This action might not be possible to undo. Are you sure you want to continue?
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
lessthan 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 oneone 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
modelcompleteness, 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:
ONEONE. 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 oneone correspondence between two sets is a relation which pairs each
member of either set with exactly one member of the other. For example, the
uppercase letters of the alphabet can be paired with the lowercase letters:
ABCDEFGHIJKLMNOPQRSTUVWXYZ
abcdefghijklmnopqrstuvwxyz
So, the standard theory says, the set of uppercase letters is the same size
as the set of lowercase 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 onetoone:
. . . −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 nth prime with the nth
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 twoplace 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 longwinded 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 ONEONE 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 oneone correspondences. More speciﬁcally, saying
that two sets are similar iﬀ they are in oneone 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 oneone 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 oneone 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 ONEONE as a theory; indeed, he oﬀered an argument for this theory.
1.2 Cantor’s argument
Cantor bases his argument for ONEONE 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 oneone 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 ONEONEa and ONEONEb.
M ∼ N → [M[ = [N[ (ONEONEa)
[M[ = [N[ → M ∼ N (ONEONEb)
ONEONEa 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 oneone 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.
ONEONEb 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 manyone and not oneone. 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 ONEONEa only works by assigning all non
empty sets the same, onemembered, cardinal number. But, this contradicts
ONEONEb.
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 ONEONEa.
There is no way to repair Cantor’s argument. Rather than leading to a
justiﬁcation of ONEONE, 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
ONEONEb. Similarly, (1.8) contradicts ONEONEa; 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 oneone correspondence. So
ONEONE 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 5051]
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 oneone correlation, seems in recent years to
have gained widespread acceptance among mathematicians” [Frege, pp. 7374].
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. 304305]
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 ONEONE as just a deﬁnition and
others embrace it as a theory, the logicists have it both ways: adding ONEONE
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 ONEONE 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 ONEONE is the
correct theory of size, then this second condition follows from the ﬁrst, since the
existence of oneone correspondences will be preserved under a oneone 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 ONEONE 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 ONEONE 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 ONEONE,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 ONEONE 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 ONEONE 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 ONEONE, 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 ONEONE 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. QUASILOGICAL
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 quasilogical 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 Fer than y, then y is not Fer 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 ﬁrstorder 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 quasilogical principles,
consistency is no problem. After all, what do the quasilogical 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 quasilinear 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 nonstandard 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 threeplace 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 quasilinear ordering
which extends the partial ordering given by the proper subset relation. Since
any partial ordering can be extended to a quasilinear 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 ONEONE 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 ONEONE 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 ONEONE, 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 oneplace 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 twoplace 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 nonlogical 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 oneto
one correspondence cannot be expressed in this language, Cantor’s principle
ONEONE 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 natom atomic boolean algebra is isomorphic to any
standard ﬁnite interpretation of L
C
with an nelement 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
nonstandard 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 ncongruence class iﬀ x = [nk +m] for some m, where
0 ≤ m < n.
b. x is an nquasicongruence class iﬀ x is the union of ﬁnitely many
ncongruence classes.
c. x is a congruence class iﬀ x is an ncongruence class for some n.
d. x is a quasicongruence class iﬀ x is an nquasicongruence class
for some n.
e. QC
n
= ¦ x [ x is an nquasicongruence class ¦
f. QC =
n>0
QC
n
Examples.
a. The set of evens, [2n], and the set of odds, [2n + 1], are 2congruence
classes.
b. [3k + 2] is a 3congruence class.
c. N is a 1congruence class.
d. N is an nquasicongruence 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 ncongruence classes. If x is the union of m of these
classes, then N − x is the union of the remaining (n − m) ncongruence
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 nquasicongruence 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 settheoretic 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 settheoretic 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 quasicongruence
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
+θ
2
= ¸ρ
1
+ρ
2
, δ
1
+δ
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
nquasicongruence class, x
. So x
is the union of k ≤ n nquasicongruence
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 welldeﬁned.
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 quasicongruence classes are near each other.)
b. Any set is near at most one quasicongruence 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 ncongruence 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 quasicongruence 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 quasicongruence class, then
a. C(x) is the quasicongruence 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 ncongruence class is a disjoint union of mcongruence 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 quasilinear 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
ncongruence 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
+δ
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 nonatomic 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 511, 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 nonempty 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 Zgroups 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 oneplace 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 nplace 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 nplace 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
(τ
∀
(x
1
) ∧ . . . ∧ τ
∀
(x
n
)) →
∃y
1
(τ
∀
(y
1
) ∧ ∀y
2
(τ
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 nonempty. 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 τreducible 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 τreducible to T
2
.
c. T
1
is uniformly τreducible 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 τreducible 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
(τ
∀
(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 τreducible 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 τreducible 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 Zgroups, 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 Nsemigroups 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 Zgroups 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
Nsemigroups, and
(iv) The following additional axioms
(y ≤ z ∧ x ≤ x +z) → x +y ≤ x +z) (13)
x ≤ I (14)
The theory of Zgroups 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 Nsemigroups is model complete.
b. The theory of Zgroups 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. 4448]. 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 Nsemigroups 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 Nsemigroups in the language which contains,
besides the constant symbols in the original theory, an individual constant,
I. The theory of Nsemigroups is modelcomplete in this language, by
4.2.2b.
We claim that the theory of Zgroups 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 Zgroups 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
Zgroups 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 τimage 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 τreducible to Zgm. The reduction is uniform since each
model / of PCA has the same domain as its Zgmmodel. 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
nonatomic, 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 quasicongruence 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
+θ
2
= ¸ρ
1
+ρ
2
, δ
1
+δ
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 ncongruence classes for a
given n.
Deﬁnition 5.2.2. If f is total and congruous, then
a. If x is an ncongruence 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 ncongruence 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 ncongruence classes are assigned sizes ¸1/n, δ), but δ is no
longer 0 in all cases, as in Q. Instead, the ﬁrst f(n) ncongruence classes are
each one atom larger that the remaining (n −f(n)) ncongruence 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 nfactor of x;
for i > m, x
i
is a common nfactor 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 2congruence
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 6congruence
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 3congruence classes are charmed
except for [6n + 4] and [6n + 5].
In any case, [2n] will include the same number of charmed 3congruence
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 universalexistential theory; so it is preserved under unions of
chains. (See Appendix, Fact B.5).
b. Each ncongruence class can be partitioned into k nkcongruence 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 0extends 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 nfactor of
˙
/ has an (m, k)partition.
c. Every charmed nfactor of
˙
/ has an (m, k + 1)partition.
Proof.
a. Mod
m,k
(x) says that x has an (m, k)partition.
b. All common nfactors are the same size and satisfy the same Mod
m,k
predicate. So, suppose that each common nfactor has k charmed m
factors and m−k common mfactors.
By (a), / has f(n) charmed nfactors and n − f(n) common nfactors.
Partitioning each of the nfactors into m subsets of roughly the same size
yields an nmpartition of /; the charmed mfactors of the nfactors are
the charmed nm factors of / and the common mfactors of the nfactors
are the common nm factors of /.
Each of the common n factors has k charmed mfactors and each of the
charmed nfactors has k + 1 charmed mfactors. In all, there are
(n −f(n))k +f(n)(k + 1)
=nk +f(n)
72 5. COMPLETENESS OF CA
charmed mfactors among the nfactors of /.
But, by (a) again, there are f(nm) charmed nmfactors of /. So
f(nm) = nk +f(n)
and k = (f(nm) −f(n))/n
c. Immediate from (b), since every charmed nfactor is one atom larger than
each common nfactor.
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 0extends 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 nfactors are charmed and n! −f(n!) are common.
By Fact 5.2.12, each charmed nfactor has an (n + 1, k + 1)partition and each
common nfactor 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 nfactors
and k(n! −f(n!)) charmed n + 1 factors from the common nfactors
In all, then , the number of charmed nfactors 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 nfactors 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 nodeatoms, 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 nodeatoms 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 nodeatoms 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 quasicongruence 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 quasicongruence classes; ﬁnite sets of nodeatoms correspond to the ﬁnite
subsets in Q.
Deﬁnition 5.2.15.
a. x is a quasinodal 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. 3.2.143.2.18)
Still following in the footsteps of chapter 3, we can characterize
˙
ˆ
A as the
collection of sets /near quasinodal 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 quasinodal set of /.
b.
ˆ
/ is an atomic boolean algebra whose atoms are the nodeatoms 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 quasinodal 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 quasinodal 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
quasinodal 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 quasinodal 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 0extends 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 0extends
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 leftmost 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 leftmost 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 quasicongruence class, then
x = (x
∪ ∆
1
(x)) −∆
2
(x)
where x
is a quasicongruence 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 oneone.
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 quasilinear
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 nonempty.
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 nonprincipal ultraﬁlter contains no ﬁnite sets. [Bell, Ch.6, lemma
1.3, p.108]
b. A nonprincipal 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 nonprincipal 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 nonprincipal, then /
F
is an outpacing model.
Theorem 6.2.5. If F is a nonprincipal 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
(τ
1
) ∪ /
k
(τ
2
)
= (/
F
(τ
1
) ∩ A
k
) ∪ (/
F
(τ
2
) ∩ A
k
)
= (/
F
(τ
1
) ∪ /
F
(τ
2
)) ∩ A
k
= /
F
(τ
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 nonatomic, 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 nonprincipal 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.
8792]), 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 nonuniversal formulas
were likewise satisﬁed.
Fortunately, these special means were available; the only nonuniversal 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 nonprincipal 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 nonprincipal 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
nonprincipal 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 9membered runs.) So, x
1
(n)
is the ﬁrst element in the nth run of x and x
2
(n) is the ﬁrst element after
the nth 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 nonprincipal 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
kcongruence 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
nonprincipal 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 nonprincipal 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 ωsequence.
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 nonprincipal 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 nonlogical 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 jth 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 modeltheoretic 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 nplace predicate of L, a set of ntuples over
˙
/, and to each nargument
function symbol of L, an nary 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 ﬁrstorder 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 universalexistential 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 universalexistential 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. NorthHolland 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. McGrawHill,
New York, 1965.
[Kurosh] Kurosh, A.G. Lectures in General Algebra. Pergamon Press,
New York, 1965.
[Monk] Monk, J. D. Mathematical Logic SpringerVerlag, New York,
1975.
[Robinson] Robinson, A. and Zakon, E. Transactions of the American
Mathematical Society. 96(1960), pp. 222236.
[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. NorthHolland 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 nfactor, 68
class size, language of, 21
classes, language of, 20
common nfactor, 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
ncongruence class, 27
nquasicongruence class, 27
Nsemigroups, 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
oneone correspondence, 1
outpaces, 83
outpacing model, 84
partial, 40
PCA, 61
prime model, 104
primitive formula, 104
principal, 87
proves, 103
quasicongruence class, 27
quasilogical principles, 11
quasinodal, 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
τreducible, 50
theory, 103
theory of
abelian groups with identity, 51
cancellable abelian semigroups
with identity, 51
Nsemigroups, 52
Zgroups, Zg, 51
Zgroups modulo I, 52
total, 40
translation, 48
trichotomy, 12
ultraﬁlter, 87
uniformly τreducible, 50
union of a chain, 103
universal, 104
universalexistential, 104
Zgroups, Zg, 51
Zgroups modulo I, 52
0extends, 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 lessthan 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 oneone 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 modelcompleteness, 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
for directing me to some useful literature. 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. gaps. 1981 It was certiﬁed and accepted by Prof. whatever errors. Fred M.ii him thanks for ﬁnding (and sometimes correcting) my errors. and for his insistence on ﬁlling in mere details. Katz June. typographical corrections. Nevertheless. and simply for being interested. The reader should join me in thanking him for his help with style and organization. and omissions of some elementary (but very long) proofs. or infelicities remain are solely the responsibility of the author. this is the text of my dissertation. George Boolos. Note Except for some minor wording changes. Thesis Supervisor and Chairperson of the Department of Linguistics and Philosophy. 2001 .
.2 Cantor’s argument . . . . . . . . 20 3. . . . . . . . .2 Addition of Set Sizes . . . . . . . . 2 The General Theory 2. 66 5. . 5 Completeness of CA 63 5. . . . . . . . . . . . . . . . . 38 4 The 4. . . . . 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 4. . 1. . . . . . . . . . . . . .2 Prime models for CA .1 The outpacing principle . .2 4. . . . . . . . 3 A Formal Theory of Class Size 19 3. . . . . . 83 6. . . . . . . . . . . . . . . . . . . . . .2 CA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 CS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Size and density . . . . . . . .2 Models of CS and Outpacing .1 Model completeness of CA . . . . . . 79 6 Sets of Natural Numbers 83 6. . . . . . . . . . . Remarks on showing that PCA axiomatizes PCS 45 45 47 53 61 . PCA is model complete . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i 1 1 3 6 7 9 9 16 . Some models and theories . . . . . . . . . . . . . . . . . . .1 The Problem . . . . . . . . . . . . . . . . . . . . . . . . . . .1 The Theory CORE . . . .Axioms for CS . . . . . .3 Summary . . . . . . . . . . . . . . .3 Cantor and the logicists 1. 94 iii . . 22 3. . . . . . . .The Theory of Class Size . . . . . . . . . . . . . . . . . . . . 86 6. . . . . .4 Pure Theory of Class Sizes Size models and PCS . . . . . . . . . . . . . . .3 Remarks on showing that CA axiomatizes CS . . . . . . . . . . . . . . . . . 1. . 63 5. . . . . . . . . . . . . . . . . . .4 Aims and outline . .Contents Preface 1 Introduction 1. . . . . . . . . . . . . . . .3 4.
103 C Model Completeness . . . . 104 Bibliography Index of Symbols Index of Theories Subject Index 107 108 110 112 . . . 101 B Model Theory . . . . . . .iv CONTENTS Appendix 101 A Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
CANTOR< . . n 2n . The theory goes beyond intuitions. The prejudice stems from Cantor’s theory of set size. naive and otherwise. I spend a lot of time defending the theory because no one who comes to the matter comes without prejudice. the set of uppercase letters is the same size as the set of lowercase letters..... Fine and good. y.. A oneone correspondence between two sets is a relation which pairs each member of either set with exactly one member of the other. but not the same size as y itself. A set.. the uppercase letters of the alphabet can be paired with the lowercase letters: ABCDEFGHIJKLMNOPQRSTUVWXYZ abcdefghijklmnopqrstuvwxyz So. . −3 −6 −2 −4 −1 −2 1 0 0 1 2 2 4 3 6 . Cantor’s theory consists of just two principles: ONEONE.1 The Problem This paper proposes a theory of set size which is based on intuitions. so it needs both justiﬁcation and defense. just in case x is the same size as some subset of y. Two sets are the same size just in case there is a oneone correspondence between them. as theories will.... −n −2n . it is so clearly true that anyone who comes to the matter without prejudice will accept it. For example. which is as old as sets themselves and so widely held as to be worthy of the name the standard theory. the standard theory says. 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 onetoone: . I spend very little time justifying the theory.. . ....1 Introduction 1. is smaller than a set.. x... .
The theory proposed here accommodates these bits of common sense reasoning. common sense chokes on the standard theory. • The number of α’s is greater than the number of β’s. and A > B is read as A is larger than B. while the prime numbers are few and far between. A ∼ B is read as A is the same size as B. to the readings of the three predicates given above. In the ﬁrst case. it’s the only conclusion common sense allows. • The size of A is larger than the size of B. we use three twoplace predicates: <. INTRODUCTION Similarly. have) the same size. • The number of α’s is the same as the number of β’s. To state this theory. even into the inﬁnite. ∼. If A and B name sets. . This is just to say that common sense seems to follow: SUBSET. then the ﬁrst is larger than the second. • There are just as many α’s than β’s. given the number theory. you need a little number theory in addition to common sense. Incidentally. it is still clear to common sense that the former is smaller than the latter. to use this reasoning. If one set properly includes another. assuming that A is the set of α’s and B is the set of β’s: a. The integers contain all of the even integers and then some. b. It maintains SUBSET and a few and far between principle and much else besides. In both of these cases. c. common sense holds that the set of integers is larger than the set of even integers. Though the set of primes is not contained in the set of even integers. • The number of α’s is less than the number of β’s. we assume throughout this thesis that the following schemata are equivalent. but. where it comes up against the standard theory. So it’s just good common sense to believe there are more of the former than the latter. Common sense can make decisions without help from SUBSET. • A and B are (or. then • • • A < B is read as A is smaller than B. item by item. the standard theory says that the set of positive even integers is the same size as the set of prime numbers: pair the nth prime with the nth positive even number. No doubt. • There are more α’s than β’s. and >. • There are fewer α’s than β’s. One out of every two integers is even. • The size of A is smaller than the size of B.2 1.
The relation picked out is well studied and well worth the study. In addition. More speciﬁcally. In stating ONEONE and CANTOR< . This is true. per se. so it might be wondered whether the standard theory is really so standard. depends on neither the particular elements it contains nor on how those elements are arranged: The cardinal number. M . Though Cantor’s theory is usually taken as a theory of set size.2 Cantor’s argument Cantor bases his argument for ONEONE on the idea that the size of a set. Statements about sizes can be translated in familiar and longwinded ways into statements about sets. Whether or not similarity is coextensive with being the same size. They use the term in just the way that we use size and slide freely among (A). saying that two sets are similar iﬀ they are in oneone correspondence can either be taken as a claim about size or be regarded as a mere deﬁnition. Neither does it insure that two sets have the same cardinal number just in case they are in oneone correspondence. p. he oﬀered an argument for this theory. 1. of Cantor. Cantor’s. All of this is to say that the interest in oneone correspondence has not been sustained solely by its identiﬁcation with the notion of size. To ﬂesh out this notion of double abstraction. who oﬀered ONEONE as a theory. (B). Cantor reduces it to a second abstraction operator. The technical brilliance of the theory attests to this: it has given us the transﬁnite hierarchy.g. Hence. is rarely used in connection with Cantor’s theory. 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 can also be taken as just a theory of oneone correspondences. 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 term size. In the literature.2. the deﬁnition is worth making. indeed. the term cardinal number (sometimes just number ) is used most frequently. If someone introduces cardinal number as a deﬁned predicate or as part of a contextual deﬁnition (e. it’s cardinal number. “We say that two sets have the same cardinal number just in case . [Cantor. CANTOR’S ARGUMENT 3 Regarding this last group. one which works on the elements of sets rather than the sets . and (C).86] But to say that the cardinal number of a set does not depend on certain things is not to say what the cardinal is. ”). with two principles about set size. But most mathematicians and philosophers do not use cardinal number as a mere abbreviation. and much else. in particular. however. the continuum problem.1. . nor that there are not really such things. we emphasize that we are not arguing that there really are such things as set sizes. Cantor used the terms power and cardinal number rather than size. I have identiﬁed the standard theory. though we will not bother to do so. there is no point in discussing whether that person is right about size. denying that they are the same does not endanger the theory of oneone correspondences.
80] I have parenthesized the expressions above where Cantor describes cardinal numbers as mental entities. [Cantor. M . [Cantor. INTRODUCTION Every element.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. I can only make sense of his arguments insofar as he treats cardinal numbers as sets: he refers to them as ‘deﬁnite aggregates’. if we abstract from its nature becomes a unit.86] According to Jourdain. and M  as the cardinal number of M .1) gives a deﬁnition of cardinal number. says Cantor. Nevertheless. (p. is a set composed of units (which has existence in our minds as an intellectual image or projection of M .86] and so. concludes Cantor: The cardinal number.1) (1.” [Cantor. Unit (x) as x is a unit. m. (ONEONEa) (ONEONEb) M ∼ N → M  = N  M  = N  → M ∼ N ONEONEa is true.4 themselves: 1. of M with f (m) M transforms into N without change of cardinal number. supposes that they have elements. then in replacing each element. p. p.2) (1. So (1.) [Cantor. The following three statements seem to express Cantor’s intent: (1. and employs mappings between cardinal numbers and other sets. whereas its cardinal number is an abstract picture of it in our mind. Cantor distinguised very sharply between an aggregate and a cardinal number that belongs to it: “Is not an aggregate an object outside us. 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.80] and so. in terms of the operation of abstraction.88) . if f is a oneone mapping from M onto N . m. from which Cantor proves both ONEONEa and ONEONEb. p.
