The Identity Problem for Realist Structuralism^

JUKKA KERANEN*

Recent decades have seen the emergence of a new strand of philosophy of

mathematics known as structuralism. A central aim of the structuralist

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

project is to furnish an account of ontology that elaborates the idea that
mathematical objects are individuated solely by the relations they have to
one another. The basic realist elaboration of this idea is that mathemati-
cal objects are places in relational structures. In this paper, we will argue
that in spite of its many attractions, realist structuralism must ultimately
be rejected. On the one hand, we will show that mathematical structures
typically contain places that are indiscernible if individuated solely by the
relations they have to one another. On the other hand, we will argue that
any account of place-identity available to the realist structuralist entails
that relationally indiscernible places are identical. Since she maintains that
mathematical singular terms denote places in structures, the realist struc-
turalist is therefore committed to saying that, for example, 1 = —1 in the
group of integers (Z, +). We will call this predicament the identity problem.
We will argue that the identity problem vitiates the realist structuralist ac-
count of ontology beyond repair and conclude that nominalism is the only
potentially viable version of structuralism.

1. Structuralism about Mathematics

Philosophical structuralism about mathematics has two basic motivations,
both of which can be found in Benacerraf. On the ontological front, struc-
turalism can be seen as a response to the problem of 'multiple reductions'
raised in 'What numbers could not be' [1965]. On the epistemological
front, structuralism can be seen as a response to the well-known dilemma

T Previous versions of this paper have been presented at the Annual Meeting of the
Association for Symbolic Logic in San Diego, 1999; the 11 t h International Congress of
Logic, Methodology and Philosophy of Science in Krakow, Poland, 1999; and elsewhere.
Special thanks are due to Lisa Belicka, Ken Manders, and John McDowell for their
invaluable assistance. I would also like to thank Charles Parsons, Michael Resnik, and
Stewart Shapiro for stimulating conversations about the issues addressed herein.
* Department of Philosophy, 1001 Cathedral of Learning, University of Pittsburgh,
Pittsburgh, Pennsylvania 15260 U. S. A. keranen+Opitt.edu

PHILOSOPHIA MATHEMATICA (3) Vol. 9 (2001), pp. 308-330.

 IDENTITY PROBLEM 309

in 'Mathematical truth' [1973]. We begin by briefly revisiting these two

fountainheads of structuralism.
The central argument of 'What numbers could not be' is aptly sum-
marized by Benacerraf himself: 'If numbers are sets, then they must be
particular sets, for each set is some particular set. But if the number 3 is
really one set rather than another, it must be possible to give some cogent
reason for thinking so' (Benacerraf [1965] in Benacerraf and Putnam [1983],
p. 284). However, '[t]here is no way connected with the reference of number
words that will allow us to choose among [the candidate reductions of num-
bers to sets], for the accounts differ at places where there is no connection
whatever between the features of the accounts and our uses of the words in

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

question' (op. tit., p. 285). Hence, concludes Benacerraf, numbers are not
sets. He then extends the argument to the conclusion that numbers are not
objects at all. His reasoning turns on the idea that 'in giving the properties
[... ] of numbers you merely characterize an abstract structure...' (ibid.).
Thus, according to Benacerraf, if numbers are indeed objects, they must be
identified with some particular system that has the appropriate structure.
But, again, numerous systems have this structure.1 Since numbers 'have no
properties other than those relating them to other "elements" of the same
structure' (ibid.), the choice between the candidate systems would have
to be based on such intra-structural, relational properties. However, given
that all the candidate systems are identical with respect to these properties,
we have no more reason to identify the numbers with any one particular
system than with any other. In particular, numbers are not objects at all
because 'there is no more reason to identify any individual number with
any one particular object than with any other' (op. cit., p. 291).
In sum, numbers are individuated by their intra-structural relational
properties; talk of individual numbers without reference to the number
structure is nonsense. This is a powerful insight. Indeed, this is where
structuralism begins. What remains less clear, however, is why we could
not nevertheless treat numbers as objects. Since all the candidate systems
have the same relational 'structure', why could we not treat that struc-
ture itself as the number system, its constituent places as the numbers?2
Benacerraf did not explore this possibility. He thought that in order for
the places of a given structure to count as bona fide objects, 'it must be
possible to individuate [them] independently of the role they play in that
structure' (op. tit., p. 291). Benacerraf gave no justification for taking

1
The von Neumann and Zermelo ordinals are the standard examples.
2
By a 'structure' we will understand an abstract object consisting of a collection D of
'places' and a collection {R\} of relations on those places. Thus, formally, a structure
is an ordered pair of the form (D, {R\ '• A 6 o}), where D is a non-empty collection of
places, a a cardinal number, and {R\ : A € a} a family of relations on D; A is the arity
of the relation R\. See Weaver (1998).
310 KERANEN

this line of thought, and, indeed, many philosophers of mathematics have

subsequently rejected it. Accordingly, the idea that numbers are individ-
uated by their intra-structural relational properties has been developed in
two directions: nominalist and realist. The nominalist structuralists (such
as Hellman) have followed Benacerraf and maintained that no particular
system deserves the honorific 'numbers'. More generally, there is no specif-
ically 'mathematical' ontology: a mathematical theory is about any system
of objects, abstract or concrete, that satisfies its axioms. Thus, mathemat-
ical statements are not to be taken at face value: apparent singular terms,
such as numerals, are really disguised bound variables ranging over ele-
ments in systems. The realist structuralists (such as Resnik and Shapiro),

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

in contrast, have maintained that there is a specifically mathematical on-
tology that consists of places in abstract structures. Thus, mathematical
statements are to be taken at face value: apparent singular terms denote
places in structures and as such, are properly referential.
Many realist structuralists have suggested that their view offers a way of
resolving Benacerraf's well-known epistemological dilemma in 'Mathemat-
ical truth' [1973]: If mathematical objects are abstract entities and hence
cannot be causally interacted with, how can we fit mathematical knowl-
edge into a broadly naturalistic account of epistemology? For the realist
structuralist, epistemology of mathematics begins with the idea that since
mathematical objects belong to structured patterns, it is our pattern cog-
nition that enables us to acquire knowledge of mathematical objects. The
initial epistemic access to mathematical structures is provided by causal
contact with concrete patterns from which the structures themselves are
'abstracted'; after this, the details vary considerably. For the nominalist
structuralist, the story begins in a similar fashion. In her case, however,
the epistemic access is to various systems rather than to structures as such
since, strictly speaking, structures do not exist. Again, the further details
vary, and as Hale [1996] has pointed out, difficulties abound.
We now have two basic distinctions in view: the ontological and the epis-
temological motivations of structuralism on the one hand, and the nominal-
ist and the realist variants of structuralism on the other. In the remainder
of this paper, we will explore the ontological and semantic issues the iden-
tity problem raises for the realist structuralist.3 We will maintain that the
identity problem vitiates realist structuralism beyond repair and conclude
that nominalist structuralism is the only potentially viable elaboration of
the Benacerrafian intuitions.

