Fred M.

Katz
Sets and Their Sizes
New York, New York
Fred M. Katz (fred@katz.com)
Logic & Light, Inc.
New York, NY 10028
USA
Sets and Their Sizes
Mathematics Subject Classification (2001): 03E10, 54A25
c _1981, 2001 Fred M. Katz.
All rights reserved. This work may not be translated or copied in whole or in
part without the written permission of the publisher (Logic & Light, Inc. 400
East 84th Street, New York, NY 10028, USA), except for brief excerpts in con-
nection with reviews or scholarly analysis.
The author hereby grants to M.I.T. permission to reproduce and distribute
copies of this thesis document in whole or in part.
Preface
Cantor’s theory of cardinality violates common sense. It says, for example, that
all infinite sets of integers are the same size. This thesis criticizes the arguments
for Cantor’s theory and presents an alternative.
The alternative is based on a general theory, CS (for Class Size). CS
consists of all sentences in the first order language with a subset predicate and a
less-than predicate which are true in all interpretations of that language whose
domain is a finite power set. Thus, CS says that less than is a linear ordering
with highest and lowest members and that every set is larger than any of its
proper subsets. Because the language of CS is so restricted, CS will have infinite
interpretations. In particular, the notion of one-one correspondence cannot
be expressed in this language, so Cantor’s definition of similarity will not be in
CS, even though it is true for all finite sets.
We show that CS is decidable but not finitely axiomatizable by characterizing
the complete extensions of CS. CS has finite completions, which are true only
in finite models and infinite completions, which are true only in infinite models.
An infinite completion is determined by a set of remainder principles, which
say, for each natural number, n, how many atoms remain when the universe is
partitioned into n disjoint subsets of the same size.
We show that any infinite completion of CS has a model over the power set
of the natural numbers which satisfies an additional axiom:
OUTPACING. If initial segments of A eventually become smaller
than the corresponding initial segments of B, then A is smaller than
B.
Models which satisfy OUTPACING seem to accord with common intuitions
about set size. In particular, they agree with the ordering suggested by the
notion of asymptotic density.
Acknowledgements
This dissertation has been made possible by Cantor — who invented set theory,
Tarski — who discovered model theory, Abraham Robinson — who introduced
model-completeness, and George Boolos — who made all of these subjects ac-
cessible to me. Prof. Boolos read several handwritten versions and as many
typed versions of this thesis, with care and patience each time around. I owe
i
ii
him thanks for finding (and sometimes correcting) my errors, for directing me
to some useful literature, and simply for being interested. The reader should
join me in thanking him for his help with style and organization, and for his
insistence on filling in mere details. Nevertheless, whatever errors, gaps, or
infelicities remain are solely the responsibility of the author.
Note
Except for some minor wording changes, typographical corrections, and omis-
sions of some elementary (but very long) proofs, this is the text of my disserta-
tion.
This dissertation was submitted in partial fulfillment of the requirements for
the degree of Doctor of Philosophy at the Massachusetts Institute of Technology
in April, 1981
It was certified and accepted by Prof. George Boolos, Thesis Supervisor and
Chairperson of the Department of Linguistics and Philosophy.
Fred M. Katz
June, 2001
Contents
Preface i
1 Introduction 1
1.1 The Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Cantor’s argument . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Cantor and the logicists . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Aims and outline . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2 The General Theory 9
2.1 The Theory CORE . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Addition of Set Sizes . . . . . . . . . . . . . . . . . . . . . . . . . 16
3 A Formal Theory of Class Size 19
3.1 CS - The Theory of Class Size . . . . . . . . . . . . . . . . . . . . 20
3.2 CA - Axioms for CS . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Remarks on showing that CA axiomatizes CS . . . . . . . . . . . 38
4 The Pure Theory of Class Sizes 45
4.1 Size models and PCS . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.2 Some models and theories . . . . . . . . . . . . . . . . . . . . . . 47
4.3 PCA is model complete . . . . . . . . . . . . . . . . . . . . . . . 53
4.4 Remarks on showing that PCA axiomatizes PCS . . . . . . . . . 61
5 Completeness of CA 63
5.1 Model completeness of CA . . . . . . . . . . . . . . . . . . . . . . 63
5.2 Prime models for CA . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6 Sets of Natural Numbers 83
6.1 The outpacing principle . . . . . . . . . . . . . . . . . . . . . . . 83
6.2 Models of CS and Outpacing . . . . . . . . . . . . . . . . . . . . 86
6.3 Size and density . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
iii
iv CONTENTS
Appendix 101
A Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
B Model Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
C Model Completeness . . . . . . . . . . . . . . . . . . . . . . . . . 104
Bibliography 107
Index of Symbols 108
Index of Theories 110
Subject Index 112
1
Introduction
1.1 The Problem
This paper proposes a theory of set size which is based on intuitions, naive and
otherwise. The theory goes beyond intuitions, as theories will, so it needs both
justification and defense. I spend very little time justifying the theory; it is so
clearly true that anyone who comes to the matter without prejudice will accept
it. I spend a lot of time defending the theory because no one who comes to the
matter comes without prejudice.
The prejudice stems from Cantor’s theory of set size, which is as old as sets
themselves and so widely held as to be worthy of the name the standard theory.
Cantor’s theory consists of just two principles:
ONE-ONE. Two sets are the same size just in case there is a one-
one correspondence between them.
CANTOR
<
. A set, x, is smaller than a set, y, just in case x is the
same size as some subset of y, but not the same size as y itself.
A one-one correspondence between two sets is a relation which pairs each
member of either set with exactly one member of the other. For example, the
upper-case letters of the alphabet can be paired with the lower-case letters:
ABCDEFGHIJKLMNOPQRSTUVWXYZ
abcdefghijklmnopqrstuvwxyz
So, the standard theory says, the set of upper-case letters is the same size
as the set of lower-case letters. Fine and good.
The standard theory also says that the set of even numbers is the same size
as the set of integers since these two sets can also be paired off one-to-one:
. . . −n . . . −3 −2 −1 0 1 2 3 . . . n . . .
. . . −2n . . . −6 −4 −2 0 2 4 6 . . . 2n . . .
1
2 1. INTRODUCTION
Similarly, the standard theory says that the set of positive even integers is
the same size as the set of prime numbers: pair the n-th prime with the n-th
positive even number. In both of these cases, common sense chokes on the
standard theory.
In the first case, common sense holds that the set of integers is larger than
the set of even integers. The integers contain all of the even integers and then
some. So it’s just good common sense to believe there are more of the former
than the latter. This is just to say that common sense seems to follow:
SUBSET. If one set properly includes another, then the first is
larger than the second.
even into the infinite, where it comes up against the standard theory.
Common sense can make decisions without help from SUBSET. Though
the set of primes is not contained in the set of even integers, it is still clear to
common sense that the former is smaller than the latter. One out of every two
integers is even, while the prime numbers are few and far between. No doubt, to
use this reasoning, you need a little number theory in addition to common sense;
but, given the number theory, it’s the only conclusion common sense allows.
The theory proposed here accommodates these bits of common sense rea-
soning. It maintains SUBSET and a few and far between principle and much
else besides. To state this theory, we use three two-place predicates: <, ∼, and
>. If A and B name sets, then
• A < B is read as A is smaller than B,
• A ∼ B is read as A is the same size as B, and
• A > B is read as A is larger than B.
Incidentally, we assume throughout this thesis that the following schemata
are equivalent, item by item, to the readings of the three predicates given above,
assuming that A is the set of α’s and B is the set of β’s:
a. • There are fewer α’s than β’s.
• There are just as many α’s than β’s.
• There are more α’s than β’s.
b. • The number of α’s is less than the number of β’s.
• The number of α’s is the same as the number of β’s.
• The number of α’s is greater than the number of β’s.
c. • The size of A is smaller than the size of B.
• A and B are (or, have) the same size.
• The size of A is larger than the size of B.
1.2. CANTOR’S ARGUMENT 3
Regarding this last group, we emphasize that we are not arguing that there
really are such things as set sizes, nor that there are not really such things.
Statements about sizes can be translated in familiar and long-winded ways into
statements about sets, though we will not bother to do so.
I have identified the standard theory, Cantor’s, with two principles about set
size. The term size, however, is rarely used in connection with Cantor’s theory;
so it might be wondered whether the standard theory is really so standard. In
stating ONE-ONE and CANTOR
<
, Cantor used the terms power and cardinal
number rather than size. In the literature, the term cardinal number (some-
times just number) is used most frequently. If someone introduces cardinal
number as a defined predicate or as part of a contextual definition (e.g. “We
say that two sets have the same cardinal number just in case . . . ”), there is
no point in discussing whether that person is right about size.
Though Cantor’s theory is usually taken as a theory of set size, it can also
be taken as just a theory of one-one correspondences. More specifically, saying
that two sets are similar iff they are in one-one correspondence can either be
taken as a claim about size or be regarded as a mere definition. Whether or
not similarity is coextensive with being the same size, the definition is worth
making. The relation picked out is well studied and well worth the study. The
technical brilliance of the theory attests to this: it has given us the transfinite
hierarchy, the continuum problem, and much else. In addition, the theory has
consequences which do not prima facie seem to have anything to do with size
or similarity: the existence of transcendental numbers comes to mind. All of
this is to say that the interest in one-one correspondence has not been sustained
solely by its identification with the notion of size. Hence, denying that they are
the same does not endanger the theory of one-one correspondences, per se.
But most mathematicians and philosophers do not use cardinal number as
a mere abbreviation. They use the term in just the way that we use size and
slide freely among (A), (B), and (C). This is true, in particular, of Cantor, who
offered ONE-ONE as a theory; indeed, he offered an argument for this theory.
1.2 Cantor’s argument
Cantor bases his argument for ONE-ONE on the idea that the size of a set, it’s
cardinal number, depends on neither the particular elements it contains nor on
how those elements are arranged:
The cardinal number, [M[, of a set M (is the general concept which)
arises from M when we make abstraction of the nature of its elements
and of the order in which they are given. [Cantor, p.86]
But to say that the cardinal number of a set does not depend on certain things
is not to say what the cardinal is. Neither does it insure that two sets have
the same cardinal number just in case they are in one-one correspondence. To
flesh out this notion of double abstraction, Cantor reduces it to a second ab-
straction operator, one which works on the elements of sets rather than the sets
4 1. INTRODUCTION
themselves:
Every element, m, if we abstract from its nature becomes a unit.
[Cantor, p.86]
and so, concludes Cantor:
The cardinal number, [M[, is a set composed of units (which has
existence in our minds as an intellectual image or projection of M.)
[Cantor, p.86]
According to Jourdain, Cantor
distinguised very sharply between an aggregate and a cardinal num-
ber that belongs to it: “Is not an aggregate an object outside us,
whereas its cardinal number is an abstract picture of it in our
mind.” [Cantor, p.80]
I have parenthesized the expressions above where Cantor describes cardinal
numbers as mental entities. Nevertheless, I can only make sense of his arguments
insofar as he treats cardinal numbers as sets: he refers to them as ‘definite
aggregates’, supposes that they have elements, and employs mappings between
cardinal numbers and other sets.
The following three statements seem to express Cantor’s intent:
[M[ = ¦ y [ ∃x ∈ M ∧ y = Abstract(x) ¦ (1.1)
∀x∃y(y = Abstract(x) ∧ Unit(y)) (1.2)
∀M∀y(y ∈ [M[ → Unit(y)) (1.3)
Abstract (x) is to be read as the result of abstracting from the element
x, Unit (x) as x is a unit, and [M[ as the cardinal number of M.
So (1.1) gives a definition of cardinal number, in terms of the operation of
abstraction, from which Cantor proves both ONE-ONEa and ONE-ONEb.
M ∼ N → [M[ = [N[ (ONE-ONEa)
[M[ = [N[ → M ∼ N (ONE-ONEb)
ONE-ONEa is true, says Cantor, because
the cardinal number [M[ remains unaltered if in the place of one
or many or even all elements m of M other things are substituted.
[Cantor, p.80]
and so, if f is a one-one mapping from M onto N, then in replacing each element,
m, of M with f(m)
M transforms into N without change of cardinal number. (p.88)
1.2. CANTOR’S ARGUMENT 5
In its weakest form, the principle Cantor cites says that if a single element of
M is replaced by an arbitrary element not in M then the cardinal number of
the set will remain the same. That is,
(1.4) (a ∈ M ∧ b / ∈ M)
∧ N = ¦ x [ (x ∈ M ∧ x / ∈ a) ∨ x ∈ b ¦
→ [M[ = [N[
The reasoning is clear: so far as the cardinal number of a set is concerned, one
element is much the same as another. It is not the elements of a set, but only
their abstractions, that enter into the cardinal number of a set. But abstractions
of elements are just units; so one is much the same as another.
ONE-ONEb is true, Cantor says, because
. . . [M[ grows, so to speak, out of M in such a way that from every
element m of M a special unit of [M[ arises. Thus we can say that
M ∼ [M[.
So, since a set is similar to its cardinal number, and similarity is an equivalence
relation, two sets with same cardinal number are similar. Unless each element
of a set abstracts to a ‘special’, i.e. distinct, unit, the correspondence from M
to its cardinal number will be many-one and not one-one. A weak version of
this principle is:
M = ¦a, b¦ ∧ a ,= b
→ [M[ = ¦Abstract(a), Abstract(b)¦
∧ Abstract(a) ,= Abstract(b)
(1.5)
These two arguments do one another in. (1.4) says that replacing an element
of a set with any element not in the set does not affect the cardinality. But, by
the definition of [M[, (1.1), this means that
(1.6) ∀x∀y(Abstract(x) = Abstract(y))
For, consider an arbitrary pair of elements, a and b. Let M = ¦a¦ and let
N = ¦b¦. So, the conditions of (1.4) are met and [M[ = [N[. But [M[ =
¦Abstract(a)¦ and [N[ = ¦Abstract(b)¦, by (1.1). So Abstract(a) = Abstract(b).
Generalizing this argument yields (1.6).
So Cantor’s argument for ONE-ONEa only works by assigning all non-
empty sets the same, one-membered, cardinal number. But, this contradicts
ONE-ONEb.
Conversely, the argument that a set is similar to its cardinal number relies
on (1.5), which entails
(1.7) ∀x∀y(x ,= y → Abstract(x) ,= Abstract(y))
6 1. INTRODUCTION
assuming only that any two objects can constitute a set. But if the abstractions
of any two elements are distinct, then no two sets have the same cardinal number
as defined by (1.1), contra ONE-ONEa.
There is no way to repair Cantor’s argument. Rather than leading to a
justification of ONE-ONE, Cantor’s definition of cardinal number is sufficient
to refute the principle. The negation of (1.6) is:
(1.8) ∃x∃y(Abstract(x) ,= Abstract(y))
So one of (1.6) and (1.8) must be true. We have shown that (1.6) contradicts
ONE-ONEb. Similarly, (1.8) contradicts ONE-ONEa; if a and b have distinct
abstractions, then ¦a¦ and ¦b¦ have distinct cardinal numbers, ¦Abstract(a)¦
and ¦Abstract(b)¦, despite the fact that they are in one-one correspondence. So
ONE-ONE is false whether (1.6) or its negation, (1.8), is true.
1.3 Cantor and the logicists
Though both Frege and Russell accepted Cantor’s theory of cardinality, neither
accepted Cantor’s argument. Frege spends an entire chapter of the Grundlagen
mocking mathematicians from Euclid to Schroder for defining numbers as sets
of units. He neatly summarizes the difficulty with such views:
If we try to produce the number by putting together different dis-
tinct objects, the result is an agglomeration in which the objects
remain still in possession of precisely those properties which serve to
distinguish them from one another, and that is not the number. But
if we try to do it in the other way, by putting together identicals,
the result runs perpetually together into one and we never reach a
plurality . . .
The word ‘unit’ is admirably adapted to conceal the difficulty . . . We
start by calling the things to be numbered ‘units’ without detracting
from their diversity; then subsequently the concept of putting to-
gether (or collecting, or uniting, or annexing, or whatever we choose
to call it) transforms itself into arithmetical addition, while the con-
cept word ‘unit’ changes unperceived into the proper name ‘one’.
[Frege, pp 50-51]
These misgivings about units do not prevent Frege from basing his definition
of ‘number’ and his entire reduction of arithmetic on Cantor’s notion of one-
one correspondence. “This opinion”, says Frege, “that numerical equality or
identiy must be defined in terms of one-one correlation, seems in recent years to
have gained widespread acceptance among mathematicians” [Frege, pp. 73-74].
Frege cites Schroder, Kossak, and Cantor.
Russell displays similar caution about Cantor’s argument [Russell, p.305] and
similar enthusiasm for his theory (see the quote at the beginning of Chapter 2,
for example.)
1.4. AIMS AND OUTLINE 7
Of course, Frege and Russell cleaned up Cantor’s presentation of the theory.
Russell, for example, notes that Cantor’s statement (1) is not a ‘true definition’
and
merely presupposes that every collection has some such property as
that indicated — a property, that is to say, independent of its terms
and their order; depending, we might feel tempted to add, only upon
their number. [Russell, pp. 304-305]
So Russell, and similarly Frege, relied upon the principle of abstraction to obtain
a ‘formal definition’ of cardinal numbers, in contrast to Cantor, who had “taken”
number “to be a primitive idea” and had to rely on “the primitive proposition
that every collection has a number.” [Russell, p.305]
So, while some people regard Cantor’s ONE-ONE as just a definition and
others embrace it as a theory, the logicists have it both ways: adding ONE-ONE
as a formal definition to set theory (or, as they would call it, logic) they have no
obligation to defend it and can steer clear of peculiar arguments about units; at
the same time, they can advance it as a great lesson for simple common sense.
The logicists’ adoption of Cantor’s theory of cardinality needs no great ex-
planation: it came with set theory and, to a large extent, motivated set theory
and determined its research problems. But there are two specific reasons that
they should have seized upon ONE-ONE and CANTOR
<
. First, they both
have the form of definitions, no matter how they are intended. So the notion of
cardinality is born reduced.
Second, Cantor’s theory clears the way for other reductions. Suppose, for
example, you wish to reduce ordered pairs to sets. Well, you have to identify
each ordered pair with a set and define the relevant properties and relations
among ordered pairs in terms of properties and relations among sets. One
of the relations that has to be maintained is identity; so each ordered pair
must be identified with a distinct set. In addition, the relative sizes of sets of
ordered pairs should be preserved under translation. But, if ONE-ONE is the
correct theory of size, then this second condition follows from the first, since the
existence of one-one correspondences will be preserved under a one-one mapping.
1.4 Aims and outline
It would be naive to suppose that people’s faith in Cantor’s theory would be
shaken either by refuting specific arguments for ONE-ONE or by associating the
acceptance of that theory with a discredited philosophy of mathematics. Such
points may be interesting, but in the absence of an alternative theory of size
they are less than convincing.
This dissertation presents such an alternative. Chapter 2 canvasses common
sense intuitions for some basic principles about set size. Chapter 3 reorganizes
those principles into a tidy set of axioms, offers an account of where the intu-
itions come from (viz. known facts about finite sets), and mines this source for
additional principles. Chapters 4 and 5 prove that the theory so obtained is
8 1. INTRODUCTION
complete, in the sense that it embraces all facts about finite sets of a certain
kind (i.e. expressible in a particular language). Finally, Chapter 6 elaborates
additional principles that concern only sets of natural numbers and demon-
strates that these additional principles, together with the theory in Chapter 3,
are satisfiable in the domain of sets of natural numbers.
2
The General Theory
The possibility that whole and part may have the same number of
terms is, it must be confessed, shocking to common sense . . . Common
sense, therefore, is here in a very sorry plight; it must choose between
the paradox of Zeno and the paradox of Cantor. I do not propose
to help it, since I consider that, in the face of the proofs, it ought to
commit suicide in despair. [Russell, p. 358]
Is common sense confused about set size, as Russell says, or is there a way
of elaborating on common sense to get a plausible and reasonably adequate
theory of cardinality? To be plausible, a theory should at least avoid principles
and consequences which violate common sense. To be reasonably adequate, a
theory has to go beyond bare intuitions: it should not rest with trivialities and
it should answer as many questions about set size as possible, though it need
not be complete. Plausibility and adequacy are conflicting demands: the first
says that there should not be too many principles (no false ones, consistency),
the second that there should not be too few principles.
In the Introduction, I argued that a coherent theory of cardinality has to
contain some principles that refer to the kinds of objects in sets, pace Cantor. In
this chapter, however, I want to see how far we can go without such principles;
i.e. how much can you say about smaller than without using predicates (other
than identity) which relate the members of the sets being compared? I shall
begin by stating a number of principles and explaining why they are included
in the general theory.
2.1 The Theory CORE
First, there is the SUBSET principle:
SUBSET. If x is a proper subset of y, then x is smaller than y.
The reason for including SUBSET should be obvious. What has prompted the
search for an alternative to the standard theory of cardinality is the conflict
9
10 2. THE GENERAL THEORY
between ONE-ONE and SUBSET. Now, it is often said that common sense
supports both of these principles and that it is in doing so that common sense
is confused. From this it is supposed to follow that common sense cannot be
relied upon, so we should opt for ONE-ONE,with the technical attractions that
it provides.
But, there’s a difference in the way that common sense supports these two
principles. There is no doubt that you can lead an unsuspecting person to
agree to ONE-ONE by focusing their attention on forks and knives, husbands
and wives, and so forth: i.e. finite sets. With carefully chosen examples, say
the odds and the evens, you might even convince someone that ONE-ONE is
true for infinite sets, too. Now, I do not think that such guile is needed to
lead someone to agree to SUBSET, but that’s not what my argument depends
on. The argument hinges on a suggestion about how to resolve cases where
mathematical intuitions seem to conflict. The suggestion is to see what happens
with particular cases on which the principles conflict before you’ve lead someone
to agree to either of the general statements.
So, if you want to find out what common sense really thinks about SUBSET
and ONE-ONE, you would present people with pairs of infinite sets, where one
was a proper subset of the other. I’ve actually tried this, in an unscientific way,
and what I’ve gotten, by and large, is what I expected: support for SUBSET.
(By and large because many people think that all infinite sets have the same
size: Infinity.)
Naturally, I would not venture that this sort of technique, asking people, is
any way to find out which of SUBSET and ONE-ONE is true. People’s intu-
itions about mathematics are notoriously unreliable, not to mention inconstant.
Of course, harping on this fact might engender some unwarranted skepticism
about mathematics. What I am suggesting is that there might be a rational
way of studying mathematical intuitions and that we should at least explore
this possibility before proclaiming common sense to be hopelessly confused on
mathematical matters.
So, SUBSET, all by itself seems to be a plausible alternative to Cantor’s
theory. though it surely is not enough. Given just this principle, it’s possible
that one set is smaller than another just in case it is a proper subset of the
other. It’s clear that we need additional principles. All of the following seem
worthy (where x < y is to be read as x is smaller than y, x ∼ y is to be read
as x is the same size as y, x > y as x is larger than y. The Appendix gives
a full account of the notations used throughout this thesis.)
2.1. THE THEORY CORE 11
Theory 2.1.1. QUASI-LOGICAL
x < y → y < x (ASYM
<
)
x > y → y > x (ASYM
>
)
(x < y ∧ y < z) → x < z (TRANS
<
)
(x > y ∧ y > z) → x > z (TRANS
>
)
x ∼ y → Indisc(x, y) (INDISC

)
x ∼ x (REF

)
x ∼ y → y ∼ x (SYM

)
(x ∼ y ∧ y ∼ z) → x ∼ z (TRANS

)
x > y ↔ y < x (DEF
>
)
where Indisc(x, y) abbreviates ∀z((z < x ↔ z < y) ∧ (z > x ↔ z > y))
We call the principles listed above quasi-logical principles because it is
tempting to defend them as logical truths. Consider the first principle, for
example, in unregimented English:
ASYM
<
. If x is smaller than y, then y is not smaller than x.
This sentence can be regarded as an instance of the schema:
ASYM
F
. If x is F-er than y, then y is not F-er than x.
where F is to be replaced by an adjective from which comparatives can be
formed, e.g. ‘tall’, ‘short’, ‘happy’, but not ‘unique’ or ‘brick’. It appears that
every instance of this schema is true, so it could be maintained that each is true
in virtue of its form, that each is a logical truth.
The other principles might be defended in the same way, though the schema
for INDISC

would have to be restricted to triples of corresponding compara-
tives, for example: is smaller than, is larger than, is the same size as.
But using such observations to support these principles would be problematic
for two reasons. First, it would require taking positions on many questions
about logical form and grammatical form which would take us far afield and,
possibly, antagonize first-order logicians. Second, there are some instances of
the schemata that make for embarassing counterexamples: ‘further east than’
(in a round world) and ‘earlier than’ (in, I’m told, a possible world).
So, it might be that casting the principles above as instances of the appro-
priate schema would only explain why they are part of common sense. What
remains clear is that a theory of cardinality which openly denied any of these
principles would be implausible: it would be ridiculed by common sense and
mathematical sophisticates alike. I can just barely imagine presenting a theory
which, for fear of inconsistency, withheld judgment on one or more of these
principles. But to do so without good reason would be counterproductive. It
seems that if a case could be made that these statements, taken together, are
inconsistent with SUBSET, that would be good reason to say that there is no
12 2. THE GENERAL THEORY
reasonably adequate alternative to Cantor’s theory. Since my goal is to counter
such a conclusion, it seems that the proper strategy is to include such seemingly
obvious truths and to show that the resulting theory is consistent. So the strat-
egy here is not to adduce principles and argue for the truth of each. This would
be impossible, given that the principles are logically contingent. Instead, our
approach is to canonize what common sense holds to be true about cardinality
and show that the result is consistent and reasonably adequate.
As long as we restrict ourselves to SUBSET and the quasi-logical principles,
consistency is no problem. After all, what do the quasi-logical principles say?
Only that smaller than is a partial ordering, the larger than is the converse
partial ordering, that the same size as is an equivalence relation, and that
sets of the same size are indiscernible under the partial orderings. So, if we
are given any domain of sets, finite or infinite, we get a model for our theory
by assigning to < the relation of being a proper subset of, assigning to > the
relation of properly including, and assigning to ∼ the identity relation. Since
common sense knows that different sets can be the same size, there must be
some additional principles to be extracted from common sense.
We shall now consider some principles which cannot be regarded as quasi-
logical.
First, there is the principle of trichotomy.
(TRICH) x < y ∨ x ∼ y ∨ x > y
which says that any two sets are comparable in size. While a theory of set size
which excluded TRICH might escape ridicule, it would surely be regarded with
suspicion. Indeed, if the principles of common sense were incompatible with
TRICH, this would undoubtedly be used to discredit them.
Second, there is the representation principle.
(REP
<
) x < y → ∃x

(x

∼ x ∧ x

⊂ y)
which says that if a set, x, is smaller than another, y, then x is the same size
as some proper subset of y. Now, this is a principle which common sense has
no particular feelings about. Analogous statements about physical objects are
neither intuitive nor very clearly true. For example, (1) does not stand a chance
of being regarded as true:
(1) If one table is smaller than another, then the first is the same
size as some proper part of the second.
if ‘part’ is taken to mean ‘leg or top or rim or ...’. Even if common sense can
be persuaded to take particles and arbitrary fusions of such as parts of tables,
no one should condemn its residual caution about (1). If REP
<
is true, then it
seems to be an interesting and special fact about sets.
REP
<
was originally included in this theory for technical reasons; it makes
it easier to reduce the set of axioms already presented and it provides a basis for
several principles not yet presented. REP
<
may be open to doubt, but it is not
2.1. THE THEORY CORE 13
a principle that Cantorians could complain about, for it is entailed by Cantor’s
definition of <:
(CANTOR
<
) x < y ↔ (x ∼ y) ∧ ∃x

(x

∼ x ∧ x

⊂ y)
If CANTOR
<
is regarded as a principle instead of a definition, then it is entailed
by the principles we have already mentioned:
If x < y, then (x ∼ y), by INDISC

and ASYM
<
. By REP
<
,
some proper subset of y, say x

, must be the same size as x. But
x

< y, by SUBSET: so x < y

, by INDISC

.
Conversely, if x

∼ x and x

⊂ y, then x

< y, by SUBSET. So
x < y, by INDISC

.
There are more principles to come, but before proceeding, I’d like to take stock
of what we already have. First, I want to reduce the principles mentioned above
to a tidy set of axioms. Second, I want to estimate how far we’ve gone.
The entire set of principles already adopted are equivalent to the following,
which will be referred to as the core theory.
Theory 2.1.2. CORE, the core theory,
x ⊂ y → x < y (SUBSET)
x ∼ y ↔ Indisc(x, y) (DEF

)
(x ∼ y ∧ y ∼ z) → x ∼ z (TRANS

)
x > y ↔ y < x (DEF
>
)
(x < x) (IRREF
<
)
x < y ∨ x ∼ y ∨ x > y (TRICH)
The only axiom in CORE that has not already been introduced is DEF

,
which is logically equivalent to the conjunction of INDISC

and its converse

INDISC:
x ∼ y → Indisc(x, y) (INDISC

)
Indisc(x, y) → x ∼ y (

INDISC)

INDISC says that if two sets fail to be the same size, then their being different
in size is attributable to the existence of some set which is either smaller than
one but not smaller than the other, or larger than one but not larger than the
other.
Theorem 2.1.3. Let T = QUASI; SUBSET; REP
<
. Then T ≡ CORE.
Proof.
¬ T ¬ CORE. We only need to show that

INDISC is entailed by T.
Suppose (x ∼ y). So x < y or y < x, by TRICH. But (x < x) and
(y < y), by IRREF
<
.
14 2. THE GENERAL THEORY
¬ CORE ¬ T
(i) TRANS
<
. If y < z, there is a y

such that y

∼ y and y

⊂ z by
REP
<
. If x < y, then x < y

because y

∼ y, by DEF

. So, there
is an x

such that x

∼ x and x

⊂ y

. So x

⊂ z and, by SUBSET,
x

< z. But then x < z by DEF

.
(ii) ASYM
<
. If x < y and y < x, then x < x, by TRANS
<
, contra
IRREF
<
.
(iii) TRANS
>
, ASYM
>
, and IRREF
>
follow from the corresponding
principles for < and DEF
>
.
(iv) INDISC

, SYM

, TRANS

, and REF

are logical consequences of
DEF

.
CORE is consistent. In fact, two kinds of models satisfy CORE.
Definition 2.1.4.
a. / is a finite class model with basis x iff
(i)
˙
/ = T(x), where x is finite.
(ii) / a ⊂ b iff a ⊂ b
(iii) / a < b iff [a[ < [b[
b. / is a finite set model iff
(i)
˙
/ = ¦ x [ x is a finite subset of Y ¦, for some infinite Y .
(ii) / a ⊂ b iff a ⊂ b
(iii) / a < b iff [a[ < [b[
Models can be specified by stipulation the smaller than relation since larger
than and same size as are defined in terms of smaller than.
Fact 2.1.5.
a. If / is a finite class model, then / CORE.
b. If / is a finite set model, then / CORE.
Proof. In both cases, the finite cardinalities determine a quasi-linear ordering
of the sets in which any set is higher than any of its proper subsets.
The normal ordering of finite sets is, in fact, the only one that satisfies
CORE. By adding TRICH we have ruled out all non-standard interpretations
of <.
Theorem 2.1.6. Suppose that / is a model such that
2.1. THE THEORY CORE 15
a. If x ∈
˙
/, then x is finite,
b. If x ∈
˙
/ and y ⊂ x, then y ∈
˙
/, and
c. / CORE.
Then / a < b iff [a[ < [b[
Proof. We shall prove (*) by induction on n:
If [a[ = n, then / (a ∼ b) iff [b[ = n (*)
Suppose n = 0
then a = ∅
But if [b[ = 0
then / (a ∼ b) by REF

And if [b[ , = 0
then a ⊂ b
So / (a < b) by SUBSET
So / (a ∼ b) by INDISC

Now, suppose that (*) is true for all i ≤ n:
If [a[ = [b[ = n + 1
then / (a ∼ b), as follows:
Suppose / (a < b)
So / (a

∼ a) for some a

⊂ b by REP
<
But if / a

⊂ b
then [a

[ ≤ n
So [a[ ≤ n by (*), contra our hypothesis.
But if [a[ = n + 1
and / (a ∼ b)
then [b[ ≥ n + 1 by induction.
But if [b[ > n + 1
pick b

⊂ b, b

∈ A,
with [b

[ = n + 1 by condition (b)
So / (b

< b) by SUBSET
and / (b

∼ a)
So / (a < b) contra our supposition.
So [b[ = n + 1
16 2. THE GENERAL THEORY
2.2 Addition of Set Sizes
We shall now extend CORE to an account of addition of set sizes. Since the
domains of our intended models contain only sets and not sizes of sets, we have
to formulate our principles in terms of a three-place predicate true of triples of
sets: Sum (x, y, z) is to be read as the size of z is the sum of the sizes of x
and y .
The following principles are sufficient for a theory of addition:
Theory 2.2.1. Addition
Functionality of addition
Sum(x, y, z) → (x ∼ x

↔ Sum(x

, y, z) (a)
Sum(x, y, z) → (y ∼ y

↔ Sum(x, y

, z) (b)
Sum(x, y, z) → (z ∼ z

↔ Sum(x, y, z

) (c)
Addition for disjoint sets
x ∩ y = ∅ → Sum(x, y, x ∪ y) (DISJ
+
)
Monotonicity of Addition
Sum(x, y, z) → x < z ∨ x ∼ z (MONOT)
FUNC
+
says that sets bear Sum relations to one another by virtue of their
sizes alone. This condition must clearly be met if Sum is to be read as specified
above.
DISJ
+
tries to say what function on sizes the Sum relation captures by fixing
the function on paradigm cases: disjoint sets. But FUNC
+
and DISJ
+
leave
open the possibility that addition is cyclic: suppose we begin with a finite class
model whose basis has n elements and assign to Sum those triples ¸x, y, z) where
[z[ = ([x[ +[y[) mod (n + 1)
Both FUNC
+
and DISJ
+
will be satisfied, though the interpretation of Sum does
not agree with the intended reading. MONOT rules out such interpretations.
Given an interpretation of ’<’ over a power set there is at most one way
of interpreting Sum which satisfies ADDITION. We shall show this by proving
that ADDITION and CORE entail DEF
+
:
(DEF
+
) Sum(x, y, z) ↔
∃x

∃y

(x ∼ x

∧ y ∼ y

∧ x

∩ y

= ∅ ∧ x

∪ y

= z)
DEF
+
says that the extension of Sum is determined by the extension of <.
2.2. ADDITION OF SET SIZES 17
A model of CORE must satisfy an additional principle, DISJ

, if Sum is to
be interpreted in a way compatible with ADDITION,
(DISJ

) (x ∼ x

∧ y ∼ y

∧ x ∩ y = ∅ ∧ x

∩ y

= ∅)
→ (x ∪ y) ∼ (x

∪ y

)
(Note: The proofs in this chapter will use boolean principles freely, despite the
fact that we have not yet introduced them.)
Theorem 2.2.2. CORE + ADDITION ¬ DISJ

Proof.
Suppose (x ∼ x

∧ y ∼ y

∧ x ∩ y = ∅ ∧ x

∩ y

= ∅)
then Sum(x, y, x ∪ y)
and Sum(x

, y

, x

∪ y

) by DISJ
+
So Sum(x

, y, x ∪ y) by FUNC
+
(a)
So Sum(x

, y

, x ∪ y) by FUNC
+
(b)
So x ∪ y ∼ x

∪ y

by FUNC
+
(c)
If the minimal conditions on addition are to be satisfied in a model of CORE,
then Sum has to be definable by DEF
+
.
Theorem 2.2.3. CORE + ADDITION ¬ DEF
+
Proof. (⇒)
Suppose Sum(x, y, z)
So x < z ∨ x ∼ z by MONOT
But if x ∼ z,
Let x

= z
and y

= ∅;
then Sum(x

, y

, z) by DISJ
+
So Sum(x, y

, z) by FUNC
+
(a)
So y

∼ y by FUNC
+
(b)
And if x < z
pick x

⊂ z
with x

∼ x by REP
<
.
Let y

= z −x

But y

∼ y, as before.
18 2. THE GENERAL THEORY
(⇐)
Suppose (x ∼ x

∧ y ∼ y

∧ x

∩ y

= ∅ ∧ x

∪ y

= z)
then Sum(x

, y

, z) by DISJ
+
So Sum(x, y

, z) by FUNC
+
(a)
So Sum(x, y, z) by FUNC
+
(b)
Theory 2.2.4. EXCORE, the extended core, is CORE ∪ ADDITION.
Fact 2.2.5. The following are consequences of EXCORE:
x ∩ y
1
= x ∩ y
2
= ∅ ∧ y
1
∼ y
2
→ x ∪ y
1
∼ x ∪ y
2
x ∩ y
1
= x ∩ y
2
= ∅ ∧ y
1
< y
2
→ x ∪ y
1
< x ∪ y
2
x
1
∩ y
1
= x
1
∩ y
2
= ∅ ∧ x
1
∼ x
2
∧ y
1
< y
2
→ x
1
∪ y
1
< x
2
∪ y
2
x ⊂ z ∧ y ⊂ z ∧ x < y → (z −y) < (z −x)
x ⊂ z ∧ y ⊂ z ∧ x ∼ y → (z −y) ∼ (z −x)
x < y → ∃y

(y

∼ y ∧ x ⊂ y

)
x ∼ y → x −(x ∩ y) ∼ y −(x ∩ y)
x < y → x −(x ∩ y) < y −(x ∩ y)
Proof. The proofs are elementary.
3
A Formal Theory of Class
Size
In the preceding chapter, we searched for principles that accord with pre-
Cantorian ideas about sizes of sets. We produced several such principles, consti-
tuting EXCORE, and found two kinds of interpretations which satisfied these
principles. In one case, the domains of the interpretations were finite power
sets. In the other case, the domains consisted of all finite subsets of a given
infinite set. Insofar as both kinds of interpretation have quantifiers ranging over
finite sets, we may say that they demonstrate that EXCORE is true when it is
construed as being about finite sets.
Our goal is to show that this general theory of set size can be maintained for
infinite sets as well as for finite sets. We shall show this by constructing a model
for the general theory whose domain is the power set of the natural numbers.
But the ability to construct such a model is interesting only to the extent that
the general theory it satisfies is reasonably adequate. Suppose, for example, that
we offered as a general theory of size the axioms of CORE other than REP
<
.
Call this theory T. So T just says that smaller than is a quasi-linear ordering
which extends the partial ordering given by the proper subset relation. Since
any partial ordering can be extended to a quasi-linear ordering, T has a model
over T(N). But unless we have a guarantee that the model constructed will
satisfy, say, DISJ

, the existence of the model does not rule out the possibility
that T is incompatible with DISJ

. For a particular principle, φ, in this case
DISJ

, we may take one of three tacks: (1) add φ to T, obtaining T

, and show
that T

has a model over T(N); (2) show that φ is inconsistent with T and argue
that T is somehow more fundamental or more intuitive than φ; (3) acquiesce
in ignorance of whether T and φ are compatible and argue that if they are
incompatible, then T should be maintained anyway.
Below, we deal with DISJ

as in (1), since DISJ

is in EXCORE. Cantor’s
principle ONE-ONE is dealt with as in case (2). It seems futile to try to rule
out the need to resort to the third approach for any cases at all, but we can
19
20 3. A FORMAL THEORY OF CLASS SIZE
reduce this need to the extent that we include in our general theory, T, as many
plausible statements as possible.
We cannot construct T by taking all statements which are true for finite
sets. Not only is ONE-ONE such a statement, but using the notion of all
statements true for finite sets presupposes that we have some idea of the
range of all statements. To avoid the problems involved in speaking of all
statements, we might instead settle for all statements in L, where L is some
judiciously chosen language. To avoid ONE-ONE, T must fall short of the full
expressive power of the language of set theory.
Consider, now, the axioms in EXCORE. Other than size relations, these
axioms involve only boolean operations and inclusion relations among sets. They
do not use the notions ordered pair, relation, or function. In short, the only set
theory implicit in these axioms is boolean algebra, or a sort of Venn diagram
set theory. This is not to say the axioms do not apply to relations, functions,
or other sets of ordered pairs, but only that they do not refer to these sorts of
objects as such.
In the next section, we define a language just strong enough to express
EXCORE. We then construct a theory by taking all statements in that lan-
guage which are true over all finite power sets. By drawing statements only from
this relatively weak language, we arrive at a theory which can be satisfied over
infinite power sets. But, since we include in the language all statements of the
language which are true over any finite power set, we know that no statement in
that language can arise as something which ought to be true over infinite power
sets but might be incompatible with our theory.
There remains the possibility that we could follow the same strategy with a
more expressive language, though it would have to remain less expressive that
the full language of set theory. In fact, such a language can be obtained by
including a notion of the product of set sizes. This in turn opens the possibility
of a succession of richer languages and a corresponding succession of stronger
theories of size. At this point, the possible existence of any such hierarchy is
sheer speculation; we mention it only to emphasize that no claim is made here
that we have the strongest possible general theory of set size.
3.1 CS - The Theory of Class Size
The theories discussed in this paper will be formulated within first order predi-
cate logic with identity. To specify the language in which a theory is expressed,
then, we need only list the individual constants, predicates, and operation sym-
bols of the language and stipulate the rank, or number of argument places, for
each predicate and each operation symbol.
Definition 3.1.1.
a. L
C
, L
C
, the language of classes, is the first order language with
individual constants ∅ and I, the one-place predicate, Atom, the two-
3.1. CS - THE THEORY OF CLASS SIZE 21
place predicate ⊂, and the two place operation symbols, −, ∩, and
∪.
b. L
<
, the language of size,, is the first order language with the one-
place predicate Unit, the two-place predicates < and ⊂, and the three
place predicate Sum.
c. L
C<
, the language of class size, is the first order language contain-
ing just the non-logical constants in L
C
and L
<
.
Following the strategy outlined above, we define the theory of class size in
terms of interpretations of L
C<
over finite power sets.
Definition 3.1.2.
a. If L
C
⊂ L, then / is a standard interpretation of L iff
(i)
˙
/ = T(x), for some x, and
(ii) / assigns the usual interpretations to all constants of L
C
:
/(I) = x,
/(∅) = ∅,
/ a ⊂ b iff a ⊂ b
b. If / is a standard interpretation and
˙
/ = T(x), then x is the basis
of /, B(/).
Definition 3.1.3. / is a standard finite interpretation of L
C<
iff
a. / is a standard interpretation of L
C<
,
b. / has a finite basis, and
c.
/ a < b iff [a[ < [b[,
/ a ∼ b iff [a[ = [b[, and
/ Unit(a) iff [a[ = 1
Definition 3.1.4. CS, the theory of class size, is the set of all sentences
of L
C<
which are true in all standard finite interpretations of L
C<
.
By drawing only on principles which can be stated in L
C<
we have at least
ruled out the most obvious danger of paradox. Since the notion of one-to-
one correspondence cannot be expressed in this language, Cantor’s principle
ONE-ONE will not be included in the theory CS, even though it is true over
any finite power set.
22 3. A FORMAL THEORY OF CLASS SIZE
Since CS has arbitarily large finite models, it has infinite models. It is not
obvious that CS has standard infinite models, in which the universe is an infinite
power set. In Chapter 6 we show that such models do exist.
The present chapter is devoted to getting a clearer picture of the theory CS.
Section 2 develops a set of axioms, CA, for CS. Section 3 outlines the proof
that CA does indeed axiomatize CS. This proof is presented in Chaper 5, after
a slight detour in Chapter 4.
3.2 CA - Axioms for CS
Here we shall develop a set of axioms, CA, for the theory CS. This will be done
in several stages.
3.2.1 BA- Axioms for atomic boolean algebra
We’ll begin with the obvious. Since all of the universes of the interpretations
mentioned in the definition of CS are power sets, they must be atomic boolean
algebras and, so, must satisfy BA:
Definition 3.2.1. BA, the theory of atomic boolean algebras, is the
theory consisting of the following axioms:
x ∪ y = y ∪ x
x ∩ y = y ∩ x
x ∪ (y ∪ z) = (x ∪ y) ∪ z)
x ∩ (y ∩ z) = (x ∩ y) ∩ z)
x ∩ (y ∪ z) = (x ∩ y) ∪ (y ∩ z)
x ∪ (y ∩ z) = (x ∪ y) ∩ (y ∪ z)
x ∩ (I −x) = ∅
x ∪ (I −x) = I
x ⊂ y ↔ (x ∪ y) = y ∧ x ,= y
Atom(x) ↔ (∀y)(y ⊂ x → y = ∅)
x ,= ∅ → (∃y)(Atom(y) ∧ (y ⊂ x ∨ y = x)
These axioms are adapted from [Monk, Def 9.3, p. 141 and Def. 9.28, p.
151].
BA is clearly not a complete axiomatization of CS, since BA does not in-
volve any size notions. But BA does entail all the sentences in CS which do
not themselves involve size notions. To show this we need to draw on some
established facts about the complete extensions of BA (in the language L
C
).
The key idea here is that complete extensions of BA can be obtained either by
stipulating the finite number of atoms in a model or by saying that there are
infinitely many atoms.
3.2. CA - AXIOMS FOR CS 23
Definition 3.2.2. For n ≥ 1,
a. ATLEAST
n
is a sentence which says that there are at least n atoms:
∃x
1
...∃x
n
(Atom(x
1
) ∧ ... ∧ Atom(x
n
) ∧ ¦ x
i
,= x
j
[ 0 < i < j ≤ n¦)
b. EXACTLY
n
is a sentence which says that there are exactly n atoms:
ATLEAST
n
∧ ATLEAST
n+1
c. INF is a set of sentences which is satisfied in all and only infinite
models of BA:
INF = ¦ ATLEAST
n
[ n > 1 ¦
d. BA
n
= BA; EXACTLY
n
e. BAI = BA∪ INF
Fact 3.2.3.
a. For n ≥ 1, BA
n
is categorical. [Monk, Cor 9.32, p.152]
b. For n ≥ 1, BA
n
is complete. (Immediate from above)
c. For n ≥ 1, an n-atom atomic boolean algebra is isomorphic to any
standard finite interpretation of L
C
with an n-element basis. [Monk,
Prop. 9.30, p.151]
d. BA is complete. [Monk, Theorem 21.24, p.360]
e. BAI and the theories BA
n
, n ≥ 1, are the only complete, consistent
extensions of BA.
Fact 3.2.4. If BAI ¬ φ, then φ is true in some finite model of BA.
Proof. BA∪ INF ¬ φ. By compactness, then, there is a k such that
BA∪ ¦ ATLEAST
n
[ 1 ≤ n ≤ k ¦ ¬ φ
So φ is true in any atomic boolean algebra with more than k atoms.
Theorem 3.2.5. If CS ¬ φ, and φ ∈ L
C
, then BA ¬ φ.
Proof. If BA φ, then φ is true is some atomic boolean algebra, /.
If / is finite, then / is isomorphic to some standard finite interpration /

of L
C
. But, then φ is true in /

, so φ is not true in /

and φ / ∈ CS.
If / is infinite, then φ is consistent with BAI. But BAI is complete, so
BAI ¬ φ. By 3.2.4, φ is true in some finite model / of BA. Hence, φ is
true in some standard finite interpretation of L
C
and, again, φ / ∈ CS.
24 3. A FORMAL THEORY OF CLASS SIZE
3.2.2 Size principles
Here, we just gather the principles presented above as EXCORE:
Theory 3.2.6. CORE SIZE consists of the following axioms:
x ⊂ y → x < y (SUBSET)
x > y ↔ y < x (DEF
>
)
x ∼ y ↔ Indisc(x, y) (DEF

)
(x < x) (IRREF
<
)
x < y ∨ x ∼ y ∨ x > y (TRICH)
Unit(x) ↔ Atom(x) (DEF
1
)
Sum(x, y, z) ↔ ∃x

