You are on page 1of 184

Gerhard Preyer, Georg Peter (Eds.

)
Philosophy of Mathematics

LOGOS
Studien zur Logik, Sprachphilosophie und Metaphysik

Herausgegeben von / Edited by


Volker Halbach x Alexander Hieke
Hannes Leitgeb x Holger Sturm
Band 13 / Volume 13

Gerhard Preyer, Georg Peter (Eds.)

Philosophy of Mathematics
Set Theory, Measuring Theories,
and Nominalism

Bibliographic information published by Die Deutsche Bibliothek


Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliographie;
detailed bibliographic data is available in the Internet at http://dnb.ddb.de
North and South America by
Transaction Books
Rutgers University
Piscataway, NJ 08854-8042
trans@transactionpub.com

United Kingdom, Ireland, Iceland, Turkey, Malta, Portugal by


Gazelle Books Services Limited
White Cross Mills
Hightown
LANCASTER, LA1 4XS
sales@gazellebooks.co.uk

Livraison pour la France et la Belgique:


Librairie Philosophique J.Vrin
6, place de la Sorbonne ; F-75005 PARIS
Tel. +33 (0)1 43 54 03 47 ; Fax +33 (0)1 43 54 48 18
www.vrin.fr

2008 ontos verlag


P.O. Box 15 41, D-63133 Heusenstamm
www.ontosverlag.com
ISBN: 978-3-86838-009-5
No part of this book may be reproduced, stored in retrieval systems or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming, recording or
otherwise without written permission from the Publisher, with the exception of any material supplied
specifically for the purpose of being entered and executed on a computer system, for exclusive use of
the purchaser of the work
Printed on acid-free paper
ISO-Norm 970-6
FSC-certified (Forest Stewardship Council)
This hardcover binding meets the International Library standard
Printed in Germany
by buch bcher dd ag

Contents

Contents
Preface
Gerhard Preyer, Georg Peter ...............................................................

Part I
Set Theory, Inconsistency, and Measuring Theories
Douglas Patterson
Representationalism and Set-Theoretic Paradox ..................................

11

Mark Colyvan
Whos Afraid of Inconsistent Mathematics? ........................................ 28
Andrew Arana

Logical and Semantic Puritiy .............................................................. 40


Wilhelm K. Essler
On Using Measuring Numbers according to Measuring Theories ...... 53

Part II
The Challenge of Nominalism
Jody Azzouni
The Compulsion to Believe: Logical Inference and Normativity ......... 73
Otvio Bueno
Nominalism and Mathematical Intuition ........................................... 93
Yvonne Raley
Jobless Objects: Mathematical Posits in Crisis .................................... 112
Susan Vineberg
Is Indispensability Still a Problem for Fictionalism? ............................ 132

Contents

Part III
Historical Background
Madeline Muntersbjorn
Mill, Frege and the Unity of Mathematics ......................................... 147
Raaella De Rosa and Otvio Bueno
Descartes on Mathematical Essences .................................................. 164
Editors and Contributors .................................................................. 183

Preface

Preface
One main interest of philosophy is to become clear about the assumptions,
premisses and inconsistencies of our thoughts and theories. And even for a formal language like mathematics it is controversial if consistency is achievable or
necessary like the articles in the rst part of the publication show. Also the role
of formal derivations, the role of the concept of apriority, and the intuitions
of mathematical principles and properties need to be discussed. The second
part is a contribution on nominalistic and platonistic views in mathematics,
like the indispensability argument of W. v. O. Quine H. Putnam and the
makes no dierence argument of A. Baker. Not only in retrospect, the third
part shows the problems of Mill, Freges and the unity of mathematics and
Descartess contradictional conception of mathematical essences.
Together, these articles give us a hint into the relationship between mathematics and world, that is, one of the central problems in philosophy of mathematics and philosophy of science.
This book was planed by the project Protosociology at the Goethe-University
Frankfurt am Main, Germany. The digital version is published by Protosociology. An International Journal of Interdisciplinary Research, Vol. 25 (2008). We
would like to thank our contributors for their support in realizing this publication. The book represents a continuation of our research in analytic philosophy
and semantics in the journal and of the project.
Gerhard Preyer and Georg Peter,
Frankfurt am Main, Germany

Preface

Representationalism and Set-Theoretic Paradox

Part I
Set Theory, Inconsistency, and
Measuring Theories

10

Douglas Patterson

Representationalism and Set-Theoretic Paradox

Representationalism and Set-Theoretic


Paradox
Douglas Patterson
Abstract
I defend the settist view that set theory can be done consistently without any form of
distinction between sets and classes (by whatever name), if we think clearly about belief
and the expression of beliefand this, furthermore, entirely within classical logic. Standard
arguments against settism in classical logic are seen to fail because they assume, falsely, that
expressing commitment to a set theory is something that must be done in a meaningful
language, the semantics of which requires, on pain of Russellian paradox, a more powerful
set theory. I explore the consequences of this response to the standard argument against
classical logic settism for our notion of belief, and argue that what is revealed is that
representationalist theories of belief cannot be right as long as it is possible to believe that
every set is self-identical.

For a number of years now I have been developing an account of the semantic
paradoxes along the following lines (2007a, b, forthcoming). Consider a sentence like Liar, which reads Liar is not true. Apparent truths about meaninge.g. that Liar refers to Liar is not true, that is not true applies to true
sentences, that meaningful declarative sentences are true when what they say is
the casejointly imply a contradiction in the presence of sentences like Liar.
My view is that what this shows is that the semantics of natural language that
speakers of natural language are inclined to believe is simply logically false. The
centerpiece of the strategy is an argument that communication requires only
that speakers cognize the same semantic theory, rather than that they cognize
a true one, so that we can account for everything simply by allowing that the
paradoxes show that the cognized theory is in fact untrue.
As long as I have been at this I have had it in mind that it would be interesting
to consider whether a view of my sort could be put to work in addressing the set
theoretic paradoxes and related problems about unrestricted quantication
interesting because there are signicant enough dierences that the account
of the semantic paradoxes cannot simply be transferred over without modication. Two major dierences are these. First, the account of the semantic
paradoxes rests on a thoroughgoing non-factualism about meaning in natural
language, while I have no interest in being a non-factualist about sets. Second,
the account of the semantic paradoxes works by explaining how beliefs about

12

Douglas Patterson

meaning that seemed paradoxical are in fact merely false, while the set theoretic
paradoxes, for reasons to be explained below, threaten more deeply to show
that certain beliefs we appear to have are in fact paradoxical.
The current paper represents my rst attempt to work out an account of
the set theoretic paradoxes. The main discovery is that the culprit in the usual
conundrums about discourse about sets is what I will call representationalism
abut belief: the view that belief is a relation to a representation of some sorta
proposition, or a sentence in a natural or mental languagethat itself stands
in semantic relations. (I take Field 1978 to be a classic statement of the view I
have in mind.) Set representationalism aside, and the paradoxes of set theory
can be unknotted without residue, the result being, in Booloss terms (1998, 36),
a pure settist view, set entirely within classical logic, with no need to appeal
to set-class distinctions, higher order quantication or anything else. In what
follows I will call the view I hope to defend classical logic settism (CLS).
1. Consider three familiar questions about how consistently to describe the
universe of sets. The rst concerns nave abstraction, the principle that every
predicate determines a set of just those things of which it is true; schematically,
that the instances of:
(y)(x) (x y )
with y not free in what is substituted for , are all true. It can be hard to
see what could be wrong with the principle when one considers a chain of
inferences like this is red, so it is one of the red things, so it is a member of
the set of red things. Yet Russells paradox appears to show that we need only
consider the instance substituting x x for to show that something must
go wrong somewhere.
The second question, made famous by F. P. Ramsey, concerns the relation
between the set theoretic paradoxesthe Burali-Forti and so on as well as
the Russelland the semantic paradoxes such as the Liar and the Grelling.
There is a question here because the play with schemataall of the instances
are trueand claims such as every predicate determines a set have a semantic look to them. This motivates thinking that somehow the set theoretic paradoxes are really other manifestations of whatever is at work in the
semantic paradoxes. Yet unlike the actual semantic paradoxes, the arguments
in which contradictions are derived in the set theoretic paradoxes dont have
any explicitly metalinguistic premises about the semantic properties of expres-

Representationalism and Set-Theoretic Paradox

sions. This can make it look as though something we do not understand is


going on in the set theoretic universe itself, and not just in our talk about it.
So the relation between the two sorts of paradox seems to demand further
investigation.
Finally, there is a question about how we are to construe the languages in
which we state our own set theories, even when these theories themselves are
restricted in what sets they countenance so as to be consistent (as far as we
can tell, that is). For suppose we accept some theory as the correct theory of
setsand by that I mean the correct theory of all the sets there are. One may, it
seems, write out this theory in some language and publish it, thereby attempting to inform others about everything there is in the universe of sets. Surely
if this act makes sense, it will be thought, the language in which we do so is
meaningful, and the standard way to take language, especially mathematical
language, as meaningful is to take its expressions and modes of composition
as having sets, ordered n-tuples etc., and functions therefrom and thereto as
their semantic values, and the sentences of the language as having their truthconditions determined from these in the usual way.
The problem here is that it seems to be a quite general feature of the enterprise of construing the language in which a particular set theory is expressed
as meaningful in this manner that we appeal to more sets than the existence
of which that set theory implies. The problems here will depend on the details
of the set theory in question to some extent, but to take a simple version of
the most common form of the problem, standard set theories avoid paradox
at least in part by implying that there is no set of all sets. But now consider a
language in which such a theory may be stated. The language, it would seem,
has a negation operatorafter all, we are able to say that there is not a set of
all sets. But, when it comes to open sentences with one free variable, a negation operator forms from one of them a sentence satised by all of the objects
that do not satisfy the original open sentence. But if the result makes sense in
the manner upon which the semantics insistsan open sentence in one free
variable is meaningful if and only if there is a set of objects that satisfy itthen
unrestricted passage to the complement makes sense for the sets over which the
variables of the language range. But unrestricted passage to the complement
only makes sense against the background of some totality of sets; otherwise
no set has a complement. So we are faced with a choice: either we cannot so
much as construe the language in which our theory is expressed as having a
standard negation, in which case it is hard to see how the theory expresses the
claim that there is no universal set, or we have to admit that there is after all a

14

Douglas Patterson

universal set relative to the theory and its language: that is, that there is some
identiable set of all the sets over which the variables of the language range.
This, in turn, cannot be the end of it: as we already know, if there really is a
universal set, everything is the case. So our only choice is to take the original
theory as simply incomplete as a story about what sets there are: there is a set
that is universal for it, but not actually universal; the theory holds of sets but
is interpreted within a proper class, holds at ranks lower than the rst inaccessible, and so onweve all heard stories of this sort before. And, of course,
the same reasoning will go through for whatever expanded set theory we take
ourselves as using to do semantics for the language of the original theory.
2. In order to motivate my own approach to these problems, I will survey a
representative sample of some others. I review these matters not to say anything
novel, nor to provide knock-down arguments against the approaches surveyed,
but simply as a representative assay of the costs theorists are willing to pay in
order to solve the three interrelated problems about set theory we are here to
discuss. Given that my aim will not be to convict any of these approaches of
incoherence or even to argue outright that they should not be accepted, my
summaries will be brief and will focus on the most pertinent elements of what
they uncontroversially entail.
A standard thought, with respect to our problems, is to hold that though not
every predicate determines a set, every predicate determines a class. Often not
much more is said about classes than that nave abstraction is valid for them,
and that at least some of them can not be members of anything. The terminology shifts; sometimes one has mathematical and logical sets, sometimes sets
and extensions; of late Field, discussed below, has sets and properties; the
distinction is nevertheless always between sets as what a consistent set theory
countenances and the semantic values of arbitrary predicates conceived of as
derivable by nave abstraction.
This view addresses our three problems or it would not have found so many
proponents. On it: (1) Nave abstraction is valid for classes, and it is just a
confusion of sets with classes that leads to Russells paradox. (2) The set theoretic paradoxes are sui generis and arise not from semantics, but, again, from
confusing sets with classes. Finally, (3), we do semantics for the language of
genuine set theory (e.g. Z or some reasonable extension; since it doesnt matter, I will just assume the theory in question is ZF) by taking its variables to
range over the members of the class of all sets. ZF is granted its status as the
theory of all sets despite our ability to do semantics for the language in which

Representationalism and Set-Theoretic Paradox

it is expressed, since that semantics posits a universal class (extension, propery,


etc.), not a universal set.
The result is a tidy package if there is some plausible distinction between sets
and classes. At this point we may choose among theories based on some form
of the distinction which retain classical logic, those that reject classical logic
in the course of maintaining some such distinction, and theories that simply
reject the distinction and do all of the work by revising logic. For examples
here we may take the standard picture advocated by von Neumann at the basis
of systems such as NBG, the hybrid views of Maddy and more recently Field,
and nally Priests dialeithic set theory.
Von Neumann (1925) distinguishes his classes from his sets in terms of size;
the proper (that is, non-set) classes are those collections that can mapped onto
the collection of all collections that can be members of further collections. The
distinction, clearly, depends on the distinction between collections that can be
members and those that cannot, and the reason for making such a distinction
at all, as Maddy notes, is purely the avoidance of paradox (Maddy 1983, 120).
Given the distinction, that proper classes cannot be members of sets can be
understood in terms of the fact that at no rank in the iterative hierarchy are all
of their members generated. Nevertheless, as Maddy writes:
The above considerations do show why these proper classes cannot be elements of sets, but they do not rule out the possibility that proper classes
might be elements of other proper classes, In fact, this possibility is strongly
indicated by the logical notion. It seems clear that the class of all innite collections is not only a member of the class of all collections with more than
three elements, but also a member of itself (121).

Thus, in addition to being undermotivated on its own, the ban on classes


being members of anything, and not merely on their being members of sets,
seems precisely to rule out their functioning as the semantic values of arbitrary
predicates, as Maddy emphasizes here: if is an innite collection is true of
anything, one of the things of which it is true is the collection of things of
which it is true. Thus:
it becomes very dicult to see why this layer of proper classes atop V is not
just another stage of sets we forgot to include. It looks just like another rank;
saying it is not seems arbitrary. The only dierence we can point to is that
the proper classes are banned from set membership, but so is the th rank
banned from membership in sets of rank less than . Because the classes look
so much like just another layer of sets, most set theorists simply think of the
proper classes of a weak system like VNBG as a metamathematical shorthand,

16

Douglas Patterson
and those of the stronger MK as subsets of a suitably chosen high rank. (For
example, take V to be R for the rst inaccessible cardinal, and take the
proper classes to be (R) R.) Let me express the objection I am making here to theories of this sort by saying they draw no signicant dierence
between sets and classes (122).

This dissatisfaction, I think, is something that nearly everyone feels with the
distinction between sets and classes as classically made in NBG and related
systems.
Maddys response is to cleave more closely to the intuition, one she attributes
(1983, 118) to Knig, that the fundamental distinction is in fact one between sets
as combinatorially determined in accordance with the iterative conception
expressed in Z and its extensions, and classes as extensions of properties, that
is, arbitrary conditions. Maddy bases her treatment on Kripkes well-known
treatment of the liar paradox within the parameters of Kleene three-valued
logic, and so part of the deal is signicant logical revision; Maddy takes it as
an additional feature of the distinction between sets and classes that the membership relation among classes is not everywhere dened: just as on Kripkes
construction certain sentences (intuitively, those like this sentence is false)
are assigned the middle value, so on Maddys construction claims about what
is a member of what likewise sometimes get the middle value.
As much as an advance as Maddys treatment might be, there is reason to be
dissatised with the Kleene logics lack of a reasonable conditional and set of
theorems. Since of the values 1, and 0 only 1 is designated, not even A A
is a theorem, since it is equivalent to ~A v A, which gets when A does. Field,
in recent work, has attempted to save what he calls nave property theory
along Maddys lines (2004, 81) within a logic with a reasonable conditional and
a set of theorems that includes all the instances of nave abstraction for properties. Like Maddy and von Neumann, Field wishes to maintain the distinction
between sets as conceived in ZF and classes, in his case properties conceived
of as the extensions of arbitrary open sentences:
According to the nave theory of properties, for every predicate (x) there is
a corresponding property x(x). Moreover, this property x(x) is instantiated by an object o i (o). More generally, the nave theory involves the
following nave comprehension schema:
NC. u1uny[Property(y) & x(x instantiates y (x, u1un))] (Field
2004, 78)

Field shows that the nave theory of properties may consistently be added to any
consistent theory T within his version of many valued logic which, as its main

Representationalism and Set-Theoretic Paradox

achievement, has a conditional that allows all the instances of NC to be true


even in the presence of component sentences that take undesignated values.
For all its nice properties, however, Fields conditional is crafted to avoid a
property theoretic version of the Curry paradox:
(NC) implies the existence of a Curry property K, for which x[x K
(x x )] where is any absurdity you like. So K K (K K );
that is
(i) K K (K K )
and
(ii) (K K ) K K.
But in many logics of we have the contraction rule A (A B)  A
B, on which (i) implies
(i*) K K
But this with (ii) leads to K K by modus ponens; and another application of
modus ponens leads from that and (i*) to .
Unless we restrict modus ponens (and it turns out that very drastic restrictions
of it would be required), we need to restrict the contraction rule. This requires
further restrictions on the logic as well. For instance, given that we are keeping
modus ponens in the form A, A B B, we certainly have A, A (A
B)  B simply by using modus ponens twice; so to prevent contraction, we
cannot have the generalized -introduction meta-rule (82).

Thus Field will grant one nave property theory, but at the price of the general
invalidity of conditional proof; Fields thus does not express the consequence relation of the underlying logic. Field supports the view with an ingenious semantic treatment which would merit involved discussion, but for our
purposes here, these basic features are what we need note.
Now one might react to Fields proposal (and Maddys, for that matter)
by noting that if one is willing to revise logic to retain views in set theory,
one might do well simply to jettison the distinction between sets and classes
(logical sets, properties, etc.) entirely and try to do all the work with logical
revision. The only motivation for the distinction in the rst place is to retain
both nave abstraction and a treatment of sets suitable for use in mathematics.
What if, given a suitable revision of logic, one could do without a distinction
between the objects about which nave abstraction is correct and the objects
of which a mathematical theory of sets treats? This is the cure Priests nds in
his dialeitheic treatment of set theory (Priest 2006, chs. 2 and 18). On Priests
view sentences can be both true and false; true sentences are to be asserted
whether or not they are also false. Since sentences can be both true and false,
disjunctive syllogism is invalid (since one cannot conclude from the truth of a

18

Douglas Patterson

disjunction and the falsehood of one of its disjuncts that the other disjunct is
trueit could be that the original false disjunct is also true), and since this is
so, ex falso is invalid. Given this, Priest can maintain that some contradictions
are true without logical triviality.
As applied to set theory, the view is then that nave abstraction (plus extensionality) is the correct set theory, and it is just that some theorems of nave
set theory are both true and false. As for the Curry, Priest like Field works up
a suitable conditional for which absorption (that is, A (A B)  A B)
fails (2006, ch. 6). The result is nave set theory in a paraconsistent but logically
non-trivial form, and all this without any problematic distinction between sets
and classes. The price, as with Field and Maddy, is in the logic, though in this
case it is somewhat higher. Without contraction, we dont have full conditional
proof, as we have just seen Field explain. In addition one gets Priests true
contradictions and loses disjunctive syllogism. Still, unlike Field, Priest needs
no distinction between sets and classes:
Proper classes, if we are to take them seriously (and not just as faons de parler),
are a masquerade. The cumulative hierarchy was supposed to be an analysis of
the notion of set. It is supposed to contain all sets. If we are forced to admit
that there are sets outside the hierarchy, this just shows that the analysis is
wrong. And calling them by a dierent name is just a trivial evasion. Moreover,
the insistence that proper classes cannot be members of other collections can
have no satisfactory rationale. If they are determinate collections with determinate members, there is no reason why we should not consider them to be
members of other collections, for example their singletons (2006, 34)

Given the intuitive implausibility of a distinction between sets and classes, this
is to be counted as a signicant point in Priests favor.
3. The above survey, to my knowledge, accurately represents extant approaches: according to extant approaches, if you want to think consistently (for Priest,
non-trivially) about the universe of sets while keeping an eye on nave abstraction and the semantics of the language of your own theory, you will have to
accept an under-motivated distinction between sets and something else, signicant revisions in classical logic, or both.1 We can spot the confusion responsible
1

One apparent exception here is the approach based on higher order logic introduced of late by
Rayo and Williamson (Williamson 2003, Rayo and Williamson 2003). Prima facie, of course,
it doesnt improve much on the set-class distinction to say that one can do semantics for all
sets by insisting that the interpretants of higher-order quantiers arent sets, but Williamson
and Rayo have an answer: construe higher-order quantication, following Boolos (1998, essay

Representationalism and Set-Theoretic Paradox

for the impasse, and begin to see the way out, if we return and reconsider the
argument I oered above when discussing the problem of how to do semantics
set theoretically for a language in which we express what we take to be the
complete account of the set theoretic universe. Consider this paragraph from
Priest:
Take a rst order language and consider the denition of logical validity for
that language:
 i in every interpretation in which all the members of are true, is
true.

An interpretation is a set theoretic entity whose domain is an arbitrary


set. Hence there are interpretations of arbitrarily high rank. Thus, the
denition of validity has quantiers that range over the universe of
sets, which is not, according to the cumulative hierarchy, a set. But
now consider the language in which this denition is given. Normally
this is just a fragment of mathematical English. What is its semantics?
Obviously it has as semantics, since we make perfectly meaningful and
true assertions in it. No coherent answer can be given, at least if we
adhere to the cumulative hierarchy. For an interpretation is a pair, D,
I, where D here is the domain of quantication. But D is not a set,
so this is just nonsense (36).
This is a succinct presentation of what we can call the argument from semantic
ascent against CLS.
The crucial move in the argument comes with Priests emphatic claim that
obviously it has a semantics, since we make perfectly meaningful and true assertions in it. The wording of the claim, and the emphasis used, insinuate that
the claim is just obviously and trivially true, but it is not. The claim presupposes
that since we are obviously able to use language to do things like express our
acceptance of a certain denition of validity for the language of our chosen
set theory, this language must have a semantics itself. What is not trivial here
is the connection between expressing our belief that the set theoretic universe
is thus-and-so characterized, and our doing so in a language that can receive
a standard semantic treatment. Before I explain why the assumption can and
should be denied, let us note its appearance in Field. Toward the end of his
article, Field notes the way in which nave property theory assigns a property
4) as plural quantication that introduces no new entities at all. As Rayo admits (2006) there
are problems with the approach as one goes beyond the second order. Ill have to leave this
issue open here.

20

Douglas Patterson

to every predicate while at the same time standard set theory, which Field
endorses, simply denies the existence of some of the sets that would putatively
correspond to these properties and predicates. He writes:
But if there is no need for a nave theory of sets, why is there a need for a nave
theory of properties, and for a nave theory of satisfaction? Was this paper a
wasted eort?
In fact the case of properties (on at least one conception of them) and of
satisfaction are totally dierent from the case of sets. For the way we solve
the paradoxes of nave set theory in ZF is to deny the existence of the alleged
set: for instance, there simply is no set of all sets that do not have themselves
as members. The analogous paradox in the case of the theory of satisfaction
involves the expression is not true of itself , and if we were to try to solve the
paradox on strictly analogous lines, we would have to deny the existence of the
expression! That would be absurd: after all, I have just exhibited it (103).

As with Priest, the crucial assumption is at work here, but is misrepresented


as something trivially true. The point, if Fields view is to be vindicated, cant
merely be that a string of characters has been exhibitedthat wouldnt show
anythingbut that a string of characters that has a certain meaning has been
exhibited. Now, of course, ordinary speakers of English do take is not true of
itself to have a meaning such that, if an expression had that meaning, Grellings paradox would be a sound argument. But we might simply take this to
show that ordinary speakers have false beliefs about meaning. Here there is a
move from the fact that speakers reliably take certain expressions to have certain
meanings to the claim that, in fact, these expressions do have these meanings.
Common to both Priest and Field is the uncritical assumption that familiar
expressions used to communicate familiar thoughts must bear the meanings
they seem to speakers to bear.
It is this which I believe we can question; in fact, I believe that we can reject
it at no theoretical cost at all, save an initial insult to intuition. Seeing this requires a simple bit of clear thought of what one does in communicating about
set theory (or anything else). Suppose that you want to let someone who seems
confused about Russells paradox know that there is no set of all sets. What
will you say? Well, of course, it depends on to whom you are talking. If it is a
speaker of English, you might say:
There is no set of all sets.
If you are dealing with a monolingual speaker of French, you might say:

Representationalism and Set-Theoretic Paradox

Il ny as pas densemble de tous les ensembles.


If it is someone reasonably familiar with standard logical notation, you might
write out:
~(x)(Set(x) & (y)(Set(y) y x))
while if it is someone whom you know to be mostly competent as a speaker of
English, save thinking that cat means what most people think set means
and vice versa, you might, if you do not want to bother correcting his English
before getting down to business, say:
There is no cat of all cats.
The point is that to express your set-theoretic belief, and, as you hope, get your
interlocutor to accept it, it is neither necessary nor sucient that you use a
sentence that intuitively means that there is no set of all sets. A sentence that
your interlocutor merely thinks means that there is no set of all sets will do,
and indeed even this is not necessary: suppose, for example, that you and your
interlocutor have, as some sort of game, agreed to treat sentences as though
our set-cat confused person were right about them. You might play the
game a whileI fed my set this morning, heh heh, Thats a nice chess cat
you have there, hee hee hee, I wonder if your set is going to have sittens?
Har haruntil falling into a serious discussion of the set-theoretic paradoxes,
still playing the set-cat imitation game and saying, eventually, we have to
conclude that there is no cat of all cats.
It is sucient for communication that the hearer believe that the sentence
uttered have the truth-condition the speaker intends to indicate obtains. All
that is necessary for communication is that speaker and hearer agree with respect
to the truth conditions they will treat an uttered sentence as havingwhether
this agreement be explicit, tacit, or even wholly accidental or based on pretense
(Graham 2000, Pettit 2002, Gross 2005 and Woodbridge 2005, 2006 all stress
related points in highly complementary ways). It follows from this point that
the sentences that we use to express our belief in a set theory do not have to
mean anything at all: it is sucient that we think they mean things about sets,
and, indeed, sucient that we act as though they mean things about sets. If
these do not mean anything, then, it need not be possible to give them a settheoretic semantics. And if it need not be possible to give the language in which

22

Douglas Patterson

we express our belief about all of the sets there are a set-theoretic semantics,
then our dilemma is gone: we are not caught between admitting that what we
thought was the complete theory of sets is incomplete and giving an inconsistent semantics for the language in which we express it. We thus do not need
to try to wriggle out of the dilemma by revising logic or resting everything on
a distinction between the sets of which our set theory is a complete theory,
and the non-sets that ll out the domain of quantication of the semantics of
the language this theory.
If we accept this view, we are allowed to admit that the following is simply
true: any set theory for the language of which you can give a standard set-theoretic semantics is an incomplete theory of sets. If you think that your favorite
set theory is the whole story, you had better not think that you can state it in
a language that you can give a set theoretic semantics. Fortunately, you do not
need to: you can convey to others what it is that you believe about set theory
by using sentences that they take to have certain truth conditions, or that they
take you to take them to have, or even that you both agree to pretend to have
those truth conditions. The sentences do not need to have those truth conditions to get the job done; all that is necessary is a match between the condition
to the obtaining of which you intend to commit by uttering a sentence and
the condition your interlocutor treats (even as a pretense) as necessary and
sucient for you to have spoken truly.
The alternative, then, to revising logic or accepting some sort of set-class
distinction, is to accept an error theory of other peoples beliefs about the
meaning of sentences of the language of (whatever you take to be) the complete story about sets, and a pretense-theoretic account of your own. That is
over-simplied, of course: the correct story about yourself in your unreective
moments may be an error-theoretic one, just as the correct story about some
others, in their reective moments, may be a pretense-theoretic one; the point
is clear enough. All the sets there are (according to you) are merely all the sets
you believe in by accepting your favorite set theory; no further sets need to
exist for you to believe that, and no further sets need to exist for you to express
the belief to someone else.
To review my responses to the three problems with which I began, they
are as follows. The answer as to (3) how one can do set theoretic semantics
for the language of complete theory of sets is that one cant and need not.
As for (1), nave abstraction is simply false, since it is inconsistent; speakers are inclined to accept it, however, because the idea that every predicate
determines a set is a feature of nave semantics. This leads into issue (2), the

Representationalism and Set-Theoretic Paradox

relation of the set theoretic paradoxes to the semantic paradoxes. On the view
I oer, both sorts of paradox ultimately rest on false theses about meaning
and belief, and not, e.g., on failure to recognize that logic is really three valued or paraconsistent. In the case of the semantic paradoxes, the false thesis
is what I call, in Patterson forthcoming, belief in the epistemic transparency
of meaning, the idea that competent speakers cannot be wrong about what
expressions of the language with which they are competent mean. In the case
of the set theoretic paradoxes, as I will explain, the problem goes a bit deeper,
to the point where it touches basic issues about what it is to have a belief at
all.
The obvious objection, of course, is that the words we use to convey our
ideas about set theory (or cats, for that matter) seem meaningful and therefore
must be meaningful. See 2007a and 2007b for my remarks on this. For now,
the main point is just that I have not denied that our assertions express our
beliefs and are in that sense meaningful communicative acts; I have denied only
that communication requires the words we use actually to mean what we take
them to mean. Likewise, I have not denied that there is a way the set theoretic
universe is and that one can come to believe it to be that way; the present view
is entirely realistic about sets and set theory.
4. I have taken on a range of related commitments about the relation of belief
to language. My basic commitment, obviously, is that its possible to believe
that p without bearing certain understanding-constituting relations to some
sentence that means that p. This is familiar enough and is maintained by anyone who accepts that common household pets have beliefs. What is maybe a
bit more radical is my extension of this thesis even to abstract thought. Note,
however, that it is compatible with my story that it is not possible, at least for
certain kinds of thinking beings (say, us), to have beliefs about abstract topics
without being related to sentences that seem to them to have these meanings.
This strikes me as a plausible claim about human psychology, that people cannot come to have the belief that there is no set of all ordinals without starting
from beliefs about collections of toys, working their way up through sentences
that they take to be about sets, and arriving, after a while, at abstract thought
about sets. It is not for me to pronounce on that psychological matter, but
whatever the right psychological story is about the development of cognition about set theory and other abstract topics, my account will be perfectly
compatible with it, because my point will be that thinking that sentences are
meaningful does all the cognitive work while the putative meanings of these

24

Douglas Patterson

sentences do none: they are always explanatorily screened o by the attitudes


of subjects.
The more signicant, but still defensible, conception of belief I need is this:
belief is not a relation to an item with semantic properties. That belief involves
such a relation is the view I called representationalism in the introduction,
and of which I take Field 1978 to be a classic statement. Perhaps the three most
common forms of representationalism are the views (1) that belief is a relation
to a proposition that itself has truth conditions,2 (2) that belief is a relation
to a sentence of natural language that has a meaning, and (3) that belief is a
relationship to an inner representational statea sentence of the language of
thought, if one prefersthat again has a meaning. Each view, of course, will
be spelled out further by its proponents, (1) involving claims about how thinkers come into relation to propositions, (2) involving views about the relation
between belief and language, and (3) involving claims about how cognition
is implemented in physical systems. Obviously, I deny (2). But (1) and (3)
remain, and either threatens to undermine my approach, for obvious reasons.
If to believe that p is to bear a relationship to a representation r that is true i
p, and if being true i p is construed in terms of set-theoretic semantics, then
the move from language to thought has netted us nothing. For belief will then
be a relationship to an interpreted entity (in the case of a proposition one that
bears an interpretation necessarily on the usual views), and we will be back
in the problem raised by the argument from semantic ascent against CLS: if
an interpretation is a function from a representation to what is represented,
and functions are set theoretic entities, then no interpretation can coherently
be taken to treat a representation as truly representing all sets as having some
feature in common.
If belief is a relation to a representation, then, there is no such thing as even
believing something about all sets, and hence is it absolutely impossible to
believe the truth about the universe of setsindeed, absolutely impossible to
believe anything true about the entirety of the universe of sets, even that every
set is identical to itself. I take that last claim to be a reductio of the view that
belief is a relation to a representation. I will rst explain what view of belief I
would put in its place, and I will second say something about what belief has
to do with propositions and with inner computational states. After that I will
address some residual questions about the treatment of belief I require.
2

The view I reject, then, is that belief is a relation to a proposition construed as something
that has a truth condition. I have no problem with the conception of a belief as a relation to
something that is a truth conditione.g. a set of possible worlds.

Representationalism and Set-Theoretic Paradox

What alternative do we have if we do not hold that belief is a relationship


between a thinker and some representation? Plenty of options are open; e.g.
the encyclopedia entry Schwitzgebel 2006 gives me at least dispositionalism and interpretivism as alternatives. I cant say much in the space I have
here about the view I favor, though of course I am committed to doing so
somewhere, soon. (My primary inuences here have long been Stalnaker 1984
and Collins 1987.) I will say the following. When you believe that the empty
set has a singleton, you are in a state such that you are right if the empty set
has a singleton, and wrong if it isnt. You bear a statusbeing right or being
wrongdepending on how things are with the empty set. No representational
intermediary is required. There is a complex story to be told about how you
get into this position, but the position you are in is itself simple in this way. It
is not that you bear a relation to a representation that has a truth condition;
it is that you yourself are either mistaken or not depending on how things are
with the empty set.
Obviously, ultimately I owe a complete story about belief. In this paper I
am simply noting the commitments of my treatment of the paradoxes, and
there you have them. Let me now say a little about propositions, and then a
little about mental representations. As for propositions, a distinction is crucial
to clear thought about their role in belief, a distinction between, on the one
hand, the idea that beliefs (and other attitudes) are individuated by relations
to propositions construed as set theoretic objects, and, on the other hand, the
idea that beliefs derive their intentionality from their relations to propositions
so construed. As for the rst idea, there is nothing wrong with thinking that
the individuation of beliefs runs by way of relations to certain abstracta. This
rst idea doesnt support the second at all, for reasons already entered: if beliefs
derive their intentionality from relations to propositions which themselves are
semantically interpreted, it is impossible to believe even, of every set, that it
is self-identical. I do not pretend that this is a complete treatment of the distinction; in another place I intend to discuss the matter, and in particular its
relation to standard Skolemite considerations (see Lavine 2006) more fully.
As for mental representation, the story is similar. I would be the last to deny
that empirical science may show, and to a signicant extent has shown, that on
the best theories of human cognition, human cognitive states as implemented
in the brain are highly systematic in ways that implement the inferential and
representational systematicity of human thought. Beliefs may well be such that
(at least relative to a believer or a type of believer) a given belief type is tokened
when and only when the believers underlying cognitive apparatus is in a given

26

Douglas Patterson

computational state. Indeed, I nd it hard to imagine how that claim could


be false on its most straightforward readings. What I deny is that the intentionality of belief derives from relations to these computational states where
these, in turn, get a standard semantic interpretation. If my belief that there
is no set of all sets is a relation to a sentence in the language of thought that
means that there is no set of all sets, then we are back in the soup: the language
of thought will need a set theoretic semantics, and a condition the semantics
being coherent will be representing me as not really having a belief about all
sets. What is empirical is how cognition implemented. What is not empirical is
the theory that the aboutness of belief derives from the aboutness of these
underlying computational states. Again, I take the paradoxes of set theory and
unrestricted quantication to be reductios of that latter view. Clearly there is
more to say here, too; again, I am just noting commitments that I intend to
discuss more fully elsewhere.
5. The real culprit in the set theoretic paradoxes is the representationalist
view that one believes or expresses things about sets by bearing relations to
representations that themselves are semantically interpreted. Neither believing
something about sets nor conveying these beliefs to another can be construed
as a relation to an interpreted representation if one wants to have a coherent
view. When we see that we can give this up, we can see the argument from
semantic ascent against CLS is simply unsound. Interestingly, then, the real
work in solving the set theoretic paradoxes is to be done in the philosophy of
mindnot, as most have thought, in set theory or in logic.

References
Boolos, George. 1998. Logic, Logic and Logic. Cambridge, MA: Harvard University
Press.
Collins, Arthur. 1987. The Nature of Mental Things. Notre Dame, IN: Notre Dame
University Press.
Field, Hartry. 1978. Mental Representation. Erkenntnis 13: 961.
Field, Hartry. 2003. The Semantic Paradoxes and the Paradoxes of Vagueness. In Liars
and Heaps: New Essays on Paradox, edited by J. C. Beall, 262311. New York: Oxford
University Press.
Field, Hartry. 2004. The Consistency of the Nave Theory of Properties. The Philosophical Quarterly 54, n. 214: 78104.

Representationalism and Set-Theoretic Paradox

Graham, Peter. 2000. Conveying Information. Synthese 12: 365392.


Lavine, Shaughan. 2006. Something About Everything: Universal Quantication in
the Universal Sense of Universal Quantication. In Rayo and Uzquiano 2006.
Maddy, Penelope. 1983. Proper Classes. The Journal of Symbolic Logic 48: 113139.
Patterson, Douglas. 2007a. Understanding the Liar. In Revenge of the Liar: New Essays on the Paradox, edited by J. C. Beall, 184224. New York: Oxford University
Press.
Patterson, Douglas. 2007b. Inconsistency Theories: The Signicance of Seamantic Ascent. Inquiry 50, 6: 575589.
Patterson, Douglas. Forthcoming. Inconsistency Theories of Semantic Paradox. Philosophy and Phenomenological Research.
Priest, Graham. 2006. In Contradiction: A Study of the Transconsistent, 2nd ed. New York:
Oxford University Press.
Rayo, Augustn. 2006. Beyond Plurals. In Rayo and Uzquiano 2006.
Rayo, Augustn and Uzquiano, Gabriel. 2006. Absolute Generality. New York: Oxford
University Press.
Rayo, Augustn and Williamson, Timothy. 2003. A Completeness Theorem for Unrestricted First Order Languages. In Liars and Heaps: New Essays on Paradox, edited
by J. C. Beall, 33 1356. New York: Oxford University Press.
Schwitzgebel, Eric. Belief. The Stanford Encyclopedia of Philosophy. http://plato.stanford.edu/entries/belief.
Stalnaker, Robert. 1984. Inquiry. Cambridge, MA: M.I.T. Press.
von Neumann, John. 1925. An Axiomatization of Set Theory. In From Frege to Gdel: A
Source Book in Manthematical Logic, 18791931, edited by J. van Heijenoort. Cambridge, MA: Harvard University Press, 1967.
Williamson, Timothy. 2003. Everything. Philosophical Perspectives, 17, Language and
Philosophical Linguistics: 415465.
Woodbridge, James. 2005. Truth as a Pretense. In Fictionalism in Metaphysics, edited
by M. Kalderon, 134177. New York: Oxford University Press.
Woodbridge, James. 2006. Propositions as Semantic Pretense. Language and Communication 26: 343355.

28

Mark Colyvan

Whos Afraid of Inconsistent Mathematics?


Mark Colyvan

Abstract
Contemporary mathematical theories are generally thought to be consistent. But it hasnt
always been this way; there have been times in the history of mathematics when the consistency of various mathematical theories has been called into question. And some theories,
such as nave set theory and (arguably) the early calculus, were shown to be inconsistent.
In this paper I will consider some of the philosophical issues arising from inconsistent mathematical theories.

1. A Five Line Proof of Fermats Last Theorem


Fermats Last Theorem says that there are no positive integers x, y, and z, and
integer n > 2, such that xn + yn = zn. This theorem has a long and illustrious
history but was nally proven in the 1990s by Andrew Wiles. Despite the apparent simplicity of the theorem itself, the proof runs over a hundred pages,
invokes some very advanced mathematics (the theory elliptic curves, amongst
other things), and is understandable to only a handful of mathematicians.1 But
now consider the following proof.
Fermats Last Theorem (FLT): There are no positive integers x, y, and z, and
integer n > 2, such that xn + yn = zn.
Proof: Let R stand for the Russell set, the set of all sets that are not members
of themselves: R = {x : xx}. It is straightforward to show that this set is both
a member of itself and not a member of itself: RR and RR. Since RR, it
follows that RR or FLT. But since RR, by disjunctive syllogism, FLT.
This proof is short, easily understood by anyone with just a bit of high-school
mathematics. Moreover, the proof was available to mathematicians well before
Wiles groundbreaking research. Why wasnt the above proof ever advanced?
One reason is that the proof invokes an inconsistent mathematical theory,
namely, nave set theory. This theory was shown to be inconsistent toward the
end of the 19th century. The most famous inconsistency arising in it was a para1

See S. Singh, Fermats Last Theorem: The Story of a Riddle that Confounded the Worlds Greatest
Minds for 358 Years, London 1997, for a popular account of Fermats Last Theorem.

Whos Afraid of Inconsistent Mathematics?

dox due to Bertrand Russell. I invoked Russells paradoxical set in the above
proof.2 Paradoxes such as Russells (and, to a lesser extent, others such as the
Burali-Forti ordinal paradox and Cantors cardinality paradox) led to a crisis in
mathematics at the turn of the 20th Century. This, in turn, led to many years of
sustained work on the foundations of mathematics. In particular, a huge eort
was put into nding a consistent (or at least not known-to-be-inconsistent)
replacement for nave set theory. The generally-agreed-upon replacement is
Zermelo-Fraenkel set theory with the axiom of choice (ZFC).3
But the inconsistency of nave set theory cannot be the whole story of why
the above proof of Fermats Last Theorem was never seriously advanced. After
all, there was a period of some 30 odd years between the discovery of Russells
paradox and the development of ZFC. Mathematicians did not shut up shop
until the foundational questions were settled. They continued working, using
nave set theory, albeit rather cautiously. Moreover, it might be argued that
many mathematicians to this day, still use nave set theory.4 In summary, we
have a situation where mathematicians knew about the paradoxes and they
continued to use a known-to-be-inconsistent mathematical theory in the
development of other branches of mathematics and in applications beyond
mathematics.
This raises a number of interesting philosophical questions about inconsistent mathematics, its logic and its applications. Ill pursue two of these issues
in this paper. The rst concerns the logic used in mathematics. It is part of the
accepted wisdom that in mathematics, classical logic is king. Despite a serious challenge from the intuitionists in the early part of the twentieth century,
classical logic is generally thought to have prevailed. But now we have a new
challenge from logics more tolerant to inconsistency, so-called paraconsistent
logics. In the next section I will give a brief outline of paraconsistent logics and
discuss their relevance for the question of the appropriate logic for mathemat2
3

The paradox is that the Russell set both is and is not a member of itself.
See M. Giaquinto, The Search for Certainty: A Philosophical Account of Foundations of
Mathematics, Oxford 2002, for an account of the history and H. B. Enderton, Elements of
Set Theory, New York 1997, for details of ZFC set theory.
After all, so long as you are careful to skirt around the known paradoxes of nave set theory, it
can be safely used in areas such as analysis, topology, algebra and the like. Most mathematical
proofs, outside of set theory, do not explicitly state the set theory being employed. Moreover,
typically these proofs do not show how the various set-theoretic constructions are legitimate
according to ZFC. This suggests, at least, that the background set theory is nave, where there
are less restrictions on set-theoretic constructions. See Enderton, 1997 and P. R. Halmos,
Nave Set Theory, New York 1974, for the details.