3
We should note that while our considerations are often framed in semantic terms, the
core issue of this paper is meant to be genuinely ontological (if you do not believe that
there is such a distinction to be drawn, you are of course welcome to ignore this remark).
 IDENTITY PROBLEM 311

2. Realist Structuralism
Our strategy will be to argue that the realist structuralist cannot solve the
identity problem without compromising the very motivation of her theory.
Thus, we should now indicate why a structuralist might want to be a realist
about ontology. Since Shapiro's Philosophy of Mathematics [1997] is the
most recent and perhaps the most clearly articulated account of realist
structuralism to date, we will focus on the motivations of realism as they
emerge in Shapiro.
The most important motivation for Shapiro's realism is his desire to
adopt semantic anti-revisionism: we should, whenever possible, take math-

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

ematical statements at face value and, in particular, regard mathematical
singular terms as properly referential.4 As we noted above, realism about
ontology provides a natural setting for such anti-revisionism which, in turn,
can be motivated as follows. First, we have the familiar theme documented
by Benacerraf in his 'Mathematical truth'. Assuming that we can adopt
the Tarskian referential account of truth for the language of mathematics as
well as for the language of empirical science, we can furnish a uniform, ac-
count of truth for mathematics and science (Shapiro [1997], p. 13). Second,
like many realists, Shapiro thinks that philosophy of mathematics should
be a descriptive-explicatory or, as he puts it, an 'interpretative' enterprise
rather than a normative one (ibid., p. 51). Respecting the actual practices
of working mathematicians, including what they take the semantic struc-
ture of their assertions to be, is therefore inherently attractive to Shapiro.
Finally, building on his semantic anti-revisionism, Shapiro offers the fol-
lowing further motivation for his realism. The hallmark of mathematical
structures vis-a-vis the non-mathematical ones is that they are 'freestand-
ing': mathematical structures can be 'exemplified by objects beyond those
of the original ontology' of the theory in question (ibid., p. 126). Now, in
the nominalist accounts, the ontology of structures reduces to the ontol-
ogy of the various systems. The realist accounts, by contrast, allow us to
consider structures as entities on their own right and hence independently
of any particular system. Thus, maintains Shapiro, realist structuralism
provides a better framework for bringing out the freestanding character of
mathematical structures (ibid., p. 100).5

4
Anti-revisionism as such is one of the overarching themes of Shapiro's book and his
espousal of realism reflects it in a number of ways (see Shapiro [1997], p. 30).
5
The Quine-Putnam 'indispensability argument' is often viewed as the best, possibly
even the only serious motivation for adopting realism about ontology (see, for example,
Colyvan [1998]). Indeed, now that the dust raised by Field [1980] and Chihara [1973|
has settled, the indispensability argument appears to be experiencing something of a
second coming. It is then worth noting that Shapiro, for one, does not think that the
indispensability argument all by itself provides a sufficient motivation for realism (see
Shapiro |1997), p. 46).
312 KBRANEN

3. Accounts of Identity
In this section we will argue that the realist structuralist must include in
her theory an 'account of identity' for the places of any given structure. To
facilitate our argument, we will first explain what an account of identity is
meant to accomplish and then proceed to distinguish two ways of furnishing
such an account.
Traditionally, what there is falls into two categories: objects and entities.
Objects, unlike entities, have determinate 'identity conditions': Given two
singular terms 'a' and '&' that denote objects of a given kind, there is
a definite fact as to whether or not the identity statement 'a = 6' is true

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

(think apples). By contrast, given two singular terms 'a' and 'b' that denote
entities, there need be no definite fact as to whether or not 'a = b' is
true (think waves). Thus, one might call objects 'properly individuated'
entities. In what follows, we shall assume that the realist wants to construe
mathematical objects as objects in this sense: there is a definite fact as to
whether two, say, numerals denote the same number.6
Suppose, then, that L is an interpreted language and 'a' and 'b' two
singular terms in L that denote objects. What should an account of iden-
tity for the domain of discourse of L accomplish? There are two basic
requirements. First, given what we have just said, an account of identity
for the domain of discourse of L should specify the circumstances under
which 'a = 6' is true in L by specifying what sort of facts govern the iden-
tity of the objects referred to in L. In other words, an account of identity
should specify the circumstances under which 'a' and '6' denote the same
object. Second, although the first requirement is phrased in terms of sin-
gular terms, an account of identity for the domain of discourse of L should
obviously apply to all objects in that domain, not just to the ones with
singular terms denoting them.7
Given a theory T and a language L, how might one formulate an account
of identity in T for the domain of discourse of IP. In view of the two
requirements above, the following very naturally suggests itself: given two
quantifiers ' W and 'Vj/1 ranging over the domain of discourse of L, just
complete the blank in the 'identity schema'

(IS) VxVy (i = 2/<=> ).

Note that the resulting account of identity satisfies the two requirements
above. The first requirement is satisfied because for any two singular terms
6
We will not take up the issue whether there really are entities that are not objects.
See, for example, Lowe [1995].
7
To fix our philosophical grammar, an account of identity is an account in some
particular philosophical theory T for the domain of discourse of some particular language
L.
 IDENTITY PROBLEM 313

'a' and 'b' in L, there is a substitution instance (a = b •& ) of (IS) and so,
when suitably completed, (IS) will specify the circumstances under which
'a' and 'b' denote the same object. The second requirement is satisfied
because (IS) applies to all objects in the domain of discourse of L, not just
to the ones with singular terms denoting them.
We will now distinguish two ways in which a philosophical theory might
have us complete the blank in (IS) and thus, two ways of furnishing an
account of identity. To the best of our knowledge, all extant theories of on-
tology maintain that the identity of objects is governed by their properties.
To put it in the most general terms possible, 'a = b' is true if and only if for
all properties <j> in some class <£, <p(a) if and only if <p(b). Depending on wha

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

sort of properties the theory T takes the class <fr to include, the account
of identity it gives will be either a general-property account or a haecceity
account. A general property is a property that can be possessed by more
than one entity, whereas a haecceity is a property that can be possessed by
one entity alone. For example, 'being yellow' is a general property, while
'being identical to a' is not; indeed, 'the haecceity of a' is often cashed out
in this way.8 Thus, given a language L, a general-property account says
that the identity of the objects referred to in L is governed by their general
properties. Hence, the blank in (IS) should be completed with some expres-
sion logically equivalent to V ' </>(</> € $ => (<fr(x) •&• </>(y)))\ where 'V0' ranges
over the properties countenanced by L; $ is the class of general properties
that according to T governs the identity of the objects referred to in L.9
Similarly, given a language L, a haecceity account says that the identity
of the objects referred to in L is governed by their haecceities. Hence, the
only way to complete the blank in (IS) is 'Hx — Hy\ There just is nothing
more informative to say. General-property accounts have been advocated,
among others, by Leibniz, Russell, and Ayer. Haecceity accounts are found
primarily in medieval philosophy, famously in Duns Scotus, but also in Kant
and Peirce. For the present purposes, we shall assume that there are no
other plausible candidates for an account of identity.
Now that we have explained what an account of identity is meant to
accomplish, it should seem clear that the realist structuralist must furnish
some such account for the places of any given structure. For, the realist
has set out to provide a primitive, self-contained theory of mathematical
objects qua places in structures. While the nominalist structuralist takes
mathematical ontology to be reducible to, say, set theory or modal model
theory,10 the realist structuralist takes mathematical ontology to be irre-
ducibly structural. For example, for the 'set-theoretic' nominalist struc-
8
See Adams (1979].
9
For example, essenlialist accounts maintain that 'a = 6' is true if and only if a and
b have the same essential properties (whatever those may be).
10
See, for example, Shapiro [1997], pp. 90-92.
314 KERANEN