∃y

( (DEF
+
)
x ∼ x

∧ y ∼ y

∧ x

∩ y

= ∅
∧ x

∪ y

= z)
(x ∼ x

∧ y ∼ y

(DISJ

)
∧x

∩ y

= ∅
∧x

∪ y

= z) → x ∪ y ∼ x

∪ y

Combining the principles of boolean algebra and the size principles, we ob-
tain our first serious attempt at a general theory of size:
Definition 3.2.7. BASIC, the basic theory, is defined as:
BA∪ SIZE
3.2.3 Division principles
BASIC is not a complete axiomatization of CS. In this section we shall exhibit
an infinite number of principles which need to be added to BASIC in order to
axiomatize CS. When we are done, we will have an effective set of sentences,
CA, the Class Size Axioms, though we will not prove that CA ≡ CS until
Chapter 5.
To show that the new principles really do need to be added, we’ll need some
non-standard models of BASIC. These models will be similar in that (1)their
universes will be subsets of T(N), (2) their atoms will be singletons in T(N),
and (3) all boolean symbols will receive their usual interpretations. The models
will, however, include different subsets of N and assign different size orderings
to these sets.
In Chapter 6, these models of BASIC will reappear as submodels of various
standard models of CS over T(N). So, in addition to the immediate purpose of
establishing independence results, these models provide a glimpse of how sets
of natural numbers are ordered by size.
3.2. CA - AXIOMS FOR CS 25
Every standard finite interpretation, /, of L
C<
satisfies exactly one of the
following:
∃x∃y(x ∼ y ∧ x ∩ y = ∅ ∧ x ∪ y = I) (EVEN)
∃x∃y∃y(x ∼ y ∧ x ∩ y = x ∩ z = y ∩ z = ∅
∧Atom(z) ∧ x ∪ y ∪ z = I)
(ODD)
/ EVEN if [B(/)[ is even and / ODD if [B(/)[ is odd. But BASIC
(EVEN∨ ODD). Consider the model T whose universe consists of all only the
finite and cofinite subsets of N, where, for a, b ∈
˙
T:
T (a < b) iff a and b are both finite and [a[ < [b[
or a and b are both cofinite and [N −a[ > [N −b[
or a is finite and b is infinite.
T is a model of BASIC. But neither EVEN nor ODD is true in T, for any
two sets that are the same size are either both finite, in which case their union
is also finite, or both cofinite, in which case they cannot be disjoint.
So EVEN∨ODD is in CS, but not entailed by BASIC. As you might suspect,
this is just the tip of the iceberg of principles missing from an axiomatization of
CS. Informally, we can extend T to a model that satisfies EVEN by including
the set of even numbers and the set of odd numbers and making them the same
size. To round out the result to a model of BASIC, we also need to include all
sets which are near the set of evens or the set of odds, i.e. those that differ
from the evens or odds by a finite set. With these additions made, the new
model will be closed under boolean operations and will satisfy BA. There is
a (unique) way of ordering these added sets by smaller than that will satisfy
BASIC: rank them according to the size and direction of their finite difference
from the odds or the evens. So, we can construct an infinite model of BASIC ;
(EVEN ∨ ODD).
But this model will still not satisfy CS, as we can see by generalizing the
argument above. (EVEN∨ODD) says that the universe is roughly divisible by
two: EVEN says that the universe is divisible by two without remainder. ODD
says that there is a remainder of a single atom. We can construct a similar
statement that says the universe is roughly divisible by three — with remainder
0, 1, or 2. As with (EVEN ∨ ODD), this statement is in CS but is satisfied by
neither our original model nor the model as amended. Again, we can extend
the model and again we can produce a statement of CS which is false in the
resulting model.
We now formalize this line of reasoning.
Definition 3.2.8. If 0 ≤ m < n, then
26 3. A FORMAL THEORY OF CLASS SIZE
a. MOD
n,m
is the sentence
∃x
1
...∃x
n
∃y
1
. . . ∃y
m
(
x
1
∼ x
2
∼ . . . ∼ x
n
∧ Atom(y
1
) ∧ . . . ∧ Atom(y
m
)

1≤i<j≤n
(x
i
∩ x
j
= ∅)

1≤i<j≤m
(y
i
∩ y
j
= ∅)
∧ (x
1
∪ . . . x
n
) ∩ (y
1
∪ . . . y
m
) = ∅
∧ (x
1
∪ . . . x
n
) ∪ (y
1
∪ . . . y
m
) = I)
MOD
n,m
says that the universe is divisible into n sets of the same
size with m atoms remaining.
b. DIV
n
is the sentence
MOD
n,0
∨ . . . ∨ MOD
n,n−1
Fact 3.2.9. If 0 ≤ m < n and / is a standard finite interpretation of L
C<
,
then
/ MOD
n,m
iff [B(/)[ ≡ m mod n (a)
/ DIV
n
(b)
CS ¬ DIV
n
(c)
Theorem 3.2.10. BASIC CS
Proof. If n > 1, BASIC DIV
n
. The model, T defined above satisfies BASIC
but not DIV
n
, for any n finite sets have a finite union and any two cofinite sets
overlap. So, BASIC CS, by 3.2.9c.
We could consider adding all DIV
n
sentences to BASIC in the hope that this
would yield a complete set of axioms for CS. We did consider this, but it does
not work. To demonstrate this, we need some independence results for sets of
DIV
n
sentences.
Definition 3.2.11. If J is a set of natural numbers, then
a. DIV
J
= ¦ DIV
n
[ n ∈ J ¦
b. BDIV
J
= BASIC∪ DIV
J
c. BDIV
j
= BDIV
{j}
Our independence results will be obtained by constructing models of BASIC
which satisfy specific sets of DIV sentences. To build such models from subsets
of N, we shall include sets which can be regarded as fractional portions of N.
3.2. CA - AXIOMS FOR CS 27
Definition 3.2.12. For n ≥ 0,
a. x is an n-congruence class iff x = [nk +m] for some m, where
0 ≤ m < n.
b. x is an n-quasi-congruence class iff x is the union of finitely many
n-congruence classes.
c. x is a congruence class iff x is an n-congruence class for some n.
d. x is a quasi-congruence class iff x is an n-quasi-congruence class
for some n.
e. QC
n
= ¦ x [ x is an n-quasi-congruence class ¦
f. QC =

n>0
QC
n
Examples.
a. The set of evens, [2n], and the set of odds, [2n + 1], are 2-congruence
classes.
b. [3k + 2] is a 3-congruence class.
c. N is a 1-congruence class.
d. N is an n-quasi-congruence class for every n > 0:
N = [nk + 0] ∪ . . . ∪ [nk +n −1]
Fact 3.2.13.
a. If x ∈ QC
n
and y ∈ QC
n
, then x ∪ y ∈ QC
n
.
b. If x ∈ QC
n
, then N −x ∈ QC
n
.
Proof.
a. Suppose x = a
1
∪. . . a
k
and y = b
1
∪. . . b
j
. Then x∪y = a
1
∪. . . a
k
∪b
1

. . . b
j
.
b. N itself is the union of n-congruence classes. If x is the union of m of these
classes, then N − x is the union of the remaining (n − m) n-congruence
classes.
Definition 3.2.14. x is near y, NEAR(x,y), iff x − y and y − x are both
finite.
Fact 3.2.15. x is near y iff there are finite sets, w
1
and w
2
such that
x = (y ∪ w
1
) −w
2
.
28 3. A FORMAL THEORY OF CLASS SIZE
Proof.
⇒ Let w
1
= x −y and let w
2
= y −x.
⇐ Suppose x = (y ∪ w
1
) − w
2
. Then x − y ⊆ w
1
and y − x ⊆ w
2
, so x − y
and y −x are finite.
Fact 3.2.16. If x
1
⊆ x ⊆ x
2
, x
1
is near y, and x
2
is near y, then x is near
y.
Proof. Since x ⊆ x
2
, (x − y) ⊆ (x
2
− y). But (x
2
− y) is finite, so (x − y) is
finite.
Since x
1
⊆ x, (y−x) ⊆ (y−x
1
). But (y−x
1
) is finite, so (y−x) is finite.
Fact 3.2.17. NEAR is an equivalence relation.
Proof.
a. x is near x, since x −x = ∅, which is finite.
b. If x is near y, then y is near x. Immediate.
c. Suppose x is near y and y is near z. Note that
(z −x) = ((z ∩ y) −x) ∪ ((z −y) −x)
But (z −y) −x is finite because (z −y) is finite and ((z ∩ y) −x) is finite
because ((z ∩ y) − x) ⊆ (y − x), which is finite. So the union, (z − x), is
finite.
Similarly,
(x −z) = ((x ∩ y) −z) ∪ ((x −y) −z)
where ((x∩y) −z) ⊆ (y −z) and ((x−y) −z ⊆ (x−y). So (x−z) is also
finite.
Hence, x is near z.
Fact 3.2.18.
a. If x
1
is near x
2
, then x
1
∪ y is near x
2
∪ y.
b. If x
1
is near x
2
and y
1
is near y
2
, then x
1
∪ y
1
is near x
2
∪ y
2
.
c. If x is near y, then N −x is near N −y
Proof.
3.2. CA - AXIOMS FOR CS 29
a. Since x
1
is near x
2
, x
1
−x
2
and x
2
−x
1
are finite. But
(x
1
∪ y) −(x
2
∪ y) ⊆ (x
1
−x
2
)
and (x
2
∪ y) −(x
1
∪ y) ⊆ (x
2
−x
1
)
b. By (a), x
1
∪ y
1
is near x
2
∪ y
1
, which is near x
2
∪ y
2
. So x
1
∪ y
1
is near
x
2
∪ y
2
, by transitivity.
c. (N − x) − (N − y) = y − x and (N − y) − (N − x) = x − y. So if x and y
are near each other, so are their complements.
We can now define the domains of the models we will use to establish the
independence results.
Definition 3.2.19.
a. Q
n
= ¦ y [ y is near an n-quasi-congruence class ¦
b. Q =

n>0
Q
n
Examples.
a. Q
1
= ¦ y ⊆ N [ y is finite or cofinite ¦.
b. Q
2
= ¦ y ⊆ N [ y is finite, cofinite, near [2n] or near [2n + 1] ¦.
Q
2
is the domain of the model constructed above to satisfy DIV
2
.
Fact 3.2.20. If A is a class of sets such that
a.

A ∈ A,
b. If x ∈ A, then

A−x ∈ A, and
c. If x ∈ A and y ∈ A, then x ∪ y ∈ A,
then A forms a boolean algebra under the usual set-theoretic operations,
where I is interpreted as A. [Monk, Def. 9.1, p.141 and Corr 9.4, p.142]
Theorem 3.2.21. If n > 0, then Q
n
forms an atomic boolean algebra under
the usual set-theoretic operations.
Proof. Using Fact 3.2.20:
a.

¦Q
n
¦ = N, since N ∈ ¦Q
n
¦ and if x ∈ Q
n
, then x ⊆ N.
b. If x ∈ Q
n
, then
Suppose x is near y
and y ∈ QC
n
So N −y ∈ QC
n
by 3.2.13b
and N −x is near N −y by 3.2.18b
So N −x ∈ Q
n
.
30 3. A FORMAL THEORY OF CLASS SIZE
c.
Suppose x ∈ Q
n
and y ∈ Q
n
then x is near x

and x

∈ QC
n
for some x

and y is near y

and y

∈ QC
n
for some y

But then x

∪ y

∈ QC
n
by 3.2.13a
and x ∪ y is near x

∪ y

by 3.2.18b
So x ∪ y ∈ QC
n
Thus, Q
n
is a boolean algebra. Moreover, every singleton is in Q
n
since all
singletons are near ∅. So Q
n
is an atomic boolean algebra.
We now define a size function on all sets which are near quasi-congruence
classes. The sizes assigned to sets are ordered pairs. The first member is a
rational between 0 and 1 which represents the density of the set. The second
member is an integer which represents the finite (possibly negative) deviation
of a set from average sets of the same density. First, we define the ordering
and arithmetic for these sizes with the intention of inducing the size ordering
and Sum relation for sets from the assignment of sizes to sets.
Definition 3.2.22.
a. A size is an ordered pair ¸ρ, δ), where ρ is rational and δ is an integer.
b. If θ
1
= ¸ρ
1
, δ
1
) and θ
2
= ¸ρ
2
, δ
2
) are sizes, then
(i) θ
1
< θ
2
iff ρ
1
< ρ
2
or (ρ
1
= ρ
2
∧ δ
1
< δ
2
).
(ii) θ
1

2
= ¸ρ
1

2
, δ
1

2
)
Only some of these sizes will actually be assigned to sets. Specifically, a size
will be assigned to a set only if 0 ≤ ρ ≤ 1. Moreover, if ρ = 0, then δ ≥ 0 and
if ρ = 1, then δ ≤ 0.
Our intention in assigning sizes to sets is as follows: Suppose x is near an
n-quasi-congruence class, x

. So x

is the union of k ≤ n n-quasi-congruence
classes. The set x

has density k/n and this is the value, ρ, assigned to x. The
δ value assigned to x is the finite number of elements added to or removed from
x

to obtain x.
The definitions and facts below formalize this intention and demonstrate
that the assignment of sizes to sets is well-defined.
Fact 3.2.23.
3.2. CA - AXIOMS FOR CS 31
a. If x ∈ QC, y ∈ QC, and x ,= y, then x is not near y. (That is, no
two quasi-congruence classes are near each other.)
b. Any set is near at most one quasi-congruence class.
Proof.
a. Let n be the least number such that
x ∈ QC
n
and y ∈ QC
n
So each is a disjoint union of n-congruence classes:
x = x
1
∪ . . . ∪ x
j
and y = y
1
∪ . . . ∪ y
k
Suppose a ∈ x −y
then a ∈ x
i
for some i.
But a / ∈ y
j
, for any y.
So x
i
,= y
j
, for any j
So x
i
⊆ x −y
So x −y is infinite, since x
i
is infinite
Similarly, if a ∈ y −x, then y −x is infinite.
b. If x were near two quasi-congruence classes, the two would have to be near
each other, since NEAR is transitive. But this is impossible by (a).
Definition 3.2.24. If x is near a quasi-congruence class, then
a. C(x) is the quasi-congruence class near x.
b. ∆
1
(x) = x −C(x).
c. ∆
2
(x) = C(x) −x.

1
(x) and ∆
2
(x) are finite and
x = (C(x) ∪ ∆
1
(x)) −∆
2
(x)
Definition 3.2.25. If x ∈ QC, then
a. α(x) = the least n such that x ∈ QC
n
.
b. β(x) = the unique k such that x is the disjoint union of k α(x)-
congruence classes.
32 3. A FORMAL THEORY OF CLASS SIZE
Examples.
a. If x = [2n + 1], α(x) = 2 and β(x) = 1.
b. If x = [4n + 1] ∪ [4n + 2], α(x) = 4 and β(x) = 2.
c. If x = [4n + 1] ∪ [4n + 3], α(x) = 2 and β(x) = 1, since x = [2n + 1].
Definition 3.2.26. If x ∈ Q, then
ρ(x) =
β(C(x))
α(C(x))
δ(x) = [∆
1
(x)[ −[∆
2
(x)[
θ(x) = ¸ρ(x), δ(x))
We can, at last, define the models to be used in our independence proof.
Definition 3.2.27. For n > 0, Q
n
is the interpretation Q of L
C<
in which
boolean symbols receive their usual interpretation and
˙
Q = Q
n
Q x < y iff θ(x) < θ(y)
Q x ∼ y iff θ(x) = θ(y)
Q Unit(x) iff θ(x) = ¸0, 1)
Q Sum(x, y, z) iff θ(z) = θ(x) +θ(y)
To show that the models Q
n
satisfy BASIC, we will need the following facts
about congruence classes.
Fact 3.2.28.
a. If x = [an +b] and a
2
= ac, then
x =
_
0≤i<c
[a
2
n + (ia +b)]
b. If x ∈ QC
n
and m = kn, then x ∈ QC
m
.
c. If x ∈ QC and y ∈ QC, there is an n such that x ∈ QC
n
and y ∈ QC
n
.
Proof.
a. If k ∈ [a
2
n + (ia +b)] for some i, 0 ≤ i < c, then, for some n
1
,
k = a
2
n
1
+ia +b
= a(cn
1
) +ia +b
= a(cn
1
+i) +b
So k ∈ x.
3.2. CA - AXIOMS FOR CS 33
If k ∈ x, there is an n
1
such that k = an
1
+ b. Let n
2
be the greatest n
such that a
2
n ≤ k and let k

= k −a
2
n
2
.
Since k ≡ b mod a
and a
2
n
2
≡ 0 mod a
then k

≡ b mod a
So k

= ia +b, where 0 ≤ i < c
But k = a
2
n
2
+k

= a
2
n
2
+ia +b
So k ∈ [a
2
n + (ia +b)]
b. By (a), each n-congruence class is a disjoint union of m-congruence classes.
c. Suppose x ∈ QC
n1
and y ∈ QC
n2
. Then, by (b), both x and y are in
QC
n1n2
.
Theorem 3.2.29. For any n > 0, Q
n
BASIC
Proof. By 3.2.21, Q
n
is an atomic boolean algebra; so Q
n
BA. The <-relation
of Q
n
is induced from the linear ordering of sizes; so it is a quasi-linear ordering
and IRREF
<
, TRICH, and DEF

are satisfied. As for the remaining axioms:
a. SUBSET: Suppose x ⊂ y. If C(x) = C(y) then ρ(x) = ρ(y) and ∆
1
(x) ⊆

1
(y) and ∆
2
(x) ⊆ ∆
2
(y), where at least one of these inclusions is proper.
So δ(x) < δ(y).
But if C(x) ,= C(y), then C(x) ⊂ C(y), so ρ(x) < ρ(y).
In either case, θ(x) < θ(y), so Q
n
x < y.
b. REP
<
: Suppose Q
n
x < y. So θ(x) = ¸k
1
/n, δ
1
), θ(x) = ¸k
2
/n, δ
2
), and
either k
1
< k
2
or δ
1
< δ
2
.
We want to find some x

⊂ y such that Q
n
x ∼ x

.
If k
1
= k
2
> 0, then y must be infinite; so, x

can be obtained by removing
δ
2
−δ
1
atoms from y.
If k
1
= k
2
= 0, then 0 ≤ δ
1
< δ
2
; so, again, x

can be obtained by removing
δ
2
−δ
1
atoms from y.
If k
2
> k
1
> 0, then let y
1
be the union of k
1
n-congruence classes
contained in C(y). So y − y
1
is infinite and y
1
− y is finite. Let y
2
=
y
1
−(y
1
−y) = y
1
∩y. So θ(y
2
) = ¸k
1
/n, −δ
3
) where δ
3
= [y
1
−y[. Finally,
let δ
4
= δ
3

1
and x

= y
2
∪ y
3
, where y
3
⊆ y
1
−y with [y
3
[ = δ
4
.
If k
1
= 0, then x is finite. If k
2
> 0, then y is infinite, so there is no
problem. If k
2
= 0, then y is finite, but has more members than x, since
δ
1
< δ
2
. So, let x

be any proper subset of y with δ
1
members.
34 3. A FORMAL THEORY OF CLASS SIZE
c. DISJ

: It is enough to show that if x and y are disjoint, then θ(x ∪ y) =
θ(x) +θ(y). We need the following three facts:
(i) C(x ∪ y) = C(x) ∪ C(y). (See Fact 3.2.18b.)
(ii) ∆
1
(x ∪ y) = (∆
1
(x) ∪ ∆
1
(y)) −(C(x) ∪ C(y)).
If z ∈ x ∪ y but a / ∈ C(x ∪ y), then a ∈ ∆
1
(x) or a ∈ ∆
1
(y); any
element of ∆
1
(x) is also in ∆
1
(x∪y) unless it is in C(y); any element
of ∆
1
(y) is also in ∆
1
(x ∪ y) unless it is in C(x).
(iii) ∆
2
(x ∪ y) = (∆
2
(x) ∪ ∆
2
(y)) −(∆
1
(x) ∪ ∆
1
(y)).
Note that if x ∈ Q, y ∈ Q, and x ∩ y = ∅, then C(x) ∩ C(y) = ∅;
otherwise, C(x) and C(y) have an infinite intersection.
Hence, if a ∈ C(x) ∪ C(y) −(x ∪ y)
then a ∈ C(x) −x = ∆
2
(x)
or a ∈ C(y) −y = ∆
2
(y)
And if a ∈ ∆
2
(x), then a ∈ ∆
2
(x ∪ y)
unless a ∈ ∆
1
(y)
And if a ∈ ∆
2
(y), then a ∈ ∆
2
(x ∪ y)
unless a ∈ ∆
1
(x)
.
From (ii) we obtain (iia):
(iia) [∆
1
(x ∪ y)[ = [∆
1
(x) ∪ ∆
1
(y)[
−[(∆
1
(x) ∪ ∆
1
(y) ∩ (C(x) ∪ C(y))[
and from (iii) we obtain (iiia):
(iiia) [∆
2
(x ∪ y)[ = [∆
2
(x) ∪ ∆
2
(y)[
−[∆
1
(x) ∪ ∆
1
(y) ∩ (∆
2
(x) ∪ ∆
2
(y))[
But ∆
1
(x) and ∆
1
(y) are disjoint, since x and y are disjoint. So:
[∆
1
(x) ∪ ∆
1
(y)[ = [∆
1
(x)[ +[∆
1
(y)[ = δ
1
(x) +δ
1
(y) (iv)
Since ∆
2
(x) and ∆
2
(y) are contained, respectively, in C(x) and C(y),
which are disjoint, they are also disjoint. So:
[∆
2
(x) ∪ ∆
2
(y)[ = δ
2
(x) +δ
2
(y) (v)
3.2. CA - AXIOMS FOR CS 35
So,
δ
1
(x ∪ y)−δ
2
(x ∪ y)
= (δ
1
(x) +δ
1
(y)) −(δ
2
(x) +δ
2
(y))
= (δ
1
(x) −δ
2
(x)) + (δ
1
(y) −δ
2
(y))
= δ(x) +δ(y)
Since ρ(x ∪ y) = ρ(x) +ρ(y), by (i), we know that θ(x ∪ y) = θ(x) +θ(y).
d. DEF
+
:
Suppose Q
n
Sum(x, y, z)
So θ(z) = θ(x) +θ(y)
Clearly θ(x) ≤ θ(z)
Assume θ(x) = θ(z)
then θ(y) = ¸0, 0)
So y = ∅
Let x

= z, y

= ∅ to satisfy DEF
+
Assume θ(x) < θ(z)
then Q
n
x

∼ x
and x

⊂ z for some x

since Q
n
REP
<
Let y

= z −x

So z = x

y

Claim θ(y) = θ(y

) so Q
n
y ∼ y

For θ(x

) +θ(y

) = θ(z) by DISJ

But θ(x

) = θ(x)
So θ(y

) = θ(z) −θ(x) = θ(y)
(Cancellation is valid for sizes because it is valid for rationals and integers.)
Conversely,
If θ(z) = θ(x

) +θ(y

)
and θ(x

) = θ(x)
and θ(y

) = θ(y)
then θ(z) = θ(x) +θ(y)
Theorem 3.2.30. For any n > 0, Q
n
DIV
m
iff m [ n.
Proof.
36 3. A FORMAL THEORY OF CLASS SIZE
⇒ θ(N) = ¸1, 0). Hence, if m disjoint sets of the same size exhaust N, they
must each have size ¸1/m, 0). But if x ∈ Q
n
, then θ(x) = ¸a/n, b), for
integral a and b. So b = 0 and a = n/m.
⇐ For each i, 0 ≤ i < n, let A
i
= [nk +i]. So N =

0≤i<n
A
i
.
If i ,= j, then A
i
∩ A
j
= ∅ and Q
n
(A
i
∼ A
j
) since θ(A
i
) = θ(A
j
) =
¸1/n, 0).
Letting p = n/m, group the n sets A
i
into m collections with p members
in each:
B
1
, . . . , B
m
Letting b
j
=

B
j
for 1 ≤ j ≤ m, we have b
j
∈ Q
n
and θ(b
j
) = ¸p/n, 0) =
¸1/m, 0).
Furthermore, b
1
∪ . . . b
m
= N.
Definition 3.2.31. If J ,= ∅ and J is finite, then the least common mul-
tiple of J, µ(J), is the least k which is divisible by every member of J.
Remark. µ(J) always exists since the product of all members of J is di-
visible by each member of J. Usually, the product is greater than µ(J).
Corollary 3.2.32. If J is finite, then
a. If BDIV
n
¬ DIV
m
, then m [ n.
b. If BDIV
J
¬ DIV
m
, then m [ µ(J).
c. There are only finitely many m for which BDIV
J
¬ DIV
m
.
Proof. Based on Theorem 3.3.20:
a. If m n, then Q
n
BDIV
n
; DIV
m
b. Q
µ(J)
BDIV
J
since it satisfies DIV
j
for each j ∈ J. But, if m µ(J),
then Q
µ(J)
DIV
m
.
c. Immediate from (b), since only finitely many m divide µ(J).
We are now ready to show that BDIV
n
CS by finding a sentence in CS
which entails infinitely many DIV
n
sentences. Such sentences can be produced
by generalizing the notion of divisibility to all sets instead of applying it only
to the universe.
Definition 3.2.33.
3.2. CA - AXIOMS FOR CS 37
a. If 0 ≤ n, then Times
n
(x, y) is the formula:
∃x
0
. . . ∃x
n
[x
0
= ∅ ∧ x
n
= y ∧

1≤i≤n
Sum(x
i−1
, x, x
i
)]
Times
n
(x, y) says that y is the same size as the Sum of n sets, each
the same size as x.
b. If 0 ≤ m < n, then Mod
n,m
(z) is the formula:
∃x∃y∃v∃w[Times
n
(x, v) ∧ Unit(y) ∧ Times
m
(y, w) ∧ Sum(v, w, z)]
Mod
n,m
(z) says that z can be partitioned in n sets of the same size
and m atoms.
c. Div
n
(z) is the formula
Mod
n,0
(z) ∨ . . . ∨ Mod
n,n−1
(z)
d. ADIV
n
is the sentence
∀xDiv
n
(x)
Remark. We have taken this opportunity to formulate the divisibility
predicates purely in terms of size predicates. Notice that in the presence
of BASIC,
MOD
n,m
≡ Mod
n,m
(I)
and DIV
n
≡ Div
n
(I)
Fact 3.2.34. CS ¬ ADIV
n
, for every n.
Proof. Every set in every standard finite interpretation is a finite set, and all
finite sets are roughly divisible by every n.
Fact 3.2.35. BASIC; ADIV
n
¬:
a. ADIV
n
m, for all m
b. DIV
n
, and
c. DIV
n
m, for all m.
Proof.
a. By induction on m: if m = 1, then n
m
= n, so ADIV
n
¬ ADIV
n
m.
If T ¬ ADIV
n
k, / T, and x ∈
˙
/, then x can be partitioned into n
k
sets of the same size and i atoms, where i < n
k
. Each non-atomic set in
the partition can be further partitioned into n sets of the same size and
atoms, where j < n. Thus, we have partitioned x into n
k
n sets of the
same size and n
k
j + i atoms. But n
k
n = n
k+1
and, since i < n
k
and
j < n, n
k
j +i < n
k+1
. Hence, / ADIV
n
k+1.
38 3. A FORMAL THEORY OF CLASS SIZE
b. Obvious.
c. Immediate from (a) and (b).
Theorem 3.2.36. BDIV
N
ADIV
n
for any n > 1.
Proof. If BDIV
N
¬ ADIV
n
, then by compactness there is a finite set J such
that BDIV
J
¬ ADIV
n
. But then BDIV
J
¬ DIV
n
k for every k, by Fact 3.2.35c.
But this contradicts Fact 3.2.32, which says that BDIV
J
entails only finitely
many DIV
n
sentences.
So, even if we add all of the DIV sentences to BASIC, we are left with a
theory weaker than CS. Since this weakness has arisen in the case of ADIV
sentences, it is reasonable to attempt an axiomatization of CS as follows:
Definition 3.2.37. CA ≡ BASIC∪ ¦ ADIV
n
[ n > 0 ¦
The remainder of this chapter and the next two are devoted to showing that
CA is, indeed, a complete set of axioms for CS.
3.3 Remarks on showing that CA axiomatizes
CS
We know that CS ¬ CA and we want to show that CA ≡ CS, i.e. that CA ¬ CS.
To do so, it will be sufficient to show that every consistent extension of CA is
consistent with CS.
Fact 3.3.1.
a. (Lindenbaum’s lemma) Every consistent theory has a consistent,
complete extension.
b. If every consistent, complete extension of T
2
is consistent with T
1
,
then T
2
¬ T
1
.
Proof.
a. [Monk, Theorem 11.13, p.200]
b. Suppose that T
1
¬ φ, T
2
φ. Then T = T
2
; φ is consistent. T has a
consistent, complete extension, T

, by Lindenbaum’s lemma. Since T
2

T, T

is also a consistent, complete extension of T
2
. But T

is not consistent
with T
1
.
Definition 3.3.2. T

is a completion of T iff T

is a complete, consistent
extension of T.
3.3. REMARKS ON SHOWING THAT CA AXIOMATIZES CS 39
To prove that every completion of CA is consistent with CS, we define two
kinds of completions of a theory.
Definition 3.3.3.
a. T

is a finite completion of T iff T

is true in some finite model of
T.
b. T

is an infinite completion of T iff T

is true in some infinite model
of T.
Fact 3.3.4. If T

is a completion of T and BA ⊆ T, then (T

is a finite
completion of T iff T

is not an infinite completion of T.)
Proof.
⇒ Suppose / is a finite model of T

with n atoms. Since T

is complete,
T

¬ EXACTLY
n
. So T

¬ ATLEAST
n+1
and has no infinite models.
⇐ T

¬ ATLEAST
n
for every n, so T

has no finite models.
Fact 3.3.5.
a. Every finite completion of CA is equivalent to CA;EXACTLY
n
, for
some n.
b. Every finite completion of CA is consistent with CS.
Proof.
a. CA ¬ BASIC and, by Theorem 2.1.6, BASIC is categorical in every finite
power.
b. The model / of CA; EXACTLY
n
is a standard finite interpretation. So
/ CS.
Definition 3.3.6.
a. CAI = CA∪ INF
b. CSI = CS ∪ INF
So, to show that every completion of CA is consistent with CS, we may now
concentrate on showing that every completion of CAI is consistent with CSI.
What, then, are the completions of CAI? Recall that CA entails DIV
n
for
every n > 0, where DIV
n
is
MOD
n,0
∨ . . . ∨ MOD
n,n−1
Any completion, T, of CAI has to solve the disjunction DIV
n
for each n; T
has to entail one of the disjuncts. Remainder theories specify, for each n, the
number of atoms remaining when the universe is divided into n disjoint sets of
the same size.
40 3. A FORMAL THEORY OF CLASS SIZE
Definition 3.3.7. Remainder functions and remainder theories
a. f : N
+
→ N is a remainder function iff 0 ≤ f(n) < n for all n ∈
Dom(f). (Henceforth f ranges over remainder functions.)
b. f is total iff Dom(f) = N
+
; otherwise f is partial.
c. f is finite iff Dom(f) is finite.
d. n is a solution for f iff for any i ∈ Dom(f), n ≡ f(i) mod i.
e. f is congruous iff for any i, j ∈ Dom(f), then gcd(i, j) [ f(i) − f(j);
otherwise, f is incongruous.
f. The remainder theory specified by f, RT
f
, is the set of sentences
¦ MOD
n,m
[ f(n) = m¦.
g. If T is a theory, T
f
= T ∪ RT
f
.
Chapter 5 will show that if f is total, CAI
f
is complete and that these are
the only complete extensions of CAI. In this section, we will show that CAI
f
is
consistent just in case CSI
f
is consistent.
Fact 3.3.8.
a. If f is finite, then f has a solution iff f is congruous iff f has in-
finitely many solutions. [Griffin, Theorem 5-11, p.80]
b. f is congruous iff every finite restriction of f is congruous.
c. There are congruous f without any solutions. (Let f(p) = p − 1,
for all primes p. Any solution would have to be larger than every
prime.)
Theorem 3.3.9.
a. If f is finite, then CS
f
is consistent iff f is congruous.
b. CS
f
is consistent iff f is congruous.
c. CSI
f
is consistent iff f is congruous.
Proof.
a. (⇒)Let φ =
_
RT
f
, Since CS; φ is consistent, there is some n such that
/
n
φ. So n is a solution of f and, hence, f is congruous by 3.3.8a.
(⇐)If f is congruous, f has a solution, n. So /
n
CS; φ
b. (⇒)For every finite restriction, g, of f, CS
g
is consistent. By (a), each
such g is congruous. Hence, f is congruous by 3.3.8b.
(⇐)Every finite restriction, g, of f is congruous. So CS
g
is consistent, by
(a). By compactness, then, CS
f
is consistent.
3.3. REMARKS ON SHOWING THAT CA AXIOMATIZES CS 41
c. (⇒)If CSI
f
is consistent, so is CS
f
. So, by (b), f is congruous.
(⇐)By compactness, it is sufficient to show that every finite subtheory, T,
of CSI
f
is consistent. But, if T is such a theory, then
T ⊆ CS
g
∪ ¦ ATLEASTi [ i < n¦
for some n and some finite restriction, g, of f. Since f is congruous, g is
as well, by 3.3.8b. So g has arbitrarily large solutions and CS
g
has finite
models large enough to satisfy T. Hence, T is consistent.
We now want to prove a similar theorem for CA, our proposed axiomatization
of CS. To do this, we must first establish that certain sentences are theorems
of CA.
Lemma 3.3.10. If n [ m, 0 ≤ q < n, and p ≡ q mod n, then
CA ¬ MOD
m,p
→ MOD
m,q
.
Proof. Suppose / MOD
m,p
, k
1
= m/n, and p = k
2
n + q. So, B(/) can be
partitioned into m sets of the same size
b
1,1
. . . b
1,n
b
2,1
. . . b
2,n
b
k1,1
. . . b
k1,n
and p atoms
a
1,1
. . . a
1,n
a
2,1
. . . a
2,n
a
k2,1
. . . a
k2,n
c
1
. . . . . . c
q
For 1 ≤ i ≤ n, let
B
i
=
_
1≤j≤k1
b
j,n

_
1≤j≤k2
a
j,n
Since / DISJ

, / B
i
∼ B
j
, for 1 ≤ i, j ≤ n. Furthermore,
B(/) =
_
1≤i≤n
B
i
∪ (c
1
. . . c
q
)
So, / MOD
n,q
.
Lemma 3.3.11. If 0 ≤ p < q < m, then
CA ¬ MOD
m,p
→ MOD
m,q
42 3. A FORMAL THEORY OF CLASS SIZE
Proof. Suppose / MOD
m,p
∧ MOD
m,q
. Then:
/ x
1
∪ . . . x
m
∪ a
1
∪ . . . a
p
= I
/ y
1
∪ . . . y
m
∪ b
1
∪ . . . b
q
= I
where the a’s and b’s are atoms and the x’s (y’s) are disjoint sets of the same
size in /. Let
X = x
1
∪ ∪ x
m
Y = y
1
∪ ∪ y
m
A = a
1
∪ ∪ a
p
B = b
1
∪ ∪ b
p
B

= b
p+1
∪ ∪ b
q
We claim that y
1
< x
1
. For, if y
1
∼ x
1
, then X ∪ A ∼ Y ∪ B and if x
1
< y
1
,
then X ∪ A < Y ∪ B; neither is possible since Y ∪ B ⊂ I = X ∪ a. So y
i
< x
i
for 1 ≤ i ≤ m. Since / REP
<
, there is a proper subset y

i
of x
i
which is the
same size at y
i
.
Let z
i
= x
i
−y

i
and Y

= y

i
∪ ∪ y

m
So Y

∪ Z = X = I −A
and Y

∪ B

= I −B.
But A ∼ B, since each is the disjoint union of p atoms. Thus (I − A) ∼
(I − B), by RC

, so Y

∪ Z ∼ Y ∪ B

. But Y

∼ Y , since for each component
of Y , there is a component of Y

of the same size. So Z ∼ B

, by RC

.
But Z must be larger than B

, for B

is the union of fewer than m atoms
while Z is the union of m non-empty sets. So, the original supposition entails a
contradiction.
Theorem 3.3.12.
a. If f is finite, then CA
f
is consistent iff f is congruous.
b. CA
f
is consistent iff f is congruous.
c. CAI
f
is consistent iff f is congruous.
Proof.
a. (⇒)Supposing that f is incongruous, there exist i, j, and k such that
k = gcd(i, j) and k (f(i) −f(j)). We will show that
CA ¬ (MOD
i,f(i)
∧ MOD
j,f(j)
) (*)
3.3. REMARKS ON SHOWING THAT CA AXIOMATIZES CS 43
from which it follows that CA
f
is inconsistent.
Let p and q be such that
0 ≤ p, q < k,
f(i) ≡ p mod k, and
f(j) ≡ q mod k
By lemma 3.3.10, we have
CA ¬ MOD
i,f(i)
→ MOD
k,p
, and (1)
CA ¬ MOD
j,f(j)
→ MOD
k,q
(2)
since k [ i and k [ j. Since k f(i) −f(j), p ,= q. So lemma 3.3.11 yields
CA ¬ MOD
k,p
→ MOD
k,q
(3)
From (1), (2), and (3) we may conclude (*).
(⇐)Follows from 3.3.9a since CA
f
⊆ CS
f
.
b. See proof of 3.3.9b.
c. See proof of 3.3.9c.
Corollary 3.3.13. CAI
f
is consistent iff CSI
f
is consistent.
Proof. Immediate from 3.3.9c and 3.3.12c.
It might help to review our strategy before presenting the difficult parts of
the proof that CA ≡ CS. The main objective is (1), which follows from (2) by
3.3.1b.
(1) CA ¬ CS
(2) Every completion of CA is consistent with CS.
We already know that the finite completions of CA are consistent with CS
(see 3.3.5c) and that if CAI
f
is consistent, then CSI
f
is also consistent (see
3.3.12). So (2) is a consequence of (3).
(3) If T is a completion of CAI, then T ≡ CAI
f
for some total,
congruous f.
To establish (3), it is sufficient to prove (4) because every completion of CAI
entails CAI
f
for some total f.
(4) If f is total and congruous, then CAI
f
is complete.
44 3. A FORMAL THEORY OF CLASS SIZE
To prove (4), we invoke the prime model test: If T is model complete and T
has a prime model, then T is complete (see Appendix, Fact C.3). So (4) follows
from (5) and (6).
(5) If f is total and congruous, then CAI
f
has a prime model.
(6) For any f, CAI
f
is model complete.
Finally, since any extension of a model complete theory is also model com-
plete (see Appendix, Fact C7a), we can infer (6) from (7).
(7) CAI is model complete.
So (1) follows from (5) and (7).
The proof outlined here will be carried out in Chapter 5. But first, we
consider a simpler theory, PSIZE, which deals only with sizes of sets and ignores
boolean relations. Chapter 4 formulates PSIZE and establishes that it is model
complete, a result we need for showing that CAI is model complete.
4
The Pure Theory of Class
Sizes
CS is about sets; it makes claims about sets in terms of their boolean relations
and their size relations. In this chapter, we identify a theory, PCS (the pure
theory of class sizes), which is not about sets, but only about sizes of sets. Sizes,
here, are equivalence classes of sets which have the same size. PCS is formulated
in the language of size relations, L
<
.
PCS is worth examining in its own right, for if number theory is the theory
of cardinal numbers, then PCS is our version of number theory. But our main
reason for introducing PCS is to aid the proof that CA is model complete. For
this reason,we only give a sketchy treatment of PCS itself.
Section 1 defines PCS and develops a set of axioms, PCA, for it, as follows:
for each model, /, of BASIC, the size model, o
A
, consists of equivalence classes
drawn from
˙
/ under the same size relation; PCS is the set of statements true
in o
A
for any standard finite model, /. PCA consists of a theory, PSIZE, which
holds in o
A
whenever / BASIC and a set of divisibility principles.
Using some results about model theory in section 2, Section 3 establishes
that PCA is model complete, the main result of this chapter and the only result
needed for subsequent proofs. This is done by reducing PCA to the theory Zgm,
whose models are Z-groups taken modulo some specific element.
Finally, section 4 indicates how PCA could be shown to axiomatize PCS.
This method is the same outlined in chapter 3 to show that CA ≡ CS.
4.1 Size models and PCS
Definition 4.1.1. Suppose / BASIC.
a. If x is a member of
˙
/, then σ
A
[x] is the size of x in /:
σ
A
[x] = ¦ y [ / (x ∼ y) ¦
45
46 4. THE PURE THEORY OF CLASS SIZES
b. o
A
, the size model for /, is the interpretation of L
<
whose domain
is
_
σ
A
[x] [ x ∈
˙
/
_
where
o
A
σ
A
[x] ∼ σ
A
[y] iff / x ∼ y
o
A
σ
A
[x] < σ
A
[y] iff / x < y
o
A
Sum(σ
A
[x], σ
A
[y], σ
A
[z]) iff / Sum(x, y, z), and
o
A
Unit(σ
A
[x]) iff / Atom(x)
c. The one-place operator ˜ x is to be read as the complementary size of x:
o
A
(y = ˜ x) iff o
A
Sum(x, y, I)
These interpretations are well defined because the predicates are satisfied
by elements of
˙
/ in virtue of their sizes. For example, if / Sum(x, y, z) and
/ z ∼ z

, then / Sum(x, y, z

).
Definition 4.1.2. PCS, the pure theory of class sizes, consists of all
sentences of L
<
which are true in the size model of every standard finite
interpretation of L
C<
.
Definition 4.1.3. PSIZE, the theory of sizes, consists of the following
axioms:
Order axioms
(x < x) (IRREF
<
)
x < y ∧ y < z → x < z (TRANS
<
)
x ∼ y ↔ x = y (UNIQ

)
x ≤ I (MAX)
0 ≤ x (MIN)
x < y ∨ x ∼ y ∨ y < x (TRICH)
Unit axioms
Unit(x) ↔ (y < x ↔ y = 0)
∃xUnit(x)
4.2. SOME MODELS AND THEORIES 47
Sum axioms
Sum(x, 0, x) (IDENT)
Sum(x, y, z) ↔ Sum(y, x, z) (COMM)
Sum(x
1
, y, z
1
) ∧ Sum(x
2
, y.z
2
) → (MONOT)
(x
1
< x
2
↔ z
1
< z
2
)
Sum(x, y, w
1
) ∧ Sum(w
1
, z, w) ∧ Sum(y, z, w
2
) (ASSOC)
→ Sum(x, w
2
, w)
∃zSum(x, y
1
, z) ∧ y
2
≤ y
1
→ ∃zSum(x, y
2
, z) (EXIST
+
)
x ≤ z → ∃ySum(x, y, z) (EXIST

)
Sum(x, ˜ x, I) (COMP)
Fact 4.1.4. If / BASIC, then o
A
PSIZE
Fact 4.1.5. PSIZE PCS.
Proof. PSIZE fails to axiomatize PCS for the same reason that BASIC fails to
axiomatize CS: the lack of divisibility principles.
We offer PCA as an axiomatic version of PCS:
Definition 4.1.6. PCA, the pure class size axioms, is defined as:
PSIZE ∪ ¦ ADIV
n
[ n > 0 ¦
Fact 4.1.7. If / CA, then o
A
PCA
If / is a standard finite interpretation of BASIC with n atoms, then the
elements of o
A
can be regarded as the sequence 0, . . . , n with the usual ordering,
where
o
A
Unit(x) iff x = 1
and
o
A
Sum(i, j, k) iff (i +j) = k ≤ n
Once n is fixed, this is the only interpretation allowed by the axioms PSIZE. In
particular, MONOT rules out the interpretation in which Sum(i, j, k) is satisfied
just in case (i +j) ≡ k mod n + 1.
4.2 Some models and theories
We shall use Theorem 4.2.1a to show that PCA is model complete and Theorem
4.2.1b to show, in chapter 5, that CAI is model complete.
Theorem 4.2.1.
a. If T satisfies Monk’s Condition, then T is model complete.
48 4. THE PURE THEORY OF CLASS SIZES
Monk’s Condition. If / T, B T, / ⊆ B, and ( is a
finitely generated submodel of B, then there is an isomor-
phism, f : ( → / such that if x ∈
˙
( ∩
˙
/, then f(x) = x. f is
a Monk mapping.
b. If T is model complete and L
T
has no function symbols, then T
satisfies Monk’s Condition.
Proof.
a. [Monk, p.359]
b. If L
T
has no function symbols, then any finitely generated structure over
L
T
is finite. So, suppose
˙
( contains a
1
, . . . , a
n
(from
˙
/) and b
1
, . . . , b
m
(from
˙
B-
˙
/). Let φ
1
be the diagram of ( and obtain φ
2
from φ
1
by sub-
stituting the variable x
i
for each constant a
i
and the variable y
i
for each
constant b
i
. Finally, let φ
3
be
∃y
1
. . . ∃y
m
φ
2
So, φ
3
is a primitive formula. B φ
3
(a
1
, . . . , a
n
), so / does as well, by
Fact C.5d in the Appendix. So, to obtain the desired isomorphism, map
the a
i
’s into themselves and map the b
i
’s into a sequence of elements of
˙
/
which can stand in for the existentially quantified variables of φ
3
.
In chapter 5, we use Monk’s Theorem to infer the model completeness of the
theory CA from that of PCA. To establish the model completeness of PCA, we
use Fact 4.2.2.
Fact 4.2.2.
a. If T
1
is model complete and T
1
¬ T
2
, then T
2
is also model complete.
b. If T is model complete in L, and L

is an expansion of L by adjoining
new individual constants, then T is model complete in L

. [Monk,
p.355]
Definition 4.2.3. Suppose L
1
and L
2
are first order languages and L
12
=
L
1
−L
2
. A translation (simple translation) of L
1
into L
2
is a function,
τ, which:
a. assigns to the universal quantifier a (quantifier free) formula, τ

, of
L
2
with exactly one free variable.
b. assigns to each n-place predicate, P, in L
12
a (quantifier free) for-
mula, τ
P
, of L
2
with exactly n free variables, and
c. assigns to each n-place function symbol, O, in L
12
a (quantifier free)
formula, τ
O
, of L
2
with exactly (n + 1) free variables.
4.2. SOME MODELS AND THEORIES 49
Definition 4.2.4. If τ is a translation of L
1
into L
2
, then τ extends to all
formulae of L
1
as follows:
a. Predicates and function symbols of L
2
are translated into themselves.
b.
τ(φ ∨ ψ) = τ(φ) ∨ τ(ψ)
τ(φ ∧ ψ) = τ(φ) ∧ τ(ψ)
τ(φ) = τ(φ)
τ(∀xφ) = ∀x(τ

(x) → φ)
τ(∃xφ) = ∃x(τ

(x) ∧ φ)
Definition 4.2.5. If τ is a translation of L
1
into L
2
, then
a. The functional assumptions of τ are the sentences:
∀x
1
. . . ∀x
n


(x
1
) ∧ . . . ∧ τ

(x
n
)) →
∃y
1


(y
1
) ∧ ∀y
2

O
(x
1
, . . . , x
n
, y
2
) ↔ y
1
= y
2
))
where O is a function symbol in L
1
but not in L
2
.
b. The existential assumption of τ is
∃xτ

(x)
The functional assumptions of a translation say that the formulas which
translate function symbols yield unique values within the relevant part of the
domain when given values in the relevant part of the domain. The relevant
part of the domain is the set of elements which satisfy the interpretation of
universal quantifier. The existential assumption of a translation says that that
subdomain is non-empty. Notice that the existential and functional assumptions
of a translation are sentences of L
2
.
A translation from L
2
into L
1
induces a mapping from interpretations of L
2
into interpretations of L
1
.
Definition 4.2.6. If τ is a translation from L
1
into L
2
and B is an inter-
pretation of L
2
which satisfies the existential and functional assumptions
of τ, then τ(B) is the interpretation / of L
1
such that:
a.
˙
/ is the set of elements of
˙
B which satisfy τ

