This action might not be possible to undo. Are you sure you want to continue?
David Lewis
With an appendix by
John P. Burgess, A. P. Hazen, .
and David Lewis
Basil Blackwell
Copyright © David Lewis 1991
First published 1991
Basil Blackwell Ltd
108 Cowley Road, Oxford. OX4 1JF, UK
Basil Blackwell, Inc.
3 Cambridge Center
Cambridge, Massachusetts 02142. USA
All rights reserved. Except for the quotation of short passages for the
purposes of criticism and review, no part of this publication may be
reproduced, stored in a retrieval system, or transmitted, in any form or
by any means, electronic. mechanical, photocopying, recording or
otherwise, without the prior permission of the publisher.
Except in the United States of America. this book is sold subject to the
condition that it shall not, by way of trade or otherwise, be lent, resold,
hired out, or otherwise circulated without the publisher's prior consent in
any form of binding or cover other than that in which it is published and
without a similar condition including this condition being imposed on
the subsequent purchaser.
British Library Cataloguing in.Publication Data
A CIP catalogue record for this book is available from the British
Library.
Library of Congress Cataloging in Publication Data
Lewis, David K.
Parts of classes/David Lewis.
p. cm.
Includes bibliographical references.
ISBN Q..631176551  ISBN Q..63117656X (pbk.)
1. Set theory. 2. Whole and parts (philosophy) L Title.
QA248.L48 1991
511.3'22  dc20 8949771
Typeset in 12 on 14pt Bembo
by Graphicraft Typesetters Ltd., Hong Kong
Printed in Great Britain by Billings & Sons Ltd., Worcester
QA 24B .lAB 1991
CIP
Contents
Preface
VlI
1 Classes Apart
1
1.1 Fusions and Classes
1
1.2 Classes and their Subclasses 3
1.3 Are there any other Parts of Classes? 6
1.4 The Nun Set 10
1.5 Consequences of the Main Thesis 15
1.6 More Redefinitions 17
1.7 Sethood and Proper Classes 18
1.8 A Map of Reality
19
1.9 Nominalistic Set Theory Revisited
21
2 The Trouble with Classes
29
2.1 Mysterious Singletons 29
2.2 Van Inwagen's Tu Quoque 35
2.3 On Relations that Generate 38
2.4 Quine on Urelements and Singletons 41
2.5 The Lasso Hypothesis
42
v
3
4
Contents
2.6 Ramsifying out the Singleton Function 45
2.7 Metaphysics to the Rescue? 54
2.8 Credo 55
A Framework for Set Theory
3.1 Desiderata for a Framework
3.2 Plural Quantification
3.3 Choice
3.4 Mereology
3.5 Strife over Mereology
3.6 Composition as Identity
3.7 Distinctions of Size
Set Theory for Mereologists
4.1 The Size of Reality
4.2 Mereologized Arithmetic
4.3 Four Theses Regained
4.4 Set Theory Regained
4.5 Ordinary Arithmetic and Mereologized
Arithmetic
4.6 What's in a Name?
4.7 Intermediate Systems
4.8 Completeness of Me reologi zed
Arithmetic
61
61
62
71
72
75
81
88
93
93
95
98
100
107
109
112
113
Appendix on Pairing
121
by John P. Burgess, A. P. Hazen,
and David Lewis
Index
VI
151
Preface
There is more mereology in set theory than we usually think.
The parts of a class are exactly the subclasses (except that, for
this purpose, the null set should not count as a class). The
notion of a singleton, or unit set, can serve as the distinctive
primitive of set theory. The rest is mereology:a class is the
fusion of its singleton subclasses, something is a member of a
class iff its singleton is part of that class. If we axiomatize set
theory with singleton as primitive (added to an ontologically
innocent framework of plural quantification and mereo
logy), our axioms for 'singleton' closely resemble the Peano
axioms for 'successor'. From these axioms, we can regain
standard iterative set theory.
Alas, the notion of a singleton was never properly ex
plained: talk of collecting many into one does not apply to
onemembered sets, and in fact introduces us only to the
mereology in set theory. I wonder how it is possible for us to
understand the primitive notion of singleton, if indeed we
really do.
In March 1989, after this book was mostly written, I learned
belatedly that its main thesis had been anticipated in the
Vll
Preface
'ensemble theory' of Harry C. Bunt.
1
He says, as I do, that
the theory of part and whole applies to classes; that subclasses
are the parts of classes, and hence that singletons  unit classes
 are the minimal parts of classes; and that, given the theory
of part and whole, the membersingleton relation may
replace membership generally as the primitive notion of set
theory. There is some significant difference from the start,
since Bunt accepts a null individual and rejects individual
atoms. But the more important differences come later: in my
discussion of the philosophical consequences of the main the
sis, and in our quite different ways of formulating set theory
within a mereological framework. I think there is difference
enough to justify retelling the story.
Later in 1989 I had another surprise. I learned how we can,
in effect, quantify over relations without benefit of the re
sources of set theory. (All we need is the framework: plural
quantification and mereology.) John P. Burgess discovered
one way to do it; A. P. Hazen discovered a different way. It
was not quite too late to add an appendix, jointly written
with Burgess and Hazen, in which these methods are pre
sented and applied. They open the way to a new 'structuralist'
interpretation of set theory, on which the primitive notion of
singleton vanishes and my worries about whether it is well
understood vanish too.
Good news; but not a full answer to my worries. The set
theory of the future may go structuralist, if it likes. But I
can't very well say that set theory was implicitly structuralist
all along, even before the discoveries were made that opened
1 Harry C. Bunt, Mass Terms and Model Theoretic Semantics (Cambridge
University Press, 1985), pp. 53 72, 233  301.
Vl1l
Preface
the way. The set theory of the present which means the
bulk of presentday mathematics rests on the primitive
notion of singleton. Is it rotten in its foundation? I dare not
make that accusation lightly! Philosophers who repudiate all
that they cannot understand have very often gone astray.
Maybe my worries are misguided; maybe somehow, I know
not how, we have understood the membersingleton relation
(and with it membership generally) well enough all along. If
so, structuralism is a real solution to an unreal problem, and
we might as well go on as before.
The book reflects my indecision. I whinge at length about
the primitive notion of singleton (section 2.1), but then I
mock those philosophers who refuse to take mathematics as
they find it (section 2.8). The book mostly proceeds under
the working assumption that singleton is a legitimate primi
tive notion. But also it presents the structuralist alternative, in
section 2.6 and in the appendix. (What's said taking singleton
as primitive will not in any case go to waste. All but section
4.6 admits of structuralist reinterpretation.) Structuralism is
at least a very welcome fallback. But I would much prefer a
good answer to my worries about primitive singleton, so
that I could in good conscience take mathematics as I find it.
I thank all those who have helped by discussion of this
material, especially D. M. Armstrong, Donald Baxter, Paul
Benacerraf, George Boolos, Phillip Bricker,John P. Burgess,
M. J. Cresswell, Jennifer Davoren, Hartry Field, A. P.
Hazen, Daniel Isaacson, Mark Johnston, Graham Oppy,
Hilary Putnam, W. V. Quine, Denis Robinson, Barry
Taylor, and Peter van Inwagen. I thank Nancy Etchemendy
for preparing the figures in the appendix. I am indebted
to the Melbourne Semantics Group, where this material
was first presented in 1984; to Stanford University, where
IX
Preface
a more developed version was presented as the 1988 Kant
Lectures; to Harvard University, for research support under
a Santayana Fellowship in 1988; and to the Boyce Gibson
Memorial Library.
x
David Lewis
Princeton, January 1990
1
Taking Classes Apart
1.1 Fusions and Classes
Mereology is the theory of the relation of part to whole, and
kindred notions.
1
One of these kindred notions is that of
a mereological fusion, or sum: the whole composed of some
given parts.
The fusion of all cats is that large, scattered chunk of cat
stuff which is composed of all the cats there are, and nothing
else. It has all cats as parts. There are other things that have all
cats as parts. But the catfusion is the least such thing: it is
included as a part in any other one.
It does have other parts too: all catparts are parts of it, for
instance catwhiskers, catcells, catquarks. For parthood is
transitive; whatever is part of a cat is thereby part of a part of
the catfusion, and so must itself be part of the catfusion.
The catfusion has still other parts. We count it as a part of
itself: an improper part, a part identical to the whole. But also
1 See sections 3.4 and 3.5 for further discussion of mereology.
I
Taking Classes Apart
it has plenty of proper parts  parts not identical to the whole
 besides the cats and catparts already mentioned. Lesser
fusions of cats, for instance the fusion of my two cats Magpie
and Possum, are proper parts of the grand fusion of all cats.
Fusions of catparts are parts of it too, for instance the fusion
of Possum's paws plus whiskers, or the fusion of all
cattails wherever they be. Fusions of several cats plus several
catparts are parts of it. And yet the catfusion is made of
nothing but cats, in this sense: it has no part that is entirely
distinct from each and every cat. Rather, every part of it
overlaps some cat.
We would equivalently define the catfusion as the thing
that overlaps all and only those things that overlap some cat.
Since all and only overlappers of cats are overlappers of
catparts, the fusion of all cats is the same as the fusion of all
catparts. It is also the fusion of all catmolecules, the fusion
of all catparticles, and the fusion of all things that are either
catfronthalves or catbackhalves. And since all and only
overlappers of cats are overlappers of catfusions, the fusion
of all cats is the same as the fusion of all catfusions.
The class of all cats is something else. It has all and only
cats as members. It has no other members. Catparts such as
whiskers or cells or quarks are parts of members of it, but they
are not themselves members of it, because they are not whole
cats. Catparts are indeed members of the class of all cat
parts, but that's a different class. Fusions of several cats are
fusions of members of the class o'f all cats, but again they are
not themselves members of it. They are members of the class
of catfusions, but again that's a different class.
The class of As and the class of Bs can never be identical
unless the As are all and only the Bs; whereas the fusion of the
As and the fusion of the Bs can· be identical even when none
of the As is a B. Therefore we learn not to identify the class of
As with the fusion of As, and the class of Bs with the fusion
2
Classes and their Subclasses
of Bs, lest we identify two different classes with one single
fusion.
A member of a member of something is not, in general, a
member of it; whereas a part of a part of something is always
a part of it. Therefore we learn not to identify membership
with the relation of part to whole.
So far, so good. But I used to think, and so perhaps did
you, that we learned more. We learned to distinguish two
entirely different ways to make one thing out of many: the
way that made one fusion out of many parts, versus the way
that made one class out of many members. We learned that
fusions and classes were two quite different kinds of things, so
that no class was ever a fusion. We learned that 'the part
whole relation applies to individuals, not sets. '2 We even
learned to call mereology 'The Calculus ofIndividuals'!
All that was a big mistake. Just because a class isn't the
mereological fusion of its members, we shouldn't jump to
the conclusion that it isn't a fusion. Just because one class
isn't composed mereologically out of its many members, we
shouldn't jump to the conclusion that there must be some
unmereological way to make one out of many. Just because
a class doesn't have all and only its members as parts, we
shouldn't jump to the conclusion that a class has no parts.
1.2 Classes and their Subclasses
Mereology does apply to classes. For classes do have parts:
their subclasses. Maybe they have other parts as well; that re
mains to be seen. But for now, we have this
2 So said David Lewis in Philosophical Papers, vol. I (Oxford University
Press, 1983), p. 40. I might at least have granted that the partwhole
relation applies to classes in a trivial way: even if a class has no proper
parts, as I then thought, at least it should have itself as an improper part.
3
Taking Classes Apart
First Thesis: One class is a part of another iff the first is a
subclass of the second.
To explain what the First Thesis means, I must hasten to
tell you that my usage is a little bit idiosyncratic. By 'classes'
I mean things that have members. By 'individuals' I mean
things that are members, but do not themselves have mem
bers. Therefore there is no such class as the null class. I don't
mind calling some memberless thing  some individual the
null set. But that doesn't make it a memberless class. Rather,
that makes it a 'set' that is not a class. Standardly, all sets are
classes and none are individuals. I am sorry to stray, but I
must if I am to mark the line that matters: the line between
the membered and the memberless. (Or I could impose novel
coinages on you, which would probably be still more annoy
ing.) Besides, we had more than' enough words. I can hijack
'class' and 'individual', and still leave other words unmo
lested to keep their standard meanings. As follows: a proper
class is a class that is not a member of anything; a set is either
the null set or else a class that is not a proper class; and an
urelement is any individual other than the null set.
3
My First Thesis. therefore, has nothing to say yet about the
null set. It does not say whether the null set is part of any
classes, nor whether any classes are part of the null set. I shall
take up those questions later. Now that you understand what
the First Thesis means, what can I say in its favour?
First, that it conforms to common speech. It does seem
natural to say that a subclass is part of a class: the class of
women is part of the class of human beings, the class of even
3 Warning: I introduced a different idiosyncratic usage of 'class' in my
book On the Plurality q{ Worlds (Blackwell, 1986), pp. 50ln.
4
I
Classes and their :}uf)ctasses
numbers is part of the class of natural numbers, and so on.
4
Likewise it seem natural to say that a hyperbola has two sep
arate parts and not to take that back when we go on to say
that the hyperbola is a class of xy pairs. To a German, the
First Thesis might seem almost tautological: a standard word
for 'subset' is 'Teilmenge', literally 'partset'. Sometimes, for
instance in the writings of Cantor and Zermelo, the word is
just'Teil', sometimes with an implication that the subset is
proper, or proper and nonempty. (Here I am indebted to
Ignacio Angelelli.) The devious explanation of what we say
is that we speak metaphorically, guided by an analogy of for
mal character between the partwhole relation and the subclass
relation. The straightforward explanation is that subclasses
just are parts of classes, we know it, we speak accordingly.
Second, the First Thesis faces no formal obstacles. We
learned, rightly, that membership could not be (a special case
of) the partwhole relation because of a difference in formal
character. But the subclass relation and the partwhole re
lation behave alike. Just as a part of a part is itself a part, so a
subclass of a subclass is itself a subclass; whereas a member of
a member is not in general a member. Just as a whole divides
exhaustively into parts in many different ways, so a class
divides exhaustively into subclasses in many different ways;
whereas a class divides exhaustively into members in only
one way. We have at the very least an analogy of formal
character, wherefore we are free to claim that there is more
than a mere analogy.
Finally, I hope to show you that the First Thesis will prove
fruitful. Set theory is peculiar. It all seems so innocent at first!
We need only accept that when there are many things, then
4 As noted in D. M. Armstrong, Universals and Scientific Realism (Cam
bridge University Press, 1978), vol. II, pp. 367.
Taking Classes Apart
also there is one thing  the class which is just the many
taken together. It's hard to object to that. But it turns out
later that this manyintoone can't always work, on pain of
contradiction yet it's just as h,ard to object to it when it
doesn't work as when it does. What's more, the innocent
business of making many into one somehow transforms itself
into a remarkable making of one into many. Given just one
single individual, Magpie or Possum or the null set or what
you will, suddenly we find ourselves committed to a vast
hierarchy of classes built up from it. Not so innocent after all!
It is exactly this ontological extravagance that gives set the
ory its welcome mathematical power. But, like it or not, it's
far from what we bargained for when we first agreed that
many can be taken together as one. We could understand set
theory much better if we could separate the innocent Dr
Jekyll from the extravagant and powerful Mr Hyde. The
First Thesis will be our first, and principal, step towards that
separation.
1.3 Are there any other Parts of Classes?
The First Thesis leaves it open that classes might have other
parts as well, besides their subclasses. Maybe classes some
times, or always, have individuals as additional parts: the null
set, cat Magpie, Possum's tail (and with it all the tailsegments,
cells, quarks, and what not that are parts of Possum's tail). To
settle the question, I advance this
Second Thesis: No class has any part that is not a class.
The conjunction of the First and Second Thesis is our
6
I
Are there any other Parts of Classes?
Main Thesis: The parts of a class are all and only its
subclasses.
But the Second Thesis seems to me far less evident than the
First; it needs an argument. And an argument needs premises.
My premises will be the First Thesis, plus three more.
Division Thesis: Reality divides exhaustively into indi
viduals and classes.
Priority Thesis: No class is part of any individual.
Fusion Thesis: Any fusion of individuals is itself an indi
viduaL
Roughly speaking, the Division Thesis says that there is
nothing else except individuals and classes. But that is not
exactly right. If we thought that Reality divided exhaus
tively into animal, vegetable, and mineral, that would not
mean that there was no such thing as a salt beef sandwich.
The sandwich is no counterexample, because the sandwich
itself divides: the beef is animal, the bread is vegetable, and
the salt is mineraL Likewise, the Division Thesis permits
there to be a mixed thing which is neither an individual nor a
class, so long as it divides exhaustively into individuals and
classes. I accept a principle of Unrestricted Composition: when
ever there are some things, no matter how many or how
unrelated or how disparate in character they may be, they
have a mereological fusion. (See sections 3.4 and 3.5 for
further discussion.) That means that if I accept individuals
and I accept classes, I have to accept mereological fusions of
individuals and classes. Like the mereological fusion of the
front half of a trout plus the back half of a turkey, which is
7
Taking Classes Apart
neither fish nor fowl, these things can be mostly ignd'red.
They can be left out of the domains of all but our most unre
stricted quantifying. They resist concise classification: all we
can say is that the salt beef sandwich is part animal, part veg
etable, part mineral; the troutturkey is part fish and part
fowl; and the mereological fusion of Possum plus the class
of all catwhiskers is part individual and part class. Likewise,
Reality itself  the mereological fusion of everything  is
mixed. It is neither individual nor class, but it divides exhaus
tively into individuals and classes. Indeed, it divides into one
part which is the most inclusive individual and another
which is the most inclusive class.
If we accept the mixed fusions of individuals and classes,
must we also posit some previously ignored classes that have
these mixed fusions as members? No; we can hold the mixed
fusions to be ineligible for membership. Mixed fusions are
forced upon us by the principle of Unrestricted Composi
tion. Classes containing them are not likewise forced upon us
by a corresponding principle of unrestricted classformation.
That principle is doomed in any case: we dare not say that
whenever there are some things, there is a class of them,
because there can be no class of all nonselfmembers. Nor
are classes containing mixed fusions forced upon us in any
other way. Let us indulge our offhand reluctance to believe
in them.
All I can say to defend the Division Thesis, and it is weak,
is that as yet we have no idea of an y third sort of thing that is
neither individual nor class nor mixture of the two. Remem
ber what an individual is: not necessarily a commonplace in
dividual like Magpie or Possum, or a quark, or a spacetime
point, but anything whatever that has no members but is a
member. If you believe in some remarkable nonclasses 
universals, tropes, abstract simple states of affairs, God, or
what you will it makes no difference. They're still individuals,
8
Are there any other Parts of Classes?
however remarkable, so long as they're members of
classes and not themselves classes. Maybe the mixed fusions
are disqualified from membership. Maybe some classes are
'proper classes' and are disqualified from membership. But
rejecting the Division Thesis means positing some new and
hitherto unheardof disqualification, applicable this time to
things that neither are classes nor have classes as parts, that
can make things ineligible to be members. I wouldn't object
to such a novel proposal, if there were some good theoretical
reason for it. But so long as we have no good reason to inno
vate, let conservatism rule.
The Priority Thesis and the Fusion Thesis reflect our vague
notion that somehow the individuals are 'basic' and 'self
contained' and that the classes are somehow a 'superstruc
ture'; 'first' we have individuals and the classes come 'later'.
(In some sense. But it's not that God made the individuals
on the first day and the classes not until the second.) Indeed,
these two theses may be all the sense that we can extract from
that notion. We don't know what classes are made ofthat's
what we want to figure out. But we do know what indi
viduals are made of: they're made of various smaller individ
uals, and nothing else.
From the First Thesis, the Division Thesis, the Priority
Thesis, and the Fusion Thesis, our Second Thesis follows.
Proof Suppose the Second Thesis false: some class x has a
part y that is not a class. If y is an individual, x has an individ
ual as part; if y is a mixed fusion of an individual and a class,
then again x has an individual as part; and by the Division
Thesis those are the only possibilities. Let z be the fusion of
all individuals that are part of x. Then z is an individual, by
the Fusion Thesis. Now consider the difference x z, the
residue that remains of x after z is removed. (It is the fusion
of all parts of x that do not overlap z.) Then x  z has no
9
Taking Classes Apart
individuals as parts, so it is not an individual or a mixed
fusion. By the Division Thesis, it must be a class. We now
have that x is the fusion of class x  z with an individual z.
Since x z is part of x, and not the whole of x (else there
wouldn't have been any z to remove), we have that the class
x z is a proper part of the class x. So, by the First Thesis,
x z must be a proper subclass of x. Then we have v, a
member of x but not of x  z. According to standard set the
ory, we then have u, the class with v as its only member. By
the First Thesis, u is part of x but not of x  z; by the Priority
Thesis, u is not part of z; so u has some proper part w that
does not overlap z. No individual is part of w; so by the Div
ision Thesis, w is a class. By the First Thesis, w is a proper
subclass of u. But u, being onemembered, has no proper sub
class. This completes a reductio. QED
1.4 The Null Set
A consequence of our Second Thesis is that classes do not
have the null set as part. Because it was a memberless mem
ber, we counted it as an individual, not a class; therefore
it falls under our denial that individuals ever are parts of
classes. To be sure, the null set is incfuded in any class x: all
its members  all none of them  are among x's members.
But it never can be a subclass if it is not even a class. Were we
hasty? Should we amend the Second Thesis, and the premises
whence we derived it, to let the null set be a part of classes
after all? I think not.
Some mereologists
5
posit a 'null individual' meant to play
5 For one, R. M. Martin, 'A Homogeneous System for Formal Logic',
Journal of Symbolic Logic, 8 (1943), pp. 123, especially p. 3; and 'Of Time
and the Null Individual',Journal of Philosophy, 62 (1965), pp. 72335. For
another, Bum, Mass Terms and Model Theoretic Semantics, pp. 567.
10
The Null Set
a role in mereology that corresponds to the role of the null
set in set theory. The null individual is supposed to be part of
everything whatever. So if we take the fusion of the null
individual and something else, we just get the something else
back again, because the null individual was part of it already.
And even if two things don't overlap in any ordinary way,
they still have the null individual as a common part. And it
has no proper parts; for if x is part of the null individual, the
null individual also is part of x, wherefore x and the null indi
vidual are identical. If we accepted the null individual, no
doubt we would identify the null set with it, and so conclude
that the null set is part of every class.
But it is wellnigh unintelligible how anything could be
have as the null individual is said to behave. It is a very queer
thing indeed, and we have no good reason to believe in it.
Such streamlining as it offers in formulating mereology
namely, that intersections of things come out welldefined
even when they shouldn't  can well be done without.
Therefore, reject the null individual; look elsewhere for the
null set.
Should we perhaps reject the null set as we1l? Is it another
misguided posit, meant to streamline the formulation of set
theory by behaving in peculiar ways? I think not. It was mis
guided, perhaps, to call it a 'set'; and I have halfremedied
that by declining at least to call it a 'class'. But the null set's
behaviour is not, after all, so very peculiar. It is included in
every class just because it lacks members and lacking
members is not so queer, all individuals do it. Also, the null
set is more use than the null individual would be, and some
of its services are less easily done without.
The null set serves, first, as a denotation of last resort for
classterms that fail to denote classes. It is the settheoretical
intersection of x and y, where x and y have no members in
common (just as the null individual was invented to be the
1 I
Taking Classes Apart
mereological intersection of x and y when x and y didn't
really overlap). It is the class of all selfmembers, since there
are no selfmembers; it is the class of real numbers x such that
x
2
+ 1 = o. This service is a mere convenience. It would be
better to do without it than to purchase it by believing in
some queer entity.
But, second, the null set also serves as an element of last
resort. Suppose you repudiate the null set, and make do en
tirely with classes generated from some ordinary individual
 let it be cat Possum. And suppose you reduce mathemat
ical objects to classes, as usual  to do that, it matters not
whether you begin with the null set or with Possum. Instead
of the pure sets, you will have Possum himself; you will have
Possum's singleton, or unit class, that contains him as its sole
member; you will have the class of Possum's singleton and
Possum himself; you will have the new singleton of that
class; you will have the class of that new singleton and the old
singleton and Possum; and so on ad infinitum and far beyond,
until you have enough modelling clay to make the whole of
mathematics. But now your confidence in the existence of
the entire mathematical realm  all those numbers, matrices,
curves, homeomorphisms, the lot  rests on your confidence
that Possum exists! If I was fibbing when I said I had a cat of
that name, shall mathematics fall? And where was mathemat
ics before Possum was born to redeem it? If mathematics is to
be safe, it had better rest on a surer foundation. Possum is not
a sure thing. The null set is. You'd better believe in it, and
with the utmost confidence; for then you can believe with
equal confidence in its singleton, the class of that singleton
and the null set, the new singleton of that class, the class of
that new singleton and the old singleton and the null set, and
so on until you have enough modelling clay to make the
whole of mathematics.
(The null individual could do us no corresponding service.
12
The Null Set
If we have the null set, we have morethancountless other
things besides. If we had the null individual, we would have
the null individual and that would be that. Mereology does
not spin extravagant realms of being out of just one single
thing.)
Must we then accept the null set as a most extraordinary
individual, a little speck of sheer nothingness, a sort of black
hole in the fabric of Reality itself?
Not really. We needn't be ontologically serious about the
null set. It is useful to have a name that is guaranteed to de
note some individual, but it needn't be a special individual
with a whiff of nothingness about it. Ordinary individuals
suffice. In fact, any individual will do for the null set  even
Possum. Like any individual, he has the main qualification
for the job  memberlessness. As for the second qualification,
guaranteed existence, that is not really a qualification of the
jobholder itself, rather it is a requirement on our method of
selection. To guarantee that we will select some existing indi
vidual to be the null set, we don't need to select something
that's guaranteed to exist. It is enough to make sure that
we select from among whatever things happen to exist. We
can select Possum, contingent though he be, so long as our
method of selection would have found something else had
Possum not been there to find.
(And what if there had existed no ordinary individuals
whatsoever?  In that case, maybe we can let mathematics
fall. Just how much security do we really need?)6
6 I myself hold a thesis of plurality of worlds according to which (1)
Possum could not have failed to exist altogether, though he might have
been off in some other world and not a world mate of ours; and (2) it is
necessary that there be something  some individual  and not rather
nothing. But I shall not rely on this thesis here, for fear that not all of you
accept it.
Taking Classes Apart
The choice of an individual to serve as the null set is arbi
trary. An arbitrary choice might be left unmade. We could
say that what's true is what's true on all ways of making the
choice; what's false is what's false on all ways; and what's
true on some and false on others could be left indetermi
nate as to truth.
7
If I left the choice unmade, 'null set' would
be harmlessly ambiguous: Possum would be not unequivo
cally not the null set, likewise Magpie, and likewise all other
individuals.
But I prefer not to leave the arbitrary choice unmade.
Instead, I make it arbitrarily! as follows. Instead of the un
helpful definition of the null set as the set without members, J
adopt this
Redefinition: The null set is the fusion of all individuals.
That's an easy selection to specify; and it's guaranteed to sel
ect an individual to be the null set if there exist any individ
uals at all. It also introduces a handy name for one of the
main subdivisions of Reality. For what it's worth, it respects C
our 'intuition' that the null set is no place in particular, no
more one place than another. It's as far as can be from the
notion that the null set is a speck of nothingness, and that's all
to the good.
It may be that this choice of the null set offends against
some 'intuition' you had that the null set was small and non
descript. If you think I am offering you a substitute for the
null set as you have conceived of it hitherto, no harm done.
We may agree to disagree, and proceed. But for myself, I
think such an 'intuition' has no ground or authority, and the
7 Here I apply Bas van Fraassen's method of ' super valuations'. See van
Fraassen, 'Singular Terms, TruthValue Gaps and Free Logic', Journal oj
Philosophy, 63 (1966), pp. 48195.
14
Consequences of the Main Thesis
fusion of all individuals is perfectly suitable to serve as the
null set. .
If the null set is the fusion of all individuals, then, by the
Division, Priority, and Fusion Theses, the individuals are all
and only the parts of the null set. And the ure1ements, indi
viduals other than the null set, are the proper part of the
null set.
1.5 Consequences of the Main Thesis
?ur Main Thesis says that the parts of a class are all and only
Its subclasses. This applies, in particular, to onemembered
classes: unit classes, or singletons. Possum's singleton has
Possum as its sole member. It has no subclasses except itself
!herefore it is a mereologi,cal atom: it has no parts except
Itself, no proper parts. Likewise the singleton of Possum's
singleton is an atom; and likewise for any other singleton.
Anything that can be a member of a class has a singleton:
every individual has a singleton, and so does every set. The
only things that lack singletons are the proper classes classes
that are not members of anything, and afortiori not members
of singletons  and those mixed things that are part individ
ual and part class. And, of course, nothing has two singletons.
So the singletons correspond oneone with the individuals
and sets.
A class has its singleton subclasses as atomic parts, one for
each of its members. It usually has other parts that are not
singletons, namely subclasses with more than one member.
However, if x is part of a class y, then x must have one or
more singletons as parts, else x could have no members,
could not be a class, and so could not be a subclass of y. In
fact x must consist entirely of singletons, else the rest of x
would be a part of y with no singletons as parts, so a part of a
15
Taking Classes Apart
class that is not a subclass. A class is the union, and hence the
fusion, of the singletons of its members. For example, the
class of the two cats Possum and Magpie is the fusion of
Possum's singleton and Magpie's singleton. The class whose
three members are Possum, Magpie's singleton, and the
aforementioned class is the fusion of Possum's singleton,
Magpie's singleton's singleton, and the singleton of the afore
mentioned class. And so it goes.
If we take the notion of a singleton henceforth as primi
tive, we have this new definition of a class.
Redefinition: A class is any fusion of singletons.
The members of a class are exactly those things whose
singletons are subclasses of it; and so, by our Main Thesis, are
exactly those things whose singletons are parts of it. An indi
vidual has no members and, by our Priority Thesis, it has no
singletons as parts. The remaining case, by our Division The
sis, is a mixed fusion, part individual and part class. It does
have singletons as parts, yet probably we would not want to
say that it has members. So when we define membership
(formerly primitive) in terms of the notion of singleton, we
must write in a restriction to classes.
8
Dtifinition: x is a member of y iff Y is a class and the single
ton of x is part of y.
By the Division and Priority Theses, something is an in
dividual iff it has no class as part; and it has a fusion of
8 Bunt does without this restriction, and accordingly grants that his con
cept of membership differs from that in set theory. See Bunt, Mass Terms
and Model Theoretic Semantics, p. 61.
More Redefinitions
singletons as part iff it has one or more singletons as parts. So
we have another new definition.
Redefinition: An individual is anything that has no single
tons as parts.
Equivalently, since singletons are atoms, an individual is any
thing that overlaps no singletons; equivalently, since classes
are fusions of singletons, an individual is anything that over
laps no classes. Hence the null set, as we have arbitrarily cho
sen it, is the fusion of all things that do not overlap the fusion
of all singletons; which is to say that it is the mereological dif
ference, Reality minus that fusion. It is what's left of Reality
after all the singletons are removed.
1.6 More Redtdinitions
Once we redefine membership and null set, we can proceed
as usual to all other notions of set theory. But sometimes it is
instructive not to go the long way round via membership,
but rather to define other notions directly. Unfortunately,
the special case of the null set keeps complicating the story.
Inclusion. The main case is that a class includes a subclass
by having it as a part. But also a class includes the null set,
which is not one of its subclasses and not one of its parts. The
null set is not a class, but it too includes the null set. And al
though a class, or the null set, may be part of a mixed fusion,
we probably would not want to call that a case of inclusion.
So we have:
Redefinition: y includes x iff (1) x is the null set and y
is the null set or a class, or (2) x and yare classes and x is
part ofy.
Taking Classes Apart
Union. The main case is that a union of classes is their
fusion. But the null set also may enter into unions; and other
individuals, or mixed fusions, may not. So we have this:
Redtifinition: The union of one or more things is defined
iff each of them is either a class or the null set; it is the
nul1 set if each of them is the null set, otherwise it is the
fusion of those of them that are classes.
1.7 Sethood and Proper Classes
A class is a set iff it is a member of something. If x is a mem
ber of something, it must have a singleton to be part of that
thing, and if x has a singleton, the singleton is something that
x is a member of. So a class is a set if it has a singleton, other
wise it is a proper class. But the null set also counts as a set, so
sethood requires another disjunctive definition.
Redtifinition: Something is a set iff either it is a class that
has a singleton, or else it is the null set.
Redtifinition: A proper class is a class that has no singleton.
For instance, the class of all sets that are nonselfmembers
had better not be a set, on pain of Russell's paradox. Al
though it is indeed a nonselfmember (everything is, accord
ing to the standard principle of Fundierung), that won't make
it a selfmember unless it is a set. So it isn't; it is a proper class,
it has no singleton, and it cannot be a member of anything.
We dare not allow a set of all sets that are nonselfmembers
,
but there are two alternative ways to avoid it. One way
would be to restrict composition: we have all the sets that are
nonselfmembers and we have a singleton of each of these
18
A Map of Reality
sets, but somehow we have no fusion of all these singletons,
so we have no class of all these sets. But there is no good in
dependent reason to restrict composition. Mereology per se is
unproblematic, and not to blame for the settheoretical para
doxes; so it would be unduly drastic to stop the paradoxes by
mutilating mereology, if there is any other remedy. (Just
as it would be unduly drastic to solve problems in quantum
physics by mutilating logic, or problems in the philosophy of
mind and language by mutilating mathematics.) The better
remedy, which I have adopted, is to restrict not composition,
but rather the making of singletons. We can fuse all the single
tons of sets that are nonselfmembers, thereby obtaining a
proper class of sets, but this proper class does not in turn have
a singleton. Unlike composition, the making of singletons is
illunderstood to begin with, so we should not be surprised
or disturbed to find that it needs restricting.
I do not say, note well, that we posit the proper classes
because of their utility in formulating powerful systems of set
theory. George Boolos has argued, and I agree, that we can
get all the power we need by resorting to plural quantifi
cation.
9
And yet we have the proper classes willynilly, be
they useful or be they useless. We do not go out of our way
to posit them; rather we can't keep them away, given our
Main Thesis and Unrestricted Composition.
1.8 A Map of Reality
Figure 1 is a map of all of Reality. The dots are mer:eological
atoms. Those above the gap are singletons; those below are
9 See Boolos, 'To Be is To Be the Value of A Variable (or To Be Some
Values of Some Variables)" Journal of Philosophy, 81 (1984), pp. 43049;
'Nominalist Platonism', Philosophical Review, 94 (1985). pp. 32744.
Taking Classes Apart
••..••.•.•.••.•.....................................................
•.•..........................................
...............................
.....................
................
........
Figure 1
atomic individuals. The blob at the bottom is atomless gunk:
an individual whose parts all have further proper parts. Any
thing made of the dots above the gap is a class. The biggest
class is the thing made of all the dots above the gap. It is the
universal class; every singleton is part of it, therefore every
thing is a member of it, save only those things that are not
members of anything. It is a proper class, not a member of
anything, and not a set; by the standard principle of Limi
tation of Size, it is too big to be a set. Many other classes like
wise are too big to be sets. The smaller classes, however, are
sets; and these classes have their own singletons, which are
dots somewhere above the gap. A set is a member of just
those classes that have its singleton as a part; and by the stan
dard principle of Fundierung, it can never itself be one of the
classes that its singleton is part of. To make sure of this, let a
set's singleton always go just above the set itself. (That's
why there's no top, and that's why there are more and more
dots as ,we ascend.) The smallest classes are the singletons,
each one being just one dot, and of course these are all small
enough to be sets.
Anything made of the dots below the gap, the blob of
20
Nominalistic Set Theory Revisited
gunk, or some of each is an individuaL (Maybe there isn't
any gunk; or maybe there aren't any atomic individuals; or
ma ybe there are both, and that is the case shown. )10 The
biggest one is the null set, the fusion of all individuals; it is
made of all the atomic individuals plus all of the atomless
gunk. Each individual has its singleton, somewhere up just
above the gap, and is a member of all classes that have that
singleton as a part. That is why the part of the map above the
gap must be so much bigger: so that there will be enough
singletons to go around.
Straddling the gap are the 'salt beef sandwiches': the mixed
fusions, ignored but undenied, consisting partly of dots or
gunk from below and partly of dots from above. These have
no singletons and are not members of anything.
1.9 Nominalistic Set Theory Revisited
Years ago, I wrote a paper called 'Nominalistic Set Theory'.ll
By 'nominalistic' I meant 'setless', in the Harvard fashion;
nominalism in the traditional sense of the word  repudiation
of universals  had nothing to do with it. I argued that we
could imitate set theory, to a limited extent, by mereological
means.
One reason why the fusion of several things is different
10 Bunt disagrees; he takes it as axiomatic that every atom is a singleton
(Mass Terms and Model Theoretic Semantics, p. 60). So his individuals, if
there arc any, must consist of gunk; for instance he cannot countenance
spacetime points (unless they are singletons). I find this assumption hasty.
Later he reopens the question (p. 291), but only for purposes of a formal
comparison of systems.
11 David Lewis, 'Nominalistic Set Theory, Nous, 4 (1970), pp. 22540.
21
Taking Classes Apart
from the of them is that a fusion does not, in general,
have a into parts. However, it may
happen that If we dIStingUIsh some parts of a fusion as nice
then a fusion will have a unique decomposition into
nIce parts.
. Cats are nice. A fusion of cats has a unique decomposition
Into cats; the fusion of cats c
1
• c
2
• c
3
•• •• is identical to the
fusion of cats d1• d
2
• d
3
•• •• only if each if the cs is one of the ds
and each of the ds is one of the cs. That is because one cat is
never a proper part of another, and one cat is never a proper
part of the fusion of several others. More generally. it is be
cause cats never overlap.
Maximal spatiotemporally connected parts are nice. For
sake of. this example. let us assume that spatiotemporal
things C?nSlSt entirely of pointsized particulars; igriore uni
versals, If such there be. Suppose we have some notion of
what it means two things to touch spatiotemporally:
roughly. that pOints of one are arbitrarily close to points of
the other. Then something x is connected iff there are no y and
z such that x is the fusion of y and z and y does not touch z' x
is a maximal connected part of w iff x is a connected part of w
that is a proper part of any other connected part of w. No
two connected parts of a fusion overlap, so a fusion
has a UnIque .decomposition into its maximal connected parts.
atoms are nice. No two atoms overlap,
so any fusIon of atoms has a unique decomposition into
its atoms.
on some specific definition of 'nice part'. Then fusions
?f .nIce parts imitate classes, and the relation 'nice part of'
Imitates the membership relation, in this way: whenever we
have some suitable things, we will have a fusion that has all
and only those things as nice parts. Say it like this: we have a
pseudoclass that has all and only those things as its pseudo
members.
22
r
,
Nominalistic Set Theory Revisited
This first stab at setless theory doesn't go far, because the
only things that can qualify as pseudomembers of pseudo
classes are the things that will be nice parts. If nice parts are
maximal connected parts, for instance, then only a connected
thing can ever be a pseudomember of a pseudoclass. If
we try to put in a disconnected thing, we cannot get that
thing itself back as a pseudomember. The closest we can
come is to get back its several maximal connected parts.
That's the best case. Sometimes other unwelcome things hap
pen instead.
To improve matters, we resort to coding. Define some
scheme of coding on which things that are not themselves
nice may nevertheless have nice codes. Suppose we can en
code the things we want to have as the pseudomembers of
our pseudoclasses. Then we can say that x is a pseudo
4
member
of w, relative to a given definition of 'nice part' and a given
scheme of coding, iff some nice part of w is a code for x. In
'Nominalistic Set Theory', I chose maximal connected parts
as my 'nice parts' and tried various geometrically definable
codings. (The 'first stab' was the case where I chose the cod
ing to be identity.) I got imitations of set theory, but they
worked only under especially favourable conditions.
Example. Suppose the things of interest to us are made out
of points in an infinite grid. Each point in the grid is next to
eight others: two vertical neighbours, two horizontal, four
diagonal. Two things touch iff some point of one is next to
some point of the other. Nice parts are maximal connected
parts. The interior of something consists of those points of it
which are not next to any points that are not parts of it.
Something x is a code for something y iff y is the interior of
x. If we have disconnected things, we can make connected
codes for them by wrapping them in skins and running
strings from one part to the other. Thus: suppose we have
two disconnected things a and b, each composed of the points
Taking Classes Apart
· . . . . . . . . . . . . . . . . . . . . . . .
· . . . . . . . . . .
• jaal. • • • • • • ["bbl. • • · .
. . .
• • ....... .. .
• • •
· .
·raal·
.
· . . . .
· . . . .
• [bbl.
.
• •
. .
· . . . . . . . . . . . . . . . . .
. . . .
· . . . . . . . . . .
Figure 2
· . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . .
· . . .
• c c c c c c c c c c c c c c c c • • • • • • • • • . . . . .
· . • c·· • • • • • • • • • • • • c • • • • • • • • • . . . . .
· . • c • • • • • • • • • • • • • • c • • • • • • • • • • • • • •
'
· .
d d d d·· • • • c c c c • • • • • d d d d
d/b bid. • • • • cJa3lc • • • • • dfbbld
• ·cccc··
• • c[a3lc • •
• • •
· . .
· .
· . .
· .
• • •
db b d·· • • • • • •••
d d d d·· • • • c c c c • • • • • d d d d • •
• • • • • • • • • • • '. • • d • • d r
• ·cccc·· • • •
. . . . . . . . . . .. .....
• • • • • • • • • • • • • • d • • • • • • • • • • • • • • d • • • • •
· . . . . . .
• • • • • • • d·· • • • • • • • • • • • • d •
· . • •
• • • • • • • d·· • • • • • • • • • • • • d •
· . . . . . .
• •
· .
• • • • • • • d • • • • • • • • • • • • • • d •
· . . . . . .
• •
· .
••••••• dddddddddddddddd·
· .
· . . . . . .
· .
· . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
Figure 3
so labelled as shown in figure 2. Then we can form a
pseudoclass as shown in figure 3. The fusion a + c is a con
nected code for a, the fusion b + d is a connected code for
b, these two codes are the two nice parts of the fusion
a + c + b + d, wherefore that fusion is a pseudoclass of a and
b. The coding is redundant: the strings could have taken dif
ferent routes. So we cannot speak of the pseudoclass of a and
b. Any fusion whose nice parts are codes for a and for b is a
pseudoclass of a and b.
Besides the lack of a unique decomposition, another reason
· . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
· . . eeeeeeeeeeeeeeeeee· · . . . . . . · . . .
· . . ecccccccccccccccce
· .
· . . · . . · . . .
· . . e c eeeeeeeeeeeeee c e • · . . . . . . · . . .
• • • e c e • · . . . • • • • • •
• e
c e • · . . . · . . · . . .
· . . e c e •
• • • • • • •
. . . .
ec
e •
· .
. . . . .
· . . .
· . . e c
e •
· . . . . .
• • • • • e
• · . . · . . .

e e e e c e •
· .
eeeee
ceeee·
· .
eeeeee
,
• • e d d d de· • • e
'
dddde e c c c c e • cccce· • • e
• •
• e dlb bid e e •
• • e
dlb bid e
ecaace · . . edbbde·· ·e c a ace
• • • e
dbbde
ecccce • • • e··
• e
c c c c e··
• e
dddde
eeeeee· • • eeeed e • • ·eeeeee·
· .
e d'eee e
'
d
d er;;;
· . . . . · . . . . . . e e • • · . . .
· .
· . . . e
· . . . .
· .
· . .
• • e
d e
· .
• •
. .
· .
· . .
• e
d e •
· .
· . . . . • • • • • • • e
d e • • · .
. .
· .
· . .
• e
d e •
· .
· . . . · . . • • • •• ed Ie e e e e e e e e e e e e e d e •
· .
• • • • • • • · . . . . edddddddddddddddd e •
· .
· . . .
· .
• • • • • • eeeeeeeeeeeeeeeeee·
· .
· .............. \ . · . . . . . . . . . .
· . . . . . . . . . . . . . . . . . · . . . . . . . .
· . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
• • • j j j j j j j j j j j j j j j j j j • • • • • • • • • • • •
• • • j h h h h h h h h h h h h h h h h j • • • • • • • • • • • •
• • • j h j j j j j j j j j j j j j j h j • • • • • • • • • • • • •
• • • j h j • • • • • • • • • • • • j h j • • • • • • • • • • • •
• • • j h j • • • • • • • • • • • • j h j • • • • • • • • • • • •
•• ejhj •• •••••••••
""jj""j j h j • • • j j j j j j I· .. j h j j j j ••• j j j j j j
j'hhil h j • • • j iii i j j h hi1i1 j ••• j 1 1 1 1 j
j h[Tf]h j • •• j i!"iiP j j j j j h[Tf]h j ••• j
j hluJhJ ••• j iLuJi j';=;=; j hluJh j ••• j J
jhhhhj ••• j iii ij.· .jhhhhj···j iii ij
j j j j j j • • • jjjj i j • • • j j j j j j ••• j iTlll
••••••••• •••••••••••• jiJ •••
•••••••••••• jij •• ···········jij···
.••....•.••• jij············jij···
'kkk" ••••••••• j ijjjjjjjjjjjjjj ij···
k k k • • • • • • • • • j iii iii iii iii iii i j • • •
k k k • • • • • • • • • j j j j j j j j j j j j j j j j j j • • •
'
· . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Figure 4
Taking Classes Apart
wh y the fusion of several things differs from the class of them
is that fusion offers nothing to correspond to the settheoretical
hierarchy of individuals, classes of individuals, classes of
classes of individuals (or mixed classes, containing classes of
individuals and also individuals), and so on, ad infinitum and
beyond. But if coding can be iterated, then fusion plus cod
ing do offer at least the beginning of a hierarchy. We saw
that a + c + b + d  call it P  was a pseudoclass of a and b. By
attaching a skin and strings, as shown in figure 4, we can
make a+c+b+d+e, a code for P. Likewisej+h+g+i
call it Q  is a pseudoclass ofj and g andj+ h + g+ i + j is a
code for Q. Now a+c+b+d+e+j+h+g+i+j, which has
these two codes as its nice parts, is the pseudoclass of P and
Q. And we could go on a little longer. There's a problem:
when P turns up as a pseudomember of the pseudoclass of P
and Q, how can we tell that it's meant to be taken as a code
for a pseudoclass, and further broken up into nice parts and
decoded? How could we tell whether k, if it turned up as a
pseudomember of some other pseudoclass, was meant to be
further decoded? It could be: it is a code for its central point.
It is easy to think of more problems. What if the result of
going up the hierarchy is that our grid gets too crowded to
make room for any more strings and skins? (How much
would it help to work in more than two dimensions?) What
if we want to make a pseudoclass of things that touch? Or
almost touch? Or worse, overlap? What if we want to make
a pseudoclass of a disconnected thing and something that
completely surrounds one of its parts? You can define ever
more complicated geometrical codings to cover ever more
difficult cases. You can while away many an hour. But you
cannot hope to arrive at a mereologicalcumgeometrical
reconstruction of the entire realm of mathematical objects.
To see this, it is enough to remember that, out of a countable
26
Nominalistic Set Theory Revisited
grid of points; there are only continuum many different fu
sions; but the mathematical realm is much bigger than that.
But imagine now that we could have a coding for free,
and never have to to define it. Farewell honest toill What
conditions would we like it to meet?
We'd like to avoid crowding; so let the codes be small. In
fact, let them be atoms; because whenever we have a fusion
of atoms, its atoms are nice parts of it. (Something that is a
maximal connected part of one fusion may not be a maximal
connected part of another; its niceness is relative to the fu
sion. Not so for the niceness of atoms.)
We must avoid ambiguity. We'd also like to avoid redun
dancy, in order to use our supply of codes efficiently. So let
the encoded things and their codes correspond oneone.
We'd like to avoid any confusion between our original
individuals and our codes, or fusions of codes. So let the
codes be entirely distinct from the individuals.
We'd like to be able to iterate coding and build up a hier
archy. So let the codes, and the fusions of codes, have their
own codes in turn  so far as possible. But there must be a
limitation. Not all codefusions can have codes.
Proof We apply the reasoning of Cantor's Theorem and
Russell's Paradox. Call a codefusion normal iff it has a code
and its code is not part of it. By Unrestricted Composition,
we have the fusion of all codes of normal codefusions; call it
n. Has n a code? Then by definition of n, n's code is part of n
iff n is normal. By definition of normality, n's code is part of
n iff n is not normal. We conclude that n is a codefusion that
has no code. QED
In fact, if we are not to run out of codes, the great majority
of codefusions must remain codeless. One way to ration a
Taking Classes Apart
limited supply of codes is to allocate them only to the com
parativel.y small codefusions. Then the great majority of
would go without, because the great majority
of codefusIons are not smalL To take a miniature example: if
we had countably many codes, then each of the countably
finite codefusions could have one, leaving the con
tinuum many infinite codefusions uncoded. To be sure, this
of Limitation of Size is not the only possible way of
selecting a favoured few codefusions to receive codes. Alter
native principles of rationing might allow codes to some few
of the infinite fusions, or deny codes to some or all of the fi
nite ones. But standard iterative set theory uses Limitation of
Size, and that is what our setless imitation is meant to imitate.
Codefusions are pseudoclasses. Those with codes of their
own are pseudosets, the rest are proper pseudoclasses. Also
there is the null pseudoset, something or other that has no
codes as parts and therefore has no pseudomembers. When
ever something has a code, that code is a pseudosingleton.
only heaven would grant us an ideal coding, one that
satIsfied all the desiderata just listed, and one that provided
codes for all manner of ordinary individuals, then the pro
gram of setless set theory would succeed. The pseudoclasses
made by mereological fusion of codes would be every bit as
good as the real thing.
I say they would be the real thing. If setless set theory
works that well, its classes are no longer pseudo. There is
nothing we know about the real classes that distinguishes
them from the 'imitations'. Somehow, I know not how
heaven has granted us an ideal coding. We call the
'singletons'. Classes are codefusions, with singletons as their
nice  that is, atomic  parts. Something is a member of a
class iff some nice part of that class is a code for that thing.
2
The Trouble with
Classes
2.1 Mysterious Singletons
Cantor taught that a set is a 'many, which can be thought of
as one, i.e., a totality of definite elements that can be com
bined into a whole by a law'. To this day, when a student is
first introduced to set theory, he is apt to be told something
similar. 'A set is a collection of objects .... [It] is formed by
gathering together certain objects to form a single object'
(Shoenfield). A set or class is 'constituted by objects thought
of together' (Kleene). 'Roughly speaking, a set is a collection
of objects and is thought to have an independent existence
of its own ... ' (Robbin). Maybe also, or instead, he will be
given some familiar examples: ' ... for example, the set of all
even numbers is considered to be just as real as any particular
even number such as 2 or 16' (Robbin); 'A pack of wolves, a
bunch of grapes, or a flock of pigeons are all examples of sets
of things' (Halmos).l
1 Georg Cantor, Gesammelte Abhandlungen mathematischen und philosophi
schen Inhalts, cd. Ernst Zermdo (Springer, 1932), p. 204; Joseph R.
The Trouble with Classes
But after a time, the unfortunate student is told that some
classes  the singletons  have only a single member. Here is
a just cause for student protest, if ever there was one. This
time, he has no. 'many'. He has no elements or objects  I
stress the plural to be 'combined' or 'collected' or 'gathered
together' into one, or to be 'thought of together as one'.
Rather, he has just one single thing, the element, and he has
another single thing, the singleton, and nothing he was told
gives him the slightest guidance about what that one thing
has to do with the other. Nor did any of those familiar ex
amples concern singlemembered sets. His introductory les
son just does not apply.
(He might think: whatever it is that you do, in action or in
thought, to make several things into a class, just do the same
to a single thing and you make it into a singleton. Daniel
Isaacson suggests this analogy: How do you make several
paintings into an art collection? Maybe you make a plan, you
buy the paintings, you hang them in a special room, you
even publish a catalogue. If you do the same, but your
money runs out after you buy the first painting on your list,
you have a collection that consists of a single painting.  But
this thought is worse than useless. For all those allusions to
human activity in the forming of classes are a bum steer.
Sooner or later our student will hear that there are countless
classes, most of them infinite and miscellaneous, so that the
vast majority of them must have somehow got 'formed'
with absolutely no attention or assistance from us. Maybe
we've formed a general concept of classes, or a theory of
them, or some sort of sketchy mental map of the whole of
Shoenficld, Mathematical Logic (AddisonWesley, 1967), p. 238; Stephen
C. Kleene, Mathematical Logic (John Wiley, 1967), p. 135; Joel W.
Robbin, Mathematical Logic: A First Course (W. A. Benjamin, 1969), p.
171; Paul R. Halmos, Naive Set Theory (Van Nostrand, 1960), p. 1.
Mysterious Singletons
settheoretical Reality. Maybe we've formed a few mental
representations of a few very special classes. But there just
cannot be anything that we've done to all the classes one at a
time. The job is far too big for us. Must set theory rest on
theology?)
We were told nothing about the nature of the singletons,
and nothing about the nature of their relation to their ele
ments. That might not be quite so bad if the singletons were
a very special case. At least we'd know about the rest of the
classes. But since all classes are fusions of singletons, and
nothing over and above the singletons they're made of, our
utter ignorance about the nature of the singletons amounts
to utter ignorance about the nature of classes generally. We
understand how bigger classes are composed of their single
ton atoms. That's the easy part: just mereology. That's where
we get the many into one, the combining or collecting or
gathering. Those introductory remarks (apart from the mis
guided allusions to human activity) introduced us only to the
mereology in set theory. But as to what is distinctively set
theoretical  the singletons that are the building blocks of all
classes  they were entirely silent. Dr Jekyll was there to wel
come us. Mr Hyde kept hidden. What do we know about
singletons when we know only that they are atoms, and
wholly distinct from the familiar individuals? What do we
know about other classes, when we know only that they are
composed of these atoms about which we know next to
nothing?
Set theory has its unofficial axioms, traditional remarks
about the nature of classes. They are never argued, but are
passed along heedlessly from one author to another. One of
these unofficial axioms says that the classes are nowhere:
they are outside of space and time. But why do we think
this? Perhaps because, wherever we go, we never see them or
stumble over them. But ma ybe they are invisible and intangible.
The Trouble with Classes
Maybe they can share their locations with other things.
Maybe they are somewhere. Where? If Possum's singleton
were elsewhere than Possum himself, it would presumably
be an obnoxiously arbitrary matter where it was. If it's in
Footscray, why there instead of Burwood? But perhaps
Possum's singleton is just where Possum is. Perhaps, indeed,
every singleton is just where its member is. Since members of
singletons occupy extended spatiotemporal regions, and single
tons are atoms, that would have to mean that something
can occupy an extended region otherwise than by having
different parts that occupy different parts of the region, and
that would certainly be peculiar. But not more peculiar, I
think, than being nowhere at all we get a choice of equal
evils, and cannot reject either hypothesis by pointing to the
repugnancy of the other. If every singleton was where its
member was, then, in general, classes would be where their
members were. The class of Magpie and Possum would be
divided: the part of it that is Magpie's singleton would ocupy
the region where Magpie is, the part that is Possum's single
ton would occupy the region where Possum is, and so the
entire class would occupy the entire spatiotemporal region
where Magpie and Possum are.
Perhaps when we say that classes are outside of space and
time, it is especially the 'pure' classes that we have in mind.
These are the classes built up from the null set and nothing
else: the null set, its singleton, its singleton's singleton, the
class of the null set and its singleton, and so on. Maybe we
think that these classes, at any rate, are nowhere; because we
start by thinking that classes are where their members are, we
conclude that a class with no members must be nowhere, and
finally we conclude that all the pure classes must share the
location or un location of the null set that is their common
ancestor. But even if genuine classes are where their members
are, the null set is no genuine class, only a 'set' by courtesy. If
;2
Mysterious Singletons
the null set is the fusion of all individuals, as I arbitrarily
stipulated, then it is everywhere, rather than nowhere. If the
pure classes share its location, they too are everywhere 
nowhere in particular, to be sure, but by no means outside of
space and time.
I don't say the classes are in space and time. I don't say they
aren't. I say we're in the sad fix that we haven't a clue
whether they are or whether they aren't. We go much too
fast from not knowing whether they are to thinking we
know they are not, just as the conjurer's dupes go too fast
from not seeing the stooge's head to thinking they see that
the stooge is headless.
Another unofficial axiom says that classes have nothing
much by way of intrinsic character. We know that's not
quite right: to be an atom, or to be a fusion of atoms, or to be
a fusion of exactly seventeen atoms, are matters of intrinsic
character. However, these are not matters that distinguish
one singleton from another, or one seventeenmembered
class (a seventeenfold fusion of singletons) from another. Are
all singletons exact intrinsic duplicates? Or do they some
times, or do they always, differ in their intrinsic character? If
they do, do those differences in any way reflect differenc6s
between the character of their members? Do they involve
any of the same qualities that distinguish individuals from
one another? Again we cannot argue the case one way or the
other, and if we think we know that classes have no distinc
tive intrinsic character, probably that's like thinking we
know that the stooge is headless.
