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

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

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

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

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

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

The term size. 1.2. 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. . But most mathematicians and philosophers do not use cardinal number as a mere abbreviation. the term cardinal number (sometimes just number ) is used most frequently.1. we emphasize that we are not arguing that there really are such things as set sizes. with two principles about set size. and much else. |M |. one which works on the elements of sets rather than the sets . The relation picked out is well studied and well worth the study. ”). 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. the definition is worth making. Cantor’s. Statements about sizes can be translated in familiar and long-winded ways into statements about sets. CANTOR’S ARGUMENT 3 Regarding this last group. Cantor used the terms power and cardinal number rather than size. p. in particular. 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. who offered ONE-ONE as a theory. The technical brilliance of the theory attests to this: it has given us the transfinite hierarchy. They use the term in just the way that we use size and slide freely among (A). . I have identified the standard theory. Neither does it insure that two sets have the same cardinal number just in case they are in one-one correspondence. per se. depends on neither the particular elements it contains nor on how those elements are arranged: The cardinal number. If someone introduces cardinal number as a defined predicate or as part of a contextual definition (e. 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. it’s cardinal number. denying that they are the same does not endanger the theory of one-one correspondences. This is true. Hence. (B). the continuum problem. “We say that two sets have the same cardinal number just in case . [Cantor. Though Cantor’s theory is usually taken as a theory of set size. there is no point in discussing whether that person is right about size.g. of Cantor. it can also be taken as just a theory of one-one correspondences. Cantor reduces it to a second abstraction operator. In the literature. he offered an argument for this theory. indeed. To flesh out this notion of double abstraction. 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.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. In addition. Whether or not similarity is coextensive with being the same size. In stating ONE-ONE and CANTOR< . however. More specifically. and (C). is rarely used in connection with Cantor’s theory. though we will not bother to do so. nor that there are not really such things.

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

that enter into the cardinal number of a set. by the definition of |M |. b} ∧ a = b → |M | = {Abstract(a).1). which entails (1. (1. so to speak. one-membered.1). the argument that a set is similar to its cardinal number relies on (1. So. Generalizing this argument yields (1.4) says that replacing an element of a set with any element not in the set does not affect the cardinality. . A weak version of this principle is: M = {a. and similarity is an equivalence relation. this contradicts ONE-ONEb.5) These two arguments do one another in. consider an arbitrary pair of elements. |M | grows. a and b. the conditions of (1. But abstractions of elements are just units. distinct. since a set is similar to its cardinal number. out of M in such a way that from every element m of M a special unit of |M | arises. ONE-ONEb is true. the correspondence from M to its cardinal number will be many-one and not one-one. (1. Unless each element of a set abstracts to a ‘special’. 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. by (1.e. Abstract(b)} ∧ Abstract(a) = Abstract(b) (1. So Abstract(a) = Abstract(b).2. But. Let M = {a} and let N = {b}. But. Cantor says. So Cantor’s argument for ONE-ONEa only works by assigning all nonempty sets the same. That is. but only their abstractions.1. (1. two sets with same cardinal number are similar.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. It is not the elements of a set.5). .6) ∀x∀y(Abstract(x) = Abstract(y)) For. cardinal number. i. so one is much the same as another. But |M | = {Abstract(a)} and |N | = {Abstract(b)}.4) are met and |M | = |N |. Conversely. unit. because . this means that (1. one element is much the same as another.7) ∀x∀y(x = y → Abstract(x) = Abstract(y)) .6). Thus we can say that M ∼ |M |.

Kossak. There is no way to repair Cantor’s argument. for example.6) or its negation.1). 1.8) contradicts ONE-ONEa. then subsequently the concept of putting together (or collecting. “that numerical equality or identiy must be defined in terms of one-one correlation. The word ‘unit’ is admirably adapted to conceal the difficulty . But if the abstractions of any two elements are distinct. and Cantor. . neither accepted Cantor’s argument. is true.8) ∃x∃y(Abstract(x) = Abstract(y)) So one of (1.8). if a and b have distinct abstractions. We start by calling the things to be numbered ‘units’ without detracting from their diversity. 73-74]. INTRODUCTION assuming only that any two objects can constitute a set. p. 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.8) must be true.6 1. then no two sets have the same cardinal number as defined by (1.6) is: (1. . Frege cites Schroder. (1.305] and similar enthusiasm for his theory (see the quote at the beginning of Chapter 2. . despite the fact that they are in one-one correspondence. and that is not the number. or annexing. He neatly summarizes the difficulty with such views: If we try to produce the number by putting together different distinct objects. [Frege. seems in recent years to have gained widespread acceptance among mathematicians” [Frege.6) contradicts ONE-ONEb. Similarly. “This opinion”. We have shown that (1. contra ONE-ONEa. {Abstract(a)} and {Abstract(b)}. So ONE-ONE is false whether (1. says Frege. while the concept word ‘unit’ changes unperceived into the proper name ‘one’. the result runs perpetually together into one and we never reach a plurality .6) and (1. by putting together identicals. The negation of (1.3 Cantor and the logicists Though both Frege and Russell accepted Cantor’s theory of cardinality.) . Cantor’s definition of cardinal number is sufficient to refute the principle. or whatever we choose to call it) transforms itself into arithmetical addition. But if we try to do it in the other way. pp. then {a} and {b} have distinct cardinal numbers. . (1. Frege spends an entire chapter of the Grundlagen mocking mathematicians from Euclid to Schroder for defining numbers as sets of units. Russell displays similar caution about Cantor’s argument [Russell. 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. Rather than leading to a justification of ONE-ONE. or uniting.

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. 1. Second. for example. Cantor’s theory clears the way for other reductions.4. offers an account of where the intuitions come from (viz. they can advance it as a great lesson for simple common sense. to a large extent. Chapter 3 reorganizes those principles into a tidy set of axioms. So the notion of cardinality is born reduced. 304-305] So Russell. logic) they have no obligation to defend it and can steer clear of peculiar arguments about units . 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. Chapter 2 canvasses common sense intuitions for some basic principles about set size. But there are two specific reasons that they should have seized upon ONE-ONE and CANTOR< . and mines this source for additional principles.” [Russell. the logicists have it both ways: adding ONE-ONE as a formal definition to set theory (or. pp. Such points may be interesting. First. Russell. The logicists’ adoption of Cantor’s theory of cardinality needs no great explanation: it came with set theory and. motivated set theory and determined its research problems. only upon their number. at the same time. who had “taken” number “to be a primitive idea” and had to rely on “the primitive proposition that every collection has a number. and similarly Frege. [Russell. relied upon the principle of abstraction to obtain a ‘formal definition’ of cardinal numbers. for example. But. in contrast to Cantor. In addition. 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.305] So. no matter how they are intended. since the existence of one-one correspondences will be preserved under a one-one mapping. you wish to reduce ordered pairs to sets. Frege and Russell cleaned up Cantor’s presentation of the theory. we might feel tempted to add. known facts about finite sets). p. so each ordered pair must be identified with a distinct set. that is to say. but in the absence of an alternative theory of size they are less than convincing.1. Suppose. as they would call it. then this second condition follows from the first. Well. the relative sizes of sets of ordered pairs should be preserved under translation. Chapters 4 and 5 prove that the theory so obtained is . they both have the form of definitions. if ONE-ONE is the correct theory of size. depending. independent of its terms and their order. One of the relations that has to be maintained is identity . This dissertation presents such an alternative. AIMS AND OUTLINE 7 Of course.

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

. it ought to commit suicide in despair. 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. . the second that there should not be too few principles. pace Cantor. or is there a way of elaborating on common sense to get a plausible and reasonably adequate theory of cardinality? To be plausible. I want to see how far we can go without such principles. since I consider that. then x is smaller than y. [Russell.2 The General Theory The possibility that whole and part may have the same number of terms is. If x is a proper subset of y. I do not propose to help it. shocking to common sense . though it need not be complete. p. In the Introduction. 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. 2. 358] Is common sense confused about set size. as Russell says. To be reasonably adequate. First. i. in the face of the proofs. therefore. In this chapter.1 The Theory CORE SUBSET. I argued that a coherent theory of cardinality has to contain some principles that refer to the kinds of objects in sets. Common sense. consistency). it must be confessed. a theory should at least avoid principles and consequences which violate common sense.e. there is the SUBSET principle: 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 . is here in a very sorry plight. Plausibility and adequacy are conflicting demands: the first says that there should not be too many principles (no false ones. however. it must choose between the paradox of Zeno and the paradox of Cantor.

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

where F is to be replaced by an adjective from which comparatives can be formed. that each is a logical truth.1.2. though the schema for INDISC∼ would have to be restricted to triples of corresponding comparatives. are inconsistent with SUBSET. The other principles might be defended in the same way. ‘short’. taken together. for fear of inconsistency. If x is F -er than y. ‘tall’. it would require taking positions on many questions about logical form and grammatical form which would take us far afield and. there are some instances of the schemata that make for embarassing counterexamples: ‘further east than’ (in a round world) and ‘earlier than’ (in. for example: is smaller than. First. THE THEORY CORE Theory 2. I can just barely imagine presenting a theory which. a possible world). If x is smaller than y. antagonize first-order logicians. is the same size as. 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. This sentence can be regarded as an instance of the schema: ASYMF . that would be good reason to say that there is no . withheld judgment on one or more of these principles. 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. then y is not F -er than x. e.1. I’m told. ‘happy’. but not ‘unique’ or ‘brick’. So. is larger than. then y is not smaller than x. Consider the first principle. Second.g. But to do so without good reason would be counterproductive. But using such observations to support these principles would be problematic for two reasons.1. possibly. it might be that casting the principles above as instances of the appropriate schema would only explain why they are part of common sense. 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. so it could be maintained that each is true in virtue of its form. y) x∼x x∼y→y∼x (x ∼ y ∧ y ∼ z) → x ∼ z x>y↔y<x 11 where Indisc(x. It seems that if a case could be made that these statements. It appears that every instance of this schema is true. in unregimented English: ASYM< . for example.

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

I want to estimate how far we’ve gone. by SUBSET. So x < y. .3. (SUBSET) (DEF∼ ) (TRANS∼ ) (DEF> ) (IRREF< ) (TRICH) x⊂y→x<y x ∼ y ↔ Indisc(x. REP< . by INDISC∼ . Let T = QUASI. if x ∼ x and x ⊂ y. There are more principles to come. then x < y. By REP< . First. Proof. Conversely. then ¬(x ∼ y). The entire set of principles already adopted are equivalent to the following. by INDISC∼ . which is logically equivalent to the conjunction of INDISC∼ and its converse ∼ INDISC: (INDISC∼ ) (∼ INDISC) ∼ x ∼ y → Indisc(x. Suppose ¬(x ∼ y). by INDISC∼ and ASYM< . CORE. Then T ≡ CORE. which will be referred to as the core theory. but before proceeding. Theory 2.1. We only need to show that ∼ INDISC is entailed by T .1. y) → x ∼ y INDISC says that if two sets fail to be the same size.1. must be the same size as x. T CORE. I want to reduce the principles mentioned above to a tidy set of axioms. I’d like to take stock of what we already have. But x < y. SUBSET. y) Indisc(x. the core theory. then it is entailed by the principles we have already mentioned: If x < y. by SUBSET: so x < y .2. 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 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. Second. by TRICH. say x . or larger than one but not larger than the other. But ¬(x < x) and ¬(y < y). 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.2. So x < y or y < x. by IRREF< . some proper subset of y. THE THEORY CORE 13 a principle that Cantorians could complain about. Theorem 2.

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

˙ ˙ b. |b | = n + 1 A (b < b) A (b ∼ a) A (a < b) |b| = n + 1 by (*). and c. If x ∈ A and y ⊂ x. .1. We shall prove (*) by induction on n: (*) Suppose then But if then And if then So So If |a| = n.2. contra our hypothesis. THE THEORY CORE ˙ a. by REP< by induction. then y ∈ 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). If x ∈ A. a < b iff |a| < |b| 15 Proof. by condition (b) by SUBSET contra our supposition. 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. A Then A CORE. b ∈ A. 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. then x is finite.

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

CORE + ADDITION Proof.2. x ∪ y ) Sum(x . CORE + ADDITION Proof. Suppose then and So So So (x ∼ x ∧ y ∼ y ∧ x ∩ y = ∅ ∧ x ∩ y = ∅) Sum(x. then Sum has to be definable by DEF+ . y. as before. y . z) x<z∨x∼z x ∼ z. y .2. z) Sum(x. Sum(x . x =z y = ∅. (DISJ∪ ) (x ∼ x ∧ y ∼ y ∧ x ∩ y = ∅ ∧ x ∩ y = ∅) → (x ∪ y) ∼ (x ∪ y ) (Note: The proofs in this chapter will use boolean principles freely.) Theorem 2. y. z) y ∼y x<z x ⊂z x ∼x y =z−x y ∼ y.2. Theorem 2. DISJ∪ . despite the fact that we have not yet introduced them. x ∪ y) Sum(x . DEF+ by MONOT by DISJ+ by FUNC+ (a) by FUNC+ (b) by REP< . ADDITION OF SET SIZES 17 A model of CORE must satisfy an additional principle. y. .3.2. 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. (⇒) Suppose So But if Let and then So So And if pick with Let But Sum(x. x ∪ y) Sum(x . if Sum is to be interpreted in a way compatible with ADDITION. y .2. y .

z) by DISJ+ by FUNC+ (a) by FUNC+ (b) Theory 2.4. THE GENERAL THEORY (x ∼ x ∧ y ∼ y ∧ x ∩ y = ∅ ∧ x ∪ y = z) Sum(x . is CORE ∪ ADDITION. → 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) . 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.18 (⇐) Suppose then So So 2. z) Sum(x. z) Sum(x. the extended core. The proofs are elementary. y.5. Fact 2. y .2.2. y . EXCORE.