1). So. Unless each element of a set abstracts to a ‘special’. by the deﬁnition of M . unit. .6). by (1. But. onemembered. so one is much the same as another. i. and similarity is an equivalence relation. Generalizing this argument yields (1. which entails (1. That is. . out of M in such a way that from every element m of M a special unit of M  arises. A weak version of this principle is: M = {a. ONEONEb is true.6) ∀x∀y(Abstract(x) = Abstract(y)) For. CANTOR’S ARGUMENT 5 In its weakest form. So Cantor’s argument for ONEONEa only works by assigning all nonempty sets the same. Abstract(b)} ∧ Abstract(a) = Abstract(b) (1. the correspondence from M to its cardinal number will be manyone and not oneone. consider an arbitrary pair of elements.2. It is not the elements of a set. So. But abstractions of elements are just units. cardinal number. b} ∧ a = b → M  = {Abstract(a). But M  = {Abstract(a)} and N  = {Abstract(b)}. one element is much the same as another. since a set is similar to its cardinal number.7) ∀x∀y(x = y → Abstract(x) = Abstract(y)) . Let M = {a} and let N = {b}. (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. this means that (1. (1.4) says that replacing an element of a set with any element not in the set does not aﬀect the cardinality. the argument that a set is similar to its cardinal number relies on (1. Thus we can say that M ∼ M . two sets with same cardinal number are similar. that enter into the cardinal number of a set. distinct. So Abstract(a) = Abstract(b). But. but only their abstractions. M  grows. Conversely.e. because .5). 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. Cantor says.1).1. (1. the conditions of (1.4) are met and M  = N . a and b. so to speak.5) These two arguments do one another in. this contradicts ONEONEb.
then {a} and {b} have distinct cardinal numbers. the result runs perpetually together into one and we never reach a plurality . We have shown that (1. Similarly.6) contradicts ONEONEb. then no two sets have the same cardinal number as deﬁned by (1. 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. (1. or whatever we choose to call it) transforms itself into arithmetical addition.1). says Frege. Frege spends an entire chapter of the Grundlagen mocking mathematicians from Euclid to Schroder for deﬁning numbers as sets of units. Rather than leading to a justiﬁcation of ONEONE. INTRODUCTION assuming only that any two objects can constitute a set. seems in recent years to have gained widespread acceptance among mathematicians” [Frege. The negation of (1. The word ‘unit’ is admirably adapted to conceal the diﬃculty .6) or its negation. pp 5051] 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. for example. {Abstract(a)} and {Abstract(b)}.3 Cantor and the logicists Though both Frege and Russell accepted Cantor’s theory of cardinality. “This opinion”. Kossak. p.) . pp. Cantor’s deﬁnition of cardinal number is suﬃcient to refute the principle. and Cantor. is true. 1.8) contradicts ONEONEa. 7374]. . or uniting. . “that numerical equality or identiy must be deﬁned in terms of oneone correlation. Frege cites Schroder. . Russell displays similar caution about Cantor’s argument [Russell.6) is: (1. while the concept word ‘unit’ changes unperceived into the proper name ‘one’. But if the abstractions of any two elements are distinct. neither accepted Cantor’s argument.8).8) must be true.6 1. (1.6) and (1. then subsequently the concept of putting together (or collecting. So ONEONE is false whether (1. There is no way to repair Cantor’s argument. by putting together identicals.305] and similar enthusiasm for his theory (see the quote at the beginning of Chapter 2. He neatly summarizes the diﬃculty with such views: If we try to produce the number by putting together diﬀerent distinct objects. and that is not the number. We start by calling the things to be numbered ‘units’ without detracting from their diversity. or annexing.8) ∃x∃y(Abstract(x) = Abstract(y)) So one of (1. contra ONEONEa. . despite the fact that they are in oneone correspondence. [Frege. But if we try to do it in the other way. if a and b have distinct abstractions.
independent of its terms and their order. So the notion of cardinality is born reduced. 304305] So Russell. Well. and similarly Frege. to a large extent. oﬀers an account of where the intuitions come from (viz. Suppose. that is to say. the logicists have it both ways: adding ONEONE as a formal deﬁnition to set theory (or. The logicists’ adoption of Cantor’s theory of cardinality needs no great explanation: it came with set theory and. depending. for example. at the same time. Frege and Russell cleaned up Cantor’s presentation of the theory. In addition.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 ONEONE or by associating the acceptance of that theory with a discredited philosophy of mathematics. logic) they have no obligation to defend it and can steer clear of peculiar arguments about units . in contrast to Cantor.4. pp. you wish to reduce ordered pairs to sets. Such points may be interesting. relied upon the principle of abstraction to obtain a ‘formal deﬁnition’ of cardinal numbers. Second. known facts about ﬁnite sets). we might feel tempted to add. as they would call it. Chapters 4 and 5 prove that the theory so obtained is . the relative sizes of sets of ordered pairs should be preserved under translation.1. First. 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. But. Russell. so each ordered pair must be identiﬁed with a distinct set. no matter how they are intended. if ONEONE is the correct theory of size. Chapter 2 canvasses common sense intuitions for some basic principles about set size. AIMS AND OUTLINE 7 Of course. only upon their number. they can advance it as a great lesson for simple common sense. This dissertation presents such an alternative. who had “taken” number “to be a primitive idea” and had to rely on “the primitive proposition that every collection has a number. 1.” [Russell. Cantor’s theory clears the way for other reductions. [Russell. for example. 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. and mines this source for additional principles. then this second condition follows from the ﬁrst. p. since the existence of oneone correspondences will be preserved under a oneone mapping. motivated set theory and determined its research problems. One of the relations that has to be maintained is identity .305] So. while some people regard Cantor’s ONEONE as just a deﬁnition and others embrace it as a theory. Chapter 3 reorganizes those principles into a tidy set of axioms. But there are two speciﬁc reasons that they should have seized upon ONEONE and CANTOR< . they both have the form of deﬁnitions. but in the absence of an alternative theory of size they are less than convincing.
INTRODUCTION complete. Finally. expressible in a particular language). together with the theory in Chapter 3. in the sense that it embraces all facts about ﬁnite sets of a certain kind (i.e. Chapter 6 elaborates additional principles that concern only sets of natural numbers and demonstrates that these additional principles. . are satisﬁable in the domain of sets of natural numbers.8 1.
the second that there should not be too few principles. shocking to common sense . in the face of the proofs. therefore. 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. it ought to commit suicide in despair. since I consider that.2 The General Theory The possibility that whole and part may have the same number of terms is. or is there a way of elaborating on common sense to get a plausible and reasonably adequate theory of cardinality? To be plausible. To be reasonably adequate. I do not propose to help it. Plausibility and adequacy are conﬂicting demands: the ﬁrst says that there should not be too many principles (no false ones. I argued that a coherent theory of cardinality has to contain some principles that refer to the kinds of objects in sets. p. it must be confessed. . In this chapter. . however. I want to see how far we can go without such principles. [Russell. then x is smaller than y. as Russell says.1 The Theory CORE SUBSET. Common sense. a theory should at least avoid principles and consequences which violate common sense. First. is here in a very sorry plight. how much can you say about smaller than without using predicates (other than identity ) which relate the members of the sets being compared? I shall begin by stating a number of principles and explaining why they are included in the general theory. pace Cantor. there is the SUBSET principle: The reason for including SUBSET should be obvious. i. 358] Is common sense confused about set size. though it need not be complete. consistency). What has prompted the search for an alternative to the standard theory of cardinality is the conﬂict 9 . 2. it must choose between the paradox of Zeno and the paradox of Cantor. In the Introduction.e. If x is a proper subset of y.
I would not venture that this sort of technique. all by itself seems to be a plausible alternative to Cantor’s theory. From this it is supposed to follow that common sense cannot be relied upon. (By and large because many people think that all inﬁnite sets have the same size: Inﬁnity. and so forth: i. by and large. With carefully chosen examples. The argument hinges on a suggestion about how to resolve cases where mathematical intuitions seem to conﬂict. There is no doubt that you can lead an unsuspecting person to agree to ONEONE by focusing their attention on forks and knives. It’s clear that we need additional principles. though it surely is not enough. too.) Naturally. I do not think that such guile is needed to lead someone to agree to SUBSET. So. but that’s not what my argument depends on. is any way to ﬁnd out which of SUBSET and ONEONE is true. you might even convince someone that ONEONE is true for inﬁnite sets. x > y as x is larger than y. if you want to ﬁnd out what common sense really thinks about SUBSET and ONEONE.10 2. there’s a diﬀerence in the way that common sense supports these two principles. x ∼ y is to be read as x is the same size as y. People’s intuitions about mathematics are notoriously unreliable. I’ve actually tried this. 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. it’s possible that one set is smaller than another just in case it is a proper subset of the other. so we should opt for ONEONE.with the technical attractions that it provides.e. Given just this principle. say the odds and the evens. not to mention inconstant.) . asking people. So. where one was a proper subset of the other. SUBSET. harping on this fact might engender some unwarranted skepticism about mathematics. The Appendix gives a full account of the notations used throughout this thesis. in an unscientiﬁc way. 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. is what I expected: support for SUBSET. THE GENERAL THEORY between ONEONE and SUBSET. ﬁnite sets. All of the following seem worthy (where x < y is to be read as x is smaller than y. and what I’ve gotten. Now. you would present people with pairs of inﬁnite sets. Of course. But. husbands and wives. 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.
1. y) abbreviates ∀z((z < x ↔ z < y) ∧ (z > x ↔ z > y)) We call the principles listed above quasilogical principles because it is tempting to defend them as logical truths. The other principles might be defended in the same way. that each is a logical truth. ‘short’. where F is to be replaced by an adjective from which comparatives can be formed.1. THE THEORY CORE Theory 2. It seems that if a case could be made that these statements. ‘happy’. If x is smaller than y. it would require taking positions on many questions about logical form and grammatical form which would take us far aﬁeld and. then y is not smaller than x. I can just barely imagine presenting a theory which. But to do so without good reason would be counterproductive. antagonize ﬁrstorder logicians. is larger than. 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. in unregimented English: ASYM< . so it could be maintained that each is true in virtue of its form. possibly.2. a possible world). withheld judgment on one or more of these principles. ‘tall’. then y is not F er than x. for example: is smaller than. that would be good reason to say that there is no . If x is F er than y. taken together. But using such observations to support these principles would be problematic for two reasons. Second. I’m told. So. though the schema for INDISC∼ would have to be restricted to triples of corresponding comparatives. First. This sentence can be regarded as an instance of the schema: ASYMF .1. e. QUASILOGICAL (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. is the same size as. are inconsistent with SUBSET.g. for fear of inconsistency. but not ‘unique’ or ‘brick’. It appears that every instance of this schema is true. y) x∼x x∼y→y∼x (x ∼ y ∧ y ∼ z) → x ∼ z x>y↔y<x 11 where Indisc(x. for example. it might be that casting the principles above as instances of the appropriate schema would only explain why they are part of common sense. Consider the ﬁrst principle. there are some instances of the schemata that make for embarassing counterexamples: ‘further east than’ (in a round world) and ‘earlier than’ (in.
. 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. this is a principle which common sense has no particular feelings about. and that sets of the same size are indiscernible under the partial orderings. Even if common sense can be persuaded to take particles and arbitrary fusions of such as parts of tables. there must be some additional principles to be extracted from common sense. then the ﬁrst is the same size as some proper part of the second. but it is not . Analogous statements about physical objects are neither intuitive nor very clearly true. what do the quasilogical principles say? Only that smaller than is a partial ordering. there is the principle of trichotomy. this would undoubtedly be used to discredit them. While a theory of set size which excluded TRICH might escape ridicule. is smaller than another. This would be impossible.. there is the representation principle. If REP< is true. Second. So. it would surely be regarded with suspicion. For example. we get a model for our theory by assigning to < the relation of being a proper subset of. ﬁnite or inﬁnite. if we are given any domain of sets. As long as we restrict ourselves to SUBSET and the quasilogical principles. assigning to > the relation of properly including. First. Since common sense knows that diﬀerent sets can be the same size. After all. then x is the same size as some proper subset of y. So the strategy here is not to adduce principles and argue for the truth of each. consistency is no problem. if ‘part’ is taken to mean ‘leg or top or rim or . and assigning to ∼ the identity relation. REP< was originally included in this theory for technical reasons. no one should condemn its residual caution about (1). if the principles of common sense were incompatible with TRICH. Indeed. that the same size as is an equivalence relation. it seems that the proper strategy is to include such seemingly obvious truths and to show that the resulting theory is consistent. Since my goal is to counter such a conclusion. x. y. REP< may be open to doubt. given that the principles are logically contingent. (1) does not stand a chance of being regarded as true: (1) If one table is smaller than another. Now. the larger than is the converse partial ordering. (TRICH) x<y∨x∼y∨x>y which says that any two sets are comparable in size.’. then it seems to be an interesting and special fact about sets.12 2. (REP< ) x < y → ∃x (x ∼ x ∧ x ⊂ y) which says that if a set. it makes it easier to reduce the set of axioms already presented and it provides a basis for several principles not yet presented. THE GENERAL THEORY reasonably adequate alternative to Cantor’s theory. We shall now consider some principles which cannot be regarded as quasilogical.
by SUBSET. But x < y. I want to estimate how far we’ve gone. by TRICH. Let T = QUASI.1. y) → x ∼ y INDISC says that if two sets fail to be the same size. must be the same size as x. SUBSET. I’d like to take stock of what we already have. which will be referred to as the core theory. by SUBSET: so x < y .1. then it is entailed by the principles we have already mentioned: If x < y. Then T ≡ CORE. There are more principles to come. Suppose ¬(x ∼ y). T CORE.2. So x < y. which is logically equivalent to the conjunction of INDISC∼ and its converse ∼ INDISC: (INDISC∼ ) (∼ INDISC) ∼ x ∼ y → Indisc(x. but before proceeding. First. or larger than one but not larger than the other. The entire set of principles already adopted are equivalent to the following. by IRREF< . Proof. then x < y. So x < y or y < x. (SUBSET) (DEF∼ ) (TRANS∼ ) (DEF> ) (IRREF< ) (TRICH) x⊂y→x<y x ∼ y ↔ Indisc(x. REP< . Conversely. THE THEORY CORE 13 a principle that Cantorians could complain about.2. the core theory. . by INDISC∼ . if x ∼ x and x ⊂ y. y) Indisc(x. by INDISC∼ . 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∼ . We only need to show that ∼ INDISC is entailed by T .1. Theory 2. 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. by INDISC∼ and ASYM< . I want to reduce the principles mentioned above to a tidy set of axioms. Second. say x . Theorem 2. CORE. By REP< . then ¬(x ∼ y). 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. some proper subset of y.3. But ¬(x < x) and ¬(y < y).
x < z. Deﬁnition 2. by TRANS< . If y < z. there is a y such that y ∼ y and y ⊂ z by REP< . (ii) ASYM< . Proof. and REF∼ are logical consequences of DEF∼ . So x ⊂ z and. the only one that satisﬁes CORE. contra IRREF< . two kinds of models satisfy CORE. for some inﬁnite Y . Fact 2.6. a. A is a ﬁnite class model with basis x iﬀ ˙ (i) A = P(x).1.14 CORE T 2. the ﬁnite cardinalities determine a quasilinear ordering of the sets in which any set is higher than any of its proper subsets. TRANS∼ . If A is a ﬁnite class model. CORE is consistent.1.4. (ii) A (iii) A a ⊂ b iﬀ a ⊂ b a < b iﬀ a < b Models can be speciﬁed by stipulation the smaller than relation since larger than and same size as are deﬁned in terms of smaller than. ASYM> . If A is a ﬁnite set model. a. where x is ﬁnite. In fact. If x < y.5. then A b. So. In both cases. (iii) TRANS> . By adding TRICH we have ruled out all nonstandard interpretations of <. If x < y and y < x. THE GENERAL THEORY (i) TRANS< . (iv) INDISC∼ . then x < x. then A CORE. and IRREF> follow from the corresponding principles for < and DEF> . by DEF∼ . by SUBSET. (ii) A (iii) A a ⊂ b iﬀ a ⊂ b a < b iﬀ a < b b. there is an x such that x ∼ x and x ⊂ y . Suppose that A is a model such that . then x < y because y ∼ y. The normal ordering of ﬁnite sets is. Theorem 2.1. CORE. SYM∼ . in fact. But then x < z by DEF∼ . A is a ﬁnite set model iﬀ ˙ (i) A = { x  x is a ﬁnite subset of Y }.
then y ∈ A. 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. by REP< by induction. ˙ ˙ b. by condition (b) by SUBSET contra our supposition. a < b iﬀ a < b 15 Proof. THE THEORY CORE ˙ 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). contra our hypothesis. and c. A Then A CORE.1. b ∈ A. . If x ∈ 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. b  = n + 1 A (b < b) A (b ∼ a) A (a < b) b = n + 1 by (*). We shall prove (*) by induction on n: (*) Suppose then But if then And if then So So If a = n. If x ∈ A and y ⊂ x.2. then x is ﬁnite.
y. z) → (x ∼ x ↔ Sum(x . We shall show this by proving that ADDITION and CORE entail DEF+ : (DEF+ ) Sum(x. 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. we have to formulate our principles in terms of a threeplace predicate true of triples of sets: Sum (x. y. z) → (y ∼ y ↔ Sum(x. x ∪ y) FUNC+ says that sets bear Sum relations to one another by virtue of their sizes alone. y. y. z) Sum(x. Since the domains of our intended models contain only sets and not sizes of sets. y. .16 2. THE GENERAL THEORY 2.1. z) is to be read as the size of z is the sum of the sizes of x and y . z) → x < z ∨ x ∼ z x ∩ y = ∅ → Sum(x. z ) Addition for disjoint sets (DISJ+ ) Monotonicity of Addition (MONOT) Sum(x. y.2. Addition Functionality of addition (a) (b) (c) Sum(x. 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.2 Addition of Set Sizes We shall now extend CORE to an account of addition of set sizes. y. DISJ+ tries to say what function on sizes the Sum relation captures by ﬁxing the function on paradigm cases: disjoint sets. The following principles are suﬃcient for a theory of addition: Theory 2. This condition must clearly be met if Sum is to be read as speciﬁed above. 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. y. z) Sum(x. z) → (z ∼ z ↔ Sum(x. Given an interpretation of ’<’ over a power set there is at most one way of interpreting Sum which satisﬁes ADDITION.
CORE + ADDITION Proof. x ∪ y ) Sum(x . 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. Suppose then and So So So (x ∼ x ∧ y ∼ y ∧ x ∩ y = ∅ ∧ x ∩ y = ∅) Sum(x. y. as before.2.2.2. y.2.3. y . despite the fact that we have not yet introduced them. y . y . . x ∪ y) Sum(x . Theorem 2. y. then Sum has to be deﬁnable by DEF+ . x =z y = ∅. z) Sum(x. z) y ∼y x<z x ⊂z x ∼x y =z−x y ∼ y.2. ADDITION OF SET SIZES 17 A model of CORE must satisfy an additional principle.) Theorem 2. x ∪ y) Sum(x . Sum(x . y . 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. z) x<z∨x∼z x ∼ z. DEF+ by MONOT by DISJ+ by FUNC+ (a) by FUNC+ (b) by REP< . DISJ∪ . CORE + ADDITION Proof.
is CORE ∪ ADDITION.18 (⇐) Suppose then So So 2. THE GENERAL THEORY (x ∼ x ∧ y ∼ y ∧ x ∩ y = ∅ ∧ x ∪ y = z) Sum(x . y. → 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 extended core.2. y .2. 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. The proofs are elementary. Fact 2.5. y . EXCORE. z) Sum(x. z) Sum(x. z) by DISJ+ by FUNC+ (a) by FUNC+ (b) Theory 2.4.
DISJ∪ . Cantor’s principle ONEONE is dealt with as in case (2). But the ability to construct such a model is interesting only to the extent that the general theory it satisﬁes is reasonably adequate. for example. Since any partial ordering can be extended to a quasilinear ordering. obtaining T . In the other case. constituting EXCORE. Suppose. Below. In one case. but we can 19 . Our goal is to show that this general theory of set size can be maintained for inﬁnite sets as well as for ﬁnite sets. the existence of the model does not rule out the possibility that T is incompatible with DISJ∪ . Call this theory T. and show that T has a model over P(N). in this case DISJ∪ . then T should be maintained anyway.3 A Formal Theory of Class Size In the preceding chapter. since DISJ∪ is in EXCORE. φ. T has a model over P(N). and found two kinds of interpretations which satisﬁed these principles. We shall show this by constructing a model for the general theory whose domain is the power set of the natural numbers. we may say that they demonstrate that EXCORE is true when it is construed as being about ﬁnite sets. We produced several such principles. we may take one of three tacks: (1) add φ to T. (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. But unless we have a guarantee that the model constructed will satisfy. say. the domains of the interpretations were ﬁnite power sets. Insofar as both kinds of interpretation have quantiﬁers ranging over ﬁnite sets. we searched for principles that accord with preCantorian ideas about sizes of sets. we deal with DISJ∪ as in (1). So T just says that smaller than is a quasilinear ordering which extends the partial ordering given by the proper subset relation. that we oﬀered as a general theory of size the axioms of CORE other than REP< . It seems futile to try to rule out the need to resort to the third approach for any cases at all. the domains consisted of all ﬁnite subsets of a given inﬁnite set. For a particular principle.