2
2 Some philosophers propose that things have their qualities by hav
ing them as parts. The qualities might be repeatable universals, as in the
main system of Nelson Goodman, The Structure of Appearance (Harvard
University Press, 1951); or they might themselves be particular, as in
Donald C. Williams, 'On the Elements of Being' in Principles of Empirical
Realism (Charles Thomas. 1966), and Keith Campbell, Abstract Particulars
H
The Trouble with Classes
Sometimes our offhand opinions about the nature of classes
don't even agree with one another. When Nelson Goodman
finds the notion of classes 'essentially incomprehensible' and
says that he 'will not willingly use apparatus that peoples his
world with a host of ethereal, platonic, pseudo entities',3 we
should ask which are they: ethereal or platonic? The ether is
everywhere, and one bit of it is pretty much like another;
whereas the forms are nowhere, and each of them is unique.
Ethereal entities are 'light, airy or tenuous', says the diction
ary, whereas the forms are changeless and most fully real
(whatever that means). Ifwe knew better whether the classes
were more fittingly called 'ethereal' or 'platonic', that would
be no small advance!
Because we know so little about the singletons, we are ill
placed even to begin to understand the relation of a thing to
its singleton. We know what to call it, of course  member
ship  but that is all. Is it an external relation, like the spatia
temporal relations of distance? Or is it an internal relation,
like a relation of intrinsic similarity or difference? Or is it a
combination of the two? Or something else altogether? It
cannot be entirely an internal relation: when two individuals
are exact intrinsic duplicates, as sometimes happens, they can
not differ in their internal relations to any third thing, yet
something is the singleton of one but not the other. If classes
are outside space and time, then membership cannot be at all
a matter of spatiotemporal relations between the thing and
the singleton. Even if classes are where their members are, it
cannot be entirely a matter of spatiotemporal relations: Possum
and his singleton occupy the same region on this hypothesis,
(Blackwell, 1990). A mereologically atomic singleton could have no
qualities as proper parts. Still it would not follow that a singleton had no
qualitative character. It might be a quality.
3 Goodman, The Structure of Appearance, section II. 2.
34
Van [nwagen's Tu Quoque
and so are just alike in their spatiotemporal relation to
any third thing, yet something is the singleton of one and
not the other. (N or will it help to add an extra quasispatial
dimension of , rank' in the settheoretical hierarchy  the ver
tical in our map of Reality  because the singleton of the
singleton of the singleton of Possum and the singleton of the
class of Possum and his singleton will be at the same rank, as
well as the same spatiotemporallocation.)
We cannot use the internal structure of a singleton to encode
the information which thing it has as its member; because a
singleton, being an atom, has no internal structure. Nor can
we use familiar qualities that the singleton might conceivably
share with familiar individuals. There are different single
tons for all the individuals and all the sets. There just aren't
enough familiar qualities to encode that much information.
Even if we suppose that singletons have spatiotemporalloca
tions to help carry the information, we still fall very far short.
It seems that we have no alternative but to suppose that the
relation of singleton to member holds in virtue of qualities
or external relations of which we have no conception what
soever. Yet we think we do somehow understand what it
means for a singleton to have a member!
Finally, it's no good saying that a singleton has x as its
member because it has the property: being the singleton of x.
That's just to go in a circle. We've named a property; but all
we know about the property that bears this name is that it's
the property, we know not what, that distinguishes the single
ton of x from all other singletons.
2.2 Van Inwagen' 5 Tu Quoque
Elsewhere I have complained about the difficulty of under
standing the relation that allegedly relates concrete things to
the abstract simple possibilities  propositions or properties 
35
The Trouble with Classes
that they realize. Peter van Inwagen replied with a tu quoque:
I accept set theory, yet the relation of member to set seems
difficult to understand for exactly the same reasons.
4
Van
Inwagen concludes that my complaints are as damaging to
the set theory I accept (and to classical mathematics as well) as
they are to the doctrine of abstract simple possibilities I reject.
Why should they be lethal to the latter and not the former?
His point is very well taken, and my present complaints
against membership (or rather, against the special case of the
membersingleton relation) amount to my endorsement of
it. But are the parallel complaints lethal, or are they not?
I suppose they're equally formidable in both cases, but not
quite lethal in either case. It's a nasty predicament to claim
that you somehow understand a primitive notion, although
you have no idea how you could possibly understand it.
That's the predicament I'm in when I accept the notion of
singleton, and that's the predicament I claim others are in if
they accept the alleged notion of realization of abstract sim
ple possibilities. Either is bad, and both together are worse.
What price escape? It seems (and van Inwagen agrees) that
rejecting the notion of singleton means rejecting presentday
settheoretical mathematics. Hot though it is in the frying
pan, that fire is worse. But in the case of the abstract simple
possibilities, escape can be had by accepting the doctrine
of plurality of worlds. That's much easier to understand
though not very easy to believe. This time, say I, the price
right  escape is worth it. The difference is almost entirely
between the fires, not between the frying pans.
Almost entirely: I have to add that the tu quoque is imper
fect. I posed a dilemma: is the relation of the concrete world
4 For my complaint, see Lewis, On the Plurality of Worlds, pp. 17691;
for his reply, see Peter van Inwagen, 'Two Concepts of Possible Worlds',
Midwest SiudiesinPhilosophy,11 (1986),pp. 185213.especiallypp. 207 10.
. '
···r··
1
.' .....•..
Van Inwagen's Tu Quoque
to the abstract possibilities internal or external? So far as the
'internal' hom goes, the tu quoque is complete. Not so for the
'external' hom  but there's enough left of the tu quoque, even
there, to drive home how nasty it is to have to believe in the
membersingleton relation.
Taking the 'external' hom, one big part of my complaint
was that we had no independent grasp on what sort of rela
tion we were dealing with. That applies equally to the case of
the singletons. But another part of my complaint was that we
get a mysterious necessary connection: if the concrete world
has a certain intrinsic qualitative character, it must stand in the
external relation to certain abstract simples and not others.
Why? We would expect, as in the case of shape and distance,
say, that intrinsic character and external relations can vary
independently.
Can you turn a parallel complaint against the singletons?
Try it thus.
5
Suppose the relation of member to singleton is
external. Why must Possum be a member of one singleton
rather than another? Why isn't it contingent which singleton
is his?  But to this I have a reply. On my theory of modality
the question becomes: why doesn't some otherworldly
counterpart of Possum have a singleton which isn't a coun
terpart of the singleton that Possum actually has? And my
answer is: what makes one singleton a counterpart of another
exactly is their having counterpart members. Or rather, I say
that it's a flexible matter which things we count as counter
parts; one way, the answer just given holds; other ways, it's
not even true that a counterpart of Possum must have a
counterpart of his singleton.
6
5 Van Inwagen, 'Two Concepts of Possible Worlds', p. 210.
6 On counterparts, see David Lewis 'Counterpart Theory and Quan
tified Modal Logic',}ournal of Philosophy, 65 (1968), pp. 11326; On the
Plurality of Worlds, passim.
37
The Trouble with Classes
Now, can the defender of abstract possibilities answer
me in a parallel way?  No, because the two complaints in
volve different kinds of necessary connection. The complaint
against singletons involves a necessary connection between
external relations and the identity of the relatum; whereas the
complaint against abstract simple possiblilities involves a nec
essary connection between external relations and the qualita
tive character of the relatum. The method of counterparts does
not apply to the latter problem.
2.3 On Relations that Generate
I dislike classes because they burden us with mysteries. Nelson
Goodman dislikes classes for a different reason. In 'A World
of Individuals',
7
he advances a compelling principle: the
'generating relation' of a system should never generate two
different things out of the very same material. There should
be no difference without a difference in content. There has to
be something to make the difference, and it has to be some
thing built, directly or indirectly, into one thing but not the
other. We can state this more precisely. Let R be the generat
ing relation; if we start with some sort of nontransitive
relation of direct or immediate generation, let R be not that
but rather its ancestral. A minimal element is something to
which nothing else bears R. The content of a thing consists of
7 In Philosophy oj Mathematics: Selected Readings, ed. Paul Benacerraf and
Hilary Putnam (Prentice Hall, 1964); and in Nelson Goodman, Problems
and Projects (BobbsMerrill, 1972). The paper was originally published in
two parts: 'A World of Individuals' in I. M. Bochenski, Alonzo Church,
and Nelson Goodman, The Problem oj Universals (University of Notre
Dame Press, 1956); and the important appendix as a separate note 'On
Relations that Generate', Philosophical Studies, 9 (1958), pp. 65 6.
On Relations that Generate
those minimal elements that bear R to it (or of the thing
itself, if it is already minimal). Then Goodman's principle
requires that no two things have exactly the same content.
Otherwise, R is not acceptable as a generating relation.
If we take the proper part relation as our generating re
lation, Goodman's principle is satisfied: it is a principle of
mereology that no two things consist of exactly the same
atoms. (Not if the two things consist entirely of atoms, any
way. The principle needs revision if we accept atom less gunk
as a genuine possibility.) If we start with membership, on the
other hand, and take its ancestral as our generating relation,
and accept anything remotely resembling the normal exist
ence axioms of set theory, then we find Goodman's principle
violated in a big way. Countless things are generated out
of exactly the same content. Possum's singleton, his single
ton's singleton, the class of him and his singleton, and ever so
many more things all are generated out of Possum alone.
There is difference galore without difference of content.
Roughly, I want to say that I accept Goodman's principle
(ignoring gunk), but I deny that the ancestral of membership
is truly a generating relation. But then I have some urgent
explaining to do, because Goodman stipulates that the an
cestral of membership is a generating relation in a system
founded on set theory. What is a generating relation, any
way?
Goodman characterizes the idea in a preliminary way with
a volley of nearsynon yms. A generating relation is one where
by entities are 'generated', or 'comprised of', or 'made up of',
or 'composed of' other entities;, and its converse is a relation
whereby one entity 'breaks down into' others. But then he
proceeds to an example in which it is stipulated that the gen
erating relations of two systems are in one case the proper
part relation and in the other case the ancestral of member
ship  the challenge is to find out which is which. Finally (in
39
The Trouble with Classes
an appendix originally published apart from the rest of the
paper) he gives the official definition: 'the generating relation
of a system is the proper part relation or the ancestral of ...
membership or the logical sum of the two, as they occur in
the system'. Given this stipulation, of course I cannot deny
that the ancestral of membership is a generating relation;
given the stipulation, I deny instead that Goodman's prin
ciple has any force. It says only that if a relation is either the
proper part relation, or the ancestral of membership, or the
logical sum of the two, then it must obey a certain condition
which the proper part relation does obey and the ancestral of
membership (or the sum of the two) does not. Why should
anyone believe that, unless because he renounces classes
already?
But if we go against Goodman's wishes and ignOf'e his
official stipulation, then we have an argument more worthy
of attention. Suppose we thought that we had a broad notion
of composition (or 'generation', or 'making up', or what you
will) and that mereological composition was only one spe
cies of composition among others. Then we could very well
think that every kind of composition should obey Goodman's
principle, even if there were some sort of unmereological
composition alongside the mereological sort. If so, we
should conclude that generation ,of classes via the ancestral of
membership is not  contrary perhaps to our first impressions
a legitimate sort of unmereological composition.
And with that I agree. But not because I think it illegit
imate. Rather, because the generation of classes is not an
unmereological sort of composition. In so far as it is unmereo
logical, it isn't composition; in so far as it is composition, it is
mereological. When something is a member of a class, first
the thing is a member of its singleton  that part is not com
position at all. Then its singleton is part of the class  that part
is mereological.
Quine on Urelements and Singletons
When the thing is a member of its singleton, there isn't
one thing made out of many; there is one thing made out of
one thing. Further, the thing and its singleton are entirely dis
tinct. (If member x and singleton y were not entirely distinct,
then, since y is an atom, either x and y would be identical, or
y would be a proper part of x and x would be a class, or y
would be a proper part of x and x would be a fusion of a class
and an individual. In the first and second cases, x would be a
member of itself, which is impossible; in the third case, x
could not be a member of anything.) What does this relation
of one thing to another wholly distinct thing have to do with
composition, in no matter how broad a sense? You might
as well say that marrying someone is an odd sort of com
position, in which your spouse is generated from you! For
better or worse, the ancestral of membership is officially a
generating relation. But if it's a generating relation in the
(vague) sense given by Goodman's preliminary volley of
nearsynonyms, you might as well say that the ancestral of
'being married to someone who is part of' is a generating
relation too. Apart from the fact that marriage is better un
derstood, the cases are parallel.
2.4 Quine on Urelements and Singletons
Quine has discovered a nice way to remove all mystery
about the nature of some singletons.
8
He reinterprets the pre
dicate of membership when applied to urelements (for him,
'individuals', since he does not call the null set an individual)
to mean what we would originally have called membership
oridentity. He thereby counts urelements as selfmembers;
thereby he identifies each urelement with its singleton;
8 See W. V. Quine, Set Theory and Its Logic (Harvard University Press,
1963), pp. 314.
The Trouble with Classes
indeed, he defines an urelement (an 'individual') as something
identical to its singleton. If Possum is his own singleton, then
also his singleton is his singleton's singleton, and so he himself
is his singleton's singleton; likewise he is his singleton's
singleton's singleton, and so on. So we have no more mys
tery about the nature of any of these singletons than we do
about Possum himself
Quine's plan cannot be extended, else we give most classes
spurious members. If the null set were its own singleton, it
would have itself for a member, though it was supposed to
be memberless. If the class of Magpie and Possum were its
own singleton, it would have itself as a third member along
with Magpie and Possum. And so on. So if we followed
Quine, we could at best remove only some of our mysteries
about the nature of singletons and their relation to their
members.
But in fact we cannot even begin to follow Quine. Cats
are urelements, so for Quine they are their own singletons.
The molecules of which cats are composed also are urele
ments, so again their own singletons. The class of cats, for us,
is the fusion of catsingletons; so if cats are their own single
tons, that class is the fusion of cats. Likewise the class of cat
molecules, if these too are their own singletons, is the fusion
of catmolecules. The fusion of cats and the fusion of cat
molecules are identical. But the class of cats and the class of
catmolecules are not identical. Sad to say, Quine's idea is not
for us. Nice as it is, I submit that our mereology of classes is
nicer; and the two cannot coexist on pain of collapse.
2.5 The Lasso Hypothesis
We learnt in school to picture a class by drawing pictures of
its members, and then drawing a picture of a lasso around
The Lasso Hypothesis
them. What if this were the exact truth of the matter? Maybe
the singleton of something x is not, after all, an atom; but
rather consists of x plus a lasso. That gives singletons an in
ternal structure after all; and so might seem to shed light, in a
speculative way, both on the nature of the singletons and on
their relation to their members.
(Picture it a different way, if you'd rather: a singleton is its
member encapsulated. What surrounds the member is not a
soft loop but a hard shell. So tough is this shell that when we
try to break a class down into its smallest parts, we do not
quite succeed. We get the singleton subclasses, but these are
not yet genuine mereological atoms. Rather they are the en
capsulated members of the class, each one consisting of mem
ber plus shell, which have remained unsplit despite our best
efforts.  This picture needs more than a grain of salt! For
when something consists of parts, that is not because of any
thing we have done to break it into parts. It makes no differ
ence whether we are able to separate the parts, or whether it
even makes sense to talk of separating the parts. So ignore the
breaking, ignore the failing to break, ignore the toughness of
the shells; what's left is just the hypothesis that the singleton
of x consists of x plus something else. Whether you call this
something else a 'shell' or a 'lasso' is neither here nor there.)
The Lasso Hypothesis says that a singleton has proper
parts, namely its lasso, its member, and its member's proper
parts, if any. So it contradicts our Main Thesis, which says
that a class has no parts but its subclasses, and which thereby
implies that a singleton has no proper parts. So it must also
contradict one of the premises whence the Main Thesis
follows. I take it to contradict the Division Thesis: the lasso
would be neither individual nor class nor mixed fusion of the
two. But the Division Thesis could go, in a good cause; all I
said in its favour was that we had no good reason to abandon
it; and if abandoning it could buy us clarification of how
43
The Trouble with Classes
a singleton' is related to its member, that would be a good
reason indeed!
But now ask: if the singleton of x consists of x plus a lasso,
and the singleton of y consists of y plus a lasso, can it ever be
that the same lasso is used twice over? (Can it even be that
one single lasso will do for making all the singletons?) Or
must it be a different lasso each time, one lasso per singleton?
The answer is that it must be a different lasso each time.
Proof Say that lasso L fits x iff the singleton of x is the
fusion L + x. Suppose for reductio that we have two different
things x and y, and yet one lasso L fits both: x's singleton is
L + x, y's singleton is L + y. Case 1: Either x and yare both
individuals, or else they are both classes. If they're classes
they're sets, and their union is a set too. Either way, another
thing that has a singleton is the fusion x + y; its singlefon is
M+ (x+ y). (We needn't ask whether M and L are the same
lasso.) Now the class whose members are x and x + y is the
fusi.on of the two singletons, that is (L + x) + (M + (x + y»,
whIch reduces to L + M + x + y; and likewise the class whose
members are y and x + y is (L + y) + (M + (x + y», which
also reduces to L + M + x + y. So the classes are identical. It
follows that x and yare identical, contra our supposition.
Case 2: One of x and y, let it be x, is an individual; the other,
y, i.s a class. Then the class of x and y is (L + x) + (L + y),
whIch reduces to L + x + y; and the union of y with the
singleton of x is (L + x) + y, which also reduces to L + x + y.
two are identical. It follows that either class y
IS Identical to mdIvidual x, or else class y is a member of itself,
both of which are impossible. This completes the reductio.
QED
Lassos correspond oneone to singletons, and therefore to
members of singletons. Our old questions were: What, if
44
Ramsifying out the Singleton Function
anything, do we know about the nature of singletons? How
is a singleton related to its member? Our new questions are:
What, if anything, do we know about the nature of lassos?
How is a lasso related to the thing it fits? We are no better off
if we adopt the Lasso Hypothesis.
In fact, nothing changes at all. Whether we accept the
Lasso Hypothesis or whether we reject it, we still say that to
each x that has a singleton, there corresponds something y
that is wholly distinct from x, and also there corresponds the
fusion y + x. We could call y the 'lasso' and y + x the 'single
ton', and so accept the Lasso Hypothesis. Or we could call y
the 'singleton' and thereby reject the Lasso Hypothesis, and
then we could call y + x the fusion (perhaps mixed) of the
singleton and its member. What's in a name?
2.6 Ramsifying out the Singleton Function
We know nothing, so I lament, about the nature of the primi
tive relation between things and their singletons. What we
do know, though, is that this relation satisfies certain struc
tural conditions set forth in the axioms of set theory; and that
it satisfies certain other conditions, for instance, that ordinary
things such as cats, quarks, and spacetime points should turn
out to be among the things that have no singletons as parts.
Might that be all we need to know? Paul Fitzgerald has sug
gested that it is.
9
(Or rather, he has suggested this for the
membership relation generally. But if I am right that x is a
member of a class y just when x is a member of a singleton
that is part of y, and if I am right that the relation of part to
whole is not itself mysterious, then we can confine our attention
9 Paul Fitzgerald, 'Meaning in Science and Mathematics' in PSA 1974,
cd. R. S. Cohen et al. (Reidel, 1976), section IV.
45
The Trouble with Classes
to the special case of member and singleton.) We needn't
pretend to speak unequivocally of the function that takes
members to singletons. Rather, any function that conforms
to the appropriate conditions shall count as a singleton func
tion. The content of set theory is that there exists some such
function.
Axiomatize set theory with 'singleton' as the only set
theoretical primitive. Write down a system of structural,
mathematical axioms chosen to yield standard iterative set
theory. Make the axioms strong enough to rule out Skolem
ized interpretations with a countable domain, since these
are obviously unintended. (We shall see in section 4.2 what
such an axiom system should look like. Besides 'singleton'
and elementary logic, there must of course be other primitive
apparatus. But we may take that to be interpreted once and
for all. It will be neither settheoretical nor mysterious.) Add
unofficial axioms as well, embodying those of our customary
metaphysical opinions about classes that deserve credence.
Include, for instance, an axiom stating that ordinary things 
cats, etc.  are individuals, rather than classes or mixed fu
sions. Omit, however, those unofficial axioms which declare
that singletons are outside space and time, that they are eth
ereal, or that they are platonic. Now conjoin all the axioms,
mathematical and unofficial, into one single sentence:
... singleton ... singleton ... singleton ...
The idea is to regard the word 'singleton' here not as an un
equivocally meaningful primitive, but as a functionvariable
bound by an invisible quantifier. To give the content of the
conjoined axioms more explicitly, we take the Ramsey sen
tence:
For some S: .. . S ... S . . . S . ..
!.
Ramsifying out the Singleton Function
And if the word 'singleton' appears in some other sentence,
Possum's singleton has a singleton
we should supply the quantifier and the structural conditions
that restrict it, thus:
For all S: if ... S . .. S . .. S . .. , then Possum's S has
an S.
(Or better, thus:
For some S: ... S ... S ... S ... ; and
for all S: if. .. S . .. S . .. S . .. , then Possum's S has
an S
so that the existence of suitable values of S is affirmed rather
than just presupposed. That brings our treatment of the con
joined axioms and our treatment of other sentences into line.)
Think of it this way: the word 'singleton' is highly equivo
cal, since all interpretations of it that satisfy the conjoined
axioms are equally 'intended' interpretations. Though it ap
pears in the guise of a primitive constant, it is no b.ette.r t?an
a variable  albeit a variable restricted to a certam hmlted
range of admissible values. The equivocal sentence about
Possum is true simpliciter iff it is true on all its intended
interpretations (and there are some); equivalently, true on
admissible values (and there are some) of its disguised van
able.lO
10 On tolerating equivocation, see van Fraassen, 'Singular Terms, Truth
Value Gaps and Free Logic' on supervaluations. On Ramsification, see F.
P. Ramsey, 'Theories', in Foundations (Routledge and Kegan Paul, 1978);
or Rudolf Carnap, Philosophical Foundations oj Physics (Basic Books,
1966), section 26.
47
The Trouble with Classes
A parallel 'structuralist' account of arithmetic is familiar
and attractive. We have the primitive notion 'successor'. (It
can be our only arithmetical primitive. Numbers are those
things that have successors. Zero is the number which is not
a successor.) We have structural conditions, set forth in the
Peano axioms, that the successor function is supposed to
satisfy. But beyond that, alas, we haven't a clue about the
nature of.the function. (Or maybe we are willing to
reduce to set theory, and claim  pace my present
lamentatIOns  to understand that. Then our predicament is
that we have too many conflicting ideas about what the suc
cessor function might be, and no way to choose.) The struc
turalist's solution is to Ramsify out the successor function.
He says that the word 'successor' is equivocal, no better than
a (restricted) variable: any interpretation of it that satisfies the
Peano axioms shall count as 'intended'. The content of the
axioms is just that there exists some intended interpretation.
in which the word 'successor' occurs is true simpli
CIter Iff It IS true on all intended interpretations; that is, true
for all admissible values of its disguised variable.
The structuralist about arithmetic needn't scratch his head
about the unknown nature of the numbersuccessor relation'
the structuralist about set theory needn't scratch his head
the nature of the membersingleton relation. Any
functIOn that satisfies the stipulated conditions will do. We
talk equivocally about all such functions. There's no secret
fact about which one is the right one, the one we really mean.
So sa ys the structuralist.
I'm Structuralism is not a such a bad position
ab?ut arIthmetic, though it would be better to respect our
conviction that 'successor' has an unequivocal mean
mg. But structuralism for set theory does not fare so well.
In the first place, even if we could remove all mystery
about the nature of the membersingleton relation, we would
"
:::r
1
,'
,
Ramsifying out the Singleton Function
still have a mystery about the nature of the singletons. Note
that we can say this, even if 'singleton' means nothing defi
nite. For on all intended interpretations of 'singleton' alike,
the same things will count as singletons: namely, all the mereo
logical atoms, except for those that are parts of the ordinary
things that, by axiomatic stipulation, were supposed to count
as individuals.
The structuralist might retreat by dropping the unofficial
axiom which says that singletons may not turn out to be
among the atomic parts of ordinary things. Then some of the
things that pass for singletons under some intended interpre
tations might be ordinary and unmysterious. But not nearly
enough of them.
Here structuralism about arithmetic is better off. If we
want enough ordinary things to make room for an intended
interpretation of 'successor', countably many will do. It is
to believe there are enough. Probably there are count
ably many spacetime points (or pointsized momentary
parts of particles) in this world alone, not to mention what
the other worlds may have to offer. Maybe in this world, and
surely in all the worlds together, there are at least countably
many quarks, and countably many cats. So, unless we stipu
late otherwise when we lay down the axioms, the things
that pass for numbers under various intended interpretations
of 'successor' might secretly be ordinary things, points or
quarks or cats.
But standard iterative set theory  let alone large cardinal
axioms!  makes far stronger demands on the size of Reality.
There must be more than countably many atoms, more than
continuum many, more than two to the continuum power
many, more than two to the two to the continuum, ... and
all the demands in this sequence are but the barest beginning.
It is hard to think where there could be that many ordin
ary atoms. So if there are enough things, and in particular
49
The Trouble with Classes
enough atoms, to make room for an intended interpreta
tion of 'singleton', then the overwhelming majority of these
things must be extraordinary. The preponderant part of
Reality must consist of unfamiliar, unobserved things, whose
existence would have gone unsuspected but for our accep
tance of set theory, and of whose intrinsic nature we know
nothing. It's like the astronomers' problem of the missing
mass of the cosmos  but it's far, far worse.
The structuralist might retreat once more. He might for
mulate the axioms in an elementary way that would not,
after all, rule out the undemanding Skolemized interpreta
tions.
ll
Then if he claimed that those interpretations were,
after all, intended, he might conclude that ordinary things
were abundant enough to meet his needs. I protest: those
interpretations are not intended, and to say otherwise is to
to suit our philosophy. It's subtle
latton, but It s stIll unacceptable. And it's no excuse that there
is a philosophical problem about how we grasp the difference
between intended and Skolemized interpretations of set the
ory. We do grasp it. Any philosophy of thought and langu
age that says we can't thereby stands refuted.
No retreat, then. Settheoretical structuralism must
acknowledge the mystery of the missing mass. At most
it may hope to remove the mystery of the member
singleton function.
But in the second place, what can it mean to say  not yet
presupposing set theory when we say it  that there exists a
suitable function? We understand· what makes a function
suitable as an interpretation of 'singleton', but what's a
11 If he did this by taking all instances of certain schemata as axioms and
if he did not abbreviate by means of substitutional 'quantification', 'then
his conjuction of axioms would be an infinitely long sentence. I see no
harm in that.
Ramsifying out the Singleton Function
function? A set theorist will say that a function is a special case
of a relation, and a relation is a class of ordered pairs. Some
times it's a set of pairs; but in the case of the singleton func
tion, it's a proper class. But our structuralist cannot agree, not
at this stage of the game, because he does not grant that the
words 'class of ordered pairs' have any defmite meaning.
Should the structuralist adopt some alternative theory of
functions, and of relations generally? (1) There are various
other settheoretical analyses. But whatever advantages these
may have, they are one and all unavailable at this stage of the
game. (2) A theory which says that relations are sui generis
individuals, but makes them classlike in all but name, can be
expected to end in a predicament differing only in name
from our present predicament. If relations are classlike, a
plausible new 'Main Thesis' will say that the parts of a rela
tion are all and only its subrelations; and we will be left
scratching our heads over what it means for a minimal subre
lation, instantiated just once, to relate one thing to another.
Some neostructuralist might propose to Ramsify out this
primitive notion. . .. (3) A very different sort of theory of
relations as sui generis individuals treats them as elements of
being, building blocks of nature, constitutive of the qualitat
ive character of things composed of interrelated parts. Spa
tiotemporal relations between the parts of a thing constitute
that thing's shape and size, for instance. But we cannot ex
plain the membersingleton relation in terms of any known
aspects of the qualitative character of things, and it seems
overbold to respond by positing new, hidden aspects of
qualitative character. What's good about structuralism is that
it would allow us to get by (at the cost of losing uniqueness)
with the sort of relations that pair things up arbitrarily and
have nothing to do with qualitative character. (4) Some talk
of relations may just be substitutional 'quantification' over
relational predicates. But the only relevant predicates we
The Trouble with Classes
have are 'member of' and 'singleton of', and if we are struc
turalists we will regard these as unsuitable substituends be
cause they lack definite meaning. (5) Or we may quantify
over relational concepts; but that won't help unless we
already have at least one definite concept suited to be the
concept of membership in a singleton. I cannot survey all the
theories that our structuralist might conceivably try, but it
seems that we may dismiss the obvious candidates. Whatever
their merits otherwise, they do not meet his present need.
He may solve part of his problem if he says not 'there is a
class of pairs' but just 'there are some pairs'. He can insist.
with Boolos, that plural quantification is primitively intelli
gible and need not and cannot be reduced to singular quan
tification over classes or anything else. I agree (see section
3.2).
But how, at this stage of the game, is the structuralist
entitled to speak of ordered pairs? We usually define pairing
settheoretically, in various wellknown ways. These defi
nitions are not available to the structuralist. He denies that
the primitive notion of set theory 'member' or 'singleton'
has any definite meaning, so 'pair' defined in settheoretical
terms is no better off. If we need a settheoretical definition
of 'ordered pair', settheoretical structuralism can succeed
only if we understand settheoretical notions to begin with.
Only if we don't need it can we have it.
The structuralist might claim to have a primitive under
standing of 'ordered pair'. But if so, why bother to be a
structuralist? If he has pairing, he has selfpairing. And if he
has selfpairing, he might as well define the singleton of x
as the selfpair of x and x. Given reasonable axioms on primi
tive pairing, we would expect selfpairs to behave as single
tons should. In requiring such axioms, he is no worse off
than those of us who must lay down axioms directly on a
r
!
Ramsifying out the Singleton Function
primitive notion of singleton, or on a primitive notion of
membership.
The game is not yet over. There might be some third way,
neither settheoretical nor primitive, to introduce pairing,
and thereby to simulate quantification over relations, and
thereby to get rid of the primitive notion of singleton. In fact
this can be done, thanks to new work by John P. Burgess
and A. P. Hazen. The appendix tells how. If we wish, we
may have a new, structuralist set theory. First the general
memberclass relation gives way to the special case, member
singleton; then we Ramsify out membersingleton; and so
we are left with no settheoretical primitive at all. Farewell,
then, to my complaints that the membersingleton relation is
illunderstood. No worries, that which is not there requires
no understanding.
Unfortunately, the structuralist redemption of set theory
is not retroactive. Structuralist set theory really is new. It
would be highhanded anachronism to claim that set theory
was structuralist all along. It took the work of Burgess and
Hazen to turn settheoretical structuralism into an available
option (if I am right to dismiss versions of structuralism that
purport to quantify over relations taken as sui generis indivi
duals). Before mid1989, nobody even knew how to be a set
theoretical structuralist. To be a structuralist, you quantify
over relations; to be a settheoretical structuralist, you do so
before you are entitled to the resources of set theory; and that
is what we've only now learned how to do.
Arithmetical structuralism is a different case. Mathemati
cians have long known that the universe of mathematical
objects affords countless that satisfy the structural
conditions for a successor function. These functions are sets
(set theory itself being taken for granted) so there is no prob
lem about quantifying over them. The technical prerequisites
53
The Trouble with Classes
for Ramsifying out the successor function have long been
common knowledge. So it is not anachronistic to say that
arithmetic, as understood among mathematicians, is already
structuralist. (Laymen are another question.) It is even some
what plausible. Or at least it is somewhat plausible to say that
arithmetic is not determinately not structuralist: even if math
ematicians seldom or never avow arithmetical structuralism
at least it fits, well enough, the ways they mostly talk and
think. Not so for settheoretical structuralism. It does not fit
presentday mathematical practice, because that practice does
not include any knowledge of how to Ramsify out the
singleton function.
To some extent, settheoretical structuralism changes the
subject. If we want to examine set theory as we find it now,
we have to concede that it claims a primitive understanding
of membership; and even when the general case of member
ship reduces via mereology to the special case of membership
in singletons, still that special case is primitive. Despite all my
misgivings over the notion of singleton, I am not fully con
vinced that structuralist revolution is the right response. I
want to carryon examining set theory as we find it. There
fore Ilea ve structuralism as unfinished business. Our work
ing assumption for the rest of this book, until the appendix,
shall be that the membersingleton relation is indeed primitive.
2.7 Metaphysics to the Rescue?
We can expect to hear from several systematic metaphysi
cians, each one offering to clarify the notion of singleton by
subsuming it, in his own way, in his own system.
The first of them, of course, is the orthodox set theorist
who tells us that a singleton is a special case of a class. True.
Not helpful.
54
r
Metaphysics to the Rescue?
Next comes the one who tells us that pairing is primitive
and that Possum's singleton is just the pair of Possum and
Possum. False, I guess. Why favour the novel reduction of set
theory to pairing (plus mereology) over the standard reduc
tion of pairing to set theory?
Next comes one who tells us that Possum's singleton is
Possum's haecceity: a special, nonqualitative property that
can belong only to Possum.
12
True, I say, because it's part of
a settheoretical conception of properties  a property is any
class of possibilia, and Possum's singleton is one such class
without which 'nonqualitative property' would be a contra
diction in terms. But this does more to explain 'haecceity'
then 'singleton'.
Next comes one who tells us that Possum's singleton is an
abstraction from several coextensive qualitative properties,
each one had by Possum alone. By 'property' he does not
mean a class of actual and possible instances; by 'abstraction'
he does not mean the taking of equivalence classes. My most
urgent worry is that I have no· guarantee that there is any
qualitative property had by Possum alone. And if there is  if
he is indeed a very special cat  it may be otherwise for other
things that have singletons.
Next comes one who tells us that for each thing that has a
property, there is a particular case of the property, a trope,
that belongs to it alone. If Possum does have an exact dupli
cate, the two cats have duplicate tropes, not identical tropes.
Then maybe Possum's singleton is one of his tropes: his own
12 John Bigelow, The Reality of Numbers: A Physicalist's Philosophy of
Mathematics (Oxford University Press, 1988), section 16, presents a theory
of sets in general as plural haecceities. Thence we have the special case that
a singleton is a singular haecceity. I do not know whether Bigelow could
consistently say that a plural haecceity is a fusion of singular haecceities,
and thereby endorse our Main Thesis. Anyhow, he doesn't.
5 5
The Trouble with Classes
particular selfidentity, maybe, or perhaps his own particular
cathood.
13
But even if we should believe in some tropes, par
ticular selfidentities seem de trop, and so do mereologically
atomic tropes had by thoroughly nonatomic cats.
Last comes one who tells us that the world is the totality of
facts, not of things; that although facts do in some sense have
'constituents', they are composed of these constituents in an
altogether unmereological way, and are  at least sometimes
 mereologically atomic; and that Possum's singleton is to be
identified with some fact about him. Perhaps it should be the
(mereologically atomic) fact that he is selfidentical, or that
he is a cat. Or perhaps it should just be the fact that he is one
single thing, a unit; then unit classes would turn out to be
facts of unithood.
14
Again, even granted an ontology of
facts, I suspect it ought to be too sparse to afford atomic facts
13 On tropes in general, see Williams, 'On the Elements of Being'; and
Campbell, Abstract Particulars. Williams proposed the identification of
singletons with selfidentity tropes in lectures at Harvard circa 1963; but
very tentatively, and as part of a broader plan that conformed to our First
Thesis but not to our Second Thesis. The class of Magpie and Possum, for
instance, would on Williams' plan be the fusion of three tropes: Magpie's
selfidentity (her singleton), Possum's selfidentity (his singleton), and
also the particular nonidentity between Magpie and Possum.
14 See Peter Forrest and D. M. Armstrong, 'The Nature of Numbers',
Philosophical Papers, 16 (1987), pp. 16586; and Armstrong, 'Classes are
States of Affairs', forthcoming in Mind, 100 (1991). The latter paper gives
exactly the proposal now under consideration; the former differs in one
relatively minor way, as follows. Although it takes a class to be a fusion of
some facts (or 'states of affairs') pertaining to its several members, it
allows that the same thing might belong to different classes in virtue of
different facts pertaining to it. If x is an F and also a G, Fx might be the
fact whereby x belongs to the class of Fs, but Gx might be the fact
whereby x belongs to the class of Gs. Then neither fact is once and for all
the singleton that is a common part of both classes.
Credo
about nonatomic things. Further, mereology looks to be the
general theory of composition, not the theory of one special
kind of composition. Therefore I find unmereological 'com
position' profoundly mysterious. After expelling it from set
theory, I scarcely want to welcome it back via the anatomy
of facts.
I am no enemy of systematic metaphysics. Nor do I im
agine that the reservations I've expressed are at all conclusive
against the several proposals. However, I want to pose a chal
lenge to them all.
We were stumped by several questions about singletons.
Where, if anywhere, are they? What is their intrinsic nature?
Do they differ qualitatively from another? Is the relation
of member to singleton founded on the qualitative nature of
the relata, or is it more like a distance relation, or is it a mix
ture, or something else altogether?
Let each of the metaphysicians tell us his answer to these
questions, if indeed his theory of singletons yields answers. If
it does, well and good. Let's see it done. But ifit doesn't, then
his subsumption of singletons has not dispelled their mystery.
Rather, his subsumption shows that the mystery of single
tons also bedevils the classes, or pairs, or haecceities, or ab
stractions from qualitative properties, or tropes, or facts. If
he has shown that, he has not helped us and he has done him
self no favour.
2.8 Credo
Singletons, and therefore all classes, are profoundly mysteri
ous. Mysteries are an onerous burden. Should we therefore
dump the burden by dumping the classes? If classes do not
exist, we needn't puzzle over their mysterious nature. If we
renounce classes, we are set free.
57
The Trouble with Classes
No; for set theory pervades modern mathematics. Some
special branches and some special styles of mathematics can
perhaps do without, but most of mathematics is into set
theory up to its ears. If there are no classes, than there are
no Dedekind cuts, there are no homeomorphisms, there are
no complemented lattices, there are no probability distribu
tions, .... For all these things are standardly defined as one
or another sort of class. If there are no classes, then our math
ematics textbooks are works of fiction, full of false 'theo
rems'. Renouncing classes means rejecting mathematics.
That will not do. Mathematics is an established, going con
cern. Philosophy is as shaky as can be. To reject mathematics
for philosophical reasons would be absurd. If we philos
ophers are sorely puzzled by the classes that constitute math
ematical reality, that's our problem. We shouldn't expect
mathematics to go away to make our life easier. Even if we
reject mathematics gently explaining how it can be a most
useful fiction, 'good without being true'15 we still reject it,
and that's still absurd. Even if we hold onto some mutilated
fragments of mathematics that can be reconstructed with
out classes, if we reject the bulk of mathematics that's
still absurd.
16
15 As in Hartry Field, Science Without Numbers (Princeton University
Press, 1980).
16 Does settheoretical structuralism reject mathematics? It may seem
not: it demands no change in the practice of mathematics, it counts the
theorems as true, it says that classes exist, and it grants that speaking of
singletons and their members makes sense. But it does demand wholesale
reinterpretation. Whether acceptanceonlyunderreinterpretation is tan
tamount to rejection is a vague matter it depends on how drastic the
reinterpretation is. ( myself would not call the structuralist reinterpreta
tion a rejection of mathematics; I do not dismiss it as absurd. Still, it does
challenge the established understanding of mathematics for philosophical
reasons, and all such challenges deserve suspicion.
Credo
That's not an argument, I know. Rather, I'm moved to
laughter at the thought of how presumptuous it would be to
reject mathematics for philosophical reasons. How would
you like the job of telling the mathematicians that they must
change their ways, and abjure countless errors, now that phil
osophy has discovered that there are no classes? Can you tell
them, with a straight face, to follow philosophical argument
wherever it may lead? If they challenge your credentials, will
you boast of philosophy's other great discoveries: that mo
tion is impossible, that a Being than which no greater can be
conceived cannot be conceived not to exist, that it is unthink
able that anything exists outside the mind, that time is unreal,
that no theory has ever been made at all probable by evi
dence (but on the other hand that an empirically ideal theory
cannot possibly be false), that it is a wideopen scientific ques
tion whether anyone has ever believed anything, and so on,
and on, ad nauseam?
Not me! And so I have to say, gritting my teeth, that
somehow, I know not how, we do understand what it means
to speak of singletons. And somehow we know that ordin
ary things have singletons, and singletons have singletons,
and fusions of singletons sometimes have singletons. We
know even that singletons comprise the predominant part of
Reality.
59
3
A Framework for
Set Theory
3.1 Desideratafor a Framework
Let us subdue our scepticism, and have faith in the teachings
of settheoretical mathematics. Let us accept the orthodox
iterative conception of set, including the part of it that es
capes elementary formulation. We want to end with nothing
out of the ordinary.
But let us henceforth begin with singleton rather than mem
bership as the primitive notion of set theory. We can leave it
to mereology to make manymembered classes by fusing
together singletons. And in formulating set theory, let us dis
entangle the part that characterizes the notion of singleton
from the part that is just mereology.l (Let us separate Mr
Hyde from Dr Jekyll.) Before we advance to set theory itself,
1 Bunt, Mass Terms and Model Theoretic Semantics, also formulates set
theory with the partwhole relation and the membersingleton ('unicle')
relation as primitive. But both primitives figure together in his axioms
almost from the beginning; there is no attempt to set up mereology first
and then add on the theory of singletons.
A Framework for Set Theory
we must have in place the framework to which we shall add it.
This framework will be topicneutral, as logic is. It will be
devoid of set theory, as logic is. It will be ontologically inno
cent, as logic is. It will be fully and precisely understood, as
logic is. I would have liked to call it 'logic', in fact. But that is
not its name, and, with names, possession is nine points of the
law.
It will not be quite like logic, though. It will have a lot of
mathematical power, as is shown in the appendix. (I com
mend it to those who dare to cut their mathematics to suit
their philosophy.) In return, it will not admit of complete
axiomatization. So we shall help ourselves to evident princi
ples of the framework as needed, not stopping to make any
official choice among its incomplete fragments. Our interest
in axioms will concern not the framework, but rather the set
theoretical addition.
3.2 Plural Quantification
Besides the elementary logical apparatus of truth functions,
identity, and ordinary singular quantification, our framework
also shall be equipped with apparatus of plural quantification.
We shall have plural pronouns (that is, variables) and plural
quantifiers to bind them. We shall have a copula to link singu
lars with plurals. When we go beyond the vocabulary of
the framework itself, we may have plural terms, and we may
have predicates or functors (defined or primitive) that take
plural arguments. This apparatus is not common in formal
languages, but we know it well as masters of ordinary Eng
lish. Let us remind ourselves of what we know how to say.2
2 In this section I closely follow the lead of George Boolos, 'To Be is To
Be the Value of A Variable (or To Be Some Values of Some Variables),
62
Plural Quantification
A famous example: 'Some critics admire only one another.'
Or in less abbreviated form: 'There are some critics such that
each of them admires only other ones of them.' Or in long
winded regimentation: 'There are some things such that each
of them is a critic; and such that for all x and y, if x is one
of them and x admires y, then y is one of them and y is not
identical to x.'
An example, from Boolos, to illustrate the power of plural
quantification: Napoleon is Peter's ancestor iff, whenever
there are some people such that each parent of Peter is one of
them, and each parent of one of them is one of them, then
Napoleon is one of them.
to show the evident triviality of a principle of
plural 'comprehension': If there is at least one cat, then there
are some things that are all and only the cats. (Regimented:
... then there are some things such that, for all x, x is one of
them iff x is a cat.) Likewise, if there is at least one set, then
there are some things that are all and only the sets. If there is
and 'Nominalist Platonism'. I take it that I agree fully with him on sub
stantive questions about plural quantification, though (as noted later) I
would make less than he does of the connection with secondorder logic.
Other relevant writings on plurals include Bertrand Russell, Principles of
Mathematics (Cambridge University Press, 1903), sections 70 and 74; Max
Black, 'The Elusiveness of Sets', Review of Metaphysics, 24 (1971), pp.
61436; Eric Stenius. 'Sets', Symhese, 27 (1974), pp. 161 88; Adam
Morton, 'Complex Individuals and Multigrade Relations', Nous, 9
(1975), pp. 30918; Gerald J. Massey, 'Tom. Dick, and Harry, and All
the King's Men', American Philosophical Quarterly, 13 (1976), pp. 89 107;
D. M. Armstrong, Universals and Scientific Realism, vol. I, pp. 324;
Richard Sharvy, 'A More General Theory of Definite Descriptions',
Philosophical Review, 89 (1980), pp. 607 24; Peter M. Simons, 'Plural
Reference and Set Theory', in Parts and Moments: Studies in Logic and
Formal Ontology, ed. Barry Smith (Philosophia Verlag, 1982); Stephen
Pollard, 'Plural Quantification and the Iterative Conception of Set',
Philosophy Research Archives, 11 (1986), pp. 57987.
A Framework for Set Theory
at least one nonselfmembered class, then there are some
things that are all and only the nonselfmembered classes.
Exam pIes to illustrate the use of plural predicates: Isaac has
written some books iff there are some things such that he has
written them and they are books. He has written more than
450 books iff, whenever there are some books and they are at
most 450 in number, then he has written some book that is
not one of them. Likewise, he has written uncountably many
books iff, whenever there are some books and they are
countable in number, then he has written some book that is
not one of them. (I presuppose, safely I think, that there exists
at least one book.) The plural 'they are books' and 'Isaac has
written them' reduce instantly to the singular: 'each one of
them is a book', 'Isaac has written each one of them'. The
plural 'they are at most 450' reduces less instantaneously: 'for
some Xl"'" for some X
450
' each of them either is Xl or ...
or is X450" As for 'they are countable', that might reduce
with the aid of apparatus not yet introduced  set theory, for
instance or might remain primitive.
Examples of plural terms: 'The Dubliners', 'The Clancy
Brothers and Tommy Makem', 'Andy Irvine and Paul
Brady', 'the fans of the Chieftains'. The first is a plural proper
name, the rest are eliminable plural definite descriptions. The
fans of the Chieftains are many; for instance, iff there are
some things, and they are many, and, for all x, X is one of
them iff x is a fan of the Chieftains. The Clancy Brothers and
Tommy Makem sing together iff there are some things, and
they sing together, and, for all x, x is one of them iff either x
is one of the Clancy Brothers or x is Tommy Makem.
Examples to deter us from confusing plural quantification
with substitutional 'quantification' over predicates, or with
singular quantification over sets or classes or properties: We
can say without apparent contradiction  and indeed truly
that there are some things such that for no predicate are they
64
Plural Quantification
all and only the satisfiers of that predicate. There are some
things such that for no set are they all and only its members;
and some things such that for no proper class are they all and
only its members; and some things such that for no property
are they all only its instances. .
Indeed, a nominalist well might say, falsely by my lights
but without apparent contradiction, that although there are
some people such that Peter's parents are among them and
Napoleon isn't, and such that the parents of anyone of them
are among them, yet there is no set or class of these people,
and no property that distinguishes them from other.
There are just people, says our nominalist, and other mdlvld
uals, many of them; but no sets or classes or properties.
It is customary to take for granted that plural quantification
must really be singular. Plurals, so it is said, are the
whereby ordinary language talks about classes. Accordmg
to this dogma, he who says that there are the cats can only
mean, never mind that he professes nominalism, that there is
the class of cats. And if you say that there are the nonself
membered classes, you can only mean, never mind that you
know better, that there is the class of nonselfmembered
classes.
If that's what you mean, what you say cannot be true: the
supposed class is a member of itself iff it isn't, so there can be
no such class. (Likewise there cannot be the set of all nonself
membered sets; and if there are classlike whatnots that are
not exactly classes or sets,· there cannot be the whatnot of all
nonselfmembered whatnots.) To translate your seemingly
true plural quantification into a contradictory singular quan
tification is to impute error grave and hidden error.
We may well look askance at this ,Even the
singularist may. He might dodge it ,by some
imaginative interpretation of others words with some
6;
A Framework for Set Theory
imaginative ontology. When you say that there are the non
selfmembered classes, the singularist sa ys:
By dogma, you must mean that there is something that
contams all and only the nonselfmembered classes. But this
something may not be what you ordinarily call a 'class'. I think,
and I say that really you must think so too, that beyond the sets,
and beyond the ordinary proper classes, there are some truly
awesome 'classes'  if I may call them that  which contain even
the proper classes as members. You must have meant that there
is an awesome 'class' that contains all and only the nonself
membered sets and proper classes. That's no contradiction. [
believe it  and you do too.
So you splutter, and eventually manage to say that you are
not now, and never have been, a believer in awesome 'classes'.
He says that you have just affirmed the existence of one
by plural quantification understood according to the
larist dogma.
When the nominalist embraces plural quantification and
repudiates sets, he plays into the singularist's hands. The sin
says: 'Your plural quantification corresponds to my
smgular quantification over sets of individuals. How do you
from me?' When Boolos embraces plural quantifi
catIon and repudiates proper classes, he plays into the singular
ist's hands.
3
The singularist says: 'Your plural quantification
corresponds to my singular quantification over sets and pro
per classes. How do you differ from me?' And when you em
brace plural quantification and repudiate awesome 'classes'
same again. The singularist says: 'Your plural
tlficatlon corresponds to my singular quantification over sets,
3 Michael D. Resnik, 'Secondorder Logic Still Wild', Journal of
PhIlosophy, 85 (1988), pp. 7587.
66
Plural Quantification
proper classes, and awesome "classes". How do you differ
from me?'
The answer is the same every time. Maybe we're right in
our various repudiations, maybe we're wrong. Suppose you
repudiate sets. But there are still individuals, and they are
available to be plurally quantified over. You may truly say
that there are some things that are the cats. If you're right and
there are no sets, then your true existential plural quantifi
cation over individuals cannot be the true existential plural
quantification over sets that the singularist says it is  because
ex hypothesi that singular quantification is false. If you're
wrong and there are sets after all, proceed to the next case.
Suppose there are sets, but you still repudiate proper
classes. If so, sets are available to be plurally quantified over.
You may truly say that there are some things that are all the
sets, or that there are some sets with no bound on their ranks,
or that there are all the sets that are nonselfmembers. But
there is no set of these sets. If you're right and there are no
proper classes, then your true existential plural quantification
over sets cannot be the true singular quantification over pro
per classes that the singularist says it is because ex hypothesi
that singular quantification is false. If you're wrong and there
are proper classes after all, proceed to the next case.
Suppose there are proper classes, but you still repudiate
awesome' classes'. If so, proper classes are available to be plu
rally quantified over. You may truly say that there are some
proper classes, but there is no set or proper class that has any
of them as members; or that there are some things that are all
the sets and proper classes that are nonselfmembers. If
you're right and there are no awesome 'classes', then your
true plural quantification over proper classes cannot be the
true singular quantification over awesome 'classes' that the
singularist says it is because ex 11ypothesi that singular
A Framework for Set Theory
quantification is false. If you're wrong and there are awe
some 'classes' after all, proceed to the next case.
Suppose there are awesome 'classes'; then they are avail
able to be plurally quantified over; you can truly say that
there are some things that are all and only the sets and proper
classes and awesome 'classes' that are nonselfmembers.
Maybe the singularist says that way out beyond even the
awesome 'classes' ....
And so it goes. But let's cut a long story short. Whatever
classlike things there may be altogether, holding none in
reserve, it seems we can truly say that there are those of them
that are nonselfmembers. Maybe the singularist replies that
some mystical censor stops us from quantifying over absol
utely everything without restriction. Lo, he violates his own
stricture in the very act of proclaiming it!
We embrace plural quantification over all the things there
even those, if any, that we may now be wrongly repu
dIatmg. How do we differ from the singularist, and from
ourselves as he interprets us? We differ about what's true
under anyone fixed hypothesis about what there is and what
there isn't.
. ,!,he singularist dogma has the weight of common opinion
m Its favour. Quantification would be a simpler affair if it
were true. Those are two weighty reasons to believe it. But
not weighty enough. It imputes deviousness where there is
no deviousness. It imputes ontological commitment where
there is no ontological commitment. It imputes falsehood
where there is no falsehood, and contradiction where there is
no contradiction  unless instead it imputes restriction where
there is no restriction, and evermoreextravagant ontologi
cal commitment. Reject it, therefore. Plural quantification is
irreducibly plural. It is not ordinary singular quantification
over special plural things  not even when there are special
68
Plural Quantification
plural things, namely classes, to be had. Rather, it is a special
way to quantify over whatever things there may be to quan
tify over. Plural quantification, like singular, carries onto
logical commitment only to whatever may be quantified over.
It is devoid of set theory and it is ontologically innocent.
It may help to break the grip of singularism if we see how
some plural quantification can be translated without dragging
in any sets, classes, or properties. In the case of the prolific
author, we quantified plurally over books. Let us eliminate
the plurals and still quantify just over books. 'Whenever
there are at most 450 books, some book by Isaac is not
among them' becomes 'For any book Xl' for any book X
2
,
... , for any book X
450
, some book by Isaac is not Xl and not
X
2
and ... and not X450" One plural quantifier over books
gives way to 450 singular quantifiers over books. The next
case is the same, but this time we need a language that allows
infinite blocks of quantifiers and infmite conjunctions. 'When
ever there are countably many books, some book by Isaac is
not among them' becomes 'For any book Xl' for any book
X
2
, • •• , some book by Isaac is 'not Xl and not X
2
and ... '
where this time the ellipses abbreviate infinite sequences.
Can we say that plural quantifiers in general are abbre
viations for long blocks of singular quantifiers, so long that
their limited length never imposes any numerical restriction?
I fear not. Before long, we'll have to concede that the abbre
viated expressions exist only as settheoretical sequences, and
then I do not think we may mention them in explaining
apparatus that purports to be devoid of set theory. What's
worse, we'll eventually have to concede that the abbreviated
expressions are too long to exist at all, even as sequences. And
then they just don't exist, so we have no right to mention
them at all. The reduction to quantifierblocks is instructive,
A Framework for Set Theory
but yields only a skimpy fragment of plural quantification.
The rest must remain primitive.
4
Boolos identifies plural quantification with secondorder quan
tification: monadic, full secondorder quantification, with a
restriction against vacuous variables. That will do to tell the
experts what a formal language with plural quantification
looks like, and what it can do. But I fear the identification
may be misleading, though in no way has it misled Boolos
himself. So I prefer to play it down.
(1) In so far as we already regarded secondorder logic as
set theory (or property theory) in sheep's clothing, it encour
ages singularism.
(2) It hints that the third, fourth, and higher orders can
not be far behind but what mjght plurally plural quantifi
cation be? (Infinite blocks of plural quantifiers?  That will
4 Morton, 'Complex Individuals and Multigrade Relations', treats plural
quantifiers as blocks of singular quantifiers safe enough, if he has set
theory at his disposal and yet applies this treatment only to plural quan
tifiers over individuals, and if there are not too many individuals.
He joins this with the idea that we could analyse statements ostensibly
about composite entities in terms of plural quantification over atoms
alone. Then if we liked Morton does not advocate it! we could go on
to deny that composition ever takes place. We could claim that 'the cat is
on the mat' is true in virtue of the arrangement of many atoms, though
strictly speaking there are no such composite entities as cats and mats. But
even if we could save the truth of everyday statements, still this would be
far from believable: T are many atoms, not one composite thing. Fur
ther, it leaves us vulnerable to an argument akin to Malezieu's (see David
Hume, A Treatise of Human Nature (1739), book I, part II, section II): if all
composite entities are to be eliminated in favour of their parts, then we
are forced to deny a priori that there can be such a thing as atomless gunk;
because if there were, its parts in turn would be composite ad infinitum.
Further, and fatally for purposes of my present project, it gives us no way
to quantify plurally over mereological fusions.
Choice
be only a skimpy third order, and no start at all on the
fourth.)
(3) It also hints that polyadic sec.ondorder quantifica
tion cannot be far behind. But, in fact, that takes additional
resources. We need plural quantification over ordered pairs,
triples, etc.; and we need some method of pairing. Even the
methods of the appendix, which do without set theory (and
primitive pairing), still require mereology and an infinite
supply of atoms.
(4) Secondorder quantification, unless especially restric
ted against vacuity, corresponds not to 'there are one or more
things' but rather to 'there are zero or more things'. Because
of this mismatch, regimented translations from unrestricted
secondorder logic into English come out abominably convo
luted. Certainly there's something that's hard to understand!
The hasty reader may think it's the plural quantification
that's to blame.
3.3 Choice
Even before we reach mereology, let alone the settheoretical
notion of singleton, we have some of the power that's nor
mally credited to set theory. We have already seen how to
take ancestrals. Also, we can state  and should, of course,
affirm  versions of the Axiom of Choice. Here are two.