.
b. / interprets all predicates and function symbols common to L
1
and
L
2
in the same way that B does.
50 4. THE PURE THEORY OF CLASS SIZES
c. / interprets all predicates and function symbols in L
12
in accordance
with the translations assigned by τ.
/ P(x) iff B τ
P
(x)
/ y = O(x) iff B τ
O
(x, y)
The following condition on theories allows us to infer the model completeness
of one from the model completeness of the other.
Definition 4.2.7.
a. If τ is a translation from L
1
to L
2
, then T
1
is τ-reducible to T
2
iff
for every model / of T
1
there is a model B of T
2
such that / = τ(B).
b. T
1
is reducible (simply reducible) to T
2
iff there is a (simple) trans-
lation, τ, for which T
1
is τ-reducible to T
2
.
c. T
1
is uniformly τ-reducible to T
2
iff for any models, /
1
and B
1
,
such that /
1
T
1
and B
1
T
1
, and /
1
⊆ B
1
, there exist models /
2
and B
2
such that /
2
T
2
and B
2
T
2
, and /
2
⊆ B
2
/
1
= τ(/
2
) and
B
1
= τ(B
2
),
Lemma 4.2.8. Suppose that T
1
is τ-reducible to T
2
and that /
1
= τ(/
2
).
Then, for any primitive formula, φ, of L
1
and any sequence, x ∈ A
1
,
/
1
φ(x) iff /
2
τ
φ
(x)
Proof. Suppose
φ(x) = ∃y
1
. . . ∃y
n
φ

(x)
where φ

(x) is a conjunction of atomic formulae and negations of atomic formu-
lae. Then
/
1
φ(x) iff /
1
∃y
1
. . . ∃y
n
φ

(x, y
1
, . . . , y
n
)
iff /
1
φ

(x, b
1
, . . . , b
n
), for b
i

˙
/
iff /
2
τ
φ
(x, b
1
, . . . , b
n
)
iff /
2
∃y
1
. . . ∃y
n


(y
1
) ∧ . . . ∧ τ

(y
n
) ∧ φ

(x, y
1
, . . . , y
n
))
iff /
2
τ
φ
(x)
Theorem 4.2.9. If T
2
is model complete, τ is a simple translation from
L
T1
to L
T2
, and T
1
is uniformly τ-reducible to T
2
, then T
1
is also model
complete.
4.2. SOME MODELS AND THEORIES 51
Proof. By Fact C.5d in the Appendix, it is enough to show that given models
/
1
and B
1
of T
1
, where /
1
⊆ B
1
and a primitive formula, φ:
If B
1
φ(x) for x ∈
˙
/
then /
1
φ(x)
Since T
1
is uniformly τ-reducible to T
2
, there are models of T
2
, /
2
⊆ B
2
, where
/
1
= τ(/
2
) and B
1
= τ(B
2
)
Since B
1
φ(x) by assumption
then B
2
τ
φ
(x) by lemma 4.2.8
So /
2
τ
φ
(x) since T
2
is model complete
and /
1
φ(x) by lemma 4.2.8
We shall now define several theories, all more or less familiar, which will
serve in showing that our theory of size is model complete.
Definition 4.2.10.
a. The theory of abelian groups with identity has the following axioms
x + (y +z) = (x +y) +z (1)
x +y = y +x (2)
x + 0 = x (3)
∃y(x +y = 0) (4)
b. The theory of cancellable abelian semigroups with identity con-
sists of (1), (2), and (3) above and:
x +y = x +z → y = z (4

)
c. The axioms of simple order are:
x ≤ y ∧ y ≤ z → x ≤ z (5)
x ≤ y ∧ y ≤ x → x = y (6)
x ≤ x (7)
x ≤ y ∨ y ≤ x (8)
d. The theory of Z-groups, Zg , has the following axioms
(i) The axioms for abelian groups with identity,
(ii) The axioms for simple order,
52 4. THE PURE THEORY OF CLASS SIZES
(iii) The following additional axioms:
y ≤ z → x +y ≤ x +z (9)
1 is the least element greater than 0 (10)
∀x∃y(ny = x ∨ . . . ∨ ny = x + (n −1)) (11)
for each positive n, where ny stands for:
y + +y
. ¸¸ .
n times
e. The theory of N-semigroups has the following axioms:
(i) The axioms for cancellable abelian semigroups with identity,
(ii) The axioms of simple order,
(iii) Axioms (9) and (10) of Zg, and
(iv) The additional axiom:
0 ≤ x (12)
f. The theory of Z-groups modulo I, Zgm, consists of the following
axioms:
(i) The axioms for abelian groups with identity,
(ii) The axioms for simple order
(iii) Axioms (10) and (11) of Zg, axiom (12) from the theory of
N-semigroups, and
(iv) The following additional axioms
(y ≤ z ∧ x ≤ x +z) → x +y ≤ x +z) (13)
x ≤ I (14)
The theory of Z-groups is taken from [Chang, p. 291]
Fact 4.2.11.
a. Zg is the complete theory of ¸Z, +, 0, 1, ≤). [Chang, p.291]
b. Zg is model complete.[Robinson]
Theorem 4.2.12.
a. The theory of N-semigroups is model complete.
b. The theory of Z-groups modulo I is model complete.
4.3. PCA IS MODEL COMPLETE 53
Proof.
a. Every abelian semigroup with cancellation can be isomorphically embed-
ded in an abelian group [Kurosh, pp. 44-48]. It is clear from the con-
struction in [Kurosh] that if the semigroup is ordered, the abelian group
in which it is embedded may also be ordered and that the elements of
the semigroup will be the positive elements of the group. Moreover, the
(rough) divisibility of the elements in the semigroup will be carried over
to the group.
Consequently, the theory of N-semigroups is uniformly reducible to the
theory of Z groups by the translation:
τ

= 0 ≤ x
Since the latter is model complete, so is the former, by 4.2.9.
b. First, consider the theory of N-semigroups in the language which contains,
besides the constant symbols in the original theory, an individual constant,
I. The theory of N-semigroups is model-complete in this language, by
4.2.2b.
We claim that the theory of Z-groups modulo I is uniformly reducible to
this new theory by the following translation:
τ

= x ≤ I
τ
+
= x +y = z ∨ x +y = I +z
(The construction: given a model of Zgm, stack up ω many copies of the
model, assigning interpretations in the obvious way. The result is an N-
semigroup and the original model is isomorphic to the first copy of itself.)
In the next section, we use Theorem 4.2.9 to show that PSIZE is model
complete, by reducing it to the theory of Z-groups modulo I. Chapter 5 uses
Monk’s Theorem to show that CA is model complete.
4.3 PCA is model complete
To show that PCA is model complete, we shall reduce it to Zgm, the theory of
Z-groups with addition taken modulo some constant (see 4.2.11). The model
completeness of PCA then follows from the model completeness of Zgm (Fact
4.2.12) and Theorem 4.2.9. Specifically, we shall show that every model of PCA
is the τ-image of a model of Zgm, where τ is the following translation.
54 4. THE PURE THEORY OF CLASS SIZES
Definition 4.3.1. Let τ be the translation from L
PSIZE
to L
Zgm
where:
τ

= x = x
τ
Sum(x,y,z)
= x +y = z ∧ x ≤ z
τ
Unit(x)
= x = 1
τ
˜ x
= x +y = I
τ
x<y
= x ≤ y ∧ x ,= y
τ

= x = 0
Given a model, /, of PCA, we can construct a model, Z
A
, of Zgm directly:
Definition 4.3.2. If / PSIZE, then Z
A
is the interpretation of L
Zgm
in
which:
˙
Z
A
=
˙
/ (a)
Z
A
(x ≤ y) iff / (x < y ∨ x = y) (b)
Z
A
(I) = /(I) (c)
Z
A
(0) = /(0) (d)
Z
A
(x +y = z) iff / Sum(x, y, z) (e)
or / ∃a∃wUnit(a)
∧ Sum(˜ x, ˜ y, w) ∧ Sum(w, a, ˜ z)
Z
A
(x = 1) iff / Unit(x) (f)
Fact 4.3.3j establishes that 4.3.2 gives Sum a functional interpretation. The-
orem 4.3.8 establishes that Z
A
Zgm, on the basis of the intervening facts: 4.3.3
deals with the model / of PSIZE; 4.3.4 deals with the corresponding model Z
A
;
4.3.6 verifies some connections between / and Z
A
.
Fact 4.3.3. The following are theorems of PSIZE. (Sm(x, y) abbreviates
∃zSum(x, y, z). )
4.3. PCA IS MODEL COMPLETE 55
Sum(x, y, z
1
) ∧ Sum(x, y, z
2
) → z
1
= z
2
(UNIQ
+
) (a)
Sum(x, y
1
, z) ∧ Sum(x, y
2
, z) → y
1
= y
2
(UNIQ

) (b)
Sum(x, y, w
1
) & Sum(w
1
, z, w) → (ASSOC2) (c)
∃w
2
(Sum(y, z, w
2
) ∧ Sum(x, w
2
, w))
˜
˜ x = x (d)
x < y ↔ ˜ y < ˜ x (e)
x < y ∧ y < ˜ y → x < ˜ x (f)
Sm(x, y) ∨ Sm(˜ x, ˜ y) (g)
Sum(x, y, z) → Sum(˜ z, y, ˜ x) (h1)
Sum(x, y, z) → Sum(x, ˜ z, ˜ y) (h2)
Sum(x, y, z
1
) ∧ Sum(˜ x, ˜ y, ˜ z
2
) → (z
1
= z
2
= I) (i)
∃z
1
Sum(x, y, z
1
) ↔ (j)
∃a∃w∃z
2
(Unit(a) ∧ Sum(˜ x, ˜ y, w) ∧ Sum(w, a, ˜ z
2
))
Proof. The proofs are elementary.
In addition to the theorems of PSIZE listed in 4.3.3, we require a battery of
tedious facts about the model Z
A
. We shall state these in terms of the model
´
Z
A
, an expansion of both / and Z
A
, which interprets two additional operators,
as follows:
Definition 4.3.4. Given a model / of PSIZE,
´
Z
A
is the expansion of Z
A
induced by the definitions in 4.3.2 together with (a) and (b):
´
Z
A
y = −x iff Z
A
x +y = 0 (a)
´
Z
A
z = x −y iff Z
A
z = x +−y (b)
Remark. Given a model / of PSIZE: for each x in
˙
Z
A
, there is a unique
y such that:
Z
A
x +y = 0
Proof.
a. If x = 0, then Z
A
x +y = 0 iff y = 0.

If Z
A
0 +y = 0
then / Sum(∅, y, ∅) by 4.3.2e
But / Sum(∅, ∅, ∅) by IDENT
So Z
A
y = 0 by UNIQ

56 4. THE PURE THEORY OF CLASS SIZES
⇐ Obvious
b. If x > 0, there is a unique y which satisfies
Sum(˜ x, a, y), where Unit (a) (*)
since there is a unique unit and / UNIQ
+
.
But Z
A
x +y = 0 iff / (*).

If Z
A
x +y = 0
then / Sum(˜ x, ˜ y, w)
and / Sum(w, a,
˜
∅) since / Sum(x, y, 0)
But /
˜
∅ = I
So / w = ˜ a by COMP and UNIQ

So / Sum(˜ x, ˜ y, ˜ a)
So / Sum(˜ x, a, y) by 4.3.3d and h2

If / (*)
then / Sum(˜ x, ˜ y, ˜ a) by 4.3.3h2
and / Sum(˜ a, a,
˜
∅) by COMP
So Z
A
x +y = 0 by 4.3.2e
Fact 4.3.5.
´
Z
A
satisfies the following:
−(x +y) = −x +−y (a)
−(x −y) = y −x (b)
(x +y) −y = x (c)
−(−x) = x (d)
x ,= 0 ∧ x < y → −y < −x (e)
Proof. Omitted.
Fact 4.3.6.
´
Z
A
satisfies the following:
a. Sm(a, b) ↔ Sum(a, b, a +b)
b. c ≤ b ↔ Sum(c, b −c, b)
c. Sum(a, b, c) ∧ 0 < b → Sum(a, −c, −b)
Sum(a, b, c) ∧ 0 < a → Sum(−c, b, −a)
4.3. PCA IS MODEL COMPLETE 57
d. Sm(a, b) → (Sm(−a, −b) ∨ b = −a)
Proof.
a. (⇒)
Suppose Sm(a, b)
So Sum(a, b, w) for some w
So (a +b) = w by 4.3.2e
So Sum(a, b, a +b)
(⇐)Obvious.
b. (⇒)
Suppose c ≤ b
then Sum(c, w, b) for some w by EXIST

So b = c +w by (a) and UNIQ
+
So b −c = (c +w) −c
So b −c = w by 4.3.5c
So Sum(c, b −c, b)
(⇐)
Suppose Sum(c, b −c, b)
But Sum(c, 0, c) by IDENT
So 0 ≤ (b −c) ↔ b ≤ c by MONOT and UNIQ
+
So b ≤ c by MIN
c.
Suppose Sum(a, b, c)
and 0 < b
then c = (a +b) by (a) and UNIQ
+
So c −b = (a +b) −b
So c −b = a by 4.3.5c
So −b = a −c by 4.3.5c
So Sum(a, −c, −b) if Sm(a, −c) by (a)
But if a = 0
then Sm(a, −c) by IDENT
And if a ,= 0
then a < c, since 0 < b by MONOT and UNIQ
+
58 4. THE PURE THEORY OF CLASS SIZES
So −c < −a by 4.3.5e
So −c ≤ ˜ a by 4.3.4
But Sm(a, ˜ a) by IDENT
So Sm(a, −c) by EXIST
+
d.
Suppose Sm(a, b)
But b = (a +b) −a, by 4.3.5c
So Sum(a, (a +b) −a, a +b), by (a)
So (a ≤ a +b), by (b)
So (a +b) < a, by TRICH
But if (a +b) ,= 0
then −a < −(a +b), by 4.3.5e
So −a < −a +−b, by 4.3.5a
So Sum(−a, (−a +−b) −(−a), −a +−b), by (b)
But (−a +−b) −(−a) = −b, by 4.3.5c
So Sum(−a, −b, −a +−b)
So Sm(−a, −b)
And if (a +b) = 0
then b = −a
We can now show that Z
A
Zgm. The only real difficulty arises in verifying
that addition is associative. We need the following lemma.
Lemma 4.3.7. Z
A
(a +c) + (b −c) = a +b
Proof. We shall work through successively more general cases:
a. When Sm(a, b) and c ≤ b
We know Sum(b −c, c, b) by 4.3.6b
and Sum(b, a, a +b) by 4.3.6a
So ∃w(Sum(c, a, w)
∧ Sum(b −c, w, a +b) by ASSOC2
So w = c +a by 4.3.6a and UNIQ
+
and a +b = (b −c) + (c +a) by 4.3.6a and UNIQ
+
So a +b = (a +c) + (b −c)
4.3. PCA IS MODEL COMPLETE 59
b. When Sm(a, b) and Sm(a, c)
If c ≤ b, Case (a) applies directly
So assume b < c
then a +c = (a +b) + (c −b) by Case (a)
So (a +c) −(c −b) = a +b by 4.3.5c
So (a +c) +−(c −b) = a +b by 4.3.4b
So (a +b) + (b −c) = a +b by 4.3.5b
c. Sm(a, b)
If Sm(a, c) Case (b) applies
So assume Sm(a, c) and thus b < c by EXIST
+
So Sum(b, c −b, c) by 4.3.6b
and 0 < c −b by MONOT
So Sum(b, −c, −(c −b)) by 4.3.6c
So Sum(b, −c, b −c) by 4.3.5b
So Sum(−c, b, b −c) (c1)
Either Sm(−a, −c) or c = −a since Sm(a, c) by 4.3.6d
Assume Sm(−a, −c)
then Sum(−a, −c, −a +−c) by 4.3.6a
So Sum(−a, −c, −(a +c)) by 4.3.5a
and 0 < −a
So Sum(−(−(a +c)), −c, −(−a)) by 4.3.6c
So Sum(a +c, −c, a) by 4.3.5d (c2)
Assume c = −a
then a +c = 0
and a = −c
Hence Sum(a +c, −c, a) by (c2)
and Sum(−c, b, b −c) by (c1)
and Sum(a, b, a +b) since Sm(a, b)
So Sum(a +c, b −c, a +b) by ASSOC
So (a +c) + (b −c) = a +b by 4.3.6a
d. Whenever
Suppose Sm(a, b), for otherwise case c applies.
60 4. THE PURE THEORY OF CLASS SIZES
Either Sm(−a, −b) or b = −a by 4.3.6d
Assume b = −a
then a +b = 0
and b −c = (−a) −c
= −a +−c
= −(a +c)
So (a +b) + (b −c)
= (a +c) +−(a +c)
= 0 = a +b
Assume Sm(−a, −b)
then −a +−b = (−a +−c) + (−b −−c) by case c
So −(a +b) = −(a +c) +−(b −c) by 4.3.5a
So −(a +b) = −((a +c) + (b −c)) by 4.3.5a
So a +b = (a +c) + (b −c) by 4.3.5d
Theorem 4.3.8. Z
A
Zgm
Proof. The only axiom for abelian groups that needs further verification is
associativity. We prove this using lemma 4.3.7:
x + (y +z) = (x +y) + ((y +z) −y), by 4.3.7
= (x +y) +z, by 4.3.5c
The ordering axioms of Zgm are satisfied in Z
A
because Z
A
uses the same
ordering as /, / PSIZE, and PSIZE includes the same ordering axioms.
Z
A
satisfies axiom (10) of Zgm because /satisfies the UNIT axiom of PSIZE.
The divisibility of elements in Z
A
required by axiom (11) of Zgm is guaran-
teed by the fact that / satisfies the divisibility principles of PCA.
Similarly, Z
A
satisfies axioms (12), and (14) of Zgm because / satisfies MIN
and MAX.
Axiom (13) of Zgm is:
y ≤ z ∧ x ≤ x +z → x +y ≤ x +z
If Z
A
x ≤ x +z
then / Sum(x, z, x +z)
Since Z
A
y < z
then / Sum(x, y, x +y) by EXIST
+
and / x +y ≤ x +z by MONOT
So Z
A
x +y ≤ x +z
4.4. REMARKS ON SHOWING THAT PCA AXIOMATIZES PCS 61
So, we PCA is τ-reducible to Zgm. The reduction is uniform since each
model / of PCA has the same domain as its Zgm-model. So, we may conclude
that PCA is model complete.
Theorem 4.3.9.
a. PCA is model complete.
b. PCA satisfies Monk’s condition.
Proof.
a. Apply Theorem 4.2.9.
b. Immediate from (a) and 4.2.1(b).
4.4 Remarks on showing that PCA axiomatizes
PCS
To prove that PCA ≡ PCS, we could follow the method outlined at the end of
chapter 3 for showing that CA ≡ CS.
We already know that PCS ¬ PCA, so we need only prove (1), which follows
from (2) by 3.3.1b.
(1) PCA ¬ PCS
(2) Every completion of PCA is consistent with PCS.
But (2) is equivalent to the conjunction of (2a) and (2b).
(2a) Every finite completion of PCA is consistent with PCS.
(2b) Every infinite completion of PCA is consistent with PCS.
The finite completions of PCA are the theories PCA; EXACTLY
n
. But
PCA; EXACTLY
n
is true in o
A
, where / is the standard finite interpretation
of L
C<
containing n atoms. So, the finite completions of PCA are consistent
with PCS. (Formally, we would have to define EXACTLY
n
in terms of units
rather than atoms.)
Letting PCAI = PCA+INF and PCSI = PCS +INF, (2b) is a consequence
of (3a) and (3b).
(3a) If PCAI
f
is consistent, then PCSI
f
is consistent.
(3b) If T is a completion of PCAI, then T ≡ PCAI
f
, for some f.
62 4. THE PURE THEORY OF CLASS SIZES
To prove (3a), we would have to prove analogues of 3.3.9 through 3.3.13 for
PCA and PCS. This seems straightforward, but tedious. The trick is to show
that enough axioms about size have been incorporated in PCA to establish the
entailments among MOD statements.
To prove (3b), it is sufficient to demonstrate (4), because every completion
of PCAI entails PCAI
f
for some total f.
(4) If f is total, then PCAI
f
is complete.
But PCA is model complete, so only (5) remains to be shown.
(5) If f is total and congruous, then PCAI
f
is has a prime model.
We will not construct prime models for the extensions of PCAI. It’s apparent
that the size models of the prime models for CAI
f
would do nicely. Alternatively,
the construction could be duplicated in this simpler case.
5
Completeness of CA
5.1 Model completeness of CA
We shall show that CA is model complete by showing that it satisfies Monk’s
criterion (see 4.3.1). So, given Assumption 5.1.1, we want to prove 5.1.2.
Assumption 5.1.1. / CA, B CA, / ⊆ B , and ( is a finitely generated
substructure of B.
Theorem 5.1.2. There is an isomorphic embedding, f : ( → /, where
f(x) = x if x ∈
˙
( ∩
˙
/
The Monk mappings for PCA can serve as a guide in constructing Monk
mappings for CA. The existence of Monk mappings for PCA tells us that we
can find elements with the right sizes. DISJ

and REP
<
then allow us to find
elements with those sizes that fit together in the right way.
Strictly speaking, o
A
is not a submodel of o
B
, so we cannot apply Monk’s
Theorem directly. But, let
o
B
(X) = the submodel of o
B
whose domain is ¦ σ
B
[x] [ x ∈ X¦
Then, clearly,
o
A
· o
B
(A) ⊆ o
B
, and
o
C
· o
B
(C), a finitely generated submodel of o
B
.
Monk’s Theorem applies directly to o
B
(A), o
B
, and o
B
(C), so we may con-
clude 5.1.3:
Fact 5.1.3. There is an isomorphic embedding, g : o
C
→ o
A
, where
g(σ
B
[x]) = σ
A
[x] for all x ∈
˙
( ∩
˙
/.
63
64 5. COMPLETENESS OF CA
That is to say, the sizes of elements in ( can be embedded in the sizes of
elements in /. It remains to be shown that the elements of ( themselves can be
mapped into / by a function which preserves boolean relations as well as size
relations.
( is a finite, and hence atomic, boolean algebra, though its atoms need not
be atoms of B; indeed, if B is infinite, there must be some atoms of ( which are
not atoms of B since the union of all atoms of ( is the basis of B.
Definition 5.1.4. d is a molecule iff d ∈
˙
( ∩
˙
/ and no proper subset of d
is in
˙
( ∩
˙
/.
˙
( ∩
˙
/ is a boolean algebra whose basis is the same as the basis of (. So
every atom of ( is included in some molecule. The embedding f has to map
each molecule to itself. Moreover, f has to be determined by its values on (
since f must preserve unions. In fact, the atoms of ( can be partitioned among
the molecules. So, if
d = b
1
∪ . . . ∪ b
n
where d is a molecule and b
1
, . . . , b
n
are the atoms of ( contained in d, then d
must also be the (disjoint) union of f(b
1
), . . . , f(b
n
). If this condition is satisfied,
f will preserve boolean relations. f must also select images with appropriate
sizes.
Proof. Proof of 5.1.2 Given a molecule, d, let b
1
, . . . , b
n
be the atoms of C
contained in d. For each b
i
, let c
i
be some member of g(σ
B
[b
i
]) These elements
will be elements of
˙
/, since g yields sizes in / whose members are in
˙
/. These
elements have the right sizes, as we will show below, but they are in the wrong
places. We have no guarantee that they are contained in the molecule d. So we
still need to show that there are disjoint elements of
˙
/, a
1
, . . . , a
n
whose union
is d and whose sizes are the same as those of b
1
, . . . , b
n
, respectively. Well,
Suppose d = b
1
∪ . . . ∪ b
n
So ( Sum(b
1
, . . . , b
n
, d)
So o
C
Sum(σ
B
[b
1
], . . . , σ
B
[b
n
], σ
B
[d])
So o
A
Sum(g(σ
B
[b
1
]), . . . , g(σ
B
[b
n
]), g(σ
B
[d]))
So o
A
Sum(c
1
, . . . , c
n
, d)
since each c
i
∈ g(σ
B
[b
i
]) and d ∈ g(σ
B
[d]) = σ
A
[d]. So, the existence of
a
1
, . . . , a
n
, as above, is guaranteed because / DEF
+
.
Now, let f(b
i
) = a
i
for each b
i
in the molecule d. Repeating this procedure
for each molecule yields a value of f for each atom of (. Finally, if x ∈
˙
( is
non-atomic, then
x = b
1
∪ . . . ∪ b
k
where each b
k
is atomic. So let
f(x) = f(b
1
) ∪ . . . ∪ f(b
k
)
5.1. MODEL COMPLETENESS OF CA 65
f satisfies the requirements of Theorem 5.1.2: Boolean relations are preserved
by f because the function is determined by its values on the atoms of (; f
maps elements of C ∩ A into themselves because the set of atoms contained in
each of these molecules is mapped into a disjoint collection of elements of
˙
/
whose union is the same molecule; so, we need only show that f preserves size
relations. To do so, we invoke lemma 5.1.6, below:
( x < y iff / f(x) < f(y) (a)
( x ∼ y iff / f(x) ∼ f(y) (b)
( Sum(x, y, z) iff / Sum(f(x), f(y), f(z)) (c)
( Unit(x) iff / Unit(y) (d)
Proof of (a): (The others are similar).
( x < y iff o
B
σ
B
[x] < σ
B
[y]
iff o
A
g(σ
B
[x]) < g(σ
B
[y])
iff o
A
σ
A
[f(x)] < σ
A
[f(y)] by 5.1.6
iff / f(x) < f(y)
So, we may conclude:
Theorem 5.1.5. CA is model complete.
Lemma 5.1.6. For all x in
˙
(,
σ
A
[f(x)] = g(σ
B
[x])
Proof. Suppose
x = b
1
∪ ∪ b
n
where each b
i
is an atom of
˙
(.
So B Sum(b
1
, . . . , b
n
, x) since B DISJ

So o
B
Sum(σ
B
[b
1
], . . . , σ
B
[b
n
], σ
B
[x])
So o
A
Sum(g(σ
B
[b
1
]), . . . , g(σ
B
[b
n
]), g(σ
B
[x])) since o
A
⊆ o
B
But g(σ
B
[b]) = σ
A
[f(b)] by the choice of f(b).
So o
A
Sum(σ
A
[f(b
1
)], . . . , σ
A
[f(b
n
)], g(σ
B
[x]))
But / Sum(f(b
1
), . . . , f(b
n
), f(x)) since / DISJ

So o
A
Sum(σ
A
[f(b
1
)], . . . , σ
A
[f(b
n
)], σ
A
[f(x)])
So g(σ
B
[x]) = σ
A
[f(x)] since o
A
UNIQ
+
66 5. COMPLETENESS OF CA
5.2 Prime models for CA
For each total, congruous remainder function, f, we want to find a prime model,
Q
f
, for CAI
f
. All of these prime models can be defined over the class, Q, of
sets near quasi-congruence classes (see section 3.2). For different remainder
functions, we need to assign different size relations over Q. Section 5.2.1 defines
the structures Q
f
and verifies that each satisfies the respective theory, CAI
f
.
Section 5.2.2 defines, for each model of CAI
f
, a submodel, or shell. Section
5.2.3 shows that Q
f
is isomorphic to the shell of any model of CAI
f
.
5.2.1 The standard protomodels
The construction here is a more elaborate version of the construction of Q in
chapter 3 (see 3.2.19). Q turns out to be Q
f
, where f(n) = 0 for all n. As in
the case of Q, the models Q
f
and their copies in arbitrary models of CAI are
unions of chains.
The sizes assigned to elements of Q to induce Q
f
for a total, congruous
remainder function, f, are more elaborate than those used in the definition of
Q (see 3.2.22), but they are employed in substantially the same way:
Definition 5.2.1.
a. A size is an ordered pair ¸ρ, δ), where both ρ and δ are rational.
b. If θ
1
= ¸ρ
1
, δ
1
) and θ
2
= ¸ρ
2
, δ
2
) are sizes, then
(i) θ
1
< θ
2
iff ρ
1
< ρ
2
or ρ
1
= ρ
2
and δ
1
< δ
2
.
(ii) θ
1

2
= ¸ρ
1

2
, δ
1

2
)
(cf 3.2.22)
To assign sizes for Q
f
we rely on the representation of sets in Q defined in
3.2.24. Q
f
, unlike Q, assigns different sizes to the n-congruence classes for a
given n.
Definition 5.2.2. If f is total and congruous, then
a. If x is an n-congruence class, x = [nk +i], then
θ
f
(x) =
_
¸
1
n
,
n−f(n)
n
), if i < f(n),
¸
1
n
,
−f(n)
n
), if f(n) ≤ i.
b. If x ∈ QC
n
, so that x is the disjoint union of n-congruence classes,
x
1
∪ . . . ∪ x
k
then
θ
f
(x) = θ
f
(x
1
) + +θ
f
(x
k
)
5.2. PRIME MODELS FOR CA 67
c. If x is finite, then
θ
f
(x) = ¸0, [x[)
d. If x ∈ Q, so that x can be represented as
(C(x) ∪ ∆
1
(x)) −∆
2
(x)
as in 3.2.24, then
θ
f
(x) = θ
f
(C(x)) +θ
f
(∆
1
(x)) −θ
f
(∆
2
(x))
= θ
f
(C(x)) +¸0, [∆
1
(x)[) −¸0, [∆
2
(x)[)
= θ
f
(C(x)) +¸0, [∆
1
(x)[ −[∆
2
(x)[)
Intuitively, all n-congruence classes are assigned sizes ¸1/n, δ), but δ is no
longer 0 in all cases, as in Q. Instead, the first f(n) n-congruence classes are
each one atom larger that the remaining (n −f(n)) n-congruence classes.
The desired models of CAI may now be defined:
Definition 5.2.3. If f is total and congruous, then the standard proto-
model for f is Q
f
, where
˙
Q
{
= Q
Q
f
x < y iff θ
f
(x) < θ
f
(y)
Q
f
x ∼ y iff θ
f
(x) = θ
f
(y)
Q
f
Unit(x) iff θ
f
(x) = ¸0, 1)
Q
f
Sum(x, y, z) iff θ
f
(z) = θ
f
(x) +θ
f
(y)
(cf 3.2.27)
To verify that the structure Q
f
is a model of CAI
f
, for total and congruous
f, we exhibit each such model as the union of a chain of models.
Definition 5.2.4. If f is total and congruous, then Q
f,n
is the submodel
of Q
f
whose domain is Q
n
.
Fact 5.2.5. If f is total and congruous and n > 0, then Q
f,n
BASIC.
Proof. The proof can be obtained from the proof of Theorem 3.2.29 by substi-
tuting:
Q
f,n
for Q
n
,
θ
f
(x) for θ(x),
ρ
f
(x) for ρ(x), and
δ
f
(x) for δ(x)
68 5. COMPLETENESS OF CA
The following notion is helpful in understanding our constructions.
Definition 5.2.6. Suppose / BASIC, x ∈
˙
/, and 0 ≤ m < n. Then an
(n.m)-partition of x in / is a sequence x
1
, . . . , x
n
, where
a. x = x
1
∪ . . . ∪ x
n
b. If 0 < i < j ≤ n, then x
i
∩ x
j
= ∅, and
c. The x
i
are approximately the same size:
(i) If 0 < i < j ≤ m, then x
i
∼ x
j
,
(ii) If m < i < j ≤ n, then x
i
∼ x
j
,
(iii) If 0 < i ≤ m < j ≤ n, then Sum(x
i
, a, x
j
), for any atom a,
In other words, x is partitioned among n pairwise disjoint sets which are
roughly the same size: each of the first m is one atom larger than each of
the remaining n −m. If 0 < i ≤ m, x
i
is called a charmed n-factor of x;
for i > m, x
i
is a common n-factor of x.
The sequence Q
f,n
does not constitute a chain of models. For example Q
f,3
is not an extension of Q
f,2
. But this sequence does harbor a chain of models:
Fact 5.2.7. If n > m, then Q
f,n!
⊆ Q
f,m!
.
Proof. It will be clearer, and easier, to establish this by example rather than by
formal proof. Letting n = 2 and m = 3, we want to show that the 2-congruence
classes have the same size relations in Q
f,6
as they do in Q
f,2
, for any f. The
other elements of Q
f,6
will then fall into place, since size relations are determined
by the representations of each set, x, as C(x), ∆
1
(x), and ∆
2
(x).
Suppose that f(2) = 0, so that
Q
f,2
[2n] ∼ [2n + 1]
Since f is congruous, f(6) ∈ ¦0, 2, 4¦. If f(6) = 0, then all of the 6-congruence
classes are common. If f(6) = 2, then [6n] and [6n + 1] are the only charmed 3-
congruence classes. If f(6) = 4, then all of the 3-congruence classes are charmed
except for [6n + 4] and [6n + 5].
In any case, [2n] will include the same number of charmed 3-congruence
classes as [2n + 1], so
Q
f,6
[2n] ∼ [2n + 1]
Suppose, however that f(2) = 1, so that
Q
f,2
[2n] is one atom larger than [2n + 1]
Here, f(6) ∈ ¦1, 3, 5¦, since f is congruous. In any case, [2n] contains exactly
one more charmed congruence class than [2n + 1], so
Q
f,6
[2n] is one atom larger than [2n + 1]
So it goes in general.
5.2. PRIME MODELS FOR CA 69
Fact 5.2.7 allows us to regard Q
f
as the union of a chain of models:
Fact 5.2.8. If f is total and congruous, then
Q
f
=
_
n>0
Q
f,n!
Fact 5.2.9. If f is total and congruous, then
a. Q
f
BASIC
b. Q
f
ADIV
k
, for k > 0
c. Q
f
MOD
n,f(n)
, for n > 0
d. Q
f
CAI
f
Proof.
a. BASIC is a universal-existential theory; so it is preserved under unions of
chains. (See Appendix, Fact B.5).
b. Each n-congruence class can be partitioned into k nk-congruence classes.
c. Q
f,n
MOD
n,f(n)
, by definition. Since MOD
n,f(n)
is an existential sen-
tence, it is preserved under extensions.
d. Immediate from (a), (b), and (c).
5.2.2 Shells of models
To embed the model Q
f
into an arbitrary model, /, of CAI
f
, we must find a
smallest submodel, B, of / which satisfies CAI. Clearly, the basis of /, call it
x
0
, must be included in B, since the symbol I must refer to the same set in the
submodel as it does in the model. But, if the basis of / is in
˙
B and B CAI,
then
˙
B must contain two disjoint sets of roughly the same size whose union is
the basis of /, x
0
. Pick such a pair, x
1
and x
2
, to include in
˙
B. Whether these
are exactly the same size or differ by an atom depends on whether / MOD
2,0
or / MOD
2,1
.
B must also satisfy ADIV
3
. We can aim for this by placing in
˙
B three disjoint
sets, x
11
, x
12
, and x
13
whose union is x
1
and another three disjoint sets, x
21
,
x
22
, and x
23
whose union is x
2
. The existence of such sets is assured because
/ ADIV
3
. Again, the exact size relations will be determined by which MOD
principles are satisfied in /. Insuring that x
1
and x
2
are each divisible by 3,
also guarantees that x
0
is divisible by 3: the three unions
x
11
∪ x
21
, x
12
∪ x
22
, and x
13
∪ x
23
will be roughly the same size and will exhaust x
0
.
70 5. COMPLETENESS OF CA
We can continue this process indefinitely, dividing each set introduced at
stage n into n + 1 roughly equal subsets at stage n + 1. This will produce an
infinite tree, bearing sets. The deeper a node is in this tree, the smaller the set
it bears and the greater the number of successors among which this set will be
partitioned.
This great tree of sets will not form a boolean algebra, nor will it be closed
under finite unions. A boolean algebra could be obtained by including both the
node sets and their finite unions, but this would still not be an atomic boolean
algebra, which is what we are looking for. We cannot correct for this problem
by including in
˙
B all atoms of /: / may have uncountably many atoms while B,
to be a prime model, must be countable. We leave the solution of this problem
to the formal construction.
The formal proof proceeds as follows: First, we define a tree, i.e. the set of
nodes on which we shall hang both the components of the successive partitions
described above and the atoms of the submodel being constructed. Second, we
present the construction which, given a model / of CAI
f
, assigns a node set
A
λ
and a node atom, a
λ
, to each node, λ. Third, we define the shell of / as
the submodel of / generated by the collection of node sets and node atoms. In
the next section, we show that the shell of / is isomorphic to Q
f
.
First, the tree:
Definition 5.2.10.
a. A node is a finite sequence ¸n
1
, . . . , n
k
), where k > 0 and for all i ≤ k,
n
i
< i. (The variables λ and κ range over nodes.)
b. If λ = ¸n
1
, . . . , n
k
), then
(i) The length, or depth, of λ, L(λ) is k.
(ii) If 1 ≤ i ≤ k, then λ(i) = n
i
.
(iii) λ • m = ¸n
1
, . . . , n
k
, m)
c. λ extends κ iff L(λ) > L(κ) and, for 1 ≤ i ≤ L(κ), λ(i) = κ(i).
d. λ 0-extends κ iff λ extends κ and, for L(κ) < i ≤ L(λ), λ(i) = 0.
Nodes constitute the vertices of an infinite tree in which ¸0) is the root and
λ dominates κ iff κ extends λ. The number of immediate descendants of a node
grows as the depth of the node increases.
We shall now assign a set to each node by repeatedly partitioning the basis
of /. At the same time we shall assign an atom to each node.
Definition 5.2.11. Given / = /
f
CAI
f
, where f is a total, congruous
remainder function:
a. Let A
0
be the basis of /, and let a
0
be any atom of /.
5.2. PRIME MODELS FOR CA 71
b. Suppose that A
λ
and a
λ
have been chosen. Let m = L(λ) and let
k =
f((m+ 1)!) −f(m!)
m!
Since f is congruous, k is an integer (see Definition 3.3.7e). Let
A
λ•0
, . . . , A
λ•m
be an (m+1, k) partition of A
λ
if A
λ
is common or an (m+1, k +1)
partition of A
λ
if A
λ
is charmed. Fact 5.2.12 guarantees that such
partitions exist. In either case, choose A
λ•0
so that it contains a
λ
.
This is always possible because A
λ
contains a
λ
.
c. Let a
λ•0
= a
λ
. If 0 < i ≤ m, let a
λ•i
be any atomic subset of A
λ•i
.
The sets A
λ
will be referred to as node sets and the atoms a
λ
as node
atoms.
Fact 5.2.12. Suppose / CAI
f
, n > 0, m > 0, and
k =
f(nm) −f(n)
n
Then
a. x has an (m, k)-partition iff / Mod
m,k
(x)
b. Every common n-factor of
˙
/ has an (m, k)-partition.
c. Every charmed n-factor of
˙
/ has an (m, k + 1)-partition.
Proof.
a. Mod
m,k
(x) says that x has an (m, k)-partition.
b. All common n-factors are the same size and satisfy the same Mod
m,k
predicate. So, suppose that each common n-factor has k charmed m-
factors and m−k common m-factors.
By (a), / has f(n) charmed n-factors and n − f(n) common n-factors.
Partitioning each of the n-factors into m subsets of roughly the same size
yields an nm-partition of /; the charmed m-factors of the n-factors are
the charmed nm factors of / and the common m-factors of the n-factors
are the common nm factors of /.
Each of the common n factors has k charmed m-factors and each of the
charmed n-factors has k + 1 charmed m-factors. In all, there are
(n −f(n))k +f(n)(k + 1)
=nk +f(n)
72 5. COMPLETENESS OF CA
charmed m-factors among the n-factors of /.
But, by (a) again, there are f(nm) charmed nm-factors of /. So
f(nm) = nk +f(n)
and k = (f(nm) −f(n))/n
c. Immediate from (b), since every charmed n-factor is one atom larger than
each common n-factor.
The facts listed in 5.2.13 should clarify this constuction. All of them can be
established by induction on the depth of nodes.
Fact 5.2.13.
a. A
λ
⊂ A
κ
iff κ extends λ or κ = λ.
b. If A
λ
= A
κ
, then κ = λ.
c. There are n! nodes (and, hence, node sets) of depth n.
d. If i ,= j, then A
λ•i
∩ A
λ•j
= ∅.
e. Any two node sets of the same depth are disjoint.
f. Each node set is the disjoint union of its immediate descendants.
g. Each node set is the disjoint union of all of its descendants at any
given depth.
h. a
λ
⊂ A
κ
iff λ extends κ.
i. a
λ
= a
κ
iff one of λ and κ 0-extends the other.
j. Every node set contains infinitely many node atoms.
k. For any n, the node sets of depth n form an (n!, f(n!))-partition of
the basis of /.
Proof. We demonstrate (k) by induction on n.: If n = 1, then n! = 1, f(n!) = 0,
and A
λ
0
is the basis of /. So (k) holds because any set is a (1,0)-partition of
itself.
Assume that (k) holds for n. Then (k) also holds for n+1: each node set of
depth n has n + 1 immediate descendants; so, there are (n!)(n + 1) = (n + 1)!
node sets of depth n + 1.
5.2. PRIME MODELS FOR CA 73
Furthermore, f(n!) of the n-factors are charmed and n! −f(n!) are common.
By Fact 5.2.12, each charmed n-factor has an (n + 1, k + 1)-partition and each
common n-factor has an (n + 1, k)-partition, where
k =
f(n!(n + 1)) −f(n!)
n!
=
f((n + 1)!) −f(n!)
n!
(substituting n! for n and n + 1 for m).
So, there are
(k + 1)f(n!) charmed n + 1 factors from the charmed n-factors
and k(n! −f(n!)) charmed n + 1 factors from the common n-factors
In all, then , the number of charmed n-factors is:
(k + 1)f(n!) +k(n! −f(n!))
= kf(n!) +f(n!) +n!k −f(n!)k
= f(n!) +n!k
= f(n!) +f((n!)(n + 1)) −f(n!)
= f(n!)(n + 1))
= f((n + 1)!)
So, the n+1 factors of the n-factors of the basis form an ((n+1)!, f((n+1)!))-
partition of the basis. That is to say, (k) holds for n + 1.
Given a collection of node sets, A
λ
, and node-atoms, a
λ
, from a model, /,
of CAI
f
, we can now construct a submodel, B, of / which is isomorphic to Q
f
.
Definition 5.2.14. Suppose f is total and congruous, that / CAI
f
and
A
λ
and a
λ
are the node sets and node-atoms of / produced by construction
5.1.12. Then, the shell of / is the submodel of / generated by ¦A
λ
, a
λ
¦.
For the remainder of this chapter, we will regard as fixed:
a. f, a total, congruous remainder function
b. /, a model of CAI
f
,
c. ¦A
λ
, a
λ
¦, a set of node sets and node-atoms produced by the construction
above.
d.
ˆ
/, the shell of / generated from ¦A
λ
, a
λ
¦
74 5. COMPLETENESS OF CA
To show that
ˆ
/ · Q
f
, we need a sharper characterization of the elements
of
ˆ
/. Recall from 3.2.24 that each member, x, of Q has a unique representation
as
(C(x) ∪ ∆
1
(x)) −∆
2
(x)
where C(x) is a quasi-congruence class, ∆
1
(x) and ∆
2
(x) are finite sets and

1
(x) ∩ C(x) = ∅

1
(x) ∩ ∆
2
(x) = ∅

2
(x) ⊆ C(x)
We can obtain a similar representation for elements of
ˆ
/: the node sets play
the role of (some of) the congruence classes; finite unions of node sets correspond
to the quasi-congruence classes; finite sets of node-atoms correspond to the finite
subsets in Q.
Definition 5.2.15.
a. x is a quasi-nodal set of / iff it is the union of finitely many node
sets (iff it is the union of finitely many node sets at a given depth).
b. x is an /-finite set iff it is a finite set of node atoms.
c. If x ∈ A and y ∈ A, then x is /-near y iff both x − y and y − x are
/-finite sets,
(cf. 3.2.14-3.2.18)
Still following in the footsteps of chapter 3, we can characterize
˙
ˆ
A as the
collection of sets /-near quasi-nodal sets. Analogues of 3.2.15 through 3.2.18
obtain for /-nearness.
Fact 5.2.16.
a. x ∈
˙
ˆ
A iff x is /-near some quasi-nodal set of /.
b.
ˆ
/ is an atomic boolean algebra whose atoms are the node-atoms a
λ
.
c. If x ∈
˙
ˆ
A, then x has a unique representation as
(C(x) ∪ ∆
1
(x)) −∆
2
(x)
where C(x) is a quasi-nodal set disjoint from ∆
1
(x) and including

2
(x), both of which are /-finite sets.
Proof.
a. (⇒)
ˆ
/ is generated from node sets and node atoms via the boolean oper-
ations, each of which preserves /-nearness to quasi-nodal sets.
(⇐)
ˆ
/ must contain finite unions of node sets as well as /-finite sets; so it
must also contain sets obtained by adding or removing /-finite sets from
quasi-nodal sets.
5.2. PRIME MODELS FOR CA 75
b. The proof parallels that of Theorem 3.2.21 exactly.
c. Let C(x) be the quasi-nodal set which is /-near x (see 3.2.23); Let

1
(x) = x −C(x); and, let ∆
2
(x) = C(x) −x.
5.2.3 The embeddings
To embed Q
f
into /, we first describe Q in terms of node sets and node atoms.
In effect, we are performing the construction 5.2.12 on Q
f
, but with two dif-
ferences: first, we are stipulating which (n, m)-partitions to use at each level;
second, we are selecting node atoms so that every singleton in Q is the node
atom for some node. This latter condition guarantees that the shell of Q
f
will
be Q
f
itself.
Definition 5.2.17. Suppose λ is a node. Then
a. If L(λ) = k, then
Q
λ
= [k!n +m]
where
m =
k

i=1
λ(i)(i −1)!
b. The depth of Q
λ
is L(λ).
Examples.
Q
0
= [n]
Q
0,0
= [2n]
Q
0,1
= [2n + 1]
Q
0,1,0
= [6n + 1]
Q
0,0,2
= [6n + 4]
Q
0,1,2
= [6n + 5]
Definition 5.2.18.
a. ι(λ) = the least n ∈ Q
λ
.
b. q
λ
= ¦ι(λ)¦
Fact 5.2.19.
a. If λ = ¸n
1
, . . . , n
k
), then
ι(λ) =
k

i=1
(i −1)!n
i
76 5. COMPLETENESS OF CA
b. ι(λ) = ι(κ) iff λ = κ or one 0-extends the other.
c. For every n, there’s a λ such that n = ι(λ).
d. For every n, there are infinitely many λ such that n = ι(λ).
e. For every n, there’s a λ such that n = ι(κ) iff κ = λ or κ 0-extends
λ.
Fact 5.2.20.
a. At each depth, n, the ι(λ) take on all and only values less than n!.
b. If ι(λ) = k, then all nodes along the left-most branch descending from
λ also have the value k. These are the only nodes below λ with the
value k.
c. Every natural number is the value of all and only those nodes along
the left-most branch descending from some node.
Though for a given natural number n, there will be infinitely many nodes λ
for which ι(λ) = n, we can associate with each natural number a shortest (i.e.
shallowest) node for which ι(λ) = n.
Definition 5.2.21. λ
n
= the shortest λ such that ι(λ) = n.
Fact 5.2.22.
a. ι(λ
n
) = n
b. λ extends λ
ι(λ)
c. λ
ι(λn)
= λ
n
d. λ
ι(λ)
= λ iff λ =< 0 > or λ(L(λ)) ,= 0 (ie a node, λ, will be the highest
node with a certain value just in case λ is not the leftmost immediate
descendant of its parent.)
Each of the points listed in Fact 5.2.13 hold for the sets Q
λ
and q
λ
. That
is to say, the Q
λ
can be regarded as node sets and the q
λ
as node atoms
for any model Q
f
. Notice, especially, that 5.2.13k holds.
We may, finally, define the embedding of Q
f
into /:
Fact 5.2.23. If x ∈ Q, there is a unique y ∈
˙
ˆ
A such that
∀λ(q
λ
⊆ x ↔ a
λ
⊆ y)
Proof. There is at most one such y, by Fact 5.2.16. To show that there is such
a y, suppose first that x ∈ QC.
Then there is some n such that
x = x
1
∪ ∪ x
k
5.2. PRIME MODELS FOR CA 77
where each x
i
is an (n!)-congruence class and hence a node set in Q
f
. So
x = Q
λ1
∪ ∪ Q
λ
k
Now, let
y = A
λ1
∪ ∪ A
λ
k
So y ∈
˙
ˆ
A and:
q
λ
⊂ x iff q
λ
⊂ Q
λ
i
, for some i
iff λ extends λ
i
by 5.2..13h
iff a
λ
⊂ A
λi
iff a
λ
⊂ y.
If x is not a quasi-congruence class, then
x = (x