turalist, mathematical objects are in the first instance sets. In her theory,
the Axiom of Extensionality constitutes an account of identity for mathe-
matical objects qua sets: two singular terms 's' and 'r' denoting sets denote
the same set if and only if the sets they denote have the same members. For
the realist structuralist, by contrast, mathematical objects are in the first
instance places in structures. Hence, she must include in her theory an in-
digenous account of identity for mathematical objects qua places. Indeed,
as Shapiro tells us in a preamble to his theory of structures:
Quine's thesis is that within a given theory, language, or framework, there
should be definite criteria for identity among its objects. There is no reason
for structuralism to be the single exception to this. (Shapiro [1997], p. 92)

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

On the other hand,
each structure has a collection of 'places' and relations on those places... The
theory 'of structures' thus has a second sort of variable that ranges over places
in structures. (Shapiro [1997], p. 93)
But, to be an object is to be a value of a variable, and hence, Shapiro has
certainly committed himself to giving an account of identity for the places
of any given structure. Regrettably, he docs not provide such an account.
Given, then, that the realist structuralist must furnish an account of
identity for places, she must adopt either a general-property account or a
haecceity account. We will now argue that, in fact, the realist structuralist
can adopt neither kind of account. On the one hand, any general-property
account available to the realist structuralist results in the identity prob-
lem. On the other hand, any haecceity account directly conflicts with the
spirit and motivations of realist structuralism. To facilitate our argument,
we will adopt the following conception of the ontological project of realist
structuralism: the realist structuralist begins with a specific system S and
tries to give an account of identity for the structure S of that system.11 It
seems to us that this conception of her project should be entirely acceptable
to the realist structuralist; after all. we are giving her all the set-theoretic
resources she might hope for.12 We are merely asking her to tell us how

11
We need to make three definitions. 1) The language of S is the language L of the
theory of which S is a model, interpreted to be about S. In what follows, we will assume
that all the individual constants in L have unique well-defined referents and that L
contains as many individual constants as necessary. 2) An account of identity for the
structure S is an account of identity for the places of S. 3) The structure of S is a
structure in which each place is individuated in terms of the relations the element in S
occupying it has to the other elements in S. For now, we will remain neutral on whether
several elements in S can occupy the same place of S (see §4).
12
An opponent of realist structuralism might urge that we are being too charitable.
After all, it is not clear whether the realist structuralist is always entitled to a system of
independently identifiable elements, especially if set theory itself is to be brought under
the auspices of structuralism. As we shall see in §4, however, we are not allowing the
realist structuralist to individuate places in terms of (previously individuated) elements
in systems.
 IDENTITY PROBLEM 315

she would complete the blank in the identity schema (IS) for the structure
of 5.

4. The Identity Problem

Assuming now that the realist structuralist wants to give a general-property
account of identity for places, we will argue that she can only employ intra-
structural relational properties in her account. Hence, if 'a' and 'b' denote
places in the same structure, the realist structuralist must view their deno-
tations as identical whenever they have the same intra-structural relational
properties. In many cases, this will force her to endorse identity statements

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

that any sound account of mathematical ontology must maintain are false;
again, we will call this the identity problem.
It seems clear that the realist structuralist can only employ intra-struc-
tural relational properties in her account of identity for places. This follows
directly from the fundamental structuralist intuition we found in Benacer-
raf: 'mathematical objects have no properties other than those relating
them to other "elements" of the same structure' ([1965]; Benacerraf and
Putnam [1983], p. 285). The contemporary realist structuralists, such as
Parsons, Resnik, and Shapiro, seem to agree wholeheartedly. Parsons has it
that' [mathematical objects] have no more to them than can be expressed in
terms of the basic relations of the structure [to which they belong]' (Parsons
[1990], p. 303). Resnik famously tells us that '[as] positions in structures,
[mathematical objects] have no identity or features outside a structure'
(Resnik [1981], p. 530). Finally, Shapiro agrees that 'the essence of a natu-
ral number is its relations to other natural numbers' (Shapiro [1997], p. 72).
Indeed, 'there is no more to the individual numbers "in themselves" than
the relations they bear to each other' (Shapiro [1997], p. 73).
We are now almost ready to formulate a realist structuralist account of
identity for places. We only need to consider where the intra-structural
relations come from. After all, usually when we are presented with a struc-
ture, it is the places that are seen as primary, and the relations are described
in terms of these places (as, say, ordered pairs). Here, however, it is the
relations that are seen as primary, and the places are to be described in
terms of these relations. Recall, then, that we are allowing the realist struc-
turalist to begin with a particular system S the structure S of which we
are asking her to describe. Now, the realist structuralist wants S to be the
'structure of S in the sense that each place of S is individuated in terms
of the relations the element in S occupying it has to the other elements
in S.13 According to Shapiro, for example, 'every [place] is characterized

13
There are typically several distinct systems that all exemplify the same structure and
hence, several distinct elements that occupy the same place. We must ensure that a place
occupied by several elements drawn from distinct systems is unambiguously individuated,
no matter which particular system we start with. See constraint (1) below.
316 KERANEN

completely in terms of how its occupant relates to the occupants of the

other [places] in the structure' (Shapiro [1997], p. 100). Thus, the most
obvious thing to say is that the relations that govern the identity of the
places of S are nothing but the very relations the elements in S have to one
another. In this way, each place of S is indeed individuated by the relations
its occupant has to the occupants of the other places, but, equally, by the
relations it has to the other places themselves.14
Thus, the realist structuralist will complete (IS) as follows. Given a
system S, and given two quantifiers 'Vi' and 'Vy' ranging over the places
of the structure S of S,

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

(STR) VxVy(z = y& V<j>{cj> e $ =*> (</>(*) <=> tffo)))),
where 'V0' is taken to be ranging over the relational properties in S.15
In order to keep in agreement with the spirit and motivations of realist
structuralism, we must be careful about which properties we admit into
the set <fr. We maintain that there are two crucial constraints:
(1) No property the specification of which essentially involves an individual
constant denoting an element in S may be admitted. There are two
initial reasons for imposing this constraint:
a) The relations in S that govern the identity of places are relations in
S also. But, since the elements in S are not identical with the places
of S, having a relation to an element in 5 would not be an intra-
structural relation in S. Hence, relations that can only be specified by
using individual constants denoting elements in 5 are inadmissible. In
particular, the extra-structural relation 'being occupied by a', where a
is an element in S, is certainly inadmissible. The shortcut this property
provides is only available to the nominalist structuralist (see below).
b) The realist structuralist wants to view S as freestanding, independent
of any particular system exemplifying it (see §2). Hence, since an
individual constant that denotes an element in 5 denotes an element
in a particular system, properties that can only be specified by using
individual constants denoting elements in S are inadmissible.
(2) No property the specification of which essentially involves an individual
constant denoting a place of S may be admitted. There is one initial
reason for imposing this constraint:
14
One might wonder where these relations come from if the realist structuralist is not
allowed to begin with some particular system S. Geoffrey Hellman explores this issue in
his 'Three varieties of mathematical structuralism' [2001] and concludes that herein lies
a further reason to suspect that realist structuralism is untenable.
15
Note that in order to make sense of the quantifiers 'Vi' and 'Vj/', we must suppose
that the places of S exist and are properly individuated objects rather than mere entities.
In what follows, we shall make this supposition, but only for the sake of convenience; as
we shall see below, it is not a genuine concession.
 IDENTITY PROBLEM 317