Both are schemata. To affirm them is to affirm whatever sen
tences can be made by filling in the blank, using any vocabu
lary that may later come to hand (adjusting the number of
the verb), and then prefixing universal quantifiers, singular
or plural, to bind any free variables.
First Choice Schema: If there are some things, and each of
them ... some things, and no two of them ... the same
A Framework for Set Theory
things, then there are some things such that each of the
former things ... exactly one of the latter things.
Second Choice Schema: If nothing ... itself, and if when
ever x . .. y and y . .. z then x . .. z, and if there are
some things such that each of them '" another one of
them, then also there are some of those things such that
(1) among the latter things also, each one ... another
one; (2) whenever x and y are two of the latter things,
then either x . .. y or y . .. x.
The Second Choice Schema is a mouthful, but it says roughly
this: if there is a partial ordering with no last term, there is
a linear subordering of it still with no last term. But only
roughly; because the Schema carries no commitment to any
such entities as orderings.
3.4 Mereology
To finish our framework, we add the apparatus of mereo
logy.
5
We have a choice of primitives. We could begin with
'part' and go on as follows.
5 Mereology was developed by Stanislaw LeSniewski in several papers in
Polish in 1916 and in 19271931: 'Podstawy Og61nej Teoryi Mnogoki
I', Prace Polskiego Kola Naukowego w Moskwie: Sekcya Matematyczno
przyrodnicza, 2 (1916); and '0 Podstawach Matematyki'. Przeglad
Filozojiczny, 30 (1927), pp. 164206; 31 (1928), pp. 26191; 32 (1929),
pp. 60101; 33 (1930), pp. 77105; and 34 (1931), pp. 14270. The
1927 1931 papers appear in abridged English translation as 'On the
Foundations of Mathematics', Topoi, 2 (1983), pp. 352.
LeSniewski's mereology is founded upon, and includes, his Ontology.
In Ontology there is a kind of quantification that can be interpreted,
plausibly but without benefit of direct textual support, as plural; and
Definitions: x and y overlap iff they have some common
part. Iff not, they are (entirely) distinct.
(Those who hijack the word 'distinct' to mean merely 'non
identical' may say 'disjoint' to mean 'nonoverlapping'.)
Definition: Something is afusion of some things iff it has
all of them as parts and has no part that is distinct from
each of them.
Or we could begin with 'overlaps', 'distinct', or 'fusion'
instead.
there is a copula that can be read 'is one of'. (Here the plural subsumes
not only the singular but also the empty: 'some things' must mean 'some
zero or one or more things'.) See P. M. Simons, 'On Understanding
LeSniewski', History and Philosophy of Logic, 3 (1983), pp. 16591. Under
this pluralist reading, LeSniewski's original version of mereology includes
the whole of my 'framework'.
(Beware: LeSniewski thinks of mereology as affording an interpretation
of the language of set theory. His term 'collective class' corresponds to
our term 'fusion', not to the term 'class' in any sense now current.)
For other expositions of mereology in English, see A. N. Prior, Formal
Logic (Oxford University Press, 1955), section [[[.4; Alfred Tarski, 'Foun
dations of the Geometry of Solids' in Logic, Semantics, Metamathematics
(Oxford University Press, 1956); Tarski's Appendix E in]. H. Woodger,
The Axiomatic Method in Biology (Cambridge University Press, 1937);
Nelson Goodman and Henry Leonard, 'The Calculus of Individuals and
Its Uses', Journal of Symbolic Logic, 5 (1940), pp. 4555; Goodman, The
Structure of Appearance, chapter II; Rolf A. Eberle, Nominalistic Systems
(Reidel, 1970); and, for a very thorough survey, Peter Simons, Parts: A
Study in Ontology (Oxford University Press, 1987), chapters 1 and 2.
If mereology is the general theory of part and whole, then it is a frag
ment of the 'ensemble theory' of Bunt, Mass Terms and Model Theoretic
Semantics. Bunt himself, however, reserves the term 'mereology' for a
restricted theory of part and whole, applying only to individuals (see pp.
712,288301).
73
A Framework for Set Theory
Definitions: x is part of y iff everything that overlaps x
also overlaps y; or iff everything distinct from y also is
distinct from x; or iff y is a fusion of x and something z.
The basic axioms of mereology, besides whatever it may
take to close the circle of interdefinition, are three.
Transitivity: If x is part of some part of y, then x is part
ofy.
Unrestricted Composition: Whenever there are some things,
then there exists a fusion of those things.
Uniqueness of Composition: It never happens that the
same things have two different fusions.
In view of the second and third axioms, we are entitled to say
'the fusion of', and rely on it to be a functor defined for any
plural argument whatever.
These axioms do not settle all questions that can be raised
in the language of mereology. Does Reality consist entirely
of atomless gunk? Entirely of atoms? Or some of each?
When we add set theory, then we shall conclude that there
are atoms  namely, the singletons but at this point we
don't say. It will remain undecided whether there is gunk.
And if there are atoms, how many are there? Using the re
sources of elementary logic, we can express such hypotheses
as that there are more than seven, or that there are exactly
seventeen, or that there are fewer than eighty. Using the
resources of plural quantification, as we shall soon see, we can
express hypotheses that discriminate between infinite sizes.
But the basic axioms of mereology are silent about which of
these hypotheses are true.
74
Strife over Mereology
3.5 Str!fe over Mereology
I myself take mereology to be perfectly understood, unprob
lematic, and certain. This is a minority opinion. Many phi
losophers view mereology with the gravest suspicion.
Sometimes they suspect that originally the notion of part
and whole was understood not as topicneutral, but rather as
a spatiotemporal or merely spatial notion. They conclude
that any application of it to things not known to be in space
and time is illicit. The original idea, supposedly, was that x is
part of y iff Y is wherever x is.
That is wrong thrice over. In the first place, it is common
enough, and not especially philosophical, to apply mereology
to things without thinking that they are in space and time.
Trigonometry is said to be part of mathematics; God's
foreknowledge is said to be part of His omniscience. In the
second place, the proposed definition can go wrong even for
things that are in space and time. Suppose two angels dance
forever on the head of one pin. At every moment, each
occupies the same place as the other. Still, they are two dis
tinct proper parts of the total angelic content of their shared
region. Suppose it turned out that the three quarks of a pro
ton are exactly superimposed, each one just where the others
are and just where the proton is. (And suppose the three
quarks each last just as long as the proton.) Still the quarks are
parts of the ptoton, but the proton is not part of the quarks
and the quarks are not part of each other. Suppose a material
thing occupies a region of substantival spacetime; it does not
follow (though it just might be true) that the region is part of
the thing, still less that the region and the thing are parts of
each other and therefore identical. In the third place, the
proposed definition presupposes a prior notion of part and
75
A Framework for Set Theory
whole, because 'wherever x is' had better mean 'wherever
some part of x is'.
Sometimes, philosphers' suspicions against mereology
amount to guilt by association. Not just any controversial
thesis that uses mereological notions and wins the adherence
of some mereologists is a thesis of mereology per se. (1) Mere
ology is silent about whether all things are spatiotemporal.
(2) It is silent about whether spatiotemporal things may have
parts that occupy no less of a region than the whole does
for instance, about whether the quarks that make up the
proton might be exactly superimposed; or about whether
universals in rebus, wholly present wherever they are instan
tiated, might be among, the proper parts of ordinary particu
lars; or about whether, alternatively, ordinary particulars are
bundles of particular tropes, so that an electron would be
composed of its chargetrope, its masstrope, and a few
more.
6
(3) Mereology is silent about whether something
wholly present in one region may also be wholly present in
another. For better or worse it does not forbid recurrent
universals, or enduring things wholly present at different
times, or a singleton atom that is where its extended member
is by being at every point of an extended region, or the un
divided omnipresence of God. (4) Finally, if something oc
cupies a region, mcreology per se does not demand that each
part of the occupied region must be occupied by some part 
proper or improper of the occupying thing. If not, that's a
second way for a singleton atom to be where its extended
member is.
Sometimes temporal or modal problems about mereology
6 On universals in reblH, see Armstrong, Universals and Scientific Realism;
or the main system of Goodman, The Structure oj Appearance. On tropes. see
Williams, 'On the Elements of Being'; and Campbell, Abstract Particulars.
Strife over Mereology
arouse suspicion. It seems that something and its proper part
can be first different and later identical, or actually different
but possibly identical. Then from the standpoint of the later
time, or the un actualized possibility, it seems we have one
thing that used to be, or could have been, not identical to
itself!
I have discussed these problems elsewhere.
7
Briefly, they
are instances of broader problems about temporary and acci
dental instrinsic properties. Those problems are not the wages
of mereology. They arise for everyone. The temporal prob
lems are best solved by accepting the doctrine of temporal
parts. Other solutions (equally consistent with mereology)
do exist, but have repugnant consequences, for instance, that
I am a changeless soul, so not at all the sort of thing I think I
am. The modal problems are best solved by denying the
overlap of possible worlds. (Apart from shared universals, if
such there be.) We can still say that our Hubert Humphrey
he himself! wins the presidency at another world. But this
can't mean that he himself is part of that world as well as
this one. What can it mean?  Two theories explain it with
out overlap. My own theory says that Humphrey wins at
another world by having a winning counterpart there, a
counterpart not identical to Humphrey himself. A more
popular rival theory says that possible worlds are abstract
representations; some of them misrepresent our Humphrey
as a winner, thereby ascribe to him the property of winning,
and thereby confer upon him the property of winning at
(better, winning according to) that other world. When a
possible world misrepresents Humphrey as winning, no
7 Lewis, On the Plurality oj Worlds, chapter 4. For related discussions see
Lewis, 'Counterparts of Persons and Their Bodies', Journal oj Philosophy,
68 (1971), pp. 20311; and 'Rearrangement of Particles: Reply to Lowe',
Analysis, 48 (1988), pp. 6572.
77
A Framework for Set Theory
Humphreyasmisrepresented exists as part of that possible
world, or anywhere else. What's true is that he exists accord
ing to that possible world.
Sometimes it is the axiom of Uniqueness of Composition
that arouses suspicion. It says there is no difference without a
differencemaker: x and yare identical unless there is some
thing to make the difference between them by being part
of one but not the other. Against this it might be said, for
instance, that the two words 'master' and 'stream' are made
of the same six letters. No; various things may happen, but
none of them is a violation of the axiom. (1) There is a
'master' inscription and there is a 'stream' inscription, and each
one is made of six letterinscriptions. Then the 'a' in 'master'
and the 'a' in 'stream' are not the same, though they are alike
in shape. (2) The two words might be written crossword
fashion so that they share the same 'a'inscription; but then
the 'm' inscriptions of the two words would not be the same.
(3) One single inscription, made of six letterinscriptions,
might be both a 'stream'inscription and also a 'master'in
peculiarorderinscription. (4) The letters might be movable
pieces of plastic, formed first into a 'master' inscription and
then rearranged into a 'stream'inscription. Then the two
wordinscriptions are made of different temporal parts of the
same six persisting letterinscriptions. (5) Suppose, as I think
wrongly, that the six plastic letters do not have temporal
parts. Instead they persist by enduring identically, wholly
present at different times. Then the unique thing composed
of the six letters is a 'master'inscription relative to one time
and a 'stream' incription relative to another. There's no kind
of inscription that it is simpliciter; nor, on this supposition, do
we have two different inscriptions of the two words. (6) Sup
pose the wordtypes 'master' and 'stream' are taken to be
each the fusion of all its inscriptions; and likewise for letter
types. (To avoid confusion, we'd better take only those
7
8
r
Strife over Mereology
inscriptions which contrast with their surroundings.) Then the
two wordtypes are each made of parts of the same six
lettertypes; but not the same parts. (7) Suppose instead that
the wordtypes are taken as settheoretic constructions out of
the six lettertypes sequences, perhaps. Then indeed two
things are generated settheoretically out of the same six
members. It is just this power to make the most of limited
material that makes settheoretic constructions worthwhile.
But they are not cases of composition, so I claim, because
what is going on is not just composition. Partly it is compo
sition; partly it is the making of singletons, which makes one
thing out of one thing, not out of many. I conclude that
words and their letters, and parallel examples, are no threat
to Uniqueness of Composition. 8
Most of all, it is the axiom of Unrestricted Composition
that arouses suspicion. I say that whenever there are some
things, they have a fusion. Whenever! It doesn't matter how
many or disparate or scattered or unrelated they are. It
doesn't matter whether they are all and only the satisfiers of
some description. It doesn't matter whether there is any set,
or even any class, of them. (Here's where plural quantification
S But what of another way of taking the example? (8) A wordtype is a
structural universal instantiated by the inscriptions; it is made of simpler
structural universals, namely the six lettertypes together with rclations of
juxtaposition; these relations, as well as the lettertypes, are involved alike
in both wordtypes. This really would go against the principle no mis
understanding this time. But to accept this as a counterexample, we must
first accept a theory of structural universals which I take to be far from
established. Rather than taking the case as an objection to the principle of
Uniqueness of Composition, I go the other way: I object to structural
universals exactly because they require some sort of unmcreological
'composition' that violates the principle. See David Lewis, 'Against
Structural Universals', Australasian Journal of Philosophy, 64 (1986), pp.
2546.
79
A Framework for Set Theory
pays its way, for better or worse.) There is still a fusion.
So I am committed to all manner of unheardof things:
troutturkeys, fusions of individuals and classes, all the
world's styrofoam, and many, many more. We are not
accustomed to speak or think about such things. How is it
done? Do we really have to?
It is done with the greatest of ease. It is no problem to
describe an unheardof fusion. It is nothing over and above
its parts, so to describe it you need only describe the parts.
Describe the character of the parts, describe their interre
lation, and you have ipso facto described the fusion. The
troutturkey in no way defies description. It is neither fish
nor fowl, but it is nothing else: it is part fish and part fowl. It
is neither here nor there, so where is it?  Partly here, partly
there. That much we can say, and that's enough. Its character
is exhausted by the character and relations of its parts.
I never said, of course, that a troutturkey is no different
from an ordinary, muchheardof thing. It is inhomogene
ous, disconnected, and not in contrast with its surroundings.
(Not along some of its borders.) It is not cohesive, not caus
ally integrated, not a causal unit in its impact on the rest of
the world. It is not carved at the joints. But none of that has
any bearing on whether it exists.
If you wish to ignore it, of course you may. Only if you
speak with your quantifiers wide open must you affirm the
troutturkey's existence. If, like most of us all the time and all
of us most of the time, you quantify subject to restrictions,
then you can leave it out. You can declare that there just does
n()t exist any such thing  except, of course, among the things
you're ignoring.
Doing away with queer fusions by restricting composition
cannot succeed, unless we do away with too much else be
sides. For many respects of queerness are matters of degree.
But existence cannot be a matter of degree. If you say there
80
r
Composition as Identity
is something that exists to a diminished degree, once you've
said 'there is' your game is up. Existence is not some special
distinction that befalls some of the things there are. Existence
just means being one of the things there are, nothing else. The
fuzzy line between less queer and more queer fusions cannot
possibly coincide with the sharp edge where existence gives
out and nothing lies beyond. A restriction on your quanti
fiers, on the other hand, may be as fuzzy as you please.
9
3.6 Composition as Identity
So I claim that mereology is legitimate, unproblematic, fully
and precisely understood. All suspicions against it are mis
taken. But I claim more still. Mereology is ontologically
innocent.
To be sure, if we accept mereology, we are committed
to the existence of all manner of mereological fusions. But
given a prior commitment to cats, say, a commitment to
catfusions is not afurther commitment. The fusion is nothing
over and above the cats that compose it. It just is them. They
just are it. Take them together or take them separately, the
cats are the same portion of Reality either way. Commit
yourself to their existence all together or one at a time, it's
the same commitment either way. If you draw up an inven
tory of Reality according to your scheme of things, it would
be double counting to list the cats and then also list their
fusion. In general, if you are already committed to some
things, you incur no further commitment when you affirm
9 For a fuller statement of this argument against restricting composition,
see Lewis, On the Plurality of Worlds, pp. 21213. For a rejoinder, see
Peter van Inwagen, 'When are Objects Parts?', Philosophical Perspectives, 1
(1987), pp. 2147, especially pp. 405.
8r
A Framework for Set Theory
the existence of their fusion. The new commitment is redun
dant, given the old one.
For the most part, if you are committed to the existence of
a certain thing or things, and then you become committed to
the existence of something that bears a certain relation to it
or them, that is indeed a further commitment. If you incur a
commitment to the of, or the nextdoor neighbour of,
or the weight in grams of, or the shadow of, or the single
ton of, something you were committed to already, you have
made a further commitment. It is not redundant. But the
relation of identity is different. If you are already committed
to the existence of cat Possum, and then affirm that there
exists something identical to Possum, that is not a further
commitment. I say that composition  the relation of part
to whole, or, better, the manyone relation of many parts to
their fusion  is like identity. The 'are' of composition is, so
to speak, the plural form of the 'is' of identity. Call this the
Thesis of Composition as Identity. It is in virtue of this thesis
that mereology is ontologically innocent: it commits us only
to things that are identical, so to speak, to what we were
committed to before.
In endorsing Composition as Identity, I am following the
lead of D. M. Armstrong and Donald Baxter. Armstrong
takes strict identity and strict difference as the endpoints of a
spectrum of cases, with cases of more or less extensive over
lap in between. Overlap subsumes partwhole as a special
case: it may be x itself, or y itself, that is a common part of x
and y. Two adjoining terra_ce houses that share a common
wall
are not identical, but they are not completely distinct from each
other either. They are partially identical, and this partial identity
takes the form of having a common part. Australia and New
South Wales are not identical, but they are not completely
82
Composition as Identity
distinct from each other. They are partially identical, and this
partial identity takes the form of the wholepart 'relation' ....
Partial identity admits of at least roughandready degree.
Begin with New South Wales and then take larger and larger
portions of Australia. One is approaching closer and closer to
complete identity with Australia.
1o
Baxter puts it this way:
The whole is the many parts counted as one thing. On this view
there is no one thing distinct from each of the parts which is the
whole. Rather, the whole is simply the many parts with their
distinctness from each other not mattering. This ... is not to
deny the existence of the whole. It is merely to deny the
additional existence of the whole ....
Suppose a man owned some land which he divides into six
parcels. Overcome with enthusiasm for [the denial of Com
position as Identity I he might try to perpetrate the following
scam. He sells off the six parcels while retaining ownership of
the whole. That way he gets some cash while hanging on to his
land. Suppose the six buyers of the parcels argue that they
jointly own the whole and the original owner now owns
nothing. Their argument seems right. But it suggests that the
whole was not a seventh thing.ll
Indeed! I grant that no one of the six parcels by itself is ident
ical to the original block of land. Still, there is a good sense in
which the six parcels and the original block are the very same
thing. They are it, it is them. You can't sell them without sell
ing it, because you can't sell it without selling it.
10 Armstrong, Universals and Scientific Realism, vol. II, pp. 37  8.
11 Donald Baxter 'Identity in the Loose and Popular Sense', Mind, 97
(1988), p. 579. Also see Donald Baxter, 'ManyOne Identity', Philosophi
cal Papers, 17 (1988), pp. 193216.
A Framework for Set Theory
A doubter might seek to trivialize Composition as Ident
ity, thus:
Of course the terrace houses are partially identical, and so are
New South Wales and Australia, if by that you just mean that
something that is part of one is identical to something that is
part of the other. Of course the six parcels are identical to the
original block, if by that you just mean that the fusion of the
parcels is identical to the block. But that shows nothing about
the innocence of mereology. You could do the same trick with
any old relations. Possum is 'maternally identical' to Magpie, if
by that you just mean that something that is mother of one is
identical to something that is mother of the other. The six
numbers 2, 8, 4, 7, 6 and 3 are 'on average' identical to the one
number 5, if by that you just mean that the average of the six is
identical to the one. So what?
But the doubter misses the point. Never mind that mereo
logical relations can be equivalently restated so as to drag in
identity. Anything can be, as he said. The.real point is that
the mereological relations (however restated) are something
special. They are unlike the samemother relation or the
a verageof relation. Rather, they are strikingly analogous to
ordinary identity, the oneone relation that each thing bears
to itself and to nothing else. So strjking is this analogy that it
is appropriate to mark it by speaking of mereological rela
tions the manyone relation of composition, the oneone
relations of part to whole and of overlap  as kinds of identity.
12 Here I part company with Baxter. Composition as Identity comes in
different versions. My version, as just stated, is analogical. A stronger, and
stranger, version would disdain mere analogy. It would insist that there is
just one kind of identity, the ordinary oneone kind, and that compo
sition involves this kind of identity. It is this view, or something very like
it, that Baxter defends. His ingenious defence makes better sense of the
view than I'd have thought possible; but in the end I'm unconvinced, and
so rest content with mere analogy.
Composition as Identity
Ordinary identity is the special, limiting case of identity
in the broadened sense.
12
The analogy has many aspects. Four of them have come to
our attention already. The fust is the ontological innocence
of mereology: just as it is redundant to say that Possum exists
and something identical to him exists as well, so likewise it
is redundant to say that Possum and Magpie both exist and
their fusion exists as well.
Unrestricted composition is a second aspect of the analogy.
If Possum exists, then automatically something identical to
Possum exists; likewise if Possum and Magpie exist then au
tomatically their fusion exists. just as Possum needn't satisfy
any special conditions in order to have something identical to
him, so Possum and Magpie needn't satisfy any special condi
tions in order to have a fusion.
Uniqueness of Composition is a third aspect of the anal
ogy. Just as there cannot be two different things both ident
ical to Possum, likewise there cannot be two different fusions
of Magpie and Possum. A kind of transitivity applies, with
Magpie and Possum together as a plural middle term. If x is
them, and they are y, then x is y.
The ease of describing fusions is a fourth aspect of the
analogy. Describe Possum fully, and thereby you fully de
scribe whatever is identical to Possum. Describe Magpie and
Possum fully  the character of each, and also their inter
relation  and thereby you fully describe their fusion. Like
wise for relational description: specify the location of Magpie
and Possum, and thereby you specify the location of their
fusion. Specify the present ownership of the six parcels and
thereby you specify the ownership of the original block of
land.
A fifth aspect of the analogy has to do with multiple
location. If Mary's lamb goes everywhere that Mary goes,
and if this is so not just as a matter of fact but as a matter of
85
A Framework for Set Theory
absolute necessity, we have a highly mysterious necessary
connection between distinct existences. But if it turns out
that Mary and the lamb are identical, then there is no mys
tery at all about their inseparability. Likewise if it turns out
that the lamb is part of Mary, and if Mary is wholly present
wherever she goes, then again the inseparability is automatic,
and in no way mysterious. In general, if a fusion is multiply
located wholly present at different places or times or poss
ible worlds then also its parts must be multiply located.
Since it just is them, it cannot appear anywhere without
them.
It would be nice to illustrate this by example. Perhaps I
can, if there are multiply located universals. If a conjunctive
universal P & Q is the fusion of its conjuncts P and Q,13 then
it is automatic and unmysterious that both P and Q must
appear wherever P & Q does.
As for alleged examples that involve multiply located
particulars, instead of universals, I definitely disbelieve in
those. To be sure, the same road may have a lane on the hilly
stretch that it lacks on the plain, the same man may have
more teeth when young than when old, or he may actually
have more toes than he might have had. But to these cases I
apply the principle contrapositively. Because the thing varies
with respect to its parts, I conclude that we do not have
genuine multiple location. It is not the same thing wholly
13 Armstrong, Universals and Scientific Realism, vol. rr, p. 36, says that P is
part of P & Q. There is a problem: the fusion of P and Q exists whether or
not P and Q are coinstantiated, but according to Armstrong we have no
conjunctive universal P & Q unless P and Q are coinstantiated. Does this
show that conjunctive universals cannot be the fusions of their conjuncts?
No; it only shows that if P and Q are not coinstantiated, their fusion is not
in that case a universal. But we can perfectly well say that whether the
fusion is a universal or not depends on whether P and Q are coinstan
tiated.
86
Composition as Identity
present at multiple locations. It is the same thing partially
present, or different things wholly present.
This completes the analogy that I take to give the meaning
of Composition as Identity. But alas, it has its limits. In the
first place, I know of no way to generalize the definition of
ordinary oneone identity in terms of plural quantification.
We know that x and yare identical iff, whenever there are
some things, x is one of them iff y is one of them. But if y is
the fusion of the xs, then there are some things such that each
of the xs is one of them and y is not; and there are some
things such that y is one of them but none of the xs is. And in
the second place, even though the many and the one are the
same portion of Reality, and the character of that portion is
given once and for all whether we take it as many or take it
as one, still we do not really have a generalized principle of
indiscernibility of identicals. It does matter how you slice it 
not to the character of what's described, of course, but to the
form of the description. What's true of the many is not
exactly what's true of the one. After all they are many while
it is one. The number of the many is six, as it might be,
whereas the number of the fusion is one. And the singletons
of the many parts are wholly distinct from the singleton of
the one fusion. That is how we can have set theory.
Plural quantification is innocent: we have many things, we
speak of them as many, in no way do we mention one thing
that is the many taken together. Mereology is innocent in a
different way: we have many things, we do mention one
thing that is the many taken together, but this one thing is
nothing different from the many. Set theory is not innocent.
Its trouble has nothing to do with gathering many into one.
Instead, its trouble is that when we have one thing, then
somehow we have another wholly distinct thing, the single
ton. And another, and another, ... ad infinitum. But that's the
price for mathematical power. Pay it.
87
A Framework for Set Theory
3.7 Distinctions of Size
Our framework, as it turns out, affords the means to define
distinctions in the size of things. up to a point, of course, this
can be done in elementary logic: using identity, we can de
fine 'thing with more than seventeen atoms' and the like. But
with the resources of plural quantification and mereology,
we can do much more.
We begin with the distinction between finite and infi
nite things. An infinite thing is one with infinitely many
parts; in that case, its parts will admit of an endless ordering;
a partial ordering will do; and we have one ready to hand,
namely the relation of part to whole.
Definitions: x is infinite iff x is the fusion of some things,
each of which is a proper part of another. Otherwise x is
finite.
Note that any bit of atomless gunk is infinite: it is the fusion
of all its proper parts, each of which is a proper part of an
other. Likewise anything that has an atomless part is infinite.
A thing that consists entirely of atoms, .on the other hand,
may be either finite or infinite. An obvious consequence of
the definition is that an atom is finite.
There are other wa ys of defining infinity. One is Dedekind' s:
when we have infinitely many things, then all those things
correspond oneone with only some of those things. This
may seem useless until we have ordered pairs. Useless as a
definition, yes; but at least we may affirm it as a principle of
the framework.
Dedekind Schema: If x is a proper part of y, and if each
atom of y . .. exactly one atom of x, and if each atom of
88
Distinctions of Size
x is such that exactly one atom of y . .. it, then y is infi
nite.
(Note that if x includes all the atoms of y, leaving out only
some atorriless gunk, then we do not know that y has infi
nitely many atoms. But then y is infinite because of the
gunk.)
Next, a relative distinction. We may take all the atoms of
Reality, however many of them there may be, to set a stan
dard of size. 'Size' here is measured by atoms; atomless gunk,
if there is any, won't count. Roughly, I want to say that
something is 'large' iff it has as many atoms as there are in all
of Reality; otherwise 'small'. Without benefit of ordered
pairs, I cannot say quite that. But I can say that something has
at least as many atoms as there are in all the rest of Reality;
which comes to the same thing, except in the special case
contrary to set theory, and to any physics that posits points of
the spacetime continuum  that Reality has only finitely
manyatoms.
14
Definitions: x is large iff there are some things such that
(1) no two of them overlap, (2) their fusion is the whole
of Reality, and (3) each of them exactly one
atom that is part of x and at most one other atom.
Otherwise x is small.
It follows immediately that any part of a small thing is small.
We may affirm principles of the framework that connect
14 A price we pay for not saying just what we want to say concerns the
smallness of atoms. We'd want it to come out that a single atom is small,
unless it is the only atom; in other words, that an atom is small iff there
are two or more atoms. But instead it turns out that an atom is small iff
there are three or more atoms.
A Framework for Set Theory
the distinction of infinite versus finite to the distinction of
large versus small. If there is something infinite that consists
entirely of atoms, then (1) any finite thing is small, and in
deed (2) any fusion of a small thing with a finite thing is
small. (The small thing may itself be finite, and if so the fu
sion is finite too; or the small thing may be small and infinite,
and if so the fusion has the same small infinite size.)
We may also affirm, as a principle of the framework, this
Replacement Schema (singular version): If each atom of a
thing x . .. exactly one atom of a thing y, and if for each
atom of y there is an atom of x that ... it, and if x is
small, then y is small.
(That is, again, we affirm each sentence made by filling in the
blank and prefixing universal quantifiers as required.) Set
theorists beware: the schema is not set theoretical, and it in no
way constrains the size of Reality.
I would like to introduce plural predicates 'few' and
'many', corresponding to the singular predicates 'small' and
'large', where again the atoms in all of Reality set the stan
dard.
15
But without ordered pairs, I cannot do this in full
generality; so my definition applies only to things whose
fusion does not fill up too much bf Reality. (Two halves of
Reality, though only two, won't count as few!)
15 Beware; the plural distinction of few and many differs in one way
from the singular distinction of small and large. Unless there are but
finitely many atoms, 'large' is a single size: all large things are equal, each
having as many atoms as there are in all of Reality. But 'many' is not a
single size. There are many atoms, and many fusions of atoms, but there
are more of the latter than of the former. The atoms are barely many, the
fusions of atoms are more than barely many.
Distinctions of Size
Definition: Suppose we have some things such that some
large thing does not overlap any of them. Then they
are few iff there is some small thing x, and there are
some things, such that (1) x does not overlap the fusion
of the former things, (2) each of the latter things is the
fusion of one of the former things and one atom of x,
(3) for each of the former things, one of the latter things
is the fusion of it and one atom of x, and (4) no atom of
x is part of two or more of the latter things. Otherwise
they are many.
We connect our plural and singular predicates by afflrming,
as a principle of the framework, that something is small iff its
atoms are few. And we may affirm, as a principle of the
framework, this
Replacement Schema (plural version): Given some things,
and given some other things (not necessarily different),
if each of the former things ... exactly one of the latter
things, and if for each of the latter things there is one of
the former things that ... it, and if the former things are
few, then the latter things are few.
Much more could be done, within the vocabulary of the
framework, to define distinctions of size (see the appendix).
Further, we could look for an elegant systematic theory that
would yield, as axiom or as theorem, each of the 'principles
of the framework' that I have affirmed in passing, and others
besides. But we have done enough to be going on with. We
have in store just what we shall need later.
4
Set Theory for
Mereologists
4.1 The Size of Reality
Now that we have our framework, it is time for an end to
our ontolgical innocence. We shall take on· our commitments
in two stages. First, holding the primitive notion of singleton
still in abeyance, we shall assert that there is a very great deal
of something. We can't say yet just what this great deal of
somthing consists of, but of course  to jump the gun  we
know it must consist predominantly of singletons.
We may affirm three hypotheses about. the size of Reality.
These are not innocent principles of the framework, but
rather are forced upon us by our acceptance of orthodox,
settheoretical mathematics. The first two correspond, as we
shall see later, to the settheoretical axioms of Power Sets and
Unions .
. Hypothesis P: If something is small, then its parts are
few.
Hypothesis U: If some things are small and few, their
fusion is small.
93
Set Theory for Mereologists
Hypotheses P and U would be true if Reality were barely
infinite, with a countable infinity of atoms. 'Small' and 'few'
would then mean 'finite'. Hypothesis P would say, truly, that
a finite thing has only fmitely many parts. Hypothesis U
would say, truly, that the fusion of finitely many finite things
is still finite.
Each hypothesis also can be true at certain larger infinite
sizes. Hypothesis P would be true, for instance, if the size of
Reality were like the limit cardinal bethomega. Hypothesis
U would be true, for instance, if the size of Reality were like
bethone (the continuum), or alephone, or bethseventeen.
The problem comes in demanding that the two be true to
gether. That happens, we saw, ifReality is countably infinite.
It does not happen again until we reach an 'inaccessible' infi
nite size
l
that transcends our commonplace alephs and beths
in much the same way that those transcend mere finitude.
The story is well know as told in set theory; but our frame
work of plural quantification and mereology has enough
power that we can tell it there too.
If we want orthodox mathematics, we cannot be content
with mere countable infinity. We must demand a little more.
But with Hypotheses P and U in place, to demand a little is
willynilly to demand a lot.
Hypothesis I: Some fusion of atoms is infinite and yet
small.
By itself, Hypothesis I is undemanding. It excludes only the
smallest of infinite sizes. It is only when combined with
Hypotheses P and U that it demands the high jump. What
might there be uncountably and inaccessibly many of? Not
I 'Strongly'inaccessible.
94
Mereologized Arithmetic
cats, not quarks, not spacetime points, it's safe to say! So
now it's time for the primitive notion of singleton.
4.2 Mereologized Arithmetic
Our settheoretical primitive can at first be taken as a two
place predicate: x is a singleton of y. But the first of our
axioms for it,
Functionality: Nothing has two different singletons.
straightway entitles us to treat it henceforth as a functor, and
speak of the singleton of something. (A singleton, of course,
is something that is the singleton of something.) Our next
axiom specifies the domain of the singleton function.
Domain: Any part of the null set has a singleton; any
singleton has a singleton; any small fusion of singletons
has a singleton; and nothing else has a singleton.
Our next axiom specifies that different t.hings have distinct
singletons, and that singletons are distinct from parts of the
null set. The axiom demands distinctness in the strong sense
of nonoverlap.
Distinctness: No two things have overlapping singletons,
nor does any part of the null set overlap any singleton.
Distinctness in the sense of mere nonidentity follows: noth
ing is the singleton of two different things, nor is any single
ton part of the null set. Finally we have an axiom of induction:
it says that we can get to anything there is by starting with
95
Set Theory for Mereologists
the null set and its parts, and iterating the operations of
singleton and fusion.
Induction: If there are some things, if every part of the
null set is one of them, if every singleton of one of them
is one of them, and if every fusion of some of them is
one of them, then everything is one of them.
In the axioms as stated, 'null set' looks like an unacknowl
edged second primitive. Let us set that right. By Induction,
we have that everything is either a part of the null set, a
fusion of singletons, or a fusion of a part of the null set and
some singletons. Distinctness implies that the null set has no
singletons as parts. Hence it follows from the axioms that the
null set may be defined as the fusion of all things that have
no singletons as parts, and that there is such a thing. So the
axioms could have been rewritten just in terms of 'singleton '.
A singleton must be an atom.
Proof What could be a part of it? Only a part of the null
set, a fusion of singletons, or a fusion of a part of the null set
and some singletons  for those are the only things there are.
But by Distinctness, it cannot have as a part either any part of
the null set or any singleton other than itself. So it can have
no part except itself. QED
Once we know that singletons are atoms, we can show that
something that consists entirely of atoms, namely the fusion
of all singletons, is infinite.
Proof By Domain, there are singletons, for instance the
singleton of the null set, and all singletons have singletons.
But by Distinctness, some singletons, for instance the single
ton of the null set, are not singletons of singletons. So if x is
Mereologized Arithmetic
the fusion of all singletons of singletons, and y is the fusion
of all singletons, then x is a proper part of y. Further, each
singleton has as singleton exactly one singleton of a singleton,
by Functionality; that is, each atom of y has as singleton
exactly one atom of x. Further, each singleton of a singleton
is the singleton of exactly one singleton, by Distinctness; that
is, each atom of x is such that exactly one atom of y has it as
singleton. So y is infinite, by the Dedekind Schema. QED
Once we know that something that consists of atoms is infi
nite, we know that anything finite is small. Singletons, being
atoms, are finite. So singletons are small. Hence singletons
themselves are subsumed under small fusions of singletons,
and Domain looks to have a redundant clause.
2
The things
that have singletons are exactly the parts of the null set and
the small fusions of singletons.
Recall our previous definitions in terms of 'singleton', and
let them be taken into mereologized arithmetic. We defined
a class as a fusion of singletons, and the members of a class as
things whose singletons are parts of that class. We defined
an individual as anything that has no singletons as parts. We
2 So why did I write it that way? Because I to assume that each
singleton had a singleton before I could establish that something consisting
of atoms was infinite and hence that singletons were small. Without the
assumption, I could have shown that there were two atoms: namely the
null set and its singleton, if the null set is an atom, and otherwise the
singletons of the null set and one of its proper parts. But I could not have
shown that there were three; and only if there are three does it follow that
atoms are small (see section 3.7, note 14).
To see this, note that if the clause 'any singleton has a singleton' were
stricken from Domain, then all the axioms of mereologized arithmetic
would be satisfied if there existed just two atoms, the null set and its
singleton; their fusion; and nothing else. There would be no sma)) fusions
of singletons. hence no singletons thereof, because the solitary singleton,
accompanied by just one other atom, would not be small.
97
Set Theory for Mereologists
defined the null set as the fusion of all things that have no
singletons as parts  the same defmition that now follows
from our axioms. We defmed an urelement as something that
has no singletons as parts, but is part of something else of
which the same is true, i.e. an individual other than the null
set. We defmed a set as either the null set or a class that has a
singleton.
Accordingly, a class is a set iff it is small. A large class is a
proper class, it lacks a singleton, and it cannot be a member
of anything. The things that can be members are exactly the
sets and the individuals; in other words, the small classes, the
null set, and the urelements. Here is the principle of Limi
tation of Size, in an uncommonly explicit form. It is a strong
form of the principle, following von Neumann: not only
does it say that the proper classes are larger than all others,
also it says that an proper classes are the same size namely,
the largest size.
4.3 Four Theses Regained
At the start, I advanced four intuitive theses about the mere
ology of classes. Arguing from these theses together with or
thodox set theory, I concluded that the class of cats was the
fusion of singletons of cats, that Possum was a member of just
those classes that had his singleton as a part, and so forth.
Now I am arguing in the opposite direction: starting from
the axioms and definitions of mereologized arithmetic, plus
principles of the framework, I want to bring us back to
where we started. First we regain the original four theses.
Then we shall regain orthodox set theory.
First Thesis: One class is part of another iff the first is a
subclass of the second.
Four Theses Regained
Proof Left to right. Suppose class x is part of class y and z is
a member of x. By our present definition of membership, z's
singleton is part of x and hence part of y. So z is member of
y. Right to left. Prove the contrapositive. Suppose class x is
not a part of class y. Since x is a fusion of singletons, by defi
nition of class, and singletons are atoms, some singleton is
part of x but not of y. Let it be z's singleton; by definition of
membership, z is a member of x but not of y, so x is not a
subclass of y. QED
Division Thesis: Reality divides exhaustively into individ
uals and classes; in other words, everything is an indi
vidual, a class, or a mixed fusion of individual and class.
Proof By our present defmitions, that means that every
thing either has no singletons as parts, or is a fusion of single
tons, or is a fusion of something with no singletons as parts
and some singletons. We've already proved that everything
either is a part of the null set, or is a fusion of singletons, or is
a fusion of a part of the null set and some singletons. And by
defmitions and mereology, something is part of the null set
iff it has no singletons as parts. QED
Priority Thesis: No class is part of any individual.
Proof That means that no fusion of singletons is part of any
thing that has no singletons as parts, which holds by mere
ology. QED
Fusion Thesis: Any fusion of individuals is an individual.
Proof That means that a fusion of things without singletons
as parts itse1fhas no singletons are parts, which holds by mere
ology plus the fact that singletons are atoms. QED
99
Set Theory for Mereologists
These four theses, we recall, imply our
Main Thesis: The parts of a class are an and only its
subclasses.
4.4 Set Theory Regained
From the axioms and definitions of mereologized arithmetic,
plus principles of the framework, we next derive versions of
the standard axioms for iterative set theory. In some cases
these versions will be stronger than usual, because schematic
formulations will give way to plural quantification. In other
. cases they will be weaker than usual, because they will be
conditional on our hypotheses about the size of Reality.
Having affirmed the hypotheses, however, we are committed
also to the unconditional versions.
Null Set: The nu]] set is a set with no members.
Proof It is a set by definition. It has no members because, by
Distinctness, it has no singletons as parts. QED
Extensionality: No two classes have the same members;
no class has the same members as the nun set.
Proof If two classes had the same members, they would be
two different fusions of the same singletons, contrary to
Uniqueness of Composition. Every class has members, unlike
the null set. QED
Pair sets: If each of x and y is an individual or a set, then
there exists a set of x and y.
100
Set Theory Regained
Proof Each of x and y has a singleton, and the fusion of
these singletons is the class of x and y. We have seen that
singletons are finite and that they are small. Since there is
something infinite that consists of atoms, the fusion of two
singletons something small and something finite  is smal1.
The class of x and y is small, and hence a set. QED
Aussonderung: Given a set x, and given some things,
there is a set of all and only those of the given things
that are members of x.
Proof If x is the nun set, or if none of the given things are
members of x, then the required set is the null set. Otherwise
x is a small class; all members of x have singletons; so by
Unrestricted Composition we have the fusion of all single
tons of members of x that are among the given things. This is
the class of all members of x that are among the given things.
It is part of the small class x, hence small, hence a set. QED
Ordinarily, following Skolem, Aussonderung is stated as an
axiom schema. This amounts to a special case of the plurally
quantified version: given a set x, and given some things that
are all and only the satiifiers if suchandsuch formula, there is a
set of all and only those of the given things that are members
of x. (Or we may get the same effect in two stages: first, a
version of Aussonderung that tells us that given a set x and a
class y, there is a set of the common members of x and y; and
second a classcomprehension schema to tell us that there is a
class  perhaps proper of all and only the sets and urelements
that satisfy suchandsuch formula.) There are normally re
strictions, more or less stringent, on what the formula may
be. Our plurally quantified Aussonderung is stronger. But
beware: this is useless strength, when it comes to proving
theorems in a formal system. For the formal system in question
101
Set Theory for Mereologists
will have to include some axiom system, inevitably par
tial, for the framework. What we get out of our strong
Aussonderung depends on what we can put into it. Recall that
we can write principles of plural 'comprehension': if there
is at least one thing that ... , then there are some things
that are all and only the things that .... We noted that such
principles are ontologically innocent and altogether trivial.
Nevertheless, an axiom system for the framework must sup
ply them; and if, being incomplete, it fails to supply enough
of them, that will cramp our style as surel y as if Aussonderung
itself had been given as a mere schema. So the point of using
plural quantification here is not to increase our power to
prove things, but rather to tell the whole truth.
The same comment will apply, mutatis mutandis, to plurally
quantified versus schematic formulations of Replacement
and Choice.
Replacement: If there are ordered pairs whereby
each member of a class x is paired with exactly one
member of a class y, and if for each member of y there is
a member of x that is paired with it, and if x is a set,
y is a set.
We assume that pairing is defmed here In some standard set
theoretical way, say that of Kuratowski: the (ordered) pair of
u and v is the twomembered set whose members are the
singleton of u and the set of u and v.
Proof Fill the blank in the Replacement Schema (singular
version) with 'is such that its member is paired by one of the
given pairs with the member of'. Recall that the atoms, i.e.
the singleton subclasses, of x and y correspond oneone with
their members, by Functionality and Distinctness. We con
clude that if x is small then so is y. So Y is a set if x is. QED
102
Set Theory Regained
Fundierung: No class intersects each of its own members.
A counterexample to Fundierung might be an infinite class of
XI' x
2
, X
3
,···, where XI contained X
2
as a member, and
X
2
contained X
3
, and so on; or a finite class of XI, X
2
, ••• ,
Xn' where XI contained X
2
, and ... and finally Xn contained
XI; or simply a single thing that was its own singleton.
Fundierung says that none of these things can happen.
Proof Call something grounded iff it belongs to no class that
is a counterexample to Fundierung. We show by Induction
that everything is grounded, and conclude that Fundierung
can have no counterexamples. First, an individual is ground
ed: it has no members, intersects nothing, and so intersects no
class that contains it. Second, if y is grounded, so is its single
ton; else the singleton belongs to, and hence intersects, a
counterexample class C, in which case y must belong to C,
contra the groundedness of y. Third, a fusion of grounded
things is grounded. If the fusion is an individual, we already
know that it is grounded. If it is a mixed fusion of individual
and class, or a proper class, it is grounded because it isn't a
member of anything. The remaining case is that the fusion
is a small class, in which case the given grounded things are
small classes too. Suppose for reductio that the fusion belongs
to, and therefore intersects, a counterexample class C. Then
one of the given grounded things, call it g, also intersects C.
Let C+ be the class that contains all the members of C, and
g as well. Is C+ another counterexample class? No, because
it contains the grounded g. But yes, because each member
of C+  the members of C, and g as well intersects C+ by
intersecting C. This completes the reductio, and so completes
the proof that everything is grounded. QED
Choice: Suppose X is a class, and suppose there are some
ordered pairs whereby each member of X is paired with
10
3
Set Theory for Mereologists
at least one thing, and no two members of x are
paired with the same thing. Then there is a class y
such that each member of x is paired with exactly one
member ofy.
Proof Apply the First Choice Schema, filling in the blank
with 'is paired by one of the given pairs with'. (Note that
things which appear in pairs must have singletons, hence,
whenever we have some of them, we have a class.) QED
Power Sets: Given Hypothesis P about the size of
Reality, then if x is a set, there is a set of all subsets of x.
Proof Since x is a set, x is small. Then the subclasses of x are
also small; so they are sets, and they have singletons. The
fusion y of these singletons is the class of all subclasses of x.
The fusion z of y with the singleton of the null set one
further atom  is the dass of all subsets of z. By Hypothesis
P, the parts of x  that is, its subclasses are few. Then by
the Replacement Schema (plural version) their singletons are
few. These are the atoms of y, so y is smalL We recall that
since the fusion of all singletons is infinite and consists of
atoms, the fusion of something small with one further atom,
which is finite, is still small. So z is small, and therefore is a
set. QED
Unions: Given Hypothesis U about the size of Reality,
then if x is a set, there is a set of all members of mem
bers of x.
3
3 It is not really necessary to take Unions in this conditional form. Azriel
Levy proved that unconditional Unions is redundant in von Neumann's
set theory; see Levy, 'On von Neumann's Axiom System for Set The
ory', American Mathematical Monthly, 75 (1968), pp. 7623. John P.
104
Set Theory Regained
Proof If all members of x are individuals (the null set or
proper parts thereof) then the required set is the null set.
Otherwise, consider those classes that are members of x. Let y
be their fusion; then y is the class of members of members of
x. Each of these classes, since it is a member of something,
Burgess has shown that an improved version of Levy's proof, by
passing Levy's use of unconditional Power Sets and therefore not relying
on Hypothesis P, is available in the present system as welL We knew
already that the axioms of mereologized arithmetic had implications
concerning the size of Reality; Burgess's result indicates that they imply
Hypothesis U.
Here is part, but only part, of a direct proof that mereologized arith
metic implies Hypothesis U. First, I conjecture that
(*) If there are infinitely many atoms, and if some large thing is the
fusion of a few small things, then Reality itself is the fusion of a few
small things.
follows from evident principles of the framework. (I fear it is not a suffi
ciently evident principle in its own right. I believe it in virtue of a suspect
analogy with what's true of things smaller than Reality.) Assume (*).
Suppose for reductio that mereologized arithmetic holds but Hypothesis U
is false. Then, since by mereologized arithmetic there are infinitely many
atoms, we have by (*) that Reality is the fusion. of a few small things.
Then anything is the fusion of a few small things: namely, its intersections
with some of the few small things whose fusion is Reality (whichever
ones of them it overlaps). Given mereologized arithmetic, that enables us
to code all things, and in particular all fusions of atoms, by single atoms;
which is impossible, by the reasoning of Cantor's Theorem and Russell's
. Paradox.
The coding goes thus. First, we can code any x by a class y, such that if
x is small, y is small too: if any singletons are parts of x, y contains them,
and if any individuals arc parts of x, y contains their fusion. If an arbitrary
thing is the fusion of a few small things, we can code it by the few small
classes that code those things, then by the few singletons of those classes,
then by the fusion of those singletons, and finally by the singleton of that
fusion, which is an atom. QED
10
5
Set Theory for Mereologists
must be small. Further, there must be few of them, since the
class of them is part of the small class x. So y is the fusion of a
few small things, hence small by Hypothesis U. Therefore y
is a set. QED
Infinity: Given Hypothesis I about the size of Reality,
then there is a nesting with no greatest member, the
union of which is a set.
(A nesting is a class of sets, such that, whenever x and yare
two members of it, either x is properly included in y or y is
properly included in x.)
Proof By Hypothesis I, there are some things such that each
of them is a proper part of another, and such that their fusion
is a small fusion of stoms. Each of these things must be a small
fusion of atoms. Each atom must be either an individual or
a singleton, and in either case will have a singleton. Replace
each atom by its singleton: that will not affect the smallness
of the things and their fusion, nor will it affect the fact that
each is a proper part of another.
N ow we can say that there are some .sets such that each
of them is properly included in another, and such that their
union is a set. Apply the Second Choice Schema, filling in
the blank with 'is properly included in', and we may con
clude that there are some of the given sets  the class of these
will be the desired nesting  such that (1) each of them is
properly included in another one, so that none is greatest;
and (2) whenever x and yare two of them, either x is prop
erly included in y or y is properly included in x. Further,
since these sets are some of the previous sets whose union was
a set, the union of these sets is part of a set, hence small, and
hence a set. QED
106
Ordinary Arithmetic and Mereologized Arithmetic
This is not the most familiar formulation of a settheoretical
axiom of infinity, to be sure; but with the set theory now at
hand, we know that we can get to various more familiar
verSIons.
What we have, given Hypothesis I, is a set that is infinite
under one possible settheoretical definition of infinity, and
also under our original mereological definition. We already
knew that we had an infinite class  namely, the fusion of all
singletons, which is the large class of all sets and individuals
but not yet an infinite set. For all we knew without Hypoth
esis I, even given Hypotheses P and U, it might have been
that all sets were finite, and only proper classes were infinite.
4.5 Ordinary Arithmetic and Mereologized Arithmetic
We could write the Peano axioms for the arithmetic of the
natural numbers, with 'successor' as primitive, as follows.
Functionality: Nothing has two different successors.
Domain: Zero has a successor; any successor has a suc
cessor; and nothing else has a successor.
Distinctness: No two things have identical successors, nor
is zero identical to any successor.
Induction: If there are some things, if zero is one of them,
and if every successor of one of them is one of them,
then everything is one of them.
In the axioms as stated, 'zero' looks like an unacknowl
edged second primitive. Let us set that right. Induction im
plies that everything is either zero or a successor; Distinctness
Set Theory for Mereologists
says that zero is not a successor; hence it follows from the
axioms that zero may be defined as the thing that is not a suc
cessor, and that there is such a thing. So the axioms could
have been rewritten just in terms of 'successor'.
Four steps take us from ordinary to mereologized arithme
tic. The first step is the crucial one: stipulate that besides get
ting new numbers from old ones by succession, we can also
get new 'numbers' from old by fusion. Specifically, by tak
ing small fusions of successors. The numbers so obtained, like
other numbers, have successors of their own; these successors
have their own successors, and they join in fusions; and so it
goes. We have oldfashioned numbers 0, 1,2, 3, ... ; and also
the new 'number' that is the fusion of 1 and 3; and the suc
cessor of that fusion; amd the fusion of that successor with 2;
and the successor of that second fusion .... Taking this step
means adding clauses to two axioms. The extra clause for
Domain is 'any small fusion of successors has a successor'.
The extra clause for Induction is 'if every fusion of some of
them is one of them'.
Wh y not: 'if every small fusion of some of them is one of
them'? That would correspond directly to our new way of
generating numbers. But when we say 'everything', do we
mean everything, or do we just mean everything in some
restricted domain? In the case of ordinary arithmetic, we had
better just mean everything in a restricted domain  we're
not Pythagoreans! But by the time we're done mereologiz
ing arithmetic, we may hope to say 'everything' and mean
it without restriction. Speaking unrestrictedly, there are large
fusions  for instance, Reality  as well as small ones. So to
be assured that we get everything, large or small, we must
close under fusion generally.
The second step stipulates that the new numbers really are
new. We wouldn't want it to turn out that 17 is really the
fusion of 6 and 42; or that zero is really the fusion of 17 with
108
What's in a Name?
the successor of the fusion of 9 and 6. So we amend Distinct
ness: in both its clauses, we replace distinctness in the sense
of mere nonidentity with distinctness in the strong sense of
nonoverlap.
The third step allows, in effect, that we may have not just
one zero but many: many starting points that are neither
nor fusions of successors. If we're quantifying
unrestnctedly, now, we need to make a place for cats and
quarks and points, and presumably these things are not suc
cessors or fusions thereof. We assume further that if there are
many zeros, they are all and only the parts of one big zero.
So we may reserve the proper name 'zero' henceforth for the
one big zero, and speak of all the 'zeros' (including the big
one) henceforth as 'parts of zero'. So we amend Domain
and Induction by writing 'any part of zero:
Instead of 'zero'. (Strictly speaking, we needn't amend Dis
tinctness: not overlapping any part of zero is equivalent to
not overlapping zero. But I give uniformity of the amend
ments precedence over simplicity of the final formulation.)
The final step, I claim, is not substantive but merely ver
bal: we write 'singleton' instead of 'successor', and 'null set'
instead of ' zero', throughout.
4.6 What's in a Name?
 that arouse incredulity! Surely
sIngleton IS one concept, successor IS another? But this is to
overestimate how rich a conception we have of the member
singleton function and the numbersuccessor function. All
there is to our understanding in either case (apart from some
via negativa, to the effect, for instance, that Possum is neither
a . singl.eton nor a successor) is just a formal modus operandi,
gIven In one case by the axioms of mereologized arithmetic
Set Theory for Mereologists
and in the other case by the axioms of ordinary arithmetic.
(Hence the appeal of structuralism.) And what's said about
the modus operandi of 'successor' is that, within a certain
restricted domain, it's just like that of 'singleton'. The axioms
of ordinary arithmetic, which quantify restrictedly, are silent
about what does or doesn't go on outside that restriction.
They do not deny that things falling outside the restriction
may have successors. For all that arithmetic tells us, maybe
Possum has a successor, and Magpie has a successor, and
maybe the fusion of Possum's successor and Magpie's has a
successor, .... In short, there's nothing in our meager con
ceptions of the membersingleton function and the number
successor function to prove them different, and simplicity
favours the hypothesis that they are the same. And if they
are the same, then it matters not at all whether we say 'single
ton' or whether we say 'successor'. A singleton by any other
name would smell as sweet.
Suppose Gregson tells Holmes that a thief is plundering
Piccadilly, and the police know nothing about him except
for his distinctive modus operandi. In particular, it is not
known whether he operates anywhere else. And Lestrade
tells Holmes that a thief is looting all of London, and the
police know nothing about him except for his distinctive
modus operandi. And 10, the modus operandi of the London
looter is just the same as that of the Piccadilly plunderer! Or
rather, the modus operandi of the looter, when he's looting in
Piccadillylike places, is just like that of the plunderer. Now
even Watson may arrive at the working hypothesis that these
scoundrels are one and the same. And we would do well to
follow his example.
I am not just saying, as Zermelo did and as structuralists
do, that if we take the null set as zero and the singleton func
tion as successor, then we get a settheoretical model of
110
What's in a Name?
arithmetic  one such model among many, the socalled
'Zermelo numbers'. And I am not just saying, as Quine
might, that when we have many models we can choose one
aribtraril y, and I happen to choose this one.
Rather, I am saying this. Assume, pace structuralism, that
we should take 'singleton' as primitive. Then we must admit
that somehow we have managed to give 'singleton' an un
equivocal meaning. But if we did it for 'singleton', why not
also for 'successor'? What's the good of being a parttime
structuralist, for arithmetic but not for set theory? You get
the worst of both worlds. You bear the burden of admitting
that somehow, you know not how, such feats of meaning
giving are possible. Yet you still bear the burden of denying
our naive conviction that 'successor' has in fact been given an
unequivocal meaning. You might as well be hanged for two
sheep as for one. And if we have somehow given meaning
both to 'singleton' and 'successor', then the obvious hypo
thesis, what with the match of formal modus operandi and the
lack of known differences, is that we gave them both the
same meaning.
What sets Zermelo's modelling of arithmetic apart from
von Neumann's and all the rest? It is Zermelo's that identifies
the primitive of arithmetic with an appropriately primitive
notion of set theory. Not, indeed, the notion usually chosen
as primitive  but one that could and should have been. If
there is something about certain meanings that especially
suits them to become the meanings of primitive terms  and
if there isn't, it's hard to understand how determinate mean
ing in thought and language is possible at all
4
 then it is to
4 See David Lewis 'New Work for a Theory of Universals', Australasian
Journal oj Philosophy, 61 (1983), pp. 34377; and 'Putnam's Paradox',
Australasian Journal oj Philosophy, 62 (1984), pp. 22136.