30

Mark Colyvan

ics. I will suggest that not only are such logics appropriate, but they may already
be the logic of choice amongst the mathematical community.
The second general topic I will discuss concerns applications of inconsistent
mathematics, both within mathematics itself and in empirical science. There
are many questions here but I will focus on two: how can an inconsistent
theory apply to a presumably consistent world?; and what do the applications
of inconsistent mathematical theories tell us about what exists? But before we
broach such philosophical matters, I will rst present a couple of examples of
inconsistent mathematical theories.

2. Inconsistent Mathematics
We have already seen Russells paradox, the paradox arising from the set of
all sets that are not members of themselves: R = {x : xx}. The paradox arises
because of an axiom of nave set theory known as unrestricted comprehension.
This axiom says that for every predicate, there is a corresponding set. So, for
example, there is the predicate is a cat and there is the set of all cats; there is
the predicate is a natural number and there is the set of all natural numbers.
So far, so good. The trouble starts when we consider predicates such as is a
set or is a non-self-membered set. If there are sets corresponding to these
two predicates, we get Cantors cardinality paradox and Russells paradox, respectively. Cantors cardinality paradox starts by assuming that there is a set of
all sets, , with cardinality5 . Now consider the power set of : (). Cantors theorem can be invoked to show that the cardinality of () is strictly
greater than the cardinality of . But is the set of all sets and so must have
cardinality at least as large as any set of sets. Since () is a set of sets, we
have a contradiction.
The nave axiom of unrestricted comprehension was seen to be the culprit
in all the paradoxes, and mathematicians set about nding ways to limit the
scope of this overly powerful principle. One obvious suggestion is to simply
ban the problematic setslike the set of all sets and Russells set. This, however,
is clearly ad hoc. Slightly better is to ban all sets that refer to themselves (either
explicitly or implicitly) in their own specication. The generally-agreed-upon
solution achieves the latter by invoking axioms that insure that such problem5

This, in a mathematically precise sense, is the size of the set.

Whos Afraid of Inconsistent Mathematics?

atic sets (and others as well) cannot be formed. This is ZFC. The basic idea is
to have a hierarchy of sets, where sets can only be formed from sets of a lower
levela set cannot have itself as a member, for instance, because that would involve collecting sets from the same level. Nor can there be a set of all setsonly
a set of all sets from lower down in the hierarchy. ZFC has not engendered any
paradoxes but it has the look and feel of a theory designed to avoid disaster
rather than a natural successor to nave set theory. More on this later.
Another important example of an inconsistent mathematical theory is the
early calculus. When the calculus was rst developed in the late 17th century
by Newton and Leibniz, it was fairly straightforwardly inconsistent. It invoked
strange mathematical items called innitesimals (or uxions). These items are
supposed to be changing mathematical entities that approach zero. The problem is that in some places these entities behave like real numbers close to zero
but in other places they behave like zero. Take an example from the early calculus: dierentiating a polynomial such as f(x) = ax2 + bx + c.6
f (x + ) f (x )

(1)

a( x + )2 + b( x + ) + c ( ax 2 + bx + c )

(2)

2ax + 2 + b

(3)

f ( x ) =

= 2ax + b +

(4)

= 2ax + b

(5)

Here we see that at lines 13 the innitesimal is treated as non-zero, for otherwise we could not divide by it. But just one line later we nd that 2ax + b +
= 2ax + b, which implies that = 0. The dual nature of such innitesimals
can lead to trouble, at least if care is not exercised. After all, if innitesimals
behave like zero in situation like lines 4 and 5 above, why not allow:
6

The omission of the limit lim0 from the right-hand side on the rst four lines of the following
calculation is deliberate. Such limits are a modern development. At the time of Newton and
Leibniz, there was no rigorous theory of limits; dierentiating from rst principles was along
the lines presented here.

32

Mark Colyvan

2=3
then divide by to yield
2 = 3?
This illustrates how easily trouble can arise and spread if 17th and 18th century
mathematicians werent careful. There were rules about how these inconsistent
mathematical objects, innitesimals, were to be used. And according to the
rules in question, the rst calculation above is legitimate but the second is
not. No surprises there. But one can quite reasonably ask after the motivation
for the rules in question. Such rules about what is legitimate and what is not
require motivation beyond what does and what does not lead to trouble.
The calculus was eventually, and gradually, made rigorous by the work of Bolzano, Cauchy, Weierstrass, and others7 in the 19th century. This was achieved
by a rigorous () denition of limit.8 So, to be clear, I am not claiming that
there are any ongoing consistency problems for the calculus. The point is simply that for over a hundred years mathematicians and physicists worked with
what would seem to be an inconsistent theory of calculus.9

3. Is the Appropriate Logic for Mathematics Paraconsistent?


Classical logic has it that an argument form known as ex contradictione quodlibet or explosion is valid. The argument form was used in my proof of Fermats
Last Theorem at the beginning of this paper. According to explosion any arbitrary proposition follows from a contradiction.10 Logics in which this argument
7 M. Kline, Mathematical Thought from Ancient to Modern Times, New York 1972.
8 More recently there has been a revival of something like the original innitesimal idea by
A. Robinson, Non-standard Analysis, Amsterdam 1966, and J. H. Conway, On Numbers and
Games, New York 1976, and even an explicitly inconsistent theory of innitesimals by C.
Mortensen, Inconsistent Mathematics, Dortrecht 1995.
9 There are also cases where explicitly inconsistent, but non-trivial, theories have been
developed. See R. K. Meyer, Relevant Arithmetic, Bulletin of the Section of Logic of the Polish
Academy of Sciences 1976, 5:133137; R. K. Meyer, and C. Mortensen, Inconsistent Models
for Relevant Arithmetic, Journal of Symbolic Logic 1984, 49: 917929; C. Mortensen, 1995;
G. Priest, Inconsistent Models of Arithmetic Part I: Finite Models, Journal of Philosophical
Logic 1997, 26(2): 223235; and G. Priest, Inconsistent Models of Arithmetic Part II: The
General Case, Journal of Symbolic Logic 2000, 65: 15191529.
10 The negation of Fermats Last Theorem, or anything else can be proven just as easily, and
with pretty much the same proof as the one I opened with.

Whos Afraid of Inconsistent Mathematics?

form is valid are said to be explosive. A paraconsistent logic is one that is not
explosive. That is, in a paraconsistent logic at least one proposition does not
follow from a contradiction. Ex contradictione quodlibet is invalid according
to such logics.
There are many paraconsistent logics in the market place but let me sketch
the details of one, just to make the discussion concrete. The logic LP, is a threevalued logic with values 0, i, and 1 (here 1 is true, 0 is false and i is the other
value, quite reasonably interpreted as both true and false). So far nothing
unusual; several logics have three values. The interesting feature of LP is that
the crucial notion of validity is dened in terms of preservation of two of the
truth values: an argument is valid if whenever the truth value of the premises
are not 0, the truth value of the conclusion is not 0.11 We also need to dene
the operator tables for the logical connectives (i.e., dene how conjunctions,
disjunctions, and negations get their truth values).12 Negation, conjunctions
and disjunction (respectively) are given by the following tables:13

1
i
0

0
i
1

1
i
0

1
1
i
0

i
i
i
0

0
0
0
0

1
i
0

1
1
i
1

i
i
i
i

0
1
i
0

From these we see that if some sentence P has the truth value i, its negation,
P, also has the value i, and so does the conjunction of the two: PP. Now
11 This denition of validity is a natural extension of the usual denition of validity in classical
logic: an argument is valid if whenever the premises are true, the conclusion is also true.
The change of focus from truth to non-falsity does not matter in classical logic, since there
are only two truth values (non-falsity and truth are the same thing). But in a three-valued
logic, this change of focus to non-truth makes all the dierence.
12 See JC Beall, and B. C. van Fraassen, Possibilities and Paradox, Oxford 2003; G. Priest, Worlds
Possible and Impossible: An Introduction to Non-Classical Logic, Cambridge 2001; or G. Priest
and K. Tanaka, Paraconsistent Logic, in E. N. Zalta (ed.), The Stanford Encyclopedia of
Philosophy 2004, (Winter 2004 Edition), URL http://plato.stanford.edu/archives/win/2004/
entries/logic-paraconsistent for full details and further discussion. The operator tables are the
same as for the Kleene strong logic K3.
13 These operator tables dene negation (), conjunction (), and disjunction () respectively.
They are read as follows: (i) in the rst table, read the right-hand column as giving the truth
values of the unnegated proposition and the left-hand column as giving the corresponding
truth value for the negation; (ii) in the second and third tables, read the top row and the left
column (the ones separated from the main table by horizontal and vertical lines, respectively)
to represent the truth values of the two conjuncts/disjuncts and the corresponding entry of
the main table gives the truth value of the conjunction/disjunction.

34

Mark Colyvan

take some false sentence Q (i.e., whose truth value is 0) and consider the argument from PP to Q. In LP this argument is invalid, since the premise
PP does not have the truth value 0 and yet the conclusion Q does have the
truth value 0. In this logic the proof of Fermats Last theorem that I gave
earlier is invalid.
Whats the philosophical signicance of all this? Well, it might just be that,
mathematicians were never tempted by the above proof of Fermats Last Theorem because the appropriate logic of mathematical proofs is a paraconsistent
one. Perhaps this sounds implausible. Surely all we need to do is ask a mathematician which logic they use and surely theyll all answer classical logic (or
perhaps intuitionistic logic). For various reasons it might be interesting to
conduct such sociological research of mathematicians beliefs but it will not
help us answer the question at hand about the logic of mathematics. Our question is which logic do mathematicians actually use, and this is determined by
mathematical practice, not by what mathematicians claim they use. (Indeed,
most mathematicians are not experts in the dierences between the various
logics available.)
Perhaps, mathematicians dont use a paraconsistent logic but, rather, just
avoid proofs like the ve-line proof of FLT given earlier. Indeed, they might
steer clear of contradictions generally. The latter is hard to do, though, when
youre working in a theory thats known to be inconsistent. But perhaps part
of what it takes to be a good mathematician is to recognise, not just valid
proofs, but also sensible ones. On this suggestion, the proof I opened with
might be formally valid but its not sensible, since it involves a contradiction
(it takes a contradiction as a premise). But this wont do as a response. First,
the contradiction in question can be proven fairly straightforwardly in a very
rigorous way from, what was at the time, the best available theory of sets; its not
some implausible proposition without any support. Second, not all arguments
involving contradictions (or taking contradictions as premises) are defective.
Take the argument from PP therefore PP. Surely this is both valid and
sensible. Putting these issues aside, the most serious problem with this line of
response is that the notion of a sensible proof is in need of clarication. The
advocate of a paraconsistent logic has no such problem here; they have only
the one notion: (paraconsistent) validity and the proof in question fails to be
valid.
Even if mathematicians do use classical logic but exercise some (ill-dened)
caution about what proofs to accept above and beyond the valid ones, perhaps
they ought to use a paraconsistent logic. As Ive already suggested, one reason

Whos Afraid of Inconsistent Mathematics?

for thinking this is that the paraconsistent approach provides a more natural
way to block the undesirable proofs. But there are other reasons to entertain
a paraconsistent logic. There are many situations in mathematics where the
consistency of a theory is called into question but without a demonstration of
any inconsistency. Consider, for example, the earliest uses of complex numbers,
numbers of the form x + yi, where i  1 and x and y are real numbers. There
was a great deal of debate about whether it was inconsistent or just weird to
entertain the square root of negative numbers.14 Moreover, it was not just the
status of complex analysis that was at issue. If the theory of complex analysis
turned out to be inconsistent, everything that depended on it, such as some
important results in real analysis, would also be in jeopardy. Adopting a paraconsistent logic is a kind of insurance policy: it stops the rot from spreading
too swiftly and too farwhether or not you know about the rot.
Perhaps the most interesting reason to entertain a paraconsistent logic in
mathematics is that with such a logic in hand, nave set theory and nave
innitesimal calculus can be rescued.15 There is no need to adopt their more
mathematically sophisticated replacements: ZFC and modern calculus. There
are a couple of pay-os here. First, both nave set theory and nave innitesimal
calculus are easier to teach and learn than their modern successors. In nave
set theory there is no need to deal with complicated axioms designed to block
the paradoxes; the easily understood and intuitive unrestricted comprehension
is allowed to stand. With nave calculus there is no need to concern oneself
with the subtle modern () denition of limit; innitesimals are allowed
back in the picture.16 The second pay-o is related to the rst and concerns
the intuitiveness of the theories in question. At least in the case of set theory,
the nave theory is more intuitive. ZFC, for all its great power and acceptance,
remains unintuitive and even ad hoc. There is no doubt that nave set theory is
the more natural theory. Similar claims could be advanced in relation to nave
innitesimal calculus over modern calculus, though the case is not as strong
here.

14 See M. Kline, 1972, for some of the relevant history of this debate.
15 C. Mortensen, 1995.
16 As they are in non-standard analysis, but non-standard analysis is also rather dicult to teach
and learn.

36

Mark Colyvan

4. Applying Inconsistent Mathematics


I now turn to application of inconsistent mathematics. There are many interesting issues here, and Ill say just a little about a few of these. The rst issue
is that inconsistent mathematics adds a new twist to an old problem known
as the unreasonable eectiveness of mathematics.17 The puzzle is to explain
how an a priori discipline like mathematics can nd applications in a posteriori
science. As Mark Steiner puts it:
[H]ow does the mathematiciancloser to the artist than the explorerby
turning away from nature, arrive at its most appropriate descriptions?18

This problem has attracted the attention of physicists and mathematicians,


but few philosophers have been drawn to it. Part of the reason for this is that
several of the philosophers who have written on the problem seem to think
that something like the following holds, and is all thats required in order to
explain the puzzle.
Mathematicians develop structures, often motivated by, or at least inspired by,
physical structures. The mathematicians structures then (unsurprisingly) turn
out to be similar (or even isomorphic) to various physical structures.19

But the fact that inconsistent mathematics, such as the early calculus, nds
wide and varied applications in empirical science, raises problems for this line
of thought. After all, assuming, as most of us do, that the world is consistent,
how can an inconsistent mathematical theory be similar in structure to something thats consistent? There is a serious mismatch here. It certainly cannot
be that the inconsistent mathematics in question is isomorphic to the world,
unless one is prepared to countenance the possibility that the world itself is
inconsistent. Im not suggesting that the above thought about how to dissolve
17 See the original paper on this, E. P. Wigner, The Unreasonable Eectiveness of Mathematics
in the Natural Sciences, Communications on Pure and Applied Mathematics 1960, 13: 14, as
well as M. Colyvan, The Miracle of Applied Mathematics, Synthese 2001, 127: 265278; M.
Colyvan, Mathematics and the World, in A. D. Irvine (ed.), Handbook of the Philosophy of
Science Volume 9: Philosophy of Mathematics, North Holland forthcoming; and M. Steiner,
The Applicability of Mathematics as a Philosophical Problem, Cambridge MA 1998.
18 M. Steiner, The Applicability of Mathematics, Philosophia Mathematica 1995, 3:129156,
see p. 154.
19 See, for example, M. Balaguer, Platonism and Anti-Platonism in Mathematics, New York
1998, pp. 142144, and P. Maddy, Second Philosophy: A Naturalistic Method, Oxford 2007,
pp. 329343, for views along these lines.

Whos Afraid of Inconsistent Mathematics?

the puzzle of the unreasonable eectiveness of mathematics is completely o


the mark, just that it cannot be the whole story.20
The second issue in relation to applying inconsistent mathematics takes us
into metaphysics. There is a much-discussed argument in the philosophy of
mathematics known as the indispensability argument. This is an argument for
belief in the reality of mathematical objectsPlatonismfrom the fact that
mathematical theories are indispensable to our best scientic theories.21 According to this line of thought, we should be committed to the existence of
all and only the entities that are indispensable to our best scientic theories
and, as it turns out, mathematical entities are indispensable to these theories.
This leads to the conclusion that we ought to believe in the existence of mathematical entities, along with electrons, dark matter, pulsars and other entities
indispensable to our best scientic theories. Again, applications of inconsistent
mathematics adds a new twist. There have been times when inconsistent mathematical theories (most notably the early calculus) have been indispensable to a
broad range of scientic theories. 17th and 18th century calculus was indispensable to mechanics, electromagnetic theory, gravitational theory, heat conduction and the list goes on. It seems that if one subscribes to the indispensability
argument (as I do) then theres a rather unpalatable conclusion beckoning:
sometimes we ought to believe in the existence of inconsistent objects.22
It is not clear what to make of this argument for the existence of inconsistent
objects. Is it a reductio of the original indispensability argument? Does it tell us
that consistency should be an overriding constraint in such matters? If so, on
what grounds? Perhaps it is not as crazy as it sounds to believe in inconsistent
mathematical objects. It is fair to say that the jury is still out on these issues,
with much more work and detailed examination of case studies required before
a sensible verdict can be delivered.
20 It is also worth noting that sometimes, when there is concern over the consistency of a
mathematical theory (such as the early use of complex numbers), condence in the theory
increased when the theory was found to enjoy widespread applications.
21 See M. Colyvan, The Indispensability of Mathematics, New York 2001; M. Colyvan,
Indispensibility Arguments in the Philosophy of Mathematics, in E. N. Zalta (ed.), The
Stanford Encyclopedia of Philosophy, (Spring 2008 edition, forthcoming), URL=<http://plato.
stanford.edu/archives/spr2008/entries/mathphil-indis/>; H. Putnam, Philosophy of Logic,
New York 1971; and W. V. Quine, Success and Limits of Matematization, in Theories and
Things, Cambridge 1981 for details
22 M. Colyvan, The Ontological Commitment of Inconsistent Theories, Philosophical
Studies, forthcoming; and C. Mortensen, Inconsistent Mathematics: Some Philosophical
Implications, in A. D. Irvine (ed.), Handbook of the Philosophy of Science Volume 9: Philosophy
of Mathematics, North Holland, forthcoming.

38

Mark Colyvan

Finally, there has been some very interesting work on using inconsistent
mathematical theoriesmore specically, inconsistent geometryto model
inconsistent pictures such as those of M. C. Escher and Oscar Reutersvaard
(e.g., Eschers Belvedere). Chris Mortensen23 has argued convincingly that
consistent mathematical theories24 of such pictures do no do justice to the
cognitive dissonance associated with seeing such pictures as impossible. Arguably, the dissonance arises from the perceiver of such a picture constructing
an inconsistent mental model of the situationan impossible spatial geometry. Any consistent mathematical representation of this inconsistent cognitive
model will fail to capture its most important quality, namely its impossibility. Inconsistent mathematics, on the other hand, can faithfully represent the
inconsistent spatial geometry being contemplated by the perceiver and thus
serve as a useful tool in exploring such phenomena further. These applications of inconsistent mathematics should hold interest beyond philosophy.
Indeed there are immediate applications in cognitive science and psychology.
But such work is very new and the full import of it has not yet been properly
appreciated.25

5. Conclusion
Inconsistent mathematics has received very little attention in mainstream philosophy of mathematics and yet, as I have argued here, there are several interesting philosophical issues raised by it. Moreover, some of these issuessuch as
the ontological commitments of inconsistent mathematical theories and the use
of paraconsistent logic as the logic for mathematicsbear directly on contemporary debates in philosophy of mathematics. Other issuessuch as the application of inconsistent mathematics to model inconsistent picturespromise
to take philosophy of mathematics in new and fruitful directions. For my
money, though, the biggest issue concerns possible insights into the relation23 C. Mortensen, Peeking at the Impossible, Notre Dame Journal of Formal Logic, 38(4):
527534; C. Mortensen, Inconsistent Mathematics, in E. N. Zalta (ed.), The Stanford
Encyclopedia of Philosophy, (Fall 2004 edition), URL=<http://plato.stanford.edu/archives/
fall/2004/entries/mathematics-inconsistent/>; and C. Mortensen forthcoming.
24 Such as in L. S. Penrose and R. Penrose, Impossible Objects, a Special Kind of Illusion,
British Journal of Psychology 1958, 49: 3133; and R. Penrose, On the Cohomology of
Impossible Pictures, Structural Topology 1991, 17: 1116.
25 Although see C. Mortensen forthcoming.

Whos Afraid of Inconsistent Mathematics?

ship between mathematics and the world. This is a central problem for both
philosophy of mathematics and philosophy of science.
I believe that there is a great deal to be learned about the role of mathematical modelsboth consistent and inconsistentin scientic theories, by paying closer attention to the use of inconsistent mathematics in applications.
Perhaps focussing our attention on the consistent mathematical theories has
misled us to some extent. If this is right, we wont have the complete picture
of the mathematicsworld relationship until we understand how inconsistent
mathematics can be so useful in scientic applications.26

26 Id like to thank Stephen Gaukroger and Audrey Yap for helpful conversations on the history
of the calculus, and Adam La Caze and Fabien Medvecky for comments on an earlier draft.
I have also beneted from several conversations with Chris Mortensen about inconsistent
mathematics. Work on this paper was funded by an Australian Research Council Discovery
Grant (grant number DP0209896).

40

Andrew Arana

Logical and Semantic Puritiy


Andrew Arana

Abstract
I distinguish two dierent views on what makes a proof of a theorem pure, rstly by characterizing them abstractly, and secondly by showing that in practice the views dier on what
proofs qualify as pure.

Many mathematicians have sought pure proofs of theorems. There are dierent takes on what a pure proof is, though, and its important to be clear on
their dierences, because they can easily be conated. In this paper I want to
distinguish between two of them.
I want to begin with a classical formulation of purity, due to Hilbert:
In modern mathematics one strives to preserve the purity of the method, i.e.
to use in the proof of a theorem as far as possible only those auxiliary means
that are required by the content of the theorem.1

A pure proof of a theorem, then, is one that draws only on what is required
by the content of the theorem.
I want to continue by distinguishing two ways of understanding required by
the content of [a] theorem, and hence of understanding what counts as a pure
proof of a theorem. Ill then provide three examples that I think show how these
two understandings of content-requirement, and thus of purity, diverge.

1. Logical purity
The rst way of understanding purity that I want to consider takes what is required by the content of [a] theorem to be just what suces for proving that
1

Translation in [25], pp. 3934. The original ([16], pp. 3156) reads, In der modernen Mathematik wird solche Kritik sehr hug gebt, wobei das Bestreben ist, die Reinheit der Methode
zu wahren, d.h. beim Beweise eines Satzes wo mglich nur solche Hlfsmittel zu benutzen,
die durch den Inhalt des Satzes nahe gelegt sind. Hilbert continues by remarking that
Dieses Bestreben ist oft erfolgreich und fr den Fortschritt der Wissenschaft fruchtbar
gewesen. Hilbert seems to have had in mind recent work on circle quadrature and the parallel
postulate.

Logical and Semantic Puritiy

theorem. The ideal is what Hilbert pursued in his Grundlagen der Geometrie: to
determine which of the axioms he gave for geometry are sucient for proving
interesting geometric theorems, such that if any of those axioms were left out,
the theorem would no longer follow.2 As an rst approximation, then, this ideal
can be made more precise by dening a set of axioms S as logically minimal
for a theorem P just in case S proves P, but no proper subset of S proves P; and
a proof of a theorem P as logically pure if it is a proof of P from a logically
minimal subset S of a set of axioms T. This is only an approximation, because
if we allow as a set of axioms, say, the conjunction of Hilberts axioms of geometry, then every theorem provable from Hilberts axioms has a logically pure
proof from that single conjoined axiom. To avoid this trivialization, wed need
to restrict our attention to the sorts of axiomatic theories that arise in ordinary
practice. What I have in mind are ordinary examples like Hilberts axioms for
geometry, or the Peano axioms for arithmetic. We dont at present have a convincing way to characterize completely non-trivial axiomatic theories. So Ill
just leave this as an approximation, but one that I take is clear enough.
For a theorem P and a set of axioms T, there may be several dierent logically
pure proofs of P, since there may be several logically minimal subsets of T for
P. Furthermore, for a given theorem there may be several good candidates for
axiom sets T relative to which we can search for logically pure proofs.3
One way to pinpoint what suces for proving a given theorem is to nd a
set of axioms that is logically equivalent to that theorem (over a logically weak
base theory). This is what is done in reverse mathematics as developed by
Harvey Friedman and Stephen Simpson: starting with a mathematical theorem
and an interesting collection of set-theoretic axiom sets, we try to determine
which of these axiom sets is logically equivalent to that theorem (over a base
theory that is set-theoretically weaker than the theorem and axiom sets under
consideration).4 When successfully carried out, reverse mathematics locates
both necessary and sucient conditions for a given theorem, and thus locates
the logically weakest axiom set (among a given set of candidate axiom sets) for
proving a given theorem. Lets call a proof from such a set of axioms strongly
logically pure.
In this paper Im just concerned with logical purity, not strong logical purity.
Advocates of logical purity have not always been clear enough about which
2
3
4

For more on Hilberts interest in purity in his geometric work, see [13].
Cf. [26], p. 20, for some discussion of this point.
Cf. [10], [30]. The name reverse is due to the part of the project in which we prove the
axiom set from the theorem.

42

Andrew Arana

of the two projects theyre pursuing. For instance, Pambuccian writes that in
geometry one would want to know what axioms are needed to prove a particular theorem, an enterprise that might be called reverse geometry.5 Aligning
his work with reverse mathematics suggests that he seeks strong logical purity,
rather than merely logical purity. This suggestion is supported by his citing the
following passage of Hilberts as articulating his view on purity:
By the axiomatic study of mathematical truth I understand an investigation
which does not aim to discover new or more general theorems with the help of
given truths, but rather the position of a theorem within the system of known
truths and their logical connections in a way that indicates clearly which conditions are necessary and sucient for the grounding of that truth.6

However, I think the details of Pambuccians work bears out that hes primarily
interested in logical purity, rather than strong logical purity. (I also dont think
its clear that Hilbert was interested in logical purity as opposed to another type
of purity, but Ill return to this in the next section.)

2. Semantic purity
In contrast to the logical reading of required by the content of [a] theorem considered in the last section, consider the following semantic reading:
namely, whatever must be understood or accepted in order to understand that
theorem.7 These concepts and truths are the conditions for understanding the
theorem, and so are part of its content. A proof of a theorem, then, is semantically pure, if it draws only on what must be understood or accepted in order
to understand that theorem.
Its dicult to say precisely what must be understood and accepted in order
to understand a given theorem. Its easier to focus on specic cases in order to
5
6

Cf. [25], p. 393. For more on reverse geometry, see also [26], p. 19.
Cf. [15], p. 50. My translation. The original reads, Unter der axiomatischen Erforschung
einer mathematischen Wahrheit verstehe ich eine Untersuchung, welche nicht dahin zieht,
im Zusammenhange mit jener Wahrheit neue oder allgemeinere Stze zu entdecken, sondern
die vielmehr die Stellung jenes Satzes innerhalb des Systems der bekannten Wahrheiten und
ihren logischen Zusammenhang inder Weise klarzulegen sucht, da sich sicher angeben lt,
welche Voraussetzungen zur Begrndung jener Wahrheit notwendig und hinreichend sind.
Pambuccian cites this passage in [26], p. 19.
For more on this semantic reading, see [1], pp. 46.

Logical and Semantic Puritiy

see how this is supposed to work. Ill focus here on H.S.M. Coxeters work on
Sylvesters problem, which says:
Let n given points have the property that the straight line joining any two of
them passes through a third of the given points. Show that the n points lie
on a straight line.8

What must be understood and accepted in order to understand Sylvesters


problem? The surface grammar of the problem concerns points, straight lines,
and the incidence of points on straight lines. So we must understand and accept
denitions, including axioms, of these concepts. Coxeter thought a line segment is, by denition, the set of points between two points.9 So to understand
Sylvesters problem, Coxeter thought it was necessary to understand and accept
incidence and betweenness axioms for geometry; i.e., the theory he called ordered geometry.10 Hence a solution to Sylvesters problem in ordered geometry
would be semantically pure; and indeed Coxeter found such a solution.11
Coxeters understanding of straight line isnt the only possible one, of course.
Dierential geometers sometimes dene straight line as the shortest line between two given points. Consequently, for them distance, and not betweenness, must be understood to understand Sylvesters problem. I dont know a
plausible way of establishing that Coxeter is right and the dierential geometers
are wrong. We dont need to do this for our purposes here, though. Its enough
to note that Coxeters understanding makes explicit a single communitys un8 This problem was originally posed by J.J. Sylvester in [31], hence the name. Ive given Erds
formulation from [9], p. 65.
9 Coxeter writes ([4], p. 176), The essential idea [for problems such as Sylvesters] is intermediacy (or betweenness), which Euclid used in his famous denition: A line (segment) is that
which lies evenly between its ends. This suggests the possibility of regarding intermediacy as a
primitive concept and using it to dene a line segment as the set of all points between two
given points. I want to note that in addition to relying uncharacteristically on the authority
of Euclid, Coxeter uses an unusual translation of Euclids denition I.4. For instance, Heaths
translation reads, A straight line is a line which lies evenly with the points on itself.
10 Coxeter presents the axioms of ordered geometry, adapted from Pasch and Veblens earlier
treatments, in [4], pp. 1778. In particular, Coxeters presentation follows Veblens in [32],
with some minor rearrangement. Veblen notes that his axioms presuppose only the validity
of the operations of logic and of counting (ordinal number) (p. 344). Why must logic and
principles governing the nite ordinal arithmetic of counting be understood and accepted in
order to understand Sylvesters problem? I think Veblen and Coxeter would say that anyone
who failed to understand and accept these principles would not completely understand any
mathematical problems. I agree with this point of viewthough teasing out exactly what
logical principles are involved here would be challenging.
11 The proof is given at [4], pp. 1812.

44

Andrew Arana

derstanding, and to focus on semantic purity relative to that communitys


understanding.
While Pambuccian gives a logical reading of purity as formulated by Hilbert,
I want to make clear that the semantic reading is also consistent with Hilberts
formulation. The semantic purist believes that she seeks proofs of theorems
that are restricted to what is required by the content of the theorem. She
agrees with Hilbert that in seeking a pure proof of a given theorem, she is
seeking conditions [that] are necessary and sucient for the grounding of
that truth. She just seeks a semantic grounding relation, rather than a purely
logical one.

3. Distinguishing logical from semantic purity


One question Im not going to address here is why we should value logically
pure proofs of theorems over logically impure proofs. Nor will I address here
the value of semantic purity. These are important questions, but my goal in
this paper is simply to distinguish sharply logical purity from semantic purity.
Ill do so demonstrating two theses:
Some results require more concepts and/or proposition to be proved than
to be understood.
Some results require more concepts and/or proposition to be understood
than to be proved.
Ill demonstrate these by presenting three case studies.

3.1 The casus irreducibilis


Consider the problem of nding exact solutions to cubic polynomial equations
with rational coecients that have three real roots, e.g. x3x = 0, since x3x =
x(x + 1)(x1). What must we understand and accept to understand this problem?
Rational polynomials are built up using the six algebraic operations, namely,
addition, subtraction, multiplication, division, exponentiation, and extraction
of roots. So we must understand those, and accept the usual algebraic laws
thought to govern these operations (e.g. commutativity, distributivity) as denitions of those operations. In addition, we must understand how to carry out
these operations on rational numbers. But we neednt understand how to carry

Logical and Semantic Puritiy

out these operations on all rational numbers. We dont need to understand how
to divide by zero, for instance. We also neednt understand how to extract the
square root of a negative number: as evidence for this I point out that the early
workers on this problem (e.g. Cardano, Bombelli) didnt understand how to
do this, or what it would mean to nd the square root of a negative number;
and yet they understood the problem of exact cubic solution.
Ill next explain whats needed to solve, rather than just understand, this problem. Firstly, note that we can restrict our attention to cubics of the form x3 = qx
+ r (such as x3 = 15x + 4), since every cubic can be put in this form by a change
of variables.12 We can then use Cardanos formula to solve these cubics:
x3

q 1
q 1

3D 3
3D
2 18
2 18

where D is the discriminant of the cubic.13


To understand Cardanos formula, we need to understand only what we
need to understand in order to understand the problem of solving cubics with
rational coecients and three real roots. It would be usable in a semantically
pure solutionexcept for the following problem. It turns out that (x1x2)(x1x3)
(x2 x3) = D . So when all three roots x1, x2, x3 are real, it follows that D > 0.
But then to nd a solution using Cardanos formula, we will have to evaluate 3D , the square root of a negative number and hence imaginary. In
applying Cardanos formula to solving cubics with three real roots, we must
use imaginary numbers along the way. This case has historically been known
as the casus irreducibilis, or irreducible case, because of the diculty in
showing that these imaginary terms reduce to real terms. Because of the casus
irreducibilis, Cardanos formula isnt usable in a semantically pure solution. Its
use in the casus irreducibilis involves imaginary numbers, and these neednt be
understood to understand our problem.14
12 That is, all cubics of the general form z3 + a1z2 + a2z + a3 = 0 can be put in the form x3 = qx +
r using the substitution z = x a1 / 3.
13 For cubics in our restricted form, the discriminant D is 4q3 27r2. Cardanos formula will
locate one of the three real roots; we can use the quadratic formula to nd the other two.
14 One could object that complex numbers are just ordered pairs of real numbers, and so ought
to be usable in a semantically pure solution to cubics in the casus irreducibilis. This objection misses the point that to dene complex numbers as pairs of reals, one must dene the
algebraic operations for complex numbers in terms of the algebra of ordered pairs of reals.
But these peculiar denitions on ordered pairs of reals exceed what must be understood and
accepted in order to understand ordered pairs of reals.

46

Andrew Arana

After the discovery of the casus irreducibilis in the sixteenth century, mathematicians found other methods for solving cubics. These include the geometric solutions of the seventeenth century15 and the innite series solutions of the
eighteenth century.16 Mathematicians began to wonder if there is a semantically
pure solution.17 In 1892, Otto Hlder answered this question, using Galois
theory to show that there can be no solution to cubics in the casus irreducibilis
that avoids using complex numbers, while at the same time using just nitely
many instances of the six algebraic operations.18
Hlders result demonstrates that to solve cubics in the casus irreducibilis
exactly, imaginary numbers must be used. Since complex numbers neednt be
understood in order to understand this problem, we have an example where
15 Vite and Descartes were able to avoid the use of complex numbers by constructing the solutions to these cubics as the lengths of sides of triangles determined by the cubics. Vites work
on the casus irreducibilis is located in two places: his 1593 text Supplementum Geometriae, and
his 1615 text De Aequationum Recognitione et Emendatione Tractatus Duo. Both are available in
translation in [33]. Descartes work on the casus irreducibilis is in Book III of La Gomtrie
([7], pp. 21516). This trigonometric construction can be expressed as an equation, by making use of the non-algebraic trigonometric operations cos and arccos; though, interestingly,
this formula requires complex numbers in cases where Cardanos solution avoids them. This
geometric method isnt semantically pure, as it requires understanding either geometry or
non-algebraic operations.
16 Newton and Leibniz both attempted to use innite series to avoid complex algebra for
solving cubics in the casus irreducibilis, but the idea came to full fruition a little later in
work of Franois Nicole and Alexis Clairaut. For Newtons work on this, see his letter to
Collins, dated 6/20/1674, in [29]. For Leibniz work, see a 1675 (approx.) letter to Wallis,
in [22]. Nicoles work may be found in [23] and [24], while Clairauts extension of Nicoles
work may be found in [3]. Nicoles papers contain the essential details, applying Newtons
binomial theorem to Cardanos equation, except that he missed a few details about the
conditions under which binomial series converge. Clairaut corrected those mistakes in his
1746 algebra textbook. This method isnt semantically pure, since it requires understanding
innite series.
17 For instance, Lagrange wrote that the irreducible case of equations of the third degreeis
constantly giving rise to unprotable inquiries with a view to reducing the imaginary form
to a real form andpresents in algebra a problem which may be placed upon the same footing with the famous problems of the duplication of the cube and the squaring of the circle
in geometry. ([21], p. 62) Proof of the impossibility of doubling the cube and squaring the
circle using just straightedge and compass was still thirty years away at the time Lagrange
gave his lectures, but their impossibility was generally accepted in his time as fact. The passage suggests that a similar attitude had taken hold concerning the casus irreducibilis.
18 For Hlders proof, see [17]; see also Hlders commentary in [18]. Hlders impossibility
theorem may be stated precisely as follows: if a cubic equation x3 + qx + r = 0 has three real,
unequal roots and is irreducible over the eld F = Q(q, r), then it is not solvable by real
radicals.

Logical and Semantic Puritiy

what is needed to solve a problem exceeds what is needed to understand it.


This demonstrates Thesis 1, that some results require more concepts and/or
proposition to be proved than to be understood.

3.2. The innitude of primes and fragments of arithmetic


Though it sounds strange, some results can be proved using fewer resources
than are needed to understand it. I want to consider one example of this from
arithmetic.
The innitude of primes (IP) asserts that for all natural numbers a, there exists
a natural number b > a such that b is prime. It was proved by Euclid in Elements
Book IX, Proposition 20. To understand IP, we must understand what a natural
number is. Its been argued that to do so, we must understand and accept the second-order induction axiom X [0 X y( y X S ( y ) X ) x ( x X )] ,
where S is the successor function and X ranges over all subsets of the natural
numbers.19 This is controversial, but lets temporarily grant the claim and focus on its consequences. IP can be stated formally in the language of rstorder arithmetic, and Euclids proof can be carried out in rst-order Peano
Arithmetic (PA) with only minor modications.20 But PA is weaker than second-order arithmetic: instead of an induction axiom for every subset of the
natural numbers, it includes countably many induction axioms of the form
y [(0, y ) x (( x , y ) (S ( x ), y )) x( x , y )] , for each formula ( x , y )
in the language of rst-order arithmetic. That is, PA includes induction axioms
for each subset of the natural numbers that is denable by a rst-order formula,
but there are only countably many of these, and there are uncountably many
subsets of the natural numbers. So PA requires less to be understood and accepted than IP does. Yet IP can be proved in PA. So less is needed to prove IP
than to understand it.
The argument I just gave depended on the claim that the second-order induction axiom must be understood and accepted in order to understand what a
natural number is. Lets drop that claim, and suppose instead that understanding and accepting rst-order PA is necessary and sucient for understanding
what a natural number is. But IP can be proved in logically weak fragments
of PA, specically by weakening the induction schema. It is straightforward
to check that the Euclidean proof, when formalized in PA, uses the induction
19
20

Cf. [19], pp. 2046, and [20], p. 42, for example.


The formalized innitude of primes is ab[b > a x[y( x y = b ) ( x = 1 x = b )]].

48

Andrew Arana

schema for formulas of complexity at most 1. That is, the proof can be carried
out in I1, which is PA with the induction schema restricted to 1 formulas.
On the other hand, the Euclidean proof cannot be carried out in I0, which
is PA with the induction schema restricted to formulas with just bounded
quantiers.21 It is open whether IP can be proved in I0.22 However it is known
that the Euclidean proof can be carried out using bounded induction provided
that we add another axiom asserting the totality of the exponential relation,
resulting in a theory called I0(exp).23 So IP can be proved in I0(exp), but
understanding and accepting I0(exp) isnt sucient for understanding what
a natural number is. Once again, we have that less is needed to prove IP than
to understand it.
Thus, in either case we have demonstrated Thesis 2, that some results require
more concepts and/or proposition to be understood than to be proved.

3.3. Gdel sentences


Finally, I want to consider the case of Gdel sentences. As is well-known, there
are sentences expressed in the language of PA that are unprovable in PA, but
21 This is because in the Euclidean proof, we suppose that we have a nite enumeration
p1 , p2 , , pn of all the primes, and generate the quantity Q  ( p1 p2  pn ) 1 towards
showing that there is another prime not on this list. But to show that Q exists, we need that
multiplication is total, and the usual proof of this uses 1-induction. How much induction is
needed to prove this? That is an open question. However, it is known that in I0 it is unprovable that every product of primes exists (cf. [5], p. 13). This follows from Parikhs result (in
[27]) that every 0-denable function that is provably total in I0 has polynomial growth.
Cf. [6] pp. 1647 for more on what is known concerning the rate of growth of the function
yielding products of primes in I0.
22 In his dissertation, A. Woods [34] was able to solve IP in I0 together with a weak version of
the pigeonhole principle. He did not present a modied version of the Euclidean resolution,
but instead, a modied version of a solution due to Sylvester. Woods theory, called I0+
PHP , is logically weaker than I0(exp), in that I0(exp) proves I0+ PHP but not vice-versa
(cf. [28]; also [6], pp. 1624). Later Paris, Wilkie, and Woods replaced Woods earlier proof
with one using an even weaker version of the pigeonhole principle (cf. [28]; also [6], pp.
1624).
23 Cf. [6], p. 153. I0(exp) is sometimes studied under the name EFA, for Elementary Function Arithmetic (cf. [11]) or EA, for Elementary Arithmetic (cf. [2]). In [2] Avigad explains
Harvey Friedmans conjecture that every result in elementary number theory (for instance,
every result in Hardy and Wright [14], a canonical elementary text in number theory) can
be proved in I0(exp)(cf. [6], p. 149n1). Due to a result of Gdel [12], we know that the
exponential relation is denable in (N, 1, S, +, x). For a reasonably explicit denition of this
type, see [8], pp. 2769.

Logical and Semantic Puritiy

are provable in stronger formal systems (such as ZFC). In virtue of being expressed in the language of PA, Gdel sentences have arithmetical content, and
as a result, a grasp of the arithmetical denitions provided by the axioms of
PA suces for understanding these sentences. So Gdel sentences are another
example where more is needed for proof than understanding.
I want to consider two objections to this. The rst is based on a view of
Daniel Isaacson. Isaacson writes that the only way to see the arithmetical truth
of the Gdel sentence is to see it as having coded metamathematical content,
i.e. to see that it says of itself that it isnt provable in whichever axiom system
is being considered. I have my qualms about this point24, but lets grant it for
the sake of argument. Isaacson argues that this shows that the Gdel sentence
doesnt have arithmetical content, and so to be understood requires understanding and accepting higher-order truths, such as those of set theory. This is
because he believes that the type of content a sentence has depends on what
must be grasped to perceive that that sentence is true. As he writes:
[A] truth expressed in the (rst-order) language of arithmetic is arithmetical
just in case its truth is directly perceivable on the basis of our (higher-order)
articulation of our grasp of the structure of the natural numbers or directly
arithmetical.25
Isaacson explains that by the grasp of the structure of the natural numbers
clause he has in mind axioms, while by the second clause he has in mind
theorems.
If Isaacson were correct, it would undermine my claim that more is needed
to prove Gdel sentences than to understand them. I dont think he is correct, though, as I want to show by two dierent replies. Firstly, Isaacsons
view renders obviously arithmetic sentences like the Goldbach conjecture unarithmetical, since there is at present no reason to believe its truth (or the truth
of its negation, should it turn out to be false) is like an axiom in being directly
perceivable just from our grasp of the structure of the natural numbers, and
24 The Gdel sentence for PA can be proved in ZFC. Its plausible that a person who knew
nothing about metamathematics, but had a command of set theory, would encounter the
Gdel sentence for PA, but would not recognize it as such. She could then prove the Gdel
sentence without seeing it as having coded metamathematical content.
Isaacson moderates his view later in the paper, saying that Gdel sentences can and must
be shown to be true by an argument in terms of truths concerning some higher-order notions, including essentially set-theoretical principles (pp. 2201). But once again he cites
the relationship of coding as the rigid link between the arithmetical and the higher-order
truths, and I dont see why we should believe this, as I explained above.
25 Cf. [19], p. 217.