∪ ∆
1
(x)) −∆
2
(x)
where x

is a quasi-congruence class. Let y

be the element of
˙
ˆ
A corresponding
to x

, as described above, and let
y = (y

∪ ¦ a
λ
[ q
λ
⊆ ∆
1
(x) ¦) −¦ a
λ
[ q
λ
⊆ ∆
2
(x) ¦
Definition 5.2.24. The nodal embedding of Q into
˙
/ is defined by letting
g(x ∈ Q) be the y ∈
˙
ˆ
A such that
∀λ(a
λ
⊆ y ↔ q
λ
⊆ x)
Fact 5.2.25.
a. If x ∈ Q, then
g(x) = (g(C(x)) ∪ g(∆
1
(x))) −g(∆
2
(x))
b. g is one-one.
c. g maps Q onto
˙
ˆ
A.
Proof.
a. Immediate from the proof of 5.2.23.
b. Suppose g(x) = y = g(x

). Then
∀λ(q
λ
⊆ x ↔ a
λ
⊆ y ↔ q
λ
⊆ x

)
But every integer is ι(λ) for some λ, so x = x

.
78 5. COMPLETENESS OF CA
c. Obvious.
Theorem 5.2.26. The nodal embedding, g, of Q into / is an isomorphism
of Q
f
onto
ˆ
/.
Proof.
a.
Q
f
x ⊆ y iff ∀λ(q
λ
⊂ x → q
λ
⊂ y)
iff ∀λ(a
λ
⊆ g(x) → a
λ
⊆ g(y)) (by 5.2.24)
iff
ˆ
/ g(x) ⊆ g(y) (by 5.2.16)
b. As in (a), g can be shown to preserve ∅, I, unions, intersections, relative
complements, and proper subsets.
c. Let nc(z, n) be the number of charmed node sets of level n contained in
C(z). And let n, below, be the least k such that x and y are both unions
of node sets of depth k.
Q
f
x ∼ y iff θ
f
(x) = θ
f
(y)
iff nc(x, n) −([∆
1
(x)[ −[∆
2
(x)[)
= nc(y, n) −([∆
1
(y)[ −[∆
2
(y)[)
iff nc(g(x), n) −([g(∆
1
(x))[ −[g(∆
2
(x))[)
= nc(g(y), n) −([g(∆
1
(y))[ −[g(∆
2
(y))[)
since Q
λ
is charmed iff A
λ
is charmed
and g preserves boolean relations
iff / g(x) ∼ g(y)
iff
ˆ
/ g(x) ∼ g(y) since
ˆ
/ ⊆ /
d.
Q
f
x < y iff Q
f
x ∼ x

⊂ y
for some x

∈ Q since Q
f
REP
<
iff / g(x) ∼ g(x

) ⊂ g(y) by (b) and (c)
iff / g(x) < g(y) since / SUBSET, INDISC

iff
ˆ
/ g(x) < g(y) since
ˆ
/ ⊆ /
5.3. SUMMARY 79
e.
Q
f
Sum(x, y, z) iff Q
f
(x

∪ y

= z
∧ x ∼ x

∧ y ∼ y

∧ x

∩ y

= ∅) since Q
f
DEF
+
iff
ˆ
/ (g(x

) ∪ g(y

) = g(z)
∧ g(x) ∼ g(x

)
∧ g(y) ∼ g(y

)
∧ g(x

) ∩ g(y

) = ∅) by (b) and (c)
iff / (g(x

) ∪ g(y

) = g(z)
∧ g(x) ∼ g(x

)
∧ g(y) ∼ g(y

)
∧ g(x

) ∩ g(y

) = ∅)
iff / Sum(g(x), g(y), g(z)) since / DEF
+
iff
ˆ
/ Sum(g(x), g(y), g(z)) since
ˆ
/ ⊆ /
So, each model of CAI
f
has a submodel isomorphic to Q
f
, and we may
conclude:
Corollary 5.2.27. If f is total and congruous, then CAI
f
has a prime
model.
5.3 Summary
We can now draw our final conclusions about CA, CS, and their completions.
Theorem 5.3.1. If f is a total, congruous remainder function, then CAI
f
is consistent and complete.
Proof. CAI
f
is consistent because f is congruous, by 3.3.12c. Since CAI
f
is
model complete, by 5.1.6, and has a prime model, by 5.2.27, the prime model
test (Appendix, Fact C.3) applies. So CAI
f
is complete.
Corollary 5.3.2. It T is a completion of CAI, then T ≡ CAI
f
, for some
congruous f.
Proof. For each n > 0, T ¬ MOD
n,i
for exactly one i, 0 ≤ i < n:
Since T is complete, T ¬ MOD
n,i
or T ¬ MOD
n,i
for each such i. But if
T ¬ (MOD
n,0
∧ . . . ∧ MOD
n,n−1
), then T is inconsistent, since T ¬ CAI and
CAI ¬ DIV
n
. Hence T ¬ MOD
n,i
for at least one one i, 0 ≤ i < n.
But suppose T ¬ MOD
n,i
and T ¬ MOD
n,j
where 0 ≤ i ,= j < n. Again, T
would be inconsistent, for CA ¬ MOD
n,i
→ MOD
n,j
(see Lemma 3.3.11).
80 5. COMPLETENESS OF CA
So, let f(n) = m iff T ¬ MOD
n,m
. Then T ¬ CAI
f
and, since CAI
f
is
complete, CAI
f
¬ T.
Corollary 5.3.3.
a. Every completion of CA is consistent with CS.
b. CA ≡ CS
Proof.
a. Follows from 5.3.2 and 3.3.13.
b. Follows from (a) and 3.3.2b, given that CS ¬ CA.
Theorem 5.3.4.
a. For n > 0, CS; EXACTLY
n
is decidable.
b. CS is decidable.
Proof.
a. CS; EXACTLY
n
¬ φ iff /
n
φ. But /
n
is a finite model.
b. To determine whether CS ¬ φ, alternate between generating theorems of
CA and testing whether /
n
φ.
Corollary 5.3.5.
a. CSI has 2
ω
completions.
b. For total f, CSI
f
is decidable iff f is decidable.
Proof.
a. There are 2
ω
remainder functions whose domain is the set of prime num-
bers. Each such function is congruous, so each corresponds to a consistent
extension of CSI. By Lindenbaum’s lemma, each of these extensions has
a consistent and complete extension.
b. (⇒)If f is decidable, then CAI
f
is recursively enumerable. But CAI
f
is
complete, so it is decidable.
(⇐)To calculate f(n), determine which MOD
n,m
sentence is in CSI
f
.
Theorem 5.3.6. There is no sentence, φ, such that T = CA; φ is consistent
and T has only infinite models.
5.3. SUMMARY 81
Proof. If φ is true only in infinite models of CA, then φ is true in all finite
models of CA, so φ ∈ CS. But CA ≡ CS, so CA; φ is inconsistent.
Theorem 5.3.7. CS is not finitely axiomatizable.
Proof. Suppose CS ¬ φ. So CA ¬ φ and, by compactness, (BASIC ∪ T) ¬ φ
for some finite set of ADIV
n
principles, T = ¦ ADIV
n
[ n ∈ J ¦. Let K =
¦ n [ every prime factor of n is a member of J ¦.
Let / be a model with domain

k∈K
Q
k
, in which size relations are deter-
mined in accordance with the size function, θ, defined in 3.2.26. We claim the
following without proof:
/ BASIC (1)
/ ADIV
j
for all j ∈ J (2)
/ ADIV
k
(3)
By (1) and (2), / (BASIC ∪ T), so / φ. But by (3), / CS. Hence
φ CS.
82 5. COMPLETENESS OF CA
6
Sets of Natural Numbers
CS has standard finite models since it consists of sentences true in all such
models. It has infinite models Q
n
and Q
f
. In this chapter we will show that
CS has infinite standard models over T(N).
An ordering of T(N) that satisfies CS will not necessarily appear reasonable.
For example, some such orderings say that there are fewer even numbers than
prime numbers (see below 6.2.13). To rule out such anomalies, we introduce
a principle, OUTPACING, in section 1. OUTPACING mentions the natural
ordering of N and applies only to subsets of N. Section 2 establishes that
OUTPACING can be satisfied jointly with any consistent extension of CS in
a model whose domain is T(N). So CS ; OUTPACING does not fix the size
relations over T(N).
6.1 The outpacing principle
Throughout this chapter, x and y will range over T(N).
Definition 6.1.1. x outpaces y just in case the restriction of x to any
sufficiently large initial segment of N is larger than the corresponding
restriction of y, that is, iff:
∃n∀m(m > n → [x
[m]
[ > [y
[m]
[)
Notice that the size comparison between the two restricted sets will always
agree with the comparison of their normal cardinalities since all initial segments
of N are finite.
We employ this notion to state a sufficient condition for one set of natural
numbers to be larger than another:
OUTPACING. If x outpaces y, then x > y.
The general motivation behind this principle should be familiar. We extrapolate
from well understood finite cases to puzzling infinite cases. But we should also
83
84 6. SETS OF NATURAL NUMBERS
emphasize, again, that this extrapolation cannot be done in any straightfor-
ward, mechanical way without risking contradiction. We cannot, for example,
strengthen the conditional to a biconditional, thus:
(1) x > y iff x outpaces y.
This revised principle conflicts with CS, for outpacing is not a quasi-linear
ordering. For example, neither [2n] nor [2n + 1] outpaces the other since each
initial segment ¦0, . . . , 2n + 1¦ of N contains n evens and n odds. But the two
are discernible under outpacing, since [2n] outpaces [2n + 2] while [2n + 1] does
not.
There is another point that underlines the need for care in extrapolating
from finite cases to infinite cases: we cannot just use (2):
(2) If, given any finite subset z of N, x restricted to z is larger
than y restricted to z, then x > y.
Though (2) is true, its antecedent is only satisfied when y ⊂ x.
So, there are many statements that assert of infinite cases what is true of
finite cases. Some of these conflict with one another. Others are too weak to
be helpful. It is doubtful whether there is any mechanical way to decide which
of these statements are true. The best we can do is propose plausible theories,
determine whether they are consistent, and see how far they go.
Definition 6.1.0. / is an outpacing model iff
˙
/ = T(N)
/ BASIC, and
/ OUTPACING
There is a slight difficulty in saying that an interpretation of L
C<
satisfies
OUTPACING . Since OUTPACING involves the < relation over N, it cannot
be expressed in L
C<
. We shall finesse this problem by regarding OUTPACING
as the (very large) set of sentences
¦ b < a [ a outpaces b ¦
Fact 6.1.1. Every outpacing model satisfies the following
a. [2n] > [3n]
b. [3n] > [4n] > [5n] > . . .
c. If k > 0, then [kn] >
_
n
2
¸
Proof.
a. [2n] has at least (k−1)/2 members less than or equal to k for any given k.
[3n] has at most k/3+1 such members. If k > 4, then (k −1)/2 > k/3+1
6.1. THE OUTPACING PRINCIPLE 85
b. Similar to (1).
c. If m = k
2
, both [kn] and
_
n
2
¸
have exactly k members less than m. For
m > k(k + 1), N
[m]
will have more members in [kn] than in
_
n
2
¸
.
Fact 6.1.2. Every outpacing model satisfies the following
a. [2n] ≥ [2n + 1]
b. [2n + 1] ≥ [2n + 2]
c. If [2n] > [2n + 1], then [2n + 1] ∼ [2n + 2].
Proof.
a. By TRICH, it is sufficient to show that [2n] < [2n + 1]. If it were, then,
by REP
<
, there is a y such that y ∼ [2n] and y ⊂ [2n + 1]. But any
proper subset of [2n + 1] is outpaced by, and hence smaller than [2n]. Let
k be the least odd number not in y. Then [2n] leads y at k+1 and y never
catches up. So there is no y such that [2n] ∼ y ⊂ [2n + 1].
b. Similar to that of (1).
c. Note that [2n] = [2n + 2] ∪ ¦0¦ and that BASIC (*).
(*) (y ⊂ x ∧ z ≤ y ∧ Atom(z

) ∧ z ⊂ x ∧ x = z ∪ z

) → y ∼ z
Let x = [2n], y = [2n + 1], z = [2n + 2] and apply (*).
The two alternatives left open in 6.1.2 correspond to the possibilities that N
may be odd or even: if [2n] ∼ [2n + 1], then N is even, if [2n + 1] ∼ [2n + 2],
then N is odd. In section 6.2, we show that both of these possibilities can be
realized in standard models over T(N). Here, we generalize 6.1.2 to similar
cases, including other congruence classes.
Definition 6.1.3. ¸x, y) is an alternating pair iff x and y are infinite and
for all i > 0, x
i
< y
i
< x
i+1
.
Fact 6.1.4. If ¸x, y) is an alternating pair, then in any outpacing model:
x ∼ y ∨ x > y ∼ (x −x
1
)
Proof. The argument for 6.1.2 applies here since the only facts about [2n] and
[2n + 1] used hold by virtue of these sets forming an alternating pair.
Theorem 6.1.5. For a given k > 0, let A
i
= [kn +i] for each i < k. Then:
a. If 0 ≤ i < j < k, then ¸A
i
, A
j
) is an alternating pair.
86 6. SETS OF NATURAL NUMBERS
b. There is a p, 0 < p ≤ k such that
(i) If i < j < p, then A
i
∼ A
j
.
(ii) If p ,= k, then A
0
∼ A
p
∪ ¦0¦.
(iii) If p ≤ i < k, then A
i
∼ A
p
.
(See example below).
Proof.
a. A
i
(n) = kn +i,A
j
(n) = kn +j, A
i
(n +1) = kn +i +k and i < j < k +i.
b. If A
0
> A
i
for some i, let p be the least such i; otherwise, let p = k.
(i) For 0 ≤ i < p, ¸A
0
, A
i
) is an alternating pair. So either A
0
> A
i
or
A
0
∼ A
i
by 6.1.4. But A
0
≤ A
i
by the selection of p, so A
0
∼ A
i
. (i)
follows by TRANS

.
(ii) Immediate from 6.1.4 since ¸A
0
, A
p
) is an alternating pair and A
0
>
A
p
.
(iii) A
0
> A
p
≥ A
i
if i ≥ p. So A
0
> A
i
. Hence A
i
∼ A
0
−¦0¦ ∼ A
p
. So
A
i
∼ A
p
.
Example. Let k = 4, so A
i
= [4n +i] for i = 0, 1, 2, 3. Then one of the
following situations obtains:
A
0
∼ A
1
∼ A
2
∼ A
3
> [4n + 4]
A
0
> A
1
∼ A
2
∼ A
3
∼ [4n + 4]
A
0
∼ A
1
> A
2
∼ A
3
∼ [4n + 4]
A
0
∼ A
1
∼ A
2
> A
3
∼ [4n + 4]
6.2 Models of CS and Outpacing
In this section, we construct models of CS over T(N) that satisfy OUTPACING.
Outpacing models will be constructed out of finite models of CS by a tech-
nique which is very much like the ultraproduct construction common in model
theory, though the application here demands some important differences.
Definition 6.2.1.
a. L
N
, the language of subsets of N, is the first order language which
results from adding to L
C<
, as individual constants, a name for each
subset of N.
b. /
n
is the standard finite interpretation of L
N
over T(N
[n]
) in which:
/
n
(a) = a ∩ N
[n]
= a
[n]
for each a ⊆ N.
6.2. MODELS OF CS AND OUTPACING 87
Definition 6.2.2. (cf. [Monk, Def. 18.15, p318] If X is a set and F ⊆
T(X), then
a. F has the finite intersection property iff the intersection of any
finite subset of F is non-empty.
b. F is a filter over X iff
(i) F ,= ∅
(ii) If a ∈ F and a ⊆ b, then b ∈ F, and
(iii) If a ∈ F and b ∈ F, then a ∩ b ∈ F
c. F is an ultrafilter over X iff
(i) F is a filter over X
(ii) X ∈ F, and
(iii) if Y ⊆ X, then either Y ∈ F or (X −Y ) ∈ F.
d. An ultrafilter, F, over X is principal iff there is some x ∈ F such
that F = ¦ a ⊆ X [ x ∈ a ¦
Fact 6.2.3.
a. A non-principal ultrafilter contains no finite sets. [Bell, Ch.6, lemma
1.3, p.108]
b. A non-principal ultrafilter over X contains all cofinite subsets of X.
c. If F ⊆ T(X) and F has the finite intersection property, then there is
an ultrafilter over X which includes F. [Monk, Prop. 18.18, p.319]
d. If Y ⊆ X and Y is infinite, then there is a non-principal ultrafilter
over X which contains Y .
Definition 6.2.4. If F is an ultrafilter over N, then /
F
is the interpreta-
tion of L
C<
in which
a.
˙
/
F
= T(N),
b. /
F
a < b iff
_
k [ a
[k]
< b
[k]
_
∈ F, and similarly for other predicates.
c. Boolean symbols receive the usual interpretation.
Our main result is that if F is non-principal, then /
F
is an outpacing model.
Theorem 6.2.5. If F is a non-principal ultrafilter over N, then
a. If τ is a term of L
N
, then
/
k
(τ) = /
F
(τ) ∩ A
k
= /
F
(τ)
[k]
88 6. SETS OF NATURAL NUMBERS
b. If φ is a quantifier free formula of L
N
, then
/
F
φ iff ¦ k [ /
k
φ¦ ∈ F
c. If φ is a universal formula of L
N
and /
k
φ for every k, then
/
F
φ.
d. /
F
REP
<
.
e. /
F
BASIC.
f. /
F
ADIV
n
, for every n.
Proof.
a. By induction on the structure of τ:
(i) If τ is a constant, τ = a for some a ⊂ N. So A
k
(τ) = a
[n]
by 6.2.1b.
(ii) If τ = τ
1
∪ τ
2
,
/
k
(τ) = /
k

1
) ∪ /
k

2
)
= (/
F

1
) ∩ A
k
) ∪ (/
F

2
) ∩ A
k
)
= (/
F

1
) ∪ /
F

2
)) ∩ A
k
= /
F

1
∪ τ
2
) ∩ A
k
The proofs for intersections and relative complements are similar.
b. By induction on the structure of φ:
(i) If φ = a ⊂ b, then /
F
φ iff a ⊂ b
If a ⊂ b, then there is a k ∈ b but not ∈ a. So if n > k, a
[n]
⊂ b
[n]
.
Hence, ¦ n [ /
F
φ¦ is cofinite and, by 4.2.3b, in F.
Conversely, if
_
n [ a
[n]
⊂ b
[n]
_
∈ F, then it is infinite. So, there
cannot be a k in a but not in b; otherwise a
[n]
would not be included
in b
[n]
for any n greater than k. So a ⊆ b. But, clearly a ,= b, so
a ⊂ b.
(ii) If φ = a = b, then
/
F
φ iff a = b
iff a
[n]
= b
[n]
for all n.
iff ¦ n [ /
n
a = b ¦ = N
iff ¦ n [ /
n
a = b ¦ ∈ F
since if /
n
a = b and k > n, then /
k
a = b.
(iii) /
F
a < b iff ¦ k [ A
k
(a < b) ¦ ∈ F. Immediate from 6.2.4b.
6.2. MODELS OF CS AND OUTPACING 89
(iv) If φ is non-atomic, then, since F is an ultrafilter:
/
F
φ
1
∧ φ
2
iff /
F
φ
1
and /
F
φ
2
iff ¦ k [ /
k
φ
1
¦ ∈ F and ¦ k [ /
k
φ
2
¦ ∈ F
iff ¦ k [ /
k
φ
1
∧ φ
2
¦ ∈ F
/
F
φ iff /
F
φ
iff ¦ k [ /
k
φ¦ / ∈ F
iff ¦ k [ /
k
φ¦ ∈ F
c.
Suppose /
k
∀xφ(x), for all k
then /
k
φ(a), for all a, for all k
So /
k
φ(a), for all k, for all a
So /
F
φ(a), for all a by (b)
So /
F
∀xφ(x)
d. Suppose /
F
(a < b). Construct a

, a subset of b, for which /
F
(a ∼ a

)
as follows:
Let K =
_
k [ a
[k]
< b
[k]
_
so, K ∈ F.
= ¦k
1
, . . . , k
i
, . . .¦ where the k
i
’s are in strictly increasing order.
Let a
0
= ∅
a
i+1
= a
i

_
the n greatest members of b
[ki+1]
not in a
i
_
,
where n = [a
[ki+1]
[ −[a
i
[
a

=
_
a
i
Then a

⊂ b since each a
i
draws its new members from b.
Claim: If k ∈ K, then a

[k]
∼ a
[k]
.
Hence: /
F
a

∼ a since they are the same size over some set which
contains K and is, thus, if F.
e. Immediate from (c) and (d): BASIC is equivalent to a set of universal
sentences, together with ATOM and REP
<
. (/
F
ATOM because it
contains all singletons of natural numbers.)
f. For 0 < i ≤ n, x an infinite subset of N, let x
i
= ¦ x
kn+i−1
[ k ∈ N¦.
The n sets, x
i
, partition x. Furthermore, these form an alternating n-
tuple, in the manner of the congruence classes modulo n (see Theorem
6.1.5). As in 6.1.5, these sets are approximately equal in size and /
F

ADIV
n
(x).
90 6. SETS OF NATURAL NUMBERS
Corollary 6.2.6. If F is a non-principal ultrafilter over N, then
a. /
F
CAI, and
b. /
F
CSI
Proof.
a. Immediate from 6.2.5e and 6.2.5f.
b. Immediate from (a) and 5.3.3b.
The proof of 6.2.6 is modeled on the usual ultraproduct construction, but
is not quite the same. In the usual construction(see, for example, [Bell, pp.
87-92]), a model is built by first taking the product of all factors (in this case,
the /
k
), which results in a domain whose elements are functions from the index
set (N, here) to elements of the factors. These functions are then gathered
into equivalence classes (by virtue of agreeing almost everywhere, i.e. on some
member of the filter) and the reduced ultraproduct is defined by interpreting
the language over these equivalence classes. The model so constructed, which
we’ll call

/
i
/F, has the handy property that it satisfies any formula which is
satisfied by almost all factors, and certainly any formula which is satisfied in all
the factors. This is handy because, given that each of the /
k
satisfies CS, we
can immediately conclude that

/
i
/F satisfies CS.
Unfortunately,

/
i
/F is not the model we wanted: its elements are not
subsets of N, but equivalence classes of functions from N to finite subsets of
N. There is, indeed, a natural mapping from subsets of N to such elements,
and this mapping would have allowed us to identify a model over T(N) as a
submodel of

/
i
/F; but only a submodel. So, had we constructed the reduced
ultraproduct, we would have then been able to infer that the part of that model
which held our interest satisfied all universal formulas of CS; we still would
have had to resort to special means to show that the non-universal formulas
were likewise satisfied.
Fortunately, these special means were available; the only non-universal ax-
ioms of CAI could be verified in the constructed model more or less directly, and
the completeness proof of the last chapter allowed us to infer that all formulas
true in all of the factors are true in the model /
F
, after all.
For a given F, there will be many cases where /
F
a < b even though
b does not outpace a. This will happen whenever
_
k [ a
[k]
< b
[k]
_
∈ F but is
not cofinite. Consider a familiar example: Let a = [2n + 1] and b = [2n]. Then
a
[k]
< b
[k]
iff k ∈ [2n], for if we count the even numbers and the odd numbers
up to some even number, there will always be one more even and if we count up
to some odd number, there will always be the same number of evens and odds.
[2n] is neither finite nor cofinite, so it may or may not be in F, If [2n] ∈ F,
/
F
[2n + 1] < [2n]. Otherwise, [2n + 1] ∈ F, so /
F
[2n + 1] ∼ [2n].
6.2. MODELS OF CS AND OUTPACING 91
The construction of a model /
F
from any non-principal ultrafilter, F, sug-
gests that there are many outpacing models unless different ultrafilters can yield
the same model. We will first show that this qualification is not needed.
Theorem 6.2.7. If F
1
and F
2
are distinct non-principal ultrafilters over
N, then /
F1
,= /
F2
.
Proof. (See below.)
To show this, we will show that the presence of a set in an ultrafilter makes a
direct, personalized contribution to the model /
F
. Putting this in another way,
there is a set of decisions that must be made in constructing an outpacing model;
each decision may go either way, though the decisions are not independent of
each other. Furthermore, each decision is made for a model /
F
by the presence
or absence of a particular set in F.
Definition 6.2.8. If x ⊆ N, then x
+
= ¦ i + 1 [ i ∈ x¦.
A pair of sets, ¸x, x
+
), can sometimes be an alternating pair, but this is not
always the case.
Fact 6.2.9. ¸x, x
+
) is an alternating pair iff x is infinite and there is no
n ∈ x such that (n+1) ∈ x. That is, no two consecutive numbers are in x.
Nevertheless, pairs ¸x, x
+
) are like alternating pairs in the following way:
Lemma 6.2.10. If x is infinite, x1 = (x − x
+
), x2 = (x
+
− x), F is a
non-principal ultrafilter, and / = /
F
, then
a. ¸x1, x2) is an alternating pair.
b. / x ∼ x
+
or / x > x
+
∼ x −¦x
1
¦
c. / x > x
+
iff x ∈ F.
Proof.
a. Let a run of x be a maximal consecutive subset of x. (So [2n] has only 1-
membered runs, while N−[10n + 1] has only 9-membered runs.) So, x
1
(n)
is the first element in the n-th run of x and x
2
(n) is the first element after
the n-th run of x.
b. We know from (a) that x
2
∼ x
1
or x
2
∼ x
1
−x
1
(1). But the disjoint union
of x ∩ x
+
with x
1
or x
2
, respectively, yields x or x
+
. So (b) follows from
DISJ

.
c. We need only prove (*):
[x
[n]
[ > [x
+
[n]
[ iff n ∈ x (*)
Informally: x
[n]
first becomes greater than x
+
[n]
for n = x(1), since x(1) / ∈
x
+
because (x
1
−x
1
(1) ⊂ x). Throughout the first run of x, x maintains
its lead, losing it only at the least n for which n / ∈ x (since (n−1) ∈ x, so
n ∈ x
+
). This pattern repeats during successive runs of x.
92 6. SETS OF NATURAL NUMBERS
We can now prove our main result, Theorem 6.2.7:
Proof. Without loss of generality, we can suppose there is a set, x, such that
x ∈ F
1
and x / ∈ F
2
. By 6.2.10c, /
F1
x > x
+
and /
F2
x ∼ x
+
Theorem 6.2.7 allows us to improve upon some previous results. For example,
we can show that either of the alternatives in 6.2.10b can obtained for any
alternating pair.
Theorem 6.2.11.
a. If neither x nor y outpaces the other, then there is an ultrafilter F,
such that /
F
(x ∼ y).
b. If ¸x, y) is an alternating pair, then there is an ultrafilter F, such
that /
F
(x > y).
Proof.
a. Let J =
_
k [ [x
[k]
[ = [y
[k]
[
_
. Since neither x nor y outpaces the other, J
is infinite. By 6.2.3d, let F be a non-principal ultrafilter which contains
J. Then /
F
x ∼ y.
b. If ¸x, y) is an alternating pair, so is ¸y, x −x(1)). So, by (a) there is an F
such that /
F
y ∼ (x −x(1). But then /
F
x > y.
Theorem 6.2.12. Every infinite completion of CS has an outpacing model.
Proof. Recall that every infinite completion of CS is equivalent to CSI
f
for
some total and congruous remainder function, f. (See 3.6.2.)
Given such an f, let G
k
= [kn +f(k) + 1] for each k > 0. Then (*) holds
for each k:
(*) G
k
=
_
n [ /
n
MOD
k,f(k)
_
Let G = ¦ G
k
[ k > 0 ¦.
The intersection of any finite subset, H, of G is infinite. If H is a finite
subset of G, then H = ¦ G
k
[ k ∈ J ¦, where J is some finite subset of N
+
. So
H = ¦ n + 1 [ k ∈ J → n ≡ f(k) mod k ¦. But f is congruous, so the restriction
of f to the finite domain J has infinitely many solutions (see 3.3.8a).
Since the intersection of any finite subset of G is infinite, there is a non-
principal ultrafilter F such that G ⊆ F. By (*), /
F
MOD
k,f(k)
, so /
F

CSI
f
.
6.2. MODELS OF CS AND OUTPACING 93
On the basis of 6.2.11 we noted that there are even and odd outpacing
models; we can now extend that observation to moduli other than 2. More
specifically, all of the possibilities listed in 6.1.5 for the relative sizes of the
k-congruence classes are obtainable in outpacing models.
This section has explored the existence and variety of outpacing models.
Three comments are in order before we turn to the common structure of out-
pacing models.
First, even it T is an infinite completion of CS, there is no unique outpacing
model which satisfies T. This would be true only if fixing the congruence classes
determined whether x ∼ x
+
or x > x
+
for every x ⊆ N. That all such choices
are not determined by a remainder theory can be seen intuitively, perhaps, by
considering x =
_
n
2
¸
: any finite set of congruences has infinitely many solutions
that are squares and infinitely many that are not; so whether x ∈ F is an
independent choice. Also, there are only 2
ω
remainder functions while there are
2
2
ω
non-principal ultrafilters over N[Bell, Ch. 6, Theorem 1.5], each yielding a
different outpacing model.
Second, it is not clear whether every outpacing model can be obtained by the
construction of 6.2.4. Lemma 6.2.10c may suggest that any outpacing model,
/, is /
F
for F
A
= ¦ x [ / x > x
+
¦, but it should not. To establish that / =
/
F
, both (1) and (2) are necessary.
(1) If / is an outpacing model, then F
A
is a non-principal ultra-
filter.
(2) If F
A
= F
B
, then / = B
I have not been able to prove (1) or (2). If (1) is false, then clearly / ,= /
F
.
But even if (1) is true, two outpacing models may agree about all pairs ¸x, x
+
)
but disagree elsewhere. At most one of them is obtainable by our construction.
So (1) and (2) are open problems.
Finally, though it may be extraneous to show that OUTPACING is inde-
pendent, we will do so.
Theorem 6.2.13. There are standard models of CS over T(N) which do
not satisfy OUTPACING.
Proof. Suppose that is a linear ordering under which N forms an ω-sequence.
Then we could define x -outpaces y as follows:
∃n∀m[n m → [ ¦ k [ k ∈ x ∧ k m¦ [ > [ ¦ k [ k ∈ x ∧ k n¦ []
and define the principle:
OUTPACING . If x -outpaces y, then x > y.
Modifying 6.2.4, we could produce standard models of CS over T(N) which
satisfy OUTPACING and these will not, in general, satisfy OUTPACING.
94 6. SETS OF NATURAL NUMBERS
Suppose, for example, that is the ordering:
p
1
, q
1
, . . . , p
k
, q
k
, . . .
where p is the set of prime numbers and q is its complement. In any model
of OUTPACING , p and q will be nearly the same size, as the evens and the
odds are in normal outpacing models. But the evens are much smaller than q,
so p > [2n] and OUTPACING is false in OUTPACING models.
6.3 Size and density
When number theorists talk about the sizes of sets of natural numbers, they do
not content themselves with speaking of the (Cantorian) cardinalities of these
sets. Since they often want to compare infinite subsets of N, they need a more
discriminating notion.
One notion they use is asymptotic density. The asymptotic density of a
set, x, of natural numbers is the limit, if there is one, of [x
[n]
[ as n grows. For
example, the asymptotic density of [2n] is 1/2. From now on we shall use the
term ‘density’ for asymptotic density.
In this section, we compare the ordering of T(N) given by density to the
orderings given by CS and OUTPACING.
Definition 6.3.1.
a. ˆ x
i
=
|x
[i]
|
i
, the fraction of numbers less than or equal to i that are
members of x.
b. If x ⊆ y ,= ∅, then ρ(x, y), the density of x in y, is the limit, if it
exists of
ω
lim
i=y1
ˆ x
i
/ˆ y
i
That is,
ρ(x, y) = r iff ∀(δ > 0)∃n(∀i > n)(r −δ <
ˆ x
i
ˆ y
i
< r +δ)))
c. The density of x, ρ(x), is the density of of x in N, if x has a density
in N.
d. If x ⊆ y ,= ∅, then x converges in y, cvg(x, y), iff x has a density in
y; otherwise, x diverges in y.
e. x converges, cvg(x), iff x converges in N. x diverges iff x diverges
in N.
Fact 6.3.2.
a. If x is finite, ρ(x) = 0.
6.3. SIZE AND DENSITY 95
b. If x is cofinite, ρ(x) = 1.
c.
ρ([2n]) = 1/2
ρ([4n] , [2n]) = 1/2
ρ([4n]) = 1/4
d. If cvg(x, y) and cvg(y, z), then cvg(x, z) and ρ(x, z) = ρ(x, y)ρ(y, z).
Fact 6.3.3.
a. There are divergent sets.
b. If 0 ≤ r ≤ 1, there is a set with density r.
c. If 0 ≤ r ≤ 1 and y is infinite, there is a set with density r in y.
Proof.
a. Let x =
_
i [ ∃n(10
2n
≤ i < 10
2n+1
_
. So x contains all numbers between
0 and 9, between 100 and 999, between 10,000 and 99,999, and so forth.
If n > 1, then ˆ x
10
2n
≤ .1 and ˆ x
10
2n+1
≥ .9. So ˆ x
k
cannot have a limit.
b. Suppose r is given. Construct the set x as follows:
x
0
= ∅
x
i+1
=
_
x
i
, if ˆ x
i
≥ r;
x
i
; i + 1, otherwise.
x =
_
x
i
c. Modify the construction for (b) in the obvious ways.
Theorem 6.3.4. Suppose that both x and y converge in z. Then, if
ρ(x, z) < ρ(y, z), y outpaces x.
96 6. SETS OF NATURAL NUMBERS
Proof.
Let b = (ρ(y, z) −ρ(x, z))/3
Let n
1
= the least n such that for i > n, −b < ˆ x
i
/ˆ z
i
−ρ(x, z) < b
and let n
2
= the least n such that for i > n, −b < ˆ y
i
/ˆ z
i
−ρ(y, z) < b
So for any i > n
1
+n
2
,
we have ˆ x
i
/ˆ z
i
< ρ(x, z) +b
and ˆ y
i
/ˆ z
i
> ρ(y, z) −b
But ρ(x, z) +b < ρ(y, z) −b,
So ˆ x
i
/ˆ z
i
< ˆ y
i
/ˆ z
i
So ˆ x
i
< ˆ y
i
So [x
[i]
[ < [y
[i]
[
We can use the relation between density and outpacing to draw conclusions
about densities that have nothing to do with outpacing, as in Theorem 6.3.5.
Theorem 6.3.5. If ρ(x, z
1
) < ρ(y, z
1
), and both x and y converge in z
2
,
then ρ(x, z
2
) ≤ ρ(y, z
2
)
Proof. Since ρ(x, z
1
) < ρ(y, z
1
), y outpaces x. But if ρ(x, z
2
) > ρ(y, z
2
), then x
outpaces y. So, ρ(x, z
2
) ≤ ρ(y, z
2
).
We cannot strengthen the consequent of 6.3.5 to say that ρ(x, z
2
) < ρ(y, z
2
):
Let z
1
be the set of primes, let x contain every third member of z
1
, and let y
be z
1
−x. Then ρ(x, z
1
) = 1/3 and ρ(y, z
1
) = 2/3, but ρ(x, N) = ρ(y, N) = 0.
Theorem 6.3.4 implies that in any outpacing model, sets with distinct den-
sities will have distinct sizes. Even if two sets have the same density, they will
differ in size if they have the same density in some common set. So, from Facts
6.3.b and (c), we can begin to appreciate how precise an ordering outpacing
models provide:
Fact 6.3.6. If / is an outpacing model, then
a. there are uncountably many sizes of sets in /, and
b. if 0 ≤ r ≤ 1, then even among sets with density r, there are uncount-
ably many sizes in /.
Proof.
a. Immediate from Fact 6.3.3b and Theorem 6.3.4.
b. Let x be an infinite set with density r, let y be an infinite subset of x,
where ρ(y, x) = 0, and let z = x−y. There are uncountably many subsets
of y with distinct densities in y, though ρ(y
1
, x) = 0, for any y
1
⊂ y.
6.3. SIZE AND DENSITY 97
Suppose now that y
1
⊂ y, y
2
⊂ y, and ρ(y
1
) < ρ(y
2
). Then / y
1
< y
2
,
by 6.3.4. so / z ∪ y
1
< z ∪ y
2
, by DISJ

.
But ρ(z ∪ y
1
) = ρ(z ∪ y
2
) = ρ(x), since z was obtained by removing from
x a set with density 0 relative to x.
6.3.1 The Convexity Problem
It’s tempting to infer from these results that the extremely fine ordering of sets
by size (or, rather, any such ordering which is realized in an outpacing model)
is both a refinement and a completion of the ordering suggested by density: a
refinement because it preserves all differences in size which are captured by the
notion of density, a completion because all sets are located in a single, linear
ordering of sizes.
But the situation is really not so clear. It is evident that the size ordering
over T(N) in any outpacing is a refinement of the ordering by cardinality: if
x has a smaller cardinal number than y, then x is smaller than y, though two
sets with the same cardinal number may have different sizes. We can regard
the cardinality of a set in T(N) as determined by its size, though different sizes
may yield the same cardinal number. We shall express this fact by saying that,
at least when we focus on T(N), cardinality is a function of size. (It is not at
all clear that this is true in any power set.)
Now, we want to know whether the density of a set is a function of its
size. It turns out that a negative answer is compatible with our theory (CS
and OUTPACING), while an affirmative answer may or may not be consistent.
First, we will give a more precise formulation of this problem; second, we will
show that a negative answer is consistent; finally, we’ll discuss the consistency of
an affirmative answer. In passing, we’ll explain why this is called the convexity
problem.
Consider (1) and (1a):
(1) If x ∼ y and ρ(x) = r, then ρ(y) = r.
(1a) If x ∼ y and ρ(x) = r and cvg(y), then ρ(y) = r.
(1a) is an immediate consequence of Theorem 6.3.4. For if y converges, there is
some r
2
= ρ(y); if r < r
2
, then x < y and if r > r
2
, then x > y. But x ∼ y, so
r
2
= r.
So, the questionable part of (1) can be expressed as (2):
(2) If x ∼ y and x converges, then y converges.
Theorem 6.3.4 insures that (1) just in case (2). Recalling that sets may have the
same density even though they differ in size, we may consider two additional
formulations:
(3) If ρ(x) = ρ(y) and x < z < y, then ρ(z) = ρ(x)
98 6. SETS OF NATURAL NUMBERS
(4) If ρ(x) = ρ(y) and x < z < y, then z converges.
(3) and (4) are equivalent for the same reasons that (1) and (2) are equivalent.
Fact 6.3.7. An outpacing model satisfies (1) iff it satisfies (3).
Proof.
⇒ Suppose that x < y < z and ρ(x) = ρ(z).
By REP
<
, there are two sets, x

and y

, for which
x

⊂ y

⊂ z, x ∼ x

, y ∼ y

.
Since x ∼ x

and ρ(x) = ρ(z)
then ρ(x

) = ρ(z) by (1)
So ρ(z −x

) = 0
and ρ(z −y

) = 0
But y

= z −(z −y

)
So ρ(y

) = ρ(z)
So ρ(y) = ρ(z) by (1)
⇐ Suppose that x ∼ y and ρ(x) = r.
If x = ∅ or x = N, then y = x, so ρ(y) = ρ(x)
To apply (3), we need to find two sets, x
1
and x
2
such that
ρ(x
1
) = ρ(x
2
) = r = ρ(x)
and
x
1
< y < x
2
Assuming that x ,= ∅ and x ,= N, let
x
1
= x −a, for some a ∈ x
and
x
2
= x ∪ b, for some b / ∈ x.
Then x
1
< x < x
2
, by SUBSET, and x
1
< y < x
2
, by INDISC

.
Since adding or removing a single element has no effect on the density of
a set,
ρ(x
1
) = ρ(x) = ρ(x
2
)
So, by (3), ρ(y) = r.
6.3. SIZE AND DENSITY 99
The question at hand is called the ‘convexity problem’ because of the for-
mulation in (3). In geometry, a figure is convex if, given any two points in the
figure, any point between them (ie on the line segment from one to the other) is
also in the figure. Applying this notion in the obvious way to the size ordering,
(3) says that the class of sets having a given density is convex. By theorems
6.3.4 and 6.3.7, (1), (2) and (4) say the same thing.
We regret that the only thing we know about (3) is that it may be false:
Theorem 6.3.8. The negation of (1) is satisfied in some outpacing model.
Proof. Let x be a set with density r and let y be a divergent set which neither
outpaces nor is outpaced by x. Let K be
_
n [ x
[n]
= y
[n]
_
K is infinite, so K is a member of some non-principal ultrafilter, F. /
F
x ∼ y,
so (1) is false in /
F
.
Open Problem: Is (1) consistent with CS and OUTPACING?
100 6. SETS OF NATURAL NUMBERS
Appendix
A Notation
A.1 Predicate Logic
The formal theories discussed in this thesis are theories with standard formal-
ization, in the sense of [Tarski, p.5]. That is, they are formalized in first order
predicate logic with identity and function symbols. The following notation is
used for the predicate calculus:
∧ and
∨ or
not
→ if ... then
↔ if and only if
= identical
,= not identical
∃x there is an x
∀x for all x
Conjunctions and disjunctions of sets of sentences are represented by:

... φ ...
φ
and

... φ ...
φ
A first order language is determined by its non-logical constant symbols,
in the usual way. These may be predicates, individual constants, or function
symbols. In most cases, the ranks of the symbols will accord with their familiar
uses and we do not stipulate them.
A schematic function is a function whose range is a set of formulae. DIV
n
,
for example, maps natural numbers into sentences (see 3.2.8). The arguments
of such functions are indicated by subscripts.
101
102 APPENDIX
A.2 Set Theory
For first order languages with boolean operations and predicates, we use the
following notation:
I the universe
∅ the empty set
x ∪ y the union of x and y
x ∩ y the intersection of x and y
x −y the relative complement of y in x
x ⊆ y x is contained in y
x ⊂ y x is a proper subset of y
These symbols are also used for set theoretic relations, outside of first order
languages.
In addition, we use the following:
x ∈ y x is a member of y
T(x) the power set of x
¦ x [ φ(x) ¦ the set of x’s which are φ
x; a the union of x and a
x
[n]
the members of x less than or equal to n
N the set of natural numbers: 0, 1, ...
N
+
N −0
Z the set of integers
ω the smallest infinite cardinal
2
ω
2 to the ω
A.3 Arithmetic
For arithmetic, we use the following:
i +j i plus j
i −j i minus j
ij i times j
i
j
i to the j-th power
i! i factorial
i [ j i divides j
gcd(i, j) greatest common divisor of i and j
n ≡ m mod j n is congruent to m modulo j
B. MODEL THEORY 103
We use the following notation for sets of natural numbers:
[...n...] = ¦ k [ ∃n(k = ...n...) ¦
For example:
[kn +j] = the set of numbers congruent to j modulo k
_
n
2
¸
= the set of squares of natural numbers
B Model Theory
This section lists the model-theoretic notions and results assumed in the text
and presents our notation for these notions.
An interpretation, /, of a first order language, L, consists of a domain,
˙
/,
and a function which assigns to each individual constant of L a member of
˙
/,
to each n-place predicate of L, a set of n-tuples over
˙
/, and to each n-argument
function symbol of L, an n-ary function defined over and yielding values in
˙
/.
We assume the notion of satisfaction as defined in [Chang, section 1.3] and
use
/ φ to mean that / satisfies φ .
Definition B.1. Familiar notions about models.
a. / is a submodel of B (/ ⊆ B).
b. B is an extension of / (/ ⊆ B).
c. The submodel of B generated by X, where X is a subset of
˙
B.
d. / is isomorphic to B (/ · B).
e. f is an isomorphic embedding of / into B.
f. B is an elementary extension of /.
g. ¦/
i
¦ is a chain of models.
h. / =

¦/
i
¦; the union of a chain of models.
Definition B.2. Notions about theories:
a. A theory is a set of first-order sentences.
b. The language of a theory, L
T
.
c. T proves φ (T ¬ φ);
T proves T
2
(T ¬ T
2
).
d. T is complete.
e. T is consistent.
104 APPENDIX
f. T is categorical.
g. T is equivalent to T
2
(T ≡ T
2
).
Fact B.3. The following are well known facts:
a. If T ¬ φ, then some finite subset of T proves φ (compactness).
b. If T is complete and T; φ is consistent, then T ¬ φ.
c. If T is categorical, then T is complete.
Definition B.4. More familiar notions:
a. T is existential iff it is equivalent to a set of prenex sentences, none
of which have universal quantifiers.
b. T is universal iff it is equivalent to a set of prenex sentences, none
of which have existential quantifiers.
c. T is universal-existential iff it is equivalent to a set of prenex sen-
tences, each of which has all of its universal quantifiers preceding
any of its existential quantifiers.
d. φ is a primitive formula iff φ is an existential formula in prenex
form whose matrix is a conjunction of atomic formulas and nega-
tions of atomic formulas.
Fact B.5. Fairly familiar facts:
a. If T is existential, / T, and / ⊆ B, then B T.
b. If T is universal and / T, then any submodel of / also satisfies T.
c. If T is universal-existential and /
i
T, for all i > 0, then