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

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

is the first order language containing just the non-logical constants in LC and L< . . a ∼ b iff |a| = |b|. b. we define the theory of class size in terms of interpretations of LC< over finite power sets. and ∪. and the three place predicate Sum. If LC ⊂ L. If A is a standard interpretation and A = P(x). is the set of all sentences of LC< which are true in all standard finite interpretations of LC< . Following the strategy outlined above.. Cantor’s principle ONE-ONE will not be included in the theory CS. the two-place predicates < and ⊂. and (ii) A assigns the usual interpretations to all constants of LC : A(I) = x. then x is the basis of A. and Unit(a) iff |a| = 1 Definition 3. −. CS . ∩. By drawing only on principles which can be stated in LC< we have at least ruled out the most obvious danger of paradox. then A is a standard interpretation of L iff ˙ (i) A = P(x).3. and the two place operation symbols. Since the notion of one-toone correspondence cannot be expressed in this language. the language of class size.1. and c. the language of size. Definition 3. LC< . a ⊂ b iff a ⊂ b A ˙ b. b.THE THEORY OF CLASS SIZE 21 place predicate ⊂. Definition 3.1. B(A). A A A a < b iff |a| < |b|. L< . c.3.1. A is a standard finite interpretation of LC< iff a. even though it is true over any finite power set.2. A is a standard interpretation of LC< . A has a finite basis. for some x. is the first order language with the oneplace predicate Unit. a. the theory of class size. CS.1. A(∅) = ∅.4.

they must be atomic boolean algebras and. for CS. BA. 9. so.1.2. must satisfy BA: Definition 3. 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. BA is clearly not a complete axiomatization of CS. after a slight detour in Chapter 4. since BA does not involve any size notions. in which the universe is an infinite power set. Since all of the universes of the interpretations mentioned in the definition of CS are power sets. Section 2 develops a set of axioms. Section 3 outlines the proof that CA does indeed axiomatize CS.22 3.Axioms for atomic boolean algebra We’ll begin with the obvious. 151].Axioms for CS Here we shall develop a set of axioms. it has infinite models.2. But BA does entail all the sentences in CS which do not themselves involve size notions. 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. . 3. CA.2 CA . the theory of atomic boolean algebras. p. This proof is presented in Chaper 5.1 BA. The present chapter is devoted to getting a clearer picture of the theory CS.28. This will be done in several stages. In Chapter 6 we show that such models do exist. CA. Def 9. A FORMAL THEORY OF CLASS SIZE Since CS has arbitarily large finite models.3. 141 and Def. It is not obvious that CS has standard infinite models. for the theory CS. To show this we need to draw on some established facts about the complete extensions of BA (in the language LC ). p. 3.

. then ¬φ is consistent with BAI. / If A is infinite. BAn is complete.2. 23 a.152] b.4. so BAI ¬φ. EXACTLYn is a sentence which says that there are exactly n atoms: ATLEASTn ∧ ¬ATLEASTn+1 c. A. CA . ATLEASTn is a sentence which says that there are at least n atoms: ∃x1 . Prop.3. Proof. Fact 3. p.360] e. (Immediate from above) c. 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. BA ∪ INF φ. / . BAI and the theories BAn . For n ≥ 1. Hence. By 3. 9. Cor 9. If BAI Proof. 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. then. φ. then ¬φ is true is some atomic boolean algebra. ¬φ is true in some standard finite interpretation of LC and. By compactness. BAn is categorical. again..∃xn (Atom(x1 ) ∧ ..151] d. For n ≥ 1. EXACTLYn e. For n ≥ 1. p.24. But BAI is complete. are the only complete. φ ∈ CS. Theorem 3. so φ is not true in A and φ ∈ CS. p. then φ is true in some finite model of BA.2.AXIOMS FOR CS Definition 3. If A is finite. ¬φ is true in some finite model A of BA. For n ≥ 1. an n-atom atomic boolean algebra is isomorphic to any standard finite interpretation of LC with an n-element basis. a. [Monk. BA is complete. [Monk.2. then ¬φ is true in A . n ≥ 1.2. consistent extensions of BA. then BA φ.4.2. [Monk. Theorem 21.30.5.3.2.2. If CS φ.32.. ∧ Atom(xn ) ∧ { xi = xj | 0 < i < j ≤ n }) b. BAn = BA. If BA φ. But. then A is isomorphic to some standard finite interpration A of LC .

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. we’ll need some non-standard models of BASIC.24 3. To show that the new principles really do need to be added. The models will. though we will not prove that CA ≡ CS until Chapter 5. CA. we will have an effective set of sentences.7. we obtain our first serious attempt at a general theory of size: Definition 3. 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. .6.3 Division principles BASIC is not a complete axiomatization of CS. include different subsets of N and assign different size orderings to these sets. 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.2 Size principles Here. these models of BASIC will reappear as submodels of various standard models of CS over P(N). In this section we shall exhibit an infinite number of principles which need to be added to BASIC in order to axiomatize CS. A FORMAL THEORY OF CLASS SIZE 3. is defined as: BA ∪ SIZE 3.2. In Chapter 6. the basic theory. y. So. however.2. and (3) all boolean symbols will receive their usual interpretations. we just gather the principles presented above as EXCORE: Theory 3. the Class Size Axioms. y) ¬(x < x) x<y∨x∼y∨x>y Unit(x) ↔ Atom(x) Sum(x.2. (2) their atoms will be singletons in P(N). When we are done. BASIC. These models will be similar in that (1)their universes will be subsets of P(N).2.

this statement is in CS but is satisfied by neither our original model nor the model as amended. we can extend the model and again we can produce a statement of CS which is false in the resulting model. 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. (EVEN∨ODD) says that the universe is roughly divisible by two: EVEN says that the universe is divisible by two without remainder. We now formalize this line of reasoning. But this model will still not satisfy CS.3.2. but not entailed by BASIC. in which case they cannot be disjoint. As you might suspect. So EVEN∨ODD is in CS. or both cofinite. So.2. 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. F is a model of BASIC.8.e. ODD says that there is a remainder of a single atom. in which case their union is also finite. 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. we can construct an infinite model of BASIC . As with (EVEN ∨ ODD). Definition 3. CA . where. we also need to include all sets which are near the set of evens or the set of odds. Consider the model F whose universe consists of all only the ˙ finite and cofinite subsets of N. (EVEN ∨ ODD). as we can see by generalizing the argument above.AXIOMS FOR CS 25 Every standard finite interpretation. the new model will be closed under boolean operations and will satisfy BA. Informally. this is just the tip of the iceberg of principles missing from an axiomatization of CS. We can construct a similar statement that says the universe is roughly divisible by three — with remainder 0. A. To round out the result to a model of BASIC. With these additions made. i. for any two sets that are the same size are either both finite. 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. Again. for a. then . those that differ from the evens or odds by a finite set. or 2. 1. But BASIC (EVEN ∨ ODD). But neither EVEN nor ODD is true in F. If 0 ≤ m < n.

BASIC CS. . BASIC DIVn .10. xn ) ∪ (y1 ∪ . then a. Definition 3.26 a.2. . . . DIVJ = { DIVn | n ∈ J } b. but it does not work. by 3. ∧ Atom(ym ) ∧ 1≤i<j≤n (xi ∩ xj = ∅) (yi ∩ yj = ∅) 1≤i<j≤m ∧ ∧ (x1 ∪ .n−1 Fact 3. . ym ) = I) MODn. If 0 ≤ m < n and A is a standard finite interpretation of LC< . . . F defined above satisfies BASIC but not DIVn . xn ) ∩ (y1 ∪ .. . .2.2. then (a) (b) (c) A A CS MODn. . . ∼ xn ∧ Atom(y1 ) ∧ .9.9c. If n > 1. we shall include sets which can be regarded as fractional portions of N.11. . MODn. We could consider adding all DIVn sentences to BASIC in the hope that this would yield a complete set of axioms for CS. We did consider this. So. To demonstrate this. .2. If J is a set of natural numbers. we need some independence results for sets of DIVn sentences. . BDIVJ = BASIC ∪ DIVJ c.0 ∨ .m iff |B(A)| ≡ m mod n DIVn DIVn CS Theorem 3. . BDIVj = BDIV{j} Our independence results will be obtained by constructing models of BASIC which satisfy specific sets of DIV sentences. A FORMAL THEORY OF CLASS SIZE ∃x1 . BASIC Proof.m is the sentence 3. To build such models from subsets of N.m says that the universe is divisible into n sets of the same size with m atoms remaining. DIVn is the sentence MODn. for any n finite sets have a finite union and any two cofinite sets overlap. .∃xn ∃y1 . b. ym ) = ∅ ∧ (x1 ∪ . The model. ∨ MODn. . ∃ym ( x1 ∼ x2 ∼ ..

d. [2n]. N itself is the union of n-congruence classes. b. b. . Proof. . . QCn = { x | x is an n-quasi-congruence class } f. . then x ∪ y ∈ QCn . 27 a. Fact 3.13. and the set of odds. If x is the union of m of these classes. x is a congruence class iff x is an n-congruence class for some n. . a. where 0 ≤ m < n. N is an n-quasi-congruence class for every n > 0: N = [nk + 0] ∪ . [3k + 2] is a 3-congruence class. c.12. NEAR(x. Suppose x = a1 ∪ . w1 and w2 such that x = (y ∪ w1 ) − w2 . x is a quasi-congruence class iff x is an n-quasi-congruence class for some n. The set of evens. ak ∪ b1 ∪ . ak and y = b1 ∪ .2. . ∪ [nk + n − 1] Fact 3. a. [2n + 1]. n>0 QCn Definition 3.15. then N − x is the union of the remaining (n − m) n-congruence classes. c. iff x − y and y − x are both finite.2. . a. then N − x ∈ QCn . If x ∈ QCn and y ∈ QCn . QC = Examples. b.2. x is an n-congruence class iff x = [nk + m] for some m. . b.14. bj . e. x is an n-quasi-congruence class iff x is the union of finitely many n-congruence classes. x is near y. If x ∈ QCn .y). .AXIOMS FOR CS Definition 3. For n ≥ 0. d. bj .2. . CA . N is a 1-congruence class.3. x is near y iff there are finite sets. . Then x ∪ y = a1 ∪ . are 2-congruence classes.2.

But (x2 − y) is finite. ⇐ Suppose x = (y ∪ w1 ) − w2 . then y is near x. so (x − y) is finite. NEAR is an equivalence relation. Proof. 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).28 Proof. Then x − y ⊆ w1 and y − x ⊆ w2 . If x is near y. since x − x = ∅. x is near x. c. c. then x is near y. But (y − x1 ) is finite. so x − y and y − x are finite. x1 is near y. which is finite.16. If x is near y. a. (y − x) ⊆ (y − x1 ). A FORMAL THEORY OF CLASS SIZE ⇒ Let w1 = x − y and let w2 = y − x. If x1 is near x2 and y1 is near y2 . . Proof. If x1 ⊆ x ⊆ x2 . Fact 3. then x1 ∪ y is near x2 ∪ y. Similarly. then x1 ∪ y1 is near x2 ∪ y2 . If x1 is near x2 . Suppose x is near y and y is near z. a. So the union. x is near z. b. (z − x). Fact 3.17. 3. Since x1 ⊆ x. So (x − z) is also finite. is finite. so (y − x) is finite.2. Hence. which is finite.2. (x − z) = ((x ∩ y) − z) ∪ ((x − y) − z) where ((x ∩ y) − z) ⊆ (y − z) and ((x − y) − z ⊆ (x − y). and x2 is near y. Fact 3. then N − x is near N − y Proof.2. Since x ⊆ x2 .18. (x − y) ⊆ (x2 − y). b. Immediate.

By (a). But (x1 ∪ y) − (x2 ∪ y) ⊆ (x1 − x2 ) and (x2 ∪ y) − (x1 ∪ y) ⊆ (x2 − x1 ) 29 b.2. c. Using Fact 3. Qn = { y | y is near an n-quasi-congruence class } b. (N − x) − (N − y) = y − x and (N − y) − (N − x) = x − y. If x ∈ A. b.20: a. A ∈ A. 9. then x ∪ y ∈ A. then Qn forms an atomic boolean algebra under the usual set-theoretic operations. If n > 0.2. Definition 3. p. We can now define the domains of the models we will use to establish the independence results.3. then A forms a boolean algebra under the usual set-theoretic operations. Q = Examples.1. {Qn } = N. p. then Suppose and So and So x is near y y ∈ QCn N − y ∈ QCn N − x is near N − y N − x ∈ Qn . cofinite. Proof. A − x ∈ A. If A is a class of sets such that a. x1 − x2 and x2 − x1 are finite.2. x1 ∪ y1 is near x2 ∪ y1 .13b by 3. Q2 = { y ⊆ N | y is finite.2. b. So if x and y are near each other. a. Since x1 is near x2 . where I is interpreted as A. near [2n] or near [2n + 1] }.20. by transitivity. since N ∈ {Qn } and if x ∈ Qn .AXIOMS FOR CS a.2.141 and Corr 9. Fact 3. and n>0 Qn b. So x1 ∪ y1 is near x2 ∪ y2 . If x ∈ Qn . Q2 is the domain of the model constructed above to satisfy DIV2 . Def.2. then x ⊆ N. which is near x2 ∪ y2 .19.2. CA . a. so are their complements.18b . by 3. Q1 = { y ⊆ N | y is finite or cofinite }.142] Theorem 3. [Monk.21.4. If x ∈ A and y ∈ A. then c.

