ β α ∈ L(α + 1).

Claim 1. β = L(β) ∩ ON.

Proof of Claim 1. If α ∈ β, then α ∈ L(α + 1) ⊆ L(β). If α ∈ L(β),
then α ∈ β because otherwise α = β or β ∈ α, which contradicts
β∈/ L(β) from (1).
Claim 2. ∀x x ∈ ON iff

[(∀u ∈ x ∀v ∈ u v ∈ x) ∧ (∀u ∈ x ∀v ∈ x (u ∈ v ∨ v ∈ u ∨ u = v))

∧(∀u ∈ x ∀v ∈ x ∀w ∈ x (u ∈ v ∧ v ∈ w → u ∈ w))].

Proof of Claim 2. The statement says that x is an ordinal iff x is a tran-

sitive set and the ordering ∈ on x is transitive and satisfies trichotomy.
This is true since ∈ is automatically well founded.

The importance of this claim is that this latter formula, call it Φ(x), is
40 and hence absolute for transitive sets.
We have:

β = L(β) ∩ ON = {x ∈ L(β) : x is an ordinal}

= {x ∈ L(β) : Φ(x)}
= {x ∈ L(β) : ΦL(β+1) (α)}
∈ cl(L(β)) using Theorem 44
= L(β + 1).
108 CHAPTER 12. CONSTRUCTIBILITY

Lemma.

1. For each ordinal β, β = L(β) ∩ ON.

2. ON ⊆ L.

Proof. This is easy from the previous lemmas.

Lemma.

1. If W is a finite subset of X then W ∈ cl(X).

2. If W is a finite subset of L(β) then W ∈ L(β + 1).

Proof.

1. We apply Theorem 44 to the formula “x = w0 ∨ · · · ∨ x = wn ”, where

W = {w1 , w2 , . . . , wn }.

2. This follows immediately from (1).

Lemma.

1. If X is infinite then |cl(X)| = |X|.

2. α ≥ ω then |L(α)| = |α|.

Proof.

1. By Theorem 44 we can construct an injection cl(X) → ω × {k X : k ∈

S
ω}. Hence, |X| ≤ |cl(X)| ≤ max (ℵ0 , |{k X : k ∈ ω}|) = |X|.
 109

2. We proceed by induction, beginning with the case α = ω. We first note

that from the previous lemma, we have L(n) = R(n) for each n ∈ ω.
Therefore,
[
|L(ω)| = | {L(n) : n ∈ ω}|
= max (ℵ0 , sup {|L(n)| : n ∈ ω})
= max (ℵ0 , sup {|R(n)| : n ∈ ω})
= ℵ0 .

For the successor case,

|L(β + 1)| = |L(β)| by (1)

= |β| by inductive hypothesis
= |β + 1| since β is infinite .

And if δ is a limit ordinal then

[
|L(δ)| = | {L(β) : β ∈ δ}|
= max (|δ|, sup {|L(β)| : β ∈ δ})
= max (|δ|, sup {|(β)| : β ∈ δ}) by inductive hypothesis
= |δ|.

Lemma. (∀x) [x ⊆ L → (∃y ∈ L)(x ⊆ y)].

Proof. x ⊆ L means that ∀u ∈ x ∃α ∈ ON x ∈ L(α). By the Axiom of

Replacement,

∃z z = {α : (∃u ∈ x)(α is the least ordinal such that u ∈ L(α))}.

Let β = sup z; then β ∈ ON and for each u ∈ x, there is α ≤ β such that

u ∈ L(α) ⊆ L(β). Since L(β) ∈ L(β + 1) ⊆ L, we can take y = L(β).

Remark. The above lemma is usually quoted as “L is almost universal”.

110 CHAPTER 12. CONSTRUCTIBILITY

Lemma. L |= V = L.

Proof. This is not the trivial statement

∀x ∈ L x ∈ L

but rather
∀x ∈ L (x ∈ L)L
which is equivalent to (∀x ∈ L)(∃α ∈ ON x ∈ L(α))L ; which is, in turn, since
ON ⊆ L, equivalent to (∀x ∈ L)(∃α ∈ ON)(x ∈ L(α))L .

This latter statement is true since “x ∈ L(α)” is a 40 formula when

written out in full in LOST, and since L is transitive.

For each Axiom Φ of ZFC we have:

Theorem 50. Φ

L |= Φ.