¦/
i
¦ T.
C Model Completeness
This section presents the definition of model completeness and some basic
results needed in Chapters 4 and 5.
There are several ways to show that a theory is complete. We use only one,
which is based on A. Robinson’s notion of model completeness.
Definition C.1. T is model complete iff T is consistent and for any two
models, / and B of T, / ⊆ B iff / is an elementary submodel of B.
[Monk, p. 355].
A model complete theory is not necessarily complete, unless the theory has
a prime model.
Definition C.2. / is a prime model of T if / T and / can be embedded
in any model of T. [Monk, p.359]
C. MODEL COMPLETENESS 105
Fact C.3. If T is model complete and T has a prime model, then T is
complete.
To show that a theory is model complete we rely, directly or indirectly, on a
theorem of Monk’s (see 4.2.1a), which is based on some equivalent formulations
of model completeness:
Definition C.4.
a. The /-expansion of L is the result of adding to L a constant for
each element of
˙
/.
b. The diagram of / is the set of atomic sentences and negations of
atomic sentences of the /-expansion of the language of / which are
true in /.
Fact C.5. The following are equivalent [Monk, p. 356]:
a. T is model complete.
b. For every model, /, of T and every /-expansion L

of L, the T ∪
L

-diagram of L is complete.
c. If / and B are models of T, / ⊆ B, φ is a universal formula, x ∈ /,
and / φ(x), then B φ(x).
d. If / and B are models of T, / ⊆ B, φ is a primitive formula, x ∈ /,
and B φ(x), then / φ(x).
106 APPENDIX
Bibliography
[Bell] Bell and Slomson. Models and Ultraproducts. North-
Holland Publishing Company, Amsterdam, 1971.
[Cantor] Cantor, G. Contributions to the Foundations of the The-
ory of Transfinite Numbers. Dover Publications, New York,
1955.
[Chang] Chang and Keisler. Model Theory. North-Holland Publishing
Company, Amsterdam, 1973.
[Dauben] Dauben, J.W. Georg Cantor: His Mathematics and Phi-
losophy of the Infinite. Princeton University Press, Prince-
ton, New Jersey, 1979.
[Enderton] Enderton, H. A Mathematical Introduction to Logic. Aca-
demic Press, Inc., Orlando, Florida, 1972.
[Frege] Frege, G. The Foundations of Arithmetic. Evanston, Illi-
nois: Northwestern University Press, 1968.
[Griffin] Griffin, H. Elementary Theory of Numbers. McGraw-Hill,
New York, 1965.
[Kurosh] Kurosh, A.G. Lectures in General Algebra. Pergamon Press,
New York, 1965.
[Monk] Monk, J. D. Mathematical Logic Springer-Verlag, New York,
1975.
[Robinson] Robinson, A. and Zakon, E. Transactions of the American
Mathematical Society. 96(1960), pp. 222-236.
[Russell] Russell, B. The Principles of Mathematics. George Allen
and Unwin, London, 1903. Second edition, reprinted by W.W.
Norton and Company, New York, 1937.
[Tarski] Tarski, A. Undecidable Theories. North-Holland Publishing
Company, Amsterdam, 1968.
107
Index of Symbols
Logic
φ ∧ ψ, 101
φ ∨ ψ, 101
φ → ψ, 101
φ, 101
φ ↔ ψ, 101
∀xφ, 101
∃xφ, 101
Set Operations
x ∪ y, 102
x ∩ y, 102
x −y, 102
x; y, 102
x y, 102
T(x), 102
[x[, 102
x
[k]
, 102
ˆ x
k
, 94
Arithmetic
k [ n, 102
k!, 102
[in +j], 102
Size ordering
x ∼ y, 1
x < y, 1
x ≤ y, 1
x > y, 1
x ≥ y, 1
Model Theory
T ¬ φ, 103
T φ, 103
T
1
≡ T
2
, 103
/ φ, 103
/ φ , 103
/ ⊆ B, 103
Models and Domains
/, 103
˙
/, 103
B(/), 21
/
n
, 86
/
F
, 87
ˆ
/, 73
Q, 29
Q
f
, 67
Q
f,n
, 67
Q
n
, 29
Q
n
, 32
QC, 27
QC
n
, 27
N, 25
o
A
, 45
Z
A
, 54
´
Z
A
, 55
Sizing
˜ x, 45
C(x), 31

1
(x), 31

2
(x), 31
ρ(x), 32
δ(x), 32
θ(x), 32
θ
f
(x), 66
α(x), 31
β(x), 31
ρ(x, y), 94
σ
A
[x], 45
cvg(x, y), 94
108
INDEX OF SYMBOLS 109
cvg(x), 94
Node Construction
λ, 70
L(λ), 70
λ • m, 70
A
λ
, 70
a
λ
, 70
ˆ
/, 73
ι(λ), 75
q
λ
, 75
λ
n
, 76
Index of Theories
Languages
L
C
, 20
L
C<
, 20
L
<
, 20
L
N
, 86
Predicates
Abstract(x), 1
Atom(x), 20
Div
n
(x) , 36
Indisc(x, y), 10
Mod
n,m
(x), 36
Sm(x, y), 54
Sum(x, y, z), 20
Unit(x), 20
Times
n
(x, y), 36
Theories
ADIV
n
, 36
ASSOC, 46
ASSOC2, 54
ASYM
<
, 10
ASYM
F
, 10
ASYM
>
, 10
ATLEAST
n
, 23
BA, 22
BA
n
, 23
BAI, 23
BASIC, 24
BDIV
J
, 26
BDIV
n
, 26
CA, 38
CAI, 39
CAI
f
, 40
CORE, 13
CS, 21
CSI, 39
CANTOR
<
, 9
DEF
+
, 16
DEF
+
, 24
DEF

, 13
DEF
>
, 10
DISJ

, 24
DISJ
+
, 16
FUNC
+
, 16
DIV
J
, 26
DIV
n
, 25
EVEN, 9
EXACTLY
n
, 23
EXCORE, 18
INDISC

, 10
IRREF
<
, 13
INF, 23
MAX, 46
MIN, 46
MOD
n,m
, 25
MONOT, 16
ODD, 9
OUTPACING, 83
PCA, 47
PCS, 46
PSIZE, 46
QUASI, 10
REF

, 10
REP
<
, 9
RT
f
, 40
SIZE, 24
SYM

, 10
T
f
, 40
TRANS

, 10
TRANS
<
, 10
TRANS
>
, 10
TRICH, 13
110
INDEX OF THEORIES 111
UNIQ

, 54
UNIQ
+
, 54
UNIQ

, 46
Zg, 51
Zgm, 51
Subject Index
/-finite set, 74
/-near, 74
abelian groups with identity, 51
alternating pair, 85
asymptotic density, 94
basis, 21
cancellable abelian semigroups with
identity, 51
categorical, 104
chain, 103
charmed n-factor, 68
class size, language of, 21
classes, language of, 20
common n-factor, 68
compactness, 104
complete, 103
completion, 38
congruence class, 27
congruous, 40
consistent, 103
converges, 94
density, 94
depth, 70, 75
diagram, 105
diverges, 94
elementary extension, 103
equivalent, 104
existential, 104
existential assumption, 49
expansion, 105
extends, 70
extension, 103
filter, 87
finite, 40
finite class model, 14
finite completion, 39
finite intersection property, 87
finite set model, 14
functional assumptions, 49
generated, 103
incongruous, 40
infinite completion, 39
interpretation, 103
isomorphic, 103
isomorphic embedding, 103
language of
class size (L
C<
), 21
classes (L
C
), 20
size (L
<
), 21
subsets of N(L
N
), 86
language of a theory, 103
least common multiple, 36
length, 70
model complete, 104
molecule, 64
Monk mapping, 48
Monk’s Condition, 47
n-congruence class, 27
n-quasi-congruence class, 27
N-semigroups, 52
near, 27
(n.m)-partition, 68
nodal embedding, 77
node, 70
node atom, 70
112
SUBJECT INDEX 113
node atoms, 71
node set, 70
node sets, 71
one-one correspondence, 1
outpaces, 83
outpacing model, 84
partial, 40
PCA, 61
prime model, 104
primitive formula, 104
principal, 87
proves, 103
quasi-congruence class, 27
quasi-logical principles, 11
quasi-nodal, 74
reducible, 50
remainder function, 40
remainder theory, 40
representation principle, 12
roughly divisible, 25
run, 91
satisfaction, 103
schematic function, 101
shell, 70, 73
simple order, 51
simple translation, 48
simply reducible, 50
size, 30, 45, 66
size model, 45, 46
size, language of, 21
solution, 40
standard finite interpretation, 21
standard interpretation, 21
standard protomodel, 67
submodel, 103
subsets of N, language of, 86
τ-reducible, 50
theory, 103
theory of
abelian groups with identity, 51
cancellable abelian semigroups
with identity, 51
N-semigroups, 52
Z-groups, Zg, 51
Z-groups modulo I, 52
total, 40
translation, 48
trichotomy, 12
ultrafilter, 87
uniformly τ-reducible, 50
union of a chain, 103
universal, 104
universal-existential, 104
Z-groups, Zg, 51
Z-groups modulo I, 52
0-extends, 70

Fred M. Katz (fred@katz.com) Logic & Light, Inc. New York, NY 10028 USA

Sets and Their Sizes
Mathematics Subject Classification (2001): 03E10, 54A25

c 1981, 2001 Fred M. Katz. All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Logic & Light, Inc. 400 East 84th Street, New York, NY 10028, USA), except for brief excerpts in connection with reviews or scholarly analysis. The author hereby grants to M.I.T. permission to reproduce and distribute copies of this thesis document in whole or in part.

Preface
Cantor’s theory of cardinality violates common sense. It says, for example, that all infinite sets of integers are the same size. This thesis criticizes the arguments for Cantor’s theory and presents an alternative. The alternative is based on a general theory, CS (for Class Size ). CS consists of all sentences in the first order language with a subset predicate and a less-than predicate which are true in all interpretations of that language whose domain is a finite power set. Thus, CS says that less than is a linear ordering with highest and lowest members and that every set is larger than any of its proper subsets. Because the language of CS is so restricted, CS will have infinite interpretations. In particular, the notion of one-one correspondence cannot be expressed in this language, so Cantor’s definition of similarity will not be in CS, even though it is true for all finite sets. We show that CS is decidable but not finitely axiomatizable by characterizing the complete extensions of CS. CS has finite completions, which are true only in finite models and infinite completions, which are true only in infinite models. An infinite completion is determined by a set of remainder principles, which say, for each natural number, n, how many atoms remain when the universe is partitioned into n disjoint subsets of the same size. We show that any infinite completion of CS has a model over the power set of the natural numbers which satisfies an additional axiom: OUTPACING. If initial segments of A eventually become smaller than the corresponding initial segments of B, then A is smaller than B. Models which satisfy OUTPACING seem to accord with common intuitions about set size. In particular, they agree with the ordering suggested by the notion of asymptotic density.

Acknowledgements
This dissertation has been made possible by Cantor — who invented set theory, Tarski — who discovered model theory, Abraham Robinson — who introduced model-completeness, and George Boolos — who made all of these subjects accessible to me. Prof. Boolos read several handwritten versions and as many typed versions of this thesis, with care and patience each time around. I owe i

whatever errors. Nevertheless. and for his insistence on filling in mere details. Katz June. This dissertation was submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy at the Massachusetts Institute of Technology in April. The reader should join me in thanking him for his help with style and organization. Thesis Supervisor and Chairperson of the Department of Linguistics and Philosophy.ii him thanks for finding (and sometimes correcting) my errors. and omissions of some elementary (but very long) proofs. 2001 . typographical corrections. or infelicities remain are solely the responsibility of the author. 1981 It was certified and accepted by Prof. George Boolos. for directing me to some useful literature. gaps. Fred M. this is the text of my dissertation. Note Except for some minor wording changes. and simply for being interested.

. 63 5. . . . . . . . 2. . . . . . . . . .2 Models of CS and Outpacing . . 2 The General Theory 2. . . . .4 Pure Theory of Class Sizes Size models and PCS . . .The Theory of Class Size . 66 5. . . . . . . . . . . . 1. . . . . Remarks on showing that PCA axiomatizes PCS 45 45 47 53 61 . . . . . . 83 6. . . .1 CS . . . . . . . 79 6 Sets of Natural Numbers 83 6. . . . . . . . . . . .3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . 20 3. . . . . . . . . . . . . . . . . . . . . .1 Model completeness of CA . . . . . . . . . . . . . . . . . . . . . . . . . 3 A Formal Theory of Class Size 19 3. . . . . . . . . . . . . . . . . . . . . . PCA is model complete . . . . . . . . . . . . . . .4 Aims and outline . . 86 6. . .2 CA . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Size and density . . . . . . . . . . . . . . . . . . .1 The outpacing principle . . . . . .Axioms for CS . . . . . . . . . . . . . . . . . .2 Prime models for CA . . . 22 3. . . . . . . . . . . . .2 Addition of Set Sizes . . 94 iii . . . . . . . .3 Remarks on showing that CA axiomatizes CS . . . . .2 4. . . . . .1 The Problem . . . . . . . . . .Contents Preface 1 Introduction 1. . . . . . . . . . . .3 4. . . . . . . 38 4 The 4. . . . . . . . . . . . . . . . . . . . . . . . . . .2 Cantor’s argument . . . . . . .1 4. . . . . . . . . . . . . Some models and theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Cantor and the logicists 1. . . 5 Completeness of CA 63 5. . . . . . . . . . . . . .1 The Theory CORE . . . . . . . . . . i 1 1 3 6 7 9 9 16 . . . . . . . . . . . . 1. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . 103 C Model Completeness . . . . . . . . . . . . . . . . . 104 Bibliography Index of Symbols Index of Theories Subject Index 107 108 110 112 . . . . . . .iv CONTENTS Appendix 101 A Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 B Model Theory . . .

so it needs both justification and defense. The prejudice stems from Cantor’s theory of set size.. For example. Two sets are the same size just in case there is a oneone correspondence between them. but not the same size as y itself.1 Introduction 1. n 2n . it is so clearly true that anyone who comes to the matter without prejudice will accept it. as theories will. The standard theory also says that the set of even numbers is the same size as the set of integers since these two sets can also be paired off one-to-one: . y. . naive and otherwise. .. I spend a lot of time defending the theory because no one who comes to the matter comes without prejudice. −3 −6 −2 −4 −1 −2 1 0 0 1 2 2 4 3 6 . .. the set of upper-case letters is the same size as the set of lower-case letters.. Fine and good.1 The Problem This paper proposes a theory of set size which is based on intuitions. A set..... just in case x is the same size as some subset of y. . CANTOR< ... which is as old as sets themselves and so widely held as to be worthy of the name the standard theory. is smaller than a set. The theory goes beyond intuitions.. x... the standard theory says. the upper-case letters of the alphabet can be paired with the lower-case letters: ABCDEFGHIJKLMNOPQRSTUVWXYZ abcdefghijklmnopqrstuvwxyz So.. −n −2n . .. Cantor’s theory consists of just two principles: ONE-ONE. A one-one correspondence between two sets is a relation which pairs each member of either set with exactly one member of the other. I spend very little time justifying the theory..

To state this theory. No doubt. ∼.2 1. • The number of α’s is the same as the number of β’s. • The size of A is larger than the size of B. If one set properly includes another. • There are fewer α’s than β’s. even into the infinite. In the first case. INTRODUCTION Similarly. b. • The number of α’s is greater than the number of β’s. we assume throughout this thesis that the following schemata are equivalent. Incidentally. This is just to say that common sense seems to follow: SUBSET. . If A and B name sets. have) the same size. to use this reasoning. • There are just as many α’s than β’s. It maintains SUBSET and a few and far between principle and much else besides. it is still clear to common sense that the former is smaller than the latter. Common sense can make decisions without help from SUBSET. • The number of α’s is less than the number of β’s. and A > B is read as A is larger than B. you need a little number theory in addition to common sense. The integers contain all of the even integers and then some. the standard theory says that the set of positive even integers is the same size as the set of prime numbers: pair the n-th prime with the n-th positive even number. A ∼ B is read as A is the same size as B. One out of every two integers is even. common sense chokes on the standard theory. where it comes up against the standard theory. • The size of A is smaller than the size of B. given the number theory. common sense holds that the set of integers is larger than the set of even integers. In both of these cases. we use three two-place predicates: <. and >. to the readings of the three predicates given above. but. item by item. • There are more α’s than β’s. it’s the only conclusion common sense allows. assuming that A is the set of α’s and B is the set of β’s: a. then the first is larger than the second. • A and B are (or. The theory proposed here accommodates these bits of common sense reasoning. while the prime numbers are few and far between. So it’s just good common sense to believe there are more of the former than the latter. Though the set of primes is not contained in the set of even integers. then • • • A < B is read as A is smaller than B. c.

. ”). If someone introduces cardinal number as a defined predicate or as part of a contextual definition (e. But most mathematicians and philosophers do not use cardinal number as a mere abbreviation. we emphasize that we are not arguing that there really are such things as set sizes. They use the term in just the way that we use size and slide freely among (A). of a set M (is the general concept which) arises from M when we make abstraction of the nature of its elements and of the order in which they are given. Cantor’s. is rarely used in connection with Cantor’s theory. the definition is worth making. . and (C). however. Whether or not similarity is coextensive with being the same size. Neither does it insure that two sets have the same cardinal number just in case they are in one-one correspondence. To flesh out this notion of double abstraction.g. The term size.1. All of this is to say that the interest in one-one correspondence has not been sustained solely by its identification with the notion of size. and much else. 1. CANTOR’S ARGUMENT 3 Regarding this last group. who offered ONE-ONE as a theory. he offered an argument for this theory. the term cardinal number (sometimes just number ) is used most frequently. the continuum problem. per se. I have identified the standard theory. In addition. p. depends on neither the particular elements it contains nor on how those elements are arranged: The cardinal number. it can also be taken as just a theory of one-one correspondences. saying that two sets are similar iff they are in one-one correspondence can either be taken as a claim about size or be regarded as a mere definition. More specifically. of Cantor. Statements about sizes can be translated in familiar and long-winded ways into statements about sets. Though Cantor’s theory is usually taken as a theory of set size. This is true. [Cantor. nor that there are not really such things. Cantor used the terms power and cardinal number rather than size. Cantor reduces it to a second abstraction operator. In the literature. “We say that two sets have the same cardinal number just in case . the theory has consequences which do not prima facie seem to have anything to do with size or similarity: the existence of transcendental numbers comes to mind. |M |. it’s cardinal number. The technical brilliance of the theory attests to this: it has given us the transfinite hierarchy. in particular. so it might be wondered whether the standard theory is really so standard.2 Cantor’s argument Cantor bases his argument for ONE-ONE on the idea that the size of a set.2. (B). Hence. there is no point in discussing whether that person is right about size. one which works on the elements of sets rather than the sets .86] But to say that the cardinal number of a set does not depend on certain things is not to say what the cardinal is. with two principles about set size. The relation picked out is well studied and well worth the study. denying that they are the same does not endanger the theory of one-one correspondences. indeed. In stating ONE-ONE and CANTOR< . though we will not bother to do so.

if we abstract from its nature becomes a unit. and employs mappings between cardinal numbers and other sets.1) gives a definition of cardinal number.80] and so.1) (1. p. INTRODUCTION Every element. m.2) (1. [Cantor. [Cantor. I can only make sense of his arguments insofar as he treats cardinal numbers as sets: he refers to them as ‘definite aggregates’.3) |M | = { y | ∃x ∈ M ∧ y = Abstract(x) } ∀x∃y(y = Abstract(x) ∧ Unit(y)) ∀M ∀y(y ∈ |M | → Unit(y)) Abstract (x) is to be read as the result of abstracting from the element x.88) . because the cardinal number |M | remains unaltered if in the place of one or many or even all elements m of M other things are substituted. (ONE-ONEa) (ONE-ONEb) M ∼ N → |M | = |N | |M | = |N | → M ∼ N ONE-ONEa is true.) [Cantor. m. whereas its cardinal number is an abstract picture of it in our mind.4 themselves: 1. then in replacing each element. p. p. if f is a one-one mapping from M onto N . is a set composed of units (which has existence in our minds as an intellectual image or projection of M . from which Cantor proves both ONE-ONEa and ONE-ONEb.86] According to Jourdain.80] I have parenthesized the expressions above where Cantor describes cardinal numbers as mental entities. and |M | as the cardinal number of M . |M |.” [Cantor. So (1. p. Unit (x) as x is a unit. Cantor distinguised very sharply between an aggregate and a cardinal number that belongs to it: “Is not an aggregate an object outside us. The following three statements seem to express Cantor’s intent: (1. in terms of the operation of abstraction. concludes Cantor: The cardinal number. says Cantor. (p.86] and so. of M with f (m) M transforms into N without change of cardinal number. Nevertheless. supposes that they have elements.

so to speak. But abstractions of elements are just units. Cantor says. because . two sets with same cardinal number are similar. (1.4) are met and |M | = |N |. this contradicts ONE-ONEb. Abstract(b)} ∧ Abstract(a) = Abstract(b) (1. a and b. by (1. and similarity is an equivalence relation.1). That is. A weak version of this principle is: M = {a. Thus we can say that M ∼ |M |. by the definition of |M |.4) (a ∈ M ∧ b ∈ M ) / ∧ N = { x | (x ∈ M ∧ x ∈ a) ∨ x ∈ b } / → |M | = |N | The reasoning is clear: so far as the cardinal number of a set is concerned. So Abstract(a) = Abstract(b). the correspondence from M to its cardinal number will be many-one and not one-one.1. .4) says that replacing an element of a set with any element not in the set does not affect the cardinality.6). CANTOR’S ARGUMENT 5 In its weakest form. the principle Cantor cites says that if a single element of M is replaced by an arbitrary element not in M then the cardinal number of the set will remain the same. but only their abstractions.6) ∀x∀y(Abstract(x) = Abstract(y)) For. unit. (1. But. that enter into the cardinal number of a set. Conversely. one element is much the same as another. It is not the elements of a set. So Cantor’s argument for ONE-ONEa only works by assigning all nonempty sets the same. Unless each element of a set abstracts to a ‘special’.2. which entails (1. |M | grows. out of M in such a way that from every element m of M a special unit of |M | arises. so one is much the same as another. But |M | = {Abstract(a)} and |N | = {Abstract(b)}. . Generalizing this argument yields (1. cardinal number. since a set is similar to its cardinal number.5) These two arguments do one another in. Let M = {a} and let N = {b}. i. consider an arbitrary pair of elements. this means that (1. So. the argument that a set is similar to its cardinal number relies on (1. one-membered.5).e. the conditions of (1.7) ∀x∀y(x = y → Abstract(x) = Abstract(y)) . (1. But.1). distinct. So. ONE-ONEb is true. b} ∧ a = b → |M | = {Abstract(a).

while the concept word ‘unit’ changes unperceived into the proper name ‘one’. Kossak. then subsequently the concept of putting together (or collecting. says Frege. . .6) or its negation. then {a} and {b} have distinct cardinal numbers. The negation of (1. There is no way to repair Cantor’s argument. [Frege. the result is an agglomeration in which the objects remain still in possession of precisely those properties which serve to distinguish them from one another. by putting together identicals. We have shown that (1. Frege cites Schroder. “that numerical equality or identiy must be defined in terms of one-one correlation.6) contradicts ONE-ONEb. INTRODUCTION assuming only that any two objects can constitute a set. Russell displays similar caution about Cantor’s argument [Russell. or whatever we choose to call it) transforms itself into arithmetical addition. The word ‘unit’ is admirably adapted to conceal the difficulty . despite the fact that they are in one-one correspondence. then no two sets have the same cardinal number as defined by (1.305] and similar enthusiasm for his theory (see the quote at the beginning of Chapter 2. . Cantor’s definition of cardinal number is sufficient to refute the principle.3 Cantor and the logicists Though both Frege and Russell accepted Cantor’s theory of cardinality.1).8) contradicts ONE-ONEa. (1. pp 50-51] These misgivings about units do not prevent Frege from basing his definition of ‘number’ and his entire reduction of arithmetic on Cantor’s notion of oneone correspondence.6 1. 1.8) ∃x∃y(Abstract(x) = Abstract(y)) So one of (1. p.) . “This opinion”. But if the abstractions of any two elements are distinct. . {Abstract(a)} and {Abstract(b)}. or annexing. the result runs perpetually together into one and we never reach a plurality . He neatly summarizes the difficulty with such views: If we try to produce the number by putting together different distinct objects. contra ONE-ONEa. Similarly.8). So ONE-ONE is false whether (1. for example. and Cantor. neither accepted Cantor’s argument. and that is not the number. We start by calling the things to be numbered ‘units’ without detracting from their diversity. But if we try to do it in the other way. (1.6) and (1. seems in recent years to have gained widespread acceptance among mathematicians” [Frege.8) must be true. pp. Frege spends an entire chapter of the Grundlagen mocking mathematicians from Euclid to Schroder for defining numbers as sets of units. if a and b have distinct abstractions. is true. 73-74]. Rather than leading to a justification of ONE-ONE.6) is: (1. or uniting.

motivated set theory and determined its research problems. in contrast to Cantor. 1. as they would call it.1. the relative sizes of sets of ordered pairs should be preserved under translation. no matter how they are intended. notes that Cantor’s statement (1) is not a ‘true definition’ and merely presupposes that every collection has some such property as that indicated — a property. to a large extent. Cantor’s theory clears the way for other reductions. only upon their number. relied upon the principle of abstraction to obtain a ‘formal definition’ of cardinal numbers. independent of its terms and their order. But. so each ordered pair must be identified with a distinct set. since the existence of one-one correspondences will be preserved under a one-one mapping. then this second condition follows from the first. Chapter 2 canvasses common sense intuitions for some basic principles about set size. for example. the logicists have it both ways: adding ONE-ONE as a formal definition to set theory (or. they can advance it as a great lesson for simple common sense. if ONE-ONE is the correct theory of size.4. Chapter 3 reorganizes those principles into a tidy set of axioms. Well. Russell. Frege and Russell cleaned up Cantor’s presentation of the theory. for example. who had “taken” number “to be a primitive idea” and had to rely on “the primitive proposition that every collection has a number. pp. But there are two specific reasons that they should have seized upon ONE-ONE and CANTOR< . [Russell.” [Russell. Such points may be interesting. In addition. offers an account of where the intuitions come from (viz. Suppose. at the same time. and mines this source for additional principles. but in the absence of an alternative theory of size they are less than convincing. you have to identify each ordered pair with a set and define the relevant properties and relations among ordered pairs in terms of properties and relations among sets. you wish to reduce ordered pairs to sets. that is to say. and similarly Frege. First. p.305] So. One of the relations that has to be maintained is identity . logic) they have no obligation to defend it and can steer clear of peculiar arguments about units . So the notion of cardinality is born reduced. Second. The logicists’ adoption of Cantor’s theory of cardinality needs no great explanation: it came with set theory and. they both have the form of definitions. 304-305] So Russell. This dissertation presents such an alternative. while some people regard Cantor’s ONE-ONE as just a definition and others embrace it as a theory.4 Aims and outline It would be naive to suppose that people’s faith in Cantor’s theory would be shaken either by refuting specific arguments for ONE-ONE or by associating the acceptance of that theory with a discredited philosophy of mathematics. Chapters 4 and 5 prove that the theory so obtained is . AIMS AND OUTLINE 7 Of course. known facts about finite sets). we might feel tempted to add. depending.

Chapter 6 elaborates additional principles that concern only sets of natural numbers and demonstrates that these additional principles. are satisfiable in the domain of sets of natural numbers. . INTRODUCTION complete.8 1. in the sense that it embraces all facts about finite sets of a certain kind (i. expressible in a particular language). together with the theory in Chapter 3.e. Finally.

or is there a way of elaborating on common sense to get a plausible and reasonably adequate theory of cardinality? To be plausible.1 The Theory CORE SUBSET. consistency). is here in a very sorry plight. how much can you say about smaller than without using predicates (other than identity ) which relate the members of the sets being compared? I shall begin by stating a number of principles and explaining why they are included in the general theory. in the face of the proofs. it must choose between the paradox of Zeno and the paradox of Cantor. First. To be reasonably adequate. p. shocking to common sense . If x is a proper subset of y. a theory should at least avoid principles and consequences which violate common sense. I do not propose to help it. . 358] Is common sense confused about set size. since I consider that. .e. there is the SUBSET principle: The reason for including SUBSET should be obvious. then x is smaller than y. pace Cantor. as Russell says. What has prompted the search for an alternative to the standard theory of cardinality is the conflict 9 .2 The General Theory The possibility that whole and part may have the same number of terms is. it must be confessed. 2. Plausibility and adequacy are conflicting demands: the first says that there should not be too many principles (no false ones. therefore. Common sense. the second that there should not be too few principles. In the Introduction. I want to see how far we can go without such principles. i. I argued that a coherent theory of cardinality has to contain some principles that refer to the kinds of objects in sets. it ought to commit suicide in despair. though it need not be complete. In this chapter. a theory has to go beyond bare intuitions: it should not rest with trivialities and it should answer as many questions about set size as possible. [Russell. however.

and what I’ve gotten. The Appendix gives a full account of the notations used throughout this thesis. though it surely is not enough. asking people. But. x > y as x is larger than y. is any way to find out which of SUBSET and ONE-ONE is true. With carefully chosen examples. Now. husbands and wives. People’s intuitions about mathematics are notoriously unreliable. not to mention inconstant. is what I expected: support for SUBSET. Of course. where one was a proper subset of the other. finite sets. SUBSET. x ∼ y is to be read as x is the same size as y.) Naturally. and so forth: i. Given just this principle. So. harping on this fact might engender some unwarranted skepticism about mathematics. Now.with the technical attractions that it provides. The argument hinges on a suggestion about how to resolve cases where mathematical intuitions seem to conflict. (By and large because many people think that all infinite sets have the same size: Infinity. if you want to find out what common sense really thinks about SUBSET and ONE-ONE. From this it is supposed to follow that common sense cannot be relied upon. but that’s not what my argument depends on. in an unscientific way. There is no doubt that you can lead an unsuspecting person to agree to ONE-ONE by focusing their attention on forks and knives. so we should opt for ONE-ONE. it is often said that common sense supports both of these principles and that it is in doing so that common sense is confused. I would not venture that this sort of technique. by and large. you might even convince someone that ONE-ONE is true for infinite sets. I do not think that such guile is needed to lead someone to agree to SUBSET. all by itself seems to be a plausible alternative to Cantor’s theory. What I am suggesting is that there might be a rational way of studying mathematical intuitions and that we should at least explore this possibility before proclaiming common sense to be hopelessly confused on mathematical matters. The suggestion is to see what happens with particular cases on which the principles conflict before you’ve lead someone to agree to either of the general statements.10 2. you would present people with pairs of infinite sets. too. THE GENERAL THEORY between ONE-ONE and SUBSET.e. All of the following seem worthy (where x < y is to be read as x is smaller than y. there’s a difference in the way that common sense supports these two principles.) . I’ve actually tried this. So. It’s clear that we need additional principles. say the odds and the evens. it’s possible that one set is smaller than another just in case it is a proper subset of the other.

1. If x is F -er than y. so it could be maintained that each is true in virtue of its form. THE THEORY CORE Theory 2. for fear of inconsistency. ‘short’. there are some instances of the schemata that make for embarassing counterexamples: ‘further east than’ (in a round world) and ‘earlier than’ (in. withheld judgment on one or more of these principles.2. ‘happy’. for example. a possible world). e. What remains clear is that a theory of cardinality which openly denied any of these principles would be implausible: it would be ridiculed by common sense and mathematical sophisticates alike. y) abbreviates ∀z((z < x ↔ z < y) ∧ (z > x ↔ z > y)) We call the principles listed above quasi-logical principles because it is tempting to defend them as logical truths. y) x∼x x∼y→y∼x (x ∼ y ∧ y ∼ z) → x ∼ z x>y↔y<x 11 where Indisc(x. possibly. but not ‘unique’ or ‘brick’. it might be that casting the principles above as instances of the appropriate schema would only explain why they are part of common sense. Second. I can just barely imagine presenting a theory which. It seems that if a case could be made that these statements. where F is to be replaced by an adjective from which comparatives can be formed. that would be good reason to say that there is no . that each is a logical truth. The other principles might be defended in the same way. First. are inconsistent with SUBSET. is larger than. then y is not F -er than x. If x is smaller than y. taken together. So. But to do so without good reason would be counterproductive. But using such observations to support these principles would be problematic for two reasons. then y is not smaller than x. it would require taking positions on many questions about logical form and grammatical form which would take us far afield and. for example: is smaller than. QUASI-LOGICAL (ASYM< ) (ASYM> ) (TRANS< ) (TRANS> ) (INDISC∼ ) (REF∼ ) (SYM∼ ) (TRANS∼ ) (DEF> ) x < y → ¬y < x x > y → ¬y > x (x < y ∧ y < z) → x < z (x > y ∧ y > z) → x > z x ∼ y → Indisc(x.g.1. I’m told. It appears that every instance of this schema is true. is the same size as. ‘tall’. in unregimented English: ASYM< . This sentence can be regarded as an instance of the schema: ASYMF . though the schema for INDISC∼ would have to be restricted to triples of corresponding comparatives. Consider the first principle. antagonize first-order logicians.1.

finite or infinite. our approach is to canonize what common sense holds to be true about cardinality and show that the result is consistent and reasonably adequate. there is the representation principle. Analogous statements about physical objects are neither intuitive nor very clearly true. This would be impossible. then the first is the same size as some proper part of the second. there must be some additional principles to be extracted from common sense. if ‘part’ is taken to mean ‘leg or top or rim or . given that the principles are logically contingent. it would surely be regarded with suspicion.. but it is not .. we get a model for our theory by assigning to < the relation of being a proper subset of. that the same size as is an equivalence relation. what do the quasi-logical principles say? Only that smaller than is a partial ordering. If REP< is true. (REP< ) x < y → ∃x (x ∼ x ∧ x ⊂ y) which says that if a set. no one should condemn its residual caution about (1). Since my goal is to counter such a conclusion. there is the principle of trichotomy. y. then it seems to be an interesting and special fact about sets. First. Even if common sense can be persuaded to take particles and arbitrary fusions of such as parts of tables. So the strategy here is not to adduce principles and argue for the truth of each.12 2. Now. Instead. After all. THE GENERAL THEORY reasonably adequate alternative to Cantor’s theory. assigning to > the relation of properly including. Since common sense knows that different sets can be the same size. Second. As long as we restrict ourselves to SUBSET and the quasi-logical principles. if we are given any domain of sets. it seems that the proper strategy is to include such seemingly obvious truths and to show that the resulting theory is consistent. and assigning to ∼ the identity relation. REP< may be open to doubt. the larger than is the converse partial ordering. (1) does not stand a chance of being regarded as true: (1) If one table is smaller than another. x. So. consistency is no problem. While a theory of set size which excluded TRICH might escape ridicule. if the principles of common sense were incompatible with TRICH. this is a principle which common sense has no particular feelings about. REP< was originally included in this theory for technical reasons. Indeed.’. (TRICH) x<y∨x∼y∨x>y which says that any two sets are comparable in size. is smaller than another. this would undoubtedly be used to discredit them. then x is the same size as some proper subset of y. For example. and that sets of the same size are indiscernible under the partial orderings. We shall now consider some principles which cannot be regarded as quasilogical. it makes it easier to reduce the set of axioms already presented and it provides a basis for several principles not yet presented.

Proof.2.2. Theorem 2. Then T ≡ CORE.1. if x ∼ x and x ⊂ y. which is logically equivalent to the conjunction of INDISC∼ and its converse ∼ INDISC: (INDISC∼ ) (∼ INDISC) ∼ x ∼ y → Indisc(x. but before proceeding. by INDISC∼ . by SUBSET: so x < y . I want to reduce the principles mentioned above to a tidy set of axioms. So x < y or y < x. Let T = QUASI. We only need to show that ∼ INDISC is entailed by T . for it is entailed by Cantor’s definition of <: (CANTOR< ) x < y ↔ ¬(x ∼ y) ∧ ∃x (x ∼ x ∧ x ⊂ y) If CANTOR< is regarded as a principle instead of a definition. (SUBSET) (DEF∼ ) (TRANS∼ ) (DEF> ) (IRREF< ) (TRICH) x⊂y→x<y x ∼ y ↔ Indisc(x. Theory 2. By REP< . or larger than one but not larger than the other. So x < y. The entire set of principles already adopted are equivalent to the following. say x . REP< . by INDISC∼ . Conversely. must be the same size as x. by INDISC∼ and ASYM< . THE THEORY CORE 13 a principle that Cantorians could complain about. y) (x ∼ y ∧ y ∼ z) → x ∼ z x>y↔y<x ¬(x < x) x<y∨x∼y∨x>y The only axiom in CORE that has not already been introduced is DEF∼ . then it is entailed by the principles we have already mentioned: If x < y. I want to estimate how far we’ve gone. But x < y. Suppose ¬(x ∼ y). There are more principles to come. . then x < y. I’d like to take stock of what we already have. CORE. Second. then their being different in size is attributable to the existence of some set which is either smaller than one but not smaller than the other. But ¬(x < x) and ¬(y < y). First. then ¬(x ∼ y). the core theory. which will be referred to as the core theory. some proper subset of y. by SUBSET. T CORE. y) Indisc(x.1.3. by IRREF< . y) → x ∼ y INDISC says that if two sets fail to be the same size.1. by TRICH. SUBSET.

TRANS∼ . In both cases. Theorem 2. (iv) INDISC∼ . then x < x. SYM∼ . The normal ordering of finite sets is.1. CORE is consistent. But then x < z by DEF∼ . then A CORE. Definition 2. So x ⊂ z and. and IRREF> follow from the corresponding principles for < and DEF> .4. contra IRREF< . and REF∼ are logical consequences of DEF∼ . Fact 2.1. By adding TRICH we have ruled out all non-standard interpretations of <. A is a finite class model with basis x iff ˙ (i) A = P(x). In fact.14 CORE T 2. (ii) ASYM< . A is a finite set model iff ˙ (i) A = { x | x is a finite subset of Y }.6. a. the finite cardinalities determine a quasi-linear ordering of the sets in which any set is higher than any of its proper subsets. in fact.1. then x < y because y ∼ y. there is an x such that x ∼ x and x ⊂ y . the only one that satisfies CORE. (iii) TRANS> . by SUBSET. for some infinite Y . If x < y and y < x. there is a y such that y ∼ y and y ⊂ z by REP< . CORE. then A b. If A is a finite set model. x < z. where x is finite. a.5. two kinds of models satisfy CORE. Suppose that A is a model such that . If y < z. So. Proof. THE GENERAL THEORY (i) TRANS< . (ii) A (iii) A a ⊂ b iff a ⊂ b a < b iff |a| < |b| b. (ii) A (iii) A a ⊂ b iff a ⊂ b a < b iff |a| < |b| Models can be specified by stipulation the smaller than relation since larger than and same size as are defined in terms of smaller than. ASYM> . by TRANS< . If A is a finite class model. by DEF∼ . If x < y.

b ∈ A. suppose that (*) is true for all i ≤ n: If then Suppose So But if then So But if and then But if pick with So and So So |a| = |b| = n + 1 A (a ∼ b).2. |b | = n + 1 A (b < b) A (b ∼ a) A (a < b) |b| = n + 1 by (*). We shall prove (*) by induction on n: (*) Suppose then But if then And if then So So If |a| = n. . contra our hypothesis. A Then A CORE. then y ∈ A.1. THE THEORY CORE ˙ a. a < b iff |a| < |b| 15 Proof. by REP< by induction. then A n=0 a=∅ |b| = 0 A (a ∼ b) |b| = 0 a⊂b A (a < b) A (a ∼ b) (a ∼ b) iff |b| = n by REF∼ by SUBSET by INDISC∼ Now. as follows: A (a < b) A (a ∼ a) for some a ⊂ b A a ⊂b |a | ≤ n |a| ≤ n |a| = n + 1 A (a ∼ b) |b| ≥ n + 1 |b| > n + 1 b ⊂ b. If x ∈ A. If x ∈ A and y ⊂ x. by condition (b) by SUBSET contra our supposition. then x is finite. ˙ ˙ b. and c.

16 2. THE GENERAL THEORY 2. y. though the interpretation of Sum does not agree with the intended reading. z) → (y ∼ y ↔ Sum(x. y. y. z) → (z ∼ z ↔ Sum(x. y. y. x ∪ y) FUNC+ says that sets bear Sum relations to one another by virtue of their sizes alone. y. Since the domains of our intended models contain only sets and not sizes of sets. DISJ+ tries to say what function on sizes the Sum relation captures by fixing the function on paradigm cases: disjoint sets. z where |z| = (|x| + |y|) mod (n + 1) Both FUNC+ and DISJ+ will be satisfied. z ) Addition for disjoint sets (DISJ+ ) Monotonicity of Addition (MONOT) Sum(x. MONOT rules out such interpretations. . This condition must clearly be met if Sum is to be read as specified above. y. Addition Functionality of addition (a) (b) (c) Sum(x.2 Addition of Set Sizes We shall now extend CORE to an account of addition of set sizes. z) → (x ∼ x ↔ Sum(x . The following principles are sufficient for a theory of addition: Theory 2. y. z) is to be read as the size of z is the sum of the sizes of x and y . y . z) Sum(x.2. y. We shall show this by proving that ADDITION and CORE entail DEF+ : (DEF+ ) Sum(x.1. we have to formulate our principles in terms of a three-place predicate true of triples of sets: Sum (x. Given an interpretation of ’<’ over a power set there is at most one way of interpreting Sum which satisfies ADDITION. y. z) ↔ ∃x ∃y (x ∼ x ∧ y ∼ y ∧ x ∩ y = ∅ ∧ x ∪ y = z) DEF+ says that the extension of Sum is determined by the extension of <. But FUNC+ and DISJ+ leave open the possibility that addition is cyclic: suppose we begin with a finite class model whose basis has n elements and assign to Sum those triples x. z) → x < z ∨ x ∼ z x ∩ y = ∅ → Sum(x. z) Sum(x.

. if Sum is to be interpreted in a way compatible with ADDITION. z) x<z∨x∼z x ∼ z. Theorem 2. (DISJ∪ ) (x ∼ x ∧ y ∼ y ∧ x ∩ y = ∅ ∧ x ∩ y = ∅) → (x ∪ y) ∼ (x ∪ y ) (Note: The proofs in this chapter will use boolean principles freely. x ∪ y) Sum(x . Sum(x .3. then Sum has to be definable by DEF+ . x ∪ y) Sum(x . x =z y = ∅. y. y .2. as before. CORE + ADDITION Proof.) Theorem 2. despite the fact that we have not yet introduced them.2. y. (⇒) Suppose So But if Let and then So So And if pick with Let But Sum(x. y . DISJ∪ .2. z) y ∼y x<z x ⊂z x ∼x y =z−x y ∼ y. CORE + ADDITION Proof. Suppose then and So So So (x ∼ x ∧ y ∼ y ∧ x ∩ y = ∅ ∧ x ∩ y = ∅) Sum(x.2. DEF+ by MONOT by DISJ+ by FUNC+ (a) by FUNC+ (b) by REP< .2. x ∪ y) x∪y ∼x ∪y by DISJ+ by FUNC+ (a) by FUNC+ (b) by FUNC+ (c) DISJ∪ If the minimal conditions on addition are to be satisfied in a model of CORE. z) Sum(x. y . ADDITION OF SET SIZES 17 A model of CORE must satisfy an additional principle. x ∪ y ) Sum(x . y. y .

y .2.4. EXCORE.5. z) by DISJ+ by FUNC+ (a) by FUNC+ (b) Theory 2. Fact 2.2. is CORE ∪ ADDITION. The proofs are elementary. z) Sum(x. y. the extended core. THE GENERAL THEORY (x ∼ x ∧ y ∼ y ∧ x ∩ y = ∅ ∧ x ∪ y = z) Sum(x . z) Sum(x. The following are consequences of EXCORE: x ∩ y1 = x ∩ y2 = ∅ ∧ y1 ∼ y2 x ∩ y1 = x ∩ y2 = ∅ ∧ y1 < y2 x1 ∩ y1 = x1 ∩ y2 = ∅ ∧ x1 ∼ x2 ∧ y1 < y2 x⊂z∧y ⊂z∧x<y x⊂z∧y ⊂z∧x∼y x<y x∼y x<y Proof. → x ∪ y1 ∼ x ∪ y2 → x ∪ y1 < x ∪ y2 → x1 ∪ y1 < x2 ∪ y2 → (z − y) < (z − x) → (z − y) ∼ (z − x) → ∃y (y ∼ y ∧ x ⊂ y ) → x − (x ∩ y) ∼ y − (x ∩ y) → x − (x ∩ y) < y − (x ∩ y) .18 (⇐) Suppose then So So 2. y .

We shall show this by constructing a model for the general theory whose domain is the power set of the natural numbers. the domains consisted of all finite subsets of a given infinite set. the existence of the model does not rule out the possibility that T is incompatible with DISJ∪ . the domains of the interpretations were finite power sets. For a particular principle. We produced several such principles. Our goal is to show that this general theory of set size can be maintained for infinite sets as well as for finite sets. then T should be maintained anyway. (2) show that φ is inconsistent with T and argue that T is somehow more fundamental or more intuitive than φ. we deal with DISJ∪ as in (1). that we offered as a general theory of size the axioms of CORE other than REP< . and found two kinds of interpretations which satisfied these principles. for example. Since any partial ordering can be extended to a quasi-linear ordering. say. It seems futile to try to rule out the need to resort to the third approach for any cases at all. but we can 19 . In the other case. Cantor’s principle ONE-ONE is dealt with as in case (2). constituting EXCORE. we may take one of three tacks: (1) add φ to T. In one case. But unless we have a guarantee that the model constructed will satisfy. we searched for principles that accord with preCantorian ideas about sizes of sets. Suppose. So T just says that smaller than is a quasi-linear ordering which extends the partial ordering given by the proper subset relation. DISJ∪ . since DISJ∪ is in EXCORE. (3) acquiesce in ignorance of whether T and φ are compatible and argue that if they are incompatible. obtaining T . Call this theory T. Insofar as both kinds of interpretation have quantifiers ranging over finite sets. and show that T has a model over P(N). we may say that they demonstrate that EXCORE is true when it is construed as being about finite sets. But the ability to construct such a model is interesting only to the extent that the general theory it satisfies is reasonably adequate.3 A Formal Theory of Class Size In the preceding chapter. T has a model over P(N). φ. Below. in this case DISJ∪ .

To avoid the problems involved in speaking of all statements. the one-place predicate. relation. for each predicate and each operation symbol. 3.1 CS . but only that they do not refer to these sorts of objects as such. we arrive at a theory which can be satisfied over infinite power sets. is the first order language with individual constants ∅ and I. predicates. In fact. we know that no statement in that language can arise as something which ought to be true over infinite power sets but might be incompatible with our theory. By drawing statements only from this relatively weak language. we need only list the individual constants. This is not to say the axioms do not apply to relations. since we include in the language all statements of the language which are true over any finite power set. We then construct a theory by taking all statements in that language which are true over all finite power sets. the two- .1. Not only is ONE-ONE such a statement. Consider. we might instead settle for all statements in L.The Theory of Class Size The theories discussed in this paper will be formulated within first order predicate logic with identity. we mention it only to emphasize that no claim is made here that we have the strongest possible general theory of set size. LC . In the next section. a. T must fall short of the full expressive power of the language of set theory. we define a language just strong enough to express EXCORE.1. where L is some judiciously chosen language. the language of classes. This in turn opens the possibility of a succession of richer languages and a corresponding succession of stronger theories of size. There remains the possibility that we could follow the same strategy with a more expressive language. such a language can be obtained by including a notion of the product of set sizes. But. To specify the language in which a theory is expressed. the axioms in EXCORE.20 3. Definition 3. though it would have to remain less expressive that the full language of set theory. these axioms involve only boolean operations and inclusion relations among sets. In short. or other sets of ordered pairs. LC . To avoid ONE-ONE. the possible existence of any such hierarchy is sheer speculation. the only set theory implicit in these axioms is boolean algebra. A FORMAL THEORY OF CLASS SIZE reduce this need to the extent that we include in our general theory. and operation symbols of the language and stipulate the rank. T. Atom. functions. We cannot construct T by taking all statements which are true for finite sets. now. then. They do not use the notions ordered pair. or number of argument places. At this point. or function. but using the notion of all statements true for finite sets presupposes that we have some idea of the range of all statements. or a sort of Venn diagram set theory. Other than size relations. as many plausible statements as possible.