The δ value assigned to x is the finite number of elements added to or removed from x to obtain x.18b Thus. Fact 3.22. a size will be assigned to a set only if 0 ≤ ρ ≤ 1. The first member is a rational between 0 and 1 which represents the density of the set. We now define a size function on all sets which are near quasi-congruence classes.13a by 3. The definitions and facts below formalize this intention and demonstrate that the assignment of sizes to sets is well-defined. a. every singleton is in Qn since all singletons are near ∅. then δ ≤ 0. assigned to x.2. Specifically. δ1 + δ2 Only some of these sizes will actually be assigned 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. Suppose and then and and and But then and So x ∈ Qn y ∈ Qn x is near x 3. So Qn is an atomic boolean algebra. The sizes assigned to sets are ordered pairs. ρ. δ2 are sizes. If θ1 = ρ1 .30 c.2. Our intention in assigning sizes to sets is as follows: Suppose x is near an n-quasi-congruence class. 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. δ1 and θ2 = ρ2 . Definition 3. then δ ≥ 0 and if ρ = 1. (ii) θ1 + θ2 = ρ1 + ρ2 . if ρ = 0. 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. The set x has density k/n and this is the value. then (i) θ1 < θ2 iff ρ1 < ρ2 or (ρ1 = ρ2 ∧ δ1 < δ2 ). Qn is a boolean algebra. δ . A size is an ordered pair ρ. x . Moreover.2. Moreover.2. b. So x is the union of k ≤ n n-quasi-congruence classes. where ρ is rational and δ is an integer.23.

24.3. If x ∈ QC. / xi = yj . . Any set is near at most one quasi-congruence class.AXIOMS FOR CS 31 a. b. then y − x is infinite.2.) b. a ∈ yj . since xi is infinite Similarly. ∪ yk Suppose then But So So So a∈x−y a ∈ xi for some i. then x is not near y. y ∈ QC. . then a. since NEAR is transitive. if a ∈ y − x. ∆1 (x) = x − C(x). (That is.2. the two would have to be near each other. ∪ xj and y = y1 ∪ . 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. Definition 3. . b. But this is impossible by (a). c. If x ∈ QC. for any y. no two quasi-congruence classes are near each other. . b. α(x) = the least n such that x ∈ QCn . If x were near two quasi-congruence classes. . β(x) = the unique k such that x is the disjoint union of k α(x)congruence classes. ∆2 (x) = C(x) − x.2.25. CA . and x = y. a. ∆1 (x) and ∆2 (x) are finite and x = (C(x) ∪ ∆1 (x)) − ∆2 (x) Definition 3. If x is near a quasi-congruence class. then a. Proof. C(x) is the quasi-congruence class near x.

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)

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. Qn Proof.AXIOMS FOR CS So. by (i). y. z) θ(z) = θ(x) + θ(y) θ(x) ≤ θ(z) θ(x) = θ(z) θ(y) = 0.2.3. CA . we know that θ(x ∪ y) = θ(x) + θ(y). 0 y=∅ x = z.30. DEF+ : Suppose So Clearly Assume then So Let Assume then and Let So Claim For But So Qn Sum(x. 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.2. DIVm iff m | n. . 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.) Conversely.

Based on Theorem 3. . c. bm = N. then the least common multiple of J. DIVm . If BDIVn b. Remark. Qµ(J) BDIVJ since it satisfies DIVj for each j ∈ J. Corollary 3. ⇐ For each i. 0 ≤ i < n. Hence. 0 . If BDIVJ DIVm . If m n. 0 . then Ai ∩ Aj = ∅ and Qn 1/n. we have bj ∈ Qn and θ(bj ) = p/n. if m µ(J). Definition 3. then Qµ(J) DIVm . (Ai ∼ Aj ) since θ(Ai ) = θ(Aj ) = Letting p = n/m. the product is greater than µ(J). Such sentences can be produced by generalizing the notion of divisibility to all sets instead of applying it only to the universe. Definition 3.33. . then Qn BDIVn . So N = If i = j. We are now ready to show that BDIVn CS by finding a sentence in CS which entails infinitely many DIVn sentences. group the n sets Ai into m collections with p members in each: B1 . they must each have size 1/m. if m disjoint sets of the same size exhaust N.36 3. . then m | n. Immediate from (b). b . But if x ∈ Qn .3.2. is the least k which is divisible by every member of J. 0 = Furthermore. then a. .2. . 0 . Bj for 1 ≤ j ≤ m. There are only finitely many m for which BDIVJ Proof. Usually. ¬DIVm b. c. 0 .32. let Ai = [nk + i]. µ(J) always exists since the product of all members of J is divisible by each member of J. But. . then m | µ(J). If J = ∅ and J is finite. since only finitely many m divide µ(J). . for integral a and b. A FORMAL THEORY OF CLASS SIZE ⇒ θ(N) = 1.31. then θ(x) = a/n. µ(J).2. If J is finite. DIVm . Bm Letting bj = 1/m. b1 ∪ .20: a. So b = 0 and a = n/m. 0≤i<n Ai .

CS ADIVn . DIVnm .m ≡ Modn. c. b. x.m (I) and DIVn ≡ Divn (I) Fact 3. and all finite sets are roughly divisible by every n. and x ∈ A.2.m (z) is the formula: ∃x∃y∃v∃w[T imesn (x. By induction on m: if m = 1. A T . for all m.35.AXIOMS FOR CS a. . If 0 ≤ n. Thus. Notice that in the presence of BASIC. ADIVn is the sentence ∀xDivn (x) Remark. we have partitioned x into nk n sets of the same size and nk j + i atoms. . for every n. since i < nk and j < n. ∨ Modn. then Timesn (x. Divn (z) is the formula Modn. A ADIVnk+1 . then x can be partitioned into nk sets of the same size and i atoms.2. a. Proof.34. y) says that y is the same size as the Sum of n sets.0 (z) ∨ . y) is the formula: ∃x0 . ADIVn : a. so ADIVn ADIVnm . each the same size as x.2. w. ∃xn [x0 = ∅ ∧ xn = y ∧ 1≤i≤n 37 Sum(xi−1 . DIVn . then nm = n. w) ∧ Sum(v. then Modn. for all m b. ADIVnm . If 0 ≤ m < n. Hence. ˙ If T ADIVnk . and c. We have taken this opportunity to formulate the divisibility predicates purely in terms of size predicates. . Every set in every standard finite interpretation is a finite set. where j < n. Fact 3.n−1 (z) d. . where i < nk . v) ∧ Unit(y) ∧ T imesm (y.3. xi )] Timesn (x. But nk n = nk+1 and. MODn. z)] Modn. BASIC. Proof. . CA .m (z) says that z can be partitioned in n sets of the same size and m atoms. Each non-atomic set in the partition can be further partitioned into n sets of the same size and atoms. nk j + i < nk+1 .

complete extension of T2 . indeed. by Lindenbaum’s lemma. a complete set of axioms for CS. Proof. a.3. Theorem 3.2.32. Theorem 11. BDIVN ADIVn for any n > 1. a.3. But T is not consistent with T1 . it is reasonable to attempt an axiomatization of CS as follows: Definition 3. p.2. Then T = T2 .1. T2 φ. Suppose that T1 φ. Obvious. CA ≡ BASIC ∪ { ADIVn | n > 0 } The remainder of this chapter and the next two are devoted to showing that CA is. 3.38 b. [Monk.35c.2. But then BDIVJ DIVnk for every k. Definition 3. T . 3. complete extension. But this contradicts Fact 3. complete extension. which says that BDIVJ entails only finitely many DIVn sentences. Fact 3. by Fact 3. ¬φ is consistent. it will be sufficient to show that every consistent extension of CA is consistent with CS. Since this weakness has arisen in the case of ADIV sentences.2.3 Remarks on showing that CA axiomatizes CS We know that CS CA and we want to show that CA ≡ CS.36. we are left with a theory weaker than CS.2. If BDIVN ADIVn . even if we add all of the DIV sentences to BASIC. i. T has a consistent. complete extension of T2 is consistent with T1 . If every consistent. Immediate from (a) and (b). then T2 T1 . b. Proof. To do so.37. T is a completion of T iff T is a complete. consistent extension of T . then by compactness there is a finite set J such that BDIVJ ADIVn . So. . T is also a consistent. A FORMAL THEORY OF CLASS SIZE c.200] b. (Lindenbaum’s lemma) Every consistent theory has a consistent. that CA CS.e.13. Since T2 ⊆ T .

T is an infinite completion of T iff T is true in some infinite model of T .3. So T ¬ATLEASTn+1 and has no infinite models.3. T is a finite completion of T iff T is true in some finite model of T. . ⇒ Suppose A is a finite model of T with n atoms. are the completions of CAI? Recall that CA entails DIVn for every n > 0. CA BASIC and. b. Proof. . then (T is a finite completion of T iff T is not an infinite completion of T .EXACTLYn . If T is a completion of T and BA ⊆ T . The model A of CA. EXACTLYn is a standard finite interpretation. Definition 3. Remainder theories specify.4. Fact 3. where DIVn is MODn. for each n. for some n.6. CAI = CA ∪ INF b. b. so T has no finite models.3. REMARKS ON SHOWING THAT CA AXIOMATIZES CS 39 To prove that every completion of CA is consistent with CS. ⇐ T ATLEASTn for every n.3. T .5.1. Definition 3. then. CSI = CS ∪ INF So. a. So A CS. a. the number of atoms remaining when the universe is divided into n disjoint sets of the same size. . of CAI has to solve the disjunction DIVn for each n. Every finite completion of CA is equivalent to CA. by Theorem 2. Every finite completion of CA is consistent with CS.3. 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. b. T has to entail one of the disjuncts.) Proof. we define two kinds of completions of a theory. Fact 3. ∨ MODn.3.0 ∨ . a.6. BASIC is categorical in every finite power. T EXACTLYn . What.n−1 Any completion. a. Since T is complete.3.

g. Theorem 5-11. e. f.3. of f is congruous. Hence. a. c. then. [Griffin. then f has a solution iff f is congruous iff f has infinitely many solutions. c. c. CSf is consistent. Remainder functions and remainder theories a.3.8b. (⇐)If f is congruous. d. So CSg is consistent. CAIf is complete and that these are the only complete extensions of CAI. Tf = T ∪ RTf . (⇒)For every finite restriction. p. Proof. CSg is consistent. A FORMAL THEORY OF CLASS SIZE Definition 3. f is congruous iff for any i. (Let f (p) = p − 1. If T is a theory. f is finite iff Dom(f ) is finite. Chapter 5 will show that if f is total.8a. By (a). f is congruous iff every finite restriction of f is congruous. There are congruous f without any solutions. Fact 3. . RTf .40 3. is the set of sentences { MODn. a. a. n. φ is consistent. (⇐)Every finite restriction. f is congruous by 3. there is some n such that An φ. φ b. b.m | f (n) = m }. for all primes p. g. CSIf is consistent iff f is congruous.80] b. In this section. n ≡ f (i) mod i. (Henceforth f ranges over remainder functions.9. (⇒)Let φ = RTf . f is total iff Dom(f ) = N+ . then CSf is consistent iff f is congruous. of f . f : N+ → N is a remainder function iff 0 ≤ f (n) < n for all n ∈ Dom(f ). CSf is consistent iff f is congruous. each such g is congruous. By compactness.3. If f is finite. otherwise f is partial. we will show that CAIf is consistent just in case CSIf is consistent. So An CS. g. otherwise. The remainder theory specified by f . Any solution would have to be larger than every prime.8. f has a solution.3. then gcd(i.3.) b. j ∈ Dom(f ). So n is a solution of f and. If f is finite. hence. by (a). n is a solution for f iff for any i ∈ Dom(f ). f is incongruous. Since CS.7. f is congruous by 3.) Theorem 3. j) | f (i) − f (j).

. . . Since f is congruous. of f . b1. ak2 .1 . To do this. g is as well.11. then CA MODm.1 . by 3. then CA MODm. and p = k2 n + q. k1 = m/n. REMARKS ON SHOWING THAT CA AXIOMATIZES CS c. f is congruous. But.n ak2 .p . T . b2.1 . our proposed axiomatization of CS. We now want to prove a similar theorem for CA. . .3. . . Hence.n bk1 .p → ¬MODm. . we must first establish that certain sentences are theorems of CA.q . for 1 ≤ i.q .p → MODm.10. .3. . T is consistent. cq For 1 ≤ i ≤ n. Proof. . .8b. B(A) = 1≤i≤n Bi ∪ (c1 . So g has arbitrarily large solutions and CSg has finite models large enough to satisfy T . bk1 .3. . . Suppose A MODm. So. it is sufficient to show that every finite subtheory. . Furthermore. of CSIf is consistent.n b2. So. .n c1 . Lemma 3.1 .n and p atoms a1. B(A) can be partitioned into m sets of the same size b1. 0 ≤ q < n.1 . (⇒)If CSIf is consistent.n ∪ 1≤j≤k2 aj.3.n Since A DISJ∪ .3. if T is such a theory. a1. If 0 ≤ p < q < m. j ≤ n. g.q . If n | m. so is CSf .n a2. let Bi = 1≤j≤k1 bj. then T ⊆ CSg ∪ { AT LEAST i | i < n } for some n and some finite restriction. A MODn.1 . cq ) So. and p ≡ q mod n. . by (b). 41 (⇐)By compactness. . a2. . A Bi ∼ Bj . Lemma 3.

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

and f (j) ≡ q mod k By lemma 3. p = q.3. we have (1) (2) CA MODi.3. CAIf is consistent iff CSIf is consistent. So lemma 3.3.3.3.p .3.1b.3.3. See proof of 3. f (i) ≡ p mod k.q 43 since k | i and k | j. To establish (3). (4) If f is total and congruous.9c.3. .12c. Since k f (i) − f (j). (1) (2) CA CS Every completion of CA is consistent with CS.10. c.3. and CA MODj.3. (⇐)Follows from 3. The main objective is (1).p → ¬MODk.3.f (j) → MODk. (3) If T is a completion of CAI. it is sufficient to prove (4) because every completion of CAI entails CAIf for some total f . then T ≡ CAIf for some total. Corollary 3. (2). which follows from (2) by 3.5c) and that if CAIf is consistent. Let p and q be such that 0 ≤ p. then CSIf is also consistent (see 3.12).11 yields (3) CA MODk.9a since CAf ⊆ CSf . So (2) is a consequence of (3). congruous f . REMARKS ON SHOWING THAT CA AXIOMATIZES CS from which it follows that CAf is inconsistent. See proof of 3. and (3) we may conclude (*). It might help to review our strategy before presenting the difficult parts of the proof that CA ≡ CS.9c and 3. Immediate from 3.13. Proof.9b. We already know that the finite completions of CA are consistent with CS (see 3.q From (1). then CAIf is complete.3.f (i) → MODk. b. q < k.