a) The relations in S that govern the identity of places are relations in 5

also. But, since the places of S are not identical with the elements in S,
having a relation to a place of S would not be an infra-systemic relation
in 5. Hence, strictly speaking, relations that can only be specified by
using individual constants denoting places of S are not available in the
first place.
In sum, only the properties that can be specified by formulae in one free
variable and without individual constants may be admitted into the set $.
For example, the property specified by 3y (x + x = y) and the property
specified by x + x = x would be admissible. We take this to be in a good

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

agreement with the spirit of realist structuralism. The realist wants to be
able to say both that (a) one can individuate places without appealing to
elements in any particular system, and yet (b) in some sense places are
individuated by the very relations their occupants have to one another.
Our constraints (1) and (2), respectively, allow these two desiderata to be
simultaneously satisfied.
We are now ready to state the identity problem. Given a structure
S, the schema (STR) says that two singular terms denoting places of S
denote the same place precisely when their referents have the same intra-
structural relational properties that can be specified without using indi-
vidual constants.16 Ontologically speaking, two elements in any system S
exemplifying S occupy the same place of S precisely when they have the
same intra-systemic relational properties that can be specified without us-
ing individual constants. When considered abstractly, this result sounds
rather plausible. The problem is that it does not at all square with the use
of the identity predicate in mathematical discourse. For example, since 1
and —1 in any system (Z, +) have the same intra-systemic relational prop-
erties, the realist structuralist must view ' 1 ' and ' —1' in the language of the
structure (Z, +) as co-referential terms.-Thus, since structures are meant
to be the subject matter of mathematics, and places in structures the ref-
erents of mathematical singular terms, the realist structuralist must adopt
an aggressively revisionist stance towards the surface grammar of mathe-
matical theories. Since 1 = —1 in (Z, +), as far as addition of integers is
concerned, she must maintain that 1 = —1, period. Thus, on the one hand,
the realist account of places fails to underwrite a vast number of mathe-
matical truths and underwrites blatant falsehoods. On the other hand, the
realist account implies that most mathematical theories are inconsistent.
For example, since 1 and —1 are co-referential, we have 0 = 1 + (—1) =
1 + 1 = 2. But, 0 and 2 do not have the same intra-systemic relational
properties and hence, according to (STR), it is not the case that 0 = 2.

16
Incidentally, this entails a (rather strong) form of the 'Leibniz principle' of the identity
of indisceraibles.
318 KERANEN

Thus, the realist structuralist must maintain that any theory that captures
the mathematician's conception of the group (Z, + ) is inconsistent. This,
then, is the identity problem: the realist structuralist's general-property
account of identity for places forces her to endorse identity statements that
any sound account of mathematical ontology must maintain are false.
The foregoing argument depends critically on the claim that only the
properties that can be specified by formulae in one free variable and with-
out individual constants may be admitted into the set $ of individuating
properties. Accordingly, we will now discuss a further reason why this claim
should be accepted. Recall that structuralism is in part motivated by the
familiar observation that mathematical structures can only be identified

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

'up to an isomorphism'. Thus, the places of the structure S must be indi-
viduated by properties that are invariant under the automorphisms of S.17
For, any property that is not invariant in this sense would not figure in
the individuation of S itself and hence, presumably, would not figure in the
individuation of the places of S.18 But, as we will now show, a property
is guaranteed to be invariant under the automorphisms of S if and only if
it can be specified by formulae in one free variable and without individual
constants.
On the one hand, in order to show that all such properties are guar-
anteed to be invariant under the automorphisms of S, we only need to
recall the following elementary model-theoretic fact. If A and B are mod-
els of some given theory T, and if / : A —> B is an isomorphism, a sequence
a\,..., an in A satisfies the formula <j){x\,..., xn) if and only if the sequence
/ ( a i ) , . . . , f(an) in B satisfies <fi(x\,..., xn). In particular, if / : S —» S is
an automorphism, a in S satisfies the formula <j>(x) if and only if /(a) sat-
isfies (j)(x). Thus, a in S has the property specified by <p(x) if and only if
/(a) has that property.
On the other hand, in order to show that only such properties are guar-
anteed to be invariant under the automorphisms of S, we need to show
that each of the following four classes contains some properties that are not
invariant in this sense: (1) extra-structural properties, (2) non-relational

17
If A and B are models of some given theory T with sig(-4) = sig(B), an isomorphism
f: A —» B is a bijective m a p / : dom(A) —* dom(B) such that for any a i , . . . , a n in
dom(/l) 1 we have
(i) for any relation symbol R, RA(a\,... ,cin) if and only if RB(f(a.\),. ••,J(an));
(ii) for any operation symbol +, f(+A{ai, • • • ,dn)) = +fl(/(ai),- ••,/(««)); and
1
(iii) for any constant symbol c, /(c' ) = r .
An automorphism is such a map with A = B. A property R is said to be 'invariant
under' the automorphism / : A —» A when any element a in A has R if and only if its
image /(a) has R.
18
Indeed, some realist structuralists have defined a 'structure' to be an isomorphism
class of systems. Similarly, some nominalist structuralists have explored the idea that
two mathematical systems are identical as structures precisely when they are isomorphic.
See Shapiro [1997J, p. 91.
 IDENTITY PROBLEM 319

properties, (3) intra-structural relational properties the specification of

which essentially involves individual constants denoting places of S, and
(4) intra-structural relational properties the specification of which essen-
tially involves individual constants denoting elements in S. This is quite
easily done. Recall that the group of integers (Z, +) has an automorphism
sending a; to —a;.
(1) Extra-structural properties. The number 1 in (Z, + ) has the extra-
structural property ' being the cardinality of the set of stars in our
solar system' whereas —1 does not have this property.
(2) Non-relational properties. The number 1 in (Z, +) has the non-rela-
tional property 'being identical to 1' whereas —1 does not have this