T. Not only is ONEONE such a statement. where L is some judiciously chosen language. but only that they do not refer to these sorts of objects as such. To avoid ONEONE. the possible existence of any such hierarchy is sheer speculation. the two . At this point. relation. We then construct a theory by taking all statements in that language which are true over all ﬁnite power sets. To avoid the problems involved in speaking of all statements. In short. LC . In fact. such a language can be obtained by including a notion of the product of set sizes. or number of argument places. is the ﬁrst order language with individual constants ∅ and I. now.The Theory of Class Size The theories discussed in this paper will be formulated within ﬁrst order predicate logic with identity. for each predicate and each operation symbol. or other sets of ordered pairs.1. This is not to say the axioms do not apply to relations. But.1. Atom.1 CS . They do not use the notions ordered pair. since we include in the language all statements of the language which are true over any ﬁnite power set. we arrive at a theory which can be satisﬁed over inﬁnite power sets. LC . we deﬁne a language just strong enough to express EXCORE. but using the notion of all statements true for ﬁnite sets presupposes that we have some idea of the range of all statements. To specify the language in which a theory is expressed. In the next section. the oneplace predicate. we mention it only to emphasize that no claim is made here that we have the strongest possible general theory of set size. the language of classes. A FORMAL THEORY OF CLASS SIZE reduce this need to the extent that we include in our general theory. or a sort of Venn diagram set theory. functions. we need only list the individual constants. we know that no statement in that language can arise as something which ought to be true over inﬁnite power sets but might be incompatible with our theory. By drawing statements only from this relatively weak language.20 3. 3. predicates. the only set theory implicit in these axioms is boolean algebra. Deﬁnition 3. we might instead settle for all statements in L. a. Consider. and operation symbols of the language and stipulate the rank. T must fall short of the full expressive power of the language of set theory. these axioms involve only boolean operations and inclusion relations among sets. the axioms in EXCORE. This in turn opens the possibility of a succession of richer languages and a corresponding succession of stronger theories of size. though it would have to remain less expressive that the full language of set theory. There remains the possibility that we could follow the same strategy with a more expressive language. or function. then. We cannot construct T by taking all statements which are true for ﬁnite sets. as many plausible statements as possible. Other than size relations.
THE THEORY OF CLASS SIZE 21 place predicate ⊂. then x is the basis of A. and ∪.3. and Unit(a) iﬀ a = 1 Deﬁnition 3. If A is a standard interpretation and A = P(x). L< . a. a ∼ b iﬀ a = b. B(A). A is a standard ﬁnite interpretation of LC< iﬀ a. A A A a < b iﬀ a < b. is the ﬁrst order language with the oneplace predicate Unit. and the two place operation symbols.1.1. is the set of all sentences of LC< which are true in all standard ﬁnite interpretations of LC< . the language of size. even though it is true over any ﬁnite power set. and the three place predicate Sum. for some x. we deﬁne the theory of class size in terms of interpretations of LC< over ﬁnite power sets.4. A is a standard interpretation of LC< . Following the strategy outlined above. A has a ﬁnite basis. b. CS . Since the notion of onetoone correspondence cannot be expressed in this language.1.3. A(∅) = ∅. . ∩.2.. is the ﬁrst order language containing just the nonlogical constants in LC and L< . CS. a ⊂ b iﬀ a ⊂ b A ˙ b. the theory of class size. b. and (ii) A assigns the usual interpretations to all constants of LC : A(I) = x. c. Deﬁnition 3. Cantor’s principle ONEONE will not be included in the theory CS. the language of class size. By drawing only on principles which can be stated in LC< we have at least ruled out the most obvious danger of paradox. Deﬁnition 3.1. then A is a standard interpretation of L iﬀ ˙ (i) A = P(x). LC< . If LC ⊂ L. and c. the twoplace predicates < and ⊂. −.
A FORMAL THEORY OF CLASS SIZE Since CS has arbitarily large ﬁnite models. must satisfy BA: Deﬁnition 3. 151].2. Since all of the universes of the interpretations mentioned in the deﬁnition of CS are power sets. 3. the theory of atomic boolean algebras. But BA does entail all the sentences in CS which do not themselves involve size notions. after a slight detour in Chapter 4. in which the universe is an inﬁnite power set. In Chapter 6 we show that such models do exist.3. so. 9. BA. This will be done in several stages.2. BA is clearly not a complete axiomatization of CS. since BA does not involve any size notions.Axioms for atomic boolean algebra We’ll begin with the obvious.2 CA . 3. Section 2 develops a set of axioms. 141 and Def. . 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.22 3. The present chapter is devoted to getting a clearer picture of the theory CS. To show this we need to draw on some established facts about the complete extensions of BA (in the language LC ).1 BA. p.1. is the theory consisting of the following axioms: x∪y =y∪x x∩y =y∩x x ∪ (y ∪ z) = (x ∪ y) ∪ z) x ∩ (y ∩ z) = (x ∩ y) ∩ z) x ∩ (y ∪ z) = (x ∩ y) ∪ (y ∩ z) x ∪ (y ∩ z) = (x ∪ y) ∩ (y ∪ z) x ∩ (I − x) = ∅ x ∪ (I − x) = I x ⊂ y ↔ (x ∪ y) = y ∧ x = y Atom(x) ↔ (∀y)(y ⊂ x → y = ∅) x = ∅ → (∃y)(Atom(y) ∧ (y ⊂ x ∨ y = x) These axioms are adapted from [Monk. It is not obvious that CS has standard inﬁnite models. Def 9. CA. CA. for CS. Section 3 outlines the proof that CA does indeed axiomatize CS. This proof is presented in Chaper 5.Axioms for CS Here we shall develop a set of axioms. they must be atomic boolean algebras and. for the theory CS. it has inﬁnite models. p.28.
2. Theorem 21.3. BAI and the theories BAn . BAn is categorical. EXACTLYn is a sentence which says that there are exactly n atoms: ATLEASTn ∧ ¬ATLEASTn+1 c.151] d.152] b. BA ∪ INF φ. BA is complete. For n ≥ 1. If BAI Proof. p. By 3. CA .2.32. φ ∈ CS. a.2.2. 9.5... For n ≥ 1. Proof. and φ ∈ LC . an natom atomic boolean algebra is isomorphic to any standard ﬁnite interpretation of LC with an nelement basis. By compactness. If A is ﬁnite. ∧ Atom(xn ) ∧ { xi = xj  0 < i < j ≤ n }) b. Prop. [Monk. For n ≥ 1. For n ≥ 1. then BA φ. Theorem 3. / If A is inﬁnite. then φ is true in some ﬁnite model of BA... then ¬φ is true in A .2. then A is isomorphic to some standard ﬁnite interpration A of LC . 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.3. [Monk. so φ is not true in A and φ ∈ CS.4. Cor 9.360] e. [Monk. ATLEASTn is a sentence which says that there are at least n atoms: ∃x1 .∃xn (Atom(x1 ) ∧ . p. again. n ≥ 1. BAn is complete.30. EXACTLYn e. consistent extensions of BA. Hence.24. But BAI is complete. INF is a set of sentences which is satisﬁed in all and only inﬁnite models of BA: INF = { ATLEASTn  n > 1 } d. so BAI ¬φ. φ. If BA φ. then ¬φ is consistent with BAI.4. 23 a.2. ¬φ is true in some standard ﬁnite interpretation of LC and. BAn = BA. Fact 3. are the only complete. BAI = BA ∪ INF Fact 3. / . (Immediate from above) c. A.2. then ¬φ is true is some atomic boolean algebra.AXIOMS FOR CS Deﬁnition 3. p. then. But. ¬φ is true in some ﬁnite model A of BA. If CS φ.
2.2. The models will.2. include diﬀerent subsets of N and assign diﬀerent size orderings to these sets. (2) their atoms will be singletons in P(N). is deﬁned as: BA ∪ SIZE 3. A FORMAL THEORY OF CLASS SIZE 3. the Class Size Axioms. These models will be similar in that (1)their universes will be subsets of P(N). these models of BASIC will reappear as submodels of various standard models of CS over P(N).24 3. though we will not prove that CA ≡ CS until Chapter 5. So.2. however. CA. BASIC. In Chapter 6. we just gather the principles presented above as EXCORE: Theory 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. In this section we shall exhibit an inﬁnite number of principles which need to be added to BASIC in order to axiomatize CS. 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. y) ¬(x < x) x<y∨x∼y∨x>y Unit(x) ↔ Atom(x) Sum(x.7. When we are done. and (3) all boolean symbols will receive their usual interpretations.3 Division principles BASIC is not a complete axiomatization of CS. To show that the new principles really do need to be added. y. the basic theory. these models provide a glimpse of how sets of natural numbers are ordered by size. we will have an eﬀective set of sentences. we obtain our ﬁrst serious attempt at a general theory of size: Deﬁnition 3. . in addition to the immediate purpose of establishing independence results. we’ll need some nonstandard models of BASIC.2 Size principles Here.6.
(EVEN ∨ ODD). (EVEN∨ODD) says that the universe is roughly divisible by two: EVEN says that the universe is divisible by two without remainder. 1. 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. we also need to include all sets which are near the set of evens or the set of odds. where.2. in which case their union is also ﬁnite.8. So EVEN∨ODD is in CS. A.2. this is just the tip of the iceberg of principles missing from an axiomatization of CS. ODD says that there is a remainder of a single atom. Again. we can construct an inﬁnite model of BASIC . But this model will still not satisfy CS. we can extend the model and again we can produce a statement of CS which is false in the resulting model. 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. or 2. F is a model of BASIC. As with (EVEN ∨ ODD). To round out the result to a model of BASIC. in which case they cannot be disjoint. We can construct a similar statement that says the universe is roughly divisible by three — with remainder 0. So. But BASIC (EVEN ∨ ODD). or both coﬁnite. for any two sets that are the same size are either both ﬁnite. for a.e. If 0 ≤ m < n. 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. i. 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.AXIOMS FOR CS 25 Every standard ﬁnite interpretation.3. CA . this statement is in CS but is satisﬁed by neither our original model nor the model as amended. 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. but not entailed by BASIC. But neither EVEN nor ODD is true in F. as we can see by generalizing the argument above. Consider the model F whose universe consists of all only the ˙ ﬁnite and coﬁnite subsets of N. Deﬁnition 3. Informally. As you might suspect. then . We now formalize this line of reasoning.
n−1 Fact 3. we need some independence results for sets of DIVn sentences. . A FORMAL THEORY OF CLASS SIZE ∃x1 . DIVJ = { DIVn  n ∈ J } b. The model. b. To build such models from subsets of N. We did consider this. MODn.m is the sentence 3. . by 3. If n > 1. . but it does not work.9c. ∃ym ( x1 ∼ x2 ∼ . . BDIVj = BDIV{j} Our independence results will be obtained by constructing models of BASIC which satisfy speciﬁc sets of DIV sentences. . . If J is a set of natural numbers.11. for any n ﬁnite sets have a ﬁnite union and any two coﬁnite sets overlap. .10. BASIC CS. So. .∃xn ∃y1 . xn ) ∩ (y1 ∪ . . then (a) (b) (c) A A CS MODn. ym ) = ∅ ∧ (x1 ∪ .26 a. ∧ Atom(ym ) ∧ 1≤i<j≤n (xi ∩ xj = ∅) (yi ∩ yj = ∅) 1≤i<j≤m ∧ ∧ (x1 ∪ . . We could consider adding all DIVn sentences to BASIC in the hope that this would yield a complete set of axioms for CS..2. Deﬁnition 3. . If 0 ≤ m < n and A is a standard ﬁnite interpretation of LC< . ym ) = I) MODn. BASIC Proof. . BDIVJ = BASIC ∪ DIVJ c. DIVn is the sentence MODn. To demonstrate this. ∨ MODn. . . BASIC DIVn . then a.2.9. . .0 ∨ .2. ∼ xn ∧ Atom(y1 ) ∧ .m says that the universe is divisible into n sets of the same size with m atoms remaining.. F deﬁned above satisﬁes BASIC but not DIVn .2. . we shall include sets which can be regarded as fractional portions of N. xn ) ∪ (y1 ∪ .m iﬀ B(A) ≡ m mod n DIVn DIVn CS Theorem 3.
then N − x ∈ QCn .3. . [3k + 2] is a 3congruence class. .2. For n ≥ 0. [2n]. QC = Examples. then x ∪ y ∈ QCn . a. are 2congruence classes. 27 a. b. . b. e. x is an nquasicongruence class iﬀ x is the union of ﬁnitely many ncongruence classes.AXIOMS FOR CS Deﬁnition 3.13.2.14. QCn = { x  x is an nquasicongruence class } f. d. The set of evens. . w1 and w2 such that x = (y ∪ w1 ) − w2 . . b. If x is the union of m of these classes. x is near y iﬀ there are ﬁnite sets. . n>0 QCn Deﬁnition 3. x is a quasicongruence class iﬀ x is an nquasicongruence class for some n. N is a 1congruence class. x is a congruence class iﬀ x is an ncongruence class for some n.y). a. d. c. Fact 3. . Proof. bj . c.12. then N − x is the union of the remaining (n − m) ncongruence classes. . ak ∪ b1 ∪ . CA . ak and y = b1 ∪ . If x ∈ QCn and y ∈ QCn . a.2. . ∪ [nk + n − 1] Fact 3. If x ∈ QCn . .15. N is an nquasicongruence class for every n > 0: N = [nk + 0] ∪ . N itself is the union of ncongruence classes.2. Suppose x = a1 ∪ . x is near y. and the set of odds. where 0 ≤ m < n. [2n + 1]. x is an ncongruence class iﬀ x = [nk + m] for some m. . iﬀ x − y and y − x are both ﬁnite. NEAR(x.2. bj . b. Then x ∪ y = a1 ∪ .
Since x1 ⊆ x. Fact 3. which is ﬁnite. a. so (x − y) is ﬁnite. x is near z. then x1 ∪ y is near x2 ∪ y. b.18. Since x ⊆ x2 .17. b. If x is near y. So (x − z) is also ﬁnite. If x1 is near x2 . which is ﬁnite. Immediate. 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). is ﬁnite. (x − z) = ((x ∩ y) − z) ∪ ((x − y) − z) where ((x ∩ y) − z) ⊆ (y − z) and ((x − y) − z ⊆ (x − y). c. Fact 3. Then x − y ⊆ w1 and y − x ⊆ w2 . (z − x).2. A FORMAL THEORY OF CLASS SIZE ⇒ Let w1 = x − y and let w2 = y − x. But (y − x1 ) is ﬁnite. x is near x. Proof. If x1 is near x2 and y1 is near y2 . Proof. since x − x = ∅.16. . 3. a. If x is near y. Hence. Similarly. NEAR is an equivalence relation. then x is near y. and x2 is near y. then x1 ∪ y1 is near x2 ∪ y2 .28 Proof.2. ⇐ Suppose x = (y ∪ w1 ) − w2 . so x − y and y − x are ﬁnite. (x − y) ⊆ (x2 − y). x1 is near y. so (y − x) is ﬁnite. c.2. then y is near x. So the union. Fact 3. (y − x) ⊆ (y − x1 ). Suppose x is near y and y is near z. then N − x is near N − y Proof. But (x2 − y) is ﬁnite. If x1 ⊆ x ⊆ x2 .
b. So if x and y are near each other. If x ∈ A. p. If A is a class of sets such that a. a. A − x ∈ A. then x ⊆ N. (N − x) − (N − y) = y − x and (N − y) − (N − x) = x − y. then Qn forms an atomic boolean algebra under the usual settheoretic operations. But (x1 ∪ y) − (x2 ∪ y) ⊆ (x1 − x2 ) and (x2 ∪ y) − (x1 ∪ y) ⊆ (x2 − x1 ) 29 b. Since x1 is near x2 . then A forms a boolean algebra under the usual settheoretic operations. We can now deﬁne the domains of the models we will use to establish the independence results. By (a). since N ∈ {Qn } and if x ∈ Qn . Using Fact 3.2.3. [Monk. then c. by transitivity. Qn = { y  y is near an nquasicongruence class } b. then Suppose and So and So x is near y y ∈ QCn N − y ∈ QCn N − x is near N − y N − x ∈ Qn . c. Q = Examples. then x ∪ y ∈ A. Def. Q2 is the domain of the model constructed above to satisfy DIV2 . which is near x2 ∪ y2 . Q2 = { y ⊆ N  y is ﬁnite. A ∈ A. so are their complements. and n>0 Qn b. Q1 = { y ⊆ N  y is ﬁnite or coﬁnite }. Deﬁnition 3.19. If x ∈ A and y ∈ A.2. coﬁnite.2. So x1 ∪ y1 is near x2 ∪ y2 .141 and Corr 9. x1 ∪ y1 is near x2 ∪ y1 . If n > 0. CA .13b by 3. where I is interpreted as A.2.2. 9.1.2. p. {Qn } = N.20: a.AXIOMS FOR CS a. b. near [2n] or near [2n + 1] }. Fact 3. by 3.18b . a.4. Proof.20.142] Theorem 3. x1 − x2 and x2 − x1 are ﬁnite.21. If x ∈ Qn .2.
We now deﬁne a size function on all sets which are near quasicongruence classes. every singleton is in Qn since all singletons are near ∅. If θ1 = ρ1 . δ . The set x has density k/n and this is the value. Fact 3.18b Thus. First. Our intention in assigning sizes to sets is as follows: Suppose x is near an nquasicongruence class.23. Suppose and then and and and But then and So x ∈ Qn y ∈ Qn x is near x 3. A size is an ordered pair ρ. Qn is a boolean algebra. The deﬁnitions and facts below formalize this intention and demonstrate that the assignment of sizes to sets is welldeﬁned. a.2. Deﬁnition 3.2. if ρ = 0. The sizes assigned to sets are ordered pairs. δ1 and θ2 = ρ2 . ρ. So Qn is an atomic boolean algebra. Speciﬁcally. δ1 + δ2 Only some of these sizes will actually be assigned to sets.2. So x is the union of k ≤ n nquasicongruence classes. assigned to x. x . 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.13a by 3. . Moreover. a size will be assigned to a set only if 0 ≤ ρ ≤ 1. then (i) θ1 < θ2 iﬀ ρ1 < ρ2 or (ρ1 = ρ2 ∧ δ1 < δ2 ). then δ ≥ 0 and if ρ = 1. The ﬁrst member is a rational between 0 and 1 which represents the density of the set. 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. b.2.22. The δ value assigned to x is the ﬁnite number of elements added to or removed from x to obtain x. then δ ≤ 0. (ii) θ1 + θ2 = ρ1 + ρ2 . Moreover. where ρ is rational and δ is an integer. δ2 are sizes. The second member is an integer which represents the ﬁnite (possibly negative) deviation of a set from average sets of the same density.30 c.
If x ∈ QC. since NEAR is transitive. . . b.2. ∪ xj and y = y1 ∪ . y ∈ QC. since xi is inﬁnite Similarly. no two quasicongruence classes are near each other. then x is not near y. then y − x is inﬁnite. ∪ yk Suppose then But So So So a∈x−y a ∈ xi for some i. .24. then a. If x were near two quasicongruence classes. for any j xi ⊆ x − y x − y is inﬁnite. ∆2 (x) = C(x) − x. β(x) = the unique k such that x is the disjoint union of k α(x)congruence classes.) b. if a ∈ y − x. Deﬁnition 3. But this is impossible by (a).2. b.AXIOMS FOR CS 31 a. a. α(x) = the least n such that x ∈ QCn . and x = y.3. c. Let n be the least number such that x ∈ QCn and y ∈ QCn So each is a disjoint union of ncongruence classes: x = x1 ∪ . Any set is near at most one quasicongruence class. (That is. If x ∈ QC. then a. the two would have to be near each other. / xi = yj . .25. If x is near a quasicongruence class. . for any y. ∆1 (x) = x − C(x). ∆1 (x) and ∆2 (x) are ﬁnite and x = (C(x) ∪ ∆1 (x)) − ∆2 (x) Deﬁnition 3. C(x) is the quasicongruence class near x. Proof. a ∈ yj . CA .2. b.