A is a standard interpretation of LC< . CS. the language of class size. the two-place predicates < and ⊂. the theory of class size. Definition 3. . LC< . is the set of all sentences of LC< which are true in all standard finite interpretations of LC< . for some x. ∩. If LC ⊂ L. Since the notion of one-toone correspondence cannot be expressed in this language. b. A(∅) = ∅. is the first order language containing just the non-logical constants in LC and L< . CS . and the three place predicate Sum. a.1. If A is a standard interpretation and A = P(x). Cantor’s principle ONE-ONE will not be included in the theory CS. even though it is true over any finite power set. the language of size. By drawing only on principles which can be stated in LC< we have at least ruled out the most obvious danger of paradox.1. then x is the basis of A. is the first order language with the oneplace predicate Unit. and (ii) A assigns the usual interpretations to all constants of LC : A(I) = x. and ∪. A is a standard finite interpretation of LC< iff a.1. Following the strategy outlined above. A A A a < b iff |a| < |b|.3. Definition 3. then A is a standard interpretation of L iff ˙ (i) A = P(x). a ⊂ b iff a ⊂ b A ˙ b. c. L< .3. and the two place operation symbols..2. B(A).1. a ∼ b iff |a| = |b|. and c. we define the theory of class size in terms of interpretations of LC< over finite power sets.4. −. b. and Unit(a) iff |a| = 1 Definition 3. A has a finite basis.THE THEORY OF CLASS SIZE 21 place predicate ⊂.

It is not obvious that CS has standard infinite models. Def 9. This proof is presented in Chaper 5. 3. in which the universe is an infinite power set. 141 and Def.1 BA. In Chapter 6 we show that such models do exist. . the theory of atomic boolean algebras. so. must satisfy BA: Definition 3. Section 2 develops a set of axioms. p.2.28. 9. BA is clearly not a complete axiomatization of CS. A FORMAL THEORY OF CLASS SIZE Since CS has arbitarily large finite models. The key idea here is that complete extensions of BA can be obtained either by stipulating the finite number of atoms in a model or by saying that there are infinitely many atoms. CA.2. Since all of the universes of the interpretations mentioned in the definition of CS are power sets. since BA does not involve any size notions.3. Section 3 outlines the proof that CA does indeed axiomatize CS. To show this we need to draw on some established facts about the complete extensions of BA (in the language LC ).2 CA . The present chapter is devoted to getting a clearer picture of the theory CS. This will be done in several stages. CA. they must be atomic boolean algebras and.Axioms for CS Here we shall develop a set of axioms. But BA does entail all the sentences in CS which do not themselves involve size notions. 3.Axioms for atomic boolean algebra We’ll begin with the obvious. BA. after a slight detour in Chapter 4. for the theory CS.22 3. for CS.1. is the theory consisting of the following axioms: x∪y =y∪x x∩y =y∩x x ∪ (y ∪ z) = (x ∪ y) ∪ z) x ∩ (y ∩ z) = (x ∩ y) ∩ z) x ∩ (y ∪ z) = (x ∩ y) ∪ (y ∩ z) x ∪ (y ∩ z) = (x ∪ y) ∩ (y ∪ z) x ∩ (I − x) = ∅ x ∪ (I − x) = I x ⊂ y ↔ (x ∪ y) = y ∧ x = y Atom(x) ↔ (∀y)(y ⊂ x → y = ∅) x = ∅ → (∃y)(Atom(y) ∧ (y ⊂ x ∨ y = x) These axioms are adapted from [Monk. 151]. p. it has infinite models.

φ. then φ is true in some finite model of BA.. If BAI Proof. Fact 3.30.. / .2. 9. an n-atom atomic boolean algebra is isomorphic to any standard finite interpretation of LC with an n-element basis. then. ¬φ is true in some finite model A of BA.32.AXIOMS FOR CS Definition 3. a. so φ is not true in A and φ ∈ CS. If A is finite.2. Hence. 23 a. ATLEASTn is a sentence which says that there are at least n atoms: ∃x1 .152] b. then BA φ.24. [Monk. / If A is infinite. If CS φ.3.360] e.151] d. BAn is complete. EXACTLYn e. then ¬φ is consistent with BAI. p. Cor 9. For n ≥ 1.2. For n ≥ 1. Prop. ∧ Atom(xn ) ∧ { xi = xj | 0 < i < j ≤ n }) b. so BAI ¬φ. BAI and the theories BAn .2. A. n ≥ 1. then A is isomorphic to some standard finite interpration A of LC . BAn = BA. ¬φ is true in some standard finite interpretation of LC and. For n ≥ 1. If BA φ.. BAn is categorical. and φ ∈ LC . BAI = BA ∪ INF Fact 3. INF is a set of sentences which is satisfied in all and only infinite models of BA: INF = { ATLEASTn | n > 1 } d. then ¬φ is true is some atomic boolean algebra. BA is complete.5. [Monk. But BAI is complete.2. By compactness. p. again.2. then ¬φ is true in A . p. φ ∈ CS.∃xn (Atom(x1 ) ∧ .4.2. (Immediate from above) c. there is a k such that BA ∪ { ATLEASTn | 1 ≤ n ≤ k } φ So φ is true in any atomic boolean algebra with more than k atoms. Theorem 3. But. Theorem 21. EXACTLYn is a sentence which says that there are exactly n atoms: ATLEASTn ∧ ¬ATLEASTn+1 c. Proof.4.. For n ≥ 1. consistent extensions of BA. [Monk. BA ∪ INF φ. CA . are the only complete. By 3.3.

these models of BASIC will reappear as submodels of various standard models of CS over P(N). the basic theory. BASIC. z) ↔ ∃x ∃y ( x∼x ∧y ∼y ∧x ∩y =∅ ∧ x ∪ y = z) (DISJ∪ ) (x ∼ x ∧ y ∼ y ∧x ∩ y = ∅ ∧x ∪ y = z) → x ∪ y ∼ x ∪ y Combining the principles of boolean algebra and the size principles. we obtain our first serious attempt at a general theory of size: Definition 3.2. though we will not prove that CA ≡ CS until Chapter 5.24 3. In Chapter 6. y) ¬(x < x) x<y∨x∼y∨x>y Unit(x) ↔ Atom(x) Sum(x. A FORMAL THEORY OF CLASS SIZE 3.6.7.2. include different subsets of N and assign different size orderings to these sets. in addition to the immediate purpose of establishing independence results.2. CA.2 Size principles Here.2. we will have an effective set of sentences. . To show that the new principles really do need to be added. we just gather the principles presented above as EXCORE: Theory 3. When we are done. The models will. however. y. CORE SIZE consists of the following axioms: (SUBSET) (DEF> ) (DEF ) (IRREF< ) (TRICH) (DEF1 ) (DEF ) + ∼ x⊂y→x<y x>y↔y<x x ∼ y ↔ Indisc(x. the Class Size Axioms. These models will be similar in that (1)their universes will be subsets of P(N). and (3) all boolean symbols will receive their usual interpretations. these models provide a glimpse of how sets of natural numbers are ordered by size. (2) their atoms will be singletons in P(N).3 Division principles BASIC is not a complete axiomatization of CS. is defined as: BA ∪ SIZE 3. So. we’ll need some non-standard models of BASIC. In this section we shall exhibit an infinite number of principles which need to be added to BASIC in order to axiomatize CS.

this statement is in CS but is satisfied by neither our original model nor the model as amended. the new model will be closed under boolean operations and will satisfy BA.e. So EVEN∨ODD is in CS. But this model will still not satisfy CS. for a. As you might suspect. A. in which case they cannot be disjoint. ODD says that there is a remainder of a single atom. F is a model of BASIC. So. Definition 3. or both cofinite.2. or 2. b ∈ F: F (a < b) iff a and b are both finite and |a| < |b| or a and b are both cofinite and |N − a| > |N − b| or a is finite and b is infinite. as we can see by generalizing the argument above. We can construct a similar statement that says the universe is roughly divisible by three — with remainder 0.8.2. those that differ from the evens or odds by a finite set. Consider the model F whose universe consists of all only the ˙ finite and cofinite subsets of N. for any two sets that are the same size are either both finite.AXIOMS FOR CS 25 Every standard finite interpretation. where. We now formalize this line of reasoning. With these additions made. but not entailed by BASIC. But BASIC (EVEN ∨ ODD). then . we also need to include all sets which are near the set of evens or the set of odds. this is just the tip of the iceberg of principles missing from an axiomatization of CS. As with (EVEN ∨ ODD). If 0 ≤ m < n. of LC< satisfies exactly one of the following: (EVEN) (ODD) ∃x∃y(x ∼ y ∧ x ∩ y = ∅ ∧ x ∪ y = I) ∃x∃y∃y(x ∼ y ∧ x ∩ y = x ∩ z = y ∩ z = ∅ ∧Atom(z) ∧ x ∪ y ∪ z = I) A EVEN if |B(A)| is even and A ODD if |B(A)| is odd. we can extend the model and again we can produce a statement of CS which is false in the resulting model. in which case their union is also finite. But neither EVEN nor ODD is true in F. Again. we can extend F to a model that satisfies EVEN by including the set of even numbers and the set of odd numbers and making them the same size. Informally. (EVEN ∨ ODD). To round out the result to a model of BASIC. 1. i.3. we can construct an infinite model of BASIC . (EVEN∨ODD) says that the universe is roughly divisible by two: EVEN says that the universe is divisible by two without remainder. There is a (unique) way of ordering these added sets by smaller than that will satisfy BASIC: rank them according to the size and direction of their finite difference from the odds or the evens. CA .

To build such models from subsets of N. BASIC Proof.2.. xn ) ∩ (y1 ∪ . BASIC DIVn .0 ∨ . b. BDIVJ = BASIC ∪ DIVJ c.2. The model. . DIVn is the sentence MODn. we shall include sets which can be regarded as fractional portions of N. ym ) = I) MODn. ∧ Atom(ym ) ∧ 1≤i<j≤n (xi ∩ xj = ∅) (yi ∩ yj = ∅) 1≤i<j≤m ∧ ∧ (x1 ∪ .m is the sentence 3. then (a) (b) (c) A A CS MODn. . ∼ xn ∧ Atom(y1 ) ∧ . by 3. we need some independence results for sets of DIVn sentences. . DIVJ = { DIVn | n ∈ J } b. If J is a set of natural numbers. . but it does not work. . for any n finite sets have a finite union and any two cofinite sets overlap. BDIVj = BDIV{j} Our independence results will be obtained by constructing models of BASIC which satisfy specific sets of DIV sentences. So. If 0 ≤ m < n and A is a standard finite interpretation of LC< . A FORMAL THEORY OF CLASS SIZE ∃x1 . To demonstrate this.m says that the universe is divisible into n sets of the same size with m atoms remaining.10. ∨ MODn. . .9c. . . . xn ) ∪ (y1 ∪ . .∃xn ∃y1 . . .9. ∃ym ( x1 ∼ x2 ∼ . .2.n−1 Fact 3. We could consider adding all DIVn sentences to BASIC in the hope that this would yield a complete set of axioms for CS. .26 a.11. then a.2. Definition 3. ym ) = ∅ ∧ (x1 ∪ . .. MODn. We did consider this. F defined above satisfies BASIC but not DIVn . BASIC CS.m iff |B(A)| ≡ m mod n DIVn DIVn CS Theorem 3. . If n > 1.

d.2. . Suppose x = a1 ∪ . NEAR(x.15. QCn = { x | x is an n-quasi-congruence class } f. 27 a.13.2. N itself is the union of n-congruence classes. Fact 3. x is a congruence class iff x is an n-congruence class for some n. b.14. bj . If x ∈ QCn .AXIOMS FOR CS Definition 3. . x is an n-quasi-congruence class iff x is the union of finitely many n-congruence classes. ak and y = b1 ∪ . e. N is an n-quasi-congruence class for every n > 0: N = [nk + 0] ∪ . [2n + 1]. b. then x ∪ y ∈ QCn . are 2-congruence classes. x is a quasi-congruence class iff x is an n-quasi-congruence class for some n. . . a. ∪ [nk + n − 1] Fact 3. For n ≥ 0. iff x − y and y − x are both finite. a. CA . x is an n-congruence class iff x = [nk + m] for some m.2. The set of evens. . d.y). If x ∈ QCn and y ∈ QCn . a. . . . Then x ∪ y = a1 ∪ . . and the set of odds. If x is the union of m of these classes. [3k + 2] is a 3-congruence class. . bj . ak ∪ b1 ∪ . x is near y iff there are finite sets. b. where 0 ≤ m < n. c.12. n>0 QCn Definition 3. c. w1 and w2 such that x = (y ∪ w1 ) − w2 . then N − x ∈ QCn . N is a 1-congruence class.3. then N − x is the union of the remaining (n − m) n-congruence classes. b. x is near y. [2n]. QC = Examples. Proof.2. .2.

x is near z.16. Hence. If x is near y. But (y − x1 ) is finite. x is near x. since x − x = ∅. But (x2 − y) is finite. is finite. then x1 ∪ y is near x2 ∪ y. If x1 is near x2 . c. Suppose x is near y and y is near z.17. Immediate. A FORMAL THEORY OF CLASS SIZE ⇒ Let w1 = x − y and let w2 = y − x. Note that (z − x) = ((z ∩ y) − x) ∪ ((z − y) − x) But (z − y) − x is finite because (z − y) is finite and ((z ∩ y) − x) is finite because ((z ∩ y) − x) ⊆ (y − x). which is finite. so (y − x) is finite. . Fact 3. then y is near x. So (x − z) is also finite. x1 is near y. Then x − y ⊆ w1 and y − x ⊆ w2 . (x − z) = ((x ∩ y) − z) ∪ ((x − y) − z) where ((x ∩ y) − z) ⊆ (y − z) and ((x − y) − z ⊆ (x − y). 3. so (x − y) is finite.2. a. If x1 ⊆ x ⊆ x2 . ⇐ Suppose x = (y ∪ w1 ) − w2 . Since x1 ⊆ x. then x is near y. So the union.2. b.18. (y − x) ⊆ (y − x1 ). Proof. NEAR is an equivalence relation. If x is near y. and x2 is near y. If x1 is near x2 and y1 is near y2 . Proof. c. a. Similarly. Fact 3. which is finite.2. then N − x is near N − y Proof.28 Proof. Since x ⊆ x2 . (x − y) ⊆ (x2 − y). so x − y and y − x are finite. (z − x). then x1 ∪ y1 is near x2 ∪ y2 . Fact 3. b.

Using Fact 3.2. c. 9. since N ∈ {Qn } and if x ∈ Qn . a. Q = Examples. Definition 3.2.142] Theorem 3.2. where I is interpreted as A. CA . But (x1 ∪ y) − (x2 ∪ y) ⊆ (x1 − x2 ) and (x2 ∪ y) − (x1 ∪ y) ⊆ (x2 − x1 ) 29 b. by transitivity. then x ∪ y ∈ A. If x ∈ A and y ∈ A.19.3. So x1 ∪ y1 is near x2 ∪ y2 . (N − x) − (N − y) = y − x and (N − y) − (N − x) = x − y. Q2 = { y ⊆ N | y is finite. p.2. a. b. If A is a class of sets such that a. [Monk.2.2. then c. Q2 is the domain of the model constructed above to satisfy DIV2 . and n>0 Qn b. A − x ∈ A. then A forms a boolean algebra under the usual set-theoretic operations. near [2n] or near [2n + 1] }. By (a).21.AXIOMS FOR CS a. then x ⊆ N. Def.4. If n > 0. So if x and y are near each other. Proof.13b by 3. Q1 = { y ⊆ N | y is finite or cofinite }.20. {Qn } = N. then Qn forms an atomic boolean algebra under the usual set-theoretic operations. x1 − x2 and x2 − x1 are finite.2.1. We can now define the domains of the models we will use to establish the independence results. then Suppose and So and So x is near y y ∈ QCn N − y ∈ QCn N − x is near N − y N − x ∈ Qn . If x ∈ Qn . so are their complements. p. Qn = { y | y is near an n-quasi-congruence class } b. x1 ∪ y1 is near x2 ∪ y1 .20: a. by 3. cofinite.18b . If x ∈ A. Fact 3. b. which is near x2 ∪ y2 . Since x1 is near x2 . A ∈ A.141 and Corr 9.

The first member is a rational between 0 and 1 which represents the density of the set. δ . assigned to x.2. δ2 are sizes. ρ. (ii) θ1 + θ2 = ρ1 + ρ2 . So Qn is an atomic boolean algebra. then (i) θ1 < θ2 iff ρ1 < ρ2 or (ρ1 = ρ2 ∧ δ1 < δ2 ). First. if ρ = 0. The set x has density k/n and this is the value. Moreover. The sizes assigned to sets are ordered pairs. then δ ≤ 0. where ρ is rational and δ is an integer. Specifically. every singleton is in Qn since all singletons are near ∅. Our intention in assigning sizes to sets is as follows: Suppose x is near an n-quasi-congruence class. We now define a size function on all sets which are near quasi-congruence classes. Qn is a boolean algebra. .13a by 3. If θ1 = ρ1 .2.2. δ1 + δ2 Only some of these sizes will actually be assigned to sets. a. b. δ1 and θ2 = ρ2 .23. Fact 3. Moreover. Suppose and then and and and But then and So x ∈ Qn y ∈ Qn x is near x 3. x .30 c.18b Thus. So x is the union of k ≤ n n-quasi-congruence classes. The definitions and facts below formalize this intention and demonstrate that the assignment of sizes to sets is well-defined.22. The δ value assigned to x is the finite number of elements added to or removed from x to obtain x. A size is an ordered pair ρ. A FORMAL THEORY OF CLASS SIZE x ∈ QCn for some x y is near y y ∈ QCn for some y x ∪ y ∈ QCn x ∪ y is near x ∪ y x ∪ y ∈ QCn by 3. then δ ≥ 0 and if ρ = 1. Definition 3. we define the ordering and arithmetic for these sizes with the intention of inducing the size ordering and Sum relation for sets from the assignment of sizes to sets. The second member is an integer which represents the finite (possibly negative) deviation of a set from average sets of the same density.2. a size will be assigned to a set only if 0 ≤ ρ ≤ 1.

If x ∈ QC. Let n be the least number such that x ∈ QCn and y ∈ QCn So each is a disjoint union of n-congruence classes: x = x1 ∪ . for any j xi ⊆ x − y x − y is infinite. α(x) = the least n such that x ∈ QCn . b. y ∈ QC. since xi is infinite Similarly. then a. if a ∈ y − x. the two would have to be near each other. . . CA . Proof.2. But this is impossible by (a). b. c.2. ∪ yk Suppose then But So So So a∈x−y a ∈ xi for some i. no two quasi-congruence classes are near each other. ∆1 (x) and ∆2 (x) are finite and x = (C(x) ∪ ∆1 (x)) − ∆2 (x) Definition 3. . If x ∈ QC. If x were near two quasi-congruence classes. b. If x is near a quasi-congruence class. . Any set is near at most one quasi-congruence class. a ∈ yj . C(x) is the quasi-congruence class near x. since NEAR is transitive.2. . then a.3. for any y.24.25. and x = y. / xi = yj . then x is not near y. ∆1 (x) = x − C(x). ∆2 (x) = C(x) − x. Definition 3. then y − x is infinite. (That is. a. β(x) = the unique k such that x is the disjoint union of k α(x)congruence classes.) b.AXIOMS FOR CS 31 a. ∪ xj and y = y1 ∪ .

32 Examples.

3. A FORMAL THEORY OF CLASS SIZE

a. If x = [2n + 1], α(x) = 2 and β(x) = 1. b. If x = [4n + 1] ∪ [4n + 2], α(x) = 4 and β(x) = 2. c. If x = [4n + 1] ∪ [4n + 3], α(x) = 2 and β(x) = 1, since x = [2n + 1]. Definition 3.2.26. If x ∈ Q, then ρ(x) = β(C(x)) α(C(x)) δ(x) = |∆1 (x)| − |∆2 (x)| θ(x) = ρ(x), δ(x)

We can, at last, define the models to be used in our independence proof. Definition 3.2.27. For n > 0, Qn is the interpretation Q of LC< in which boolean symbols receive their usual interpretation and ˙ Q = Qn x < y iff θ(x) < θ(y) x ∼ y iff θ(x) = θ(y)

Q

Q Q Q Unit(x) iff θ(x) = 0, 1 Sum(x, y, z) iff θ(z) = θ(x) + θ(y)

To show that the models Qn satisfy BASIC, we will need the following facts about congruence classes. Fact 3.2.28. a. If x = [an + b] and a2 = ac, then x=
0≤i<c

[a2 n + (ia + b)]

b. If x ∈ QCn and m = kn, then x ∈ QCm . c. If x ∈ QC and y ∈ QC, there is an n such that x ∈ QCn and y ∈ QCn . Proof. a. If k ∈ [a2 n + (ia + b)] for some i, 0 ≤ i < c, then, for some n1 , k = a2 n1 + ia + b = a(cn1 ) + ia + b = a(cn1 + i) + b So k ∈ x.

3.2. CA - AXIOMS FOR CS

33

If k ∈ x, there is an n1 such that k = an1 + b. Let n2 be the greatest n such that a2 n ≤ k and let k = k − a2 n2 . Since and then So But So k ≡ b mod a a2 n2 ≡ 0 mod a k ≡ b mod a k = ia + b, where 0 ≤ i < c k = a2 n2 + k = a2 n2 + ia + b k ∈ [a2 n + (ia + b)]

b. By (a), each n-congruence class is a disjoint union of m-congruence classes. c. Suppose x ∈ QCn1 and y ∈ QCn2 . Then, by (b), both x and y are in QCn1 n2 .

Theorem 3.2.29. For any n > 0, Qn

BASIC

Proof. By 3.2.21, Qn is an atomic boolean algebra; so Qn BA. The <-relation of Qn is induced from the linear ordering of sizes; so it is a quasi-linear ordering and IRREF< , TRICH, and DEF∼ are satisfied. As for the remaining axioms: a. SUBSET: Suppose x ⊂ y. If C(x) = C(y) then ρ(x) = ρ(y) and ∆1 (x) ⊆ ∆1 (y) and ∆2 (x) ⊆ ∆2 (y), where at least one of these inclusions is proper. So δ(x) < δ(y). But if C(x) = C(y), then C(x) ⊂ C(y), so ρ(x) < ρ(y). In either case, θ(x) < θ(y), so Qn x < y.

b. REP< : Suppose Qn x < y. So θ(x) = k1 /n, δ1 , θ(x) = k2 /n, δ2 , and either k1 < k2 or δ1 < δ2 . We want to find some x ⊂ y such that Qn x∼x. If k1 = k2 > 0, then y must be infinite; so, x can be obtained by removing δ2 − δ1 atoms from y. If k1 = k2 = 0, then 0 ≤ δ1 < δ2 ; so, again, x can be obtained by removing δ2 − δ1 atoms from y. If k2 > k1 > 0, then let y1 be the union of k1 n-congruence classes contained in C(y). So y − y1 is infinite and y1 − y is finite. Let y2 = y1 − (y1 − y) = y1 ∩ y. So θ(y2 ) = k1 /n, −δ3 where δ3 = |y1 − y|. Finally, let δ4 = δ3 + δ1 and x = y2 ∪ y3 , where y3 ⊆ y1 − y with |y3 | = δ4 . If k1 = 0, then x is finite. If k2 > 0, then y is infinite, so there is no problem. If k2 = 0, then y is finite, but has more members than x, since δ1 < δ2 . So, let x be any proper subset of y with δ1 members.

34

3. A FORMAL THEORY OF CLASS SIZE c. DISJ∪ : It is enough to show that if x and y are disjoint, then θ(x ∪ y) = θ(x) + θ(y). We need the following three facts: (i) C(x ∪ y) = C(x) ∪ C(y). (See Fact 3.2.18b.) (ii) ∆1 (x ∪ y) = (∆1 (x) ∪ ∆1 (y)) − (C(x) ∪ C(y)). If z ∈ x ∪ y but a ∈ C(x ∪ y), then a ∈ ∆1 (x) or a ∈ ∆1 (y); any / element of ∆1 (x) is also in ∆1 (x ∪ y) unless it is in C(y); any element of ∆1 (y) is also in ∆1 (x ∪ y) unless it is in C(x). (iii) ∆2 (x ∪ y) = (∆2 (x) ∪ ∆2 (y)) − (∆1 (x) ∪ ∆1 (y)). Note that if x ∈ Q, y ∈ Q, and x ∩ y = ∅, then C(x) ∩ C(y) = ∅; otherwise, C(x) and C(y) have an infinite intersection.

Hence, if then or And if unless And if unless

a ∈ C(x) ∪ C(y) − (x ∪ y) a ∈ C(x) − x = ∆2 (x) a ∈ C(y) − y = ∆2 (y) a ∈ ∆2 (x), then a ∈ ∆2 (x ∪ y) a ∈ ∆1 (y) a ∈ ∆2 (y), then a ∈ ∆2 (x ∪ y) a ∈ ∆1 (x)

. From (ii) we obtain (iia): (iia) |∆1 (x ∪ y)| = |∆1 (x) ∪ ∆1 (y)| − |(∆1 (x) ∪ ∆1 (y) ∩ (C(x) ∪ C(y))|

and from (iii) we obtain (iiia): (iiia) |∆2 (x ∪ y)| = |∆2 (x) ∪ ∆2 (y)| − |∆1 (x) ∪ ∆1 (y) ∩ (∆2 (x) ∪ ∆2 (y))| But ∆1 (x) and ∆1 (y) are disjoint, since x and y are disjoint. So: (iv) |∆1 (x) ∪ ∆1 (y)| = |∆1 (x)| + |∆1 (y)| = δ1 (x) + δ1 (y)

Since ∆2 (x) and ∆2 (y) are contained, respectively, in C(x) and C(y), which are disjoint, they are also disjoint. So: (v) |∆2 (x) ∪ ∆2 (y)| = δ2 (x) + δ2 (y)

.30. y = ∅ to satisfy DEF+ θ(x) < θ(z) Qn x ∼ x x ⊂ z for some x y =z−x z=x y y∼y by DISJ∪ θ(y) = θ(y ) so Qn since Qn REP< θ(x ) + θ(y ) = θ(z) θ(x ) = θ(x) θ(y ) = θ(z) − θ(x) = θ(y) (Cancellation is valid for sizes because it is valid for rationals and integers. Qn Proof.2. d. If and and then θ(z) = θ(x ) + θ(y ) θ(x ) = θ(x) θ(y ) = θ(y) θ(z) = θ(x) + θ(y) Theorem 3. CA .AXIOMS FOR CS So. by (i). z) θ(z) = θ(x) + θ(y) θ(x) ≤ θ(z) θ(x) = θ(z) θ(y) = 0. δ1 (x ∪ y)−δ2 (x ∪ y) = (δ1 (x) + δ1 (y)) − (δ2 (x) + δ2 (y)) = (δ1 (x) − δ2 (x)) + (δ1 (y) − δ2 (y)) = δ(x) + δ(y) 35 Since ρ(x ∪ y) = ρ(x) + ρ(y). For any n > 0. we know that θ(x ∪ y) = θ(x) + θ(y). 0 y=∅ x = z. DIVm iff m | n.3. y.) Conversely.2. DEF+ : Suppose So Clearly Assume then So Let Assume then and Let So Claim For But So Qn Sum(x.

bm = N. Bm Letting bj = 1/m. b . So b = 0 and a = n/m. then the least common multiple of J. But if x ∈ Qn . A FORMAL THEORY OF CLASS SIZE ⇒ θ(N) = 1. DIVm . ¬DIVm b. If m n. If J is finite. b1 ∪ . Definition 3. µ(J) always exists since the product of all members of J is divisible by each member of J. 0 .2.36 3. ⇐ For each i. if m disjoint sets of the same size exhaust N. DIVm . then Ai ∩ Aj = ∅ and Qn 1/n. If J = ∅ and J is finite. c. .32.2. 0 ≤ i < n. c. Immediate from (b). If BDIVn b. for integral a and b. then θ(x) = a/n. . .2. they must each have size 1/m. . . Remark. the product is greater than µ(J). 0 . then m | n. 0≤i<n Ai . Usually. then m | µ(J).3. Hence. There are only finitely many m for which BDIVJ Proof. 0 . if m µ(J).33. let Ai = [nk + i]. then a. . group the n sets Ai into m collections with p members in each: B1 .31. If BDIVJ DIVm . is the least k which is divisible by every member of J. Qµ(J) BDIVJ since it satisfies DIVj for each j ∈ J. since only finitely many m divide µ(J). Such sentences can be produced by generalizing the notion of divisibility to all sets instead of applying it only to the universe. Based on Theorem 3. Corollary 3. 0 = Furthermore. We are now ready to show that BDIVn CS by finding a sentence in CS which entails infinitely many DIVn sentences. . Definition 3. 0 . then Qn BDIVn . µ(J). then Qµ(J) DIVm . So N = If i = j. Bj for 1 ≤ j ≤ m. we have bj ∈ Qn and θ(bj ) = p/n.20: a. But. (Ai ∼ Aj ) since θ(Ai ) = θ(Aj ) = Letting p = n/m.

Proof. Notice that in the presence of BASIC. Divn (z) is the formula Modn.2. xi )] Timesn (x. ∨ Modn. . each the same size as x. so ADIVn ADIVnm . ADIVn : a. then nm = n.2. . DIVnm . Hence. Thus.n−1 (z) d. A ADIVnk+1 .3. then x can be partitioned into nk sets of the same size and i atoms. If 0 ≤ n. A T . Every set in every standard finite interpretation is a finite set. .m (z) is the formula: ∃x∃y∃v∃w[T imesn (x. and x ∈ A. DIVn .m (z) says that z can be partitioned in n sets of the same size and m atoms.34. CS ADIVn . we have partitioned x into nk n sets of the same size and nk j + i atoms. BASIC. w) ∧ Sum(v.2. CA . We have taken this opportunity to formulate the divisibility predicates purely in terms of size predicates. Each non-atomic set in the partition can be further partitioned into n sets of the same size and atoms. for all m b. But nk n = nk+1 and. ∃xn [x0 = ∅ ∧ xn = y ∧ 1≤i≤n 37 Sum(xi−1 . y) says that y is the same size as the Sum of n sets. then Timesn (x. c. . a.0 (z) ∨ . where i < nk . ADIVnm . By induction on m: if m = 1. nk j + i < nk+1 . Fact 3. and c. If 0 ≤ m < n. where j < n. w.m ≡ Modn. x. MODn. y) is the formula: ∃x0 . for every n. b. v) ∧ Unit(y) ∧ T imesm (y. . z)] Modn. for all m. ADIVn is the sentence ∀xDivn (x) Remark. since i < nk and j < n. Proof. then Modn. and all finite sets are roughly divisible by every n.m (I) and DIVn ≡ Divn (I) Fact 3.35.AXIOMS FOR CS a. ˙ If T ADIVnk .

then by compactness there is a finite set J such that BDIVJ ADIVn .2. A FORMAL THEORY OF CLASS SIZE c. CA ≡ BASIC ∪ { ADIVn | n > 0 } The remainder of this chapter and the next two are devoted to showing that CA is. 3. Since T2 ⊆ T .35c.36. which says that BDIVJ entails only finitely many DIVn sentences. complete extension of T2 is consistent with T1 . by Fact 3. ¬φ is consistent. If every consistent.2.38 b. To do so. .e. Proof. (Lindenbaum’s lemma) Every consistent theory has a consistent. If BDIVN ADIVn .1.37. a. a complete set of axioms for CS.3 Remarks on showing that CA axiomatizes CS We know that CS CA and we want to show that CA ≡ CS. So. a. T . p.3. T is also a consistent. consistent extension of T . by Lindenbaum’s lemma. Theorem 3. [Monk. But then BDIVJ DIVnk for every k. T is a completion of T iff T is a complete. it will be sufficient to show that every consistent extension of CA is consistent with CS. Suppose that T1 φ. But this contradicts Fact 3. T has a consistent. Definition 3. Fact 3. complete extension of T2 . complete extension.32. i. T2 φ. Theorem 11.200] b. that CA CS. indeed. 3. we are left with a theory weaker than CS. Obvious. But T is not consistent with T1 . then T2 T1 . b.3.2.2. BDIVN ADIVn for any n > 1. even if we add all of the DIV sentences to BASIC.13. complete extension. it is reasonable to attempt an axiomatization of CS as follows: Definition 3. Immediate from (a) and (b).2. Proof. Then T = T2 . Since this weakness has arisen in the case of ADIV sentences.

3. for each n. The model A of CA.1. So T ¬ATLEASTn+1 and has no infinite models. of CAI has to solve the disjunction DIVn for each n. Remainder theories specify. So A CS. . we define two kinds of completions of a theory. b.0 ∨ .3. to show that every completion of CA is consistent with CS. b. CA BASIC and. . the number of atoms remaining when the universe is divided into n disjoint sets of the same size. Every finite completion of CA is equivalent to CA. a.3. a. where DIVn is MODn. then. CSI = CS ∪ INF So. a. CAI = CA ∪ INF b. we may now concentrate on showing that every completion of CAI is consistent with CSI. Proof. Definition 3.3.6.) Proof. T is an infinite completion of T iff T is true in some infinite model of T . ⇒ Suppose A is a finite model of T with n atoms. Fact 3. so T has no finite models. What. T EXACTLYn . .6.3. T . T is a finite completion of T iff T is true in some finite model of T. Every finite completion of CA is consistent with CS. Fact 3. T has to entail one of the disjuncts. b. Since T is complete. EXACTLYn is a standard finite interpretation. If T is a completion of T and BA ⊆ T .4.3. by Theorem 2.n−1 Any completion. then (T is a finite completion of T iff T is not an infinite completion of T . for some n.EXACTLYn .5. Definition 3. BASIC is categorical in every finite power. ⇐ T ATLEASTn for every n. a. ∨ MODn. REMARKS ON SHOWING THAT CA AXIOMATIZES CS 39 To prove that every completion of CA is consistent with CS.3. are the completions of CAI? Recall that CA entails DIVn for every n > 0.

RTf . So n is a solution of f and. f is total iff Dom(f ) = N+ . f. (⇒)For every finite restriction. CSf is consistent. of f . (⇒)Let φ = RTf . we will show that CAIf is consistent just in case CSIf is consistent. d. The remainder theory specified by f . otherwise. a.9. then CSf is consistent iff f is congruous. (⇐)Every finite restriction. f is finite iff Dom(f ) is finite. of f is congruous. otherwise f is partial. CAIf is complete and that these are the only complete extensions of CAI. (Henceforth f ranges over remainder functions. Chapter 5 will show that if f is total. If T is a theory.40 3. n is a solution for f iff for any i ∈ Dom(f ). φ b.) b.7. then. f is congruous by 3.8.80] b. f is congruous by 3. a. g. then f has a solution iff f is congruous iff f has infinitely many solutions. p. f is congruous iff for any i. there is some n such that An φ. If f is finite. Proof.) Theorem 3.3. CSf is consistent iff f is congruous.3. c. f is congruous iff every finite restriction of f is congruous. by (a).8b. φ is consistent. Tf = T ∪ RTf .3. is the set of sentences { MODn. If f is finite. j) | f (i) − f (j). By compactness. [Griffin. In this section. j ∈ Dom(f ). CSg is consistent.8a. f has a solution. a. f : N+ → N is a remainder function iff 0 ≤ f (n) < n for all n ∈ Dom(f ). Since CS. n ≡ f (i) mod i. Fact 3. (⇐)If f is congruous. e. (Let f (p) = p − 1. for all primes p. Hence. CSIf is consistent iff f is congruous. . Any solution would have to be larger than every prime. c. So CSg is consistent. hence.3.m | f (n) = m }.3. n. g. g. So An CS. then gcd(i. each such g is congruous. There are congruous f without any solutions. b. c. By (a). f is incongruous. Remainder functions and remainder theories a. Theorem 5-11. A FORMAL THEORY OF CLASS SIZE Definition 3.

by (b).1 .1 . . it is sufficient to show that every finite subtheory.p . g. . Proof.p → MODm. . A Bi ∼ Bj .3. then CA MODm. . . and p = k2 n + q. So. So g has arbitrarily large solutions and CSg has finite models large enough to satisfy T . .q . .3.3.1 .p → ¬MODm. But.n a2. REMARKS ON SHOWING THAT CA AXIOMATIZES CS c. we must first establish that certain sentences are theorems of CA.3. if T is such a theory.1 .n ∪ 1≤j≤k2 aj.q . . and p ≡ q mod n. a2.8b. Furthermore. Suppose A MODm.n and p atoms a1. cq ) So. To do this. . B(A) = 1≤i≤n Bi ∪ (c1 . cq For 1 ≤ i ≤ n. f is congruous.1 . g is as well. then T ⊆ CSg ∪ { AT LEAST i | i < n } for some n and some finite restriction. for 1 ≤ i. Lemma 3. A MODn. b1.10. bk1 .q . j ≤ n.n ak2 .n bk1 . a1. .1 . . So. . . B(A) can be partitioned into m sets of the same size b1. let Bi = 1≤j≤k1 bj. ak2 . our proposed axiomatization of CS. b2. T is consistent.3.n c1 . If n | m. T . by 3. Hence.11. of CSIf is consistent. 0 ≤ q < n. . .n Since A DISJ∪ . k1 = m/n. . . Lemma 3. If 0 ≤ p < q < m. Since f is congruous.n b2. then CA MODm. so is CSf . We now want to prove a similar theorem for CA. 41 (⇐)By compactness. (⇒)If CSIf is consistent. . of f . .

CAf is consistent iff f is congruous. A FORMAL THEORY OF CLASS SIZE MODm. Thus (I − A) ∼ (I − B). . a. . then X ∪ A < Y ∪ B. the original supposition entails a contradiction. bq = I where the a’s and b’s are atoms and the x’s (y’s) are disjoint sets of the same size in A. and k such that k = gcd(i. . But Y ∼ Y . for B is the union of fewer than m atoms while Z is the union of m non-empty sets. Let X = x1 ∪ · · · ∪ xm Y = y1 ∪ · · · ∪ ym A = a1 ∪ · · · ∪ ap B = b1 ∪ · · · ∪ bp B = bp+1 ∪ · · · ∪ bq We claim that y1 < x1 . For. by RC∼ . Suppose A 3. . Since A REP< . . if y1 ∼ x1 . ap = I y1 ∪ . But Z must be larger than B . If f is finite. c. j. But A ∼ B.p ∧ ¬MODm. there is a component of Y of the same size. . We will show that (*) CA ¬(MODi. Let and So and zi = xi − yi Y = yi ∪ · · · ∪ ym Y ∪Z =X =I −A Y ∪ B = I − B. neither is possible since Y ∪ B ⊂ I = X ∪ a. there is a proper subset yi of xi which is the same size at yi . (⇒)Supposing that f is incongruous. So yi < xi for 1 ≤ i ≤ m.12.q . since for each component of Y . Then: A A x1 ∪ . then X ∪ A ∼ Y ∪ B and if x1 < y1 . a. Proof. . so Y ∪ Z ∼ Y ∪ B . j) and k (f (i) − f (j)).f (j) ) . So Z ∼ B . ym ∪ b1 ∪ . Theorem 3.f (i) ∧ MODj. by RC∼ . there exist i. then CAf is consistent iff f is congruous.42 Proof. xm ∪ a1 ∪ . b. CAIf is consistent iff f is congruous. since each is the disjoint union of p atoms. . So.3.

Let p and q be such that 0 ≤ p. So (2) is a consequence of (3).9c. It might help to review our strategy before presenting the difficult parts of the proof that CA ≡ CS.f (j) → MODk. c. b.13.11 yields (3) CA MODk.3.3.1b.9b. (4) If f is total and congruous. Immediate from 3.3.p → ¬MODk.q From (1).3. we have (1) (2) CA MODi.12c. See proof of 3. See proof of 3. and (3) we may conclude (*). (3) If T is a completion of CAI.10. then CSIf is also consistent (see 3.3. . (2).3.3.3.3. (⇐)Follows from 3.12).9a since CAf ⊆ CSf . REMARKS ON SHOWING THAT CA AXIOMATIZES CS from which it follows that CAf is inconsistent. We already know that the finite completions of CA are consistent with CS (see 3. To establish (3).3. Proof. CAIf is consistent iff CSIf is consistent. then CAIf is complete. f (i) ≡ p mod k.p .3.9c and 3. The main objective is (1). q < k.3. Since k f (i) − f (j).q 43 since k | i and k | j.3. p = q. congruous f . it is sufficient to prove (4) because every completion of CAI entails CAIf for some total f . and CA MODj. (1) (2) CA CS Every completion of CA is consistent with CS. which follows from (2) by 3.5c) and that if CAIf is consistent. So lemma 3. Corollary 3.f (i) → MODk. then T ≡ CAIf for some total. and f (j) ≡ q mod k By lemma 3.

(7) CAI is model complete. Fact C. since any extension of a model complete theory is also model complete (see Appendix. Finally. then T is complete (see Appendix.44 3. But first. So (4) follows from (5) and (6). we consider a simpler theory. A FORMAL THEORY OF CLASS SIZE To prove (4).3). (5) (6) If f is total and congruous. we can infer (6) from (7). PSIZE. which deals only with sizes of sets and ignores boolean relations. The proof outlined here will be carried out in Chapter 5. Fact C7a). then CAIf has a prime model. CAIf is model complete. a result we need for showing that CAI is model complete. For any f . we invoke the prime model test: If T is model complete and T has a prime model. . So (1) follows from (5) and (7). Chapter 4 formulates PSIZE and establishes that it is model complete.

PSIZE. This method is the same outlined in chapter 3 to show that CA ≡ CS. here. we identify a theory. Definition 4.we only give a sketchy treatment of PCS itself. then PCS is our version of number theory. section 4 indicates how PCA could be shown to axiomatize PCS. whose models are Z-groups taken modulo some specific element.1.4 The Pure Theory of Class Sizes CS is about sets. But our main reason for introducing PCS is to aid the proof that CA is model complete. PCS is the set of statements true in SA for any standard finite model. Using some results about model theory in section 2. but only about sizes of sets.1 Size models and PCS BASIC. PCS (the pure theory of class sizes). 4. for it. are equivalence classes of sets which have the same size. L< . This is done by reducing PCA to the theory Zgm. it makes claims about sets in terms of their boolean relations and their size relations. In this chapter. then σA [x] is the size of x in A: σA [x] = { y | A 45 (x ∼ y) } . consists of equivalence classes ˙ drawn from A under the same size relation. PCS is worth examining in its own right. as follows: for each model. Section 3 establishes that PCA is model complete. which is not about sets. Suppose A ˙ a. Finally. Sizes. If x is a member of A. PCA. of BASIC. the size model. A. PCA consists of a theory. For this reason. PCS is formulated in the language of size relations. the main result of this chapter and the only result needed for subsequent proofs. which holds in SA whenever A BASIC and a set of divisibility principles. Section 1 defines PCS and develops a set of axioms.1. A. SA . for if number theory is the theory of cardinal numbers.

is the interpretation of L< whose domain ˙ is σA [x] | x ∈ A where SA σA [x] ∼ σA [y] iff A SA σA [x] < σA [y] iff A Sum(σA [x].2. y. then A Sum(x. y. Definition 4. z ). PCS. the pure theory of class sizes. The one-place operator x is to be read as the complementary size of x: ˜ SA (y = x) iff SA ˜ Sum(x. σA [z]) iff A SA Unit(σA [x]) iff A x∼y x<y Sum(x.3. the size model for A. y. the theory of sizes. z) and A z ∼ z . consists of the following axioms: Order axioms (IRREF< ) (TRANS< ) (UNIQ∼ ) (MAX) (MIN) (TRICH) Unit axioms Unit(x) ↔ (y < x ↔ y = 0) ∃xUnit(x) ¬(x < x) x<y∧y <z →x<z x∼y↔x=y x≤I 0≤x x<y∨x∼y∨y <x . For example. y. if A Sum(x. THE PURE THEORY OF CLASS SIZES b. SA . consists of all sentences of L< which are true in the size model of every standard finite interpretation of LC< . σA [y]. Definition 4. z). PSIZE.46 4.1. I) These interpretations are well defined because the predicates are satisfied ˙ by elements of A in virtue of their sizes.1. and Atom(x) SA c.

1.z2 ) → (x1 < x2 ↔ z1 < z2 ) Sum(x.4.2 Some models and theories We shall use Theorem 4. j. y1 . .4. in chapter 5.7. n with the usual ordering.2. We offer PCA as an axiomatic version of PCS: Definition 4. w2 ) → Sum(x.6.1a to show that PCA is model complete and Theorem 4. . where SA Unit(x) iff x = 1 and SA Sum(i. y.1b to show.2. x. y2 . x) Sum(x. z) Sum(x1 . If T satisfies Monk’s Condition. that CAI is model complete. the pure class size axioms. w) ∃zSum(x. y. SOME MODELS AND THEORIES Sum axioms (IDENT) (COMM) (MONOT) (ASSOC) (EXIST+ ) (EXIST− ) (COMP) Fact 4. w2 . y. z) ↔ Sum(y. w) ∧ Sum(y.2. then SA PCA If A is a standard finite interpretation of BASIC with n atoms. . z. then SA PCS. z) x ≤ z → ∃ySum(x.5. z. In particular. PSIZE Sum(x. 0. j. 4.2. Theorem 4. . k) iff (i + j) = k ≤ n Once n is fixed. z) ∧ y2 ≤ y1 → ∃zSum(x. z1 ) ∧ Sum(x2 . If A Fact 4. MONOT rules out the interpretation in which Sum(i. PSIZE fails to axiomatize PCS for the same reason that BASIC fails to axiomatize CS: the lack of divisibility principles. x. then the elements of SA can be regarded as the sequence 0. this is the only interpretation allowed by the axioms PSIZE. w1 ) ∧ Sum(w1 . is defined as: PSIZE ∪ { ADIVn | n > 0 } Fact 4. then T is model complete.1.1. If A CA. k) is satisfied just in case (i + j) ≡ k mod n + 1. y. I) ˜ BASIC.1. PCA. a. PSIZE 47 Proof. z) Sum(x. .1. y.

THE PURE THEORY OF CLASS SIZES Monk’s Condition. . To establish the model completeness of PCA. assigns to each n-place function symbol.359] b.2. So. A ⊆ B. .2. b. . an (from A) and b1 . O.355] Definition 4. In chapter 5.2.2. Suppose L1 and L2 are first order languages and L12 = L1 − L2 . . If T is model complete in L. in L12 a (quantifier free) formula. of L2 with exactly one free variable. assigns to each n-place predicate. which: a. and L is an expansion of L by adjoining new individual constants. and c. . a. . bm ˙ ˙ (from B-A). then T2 is also model complete. τP . . of L2 with exactly (n + 1) free variables. So. . ∃ym φ2 So. Fact 4. of L2 with exactly n free variables. then there is an isomor˙ ˙ phism. If T1 is model complete and T1 T2 . If LT has no function symbols.2. B φ3 (a1 . f : C → A such that if x ∈ C ∩ A. τO . let φ3 be ∃y1 . we use Monk’s Theorem to infer the model completeness of the theory CA from that of PCA. then T is model complete in L . . we use Fact 4. φ3 is a primitive formula. .3. Finally. [Monk. . and C is a finitely generated submodel of B. assigns to the universal quantifier a (quantifier free) formula. so A does as well. a. . If T is model complete and LT has no function symbols. τ .5d in the Appendix. then T satisfies Monk’s Condition. an ). p. b. P . in L12 a (quantifier free) formula. A translation (simple translation) of L1 into L2 is a function. If A T . . . b.48 4. f is a Monk mapping. by Fact C. Let φ1 be the diagram of C and obtain φ2 from φ1 by substituting the variable xi for each constant ai and the variable yi for each constant bi . to obtain the desired isomorphism. . then any finitely generated structure over ˙ ˙ LT is finite. p. then f (x) = x. τ∀ . B T . suppose C contains a1 . [Monk. Proof. map ˙ the ai ’s into themselves and map the bi ’s into a sequence of elements of A which can stand in for the existentially quantified variables of φ3 .