50

Andrew Arana

we know of no proof of it from any truths at all at present, arithmetic or not. I


dont think we should accept a view that implies that the Goldbach conjecture
is un-arithmetical. Secondly, its true that we cant see that the Gdel sentence
is a Gdel sentence without grasping its coded metamathematical content;
but we can grasp it as a universally quantied sentence in the language of PA,
without seeing that its a Gdel sentence. Its true that our reason for interest in
the Gdel sentence may be that it is a Gdel sentence, rather than because we
encountered it in ordinary arithmetical work, but we shouldnt conclude from
that contingency that Gdel sentences arent arithmetical. After all, we still
could encounter sentences independent from PA in the course of future work in
mainstream number theory, without knowing beforehand that these sentences
are independent of PA. The arithmetic character of a problem is independent
of the reasons we have for choosing to solve that problem.
The second objection to my claim that Gdel sentences require more for
proof than for understanding, is the following: Gdel sentences expressed in
the language of PA are, practically speaking, unintelligible because they contain
so many symbols; but when expressed in higher-order terms (such as in set
theory), they become intelligible.26 So, practically speaking, grasp of the axioms
of PA is not sucient for understanding Gdel sentences.
I dont think we should be troubled by this objection either. For this observation holds not just for Gdel sentences, but for many obviously number-theoretic sentences, for instance those involving exponentiation (such as Fermats
last theorem). It is possible to express exponentiation in the language of PA,
using the Gdel -function, but the sentences resulting from the subsequent
substitution will be quite long and complex, compared to the simple sentence
with which we began. So if the gain in intelligibility resulting from using higher-order terms tells against a sentence expressible in the language of PA being
arithmetic, then we will have to conclude that many statements of elementary
number theory will also fail to be arithmetic. Since I think this is implausible,
I reject this second objection.
I conclude that the case of Gdel sentences demonstrate Thesis 1, that some
results require more concepts and/or proposition to be proved than to be understood.
Thanks to Paolo Mancosu and Doug Patterson.
26 Isaacson makes a related point, using the observation that higher-order notions can be
essential for shortening an otherwise unsurveyable proof. (p. 221)

Logical and Semantic Puritiy

References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]

Andrew Arana. On formally measuring and eliminating extraneous notions


in proofs. Philosophia Mathematica, 2008. Forthcoming.
Jeremy Avigad. Number theory and elementary arithmetic. Philosophia Mathematica, 11:257284, 2003.
A.C. Clairaut. Elemens dAlgebre. Guerin, Paris, 1746.
H. S. M. Coxeter. Introduction to Geometry. Wiley, second edition, 1989.
Paola DAquino. Local behaviour of the Chebyshev theorem in models of I0.
J. Symbolic Logic, 57(1):1227, 1992.
Paola DAquino. Weak fragments of Peano arithmetic. In The Notre Dame
Lectures, volume 18 of Lecture Notes In Logic, pages 149185. Association for
Symbolic Logic, Urbana, IL, 2005.
Ren Descartes. La Gomtrie. Dover Publications, 1954. Translated by D.E.
Smith and M.L. Latham, 1925.
Herbert B. Enderton. A mathematical introduction to logic. Harcourt/Academic Press, Burlington, MA, second edition, 2001.
Paul Erds. Problem 4065. American Mathematical Monthly, 50(1):65, January 1943.
Harvey M. Friedman. Systems of second order arithmetic with restricted
induction. I. Journal of Symbolic Logic, 41(2):5578, June 1976.
Harvey M. Friedman. A Strong Conservative Extension of Peano Arithmetic.
In H. J. Keisler J. Barwise and K. Kunen, editors, Proceedings of the 1978 Kleene
Symposium, pages 113122. North-Holland, 1980.
Kurt Gdel. Uber formal unentscheidhare Stze der Principia Mathematica
und verwandter Systeme I. In Solomon Feferman et. al., editor, Collected
Works, volume 1, pages 145195. Oxford University Press, 1986.
Michael Hallett. Reections on the Purity of Method in Hilberts Grundlagen der Geometrie. In Paolo Mancosu, editor, The Philosophy of Mathematical
Practice. Oxford University Press, 2008.
G. H. Hardy and E. M. Wright. An introduction to the theory of numbers.
Oxford University Press, New York, fth edition, 1979.
David Hilbert. ber den Satz von der Gleichheit der Basiswinkel im gleichschenkligen Dreieck. Proceedings of the London Mathematical Society,
35:5068, 1902/03.
David Hilbert. Lectures on the foundations of geometry, 189 11902, volume 1.
Springer, 2004.
Otto Hlder. ber den Casus Irreducibilis bei der Gleichung dritten Grades.
Mathematische Annalen., 38:307312, 1892.
Otto Hlder. Encyklopdie der mathematischen Wissenschaften. Band I. Algebra
und Zahlentheorie. Teil 1. B. Algebra. Heft 3. Teil 27. Casus irreducibilis der
kubischen Gleichung. W. F. Meyer, Leipzig, 18981904.
Daniel Isaacson. Arithmetical truth and hidden higher-order concepts. In
W.D. Hart, editor, The Philosophy of Mathematics, pages 203224. Oxford

52

Andrew Arana

University Press, 1996.


[20] Richard Kaye. Models of Peano Arithmetic. Oxford University Press, Oxford,
1991.
[21] Joseph-Louis Lagrange. Lectures on elementary mathematics. Open Court,
Chicago, 1898. Based on lectures given to LEcole Normale in 1795, and originally published in Journal de LEcole Polytechnique in 1812. Translated by T.J.
McCormack.
[22] Gottfried Wilhelm Leibniz. Letter to Wallis, 1675 (approx.). In Gothofredi
Guillelmi Leibnitii Opera Omnia. Apud Fratres de Tournes, Geneva, 1768.
[23] Franois Nicole. Sur le cas irrductible du troisime degr. Memoires de
LAcademie Royal, pages 97102, 1738.
[24] Franois Nicole. Sur les equations du troisime degr. Memoires de LAcademie
Royal, pages 244259, 1738.
[25] Victor Pambuccian. Fragments of Euclidean and hyperbolic geometry. Scientiae mathematicae Japonicae, 53(2):361400, March 2001.
[26] Victor Pambuccian. Euclidean geometry problems rephrased in terms of midpoints and pointreections. Elemente der Mathematik, 60:1924, 2005.
[27] Rohit Parikh. Existence and feasibility in arithmetic. J. Symbolic Logic,
36:494508, 1971.
[28] J. B. Paris, A. J. Wilkie, and A. R. Woods. Provability of the pigeonhole
principle and the existence of innitely many primes. J. Symbolic Logic,
53(4):12351244, 1988.
[29] Stephen Rigaud, editor. Correspondence of scientic men of the seventeenth century, including letters of Barrow, Flamstead, Wallis, and Newton, volume 2.
Oxford University Press, Oxford, 1841.
[30] Stephen G. Simpson. Subsystems of Second Order Arithmetic. Perspectives in
Mathematical Logic. Springer, 1999.
[31] J. J. Sylvester. Mathematical Question 11851. Educational Times, 59:98, 1893.
[32] Oswald Veblen. A system of axioms for geometry. Transactions of the American
Mathematical Society, 5(3):343384, July 1904.
[33] Franois Vite. The Analytic Art. Kent State University Press, 1983. Translated
by T. Witmer.
[34] Alan Woods. Some problems in logic and number theory and their connections.
PhD thesis, University of Manchester, 1981.

On Using Measuring Numbers according to Measuring Theories

On Using Measuring Numbers according to


Measuring Theories
Wilhelm K. Essler

Abstract
It was shown by Frege that four of the ve axioms of Peano can be regarded as analytical
truths; and it was shown by Russell that the remaining axiom cannot be regarded as being
analytically true or even as being analytically false, that this axiom thus is to be regarded as
a synthetic statement. In using the concept of apriority in the sense of Reichenbach, it can
be shown that this synthetic axiom is to be regarded as an apriorical truth within the usual
background theory of measuring theories, which are used not as generalizations of empirical
results but as not moreover provable preconditions of receiving measuring results and
of ordering these results. Furthermore, the systems of numbers, starting with the natural
numbers, are developed in a way such that the pre-rational numbers but not the rational
ones turn out to be those ones which are used in performing measurements according to
such theories, while the pre-real numbers but not the real ones then turn out to be
those ones which are used in using such measuring theories together with their background
theories for purely theoretical reasons.

As it was pointed out already by Reichenbach1, Kants concept syntheticapriorical may be understood in a relative way, but also in an absolute way:
(1) In its relative meaning, the sentence A is a synthetic-apriorical truth is
to be understood as a statement related to some xed case of application, i.e. in
the sense of A, being a synthetic judgement, is used in some given situation
of application as an apriorical truth.
(2) In its absolute meaning, the sentence A is a synthetic-apriorical truth is,
however, to be understood without such being related to any case of application, receiving its validity therefore not from some suitable case of applying a
given background theory in order to receive empirical results, but in the sense
of A, being synthetic truth, is provable by purely apriorical means, whichin
establishing its truthare furthermore proving its necessity.
Like Reichenbach, I am using these methodological instruments apriorical
und synthetic not in thelogically unmaintainablesense of (2), but in
thephilosophically very usefulsense of (1)2. And in this very sense of (1),
1
2

See Reichenbach (1920).


In fact, I used them in this way since 1966. And it was about 1990, when Andreas Kamlah

54

Wilhelm K. Essler

in dening the fundamental concepts of the ve Peano axioms in the FregeRussell-sense (= fr), four of them turn out to be analytically true sentences on
natural numbers:
Dffr-1 Some class F is a member of 0 if F is empty, i.e., i F itself does not
contain any object. Therefore, 0 is the set of those classes which are
empty.
Dffr-2 Some class F is a member of the successor of a natural number N
i, omitting some object out of F, the remaining class is a member
of N. Therefore, the successor of a natural number N is the set of
those classes for which, omitting resp. one object out of them, the
remaining classes are members of N.
Dffr-3 Some set N is member of the class of natural numbers i N is element of every set P containing 0 and, for every element N of P, also
its successor. Therefore, the class of natural numbers is the smallest
class of those sets P containing 0 and, for every element of P, also its
successor.
In using higher order logic, the following four theorems are derivable out of
these three denitions:3
Thfr-1
Thfr-2
Thfr-3
Thfr-5

0 is a natural number.
For every natural number, its successor is a natural number, too.
0 is dierent from each successor of a natural number.
The class of natural numbers is partor: subclassof every class
P containing 0 and to each member also its successor.

But the fourth axiom of Peano, stating the innity of the class of natural
numbers, i.e.:
Axfr-4 For every two dierent natural numbers, its resp. successors are
dierent, too.
is not deducible out of that three denitions; and it is therefore not an analyti-

informed me on its similarity to Reichenbachs usage in his maiden book (1920).


And when two people are seeing something independently from one another, than this thing
may be regarded as being not a mere dream but as something which should be dealt with.
See, e.g., Essler-Brendel (1993).

On Using Measuring Numbers according to Measuring Theories

cal statement in the Frege-Russell approach, but a synthetic one.4


Up to this point of that approach, it is not decidable, whether this synthetic
axiom Axfr-4 is either an apriorical or an aposterioricali.e.: an empirical
statement. Russell himself seemed to have regarded it as an empirical truth.
But taking it as an idealisationi.e. as an assertion which is not intended to
be tested by empirical means but as a judgement stated in order to establish a
measuring theory for some science in which mathematics is applied , then
without any diculties it may be regarded as an apriorical truth in the sense
of (1). The following considerations intend to show, how this is to be seen in
aspecic case of application, i.e.: in considering the background theory of
measuring extensive additive quantities, being regarded as a theory according to whose applications the attributes of objects are determined by using
empirical means.
Measuring some object x of a suitable eld F of the universe U involves a
magnitude g concerning which the value g(x) of this object is to be determined.
Historically, the rst axiomatic characterization of the abstract attributes of
such magnitudes was given by Hlder.5 These are his seven Axioms:
I. Wenn zwei Grssen a und b gegeben sind, so ist entweder a mit b
identisch (a = b, b = a), oder es ist a grsser als b, und b kleiner als
a (a > b, b < a), oder umgekehrt b grsser als a, und a kleiner als b;
diese drei Flle schliessen sich aus.
II. Zu jeder Grsse giebt es eine kleinere.
III. Zwei Grssen a und b, die auch identisch sein knnen, ergeben in
einer bestimmten Reihenfolge eine eindeutig bestimmte Summe a
+ b.
IV. a + b ist grsser als a und grsser als b.
V. Ist a < b, so giebt es ein x so, dass a + x = b, und ein y so, dass y + a
= b.
VI. Es ist stets (a + b) + c = a + (b) + c).
VII. Wenn alle Grssen in zwei Classen so eingetheilt sind, dass jede
Grsse einer und nur einer Classe zugewiesen ist, dass jede Classe
Grssen enthlt, und dass jede Grsse der ersten Classe kleiner ist als
4
5

Most probably, this was seen already by Frege. He therefore tried to create an innite universe
by applying ontologicaland in this sense: aprioricalmeans only.
See Hlder (1901). It was Suppes (1951), who discovered this important paper, which otherwise surely were unknown up to now to philosophers as well as to mathematicians. NB: A
predecessor of Hlder (1901) is Grassmann (1862).

56

Wilhelm K. Essler

jede Grsse der zweiten, so existirt eine Grsse derart, dass jedes
< zur ersten, und jedes > zur zweiten Classe gehrt. selbst
kann, je nach dem gegebenen Fall, zur einen oder zur anderen Classe
gehren.
Now it seems that Hlder, in formulating these axioms, was regarding (A)
sometimes the objects of the eldor classof applications, perhaps together
with ist values w.r.t. that quantity g, but (B) sometimes the equivalence classes
of these objects of F, i.e. the classes of objects obtainig the same value w.r.t
that given quantity. But we will deal here only with case (A), thereby dividing and weakening this quantity g into the pair of relations smaller-than and
equal-to.
Let the function , which Hlder denotes by +, be the operation of composing two objects a and b of F to the object a 0 b of F. The abstract attributes
of on F are to be formulated in a similar way as those for the natural numbers
were fomulated by Peano. But the specic attributes of this operation are
widely varying, depending on the given eld F as well as on the specic content
of equal-to and smaller-than on F, in short: of the specic comparative order
<equal-to, smaller-than, F>.
Now an object d of F is composed out of a set of objects being parts of d,
which themselves need not be elementary ones; but given an elementary object
c of F, then, of course, the set of its parts is the unit class of c. Such a set of
composing parts of x is the smallest set to be obtained in that way, as is shown
by the following denition:6
Dfhd-1

Some object w of F is element of the set of composing parts of the


object x of F i w is element of every subset H of F containing x as
well as every y and z of F, for which also y z i.e.: the compound
of y and zis within it.

In order to compose an object out of two other ones these other ones have to
be of completely dierent kind, i.e. they must not possess any comon part:
Dfhd-2 Any two objects x and y of F are of completely dierent kind i
there is no component of them within both of them.
6

The deniens of that denition formulates the abstract attributes of this set as those of a
cylindric set.

On Using Measuring Numbers according to Measuring Theories

Furthermore, being a consequence of the following Hlder axioms, the attributes of the relation smaller-than are depending completely on that ones
of the relation equal-to and the operation of composing two objects to a new
one. We therefore reduce the fundamental vocabulary of this axiom system in
dening in advance, what lateron can be proved:
Dfhd-3 Some x of F is smaller than some y of F if there exists some z of F
such that the result of adding z to this x or adding x to this z is (in
its quantity) equal to y.
Then Hlders axioms may be refomulated for the objects of the eld of applications as follows:7
Axhd-0 (a) The operation of composing is on F conserving the order
of the relation equal-to; i.e.: Adding equivalent objects to
equivalent ones leads to equivalent compounds.
(b) The operation of composing is on F conserving the order of
the relation smaller-than; i.e.: Adding a smaller object to a
smaller one and a larger object to a larger one, than the rst
compound is still smaller than the second one.
Axhd-I (a) Equal-to is on F external-connex to smaller-than.
(b) Equal-to is on F exclusive to smaller-than.8
(c) Smaller-than is on F asymmetrical.
(d) Equal-to is on F symmetrical.
(e) Smaller-than is on F transitive.
(f ) Equal-to is on F transitive.
Axhd-II There is no smallest object in F. (This implies that F does not contain completely elementary particles, i.e. undividable atoms).
Axhd-III For every two objects of F being of completely dierent kind, their
compound is also in F.
7
8

This rst pair of axioms is not mentioned by Hlder but obviously tacitly presupposed by him.
The pair of relations smaller-than and equal-to are a comparative order on F. They are related
one to another in the following manner: They are exclusive (Excl) one to one another; i.e.: If
between two objects of F the relation equal-to holds, than smaller-than does not hold between
them. And they are external-connex (ExtCon) one to another; i.e.: If between two such objects
the relation equal-to does not hold, than either the relation smaller-than or its converse holds
between them.
NB: The rst four parts of this axiom are mentioned by Hlder in a footnote, whereas the
last two ones are obviously tacitly presupposed by him.

58

Wilhelm K. Essler

Axhd-IV (a) Adding some object of F to another one of F leads to a larger


result.9 (This implies that there is no object in F whose g-value
is 0).
(b) Every object of F may be doubled in its value. (This implies that
there exists no largest object in F).
Axhd-V For every two objects of F being of dierent quantity there exists
the dierence between them.
Axhd-VI (a) The operation of combining is, in combining three objects of
F, in F associative.
(b) The operation of combining is, in combining two objects of
F, in F commutative.
Axhd-Ded The eld F is complete concerning the relation smaller-than; i.e. it
contains every Dedekindian cut.
Two remarks may be in order at that point:
(1) The Axhd-IV expresses the innity of F in the sense of the innite sequence
of the natural numbers: Given some object a of F, there exists another object
of F of doubled quantity, and to this second one again another one with still
doubled quantity, and so on ad innitum. Therefore, the fourth axiom of Peano
is established in this apriorically accepted background theory of measurement
as a theorem:
Thhd-4 For every two dierent natural numbers, its resp. successors are different, too. (This implies that non of them is empty, i.e. that there
exist classes of every nite size.)
Of course, Thhd-4 is not deduced out of mere denitions and is therefore not an
analytical truth but a synthetic one, in regarding the Hlder axioms as true statements. And in regarding them as not being received by empirical means like
induction with regard to observations10 buton the contraryas constituting
the domain of establishing empirical truths, they thenin applying this axiom
systemare to be regarded as being non-empirical truths in this system, i.e.
apriorical ones in this system of synthetic statements.
(2) The Dedekindian Principle entails the Archimedian Principle which
guarantees that no object of F is inaccessible by the other ones, e.g. via being
9 Because of Axhd-VI, the vice versa need not be formulated.
10 It need not be justied separately that, e.g., Axhd-Ded cannot be justied by any empirical
means an cannot be regarded therefore as an empiricalresp. aposteriorical truth.

On Using Measuring Numbers according to Measuring Theories

repeatedly doubled; and this entails a strengthened assertion of the innity of


F:
For every object x of F, there exists such a set of duplicates in F, each
of its members being equal in magnitude to x and thus also one to
another. The cardinal number M of this set may be nite in each
cases of application;11 but there does not exist an upper bound of
niteness. Then, according to the laws of higher order logic, this is
logically equivalent to say that there exists an innite set G of duplicats of F, each of its members being equal in magnitude to x and
thus also one to another.
Given such an arbitrary x of F, because of the axiom of choice there exists a
choice function f which is ordering a denumerable innite subset H of G; thus,
f is mapping the set of natural numbers onto that subset H; now, H is also a
subset of the equivalence class of x, i.e. of the class of those objects of F which
are equal to x. We then dene recursively the concept of M-fold application
of the operation to x:
Dfhd-4 Let x be some arbitrary object of F; let H be some arbitrary denumerable innite subset of the equivalence class of x but without x; let
there be any numeration f of H; and let there be any naturalresp.
nite cardinalnumber M establishing the segment of the initial
M members of H. Then the M-fold application of is to be distinguished according to: (a) If nothing is added, then the compound
is, of course, nothing but x itself; (b) if something new is added, it
is added to the previous compound.
The Archimedian Principle, being a theorem of this axiom system, may now
be formulated as follows:
ThArch Given two arbitrary objects x and y of F, whereby x is smaller than
y; then there exists a denumerable innite subclass of other objects of
F which are equal to x and a denumeration f of it as well as a natural
number M such that adding successively the initial M members of
this sequence f leads to a compound which is not smaller than x, i.e.:
which is equal-to or even larger-than x.
11 The natural numbers are in fact nothing but the nite cardinal numbers.

60

Wilhelm K. Essler

Again, two remarks are in order here:


(1) If we substitute Axhd-Ded by Thhd-Arch, we then need rational numbers only
instead of real numbers, which are required by Axhd-Ded, since for the mere
purposes of measuring, the rational numbers are completely sucient.
(2) This Thhd-Arch indicates how measurements are to be performed, namely
according to the advice: Take some suitable choosen object u of F to be used
then as the unit element of the scales; and let the scales of dierent lengths
be established by the elements of the equivalence class of u according to the
multiplication M of u; and given another object x of F to be measured by
such an M-fold multiplication of u. Then the smallest M-fold multiplication
of u, which is equal-to or larger-than x, is the interval within the value of the
quantity of x is to be determined.
In proceeding now from this background theory of measurement to the
rules of measuring, we have to establish the apriorical structure of the extensive
quantity g in relation to these axioms of Hlder.
Usually, an additive quantity12 is introduced by adding to such a background
theory like the Hlder system a measuring theory of the following kind:13
Axasc-1 For objects of F being equal one to another, the quantity g of both,
measured with u, is the same.
Axasc-2 For two objects of F being of completely dierent kind, the quantities of their compound is the same as the sum of the quantitiy of
each of them.
Axasc-3 Take some suitable specic u of F, take some measuring number
r, and determine r as the measuring number of the g-value of u,
measured by itself.
Concerning the unit object u, it is suitable to choose it as an object of F which
is easily accessable and to which there exist a large equivalence class whose elements are easily accessible, too.
Concerning the number r, the decision is to be made in a related sense according to simplicity in using it in calculating. Therefore, e.g.
510511 or 41/97
12 There exist also non-additive quantities; but it seems that all of them are to be established
via additive ones.
13 See, e.g., Hempel (1965), Carnap (1926), Carnap (1966), Suppes (1951), Essler-Labude-Ucsnay (2000).

On Using Measuring Numbers according to Measuring Theories

are no suitable candidates for r to be used in performing calculations. Usually,


one of the numbers
103s, where s is one of the numbers 0, 1, 2, 3,
is taken as such an r, at least in using the decimal system of numbers.
The nal question is now to be answered, how in this approach the set of
measuring numbers is to be introduced in relation to the set of natural numbers.
If F were consisting of compounds combined from undividable elements,
then the measuring numbers could be identied with the natural numbers.
For in this case, the quantities of all compounds were to be determinated as
/tHu(n) for some respective natural number N, whereby u is one of these
undividable objects. But the Hlder system excludes the existence of such
atoms.14
No way out is the procedure of using the integer numbersor: the whole
numbersinstead of the natural numbers. For in the sense of foundational research, integer numbers are nothing but functions of pairs of natural numbers,
namely the addition of the respective dierences of them, therefore non-negative dierences as well as negative ones. Thereby, the non-negative dierences
are totally dened functions, whereas the negative ones are partially denes
only; but, because of Thhd-4the theorem of innity in the Hlder system
none of them is completely undened.15
The value of such an integer number , applied to a natural number L,
is then determinated according to the equation: (L) = (L + M)N = L +
(MN) for some ordered pair M and N of natural numbers, concerning their
dierence. And, since F is innite according to Thhd-4, such natural numbers do
exist, which means here: they are notfrom a certain point on alltogether
identical with the empty set of second level.
In this sense, integer numbers are the procedure of adding dierences of
natural numbers M and N: If M may be 0, then these whole numbers are
non-positive ones; and if N may be 0, then they are non-negative ones. We
14 Taking something as atoms, like the substances of the periodic system in chemistry, and
using also the name atoms for them, it usually turns out lateron that these atoms are in
fact dividable.
15 For details see Essler-Brendel (1993), Kap. X, Abschn. 1214. This being not completely
undened is essential; for otherwise they would be empty andin being the empty class
therefore identically one to another.

62

Wilhelm K. Essler

therefore, according to this approach of apriorically justifying the systems of


numbers, may dene the concept integer number as follows:16
Dffr-4 Some function is an integer number i there exist natural numbers M and N such that for every natural number L, whereby some
natural number K satises K + N = L + M, we get: (L) = (L + M)
- N.
From the point of view of the integer numbers, the natural numbers are occurring somehow as pre-integer numbers. These pre-integer numbers are, of course,
ontologically dierent from the non-negative integer numbers. For the former
ones are abstract entities of level 2, whereas the latter onesconcerning this
foundational approach in the sense of Frege and Russellare functions of them
and therefore entities of level 3. But there is, nevertheless a simple and thus
natural bijective mapping of the one set onto the other one, in the sense:
0  sc0 (= 0 = indentity functions restricted to natural numbers)
1  sc1 (= +1 = sc = successor of )
2  sc2 (= +2 = sc sc = successor of successor of )

It is this mapping, according to which here it may be said: the set of natural
numbers is embedded into the class of integer numbers.
Obviously, the integer numbers do not measure any object of F concerning
its additive quantity g, too. We therfore now go ahead to the rational numbers.
But in regarding thereby all the positive as well as the non-positive rational
numbers as to be established in a manner like integer numbers were established, we in advance have to look for a suitable set of proportional respective
pre-rational numbers, regarding them thereby as directly related to the innite
set of natural numbers.
These proportional numbers are, of course, not negative ones. But, astonishingly, they furthermore do even not contain a number Zero. For the set F of
physical objects does not contain an element being the smallest one concerning
the quantity g. Therefore, each object u of F, being chosen as the standard object for performing measurements according to Axasc-3, will somewhen turn out
of being too large with regard to the object d to be measured, or of being not
16 In dening them in this way, each single integer number contains: (1) the structure of this
adding-a-dierence, (2) the minuend M, and (3) the subtrahend N.

On Using Measuring Numbers according to Measuring Theories

comparable with it in using natural numbers only because of other reasons.


Let, e.g., g be the length of the objects of F, being measured in comparing
it to the standard object m.17 Suppose, we have to measure the length of an
object y from which we are told it to be a duplicate of either a USA-yard or a
UK-yard. In measuring this y in relation to m, we thenin accordance with
the Archimedian Principle and in applying itwill empirically determine its
length step by step, in this case: in 7 steps, beginning with the 0-th step: The
0-th step is that part of the number to be determined, where no (= 0) fractions of
m are used.
(0) At the 0-th step, the length of y, measured with elements of equal lenths
to m, is determined as being larger than 0 but not larger than 1.18
(1) At the 1-st step, the length of y, measured with elements of equal length
to dm, is determined as being larger than 09 but not larger than 10.
Hereby, 10 duplicates of dm, combined together, are of equal length
to m, whereby 09 of them are needed to determine the lower bound
of the value at this step.
(2) At the 2-nd step, the length of y, measured with elements of equal
length to cm, is determined as being larger than 091 but not larger than
092. Hereby, 100 duplicates of cm, combined together, are of equal
length to m, whereby 092 of them are needed to determine the lower
bound of the value at this step.
(3) At the 3-rd step, the length of y, measured with elements of equal
length to mm, is determined as being larger than 0914 but not larger
than 0915. Hereby, 1000 duplicates of mm, combined together, are of
equal length to m, whereby 0914 of them are needed to determine the
lower bound of the value at this step.
Up to now, we still are not able to decide the kind of yard we are trying to
determine; and since this is not the subject of this paper, we will leave this
questian open. But instead, we will advance in answering the question, how
17 This object m is kept in Breteuil near Paris. In 1795 it was dened by the French parliament
UIFNUPCFNPSFFYBDUMZUIFEJTUBODFPGUXPNBSLTBUNUPCFt-6th part of an
earth-meridian. 1875 this was accepted by 35 nations. But 1960, it was redened as: 1 m being
1.650.763,73 wavelengths of the orange spectrum colour of Krypton.
By the way: The length of the yard is dened in dependence of the meter, namely: 1 USA-yard
being the length of 0,914402 m, and 1 UK-yard being the length of 0,914339 m.
18 Since the set of proportional numbers is in fact without a number Zero, the way of speaking
larger than 0, as I used it above, must not be analyzed here too exactly.

64

Wilhelm K. Essler

the proportional numbers are to be identied according to this approach. In


rewriting these four steps, we obtain:
This nite sequence of nested intervals may be rewritten as half-open intervalls of the following kind:


  t-0 t-0> t-1 t-1> t-2 t-2> t


10-3 t-3]>

Regarding only the lower bounds of this nite sequence, we receive the following basic sequence:
<0; 0,9; 0,91; 0,914>
This nite sequence may be regarded as the initial part of all those innite
sequences, whose rst members are identical with that ones. Because of the
niteness of the life of every man as well as of the mankind, we never will get
more than such a nite sequence, how ever this sequence will increase in the
future; and every nite sequence of that kind is nothing but a proportion of
two natural numbers M and N, which therefore determine again a function
to be applied to natural numbers:
Dffr-5 Some function q is a proportional number i there exist positive
natural numbers M and N such that for every natural number L,
XIFSFCZTPNFOBUVSBMOVNCFS,TBUJTmFT,t/-t. XFHFU
R -
 -t.
/w
The requirement 0 N need not be justied. The requirement M 0
excludes certain degenerations of the proportional number q which otherwise
NJHIUPDDVS"OEUIFSFRVJSFNFOUiTPNFOBUVSBMOVNCFS,TBUJTmFT,t/
-t.wJOEJDBUFTUIFEPNBJO XIFSFJOUIJTGVODUJPORJTEFmOFEXJUISFTQFDU
to its arguments L of natural numbers, according to this approach.
These proportional numbers are partially dened functions only; but none of
them is completely undened. For taking a natural number L, e.g., as identical
with N, then there exists some natural number Knamely Msuch that this
requirement is satised.
In this way, the proportional number q is nothing but the function of multiplying the fraction M/N; and the value q(L) ist therefore identically with
- t .  /
 PSUP CF MFTT WBHVFXJUI - t .
  / 'PS FWFSZ QVSQPTF

On Using Measuring Numbers according to Measuring Theories

of measuring additive quantitiesi.e. for performing the measuring theory of


additive quantities, no other numbers than these proportional numbers are
required.
The set of rational numbers isaccording to this kind of approachto be
established comparable to the set of whole numbers, namely as the operation
of adding dierences:
Dffr-6 Some function is a rational number i there exist proportional
numbers p and q such that for every proportional number l, whereby
some natural number k satises k + q = l + p, we get: (l) = (l + p) q.
And again, neither the proportional numbers nor the rational numbers are
satisfying Hlders system, especially with regard to Axhd-Ded. Therefore, we
have to proceed from nite sequencesi.e. from proportional numbersto
converging innite sequencesi.e. to measuring numbersin order to receive
a system of numbers satisfying also the theory of measurement and not only
the applications of this theory.
Usually, the real numbers are determined via Dedekindian cuts or via converging innite sequences. The easiest way is, in fact, to dene them via Dedekinds procedure. But with regard to fundamental research, this approach does
not show immediately the connections of proportional numbers to the measuring numbers respectively the rational numbers to the real numbers. And
determinating them via converging innite sequences entails, as will be shown
later, in advance the dividing the set of such sequences into suitable equivalence
classes, which is all but a simple procedure. Therefore, in this approach the way
of nesting intervals is used.
In order to develop the instruments of nested intervalls, rst of all some
auxiliary concepts have to be explained:19
Dffr-7 Some function f, mapping some set P into some set Q, is a sequence
i P is a subset of the set of natural numbers, whereby f is a surjective
mapping of P onto Q.
Dffr-8 The initional part below the natural number N is the set of natural
numbers which are smaller than N.20
19 The following denitions are to some parts less generall as they are used in mathematics; for
the are dened for the purpose in question only.
20 Note: The initional part below 0 is the empty set of level 3.

66

Wilhelm K. Essler

Dffr-9 Some function f, mapping some set P into some set Q, is a nite
sequence i f is a seqence of this kind whereby P is an initional part
of the set of natural numbers in its natural order.
Dffr-10 Some function f, mapping some set P into some set Q, is an innite
sequence i f is such a sequence whereby P is the complete set of
natural numbers.
Dffr-11 Some function f, mapping some set P into some set Q, is a converging sequence i f is an innite sequence such that for every proportional number (be it larger or smalles or very small) there exists some
step within this sequence from where on the dierences between the
f-value of this step and every following step is smaller that that the
given proportional number.
But dierent sequences may converge to the same point. In adding to some converging sequence a sequence converging against Zero, we receive some other
sequence converging to the same point as the former one did. They all can
be put together into the equivalence class of the former one; but most of the
elements of this class do not show any similarity to the results of measuring,
i.e. to its initial parts, to the respectivenite sequence of increasingnite
sequences. The nested intervals, on the contrary, are avoiding these non-intended circumstances. And, furthermore, since they are consisting of a pair of
converging innite seqences, they need not worry how to handle with, e.g.,
the sequences 0,999999 and 1,000000, which are converging to the same
point; for they are nothing but the step-by-step intervals of each of the respective steps of the left and of the right sequence.
In establishing the measuring numbers with regard to the proportional ones,
we have to dene the concepts intervall (of proportional numbers), sequence (of them) and converging sequence (of them):
Dffr-12 The half-open intervall of proportional numbers between p and q
is the set of the proportional numbers k, which are larger than p but
not larger than q.
Dffr-13 Some function f, mapping some set P into some set Q, is an innite sequence of half-open Intervalls i P is the set of natural numbers, and f is a surjective mapping P onto Q, whereby the elements
of Q are half-open intervals of propertional numbers such that,
for every natural number N, the interval f(N + 1) is a subclass of
f(N).

On Using Measuring Numbers according to Measuring Theories

The last condition guarantees that the measuring values of each step are carried on within all of the following steps of measuring. In this sense, we now are
dening the concept measuring number by using topological means:
Dffr-14 Some function f, mapping some set P into some set Q, is a measuring number i f is an innite sequence of half-open intervals which
are converging in the following sense: For every proportional number k there exists a natural number M, such that for every natural
number N larger than M the dierenc qp of the right and left
bound of f(N) is smaller than k.
In order to obtain the usual kind of designating these measuring numbers,
we dene:
Dffr-15 Some r is a measuring number i there exists some function f mapping the set of natural numbers into some set Q, whereby r is the
ordered pair of f and Q, this ordered pair being thereby a measuring
number (in the former sense).
This set of measuring numbers contains exactly these numbers, which are
required by the Hlder system and the three axioms of additive quantities
as values for the objects of F, in idealizing the results of measuring these objects. Therefore, also this set of measuring numbers does not contain a number
Zero.
In order to handle with these idealized results in empirical theories to be
established in accordance with these measuring numbers, we have to proceed
from them to the real numbers, regarding them again as the operations of adding dierences of measuring numbers:
Dffr-16 Some function is a real number i there exist measuring numbers
r and v such that for every measuring number s, whereby some measuring number w satises w + v = s + r, we get: (s) = (s + r) - v.
The philosophical meaning of these real numbers, established apriorically with
regard to the Hlder system, when used then in connecting measuring numbers in empirical theories, is up to now not analyzed in all its details, as far as
I am aware of it. And the same holds for complex numbers, those functions of
an ordered pair of real numbers, which are used in electrodynamics as well as

68

Wilhelm K. Essler

in quantum mechanics. The sense of the statements of these empirical theories


will become clear in all details as soon as the philosophical meaning of complex numbers in relation to real numbers will been analyzed according to
this foundational procedure.

Acknowledgments
I would like to thank MA Joachim Labude and Dr. Stephanie Ucsnay for
helpful comments concerning an earlier version of the Hlder system, as well
as Prof. Dr. Elke Brendel and Prof. Dr. Rosa F. Martinez Cruzado for helpful
comments concerning an earlier version of that developmet of the systems of
numbers.

References:
Carnap, R. Physikalische Begrisbildung Karlsruhe 11926, Darmstadt 21966
Carnap, R. Philosophical Foundations of Physics New YorkLondon 1966, dt.:
Einfhrung in die Philosophie der Naturwissenschaft Mnchen 1969
Dedekind, R. Was sind und was sollen Zahlen? 1988, abgedr.: Gesammelte mathematische Werke 3, Braunschweig 1932, 335390
Essler, W. K. Analytische Philosophie I Stuttgart 1972
Essler, W. K.Brendel, E.Martinez Cruzado, R. F. Grundzge der Logik II
Frankfurt/M. 19873
Essler, W. K.Brendel, E. Grundzge der Logik II Frankfurt/M. 19934
Frege, G. Die Grundlagen der Arithmetik, eine logisch-mathematische Untersuchung
ber den Begri der Zahl Breslau 11884, Darmstadt 21961, Hamburg 19863
Grassmann, H. Die einfachen Verknpfungen extensiver Grssen 18622, abgedr.:
F. Engel (Hrsg.) Hermann Gassmanns gesammelte mathematisch-physikalische
Werke II/1, Leipzig 1896
Hempel, C. G. Fundamentals of Concept Formation in Empirical Science Chicago
1952, dt.: Grundzge der Begrisbildung in der empirischen Wissenschaft Dsseldorf 1974
Hlder, O. Die Axiome der Quantitt und die Lehre vom Mass, in: Berichte ber
die Verhandlungen der kniglich schsischen Gesellschaft der Wissenschaften zu
Leipzig, Mathematisch-physikalische Classe 53 (1091) 164
Kant, I. Kritik der reinen Vernunft Riga 17811 (= A), 17872 (= B)
Kant, I. Metaphysische Anfangsgrnde der Naturwissenschaft Riga 1786
Peano, G. Formulaire de mathmatiques Turin 1895
Reichenbach, H. Relativittstheorie und Erkenntnis Apriori Berlin 1920

On Using Measuring Numbers according to Measuring Theories

Russell, B. The Principles of Mathematics Northampton 11903, 21972


Suppes, P. A Set of Independent Axioms for Extensive Quantities 1951, abgedr.: Suppes, P. Studies in the Methodology and Foundations of Science Dordrecht 1969,
3645

70

Wilhelm K. Essler

The Compulsion to Believe: Logical Inference and Normativity

Part II
The Challenge of Nominalism

72

Jody Azzouni

The Compulsion to Believe: Logical Inference and Normativity

The Compulsion to Believe: Logical


Inference and Normativity
Jody Azzouni
Abstract
The interaction between intuitions about inference, and the normative constraints that
logical principles applied to mechanically-recognizable derivations impose on (informal)
inference, is explored. These intuitions are evaluated in a clear testcase: informal mathematical proof. It is argued that formal derivations are not the source of our intuitions of
validity, and indeed, neither is the semantic recognition of validity, either as construed
model-theoretically, or as driven by the subject-matter such inferences are directed towards.
Rather, psychologically-engrained inference-packages (often opportunistically used by mathematicians) are the source of our sense of validity. Formal derivations, or the semantic
construal of such, are after-the-fact norms imposed on our inference practices.

1
Mathematical proof amazed ancient Greeks. Here was a methodreasoning
from assumptions to unexpected new results. Furthermore, one saw that the
conclusions had to follow. On my reading of Platos Menoand his other
dialoguesthe Greek discovery (of deduction) not only provoked Plato to
the hopes of nally resolving ethical dierences (by importing the method of
reasoning from geometry), but also provided himby means of a best explanation for why mathematical proof workssupport for reincarnation.
Those were glorious days for philosophy, werent they? So much seemed possible then by sheer reasoning aloneand there are still philosophers living o
the meager echoes of that project. But some thousand years later most of us
arecomparatively speakingrather jaded about deduction; indeed, many
philosophers, sociologists of knowledge, and others, are jaded enough to nd
tempting social constructivist views about mathematical proof. Social constructivists take mathematical proof as no dierentsociologically speaking
from other practices that humans conform to: cuisine, tacit restrictions on
polite conversation, linguistic rules, and so on. On such views, the plethora of
alternative logicsand within themthe plethora of alternative mathematical
systems, that were such a shocking discovery of the twentieth century, should
have been expected; indeed, only sheer historical (and contingent) facts are

74

Jody Azzouni

available to explain why mathematics took the particular developmental trajectory it took, and why it was tacitly based upon the particular logic (until the
twentieth century) that it was based upon. Reason, on such a view, is a kind
of fashionable dress of culturescanons of reason, too, ebb and ow among
peoples.
Social constructivist views, however, dont recognize how unusual mathematicssociologically speakingis. I must be brief;1 but its striking how,
in contrast to politics and religion (and philosophy, for that matter), doctrinal mistakes lead in the latter cases to new views, or to new standards that
views should presuppose, whereas in mathematics mistakes in proofs are
eliminatedeven if undetected for many years. One striking piece of evidence that mathematical proof, during its thousands-of-years development,
remained largely within the connes of a particular (although tacit) logic, was
that the grand regimentation of it by logicistsFrege, and later, Russell and
Whiteheadlargely succeeded with respect to the mathematics of the time.
Indeed, the plethora of alternative mathematicsbased on dierent logics, and dierent substantive mathematical principlesbecame a topic for
mathematical exploration largely because logicists had made what appeared
to be the logic of mathematical proof (indeed logic tout court) explicit, and so
practitioners couldfor the rst timeconsider changing the rules.
Call the following theses the traditional view: (i) informal mathematical
proofs, though in the vernacular, correspond to derivations of formal languages
(perhaps by being abbreviations), where derivations are mechanically-recognizable constructions without missing steps2; (ii) this (tacit) correspondence
explains the uniqueness of mathematics as a social practice. The properties
of informal mathematical practiceincluding its apparent imperviousness to
changes in its logicis explained by mathematicians (when constructing or
reading informal proofs) actually grasping derivations (in formal systems)
corresponding to these proofs.
The traditional view satises thricewise: First, it provides causal machinery
derivationsfor explaining the uniqueness of mathematics as a social practice;
second, via those same derivations, it provides normative standards by which
1
2

See Azzouni 2006, chapter 6, for the longer version.


I argue in my 2005 that the background logic of derivations, at least with respect to traditional
mathematics, induces the same consequence relation as the rst-order predicate calculus
but this isnt required of the traditional view. If, however, the traditional view takes the formal
language as one specic language, then its refuted by Gdels theorem. Read the traditional
view, therefore, as taking the formal derivations in question as belonging to an open-ended
family of formal languages.