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 ncongruence class is a disjoint union of mcongruence 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 quasilinear 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 ncongruence 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)
) Conversely.3. 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. d.30. z) θ(z) = θ(x) + θ(y) θ(x) ≤ θ(z) θ(x) = θ(z) θ(y) = 0. by (i). 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).2. 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. we know that θ(x ∪ y) = θ(x) + θ(y). For any n > 0.AXIOMS FOR CS So. CA . 0 y=∅ x = z. DEF+ : Suppose So Clearly Assume then So Let Assume then and Let So Claim For But So Qn Sum(x. . y. Qn Proof. DIVm iﬀ m  n.2.
then a. ⇐ For each i. . If J is ﬁnite. c. then the least common multiple of J. Corollary 3. 0 = Furthermore. group the n sets Ai into m collections with p members in each: B1 . There are only ﬁnitely many m for which BDIVJ Proof.2. µ(J) always exists since the product of all members of J is divisible by each member of J. we have bj ∈ Qn and θ(bj ) = p/n. Deﬁnition 3. since only ﬁnitely many m divide µ(J). Remark. So b = 0 and a = n/m. let Ai = [nk + i]. 0 ≤ i < n. So N = If i = j. Hence. 0 . 0 .32. then Qµ(J) DIVm .20: a.2. If m n.31. . DIVm . If BDIVJ DIVm . . c. then Qn BDIVn . Immediate from (b). Bm Letting bj = 1/m. if m disjoint sets of the same size exhaust N. Usually. (Ai ∼ Aj ) since θ(Ai ) = θ(Aj ) = Letting p = n/m. is the least k which is divisible by every member of J. then m  n. then Ai ∩ Aj = ∅ and Qn 1/n.2. We are now ready to show that BDIVn CS by ﬁnding a sentence in CS which entails inﬁnitely many DIVn sentences. . . if m µ(J).36 3.3. Deﬁnition 3. . then θ(x) = a/n. If J = ∅ and J is ﬁnite. Based on Theorem 3. If BDIVn b. But. ¬DIVm b. DIVm . 0 . b . Bj for 1 ≤ j ≤ m. they must each have size 1/m. A FORMAL THEORY OF CLASS SIZE ⇒ θ(N) = 1. .33. the product is greater than µ(J). But if x ∈ Qn . then m  µ(J). bm = N. 0≤i<n Ai . Qµ(J) BDIVJ since it satisﬁes DIVj for each j ∈ J. for integral a and b. Such sentences can be produced by generalizing the notion of divisibility to all sets instead of applying it only to the universe. b1 ∪ . µ(J). 0 .
y) is the formula: ∃x0 .AXIOMS FOR CS a. CS ADIVn .35. Each nonatomic set in the partition can be further partitioned into n sets of the same size and atoms. and all ﬁnite sets are roughly divisible by every n. ∃xn [x0 = ∅ ∧ xn = y ∧ 1≤i≤n 37 Sum(xi−1 . w. then x can be partitioned into nk sets of the same size and i atoms. A ADIVnk+1 . ADIVn is the sentence ∀xDivn (x) Remark. . xi )] Timesn (x. and x ∈ A. and c. b. . y) says that y is the same size as the Sum of n sets. DIVnm . v) ∧ Unit(y) ∧ T imesm (y. Proof. Thus. since i < nk and j < n. Notice that in the presence of BASIC. Divn (z) is the formula Modn. so ADIVn ADIVnm . z)] Modn. BASIC. nk j + i < nk+1 . ADIVnm .2. We have taken this opportunity to formulate the divisibility predicates purely in terms of size predicates. Fact 3. If 0 ≤ n. then nm = n. ˙ If T ADIVnk .m ≡ Modn. Proof. . where i < nk .34. Every set in every standard ﬁnite interpretation is a ﬁnite set.m (I) and DIVn ≡ Divn (I) Fact 3. CA .2. then Timesn (x. for all m b. where j < n. each the same size as x.3. A T . then Modn. we have partitioned x into nk n sets of the same size and nk j + i atoms.m (z) is the formula: ∃x∃y∃v∃w[T imesn (x. w) ∧ Sum(v. . . DIVn . a. c. ∨ Modn. If 0 ≤ m < n. Hence. By induction on m: if m = 1. But nk n = nk+1 and. for all m. x.m (z) says that z can be partitioned in n sets of the same size and m atoms. ADIVn : a. MODn.n−1 (z) d. for every n.0 (z) ∨ .2.
3 Remarks on showing that CA axiomatizes CS We know that CS CA and we want to show that CA ≡ CS. complete extension of T2 is consistent with T1 . indeed. Since T2 ⊆ T . So.2. Proof. b. we are left with a theory weaker than CS. Deﬁnition 3.1. even if we add all of the DIV sentences to BASIC. Then T = T2 . T is also a consistent. 3.3.e.13. i.32. If every consistent. it will be suﬃcient to show that every consistent extension of CA is consistent with CS. that CA CS. Theorem 11. complete extension of T2 . p.35c. CA ≡ BASIC ∪ { ADIVn  n > 0 } The remainder of this chapter and the next two are devoted to showing that CA is. T2 φ. But T is not consistent with T1 . it is reasonable to attempt an axiomatization of CS as follows: Deﬁnition 3.37.2. Obvious.3.2. by Lindenbaum’s lemma. Fact 3.38 b. .2. Theorem 3. [Monk. complete extension. consistent extension of T . then by compactness there is a ﬁnite set J such that BDIVJ ADIVn . T has a consistent.2. 3. then T2 T1 . Suppose that T1 φ. complete extension. To do so. which says that BDIVJ entails only ﬁnitely many DIVn sentences. T is a completion of T iﬀ T is a complete. Since this weakness has arisen in the case of ADIV sentences.200] b. BDIVN ADIVn for any n > 1. a complete set of axioms for CS. But then BDIVJ DIVnk for every k. a. But this contradicts Fact 3. a. ¬φ is consistent. If BDIVN ADIVn . by Fact 3. (Lindenbaum’s lemma) Every consistent theory has a consistent. A FORMAL THEORY OF CLASS SIZE c.36. Immediate from (a) and (b). T . Proof.
b. we may now concentrate on showing that every completion of CAI is consistent with CSI. where DIVn is MODn. Fact 3. we deﬁne two kinds of completions of a theory. . Deﬁnition 3.5. by Theorem 2. Remainder theories specify. ⇒ Suppose A is a ﬁnite model of T with n atoms. CA BASIC and. The model A of CA. . Every ﬁnite completion of CA is consistent with CS. T is a ﬁnite completion of T iﬀ T is true in some ﬁnite model of T. . T is an inﬁnite completion of T iﬀ T is true in some inﬁnite model of T .3.6. the number of atoms remaining when the universe is divided into n disjoint sets of the same size.3. are the completions of CAI? Recall that CA entails DIVn for every n > 0.3. of CAI has to solve the disjunction DIVn for each n. then (T is a ﬁnite completion of T iﬀ T is not an inﬁnite completion of T . REMARKS ON SHOWING THAT CA AXIOMATIZES CS 39 To prove that every completion of CA is consistent with CS. to show that every completion of CA is consistent with CS. What. ⇐ T ATLEASTn for every n. a. Proof. T EXACTLYn .EXACTLYn .1.6. T has to entail one of the disjuncts.3.0 ∨ . then. a. b.4. Deﬁnition 3. So T ¬ATLEASTn+1 and has no inﬁnite models. CSI = CS ∪ INF So.n−1 Any completion. Fact 3. for some n. ∨ MODn. for each n. So A CS. so T has no ﬁnite models. EXACTLYn is a standard ﬁnite interpretation.) Proof. BASIC is categorical in every ﬁnite power. a. b. Every ﬁnite completion of CA is equivalent to CA.3. a. If T is a completion of T and BA ⊆ T . Since T is complete. T .3.3. CAI = CA ∪ INF b.
8. The remainder theory speciﬁed by f . (⇐)If f is congruous. f is congruous by 3.9. of f is congruous. a. j ∈ Dom(f ). Proof. [Griﬃn.3. CSg is consistent. (Let f (p) = p − 1.3. f is total iﬀ Dom(f ) = N+ . d. CSIf is consistent iﬀ f is congruous. φ is consistent. Remainder functions and remainder theories a. c. then f has a solution iﬀ f is congruous iﬀ f has inﬁnitely many solutions. If f is ﬁnite.3.) Theorem 3.8b. a. c. Fact 3. there is some n such that An φ. If T is a theory.) b.7. then. (Henceforth f ranges over remainder functions. is the set of sentences { MODn.3. (⇒)Let φ = RTf . Hence. otherwise. f has a solution. Tf = T ∪ RTf . b. f is ﬁnite iﬀ Dom(f ) is ﬁnite. Since CS. we will show that CAIf is consistent just in case CSIf is consistent. then gcd(i. There are congruous f without any solutions. for all primes p. By (a). hence. g. If f is ﬁnite. So n is a solution of f and. e. n is a solution for f iﬀ for any i ∈ Dom(f ). Chapter 5 will show that if f is total. f : N+ → N is a remainder function iﬀ 0 ≤ f (n) < n for all n ∈ Dom(f ).80] b. j)  f (i) − f (j). CAIf is complete and that these are the only complete extensions of CAI. . f.3. CSf is consistent.8a. g. of f . a. By compactness. RTf . f is congruous iﬀ every ﬁnite restriction of f is congruous. CSf is consistent iﬀ f is congruous. (⇐)Every ﬁnite restriction. A FORMAL THEORY OF CLASS SIZE Deﬁnition 3. then CSf is consistent iﬀ f is congruous. (⇒)For every ﬁnite restriction. c. So An CS.m  f (n) = m }. each such g is congruous.40 3. n ≡ f (i) mod i. So CSg is consistent. n. Any solution would have to be larger than every prime. f is congruous iﬀ for any i. In this section. g. otherwise f is partial. φ b. Theorem 511. f is incongruous. by (a). p. f is congruous by 3.
. .n Since A DISJ∪ . 41 (⇐)By compactness. . if T is such a theory. then CA MODm. So. T .3. and p = k2 n + q. g is as well.3. f is congruous. Since f is congruous. k1 = m/n. cq For 1 ≤ i ≤ n.1 . it is suﬃcient to show that every ﬁnite subtheory. But.3. . Lemma 3. we must ﬁrst establish that certain sentences are theorems of CA. bk1 . So.1 . b2. T is consistent. Hence.1 . then T ⊆ CSg ∪ { AT LEAST i  i < n } for some n and some ﬁnite restriction. 0 ≤ q < n.p → MODm.q . We now want to prove a similar theorem for CA.n a2. of CSIf is consistent. of f . a2.n ∪ 1≤j≤k2 aj. our proposed axiomatization of CS. by (b).n bk1 .q .3.n ak2 . . .p → ¬MODm. . REMARKS ON SHOWING THAT CA AXIOMATIZES CS c. So g has arbitrarily large solutions and CSg has ﬁnite models large enough to satisfy T . A Bi ∼ Bj . To do this. . then CA MODm.3. A MODn. Lemma 3. Suppose A MODm.n b2. . let Bi = 1≤j≤k1 bj. cq ) So.8b.1 . j ≤ n.1 . ak2 .p . . .1 . B(A) = 1≤i≤n Bi ∪ (c1 . Furthermore.11. .n and p atoms a1. If 0 ≤ p < q < m. . . so is CSf . .10. If n  m. B(A) can be partitioned into m sets of the same size b1. b1.q . g.n c1 . for 1 ≤ i. . and p ≡ q mod n. (⇒)If CSIf is consistent. . . . Proof. a1. by 3.
since each is the disjoint union of p atoms. Suppose A 3. there is a component of Y of the same size. xm ∪ a1 ∪ . CAIf is consistent iﬀ f is congruous. Proof. ap = I y1 ∪ . . so Y ∪ Z ∼ Y ∪ B . But Z must be larger than B .f (i) ∧ MODj. But A ∼ B.42 Proof. b. Since A REP< . But Y ∼ Y . there is a proper subset yi of xi which is the same size at yi . c. for B is the union of fewer than m atoms while Z is the union of m nonempty sets. Let and So and zi = xi − yi Y = yi ∪ · · · ∪ ym Y ∪Z =X =I −A Y ∪ B = I − B. We will show that (*) CA ¬(MODi. by RC∼ . Let X = x1 ∪ · · · ∪ xm Y = y1 ∪ · · · ∪ ym A = a1 ∪ · · · ∪ ap B = b1 ∪ · · · ∪ bp B = bp+1 ∪ · · · ∪ bq We claim that y1 < x1 . and k such that k = gcd(i. So Z ∼ B . .f (j) ) . then X ∪ A ∼ Y ∪ B and if x1 < y1 . the original supposition entails a contradiction. . . . Theorem 3. ym ∪ b1 ∪ . neither is possible since Y ∪ B ⊂ I = X ∪ a. For. if y1 ∼ x1 . j.p ∧ ¬MODm. . (⇒)Supposing that f is incongruous. So yi < xi for 1 ≤ i ≤ m.12. by RC∼ .q . there exist i. Then: A A x1 ∪ . 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. . If f is ﬁnite. So. CAf is consistent iﬀ f is congruous. a.3. Thus (I − A) ∼ (I − B). A FORMAL THEORY OF CLASS SIZE MODm. a. then X ∪ A < Y ∪ B. since for each component of Y . . then CAf is consistent iﬀ f is congruous. j) and k (f (i) − f (j)).
13. Let p and q be such that 0 ≤ p. (⇐)Follows from 3.q From (1). c.12).3.10.3. congruous f . and f (j) ≡ q mod k By lemma 3. then CSIf is also consistent (see 3.9b. (2).3. (1) (2) CA CS Every completion of CA is consistent with CS.5c) and that if CAIf is consistent. it is suﬃcient to prove (4) because every completion of CAI entails CAIf for some total f .12c. It might help to review our strategy before presenting the diﬃcult parts of the proof that CA ≡ CS. To establish (3).p .1b.f (j) → MODk.3. See proof of 3.3. p = q.3.3. Corollary 3.9c.3. (4) If f is total and congruous. .11 yields (3) CA MODk.3. q < k. b. and (3) we may conclude (*). Proof. We already know that the ﬁnite completions of CA are consistent with CS (see 3.q 43 since k  i and k  j. we have (1) (2) CA MODi.9a since CAf ⊆ CSf . then CAIf is complete.9c and 3.3.3. CAIf is consistent iﬀ CSIf is consistent.3. So (2) is a consequence of (3). which follows from (2) by 3. REMARKS ON SHOWING THAT CA AXIOMATIZES CS from which it follows that CAf is inconsistent.p → ¬MODk. and CA MODj.3. Since k f (i) − f (j). (3) If T is a completion of CAI. So lemma 3. then T ≡ CAIf for some total. The main objective is (1).f (i) → MODk. See proof of 3. Immediate from 3. f (i) ≡ p mod k.
we can infer (6) from (7). The proof outlined here will be carried out in Chapter 5. Chapter 4 formulates PSIZE and establishes that it is model complete. which deals only with sizes of sets and ignores boolean relations. (5) (6) If f is total and congruous. since any extension of a model complete theory is also model complete (see Appendix. So (4) follows from (5) and (6). Finally. CAIf is model complete. So (1) follows from (5) and (7). . Fact C7a). PSIZE. Fact C. we consider a simpler theory. For any f . But ﬁrst. then T is complete (see Appendix.3). (7) CAI is model complete.44 3. then CAIf has a prime model. a result we need for showing that CAI is model complete. we invoke the prime model test: If T is model complete and T has a prime model. A FORMAL THEORY OF CLASS SIZE To prove (4).
PCS is the set of statements true in SA for any standard ﬁnite model. consists of equivalence classes ˙ drawn from A under the same size relation. But our main reason for introducing PCS is to aid the proof that CA is model complete. are equivalence classes of sets which have the same size. Suppose A ˙ a. In this chapter.we only give a sketchy treatment of PCS itself. but only about sizes of sets. of BASIC. for if number theory is the theory of cardinal numbers. A. L< .1. PSIZE. Finally. PCS (the pure theory of class sizes). it makes claims about sets in terms of their boolean relations and their size relations. which holds in SA whenever A BASIC and a set of divisibility principles. we identify a theory. Using some results about model theory in section 2. 4. PCS is formulated in the language of size relations. the size model.1 Size models and PCS BASIC. Deﬁnition 4. whose models are Zgroups taken modulo some speciﬁc element. for it. This is done by reducing PCA to the theory Zgm. This method is the same outlined in chapter 3 to show that CA ≡ CS. PCS is worth examining in its own right. For this reason. A. here. PCA. as follows: for each model. which is not about sets. the main result of this chapter and the only result needed for subsequent proofs. Section 1 deﬁnes PCS and develops a set of axioms. SA . then PCS is our version of number theory.1.4 The Pure Theory of Class Sizes CS is about sets. Sizes. If x is a member of A. section 4 indicates how PCA could be shown to axiomatize PCS. Section 3 establishes that PCA is model complete. PCA consists of a theory. then σA [x] is the size of x in A: σA [x] = { y  A 45 (x ∼ y) } .
consists of all sentences of L< which are true in the size model of every standard ﬁnite interpretation of LC< .1. z).3. then A Sum(x. 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 .46 4. and Atom(x) SA c. the size model for A. 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]. y. the theory of sizes.2. I) These interpretations are well deﬁned because the predicates are satisﬁed ˙ by elements of A in virtue of their sizes. SA . z) and A z ∼ z . z ). y.1. y. PCS. the pure theory of class sizes. PSIZE. Deﬁnition 4. σA [y]. σA [z]) iﬀ A SA Unit(σA [x]) iﬀ A x∼y x<y Sum(x. The oneplace operator x is to be read as the complementary size of x: ˜ SA (y = x) iﬀ SA ˜ Sum(x. Deﬁnition 4. For example. if A Sum(x. y. THE PURE THEORY OF CLASS SIZES b.
z) ∧ y2 ≤ y1 → ∃zSum(x. in chapter 5.5. .2. that CAI is model complete. a.2. z. PSIZE Sum(x. I) ˜ BASIC. w2 . y. w1 ) ∧ Sum(w1 . 0. z) ↔ Sum(y. x. In particular. y. . y. . this is the only interpretation allowed by the axioms PSIZE. If T satisﬁes Monk’s Condition.6. y.1. w2 ) → Sum(x. z) Sum(x. PSIZE fails to axiomatize PCS for the same reason that BASIC fails to axiomatize CS: the lack of divisibility principles.1a to show that PCA is model complete and Theorem 4. where SA Unit(x) iﬀ x = 1 and SA Sum(i. k) iﬀ (i + j) = k ≤ n Once n is ﬁxed.z2 ) → (x1 < x2 ↔ z1 < z2 ) Sum(x. w) ∃zSum(x. then T is model complete.1. n with the usual ordering. then SA PCA If A is a standard ﬁnite interpretation of BASIC with n atoms.1.7.2. z) Sum(x1 . z1 ) ∧ Sum(x2 .2. x) Sum(x. w) ∧ Sum(y. j. z) x ≤ z → ∃ySum(x. SOME MODELS AND THEORIES Sum axioms (IDENT) (COMM) (MONOT) (ASSOC) (EXIST+ ) (EXIST− ) (COMP) Fact 4.4. x. y. If A Fact 4. . k) is satisﬁed just in case (i + j) ≡ k mod n + 1. z. We oﬀer PCA as an axiomatic version of PCS: Deﬁnition 4. . MONOT rules out the interpretation in which Sum(i.1.4. Theorem 4. j. then the elements of SA can be regarded as the sequence 0. then SA PCS. is deﬁned as: PSIZE ∪ { ADIVn  n > 0 } Fact 4.1. 4. PSIZE 47 Proof. PCA.1b to show. y1 . If A CA.2 Some models and theories We shall use Theorem 4. y2 . the pure class size axioms.
and L is an expansion of L by adjoining new individual constants.359] b. To establish the model completeness of PCA.2. ∃ym φ2 So. In chapter 5. . so A does as well. f : C → A such that if x ∈ C ∩ A.2. So. .48 4.5d in the Appendix. If T is model complete and LT has no function symbols.2. . Proof. If T1 is model complete and T1 T2 . of L2 with exactly n free variables. 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 . φ3 is a primitive formula. and c. in L12 a (quantiﬁer free) formula. by Fact C. of L2 with exactly one free variable. f is a Monk mapping. THE PURE THEORY OF CLASS SIZES Monk’s Condition.2. . p. a. τ∀ . which: a. τP . A ⊆ B. a. . Finally. If T is model complete in L. . Fact 4.3. . Suppose L1 and L2 are ﬁrst order languages and L12 = L1 − L2 . of L2 with exactly (n + 1) free variables. then T satisﬁes Monk’s Condition. τO . B T . we use Monk’s Theorem to infer the model completeness of the theory CA from that of PCA. O. an (from A) and b1 . [Monk. assigns to the universal quantiﬁer a (quantiﬁer free) formula. If A T . [Monk. we use Fact 4. P . then T2 is also model complete. . . p. b. A translation (simple translation) of L1 into L2 is a function. . then T is model complete in L . to obtain the desired isomorphism. and C is a ﬁnitely generated submodel of B. in L12 a (quantiﬁer free) formula. If LT has no function symbols. then any ﬁnitely generated structure over ˙ ˙ LT is ﬁnite. suppose C contains a1 . assigns to each nplace function symbol. an ). So.2. . b. b. τ . .355] Deﬁnition 4. let φ3 be ∃y1 . . . bm ˙ ˙ (from BA). assigns to each nplace predicate. 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 . then there is an isomor˙ ˙ phism. . B φ3 (a1 . then f (x) = x.