. . The functional assumptions of τ are the sentences: ∀x1 .4. A is the set of elements of B which satisfy τ∀ . then τ extends to all formulae of L1 as follows: a. If τ is a translation of L1 into L2 . b. Notice that the existential and functional assumptions of a translation are sentences of L2 . b.5.2. A interprets all predicates and function symbols common to L1 and L2 in the same way that B does. . A translation from L2 into L1 induces a mapping from interpretations of L2 into interpretations of L1 . then τ (B) is the interpretation A of L1 such that: ˙ ˙ a. The existential assumption of τ is ∃xτ∀ (x) The functional assumptions of a translation say that the formulas which translate function symbols yield unique values within the relevant part of the domain when given values in the relevant part of the domain. τ (φ ∨ ψ) = τ (φ) ∨ τ (ψ) τ (φ ∧ ψ) = τ (φ) ∧ τ (ψ) τ (¬φ) = ¬τ (φ) τ (∀xφ) = ∀x(τ∀ (x) → φ) τ (∃xφ) = ∃x(τ∀ (x) ∧ φ) Definition 4. . ∧ τ∀ (xn )) → ∃y1 (τ∀ (y1 ) ∧ ∀y2 (τO (x1 . . b. Predicates and function symbols of L2 are translated into themselves. . If τ is a translation from L1 into L2 and B is an interpretation of L2 which satisfies the existential and functional assumptions of τ . SOME MODELS AND THEORIES 49 Definition 4.4.2. If τ is a translation of L1 into L2 . then a. Definition 4.2. y2 ) ↔ y1 = y2 )) where O is a function symbol in L1 but not in L2 . .2. The relevant part of the domain is the set of elements which satisfy the interpretation of universal quantifier. . The existential assumption of a translation says that that subdomain is non-empty. . xn .6. ∀xn (τ∀ (x1 ) ∧ .

. A1 and B1 . . . a. . . then T1 is τ -reducible to T2 iff for every model A of T1 there is a model B of T2 such that A = τ (B). T1 is uniformly τ -reducible to T2 iff for any models. A1 Proof.2.50 4. . Then.2. there exist models A2 and B2 such that A2 T2 and B2 T2 . . then T1 is also model complete. . b1 . Suppose φ(x) = ∃y1 . . . A interprets all predicates and function symbols in L12 in accordance with the translations assigned by τ . y) A The following condition on theories allows us to infer the model completeness of one from the model completeness of the other. T1 is reducible (simply reducible) to T2 iff there is a (simple) translation. c. yn )) iff A2 τφ (x) Theorem 4. b. ∃yn φ (x) where φ (x) is a conjunction of atomic formulae and negations of atomic formulae. . . . b1 . If T2 is model complete. . x ∈ A1 . τ is a simple translation from LT1 to LT2 . τ . for which T1 is τ -reducible to T2 . ∃yn (τ∀ (y1 ) ∧ . . of L1 and any sequence. for bi ∈ A φ(x) iff A2 τφ (x) τφ (x. Suppose that T1 is τ -reducible to T2 and that A1 = τ (A2 ).9. Then A1 φ(x) iff A1 iff A1 iff A2 iff A2 ∃y1 . . . . such that A1 T1 and B1 T1 . .7. and A2 ⊆ B2 A1 = τ (A2 ) and B1 = τ (B2 ). Lemma 4. y1 . y1 . .2. . for any primitive formula.8. THE PURE THEORY OF CLASS SIZES c. . and T1 is uniformly τ -reducible to T2 . yn ) ˙ φ (x. φ. If τ is a translation from L1 to L2 . . . A P (x) iff B y = O(x) iff B τP (x) τO (x. and A1 ⊆ B1 . Definition 4. ∧ τ∀ (yn ) ∧ φ (x. bn ) ∃y1 . . bn ). ∃yn φ (x.

has the following axioms (i) The axioms for abelian groups with identity. where A1 = τ (A2 ) and B1 = τ (B2 ) Since then So and B1 B2 A2 A1 φ(x) τφ (x) τφ (x) φ(x) by assumption by lemma 4. φ: If then B1 A1 ˙ φ(x) for x ∈ A φ(x) Since T1 is uniformly τ -reducible to T2 .4. Definition 4. (2). SOME MODELS AND THEORIES 51 Proof. it is enough to show that given models A1 and B1 of T1 . where A1 ⊆ B1 and a primitive formula.8 We shall now define several theories. there are models of T2 . A2 ⊆ B2 .5d in the Appendix. The theory of cancellable abelian semigroups with identity consists of (1). .10. a. (ii) The axioms for simple order.2.2. The theory of abelian groups with identity has the following axioms (1) (2) (3) (4) x + (y + z) = (x + y) + z x+y =y+x x+0=x ∃y(x + y = 0) b.2. and (3) above and: (4 ) x+y =x+z →y =z c. By Fact C. Zg . all more or less familiar. which will serve in showing that our theory of size is model complete. The theory of Z-groups.2.8 since T2 is model complete by lemma 4. The axioms of simple order are: (5) (6) (7) (8) x≤y∧y ≤z →x≤z x≤y∧y ≤x→x=y x≤x x≤y∨y ≤x d.

. The theory of Z-groups modulo I. axiom (12) from the theory of N-semigroups. Zgm. [Chang. p. b. . a. . and (iv) The additional axiom: (12) 0≤x f. where ny stands for: y + ··· + y n times e. ∨ ny = x + (n − 1)) for each positive n. +. p. THE PURE THEORY OF CLASS SIZES (iii) The following additional axioms: (9) (10) (11) y ≤z →x+y ≤x+z 1 is the least element greater than 0 ∀x∃y(ny = x ∨ .2. (ii) The axioms for simple order (iii) Axioms (10) and (11) of Zg. The theory of N-semigroups is model complete. consists of the following axioms: (i) The axioms for abelian groups with identity. Zg is the complete theory of Z.11. (iii) Axioms (9) and (10) of Zg.291] b. ≤ . a. The theory of N-semigroups has the following axioms: (i) The axioms for cancellable abelian semigroups with identity.[Robinson] Theorem 4. (ii) The axioms of simple order.52 4. 1.12. Zg is model complete.2. 291] Fact 4. The theory of Z-groups modulo I is model complete. and (iv) The following additional axioms (13) (14) (y ≤ z ∧ x ≤ x + z) → x + y ≤ x + z) x≤I The theory of Z-groups is taken from [Chang. 0.

where τ is the following translation.12) and Theorem 4. we shall show that every model of PCA is the τ -image of a model of Zgm. Specifically. stack up ω many copies of the model.2b.9 to show that PSIZE is model complete. Moreover. we use Theorem 4. the (rough) divisibility of the elements in the semigroup will be carried over to the group. It is clear from the construction in [Kurosh] that if the semigroup is ordered.9. an individual constant. besides the constant symbols in the original theory.) In the next section. Consequently. 53 a. 4. the theory of N-semigroups is uniformly reducible to the theory of Z groups by the translation: τ∀ = 0 ≤ x Since the latter is model complete. I.2.2.11). by 4.2. so is the former. the abelian group in which it is embedded may also be ordered and that the elements of the semigroup will be the positive elements of the group. we shall reduce it to Zgm.2. by 4. Every abelian semigroup with cancellation can be isomorphically embedded in an abelian group [Kurosh. . 44-48].4.3. PCA IS MODEL COMPLETE Proof. by reducing it to the theory of Z-groups modulo I. The theory of N-semigroups is model-complete in this language. assigning interpretations in the obvious way.2. We claim that the theory of Z-groups modulo I is uniformly reducible to this new theory by the following translation: τ∀ = x ≤ I τ+ = x + y = z ∨ x + y = I + z (The construction: given a model of Zgm.3 PCA is model complete To show that PCA is model complete. b. pp. the theory of Z-groups with addition taken modulo some constant (see 4.2. The result is an Nsemigroup and the original model is isomorphic to the first copy of itself. consider the theory of N-semigroups in the language which contains.9. The model completeness of PCA then follows from the model completeness of Zgm (Fact 4. Chapter 5 uses Monk’s Theorem to show that CA is model complete. First.

(Sm(x. y .54 4.2. A. Fact 4. z ) x ˜ ˜ (f) ZA (x = 1) iff A Unit(x) Fact 4.8 establishes that ZA Zgm. 4. y. ZA .3. ) . on the basis of the intervening facts: 4.3. of PCA. Let τ be the translation from LPSIZE to LZgm where: τ∀ = x = x τSum(x. then ZA is the interpretation of LZgm in (a) (b) (c) (d) (e) ZA ˙ Z˙A = A ZA (x ≤ y) iff A (x < y ∨ x = y) ZA (I) = A(I) ZA (0) = A(0) (x + y = z) iff A Sum(x. y) abbreviates ∃zSum(x.4 deals with the corresponding model ZA .6 verifies some connections between A and ZA . If A which: PSIZE. z) or A ∃a∃wUnit(a) ∧ Sum(˜.3. The following are theorems of PSIZE.3 deals with the model A of PSIZE. we can construct a model. w) ∧ Sum(w.3.2 gives Sum a functional interpretation. z).3.3. THE PURE THEORY OF CLASS SIZES Definition 4. 4. of Zgm directly: Definition 4. Theorem 4.3.1. y.3j establishes that 4. a.z) = τUnit(x) = τx = ˜ τx<y = τ∅ = x+y =z∧x≤z x=1 x+y =I x≤y∧x=y x=0 Given a model.3.3.y.3.

w) ∧ Sum(w. y. z2 ) → (z1 = z2 = I) x ˜ ˜ ∃z1 Sum(x.2 together with (a) and (b): (a) (b) ZA ZA y = −x iff ZA z = x − y iff ZA x+y =0 z = x + −y Remark. y1 . w)) ˜ x=x ˜ x<y↔y<x ˜ ˜ x<y∧y <y →x<x ˜ ˜ Sm(x. z2 ) → z1 = z2 Sum(x. y. y. We shall state these in terms of the model ZA . y ) x ˜ Sum(x. ∅) A Sum(∅. there is a unique y such that: ZA x + y = 0 Proof. w) → (ASSOC2) ∃w2 (Sum(y.4. y) ∨ Sm(˜. z) → Sum(˜. In addition to the theorems of PSIZE listed in 4. z. x) z ˜ Sum(x. z1 ) ∧ Sum(˜. Given a model A of PSIZE: for each x in Z˙A . ∅) ZA y = 0 x + y = 0 iff y = 0. y . z. y2 . PCA IS MODEL COMPLETE 55 (a) (b) (c) (d) (e) (f) (g) (h1) (h2) (i) (j) Sum(x. The proofs are elementary. z2 )) x ˜ ˜ Proof. y. a. z .3. y.3. z) → Sum(x.2e by IDENT by UNIQ− .3. z) → y1 = y2 (UNIQ+ ) (UNIQ− ) Sum(x. y. z1 ) ∧ Sum(x. z) ∧ Sum(x.3. as follows: Definition 4. Given a model A of PSIZE. w2 ) ∧ Sum(x. y. z1 ) ↔ ¬∃a∃w∃z2 (Unit(a) ∧ Sum(˜. y. w1 ) & Sum(w1 . w2 . ZA is the expansion of ZA induced by the definitions in 4. which interprets two additional operators. a. by 4.4.3. y. y . If x = 0. y ) ˜ ˜ Sum(x. an expansion of both A and ZA . then ZA ⇒ If then But So ZA 0 + y = 0 A Sum(∅. ∅.3. we require a battery of tedious facts about the model ZA .

3. a.3. Sm(a. b − c. since there is a unique unit and A But ZA ⇒ If then and But So So So ⇐ If then and So A A A ZA (*) Sum(˜. c) ∧ 0 < a → Sum(−c. a. ˜ ∅) A A A A ˜=I ∅ w=a ˜ Sum(˜. since A Sum(x.6. THE PURE THEORY OF CLASS SIZES b. y) x x + y = 0 iff A (*). a. where Unit (a) x UNIQ+ . there is a unique y which satisfies (*) Sum(˜. a + b) b. a) x ˜ ˜ Sum(˜. Sum(a.56 ⇐ Obvious 4. b. c ≤ b ↔ Sum(c. c) ∧ 0 < b → Sum(a.5. If x > 0. −a) −(x + y) = −x + −y −(x − y) = y − x (x + y) − y = x −(−x) = x x = 0 ∧ x < y → −y < −x . b. ZA satisfies the following: a.3d and h2 by 4. −b) Sum(a. ZA satisfies the following: (a) (b) (c) (d) (e) Proof. b) c. y). Omitted.3. y . b. a) x ˜ ˜ Sum(˜. 0) by COMP and UNIQ− by 4. Fact 4. −c.3h2 by COMP by 4.2e Fact 4. b) ↔ Sum(a. y. b. y .3. a. y . w) x ˜ A Sum(w.3. ˜ a ∅) x+y =0 ZA x + y = 0 A Sum(˜.

5c by (a) by IDENT by MONOT and UNIQ+ . b) Sum(c. −c) a=0 Sm(a. a.3. b.5c by 4. c) 0 ≤ (b − c) ↔ b ≤ c b≤c c≤b Sum(c. b. −b) if Sm(a.2e (⇐)Obvious. b − c.5c by IDENT by MONOT and UNIQ+ by MIN by (a) and UNIQ+ by 4. −c. w. a + b) 57 by 4. b) → (Sm(−a. (⇒) Suppose So So So Sm(a. since 0 < b Sum(c. b) Sum(a. b − c. 0.3.3. w) for some w (a + b) = w Sum(a. −b) ∨ b = −a) Proof. c) 0<b c = (a + b) c − b = (a + b) − b c−b=a −b=a−c Sum(a.3.3. b) for some w b=c+w b − c = (c + w) − c b−c=w Sum(c. −c) a=0 a < c. ¬Sm(a. (⇒) Suppose then So So So So (⇐) Suppose But So So c. PCA IS MODEL COMPLETE d. b) by EXIST− by (a) and UNIQ+ by 4.4. b. b. Suppose and then So So So So But if then And if then Sum(a.

−b) (a + b) = 0 b = −a by by by by by by by by 4. (−a + −b) − (−a). We need the following lemma.6a and UNIQ+ by 4. −b. b) b = (a + b) − a.4 IDENT EXIST+ ¬Sm(a.6a and UNIQ+ . −a + −b) Sm(−a.5c (a) (b) TRICH 4. Lemma 4. Suppose But So So So But if then So So But So So And if then − c < −a −c≤a ˜ Sm(a. (a + b) < a.5e 4. a + b). −c) 4.3.6b by 4.3. a + b) ∃w(Sum(c. b) Sum(b. b) and c ≤ b We know and So So and So Sum(b − c. a + b) w =c+a a + b = (b − c) + (c + a) a + b = (a + c) + (b − c) by 4. Sum(−a. The only real difficulty arises in verifying that addition is associative.3. a.6a by ASSOC2 by 4.3.3. −a + −b).3.3.5c We can now show that ZA Zgm. ¬Sum(a. Sum(−a.58 So So But So d. THE PURE THEORY OF CLASS SIZES by by by by 4. (a + b) = 0 − a < −(a + b).3. a.5a (b) 4. − a < −a + −b. When Sm(a.5e 4. We shall work through successively more general cases: a. w. ¬(a ≤ a + b).3. c. ZA (a + c) + (b − c) = a + b Proof.7.3. w) ∧ Sum(b − c.3. (a + b) − a. a) ˜ Sm(a. (−a + −b) − (−a) = −b.

4b 4. c − b. −c. −(a + c)) 0 < −a Sum(−(−(a + c)).6c by 4.6c 4. b. b) If So assume So and So So So Either Assume then So and So So Assume then and Hence and and So So d. b) and Sm(a. for otherwise case c applies.6d by 4. c) Sm(−a. −c. −a + −c) Sum(−a. b) by ASSOC by 4. Sm(a. −c. −c) or c = −a since ¬Sm(a.5d (c2) by (c2) by (c1) since Sm(a.3.5b by by by by by (c1) EXIST+ 4. b.4.3.3.3.3. −c.3. b − c) Sm(−a. −c. a + b) Sum(a + c. Whenever Suppose ¬Sm(a. b − c) Sum(a. −(−a)) Sum(a + c. c) and thus b < c Sum(b. a + b) (a + c) + (b − c) = a + b c ≤ b. b − c) Sum(−c. b − c. −c. Sm(a.3. b. PCA IS MODEL COMPLETE b. −c) Sum(−a.6a by 4. −c.3.6a .5a by 4. When Sm(a. Case (a) applies directly b<c a + c = (a + b) + (c − b) (a + c) − (c − b) = a + b (a + c) + −(c − b) = a + b (a + b) + (b − c) = a + b 59 by by by by Case (a) 4.3. a) Sum(−c. −(c − b)) Sum(b.6b MONOT 4.3. c) 0<c−b Sum(b. b).5b by 4. c) Case (b) applies ¬Sm(a.3. c) If So assume then So So So c.5c 4.3.3. a) c = −a a+c=0 a = −c Sum(a + c.

A PSIZE. ZA satisfies axiom (10) of Zgm because A satisfies the UNIT axiom of PSIZE.7: x + (y + z) = (x + y) + ((y + z) − y). −b) or b = −a b = −a a+b=0 b − c = (−a) − c = −a + −c = −(a + c) (a + b) + (b − c) = (a + c) + −(a + c) =0=a+b Sm(−a. The only axiom for abelian groups that needs further verification is associativity.3.5a 4.3.5d Theorem 4.3.7 = (x + y) + z. Similarly.3. ZA Zgm Proof. by 4. and PSIZE includes the same ordering axioms.5c The ordering axioms of Zgm are satisfied in ZA because ZA uses the same ordering as A. y. Axiom (13) of Zgm is: y ≤z∧x≤x+z →x+y ≤x+z If then Since then and So ZA x ≤ x + z A Sum(x. We prove this using lemma 4.3.3.5a 4. THE PURE THEORY OF CLASS SIZES Sm(−a. z. ZA satisfies axioms (12).8. and (14) of Zgm because A satisfies MIN and MAX.3. x + y) A x+y ≤x+z ZA x + y ≤ x + z by EXIST+ by MONOT .60 Either Assume then and 4. −b) − a + −b = (−a + −c) + (−b − −c) − (a + b) = −(a + c) + −(b − c) − (a + b) = −((a + c) + (b − c)) a + b = (a + c) + (b − c) by 4. by 4. x + z) ZA y < z A Sum(x.3. The divisibility of elements in ZA required by axiom (11) of Zgm is guaranteed by the fact that A satisfies the divisibility principles of PCA.6d So Assume then So So So by by by by case c 4.

. We already know that PCS PCA. we PCA is τ -reducible to Zgm. a.1(b).2. So. which follows from (2) by 3. (3a) (3b) If PCAIf is consistent. (Formally. a.4 Remarks on showing that PCA axiomatizes PCS To prove that PCA ≡ PCS. we may conclude that PCA is model complete. then PCSIf is consistent. Immediate from (a) and 4. we would have to define EXACTLYn in terms of units rather than atoms. Every infinite completion of PCA is consistent with PCS. EXACTLYn is true in SA .9. for some f .4.3. where A is the standard finite interpretation of LC< containing n atoms. the finite completions of PCA are consistent with PCS. (1) (2) PCA PCS Every completion of PCA is consistent with PCS. EXACTLYn . we could follow the method outlined at the end of chapter 3 for showing that CA ≡ CS.) Letting PCAI = PCA + INF and PCSI = PCS + INF.3. then T ≡ PCAIf . But PCA. b.9. PCA satisfies Monk’s condition. (2b) is a consequence of (3a) and (3b).4. 4. The finite completions of PCA are the theories PCA. b. If T is a completion of PCAI. REMARKS ON SHOWING THAT PCA AXIOMATIZES PCS 61 So.2. Apply Theorem 4. But (2) is equivalent to the conjunction of (2a) and (2b). The reduction is uniform since each model A of PCA has the same domain as its Zgm-model. Theorem 4. So.1b. (2a) (2b) Every finite completion of PCA is consistent with PCS. so we need only prove (1). Proof. PCA is model complete.

we would have to prove analogues of 3. then PCAIf is complete. but tedious. then PCAIf is has a prime model. .3.13 for PCA and PCS. because every completion of PCAI entails PCAIf for some total f . the construction could be duplicated in this simpler case. (4) If f is total. so only (5) remains to be shown.62 4. Alternatively.3. It’s apparent that the size models of the prime models for CAIf would do nicely. To prove (3b). This seems straightforward. The trick is to show that enough axioms about size have been incorporated in PCA to establish the entailments among MOD statements. We will not construct prime models for the extensions of PCAI.9 through 3. THE PURE THEORY OF CLASS SIZES To prove (3a). it is sufficient to demonstrate (4). But PCA is model complete. (5) If f is total and congruous.

1. There is an isomorphic embedding. and C is a finitely generated Theorem 5. SA SC SB (A) ⊆ SB . Monk’s Theorem applies directly to SB (A).1. DISJ∪ and REP< then allow us to find elements with those sizes that fit together in the right way. But. we want to prove 5. SA is not a submodel of SB . B CA. SB .1). let SB (X) = the submodel of SB whose domain is { σB [x] | x ∈ X } Then. A substructure of B. where ˙ ˙ f (x) = x if x ∈ C ∩ A The Monk mappings for PCA can serve as a guide in constructing Monk mappings for CA.1. clearly.3.2.3: Fact 5. 63 . The existence of Monk mappings for PCA tells us that we can find elements with the right sizes.1.1.1. and SB (C). Assumption 5. and SB (C). Strictly speaking. so we cannot apply Monk’s Theorem directly. so we may conclude 5. given Assumption 5. a finitely generated submodel of SB . A ⊆ B . f : C → A.1. g : SC → SA .1 Model completeness of CA We shall show that CA is model complete by showing that it satisfies Monk’s criterion (see 4. So.3.5 Completeness of CA 5.2. There is an isomorphic embedding.1. CA. where ˙ ˙ g(σB [x]) = σA [x] for all x ∈ C ∩ A.

64

5. COMPLETENESS OF CA

That is to say, the sizes of elements in C can be embedded in the sizes of elements in A. It remains to be shown that the elements of C themselves can be mapped into A by a function which preserves boolean relations as well as size relations. C is a finite, and hence atomic, boolean algebra, though its atoms need not be atoms of B; indeed, if B is infinite, there must be some atoms of C which are not atoms of B since the union of all atoms of C is the basis of B. ˙ ˙ Definition 5.1.4. d is a molecule iff d ∈ C ∩ A and no proper subset of d ˙ ∩ A. ˙ is in C ˙ ˙ C ∩ A is a boolean algebra whose basis is the same as the basis of C. So every atom of C is included in some molecule. The embedding f has to map each molecule to itself. Moreover, f has to be determined by its values on C since f must preserve unions. In fact, the atoms of C can be partitioned among the molecules. So, if d = b1 ∪ . . . ∪ bn where d is a molecule and b1 , . . . , bn are the atoms of C contained in d, then d must also be the (disjoint) union of f (b1 ), . . . , f (bn ). If this condition is satisfied, f will preserve boolean relations. f must also select images with appropriate sizes. Proof. Proof of 5.1.2 Given a molecule, d, let b1 , . . . , bn be the atoms of C contained in d. For each bi , let ci be some member of g(σB [bi ]) These elements ˙ ˙ will be elements of A, since g yields sizes in A whose members are in A. These elements have the right sizes, as we will show below, but they are in the wrong places. We have no guarantee that they are contained in the molecule d. So we ˙ still need to show that there are disjoint elements of A, a1 , . . . , an whose union is d and whose sizes are the same as those of b1 , . . . , bn , respectively. Well, Suppose So So So So d = b1 ∪ . . . ∪ bn C Sum(b1 , . . . , bn , d) SC Sum(σB [b1 ], . . . , σB [bn ], σB [d]) SA Sum(g(σB [b1 ]), . . . , g(σB [bn ]), g(σB [d])) SA Sum(c1 , . . . , cn , d)

since each ci ∈ g(σB [bi ]) and d ∈ g(σB [d]) = σA [d]. So, the existence of a1 , . . . , an , as above, is guaranteed because A DEF+ . Now, let f (bi ) = ai for each bi in the molecule d. Repeating this procedure ˙ for each molecule yields a value of f for each atom of C. Finally, if x ∈ C is non-atomic, then x = b1 ∪ . . . ∪ bk where each bk is atomic. So let f (x) = f (b1 ) ∪ . . . ∪ f (bk )

5.1. MODEL COMPLETENESS OF CA

65

f satisfies the requirements of Theorem 5.1.2: Boolean relations are preserved by f because the function is determined by its values on the atoms of C; f maps elements of C ∩ A into themselves because the set of atoms contained in ˙ each of these molecules is mapped into a disjoint collection of elements of A whose union is the same molecule; so, we need only show that f preserves size relations. To do so, we invoke lemma 5.1.6, below: (a) (b) (c) (d) C C C x < y iff A x ∼ y iff A f (x) < f (y) f (x) ∼ f (y) Sum(f (x), f (y), f (z)) Unit(y)

Sum(x, y, z) iff A C Unit(x) iff A

Proof of (a): (The others are similar).

C

x < y iff iff iff iff

SB σB [x] < σB [y] SA g(σB [x]) < g(σB [y]) SA σA [f (x)] < σA [f (y)] A f (x) < f (y)

by 5.1.6

So, we may conclude: Theorem 5.1.5. CA is model complete. ˙ Lemma 5.1.6. For all x in C, σA [f (x)] = g(σB [x]) Proof. Suppose x = b1 ∪ · · · ∪ bn ˙ where each bi is an atom of C. So So So But So But So So B Sum(b1 , . . . , bn , x) SB Sum(σB [b1 ], . . . , σB [bn ], σB [x]) SA Sum(g(σB [b1 ]), . . . , g(σB [bn ]), g(σB [x])) g(σB [b]) = σA [f (b)] SA Sum(σA [f (b1 )], . . . , σA [f (bn )], g(σB [x])) A Sum(f (b1 ), . . . , f (bn ), f (x)) SA Sum(σA [f (b1 )], . . . , σA [f (bn )], σA [f (x)]) g(σB [x]) = σA [f (x)] since B DISJ∪

since SA ⊆ SB by the choice of f (b). since A since SA DISJ∪ UNIQ+

66

5. COMPLETENESS OF CA

5.2

Prime models for CA

For each total, congruous remainder function, f , we want to find a prime model, Qf , for CAIf . All of these prime models can be defined over the class, Q, of sets near quasi-congruence classes (see section 3.2). For different remainder functions, we need to assign different size relations over Q. Section 5.2.1 defines the structures Qf and verifies that each satisfies the respective theory, CAIf . Section 5.2.2 defines, for each model of CAIf , a submodel, or shell. Section 5.2.3 shows that Qf is isomorphic to the shell of any model of CAIf .

5.2.1

The standard protomodels

The construction here is a more elaborate version of the construction of Q in chapter 3 (see 3.2.19). Q turns out to be Qf , where f (n) = 0 for all n. As in the case of Q, the models Qf and their copies in arbitrary models of CAI are unions of chains. The sizes assigned to elements of Q to induce Qf for a total, congruous remainder function, f , are more elaborate than those used in the definition of Q (see 3.2.22), but they are employed in substantially the same way: Definition 5.2.1. a. A size is an ordered pair ρ, δ , where both ρ and δ are rational. b. If θ1 = ρ1 , δ1 and θ2 = ρ2 , δ2 are sizes, then (i) θ1 < θ2 iff ρ1 < ρ2 or ρ1 = ρ2 and δ1 < δ2 . (ii) θ1 + θ2 = ρ1 + ρ2 , δ1 + δ2 (cf 3.2.22) To assign sizes for Qf we rely on the representation of sets in Q defined in 3.2.24. Qf , unlike Q, assigns different sizes to the n-congruence classes for a given n. Definition 5.2.2. If f is total and congruous, then a. If x is an n-congruence class, x = [nk + i], then θf (x) =
1 n−f (n) n, n 1 −f (n) , n , n

,

if i < f (n), if f (n) ≤ i.

b. If x ∈ QCn , so that x is the disjoint union of n-congruence classes, x1 ∪ . . . ∪ xk then θf (x) = θf (x1 ) + · · · + θf (xk )

|∆1 (x)| − |∆2 (x)| 67 Intuitively. and δ(x) . If f is total and congruous. for total and congruous f . The desired models of CAI may now be defined: Definition 5. all n-congruence classes are assigned sizes 1/n. but δ is no longer 0 in all cases. then Qf. θ(x). then θf (x) = 0. If f is total and congruous.n BASIC. PRIME MODELS FOR CA c.4. then the standard protomodel for f is Qf .2. z) iff θf (z) = θf (x) + θf (y) Qf (cf 3. |∆2 (x)| = θf (C(x)) + 0.24. Proof. |∆1 (x)| − 0.n is the submodel of Qf whose domain is Qn . If x ∈ Q. |x| d.2.2.27) To verify that the structure Qf is a model of CAIf . we exhibit each such model as the union of a chain of models. Definition 5.3.n θf (x) ρf (x) δf (x) for for for for Qn .29 by substituting: Qf.2. so that x can be represented as (C(x) ∪ ∆1 (x)) − ∆2 (x) as in 3. 1 Sum(x.5. then θf (x) = θf (C(x)) + θf (∆1 (x)) − θf (∆2 (x)) = θf (C(x)) + 0. then Qf. Instead. The proof can be obtained from the proof of Theorem 3. δ .2. y.5. the first f (n) n-congruence classes are each one atom larger that the remaining (n − f (n)) n-congruence classes. as in Q. ρ(x). where ˙ Q{ = Q Qf x < y iff θf (x) < θf (y) Qf x ∼ y iff θf (x) = θf (y) Qf Unit(x) iff θf (x) = 0.2.2. Fact 5. If f is total and congruous and n > 0. If x is finite.

˙ Definition 5. Suppose A BASIC.2 . But this sequence does harbor a chain of models: Fact 5. .68 5. Then an (n. so Qf. In any case.6. x ∈ A.2 . then Sum(xi . since size relations are determined by the representations of each set. 4}. xj ). then all of the 6-congruence classes are common.3 is not an extension of Qf. then [6n] and [6n + 1] are the only charmed 3congruence classes.6 [2n] ∼ [2n + 1] Suppose. x. Letting n = 2 and m = 3. If f (6) = 2. as C(x).7.2 [2n] is one atom larger than [2n + 1] Here. (ii) If m < i < j ≤ n. The sequence Qf. 3. If 0 < i ≤ m. xi is called a charmed n-factor of x. . for any f . The other elements of Qf. Suppose that f (2) = 0. we want to show that the 2-congruence classes have the same size relations in Qf. 5}. so Qf. since f is congruous.2. For example Qf. The xi are approximately the same size: (i) If 0 < i < j ≤ m. where a. then xi ∼ xj . xi is a common n-factor of x. x is partitioned among n pairwise disjoint sets which are roughly the same size: each of the first m is one atom larger than each of the remaining n − m. . then all of the 3-congruence classes are charmed except for [6n + 4] and [6n + 5]. for i > m. If f (6) = 0. .m! . If n > m. ∪ xn b. however that f (2) = 1.n does not constitute a chain of models.2 [2n] ∼ [2n + 1] Since f is congruous.m)-partition of x in A is a sequence x1 . then xi ∼ xj .n! ⊆ Qf. xn . (iii) If 0 < i ≤ m < j ≤ n. . f (6) ∈ {0. Proof. then xi ∩ xj = ∅. so that Qf. COMPLETENESS OF CA The following notion is helpful in understanding our constructions.6 [2n] is one atom larger than [2n + 1] So it goes in general. If f (6) = 4. so that Qf.2. ∆1 (x).6 will then fall into place. In any case. a.6 as they do in Qf. If 0 < i < j ≤ n. . In other words. . [2n] contains exactly one more charmed congruence class than [2n + 1]. and 0 ≤ m < n. for any atom a. to establish this by example rather than by formal proof. and ∆2 (x). then Qf. It will be clearer. [2n] will include the same number of charmed 3-congruence classes as [2n + 1]. and easier. and c. 2. f (6) ∈ {1. x = x1 ∪ .

2. d. A.7 allows us to regard Qf as the union of a chain of models: Fact 5. Fact B. Qf b. If f is total and congruous. (b). we must find a smallest submodel.n MODn.2 Shells of models To embed the model Qf into an arbitrary model. We can aim for this by placing in B three disjoint sets. x22 . If f is total and congruous. of A which satisfies CAI. x0 .5). call it x0 . Qf c. of CAIf .f (n) . since the symbol I must refer to the same set in the ˙ submodel as it does in the model. then a. Immediate from (a). if the basis of A is in B and B CAI. b.f (n) is an existential sentence.1 . But. Each n-congruence class can be partitioned into k nk-congruence classes. . to include in B.5. then Qf = n>0 69 Qf.9.2. Pick such a pair.f (n) . also guarantees that x0 is divisible by 3: the three unions x11 ∪ x21 . PRIME MODELS FOR CA Fact 5. x11 . ˙ must contain two disjoint sets of roughly the same size whose union is then B ˙ the basis of A. (See Appendix. it is preserved under extensions. x1 and x2 . by definition. Qf. Whether these are exactly the same size or differ by an atom depends on whether A MOD2.2. ˙ B must also satisfy ADIV3 . Insuring that x1 and x2 are each divisible by 3. the exact size relations will be determined by which MOD principles are satisfied in A.2.n! Fact 5. BASIC ADIVk . x12 ∪ x22 . Clearly. so it is preserved under unions of chains. c. x12 . B. and x23 whose union is x2 . The existence of such sets is assured because A ADIV3 . for n > 0 CAIf 5. Qf d.0 or A MOD2. Since MODn. for k > 0 MODn. the basis of A. a. Qf Proof. BASIC is a universal-existential theory.2. must be included in B. x21 . Again.8. and x13 ∪ x23 will be roughly the same size and will exhaust x0 . and (c). and x13 whose union is x1 and another three disjoint sets.

we show that the shell of A is isomorphic to Qf . nk . where f is a total. then λ(i) = ni . This great tree of sets will not form a boolean algebra. d. to each node. Let A 0 CAIf .) b. Definition 5. . to be a prime model. and let a 0 . nk . Third.e. where k > 0 and for all i ≤ k. Given A = Af remainder function: a. In the next section. the smaller the set it bears and the greater the number of successors among which this set will be partitioned. L(λ) is k. A node is a finite sequence n1 . we present the construction which.2.11. of λ. (The variables λ and κ range over nodes. . . λ 0-extends κ iff λ extends κ and. the set of nodes on which we shall hang both the components of the successive partitions described above and the atoms of the submodel being constructed. We leave the solution of this problem to the formal construction. bearing sets. If λ = n1 . . Nodes constitute the vertices of an infinite tree in which 0 is the root and λ dominates κ iff κ extends λ. ni < i. At the same time we shall assign an atom to each node. λ. We cannot correct for this problem ˙ by including in B all atoms of A: A may have uncountably many atoms while B. dividing each set introduced at stage n into n + 1 roughly equal subsets at stage n + 1. . . nor will it be closed under finite unions. This will produce an infinite tree. must be countable. assigns a node set Aλ and a node atom. nk . for 1 ≤ i ≤ L(κ). λ(i) = κ(i).70 5. A boolean algebra could be obtained by including both the node sets and their finite unions. .2.10. (ii) If 1 ≤ i ≤ k. λ(i) = 0. m c. the tree: Definition 5. . congruous be any atom of A. aλ . . a. i. . given a model A of CAIf . The number of immediate descendants of a node grows as the depth of the node increases. The formal proof proceeds as follows: First. λ extends κ iff L(λ) > L(κ) and. for L(κ) < i ≤ L(λ). . First. which is what we are looking for. (iii) λ • m = n1 . then (i) The length. COMPLETENESS OF CA We can continue this process indefinitely. or depth. Second. but this would still not be an atomic boolean algebra. we define the shell of A as the submodel of A generated by the collection of node sets and node atoms. be the basis of A. we define a tree. . The deeper a node is in this tree. We shall now assign a set to each node by repeatedly partitioning the basis of A.

let aλ•i be any atomic subset of Aλ•i . choose Aλ•0 so that it contains aλ . m > 0.k predicate. k is an integer (see Definition 3. If 0 < i ≤ m. suppose that each common n-factor has k charmed mfactors and m − k common m-factors. The sets Aλ will be referred to as node sets and the atoms aλ as node atoms. By (a).2. k)-partition. .5. Let m = L(λ) and let k= f ((m + 1)!) − f (m!) m! 71 Since f is congruous.k (x) says that x has an (m. Proof. In either case. All common n-factors are the same size and satisfy the same Modm. Fact 5. Suppose that Aλ and aλ have been chosen. a.12 guarantees that such partitions exist. x has an (m. n > 0. k) partition of Aλ if Aλ is common or an (m + 1. In all. b. Every charmed n-factor of A has an (m. Each of the common n factors has k charmed m-factors and each of the charmed n-factors has k + 1 charmed m-factors.12. and k= Then a. k)-partition iff A Modm. .3. . k + 1)-partition.k (x) f (nm) − f (n) n ˙ b. This is always possible because Aλ contains aλ . the charmed m-factors of the n-factors are the charmed nm factors of A and the common m-factors of the n-factors are the common nm factors of A. Suppose A CAIf . k)-partition. Aλ•m be an (m + 1. Let Aλ•0 . A has f (n) charmed n-factors and n − f (n) common n-factors. ˙ c.2. c.2. Every common n-factor of A has an (m. PRIME MODELS FOR CA b. Let aλ•0 = aλ . there are (n − f (n))k + f (n)(k + 1) =nk + f (n) .7e). Fact 5. k + 1) partition of Aλ if Aλ is charmed. Partitioning each of the n-factors into m subsets of roughly the same size yields an nm-partition of A. So. . Modm.

aλ ⊂ Aκ iff λ extends κ. COMPLETENESS OF CA charmed m-factors among the n-factors of A. Immediate from (b). There are n! nodes (and. by (a) again. then Aλ•i ∩ Aλ•j = ∅.13 should clarify this constuction. then n! = 1. . Any two node sets of the same depth are disjoint. For any n.72 5. If Aλ = Aκ . there are f (nm) charmed nm-factors of A. so. there are (n!)(n + 1) = (n + 1)! node sets of depth n + 1. and Aλ 0 is the basis of A. f (n!))-partition of the basis of A. Then (k ) also holds for n + 1: each node set of depth n has n + 1 immediate descendants.2. If i = j. g. d. node sets) of depth n.2. b. then κ = λ. f. Each node set is the disjoint union of its immediate descendants. since every charmed n-factor is one atom larger than each common n-factor. So f (nm) = nk + f (n) k = (f (nm) − f (n))/n and c.0)-partition of itself. e. i. Aλ ⊂ Aκ iff κ extends λ or κ = λ. Fact 5. a. So (k ) holds because any set is a (1. the node sets of depth n form an (n!. hence. Proof. h.13. The facts listed in 5. f (n!) = 0.: If n = 1. c. But. k. Each node set is the disjoint union of all of its descendants at any given depth. All of them can be established by induction on the depth of nodes. aλ = aκ iff one of λ and κ 0-extends the other. Assume that (k ) holds for n. Every node set contains infinitely many node atoms. We demonstrate (k) by induction on n. j.

the shell of A generated from {Aλ . a model of CAIf . a set of node sets and node-atoms produced by the construction above. aλ . the n + 1 factors of the n-factors of the basis form an ((n + 1)!. from a model. where k= f (n!(n + 1)) − f (n!) n! f ((n + 1)!) − f (n!) n! = (substituting n! for n and n + 1 for m). congruous remainder function b. For the remainder of this chapter.2. aλ }. a total. of CAIf . Aλ . aλ }. there are (k + 1)f (n!) charmed n + 1 factors from the charmed n-factors and k(n! − f (n!)) charmed n + 1 factors from the common n-factors In all. So. A. aλ } . That is to say. of A which is isomorphic to Qf . f . Definition 5. and node-atoms. then . ˆ d. B.12. (k ) holds for n + 1. each charmed n-factor has an (n + 1.2. A. f ((n + 1)!))partition of the basis. k)-partition. PRIME MODELS FOR CA 73 Furthermore. {Aλ . Then. Suppose f is total and congruous.1. that A CAIf and Aλ and aλ are the node sets and node-atoms of A produced by construction 5.12. the shell of A is the submodel of A generated by {Aλ . c. A. By Fact 5.14. k + 1)-partition and each common n-factor has an (n + 1.5. we will regard as fixed: a. we can now construct a submodel.2. f (n!) of the n-factors are charmed and n! − f (n!) are common. the number of charmed n-factors is: (k + 1)f (n!) + k(n! − f (n!)) = kf (n!) + f (n!) + n!k − f (n!)k = f (n!) + n!k = f (n!) + f ((n!)(n + 1)) − f (n!) = f (n!)(n + 1)) = f ((n + 1)!) So. Given a collection of node sets.

18 obtain for A-nearness. a.15 through 3. ˆ (⇐)A must contain finite unions of node sets as well as A-finite sets. then x has a unique representation as (C(x) ∪ ∆1 (x)) − ∆2 (x) where C(x) is a quasi-nodal set disjoint from ∆1 (x) and including ∆2 (x).2. COMPLETENESS OF CA ˆ To show that A Qf . x ∈ A iff x is A-near some quasi-nodal set of A. 3. then x is A-near y iff both x − y and y − x are A-finite sets.74 5.2.18) ˙ ˆ Still following in the footsteps of chapter 3. x is an A-finite set iff it is a finite set of node atoms. we need a sharper characterization of the elements ˆ of A. .16.14-3.2. each of which preserves A-nearness to quasi-nodal sets.15. Recall from 3. A is an atomic boolean algebra whose atoms are the node-atoms aλ . Proof. b. If x ∈ A and y ∈ A. Fact 5. finite sets of node-atoms correspond to the finite subsets in Q. (⇒)A is generated from node sets and node atoms via the boolean operations.2. Analogues of 3. ˆ b.2. ˙ ˆ a.2. ˆ a. c. so it must also contain sets obtained by adding or removing A-finite sets from quasi-nodal sets.24 that each member. ˙ ˆ c. If x ∈ A. of Q has a unique representation as (C(x) ∪ ∆1 (x)) − ∆2 (x) where C(x) is a quasi-congruence class. finite unions of node sets correspond to the quasi-congruence classes. x is a quasi-nodal set of A iff it is the union of finitely many node sets (iff it is the union of finitely many node sets at a given depth). (cf. x. Definition 5. we can characterize A as the collection of sets A-near quasi-nodal sets.2. both of which are A-finite sets. ∆1 (x) and ∆2 (x) are finite sets and ∆1 (x) ∩ C(x) = ∅ ∆1 (x) ∩ ∆2 (x) = ∅ ∆2 (x) ⊆ C(x) ˆ We can obtain a similar representation for elements of A: the node sets play the role of (some of) the congruence classes.

nk .18. In effect.2. . If L(λ) = k. we are performing the construction 5. then k ι(λ) = i=1 (i − 1)!ni . we are stipulating which (n. b. Examples.12 on Qf .0 Q 0. m)-partitions to use at each level.2. . ι(λ) = the least n ∈ Qλ .2.2.21 exactly. This latter condition guarantees that the shell of Qf will be Qf itself. a. we first describe Q in terms of node sets and node atoms. but with two differences: first.23). 5. PRIME MODELS FOR CA b. The proof parallels that of Theorem 3.17. Then a.2 Q 0.2.2 = [n] = [2n] = [2n + 1] = [6n + 1] = [6n + 4] = [6n + 5] Definition 5. qλ = {ι(λ)} Fact 5.2.1 Q 0.2. we are selecting node atoms so that every singleton in Q is the node atom for some node.3 The embeddings To embed Qf into A. second.1. Let C(x) be the quasi-nodal set which is A-near x (see 3. Definition 5. 75 c. a.2. Suppose λ is a node. Let ∆1 (x) = x − C(x). .0 Q 0. If λ = n1 . and.5. Q0 Q 0. then Qλ = [k!n + m] where k m= i=1 λ(i)(i − 1)! b.0. The depth of Qλ is L(λ). .19.1. let ∆2 (x) = C(x) − x.

We may. That is to say.76 5. Definition 5. that 5. suppose first that x ∈ QC. b.13 hold for the sets Qλ and qλ .2. there will be infinitely many nodes λ for which ι(λ) = n. shallowest) node for which ι(λ) = n. If ι(λ) = k. Notice. λ extends λι(λ) c.2. Every natural number is the value of all and only those nodes along the left-most branch descending from some node. Fact 5. COMPLETENESS OF CA b. e. There is at most one such y. To show that there is such a y. there is a unique y ∈ A such that ∀λ(qλ ⊆ x ↔ aλ ⊆ y) Proof.e.13k holds. c. d. define the embedding of Qf into A: ˙ ˆ Fact 5. by Fact 5. λι(λn ) = λn d. For every n. For every n.2. λn = the shortest λ such that ι(λ) = n. These are the only nodes below λ with the value k. Fact 5.2. λ. a.23.2. Then there is some n such that x = x1 ∪ · · · ∪ xk . there’s a λ such that n = ι(λ).) Each of the points listed in Fact 5. n. then all nodes along the left-most branch descending from λ also have the value k.16. a. there are infinitely many λ such that n = ι(λ). Though for a given natural number n. the Qλ can be regarded as node sets and the qλ as node atoms for any model Qf . finally. c. the ι(λ) take on all and only values less than n!.2. ι(λ) = ι(κ) iff λ = κ or one 0-extends the other.21. λι(λ) = λ iff λ =< 0 > or λ(L(λ)) = 0 (ie a node. If x ∈ Q. At each depth. there’s a λ such that n = ι(κ) iff κ = λ or κ 0-extends λ.22. we can associate with each natural number a shortest (i.2. especially.20. will be the highest node with a certain value just in case λ is not the leftmost immediate descendant of its parent. ι(λn ) = n b. For every n.

˙ ˆ c. If x is not a quasi-congruence class. Then ∀λ(qλ ⊆ x ↔ aλ ⊆ y ↔ qλ ⊆ x ) But every integer is ι(λ) for some λ. b. g is one-one. PRIME MODELS FOR CA where each xi is an (n!)-congruence class and hence a node set in Qf . g maps Q onto A. Suppose g(x) = y = g(x ).24.2.25. Let y be the element of A corresponding to x . The nodal embedding of Q into A is defined by letting ˙ ˆ g(x ∈ Q) be the y ∈ A such that ∀λ(aλ ⊆ y ↔ qλ ⊆ x) Fact 5. for some i iff λ extends λi iff aλ ⊂ Aλi iff aλ ⊂ y. a. let y = Aλ1 ∪ · · · ∪ Aλk ˙ ˆ So y ∈ A and: qλ ⊂ x iff qλ ⊂ Qλi . Proof.2.13h 77 ˙ ˆ where x is a quasi-congruence class.23.2. Immediate from the proof of 5.5. . so x = x . If x ∈ Q.2. and let y = (y ∪ { aλ | qλ ⊆ ∆1 (x) }) − { aλ | qλ ⊆ ∆2 (x) } ˙ Definition 5. a. as described above.2.. then x = (x ∪ ∆1 (x)) − ∆2 (x) by 5. then g(x) = (g(C(x)) ∪ g(∆1 (x))) − g(∆2 (x)) b. So x = Qλ1 ∪ · · · ∪ Qλk Now.

16) b. Qf x ⊆ y iff ∀λ(qλ ⊂ x → qλ ⊂ y) iff ∀λ(aλ ⊆ g(x) → aλ ⊆ g(y)) ˆ iff A g(x) ⊆ g(y) (by 5. n) be the number of charmed node sets of level n contained in C(z). As in (a). unions. n) − (|∆1 (x)| − |∆2 (x)|) = nc(y.2. g can be shown to preserve ∅. of Q into A is an isomorphism ˆ of Qf onto A.78 c.24) (by 5. Qf x < y iff Qf iff A iff A ˆ iff A x∼x ⊂y for some x ∈ Q g(x) ∼ g(x ) ⊂ g(y) g(x) < g(y) g(x) < g(y) since A ˆ since A ⊆ A since Qf REP< by (b) and (c) SUBSET. n) − (|∆1 (y)| − |∆2 (y)|) iff nc(g(x). below. n) − (|g(∆1 (x))| − |g(∆2 (x))|) = nc(g(y). Qf x ∼ y iff θf (x) = θf (y) iff nc(x. a.2. intersections. And let n. relative complements. and proper subsets.26. Proof. COMPLETENESS OF CA Theorem 5. Let nc(z. g. INDISC∼ ˆ since A ⊆ A . c. 5. The nodal embedding.2. be the least k such that x and y are both unions of node sets of depth k. I. Obvious. n) − (|g(∆1 (y))| − |g(∆2 (y))|) since Qλ is charmed iff Aλ is charmed and g preserves boolean relations iff A g(x) ∼ g(y) ˆ iff A g(x) ∼ g(y) d.