The Compulsion to Believe: Logical Inference and Normativity

informal mathematical proofs are (ultimately) to be judged by practitioners as


correct or incorrect; and so, third, it uses the phenomenological impression of
compulsionthat mathematical proof induces in cognoscentito connect
the perception of causal machinery (derivations) to normative standards (also
derivations). How nice when the descriptive and the normative dovetail so
compatibly.3
And yet how suspicious. (Philosophy begins in wonder, but that pleasant
sensation never lasts.) I confess to being tempted by the traditional view4so
describedeven long after I had learned to resist attributing to us the magical
powers purported knowledge of abstracta invite: that somehow, by the magic
of pure descriptions, we grasp truths about eternally aloof objects. Descartes,
we remember, began modern philosophy with a similar welding of the (psychological) impression of compulsion with the recognition of a standard (clarity
and distinctness) for recognizing metaphysical necessity. No surprise that he
was a rst-rate mathematician.

2
Whats wrong with the traditional view? This: It requires mathematicians to
have enough of a grip on theotherwise unexplicatedderivations (taken
to correspond to informal proofs) to explain in terms of those derivations the
sense of validity an informal proof induces in its readers. Because informal
mathematical proofs alwaysfrom the point of view of the formal derivationskip numerous steps, this is possible only if the missing steps (via the
mathematicians awareness of whats missing) are causally active in the phenomenology of mathematical proof. That is, perceptionin some senseof the
derivations that correspond to informal proofs can thus be the source of the
compulsion induced by informal mathematical proofs only ifsomehow
3

Although the phenomenological sensation of the compulsion to believe is the central topic
of this paper, by no means is it my full story of how mathematicians are convinced that a
proof for a result exists, nor even the full story of how a mathematicianwhen surveying a
proofis convinced of its validity. The division of intellectual labor within mathematics itself
(see my 1994, Part III, 2) operating as it does both with respect to results the mathematician
presumes the truth of, and even during the cognizing of a particular proof (where some steps
in a proof are accepted on authority), already shows this.
Indications of backsliding to the traditional view may remain visible in my 2006; theyre, I
hope, absent from my 2005which was written after.

76

Jody Azzouni

the tacit recognition of the missing steps gives enough of a grip on the course
of a derivation corresponding to an informal proof to explain why the mathematician feels the conclusion follows from the premises. Only in this way will
the normative standards that corresponding derivations supply to informal
proofthat such derivations are themselves the standards by which mistakes
in informal proofs are recognizedactually be operative in the recognition (on
the part of mathematicians) that a proof is valid. (And only in this way can the
uniqueness of mathematics as a social practice be explained by the perception
by mathematicians of correlated derivations.)
Ive come to believe that the requirements just laid out are impossible to
meet. Therefore, the normative role of the correlated derivations isnt connected
to the phenomenology of compulsion that informal mathematical proof induces. Furthermore, I suggest this coming apart of a psychological impression,
and the normative standards that its supposed to be an impression of, is widespread: Philosophers oftenas the Cartesian example indicatestreat psychological compulsion (that something seems like it must be a certain way) as a
kind of perception (e.g., that of the metaphysics of possibility and necessity),
and therefore as havingfor that reasonnormative force. How this fails with
respect to informal mathematical proof illustrates a general phenomenon.
Consider a derivation thats to replace an informal proof. When such is
constructed, not only will it be very much longer than the originalinvolving
syntactic manipulations that a mathematician couldnt even be aware of, but
it will be padded with additional assumptions that mathematicians alsooftencouldnt be aware of.5 But theres an interesting phenomenological point
about the relationship of the formalized proof to its unformalized cousin: Even
if we understand that each step in a derivation follows from earlier ones, that
knowledge neednt contribute to our understanding of the informal proof;
rather, one often gets lost in the details of the formal derivation, and cant tell
what the main ideas are. So, at least phenomenologically, it seems that the
source of compulsionthe sense that an informal proof is validisnt due to a
perception of the correlated formal proof. Indeed, the epistemic process is usu5

Mackenzie 2005 is a news-brief about proof assistants: software that checks proofs that
have been formalized into appropriate (mechanically checkable) form. We nd that all in
all, people who have used proof veriers say they can formalize about a page of textbook
mathematics in a week. There is also an anecdote in Moorehead 1992, p. 92, that Whitehead
estimated completing Principia Mathematica would take a short period of one year, a tenfold
underestimation by someone who had published much mathematics.
I chose the word couldnt deliberately. Often the explicitation ofheretofore tacitassumptions underlying an informal proof are matters of major mathematical discovery.

The Compulsion to Believe: Logical Inference and Normativity

ally the reverse: One understands why a formal proof is possible only because
of the way that it was constructedstarting from an informal proof.
One might attempt to save the purported causal role of derivations in the recognition of the validity of informal proofs by borrowing a page from linguistics.
A truism in that eld is that our ability to distinguish grammatical from ungrammatical sentences is due to complex (subconscious) processing. Strikingly,
whats processed (be it rules, or whatever) is so inaccessible to introspection
that one can only discover it empirically. And this means that were one to see
a description of the mechanisms by which one distinguishes grammatical from
ungrammatical sentences, they would remain introspectively alien: One would
fail to see how such contributed to ones understanding of grammar.
The traditional view still has a hope if one or another sort of formal derivation neurophysiologically (as it were) underlies the mathematicians grasp
of informal proof. Unfortunately, the empirical prospects for this hypothesis
arent good: As we gain an understanding of the neurophysiological bases of
mathematical abilities, the result isnt the discovery that the grasping of derivations of one or another formal system lies in back of our abilities; rather, its
that there is a patchwork of narrow modularized capacities that are brought
to bear on mathematics. These capacitiesdispositions, in some senseare
proving to be fragmentedly piecemeal in their scope, and ones that, in addition
to enabling mathematical task-solvinge.g., addingare equally the source
of common mathematical errors.6
So informal mathematical proofs are perspicuousand therefore often clearly communicate the logical status (validity or otherwise) of an informal proof,
in contrast to formal derivations that are impenetrable except in the step-bystep mechanical sense that each step follows from earlier ones. Indeed, even
in the case of an informal proof, ones sense that its valid often precedes a close
examination of its steps; one gets the sense of validity rston the basis of a
broad overviewand then looks to the steps to see how the trick is turned.
The foregoing leaves us with two things that need explanation. The rst is
where the phenomenological feel of psychological compulsionthe it must
be that we feel when recognizing the validity of an inferenceis coming from:
What (causally) is it about a good informal proof that compels assent? And
second (and notice these queries will now have answers that arent linked) an explanation of the normative status of strict derivations must be given as well.
6

See, e.g., Dehaene 1991 or Dahaene 1997. Notice the point: these dispositions are as much
the foundation of our competence (in those aspects of mathematics they apply to) as they
are the foundation of our incompetence.

78

Jody Azzouni

First the psychological compulsion (the sense of validity of informal proof ):


We come equippedneurophysiologicallywith inference packages.7 Inference packages are topic-specic, bundled, sets of principles naturally applied to certain areas: various visualization capabilities, language-manipulation
capacities, kinesthetic abilities, and so on. For example, we can spontaneously
visualize how line gures change when moved about on a surface. We recognize thisin partby factoring in the curvature of the surface they are on;
but success in this endeavor occurs without any introspective grip on what
inferential principles were using. That is to say, we cant explicitly formulate
generalizations about the curvature of surfaces and how that curvature aects
properties of gures on the surfaces.8 Intuition, as mathematicians use the
term, involves inference packageswhich amount to our grasping bundles
of principles without necessarily being able to distinguish specic assumptions9; so does the elusive understanding that some proofs provide, and others
dont.
Inference packagesthough genuinely syntactic insofar as they involve the
manipulation of sets of bundled principles to enable the drawing of implications from those principlesdont introspectively present as syntactic for two
reasons. The rst is that inference packages are psychologically designed for
specic situationse.g., geometric visualizations: Thus the concepts involved
in such packagesintrospectively speakingseem to come intrinsically attached to xed subject mattersindeed they seem to come with an implicit
interpretation; this is something that concepts governed by syntactic rules
dont come with. In order for a mathematician to shift the application of an inference package from one (mathematical) domain to another, she mustoften
with dicultyreinterpret the concepts in an inference package.10 Second,
7 I rst described inference packages in my 2005.
8 Indeed, it took mathematical talent of a pretty high order to determine what these generalizations look like.
9 A qualication. This way of putting the matter makes it sound like its simply a fact that a
bundle of principleslogically equivalent to the inference package as a wholeis the same
as that package. This isnt quite my view. Rather, they are deemed identical in light of the
later construal of the package (in terms of the various principles its identied with). Also,
the packagein operationcan at times deviate from what we take (from the vantage point
of the various later principles) to be the right answer. See the last three paragraphs of section
3 for a discussion of the rst point. My second point is touched on in the last paragraph of
this paper, and discussed more fully in my 2005.
10 An example: I may mentally depict four-dimensional geometric spaces via visualization of
a three-dimensional space where the points are endowed within additionreal-valued
temperatures ranging from -d to +d.

The Compulsion to Believe: Logical Inference and Normativity

inferring via such packages is accompanied by a psychological compulsion to


believe that if this is the case, then that must be the case. Such a compulsion
always accompanying inferences bundled with an implicit subject-matter gives
the impression that the subject matterwhat the inferences are aboutis
itself forcing the conclusions. This phenomenological impression, it should be
noted, is rarely present during mechanical manipulation (according to rules)
to determine that something follows from something else.11
Ive described the compulsion in inference (in much of mathematical proof )
as including the impression of the subject matter forcing conclusions from
premises. And, I think, this is why the thought of making mathematical proofs
explicit not only required every step be present, but was also accompanied
by the idea that doing so would supply a complete (interpreted) language for
mathematicsa complete set of primitive mathematical concepts. The rules
governing such concepts were to be (semantically) transparent: They would
capture (fully) what such concepts meant. (So each concept would come with
a set of denitional axioms.) In turn, every step in a proof being present, plus
all the concepts in any proof being drawn from this complete set of concepts,
would result in proofs that would make transparent how the psychological
compulsionthe recognition of validitypresent in our perception of informal proofs resulted from our tacit recognition of the underlying complete
proofs.12
On my pictureon the contrarythe source of the compulsive impression of validity is due to the inference packages that psychologically enable
11 Hacking (1973, p. 20203) notes the operation of this distinction in what he takes as the
emergence of our contemporary notion of proof in Leibniz. He writes: Geometrical demonstrations can appear to rely on their content. Their validity may seem to depend on facts
about the very shapes under study, and whose actual construction is the aim of the traditional
Euclidean theorems. He adds: [a]lgebra is specically a matter of getting rid of some
content. Hence in virtue of Descartess discovery, geometrical proof can be conceived as
purely formal. Its a contingent psychological fact that inference packagesthat enable the
algebraic manipulation of proofsseem to lack content. But given this is true (and it seems
to be), geometrical proofs almost always seem explanatory and to provide understanding
(for why something is the case) in ways that algebraic ones dont.
12 If mathematical is replaced with logical, as it should be on Freges view, then this
nearly enoughis Freges project (and indeed, that of Russell and Whitehead). Freges logical language is interpreted. The contemporary vision of logical languages as open to the
reinterpretation of their nonlogical vocabulary comes later. Gdels theoremin various
formsshatters the logicist project; but Im focusing on the epistemic drawbacks of that
project, specically with respect to the attempt to analyze what we are recognizing when we
recognize a successful informal mathematical proof as such.

80

Jody Azzouni

mathematicians to construct and understand informal proofs: Formal derivations are too far away (psychologically speaking) to play a causal role in the
psychological story of how the ordinary mathematician either constructs or
understands informal proofs.13

3
The second bit to be explainedrecallis how, despite the absence of a causal role, derivations nevertheless came to play a normative role in informal
inference: why we currently take them to embody standards of correctness/
incorrectness for informal mathematical proof. Heres how that happened. Its
already a normative given in mathematical practice that an informal proof is
to be faulted if (i) it relies on substantial assumptions that are tacit, or (ii) if
it skips steps that are nontrivial to establish.14 Thus the status of a successful
informal proof is seen as promissory in the sense that should the explicitating
of tacit details reveal a non sequitor or a false assumption, the proof is taken
to have failed.
Mathematicians (like all of us) take the ability to engage in a complicated
activity that apparently involves many presuppositions to indicate thatin
some sensewe have a (tacit) grip on all the presuppositions. Thus its easy to
think that a mathematicians understanding of an informal proof turns on a
tacit grasp of a version of that proof without the missing steps, and without the
missing assumptions. So part of the story for why derivations operate as norms
for ordinary mathematical proof is an error theory: Given that one takes an
ordinary mathematical proof to be skipping steps (given that one accepts the
model that an informal mathematical proof requires lling out), one was
and isroutinely mistaken about how much is missing (how much is skipped)
13 However, the contemporary role of the computer in mathematical practice has induced
a new causal role for (tokens of ) formal derivations in ordinary mathematical practice
although that causal role isnt psychological. Computer proofs provide warrants that certain
(mathematical) results are truesuch is based on empirical results that computers have
veried certain derivations: This may take the form of good empirical reasons to think that
a computer has actually constructed a token of a formal derivation.
14 Its not seen as creative to ll in missing steps in proofsthats left to textbooks. But new
proofs of established results are of interest to creative mathematicians if they use a signicantly dierent approach; and, of course, substantial and nontrivial are professional
judgments.

The Compulsion to Believe: Logical Inference and Normativity

in such.15 That derivations have the status of norms for informal proofs is also
due, however, to there being no principled stopping point in the explicitation
of an informal proof earlier than a (formal) derivation that can be taken to
correspond to it. Only in such a derivation does the process of possible analysis
seem nished: only there is every step present, and every concept that was
tacitly involved in the informal proof now explicit.16
Notice that this normative role doesnt require derivations to have a causal
(psychological) role in how the mathematician recognizes errors in informal
proofs. That can be explained not by requiring psychological access to a strict
derivation that fully explicitates the informal proof, but by access to equally
informal explicitations of proofs that ll in (some) missing steps or assumptions.
One may worry that this explanationcontrary to the advertisement in the
last paragraphnevertheless (surreptitiously) brings in perception of formal
derivations. The recognition of gaps in a proof must involve a sense of what
a gapless version of that proof would be like. (And if that requires sensing
somehowthe formal derivation corresponding to the proof then such are
back in the picture.)17 What, therefore, is the source of the thought that informal proofs are missing steps, and what is it that allows the mathematician to
regard a proof that lls out some of these gaps to be an elucidation of the
original proof, as opposed to something new?18 One point of this concern is
that the answer to this question shouldnt turn on a tacit perception of the
goal (the more explicit proof, and ultimately, a formal derivation).
I dont want to suggestbecause I doubt its truethat as one comprehends
and becomes convinced of any particular informal proof, that one necessarily
hasthen and therea perception of its gaps. Sometimes, of course, thats
true. We often have the sense that steps have been skipped (and not just in
mathematics, of course), and we often request that some of these be lled in.
15 Recall footnote 5.
16 Nevertheless, there is latitude in what derivation corresponds to an informal proof because,
whenfrom the perspective of formal derivationssteps are missing, there are often nonequivalent ways of traversing the gaps. This hasnt aected, however, the normative status of
derivations. If one or another derivation corresponds to an informal proof, and the concepts
made explicit arent too arduous for mathematicians to be taken to have presupposed, the
informal proof is taken as corresponding to that proof.
17 We seem to be tripping over the paradox Plato (1961, p. 363) mentions in 8081 of the
Meno.
18 My thanks to Nancy Bauer, Sylvain Bromberger, and Eric Swanson for this particular formulation of the question.

82

Jody Azzouni

Here, one can (often) rely on the idea that a fuller explicitation of the argument
is playing a psychological role: One neednt, however, take that fuller explicitation to be anything like a formal derivation.
But apart from this, there are various models in the practice of mathematicsones that arise quite earlythat are taken to mark out in a clear way a
contrast between whats explicit and tacit in an informal proof, and in a
way that oers a contrast of completeness for informal proofs that supplements
the above perception of gaps in arguments.19
Consider sheer calculation; one rst learns about multiplication by its relation to addition, and about addition by its relation to counting20; furthermore,
one sees how various errors can arise, both at the ground level (by inadvertantly
skipping a numeral, or an object), and by introducing shortcuts (in addition
and multiplication). One, therefore, hasin the informal contexta full characterization of how mistakes arise, and how, by utilizing other methods, one
can triangulate access to right answers by means of multiple approaches. The
importance of this triangulation through multiple approaches is that various
mistakes (in the dierent approaches) dont coordinate into systematic (and
thus uncorrectable) errors.
Syllogistic reasoning, on the other hand, seems to exhibit entirely explicit
reasoning: Valid inferences are recognized by sheer grammatical form. Here it
looks like the analysis of a (quite short) proof has come to an end: There are
no missing steps. Most ordinary mathematical reasoning, of course, doesnt
look anything like this.21 Finally, there is also the example of compass and
straightedge constructions in Euclidean geometry.22 Here too, a mechanical
proof-system is in place; and proofs are seen as incomplete only in the tame
respect either that there are assumptions that one suspects should be instead
proven (e.g., the fth postulate), or that there are cases missing.23 Proofs in
19 These arise quite early both in the sense that one runs across such cases early in ones mathematical education, and in the sensehistoricallythat they arose early in the development
of mathematics.
20 Im not speaking of recursive denitions; I mean the related informal point that, e.g., the
adding of 17 to 15 can be executed by counting 17 items starting with the word 16; similarly, that multiplying 6 by 7 amounts to taking 6 seven times, i.e., counting 6 items seven
times.
21 Its signicant, however, that syllogistic reasoning turns outfrom the perspective of the
rst-order calculusas incompletely analyzed because there are missing connectives. I make
something of this shortly.
22 See my 2004.
23 That is, there is sensitivity to the danger of a mismatch between the cases depicted by a
diagram, and the cases actually under consideration.

The Compulsion to Believe: Logical Inference and Normativity

other branches of (traditional) mathematics clearly dier, both in the absence


of intuitively-justied mechanical-methods of reasoning, and in the absence of
clear signs that all the steps and assumptions are largely present.
That these exemplars place external pressureby comparisonon ones
sense that other informal proofs (in mathematics) are gapless is compatible
with such exemplars themselves subsequently being inexplicit vis--vis formal
derivation. But one can ask why explicitations of proofsones that are supplemented by extra assumptions, and additional stepsare seen as explicitations.
For example, imagine that a particular assumption is analyzed into several subassumptions that can subsequently be separated. Why is this seen as a matter
of more fully analyzing the proof, as opposed to being a replacement of the
original proof by a new one with dierent assumptions (some of which imply
the original assumption)?
I claim that, strictly speaking, there is no fact of the matter about whether
an explicitation of a proof really is an explicitation of that proof, as opposed
to a stipulative embedding of it into a dierent proof.24 Rather, it suces to
point to the reasons for why mathematicians will embrace such embeddings,
and (consequently) take them to be explicitations.25 First, there is our tendency
to bundle assumptions together, and infer conclusions from them as a group.
This case, which can reasonably be taken as one in which we do have psychological access to the separable assumptions, is confounded with cases where we
dont. Thus, an analysis of a proofthat analyzes it into additional steps and
assumptions (that we recognize to imply steps in the original proof )is always
presumed to be an analysis of the original proof.
Its important that such explicitations (almost always) respect the original
proofand so nothing is lost; but because (from the point of view of the
explicitation) assumptions have been made explicitseparated out from one
another in certain casesand inferential steps have also been made explicit,
theorem-proving capacities are greatly increased. This is because one is no
longer restricted to applying what amounts to the principles only as a group.
In general, explicitating increases proof-theoretic strength.
Its also worth stressing that the explicitation of proofs often involves a
supplementation of the concepts involved in the proofe.g., number, function, integral, and so on, so that proofs in one language are (often implicitly)
24
25

Recall the qualication of footnote 9.


These positive reasons are exactly the same ones that drive us, when working with explicit
algorithmic systems, to embed algorithmic systems conservatively in stronger ones. See Azzouni 1994, Part II, 6.

84

Jody Azzouni

assimilated to proofs in a dierent language.26 In this case, strictly speaking,


new assumptions have come into play, although they areespecially in the
context of informal mathematicsnot always seen that way.27 Its important
to the understanding of mathematical practice to recognize that there is no
sharp introspective distinction between analyzing a proof to tease out tacit assumptions that one hasin principleintrospective access to (because they
are assumptions that, at one stage, one has learned to bundle together), and
where such analysis amounts to embedding that proof in another so that the
result is a genuine supplementation. The simultaneous demands to conserve
already established results, and at the same time to develop new and interesting mathematics, work together to make such a distinction irrelevant to
mathematical practice.

4
There is another aspect of the normative role of formal derivations that may
seem still unexplained by the foregoing. This is that we often recognize the standards in one or another practice as ones easily changed. Say that the standards
in such cases have weak normative role, and distinguish this from strong
normative role, where the standards are perceived as unchangeable. Trac
laws have only weak normative role: We recognize such laws can be changed,
even though changesin certain caseswont be good.28 But the (classical)
logical principles governing derivations that we take to tacitly govern informal
mathematical proofs seem to have strong normative role. Classical logic strikes
many to be a standard for reasoning that we cant drop. This is a large part of
the intuition that many have that such principles are a priori.
For such, our recognition of validity isnt like the recognition of a standard
we happen to have. Instead, they react to proposals of alternative logics with
the baement suitable towards an incomprehensible or irrational suggestion.
They recognize, of course, that an alternative logicone that allows true con26 As a result, some proofs, with respect to an earlier set of concepts, can become special cases
with respect to the later concepts. Lakatos 1976 is a famous discussion of this phenomenon.
27 See section 7.11 of my 2006.
28 We could reverse the role of red and green trac lights. Given the dierential hard-wired
responses we have to these colors, the result wouldnt be as optimal as our current conventions.

The Compulsion to Believe: Logical Inference and Normativity

tradictions, saycould be adopted for proofs and reasoning; they recognize,


that is, that such an alternative logic would allow practitioners the mechanical
recognition of the validities so dened. But they balk at the idea that such
validities would therefore make sense.29
And yet, its striking how many philosophers (and logicians) lack this otherwise extraordinarily powerful intuition. Given the presence or absence of such
a powerful phenomenological impression operating in the background of ones
views about the status of classical logic, its no surprise that those, on the one
hand, who explain (away) the source of the intuitions that purport to give classical logical principles strong normative role, often provide explanations that
to those gripped by such intuitionsseem to miss the point (and force) of the
intuitions. On the other hand, if those gripped by such intuitions go beyond
the mere assertion of them (the mere assertion that classical logical principles
have strong normative role) by oering a justication for that role, they do so
in ways that seemto their opponentsobviously circular. I givein the rest
of this section, and in the nextsome illustrations of this (depressing) aspect
of the debate about the strong normative role of classical logic.30
Consider the suggestion that logical connectives have properties (e.g.,
Gentzen introduction and elimination rules, or axioms) by virtue of their
meanings. This isnt an old view quietly buried with the logical positivistsits
still au courant here and there. But for those who feel the normative compulsion of the rules that govern the classical logical idioms, this characterization
is too shallow: Meaning is a matter of stipulation; and if notbecause meanings attributed to terms are so attributed to capture antecedent usagestill
(without further grounds), antecedent usage is arbitrary, and could have been
dierent. Construing logical connectives as having the properties they have
because of the (antecedent) meanings they have procuresat bestweak normative role.31
29 Frege (1967, p. 14) expresses this widely shared sentiment: But what if beings were even
found whose laws of thought atly contradicted ours and therefore frequently led to contrary
results even in practice? The psychological logician could only acknowledge the fact and say
simply: those laws hold for them, these laws hold for us. I should say: we have here a hitherto
unknown type of madness.
30 An hypothesis (I have no idea how one would empirically test this claim): Those with this
intuition reasonusuallyvia inference packages. Those without it usually reason formally
or quasi-formally.
31 A way to see this is to note that contradictions arent meaningless. (Their meaningfulness
is marked by our recognition that they must be false.) Neither are liar paradoxes. (Their
meaningfulness is marked by the nausea-inducing recognition, based on what they seem

86

Jody Azzouni

OthersQuine notablydiagnose aprioristic intuitions about classical


logic as due to the central role classical logic (currently) has in our web of
belief.32 To explore alternative logics is to explore ways of possibly dissolving
such intuitions away by the exploration of an alternative web of beliefs with
an alternative logic at its center. The baement, when initially faced with
an alternative logic, therefore, is due to insucient practice in an alternative
mindset: live long enough amidst the inferences of an alternative logic and such
atavistic intuitions (eventually) wither away.
This Quinean explanation oers a promissory note: Our initial fear of
what looks to be a hitherto unknown form of madness vanishes with practice. Thus, on this view, one could say that the (apparent) impression that
classical logical principles have strong normative role is merely the perception of their weak normative role accompanied by lack of (logical) imagination. This explanation can be supported by the recognition that logical
principlesin practicearent applied as unanalyzable units rejected or accepted only as wholes. Rather, the shift to an alternative logic isfor certain classical laws of logica change in status from that of a topic-neutral
principle, applying to any subject matter whatsoever, to a topic-specic law
applicable only in special circumstances. Thus much of the specic reasoning
to say, that they must be false and true.) Similarly, Gentzen rules governing connectives
(such as Priors infamous tonk) arent meaningless either (pace Tennant 2005)the rules
governing such a connective allow the inference of every sentence. Such may be pragmatically undesirable, or trivial; but the charge of meaninglessness is a surreptitious violation
of our intuitions about meaning that otherwise clearly allow us to make sense of such
items.
Some may feel the paragraphthis footnote is appended togives attempts to root logic
in meaning short shrift. Quine (1986, p. 81) once wrote in passing: Here, evidently, is the
deviant logicians predicament: when he tries to deny the doctrine he only changes the
subject. Lets grant the point: what does it show? Not much because the issue should be
about the possibility of dropping one set of topic-neutral devices that are constitutive of
inference for another set. This issueconstrued this wayis the focus of many of those who
take alternative logics seriously. Who cares, therefore, whether or not the paraconsistent
or intuitionistic and has the same meaning as the classical and? Whats important,
rather, is (i) that there are (formal) languages in which nonclassical operators play analogous
roles to those played by classical connectives and quantiers, and (ii) that in many cases,
such competing logical operators cannot hold simultaneously sway over the same formal
languages without collapsing into one another. (See Harris 1982 for the case of intuitionistic
and classical logic.)
32 Nagel (1997, p. 61) expresses a Quinean thought as the antecedent of an argument that
Quine himself would reject: Certain forms of thought cant be intelligibly doubted because
they force themselves into every attempt to think about anything.

The Compulsion to Believe: Logical Inference and Normativity

allowed in a classical setting is retained if the logic shiftsbut is now labeled


dierently.33
For those gripped by the intuition of the strong normative role of classical
logic, these suggested explaining aways dont touch something fundamental
in the perception of strong normative role of (classical) logical principles: what
we might call local perceptions of validity. Consider:
All men are mortal,
Socrates is a man,
Therefore: Socrates is mortal.
The forceful intuition of validity accompanying the understanding of this
inference doesnt seem due either to the centrality of the form of reasoning
depictedalthough such a form is central and generalizable (and recognizable as such because of validity-preserving, and yet arbitrary, substitutions for
Socrates, mortal, and man)nor does it seem due to the meaning of the
words All and is. The meaning of these words does enable the expression of
this validity, but it doesnt seem the validity is due (solely) to the words having
these meanings.
It might be thought that the just-described understanding of how logic can
changethat a classical logical principle can lose its status as topic-neutral,
while leaving undisturbed many if not most specic instances of itresponds
to this local intuition of validity. It doesnt. We have the same compulsive feeling with respect to those (specic) inferences that are to be disallowed in the
switch to an alternative logic. When faced with the liar paradox, for example,
the compulsion is to unearth and deny a (perhaps tacit) assumption; in so doing the paradox is hoped to be revealed as only apparent (e.g., the liar-paradoxexpression doesnt express a proposition). One is compelled to reject that the
sentence can be both true and false. Similar remarks apply to the intuitionistic
remodeling of the notion of negation: One feels that double-negation inferences in every case are rightand regardless of the many observations the
intuitionist oers about the supposed treachery of innity.

33 The move from classical logic to intuitionism or to a paraconsistent logic should be seen
this way: e.g., the law of double negationin intuitionismis a law applied only in special
circumstances. See Azzouni and Armour-Garb 2005.

88

Jody Azzouni

5
On the other hand, when those gripped by the impression of the strong normative role of classical logic try to justify that role, they reach for the idea that the
job of (deductive) inference is truth-preservation. Whats special about the
forms of words, and their meanings, when used to express principles of classical
logic, and that isnt special to other forms of words and meaningsthat we
might make upis that (with respect to a classical form of reasoning) if the
premises are true, then the conclusion must be true. So the idea would be that
the strong normative role of classical logic traces back to a semantic property
of the principles of classical logic: that those principles of inference are truth
preserving, and the alternatives arent.34
Unfortunately, the truth idiom is far too promiscuous to exclusively support a
strong normative role for classical logic. Any way we have of characterizing the
truth idiomeither in terms of the laws it obeys (e.g., all instances of Snow
is white, is true if and only if snow is white) or in terms of metaphysically-rich
characterizations of truth involving one or another form of correspondence
(e.g., to facts, or to objects bearing such and such properties, and so on)is
too weak by itself to do any work. One needs the very principles of logic that
are supposedly being given strong normative role by the characterization of
truth-preservation. One might deny this is problematical by invoking Nagels
claim cited in footnote 32. In this case (as its sometimes said) utilizing the
logical principles in the characterization of truth preservation that in turn is to
justify the strong normative role of those same principles is virtuously circular. But virtuous circularity wont procure strong normative role because one
needs to justify classical logical principles against competitors thatit must be
claimedthemselves dont preserve truth. Unfortunately, a notion of truth
preservation can be crafted for any alternative logicif we only substitute the
principles of the alternative logic for the classical ones in the characterization of
what truth preservation comes to. There is no escaping this by suggesting that
the notion of truth preservation so described is dierent for dierent logics,
since its only dierent because of the diering logical principles accompanying
the otherwise same notion of truth.
Its true that what drives belief in the strong normative role of classical logic
is our intuitive sense of the validity of classical principles; in particular, what
34 Ive cast the point in terms of truth-preserving inferences; but it can be cast in terms of
truthif one is thinking of logical principles as statements rather than as licenses for inference. The strategy isessentiallythe same one.

The Compulsion to Believe: Logical Inference and Normativity

we seem to sense is a semantic fact about inferences licensed by classical logical principles, and not merely the syntactic fact that certain rules have been
correctly applied. This is the source of the impression that if the premises of a
syllogism are true, then the conclusion must be true.
But what does this intuition amount to? The modal thought involved (must
be) seems to be: It cant be otherwise. That is, we cant see how the premises
could be true and the conclusion false. But this isnt a positive characterization
of anythingit only expresses that we fail to see how something could be. One
way of trying to give a positive characterization of this intuition is to take it as a
perception of a genuine modality: We recognize that, regardless of how the world
might be, if the premises are true, then the conclusion is true as well.
Attempting to so construe the intuition of the strong normativity of classical logical principles is too demanding (on us) in two ways. First, it takes us
as recognizing (somehow) that varying the world in all sorts of ways (while
keeping the premises true) keeps the conclusion true as well. This, to put it
mildly, seems hard to do.35 Second, when we try to systematically (and rigorously) characterize this suggested route to validityby introducing a semantics
(a model theory) that (in some sense) varies the world in the ways needed, we
again require the use of the very logical principles the characterization of validity is supposed to underwrite. But (also again), since classical logical principles
are being pitted against alternatives, this strategy is useless. (We have no grasp
of how the world can vary apart from a characterization in terms of whatever
logical principles we use: Dierent logical principles allow the world to vary
in ways quite dierent from how the classical principles allow the world to
vary.)
These considerations suggest that if we try to give a positive characterization
of the intuition of the strong normative role of classical logic, we fall back on
the very principles that we are trying to provide a strong normative role for.
This isnt a virtuous circle in a context where proponents of alternative logics
can help themselves to exactly the same strategyand with the same (apparent) degree of success.

35 Do we, as it were, imagine the whole world going through all sorts of variations? Really?

90

Jody Azzouni

6
A sheerly negative intuitionwe cant see how the premises can be true and
the conclusion falseinvites diagnosing the intuition that classical logic has
strong normative role. Thus, the apparent stando between those who explain
away intuitions of strong normative role for classical logic and those who
justify such intuitions is more problematic for the latter.36 In conclusion
thereforeI present one way of so diagnosing these intuitions. This is to take
seriously a point made implicitly earlier, but not so far stated loudly: that we
have no (introspective) grip on the principles we use to reasonother than the
brute sense of compulsion induced in us when we reasonand so (here is the
diagnosis) its no surprise we cant imagine how alternative forms of reasoning
are possible.
Here are the pieces needed to explain away intuitions of strong normative
role for classical logic. First, one gives a nativist explanation for why we feel
the compulsion to reason as classical principles dictate. Such an explanation,
of course, doesnt require that the principles themselves be hardwired in us
neurophysiologically; the view can get by with the weaker assumption that we
have certain (hardwired) mental tendencies, which given the right nurturing,
cause the emergence of dispositions to reason in accord with classical logical
principles. Second, our use of such principles in reasoning remains tacit even
when we know (empirically) what those principles are. That is, our conscious
grasp of these logical principles amounts only to the brute compulsion to believe
that if something is the case then something else must be the case as well.
This suces to explain the impression of strong normative role for classical
principles. If the feeling of brute compulsion is relatively rigidthen even if
we practice the inferences of an alternative logic for the rest of our liveswe
are nevertheless never to have the feeling of understanding (that it must be
this way) that we have when we reason according to the dictates of that compulsion.37
36 Some might think this is unfair. But its hard to see how to sustain the purported perception of strong normative role for classical logical principles except via an argument that a
genuine perception of validity (i.e., truth preservation) is taking place. On the other hand,
Quines attempts to explain away intuitions of strong normative role for classical logic hardly
exhaust the strategic options for those attempting to so explain away such intuitions. The
diagnostician has more philosophical resources than the justier (at least in this case).
37 This dramatic remark should be qualied: The feeling of understanding will be absent
when the dictates of the alternative logic actually deviate (in specic cases) from classical
principles.

The Compulsion to Believe: Logical Inference and Normativity

Its worth adding this last reassuring point: In practice we disallow brute
intuitions of validity, no matter how powerful they are. Our view of certain of
Aristotles syllogisms takes exactly this form: We diagnose the intuition of validity in such cases by locating a special assumption. We are willing to also explain
away (fallacious) probabilistic intuitionsones psychologically every bit as
powerful as the ones that grip us when we reason in accord with classical logical
principles. In practice, we recognize that intuitions of validityno matter how
powerfulare at best prima facie. The normativity that, in other moods, we
presume such intuitions to be indications of, is actually a moving target to be
decided ultimately (and instead) by our (collective) pragmatic needs.38

Bibliography
Azzouni, Jody 1994. Metaphysical myths, mathematical practice: the ontology and epistemology of the exact sciences. Cambridge: Cambridge University Press.
Azzouni, Jody 2004. Proof and ontology in Euclidean mathematics. In New trends in the
history and philosophy of mathematics, edited by T. H. Kjeldsen, S. A. Pedersen, and
L. M. Sonne-Hansen, 11733. Denmark: University Press of Southern Denmark.
Azzouni, Jody 2005. Is there still a sense in which mathematics can have foundations? In
Essays on the foundations of mathematics and logic, edited by G. Sica, 947. Monza,
Italy: Polimetrica International Scientic Publisher.
Azzouni, Jody 2006. Tracking reason: proof, consequence, and truth. Oxford: Oxford
University Press.
Azzouni, Jody, and Bradley Armour-Garb 2005. Standing on common ground. Journal
of Philosophy CII(10): 532544.
Bloor, David 1983. Wittgenstein: A social theory of knowledge. New York: Columbia
University Press.
Dehaene, S. 1991. Numerical cognition. Oxford: Basil Blackwell.
Dehaene, S. 1997. The number sense. Oxford: Oxford University Press.
Frege, G. 1967. The basic laws of arithmetic, trans. J.L. Austin. Oxford: Basil Black38 My thanks to the audience and participants at the logicism session of the 2006 joint meeting
of the North Carolina Philosophical Society and the South Carolina Society for Philosophy, and
to the attending members at the February 25th meeting of the Massachusetts Bay Philosophy
Alliance, where I gave earlier versions of this paper. Especial thanks are due to Nancy Bauer,
Avner Baz, Sylvain Bromberger, Otvio Bueno, Jenn Fisher, Thomas Hofweber, Je McConnell, Sarah McGrath, Michael D. Resnik, and Eric Swanson. I also read penultimate versions
of the paper on April 8, 2006 at the joint meeting of NJRPA and LIPS, and also elded
questions on the paper at a departmental presentation of it at Tufts University. My thanks
to everyone present at both occasions. Finally, my thanks to Agustn Rayo for conversations
related to this topic, and for his drawing my attention to Harris 1982.

92

Jody Azzouni

well.
Hacking, Ian 1973. Leibniz and Descartes: Proof and eternal truths. In his (2002) Historical ontology, 20013. Harvard: Harvard University Press.
Harris, J.H. 1982. Whats so logical about the logical axioms? Studia Logica 41:
15971.
Lakatos, Imre 1976. Proofs and refutations: The logic of mathematical discovery. Cambridge: Cambridge University Press.
Mackenzie, Dana 2005. What in the name of Euclid is going on here? Science 307,
March 4.
Moorehead, Caroline 1992. Bertrand Russell: A life. New York: Viking.
Nagel, Thomas 1997. The last word. Oxford: Oxford University Press.
Plato 1963. The Meno. In The collected dialogues of Plato, edited by E. Hamilton and H
Cairns, 35384. Princeton: Princeton University Press.
Quine, W.V. 1986. Philosophy of Logic, 2nd edition. Harvard: Harvard University
Press.
Tennant, Neil 2005. Rule-circularity and the justication of deduction. The Philosophical Quarterly 55(221): 62548.

Nominalism and Mathematical Intuition

Nominalism and Mathematical Intuition


Otvio Bueno

Abstract
As part of the development of an epistemology for mathematics, some Platonists have defended the view that we have (i) intuition that certain mathematical principles hold, and (ii)
intuition of the properties of some mathematical objects. In this paper, I discuss some diculties that this view faces to accommodate some salient features of mathematical practice. I
then oer an alternative, agnostic nominalist proposal in which, despite the role played by
mathematical intuition, these diculties do not emerge.

1. Introduction
For the purpose of this paper, Ill take mathematical intuition to be any sort
of intuition involved in mathematical activity. The intuition in question may
be invoked in grasping the truth of certain mathematical statements (whether
they are taken to be axioms or not); constructing and evaluating proofs; assessing the cogency of the use of certain pictures, templates, or diagrams in
a proof; or appreciating the reasonableness and fruitfulness of certain mathematical denitions. Clearly, mathematical intuition plays a central epistemological function: its supposed to help us obtain knowledge of certain basic
mathematical facts (typically, those that are described in certain mathematical
principles or axioms). And its common to nd the development of accounts
of mathematical intuition as part of a defense of Platonism. The crucial idea is
that we have intuition of certain mathematical facts}facts about mathematical objects and their relations}and we then extend that basic mathematical
knowledge to other, more complex, recherch facts.
There have been extensive discussions of mathematical intuition in the literature.1 Although I wont review the discussions here, I will examine a prominent
Platonist conception, and raise some diculties that it faces vis--vis mathematical practice. My main goal is to examine whether a certain conception of
mathematical intuition can support a particular, agnostic form of nominalism.
1

For insightful accounts, see Parsons [1980], Parsons [2008] (particularly Chapter 5), and
Giaquinto [2007].

94

Otvio Bueno

If the considerations discussed below are nearly correct, theres no reason to


think that mathematical intuition is inevitably linked to a Platonist picture.
Along the way, a new way of conceptualizing nominalism emerges.

2. Mathematical intuition
2.1. Mathematical intuition: Some general features and a dilemma
What is the role of mathematical intuition? To answer this question, we need
rst to be clear about what we take intuition to be. What follows, although
certainly not comprehensive, should give us a rough indication of some features
involved in an account of intuition.
First, whatever intuition turns out to be, it is certainly fallible: it oers no
conclusive account of the truth of the mathematical statements under consideration. Similarly to any other cognitive process, intuition may turn out
to be mistaken after all. This is not a huge constraint, given that fallibilism
is the rule, rather than the exception, in epistemological matters. A fallibilist
stance includes mathematics, despite attempts to provide infallible strategies
of knowledge generation in this domain.
Second, what is the object of intuition, that is, to which kinds of things does
intuition apply? It seems that it applies to concepts, but also to constructions,
mental models, inscriptions, and patterns of various kinds. If we think of
intuition as something that applies to concrete entities, it becomes unproblematic how we can have intuition of so many things. However, it also becomes
unclear how intuition of concrete objects can give us any grounds for belief
in claims about abstract entities}those that are referred to in statements of
mathematical theorems.
As a result, a dilemma emerges at this point: Either intuition applies to
concrete entities, or it doesnt. Suppose that intuition does apply to concrete
entities, that is, we form mathematical intuitions by considering concrete entities (such as templates, inscriptions, diagrams, and drawings). In this case, its
unclear how exactly the intuition of such entities can provide any information
about an independently existing domain of abstract entities or structures. After
all, there is no information channel between the concrete entities we experience
(e.g. a diagram) and the mathematical objects and relations that these concrete
entities stand for (e.g. the particular relations among the elements of a group).

Nominalism and Mathematical Intuition

Obviously, there is no causal connection between the concrete and the abstract
objects in question. Its then not clear just how intuition of concrete objects
can provide information about causally inert abstract objects and relations. In
fact, we may not be able to maintain even something weaker, namely, that our
intuitions of concrete entities generate grounds for belief in the existence of
abstract entities and structures, or can justify claims about them.
Alternatively, if intuition doesnt apply to concrete entities, but only to abstract ones (such as Fregean concepts, mathematical structures, etc.), it becomes
unclear how we can have such an intuition in the rst place. How does the
intuition of an abstract object operate? And how can it give us knowledge, or at
least justication, of our claims about mathematical objects and structures?
To answer worries of this kind, Platonists have developed detailed accounts
of mathematical intuition. Ill consider a prominent proposal in turn.