I I I
Set Theory for Mereologists
be expected that, whenever possible given constraints on for
mal modus operandi, one primitive term should take on the
meaning that already belongs to another primitive term.
4.7 Intermediate Systems
Between ordinary arithmetic and mereologized arithmetic
there lie intermediate systems. One is pure mereologized
arithmetic. Take only the first two of our four steps (plus the
fourth if you like) to get
Functionality: Nothing has two different successors.
Domain: Zero has a successor; any successor has a suc
cessor; any small fusion of successors has a successor; and
nothing else has a successor.
Distinctness: No two things have overlapping successors,
nor does zero overlap any successor.
Induction: If there are some things, if zero is one of them,
if every successor of one of them is one of them, and
if every fusion of some of them isone of them, then
everything is one of them ..
This 'everything' is presumably said under a restriction; a
restriction which omits most of the nonsuccessors we think
there are, but within which Unrestricted Composition holds.
Applying our definitions of class, membership, and so on
(written with 'successor' or 'singleton', it doesn't matter), we
get pure set theory: set theory without urelements, in which
the only memberless things are the null set, proper classes,
and mixed fusions of the null set and classes.
The other intermediate system differs right at the beginning.
112.
Completeness of Mereologized Arithmetic
We take fusions of numbers to get new numbers, but
only when the numbers fused meet a special condition. Not
only must it be a small fusion of successors; also, it must be
an initial segment without a last term. I skip the definition,
which exploits the resources of plural quantification to define
precedence, and rest content with examples. One fusible seg
ment consists of the numbers 1, 2, 3, ... ; its fusion is the new
'number' omega. Omega has a successor, omega + 1. The
next fusible segment consists of the numbers 1, 2, 3, .. "
omega + 1, omega + 2, omega + 3, .... Its fusion is the num
ber omega + omega. Then comes omega + omega + 1, ....
And so on, as long as the fusions remain small. In short, we
have the system of ordinal arithmetic; the ordinals are zero,
successors, and limit ordinals, with each limit ordinal taken as
the fusion of all preceding successors. We adjust the axioms
of Domain and Induction to accommodate the limit ordinals
and their successors, and strengthen Distinctness by putting
nonoverlap in place of nonidentity.
If we embed ordinal arithmetic in full mereologized arith
metic, we find that although we are still following Zermelo's
plan when it comes to successors  'successor' means 'sing
leton' we are following von Neumann's plan when it
comes to limit ordinals. Omega, the fusien of 1, 2, 3, ... , is
the union of O's singleton, 1 's singleton, 2's singleton, ... ;
which makes it the class of its predecessors 0, 1, 2, ..... Like
wise omega + omega is the fusion of 1, 2, 3, ... , omega + 1,
omega + 2, omega + 3, ... ; which makes it the class of its
predecessors 0, 1, 2, ... , omega, omega + 1, omega + 2, ....
4.8 Completeness of Mereologized Arithmetic
Let's imagine a fictitious history of geometry: there never
was a parallel postulate. Instead there was the parallel conjecture:
II;
Set Theory for Mereologists
a perennial problem, never proved and never refuted.
After millenia, someone proposed a diagnosis. 'The trouble
may be,' he said, 'that the accepted axioms do not charac
terize the primitive geometrical notion of congruence well
enough to settle the question. Perhaps there are two rival
congruence relations, congruence
l
and congruence
2
• both
satisfying our accepted axioms of geometry' not including
any parallel postulate  'and the reason we cannot settle the
parallel conjecture is that it is true for congruence
1
but false
for congruence
2
.' This diagnosis would have been exactly
right.
The case of arithmetic is different. What if someone pro
posed a parallel diagnosis of our failure to settle Goldbach's
conjecture? 'The trouble may be that the accepted axioms do
not characterize the primitive notion of successor well enough.
Perhaps there are two rival successor relations, successor 1 and
successor 2' both satisfying the Peario axioms; and the reason
we cannot settle Goldbach's conjecture is that it is true for
successor 1 and false for successor 2.' This time, the diagnosis is
wrong. One reason why is that the Peano axioms charac
terize the successor relation up to isomorphism. I mean, of
course, the Peano axioms in their full strength, not some
elementary approximation thereof: InduGtion is to be stated
by means of plural quantification. If we have two rival suc
cessor functions, both satisfying those axioms, they must be
structurally alike. Each must be the image of the other under
a oneone correspondence. That means that their difference
cannot affect the truth of Goldbach's conjecture, or any other
statement formulated in the language of arithmetic.
s
Godel has not been vanquished, of course. It is as true as
5 See any discussion of the 'categoricity of secondorder arithmetic', for
instance in Robbin, Mathematical Logic, pp. 161  3.
114
Completeness of Mereologized Arithmetic
ever that any consistent formal system of arithmetic, even if
it has all the Peano axioms, must omit some truths that ought
to be theorems. But arithmetic is not to blame. The trouble,
rather, is that the framework of plural quantification does
not admit of any complete formalization. The distinctively
arithmetical part of the system is complete.
Now, how about mereologized arithmetic? Is it like the
case of geometry without a parallel postulate? Or is it like the
case of arithmetic? Since mereologized arithmetic is set the
ory, we certainly have unsettled conjectures galore. And we
found plenty of cause to complain that we lack any good
understanding of the primitive membersingleton relation.
We may well ask whether our incomplete understanding of
the primitive notion of set theory bears some of the blame
for our incomplete knowledge of what's true in set theory.
But in fact it doesn't. The case of mereologized arithmetic
that is, set theory; that is, mathematics  is like the case of'
ordinary arithmetic. Our philosophical understanding of the
concept of a singleton is already good enough for purposes
of mathematics. Better understanding would do nothing to
remedy our mathematical ignorance. The trouble lies else
where.
Suppose we have two rival singleton functions, singleton
l
and singleton2> both satisfying the stated axioms of mereo
logized arithmetic. We ask whether some unsettled conjec
ture in set theory might be true for singleton
1
and false for
singleton
2
•
No matter how mystified we may be about the nature of
singletons, still we somehow seem to know that cats, for
instance, or puddles or quarks or spacetime points, are not
singletons and have no singletons as parts. Even without any
understanding of what singletons are, we seem to know
somehow just how much of Reality is the singletonfree
zone. That limits the difference between our two rival
Set Theory for Mereologists
singleton functions: we may assume that they do not disagree
about the demarcation between individuals and everything
else. The individuals
l
are exactly the individuals
2
• (As illus
trated here, we mark defined terms with subscripts to in
dicate whether they are defined in terms of singleton or
. I 1
smg eton
2
. To be an individual
l
is to have no singletons
t
as
parts, to be an individual
2
is to have no singletons
2
as parts.) It
follows, since singletons are exactly the atoms that are not
individuals, that the singletons
l
are exactly the singletons
2
.
So we may suppress the subscripts when we classify things
as 'individuals' and 'singletons'; but of course the subscripts
must return when we speak of the singleton of something.
As well as the demarcation of the individuals, the frame
work also is settled. We are quantifying, once and for all,
over everything; so the domain of quantification is fixed.
The interpretation of the apparatus of the framework, plural
quantification and mereology as well as elementary logic, is
fixed. That means that singleton
l
and singleton
2
cannot
fer in. what they make true concerning the size of Reality.
That IS a framework matter, to which the interpretation of
'singleton' is irrelevant.
6
Then, I say, singleton land singleton
2
differ only by a per
mutation of singletons. They are structurally alike. So they
cannot differ in any way that matters to the truth of any
statement in the language of mereologized arithmetic. Given
the fixed demarcation of the individuals, given the fixed
framework and consequently the fixed size of Reality, the
6 it important that we reject the singularist dogma that plural
quantIfication must really be singular quantification over classes, or at any
rate over some sort of classlike entities. A singularist has no business
the in.terpretation of plural quantification fixed while regarding
the Interpretation of the settheoretical primitive as unsettled. See Thomas
Weston, 'Kreisel, the Hypothesis, and Second Order Set
Theory', Journal oj Philosophical Logic, 5 (1976), pp. 28198.
116
Completeness of Mereologized Arithmetic
axioms of mereologized arithmetic suffice to characterize the
primitive notion of singleton up to isomorphism  indeed,
up to automorphism.
7
It remains to state this properly, and to prove it. A relation
of singletons is a class, perhaps proper, of ordered pairs of
singletons.
8
It is a permutation of singletons iff each singleton
is the first term of exactly one pair and the second term of
exactly one pair. If P is a permutation of singletons, single
tont and singleton
2
differ by the permutation P iff
(1) whenever x is an individual, P pairs the singleton
t
of
x with the singleton
2
of x; and (2) whenever x and yare
fusions of singletons, and P pairs each singleton in x
7 There are known results concerning the almostcategoricity of second
order set theory: one model is not necessarily isomorphic to another, but
at least one is isomorphic to an initial segment of the other. These are dis
cussed; for instance, in Weston, 'Kreisel, the Continuum Hypothesis, and
Second Order Set Theory'. It is our holding fixed of the framework that
takes us from almostcategoricity to categoricity simpliciter.
If we had declined to hold fixed the demarcation of individuals, then
the rival singleton functions could have differed structutally. Something
that is a singleton) could also have been an individual
2
atom, or vice versa.
So the two rival singleton functions might disagreed about how
many individual atoms there were, and in this way fallen short of is om or
phism. The results concerning almostcategoricity avoid such problems
by confining themselves to pure set theory; but we cannot take this
course if we want to hold the domain of quantification fixed by quan
tifying unrestrictedly over everything.
8 In section 2.6, I asked how we were entitled to speak of ordered pairs
before we were given, unequivocally, some singleton function. So far
(but see the appendix) we can introduce pairing only by a settheoretic
definition. Then how can we speak of ordered pairs now? Because now
I have supposed that we are given. unequivocally, each of the two rival
singleton functions. Our problem is not privation but overabundance.
We have only to make an arbitrary choice: as it might be, that our
ordered pairs shall be Kuratowski pairs).
Set Theory for Mereologists
with a singleton in y, and for each singleton in y, P pairs
some singleton in x with it, and x has a singleton
t
, and y
has a singleton
2
, then P pairs the singleton
t
of x with
the singleton
2
of y.
We define P to be the least relation that satisfies the closure
conditions (1) and (2). So, whenever P pairs one thing with
another, it must do so either in virtue of clause (1) or in vir
tue of clause (2). It remains to prove that P is a permutation
of singletons.
Proof Call a singleton good iff P pairs it with exactly one
singleton; and, in general, call something good iff every single
ton that is part of it is good. We use Induction for single
tonI to show that everything is good. Individuals are good,
vacuously, because they have no singletons as parts. Fusions
of good things are good. It remains to show that if x is good
and y is x's singleton 1 , then y is also good. Given that x has a
singleton l' X cannot be either a large fusion of singletons or a
mixed fusion. We have two remaining cases.
Case 1: If x is an individual, P pairs y with x's singleton
2
by clause (1); and with nothing else, since an individual
cannot be a fusion of singletons as well, and hence clause
(2) cannot apply to x.
Case 2: If x is a small fusion of singletons, then let z be the
fusion of all singletons which P pairs with singletons that are
parts of x. Since x is good, the singletons that are part of it
are good; by that and Replacement, we have that z also is
small. So z has a singleton
2
• call it w, and P pairs y with w by
clause (2). Now suppose for reductio that P also pairs y with v,
v i= w. Since v is a singleton, let it be the singleton
2
of u, and
u i= z. How did y and v get paired? Not by clause (1): x is a
fusion of singletons, hence not an individual. By clause (2),
118
Completeness of Mereologized Arithmetic
then. So u is a fusion of singletons, and P pairs each singleton
in x with a singleton in u, and for each singleton in u, P pairs
some singleton in x with it. It follows, by definition of z and
goodness of the singletons in x, that u = z. Contradiction.
This completes the inductive proof that everything is
good, and in particular that every singleton is good. That is:
for every singleton, P pairs it with exactly one singleton. In
just the same way, except that we use Induction now for
singleton
2
, we show that, for every singleton, P pairs exactly
one singleton with it. Therefore P is a permutation of single
tons. QED
To see better what it means that singleton 1 and singleton
2
differ by a permutation of singletons in the sense given by
clauses (1) and (2), it is helpful to note that, in a derivative
sense, P permutes not only the singletons but everything else
as well. Say that P maps x to y iff y is the fusion of all in
dividuals that are part of x, together with all singletons that P
pairs with singletons that are part of x. It is easy to see, once
we know that P is a permutation of singletons, that P
maps each thing to one and only one thing, and that, for each
thing, P maps something to it.
Mereological operations are preserved; if P maps x to y,
and z to w, then x is part of z iff y is part of w. There are two
cases in which x has a singleton 1 : when it is an individual and
when it is a small class. In the second case, if P maps x to y, y
also must be a small class and therefore must have a single
ton2' Clause (1) says that in the first case, when P maps x to
itself, then P maps x's singleton 1 to x's singleton
2
. Clause (2)
says that in the second case, if P maps x to y, then P maps x's
singleton
t
to y's singleton
2
. So, in general, if P maps x to
y, then P maps the singleton
t
of x (if such there be) to the
singleton
2
of y. The sense in which singleton
t
and singleton2
are structurally alike is, loosely speaking, that singleton
2
is
119
Set Theory for Mereologists
the image of singleton
t
under a mereologypreserving per
mutation of everything.
9
That is why no statement in the language of mereologized
arithmetic can be true for singleton
t
and false for singleton
z
.
No choice between the two rivals could tell us more than we
know already about what's true in set theory. The sources
of mathematical ignorance lie elsewhere: in our ignorance of
the size of Reality, and in our irremediable lack of a com
plete axiom system for plural quantification.
9 This is loose speaking because there can be no such thing as a permu
tation of everything. Not even a proper class can have as many members
as there are pairs of things. That is why, hitherto, I shunned the noun and
preferred the verb: P permutes everything.
There are ways, however, to simulate quantification over such relations
as a permutation of everything (see the appendix).
120
..
Appendix on Pairing
by John P. Burgess, A. P. Hazen, and David Lewis
Introduction
Suppose we have the resources of plural quantification and
mereology, as in chapter 3; but no primitive singleton func
tion. Can we somehow get the effect of quantifying over
relations? Equivalently, can we get the effect of plural quan
tification over ordered 'tuples? We can, as it turns out, if
we have infinitely many atoms, and not too much atom less
gunk. It can be done in two different ways, or three if we
count one that is just a hybrid of the othertwo.
Until further notice, we shall assume that everything con
sists entirely of atoms. Later we shall extend both methods to
cover the case where there is a certain amount of atomless
gunk as well.
The Method of Double Images (Burgess, 1989)
The foremost 'paradox of infinity' is that the part may equal
the whole. Infinite Reality contains infinite proper parts that
are microcosms, each one containing within itself images of
121
Appendix on Pairing
all the things there are. Suppose we have one such micro
cosm; and also a second, wholly distinct from the first.
1
All
things have images within the first microcosm, call them first
images; and all things have images within the second, call
them second images. Any first image is distinct from any sec
ond image. Then if P1 is Possum's first image and m
2
is
Magpie's second image, the fusion P 1 + m
2
unequivocally codes
the pair of Possum and Magpie, taken in that order. Such
fusions of first and second images may serve in general as
ordered pairs.
To make good sense of the plan just sketched, we must
first explain what we mean by our talk of 'images'. We can
not yet say directly that there exists a oneone mapping from
all atoms into the atoms of the first microcosm, and there
exists another oneone mapping from all the atoms into the
atoms of the second microcosm. Discovering how to say
such things is exactly our present task.
But we can speak, in effect, of oneone mappings between
the atoms of distinct things. In this special case, we can specify
a relation without recourse to ordered pairs. We can use un
ordered pairs of atoms: twoatom fusions, or diatoms. We
quantify plurally over them.
2
Definition: Diatoms Y map X oneone into Y iff x and yare
distinct; any two of Yare distinct; each one of Y consists
of an atom of x and an atom of y; and x is part of the
fusion of Y.
1 We follow Lewis's usage: 'distinct' means 'nonoverlapping' or 'dis
joint', rather than 'nonidentical'. (See section 3.4.)
2 Soon we shall meet strings of several plural quantifiers, so to avoid con
fusion we supplement the plural pronouns of English with capital letters
used as plural variables. Sometimes we also use capital letters as singular
variables over relations; for all such variables will tum plural upon trans
lation.
122
The Method of Double Images
We cannot tell the direction of mapping just by looking at a
diatom, but since x and yare given along with the diatoms
Y, we know that we are to take the diatom as mapping the
atom of x to the atom of y and not vice versa. The stipulation
that x and yare distinct makes sure we will never be confused
by atoms that are common to both.
So far, we are limited to the special case that the domain
and range are distinct. But we can get around this limitation
by taking mappings in tandem, as follows. Suppose that Xl
and X
2
are a partition of x: that is, they are distinct, and their
fusion is x. Suppose likewise that Y1 and Y2 are a partition of
y. Suppose that the diatoms Y
1
map Xl oneone into Y1' and
the diatoms Y
2
map X2 oneone into Y2' Then Y
I
from Xl to
Y1> in tandem with Y
2
from X2 to Y2' together map X one
one into y. Note that we have written our tandem mapping
as an eightplace relation (with two places plural). We have
not just merged the diatoms Y
1
and the diatoms .Y
2
• We have
retained the information that Y
1
are to be taken as mapping
from Xl to Y1 and Y
2
are to be taken as mapping from X2 to
Y2' The limitation to distinct domains applies to Xl and
Y1; it still applies to X
2
and Y2' But it does not apply to X and
y. They may overlap. They may even be identical, as shown
in figure 1: Xl = Y2' X
2
= Y1' so x= y.
Figure 1
12
3
Appendix on Pairing
By taking mappings in tandem, twice over, we can map
the whole of Reality oneone into each of two distinct
microcosms. We first affirm the following as a framework
principle. (It might be derived from other, more evident
principles; but since we can have no complete axiom system
for the framework, and we have not chosen any particular
partial axiom system, we refrain from doing so.)
Trisection: If Reality is infinite, and consists entirely of
atoms, then there exist some x, y, z, and some X, Y, Z,
such that x, y, and z partition Reality; X maps y + z
oneone into x; Y maps x + z oneone into y; and Z
maps x+ y oneone into z.
Figure 2
Given any such x, y, z, X, Y, Z, we proceed as shown in
figure 2. First, we divide Y and y. Let Y
x
be those diatoms of
Y that have an atom of x as part and let Yx be the fusion of
those atoms of y that are part of diatoms of Y
x
; then the
diatoms Y
x
map x oneone into Yx' Define Y
z
and yz in the
same way; the diatoms Y
z
map z oneone into yz. Now X
from z + y to x, in tandem with Y
x
from x to Yx, together
124
The Method of Double Images
map the whole of Reality oneone into x + Yx' So x + Yx
contains our first microcosm. Likewise Z from x + y to z, in
tandem with Y
z
from z to yz, together map the whole of
Reality oneone into z + y z' So z + Y z contains our second
microcosm. The two microcosms are distinct.
N ow we can make good our talk of images. We start with
images of atoms, and go on to images of fusions of atoms. All
imaging is relative to a given x, y, z, X, Y, Z, assumed to
satisfy the conditions stated in the principle of Trisection.
Definition: Atom v is the first image (with respect to x, y,
z, X, Y, Z) of atom u iff either the diatom u + v is one of
X and u is part of y or of z, or else the diatom u + v is
one of Y and u is part of x. The first image (with respect
to ... ) of any fusion of atoms is the fusion of the first
images of its atoms.
Likewise, mutatis mutandis, for the definition of second im
ages. Any fusion of atoms has a unique first image and a
unique second image. All first images are parts of x + Yx and all
second images are parts of z + yz. Hence any first image is
distinct from any second image. Finally, no two different
things ever have the same first or second image. Now we de
fine our ordered pairs.
Definition: The bpair (with respect to ... ) oj j and s is the
fusion of the first image of j and the second image of s.
For any j and s, there is a bpair ofj and s. A bpair is unam
biguous: if p is the bpair of j and s, and if p' is the bpair off'
and s', then p = p' only ifj= l' and s= s'.
By repeated pairing, we have ordered 'tuples of other
125
Appendix on Pairing
degrees. Let the btriple off, s, and t be the bpair off and p,
where p in turn is the bpair of sand t. (We must be told, of
course, whether such a thing is meant to be taken as a pair or
as a triple.) Similarly we have ordered quadruples, quin
tuples, ....
Roughly speaking, a singular quantifier over relations may
be replaced by a plural quantifier over ordered 'tuples of the
appropriate degree. Instead of saying 'there is a (dyadic)
relation', say 'there are some bpairs'; and similarly for tri
adic, tetradic, ... relations.
But only roughly. The complication is that our bpairs
do not have the wherewithal for decoding built into them.
In this respect they are inferior to the settheoretical ordered
pairs of Wiener or Kuratowski. A quantifier over bpairs must
therefore be accompanied by quantifiers over the where
withal for decoding. If we have a singular existential quan
tifier over dyadic relations, and within its scope one or more
atomic formulas,
For some relation R: }Rk ,
then our translation must take the cumbersome form
For some x, y, z, X, Y, Z such that [here follow the
conditions stated in Trisection), for some R such that
each one of R is a bpair (with respect to x, y, z, X, Y,
Z): the bpair of} and k (with respect to x, y, z, X,
Y, Z) is one of R 
with six extra initial quantifiers besides the plural quantifier
over pairs.
3
Similarly, if we translate a singular universal
3 The order of initial existential quantifiers does not matter; if we swap
the x and z quantifiers, and swap the X and Z quantifiers, we get a
126
The Method of Extraneous Ordering
quantifier over dyadic relations, we get a string of seven uni
versal quantifiers.
The same complication will arise, mutatis mutandis, for all
other methods of pairing considered in this appendix, though
the wherewithal for decoding the pairs will look different
in different cases. It is to be understood, therefore, that any
plural quantifier over pairs will be accompanied by a string
of other quantifiers, mostly plural. For the most part we shall
leave these accom paniments tacit.
The Method of Extraneous Ordering (Hazen, 1989)
There is a second special case in which we can specify a rela
tion without recourse to ordered pairs. Take the parentchild
relation among people. If we know which people are older
than which, we can specify the parentchild relation by
means of the unordered parentchild pairs. The parent is
always the older one. We lack the builtin ordering whereby
the first term of an ordered pair precedes the second, but the
sentence equivalent to the original translation. But the bpair of s andJ with
respect to x, y, z, X, Y, Z is the same thing as the bpair ofJ and 5 with
respect to z, y, x, Z, Y, X. Does this mean that, after all, we have lost
track of the direction of the pairs and the relation? Not in any sense that
matters. Consider sentences
(1 )
(2)
(3)
For some R: :iRk mRn
For some R: kRj fiRm
For some R: nRm
with no occurrences of'R' except those in the displayed atomic formulas.
To be sure, our translations of (1) and (2) are equivalent. So they should
be. The original (1) and (2) also are equivalent: whenever there is a
relation, there is also its converse. What matters is that the translations of
(1) and (3) are not equivalent. Within the scope of the quantifier, we still
keep track of sameness and difference in the direction of the relation.
12
7
Appendix on Pairing
extraneous ordering by age serves as a substitute.
The relation of respect among people is more of a prob
lem. Given an unordered pair of two people such that one
respects the other, we cannot tell whether it is the older who
respects the younger, the younger who respects the older, or
whether the respect is mutual. Also, some people respect
themselves and some do not.
4
However, we can specify the
relation of respect by dividing it into three subrelations. First,
there is the relation of respect from older to younger; it can
be given by its unordered pairs, just as the parentchild rela
tion can. Second, there is the relation of respect from
younger to older, and it too can be given by its unordered
pairs. Third, there is the relation of selfrespect; it can be
given by its unordered identitypairs. More simply, it can be
given just by giving its ituples, the selfrespecting people
themselves.
5
The case of a triadic relation R among people is cum ber
some, but no different in principle. We divide it into thirteen
subrelations: six triadic, six dyadic, and one monadic. They
are given as follows. (We write '0' for 'is older than'.)
4 What of the case that someone respects someone else of precisely the
same age? We may assume, at safe odds of infinity to one, that this never
happens  not when the ordering by age has been specified in a suffi
ciently precise (and artificial) fashion.
S We could instead divide respect into just two subrelations: the relation
of respect for someone not older than oneself, and the relation of respect
for someone not younger than oneself. Each subrelation is given by its
unordered pairs; however, some of these 'pairs' will be identity pairs of
selfrespecting persons, or ltuples, or simply the selfrespecting persons
themselves. This variation would be easier if we had to do only with
dyadic relations, but when we go on to triadic relations it is much more
trouble than it is worth.
128
The Method of Extraneous Ordering
R1XyZ iff Rxyz & xOyOz R7XY iff Rxxy & xOy
R2xyziff Rxyz & xOzOy R8Xy iff Rxxy & yOx
R3XyZ iff Rxyz & yOxOz R9xy iff Rxyy & xOy
R4xyZ iff Rxyz & yOzOx RlOxy iff Rxyy & yOx
Rsxyz iff Rxyz & zOxOy RllXy iff Rxyx & xOy
R6xyZ iff Rxyz & zOyOx R12xy iff Rxyx & yOx
R13X iff Rxxx
Each of Rl ... R6 can be given by its unordered triples; each
of R
7
• •• R12 by its unordered pairs; and R13 by its ituples.
To what extent can we use this strategy within the frame
work of plural quantification and mereology, without ben
efit of set theory? So far, without benefit of set theory, we
do not even have unordered pairs or triples of arbitrary
things. People are one special case, because a fusion of people
divides into people in only one way. Atoms are another spe
cial case, for the same reason. (A more important case, since
more of Reality consists of atoms than consists of people.)
Diatoms serve as unordered pairs. Likewise threeatom fu
sioris, triatoms, serve as unordered triples; and atoms them
selves serve as ituples.
In the case of the atoms, unlike the case of people, we
don't know how to specify an extraneous ordering. But we
don't have to. It is enough to say that one exists. Then we can
proceed relative to an unspecified value of the quantified
variable. Or rather, values; for the quantified variable is plu
ral. In the case of atoms (or in the case of people, but not in
general) we can give an ordering by giving all fusions of its
initial segments.
Definition: 0 are nested iff, for any two of 0, one of
them is part of the other.
Appendix on Pairing
Definition: Atom x precedes atom y with respect to 0 iff
there is some one of 0 that contains x but not y.
Definition: 0 order the atoms iff 0 are nested; the fusion
of 0 contains all atoms; and for any two atoms, one
precedes the others with respect to O.
We might affirm, as a principle of the framework, the exist
ence of some 0 which order the atoms. Instead we affirm a
stronger principle, a version of the Axiom of Choice (different
from those stated in section 3.3), from which that follows.
Definition: 0 wellorder the atoms iff 0 order the atoms
and, whenever there are some atoms X, some one of X
precedes all the others with respect to O.
Wellordering Principle: Some 0 wellorder the atoms.
Now we can translate a quantifier over relations of atoms.
For the dyadic case, we translate
as
For some relation R: , jRk,
For some 0 that order the atoms, for some diatoms R
1
,
for some diatoms R
2
, for some atoms R3: either
j+ k is one of Rl andj precedes k with respect to 0, or
j + k is one of R2 and k precedes j with respect to 0, or
j= k andj is one of R
3
;
The Method of Extraneous Ordering
and similarly when the quantifier is universal. 6 For the triadic
case we have a string of fourteen plural quantifiers: first
comes the 0 quantifier, then the R
1
• •• R6 quantifiers over
triatoms, then the R
7
• •• R12 quantifiers over diatoms, and
finally the R13 quantifier over atoms. An atomic formula
'Rhjk' within the scope of the quantifier becomes a disjunc
tion with thirteen disjuncts.
We spare you the tetradic case; because once we can quan
tify over triadic relations, we can introduce ordered pairs of
atoms. Then we can proceed to ordered triples, quadruples,
... of atoms by repeated pairing; and we can translate all
further quantifiers over relations among atoms by plural
quantifiers over 'tuples.
Our ordered pairs of atoms will themselves be atoms. But
we need not specify our pairing relation; it is enough to say
that one exists.7 But we do not say exactly that; rather, we
say its translation. The requisite definition and principle are
given below in un translated form, with a quantifier over
triadic relations of atoms. What we really affirm as a prin
ciple of the framework is the translation of the principle as
written.
Definitions: P is a pairing relation (on atoms) iff P is a tri
adic relation of atoms; for any atoms f and s, there is a
6 One might worry that we lose track of the direction of the relation,
since it would be reversed if we reversed the ordering 0 or if we swapped
the Rl and R2 quantifiers. But the reply given in note 3 applies, mutatis
mutandis.
7 Instead of our one triadic pairing relation, we could take two dyadic
unpairing relations: the one that yields the fIrSt term, and the one that
yields the second term. Together, these give back the pairing relation.
Had we chosen this alternative, it would have been worthwhile to adopt
the variation considered in note 5.
Appendix on Pairing
unique p such that Pfsp; if Pfsp and Pf's'p, thenf= f' and
5 = s'. The hpair off and 5 (with respect to P) is the p
such that Rjsp.
Pairing: If Reality is infinite, and consists entirely of
atoms, there exists a pairing relation.
Now our new translation of
For some relation R: jRk,
is our old translation of
For some hpairs R, with respect to some pairing rela
tion P: the hpair of j and k (with respect to P) is
oneofR;
and similarly when the quantifier is universal; and similarly
in the triadic, tetradic, ... cases. We need onl y use an
extraneous ordering once, to legitimize the introduction of
hpairs by quantifying over pairing relations. Thereafter we
use the hpairs.
So far, we have a pairing relation that only yields hpairs
of single atoms. But we may extend the definition of h
pairing from atoms to fusions. Given two fusions f and 5 of
atoms, we may pair them atom by atom, as follows.
8
Definition: The hpair off and 5 (with respect to P) is the
fusion of all hpairs (with respect to P) of an atom off
and an atom of s.
8 Here we borrow a device that is known in other uses, for instance in
higherorder arithmetic. We thank W. V. Quine for pointing out to us
that it would serve our present purpose.
Adding Gunk
Since hpairs of atoms are atoms, we can recover them from
their fusion. Then we can recover f as the fusion of the atoms
which are the first terms of these hpairs, and 5 as the fusion
of the atoms which are the second terms. So hpairing of
fusions is unambiguous.
We obtain ordered 'tuples of other degrees by repeated
hpairing. We can translate a singular quantifier over rela
tions between fusions of atoms by a plural quantifier (suit
ably accompanied) over such 'tuples.
A Hybrid Method
We could begin as the Method of Extraneous Ordering does,
and finish as the Method of Double Images does. Once we
can quantify over dyadic relations of atoms, however we do
can state a framework principle as follows: If Reality is
mfimte, and consists entirely of atoms, then there exist two
relations Rl and R2 mapping Reality oneone into two dis
tinct proper parts of Reality. With respect to any such Rl
and R
z
, we can define the first and second images of a given
thing as its images under Rl and R
z
respectively. We can
define the hbpair off and s as the fusion of the first image off
and the second image of s. We go on as usual to 'tuples of
higher degree and to a translation of singular quantifiers over
relations.
Adding Gunk
Now we withdraw the assumption that Reality consists
entirely of atoms. Maybe the fusion of all atoms is only part
of and the rest is atom less gunk, so that any part
of It has proper parts in turn. If so, all our methods so far
Appendix on Pairing
apply only to the atomic part of Reality. They must all be
extended.
To that end, we note still another special case in which a
relation may be given by unordered pairs: the case of a re
lation that holds between atomless things and atoms. The rela
tion is given by the gunkandoneatom fusions such that the
gunk bears the relation to the atom. In general, any fusion
decomposes uniquely into its gunk and its atoms.
Definition: The maximal atom less part and the maximal
atomic part of any given thing are, respectively, the fu
sion of all its atomless parts, if any, and the fusion of all
its atoms, ifany.
A gunkandoneatom fusion has maximal parts of both
kinds, and the maximal atomic part is a single atom. Now we
want to say, by plural quantification over gunkandone
atom fusions, that there is a oneone correspondence where
by every atomless thing has a representative atom. That will
be so, provided there are enough atoms to go around; in
other words, provided there is not too much atomless gunk.
NOHoomuchgunk Hypothesis: There are some things G
such that each one of G is the fusion of an atomless thing
and exactly one atom; every atomless thing is the maxi
mal atom less part of exactly one of G; and no two of G
have an atom as a common part.
(The hypothesis follows from Hypothesis P of section 4.1.
The fusion of all atomless things is small; so by Hypothesis P
its parts are few; these parts are all and only the atomless
things; and by the definition of 'few', we have the desired G.)
Let A(x) be the maximal atomic part of x, if x has any
atoms. And, given G satisfying the conditions of the
Adding Gunk
Nottoomuchgunk Hypothesis, let G(x) be the atom such
that its fusion with the maximal atomless part of x is one of
G, if x has any atomless parts.
We have our methods of introducing ordered pairs, triples,
etc, within the atomic part of Reality. Whichever meth
od we choose, our pairs, triples, etc. are fusions of atoms.
Choose some one of our methods of pairing, and hold it
fixed in what follows.
We can encode arbitary things by ordered pairs of fusions
of atoms. Something x may consist entirely of atoms, so that
G(x) is undefined; or it may be atom less gunk, so that A(x) is
undefined; or it may be mixed, part atoms and part atomless.
Let a, g, m be three arbitrarily chosen atoms, used as markers
to distinguish the three cases.
Definition: The code of x (with respect to G, a, g, m) is
the pair of a and A(x), if G(x) is undefined; or the pair
of g and G(x), if A(x) is undefined; or the pair of m and
the pair of A(x) and G(x), otherwise.
Now we can introduce new ordered pairs, available not only
for fusions of atoms but for things of all kinds, as ordered
pairs of these codes.
Definition: The new pair off and s is the pair of the code
off and the code of s. .
We obtain new triples, quadruples, ... by repeated pairing.
We translate a singular quantifier over relations between
arbitrary things into a plural quantifier over the new 'tuples.
(The accompanying quantifiers over the wherewithal for
decoding the new 'tuples now include quantifiers over G, a,
g, m, as above, as well as the wherewithal for decoding the
old pairs.) Henceforth we may quantify over relations, freely
Appendix on Pairing
and without comment. Knowing, as we do, how such quan
tifiers may be translated, we need not bother actually to
translate them.
Megethology9
Sections 3.7 and 4.1 used the resources of the framework to
express distinctions and hypotheses concerning the size of
Reality. Now that the resource:; of the framework turn out
to include quantification over relations, the job can be done
over in an easier and more familiar way.
Definition: X and Yare equinumerous iff there exists a
relation R such that each one of X bears R to exactly
one of Y, and for each one of Y, exactl y one of X bears
R to it.
Definitions: Yare many parts of x iff Yare parts of x, and
the atoms of x and some of Yare equinumerous. Yare
few parts of x iff Yare parts of x but not many parts of x.
Definitions: Something is a large part of x iff its atoms are
many parts of x; a small part of x iff its atoms are few
parts of x.
Corresponding to Hypotheses p, U, and I of section 4.1, we
can write down three conditions on an arbitrary thing x.
Taken together, they come rise a definition of inaccessible
SIze.
Condition P: Whenever something is a small part of x, its
parts are few parts of x.
9 From megethos 'size' + logos 'doctrine'.
Megetholo£y
Condition U: Whenever some things are few parts of x
and small parts of x, their fusion is a small part of x.
Condition I: Some fusion of atoms of x is infinite and yet
is a small part of x. (Something infmite is called uncount
able iff it satisfies Condition I, otherwise countable.)
Definition: x is of inaccessible size iff x satisfies Conditions
P, U, and I.
Standard set theory requires that Reality be of inacessible
size. But we may speculate, if we wish, that it is larger stilL
Hypothesis IC: Some small part of Reality is of
inaccessible size.
This is the framework version of a modest 'large cardinal'
axiom. Given mereologized arithmetic, it implies that there
is a set of inaccessible size, hence that there exists an
inaccessible cardinal.
We can also state a less modest hypothesis, as follows. Sup
pose we have a measure, for instance the familiar Lebesgue
measure of area. We can then distinguish regions of measure
zero, 'negligible' regions, from others. Sometimes thisquali
tative distinction is all we want from a measure, and when it
is, we can conflate the measure itself with the distinction be
tween negligible regions and others. Measures in this purely
qualitative sense can be characterized in the framework.
Definition: N measure x iff the fusion of N is x; x is not
one of N; every part of one of N is itself one of N; and
whenever y and z are among N, then the fusion of y and
z is one ofN.
137
Appendix on Pairin,{!
When N measure x, N may also meet further conditions.
Defintion: N are fully additive on x iff, whenever Yare
few parts of x and Yare among N, then the fusion of Y
is one of N.
Definition: N are maximal on x iff, whenever y and z
partition x, either y or z is one of N.
Definition: x is measurable iff, for some measure N on x,
N are both fully additive and maximal on x.
A countable fusion of atoms is measurable.
10
Are larger
things ever measurable? Not unless they are very much
larger, as it turns out. An uncountable measurable thing
would have to be of more than merely inaccessible size. But
we may speculate, if we wish, that Reality is measurable. Or,
stronger still,
Hypothesis MC: Some small part of Reality is measur
able but uncountable.
Given mereologized arithmetic, this hypothesis implies that
there exists a measurable cardinal.H
10 If x is an infinite fusion of atoms, countable or not, the finite parts of x
measure x. Whenever N measure x but fail to satisfy maximality, then N
may be extended to N+ which also measure x and do satisfy maximality;
this is shown by an argument requiring a wellordering of the parts
of x. (It can be carried out when we have set theory, or else it requires
additional principles of the framework.) Finally, we note that when x is
countable, any measure on x must be fully additive; for in that case 'few'
means 'finite', so full additively adds nothing to the final clause in the
original definition of measure.
11 For a settheoretical discussion of measurable cardinals, see Shoen field,
Mathematical Logic, section 9.1 O.
Ramsifying out the Singleton Function
As well as asking what is the largest size Reality has to
offer, we can ask also about intermediate sizes. The Gen
eralized Continuum Hypothesis, which addresses this ques
tion, is usually taken as a hypothesis about the sets. But we
can state it in the framework.
Generalized Continuum Hypothesis: If x is an infinite
fusion of atoms, and Yare some of the parts of x, then
either Y and all the parts of x are equinumerous or else
Yand some of the atoms of x are equinumerous.
The Continuum Hypothesis, ungeneralized, is the special
case in which we stipulate that x is countable.
Ramsifying out the Singleton Function, Continued
Now that we may quantify over relations, secure in our
knowledge of how to translate these quantifications using
only the resources of the framework, we may return to the
unfinished business of section 2.6. Adapting a suggestion of
Paul Fitzgerald, we considered a way to get rid of the primi
tive membersingleton relation. We could Ramsify it out.
Take some axioms, perhaps the axioms of mereologized
arithmetic in section 4.2, with 'singleton' as their only set
theoretical primitive.
12
Conjoin them into a sentence saying
that the membersingleton relation satisfies certain structural
(and perhaps other) conditions.
... singleton ... singleton ... singleton ...
12 To put them in primitive notation, uniformly replace 'null set' by
'fusion of all things that have no singletons as parts'; then uniformly
replace 'singleton', when used nonrelationally, by 'thing that is a single
ton of something'; and add an extra axiom saying that something has no
singletons as parts.
Appendix on Pairing
We may replace it by (a translation of) its Ramsey
sentence
For some S: ... S ... S ... S ...
which says just that some relation satisfies those conditionsP
Where there is one suitable relation, there will be many.
(As many as there are permutations of the singletons.) To
claim a primitive understanding of the membersingleton
relation is to think that one of all these relations, and only
one, is the one we had in mind all along. To retreat to set
theoretical structuralism
14
is to think that the suitable rela
tions are all on a par, with nothing to distinguish the one real
membersingleton relation trom the horde of pretenders. In
that case, any further sentence about 'the' membersingleton
relation
singletonsingletonsingleton
should be understood as tacitly general: it says that some
thing holds for all relations that satisfy the given conditions.
13 Had we had been content to take membership generally as a relation
between atoms, we could have Ramsifled membership as soon as we
were in a position to quantify over dyadic relations of atoms; we would
have needed only the first part of the Method of Extraneous Ordering.
The axioms to be Ramsified would not have been those of mereologized
arithmetic, but would have taken membership as primitive in the stan
dard way. This would have come closer to Fitzgerald's original sugges
tion. But it would have meant abandoning the thesis that a class has its
subclasses as parts, not to mention our usual supposition that nonatomic
individuals are sometimes members of classes.
14 So called because it takes settheoretical mathematics to consist of
generalizations over all the many structurally similar membersingleton
relations; not because it posits a new sort of an entity, an 'abstract struc
ture' common to all those relations.
140
Ramsifying out the Singleton Function
:n
e
. may replace it by (a translation of) an explicit general
IZatIon.
For all S: if. .. S ... S ... S ... , then
(We might wish to conjoin the Ramsey sentence to
the generalization, to prevent vacuous truth in case no suit
able relation exists.) If the further sentence followed from the
axioms, plus principles and hypotheses of the framework
. also the generalization fol1ows from those auxiliar;
prmClpies and hypotheses. (Compare first order logic: if T(s)
foHows from A (5) and F l' F 2' ..• and the constant 5 does not
occur in the Fs, then the generalization
For an x: if A (x) then T(x)
foHows from the Fs alone.)
Structuralism proposes to reconstrue all theorems of set
theory  that is, all theorems of the whole of settheoretical
mathematics  as the corresponding generalizations. This
reconstrual is a painless reform. No previously accepted the
orem must be denied; no previously accepted method of
proof must be renounced. Mathematics may go on just as
before. Only our understanding of it needs revision.
(And yet structuralism is a reform; and it is a reform that
serves little purpose if we really did have a primitive under
standing of the membersingleton relation; and so it is a
rebuke to the old mathematics that was content to take mem
bership as primitive. A modest philosopher might well hesi
tate to rebuke the old mathematics. He might think that if
he cannot understand how it was entitled to its primitive
membersingleton relation, most likely the fault is his.)
The philosophical reward of structuralism is that it
Appendix on Pairing
bypasses all doubt about whether the primitive member
singleton relation is well enough understood. You might
complain (as Lewis does in section 2.1) that your introduc
tory lesson in set theory just does not apply to the case of
membership in a singleton. You might complain that all you
were told amounts only to this: it's just like collecting many
into one, only without collecting many into one. You might
complain that no known theory of intentionality explains
how you can have in mind just one out of all the suitable
relations. You might complain that there's no hope of reduc
ing the membersingleton relation to any familiar properties
and relations, because those familiar properties and relations
cannot distinguish inaccessibly many atoms. Whatever the
merit of these complaints, structuralism makes them moot. It
says there is no primitive notion that needs understanding; all
the suitable relations are equally 'the' membersingleton rela
tion; we needn't try to wrap our minds around some special
one of them.
But structuralism is no panacea. It does not put paid to
all philosophical complaints that might be brought against
settheoretical mathematics. If you complain not about the
'ideology' but about the ontology of set theory, structuralism
doesn't help. Whether the singleton function is primitive or
whether it is Ramsified, set theory still requires the hypoth
esis that there are inaccessibly many atoms. Only very few of
them can be ordinary atoms, of the kinds we know and name
by causal interaction. The rest are mysterious. We know
nothing about their whereabouts, or lack of it. We know
nothing about their intrinsic qualitative character, or lack of
it. We have to believe in these mysterious extra atoms, on
pain of rebelling against established mathematics. Structural
ism does nothing to relieve the burden.
In fact, it makes matters worse. We might have been
attracted to the speculation that singletons are where their
142
Ramsifying out the Singleton Function
members are, and perhaps even share the qualitative charac
ter of their members. (We might have hoped to answer,
somehow, the question how atoms can share the location or
character of nonatoms.) But structuralism makes nonsense
of this speculation. If all suitable relations alike are member
singleton relations, then a singleton does not have its mem
ber once and for all. For anything that can be a member, for
anything that can be a singleton, there will be some member
singleton relation that pairs the one with the other. The atom
that is Possum's singleton under one suitable relation is
Magpie's singleton under another. Shall it share Possum's
location and character, or shall it share Magpie's?
If you complain (as Hazen does) about the non
constructiveness of settheoretical mathematics, again struc
turalism doesn't help. All it does is redirect your complaint
from set theory to plural quantification and mereology. If
you will not affirm an existential statement until you see a con
firming instance  and all the more if you question whether
quantification over un constructed mathematical objects is
meaningful you should not like the Ramsey sentence of
mereologized arithmetic any better than you liked the orig
inal axioms. In fact the Ramsey sentence is worse by one
initial quantifier  or rather, under translation, by one long
string of quantifiers. You should like the Axiom of Choice
no better as a principle of the framework than as an axiom of
set theory. And you have a second reason to challenge the
positing of inaccessibly many atoms, quite apart from any
qualms about their unknown location and character.
If you simply complain that the familiar axioms of set the
ory are not a selfevident foundation for mathematics, again
structuralism doesn't help. The Ramsey sentence of merea
logized arithmetic is far from selfevident, and the hypothesis
that there are inaccessibly many atoms is downright surpris
ing. In the case of the Ramsey sentence, there is some room
143
Appendix on Pairing .
for improvement; in the next section, we shall see how to
regain it from a seemingly weakened version. Even so, there
is little hope for a selfevident foundation. For better or
worse, it is the edifice that justifies the foundation or noth
ing does.
Equivalence under Ramsijication
Structuralism also brings a mathematical reward. (So it does
not quite leave mathematics unchanged. Nothing is lost, but
something is gained.) It lets us weaken the axioms of mere
ologized arithmetic. Without Ramsification, the new, weak
ened axioms are not· equivalent to the old axioms. If a
relation satisfies the new axioms, it does not follow that the
same relation satisfies the old ones. Y if some relation satis
fies the new axioms, it may follow that some relation sa
tisfies the old ones. (The converse is trivial, given that the old
axioms imply the new ones.) If so, the old and new Ramse:
sentences are equivalent. And if all that we really assert IS
the Ramsey sentence, as structuralism claims, then the new
axioms are interchangeable with the old. There is no substan
tive question which are right.
15
Two simultaneous weakenings (at least) are possible. One
is to drop the axiom of Induction, and put in its place the
much weaker
Atomicity: Every singleton is an atom.
(In section 4.2, Atomicity was derived using Induction.) The
other is to retain the positive part of the axiom of Domain
15 For a general discussion of equivalence under Ramsification, see Jane
English, 'Underdetermination: Craig and Ramsey', Journal of Philosophy,
70 (1973), pp. 45362.
144
Equivalence under Ramsification
Domain +: Any part of the null set has a singleton; any
singleton has a singleton; any small fusion of singletons
has a singleton,
and drop the negative final clause, which says that nothing
else has a singleton. Dropping Induction means leaving it
open that there might be nonwellfounded sets, involved in
infinite descents of membership}6 Dropping the negative
clause in the axiom of Domain means leaving it open that
there might be scattered exceptions to the principle of Limi
tation of Size: occasional cases in which a large fusion of
singletons has a singleton. (Scattered exceptions only: almost
all classes must lack singletons, else there would not be
enough singletons to go around; almost all classes are large;
therefore almost all large classes must lack singletons.) Drop
ping the negative clause also means leaving it open that there
might be singletons of mixed fusions of individuals and
classes, pace the invocation of 'our offhand reluctance to
believe in them' in section 1.3. If we think we have a primi
tive understanding of the membersingleton relation, it is a
substantive question whether the relation we have in mind
really does allow any of these things to happen; whether it
satisfies the original axioms of mereologized arithmetic as
well as the weakened axioms. But if we are structuralists, and
think we did not have anyone relation in mind, that question
makes no sense. We can weaken the axioms without loss.
Proof Let singleton
l
be a relation that satisfies Functional
ity, Domain +, Distinctness, and Atomicity, but perhaps not
the negative clause of Domain and perhaps not Induction.
Then there is a relation singleton3 that satisfies all the axioms
16 See Peter Aczel, Non Welljounded Sets (Center for the Study of
Language and Information, Stanford University. 1988).
145
Appendix on Pairing
of mereologized arithmetic. (We subscript defined terms of
mereologized arithmetic to indicate which singleton relation
they correspond to.)
Call a singleton
t
required iff, whenever there are some
atoms such that every singleton
t
of an individual
t
is one of
them, and every singleton
t
of one of them is one of them,
and every singleton
t
of a small fusion of them is one of them,
then it is one of them. Atomicity ensures that singletons
t
are
indeed atoms. Take the class of nonselfmembers
t
whose
singletons
t
are required. It is large; else it would have a
singleton
t
by Domain +, and its singleton
t
would be re
quired, so it would be a selfmember
t
iffit were not.
So all the required singletons
t
are many, so they are equi
numerous with all the singletons
t
, so there is a relation that
maps the required singletons
t
oneone onto all the single
tons
t
• Let us extend it to a relation M that also maps all in
dividuals
t
onto themselves. Let singleton
2
be the restriction of
singleton
t
to the required singletons
t
and the individuals
t
,
and let singleton 3 be the image of singleton
2
under M. It is
easily shown that singleton
3
satisfies Functionality, Domain,
Distinctness, and Induction. QED
A second use of equivalence under Ramsification is of meta
physical rather than mathematical interest. Recall the dis
cussion of 'unofficial axioms' of set theory in sections 2.1 and
2.6. You might think that the 'structural' axioms of
logized arithmetic are not really enough to characterize a
suitable membersingleton relation. You might want a meta
physical characterization as well. There may be little positive
to say, but at least there is a via negativa: you can insist that no
singleton is part of any material thing, or part of spacetime,
or part of any platonic form, or part of any spirit (either tem
poral or eternal), or .... If you think this via negativa is an
essential ingredient in our conception of'singletons, you
146
Equivalence under Ramsification
could insist on incorporating it into the axioms. The result
ing conjunction of axioms will be a strengthening of the
axioms of mereologized arithmetic; and the resulting
Ramsey sentence will go beyond the vocabulary of the frame
work. Although the primitive notion 'singleton' wiIl be
Ramsified out, such terms as 'material', 'spacetime', 'pla
tonic form', 'spirit', ... will remain. This sort of Ramsey sen
tence says that there is a relation that is suitable not only
mathematically but also metaphysicallyP
But if structuralism is right, the strengthening may be
iIlusory. Suppose for definiteness that we contemplate adding
one 'unofficial' axiom of the form
XAxiom: An atom is an individual atom iff it is one
of the atoms X [somehow specified], otherwise it is a
singleton.
And suppose we are prepared to affirm, as an auxiliary
hypothesis, that only few atoms are among X, and hence that
many atoms are not. (That means it would be foolhardy to
specify X as 'the atoms located in spacetime' or 'the atoms
having some qualitative character'; such a specification would
go beyond classifying the atoms, and would advance the
daring speculation that some  indeed, many  atoms lack
location or character.) Assume, as usual, that Reality is of
inaccessible size. Then the XAxiom is redundant under
Ramsification. It is not a substantive issue whether to take it
or leave it. We get equivalent Ramsey sentences either way.
17 Further, if we strengthen the axioms by giving a necessary and suffi
cient condition for an atom to be an individual, then we can apply the
result of section 4.8: two membersingleton relations that both satisfy the
strengthened axioms differ only by a permutation of singletons.
147
Appendix on Pairing
Proof Let singleton
3
be a relation that satisfies all the ax
ioms of mereologized arithmetic, but perhaps not the X
Axiom. Then there is a relation singletons that satisfies the
axioms of mereologized arithmetic and the Xaxiom as well.
1s
Let n be the fusion of the atoms X and whatever atom less
gunk there may be. The Xaxiom tells us that n ought to end
up as the null set. We have assumed that n is small.
Since we have all of mereologized arithmetic and the
hypothesis that Reality is of inaccessible size, we may use set
theory freely. There is a rank3' call it R, big enough so that
(1) for some atoms Yat rank3 R, Yare equinumerous with
the parts of n; and (2) for some atoms Z below rank R, Z are
equinumerous with X. Let m be the fusion of the atoms Z
and whatever gunk there may be. There is a relation single
ton
4
that maps the parts of m oneone onto Y. Say that an
atom is above Y iff, whenever there are some atoms and every
atom of Y is one of them, and every singleton of one of them
is one of them, and every singleton of a small fusion of them
is one of them, then it is one of them. Let the relation single
tons be the restriction of singleton
3
to atoms above Y. Let the
relation singleton
6
be the union of singleton
4
and singletons.
The atoms above Yare many; the atoms above Y together
with the atoms Z are likewise many. So we have a relation
that maps the atoms above Y together with the atoms Z
oneone onto all atoms; and let us extend it to a relation M
that also maps all atomless things onto themselves. Let single
ton, be the image of singleton
6
under this M. It is easily
shown that singleton, satisfies all the axioms of mereologized
arithmetic. It does not satisfy the XAxiom (unless by luck);
18 As the subscripts suggest, this proof may be joined to the previous one
to show that the weakening of mereologized arithmetic just considered is
equivalent under Ramsification (given the specified hypotheses) to the
strengthening of mereologized arithmetic which adds an XAxiom.
14
8
Equivalence under Ramsijication
however the individual, atoms are equinumerous with the
atoms X that ought to be individual atoms.