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

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

PCS. z ). Definition 4.2. 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 . SA . y. the pure theory of class sizes. y. For example. 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]. PSIZE. The one-place operator x is to be read as the complementary size of x: ˜ SA (y = x) iff SA ˜ Sum(x. and Atom(x) SA c. the theory of sizes. y.46 4. if A Sum(x. σA [z]) iff A SA Unit(σA [x]) iff A x∼y x<y Sum(x. Definition 4. then A Sum(x. σA [y]. z). I) These interpretations are well defined because the predicates are satisfied ˙ by elements of A in virtue of their sizes. consists of all sentences of L< which are true in the size model of every standard finite interpretation of LC< .1. z) and A z ∼ z . THE PURE THEORY OF CLASS SIZES b. the size model for A.1.3. y.

4. z) Sum(x. . . z) Sum(x1 . . then SA PCS. the pure class size axioms.1a to show that PCA is model complete and Theorem 4. then T is model complete. MONOT rules out the interpretation in which Sum(i. y2 . Theorem 4.4. y. If T satisfies Monk’s Condition. z) x ≤ z → ∃ySum(x. y.1. w) ∧ Sum(y. k) iff (i + j) = k ≤ n Once n is fixed. y. this is the only interpretation allowed by the axioms PSIZE. z.2.z2 ) → (x1 < x2 ↔ z1 < z2 ) Sum(x. y1 . z) ↔ Sum(y. w2 ) → Sum(x. where SA Unit(x) iff x = 1 and SA Sum(i. j. I) ˜ BASIC.2. that CAI is model complete. w1 ) ∧ Sum(w1 . x. j.1. y. w) ∃zSum(x. then the elements of SA can be regarded as the sequence 0. w2 . z. SOME MODELS AND THEORIES Sum axioms (IDENT) (COMM) (MONOT) (ASSOC) (EXIST+ ) (EXIST− ) (COMP) Fact 4.1. PSIZE fails to axiomatize PCS for the same reason that BASIC fails to axiomatize CS: the lack of divisibility principles. If A CA. . z1 ) ∧ Sum(x2 . z) ∧ y2 ≤ y1 → ∃zSum(x. is defined as: PSIZE ∪ { ADIVn | n > 0 } Fact 4. in chapter 5. PSIZE Sum(x. 4. a. PCA. We offer PCA as an axiomatic version of PCS: Definition 4. n with the usual ordering.1.1b to show.2. x) Sum(x.7. k) is satisfied just in case (i + j) ≡ k mod n + 1.6. then SA PCA If A is a standard finite interpretation of BASIC with n atoms. y. If A Fact 4. In particular.1.5. x.2. 0. . PSIZE 47 Proof.2 Some models and theories We shall use Theorem 4.

f is a Monk mapping. a. in L12 a (quantifier free) formula. assigns to each n-place predicate. and c. So.48 4. and C is a finitely generated submodel of B. then there is an isomor˙ ˙ phism.355] Definition 4. τO .5d in the Appendix. [Monk. . let φ3 be ∃y1 . P . f : C → A such that if x ∈ C ∩ A. Fact 4.359] b.2. So.2. ∃ym φ2 So. then T2 is also model complete.3. . b. by Fact C.2. to obtain the desired isomorphism. p. we use Fact 4. of L2 with exactly (n + 1) free variables. of L2 with exactly n free variables. . . b. . then any finitely generated structure over ˙ ˙ LT is finite. . in L12 a (quantifier free) formula. If LT has no function symbols. . then f (x) = x. If T is model complete and LT has no function symbols. If T is model complete in L. assigns to the universal quantifier a (quantifier free) formula. we use Monk’s Theorem to infer the model completeness of the theory CA from that of PCA. . τ∀ . Suppose L1 and L2 are first order languages and L12 = L1 − L2 . . suppose C contains a1 . then T is model complete in L . an ). b. To establish the model completeness of PCA. O. of L2 with exactly one free variable.2. assigns to each n-place function symbol. If A T . which: a. B T . A ⊆ B. If T1 is model complete and T1 T2 . 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 . . . Proof. . φ3 is a primitive formula. Finally. a. τP . τ . THE PURE THEORY OF CLASS SIZES Monk’s Condition. B φ3 (a1 . then T satisfies Monk’s Condition. p.2. [Monk. so A does as well. . . an (from A) and b1 . . bm ˙ ˙ (from B-A). and L is an expansion of L by adjoining new individual constants. In chapter 5. A translation (simple translation) of L1 into L2 is a function. 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 .

6.2. 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. xn . 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 . 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. then τ (B) is the interpretation A of L1 such that: ˙ ˙ a. The existential assumption of a translation says that that subdomain is non-empty. ∀xn (τ∀ (x1 ) ∧ . y2 ) ↔ y1 = y2 )) where O is a function symbol in L1 but not in L2 .2. Notice that the existential and functional assumptions of a translation are sentences of L2 . b.4. then a. Predicates and function symbols of L2 are translated into themselves. .2. Definition 4. A translation from L2 into L1 induces a mapping from interpretations of L2 into interpretations of L1 . The relevant part of the domain is the set of elements which satisfy the interpretation of universal quantifier.2. SOME MODELS AND THEORIES 49 Definition 4. The functional assumptions of τ are the sentences: ∀x1 . If τ is a translation of L1 into L2 .5.4. . . . 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 . ∧ τ∀ (xn )) → ∃y1 (τ∀ (y1 ) ∧ ∀y2 (τO (x1 . b. . . b. A interprets all predicates and function symbols common to L1 and L2 in the same way that B does.

Lemma 4. A1 Proof. for which T1 is τ -reducible to T2 . . and T1 is uniformly τ -reducible to T2 . If T2 is model complete. yn )) iff A2 τφ (x) Theorem 4. φ. Then.2. . y1 . and A2 ⊆ B2 A1 = τ (A2 ) and B1 = τ (B2 ). ∃yn (τ∀ (y1 ) ∧ . . ∃yn φ (x) where φ (x) is a conjunction of atomic formulae and negations of atomic formulae.2. yn ) ˙ φ (x. . c. . . Suppose that T1 is τ -reducible to T2 and that A1 = τ (A2 ). Suppose φ(x) = ∃y1 . ∧ τ∀ (yn ) ∧ φ (x. for bi ∈ A φ(x) iff A2 τφ (x) τφ (x. . . b1 . .9. . T1 is reducible (simply reducible) to T2 iff there is a (simple) translation. then T1 is also model complete. Definition 4. b1 . there exist models A2 and B2 such that A2 T2 and B2 T2 . . 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). . . . y) A The following condition on theories allows us to infer the model completeness of one from the model completeness of the other. . b. A P (x) iff B y = O(x) iff B τP (x) τO (x. A1 and B1 . for any primitive formula. . bn ).7. a. bn ) ∃y1 . . . and A1 ⊆ B1 . .2. . y1 . of L1 and any sequence. Then A1 φ(x) iff A1 iff A1 iff A2 iff A2 ∃y1 . T1 is uniformly τ -reducible to T2 iff for any models. τ is a simple translation from LT1 to LT2 .8. τ . . . A interprets all predicates and function symbols in L12 in accordance with the translations assigned by τ . If τ is a translation from L1 to L2 . ∃yn φ (x. . x ∈ A1 . THE PURE THEORY OF CLASS SIZES c. .50 4. such that A1 T1 and B1 T1 . .

Zg . and (3) above and: (4 ) x+y =x+z →y =z c.2.2.2. SOME MODELS AND THEORIES 51 Proof. 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.8 We shall now define several theories. φ: If then B1 A1 ˙ φ(x) for x ∈ A φ(x) Since T1 is uniformly τ -reducible to T2 . (ii) The axioms for simple order. A2 ⊆ B2 . where A1 ⊆ B1 and a primitive formula. all more or less familiar.5d in the Appendix. 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.8 since T2 is model complete by lemma 4. 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. has the following axioms (i) The axioms for abelian groups with identity. a. The theory of Z-groups. By Fact C. The theory of cancellable abelian semigroups with identity consists of (1).2. there are models of T2 . .10. Definition 4. it is enough to show that given models A1 and B1 of T1 . (2). which will serve in showing that our theory of size is model complete.4.

Zgm. and (iv) The additional axiom: (12) 0≤x f.291] b. . The theory of N-semigroups is model complete. ∨ ny = x + (n − 1)) for each positive n. a.[Robinson] Theorem 4. (iii) Axioms (9) and (10) of Zg. 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 ∨ . [Chang. . The theory of N-semigroups has the following axioms: (i) The axioms for cancellable abelian semigroups with identity. (ii) The axioms of simple order.12.2. ≤ . where ny stands for: y + ··· + y n times e. a. 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. The theory of Z-groups modulo I. The theory of Z-groups modulo I is model complete. (ii) The axioms for simple order (iii) Axioms (10) and (11) of Zg.52 4. Zg is model complete. p. 291] Fact 4. 0. 1. Zg is the complete theory of Z. consists of the following axioms: (i) The axioms for abelian groups with identity.11.2. b. +. p. . axiom (12) from the theory of N-semigroups.

Moreover. the theory of Z-groups with addition taken modulo some constant (see 4. 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. by reducing it to the theory of Z-groups modulo I. assigning interpretations in the obvious way. It is clear from the construction in [Kurosh] that if the semigroup is ordered. Specifically.3. the (rough) divisibility of the elements in the semigroup will be carried over to the group. Every abelian semigroup with cancellation can be isomorphically embedded in an abelian group [Kurosh. Consequently. Chapter 5 uses Monk’s Theorem to show that CA is model complete. consider the theory of N-semigroups in the language which contains. The model completeness of PCA then follows from the model completeness of Zgm (Fact 4.4. an individual constant. we use Theorem 4.9.2. The theory of N-semigroups is model-complete in this language.9.2. 53 a. we shall show that every model of PCA is the τ -image of a model of Zgm.3 PCA is model complete To show that PCA is model complete. First.11). 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. b. we shall reduce it to Zgm. besides the constant symbols in the original theory. 44-48]. by 4.2. 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. so is the former.2. where τ is the following translation. by 4.2. pp.2b.2.9 to show that PSIZE is model complete.) In the next section. . I.12) and Theorem 4. stack up ω many copies of the model. The result is an Nsemigroup and the original model is isomorphic to the first copy of itself. PCA IS MODEL COMPLETE Proof.

2 gives Sum a functional interpretation. ZA .3.3. y.3. Theorem 4. z). y) abbreviates ∃zSum(x. (Sm(x. on the basis of the intervening facts: 4.3j establishes that 4. w) ∧ Sum(w. a. THE PURE THEORY OF CLASS SIZES Definition 4.3.3. z ) x ˜ ˜ (f) ZA (x = 1) iff A Unit(x) Fact 4.1.3.6 verifies some connections between A and ZA . y .3. 4.8 establishes that ZA Zgm.2. 4. of Zgm directly: Definition 4. 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.54 4.4 deals with the corresponding model ZA . The following are theorems of PSIZE.3 deals with the model A of PSIZE. we can construct a model. y. A.3. z) or A ∃a∃wUnit(a) ∧ Sum(˜. Fact 4.3. ) . of PCA.y.3.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. Let τ be the translation from LPSIZE to LZgm where: τ∀ = x = x τSum(x. If A which: PSIZE.

z1 ) ∧ Sum(˜. y. ZA is the expansion of ZA induced by the definitions in 4. y. y. as follows: Definition 4. w1 ) & Sum(w1 . z) → Sum(x. w) → (ASSOC2) ∃w2 (Sum(y. z . y. a. z) → Sum(˜. y . w) ∧ Sum(w. an expansion of both A and ZA . z.2e by IDENT by UNIQ− . we require a battery of tedious facts about the model ZA . y ) ˜ ˜ Sum(x. then ZA ⇒ If then But So ZA 0 + y = 0 A Sum(∅.4. y.3.3. y) ∨ Sm(˜. Given a model A of PSIZE. ∅) A Sum(∅. there is a unique y such that: ZA x + y = 0 Proof. y.4. y2 . y ) x ˜ Sum(x. y. y. ∅) ZA y = 0 x + y = 0 iff y = 0. PCA IS MODEL COMPLETE 55 (a) (b) (c) (d) (e) (f) (g) (h1) (h2) (i) (j) Sum(x. which interprets two additional operators. x) z ˜ Sum(x. y1 .3. z) → y1 = y2 (UNIQ+ ) (UNIQ− ) 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. z2 )) x ˜ ˜ Proof.3. Given a model A of PSIZE: for each x in Z˙A . ∅.3. z) ∧ Sum(x. z2 ) → (z1 = z2 = I) x ˜ ˜ ∃z1 Sum(x. z. w)) ˜ x=x ˜ x<y↔y<x ˜ ˜ x<y∧y <y →x<x ˜ ˜ Sm(x. z1 ) ∧ Sum(x. We shall state these in terms of the model ZA .3. w2 ) ∧ Sum(x. a. z2 ) → z1 = z2 Sum(x. w2 . y . by 4. If x = 0. In addition to the theorems of PSIZE listed in 4. z1 ) ↔ ¬∃a∃w∃z2 (Unit(a) ∧ Sum(˜. The proofs are elementary. y.

c) ∧ 0 < a → Sum(−c. there is a unique y which satisfies (*) Sum(˜. since A Sum(x. ZA satisfies the following: a.56 ⇐ Obvious 4.3. Fact 4. ZA satisfies the following: (a) (b) (c) (d) (e) Proof. ˜ ∅) A A A A ˜=I ∅ w=a ˜ Sum(˜. y . y . ˜ a ∅) x+y =0 ZA x + y = 0 A Sum(˜. b − c. a. a. y) x x + y = 0 iff A (*). where Unit (a) x UNIQ+ . THE PURE THEORY OF CLASS SIZES b. y .3d and h2 by 4. a) x ˜ ˜ Sum(˜. y).3.5. 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(˜. w) x ˜ A Sum(w. −c. c) ∧ 0 < b → Sum(a. a + b) b. −b) Sum(a. Sm(a.2e Fact 4.3. y. c ≤ b ↔ Sum(c. b.3h2 by COMP by 4. If x > 0.3.6. a. a. b) c. 0) by COMP and UNIQ− by 4. b) ↔ Sum(a.3. Omitted. −a) −(x + y) = −x + −y −(x − y) = y − x (x + y) − y = x −(−x) = x x = 0 ∧ x < y → −y < −x . b. a) x ˜ ˜ Sum(˜. Sum(a. b. b.