2.2. Mathematical intuition: Robustness and Gdel


For Kurt Gdel, the truth of basic mathematical axioms can be obtained directly by intuition (see Gdel [1964], and Maddy [1990]). In fact, not only the
principles of arithmetic, but also the axioms of set theory can be apprehended
directly by intuition. We have, Gdel claims, something like a perception of
the objects of set theory ([1964], p. 485). That is, we are able to perceive
these objects as having certain properties and lacking others, in a similar way
that we perceive physical objects around us. That we have such a perception of
set-theoretic objects is supposed to be seen from the fact that the axioms [of
set theory] force themselves upon us as being true (Gdel [1964], p. 485).
But how exactly does the fact that set-theoretic axioms force themselves
upon us support the claim that we perceive the objects of set theory? Gdel
seemed to have a broad conception of perception, and when he referred to the
objects of set theory, he thought that we perceived the concepts involved
in the characterization of these objects as well. The point may seem to be
strange at rst. With further reection, however, it is not unreasonable. In
fact, an analogous move can be made in the case of the perception of physical
objects. For example, in order for me to perceive a tennis ball, and recognize
it as a tennis ball, I need to have the concept of tennis ball. Without the latter
concept, at best Ill perceive a round yellow thing}assuming I have these concepts. Similarly, I wouldnt be able to recognize certain mathematical objects
as objects of set theory unless I had the relevant concepts. The objects couldnt

96

Otvio Bueno

be perceived to be set-theoretic except if the relevant concepts were in


place.
Now, in order to justify the perception of set-theoretic objects from the
fact that the axioms of set theory are forced upon us as being true, Gdel needs
to articulate a certain conception of rational evidence (see Parsons [2008],
pp. 14648). On Gdels view, in order for us to have rational evidence for a
proposition}such as an axiom of set theory}we need to make sense of the
concepts that occur in that proposition. In making sense of these concepts, we
are perceiving them. Mathematical concepts are robust in their characterization, in the sense that what they stand for is not of our own making. Our
perception of physical objects is similarly robust. If there is no pink elephant
in front of me right now, I cannot perceive one.2 And you cannot fail to perceive the letters of this sentence as you read it, even though you might not be
thinking about the letters as you read the sentence, but what the latter stands
for. The analogy between sense perception and the perception of concepts is
grounded on the robustness of both.
The robustness requires that I perceive what is the case, although I can, of
course, be mistaken in my perception. For instance, as I walk on the street, I see
a bird by a tree. I nd it initially strange that the bird doesnt move as I get closer
to the tree; just to nd out, when I get close enough, that there was no bird
there, but a colorful piece of paper. I thought I had perceived a bird, when in
fact I perceived something else. The perception, although robust}something
was perceived, after all}is fallible. But I still perceived what was the case: a
piece of paper in the shape of a bird. I just mistook that for a bird, something
I corrected later. Similarly, the robustness of our perception of the concepts
involved in an axiom of set theory is part of the account of how that axiom can
force itself upon us as being true. By making sense of the relevant set-theoretic
concepts, we perceive the latter and the connections among them. In this
way, we perceive what is the case among the sets involved. Of course, similarly
to the case of sense perception, we may be mistaken about what we think we
perceive}that is part of the fallibility of the proposal. But the perception is,
nevertheless, robust. We perceive something that is true.
This account of perception of mathematical concepts and objects is, in fact,
an account of mathematical intuition. Following Charles Parsons ([2008], pp.
138143), we should note that we have intuition of two sorts of things. We have
2

I can imagine one, but thats an entirely dierent story, since imagination and perception have
very dierent functions, and each has its own phenomenology. I can perhaps hallucinate that
there is a pink elephant in front of me, but again that wouldnt be to perceive an elephant.

Nominalism and Mathematical Intuition

intuition of objects (e.g. the intuition of the objects of arithmetic), and we have
intuition that some proposition is true (e.g. the intuition that the successor of
a natural number is also a natural number is true). The former can be called
intuition of, and the latter intuition that. In the passage quoted in the rst
paragraph of this section, Gdel seems to be using the intuition that the axioms
of set theory force themselves upon us to support the corresponding intuition
of the objects of set theory. The robustness of both intuitions involved here is
a central feature of the account.
The Gdelian overall strategy involves two parallel steps. First, we have
the intuition of basic mathematical facts, through which the truth of certain
mathematical axioms and the existence of the corresponding objects are apprehended. But not every mathematical fact can be apprehended in this way.
In some cases, we may not have any clear intuition of the truth of the relevant
principles. A second step is then invoked. We draw consequences from the relevant mathematical principles, and assess the signicance of the results that are
established based on such principles. If by invoking the latter signicant results
are proved, the principles will receive indirect conrmation}analogous to the
conrmation that empirical hypotheses receive at the end of an experiment in
science. With this second step, although the truth of the relevant mathematical
principles is not established, at least the consequences obtained provide some
support for the principles involved. Taken together, we have here a Platonist
epistemology for mathematics.

2.3. Troubles with mathematical intuition: A Fregean predicament


Some diculties, however, still need to be addressed. They emerge from the
Platonist commitments involved in the proposal. Despite the robustness of
mathematical intuition, the issue arises as to whether we have good grounds
to believe that the objects we take to have intuition of are indeed the objects of
that intuition. In other words, is it possible that although we believe we have
intuition of certain mathematical entities, we are, in fact, apprehending other
such objects, or perhaps none at all? Given the admitted fallibility of mathematical intuition}the fact that the outcome of an intuition may turn out
to be false}its not clear that these possibilities can be excluded. But if these
possibilities cant be ruled out, the Platonist account doesnt deliver the result it
was supposed to oer. Ill illustrate the diculties by exploring two signicant
cases in mathematical practice: one is a foundational study in arithmetic (the

98

Otvio Bueno

Fregean predicament); the other is a study in the early development of the


calculus (with the introduction of innitesimals by Leibniz).
Despite the robustness of mathematical intuition, its possible that we have
intuition of certain mathematical principles that turn out to be inconsistent.
So, although we think we are apprehending the truth of a certain principle, we
are, in fact, dealing with inconsistent objects. I think this was precisely Freges
predicament. In his logicist reconstruction of arithmetic, Frege tried to show
how natural numbers could be constructed from logic and denitions alone
(Frege [1974]). The construction is very ingenious. Frege started from concepts (which, on his view, are abstract, mind independent entities), and their
extensions, comprised by the objects that fall under that concept. Roughly,
the objects that have the property described by a concept are those that fall
under that concept. Consider the concept of a natural number. Ask yourself:
how many objects fall under the concept not identical to itself? The answer, of
course, is zero. That characterizes the number zero. How many objects fall
under the concept identical to zero? The answer, once again, is clear: only one
object (namely, 0). We have now characterized the number one. How many
objects fall under the concept identical to one or zero? Precisely two objects: 0
and 1. That characterizes the concept 2. And so on. This is, very briey, Freges
characterization of natural numbers, and the crucial work is in specifying precisely how the and so on goes (Frege [1974]).3
Central to Freges strategy was the use of what is now called Humes Principle: two concepts are equinumerous if, and only if, there is a one-to-one
correspondence between them. Humes Principle was used at various crucial
points; for instance, to show that the number 0 is dierent from the number
1. As we saw, the concept 0 is characterized in terms of the number of objects
that fall under the concept not identical to itself, and the concept 1, in turn, is
characterized in terms of the concept identical to 0. Now, given that nothing
falls under the concept not identical to itself, and only one object falls under
the concept identical to 0, by Humes Principle, these two concepts are not
equinumerous. As a result, 0 is distinct from 1.
But how can one establish that Humes Principle is true? Frege thought
he could derive Humes Principle from a basic logical law (called Basic Law
V). According to this law, the extension of the concept F is the same as the
extension of the concept G if, and only if, the same objects fall under the
3

Operations over natural numbers, such as addition and subtraction, can also be characterized,
and arithmetic can be perfectly developed in this way (see Frege [1974], Boolos [1998], and
Hale and Wright [2001]).

Nominalism and Mathematical Intuition

concepts F and G. Basic Law V seemed to be a fundamental logical law, dealing with concepts, their extensions, and their identity. It had the right sort of
generality and analytic character that was needed for a logicist foundation of
arithmetic.
There is only one problem. Basic Law V turns out to be inconsistent. It immediately raises Russells paradox if we consider the concept is not a member
of itself . To see why this is the case, suppose that there is such a thing as the
set composed by all the sets that are not members of themselves. Lets call this
set R (for Russell). Now lets consider whether R is a member of R. Suppose
that it is. In this case, we conclude that R is not a member of R, given that, by
denition of R, Rs members are those sets that are not members of themselves.
Suppose, in turn, that R is not a member of R. In this case, we conclude that R is
a member of R}since this is precisely what it takes for the set R to be a member
of R. Thus, R is a member of R if, and only if, R is not a member of R. It then
immediately follows that R is and is not a member of R}a contradiction.
Someone may say that this argument just shows that there isnt such a thing
as the Russell set R after all.4 So, what is the big deal? The problem is that, as
Russell also found out, it follows from Freges system using suitable denitions that there is a set of all sets that are not members of themselves. Given
the argument above establishing that there isnt such a set, we have a contradiction. Freges original reconstruction of arithmetic in terms of logic was in
trouble.
This is a familiar case. But it raises a problem for the epistemological model
of mathematical intuition. Did Frege have an intuition of the truth of Basic
Law V? If so, we would have to conclude that mathematical intuition can be
highly unreliable}unless we are committed to the truth of inconsistent principles, which was certainly not the case with Frege (or Gdel). If on this model
of mathematical intuition, Frege didnt have an intuition of the truth of Basic
Law V, we need to be given an independently motivated account of why no
intuition was involved here. It would certainly be ad hoc to claim that no intu4

I am assuming here, with Frege and Russell, that the underlying logic is classical. In particular,
classical logic has the feature that everything follows from a contradiction}a principle that
is often called explosion. However, there are non-classical logics in which this is not the case;
that is, on these logics, not everything follows from a contradiction. These logics are called
paraconsistent (see, e.g., da Costa, Krause, and Bueno [2007], and Priest [2006]). If we
adopt a paraconsistent logic, we can then study the properties of the Russell set in a suitable
paraconsistent set theory (see, again, da Costa, Krause, and Bueno [2007]). Of course, one
need not be a Platonist about such a set}or any other, for that matter (see, e.g., Azzouni
[2004]).

100

Otvio Bueno

ition was at play since the principle in question was inconsistent. After all, the
inconsistency of Basic Law V was discovered only after Freges reconstruction
of arithmetic had already been developed; hence, the need for an independent
explanation. In either case, however, we have problems.
But not everything was lost. Although Frege acknowledged the problem that
Russell raised, and tried to x it by introducing a new, consistent principle, his
solution ultimately didnt work. After all, the principle that Frege introduced as
a replacement for Basic Law V, although logically consistent, was inconsistent
with the claim that there are at least two distinct numbers. Since the latter claim
was true in Freges system, the proposed principle was clearly unacceptable (for
an illuminating discussion, see Boolos [1998]). There was, however, a solution
available to Frege. He could have jettisoned the inconsistent Basic Law V, and
adopted Humes Principle as his basic principle instead. Given that the only
use that Frege made of Basic Law V was to derive Humes Principle, if the latter
were assumed as basic, one could then run, in a perfectly consistent manner,
Freges reconstruction of arithmetic. In fact, we could then credit Frege with
the theorem to the eect that arithmetic can be derived in a system like Freges
from Humes Principle alone. Freges approach could then be extended to other
branches of mathematics.5
Could the truth of Humes Principle be apprehended by mathematical
intuition? As we saw, according to this principle, two concepts are equinumerous if, and only if, there is a one-to-one correspondence between them.
Humes Principle has essentially the same form as Basic Law V, according
to which, as noted, the extension of the concept F is the same as the extension of the concept G if, and only if, the same objects fall under the concepts F and G. The principles provide conditions for sameness of number
or sameness of extension for concepts, by specifying suitable conditions on
the extension of such concepts: respectively, the existence of a one-to-one
correspondence between the concepts, or the same objects falling under the
relevant concepts. Given the structural similarity between these principles,
it is hard to see, without hindsight, how we could know one of them by intuition and fail to know the other. But given that these principles are indeed
signicantly dierent (one is inconsistent, the other is not!), an account is
needed as to how such a dierence could be found based on mathematical
intuition.
5

To implement a program along these lines is one of the central features of the neo-Fregean
approach to the philosophy of mathematics (see, e.g., Hale and Wright [2001], and, for a
discussion, Boolos [1998]).

Nominalism and Mathematical Intuition

Of course, Frege himself didnt think that intuition was needed to accommodate our knowledge of the principles of arithmetic. After all, he thought
that arithmetic was analytic, and not synthetic, and hence there was no place
for such an intuition. The point of this example is to indicate a diculty for
the Platonist who claims that mathematical principles, such as those articulated
in arithmetic, are known by mathematical intuition. Given the inconsistency
involved in Freges account, to have a mathematical intuition of the truth of
Basic Law V would amount to a case in which we think that we are apprehending certain objects when, in fact}if we assume a consistent approach
to arithmetic}no such objects are being apprehended. After all, there are
no inconsistent objects to be apprehended. Alternatively, if an inconsistent
approach to arithmetic is not ruled out, when we apprehend the truth of
Basic Law V we would then apprehend the truth of an inconsistent principle! Perhaps for those who believe in the existence of true contradictions
(see Priest [2006]) this is not a problem. But it will be a problem for everyone
else.

2.4. Troubles with mathematical intuition: Thinking about innitesimals


Consider now the case of innitesimals (see Robinson [1967], Bell [2005],
Bueno [2007], and Colyvan [2008]). Introduced in the 17th century as part of
the early development of the calculus, innitesimals were thought of as positive
numbers that were smaller than any number. They were introduced as computational devices, in order to help the derivation of certain results, and were
thought of by Leibniz, for example, as useful ctions}helpful constructs to
whose existence we should not be committed (Leibniz [1716]). Part of the skepticism about innitesimals emerged from their peculiar behavior. In some contexts, they were indeed thought of as strictly positive numbers. But since their
value was so small}in fact, smaller than any positive number}innitesimals
were also thought of as being zero. The trouble is that these two inconsistent
characterizations were often found in the same derivation!
As an illustration, Ill examine a typical example of the use of innitesimals
in the 17th century, in dierentiation from rst principles, before the modern
theory of limits had been developed (see Colyvan [2008] for a fascinating
discussion). Consider, for instance, the dierentiation of a polynomial of the
form: f(x) = ax2 + bx + c. The dierentiation has the general form (where D is
an innitesimal):

102

Otvio Bueno

(Di) f a(x) = f(x +D) f(x)



D

If we now apply the polynomial above to (Di), we have:


f a(x) = a(x + D)2 + b(x + D) + c (ax2 + bx + c)



D
= 2axD +D2 + bD
D
= 2ax + b + D
= 2ax + b

In the rst two lines of this derivation, D was denitely taken as a number different from zero, otherwise it would not be possible to divide by it. However,
in the next to last line, D was clearly taken to be zero, otherwise the identity
with the last line would be false. Thus, D was both identical with zero and nonidentical with zero. We have an inconsistency.
Given the inconsistent properties of innitesimals, its not surprising that
their status has been controversial from the start. Even Leibniz acknowledged
the need for devising a strategy to dispense with such objects (Leibniz [1701]).
Without care, the inconsistency of innitesimals could lead to trouble. After
all, if an innitesimal D can be taken as zero in some contexts, we could have
that:
1xD=2xD
And then, if we assume that D is non-zero, we could then divide by D, and
establish that:
1 = 2 (!).
But this is clearly unacceptable. Some rules are needed to manipulate such
objects in order to avoid triviality.
Its important to note that despite the peculiar behavior of innitesimals,
mathematicians in the 17th century didnt derive mistaken results such as the
one above. They had rules about which inferences involving innitesimals were
allowed and which were blocked. These rules, however, clearly needed to be
properly motivated. But how could that be done?

Nominalism and Mathematical Intuition

When Leibniz introduced, in 1684, the dierential calculus in his Nova


Methodus, he formulated the concept of dierential without mentioning innitesimals. In this way, presumably he could try to avoid foundational worries
about the latter. He introduced a group of rules of dierentiation, adopting,
thus, an algebraic approach to the subject. Interestingly enough, he never
oered a proof for the rules (Bell [2005]). They are:
da
d(ax)
d(x + y z)
d(xy)
d(x/y)
d(xp)

= 0, where a is a constant;
= a dx, where a is a constant;
= dx + dy dz
= x dy + y dx
= x dy y dx
y2
= pxp 1dx, also for fractional p.

But it turns out that innitesimals were still presupposed, given the particular
interpretation of the formalism oered by Leibniz in terms of suitable curves.
In fact, if a curve is determined by the variables x and y, then Leibniz took dx
and dy to be innitesimal dierences, or dierentials, between the values x and
y. In this case, dx/dy was taken as the ratio of the two, and for Leibniz this is
the slope of the curve at the corresponding point. Moreover, Leibnizs denition of tangent uses innitely small distances (see Bell [2005]). So, in the end,
by thinking about potentially inconsistent innitesimals, Leibniz was able to
devise correct rules of the dierential calculus.
Is it possible that the rules above were reached by mathematical intuition?
The rules are, of course, formally correct. But the process by which they were
generated involved objects}the innitesimals}that can behave inconsistently. Can we have robust intuitions about inconsistent objects? Its unclear
whether we can. And even if we could, how reliable would such intuitions be?
Given the inconsistent nature of the objects in question, the intuitions are
likely to be inconsistent.
Alternatively, we could explore the idea that innitesimals become inconsistent only if their dual nature}being zero, and being strictly greater than zero}were invoked simultaneously. Every old person was once
young (that is, non-old), but that doesnt entail that someone is both old
and young. As long as the inconsistent components are never used simultaneously, perhaps there is no trouble. But, as we saw, using the inconsistent components at the same time was part of the way in which inni-

104

Otvio Bueno

tesimals were sometimes put to work. So, this maneuver will not help in
general.
Finally, perhaps we can think that we are having an intuition of the consistent components of innitesimals, while we are actually having an intuition
of the consistent behavior of a dierential mathematical operator, such as the
one satisfying Leibnizs rules above. In this case, even though the result of the
intuition is formally correct, this is just a fortuitous accident, given that the
intuition is not of the objects that we think we are having an intuition of. Since
all of these cases raise diculties for the reliability of mathematical intuition
as an intuition of existing mathematical objects, they are problematic for the
Platonist.

3. Agnostic nominalism and mathematical intuition


Despite its shortfalls, can the account of mathematical intuition discussed
above help a nominalist about mathematics? With suitable amendments, I
think it can. But, to begin with, its important to be clear about what nominalism is. I dont take it to be the negative, skeptical claim to the eect that
mathematical objects do not exist}although this is the usual formulation of
the view (see, e.g., Field [1980] and Field [1989]). On the conception I favor,
nominalism is an agnostic, not a skeptical, proposal. After all, its not clear
how we could establish the non-existence of mathematical objects}particularly
without begging the question against the Platonist, who of course insists that
such objects exist. The fact that we have no causal access to mathematical
objects, nor can we locate them in space and time, gives us no reason to believe in the non-existence of such objects. As the Platonist would remind us,
mathematical objects are not supposed to be identied in this way. And if one
insists that the only existing things are those that are causally accessible to us,
we would simply be begging the question against the Platonist, who has less
restrictive ontological constraints.
But perhaps one could try to establish a weaker claim. Rather than insisting
that mathematical objects do not exist, we could argue that we dont have good
reason to believe in their existence. But even if the latter claim were established,
it wouldnt settle the issue regarding the non-existence of mathematical entities.
Given that mathematical objects are taken to be independent of us, the lack of
reason for us to believe in their existence is not enough to guarantee their non-

Nominalism and Mathematical Intuition

existence. Moreover, the fact that such entities are causally isolated and are not
located in space and time provides additional reasons to think that to establish
the existence of these entities, we dont invoke any sort of empirical investigation. In any case, we shouldnt expect their existence to depend on us.
A move along these lines was oered by Hartry Field, by undermining what
he took to be the only non-question-begging argument for the belief in the
existence of mathematical objects: the indispensability argument (see Field
[1980]). According to this argument (see Quine [1960], Putnam [1971], and
Colyvan [2001]), we ought to be ontologically committed to all and only those
entities that are indispensable to our best theories of the world, and (reference
to) mathematical entities is indispensable to these theories. Field [1980] tried
to undermine the claim that reference to mathematical entities was indispensable to our best theories, by providing a particular reformulation of Newtonian
gravitational theory in which no quantication over mathematical entities was
involved. However, even if Field succeeded in the case of Newtonian physics,
its unclear how to extend his approach to other physical theories, particularly
quantum mechanics (see, e.g., Malament [1982], and Bueno [2003]). And even
if this worked out, and Field successfully showed how to dispense with reference to mathematical objects in applied mathematics, its unclear that this
would have actually showed that mathematical objects dont exist. At best, this
shows that we dont need to refer to them when doing applied mathematics. For
the traditional Platonist, however, the existence of these objects is independent
of any use we may happen to make of them.
The central point of agnostic nominalism is that we need not take mathematical objects to exist in order to make sense of mathematical practice. Perhaps mathematical objects exist, perhaps they dont. Nothing in mathematical
practice seems to require the existence of these entities. This agnostic form
of nominalism has signicant virtues over the traditional characterization of
nominalism, in that its not committed to the non-existence of mathematical
objects, and the diculty of actually establishing that such objects do not exist
without begging the question against the Platonist.
Suppose that we claimed that the things that exist are only those that are ontologically independent of our linguistic practices and psychological processes
(see Azzouni [2004]). This is, clearly, a broader and much more reasonable
criterion for existence than the causal accessibility criterion we discussed above.
But, if used against the Platonist, its not clear that it fares much better than the
latter criterion. After all, on the Platonists conception, mathematical objects
are ontologically independent of us. We dont make them up by our linguistic

106

Otvio Bueno

practices, given that, for the Platonist, numbers, sets, and functions would
have existed even if no humans did. Nor do we create mathematical objects by
thinking of them; after all, for the Platonist, their existence is not dependent
on us. So, if we tried to use ontological independence as an argument against
the Platonist, we would have to conclude that mathematical objects do exist!
Thus, invoking such a criterion of existence doesnt succeed as a refutation of
Platonism; quite the contrary.
None of these problems are faced by agnostic nominalism, since the view
is not committed to establishing the falsity of Platonism. The view is, after
all, agnostic about the existence of mathematical objects. Moreover, for the
agnostic nominalist, settling the latter issue is not required to make sense of
mathematics and its practice, given that the existence of mathematical objects
plays not role in the description of that practice.6
We can now return to the discussion of mathematical intuition, indicating
in which way it can support a nominalist view}at least of the agnostic sort.
The central idea, from the viewpoint of agnostic nominalism, is to emphasize
that the intuitions involved in mathematical practice emerge from concrete
objects: inscriptions, drawings, diagrams, templates. It is from the consideration of these entities that mathematicians are able to form intuitions about
mathematical objects and relations. Of course, because the agnostic nominalist is dealing with concrete objects, no steps are taken beyond the boundaries
of nominalism. But how exactly can mathematicians form intuitions about
abstract objects and relations by considering concrete entities? Isnt there a gap
between the concrete inscriptions and the abstract pattern that the inscriptions
partially represent? As we saw, the Platonist account of mathematical intuition
was meant to overcome this diculty, trying to oer a story as to how concrete
inscriptions can reliably stand for abstract structures.
But there is a gap to be bridged only if there are abstract objects in need of
being grasped. For those who are not committed to the Platonist picture, the
issue doesnt even arise. As noted, for the agnostic nominalist, not being committed to the existence of mathematical objects doesnt entail that such objects
dont exist, only that their existence plays no role in how we come to know
certain mathematical facts. But what are mathematical facts in this case? The
6

If reference to existing mathematical objects turns out to be unnecessary to make sense of


mathematical practice, shouldnt we simply deny that such objects exist? Again, I dont think
we are entitled to this conclusion. After all, whether mathematical objects are useful or not
in mathematical practice is a pragmatic matter. These objects may turn out to exist despite
their dispensability to us.

Nominalism and Mathematical Intuition

agnostic nominalist understands these facts as facts about what follows from
certain assumptions regarding a given domain. The domain is specied by the
introduction of some comprehension principles, principles that determine the
meaning of the mathematical terms involved, and how to operate with the
mathematical concepts in question. Mathematical facts are not understood
as facts about existing mathematical objects, but as facts about what follows
from certain principles and relations among concepts (which are, of course,
not understood as abstract entities).
Consider, for example, the concept of a metric. In order to operate mathematically with this concept, we need rst to characterize it, that is, specify
which conditions the concept satises. But before we can introduce sensibly the concept of a metric, we need to have already introduced some other
concepts}in particular, in the usual study of metric spaces in real analysis, the
concept of real numbers. Of course, to introduce these concepts, suitable comprehension principles would have to be introduced as well. The procedure is
ubiquitous in mathematics. So, lets assume a background system of real analysis with its suitable comprehension principles. We can then specify a metric d
as a two-place function dened on the Cartesian product of a nonempty set S
with values in the set of real numbers, as follows: d is always positive (d(x, y)
r 0, for every x and y in S); d has value 0 precisely when its arguments are the
same (d(x, y) = 0 if, and only if, x = y), and d satises the properties of symmetry
(d(x, y) = d(y, x), for every x and y in S), and triangle inequality (d(x, z) + d(z,
y) r d(x, y), for every x, y and z in S).
Once these conditions are formulated, we can then determine what follows
from them. We can determine some of the facts about a metric. These facts
will depend on a number of additional components. In particular, the facts will
depend on the logic that is assumed in the derivations}or, given that a logic
is hardly ever made explicit in mathematical practice, on some rough inference
principles that are invoked. The facts will also depend on additional denitions, which introduce new concepts and rene old ones, and these concepts
are then used in proofs and in the statement of theorems. Finally, the facts in
some cases will depend on additional conditions specied in the assumptions
of a theorem.
For example, after formulating the notion of a metric, we can introduce the
concept of a metric space: the pair (S, d), where S is a non-empty set and d
is a metric. We can also introduce the concept of a sequence {xk} in S, and of
convergence of the sequence {xk} to x in S with respect to the metric d. The
sequence converges as long as limk d(xk, x) = 0. We can then prove that if a

108

Otvio Bueno

sequence in a metric space converges, it converges to a unique point; that is, if


lim xk = x and limxk = y, then x = y. After all, given that the metric satises the
triangle inequality, d(x, xk) + d(xk, y) r d (x, y), for all k, it follows that d(x, y) =
0. Thus, since d (x, y) = 0 if, and only if, x = y, we obtain the result.
This is, of course, a simple fact about sequences in a metric space. And its
very tempting to think of it as a fact about objects, such as sequences and metric
spaces. There is nothing wrong with that as long as we dont reify the objects,
and suddenly start thinking that they exist. The objects are indeed introduced
via the relevant comprehension principles, but their existence plays no role
in the account. What matters is how the objects have been characterized and
which conclusions can be drawn about them. The facts in question are facts
about what follows from the comprehension principles involved.
At this point, mathematical intuitions can be helpful. Mathematicians
can have such intuitions about a fruitful way of introducing mathematical
denitions}one that seems to lead to more interesting, powerful results.
Deep theorems that originally emerged from certain mathematical denitions
are eventually taken as new denitions of the mathematical concepts under
consideration (several concepts in real analysis illustrate this situation). By exploring the intuition of these concepts, mathematicians can determine which
among various denitions seems to be the most fruitful}although, in many
cases, there is no obvious candidate that ts the bill.7
So, on the agnostic nominalist picture, when we have a mathematical intuition we have an intuition of some basic mathematical relations among objects
that were introduced via suitable comprehension principles. For the agnostic
nominalist, objects that are so introduced are not taken to exist independently
of the comprehension principles in question, given that without the latter, the
conditions that characterize the objects are not specied. Moreover, objects
introduced via comprehension principles are introduced as entities that are
not located in space-time. Since, typically, there is no specication for time
or space in the comprehension principles, the agnostic nominalist can explain
why mathematical objects are, thus, introduced as abstract objects.8 As a result,
7

Diagrams are often used in these contexts to help gure out the consequences of certain denitions. In some cases, diagrams can be as eective as usual, linguistic proofs, even though,
typically, a proof}particularly of complex, subtle results}will not rest simply on a diagram.
(For a fascinating study of the role of visual thinking in various branches of mathematics, see
Giaquinto [2007].)
Incidentally, this feature of the comprehension principles also allows the agnostic nominalist to explain why mathematical theories are not typically refuted by experiments. Since the
terms in these theories typically do not refer to events in space and time, they need to be

Nominalism and Mathematical Intuition

the agnostic nominalist can verbally agree with the Platonists suggestion to
the eect that mathematical objects are abstract}or would be if these objects
happen to exist.
How can an agnostic nominalist use the}suitably nominalized}account of
mathematical intuition to accommodate the diculties faced by the Platonist
in light of examples such as Freges reconstruction of arithmetic and the use of
innitesimals in the early development of the calculus? Given that, on the agnostic nominalist picture, the commitment to the existence of abstract objects
drop out of the picture, we are no longer committed to believing that we have
intuition of objects that exist independently of the relevant comprehension
principles. So, the intuitions are tied to the principles in question, and not to
independently existing objects. Thus, we cannot have intuition of some other
objects but those that are introduced by the relevant comprehension principles,
since the principles introduce the objects. The mismatch found in the two cases
above can only emerge in a Platonist picture that posits mathematical objects
that exist independently of any description. Once this picture is no longer
in place, the diculties faced by the Fregean predicament and by the use of
innitesimals dont get o the ground.
But can we have intuition of inconsistent mathematical objects? On an agnostic nominalist picture, nothing precludes this possibility. If the comprehension principles that guide the introduction of certain mathematical objects turn
out to be inconsistent, we will be dealing with inconsistent objects. (But, on the
agnostic nominalist view, we need not be committed to the existence of these
objects!) In fact, this was the Fregean predicament, and what was going on when
innitesimals were originally introduced. Interestingly, in none of these cases,
triviality emerged. That is, the results that were actually established}Freges
derivation of arithmetic from Humes Principle and Leibnizs introduction of
the rules of dierentiation}were correct, in the sense that they can be derived
in consistent theories. As paving the way for consistent successors, intuition of
inconsistent objects may turn out to be quite helpful.
But without the commitment to the existence of mathematical objects, what
is the point of invoking mathematical intuition? As noted above, as part of the
development of certain mathematical theories, particularly in the introduction
of suitable comprehension principles and in the renement of mathematical
concepts, mathematical intuition does play a role. But this role need not be
in the justication of mathematical results, if we assume that the justication
suitably interpreted in order to have any empirical import.

110

Otvio Bueno

requires the existence of the objects in question. Its a role in formulating the
principles to begin with (e.g. by selecting alternative formulations), and exploring what follows from them.

4. Conclusion
I started the discussion above by identifying a few central features of a Platonist
account of mathematical intuition. In particular, I highlighted the robustness
of such an intuition as a crucial component of its epistemological signicance.
I then discussed some diculties faced by the Platonist account in light of
some salient features of mathematical practice. I examined the possibility that
we stumble into inconsistency when having an intuition of certain objects (as
in the Fregean case), and the diculty of explaining how we can have correct
mathematical intuitions about objects that may have inconsistent properties
(as in the case of the use of innitesimals in the early development of the
calculus).
As an alternative, an agnostic nominalist view was oered in which mathematical intuition played a role in selecting suitable denitions and in shaping
comprehension principles. But given that the agnostic nominalist does not
assume the existence of mathematical objects, the diculties raised against
the Platonists account of mathematical intuition do not emerge. At least in
outline, an alternative, nominalist conception of mathematical intuition is
then available.

References
Azzouni, J. [2004]: Deating Existential Consequence: A Case for Nominalism. New
York: Oxford University Press.
Bell, J.L. [2005]: The Continuous and the Innitesimal in Mathematics and Philosophy.
Milan: Polimetrica.
Benacerraf, P., and Putnam, H. (eds.) [1983]: Philosophy of Mathematics: Selected Readings. (2nd edition.) Cambridge: Cambridge University Press.
Boolos, G. [1998]: Logic, Logic, and Logic. Cambridge, MA: Harvard University
Press.
Bueno, O. [2003]: Is It Possible to Nominalize Quantum Mechanics?, Philosophy of
Science 70, pp. 14241436.

Nominalism and Mathematical Intuition

Bueno, O. [2007]: Incommensurability in Mathematics, in Van Kerkhove and Van


Bendegem (eds.) [2007], pp. 83105.
Colyvan, M. [2001]: The Indispensability of Mathematics. New York: Oxford University
Press.
Colyvan, M. [2008]: Who Is Afraid of Inconsistent Mathematics?, forthcoming in
this issue of Protosociology.
da Costa, N.C.A., Krause, D., and Bueno, O. [2007]: Paraconsistent Logics and
Paraconsistency, in Jacquette (ed.) [2007], pp. 79 1911.
Field, H. [1980]: Science without Numbers: A Defense of Nominalism. Princeton, N.J.:
Princeton University Press.
Field, H. [1989]: Realism, Mathematics and Modality. Oxford: Blackwell.
Frege, G. [1974]: Foundations of Arithmetic. (Translated by J.L. Austin.) Oxford: Basil
Blackwell.
Giaquinto, M. [2007]: Visual Thinking in Mathematics. Oxford: Oxford University
Press.
Gdel, K. [1964]: What is Cantors Continuum Problem?, in Benacerraf and Putnam
(eds.) [1983], pp. 470485.
Hale, B., and Wright, C. [2001]: The Reasons Proper Study. Oxford: Oxford University
Press.
Jacquette, D. (ed.) [2007]: Philosophy of Logic. Amsterdam: North-Holland.
Lakatos, I. (ed.) [1967]: Problems in the Philosophy of Mathematics. Amsterdam: NorthHolland.
Leibniz, G.W. [1701]: Mmoire de M.G.G. Leibniz touchant son sentiment sur le
calcul direntiel. Journal de Trvoux, in Leibniz [1858], p. 350.
Leibniz, G.W. [1716]: Letter to Dangicourt, sur les monades et le calcul innitsimal
etc. (September 11), in Leibniz [1789], vol. 3, pp. 499502.
Leibniz, G.W. [1789]: Opera Omnia. (Edited by L. Dutens.) Geneva.
Leibniz, G.W. [1858]: Mathematische Schriften. (Volume 6 of Die Philosophischen Scriften von G.W. Leibniz, edited by C.I. Gerhardt.) Berlin.
Maddy, P. [1990]: Realism in Mathematics. Oxford: Clarendon Press.
Malament, D. [1982]: Review of Field [1980], Journal of Philosophy 79, pp. 523534.
Parsons, C. [1980]: Mathematical Intuition, Proceedings of the Aristotelian Society 80,
pp. 145168.
Parsons, C. [2008]: Mathematical Thought and its Objects. Cambridge: Cambridge
University Press.
Priest, G. [2006]: In Contradiction. (2nd edition.) Oxford: Oxford University Press.
Putnam, H. [1971]: Philosophy of Logic. New York: Harper and Row. (Reprinted in
Putnam [1979], pp. 323357.)
Putnam, H. [1979]: Mathematics, Matter and Method. Philosophical Papers, volume 1.
(Second edition.) Cambridge: Cambridge University Press.
Quine, W.V.O. [1960]: Word and Object. Cambridge, Mass.: The MIT Press.
Robinson, A. [1967]: The Metaphysics of the Calculus, in Lakatos (ed.) [1967], pp.
2746.
Van Kerkhove, B., and Van Bendegem, J.P. (eds.) [2007]: Perspectives on Mathematical
Practices. Dordrecht: Springer.

112

Yvonne Raley

Jobless Objects:
Mathematical Posits in Crisis
Yvonne Raley
Abstract
This paper focuses on an argument against the existence of mathematical objects called
the Makes No Dierence Argument (MND). Roughly, MND claims that whether or
not mathematical objects exist makes no dierence, and that therefore, we have no reason
to believe in them. The paper analyzes four dierent versions of MND for their merits.
It concludes that the defender of the existence of mathematical objects (the mathematical
Platonist) does have some retorts to the rst three versions of MND, but that no adequate
reply is possible to the fourth, and most crucial, version of MND. That version argues that
mathematical objects make no dierence to our epistemic processes: they play no role in the
process of obtaining mathematical knowledge.

I. Introduction
As the debate over the existence of mathematical objects continues, a new
type of argument has surfaced as a challenge to mathematical Platonism. Alan
Baker (2003) calls it the Makes No Dierence argument (MND). Roughly,
the Makes No Dierence argument says that whether or not mathematical
objectsobjects that are said to be neither spatial nor temporal, and that
are causally inertexist makes no dierence, and that therefore, we have no
reason to believe in them. How is this lack of dierence-making cashed out?
Here, the literature oers various options. Horgan (1987: 281, 282) has it that
the worlds spatio-temporal causal nexus would be unaltered if sets did not
exist. For Ellis (1990: 113), the world we can know about would be the same
if there were no abstract objects. Azzouni (1994: 56) believes that if mathematical objects ceased to exist, [m]athematical work would go on
as usual. For Balaguer (1999: 113), if there were no such things as abstract
objects, science would be practiced exactly as it is right now. And lastly, Baker
(2003: 247) describes MND as saying that [i]f there were no mathematical objects, then this would make no dierence in the concrete, physical
world.1
1

Not everyone on this list endorses the Makes-No-Dierence argument, however.

Jobless Objects: Mathematical Posits in Crisis

Given this family of similar, but far from identical, formulations of MND,
tracing the argument to a point of origination is dicult. But theres another
reason for this. The Makes-No-Dierence argument is not actually one argument, and the quotations provided in the previous paragraph give us a rst
indication of this. For what it is that mathematical objects are supposed to
make a dierence to, depends very much on the purported job mathematical
objects are taken to have in the rst place.
As I see the matter, there are four distinct arguments that hide beneath the
umbrella of MND; and they need to be disentangled. That is what I propose
to do here. In separating out these four versions of MND, I am especially
interested in the replies available to the defender of the existence of mathematical objects the mathematical Platonist. The dialectic that will emerge is this:
Initially, the Platonists position will actually look quite promising, as he has
at least some responses to the rst three versions of MND (sections II, III,
and IV). However, as we investigate further how the claim that mathematical
objects make no dierence might be understood, we will see that the rst three
versions of MND dont really get at the heart of the debate. It is only the fourth
and last version of MND that succeeds in locating the focal point of the issue:
whether or not mathematical objects have an epistemic role (section V). In my
view, when it comes to the question of whether or not mathematical objects
play a role in the process of obtaining mathematical knowledge, the Platonist
loses his initial advantage. Such objects dont play a role. Consequently, whether
or not mathematical objects exist really makes no dierence. Thus the proponent of MND wins: without an epistemic role for mathematical objects, we
have no reason to believe in their existence.

II. MND and Causal Inertness


The rst version of MND is based on the assumption that mathematical objects are causally inert. This assumption is well-entrenched in the philosophy
of mathematics literature and it is sometimes called the standard platonistic
view. For example, Cheyne and Pigdon (1996: 639) take the standard platonistic position to include the claim that platonic objects lack spatio-temporal
location and causal powers.
The fact that mathematical objects are understood to be causally inert lends
itself to a quick and dirty version of MND: If the objects of mathematics are

114

Yvonne Raley

causally inert, then they cannot aect any objects or processes in the concretephysical world. Therefore, we have no reason to believe in their existence. So,
in this version of MND, to make a dierence, a mathematical object has to
be causally eacatious.
Strands of this argument are discussed in Horgan (1987: 281, 282): Since sets
are not supposed to be part of the worlds spacio-temporal causal nexus, that
nexus would be exactly as it is whether sets existed or not, and in Balaguer
(1998: 133):
Empirical science knows, so to speak, that mathematical objects are causally
inert. That is, it does not assign any causal role to any mathematical entity.
Thus, it seems that empirical science predicts that the behavior of the physical world is not dependent in any way upon the existence of mathematical
objects.2

How can a Platonist respond to this version of MND? In my view, the best
response (and probably the most common) is to simply concede the point.
If mathematical objects are causally inert, then of course they cant aect any
objects or processes in the physical world. (Thats why I called this version of
MND the quick and dirty version). But the Platonist can go on to say that
this is not the sort of role he had in mind for his objects in the rst place. After
all, it is a role he has dened away by his own assumptions.
The Platonist therefore has to take a dierent approach to delineating a role
for mathematical objects. To see what he might have to oer, we have to go
on a short tangent and explore a proposed counter-example to the view that
mathematical objects are causally inert. A detailed evaluation of this counterexample will reveal a possible role for mathematical objects that avoids the issue
of their nature. It will also lead us to a second version of MND.
The counter-example I have in mind is due to Cheyne and Pigdon (1996).
Cheyne and Pigdon, in their paper, focus on the question of whether the
Quine-Putnam indispensability thesis establishes standard Platonism about
mathematics. The indispensability thesis is cashed out as the claim that mathematical objects are just as indispensable to science as theoretical entities like
electrons. Electron theory quanties over numbers, just as it quanties over
electrons (1996: 640).
2

I dont know how Balaguer thinks hes established what empirical science knows, or even what
it knows, so to speak about the causal inertness of mathematical objects (who knows this,
how do they know it?). But, as I said above, the view is certainly presumed in the philosophical
literature, which is all we need for our purposes.

Jobless Objects: Mathematical Posits in Crisis

Cheyne and Pigdon then claim that even if we grant that mathematical
objects are indispensable to science (which they do grant), then standard Platonism is in trouble. Heres why:
If we are genuinely unable to leave those objects out of our best theory of
what the world is like (at least, that part of the world with which we causally
interact), then they must be responsible in some way for that worlds being
the way it is. In other words, their indispensability is explained by the fact that
they are causally aecting the world, however indirectly. (1996: 641)

As is well known, and as Cheyne and Pigdon correctly observe, the indispensability thesis is a claim about what sorts of objects our best scientic theories
need to quantify over. It then says that whatever objects we do quantify over are
objects that we have to take as real. In short, what it means for an object to be
indispensable to science is for that object to essentially fall within the range of
the (objectual) existential quantier once our best theories are written down.
But how does indispensability, which just involves quantication, bring
causal powers to the objects so quantied over? Cheyne and Pigdon oer support for this view with the following example:
The fact that there are three cigarette butts in the ashtray causes Sherlock to
deduce that Moriarty is the murderer, and that if there had been more or fewer
butts he would have deduced otherwise. (1996: 642)

Cheyne and Pigdon then say that the fact that there are three cigarette butts in
the ashtray is causal, and that the number three is an indispensable constituent
of this fact (1996: 642). 3 Furthermore, they say that the number three makes
a causal dierence because had there been a dierent number of cigarettes in
the ashtray, dierent eects could be expected (1996: 642).
Cheyne and Pigdon want to use this example to challenge to the standard
Platonist like this: Our challenge to Platonists is for them to provide an explanation for the indispensability of objects whose presence (they claim) makes no
causal dierence (1996: 642). But of course, that challenge has long been met,
because we already have an explanation for the indispensability of mathematical objects: mathematical objects are indispensable because we cannot rewrite
the language of science in such a way that avoids quantication over them! So
the Platonist need not explain anything. The indispensability thesis is a claim
3

In a footnote, Cheyne and Pigdon are quick to point out that [the number three] isnt, of
course (indispensable). They believe that more likely cases of indispensability will be found
in General Relativity or quantum mechanics.

116

Yvonne Raley

about what sorts of entities have to be quantied over in (the best version of )
scientic discourse. It is not a claim about what sorts of entities are causally efcacious. The indispensability thesis, per se, does not require the indispensable
objects to have any particular properties. The properties they have are dictated
by the scientic discourse. And, when it comes (say) to numbers, there isnt
anything in scientic discourse that requires them to be causal.
Theres another point to be raised here. As footnote 3 states, in the example
Cheyne and Pigdon oer, the number three is not actually indispensable the
number three need not be quantied over. Now compare the example Cheyne
and Pigdon oer with an example that involves a universal such as redness.
In many sentences, such universals are also not quantied over, e.g.: The
fact that Lucinda wore a red dress causes Sherlock to deduce that Moriarty
is the murderer (and if Lucinda had worn a dierent color dress, he would
have deduced otherwise). If, by analogy to Cheynes and Pigdons example, we
further suppose that this fact is causal, and that the redness of Lucindas dress
is a constituent of this fact, wouldnt we then have to say that the property of
redness exists even though it is not quantied over?4
If that is what we say, then the indispensability of an object (or a universal,
what have you) is not necessary for a commitment to the existence of that
object. That is, quantication over an object isnt necessary to be committed
to it. But if (instead) we say that since redness is not quantied over, we are
not committed to its existence, then (in any case) its being causally ecacious
isnt crucial to what we are (or arent) ontologically committed to. The upshot:
the indispensability of mathematical objects and the question of whether they
have causal powers (or not) are entirely independent from one another.