So there is a permutation of atoms that maps the individ
ual, atoms onto X. Let us extend it to a relation N that also
maps all atom less things onto themselves. Let singletons be
the image of singleton, under N. It is easily shown that
singletons satisfies all the axioms of mereologized arithmetic,
and the XAxiom as well. QED
149·
Index
acts of setforming, 2931
Aczel, Peter, 145
Angelelli, Ignacio,S
arithmetic, 48,534,10713,
11315
Armstrong, D. M., ix,S, 567,
63,76,823,86
Atomicity, Axiom of, 144
atoms, definition of, 15
Aussonderung, Axiom of, 1012
awesome 'classes', 668
axioms, for arithmetic, vii, 48,
107; for mereologized
arithmetic, vii, 46, 62, 956,
112,139; for mereology, 74; for
ordinal arithmetic, 113; for set
theory, 1007; unofficial,
314,46,49,109,11516,
1467
Baxter, Donald, ix, 824
Benacerraf, Paul, ix
Bigelow, John, 55
151
Black, Max, 63
Boolos. George, ix, 19. 52,623,
66, 70
Bricker, Phillip, ix
Bunt, Harry c., viiviii, 10, 16,
21,61,73
Burgess,John P., viii, ix, 53,
1045, 1217
Campbell, Keith, 334, 76
Cantor,Georg,5,27,29,105
cardinals, see large cardinals, size
Carnap, Rudolf, 47 '
categoricity, 11320, 147
character of singletons, 335, 57,
1423, 147
Choice, Axiom of, 712, 102,
1034, 130, 143
class, definitions of, 4, 16,97
classlike things, 51,65, 116
coding of pseudomembers, 238
completeness, 11320, 147
Index
composition, as identity, 817;
uniqueness of, 389, 74, 789,
85,100; unmereological, 3,
3841,57,79; unrestricted,
78,1819,74,7981,85,101,
112
connected parts, 227
conservatism, ix, 9, 19,50,54,
589,141
constructiveness, 143
Continuum Hypothesis, 139
counterparts, 378, 77
Cresswell, M. j., ix
Davoren,Jennifer, ix
decomposition, uniqueness or
multiplicity of, 23, 5, 22. 24
Dedekind, Richard, 58,889; see
also Peano axioms
difference, mereological, 9, 17
Distinctness, Axiom of. 95,
1079,112
Division Thesis, 79.15,16,
434,99
Domain. Axiom of, 95, 1079,
112,1446
double images, 1217,133
Eberle, Rolf, 73
empty set, see null set
English,Jane,l44
equivalence under Ramsification,
1449
Etchemendy, Nancy, ix
Extensionality, Axiom of, 100
external relations, 345, 378
extraneous orderings, 12733
facts, 8, 567
few and many, definitions of,
901.136
Field, Hartry, ix, 58
finite size, 889, 945, 967, 107
First Thesis, 46, 910, 56, 98
Fitzgerald, Paul, 45, 139, 140
Forrest, Peter, 56
framework, see mereology. plural
quantification
Functionality, Axiom of. 95, 107,
112
functions, see relations
Fundierung, Axiom of, 18,20, 103;
see also Induction
fusion, definitions of, 12. 73
Fusion Thesis, 7, 9, 15,99
Generalized Continuum
Hypothesis, 139
generating relations, 3841
geometry, 11314
God, 8,9,31,75,76
Godel's incompleteness theorem,
11415
Goldbach's conjecture. 114
Goodman, Nelson, 33,34,3841,
73, 76
gunk,201,39, 70, 74.88,89,
121.1336
haecceities, 55, 57
Halmos, Paul R., 2930
Hazen, A. P., viii, ix, 53, 12733,
143
hierarchy of classes, 6, 1213,20,
26,278,32,35
Holmes, Sherlock, 110
Index
Hume, David, 70
Hypotheses I, P, and U, 934,
1047,134,1489
identity, composition as, 817
images, 1217, 133
improper parts, definition of, 1
inaccessible size, 94, 1367, 142,
143, 1478
inclusion, 10, 11, 17
individuals, definitions of. 4, 15,
17,97; as parts of classes, 6,
425
Induction, Axiom of, 96, 107, 109,
112,113, 1446
in finitary sentences, 50, 69, 701
infinite and finite, 889, 945,
967,1067
Infinity. Axiom of, 1067
internal relations, 347
Isaacson, Daniel. ix, 30
Jekyll and Hyde, 6, 31, 61
Johnston, Mark, ix
Kleene, Stephen c., 2930
Kuratowski, Kazimierz, 102, 117,
126
large cardinals, 49, 1378
large and small, definitions of,
8991, 136
Lasso Hypothesis, 425
Leonard, Henry, 73
Lesniewski, Stanislaw, 723
Levy, Azriel, 1045
Lewis, David, 3,4,21, 77, 79, 81,
111
limitation of size, 20, 28, 98, 145
Magpie, 2,6,8, 14, 16,32,42,56,
84,85,110,122,143
Main Thesis, 7,1517,43,55,100
Malezieu, Nicolas de, 70
many and few, definitions of,
901,136
Martin, Richard M., 10
Massey, Gerald, 63
mathematics as set theory, ix, 6,
12,267,36,534,58,87,93,
94,115,141,1434
maximal connected parts, 227
measurable cardinals, 1378
megethology, 27, 4950, 74,
935, 120, 1369
membership, definition of, 16, 97
mental acts of setforming, 2931
mereologized arithmetic, 958
mereology, 13,7287, and
passim
metaphysics, 547; see also
unofficial axioms of set theory
microcosms, 1217
mixed fusions, 78, 15, 17,21,45,
46,80,112,145
Morton, Adam, 63, 70
mystery about sets, vii, 56,
2938,412,45,4850,53,
547
nice parts, 223, 27
nominalism, 21, 65, 667
nominalistic set theory, 218
nonselfmembers, 8, 1819,27,
634,658
null individual, viii, 1012, 1213
null set, 4, 1015, 17, 18, 21,
323,956,97,98,100,139
Null Set, Axiom of, 100
Index
ontological innocence, 62, 689,
817,93,102
Ontology, LeSniewski's, 723
Oppy, Graham, ix
orderings, extraneous, 12733
ordinals, 11213
overlap, definition of, 73
Pairing, Principle of, 132
pairs, 523,55,57, 71, 102, 117,
122, 1257, 1313, 135
Pair Sets, Axiom of, 1001
parallel postulate, 11314, 115
Peano axioms, vii, 48, 956,
1079,11213,11415,1446
permutations of singletons,
11620
philosophers, follies of, ix, 59
plural quantification, 19,52,
6271,723,87,1012,115,
116, 120, and passim
Pollard, Stephen, 63
possibilities, as abstract
representations, 358, 778; as
worlds, 13,36,37,49,77,86
Possum, 2, 6, 8, 12, 13, 14, 15, 16,
32,35,37,39,42,47,556,82,
84,85,98,109,110,122,143
Power Sets, Axiom of, 93, 104,
105
Prior, A. N., 73
Priority Thesis, 7, 9, 15, 16,99
proper classes, 4, 9, 15, 1819,20,
65,668,98,112,120
proper parts, definition of, 2
properties, 8, 21, 22, 334, 557,
76, 79, 86; see also relations
pseudomembers and
pseudoclasses, 218
pure classes, 1213,323, 112,
117
Putnam, Hilary, ix
qualitative character of singletons,
335,57,1423,147
Quine, Willard V., ix, 412, 111,
132
Ramsey, Frank P., 47
Ramsey sentences, 4654, 13949
rank, 35, 67, 148
Reality, map of, 1921
relations, existence of, 501;
generating, 3841; internal and
external, 368; quantifying
over, viii, 45 54, 12149;
theories of, 512
Replacement, 90, 91, 102
Resnik, Michael, 66
Robbin,joel W., 2930,114
Robinson, Denis, ix
Russell, Bertrand, 63; see also
nonselfmembers
secondorder systems, 63, 66,
,701,114,116,117
Second Thesis, 610, 56
sets, definitions of, 4, 18, 98
Sharvy, Richard, 63
Shoenfield,joseph R., 2930,138
Simons, Peter, 63, 73
singleton, 12, 15, and passim;
identified with its member,
412; as primitive, viiix,
16,31,457,52,54,61,95,
10912,115,13942,145,147
singularism, 659, 70, 116
size, finite and infinite, 889,
154 .
945,967,1067;
inaccessible, 94, 1367, 142,
143,1478; large and small,
8992, 136; limitation of, 20,
28,98,145; of reality, 27,
4950,74,935,120,1369
Skolem, Thoralf, 46, 50, 101
small and large, definitions of,
8991, 136
spacetime, 14,315,46,756,
78,857,89,1423,147
states of affairs, 8, 567
Stenius, Eric, 63
structuralism, viiiix, 4554, 58,
11 011, 13949
substitutional 'quantification', 50,
512,645
successor relation, 48, 49, 534,
10713,114
supervaluations, 14,47
tandem, mappings in, 1235
Tarski, Alfred, 73
Taylor, Barry, ix
time, see spacetime
transitivity, 1,3,5, 74
Trisection, Principle of, 124
tropes, 8, 334, 51, 556, 57, 76
Index
15 5
unicle relation, 61; see also
singleton
union, 18
Unions, Axiom of, 93, 1045
Uniqueness of Composition,
389,74,789,85,100
unit class, see singleton
unithood facts, 567
universals, 8, 21, 22, 334, 51, 76,
79,86
unofficial axioms of set theory,
314,46,49,109,11516
Unrestricted Composition, 78,
1819,74,7981,85,101,112
urelements, 4,15,412,98
van Fraassen, Bas, 14,47
van Inwagen, Peter, ix, 358, 81
von Neumann, john, 98,104,111,
113
Wellordering Principle, 130
Weston, Thomas, 116, 117
Wiener, Norbert, 126
Williams, Donald c., 33, 56, 76
Zermelo, Ernst, 5, 11011
Copyright © David Lewis 1991 First published 1991 Basil Blackwell Ltd 108 Cowley Road, Oxford. OX4 1JF, UK Basil Blackwell, Inc. 3 Cambridge Center Cambridge, Massachusetts 02142. USA All rights reserved. Except for the quotation of short passages for the purposes of criticism and review, no part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic. mechanical, photocopying, recording or otherwise, without the prior permission of the publisher. Except in the United States of America. this book is sold subject to the condition that it shall not, by way of trade or otherwise, be lent, resold, hired out, or otherwise circulated without the publisher's prior consent in any form of binding or cover other than that in which it is published and without a similar condition including this condition being imposed on the subsequent purchaser.
Contents
Preface
VlI
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9
Classes Apart Fusions and Classes Classes and their Subclasses Are there any other Parts of Classes? The Nun Set Consequences of the Main Thesis More Redefinitions Sethood and Proper Classes A Map of Reality Nominalistic Set Theory Revisited
1 1 3 6
10
British Library Cataloguing in.Publication Data
A CIP catalogue record for this book is available from the British Library.
Library of Congress Cataloging in Publication Data
Lewis, David K. Parts of classes/David Lewis. p. cm. Includes bibliographical references. ISBN Q..631176551  ISBN Q..63117656X (pbk.) 1. Set theory. 2. Whole and parts (philosophy) L Title. QA248.L48 1991 511.3'22  dc20 8949771 CIP
15 17 18 19 21 29 29 35 38 41 42
2 The Trouble with Classes 2.1 Mysterious Singletons 2.2 Van Inwagen's Tu Quoque 2.3 On Relations that Generate 2.4 Quine on Urelements and Singletons 2.5 The Lasso Hypothesis
v
Typeset in 12 on 14pt Bembo by Graphicraft Typesetters Ltd., Hong Kong Printed in Great Britain by Billings & Sons Ltd., Worcester
QA
24B .lAB 1991
Contents
2.6 2.7 2.8 3
Ramsifying out the Singleton Function Metaphysics to the Rescue? Credo
45 54 55 61 61 62 71 72 75 81 88 93 93 95 98 100 107 109 112
A Framework for Set Theory 3.1 Desiderata for a Framework 3.2 Plural Quantification 3.3 Choice 3.4 Mereology 3.5 Strife over Mereology 3.6 Composition as Identity 3.7 Distinctions of Size
Preface
4
Set Theory for Mereologists 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 The Size of Reality Mereologized Arithmetic Four Theses Regained Set Theory Regained Ordinary Arithmetic and Mereologized Arithmetic What's in a Name? Intermediate Systems Completeness of Me reologi zed Arithmetic
113
121
Appendix on Pairing by John P. Burgess, A. P. Hazen, and David Lewis Index
There is more mereology in set theory than we usually think. The parts of a class are exactly the subclasses (except that, for this purpose, the null set should not count as a class). The notion of a singleton, or unit set, can serve as the distinctive primitive of set theory. The rest is mereology:a class is the fusion of its singleton subclasses, something is a member of a class iff its singleton is part of that class. If we axiomatize set theory with singleton as primitive (added to an ontologically innocent framework of plural quantification and mereology), our axioms for 'singleton' closely resemble the Peano axioms for 'successor'. From these axioms, we can regain standard iterative set theory. Alas, the notion of a singleton was never properly explained: talk of collecting many into one does not apply to onemembered sets, and in fact introduces us only to the mereology in set theory. I wonder how it is possible for us to understand the primitive notion of singleton, if indeed we really do. In March 1989, after this book was mostly written, I learned belatedly that its main thesis had been anticipated in the
Vll
151
VI
Preface
'ensemble theory' of Harry C. Bunt. 1 He says, as I do, that the theory of part and whole applies to classes; that subclasses are the parts of classes, and hence that singletons  unit classes  are the minimal parts of classes; and that, given the theory of part and whole, the membersingleton relation may replace membership generally as the primitive notion of set theory. There is some significant difference from the start, since Bunt accepts a null individual and rejects individual atoms. But the more important differences come later: in my discussion of the philosophical consequences of the main thesis, and in our quite different ways of formulating set theory within a mereological framework. I think there is difference enough to justify retelling the story. Later in 1989 I had another surprise. I learned how we can, in effect, quantify over relations without benefit of the resources of set theory. (All we need is the framework: plural quantification and mereology.) John P. Burgess discovered one way to do it; A. P. Hazen discovered a different way. It was not quite too late to add an appendix, jointly written with Burgess and Hazen, in which these methods are presented and applied. They open the way to a new 'structuralist' interpretation of set theory, on which the primitive notion of singleton vanishes and my worries about whether it is well understood vanish too. Good news; but not a full answer to my worries. The set theory of the future may go structuralist, if it likes. But I can't very well say that set theory was implicitly structuralist all along, even before the discoveries were made that opened
Preface
the way. The set theory of the present which means the rests on the primitive bulk of presentday mathematics notion of singleton. Is it rotten in its foundation? I dare not make that accusation lightly! Philosophers who repudiate all that they cannot understand have very often gone astray. Maybe my worries are misguided; maybe somehow, I know not how, we have understood the membersingleton relation (and with it membership generally) well enough all along. If so, structuralism is a real solution to an unreal problem, and we might as well go on as before. The book reflects my indecision. I whinge at length about the primitive notion of singleton (section 2.1), but then I mock those philosophers who refuse to take mathematics as they find it (section 2.8). The book mostly proceeds under the working assumption that singleton is a legitimate primitive notion. But also it presents the structuralist alternative, in section 2.6 and in the appendix. (What's said taking singleton as primitive will not in any case go to waste. All but section 4.6 admits of structuralist reinterpretation.) Structuralism is at least a very welcome fallback. But I would much prefer a good answer to my worries about primitive singleton, so that I could in good conscience take mathematics as I find it. I thank all those who have helped by discussion of this material, especially D. M. Armstrong, Donald Baxter, Paul Benacerraf, George Boolos, Phillip Bricker,John P. Burgess, M. J. Cresswell, Jennifer Davoren, Hartry Field, A. P. Hazen, Daniel Isaacson, Mark Johnston, Graham Oppy, Hilary Putnam, W. V. Quine, Denis Robinson, Barry Taylor, and Peter van Inwagen. I thank Nancy Etchemendy for preparing the figures in the appendix. I am indebted to the Melbourne Semantics Group, where this material was first presented in 1984; to Stanford University, where
IX
1 Harry C. Bunt, Mass Terms and Model Theoretic Semantics (Cambridge University Press, 1985), pp. 53 72, 233  301.
Vl1l
whatever is part of a cat is thereby part of a part of the catfusion. It has all cats as parts.1 Fusions and Classes Mereology is the theory of the relation of part to whole.4 and 3. But also 1 See sections 3. The fusion of all cats is that large. 1 One of these kindred notions is that of a mereological fusion. I x . a part identical to the whole. and so must itself be part of the catfusion. for research support under a Santayana Fellowship in 1988. It does have other parts too: all catparts are parts of it. or sum: the whole composed of some given parts. for instance catwhiskers. and nothing else.Preface a more developed version was presented as the 1988 Kant Lectures. January 1990 1 Taking Classes Apart 1. scattered chunk of catstuff which is composed of all the cats there are. and to the Boyce Gibson Memorial Library. and kindred notions. We count it as a part of itself: an improper part.5 for further discussion of mereology. catcells. But the catfusion is the least such thing: it is included as a part in any other one. The catfusion has still other parts. to Harvard University. David Lewis Princeton. There are other things that have all cats as parts. For parthood is transitive. catquarks.
so good. we shouldn't jump to the conclusion that it isn't a fusion. Therefore we learn not to identify the class of As with the fusion of As.Taking Classes Apart it has plenty of proper parts . It has all and only cats as members. the fusion of all catparticles. But I used to think. Since all and only overlappers of cats are overlappers of catparts. We would equivalently define the catfusion as the thing that overlaps all and only those things that overlap some cat. We learned that fusions and classes were two quite different kinds of things. that we learned more. 40. a member of it. And yet the catfusion is made of nothing but cats. Fusions of catparts are parts of it too. lest we identify two different classes with one single fusion. we have this 2 So said David Lewis in Philosophical Papers. whereas the fusion of the As and the fusion of the Bs can· be identical even when none of the As is a B. we shouldn't jump to the conclusion that there must be some unmereological way to make one out of many. I might at least have granted that the partwhole relation applies to classes in a trivial way: even if a class has no proper parts. the fusion of all cats is the same as the fusion of all catfusions. But for now. and so perhaps did you. that remains to be seen. because they are not whole cats. and the fusion of all things that are either catfronthalves or catbackhalves. but that's a different class. Lesser fusions of cats. For classes do have parts: their subclasses. for instance the fusion of my two cats Magpie and Possum. It has no other members.besides the cats and catparts already mentioned. Catparts such as whiskers or cells or quarks are parts of members of it. Fusions of several cats plus several catparts are parts of it. not sets. vol. Therefore we learn not to identify membership with the relation of part to whole. We learned to distinguish two entirely different ways to make one thing out of many: the way that made one fusion out of many parts. They are members of the class of catfusions. 3 . Just because one class isn't composed mereologically out of its many members. are proper parts of the grand fusion of all cats. 1. every part of it overlaps some cat. and the class of Bs with the fusion 2 Classes and their Subclasses of Bs.2 Classes and their Subclasses Mereology does apply to classes. I (Oxford University Press. It is also the fusion of all catmolecules. '2 We even learned to call mereology 'The Calculus ofIndividuals'! All that was a big mistake. Maybe they have other parts as well. And since all and only overlappers of cats are overlappers of catfusions. The class of all cats is something else. p. in this sense: it has no part that is entirely distinct from each and every cat. so that no class was ever a fusion. for instance the fusion of Possum's paws plus whiskers. Just because a class doesn't have all and only its members as parts. in general. The class of As and the class of Bs can never be identical unless the As are all and only the Bs. versus the way that made one class out of many members. 1983). Catparts are indeed members of the class of all catparts. or the fusion of all cattails wherever they be. but again that's a different class. So far. Rather. but they are not themselves members of it.parts not identical to the whole . as I then thought. at least it should have itself as an improper part. but again they are not themselves members of it. Just because a class isn't the mereological fusion of its members. We learned that 'the partwhole relation applies to individuals. whereas a part of a part of something is always a part of it. Fusions of several cats are fusions of members of the class o'f all cats. the fusion of all cats is the same as the fusion of all catparts. A member of a member of something is not. we shouldn't jump to the conclusion that a class has no parts.
that it conforms to common speech. the First Thesis faces no formal obstacles. By 'classes' I mean things that have members. guided by an analogy of formal character between the partwhole relation and the subclass relation. pp. wherefore we are free to claim that there is more than a mere analogy. nor whether any classes are part of the null set. what can I say in its favour? First. 3 Classes and their :}uf)ctasses I numbers is part of the class of natural numbers. and still leave other words unmolested to keep their standard meanings. By 'individuals' I mean things that are members.the null set. therefore. we know it. Set theory is peculiar. Just as a whole divides exhaustively into parts in many different ways. whereas a member of a member is not in general a member. whereas a class divides exhaustively into members in only one way.) Besides. It does seem natural to say that a subclass is part of a class: the class of women is part of the class of human beings. the class of even Warning: I introduced a different idiosyncratic usage of 'class' in my book On the Plurality q{ Worlds (Blackwell. the word is just'Teil'.some individual. that makes it a 'set' that is not a class. We learned. then 4 As noted in D. It all seems so innocent at first! We need only accept that when there are many things. 367. The straightforward explanation is that subclasses just are parts of classes. (Here I am indebted to Ignacio Angelelli. 4 . 1978). Now that you understand what the First Thesis means. and an urelement is any individual other than the null set. or proper and nonempty. I must hasten to tell you that my usage is a little bit idiosyncratic. To a German. for instance in the writings of Cantor and Zermelo. 3 My First Thesis. 1986). but I must if I am to mark the line that matters: the line between the membered and the memberless. and so on. I can hijack 'class' and 'individual'. the First Thesis might seem almost tautological: a standard word for 'subset' is 'Teilmenge'. But that doesn't make it a memberless class. But the subclass relation and the partwhole relation behave alike. Second. Armstrong. As follows: a proper class is a class that is not a member of anything. Standardly. vol. To explain what the First Thesis means. (Or I could impose novel coinages on you. all sets are classes and none are individuals. It does not say whether the null set is part of any classes. sometimes with an implication that the subset is proper. Sometimes.) The devious explanation of what we say is that we speak metaphorically. Therefore there is no such class as the null class. which would probably be still more annoying. I don't mind calling some memberless thing . I hope to show you that the First Thesis will prove fruitful. We have at the very least an analogy of formal character. Universals and Scientific Realism (Cambridge University Press. I am sorry to stray. that membership could not be (a special case of) the partwhole relation because of a difference in formal character.Taking Classes Apart First Thesis: One class is a part of another iff the first is a subclass of the second. so a class divides exhaustively into subclasses in many different ways. literally 'partset'. M. we speak accordingly. Finally. but do not themselves have members. Rather. I shall take up those questions later. 4 Likewise it seem natural to say that a hyperbola has two separate parts and not to take that back when we go on to say that the hyperbola is a class of xy pairs. rightly. II. 50ln. Just as a part of a part is itself a part. so a subclass of a subclass is itself a subclass. has nothing to say yet about the null set. we had more than' enough words. pp. a set is either the null set or else a class that is not a proper class.
Taking Classes Apart
also there is one thing  the class which is just the many taken together. It's hard to object to that. But it turns out later that this manyintoone can't always work, on pain of contradiction yet it's just as h,ard to object to it when it doesn't work as when it does. What's more, the innocent business of making many into one somehow transforms itself into a remarkable making of one into many. Given just one single individual, Magpie or Possum or the null set or what you will, suddenly we find ourselves committed to a vast hierarchy of classes built up from it. Not so innocent after all! It is exactly this ontological extravagance that gives set theory its welcome mathematical power. But, like it or not, it's far from what we bargained for when we first agreed that many can be taken together as one. We could understand set theory much better if we could separate the innocent Dr Jekyll from the extravagant and powerful Mr Hyde. The First Thesis will be our first, and principal, step towards that separation.
I
Are there any other Parts of Classes?
Main Thesis: The parts of a class are all and only its subclasses.
But the Second Thesis seems to me far less evident than the First; it needs an argument. And an argument needs premises. My premises will be the First Thesis, plus three more.
Division Thesis: Reality divides exhaustively into individuals and classes. Priority Thesis: No class is part of any individual. Fusion Thesis: Any fusion of individuals is itself an individuaL
Roughly speaking, the Division Thesis says that there is nothing else except individuals and classes. But that is not exactly right. If we thought that Reality divided exhaustively into animal, vegetable, and mineral, that would not mean that there was no such thing as a salt beef sandwich. The sandwich is no counterexample, because the sandwich itself divides: the beef is animal, the bread is vegetable, and the salt is mineraL Likewise, the Division Thesis permits there to be a mixed thing which is neither an individual nor a class, so long as it divides exhaustively into individuals and classes. I accept a principle of Unrestricted Composition: whenever there are some things, no matter how many or how unrelated or how disparate in character they may be, they have a mereological fusion. (See sections 3.4 and 3.5 for further discussion.) That means that if I accept individuals and I accept classes, I have to accept mereological fusions of individuals and classes. Like the mereological fusion of the front half of a trout plus the back half of a turkey, which is
7
1.3
Are there any other Parts of Classes?
The First Thesis leaves it open that classes might have other parts as well, besides their subclasses. Maybe classes sometimes, or always, have individuals as additional parts: the null set, cat Magpie, Possum's tail (and with it all the tailsegments, cells, quarks, and what not that are parts of Possum's tail). To settle the question, I advance this
Second Thesis: No class has any part that is not a class.
The conjunction of the First and Second Thesis is our
6
Taking Classes Apart
neither fish nor fowl, these things can be mostly ignd'red. They can be left out of the domains of all but our most unrestricted quantifying. They resist concise classification: all we can say is that the salt beef sandwich is part animal, part vegetable, part mineral; the troutturkey is part fish and part fowl; and the mereological fusion of Possum plus the class of all catwhiskers is part individual and part class. Likewise, Reality itself  the mereological fusion of everything  is mixed. It is neither individual nor class, but it divides exhaustively into individuals and classes. Indeed, it divides into one part which is the most inclusive individual and another which is the most inclusive class. If we accept the mixed fusions of individuals and classes, must we also posit some previously ignored classes that have these mixed fusions as members? No; we can hold the mixed fusions to be ineligible for membership. Mixed fusions are forced upon us by the principle of Unrestricted Composition. Classes containing them are not likewise forced upon us by a corresponding principle of unrestricted classformation. That principle is doomed in any case: we dare not say that whenever there are some things, there is a class of them, because there can be no class of all nonselfmembers. Nor are classes containing mixed fusions forced upon us in any other way. Let us indulge our offhand reluctance to believe in them. All I can say to defend the Division Thesis, and it is weak, is that as yet we have no idea of an y third sort of thing that is neither individual nor class nor mixture of the two. Remember what an individual is: not necessarily a commonplace individual like Magpie or Possum, or a quark, or a spacetime point, but anything whatever that has no members but is a member. If you believe in some remarkable nonclasses universals, tropes, abstract simple states of affairs, God, or what you will it makes no difference. They're still individuals, 8
Are there any other Parts of Classes?
however remarkable, so long as they're members of classes and not themselves classes. Maybe the mixed fusions are disqualified from membership. Maybe some classes are 'proper classes' and are disqualified from membership. But rejecting the Division Thesis means positing some new and hitherto unheardof disqualification, applicable this time to things that neither are classes nor have classes as parts, that can make things ineligible to be members. I wouldn't object to such a novel proposal, if there were some good theoretical reason for it. But so long as we have no good reason to innovate, let conservatism rule. The Priority Thesis and the Fusion Thesis reflect our vague notion that somehow the individuals are 'basic' and 'selfcontained' and that the classes are somehow a 'superstructure'; 'first' we have individuals and the classes come 'later'. (In some sense. But it's not that God made the individuals on the first day and the classes not until the second.) Indeed, these two theses may be all the sense that we can extract from that notion. We don't know what classes are made ofthat's what we want to figure out. But we do know what individuals are made of: they're made of various smaller individuals, and nothing else. From the First Thesis, the Division Thesis, the Priority Thesis, and the Fusion Thesis, our Second Thesis follows.
Proof Suppose the Second Thesis false: some class x has a part y that is not a class. If y is an individual, x has an individual as part; if y is a mixed fusion of an individual and a class, then again x has an individual as part; and by the Division Thesis those are the only possibilities. Let z be the fusion of all individuals that are part of x. Then z is an individual, by the Fusion Thesis. Now consider the difference x z, the residue that remains of x after z is removed. (It is the fusion of all parts of x that do not overlap z.) Then x  z has no
9
Taking Classes Apart
individuals as parts, so it is not an individual or a mixed fusion. By the Division Thesis, it must be a class. We now have that x is the fusion of class x  z with an individual z. Since x z is part of x, and not the whole of x (else there wouldn't have been any z to remove), we have that the class x z is a proper part of the class x. So, by the First Thesis, x z must be a proper subclass of x. Then we have v, a member of x but not of x  z. According to standard set theory, we then have u, the class with v as its only member. By the First Thesis, u is part of x but not of x  z; by the Priority Thesis, u is not part of z; so u has some proper part w that does not overlap z. No individual is part of w; so by the Division Thesis, w is a class. By the First Thesis, w is a proper subclass of u. But u, being onemembered, has no proper subclass. This completes a reductio. QED
The Null Set
a role in mereology that corresponds to the role of the null set in set theory. The null individual is supposed to be part of everything whatever. So if we take the fusion of the null individual and something else, we just get the something else back again, because the null individual was part of it already. And even if two things don't overlap in any ordinary way, they still have the null individual as a common part. And it has no proper parts; for if x is part of the null individual, the null individual also is part of x, wherefore x and the null individual are identical. If we accepted the null individual, no doubt we would identify the null set with it, and so conclude that the null set is part of every class. But it is wellnigh unintelligible how anything could behave as the null individual is said to behave. It is a very queer thing indeed, and we have no good reason to believe in it. Such streamlining as it offers in formulating mereology namely, that intersections of things come out welldefined even when they shouldn't  can well be done without. Therefore, reject the null individual; look elsewhere for the null set. Should we perhaps reject the null set as we1l? Is it another misguided posit, meant to streamline the formulation of set theory by behaving in peculiar ways? I think not. It was misguided, perhaps, to call it a 'set'; and I have halfremedied that by declining at least to call it a 'class'. But the null set's behaviour is not, after all, so very peculiar. It is included in and lacking every class just because it lacks members members is not so queer, all individuals do it. Also, the null set is more use than the null individual would be, and some of its services are less easily done without. The null set serves, first, as a denotation of last resort for classterms that fail to denote classes. It is the settheoretical intersection of x and y, where x and y have no members in common (just as the null individual was invented to be the
1I
1.4
The Null Set
A consequence of our Second Thesis is that classes do not have the null set as part. Because it was a memberless member, we counted it as an individual, not a class; therefore it falls under our denial that individuals ever are parts of classes. To be sure, the null set is incfuded in any class x: all its members  all none of them  are among x's members. But it never can be a subclass if it is not even a class. Were we hasty? Should we amend the Second Thesis, and the premises whence we derived it, to let the null set be a part of classes after all? I think not. Some mereologists 5 posit a 'null individual' meant to play
5 For one, R. M. Martin, 'A Homogeneous System for Formal Logic', Journal of Symbolic Logic, 8 (1943), pp. 123, especially p. 3; and 'Of Time and the Null Individual',Journal of Philosophy, 62 (1965), pp. 72335. For another, Bum, Mass Terms and Model Theoretic Semantics, pp. 567.
10
Taking Classes Apart
mereological intersection of x and y when x and y didn't really overlap). It is the class of all selfmembers, since there are no selfmembers; it is the class of real numbers x such that x 2 + 1 = o. This service is a mere convenience. It would be better to do without it than to purchase it by believing in some queer entity. But, second, the null set also serves as an element of last resort. Suppose you repudiate the null set, and make do entirely with classes generated from some ordinary individual  let it be cat Possum. And suppose you reduce mathematical objects to classes, as usual  to do that, it matters not whether you begin with the null set or with Possum. Instead of the pure sets, you will have Possum himself; you will have Possum's singleton, or unit class, that contains him as its sole member; you will have the class of Possum's singleton and Possum himself; you will have the new singleton of that class; you will have the class of that new singleton and the old singleton and Possum; and so on ad infinitum and far beyond, until you have enough modelling clay to make the whole of mathematics. But now your confidence in the existence of the entire mathematical realm  all those numbers, matrices, curves, homeomorphisms, the lot  rests on your confidence that Possum exists! If I was fibbing when I said I had a cat of that name, shall mathematics fall? And where was mathematics before Possum was born to redeem it? If mathematics is to be safe, it had better rest on a surer foundation. Possum is not a sure thing. The null set is. You'd better believe in it, and with the utmost confidence; for then you can believe with equal confidence in its singleton, the class of that singleton and the null set, the new singleton of that class, the class of that new singleton and the old singleton and the null set, and so on until you have enough modelling clay to make the whole of mathematics. (The null individual could do us no corresponding service.
12
The Null Set
If we have the null set, we have morethancountless other things besides. If we had the null individual, we would have the null individual and that would be that. Mereology does not spin extravagant realms of being out of just one single thing.) Must we then accept the null set as a most extraordinary individual, a little speck of sheer nothingness, a sort of black hole in the fabric of Reality itself? Not really. We needn't be ontologically serious about the null set. It is useful to have a name that is guaranteed to denote some individual, but it needn't be a special individual with a whiff of nothingness about it. Ordinary individuals suffice. In fact, any individual will do for the null set  even Possum. Like any individual, he has the main qualification for the job  memberlessness. As for the second qualification, guaranteed existence, that is not really a qualification of the jobholder itself, rather it is a requirement on our method of selection. To guarantee that we will select some existing individual to be the null set, we don't need to select something that's guaranteed to exist. It is enough to make sure that we select from among whatever things happen to exist. We can select Possum, contingent though he be, so long as our method of selection would have found something else had Possum not been there to find. (And what if there had existed no ordinary individuals whatsoever?  In that case, maybe we can let mathematics fall. Just how much security do we really need?)6
6 I myself hold a thesis of plurality of worlds according to which (1) Possum could not have failed to exist altogether, though he might have been off in some other world and not a world mate of ours; and (2) it is necessary that there be something  some individual  and not rather nothing. But I shall not rely on this thesis here, for fear that not all of you accept it.
the individuals are all and only the parts of the null set. and the 7 C Here I apply Bas van Fraassen's method of ' super valuations'. 7 If I left the choice unmade. no more one place than another. Journal oj Philosophy. 63 (1966). It also introduces a handy name for one of the main subdivisions of Reality. A class has its singleton subclasses as atomic parts. Instead of the unhelpful definition of the null set as the set without members. and what's true on some and false on others could be left indeterminate as to truth. An arbitrary choice might be left unmade. are the proper part of the null set. one for each of its members. 1. Instead. It's as far as can be from the notion that the null set is a speck of nothingness. It may be that this choice of the null set offends against some 'intuition' you had that the null set was small and nondescript.Taking Classes Apart The choice of an individual to serve as the null set is arbitrary. Possum's singleton has Possum as its sole member. So the singletons correspond oneone with the individuals and sets. If the null set is the fusion of all individuals. and proceed. And. and Fusion Theses. We could say that what's true is what's true on all ways of making the choice. or singletons. Priority. then x must have one or more singletons as parts. nothing has two singletons. Anything that can be a member of a class has a singleton: every individual has a singleton. But I prefer not to leave the arbitrary choice unmade. I think such an 'intuition' has no ground or authority. and likewise all other individuals. pp.5 Consequences of the Main Thesis Redefinition: The null set is the fusion of all individuals. And the ure1ements. by the Division. It has no subclasses except itself !herefore it is a mereologi. else the rest of x would be a part of y with no singletons as parts. But for myself.and those mixed things that are part individual and part class. no proper parts. TruthValue Gaps and Free Logic'. J adopt this Consequences of the Main Thesis fusion of all individuals is perfectly suitable to serve as the null set. to onemembered classes: unit classes. individuals other than the null set. of course. then. so a part of a 14 15 . 'Singular Terms. That's an easy selection to specify. and so does every set. We may agree to disagree. If you think I am offering you a substitute for the null set as you have conceived of it hitherto. no harm done. The only things that lack singletons are the proper classes classes that are not members of anything. In fact x must consist entirely of singletons. This applies. likewise Magpie. 48195. if x is part of a class y. it respects our 'intuition' that the null set is no place in particular. See van Fraassen. For what it's worth. namely subclasses with more than one member. and so could not be a subclass of y. could not be a class. and that's all to the good. It usually has other parts that are not singletons. and likewise for any other singleton. what's false is what's false on all ways. in particular. However. and afortiori not members of singletons . ?ur Main Thesis says that the parts of a class are all and only Its subclasses. I make it arbitrarily! as follows. 'null set' would be harmlessly ambiguous: Possum would be not unequivocally not the null set. . else x could have no members. and it's guaranteed to select an individual to be the null set if there exist any individuals at all. Likewise the singleton of Possum's singleton is an atom.cal atom: it has no parts except Itself.
And so it goes.Theoretic Semantics. 8 Once we redefine membership and null set. it has no singletons as parts. and the singleton of the aforementioned class. For example. Redefinition: A class is any fusion of singletons. Magpie's singleton's singleton. More Redefinitions singletons as part iff it has one or more singletons as parts. Mass Terms and Model. the class of the two cats Possum and Magpie is the fusion of Possum's singleton and Magpie's singleton. and hence the fusion. we have this new definition of a class. 8 1. by our Division Thesis. is a mixed fusion. Hence the null set. an individual is anything that overlaps no classes. A class is the union. See Bunt. The remaining case. Magpie's singleton. And although a class. and the aforementioned class is the fusion of Possum's singleton. by our Main Thesis. or (2) x and yare classes and x is part ofy. as we have arbitrarily chosen it. Unfortunately. which is not one of its subclasses and not one of its parts. Redefinition: An individual is anything that has no singletons as parts. something is an individual iff it has no class as part. and so. The members of a class are exactly those things whose singletons are subclasses of it. we probably would not want to call that a case of inclusion. The main case is that a class includes a subclass by having it as a part. Inclusion. but it too includes the null set. Equivalently. or the null set. is the fusion of all things that do not overlap the fusion of all singletons. Reality minus that fusion. So we have another new definition. So when we define membership (formerly primitive) in terms of the notion of singleton. By the Division and Priority Theses. since singletons are atoms. It does have singletons as parts. equivalently. The null set is not a class.Taking Classes Apart class that is not a subclass. p. . of the singletons of its members. and it has a fusion of Bunt does without this restriction. we can proceed as usual to all other notions of set theory. 61. It is what's left of Reality after all the singletons are removed. are exactly those things whose singletons are parts of it. part individual and part class. we must write in a restriction to classes. since classes are fusions of singletons. by our Priority Thesis. But sometimes it is instructive not to go the long way round via membership. So we have: Redefinition: y includes x iff (1) x is the null set and y is the null set or a class. yet probably we would not want to say that it has members. The class whose three members are Possum. which is to say that it is the mereological difference. and accordingly grants that his concept of membership differs from that in set theory. an individual is anything that overlaps no singletons. But also a class includes the null set. may be part of a mixed fusion. the special case of the null set keeps complicating the story. If we take the notion of a singleton henceforth as primitive.6 More Redtdinitions Dtifinition: x is a member of y iff Y is a class and the singleton of x is part of y. but rather to define other notions directly. An individual has no members and.
otherwise it is the fusion of those of them that are classes. it is the nul1 set if each of them is the null set. but somehow we have no fusion of all these singletons. or else it is the null set. Redtifinition: A proper class is a class that has no singleton. We do not go out of our way to posit them. The main case is that a union of classes is their fusion. Those above the gap are singletons. that we can get all the power we need by resorting to plural quantification. be they useful or be they useless. For instance. according to the standard principle of Fundierung). is to restrict not composition. if there is any other remedy.7 Sethood and Proper Classes A class is a set iff it is a member of something. and I agree. and if x has a singleton. We dare not allow a set of all sets that are nonselfmembers . or mixed fusions. it must have a singleton to be part of that thing. so we should not be surprised or disturbed to find that it needs restricting. the singleton is something that x is a member of. so sethood requires another disjunctive definition. Redtifinition: Something is a set iff either it is a class that has a singleton.Taking Classes Apart Union. But the null set also counts as a set. on pain of Russell's paradox. Mereology per se is unproblematic.) The better remedy. Unlike composition. that we posit the proper classes because of their utility in formulating powerful systems of set theory. it is a proper class. otherwise it is a proper class. So we have this: A Map of Reality sets. The dots are mer:eological atoms. it has no singleton. note well. (Just as it would be unduly drastic to solve problems in quantum physics by mutilating logic. but this proper class does not in turn have a singleton. which I have adopted. 43049. pp. Although it is indeed a nonselfmember (everything is. the making of singletons is illunderstood to begin with. so it would be unduly drastic to stop the paradoxes by mutilating mereology. 94 (1985). so we have no class of all these sets.8 A Map of Reality Figure 1 is a map of all of Reality. I do not say. and not to blame for the settheoretical paradoxes. but there are two alternative ways to avoid it. thereby obtaining a proper class of sets. and other individuals. rather we can't keep them away. We can fuse all the singletons of sets that are nonselfmembers. 'To Be is To Be the Value of A Variable (or To Be Some Values of Some Variables)" Journal of Philosophy. the class of all sets that are nonselfmembers had better not be a set. One way would be to restrict composition: we have all the sets that are nonselfmembers and we have a singleton of each of these 18 1. 9 And yet we have the proper classes willynilly. may not. pp. 1. If x is a member of something. So it isn't. Philosophical Review. 'Nominalist Platonism'. George Boolos has argued. that won't make it a selfmember unless it is a set. and it cannot be a member of anything. . But the null set also may enter into unions. but rather the making of singletons. given our Main Thesis and Unrestricted Composition. or problems in the philosophy of mind and language by mutilating mathematics. those below are 9 See Boolos. But there is no good independent reason to restrict composition. 32744. Redtifinition: The union of one or more things is defined iff each of them is either a class or the null set. So a class is a set if it has a singleton. 81 (1984).
.. or some of each is an individuaL (Maybe there isn't any gunk... •. Each individual has its singleton........... it is made of all the atomic individuals plus all of the atomless gunk. It is the universal class... p...... it is too big to be a set.... or maybe there aren't any atomic individuals.... but only for purposes of a formal comparison of systems....... somewhere up just above the gap.. 291)... by the standard principle of Limitation of Size. every singleton is part of it... are sets.. ••......... A set is a member of just those classes that have its singleton as a part.9 Nominalistic Set Theory Revisited Years ago. must consist of gunk... That is why the part of the map above the gap must be so much bigger: so that there will be enough singletons to go around.. in the Harvard fashion.. each one being just one dot.. It is a proper class..... and by the standard principle of Fundierung. ignored but undenied. to a limited extent. save only those things that are not members of anything.... I find this assumption hasty.••... Anything made of the dots below the gap. Anything made of the dots above the gap is a class.... and these classes have their own singletons. 21 . (That's why there's no top.... for instance he cannot countenance spacetime points (unless they are singletons). The blob at the bottom is atomless gunk: an individual whose parts all have further proper parts..Theoretic Semantics. )10 The biggest one is the null set.. therefore everything is a member of it....... . So his individuals.....Taking Classes Apart Nominalistic Set Theory Revisited gunk. pp..) The smallest classes are the singletons..•... let a set's singleton always go just above the set itself..... 4 (1970)..... 'Nominalistic Set Theory... by mereological means. and is a member of all classes that have that singleton as a part.had nothing to do with it. Many other classes likewise are too big to be sets. I wrote a paper called 'Nominalistic Set Theory'..... 11 David Lewis...... To make sure of this. it can never itself be one of the classes that its singleton is part of. I argued that we could imitate set theory. the fusion of all individuals. Nous....... The smaller classes. The biggest class is the thing made of all the dots above the gap...... which are dots somewhere above the gap. 60)..repudiation of universals . These have no singletons and are not members of anything...•... or ma ybe there are both. 22540.. nominalism in the traditional sense of the word .....••.. the blob of 20 1.. Straddling the gap are the 'salt beef sandwiches': the mixed fusions.•. Figure 1 atomic individuals.. not a member of anything......... if there arc any....•.... Later he reopens the question (p....... and of course these are all small enough to be sets.... and that is the case shown. .ll By 'nominalistic' I meant 'setless'... and that's why there are more and more dots as .... and not a set.....we ascend.. .. consisting partly of dots or gunk from below and partly of dots from above..... he takes it as axiomatic that every atom is a singleton (Mass Terms and Model. .. however.. One reason why the fusion of several things is different 10 Bunt disagrees.
(The 'first stab' was the case where I chose the coding to be identity. For sake of. have a into parts. Then fusions ?f . in general. No two atoms overlap. four diagonal. Each point in the grid is next to eight others: two vertical neighbours. Two things touch iff some point of one is next to some point of the other.) I got imitations of set theory. that pOints of one are arbitrarily close to points of the other. Say it like this: we have a pseudoclass that has all and only those things as its pseudomembers. If nice parts are maximal connected parts. we will have a fusion that has all and only those things as nice parts. The closest we can come is to get back its several maximal connected parts. However.Taking Classes Apart from the of them is that a fusion does not. Suppose the things of interest to us are made out of points in an infinite grid. and one cat is never a proper part of the fusion of several others. That is because one cat is never a proper part of another. Something x is a code for something y iff y is the interior of x. it is because cats never overlap. Nice parts are maximal connected parts. for instance. More generally. I chose maximal connected parts as my 'nice parts' and tried various geometrically definable codings.decomposition into its maximal connected parts. Suppose we can encode the things we want to have as the pseudomembers of our pseudoclasses. If we have disconnected things. Maximal spatiotemporally connected parts are nice. but they worked only under especially favourable conditions. Thus: suppose we have two disconnected things a and b. A fusion of cats has a unique decomposition Into cats. Sometimes other unwelcome things happen instead. so any fusIon of atoms has a unique decomposition into its atoms. atoms are nice. this example. it may happen that If we dIStingUIsh some parts of a fusion as nice then a fusion will have a unique decomposition into nIce parts. Then we can say that x is a pseudo member of w. 22 Nominalistic Set Theory Revisited This first stab at setless theory doesn't go far. so a fusion has a UnIque . Cats are nice. we can make connected codes for them by wrapping them in skins and running strings from one part to the other. Define some scheme of coding on which things that are not themselves nice may nevertheless have nice codes. because the only things that can qualify as pseudomembers of pseudoclasses are the things that will be nice parts. in this way: whenever we have some suitable things. Then something x is connected iff there are no y and z such that x is the fusion of y and z and y does not touch z' x is a maximal connected part of w iff x is a connected part of w that is a proper part of any other connected part of w. each composed of the points 4 r . No two connected parts of a fusion overlap. on some specific definition of 'nice part'. igriore universals. Suppose we have some notion of what it means two things to touch spatiotemporally: roughly. let us assume that spatiotemporal things C?nSlSt entirely of pointsized particulars.nIce parts imitate classes. two horizontal. iff some nice part of w is a code for x. That's the best case. then only a connected thing can ever be a pseudomember of a pseudoclass. Example. If we try to put in a disconnected thing. we resort to coding. and the relation 'nice part of' Imitates the membership relation. To improve matters. The interior of something consists of those points of it which are not next to any points that are not parts of it. If such there be. . In 'Nominalistic Set Theory'. relative to a given definition of 'nice part' and a given scheme of coding. the fusion of cats c1 • c2 • c3 •• •• is identical to the fusion of cats d1 • d2 • d3 •• •• only if each if the cs is one of the ds and each of the ds is one of the cs. . we cannot get that thing itself back as a pseudomember.
.. . .... · . •• . .. · . . . •••• ·. . ·. . . . . . •• •• •• . . ·.. . .... . . · .. · • • • • • • ·. .. . .. . . . . · .. ..· .. . . . ·. . . .. . . . • • • • • • db b d·· ·cccc·· • • • d d d d·· ·. .. . . . . \.. . . . . · ... . · . . ·. . . . .... · . . . .. . . these two codes are the two nice parts of the fusion a + c + b + d... .. . . . .. .. .. . . • • · . . .. .. • ·. . . . . . . . . . . · .· . · . the fusion b + d is a connected code for b. . . . .. · . . . • • . Then we can form a pseudoclass as shown in figure 3. . · .c e • e e e eeeee 'e c c c c e • • • e d d d de· • • e cccce· • • e dddde • • • e dlb bid e • e dlb bid e e ecaace edbbde·· ·e c a ace • • • e dbbde ecccce e · · • e c c c c e · · • e dddde e d'eee e eeeed e ·eeeeee· eeeeee· 'e d er.. . •[bbl. . · . .. . . . . . . . · . .. . .... ede d e • ed e • • • e • • ede • ed e • ed Ie e e e e e e e e e e e e e d e • edddddddddddddddd e • eeeeeeeeeeeeeeeeee· • • • • • • • • • • • • • • • j j j j j j h h h h j h j j j j h j • • j h j • • j h j • • j h j • • j j hh j j • • • • j h j • • j h j • • j h j • • j h j • • j h j • • j h j • • j j hh j h j h j h j j j j j • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • Figure 3 so labelled as shown in figure 2. . ·. . .... . · · . The coding is redundant: the strings could have taken different routes. · . . · . · .. . .. . . .·. . .. ·. . . ·. · .... . . . . . .••• jij············jij··· . . . · .... . . . · . · . . . . . . • • • · . ... ·.. . . • • • • • •["bbl.. .. • • • c c c c • • • • • dddd ' . .. . .. .. . . eeeeeeeeeeeeeeeeee· ecccccccccccccccce e c eeeeeeeeeeeeee c e • e c e • • ec e • e c e • ec e • e c e • • • • • • e eeeeee ceeee· e . . . ·. . · • •• •• • · . . ... . d • • • • d r• • • • • • • • • • • • • • d • • • • • • • • • • • • • • d • • • • • • • • • • • • d • • • • • • • • • • • • • • d • ••••••• dddddddddddddddd· • • • cJa3lc • • • • • dfbbld • • • • • ••• • • • c c c c • • • • • dddd • • ·. · . . . . . Any fusion whose nice parts are codes for a and for b is a pseudoclass of a and b. .=. . .. . . . .. . • • • • • • • • • • • '. . . . . . Figure 2 Taking Classes Apart · . . .. ... . . . . . •• •• •• •• •• •• •• d·· •• •• •• •• •• •• •• •• •• •• •• •• dd •• · . . . · . .. . · .... .. . . . • • • . · . .. another reason ""jj""j j •• ejhj •• ••••••••• hj • • • j j j j j j j h j j j j ••• j j j j j j j'hhil h j • • • j i i i i j j h j ••• j 1 1 1 1 j j h[Tf]h j • •• j j j j j j h[Tf]h j ••• j j hluJhJ ••• j iLuJi j'. . . · . . . ·. . .. . . . . . . . . .. ·. . · . . .••. · . •• •• · . d·· •• ·. .. . . .. ... .. . .. . · . .. . •• ·. .. . · . j h h h h j · · · j i i i i j j j j j j j • • • j j j j i j • • • j j j j j j ••• j i T l l l ••••••••• • • • • • • • • • • • • j i J ••• • • • • • • • • • • • • jij •• ···········jij··· I· . . . . . .·. . . .=. . . . . · . .. . . .·. . . . . . . . . . · .. • •• • . . . •• •• •• •• • • · .. ... .. ..·... . . . .. . · ·• • · . . ... .. . . . . • jaal.. •• · . .. The fusion a + c is a connected code for a.. .•.. Figure 4 j ijjjjjjjjjjjjjj ij··· k k k • • • • • • • • • j iii iii iii iii iii i j • • • kkk • • • • • • • • • j j j j j j j j j j j j j j j j j j • • • ' 'kkk" ••••••••• . ••. • • ·. So we cannot speak of the pseudoclass of a and b. wherefore that fusion is a pseudoclass of a and b.. ... .... . .. · . · . . .. . · . · . . .. . . . . . • • · . •• ·raal· · . ·. . . .... • cc c c c c c c c cc c cc c c • • • • • • • • • • c·· • • • • • • • • • • • • c • • • • • • • • • • c • • • • • • • • • • • • • • c • • • • • • • • • • • • • • • • • • ·cccc·· d d d d·· • c[a3lc • • d/b bid. · . .·. .. . · . j hluJh j ••• j J jhhhhj ••• j i i i i j .. .. . ·. . .. . . ... . . . • ·. . .. . .. Besides the lack of a unique decomposition. .. ... .. .. .. . . . i!"iiP hi1i1 · . . . . . . · . . . . . · ..
a code for P. if we are not to run out of codes.so far as possible. there are only continuum many different fusions. so let the codes be small. because whenever we have a fusion of atoms. What if the result of going up the hierarchy is that our grid gets too crowded to make room for any more strings and skins? (How much would it help to work in more than two dimensions?) What if we want to make a pseudoclass of things that touch? Or almost touch? Or worse. Not all codefusions can have codes. but the mathematical realm is much bigger than that.was a pseudoclass of a and b. and the fusions of codes. then fusion plus coding do offer at least the beginning of a hierarchy. There's a problem: when P turns up as a pseudomember of the pseudoclass of P and Q. and further broken up into nice parts and decoded? How could we tell whether k. To see this.call it P . and never have to to define it. let them be atoms. But there must be a limitation. n's code is part of n iff n is normal. By Unrestricted Composition. have their own codes in turn . we have the fusion of all codes of normal codefusions. Farewell honest toill What conditions would we like it to meet? We'd like to avoid crowding. out of a countable Nominalistic Set Theory Revisited grid of points. classes of individuals. classes of classes of individuals (or mixed classes. By definition of normality. But if coding can be iterated. Now a+c+b+d+e+j+h+g+i+j. it is enough to remember that. was meant to be further decoded? It could be: it is a code for its central point. the great majority of codefusions must remain codeless. is the pseudoclass of P and Q. its atoms are nice parts of it. how can we tell that it's meant to be taken as a code for a pseudoclass. n's code is part of n iff n is not normal. Has n a code? Then by definition of n. ad infinitum and beyond. You can while away many an hour.) We must avoid ambiguity. and so on. By attaching a skin and strings. It is easy to think of more problems. QED In fact. But you cannot hope to arrive at a mereologicalcumgeometrical reconstruction of the entire realm of mathematical objects. (Something that is a maximal connected part of one fusion may not be a maximal connected part of another. So let the codes be entirely distinct from the individuals. Likewisej+h+g+icall it Q . So let the encoded things and their codes correspond oneone. Not so for the niceness of atoms. overlap? What if we want to make a pseudoclass of a disconnected thing and something that completely surrounds one of its parts? You can define ever more complicated geometrical codings to cover ever more difficult cases. One way to ration a 26 . We'd also like to avoid redundancy. in order to use our supply of codes efficiently.is a pseudoclass ofj and g andj+ h + g+ i + j is a code for Q. We'd like to avoid any confusion between our original individuals and our codes. as shown in figure 4. if it turned up as a pseudomember of some other pseudoclass. So let the codes. We conclude that n is a codefusion that has no code. its niceness is relative to the fusion. which has these two codes as its nice parts. In fact. call it n. we can make a+c+b+d+e. Proof We apply the reasoning of Cantor's Theorem and Russell's Paradox. But imagine now that we could have a coding for free. containing classes of individuals and also individuals). And we could go on a little longer. We saw that a + c + b + d . We'd like to be able to iterate coding and build up a hierarchy. Call a codefusion normal iff it has a code and its code is not part of it.Taking Classes Apart wh y the fusion of several things differs from the class of them is that fusion offers nothing to correspond to the settheoretical hierarchy of individuals. or fusions of codes.