5c by (a) by IDENT by MONOT and UNIQ+ .3. c) 0<b c = (a + b) c − b = (a + b) − b c−b=a −b=a−c Sum(a. since 0 < b Sum(c.4. −c. w) for some w (a + b) = w Sum(a. b) for some w b=c+w b − c = (c + w) − c b−c=w Sum(c. b.5c by IDENT by MONOT and UNIQ+ by MIN by (a) and UNIQ+ by 4.3. c) 0 ≤ (b − c) ↔ b ≤ c b≤c c≤b Sum(c. −c) a=0 Sm(a. ¬Sm(a. b) Sum(c. (⇒) Suppose So So So Sm(a. b − c. (⇒) Suppose then So So So So (⇐) Suppose But So So c.5c by 4. w. b) by EXIST− by (a) and UNIQ+ by 4. −c) a=0 a < c. PCA IS MODEL COMPLETE d. a. b − c. b) Sum(a. b) → (Sm(−a. a + b) 57 by 4.3. b. −b) ∨ b = −a) Proof.3. b. −b) if Sm(a. 0.3. b.2e (⇐)Obvious. Suppose and then So So So So But if then And if then Sum(a.

−a + −b). c. Suppose But So So So But if then So So But So So And if then − c < −a −c≤a ˜ Sm(a.3.3. a) ˜ Sm(a. Lemma 4. a.3. ¬Sum(a.6a by ASSOC2 by 4. (−a + −b) − (−a) = −b. (a + b) < a.6a and UNIQ+ by 4.6a and UNIQ+ . (a + b) − a. Sum(−a. −b.3. (−a + −b) − (−a).5e 4.3.3. w) ∧ Sum(b − c.3.3. a + b) ∃w(Sum(c.58 So So But So d. w. b) b = (a + b) − a. When Sm(a.5a (b) 4. We shall work through successively more general cases: a. We need the following lemma. b) Sum(b.5c (a) (b) TRICH 4.3. ¬(a ≤ a + b).4 IDENT EXIST+ ¬Sm(a.3. a.6b by 4.7. (a + b) = 0 − a < −(a + b). Sum(−a. −a + −b) Sm(−a. ZA (a + c) + (b − c) = a + b Proof.5c We can now show that ZA Zgm. −b) (a + b) = 0 b = −a by by by by by by by by 4. − a < −a + −b.5e 4. a + b) w =c+a a + b = (b − c) + (c + a) a + b = (a + c) + (b − c) by 4.3. b) and c ≤ b We know and So So and So Sum(b − c. a + b). −c) 4. THE PURE THEORY OF CLASS SIZES by by by by 4. The only real difficulty arises in verifying that addition is associative.

c) 0<c−b Sum(b.3. Sm(a.4b 4. Whenever Suppose ¬Sm(a. −c.5b by 4. −a + −c) Sum(−a. b). b − c) Sm(−a. b − c.3.5c 4.5d (c2) by (c2) by (c1) since Sm(a. a + b) (a + c) + (b − c) = a + b c ≤ b. −(a + c)) 0 < −a Sum(−(−(a + c)).6c 4. a) Sum(−c. −c. b) by ASSOC by 4. −c. c) and thus b < c Sum(b. −c) Sum(−a. c) Case (b) applies ¬Sm(a.4. for otherwise case c applies. −c.3.5b by by by by by (c1) EXIST+ 4. b − c) Sum(−c. c − b. Sm(a.5a by 4.3.3. −c.3. −c) or c = −a since ¬Sm(a.3.6b MONOT 4.3. −c. b.3. b.3. a + b) Sum(a + c. a) c = −a a+c=0 a = −c Sum(a + c. 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. When Sm(a. PCA IS MODEL COMPLETE 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. −c. −(−a)) Sum(a + c. −(c − b)) Sum(b.3. c) Sm(−a. c) If So assume then So So So c. b − c) Sum(a.6a . b.6a by 4.6d by 4. b) and Sm(a.3.6c by 4.

The only axiom for abelian groups that needs further verification is associativity.6d So Assume then So So So by by by by case c 4.3. x + z) ZA y < z A Sum(x. We prove this using lemma 4. z.7: x + (y + z) = (x + y) + ((y + z) − y). by 4.5a 4.7 = (x + y) + z.3. x + y) A x+y ≤x+z ZA x + y ≤ x + z by EXIST+ by MONOT .5d Theorem 4. 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.5c The ordering axioms of Zgm are satisfied in ZA because ZA uses the same ordering as A. −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 4.3. Similarly.3. THE PURE THEORY OF CLASS SIZES Sm(−a.60 Either Assume then and 4. 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.3.3.3.5a 4. −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.8. A PSIZE. ZA satisfies axioms (12). ZA Zgm Proof.3. and PSIZE includes the same ordering axioms. and (14) of Zgm because A satisfies MIN and MAX. ZA satisfies axiom (10) of Zgm because A satisfies the UNIT axiom of PSIZE. by 4. y.

the finite completions of PCA are consistent with PCS. b. EXACTLYn .4. The reduction is uniform since each model A of PCA has the same domain as its Zgm-model. The finite completions of PCA are the theories PCA. Theorem 4. If T is a completion of PCAI. EXACTLYn is true in SA . Apply Theorem 4. (1) (2) PCA PCS Every completion of PCA is consistent with PCS.4.3. We already know that PCS PCA. Every infinite completion of PCA is consistent with PCS. Proof.1(b).) Letting PCAI = PCA + INF and PCSI = PCS + INF. we could follow the method outlined at the end of chapter 3 for showing that CA ≡ CS. (Formally. a. which follows from (2) by 3.9. (2a) (2b) Every finite completion of PCA is consistent with PCS.3. we may conclude that PCA is model complete. so we need only prove (1).9. But PCA. So. where A is the standard finite interpretation of LC< containing n atoms. But (2) is equivalent to the conjunction of (2a) and (2b). a. REMARKS ON SHOWING THAT PCA AXIOMATIZES PCS 61 So. (2b) is a consequence of (3a) and (3b). So.4 Remarks on showing that PCA axiomatizes PCS To prove that PCA ≡ PCS. (3a) (3b) If PCAIf is consistent. we would have to define EXACTLYn in terms of units rather than atoms. we PCA is τ -reducible to Zgm. b. 4. PCA is model complete.2. for some f . then T ≡ PCAIf . .1b. PCA satisfies Monk’s condition. then PCSIf is consistent. Immediate from (a) and 4.2.

so only (5) remains to be shown. (4) If f is total.3.3. we would have to prove analogues of 3.9 through 3. This seems straightforward. because every completion of PCAI entails PCAIf for some total f . But PCA is model complete. but tedious. the construction could be duplicated in this simpler case.13 for PCA and PCS. It’s apparent that the size models of the prime models for CAIf would do nicely. THE PURE THEORY OF CLASS SIZES To prove (3a). then PCAIf is complete. To prove (3b). . Alternatively.62 4. (5) If f is total and congruous. We will not construct prime models for the extensions of PCAI. then PCAIf is has a prime model. The trick is to show that enough axioms about size have been incorporated in PCA to establish the entailments among MOD statements. it is sufficient to demonstrate (4).

There is an isomorphic embedding. B CA. SA SC SB (A) ⊆ SB . let SB (X) = the submodel of SB whose domain is { σB [x] | x ∈ X } Then. Strictly speaking.2. Monk’s Theorem applies directly to SB (A). a finitely generated submodel of SB . so we cannot apply Monk’s Theorem directly.1). There is an isomorphic embedding. so we may conclude 5.1.1. f : C → A.1.1 Model completeness of CA We shall show that CA is model complete by showing that it satisfies Monk’s criterion (see 4.1. 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. and SB (C). SB .3: Fact 5. CA.3.1. DISJ∪ and REP< then allow us to find elements with those sizes that fit together in the right way. But. So.1.3. Assumption 5. SA is not a submodel of SB . A ⊆ B . clearly. The existence of Monk mappings for PCA tells us that we can find elements with the right sizes.5 Completeness of CA 5. and C is a finitely generated Theorem 5. where ˙ ˙ g(σB [x]) = σA [x] for all x ∈ C ∩ A.1.2. 63 . given Assumption 5. g : SC → SA . A substructure of B. we want to prove 5. and SB (C).

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 )

|x| d. for total and congruous f . so that x can be represented as (C(x) ∪ ∆1 (x)) − ∆2 (x) as in 3. Proof.27) To verify that the structure Qf is a model of CAIf . 1 Sum(x.2.n is the submodel of Qf whose domain is Qn .29 by substituting: Qf.2. The desired models of CAI may now be defined: Definition 5. and δ(x) . PRIME MODELS FOR CA c.24. 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. |∆1 (x)| − |∆2 (x)| 67 Intuitively. ρ(x).2. The proof can be obtained from the proof of Theorem 3. then the standard protomodel for f is Qf . |∆1 (x)| − 0.5.2. Instead. |∆2 (x)| = θf (C(x)) + 0. then θf (x) = 0. we exhibit each such model as the union of a chain of models.2.4.3. then θf (x) = θf (C(x)) + θf (∆1 (x)) − θf (∆2 (x)) = θf (C(x)) + 0.5.2. y. θ(x). Definition 5. If f is total and congruous. δ . all n-congruence classes are assigned sizes 1/n. then Qf. If x ∈ Q. Fact 5. as in Q.n θf (x) ρf (x) δf (x) for for for for Qn . the first f (n) n-congruence classes are each one atom larger that the remaining (n − f (n)) n-congruence classes. z) iff θf (z) = θf (x) + θf (y) Qf (cf 3. If f is total and congruous and n > 0. then Qf. but δ is no longer 0 in all cases.n BASIC.2. If f is total and congruous. If x is finite.

since f is congruous. If 0 < i ≤ m. 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. 4}. then Qf. If f (6) = 0.6 will then fall into place. If 0 < i < j ≤ n. [2n] will include the same number of charmed 3-congruence classes as [2n + 1]. then all of the 3-congruence classes are charmed except for [6n + 4] and [6n + 5]. In other words. x = x1 ∪ . so Qf.7.m! . 2. then xi ∩ xj = ∅. 3. (ii) If m < i < j ≤ n. Suppose A BASIC. for any atom a. . If f (6) = 4. The sequence Qf. x. and easier. .n! ⊆ Qf. f (6) ∈ {1.2 .3 is not an extension of Qf. we want to show that the 2-congruence classes have the same size relations in Qf. Proof. 5}. If n > m. The xi are approximately the same size: (i) If 0 < i < j ≤ m.2. xn . ˙ Definition 5. ∆1 (x). and 0 ≤ m < n.2 [2n] ∼ [2n + 1] Since f is congruous. .6 [2n] ∼ [2n + 1] Suppose. then xi ∼ xj . xj ). It will be clearer. then xi ∼ xj . where a. a. for i > m. ∪ xn b. Suppose that f (2) = 0. xi is called a charmed n-factor of x. [2n] contains exactly one more charmed congruence class than [2n + 1]. then all of the 6-congruence classes are common. x ∈ A. (iii) If 0 < i ≤ m < j ≤ n.m)-partition of x in A is a sequence x1 . as C(x). Letting n = 2 and m = 3. however that f (2) = 1.6 as they do in Qf.6. then Sum(xi . . But this sequence does harbor a chain of models: Fact 5. to establish this by example rather than by formal proof. . In any case. In any case. so Qf. and ∆2 (x). and c. since size relations are determined by the representations of each set. then [6n] and [6n + 1] are the only charmed 3congruence classes. If f (6) = 2. xi is a common n-factor of x.6 [2n] is one atom larger than [2n + 1] So it goes in general. COMPLETENESS OF CA The following notion is helpful in understanding our constructions. . so that Qf. so that Qf. f (6) ∈ {0.2 . Then an (n. The other elements of Qf.2 [2n] is one atom larger than [2n + 1] Here.2. for any f . For example Qf. .n does not constitute a chain of models.68 5.

x12 ∪ x22 . the basis of A.1 . if the basis of A is in B and B CAI.f (n) .2 Shells of models To embed the model Qf into an arbitrary model. x1 and x2 .8. Immediate from (a). d.2.5. . and x13 ∪ x23 will be roughly the same size and will exhaust x0 .7 allows us to regard Qf as the union of a chain of models: Fact 5. it is preserved under extensions. PRIME MODELS FOR CA Fact 5. Since MODn. then Qf = n>0 69 Qf. for k > 0 MODn. Again. Pick such a pair. c. and x23 whose union is x2 . If f is total and congruous. x0 . Qf Proof. Fact B. for n > 0 CAIf 5. call it x0 . b. of CAIf .0 or A MOD2. A. to include in B. of A which satisfies CAI. Clearly. Qf c. (b). ˙ B must also satisfy ADIV3 . then a. so it is preserved under unions of chains.2. But.2. If f is total and congruous.2. must be included in B.f (n) is an existential sentence. x22 .n! Fact 5. and (c). BASIC is a universal-existential theory. Whether these are exactly the same size or differ by an atom depends on whether A MOD2. Insuring that x1 and x2 are each divisible by 3. we must find a smallest submodel. B. and x13 whose union is x1 and another three disjoint sets. Each n-congruence class can be partitioned into k nk-congruence classes. We can aim for this by placing in B three disjoint sets. by definition. also guarantees that x0 is divisible by 3: the three unions x11 ∪ x21 .9. BASIC ADIVk . Qf d. ˙ must contain two disjoint sets of roughly the same size whose union is then B ˙ the basis of A. x12 .f (n) . (See Appendix.n MODn. x21 . since the symbol I must refer to the same set in the ˙ submodel as it does in the model. Qf. a.2. the exact size relations will be determined by which MOD principles are satisfied in A. x11 . Qf b.5). The existence of such sets is assured because A ADIV3 .