III. MND and Indispensability


Having thus separated questions about indispensability from questions about
the causal inertness of mathematical objects, we can now see our way to another
version of MND. This version relies on the indispensability thesis directly, and
avoids any questions about the nature of mathematical objects. This version
of MND surfaces in Baker (2003: 254): If mathematics is dispensable for science, then (No-Dierence) is true Baker understands the indispensability
4

This is precisely what Quine was trying to avoid when he used his criterion to argue against
McXs claim that there are universals. (Quine, 1948: 10.)

Jobless Objects: Mathematical Posits in Crisis

thesis in the same way Cheyne and Pigdon do, namely as the thesis that science cannot be formulated without quantifying over mathematical objects
(2003:258).5 So our indispensability version of MND might therefore read: if
scientic discourse can be formulated without quantifying over mathematical
objects, then these objects do not play a role in science. Their existence does
not make a dierence.
But can scientic discourse be formulated without quantifying over mathematical objects? The answer to that seems to be no. And Cheyne and Pigdon
(among many others) have explained why: the best eort (so far) to nominalize
mathematics began and ended with Newtonian physics. Famously, Field (1980)
tried to show that we need not quantify over numbers in Newtonian physics
(which is, however useful, not even true), and even if we grant that he was
successful, this leaves Quantum Mechanics and General Relativity untouched.
Therefore, our best scientic theories are stuck with quantication over numbers, and other mathematical abstracta.
On this count, therefore, the Platonist seems to win, for the antecedent to
the indispensability version of MND cannot be established. However, theres
an important objection to using the indispensability thesis to carve out a role
for mathematical objects: what is indispensable to science are quantications,
which involve mathematical terms. This means that it is those terms appearing
within the scope of the (objectual) existental quantier in sentences that are
indispensable. Hence, talk of indispensability, rst and foremost, is talk about
the indispensable role that mathematical terms play in scientic discourse. So,
strictly speaking, it is not purported objects that are indispensable, but a kind
of language. This is not to say that a connection between terms and objects
cannot be argued for, but it requires that the Platonist avail himself of an additional tool. The way the objects come in (and the above quotations indicate
this) is via Quines criterion for ontological commitment. As Quine (1948: 13)
has put it: To be assumed an entity is, purely and simply, to be reckoned as
a value of a variable.6
This is an important point, and it deserves further attention. But rst, we
will look at two other ways in which MND could be understood. As it turns
5
6

Bakers example, in the passage in question, comes from general relativity.


A bit further down in the same paragraph, Quine connects ontological commitment to the
quantiers explicitly: The variables of quantication something, nothing, everything
range over our whole ontology, whatever it may be; and we are convicted of a particular ontological presupposition if, and only if, the alleged presuppositum has to be reckoned among
the entities over which our variables range in order to render one of our armations true.

118

Yvonne Raley

out, once we are done examining the dierent strands of argument that hide
beneath the umbrella of MND, Quines criterion will surface twice more as
a tool which the Platonist must avail himself in order to etch out a role for
mathematical objects. We will therefore leave the work of Quines criterion in
this debate for last.

IV. MND and Mathematical Truth


Another role mathematical objects have often been assigned is that of grounding the truth of mathematical claims. In denying this role, MND might be
formulated like so: whether or not mathematical objects exist makes no difference to the truth of mathematical claims. In order to evaluate the strength
of this version of MND, let us look more carefully at what its proponent is
up against.
While the tradition that mathematical objects ground the truth of mathematical claims can be traced all the way back to Plato, consider a more recent
formulation of the idea, which is due to Benacerraf. According to Benacerraf
(1973: 405), just as the sentence There are at least three large cities older than
New York is true if and only if there really are three large cities that are older
than New York, the sentence There are at least three perfect numbers greater
than 17 is true if and only if there really are at least three perfect numbers
greater than 17. Thus, the truth conditions of the mathematical sentence seem
to necessitate the existence of mathematical objects that bear certain relations
to one another. On this view, in other words, the role mathematical objects
play is to provide the conditions under which mathematical sentences are
true.
Note that Benacerraf s own account, which Ive here summarized briey,
actually relies on Quines criterion for ontological commitment. Heres how
Benacerraf spells out the truth conditions for the rst sentence:
[It] will be true if and only if the thing named by the expression replacing a
(New York) bears the relation designated by the expression replacing R (1
is older than 2) to at least three elements (of the domain of discourse of the
quantiers) which satisfy the predicates replacing F and G (large and city
respectively). (1973: 405, emphasis mine.)

Benacerraf then suggests that the truth conditions of the mathematical sentence can be explicated analogously, since the two sentences have the same

Jobless Objects: Mathematical Posits in Crisis

grammatical structure.7 Therefore, the way in which the truth of mathematical sentences necessitates the existence of mathematical objects is via Quines
criterion. Mathematical objects, on this view, exist because the truth conditions
for mathematical statements require the domains of the quantiers in those
statements to range over mathematical objects.
As just noted, this way of delineating a role for mathematical objects comes
up against the same issue we encountered in the previous section: it is Quines
criterion for ontological commitment that connects the mathematical terms
used in mathematical sentences to mathematical objects.
But there is also another way in which the Platonist can try to connect mathematical terms and their objects. He can do so via a correspondence theory of
truth. For example, if one assumes that the sentences There are at least three
large cities older than New York and There are at least three perfect numbers
greater than 17 are true because of how the world is, then it would appear
that numbers (and other mathematical objects), just like cities, tables, chairs
etc., are part of furniture of the world. So if the Platonist avails himself of the
correspondence theory of truth, it seems that he can argue for a role of the
objects of mathematics in grounding mathematical truths without the help of
Quines criterion.
Making sense of this view, however, requires a cashing out of the notion of correspondence. Specically, what is needed is an account of truth
that explicates the correspondence relation between mathematical sentences and the mathematical objects which (supposedly) provide the conditions for their truth. Notoriously, explicating the correspondence relation
between mathematical objects and mathematical sentences is no easy feat.
Correspondence accounts of truth have been unpopular precisely because
the spelling out of the correspondence relation has been such an obstacle.
And abstract objects, such as the objects of mathematics, pose the largest
problem.
It is interesting that the truthmaker theory, the perhaps most sustained eort
of developing a correspondence theory of truth, attempts to circumvent, rather
than face that problem. Armstrong (2004), one of the strongest defenders of
the truthmaker theory, argues that numbers are not needed as truthmakers.
His reasons for this will prove instructive for our purposes.
7

We would be careless, however, to simply assume that Benacerraf endorses the view that mathematical objects ground the truths of mathematical sentences. As is well-known, Benacerraf
himself presents this way of understanding mathematical truth as a problem because it is
incompatible with the causal theory of knowledge.

120

Yvonne Raley

Armstrong (2004: 5) characterizes his truthmaker theory like this: [t]he idea
of a truthmaker for a particular truth, then, is just some existent, some portion
of reality, in virtue of which that truth is true. So the truthmaking relation
is not a causal relation, rather, sentences (or for Armstrong, propositions), are
true in virtue of what obtains in independent reality.
While it might be thought that the fact that the truthmaker relation is not a
causal relation for Armstrong at least opens the door for objects without causal
powers, he opts for an ontology without them.8 His reason:
To nd truthmakers for certain truths, or sorts of truths, one wants to postulate entities that stand in various more or less complex relations of correspondence to these truths. But one wants these entities to be such that we
can know, or at least have rational belief, that such entities exist. The entities
must be such that they are epistemically accessible. (2004: 37)9

Why does Armstrong insist on positing only those entities that we can know
about as truthmakers? As I see it, the reason for this is that if we dont, we are
stripped of any legitimate reason for supposing that any sentences are true in
the rst place. It is thus unavoidable to tie talk of truth to epistemology in
this way.10
While Armstrong (2004: 37) cashes out epistemic access mainly in terms of a
causal, or nomic link between the postulated truthmakers and the truths that
we can fairly claim to know or to rationally believe, this link can be understood
more widely, namely in terms of a reliabilist account of how we know about
objects: For S to know that p, the knowledge must have been produced by a
reliable process.11
Field (1989: 26) in couching Benacerraf s challenge (how we can know that
mathematical sentences are true if the only way to know about objects is by
8 See Armstrong, 2004: 112118, for how this is done. I am not here endorsing the truthmaker
theory.
9 This is also the point that Benacerraf is worried about. On a causal theory of knowledge, it
seems impossible that we could ever know the truths of mathematics.
10 Note that this does not make the truth of a sentence dependent upon knowledge of it.
11 Armstrongs text permits such an interpretation as well (see Armstrong, 2004: 37). Another
point: cashing out Armstrongs point in terms of a reliabilist account of how we come to
know objects also avoids the kind of reply the Platonist can make to the Eleaticicist: since
mathematical objects are causally inert, they cannot make a contribution to the causal/nomic
order of the world, but this does not mean that they might not be known in ways that dont
involve such a contribution. However, on a reliabilist account of knowledge, these objects
would not necessarily have to make a causal contribution in order for us to know about
them.

Jobless Objects: Mathematical Posits in Crisis

causal interaction with them) in reliabilist language, has put the problem rather
nicely:
But Benacerraf s challenge is to provide an account of the mechanisms
that explain how our beliefs about these remote entities can so well reect the
facts about them. The idea is that if it appears in principle impossible to explain
this, then that tends to undermine the belief in mathematical entities, despite
whatever reason we might have for believing in them.

To me, Fields observation also gets at the heart of where the role of mathematical objects might ultimately be sought and therefore, the way in which MND
might be most plausibly understood: as the view that the objects of mathematics make no dierence to how we come to know mathematical truths.
Before turning to this, we need to evaluate the success of the truth-version
of MND vis--vis the observations made here. Do mathematical objects play a
role in grounding the truth of mathematical claims? As we have seen, in order
to win this argument, the Platonist needs to establish a connection between
the mathematical terms used in mathematical sentences and their objects. He
can do this either by invoking Quines criterion (this will be discussed in section VI), or he has to provide an account of correspondence truth that explains
the relationship between mathematical terms and mathematical objects. As we
have seen, the most inuential account of correspondence truth the truthmaker theory, actually tries to circumvent having to do that, and for legitimate
reasons: one rst needs to explain how our beliefs about mathematical objects
accurately reects their properties. And for this, one in turn rst needs to show
that we can come to know about mathematical objects and their properties
in a reliable way. Let us explore this path, and the last version of MND that
it leads to.

V. MND and Mathematical Knowledge


We begin by providing a little more content to this epistemic version of MND.
In the opening line to his 1994 (3), Azzouni describes mathematical practice in
one succinct line: The mathematician proves truths.12 Later in the same work,
Azzouni oers the following thought experiment: Imagine that mathematical
12 Whether or not this is all a mathematician does will not concern us here.

122

Yvonne Raley

objects ceased to exist sometime in 1968. Mathematical work went on as usual.


Why wouldnt it?13
Why would mathematical practice go on as usual? According to Azzouni,
[g]iven standard mathematical practice, there seems to be no epistemic role
for mathematical objects.14 And on the next page: It is not merely that mathematical objects do not seem causally involved in the processes we use to learn
about their properties: it is that they seem to play no role at all.
In contrast, the objects of science clearly have an epistemic role. With some
of these objects, we can interact in fairly direct causal ways (we can touch
them, see them, etc.). Others the more theoretical objects like electrons
we interact with by designing instruments that are causally sensitive to their
machinations (1994: 53).15 Mathematical objects do not play such a role. We
dont interact with them, either directly, or indirectly via the use of instruments. Furthermore, I would add that we dont use mathematical objects to
justify the truth or falsity of mathematical statments that work is done by
mathematical proof: it is the proof that is referred to in the justication of
mathematical statements not the mathematical objects (or their properties).
Nor do mathematical objects or our interactions with them explain why we
believe that some mathematical statements are true and others false. We believe
in the truth of mathematical statements when (and because) a proof has been
provided.16
Here, then, is a way we can formulate an epistemic version of MND: mathematical objects play no role whatsoever in the process by which we come to
know mathematical truths. Therefore, we have no reason to believe that mathematical objects exist.
Does the Platonist have a reply to this? Well, at one point, mathematical
Platonists claimed that we grasped mathematical objects via a faculty of mathematical intuition, and it was this grasping of the mathematical objects and
their properties that provided justications for believing in the truth of certain
mathematical statements. This story may seem to best t basic mathematical
truths, such as that the number two is even or that circles are round. It might
be argued that no proof for such truths appears to be necessary, yet, we seem
to know them just the same. So, couldnt it be said that we know these truths
13 Azzouni, 1994: 56. Azzouni also discusses this thought experiment in his 2000. Baker (2003)
addresses it as well.
14 Azzouni, 1994: 55. Azzouni calls this the epistemic role puzzle.
15 Azzouni discusses this in more detail in his 1997.
16 I will say more about the role of proof in a moment.

Jobless Objects: Mathematical Posits in Crisis

because we intuitively grasp, or understand, the facts about these mathematical objects? If that were the case, this would delineate a role for mathematical
objects after all: these objects would be involved in the epistemic process via
the faculty of intuition.
A commonand quite reasonableobjection to the idea of resorting to a
faculty of mathematical intuition as a way of explaining mathematical knowledge is that it doesnt actually explain anything, but instead, pushes the problem one step further back. After all, what is this mysterious faculty supposed to
be? Thankfully, the current version of the position that we know mathematical
truths via intuition has abandoned the view that intuition is a special faculty.
Rather, intuition is understood as an ordinary cognitive process. For example,
following in the footsteps of Katz and Bonjour, McEvoy (2004: 433) sees mathematical intuition as
the fallible faculty of reason, representing the most basic elements of mathematical reality our ability to reason enables us to establish basic mathematical truths, and produces in us belief in these truths without any causal
or experiential connection between minds and abstract objects.

McEvoy argues that it is by accessing and examining mathematical concepts,


not mathematical objects, that we come to know the truths about the abstract
objects that these concepts represent. So, for instance, an analysis of the concepts of four and composite reveals that it could not be the case that four
was not composite (2004: 433). When it comes to such basic mathematical
truths, then, our capacity to reason does not have to be understood in terms
of providing an elaborate proof, but rather, can be understood in terms of a
simple conceptual analysis.
Mathematical concepts, for McEvoy, are concepts of abstract objects. These
concepts represent the mathematical objects and their properties.17 So a direct
causal link between us and abstract objects is not necessary for us to understand the truths about them. It might be objected that such a link is needed
to acquire the mathematical concepts in the rst place, but here McEvoy has
a good response. In his view, the Platonist can allow that our mathematical
concepts originate from our empirical interactions with physical objects that
approximate geometrical shapes, or that instantiate arithmetical structure.
[2004: 433, emphasis mine]18
17 As the reader will no doubt have noticed, a lot is packed into this seemingly benign notion
of representation. We will unpack it shortly.
18 A similar explanation is also oered by Resnik, 1997, chapter 9.

124

Yvonne Raley

Since physical shapes instantiate our concepts only approximately (e.g. there
are no perfect circles), and since our arithmetical concepts outstrip the number
of, and relations between, physical objects (e.g. not all numbers are instantiated), we realize that our mathematical concepts must be concepts of abstract
objects (2004: 433). So as we become aware that circles are not perfectly round,
we see that the mathematical concepts we have cannot really be concepts of
objects in physical space. Therefore, we reason these concepts must represent
the properties of abstract objects.
In this way, McEvoy believes he has oered an a priori and reliabilist account
of how we gain mathematical knowledge of abstract objects an account that
does not require these objects to causally interact with us; and that does not
presume any other experiential connection to the object (whatever that may
come to). The account is a priori in the sense that mathematical knowledge is
produced by a process of reasoning about concepts (and not about empirical
objects). It is reliabilist because the reasoning process by which this knowledge
is produced itself is reliable.19 Lastly, as McEvoy says, his account of mathematical knowledge is also fallibilist. This is because mathematical knowledge can be
revised based on a priori considerations.
The question now arises whether or not this new account of intuition succeeds in providing an epistemic role for mathematical objects. My view is that
it does not. In order to see this, let us go through the argument McEvoy has
presented step by step. The rst step is that of concept acquisition. We learn
about mathematical concepts by learning about empirical objects and their
concepts. So far, so good. But no mathematical objects are involved in this
process, by McEvoys own description. The second step is that of reasoning
about these concepts, or analyzing them, and coming to understand truths
about them. This, too, does not require any mathematical objects.
Let me add one other ingredient to the process of gaining mathematical
knowledge that I think is unobjectionable. In order to come to know more
complex mathematical truths, we (or rather, the mathematicians) provide
proofs. (I will here understand proof in a fairly loose sense, not requiring
that each assumption and every logical step of the proof are written down. A
proof, in this sense, is the sort of thing one nds in a standard mathematics
textbook.)
Still, so far, no part of the process described involves mathematical objects.
In fact, that may be the very reason why this process of acquiring mathematical
19 McEvoy, 2004: 428.

Jobless Objects: Mathematical Posits in Crisis

knowledge, as detailed by McEvoy, is reliable. It is reliable because it involves


no access to, and no interaction with, abstract objects. And this is so even if we
dont understand access or interaction in a causal way.
One might think that theres a way that mathematical objects enter the
picture yet, albeit in a much more indirect way. McEvoy states, correctly, that mathematical concepts are not really instantiated in the physical
world. The geometrical objects we encounter in physical space do not have
exactly the properties that our mathematical formulas attribute to them,
and the vast majority of numbers are not instantiated by collections of objects. Therefore, McEvoy reasons, these concepts must be representations of
abstracta.
Lets play along with this for a moment. Lets assume, for the sake of the
argument, that mathematical objects do exist, and that they indeed have the
properties that we attribute to them. Let us also assume that we dont actually know this, because as McEvoy himself describes the process of gaining
mathematical knowledge, these objects arent involved: they are not interacted
with in any way.
Heres how I want to press my objection: if all this is true, then what reliable
way do we have of knowing that the concepts we have acquired from interaction with (mere) physical objects correctly represent the objects of mathematics
and their properties? In other words, why doesnt Fields challenge reappear at
this point? How do we explain that what we believe is true of mathematical
objects (by abstraction from and extension of what we have learned by interaction with physical objects) really is true of mathematical objects?
We can certainly grant that physical circles, say, the circle that I carefully
draw on the board with the help of a ruler, are (approximately) round. But
how does it follow from this that the abstract object we take ourselves to refer
to with the name of circle is round? How do we even know that among
the abstract objects there is an object that represents the properties we have
derived from studying (and idealizing) the properties of physical objects? This
may sound like a very odd question to ask. But the only reason why it sounds
odd is because thats what we mean by circle. The word circle is what
we use to name round objects. Our having settled on that name, however,
doesnt force the existence of an abstract object that ts the description we have
provided.
Let us pursue this objection by working with an example. The example will
look intuitively dis-analogous, but further examination will prove this to be a
chimera. This strategy will, I hope, reveal how strange the Platonists picture

126

Yvonne Raley

really is. So, here goes. Imagine a philosophy student, call him Kitz. Kitz believes that there are abstract objects called quirks, and that these quirks have
certain properties, a, b, and c. Among these properties, Kitz believes, is the
property of being spiy. Kitz also believes, however, that we cannot interact
with abstract objects in any way. We can interact only with physical objects
and their properties. By abstracting away from the imperfections about these
objects and their properties as we encounter them in the physical world, we
realize, Kitz thinks, that there must be abstract objects which represent these
properties perfectly.
Kitz now sets out to write a dissertation about quirks and their spiyness.
But his advisor, while a Platonist, has some doubts that Kitz will actually pass
his defense. So he challenges Kitz to explain how he knows that quirks, that
is, perfectly spiy objects, are really among the realm of abstract objects. Kitz
spells his reasoning out like this: when he was a child, Kitz had received a very
nice pair of red boots for Christmas, and he thought to himself, now these are
really spiy. His new red boots were, of course, empirical objects, and he had
causally interacted with them. Kitz also possessed the concept of spiy. Spiy,
he had learned, was the property of being exceptionally cool, and this property
most certainly applied to his new red boots.
Later on, having sat in metaphysics classes and learned about abstract objects,
Kitz came to believe that abstract objects are real, and that they instantiate
properties he has acquired in the real world. So, he thought to himself, there
must certainly be an abstract object that instantiates the property of spiyness
and, for lack of a better name, he named the object quirk. (He would have
liked to name the object circle but that name was already taken.)
Clearly, Kitzs reasoning is just an instance of bootstrapping. We can grant
every single one of Kitz claims: how he acquired the concept of spiy, that he
used the concept correctly, that his red boots were spiy. In other words, we
can grant that everything Kitz says about the physical world is true. We can also
grant (just to be nice) that there are abstract objects. We can even allow Kitz
to pick a hitherto unutilized sound, spelled out as quirk, to name an object.
Still, it just does not follow that there is an abstract object named by quirk
which has the property of being spiy.
Fast forward to Kitzs dissertation defense. (By sheer obstinacy, Kitz has gone
on to complete his dissertation about quirks, and his advisor, tired of opposing
Kitz quest, has decided to just let him have it at the defense.) Here are the
objections that are raised by his examination committee (all of whom, lets say,
are mathematical Platonists).

Jobless Objects: Mathematical Posits in Crisis

Examiner 1: The case for mathematical objects is much stronger than your
case for quirks. Take circles: we realize that physical circles arent perfectly round, that is, that they only approximate the mathematical
properties we attribute to circles. So, we reason that there must be an
(abstract) object which does have the properties. This reasoning does
not apply to quirks.
Kitz: The way I see it, no object in the real world is exceptionally cool to
the perfect degree. Only an abstract object could possess this quality
perfectly. So the same reasoning does apply.
Examiner 2: Spiy is vaguely dened. Mathematical properties are not
vague. So mathematical terms pick out abstract objects, but spiy
does not succeed in picking out anything.
Kitz: What is the argument for saying that vague terms cannot pick out
abstract objects? Consider the pre-theoretical notion of circle as
the Babylonians might have had it. They did not possess the formula
we have for working out the area of a circle, so we might argue that
they did not have a complete understanding of the exact properties
of circles. Therefore, their concept of circle was vaguely dened. But
I should think when the Babylonians talked about circles they
referred to the same abstract object we refer to when we talk about
circles.
Examiner 3: Mathematical objects have their properties necessarily. They
cannot be otherwise. Quirks are not necessarily spiy. They could be
otherwise.
Kitz: No, quirks are necessarily spiy. I cant imagine them being any other
way.
Examiner 3: We are not talking about what we can or cannot imagine. We
are talking about logical necessity.
Kitz: Im not talking about what we can imagine any more than you are.
Examiner 3: But you are simply stipulating that quirks are necessarily
spiy.
Kitz: Arent you just stipulating as well? Consider basic mathematical truths
postulates, if you like. These are the sorts of things that are assumed
in a proof, not the theorems they result in. Take one of Euclids postulates, the one that says that the shortest distance between two points is
a straight line. Thats what we mean by the concept of a straight line.
Its the shortest distance between two points. Thats why it cant be
otherwise. We have stipulated that thats what we mean by the words.

128

Yvonne Raley

Similarly, I stipulate that quirks are the sorts of things that are perfectly
spiy. So, it cant be otherwise. Digging in his heels, he adds: prove
me wrong!
Examiner 3: Stipulation does not bring into existence an object to which the
term refers. By stipulation, you certainly dont bring into existence an
object that is mind-independent, which we assume abstract objects to
be.20 Consider what Benacerraf has to say about stipulation: stipulation makes no connection between the propositions and their subject
matter [the abstract object] stipulation does not provide for truth.21
So how do you know that what you have stipulated to be true of the
object really is true of the object?
Kitz: How do you know this with regard to the objects of mathematics?
Really, I just dont see any dierence between the case for mathematical
objects and my object. So if you grant that mathematical objects exist,
then you must grant the existence of quirks as well.
Well, I dont see a dierence between the two cases either, and that was the
point of the exercise, of course. However, I disagree with Kitz that he has made
a case for quirks. Rather, I think matters are exactly the other way around.
We cannot grant that a case has been made that anything Kitz says or believes
about quirks is true of any abstract object. Such an object might exist, and it
might not, but this makes no dierence to Kitzs statements or beliefs. And
the same is true for mathematical objects. The cases are exactly analogous,
and so the epistemic version of MND stands up. Whether or not mathematical objects exist makes no dierence to the process of how we come to know
mathematical truths.

VI. The Role of Quines Criterion


So, does this mean there is absolutely no role for mathematical objects? Before we can come to this conclusion, there is a loose end to consider. Above,
we noted that according to McEvoy, mathematical concepts represent abstract
objects. They have to represent abstract objects, he says, because they couldnt
represent physical ones.
20 For a more detailed discussion of this claim, see Azzounis 2000.
21 Benacerraf 1973: 419.

Jobless Objects: Mathematical Posits in Crisis

We can certainly grant that mathematical concepts dont represent physical


objects, but why does it follow from this that they must represent abstract
objects? The answer, again, lies in the application of Quines criterion: the
mathematical terms we use fall within the range of the (objectual) existential
quantier, and therefore, we seem to be committed to their existence. The real
work of tying mathematical terms to mathematical objects, it appears, is done
by Quines criterion for ontological commitment. However indirect this role
may be, unless the existence conferring role of Quines criterion can be challenged, we will be saddled with the conclusion that mathematical objects do
exist (even if their role is purely semantic).
But the proponent of MND is not out of options, for Quines doctrine has,
in fact, been challenged. For example, Yablo (2000: 304) has argued that the
quantiers must be understood metaphorically, not literally. Although this
metaphorical use cannot always be paraphrased away, theres no reason to
think that a metaphorical use of the quantiers is ontologically committing.
More recently, Azzouni (2004: 54) has questioned the claim that the objectual
existential quantier must be understood as ontologically committing. The
objectual existential quantier in the object-language, he says, is read as ontologically committing by virtue of a domain of objects they range over (as weve
seen). But that the objects in that domain are objects we should be committed to
is determined by a reading of the existential quantiers in the meta-language
that provide the semantics for the object-language quantiers. Nothing (in that
meta-language or elsewhere) forces such a reading on us. If this is correct, then
it is not true that the (objectual) existential quantier must be understood as
ontologically committing.
Whether or not we are committed to the entities that we quantify over is not
going to be easily settled. What is needed is a careful evaluation of the arguments for and against Quines criterion.22 What is further needed is an evaluation of how we express our ontological commitments in English. Take, again,
Benacerraf s mathematical sentence: There are at least three perfect numbers
greater than 17. When we utter that sentence in ordinary English, are we really
expressing a commitment to the existence of three perfect numbers? Or are we
simply committing ourselves to a truth (a truth we can justify by providing a
mathematical proof )?
Rather than try to provide a quick, and therefore glib, answer to these questions in a few closing paragraphs, I want to end this paper with a challenge
22 See, also, Raley 2007.

130

Yvonne Raley

of my own: Would it really be such a bad thing if mathematical terms didnt


refer? What, indeed is the reason for thinking that mathematical terms must
represent objects?
This challenge isnt merely a verbal one. As I have tried to show in this paper,
mathematical objects do not make a dierence. And why should we think that
our mathematical terms must represent objects when there is absolutely no role
that these objects could play other than the sheer role of being the referents of
our terms? If these objects did have a role, then yes, I think we would have to
conclude that our mathematical terms refer (and thats indeed, the direction in
which the argument should go). But we are not at all forced to this conclusion
if mathematical objects do not make a dierence.

Bibliography
Armstrong, David M. 2004. Truth and Truthmakers. Cambridge, MA: Cambridge
University Press.
Azzouni, Jody. 1994. Metaphysical Myths, Mathematical Practice. Cambridge, MA:
Cambridge University Press.
1997. Thick Epistemic Access. The Journal of Philosophy XCIV, 9: 472484.
1998. On On What There is, Pacic Philosophical Quarterly 79: 118.
2000. Stipulation, Logic and Ontological Independence. Philosophica Mathematica, Vol. 8: 225243.
2004. Deating Existential Consequence: A Case for Nominalism. New York, NY:
Oxford University Press.
Baker, Alan. 2003. Does the Existence of Mathematical Objects Make a Dierence,
Australasian Journal of Philosophy 81, No. 2: 246264.
Balaguer, Mark. 1998. Platonism and Anti-Platonism in Mathematics, New York, NY:
Oxford University Press.
1999. Review of Michael Resniks Mathematics as a Science of Patterns. Philosophica
Mathematica, Vol. 7: 108126.
Benacerraf, Paul. 1973. Mathematical Truth. In Philosophy of Mathematics: Selected Readings, Second Edition, edited by Paul Benacerraf and Hilary Putnam. Cambridge,
MA: Cambridge University Press, 1983: 403420.
Cheyne, Colin and Charles Pigden. 1996. Pythagorean Powers, or A Challenge to
Platonism, Australasian Journal of Philosophy 76: 63945.
Ellis, Brian. 1990. Truth and Objectivity. Oxford, England: Basil Blackwell.
Field, Hartry. 1980. Science Without Numbers. Princeton, NJ: Princeton University
Press, Princeton.
1989. Realism, Mathematics & Modality. Cambridge, MA: Blackwell Publishing.

Jobless Objects: Mathematical Posits in Crisis

Horgan, Terence. 1987. Discussion: Science Nominalized Properly. Philosophy of Science 54: 28 182.
McEvoy, Mark. 2004. Is Reliabilism Compatible with Mathematical Knowledge?
Philosophical Forum XXXY, No. 4: 423437.
Quine, W.V., 1948, On What There is, in Quine, From A Logical Point of View, Harvard University Press, Cambridge, MA, 1980: 120.
Raley, Yvonne. 2007. Ontology, Commitment, and Quines Criterion. Philosophica
Mathematica 15: 27 1290.
Resnik, Michael. 1997. Mathematics as a Science of Patterns. New York, NY: Oxford
University Press.
Yablo, Stephen. 2000, A Paradox of Existence, in Everett, A. and Hofweber, T. (eds.),
Empty Names, Fiction and the Puzzles of Non-Existence. CSLI Publications, Stanford,
CA: 275312.

132

Susan Vineberg

Is Indispensability Still a Problem for


Fictionalism?
Susan Vineberg

Abstract
For quite some time the indispensability arguments of Quine and Putnam were considered
a formidable obstacle to anyone who would reject the existence of mathematical objects.1
Various attempts to respond to the indispensability arguments were developed, most notably
by Chihara and Field.2 Field tried to defend mathematical ctionalism, according to which
the existential assertions of mathematics are false, by showing that the mathematics used in
applications is in fact dispensable. Chihara suggested, on the other hand, that mathematics makes true existential assertions, but that these can be interpreted so as to remove the
commitment to abstract objects. More recently, there have been various attempts to show
that the indispensability arguments contain assumptions that are conceptually misguided
in ways having little to do with mathematical content.3 All of this work is of considerable
interest, and the result has been a gathering consensus that the indispensability arguments, as
put forth by Quine and Putnam, do not provide convincing reason to accept mathematical
realism. The focus here will be on the ways of responding to the indispensability arguments,
and in particular on the obstacles to ctionalism that remain after the versions of Quine
and Putnam are undercut.

The Quine-Putnam Indispensability Argument for


Mathematical Realism
It is common to nd references to the Quine-Putnam indispensability argument in contemporary discussions of mathematical realism. Although their
arguments are often run together, it will be useful in evaluating some recent
1

2
3

See, Putnam, H. (1979). Philosophy of Logic. Mathematics, Matter and Method: Philosophical Papers vol. 1. Cambridge, Cambridge University Press: 323357. Quine, W. V. O. (1969).
Existence and Quantication. Ontological Relativity and Other Essays. New York, Columbia
University Press: 9 1113.
See, Chihara, C. (1990). Constructibility and Mathematical Existence. Oxford, Oxford University Press. Field, H. H. (1980). Science Without Numbers. Oxford, Basil Blackwell.
See, Maddy, P. (1997). Naturalism in Mathematics. New York, Oxford University Press., Sober,
E. (1993). Mathematics and Indispensability. The Philosophical Review 102(1): 3557.

Is Indispensability Still a Problem for Fictionalism?

responses to see that they are distinct. Each argument will be presented briey,
followed by a discussion of some responses and the extent to which they address each argument.

Quines Argument
Quines indispensability argument stresses the idea that we have indirect empirical evidence for the mathematical claims used in science. According to
Quine, we have good grounds for believing that our best scientic theories are
true. In his view, the reasons for accepting the existence of molecules, atoms,
and quarks are analogous to the reasons for accepting the common sense belief
that there are physical objects, namely that the best theories of our experience
postulate such entities. The fact that a theory organizes our experience is what
constitutes evidence for the objects of that theory. Thus, to have reason to believe that a theory is true just is for it to be the best theory we have that organizes
and explains our experience as a whole (Quine 1976). Since a considerable body
of mathematics is indispensable to our scientic theories, it receives indirect
empirical support, in accordance with the hypothetico-deductive method, for
being part of our best overall account of experience. Quine takes it that the variables in our mathematical theories range over abstract entities and so we must
accept the existence of abstract entities in accepting the truth of our scientic
theories that make use of mathematics (Quine 1969). Quines indispensability
argument can be summed up as follows:
(1a) We have (empirical) evidence for all of the entities quantied over in
our best theories.
(2a) Our best theories quantify over mathematical entities, indeed mathematics is indispensable to our best theories.
(3a) So, we have reason to believe in mathematical entities.
For Quine, the evidence for mathematical objects comes from the fact that they
gure in our best overall theory that accounts for the observational data. Maddy
(Maddy 1992) has argued, using a number of examples, that the conrmational
holism, upon which this depends, is at odds with scientic practice, and hence
with Quines naturalism, according to which scientic standards provide the
grounds for theory acceptance. If we accept naturalism, as Maddy understands
it, whereby the record of scientic practice seems to be the last word as to

134

Susan Vineberg

what constitutes appropriate scientic standards, then the holism upon which
Quines indispensability argument depends cannot be maintained.
One might respond by arguing that some form of naturalism is compatible
with conrmational holism. However, the hypothetico-deductive account of
conrmation that lies at the heart of Quines treatment of evidence is also seriously awed, for reasons independent of holism or naturalism, and has been
dismissed by philosophers of science as thoroughly inadequate. Moreover, it is
doubtful that any plausible analysis of conrmation will itself support holism
or the claim that mathematical truths should be taken to be conrmed by their
use in science (Vineberg 1998). Indeed, as Sober argues, on the contrastive
view of conrmation, which he endorses, the indispensability of mathematics
is actually inconsistent with the idea that it receives conrmation through its
role in well-conrmed physical theory (Sober 1993).4 While there remains the
possibility of defending the view that mathematics receives empirical support
through its use in science, this position is simply not justied by contemporary
accounts of evidential support.

Putnams Argument
Putnam (Putnam 1979a; Putnam 1979b) does not deny that the mathematics
that gures in scientic theories receives empirical support, but instead emphasizes that our physical theories require mathematical equations for their formulation. He observes that Newtons law of gravitation, which states that the
gravitational force acting on two bodies is directly proportional to the product
of their masses and inversely proportional to the square of the distance between
them, has a mathematical structure, and claims that it cannot be stated in a
nominalistic language, that is, one in which no reference is made to numbers,
functions, sets, etc. If this is right, our scientic theories refer to mathematical
4

On Sobers view, conrming mathematical truths through their use in scientic theories
would require contrasting those theories with ones that do not contain mathematics. If
mathematics is truly indispensable to our empirical theories then it cannot receive empirical
conrmation. Only Putnams version of the argument assumes that mathematics is strictly
indispensable; Quine at most assumes that mathematics is indispensable to our best physical
theories. In any case, it is clear that Quine would reject the contrastive account of evidence
according to which empirical data for a hypothesis must discriminate between it and some
alternative, since he maintains that we have empirical reason to accept the theory that there
are physical objects over the idea that there are simply collections of sense data, even though
our experience does not discriminate between the views.

Is Indispensability Still a Problem for Fictionalism?

entities, and so accepting scientic theories as true appears to require accepting


the existence of mathematical entities. Furthermore, Putnam argued that we
do indeed have reason to believe that our best scientic theories are true, for
otherwise their demonstrated success would be a miracle (Putnam, 1979b). His
argument may be put as follows:
(1b) We have reason to accept our best scientic theories as true.
(2b) Our best scientic theories cannot be formulated without mathematics.
(3b) If mathematically formulated theories are to be taken as true, then
we must accept mathematics as true.
(3b) Accepting mathematics as true requires accepting mathematical entities.
(4b) So, we must accept mathematical entities.
This argument for mathematical realism clearly survives the challenges to conrmational holism. Even if theories do not receive conrmation as a whole,
as long as there are theoretical claims that we regard as receiving empirical
support, which cannot be stated without mathematics, it would appear that
we must accept mathematical realism. For Putnam, the empirical grounds for
accepting our best scientic theories carry over to provide indirect empirical
support for mathematical objects. Suppose however that we reject altogether
the idea that mathematics is conrmed through its use in science. There is
an indispensability argument, closely related to Putnams, which does not require that the mathematics employed in stating physical theories is empirically
conrmed. Insofar as we are committed to maintaining that at least some
mathematically formulated claims of physical theory are true, it would seem
that we are committed to mathematical objects regardless of whether we take
mathematics as conrmed through such use. That is, the use of mathematics
in formulating theoretical claims that we regard as true appears to presuppose
the existence of mathematical objects.
Understood along these lines, the indispensability argument more closely
resembles Freges argument that platonism is necessary in order to make sense
of mathematical applications.5 Frege argued that the legitimacy of inferences
that use mathematics, including simple arithmetical inferences, depends upon
the existence of mathematical objects. Consider
5

Frege, G. (1978). The Foundations of Arithmatic. Oxford, Blackwell.

136

Susan Vineberg

There are 2 apples in the refrigerator.


There are 3 oranges in the refrigerator.
There are no other fruits in the refrigerator.
2+3 =5

There are 5 pieces of fruit in the refrigerator.


Frege argued that our justication for drawing the conclusion above requires
an analysis in which the numeral 2 refers to an abstract object. The key point
here is that that applying mathematics seems to involve drawing inferences
from mathematical premises. If those premises arent true, then the inferences
do not appear justied.

Antirealist Responses to the Indispensability Arguments


A number of antirealist responses that were directed towards the indispensability arguments of Quine and Putnam have force against the broad spectrum
of indispensability arguments. For example, Chihara maintains that inferences
involving mathematics can be formulated by appealing to modal notions in
such a way that the premises are true, but do not involve quantication over
mathematical objects, and that mathematical formulations of physical theory
do not presuppose mathematical objects. This response does not appeal to ideas
about conrmation or scientic practice, and works equally against all versions
of the indispensability argument considered here. Azzouni also argues that the
truth of mathematics does not require taking a realist stance towards mathematical objects, by attempting to undercut Quines account of ontological
commitment, namely that we are committed to those objects that are quantied over in statements that we take as true. (Azzouni 2004).6 Both Azzouni
and Chihara assume that the use of mathematics in science requires taking
mathematical statements as true, although what such truth amounts to diers
from the platonistic view. As such, their views depart from pure or traditional
ctionalism, of the sort proposed by Field, according to which mathematics
is to be regarded as strictly false. While one might classify all of these views as
being forms of ctionalism, as Balaguer does, an important dierence concerns
6

This is perhaps the most direct response, and, if cogent, would undermine indispensability
arguments across the board, and not just those used to justify mathematical realism.

Is Indispensability Still a Problem for Fictionalism?

the ways of responding to the indispensability arguments that the views permit.
Those who, with Field, maintain full blown ctionalism must explain the use
of mathematics in drawing reliable inferences in science, whereas those who
maintain the truth of mathematics do not face this additional hurdle. Defending such a view requires an account of how the application of mathematics
works. Although this additional diculty makes pure ctionalism less attractive in some ways than the approaches of Chihara and Azzouni, their views face
questions as well. But in any case, it is worthwhile to explore the ctionalists
prospects for accounting for the use of mathematics in science.
Field addresses the arguments given by Quine and Putnam, but his approach
yields a response to the other version of the indispensability argument discussed above as well. His strategy for defending ctionalism involves establishing that scientic theories formulated using mathematics can be reformulated
in nominalistic terms (Field 1980). It is the nominalized theories that on Fields
view may be said to be strictly true. The other key plank of his program involves showing that the use of mathematics in science can be viewed as a mere
shortcut to logical derivations within nominalized science. Insofar as there are
nominalized versions of scientic theories, scientic inferences can be taken
as deductions from the true premises of nominalized science, though standard
mathematical theory will typically be used as a shortcut.
There have been numerous criticisms of Fields program one of which is
of signicance here, namely that there is considerable doubt that appropriate nominalized versions of science are available, such that the application of
mathematics can be explained in the way that Field envisioned.7 There is some
plausibility to the idea that the mathematical structures in which scientic
theories are cast could be reconstructed out of physical objects, however it is
considerably less clear that such structures can be assumed to exist as such in
the physical world. Even in the case of the (relatively) simple theory of Newtonian Gravitation, for which Field did construct a nominalized version, there is
considerable doubt that the constituents of the relevant physical structures, in
this case space-time points, are indeed physical objects. It is even less clear that
quantum mechanics, or even simple economic theories, can be reinterpreted
in terms of existing physical structures.8
7
8

See Chihara (1990).


Balaguer (Balaguer, M. (1998). Platonism and Anti-Platonism in Mathematics. New York,
Oxford University Press) suggests that quantum mechanics can be nominalized, despite criticism of Fields program claiming this cannot be accomplished. Balaguers approach involves
constructing objects that are isomorphic to Hilbert spaces out of what he claims are nominal-

138

Susan Vineberg

It is true that many uses of mathematics are dispensable, and this fact does cut
against Putnams version of the indispensability argument. Recall that one of
the premises of the argument is that our best theories are to be regarded as true,
where this is to be understood as requiring belief in all of the entities quantied
over in stating the theory. However, the fact that there are alternative formulations of some physical theories that do not quantify over mathematical objects,
provides evidence against Putnams claim that our best scientic theories simply cannot be formulated without mathematics. It further undercuts the idea
that the success of our theories is a reason to regard them as strictly or literally
true in all respects. Even Fields partial success in showing that mathematics
is dispensable undermines Putnams original argument. But, such partial success is insucient to show that the use of mathematics in science does not in
some way presuppose its truth, since there remain applications that have not
be nominalized. For Fields strategy to succeed in showing that ctionalism is
viable, he must show that mathematics is entirely dispensable in science. Even
if this were correct, and that seems doubtful, it could at best be established on
a temporary basis, for each genuinely new use of mathematics would require
an argument as to its dispensability. This suggests the need for a dierent way
of defending ctionalism that would work for as yet unformulated theories.