Proof. Transitivity of L automatically gives Equality, Extensionality, Exis-

tence and Foundation. We get Infinity since ω ∈ L and “z = N” is a 40
formula.

For Comprehension, let Φ be any formula of LOST; we wish to prove

∀y ∈ L ∀w0 ∈ L . . . ∀wn ∈ L ∃z ∈ L z = {x ∈ y : ΦL (x, y, w0 , . . . , wn )}

since L is transitive.

Fix y, w0 , . . . , wn and α ∈ ON such that {y, w}

~ ⊆ L(α). By the Levy
Reflection Principle, there is some β > α such that Φ is absolute between L
and L(β).

By Theorem 44, there is an n ∈ ω such that

~ = {x ∈ L(β) : ΦL(β) (x, y, w)}.

G(n, L(β), y, w) ~
 111

and so by definition, {x ∈ L(β) : ΦL(β) (x, y, w)}

~ ∈ L(β + 1). Now by
L(β)
absoluteness, {x ∈ L(β) : Φ (x)} = {x ∈ L(β)Φ (x)}. So we have
L

{x ∈ L(β) : ΦL (x, y, w)}

~ ∈ L(β + 1).

Moreover, since y ∈ L(β + 1),

{x ∈ y : ΦL (x, y, w)}
~ = y ∩ {x ∈ L(β + 1) : ΦL (x, y, w)}
~
∈ L(β + 2)

and since L(β + 2) ⊆ L we are done.

For the Power Set Axiom, we must prove that (∀x ∃z z = {y : y ⊆ x})L .
That is, ∀x ∈ L ∃z ∈ L z = {y : y ∈ L and y ⊆ x}. Fix x ∈ L; by the Power
Set Axiom and the Axiom of Comprehension we get

∃z 0 z 0 = {y ∈ P(x) : y ∈ L ∧ y ⊆ x} = {y : y ∈ L ∧ y ⊆ x}.

By the previous lemma L is almost universal and z 0 ⊆ L so

∃z 00 ∈ L z 0 ⊆ z 00 .

So z 0 = z 0 ∩ z 00 = {y ∈ z 00 : y ∈ L ∩ y ⊆ x}. By the fact that the Axiom of

Comprehension holds relativised to L we get

(∃z z = {y ∈ z 00 : y ⊆ x})L ;

i.e.,

∃z ∈ L z = {y ∈ z 00 : y ∈ L ∧ y ⊆ x}
= {y : y ∈ L ∧ y ⊆ x}

The Union Axiom and the Replacement Scheme are treated similarly. To
prove (the Axiom of Choice) L , we will show that the Axiom of Choice follows
from the other axioms of ZFC with the additional assumption that V = L.

It suffices to prove that for each α ∈ ON there is a β ∈ ON and a

surjection fα : bα → L(α).
112 CHAPTER 12. CONSTRUCTIBILITY

To do this we define fα recursively. Of course f0 = β0 = ∅ = L(0). If α

is a limit ordinal, then we let
X
βα = {β :  < α}
P
and fα (σ) = fδ (τ ) where σ = {β :  < δ} + τ and τ < βδ .

If α = γ + 1 is a successor ordinal, use fγ to generate a well ordering of

L(γ)
S k and use this well ordering to generate a lexicographic well ordering of
¯
{ (L(γ)) : k ∈ ω} and use this to obtain an ordinal βα and a surjection
[
f¯γ : β¯γ → {k (L(γ)) : k ∈ ω}.

Now let βα = βγ+1 = β¯γ × ω and let

fα : βα → L(α) = {G(n, L(α), ~y ) : n ∈ ω and ∃k ∈ ω ~y ∈k L(γ)}

be defined by fα (σ) = G(n, L(γ), f¯γ (τ )) where σ = β¯γ × n + τ , τ < β¯γ .

This completes the proof of Theorem 50 Φ and motivates calling “V = L”

the Axiom of Constructibility.

Remark. V = L is consistent with ZFC in the sense that no finite subcollec-

tion of ZFC can possibly prove V 6= L; To see this, suppose

{Ψ0 , . . . , Ψn } ` V 6= L.

Then
ΨL0 , . . . , ΨLn } ` (V 6= L)L .
by Theorem 50. This contradicts the preceding lemma.
Remark. Assuming V = L we actually can find a formula Ψ(x, y) which gives
a well ordering of the universe.