λ(i) = 0.) b. but this would still not be an atomic boolean algebra. L(λ) is k. We cannot correct for this problem ˙ by including in B all atoms of A: A may have uncountably many atoms while B. λ extends κ iff L(λ) > L(κ) and. This great tree of sets will not form a boolean algebra. This will produce an infinite tree. The deeper a node is in this tree. Second. given a model A of CAIf . We leave the solution of this problem to the formal construction. for 1 ≤ i ≤ L(κ). to each node. ni < i. In the next section. we present the construction which. we define the shell of A as the submodel of A generated by the collection of node sets and node atoms. Nodes constitute the vertices of an infinite tree in which 0 is the root and λ dominates κ iff κ extends λ. . COMPLETENESS OF CA We can continue this process indefinitely. or depth. The number of immediate descendants of a node grows as the depth of the node increases. for L(κ) < i ≤ L(λ). (iii) λ • m = n1 . i. be the basis of A. d. to be a prime model. The formal proof proceeds as follows: First. . bearing sets. of λ. a. we define a tree. λ 0-extends κ iff λ extends κ and. . congruous be any atom of A.10. and let a 0 .70 5. which is what we are looking for. aλ . dividing each set introduced at stage n into n + 1 roughly equal subsets at stage n + 1. (The variables λ and κ range over nodes. where k > 0 and for all i ≤ k. λ. Let A 0 CAIf . If λ = n1 . At the same time we shall assign an atom to each node. First. λ(i) = κ(i). Definition 5. . A boolean algebra could be obtained by including both the node sets and their finite unions. m c. .11. . nk . (ii) If 1 ≤ i ≤ k. we show that the shell of A is isomorphic to Qf .e. the tree: Definition 5.2. . nk . then (i) The length. Third. We shall now assign a set to each node by repeatedly partitioning the basis of A. . the smaller the set it bears and the greater the number of successors among which this set will be partitioned. assigns a node set Aλ and a node atom. nor will it be closed under finite unions. . where f is a total. A node is a finite sequence n1 . . . . then λ(i) = ni . 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. nk .2. must be countable. Given A = Af remainder function: a.

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

: If n = 1. For any n. and Aλ 0 is the basis of A. If Aλ = Aκ . by (a) again. Aλ ⊂ Aκ iff κ extends λ or κ = λ. there are (n!)(n + 1) = (n + 1)! node sets of depth n + 1.72 5. There are n! nodes (and.13 should clarify this constuction. b. c. So f (nm) = nk + f (n) k = (f (nm) − f (n))/n and c. So (k ) holds because any set is a (1. g. hence. . f. since every charmed n-factor is one atom larger than each common n-factor. Then (k ) also holds for n + 1: each node set of depth n has n + 1 immediate descendants. e. then n! = 1. Every node set contains infinitely many node atoms. COMPLETENESS OF CA charmed m-factors among the n-factors of A. aλ = aκ iff one of λ and κ 0-extends the other. the node sets of depth n form an (n!. then Aλ•i ∩ Aλ•j = ∅. Fact 5. i. j. The facts listed in 5. there are f (nm) charmed nm-factors of A. a. k.13. aλ ⊂ Aκ iff λ extends κ. Any two node sets of the same depth are disjoint. Each node set is the disjoint union of all of its descendants at any given depth. Assume that (k ) holds for n. so. h. Each node set is the disjoint union of its immediate descendants. node sets) of depth n.2. f (n!))-partition of the basis of A.0)-partition of itself. d. then κ = λ. f (n!) = 0. But.2. If i = j. Proof. Immediate from (b). We demonstrate (k) by induction on n. All of them can be established by induction on the depth of nodes.

2.5. By Fact 5.2. the n + 1 factors of the n-factors of the basis form an ((n + 1)!. f (n!) of the n-factors are charmed and n! − f (n!) are common. A. PRIME MODELS FOR CA 73 Furthermore. {Aλ . aλ . 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. 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. where k= f (n!(n + 1)) − f (n!) n! f ((n + 1)!) − f (n!) n! = (substituting n! for n and n + 1 for m). f . A. ˆ d.12. we will regard as fixed: a. Definition 5. a total.12. and node-atoms. f ((n + 1)!))partition of the basis. a model of CAIf . from a model.1. of CAIf . congruous remainder function b. c. k)-partition. So. each charmed n-factor has an (n + 1.14. we can now construct a submodel. k + 1)-partition and each common n-factor has an (n + 1. That is to say. of A which is isomorphic to Qf . B. the shell of A is the submodel of A generated by {Aλ . a set of node sets and node-atoms produced by the construction above. aλ }. A. Aλ . (k ) holds for n + 1. Then. Given a collection of node sets.2. aλ }. Suppose f is total and congruous. For the remainder of this chapter. that A CAIf and Aλ and aλ are the node sets and node-atoms of A produced by construction 5. aλ } . then . the shell of A generated from {Aλ .

b. Recall from 3. If x ∈ A and y ∈ A. x ∈ A iff x is A-near some quasi-nodal set of A. x. ˆ b.16. we can characterize A as the collection of sets A-near quasi-nodal sets.2. If x ∈ A.2. 3. c.14-3. ∆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.2. 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). ˙ ˆ a. A is an atomic boolean algebra whose atoms are the node-atoms aλ .2. each of which preserves A-nearness to quasi-nodal sets. (⇒)A is generated from node sets and node atoms via the boolean operations. finite sets of node-atoms correspond to the finite subsets in Q.74 5.2. Fact 5. ˙ ˆ c.15 through 3. so it must also contain sets obtained by adding or removing A-finite sets from quasi-nodal sets. Definition 5. 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).18 obtain for A-nearness.24 that each member. ˆ a.2. both of which are A-finite sets. Proof. a. finite unions of node sets correspond to the quasi-congruence classes. we need a sharper characterization of the elements ˆ of A. Analogues of 3. x is an A-finite set iff it is a finite set of node atoms. (cf.18) ˙ ˆ Still following in the footsteps of chapter 3. of Q has a unique representation as (C(x) ∪ ∆1 (x)) − ∆2 (x) where C(x) is a quasi-congruence class.2. COMPLETENESS OF CA ˆ To show that A Qf . then x is A-near y iff both x − y and y − x are A-finite sets.15.

5. then k ι(λ) = i=1 (i − 1)!ni . m)-partitions to use at each level. In effect. Let C(x) be the quasi-nodal set which is A-near x (see 3.18. PRIME MODELS FOR CA b.0. we are stipulating which (n.2 = [n] = [2n] = [2n + 1] = [6n + 1] = [6n + 4] = [6n + 5] Definition 5.2.2 Q 0. . Let ∆1 (x) = x − C(x). Q0 Q 0. second.21 exactly.19.2. Then a. nk .0 Q 0. qλ = {ι(λ)} Fact 5. we are selecting node atoms so that every singleton in Q is the node atom for some node. . let ∆2 (x) = C(x) − x.0 Q 0. and.17.2. .23). The proof parallels that of Theorem 3. but with two differences: first.2.1.3 The embeddings To embed Qf into A. a. Suppose λ is a node. we are performing the construction 5. This latter condition guarantees that the shell of Qf will be Qf itself. we first describe Q in terms of node sets and node atoms. b. If L(λ) = k. ι(λ) = the least n ∈ Qλ . Examples.2. then Qλ = [k!n + m] where k m= i=1 λ(i)(i − 1)! b.2. 75 c.1.2.5. a.12 on Qf .2. The depth of Qλ is L(λ). Definition 5. . If λ = n1 .1 Q 0.

λι(λ) = λ iff λ =< 0 > or λ(L(λ)) = 0 (ie a node. a. λι(λn ) = λn d.2. there are infinitely many λ such that n = ι(λ). We may. e.2. λ. There is at most one such y. then all nodes along the left-most branch descending from λ also have the value k.76 5. These are the only nodes below λ with the value k. ι(λn ) = n b. by Fact 5. there’s a λ such that n = ι(κ) iff κ = λ or κ 0-extends λ. a. That is to say. shallowest) node for which ι(λ) = n. there’s a λ such that n = ι(λ).22.2.e. there is a unique y ∈ A such that ∀λ(qλ ⊆ x ↔ aλ ⊆ y) Proof. λn = the shortest λ such that ι(λ) = n. c. there will be infinitely many nodes λ for which ι(λ) = n.16. For every n. Then there is some n such that x = x1 ∪ · · · ∪ xk .13 hold for the sets Qλ and qλ . If ι(λ) = k. Though for a given natural number n. For every n. Fact 5. To show that there is such a y. the ι(λ) take on all and only values less than n!. d.21. If x ∈ Q.23.2.2.2. For every n.2. COMPLETENESS OF CA b. Fact 5.13k holds. Definition 5. finally. At each depth. n. we can associate with each natural number a shortest (i.20. the Qλ can be regarded as node sets and the qλ as node atoms for any model Qf . λ extends λι(λ) c. Notice. Every natural number is the value of all and only those nodes along the left-most branch descending from some node. that 5. will be the highest node with a certain value just in case λ is not the leftmost immediate descendant of its parent. c. especially. suppose first that x ∈ QC. define the embedding of Qf into A: ˙ ˆ Fact 5. b. ι(λ) = ι(κ) iff λ = κ or one 0-extends the other.) Each of the points listed in Fact 5.

b. Proof. Suppose g(x) = y = g(x ). and let y = (y ∪ { aλ | qλ ⊆ ∆1 (x) }) − { aλ | qλ ⊆ ∆2 (x) } ˙ Definition 5. as described above. then g(x) = (g(C(x)) ∪ g(∆1 (x))) − g(∆2 (x)) b. 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 5.25. Immediate from the proof of 5. g maps Q onto A.2.24. a. ˙ ˆ c.13h 77 ˙ ˆ where x is a quasi-congruence class. . 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 . If x is not a quasi-congruence class. so x = x . If x ∈ Q. then x = (x ∪ ∆1 (x)) − ∆2 (x) by 5. So x = Qλ1 ∪ · · · ∪ Qλk Now.2.2.2.. g is one-one.23.2. PRIME MODELS FOR CA where each xi is an (n!)-congruence class and hence a node set in Qf . Let y be the element of A corresponding to x . Then ∀λ(qλ ⊆ x ↔ aλ ⊆ y ↔ qλ ⊆ x ) But every integer is ι(λ) for some λ.5.

a. of Q into A is an isomorphism ˆ of Qf onto A.24) (by 5.2. relative complements. unions. As in (a). g can be shown to preserve ∅.2. n) − (|∆1 (x)| − |∆2 (x)|) = nc(y. INDISC∼ ˆ since A ⊆ A . I. n) − (|g(∆1 (x))| − |g(∆2 (x))|) = nc(g(y). 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.2. g. Qf x ∼ y iff θf (x) = θf (y) iff nc(x. The nodal embedding. COMPLETENESS OF CA Theorem 5. n) be the number of charmed node sets of level n contained in C(z). 5. below. 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. be the least k such that x and y are both unions of node sets of depth k. and proper subsets.78 c. Proof. c. 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. And let n. Obvious.26. n) − (|∆1 (y)| − |∆2 (y)|) iff nc(g(x). Let nc(z. intersections.16) b.

i for exactly one i. then T is inconsistent.n−1 ).j (see Lemma 3. by 5. But if T (¬MODn. SUMMARY e. for some congruous f .12c. Theorem 5. If f is a total.3. .3. Again. CS.5.27.27. and their completions. 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).0 ∧ . y.i for each such i. congruous remainder function. g(z)) Sum(g(x). the prime model test (Appendix. CAIf is consistent because f is congruous.6. and has a prime model. Qf Sum(x. If f is total and congruous. then CAIf is consistent and complete. . T would be inconsistent.3. Fact C. 0 ≤ i < n.11).i for at least one one i. since T CAI and CAI DIVn . Hence T MODn. 0 ≤ i < n: Since T is complete.1. g(y).2.2.j where 0 ≤ i = j < n.3 Summary We can now draw our final conclusions about CA.3. 5. and we may conclude: Corollary 5. For each n > 0. g(z)) since Qf 79 DEF+ by (b) and (c) since A DEF+ ˆ since A ⊆ A So. Corollary 5.3) applies. each model of CAIf has a submodel isomorphic to Qf . So CAIf is complete. Since CAIf is model complete.1. Proof.i or T ¬MODn. Proof. by 5. for CA MODn.2.i → ¬MODn. ∧ ¬MODn. then CAIf has a prime model. But suppose T MODn.3. g(y). by 3. T MODn. . It T is a completion of CAI.i and T MODn. then T ≡ CAIf . T MODn.