Balaguers Defense of Fictionalism


Recognizing that there remains some question about whether mathematics
is genuinely indispensable, Balaguer has tried to give a ctionalist account of
mathematical applications on the assumption that mathematics is not entirely
dispensable (Balaguer 1998). Assuming that mathematics is indispensable, the
thought is that a complete description of the physical world will contain statements that quantify over mathematical objects. In particular, there will be
mixed statements that refer both to physical states and mathematical objects.
As an example Balaguer gives
(A) The physical system S is forty degrees Celsius.
istic entities, but it is unclear that the structures so constructed are physically real. Balaguer
only provides a sketch of how this might be done, and it can be objected that he appeals to
structures, such as propensities, which are not genuinely nominalistic.

Is Indispensability Still a Problem for Fictionalism?

In this case, it may seem that we can nd a nominalistic reformulation, but


that doesnt matter; the sentence above is supposed to be a stand in for whatever
indispensable applications there are. Since any mathematical objects are causally independent of physical objects, Balaguer argues that while (A) expresses
a mixed fact about numbers and physical objects, it cannot be a bottom-level
fact. It must be that the facts expressed by sentences such as (A) supervene on
more basic facts that are not mixed; there must be underlying physical facts,
which are independent of the mathematical facts, that stand behind the truth
of sentences such as (A). On Balaguers view, mathematics can be used to represent physical facts. However, since there must be underlying physical facts
that stand behind any mixed fact expressed by such a representation, which
are independent of any mathematical facts, we are free to disbelieve any such
mathematical facts and embrace only the underlying physical facts.
For Balaguer mixed statements such as (A) have a platonistic content (P)
and a nominalistic content (N), corresponding to the underlying platonistic
and nominalistic facts that would have to obtain for (A) to be true. According
to the argument above, we could accept just the nominalistic content (N).
Assuming that the mathematics in statements like (A) is indispensable, it will
be impossible to nd a sentence that expresses just the nominalistic content
of (A). The position is awkward at best, for we are left having to say that it
is the physical content of statements such as (A) that are conrmed through
observation, but that we cannot say what exactly has been so conrmed. This
is especially problematic for someone such as Balaguer who embraces scientic
realism. Given that Balaguer must say that theoretical statements that make
indispensable use of mathematics are strictly false, he certainly cannot maintain the most stringent form of scientic realism, namely that the statements
of contemporary theory are strictly and literally true. This by itself is not very
troubling, since such stringent realism is highly implausible. The problem for
Balaguer though is that while he might maintain that theoretical entities play
a causal explanatory role in accounting for observation, in accordance with
scientic realism, he cannot characterize the properties of those entities because
in many cases that seems to make indispensable use of mathematics. What we
are left with is a vague and anemic form of realism.
Balaguers main discussion centers on the descriptive role of mathematics
in characterizing physical laws. However, he claims that the inferential role of
mathematics is covered easily by his account. He acknowledges,
I do need it to be the case that if we have a (sound) argument for C that takes
P1Pn as premises and that is formulated in platonistic terms, so that at least

140

Susan Vineberg
one member of {P1Pn } and perhaps also C-refers to, or quanties over,
mathematical objects, then whenever the nominalistic content of {P1Pn} is
true, the nominalistic content of C is true. (Balaguer 1998, pg. 2023)

Balaguer claims that this condition obtains because the platonistic and nominalistic contents concern dierent realms. While it seems plausible that physical facts should not depend on non-physical facts, there is no convincing argument here that there cannot be bottom-level mixed facts. Furthermore, this
defense of the legitimacy of drawing nominalistic conclusions from platonistic
premises leaves much to be desired. To be sure, Frege would not have regarded
the vague contents on which such reasoning is supposed to rest as belonging
to a legitimate account of inference. It is really quite mysterious as to what
these nominalistic contents are, and how we are to understand, in a satisfactory way, what it is for our nominalistic conclusions to follow from them.
Indeed, it is far from clear that such contents are in any way nominalistically
acceptable.
It is also perplexing as to how, on this view, inferences involving mathematics yield correct predictions. The conclusions inferred are taken to be false, and
the true physical claims, with which they are associated, cannot be stated. The
usual sort of account of how theories are conrmed becomes thoroughly problematic. In conrming a theory T, we look to its consequences and compare
them with empirical data. Observational data need not always be taken as true;
we may work with corrected data, or upon occasion disregard it entirely. But,
in deciding whether or not to retain a theory T on the basis of empirical data,
we accept the data. However, the data that we use to assess the theory would
typically be given a mathematical representation, i.e. it might look something
like (A), and hence on Balaguers account would have to be taken as false.
What is apparently needed for conrming theories are true descriptions of
the empirical facts, but assuming mathematics is indispensable some of these
needed descriptions will simply be unavailable.

Towards a Fictionalist Account of Mathematics


Can a ctionalist who accepts the indispensability of mathematics in formulating scientic theory accept scientic realism without appealing to something
such as Balaguers mysterious physical contents? A line of defense for the sort
of ctionalism that Balaguer wants to maintain seems possible that permits a

Is Indispensability Still a Problem for Fictionalism?

more robust form of scientic realism, but also captures an important feature of
mathematical application and the contemporary treatment of scientic theories. An outline of the approach is sketched below.
Many philosophers of science, realists and antirealists alike, have urged that
scientic theories should be thought of in terms of models. Note that the very
idea of theories as models appears congenial to the ctionalist stance. Models
are not intended to be exact replicas of the phenomena modeled, but rather
involve varying degrees of idealization. In practice, if not of necessity, some
features of the phenomena are singled out for attention. Those features of the
model that are not directly concerned with the features of interest need not
resemble aspects of the phenomena so modeled. Further, where it is the relationships between objects that are targeted, there seems little diculty with
regarding the objects of the model as ctional. There is then some plausibility
to the idea that mathematical objects could be used to model physical objects
and relations without a corresponding commitment to mathematical objects.
The idea that theories are to be characterized in terms of models has been
formulated in a variety of dierent ways, accompanied by a varying array of
attitudes towards theoretical commitment. One position that is compatible
with at least a form of scientic realism is Gieres (Giere 1999). Giere emphasizes
that constructing models involves considerable idealization. He writes that
theoretical hypotheses have the following form:
The designated real system is similar to the proposed model in specied respects
and to specied degrees. (Giere 1999, pg. 179)

Theoretical statements, in his view, are only true of the models, not of the
world. If this account of theories is correct, then theoretical statements are
not to be understood as strictly and literally true, but only true of our models.
Notice that on such a view we can account easily for the legitimacy of much
scientic reasoning. Our reasoning proceeds from theoretical statements about
our models to other statements about the models. Such reasoning is to be
understood as about our models and not literally about the world. Of course,
we must also account for inferences from the models to the world, but at least
we can say that mathematically formulated laws are to be understood as claims
about the models and further that much of the use of mathematics in science
just involves reasoning about them. Typically, there will be a hierarchy of models, but then the reasoning concerns the relationship between one model and
another. Such reasoning ts within an if-thenist account of mathematics, and
as such presents no diculty for the ctionalist.

142

Susan Vineberg

Although initially promising for accommodating a ctionalist view of mathematics, Gieres account of theories faces a serious problem that has implications for the mathematical anti-realist. While his is a general account, and different sorts of models will be related to the world in dierent ways, it is clearly
unsatisfactory to characterize the relationship between a successful model and
the world merely as one of similarity. Many discarded models could be regarded
as having structural features that are similar to the structural features of the
world.
Perhaps a more satisfactory account of the relationship between models and
the empirical world can be obtained by focusing on the procedures by which
the empirical data are given a mathematical representation. The idea would
be that there is a natural and systematic way in which our measurements are
represented mathematically, which provides the content to their claim of being
representations. What is required of models is that, modulo an appropriate
margin of error, there must be an isomorphism (or in many cases a homomorphism), between the model and the representation of our measurements.
Under this conception, the mathematically formulated laws that are true in our
models, would not be strictly and literally true of the world, but could be said to
be approximately true, and would be linked to physical facts through canonical measurement procedures. The view is compatible with realism in that our
models can be taken to represent the causal properties of both the observable
and unobservable physical objects postulated by our theories.
A line of this sort seems to be needed to account for the use of modeling in
science, irrespective of the issues that arise for antirealists in the philosophy
of mathematics. The picture further provides a means of making sense of assumptions made in the course of theorizing that scientists do not concern
themselves with justifying, and with certain kinds of idealization (for example
treating objects as points), which they never attempt to relax. Moreover, since
mathematical inferences are conned to reasoning concerning models, there
is no need to suppose that there is some special physical content underlying
the premises and conclusions involved. The position will be unsatisfactory to
a strict scientic realist, and it remains to be seen whether every application
of mathematics can be treated as either as dispensable or as involving idealization. Whatever the result, further investigation promises to reveal more about
science and how mathematics is applied.

Is Indispensability Still a Problem for Fictionalism?

References
Azzouni (2004). Deating Existential Commitment: A Case for Nominalism, Oxford.
Balaguer, M. (1998). Platonism and Anti-Platonism in Mathematics. New York, Oxford
University Press.
Chihara, C. (1990). Constructibility and Mathematical Existence. Oxford, Oxford University Press.
Field, H. H. (1980). Science Without Numbers. Oxford, Basil Blackwell.
Frege, G. (1978). The Foundations of Arithmatic. Oxford, Blackwell.
Giere, R. N. (1999). Science Without Laws. Chicago, University of Chicago Press.
Maddy, P. (1992). Indispensability and Practice. The Journal of Philosophy 89(6):
27589.
Maddy, P. (1997). Naturalism in Mathematics. New York, Oxford University Press.
Putnam, H. (1979a). Philosophy of Logic. Mathematics, Matter and Method: Philosophical
Papers vol. 1. Cambridge, Cambridge University Press: 323357.
Putnam, H. (1979b). What is Mathematical Truth? Mathematics, Matter and Method:
Philosophical Papers vol 1. Cambridge, Cambridge University Press. 2: 6078.
Quine, W. V. O. (1969). Existence and Quantication. Ontological Relativity and Other
Essays. New York, Columbia University Press: 9 1113.
Quine, W. V. O. (1976). Posits and Reality. The Ways of Paradox and Other Essays.
Cambridge MA, Harvard: 246254.
Sober, E. (1993). Mathematics and Indispensability. The Philosophical Review 102 (1):
3557.
Vineberg, S. (1998). Indispensability Arguments and Scientic Reasoning.
Taiwanese Journal for Philosophy and the History of Science (no. 10): 117140.

144

Susan Vineberg

Mill, Frege and the Unity of Mathematics

Part III
Historical Background

146

Madeline Muntersbjorn

Mill, Frege and the Unity of Mathematics

Mill, Frege and the Unity of


Mathematics
Madeline Muntersbjorn

Abstract:
This essay discusses the unity of mathematics by comparing the philosophies of Mill and
Frege. While Mill is remembered as a progressive social thinker, his contributions to the
development of logic are less widely heralded. In contrast, Frege made important and
lasting contributions to the development of logic while his social thought, what little is
known of it, was very conservative. Two theses are presented in the paper. The rst is that
in order to pursue Mills progressive sociopolitical project, one must embrace Freges distinction between logic and psychology. The second thesis is that in order to pursue Freges
project of accounting for the unity of mathematics, we must understand mathematics as
a human activity and consider the role that history and psychology play in the growth of
mathematics.

This essay considers the unity of mathematics and the relationship between
psychology and logic by contrasting the views of John Stuart Mill (18061873)
and Gottlob Frege (18481925). This contrast is instructive for several reasons.
Tradition has it that these two men held radically distinct views on mathematics, logic and language and that Freges more rigorous realism decisively won
out over Mills nave naturalism. As with many traditional stories, a closer look
reveals the extent to which the central plot is plausible yet overly simplistic.
One thesis articulated below is that the traditional story is correct because,
as it turns out, in order to pursue Mills sociopolitical vision for humankind,
we must cultivate a commitment to something like Freges laws of thought
as a domain of inquiry that logicians pursue independent of psychology. The
other thesis developed belowand by far the more controversial claimis
that in order to pursue Freges project of explaining the unity exhibited by
mathematics, a discussion of human practices must be included as part of our
explanation. The two theses may be summed up as follows: While there are
many good reasons to distinguish between logic and psychology, articulating
a satisfying account of the unity of mathematics is not one of them. For the
relation between the laws of thought and the reliable habits of thinking people
is like that between the chicken and the eggthey are so dependent upon one

148

Madeline Muntersbjorn

another for their existence that it is impossible to determine which comes


rst.1
Identifying the respective philosophies of mathematics of Mill and Frege is
relatively easy for their views have acquired widely used labels. Mill saw logic
as evidence of a unity of mind exhibited by all human beings. The unity that
mathematics exhibits results from our shared abilities and experiencesindependent of whether what these experiences have in common are articulated
explicitly as such or not. For example, when we say that two quantities dier,
what we mean is that dierences in size excite dierent physical sensations in
us:
But my object is to show, that when we say of two things that they dier
in quantity, just as when we say that they dier in quality, the assertion is
always grounded on a dierence in the sensations which they excite. Nobody,
I presume, will say, that to see, or to lift, or to drink, ten gallons of water,
does not include in itself a dierent set of sensations from those of seeing,
lifting, or drinking one gallon; or that to see or handle a foot rule, and to
see or handle a yard-measure made exactly like it, are the same sensations. I
do not undertake to say what the dierence in the sensations is. Everybody
knows, and nobody can tell; But the dierence, so far as cognizable by our
faculties, lies in the sensations. Whatever dierence we say there is in the
things themselves, is in this, as in all other cases, grounded, and grounded
exclusively, on a dierence in the sensations excited by them. (Book I, Ch. 3,
12: p. 46; emphasis added.)

Mills empirical view in the philosophy of mathematics is known as psychologism, the belief that all mathematical relationships have their origins in ubiquitous facts of human psychology. One gloss on this passage is that what we
mean when we say 10 > 1 is that anyone who lifts ten gallons of water will
experience it dierently from the experience of lifting one gallon. Frege read
Mill as oering this argument, which he scorned and rightfully so. For while
a water-lifting experience may give us some intuitions about relative size in
general, no experiments with buckets could teach us the dierence between,
say, 13,623 drops of water and 13,632 drops. Yet, we are cognizable of their
dierence, and to a much greater degree of precision, despite having never
interacted with any collections of objects corresponding to these two measures.
1

The Frege Mill contrast rst occurred to me as I considered the distinction between internal
and external history of mathematics. I am grateful to my colleagues at UNAM for feedback
on an earlier version of this paper presented at Coloquio Prctica Matemtica y Explicacin in
June 2005. Just as this paper was going to press in May of 2008, my colleague John Sarnecki
gave me valuable feedback.

Mill, Frege and the Unity of Mathematics

A more charitable reading would take the references to water less literally
but would still need to explain, in more detail, what Mill meant by grounded
exclusively.
More sympathetic readers of Mill suggest that the psychologism label is
undeserved:
All inferences are matters of psychological fact. For this Mill was later accused of the sin of psychologism. But this is unfair to Mill. The latter is not
claiming that the laws of logic are part of the subject-matter of the empirical
science of psychology. He is arguing, rather, that the laws of logic, of both
deductive logic and inductive logic, are normative, rules or standards about
how we ought to reason, or, at least, about how we ought to reason given that
we have a concern for matter-of-fact truth.2

If we limit the scope of the label psychologism to the view that no demarcation may be drawn between descriptive and normative inquiry, the label is
perhaps undeserved. But if psychologism is the view that relations of inequality,
such as 13,623 < 13,632, are grounded exclusively in the same kinds of experiences
that lead us to believe 1 + 2 = 3 (i.e., one pebble plus two pebbles equals three
pebbles), then Mills views are correctly labeled and Freges critique still holds.
While Wilson may yet have a point about the overlooked role of normativity in
Mills system as a whole, what Mill says about the dierence between quantity
(one gallon vs. ten) and quality (water vs. wine) exemplies what most people
mean by psychologism: What is the real distinction between the two cases?
It is not within the province of Logic to analyse it; nor to decide whether it
is susceptible of analysis or not (op cit.). Mills view on this point follows
from the leading shortcoming of his System of Logic, which we can nd on the
very rst page, in the preface to the rst edition where he outlines the books
contents:
In the portion of the work which treats of Ratiocination the author has not
deemed it necessary to enter into technical details which may be obtained
in so perfect a shape from the existing treatises on what is termed the Logic
of the Schools.

Frege devoted his life to undertaking the project Mill disdained by entering
into the technical details of ratiocination, or what we now call deductive,
rather than inductive, logic. Frege recognized that most 19th C. logic treatises
were in far from perfect shape. Today, so much has been learned about the
2

Wilson, Fred, John Stuart Mill, The Stanford Encyclopedia of Philosophy (Fall 2007 Edition),
Edward N. Zalta (ed.), URL = <http://plato.stanford.edu/archives/fall2007/entries/mill/>.

150

Madeline Muntersbjorn

dierences between quantities and qualities through the use of Freges formal
logic of properties and relations that it is hard to overstate his contribution to
the growth of mathematics and logic in the twentieth-century.
For Frege the diversity of human experience was evidence against psychologism. Like Mill, Frege saw the unity of mathematics as a fact to be explained.
Unlike Mill, Frege thought that appeals to experience threaten to undermine
this unity as a myth. True believers in the objective unity of mathematics must
seek its source elsewhere:
No, arithmetic has nothing at all to do with sensations. Just as little has it to
do with mental images, compounded from the traces of earlier sense impressions. The uctuating and indeterminate nature of these forms stands in stark
contrast to the determinate and xed nature of mathematical concepts and
objects. (Grundlagen V-VI; p. 87)3

Frege sought the unity of mathematics in formal logic, a discipline that must
be practiced in isolation of psychology:
There must be a sharp separation of the psychological from the logical, of the
subjective from the objective. (Ibid. X; p. 90.)

Frege had a sharp wit and vigorous style; psychologism, in general, and Mill,
in particular, were among his favorite targets.4 Those who would try to explain
the unity of mathematics by appeal to cognitive development or human history
are wasting their time for the studies will, necessarily, be in vain:
What are we then to say when someone, instead of carrying on this work
where it still seems incomplete, ignores it entirely, and enters the nursery or
takes himself back to the earliest conceivable stage of human development,
in order there to discover, like John Stuart Mill, some gingerbread or pebble
arithmetic! It remains only to ascribe to the avor of the cake a special meaning for the concept of number. This is surely the exact opposite of a rational
procedure and in any case as unmathematical as it could possibly be. It may
well be that the history of discoveries is useful in many cases as preparation
for further research; but it should not aspire to take its place. (Ibid. VII-VIII;
pp. 8889.)
3
4

For Frege quotes, the original location of the extract is given, followed by a page number to
Beaneys (1997) translation in The Frege Reader.
If not his most favorite target: There is hardly a piece of writing by Fregebe it book or letter,
article or reviewwhere he misses the opportunity to stigmatize the evil of psychologism. An
unmistakable mark that a philosopher has fallen prey to this infection is his tendency to blur
certain distinctions, for example, between the laws of thinking and the laws of thought
(Picardi 1996, p. 307).

Mill, Frege and the Unity of Mathematics

For Frege, logic is the study of the objective truths that must be true while
psychology is the study of subjective facts that just so happen to be true.
Hence Freges philosophy of mathematics came to be known as logicism, the
belief that all of mathematics derives from and thus may be reduced to purely
logical relationships since, like logic, mathematics is the study of objective
truths. Logicism did not become a very popular position in the philosophy of
mathematics, for several reasons, not the least of which was a growing sense
of its unfeasibility in light of 20th C. developments within formal logic itself.
Nevertheless, Freges writings on mathematics, thought and language continue
to be studied by contemporary epistemologists.
Indeed, Frege is frequently hailed as the founder of the analytic tradition
as cultivated by Bertrand Russell and other philosophers in the 20th C. However deserved this label is, and I see no reason to dispute it, Freges role in
this regard does not imply that we must thus read Frege as winner and
Mill as loser in an imagined competition between rival thinkers, if only
because the history of ideas is not a winner-take-all competition. We must always recall what Herbert Buttereld noted in (1931), in his discussion of Whig
history:
Instead of seeing the modern world emerge as the victory of the children of
light over the children of darkness in any generation, it is at least better to see
it emerge as the result of a clash of wills, a result which often neither party
wanted or even dreamed of, a result which indeed in some cases both parties would equally have hated, but a result for the achievement of which the
existence of both and the clash of both were necessary. (P. 28.)

The Mill vs. Frege contrast is an illustrative conrming instance of Butterelds


historiographical hypothesis, if only because it is not entirely clear whether either one qualies as a child of light. Identifying their respective philosophies
of mind is more dicult, as their views do not have canonical labels with which
they are associatedor, more accurately, when particular labels are applied to
their philosophies of mind they are more frequently seen as contentious. Did
Mill subscribe to the identity theory of mind or some other form of naturalism?
Was Frege a Cartesian rationalist or Platonic realist? While the labels psyhologism and logicism have the advantage of being more or less cotemporaneous
with the lives of our subjects, the labels invoked in these questions are more
anachronistic and, for this reason, have limited relevance. For this essay, as an example of protosociology, I eschew the quest for the best labels and instead sketch
their views on humanity roughly by contrasting their social philosophies. For if
Frege is one of the children of light in the history of the philosophy of math-

152

Madeline Muntersbjorn

ematics, when it comes to political theory, Mills work shines by far the more
brightly.
Mill worked as a member of the British parliament and was an original and
inspiring political thinker. As an abolitionist and suragist he argued for equal
rights among all people independent of sex and origin. In his (1869) book, The
Subjection of Women, Mill decries conservative systems that distribute privileges
and liberties on the basis of birthright to kinds of people, rather than to individuals on the basis of ability and inclination:
[T]he tendencies of progressive human society, aord not only no presumption in favour of this system of inequality of rights, but a strong one against
it; and that, so far as the whole course of human improvement up to this
time, the whole stream of modern tendencies, warrants any inference on the
subject, it is, that this relic of the past is discordant with the future, and must
necessarily disappear. (P. 139.) 5

The chief dierence between modernity and history is the extent to which this
relic has started to disappear, such that,
human beings are no longer born to their place in life, and chained down
by an inexorable bond to the place they are born to, but are free to employ
their faculties, and such favourable chances as oer, to achieve the lot which
may appear to them most desirable. (Ibid.)

Except, of course, those who are born female. This particular relic lingers most
egregiously around the world and inspired Mill to write the book under consideration. Politically, Mill was a man ahead of his time, in both his commitment
to womens rights and his insistence on the pre-eminent value of justice:
In regard, however, to the larger question, the removal of womens disabilitiestheir recognition as the equals of men in all that belongs to citizenshipthe opening to them of all honourable employments, and of the training and education which qualies for those employmentsthere are many
persons for whom it is not enough that the inequality has no just or legitimate
defense; they require to be told what express advantage would be obtained by
abolishing it. To which let me rst answer, the advantage of having the most
universal and pervading of all human relations regulated by justice instead of
injustice. (Ibid., p. 206.)

His voluminous writings on social issues, especially Utilitarianism (1859) and


On Liberty (1861) are frequently assigned readings in ethics courses. His sub5

Page numbers refer to the Modern Library Paperback (2002) collection of Mills most celebrated books.

Mill, Frege and the Unity of Mathematics

stantial (1843) treatise on logic, A System of Logic: Ratiocinative and Inductive,


is less widely studied today. When published, it was the most sophisticated
articulation yet of what we now call the philosophy of science; the endorsement on the back cover of the 2002 facsimile of the 8th edition of 1891 claims
that the 600+-page volume was an instant cult book. Contemporary introductory logic textbooks often include a section on Mills Methods, strategies
for causal reasoning, for these are now considered the most valuable parts of his
overall system. In brief, Mill sought to improve the life of humankind through
the exercise of reason and the evaluation of evidence. When we consider his
philosophy as a whole, we see how his commitment to humanity as a kind of
being whose nature can be studied as such informed his political theory and
epistemology.
In contrast, Frege was neither an original nor an inspiring political thinker.
Not only was he not ahead of his time, he deeply distrusted modernity and was
nostalgic for a noble past that probably never existed. What meager evidence
we have of his political theory is limited to a few journal pages written very
late in his life, between the two world wars. Like many of his fellow citizens
at that time and place, Frege did not believe in anything like equal rights for
all people. He believed in the unrivaled superiority of aristocratic protestant
Germans, not because he had compared them to other peoples of the world and
found them to be superior, but because he grew up with them and preferred
them out of emotional attachment, which he saw as the origin of genuine
patriotism:
It is not really true that the child compares several mothers with one another
and then by the most impartial investigation possible recognizes his mother
as the best; such an impartial investigation does not happen at all, only a
prejudice in favor of the real mother is at play. He who nds it necessary
to conduct an unbiased examination of all peoples to make up his mind as to
who is best doesnt know true patriotism. The question here is not about a
judgment in the sense of logic, not about considering something as true, but
about ones feelings and inner attitude. Only Feeling [Gemt] participates,
not Reason, and it speaks freely, without having spoken to Reason beforehand
for counsel. And yet, as times, it appears that such a participation of Feeling
[Gemt] is needed to be able to make sound, rational judgments in political
matters. (P. 337.)

In other words, while sensations play no role in the rational inquiry into the
truth, they have a necessary role to play in our political lives wherein we ought
to allow our passions to inuence our judgment. Of course, the problem with
Freges reasoning is that not all peoples are equally worthy of respect; some

154

Madeline Muntersbjorn

mothers abuse their children! World War II, from the Anschluss to Die Endlsung, or nal solution, reveals just how dangerous and irrational an overreliance on familiarity and feeling can be in the course of political aairs.
The family business into which Frege was born was a private girls school
and his mother took over as principle of the school when his father died. It is
likely that, like Mill, he supported the idea of educating girls, but only those
girls whose parents could aord the tuition and came from the right sort of
families. For, unlike Mill, Frege saw nothing amiss in political privilege, or
lack thereof, based on birthright and thought monarchy was the best model of
government. For in a legitimate monarchy,
Political experience and insight can always be handed down from father to
son, and so a centuries-old, ever more complete and improved treasure of
political experience and insight can be accumulated. (Ibid., p. 337.)

On Freges view, the only disadvantages to dynastic rule were (1) a coddling
court life, for which the antidote is military upbringing; and (2) the tendency for foreign princesses to marry into the family, a practice which is not
only dangerous but also unnecessary, for, In Germany we had within our
princely houses an aristocracy which could have supplied wives to the imperial
family (ibid., p. 338). Presumably, these wives would not only have the right
bloodline, but would also be educated at the nest separate but equal schools
for German girls.6
Frege not only distrusted the intrusion of foreign princesses into the imperial family, but also the inuence of French and English intellectuals on German political thought. Pre-eminent Frege scholar Dummett wrote that he was
shocked by the contents of the diary, insofar as it shows his intellectual hero
to have been a man of extreme right-wing opinions, bitterly opposed to the
parliamentary system, democrats, liberals, Catholics, the French and, above
all, Jews, who he thought ought to be deprived of political rights and, preferably, expelled from Germany. When I rst read the diary, many years ago, I
was deeply shocked, because I had revered Frege as an absolutely rational man,
if, perhaps, not a very likeable one. (Ibid., pp. 303304.)

When we view Freges writings as a whole, we can see how his lack of faith
in human experience as something all people share informed his political
theory as well as his epistemology. In other words, he rejected psychologism
because the view presupposed the existence of something Frege did not be6

For a quick overview of Freges Life and Work, see Thiel and Beaney (2005).

Mill, Frege and the Unity of Mathematics

lieve in, namely a unity of mind shared by all human beings independent of
origin.
Suppose Mill is right and that there are good reasons to believe that the
unity of mind is not a myth and that no particular kind of human being is
so distinct from the rest that their lot in life should be determined by their
origin: Would these facts imply that we must deny Freges distinction between
the objective laws of thought and the subjective habits of thinking people?
No, emphatically not, because the unity of mind hypothesis requires that we
demarcate between logic and psychology, if only for pragmatic reasons. Freges
lack of faith in shared human experience was not his only reason for rejecting
psychologism; there are better reasons.
One reason often given is that the distinction between logic and psychology
makes communication possiblefor if our ideas were limited to our experiences, and we did not have access to any truths beyond experience, how could
we share ideas? For my ideas would be based on my experiences while your ideas
would be based on your experiences. Since our experiences are necessarily distinct, the only way we could communicate, then, would be if there were some
ideas that were not based in any experiences, such as those studied in logic.
But the communication argument is really just another denial of the unity of
mind hypothesis, for it understands all experience to be personal, unique and
subjective. Some of our experiences may be unique, but this does not mean
that all of our experiences are necessarily so distinct as to have nothing at all in
common with one another. There may be some experiences that are shared in
the same way by more than one personor at least there is no reason to believe
that its impossible and lots of evidence that it happens all the time.
On my view, the best reason for the distinction between logic and psychology
is that it makes the pursuit of progressive social reform possible. Frege argued
that those who deny the distinction between how people actually think and
how people ought to think will not be able to distinguish between reliable and
unreliable inferences. Consistent criteria of correct reasoning are necessary for
making judgments about the relative reliability of judgment-making strategies.
Even those who dont believe in the transcendent existence of analytic truths
or self-evident axioms can agree that criteria of logical correctness should be
as culturally-neutral as possible. For example, if you want to promote equal
rights as a public good you need an account of what equal means that is not
limited to what passes for equality in actual human aairs. Making implicit
criteria of judgment explicit by articulating our experiences may play a role at
the outset of critical inquiry, but these articulations can never be the end. For

156

Madeline Muntersbjorn

we must subsequently assess our criteria for soundness by looking for the gaps
in the proofs or holes in our arguments where ungrounded assumptions
threaten the integrity of our inquiry.
As Arp (2005) argues, there are compelling pragmatic reasons involving the
articulation of normative criteria for rational thought that require us to make a
distinction between how people actually think and how they ought to think:
The case can be made that such notions as The nature of AIDS and What
is best for my child, as well as the propositions communicated about these
notions, are not only nonreducible to the beliefs of a particular thinking
community, but also, actually are abstract objects having a truth-value that
is discoverable. (P. 33.)

For these reasons,


It is good to have Platonists in the scientic or philosophic community reminding these communities not to rest on the laurels of coherentist pragmatism; the question will always remain as to whether coherence is enough.
(P. 35.)

Thus, in an exquisite irony of intellectual history, it turns out that in order to


pursue Mills vision of a more just society, we must cultivate a commitment
to something like Freges laws of thought as objective patterns of reasoning
that any rational agent exhibits, insofar as she is rational, and as a domain of
inquiry logicians can pursue independent of the social sciences.
This thesis, the rst of the two oered in this essay, is endorsed by many
philosophers today, especially those in the analytic tradition such as myself
who teach logic as a calling and not just as a living. The second thesis is more
controversialthough admittedly, few people outside the analytic tradition
have an opinion on the matter one way or the other. Suppose both Mill and
Frege are right and that there are good reasons to believe that the unity of
mathematics is not a myth either and that, consequently, all branches of
mathematics trace back to common roots: Is this unity best explained by an
appeal to logic? Must we reject Mills belief that the common roots of all mathematical branches may be traced back to human experience? Not necessarily.
Its possible that what binds mathematics together as such is a combination of
distinctive kinds of human practicessome intuitive, others symbolic, many
explanatory. Freges distinction between the laws of thought, on the one hand,
and the mental habits of thinking people, on the other, while necessary in some
explanatory contexts, is not sucient in all contexts. One can commit to the

Mill, Frege and the Unity of Mathematics

clarity and practical necessity of this distinction while, at the same time, advocate a more interdisciplinary approach to explaining the unity of mathematics.
In other words, it is possible to distinguish between logic and psychology and,
at the same time, study the causal relationships between them by writing the
history of mathematics as an interactive dialectic between the two domains.
The approach advocated here is inspired by Beaneys dialectical reconstruction as introduced in his (1996) Frege: Making Sense, where he addresses the
challenge of writing the intellectual history of someone who disdained intellectual history as irrelevant to critical inquiry or to an increased understanding
of logical relations (pp. 35).
As noted as the outset of this essay, logic and psychology are distinct disciplines for good reasons, but accounting for the unity of mathematics is not one
of them. Any scholar who wishes to explain the unity of mathematics over time
must incorporate insights from both disciplines into her research. The relationship between how we actually think and how we ought to think is not like
a static structure, wherein our reliable habits rest for all time on foundations
of self-evident axioms. Nor are our reliable habits simply the by-products of
natural selection on a particular form of neuronal cognitive architecture. The
laws of thought neither fall from heaven above nor trickle up from our bodies
below, waiting to be discovered by either logicians or psychologists working in
isolation. This imagery is taken from Van Kerkhove and Myens (2002) critique
of Lako and Nuez (2001):
Mathematics now becomes determined by a xed realm of entities, no longer
situated in Platos heaven, but constituted by the mechanics of the mind:
mathematical structure has been moved from heaven into our heads. Humanity appears to lose deliberate control over mathematics to anonymous brain
mechanisms. This means that in the end, we are out of touch with the world
of mathematics, now not because its up above in Platos heaven, but instead
because it is buried deep down in ourselves. (P. 360361.)

As noted at the outset of this essay, the relationship between logic and psychology is like the relationship between the chicken and the egg: it is very dicult
to say which comes rst because they are so dependent upon one another.
Van Kerkhove and Myen alert us to the dangers of understanding mathematical
content as grounded exclusively in either purely concrete or abstract realms.
Both approaches threaten to overly privilege one side of the distinction between what we think mathematics is about and what mathematics really
is about, to the exclusion of the other, while creating puzzles as to how our
thoughts engage genuine mathematical content.

158

Madeline Muntersbjorn

As more and more scholars look to actual mathematical practices as relevant


to the cultivation of a satisfying account of mathematical content, emergentist ontologies have been proposed wherein mathematical objects are taken
to be real, but not eternal, phenomena that come into existence, in part, as
a result of human activity.7 An early statement of the mutual interdependence between what exists and what can be said may be found in Shapiro
(1989):
The mathematical universe does not come to us, nor does it exist, already
divided into objects, waiting to be studied. If anything, it is the other way
aroundthe type of discourse and its allowed inferences determine (at least
in part) the nature of the objects. The mutual dependence of object and logic
is clearest when one focuses on dynamic mathematical practice, in which the
inferences allowed in the reasoning of a branch of mathematics are directly
related to the sorts of moves available to the ideal constructor. (P. 24.)

What is needed are better accounts not only of the relationship between logic
and psychology in the emergence of new mathematics over time, but also more
ne-grained accounts of the object-process duality inherent in mathematical
content, a duality which may go a long way to explaining the metaphysical
challenges mathematics has long posed for philosophers.
Consider, for example, how Wells (2003) explains this tension in his Handbook of Mathematical Discourse in his object-process duality entry:
Mathematicians thinking about a mathematical concept will typically hold it
in mind both as a process and an object. As a process, it is a way of performing mathematical actions in stages. But this process can then be conceived as
a mathematical object, capable for example of being an element of a set or
the input to another process. Thus the sine function, like any function, is a
process that associates to each number another number, but it is also an object
which you may be able to dierentiate and integrate. The mental operation
that consists of conceiving of a process as an object is called encapsulation, or
sometimes reication or entication. (P. 180.)

The duality that Wells identies does not necessarily accrue to all mathematical
processesat least I see no reason to presume any such universality of duality
throughout the mathematical universe. On my view, mathematical objects
7

Hersh (2005) he writes, And at last, in 2003 and 2004, a few philosophers are also recognizing that mathematical objects are real and are our creations. (Jessica Carter, Ontology and
Mathematical Practice, Philosophia Mathematica 12 (3) 2004; M. Panza, Mathematical
Proofs, Synthese 134 2003; M. Muntersbjorn, Representational Innovation and Mathematical Ontology, Synthese 134 2003) (pp. viii-ix).

Mill, Frege and the Unity of Mathematics

come into being over time as implicit practices for manipulating extant mathematical expressions become explicit articulations in and of themselves. Some,
but not all, of these articulations may be taken to refer to objects as part of
the process of mathematical explanation. Those that do, like functions, were
once symbolic means to other ends that became objects of study in themselves
as part of the process of solving further problems, including, as Freges own
contributions to the growth of mathematics show, the problem of explaining
the unity of mathematics.
The line between logic and psychology must be drawn, but how it gets drawn
is going to vary over time and place. For this reason, we need to beware of some
common false beliefs people sometimes draw from this distinction, especially
when they fall into the trap of believing they know what the laws of thought
are for all time. All variety of specious projects with dubious prospects, from
the search for extra-terrestrial intelligence to mathematical proofs of intelligent
design, rely on Freges distinction concomitant with a particular view of what
intelligence must be like, typically in the form of an unreective faith in the
resemblance of ones own mental habits to the laws of thought. An independent
criterion of correct reasoning is a necessary prerequisite for making judgments
about the reliability of judgments. However, even Frege had misgivings about
the extent to which we can express the laws of correct reasoning, if only because
we will be limited to the use of some kind of human language. Indeed, Frege
thought that our reliance on concrete means to represent these laws limited
our capacity to express thought in a pure form. While his concept-script is an
improvement in notation, as is surely inevitable in the case of external means
of representation, even this cannot make thought pure again (Begrisschrift
VII, p. 51).
Neither Frege nor Mill were so nave as to subscribe to such an unreective
faith in the intellectual products of human inquiry. Both philosophers warned
that we must not be misled by the existence of nouns in our language into
thinking that objects must exist that are named by these words because we have
no reason to presume that what actually exists will ever match up, exactly, with
any cultures list of substantives. Another largely overlooked point of agreement
between these two philosophers is their shared commitment to the instrumental role of language as a tool of discovery as well as communication:
Logic, then, is the science of the operations of the understanding which are
subservient to the estimation of evidence: both the process itself of advancing
from known truths to unknown, and all other intellectual operations in so
far as auxiliary to this. It includes, therefore, the operation of Naming; for

160

Madeline Muntersbjorn
language is an instrument of thought, as well as a means of communicating
our thoughts. (Introduction 7: pp. 67.)

Frege took the idea of language as an instrument of thought seriously. Like


Mill, Freges rst major book was a study in logic which he called Begrisschrift,
or concept-script. Unlike Mill, Freges book was a failureat least at rst.
Freges major innovation was a formal language that takes patience and skill to
master. Despite numerous innovations in the actual symbolization employed
in modern textbooks over Freges original notational scheme, the study of
predicate logic generates much gnashing of teeth among students of mathematics and philosophy today. Frege warned his readers of the work was in store for
them in the introduction to his book:
I believe I can make the relationship of my Begrisschrift to ordinary language
clearest if I compare it to that of the microscope to the eye. The latter, due
to the range of its applicability, due to the exibility with which it is able
to adapt to the most diverse circumstances, has a great superiority over the
microscope. Considered as an optical instrument, it admittedly reveals many
imperfections, which usually remain unnoticed only because of its intimate
connection with mental life. But as soon as scientic purposes place great
demand on sharpness and resolution, the eye turns out to be inadequate. The
microscope, on the other hand, is perfectly suited for just such purposes, but
precisely because of this is useless for all others. (Op cit.V; p. 49.)

Unfortunately for Frege, few of his readers were up to the challenge until
Russell championed Freges vision. While the details are neither transparent
nor easily articulated, understanding language as an instrument of thought
whereby we gain access to logical reality helps us better understand both Shapiros ideal constructor and the processes whereby mathematical practices
give rise to mathematical content.8
Contemporary philosophers, especially those with nominalist intuitions,
reject both the psychologism of Mill and the logicism of Frege. Signicantly,
nominalists do not recognize a unity of mathematics. They demarcate sharply
between the methods of science and the study of mathematics and distinguish
between pure mathematics. Nominalists are eager to distinguish between pure
mathematics (where nothing really exists) and applied mathematics (using
mathematical languages to re-state problems involving real things). Even mathematics itself is often divided into two kindsthe mathematics that people
8

See also Sfard (1998) wherein she shows how the semantic space created by the introduction
of a new signier may be replenished with content coming from within the mathematical
discourse itself (p. 85).

Mill, Frege and the Unity of Mathematics

actually practice, wherein they naively refer to objects as if they existed, and an
imagined mathematics that has been nominalized or puried of any references to any things.9 Consider, for example, this passage from Azzouni (1994):
A crucial part of the practice of empirical science is constructing means of access to (many of ) the objects that constitute the subject matter of that science.
Certainly this is true of theoretical objects such as subatomic particles, black
holes, genes, and so on. Empirical scientists attempt to interact with most
of the theoretical objects they deal with, and it is almost never a trivial matter
to do so. Scientic theory and engineering know-how are invariably engaged
in such attempts, which are often ambitious and expensive. Nothing like this
seems to be involved in mathematics. At best the process the mathematician
uses to engage mathematical objects seems to be like introspection. This may
explain the nineteenth-century temptation to reduce mathematical objects to
psychological ones. (P. 5.)

As both Mill and Frege knew, formal mathematical languages have more than
one function. Not only do they communicate ideas, they also help us make
discoveries, and, in particular, discoveries about relationships between ideas.
Contra nominalism, formal languages provide us with access to abstract objects. Access is not granted exclusively through simplistic subjective processes of
introspection, but via complex inter-subjective practices of articulation and
explanation. While introspection may play a role, even more kinds of mathematical activity involve constructing shared means of access to mathematical reality, which includes object-process dualities, as cultivated into existence
via the judicious manipulation of formal languages.
The more scholars pay attention to mathematical practice, the more they
realize that the processes mathematicians use to engage mathematical objects
are more like the processes empirical scientists use than most philosophers
realize. The mathematicians reliance on pencil and paper may mean that their
projects cost less than those of their natural science colleagues. But time is
moneyespecially uninterrupted time to draw sketch after sketch until the
proof begins to take shapeand mathematical theory and problem-solving
know-how are invariably engaged in such attempts. Mathematical projects
tend to be ambitious and rarely trivial insofar as proof practices are intended
to generate lasting results, intelligible to anyone who takes the time to master
the methods and the material. Recall the objection to Mills views of quantity
about the drops of water. The argument claims that we have no experiences on
9

Cf. Vinebergs critique of Fields response to the indispensability argument in this volume.

162

Madeline Muntersbjorn

which to derive an understanding of the dierence between 13,623 and 13,632.


But this objection ignores our experiences with the numerical representations
themselves. It is relatively easy to contrive an experience to teach inequality as
a real relationship in the world. All we need do is take ourselves back to the
nursery and distribute a treat inequitably among the children and see how
their experiences of less than inform their howls of protest. Can we contrive
experiences to teach people that 13,623 is smaller than 13,632? We not only can,
we dotheyre called math classes! The cultivation of a more sophisticated
mathematical understanding requires more specialized mathematical experiences, more contrived and self-conscious than Mills overly simplistic examples
suggested. If we want to explain the unity of mathematics, we need to account
for the diversity of mathematical experiences, including the experiences of developing and manipulating formal mathematical languages, both as a means of
communicating ideas between individuals as well as an instrument of discovery
within mathematical communities.