. To this day.e. then the program of setless set theory would succeed. the rest are proper pseudoclasses. a bunch of grapes. Codefusions are pseudoclasses. I know not how heaven has granted us an ideal coding. i. and that is what our setless imitation is meant to imitate. cd. 'Roughly speaking. To be sure.that is. he is apt to be told something similar. ' (Robbin). with singletons as their nice . 'A pack of wolves. or instead. or a flock of pigeons are all examples of sets of things' (Halmos). he will be given some familiar examples: ' .Taking Classes Apart limited supply of codes is to allocate them only to the comparativel.. I say they would be the real thing. leaving the continuum many infinite codefusions uncoded. Alternative principles of rationing might allow codes to some few of the infinite fusions. because the great majority of codefusIons are not smalL To take a miniature example: if we had countably many codes. that code is a pseudosingleton. for example. and one that provided codes for all manner of ordinary individuals. which can be thought of as one. [It] is formed by gathering together certain objects to form a single object' (Shoenfield). 1932). Something is a member of a class iff some nice part of that class is a code for that thing.. only heaven would grant us an ideal coding. Then the great majority of would go without. Gesammelte Abhandlungen mathematischen und philosophischen Inhalts. 2 The Trouble with Classes 2. 204. Those with codes of their own are pseudosets. this of Limitation of Size is not the only possible way of selecting a favoured few codefusions to receive codes. Joseph R.. A set or class is 'constituted by objects thought of together' (Kleene)..parts. atomic . The pseudoclasses made by mereological fusion of codes would be every bit as good as the real thing. Classes are codefusions. a totality of definite elements that can be combined into a whole by a law'. We call the 'singletons'. Ernst Zermdo (Springer.. If setless set theory works that well. Somehow. But standard iterative set theory uses Limitation of Size.y small codefusions..1 Mysterious Singletons Cantor taught that a set is a 'many. Maybe also..l 1 Georg Cantor. p. Also there is the null pseudoset. the set of all even numbers is considered to be just as real as any particular even number such as 2 or 16' (Robbin). 'A set is a collection of objects . Whenever something has a code. its classes are no longer pseudo. then each of the countably finite codefusions could have one. . when a student is first introduced to set theory. There is nothing we know about the real classes that distinguishes them from the 'imitations'. one that satIsfied all the desiderata just listed. a set is a collection of objects and is thought to have an independent existence of its own . something or other that has no codes as parts and therefore has no pseudomembers. or deny codes to some or all of the finite ones.
they were entirely silent. but your money runs out after you buy the first painting on your list. wherever we go. our utter ignorance about the nature of the singletons amounts to utter ignorance about the nature of classes generally. That's where we get the many into one. . Dr Jekyll was there to welcome us. we never see them or stumble over them. you hang them in a special room. Joel W.the singletons that are the building blocks of all classes . or some sort of sketchy mental map of the whole of Shoenficld. you even publish a catalogue. 238. That might not be quite so bad if the singletons were a very special case. But as to what is distinctively settheoretical . 'many'. 171. to make several things into a class. What do we know about singletons when we know only that they are atoms. Halmos. so that the vast majority of them must have somehow got 'formed' with absolutely no attention or assistance from us. when we know only that they are composed of these atoms about which we know next to nothing? Set theory has its unofficial axioms. At least we'd know about the rest of the classes. But why do we think this? Perhaps because. p. and nothing about the nature of their relation to their elements. Robbin. Paul R. traditional remarks about the nature of classes.I stress the plural. Mr Hyde kept hidden. The job is far too big for us. Stephen C. Nor did any of those familiar examples concern singlemembered sets.The Trouble with Classes But after a time.But this thought is worse than useless. the singleton. . and he has another single thing. Daniel Isaacson suggests this analogy: How do you make several paintings into an art collection? Maybe you make a plan. Mysterious Singletons settheoretical Reality. If you do the same. Sooner or later our student will hear that there are countless classes. But there just cannot be anything that we've done to all the classes one at a time.to be 'combined' or 'collected' or 'gathered together' into one. But since all classes are fusions of singletons. 135. Naive Set Theory (Van Nostrand. We understand how bigger classes are composed of their singleton atoms. Maybe we've formed a few mental representations of a few very special classes. most of them infinite and miscellaneous. but are passed along heedlessly from one author to another. and nothing he was told gives him the slightest guidance about what that one thing has to do with the other. This time. They are never argued. His introductory lesson just does not apply. the element. Benjamin. and wholly distinct from the familiar individuals? What do we know about other classes. One of these unofficial axioms says that the classes are nowhere: they are outside of space and time. Rather. the combining or collecting or gathering.have only a single member. Mathematical Logic: A First Course (W. just do the same to a single thing and you make it into a singleton. or a theory of them. or to be 'thought of together as one'. 1967). Must set theory rest on theology?) We were told nothing about the nature of the singletons. you have a collection that consists of a single painting. in action or in thought. the unfortunate student is told that some classes . Mathematical Logic (AddisonWesley. Maybe we've formed a general concept of classes.the singletons . if ever there was one. Kleene. For all those allusions to human activity in the forming of classes are a bum steer. He has no elements or objects . you buy the paintings. That's the easy part: just mereology. Those introductory remarks (apart from the misguided allusions to human activity) introduced us only to the mereology in set theory. A. (He might think: whatever it is that you do. p. p. But ma ybe they are invisible and intangible. p. he has no. 1. 1967). and nothing over and above the singletons they're made of. Here is a just cause for student protest. 1969). 1960). he has just one single thing. Mathematical Logic (John Wiley.
it is especially the 'pure' classes that we have in mind. and if we think we know that classes have no distinctive intrinsic character. at any rate. do those differences in any way reflect differenc6s between the character of their members? Do they involve any of the same qualities that distinguish individuals from one another? Again we cannot argue the case one way or the other. 'On the Elements of Being' in Principles of Empirical Realism (Charles Thomas.2 H . the class of the null set and its singleton. Since members of singletons occupy extended spatiotemporal regions. or to be a fusion of atoms. I think. If Mysterious Singletons the null set is the fusion of all individuals. If every singleton was where its member was. The class of Magpie and Possum would be divided: the part of it that is Magpie's singleton would ocupy the region where Magpie is. Perhaps when we say that classes are outside of space and time. I don't say the classes are in space and time. as in Donald C. and singletons are atoms. and that would certainly be peculiar. but by no means outside of space and time. Abstract Particulars . the null set is no genuine class. 2 2 Some philosophers propose that things have their qualities by having them as parts. just as the conjurer's dupes go too fast from not seeing the stooge's head to thinking they see that the stooge is headless. I say we're in the sad fix that we haven't a clue whether they are or whether they aren't. or one seventeenmembered class (a seventeenfold fusion of singletons) from another. indeed. rather than nowhere. The qualities might be repeatable universals. in general. only a 'set' by courtesy. and so on. as I arbitrarily stipulated. We know that's not quite right: to be an atom. it would presumably be an obnoxiously arbitrary matter where it was. But not more peculiar. then it is everywhere. Another unofficial axiom says that classes have nothing much by way of intrinsic character. or to be a fusion of exactly seventeen atoms. are matters of intrinsic character. I don't say they aren't. are nowhere. differ in their intrinsic character? If they do. and finally we conclude that all the pure classes must share the location or un location of the null set that is their common ancestor. then. Where? If Possum's singleton were elsewhere than Possum himself. classes would be where their members were. 1951). and Keith Campbell. every singleton is just where its member is. We go much too fast from not knowing whether they are to thinking we know they are not.The Trouble with Classes Maybe they can share their locations with other things. or they might themselves be particular. Are all singletons exact intrinsic duplicates? Or do they sometimes. because we start by thinking that classes are where their members are. If it's in Footscray. However. or do they always. they too are everywhere nowhere in particular. than being nowhere at all we get a choice of equal evils. its singleton. we conclude that a class with no members must be nowhere. Maybe we think that these classes. and so the entire class would occupy the entire spatiotemporal region where Magpie and Possum are. as in the main system of Nelson Goodman. Maybe they are somewhere. its singleton's singleton. If the pure classes share its location. probably that's like thinking we know that the stooge is headless. that would have to mean that something can occupy an extended region otherwise than by having different parts that occupy different parts of the region. to be sure. the part that is Possum's singleton would occupy the region where Possum is. these are not matters that distinguish one singleton from another. 1966). why there instead of Burwood? But perhaps Possum's singleton is just where Possum is. But even if genuine classes are where their members are. These are the classes built up from the null set and nothing else: the null set. and cannot reject either hypothesis by pointing to the repugnancy of the other. Williams. Perhaps. The Structure of Appearance (Harvard University Press.
says the dictionary. That's just to go in a circle. because a singleton.the vertical in our map of Reality .2 Van Inwagen' 5 Tu Quoque Elsewhere I have complained about the difficulty of understanding the relation that allegedly relates concrete things to the abstract simple possibilities . 1990). If classes are outside space and time. like the spatiatemporal relations of distance? Or is it an internal relation. It seems that we have no alternative but to suppose that the relation of singleton to member holds in virtue of qualities or external relations of which we have no conception whatsoever. they cannot differ in their internal relations to any third thing.) We cannot use the internal structure of a singleton to encode the information which thing it has as its member. There just aren't enough familiar qualities to encode that much information. Ethereal entities are 'light. Van [nwagen's Tu Quoque and so are just alike in their spatiotemporal relation to any third thing. Nor can we use familiar qualities that the singleton might conceivably share with familiar individuals. we are ill placed even to begin to understand the relation of a thing to its singleton. platonic. Still it would not follow that a singleton had no qualitative character. Yet we think we do somehow understand what it means for a singleton to have a member! Finally. Even if we suppose that singletons have spatiotemporallocations to help carry the information.membership . of course . Is it an external relation. 2. We know what to call it. yet something is the singleton of one but not the other. as sometimes happens. we know not what. 2.3 we should ask which are they: ethereal or platonic? The ether is everywhere. and each of them is unique. it cannot be entirely a matter ofspatiotemporal relations: Possum and his singleton occupy the same region on this hypothesis. Even if classes are where their members are. has no internal structure. that distinguishes the singleton of x from all other singletons. whereas the forms are changeless and most fully real (whatever that means). A mereologically atomic singleton could have no qualities as proper parts. rank' in the settheoretical hierarchy . as well as the same spatiotemporallocation. airy or tenuous'. We've named a property. pseudo entities'. it's no good saying that a singleton has x as its member because it has the property: being the singleton of x. and one bit of it is pretty much like another.but that is all. It might be a quality.because the singleton of the singleton of the singleton of Possum and the singleton of the class of Possum and his singleton will be at the same rank. that would be no small advance! Because we know so little about the singletons. but all we know about the property that bears this name is that it's the property. (N or will it help to add an extra quasispatial dimension of . then membership cannot be at all a matter of spatiotemporal relations between the thing and the singleton. being an atom.The Trouble with Classes Sometimes our offhand opinions about the nature of classes don't even agree with one another. When Nelson Goodman finds the notion of classes 'essentially incomprehensible' and says that he 'will not willingly use apparatus that peoples his world with a host of ethereal. The Structure of Appearance. 3 Goodman. like a relation of intrinsic similarity or difference? Or is it a combination of the two? Or something else altogether? It cannot be entirely an internal relation: when two individuals are exact intrinsic duplicates.propositions or properties  34 35 . There are different singletons for all the individuals and all the sets. whereas the forms are nowhere. section II. yet something is the singleton of one and not the other. Ifwe knew better whether the classes were more fittingly called 'ethereal' or 'platonic'. we still fall very far short. (Blackwell.
see David Lewis 'Counterpart Theory and Quantified Modal Logic'. Taking the 'external' hom. 210. 4 Van Inwagen concludes that my complaints are as damaging to the set theory I accept (and to classical mathematics as well) as they are to the doctrine of abstract simple possibilities I reject. 37 . I posed a dilemma: is the relation of the concrete world 4 For my complaint. Or rather. it must stand in the external relation to certain abstract simples and not others. 5 Suppose the relation of member to singleton is external. the tu quoque is complete. say. for his reply. What price escape? It seems (and van Inwagen agrees) that rejecting the notion of singleton means rejecting presentday settheoretical mathematics. 6 5 6 Van Inwagen. other ways. see Peter van Inwagen. 207 10.escape is worth it. and my present complaints against membership (or rather. 17691. This time.The Trouble with Classes that they realize. passim. Why should they be lethal to the latter and not the former? His point is very well taken. 'Two Concepts of Possible Worlds'. see Lewis. the answer just given holds. that fire is worse. .But to this I have a reply.especiallypp. It's a nasty predicament to claim that you somehow understand a primitive notion.but there's enough left of the tu quoque. or are they not? I suppose they're equally formidable in both cases. 11326. On the Plurality of Worlds. that intrinsic character and external relations can vary independently. Why? We would expect. That's much easier to understand though not very easy to believe.11 (1986). Why must Possum be a member of one singleton rather than another? Why isn't it contingent which singleton is his? . Almost entirely: I have to add that the tu quoque is imperfect. to drive home how nasty it is to have to believe in the membersingleton relation. But in the case of the abstract simple possibilities. pp. escape can be had by accepting the doctrine of plurality of worlds. On my theory of modality the question becomes: why doesn't some otherworldly counterpart of Possum have a singleton which isn't a counterpart of the singleton that Possum actually has? And my answer is: what makes one singleton a counterpart of another exactly is their having counterpart members. yet the relation of member to set seems difficult to understand for exactly the same reasons. although you have no idea how you could possibly understand it.. say I. Peter van Inwagen replied with a tu quoque: I accept set theory. 65 (1968). but not quite lethal in either case.'. Not so for the 'external' hom . On counterparts. one way. even there. pp. 1 ' Van Inwagen's Tu Quoque to the abstract possibilities internal or external? So far as the 'internal' hom goes. as in the case of shape and distance. That's the predicament I'm in when I accept the notion of singleton. On the Plurality of Worlds. That applies equally to the case of the singletons. ···r··. one big part of my complaint was that we had no independent grasp on what sort of relation we were dealing with. 'Two Concepts of Possible Worlds'. Midwest SiudiesinPhilosophy. and both together are worse.pp. •. I say that it's a flexible matter which things we count as counterparts. But are the parallel complaints lethal. the price right . 185213.. not between the frying pans. against the special case of the membersingleton relation) amount to my endorsement of it. it's not even true that a counterpart of Possum must have a counterpart of his singleton. Either is bad. The difference is almost entirely between the fires. p. But another part of my complaint was that we get a mysterious necessary connection: if the concrete world has a certain intrinsic qualitative character. and that's the predicament I claim others are in if they accept the alleged notion of realization of abstract simple possibilities. Can you turn a parallel complaint against the singletons?Try it thus. Hot though it is in the frying pan.}ournal of Philosophy.
the challenge is to find out which is which. and the important appendix as a separate note 'On Relations that Generate'. anyway? Goodman characterizes the idea in a preliminary way with a volley of nearsynon yms. Finally (in 2. There should be no difference without a difference in content. Then Goodman's principle requires that no two things have exactly the same content. 39 . Possum's singleton. What is a generating relation. then we find Goodman's principle violated in a big way. On Relations that Generate those minimal elements that bear R to it (or of the thing itself. because Goodman stipulates that the ancestral of membership is a generating relation in a system founded on set theory. Philosophical Studies. 7 he advances a compelling principle: the 'generating relation' of a system should never generate two different things out of the very same material. Goodman's principle is satisfied: it is a principle of mereology that no two things consist of exactly the same atoms. The principle needs revision if we accept atom less gunk as a genuine possibility. I want to say that I accept Goodman's principle (ignoring gunk). and its converse is a relation whereby one entity 'breaks down into' others. let R be not that but rather its ancestral. There is difference galore without difference of content. and it has to be something built. or 'comprised of'. because the two complaints involve different kinds of necessary connection. But then he proceeds to an example in which it is stipulated that the generating relations of two systems are in one case the proper part relation and in the other case the ancestral of membership . The Problem oj Universals (University of Notre Dame Press. The content of a thing consists of 7 In Philosophy oj Mathematics: Selected Readings. A minimal element is something to which nothing else bears R. Bochenski. Problems and Projects (BobbsMerrill. his singleton's singleton. Otherwise.3 On Relations that Generate I dislike classes because they burden us with mysteries. (Not if the two things consist entirely of atoms. anyway. and in Nelson Goodman. directly or indirectly. pp. Roughly.The Trouble with Classes Now. ed. Countless things are generated out of exactly the same content. and take its ancestral as our generating relation. The paper was originally published in two parts: 'A World of Individuals' in I. whereas the complaint against abstract simple possiblilities involves a necessary connection between external relations and the qualitative character of the relatum. But then I have some urgent explaining to do. A generating relation is one whereby entities are 'generated'. Alonzo Church. 1964). and Nelson Goodman. M. There has to be something to make the difference. R is not acceptable as a generating relation. or 'composed of' other entities.. if we start with some sort of nontransitive relation of direct or immediate generation. Nelson Goodman dislikes classes for a different reason. 9 (1958). The complaint against singletons involves a necessary connection between external relations and the identity of the relatum. 65 6. We can state this more precisely. Paul Benacerraf and Hilary Putnam (Prentice Hall. can the defender of abstract possibilities answer me in a parallel way? . or 'made up of'. The method of counterparts does not apply to the latter problem. If we take the proper part relation as our generating relation. the class of him and his singleton. In 'A World of Individuals'. 1972). 1956). and ever so many more things all are generated out of Possum alone. on the other hand. Let R be the generating relation. and accept anything remotely resembling the normal existence axioms of set theory. but I deny that the ancestral of membership is truly a generating relation.No. if it is already minimal). into one thing but not the other.) If we start with membership.
The Trouble with Classes an appendix originally published apart from the rest of the paper) he gives the official definition: 'the generating relation of a system is the proper part relation or the ancestral of . Given this stipulation. I deny instead that Goodman's principle has any force. or y would be a proper part of x and x would be a fusion of a class and an individual. or the ancestral of membership. (If member x and singleton y were not entirely distinct. which is impossible. Quine. In so far as it is unmereological.of classes via the ancestral of membership is not . 1963). 8 . See W.. Quine on Urelements and Singletons When the thing is a member of its singleton. the ancestral of membership is officially a generating relation. x could not be a member of anything. it is mereological. first the thing is a member of its singleton . Then we could very well think that every kind of composition should obey Goodman's principle.that part is not composition at all. Why should anyone believe that. V.) What does this relation of one thing to another wholly distinct thing have to do with composition. thereby he identifies each urelement with its singleton.4 Quine on Urelements and Singletons Quine has discovered a nice way to remove all mystery about the nature of some singletons. Rather. Then its singleton is part of the class . the thing and its singleton are entirely distinct. then we have an argument more worthy of attention. since y is an atom. Set Theory and Its Logic (Harvard University Press. because the generation of classes is not an unmereological sort of composition. or 'making up'. or what you will) and that mereological composition was only one species of composition among others. But not because I think it illegitimate. there isn't one thing made out of many. it isn't composition. you might as well say that the ancestral of 'being married to someone who is part of' is a generating relation too. even if there were some sort of unmereological composition alongside the mereological sort. He thereby counts urelements as selfmembers. in no matter how broad a sense? You might as well say that marrying someone is an odd sort of composition. in which your spouse is generated from you! For better or worse. But if it's a generating relation in the (vague) sense given by Goodman's preliminary volley of nearsynonyms. Apart from the fact that marriage is better understood. In the first and second cases. in so far as it is composition. or the logical sum of the two. as they occur in the system'. If so. then. It says only that if a relation is either the proper part relation. since he does not call the null set an individual) to mean what we would originally have called membershiporidentity. When something is a member of a class. we should conclude that generation . unless because he renounces classes already? But if we go against Goodman's wishes and ignOf'e his official stipulation. then it must obey a certain condition which the proper part relation does obey and the ancestral of membership (or the sum of the two) does not. And with that I agree. 8 He reinterprets the predicate of membership when applied to urelements (for him. either x and y would be identical. Further. 'individuals'. given the stipulation. or y would be a proper part of x and x would be a class. pp. Suppose we thought that we had a broad notion of composition (or 'generation'. x would be a member of itself. of course I cannot deny that the ancestral of membership is a generating relation.contrary perhaps to our first impressions a legitimate sort of unmereological composition.. 2. there is one thing made out of one thing. membership or the logical sum of the two. in the third case.that part is mereological. 314. the cases are parallel.
If Possum is his own singleton. What surrounds the member is not a soft loop but a hard shell. and so he himself is his singleton's singleton. and so might seem to shed light. I submit that our mereology of classes is nicer. so for Quine they are their own singletons.This picture needs more than a grain of salt! For when something consists of parts. The molecules of which cats are composed also are urelements. that class is the fusion of cats. he defines an urelement (an 'individual') as something identical to its singleton. and the two cannot coexist on pain of collapse.) The Lasso Hypothesis says that a singleton has proper parts. That gives singletons an internal structure after all. I take it to contradict the Division Thesis: the lasso would be neither individual nor class nor mixed fusion of the two. ignore the toughness of the shells. its member. is the fusion of catsingletons. but these are not yet genuine mereological atoms. and which thereby implies that a singleton has no proper parts. in a speculative way. else we give most classes spurious members. We get the singleton subclasses. So it contradicts our Main Thesis. if these too are their own singletons. is the fusion of catmolecules. if you'd rather: a singleton is its member encapsulated. It makes no difference whether we are able to separate the parts. then also his singleton is his singleton's singleton. What if this were the exact truth of the matter? Maybe the singleton of something x is not. Sad to say. Rather they are the encapsulated members of the class. So ignore the breaking. The Lasso Hypothesis them. Quine's idea is not for us. and if abandoning it could buy us clarification of how 43 2. Likewise the class of catmolecules. we do not quite succeed. and so on. we could at best remove only some of our mysteries about the nature of singletons and their relation to their members. in a good cause. which says that a class has no parts but its subclasses. So tough is this shell that when we try to break a class down into its smallest parts. Cats are urelements. which have remained unsplit despite our best efforts. so again their own singletons. all I said in its favour was that we had no good reason to abandon it. likewise he is his singleton's singleton's singleton. But the class of cats and the class of catmolecules are not identical. if any. though it was supposed to be memberless. The fusion of cats and the fusion of catmolecules are identical. or whether it even makes sense to talk of separating the parts. So if we followed Quine. If the class of Magpie and Possum were its own singleton. ignore the failing to break. an atom. and then drawing a picture of a lasso around . . what's left is just the hypothesis that the singleton of x consists of x plus something else. But in fact we cannot even begin to follow Quine. So it must also contradict one of the premises whence the Main Thesis follows. but rather consists of x plus a lasso. each one consisting of member plus shell.5 The Lasso Hypothesis We learnt in school to picture a class by drawing pictures of its members. for us. so if cats are their own singletons. So we have no more mystery about the nature of any of these singletons than we do about Possum himself Quine's plan cannot be extended. (Picture it a different way. The class of cats. after all. Whether you call this something else a 'shell' or a 'lasso' is neither here nor there. that is not because of anything we have done to break it into parts. and its member's proper parts. But the Division Thesis could go. it would have itself as a third member along with Magpie and Possum. both on the nature of the singletons and on their relation to their members. it would have itself for a member. If the null set were its own singleton. namely its lasso. And so on. Nice as it is.The Trouble with Classes indeed.
Or we could call y the 'singleton' and thereby reject the Lasso Hypothesis.) Now the class whose members are x and x + y is the fusi. i. and the singleton of y consists of y plus a lasso. do we know about the nature of lassos? How is a lasso related to the thing it fits? We are no better off if we adopt the Lasso Hypothesis. the other. R. and also there corresponds the fusion y + x. and their union is a set too. that ordinary things such as cats. Case 2: One of x and y. that would be a good reason indeed! But now ask: if the singleton of x consists of x plus a lasso. What we do know. What's in a name? QED Proof Say that lasso L fits x iff the singleton of x is the fusion L + x. and that it satisfies certain other conditions. let it be x. for instance. or else they are both classes. is that this relation satisfies certain structural conditions set forth in the axioms of set theory. which also reduces to L + M + x + y. Case 1: Either x and yare both individuals. can it ever be that the same lasso is used twice over? (Can it even be that one single lasso will do for making all the singletons?) Or must it be a different lasso each time. y's singleton is L + y.6 Ramsifying out the Singleton Function We know nothing. we still say that to each x that has a singleton. and then we could call y + x the fusion (perhaps mixed) of the singleton and its member. two are identical. contra our supposition. whIch reduces to L + M + x + y. then we can confine our attention 9 Paul Fitzgerald. and if I am right that the relation of part to whole is not itself mysterious. Lassos correspond oneone to singletons. This completes the reductio. another thing that has a singleton is the fusion x + y. Might that be all we need to know? Paul Fitzgerald has suggested that it is. Either way. cd. 1976). In fact. Our old questions were: What. one lasso per singleton? The answer is that it must be a different lasso each time. both of which are impossible. 2. Whether we accept the Lasso Hypothesis or whether we reject it. We could call y the 'lasso' and y + x the 'singleton'. if 44 45 . Ramsifying out the Singleton Function anything. so I lament. do we know about the nature of singletons? How is a singleton related to its member? Our new questions are: What. Suppose for reductio that we have two different things x and y. So the classes are identical. he has suggested this for the membership relation generally. is an individual. But if I am right that x is a member of a class y just when x is a member of a singleton that is part of y. (We needn't ask whether M and L are the same lasso. It follows that x and yare identical. Then the class of x and y is (L + x) + (L + y). 'Meaning in Science and Mathematics' in PSA 1974. and spacetime points should turn out to be among the things that have no singletons as parts. and yet one lasso L fits both: x's singleton is L + x. and likewise the class whose members are y and x + y is (L + y) + (M + (x + y». if anything. about the nature of the primitive relation between things and their singletons. which also reduces to L + x + y. and therefore to members of singletons. quarks. there corresponds something y that is wholly distinct from x. It follows that either class y IS Identical to mdIvidual x. its singlefon is M+ (x+ y).s a class. (Reidel. or else class y is a member of itself. that is (L + x) + (M + (x + y». 9 (Or rather.on of the two singletons. and the union of y with the singleton of x is (L + x) + y. S. If they're classes they're sets. Cohen et al. section IV. though.The Trouble with Classes a singleton' is related to its member. whIch reduces to L + x + y. nothing changes at all. y. and so accept the Lasso Hypothesis.
an axiom stating that ordinary things cats. 'Singular Terms. To give the content of the conjoined axioms more explicitly. however.The Trouble with Classes to the special case of member and singleton. etc. any function that conforms to the appropriate conditions shall count as a singleton function. S .are individuals. we take the Ramsey sentence: For some S: . !. or that they are platonic. .. singleton .... for instance. . then Possum's S has an S... or Rudolf Carnap. see van Fraassen.albeit a variable restricted to a certam hmlted range of admissible values. see F. equivalently. Besides 'singleton' and elementary logic.lO 10 On tolerating equivocation.. thus: For some S: . singleton . . in Foundations (Routledge and Kegan Paul.. and for all S: if. since all interpretations of it that satisfy the conjoined axioms are equally 'intended' interpretations. S . Include. Ramsey. S . 47 .. Though it appears in the guise of a primitive constant. (Or better. Axiomatize set theory with 'singleton' as the only settheoretical primitive. 'Theories'.. into one single sentence: . .. singleton . But we may take that to be interpreted once and for all. Possum's singleton has a singleton we should supply the quantifier and the structural conditions that restrict it. since these are obviously unintended. . (We shall see in section 4..) Think of it this way: the word 'singleton' is highly equivocal. . S . The equivocal sentence about Possum is true simpliciter iff it is true on all its intended interpretations (and there are some). TruthValue Gaps and Free Logic' on supervaluations. .. but as a functionvariable bound by an invisible quantifier. . Now conjoin all the axioms. S . . 1978).2 what such an axiom system should look like. That brings our treatment of the conjoined axioms and our treatment of other sentences into line..) We needn't pretend to speak unequivocally of the function that takes members to singletons. On Ramsification. . S . section 26. rather than classes or mixed fusions. It will be neither settheoretical nor mysterious.. . S .r t?an a variable ... S . true on admissible values (and there are some) of its disguised vanable.ette. . P. S . 1966).. S . Rather. . . embodying those of our customary metaphysical opinions about classes that deserve credence.. S .. then Possum's S has an S so that the existence of suitable values of S is affirmed rather than just presupposed. there must of course be other primitive apparatus.. Make the axioms strong enough to rule out Skolemized interpretations with a countable domain. mathematical and unofficial.. thus: For all S: if . that they are ethereal. those unofficial axioms which declare that singletons are outside space and time.) Add unofficial axioms as well. Write down a system of structural. The content of set theory is that there exists some such function. Philosophical Foundations oj Physics (Basic Books. Ramsifying out the Singleton Function And if the word 'singleton' appears in some other sentence. The idea is to regard the word 'singleton' here not as an unequivocally meaningful primitive.. Omit... mathematical axioms chosen to yield standard iterative set theory.... S . .. it is no b.
the things that pass for numbers under various intended interpretations of 'successor' might secretly be ordinary things.the function. But beyond that. more than two to the two to the continuum. though it would be better to respect our conviction that 'successor' has an unequivocal meanmg. were supposed to count as individuals. Any functIOn that satisfies the stipulated conditions will do. set forth in the Peano axioms. even if 'singleton' means nothing definite. Then our predicament is that we have too many conflicting ideas about what the successor function might be. except for those that are parts of the ordinary things that. The structuralist might retreat by dropping the unofficial axiom which says that singletons may not turn out to be among the atomic parts of ordinary things. not to mention what the other worlds may have to offer. Numbers are those things that have successors.) The structuralist's solution is to Ramsify out the successor function. . There's no secret fact about which one is the right one. It is to believe there are enough. The structuralist about arithmetic needn't scratch his head about the unknown nature of the numbersuccessor relation' the structuralist about set theory needn't scratch his head the nature of the membersingleton relation.pace my present reduce lamentatIOns . Maybe in this world. So if there are enough things. even if we could remove all mystery about the nature of the membersingleton relation. It is hard to think where there could be that many ordinary atoms.. The content of the axioms is just that there exists some intended interpretation.The Trouble with Classes A parallel 'structuralist' account of arithmetic is familiar and attractive. (It can be our only arithmetical primitive. and all the demands in this sequence are but the barest beginning. But structuralism for set theory does not fare so well. I'm Structuralism is not a such a bad position ab?ut arIthmetic.to understand that. In the first place. the same things will count as singletons: namely. and no way to choose. So. more than continuum many. Then some of the things that pass for singletons under some intended interpretations might be ordinary and unmysterious. by axiomatic stipulation. If we want enough ordinary things to make room for an intended interpretation of 'successor'. We talk equivocally about all such functions. We have the primitive notion 'successor'. Note that we can say this.makes far stronger demands on the size of Reality. (Or maybe we are willing to to set theory. Probably there are countably many spacetime points (or pointsized momentary parts of particles) in this world alone. more than two to the continuum power many. all the mereological atoms. There must be more than countably many atoms. true for all admissible values of its disguised variable. So sa ys the structuralist. countably many will do. and in particular 49 . Here structuralism about arithmetic is better off. the one we really mean. For on all intended interpretations of 'singleton' alike. Zero is the number which is not a successor. that is. and countably many cats.. and surely in all the worlds together. alas. there are at least countably many quarks. . in which the word 'successor' occurs is true simpliCIter Iff It IS true on all intended interpretations. no better than a (restricted) variable: any interpretation of it that satisfies the Peano axioms shall count as 'intended'. But standard iterative set theory . we haven't a clue about the nature of. and claim . we would " :::r 1 .let alone large cardinal axioms! . He says that the word 'successor' is equivocal. that the successor function is supposed to satisfy. But not nearly enough of them.' Ramsifying out the Singleton Function still have a mystery about the nature of the singletons. points or quarks or cats.) We have structural conditions. unless we stipulate otherwise when we lay down the axioms.
We do grasp it. far worse. then the overwhelming majority of these things must be extraordinary. (3) A very different sort of theory of relations as sui generis individuals treats them as elements of being. rule out the undemanding Skolemized interpretall tions.The Trouble with Classes enough atoms.. He might formulate the axioms in an elementary way that would not. unobserved things. Ramsifying out the Singleton Function function? A set theorist will say that a function is a special case of a relation. At most it may hope to remove the mystery of the membersingleton function. a plausible new 'Main Thesis' will say that the parts of a relation are all and only its subrelations. they are one and all unavailable at this stage of the game. it's a proper class. The preponderant part of Reality must consist of unfamiliar. Settheoretical structuralism must acknowledge the mystery of the missing mass. and we will be left scratching our heads over what it means for a minimal subrelation. constitutive of the qualitative character of things composed of interrelated parts. what can it mean to say . and to say otherwise is to to suit our philosophy. . not at this stage of the game. then. (4) Some talk of relations may just be substitutional 'quantification' over relational predicates.that there exists a suitable function? We understand· what makes a function suitable as an interpretation of 'singleton'. for instance.but it's far. Then if he claimed that those interpretations were. can be expected to end in a predicament differing only in name from our present predicament. and of relations generally? (1) There are various other settheoretical analyses. Some neostructuralist might propose to Ramsify out this primitive notion. Any philosophy of thought and language that says we can't thereby stands refuted. But in the second place. Sometimes it's a set of pairs. But the only relevant predicates we . But whatever advantages these may have. Spatiotemporal relations between the parts of a thing constitute that thing's shape and size. intended. after all. What's good about structuralism is that it would allow us to get by (at the cost of losing uniqueness) with the sort of relations that pair things up arbitrarily and have nothing to do with qualitative character. . to relate one thing to another. 'then his conjuction of axioms would be an infinitely long sentence. to make room for an intended interpretation of 'singleton'. I see no harm in that. And it's no excuse that there is a philosophical problem about how we grasp the difference between intended and Skolemized interpretations of set theory. No retreat. hidden aspects of qualitative character. But we cannot explain the membersingleton relation in terms of any known aspects of the qualitative character of things. he might conclude that ordinary things were abundant enough to meet his needs. because he does not grant that the words 'class of ordered pairs' have any defmite meaning. If relations are classlike. after all. but what's a 11 If he did this by taking all instances of certain schemata as axioms and if he did not abbreviate by means of substitutional 'quantification'. but It s stIll unacceptable. and it seems overbold to respond by positing new. but makes them classlike in all but name. (2) A theory which says that relations are sui generis individuals. instantiated just once. The structuralist might retreat once more. whose existence would have gone unsuspected but for our acceptance of set theory. Should the structuralist adopt some alternative theory of functions. and of whose intrinsic nature we know nothing. But our structuralist cannot agree.not yet presupposing set theory when we say it . It's subtle latton. I protest: those interpretations are not intended. building blocks of nature. It's like the astronomers' problem of the missing mass of the cosmos . and a relation is a class of ordered pairs. but in the case of the singleton function.
Structuralist set theory really is new. In fact this can be done. I cannot survey all the theories that our structuralist might conceivably try. No worries. is the structuralist entitled to speak of ordered pairs? We usually define pairing settheoretically. and if we are structuralists we will regard these as unsuitable substituends because they lack definite meaning. He can insist. And if he has selfpairing. Whatever their merits otherwise. These functions are sets (set theory itself being taken for granted) so there is no problem about quantifying over them. The structuralist might claim to have a primitive understanding of 'ordered pair'. It took the work of Burgess and Hazen to turn settheoretical structuralism into an available option (if I am right to dismiss versions of structuralism that purport to quantify over relations taken as sui generis individuals). nobody even knew how to be a settheoretical structuralist. and thereby to simulate quantification over relations. If we need a settheoretical definition of 'ordered pair'. to my complaints that the membersingleton relation is illunderstood. or on a primitive notion of membership. at this stage of the game. Burgess and A. to be a settheoretical structuralist. structuralist set theory. But if so. Mathematicians have long known that the universe of mathematical that satisfy the structural objects affords countless conditions for a successor function. he is no worse off than those of us who must lay down axioms directly on a r ! Ramsifying out the Singleton Function primitive notion of singleton. so 'pair' defined in settheoretical terms is no better off. we may have a new. settheoretical structuralism can succeed only if we understand settheoretical notions to begin with. I agree (see section 3. you quantify over relations. you do so before you are entitled to the resources of set theory. why bother to be a structuralist? If he has pairing. Hazen.The Trouble with Classes have are 'member of' and 'singleton of'. These definitions are not available to the structuralist. Given reasonable axioms on primitive pairing. There might be some third way. (5) Or we may quantify over relational concepts. but it seems that we may dismiss the obvious candidates. First the general memberclass relation gives way to the special case. but that won't help unless we already have at least one definite concept suited to be the concept of membership in a singleton.2). the structuralist redemption of set theory is not retroactive. and that is what we've only now learned how to do. thanks to new work by John P. they do not meet his present need. He denies that the primitive notion of set theory 'member' or 'singleton'has any definite meaning. and thereby to get rid of the primitive notion of singleton. in various wellknown ways. Only if we don't need it can we have it. Arithmetical structuralism is a different case. The game is not yet over. then we Ramsify out membersingleton. and so we are left with no settheoretical primitive at all. The technical prerequisites 53 . He may solve part of his problem if he says not 'there is a class of pairs' but just 'there are some pairs'. The appendix tells how. It would be highhanded anachronism to claim that set theory was structuralist all along. Before mid1989. neither settheoretical nor primitive. If we wish. that plural quantification is primitively intelligible and need not and cannot be reduced to singular quantification over classes or anything else. Unfortunately. But how. we would expect selfpairs to behave as singletons should. he has selfpairing. with Boolos. that which is not there requires no understanding. To be a structuralist. membersingleton. Farewell. P. In requiring such axioms. then. to introduce pairing. he might as well define the singleton of x as the selfpair of x and x.
in his own way. in his own system. If we want to examine set theory as we find it now.a property is any class of possibilia. nonqualitative property that can belong only to Possum. each one offering to clarify the notion of singleton by subsuming it. Why favour the novel reduction of set theory to pairing (plus mereology) over the standard reduction of pairing to set theory? Next comes one who tells us that Possum's singleton is Possum's haecceity: a special.if he is indeed a very special cat . And if there is . Then maybe Possum's singleton is one of his tropes: his own 2. Next comes one who tells us that for each thing that has a property. not identical tropes. False. the two cats have duplicate tropes. I am not fully convinced that structuralist revolution is the right response. Our working assumption for the rest of this book. and thereby endorse our Main Thesis. by 'abstraction' he does not mean the taking of equivalence classes. the ways they mostly talk and think. To some extent. It does not fit presentday mathematical practice. well enough. 1988). The Reality of Numbers: A Physicalist's Philosophy of Mathematics (Oxford University Press. Not helpful. each one had by Possum alone. Despite all my misgivings over the notion of singleton. 54 55 . he doesn't. and Possum's singleton is one such classwithout which 'nonqualitative property' would be a contradiction in terms. Not so for settheoretical structuralism. True.) It is even somewhat plausible. section 16.it may be otherwise for other things that have singletons. presents a theory of sets in general as plural haecceities. My most urgent worry is that I have no· guarantee that there is any qualitative property had by Possum alone. is already structuralist. of course. I want to carryon examining set theory as we find it. I say. The first of them. Therefore Ilea ve structuralism as unfinished business. I do not know whether Bigelow could consistently say that a plural haecceity is a fusion of singular haecceities. Or at least it is somewhat plausible to say that arithmetic is not determinately not structuralist: even if mathematicians seldom or never avow arithmetical structuralism at least it fits. So it is not anachronistic to say that arithmetic. still that special case is primitive. that belongs to it alone. By 'property' he does not mean a class of actual and possible instances.7 Metaphysics to the Rescue? We can expect to hear from several systematic metaphysicians. is the orthodox set theorist who tells us that a singleton is a special case of a class. 12 True. and even when the general case of membership reduces via mereology to the special case of membership in singletons. settheoretical structuralism changes the subject. because it's part of a settheoretical conception of properties . until the appendix. a trope. shall be that the membersingleton relation is indeed primitive. there is a particular case of the property. Next comes one who tells us that Possum's singleton is an abstraction from several coextensive qualitative properties. (Laymen are another question. Thence we have the special case that a singleton is a singular haecceity. r Metaphysics to the Rescue? Next comes the one who tells us that pairing is primitive and that Possum's singleton is just the pair of Possum and Possum. because that practice does not include any knowledge of how to Ramsify out the singleton function. 12 John Bigelow. we have to concede that it claims a primitive understanding of membership. If Possum does have an exact duplicate. But this does more to explain 'haecceity' then 'singleton'. as understood among mathematicians. Anyhow.The Trouble with Classes for Ramsifying out the successor function have long been common knowledge. I guess.
they are composed of these constituents in an altogether unmereological way. Let's see it done. we are set free. if indeed his theory of singletons yields answers. Although it takes a class to be a fusion of some facts (or 'states of affairs') pertaining to its several members. then unit classes would turn out to be facts of unithood. are profoundly mysterious. The class of Magpie and Possum. or haecceities. Armstrong. would on Williams' plan be the fusion of three tropes: Magpie's selfidentity (her singleton). if anywhere. maybe. or pairs. his subsumption shows that the mystery of singletons also bedevils the classes. are they? What is their intrinsic nature? Do they differ qualitatively from another? Is the relation of member to singleton founded on the qualitative nature of the relata. Possum's selfidentity (his singleton). 100 (1991). a unit. Rather. Nor do I imagine that the reservations I've expressed are at all conclusive against the several proposals. Last comes one who tells us that the world is the totality of facts. Credo about nonatomic things. If he has shown that. Perhaps it should be the (mereologically atomic) fact that he is selfidentical. Therefore I find unmereological 'composition' profoundly mysterious. mereology looks to be the general theory of composition. that although facts do in some sense have 'constituents'. Mysteries are an onerous burden. and as part of a broader plan that conformed to our First Thesis but not to our Second Thesis. forthcoming in Mind. However. M. Philosophical Papers. and Armstrong. even granted an ontology of facts. and are . 2. The latter paper gives exactly the proposal now under consideration. as follows. and therefore all classes. for instance. Should we therefore dump the burden by dumping the classes? If classes do not exist. We were stumped by several questions about singletons. 14 See Peter Forrest and D. particular selfidentities seem de trop. but Gx might be the fact whereby x belongs to the class of Gs. Or perhaps it should just be the fact that he is one single thing. 14 Again. Where. I scarcely want to welcome it back via the anatomy of facts. 'Classes are States of Affairs'. or perhaps his own particular cathood. Fx might be the fact whereby x belongs to the class of Fs. or is it more like a distance relation. the former differs in one relatively minor way. If it does. not the theory of one special kind of composition. see Williams. 'On the Elements of Being'. or something else altogether? Let each of the metaphysicians tell us his answer to these questions. but very tentatively. I suspect it ought to be too sparse to afford atomic facts 13 On tropes in general. 'The Nature of Numbers'. Abstract Particulars. and that Possum's singleton is to be identified with some fact about him. 16586. I am no enemy of systematic metaphysics. or tropes. After expelling it from set theory. or facts. 13 But even if we should believe in some tropes. Further. Williams proposed the identification of singletons with selfidentity tropes in lectures at Harvard circa 1963. or that he is a cat. Then neither fact is once and for all the singleton that is a common part of both classes. I want to pose a challenge to them all. But ifit doesn't. pp. and Campbell. we needn't puzzle over their mysterious nature. and also the particular nonidentity between Magpie and Possum.8 Credo Singletons. 16 (1987). If x is an F and also a G. and so do mereologically atomic tropes had by thoroughly nonatomic cats. then his subsumption of singletons has not dispelled their mystery.mereologically atomic. or abstractions from qualitative properties. 57 . well and good.The Trouble with Classes particular selfidentity. he has not helped us and he has done himself no favour. or is it a mixture.at least sometimes . If we renounce classes. not of things. it allows that the same thing might belong to different classes in virtue of different facts pertaining to it.
1980). But it does demand wholesale reinterpretation. If there are no classes. Still. Even if we hold onto some mutilated fragments of mathematics that can be reconstructed without classes. and all such challenges deserve suspicion. and fusions of singletons sometimes have singletons. it says that classes exist. and on. I know. Credo That's not an argument. We shouldn't expect mathematics to go away to make our life easier. that a Being than which no greater can be conceived cannot be conceived not to exist. I do not dismiss it as absurd. 'good without being true'15 we still reject it. will you boast of philosophy's other great discoveries: that motion is impossible. I'm moved to laughter at the thought of how presumptuous it would be to reject mathematics for philosophical reasons. and abjure countless errors. 59 . 16 Does settheoretical structuralism reject mathematics? It may seem not: it demands no change in the practice of mathematics. that it is unthinkable that anything exists outside the mind. Science Without Numbers (Princeton University Press. it counts the theorems as true. 16 15 As in Hartry Field. there are no homeomorphisms.The Trouble with Classes No. and it grants that speaking of singletons and their members makes sense. for set theory pervades modern mathematics. I know not how. If we philosophers are sorely puzzled by the classes that constitute mathematical reality. it does challenge the established understanding of mathematics for philosophical reasons. with a straight face. ( myself would not call the structuralist reinterpretation a rejection of mathematics. How would you like the job of telling the mathematicians that they must change their ways. there are no complemented lattices. To reject mathematics for philosophical reasons would be absurd. full of false 'theorems'. then our mathematics textbooks are works of fiction. . And somehow we know that ordinary things have singletons. that time is unreal. For all these things are standardly defined as one or another sort of class. Renouncing classes means rejecting mathematics. If there are no classes. Even if we reject mathematics gently explaining how it can be a most useful fiction... Whether acceptanceonlyunderreinterpretation is tantamount to rejection is a vague matter it depends on how drastic the reinterpretation is. ad nauseam? Not me! And so I have to say. and singletons have singletons. gritting my teeth.. that no theory has ever been made at all probable by evidence (but on the other hand that an empirically ideal theory cannot possibly be false). Some special branches and some special styles of mathematics can perhaps do without. that's our problem. and that's still absurd. but most of mathematics is into set theory up to its ears. Mathematics is an established. to follow philosophical argument wherever it may lead? If they challenge your credentials. Philosophy is as shaky as can be. if we reject the bulk of mathematics that's still absurd. that it is a wideopen scientific question whether anyone has ever believed anything. We know even that singletons comprise the predominant part of Reality. That will not do. Rather. there are no probability distributions. that somehow. than there are no Dedekind cuts. now that philosophy has discovered that there are no classes? Can you tell them. and so on. going concern. we do understand what it means to speak of singletons.
We can leave it to mereology to make manymembered classes by fusing together singletons. But let us henceforth begin with singleton rather than membership as the primitive notion of set theory. 1 Bunt.) Before we advance to set theory itself. also formulates set theory with the partwhole relation and the membersingleton ('unicle') relation as primitive. there is no attempt to set up mereology first and then add on the theory of singletons.l (Let us separate Mr Hyde from Dr Jekyll. And in formulating set theory. But both primitives figure together in his axioms almost from the beginning. We want to end with nothing out of the ordinary.Theoretic Semantics. Let us accept the orthodox iterative conception of set. and have faith in the teachings of settheoretical mathematics. including the part of it that escapes elementary formulation. Mass Terms and Model.3 A Framework for Set Theory 3. let us disentangle the part that characterizes the notion of singleton from the part that is just mereology. .1 Desideratafor a Framework Let us subdue our scepticism.
and All the King's Men'. pp. Symhese. and we may have predicates or functors (defined or primitive) that take plural arguments. 24 (1971). 'Tom. 89 107. 13 (1976). then Napoleon is one of them. vol. identity. we may have plural terms. then there are some things such that. I would have liked to call it 'logic'. If there is and 'Nominalist Platonism'. Massey. D. it will not admit of complete axiomatization. then y is one of them and y is not identical to x. American Philosophical Quarterly. Peter M. for all x. 1903). 30918. Stephen Pollard. from Boolos.A Framework for Set Theory we must have in place the framework to which we shall add it. 'Complex Individuals and Multigrade Relations'.' Or in longwinded regimentation: 'There are some things such that each of them is a critic. as logic is. but we know it well as masters of ordinary English. 9 (1975). 1982). pp. and Harry. 'To Be is To Be the Value of A Variable (or To Be Some Values of Some Variables). as logic is.2 Plural Quantification Besides the elementary logical apparatus of truth functions. (I commend it to those who dare to cut their mathematics to suit their philosophy. Max Black. but rather the settheoretical addition. 'The Elusiveness of Sets'. I take it that I agree fully with him on substantive questions about plural quantification. Eric Stenius. Let us remind ourselves of what we know how to say.. Philosophical Review. 161 88. pp. It will be ontologically innocent. though.) Likewise. Universals and Scientific Realism. to illustrate the power of plural quantification: Napoleon is Peter's ancestor iff. It will not be quite like logic.) In return. x is one of them iff x is a cat. 27 (1974). Other relevant writings on plurals include Bertrand Russell. Simons. Review of Metaphysics. Gerald J. possession is nine points of the law. 3. and each parent of one of them is one of them. and. 89 (1980). 607 24. Barry Smith (Philosophia Verlag. and such that for all x and y. 'Sets'. our framework also shall be equipped with apparatus of plural quantification. Armstrong. in fact. Philosophy Research Archives. as logic is. pp. whenever there are some people such that each parent of Peter is one of them. sections 70 and 74. 'Plural Quantification and the Iterative Conception of Set'. Adam Morton. as is shown in the appendix. 11 (1986). pp. When we go beyond the vocabulary of the framework itself. pp. Dick. We shall have plural pronouns (that is. as logic is. But that is not its name. if x is one of them and x admires y. if there is at least one set. pp. 61436. in Parts and Moments: Studies in Logic and Formal Ontology. though (as noted later) I would make less than he does of the connection with secondorder logic. 62 . not stopping to make any official choice among its incomplete fragments. M. to show the evident triviality of a principle of plural 'comprehension': If there is at least one cat.' Or in less abbreviated form: 'There are some critics such that each of them admires only other ones of them. Principles of Mathematics (Cambridge University Press. Richard Sharvy. Nous. Our interest in axioms will concern not the framework. with names. 57987.2 2 In this section I closely follow the lead of George Boolos. Plural Quantification A famous example: 'Some critics admire only one another. It will be devoid of set theory. So we shall help ourselves to evident principles of the framework as needed. This apparatus is not common in formal languages. 'Plural Reference and Set Theory'. then there are some things that are all and only the sets. then there are some things that are all and only the cats. This framework will be topicneutral. I. 'A More General Theory of Definite Descriptions'. ed.. (Regimented: . We shall have a copula to link singulars with plurals. variables) and plural quantifiers to bind them. and ordinary singular quantification. It will have a lot of mathematical power. It will be fully and precisely understood.' An example. 324.
and no property that distinguishes them from other. what you say cannot be true: the supposed class is a member of itself iff it isn't. and they are many. The fans of the Chieftains are many. Likewise. X is one of them iff x is a fan of the Chieftains. so there can be no such class. many of them. and. There are some things such that for no set are they all and only its members. Indeed. It is customary to take for granted that plural quantification must really be singular. and they sing together. never mind that you know better. and. 'the fans of the Chieftains'. . or with singular quantification over sets or classes or properties: We can say without apparent contradiction . And if you say that there are the nonselfmembered classes. then he has written some book that is not one of them.) To translate your seemingly true plural quantification into a contradictory singular quantification is to impute error grave and hidden error. for all x. then he has written some book that is not one of them.. We may well look askance at this .by imaginative interpretation of others words with some 64 6. iff there are some things. There are just people. that although there are some people such that Peter's parents are among them and Napoleon isn't. whenever there are some books and they are countable in number. that might reduce with the aid of apparatus not yet introduced . never mind that he professes nominalism. he who says that there are the cats can only mean. He has written more than 450 books iff. x is one of them iff either x is one of the Clancy Brothers or x is Tommy Makem. that there is the class of nonselfmembered classes. 'Andy Irvine and Paul Brady'. and if there are classlike whatnots that are not exactly classes or sets. for instance. that there is the class of cats. The Clancy Brothers and Tommy Makem sing together iff there are some things.· there cannot be the whatnot of all nonselfmembered whatnots. but no sets or classes or properties. (Likewise there cannot be the set of all nonselfmembered sets.) The plural 'they are books' and 'Isaac has written them' reduce instantly to the singular: 'each one of them is a book'. Examples of plural terms: 'The Dubliners'. . for instance or might remain primitive. Plurals. yet there is no set or class of these people. whenever there are some books and they are at most 450 in number. He might dodge it . and some things such that for no property are they all only its instances. then there are some things that are all and only the nonselfmembered classes. The plural 'they are at most 450' reduces less instantaneously: 'for some Xl"'" for some X 450 ' each of them either is Xl or . that there exists at least one book. Accordmg to this dogma. or is X450" As for 'they are countable'. a nominalist well might say. falsely by my lights but without apparent contradiction. and some things such that for no proper class are they all and only its members. are the whereby ordinary language talks about classes. and such that the parents of anyone of them are among them. you can only mean.and indeed trulythat there are some things such that for no predicate are they Plural Quantification all and only the satisfiers of that predicate. (I presuppose.Even the some singularist may. he has written uncountably many books iff. The first is a plural proper name..A Framework for Set Theory at least one nonselfmembered class. the rest are eliminable plural definite descriptions. says our nominalist. 'Isaac has written each one of them'. Examples to deter us from confusing plural quantification with substitutional 'quantification' over predicates. for all x.set theory. safely I think. and other mdlvlduals. 'The Clancy Brothers and Tommy Makem'. so it is said. If that's what you mean. Exam pIes to illustrate the use of plural predicates: Isaac has written some books iff there are some things such that he has written them and they are books.
and beyond the ordinary proper classes. or that there are all the sets that are nonselfmembers. or that there are some things that are all the sets and proper classes that are nonselfmembers. but you still repudiate proper classes. then your true plural quantification over proper classes cannot be the true singular quantification over awesome 'classes' that the singularist says it is because ex 11ypothesi that singular 66 . You may truly say that there are some things that are all the sets. sets are available to be plurally quantified over. 7587. Maybe we're right in our various repudiations.because ex hypothesi that singular quantification is false. If you're wrong and there are sets after all. But there is no set of these sets. The sinsays: 'Your plural quantification corresponds to my smgular quantification over sets of individuals. a believer in awesome 'classes'. Plural Quantification So you splutter. If so. But this something may not be what you ordinarily call a 'class'. Suppose there are proper classes. If you're wrong and there are proper classes after all. Resnik. If you're right and there are no awesome 'classes'. Suppose there are sets. He says that you have just affirmed the existence of one by plural quantification understood according to the larist dogma. the singularist sa ys: By dogma. proceed to the next case. then your true existential plural quantification over sets cannot be the true singular quantification over proper classes that the singularist says it is because ex hypothesi that singular quantification is false. If you're right and there are no sets. How do you differ from me?' The answer is the same every time. How do you from me?' When Boolos embraces plural quantificatIon and repudiates proper classes. But there are still individuals. Suppose you repudiate sets.which contain even the proper classes as members. and I say that really you must think so too. The singularist says: 'Your plural tlficatlon corresponds to my singular quantification over sets. you must mean that there is something that contams all and only the nonselfmembered classes. I think. Journal of PhIlosophy.if I may call them that . then your true existential plural quantification over individuals cannot be the true existential plural quantification over sets that the singularist says it is . maybe we're wrong.and you do too. If so. or that there are some sets with no bound on their ranks. there are some truly awesome 'classes' . [ believe it . and never have been. You may truly say that there are some proper classes. 3 Michael D.A Framework for Set Theory imaginative ontology. that beyond the sets. and awesome "classes". he plays into the singularist's hands. When the nominalist embraces plural quantification and repudiates sets. and they are available to be plurally quantified over. 'Secondorder Logic Still Wild'. You must have meant that there is an awesome 'class' that contains all and only the nonselfmembered sets and proper classes. 85 (1988). proper classes are available to be plurally quantified over. but there is no set or proper class that has any of them as members. pp. he plays into the singularist's hands. That's no contradiction. proceed to the next case. If you're right and there are no proper classes. How do you differ from me?' And when you embrace plural quantification and repudiate awesome 'classes' same again. proper classes. 3 The singularist says: 'Your plural quantification corresponds to my singular quantification over sets and proper classes. You may truly say that there are some things that are the cats. and eventually manage to say that you are not now. but you still repudiate awesome' classes'. When you say that there are the nonselfmembered classes.
so we have no right to mention them at all. some book by Isaac is 'not Xl and not X 2 and . The reduction to quantifierblocks is instructive. if any. proceed to the next case.. Plural quantification. Reject it. It is not ordinary singular quantification over special plural things . therefore. so long that their limited length never imposes any numerical restriction? I fear not. But not weighty enough. . you can truly say that there are some things that are all and only the sets and proper classes and awesome 'classes' that are nonselfmembers. It may help to break the grip of singularism if we see how some plural quantification can be translated without dragging in any sets. like singular. And so it goes. Suppose there are awesome 'classes'. but this time we need a language that allows infinite blocks of quantifiers and infmite conjunctions. Maybe the singularist says that way out beyond even the awesome 'classes' . 'Whenever there are at most 450 books. It imputes ontological commitment where there is no ontological commitment. Rather.. Those are two weighty reasons to believe it. Let us eliminate the plurals and still quantify just over books. and from ourselves as he interprets us? We differ about what's true under anyone fixed hypothesis about what there is and what there isn't. some book by Isaac is not Xl and not X 2 and . and not X450" One plural quantifier over books gives way to 450 singular quantifiers over books. holding none in reserve. even as sequences.unless instead it imputes restriction where there is no restriction. What's worse. Lo. And then they just don't exist. carries ontological commitment only to whatever may be quantified over...A Framework for Set Theory quantification is false. 'Whenever there are countably many books.. If you're wrong and there are awesome 'classes' after all.. . classes. • •• . Can we say that plural quantifiers in general are abbreviations for long blocks of singular quantifiers. and evermoreextravagant ontological commitment. or properties. that we may now be wrongly repudIatmg. 68 . In the case of the prolific author. we quantified plurally over books. to be had. How do we differ from the singularist. some book by Isaac is not among them' becomes 'For any book Xl' for any book X 2 . But let's cut a long story short. .. Whatever classlike things there may be altogether. he violates his own stricture in the very act of proclaiming it! We embrace plural quantification over all the things there even those. it seems we can truly say that there are those of them that are nonselfmembers.not even when there are special Plural Quantification plural things. we'll have to concede that the abbreviated expressions exist only as settheoretical sequences. and then I do not think we may mention them in explaining apparatus that purports to be devoid of set theory. and contradiction where there is no contradiction . some book by Isaac is not among them' becomes 'For any book Xl' for any book X 2 . we'll eventually have to concede that the abbreviated expressions are too long to exist at all. then they are available to be plurally quantified over. It imputes falsehood where there is no falsehood. It imputes deviousness where there is no deviousness. ' where this time the ellipses abbreviate infinite sequences. . Before long. for any book X 450 .he singularist dogma has the weight of common opinion m Its favour.. The next case is the same.!. it is a special way to quantify over whatever things there may be to quantify over. Maybe the singularist replies that some mystical censor stops us from quantifying over absolutely everything without restriction.. It is devoid of set theory and it is ontologically innocent. Quantification would be a simpler affair if it were true. namely classes. Plural quantification is irreducibly plural.
Even the methods of the appendix. First Choice Schema: If there are some things. Further. it gives us no way to quantify plurally over mereological fusions. its parts in turn would be composite ad infinitum. though in no way has it misled Boolos himself. (2) It hints that the third. Certainly there's something that's hard to understand! The hasty reader may think it's the plural quantification that's to blame. in fact. He joins this with the idea that we could analyse statements ostensibly about composite entities in terms of plural quantification over atoms alone.and should. it encourages singularism. 3. unless especially restricted against vacuity. some things.That will 4 Morton.. section II): if all composite entities are to be eliminated in favour of their parts. it leaves us vulnerable to an argument akin to Malezieu's (see David Hume. the same . and if there are not too many individuals. Choice be only a skimpy third order. full secondorder quantification. A Treatise of Human Nature (1739). we can state . still this would be far from believable: T are many atoms. 'Complex Individuals and Multigrade Relations'.. We could claim that 'the cat is on the mat' is true in virtue of the arrangement of many atoms. The rest must remain primitive. Because of this mismatch. That will do to tell the experts what a formal language with plural quantification looks like. to bind any free variables. and no two of them .versions of the Axiom of Choice. we have some of the power that's normally credited to set theory. and no start at all on the fourth. We have already seen how to take ancestrals. of course. But even if we could save the truth of everyday statements. which do without set theory (and primitive pairing). and each of them . and higher orders cannot be far behind but what mjght plurally plural quantification be? (Infinite blocks of plural quantifiers? ..) (3) It also hints that polyadic sec. regimented translations from unrestricted secondorder logic into English come out abominably convoluted. and fatally for purposes of my present project. let alone the settheoretical notion of singleton. treats plural safe enough. corresponds not to 'there are one or more things' but rather to 'there are zero or more things'.A Framework for Set Theory but yields only a skimpy fragment of plural quantification. still require mereology and an infinite supply of atoms. with a restriction against vacuous variables. then we are forced to deny a priori that there can be such a thing as atomless gunk. part II. (1) In so far as we already regarded secondorder logic as set theory (or property theory) in sheep's clothing. Both are schemata. and then prefixing universal quantifiers. But. Also. So I prefer to play it down. etc. Then if we liked Morton does not advocate it! we could go on to deny that composition ever takes place. using any vocabulary that may later come to hand (adjusting the number of the verb). affirm .. But I fear the identification may be misleading. not one composite thing. because if there were. if he has set quantifiers as blocks of singular quantifiers theory at his disposal and yet applies this treatment only to plural quantifiers over individuals. Further.3 Choice Even before we reach mereology. and we need some method of pairing. triples. We need plural quantification over ordered pairs. singular or plural. that takes additional resources.ondorder quantification cannot be far behind. Here are two. though strictly speaking there are no such composite entities as cats and mats. 4 Boolos identifies plural quantification with secondorder quantification: monadic.. fourth. and what it can do. (4) Secondorder quantification. To affirm them is to affirm whatever sentences can be made by filling in the blank. book I.