We denote by ΦL the conjunction of a finite number of axioms of ZFC

conjoined with “V = L” such that ΦL implies all our lemmas and theorems
about ordinals and ensures that x ∈ L(α) is equivalent to some 40 formula
 113

(but I think we have already defined it to be 40 ). In particular, x ∈ ON will

be equivalent to a 40 formula.

Furthermore, we explicitly want ΦL to imply that ∀α ∈ ON ∃z z = L(α)

and that there is no largest ordinal.

We shall use the abbreviation o(M ) = ON ∩ M .

Lemma. ∀M (M is transitive and ΦM

L → M = L(o(M ))).

Proof. Let M be transitive such that M |= ΦL . Note that o(M ) ∈ ON. We

have M |= ∀α ∈ ON ∃z z = L(α). So,

∀α ∈ o(M ) M |= ∃z z = L(α)
⇒∀α ∈ o(M ) ∃z ∈ M M |= z = L(α)
⇒∀α ∈ o(M ) ∃z ∈ M z = L(α)
⇒∀α ∈ o(M ) L(α) ∈ M
⇒∀α ∈ o(M ) L(α) ⊆ M.

Since M |= ΦL , o(M ) is a limit ordinal and hence

[
L(o(M )) = {L(α) : α ∈ o(M )} ⊆ M.

Now let a ∈ M . Since M |= V = L we have

M |= ∀x ∃y ∈ ON x ∈ L(y)
⇒M |= ∃y ∈ ON a ∈ L(y)
⇒∃α ∈ o(M ) M |= a ∈ L(α)
⇒∃α ∈ o(M ) a ∈ L(α)
⇒a ∈ L(o(M )).

Lemma. χC

If ON ⊆ C, C is transitive, and ΦC
L , then C = L.
114 CHAPTER 12. CONSTRUCTIBILITY

Proof. The proof is similar to that of the previous lemma.

Theorem 51. (K. Gödel)

If V = L then GCH holds.

Proof. We first prove the following:

Claim. ∀α ∈ ON P(L(α)) ⊆ L(α+ ).
Proof of Claim. This is easy for finite α, since L(n) = R(n) for each n ∈ ω.

Let’s prove the claim for infinite α ∈ ON. Let X ∈ P(L(α)); we will
show that X ∈ L(α+ ).

Let A = L(α) ∪ {X}. A is transitive and |A| = |α|.

By the Levy Reflection Principle, there is a β ∈ ON such that both

A ⊆ L(β) and L(β) |= ΦL , where ΦL is the formula introduced earlier.

Now use the Lowenheim-Skolem-Tarski Theorem to obtain an elementary

submodel K ≺ L(β) such that A ⊆ K and |K| = |A| = |α| so by elementarily
we have K |= ΦL .

Now use the Mostowski Collapsing Theorem to get a transitive M such

that K ∼
= M . Since A is transitive, the isomorphism is the indentity on A
and hence A ⊆ M . We also get M |= ΦL and |M | = |α|.

Now we use the penultimate lemma to infer that M = L(o(M )). Since
|M | = |α| we have |o(M )| = |α| so that o(M ) < |α+ |.

Hence A ⊆ M = L(o(M )) ⊆ L(α+ ), so that X ∈ L(α+ ).

We now see that the GCH follows from the claim. For each cardinal κ
we have κ ⊆ L(κ) so that |P(κ)| ≤ |P(L(κ))| ≤ |L(κ+ ).

Since |L(κ+ )| = κ+ we have |P (κ)| = κ+ .

 115

We now turn our attention to whether V = L is true.

Let µ be a cardinal and let U be an ultrafilter over µ. Recalling that

µ
V = {f : f : µ → V}, let ∼U be a binary relation on µ V defined by

f ∼U g iff {α ∈ µ : f (α) = g(α)} ∈ U.

It is easy to check that ∼U is an equivalence relation.

µ
For each f ∈ V let ρ(f ) be the least element of

{α ∈ ON : rank(g) = α ∧ f ∼U g}.
µ
Let [f ] = {g ∈ R(ρ(f ) + 1) : g ∼U f } and let U LTU V = {[f ] : f ∈ V}.

Define a relation ∈U on U LTU V by

[f ] ∈U [g] iff {α ∈ µ : f (α) ∈ g(α)} ∈ U.

It is easy to check that ∈U is well defined.

For each cardinal κ, we use the abbreviation

[X]<κ = {Y ⊆ X : |Y | < κ}.