A is the set of elements of B which satisfy τ∀ . then τ (B) is the interpretation A of L1 such that: ˙ ˙ a.2. . xn . A interprets all predicates and function symbols common to L1 and L2 in the same way that B does.5. 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. Predicates and function symbols of L2 are translated into themselves. The functional assumptions of τ are the sentences: ∀x1 . then a. A translation from L2 into L1 induces a mapping from interpretations of L2 into interpretations of L1 . .4. . . ∧ τ∀ (xn )) → ∃y1 (τ∀ (y1 ) ∧ ∀y2 (τO (x1 .2. . . 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. ∀xn (τ∀ (x1 ) ∧ . If τ is a translation of L1 into L2 . If τ is a translation from L1 into L2 and B is an interpretation of L2 which satisﬁes the existential and functional assumptions of τ . . SOME MODELS AND THEORIES 49 Deﬁnition 4. .4. b. b. The existential assumption of a translation says that that subdomain is nonempty.2. Deﬁnition 4. The relevant part of the domain is the set of elements which satisfy the interpretation of universal quantiﬁer. then τ extends to all formulae of L1 as follows: a.6. If τ is a translation of L1 into L2 . y2 ) ↔ y1 = y2 )) where O is a function symbol in L1 but not in L2 . Notice that the existential and functional assumptions of a translation are sentences of L2 .
. 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). yn ) ˙ φ (x. y) A The following condition on theories allows us to infer the model completeness of one from the model completeness of the other. . T1 is reducible (simply reducible) to T2 iﬀ there is a (simple) translation. for which T1 is τ reducible to T2 . . . Deﬁnition 4. b1 . φ.7. T1 is uniformly τ reducible to T2 iﬀ for any models. . THE PURE THEORY OF CLASS SIZES c. τ is a simple translation from LT1 to LT2 .8. . there exist models A2 and B2 such that A2 T2 and B2 T2 . yn )) iﬀ A2 τφ (x) Theorem 4.2. A P (x) iﬀ B y = O(x) iﬀ B τP (x) τO (x. . . . . ∃yn (τ∀ (y1 ) ∧ . . and T1 is uniformly τ reducible to T2 .9. for any primitive formula. and A1 ⊆ B1 . such that A1 T1 and B1 T1 . A1 Proof. . τ . bn ) ∃y1 . ∃yn φ (x. c. Suppose that T1 is τ reducible to T2 and that A1 = τ (A2 ). . . of L1 and any sequence. . If T2 is model complete. b1 . . ∧ τ∀ (yn ) ∧ φ (x. Then. then T1 is also model complete. Lemma 4. bn ).2. . . . b. and A2 ⊆ B2 A1 = τ (A2 ) and B1 = τ (B2 ). ∃yn φ (x) where φ (x) is a conjunction of atomic formulae and negations of atomic formulae. Then A1 φ(x) iﬀ A1 iﬀ A1 iﬀ A2 iﬀ A2 ∃y1 . . y1 . . Suppose φ(x) = ∃y1 . . for bi ∈ A φ(x) iﬀ A2 τφ (x) τφ (x. x ∈ A1 . A interprets all predicates and function symbols in L12 in accordance with the translations assigned by τ .2. a. . . . y1 .50 4. If τ is a translation from L1 to L2 . A1 and B1 .
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. (ii) The axioms for simple order. all more or less familiar.10. Deﬁnition 4.8 We shall now deﬁne several theories. Zg .5d in the Appendix. which will serve in showing that our theory of size is model complete. it is enough to show that given models A1 and B1 of T1 . A2 ⊆ B2 . a.2.8 since T2 is model complete by lemma 4.4.2. By Fact C. where A1 ⊆ B1 and a primitive formula. The theory of cancellable abelian semigroups with identity consists of (1).2. and (3) above and: (4 ) x+y =x+z →y =z c. 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. there are models of T2 . has the following axioms (i) The axioms for abelian groups with identity. . The theory of Zgroups. (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. φ: If then B1 A1 ˙ φ(x) for x ∈ A φ(x) Since T1 is uniformly τ reducible to T2 .
∨ ny = x + (n − 1)) for each positive n. p. The theory of Zgroups modulo I is model complete.291] b.52 4. The theory of Nsemigroups is model complete. b. The theory of Nsemigroups has the following axioms: (i) The axioms for cancellable abelian semigroups with identity. and (iv) The following additional axioms (13) (14) (y ≤ z ∧ x ≤ x + z) → x + y ≤ x + z) x≤I The theory of Zgroups is taken from [Chang. 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 ∨ .[Robinson] Theorem 4. Zg is the complete theory of Z. 291] Fact 4.11. The theory of Zgroups modulo I. +. . consists of the following axioms: (i) The axioms for abelian groups with identity. . p. (ii) The axioms of simple order. 0. [Chang. 1. Zgm. and (iv) The additional axiom: (12) 0≤x f. Zg is model complete. (iii) Axioms (9) and (10) of Zg. ≤ . (ii) The axioms for simple order (iii) Axioms (10) and (11) of Zg. a.12.2. . axiom (12) from the theory of Nsemigroups. where ny stands for: y + ··· + y n times e.2. a.
by reducing it to the theory of Zgroups modulo I.2.9 to show that PSIZE is model complete. besides the constant symbols in the original theory. Every abelian semigroup with cancellation can be isomorphically embedded in an abelian group [Kurosh. 53 a. by 4. an individual constant. b. Moreover.) In the next section.4. The result is an Nsemigroup and the original model is isomorphic to the ﬁrst copy of itself. the theory of Nsemigroups is uniformly reducible to the theory of Z groups by the translation: τ∀ = 0 ≤ x Since the latter is model complete.2b. The model completeness of PCA then follows from the model completeness of Zgm (Fact 4. so is the former.2.3 PCA is model complete To show that PCA is model complete. I.9.3. stack up ω many copies of the model. consider the theory of Nsemigroups in the language which contains. PCA IS MODEL COMPLETE Proof.12) and Theorem 4. It is clear from the construction in [Kurosh] that if the semigroup is ordered. we use Theorem 4.2. the (rough) divisibility of the elements in the semigroup will be carried over to the group. pp. First.11). by 4. assigning interpretations in the obvious way. We claim that the theory of Zgroups 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.2. Consequently. .2. we shall show that every model of PCA is the τ image of a model of Zgm. where τ is the following translation.2. we shall reduce it to Zgm. 4. 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. Chapter 5 uses Monk’s Theorem to show that CA is model complete. the theory of Zgroups with addition taken modulo some constant (see 4. Speciﬁcally.9. The theory of Nsemigroups is modelcomplete in this language. 4448].
y.3.3. a.y. z) or A ∃a∃wUnit(a) ∧ Sum(˜. on the basis of the intervening facts: 4.3. 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. z ) x ˜ ˜ (f) ZA (x = 1) iﬀ A Unit(x) Fact 4.3.1. w) ∧ Sum(w.3.2. (Sm(x.3 deals with the model A of PSIZE. A.2 gives Sum a functional interpretation. y . 4.3. The following are theorems of PSIZE. ZA . z). Let τ be the translation from LPSIZE to LZgm where: τ∀ = x = x τSum(x.54 4.3.3.4 deals with the corresponding model ZA . of PCA. we can construct a model. Theorem 4.8 establishes that ZA Zgm. ) . Fact 4.3.6 veriﬁes some connections between A and ZA .3.3j establishes that 4. If A which: PSIZE. THE PURE THEORY OF CLASS SIZES Deﬁnition 4. 4. y) abbreviates ∃zSum(x.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. y.
∅) A Sum(∅. as follows: Deﬁnition 4. z) ∧ Sum(x. PCA IS MODEL COMPLETE 55 (a) (b) (c) (d) (e) (f) (g) (h1) (h2) (i) (j) Sum(x. Given a model A of PSIZE. If x = 0. a. ZA is the expansion of ZA induced by the deﬁnitions in 4. y. z1 ) ↔ ¬∃a∃w∃z2 (Unit(a) ∧ Sum(˜. z1 ) ∧ Sum(˜. we require a battery of tedious facts about the model ZA . x) z ˜ Sum(x. y ) x ˜ Sum(x. z. y.3. y ) ˜ ˜ Sum(x. In addition to the theorems of PSIZE listed in 4. w) ∧ Sum(w. z) → y1 = y2 (UNIQ+ ) (UNIQ− ) Sum(x.2e by IDENT by UNIQ− .3. which interprets two additional operators. y. a.4. then ZA ⇒ If then But So ZA 0 + y = 0 A Sum(∅. by 4. We shall state these in terms of the model ZA .4. y. z2 ) → z1 = z2 Sum(x. w)) ˜ x=x ˜ x<y↔y<x ˜ ˜ x<y∧y <y →x<x ˜ ˜ Sm(x. y. y2 . w2 . y) ∨ Sm(˜. w) → (ASSOC2) ∃w2 (Sum(y. z) → Sum(˜. ∅. there is a unique y such that: ZA x + y = 0 Proof. ∅) ZA y = 0 x + y = 0 iﬀ y = 0. z2 )) x ˜ ˜ Proof. w2 ) ∧ Sum(x. z2 ) → (z1 = z2 = I) x ˜ ˜ ∃z1 Sum(x. z.3. y. z . y. y1 . y .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. y . The proofs are elementary. z1 ) ∧ Sum(x. w1 ) & Sum(w1 .3. z) → Sum(x. Given a model A of PSIZE: for each x in Z˙A .3. an expansion of both A and ZA . y.3. y.
3. a. −c. b. since A Sum(x. −b) Sum(a. −a) −(x + y) = −x + −y −(x − y) = y − x (x + y) − y = x −(−x) = x x = 0 ∧ x < y → −y < −x . 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(˜.3. y). b) c. a) x ˜ ˜ Sum(˜. Sm(a. ZA satisﬁes the following: a.3. Sum(a. a) x ˜ ˜ Sum(˜. 0) by COMP and UNIQ− by 4.56 ⇐ Obvious 4. y . y) x x + y = 0 iﬀ A (*). ˜ ∅) A A A A ˜=I ∅ w=a ˜ Sum(˜. ˜ a ∅) x+y =0 ZA x + y = 0 A Sum(˜. y. b. there is a unique y which satisﬁes (*) Sum(˜.2e Fact 4. where Unit (a) x UNIQ+ .3. Fact 4.3h2 by COMP by 4.5. b) ↔ Sum(a.3d and h2 by 4. c) ∧ 0 < b → Sum(a. c ≤ b ↔ Sum(c. b. THE PURE THEORY OF CLASS SIZES b. a. ZA satisﬁes the following: (a) (b) (c) (d) (e) Proof. b − c. a. Omitted. y .3. w) x ˜ A Sum(w. c) ∧ 0 < a → Sum(−c.6. a + b) b. If x > 0. b. y . a.
5c by IDENT by MONOT and UNIQ+ by MIN by (a) and UNIQ+ by 4. (⇒) Suppose So So So Sm(a. w) for some w (a + b) = w Sum(a. b. 0.4. b) for some w b=c+w b − c = (c + w) − c b−c=w Sum(c.3.3. b − c. w. −c) a=0 a < c. b) Sum(c. (⇒) Suppose then So So So So (⇐) Suppose But So So c. c) 0 ≤ (b − c) ↔ b ≤ c b≤c c≤b Sum(c. PCA IS MODEL COMPLETE d. b) by EXIST− by (a) and UNIQ+ by 4. Suppose and then So So So So But if then And if then Sum(a. a + b) 57 by 4.5c by 4. −b) if Sm(a. −b) ∨ b = −a) Proof.3. b. b. ¬Sm(a.3. b − c. c) 0<b c = (a + b) c − b = (a + b) − b c−b=a −b=a−c Sum(a. b. a. b) → (Sm(−a. b) Sum(a.3. −c.2e (⇐)Obvious. since 0 < b Sum(c.5c by (a) by IDENT by MONOT and UNIQ+ . −c) a=0 Sm(a.
Lemma 4. b) and c ≤ b We know and So So and So Sum(b − c. We shall work through successively more general cases: a.5e 4.3.3. ¬(a ≤ a + b). (a + b) = 0 − a < −(a + b). Suppose But So So So But if then So So But So So And if then − c < −a −c≤a ˜ Sm(a. (−a + −b) − (−a) = −b.5c (a) (b) TRICH 4.6b by 4.3. ¬Sum(a. (a + b) − a. w) ∧ Sum(b − c. THE PURE THEORY OF CLASS SIZES by by by by 4. (a + b) < a.5c We can now show that ZA Zgm. − a < −a + −b.3.5a (b) 4. c.58 So So But So d. (−a + −b) − (−a).4 IDENT EXIST+ ¬Sm(a.6a and UNIQ+ .6a by ASSOC2 by 4. Sum(−a. a + b) ∃w(Sum(c.3.3. We need the following lemma. a. a + b) w =c+a a + b = (b − c) + (c + a) a + b = (a + c) + (b − c) by 4. −b.3.3. When Sm(a. w.3. a) ˜ Sm(a. Sum(−a. b) b = (a + b) − a. b) Sum(b.5e 4. −a + −b) Sm(−a. The only real diﬃculty arises in verifying that addition is associative.6a and UNIQ+ by 4.7. −a + −b). −b) (a + b) = 0 b = −a by by by by by by by by 4. ZA (a + c) + (b − c) = a + b Proof. a + b).3. a.3. −c) 4.
6d by 4. −c. −(c − b)) Sum(b. for otherwise case c applies. −c) Sum(−a.4b 4. b) by ASSOC by 4.4. a) c = −a a+c=0 a = −c Sum(a + c. c) Sm(−a.3. c) and thus b < c Sum(b. b − c. −c. c) 0<c−b Sum(b. −c) or c = −a since ¬Sm(a.5b by by by by by (c1) EXIST+ 4.6c by 4. c) If So assume then So So So c.5d (c2) by (c2) by (c1) since Sm(a. −(−a)) Sum(a + c.3. b).3. b − c) Sum(a. −c. −(a + c)) 0 < −a Sum(−(−(a + c)). Whenever Suppose ¬Sm(a.3. b.3. b) and Sm(a. b.5a by 4. c) Case (b) applies ¬Sm(a. Sm(a. 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. −a + −c) Sum(−a. When Sm(a.3.3.3.5b by 4. b − c) Sm(−a.6a . a + b) Sum(a + c. −c. PCA IS MODEL COMPLETE b. b. a + b) (a + c) + (b − c) = a + b c ≤ b. 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. b − c) Sum(−c. c − b. −c.6c 4. −c.3.5c 4. −c.6b MONOT 4.3.3.6a by 4.3.3. a) Sum(−c. Sm(a.
5c The ordering axioms of Zgm are satisﬁed in ZA because ZA uses the same ordering as A. The only axiom for abelian groups that needs further veriﬁcation is associativity.3.60 Either Assume then and 4. z. and (14) of Zgm because A satisﬁes MIN and MAX.3. by 4. A PSIZE.3.5a 4. ZA Zgm Proof.3. −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.5d Theorem 4. −b) − a + −b = (−a + −c) + (−b − −c) − (a + b) = −(a + c) + −(b − c) − (a + b) = −((a + c) + (b − c)) a + b = (a + c) + (b − c) by 4.7: x + (y + z) = (x + y) + ((y + z) − y). Similarly. by 4.5a 4.3. ZA satisﬁes axiom (10) of Zgm because A satisﬁes the UNIT axiom of PSIZE.7 = (x + y) + z.3. We prove this using lemma 4. and PSIZE includes the same ordering axioms.3. x + z) ZA y < z A Sum(x. 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. THE PURE THEORY OF CLASS SIZES Sm(−a.8. y. ZA satisﬁes axioms (12).6d So Assume then So So So by by by by case c 4. x + y) A x+y ≤x+z ZA x + y ≤ x + z by EXIST+ by MONOT . 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.
4.2. then PCSIf is consistent. b.1(b). EXACTLYn .4. b. So.1b. The ﬁnite completions of PCA are the theories PCA. for some f . PCA is model complete. Every inﬁnite completion of PCA is consistent with PCS.4 Remarks on showing that PCA axiomatizes PCS To prove that PCA ≡ PCS. REMARKS ON SHOWING THAT PCA AXIOMATIZES PCS 61 So.9. we may conclude that PCA is model complete. So. we would have to deﬁne EXACTLYn in terms of units rather than atoms. which follows from (2) by 3.) Letting PCAI = PCA + INF and PCSI = PCS + INF. We already know that PCS PCA. where A is the standard ﬁnite interpretation of LC< containing n atoms. Immediate from (a) and 4. then T ≡ PCAIf . (3a) (3b) If PCAIf is consistent. PCA satisﬁes Monk’s condition.3. so we need only prove (1). The reduction is uniform since each model A of PCA has the same domain as its Zgmmodel. But PCA. (Formally. . Proof. 4. (2a) (2b) Every ﬁnite completion of PCA is consistent with PCS. But (2) is equivalent to the conjunction of (2a) and (2b). (1) (2) PCA PCS Every completion of PCA is consistent with PCS. a. a. the ﬁnite completions of PCA are consistent with PCS. we could follow the method outlined at the end of chapter 3 for showing that CA ≡ CS.3.2.9. Apply Theorem 4. (2b) is a consequence of (3a) and (3b). Theorem 4. EXACTLYn is true in SA . If T is a completion of PCAI. we PCA is τ reducible to Zgm.
To prove (3b). then PCAIf is has a prime model. we would have to prove analogues of 3. It’s apparent that the size models of the prime models for CAIf would do nicely. then PCAIf is complete.3. THE PURE THEORY OF CLASS SIZES To prove (3a). We will not construct prime models for the extensions of PCAI.9 through 3. but tedious. The trick is to show that enough axioms about size have been incorporated in PCA to establish the entailments among MOD statements.13 for PCA and PCS. the construction could be duplicated in this simpler case. But PCA is model complete. .62 4.3. Alternatively. it is suﬃcient to demonstrate (4). This seems straightforward. (4) If f is total. (5) If f is total and congruous. so only (5) remains to be shown. because every completion of PCAI entails PCAIf for some total f .
so we may conclude 5.2. where ˙ ˙ f (x) = x if x ∈ C ∩ A The Monk mappings for PCA can serve as a guide in constructing Monk mappings for CA. given Assumption 5. clearly.2. let SB (X) = the submodel of SB whose domain is { σB [x]  x ∈ X } Then.1. so we cannot apply Monk’s Theorem directly. The existence of Monk mappings for PCA tells us that we can ﬁnd elements with the right sizes.3.1. f : C → A.3: Fact 5. A substructure of B.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. There is an isomorphic embedding. 63 . SA is not a submodel of SB . So. we want to prove 5. SA SC SB (A) ⊆ SB .1.1). Monk’s Theorem applies directly to SB (A).1. Assumption 5. Strictly speaking. DISJ∪ and REP< then allow us to ﬁnd elements with those sizes that ﬁt together in the right way. where ˙ ˙ g(σB [x]) = σA [x] for all x ∈ C ∩ A.1. and SB (C). and C is a ﬁnitely generated Theorem 5. There is an isomorphic embedding. a ﬁnitely generated submodel of SB . g : SC → SA . and SB (C). But.1. A ⊆ B . B CA.5 Completeness of CA 5.1. SB . CA.