Journal of Symbolic Logic. 3. 1956). 31 (1928). 'distinct'. pp. as plural. H. Alfred Tarski. 4555. 33 (1930). Definitions: x and y overlap iff they have some common part. 14270. pp. 16591. pp. History and Philosophy of Logic. 'The Calculus of Individuals and Its Uses'.) For other expositions of mereology in English.4 Mereology To finish our framework. then it is a fragment of the 'ensemble theory' of Bunt. and includes. 1937). Rolf A.) See P. 2 (1983). In Ontology there is a kind of quantification that can be interpreted. x. then there are some things such that each of the former things . Topoi. . pp.A Framework for Set Theory things. y or y . . (Those who hijack the word 'distinct' to mean merely 'nonidentical' may say 'disjoint' to mean 'nonoverlapping'.) Second Choice Schema: If nothing . and '0 Podstawach Matematyki'. Bunt himself. 3 (1983). .4. The Axiomatic Method in Biology (Cambridge University Press. and 34 (1931). 5 We have a choice of primitives. Parts: A Study in Ontology (Oxford University Press. If mereology is the general theory of part and whole. 'Foundations of the Geometry of Solids' in Logic. The 1927 1931 papers appear in abridged English translation as 'On the Foundations of Mathematics'. . or 'fusion' instead. they are (entirely) distinct. Iff not. because the Schema carries no commitment to any such entities as orderings. 352. 2 (1916). then either x . Formal Logic (Oxford University Press.. and if whenever x . Przeglad Filozojiczny. z then x . and.Theoretic Semantics. Semantics. section [[[. exactly one of the latter things. there is a linear subordering of it still with no last term. (Here the plural subsumes not only the singular but also the empty: 'some things' must mean 'some zero or one or more things'. there is a copula that can be read 'is one of'. Mass Terms and Model.. another one. 5 (1940). 77105. Under this pluralist reading. 26191.. however. But only roughly. chapters 1 and 2. His term 'collective class' corresponds to our term 'fusion'. pp. see A... not to the term 'class' in any sense now current. 32 (1929).. itself.. Or we could begin with 'overlaps'. and 73 . chapter II. 5 Mereology was developed by Stanislaw LeSniewski in several papers in Polish in 1916 and in 19271931: 'Podstawy Og61nej Teoryi Mnogoki I'. Peter Simons. We could begin with 'part' and go on as follows. his Ontology. Simons. 1970). Woodger. 60101.. pp. Nominalistic Systems (Reidel. (Beware: LeSniewski thinks of mereology as affording an interpretation of the language of set theory. z.. then also there are some of those things such that (1) among the latter things also. pp. but it says roughly this: if there is a partial ordering with no last term. The Structure of Appearance. 'On Understanding LeSniewski'. and if there are some things such that each of them '" another one of them. we add the apparatus of mereology. reserves the term 'mereology' for a restricted theory of part and whole. y and y . LeSniewski's original version of mereology includes the whole of my 'framework'. M. applying only to individuals (see pp. pp. 164206. Nelson Goodman and Henry Leonard. 30 (1927). The Second Choice Schema is a mouthful. Prace Polskiego Kola Naukowego w Moskwie: Sekcya Matematycznoprzyrodnicza. Prior. Tarski's Appendix E in].. (2) whenever x and y are two of the latter things. Eberle. Goodman. N. LeSniewski's mereology is founded upon. Metamathematics (Oxford University Press. 1987). . 1955)..288301). each one . 712. for a very thorough survey. plausibly but without benefit of direct textual support. Definition: Something is afusion of some things iff it has all of them as parts and has no part that is distinct from each of them.
or iff y is a fusion of x and something z. This is a minority opinion. Still. was that x is part of y iff Y is wherever x is. to apply mereology to things without thinking that they are in space and time. Suppose it turned out that the three quarks of a proton are exactly superimposed. how many are there? Using the resources of elementary logic. but rather as a spatiotemporal. (And suppose the three quarks each last just as long as the proton.namely. They conclude that any application of it to things not known to be in space and time is illicit. The original idea. Using the resources of plural quantification. or that there are exactly seventeen. supposedly. as we shall soon see. Uniqueness of Composition: It never happens that the same things have two different fusions. or iff everything distinct from y also is distinct from x. And if there are atoms. In the second place. we can express such hypotheses as that there are more than seven. are three. besides whatever it may take to close the circle of interdefinition.5 Str!fe over Mereology I myself take mereology to be perfectly understood. Trigonometry is said to be part of mathematics. But the basic axioms of mereology are silent about which of these hypotheses are true. It will remain undecided whether there is gunk. Suppose a material thing occupies a region of substantival spacetime. each occupies the same place as the other. Strife over Mereology 3. Suppose two angels dance forever on the head of one pin. and certain. it is common enough.) Still the quarks are parts of the ptoton. unproblematic. they are two distinct proper parts of the total angelic content of their shared region. God's foreknowledge is said to be part of His omniscience. In view of the second and third axioms. Unrestricted Composition: Whenever there are some things. 74 75 . the proposed definition presupposes a prior notion of part and Transitivity: If x is part of some part of y. Many philosophers view mereology with the gravest suspicion. the singletons but at this point we don't say. but the proton is not part of the quarks and the quarks are not part of each other. The basic axioms of mereology. These axioms do not settle all questions that can be raised in the language of mereology. the proposed definition can go wrong even for things that are in space and time.A Framework for Set Theory Definitions: x is part of y iff everything that overlaps x also overlaps y. Sometimes they suspect that originally the notion of part and whole was understood not as topicneutral. In the third place. then we shall conclude that there are atoms . then there exists a fusion of those things. still less that the region and the thing are parts of each other and therefore identical. each one just where the others are and just where the proton is. That is wrong thrice over. we can express hypotheses that discriminate between infinite sizes.or merely spatial notion. and not especially philosophical. and rely on it to be a functor defined for any plural argument whatever. it does not follow (though it just might be true) that the region is part of the thing. In the first place. Does Reality consist entirely of atomless gunk? Entirely of atoms? Or some of each? When we add set theory. we are entitled to say 'the fusion of'. then x is part ofy. At every moment. or that there are fewer than eighty.
chapter 4. ordinary particulars are bundles of particular tropes. Universals and Scientific Realism. For related discussions see Lewis. Sometimes temporal or modal problems about mereology Strife over Mereology arouse suspicion. mcreology per se does not demand that each part of the occupied region must be occupied by some part proper or improper of the occupying thing. A more popular rival theory says that possible worlds are abstract representations. but have repugnant consequences. 6572. might be among. 68 (1971). so not at all the sort of thing I think I am. 'On the Elements of Being'. But this can't mean that he himself is part of that world as well as this one. the proper parts of ordinary particulars. Not just any controversial thesis that uses mereological notions and wins the adherence of some mereologists is a thesis of mereology per se. 77 . or the main system of Goodman. alternatively. pp.Two theories explain it without overlap. a counterpart not identical to Humphrey himself. see Williams. or the un actualized possibility. 6 (3) Mereology is silent about whether something wholly present in one region may also be wholly present in another. When a possible world misrepresents Humphrey as winning. they are instances of broader problems about temporary and accidental instrinsic properties. so that an electron would be composed of its chargetrope. Journal oj Philosophy. On tropes. or actually different but possibly identical. no 7 Lewis. because 'wherever x is' had better mean 'wherever some part of x is'. if something occupies a region. Analysis. If not. Then from the standpoint of the later time. 'Counterparts of Persons and Their Bodies'. 6 On universals in reblH. The modal problems are best solved by denying the overlap of possible worlds. some of them misrepresent our Humphrey as a winner. It seems that something and its proper part can be first different and later identical. its masstrope. see Armstrong. about whether the quarks that make up the proton might be exactly superimposed. pp. and 'Rearrangement of Particles: Reply to Lowe'. The temporal problems are best solved by accepting the doctrine of temporal parts. and thereby confer upon him the property of winning at (better. thereby ascribe to him the property of winning.A Framework for Set Theory whole. 7 Briefly. What can it mean? . or could have been. philosphers' suspicions against mereology amount to guilt by association. Those problems are not the wages of mereology. if such there be. 48 (1988). winning according to) that other world.) We can still say that our Hubert Humphreyhe himself! wins the presidency at another world. Other solutions (equally consistent with mereology) do exist. for instance. not identical to itself! I have discussed these problems elsewhere. My own theory says that Humphrey wins at another world by having a winning counterpart there. wholly present wherever they are instantiated. that I am a changeless soul. it seems we have one thing that used to be. that's a second way for a singleton atom to be where its extended member is. 20311. (Apart from shared universals. (4) Finally. or the undivided omnipresence of God. The Structure ojAppearance. They arise for everyone. and a few more. or a singleton atom that is where its extended member is by being at every point of an extended region. or enduring things wholly present at different times. or about whether universals in rebus. Sometimes. or about whether. Abstract Particulars. On the Plurality oj Worlds. and Campbell. (1) Mereology is silent about whether all things are spatiotemporal. For better or worse it does not forbid recurrent universals. (2) It is silent about whether spatiotemporal things may have parts that occupy no less of a region than the whole does for instance.
(7) Suppose instead that the wordtypes are taken as settheoretic constructions out of the six lettertypes sequences.) Then the two wordtypes are each made of parts of the same six lettertypes. partly it is the making of singletons. and likewise for lettertypes. Then indeed two things are generated settheoretically out of the same six members. made of six letterinscriptions. Against this it might be said. Instead they persist by enduring identically. of them. but not the same parts. we must first accept a theory of structural universals which I take to be far from established. not out of many. (4) The letters might be movable pieces of plastic. might be both a 'stream'inscription and also a 'master'inpeculiarorderinscription. are involved alike in both wordtypes. so I claim. as I think wrongly. Australasian Journal of Philosophy. perhaps. Then the unique thing composed of the six letters is a 'master'inscription relative to one time and a 'stream' incription relative to another. 2546. What's true is that he exists according to that possible world. wholly present at different times. on this supposition. Sometimes it is the axiom of Uniqueness of Composition that arouses suspicion. we'd better take only those 78 Strife over Mereology inscriptions which contrast with their surroundings. It doesn't matter whether they are all and only the satisfiers of some description. but none of them is a violation of the axiom. (Here's where plural quantification S But what of another way of taking the example? (8) A wordtype is a structural universal instantiated by the inscriptions. or even any class. But they are not cases of composition. See David Lewis. namely the six lettertypes together with rclations of juxtaposition. and each one is made of six letterinscriptions. 8 Most of all. I say that whenever there are some things. (3) One single inscription. but then the 'm' inscriptions of the two words would not be the same. formed first into a 'master' inscription and then rearranged into a 'stream'inscription. (5) Suppose. which makes one thing out of one thing. that the six plastic letters do not have temporal parts. as well as the lettertypes. they have a fusion. for instance. I conclude that words and their letters. There's no kind of inscription that it is simpliciter. It is just this power to make the most of limited material that makes settheoretic constructions worthwhile. No. though they are alike in shape. Whenever! It doesn't matter how many or disparate or scattered or unrelated they are. do we have two different inscriptions of the two words. Then the two wordinscriptions are made of different temporal parts of the same six persisting letterinscriptions. it is made of simpler structural universals. Rather than taking the case as an objection to the principle of Uniqueness of Composition. 64 (1986). (To avoid confusion. But to accept this as a counterexample. because what is going on is not just composition. (1) There is a 'master' inscription and there is a 'stream' inscription. (2) The two words might be written crosswordfashion so that they share the same 'a'inscription. 'Against Structural Universals'. This really would go against the principle no misunderstanding this time. are no threat to Uniqueness of Composition. Then the 'a' in 'master' and the 'a' in 'stream' are not the same. that the two words 'master' and 'stream' are made of the same six letters. Partly it is composition. and parallel examples. or anywhere else. It says there is no difference without a differencemaker: x and yare identical unless there is something to make the difference between them by being part of one but not the other. nor. various things may happen.r A Framework for Set Theory Humphreyasmisrepresented exists as part of that possible world. 79 . pp. I go the other way: I object to structural universals exactly because they require some sort of unmcreological 'composition' that violates the principle. it is the axiom of Unrestricted Composition that arouses suspicion. (6) Suppose the wordtypes 'master' and 'stream' are taken to be each the fusion of all its inscriptions. It doesn't matter whether there is any set. these relations.
Existence is not some special distinction that befalls some of the things there are.) It is not cohesive.except. Describe the character of the parts. especially pp. That much we can say. I never said. of course. many more. How is it done? Do we really have to? It is done with the greatest of ease. pp. To be sure.A Framework for Set Theory pays its way. and not in contrast with its surroundings. see Peter van Inwagen. it would be double counting to list the cats and then also list their fusion. like most of us all the time and all of us most of the time. the cats are the same portion of Reality either way. see Lewis. we are committed to the existence of all manner of mereological fusions. partly there.6 Composition as Identity So I claim that mereology is legitimate. and many. For a rejoinder. of course.Partly here. if you are already committed to some things. 8r . all the world's styrofoam. They just are it. If you say there 80 r Composition as Identity is something that exists to a diminished degree. so where is it? . In general. But given a prior commitment to cats. fusions of individuals and classes. fully and precisely understood. describe their interrelation. It is neither here nor there. 9 3. Commit yourself to their existence all together or one at a time. 2147. 21213. It just is them. On the Plurality of Worlds. We are not accustomed to speak or think about such things. but it is nothing else: it is part fish and part fowl. It is inhomogeneous. It is no problem to describe an unheardof fusion. (Not along some of its borders. may be as fuzzy as you please. Existence just means being one of the things there are. on the other hand. disconnected. a commitment to catfusions is not afurther commitment. Doing away with queer fusions by restricting composition cannot succeed. nothing else. not a causal unit in its impact on the rest of the world. of course you may. Its character is exhausted by the character and relations of its parts. Take them together or take them separately. The fuzzy line between less queer and more queer fusions cannot possibly coincide with the sharp edge where existence gives out and nothing lies beyond. you incur no further commitment when you affirm 9 For a fuller statement of this argument against restricting composition. you quantify subject to restrictions. 'When are Objects Parts?'. But none of that has any bearing on whether it exists. not causally integrated. muchheardof thing. unless we do away with too much else besides.) There is still a fusion. So I am committed to all manner of unheardof things: troutturkeys. You can declare that there just does n()t exist any such thing . pp. among the things you're ignoring. If. Philosophical Perspectives. For many respects of queerness are matters of degree. If you draw up an inventory of Reality according to your scheme of things. once you've said 'there is' your game is up. It is not carved at the joints. if we accept mereology. 1 (1987). All suspicions against it are mistaken. it's the same commitment either way. Only if you speak with your quantifiers wide open must you affirm the troutturkey's existence. The troutturkey in no way defies description. unproblematic. But I claim more still. 405. It is neither fish nor fowl. If you wish to ignore it. and you have ipso facto described the fusion. It is nothing over and above its parts. A restriction on your quantifiers. for better or worse. then you can leave it out. and that's enough. that a troutturkey is no different from an ordinary. so to describe it you need only describe the parts. Mereology is ontologically innocent. But existence cannot be a matter of degree. say. The fusion is nothing over and above the cats that compose it.
the plural form of the 'is' of identity. to what we were committed to before. is not to deny the existence of the whole. You can't sell them without selling it.. Partial identity admits of at least roughandready degree. But it suggests that the whole was not a seventh thing. Armstrong takes strict identity and strict difference as the endpoints of a spectrum of cases. The new commitment is redundant. with cases of more or less extensive overlap in between. Donald Baxter 'Identity in the Loose and Popular Sense'. They are partially identical. so to speak. On this view there is no one thing distinct from each of the parts which is the whole.A Framework for Set Theory the existence of their fusion. Call this the Thesis of Composition as Identity.. pp. vol. The 'are' of composition is.. Suppose the six buyers of the parcels argue that they jointly own the whole and the original owner now owns nothing. If you incur a commitment to the of. Suppose a man owned some land which he divides into six parcels. 37 . something you were committed to already. Overlap subsumes partwhole as a special case: it may be x itself. better.. In endorsing Composition as Identity. that is indeed a further commitment. that is not a further commitment. Overcome with enthusiasm for [the denial of Composition as Identity I he might try to perpetrate the following scam. Begin with New South Wales and then take larger and larger portions of Australia. because you can't sell it without selling it. and this partial identity takes the form of having a common part. the whole is simply the many parts with their distinctness from each other not mattering. It is in virtue of this thesis that mereology is ontologically innocent: it commits us only to things that are identical. It is not redundant. I say that composition . One is approaching closer and closer to complete identity with Australia. so to speak. Philosophical Papers. and then you become committed to the existence of something that bears a certain relation to it or them. pp.8. Two adjoining terra_ce houses that share a common wall are not identical. Mind. 97 (1988). Their argument seems right. This .. He sells off the six parcels while retaining ownership of the whole. it is them. or the nextdoor neighbour of. but they are not completely distinct from each other either.ll Indeed! I grant that no one of the six parcels by itself is identical to the original block of land. the manyone relation of many parts to their fusion . 'ManyOne Identity'. or the singleton of. given the old one. 10 11 82 . there is a good sense in which the six parcels and the original block are the very same thing. Armstrong and Donald Baxter. 17 (1988). For the most part. Universals and Scientific Realism. if you are committed to the existence of a certain thing or things. 193216. That way he gets some cash while hanging on to his land. M. If you are already committed to the existence of cat Possum. or y itself. and this partial identity takes the form of the wholepart 'relation' . that is a common part of x and y.the relation of part to whole. Rather. II. Armstrong. They are it. or the shadow of. or the weight in grams of. Still.is like identity.. and then affirm that there exists something identical to Possum. Also see Donald Baxter. They are partially identical.. or. p. I am following the lead of D. But the relation of identity is different. 579. It is merely to deny the additional existence of the whole . you have made a further commitment.. but they are not completely Composition as Identity distinct from each other. 1o Baxter puts it this way: The whole is the many parts counted as one thing. Australia and New South Wales are not identical.
The. if by that you just mean that something that is mother of one is identical to something that is mother of the other. Uniqueness of Composition is a third aspect of the analogy. but in the end I'm unconvinced. 7. Anything can be. If Mary's lamb goes everywhere that Mary goes. so Possum and Magpie needn't satisfy any special conditions in order to have a fusion. that Baxter defends. the oneone relations of part to whole and of overlap . and thereby you fully describe whatever is identical to Possum. The ease of describing fusions is a fourth aspect of the analogy. If Possum exists. Describe Magpie and Possum fully . My version. and so rest content with mere analogy. Unrestricted composition is a second aspect of the analogy. as just stated. if by that you just mean that something that is part of one is identical to something that is part of the other. Likewise for relational description: specify the location of Magpie and Possum.A Framework for Set Theory A doubter might seek to trivialize Composition as Identity. 4. A stronger. and they are y. Never mind that mereological relations can be equivalently restated so as to drag in identity. 8. Composition as Identity comes in different versions. version would disdain mere analogy. 12 The analogy has many aspects. His ingenious defence makes better sense of the view than I'd have thought possible. Describe Possum fully. Specify the present ownership of the six parcels and thereby you specify the ownership of the original block of land. if by that you just mean that the average of the six is identical to the one. It is this view. A kind of transitivity applies. 6 and 3 are 'on average' identical to the one number 5. then automatically something identical to Possum exists. and so are New South Wales and Australia. A fifth aspect of the analogy has to do with multiple location. likewise there cannot be two different fusions of Magpie and Possum. Possum is 'maternally identical' to Magpie. or something very like it. they are strikingly analogous to ordinary identity. Just as there cannot be two different things both identical to Possum. The six numbers 2. and thereby you specify the location of their fusion. if by that you just mean that the fusion of the parcels is identical to the block. Rather. as he said.the character of each. is analogical. the ordinary oneone kind. and if this is so not just as a matter of fact but as a matter of 85 But the doubter misses the point.real point is that the mereological relations (however restated) are something special.as kinds of identity. and also their interrelation . Of course the six parcels are identical to the original block. 12 Here I part company with Baxter. and stranger. so likewise it is redundant to say that Possum and Magpie both exist and their fusion exists as well. They are unlike the samemother relation or the a verageof relation.and thereby you fully describe their fusion. It would insist that there is just one kind of identity. limiting case of identity in the broadened sense. If x is them. just as Possum needn't satisfy any special conditions in order to have something identical to him. So strjking is this analogy that it is appropriate to mark it by speaking of mereological relations the manyone relation of composition. So what? Composition as Identity Ordinary identity is the special. Four of them have come to our attention already. then x is y. thus: Of course the terrace houses are partially identical. . You could do the same trick with any old relations. and that composition involves this kind of identity. The fust is the ontological innocence of mereology: just as it is redundant to say that Possum exists and something identical to him exists as well. with Magpie and Possum together as a plural middle term. the oneone relation that each thing bears to itself and to nothing else. likewise if Possum and Magpie exist then automatically their fusion exists. But that shows nothing about the innocence of mereology.
I know of no way to generalize the definition of ordinary oneone identity in terms of plural quantification. if a fusion is multiply located wholly present at different places or times or possible worlds then also its parts must be multiply located. and there are some things such that y is one of them but none of the xs is. ad infinitum. or he may actually have more toes than he might have had. we have a highly mysterious necessary connection between distinct existences. rr. To be sure. but to the form of the description. . says that P is part of P & Q. Perhaps I can. Instead. Set theory is not innocent. It would be nice to illustrate this by example. the singleton.A Framework for Set Theory absolute necessity. It is the same thing partially present. in no way do we mention one thing that is the many taken together. Likewise if it turns out that the lamb is part of Mary. That is how we can have set theory. If a conjunctive universal P & Q is the fusion of its conjuncts P and Q. There is a problem: the fusion of P and Q exists whether or not P and Q are coinstantiated.. This completes the analogy that I take to give the meaning of Composition as Identity. It is not the same thing wholly 13 Armstrong. After all they are many while it is one. we speak of them as many. instead of universals. 87 86 . Since it just is them. And another. p. Pay it.13 then it is automatic and unmysterious that both P and Q must appear wherever P & Q does. In general. we do mention one thing that is the many taken together. But alas. But to these cases I apply the principle contrapositively. But if y is the fusion of the xs. its trouble is that when we have one thing. and if Mary is wholly present wherever she goes. Because the thing varies with respect to its parts. 36. the same man may have more teeth when young than when old. And in the second place. still we do not really have a generalized principle of indiscernibility of identicals. What's true of the many is not exactly what's true of the one. or different things wholly present. vol. whenever there are some things. it has its limits. Does this show that conjunctive universals cannot be the fusions of their conjuncts? No. it only shows that if P and Q are not coinstantiated. then there is no mystery at all about their inseparability. and another. the same road may have a lane on the hilly stretch that it lacks on the plain. but according to Armstrong we have no conjunctive universal P & Q unless P and Q are coinstantiated. Its trouble has nothing to do with gathering many into one.. The number of the many is six. as it might be. But if it turns out that Mary and the lamb are identical. and in no way mysterious. We know that x and yare identical iff. But we can perfectly well say that whether the fusion is a universal or not depends on whether P and Q are coinstantiated. Plural quantification is innocent: we have many things. even though the many and the one are the same portion of Reality. It does matter how you slice it not to the character of what's described. Composition as Identity present at multiple locations. then somehow we have another wholly distinct thing. In the first place. And the singletons of the many parts are wholly distinct from the singleton of the one fusion. then there are some things such that each of the xs is one of them and y is not. and the character of that portion is given once and for all whether we take it as many or take it as one. whereas the number of the fusion is one. Mereology is innocent in a different way: we have many things. As for alleged examples that involve multiply located particulars. I conclude that we do not have genuine multiple location. of course. their fusion is not in that case a universal. I definitely disbelieve in those. it cannot appear anywhere without them. x is one of them iff y is one of them. but this one thing is nothing different from the many. Universals and Scientific Realism. then again the inseparability is automatic. if there are multiply located universals. But that's the price for mathematical power.
as it turns out. I want to say that something is 'large' iff it has as many atoms as there are in all of Reality. to set a standard of size. 14 3. won't count. Without benefit of ordered pairs. its parts will admit of an endless ordering. this can be done in elementary logic: using identity.7 Distinctions of Size Our framework. . Dedekind Schema: If x is a proper part of y. But then y is infinite because of the gunk. (2) their fusion is the whole of Reality. but at least we may affirm it as a principle of the framework. This may seem useless until we have ordered pairs. exactly one atom of x. and if each atom of y . An obvious consequence of the definition is that an atom is finite.that Reality has only finitely manyatoms. we can do much more. and we have one ready to hand. We begin with the distinction between finite and infinite things. I cannot say quite that. affords the means to define distinctions in the size of things. a relative distinction. we can define 'thing with more than seventeen atoms' and the like. . it. may be either finite or infinite. and if each atom of 88 . .A Framework for Set Theory Distinctions of Size x is such that exactly one atom of y .) Next. in other words. We'd want it to come out that a single atom is small. 'Size' here is measured by atoms. Useless as a definition. namely the relation of part to whole. Definitions: x is infinite iff x is the fusion of some things. We may affirm principles of the framework that connect 14 A price we pay for not saying just what we want to say concerns the smallness of atoms. A thing that consists entirely of atoms. An infinite thing is one with infinitely many parts. in that case. atomless gunk. leaving out only some atorriless gunk. (Note that if x includes all the atoms of y. Note that any bit of atomless gunk is infinite: it is the fusion of all its proper parts. But with the resources of plural quantification and mereology. Roughly. each of which is a proper part of another. and (3) each of them exactly one atom that is part of x and at most one other atom. unless it is the only atom. each of which is a proper part of another. But instead it turns out that an atom is small iff there are three or more atoms. Otherwise x is small. that an atom is small iff there are two or more atoms. except in the special casecontrary to set theory. up to a point. We may take all the atoms of Reality. if there is any. then y is infinite. But I can say that something has at least as many atoms as there are in all the rest of Reality. otherwise 'small'.. Likewise anything that has an atomless part is infinite. a partial ordering will do. One is Dedekind' s: when we have infinitely many things. yes. which comes to the same thing. There are other wa ys ofdefining infinity.. Definitions: x is large iff there are some things such that (1) no two of them overlap. then all those things correspond oneone with only some of those things. It follows immediately that any part of a small thing is small.on the other hand. Otherwise x is finite. and to any physics that posits points of the spacetime continuum . however many of them there may be. then we do not know that y has infinitely many atoms. of course.
and if x is small. and given some other things (not necessarily different). and if for each of the latter things there is one of the former things that . this Distinctions of Size Definition: Suppose we have some things such that some large thing does not overlap any of them. each of the 'principles of the framework' that I have affirmed in passing.. (3) for each of the former things. (2) each of the latter things is the fusion of one of the former things and one atom of x. then y is small. we could look for an elegant systematic theory that would yield. it. . I cannot do this in full generality. (The small thing may itself be finite. And we may affirm. the fusions of atoms are more than barely many. as a principle of the framework.. I would like to introduce plural predicates 'few' and 'many'. that something is small iff its atoms are few. it. as a principle of the framework. exactly one atom of a thing y. such that (1) x does not overlap the fusion of the former things. as axiom or as theorem. 15 Beware. Unless there are but finitely many atoms. But 'many' is not a single size. and many fusions of atoms. won't count as few!) Replacement Schema (plural version): Given some things.. then (1) any finite thing is small. Much more could be done. and if for each atom of y there is an atom of x that . within the vocabulary of the framework. and indeed (2) any fusion of a small thing with a finite thing is small. If there is something infinite that consists entirely of atoms. then the latter things are few. we affirm each sentence made by filling in the blank and prefixing universal quantifiers as required. if each of the former things . this Replacement Schema (singular version): If each atom of a thing x . one of the latter things is the fusion of it and one atom of x.) Set theorists beware: the schema is not set theoretical. 15 But without ordered pairs. (Two halves of Reality. and if so the fusion has the same small infinite size. The atoms are barely many. 'large' is a single size: all large things are equal. and if so the fusion is finite too. again. and others besides.. as a principle of the framework. exactly one of the latter things. Further.A Framework for Set Theory the distinction of infinite versus finite to the distinction of large versus small.. Otherwise they are many. . We have in store just what we shall need later. the plural distinction of few and many differs in one way from the singular distinction of small and large. corresponding to the singular predicates 'small' and 'large'. Then they are few iff there is some small thing x. and if the former things are few. (That is. and it in no way constrains the size of Reality. so my definition applies only to things whose fusion does not fill up too much bf Reality.. and (4) no atom of x is part of two or more of the latter things. There are many atoms. each having as many atoms as there are in all of Reality.. but there are more of the latter than of the former. and there are some things.) We may also affirm. But we have done enough to be going on with. though only two. We connect our plural and singular predicates by afflrming. to define distinctions of size (see the appendix). or the small thing may be small and infinite. where again the atoms in all of Reality set the standard.
. Hypothesis U: If some things are small and few. We shall take on· our commitments in two stages. the size of Reality. Hypothesis P: If something is small. their fusion is small. to the settheoretical axioms of Power Sets and Unions . We can't say yet just what this great deal of somthing consists of. we shall assert that there is a very great deal of something.to jump the gun . but of course . as we shall see later.1 The Size of Reality Now that we have our framework. holding the primitive notion of singleton still in abeyance. These are not innocent principles of the framework. but rather are forced upon us by our acceptance of orthodox.we know it must consist predominantly of singletons. settheoretical mathematics. We may affirm three hypotheses about. First. 93 . then its parts are few. The first two correspond.4 Set Theory for Mereologists 4. it is time for an end to our ontolgical innocence.
nor is any singleton part of the null set. and speak of the singleton of something. is something that is the singleton of something. Hypothesis P would say. straightway entitles us to treat it henceforth as a functor. if the size of Reality were like the limit cardinal bethomega. By itself. 4. Our next axiom specifies that different t. The story is well know as told in set theory. It does not happen again until we reach an 'inaccessible' infinite size l that transcends our commonplace alephs and beths in much the same way that those transcend mere finitude. nor does any part of the null set overlap any singleton. that a finite thing has only fmitely many parts. Hypothesis I: Some fusion of atoms is infinite and yet small. It is only when combined with Hypotheses P and U that it demands the high jump. for instance. Hypothesis U would say. of course. It excludes only the smallest of infinite sizes. But with Hypotheses P and U in place. Hypothesis P would be true. truly. That happens. What might there be uncountably and inaccessibly many of? Not I Distinctness: No two things have overlapping singletons.2 Mereologized Arithmetic Our settheoretical primitive can at first be taken as a twoplace predicate: x is a singleton of y. and nothing else has a singleton.hings have distinct singletons. 94 95 . ifReality is countably infinite. but our framework of plural quantification and mereology has enough power that we can tell it there too. Distinctness in the sense of mere nonidentity follows: nothing is the singleton of two different things. Hypothesis I is undemanding. The axiom demands distinctness in the strong sense of nonoverlap.) Our next axiom specifies the domain of the singleton function. If we want orthodox mathematics. Mereologized Arithmetic cats. if the size of Reality were like bethone (the continuum). Functionality: Nothing has two different singletons. The problem comes in demanding that the two be true together. for instance. any singleton has a singleton. Finally we have an axiom of induction: it says that we can get to anything there is by starting with 'Strongly'inaccessible. We must demand a little more. 'Small' and 'few' would then mean 'finite'. that the fusion of finitely many finite things is still finite. Hypothesis U would be true. we cannot be content with mere countable infinity. or alephone. (A singleton. But the first of our axioms for it. it's safe to say! So now it's time for the primitive notion of singleton. or bethseventeen. to demand a little is willynilly to demand a lot.Set Theory for Mereologists Hypotheses P and U would be true if Reality were barely infinite. truly. and that singletons are distinct from parts of the null set. Domain: Any part of the null set has a singleton. not quarks. any small fusion of singletons has a singleton. we saw. with a countable infinity of atoms. not spacetime points. Each hypothesis also can be true at certain larger infinite sizes.
by the Dedekind Schema. being atoms. each atom of x is such that exactly one atom of y has it as singleton. accompanied by just one other atom. Distinctness implies that the null set has no singletons as parts. But by Distinctness. we can show that something that consists entirely of atoms. hence no singletons thereof. if the null set is an atom.7. Let us set that right. we know that anything finite is small. and nothing else. Singletons. a fusion of singletons. then everything is one of them. or a fusion of a part of the null set and some singletons. because the solitary singleton. We defined a class as a fusion of singletons. and y is the fusion of all singletons. and only if there are three does it follow that atoms are small (see section 3. there are singletons. By Induction. for instance the singleton of the null set. So the axioms could have been rewritten just in terms of 'singleton '. 'null set' looks like an unacknowledged second primitive. would not be small. some singletons. Recall our previous definitions in terms of 'singleton'. So if x is 97 . it cannot have as a part either any part of the null set or any singleton other than itself. Mereologized Arithmetic the fusion of all singletons of singletons. We defined an individual as anything that has no singletons as parts. each atom of y has as singleton exactly one atom of x. or a fusion of a part of the null set and some singletons . are not singletons of singletons. But I could not have shown that there were three. and iterating the operations of singleton and fusion.for those are the only things there are. 2 The things that have singletons are exactly the parts of the null set and the small fusions of singletons. and that there is such a thing. and if every fusion of some of them is one of them. So singletons are small. is infinite. There would be no sma)) fusions of singletons. QED Once we know that singletons are atoms. each singleton has as singleton exactly one singleton of a singleton. if every part of the null set is one of them. if every singleton of one of them is one of them. I could have shown that there were two atoms: namely the null set and its singleton. Proof What could be a part of it? Only a part of the null set. by Distinctness.Set Theory for Mereologists the null set and its parts. and the members of a class as things whose singletons are parts of that class. Hence singletons themselves are subsumed under small fusions of singletons. Further. their fusion. and Domain looks to have a redundant clause. So it can have no part except itself. Without the assumption. So y is infinite. To see this. for instance the singleton of the null set. note 14). a fusion of singletons. Induction: If there are some things. and let them be taken into mereologized arithmetic. But by Distinctness. the null set and its singleton. note that if the clause 'any singleton has a singleton' were stricken from Domain. Proof By Domain. Hence it follows from the axioms that the null set may be defined as the fusion of all things that have no singletons as parts. and all singletons have singletons. QED Once we know that something that consists of atoms is infinite. then all the axioms of mereologized arithmetic would be satisfied if there existed just two atoms. that is. A singleton must be an atom. In the axioms as stated. then x is a proper part of y. and otherwise the singletons of the null set and one of its proper parts. namely the fusion of all singletons. are finite. by Functionality. Further. each singleton of a singleton is the singleton of exactly one singleton. that is. We 2 So why did I write it that way? Because I to assume that each singleton had a singleton before I could establish that something consisting of atoms was infinite and hence that singletons were small. we have that everything is either a part of the null set.
Suppose class x is part of class y and z is a member of x. some singleton is part of x but not of y. by definition of membership. also it says that an proper classes are the same size namely. I advanced four intuitive theses about the mereology of classes. or is a fusion of singletons. I want to bring us back to where we started. Priority Thesis: No class is part of any individual. Right to left. plus principles of the framework. Proof That means that a fusion of things without singletons as parts itse1fhas no singletons are parts. Let it be z's singleton.Set Theory for Mereologists defined the null set as the fusion of all things that have no singletons as parts . a class is a set iff it is small. something is part of the null set iff it has no singletons as parts. I concluded that the class of cats was the fusion of singletons of cats.e. A large class is a proper class. Since x is a fusion of singletons. or is a fusion of a part of the null set and some singletons. following von Neumann: not only does it say that the proper classes are larger than all others. Suppose class x is not a part of class y. First Thesis: One class is part of another iff the first is a subclass of the second. by definition of class. and singletons are atoms. QED Fusion Thesis: Any fusion of individuals is an individual. First we regain the original four theses. which holds by mereology plus the fact that singletons are atoms. QED Division Thesis: Reality divides exhaustively into individuals and classes. that means that everything either has no singletons as parts. So z is member of y. Now I am arguing in the opposite direction: starting from the axioms and definitions of mereologized arithmetic. and the urelements.3 Four Theses Regained At the start. We defmed an urelement as something that has no singletons as parts. in other words. which holds by mereology.the same defmition that now follows from our axioms. z is a member of x but not of y. The things that can be members are exactly the sets and the individuals. and it cannot be a member of anything. Then we shall regain orthodox set theory. so x is not a subclass of y. It is a strong form of the principle. it lacks a singleton. but is part of something else of which the same is true. And by defmitions and mereology. and so forth. z's singleton is part of x and hence part of y. Here is the principle of Limitation of Size. an individual other than the null set. Proof By our present defmitions. or a mixed fusion of individual and class. the null set. or is a fusion of singletons. QED 4. in other words. the largest size. Arguing from these theses together with orthodox set theory. By our present definition of membership. the small classes. in an uncommonly explicit form. Prove the contrapositive. everything is an individual. that Possum was a member ofjust those classes that had his singleton as a part. Four Theses Regained Proof Left to right. or is a fusion of something with no singletons as parts and some singletons. i. QED 99 . We've already proved that everything either is a part of the null set. We defmed a set as either the null set or a class that has a singleton. a class. Accordingly. Proof That means that no fusion of singletons is part of anything that has no singletons as parts.
by Distinctness. We have seen that singletons are finite and that they are small. or if none of the given things are members of x. 4.4 Set Theory Regained From the axioms and definitions of mereologized arithmetic. hence small. and second a classcomprehension schema to tell us that there is a class .) There are normally restrictions. a version of Aussonderung that tells us that given a set x and a class y. Every class has members. all members of x have singletons. Otherwise x is a small class. so by Unrestricted Composition we have the fusion of all singletons of members of x that are among the given things. no class has the same members as the nun set. It is part of the small class x. This amounts to a special case of the plurally quantified version: given a set x. It has no members because. we are committed also to the unconditional versions. Proof If two classes had the same members. Our plurally quantified Aussonderung is stronger. and given some things that are all and only the satiifiers if suchandsuch formula. Since there is something infinite that consists of atoms. plus principles of the framework. when it comes to proving theorems in a formal system. (Or we may get the same effect in two stages: first. Having affirmed the hypotheses. cases they will be weaker than usual.perhaps proper of all and only the sets and urelements that satisfy suchandsuch formula.is smal1. and the fusion of these singletons is the class of x and y. Proof If x is the nun set. there is a set of all and only those of the given things that are members of x. hence a set. following Skolem. then the required set is the null set. because they will be conditional on our hypotheses about the size of Reality. and hence a set. on what the formula may be. 100 . But beware: this is useless strength. Set Theory Regained Proof Each of x and y has a singleton. they would be two different fusions of the same singletons. and given some things. QED Extensionality: No two classes have the same members. we next derive versions of the standard axioms for iterative set theory. QED Ordinarily. QED Pair sets: If each of x and y is an individual or a set. then there exists a set of x and y. For the formal system in question 101 Proof It is a set by definition. imply our Main Thesis: The parts of a class are an and only its subclasses. In some cases these versions will be stronger than usual. the fusion of two singletons something small and something finite . QED Aussonderung: Given a set x. Aussonderung is stated as an axiom schema. it has no singletons as parts. more or less stringent. we recall. however.Set Theory for Mereologists These four theses. there is a set of all and only those of the given things that are members of x. because schematic formulations will give way to plural quantification. there is a set of the common members of x and y. The class of x and y is small. Null Set: The nu]] set is a set with no members. In other . unlike the null set. contrary to Uniqueness of Composition. This is the class of all members of x that are among the given things.
an individual is grounded: it has no members. a counterexample class C.···.. by Functionality and Distinctness. A counterexample to Fundierung might be an infinite class of XI' x 2 . a counterexample class C..intersects C+ by intersecting C. i. First. to plurally quantified versus schematic formulations of Replacement and Choice. Replacement: If there are ordered pairs whereby each member of a class x is paired with exactly one member of a class y.. Is C+ another counterexample class? No. or simply a single thing that was its own singleton. else the singleton belongs to.. The same comment will apply. and finally Xn contained XI. then there are some things that are all and only the things that .e. y is a set. X 2 . But yes. Third.the members of C. QED Choice: Suppose X is a class. and if. call it g. Second. and hence intersects. The remaining case is that the fusion is a small class.Set Theory for Mereologists will have to include some axiom system. Then one of the given grounded things. because each member of C+ . Recall that the atoms. mutatis mutandis. QED 102 Proof Call something grounded iff it belongs to no class that is a counterexample to Fundierung. where XI contained X 2 as a member. being incomplete.. Proof Fill the blank in the Replacement Schema (singular version) with 'is such that its member is paired by one of the given pairs with the member of'. We conclude that if x is small then so is y. and . that will cramp our style as surel y as if Aussonderung itself had been given as a mere schema.. So the point of using plural quantification here is not to increase our power to prove things. intersects nothing. Let C+ be the class that contains all the members of C. Nevertheless. If the fusion is an individual. in which case the given grounded things are small classes too. it fails to supply enough of them. Suppose for reductio that the fusion belongs to. and X 2 contained X 3 . or a proper class. say that of Kuratowski: the (ordered) pair of u and v is the twomembered set whose members are the singleton of u and the set of u and v. X 3 . a fusion of grounded things is grounded. and suppose there are some ordered pairs whereby each member of X is paired with 10 3 . This completes the reductio. We assume that pairing is defmed here In some standard settheoretical way. also intersects C. . so is its singleton. Set Theory Regained Fundierung: No class intersects each of its own members. and so on. So Y is a set if x is. the singleton subclasses. ••• . and therefore intersects. Recall that we can write principles of plural 'comprehension': if there is at least one thing that .. and so completes the proof that everything is grounded. of x and y correspond oneone with their members. If it is a mixed fusion of individual and class. We noted that such principles are ontologically innocent and altogether trivial. because it contains the grounded g. and conclude that Fundierung can have no counterexamples. an axiom system for the framework must supply them. inevitably partial. in which case y must belong to C. We show by Induction that everything is grounded. for the framework. What we get out of our strong Aussonderung depends on what we can put into it. and g as well. and if for each member of y there is a member of x that is paired with it. or a finite class of XI. if y is grounded. but rather to tell the whole truth. and if x is a set. Fundierung says that none of these things can happen. we already know that it is grounded. it is grounded because it isn't a member of anything. Xn' where XI contained X 2 . contra the groundedness of y. and so intersects no class that contains it. and g as well.
there is a set of all members of members of x. there is a set of all subsets of x. which is an atom. is available in the present system as welL We knew already that the axioms of mereologized arithmetic had implications concerning the size of Reality. Each of these classes. Given mereologized arithmetic. Then anything is the fusion of a few small things: namely. its subclasses are few. then if x is a set. and therefore is a set. American Mathematical Monthly. Suppose for reductio that mereologized arithmetic holds but Hypothesis U is false. and in particular all fusions of atoms. which is finite. y contains their fusion. By Hypothesis P. Burgess's result indicates that they imply Hypothesis U. Then by the Replacement Schema (plural version) their singletons are few. QED Unions: Given Hypothesis U about the size of Reality. Then the subclasses of x are also small. First. bypassing Levy's use of unconditional Power Sets and therefore not relying on Hypothesis P. 104 follows from evident principles of the framework. which is impossible. y contains them. Proof Apply the First Choice Schema. the parts of x .) Assume (*). such that if x is small. we can code any x by a class y. x is small. so y is smalL We recall that since the fusion of all singletons is infinite and consists of atoms. Then.that is. Azriel Levy proved that unconditional Unions is redundant in von Neumann's set theory. we have a class. then y is the class of members of members of x. since by mereologized arithmetic there are infinitely many atoms. of a few small things. (I fear it is not a sufficiently evident principle in its own right. and finally by the singleton of that fusion. (Note that things which appear in pairs must have singletons. and no two members of x are paired with the same thing. Burgess has shown that an improved version of Levy's proof. First. but only part. Otherwise. So z is small. Then there is a class y such that each member of x is paired with exactly one member ofy. John P. then by the fusion of those singletons. y is small too: if any singletons are parts of x. whenever we have some of them. 3 3 It is not really necessary to take Unions in this conditional form. I conjecture that (*) If there are infinitely many atoms. then by the few singletons of those classes. filling in the blank with 'is paired by one of the given pairs with'. of a direct proof that mereologized arithmetic implies Hypothesis U. so they are sets. The fusion y of these singletons is the class of all subclasses of x. that enables us to code all things. Here is part. and they have singletons. Paradox. and if some large thing is the fusion of a few small things. is still small. 'On von Neumann's Axiom System for Set Theory'. the fusion of something small with one further atom. then Reality itself is the fusion of a few small things. we have by (*) that Reality is the fusion.Set Theory for Mereologists at least one thing. 75 (1968). its intersections with some of the few small things whose fusion is Reality (whichever ones of them it overlaps). we can code it by the few small classes that code those things. then if x is a set. 7623. hence.) QED Power Sets: Given Hypothesis P about the size of Reality. Let y be their fusion. see Levy. These are the atoms of y. consider those classes that are members of x. I believe it in virtue of a suspect analogy with what's true of things smaller than Reality. Set Theory Regained Proof If all members of x are individuals (the null set or proper parts thereof) then the required set is the null set. since it is a member of something. by single atoms. If an arbitrary thing is the fusion of a few small things. QED 10 5 .is the dass of all subsets of z. The fusion z of y with the singleton of the null set one further atom . The coding goes thus. Proof Since x is a set. by the reasoning of Cantor's Theorem and Russell's . pp. and if any individuals arc parts of x.
since these sets are some of the previous sets whose union was a set. the fusion of all singletons. Let us set that right. Distinctness: No two things have identical successors. In the axioms as stated. Infinity: Given Hypothesis I about the size of Reality. Induction: If there are some things. so that none is greatest. and in either case will have a singleton.5 Ordinary Arithmetic and Mereologized Arithmetic We could write the Peano axioms for the arithmetic of the natural numbers. Functionality: Nothing has two different successors. and (2) whenever x and yare two of them. any successor has a successor. Each atom must be either an individual or a singleton. then everything is one of them. as follows. hence small. and such that their union is a set. So y is the fusion of a few small things. to be sure.the class of these will be the desired nesting . Replace each atom by its singleton: that will not affect the smallness of the things and their fusion. and hence a set. is a set that is infinite under one possible settheoretical definition of infinity. and we may conclude that there are some of the given sets . there must be few of them. Further. which is the large class of all sets and individuals but not yet an infinite set. Each of these things must be a small fusion of atoms. and if every successor of one of them is one of them. Domain: Zero has a successor. the union of these sets is part of a set. but with the set theory now at hand. if zero is one of them. there are some things such that each of them is a proper part of another. For all we knew without Hypothesis I.Set Theory for Mereologists must be small. the union of which is a set. we know that we can get to various more familiar verSIons. 'zero' looks like an unacknowledged second primitive. and only proper classes were infinite.namely. We already knew that we had an infinite class . N ow we can say that there are some . and nothing else has a successor. filling in the blank with 'is properly included in'. given Hypothesis I. either x is properly included in y or y is properly included in x. Induction implies that everything is either zero or a successor. nor will it affect the fact that each is a proper part of another. nor is zero identical to any successor. with 'successor' as primitive. What we have. QED 106 4. hence small by Hypothesis U. and also under our original mereological definition.such that (1) each of them is properly included in another one. it might have been that all sets were finite. since the class of them is part of the small class x. then there is a nesting with no greatest member. Further. Distinctness .) Proof By Hypothesis I. even given Hypotheses P and U.sets such that each of them is properly included in another. and such that their fusion is a small fusion of stoms. Apply the Second Choice Schema. Therefore y is a set. whenever x and yare two members of it. QED Ordinary Arithmetic and Mereologized Arithmetic This is not the most familiar formulation of a settheoretical axiom of infinity. such that. either x is properly included in y or y is properly included in x. (A nesting is a class of sets.
But I give uniformity of the amendments precedence over simplicity of the final formulation. The second step stipulates that the new numbers really are new. (Strictly speaking. and 'null set' instead of ' zero'. do we mean everything. So to be assured that we get everything. We assume further that if there are many zeros. Specifically. and so it goes. amd the fusion of that successor with 2.as well as small ones. and also the new 'number' that is the fusion of 1 and 3.that arouse incredulity! Surely sIngleton IS one concept. So we may reserve the proper name 'zero' henceforth for the one big zero.we're not Pythagoreans! But by the time we're done mereologizing arithmetic. hence it follows from the axioms that zero may be defined as the thing that is not a successor. .Set Theory for Mereologists says that zero is not a successor. and speak of all the 'zeros' (including the big one) henceforth as 'parts of zero'. The first step is the crucial one: stipulate that besides getting new numbers from old ones by succession. for instance. The extra clause for Domain is 'any small fusion of successors has a successor'. 4.) The final step. The extra clause for Induction is 'if every fusion of some of them is one of them'. gIven In one case by the axioms of mereologized arithmetic 108 .for instance. we need to make a place for cats and quarks and points. by taking small fusions of successors.. We have oldfashioned numbers 0. So we amend Distinctness: in both its clauses. or that zero is really the fusion of 17 with What's in a Name? the successor of the fusion of 9 and 6. throughout. they are all and only the parts of one big zero. we can also get new 'numbers' from old by fusion. Speaking unrestrictedly. to the effect. and presumably these things are not successors or fusions thereof. and that there is such a thing. singl. . 3. Wh y not: 'if every small fusion of some of them is one of them'? That would correspond directly to our new way of generating numbers. So the axioms could have been rewritten just in terms of 'successor'. or do we just mean everything in some restricted domain? In the case of ordinary arithmetic. we needn't amend Distinctness: not overlapping any part of zero is equivalent to not overlapping zero. that Possum is neither a . in effect. and the successor of that second fusion . there are large fusions .eton nor a successor) is just a formal modus operandi. like other numbers. But when we say 'everything'. So we amend Domain and Induction by writing 'any part of zero: Instead of 'zero'.2.. Reality . that we may have not just one zero but many: many starting points that are neither nor fusions of successors. we may hope to say 'everything' and mean it without restriction. The third step allows. Taking this step means adding clauses to two axioms. large or small.. If we're quantifying unrestnctedly. We wouldn't want it to turn out that 17 is really the fusion of 6 and 42. we must close under fusion generally. The numbers so obtained. Four steps take us from ordinary to mereologized arithmetic. All there is to our understanding in either case (apart from some via negativa. and the successor of that fusion. we replace distinctness in the sense of mere nonidentity with distinctness in the strong sense of nonoverlap.. these successors have their own successors. is not substantive but merely verbal: we write 'singleton' instead of 'successor'.6 What's in a Name? . successor IS another? But this is to overestimate how rich a conception we have of the membersingleton function and the numbersuccessor function.. we had better just mean everything in a restricted domain . and they join in fusions. have successors of their own. 1. now. I claim.
Suppose Gregson tells Holmes that a thief is plundering Piccadilly. the modus operandi of the looter. And 10.. Then we must admit that somehow we have managed to give 'singleton' an unequivocal meaning. maybe Possum has a successor. A singleton by any other name would smell as sweet. (Hence the appeal of structuralism. within a certain restricted domain. is that we gave them both the same meaning. 61 (1983). are silent about what does or doesn't go on outside that restriction. pp. pace structuralism. And I am not just saying. for arithmetic but not for set theory? You get the worst of both worlds. what with the match of formal modus operandi and the lack of known differences. Now even Watson may arrive at the working hypothesis that these scoundrels are one and the same. And we would do well to follow his example. is just like that of the plunderer. when he's looting in Piccadillylike places. such feats of meaninggiving are possible. then we get a settheoretical model of 110 What's in a Name? arithmetic .for his distinctive modus operandi. And if we have somehow given meaning both to 'singleton' and 'successor'. They do not deny that things falling outside the restriction may have successors. and I happen to choose this one. and the police know nothing about him except for his distinctive modus operandi.one such model among many. as Quine might. In particular. Not. Yet you still bear the burden of denying our naive conviction that 'successor' has in fact been given an unequivocal meaning. 62 (1984). . Assume. Australasian Journal oj Philosophy. as Zermelo did and as structuralists do. why not also for 'successor'? What's the good of being a parttime structuralist. and simplicity favours the hypothesis that they are the same. it is not known whether he operates anywhere else. that we should take 'singleton' as primitive. there's nothing in our meager conceptions of the membersingleton function and the numbersuccessor function to prove them different. Australasian Journal oj Philosophy.) And what's said about the modus operandi of 'successor' is that. and the police know nothing about him except.. it's hard to understand how determinate meaning in thought and language is possible at all 4 .Set Theory for Mereologists and in the other case by the axioms of ordinary arithmetic. the notion usually chosen as primitive . You might as well be hanged for two sheep as for one. If there is something about certain meanings that especially suits them to become the meanings of primitive terms . then it matters not at all whether we say 'singleton' or whether we say 'successor'. then the obvious hypothesis. And Lestrade tells Holmes that a thief is looting all of London.but one that could and should have been. that if we take the null set as zero and the singleton function as successor. and 'Putnam's Paradox'. I am saying this. which quantify restrictedly.then it is to 4 See David Lewis 'New Work for a Theory of Universals'. indeed. What sets Zermelo's modelling of arithmetic apart from von Neumann's and all the rest? It is Zermelo's that identifies the primitive of arithmetic with an appropriately primitive notion of set theory.and if there isn't. you know not how. the socalled 'Zermelo numbers'. For all that arithmetic tells us. that when we have many models we can choose one aribtraril y. The axioms of ordinary arithmetic. But if we did it for 'singleton'.. And if they are the same. I I I . and maybe the fusion of Possum's successor and Magpie's has a successor. and Magpie has a successor. pp. In short. I am not just saying. the modus operandi of the London looter is just the same as that of the Piccadilly plunderer! Or rather. Rather. You bear the burden of admitting that somehow. it's just like that of 'singleton'. 34377. 22136.