Given an uncountable cardinal κ, an ultrafilter U is said to be κ−complete

if ∀X ∈ [U]<κ
T
X ∈ U.

An uncountable cardinal κ is said to be measurable whenever there exists

a κ−complete free ultrafilter over κ.
Lemma. If U is a countably complete ultrafilter (in particular if U is a
µ−complete ultrafilter) then ∈U is set-like, extensional and well founded.

Proof. To see that ∈U is set-like, just note that

{[g] : [g] ∈U [f ]} ⊆ R(ρ(f ) + 2).

For extentionality, suppose [f ] 6= [g]; i.e., {α ∈ µ : f (α) = g(α)} ∈

/ U.
Then either {α ∈ µ : ¬f (α) ⊆ g(α)} ∈ U or {α ∈ µ : ¬g(α) ⊆ f (α)} ∈ U.
This leads to two similar cases; we address the first.
116 CHAPTER 12. CONSTRUCTIBILITY

Pick any h ∈ µ V such that h(α) ∈ f (α) \ g(α) whenever ¬f (α) ⊆ g(α).
Then [h] ∈U [f ] and [h] ∈U [g].

To see that ∈U is well founded, suppose ∃{fn }n∈ω such that

∀n ∈ ω [fn+1 ] ∈U [fn ].

Let \
A= {{α ∈ µ : fn+1 (α) ∈ fn (α)} : n ∈ ω} ∈ U.
A ∈ U by the countable completeness of U, so that A 6= ∅. Pick any β ∈ A.
Then Fn+1 (β) ∈ fn (β) for each n ∈ ω, which is a contradiction.

We now create a Mostowski collapse of U LTU V

hU : U LTU V → MU

given by the recursion

hU ([f ]) = {hU ([g]) : [g] ∈U [f ]}

As per the Mostowski Theorem, h is an isomorphism and MU is transitive.

The natural embedding iU : V → U LTU V is given by iU (x) = [fx ] where

fx : µ → V such that fx (α) = x for all α ∈ µ.

This natural embedding iU combines with the unique isomorphism hU to

give
jU : V → M U
given by jU (x) = hU (iU (x)).

jU is called the elementary embedding generated by U, since for all for-

mulas Φ(v0 , . . . , vn ) of LOST we have:
Lemma. ∀v0 . . . ∀vn Φ(v0 , . . . , vn ) ↔ ΦMU (jU (v0 ), . . . , jU (vn )).

Proof. This follows from two claims, each proved by induction on the com-
plexity of Φ.
 117

Claim 1. ∀v0 . . . ∀vn Φ(v0 , . . . , vn ) ↔ Φ̄(iU (v0 ), . . . , iU (vn )).

Claim 2. ∀v0 . . . ∀vn Φ̄(iU (v0 ), . . . , iU (vn )) ↔ ΦMU (jU (v0 ), . . . , jU (vn )), where
Φ̄ is Φ with ∈ replaced by ∈U and all quantifiers restricted to U LTU V.

We leave the proofs to the reader.

Theorem 52. Every measurable cardinal is inaccessible.

Proof. We first prove that κ is regular. If cf (κ) = λ < κ, then κ is the union
of λ sets each smaller than κ. This contradicts the existence of a κ−complete
free ultrafilter over κ.

We now prove that if λ < κ, then |P(λ)| < κ. Suppose not; then there
is X ∈ [P(λ)]κ and a κ−complete free ultrafilter U over X. Now, for each
α ∈ λ let Aα = {x ∈ X : α ∈ x} and Bα = {x ∈ X : α ∈ / x}. Let
I = {α ∈ λ : Aα ∈ U} and J = {α ∈ λ : Bα ∈ U}. Since U is an ultrafilter,
I ∪ J = λ. Since U is κ−complete and λ < κ we have
\ \
{Aα : α ∈ I} ∩ {Bα : α ∈ J} ∈ U.

But this intersection is equal to X ∪{I}, which is either empty or a singleton,

contradicting that U is a free filter.

Lemma. Let U be a µ−complete ultrafilter over an measurable cardinal µ.

Let M = MU , h = hU , i = iU and j = jU as above. Then for each β ∈ ON
we have j(β) ∈ ON and j(β) ≥ β. Furthermore, if β < µ then j(β) = β and
j(µ) > µ.