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

property.19
(3) Intra-structural relational properties the specification of which essen-
tially involves individual constants denoting places of S. The number
1 in (Z, + ) has the property specified by x + x = 2 whereas - 1 does
not have this property.
(4) Intra-structural relational properties the specification of which essen-
tially involves individual constants denoting elements in S. Let (Z, +)
be any system exemplifying the structure (Z, + ) . The number 1 in
(Z, + ) has the property that the element in (Z, +) occupying it has
the property specified by x + x = 2, whereas —1 does not have this
property.
Thus, in all these cases, the property in question is not invariant under the
automorphism of (Z, +) sending x to —x. This proves our claim.
The property specified by x + x = 2 calls for a comment. The model-
theoretic fact quoted above may appear to (absurdly) entail that a in (Z,
+ ) satisfies the formula x + x = 2 if and only if —a in satisfies that for-
mula. Fortunately, there is no absurdity here: the fact quoted above does
not apply to x + x = 2. The automorphism sending a; to —a; is an automor-
phism of the group (Z, +), and the language of the theory of groups is just
{+, —, 0}. Thus, the formula x + x = 2 is not a well-formed formula of the
language of (Z, +). While we granted that the realist structuralist could
have as many individual constants as she needs (see fn. 11), she must em-
ploy them in a way that is consistent with her semantic anti-revisionism. If
she were to take the language of (Z, +) to contain, say, the constant '2', the
map sending x to — x would not be an automorphism of (Z, +): x + x = 2
is not invariant under such a map. Hence, the realist structuralist would
no longer be discussing the structure the mathematician has in mind when
she speaks of the group of integers. Thus, as far as the individuation of
the places of (Z, +) is concerned, we can either point out that the formula
x + x = 2 is not available in the first place, or else point out that the prop-
19
The shortcut this property provides is only available to the haecceilisl (see §6).
320 KERANEN

erty it specifies is not invariant under the automorphisms of (Z, + ) . We

have here chosen the latter route. After all, in a sufficiently rich language,
we can make perfectly good sense of the property specified by x + x = 2.
We just cannot use that property to individuate the places of the group
(Z, + ) = (Z, + , - , 0). 20 •
The examples above are meant to be suggestive of a general predica-
ment: most properties in our four classes are not invariant under the auto-
morphisms of S. As such, the realist structuralist should not be allowed to
use them. At the same time, it is clear that many properties in these four
classes are invariant under the automorphisms of S. As such, other consid-
erations notwithstanding, the realist structuralist should be allowed to use

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

them. Unfortunately for her, such properties can provide but little help in
solving the identity problem. If R is any property that is invariant under
the automorphisms of S, and if / : S —> S is an automorphism, a place a
has the property R if and only if its image f(a) has that property. But, in
many cases where a and b are indiscernible, there is an automorphism of
S mapping a to b (see §5). Hence, in many cases, adding properties that
are invariant under the automorphisms of S to the set $ would provide no
help in distinguishing indiscernible places. This occurs, for example, in the
structure (Z, -f).
The identity problem does not arise for nominalist structuralism. On
the one hand, before she ever begins discussing structures, the nominal-
ist structuralist assumes that some non-structural 'background ontology' is
already in place, and that the objects in the various systems are already
individuated in that ontology.21 On the other hand, for the nominalist
structuralist, structures are nothing but patterns exemplified by systems.
Thus, objects that are discernible in a given system 5 remain distinct when
considered as occupants of places in the structure of S. Hence, insofar as
they are considered to exist at all, places can be viewed as distinct even
when their occupants are 'structurally' indiscernible. The point is that the
nominalist structuralist, unlike the realist, can use ecira-structural prop-
erties for individuating places. For example, the set-theoretic structuralist
can simply say that the '1-place' of (Z, + ) is individuated by the property
'being occupied by the "1-element" in (Z, +)', where (Z, +) is any system
exemplifying the structure (Z, + ) .
The realist structuralist now has three options available to her. First,
she could try to furnish a general-property account of identity that allows
her to regard intra-structurally indiscernible places as distinct. Second, she

20
Note that the fact that the ring (Z, + , X, —, " 1 , 0, 1) has no identity problem, or
indeed non-trivial automorphisms, does not in any way solve the identity problem for
the group (Z, -+-).
21
Shapiro, for example, explores set theory and modal model theory as candidates for
such a background ontology (see Shapiro [1997J, p. 90-92).
 IDENTITY PROBLEM 321

could try to argue that she does not need to give a general-property account
at all. We will consider each of these1 options in §6. Finally, in view of the
difficulties with the realist account of place-identity, the structuralist might
wish to reject realism altogether and adopt nominalist structuralism. This
is the option we will ultimately recommend.

5. E x a m p l e s
The identity problem might not be a serious one were it not for the fact
that structures that fall prey to it pervade modern mathematics. In this
section we will examine some specific mathematical systems that prove

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

problematic for the realist structuralist. Our examples will fall into two
categories: algebraic and non-algebraic. Here 'algebraic' refers to systems
that have at least a group structure on them (such as groups, rings, and
fields), while 'non-algebraic' refers to systems that arc not algebraic in this
sense. We will conclude this section by isolating a particularly important
source of the identity problem: non-trivial automorphisms.
We begin with the non-algebraic case. First, any geometric pattern with a
reflectional symmetry contains places that are not distinguishable in terms
of their intra-structural relational properties. For example, consider the
two-point pattern: this is the structure exemplified by, say, any system
consisting of two distinct points in the Euclidean plane E 2 . Yet, there are
no infra-structural relations that could serve to distinguish the two places
of this structure from each other. Note that one cannot resolve the issue by
choosing a coordinate system: the two-point pattern is intended to be the
structure exemplified by any pair of points in E 2 and hence, to introduce
coordinates would be to introduce something strictly extra-structural.
Second, any graph with a reflectional symmetry contains places that are
not distinguishable in terms of their intra-structural relational properties.
For example, consider any system K exemplifying the directed graph K
with four vertices and four edges that looks like this:

a —* b
1 i

Write R.xy for 'there is an edge from x to 1/'; clearly, R is the only (non-
identity) relation in K. Let us now try to construct the structure K of K.
First, K would have to comprise four places, call them a, b, c, d, and
the single (non-identity) relation R. Second, recall that we are not allowed
to employ individual constants denoting elements in K nor ones denoting
places of K. Thus, the properties specified by the following four formulae
would have to suffice to individuate the places of K (quantification will be
over places of K):
322 KERANEN

(a) Vt
-^{y = z)&{Rxyk(RxzkVt{Rxt => (t = y)v(t = z)))),
(6) ->Rxxk3y(Ryx&.\/t(Rtx =>(t = y))&3z(Rxzk\/t(Rxt => (t = z))),
(c) ->Rxxk3y(Ryx&Mt(Rtx =>{t = y))k3z(Rxz&Mt(Rxt =>{t = z))),
(d) W->(Rxt)
^{y = z)k{Ryxk.{RzxkVt{Rtx =*{t = y)V{t = z))))).
The problem is now apparent: the properties intended to individuate b
and c are exactly the same. The places '&' and 'c' denote are 'structurally'
indiscernible and hence, by (STR), identical. Thus, the structure K of K

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

would seem to have three, not four places. More examples of this sort could
be manufactured ad nauseam.
We will now turn to the algebraic case. Again, the general problem
is perhaps best illustrated by first concentrating on a specific example.
Consider any system Z3 exemplifying the commutative group Z 3 of 'integers
modulo 3'. Z3 is a group that consists of equivalence classes of integers
under the relation

x = y if and only if x — y = 3n for some natural number n.