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 nonatomic, 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 quasicongruence 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 ncongruence classes for a given n. Deﬁnition 5.2.2. If f is total and congruous, then a. If x is an ncongruence 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 ncongruence classes, x1 ∪ . . . ∪ xk then θf (x) = θf (x1 ) + · · · + θf (xk )
n θf (x) ρf (x) δf (x) for for for for Qn . the ﬁrst f (n) ncongruence classes are each one atom larger that the remaining (n − f (n)) ncongruence classes. PRIME MODELS FOR CA c. The desired models of CAI may now be deﬁned: Deﬁnition 5. then the standard protomodel for f is Qf .3. but δ is no longer 0 in all cases.24. all ncongruence classes are assigned sizes 1/n. as in Q. θ(x).2. Proof.2.5. δ .n BASIC. If f is total and congruous and n > 0. 1 Sum(x. Fact 5.29 by substituting: Qf. and δ(x) .5. ρ(x). Deﬁnition 5. If f is total and congruous. so that x can be represented as (C(x) ∪ ∆1 (x)) − ∆2 (x) as in 3. for total and congruous f . 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. we exhibit each such model as the union of a chain of models.2. ∆1 (x) − 0. ∆1 (x) − ∆2 (x) 67 Intuitively.27) To verify that the structure Qf is a model of CAIf . If x ∈ Q. then Qf.2.2. x d. If f is total and congruous.4.n is the submodel of Qf whose domain is Qn .2. The proof can be obtained from the proof of Theorem 3. Instead. If x is ﬁnite.2. then θf (x) = 0. then θf (x) = θf (C(x)) + θf (∆1 (x)) − θf (∆2 (x)) = θf (C(x)) + 0. z) iﬀ θf (z) = θf (x) + θf (y) Qf (cf 3. y. then Qf. ∆2 (x) = θf (C(x)) + 0.
so that Qf.6 will then fall into place. In other words. a. Then an (n. and 0 ≤ m < n. x. [2n] will include the same number of charmed 3congruence classes as [2n + 1]. But this sequence does harbor a chain of models: Fact 5.m! .6. since size relations are determined by the representations of each set. (ii) If m < i < j ≤ n. for any atom a.n does not constitute a chain of models. . . we want to show that the 2congruence classes have the same size relations in Qf. . The other elements of Qf. then all of the 6congruence classes are common. so Qf. . 4}. then Qf. then all of the 3congruence classes are charmed except for [6n + 4] and [6n + 5]. for any f .m)partition of x in A is a sequence x1 . f (6) ∈ {0. to establish this by example rather than by formal proof. then [6n] and [6n + 1] are the only charmed 3congruence classes. . If 0 < i < j ≤ n.n! ⊆ Qf. then Sum(xi . The xi are approximately the same size: (i) If 0 < i < j ≤ m.7. so that Qf. In any case.3 is not an extension of Qf. For example Qf. as C(x). xi is a common nfactor of x. If f (6) = 0. If n > m. ∪ xn b. and ∆2 (x). however that f (2) = 1.2. COMPLETENESS OF CA The following notion is helpful in understanding our constructions. Suppose A BASIC. then xi ∼ xj .2 .2 [2n] ∼ [2n + 1] Since f is congruous. 3. (iii) If 0 < i ≤ m < j ≤ n. Suppose that f (2) = 0. and easier.2. 2.2 [2n] is one atom larger than [2n + 1] Here. 5}. for i > m. . In any case. f (6) ∈ {1. since f is congruous. 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 f (6) = 2.6 as they do in Qf. If f (6) = 4.6 [2n] is one atom larger than [2n + 1] So it goes in general. so Qf. and c. ˙ Deﬁnition 5. xj ).68 5. Proof. Letting n = 2 and m = 3. then xi ∼ xj . It will be clearer. then xi ∩ xj = ∅. where a.6 [2n] ∼ [2n + 1] Suppose. x ∈ A. xn . xi is called a charmed nfactor of x. The sequence Qf. If 0 < i ≤ m.2 . [2n] contains exactly one more charmed congruence class than [2n + 1]. ∆1 (x). x = x1 ∪ . .
so it is preserved under unions of chains. the basis of A. we must ﬁnd a smallest submodel. (b). and (c). and x13 whose union is x1 and another three disjoint sets. call it x0 . x1 and x2 . The existence of such sets is assured because A ADIV3 . a.8. x22 .f (n) .f (n) is an existential sentence.9. d. Clearly. x12 ∪ x22 . to include in B. We can aim for this by placing in B three disjoint sets.2. A. x12 .n MODn. PRIME MODELS FOR CA Fact 5. must be included in B. Immediate from (a). then a. Qf b.5. If f is total and congruous. BASIC ADIVk . for k > 0 MODn. also guarantees that x0 is divisible by 3: the three unions x11 ∪ x21 . Qf c. for n > 0 CAIf 5. ˙ B must also satisfy ADIV3 . Each ncongruence class can be partitioned into k nkcongruence classes.2. b.2. Whether these are exactly the same size or diﬀer by an atom depends on whether A MOD2.5).2 Shells of models To embed the model Qf into an arbitrary model.2. c. by deﬁnition.2.1 . then Qf = n>0 69 Qf. and x13 ∪ x23 will be roughly the same size and will exhaust x0 . (See Appendix. .n! Fact 5. if the basis of A is in B and B CAI. B. x11 . Qf d. If f is total and congruous. and x23 whose union is x2 . Qf Proof. Insuring that x1 and x2 are each divisible by 3. Fact B.0 or A MOD2. x0 . Qf. ˙ must contain two disjoint sets of roughly the same size whose union is then B ˙ the basis of A.7 allows us to regard Qf as the union of a chain of models: Fact 5. since the symbol I must refer to the same set in the ˙ submodel as it does in the model. Pick such a pair.f (n) . BASIC is a universalexistential theory. Since MODn. x21 . of A which satisﬁes CAI. it is preserved under extensions. the exact size relations will be determined by which MOD principles are satisﬁed in A. But. of CAIf . Again.
to be a prime model. must be countable. where f is a total. a. or depth. Given A = Af remainder function: a. dividing each set introduced at stage n into n + 1 roughly equal subsets at stage n + 1. If λ = n1 . L(λ) is k.10. . for 1 ≤ i ≤ L(κ). In the next section. we deﬁne the shell of A as the submodel of A generated by the collection of node sets and node atoms. . Deﬁnition 5. . First. nk . We leave the solution of this problem to the formal construction. . COMPLETENESS OF CA We can continue this process indeﬁnitely. λ extends κ iﬀ L(λ) > L(κ) and. ni < i. The formal proof proceeds as follows: First. (The variables λ and κ range over nodes. d. aλ . (ii) If 1 ≤ i ≤ k.11. then (i) The length. the tree: Deﬁnition 5. (iii) λ • m = n1 . nk . bearing sets. of λ. which is what we are looking for. . the smaller the set it bears and the greater the number of successors among which this set will be partitioned. λ(i) = 0. nk .2. . . for L(κ) < i ≤ L(λ).70 5. A node is a ﬁnite sequence n1 . This great tree of sets will not form a boolean algebra. Third. m c. We cannot correct for this problem ˙ by including in B all atoms of A: A may have uncountably many atoms while B. we deﬁne a tree.) b. . This will produce an inﬁnite tree. 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. assigns a node set Aλ and a node atom. Second. we show that the shell of A is isomorphic to Qf . then λ(i) = ni . to each node. but this would still not be an atomic boolean algebra. We shall now assign a set to each node by repeatedly partitioning the basis of A. i. λ. . At the same time we shall assign an atom to each node. . we present the construction which. .2.e. Nodes constitute the vertices of an inﬁnite tree in which 0 is the root and λ dominates κ iﬀ κ extends λ. Let A 0 CAIf . and let a 0 . The number of immediate descendants of a node grows as the depth of the node increases. . given a model A of CAIf . The deeper a node is in this tree. where k > 0 and for all i ≤ k. λ 0extends κ iﬀ λ extends κ and. be the basis of A. nor will it be closed under ﬁnite unions. A boolean algebra could be obtained by including both the node sets and their ﬁnite unions. congruous be any atom of A. λ(i) = κ(i).
A has f (n) charmed nfactors and n − f (n) common nfactors. Each of the common n factors has k charmed mfactors and each of the charmed nfactors has k + 1 charmed mfactors. PRIME MODELS FOR CA b. x has an (m. and k= Then a. All common nfactors are the same size and satisfy the same Modm. b. Every charmed nfactor of A has an (m. k + 1) partition of Aλ if Aλ is charmed. k)partition.12 guarantees that such partitions exist.12. . If 0 < i ≤ m.2. m > 0.k predicate.k (x) says that x has an (m. n > 0. So.2. . choose Aλ•0 so that it contains aλ . a. Suppose A CAIf . Let m = L(λ) and let k= f ((m + 1)!) − f (m!) m! 71 Since f is congruous. The sets Aλ will be referred to as node sets and the atoms aλ as node atoms. Fact 5.7e). suppose that each common nfactor has k charmed mfactors and m − k common mfactors. the charmed mfactors of the nfactors are the charmed nm factors of A and the common mfactors of the nfactors are the common nm factors of A. there are (n − f (n))k + f (n)(k + 1) =nk + f (n) . Let Aλ•0 . . ˙ c. let aλ•i be any atomic subset of Aλ•i . k)partition. k)partition iﬀ A Modm. Aλ•m be an (m + 1. Let aλ•0 = aλ . c. Suppose that Aλ and aλ have been chosen. k + 1)partition.k (x) f (nm) − f (n) n ˙ b.2. k) partition of Aλ if Aλ is common or an (m + 1. This is always possible because Aλ contains aλ . Modm.5. Every common nfactor of A has an (m. By (a).3. In all. k is an integer (see Deﬁnition 3. Partitioning each of the nfactors into m subsets of roughly the same size yields an nmpartition of A. . Fact 5. In either case. Proof.
then κ = λ. then Aλ•i ∩ Aλ•j = ∅. since every charmed nfactor is one atom larger than each common nfactor. i. and Aλ 0 is the basis of A. c. h. For any n. there are f (nm) charmed nmfactors of A.: If n = 1. j. f (n!))partition of the basis of A. Each node set is the disjoint union of all of its descendants at any given depth. Then (k ) also holds for n + 1: each node set of depth n has n + 1 immediate descendants.13 should clarify this constuction. Assume that (k ) holds for n. aλ = aκ iﬀ one of λ and κ 0extends the other. We demonstrate (k) by induction on n. hence. Immediate from (b).13. Every node set contains inﬁnitely many node atoms. Proof. COMPLETENESS OF CA charmed mfactors among the nfactors of A. So (k ) holds because any set is a (1.2. Each node set is the disjoint union of its immediate descendants. Aλ ⊂ Aκ iﬀ κ extends λ or κ = λ. g.0)partition of itself. node sets) of depth n. so. by (a) again. There are n! nodes (and. All of them can be established by induction on the depth of nodes.72 5. a. e. Fact 5. f. Any two node sets of the same depth are disjoint. k. then n! = 1. The facts listed in 5. the node sets of depth n form an (n!. aλ ⊂ Aκ iﬀ λ extends κ. b. d.2. . f (n!) = 0. So f (nm) = nk + f (n) k = (f (nm) − f (n))/n and c. If i = j. But. there are (n!)(n + 1) = (n + 1)! node sets of depth n + 1. If Aλ = Aκ .
f (n!) of the nfactors are charmed and n! − f (n!) are common. a total. f ((n + 1)!))partition of the basis. Suppose f is total and congruous. aλ . a set of node sets and nodeatoms produced by the construction above. the shell of A generated from {Aλ . A. each charmed nfactor has an (n + 1. then .14. aλ } . k + 1)partition and each common nfactor has an (n + 1. of CAIf .2. A. ˆ d. the n + 1 factors of the nfactors of the basis form an ((n + 1)!. that A CAIf and Aλ and aλ are the node sets and nodeatoms of A produced by construction 5. For the remainder of this chapter. and nodeatoms.12. B. Then. aλ }. of A which is isomorphic to Qf .2. By Fact 5. there are (k + 1)f (n!) charmed n + 1 factors from the charmed nfactors and k(n! − f (n!)) charmed n + 1 factors from the common nfactors In all. c. from a model. A.2. aλ }. (k ) holds for n + 1. Given a collection of node sets.12. 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. k)partition. the shell of A is the submodel of A generated by {Aλ . we can now construct a submodel. That is to say. a model of CAIf . we will regard as ﬁxed: a. Deﬁnition 5. {Aλ . congruous remainder function b. Aλ . the number of charmed nfactors 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.5. PRIME MODELS FOR CA 73 Furthermore. f .1.
of Q has a unique representation as (C(x) ∪ ∆1 (x)) − ∆2 (x) where C(x) is a quasicongruence class.24 that each member.2. (cf.15 through 3.2. 3. Fact 5. . x is a quasinodal 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).2. a.2. Proof. we need a sharper characterization of the elements ˆ of A. ∆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. ˆ (⇐)A must contain ﬁnite unions of node sets as well as Aﬁnite sets. (⇒)A is generated from node sets and node atoms via the boolean operations.74 5.18 obtain for Anearness. ˙ ˆ a. c. b.143.16. ﬁnite unions of node sets correspond to the quasicongruence classes. x ∈ A iﬀ x is Anear some quasinodal set of A. A is an atomic boolean algebra whose atoms are the nodeatoms aλ .2. each of which preserves Anearness to quasinodal sets. x. If x ∈ A and y ∈ A. then x is Anear y iﬀ both x − y and y − x are Aﬁnite sets. If x ∈ A.2. ˙ ˆ c.15. ˆ b. COMPLETENESS OF CA ˆ To show that A Qf . so it must also contain sets obtained by adding or removing Aﬁnite sets from quasinodal sets. ˆ a.2. both of which are Aﬁnite sets. Recall from 3. ﬁnite sets of nodeatoms correspond to the ﬁnite subsets in Q. then x has a unique representation as (C(x) ∪ ∆1 (x)) − ∆2 (x) where C(x) is a quasinodal set disjoint from ∆1 (x) and including ∆2 (x). Analogues of 3. Deﬁnition 5. x is an Aﬁnite set iﬀ it is a ﬁnite set of node atoms.18) ˙ ˆ Still following in the footsteps of chapter 3. we can characterize A as the collection of sets Anear quasinodal sets.
we are selecting node atoms so that every singleton in Q is the node atom for some node. If L(λ) = k. Examples.2 Q 0. The proof parallels that of Theorem 3.3 The embeddings To embed Qf into A. Let C(x) be the quasinodal set which is Anear x (see 3.2. . This latter condition guarantees that the shell of Qf will be Qf itself.0.2.2.1. we are stipulating which (n.2. qλ = {ι(λ)} Fact 5.2 = [n] = [2n] = [2n + 1] = [6n + 1] = [6n + 4] = [6n + 5] Deﬁnition 5.2. . 75 c. Deﬁnition 5. Then a. we are performing the construction 5. b. In eﬀect. 5. a. PRIME MODELS FOR CA b. m)partitions to use at each level. Q0 Q 0. but with two differences: ﬁrst.23). and.18.17. let ∆2 (x) = C(x) − x.5. Suppose λ is a node. then Qλ = [k!n + m] where k m= i=1 λ(i)(i − 1)! b. Let ∆1 (x) = x − C(x). . a. The depth of Qλ is L(λ). then k ι(λ) = i=1 (i − 1)!ni . If λ = n1 .2. . we ﬁrst describe Q in terms of node sets and node atoms. ι(λ) = the least n ∈ Qλ .1.0 Q 0.2. second. nk .21 exactly.12 on Qf .2.1 Q 0.0 Q 0.19.
2. there’s a λ such that n = ι(κ) iﬀ κ = λ or κ 0extends λ. e. Then there is some n such that x = x1 ∪ · · · ∪ xk . shallowest) node for which ι(λ) = n. b. ﬁnally.2. n.16. will be the highest node with a certain value just in case λ is not the leftmost immediate descendant of its parent. There is at most one such y. d.2. suppose ﬁrst that x ∈ QC. there is a unique y ∈ A such that ∀λ(qλ ⊆ x ↔ aλ ⊆ y) Proof. ι(λ) = ι(κ) iﬀ λ = κ or one 0extends the other. These are the only nodes below λ with the value k. λ extends λι(λ) c. Notice.20.13 hold for the sets Qλ and qλ . the ι(λ) take on all and only values less than n!. For every n.) Each of the points listed in Fact 5.2. λn = the shortest λ such that ι(λ) = n. λι(λn ) = λn d.2.76 5. To show that there is such a y. deﬁne the embedding of Qf into A: ˙ ˆ Fact 5.2. λι(λ) = λ iﬀ λ =< 0 > or λ(L(λ)) = 0 (ie a node. For every n. then all nodes along the leftmost branch descending from λ also have the value k. c. we can associate with each natural number a shortest (i. ι(λn ) = n b. If ι(λ) = k. c. that 5. Fact 5.e. Deﬁnition 5. especially. We may. there are inﬁnitely many λ such that n = ι(λ).13k holds.22.21. Fact 5.23. there will be inﬁnitely many nodes λ for which ι(λ) = n. If x ∈ Q. λ. a.2. by Fact 5. That is to say. the Qλ can be regarded as node sets and the qλ as node atoms for any model Qf . Though for a given natural number n. Every natural number is the value of all and only those nodes along the leftmost branch descending from some node. For every n. there’s a λ such that n = ι(λ). a. COMPLETENESS OF CA b. At each depth.
Immediate from the proof of 5. 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. then g(x) = (g(C(x)) ∪ g(∆1 (x))) − g(∆2 (x)) b. ˙ ˆ c. PRIME MODELS FOR CA where each xi is an (n!)congruence class and hence a node set in Qf . a. If x ∈ Q.2. So x = Qλ1 ∪ · · · ∪ Qλk Now. let y = Aλ1 ∪ · · · ∪ Aλk ˙ ˆ So y ∈ A and: qλ ⊂ x iﬀ qλ ⊂ Qλi . so x = x . and let y = (y ∪ { aλ  qλ ⊆ ∆1 (x) }) − { aλ  qλ ⊆ ∆2 (x) } ˙ Deﬁnition 5. Then ∀λ(qλ ⊆ x ↔ aλ ⊆ y ↔ qλ ⊆ x ) But every integer is ι(λ) for some λ. Proof.2.25. Let y be the element of A corresponding to x . .2. for some i iﬀ λ extends λi iﬀ aλ ⊂ Aλi iﬀ aλ ⊂ y. as described above. If x is not a quasicongruence class. Suppose g(x) = y = g(x ).23.2.5.13h 77 ˙ ˆ where x is a quasicongruence class.. then x = (x ∪ ∆1 (x)) − ∆2 (x) by 5. a. b. g maps Q onto A.24.2. g is oneone.
be the least k such that x and y are both unions of node sets of depth k. of Q into A is an isomorphism ˆ of Qf onto A. n) be the number of charmed node sets of level n contained in C(z). g. Proof.26.78 c.24) (by 5. below. The nodal embedding. Qf x ∼ y iﬀ θf (x) = θf (y) iﬀ nc(x. INDISC∼ ˆ since A ⊆ A . Let nc(z. and proper subsets. c. n) − (g(∆1 (x)) − g(∆2 (x))) = nc(g(y). n) − (∆1 (y) − ∆2 (y)) iﬀ nc(g(x).2.2. n) − (∆1 (x) − ∆2 (x)) = nc(y. relative complements. I. As in (a). Obvious.2. And let n. intersections. 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. 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. g can be shown to preserve ∅. unions. COMPLETENESS OF CA Theorem 5. 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. 5. a.16) b.
and their completions. the prime model test (Appendix. If f is a total. then CAIf has a prime model.27.3. .6. Qf Sum(x. So CAIf is complete.i for exactly one i. 0 ≤ i < n. Hence T MODn.3 Summary We can now draw our ﬁnal conclusions about CA. each model of CAIf has a submodel isomorphic to Qf .2. then T is inconsistent. CS.i for at least one one i.j where 0 ≤ i = j < n.i → ¬MODn. T would be inconsistent. ∧ ¬MODn.0 ∧ .5.3. CAIf is consistent because f is congruous. congruous remainder function.11).2.i and T MODn. 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).1. since T CAI and CAI DIVn . If f is total and congruous. then T ≡ CAIf . y. T MODn. Again.n−1 ). Theorem 5. Proof. Since CAIf is model complete. by 3. Corollary 5. g(y). 5. by 5.j (see Lemma 3. But if T (¬MODn.i or T ¬MODn.i for each such i.3. . SUMMARY e. g(y). .3. by 5. for CA MODn. T MODn. But suppose T MODn. Fact C.27. Proof. It T is a completion of CAI. then CAIf is consistent and complete.3) applies. g(z)) since Qf 79 DEF+ by (b) and (c) since A DEF+ ˆ since A ⊆ A So.3. For each n > 0.1. 0 ≤ i < n: Since T is complete.12c. g(z)) Sum(g(x). and has a prime model. for some congruous f .2. and we may conclude: Corollary 5.
CS. Follows from 5.6. Each such function is congruous. By Lindenbaum’s lemma.3. such that T = CA.3.2 and 3. There are 2ω remainder functions whose domain is the set of prime numbers. a. There is no sentence. so each corresponds to a consistent extension of CSI. given that CS CA. But An is a ﬁnite model. b. CSIf is decidable iﬀ f is decidable.5. (⇐)To calculate f (n). a. Every completion of CA is consistent with CS. CS. For total f . Corollary 5. .3. b.13. b. CAIf T . Theorem 5. COMPLETENESS OF CA MODn. CSI has 2ω completions. each of these extensions has a consistent and complete extension. Then T CAIf and.2b. Proof. To determine whether CS φ. Corollary 5. Proof. EXACTLYn φ iﬀ An φ.3. b.m . determine which MODn. For n > 0. CS is decidable. φ is consistent and T has only inﬁnite models. Follows from (a) and 3. since CAIf is a. EXACTLYn is decidable. But CAIf is complete.3. CA ≡ CS Proof. a.3.3.m sentence is in CSIf . a. φ. so it is decidable.80 So. 5. Theorem 5. (⇒)If f is decidable. let f (n) = m iﬀ T complete. then CAIf is recursively enumerable. a. b. alternate between generating theorems of CA and testing whether An ¬φ. b.3.4.
SUMMARY 81 Proof. But by (3). deﬁned in 3. in which size relations are determined in accordance with the size function. Theorem 5. θ. by compactness.2. If φ is true only in inﬁnite models of CA.3. then ¬φ is true in all ﬁnite models of CA. so A .26.7. Let K = { n  every prime factor of n is a member of J }. A A A BASIC ADIVj for all j ∈ J ADIVk φ. φ is inconsistent. But CA ≡ CS. So CA φ and. Suppose CS φ. We claim the following without proof: (1) (2) (3) By (1) and (2). A φ CS. (BASIC ∪ T ) φ for some ﬁnite set of ADIVn principles. Proof.5. Let A be a model with domain k∈K Qk . so CA. Hence (BASIC ∪ T ). so ¬φ ∈ CS.3. CS is not ﬁnitely axiomatizable. T = { ADIVn  n ∈ J }. A CS.