Proof. For each β ∈ ON we get, by the elementary embedding property of j,

that M |= j(β) ∈ ON; since M is transitive, j(β) ∈ ON.

Let β be the least ordinal such that j(β) ∈ β. Then M |= j(j(β)) ∈ j(β)
by elementarity, and j(j(β)) ∈ j(β) by transitivity of M . This contradicts
the minimality of β.
118 CHAPTER 12. CONSTRUCTIBILITY

Now let’s prove that j(β) = β for all β < µ by induction on β. Suppose
that j(γ) = γ for all γ < β < µ. We have
j(β) = h(i(β))
= {h([g]) : [g] ∈U i(β)}
= {h([g]) : [g] ∈U [fβ ]} where fβ (α) = β for all α ∈ µ
= {h([g]) : {α ∈ µ : g(α) ∈ fβ (α)} ∈ U}
= {h([g]) : {α ∈ µ : g(α) ∈ β} ∈ U}
= {h([g]) : ∃γ ∈ β {α ∈ µ : g(α) = γ} ∈ U} by µ − completeness of U
= {h([g]) : ∃γ ∈ β [g] = [fγ ]} where fγ (α) = γ for all α ∈ µ
= {h([fγ ]) : γ ∈ β}
= {h(i(γ)) : γ ∈ β}
= {j(γ) : γ ∈ β}
= {γ : γ ∈ β} by inductive hypothesis
Hence j(β) = β.

We now show that j(µ) > µ. Let g : µ → ON such that g(α) = α for each
α. We will show that β ∈ h([g]) for each β ∈ µ and that h([g]) ∈ j(µ).

Let β ∈ µ.
{α ∈ µ : fβ (α) ∈ g(α)} = {α ∈ µ : β ∈ α}
= µ \ (β + 1)
∈U
Hence [fβ ] ∈U [g] and so h([fβ ]) ∈ h([g]). But since β ∈ µ,
β = j(β)
= h(i(β))
= h([f (β)])
Hence β ∈ h([g]).

Now, {α ∈ µ : g(α) ∈ fµ (α)} = {α ∈ µ : α ∈ µ} = µ ∈ U. Hence

[g] ∈U [fµ ] and so h[g] ∈ h([fµ ]) = h(i(µ)) = j(µ).
 119

Theorem 53. (D. Scott)

If V = L then there are no measurable cardinals.

Proof. Assume that V = L and that µ is the least measurable cardinal; we

derive a contradiction. Let U be a µ−complete ultrafilter over µ and consider
j = jU and M = MU as above.

Since V = L we have ΦL and by elementarity of j we have ΦM

L . Note that
ΦL is a sentence; i.e., it has no free variables.

Since M is transitive, ON ⊆ M by the previous lemma. So, by an earlier

lemma M = L. So we have

L = V |= (µ is the least measurable cardinal)

and
L = M |= (j(µ) is the least measurable cardinal).
Thus L |= j(µ) = µ; i.e., j(µ) = µ, contradicting the previous theorem.

Remark. We have demonstrated the existence of an elementary embedding

j : V → M . K. Kunen has shown that there is no elementary j : V → V.

Large cardinal axioms are often formulated as embedding axioms. For

example, κ is said to be supercompact whenever

∀λ ∃j [j : V → M and j(κ) > λ and j|R(λ) = id|R(λ) and λ M ⊆ M ].

120 CHAPTER 12. CONSTRUCTIBILITY
Chapter 13

Appendices

.1 The Axioms of ZFC

Zermelo-Frankel (with Choice) Set Theory, abbreviated to ZFC, is consti-

tuted by the following axioms.

1. Axiom of Equality

∀x ∀y [x = y → ∀z (x ∈ z ↔ y ∈ z)]

2. Axiom of Extensionality

∀x ∀y [x = y ↔ ∀u (u ∈ x ↔ u ∈ y)]

3. Axiom of Existence
∃z z = ∅

4. Axiom of Pairing
∀x ∀y ∃z z = {x, y}

5. Union Axiom

∀x [x 6= ∅ → ∃z z = {w : (∃y ∈ x)(w ∈ y)]

121
122 CHAPTER 13. APPENDICES

6. Intersection Axiom

∀x [x 6= ∅ → ∃z z = {w : (∀y ∈ x)(w ∈ y)]

7. Axiom of Foundation

∀x [x 6= ∅ → (∃y ∈ x)(x ∩ y = ∅)]