6.3. EXACTLYn is decidable. There is no sentence. There are 2ω remainder functions whose domain is the set of prime numbers. so it is decidable. alternate between generating theorems of CA and testing whether An ¬φ. Corollary 5. a. then CAIf is recursively enumerable.4. a. b. But An is a finite model.m . (⇒)If f is decidable. Follows from 5. For total f . CAIf T . To determine whether CS φ. By Lindenbaum’s lemma. CS. each of these extensions has a consistent and complete extension. given that CS CA. since CAIf is a. b. Theorem 5. Each such function is congruous.80 So. Corollary 5. φ is consistent and T has only infinite models. CS is decidable. Theorem 5. Then T CAIf and. CSIf is decidable iff f is decidable.3. φ. Every completion of CA is consistent with CS.m sentence is in CSIf . b. COMPLETENESS OF CA MODn. But CAIf is complete.3. . b. let f (n) = m iff T complete. 5. a. a.2b. CSI has 2ω completions. (⇐)To calculate f (n).3. determine which MODn. Follows from (a) and 3.3. CS. so each corresponds to a consistent extension of CSI.3.13. such that T = CA. b. b. For n > 0. EXACTLYn φ iff An φ.3. Proof.5. Proof. CA ≡ CS Proof.2 and 3. a.3.

But by (3). Let K = { n | every prime factor of n is a member of J }. so ¬φ ∈ CS. We claim the following without proof: (1) (2) (3) By (1) and (2). If φ is true only in infinite models of CA. φ is inconsistent. then ¬φ is true in all finite models of CA. so CA. So CA φ and. A CS. Let A be a model with domain k∈K Qk . T = { ADIVn | n ∈ J }. A A A BASIC ADIVj for all j ∈ J ADIVk φ.3. A φ CS.2.3. so A . θ.7.26. But CA ≡ CS. (BASIC ∪ T ) φ for some finite set of ADIVn principles. Theorem 5. in which size relations are determined in accordance with the size function. by compactness. Proof. CS is not finitely axiomatizable. defined in 3. Hence (BASIC ∪ T ). Suppose CS φ. SUMMARY 81 Proof.5.

COMPLETENESS OF CA .82 5.

So CS .6 Sets of Natural Numbers CS has standard finite models since it consists of sentences true in all such models. To rule out such anomalies. If x outpaces y. some such orderings say that there are fewer even numbers than prime numbers (see below 6. in section 1.2. 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. 6. Section 2 establishes that OUTPACING can be satisfied jointly with any consistent extension of CS in a model whose domain is P(N). In this chapter we will show that CS has infinite standard models over P(N). OUTPACING does not fix the size relations over P(N). It has infinite models Qn and Qf . Definition 6. We extrapolate from well understood finite cases to puzzling infinite cases.13). But we should also 83 . we introduce a principle. x and y will range over P(N). OUTPACING mentions the natural ordering of N and applies only to subsets of N.1.1 The outpacing principle Throughout this chapter. We employ this notion to state a sufficient condition for one set of natural numbers to be larger than another: OUTPACING.1. 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. that is. OUTPACING. The general motivation behind this principle should be familiar. then x > y. An ordering of P(N) that satisfies CS will not necessarily appear reasonable. For example.

So. again. But the two are discernible under outpacing. . and see how far they go. a. Definition 6. For example. x restricted to z is larger than y restricted to z. neither [2n] nor [2n + 1] outpaces the other since each initial segment {0. The best we can do is propose plausible theories. that this extrapolation cannot be done in any straightforward. since [2n] outpaces [2n + 2] while [2n + 1] does not. strengthen the conditional to a biconditional. It is doubtful whether there is any mechanical way to decide which of these statements are true. .1. thus: (1) x > y iff x outpaces y. and A OUTPACING There is a slight difficulty in saying that an interpretation of LC< satisfies OUTPACING . then (k − 1)/2 > k/3 + 1 . Others are too weak to be helpful. [2n] > [3n] b. . SETS OF NATURAL NUMBERS emphasize.84 6. If k > 0. 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. [3n] has at most k/3 + 1 such members. [2n] has at least (k − 1)/2 members less than or equal to k for any given k. its antecedent is only satisfied when y ⊂ x. Though (2) is true. [3n] > [4n] > [5n] > . given any finite subset z of N. 2n + 1} of N contains n evens and n odds. c. A is an outpacing model iff ˙ A = P(N) A BASIC. Every outpacing model satisfies the following a. This revised principle conflicts with CS. then [kn] > n2 Proof. Since OUTPACING involves the < relation over N. then x > y. If k > 4.0.1. determine whether they are consistent. Some of these conflict with one another. We shall finesse this problem by regarding OUTPACING as the (very large) set of sentences { b < a | a outpaces b } Fact 6. there are many statements that assert of infinite cases what is true of finite cases. mechanical way without risking contradiction. .1. for example. it cannot be expressed in LC< . We cannot. for outpacing is not a quasi-linear ordering. . .

1. If m = k 2 . a. Fact 6. b. x. it is sufficient to show that ¬ [2n] < [2n + 1]. y is an alternating pair iff x and y are infinite and for all i > 0.1. then.1. y is an alternating pair. If x. So there is no y such that [2n] ∼ y ⊂ [2n + 1]. by REP< . Proof. including other congruence classes. But any proper subset of [2n + 1] is outpaced by. Let k be the least odd number not in y. If [2n] > [2n + 1]. N[m] will have more members in [kn] than in n2 . For a given k > 0. (y ⊂ x ∧ z ≤ y ∧ Atom(z ) ∧ z ⊂ x ∧ x = z ∪ z ) → y ∼ z Let x = [2n]. If it were. we show that both of these possibilities can be realized in standard models over P(N). If 0 ≤ i < j < k. Theorem 6. there is a y such that y ∼ [2n] and y ⊂ [2n + 1]. The argument for 6. y = [2n + 1].1. 85 c. The two alternatives left open in 6.2 applies here since the only facts about [2n] and [2n + 1] used hold by virtue of these sets forming an alternating pair. Aj is an alternating pair. Fact 6. Definition 6. if [2n + 1] ∼ [2n + 2].1.2. then Ai . then N is even. Note that [2n] = [2n + 2] ∪ {0} and that BASIC (*) (*). let Ai = [kn + i] for each i < k. Here.1. then N is odd. Every outpacing model satisfies the following a. we generalize 6. Similar to that of (1). and hence smaller than [2n]. both [kn] and n2 have exactly k members less than m. [2n + 1] ≥ [2n + 2] c. Similar to (1).2 to similar cases. xi < yi < xi+1 .3.5. c. . By TRICH. For m > k(k + 1). [2n] ≥ [2n + 1] b.2.2 correspond to the possibilities that N may be odd or even: if [2n] ∼ [2n + 1]. z = [2n + 2] and apply (*). THE OUTPACING PRINCIPLE b. Then: a.1. Then [2n] leads y at k + 1 and y never catches up.4. then [2n + 1] ∼ [2n + 2]. In section 6.1. then in any outpacing model: x ∼ y ∨ x > y ∼ (x − x1 ) Proof.6.

Definition 6.2 Models of CS and Outpacing In this section. Proof. as individual constants. LN . So either A0 > Ai or A0 ∼ Ai by 6. then Ai ∼ Aj . then Ai ∼ Ap .4 since A0 . (i) For 0 ≤ i < p. the language of subsets of N. a. a. 2. 0 < p ≤ k such that (i) If i < j < p. Ap is an alternating pair and A0 > Ap . 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] 6.1. But A0 ≤ Ai by the selection of p. Hence Ai ∼ A0 − {0} ∼ Ap . A0 .1. Ai is an alternating pair.86 b. so Ai = [4n + i] for i = 0. Ai (n + 1) = kn + i + k and i < j < k + i. 6. b. (iii) A0 > Ap ≥ Ai if i ≥ p. (iii) If p ≤ i < k. (ii) Immediate from 6. let p = k. then A0 ∼ Ap ∪ {0}. So Ai ∼ Ap . Let k = 4. so A0 ∼ Ai .Aj (n) = kn + j.1. otherwise.4. SETS OF NATURAL NUMBERS (ii) If p = k. 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. If A0 > Ai for some i. Example. a name for each subset of N. (i) follows by TRANS∼ . b.2. let p be the least such i. 3. we construct models of CS over P(N) that satisfy OUTPACING. Ai (n) = kn + i. There is a p. . (See example below). So A0 > Ai . is the first order language which results from adding to LC< . though the application here demands some important differences. 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. 1.

3.2.2. and similarly for other predicates. F is a filter over X iff (i) F = ∅ (ii) If a ∈ F and a ⊆ b. F . then there is an ultrafilter over X which includes F . p318] If X is a set and F ⊆ P(X). If F is a non-principal ultrafilter over N. If Y ⊆ X and Y is infinite. [Monk.6. Def. Ch. d. [Monk. a. then a ∩ b ∈ F c. F has the finite intersection property iff the intersection of any finite subset of F is non-empty. then there is a non-principal ultrafilter over X which contains Y . over X is principal iff there is some x ∈ F such that F = { a ⊆ X | x ∈ a } Fact 6. then a. p.2. and (iii) If a ∈ F and b ∈ F . 18. then either Y ∈ F or (X − Y ) ∈ F . AF a < b iff k | a[k] < b[k] ∈ F . [Bell. then b ∈ F . A non-principal ultrafilter contains no finite sets. lemma 1.15. b. A non-principal ultrafilter over X contains all cofinite subsets of X. c. If F is an ultrafilter over N. (cf.18.4. MODELS OF CS AND OUTPACING 87 Definition 6. Definition 6. c. 18. Theorem 6.108] b. F is an ultrafilter over X iff (i) F is a filter over X (ii) X ∈ F . p. If τ is a term of LN . An ultrafilter. b. If F ⊆ P(X) and F has the finite intersection property.6. then AF is the interpretation of LC< in which a.3.2.2. Prop. A˙F = P(N). Boolean symbols receive the usual interpretation.2.5. then AF is an outpacing model. then a. and (iii) if Y ⊆ X. then Ak (τ ) = AF (τ ) ∩ Ak = AF (τ ) [k] .319] d. Our main result is that if F is non-principal.

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

Construct a . for which AF k | a[k] < b[k] so. (AF ATOM because it contains all singletons of natural numbers. Furthermore.1. Let a0 = ∅ ai+1 = ai ∪ the n greatest members of b[ki+1 ] not in ai . let xi = { xkn+i−1 | k ∈ N }. for all k φ(a). MODELS OF CS AND OUTPACING (iv) If φ is non-atomic. for all a φ(a).2.1. Suppose AF as follows: Let K = Ak Ak Ak AF AF ∀xφ(x).) f. then. . then a [k] ∼ a[k] . Suppose then So So So d. xi . . together with ATOM and REP< .5. Claim: If k ∈ K. e. Hence: AF a ∼ a since they are the same size over some set which contains K and is. for all a ∀xφ(x) by (b) (a < b). partition x. for all k. thus. in the manner of the congruence classes modulo n (see Theorem 6. where n = |a[ki+1 ] | − |ai | a = ai Then a ⊂ b since each ai draws its new members from b. For 0 < i ≤ n. Immediate from (c) and (d): BASIC is equivalent to a set of universal sentences. 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. ki . . . . x an infinite subset of N. .} where the ki ’s are in strictly increasing order. for all a. for all k φ(a). K ∈ F . these form an alternating ntuple. these sets are approximately equal in size and AF ADIVn (x). The n sets. As in 6. .6. if F .5). a subset of b. (a ∼ a ) = {k1 . .

but only a submodel.90 6. 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. Otherwise. after all. [Bell. there will be many cases where AF a < b even though b does not outpace a. then a. [2n + 1] ∈ F . for if we count the even numbers and the odd numbers up to some even number. had we constructed the reduced ultraproduct. Fortunately. indeed.3b. 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 . There is. which results in a domain whose elements are functions from the index set (N. on some member of the filter) and the reduced ultraproduct is defined by interpreting the language over these equivalence classes. Immediate from 6. so it may or may not be in F . there will always be the same number of evens and odds. so AF [2n + 1] ∼ [2n]. . given that each of the Ak satisfies CS. the Ak ). So. pp. but is not quite the same.6 is modeled on the usual ultraproduct construction. and certainly any formula which is satisfied in all the factors. 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 model is built by first taking the product of all factors (in this case. Unfortunately. but equivalence classes of functions from N to finite subsets of N. we can immediately conclude that Ai /F satisfies CS. SETS OF NATURAL NUMBERS Corollary 6. has the handy property that it satisfies any formula which is satisfied by almost all factors.6.2. This will happen whenever k | a[k] < b[k] ∈ F but is not cofinite. these special means were available. for example.5e and 6. CAI. here) to elements of the factors. Then a[k] < b[k] iff k ∈ [2n]. and CSI The proof of 6.e. These functions are then gathered into equivalence classes (by virtue of agreeing almost everywhere. AF b. the only non-universal axioms of CAI could be verified in the constructed model more or less directly. and this mapping would have allowed us to identify a model over P(N) as a submodel of Ai /F .3. which we’ll call Ai /F . AF Proof. If F is a non-principal ultrafilter over N. This is handy because. If [2n] ∈ F . a. a natural mapping from subsets of N to such elements. Immediate from (a) and 5. Ai /F is not the model we wanted: its elements are not subsets of N.2.5f. b. AF [2n + 1] < [2n]. [2n] is neither finite nor cofinite. i. Consider a familiar example: Let a = [2n + 1] and b = [2n].2.2. In the usual construction(see. The model so constructed. For a given F . 87-92]).