That is, two integers belong to the same class precisely when they differ by
a multiple of three. It is easy to see that Z3 has three elements, denoted
by '[0]', '[1]', and '[2]'. The group operation of Z3 is defined by [x] + [y] =
[x + y], where '+' on the right denotes the ordinary addition of integers.
For example, we have [1] + [2] = [0] because 1 + 2 = 3 = 0. More generally,
the following 'multiplication table' holds:
+ [0] [1] [2]
[0] [0] [1] [2]
[1] [1] [2] [0]
[2] [2] [0] [1]
Let us now try to construct the structure Z 3 of Z3. Again, recall that the
three places of Z 3 are supposed to be individuated by their intra-structural
relational properties alone, and that we are not allowed to employ individual
constants denoting elements in Z3 nor ones denoting places of Z 3 . Thus, the
properties specified by the following three formulae would have to suffice
to individuate the places of Z3 (quantification will be over places of Z 3 ):
([0]) {x + X = x) & Vy(x + 3 / = y)
(W) (x + X & 3y[(x + ? / ^ x) & 3Z((Z y & {x + z
([2]) (x + X ±x) & 3y\(x + y^ x) & 3z((z j & {x + z ^x))l.
 IDENTITY PROBLEM 323

The problem is again apparent: the properties intended to individuate [1]

and [2] are exactly the same. The places ' [1]' and ' [2]' denote are 'struc-
turally' indiscernible and hence, by (STR), identical. Thus, the structure
Z 3 of Z3 would seem to have two, not three places.
This should be no surprise. Indeed, now that the problem has been
made explicit, it is easy to see that the multiplication table of Z3 should
already have alerted us to it. We had [1] + [1] = [2] and [2] + [2] = [1]
on the one hand, and [1] + [2] = [0] = [2] + [1] on the other. As far as
intra-structural relations are concerned, [1] and [2] are indiscernible. A
mathematically informed reader will have recognized that this 'structural
symmetry' reflects the fact that Z3 has a non-trivial automorphism mapping

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

[1] to [2], [2] to [1], and [0] to [0]. In general, one can show that whenever a
system A has a non-trivial automorphism, there are at least two objects in
A that cannot be distinguished in terms of their intra-systemic relational
properties. Indeed, given a non-trivial automorphism / : A —> A, we claim
that there exist two distinct elements a and b in A such that a and b have
the same intra-systemic relational properties. To prove our claim, we only
need to recall (again) the following elementary model-theoretic fact: If / :
A —» A is an automorphism, a in A satisfies the formula <p(x) if and only if
/(a) satisfies <p(x). Since / : A —» A is by assumption non-trivial, there is
at least one a in A such that a ^ /(a). This proves our claim.
We have seen that non-trivial automorphisms invariably give rise to the
structural symmetry that is at the root of the identity problem. It is worth
noting that the converse does not hold: that two elements a and b satisfy
the same formulae does not entail that there is an automorphism mapping
a to b. Hence, what one might call the 'automorphism problem' is a special
case of the identity problem, although a very important one. For, as it
turns out, systems that have non-trivial automorphisms are very common
indeed. For example, every group of order other than 1 or 2 has such
automorphisms.22 The situation is not much better with rings and fields.
For example, the field of complex numbers C has an automorphism given
by g~- i —* —i. In fact, the rational, real, and p-adic number fields are the
6nly notable examples of fields with no non-trivial automorphisms.

6. No Solace for the Realist Structuralist

The identity problem arises from the interaction of three factors: the real-
ist structuralist conception of mathematical objects, her general-property
account of identity, and the mathematical facts as we know them. Thus,
since the realist structuralist would presumably not want to revise mathe-
matical facts, there are two ways in which she could try to solve the identity
problem. On the one hand, (1) she could try to furnish a general-property
22
See, for example, Jensen and Lenzing [1989], p. 63.
324 KERANEN

account that allows her to regard intra-structurally indiscernible places as

distinct. On the other hand, (2) she could try to argue that she does not
need to give a general-property account at all. We will consider each option
in turn and contend that neither one can save realist structuralism.
(1) If the realist structuralist tries to furnish a general-property account,
there are again two main directions in which she could venture. On the
one hand, (a) she could accept the constraints we placed on the properties
available for giving an account of identity, but argue that there are proper-
ties she could use besides the ones explored above. On the other hand, she
could reject the constraints we placed, and argue that there are more prop-

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

erties she could use than we have thus far allowed. In particular, she could
argue that in giving an account of identity, she should be allowed either (b)
to use individual constants denoting places, or (c) to use extra-structural
properties or non-relational properties.
(a) If the realist structuralist accepts our constraints, she can only use
intra-structural relational properties that can be specified by formulae with
no individual constants. So far, we have only considered formulae in one
free variable because such formulae naturally express properties enjoyed by
individual places. However, there is no reason in principle why the realist
structuralist could not try to use properties enjoyed by sets consisting of
several places. Unfortunately for her, however, in the case of systems with
non-trivial automorphisms, using such n-ary properties for n greater than
one is confronted with just the same problems as using unary properties.
As we recalled in §4, if / : A —> A is an automorphism, a sequence a\,... ,an
of elements in A satisfies the formula (p(x\,..., xn) if and only if its image
f(a,\),..., f{an) satisfies that formula. Thus, if S is a system with non-
trivial automorphisms, n-ary properties for n greater than one can provide
no help in solving the identity problem for the corresponding structure S.
Thus far, we have argued that given the constraints entailed by the spirit
and motivations of her theory, the realist structuralist cannot give a viable
general-property account for places. There were two crucial constraints.
First, she could not employ any property the specification of which essen-
tially involves an individual constant denoting an element in some system S
exemplifying the structure S in question. Second, she could not employ any
property the specification of which essentially involves an individual con-
stant denoting a place of S. Let us now finally consider whether either one
of these two constraints could be relaxed without compromising the spirit
and motivations of realist structuralism. Since relaxing the first constraint
would amount to rejecting the letter of realist structuralism, we shall start
with the second one.
(b) It may seem that if individual constants denoting places are allowed,
the identity problem is immediately resolved. For example, consider the
structure (Z, +). If individual constants denoting places are allowed, 1
 IDENTITY PROBLEM 325

and —1 will not be indiscernible: 1 has the property x + x = 2 whereas —1

does not; conversely, —1 has the property x + x = —2 whereas 1 does not.
And, likewise for all the other places of (Z, +). One might consequently
think that the success of our argument depends critically on the two reasons
we gave for disallowing individual constants denoting places. We will now
argue that this is actually not the case: it would be futile for the realist
structuralist to employ properties the specification of which essentially in-
volves individual constants denoting places of S in her account of identity
forS.
We granted the realist structuralist that the places of S exist and are
properly individuated objects before she ever completes the blank in (IS);