8. Replacement Axiom Scheme

For each formula Φ(x, u, v, w1 , . . . , wn ) of the language of set theory,

∀w1 . . . ∀wn ∀x [∀u ∈ x ∃!v Φ → ∃z z = {v : ∃u ∈ x Φ}]

9. Axiom of Choice

∀X [(∀x ∈ X ∀y ∈ X (x = y ↔ x∩y 6= ∅)) → ∃z (∀x ∈ X ∃!y y ∈ x∩z)]

10. Power Set Axiom

∀x ∃z z = {y : y ⊆ x}

11. Axiom of Infinity

N 6= ON

.2 Tentative Axioms

Here is a summary of potential axioms which we have discussed but which

lie outside of ZFC.

1. Axiom of Inaccessibles

∃κ κ > ω and κ is an inaccessible cardinal

2. Continuum Hypothesis

|P(ω)| = ω1
.2. TENTATIVE AXIOMS 123

3. Generalised Continuum Hypothesis

∀κ [κ is a cardinal → |P(κ)| = κ+

4. Suslin Hypothesis
Suppose that R is a complete dense linear order without endpoints in
which every collection of disjoint intervals is countable.
Then R ∼= R.
5. Axiom of Constructibility

V=L
Index

n−tuple, 93 delta system lemma, 100

dense, 70
absolute, 88 derivative, 75
between, 88
aleph function, 68 elementary chain theorem, 104
elementary embedding, 116
beth function, 68 elementary submodel, 98
Cantor normal form, 48 epsilon number, 47
Cantor’s theorem, 64 equality axiom of, 23
cardinal, 63 equality principle, 83
inaccessible, 68 existence, axiom of, 23
measurable, 115 extensionality, axiom of, 23
regular, 67 filter, 71
singular, 67 formula
supercompact, 119 40 , 91
cardinality, 63 41 , 92
hereditary, 80 atomic, 11
chain rule, 76 bounded, 91
choice, axiom of, 28 of set theory, 11
class, 16 foundation, axiom of, 24
cofinality, 66 function, 21
completeness theorem, 12 cofinal, 66
comprehension theorem scheme, 27 order preserving, 55
constructibility
axiom of, 112 Gödel operations, 94
constructible, 105 Goodstein’s theorem, 50
continuum hypothesis, 68
generalised, 68 hyperreal, 73
cumulative hierarchy, 79 finite, 75
infinite
decimal, 70 positive, 75

124
INDEX 125

induction Ramsey’s theorem, 72

∈, 57, 78 rank, 79
finite, 34 real, 70
transfinite, 44 recursion
inductive hypothesis, 34 ∈, 78
infinitely close, 75 finite, 34
infinitesimal, 75 transfinite, 44
positive, 75 regularity, axiom of, 25
infinity, axiom of, 42 relation, 53
integer, 69 extensional, 53
intersection axiom, 24 irreflexive, 53
isomorphic, 55 total, 53
isomorphism, 55 transitive, 53
well founded, 53
König’s theorem, 67 relativisation, 86
replacement axiom scheme, 26
Leibniz transfer principle, 74 Richard’s paradox, 8
Levy reflection rrinciple, 89 Russell’s paradox, 21
Lowenheim-Skolem-Tarski theorem, 99
segment
minimal element, 53 initial, 55
model, 86 proper initial, 55
natural embedding, 73 sentence, 119
natural extension, 74 sequence, 47
finite, 47
order type, 59 infinite, 47
ordering set
linear, 54 cofinal, 66
partial, 54 hereditarily countable, 81
ordinal, 41 hereditarily finite, 81
limit, 43 partially ordered, 54
successor, 43 well ordered, 54
shuffle lemma scheme, 93
pairing, axiom of, 23 Sierpinski’s theorem, 73
power set, 29 standard part, 75
axiom, 29 subformula, 12
predecessor, 55 Suslin hypothesis, 71
pressing down lemma, 100 symbol, 11
126 INDEX

Tarski-Vaught condition, 88
term, 16
substitution of, 17
transitive, 32
transitive closure, 77
trichotomy
of N, 32
of ON, 42

ultrafilter, 72
κ−complete, 115
free, 72
non-principal, 72
ultrapower, 73
union axiom, 24

variable
bound, 13
free, 14
occurrence
bound, 13
substitution of, 13, 14

well ordering principle, 59

Zermelo’s theorem, 59
ZFC, 86