References
Arp, Robert (2005). The Pragmatic Value of Freges Platonism for the Pragmatist
Journal of Speculative Philosophy 19 (1): 2241.
Azzouni, Jody (1994). Metaphysical Myths, Mathematical Practice. Cambridge: CUP.
Buttereld, Herbert (1931). The Whig Interpretation of History. Available online at http://
www.eliohs.uni.it/testi/900/buttereld
Frege, Gottlob (1997). The Frege Reader. Michael Beaney, ed. Oxford: Blackwell.
Hersh, Reuben (2005). 18 Unconventional Essays on the Nature of Mathematics. New
York: Springer.
Lako, George and Rafael Nuez (2001) Where Mathematics Come From: How the
Embodied Mind Brings Mathematics into Being. New York: Basic Books.
Meneldsohn, Richard (1996). Diary: Written by Professor Dr Gottlob Frege in the
Time from 10 March to 9 April 1924. Edited and annotated by Gottfried Gabriel
and Wolfgang Kienzler. Inquiry 39: 303342.
Mill, John Stuart [(1891) 2002]. A System of Logic: Ratiocinative and Inductive. Honolulu: University Press of the Pacic.
-(2002). The Basic Writings of John Stuart Mill. Introduction by J. B. Schneewind.
Notes and Commentary by Dale E. Miller. New York: The Modern Library.
Picardi, Eva (1996). Freges Anti-Psychologism, in Matthias Schirn, ed. Frege: Importance and Legacy. New York: De Gruyter, pp. 307346.
Sfard, Anna (1998). Symbolizing Mathematical Reality into Beingor, How Mathematical Discourse and Mathematical Objects Create Each Other in P. Cobb, K.

Mill, Frege and the Unity of Mathematics

E. Yackel, & K. McClain, eds, Symbolizing and Communicating: Perspectives on


Mathematical Discourse, Tools, and Instructional Design. Mahwah, NJ: Erlbaum,
pp. 3798.
Shapiro, Stewart (1989). Logic, Ontology, Mathematical Practice, Synthese 79:
1350.
Thiel, Christian and Michael Beaney (2005). Freges Life and Work, in M. Beaney
and E. Reck, eds., Gottlob Frege: Critical Assessments of Leading Philosophers. Volume
I. London: Routledge, pp. 2339.
Wells, Charles (2003). A Handbook of Mathematical Discourse. West Conshohocken,
PA: Innity Publishing.

164

Raaella De Rosa and Otvio Bueno

Descartes on Mathematical Essences


Raaella De Rosa and Otvio Bueno

Abstract
Descartes seems to hold two inconsistent accounts of the ontological status of mathematical
essences. Meditation Five apparently develops a platonist view about such essences, while
the Principles seems to advocate some form of conceptualism. We argue that Descartes was
neither a platonist nor a conceptualist. Crucial to our interpretation is Descartes dispositional nativism. We contend that his doctrine of innate ideas allows him to endorse a hybrid
view which avoids the drawbacks of Gassendis conceptualism without facing the diculties
of platonism. We call this hybrid view quasi-platonism. Our interpretation explains Descartes account of the nature of mathematical essences, dissolves the tension between the two
texts, and highlights the benets of Descartes view.

Descartes seems to provide two prima facie inconsistent accounts of true and
immutable mathematical essences. Meditation Five suggests that Descartes was
a platonist about mathematical essences. The Principles suggests that he held
some kind of conceptualist view about such essences. We argue that, despite
recent defenses of either Descartes platonism or conceptualism, he was neither
a platonist nor a conceptualist.1 Crucial to our interpretation of Descartes is
his dispositional nativism. We contend that his doctrine of innate ideas allows
him to endorse a hybrid view that we will call quasi-platonism which avoids
the pitfalls of Gassendis conceptualism without falling into the troubles of platonism. Descartes account of the nature of mathematical essences is explained,
the tension between the two texts dissolved, and the benets of Descartes
considered view are explored.

1. Ideas of Mathematics qua Ideas of True and Immutable


Essences
What are ideas of mathematics according to Descartes? Or what do they rep1

For a neo-platonic interpretation of Cartesian mathematical essences, see (Schmaltz 1991) and
(Rozemond 2008), and for a conceptualist interpretation, see (Chappell 1997), (Nolan 1997)
and (Nolan 1998).

Descartes on Mathematical Essences

resent? Descartes answer to these questions is very clear, although its implications arent. Ideas of mathematical objects, like the ideas of God and the mind,
are ideas of true and immutable essences (T&IEs) and they are contrasted
with ideas of ctitious essences (FEs). This contrast emerges, inter alia, in a
letter to Mersenne, dated 16 June 1641, where Descartes writes that the ideas
of God, mind, body, triangle [] represent true, immutable and eternal essences (CSMK, 183), and that these ideas are to be distinguished from ideas
that represent FEs (CSMK, 184). In Meditation Five and First Set of Replies, the
fact that the idea of God, like the idea of a triangle, represents a T&IE rather
than a FE is the bulwark of Descartes ontological proof.
Distinguishing between FEs and T&IEs becomes then crucial for Descartes.
A FE, Descartes claims, is an essence put together by the intellect (CSM II,
83).2 To Caterus complaint that the ontological argument proves Gods existence only if we already assume it, Descartes replies that deriving the property
of something from its essence is question-begging only in the case of FEs (CSM
II, 834; 46 and 263). In the Letter to Mersenne just mentioned, he explains:
[] if from a constructed idea I were to infer what I explicitly put into it
when I was constructing it, I would obviously be begging the question; but
it is not the same if I draw out from an innate idea something which was
implicitly contained in it but which I did not at rst notice in it. (CSMK, 184,
emphasis added.)

In Meditation Five, Descartes writes that the innate ideas of God and the
triangle [] are not my invention but have their own true and immutable
natures (CSM II, 445, emphasis added). And as a way of further explicating
the notion of T&IE, he adds: When, for example, I imagine a triangle, []
there is [] a determinate [] essence [] of the triangle which is immutable
and eternal, and not invented by me or dependent on my mind (CSM II, 445,
emphasis added). Ultimately, according to Descartes, an important feature of
T&IEs (as opposed to FEs) is that they have properties I clearly recognize
whether I want or not (CSM II, 456).3
2
3

Examples are: winged horse, existing lion, triangle inscribed in a square.


It has been noticed in the literature that Descartes may have failed to give a successful criterion
for distinguishing T&IEs from FEs. (Wilson 1978), for instance, argued that Descartes gives
two dierent criteria for distinguishing T&IEs from FEs: (i) Ideas of T&IEs imply unforeseen
and unalterable consequences (see CSM II, 45). (ii) Ideas of T&IEs cannot be analyzed into
component ideas (see CSM II, 8384). Neither criterion, according to Wilson, is adequate.
(i) fails because we can easily conjure up an idea of an invented essence that has unalterable
and unforeseen implications. (For example, we can conjure up the term Onk, and dene

166

Raaella De Rosa and Otvio Bueno

If ideas of mathematics are ideas of objects having T&IEs, and the truth
of mathematical propositions depends on their conformity to the essence of
things (CSM II, 262), it becomes crucial to understand Descartes conception of the ontological status of these essences. Are the latter abstract objects,
distinct from both particular things and nite minds? Are they immanent in
particular objects? Or are they modes of conceiving of particular things, which
have no separate existence from either particular objects or nite minds?

2. Descartes on the Ontological Status of T&IEs: Two


Opposing Accounts
Descartes oers two prima facie inconsistent accounts of the nature of T&IEs.
Some passages suggest that he held a platonist view of mathematical essences,
where platonism is dened as follows:
Platonism: T&IEs are abstract entities. They exist outside space and time
and they are prior to, independent of, and distinct from both particular
existing things and the human mind. They would exist (or subsist)
even if there were no nite minds and no material things.
In Meditation Five, Descartes writes:
[] when I imagine a triangle, even if no such gure exists or has ever existed
anywhere outside my thought, there is still a determinate [] essence, or form
of the triangle which is immutable and eternal, and not invented by me or
dependent on my mind. (CSM II, 45, emphasis added.)

This passage suggests that Descartes held a platonist view, according to which
the essence of a triangle is an extra-mental, abstract entity. Certainly Gassendi
interpreted Descartes as holding this view. In criticizing the passage, Gassendi
takes himself to be opposing the view that there is a universal nature [of triit as the rst non-terrestrial life form to be discovered by man. Reproduction and nourishment would be necessary properties of Onk, whether I want it or not, even if Onk has an
invented essence; see (Wilson 1978, 17 1172.) (ii) fails also because the notion of an existing
lion [which is an example of a FE] and that of a triangle [which is an example of a T&IE]
seem to be equally analyzable (Wilson 1978, 173). See also (Doney 1993). For an illuminating
discussion of these issues, see (Edelberg 1990).

Descartes on Mathematical Essences

angle] before [any particular triangle existed] and before the intellect performed
the abstraction (CSM II, 222, emphasis added)
Anthony Kenny, over the years, has defended this platonist interpretation of
Descartes. According to Kenny, Descartes is the founder of modern platonism
because [] the geometers triangle is an eternal creature of God, with its
own immutable nature and properties, a real thing lacking only the perfection
of actual existence.4 Kennys argument is that given Descartes claim that all
eternal truths have been created by God (CSMK, 23 and CSM II, 261) by efcient causality (CSMK, 25); and given passages such as the one quoted above
from Meditation Five, Cartesian essences are abstract entities independent of
both particular existing things and the human mind.5 Hence, Descartes is a
platonist about essences.6
Assuming that this is Descartes considered view, his modern platonism
would have to answer the question that mars any kind of platonism, viz., how
can a nite mind which is spatially located have knowledge of abstract objects
that exist outside space? One could argue that since the Cartesian mind is a
thinking un-extended substance that isnt spatially located, in principle there
would be no problem in explaining its access to these abstract objects. The unlocated mind sees or grasps these objects with the minds eyes (whatever
that is).
This may be right but the original problem is not so easily disposed of. The
analogy with seeing does not help much, since it only goes so far. The mind
cant see these objects since it is in no causal connection with them. Moreover,
no mechanism that explains how the mind is connected to these objects has
been provided. And without some such mechanism, no account of the possibility of mathematical knowledge is on oer. Such mechanism had better be
reliable as well; otherwise, the resulting account of mathematical knowledge
wouldnt even be minimally adequate. Interestingly, we cant nd a single place
where Descartes tackled these issues. And maybe thats a hint that his position
on mathematical objects didnt raise these problems.
Various other passages, however, suggest a rather dierent interpretation
of Descartes views on mathematical essences. According to these passages,
4
5
6

(Kenny 1970, p. 697).


For Descartes identication of essences with eternal truths see, for example, Letter to
Mersenne, May 27, 1630 (CSMK 2526) and (Schmaltz 1991).
(Kenny 1970, 6937). For other platonist readings of Descartes view of eternal essences, see
(Wilson 1978) and (Curley 1978). For interesting criticisms of Kennys interpretation, see
(Chappell 1997) and (Nolan 1997). More recently, a neoplatonist reading of Cartesian essences
has been oered by (Schmaltz 1991) and (Rozemond 2008).

168

Raaella De Rosa and Otvio Bueno

Descartes held what could be called a conceptualist view of essences, where


conceptualism is dened as follows:
Conceptualism (broadly construed): T&IEs are not distinct from existences.
Numbers, for example, are nothing distinct from the things numbered.
When considered in the abstract or in general, they are merely modes
of thinking of things.
In his discussion of number, duration and order in the Principles, Descartes
very carefully species:
We shall have a very distinct understanding of duration, order and number,
provided we do not mistakenly tack on to them any concept of substance.
Instead we should regard the duration of a thing simply as a mode under
which we conceive the thing in so far as it continues to exist. And similarly we
should not regard order or number as anything separate from the things which
are ordered and numbered but should think of them simply as modes under
which we consider the things in question. (Principles I.55, CSM I, 211; emphasis
added.)

Later in the Principles, Descartes insists that the distinction between quantity
or number and the thing that has quantity or number is only conceptual (see
Principles II.8, CSM I 226). Finally, in a letter to an unknown correspondent,
Descartes writes about the essence of a triangle:
[] when I think of the essence of a triangle, and of the existence of the same
triangle, these two thoughts, as thoughts, even taken objectively dier modally
in the strict sense of the term mode; but the case is not the same with the
triangle existing outside thought, in which it seems to me manifest that essence
and existence are in no way distinct. The same is the case with all universals.
Thus, when I say Peter is a man, the thought by which I think of Peter diers modally from the thought by which I think of man, but in Peter himself being a man is
nothing other than being Peter. (CSMK, 280281, emphasis added)

All these passages suggest that the distinction between the existence and essence
of a triangle is conceptual rather than real. Although the ideas of the essence and
existence of a triangle are two dierent ideas capable of existing independently
of one another (they dier modally, as Descartes puts it), being a triangle is
nothing other than being this or that particular triangle. Essence and existence
are not really distinct. Essences do not exist in a realm separate from existing
things.7
7

For Descartes denition of real distinction, modal distinction and conceptual distinction,
see Principles, I.6062 (CSM I, 213215). Further evidence can be found in (Descartes 1975, 58).

Descartes on Mathematical Essences

The problem consists in assessing what brand of conceptualism we ought to


attribute to Descartes based on the above texts. One may wonder whether, or
to what extent, Descartes conceptualism is the same as Gassendis. According to Gassendi, essences are nothing but the way in which the nite mind
classies, or conceives of, things after having formed the idea based on the
observed similarities among things. In particular, for Gassendi, a triangle is
a kind of mental rule which you use to nd out whether something deserves
to be called a triangle (CSM II, 223), and whose nature is not distinct from
the intellect which, after seeing material triangles, has formed [it] (CSM II,
223). Similarly, Descartes suggests that essences and universals are only modes
of thought that we use for thinking of all individual items which resemble
each other (CSM I, 212).
However, there are some fundamental dierences between Gassendi and
Descartes. First of all, by saying that the triangle, when considered in the abstract, is nothing but a mode of thinking of particular things, Descartes doesnt
mean to suggest that the essence of the triangle, for example, is abstracted in
the sense of extracted from observation of particulars. Rather, he means
that the triangle is conceived in abstraction from (i.e. independently of ) any
particular triangles.
Secondly, the essence of a triangle, for both Gassendi and Descartes, is
whatever particular triangles have in common. But for Descartes, what these
particulars have in common is a function of the idea of the triangle; and,
therefore any property we attribute to triangles is grounded in the idea and
belongs necessarily to them. Gassendis view is the opposite. The idea is based
on the observed similarities among particular things and therefore any property we attribute to them doesnt necessarily belong to them. Consequently,
Gassendis conceptualism is best understood as the view that essences are ideas
that neither designate abstract entities nor correspond exactly to the nature of
physical objects.
Its worth noting that conceptualism la Gassendi faces at least two problems. First, if mathematical objects are nothing but modes of thinking, it is
unclear why they describe so successfully the physical world. Secondly, given
the empiricist view on the origin of mathematical ideas, it is unclear how a
conceptualist can explain the necessity of mathematical statements.
Alan Gewirth has defended a dierent conceptualist interpretation of Descartes. He argues that Descartes conceptualism is a form of Aristotelianism
since on the Aristotelian interpretation, mathematical essences are quantitative abstractions from natural substances; so, numbers are not really distinct

170

Raaella De Rosa and Otvio Bueno

from the things numbered.8 Gewirth means that, according to Aristotle, essences are in the objects; and in knowing essences, the mind abstracts from particular objects so that there is no real distinction between essences and things.
Notice, however, that this view counts as conceptualist only to the extent that
essences qua abstractions from nite things are not really distinct from them.
The distinction between particular things and essences is conceptual because
essences exist only where they are instantiated. However, essences arent mere
ideas because they are literally present in things.9
In conclusion, dierent passages in Descartes writings, and dierent interpretations of them, suggest that Descartes may have held any of the following
three accounts of essences:
(1) Platonism: Essences are abstract entities, distinct from particular things
and nite minds.
(2) Aristotelianism: Essences are in the objects. So they arent mere ideas.
But given that there are no non-instantiated essences, the distinction
between particular objects and essences is only conceptual.
(3) Conceptualism ( la Gassendi): essences are nothing but ideas that neither designate abstract entities nor correspond exactly to the nature of
physical objects.
What is Descartes considered view? In what follows, we argue that Descartes
held a theory of mathematical essences that (1) ought not to be identied
with any of the three proposals above; (2) dissolves the diculties raised for
both platonism and conceptualism ( la Gassendi) and (3) dissolves the prima facie inconsistencies between the passages from Meditation Five and the
Principles.

3. Descartes Quasi-Platonism
Tad Schmaltz, in his Platonism and Descartes View of Immutable Essences,
pointed out that there is no overt eort in the literature to resolve the tension between the two seemingly dierent accounts of essences that Descartes
8
9

(Gewirth 1970, 678).


In the past (Gueroult 1984), and more recently (Chappell 1997) and (Nolan 1997), have
defended a conceptualist reading of Descartes view on essences.

Descartes on Mathematical Essences

oers in dierent texts. As he puts it, Kenny did not confront the apparent implications of the Principles that immutable essences cannot be distinct
from created material and mental substances, while Gewirth and Gueroult
failed to explain the contention in Meditation V that immutable essences
can be identied with properties neither of nite minds nor of particular
bodies.10
Schmaltzs view is that the two accounts above can be reconciled by attributing to eternal essences and truths the same ontological status that laws have,
viz., that of divine decrees or moral entities. Like laws of nature, eternal essences and truths are divine commands imprinted in the heart of man rather
than being special entities created by God.11
We are sympathetic with Schmaltzs proposal to reconcile these two accounts
but we believe that there is a more straightforward way to do so. Descartes was,
as we will call it, a quasi-platonist about essences: his quasi-platonism consists precisely in his avoiding the pitfalls of Gassendis conceptualism without
falling into the diculties of platonism. This quasi-platonism can be more
simply explained within the framework of Descartes dispositional nativism.
Heres our view.12
Dispositional nativism, broadly construed, is the view that an idea is innate
if, and only if, the mind has the innate disposition to form that idea under
appropriate circumstances. The locus classicus of Descartes endorsement of
dispositional nativism is a passage from the Comments on a Certain Broadsheet
where Descartes claims that ideas are innate in the mind in the same way in
which certain diseases are innate in certain families. As children of those families are not born with the disease but with the disposition to contract it later
10 (Schmaltz 1991, 134).
11 See, for example, Letter to Mersenne, 15 April 1630 (CSMK, 23). In support of the ontological
status of eternal truths and laws as moral entities, Schmaltz refers to Descartes Sixth Replies
(CSM II, 294). See (Schmaltz 1991, 138).
12 We dont deny that there are sparse suggestions in (Schmaltz 1991) that Descartes doctrine of innate ideas may oer a solution to the problems of reconciling the two sets of
texts. However, Schmaltz overall underplays the role of dispositional nativism in explaining the consistency of Descartes view. His aim is to avoid the platonic interpretation of
Descartes without thereby falling into the Charybdis of the abstractionist interpretation
of Gueroult and Gewirth (Schmaltz 1991, 162163). According to Schmaltz, the abstractionist interpretation cannot accommodate the claim in Meditation V that the immutable
essence of a triangle does not depend on human thought (ibid., p. 163). However, when
Schmaltz does draw a connection between the innateness of ideas of essences and the claim
that essences are divine decrees (see, for example, ibid., p. 162), he doesnt expand on this
proposal.

172

Raaella De Rosa and Otvio Bueno

in life, so we are not born with certain ideas but with the disposition to form
them later in life. 13
Despite the problems one may raise against a dispositional form of nativism,
the pressing issue at this point is: How does Descartes dispositional nativism
(however spelled out) relate to his quasi-platonism?
We take Descartes considered view at least with regard to mathematical
essences to be as follows. Essences, when considered in the abstract, are only
modes of conceiving things as opposed to being either immanent forms (contra
Aristotle) or forms subsisting in a third realm (contra Plato). By claiming that
essences are modes of thought, Descartes is distancing himself from (at least
orthodox) Platonism and Aristotelianism.
However, in claiming that essences are modes of thought, Descartes is not
endorsing a conceptualism la Gassendi, according to which essences are ideas
formed by the nite mind after having observed particular things. Rather, these
essences are God-given or innate ideas; they are Gods way of predisposing our
minds to think of, and represent, things in certain ways.14
So Descartes view that essences are nothing but universal ideas implicitly
encloses three criticisms: (a) it opposes platonism about mathematics, since it
denies that numbers are prior to, independent of, and distinct from both material things and the human mind; (b) it opposes Aristotelianism for universals
are modes of thinking that do not have a common foundation in things;15
and, (c) it opposes Gassendis conceptualism because, according to Descartes,
essences are to be identied with innate ideas. It follows that Descartes view
cannot be identical to any of (1)-(3) above.
But one could object that claiming that, according to Descartes, essences
are to be identied with innate ideas doesnt per se rule out platonism because
ideas can be abstract objects that the nite mind apprehends. Alternatively, one
could inquire, are innate ideas abstracta, according to Descartes?
Answers to this question may vary depending on ones views on what Des13 See CSM I 303304. For a detailed account of Descartes dispositional nativism, see (De
Rosa 2000) and (De Rosa 2004).
14 (Chappell 1997) and (Nolan 1997) have defended a similar view. Despite apparent
similarities between our readings of Descartes, our argumentative strategies are dierent. According to Nolan and Chappell, Descartes is a conceptualist. But, according to
us, Descartes view is more nuanced since it contains elements of platonism. One of
the authors (De Rosa) is currently writing a second paper on the topic that explains
the theoretical reasons why Descartes cannot be regarded as either a conceptualist or a
platonist.
15 See on this, for example, (Bolton 1998).

Descartes on Mathematical Essences

cartes account of ideas is. If one is inclined to explain the intentionality of


ideas by postulating an object of thought that mediates the mind-world relation, then it is easy to think of ideas as abstract objects that the nite mind
apprehends. However, there are good grounds to believe that this was not
Descartes considered view. According to Descartes, ideas of T&IEs are general
ideas, and these ideas are mental representations (as opposed to abstracta) with
an intentional content whose properties mislead us into thinking that Descartes
is postulating abstract objects.
In particular, we contend that we make the mistake of inferring that Descartes is postulating abstract objects when he talks about ideas of T&IEs because of two features he attributes to the content of these ideas: (a) the mind
cannot manipulate the intentional content of these ideas. And (b) these ideas
are prior to, and independent of, the material things they represent. Let us
consider features (a) and (b) in turn.
In several places, Descartes talks of ideas of mathematics as having a content
that hasnt been put together by the intellect and, hence, is real to the extent that
it is mind independent (CSM II, 45 and CSM II, 8386). These are the passages
that provide the strongest support for the platonist interpretation of Descartes.
However, it is possible to read the text in an alternative way. By saying that
ideas of mathematics are ideas of true and immutable essences that are not
dependent on the mind, Descartes is saying that ideas of mathematics are
ideas whose representational contents include properties such that if they were
removed the idea would cease to be what it is. For example, the representational
content of the idea of triangle implies true and immutable properties such
as having three angles, having the sum of the interior angles equal to two right
angles and so on.
It is easy to mistake the features of these ideas for the features of abstract
objects that these ideas would stand for. It is the immutability of the properties of the representational content of ideas through which we conceptualize
objects that suggests realism about essences and (given Descartes mechanism
and anti-aristotelianism) a platonist kind of realism according to which there
are abstract entities that the mind simply apprehends.
Certainly platonism would explain the mind independence of the content
of these ideas, but inferring platonism from mind independence is hardly the
only possible explanation. The mind independence and realism of the content of these ideas can also be explained by the minds inability to conceptualize objects in ways other than the ways in which it does. In conclusion,
the mistake consists in passing o the immutable character of the represen-

174

Raaella De Rosa and Otvio Bueno

tational content through which our mind conceptualizes things for abstract
objects of which sensible things are mere copies and which the mind simply
apprehends.16
It is also undeniable that Descartes argues in several places that ideas of
geometrical gures, for example, are in the mind prior to, and independently of, the actual existence of the objects they represent. Besides the passages
from Meditation Five quoted above, there are other places where Descartes
maintains that we would have the idea of triangle even if no particular triangle ever existed. Gassendi raised the question as to whether Descartes
thought that one could have the idea of triangle (CSM II, 223), of God
and the self (CSM II, 216) if one never had any senses. In reply, Descartes
writes:17
I do not doubt that the mind [] would have had exactly the same ideas of
God and itself that it now has, with the sole dierence that they would have
been much purer and clearer. The senses often impede the mind in many of its
operations, and in no case do they help in the perception of ideas. The only thing
that prevents all of us noticing equally well that we have these ideas is that
we are too occupied with perceiving the images of corporeal things. (CSM
II, 258, emphasis added.)

According to Descartes, the senses play either no or, at best, an obstructive


role in our having the ideas of God, the self and the triangle. We would have
these ideas even if we had never had any senses. This is established by the fact
that we have clear and distinct ideas of things of which either no instances
can be found in the material world (such as the idea of God and the self ) or
16 For a similar reasoning see (Nolan 1997, 18 1184). Nolans argument is as follows. When
Descartes talks about immutable essences that do not depend on the mind he only means to
say that the ideas of geometrical objects being innate, as opposed to invented, ideas do
not depend on the mind causally. Rather, they causally depend on God. But denying that
these ideas are causally dependent on our mind does not imply denial of their ontological dependence on the mind. Conceptualism is then vindicated. Despite some similarity,
our argument is dierent. We agree with Nolan that the mind-independence of ideas of
mathematical essences does not justify attributing to Descartes a platonist kind of realism
according to which there is a realm of abstract entities that are distinct from God, nite
minds and things. But we are not so sure that this reasoning proves conceptualism. We want
to say that, according to Descartes, the content of these ideas is mind-independent because
we cannot manipulate it. And we cannot manipulate it because these ideas are given to us
by God. But if God gives us these ideas, they must also be in Gods mind. Needless to say,
this conclusion threatens Nolans conceptualist interpretation.
17 Although Descartes answer addresses only the ideas of God and the soul, what he says here
can be easily extended to the idea of triangle too.

Descartes on Mathematical Essences

no instances need to be found in the material world (such as the ideas of the
triangle and the chiliagon).
Moreover, in the same set of replies to Gassendi, Descartes makes the stronger point that the idea of triangle cannot depend on the existence of particular
triangles because it is the precondition for recognizing particular shapes in
space as triangles in the rst place (see CSM II, 26 1262).
This feature of geometrical ideas, viz., the fact that they exist in the mind
innately as the precondition for recognizing shapes in space can be, again,
easily conated with the platonist view that there is a world of abstract entities that exist prior to, and independently of, particular and sensible things.
Descartes claim that ideas of geometrical gures are prior to, and independent
of, particular things together with Descartes talk of these ideas as ideas of
essences engenders the conclusion that Cartesian essences are prior to, and
independent of particular things (in the same way in which the platonists
abstract entities are). However, this reasoning is sound only on pain of ignoring Descartes claim that essences are nothing but our ways of thinking
of things in abstraction from them. And if essences are nothing but ways of
conceiving of things, then their being prior to, and independent of, particular
things are properties belonging to the ideas alone and cannot transitively be
attributed to essences. The mistake consists in passing o the innate character of the ideas for the ontological status of the essences they would stand
for.
This concludes the presentation of our interpretation of Descartes view on
mathematical essences. But why should we call this view quasi-platonism?
The label seems to be misleading since platonism suggests the postulation of
abstract objects and weve argued that Descartes apparent belief in abstract
entities is only the result of the features of general ideas. After all, one may
argue, Descartes view on essences is conceptualist to the extent that he grants
them a being of reason and the features that he attributes to universal ideas
(nativism and realism) dont suce to make him a platonist of any sort. However, since platonism about mathematical essences is indeed associated with
realism and nativism (more commonly than conceptualism maybe), we believe
that quasi-platonism is an appropriate way to describe Descartes position
on mathematical essences. The prex quasi indicates the non-genuine character of his platonism; but the term platonism (better than conceptualism)
brings about the associations to the properties of realism and being prior to/
independent of particular things that Descartes attributed to innate ideas.
Moreover, the term quasi-platonism better captures Descartes belief that

176

Raaella De Rosa and Otvio Bueno

these innate ideas are God-given and, hence, must also be present, in some
sense, in Gods mind.18
Of course, how one refers to Descartes position is, ultimately, only a terminological issue. As long as its clear what position he held, no misunderstanding
should emerge. Descartes was neither a platonist nor a conceptualist, but he
developed an interesting alternative between these two extremes. And thats
what the term quasi-platonism is meant to convey.

4. Benets of Descartes Quasi-Platonism and the Eternity of


Mathematical Truths
In this section, we will discuss some of the benets Descartes view on essences
has over platonism and Gassendis version of conceptualism, and how Descartes
quasi-platonism is compatible with the eternity of mathematicaltruths.
Descartes view fares better than platonism in explaining the possibility of
mathematical knowledge for several reasons. First, the problem of explaining
mathematical knowledge is simply dissolved, since the objects of mathematics are ideas, or ways of thinking of the physical world.19 As Descartes writes
to Mersenne, there is no single truth that the mind cannot understand if it
turns to consider it, because these truths are inborn in our minds (CSMK,
23).
Secondly, Descartes provides a method/mechanism (namely, the skeptical
method of doubt) through which the mind can turn to consider the ideas and
truths of mathematics inborn in our minds. Thirdly, assuming that the method
of doubt can isolate the class of genuinely inborn, clear and distinct ideas, these
ideas are reliable indicators of truth given the non-deceiving nature of God.
So, for example, in Meditation Five and First Set of Replies, Descartes claims
that every property we clearly and distinctly understand to belong to the true
and immutable essence of a triangle really does in fact belong to it (CSM II,
45). This is obtained, in particular, by drawing conclusions from the content
of mathematical ideas. Given the ideas of natural number, one cannot deny
that 2 + 2 = 4 without facing a contradiction (CSM II, 25). Or we can draw
18 This is precisely the reason why Marleen Rozemond, in (Rozemond 2008), calls Descartes
view neo-platonist and argues, contra (Nolan 1997), that Cartesian essences must be something more than ideas in the minds of human beings.
19 This point is eshed out, in a more contemporary context, in (Bueno 2005).

Descartes on Mathematical Essences

out from the idea of the triangle that its three angles equal two right angles
(CSMK, 184).
Also Descartes view fares better than Gassendis conceptualism in explaining
the usefulness of mathematics for scientic theories. If geometrical objects,
according to Gassendi, are nothing but modes of thinking of things that do not
exactly correspond to them, it is unclear why they describe so successfully the
physical world. But, according to Descartes, (i) what particular triangles (for
example) have in common is a function of the idea of triangle; and therefore
any property we attribute to triangles is grounded in the idea and belongs
necessarily to them. And (ii) given the existence and non-deceiving nature of
God, material things possess all the properties which I clearly and distinctly
understand, that is, all those which viewed in general terms are comprised
within the subject of pure mathematics (CSM II, 55). So, according to Descartes, the language of mathematics describes successfully the physical world
because it captures its true nature.
Moreover, Descartes quasi-platonism explains the necessity of mathematical
statements better than Gassendis conceptualism. Since, according to Gassendi,
the truth of mathematical statements is grounded in the observation of things
such statements are only empirical generalizations. According to Descartes,
instead, mathematical truths are grounded on ideas. And since any property we
attribute to mathematical objects is grounded on the idea of these objects and
belongs necessarily to them, mathematical truths are necessary truths.20
However, one may object that even if Descartes can explain the necessity of
mathematical truths better than Gassendi, he (like Gassendi) cannot explain
the eternity and immutability of mathematical truths. After all, mathematical
essences are mental entities and the latter are neither eternal nor immutable.
Descartes has at least two possible explanations of the eternity and immutability of mathematical truths. First, the eternity of mathematical truths may
be read as a function of the immutability of the content of the innate ideas
through which we conceptualize objects. Given that the content of these ideas
contains the description of the essential properties of things, and these ideas
are God-given and not subject to our manipulation, essences are said to be
eternal or immutable.21
20 According to us, the claim that mathematical truth is a matter of conceptual entailment
among ideas is compatible with the Cartesian view that eternal truths ultimately depend on
Gods will. After all, God imprinted mathematical ideas, among others, in our minds.
21 This view has been defended by (Bennett 1994) and (Chappell 1997). See also (Wilson 1975,
120131).

178

Raaella De Rosa and Otvio Bueno

Secondly, if eternity is to be taken literally rather than as a synonym of


immutability, we may say that mathematical essences qua ideas are eternal
because they depend on the will and understanding of God (see CSM II, 261).
This second alternative makes a virtue of what is generally taken to be a vice of
Descartes account of the eternity of mathematical truths. It is a common objection that Descartes doctrine of the creation of eternal truths together with
his voluntarism is inconsistent with their eternity.22 But if what grounds the
truth of mathematical statements is Gods will and Gods will is unchangeable,
there is no room for any change in the truth-values of mathematical statements.
As a result, mathematical truths are indeed eternal.23
Note that we shouldnt interpret Descartes voluntarism as the view that it is
possible for God to will that eternal truths be otherwise, for two reasons. First,
the claim that it is possible for God to will that eternal truths be otherwise
presupposes (at least under one reading of it) that there are possibilities prior
to Gods willing them to be so, and that God decided to will one rather than
the others. But Descartes claims that it is self-contradictory to suppose that
the will of God was not indierent from eternity with respect to everything
which has happened or will ever happen; for it is impossible to imagine that
anything is thought of in the divine intellect as good or true [] prior to the
decision of the divine will to make it so (CSM II, 291). Descartes further explains that if some reason for somethings being good has existed prior to its
preordination, this would have determined God to prefer those things which
it was best to do (CSM II, 294).24 In conclusion, since divine indierence is
the precondition of Gods freedom, the scope of whats possible is determined
by what God actually wills.
Still, one may insist, the freedom of Gods will is certainly compatible with
a change in Gods will. This is certainly true in principle. However, and this
is our second reason, in a letter to Mersenne, Descartes replies to the objection
that God could change the eternal truths: Yes, he can if his will can change
(CSMK, 23). However, in many places Descartes rmly denies that Gods will
can change (see, e.g., CSM II, 281 and CSM II, 294). Thus, taking the if
above as an only if , the eternal truths cant change either.
Notice that the move above namely our grounding the eternity and immutability of mathematical truths on Gods will and understanding brings
out another reason why we believe Descartes is a quasi-platonist rather than
22 See, for example, CSM II, 281 and CSM II, 294.
23 See, for example, CSM I, 93, CSM I, 211, and CSM I, 240.
24 For a detailed defense of this view, see (Kaufman 2002).

Descartes on Mathematical Essences

a full-edged conceptualist. Conceptualist readings, like (Nolan 1997) and


(Chappell 1997), maintain that mathematical essences are primarily ideas in
human minds. However, we believe that this reading doesnt do full justice to
Descartes view that these essences are created by God. If God created these
essences as ideas in human minds, they must also exist in God, either as divine
decrees, as (Schmaltz 1991) suggests, or as objective beings in Gods mind, as
(Rozemond 2008) claims. So, ultimately, even if essences are ideas in human
minds, they are also ontologically dependent on God, and in this respect,
Descartes position is quasi-platonist.

5. The Textual Inconsistency Dissolved


Let us now show that Meditation Five is consistent with the Principles in light
of Descartes quasi-platonism and dispositional nativism.
Undeniably, some passages in Meditation Five seem to indicate that T&IEs
are platonic essences subsisting in an extra-mental and extra-physical realm,
since Descartes suggests that they are prior to, and independent of, material
things and human minds. However, in the Fifth Replies, Descartes emphasizes
that the idea of the triangle and the like do not come from the senses. In light of
this point, we can legitimately interpret Descartes claim in Meditation Five
that T&IEs are independent of the existence of particular things and nite
minds as simply making the negative point that the content of mathematical
ideas is not a construct of the mind based on the observation of things.25 The
innateness of the idea of triangle explains why Descartes says that this is the
idea of an object that has a T&IE.
The innateness of these ideas explains, moreover, why we have the idea of
the triangle even if no triangle ever existed or was ever observed by us. Again,
the claim in Meditation Five that we have these ideas even if no particular object
existed doesnt necessarily commit Descartes to platonism. The claim that the
idea of triangle is the precondition for perceiving things in space as triangles
is mistaken for the claim that the idea of triangle stands for an abstract object
existing prior to particular traingles.
Needless to say, the view that Cartesian essences are nothing but the innate ideas implicitly guiding our representation of things is compatible with
25 For a similar point, see (Schmaltz 1991).

180

Raaella De Rosa and Otvio Bueno

the view of the Principles according to which triangularity is nothing but a


mode of thought. But the emphasis on the innateness of the modes of thought
explains the sense in which we should interpret the conceptualism suggested
by the Principles. By claiming that essences are modes of thought, Descartes is
making the negative point that they are neither platonic nor aristotelian forms,
rather than the positive point that essences are constructs of the nite mind
based on the observation of things. And the innateness of the modes of thoughts
prevents the inference from the negative point to the positive point. Thats why
Descartes brand of conceptualism is dierent from Gassendis.
In conclusion, the innateness of ideas of mathematics provides a coherent
interpretation of both the Principles and Meditation Five by attributing to
Descartes a hybrid view that isnt identical with either platonism or conceptualism.

6. Conclusion
Our analysis of the ontological status of T&IEs established that they are not,
contrary to what some passages seem to suggest, entities subsisting in an abstract third realm. Rather they are innate or God-given modes of thinking
of things. T&IEs are engraved in the mind and, unknowingly to us, guide
our thoughts rather than being the object of these thoughts.
As a result, Descartes is neither a platonist nor a conceptualist about mathematical essences. His dispositional nativism allows him to hold a hybrid view,
quasi-platonism, that has the benets of solving the textual inconsistency without falling into the troubles of either platonism or conceptualism.26

References
CSM: John Cottingham, Robert Stootho and Dugald Murdoch (eds.), The Philosophical Writings of Descartes, volumes I-II. Cambridge: Cambridge University
Press, 198591.
26 We wish to thank Martha Bolton, John Carriero, Dan Garber, Paul Lodge, Antonia Lolordo,
Steven Nadler, Alan Nelson, Larry Nolan, Andrew Pessin, Marleen Rozemond, and Tad
Schmaltz for helpful comments on earlier versions of this work.

Descartes on Mathematical Essences

CSMK: John Cottingham, Robert Stootho, Dugald Murdoch and Anthony Kenny (ed.), The Philosophical Writings of Descartes. The Correspondence, volume III.
Cambridge: Cambridge University Press, 1991.
Bennett, J. 1994. Descartess Theory of Modality. The Philosophical Review 103:
639667.
Bolton, M. 1998. Universals, Essences, and Abstract Entities. In The Cambridge History
of the Seventeenth Century Philosophy, edited by D. Garber. Cambridge: Cambridge
University Press.
Bueno, O. 2005. Dirac and the Dispensability of Mathematics, Studies in History and
Philosophy of Modern Physics 36: 465490.
Chappell, V. 1997. Descartes Ontology. Topoi 16: 11 1127.
Curley, E.M. (1978): Descartes Against the Skeptics. Cambridge: Cambridge University
Press.
Descartes, R. 1975. Entretien avec Burman. Paris: Librairie Philosophique J. Vrin.
De Rosa, R. 2000. On Fodors Claim that Classical Empiricists and Rationalists Agree
on the Innateness of Ideas. Proto Sociology 14: 240269.
De Rosa, R. 2004. The Question-Begging Status of the Anti-Nativist Arguments. Philosophy and Phenomenological Research 69: 3764.
Doney, W. 1993.On Descartes Reply to Caterus. American Catholic Philosophical Quarterly 67: 413430.
Edelberg, W. 1990. The Fifth Meditation. The Philosophical Review 99: 493533.
Gewirth, A. 1970. The Cartesian Circle Reconsidered. Journal of Philosophy 67:
668685.
Gewirth, A. 1971. Descartes: Two Disputed Questions. Journal of Philosophy 68:
288296.
Gueroult, M. 1984. Descartes Philosophy Interpreted According to the Order of Reasons.
Minneapolis: University of Minnesota Press.
Kaufman, D. 2002. Descartess Creation Doctrine and Modality. Australasian Journal
of Philosophy 80: 2441.
Kenny, A. 1968. Descartes. New York: Random House.
Kenny, A. 1970. The Cartesian Circle and the Eternal Truths. Journal of Philosophy 67:
685700.
Nolan, L. 1997. The Ontological Status of Cartesian Natures. Pacic Philosophical
Quarterly 78: 169194.
Nolan, L. 1998. Descartes Theory of Universals. Philosophical Studies 89: 16 1180.
Rozemond, M. 2008. Descartes Ontology of the Eternal Truths. Forthcoming in Contemporary Perspectives on Early Modern Philosophy: Essays in Honor of Vere Chappell,
edited by P. Homan, D. Owen and G. Yae. Broadview Press.
Schmaltz, T. M. 1991. Platonism and Descartes View of Immutable Essences. Archiv
fr Geschichte Der Philosophie 73: 129170.
Wilson, M. 1978. Descartes. London: Routledge.

182

Raaella De Rosa and Otvio Bueno

Contributors

Editors and Contributors


Editors
Prof. Dr. phil. Gerhard Preyer, Professor of Sociology. Goethe-University
Frankfurt am Main, Farnkfurt am Main, Germany.
Dr. phil. Georg Peter, Project Protosociology. Goethe-University Frankfurt am
Main, Frankfurt am Main, Germany.

Contributors
Prof. Dr. Andrew Arana, Department of Philosophy, 201 Dickens Hall, Manhattan, USA.
Prof. Dr. Jody Azzouni, Department of Philosophy, Miner Hall, Tufts University, Medford MA 02155.
Prof. Dr. Otvio Bueno, Department of Philosophy, University of Miami,
Coral Gables, USA.
Prof. Dr. Mark Colyvan, Professor of Philosophy, School of Philosophical and
Historical Inquiry Director of the Sydney Centre for the Foundations of Science, Chief Investigator and Project Leader, Australian Centre of Excellence
for Risk Analysis, Chief Investigator, Research Hub for Applied Environmental
Decision Analysis, University of Sydney, Sydney, Australia.
Prof. Dr. Willy Essler, Institut fr Philosophie, Goethe-Universitt Frankfurt
am Main, Frankfurt am Main, Germany.
Prof. Dr. Marcel Fredericks, Ph.D., Department of Sociology, Loyola University, Chicago, USA.
Dr. J. Gregory Keller, Ph.D., Department of Philosophy, Indiana University
Purdue, University Indianapolis, 425 University Blvd., Indianapolis, USA.

184

Editors and Contributors

Dr. Nikola Kompa, Philosophisches Seminar, Westflische Wilhelms-Universitt, Domplatz 23, 48143 Mnster, Germany.
Dr. Steven I. Miller, School of Education/Department of Philosophy, Loyola
University Chicago, USA.
Prof. Dr. Madeline Muntersbjorn, Department of Philosophy, University of
Toledo, Ohio, USA.
Prof. Dr. Douglas Patterson, Department of Philosophy, Kansas State University, 201 Dickens Hall, Manhattan, USA.
Dr. Frank J. Perino, Associate Professor, College of Education, Northeastern
Illinois University, USA.
Prof. Dr. Raaella De Rosa, Department of Philosophy, Rutgers University,
Newark, USA.
Prof. Dr. Yvonne Raley, Associate Dean, Arts and Sciences, Philosophy Department, Felician College, 262 South Main Street, Lodi, USA.
Prof. Dr. Nicholas Rescher, Department of Philosophy, University of Pittsburgh, 1012 Cathedral of Learning Pittsburgh, USA.
Prof. Dr. Adam Sennet, UC Davis, 1238 Social Science and Humanities Bldg.,
Davis CA, USA.
Prof. Dr. Susan Vineberg, Department of Philosophy, Wayne State University,
Detroit, USA.