82 5. COMPLETENESS OF CA .
An ordering of P(N) that satisﬁes CS will not necessarily appear reasonable.1. in section 1. If x outpaces y.2. But we should also 83 .1 The outpacing principle Throughout this chapter. Deﬁnition 6. OUTPACING. Section 2 establishes that OUTPACING can be satisﬁed jointly with any consistent extension of CS in a model whose domain is P(N). then x > y. 6.13). x outpaces y just in case the restriction of x to any suﬃciently large initial segment of N is larger than the corresponding restriction of y. We extrapolate from well understood ﬁnite cases to puzzling inﬁnite cases. some such orderings say that there are fewer even numbers than prime numbers (see below 6. To rule out such anomalies. 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. In this chapter we will show that CS has inﬁnite standard models over P(N). we introduce a principle. x and y will range over P(N).1. We employ this notion to state a suﬃcient condition for one set of natural numbers to be larger than another: OUTPACING. For example. The general motivation behind this principle should be familiar. OUTPACING does not ﬁx the size relations over P(N).6 Sets of Natural Numbers CS has standard ﬁnite models since it consists of sentences true in all such models. It has inﬁnite models Qn and Qf . So CS . OUTPACING mentions the natural ordering of N and applies only to subsets of N.
thus: (1) x > y iﬀ x outpaces y. for outpacing is not a quasilinear ordering.1. Others are too weak to be helpful. A is an outpacing model iﬀ ˙ A = P(N) A BASIC.1. then x > y. This revised principle conﬂicts with CS. We shall ﬁnesse this problem by regarding OUTPACING as the (very large) set of sentences { b < a  a outpaces b } Fact 6. Since OUTPACING involves the < relation over N. [2n] has at least (k − 1)/2 members less than or equal to k for any given k. .1. Every outpacing model satisﬁes the following a. 2n + 1} of N contains n evens and n odds. [3n] > [4n] > [5n] > . If k > 4. then [kn] > n2 Proof. since [2n] outpaces [2n + 2] while [2n + 1] does not. a. given any ﬁnite subset z of N. and A OUTPACING There is a slight diﬃculty in saying that an interpretation of LC< satisﬁes OUTPACING . So. there are many statements that assert of inﬁnite cases what is true of ﬁnite cases. then (k − 1)/2 > k/3 + 1 . . for example. If k > 0. determine whether they are consistent. and see how far they go. [2n] > [3n] b. . Though (2) is true.0. Deﬁnition 6. Some of these conﬂict with one another. It is doubtful whether there is any mechanical way to decide which of these statements are true. For example. x restricted to z is larger than y restricted to z. again. SETS OF NATURAL NUMBERS emphasize. it cannot be expressed in LC< . mechanical way without risking contradiction. that this extrapolation cannot be done in any straightforward. [3n] has at most k/3 + 1 such members. . But the two are discernible under outpacing. c. We cannot. 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. The best we can do is propose plausible theories.84 6. . its antecedent is only satisﬁed when y ⊂ x. neither [2n] nor [2n + 1] outpaces the other since each initial segment {0. . strengthen the conditional to a biconditional.
Let k be the least odd number not in y. it is suﬃcient to show that ¬ [2n] < [2n + 1]. Similar to that of (1). Every outpacing model satisﬁes the following a. Theorem 6. we show that both of these possibilities can be realized in standard models over P(N). b. xi < yi < xi+1 . by REP< . y is an alternating pair iﬀ x and y are inﬁnite and for all i > 0. Proof. z = [2n + 2] and apply (*). Here. (y ⊂ x ∧ z ≤ y ∧ Atom(z ) ∧ z ⊂ x ∧ x = z ∪ z ) → y ∼ z Let x = [2n].1. If [2n] > [2n + 1]. By TRICH. But any proper subset of [2n + 1] is outpaced by. N[m] will have more members in [kn] than in n2 . x. For m > k(k + 1).1.1. Then [2n] leads y at k + 1 and y never catches up. y is an alternating pair. then [2n + 1] ∼ [2n + 2]. let Ai = [kn + i] for each i < k. If x.2 correspond to the possibilities that N may be odd or even: if [2n] ∼ [2n + 1].5. c. both [kn] and n2 have exactly k members less than m. a.2 applies here since the only facts about [2n] and [2n + 1] used hold by virtue of these sets forming an alternating pair. then in any outpacing model: x ∼ y ∨ x > y ∼ (x − x1 ) Proof. Note that [2n] = [2n + 2] ∪ {0} and that BASIC (*) (*).6.1. Aj is an alternating pair.1.1. If it were.2 to similar cases. The two alternatives left open in 6. if [2n + 1] ∼ [2n + 2]. Fact 6. Deﬁnition 6.2. If 0 ≤ i < j < k. THE OUTPACING PRINCIPLE b.2. then N is even.4. and hence smaller than [2n]. In section 6. then Ai . [2n + 1] ≥ [2n + 2] c. y = [2n + 1]. 85 c. If m = k 2 . we generalize 6. Fact 6.1. including other congruence classes. . [2n] ≥ [2n + 1] b. Similar to (1). The argument for 6. then. then N is odd. Then: a.1.3. For a given k > 0. there is a y such that y ∼ [2n] and y ⊂ [2n + 1]. So there is no y such that [2n] ∼ y ⊂ [2n + 1].
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. (See example below). . So Ai ∼ Ap . So either A0 > Ai or A0 ∼ Ai by 6.1. then Ai ∼ Aj .2 Models of CS and Outpacing In this section. let p = k. 0 < p ≤ k such that (i) If i < j < p. An is the standard ﬁnite interpretation of LN over P(N[n] ) in which: An (a) = a ∩ N[n] = a[n] for each a ⊆ N. then Ai ∼ Ap . Example.4 since A0 . There is a p. b. Hence Ai ∼ A0 − {0} ∼ Ap . (i) follows by TRANS∼ . Ai (n) = kn + i. though the application here demands some important diﬀerences. 1. (iii) If p ≤ i < k. Ai is an alternating pair. A0 . (i) For 0 ≤ i < p. 6. the language of subsets of N. is the ﬁrst order language which results from adding to LC< . (ii) Immediate from 6.1. 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. But A0 ≤ Ai by the selection of p.2. So A0 > Ai . LN .86 b. we construct models of CS over P(N) that satisfy OUTPACING. 2. 3. Deﬁnition 6. so A0 ∼ Ai . Ai (n + 1) = kn + i + k and i < j < k + i.1. If A0 > Ai for some i. Ap is an alternating pair and A0 > Ap . so Ai = [4n + i] for i = 0. SETS OF NATURAL NUMBERS (ii) If p = k. then A0 ∼ Ap ∪ {0}. a name for each subset of N.4. as individual constants. Let k = 4. a. let p be the least such i.Aj (n) = kn + j. b. otherwise. Proof. a. (iii) A0 > Ap ≥ Ai if i ≥ p.
then a. c. Def. d. MODELS OF CS AND OUTPACING 87 Deﬁnition 6.4. p.3. then a ∩ b ∈ F c.18.108] b. (cf. If Y ⊆ X and Y is inﬁnite.3. and similarly for other predicates. then AF is the interpretation of LC< in which a. Boolean symbols receive the usual interpretation.5. 18. A˙F = P(N). If τ is a term of LN . A nonprincipal ultraﬁlter contains no ﬁnite sets. b. Prop. F . a. AF a < b iﬀ k  a[k] < b[k] ∈ F . and (iii) If a ∈ F and b ∈ F . An ultraﬁlter. Our main result is that if F is nonprincipal. Ch. b. lemma 1.6.15. then either Y ∈ F or (X − Y ) ∈ F . If F is a nonprincipal ultraﬁlter over N. Deﬁnition 6. F has the ﬁnite intersection property iﬀ the intersection of any ﬁnite subset of F is nonempty.2. then b ∈ F . If F is an ultraﬁlter over N.2. If F ⊆ P(X) and F has the ﬁnite intersection property.6. p318] If X is a set and F ⊆ P(X). over X is principal iﬀ there is some x ∈ F such that F = { a ⊆ X  x ∈ a } Fact 6. Theorem 6. and (iii) if Y ⊆ X. c.2. [Monk. F is an ultraﬁlter over X iﬀ (i) F is a ﬁlter over X (ii) X ∈ F . then there is an ultraﬁlter over X which includes F .2.2. p. A nonprincipal ultraﬁlter over X contains all coﬁnite subsets of X.2. F is a ﬁlter over X iﬀ (i) F = ∅ (ii) If a ∈ F and a ⊆ b. then Ak (τ ) = AF (τ ) ∩ Ak = AF (τ ) [k] .319] d. [Bell. then AF is an outpacing model. then a. 18. then there is a nonprincipal ultraﬁlter over X which contains Y . [Monk.
4b. τ = a for some a ⊂ N. for every n.3b. if n  a[n] ⊂ b[n] ∈ F . Immediate from 6. then there is a k ∈ b but not ∈ a. BASIC. So if n > k. So. { n  AF φ } is coﬁnite and. b. then c. otherwise a[n] would not be included in b[n] for any n greater than k. SETS OF NATURAL NUMBERS b. 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. then AF φ iﬀ a = b iﬀ a[n] = b[n] for all n. a[n] ⊂ b[n] . .88 6. then AF φ iﬀ a ⊂ b If a ⊂ b. a < b iﬀ { k  Ak (a < b) } ∈ F . (i) If τ is a constant.1b. d. AF e. then it is inﬁnite. AF f.2. If φ is a universal formula of LN and Ak AF φ.2. so a ⊂ b.2. (ii) If φ = a = b . (ii) If τ = τ1 ∪ τ2 . by 4. So Ak (τ ) = a[n] by 6. Hence. But. AF Proof. then AF φ iﬀ { k  Ak φ} ∈ F φ for every k. a. ADIVn . So a ⊆ b. By induction on the structure of φ: (i) If φ = a ⊂ b . By induction on the structure of τ : REP< . there cannot be a k in a but not in b. Conversely. in F . then Ak a = b. If φ is a quantiﬁer free formula of LN . iﬀ { n  An a = b } = N iﬀ { n  An a = b } ∈ F since if An (iii) AF a = b and k > n.
for all a. for all a φ(a). these form an alternating ntuple.) f. . For 0 < i ≤ n.2. xi .5). As in 6. let xi = { xkn+i−1  k ∈ N }. . e.5. then. (a ∼ a ) = {k1 . together with ATOM and REP< .} where the ki ’s are in strictly increasing order. Furthermore. thus. ki .1. .6. a subset of b. these sets are approximately equal in size and AF ADIVn (x). Immediate from (c) and (d): BASIC is equivalent to a set of universal sentences. for all k φ(a). Hence: AF a ∼ a since they are the same size over some set which contains K and is. then a [k] ∼ a[k] . for all a ∀xφ(x) by (b) (a < b).1. (AF ATOM because it contains all singletons of natural numbers. 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. MODELS OF CS AND OUTPACING (iv) If φ is nonatomic. for all k φ(a). x an inﬁnite subset of N. . . where n = a[ki+1 ]  − ai  a = ai Then a ⊂ b since each ai draws its new members from b. in the manner of the congruence classes modulo n (see Theorem 6. for all k. partition x. if F . The n sets. for which AF k  a[k] < b[k] so. Claim: If k ∈ K. . . . Suppose AF as follows: Let K = Ak Ak Ak AF AF ∀xφ(x). Let a0 = ∅ ai+1 = ai ∪ the n greatest members of b[ki+1 ] not in ai . Construct a . K ∈ F . Suppose then So So So d.
Otherwise. a. the only nonuniversal axioms of CAI could be veriﬁed in the constructed model more or less directly. these special means were available. For a given F . for if we count the even numbers and the odd numbers up to some even number. Unfortunately. so AF [2n + 1] ∼ [2n]. These functions are then gathered into equivalence classes (by virtue of agreeing almost everywhere. after all. has the handy property that it satisﬁes any formula which is satisﬁed by almost all factors. In the usual construction(see.3. we can immediately conclude that Ai /F satisﬁes CS. and CSI The proof of 6. So. so it may or may not be in F . we still would have had to resort to special means to show that the nonuniversal formulas were likewise satisﬁed. pp.6 is modeled on the usual ultraproduct construction. we would have then been able to infer that the part of that model which held our interest satisﬁed all universal formulas of CS.2. Then a[k] < b[k] iﬀ k ∈ [2n]. . If F is a nonprincipal ultraﬁlter over N.6. a model is built by ﬁrst taking the product of all factors (in this case. SETS OF NATURAL NUMBERS Corollary 6.e. Consider a familiar example: Let a = [2n + 1] and b = [2n]. CAI. indeed. there will be many cases where AF a < b even though b does not outpace a. AF b.5f. then a.90 6. This will happen whenever k  a[k] < b[k] ∈ F but is not coﬁnite. there will always be the same number of evens and odds. This is handy because. Fortunately. which we’ll call Ai /F . b. but only a submodel.2. given that each of the Ak satisﬁes CS. there will always be one more even and if we count up to some odd number.3b. i. here) to elements of the factors. for example. which results in a domain whose elements are functions from the index set (N. but equivalence classes of functions from N to ﬁnite subsets of N. and certainly any formula which is satisﬁed in all the factors. had we constructed the reduced ultraproduct. the Ak ). AF [2n + 1] < [2n]. on some member of the ﬁlter) and the reduced ultraproduct is deﬁned by interpreting the language over these equivalence classes. and this mapping would have allowed us to identify a model over P(N) as a submodel of Ai /F . AF Proof. There is. Immediate from (a) and 5. [Bell.2.5e and 6.2. [2n] is neither ﬁnite nor coﬁnite. [2n + 1] ∈ F . Ai /F is not the model we wanted: its elements are not subsets of N. 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 . If [2n] ∈ F . The model so constructed. 8792]). but is not quite the same. Immediate from 6. a natural mapping from subsets of N to such elements.
That is.2.10. Fact 6.2. So (b) follows from DISJ∪ . so / n ∈ x+ ).2. yields x or x+ . there is a set of decisions that must be made in constructing an outpacing model. This pattern repeats during successive runs of x. Theorem 6. x2 is an alternating pair. x1. MODELS OF CS AND OUTPACING 91 The construction of a model AF from any nonprincipal ultraﬁlter.8. Furthermore.6. Nevertheless.2. since x(1) ∈ / x+ because ¬(x1 − x1 (1) ⊂ x). no two consecutive numbers are in x. personalized contribution to the model AF . b. A Proof. (So [2n] has only 1membered runs. x. But the disjoint union of x ∩ x+ with x1 or x2 . We know from (a) that x2 ∼ x1 or x2 ∼ x1 − x1 (1). but this is not always the case. A c. x1 = (x − x+ ). losing it only at the least n for which n ∈ x (since (n − 1) ∈ x. and A = AF .2. We need only prove (*): (*) x[n]  > x+ [n] x ∼ x+ or A x > x+ ∼ x − {x1 } x > x+ iﬀ x ∈ F . x1 (n) is the ﬁrst element in the nth run of x and x2 (n) is the ﬁrst element after the nth run of x. then x+ = { i + 1  i ∈ x }. can sometimes be an alternating pair. We will ﬁrst show that this qualiﬁcation is not needed. Let a run of x be a maximal consecutive subset of x. each decision is made for a model AF by the presence or absence of a particular set in F . F . each decision may go either way. F is a nonprincipal ultraﬁlter. . x+ is an alternating pair iﬀ x is inﬁnite and there is no n ∈ x such that (n + 1) ∈ x. If F1 and F2 are distinct nonprincipal ultraﬁlters over N. respectively. then a. x+ . Throughout the ﬁrst run of x.7. A pair of sets.  iﬀ n ∈ x [n] Informally: x[n] ﬁrst becomes greater than x+ for n = x(1). If x ⊆ N. suggests that there are many outpacing models unless diﬀerent ultraﬁlters can yield the same model. we will show that the presence of a set in an ultraﬁlter makes a direct. x+ are like alternating pairs in the following way: Lemma 6. pairs x. Putting this in another way.9. b. x2 = (x+ − x). x. x maintains its lead. then AF1 = AF2 . a. (See below. Proof. c.) To show this. though the decisions are not independent of each other. while N−[10n + 1] has only 9membered runs.) So. Deﬁnition 6. If x is inﬁnite.
AF MODk. x. Theorem 6. By 6.7: Proof. Every inﬁnite completion of CS has an outpacing model. such that AF (x > y). Then (*) holds for each k: (*) Gk = n  An MODk.2. such that x ∈ F1 and x ∈ F2 .2. Theorem 6.f (k) . of G is inﬁnite. where J is some ﬁnite subset of N+ . AF1 x > x+ and AF2 x ∼ x+ / Theorem 6. so AF CSIf . By 6. we can show that either of the alternatives in 6.10b can obtained for any alternating pair. such that AF (x ∼ y). so is y. If neither x nor y outpaces the other. . x − x(1) . SETS OF NATURAL NUMBERS We can now prove our main result.92 6. Proof. For example. (See 3.2. Since the intersection of any ﬁnite subset of G is inﬁnite. a. then there is an ultraﬁlter F .2. let F be a nonprincipal ultraﬁlter which contains J. y is an alternating pair.7 allows us to improve upon some previous results. so the restriction of f to the ﬁnite domain J has inﬁnitely many solutions (see 3. then there is an ultraﬁlter F .8a). If x.3d. J is inﬁnite. But then AF x > y.6. The intersection of any ﬁnite subset.11. Without loss of generality. a. by (a) there is an F such that AF y ∼ (x − x(1). there is a nonprincipal ultraﬁlter F such that G ⊆ F . By (*).10c. Theorem 6. Since neither x nor y outpaces the other. If x.12.2. Let J = k  x[k]  = y [k]  . But f is congruous. let Gk = [kn + f (k) + 1] for each k > 0. If H is a ﬁnite subset of G. b.) Given such an f . b. we can suppose there is a set.2.3. So. Recall that every inﬁnite completion of CS is equivalent to CSIf for some total and congruous remainder function. then H = { Gk  k ∈ J }. y is an alternating pair. H.2. f . So H = { n + 1  k ∈ J → n ≡ f (k) mod k }.2. Proof. Then AF x ∼ y.f (k) Let G = { Gk  k > 0 }.
Theorem 6. Lemma 6. To establish that A = AF . MODELS OF CS AND OUTPACING 93 On the basis of 6. then A = B I have not been able to prove (1) or (2).5 for the relative sizes of the kcongruence classes are obtainable in outpacing models. This would be true only if ﬁxing the congruence classes determined whether x ∼ x+ or x > x+ for every x ⊆ N . Second.10c may suggest that any outpacing model. At most one of them is obtainable by our construction. This section has explored the existence and variety of outpacing models. there is no unique outpacing model which satisﬁes T . each yielding a diﬀerent outpacing model. More speciﬁcally. though it may be extraneous to show that OUTPACING is independent. If x outpaces y. Modifying 6. perhaps. Three comments are in order before we turn to the common structure of outpacing models.5]. Proof.2. There are standard models of CS over P(N) which do not satisfy OUTPACING. we could produce standard models of CS over P(N) which satisfy OUTPACING and these will not. 6. both (1) and (2) are necessary. That all such choices are not determined by a remainder theory can be seen intuitively. But even if (1) is true. 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 (1) is false.4. So (1) and (2) are open problems. is AF for FA = { x  A x > x+ }.4. even it T is an inﬁnite completion of CS. x+ but disagree elsewhere. then x > y. then clearly A = AF . we will do so. by considering x = n2 : any ﬁnite set of congruences has inﬁnitely many solutions that are squares and inﬁnitely many that are not.13. all of the possibilities listed in 6.2. there are only 2ω remainder functions while there are ω 22 nonprincipal ultraﬁlters over N[Bell. Finally. Ch. two outpacing models may agree about all pairs x. we can now extend that observation to moduli other than 2. so whether x ∈ F is an independent choice.6. in general. Also.11 we noted that there are even and odd outpacing models. Theorem 1.2. (2) If FA = FB . but it should not.2. A. it is not clear whether every outpacing model can be obtained by the construction of 6.2. First.1. satisfy OUTPACING.2. (1) If A is an outpacing model. Suppose that is a linear ordering under which N forms an ωsequence. . then FA is a nonprincipal ultraﬁlter.