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

indeed, a realist structuralist could hardly settle for anything less (see
fn. 15). This might make it seem that there cannot possibly be any question
as to whether, say, 1 = — 1 according to her account of (Z, + ) . Indeed, it
might seem that the realist structuralist should be allowed to use individ-
ual constants denoting places of S in her account of identity for any given
structure S. The reason why this impression would be mistaken is that the
realist structuralist is supposed to be providing a primitive, indigenous ac-
count of identity for the places of any given structure (see §3). Her account
of identity is supposed to tell us which particular identity statements, ac-
cording to her theory, are true in the language of S. Thus, while we may
grant that the places of S exist and are properly individuated before the
realist structuralist ever completes the blank in (IS), we should not make
any concessions as to which particular places S contains. Given the nature
of the task she has set for herself, the realist structuralist has no license to
suppose of any particular identity statement that it is true (or false) in the
language of S until she has given an account of identity for S. Indeed, as far
as her theory is concerned, 'a— b1 is neither true nor false in the language
of S until she has given an account of identity for S. And, once she has,
' a = b' is true in the language of S if and only if her account entails that
it is.23
If this is right, it would be futile for the realist structuralist to employ
individual constants denoting places of S in her account of identity for
S. Consider again the structure (Z, + ) . The formulae x + x = 2 and
x + x = —2 could serve to distinguish 1 from —1 only provided that 2 ^ — 2
is true in the language of (Z, + ) . However, this is precisely what the realist
structuralist is not entitled to suppose until she has given an account of
identity that entails that 2 ^ —2. Thus, she is faced with a circularity

23
Assuming, of course, that the identity and distinctness of objects is not just a matter
of what there is, but also of how we choose to talk about what there is. Given that most
realist structuralists subscribe to a broadly Quinean picture of ontology, this assumption
should seem acceptable to them (see, for example, Resnik [1981] and Shapiro [1997),
pp. 92-93).
326 KERANEN

that vitiates her account. If the ontological project of realist structuralism

is taken in full seriousness, the sort of structures that are subject to the
identity problem remain problematic for the realist structuralist even if she
is allowed to employ individual constants denoting places.
(c) We have seen that if the realist structuralist is allowed to use only
intra-structural relational properties, she cannot solve the identity problem.
Thus, if she still maintains that the identity of places is governed by their
general properties, there is only one option left: The realist structuralist
could argue that she should be allowed to use extra-structural properties
or non-relational properties in giving her account of identity. For example,
she could argue that she should be allowed to use relations the specifica-

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

tion of which essentially involves individual constants denoting elements in
systems exemplifying the structure in question. The problem is that using
extra-structural properties or non-relational properties would amount to
rejecting structuralism in ontology. For, once again, structuralism can be
viewed as an elaboration of the intuition that mathematical objects 'have
no properties other than those relating them to the other "elements" of
the same structure' (Benacerraf [1965] in Benacerraf and Putnam [1983],
p. 291). Moreover, if the realist structuralist concedes that she can only
use properties that are invariant under automorphisms, she must concede
that in general she can only use intra-structural relational properties. It is
then difficult to see just how she could use extra-structural properties, let
alone non-relational ones, without compromising the very motivation, not
to say the spirit, of her theory.24

24
We should note that Robert Brandom exploits the existence of non-trivial automor-
phisms in his 'The significance of complex numbers for Frege's philosophy of mathemat-
ics' [1996]. Brandom shows that Frege's 'objectivist version of logicism' cannot work
because for Frege a singular term may be introduced only when its putative referent
can be uniquely specified (p. 296), and this is not possible when the domain in question
has non-trivial automorphisms. Brandom thinks that he has a remedy ready at hand,
however. The idea is to distinguish two types of specifiability: hypothetical and cate-
gorical. Elements of a domain are hypothetically specifiable when they are 'specifiable
(uniquely) relative to some other elements of the same domain, or assuming that others
have already been picked out' (p. 310). Elements of a domain are categorically speci-
fiable when they are 'specifiable (uniquely) by the distinctive role they play within the
domain, or in terms of their distinctive relation to what is outside the domain, to what
can be specified antecedently to the domain in question' (ibid.). Given the prevalence
of non-trivial automorphisms, suggests Brandom, one should relax Frege's uniqueness
requirement on the entitlement to introduce singular terms by insisting on hypothetical,
but not categorical specifiability (p. 313). Given, then, that the identity problem for
the realist structuralist arises for the same reason the failure of categorical specifiability
arises for Frcge, one might think that the structuralist could adopt some form of Bran-
dom's suggestion. It seems to us, however, that this thought would be wrong. This is
because Brandom's discussion operates on the level of singular reference, whereas the
realist structuralist must confront the identity problem on the level of ontology. The
question is not whether she can succeed in fixing either one of the square roots of —1 as
the referent of V. Rather, the question is whether, according to the realist structuralist
account of identity, there actually are two square roots of —1 in the structure (C, x, +).
 IDENTITY PROBLEM 327

(2) We have seen that if the realist structuralist takes the identity of
places to be governed by their general properties, she cannot solve the
identity problem. Let us now turn to the second option available to her: She
could try to argue that she does not need to give a general-property account
at all. There are again two main directions in which she could venture. On
the one hand, (a) she could eliminate the need to give a general-property
account by giving a haecceity account. On the other hand, (b) she could
try to argue that she does not need to give any account of identity.
(a) It seems that the realist structuralist cannot embrace the idea that
the identity of places is governed by their haecceities. First, haecceities
are patently non-relational items, and certainly not invariant under auto-

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

morphisms. Thus, adopting a haecceity account would amount to reject-
ing structuralism in ontology. Second, the consensus among contemporary
metaphysicians appears to be that abstract objects are a prime example of
things whose identity is governed by their general properties rather than by
their haecceities, if such there be.25 Third, it is worth recalling that most
realist structuralists tend to hold broadly Quinean views on ontology (see
fn. 23). It is doubtful, therefore, that they would be prepared to adopt a
form of substantivalism.
(b) We now come to the last, and perhaps the most radical solution the
realist structuralist could propose: not to give an account of identity at all
and maintain that the identity of places is primitive. While this may be a
logical possibility, it seems that the realist structuralist cannot embrace it.
After all, being a realist, she maintains that places in structures are properly
individuated objects. Thus, the realist structuralist should either say what
sort of facts govern the identity of places by giving an account of identity,
or else explain why it is not possible to do so. It is hard to see what such
an explanation could look like. If the identity of places is indeed governed
either by general properties or by haecceities, we know that it is possible to
give an account of identity for them. Thus, if the realist structuralist refuses
to give an account of identity, she must suppose that the identity of places
is governed neither by general properties nor by haecceities. Indeed, even if
the realist structuralist rejects all general-property accounts, it seems that
she must adopt a haecceity account rather than no account at all. For,
even if one maintains that distinct objects need not be distinguishable in

Thus, as we have argued, the realist structuralist cannot employ individual constants
that denote places of S when completing the blank in (IS) and so, she cannot distinguish
intra-structurally indiscernible places in terms of other particular places of S. It is worth
noting that the realist structuralist can of course recover 'categorical specifiability' if
she is prepared to embrace the second clause in Brandom's definition: by appealing to
extra-structural properties, she can distinguish the two square roots of —1 in the struc-
ture (C, X, +). However, as we have just argued, doing so would amount to rejecting
realist structuralism.
25
See, for example, Swinburne [1995] and Adams [1979J.
328 KERANEN

terms of their general properties, surely one.must still suppose that there is
some fact that metaphysically differentiates distinct objects: after all, how
else could they be distinct?26 And, if that fact is not provided by general
properties, it seems that it can only be provided by haecceities.
Finally, we may put the demand for an account of identity into perspec-
tive by contrasting realist structuralism with axiomatic set theory. Recall
first that the realist structuralist purports to be providing a philosophical
account of mathematical ontology. Recall also that axiomatic set theory
can be seen as the archetype of such an account. Finally, recall that the
problematic schema (STR) is a direct counterpart of the Axiom of Exten-
sionality of Zermelo-Fraenkel set theory. Now, in some sense the Axiom of