3.3 Summary We can now draw our final conclusions about CA. g(z)) since Qf 79 DEF+ by (b) and (c) since A DEF+ ˆ since A ⊆ A So. Hence T MODn. T MODn.j (see Lemma 3. for some congruous f .12c. by 3. For each n > 0.j where 0 ≤ i = j < n. y.3. then CAIf is consistent and complete.i → ¬MODn.6. 0 ≤ i < n.2. by 5. then T ≡ CAIf . If f is total and congruous.i for exactly one i.i for each such i. But suppose T MODn. It T is a completion of CAI.3) applies. Since CAIf is model complete. by 5.27.27. and we may conclude: Corollary 5. g(y). for CA MODn. CS. . z) iff Qf (x ∪ y = z ∧x∼x ∧y ∼y ∧ x ∩ y = ∅) ˆ iff A (g(x ) ∪ g(y ) = g(z) ∧ g(x) ∼ g(x ) ∧ g(y) ∼ g(y ) ∧ g(x ) ∩ g(y ) = ∅) iff A (g(x ) ∪ g(y ) = g(z) ∧ g(x) ∼ g(x ) ∧ g(y) ∼ g(y ) ∧ g(x ) ∩ g(y ) = ∅) iff A ˆ iff A Sum(g(x). g(y). since T CAI and CAI DIVn . So CAIf is complete. congruous remainder function. each model of CAIf has a submodel isomorphic to Qf . Theorem 5. But if T (¬MODn. the prime model test (Appendix. Qf Sum(x.1.0 ∧ . Proof.3.2.n−1 ). ∧ ¬MODn. . and their completions.i or T ¬MODn. T MODn. . then CAIf has a prime model. Again. CAIf is consistent because f is congruous.3. and has a prime model.11). If f is a total. T would be inconsistent.5. Fact C. 0 ≤ i < n: Since T is complete.1. then T is inconsistent. Corollary 5. 5.i and T MODn. Proof. g(z)) Sum(g(x).i for at least one one i. SUMMARY e.2.3.

a. then CAIf is recursively enumerable. . since CAIf is a. a. CAIf T . given that CS CA.2b. φ is consistent and T has only infinite models.3. Theorem 5.2 and 3.80 So.3.3. b. CS. a. EXACTLYn is decidable. a. b. For total f . By Lindenbaum’s lemma.3. b. each of these extensions has a consistent and complete extension.3.5. CSI has 2ω completions.3.3. Every completion of CA is consistent with CS. a. 5. There is no sentence. COMPLETENESS OF CA MODn.m sentence is in CSIf . EXACTLYn φ iff An φ. Follows from (a) and 3. CS is decidable. There are 2ω remainder functions whose domain is the set of prime numbers. CS. CSIf is decidable iff f is decidable.6. b. Corollary 5. such that T = CA. For n > 0. But An is a finite model. Proof. alternate between generating theorems of CA and testing whether An ¬φ. so each corresponds to a consistent extension of CSI. so it is decidable. Corollary 5. Each such function is congruous. Theorem 5. Follows from 5.4. let f (n) = m iff T complete.13. To determine whether CS φ. φ. (⇒)If f is decidable. b. Proof. But CAIf is complete. determine which MODn. CA ≡ CS Proof.m .3. b. Then T CAIf and. (⇐)To calculate f (n).

defined in 3. Hence (BASIC ∪ T ). T = { ADIVn | n ∈ J }. θ. But CA ≡ CS. so A . A φ CS. SUMMARY 81 Proof.26. We claim the following without proof: (1) (2) (3) By (1) and (2).5.2. so ¬φ ∈ CS. then ¬φ is true in all finite models of CA.3. Suppose CS φ. Let K = { n | every prime factor of n is a member of J }. But by (3). in which size relations are determined in accordance with the size function. Theorem 5. A CS. If φ is true only in infinite models of CA. by compactness. φ is inconsistent.3. So CA φ and.7. Proof. (BASIC ∪ T ) φ for some finite set of ADIVn principles. A A A BASIC ADIVj for all j ∈ J ADIVk φ. CS is not finitely axiomatizable. Let A be a model with domain k∈K Qk . so CA.

COMPLETENESS OF CA .82 5.

It has infinite models Qn and Qf . OUTPACING.13).1. The general motivation behind this principle should be familiar. then x > y.6 Sets of Natural Numbers CS has standard finite models since it consists of sentences true in all such models. Definition 6. x outpaces y just in case the restriction of x to any sufficiently large initial segment of N is larger than the corresponding restriction of y. We extrapolate from well understood finite cases to puzzling infinite cases. 6.1 The outpacing principle Throughout this chapter. iff: ∃n∀m(m > n → |x[m] | > |y [m] |) Notice that the size comparison between the two restricted sets will always agree with the comparison of their normal cardinalities since all initial segments of N are finite. To rule out such anomalies. So CS . in section 1. Section 2 establishes that OUTPACING can be satisfied jointly with any consistent extension of CS in a model whose domain is P(N). If x outpaces y. An ordering of P(N) that satisfies CS will not necessarily appear reasonable. For example.2. OUTPACING does not fix the size relations over P(N). In this chapter we will show that CS has infinite standard models over P(N). x and y will range over P(N). that is. we introduce a principle. We employ this notion to state a sufficient condition for one set of natural numbers to be larger than another: OUTPACING. some such orderings say that there are fewer even numbers than prime numbers (see below 6. But we should also 83 . OUTPACING mentions the natural ordering of N and applies only to subsets of N.1.

This revised principle conflicts with CS. 2n + 1} of N contains n evens and n odds. since [2n] outpaces [2n + 2] while [2n + 1] does not. and see how far they go. x restricted to z is larger than y restricted to z. strengthen the conditional to a biconditional. . But the two are discernible under outpacing. [3n] > [4n] > [5n] > . Though (2) is true.84 6. mechanical way without risking contradiction. it cannot be expressed in LC< . A is an outpacing model iff ˙ A = P(N) A BASIC.0. that this extrapolation cannot be done in any straightforward. [2n] has at least (k − 1)/2 members less than or equal to k for any given k. for outpacing is not a quasi-linear ordering. Some of these conflict with one another. . . It is doubtful whether there is any mechanical way to decide which of these statements are true. We shall finesse this problem by regarding OUTPACING as the (very large) set of sentences { b < a | a outpaces b } Fact 6. If k > 0. . there are many statements that assert of infinite cases what is true of finite cases. a. So. Definition 6. Others are too weak to be helpful. thus: (1) x > y iff x outpaces y. There is another point that underlines the need for care in extrapolating from finite cases to infinite cases: we cannot just use (2): (2) If. its antecedent is only satisfied when y ⊂ x. and A OUTPACING There is a slight difficulty in saying that an interpretation of LC< satisfies OUTPACING . given any finite subset z of N. Since OUTPACING involves the < relation over N. again. For example. [2n] > [3n] b. If k > 4.1. for example. SETS OF NATURAL NUMBERS emphasize.1. then (k − 1)/2 > k/3 + 1 . Every outpacing model satisfies the following a. determine whether they are consistent. We cannot. then x > y. . then [kn] > n2 Proof. [3n] has at most k/3 + 1 such members. c. neither [2n] nor [2n + 1] outpaces the other since each initial segment {0.1. . The best we can do is propose plausible theories.

it is sufficient to show that ¬ [2n] < [2n + 1]. b. Similar to that of (1). then. including other congruence classes. there is a y such that y ∼ [2n] and y ⊂ [2n + 1]. we generalize 6. then in any outpacing model: x ∼ y ∨ x > y ∼ (x − x1 ) Proof.2. xi < yi < xi+1 . we show that both of these possibilities can be realized in standard models over P(N).2 applies here since the only facts about [2n] and [2n + 1] used hold by virtue of these sets forming an alternating pair. Fact 6. [2n + 1] ≥ [2n + 2] c. y is an alternating pair iff x and y are infinite and for all i > 0. then N is even. 85 c.1. Then [2n] leads y at k + 1 and y never catches up. y = [2n + 1]. By TRICH. For a given k > 0. and hence smaller than [2n]. If m = k 2 . In section 6. The argument for 6.1. Note that [2n] = [2n + 2] ∪ {0} and that BASIC (*) (*). Then: a. Theorem 6. Every outpacing model satisfies the following a. Fact 6.2 correspond to the possibilities that N may be odd or even: if [2n] ∼ [2n + 1]. z = [2n + 2] and apply (*). So there is no y such that [2n] ∼ y ⊂ [2n + 1]. Aj is an alternating pair. c. THE OUTPACING PRINCIPLE b.2 to similar cases.5. N[m] will have more members in [kn] than in n2 .1. Here.1. [2n] ≥ [2n + 1] b. (y ⊂ x ∧ z ≤ y ∧ Atom(z ) ∧ z ⊂ x ∧ x = z ∪ z ) → y ∼ z Let x = [2n].2.4. .1. If 0 ≤ i < j < k. if [2n + 1] ∼ [2n + 2].6. If [2n] > [2n + 1]. If it were. y is an alternating pair. both [kn] and n2 have exactly k members less than m. then Ai . a.1. Let k be the least odd number not in y. Similar to (1). x. let Ai = [kn + i] for each i < k.1. then [2n + 1] ∼ [2n + 2]. Definition 6. then N is odd. by REP< .3.1. But any proper subset of [2n + 1] is outpaced by. Proof. For m > k(k + 1). The two alternatives left open in 6. If x.

Proof. then Ai ∼ Aj . is the first order language which results from adding to LC< . then Ai ∼ Ap .86 b. (i) For 0 ≤ i < p. Ap is an alternating pair and A0 > Ap . Definition 6. let p = k. the language of subsets of N. Ai (n + 1) = kn + i + k and i < j < k + i. otherwise. 1.1. But A0 ≤ Ai by the selection of p. An is the standard finite interpretation of LN over P(N[n] ) in which: An (a) = a ∩ N[n] = a[n] for each a ⊆ N. let p be the least such i. There is a p. Example. So Ai ∼ Ap . SETS OF NATURAL NUMBERS (ii) If p = k. (i) follows by TRANS∼ . b.Aj (n) = kn + j. Then one of the following situations obtains: A0 A0 A0 A0 ∼ A1 > A1 ∼ A1 ∼ A1 ∼ A2 ∼ A2 > A2 ∼ A2 ∼ A3 ∼ A3 ∼ A3 > A3 > [4n + 4] ∼ [4n + 4] ∼ [4n + 4] ∼ [4n + 4] 6. we construct models of CS over P(N) that satisfy OUTPACING. a. then A0 ∼ Ap ∪ {0}. b. Let k = 4. a name for each subset of N. Ai is an alternating pair. .4.1.2 Models of CS and Outpacing In this section. though the application here demands some important differences. LN . (ii) Immediate from 6. Hence Ai ∼ A0 − {0} ∼ Ap . Outpacing models will be constructed out of finite models of CS by a technique which is very much like the ultraproduct construction common in model theory. (iii) A0 > Ap ≥ Ai if i ≥ p. as individual constants. 0 < p ≤ k such that (i) If i < j < p. so A0 ∼ Ai .4 since A0 .1. a. Ai (n) = kn + i. 2.2. So A0 > Ai . A0 . So either A0 > Ai or A0 ∼ Ai by 6. If A0 > Ai for some i. (iii) If p ≤ i < k. so Ai = [4n + i] for i = 0. (See example below). 3. 6.

If F is an ultrafilter over N. If Y ⊆ X and Y is infinite. lemma 1.15. MODELS OF CS AND OUTPACING 87 Definition 6. If τ is a term of LN . Theorem 6. then there is an ultrafilter over X which includes F .2. then AF is an outpacing model. 18. then a ∩ b ∈ F c. Prop. and (iii) If a ∈ F and b ∈ F . Def. and (iii) if Y ⊆ X.6. b. p318] If X is a set and F ⊆ P(X).6. then a. c.3.108] b. Definition 6.18.2. Ch. F is an ultrafilter over X iff (i) F is a filter over X (ii) X ∈ F . (cf.4.2. A˙F = P(N). An ultrafilter. A non-principal ultrafilter over X contains all cofinite subsets of X. Our main result is that if F is non-principal.3. and similarly for other predicates. then a. then there is a non-principal ultrafilter over X which contains Y . Boolean symbols receive the usual interpretation. F .2. d. then b ∈ F . F is a filter over X iff (i) F = ∅ (ii) If a ∈ F and a ⊆ b. b.5.2. [Bell. then either Y ∈ F or (X − Y ) ∈ F . then Ak (τ ) = AF (τ ) ∩ Ak = AF (τ ) [k] . If F is a non-principal ultrafilter over N. 18. A non-principal ultrafilter contains no finite sets. c. [Monk. F has the finite intersection property iff the intersection of any finite subset of F is non-empty. If F ⊆ P(X) and F has the finite intersection property. then AF is the interpretation of LC< in which a. p. a. [Monk.319] d. over X is principal iff there is some x ∈ F such that F = { a ⊆ X | x ∈ a } Fact 6.2. p. AF a < b iff k | a[k] < b[k] ∈ F .

a[n] ⊂ b[n] . then there is a k ∈ b but not ∈ a. otherwise a[n] would not be included in b[n] for any n greater than k. (ii) If φ = a = b . AF e. d. Conversely.2. a. By induction on the structure of φ: (i) If φ = a ⊂ b . So a ⊆ b. there cannot be a k in a but not in b. b. iff { n | An a = b } = N iff { n | An a = b } ∈ F since if An (iii) AF a = b and k > n. AF f. Immediate from 6. So if n > k. ADIVn . So Ak (τ ) = a[n] by 6. BASIC. then c.1b. by 4. then AF φ iff a = b iff a[n] = b[n] for all n. SETS OF NATURAL NUMBERS b. .2. for every n. then it is infinite.3b. if n | a[n] ⊂ b[n] ∈ F . then Ak a = b.88 6. then AF φ iff { k | Ak φ} ∈ F φ for every k. (ii) If τ = τ1 ∪ τ2 . AF Proof. Hence. τ = a for some a ⊂ N. If φ is a quantifier free formula of LN . By induction on the structure of τ : REP< . Ak (τ ) = Ak (τ1 ) ∪ Ak (τ2 ) = (AF (τ1 ) ∩ Ak ) ∪ (AF (τ2 ) ∩ Ak ) = (AF (τ1 ) ∪ AF (τ2 )) ∩ Ak = AF (τ1 ∪ τ2 ) ∩ Ak The proofs for intersections and relative complements are similar. (i) If τ is a constant. clearly a = b.2.4b. But. then AF φ iff a ⊂ b If a ⊂ b. So. so a ⊂ b. { n | AF φ } is cofinite and. in F . If φ is a universal formula of LN and Ak AF φ. a < b iff { k | Ak (a < b) } ∈ F .

As in 6. for all k φ(a). for all a φ(a).6. for all k φ(a). these sets are approximately equal in size and AF ADIVn (x). . .) f.5). (AF ATOM because it contains all singletons of natural numbers. these form an alternating ntuple. for which AF k | a[k] < b[k] so. Claim: If k ∈ K. thus. Furthermore. in the manner of the congruence classes modulo n (see Theorem 6. for all a. . . The n sets. MODELS OF CS AND OUTPACING (iv) If φ is non-atomic. then. (a ∼ a ) = {k1 .5. xi . let xi = { xkn+i−1 | k ∈ N }. Construct a . . Hence: AF a ∼ a since they are the same size over some set which contains K and is. x an infinite subset of N. for all a ∀xφ(x) by (b) (a < b). since F is an ultrafilter: AF φ1 ∧ φ2 iff AF φ1 and AF φ2 iff { k | Ak φ1 } ∈ F and { k | Ak iff { k | Ak φ1 ∧ φ2 } ∈ F AF ¬φ iff AF φ iff { k | Ak iff { k | Ak 89 φ2 } ∈ F φ} ∈ F / ¬φ } ∈ F c. . if F . Suppose then So So So d. then a [k] ∼ a[k] . Suppose AF as follows: Let K = Ak Ak Ak AF AF ∀xφ(x). K ∈ F . where n = |a[ki+1 ] | − |ai | a = ai Then a ⊂ b since each ai draws its new members from b.2. For 0 < i ≤ n. partition x.1. Immediate from (c) and (d): BASIC is equivalent to a set of universal sentences. .} where the ki ’s are in strictly increasing order. together with ATOM and REP< . Let a0 = ∅ ai+1 = ai ∪ the n greatest members of b[ki+1 ] not in ai .1. ki . . for all k. a subset of b. e.

the only non-universal axioms of CAI could be verified in the constructed model more or less directly.5e and 6.2.3b. There is. So. then a. there will be many cases where AF a < b even though b does not outpace a. pp. [Bell. which results in a domain whose elements are functions from the index set (N.3. we can immediately conclude that Ai /F satisfies CS. but only a submodel. Unfortunately. here) to elements of the factors. but is not quite the same.90 6.2.6. [2n] is neither finite nor cofinite. we would have then been able to infer that the part of that model which held our interest satisfied all universal formulas of CS. a. there will always be the same number of evens and odds. the Ak ). Ai /F is not the model we wanted: its elements are not subsets of N. If F is a non-principal ultrafilter over N. a natural mapping from subsets of N to such elements. SETS OF NATURAL NUMBERS Corollary 6. but equivalence classes of functions from N to finite subsets of N. [2n + 1] ∈ F . these special means were available. This will happen whenever k | a[k] < b[k] ∈ F but is not cofinite. for example. Consider a familiar example: Let a = [2n + 1] and b = [2n]. Fortunately. had we constructed the reduced ultraproduct. indeed. For a given F . 87-92]). for if we count the even numbers and the odd numbers up to some even number.2. AF b. AF Proof. we still would have had to resort to special means to show that the non-universal formulas were likewise satisfied. there will always be one more even and if we count up to some odd number. AF [2n + 1] < [2n]. If [2n] ∈ F . These functions are then gathered into equivalence classes (by virtue of agreeing almost everywhere. and CSI The proof of 6. so it may or may not be in F .2. Then a[k] < b[k] iff k ∈ [2n]. which we’ll call Ai /F . given that each of the Ak satisfies CS. i. and this mapping would have allowed us to identify a model over P(N) as a submodel of Ai /F . The model so constructed. and certainly any formula which is satisfied in all the factors. so AF [2n + 1] ∼ [2n]. Otherwise. a model is built by first taking the product of all factors (in this case. on some member of the filter) and the reduced ultraproduct is defined by interpreting the language over these equivalence classes. This is handy because. Immediate from 6. . Immediate from (a) and 5. b.6 is modeled on the usual ultraproduct construction. has the handy property that it satisfies any formula which is satisfied by almost all factors. and the completeness proof of the last chapter allowed us to infer that all formulas true in all of the factors are true in the model AF . CAI. after all.5f. In the usual construction(see.e.

But the disjoint union of x ∩ x+ with x1 or x2 .2. Let a run of x be a maximal consecutive subset of x. x2 is an alternating pair.10. x+ . Fact 6. respectively. MODELS OF CS AND OUTPACING 91 The construction of a model AF from any non-principal ultrafilter. then x+ = { i + 1 | i ∈ x }. (So [2n] has only 1membered runs. while N−[10n + 1] has only 9-membered runs. so / n ∈ x+ ). x. F is a non-principal ultrafilter. x maintains its lead. (See below.8. x1 = (x − x+ ). x+ are like alternating pairs in the following way: Lemma 6.) So. If F1 and F2 are distinct non-principal ultrafilters over N. Throughout the first run of x. If x ⊆ N.2. x+ is an alternating pair iff x is infinite and there is no n ∈ x such that (n + 1) ∈ x. suggests that there are many outpacing models unless different ultrafilters can yield the same model. Proof. we will show that the presence of a set in an ultrafilter makes a direct. no two consecutive numbers are in x. x. b.7. We will first show that this qualification is not needed. yields x or x+ . c. and A = AF . x2 = (x+ − x). each decision may go either way. A c. though the decisions are not independent of each other. each decision is made for a model AF by the presence or absence of a particular set in F . We know from (a) that x2 ∼ x1 or x2 ∼ x1 − x1 (1).2. Theorem 6. Definition 6. b. then AF1 = AF2 . . x1. So (b) follows from DISJ∪ . but this is not always the case. This pattern repeats during successive runs of x. a. x1 (n) is the first element in the n-th run of x and x2 (n) is the first element after the n-th run of x. there is a set of decisions that must be made in constructing an outpacing model. Furthermore. Nevertheless. losing it only at the least n for which n ∈ x (since (n − 1) ∈ x.6. A pair of sets. Putting this in another way. A Proof. That is.2. can sometimes be an alternating pair. We need only prove (*): (*) |x[n] | > |x+ [n] x ∼ x+ or A x > x+ ∼ x − {x1 } x > x+ iff x ∈ F .) To show this. pairs x. If x is infinite. since x(1) ∈ / x+ because ¬(x1 − x1 (1) ⊂ x). then a. F . | iff n ∈ x [n] Informally: x[n] first becomes greater than x+ for n = x(1). personalized contribution to the model AF .2.9.

If x. Theorem 6. there is a nonprincipal ultrafilter F such that G ⊆ F . H. By (*). Since neither x nor y outpaces the other.f (k) Let G = { Gk | k > 0 }.2. a. we can suppose there is a set. Proof.2. by (a) there is an F such that AF y ∼ (x − x(1). Without loss of generality. (See 3. So H = { n + 1 | k ∈ J → n ≡ f (k) mod k }.2.12. But f is congruous.11. x − x(1) . Since the intersection of any finite subset of G is infinite. b. let F be a non-principal ultrafilter which contains J. then there is an ultrafilter F . J is infinite.3. If x. But then AF x > y. y is an alternating pair. x. Theorem 6. If neither x nor y outpaces the other. let Gk = [kn + f (k) + 1] for each k > 0.2.7: Proof.10c. such that AF (x ∼ y).3d. By 6. such that AF (x > y). For example. AF MODk. of G is infinite. . Let J = k | |x[k] | = |y [k] | . If H is a finite subset of G. AF1 x > x+ and AF2 x ∼ x+ / Theorem 6. f . Every infinite completion of CS has an outpacing model. Proof. where J is some finite subset of N+ . so AF CSIf .2. so the restriction of f to the finite domain J has infinitely many solutions (see 3. then there is an ultrafilter F .92 6. The intersection of any finite subset.2.8a). SETS OF NATURAL NUMBERS We can now prove our main result.6. a. Recall that every infinite completion of CS is equivalent to CSIf for some total and congruous remainder function. then H = { Gk | k ∈ J }. we can show that either of the alternatives in 6.) Given such an f .7 allows us to improve upon some previous results. So. b. Theorem 6.10b can obtained for any alternating pair. so is y. such that x ∈ F1 and x ∈ F2 .f (k) . y is an alternating pair. By 6. Then AF x ∼ y.2. Then (*) holds for each k: (*) Gk = n | An MODk.2.

5]. all of the possibilities listed in 6.1. This section has explored the existence and variety of outpacing models.2. (2) If FA = FB .2. both (1) and (2) are necessary. two outpacing models may agree about all pairs x. perhaps. Theorem 6. So (1) and (2) are open problems. Ch. x+ but disagree elsewhere. If (1) is false.2. 6. we can now extend that observation to moduli other than 2. If x -outpaces y. in general. A.11 we noted that there are even and odd outpacing models. But even if (1) is true. there is no unique outpacing model which satisfies T . Finally. First. so whether x ∈ F is an independent choice. Lemma 6.10c may suggest that any outpacing model. To establish that A = AF . even it T is an infinite completion of CS. This would be true only if fixing the congruence classes determined whether x ∼ x+ or x > x+ for every x ⊆ N . MODELS OF CS AND OUTPACING 93 On the basis of 6.4. by considering x = n2 : any finite set of congruences has infinitely many solutions that are squares and infinitely many that are not.2. it is not clear whether every outpacing model can be obtained by the construction of 6. That all such choices are not determined by a remainder theory can be seen intuitively. Proof. then FA is a non-principal ultrafilter.13. we could produce standard models of CS over P(N) which satisfy OUTPACING and these will not. (1) If A is an outpacing model. Also. At most one of them is obtainable by our construction. There are standard models of CS over P(N) which do not satisfy OUTPACING. is AF for FA = { x | A x > x+ }.5 for the relative sizes of the k-congruence classes are obtainable in outpacing models. there are only 2ω remainder functions while there are ω 22 non-principal ultrafilters over N[Bell. but it should not. Theorem 1. . Modifying 6.2.4. Suppose that is a linear ordering under which N forms an ω-sequence. then x > y. each yielding a different outpacing model. though it may be extraneous to show that OUTPACING is independent.6. More specifically. then A = B I have not been able to prove (1) or (2). then clearly A = AF . Then we could define x -outpaces y as follows: ∃n∀m[n m → |{k | k ∈ x ∧ k m}| > |{k | k ∈ x ∧ k n } |] and define the principle: OUTPACING . we will do so. Three comments are in order before we turn to the common structure of outpacing models.2. satisfy OUTPACING. Second.

a. . x.3. . e. they do not content themselves with speaking of the (Cantorian) cardinalities of these sets. . is the density of of x in N. d. Definition 6. . . If x ⊆ y = ∅. iff x converges in N. they need a more discriminating notion. x diverges in y. The asymptotic density of a set. From now on we shall use the term ‘density’ for asymptotic density. we compare the ordering of P(N) given by density to the orderings given by CS and OUTPACING. so p > [2n] and OUTPACING is false in OUTPACING models. y). if there is one. In any model of OUTPACING . ρ(x) = 0. In this section. The density of x. . qk . ρ(x). iff x has a density in y. of |x[n] | as n grows. as the evens and the odds are in normal outpacing models. the asymptotic density of [2n] is 1/2. is the limit. Since they often want to compare infinite subsets of N. SETS OF NATURAL NUMBERS is the ordering: p1 . pk . Fact 6. b. For example. But the evens are much smaller than q. the fraction of numbers less than or equal to i that are ˆ members of x. a. y) = r iff ∀(δ > 0)∃n(∀i > n)(r − δ < xi ˆ < r + δ))) yi ˆ c. if x has a density in N. q1 . then ρ(x. of natural numbers is the limit. One notion they use is asymptotic density. cvg(x).2. If x is finite. x converges. cvg(x.3. 6. the density of x in y. . that 6. .94 Suppose. x diverges iff x diverges in N. If x ⊆ y = ∅. if it exists of i=y1 [i] lim xi /ˆi ˆ y ω That is. where p is the set of prime numbers and q is its complement. ρ(x.1. xi = |xi | . otherwise. p and q will be nearly the same size.3 Size and density When number theorists talk about the sizes of sets of natural numbers. then x converges in y. y). for example.

SIZE AND DENSITY b. and so forth. Construct the set x as follows: x0 = ∅ xi . If x is cofinite. between 100 and 999. b. c. z). ρ(x) = 1. there is a set with density r in y. between 10. then cvg(x. 95 a. Suppose r is given. y) and cvg(y. There are divergent sets. If 0 ≤ r ≤ 1. Then. y)ρ(y. z) and ρ(x. c. x= xi if xi ≥ r. xi+1 = c.999. z) = ρ(x. there is a set with density r. xi . if .3. z).3. If cvg(x. Modify the construction for (b) in the obvious ways. Fact 6. a. ˆ ˆ If n > 1. i + 1. then x10 ≤ .3. [2n]) = 1/2 ρ([4n]) = 1/4 d. z).1 and x10 ˆ b.6. If 0 ≤ r ≤ 1 and y is infinite. Theorem 6. ρ(x.000 and 99. So xk cannot have a limit. Let x = i | ∃n(102n ≤ i < 102n+1 . Proof. So x contains all numbers between 0 and 9.9. ρ([2n]) = 1/2 ρ([4n] .3. y outpaces x. 2n+1 2n ≥ . z) < ρ(y. Suppose that both x and y converge in z.4. ˆ otherwise.

Immediate from Fact 6. So. − b < y i /ˆi − ρ(y.3. z2 ) ≤ ρ(y. z1 ) = 1/3 and ρ(y. But if ρ(x.3. and let y be z1 − x. xi /ˆi < ρ(x. xi /ˆi < y i /ˆi ˆ z ˆ z xi < y i ˆ ˆ |x[i] | < |y [i] | We can use the relation between density and outpacing to draw conclusions about densities that have nothing to do with outpacing. there are uncountably many sizes in A.3. then a. ρ(x. Theorem 6. we can begin to appreciate how precise an ordering outpacing models provide: Fact 6. N) = 0. as in Theorem 6. z2 ) > ρ(y. there are uncountably many sizes of sets in A. − b < xi /ˆi − ρ(x. . from Facts 6.96 Proof. y outpaces x. z2 ): Let z1 be the set of primes.3. let x contain every third member of z1 . z1 ). then x outpaces y. and b. N) = ρ(y. z1 ) < ρ(y.3b and Theorem 6.4.5. z1 ) = 2/3. a. We cannot strengthen the consequent of 6. let y be an infinite subset of x. Since ρ(x. and let z = x − y. z) < b ˆ z n2 = the least n such that for i > n. Then ρ(x. z) + b < ρ(y. z1 ).6.3. z2 ) < ρ(y. So. they will differ in size if they have the same density in some common set.5 to say that ρ(x.4 implies that in any outpacing model. then ρ(x. x) = 0. Let Let and let So we have and But So So So b = (ρ(y. SETS OF NATURAL NUMBERS n1 = the least n such that for i > n. If A is an outpacing model. If ρ(x. z) − b. z2 ). z) − ρ(x.3. but ρ(x. There are uncountably many subsets of y with distinct densities in y. Even if two sets have the same density. z2 ) Proof. z) − b ˆ z ρ(x. z) < b ˆ z for any i > n1 + n2 . z2 ). Proof. then even among sets with density r.3. z) + b ˆ z y i /ˆi > ρ(y. if 0 ≤ r ≤ 1. x) = 0.b and (c). Theorem 6.3. where ρ(y. sets with distinct densities will have distinct sizes. b. and both x and y converge in z2 . though ρ(y1 . z2 ) ≤ ρ(y. z1 ) < ρ(y. for any y1 ⊂ y.5. z))/3 6. Let x be an infinite set with density r.

3. any such ordering which is realized in an outpacing model) is both a refinement and a completion of the ordering suggested by density: a refinement because it preserves all differences in size which are captured by the notion of density. though different sizes may yield the same cardinal number. 97 y1 < y2 . (1a) is an immediate consequence of Theorem 6. But ρ(z ∪ y1 ) = ρ(z ∪ y2 ) = ρ(x). though two sets with the same cardinal number may have different sizes. we want to know whether the density of a set is a function of its size. Theorem 6. y2 ⊂ y. Then A by 6. then ρ(y) = r. SIZE AND DENSITY Suppose now that y1 ⊂ y. by DISJ∪ . But the situation is really not so clear.4 insures that (1) just in case (2).1 The Convexity Problem It’s tempting to infer from these results that the extremely fine ordering of sets by size (or. so A z ∪ y1 < z ∪ y2 . In passing. It turns out that a negative answer is compatible with our theory (CS and OUTPACING).3. rather. If x ∼ y and ρ(x) = r and cvg(y). We shall express this fact by saying that. then x < y and if r > r2 . while an affirmative answer may or may not be consistent. we may consider two additional formulations: (3) If ρ(x) = ρ(y) and x < z < y. at least when we focus on P(N). we will give a more precise formulation of this problem. So. then x > y. cardinality is a function of size. there is some r2 = ρ(y). second. But x ∼ y.3.3. and ρ(y1 ) < ρ(y2 ). if r < r2 . we’ll discuss the consistency of an affirmative answer. then ρ(y) = r. Consider (1) and (1a): (1) (1a) If x ∼ y and ρ(x) = r.6. We can regard the cardinality of a set in P(N) as determined by its size. It is evident that the size ordering over P(N) in any outpacing is a refinement of the ordering by cardinality: if x has a smaller cardinal number than y. since z was obtained by removing from x a set with density 0 relative to x. 6. we will show that a negative answer is consistent.) Now. then x is smaller than y.3. For if y converges.4. we’ll explain why this is called the convexity problem. Recalling that sets may have the same density even though they differ in size. so r2 = r. a completion because all sets are located in a single. then y converges. First. the questionable part of (1) can be expressed as (2): (2) If x ∼ y and x converges. (It is not at all clear that this is true in any power set. then ρ(z) = ρ(x) .4. linear ordering of sizes. finally.

so ρ(y) = ρ(x) To apply (3). ρ(y) = r. x and y . SETS OF NATURAL NUMBERS If ρ(x) = ρ(y) and x < z < y. then y = x. by SUBSET. and x1 < y < x2 . Fact 6. we need to find two sets. for some a ∈ x and x2 = x ∪ b. let x1 = x − a. then z converges. there are two sets. by INDISC∼ . . Since and then So and But So So x∼x ρ(x) = ρ(z) ρ(x ) = ρ(z) ρ(z − x ) = 0 ρ(z − y ) = 0 y = z − (z − y ) ρ(y ) = ρ(z) ρ(y) = ρ(z) by (1) by (1) ⇐ Suppose that x ∼ y and ρ(x) = r. ρ(x1 ) = ρ(x) = ρ(x2 ) So. y ∼ y .98 (4) 6. Proof. An outpacing model satisfies (1) iff it satisfies (3). If x = ∅ or x = N. by (3).7. ⇒ Suppose that x < y < z and ρ(x) = ρ(z). Since adding or removing a single element has no effect on the density of a set. / Then x1 < x < x2 .3. for which x ⊂ y ⊂ z. (3) and (4) are equivalent for the same reasons that (1) and (2) are equivalent. By REP< . x ∼ x . x1 and x2 such that ρ(x1 ) = ρ(x2 ) = r = ρ(x) and x1 < y < x2 Assuming that x = ∅ and x = N. for some b ∈ x.

. In geometry. F .3. The negation of (1) is satisfied in some outpacing model.8.7. any point between them (ie on the line segment from one to the other) is also in the figure. Let x be a set with density r and let y be a divergent set which neither outpaces nor is outpaced by x. given any two points in the figure. (1). Applying this notion in the obvious way to the size ordering. By theorems 6. SIZE AND DENSITY 99 The question at hand is called the ‘convexity problem’ because of the formulation in (3). a figure is convex if. Let K be n | x[n] = y [n] K is infinite.6. (2) and (4) say the same thing.4 and 6.3. so K is a member of some non-principal ultrafilter.3. (3) says that the class of sets having a given density is convex. Open Problem: Is (1) consistent with CS and OUTPACING? x ∼ y. We regret that the only thing we know about (3) is that it may be false: Theorem 6. Proof.3. AF so (1) is false in AF .

100 6. SETS OF NATURAL NUMBERS .

individual constants. φ . then if and only if identical not identical there is an x for all x Conjunctions and disjunctions of sets of sentences are represented by: φ . they are formalized in first order predicate logic with identity and function symbols.1 Notation Predicate Logic The formal theories discussed in this thesis are theories with standard formalization.5].. in the usual way...8). In most cases. DIVn . These may be predicates. The following notation is used for the predicate calculus: ∧ and ∨ or ¬ not → ↔ = = ∃x ∀x if .. or function symbols.... for example. p. maps natural numbers into sentences (see 3.. the ranks of the symbols will accord with their familiar uses and we do not stipulate them. 101 . A schematic function is a function whose range is a set of formulae. A first order language is determined by its non-logical constant symbols. That is...Appendix A A. φ .2. The arguments of such functions are indicated by subscripts. and φ . in the sense of [Tarski.

.102 APPENDIX A. we use the following notation: I the universe ∅ the empty set x ∪ y the union of x and y x ∩ y the intersection of x and y x − y the relative complement of y in x x ⊆ y x is contained in y x ⊂ y x is a proper subset of y These symbols are also used for set theoretic relations.. N+ N − 0 Z the set of integers ω the smallest infinite cardinal 2ω 2 to the ω A. we use the following: x ∈ y x is a member of y P(x) the power set of x { x | φ(x) } the set of x’s which are φ x. a the union of x and a x[n] N the members of x less than or equal to n the set of natural numbers: 0. . In addition. we use the following: i+j i−j ij i plus j i minus j i times j ij i to the j-th power i! i factorial i | j i divides j gcd(i. 1.3 Arithmetic For arithmetic.2 Set Theory For first order languages with boolean operations and predicates. j) greatest common divisor of i and j n ≡ m mod j n is congruent to m modulo j . outside of first order languages.

f is an isomorphic embedding of A into B. ˙ and a function which assigns to each individual constant of L a member of A.n. A. an n-ary function defined over and yielding values in A. the union of a chain of models. f.. ˙ c. {Ai } is a chain of models. LT . φ)... b. Definition B. a. We assume the notion of satisfaction as defined in [Chang. B is an extension of A (A ⊆ B).. g.. A is a submodel of B (A ⊆ B). Notions about theories: a. . A = {Ai }.2.) } For example: [kn + j] = the set of numbers congruent to j modulo k n2 = the set of squares of natural numbers 103 B Model Theory This section lists the model-theoretic notions and results assumed in the text and presents our notation for these notions. Definition B. of a first order language. section 1. The language of a theory. T2 ). c. MODEL THEORY We use the following notation for sets of natural numbers: [. L. where X is a subset of B.] = { k | ∃n(k = . consists of a domain..n.1. e. and to each n-argument ˙ function symbol of L. ˙ to each n-place predicate of L. Familiar notions about models. A theory is a set of first-order sentences.. A. a set of n-tuples over A.3] and use A φ to mean that A satisfies φ . ˙ An interpretation. T is complete. B is an elementary extension of A. b. e.B. d. T is consistent. h. A is isomorphic to B (A B). The submodel of B generated by X. T proves φ (T T proves T2 (T d..

If T APPENDIX φ. then some finite subset of T proves φ (compactness). Robinson’s notion of model completeness.1. then any submodel of A also satisfies T . then T is complete. If T is universal and A T . φ is consistent. Definition B. Definition C. T . A and B of T . p. d. Definition C. There are several ways to show that a theory is complete. Fairly familiar facts: a. T is universal-existential iff it is equivalent to a set of prenex sentences. A b.4. 355]. If T is universal-existential and Ai C Model Completeness This section presents the definition of model completeness and some basic results needed in Chapters 4 and 5. and A ⊆ B. g. A model complete theory is not necessarily complete. none of which have universal quantifiers. none of which have existential quantifiers. p. T is model complete iff T is consistent and for any two models. Fact B.359] T and A can be embedded . then T c.2. φ.5. A is a prime model of T if A in any model of T . A ⊆ B iff A is an elementary submodel of B. then {Ai } T. If T is categorical. φ is a primitive formula iff φ is an existential formula in prenex form whose matrix is a conjunction of atomic formulas and negations of atomic formulas. unless the theory has a prime model. b. We use only one. If T is existential.3. c. T is universal iff it is equivalent to a set of prenex sentences. c. T . Fact B. T is existential iff it is equivalent to a set of prenex sentences. More familiar notions: a. each of which has all of its universal quantifiers preceding any of its existential quantifiers. T is equivalent to T2 (T ≡ T2 ). [Monk. T is categorical. The following are well known facts: a. [Monk. b.104 f. If T is complete and T . which is based on A. for all i > 0. then B T.

T is model complete. A ⊆ B.5. The diagram of A is the set of atomic sentences and negations of atomic sentences of the A-expansion of the language of A which are true in A. directly or indirectly. d. and A φ(x). 356]: a. Fact C. To show that a theory is model complete we rely. a. For every model. . on a theorem of Monk’s (see 4. x ∈ A. then B φ(x). of T and every A-expansion L of L.1a).4. The A-expansion of L is the result of adding to L a constant for ˙ each element of A. and B φ(x). If T is model complete and T has a prime model. then A φ(x).C. A. p. b. x ∈ A. the T ∪ L -diagram of L is complete. φ is a universal formula. If A and B are models of T . c. MODEL COMPLETENESS 105 Fact C. The following are equivalent [Monk.2. b. A ⊆ B. φ is a primitive formula. which is based on some equivalent formulations of model completeness: Definition C. then T is complete.3. If A and B are models of T .

106 APPENDIX .

Tarski. Illinois: Northwestern University Press. 1968. The Principles of Mathematics. Elementary Theory of Numbers. New Jersey. G. Griffin. D. 1955. Models and Ultraproducts. Enderton. pp. 1968. New York. Cantor. 1937. North-Holland Publishing Company. NorthHolland Publishing Company. [Chang] [Dauben] [Enderton] [Frege] [Griffin] [Kurosh] [Monk] [Robinson] [Russell] [Tarski] 107 . Second edition. 1903. H. Orlando. A Mathematical Introduction to Logic. B. 1975. Contributions to the Foundations of the Theory of Transfinite Numbers. J. 1973. Russell. New York. Norton and Company. The Foundations of Arithmetic. Dauben. 222-236. McGraw-Hill. A. Amsterdam. E. Evanston. Pergamon Press. Monk. Dover Publications. J.. 1965. North-Holland Publishing Company. 96(1960). Amsterdam. Princeton University Press. 1971. 1965. Kurosh. Lectures in General Algebra. Georg Cantor: His Mathematics and Philosophy of the Infinite. Transactions of the American Mathematical Society. Model Theory. New York. Undecidable Theories. New York. 1972. Florida. A. George Allen and Unwin. London. Frege.W.G. reprinted by W. Mathematical Logic Springer-Verlag. Chang and Keisler. and Zakon. H. Academic Press. Robinson.Bibliography [Bell] [Cantor] Bell and Slomson. 1979.W. Inc. New York. Princeton. Amsterdam. G. A.

Index of Symbols Logic φ ∧ ψ. 101 φ → ψ. 27 QCn . 94 . 87 ˆ A. y). 103 T φ. 32 QC. 27 N. 1 x ≤ y. 29 Qn . 32 θf (x). 101 φ ∨ ψ. 32 δ(x). 1 x > y. 66 α(x). 102 x ∩ y. 1 Model Theory T φ. 29 Qf . 31 ∆2 (x). y. 102 x. 103 T1 ≡ T2 . y). 31 ρ(x. 1 x ≥ y. 102 x − y. 32 θ(x). 102 x y. 102 [in + j]. 31 ρ(x). 101 φ ↔ ψ. 103 B(A). 86 AF . 21 An . 31 ∆1 (x). 101 ∀xφ. 31 β(x). 55 Sizing x. 102 |x|. 67 Qn .n . 102 k!. 103 A φ. 94 ˆ Arithmetic k | n. 94 σA [x]. 25 SA . 103 Models and Domains A. 73 Q. 1 x < y. 102 P(x). 101 ¬φ. 102 xk . 103 108 A φ . 45 ˜ C(x). 101 Set Operations x ∪ y. 103 ˙ A. 67 Qf. 54 ZA . 102 x[k] . 103 A ⊆ B. 45 ZA . 45 cvg(x. 102 Size ordering x ∼ y. 101 ∃xφ.

70 Aλ . 94 Node Construction λ. 70 λ • m. 70 ˆ A. 75 λn . 70 aλ .INDEX OF SYMBOLS cvg(x). 70 L(λ). 73 ι(λ). 76 109 . 75 qλ .

46 MODn.m . 26 DIVn . 86 Predicates Abstract(x). 22 BAn . 13 DEF> . 16 DIVJ . y). 16 DEF+ . y). 10 ASYM> . 23 EXCORE. 18 INDISC∼ .Index of Theories Languages LC . 20 LC< . 16 FUNC+ . 46 QUASI. z). 54 Sum(x. 26 CA. 38 CAI. 20 L< . 23 BASIC. 36 ASSOC. 46 PSIZE. 10 DISJ∪ . 10 ATLEASTn . 9 RTf . 9 OUTPACING. 54 ASYM< . 40 SIZE. 10 TRANS> . 36 Indisc(x. 24 SYM∼ . 26 BDIVn . 23 MAX. 25 EVEN. 10 Tf . 9 EXACTLYn . y. 20 Unit(x). 20 LN . 20 Divn (x) . 24 BDIVJ . 13 INF. 47 PCS. 24 DISJ+ . 10 TRICH. 21 110 CSI. 23 BAI. 39 CAIf . 24 DEF∼ . 10 Modn. 10 IRREF< . 46 MIN. 13 CS.m (x). 9 DEF+ . 36 Theories ADIVn . 40 CORE. 20 Timesn (x. 10 REP< . 83 PCA. y). 10 REF∼ . 39 CANTOR< . 1 Atom(x). 40 TRANS∼ . 23 BA. 10 ASYMF . 16 ODD. 10 TRANS< . 25 MONOT. 36 Sm(x. 46 ASSOC2. 13 .

51 Zgm.INDEX OF THEORIES UNIQ− . 46 Zg. 51 111 . 54 UNIQ+ . 54 UNIQ∼ .

87 finite set model. 64 Monk mapping. 39 interpretation. 74 A-near. 94 elementary extension. 21 classes. 27 (n. 104 complete. 36 length. 27 congruous. 74 abelian groups with identity. 14 finite completion. 51 alternating pair. 49 expansion. 103 least common multiple. 52 near. 103 isomorphic. 70 node atom. 105 diverges.m)-partition. 77 node. 70 . 40 infinite completion. 103 completion. 86 language of a theory. 70.Subject Index A-finite set. 103 filter. 48 Monk’s Condition. 94 density. 70 model complete. 70 extension. 104 existential. 49 generated. 51 categorical. 104 molecule. 103 converges. 68 compactness. language of. 21 subsets of N(LN ). 27 n-quasi-congruence class. 103 language of class size (LC< ). 103 isomorphic embedding. 94 basis. 103 charmed n-factor. 87 112 finite. 21 cancellable abelian semigroups with identity. 38 congruence class. 20 common n-factor. 14 functional assumptions. 103 equivalent. 21 classes (LC ). 20 size (L< ). 94 depth. 68 nodal embedding. 47 n-congruence class. 68 class size. 40 consistent. 104 chain. 27 N-semigroups. 103 incongruous. 85 asymptotic density. 40 finite class model. language of. 75 diagram. 39 finite intersection property. 104 existential assumption. 105 extends.

70 . 71 node set. 46 size. 86 τ -reducible. 104 principal. 45. 103 quasi-congruence class. 48 trichotomy. 40 standard finite interpretation. 21 standard interpretation. 40 representation principle. 71 one-one correspondence. 104 primitive formula. 61 prime model. 1 outpaces. language of. 50 union of a chain. 103 universal. 45. 52 total. 40 translation.SUBJECT INDEX node atoms. 87 uniformly τ -reducible. 21 solution. 84 partial. 50 size. Zg. 103 schematic function. 51 Z-groups modulo I. 70. 101 shell. 25 run. 51 113 cancellable abelian semigroups with identity. 21 standard protomodel. 51 simple translation. 52 Z-groups. 74 reducible. 83 outpacing model. 87 proves. 52 0-extends. 104 Z-groups. 30. 91 satisfaction. 27 quasi-logical principles. 11 quasi-nodal. 40 PCA. 48 simply reducible. 70 node sets. 51 Z-groups modulo I. 12 ultrafilter. Zg. 104 universal-existential. 103 subsets of N. language of. 50 theory. 40 remainder theory. 51 N-semigroups. 50 remainder function. 66 size model. 73 simple order. 12 roughly divisible. 67 submodel. 103 theory of abelian groups with identity.