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

x − x(1) . J is infinite. But f is congruous.92 6. SETS OF NATURAL NUMBERS We can now prove our main result. y is an alternating pair. By 6. If H is a finite subset of G. a.7 allows us to improve upon some previous results. a. The intersection of any finite subset. . so is y. so the restriction of f to the finite domain J has infinitely many solutions (see 3. let F be a non-principal ultrafilter which contains J. such that AF (x > y). If neither x nor y outpaces the other. So H = { n + 1 | k ∈ J → n ≡ f (k) mod k }. (See 3. Let J = k | |x[k] | = |y [k] | . Theorem 6.2.8a). so AF CSIf . y is an alternating pair.2.2. b. Theorem 6. Proof. there is a nonprincipal ultrafilter F such that G ⊆ F . Theorem 6. If x. Since the intersection of any finite subset of G is infinite. let Gk = [kn + f (k) + 1] for each k > 0. of G is infinite.3.3d. AF1 x > x+ and AF2 x ∼ x+ / Theorem 6. x. where J is some finite subset of N+ . For example.2. we can show that either of the alternatives in 6.f (k) . such that AF (x ∼ y).2. Since neither x nor y outpaces the other. Recall that every infinite completion of CS is equivalent to CSIf for some total and congruous remainder function.2.f (k) Let G = { Gk | k > 0 }.2.7: Proof.2. H.) Given such an f .10b can obtained for any alternating pair. Then (*) holds for each k: (*) Gk = n | An MODk. AF MODk. But then AF x > y. b. then there is an ultrafilter F . such that x ∈ F1 and x ∈ F2 .10c. Every infinite completion of CS has an outpacing model. by (a) there is an F such that AF y ∼ (x − x(1).6. we can suppose there is a set. By (*). then there is an ultrafilter F . Proof. f . By 6. then H = { Gk | k ∈ J }.11. Without loss of generality. So.12. If x. Then AF x ∼ y.

Theorem 6. First.4. then clearly A = AF . If (1) is false. Three comments are in order before we turn to the common structure of outpacing models. it is not clear whether every outpacing model can be obtained by the construction of 6. x+ but disagree elsewhere. This would be true only if fixing the congruence classes determined whether x ∼ x+ or x > x+ for every x ⊆ N . Proof. but it should not. Second.1.2. then A = B I have not been able to prove (1) or (2). Modifying 6. So (1) and (2) are open problems.6. we could produce standard models of CS over P(N) which satisfy OUTPACING and these will not. Finally.10c may suggest that any outpacing model. two outpacing models may agree about all pairs x. . Suppose that is a linear ordering under which N forms an ω-sequence. perhaps. all of the possibilities listed in 6. Also.5 for the relative sizes of the k-congruence classes are obtainable in outpacing models.4. This section has explored the existence and variety of outpacing models. If x -outpaces y. But even if (1) is true.2. At most one of them is obtainable by our construction.5]. satisfy OUTPACING. To establish that A = AF . Lemma 6. both (1) and (2) are necessary. so whether x ∈ F is an independent choice.2. in general.2. there is no unique outpacing model which satisfies T . is AF for FA = { x | A x > x+ }. by considering x = n2 : any finite set of congruences has infinitely many solutions that are squares and infinitely many that are not. each yielding a different outpacing model. 6. Ch. MODELS OF CS AND OUTPACING 93 On the basis of 6. Theorem 1. even it T is an infinite completion of CS. (2) If FA = FB . That all such choices are not determined by a remainder theory can be seen intuitively.2. A.2. we can now extend that observation to moduli other than 2.13. there are only 2ω remainder functions while there are ω 22 non-principal ultrafilters over N[Bell.11 we noted that there are even and odd outpacing models. There are standard models of CS over P(N) which do not satisfy OUTPACING. then FA is a non-principal ultrafilter. (1) If A is an outpacing model. More specifically. we will do so. though it may be extraneous to show that OUTPACING is independent. then x > y. 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 .

as the evens and the odds are in normal outpacing models. x diverges iff x diverges in N. we compare the ordering of P(N) given by density to the orderings given by CS and OUTPACING. ρ(x. . b. q1 . y). If x ⊆ y = ∅. ρ(x). . they do not content themselves with speaking of the (Cantorian) cardinalities of these sets. Fact 6. is the limit. so p > [2n] and OUTPACING is false in OUTPACING models. qk . 6.3 Size and density When number theorists talk about the sizes of sets of natural numbers.2. iff x converges in N. In this section. xi = |xi | . then ρ(x. of natural numbers is the limit. y). x. if it exists of i=y1 [i] lim xi /ˆi ˆ y ω That is. . then x converges in y. is the density of of x in N. ρ(x) = 0. a. From now on we shall use the term ‘density’ for asymptotic density. One notion they use is asymptotic density. x converges. where p is the set of prime numbers and q is its complement. If x is finite. . they need a more discriminating notion. . . cvg(x). The asymptotic density of a set. a. if there is one. d. y) = r iff ∀(δ > 0)∃n(∀i > n)(r − δ < xi ˆ < r + δ))) yi ˆ c. if x has a density in N. cvg(x. . Since they often want to compare infinite subsets of N. But the evens are much smaller than q. pk . The density of x. iff x has a density in y.3. p and q will be nearly the same size. the density of x in y. that 6. e. x diverges in y. Definition 6. If x ⊆ y = ∅. the fraction of numbers less than or equal to i that are ˆ members of x. for example.3.1. otherwise. SETS OF NATURAL NUMBERS is the ordering: p1 . of |x[n] | as n grows.94 Suppose. the asymptotic density of [2n] is 1/2. For example. . In any model of OUTPACING .

there is a set with density r. xi . ˆ ˆ If n > 1.000 and 99. z). ρ(x) = 1. y outpaces x. Let x = i | ∃n(102n ≤ i < 102n+1 . z) < ρ(y. if . Suppose r is given. between 10. If 0 ≤ r ≤ 1 and y is infinite.9. Construct the set x as follows: x0 = ∅ xi . Modify the construction for (b) in the obvious ways. z) and ρ(x. If 0 ≤ r ≤ 1. If x is cofinite.6. Theorem 6. ρ([2n]) = 1/2 ρ([4n] . [2n]) = 1/2 ρ([4n]) = 1/4 d. Suppose that both x and y converge in z. So x contains all numbers between 0 and 9. ρ(x. Proof. Fact 6. If cvg(x.3. b. there is a set with density r in y.999. c.3.1 and x10 ˆ b.3. y)ρ(y. There are divergent sets. c. y) and cvg(y. z). x= xi if xi ≥ r. between 100 and 999. and so forth. Then. So xk cannot have a limit. 2n+1 2n ≥ . a. xi+1 = c. z) = ρ(x. then cvg(x. 95 a.4. i + 1. then x10 ≤ . z). ˆ otherwise. SIZE AND DENSITY b.3.

Let x be an infinite set with density r. there are uncountably many sizes of sets in A. z1 ) < ρ(y. There are uncountably many subsets of y with distinct densities in y. if 0 ≤ r ≤ 1. Proof.5 to say that ρ(x.4.3. then a. z2 ) Proof.4 implies that in any outpacing model. z))/3 6. sets with distinct densities will have distinct sizes. If A is an outpacing model. z1 ) < ρ(y. Then ρ(x.5. we can begin to appreciate how precise an ordering outpacing models provide: Fact 6.3.96 Proof. z2 ).b and (c). and both x and y converge in z2 . SETS OF NATURAL NUMBERS n1 = the least n such that for i > n. z) < b ˆ z for any i > n1 + n2 . z) + b < ρ(y. for any y1 ⊂ y. and b. z2 ) > ρ(y. from Facts 6. z2 ).3. but ρ(x. let x contain every third member of z1 .3.3. z1 ) = 1/3 and ρ(y. Even if two sets have the same density. ρ(x. z) + b ˆ z y i /ˆi > ρ(y. let y be an infinite subset of x. − b < xi /ˆi − ρ(x. and let y be z1 − x.3.3b and Theorem 6.6. Theorem 6. . 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.3. xi /ˆi < ρ(x. though ρ(y1 . then ρ(x. If ρ(x. z) < b ˆ z n2 = the least n such that for i > n. x) = 0. So. z2 ): Let z1 be the set of primes. z1 ). then x outpaces y. So. z1 ). and let z = x − y. there are uncountably many sizes in A. z2 ) ≤ ρ(y.5. Immediate from Fact 6. a. z) − ρ(x. z) − b ˆ z ρ(x. where ρ(y. z) − b. y outpaces x. b. as in Theorem 6.3. Since ρ(x. z1 ) = 2/3. then even among sets with density r. z2 ) < ρ(y. they will differ in size if they have the same density in some common set. N) = ρ(y. Theorem 6. We cannot strengthen the consequent of 6. N) = 0. But if ρ(x. Let Let and let So we have and But So So So b = (ρ(y. x) = 0. z2 ) ≤ ρ(y. − b < y i /ˆi − ρ(y.

then ρ(z) = ρ(x) . if r < r2 . then x < y and if r > r2 . then ρ(y) = r. we will show that a negative answer is consistent. by DISJ∪ . But x ∼ y. We can regard the cardinality of a set in P(N) as determined by its size.3. 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. we’ll discuss the consistency of an affirmative answer.3. So. For if y converges. But ρ(z ∪ y1 ) = ρ(z ∪ y2 ) = ρ(x). cardinality is a function of size. 6. then y converges. finally. while an affirmative answer may or may not be consistent. a completion because all sets are located in a single. SIZE AND DENSITY Suppose now that y1 ⊂ y. rather. Consider (1) and (1a): (1) (1a) If x ∼ y and ρ(x) = r.3. Then A by 6. Theorem 6.6. at least when we focus on P(N). then x is smaller than y. since z was obtained by removing from x a set with density 0 relative to x. In passing. (1a) is an immediate consequence of Theorem 6.) Now. so A z ∪ y1 < z ∪ y2 . It turns out that a negative answer is compatible with our theory (CS and OUTPACING). linear ordering of sizes. But the situation is really not so clear. y2 ⊂ y. 97 y1 < y2 .3. we want to know whether the density of a set is a function of its size. and ρ(y1 ) < ρ(y2 ). though two sets with the same cardinal number may have different sizes. If x ∼ y and ρ(x) = r and cvg(y).1 The Convexity Problem It’s tempting to infer from these results that the extremely fine ordering of sets by size (or. though different sizes may yield the same cardinal number. Recalling that sets may have the same density even though they differ in size. we will give a more precise formulation of this problem. 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.4. the questionable part of (1) can be expressed as (2): (2) If x ∼ y and x converges. so r2 = r.3. (It is not at all clear that this is true in any power set. second. we’ll explain why this is called the convexity problem. We shall express this fact by saying that. then x > y. there is some r2 = ρ(y).4 insures that (1) just in case (2). we may consider two additional formulations: (3) If ρ(x) = ρ(y) and x < z < y.4. then ρ(y) = r. First.

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

Open Problem: Is (1) consistent with CS and OUTPACING? x ∼ y. given any two points in the figure.6. Let K be n | x[n] = y [n] K is infinite. (2) and (4) say the same thing.7.3.8.3. F .3. (3) says that the class of sets having a given density is convex. By theorems 6. Let x be a set with density r and let y be a divergent set which neither outpaces nor is outpaced by x. any point between them (ie on the line segment from one to the other) is also in the figure. AF so (1) is false in AF . We regret that the only thing we know about (3) is that it may be false: Theorem 6. (1). so K is a member of some non-principal ultrafilter. SIZE AND DENSITY 99 The question at hand is called the ‘convexity problem’ because of the formulation in (3).3. a figure is convex if. Proof. In geometry.4 and 6. The negation of (1) is satisfied in some outpacing model. Applying this notion in the obvious way to the size ordering. .

100 6. SETS OF NATURAL NUMBERS .

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

. j) greatest common divisor of i and j n ≡ m mod j n is congruent to m modulo j . 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.102 APPENDIX A. In addition. 1.. N+ N − 0 Z the set of integers ω the smallest infinite cardinal 2ω 2 to the ω A.2 Set Theory For first order languages with boolean operations and predicates..3 Arithmetic For arithmetic. 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. 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. 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.

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

If T APPENDIX φ. Definition B. More familiar notions: a. T . then T is complete. We use only one. then T c. Fact B. 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.2.5. c. T is universal-existential iff it is equivalent to a set of prenex sentences. p. then {Ai } T. A b. If T is universal and A T . T is equivalent to T2 (T ≡ T2 ). Definition C. b. Definition C. which is based on A. T is categorical. φ is consistent.1. There are several ways to show that a theory is complete. p. If T is existential. then some finite subset of T proves φ (compactness). d. then any submodel of A also satisfies T . none of which have universal quantifiers.104 f. A and B of T .3.359] T and A can be embedded . b. Fact B. and A ⊆ B. The following are well known facts: a. T . none of which have existential quantifiers. Robinson’s notion of model completeness. φ. A model complete theory is not necessarily complete. for all i > 0. each of which has all of its universal quantifiers preceding any of its existential quantifiers. g. T is existential iff it is equivalent to a set of prenex sentences. 355]. A is a prime model of T if A in any model of T . c. [Monk. T is universal iff it is equivalent to a set of prenex sentences. If T is categorical. Fairly familiar facts: a. [Monk.4. T is model complete iff T is consistent and for any two models. 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. unless the theory has a prime model. then B T. If T is complete and T . A ⊆ B iff A is an elementary submodel of B.

MODEL COMPLETENESS 105 Fact C. If A and B are models of T . then B φ(x). For every model. a. d. φ is a primitive formula. x ∈ A. p. 356]: a. b. which is based on some equivalent formulations of model completeness: Definition C. The following are equivalent [Monk. and A φ(x). A ⊆ B. and B φ(x). of T and every A-expansion L of L. x ∈ A.1a). T is model complete. φ is a universal formula. If T is model complete and T has a prime model.3. directly or indirectly.C. A ⊆ B. the T ∪ L -diagram of L is complete. on a theorem of Monk’s (see 4. c. . then T is complete. then A φ(x). b. Fact C. 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. The A-expansion of L is the result of adding to L a constant for ˙ each element of A. If A and B are models of T .4.2. A.5. To show that a theory is model complete we rely.

106 APPENDIX .

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

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

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

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

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

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

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

Sign up to vote on this title
UsefulNot useful