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

Extensionality tells us just what makes a set, it codifies a key ingredient in
our understanding of what sets are in the first place. Thus, it seems that
when Zermelo-Fraenkel set theory is considered as a philosophical (rather
than a mathematical) theory, the Axiom of Extensionality is indispensable
to it: without such an axiom, the theory would not really tell us what
sets are. Likewise, it seems that when realist structuralism is considered
as a philosophical theory, a non-primitive account of place-identity is in-
dispensable to it. Without such an account, the theory would not really
tell us what places are. It seems to us that this would be an intolerable
situation.27

7. Conclusion
While the claims put forth in this paper may seem surprising at first,
they really should not. Contemporary realism about mathematics in gen-
eral, and realist structuralism in particular, have been heavily influenced
by model theory. Although some authors such as Shapiro have insisted
that model theory plays only a superficial role in the realist account of
mathematics,28 it is clear that most realists have drawn their intuitions
therefrom. In model theory one takes the various domains of discourse
26
One might dismiss this sort of consideration altogether as rabidly metaphysical. How-
ever, if one is prepared to entertain something as esoteric as the rejection of the Leibniz
principle, one is probably also prepared to accept that there must be some 'metaphysical
foundation' for the identity and difference of objects.
27
We note here one further possibility. Perhaps the realist structuralist should only in-
clude certain 'basic' structures such as the set-theoretic hierarchy within the scope of her
account of identity, leaving out structures with non-trivial automorphisms. Such 'non-
basic' structures could then be accounted for in some other way, perhaps by resorting to
some form of nominalist structuralism. We cannot examine this possibility in detail here;
we will only make two brief remarks. First, it is not obvious that all structures one would
naturally consider 'basic' lack non-trivial automorphisms. For example, the Euclidean
plane seems like a natural candidate for a 'basic' structure and yet, it has infinitely many
non-trivial automorphisms. Second, we contend that adopting a hybrid position of the
sort envisaged here would in any case amount to rejecting realist structuralism proper.
28
See Shapiro |1997), p. 47.
 IDENTITY PROBLEM 329

as given, and assumes that there is no difficulty in securing reference to

objects in these domains. This, we suspect, is the reason why problems
of the kind explored in this paper have so far gone virtually unnoticed.29
While the structuralist purports to be constructing a foundational account
of mathematical ontology, she nevertheless remains captive to the comfort-
ing picture model theory offers. As we have argued, however, as soon as the
task of furnishing ontology is taken seriously, such a complacent attitude
towards identity and reference must be rejected. And once we do reject it,
we come to realize that in the case of systems with structurally indiscernible
elements, the idea of treating the structure of such a system as an object in
its own right is incoherent. At any rate, as soon as we extract the structure

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

of a system that has structurally indiscernible elements, it collapses before
our very eyes. When treated as 'freestanding' objects, certain structures
just do not have the shape we would intuitively think.30 Another way of
putting the point is to say that Benacerraf was right all along: if mathe-
matical entities have no properties besides the ones relating them to the
other elements in the same structure, they are not properly individuated
objects at all. We can now see why he was right.

References
A D A M S , R. M. [1979]: 'Primitive thisness and primitive identity', Journal of
Philosophy 76, 5-26.
B E N A C E R R A F , P . [1965]: ' W h a t numbers could not be', Philosophical Review 74,
47-73; reprinted in Benacerraf and P u t n a m [1983], pp. 272-294.
[1973]: 'Mathematical t r u t h ', Journal of Philosophy 70, 661-679; re-
printed in Benacerraf and P u t n a m [1983], pp. 403-420.
B E N A C E R R A F , P . , and H. P U T N A M , eds. [1983]: Philosophy of Mathematics.
Second ed. Cambridge: Cambridge University Press.
B R A N D O M , R O B E R T [1996]: ' T h e significance of complex numbers for Frege's
philosophy of mathematics', Proceedings of the Aristotelian Society 9 6 , 2 9 3 -
315.
BURGESS, J O H N [1999]: Review of Stewart Shapiro [1997], Notre Dame Journal
of Formal Logic 4 0 , 283-291.
CHIIIARA, C. S. [1973]: Ontology and the Vicious-Circle Principle. Ithaca, N. Y.:
Cornell University Press.
COLYVAN, M. [1998]: 'In defense of indispensability', Philosophia Mathematica
(3) 6, 39-62.
F I E L D , H. [1980]: Science Without Numbers. Princeton: Princeton University
Press.
H A L E , B. [1996]: 'Structuralism's unpaid epistemological debts', Philosophia
Mathematica (3) 4, 124-147.

29
Virtually, but not entirely. See Burgess [1999).
30
This might arouse the suspicion that the very idea of treating structures as freestand-
ing objects is misconceived. This is not, however, the place to pursue this suspicion.
330 KERANEN

HART, W. D., ed. [1993]: The Philosophy of Mathematics. Oxford: Oxford

University Press.
HELLMAN, G. [1996]: 'Structuralism without structures', Philosophia Mathema-
tica (3) 4, 100-123.
[2001]: 'Three varieties of mathematical structuralism', Philosophia
Mathematica (3) 9, 184-211.
JENSEN, C , and H. LENZING [1989]: Model Theoretic Algebra. New York: Gor-
don and Breach Science Publishers.
LOWE, E. J. [1995]: 'The metaphysics of abstract objects', Journal of Philosophy
92, 509-524.
PARSONS, C. [1990]: 'The structuralist view of mathematical objects', Synthese

Downloaded from http://philmat.oxfordjournals.org/ at Universite Paris Diderot on May 29, 2015

84, 303-346; reprinted in Hart [1993], pp. 272-309.
RESNIK, M. [1981]: 'Mathematics as a science of patterns: Ontology and refer-
ence', Nous 15, 529-550.
SHAPIRO, S. [1997]: Philosophy of Mathematics: Structure and Ontology. Ox-
ford: Oxford University Press.
SWINBURNE, R. [1995]: 'Thisness', Australian Journal of Philosophy 73, 389-400.
WEAVER, G. [1998]: 'Structuralism and representation theorems', Philosophia
Mathematica (3) 6, 257-271.

ABSTRACT. According to realist structuralism, mathematical objects are places

in abstract structures. We argue that in spite of its many attractions, realist
structuralism must be rejected. For, first, mathematical structures typically con-
tain intra-structurally indiscernible places. Second, any account of place-identity
available to the realist structuralist entails that intra-structurally indiscernible
places are identical. Since for her mathematical singular terms denote places in
structures, she would have to say, for example, that 1 = — 1 in the group (Z, +).
We call this the identity problem and conclude that nominalism is presently the
safest route for the structuralist.