... and nothing else has a successor. nor does zero overlap any successor. and limit ordinals... 2. . Its fusion is the number omega + omega. and rest content with examples.Set Theory for Mereologists be expected that. .. Induction: If there are some things. . in which the only memberless things are the null set.. omega. Applying our definitions of class. one primitive term should take on the meaning that already belongs to another primitive term. with each limit ordinal taken as the fusion of all preceding successors. 4. If we embed ordinal arithmetic in full mereologized arithmetic. omega + 1.. . and mixed fusions of the null set and classes... proper classes. 3. the ordinals are zero. 1.. but only when the numbers fused meet a special condition. One is pure mereologized arithmetic.. 2. as long as the fusions remain small. Domain: Zero has a successor. membership.. .7 Intermediate Systems Between ordinary arithmetic and mereologized arithmetic there lie intermediate systems. Instead there was the parallel conjecture: II. . which makes it the class of its predecessors 0. we have the system of ordinal arithmetic. omega + 3. and so on (written with 'successor' or 'singleton'. . . omega + 1. and strengthen Distinctness by putting nonoverlap in place of nonidentity.. Take only the first two of our four steps (plus the fourth if you like) to get Functionality: Nothing has two different successors. omega + 1. omega + 3. omega + 2. any successor has a successor. I skip the definition. Not only must it be a small fusion of successors. any small fusion of successors has a successor. Likewise omega + omega is the fusion of 1. successors. Completeness of Mereologized Arithmetic We take fusions of numbers to get new numbers..8 Completeness of Mereologized Arithmetic Let's imagine a fictitious history of geometry: there never was a parallel postulate. omega + 2. we get pure set theory: set theory without urelements. which makes it the class of its predecessors 0. and if every fusion of some of them isone of them. 3.. 1. The next fusible segment consists of the numbers 1. a restriction which omits most of the nonsuccessors we think there are. 112. 3. .. it must be an initial segment without a last term. 4. the fusien of 1. . Omega has a successor.. 2's singleton. Distinctness: No two things have overlapping successors. it doesn't matter).. .. " omega + 1. 2.. whenever possible given constraints on formal modus operandi. 1's singleton.. . .. We adjust the axioms of Domain and Induction to accommodate the limit ordinals and their successors. .. And so on. Then comes omega + omega + 1. . if every successor of one of them is one of them.. The other intermediate system differs right at the beginning. its fusion is the new 'number' omega. but within which Unrestricted Composition holds. we find that although we are still following Zermelo's plan when it comes to successors . 2. One fusible segment consists of the numbers 1. omega + 2. In short... 3. . which exploits the resources of plural quantification to define precedence. 2. This 'everything' is presumably said under a restriction. Omega. then everything is one of them .'successor' means 'singleton' we are following von Neumann's plan when it comes to limit ordinals. . also. if zero is one of them. is the union of O's singleton. 2..
not some elementary approximation thereof: InduGtion is to be stated by means of plural quantification.3. s Godel has not been vanquished. But arithmetic is not to blame. Each must be the image of the other under a oneone correspondence. And we found plenty of cause to complain that we lack any good understanding of the primitive membersingleton relation. someone proposed a diagnosis. Suppose we have two rival singleton functions. congruence l and congruence 2 • both satisfying our accepted axioms of geometry' not including any parallel postulate . I mean. That limits the difference between our two rival 5 See any discussion of the 'categoricity of secondorder arithmetic'. successor 1 and successor 2' both satisfying the Peario axioms. The trouble. If we have two rival successor functions. or any other statement formulated in the language of arithmetic. of course. or puddles or quarks or spacetime points.is like the case of' ordinary arithmetic. are not singletons and have no singletons as parts. Even without any understanding of what singletons are. It is as true as Completeness of Mereologized Arithmetic ever that any consistent formal system of arithmetic. and the reason we cannot settle Goldbach's conjecture is that it is true for successor 1 and false for successor2. We may well ask whether our incomplete understanding of the primitive notion of set theory bears some of the blame for our incomplete knowledge of what's true in set theory. set theory. Now. pp. that is. is that the framework of plural quantification does not admit of any complete formalization. Mathematical Logic. 161 . for instance in Robbin. how about mereologized arithmetic? Is it like the case of geometry without a parallel postulate? Or is it like the case of arithmetic? Since mereologized arithmetic is set theory.' This time. still we somehow seem to know that cats. The case of arithmetic is different. What if someone proposed a parallel diagnosis of our failure to settle Goldbach's conjecture? 'The trouble may be that the accepted axioms do not characterize the primitive notion of successor well enough. Better understanding would do nothing to remedy our mathematical ignorance. must omit some truths that ought to be theorems. the diagnosis is wrong. we certainly have unsettled conjectures galore. we seem to know somehow just how much of Reality is the singletonfree zone. The case of mereologized arithmeticthat is. they must be structurally alike. Our philosophical understanding of the concept of a singleton is already good enough for purposes of mathematics. That means that their difference cannot affect the truth of Goldbach's conjecture. But in fact it doesn't.' This diagnosis would have been exactly right. After millenia. One reason why is that the Peano axioms characterize the successor relation up to isomorphism. 114 . Perhaps there are two rival successor relations. Perhaps there are two rival congruence relations.Set Theory for Mereologists a perennial problem. We ask whether some unsettled conjecture in set theory might be true for singleton 1 and false for singleton 2 • No matter how mystified we may be about the nature of singletons. both satisfying those axioms. mathematics . for instance. 'The trouble may be.' he said. The trouble lies elsewhere. even if it has all the Peano axioms. The distinctively arithmetical part of the system is complete. singleton l and singleton2> both satisfying the stated axioms of mereologized arithmetic. rather.'and the reason we cannot settle the parallel conjecture is that it is true for congruence 1 but false for congruence 2 . never proved and never refuted. of course. 'that the accepted axioms do not characterize the primitive geometrical notion of congruence well enough to settle the question. the Peano axioms in their full strength.
that our ordered pairs shall be Kuratowski pairs).Set Theory for Mereologists singleton functions: we may assume that they do not disagree about the demarcation between individuals and everything else. singleton land singleton 2 differ only by a permutation of singletons. Then how can we speak of ordered pairs now?. As well as the demarcation of the individuals. If P is a permutation of singletons. up to automorphism. unequivocally. or vice versa. We have only to make an arbitrary choice: as it might be. A relation of singletons is a class. A singularist has no business the in. I asked how we were entitled to speak of ordered pairs before we were given. I 1 smg eton 2 . 8 It is a permutation of singletons iff each singleton is the first term of exactly one pair and the second term of exactly one pair. the 6 it important that we reject the singularist dogma that plural quantIfication must really be singular quantification over classes. 5 (1976). unequivocally. then the rival singleton functions could have differed structutally. That IS a framework matter. and (2) whenever x and yare fusions of singletons. and to prove it. So far (but see the appendix) we can introduce pairing only by a settheoretic definition.Because now I have supposed that we are given. to be an individual2 is to have no singletons 2 as parts. and Second Order Set Theory'. what they make true concerning the size of Reality. Given the fixed demarcation of the individuals. pp. the framework also is settled. to which the interpretation of 'singleton' is irrelevant. See Thomas Weston. but at least one is isomorphic to an initial segment of the other. I say. the Hypothesis. The interpretation of the apparatus of the framework. plural quantification and mereology as well as elementary logic. of ordered pairs of singletons. the Continuum Hypothesis. Our problem is not privation but overabundance. so the domain of quantification is fixed. 7 It remains to state this properly. So they cannot differ in any way that matters to the truth of any statement in the language of mereologized arithmetic. we mark defined terms with subscripts to indicate whether they are defined in terms of singleton or . The results concerning almostcategoricity avoid such problems by confining themselves to pure set theory. P pairs the singleton t of x with the singleton 2 of x. some singleton function. but of course the subscripts must return when we speak of the singleton of something. 8 In section 2. These are discussed. So we may suppress the subscripts when we classify things as 'individuals' and 'singletons'. and in this way fallen short of is om orphism. Something that is a singleton) could also have been an individual 2 atom. over everything. They are structurally alike. 116 . and P pairs each singleton in x 7 There are known results concerning the almostcategoricity of secondorder set theory: one model is not necessarily isomorphic to another. in Weston. since singletons are exactly the atoms that are not individuals. but we cannot take this course if we want to hold the domain of quantification fixed by quantifying unrestrictedly over everything. So the two rival singleton functions might disagreed about how many individual atoms there were. 28198.6. singletont and singleton 2 differ by the permutation P iff (1) whenever x is an individual.indeed. given the fixed framework and consequently the fixed size of Reality. That means that singleton l and singleton 2 cannot fer in. It is our holding fixed of the framework that takes us from almostcategoricity to categoricity simpliciter. is fixed. that the singletons l are exactly the singletons 2 .terpretation of plural quantification fixed while regarding the Interpretation of the settheoretical primitive as unsettled. or at any rate over some sort of classlike entities. The individuals l are exactly the individuals 2 • (As illustrated here. and Second Order Set Theory'. We are quantifying. Completeness of Mereologized Arithmetic axioms of mereologized arithmetic suffice to characterize the primitive notion of singleton up to isomorphism .) It follows. 'Kreisel. once and for all. 6 Then. If we had declined to hold fixed the demarcation of individuals. Journal oj Philosophical Logic. 'Kreisel. each of the two rival singleton functions. perhaps proper. for instance. To be an individual l is to have no singletons t as parts.
then let z be the fusion of all singletons which P pairs with singletons that are parts of x. then x is part of z iff y is part of w. when P maps x to itself. P maps something to it. and x has a singleton t . that singleton 2 is 119 Proof Call a singleton good iff P pairs it with exactly one singleton. and with nothing else. hence not an individual. 118 . Case 1: If x is an individual. Contradiction. and. and that. We have two remaining cases. whenever P pairs one thing with another. it must do so either in virtue of clause (1) or in virtue of clause (2). In just the same way. It follows. Now suppose for reductio that P also pairs y with v. then y is also good. that u = z. In the second case. So z has a singleton 2 • call it w. and hence clause (2) cannot apply to x. P pairs exactly one singleton with it. and P pairs y with w by clause (2). if P maps x to y. and for each singleton in y. if P maps x to y. we have that z also is small. There are two cases in which x has a singleton 1 : when it is an individual and when it is a small class. It remains to prove that P is a permutation of singletons. How did y and v get paired? Not by clause (1): x is a fusion of singletons. let it be the singleton 2 of u. because they have no singletons as parts. and in particular that every singleton is good. Mereological operations are preserved. then P maps x's singleton 1 to x's singleton 2 . Fusions of good things are good. and z to w. P pairs y with x's singleton 2 by clause (1). together with all singletons that P pairs with singletons that are part of x. Say that P maps x to y iff y is the fusion of all individuals that are part of x. By clause (2). That is: for every singleton. It is easy to see. call something good iff every singleton that is part of it is good. The sense in which singleton t and singleton2 are structurally alike is. for each thing. We use Induction for singletonI to show that everything is good. Since v is a singleton. P permutes not only the singletons but everything else as well. This completes the inductive proof that everything is good. y also must be a small class and therefore must have a singleton2' Clause (1) says that in the first case. Case 2: If x is a small fusion of singletons. we show that. and y has a singleton 2 . in general. So. once we know that P is a permutation of singletons. and u i= z. So u is a fusion of singletons. P pairs it with exactly one singleton. QED To see better what it means that singleton 1 and singleton 2 differ by a permutation of singletons in the sense given by clauses (1) and (2). the singletons that are part of it are good. and for each singleton in u. loosely speaking. if P maps x to y. then P pairs the singleton t of x with the singleton 2 of y. Completeness of Mereologized Arithmetic then. So. Therefore P is a permutation of singletons. by that and Replacement. Given that x has a singleton l' X cannot be either a large fusion of singletons or a mixed fusion. that P maps each thing to one and only one thing.Set Theory for Mereologists with a singleton in y. by definition of z and goodness of the singletons in x. then P maps the singleton t of x (if such there be) to the singleton 2 of y. then P maps x's singleton t to y's singleton 2. vacuously. P pairs some singleton in x with it. except that we use Induction now for singleton 2 . P pairs some singleton in x with it. It remains to show that if x is good and y is x's singleton 1 . We define P to be the least relation that satisfies the closure conditions (1) and (2). if P maps x to y. and P pairs each singleton in x with a singleton in u. Individuals are good. it is helpful to note that. in general. since an individual cannot be a fusion of singletons as well. Since x is good. for every singleton. v i= w. in a derivative sense. Clause (2) says that in the second case.
1989) The foremost 'paradox of infinity' is that the part may equal the whole. That is why. . or three if we count one that is just a hybrid of the othertwo. Burgess. Later we shall extend both methods to cover the case where there is a certain amount of atomless gunk as well. Until further notice. and not too much atom less gunk. we shall assume that everything consists entirely of atoms. to simulate quantification over such relations as a permutation of everything (see the appendix). hitherto. Can we somehow get the effect of quantifying over relations? Equivalently. 9 That is why no statement in the language of mereologized arithmetic can be true for singleton t and false for singleton z. No choice between the two rivals could tell us more than we know already about what's true in set theory.Set Theory for Mereologists the image of singleton t under a mereologypreserving permutation of everything. It can be done in two different ways. and David Lewis Introduction Suppose we have the resources of plural quantification and mereology. There are ways. if we have infinitely many atoms. Infinite Reality contains infinite proper parts that are microcosms. however. This is loose speaking because there can be no such thing as a permutation of everything. Not even a proper class can have as many members as there are pairs of things. as it turns out. The Method of Double Images (Burgess. each one containing within itself images of 120 121 . can we get the effect of plural quantification over ordered 'tuples? We can.. Hazen. I shunned the noun and preferred the verb: P permutes everything. P. A. as in chapter 3. 9 Appendix on Pairing by John P. and in our irremediable lack of a complete axiom system for plural quantification. but no primitive singleton function. The sources of mathematical ignorance lie elsewhere: in our ignorance of the size of Reality.
and all things have images within the second. In this special case. 2 The Method of Double Images We cannot tell the direction of mapping just by looking at a diatom. the fusion P1 + m2 unequivocally codes the pair of Possum and Magpie. we must first explain what we mean by our talk of 'images'. 1 We follow Lewis's usage: 'distinct' means 'nonoverlapping' or 'disjoint'. call them second images. 1 All things have images within the first microcosm.) 2 Soon we shall meet strings of several plural quantifiers. Suppose that the diatoms Y1 map Xl oneone into Y1' and the diatoms Y2 map X2 oneone into Y2' Then YI from Xl to Y1> in tandem with Y2 from X2 to Y2' together map X oneone into y. So far. But we can speak. but since x and yare given along with the diatoms Y. Then if P1 is Possum's first image and m 2 is Magpie's second image. for all such variables will tum plural upon translation. in effect. The stipulation that x and yare distinct makes sure we will never be confused by atoms that are common to both. We have not just merged the diatoms Y1 and the diatoms . Suppose likewise that Y1 and Y2 are a partition of y. Suppose that Xl and X 2 are a partition of x: that is.Appendix on Pairing all the things there are. we are limited to the special case that the domain and range are distinct. rather than 'nonidentical'. we can specify a relation without recourse to ordered pairs. they are distinct. Such fusions of first and second images may serve in general as ordered pairs. or diatoms. They may even be identical. taken in that order. Figure 12 3 1 122 . each one of Y consists of an atom of x and an atom of y. wholly distinct from the first. and there exists another oneone mapping from all the atoms into the atoms of the second microcosm. Any first image is distinct from any second image. We cannot yet say directly that there exists a oneone mapping from all atoms into the atoms of the first microcosm. as shown in figure 1: Xl = Y2' X 2 = Y1' so x= y. any two of Yare distinct. Discovering how to say such things is exactly our present task. Sometimes we also use capital letters as singular variables over relations. But we can get around this limitation by taking mappings in tandem. Note that we have written our tandem mapping as an eightplace relation (with two places plural). They may overlap. Suppose we have one such microcosm. Definition: Diatoms Y map X oneone into Y iff x and yare distinct. so to avoid confusion we supplement the plural pronouns of English with capital letters used as plural variables. call them first images. as follows. To make good sense of the plan just sketched. We quantify plurally over them. it still applies to X 2 and Y2' But it does not apply to X and y. of oneone mappings between the atoms of distinct things. (See section 3. We can use unordered pairs of atoms: twoatom fusions. and x is part of the fusion of Y. and also a second.Y2 • We have retained the information that Y1 are to be taken as mapping from Xl to Y1 and Y2 are to be taken as mapping from X2 to Y2' The limitation to distinct domains applies to Xl and Y1. and their fusion is x.4. we know that we are to take the diatom as mapping the atom of x to the atom of y and not vice versa.
Now we define our ordered pairs. All first images are parts of x + Yx and all second images are parts of z + yz. (It might be derived from other. we can map the whole of Reality oneone into each of two distinct microcosms. X. N ow we can make good our talk of images. mutatis mutandis. we divide Y and y. Let Yx be those diatoms of Y that have an atom of x as part and let Yx be the fusion of those atoms of y that are part of diatoms of Yx . z. but since we can have no complete axiom system for the framework.. Figure 2 Given any such x. For any j and s. By repeated pairing. in tandem with Y z from z to yz. we have ordered 'tuples of other 125 . z. then p = p' only ifj=l' and s= s'. Z. z. Y maps x + z oneone into y. and consists entirely of atoms. X. Z) of atom u iff either the diatom u + v is one of X and u is part of y or of z. then there exist some x. X maps y + z oneone into x. Likewise.. Now X from z + y to x. X. assumed to satisfy the conditions stated in the principle of Trisection. such that x. y. ) ojj and s is the fusion of the first image ofj and the second image of s. no two different things ever have the same first or second image. or else the diatom u + v is Definition: Atom v is the first image (with respect to x. and some X. Y. one of Y and u is part of x. there is a bpair ofj and s. for the definition of second images. y. Z. We start with images of atoms. twice over. A bpair is unambiguous: if p is the bpair ofj and s. Y. Y.) The Method of Double Images map the whole of Reality oneone into x + Yx' So x + Yx contains our first microcosm. and if p' is the bpair off' and s'.. together 124 Definition: The bpair (with respect to . y. and we have not chosen any particular partial axiom system. together map the whole of Reality oneone into z + yz' So z + Yz contains our second microcosm. then the diatoms Yx map x oneone into Yx' Define Y z and yz in the same way. y. and z partition Reality. Any fusion of atoms has a unique first image and a unique second image. and Z maps x+ y oneone into z. We first affirm the following as a framework principle. The first image (with respect to . y. Trisection: If Reality is infinite. we refrain from doing so. in tandem with Yx from x to Yx. Z. the diatoms Yz map z oneone into yz. we proceed as shown in figure 2. The two microcosms are distinct. First. z. Finally. Y. All imaging is relative to a given x.. ) of any fusion of atoms is the fusion of the first images of its atoms. Likewise Z from x + y to z. and go on to images of fusions of atoms. more evident principles. Hence any first image is distinct from any second image.Appendix on Pairing By taking mappings in tandem.
But only roughly. Y. Z. Consider sentences (1 ) (2) (3) For some R: :iRk mRnFor some R: kRj fiRmFor some R: nRm The order of initial existential quantifiers does not matter. The complication is that our bpairs do not have the wherewithal for decoding built into them. So they should be.. The original (1) and (2) also are equivalent: whenever there is a relation. X. For the most part we shall leave these accom paniments tacit. X.with six extra initial quantifiers besides the plural quantifier over pairs. The parent is always the older one. The same complication will arise. y. y. We lack the builtin ordering whereby the first term of an ordered pair precedes the second. For some relation R: }Rk . Take the parentchild relation among people. that any plural quantifier over pairs will be accompanied by a string of other quantifiers. we get a string of seven universal quantifiers. whether such a thing is meant to be taken as a pair or as a triple. mostly plural. x. Roughly speaking. where p in turn is the bpair of sand t. 3 Similarly. Does this mean that. and similarly for triadic. What matters is that the translations of (1) and (3) are not equivalent. But the bpair ofs andJwith respect to x.. Z): the bpair of} and k (with respect to x. we have lost track of the direction of the pairs and the relation? Not in any sense that matters. Z such that [here follow the conditions stated in Trisection). but the sentence equivalent to the original translation. if we swap the x and z quantifiers. The Method of Extraneous Ordering (Hazen. It is to be understood. and within its scope one or more atomic formulas.. we can specify the parentchild relation by means of the unordered parentchild pairs. A quantifier over bpairs must therefore be accompanied by quantifiers over the wherewithal for decoding. Within the scope of the quantifier. quintuples. say 'there are some bpairs'. If we know which people are older than which. . we get a 126 with no occurrences of'R' except those in the displayed atomic formulas. after all. a singular quantifier over relations may be replaced by a plural quantifier over ordered 'tuples of the appropriate degree. relations. z. y.) Similarly we have ordered quadruples. for all other methods of pairing considered in this appendix. Y. we still keep track of sameness and difference in the direction of the relation. therefore. X. y. 1989) There is a second special case in which we can specify a relation without recourse to ordered pairs. our translations of (1) and (2) are equivalent. If we have a singular existential quantifier over dyadic relations.Appendix on Pairing degrees. z. In this respect they are inferior to the settheoretical ordered pairs of Wiener or Kuratowski. and t be the bpair off and p. there is also its converse. of course. Y. Y. (We must be told. X.. s. if we translate a singular universal 3 The Method of Extraneous Ordering quantifier over dyadic relations. Y.. Z) is one of R . 12 7 . tetradic. z. y. then our translation must take the cumbersome form For some x. . X. z. for some R such that each one of R is a bpair (with respect to x. though the wherewithal for decoding the pairs will look different in different cases. Instead of saying 'there is a (dyadic) relation'.. Let the btriple off. mutatis mutandis.. Z is the same thing as the bpair ofJ and 5 with respect to z. and swap the X and Z quantifiers. To be sure.
and the relation of respect for someone not younger than oneself. but when we go on to triadic relations it is much more trouble than it is worth. the younger who respects the older. just as the parentchild relation can. at safe odds of infinity to one. we can specify the relation of respect by dividing it into three subrelations. there is the relation of respect from younger to older. triatoms. and one monadic. People are one special case.not when the ordering by age has been specified in a sufficiently precise (and artificial) fashion. unlike the case of people. Likewise threeatom fusioris. each of R 7 • •• R12 by its unordered pairs. We divide it into thirteen subrelations: six triadic. (We write '0' for 'is older than'. we cannot tell whether it is the older who respects the younger. R6 can be given by its unordered triples. Or rather. the selfrespecting people themselves. it can be given by its unordered pairs.. 5 The case of a triadic relation R among people is cum bersome. without benefit of set theory. for any two of 0. In the case of atoms (or in the case of people. To what extent can we use this strategy within the framework of plural quantification and mereology. or simply the selfrespecting persons themselves. More simply. values. some of these 'pairs' will be identity pairs of selfrespecting persons. and atoms themselves serve as ituples. there is the relation of respect from older to younger.) The Method of Extraneous Ordering R1XyZ iff Rxyz & R2xyziff Rxyz & R3XyZ iff Rxyz & R4xyZ iff Rxyz & Rsxyz iff Rxyz & R6xyZ iff Rxyz & xOyOz R7XY iff Rxxy & xOy xOzOy R8Xy iff Rxxy & yOx yOxOz R9xy iff Rxyy & xOy yOzOx RlOxy iff Rxyy & yOx zOxOy RllXy iff Rxyx & xOy zOyOx R12xy iff Rxyx & yOx R13X iff Rxxx 4 What of the case that someone respects someone else of precisely the same age? We may assume. without benefit of set theory? So far. for the same reason.) Diatoms serve as unordered pairs. however. In the case of the atoms. we do not even have unordered pairs or triples of arbitrary things. (A more important case. S We could instead divide respect into just two subrelations: the relation of respect for someone not older than oneself. but not in general) we can give an ordering by giving all fusions of its initial segments. First. one of them is part of the other. it can be given just by giving its ituples. Each subrelation is given by its unordered pairs. 4 However.. Second. serve as unordered triples. The relation of respect among people is more of a problem. Also. Given an unordered pair of two people such that one respects the other. because a fusion of people divides into people in only one way. or whether the respect is mutual. some people respect themselves and some do not. Definition: 0 are nested iff. Atoms are another special case. we don't know how to specify an extraneous ordering. there is the relation of selfrespect. or ltuples. six dyadic. since more of Reality consists of atoms than consists of people. but no different in principle. Third. They are given as follows. This variation would be easier if we had to do only with dyadic relations. and it too can be given by its unordered pairs.Appendix on Pairing extraneous ordering by age serves as a substitute. Each of Rl . But we don't have to. that this never happens . It is enough to say that one exists. it can be given by its unordered identitypairs. and R13 by its ituples. for the quantified variable is plural. Then we can proceed relative to an unspecified value of the quantified variable. 128 .
then the R 7 • •• R12 quantifiers over diatoms. then the R 1 • •• R6 quantifiers over triatoms. of atoms by repeated pairing. for some diatoms R 1 . quadruples. Together. there is a 6 One might worry that we lose track of the direction of the relation.. . some one of X precedes all the others with respect to O. it would have been worthwhile to adopt the variation considered in note 5. for some atoms R3: . because once we can quantify over triadic relations. Definitions: P is a pairing relation (on atoms) iff P is a triadic relation of atoms. We might affirm. What we really affirm as a principle of the framework is the translation of the principle as written. Now we can translate a quantifier over relations of atoms. the existence of some 0 which order the atoms. Had we chosen this alternative. we can introduce ordered pairs of atoms. or j= k andj is one of R 3 . and finally the R13 quantifier over atoms. whenever there are some atoms X. . 6 For the triadic case we have a string of fourteen plural quantifiers: first comes the 0 quantifier. Then we can proceed to ordered triples. with a quantifier over triadic relations of atoms. it is enough to say that one exists. The requisite definition and principle are given below in un translated form. we translate For some relation R: . the fusion of 0 contains all atoms.7 But we do not say exactly that. or j + k is one of R2 and k precedes j with respect to 0. For the dyadic case. and for any two atoms. But we need not specify our pairing relation.Appendix on Pairing Definition: Atom x precedes atom y with respect to 0 iff there is some one of 0 that contains x but not y.. An atomic formula 'Rhjk' within the scope of the quantifier becomes a disjunction with thirteen disjuncts.. for any atoms f and s. these give back the pairing relation.either j+ k is one of Rl andj precedes k with respect to 0.. Definition: 0 wellorder the atoms iff 0 order the atoms and. as For some 0 that order the atoms. from which that follows. we say its translation. as a principle of the framework. But the reply given in note 3 applies. and we can translate all further quantifiers over relations among atoms by plural quantifiers over 'tuples. mutatis mutandis.jRk.3). for some diatoms R 2 . Our ordered pairs of atoms will themselves be atoms. rather. since it would be reversed if we reversed the ordering 0 or if we swapped the Rl and R2 quantifiers. The Method of Extraneous Ordering and similarly when the quantifier is universal. and the one that yields the second term. we could take two dyadic unpairing relations: the one that yields the fIrSt term. a version of the Axiom of Choice (different from those stated in section 3. 7 Instead of our one triadic pairing relation. We spare you the tetradic case. one precedes the others with respect to O. Definition: 0 order the atoms iff 0 are nested. Instead we affirm a stronger principle. Wellordering Principle: Some 0 wellorder the atoms..
and similarly in the triadic.. and consists entirely of atoms. to legitimize the introduction of hpairs by quantifying over pairing relations. we may pair them atom by atom.. as follows. So hpairing of fusions is unambiguous. 8 A Hybrid Method We could begin as the Method of Extraneous Ordering does. So far. Adding Gunk Since hpairs of atoms are atoms. we can recover them from their fusion. we can define the first and second images of a given thing as its images under Rl and R z respectively. thenf=f' and 5 = s'. The hpair off and 5 (with respect to P) is the p such that Rjsp. there exists a pairing relation. Given two fusions f and 5 of atoms.Appendix on Pairing unique p such that Pfsp. We thank W. all our methods so far . if Pfsp and Pf's'p. we have a pairing relation that only yields hpairs of single atoms. Quine for pointing out to us that it would serve our present purpose. Maybe the fusion of all atoms is only part of and the rest is atom less gunk. We obtain ordered 'tuples of other degrees by repeated hpairing. V. with respect to some pairing relation P: the hpair of j and k (with respect to P) is oneofR. Then we can recoverf as the fusion of the atoms which are the first terms of these hpairs. and finish as the Method of Double Images does. We need onl y use an extraneous ordering once. . so that any part of It has proper parts in turn. Thereafter we use the hpairs. With respect to any such Rl and R z. then there exist two relations Rl and R2 mapping Reality oneone into two distinct proper parts of Reality. Once we can quantify over dyadic relations of atoms. and consists entirely of atoms. and similarly when the quantifier is universal. is our old translation of For some hpairs R. Adding Gunk Now we withdraw the assumption that Reality consists entirely of atoms. We can define the hbpair off and s as the fusion of the first image off and the second image of s. for instance in higherorder arithmetic. But we may extend the definition of hpairing from atoms to fusions. We can translate a singular quantifier over relations between fusions of atoms by a plural quantifier (suitably accompanied) over such 'tuples. Definition: The hpair off and 5 (with respect to P) is the fusion of all hpairs (with respect to P) of an atom off and an atom of s. Now our new translation of For some relation R: jRk. 8 Here we borrow a device that is known in other uses. tetradic. and 5 as the fusion of the atoms which are the second terms. cases. Pairing: If Reality is infinite. If so. We go on as usual to 'tuples of higher degree and to a translation of singular quantifiers over relations. however we do can state a framework principle as follows: If Reality is mfimte.
A gunkandoneatom fusion has maximal parts of both kinds. every atomless thing is the maximal atom less part of exactly one of G. Definition: The maximal atom less part and the maximal atomic part of any given thing are. or the pair of m and the pair of A(x) and G(x). if x has any atoms. by repeated pairing. ifany. as ordered pairs of these codes. quadruples.1. and no two of G have an atom as a common part. They must all be extended. part atoms and part atomless. if A(x) is undefined. respectively. We can encode arbitary things by ordered pairs of fusions of atoms. etc. The relation is given by the gunkandoneatom fusions such that the gunk bears the relation to the atom. and the fusion of all its atoms. . if any. given G satisfying the conditions of the Adding Gunk Nottoomuchgunk Hypothesis. and by the definition of 'few'. or the pair of g and G(x). or it may be atom less gunk. if x has any atomless parts. Choose some one of our methods of pairing.) Let A(x) be the maximal atomic part of x. as well as the wherewithal for decoding the old pairs. in other words. our pairs. We have our methods of introducing ordered pairs. otherwise. m be three arbitrarily chosen atoms. by plural quantification over gunkandoneatom fusions. The fusion of all atomless things is small.. g. so that A(x) is undefined. provided there is not too much atomless gunk. the fusion of all its atomless parts. To that end. triples. m. provided there are enough atoms to go around. Now we want to say. so by Hypothesis P its parts are few. a. g. (The hypothesis follows from Hypothesis P of section 4. NOHoomuchgunk Hypothesis: There are some things G such that each one of G is the fusion of an atomless thing and exactly one atom. And. are fusions of atoms. (The accompanying quantifiers over the wherewithal for decoding the new 'tuples now include quantifiers over G. . We obtain new triples. Now we can introduce new ordered pairs. available not only for fusions of atoms but for things of all kinds. g. We translate a singular quantifier over relations between arbitrary things into a plural quantifier over the new 'tuples. and the maximal atomic part is a single atom. a. we note still another special case in which a relation may be given by unordered pairs: the case of a relation that holds between atomless things and atoms. and hold it fixed in what follows. within the atomic part of Reality. as above. Definition: The new pair off and s is the pair of the code off and the code of s. or it may be mixed. Let a. we have the desired G. Something x may consist entirely of atoms.Appendix on Pairing apply only to the atomic part of Reality. That will be so. m) is the pair of a and A(x). if G(x) is undefined. Definition: The code of x (with respect to G. used as markers to distinguish the three cases. triples. these parts are all and only the atomless things. any fusion decomposes uniquely into its gunk and its atoms.. In general.) Henceforth we may quantify over relations. so that G(x) is undefined. let G(x) be the atom such that its fusion with the maximal atomless part of x is one of G. etc. freely . that there is a oneone correspondence whereby every atomless thing has a representative atom. Whichever method we choose.
Taken together. Condition P: Whenever something is a small part of x. 9 From megethos 'size' + logos 'doctrine'. and whenever y and z are among N. they come rise a definition of inaccessible SIze. every part of one of N is itself one of N.Appendix on Pairing and without comment. Standard set theory requires that Reality be of inacessible size. of the framework turn out to include quantification over relations. Condition I: Some fusion of atoms of x is infinite and yet is a small part of x. it implies that there is a set of inaccessible size. U. that it is larger stilL Megethology9 Sections 3. This is the framework version of a modest 'large cardinal' axiom. their fusion is a small part of x. hence that there exists an inaccessible cardinal. Given mereologized arithmetic. Definitions: Yare many parts of x iff Yare parts of x. 137 . and when it is. Sometimes thisqualitative distinction is all we want from a measure. then the fusion of y and z is one ofN. Definition: X and Yare equinumerous iff there exists a relation R such that each one of X bears R to exactly one of Y. Hypothesis IC: Some small part of Reality is of inaccessible size. Yare few parts of x iff Yare parts of x but not many parts of x. and I of section 4. 'negligible' regions. for instance the familiar Lebesgue measure of area. (Something infmite is called uncountable iff it satisfies Condition I. and I. we can write down three conditions on an arbitrary thing x. We can also state a less modest hypothesis. a small part of x iff its atoms are few parts of x. Suppose we have a measure. we can conflate the measure itself with the distinction between negligible regions and others. But we may speculate. Definition: N measure x iff the fusion of N is x. the job can be done over in an easier and more familiar way. x is not one of N. Corresponding to Hypotheses p. otherwise countable. as we do. Measures in this purely qualitative sense can be characterized in the framework. and the atoms of x and some of Yare equinumerous. its parts are few parts of x.1 used the resources of the framework to express distinctions and hypotheses concerning the size of Reality. we need not bother actually to translate them. U. and for each one of Y. exactl y one of X bears R to it.7 and 4. We can then distinguish regions of measure zero. if we wish. as follows.) Definition: x is of inaccessible size iff x satisfies Conditions P. Megetholo£y Condition U: Whenever some things are few parts of x and small parts of x. Now that the resource:. how such quantifiers may be translated. Definitions: Something is a large part of x iff its atoms are many parts of x. from others. Knowing.1.
singleton . Ramsifying out the Singleton Function As well as asking what is the largest size Reality has to offer. this hypothesis implies that there exists a measurable cardinal.{! When N measure x. Ramsifying out the Singleton Function. whenever y and z partition x. 10 Are larger things ever measurable? Not unless they are very much larger. Whenever N measure x but fail to satisfy maximality. section 9. We could Ramsify it out. and Yare some of the parts of x. Hypothesis MC: Some small part of Reality is measurable but uncountable. and add an extra axiom saying that something has no singletons as parts.2. is the special case in which we stipulate that x is countable. Given mereologized arithmetic. we may return to the unfinished business of section 2. so full additively adds nothing to the final clause in the original definition of measure. singleton .1 O. as it turns out. whenever Yare few parts of x and Yare among N. the finite parts of x measure x. 12 Conjoin them into a sentence saying that the membersingleton relation satisfies certain structural (and perhaps other) conditions. see Shoen field. then either Y and all the parts of x are equinumerous or else Yand some of the atoms of x are equinumerous. N may also meet further conditions. either y or z is one of N. countable or not. this is shown by an argument requiring a wellordering of the parts of x.. by 'thing that is a singleton of something'.. 11 For a settheoretical discussion of measurable cardinals. for in that case 'few' means 'finite'. Definition: N are maximal on x iff. for some measure N on x.H 10 If x is an infinite fusion of atoms. with 'singleton' as their only settheoretical primitive. Or.. we note that when x is countable.. An uncountable measurable thing would have to be of more than merely inaccessible size. A countable fusion of atoms is measurable. Take some axioms.. or else it requires additional principles of the framework. Mathematical Logic. 12 To put them in primitive notation. then uniformly replace 'singleton'. is usually taken as a hypothesis about the sets. then the fusion of Y is one of N.Appendix on Pairin. .6. we considered a way to get rid of the primitive membersingleton relation. The Generalized Continuum Hypothesis. But we may speculate. if we wish. uniformly replace 'null set' by 'fusion of all things that have no singletons as parts'. stronger still.. Generalized Continuum Hypothesis: If x is an infinite fusion of atoms. which addresses this question. ungeneralized. secure in our knowledge of how to translate these quantifications using only the resources of the framework. Defintion: N are fully additive on x iff.. when used nonrelationally. then N may be extended to N+ which also measure x and do satisfy maximality. singleton . The Continuum Hypothesis. N are both fully additive and maximal on x. Definition: x is measurable iff. we can ask also about intermediate sizes. But we can state it in the framework. (It can be carried out when we have set theory. perhaps the axioms of mereologized arithmetic in section 4. . any measure on x must be fully additive.) Finally.. Continued Now that we may quantify over relations. Adapting a suggestion of Paul Fitzgerald. that Reality is measurable.
. which says just that some relation satisfies those conditionsP Where there is one suitable relation..) If the further sentence followed from the axioms. there will be many. with nothing to distinguish the one real membersingleton relation trom the horde of pretenders. but would have taken membership as primitive in the standard way.Appendix on Pairing We may replace it by (a translation of) its Ramsey sentence For some S: . plus principles and hypotheses of the framework . S .. This reconstrual is a painless reform. is the one we had in mind all along.. S . S .that is. To retreat to settheoretical structuralism 14 is to think that the suitable relations are all on a par. S . The axioms to be Ramsified would not have been those of mereologized arithmetic... an 'abstract structure' common to all those relations. prmClpies and hypotheses. No previously accepted theorem must be denied.) The philosophical reward of structuralism is that it 140 . In that case. (Compare first order logic: if T(s) foHows from A (5) and F l' F2' .also the generalization fol1ows from those auxiliar. then (We might wish to conjoin the Ramsey sentence to the generalization.. Only our understanding of it needs revision. . (As many as there are permutations of the singletons. no previously accepted method of proof must be renounced.. then the generalization For an x: if A (x) then T(x) foHows from the Fs alone. and it is a reform that serves little purpose if we really did have a primitive understanding of the membersingleton relation..) To claim a primitive understanding of the membersingleton relation is to think that one of all these relations. to prevent vacuous truth in case no suitable relation exists. Mathematics may go on just as before. we would have needed only the first part of the Method of Extraneous Ordering. ..... may replace it by (a translation of) an explicit generalIZatIon. :n Ramsifying out the Singleton Function For all S: if. S . we could have Ramsifled membership as soon as we were in a position to quantify over dyadic relations of atoms. most likely the fault is his. not because it posits a new sort of an entity. But it would have meant abandoning the thesis that a class has its subclasses as parts. 13 Had we had been content to take membership generally as a relation between atoms. all theorems of the whole of settheoretical mathematics . • and the constant 5 does not occur in the Fs. e.as the corresponding generalizations. (And yet structuralism is a reform. ..) Structuralism proposes to reconstrue all theorems of set theory . This would have come closer to Fitzgerald's original suggestion. 14 So called because it takes settheoretical mathematics to consist of generalizations over all the many structurally similar membersingleton relations. A modest philosopher might well hesitate to rebuke the old mathematics. He might think that if he cannot understand how it was entitled to its primitive membersingleton relation.. and so it is a rebuke to the old mathematics that was content to take membership as primitive. and only one. not to mention our usual supposition that nonatomic individuals are sometimes members of classes. S . any further sentence about 'the' membersingleton relation singletonsingletonsingletonshould be understood as tacitly general: it says that something holds for all relations that satisfy the given conditions.
The rest are mysterious. of the kinds we know and name by causal interaction. In fact. You might complain that there's no hope of reducing the membersingleton relation to any familiar properties and relations. In fact the Ramsey sentence is worse by one initial quantifier . We might have been attracted to the speculation that singletons are where their 142 Ramsifying out the Singleton Function members are. on pain of rebelling against established mathematics. structuralism makes them moot. it makes matters worse. For anything that can be a member. Shall it share Possum's location and character. set theory still requires the hypothesis that there are inaccessibly many atoms.and all the more if you question whether quantification over un constructed mathematical objects is meaningful. Whether the singleton function is primitive or whether it is Ramsified. If you simply complain that the familiar axioms of set theory are not a selfevident foundation for mathematics. because those familiar properties and relations cannot distinguish inaccessibly many atoms. If all suitable relations alike are membersingleton relations. under translation. We know nothing about their whereabouts. It does not put paid to all philosophical complaints that might be brought against settheoretical mathematics. and perhaps even share the qualitative character of their members. or lack of it.) But structuralism makes nonsense of this speculation. then a singleton does not have its member once and for all. for anything that can be a singleton. or shall it share Magpie's? If you complain (as Hazen does) about the nonconstructiveness of settheoretical mathematics. (We might have hoped to answer. again structuralism doesn't help. all the suitable relations are equally 'the' membersingleton relation. We have to believe in these mysterious extra atoms. But structuralism is no panacea.you should not like the Ramsey sentence of mereologized arithmetic any better than you liked the original axioms. only without collecting many into one. Only very few of them can be ordinary atoms. You might complain that all you were told amounts only to this: it's just like collecting many into one. And you have a second reason to challenge the positing of inaccessibly many atoms. somehow. there will be some membersingleton relation that pairs the one with the other. The atom that is Possum's singleton under one suitable relation is Magpie's singleton under another. by one long string of quantifiers. We know nothing about their intrinsic qualitative character. and the hypothesis that there are inaccessibly many atoms is downright surprising. If you will not affirm an existential statement until you see a confirming instance . there is some room 143 .or rather. You should like the Axiom of Choice no better as a principle of the framework than as an axiom of set theory.Appendix on Pairing bypasses all doubt about whether the primitive membersingleton relation is well enough understood. again structuralism doesn't help. All it does is redirect your complaint from set theory to plural quantification and mereology. You might complain that no known theory of intentionality explains how you can have in mind just one out of all the suitable relations. It says there is no primitive notion that needs understanding.1) that your introductory lesson in set theory just does not apply to the case of membership in a singleton. quite apart from any qualms about their unknown location and character. In the case of the Ramsey sentence. we needn't try to wrap our minds around some special one of them. structuralism doesn't help. Whatever the merit of these complaints. or lack of it. If you complain not about the 'ideology' but about the ontology of set theory. You might complain (as Lewis does in section 2. The Ramsey sentence of merealogized arithmetic is far from selfevident. the question how atoms can share the location or character of nonatoms. Structuralism does nothing to relieve the burden.
weakened axioms are not· equivalent to the old axioms. that question makes no sense. Non Welljounded Sets (Center for the Study of Language and Information. see Jane English. as structuralism claims. pace the invocation of 'our offhand reluctance to believe in them' in section 1. it is the edifice that justifies the foundation or nothing does. Journal of Philosophy. Proof Let singleton l be a relation that satisfies Functionality. then the new axioms are interchangeable with the old. And if all that we really assert IS the Ramsey sentence. which says that nothing else has a singleton. (Scattered exceptions only: almost all classes must lack singletons. But if we are structuralists. Equivalence under Ramsification Domain +: Any part of the null set has a singleton. but something is gained. Equivalence under Ramsijication Structuralism also brings a mathematical reward. any small fusion of singletons has a singleton. 144 145 . Even so. pp. therefore almost all large classes must lack singletons. For better or worse. (In section 4. 1988). and think we did not have anyone relation in mind. whether it satisfies the original axioms of mereologized arithmetic as well as the weakened axioms. it does not follow that the same relation satisfies the old ones. 45362. Nothing is lost. We can weaken the axioms without loss. Dropping Induction means leaving it open that there might be nonwellfounded sets. Stanford University. (The converse is trivial. Atomicity was derived using Induction. Y if some relation satisfies the new axioms. but perhaps not the negative clause of Domain and perhaps not Induction.2. there is little hope for a selfevident foundation. else there would not be enough singletons to go around. we shall see how to regain it from a seemingly weakened version. the old and new Ramse: sentences are equivalent. for improvement. If we think we have a primitive understanding of the membersingleton relation. the new. Distinctness. (So it does not quite leave mathematics unchanged. 70 (1973). any singleton has a singleton.Appendix on Pairing . Domain +. it is a substantive question whether the relation we have in mind really does allow any of these things to happen. If a relation satisfies the new axioms. 'Underdetermination: Craig and Ramsey'.) If so. it may follow that some relation satisfies the old ones. and put in its place the much weaker Atomicity: Every singleton is an atom.) It lets us weaken the axioms of mereologized arithmetic. almost all classes are large. There is no substantive question which are right. in the next section. and Atomicity. involved in infinite descents of membership}6 Dropping the negative clause in the axiom of Domain means leaving it open that there might be scattered exceptions to the principle of Limitation of Size: occasional cases in which a large fusion of singletons has a singleton.3. One is to drop the axiom of Induction.) Dropping the negative clause also means leaving it open that there might be singletons of mixed fusions of individuals and classes. Without Ramsification. Then there is a relation singleton3 that satisfies all the axioms 16 See Peter Aczel. given that the old axioms imply the new ones. 15 Two simultaneous weakenings (at least) are possible. and drop the negative final clause.) The other is to retain the positive part of the axiom of Domain 15 For a general discussion of equivalence under Ramsification.
QED A second use of equivalence under Ramsification is of metaphysical rather than mathematical interest. If you think this via negativa is an essential ingredient in our conception of'singletons. 'platonic form'.indeed.. 'spacetime'. so they are equinumerous with all the singletons t . 147 . 17 Further. many .. and hence that many atoms are not. There may be little positive to say.1 and 2. and the resulting Ramsey sentence will go beyond the vocabulary of the framework. that Reality is of inaccessible size. such terms as 'material'. Then the XAxiom is redundant under Ramsification. and its singleton t would be required. Let singleton 2 be the restriction of singleton t to the required singletons t and the individuals t . as usual. . (We subscript defined terms of mereologized arithmetic to indicate which singleton relation they correspond to. then it is one of them. then we can apply the result of section 4.6. or part of any platonic form. and every singleton t of a small fusion of them is one of them. You might think that the 'structural' axioms of logized arithmetic are not really enough to characterize a suitable membersingleton relation. So all the required singletons t are many. or part of any spirit (either temporal or eternal). The resulting conjunction of axioms will be a strengthening of the axioms of mereologized arithmetic.) Call a singleton t required iff. It is large. if we strengthen the axioms by giving a necessary and sufficient condition for an atom to be an individual. so it would be a selfmember t iffit were not. whenever there are some atoms such that every singleton t of an individual t is one of them. We get equivalent Ramsey sentences either way. as an auxiliary hypothesis.. otherwise it is a singleton. Atomicity ensures that singletons t are indeed atoms. It is not a substantive issue whether to take it or leave it. will remain. you 14 6 Equivalence under Ramsification could insist on incorporating it into the axioms.8: two membersingleton relations that both satisfy the strengthened axioms differ only by a permutation of singletons. Domain. the strengthening may be iIlusory. Recall the discussion of 'unofficial axioms' of set theory in sections 2. It is easily shown that singleton 3 satisfies Functionality. such a specification would go beyond classifying the atoms. and every singleton t of one of them is one of them. Take the class of nonselfmembers t whose singletons t are required.Appendix on Pairing of mereologized arithmetic. or .. and would advance the daring speculation that some . Distinctness. (That means it would be foolhardy to specify X as 'the atoms located in spacetime' or 'the atoms having some qualitative character'. This sort of Ramsey sentence says that there is a relation that is suitable not only mathematically but also metaphysicallyP But if structuralism is right.) Assume..atoms lack location or character. And suppose we are prepared to affirm. but at least there is a via negativa: you can insist that no singleton is part of any material thing. so there is a relation that maps the required singletons t oneone onto all the singletons t • Let us extend it to a relation M that also maps all individuals t onto themselves. Suppose for definiteness that we contemplate adding one 'unofficial' axiom of the form XAxiom: An atom is an individual atom iff it is one of the atoms X [somehow specified]. that only few atoms are among X. You might want a metaphysical characterization as well. and Induction. else it would have a singleton t by Domain +. and let singleton 3 be the image of singleton 2 under M. 'spirit'. Although the primitive notion 'singleton' wiIl be Ramsified out. or part of spacetime.
There is a rank3' call it R. The Xaxiom tells us that n ought to end up as the null set. be the image of singleton 6 under this M. It is easily shown that singleton. atoms onto X. Let m be the fusion of the atoms Z and whatever gunk there may be. 18 As the subscripts suggest. Yare equinumerous with the parts of n. It does not satisfy the XAxiom (unless by luck). So we have a relation that maps the atoms above Y together with the atoms Z oneone onto all atoms. We have assumed that n is small. and let us extend it to a relation M that also maps all atomless things onto themselves. Let singletons be the image of singleton. under N. Let us extend it to a relation N that also maps all atom less things onto themselves. and every singleton of a small fusion of them is one of them. Let the relation singleton6 be the union of singleton 4 and singletons. and (2) for some atoms Z below rank R. 1s Let n be the fusion of the atoms X and whatever atom less gunk there may be. and every singleton of one of them is one of them. whenever there are some atoms and every atom of Y is one of them. There is a relation singleton 4 that maps the parts of m oneone onto Y. It is easily shown that singletons satisfies all the axioms of mereologized arithmetic. big enough so that (1) for some atoms Yat rank3 R. Say that an atom is above Y iff. the atoms above Y together with the atoms Z are likewise many.Appendix on Pairing Proof Let singleton 3 be a relation that satisfies all the axioms of mereologized arithmetic. The atoms above Yare many. satisfies all the axioms of mereologized arithmetic. this proof may be joined to the previous one to show that the weakening of mereologized arithmetic just considered is equivalent under Ramsification (given the specified hypotheses) to the strengthening of mereologized arithmetic which adds an XAxiom. Equivalence under Ramsijication however the individual. Then there is a relation singletons that satisfies the axioms of mereologized arithmetic and the Xaxiom as well. we may use set theory freely. atoms are equinumerous with the atoms X that ought to be individual atoms. but perhaps not the XAxiom. QED 14 8 149· . Since we have all of mereologized arithmetic and the hypothesis that Reality is of inaccessible size. Z are equinumerous with X. So there is a permutation of atoms that maps the individual. Let the relation singletons be the restriction of singleton 3 to atoms above Y. Let singleton. and the XAxiom as well. then it is one of them.
623. 147 151 . for mereology. ix. 21. Axiom of. Peter. 147 Choice. for mereologized arithmetic.10713. 11320. 63 Boolos. ix Bigelow. ix. 55 Black. 11315 Armstrong. 51. 314. 1467 Baxter. 107. 112.73 Burgess.46. Ignacio. 144 atoms. for set theory.139. 1423. Harry c. 145 Angelelli. ix.27. definitions of. for arithmetic.11516.49. M. 48.823.S arithmetic.109. 70 Bricker. for ordinal arithmetic.105 cardinals. 74. 1045. 1012 awesome 'classes'. George.61. 956. 1007. vii. 113. 63. 712.S. 130.5.76. 19.John P. 567. 1034..86 Atomicity. 52. unofficial.. Donald. 16. ix Bunt..65. 668 axioms. vii. 334. 62. size Carnap. 10. John. 47 ' categoricity. 53. 1217 Campbell. 238 completeness. ix. 335. Max.534. 2931 Aczel. 824 Benacerraf. 102.Georg. 66. Paul. Keith. Phillip. viiviii. Axiom of. 147 character of singletons. 16. definition of. 143 class. 116 coding of pseudomembers. 46. 57. see large cardinals. Rudolf. viii.29. 11320.97 classlike things. 48. Axiom of.Index acts of setforming. 4. 15 Aussonderung. 76 Cantor. D.
definitions of.l44 equivalence under Ramsification. 70 many and few. Richard. Rolf. 70.201. 10.113. 2930 Hazen. 389. 23. 78. 901.100. 1 inaccessible size.45. 114 Goodman. unmereological. P.134.115. 3841 geometry. Paul.7981. 945. 46. 1217. 9. see also Induction fusion. ix. Adam.20. ix.1067 Infinity.20.1336 haecceities. 61 Johnston. 1367. Sherlock. 77 Cresswell.1819. 8. 634. 140 Forrest. 56. 17.54. Axiom of. 9. 45. 8.50. vii. 11415 Goldbach's conjecture. 63. 1217 mixed fusions.112 Division Thesis. 218 nonselfmembers. Richard M.99 Generalized Continuum Hypothesis. 96. Nancy. 143 hierarchy of classes. Stanislaw. 425 Induction. 73. 935. P. 1213.16. 889. 1446 in finitary sentences. 547 nice parts. 31.31. Hartry. 14. 17. 1079.100 Malezieu. 65. 3. 701 infinite and finite. 112. 967. 139 generating relations. 30 Jekyll and Hyde. 4950. Henry. 13. 73 Fusion Thesis. 74. 12. 378. 56 framework.55. 94. 12733. M. 934. 95. 69. 7.9. 789. 28. ix Extensionality. as parts of classes. 3841. 1012.278. 667 nominalistic set theory. 57 Halmos.112. 76 gunk. 4.4850.133 identity. 55. 79.89. 56.3841. 63 mathematics as set theory. definitions of. ix. 77.98. Peter. see also unofficial axioms of set theory microcosms. David. 2. 98. 6. 107. 74.21. 425 Leonard. and U.Jane.8.122. 9.136 Field. definition of. definitions of. Axiom of.35 Holmes. 100 external relations.145 Morton. 1369 membership. 227 measurable cardinals. ix. 95. 74. 7. 345.141. Azriel. 2930 Kuratowski. 1079.139 Null Set.141 constructiveness. 434. composition as.45.110. 16. 1378 megethology. Paul R.34. uniqueness of. 817 images. viii. 547.534. 6. definitions of. Axiom of. 112 connected parts. 46.58. definitions of. 6.. 109. see null set English. 1217. unrestricted.143 Main Thesis. 143. see also Peano axioms difference. Kazimierz.. Axiom of. Axiom of. 143 Continuum Hypothesis.21.80. 100 Davoren. 11. 27.39.27. 15. 1449 Etchemendy.956. 11314 God. 347 Isaacson. Mark. 1819. 73 Lesniewski.42.136 Martin. and passim metaphysics. definition of. 16. see mereology. see relations Fundierung.Jennifer.53. 142. 945. 12. 103.658 null individual. David. 94. 81. 723 Levy. 2931 mereologized arithmetic. 15. 121. A.88. 112 functions. Stephen c. 73 empty set.4. 84. ix Kleene. ix decomposition. 18.100.99 Domain. 53. 95. 1378 large and small. 5.75. 78. 102. 6. 27 nominalism.1517.15.76 Godel's incompleteness theorem. 79. 110 Hume..412. 133 improper parts.93. j. 17. ix.97..32. mereological. 8. Axiom of.74. Nicolas de. 910. 889. 1047. 2938. 97 mental acts of setforming. 70 mystery about sets. 85. 589. 117.1489 Index Magpie. 98 Fitzgerald. plural quantification Functionality. 4. 1067 internal relations.Index composition.6.1434 maximal connected parts.889.85. 1045 Lewis. 10 Massey. Gerald. 223. 15.87.267. Nelson. Axiom of. 136 Lasso Hypothesis.97. 22. 12733 . 378 extraneous orderings. 21. 49. 21.79. ix facts. 901. 50.56.85. 120. Axiom of. 107. 58. 18.101. 227 conservatism. Daniel. 58 finite size.57. 567 few and many. 126 large cardinals. 139. 958 mereology. 967. 1015. 17 Distinctness. 111 limitation of size. 26.36. 107 First Thesis. 817.43..1446 double images. 19. 145 Eberle. 20. 1213 null set. 24 Dedekind. 112. uniqueness or multiplicity of.32. 323. 8991. viii. 139 counterparts. 33. 3.7287. 17 individuals. 70 Hypotheses I. as identity. 1478 inclusion.
35. Frank P. 557. 117 Putnam. definition of.7981. 11620 philosophers. N. 86.11415. 73 Taylor. 56. 111.86 unofficial axioms of set theory.857.42. 14. Denis. 889.52. theories of.. 389.668. 22. 335. 7. 47 Ramsey sentences. 143.. 78. 63 structuralism. ix. 59 plural quantification..110. 50. 15.98 van Fraassen. 21.11516 Unrestricted Composition. 116. 62. Richard. 76. 51. 51. 76. 120. 113 Wellordering Principle. 73 Priority Thesis.1067.120. generating. quantifying over. 142. 117 Wiener. 15. ix time. john.47 tandem. 28. Graham. 73 singleton. identified with its member. 50. 15. Stephen.54. 18 Unions. 14. 98 Sharvy. 45.89. 15. 610.112 urelements. 334.109. 778.82. 98. map of. Peter. follies of.756.98. 412. Peter.86 Possum.11213.645 successor relation. 84. 358. 6. 66 Robbin.138 Simons. 1921 relations.789. as abstract representations. 104.701. Michael. 74 Trisection.85. 112. 9.145. 12149.115. 91. 46. 63 possibilities. 1313. 56. 4654. 63.74. 13. 1235 Tarski. 117. 35. 8991. limitation of. Eric. 2 properties. 523. 33. Alfred. 13949 substitutional 'quantification'. ix Russell. ix orderings. see spacetime transitivity. 9. 16.114.85. 11011 154 .116. Bertrand. 15 5 . 8.104. see also relations pseudomembers and pseudoclasses. 67.935.55.37. Hilary. 16.114 Robinson.147 singularism.57. 48. 132 Ramsey.57.120 proper parts. 1819. LeSniewski's. 334. 63 Shoenfield.47 van Inwagen. ix qualitative character of singletons.98.65. 20. see singleton unithood facts. 56 sets. Axiom of. 8. 1213. 90. 115 Peano axioms..114 supervaluations. 659. 71.145. 501. 1367. 12. inaccessible. 2. 512.109. Willard V.457.77. 11213 overlap. 101 small and large.54.1478. 58. 116. 94. 22. and passim. 6271.1012. 116 size. large and small. internal and external. 136 spacetime. finite and infinite.556. 70. 10912. Principle of. 57. 32.7 universals. Norbert. 11314. 4554. 1001 parallel postulate. 1.31. 76 Zermelo. 218 pure classes.5. 81 von Neumann. 63. mappings in. of reality. ix. 18. 2930. . 13949 rank.49.99 proper classes. 49.95.7 Stenius.2 Replacement.61. 556. 79.93.1446 permutations of singletons.joel W. as worlds.967. 1045 Uniqueness of Composition.87.85. viii. 1819. 19. 2930. 689. 105 Prior. vii. Principle of.36.13942.1369 Skolem. Ernst.52. 76 unicle relation.1423. and passim Pollard. Index 945. 10713. 124 tropes. 8. 126 Williams. 78. A. 14. 11 011. 412. 3841.723.147 Quine. 13. Donald c. 956. 4. see also singleton union. 14. 66.100 unit class.323. 130 Weston. 12733 ordinals. 93.joseph R.74.20.98. 16.111.115. 122.122. 314.. 136.143 Power Sets. 8.46. 334. 1257. extraneous. 12.47. 27. 148 Reality. definitions of. 79. Thomas. definitions of. 102. 817.. 368. existence of.39.1423.46. Axiom of. 93. 5.15. 8992.102 Ontology. 358. 135 Pair Sets. Barry. 4.412.112. Axiom of.101. 61. 102 Resnik. as primitive. 51. 4. 56. see also nonselfmembers secondorder systems. 1079. Bas.147 states of affairs.3. 21.74. definition of. 63. 4950. viiix. 73 Pairing.37. viiiix. ix.117 Second Thesis. Thoralf.315. 534. 8.Index ontological innocence.49. 48. 723 Oppy. 132 pairs.
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 reading from where you left off, or restart the preview.