. then x converges in y. if x has a density in N. that 6. In this section. If x is ﬁnite. Fact 6.2. p and q will be nearly the same size. if it exists of i=y1 [i] lim xi /ˆi ˆ y ω That is.94 Suppose. of natural numbers is the limit. Since they often want to compare inﬁnite subsets of N.3. . q1 . a. e. . for example. iﬀ x has a density in y. y) = r iﬀ ∀(δ > 0)∃n(∀i > n)(r − δ < xi ˆ < r + δ))) yi ˆ c. iﬀ x converges in N. of x[n]  as n grows. SETS OF NATURAL NUMBERS is the ordering: p1 . then ρ(x. the asymptotic density of [2n] is 1/2. is the density of of x in N.1. is the limit. we compare the ordering of P(N) given by density to the orderings given by CS and OUTPACING. x diverges in y. From now on we shall use the term ‘density’ for asymptotic density. otherwise. if there is one. But the evens are much smaller than q. y). a. b. the fraction of numbers less than or equal to i that are ˆ members of x. pk . x diverges iﬀ x diverges in N. . . ρ(x) = 0.3. . . If x ⊆ y = ∅. ρ(x). The density of x. For example. the density of x in y. x. The asymptotic density of a set. If x ⊆ y = ∅. xi = xi  . One notion they use is asymptotic density. cvg(x. qk . . where p is the set of prime numbers and q is its complement. 6. they need a more discriminating notion. x converges. cvg(x). Deﬁnition 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. ρ(x. so p > [2n] and OUTPACING is false in OUTPACING models. d. as the evens and the odds are in normal outpacing models. y). In any model of OUTPACING .
ˆ ˆ If n > 1. b. z). between 10. Fact 6. if . z). If 0 ≤ r ≤ 1 and y is inﬁnite.4. Then. xi+1 = c. z) = ρ(x. Let x = i  ∃n(102n ≤ i < 102n+1 . ρ(x. So xk cannot have a limit.3. If 0 ≤ r ≤ 1. ˆ otherwise. ρ(x) = 1. c. xi . c. Modify the construction for (b) in the obvious ways. y outpaces x. then x10 ≤ . If cvg(x. x= xi if xi ≥ r. and so forth. z) < ρ(y. i + 1. Theorem 6. ρ([2n]) = 1/2 ρ([4n] .1 and x10 ˆ b. If x is coﬁnite.999. then cvg(x. z).3. a. SIZE AND DENSITY b. Construct the set x as follows: x0 = ∅ xi .6. 2n+1 2n ≥ . between 100 and 999.000 and 99. Proof. y)ρ(y. there is a set with density r in y. There are divergent sets.9. y) and cvg(y. Suppose r is given.3. So x contains all numbers between 0 and 9. there is a set with density r. [2n]) = 1/2 ρ([4n]) = 1/4 d. Suppose that both x and y converge in z. 95 a.3. z) and ρ(x.
Let Let and let So we have and But So So So b = (ρ(y. there are uncountably many sizes in A. z1 ) < ρ(y. z1 ) = 2/3.b and (c).5. z))/3 6. Even if two sets have the same density. If ρ(x.3b and Theorem 6. z2 ). Since ρ(x. then x outpaces y. z2 ): Let z1 be the set of primes. a. z2 ) Proof.3. z) − b ˆ z ρ(x. z) < b ˆ z for any i > n1 + n2 . But if ρ(x. − b < y i /ˆi − ρ(y. Proof. We cannot strengthen the consequent of 6. then a. there are uncountably many sizes of sets in A. y outpaces x. z1 ). z) − b. let y be an inﬁnite subset of x. So. z2 ). z) + b ˆ z y i /ˆi > ρ(y. but ρ(x. z1 ). 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. x) = 0. x) = 0. N) = 0.3. z1 ) = 1/3 and ρ(y. If A is an outpacing model.96 Proof. for any y1 ⊂ y. and let z = x − y.3. .3. b.4. z2 ) > ρ(y. SETS OF NATURAL NUMBERS n1 = the least n such that for i > n. z2 ) < ρ(y. and both x and y converge in z2 . Then ρ(x.3. sets with distinct densities will have distinct sizes.5.6. So. we can begin to appreciate how precise an ordering outpacing models provide: Fact 6. N) = ρ(y. Immediate from Fact 6. and b. Let x be an inﬁnite set with density r.3. z) + b < ρ(y.4 implies that in any outpacing model. then ρ(x. they will diﬀer in size if they have the same density in some common set. if 0 ≤ r ≤ 1.5 to say that ρ(x. z2 ) ≤ ρ(y.3. z) < b ˆ z n2 = the least n such that for i > n. as in Theorem 6. let x contain every third member of z1 . z) − ρ(x. − b < xi /ˆi − ρ(x. Theorem 6. There are uncountably many subsets of y with distinct densities in y. ρ(x. z1 ) < ρ(y. though ρ(y1 . and let y be z1 − x.3. Theorem 6. z2 ) ≤ ρ(y. then even among sets with density r. from Facts 6. xi /ˆi < ρ(x. where ρ(y.
then x > y. y2 ⊂ y. second.1 The Convexity Problem It’s tempting to infer from these results that the extremely ﬁne ordering of sets by size (or.3. then ρ(z) = ρ(x) .3. by DISJ∪ .4 insures that (1) just in case (2). It turns out that a negative answer is compatible with our theory (CS and OUTPACING). we may consider two additional formulations: (3) If ρ(x) = ρ(y) and x < z < y. So. First. Consider (1) and (1a): (1) (1a) If x ∼ y and ρ(x) = r. 6.6. though diﬀerent sizes may yield the same cardinal number. rather. and ρ(y1 ) < ρ(y2 ). (It is not at all clear that this is true in any power set. we’ll explain why this is called the convexity problem. But x ∼ y.3. cardinality is a function of size. But ρ(z ∪ y1 ) = ρ(z ∪ y2 ) = ρ(x). Theorem 6. linear ordering of sizes. We can regard the cardinality of a set in P(N) as determined by its size. a completion because all sets are located in a single.3. But the situation is really not so clear. at least when we focus on P(N). SIZE AND DENSITY Suppose now that y1 ⊂ y. we’ll discuss the consistency of an aﬃrmative answer. if r < r2 . 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. Recalling that sets may have the same density even though they diﬀer in size.3. then ρ(y) = r. so A z ∪ y1 < z ∪ y2 . ﬁnally. we will show that a negative answer is consistent. the questionable part of (1) can be expressed as (2): (2) If x ∼ y and x converges. then ρ(y) = r. though two sets with the same cardinal number may have diﬀerent sizes. If x ∼ y and ρ(x) = r and cvg(y). so r2 = r. while an aﬃrmative answer may or may not be consistent. there is some r2 = ρ(y). We shall express this fact by saying that. In passing. (1a) is an immediate consequence of Theorem 6. For if y converges. since z was obtained by removing from x a set with density 0 relative to x. 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. then x < y and if r > r2 . Then A by 6. then y converges. 97 y1 < y2 .) Now. we want to know whether the density of a set is a function of its size.4.4. then x is smaller than y.
y ∼ y . 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. SETS OF NATURAL NUMBERS If ρ(x) = ρ(y) and x < z < y. ρ(y) = r. let x1 = x − a. An outpacing model satisﬁes (1) iﬀ it satisﬁes (3). by (3). x ∼ x . If x = ∅ or x = N. . By REP< . for some b ∈ x.98 (4) 6. x and y . Proof.7. then z converges. for some a ∈ x and x2 = x ∪ b. ⇒ Suppose that x < y < z and ρ(x) = ρ(z). x1 and x2 such that ρ(x1 ) = ρ(x2 ) = r = ρ(x) and x1 < y < x2 Assuming that x = ∅ and x = N. Fact 6. and x1 < y < x2 . for which x ⊂ y ⊂ z. we need to ﬁnd two sets. (3) and (4) are equivalent for the same reasons that (1) and (2) are equivalent. ρ(x1 ) = ρ(x) = ρ(x2 ) So. / Then x1 < x < x2 . there are two sets.3. so ρ(y) = ρ(x) To apply (3). by SUBSET. by INDISC∼ . Since adding or removing a single element has no eﬀect on the density of a set. then y = x.
(3) says that the class of sets having a given density is convex. SIZE AND DENSITY 99 The question at hand is called the ‘convexity problem’ because of the formulation in (3).7.8. F . (1).3.3. AF so (1) is false in AF . Open Problem: Is (1) consistent with CS and OUTPACING? x ∼ y. In geometry. (2) and (4) say the same thing.3. The negation of (1) is satisﬁed in some outpacing model. Let K be n  x[n] = y [n] K is inﬁnite. any point between them (ie on the line segment from one to the other) is also in the ﬁgure.3. Proof. . a ﬁgure is convex if. Let x be a set with density r and let y be a divergent set which neither outpaces nor is outpaced by x. We regret that the only thing we know about (3) is that it may be false: Theorem 6. By theorems 6. Applying this notion in the obvious way to the size ordering. given any two points in the ﬁgure. so K is a member of some nonprincipal ultraﬁlter.4 and 6.6.
SETS OF NATURAL NUMBERS .100 6.
. for example.. φ . and φ . These may be predicates..2. in the usual way.. the ranks of the symbols will accord with their familiar uses and we do not stipulate them. That is.. The arguments of such functions are indicated by subscripts. A schematic function is a function whose range is a set of formulae.8).. DIVn . A ﬁrst order language is determined by its nonlogical constant symbols. or function symbols.. maps natural numbers into sentences (see 3. The following notation is used for the predicate calculus: ∧ and ∨ or ¬ not → ↔ = = ∃x ∀x if .. In most cases.1 Notation Predicate Logic The formal theories discussed in this thesis are theories with standard formalization. in the sense of [Tarski.Appendix A A. φ . they are formalized in ﬁrst order predicate logic with identity and function symbols.5]. p. 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 .. 101 . individual constants..
3 Arithmetic For arithmetic. In addition.. 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: 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. 1.. . 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.102 APPENDIX A. we use the following: i+j i−j ij i plus j i minus j i times j ij i to the jth power i! i factorial i  j i divides j gcd(i.2 Set Theory For ﬁrst order languages with boolean operations and predicates. outside of ﬁrst order languages. j) greatest common divisor of i and j n ≡ m mod j n is congruent to m modulo j . N+ N − 0 Z the set of integers ω the smallest inﬁnite cardinal 2ω 2 to the ω A.
L. e. f. A is isomorphic to B (A B). The submodel of B generated by X. a. We assume the notion of satisfaction as deﬁned in [Chang. of a ﬁrst order language... A. the union of a chain of models.1. consists of a domain. b. T2 ).. A = {Ai }. LT . section 1.] = { k  ∃n(k = . B is an elementary extension of A.n.2. A. {Ai } is a chain of models. where X is a subset of B. MODEL THEORY We use the following notation for sets of natural numbers: [. d...3] and use A φ to mean that A satisﬁes φ . b. e. an nary function deﬁned over and yielding values in A. A theory is a set of ﬁrstorder sentences.. Familiar notions about models. A is a submodel of B (A ⊆ B). Notions about theories: a. Deﬁnition B. f is an isomorphic embedding of A into B. h. g. a set of ntuples over A.B. B is an extension of A (A ⊆ B). φ). The language of a theory. ˙ c. T is consistent. T proves φ (T T proves T2 (T d.. Deﬁnition B. .) } 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 modeltheoretic notions and results assumed in the text and presents our notation for these notions. and to each nargument ˙ function symbol of L. ˙ An interpretation. ˙ and a function which assigns to each individual constant of L a member of A.n. T is complete.. ˙ to each nplace predicate of L. c.
2. each of which has all of its universal quantiﬁers preceding any of its existential quantiﬁers. then {Ai } T. [Monk. unless the theory has a prime model. b. p. A and B of T . Fact B.3. Fairly familiar facts: a. A ⊆ B iﬀ A is an elementary submodel of B. More familiar notions: a. If T APPENDIX φ.104 f. Deﬁnition C. Deﬁnition B. If T is complete and T . If T is universal and A T . The following are well known facts: a. T is equivalent to T2 (T ≡ T2 ). T is universal iﬀ it is equivalent to a set of prenex sentences.4. T .5. none of which have universal quantiﬁers. c. A is a prime model of T if A in any model of T .359] T and A can be embedded . T is model complete iﬀ T is consistent and for any two models. Fact B. and A ⊆ B. then T is complete. 355]. then T c. g. b. then some ﬁnite subset of T proves φ (compactness). Deﬁnition C. d. then any submodel of A also satisﬁes T . 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. p. A model complete theory is not necessarily complete. c. Robinson’s notion of model completeness. If T is universalexistential and Ai C Model Completeness This section presents the deﬁnition of model completeness and some basic results needed in Chapters 4 and 5. for all i > 0. φ. [Monk. which is based on A. T is existential iﬀ it is equivalent to a set of prenex sentences. There are several ways to show that a theory is complete. We use only one.1. T . none of which have existential quantiﬁers. If T is existential. A b. If T is categorical. φ is consistent. T is categorical. then B T. T is universalexistential iﬀ it is equivalent to a set of prenex sentences.
T is model complete. 356]: a. c.5. b. φ is a universal formula. x ∈ A. then B φ(x). then A φ(x). If A and B are models of T . the T ∪ L diagram of L is complete. d. b. The following are equivalent [Monk. and B φ(x). φ is a primitive formula. The Aexpansion of L is the result of adding to L a constant for ˙ each element of A.3. .4. then T is complete. The diagram of A is the set of atomic sentences and negations of atomic sentences of the Aexpansion of the language of A which are true in A. A ⊆ B.C. MODEL COMPLETENESS 105 Fact C. A ⊆ B. p. A.1a). and A φ(x). x ∈ A.2. on a theorem of Monk’s (see 4. If A and B are models of T . of T and every Aexpansion L of L. directly or indirectly. For every model. If T is model complete and T has a prime model. which is based on some equivalent formulations of model completeness: Deﬁnition C. a. To show that a theory is model complete we rely. Fact C.
106 APPENDIX .
Georg Cantor: His Mathematics and Philosophy of the Inﬁnite. Mathematical Logic SpringerVerlag. 1937. Dover Publications.W. New York. H. J. George Allen and Unwin. G. A. 222236. Chang and Keisler. 1971. A. pp. Illinois: Northwestern University Press. Tarski. 1979. Amsterdam. Russell. A Mathematical Introduction to Logic. Princeton University Press. A. New York. Inc. Evanston. Orlando.Bibliography [Bell] [Cantor] Bell and Slomson. E. NorthHolland Publishing Company. Monk. B. 1975. [Chang] [Dauben] [Enderton] [Frege] [Griﬃn] [Kurosh] [Monk] [Robinson] [Russell] [Tarski] 107 . New York. 1965. Princeton. New Jersey. The Principles of Mathematics. Pergamon Press. G. Florida. Elementary Theory of Numbers. Griﬃn. 1973. 1955. Amsterdam. J. Academic Press. 96(1960). 1972.W. Frege. NorthHolland Publishing Company. Models and Ultraproducts. Model Theory. McGrawHill. The Foundations of Arithmetic. Amsterdam. Robinson. 1968. 1968. Second edition. D. H. Contributions to the Foundations of the Theory of Transﬁnite Numbers. New York. 1903. New York. 1965. Lectures in General Algebra. Cantor. Norton and Company.. and Zakon. Transactions of the American Mathematical Society. Undecidable Theories. Dauben. reprinted by W. Enderton. Kurosh.G. London. NorthHolland Publishing Company.
29 Qf . 102 x y. y. 102 x − y. 31 ∆2 (x). 103 T1 ≡ T2 .Index of Symbols Logic φ ∧ ψ. 31 ∆1 (x). 1 x < y. 27 N. 94 σA [x]. 102 k!. 103 108 A φ . 45 ZA . 101 ∃xφ. 32 δ(x). 103 A ⊆ B. 103 Models and Domains A. 102 [in + j]. 102 x ∩ y. 102 x. 101 Set Operations x ∪ y. y). 101 φ ↔ ψ. 54 ZA . 1 x ≤ y. 1 x > y. 94 . 67 Qn . 32 θ(x). 66 α(x). y). 103 ˙ A.n . 73 Q. 32 θf (x). 101 ∀xφ. 29 Qn . 25 SA . 101 φ → ψ. 31 ρ(x. 32 QC. 21 An . 102 xk . 94 ˆ Arithmetic k  n. 31 β(x). 103 A φ. 101 ¬φ. 103 B(A). 102 P(x). 67 Qf. 1 x ≥ y. 87 ˆ A. 31 ρ(x). 102 x. 101 φ ∨ ψ. 45 cvg(x. 27 QCn . 86 AF . 45 ˜ C(x). 1 Model Theory T φ. 55 Sizing x. 102 x[k] . 102 Size ordering x ∼ y. 103 T φ.
70 L(λ). 70 Aλ . 70 aλ . 73 ι(λ). 75 qλ . 70 λ • m. 70 ˆ A. 76 109 .INDEX OF SYMBOLS cvg(x). 94 Node Construction λ. 75 λn .
y. 10 TRANS> . 54 ASYM< . 13 DEF> . 40 TRANS∼ . 13 . 10 ASYMF . 10 ASYM> . 20 Unit(x). 13 INF. 9 DEF+ . 46 MIN. 39 CAIf . y). 24 BDIVJ . 22 BAn .m (x). 10 Modn. 38 CAI. 86 Predicates Abstract(x). z). 26 CA. 26 BDIVn . 10 REF∼ . 16 FUNC+ . 46 QUASI. 40 SIZE. 39 CANTOR< . 10 REP< . 10 DISJ∪ . 26 DIVn . 9 RTf . 24 SYM∼ . 10 TRICH. y). 13 CS. 83 PCA.m . 23 EXCORE. 10 ATLEASTn . 10 IRREF< . 20 Divn (x) . 20 LN . 46 MODn. 16 ODD. 23 BAI. 24 DISJ+ . 20 L< . 25 MONOT. 21 110 CSI. 47 PCS. 36 ASSOC. 24 DEF∼ . 10 TRANS< . 23 BA. 20 LC< . 9 OUTPACING. 18 INDISC∼ . 23 MAX. 9 EXACTLYn . 54 Sum(x. 23 BASIC. y). 16 DIVJ . 36 Indisc(x. 36 Theories ADIVn . 1 Atom(x). 36 Sm(x. 20 Timesn (x. 16 DEF+ . 40 CORE.Index of Theories Languages LC . 25 EVEN. 10 Tf . 46 ASSOC2. 46 PSIZE.
51 Zgm.INDEX OF THEORIES UNIQ− . 46 Zg. 54 UNIQ+ . 51 111 . 54 UNIQ∼ .
49 generated. language of. 75 diagram. 103 completion. 105 diverges. 64 Monk mapping. language of. 86 language of a theory. 20 size (L< ). 27 congruous. 36 length. 103 language of class size (LC< ). 104 existential. 51 categorical. 14 functional assumptions. 40 inﬁnite completion. 40 ﬁnite class model. 21 classes (LC ). 74 abelian groups with identity. 14 ﬁnite completion. 40 consistent. 68 class size. 21 subsets of N(LN ). 68 nodal embedding. 20 common nfactor. 103 ﬁlter. 77 node.Subject Index Aﬁnite set. 103 converges. 105 extends. 103 least common multiple. 51 alternating pair. 68 compactness. 104 chain. 103 isomorphic. 39 interpretation. 94 depth. 74 Anear. 48 Monk’s Condition. 27 Nsemigroups. 104 existential assumption. 94 density. 104 complete. 70 node atom. 39 ﬁnite intersection property. 27 (n. 27 nquasicongruence class. 52 near. 87 112 ﬁnite. 103 isomorphic embedding. 70 . 47 ncongruence class. 21 classes. 21 cancellable abelian semigroups with identity. 70 model complete. 104 molecule. 94 elementary extension. 103 charmed nfactor. 85 asymptotic density. 70. 103 equivalent.m)partition. 70 extension. 94 basis. 38 congruence class. 49 expansion. 103 incongruous. 87 ﬁnite set model.
SUBJECT INDEX node atoms. 87 proves. 51 113 cancellable abelian semigroups with identity. 91 satisfaction. 104 Zgroups. 70. 40 translation. 52 Zgroups. 103 schematic function. 103 universal. 45. 71 oneone correspondence. 50 remainder function. 71 node set. language of. 70 . 25 run. 104 universalexistential. 48 simply reducible. 50 union of a chain. 11 quasinodal. 40 standard ﬁnite interpretation. 104 principal. 21 standard protomodel. 21 solution. 12 ultraﬁlter. 27 quasilogical principles. Zg. 101 shell. 61 prime model. 51 simple translation. 70 node sets. 48 trichotomy. 40 remainder theory. 45. 103 theory of abelian groups with identity. 87 uniformly τ reducible. 73 simple order. 84 partial. 50 size. 21 standard interpretation. 83 outpacing model. 86 τ reducible. 51 Zgroups modulo I. 12 roughly divisible. 66 size model. 67 submodel. 1 outpaces. 74 reducible. 40 PCA. 30. 51 Zgroups modulo I. 51 Nsemigroups. 103 subsets of N. Zg. 50 theory. 52 total. 103 quasicongruence class. 104 primitive formula. 46 size. 52 0extends. 40 representation principle. language of.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue listening from where you left off, or restart the preview.