Professional Documents
Culture Documents
)
Philosophy of Mathematics
LOGOS
Studien zur Logik, Sprachphilosophie und Metaphysik
Philosophy of Mathematics
Set Theory, Measuring Theories,
and Nominalism
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
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
Part I
Set Theory, Inconsistency, and
Measuring Theories
10
Douglas Patterson
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-
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
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
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
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.
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).
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
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
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.
26
Douglas Patterson
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.
28
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.
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.
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
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
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
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
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.
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.
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
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.
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.
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
44
Andrew Arana
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
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.
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.
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
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
52
Andrew Arana
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
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-
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
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.
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
60
Wilhelm K. Essler
62
Wilhelm K. Essler
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.
64
Wilhelm K. Essler
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
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).
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
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
70
Wilhelm K. Essler
Part II
The Challenge of Nominalism
72
Jody Azzouni
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
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.
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
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.
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.
84
Jody Azzouni
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.
86
Jody Azzouni
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.
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.
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.
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
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).
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.
96
Otvio Bueno
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.
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.
98
Otvio Bueno
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]).
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]).
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.
102
Otvio Bueno